Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1 | /* |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2 | * linux/fs/ext4/inode.c |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3 | * |
| 4 | * Copyright (C) 1992, 1993, 1994, 1995 |
| 5 | * Remy Card (card@masi.ibp.fr) |
| 6 | * Laboratoire MASI - Institut Blaise Pascal |
| 7 | * Universite Pierre et Marie Curie (Paris VI) |
| 8 | * |
| 9 | * from |
| 10 | * |
| 11 | * linux/fs/minix/inode.c |
| 12 | * |
| 13 | * Copyright (C) 1991, 1992 Linus Torvalds |
| 14 | * |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 15 | * 64-bit file support on 64-bit platforms by Jakub Jelinek |
| 16 | * (jj@sunsite.ms.mff.cuni.cz) |
| 17 | * |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 18 | * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000 |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 19 | */ |
| 20 | |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 21 | #include <linux/fs.h> |
| 22 | #include <linux/time.h> |
Mingming Cao | dab291a | 2006-10-11 01:21:01 -0700 | [diff] [blame] | 23 | #include <linux/jbd2.h> |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 24 | #include <linux/highuid.h> |
| 25 | #include <linux/pagemap.h> |
| 26 | #include <linux/quotaops.h> |
| 27 | #include <linux/string.h> |
| 28 | #include <linux/buffer_head.h> |
| 29 | #include <linux/writeback.h> |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 30 | #include <linux/pagevec.h> |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 31 | #include <linux/mpage.h> |
Duane Griffin | e83c139 | 2008-12-19 20:47:15 +0000 | [diff] [blame] | 32 | #include <linux/namei.h> |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 33 | #include <linux/uio.h> |
| 34 | #include <linux/bio.h> |
Mingming Cao | 4c0425f | 2009-09-28 15:48:41 -0400 | [diff] [blame] | 35 | #include <linux/workqueue.h> |
Jiaying Zhang | 744692d | 2010-03-04 16:14:02 -0500 | [diff] [blame] | 36 | #include <linux/kernel.h> |
Andrew Morton | 6db26ff | 2011-01-12 16:59:13 -0800 | [diff] [blame] | 37 | #include <linux/printk.h> |
Tejun Heo | 5a0e3ad | 2010-03-24 17:04:11 +0900 | [diff] [blame] | 38 | #include <linux/slab.h> |
Theodore Ts'o | a8901d3 | 2010-12-17 10:40:47 -0500 | [diff] [blame] | 39 | #include <linux/ratelimit.h> |
Kent Overstreet | a27bb33 | 2013-05-07 16:19:08 -0700 | [diff] [blame] | 40 | #include <linux/aio.h> |
Theodore Ts'o | 9bffad1 | 2009-06-17 11:48:11 -0400 | [diff] [blame] | 41 | |
Christoph Hellwig | 3dcf545 | 2008-04-29 18:13:32 -0400 | [diff] [blame] | 42 | #include "ext4_jbd2.h" |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 43 | #include "xattr.h" |
| 44 | #include "acl.h" |
Theodore Ts'o | 9f125d6 | 2011-06-27 19:16:04 -0400 | [diff] [blame] | 45 | #include "truncate.h" |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 46 | |
Theodore Ts'o | 9bffad1 | 2009-06-17 11:48:11 -0400 | [diff] [blame] | 47 | #include <trace/events/ext4.h> |
| 48 | |
Aneesh Kumar K.V | a1d6cc5 | 2008-08-19 21:55:02 -0400 | [diff] [blame] | 49 | #define MPAGE_DA_EXTENT_TAIL 0x01 |
| 50 | |
Darrick J. Wong | 814525f | 2012-04-29 18:31:10 -0400 | [diff] [blame] | 51 | static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw, |
| 52 | struct ext4_inode_info *ei) |
| 53 | { |
| 54 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
| 55 | __u16 csum_lo; |
| 56 | __u16 csum_hi = 0; |
| 57 | __u32 csum; |
| 58 | |
Dmitry Monakhov | 171a7f2 | 2013-04-09 23:56:48 -0400 | [diff] [blame] | 59 | csum_lo = le16_to_cpu(raw->i_checksum_lo); |
Darrick J. Wong | 814525f | 2012-04-29 18:31:10 -0400 | [diff] [blame] | 60 | raw->i_checksum_lo = 0; |
| 61 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && |
| 62 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) { |
Dmitry Monakhov | 171a7f2 | 2013-04-09 23:56:48 -0400 | [diff] [blame] | 63 | csum_hi = le16_to_cpu(raw->i_checksum_hi); |
Darrick J. Wong | 814525f | 2012-04-29 18:31:10 -0400 | [diff] [blame] | 64 | raw->i_checksum_hi = 0; |
| 65 | } |
| 66 | |
| 67 | csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw, |
| 68 | EXT4_INODE_SIZE(inode->i_sb)); |
| 69 | |
Dmitry Monakhov | 171a7f2 | 2013-04-09 23:56:48 -0400 | [diff] [blame] | 70 | raw->i_checksum_lo = cpu_to_le16(csum_lo); |
Darrick J. Wong | 814525f | 2012-04-29 18:31:10 -0400 | [diff] [blame] | 71 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && |
| 72 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) |
Dmitry Monakhov | 171a7f2 | 2013-04-09 23:56:48 -0400 | [diff] [blame] | 73 | raw->i_checksum_hi = cpu_to_le16(csum_hi); |
Darrick J. Wong | 814525f | 2012-04-29 18:31:10 -0400 | [diff] [blame] | 74 | |
| 75 | return csum; |
| 76 | } |
| 77 | |
| 78 | static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw, |
| 79 | struct ext4_inode_info *ei) |
| 80 | { |
| 81 | __u32 provided, calculated; |
| 82 | |
| 83 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != |
| 84 | cpu_to_le32(EXT4_OS_LINUX) || |
| 85 | !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, |
| 86 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) |
| 87 | return 1; |
| 88 | |
| 89 | provided = le16_to_cpu(raw->i_checksum_lo); |
| 90 | calculated = ext4_inode_csum(inode, raw, ei); |
| 91 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && |
| 92 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) |
| 93 | provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16; |
| 94 | else |
| 95 | calculated &= 0xFFFF; |
| 96 | |
| 97 | return provided == calculated; |
| 98 | } |
| 99 | |
| 100 | static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw, |
| 101 | struct ext4_inode_info *ei) |
| 102 | { |
| 103 | __u32 csum; |
| 104 | |
| 105 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != |
| 106 | cpu_to_le32(EXT4_OS_LINUX) || |
| 107 | !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, |
| 108 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) |
| 109 | return; |
| 110 | |
| 111 | csum = ext4_inode_csum(inode, raw, ei); |
| 112 | raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF); |
| 113 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && |
| 114 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) |
| 115 | raw->i_checksum_hi = cpu_to_le16(csum >> 16); |
| 116 | } |
| 117 | |
Jan Kara | 678aaf4 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 118 | static inline int ext4_begin_ordered_truncate(struct inode *inode, |
| 119 | loff_t new_size) |
| 120 | { |
Theodore Ts'o | 7ff9c07 | 2010-11-08 13:51:33 -0500 | [diff] [blame] | 121 | trace_ext4_begin_ordered_truncate(inode, new_size); |
Theodore Ts'o | 8aefcd5 | 2011-01-10 12:29:43 -0500 | [diff] [blame] | 122 | /* |
| 123 | * If jinode is zero, then we never opened the file for |
| 124 | * writing, so there's no need to call |
| 125 | * jbd2_journal_begin_ordered_truncate() since there's no |
| 126 | * outstanding writes we need to flush. |
| 127 | */ |
| 128 | if (!EXT4_I(inode)->jinode) |
| 129 | return 0; |
| 130 | return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode), |
| 131 | EXT4_I(inode)->jinode, |
| 132 | new_size); |
Jan Kara | 678aaf4 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 133 | } |
| 134 | |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 135 | static void ext4_invalidatepage(struct page *page, unsigned long offset); |
Theodore Ts'o | cb20d51 | 2010-10-27 21:30:09 -0400 | [diff] [blame] | 136 | static int __ext4_journalled_writepage(struct page *page, unsigned int len); |
| 137 | static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh); |
Eric Sandeen | 5f163cc | 2012-01-04 22:33:28 -0500 | [diff] [blame] | 138 | static int ext4_discard_partial_page_buffers_no_lock(handle_t *handle, |
| 139 | struct inode *inode, struct page *page, loff_t from, |
| 140 | loff_t length, int flags); |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 141 | |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 142 | /* |
| 143 | * Test whether an inode is a fast symlink. |
| 144 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 145 | static int ext4_inode_is_fast_symlink(struct inode *inode) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 146 | { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 147 | int ea_blocks = EXT4_I(inode)->i_file_acl ? |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 148 | (inode->i_sb->s_blocksize >> 9) : 0; |
| 149 | |
| 150 | return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0); |
| 151 | } |
| 152 | |
| 153 | /* |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 154 | * Restart the transaction associated with *handle. This does a commit, |
| 155 | * so before we call here everything must be consistently dirtied against |
| 156 | * this transaction. |
| 157 | */ |
Aneesh Kumar K.V | fa5d111 | 2009-11-02 18:50:49 -0500 | [diff] [blame] | 158 | int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode, |
Jan Kara | 487caee | 2009-08-17 22:17:20 -0400 | [diff] [blame] | 159 | int nblocks) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 160 | { |
Jan Kara | 487caee | 2009-08-17 22:17:20 -0400 | [diff] [blame] | 161 | int ret; |
| 162 | |
| 163 | /* |
Theodore Ts'o | e35fd66 | 2010-05-16 19:00:00 -0400 | [diff] [blame] | 164 | * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this |
Jan Kara | 487caee | 2009-08-17 22:17:20 -0400 | [diff] [blame] | 165 | * moment, get_block can be called only for blocks inside i_size since |
| 166 | * page cache has been already dropped and writes are blocked by |
| 167 | * i_mutex. So we can safely drop the i_data_sem here. |
| 168 | */ |
Frank Mayhar | 0390131 | 2009-01-07 00:06:22 -0500 | [diff] [blame] | 169 | BUG_ON(EXT4_JOURNAL(inode) == NULL); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 170 | jbd_debug(2, "restarting handle %p\n", handle); |
Jan Kara | 487caee | 2009-08-17 22:17:20 -0400 | [diff] [blame] | 171 | up_write(&EXT4_I(inode)->i_data_sem); |
Amir Goldstein | 8e8eaab | 2011-02-27 23:32:12 -0500 | [diff] [blame] | 172 | ret = ext4_journal_restart(handle, nblocks); |
Jan Kara | 487caee | 2009-08-17 22:17:20 -0400 | [diff] [blame] | 173 | down_write(&EXT4_I(inode)->i_data_sem); |
Aneesh Kumar K.V | fa5d111 | 2009-11-02 18:50:49 -0500 | [diff] [blame] | 174 | ext4_discard_preallocations(inode); |
Jan Kara | 487caee | 2009-08-17 22:17:20 -0400 | [diff] [blame] | 175 | |
| 176 | return ret; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 177 | } |
| 178 | |
| 179 | /* |
| 180 | * Called at the last iput() if i_nlink is zero. |
| 181 | */ |
Al Viro | 0930fcc | 2010-06-07 13:16:22 -0400 | [diff] [blame] | 182 | void ext4_evict_inode(struct inode *inode) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 183 | { |
| 184 | handle_t *handle; |
Theodore Ts'o | bc965ab | 2008-08-02 21:10:38 -0400 | [diff] [blame] | 185 | int err; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 186 | |
Theodore Ts'o | 7ff9c07 | 2010-11-08 13:51:33 -0500 | [diff] [blame] | 187 | trace_ext4_evict_inode(inode); |
Jiaying Zhang | 2581fdc | 2011-08-13 12:17:13 -0400 | [diff] [blame] | 188 | |
Al Viro | 0930fcc | 2010-06-07 13:16:22 -0400 | [diff] [blame] | 189 | if (inode->i_nlink) { |
Jan Kara | 2d859db | 2011-07-26 09:07:11 -0400 | [diff] [blame] | 190 | /* |
| 191 | * When journalling data dirty buffers are tracked only in the |
| 192 | * journal. So although mm thinks everything is clean and |
| 193 | * ready for reaping the inode might still have some pages to |
| 194 | * write in the running transaction or waiting to be |
| 195 | * checkpointed. Thus calling jbd2_journal_invalidatepage() |
| 196 | * (via truncate_inode_pages()) to discard these buffers can |
| 197 | * cause data loss. Also even if we did not discard these |
| 198 | * buffers, we would have no way to find them after the inode |
| 199 | * is reaped and thus user could see stale data if he tries to |
| 200 | * read them before the transaction is checkpointed. So be |
| 201 | * careful and force everything to disk here... We use |
| 202 | * ei->i_datasync_tid to store the newest transaction |
| 203 | * containing inode's data. |
| 204 | * |
| 205 | * Note that directories do not have this problem because they |
| 206 | * don't use page cache. |
| 207 | */ |
| 208 | if (ext4_should_journal_data(inode) && |
Theodore Ts'o | 2b405bf | 2013-03-20 09:42:11 -0400 | [diff] [blame] | 209 | (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) && |
| 210 | inode->i_ino != EXT4_JOURNAL_INO) { |
Jan Kara | 2d859db | 2011-07-26 09:07:11 -0400 | [diff] [blame] | 211 | journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; |
| 212 | tid_t commit_tid = EXT4_I(inode)->i_datasync_tid; |
| 213 | |
Theodore Ts'o | d76a3a77 | 2013-04-03 22:02:52 -0400 | [diff] [blame] | 214 | jbd2_complete_transaction(journal, commit_tid); |
Jan Kara | 2d859db | 2011-07-26 09:07:11 -0400 | [diff] [blame] | 215 | filemap_write_and_wait(&inode->i_data); |
| 216 | } |
Al Viro | 0930fcc | 2010-06-07 13:16:22 -0400 | [diff] [blame] | 217 | truncate_inode_pages(&inode->i_data, 0); |
Theodore Ts'o | 1ada47d | 2013-03-20 09:39:42 -0400 | [diff] [blame] | 218 | ext4_ioend_shutdown(inode); |
Al Viro | 0930fcc | 2010-06-07 13:16:22 -0400 | [diff] [blame] | 219 | goto no_delete; |
| 220 | } |
| 221 | |
Christoph Hellwig | 907f455 | 2010-03-03 09:05:06 -0500 | [diff] [blame] | 222 | if (!is_bad_inode(inode)) |
Christoph Hellwig | 871a293 | 2010-03-03 09:05:07 -0500 | [diff] [blame] | 223 | dquot_initialize(inode); |
Christoph Hellwig | 907f455 | 2010-03-03 09:05:06 -0500 | [diff] [blame] | 224 | |
Jan Kara | 678aaf4 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 225 | if (ext4_should_order_data(inode)) |
| 226 | ext4_begin_ordered_truncate(inode, 0); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 227 | truncate_inode_pages(&inode->i_data, 0); |
Theodore Ts'o | 1ada47d | 2013-03-20 09:39:42 -0400 | [diff] [blame] | 228 | ext4_ioend_shutdown(inode); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 229 | |
| 230 | if (is_bad_inode(inode)) |
| 231 | goto no_delete; |
| 232 | |
Jan Kara | 8e8ad8a | 2012-06-12 16:20:38 +0200 | [diff] [blame] | 233 | /* |
| 234 | * Protect us against freezing - iput() caller didn't have to have any |
| 235 | * protection against it |
| 236 | */ |
| 237 | sb_start_intwrite(inode->i_sb); |
Theodore Ts'o | 9924a92 | 2013-02-08 21:59:22 -0500 | [diff] [blame] | 238 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, |
| 239 | ext4_blocks_for_truncate(inode)+3); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 240 | if (IS_ERR(handle)) { |
Theodore Ts'o | bc965ab | 2008-08-02 21:10:38 -0400 | [diff] [blame] | 241 | ext4_std_error(inode->i_sb, PTR_ERR(handle)); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 242 | /* |
| 243 | * If we're going to skip the normal cleanup, we still need to |
| 244 | * make sure that the in-core orphan linked list is properly |
| 245 | * cleaned up. |
| 246 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 247 | ext4_orphan_del(NULL, inode); |
Jan Kara | 8e8ad8a | 2012-06-12 16:20:38 +0200 | [diff] [blame] | 248 | sb_end_intwrite(inode->i_sb); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 249 | goto no_delete; |
| 250 | } |
| 251 | |
| 252 | if (IS_SYNC(inode)) |
Frank Mayhar | 0390131 | 2009-01-07 00:06:22 -0500 | [diff] [blame] | 253 | ext4_handle_sync(handle); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 254 | inode->i_size = 0; |
Theodore Ts'o | bc965ab | 2008-08-02 21:10:38 -0400 | [diff] [blame] | 255 | err = ext4_mark_inode_dirty(handle, inode); |
| 256 | if (err) { |
Eric Sandeen | 12062dd | 2010-02-15 14:19:27 -0500 | [diff] [blame] | 257 | ext4_warning(inode->i_sb, |
Theodore Ts'o | bc965ab | 2008-08-02 21:10:38 -0400 | [diff] [blame] | 258 | "couldn't mark inode dirty (err %d)", err); |
| 259 | goto stop_handle; |
| 260 | } |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 261 | if (inode->i_blocks) |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 262 | ext4_truncate(inode); |
Theodore Ts'o | bc965ab | 2008-08-02 21:10:38 -0400 | [diff] [blame] | 263 | |
| 264 | /* |
| 265 | * ext4_ext_truncate() doesn't reserve any slop when it |
| 266 | * restarts journal transactions; therefore there may not be |
| 267 | * enough credits left in the handle to remove the inode from |
| 268 | * the orphan list and set the dtime field. |
| 269 | */ |
Frank Mayhar | 0390131 | 2009-01-07 00:06:22 -0500 | [diff] [blame] | 270 | if (!ext4_handle_has_enough_credits(handle, 3)) { |
Theodore Ts'o | bc965ab | 2008-08-02 21:10:38 -0400 | [diff] [blame] | 271 | err = ext4_journal_extend(handle, 3); |
| 272 | if (err > 0) |
| 273 | err = ext4_journal_restart(handle, 3); |
| 274 | if (err != 0) { |
Eric Sandeen | 12062dd | 2010-02-15 14:19:27 -0500 | [diff] [blame] | 275 | ext4_warning(inode->i_sb, |
Theodore Ts'o | bc965ab | 2008-08-02 21:10:38 -0400 | [diff] [blame] | 276 | "couldn't extend journal (err %d)", err); |
| 277 | stop_handle: |
| 278 | ext4_journal_stop(handle); |
Theodore Ts'o | 4538821 | 2010-07-29 15:06:10 -0400 | [diff] [blame] | 279 | ext4_orphan_del(NULL, inode); |
Jan Kara | 8e8ad8a | 2012-06-12 16:20:38 +0200 | [diff] [blame] | 280 | sb_end_intwrite(inode->i_sb); |
Theodore Ts'o | bc965ab | 2008-08-02 21:10:38 -0400 | [diff] [blame] | 281 | goto no_delete; |
| 282 | } |
| 283 | } |
| 284 | |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 285 | /* |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 286 | * Kill off the orphan record which ext4_truncate created. |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 287 | * AKPM: I think this can be inside the above `if'. |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 288 | * Note that ext4_orphan_del() has to be able to cope with the |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 289 | * deletion of a non-existent orphan - this is because we don't |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 290 | * know if ext4_truncate() actually created an orphan record. |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 291 | * (Well, we could do this if we need to, but heck - it works) |
| 292 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 293 | ext4_orphan_del(handle, inode); |
| 294 | EXT4_I(inode)->i_dtime = get_seconds(); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 295 | |
| 296 | /* |
| 297 | * One subtle ordering requirement: if anything has gone wrong |
| 298 | * (transaction abort, IO errors, whatever), then we can still |
| 299 | * do these next steps (the fs will already have been marked as |
| 300 | * having errors), but we can't free the inode if the mark_dirty |
| 301 | * fails. |
| 302 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 303 | if (ext4_mark_inode_dirty(handle, inode)) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 304 | /* If that failed, just do the required in-core inode clear. */ |
Al Viro | 0930fcc | 2010-06-07 13:16:22 -0400 | [diff] [blame] | 305 | ext4_clear_inode(inode); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 306 | else |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 307 | ext4_free_inode(handle, inode); |
| 308 | ext4_journal_stop(handle); |
Jan Kara | 8e8ad8a | 2012-06-12 16:20:38 +0200 | [diff] [blame] | 309 | sb_end_intwrite(inode->i_sb); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 310 | return; |
| 311 | no_delete: |
Al Viro | 0930fcc | 2010-06-07 13:16:22 -0400 | [diff] [blame] | 312 | ext4_clear_inode(inode); /* We must guarantee clearing of inode... */ |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 313 | } |
| 314 | |
Dmitry Monakhov | a9e7f44 | 2009-12-14 15:21:14 +0300 | [diff] [blame] | 315 | #ifdef CONFIG_QUOTA |
| 316 | qsize_t *ext4_get_reserved_space(struct inode *inode) |
Mingming Cao | 60e58e0 | 2009-01-22 18:13:05 +0100 | [diff] [blame] | 317 | { |
Dmitry Monakhov | a9e7f44 | 2009-12-14 15:21:14 +0300 | [diff] [blame] | 318 | return &EXT4_I(inode)->i_reserved_quota; |
Mingming Cao | 60e58e0 | 2009-01-22 18:13:05 +0100 | [diff] [blame] | 319 | } |
Dmitry Monakhov | a9e7f44 | 2009-12-14 15:21:14 +0300 | [diff] [blame] | 320 | #endif |
Theodore Ts'o | 9d0be50 | 2010-01-01 02:41:30 -0500 | [diff] [blame] | 321 | |
Aneesh Kumar K.V | 12219ae | 2008-07-17 16:12:08 -0400 | [diff] [blame] | 322 | /* |
| 323 | * Calculate the number of metadata blocks need to reserve |
Theodore Ts'o | 9d0be50 | 2010-01-01 02:41:30 -0500 | [diff] [blame] | 324 | * to allocate a block located at @lblock |
Aneesh Kumar K.V | 12219ae | 2008-07-17 16:12:08 -0400 | [diff] [blame] | 325 | */ |
Theodore Ts'o | 01f49d0 | 2011-01-10 12:13:03 -0500 | [diff] [blame] | 326 | static int ext4_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock) |
Aneesh Kumar K.V | 12219ae | 2008-07-17 16:12:08 -0400 | [diff] [blame] | 327 | { |
Dmitry Monakhov | 12e9b89 | 2010-05-16 22:00:00 -0400 | [diff] [blame] | 328 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
Theodore Ts'o | 9d0be50 | 2010-01-01 02:41:30 -0500 | [diff] [blame] | 329 | return ext4_ext_calc_metadata_amount(inode, lblock); |
Aneesh Kumar K.V | 12219ae | 2008-07-17 16:12:08 -0400 | [diff] [blame] | 330 | |
Amir Goldstein | 8bb2b24 | 2011-06-27 17:10:28 -0400 | [diff] [blame] | 331 | return ext4_ind_calc_metadata_amount(inode, lblock); |
Aneesh Kumar K.V | 12219ae | 2008-07-17 16:12:08 -0400 | [diff] [blame] | 332 | } |
| 333 | |
Theodore Ts'o | 0637c6f | 2009-12-30 14:20:45 -0500 | [diff] [blame] | 334 | /* |
| 335 | * Called with i_data_sem down, which is important since we can call |
| 336 | * ext4_discard_preallocations() from here. |
| 337 | */ |
Aneesh Kumar K.V | 5f634d0 | 2010-01-25 04:00:31 -0500 | [diff] [blame] | 338 | void ext4_da_update_reserve_space(struct inode *inode, |
| 339 | int used, int quota_claim) |
Aneesh Kumar K.V | 12219ae | 2008-07-17 16:12:08 -0400 | [diff] [blame] | 340 | { |
| 341 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
Theodore Ts'o | 0637c6f | 2009-12-30 14:20:45 -0500 | [diff] [blame] | 342 | struct ext4_inode_info *ei = EXT4_I(inode); |
Aneesh Kumar K.V | 12219ae | 2008-07-17 16:12:08 -0400 | [diff] [blame] | 343 | |
Theodore Ts'o | 0637c6f | 2009-12-30 14:20:45 -0500 | [diff] [blame] | 344 | spin_lock(&ei->i_block_reservation_lock); |
Aditya Kali | d899024 | 2011-09-09 19:18:51 -0400 | [diff] [blame] | 345 | trace_ext4_da_update_reserve_space(inode, used, quota_claim); |
Theodore Ts'o | 0637c6f | 2009-12-30 14:20:45 -0500 | [diff] [blame] | 346 | if (unlikely(used > ei->i_reserved_data_blocks)) { |
Theodore Ts'o | 8de5c32 | 2013-02-14 15:11:41 -0500 | [diff] [blame] | 347 | ext4_warning(inode->i_sb, "%s: ino %lu, used %d " |
Theodore Ts'o | 1084f25 | 2012-03-19 23:13:43 -0400 | [diff] [blame] | 348 | "with only %d reserved data blocks", |
Theodore Ts'o | 0637c6f | 2009-12-30 14:20:45 -0500 | [diff] [blame] | 349 | __func__, inode->i_ino, used, |
| 350 | ei->i_reserved_data_blocks); |
| 351 | WARN_ON(1); |
| 352 | used = ei->i_reserved_data_blocks; |
Aneesh Kumar K.V | 6bc6e63 | 2008-10-10 09:39:00 -0400 | [diff] [blame] | 353 | } |
Aneesh Kumar K.V | 12219ae | 2008-07-17 16:12:08 -0400 | [diff] [blame] | 354 | |
Brian Foster | 97795d2 | 2012-07-22 23:59:40 -0400 | [diff] [blame] | 355 | if (unlikely(ei->i_allocated_meta_blocks > ei->i_reserved_meta_blocks)) { |
Theodore Ts'o | 01a523e | 2013-02-14 15:51:58 -0500 | [diff] [blame] | 356 | ext4_warning(inode->i_sb, "ino %lu, allocated %d " |
| 357 | "with only %d reserved metadata blocks " |
| 358 | "(releasing %d blocks with reserved %d data blocks)", |
| 359 | inode->i_ino, ei->i_allocated_meta_blocks, |
| 360 | ei->i_reserved_meta_blocks, used, |
| 361 | ei->i_reserved_data_blocks); |
Brian Foster | 97795d2 | 2012-07-22 23:59:40 -0400 | [diff] [blame] | 362 | WARN_ON(1); |
| 363 | ei->i_allocated_meta_blocks = ei->i_reserved_meta_blocks; |
| 364 | } |
| 365 | |
Theodore Ts'o | 0637c6f | 2009-12-30 14:20:45 -0500 | [diff] [blame] | 366 | /* Update per-inode reservations */ |
| 367 | ei->i_reserved_data_blocks -= used; |
Theodore Ts'o | 0637c6f | 2009-12-30 14:20:45 -0500 | [diff] [blame] | 368 | ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks; |
Theodore Ts'o | 5704265 | 2011-09-09 18:56:51 -0400 | [diff] [blame] | 369 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, |
Eric Sandeen | 72b8ab9 | 2010-05-16 11:00:00 -0400 | [diff] [blame] | 370 | used + ei->i_allocated_meta_blocks); |
Theodore Ts'o | 0637c6f | 2009-12-30 14:20:45 -0500 | [diff] [blame] | 371 | ei->i_allocated_meta_blocks = 0; |
Theodore Ts'o | 0637c6f | 2009-12-30 14:20:45 -0500 | [diff] [blame] | 372 | |
| 373 | if (ei->i_reserved_data_blocks == 0) { |
| 374 | /* |
| 375 | * We can release all of the reserved metadata blocks |
| 376 | * only when we have written all of the delayed |
| 377 | * allocation blocks. |
| 378 | */ |
Theodore Ts'o | 5704265 | 2011-09-09 18:56:51 -0400 | [diff] [blame] | 379 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, |
Eric Sandeen | 72b8ab9 | 2010-05-16 11:00:00 -0400 | [diff] [blame] | 380 | ei->i_reserved_meta_blocks); |
Theodore Ts'o | ee5f4d9 | 2010-01-01 02:36:15 -0500 | [diff] [blame] | 381 | ei->i_reserved_meta_blocks = 0; |
Theodore Ts'o | 9d0be50 | 2010-01-01 02:41:30 -0500 | [diff] [blame] | 382 | ei->i_da_metadata_calc_len = 0; |
Theodore Ts'o | 0637c6f | 2009-12-30 14:20:45 -0500 | [diff] [blame] | 383 | } |
Aneesh Kumar K.V | 12219ae | 2008-07-17 16:12:08 -0400 | [diff] [blame] | 384 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
Mingming Cao | 60e58e0 | 2009-01-22 18:13:05 +0100 | [diff] [blame] | 385 | |
Eric Sandeen | 72b8ab9 | 2010-05-16 11:00:00 -0400 | [diff] [blame] | 386 | /* Update quota subsystem for data blocks */ |
| 387 | if (quota_claim) |
Aditya Kali | 7b415bf | 2011-09-09 19:04:51 -0400 | [diff] [blame] | 388 | dquot_claim_block(inode, EXT4_C2B(sbi, used)); |
Eric Sandeen | 72b8ab9 | 2010-05-16 11:00:00 -0400 | [diff] [blame] | 389 | else { |
Aneesh Kumar K.V | 5f634d0 | 2010-01-25 04:00:31 -0500 | [diff] [blame] | 390 | /* |
| 391 | * We did fallocate with an offset that is already delayed |
| 392 | * allocated. So on delayed allocated writeback we should |
Eric Sandeen | 72b8ab9 | 2010-05-16 11:00:00 -0400 | [diff] [blame] | 393 | * not re-claim the quota for fallocated blocks. |
Aneesh Kumar K.V | 5f634d0 | 2010-01-25 04:00:31 -0500 | [diff] [blame] | 394 | */ |
Aditya Kali | 7b415bf | 2011-09-09 19:04:51 -0400 | [diff] [blame] | 395 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, used)); |
Aneesh Kumar K.V | 5f634d0 | 2010-01-25 04:00:31 -0500 | [diff] [blame] | 396 | } |
Aneesh Kumar K.V | d601430 | 2009-03-27 22:36:43 -0400 | [diff] [blame] | 397 | |
| 398 | /* |
| 399 | * If we have done all the pending block allocations and if |
| 400 | * there aren't any writers on the inode, we can discard the |
| 401 | * inode's preallocations. |
| 402 | */ |
Theodore Ts'o | 0637c6f | 2009-12-30 14:20:45 -0500 | [diff] [blame] | 403 | if ((ei->i_reserved_data_blocks == 0) && |
| 404 | (atomic_read(&inode->i_writecount) == 0)) |
Aneesh Kumar K.V | d601430 | 2009-03-27 22:36:43 -0400 | [diff] [blame] | 405 | ext4_discard_preallocations(inode); |
Aneesh Kumar K.V | 12219ae | 2008-07-17 16:12:08 -0400 | [diff] [blame] | 406 | } |
| 407 | |
Theodore Ts'o | e29136f | 2010-06-29 12:54:28 -0400 | [diff] [blame] | 408 | static int __check_block_validity(struct inode *inode, const char *func, |
Theodore Ts'o | c398eda | 2010-07-27 11:56:40 -0400 | [diff] [blame] | 409 | unsigned int line, |
| 410 | struct ext4_map_blocks *map) |
Theodore Ts'o | 6fd058f | 2009-05-17 15:38:01 -0400 | [diff] [blame] | 411 | { |
Theodore Ts'o | 24676da | 2010-05-16 21:00:00 -0400 | [diff] [blame] | 412 | if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk, |
| 413 | map->m_len)) { |
Theodore Ts'o | c398eda | 2010-07-27 11:56:40 -0400 | [diff] [blame] | 414 | ext4_error_inode(inode, func, line, map->m_pblk, |
| 415 | "lblock %lu mapped to illegal pblock " |
| 416 | "(length %d)", (unsigned long) map->m_lblk, |
| 417 | map->m_len); |
Theodore Ts'o | 6fd058f | 2009-05-17 15:38:01 -0400 | [diff] [blame] | 418 | return -EIO; |
| 419 | } |
| 420 | return 0; |
| 421 | } |
| 422 | |
Theodore Ts'o | e29136f | 2010-06-29 12:54:28 -0400 | [diff] [blame] | 423 | #define check_block_validity(inode, map) \ |
Theodore Ts'o | c398eda | 2010-07-27 11:56:40 -0400 | [diff] [blame] | 424 | __check_block_validity((inode), __func__, __LINE__, (map)) |
Theodore Ts'o | e29136f | 2010-06-29 12:54:28 -0400 | [diff] [blame] | 425 | |
Mingming Cao | f5ab0d1 | 2008-02-25 15:29:55 -0500 | [diff] [blame] | 426 | /* |
Theodore Ts'o | 1f94533 | 2009-09-30 22:57:41 -0400 | [diff] [blame] | 427 | * Return the number of contiguous dirty pages in a given inode |
| 428 | * starting at page frame idx. |
Theodore Ts'o | 55138e0 | 2009-09-29 13:31:31 -0400 | [diff] [blame] | 429 | */ |
| 430 | static pgoff_t ext4_num_dirty_pages(struct inode *inode, pgoff_t idx, |
| 431 | unsigned int max_pages) |
| 432 | { |
| 433 | struct address_space *mapping = inode->i_mapping; |
| 434 | pgoff_t index; |
| 435 | struct pagevec pvec; |
| 436 | pgoff_t num = 0; |
| 437 | int i, nr_pages, done = 0; |
| 438 | |
| 439 | if (max_pages == 0) |
| 440 | return 0; |
| 441 | pagevec_init(&pvec, 0); |
| 442 | while (!done) { |
| 443 | index = idx; |
| 444 | nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, |
| 445 | PAGECACHE_TAG_DIRTY, |
| 446 | (pgoff_t)PAGEVEC_SIZE); |
| 447 | if (nr_pages == 0) |
| 448 | break; |
| 449 | for (i = 0; i < nr_pages; i++) { |
| 450 | struct page *page = pvec.pages[i]; |
| 451 | struct buffer_head *bh, *head; |
| 452 | |
| 453 | lock_page(page); |
| 454 | if (unlikely(page->mapping != mapping) || |
| 455 | !PageDirty(page) || |
| 456 | PageWriteback(page) || |
| 457 | page->index != idx) { |
| 458 | done = 1; |
| 459 | unlock_page(page); |
| 460 | break; |
| 461 | } |
Theodore Ts'o | 1f94533 | 2009-09-30 22:57:41 -0400 | [diff] [blame] | 462 | if (page_has_buffers(page)) { |
| 463 | bh = head = page_buffers(page); |
| 464 | do { |
| 465 | if (!buffer_delay(bh) && |
| 466 | !buffer_unwritten(bh)) |
| 467 | done = 1; |
| 468 | bh = bh->b_this_page; |
| 469 | } while (!done && (bh != head)); |
| 470 | } |
Theodore Ts'o | 55138e0 | 2009-09-29 13:31:31 -0400 | [diff] [blame] | 471 | unlock_page(page); |
| 472 | if (done) |
| 473 | break; |
| 474 | idx++; |
| 475 | num++; |
Eric Sandeen | 659c600 | 2010-10-27 21:30:03 -0400 | [diff] [blame] | 476 | if (num >= max_pages) { |
| 477 | done = 1; |
Theodore Ts'o | 55138e0 | 2009-09-29 13:31:31 -0400 | [diff] [blame] | 478 | break; |
Eric Sandeen | 659c600 | 2010-10-27 21:30:03 -0400 | [diff] [blame] | 479 | } |
Theodore Ts'o | 55138e0 | 2009-09-29 13:31:31 -0400 | [diff] [blame] | 480 | } |
| 481 | pagevec_release(&pvec); |
| 482 | } |
| 483 | return num; |
| 484 | } |
| 485 | |
Dmitry Monakhov | 921f266 | 2013-03-10 21:01:03 -0400 | [diff] [blame] | 486 | #ifdef ES_AGGRESSIVE_TEST |
| 487 | static void ext4_map_blocks_es_recheck(handle_t *handle, |
| 488 | struct inode *inode, |
| 489 | struct ext4_map_blocks *es_map, |
| 490 | struct ext4_map_blocks *map, |
| 491 | int flags) |
| 492 | { |
| 493 | int retval; |
| 494 | |
| 495 | map->m_flags = 0; |
| 496 | /* |
| 497 | * There is a race window that the result is not the same. |
| 498 | * e.g. xfstests #223 when dioread_nolock enables. The reason |
| 499 | * is that we lookup a block mapping in extent status tree with |
| 500 | * out taking i_data_sem. So at the time the unwritten extent |
| 501 | * could be converted. |
| 502 | */ |
| 503 | if (!(flags & EXT4_GET_BLOCKS_NO_LOCK)) |
| 504 | down_read((&EXT4_I(inode)->i_data_sem)); |
| 505 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
| 506 | retval = ext4_ext_map_blocks(handle, inode, map, flags & |
| 507 | EXT4_GET_BLOCKS_KEEP_SIZE); |
| 508 | } else { |
| 509 | retval = ext4_ind_map_blocks(handle, inode, map, flags & |
| 510 | EXT4_GET_BLOCKS_KEEP_SIZE); |
| 511 | } |
| 512 | if (!(flags & EXT4_GET_BLOCKS_NO_LOCK)) |
| 513 | up_read((&EXT4_I(inode)->i_data_sem)); |
| 514 | /* |
| 515 | * Clear EXT4_MAP_FROM_CLUSTER and EXT4_MAP_BOUNDARY flag |
| 516 | * because it shouldn't be marked in es_map->m_flags. |
| 517 | */ |
| 518 | map->m_flags &= ~(EXT4_MAP_FROM_CLUSTER | EXT4_MAP_BOUNDARY); |
| 519 | |
| 520 | /* |
| 521 | * We don't check m_len because extent will be collpased in status |
| 522 | * tree. So the m_len might not equal. |
| 523 | */ |
| 524 | if (es_map->m_lblk != map->m_lblk || |
| 525 | es_map->m_flags != map->m_flags || |
| 526 | es_map->m_pblk != map->m_pblk) { |
| 527 | printk("ES cache assertation failed for inode: %lu " |
| 528 | "es_cached ex [%d/%d/%llu/%x] != " |
| 529 | "found ex [%d/%d/%llu/%x] retval %d flags %x\n", |
| 530 | inode->i_ino, es_map->m_lblk, es_map->m_len, |
| 531 | es_map->m_pblk, es_map->m_flags, map->m_lblk, |
| 532 | map->m_len, map->m_pblk, map->m_flags, |
| 533 | retval, flags); |
| 534 | } |
| 535 | } |
| 536 | #endif /* ES_AGGRESSIVE_TEST */ |
| 537 | |
Theodore Ts'o | 55138e0 | 2009-09-29 13:31:31 -0400 | [diff] [blame] | 538 | /* |
Theodore Ts'o | e35fd66 | 2010-05-16 19:00:00 -0400 | [diff] [blame] | 539 | * The ext4_map_blocks() function tries to look up the requested blocks, |
Theodore Ts'o | 2b2d6d0 | 2008-07-26 16:15:44 -0400 | [diff] [blame] | 540 | * and returns if the blocks are already mapped. |
Mingming Cao | f5ab0d1 | 2008-02-25 15:29:55 -0500 | [diff] [blame] | 541 | * |
Mingming Cao | f5ab0d1 | 2008-02-25 15:29:55 -0500 | [diff] [blame] | 542 | * Otherwise it takes the write lock of the i_data_sem and allocate blocks |
| 543 | * and store the allocated blocks in the result buffer head and mark it |
| 544 | * mapped. |
| 545 | * |
Theodore Ts'o | e35fd66 | 2010-05-16 19:00:00 -0400 | [diff] [blame] | 546 | * If file type is extents based, it will call ext4_ext_map_blocks(), |
| 547 | * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping |
Mingming Cao | f5ab0d1 | 2008-02-25 15:29:55 -0500 | [diff] [blame] | 548 | * based files |
| 549 | * |
| 550 | * On success, it returns the number of blocks being mapped or allocate. |
| 551 | * if create==0 and the blocks are pre-allocated and uninitialized block, |
| 552 | * the result buffer head is unmapped. If the create ==1, it will make sure |
| 553 | * the buffer head is mapped. |
| 554 | * |
| 555 | * It returns 0 if plain look up failed (blocks have not been allocated), in |
Tao Ma | df3ab17 | 2011-10-08 15:53:49 -0400 | [diff] [blame] | 556 | * that case, buffer head is unmapped |
Mingming Cao | f5ab0d1 | 2008-02-25 15:29:55 -0500 | [diff] [blame] | 557 | * |
| 558 | * It returns the error in case of allocation failure. |
| 559 | */ |
Theodore Ts'o | e35fd66 | 2010-05-16 19:00:00 -0400 | [diff] [blame] | 560 | int ext4_map_blocks(handle_t *handle, struct inode *inode, |
| 561 | struct ext4_map_blocks *map, int flags) |
Aneesh Kumar K.V | 0e855ac | 2008-01-28 23:58:26 -0500 | [diff] [blame] | 562 | { |
Zheng Liu | d100eef | 2013-02-18 00:29:59 -0500 | [diff] [blame] | 563 | struct extent_status es; |
Aneesh Kumar K.V | 0e855ac | 2008-01-28 23:58:26 -0500 | [diff] [blame] | 564 | int retval; |
Dmitry Monakhov | 921f266 | 2013-03-10 21:01:03 -0400 | [diff] [blame] | 565 | #ifdef ES_AGGRESSIVE_TEST |
| 566 | struct ext4_map_blocks orig_map; |
| 567 | |
| 568 | memcpy(&orig_map, map, sizeof(*map)); |
| 569 | #endif |
Mingming Cao | f5ab0d1 | 2008-02-25 15:29:55 -0500 | [diff] [blame] | 570 | |
Theodore Ts'o | e35fd66 | 2010-05-16 19:00:00 -0400 | [diff] [blame] | 571 | map->m_flags = 0; |
| 572 | ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u," |
| 573 | "logical block %lu\n", inode->i_ino, flags, map->m_len, |
| 574 | (unsigned long) map->m_lblk); |
Zheng Liu | d100eef | 2013-02-18 00:29:59 -0500 | [diff] [blame] | 575 | |
| 576 | /* Lookup extent status tree firstly */ |
| 577 | if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) { |
| 578 | if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) { |
| 579 | map->m_pblk = ext4_es_pblock(&es) + |
| 580 | map->m_lblk - es.es_lblk; |
| 581 | map->m_flags |= ext4_es_is_written(&es) ? |
| 582 | EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN; |
| 583 | retval = es.es_len - (map->m_lblk - es.es_lblk); |
| 584 | if (retval > map->m_len) |
| 585 | retval = map->m_len; |
| 586 | map->m_len = retval; |
| 587 | } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) { |
| 588 | retval = 0; |
| 589 | } else { |
| 590 | BUG_ON(1); |
| 591 | } |
Dmitry Monakhov | 921f266 | 2013-03-10 21:01:03 -0400 | [diff] [blame] | 592 | #ifdef ES_AGGRESSIVE_TEST |
| 593 | ext4_map_blocks_es_recheck(handle, inode, map, |
| 594 | &orig_map, flags); |
| 595 | #endif |
Zheng Liu | d100eef | 2013-02-18 00:29:59 -0500 | [diff] [blame] | 596 | goto found; |
| 597 | } |
| 598 | |
Aneesh Kumar K.V | 4df3d26 | 2008-01-28 23:58:29 -0500 | [diff] [blame] | 599 | /* |
Theodore Ts'o | b920c75 | 2009-05-14 00:54:29 -0400 | [diff] [blame] | 600 | * Try to see if we can get the block without requesting a new |
| 601 | * file system block. |
Aneesh Kumar K.V | 4df3d26 | 2008-01-28 23:58:29 -0500 | [diff] [blame] | 602 | */ |
Zheng Liu | 729f52c | 2012-07-09 16:29:29 -0400 | [diff] [blame] | 603 | if (!(flags & EXT4_GET_BLOCKS_NO_LOCK)) |
| 604 | down_read((&EXT4_I(inode)->i_data_sem)); |
Dmitry Monakhov | 12e9b89 | 2010-05-16 22:00:00 -0400 | [diff] [blame] | 605 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
Dmitry Monakhov | a4e5d88 | 2011-10-25 08:15:12 -0400 | [diff] [blame] | 606 | retval = ext4_ext_map_blocks(handle, inode, map, flags & |
| 607 | EXT4_GET_BLOCKS_KEEP_SIZE); |
Aneesh Kumar K.V | 0e855ac | 2008-01-28 23:58:26 -0500 | [diff] [blame] | 608 | } else { |
Dmitry Monakhov | a4e5d88 | 2011-10-25 08:15:12 -0400 | [diff] [blame] | 609 | retval = ext4_ind_map_blocks(handle, inode, map, flags & |
| 610 | EXT4_GET_BLOCKS_KEEP_SIZE); |
Aneesh Kumar K.V | 0e855ac | 2008-01-28 23:58:26 -0500 | [diff] [blame] | 611 | } |
Zheng Liu | f7fec03 | 2013-02-18 00:28:47 -0500 | [diff] [blame] | 612 | if (retval > 0) { |
| 613 | int ret; |
| 614 | unsigned long long status; |
| 615 | |
Dmitry Monakhov | 921f266 | 2013-03-10 21:01:03 -0400 | [diff] [blame] | 616 | #ifdef ES_AGGRESSIVE_TEST |
| 617 | if (retval != map->m_len) { |
| 618 | printk("ES len assertation failed for inode: %lu " |
| 619 | "retval %d != map->m_len %d " |
| 620 | "in %s (lookup)\n", inode->i_ino, retval, |
| 621 | map->m_len, __func__); |
| 622 | } |
| 623 | #endif |
| 624 | |
Zheng Liu | f7fec03 | 2013-02-18 00:28:47 -0500 | [diff] [blame] | 625 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
| 626 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; |
| 627 | if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) && |
| 628 | ext4_find_delalloc_range(inode, map->m_lblk, |
| 629 | map->m_lblk + map->m_len - 1)) |
| 630 | status |= EXTENT_STATUS_DELAYED; |
| 631 | ret = ext4_es_insert_extent(inode, map->m_lblk, |
| 632 | map->m_len, map->m_pblk, status); |
| 633 | if (ret < 0) |
| 634 | retval = ret; |
| 635 | } |
Zheng Liu | 729f52c | 2012-07-09 16:29:29 -0400 | [diff] [blame] | 636 | if (!(flags & EXT4_GET_BLOCKS_NO_LOCK)) |
| 637 | up_read((&EXT4_I(inode)->i_data_sem)); |
Mingming Cao | f5ab0d1 | 2008-02-25 15:29:55 -0500 | [diff] [blame] | 638 | |
Zheng Liu | d100eef | 2013-02-18 00:29:59 -0500 | [diff] [blame] | 639 | found: |
Theodore Ts'o | e35fd66 | 2010-05-16 19:00:00 -0400 | [diff] [blame] | 640 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
Zheng Liu | f7fec03 | 2013-02-18 00:28:47 -0500 | [diff] [blame] | 641 | int ret = check_block_validity(inode, map); |
Theodore Ts'o | 6fd058f | 2009-05-17 15:38:01 -0400 | [diff] [blame] | 642 | if (ret != 0) |
| 643 | return ret; |
| 644 | } |
| 645 | |
Mingming Cao | f5ab0d1 | 2008-02-25 15:29:55 -0500 | [diff] [blame] | 646 | /* If it is only a block(s) look up */ |
Theodore Ts'o | c217705 | 2009-05-14 00:58:52 -0400 | [diff] [blame] | 647 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) |
Aneesh Kumar K.V | 4df3d26 | 2008-01-28 23:58:29 -0500 | [diff] [blame] | 648 | return retval; |
| 649 | |
| 650 | /* |
Mingming Cao | f5ab0d1 | 2008-02-25 15:29:55 -0500 | [diff] [blame] | 651 | * Returns if the blocks have already allocated |
| 652 | * |
| 653 | * Note that if blocks have been preallocated |
Tao Ma | df3ab17 | 2011-10-08 15:53:49 -0400 | [diff] [blame] | 654 | * ext4_ext_get_block() returns the create = 0 |
Mingming Cao | f5ab0d1 | 2008-02-25 15:29:55 -0500 | [diff] [blame] | 655 | * with buffer head unmapped. |
| 656 | */ |
Theodore Ts'o | e35fd66 | 2010-05-16 19:00:00 -0400 | [diff] [blame] | 657 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) |
Mingming Cao | f5ab0d1 | 2008-02-25 15:29:55 -0500 | [diff] [blame] | 658 | return retval; |
| 659 | |
| 660 | /* |
Zheng Liu | a25a4e1 | 2013-02-18 00:28:04 -0500 | [diff] [blame] | 661 | * Here we clear m_flags because after allocating an new extent, |
| 662 | * it will be set again. |
Aneesh Kumar K.V | 2a8964d | 2009-05-14 17:05:39 -0400 | [diff] [blame] | 663 | */ |
Zheng Liu | a25a4e1 | 2013-02-18 00:28:04 -0500 | [diff] [blame] | 664 | map->m_flags &= ~EXT4_MAP_FLAGS; |
Aneesh Kumar K.V | 2a8964d | 2009-05-14 17:05:39 -0400 | [diff] [blame] | 665 | |
| 666 | /* |
Mingming Cao | f5ab0d1 | 2008-02-25 15:29:55 -0500 | [diff] [blame] | 667 | * New blocks allocate and/or writing to uninitialized extent |
| 668 | * will possibly result in updating i_data, so we take |
| 669 | * the write lock of i_data_sem, and call get_blocks() |
| 670 | * with create == 1 flag. |
Aneesh Kumar K.V | 4df3d26 | 2008-01-28 23:58:29 -0500 | [diff] [blame] | 671 | */ |
| 672 | down_write((&EXT4_I(inode)->i_data_sem)); |
Mingming Cao | d2a1763 | 2008-07-14 17:52:37 -0400 | [diff] [blame] | 673 | |
| 674 | /* |
| 675 | * if the caller is from delayed allocation writeout path |
| 676 | * we have already reserved fs blocks for allocation |
| 677 | * let the underlying get_block() function know to |
| 678 | * avoid double accounting |
| 679 | */ |
Theodore Ts'o | c217705 | 2009-05-14 00:58:52 -0400 | [diff] [blame] | 680 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) |
Theodore Ts'o | f232109 | 2011-01-10 12:12:36 -0500 | [diff] [blame] | 681 | ext4_set_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED); |
Aneesh Kumar K.V | 4df3d26 | 2008-01-28 23:58:29 -0500 | [diff] [blame] | 682 | /* |
| 683 | * We need to check for EXT4 here because migrate |
| 684 | * could have changed the inode type in between |
| 685 | */ |
Dmitry Monakhov | 12e9b89 | 2010-05-16 22:00:00 -0400 | [diff] [blame] | 686 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
Theodore Ts'o | e35fd66 | 2010-05-16 19:00:00 -0400 | [diff] [blame] | 687 | retval = ext4_ext_map_blocks(handle, inode, map, flags); |
Aneesh Kumar K.V | 0e855ac | 2008-01-28 23:58:26 -0500 | [diff] [blame] | 688 | } else { |
Theodore Ts'o | e35fd66 | 2010-05-16 19:00:00 -0400 | [diff] [blame] | 689 | retval = ext4_ind_map_blocks(handle, inode, map, flags); |
Aneesh Kumar K.V | 267e4db | 2008-04-29 08:11:12 -0400 | [diff] [blame] | 690 | |
Theodore Ts'o | e35fd66 | 2010-05-16 19:00:00 -0400 | [diff] [blame] | 691 | if (retval > 0 && map->m_flags & EXT4_MAP_NEW) { |
Aneesh Kumar K.V | 267e4db | 2008-04-29 08:11:12 -0400 | [diff] [blame] | 692 | /* |
| 693 | * We allocated new blocks which will result in |
| 694 | * i_data's format changing. Force the migrate |
| 695 | * to fail by clearing migrate flags |
| 696 | */ |
Theodore Ts'o | 19f5fb7 | 2010-01-24 14:34:07 -0500 | [diff] [blame] | 697 | ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE); |
Aneesh Kumar K.V | 267e4db | 2008-04-29 08:11:12 -0400 | [diff] [blame] | 698 | } |
Mingming Cao | d2a1763 | 2008-07-14 17:52:37 -0400 | [diff] [blame] | 699 | |
Aneesh Kumar K.V | 5f634d0 | 2010-01-25 04:00:31 -0500 | [diff] [blame] | 700 | /* |
| 701 | * Update reserved blocks/metadata blocks after successful |
| 702 | * block allocation which had been deferred till now. We don't |
| 703 | * support fallocate for non extent files. So we can update |
| 704 | * reserve space here. |
| 705 | */ |
| 706 | if ((retval > 0) && |
Aneesh Kumar K.V | 1296cc8 | 2010-01-15 01:27:59 -0500 | [diff] [blame] | 707 | (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)) |
Aneesh Kumar K.V | 5f634d0 | 2010-01-25 04:00:31 -0500 | [diff] [blame] | 708 | ext4_da_update_reserve_space(inode, retval, 1); |
| 709 | } |
Zheng Liu | f7fec03 | 2013-02-18 00:28:47 -0500 | [diff] [blame] | 710 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) |
Theodore Ts'o | f232109 | 2011-01-10 12:12:36 -0500 | [diff] [blame] | 711 | ext4_clear_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED); |
Theodore Ts'o | 2ac3b6e | 2009-05-14 13:57:08 -0400 | [diff] [blame] | 712 | |
Zheng Liu | f7fec03 | 2013-02-18 00:28:47 -0500 | [diff] [blame] | 713 | if (retval > 0) { |
| 714 | int ret; |
| 715 | unsigned long long status; |
| 716 | |
Dmitry Monakhov | 921f266 | 2013-03-10 21:01:03 -0400 | [diff] [blame] | 717 | #ifdef ES_AGGRESSIVE_TEST |
| 718 | if (retval != map->m_len) { |
| 719 | printk("ES len assertation failed for inode: %lu " |
| 720 | "retval %d != map->m_len %d " |
| 721 | "in %s (allocation)\n", inode->i_ino, retval, |
| 722 | map->m_len, __func__); |
| 723 | } |
| 724 | #endif |
| 725 | |
Zheng Liu | adb2355 | 2013-03-10 21:13:05 -0400 | [diff] [blame] | 726 | /* |
| 727 | * If the extent has been zeroed out, we don't need to update |
| 728 | * extent status tree. |
| 729 | */ |
| 730 | if ((flags & EXT4_GET_BLOCKS_PRE_IO) && |
| 731 | ext4_es_lookup_extent(inode, map->m_lblk, &es)) { |
| 732 | if (ext4_es_is_written(&es)) |
| 733 | goto has_zeroout; |
| 734 | } |
Zheng Liu | f7fec03 | 2013-02-18 00:28:47 -0500 | [diff] [blame] | 735 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
| 736 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; |
| 737 | if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) && |
| 738 | ext4_find_delalloc_range(inode, map->m_lblk, |
| 739 | map->m_lblk + map->m_len - 1)) |
| 740 | status |= EXTENT_STATUS_DELAYED; |
| 741 | ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, |
| 742 | map->m_pblk, status); |
| 743 | if (ret < 0) |
| 744 | retval = ret; |
Aditya Kali | 5356f26 | 2011-09-09 19:20:51 -0400 | [diff] [blame] | 745 | } |
| 746 | |
Zheng Liu | adb2355 | 2013-03-10 21:13:05 -0400 | [diff] [blame] | 747 | has_zeroout: |
Aneesh Kumar K.V | 4df3d26 | 2008-01-28 23:58:29 -0500 | [diff] [blame] | 748 | up_write((&EXT4_I(inode)->i_data_sem)); |
Theodore Ts'o | e35fd66 | 2010-05-16 19:00:00 -0400 | [diff] [blame] | 749 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
Theodore Ts'o | e29136f | 2010-06-29 12:54:28 -0400 | [diff] [blame] | 750 | int ret = check_block_validity(inode, map); |
Theodore Ts'o | 6fd058f | 2009-05-17 15:38:01 -0400 | [diff] [blame] | 751 | if (ret != 0) |
| 752 | return ret; |
| 753 | } |
Aneesh Kumar K.V | 0e855ac | 2008-01-28 23:58:26 -0500 | [diff] [blame] | 754 | return retval; |
| 755 | } |
| 756 | |
Mingming Cao | f3bd1f3 | 2008-08-19 22:16:03 -0400 | [diff] [blame] | 757 | /* Maximum number of blocks we map for direct IO at once. */ |
| 758 | #define DIO_MAX_BLOCKS 4096 |
| 759 | |
Theodore Ts'o | 2ed8868 | 2010-05-16 20:00:00 -0400 | [diff] [blame] | 760 | static int _ext4_get_block(struct inode *inode, sector_t iblock, |
| 761 | struct buffer_head *bh, int flags) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 762 | { |
Dmitriy Monakhov | 3e4fdaf | 2007-02-10 01:46:35 -0800 | [diff] [blame] | 763 | handle_t *handle = ext4_journal_current_handle(); |
Theodore Ts'o | 2ed8868 | 2010-05-16 20:00:00 -0400 | [diff] [blame] | 764 | struct ext4_map_blocks map; |
Jan Kara | 7fb5409 | 2008-02-10 01:08:38 -0500 | [diff] [blame] | 765 | int ret = 0, started = 0; |
Mingming Cao | f3bd1f3 | 2008-08-19 22:16:03 -0400 | [diff] [blame] | 766 | int dio_credits; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 767 | |
Tao Ma | 46c7f25 | 2012-12-10 14:04:52 -0500 | [diff] [blame] | 768 | if (ext4_has_inline_data(inode)) |
| 769 | return -ERANGE; |
| 770 | |
Theodore Ts'o | 2ed8868 | 2010-05-16 20:00:00 -0400 | [diff] [blame] | 771 | map.m_lblk = iblock; |
| 772 | map.m_len = bh->b_size >> inode->i_blkbits; |
| 773 | |
Anatol Pomozov | 8b0f165 | 2012-11-08 15:07:16 -0500 | [diff] [blame] | 774 | if (flags && !(flags & EXT4_GET_BLOCKS_NO_LOCK) && !handle) { |
Jan Kara | 7fb5409 | 2008-02-10 01:08:38 -0500 | [diff] [blame] | 775 | /* Direct IO write... */ |
Theodore Ts'o | 2ed8868 | 2010-05-16 20:00:00 -0400 | [diff] [blame] | 776 | if (map.m_len > DIO_MAX_BLOCKS) |
| 777 | map.m_len = DIO_MAX_BLOCKS; |
| 778 | dio_credits = ext4_chunk_trans_blocks(inode, map.m_len); |
Theodore Ts'o | 9924a92 | 2013-02-08 21:59:22 -0500 | [diff] [blame] | 779 | handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, |
| 780 | dio_credits); |
Jan Kara | 7fb5409 | 2008-02-10 01:08:38 -0500 | [diff] [blame] | 781 | if (IS_ERR(handle)) { |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 782 | ret = PTR_ERR(handle); |
Theodore Ts'o | 2ed8868 | 2010-05-16 20:00:00 -0400 | [diff] [blame] | 783 | return ret; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 784 | } |
Jan Kara | 7fb5409 | 2008-02-10 01:08:38 -0500 | [diff] [blame] | 785 | started = 1; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 786 | } |
| 787 | |
Theodore Ts'o | 2ed8868 | 2010-05-16 20:00:00 -0400 | [diff] [blame] | 788 | ret = ext4_map_blocks(handle, inode, &map, flags); |
Jan Kara | 7fb5409 | 2008-02-10 01:08:38 -0500 | [diff] [blame] | 789 | if (ret > 0) { |
Theodore Ts'o | 2ed8868 | 2010-05-16 20:00:00 -0400 | [diff] [blame] | 790 | map_bh(bh, inode->i_sb, map.m_pblk); |
| 791 | bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags; |
| 792 | bh->b_size = inode->i_sb->s_blocksize * map.m_len; |
Jan Kara | 7fb5409 | 2008-02-10 01:08:38 -0500 | [diff] [blame] | 793 | ret = 0; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 794 | } |
Jan Kara | 7fb5409 | 2008-02-10 01:08:38 -0500 | [diff] [blame] | 795 | if (started) |
| 796 | ext4_journal_stop(handle); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 797 | return ret; |
| 798 | } |
| 799 | |
Theodore Ts'o | 2ed8868 | 2010-05-16 20:00:00 -0400 | [diff] [blame] | 800 | int ext4_get_block(struct inode *inode, sector_t iblock, |
| 801 | struct buffer_head *bh, int create) |
| 802 | { |
| 803 | return _ext4_get_block(inode, iblock, bh, |
| 804 | create ? EXT4_GET_BLOCKS_CREATE : 0); |
| 805 | } |
| 806 | |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 807 | /* |
| 808 | * `handle' can be NULL if create is zero |
| 809 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 810 | struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode, |
Aneesh Kumar K.V | 725d26d | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 811 | ext4_lblk_t block, int create, int *errp) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 812 | { |
Theodore Ts'o | 2ed8868 | 2010-05-16 20:00:00 -0400 | [diff] [blame] | 813 | struct ext4_map_blocks map; |
| 814 | struct buffer_head *bh; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 815 | int fatal = 0, err; |
| 816 | |
| 817 | J_ASSERT(handle != NULL || create == 0); |
| 818 | |
Theodore Ts'o | 2ed8868 | 2010-05-16 20:00:00 -0400 | [diff] [blame] | 819 | map.m_lblk = block; |
| 820 | map.m_len = 1; |
| 821 | err = ext4_map_blocks(handle, inode, &map, |
| 822 | create ? EXT4_GET_BLOCKS_CREATE : 0); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 823 | |
Carlos Maiolino | 90b0a97 | 2012-09-17 23:39:12 -0400 | [diff] [blame] | 824 | /* ensure we send some value back into *errp */ |
| 825 | *errp = 0; |
| 826 | |
Theodore Ts'o | 0f70b40 | 2013-02-15 03:35:57 -0500 | [diff] [blame] | 827 | if (create && err == 0) |
| 828 | err = -ENOSPC; /* should never happen */ |
Theodore Ts'o | 2ed8868 | 2010-05-16 20:00:00 -0400 | [diff] [blame] | 829 | if (err < 0) |
| 830 | *errp = err; |
| 831 | if (err <= 0) |
| 832 | return NULL; |
Theodore Ts'o | 2ed8868 | 2010-05-16 20:00:00 -0400 | [diff] [blame] | 833 | |
| 834 | bh = sb_getblk(inode->i_sb, map.m_pblk); |
Wang Shilong | aebf024 | 2013-01-12 16:28:47 -0500 | [diff] [blame] | 835 | if (unlikely(!bh)) { |
Theodore Ts'o | 860d21e | 2013-01-12 16:19:36 -0500 | [diff] [blame] | 836 | *errp = -ENOMEM; |
Theodore Ts'o | 2ed8868 | 2010-05-16 20:00:00 -0400 | [diff] [blame] | 837 | return NULL; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 838 | } |
Theodore Ts'o | 2ed8868 | 2010-05-16 20:00:00 -0400 | [diff] [blame] | 839 | if (map.m_flags & EXT4_MAP_NEW) { |
| 840 | J_ASSERT(create != 0); |
| 841 | J_ASSERT(handle != NULL); |
| 842 | |
| 843 | /* |
| 844 | * Now that we do not always journal data, we should |
| 845 | * keep in mind whether this should always journal the |
| 846 | * new buffer as metadata. For now, regular file |
| 847 | * writes use ext4_get_block instead, so it's not a |
| 848 | * problem. |
| 849 | */ |
| 850 | lock_buffer(bh); |
| 851 | BUFFER_TRACE(bh, "call get_create_access"); |
| 852 | fatal = ext4_journal_get_create_access(handle, bh); |
| 853 | if (!fatal && !buffer_uptodate(bh)) { |
| 854 | memset(bh->b_data, 0, inode->i_sb->s_blocksize); |
| 855 | set_buffer_uptodate(bh); |
| 856 | } |
| 857 | unlock_buffer(bh); |
| 858 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
| 859 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
| 860 | if (!fatal) |
| 861 | fatal = err; |
| 862 | } else { |
| 863 | BUFFER_TRACE(bh, "not a new buffer"); |
| 864 | } |
| 865 | if (fatal) { |
| 866 | *errp = fatal; |
| 867 | brelse(bh); |
| 868 | bh = NULL; |
| 869 | } |
| 870 | return bh; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 871 | } |
| 872 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 873 | struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode, |
Aneesh Kumar K.V | 725d26d | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 874 | ext4_lblk_t block, int create, int *err) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 875 | { |
Theodore Ts'o | af5bc92 | 2008-09-08 22:25:24 -0400 | [diff] [blame] | 876 | struct buffer_head *bh; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 877 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 878 | bh = ext4_getblk(handle, inode, block, create, err); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 879 | if (!bh) |
| 880 | return bh; |
| 881 | if (buffer_uptodate(bh)) |
| 882 | return bh; |
Christoph Hellwig | 65299a3 | 2011-08-23 14:50:29 +0200 | [diff] [blame] | 883 | ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 884 | wait_on_buffer(bh); |
| 885 | if (buffer_uptodate(bh)) |
| 886 | return bh; |
| 887 | put_bh(bh); |
| 888 | *err = -EIO; |
| 889 | return NULL; |
| 890 | } |
| 891 | |
Tao Ma | f19d587 | 2012-12-10 14:05:51 -0500 | [diff] [blame] | 892 | int ext4_walk_page_buffers(handle_t *handle, |
| 893 | struct buffer_head *head, |
| 894 | unsigned from, |
| 895 | unsigned to, |
| 896 | int *partial, |
| 897 | int (*fn)(handle_t *handle, |
| 898 | struct buffer_head *bh)) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 899 | { |
| 900 | struct buffer_head *bh; |
| 901 | unsigned block_start, block_end; |
| 902 | unsigned blocksize = head->b_size; |
| 903 | int err, ret = 0; |
| 904 | struct buffer_head *next; |
| 905 | |
Theodore Ts'o | af5bc92 | 2008-09-08 22:25:24 -0400 | [diff] [blame] | 906 | for (bh = head, block_start = 0; |
| 907 | ret == 0 && (bh != head || !block_start); |
Theodore Ts'o | de9a55b | 2009-06-14 17:45:34 -0400 | [diff] [blame] | 908 | block_start = block_end, bh = next) { |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 909 | next = bh->b_this_page; |
| 910 | block_end = block_start + blocksize; |
| 911 | if (block_end <= from || block_start >= to) { |
| 912 | if (partial && !buffer_uptodate(bh)) |
| 913 | *partial = 1; |
| 914 | continue; |
| 915 | } |
| 916 | err = (*fn)(handle, bh); |
| 917 | if (!ret) |
| 918 | ret = err; |
| 919 | } |
| 920 | return ret; |
| 921 | } |
| 922 | |
| 923 | /* |
| 924 | * To preserve ordering, it is essential that the hole instantiation and |
| 925 | * the data write be encapsulated in a single transaction. We cannot |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 926 | * close off a transaction and start a new one between the ext4_get_block() |
Mingming Cao | dab291a | 2006-10-11 01:21:01 -0700 | [diff] [blame] | 927 | * and the commit_write(). So doing the jbd2_journal_start at the start of |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 928 | * prepare_write() is the right place. |
| 929 | * |
Jan Kara | 36ade45 | 2013-01-28 09:30:52 -0500 | [diff] [blame] | 930 | * Also, this function can nest inside ext4_writepage(). In that case, we |
| 931 | * *know* that ext4_writepage() has generated enough buffer credits to do the |
| 932 | * whole page. So we won't block on the journal in that case, which is good, |
| 933 | * because the caller may be PF_MEMALLOC. |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 934 | * |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 935 | * By accident, ext4 can be reentered when a transaction is open via |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 936 | * quota file writes. If we were to commit the transaction while thus |
| 937 | * reentered, there can be a deadlock - we would be holding a quota |
| 938 | * lock, and the commit would never complete if another thread had a |
| 939 | * transaction open and was blocking on the quota lock - a ranking |
| 940 | * violation. |
| 941 | * |
Mingming Cao | dab291a | 2006-10-11 01:21:01 -0700 | [diff] [blame] | 942 | * So what we do is to rely on the fact that jbd2_journal_stop/journal_start |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 943 | * will _not_ run commit under these circumstances because handle->h_ref |
| 944 | * is elevated. We'll still have enough credits for the tiny quotafile |
| 945 | * write. |
| 946 | */ |
Tao Ma | f19d587 | 2012-12-10 14:05:51 -0500 | [diff] [blame] | 947 | int do_journal_get_write_access(handle_t *handle, |
| 948 | struct buffer_head *bh) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 949 | { |
Jan Kara | 56d35a4 | 2010-08-05 14:41:42 -0400 | [diff] [blame] | 950 | int dirty = buffer_dirty(bh); |
| 951 | int ret; |
| 952 | |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 953 | if (!buffer_mapped(bh) || buffer_freed(bh)) |
| 954 | return 0; |
Jan Kara | 56d35a4 | 2010-08-05 14:41:42 -0400 | [diff] [blame] | 955 | /* |
Christoph Hellwig | ebdec24 | 2010-10-06 10:47:23 +0200 | [diff] [blame] | 956 | * __block_write_begin() could have dirtied some buffers. Clean |
Jan Kara | 56d35a4 | 2010-08-05 14:41:42 -0400 | [diff] [blame] | 957 | * the dirty bit as jbd2_journal_get_write_access() could complain |
| 958 | * otherwise about fs integrity issues. Setting of the dirty bit |
Christoph Hellwig | ebdec24 | 2010-10-06 10:47:23 +0200 | [diff] [blame] | 959 | * by __block_write_begin() isn't a real problem here as we clear |
Jan Kara | 56d35a4 | 2010-08-05 14:41:42 -0400 | [diff] [blame] | 960 | * the bit before releasing a page lock and thus writeback cannot |
| 961 | * ever write the buffer. |
| 962 | */ |
| 963 | if (dirty) |
| 964 | clear_buffer_dirty(bh); |
| 965 | ret = ext4_journal_get_write_access(handle, bh); |
| 966 | if (!ret && dirty) |
| 967 | ret = ext4_handle_dirty_metadata(handle, NULL, bh); |
| 968 | return ret; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 969 | } |
| 970 | |
Anatol Pomozov | 8b0f165 | 2012-11-08 15:07:16 -0500 | [diff] [blame] | 971 | static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock, |
| 972 | struct buffer_head *bh_result, int create); |
Nick Piggin | bfc1af6 | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 973 | static int ext4_write_begin(struct file *file, struct address_space *mapping, |
Theodore Ts'o | de9a55b | 2009-06-14 17:45:34 -0400 | [diff] [blame] | 974 | loff_t pos, unsigned len, unsigned flags, |
| 975 | struct page **pagep, void **fsdata) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 976 | { |
Theodore Ts'o | af5bc92 | 2008-09-08 22:25:24 -0400 | [diff] [blame] | 977 | struct inode *inode = mapping->host; |
Aneesh Kumar K.V | 1938a15 | 2009-06-05 01:00:26 -0400 | [diff] [blame] | 978 | int ret, needed_blocks; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 979 | handle_t *handle; |
| 980 | int retries = 0; |
Theodore Ts'o | af5bc92 | 2008-09-08 22:25:24 -0400 | [diff] [blame] | 981 | struct page *page; |
Theodore Ts'o | de9a55b | 2009-06-14 17:45:34 -0400 | [diff] [blame] | 982 | pgoff_t index; |
Theodore Ts'o | af5bc92 | 2008-09-08 22:25:24 -0400 | [diff] [blame] | 983 | unsigned from, to; |
Nick Piggin | bfc1af6 | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 984 | |
Theodore Ts'o | 9bffad1 | 2009-06-17 11:48:11 -0400 | [diff] [blame] | 985 | trace_ext4_write_begin(inode, pos, len, flags); |
Aneesh Kumar K.V | 1938a15 | 2009-06-05 01:00:26 -0400 | [diff] [blame] | 986 | /* |
| 987 | * Reserve one block more for addition to orphan list in case |
| 988 | * we allocate blocks but write fails for some reason |
| 989 | */ |
| 990 | needed_blocks = ext4_writepage_trans_blocks(inode) + 1; |
Theodore Ts'o | de9a55b | 2009-06-14 17:45:34 -0400 | [diff] [blame] | 991 | index = pos >> PAGE_CACHE_SHIFT; |
Theodore Ts'o | af5bc92 | 2008-09-08 22:25:24 -0400 | [diff] [blame] | 992 | from = pos & (PAGE_CACHE_SIZE - 1); |
| 993 | to = from + len; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 994 | |
Tao Ma | f19d587 | 2012-12-10 14:05:51 -0500 | [diff] [blame] | 995 | if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { |
| 996 | ret = ext4_try_to_write_inline_data(mapping, inode, pos, len, |
| 997 | flags, pagep); |
| 998 | if (ret < 0) |
Theodore Ts'o | 47564bf | 2013-02-09 09:24:14 -0500 | [diff] [blame] | 999 | return ret; |
| 1000 | if (ret == 1) |
| 1001 | return 0; |
Tao Ma | f19d587 | 2012-12-10 14:05:51 -0500 | [diff] [blame] | 1002 | } |
| 1003 | |
Theodore Ts'o | 47564bf | 2013-02-09 09:24:14 -0500 | [diff] [blame] | 1004 | /* |
| 1005 | * grab_cache_page_write_begin() can take a long time if the |
| 1006 | * system is thrashing due to memory pressure, or if the page |
| 1007 | * is being written back. So grab it first before we start |
| 1008 | * the transaction handle. This also allows us to allocate |
| 1009 | * the page (if needed) without using GFP_NOFS. |
| 1010 | */ |
| 1011 | retry_grab: |
Nick Piggin | 54566b2 | 2009-01-04 12:00:53 -0800 | [diff] [blame] | 1012 | page = grab_cache_page_write_begin(mapping, index, flags); |
Theodore Ts'o | 47564bf | 2013-02-09 09:24:14 -0500 | [diff] [blame] | 1013 | if (!page) |
| 1014 | return -ENOMEM; |
| 1015 | unlock_page(page); |
| 1016 | |
| 1017 | retry_journal: |
Theodore Ts'o | 9924a92 | 2013-02-08 21:59:22 -0500 | [diff] [blame] | 1018 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1019 | if (IS_ERR(handle)) { |
Theodore Ts'o | 47564bf | 2013-02-09 09:24:14 -0500 | [diff] [blame] | 1020 | page_cache_release(page); |
| 1021 | return PTR_ERR(handle); |
Jan Kara | cf108bc | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 1022 | } |
Tao Ma | f19d587 | 2012-12-10 14:05:51 -0500 | [diff] [blame] | 1023 | |
Theodore Ts'o | 47564bf | 2013-02-09 09:24:14 -0500 | [diff] [blame] | 1024 | lock_page(page); |
| 1025 | if (page->mapping != mapping) { |
| 1026 | /* The page got truncated from under us */ |
| 1027 | unlock_page(page); |
| 1028 | page_cache_release(page); |
Jan Kara | cf108bc | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 1029 | ext4_journal_stop(handle); |
Theodore Ts'o | 47564bf | 2013-02-09 09:24:14 -0500 | [diff] [blame] | 1030 | goto retry_grab; |
Jan Kara | cf108bc | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 1031 | } |
Theodore Ts'o | 47564bf | 2013-02-09 09:24:14 -0500 | [diff] [blame] | 1032 | wait_on_page_writeback(page); |
Jan Kara | cf108bc | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 1033 | |
Jiaying Zhang | 744692d | 2010-03-04 16:14:02 -0500 | [diff] [blame] | 1034 | if (ext4_should_dioread_nolock(inode)) |
Christoph Hellwig | 6e1db88 | 2010-06-04 11:29:57 +0200 | [diff] [blame] | 1035 | ret = __block_write_begin(page, pos, len, ext4_get_block_write); |
Jiaying Zhang | 744692d | 2010-03-04 16:14:02 -0500 | [diff] [blame] | 1036 | else |
Christoph Hellwig | 6e1db88 | 2010-06-04 11:29:57 +0200 | [diff] [blame] | 1037 | ret = __block_write_begin(page, pos, len, ext4_get_block); |
Nick Piggin | bfc1af6 | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1038 | |
| 1039 | if (!ret && ext4_should_journal_data(inode)) { |
Tao Ma | f19d587 | 2012-12-10 14:05:51 -0500 | [diff] [blame] | 1040 | ret = ext4_walk_page_buffers(handle, page_buffers(page), |
| 1041 | from, to, NULL, |
| 1042 | do_journal_get_write_access); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1043 | } |
Nick Piggin | bfc1af6 | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1044 | |
| 1045 | if (ret) { |
Theodore Ts'o | af5bc92 | 2008-09-08 22:25:24 -0400 | [diff] [blame] | 1046 | unlock_page(page); |
Aneesh Kumar K.V | ae4d537 | 2008-09-13 13:10:25 -0400 | [diff] [blame] | 1047 | /* |
Christoph Hellwig | 6e1db88 | 2010-06-04 11:29:57 +0200 | [diff] [blame] | 1048 | * __block_write_begin may have instantiated a few blocks |
Aneesh Kumar K.V | ae4d537 | 2008-09-13 13:10:25 -0400 | [diff] [blame] | 1049 | * outside i_size. Trim these off again. Don't need |
| 1050 | * i_size_read because we hold i_mutex. |
Aneesh Kumar K.V | 1938a15 | 2009-06-05 01:00:26 -0400 | [diff] [blame] | 1051 | * |
| 1052 | * Add inode to orphan list in case we crash before |
| 1053 | * truncate finishes |
Aneesh Kumar K.V | ae4d537 | 2008-09-13 13:10:25 -0400 | [diff] [blame] | 1054 | */ |
Jan Kara | ffacfa7 | 2009-07-13 16:22:22 -0400 | [diff] [blame] | 1055 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
Aneesh Kumar K.V | 1938a15 | 2009-06-05 01:00:26 -0400 | [diff] [blame] | 1056 | ext4_orphan_add(handle, inode); |
| 1057 | |
| 1058 | ext4_journal_stop(handle); |
| 1059 | if (pos + len > inode->i_size) { |
Jan Kara | b9a4207 | 2009-12-08 21:24:33 -0500 | [diff] [blame] | 1060 | ext4_truncate_failed_write(inode); |
Theodore Ts'o | de9a55b | 2009-06-14 17:45:34 -0400 | [diff] [blame] | 1061 | /* |
Jan Kara | ffacfa7 | 2009-07-13 16:22:22 -0400 | [diff] [blame] | 1062 | * If truncate failed early the inode might |
Aneesh Kumar K.V | 1938a15 | 2009-06-05 01:00:26 -0400 | [diff] [blame] | 1063 | * still be on the orphan list; we need to |
| 1064 | * make sure the inode is removed from the |
| 1065 | * orphan list in that case. |
| 1066 | */ |
| 1067 | if (inode->i_nlink) |
| 1068 | ext4_orphan_del(NULL, inode); |
| 1069 | } |
Nick Piggin | bfc1af6 | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1070 | |
Theodore Ts'o | 47564bf | 2013-02-09 09:24:14 -0500 | [diff] [blame] | 1071 | if (ret == -ENOSPC && |
| 1072 | ext4_should_retry_alloc(inode->i_sb, &retries)) |
| 1073 | goto retry_journal; |
| 1074 | page_cache_release(page); |
| 1075 | return ret; |
| 1076 | } |
| 1077 | *pagep = page; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1078 | return ret; |
| 1079 | } |
| 1080 | |
Nick Piggin | bfc1af6 | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1081 | /* For write_end() in data=journal mode */ |
| 1082 | static int write_end_fn(handle_t *handle, struct buffer_head *bh) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1083 | { |
Theodore Ts'o | 13fca32 | 2013-04-21 16:45:54 -0400 | [diff] [blame] | 1084 | int ret; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1085 | if (!buffer_mapped(bh) || buffer_freed(bh)) |
| 1086 | return 0; |
| 1087 | set_buffer_uptodate(bh); |
Theodore Ts'o | 13fca32 | 2013-04-21 16:45:54 -0400 | [diff] [blame] | 1088 | ret = ext4_handle_dirty_metadata(handle, NULL, bh); |
| 1089 | clear_buffer_meta(bh); |
| 1090 | clear_buffer_prio(bh); |
| 1091 | return ret; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1092 | } |
| 1093 | |
Zheng Liu | eed4333 | 2013-04-03 12:41:17 -0400 | [diff] [blame] | 1094 | /* |
| 1095 | * We need to pick up the new inode size which generic_commit_write gave us |
| 1096 | * `file' can be NULL - eg, when called from page_symlink(). |
| 1097 | * |
| 1098 | * ext4 never places buffers on inode->i_mapping->private_list. metadata |
| 1099 | * buffers are managed internally. |
| 1100 | */ |
| 1101 | static int ext4_write_end(struct file *file, |
| 1102 | struct address_space *mapping, |
| 1103 | loff_t pos, unsigned len, unsigned copied, |
| 1104 | struct page *page, void *fsdata) |
Aneesh Kumar K.V | f851408 | 2009-06-05 00:56:49 -0400 | [diff] [blame] | 1105 | { |
Aneesh Kumar K.V | f851408 | 2009-06-05 00:56:49 -0400 | [diff] [blame] | 1106 | handle_t *handle = ext4_journal_current_handle(); |
Zheng Liu | eed4333 | 2013-04-03 12:41:17 -0400 | [diff] [blame] | 1107 | struct inode *inode = mapping->host; |
| 1108 | int ret = 0, ret2; |
| 1109 | int i_size_changed = 0; |
| 1110 | |
| 1111 | trace_ext4_write_end(inode, pos, len, copied); |
| 1112 | if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE)) { |
| 1113 | ret = ext4_jbd2_file_inode(handle, inode); |
| 1114 | if (ret) { |
| 1115 | unlock_page(page); |
| 1116 | page_cache_release(page); |
| 1117 | goto errout; |
| 1118 | } |
| 1119 | } |
Aneesh Kumar K.V | f851408 | 2009-06-05 00:56:49 -0400 | [diff] [blame] | 1120 | |
Tao Ma | f19d587 | 2012-12-10 14:05:51 -0500 | [diff] [blame] | 1121 | if (ext4_has_inline_data(inode)) |
| 1122 | copied = ext4_write_inline_data_end(inode, pos, len, |
| 1123 | copied, page); |
| 1124 | else |
| 1125 | copied = block_write_end(file, mapping, pos, |
| 1126 | len, copied, page, fsdata); |
Aneesh Kumar K.V | f851408 | 2009-06-05 00:56:49 -0400 | [diff] [blame] | 1127 | |
| 1128 | /* |
| 1129 | * No need to use i_size_read() here, the i_size |
Zheng Liu | eed4333 | 2013-04-03 12:41:17 -0400 | [diff] [blame] | 1130 | * cannot change under us because we hole i_mutex. |
Aneesh Kumar K.V | f851408 | 2009-06-05 00:56:49 -0400 | [diff] [blame] | 1131 | * |
| 1132 | * But it's important to update i_size while still holding page lock: |
| 1133 | * page writeout could otherwise come in and zero beyond i_size. |
| 1134 | */ |
| 1135 | if (pos + copied > inode->i_size) { |
| 1136 | i_size_write(inode, pos + copied); |
| 1137 | i_size_changed = 1; |
| 1138 | } |
| 1139 | |
Zheng Liu | eed4333 | 2013-04-03 12:41:17 -0400 | [diff] [blame] | 1140 | if (pos + copied > EXT4_I(inode)->i_disksize) { |
Aneesh Kumar K.V | f851408 | 2009-06-05 00:56:49 -0400 | [diff] [blame] | 1141 | /* We need to mark inode dirty even if |
| 1142 | * new_i_size is less that inode->i_size |
Zheng Liu | eed4333 | 2013-04-03 12:41:17 -0400 | [diff] [blame] | 1143 | * but greater than i_disksize. (hint delalloc) |
Aneesh Kumar K.V | f851408 | 2009-06-05 00:56:49 -0400 | [diff] [blame] | 1144 | */ |
| 1145 | ext4_update_i_disksize(inode, (pos + copied)); |
| 1146 | i_size_changed = 1; |
| 1147 | } |
| 1148 | unlock_page(page); |
| 1149 | page_cache_release(page); |
| 1150 | |
| 1151 | /* |
| 1152 | * Don't mark the inode dirty under page lock. First, it unnecessarily |
| 1153 | * makes the holding time of page lock longer. Second, it forces lock |
| 1154 | * ordering of page lock and transaction start for journaling |
| 1155 | * filesystems. |
| 1156 | */ |
| 1157 | if (i_size_changed) |
| 1158 | ext4_mark_inode_dirty(handle, inode); |
| 1159 | |
Theodore Ts'o | 74d553a | 2013-04-03 12:39:17 -0400 | [diff] [blame] | 1160 | if (copied < 0) |
| 1161 | ret = copied; |
Jan Kara | ffacfa7 | 2009-07-13 16:22:22 -0400 | [diff] [blame] | 1162 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
Aneesh Kumar K.V | f851408 | 2009-06-05 00:56:49 -0400 | [diff] [blame] | 1163 | /* if we have allocated more blocks and copied |
| 1164 | * less. We will have blocks allocated outside |
| 1165 | * inode->i_size. So truncate them |
| 1166 | */ |
| 1167 | ext4_orphan_add(handle, inode); |
Theodore Ts'o | 74d553a | 2013-04-03 12:39:17 -0400 | [diff] [blame] | 1168 | errout: |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1169 | ret2 = ext4_journal_stop(handle); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1170 | if (!ret) |
| 1171 | ret = ret2; |
Nick Piggin | bfc1af6 | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1172 | |
Aneesh Kumar K.V | f851408 | 2009-06-05 00:56:49 -0400 | [diff] [blame] | 1173 | if (pos + len > inode->i_size) { |
Jan Kara | b9a4207 | 2009-12-08 21:24:33 -0500 | [diff] [blame] | 1174 | ext4_truncate_failed_write(inode); |
Theodore Ts'o | de9a55b | 2009-06-14 17:45:34 -0400 | [diff] [blame] | 1175 | /* |
Jan Kara | ffacfa7 | 2009-07-13 16:22:22 -0400 | [diff] [blame] | 1176 | * If truncate failed early the inode might still be |
Aneesh Kumar K.V | f851408 | 2009-06-05 00:56:49 -0400 | [diff] [blame] | 1177 | * on the orphan list; we need to make sure the inode |
| 1178 | * is removed from the orphan list in that case. |
| 1179 | */ |
| 1180 | if (inode->i_nlink) |
| 1181 | ext4_orphan_del(NULL, inode); |
| 1182 | } |
| 1183 | |
Nick Piggin | bfc1af6 | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1184 | return ret ? ret : copied; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1185 | } |
| 1186 | |
Nick Piggin | bfc1af6 | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1187 | static int ext4_journalled_write_end(struct file *file, |
Theodore Ts'o | de9a55b | 2009-06-14 17:45:34 -0400 | [diff] [blame] | 1188 | struct address_space *mapping, |
| 1189 | loff_t pos, unsigned len, unsigned copied, |
| 1190 | struct page *page, void *fsdata) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1191 | { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1192 | handle_t *handle = ext4_journal_current_handle(); |
Nick Piggin | bfc1af6 | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1193 | struct inode *inode = mapping->host; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1194 | int ret = 0, ret2; |
| 1195 | int partial = 0; |
Nick Piggin | bfc1af6 | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1196 | unsigned from, to; |
Aneesh Kumar K.V | cf17fea | 2008-09-13 13:06:18 -0400 | [diff] [blame] | 1197 | loff_t new_i_size; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1198 | |
Theodore Ts'o | 9bffad1 | 2009-06-17 11:48:11 -0400 | [diff] [blame] | 1199 | trace_ext4_journalled_write_end(inode, pos, len, copied); |
Nick Piggin | bfc1af6 | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1200 | from = pos & (PAGE_CACHE_SIZE - 1); |
| 1201 | to = from + len; |
| 1202 | |
Curt Wohlgemuth | 441c850 | 2011-08-13 11:25:18 -0400 | [diff] [blame] | 1203 | BUG_ON(!ext4_handle_valid(handle)); |
| 1204 | |
Tao Ma | 3fdcfb6 | 2012-12-10 14:05:57 -0500 | [diff] [blame] | 1205 | if (ext4_has_inline_data(inode)) |
| 1206 | copied = ext4_write_inline_data_end(inode, pos, len, |
| 1207 | copied, page); |
| 1208 | else { |
| 1209 | if (copied < len) { |
| 1210 | if (!PageUptodate(page)) |
| 1211 | copied = 0; |
| 1212 | page_zero_new_buffers(page, from+copied, to); |
| 1213 | } |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1214 | |
Tao Ma | 3fdcfb6 | 2012-12-10 14:05:57 -0500 | [diff] [blame] | 1215 | ret = ext4_walk_page_buffers(handle, page_buffers(page), from, |
| 1216 | to, &partial, write_end_fn); |
| 1217 | if (!partial) |
| 1218 | SetPageUptodate(page); |
| 1219 | } |
Aneesh Kumar K.V | cf17fea | 2008-09-13 13:06:18 -0400 | [diff] [blame] | 1220 | new_i_size = pos + copied; |
| 1221 | if (new_i_size > inode->i_size) |
Nick Piggin | bfc1af6 | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1222 | i_size_write(inode, pos+copied); |
Theodore Ts'o | 19f5fb7 | 2010-01-24 14:34:07 -0500 | [diff] [blame] | 1223 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
Jan Kara | 2d859db | 2011-07-26 09:07:11 -0400 | [diff] [blame] | 1224 | EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; |
Aneesh Kumar K.V | cf17fea | 2008-09-13 13:06:18 -0400 | [diff] [blame] | 1225 | if (new_i_size > EXT4_I(inode)->i_disksize) { |
| 1226 | ext4_update_i_disksize(inode, new_i_size); |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1227 | ret2 = ext4_mark_inode_dirty(handle, inode); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1228 | if (!ret) |
| 1229 | ret = ret2; |
| 1230 | } |
Nick Piggin | bfc1af6 | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1231 | |
Jan Kara | cf108bc | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 1232 | unlock_page(page); |
Aneesh Kumar K.V | f851408 | 2009-06-05 00:56:49 -0400 | [diff] [blame] | 1233 | page_cache_release(page); |
Jan Kara | ffacfa7 | 2009-07-13 16:22:22 -0400 | [diff] [blame] | 1234 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
Aneesh Kumar K.V | f851408 | 2009-06-05 00:56:49 -0400 | [diff] [blame] | 1235 | /* if we have allocated more blocks and copied |
| 1236 | * less. We will have blocks allocated outside |
| 1237 | * inode->i_size. So truncate them |
| 1238 | */ |
| 1239 | ext4_orphan_add(handle, inode); |
| 1240 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 1241 | ret2 = ext4_journal_stop(handle); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1242 | if (!ret) |
| 1243 | ret = ret2; |
Aneesh Kumar K.V | f851408 | 2009-06-05 00:56:49 -0400 | [diff] [blame] | 1244 | if (pos + len > inode->i_size) { |
Jan Kara | b9a4207 | 2009-12-08 21:24:33 -0500 | [diff] [blame] | 1245 | ext4_truncate_failed_write(inode); |
Theodore Ts'o | de9a55b | 2009-06-14 17:45:34 -0400 | [diff] [blame] | 1246 | /* |
Jan Kara | ffacfa7 | 2009-07-13 16:22:22 -0400 | [diff] [blame] | 1247 | * If truncate failed early the inode might still be |
Aneesh Kumar K.V | f851408 | 2009-06-05 00:56:49 -0400 | [diff] [blame] | 1248 | * on the orphan list; we need to make sure the inode |
| 1249 | * is removed from the orphan list in that case. |
| 1250 | */ |
| 1251 | if (inode->i_nlink) |
| 1252 | ext4_orphan_del(NULL, inode); |
| 1253 | } |
Nick Piggin | bfc1af6 | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1254 | |
| 1255 | return ret ? ret : copied; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1256 | } |
Mingming Cao | d2a1763 | 2008-07-14 17:52:37 -0400 | [diff] [blame] | 1257 | |
Theodore Ts'o | 9d0be50 | 2010-01-01 02:41:30 -0500 | [diff] [blame] | 1258 | /* |
Lukas Czerner | 386ad67 | 2013-03-10 22:50:00 -0400 | [diff] [blame] | 1259 | * Reserve a metadata for a single block located at lblock |
| 1260 | */ |
| 1261 | static int ext4_da_reserve_metadata(struct inode *inode, ext4_lblk_t lblock) |
| 1262 | { |
| 1263 | int retries = 0; |
| 1264 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
| 1265 | struct ext4_inode_info *ei = EXT4_I(inode); |
| 1266 | unsigned int md_needed; |
| 1267 | ext4_lblk_t save_last_lblock; |
| 1268 | int save_len; |
| 1269 | |
| 1270 | /* |
| 1271 | * recalculate the amount of metadata blocks to reserve |
| 1272 | * in order to allocate nrblocks |
| 1273 | * worse case is one extent per block |
| 1274 | */ |
| 1275 | repeat: |
| 1276 | spin_lock(&ei->i_block_reservation_lock); |
| 1277 | /* |
| 1278 | * ext4_calc_metadata_amount() has side effects, which we have |
| 1279 | * to be prepared undo if we fail to claim space. |
| 1280 | */ |
| 1281 | save_len = ei->i_da_metadata_calc_len; |
| 1282 | save_last_lblock = ei->i_da_metadata_calc_last_lblock; |
| 1283 | md_needed = EXT4_NUM_B2C(sbi, |
| 1284 | ext4_calc_metadata_amount(inode, lblock)); |
| 1285 | trace_ext4_da_reserve_space(inode, md_needed); |
| 1286 | |
| 1287 | /* |
| 1288 | * We do still charge estimated metadata to the sb though; |
| 1289 | * we cannot afford to run out of free blocks. |
| 1290 | */ |
| 1291 | if (ext4_claim_free_clusters(sbi, md_needed, 0)) { |
| 1292 | ei->i_da_metadata_calc_len = save_len; |
| 1293 | ei->i_da_metadata_calc_last_lblock = save_last_lblock; |
| 1294 | spin_unlock(&ei->i_block_reservation_lock); |
| 1295 | if (ext4_should_retry_alloc(inode->i_sb, &retries)) { |
| 1296 | cond_resched(); |
| 1297 | goto repeat; |
| 1298 | } |
| 1299 | return -ENOSPC; |
| 1300 | } |
| 1301 | ei->i_reserved_meta_blocks += md_needed; |
| 1302 | spin_unlock(&ei->i_block_reservation_lock); |
| 1303 | |
| 1304 | return 0; /* success */ |
| 1305 | } |
| 1306 | |
| 1307 | /* |
Aditya Kali | 7b415bf | 2011-09-09 19:04:51 -0400 | [diff] [blame] | 1308 | * Reserve a single cluster located at lblock |
Theodore Ts'o | 9d0be50 | 2010-01-01 02:41:30 -0500 | [diff] [blame] | 1309 | */ |
Theodore Ts'o | 01f49d0 | 2011-01-10 12:13:03 -0500 | [diff] [blame] | 1310 | static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock) |
Mingming Cao | d2a1763 | 2008-07-14 17:52:37 -0400 | [diff] [blame] | 1311 | { |
Aneesh Kumar K.V | 030ba6b | 2008-09-08 23:14:50 -0400 | [diff] [blame] | 1312 | int retries = 0; |
Mingming Cao | 60e58e0 | 2009-01-22 18:13:05 +0100 | [diff] [blame] | 1313 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
Theodore Ts'o | 0637c6f | 2009-12-30 14:20:45 -0500 | [diff] [blame] | 1314 | struct ext4_inode_info *ei = EXT4_I(inode); |
Aditya Kali | 7b415bf | 2011-09-09 19:04:51 -0400 | [diff] [blame] | 1315 | unsigned int md_needed; |
Christoph Hellwig | 5dd4056 | 2010-03-03 09:05:00 -0500 | [diff] [blame] | 1316 | int ret; |
Theodore Ts'o | 03179fe | 2012-07-23 00:00:20 -0400 | [diff] [blame] | 1317 | ext4_lblk_t save_last_lblock; |
| 1318 | int save_len; |
Mingming Cao | d2a1763 | 2008-07-14 17:52:37 -0400 | [diff] [blame] | 1319 | |
Mingming Cao | 60e58e0 | 2009-01-22 18:13:05 +0100 | [diff] [blame] | 1320 | /* |
Eric Sandeen | 72b8ab9 | 2010-05-16 11:00:00 -0400 | [diff] [blame] | 1321 | * We will charge metadata quota at writeout time; this saves |
| 1322 | * us from metadata over-estimation, though we may go over by |
| 1323 | * a small amount in the end. Here we just reserve for data. |
Mingming Cao | 60e58e0 | 2009-01-22 18:13:05 +0100 | [diff] [blame] | 1324 | */ |
Aditya Kali | 7b415bf | 2011-09-09 19:04:51 -0400 | [diff] [blame] | 1325 | ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1)); |
Christoph Hellwig | 5dd4056 | 2010-03-03 09:05:00 -0500 | [diff] [blame] | 1326 | if (ret) |
| 1327 | return ret; |
Theodore Ts'o | 03179fe | 2012-07-23 00:00:20 -0400 | [diff] [blame] | 1328 | |
| 1329 | /* |
| 1330 | * recalculate the amount of metadata blocks to reserve |
| 1331 | * in order to allocate nrblocks |
| 1332 | * worse case is one extent per block |
| 1333 | */ |
| 1334 | repeat: |
| 1335 | spin_lock(&ei->i_block_reservation_lock); |
| 1336 | /* |
| 1337 | * ext4_calc_metadata_amount() has side effects, which we have |
| 1338 | * to be prepared undo if we fail to claim space. |
| 1339 | */ |
| 1340 | save_len = ei->i_da_metadata_calc_len; |
| 1341 | save_last_lblock = ei->i_da_metadata_calc_last_lblock; |
| 1342 | md_needed = EXT4_NUM_B2C(sbi, |
| 1343 | ext4_calc_metadata_amount(inode, lblock)); |
| 1344 | trace_ext4_da_reserve_space(inode, md_needed); |
| 1345 | |
Eric Sandeen | 72b8ab9 | 2010-05-16 11:00:00 -0400 | [diff] [blame] | 1346 | /* |
| 1347 | * We do still charge estimated metadata to the sb though; |
| 1348 | * we cannot afford to run out of free blocks. |
| 1349 | */ |
Theodore Ts'o | e7d5f31 | 2011-09-09 19:14:51 -0400 | [diff] [blame] | 1350 | if (ext4_claim_free_clusters(sbi, md_needed + 1, 0)) { |
Theodore Ts'o | 03179fe | 2012-07-23 00:00:20 -0400 | [diff] [blame] | 1351 | ei->i_da_metadata_calc_len = save_len; |
| 1352 | ei->i_da_metadata_calc_last_lblock = save_last_lblock; |
| 1353 | spin_unlock(&ei->i_block_reservation_lock); |
Aneesh Kumar K.V | 030ba6b | 2008-09-08 23:14:50 -0400 | [diff] [blame] | 1354 | if (ext4_should_retry_alloc(inode->i_sb, &retries)) { |
Lukas Czerner | bb8b20e | 2013-03-10 22:28:09 -0400 | [diff] [blame] | 1355 | cond_resched(); |
Aneesh Kumar K.V | 030ba6b | 2008-09-08 23:14:50 -0400 | [diff] [blame] | 1356 | goto repeat; |
| 1357 | } |
Theodore Ts'o | 03179fe | 2012-07-23 00:00:20 -0400 | [diff] [blame] | 1358 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1)); |
Mingming Cao | d2a1763 | 2008-07-14 17:52:37 -0400 | [diff] [blame] | 1359 | return -ENOSPC; |
| 1360 | } |
Theodore Ts'o | 9d0be50 | 2010-01-01 02:41:30 -0500 | [diff] [blame] | 1361 | ei->i_reserved_data_blocks++; |
Theodore Ts'o | 0637c6f | 2009-12-30 14:20:45 -0500 | [diff] [blame] | 1362 | ei->i_reserved_meta_blocks += md_needed; |
| 1363 | spin_unlock(&ei->i_block_reservation_lock); |
Dmitry Monakhov | 39bc680 | 2009-12-10 16:36:27 +0000 | [diff] [blame] | 1364 | |
Mingming Cao | d2a1763 | 2008-07-14 17:52:37 -0400 | [diff] [blame] | 1365 | return 0; /* success */ |
| 1366 | } |
| 1367 | |
Aneesh Kumar K.V | 12219ae | 2008-07-17 16:12:08 -0400 | [diff] [blame] | 1368 | static void ext4_da_release_space(struct inode *inode, int to_free) |
Mingming Cao | d2a1763 | 2008-07-14 17:52:37 -0400 | [diff] [blame] | 1369 | { |
| 1370 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
Theodore Ts'o | 0637c6f | 2009-12-30 14:20:45 -0500 | [diff] [blame] | 1371 | struct ext4_inode_info *ei = EXT4_I(inode); |
Mingming Cao | d2a1763 | 2008-07-14 17:52:37 -0400 | [diff] [blame] | 1372 | |
Mingming Cao | cd21322 | 2008-08-19 22:16:59 -0400 | [diff] [blame] | 1373 | if (!to_free) |
| 1374 | return; /* Nothing to release, exit */ |
| 1375 | |
Mingming Cao | d2a1763 | 2008-07-14 17:52:37 -0400 | [diff] [blame] | 1376 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); |
Mingming Cao | cd21322 | 2008-08-19 22:16:59 -0400 | [diff] [blame] | 1377 | |
Li Zefan | 5a58ec8 | 2010-05-17 02:00:00 -0400 | [diff] [blame] | 1378 | trace_ext4_da_release_space(inode, to_free); |
Theodore Ts'o | 0637c6f | 2009-12-30 14:20:45 -0500 | [diff] [blame] | 1379 | if (unlikely(to_free > ei->i_reserved_data_blocks)) { |
Mingming Cao | cd21322 | 2008-08-19 22:16:59 -0400 | [diff] [blame] | 1380 | /* |
Theodore Ts'o | 0637c6f | 2009-12-30 14:20:45 -0500 | [diff] [blame] | 1381 | * if there aren't enough reserved blocks, then the |
| 1382 | * counter is messed up somewhere. Since this |
| 1383 | * function is called from invalidate page, it's |
| 1384 | * harmless to return without any action. |
Mingming Cao | cd21322 | 2008-08-19 22:16:59 -0400 | [diff] [blame] | 1385 | */ |
Theodore Ts'o | 8de5c32 | 2013-02-14 15:11:41 -0500 | [diff] [blame] | 1386 | ext4_warning(inode->i_sb, "ext4_da_release_space: " |
Theodore Ts'o | 0637c6f | 2009-12-30 14:20:45 -0500 | [diff] [blame] | 1387 | "ino %lu, to_free %d with only %d reserved " |
Theodore Ts'o | 1084f25 | 2012-03-19 23:13:43 -0400 | [diff] [blame] | 1388 | "data blocks", inode->i_ino, to_free, |
Theodore Ts'o | 0637c6f | 2009-12-30 14:20:45 -0500 | [diff] [blame] | 1389 | ei->i_reserved_data_blocks); |
| 1390 | WARN_ON(1); |
| 1391 | to_free = ei->i_reserved_data_blocks; |
| 1392 | } |
| 1393 | ei->i_reserved_data_blocks -= to_free; |
| 1394 | |
| 1395 | if (ei->i_reserved_data_blocks == 0) { |
| 1396 | /* |
| 1397 | * We can release all of the reserved metadata blocks |
| 1398 | * only when we have written all of the delayed |
| 1399 | * allocation blocks. |
Aditya Kali | 7b415bf | 2011-09-09 19:04:51 -0400 | [diff] [blame] | 1400 | * Note that in case of bigalloc, i_reserved_meta_blocks, |
| 1401 | * i_reserved_data_blocks, etc. refer to number of clusters. |
Theodore Ts'o | 0637c6f | 2009-12-30 14:20:45 -0500 | [diff] [blame] | 1402 | */ |
Theodore Ts'o | 5704265 | 2011-09-09 18:56:51 -0400 | [diff] [blame] | 1403 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, |
Eric Sandeen | 72b8ab9 | 2010-05-16 11:00:00 -0400 | [diff] [blame] | 1404 | ei->i_reserved_meta_blocks); |
Theodore Ts'o | ee5f4d9 | 2010-01-01 02:36:15 -0500 | [diff] [blame] | 1405 | ei->i_reserved_meta_blocks = 0; |
Theodore Ts'o | 9d0be50 | 2010-01-01 02:41:30 -0500 | [diff] [blame] | 1406 | ei->i_da_metadata_calc_len = 0; |
Mingming Cao | cd21322 | 2008-08-19 22:16:59 -0400 | [diff] [blame] | 1407 | } |
| 1408 | |
Eric Sandeen | 72b8ab9 | 2010-05-16 11:00:00 -0400 | [diff] [blame] | 1409 | /* update fs dirty data blocks counter */ |
Theodore Ts'o | 5704265 | 2011-09-09 18:56:51 -0400 | [diff] [blame] | 1410 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free); |
Mingming Cao | d2a1763 | 2008-07-14 17:52:37 -0400 | [diff] [blame] | 1411 | |
Mingming Cao | d2a1763 | 2008-07-14 17:52:37 -0400 | [diff] [blame] | 1412 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
Mingming Cao | 60e58e0 | 2009-01-22 18:13:05 +0100 | [diff] [blame] | 1413 | |
Aditya Kali | 7b415bf | 2011-09-09 19:04:51 -0400 | [diff] [blame] | 1414 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free)); |
Mingming Cao | d2a1763 | 2008-07-14 17:52:37 -0400 | [diff] [blame] | 1415 | } |
| 1416 | |
| 1417 | static void ext4_da_page_release_reservation(struct page *page, |
Theodore Ts'o | de9a55b | 2009-06-14 17:45:34 -0400 | [diff] [blame] | 1418 | unsigned long offset) |
Mingming Cao | d2a1763 | 2008-07-14 17:52:37 -0400 | [diff] [blame] | 1419 | { |
| 1420 | int to_release = 0; |
| 1421 | struct buffer_head *head, *bh; |
| 1422 | unsigned int curr_off = 0; |
Aditya Kali | 7b415bf | 2011-09-09 19:04:51 -0400 | [diff] [blame] | 1423 | struct inode *inode = page->mapping->host; |
| 1424 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
| 1425 | int num_clusters; |
Zheng Liu | 51865fd | 2012-11-08 21:57:32 -0500 | [diff] [blame] | 1426 | ext4_fsblk_t lblk; |
Mingming Cao | d2a1763 | 2008-07-14 17:52:37 -0400 | [diff] [blame] | 1427 | |
| 1428 | head = page_buffers(page); |
| 1429 | bh = head; |
| 1430 | do { |
| 1431 | unsigned int next_off = curr_off + bh->b_size; |
| 1432 | |
| 1433 | if ((offset <= curr_off) && (buffer_delay(bh))) { |
| 1434 | to_release++; |
| 1435 | clear_buffer_delay(bh); |
| 1436 | } |
| 1437 | curr_off = next_off; |
| 1438 | } while ((bh = bh->b_this_page) != head); |
Aditya Kali | 7b415bf | 2011-09-09 19:04:51 -0400 | [diff] [blame] | 1439 | |
Zheng Liu | 51865fd | 2012-11-08 21:57:32 -0500 | [diff] [blame] | 1440 | if (to_release) { |
| 1441 | lblk = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits); |
| 1442 | ext4_es_remove_extent(inode, lblk, to_release); |
| 1443 | } |
| 1444 | |
Aditya Kali | 7b415bf | 2011-09-09 19:04:51 -0400 | [diff] [blame] | 1445 | /* If we have released all the blocks belonging to a cluster, then we |
| 1446 | * need to release the reserved space for that cluster. */ |
| 1447 | num_clusters = EXT4_NUM_B2C(sbi, to_release); |
| 1448 | while (num_clusters > 0) { |
Aditya Kali | 7b415bf | 2011-09-09 19:04:51 -0400 | [diff] [blame] | 1449 | lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) + |
| 1450 | ((num_clusters - 1) << sbi->s_cluster_bits); |
| 1451 | if (sbi->s_cluster_ratio == 1 || |
Zheng Liu | 7d1b1fb | 2012-11-08 21:57:35 -0500 | [diff] [blame] | 1452 | !ext4_find_delalloc_cluster(inode, lblk)) |
Aditya Kali | 7b415bf | 2011-09-09 19:04:51 -0400 | [diff] [blame] | 1453 | ext4_da_release_space(inode, 1); |
| 1454 | |
| 1455 | num_clusters--; |
| 1456 | } |
Mingming Cao | d2a1763 | 2008-07-14 17:52:37 -0400 | [diff] [blame] | 1457 | } |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 1458 | |
| 1459 | /* |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 1460 | * Delayed allocation stuff |
| 1461 | */ |
| 1462 | |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 1463 | /* |
| 1464 | * mpage_da_submit_io - walks through extent of pages and try to write |
Aneesh Kumar K.V | a1d6cc5 | 2008-08-19 21:55:02 -0400 | [diff] [blame] | 1465 | * them with writepage() call back |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 1466 | * |
| 1467 | * @mpd->inode: inode |
| 1468 | * @mpd->first_page: first page of the extent |
| 1469 | * @mpd->next_page: page after the last page of the extent |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 1470 | * |
| 1471 | * By the time mpage_da_submit_io() is called we expect all blocks |
| 1472 | * to be allocated. this may be wrong if allocation failed. |
| 1473 | * |
| 1474 | * As pages are already locked by write_cache_pages(), we can't use it |
| 1475 | */ |
Theodore Ts'o | 1de3e3d | 2010-10-27 21:30:10 -0400 | [diff] [blame] | 1476 | static int mpage_da_submit_io(struct mpage_da_data *mpd, |
| 1477 | struct ext4_map_blocks *map) |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 1478 | { |
Aneesh Kumar K.V | 791b7f0 | 2009-01-05 21:50:43 -0500 | [diff] [blame] | 1479 | struct pagevec pvec; |
| 1480 | unsigned long index, end; |
| 1481 | int ret = 0, err, nr_pages, i; |
| 1482 | struct inode *inode = mpd->inode; |
| 1483 | struct address_space *mapping = inode->i_mapping; |
Theodore Ts'o | cb20d51 | 2010-10-27 21:30:09 -0400 | [diff] [blame] | 1484 | loff_t size = i_size_read(inode); |
Theodore Ts'o | 3ecdb3a | 2010-10-27 21:30:10 -0400 | [diff] [blame] | 1485 | unsigned int len, block_start; |
| 1486 | struct buffer_head *bh, *page_bufs = NULL; |
Theodore Ts'o | 1de3e3d | 2010-10-27 21:30:10 -0400 | [diff] [blame] | 1487 | sector_t pblock = 0, cur_logical = 0; |
Theodore Ts'o | bd2d021 | 2010-10-27 21:30:10 -0400 | [diff] [blame] | 1488 | struct ext4_io_submit io_submit; |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 1489 | |
| 1490 | BUG_ON(mpd->next_page <= mpd->first_page); |
Theodore Ts'o | a549984 | 2013-05-11 19:07:42 -0400 | [diff] [blame] | 1491 | memset(&io_submit, 0, sizeof(io_submit)); |
Aneesh Kumar K.V | 791b7f0 | 2009-01-05 21:50:43 -0500 | [diff] [blame] | 1492 | /* |
| 1493 | * We need to start from the first_page to the next_page - 1 |
| 1494 | * to make sure we also write the mapped dirty buffer_heads. |
Theodore Ts'o | 8dc207c | 2009-02-23 06:46:01 -0500 | [diff] [blame] | 1495 | * If we look at mpd->b_blocknr we would only be looking |
Aneesh Kumar K.V | 791b7f0 | 2009-01-05 21:50:43 -0500 | [diff] [blame] | 1496 | * at the currently mapped buffer_heads. |
| 1497 | */ |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 1498 | index = mpd->first_page; |
| 1499 | end = mpd->next_page - 1; |
| 1500 | |
Aneesh Kumar K.V | 791b7f0 | 2009-01-05 21:50:43 -0500 | [diff] [blame] | 1501 | pagevec_init(&pvec, 0); |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 1502 | while (index <= end) { |
Aneesh Kumar K.V | 791b7f0 | 2009-01-05 21:50:43 -0500 | [diff] [blame] | 1503 | nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE); |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 1504 | if (nr_pages == 0) |
| 1505 | break; |
| 1506 | for (i = 0; i < nr_pages; i++) { |
Jan Kara | f8bec37 | 2013-01-28 12:55:08 -0500 | [diff] [blame] | 1507 | int skip_page = 0; |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 1508 | struct page *page = pvec.pages[i]; |
| 1509 | |
Aneesh Kumar K.V | 791b7f0 | 2009-01-05 21:50:43 -0500 | [diff] [blame] | 1510 | index = page->index; |
| 1511 | if (index > end) |
| 1512 | break; |
Theodore Ts'o | cb20d51 | 2010-10-27 21:30:09 -0400 | [diff] [blame] | 1513 | |
| 1514 | if (index == size >> PAGE_CACHE_SHIFT) |
| 1515 | len = size & ~PAGE_CACHE_MASK; |
| 1516 | else |
| 1517 | len = PAGE_CACHE_SIZE; |
Theodore Ts'o | 1de3e3d | 2010-10-27 21:30:10 -0400 | [diff] [blame] | 1518 | if (map) { |
| 1519 | cur_logical = index << (PAGE_CACHE_SHIFT - |
| 1520 | inode->i_blkbits); |
| 1521 | pblock = map->m_pblk + (cur_logical - |
| 1522 | map->m_lblk); |
| 1523 | } |
Aneesh Kumar K.V | 791b7f0 | 2009-01-05 21:50:43 -0500 | [diff] [blame] | 1524 | index++; |
| 1525 | |
| 1526 | BUG_ON(!PageLocked(page)); |
| 1527 | BUG_ON(PageWriteback(page)); |
| 1528 | |
Theodore Ts'o | 3ecdb3a | 2010-10-27 21:30:10 -0400 | [diff] [blame] | 1529 | bh = page_bufs = page_buffers(page); |
| 1530 | block_start = 0; |
| 1531 | do { |
Theodore Ts'o | 1de3e3d | 2010-10-27 21:30:10 -0400 | [diff] [blame] | 1532 | if (map && (cur_logical >= map->m_lblk) && |
| 1533 | (cur_logical <= (map->m_lblk + |
| 1534 | (map->m_len - 1)))) { |
| 1535 | if (buffer_delay(bh)) { |
| 1536 | clear_buffer_delay(bh); |
| 1537 | bh->b_blocknr = pblock; |
| 1538 | } |
| 1539 | if (buffer_unwritten(bh) || |
| 1540 | buffer_mapped(bh)) |
| 1541 | BUG_ON(bh->b_blocknr != pblock); |
| 1542 | if (map->m_flags & EXT4_MAP_UNINIT) |
| 1543 | set_buffer_uninit(bh); |
| 1544 | clear_buffer_unwritten(bh); |
| 1545 | } |
| 1546 | |
Yongqiang Yang | 13a79a4 | 2011-12-13 21:51:55 -0500 | [diff] [blame] | 1547 | /* |
| 1548 | * skip page if block allocation undone and |
| 1549 | * block is dirty |
| 1550 | */ |
| 1551 | if (ext4_bh_delay_or_unwritten(NULL, bh)) |
Theodore Ts'o | 9749895 | 2011-02-26 14:08:01 -0500 | [diff] [blame] | 1552 | skip_page = 1; |
Theodore Ts'o | 3ecdb3a | 2010-10-27 21:30:10 -0400 | [diff] [blame] | 1553 | bh = bh->b_this_page; |
| 1554 | block_start += bh->b_size; |
Theodore Ts'o | 1de3e3d | 2010-10-27 21:30:10 -0400 | [diff] [blame] | 1555 | cur_logical++; |
| 1556 | pblock++; |
| 1557 | } while (bh != page_bufs); |
| 1558 | |
Jan Kara | f8bec37 | 2013-01-28 12:55:08 -0500 | [diff] [blame] | 1559 | if (skip_page) { |
| 1560 | unlock_page(page); |
| 1561 | continue; |
| 1562 | } |
Theodore Ts'o | cb20d51 | 2010-10-27 21:30:09 -0400 | [diff] [blame] | 1563 | |
Theodore Ts'o | 9749895 | 2011-02-26 14:08:01 -0500 | [diff] [blame] | 1564 | clear_page_dirty_for_io(page); |
Jan Kara | fe089c7 | 2013-01-28 09:38:49 -0500 | [diff] [blame] | 1565 | err = ext4_bio_write_page(&io_submit, page, len, |
| 1566 | mpd->wbc); |
Theodore Ts'o | cb20d51 | 2010-10-27 21:30:09 -0400 | [diff] [blame] | 1567 | if (!err) |
Aneesh Kumar K.V | a1d6cc5 | 2008-08-19 21:55:02 -0400 | [diff] [blame] | 1568 | mpd->pages_written++; |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 1569 | /* |
| 1570 | * In error case, we have to continue because |
| 1571 | * remaining pages are still locked |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 1572 | */ |
| 1573 | if (ret == 0) |
| 1574 | ret = err; |
| 1575 | } |
| 1576 | pagevec_release(&pvec); |
| 1577 | } |
Theodore Ts'o | bd2d021 | 2010-10-27 21:30:10 -0400 | [diff] [blame] | 1578 | ext4_io_submit(&io_submit); |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 1579 | return ret; |
| 1580 | } |
| 1581 | |
Curt Wohlgemuth | c7f5938 | 2011-02-26 12:27:52 -0500 | [diff] [blame] | 1582 | static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd) |
Aneesh Kumar K.V | c4a0c46 | 2008-08-19 21:08:18 -0400 | [diff] [blame] | 1583 | { |
| 1584 | int nr_pages, i; |
| 1585 | pgoff_t index, end; |
| 1586 | struct pagevec pvec; |
| 1587 | struct inode *inode = mpd->inode; |
| 1588 | struct address_space *mapping = inode->i_mapping; |
Zheng Liu | 51865fd | 2012-11-08 21:57:32 -0500 | [diff] [blame] | 1589 | ext4_lblk_t start, last; |
Aneesh Kumar K.V | c4a0c46 | 2008-08-19 21:08:18 -0400 | [diff] [blame] | 1590 | |
Curt Wohlgemuth | c7f5938 | 2011-02-26 12:27:52 -0500 | [diff] [blame] | 1591 | index = mpd->first_page; |
| 1592 | end = mpd->next_page - 1; |
Zheng Liu | 51865fd | 2012-11-08 21:57:32 -0500 | [diff] [blame] | 1593 | |
| 1594 | start = index << (PAGE_CACHE_SHIFT - inode->i_blkbits); |
| 1595 | last = end << (PAGE_CACHE_SHIFT - inode->i_blkbits); |
| 1596 | ext4_es_remove_extent(inode, start, last - start + 1); |
| 1597 | |
Eric Sandeen | 66bea92 | 2012-11-14 22:22:05 -0500 | [diff] [blame] | 1598 | pagevec_init(&pvec, 0); |
Aneesh Kumar K.V | c4a0c46 | 2008-08-19 21:08:18 -0400 | [diff] [blame] | 1599 | while (index <= end) { |
| 1600 | nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE); |
| 1601 | if (nr_pages == 0) |
| 1602 | break; |
| 1603 | for (i = 0; i < nr_pages; i++) { |
| 1604 | struct page *page = pvec.pages[i]; |
Jan Kara | 9b1d0998 | 2010-03-03 16:19:32 -0500 | [diff] [blame] | 1605 | if (page->index > end) |
Aneesh Kumar K.V | c4a0c46 | 2008-08-19 21:08:18 -0400 | [diff] [blame] | 1606 | break; |
Aneesh Kumar K.V | c4a0c46 | 2008-08-19 21:08:18 -0400 | [diff] [blame] | 1607 | BUG_ON(!PageLocked(page)); |
| 1608 | BUG_ON(PageWriteback(page)); |
| 1609 | block_invalidatepage(page, 0); |
| 1610 | ClearPageUptodate(page); |
| 1611 | unlock_page(page); |
| 1612 | } |
Jan Kara | 9b1d0998 | 2010-03-03 16:19:32 -0500 | [diff] [blame] | 1613 | index = pvec.pages[nr_pages - 1]->index + 1; |
| 1614 | pagevec_release(&pvec); |
Aneesh Kumar K.V | c4a0c46 | 2008-08-19 21:08:18 -0400 | [diff] [blame] | 1615 | } |
| 1616 | return; |
| 1617 | } |
| 1618 | |
Aneesh Kumar K.V | df22291 | 2008-09-08 23:05:34 -0400 | [diff] [blame] | 1619 | static void ext4_print_free_blocks(struct inode *inode) |
| 1620 | { |
| 1621 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
Theodore Ts'o | 92b9781 | 2012-03-19 23:41:49 -0400 | [diff] [blame] | 1622 | struct super_block *sb = inode->i_sb; |
Lukas Czerner | f78ee70d | 2013-04-03 23:33:30 -0400 | [diff] [blame] | 1623 | struct ext4_inode_info *ei = EXT4_I(inode); |
Theodore Ts'o | 92b9781 | 2012-03-19 23:41:49 -0400 | [diff] [blame] | 1624 | |
| 1625 | ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld", |
Theodore Ts'o | 5dee543 | 2011-09-09 19:10:51 -0400 | [diff] [blame] | 1626 | EXT4_C2B(EXT4_SB(inode->i_sb), |
Lukas Czerner | f78ee70d | 2013-04-03 23:33:30 -0400 | [diff] [blame] | 1627 | ext4_count_free_clusters(sb))); |
Theodore Ts'o | 92b9781 | 2012-03-19 23:41:49 -0400 | [diff] [blame] | 1628 | ext4_msg(sb, KERN_CRIT, "Free/Dirty block details"); |
| 1629 | ext4_msg(sb, KERN_CRIT, "free_blocks=%lld", |
Lukas Czerner | f78ee70d | 2013-04-03 23:33:30 -0400 | [diff] [blame] | 1630 | (long long) EXT4_C2B(EXT4_SB(sb), |
Theodore Ts'o | 5704265 | 2011-09-09 18:56:51 -0400 | [diff] [blame] | 1631 | percpu_counter_sum(&sbi->s_freeclusters_counter))); |
Theodore Ts'o | 92b9781 | 2012-03-19 23:41:49 -0400 | [diff] [blame] | 1632 | ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld", |
Lukas Czerner | f78ee70d | 2013-04-03 23:33:30 -0400 | [diff] [blame] | 1633 | (long long) EXT4_C2B(EXT4_SB(sb), |
Aditya Kali | 7b415bf | 2011-09-09 19:04:51 -0400 | [diff] [blame] | 1634 | percpu_counter_sum(&sbi->s_dirtyclusters_counter))); |
Theodore Ts'o | 92b9781 | 2012-03-19 23:41:49 -0400 | [diff] [blame] | 1635 | ext4_msg(sb, KERN_CRIT, "Block reservation details"); |
| 1636 | ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u", |
Lukas Czerner | f78ee70d | 2013-04-03 23:33:30 -0400 | [diff] [blame] | 1637 | ei->i_reserved_data_blocks); |
Theodore Ts'o | 92b9781 | 2012-03-19 23:41:49 -0400 | [diff] [blame] | 1638 | ext4_msg(sb, KERN_CRIT, "i_reserved_meta_blocks=%u", |
Lukas Czerner | f78ee70d | 2013-04-03 23:33:30 -0400 | [diff] [blame] | 1639 | ei->i_reserved_meta_blocks); |
| 1640 | ext4_msg(sb, KERN_CRIT, "i_allocated_meta_blocks=%u", |
| 1641 | ei->i_allocated_meta_blocks); |
Aneesh Kumar K.V | df22291 | 2008-09-08 23:05:34 -0400 | [diff] [blame] | 1642 | return; |
| 1643 | } |
| 1644 | |
Theodore Ts'o | b920c75 | 2009-05-14 00:54:29 -0400 | [diff] [blame] | 1645 | /* |
Theodore Ts'o | 5a87b7a | 2010-10-27 21:30:09 -0400 | [diff] [blame] | 1646 | * mpage_da_map_and_submit - go through given space, map them |
| 1647 | * if necessary, and then submit them for I/O |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 1648 | * |
Theodore Ts'o | 8dc207c | 2009-02-23 06:46:01 -0500 | [diff] [blame] | 1649 | * @mpd - bh describing space |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 1650 | * |
| 1651 | * The function skips space we know is already mapped to disk blocks. |
| 1652 | * |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 1653 | */ |
Theodore Ts'o | 5a87b7a | 2010-10-27 21:30:09 -0400 | [diff] [blame] | 1654 | static void mpage_da_map_and_submit(struct mpage_da_data *mpd) |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 1655 | { |
Theodore Ts'o | 2ac3b6e | 2009-05-14 13:57:08 -0400 | [diff] [blame] | 1656 | int err, blks, get_blocks_flags; |
Theodore Ts'o | 1de3e3d | 2010-10-27 21:30:10 -0400 | [diff] [blame] | 1657 | struct ext4_map_blocks map, *mapp = NULL; |
Theodore Ts'o | 2fa3cdf | 2009-05-14 09:29:45 -0400 | [diff] [blame] | 1658 | sector_t next = mpd->b_blocknr; |
| 1659 | unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits; |
| 1660 | loff_t disksize = EXT4_I(mpd->inode)->i_disksize; |
| 1661 | handle_t *handle = NULL; |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 1662 | |
| 1663 | /* |
Theodore Ts'o | 5a87b7a | 2010-10-27 21:30:09 -0400 | [diff] [blame] | 1664 | * If the blocks are mapped already, or we couldn't accumulate |
| 1665 | * any blocks, then proceed immediately to the submission stage. |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 1666 | */ |
Theodore Ts'o | 5a87b7a | 2010-10-27 21:30:09 -0400 | [diff] [blame] | 1667 | if ((mpd->b_size == 0) || |
| 1668 | ((mpd->b_state & (1 << BH_Mapped)) && |
| 1669 | !(mpd->b_state & (1 << BH_Delay)) && |
| 1670 | !(mpd->b_state & (1 << BH_Unwritten)))) |
| 1671 | goto submit_io; |
Theodore Ts'o | 2fa3cdf | 2009-05-14 09:29:45 -0400 | [diff] [blame] | 1672 | |
| 1673 | handle = ext4_journal_current_handle(); |
| 1674 | BUG_ON(!handle); |
| 1675 | |
Aneesh Kumar K.V | 79ffab3 | 2009-05-13 15:13:42 -0400 | [diff] [blame] | 1676 | /* |
Eric Sandeen | 79e8303 | 2010-07-27 11:56:07 -0400 | [diff] [blame] | 1677 | * Call ext4_map_blocks() to allocate any delayed allocation |
Theodore Ts'o | 2ac3b6e | 2009-05-14 13:57:08 -0400 | [diff] [blame] | 1678 | * blocks, or to convert an uninitialized extent to be |
| 1679 | * initialized (in the case where we have written into |
| 1680 | * one or more preallocated blocks). |
| 1681 | * |
| 1682 | * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to |
| 1683 | * indicate that we are on the delayed allocation path. This |
| 1684 | * affects functions in many different parts of the allocation |
| 1685 | * call path. This flag exists primarily because we don't |
Eric Sandeen | 79e8303 | 2010-07-27 11:56:07 -0400 | [diff] [blame] | 1686 | * want to change *many* call functions, so ext4_map_blocks() |
Theodore Ts'o | f232109 | 2011-01-10 12:12:36 -0500 | [diff] [blame] | 1687 | * will set the EXT4_STATE_DELALLOC_RESERVED flag once the |
Theodore Ts'o | 2ac3b6e | 2009-05-14 13:57:08 -0400 | [diff] [blame] | 1688 | * inode's allocation semaphore is taken. |
| 1689 | * |
| 1690 | * If the blocks in questions were delalloc blocks, set |
| 1691 | * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting |
| 1692 | * variables are updated after the blocks have been allocated. |
Aneesh Kumar K.V | 79ffab3 | 2009-05-13 15:13:42 -0400 | [diff] [blame] | 1693 | */ |
Theodore Ts'o | 2ed8868 | 2010-05-16 20:00:00 -0400 | [diff] [blame] | 1694 | map.m_lblk = next; |
| 1695 | map.m_len = max_blocks; |
Lukas Czerner | 27dd438 | 2013-04-09 22:11:22 -0400 | [diff] [blame] | 1696 | /* |
| 1697 | * We're in delalloc path and it is possible that we're going to |
| 1698 | * need more metadata blocks than previously reserved. However |
| 1699 | * we must not fail because we're in writeback and there is |
| 1700 | * nothing we can do about it so it might result in data loss. |
| 1701 | * So use reserved blocks to allocate metadata if possible. |
| 1702 | */ |
| 1703 | get_blocks_flags = EXT4_GET_BLOCKS_CREATE | |
| 1704 | EXT4_GET_BLOCKS_METADATA_NOFAIL; |
Jiaying Zhang | 744692d | 2010-03-04 16:14:02 -0500 | [diff] [blame] | 1705 | if (ext4_should_dioread_nolock(mpd->inode)) |
| 1706 | get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT; |
Theodore Ts'o | 2ac3b6e | 2009-05-14 13:57:08 -0400 | [diff] [blame] | 1707 | if (mpd->b_state & (1 << BH_Delay)) |
Aneesh Kumar K.V | 1296cc8 | 2010-01-15 01:27:59 -0500 | [diff] [blame] | 1708 | get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE; |
| 1709 | |
Lukas Czerner | 27dd438 | 2013-04-09 22:11:22 -0400 | [diff] [blame] | 1710 | |
Theodore Ts'o | 2ed8868 | 2010-05-16 20:00:00 -0400 | [diff] [blame] | 1711 | blks = ext4_map_blocks(handle, mpd->inode, &map, get_blocks_flags); |
Theodore Ts'o | 2fa3cdf | 2009-05-14 09:29:45 -0400 | [diff] [blame] | 1712 | if (blks < 0) { |
Eric Sandeen | e357063 | 2010-07-27 11:56:08 -0400 | [diff] [blame] | 1713 | struct super_block *sb = mpd->inode->i_sb; |
| 1714 | |
Theodore Ts'o | 2fa3cdf | 2009-05-14 09:29:45 -0400 | [diff] [blame] | 1715 | err = blks; |
Theodore Ts'o | ed5bde0 | 2009-02-23 10:48:07 -0500 | [diff] [blame] | 1716 | /* |
Theodore Ts'o | 5a87b7a | 2010-10-27 21:30:09 -0400 | [diff] [blame] | 1717 | * If get block returns EAGAIN or ENOSPC and there |
Theodore Ts'o | 9749895 | 2011-02-26 14:08:01 -0500 | [diff] [blame] | 1718 | * appears to be free blocks we will just let |
| 1719 | * mpage_da_submit_io() unlock all of the pages. |
Aneesh Kumar K.V | c4a0c46 | 2008-08-19 21:08:18 -0400 | [diff] [blame] | 1720 | */ |
| 1721 | if (err == -EAGAIN) |
Theodore Ts'o | 5a87b7a | 2010-10-27 21:30:09 -0400 | [diff] [blame] | 1722 | goto submit_io; |
Aneesh Kumar K.V | df22291 | 2008-09-08 23:05:34 -0400 | [diff] [blame] | 1723 | |
Theodore Ts'o | 5dee543 | 2011-09-09 19:10:51 -0400 | [diff] [blame] | 1724 | if (err == -ENOSPC && ext4_count_free_clusters(sb)) { |
Aneesh Kumar K.V | df22291 | 2008-09-08 23:05:34 -0400 | [diff] [blame] | 1725 | mpd->retval = err; |
Theodore Ts'o | 5a87b7a | 2010-10-27 21:30:09 -0400 | [diff] [blame] | 1726 | goto submit_io; |
Aneesh Kumar K.V | df22291 | 2008-09-08 23:05:34 -0400 | [diff] [blame] | 1727 | } |
| 1728 | |
Aneesh Kumar K.V | c4a0c46 | 2008-08-19 21:08:18 -0400 | [diff] [blame] | 1729 | /* |
Theodore Ts'o | ed5bde0 | 2009-02-23 10:48:07 -0500 | [diff] [blame] | 1730 | * get block failure will cause us to loop in |
| 1731 | * writepages, because a_ops->writepage won't be able |
| 1732 | * to make progress. The page will be redirtied by |
| 1733 | * writepage and writepages will again try to write |
| 1734 | * the same. |
Aneesh Kumar K.V | c4a0c46 | 2008-08-19 21:08:18 -0400 | [diff] [blame] | 1735 | */ |
Eric Sandeen | e357063 | 2010-07-27 11:56:08 -0400 | [diff] [blame] | 1736 | if (!(EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)) { |
| 1737 | ext4_msg(sb, KERN_CRIT, |
| 1738 | "delayed block allocation failed for inode %lu " |
| 1739 | "at logical offset %llu with max blocks %zd " |
| 1740 | "with error %d", mpd->inode->i_ino, |
| 1741 | (unsigned long long) next, |
| 1742 | mpd->b_size >> mpd->inode->i_blkbits, err); |
| 1743 | ext4_msg(sb, KERN_CRIT, |
Theodore Ts'o | 01a523e | 2013-02-14 15:51:58 -0500 | [diff] [blame] | 1744 | "This should not happen!! Data will be lost"); |
Eric Sandeen | e357063 | 2010-07-27 11:56:08 -0400 | [diff] [blame] | 1745 | if (err == -ENOSPC) |
| 1746 | ext4_print_free_blocks(mpd->inode); |
Aneesh Kumar K.V | 030ba6b | 2008-09-08 23:14:50 -0400 | [diff] [blame] | 1747 | } |
Theodore Ts'o | 2fa3cdf | 2009-05-14 09:29:45 -0400 | [diff] [blame] | 1748 | /* invalidate all the pages */ |
Curt Wohlgemuth | c7f5938 | 2011-02-26 12:27:52 -0500 | [diff] [blame] | 1749 | ext4_da_block_invalidatepages(mpd); |
Curt Wohlgemuth | e0fd9b9 | 2011-02-26 12:25:52 -0500 | [diff] [blame] | 1750 | |
| 1751 | /* Mark this page range as having been completed */ |
| 1752 | mpd->io_done = 1; |
Theodore Ts'o | 5a87b7a | 2010-10-27 21:30:09 -0400 | [diff] [blame] | 1753 | return; |
Aneesh Kumar K.V | c4a0c46 | 2008-08-19 21:08:18 -0400 | [diff] [blame] | 1754 | } |
Theodore Ts'o | 2fa3cdf | 2009-05-14 09:29:45 -0400 | [diff] [blame] | 1755 | BUG_ON(blks == 0); |
| 1756 | |
Theodore Ts'o | 1de3e3d | 2010-10-27 21:30:10 -0400 | [diff] [blame] | 1757 | mapp = ↦ |
Theodore Ts'o | 2ed8868 | 2010-05-16 20:00:00 -0400 | [diff] [blame] | 1758 | if (map.m_flags & EXT4_MAP_NEW) { |
| 1759 | struct block_device *bdev = mpd->inode->i_sb->s_bdev; |
| 1760 | int i; |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 1761 | |
Theodore Ts'o | 2ed8868 | 2010-05-16 20:00:00 -0400 | [diff] [blame] | 1762 | for (i = 0; i < map.m_len; i++) |
| 1763 | unmap_underlying_metadata(bdev, map.m_pblk + i); |
Theodore Ts'o | 2fa3cdf | 2009-05-14 09:29:45 -0400 | [diff] [blame] | 1764 | } |
| 1765 | |
| 1766 | /* |
Jan Kara | 03f5d8b | 2009-06-09 00:17:05 -0400 | [diff] [blame] | 1767 | * Update on-disk size along with block allocation. |
Theodore Ts'o | 2fa3cdf | 2009-05-14 09:29:45 -0400 | [diff] [blame] | 1768 | */ |
| 1769 | disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits; |
| 1770 | if (disksize > i_size_read(mpd->inode)) |
| 1771 | disksize = i_size_read(mpd->inode); |
| 1772 | if (disksize > EXT4_I(mpd->inode)->i_disksize) { |
| 1773 | ext4_update_i_disksize(mpd->inode, disksize); |
Theodore Ts'o | 5a87b7a | 2010-10-27 21:30:09 -0400 | [diff] [blame] | 1774 | err = ext4_mark_inode_dirty(handle, mpd->inode); |
| 1775 | if (err) |
| 1776 | ext4_error(mpd->inode->i_sb, |
| 1777 | "Failed to mark inode %lu dirty", |
| 1778 | mpd->inode->i_ino); |
Theodore Ts'o | 2fa3cdf | 2009-05-14 09:29:45 -0400 | [diff] [blame] | 1779 | } |
| 1780 | |
Theodore Ts'o | 5a87b7a | 2010-10-27 21:30:09 -0400 | [diff] [blame] | 1781 | submit_io: |
Theodore Ts'o | 1de3e3d | 2010-10-27 21:30:10 -0400 | [diff] [blame] | 1782 | mpage_da_submit_io(mpd, mapp); |
Theodore Ts'o | 5a87b7a | 2010-10-27 21:30:09 -0400 | [diff] [blame] | 1783 | mpd->io_done = 1; |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 1784 | } |
| 1785 | |
Aneesh Kumar K.V | bf068ee | 2008-08-19 22:16:43 -0400 | [diff] [blame] | 1786 | #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \ |
| 1787 | (1 << BH_Delay) | (1 << BH_Unwritten)) |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 1788 | |
| 1789 | /* |
| 1790 | * mpage_add_bh_to_extent - try to add one more block to extent of blocks |
| 1791 | * |
| 1792 | * @mpd->lbh - extent of blocks |
| 1793 | * @logical - logical number of the block in the file |
Jan Kara | b6a8e62 | 2013-01-28 13:06:48 -0500 | [diff] [blame] | 1794 | * @b_state - b_state of the buffer head added |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 1795 | * |
| 1796 | * the function is used to collect contig. blocks in same state |
| 1797 | */ |
Jan Kara | b6a8e62 | 2013-01-28 13:06:48 -0500 | [diff] [blame] | 1798 | static void mpage_add_bh_to_extent(struct mpage_da_data *mpd, sector_t logical, |
Theodore Ts'o | 8dc207c | 2009-02-23 06:46:01 -0500 | [diff] [blame] | 1799 | unsigned long b_state) |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 1800 | { |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 1801 | sector_t next; |
Jan Kara | b6a8e62 | 2013-01-28 13:06:48 -0500 | [diff] [blame] | 1802 | int blkbits = mpd->inode->i_blkbits; |
| 1803 | int nrblocks = mpd->b_size >> blkbits; |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 1804 | |
Eric Sandeen | c445e3e | 2010-05-16 04:00:00 -0400 | [diff] [blame] | 1805 | /* |
| 1806 | * XXX Don't go larger than mballoc is willing to allocate |
| 1807 | * This is a stopgap solution. We eventually need to fold |
| 1808 | * mpage_da_submit_io() into this function and then call |
Eric Sandeen | 79e8303 | 2010-07-27 11:56:07 -0400 | [diff] [blame] | 1809 | * ext4_map_blocks() multiple times in a loop |
Eric Sandeen | c445e3e | 2010-05-16 04:00:00 -0400 | [diff] [blame] | 1810 | */ |
Jan Kara | b6a8e62 | 2013-01-28 13:06:48 -0500 | [diff] [blame] | 1811 | if (nrblocks >= (8*1024*1024 >> blkbits)) |
Eric Sandeen | c445e3e | 2010-05-16 04:00:00 -0400 | [diff] [blame] | 1812 | goto flush_it; |
| 1813 | |
Jan Kara | b6a8e62 | 2013-01-28 13:06:48 -0500 | [diff] [blame] | 1814 | /* check if the reserved journal credits might overflow */ |
| 1815 | if (!ext4_test_inode_flag(mpd->inode, EXT4_INODE_EXTENTS)) { |
Mingming Cao | 525f4ed | 2008-08-19 22:15:58 -0400 | [diff] [blame] | 1816 | if (nrblocks >= EXT4_MAX_TRANS_DATA) { |
| 1817 | /* |
| 1818 | * With non-extent format we are limited by the journal |
| 1819 | * credit available. Total credit needed to insert |
| 1820 | * nrblocks contiguous blocks is dependent on the |
| 1821 | * nrblocks. So limit nrblocks. |
| 1822 | */ |
| 1823 | goto flush_it; |
Mingming Cao | 525f4ed | 2008-08-19 22:15:58 -0400 | [diff] [blame] | 1824 | } |
| 1825 | } |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 1826 | /* |
| 1827 | * First block in the extent |
| 1828 | */ |
Theodore Ts'o | 8dc207c | 2009-02-23 06:46:01 -0500 | [diff] [blame] | 1829 | if (mpd->b_size == 0) { |
| 1830 | mpd->b_blocknr = logical; |
Jan Kara | b6a8e62 | 2013-01-28 13:06:48 -0500 | [diff] [blame] | 1831 | mpd->b_size = 1 << blkbits; |
Theodore Ts'o | 8dc207c | 2009-02-23 06:46:01 -0500 | [diff] [blame] | 1832 | mpd->b_state = b_state & BH_FLAGS; |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 1833 | return; |
| 1834 | } |
| 1835 | |
Theodore Ts'o | 8dc207c | 2009-02-23 06:46:01 -0500 | [diff] [blame] | 1836 | next = mpd->b_blocknr + nrblocks; |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 1837 | /* |
| 1838 | * Can we merge the block to our big extent? |
| 1839 | */ |
Theodore Ts'o | 8dc207c | 2009-02-23 06:46:01 -0500 | [diff] [blame] | 1840 | if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) { |
Jan Kara | b6a8e62 | 2013-01-28 13:06:48 -0500 | [diff] [blame] | 1841 | mpd->b_size += 1 << blkbits; |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 1842 | return; |
| 1843 | } |
| 1844 | |
Mingming Cao | 525f4ed | 2008-08-19 22:15:58 -0400 | [diff] [blame] | 1845 | flush_it: |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 1846 | /* |
| 1847 | * We couldn't merge the block to our extent, so we |
| 1848 | * need to flush current extent and start new one |
| 1849 | */ |
Theodore Ts'o | 5a87b7a | 2010-10-27 21:30:09 -0400 | [diff] [blame] | 1850 | mpage_da_map_and_submit(mpd); |
Aneesh Kumar K.V | a1d6cc5 | 2008-08-19 21:55:02 -0400 | [diff] [blame] | 1851 | return; |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 1852 | } |
| 1853 | |
Aneesh Kumar K.V | c364b22 | 2009-06-14 17:57:10 -0400 | [diff] [blame] | 1854 | static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh) |
Aneesh Kumar K.V | 29fa89d | 2009-05-12 16:30:27 -0400 | [diff] [blame] | 1855 | { |
Aneesh Kumar K.V | c364b22 | 2009-06-14 17:57:10 -0400 | [diff] [blame] | 1856 | return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh); |
Aneesh Kumar K.V | 29fa89d | 2009-05-12 16:30:27 -0400 | [diff] [blame] | 1857 | } |
| 1858 | |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 1859 | /* |
Aditya Kali | 5356f26 | 2011-09-09 19:20:51 -0400 | [diff] [blame] | 1860 | * This function is grabs code from the very beginning of |
| 1861 | * ext4_map_blocks, but assumes that the caller is from delayed write |
| 1862 | * time. This function looks up the requested blocks and sets the |
| 1863 | * buffer delay bit under the protection of i_data_sem. |
| 1864 | */ |
| 1865 | static int ext4_da_map_blocks(struct inode *inode, sector_t iblock, |
| 1866 | struct ext4_map_blocks *map, |
| 1867 | struct buffer_head *bh) |
| 1868 | { |
Zheng Liu | d100eef | 2013-02-18 00:29:59 -0500 | [diff] [blame] | 1869 | struct extent_status es; |
Aditya Kali | 5356f26 | 2011-09-09 19:20:51 -0400 | [diff] [blame] | 1870 | int retval; |
| 1871 | sector_t invalid_block = ~((sector_t) 0xffff); |
Dmitry Monakhov | 921f266 | 2013-03-10 21:01:03 -0400 | [diff] [blame] | 1872 | #ifdef ES_AGGRESSIVE_TEST |
| 1873 | struct ext4_map_blocks orig_map; |
| 1874 | |
| 1875 | memcpy(&orig_map, map, sizeof(*map)); |
| 1876 | #endif |
Aditya Kali | 5356f26 | 2011-09-09 19:20:51 -0400 | [diff] [blame] | 1877 | |
| 1878 | if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es)) |
| 1879 | invalid_block = ~0; |
| 1880 | |
| 1881 | map->m_flags = 0; |
| 1882 | ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u," |
| 1883 | "logical block %lu\n", inode->i_ino, map->m_len, |
| 1884 | (unsigned long) map->m_lblk); |
Zheng Liu | d100eef | 2013-02-18 00:29:59 -0500 | [diff] [blame] | 1885 | |
| 1886 | /* Lookup extent status tree firstly */ |
| 1887 | if (ext4_es_lookup_extent(inode, iblock, &es)) { |
| 1888 | |
| 1889 | if (ext4_es_is_hole(&es)) { |
| 1890 | retval = 0; |
| 1891 | down_read((&EXT4_I(inode)->i_data_sem)); |
| 1892 | goto add_delayed; |
| 1893 | } |
| 1894 | |
| 1895 | /* |
| 1896 | * Delayed extent could be allocated by fallocate. |
| 1897 | * So we need to check it. |
| 1898 | */ |
| 1899 | if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) { |
| 1900 | map_bh(bh, inode->i_sb, invalid_block); |
| 1901 | set_buffer_new(bh); |
| 1902 | set_buffer_delay(bh); |
| 1903 | return 0; |
| 1904 | } |
| 1905 | |
| 1906 | map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk; |
| 1907 | retval = es.es_len - (iblock - es.es_lblk); |
| 1908 | if (retval > map->m_len) |
| 1909 | retval = map->m_len; |
| 1910 | map->m_len = retval; |
| 1911 | if (ext4_es_is_written(&es)) |
| 1912 | map->m_flags |= EXT4_MAP_MAPPED; |
| 1913 | else if (ext4_es_is_unwritten(&es)) |
| 1914 | map->m_flags |= EXT4_MAP_UNWRITTEN; |
| 1915 | else |
| 1916 | BUG_ON(1); |
| 1917 | |
Dmitry Monakhov | 921f266 | 2013-03-10 21:01:03 -0400 | [diff] [blame] | 1918 | #ifdef ES_AGGRESSIVE_TEST |
| 1919 | ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0); |
| 1920 | #endif |
Zheng Liu | d100eef | 2013-02-18 00:29:59 -0500 | [diff] [blame] | 1921 | return retval; |
| 1922 | } |
| 1923 | |
Aditya Kali | 5356f26 | 2011-09-09 19:20:51 -0400 | [diff] [blame] | 1924 | /* |
| 1925 | * Try to see if we can get the block without requesting a new |
| 1926 | * file system block. |
| 1927 | */ |
| 1928 | down_read((&EXT4_I(inode)->i_data_sem)); |
Tao Ma | 9c3569b | 2012-12-10 14:05:57 -0500 | [diff] [blame] | 1929 | if (ext4_has_inline_data(inode)) { |
| 1930 | /* |
| 1931 | * We will soon create blocks for this page, and let |
| 1932 | * us pretend as if the blocks aren't allocated yet. |
| 1933 | * In case of clusters, we have to handle the work |
| 1934 | * of mapping from cluster so that the reserved space |
| 1935 | * is calculated properly. |
| 1936 | */ |
| 1937 | if ((EXT4_SB(inode->i_sb)->s_cluster_ratio > 1) && |
| 1938 | ext4_find_delalloc_cluster(inode, map->m_lblk)) |
| 1939 | map->m_flags |= EXT4_MAP_FROM_CLUSTER; |
| 1940 | retval = 0; |
| 1941 | } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
Zheng Liu | d100eef | 2013-02-18 00:29:59 -0500 | [diff] [blame] | 1942 | retval = ext4_ext_map_blocks(NULL, inode, map, |
| 1943 | EXT4_GET_BLOCKS_NO_PUT_HOLE); |
Aditya Kali | 5356f26 | 2011-09-09 19:20:51 -0400 | [diff] [blame] | 1944 | else |
Zheng Liu | d100eef | 2013-02-18 00:29:59 -0500 | [diff] [blame] | 1945 | retval = ext4_ind_map_blocks(NULL, inode, map, |
| 1946 | EXT4_GET_BLOCKS_NO_PUT_HOLE); |
Aditya Kali | 5356f26 | 2011-09-09 19:20:51 -0400 | [diff] [blame] | 1947 | |
Zheng Liu | d100eef | 2013-02-18 00:29:59 -0500 | [diff] [blame] | 1948 | add_delayed: |
Aditya Kali | 5356f26 | 2011-09-09 19:20:51 -0400 | [diff] [blame] | 1949 | if (retval == 0) { |
Zheng Liu | f7fec03 | 2013-02-18 00:28:47 -0500 | [diff] [blame] | 1950 | int ret; |
Aditya Kali | 5356f26 | 2011-09-09 19:20:51 -0400 | [diff] [blame] | 1951 | /* |
| 1952 | * XXX: __block_prepare_write() unmaps passed block, |
| 1953 | * is it OK? |
| 1954 | */ |
Lukas Czerner | 386ad67 | 2013-03-10 22:50:00 -0400 | [diff] [blame] | 1955 | /* |
| 1956 | * If the block was allocated from previously allocated cluster, |
| 1957 | * then we don't need to reserve it again. However we still need |
| 1958 | * to reserve metadata for every block we're going to write. |
| 1959 | */ |
Aditya Kali | 5356f26 | 2011-09-09 19:20:51 -0400 | [diff] [blame] | 1960 | if (!(map->m_flags & EXT4_MAP_FROM_CLUSTER)) { |
Zheng Liu | f7fec03 | 2013-02-18 00:28:47 -0500 | [diff] [blame] | 1961 | ret = ext4_da_reserve_space(inode, iblock); |
| 1962 | if (ret) { |
Aditya Kali | 5356f26 | 2011-09-09 19:20:51 -0400 | [diff] [blame] | 1963 | /* not enough space to reserve */ |
Zheng Liu | f7fec03 | 2013-02-18 00:28:47 -0500 | [diff] [blame] | 1964 | retval = ret; |
Aditya Kali | 5356f26 | 2011-09-09 19:20:51 -0400 | [diff] [blame] | 1965 | goto out_unlock; |
Zheng Liu | f7fec03 | 2013-02-18 00:28:47 -0500 | [diff] [blame] | 1966 | } |
Lukas Czerner | 386ad67 | 2013-03-10 22:50:00 -0400 | [diff] [blame] | 1967 | } else { |
| 1968 | ret = ext4_da_reserve_metadata(inode, iblock); |
| 1969 | if (ret) { |
| 1970 | /* not enough space to reserve */ |
| 1971 | retval = ret; |
| 1972 | goto out_unlock; |
| 1973 | } |
Aditya Kali | 5356f26 | 2011-09-09 19:20:51 -0400 | [diff] [blame] | 1974 | } |
| 1975 | |
Zheng Liu | f7fec03 | 2013-02-18 00:28:47 -0500 | [diff] [blame] | 1976 | ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, |
| 1977 | ~0, EXTENT_STATUS_DELAYED); |
| 1978 | if (ret) { |
| 1979 | retval = ret; |
Zheng Liu | 51865fd | 2012-11-08 21:57:32 -0500 | [diff] [blame] | 1980 | goto out_unlock; |
Zheng Liu | f7fec03 | 2013-02-18 00:28:47 -0500 | [diff] [blame] | 1981 | } |
Zheng Liu | 51865fd | 2012-11-08 21:57:32 -0500 | [diff] [blame] | 1982 | |
Aditya Kali | 5356f26 | 2011-09-09 19:20:51 -0400 | [diff] [blame] | 1983 | /* Clear EXT4_MAP_FROM_CLUSTER flag since its purpose is served |
| 1984 | * and it should not appear on the bh->b_state. |
| 1985 | */ |
| 1986 | map->m_flags &= ~EXT4_MAP_FROM_CLUSTER; |
| 1987 | |
| 1988 | map_bh(bh, inode->i_sb, invalid_block); |
| 1989 | set_buffer_new(bh); |
| 1990 | set_buffer_delay(bh); |
Zheng Liu | f7fec03 | 2013-02-18 00:28:47 -0500 | [diff] [blame] | 1991 | } else if (retval > 0) { |
| 1992 | int ret; |
| 1993 | unsigned long long status; |
| 1994 | |
Dmitry Monakhov | 921f266 | 2013-03-10 21:01:03 -0400 | [diff] [blame] | 1995 | #ifdef ES_AGGRESSIVE_TEST |
| 1996 | if (retval != map->m_len) { |
| 1997 | printk("ES len assertation failed for inode: %lu " |
| 1998 | "retval %d != map->m_len %d " |
| 1999 | "in %s (lookup)\n", inode->i_ino, retval, |
| 2000 | map->m_len, __func__); |
| 2001 | } |
| 2002 | #endif |
| 2003 | |
Zheng Liu | f7fec03 | 2013-02-18 00:28:47 -0500 | [diff] [blame] | 2004 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
| 2005 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; |
| 2006 | ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, |
| 2007 | map->m_pblk, status); |
| 2008 | if (ret != 0) |
| 2009 | retval = ret; |
Aditya Kali | 5356f26 | 2011-09-09 19:20:51 -0400 | [diff] [blame] | 2010 | } |
| 2011 | |
| 2012 | out_unlock: |
| 2013 | up_read((&EXT4_I(inode)->i_data_sem)); |
| 2014 | |
| 2015 | return retval; |
| 2016 | } |
| 2017 | |
| 2018 | /* |
Theodore Ts'o | b920c75 | 2009-05-14 00:54:29 -0400 | [diff] [blame] | 2019 | * This is a special get_blocks_t callback which is used by |
| 2020 | * ext4_da_write_begin(). It will either return mapped block or |
| 2021 | * reserve space for a single block. |
Aneesh Kumar K.V | 29fa89d | 2009-05-12 16:30:27 -0400 | [diff] [blame] | 2022 | * |
| 2023 | * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set. |
| 2024 | * We also have b_blocknr = -1 and b_bdev initialized properly |
| 2025 | * |
| 2026 | * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set. |
| 2027 | * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev |
| 2028 | * initialized properly. |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2029 | */ |
Tao Ma | 9c3569b | 2012-12-10 14:05:57 -0500 | [diff] [blame] | 2030 | int ext4_da_get_block_prep(struct inode *inode, sector_t iblock, |
| 2031 | struct buffer_head *bh, int create) |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2032 | { |
Theodore Ts'o | 2ed8868 | 2010-05-16 20:00:00 -0400 | [diff] [blame] | 2033 | struct ext4_map_blocks map; |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2034 | int ret = 0; |
| 2035 | |
| 2036 | BUG_ON(create == 0); |
Theodore Ts'o | 2ed8868 | 2010-05-16 20:00:00 -0400 | [diff] [blame] | 2037 | BUG_ON(bh->b_size != inode->i_sb->s_blocksize); |
| 2038 | |
| 2039 | map.m_lblk = iblock; |
| 2040 | map.m_len = 1; |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2041 | |
| 2042 | /* |
| 2043 | * first, we need to know whether the block is allocated already |
| 2044 | * preallocated blocks are unmapped but should treated |
| 2045 | * the same as allocated blocks. |
| 2046 | */ |
Aditya Kali | 5356f26 | 2011-09-09 19:20:51 -0400 | [diff] [blame] | 2047 | ret = ext4_da_map_blocks(inode, iblock, &map, bh); |
| 2048 | if (ret <= 0) |
Theodore Ts'o | 2ed8868 | 2010-05-16 20:00:00 -0400 | [diff] [blame] | 2049 | return ret; |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2050 | |
Theodore Ts'o | 2ed8868 | 2010-05-16 20:00:00 -0400 | [diff] [blame] | 2051 | map_bh(bh, inode->i_sb, map.m_pblk); |
| 2052 | bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags; |
| 2053 | |
| 2054 | if (buffer_unwritten(bh)) { |
| 2055 | /* A delayed write to unwritten bh should be marked |
| 2056 | * new and mapped. Mapped ensures that we don't do |
| 2057 | * get_block multiple times when we write to the same |
| 2058 | * offset and new ensures that we do proper zero out |
| 2059 | * for partial write. |
| 2060 | */ |
| 2061 | set_buffer_new(bh); |
Theodore Ts'o | c820563 | 2011-04-10 22:30:07 -0400 | [diff] [blame] | 2062 | set_buffer_mapped(bh); |
Theodore Ts'o | 2ed8868 | 2010-05-16 20:00:00 -0400 | [diff] [blame] | 2063 | } |
| 2064 | return 0; |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2065 | } |
Mingming Cao | 61628a3 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2066 | |
Aneesh Kumar K.V | 62e086b | 2009-06-14 17:59:34 -0400 | [diff] [blame] | 2067 | static int bget_one(handle_t *handle, struct buffer_head *bh) |
| 2068 | { |
| 2069 | get_bh(bh); |
| 2070 | return 0; |
| 2071 | } |
| 2072 | |
| 2073 | static int bput_one(handle_t *handle, struct buffer_head *bh) |
| 2074 | { |
| 2075 | put_bh(bh); |
| 2076 | return 0; |
| 2077 | } |
| 2078 | |
| 2079 | static int __ext4_journalled_writepage(struct page *page, |
Aneesh Kumar K.V | 62e086b | 2009-06-14 17:59:34 -0400 | [diff] [blame] | 2080 | unsigned int len) |
| 2081 | { |
| 2082 | struct address_space *mapping = page->mapping; |
| 2083 | struct inode *inode = mapping->host; |
Tao Ma | 3fdcfb6 | 2012-12-10 14:05:57 -0500 | [diff] [blame] | 2084 | struct buffer_head *page_bufs = NULL; |
Aneesh Kumar K.V | 62e086b | 2009-06-14 17:59:34 -0400 | [diff] [blame] | 2085 | handle_t *handle = NULL; |
Tao Ma | 3fdcfb6 | 2012-12-10 14:05:57 -0500 | [diff] [blame] | 2086 | int ret = 0, err = 0; |
| 2087 | int inline_data = ext4_has_inline_data(inode); |
| 2088 | struct buffer_head *inode_bh = NULL; |
Aneesh Kumar K.V | 62e086b | 2009-06-14 17:59:34 -0400 | [diff] [blame] | 2089 | |
Theodore Ts'o | cb20d51 | 2010-10-27 21:30:09 -0400 | [diff] [blame] | 2090 | ClearPageChecked(page); |
Tao Ma | 3fdcfb6 | 2012-12-10 14:05:57 -0500 | [diff] [blame] | 2091 | |
| 2092 | if (inline_data) { |
| 2093 | BUG_ON(page->index != 0); |
| 2094 | BUG_ON(len > ext4_get_max_inline_size(inode)); |
| 2095 | inode_bh = ext4_journalled_write_inline_data(inode, len, page); |
| 2096 | if (inode_bh == NULL) |
| 2097 | goto out; |
| 2098 | } else { |
| 2099 | page_bufs = page_buffers(page); |
| 2100 | if (!page_bufs) { |
| 2101 | BUG(); |
| 2102 | goto out; |
| 2103 | } |
| 2104 | ext4_walk_page_buffers(handle, page_bufs, 0, len, |
| 2105 | NULL, bget_one); |
| 2106 | } |
Aneesh Kumar K.V | 62e086b | 2009-06-14 17:59:34 -0400 | [diff] [blame] | 2107 | /* As soon as we unlock the page, it can go away, but we have |
| 2108 | * references to buffers so we are safe */ |
| 2109 | unlock_page(page); |
| 2110 | |
Theodore Ts'o | 9924a92 | 2013-02-08 21:59:22 -0500 | [diff] [blame] | 2111 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
| 2112 | ext4_writepage_trans_blocks(inode)); |
Aneesh Kumar K.V | 62e086b | 2009-06-14 17:59:34 -0400 | [diff] [blame] | 2113 | if (IS_ERR(handle)) { |
| 2114 | ret = PTR_ERR(handle); |
| 2115 | goto out; |
| 2116 | } |
| 2117 | |
Curt Wohlgemuth | 441c850 | 2011-08-13 11:25:18 -0400 | [diff] [blame] | 2118 | BUG_ON(!ext4_handle_valid(handle)); |
| 2119 | |
Tao Ma | 3fdcfb6 | 2012-12-10 14:05:57 -0500 | [diff] [blame] | 2120 | if (inline_data) { |
| 2121 | ret = ext4_journal_get_write_access(handle, inode_bh); |
Aneesh Kumar K.V | 62e086b | 2009-06-14 17:59:34 -0400 | [diff] [blame] | 2122 | |
Tao Ma | 3fdcfb6 | 2012-12-10 14:05:57 -0500 | [diff] [blame] | 2123 | err = ext4_handle_dirty_metadata(handle, inode, inode_bh); |
| 2124 | |
| 2125 | } else { |
| 2126 | ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL, |
| 2127 | do_journal_get_write_access); |
| 2128 | |
| 2129 | err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL, |
| 2130 | write_end_fn); |
| 2131 | } |
Aneesh Kumar K.V | 62e086b | 2009-06-14 17:59:34 -0400 | [diff] [blame] | 2132 | if (ret == 0) |
| 2133 | ret = err; |
Jan Kara | 2d859db | 2011-07-26 09:07:11 -0400 | [diff] [blame] | 2134 | EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; |
Aneesh Kumar K.V | 62e086b | 2009-06-14 17:59:34 -0400 | [diff] [blame] | 2135 | err = ext4_journal_stop(handle); |
| 2136 | if (!ret) |
| 2137 | ret = err; |
| 2138 | |
Tao Ma | 3fdcfb6 | 2012-12-10 14:05:57 -0500 | [diff] [blame] | 2139 | if (!ext4_has_inline_data(inode)) |
| 2140 | ext4_walk_page_buffers(handle, page_bufs, 0, len, |
| 2141 | NULL, bput_one); |
Theodore Ts'o | 19f5fb7 | 2010-01-24 14:34:07 -0500 | [diff] [blame] | 2142 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
Aneesh Kumar K.V | 62e086b | 2009-06-14 17:59:34 -0400 | [diff] [blame] | 2143 | out: |
Tao Ma | 3fdcfb6 | 2012-12-10 14:05:57 -0500 | [diff] [blame] | 2144 | brelse(inode_bh); |
Aneesh Kumar K.V | 62e086b | 2009-06-14 17:59:34 -0400 | [diff] [blame] | 2145 | return ret; |
| 2146 | } |
| 2147 | |
Mingming Cao | 61628a3 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2148 | /* |
Aneesh Kumar K.V | 43ce1d2 | 2009-06-14 17:58:45 -0400 | [diff] [blame] | 2149 | * Note that we don't need to start a transaction unless we're journaling data |
| 2150 | * because we should have holes filled from ext4_page_mkwrite(). We even don't |
| 2151 | * need to file the inode to the transaction's list in ordered mode because if |
| 2152 | * we are writing back data added by write(), the inode is already there and if |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 2153 | * we are writing back data modified via mmap(), no one guarantees in which |
Aneesh Kumar K.V | 43ce1d2 | 2009-06-14 17:58:45 -0400 | [diff] [blame] | 2154 | * transaction the data will hit the disk. In case we are journaling data, we |
| 2155 | * cannot start transaction directly because transaction start ranks above page |
| 2156 | * lock so we have to do some magic. |
| 2157 | * |
Theodore Ts'o | b920c75 | 2009-05-14 00:54:29 -0400 | [diff] [blame] | 2158 | * This function can get called via... |
| 2159 | * - ext4_da_writepages after taking page lock (have journal handle) |
| 2160 | * - journal_submit_inode_data_buffers (no journal handle) |
Artem Bityutskiy | f6463b0 | 2012-07-25 18:12:04 +0300 | [diff] [blame] | 2161 | * - shrink_page_list via the kswapd/direct reclaim (no journal handle) |
Theodore Ts'o | b920c75 | 2009-05-14 00:54:29 -0400 | [diff] [blame] | 2162 | * - grab_page_cache when doing write_begin (have journal handle) |
Aneesh Kumar K.V | 43ce1d2 | 2009-06-14 17:58:45 -0400 | [diff] [blame] | 2163 | * |
| 2164 | * We don't do any block allocation in this function. If we have page with |
| 2165 | * multiple blocks we need to write those buffer_heads that are mapped. This |
| 2166 | * is important for mmaped based write. So if we do with blocksize 1K |
| 2167 | * truncate(f, 1024); |
| 2168 | * a = mmap(f, 0, 4096); |
| 2169 | * a[0] = 'a'; |
| 2170 | * truncate(f, 4096); |
| 2171 | * we have in the page first buffer_head mapped via page_mkwrite call back |
Paul Bolle | 90802ed | 2011-12-05 13:00:34 +0100 | [diff] [blame] | 2172 | * but other buffer_heads would be unmapped but dirty (dirty done via the |
Aneesh Kumar K.V | 43ce1d2 | 2009-06-14 17:58:45 -0400 | [diff] [blame] | 2173 | * do_wp_page). So writepage should write the first block. If we modify |
| 2174 | * the mmap area beyond 1024 we will again get a page_fault and the |
| 2175 | * page_mkwrite callback will do the block allocation and mark the |
| 2176 | * buffer_heads mapped. |
| 2177 | * |
| 2178 | * We redirty the page if we have any buffer_heads that is either delay or |
| 2179 | * unwritten in the page. |
| 2180 | * |
| 2181 | * We can get recursively called as show below. |
| 2182 | * |
| 2183 | * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() -> |
| 2184 | * ext4_writepage() |
| 2185 | * |
| 2186 | * But since we don't do any block allocation we should not deadlock. |
| 2187 | * Page also have the dirty flag cleared so we don't get recurive page_lock. |
Mingming Cao | 61628a3 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2188 | */ |
Aneesh Kumar K.V | 43ce1d2 | 2009-06-14 17:58:45 -0400 | [diff] [blame] | 2189 | static int ext4_writepage(struct page *page, |
Aneesh Kumar K.V | 62e086b | 2009-06-14 17:59:34 -0400 | [diff] [blame] | 2190 | struct writeback_control *wbc) |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2191 | { |
Jan Kara | f8bec37 | 2013-01-28 12:55:08 -0500 | [diff] [blame] | 2192 | int ret = 0; |
Mingming Cao | 61628a3 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2193 | loff_t size; |
Theodore Ts'o | 498e5f2 | 2008-11-05 00:14:04 -0500 | [diff] [blame] | 2194 | unsigned int len; |
Jiaying Zhang | 744692d | 2010-03-04 16:14:02 -0500 | [diff] [blame] | 2195 | struct buffer_head *page_bufs = NULL; |
Mingming Cao | 61628a3 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2196 | struct inode *inode = page->mapping->host; |
Jan Kara | 36ade45 | 2013-01-28 09:30:52 -0500 | [diff] [blame] | 2197 | struct ext4_io_submit io_submit; |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2198 | |
Lukas Czerner | a9c667f | 2011-06-06 09:51:52 -0400 | [diff] [blame] | 2199 | trace_ext4_writepage(page); |
Aneesh Kumar K.V | f0e6c98 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2200 | size = i_size_read(inode); |
| 2201 | if (page->index == size >> PAGE_CACHE_SHIFT) |
| 2202 | len = size & ~PAGE_CACHE_MASK; |
| 2203 | else |
| 2204 | len = PAGE_CACHE_SIZE; |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2205 | |
Theodore Ts'o | a42afc5 | 2010-10-27 21:30:09 -0400 | [diff] [blame] | 2206 | page_bufs = page_buffers(page); |
Aneesh Kumar K.V | c364b22 | 2009-06-14 17:57:10 -0400 | [diff] [blame] | 2207 | /* |
Jan Kara | fe38613 | 2013-01-28 21:06:42 -0500 | [diff] [blame] | 2208 | * We cannot do block allocation or other extent handling in this |
| 2209 | * function. If there are buffers needing that, we have to redirty |
| 2210 | * the page. But we may reach here when we do a journal commit via |
| 2211 | * journal_submit_inode_data_buffers() and in that case we must write |
| 2212 | * allocated buffers to achieve data=ordered mode guarantees. |
Aneesh Kumar K.V | cd1aac3 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2213 | */ |
Tao Ma | f19d587 | 2012-12-10 14:05:51 -0500 | [diff] [blame] | 2214 | if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL, |
| 2215 | ext4_bh_delay_or_unwritten)) { |
Jan Kara | f8bec37 | 2013-01-28 12:55:08 -0500 | [diff] [blame] | 2216 | redirty_page_for_writepage(wbc, page); |
Jan Kara | fe38613 | 2013-01-28 21:06:42 -0500 | [diff] [blame] | 2217 | if (current->flags & PF_MEMALLOC) { |
| 2218 | /* |
| 2219 | * For memory cleaning there's no point in writing only |
| 2220 | * some buffers. So just bail out. Warn if we came here |
| 2221 | * from direct reclaim. |
| 2222 | */ |
| 2223 | WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) |
| 2224 | == PF_MEMALLOC); |
Aneesh Kumar K.V | f0e6c98 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2225 | unlock_page(page); |
| 2226 | return 0; |
| 2227 | } |
Theodore Ts'o | a42afc5 | 2010-10-27 21:30:09 -0400 | [diff] [blame] | 2228 | } |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2229 | |
Theodore Ts'o | cb20d51 | 2010-10-27 21:30:09 -0400 | [diff] [blame] | 2230 | if (PageChecked(page) && ext4_should_journal_data(inode)) |
Aneesh Kumar K.V | 43ce1d2 | 2009-06-14 17:58:45 -0400 | [diff] [blame] | 2231 | /* |
| 2232 | * It's mmapped pagecache. Add buffers and journal it. There |
| 2233 | * doesn't seem much point in redirtying the page here. |
| 2234 | */ |
Wu Fengguang | 3f0ca30 | 2009-11-24 11:15:44 -0500 | [diff] [blame] | 2235 | return __ext4_journalled_writepage(page, len); |
Aneesh Kumar K.V | 43ce1d2 | 2009-06-14 17:58:45 -0400 | [diff] [blame] | 2236 | |
Theodore Ts'o | a549984 | 2013-05-11 19:07:42 -0400 | [diff] [blame] | 2237 | memset(&io_submit, 0, sizeof(io_submit)); |
Jan Kara | 36ade45 | 2013-01-28 09:30:52 -0500 | [diff] [blame] | 2238 | ret = ext4_bio_write_page(&io_submit, page, len, wbc); |
| 2239 | ext4_io_submit(&io_submit); |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2240 | return ret; |
| 2241 | } |
| 2242 | |
Mingming Cao | 61628a3 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2243 | /* |
Mingming Cao | 525f4ed | 2008-08-19 22:15:58 -0400 | [diff] [blame] | 2244 | * This is called via ext4_da_writepages() to |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 2245 | * calculate the total number of credits to reserve to fit |
Mingming Cao | 525f4ed | 2008-08-19 22:15:58 -0400 | [diff] [blame] | 2246 | * a single extent allocation into a single transaction, |
| 2247 | * ext4_da_writpeages() will loop calling this before |
| 2248 | * the block allocation. |
Mingming Cao | 61628a3 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2249 | */ |
Mingming Cao | 525f4ed | 2008-08-19 22:15:58 -0400 | [diff] [blame] | 2250 | |
| 2251 | static int ext4_da_writepages_trans_blocks(struct inode *inode) |
| 2252 | { |
| 2253 | int max_blocks = EXT4_I(inode)->i_reserved_data_blocks; |
| 2254 | |
| 2255 | /* |
| 2256 | * With non-extent format the journal credit needed to |
| 2257 | * insert nrblocks contiguous block is dependent on |
| 2258 | * number of contiguous block. So we will limit |
| 2259 | * number of contiguous block to a sane value |
| 2260 | */ |
Dmitry Monakhov | 12e9b89 | 2010-05-16 22:00:00 -0400 | [diff] [blame] | 2261 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) && |
Mingming Cao | 525f4ed | 2008-08-19 22:15:58 -0400 | [diff] [blame] | 2262 | (max_blocks > EXT4_MAX_TRANS_DATA)) |
| 2263 | max_blocks = EXT4_MAX_TRANS_DATA; |
| 2264 | |
| 2265 | return ext4_chunk_trans_blocks(inode, max_blocks); |
| 2266 | } |
Mingming Cao | 61628a3 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2267 | |
Theodore Ts'o | 8e48dcf | 2010-05-16 18:00:00 -0400 | [diff] [blame] | 2268 | /* |
| 2269 | * write_cache_pages_da - walk the list of dirty pages of the given |
Theodore Ts'o | 8eb9e5c | 2011-02-26 14:07:31 -0500 | [diff] [blame] | 2270 | * address space and accumulate pages that need writing, and call |
Theodore Ts'o | 168fc02 | 2011-02-26 14:09:20 -0500 | [diff] [blame] | 2271 | * mpage_da_map_and_submit to map a single contiguous memory region |
| 2272 | * and then write them. |
Theodore Ts'o | 8e48dcf | 2010-05-16 18:00:00 -0400 | [diff] [blame] | 2273 | */ |
Tao Ma | 9c3569b | 2012-12-10 14:05:57 -0500 | [diff] [blame] | 2274 | static int write_cache_pages_da(handle_t *handle, |
| 2275 | struct address_space *mapping, |
Theodore Ts'o | 8e48dcf | 2010-05-16 18:00:00 -0400 | [diff] [blame] | 2276 | struct writeback_control *wbc, |
Eric Sandeen | 72f84e6 | 2010-10-27 21:30:13 -0400 | [diff] [blame] | 2277 | struct mpage_da_data *mpd, |
| 2278 | pgoff_t *done_index) |
Theodore Ts'o | 8e48dcf | 2010-05-16 18:00:00 -0400 | [diff] [blame] | 2279 | { |
Theodore Ts'o | 4f01b02 | 2011-02-26 14:07:37 -0500 | [diff] [blame] | 2280 | struct buffer_head *bh, *head; |
Theodore Ts'o | 168fc02 | 2011-02-26 14:09:20 -0500 | [diff] [blame] | 2281 | struct inode *inode = mapping->host; |
Theodore Ts'o | 4f01b02 | 2011-02-26 14:07:37 -0500 | [diff] [blame] | 2282 | struct pagevec pvec; |
| 2283 | unsigned int nr_pages; |
| 2284 | sector_t logical; |
| 2285 | pgoff_t index, end; |
| 2286 | long nr_to_write = wbc->nr_to_write; |
| 2287 | int i, tag, ret = 0; |
Theodore Ts'o | 8e48dcf | 2010-05-16 18:00:00 -0400 | [diff] [blame] | 2288 | |
Theodore Ts'o | 168fc02 | 2011-02-26 14:09:20 -0500 | [diff] [blame] | 2289 | memset(mpd, 0, sizeof(struct mpage_da_data)); |
| 2290 | mpd->wbc = wbc; |
| 2291 | mpd->inode = inode; |
Theodore Ts'o | 8e48dcf | 2010-05-16 18:00:00 -0400 | [diff] [blame] | 2292 | pagevec_init(&pvec, 0); |
| 2293 | index = wbc->range_start >> PAGE_CACHE_SHIFT; |
| 2294 | end = wbc->range_end >> PAGE_CACHE_SHIFT; |
| 2295 | |
Wu Fengguang | 6e6938b | 2010-06-06 10:38:15 -0600 | [diff] [blame] | 2296 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
Eric Sandeen | 5b41d92 | 2010-10-27 21:30:13 -0400 | [diff] [blame] | 2297 | tag = PAGECACHE_TAG_TOWRITE; |
| 2298 | else |
| 2299 | tag = PAGECACHE_TAG_DIRTY; |
| 2300 | |
Eric Sandeen | 72f84e6 | 2010-10-27 21:30:13 -0400 | [diff] [blame] | 2301 | *done_index = index; |
Theodore Ts'o | 4f01b02 | 2011-02-26 14:07:37 -0500 | [diff] [blame] | 2302 | while (index <= end) { |
Eric Sandeen | 5b41d92 | 2010-10-27 21:30:13 -0400 | [diff] [blame] | 2303 | nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag, |
Theodore Ts'o | 8e48dcf | 2010-05-16 18:00:00 -0400 | [diff] [blame] | 2304 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1); |
| 2305 | if (nr_pages == 0) |
Theodore Ts'o | 4f01b02 | 2011-02-26 14:07:37 -0500 | [diff] [blame] | 2306 | return 0; |
Theodore Ts'o | 8e48dcf | 2010-05-16 18:00:00 -0400 | [diff] [blame] | 2307 | |
| 2308 | for (i = 0; i < nr_pages; i++) { |
| 2309 | struct page *page = pvec.pages[i]; |
| 2310 | |
| 2311 | /* |
| 2312 | * At this point, the page may be truncated or |
| 2313 | * invalidated (changing page->mapping to NULL), or |
| 2314 | * even swizzled back from swapper_space to tmpfs file |
| 2315 | * mapping. However, page->index will not change |
| 2316 | * because we have a reference on the page. |
| 2317 | */ |
Theodore Ts'o | 4f01b02 | 2011-02-26 14:07:37 -0500 | [diff] [blame] | 2318 | if (page->index > end) |
| 2319 | goto out; |
Theodore Ts'o | 8e48dcf | 2010-05-16 18:00:00 -0400 | [diff] [blame] | 2320 | |
Eric Sandeen | 72f84e6 | 2010-10-27 21:30:13 -0400 | [diff] [blame] | 2321 | *done_index = page->index + 1; |
| 2322 | |
Theodore Ts'o | 78aaced | 2011-02-26 14:09:14 -0500 | [diff] [blame] | 2323 | /* |
| 2324 | * If we can't merge this page, and we have |
| 2325 | * accumulated an contiguous region, write it |
| 2326 | */ |
| 2327 | if ((mpd->next_page != page->index) && |
| 2328 | (mpd->next_page != mpd->first_page)) { |
| 2329 | mpage_da_map_and_submit(mpd); |
| 2330 | goto ret_extent_tail; |
| 2331 | } |
| 2332 | |
Theodore Ts'o | 8e48dcf | 2010-05-16 18:00:00 -0400 | [diff] [blame] | 2333 | lock_page(page); |
| 2334 | |
| 2335 | /* |
Theodore Ts'o | 4f01b02 | 2011-02-26 14:07:37 -0500 | [diff] [blame] | 2336 | * If the page is no longer dirty, or its |
| 2337 | * mapping no longer corresponds to inode we |
| 2338 | * are writing (which means it has been |
| 2339 | * truncated or invalidated), or the page is |
| 2340 | * already under writeback and we are not |
| 2341 | * doing a data integrity writeback, skip the page |
Theodore Ts'o | 8e48dcf | 2010-05-16 18:00:00 -0400 | [diff] [blame] | 2342 | */ |
Theodore Ts'o | 4f01b02 | 2011-02-26 14:07:37 -0500 | [diff] [blame] | 2343 | if (!PageDirty(page) || |
| 2344 | (PageWriteback(page) && |
| 2345 | (wbc->sync_mode == WB_SYNC_NONE)) || |
| 2346 | unlikely(page->mapping != mapping)) { |
Theodore Ts'o | 8e48dcf | 2010-05-16 18:00:00 -0400 | [diff] [blame] | 2347 | unlock_page(page); |
| 2348 | continue; |
| 2349 | } |
| 2350 | |
Darrick J. Wong | 7cb1a53 | 2011-05-18 13:53:20 -0400 | [diff] [blame] | 2351 | wait_on_page_writeback(page); |
Theodore Ts'o | 8e48dcf | 2010-05-16 18:00:00 -0400 | [diff] [blame] | 2352 | BUG_ON(PageWriteback(page)); |
Theodore Ts'o | 8e48dcf | 2010-05-16 18:00:00 -0400 | [diff] [blame] | 2353 | |
Tao Ma | 9c3569b | 2012-12-10 14:05:57 -0500 | [diff] [blame] | 2354 | /* |
| 2355 | * If we have inline data and arrive here, it means that |
| 2356 | * we will soon create the block for the 1st page, so |
| 2357 | * we'd better clear the inline data here. |
| 2358 | */ |
| 2359 | if (ext4_has_inline_data(inode)) { |
| 2360 | BUG_ON(ext4_test_inode_state(inode, |
| 2361 | EXT4_STATE_MAY_INLINE_DATA)); |
| 2362 | ext4_destroy_inline_data(handle, inode); |
| 2363 | } |
| 2364 | |
Theodore Ts'o | 168fc02 | 2011-02-26 14:09:20 -0500 | [diff] [blame] | 2365 | if (mpd->next_page != page->index) |
Theodore Ts'o | 8eb9e5c | 2011-02-26 14:07:31 -0500 | [diff] [blame] | 2366 | mpd->first_page = page->index; |
Theodore Ts'o | 8eb9e5c | 2011-02-26 14:07:31 -0500 | [diff] [blame] | 2367 | mpd->next_page = page->index + 1; |
| 2368 | logical = (sector_t) page->index << |
| 2369 | (PAGE_CACHE_SHIFT - inode->i_blkbits); |
| 2370 | |
Jan Kara | f8bec37 | 2013-01-28 12:55:08 -0500 | [diff] [blame] | 2371 | /* Add all dirty buffers to mpd */ |
| 2372 | head = page_buffers(page); |
| 2373 | bh = head; |
| 2374 | do { |
| 2375 | BUG_ON(buffer_locked(bh)); |
Theodore Ts'o | 8eb9e5c | 2011-02-26 14:07:31 -0500 | [diff] [blame] | 2376 | /* |
Jan Kara | f8bec37 | 2013-01-28 12:55:08 -0500 | [diff] [blame] | 2377 | * We need to try to allocate unmapped blocks |
| 2378 | * in the same page. Otherwise we won't make |
| 2379 | * progress with the page in ext4_writepage |
Theodore Ts'o | 8eb9e5c | 2011-02-26 14:07:31 -0500 | [diff] [blame] | 2380 | */ |
Jan Kara | f8bec37 | 2013-01-28 12:55:08 -0500 | [diff] [blame] | 2381 | if (ext4_bh_delay_or_unwritten(NULL, bh)) { |
| 2382 | mpage_add_bh_to_extent(mpd, logical, |
Jan Kara | f8bec37 | 2013-01-28 12:55:08 -0500 | [diff] [blame] | 2383 | bh->b_state); |
| 2384 | if (mpd->io_done) |
| 2385 | goto ret_extent_tail; |
| 2386 | } else if (buffer_dirty(bh) && |
| 2387 | buffer_mapped(bh)) { |
Theodore Ts'o | 8eb9e5c | 2011-02-26 14:07:31 -0500 | [diff] [blame] | 2388 | /* |
Jan Kara | f8bec37 | 2013-01-28 12:55:08 -0500 | [diff] [blame] | 2389 | * mapped dirty buffer. We need to |
| 2390 | * update the b_state because we look |
| 2391 | * at b_state in mpage_da_map_blocks. |
| 2392 | * We don't update b_size because if we |
| 2393 | * find an unmapped buffer_head later |
| 2394 | * we need to use the b_state flag of |
| 2395 | * that buffer_head. |
Theodore Ts'o | 8eb9e5c | 2011-02-26 14:07:31 -0500 | [diff] [blame] | 2396 | */ |
Jan Kara | f8bec37 | 2013-01-28 12:55:08 -0500 | [diff] [blame] | 2397 | if (mpd->b_size == 0) |
| 2398 | mpd->b_state = |
| 2399 | bh->b_state & BH_FLAGS; |
| 2400 | } |
| 2401 | logical++; |
| 2402 | } while ((bh = bh->b_this_page) != head); |
Theodore Ts'o | 8e48dcf | 2010-05-16 18:00:00 -0400 | [diff] [blame] | 2403 | |
| 2404 | if (nr_to_write > 0) { |
| 2405 | nr_to_write--; |
| 2406 | if (nr_to_write == 0 && |
Theodore Ts'o | 4f01b02 | 2011-02-26 14:07:37 -0500 | [diff] [blame] | 2407 | wbc->sync_mode == WB_SYNC_NONE) |
Theodore Ts'o | 8e48dcf | 2010-05-16 18:00:00 -0400 | [diff] [blame] | 2408 | /* |
| 2409 | * We stop writing back only if we are |
| 2410 | * not doing integrity sync. In case of |
| 2411 | * integrity sync we have to keep going |
| 2412 | * because someone may be concurrently |
| 2413 | * dirtying pages, and we might have |
| 2414 | * synced a lot of newly appeared dirty |
| 2415 | * pages, but have not synced all of the |
| 2416 | * old dirty pages. |
| 2417 | */ |
Theodore Ts'o | 4f01b02 | 2011-02-26 14:07:37 -0500 | [diff] [blame] | 2418 | goto out; |
Theodore Ts'o | 8e48dcf | 2010-05-16 18:00:00 -0400 | [diff] [blame] | 2419 | } |
| 2420 | } |
| 2421 | pagevec_release(&pvec); |
| 2422 | cond_resched(); |
| 2423 | } |
Theodore Ts'o | 4f01b02 | 2011-02-26 14:07:37 -0500 | [diff] [blame] | 2424 | return 0; |
| 2425 | ret_extent_tail: |
| 2426 | ret = MPAGE_DA_EXTENT_TAIL; |
Theodore Ts'o | 8eb9e5c | 2011-02-26 14:07:31 -0500 | [diff] [blame] | 2427 | out: |
| 2428 | pagevec_release(&pvec); |
| 2429 | cond_resched(); |
Theodore Ts'o | 8e48dcf | 2010-05-16 18:00:00 -0400 | [diff] [blame] | 2430 | return ret; |
| 2431 | } |
| 2432 | |
| 2433 | |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2434 | static int ext4_da_writepages(struct address_space *mapping, |
Aneesh Kumar K.V | a1d6cc5 | 2008-08-19 21:55:02 -0400 | [diff] [blame] | 2435 | struct writeback_control *wbc) |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2436 | { |
Aneesh Kumar K.V | 22208de | 2008-10-16 10:10:36 -0400 | [diff] [blame] | 2437 | pgoff_t index; |
| 2438 | int range_whole = 0; |
Mingming Cao | 61628a3 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2439 | handle_t *handle = NULL; |
Aneesh Kumar K.V | df22291 | 2008-09-08 23:05:34 -0400 | [diff] [blame] | 2440 | struct mpage_da_data mpd; |
Aneesh Kumar K.V | 5e745b0 | 2008-08-18 18:00:57 -0400 | [diff] [blame] | 2441 | struct inode *inode = mapping->host; |
Theodore Ts'o | 498e5f2 | 2008-11-05 00:14:04 -0500 | [diff] [blame] | 2442 | int pages_written = 0; |
Theodore Ts'o | 55138e0 | 2009-09-29 13:31:31 -0400 | [diff] [blame] | 2443 | unsigned int max_pages; |
Aneesh Kumar K.V | 2acf2c2 | 2009-02-14 10:42:58 -0500 | [diff] [blame] | 2444 | int range_cyclic, cycled = 1, io_done = 0; |
Theodore Ts'o | 55138e0 | 2009-09-29 13:31:31 -0400 | [diff] [blame] | 2445 | int needed_blocks, ret = 0; |
| 2446 | long desired_nr_to_write, nr_to_writebump = 0; |
Theodore Ts'o | de89de6 | 2009-08-31 17:00:59 -0400 | [diff] [blame] | 2447 | loff_t range_start = wbc->range_start; |
Aneesh Kumar K.V | 5e745b0 | 2008-08-18 18:00:57 -0400 | [diff] [blame] | 2448 | struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb); |
Eric Sandeen | 72f84e6 | 2010-10-27 21:30:13 -0400 | [diff] [blame] | 2449 | pgoff_t done_index = 0; |
Eric Sandeen | 5b41d92 | 2010-10-27 21:30:13 -0400 | [diff] [blame] | 2450 | pgoff_t end; |
Shaohua Li | 1bce63d | 2011-10-18 10:55:51 -0400 | [diff] [blame] | 2451 | struct blk_plug plug; |
Mingming Cao | 61628a3 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2452 | |
Theodore Ts'o | 9bffad1 | 2009-06-17 11:48:11 -0400 | [diff] [blame] | 2453 | trace_ext4_da_writepages(inode, wbc); |
Theodore Ts'o | ba80b10 | 2009-01-03 20:03:21 -0500 | [diff] [blame] | 2454 | |
Mingming Cao | 61628a3 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2455 | /* |
| 2456 | * No pages to write? This is mainly a kludge to avoid starting |
| 2457 | * a transaction for special inodes like journal inode on last iput() |
| 2458 | * because that could violate lock ordering on umount |
| 2459 | */ |
Aneesh Kumar K.V | a1d6cc5 | 2008-08-19 21:55:02 -0400 | [diff] [blame] | 2460 | if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) |
Mingming Cao | 61628a3 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2461 | return 0; |
Theodore Ts'o | 2a21e37 | 2008-11-05 09:22:24 -0500 | [diff] [blame] | 2462 | |
| 2463 | /* |
| 2464 | * If the filesystem has aborted, it is read-only, so return |
| 2465 | * right away instead of dumping stack traces later on that |
| 2466 | * will obscure the real source of the problem. We test |
Theodore Ts'o | 4ab2f15 | 2009-06-13 10:09:36 -0400 | [diff] [blame] | 2467 | * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because |
Theodore Ts'o | 2a21e37 | 2008-11-05 09:22:24 -0500 | [diff] [blame] | 2468 | * the latter could be true if the filesystem is mounted |
| 2469 | * read-only, and in that case, ext4_da_writepages should |
| 2470 | * *never* be called, so if that ever happens, we would want |
| 2471 | * the stack trace. |
| 2472 | */ |
Theodore Ts'o | 4ab2f15 | 2009-06-13 10:09:36 -0400 | [diff] [blame] | 2473 | if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) |
Theodore Ts'o | 2a21e37 | 2008-11-05 09:22:24 -0500 | [diff] [blame] | 2474 | return -EROFS; |
| 2475 | |
Aneesh Kumar K.V | 22208de | 2008-10-16 10:10:36 -0400 | [diff] [blame] | 2476 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
| 2477 | range_whole = 1; |
Mingming Cao | 61628a3 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2478 | |
Aneesh Kumar K.V | 2acf2c2 | 2009-02-14 10:42:58 -0500 | [diff] [blame] | 2479 | range_cyclic = wbc->range_cyclic; |
| 2480 | if (wbc->range_cyclic) { |
Aneesh Kumar K.V | 22208de | 2008-10-16 10:10:36 -0400 | [diff] [blame] | 2481 | index = mapping->writeback_index; |
Aneesh Kumar K.V | 2acf2c2 | 2009-02-14 10:42:58 -0500 | [diff] [blame] | 2482 | if (index) |
| 2483 | cycled = 0; |
| 2484 | wbc->range_start = index << PAGE_CACHE_SHIFT; |
| 2485 | wbc->range_end = LLONG_MAX; |
| 2486 | wbc->range_cyclic = 0; |
Eric Sandeen | 5b41d92 | 2010-10-27 21:30:13 -0400 | [diff] [blame] | 2487 | end = -1; |
| 2488 | } else { |
Aneesh Kumar K.V | 22208de | 2008-10-16 10:10:36 -0400 | [diff] [blame] | 2489 | index = wbc->range_start >> PAGE_CACHE_SHIFT; |
Eric Sandeen | 5b41d92 | 2010-10-27 21:30:13 -0400 | [diff] [blame] | 2490 | end = wbc->range_end >> PAGE_CACHE_SHIFT; |
| 2491 | } |
Aneesh Kumar K.V | a1d6cc5 | 2008-08-19 21:55:02 -0400 | [diff] [blame] | 2492 | |
Theodore Ts'o | 55138e0 | 2009-09-29 13:31:31 -0400 | [diff] [blame] | 2493 | /* |
| 2494 | * This works around two forms of stupidity. The first is in |
| 2495 | * the writeback code, which caps the maximum number of pages |
| 2496 | * written to be 1024 pages. This is wrong on multiple |
| 2497 | * levels; different architectues have a different page size, |
| 2498 | * which changes the maximum amount of data which gets |
| 2499 | * written. Secondly, 4 megabytes is way too small. XFS |
| 2500 | * forces this value to be 16 megabytes by multiplying |
| 2501 | * nr_to_write parameter by four, and then relies on its |
| 2502 | * allocator to allocate larger extents to make them |
| 2503 | * contiguous. Unfortunately this brings us to the second |
| 2504 | * stupidity, which is that ext4's mballoc code only allocates |
| 2505 | * at most 2048 blocks. So we force contiguous writes up to |
| 2506 | * the number of dirty blocks in the inode, or |
| 2507 | * sbi->max_writeback_mb_bump whichever is smaller. |
| 2508 | */ |
| 2509 | max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT); |
Eric Sandeen | b443e73 | 2010-10-27 21:30:03 -0400 | [diff] [blame] | 2510 | if (!range_cyclic && range_whole) { |
| 2511 | if (wbc->nr_to_write == LONG_MAX) |
| 2512 | desired_nr_to_write = wbc->nr_to_write; |
| 2513 | else |
| 2514 | desired_nr_to_write = wbc->nr_to_write * 8; |
| 2515 | } else |
Theodore Ts'o | 55138e0 | 2009-09-29 13:31:31 -0400 | [diff] [blame] | 2516 | desired_nr_to_write = ext4_num_dirty_pages(inode, index, |
| 2517 | max_pages); |
| 2518 | if (desired_nr_to_write > max_pages) |
| 2519 | desired_nr_to_write = max_pages; |
| 2520 | |
| 2521 | if (wbc->nr_to_write < desired_nr_to_write) { |
| 2522 | nr_to_writebump = desired_nr_to_write - wbc->nr_to_write; |
| 2523 | wbc->nr_to_write = desired_nr_to_write; |
| 2524 | } |
| 2525 | |
Aneesh Kumar K.V | 2acf2c2 | 2009-02-14 10:42:58 -0500 | [diff] [blame] | 2526 | retry: |
Wu Fengguang | 6e6938b | 2010-06-06 10:38:15 -0600 | [diff] [blame] | 2527 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
Eric Sandeen | 5b41d92 | 2010-10-27 21:30:13 -0400 | [diff] [blame] | 2528 | tag_pages_for_writeback(mapping, index, end); |
| 2529 | |
Shaohua Li | 1bce63d | 2011-10-18 10:55:51 -0400 | [diff] [blame] | 2530 | blk_start_plug(&plug); |
Aneesh Kumar K.V | 22208de | 2008-10-16 10:10:36 -0400 | [diff] [blame] | 2531 | while (!ret && wbc->nr_to_write > 0) { |
Aneesh Kumar K.V | a1d6cc5 | 2008-08-19 21:55:02 -0400 | [diff] [blame] | 2532 | |
| 2533 | /* |
| 2534 | * we insert one extent at a time. So we need |
| 2535 | * credit needed for single extent allocation. |
| 2536 | * journalled mode is currently not supported |
| 2537 | * by delalloc |
| 2538 | */ |
| 2539 | BUG_ON(ext4_should_journal_data(inode)); |
Mingming Cao | 525f4ed | 2008-08-19 22:15:58 -0400 | [diff] [blame] | 2540 | needed_blocks = ext4_da_writepages_trans_blocks(inode); |
Aneesh Kumar K.V | a1d6cc5 | 2008-08-19 21:55:02 -0400 | [diff] [blame] | 2541 | |
Mingming Cao | 61628a3 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2542 | /* start a new transaction*/ |
Theodore Ts'o | 9924a92 | 2013-02-08 21:59:22 -0500 | [diff] [blame] | 2543 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
| 2544 | needed_blocks); |
Mingming Cao | 61628a3 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2545 | if (IS_ERR(handle)) { |
| 2546 | ret = PTR_ERR(handle); |
Theodore Ts'o | 1693918 | 2009-09-26 17:43:59 -0400 | [diff] [blame] | 2547 | ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: " |
Curt Wohlgemuth | fbe845d | 2010-05-16 13:00:00 -0400 | [diff] [blame] | 2548 | "%ld pages, ino %lu; err %d", __func__, |
Aneesh Kumar K.V | a1d6cc5 | 2008-08-19 21:55:02 -0400 | [diff] [blame] | 2549 | wbc->nr_to_write, inode->i_ino, ret); |
Namjae Jeon | 3c1fcb2 | 2011-11-07 11:01:13 -0500 | [diff] [blame] | 2550 | blk_finish_plug(&plug); |
Mingming Cao | 61628a3 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2551 | goto out_writepages; |
| 2552 | } |
Theodore Ts'o | f63e600 | 2009-02-23 16:42:39 -0500 | [diff] [blame] | 2553 | |
| 2554 | /* |
Theodore Ts'o | 8eb9e5c | 2011-02-26 14:07:31 -0500 | [diff] [blame] | 2555 | * Now call write_cache_pages_da() to find the next |
Theodore Ts'o | f63e600 | 2009-02-23 16:42:39 -0500 | [diff] [blame] | 2556 | * contiguous region of logical blocks that need |
Theodore Ts'o | 8eb9e5c | 2011-02-26 14:07:31 -0500 | [diff] [blame] | 2557 | * blocks to be allocated by ext4 and submit them. |
Theodore Ts'o | f63e600 | 2009-02-23 16:42:39 -0500 | [diff] [blame] | 2558 | */ |
Tao Ma | 9c3569b | 2012-12-10 14:05:57 -0500 | [diff] [blame] | 2559 | ret = write_cache_pages_da(handle, mapping, |
| 2560 | wbc, &mpd, &done_index); |
Theodore Ts'o | f63e600 | 2009-02-23 16:42:39 -0500 | [diff] [blame] | 2561 | /* |
André Goddard Rosa | af901ca | 2009-11-14 13:09:05 -0200 | [diff] [blame] | 2562 | * If we have a contiguous extent of pages and we |
Theodore Ts'o | f63e600 | 2009-02-23 16:42:39 -0500 | [diff] [blame] | 2563 | * haven't done the I/O yet, map the blocks and submit |
| 2564 | * them for I/O. |
| 2565 | */ |
| 2566 | if (!mpd.io_done && mpd.next_page != mpd.first_page) { |
Theodore Ts'o | 5a87b7a | 2010-10-27 21:30:09 -0400 | [diff] [blame] | 2567 | mpage_da_map_and_submit(&mpd); |
Theodore Ts'o | f63e600 | 2009-02-23 16:42:39 -0500 | [diff] [blame] | 2568 | ret = MPAGE_DA_EXTENT_TAIL; |
| 2569 | } |
Theodore Ts'o | b3a3ca8 | 2009-08-31 23:13:11 -0400 | [diff] [blame] | 2570 | trace_ext4_da_write_pages(inode, &mpd); |
Theodore Ts'o | f63e600 | 2009-02-23 16:42:39 -0500 | [diff] [blame] | 2571 | wbc->nr_to_write -= mpd.pages_written; |
Aneesh Kumar K.V | df22291 | 2008-09-08 23:05:34 -0400 | [diff] [blame] | 2572 | |
Mingming Cao | 61628a3 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2573 | ext4_journal_stop(handle); |
Aneesh Kumar K.V | df22291 | 2008-09-08 23:05:34 -0400 | [diff] [blame] | 2574 | |
Eric Sandeen | 8f64b32 | 2009-02-26 00:57:35 -0500 | [diff] [blame] | 2575 | if ((mpd.retval == -ENOSPC) && sbi->s_journal) { |
Aneesh Kumar K.V | 22208de | 2008-10-16 10:10:36 -0400 | [diff] [blame] | 2576 | /* commit the transaction which would |
| 2577 | * free blocks released in the transaction |
| 2578 | * and try again |
| 2579 | */ |
Aneesh Kumar K.V | df22291 | 2008-09-08 23:05:34 -0400 | [diff] [blame] | 2580 | jbd2_journal_force_commit_nested(sbi->s_journal); |
Aneesh Kumar K.V | 22208de | 2008-10-16 10:10:36 -0400 | [diff] [blame] | 2581 | ret = 0; |
| 2582 | } else if (ret == MPAGE_DA_EXTENT_TAIL) { |
Aneesh Kumar K.V | a1d6cc5 | 2008-08-19 21:55:02 -0400 | [diff] [blame] | 2583 | /* |
Kazuya Mio | 8de49e6 | 2011-10-20 19:23:08 -0400 | [diff] [blame] | 2584 | * Got one extent now try with rest of the pages. |
| 2585 | * If mpd.retval is set -EIO, journal is aborted. |
| 2586 | * So we don't need to write any more. |
Aneesh Kumar K.V | a1d6cc5 | 2008-08-19 21:55:02 -0400 | [diff] [blame] | 2587 | */ |
Aneesh Kumar K.V | 22208de | 2008-10-16 10:10:36 -0400 | [diff] [blame] | 2588 | pages_written += mpd.pages_written; |
Kazuya Mio | 8de49e6 | 2011-10-20 19:23:08 -0400 | [diff] [blame] | 2589 | ret = mpd.retval; |
Aneesh Kumar K.V | 2acf2c2 | 2009-02-14 10:42:58 -0500 | [diff] [blame] | 2590 | io_done = 1; |
Aneesh Kumar K.V | 22208de | 2008-10-16 10:10:36 -0400 | [diff] [blame] | 2591 | } else if (wbc->nr_to_write) |
Mingming Cao | 61628a3 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2592 | /* |
| 2593 | * There is no more writeout needed |
| 2594 | * or we requested for a noblocking writeout |
| 2595 | * and we found the device congested |
| 2596 | */ |
Mingming Cao | 61628a3 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2597 | break; |
Mingming Cao | 61628a3 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2598 | } |
Shaohua Li | 1bce63d | 2011-10-18 10:55:51 -0400 | [diff] [blame] | 2599 | blk_finish_plug(&plug); |
Aneesh Kumar K.V | 2acf2c2 | 2009-02-14 10:42:58 -0500 | [diff] [blame] | 2600 | if (!io_done && !cycled) { |
| 2601 | cycled = 1; |
| 2602 | index = 0; |
| 2603 | wbc->range_start = index << PAGE_CACHE_SHIFT; |
| 2604 | wbc->range_end = mapping->writeback_index - 1; |
| 2605 | goto retry; |
| 2606 | } |
Mingming Cao | 61628a3 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2607 | |
Aneesh Kumar K.V | 22208de | 2008-10-16 10:10:36 -0400 | [diff] [blame] | 2608 | /* Update index */ |
Aneesh Kumar K.V | 2acf2c2 | 2009-02-14 10:42:58 -0500 | [diff] [blame] | 2609 | wbc->range_cyclic = range_cyclic; |
Aneesh Kumar K.V | 22208de | 2008-10-16 10:10:36 -0400 | [diff] [blame] | 2610 | if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) |
| 2611 | /* |
| 2612 | * set the writeback_index so that range_cyclic |
| 2613 | * mode will write it back later |
| 2614 | */ |
Eric Sandeen | 72f84e6 | 2010-10-27 21:30:13 -0400 | [diff] [blame] | 2615 | mapping->writeback_index = done_index; |
Aneesh Kumar K.V | a1d6cc5 | 2008-08-19 21:55:02 -0400 | [diff] [blame] | 2616 | |
Mingming Cao | 61628a3 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2617 | out_writepages: |
Richard Kennedy | 2faf2e1 | 2009-12-25 15:46:07 -0500 | [diff] [blame] | 2618 | wbc->nr_to_write -= nr_to_writebump; |
Theodore Ts'o | de89de6 | 2009-08-31 17:00:59 -0400 | [diff] [blame] | 2619 | wbc->range_start = range_start; |
Theodore Ts'o | 9bffad1 | 2009-06-17 11:48:11 -0400 | [diff] [blame] | 2620 | trace_ext4_da_writepages_result(inode, wbc, ret, pages_written); |
Mingming Cao | 61628a3 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2621 | return ret; |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2622 | } |
| 2623 | |
Aneesh Kumar K.V | 79f0be8 | 2008-10-08 23:13:30 -0400 | [diff] [blame] | 2624 | static int ext4_nonda_switch(struct super_block *sb) |
| 2625 | { |
Eric Whitney | 5c1ff33 | 2013-04-09 09:27:31 -0400 | [diff] [blame] | 2626 | s64 free_clusters, dirty_clusters; |
Aneesh Kumar K.V | 79f0be8 | 2008-10-08 23:13:30 -0400 | [diff] [blame] | 2627 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
| 2628 | |
| 2629 | /* |
| 2630 | * switch to non delalloc mode if we are running low |
| 2631 | * on free block. The free block accounting via percpu |
Eric Dumazet | 179f7eb | 2009-01-06 14:41:04 -0800 | [diff] [blame] | 2632 | * counters can get slightly wrong with percpu_counter_batch getting |
Aneesh Kumar K.V | 79f0be8 | 2008-10-08 23:13:30 -0400 | [diff] [blame] | 2633 | * accumulated on each CPU without updating global counters |
| 2634 | * Delalloc need an accurate free block accounting. So switch |
| 2635 | * to non delalloc when we are near to error range. |
| 2636 | */ |
Eric Whitney | 5c1ff33 | 2013-04-09 09:27:31 -0400 | [diff] [blame] | 2637 | free_clusters = |
| 2638 | percpu_counter_read_positive(&sbi->s_freeclusters_counter); |
| 2639 | dirty_clusters = |
| 2640 | percpu_counter_read_positive(&sbi->s_dirtyclusters_counter); |
Theodore Ts'o | 00d4e73 | 2012-09-19 22:42:36 -0400 | [diff] [blame] | 2641 | /* |
| 2642 | * Start pushing delalloc when 1/2 of free blocks are dirty. |
| 2643 | */ |
Eric Whitney | 5c1ff33 | 2013-04-09 09:27:31 -0400 | [diff] [blame] | 2644 | if (dirty_clusters && (free_clusters < 2 * dirty_clusters)) |
Miao Xie | 10ee27a | 2013-01-10 13:47:57 +0800 | [diff] [blame] | 2645 | try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE); |
Theodore Ts'o | 00d4e73 | 2012-09-19 22:42:36 -0400 | [diff] [blame] | 2646 | |
Eric Whitney | 5c1ff33 | 2013-04-09 09:27:31 -0400 | [diff] [blame] | 2647 | if (2 * free_clusters < 3 * dirty_clusters || |
| 2648 | free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) { |
Aneesh Kumar K.V | 79f0be8 | 2008-10-08 23:13:30 -0400 | [diff] [blame] | 2649 | /* |
Eric Sandeen | c8afb44 | 2009-12-23 07:58:12 -0500 | [diff] [blame] | 2650 | * free block count is less than 150% of dirty blocks |
| 2651 | * or free blocks is less than watermark |
Aneesh Kumar K.V | 79f0be8 | 2008-10-08 23:13:30 -0400 | [diff] [blame] | 2652 | */ |
| 2653 | return 1; |
| 2654 | } |
| 2655 | return 0; |
| 2656 | } |
| 2657 | |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2658 | static int ext4_da_write_begin(struct file *file, struct address_space *mapping, |
Theodore Ts'o | de9a55b | 2009-06-14 17:45:34 -0400 | [diff] [blame] | 2659 | loff_t pos, unsigned len, unsigned flags, |
| 2660 | struct page **pagep, void **fsdata) |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2661 | { |
Eric Sandeen | 72b8ab9 | 2010-05-16 11:00:00 -0400 | [diff] [blame] | 2662 | int ret, retries = 0; |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2663 | struct page *page; |
| 2664 | pgoff_t index; |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2665 | struct inode *inode = mapping->host; |
| 2666 | handle_t *handle; |
| 2667 | |
| 2668 | index = pos >> PAGE_CACHE_SHIFT; |
Aneesh Kumar K.V | 79f0be8 | 2008-10-08 23:13:30 -0400 | [diff] [blame] | 2669 | |
| 2670 | if (ext4_nonda_switch(inode->i_sb)) { |
| 2671 | *fsdata = (void *)FALL_BACK_TO_NONDELALLOC; |
| 2672 | return ext4_write_begin(file, mapping, pos, |
| 2673 | len, flags, pagep, fsdata); |
| 2674 | } |
| 2675 | *fsdata = (void *)0; |
Theodore Ts'o | 9bffad1 | 2009-06-17 11:48:11 -0400 | [diff] [blame] | 2676 | trace_ext4_da_write_begin(inode, pos, len, flags); |
Tao Ma | 9c3569b | 2012-12-10 14:05:57 -0500 | [diff] [blame] | 2677 | |
| 2678 | if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { |
| 2679 | ret = ext4_da_write_inline_data_begin(mapping, inode, |
| 2680 | pos, len, flags, |
| 2681 | pagep, fsdata); |
| 2682 | if (ret < 0) |
Theodore Ts'o | 47564bf | 2013-02-09 09:24:14 -0500 | [diff] [blame] | 2683 | return ret; |
| 2684 | if (ret == 1) |
| 2685 | return 0; |
Tao Ma | 9c3569b | 2012-12-10 14:05:57 -0500 | [diff] [blame] | 2686 | } |
| 2687 | |
Theodore Ts'o | 47564bf | 2013-02-09 09:24:14 -0500 | [diff] [blame] | 2688 | /* |
| 2689 | * grab_cache_page_write_begin() can take a long time if the |
| 2690 | * system is thrashing due to memory pressure, or if the page |
| 2691 | * is being written back. So grab it first before we start |
| 2692 | * the transaction handle. This also allows us to allocate |
| 2693 | * the page (if needed) without using GFP_NOFS. |
| 2694 | */ |
| 2695 | retry_grab: |
| 2696 | page = grab_cache_page_write_begin(mapping, index, flags); |
| 2697 | if (!page) |
| 2698 | return -ENOMEM; |
| 2699 | unlock_page(page); |
| 2700 | |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2701 | /* |
| 2702 | * With delayed allocation, we don't log the i_disksize update |
| 2703 | * if there is delayed block allocation. But we still need |
| 2704 | * to journalling the i_disksize update if writes to the end |
| 2705 | * of file which has an already mapped buffer. |
| 2706 | */ |
Theodore Ts'o | 47564bf | 2013-02-09 09:24:14 -0500 | [diff] [blame] | 2707 | retry_journal: |
Theodore Ts'o | 9924a92 | 2013-02-08 21:59:22 -0500 | [diff] [blame] | 2708 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, 1); |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2709 | if (IS_ERR(handle)) { |
Theodore Ts'o | 47564bf | 2013-02-09 09:24:14 -0500 | [diff] [blame] | 2710 | page_cache_release(page); |
| 2711 | return PTR_ERR(handle); |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2712 | } |
| 2713 | |
Theodore Ts'o | 47564bf | 2013-02-09 09:24:14 -0500 | [diff] [blame] | 2714 | lock_page(page); |
| 2715 | if (page->mapping != mapping) { |
| 2716 | /* The page got truncated from under us */ |
| 2717 | unlock_page(page); |
| 2718 | page_cache_release(page); |
Eric Sandeen | d5a0d4f | 2008-08-02 18:51:06 -0400 | [diff] [blame] | 2719 | ext4_journal_stop(handle); |
Theodore Ts'o | 47564bf | 2013-02-09 09:24:14 -0500 | [diff] [blame] | 2720 | goto retry_grab; |
Eric Sandeen | d5a0d4f | 2008-08-02 18:51:06 -0400 | [diff] [blame] | 2721 | } |
Theodore Ts'o | 47564bf | 2013-02-09 09:24:14 -0500 | [diff] [blame] | 2722 | /* In case writeback began while the page was unlocked */ |
| 2723 | wait_on_page_writeback(page); |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2724 | |
Christoph Hellwig | 6e1db88 | 2010-06-04 11:29:57 +0200 | [diff] [blame] | 2725 | ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep); |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2726 | if (ret < 0) { |
| 2727 | unlock_page(page); |
| 2728 | ext4_journal_stop(handle); |
Aneesh Kumar K.V | ae4d537 | 2008-09-13 13:10:25 -0400 | [diff] [blame] | 2729 | /* |
| 2730 | * block_write_begin may have instantiated a few blocks |
| 2731 | * outside i_size. Trim these off again. Don't need |
| 2732 | * i_size_read because we hold i_mutex. |
| 2733 | */ |
| 2734 | if (pos + len > inode->i_size) |
Jan Kara | b9a4207 | 2009-12-08 21:24:33 -0500 | [diff] [blame] | 2735 | ext4_truncate_failed_write(inode); |
Theodore Ts'o | 47564bf | 2013-02-09 09:24:14 -0500 | [diff] [blame] | 2736 | |
| 2737 | if (ret == -ENOSPC && |
| 2738 | ext4_should_retry_alloc(inode->i_sb, &retries)) |
| 2739 | goto retry_journal; |
| 2740 | |
| 2741 | page_cache_release(page); |
| 2742 | return ret; |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2743 | } |
| 2744 | |
Theodore Ts'o | 47564bf | 2013-02-09 09:24:14 -0500 | [diff] [blame] | 2745 | *pagep = page; |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2746 | return ret; |
| 2747 | } |
| 2748 | |
Mingming Cao | 632eaea | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2749 | /* |
| 2750 | * Check if we should update i_disksize |
| 2751 | * when write to the end of file but not require block allocation |
| 2752 | */ |
| 2753 | static int ext4_da_should_update_i_disksize(struct page *page, |
Theodore Ts'o | de9a55b | 2009-06-14 17:45:34 -0400 | [diff] [blame] | 2754 | unsigned long offset) |
Mingming Cao | 632eaea | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2755 | { |
| 2756 | struct buffer_head *bh; |
| 2757 | struct inode *inode = page->mapping->host; |
| 2758 | unsigned int idx; |
| 2759 | int i; |
| 2760 | |
| 2761 | bh = page_buffers(page); |
| 2762 | idx = offset >> inode->i_blkbits; |
| 2763 | |
Theodore Ts'o | af5bc92 | 2008-09-08 22:25:24 -0400 | [diff] [blame] | 2764 | for (i = 0; i < idx; i++) |
Mingming Cao | 632eaea | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2765 | bh = bh->b_this_page; |
| 2766 | |
Aneesh Kumar K.V | 29fa89d | 2009-05-12 16:30:27 -0400 | [diff] [blame] | 2767 | if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh)) |
Mingming Cao | 632eaea | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2768 | return 0; |
| 2769 | return 1; |
| 2770 | } |
| 2771 | |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2772 | static int ext4_da_write_end(struct file *file, |
Theodore Ts'o | de9a55b | 2009-06-14 17:45:34 -0400 | [diff] [blame] | 2773 | struct address_space *mapping, |
| 2774 | loff_t pos, unsigned len, unsigned copied, |
| 2775 | struct page *page, void *fsdata) |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2776 | { |
| 2777 | struct inode *inode = mapping->host; |
| 2778 | int ret = 0, ret2; |
| 2779 | handle_t *handle = ext4_journal_current_handle(); |
| 2780 | loff_t new_i_size; |
Mingming Cao | 632eaea | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2781 | unsigned long start, end; |
Aneesh Kumar K.V | 79f0be8 | 2008-10-08 23:13:30 -0400 | [diff] [blame] | 2782 | int write_mode = (int)(unsigned long)fsdata; |
| 2783 | |
Theodore Ts'o | 74d553a | 2013-04-03 12:39:17 -0400 | [diff] [blame] | 2784 | if (write_mode == FALL_BACK_TO_NONDELALLOC) |
| 2785 | return ext4_write_end(file, mapping, pos, |
| 2786 | len, copied, page, fsdata); |
Mingming Cao | 632eaea | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2787 | |
Theodore Ts'o | 9bffad1 | 2009-06-17 11:48:11 -0400 | [diff] [blame] | 2788 | trace_ext4_da_write_end(inode, pos, len, copied); |
Mingming Cao | 632eaea | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2789 | start = pos & (PAGE_CACHE_SIZE - 1); |
Theodore Ts'o | af5bc92 | 2008-09-08 22:25:24 -0400 | [diff] [blame] | 2790 | end = start + copied - 1; |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2791 | |
| 2792 | /* |
| 2793 | * generic_write_end() will run mark_inode_dirty() if i_size |
| 2794 | * changes. So let's piggyback the i_disksize mark_inode_dirty |
| 2795 | * into that. |
| 2796 | */ |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2797 | new_i_size = pos + copied; |
Andrea Arcangeli | ea51d13 | 2011-12-13 21:41:15 -0500 | [diff] [blame] | 2798 | if (copied && new_i_size > EXT4_I(inode)->i_disksize) { |
Tao Ma | 9c3569b | 2012-12-10 14:05:57 -0500 | [diff] [blame] | 2799 | if (ext4_has_inline_data(inode) || |
| 2800 | ext4_da_should_update_i_disksize(page, end)) { |
Mingming Cao | 632eaea | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2801 | down_write(&EXT4_I(inode)->i_data_sem); |
Theodore Ts'o | f3b5929 | 2012-11-15 23:08:57 -0500 | [diff] [blame] | 2802 | if (new_i_size > EXT4_I(inode)->i_disksize) |
Mingming Cao | 632eaea | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2803 | EXT4_I(inode)->i_disksize = new_i_size; |
Mingming Cao | 632eaea | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2804 | up_write(&EXT4_I(inode)->i_data_sem); |
Aneesh Kumar K.V | cf17fea | 2008-09-13 13:06:18 -0400 | [diff] [blame] | 2805 | /* We need to mark inode dirty even if |
| 2806 | * new_i_size is less that inode->i_size |
| 2807 | * bu greater than i_disksize.(hint delalloc) |
| 2808 | */ |
| 2809 | ext4_mark_inode_dirty(handle, inode); |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2810 | } |
Mingming Cao | 632eaea | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2811 | } |
Tao Ma | 9c3569b | 2012-12-10 14:05:57 -0500 | [diff] [blame] | 2812 | |
| 2813 | if (write_mode != CONVERT_INLINE_DATA && |
| 2814 | ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) && |
| 2815 | ext4_has_inline_data(inode)) |
| 2816 | ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied, |
| 2817 | page); |
| 2818 | else |
| 2819 | ret2 = generic_write_end(file, mapping, pos, len, copied, |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2820 | page, fsdata); |
Tao Ma | 9c3569b | 2012-12-10 14:05:57 -0500 | [diff] [blame] | 2821 | |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2822 | copied = ret2; |
| 2823 | if (ret2 < 0) |
| 2824 | ret = ret2; |
| 2825 | ret2 = ext4_journal_stop(handle); |
| 2826 | if (!ret) |
| 2827 | ret = ret2; |
| 2828 | |
| 2829 | return ret ? ret : copied; |
| 2830 | } |
| 2831 | |
| 2832 | static void ext4_da_invalidatepage(struct page *page, unsigned long offset) |
| 2833 | { |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2834 | /* |
| 2835 | * Drop reserved blocks |
| 2836 | */ |
| 2837 | BUG_ON(!PageLocked(page)); |
| 2838 | if (!page_has_buffers(page)) |
| 2839 | goto out; |
| 2840 | |
Mingming Cao | d2a1763 | 2008-07-14 17:52:37 -0400 | [diff] [blame] | 2841 | ext4_da_page_release_reservation(page, offset); |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2842 | |
| 2843 | out: |
| 2844 | ext4_invalidatepage(page, offset); |
| 2845 | |
| 2846 | return; |
| 2847 | } |
| 2848 | |
Theodore Ts'o | ccd2506 | 2009-02-26 01:04:07 -0500 | [diff] [blame] | 2849 | /* |
| 2850 | * Force all delayed allocation blocks to be allocated for a given inode. |
| 2851 | */ |
| 2852 | int ext4_alloc_da_blocks(struct inode *inode) |
| 2853 | { |
Theodore Ts'o | fb40ba0 | 2009-09-16 19:30:40 -0400 | [diff] [blame] | 2854 | trace_ext4_alloc_da_blocks(inode); |
| 2855 | |
Theodore Ts'o | ccd2506 | 2009-02-26 01:04:07 -0500 | [diff] [blame] | 2856 | if (!EXT4_I(inode)->i_reserved_data_blocks && |
| 2857 | !EXT4_I(inode)->i_reserved_meta_blocks) |
| 2858 | return 0; |
| 2859 | |
| 2860 | /* |
| 2861 | * We do something simple for now. The filemap_flush() will |
| 2862 | * also start triggering a write of the data blocks, which is |
| 2863 | * not strictly speaking necessary (and for users of |
| 2864 | * laptop_mode, not even desirable). However, to do otherwise |
| 2865 | * would require replicating code paths in: |
Theodore Ts'o | de9a55b | 2009-06-14 17:45:34 -0400 | [diff] [blame] | 2866 | * |
Theodore Ts'o | ccd2506 | 2009-02-26 01:04:07 -0500 | [diff] [blame] | 2867 | * ext4_da_writepages() -> |
| 2868 | * write_cache_pages() ---> (via passed in callback function) |
| 2869 | * __mpage_da_writepage() --> |
| 2870 | * mpage_add_bh_to_extent() |
| 2871 | * mpage_da_map_blocks() |
| 2872 | * |
| 2873 | * The problem is that write_cache_pages(), located in |
| 2874 | * mm/page-writeback.c, marks pages clean in preparation for |
| 2875 | * doing I/O, which is not desirable if we're not planning on |
| 2876 | * doing I/O at all. |
| 2877 | * |
| 2878 | * We could call write_cache_pages(), and then redirty all of |
Wu Fengguang | 380cf09 | 2010-11-11 19:23:29 +0800 | [diff] [blame] | 2879 | * the pages by calling redirty_page_for_writepage() but that |
Theodore Ts'o | ccd2506 | 2009-02-26 01:04:07 -0500 | [diff] [blame] | 2880 | * would be ugly in the extreme. So instead we would need to |
| 2881 | * replicate parts of the code in the above functions, |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 2882 | * simplifying them because we wouldn't actually intend to |
Theodore Ts'o | ccd2506 | 2009-02-26 01:04:07 -0500 | [diff] [blame] | 2883 | * write out the pages, but rather only collect contiguous |
| 2884 | * logical block extents, call the multi-block allocator, and |
| 2885 | * then update the buffer heads with the block allocations. |
Theodore Ts'o | de9a55b | 2009-06-14 17:45:34 -0400 | [diff] [blame] | 2886 | * |
Theodore Ts'o | ccd2506 | 2009-02-26 01:04:07 -0500 | [diff] [blame] | 2887 | * For now, though, we'll cheat by calling filemap_flush(), |
| 2888 | * which will map the blocks, and start the I/O, but not |
| 2889 | * actually wait for the I/O to complete. |
| 2890 | */ |
| 2891 | return filemap_flush(inode->i_mapping); |
| 2892 | } |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2893 | |
| 2894 | /* |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2895 | * bmap() is special. It gets used by applications such as lilo and by |
| 2896 | * the swapper to find the on-disk block of a specific piece of data. |
| 2897 | * |
| 2898 | * Naturally, this is dangerous if the block concerned is still in the |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2899 | * journal. If somebody makes a swapfile on an ext4 data-journaling |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2900 | * filesystem and enables swap, then they may get a nasty shock when the |
| 2901 | * data getting swapped to that swapfile suddenly gets overwritten by |
| 2902 | * the original zero's written out previously to the journal and |
| 2903 | * awaiting writeback in the kernel's buffer cache. |
| 2904 | * |
| 2905 | * So, if we see any bmap calls here on a modified, data-journaled file, |
| 2906 | * take extra steps to flush any blocks which might be in the cache. |
| 2907 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2908 | static sector_t ext4_bmap(struct address_space *mapping, sector_t block) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2909 | { |
| 2910 | struct inode *inode = mapping->host; |
| 2911 | journal_t *journal; |
| 2912 | int err; |
| 2913 | |
Tao Ma | 46c7f25 | 2012-12-10 14:04:52 -0500 | [diff] [blame] | 2914 | /* |
| 2915 | * We can get here for an inline file via the FIBMAP ioctl |
| 2916 | */ |
| 2917 | if (ext4_has_inline_data(inode)) |
| 2918 | return 0; |
| 2919 | |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 2920 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) && |
| 2921 | test_opt(inode->i_sb, DELALLOC)) { |
| 2922 | /* |
| 2923 | * With delalloc we want to sync the file |
| 2924 | * so that we can make sure we allocate |
| 2925 | * blocks for file |
| 2926 | */ |
| 2927 | filemap_write_and_wait(mapping); |
| 2928 | } |
| 2929 | |
Theodore Ts'o | 19f5fb7 | 2010-01-24 14:34:07 -0500 | [diff] [blame] | 2930 | if (EXT4_JOURNAL(inode) && |
| 2931 | ext4_test_inode_state(inode, EXT4_STATE_JDATA)) { |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2932 | /* |
| 2933 | * This is a REALLY heavyweight approach, but the use of |
| 2934 | * bmap on dirty files is expected to be extremely rare: |
| 2935 | * only if we run lilo or swapon on a freshly made file |
| 2936 | * do we expect this to happen. |
| 2937 | * |
| 2938 | * (bmap requires CAP_SYS_RAWIO so this does not |
| 2939 | * represent an unprivileged user DOS attack --- we'd be |
| 2940 | * in trouble if mortal users could trigger this path at |
| 2941 | * will.) |
| 2942 | * |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2943 | * NB. EXT4_STATE_JDATA is not set on files other than |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2944 | * regular files. If somebody wants to bmap a directory |
| 2945 | * or symlink and gets confused because the buffer |
| 2946 | * hasn't yet been flushed to disk, they deserve |
| 2947 | * everything they get. |
| 2948 | */ |
| 2949 | |
Theodore Ts'o | 19f5fb7 | 2010-01-24 14:34:07 -0500 | [diff] [blame] | 2950 | ext4_clear_inode_state(inode, EXT4_STATE_JDATA); |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2951 | journal = EXT4_JOURNAL(inode); |
Mingming Cao | dab291a | 2006-10-11 01:21:01 -0700 | [diff] [blame] | 2952 | jbd2_journal_lock_updates(journal); |
| 2953 | err = jbd2_journal_flush(journal); |
| 2954 | jbd2_journal_unlock_updates(journal); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2955 | |
| 2956 | if (err) |
| 2957 | return 0; |
| 2958 | } |
| 2959 | |
Theodore Ts'o | af5bc92 | 2008-09-08 22:25:24 -0400 | [diff] [blame] | 2960 | return generic_block_bmap(mapping, block, ext4_get_block); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2961 | } |
| 2962 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2963 | static int ext4_readpage(struct file *file, struct page *page) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2964 | { |
Tao Ma | 46c7f25 | 2012-12-10 14:04:52 -0500 | [diff] [blame] | 2965 | int ret = -EAGAIN; |
| 2966 | struct inode *inode = page->mapping->host; |
| 2967 | |
Jiaying Zhang | 0562e0b | 2011-03-21 21:38:05 -0400 | [diff] [blame] | 2968 | trace_ext4_readpage(page); |
Tao Ma | 46c7f25 | 2012-12-10 14:04:52 -0500 | [diff] [blame] | 2969 | |
| 2970 | if (ext4_has_inline_data(inode)) |
| 2971 | ret = ext4_readpage_inline(inode, page); |
| 2972 | |
| 2973 | if (ret == -EAGAIN) |
| 2974 | return mpage_readpage(page, ext4_get_block); |
| 2975 | |
| 2976 | return ret; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2977 | } |
| 2978 | |
| 2979 | static int |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2980 | ext4_readpages(struct file *file, struct address_space *mapping, |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2981 | struct list_head *pages, unsigned nr_pages) |
| 2982 | { |
Tao Ma | 46c7f25 | 2012-12-10 14:04:52 -0500 | [diff] [blame] | 2983 | struct inode *inode = mapping->host; |
| 2984 | |
| 2985 | /* If the file has inline data, no need to do readpages. */ |
| 2986 | if (ext4_has_inline_data(inode)) |
| 2987 | return 0; |
| 2988 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2989 | return mpage_readpages(mapping, pages, nr_pages, ext4_get_block); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2990 | } |
| 2991 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 2992 | static void ext4_invalidatepage(struct page *page, unsigned long offset) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 2993 | { |
Jiaying Zhang | 0562e0b | 2011-03-21 21:38:05 -0400 | [diff] [blame] | 2994 | trace_ext4_invalidatepage(page, offset); |
| 2995 | |
Jan Kara | 4520fb3 | 2012-12-25 13:28:54 -0500 | [diff] [blame] | 2996 | /* No journalling happens on data buffers when this function is used */ |
| 2997 | WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page))); |
| 2998 | |
| 2999 | block_invalidatepage(page, offset); |
| 3000 | } |
| 3001 | |
Jan Kara | 53e8726 | 2012-12-25 13:29:52 -0500 | [diff] [blame] | 3002 | static int __ext4_journalled_invalidatepage(struct page *page, |
| 3003 | unsigned long offset) |
Jan Kara | 4520fb3 | 2012-12-25 13:28:54 -0500 | [diff] [blame] | 3004 | { |
| 3005 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); |
| 3006 | |
| 3007 | trace_ext4_journalled_invalidatepage(page, offset); |
| 3008 | |
Jiaying Zhang | 744692d | 2010-03-04 16:14:02 -0500 | [diff] [blame] | 3009 | /* |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3010 | * If it's a full truncate we just forget about the pending dirtying |
| 3011 | */ |
| 3012 | if (offset == 0) |
| 3013 | ClearPageChecked(page); |
| 3014 | |
Jan Kara | 53e8726 | 2012-12-25 13:29:52 -0500 | [diff] [blame] | 3015 | return jbd2_journal_invalidatepage(journal, page, offset); |
| 3016 | } |
| 3017 | |
| 3018 | /* Wrapper for aops... */ |
| 3019 | static void ext4_journalled_invalidatepage(struct page *page, |
| 3020 | unsigned long offset) |
| 3021 | { |
| 3022 | WARN_ON(__ext4_journalled_invalidatepage(page, offset) < 0); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3023 | } |
| 3024 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3025 | static int ext4_releasepage(struct page *page, gfp_t wait) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3026 | { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3027 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3028 | |
Jiaying Zhang | 0562e0b | 2011-03-21 21:38:05 -0400 | [diff] [blame] | 3029 | trace_ext4_releasepage(page); |
| 3030 | |
Jan Kara | e1c3659 | 2013-03-10 22:19:00 -0400 | [diff] [blame] | 3031 | /* Page has dirty journalled data -> cannot release */ |
| 3032 | if (PageChecked(page)) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3033 | return 0; |
Frank Mayhar | 0390131 | 2009-01-07 00:06:22 -0500 | [diff] [blame] | 3034 | if (journal) |
| 3035 | return jbd2_journal_try_to_free_buffers(journal, page, wait); |
| 3036 | else |
| 3037 | return try_to_free_buffers(page); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3038 | } |
| 3039 | |
| 3040 | /* |
Theodore Ts'o | 2ed8868 | 2010-05-16 20:00:00 -0400 | [diff] [blame] | 3041 | * ext4_get_block used when preparing for a DIO write or buffer write. |
| 3042 | * We allocate an uinitialized extent if blocks haven't been allocated. |
| 3043 | * The extent will be converted to initialized after the IO is complete. |
| 3044 | */ |
Tao Ma | f19d587 | 2012-12-10 14:05:51 -0500 | [diff] [blame] | 3045 | int ext4_get_block_write(struct inode *inode, sector_t iblock, |
Mingming Cao | 4c0425f | 2009-09-28 15:48:41 -0400 | [diff] [blame] | 3046 | struct buffer_head *bh_result, int create) |
| 3047 | { |
Jiaying Zhang | c7064ef | 2010-03-02 13:28:44 -0500 | [diff] [blame] | 3048 | ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n", |
Mingming Cao | 8d5d02e | 2009-09-28 15:48:29 -0400 | [diff] [blame] | 3049 | inode->i_ino, create); |
Theodore Ts'o | 2ed8868 | 2010-05-16 20:00:00 -0400 | [diff] [blame] | 3050 | return _ext4_get_block(inode, iblock, bh_result, |
| 3051 | EXT4_GET_BLOCKS_IO_CREATE_EXT); |
Mingming Cao | 4c0425f | 2009-09-28 15:48:41 -0400 | [diff] [blame] | 3052 | } |
| 3053 | |
Zheng Liu | 729f52c | 2012-07-09 16:29:29 -0400 | [diff] [blame] | 3054 | static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock, |
Anatol Pomozov | 8b0f165 | 2012-11-08 15:07:16 -0500 | [diff] [blame] | 3055 | struct buffer_head *bh_result, int create) |
Zheng Liu | 729f52c | 2012-07-09 16:29:29 -0400 | [diff] [blame] | 3056 | { |
Anatol Pomozov | 8b0f165 | 2012-11-08 15:07:16 -0500 | [diff] [blame] | 3057 | ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n", |
| 3058 | inode->i_ino, create); |
| 3059 | return _ext4_get_block(inode, iblock, bh_result, |
| 3060 | EXT4_GET_BLOCKS_NO_LOCK); |
Zheng Liu | 729f52c | 2012-07-09 16:29:29 -0400 | [diff] [blame] | 3061 | } |
| 3062 | |
Mingming Cao | 4c0425f | 2009-09-28 15:48:41 -0400 | [diff] [blame] | 3063 | static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset, |
Christoph Hellwig | 552ef802 | 2010-07-27 11:56:06 -0400 | [diff] [blame] | 3064 | ssize_t size, void *private, int ret, |
| 3065 | bool is_async) |
Mingming Cao | 4c0425f | 2009-09-28 15:48:41 -0400 | [diff] [blame] | 3066 | { |
Al Viro | 496ad9a | 2013-01-23 17:07:38 -0500 | [diff] [blame] | 3067 | struct inode *inode = file_inode(iocb->ki_filp); |
Mingming Cao | 4c0425f | 2009-09-28 15:48:41 -0400 | [diff] [blame] | 3068 | ext4_io_end_t *io_end = iocb->private; |
Mingming Cao | 4c0425f | 2009-09-28 15:48:41 -0400 | [diff] [blame] | 3069 | |
Theodore Ts'o | a549984 | 2013-05-11 19:07:42 -0400 | [diff] [blame] | 3070 | /* if not async direct IO or dio with 0 bytes write, just return */ |
| 3071 | if (!io_end || !size) |
| 3072 | goto out; |
Mingming | 4b70df1 | 2009-11-03 14:44:54 -0500 | [diff] [blame] | 3073 | |
Zheng Liu | 88635ca | 2011-12-28 19:00:25 -0500 | [diff] [blame] | 3074 | ext_debug("ext4_end_io_dio(): io_end 0x%p " |
Joe Perches | ace36ad | 2012-03-19 23:11:43 -0400 | [diff] [blame] | 3075 | "for inode %lu, iocb 0x%p, offset %llu, size %zd\n", |
Mingming Cao | 8d5d02e | 2009-09-28 15:48:29 -0400 | [diff] [blame] | 3076 | iocb->private, io_end->inode->i_ino, iocb, offset, |
| 3077 | size); |
Mingming Cao | 8d5d02e | 2009-09-28 15:48:29 -0400 | [diff] [blame] | 3078 | |
Theodore Ts'o | b5a7e97 | 2011-12-12 10:53:02 -0500 | [diff] [blame] | 3079 | iocb->private = NULL; |
Theodore Ts'o | a549984 | 2013-05-11 19:07:42 -0400 | [diff] [blame] | 3080 | |
| 3081 | /* if not aio dio with unwritten extents, just free io and return */ |
| 3082 | if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) { |
| 3083 | ext4_free_io_end(io_end); |
| 3084 | out: |
| 3085 | inode_dio_done(inode); |
| 3086 | if (is_async) |
| 3087 | aio_complete(iocb, ret, 0); |
| 3088 | return; |
| 3089 | } |
| 3090 | |
Mingming Cao | 4c0425f | 2009-09-28 15:48:41 -0400 | [diff] [blame] | 3091 | io_end->offset = offset; |
| 3092 | io_end->size = size; |
jiayingz@google.com (Jiaying Zhang) | 5b3ff23 | 2010-07-27 11:56:06 -0400 | [diff] [blame] | 3093 | if (is_async) { |
| 3094 | io_end->iocb = iocb; |
| 3095 | io_end->result = ret; |
| 3096 | } |
Theodore Ts'o | a549984 | 2013-05-11 19:07:42 -0400 | [diff] [blame] | 3097 | |
| 3098 | ext4_add_complete_io(io_end); |
Mingming Cao | 4c0425f | 2009-09-28 15:48:41 -0400 | [diff] [blame] | 3099 | } |
Jiaying Zhang | c7064ef | 2010-03-02 13:28:44 -0500 | [diff] [blame] | 3100 | |
Mingming Cao | 4c0425f | 2009-09-28 15:48:41 -0400 | [diff] [blame] | 3101 | /* |
| 3102 | * For ext4 extent files, ext4 will do direct-io write to holes, |
| 3103 | * preallocated extents, and those write extend the file, no need to |
| 3104 | * fall back to buffered IO. |
| 3105 | * |
Uwe Kleine-König | b595076 | 2010-11-01 15:38:34 -0400 | [diff] [blame] | 3106 | * For holes, we fallocate those blocks, mark them as uninitialized |
Theodore Ts'o | 69c499d | 2012-11-29 21:13:48 -0500 | [diff] [blame] | 3107 | * If those blocks were preallocated, we mark sure they are split, but |
Uwe Kleine-König | b595076 | 2010-11-01 15:38:34 -0400 | [diff] [blame] | 3108 | * still keep the range to write as uninitialized. |
Mingming Cao | 4c0425f | 2009-09-28 15:48:41 -0400 | [diff] [blame] | 3109 | * |
Theodore Ts'o | 69c499d | 2012-11-29 21:13:48 -0500 | [diff] [blame] | 3110 | * The unwritten extents will be converted to written when DIO is completed. |
Mingming Cao | 8d5d02e | 2009-09-28 15:48:29 -0400 | [diff] [blame] | 3111 | * For async direct IO, since the IO may still pending when return, we |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 3112 | * set up an end_io call back function, which will do the conversion |
Mingming Cao | 8d5d02e | 2009-09-28 15:48:29 -0400 | [diff] [blame] | 3113 | * when async direct IO completed. |
Mingming Cao | 4c0425f | 2009-09-28 15:48:41 -0400 | [diff] [blame] | 3114 | * |
| 3115 | * If the O_DIRECT write will extend the file then add this inode to the |
| 3116 | * orphan list. So recovery will truncate it back to the original size |
| 3117 | * if the machine crashes during the write. |
| 3118 | * |
| 3119 | */ |
| 3120 | static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb, |
| 3121 | const struct iovec *iov, loff_t offset, |
| 3122 | unsigned long nr_segs) |
| 3123 | { |
| 3124 | struct file *file = iocb->ki_filp; |
| 3125 | struct inode *inode = file->f_mapping->host; |
| 3126 | ssize_t ret; |
| 3127 | size_t count = iov_length(iov, nr_segs); |
Theodore Ts'o | 69c499d | 2012-11-29 21:13:48 -0500 | [diff] [blame] | 3128 | int overwrite = 0; |
| 3129 | get_block_t *get_block_func = NULL; |
| 3130 | int dio_flags = 0; |
Mingming Cao | 4c0425f | 2009-09-28 15:48:41 -0400 | [diff] [blame] | 3131 | loff_t final_size = offset + count; |
Zheng Liu | 729f52c | 2012-07-09 16:29:29 -0400 | [diff] [blame] | 3132 | |
Theodore Ts'o | 69c499d | 2012-11-29 21:13:48 -0500 | [diff] [blame] | 3133 | /* Use the old path for reads and writes beyond i_size. */ |
| 3134 | if (rw != WRITE || final_size > inode->i_size) |
| 3135 | return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs); |
Zheng Liu | 4bd809d | 2012-07-22 20:19:31 -0400 | [diff] [blame] | 3136 | |
Theodore Ts'o | 69c499d | 2012-11-29 21:13:48 -0500 | [diff] [blame] | 3137 | BUG_ON(iocb->private == NULL); |
Zheng Liu | 4bd809d | 2012-07-22 20:19:31 -0400 | [diff] [blame] | 3138 | |
Theodore Ts'o | 69c499d | 2012-11-29 21:13:48 -0500 | [diff] [blame] | 3139 | /* If we do a overwrite dio, i_mutex locking can be released */ |
| 3140 | overwrite = *((int *)iocb->private); |
Zheng Liu | 4bd809d | 2012-07-22 20:19:31 -0400 | [diff] [blame] | 3141 | |
Theodore Ts'o | 69c499d | 2012-11-29 21:13:48 -0500 | [diff] [blame] | 3142 | if (overwrite) { |
| 3143 | atomic_inc(&inode->i_dio_count); |
| 3144 | down_read(&EXT4_I(inode)->i_data_sem); |
| 3145 | mutex_unlock(&inode->i_mutex); |
Mingming Cao | 4c0425f | 2009-09-28 15:48:41 -0400 | [diff] [blame] | 3146 | } |
Mingming Cao | 8d5d02e | 2009-09-28 15:48:29 -0400 | [diff] [blame] | 3147 | |
Theodore Ts'o | 69c499d | 2012-11-29 21:13:48 -0500 | [diff] [blame] | 3148 | /* |
| 3149 | * We could direct write to holes and fallocate. |
| 3150 | * |
| 3151 | * Allocated blocks to fill the hole are marked as |
| 3152 | * uninitialized to prevent parallel buffered read to expose |
| 3153 | * the stale data before DIO complete the data IO. |
| 3154 | * |
| 3155 | * As to previously fallocated extents, ext4 get_block will |
| 3156 | * just simply mark the buffer mapped but still keep the |
| 3157 | * extents uninitialized. |
| 3158 | * |
| 3159 | * For non AIO case, we will convert those unwritten extents |
| 3160 | * to written after return back from blockdev_direct_IO. |
| 3161 | * |
| 3162 | * For async DIO, the conversion needs to be deferred when the |
| 3163 | * IO is completed. The ext4 end_io callback function will be |
| 3164 | * called to take care of the conversion work. Here for async |
| 3165 | * case, we allocate an io_end structure to hook to the iocb. |
| 3166 | */ |
| 3167 | iocb->private = NULL; |
| 3168 | ext4_inode_aio_set(inode, NULL); |
| 3169 | if (!is_sync_kiocb(iocb)) { |
Theodore Ts'o | a549984 | 2013-05-11 19:07:42 -0400 | [diff] [blame] | 3170 | ext4_io_end_t *io_end = ext4_init_io_end(inode, GFP_NOFS); |
Theodore Ts'o | 69c499d | 2012-11-29 21:13:48 -0500 | [diff] [blame] | 3171 | if (!io_end) { |
| 3172 | ret = -ENOMEM; |
| 3173 | goto retake_lock; |
| 3174 | } |
| 3175 | io_end->flag |= EXT4_IO_END_DIRECT; |
Theodore Ts'o | a549984 | 2013-05-11 19:07:42 -0400 | [diff] [blame] | 3176 | iocb->private = io_end; |
Theodore Ts'o | 69c499d | 2012-11-29 21:13:48 -0500 | [diff] [blame] | 3177 | /* |
| 3178 | * we save the io structure for current async direct |
| 3179 | * IO, so that later ext4_map_blocks() could flag the |
| 3180 | * io structure whether there is a unwritten extents |
| 3181 | * needs to be converted when IO is completed. |
| 3182 | */ |
| 3183 | ext4_inode_aio_set(inode, io_end); |
| 3184 | } |
| 3185 | |
| 3186 | if (overwrite) { |
| 3187 | get_block_func = ext4_get_block_write_nolock; |
| 3188 | } else { |
| 3189 | get_block_func = ext4_get_block_write; |
| 3190 | dio_flags = DIO_LOCKING; |
| 3191 | } |
| 3192 | ret = __blockdev_direct_IO(rw, iocb, inode, |
| 3193 | inode->i_sb->s_bdev, iov, |
| 3194 | offset, nr_segs, |
| 3195 | get_block_func, |
| 3196 | ext4_end_io_dio, |
| 3197 | NULL, |
| 3198 | dio_flags); |
| 3199 | |
Theodore Ts'o | a549984 | 2013-05-11 19:07:42 -0400 | [diff] [blame] | 3200 | if (iocb->private) |
Jan Kara | 4eec708 | 2013-04-11 23:56:53 -0400 | [diff] [blame] | 3201 | ext4_inode_aio_set(inode, NULL); |
Theodore Ts'o | a549984 | 2013-05-11 19:07:42 -0400 | [diff] [blame] | 3202 | /* |
| 3203 | * The io_end structure takes a reference to the inode, that |
| 3204 | * structure needs to be destroyed and the reference to the |
| 3205 | * inode need to be dropped, when IO is complete, even with 0 |
| 3206 | * byte write, or failed. |
| 3207 | * |
| 3208 | * In the successful AIO DIO case, the io_end structure will |
| 3209 | * be destroyed and the reference to the inode will be dropped |
| 3210 | * after the end_io call back function is called. |
| 3211 | * |
| 3212 | * In the case there is 0 byte write, or error case, since VFS |
| 3213 | * direct IO won't invoke the end_io call back function, we |
| 3214 | * need to free the end_io structure here. |
| 3215 | */ |
| 3216 | if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) { |
| 3217 | ext4_free_io_end(iocb->private); |
| 3218 | iocb->private = NULL; |
| 3219 | } else if (ret > 0 && !overwrite && ext4_test_inode_state(inode, |
Theodore Ts'o | 69c499d | 2012-11-29 21:13:48 -0500 | [diff] [blame] | 3220 | EXT4_STATE_DIO_UNWRITTEN)) { |
| 3221 | int err; |
| 3222 | /* |
| 3223 | * for non AIO case, since the IO is already |
| 3224 | * completed, we could do the conversion right here |
| 3225 | */ |
| 3226 | err = ext4_convert_unwritten_extents(inode, |
| 3227 | offset, ret); |
| 3228 | if (err < 0) |
| 3229 | ret = err; |
| 3230 | ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN); |
| 3231 | } |
| 3232 | |
| 3233 | retake_lock: |
| 3234 | /* take i_mutex locking again if we do a ovewrite dio */ |
| 3235 | if (overwrite) { |
| 3236 | inode_dio_done(inode); |
| 3237 | up_read(&EXT4_I(inode)->i_data_sem); |
| 3238 | mutex_lock(&inode->i_mutex); |
| 3239 | } |
| 3240 | |
| 3241 | return ret; |
Mingming Cao | 4c0425f | 2009-09-28 15:48:41 -0400 | [diff] [blame] | 3242 | } |
| 3243 | |
| 3244 | static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb, |
| 3245 | const struct iovec *iov, loff_t offset, |
| 3246 | unsigned long nr_segs) |
| 3247 | { |
| 3248 | struct file *file = iocb->ki_filp; |
| 3249 | struct inode *inode = file->f_mapping->host; |
Jiaying Zhang | 0562e0b | 2011-03-21 21:38:05 -0400 | [diff] [blame] | 3250 | ssize_t ret; |
Mingming Cao | 4c0425f | 2009-09-28 15:48:41 -0400 | [diff] [blame] | 3251 | |
Theodore Ts'o | 84ebd79 | 2011-08-31 11:56:51 -0400 | [diff] [blame] | 3252 | /* |
| 3253 | * If we are doing data journalling we don't support O_DIRECT |
| 3254 | */ |
| 3255 | if (ext4_should_journal_data(inode)) |
| 3256 | return 0; |
| 3257 | |
Tao Ma | 46c7f25 | 2012-12-10 14:04:52 -0500 | [diff] [blame] | 3258 | /* Let buffer I/O handle the inline data case. */ |
| 3259 | if (ext4_has_inline_data(inode)) |
| 3260 | return 0; |
| 3261 | |
Jiaying Zhang | 0562e0b | 2011-03-21 21:38:05 -0400 | [diff] [blame] | 3262 | trace_ext4_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw); |
Dmitry Monakhov | 12e9b89 | 2010-05-16 22:00:00 -0400 | [diff] [blame] | 3263 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
Jiaying Zhang | 0562e0b | 2011-03-21 21:38:05 -0400 | [diff] [blame] | 3264 | ret = ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs); |
| 3265 | else |
| 3266 | ret = ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs); |
| 3267 | trace_ext4_direct_IO_exit(inode, offset, |
| 3268 | iov_length(iov, nr_segs), rw, ret); |
| 3269 | return ret; |
Mingming Cao | 4c0425f | 2009-09-28 15:48:41 -0400 | [diff] [blame] | 3270 | } |
| 3271 | |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3272 | /* |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3273 | * Pages can be marked dirty completely asynchronously from ext4's journalling |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3274 | * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do |
| 3275 | * much here because ->set_page_dirty is called under VFS locks. The page is |
| 3276 | * not necessarily locked. |
| 3277 | * |
| 3278 | * We cannot just dirty the page and leave attached buffers clean, because the |
| 3279 | * buffers' dirty state is "definitive". We cannot just set the buffers dirty |
| 3280 | * or jbddirty because all the journalling code will explode. |
| 3281 | * |
| 3282 | * So what we do is to mark the page "pending dirty" and next time writepage |
| 3283 | * is called, propagate that into the buffers appropriately. |
| 3284 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3285 | static int ext4_journalled_set_page_dirty(struct page *page) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3286 | { |
| 3287 | SetPageChecked(page); |
| 3288 | return __set_page_dirty_nobuffers(page); |
| 3289 | } |
| 3290 | |
Theodore Ts'o | 74d553a | 2013-04-03 12:39:17 -0400 | [diff] [blame] | 3291 | static const struct address_space_operations ext4_aops = { |
Hisashi Hifumi | 8ab22b9 | 2008-07-28 15:46:36 -0700 | [diff] [blame] | 3292 | .readpage = ext4_readpage, |
| 3293 | .readpages = ext4_readpages, |
Aneesh Kumar K.V | 43ce1d2 | 2009-06-14 17:58:45 -0400 | [diff] [blame] | 3294 | .writepage = ext4_writepage, |
Hisashi Hifumi | 8ab22b9 | 2008-07-28 15:46:36 -0700 | [diff] [blame] | 3295 | .write_begin = ext4_write_begin, |
Theodore Ts'o | 74d553a | 2013-04-03 12:39:17 -0400 | [diff] [blame] | 3296 | .write_end = ext4_write_end, |
Hisashi Hifumi | 8ab22b9 | 2008-07-28 15:46:36 -0700 | [diff] [blame] | 3297 | .bmap = ext4_bmap, |
| 3298 | .invalidatepage = ext4_invalidatepage, |
| 3299 | .releasepage = ext4_releasepage, |
| 3300 | .direct_IO = ext4_direct_IO, |
| 3301 | .migratepage = buffer_migrate_page, |
| 3302 | .is_partially_uptodate = block_is_partially_uptodate, |
Andi Kleen | aa261f5 | 2009-09-16 11:50:16 +0200 | [diff] [blame] | 3303 | .error_remove_page = generic_error_remove_page, |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3304 | }; |
| 3305 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3306 | static const struct address_space_operations ext4_journalled_aops = { |
Hisashi Hifumi | 8ab22b9 | 2008-07-28 15:46:36 -0700 | [diff] [blame] | 3307 | .readpage = ext4_readpage, |
| 3308 | .readpages = ext4_readpages, |
Aneesh Kumar K.V | 43ce1d2 | 2009-06-14 17:58:45 -0400 | [diff] [blame] | 3309 | .writepage = ext4_writepage, |
Hisashi Hifumi | 8ab22b9 | 2008-07-28 15:46:36 -0700 | [diff] [blame] | 3310 | .write_begin = ext4_write_begin, |
| 3311 | .write_end = ext4_journalled_write_end, |
| 3312 | .set_page_dirty = ext4_journalled_set_page_dirty, |
| 3313 | .bmap = ext4_bmap, |
Jan Kara | 4520fb3 | 2012-12-25 13:28:54 -0500 | [diff] [blame] | 3314 | .invalidatepage = ext4_journalled_invalidatepage, |
Hisashi Hifumi | 8ab22b9 | 2008-07-28 15:46:36 -0700 | [diff] [blame] | 3315 | .releasepage = ext4_releasepage, |
Theodore Ts'o | 84ebd79 | 2011-08-31 11:56:51 -0400 | [diff] [blame] | 3316 | .direct_IO = ext4_direct_IO, |
Hisashi Hifumi | 8ab22b9 | 2008-07-28 15:46:36 -0700 | [diff] [blame] | 3317 | .is_partially_uptodate = block_is_partially_uptodate, |
Andi Kleen | aa261f5 | 2009-09-16 11:50:16 +0200 | [diff] [blame] | 3318 | .error_remove_page = generic_error_remove_page, |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3319 | }; |
| 3320 | |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 3321 | static const struct address_space_operations ext4_da_aops = { |
Hisashi Hifumi | 8ab22b9 | 2008-07-28 15:46:36 -0700 | [diff] [blame] | 3322 | .readpage = ext4_readpage, |
| 3323 | .readpages = ext4_readpages, |
Aneesh Kumar K.V | 43ce1d2 | 2009-06-14 17:58:45 -0400 | [diff] [blame] | 3324 | .writepage = ext4_writepage, |
Hisashi Hifumi | 8ab22b9 | 2008-07-28 15:46:36 -0700 | [diff] [blame] | 3325 | .writepages = ext4_da_writepages, |
Hisashi Hifumi | 8ab22b9 | 2008-07-28 15:46:36 -0700 | [diff] [blame] | 3326 | .write_begin = ext4_da_write_begin, |
| 3327 | .write_end = ext4_da_write_end, |
| 3328 | .bmap = ext4_bmap, |
| 3329 | .invalidatepage = ext4_da_invalidatepage, |
| 3330 | .releasepage = ext4_releasepage, |
| 3331 | .direct_IO = ext4_direct_IO, |
| 3332 | .migratepage = buffer_migrate_page, |
| 3333 | .is_partially_uptodate = block_is_partially_uptodate, |
Andi Kleen | aa261f5 | 2009-09-16 11:50:16 +0200 | [diff] [blame] | 3334 | .error_remove_page = generic_error_remove_page, |
Alex Tomas | 6476924 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 3335 | }; |
| 3336 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3337 | void ext4_set_aops(struct inode *inode) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3338 | { |
Lukas Czerner | 3d2b158 | 2012-02-20 17:53:00 -0500 | [diff] [blame] | 3339 | switch (ext4_inode_journal_mode(inode)) { |
| 3340 | case EXT4_INODE_ORDERED_DATA_MODE: |
Theodore Ts'o | 74d553a | 2013-04-03 12:39:17 -0400 | [diff] [blame] | 3341 | ext4_set_inode_state(inode, EXT4_STATE_ORDERED_MODE); |
Lukas Czerner | 3d2b158 | 2012-02-20 17:53:00 -0500 | [diff] [blame] | 3342 | break; |
| 3343 | case EXT4_INODE_WRITEBACK_DATA_MODE: |
Theodore Ts'o | 74d553a | 2013-04-03 12:39:17 -0400 | [diff] [blame] | 3344 | ext4_clear_inode_state(inode, EXT4_STATE_ORDERED_MODE); |
Lukas Czerner | 3d2b158 | 2012-02-20 17:53:00 -0500 | [diff] [blame] | 3345 | break; |
| 3346 | case EXT4_INODE_JOURNAL_DATA_MODE: |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3347 | inode->i_mapping->a_ops = &ext4_journalled_aops; |
Theodore Ts'o | 74d553a | 2013-04-03 12:39:17 -0400 | [diff] [blame] | 3348 | return; |
Lukas Czerner | 3d2b158 | 2012-02-20 17:53:00 -0500 | [diff] [blame] | 3349 | default: |
| 3350 | BUG(); |
| 3351 | } |
Theodore Ts'o | 74d553a | 2013-04-03 12:39:17 -0400 | [diff] [blame] | 3352 | if (test_opt(inode->i_sb, DELALLOC)) |
| 3353 | inode->i_mapping->a_ops = &ext4_da_aops; |
| 3354 | else |
| 3355 | inode->i_mapping->a_ops = &ext4_aops; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3356 | } |
| 3357 | |
Allison Henderson | 4e96b2d | 2011-09-03 11:51:09 -0400 | [diff] [blame] | 3358 | |
| 3359 | /* |
| 3360 | * ext4_discard_partial_page_buffers() |
| 3361 | * Wrapper function for ext4_discard_partial_page_buffers_no_lock. |
| 3362 | * This function finds and locks the page containing the offset |
| 3363 | * "from" and passes it to ext4_discard_partial_page_buffers_no_lock. |
| 3364 | * Calling functions that already have the page locked should call |
| 3365 | * ext4_discard_partial_page_buffers_no_lock directly. |
| 3366 | */ |
| 3367 | int ext4_discard_partial_page_buffers(handle_t *handle, |
| 3368 | struct address_space *mapping, loff_t from, |
| 3369 | loff_t length, int flags) |
| 3370 | { |
| 3371 | struct inode *inode = mapping->host; |
| 3372 | struct page *page; |
| 3373 | int err = 0; |
| 3374 | |
| 3375 | page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT, |
| 3376 | mapping_gfp_mask(mapping) & ~__GFP_FS); |
| 3377 | if (!page) |
Yongqiang Yang | 5129d05 | 2011-10-31 17:56:10 -0400 | [diff] [blame] | 3378 | return -ENOMEM; |
Allison Henderson | 4e96b2d | 2011-09-03 11:51:09 -0400 | [diff] [blame] | 3379 | |
| 3380 | err = ext4_discard_partial_page_buffers_no_lock(handle, inode, page, |
| 3381 | from, length, flags); |
| 3382 | |
| 3383 | unlock_page(page); |
| 3384 | page_cache_release(page); |
| 3385 | return err; |
| 3386 | } |
| 3387 | |
| 3388 | /* |
| 3389 | * ext4_discard_partial_page_buffers_no_lock() |
| 3390 | * Zeros a page range of length 'length' starting from offset 'from'. |
| 3391 | * Buffer heads that correspond to the block aligned regions of the |
| 3392 | * zeroed range will be unmapped. Unblock aligned regions |
| 3393 | * will have the corresponding buffer head mapped if needed so that |
| 3394 | * that region of the page can be updated with the partial zero out. |
| 3395 | * |
| 3396 | * This function assumes that the page has already been locked. The |
| 3397 | * The range to be discarded must be contained with in the given page. |
| 3398 | * If the specified range exceeds the end of the page it will be shortened |
| 3399 | * to the end of the page that corresponds to 'from'. This function is |
| 3400 | * appropriate for updating a page and it buffer heads to be unmapped and |
| 3401 | * zeroed for blocks that have been either released, or are going to be |
| 3402 | * released. |
| 3403 | * |
| 3404 | * handle: The journal handle |
| 3405 | * inode: The files inode |
| 3406 | * page: A locked page that contains the offset "from" |
Anatol Pomozov | 4907cb7 | 2012-09-01 10:31:09 -0700 | [diff] [blame] | 3407 | * from: The starting byte offset (from the beginning of the file) |
Allison Henderson | 4e96b2d | 2011-09-03 11:51:09 -0400 | [diff] [blame] | 3408 | * to begin discarding |
| 3409 | * len: The length of bytes to discard |
| 3410 | * flags: Optional flags that may be used: |
| 3411 | * |
| 3412 | * EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED |
| 3413 | * Only zero the regions of the page whose buffer heads |
| 3414 | * have already been unmapped. This flag is appropriate |
Anatol Pomozov | 4907cb7 | 2012-09-01 10:31:09 -0700 | [diff] [blame] | 3415 | * for updating the contents of a page whose blocks may |
Allison Henderson | 4e96b2d | 2011-09-03 11:51:09 -0400 | [diff] [blame] | 3416 | * have already been released, and we only want to zero |
| 3417 | * out the regions that correspond to those released blocks. |
| 3418 | * |
Anatol Pomozov | 4907cb7 | 2012-09-01 10:31:09 -0700 | [diff] [blame] | 3419 | * Returns zero on success or negative on failure. |
Allison Henderson | 4e96b2d | 2011-09-03 11:51:09 -0400 | [diff] [blame] | 3420 | */ |
Eric Sandeen | 5f163cc | 2012-01-04 22:33:28 -0500 | [diff] [blame] | 3421 | static int ext4_discard_partial_page_buffers_no_lock(handle_t *handle, |
Allison Henderson | 4e96b2d | 2011-09-03 11:51:09 -0400 | [diff] [blame] | 3422 | struct inode *inode, struct page *page, loff_t from, |
| 3423 | loff_t length, int flags) |
| 3424 | { |
| 3425 | ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT; |
| 3426 | unsigned int offset = from & (PAGE_CACHE_SIZE-1); |
| 3427 | unsigned int blocksize, max, pos; |
Allison Henderson | 4e96b2d | 2011-09-03 11:51:09 -0400 | [diff] [blame] | 3428 | ext4_lblk_t iblock; |
| 3429 | struct buffer_head *bh; |
| 3430 | int err = 0; |
| 3431 | |
| 3432 | blocksize = inode->i_sb->s_blocksize; |
| 3433 | max = PAGE_CACHE_SIZE - offset; |
| 3434 | |
| 3435 | if (index != page->index) |
| 3436 | return -EINVAL; |
| 3437 | |
| 3438 | /* |
| 3439 | * correct length if it does not fall between |
| 3440 | * 'from' and the end of the page |
| 3441 | */ |
| 3442 | if (length > max || length < 0) |
| 3443 | length = max; |
| 3444 | |
| 3445 | iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits); |
| 3446 | |
Yongqiang Yang | 093e6e3 | 2011-12-13 22:05:05 -0500 | [diff] [blame] | 3447 | if (!page_has_buffers(page)) |
| 3448 | create_empty_buffers(page, blocksize, 0); |
Allison Henderson | 4e96b2d | 2011-09-03 11:51:09 -0400 | [diff] [blame] | 3449 | |
| 3450 | /* Find the buffer that contains "offset" */ |
| 3451 | bh = page_buffers(page); |
| 3452 | pos = blocksize; |
| 3453 | while (offset >= pos) { |
| 3454 | bh = bh->b_this_page; |
| 3455 | iblock++; |
| 3456 | pos += blocksize; |
| 3457 | } |
| 3458 | |
| 3459 | pos = offset; |
| 3460 | while (pos < offset + length) { |
Yongqiang Yang | e260daf | 2011-10-31 17:54:36 -0400 | [diff] [blame] | 3461 | unsigned int end_of_block, range_to_discard; |
| 3462 | |
Allison Henderson | 4e96b2d | 2011-09-03 11:51:09 -0400 | [diff] [blame] | 3463 | err = 0; |
| 3464 | |
| 3465 | /* The length of space left to zero and unmap */ |
| 3466 | range_to_discard = offset + length - pos; |
| 3467 | |
| 3468 | /* The length of space until the end of the block */ |
| 3469 | end_of_block = blocksize - (pos & (blocksize-1)); |
| 3470 | |
| 3471 | /* |
| 3472 | * Do not unmap or zero past end of block |
| 3473 | * for this buffer head |
| 3474 | */ |
| 3475 | if (range_to_discard > end_of_block) |
| 3476 | range_to_discard = end_of_block; |
| 3477 | |
| 3478 | |
| 3479 | /* |
| 3480 | * Skip this buffer head if we are only zeroing unampped |
| 3481 | * regions of the page |
| 3482 | */ |
| 3483 | if (flags & EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED && |
| 3484 | buffer_mapped(bh)) |
| 3485 | goto next; |
| 3486 | |
| 3487 | /* If the range is block aligned, unmap */ |
| 3488 | if (range_to_discard == blocksize) { |
| 3489 | clear_buffer_dirty(bh); |
| 3490 | bh->b_bdev = NULL; |
| 3491 | clear_buffer_mapped(bh); |
| 3492 | clear_buffer_req(bh); |
| 3493 | clear_buffer_new(bh); |
| 3494 | clear_buffer_delay(bh); |
| 3495 | clear_buffer_unwritten(bh); |
| 3496 | clear_buffer_uptodate(bh); |
| 3497 | zero_user(page, pos, range_to_discard); |
| 3498 | BUFFER_TRACE(bh, "Buffer discarded"); |
| 3499 | goto next; |
| 3500 | } |
| 3501 | |
| 3502 | /* |
| 3503 | * If this block is not completely contained in the range |
| 3504 | * to be discarded, then it is not going to be released. Because |
| 3505 | * we need to keep this block, we need to make sure this part |
| 3506 | * of the page is uptodate before we modify it by writeing |
| 3507 | * partial zeros on it. |
| 3508 | */ |
| 3509 | if (!buffer_mapped(bh)) { |
| 3510 | /* |
| 3511 | * Buffer head must be mapped before we can read |
| 3512 | * from the block |
| 3513 | */ |
| 3514 | BUFFER_TRACE(bh, "unmapped"); |
| 3515 | ext4_get_block(inode, iblock, bh, 0); |
| 3516 | /* unmapped? It's a hole - nothing to do */ |
| 3517 | if (!buffer_mapped(bh)) { |
| 3518 | BUFFER_TRACE(bh, "still unmapped"); |
| 3519 | goto next; |
| 3520 | } |
| 3521 | } |
| 3522 | |
| 3523 | /* Ok, it's mapped. Make sure it's up-to-date */ |
| 3524 | if (PageUptodate(page)) |
| 3525 | set_buffer_uptodate(bh); |
| 3526 | |
| 3527 | if (!buffer_uptodate(bh)) { |
| 3528 | err = -EIO; |
| 3529 | ll_rw_block(READ, 1, &bh); |
| 3530 | wait_on_buffer(bh); |
| 3531 | /* Uhhuh. Read error. Complain and punt.*/ |
| 3532 | if (!buffer_uptodate(bh)) |
| 3533 | goto next; |
| 3534 | } |
| 3535 | |
| 3536 | if (ext4_should_journal_data(inode)) { |
| 3537 | BUFFER_TRACE(bh, "get write access"); |
| 3538 | err = ext4_journal_get_write_access(handle, bh); |
| 3539 | if (err) |
| 3540 | goto next; |
| 3541 | } |
| 3542 | |
| 3543 | zero_user(page, pos, range_to_discard); |
| 3544 | |
| 3545 | err = 0; |
| 3546 | if (ext4_should_journal_data(inode)) { |
| 3547 | err = ext4_handle_dirty_metadata(handle, inode, bh); |
Theodore Ts'o | decbd91 | 2011-09-06 02:37:06 -0400 | [diff] [blame] | 3548 | } else |
Allison Henderson | 4e96b2d | 2011-09-03 11:51:09 -0400 | [diff] [blame] | 3549 | mark_buffer_dirty(bh); |
Allison Henderson | 4e96b2d | 2011-09-03 11:51:09 -0400 | [diff] [blame] | 3550 | |
| 3551 | BUFFER_TRACE(bh, "Partial buffer zeroed"); |
| 3552 | next: |
| 3553 | bh = bh->b_this_page; |
| 3554 | iblock++; |
| 3555 | pos += range_to_discard; |
| 3556 | } |
| 3557 | |
| 3558 | return err; |
| 3559 | } |
| 3560 | |
Duane Griffin | 91ef4ca | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 3561 | int ext4_can_truncate(struct inode *inode) |
| 3562 | { |
Duane Griffin | 91ef4ca | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 3563 | if (S_ISREG(inode->i_mode)) |
| 3564 | return 1; |
| 3565 | if (S_ISDIR(inode->i_mode)) |
| 3566 | return 1; |
| 3567 | if (S_ISLNK(inode->i_mode)) |
| 3568 | return !ext4_inode_is_fast_symlink(inode); |
| 3569 | return 0; |
| 3570 | } |
| 3571 | |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3572 | /* |
Allison Henderson | a4bb6b6 | 2011-05-25 07:41:50 -0400 | [diff] [blame] | 3573 | * ext4_punch_hole: punches a hole in a file by releaseing the blocks |
| 3574 | * associated with the given offset and length |
| 3575 | * |
| 3576 | * @inode: File inode |
| 3577 | * @offset: The offset where the hole will begin |
| 3578 | * @len: The length of the hole |
| 3579 | * |
Anatol Pomozov | 4907cb7 | 2012-09-01 10:31:09 -0700 | [diff] [blame] | 3580 | * Returns: 0 on success or negative on failure |
Allison Henderson | a4bb6b6 | 2011-05-25 07:41:50 -0400 | [diff] [blame] | 3581 | */ |
| 3582 | |
| 3583 | int ext4_punch_hole(struct file *file, loff_t offset, loff_t length) |
| 3584 | { |
Al Viro | 496ad9a | 2013-01-23 17:07:38 -0500 | [diff] [blame] | 3585 | struct inode *inode = file_inode(file); |
Theodore Ts'o | 26a4c0c | 2013-04-03 12:45:17 -0400 | [diff] [blame] | 3586 | struct super_block *sb = inode->i_sb; |
| 3587 | ext4_lblk_t first_block, stop_block; |
| 3588 | struct address_space *mapping = inode->i_mapping; |
| 3589 | loff_t first_page, last_page, page_len; |
| 3590 | loff_t first_page_offset, last_page_offset; |
| 3591 | handle_t *handle; |
| 3592 | unsigned int credits; |
| 3593 | int ret = 0; |
| 3594 | |
Allison Henderson | a4bb6b6 | 2011-05-25 07:41:50 -0400 | [diff] [blame] | 3595 | if (!S_ISREG(inode->i_mode)) |
Allison Henderson | 7335519 | 2012-03-21 22:23:31 -0400 | [diff] [blame] | 3596 | return -EOPNOTSUPP; |
Allison Henderson | a4bb6b6 | 2011-05-25 07:41:50 -0400 | [diff] [blame] | 3597 | |
Theodore Ts'o | 26a4c0c | 2013-04-03 12:45:17 -0400 | [diff] [blame] | 3598 | if (EXT4_SB(sb)->s_cluster_ratio > 1) { |
Theodore Ts'o | bab08ab | 2011-09-09 18:36:51 -0400 | [diff] [blame] | 3599 | /* TODO: Add support for bigalloc file systems */ |
Allison Henderson | 7335519 | 2012-03-21 22:23:31 -0400 | [diff] [blame] | 3600 | return -EOPNOTSUPP; |
Theodore Ts'o | bab08ab | 2011-09-09 18:36:51 -0400 | [diff] [blame] | 3601 | } |
| 3602 | |
Zheng Liu | aaddea8 | 2013-01-16 20:21:26 -0500 | [diff] [blame] | 3603 | trace_ext4_punch_hole(inode, offset, length); |
| 3604 | |
Theodore Ts'o | 26a4c0c | 2013-04-03 12:45:17 -0400 | [diff] [blame] | 3605 | /* |
| 3606 | * Write out all dirty pages to avoid race conditions |
| 3607 | * Then release them. |
| 3608 | */ |
| 3609 | if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { |
| 3610 | ret = filemap_write_and_wait_range(mapping, offset, |
| 3611 | offset + length - 1); |
| 3612 | if (ret) |
| 3613 | return ret; |
| 3614 | } |
| 3615 | |
| 3616 | mutex_lock(&inode->i_mutex); |
| 3617 | /* It's not possible punch hole on append only file */ |
| 3618 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) { |
| 3619 | ret = -EPERM; |
| 3620 | goto out_mutex; |
| 3621 | } |
| 3622 | if (IS_SWAPFILE(inode)) { |
| 3623 | ret = -ETXTBSY; |
| 3624 | goto out_mutex; |
| 3625 | } |
| 3626 | |
| 3627 | /* No need to punch hole beyond i_size */ |
| 3628 | if (offset >= inode->i_size) |
| 3629 | goto out_mutex; |
| 3630 | |
| 3631 | /* |
| 3632 | * If the hole extends beyond i_size, set the hole |
| 3633 | * to end after the page that contains i_size |
| 3634 | */ |
| 3635 | if (offset + length > inode->i_size) { |
| 3636 | length = inode->i_size + |
| 3637 | PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) - |
| 3638 | offset; |
| 3639 | } |
| 3640 | |
| 3641 | first_page = (offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; |
| 3642 | last_page = (offset + length) >> PAGE_CACHE_SHIFT; |
| 3643 | |
| 3644 | first_page_offset = first_page << PAGE_CACHE_SHIFT; |
| 3645 | last_page_offset = last_page << PAGE_CACHE_SHIFT; |
| 3646 | |
| 3647 | /* Now release the pages */ |
| 3648 | if (last_page_offset > first_page_offset) { |
| 3649 | truncate_pagecache_range(inode, first_page_offset, |
| 3650 | last_page_offset - 1); |
| 3651 | } |
| 3652 | |
| 3653 | /* Wait all existing dio workers, newcomers will block on i_mutex */ |
| 3654 | ext4_inode_block_unlocked_dio(inode); |
| 3655 | ret = ext4_flush_unwritten_io(inode); |
| 3656 | if (ret) |
| 3657 | goto out_dio; |
| 3658 | inode_dio_wait(inode); |
| 3659 | |
| 3660 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
| 3661 | credits = ext4_writepage_trans_blocks(inode); |
| 3662 | else |
| 3663 | credits = ext4_blocks_for_truncate(inode); |
| 3664 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); |
| 3665 | if (IS_ERR(handle)) { |
| 3666 | ret = PTR_ERR(handle); |
| 3667 | ext4_std_error(sb, ret); |
| 3668 | goto out_dio; |
| 3669 | } |
| 3670 | |
| 3671 | /* |
| 3672 | * Now we need to zero out the non-page-aligned data in the |
| 3673 | * pages at the start and tail of the hole, and unmap the |
| 3674 | * buffer heads for the block aligned regions of the page that |
| 3675 | * were completely zeroed. |
| 3676 | */ |
| 3677 | if (first_page > last_page) { |
| 3678 | /* |
| 3679 | * If the file space being truncated is contained |
| 3680 | * within a page just zero out and unmap the middle of |
| 3681 | * that page |
| 3682 | */ |
| 3683 | ret = ext4_discard_partial_page_buffers(handle, |
| 3684 | mapping, offset, length, 0); |
| 3685 | |
| 3686 | if (ret) |
| 3687 | goto out_stop; |
| 3688 | } else { |
| 3689 | /* |
| 3690 | * zero out and unmap the partial page that contains |
| 3691 | * the start of the hole |
| 3692 | */ |
| 3693 | page_len = first_page_offset - offset; |
| 3694 | if (page_len > 0) { |
| 3695 | ret = ext4_discard_partial_page_buffers(handle, mapping, |
| 3696 | offset, page_len, 0); |
| 3697 | if (ret) |
| 3698 | goto out_stop; |
| 3699 | } |
| 3700 | |
| 3701 | /* |
| 3702 | * zero out and unmap the partial page that contains |
| 3703 | * the end of the hole |
| 3704 | */ |
| 3705 | page_len = offset + length - last_page_offset; |
| 3706 | if (page_len > 0) { |
| 3707 | ret = ext4_discard_partial_page_buffers(handle, mapping, |
| 3708 | last_page_offset, page_len, 0); |
| 3709 | if (ret) |
| 3710 | goto out_stop; |
| 3711 | } |
| 3712 | } |
| 3713 | |
| 3714 | /* |
| 3715 | * If i_size is contained in the last page, we need to |
| 3716 | * unmap and zero the partial page after i_size |
| 3717 | */ |
| 3718 | if (inode->i_size >> PAGE_CACHE_SHIFT == last_page && |
| 3719 | inode->i_size % PAGE_CACHE_SIZE != 0) { |
| 3720 | page_len = PAGE_CACHE_SIZE - |
| 3721 | (inode->i_size & (PAGE_CACHE_SIZE - 1)); |
| 3722 | |
| 3723 | if (page_len > 0) { |
| 3724 | ret = ext4_discard_partial_page_buffers(handle, |
| 3725 | mapping, inode->i_size, page_len, 0); |
| 3726 | |
| 3727 | if (ret) |
| 3728 | goto out_stop; |
| 3729 | } |
| 3730 | } |
| 3731 | |
| 3732 | first_block = (offset + sb->s_blocksize - 1) >> |
| 3733 | EXT4_BLOCK_SIZE_BITS(sb); |
| 3734 | stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb); |
| 3735 | |
| 3736 | /* If there are no blocks to remove, return now */ |
| 3737 | if (first_block >= stop_block) |
| 3738 | goto out_stop; |
| 3739 | |
| 3740 | down_write(&EXT4_I(inode)->i_data_sem); |
| 3741 | ext4_discard_preallocations(inode); |
| 3742 | |
| 3743 | ret = ext4_es_remove_extent(inode, first_block, |
| 3744 | stop_block - first_block); |
| 3745 | if (ret) { |
| 3746 | up_write(&EXT4_I(inode)->i_data_sem); |
| 3747 | goto out_stop; |
| 3748 | } |
| 3749 | |
| 3750 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
| 3751 | ret = ext4_ext_remove_space(inode, first_block, |
| 3752 | stop_block - 1); |
| 3753 | else |
| 3754 | ret = ext4_free_hole_blocks(handle, inode, first_block, |
| 3755 | stop_block); |
| 3756 | |
| 3757 | ext4_discard_preallocations(inode); |
Theodore Ts'o | 819c492 | 2013-04-03 12:47:17 -0400 | [diff] [blame] | 3758 | up_write(&EXT4_I(inode)->i_data_sem); |
Theodore Ts'o | 26a4c0c | 2013-04-03 12:45:17 -0400 | [diff] [blame] | 3759 | if (IS_SYNC(inode)) |
| 3760 | ext4_handle_sync(handle); |
Theodore Ts'o | 26a4c0c | 2013-04-03 12:45:17 -0400 | [diff] [blame] | 3761 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); |
| 3762 | ext4_mark_inode_dirty(handle, inode); |
| 3763 | out_stop: |
| 3764 | ext4_journal_stop(handle); |
| 3765 | out_dio: |
| 3766 | ext4_inode_resume_unlocked_dio(inode); |
| 3767 | out_mutex: |
| 3768 | mutex_unlock(&inode->i_mutex); |
| 3769 | return ret; |
Allison Henderson | a4bb6b6 | 2011-05-25 07:41:50 -0400 | [diff] [blame] | 3770 | } |
| 3771 | |
| 3772 | /* |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3773 | * ext4_truncate() |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3774 | * |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3775 | * We block out ext4_get_block() block instantiations across the entire |
| 3776 | * transaction, and VFS/VM ensures that ext4_truncate() cannot run |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3777 | * simultaneously on behalf of the same inode. |
| 3778 | * |
Justin P. Mattock | 42b2aa8 | 2011-11-28 20:31:00 -0800 | [diff] [blame] | 3779 | * As we work through the truncate and commit bits of it to the journal there |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3780 | * is one core, guiding principle: the file's tree must always be consistent on |
| 3781 | * disk. We must be able to restart the truncate after a crash. |
| 3782 | * |
| 3783 | * The file's tree may be transiently inconsistent in memory (although it |
| 3784 | * probably isn't), but whenever we close off and commit a journal transaction, |
| 3785 | * the contents of (the filesystem + the journal) must be consistent and |
| 3786 | * restartable. It's pretty simple, really: bottom up, right to left (although |
| 3787 | * left-to-right works OK too). |
| 3788 | * |
| 3789 | * Note that at recovery time, journal replay occurs *before* the restart of |
| 3790 | * truncate against the orphan inode list. |
| 3791 | * |
| 3792 | * The committed inode has the new, desired i_size (which is the same as |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3793 | * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3794 | * that this inode's truncate did not complete and it will again call |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3795 | * ext4_truncate() to have another go. So there will be instantiated blocks |
| 3796 | * to the right of the truncation point in a crashed ext4 filesystem. But |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3797 | * that's fine - as long as they are linked from the inode, the post-crash |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3798 | * ext4_truncate() run will find them and release them. |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3799 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3800 | void ext4_truncate(struct inode *inode) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3801 | { |
Theodore Ts'o | 819c492 | 2013-04-03 12:47:17 -0400 | [diff] [blame] | 3802 | struct ext4_inode_info *ei = EXT4_I(inode); |
| 3803 | unsigned int credits; |
| 3804 | handle_t *handle; |
| 3805 | struct address_space *mapping = inode->i_mapping; |
| 3806 | loff_t page_len; |
| 3807 | |
Theodore Ts'o | 19b5ef6 | 2013-04-03 21:58:52 -0400 | [diff] [blame] | 3808 | /* |
| 3809 | * There is a possibility that we're either freeing the inode |
| 3810 | * or it completely new indode. In those cases we might not |
| 3811 | * have i_mutex locked because it's not necessary. |
| 3812 | */ |
| 3813 | if (!(inode->i_state & (I_NEW|I_FREEING))) |
| 3814 | WARN_ON(!mutex_is_locked(&inode->i_mutex)); |
Jiaying Zhang | 0562e0b | 2011-03-21 21:38:05 -0400 | [diff] [blame] | 3815 | trace_ext4_truncate_enter(inode); |
| 3816 | |
Duane Griffin | 91ef4ca | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 3817 | if (!ext4_can_truncate(inode)) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3818 | return; |
| 3819 | |
Dmitry Monakhov | 12e9b89 | 2010-05-16 22:00:00 -0400 | [diff] [blame] | 3820 | ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS); |
Jiaying Zhang | c8d46e4 | 2010-02-24 09:52:53 -0500 | [diff] [blame] | 3821 | |
Theodore Ts'o | 5534fb5 | 2009-09-17 09:34:16 -0400 | [diff] [blame] | 3822 | if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC)) |
Theodore Ts'o | 19f5fb7 | 2010-01-24 14:34:07 -0500 | [diff] [blame] | 3823 | ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE); |
Theodore Ts'o | 7d8f9f7 | 2009-02-24 08:21:14 -0500 | [diff] [blame] | 3824 | |
Tao Ma | aef1c85 | 2012-12-10 14:06:02 -0500 | [diff] [blame] | 3825 | if (ext4_has_inline_data(inode)) { |
| 3826 | int has_inline = 1; |
| 3827 | |
| 3828 | ext4_inline_data_truncate(inode, &has_inline); |
| 3829 | if (has_inline) |
| 3830 | return; |
| 3831 | } |
| 3832 | |
Theodore Ts'o | 819c492 | 2013-04-03 12:47:17 -0400 | [diff] [blame] | 3833 | /* |
| 3834 | * finish any pending end_io work so we won't run the risk of |
| 3835 | * converting any truncated blocks to initialized later |
| 3836 | */ |
| 3837 | ext4_flush_unwritten_io(inode); |
| 3838 | |
Amir Goldstein | ff9893d | 2011-06-27 16:36:31 -0400 | [diff] [blame] | 3839 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
Theodore Ts'o | 819c492 | 2013-04-03 12:47:17 -0400 | [diff] [blame] | 3840 | credits = ext4_writepage_trans_blocks(inode); |
Amir Goldstein | ff9893d | 2011-06-27 16:36:31 -0400 | [diff] [blame] | 3841 | else |
Theodore Ts'o | 819c492 | 2013-04-03 12:47:17 -0400 | [diff] [blame] | 3842 | credits = ext4_blocks_for_truncate(inode); |
| 3843 | |
| 3844 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); |
| 3845 | if (IS_ERR(handle)) { |
| 3846 | ext4_std_error(inode->i_sb, PTR_ERR(handle)); |
| 3847 | return; |
| 3848 | } |
| 3849 | |
| 3850 | if (inode->i_size % PAGE_CACHE_SIZE != 0) { |
| 3851 | page_len = PAGE_CACHE_SIZE - |
| 3852 | (inode->i_size & (PAGE_CACHE_SIZE - 1)); |
| 3853 | |
| 3854 | if (ext4_discard_partial_page_buffers(handle, |
| 3855 | mapping, inode->i_size, page_len, 0)) |
| 3856 | goto out_stop; |
| 3857 | } |
| 3858 | |
| 3859 | /* |
| 3860 | * We add the inode to the orphan list, so that if this |
| 3861 | * truncate spans multiple transactions, and we crash, we will |
| 3862 | * resume the truncate when the filesystem recovers. It also |
| 3863 | * marks the inode dirty, to catch the new size. |
| 3864 | * |
| 3865 | * Implication: the file must always be in a sane, consistent |
| 3866 | * truncatable state while each transaction commits. |
| 3867 | */ |
| 3868 | if (ext4_orphan_add(handle, inode)) |
| 3869 | goto out_stop; |
| 3870 | |
| 3871 | down_write(&EXT4_I(inode)->i_data_sem); |
| 3872 | |
| 3873 | ext4_discard_preallocations(inode); |
| 3874 | |
| 3875 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
| 3876 | ext4_ext_truncate(handle, inode); |
| 3877 | else |
| 3878 | ext4_ind_truncate(handle, inode); |
| 3879 | |
| 3880 | up_write(&ei->i_data_sem); |
| 3881 | |
| 3882 | if (IS_SYNC(inode)) |
| 3883 | ext4_handle_sync(handle); |
| 3884 | |
| 3885 | out_stop: |
| 3886 | /* |
| 3887 | * If this was a simple ftruncate() and the file will remain alive, |
| 3888 | * then we need to clear up the orphan record which we created above. |
| 3889 | * However, if this was a real unlink then we were called by |
| 3890 | * ext4_delete_inode(), and we allow that function to clean up the |
| 3891 | * orphan info for us. |
| 3892 | */ |
| 3893 | if (inode->i_nlink) |
| 3894 | ext4_orphan_del(handle, inode); |
| 3895 | |
| 3896 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); |
| 3897 | ext4_mark_inode_dirty(handle, inode); |
| 3898 | ext4_journal_stop(handle); |
Alex Tomas | a86c618 | 2006-10-11 01:21:03 -0700 | [diff] [blame] | 3899 | |
Jiaying Zhang | 0562e0b | 2011-03-21 21:38:05 -0400 | [diff] [blame] | 3900 | trace_ext4_truncate_exit(inode); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3901 | } |
| 3902 | |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3903 | /* |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3904 | * ext4_get_inode_loc returns with an extra refcount against the inode's |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3905 | * underlying buffer_head on success. If 'in_mem' is true, we have all |
| 3906 | * data in memory that is needed to recreate the on-disk version of this |
| 3907 | * inode. |
| 3908 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3909 | static int __ext4_get_inode_loc(struct inode *inode, |
| 3910 | struct ext4_iloc *iloc, int in_mem) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3911 | { |
Theodore Ts'o | 240799c | 2008-10-09 23:53:47 -0400 | [diff] [blame] | 3912 | struct ext4_group_desc *gdp; |
| 3913 | struct buffer_head *bh; |
| 3914 | struct super_block *sb = inode->i_sb; |
| 3915 | ext4_fsblk_t block; |
| 3916 | int inodes_per_block, inode_offset; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3917 | |
Aneesh Kumar K.V | 3a06d77 | 2008-11-22 15:04:59 -0500 | [diff] [blame] | 3918 | iloc->bh = NULL; |
Theodore Ts'o | 240799c | 2008-10-09 23:53:47 -0400 | [diff] [blame] | 3919 | if (!ext4_valid_inum(sb, inode->i_ino)) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3920 | return -EIO; |
| 3921 | |
Theodore Ts'o | 240799c | 2008-10-09 23:53:47 -0400 | [diff] [blame] | 3922 | iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb); |
| 3923 | gdp = ext4_get_group_desc(sb, iloc->block_group, NULL); |
| 3924 | if (!gdp) |
| 3925 | return -EIO; |
| 3926 | |
| 3927 | /* |
| 3928 | * Figure out the offset within the block group inode table |
| 3929 | */ |
Tao Ma | 00d0988 | 2011-05-09 10:26:41 -0400 | [diff] [blame] | 3930 | inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; |
Theodore Ts'o | 240799c | 2008-10-09 23:53:47 -0400 | [diff] [blame] | 3931 | inode_offset = ((inode->i_ino - 1) % |
| 3932 | EXT4_INODES_PER_GROUP(sb)); |
| 3933 | block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block); |
| 3934 | iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb); |
| 3935 | |
| 3936 | bh = sb_getblk(sb, block); |
Wang Shilong | aebf024 | 2013-01-12 16:28:47 -0500 | [diff] [blame] | 3937 | if (unlikely(!bh)) |
Theodore Ts'o | 860d21e | 2013-01-12 16:19:36 -0500 | [diff] [blame] | 3938 | return -ENOMEM; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3939 | if (!buffer_uptodate(bh)) { |
| 3940 | lock_buffer(bh); |
Hidehiro Kawai | 9c83a92 | 2008-07-26 16:39:26 -0400 | [diff] [blame] | 3941 | |
| 3942 | /* |
| 3943 | * If the buffer has the write error flag, we have failed |
| 3944 | * to write out another inode in the same block. In this |
| 3945 | * case, we don't have to read the block because we may |
| 3946 | * read the old inode data successfully. |
| 3947 | */ |
| 3948 | if (buffer_write_io_error(bh) && !buffer_uptodate(bh)) |
| 3949 | set_buffer_uptodate(bh); |
| 3950 | |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3951 | if (buffer_uptodate(bh)) { |
| 3952 | /* someone brought it uptodate while we waited */ |
| 3953 | unlock_buffer(bh); |
| 3954 | goto has_buffer; |
| 3955 | } |
| 3956 | |
| 3957 | /* |
| 3958 | * If we have all information of the inode in memory and this |
| 3959 | * is the only valid inode in the block, we need not read the |
| 3960 | * block. |
| 3961 | */ |
| 3962 | if (in_mem) { |
| 3963 | struct buffer_head *bitmap_bh; |
Theodore Ts'o | 240799c | 2008-10-09 23:53:47 -0400 | [diff] [blame] | 3964 | int i, start; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3965 | |
Theodore Ts'o | 240799c | 2008-10-09 23:53:47 -0400 | [diff] [blame] | 3966 | start = inode_offset & ~(inodes_per_block - 1); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3967 | |
| 3968 | /* Is the inode bitmap in cache? */ |
Theodore Ts'o | 240799c | 2008-10-09 23:53:47 -0400 | [diff] [blame] | 3969 | bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp)); |
Wang Shilong | aebf024 | 2013-01-12 16:28:47 -0500 | [diff] [blame] | 3970 | if (unlikely(!bitmap_bh)) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3971 | goto make_io; |
| 3972 | |
| 3973 | /* |
| 3974 | * If the inode bitmap isn't in cache then the |
| 3975 | * optimisation may end up performing two reads instead |
| 3976 | * of one, so skip it. |
| 3977 | */ |
| 3978 | if (!buffer_uptodate(bitmap_bh)) { |
| 3979 | brelse(bitmap_bh); |
| 3980 | goto make_io; |
| 3981 | } |
Theodore Ts'o | 240799c | 2008-10-09 23:53:47 -0400 | [diff] [blame] | 3982 | for (i = start; i < start + inodes_per_block; i++) { |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3983 | if (i == inode_offset) |
| 3984 | continue; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 3985 | if (ext4_test_bit(i, bitmap_bh->b_data)) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3986 | break; |
| 3987 | } |
| 3988 | brelse(bitmap_bh); |
Theodore Ts'o | 240799c | 2008-10-09 23:53:47 -0400 | [diff] [blame] | 3989 | if (i == start + inodes_per_block) { |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 3990 | /* all other inodes are free, so skip I/O */ |
| 3991 | memset(bh->b_data, 0, bh->b_size); |
| 3992 | set_buffer_uptodate(bh); |
| 3993 | unlock_buffer(bh); |
| 3994 | goto has_buffer; |
| 3995 | } |
| 3996 | } |
| 3997 | |
| 3998 | make_io: |
| 3999 | /* |
Theodore Ts'o | 240799c | 2008-10-09 23:53:47 -0400 | [diff] [blame] | 4000 | * If we need to do any I/O, try to pre-readahead extra |
| 4001 | * blocks from the inode table. |
| 4002 | */ |
| 4003 | if (EXT4_SB(sb)->s_inode_readahead_blks) { |
| 4004 | ext4_fsblk_t b, end, table; |
| 4005 | unsigned num; |
Theodore Ts'o | 0d606e2 | 2013-04-23 08:59:35 -0400 | [diff] [blame] | 4006 | __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks; |
Theodore Ts'o | 240799c | 2008-10-09 23:53:47 -0400 | [diff] [blame] | 4007 | |
| 4008 | table = ext4_inode_table(sb, gdp); |
Theodore Ts'o | b713a5e | 2009-03-31 09:11:14 -0400 | [diff] [blame] | 4009 | /* s_inode_readahead_blks is always a power of 2 */ |
Theodore Ts'o | 0d606e2 | 2013-04-23 08:59:35 -0400 | [diff] [blame] | 4010 | b = block & ~((ext4_fsblk_t) ra_blks - 1); |
Theodore Ts'o | 240799c | 2008-10-09 23:53:47 -0400 | [diff] [blame] | 4011 | if (table > b) |
| 4012 | b = table; |
Theodore Ts'o | 0d606e2 | 2013-04-23 08:59:35 -0400 | [diff] [blame] | 4013 | end = b + ra_blks; |
Theodore Ts'o | 240799c | 2008-10-09 23:53:47 -0400 | [diff] [blame] | 4014 | num = EXT4_INODES_PER_GROUP(sb); |
Darrick J. Wong | feb0ab3 | 2012-04-29 18:45:10 -0400 | [diff] [blame] | 4015 | if (ext4_has_group_desc_csum(sb)) |
Aneesh Kumar K.V | 560671a | 2009-01-05 22:20:24 -0500 | [diff] [blame] | 4016 | num -= ext4_itable_unused_count(sb, gdp); |
Theodore Ts'o | 240799c | 2008-10-09 23:53:47 -0400 | [diff] [blame] | 4017 | table += num / inodes_per_block; |
| 4018 | if (end > table) |
| 4019 | end = table; |
| 4020 | while (b <= end) |
| 4021 | sb_breadahead(sb, b++); |
| 4022 | } |
| 4023 | |
| 4024 | /* |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4025 | * There are other valid inodes in the buffer, this inode |
| 4026 | * has in-inode xattrs, or we don't have this inode in memory. |
| 4027 | * Read the block from disk. |
| 4028 | */ |
Jiaying Zhang | 0562e0b | 2011-03-21 21:38:05 -0400 | [diff] [blame] | 4029 | trace_ext4_load_inode(inode); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4030 | get_bh(bh); |
| 4031 | bh->b_end_io = end_buffer_read_sync; |
Christoph Hellwig | 65299a3 | 2011-08-23 14:50:29 +0200 | [diff] [blame] | 4032 | submit_bh(READ | REQ_META | REQ_PRIO, bh); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4033 | wait_on_buffer(bh); |
| 4034 | if (!buffer_uptodate(bh)) { |
Theodore Ts'o | c398eda | 2010-07-27 11:56:40 -0400 | [diff] [blame] | 4035 | EXT4_ERROR_INODE_BLOCK(inode, block, |
| 4036 | "unable to read itable block"); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4037 | brelse(bh); |
| 4038 | return -EIO; |
| 4039 | } |
| 4040 | } |
| 4041 | has_buffer: |
| 4042 | iloc->bh = bh; |
| 4043 | return 0; |
| 4044 | } |
| 4045 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 4046 | int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4047 | { |
| 4048 | /* We have all inode data except xattrs in memory here. */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 4049 | return __ext4_get_inode_loc(inode, iloc, |
Theodore Ts'o | 19f5fb7 | 2010-01-24 14:34:07 -0500 | [diff] [blame] | 4050 | !ext4_test_inode_state(inode, EXT4_STATE_XATTR)); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4051 | } |
| 4052 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 4053 | void ext4_set_inode_flags(struct inode *inode) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4054 | { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 4055 | unsigned int flags = EXT4_I(inode)->i_flags; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4056 | |
| 4057 | inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC); |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 4058 | if (flags & EXT4_SYNC_FL) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4059 | inode->i_flags |= S_SYNC; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 4060 | if (flags & EXT4_APPEND_FL) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4061 | inode->i_flags |= S_APPEND; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 4062 | if (flags & EXT4_IMMUTABLE_FL) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4063 | inode->i_flags |= S_IMMUTABLE; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 4064 | if (flags & EXT4_NOATIME_FL) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4065 | inode->i_flags |= S_NOATIME; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 4066 | if (flags & EXT4_DIRSYNC_FL) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4067 | inode->i_flags |= S_DIRSYNC; |
| 4068 | } |
| 4069 | |
Jan Kara | ff9ddf7 | 2007-07-18 09:24:20 -0400 | [diff] [blame] | 4070 | /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */ |
| 4071 | void ext4_get_inode_flags(struct ext4_inode_info *ei) |
| 4072 | { |
Dmitry Monakhov | 84a8dce | 2010-06-05 11:51:27 -0400 | [diff] [blame] | 4073 | unsigned int vfs_fl; |
| 4074 | unsigned long old_fl, new_fl; |
Jan Kara | ff9ddf7 | 2007-07-18 09:24:20 -0400 | [diff] [blame] | 4075 | |
Dmitry Monakhov | 84a8dce | 2010-06-05 11:51:27 -0400 | [diff] [blame] | 4076 | do { |
| 4077 | vfs_fl = ei->vfs_inode.i_flags; |
| 4078 | old_fl = ei->i_flags; |
| 4079 | new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL| |
| 4080 | EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL| |
| 4081 | EXT4_DIRSYNC_FL); |
| 4082 | if (vfs_fl & S_SYNC) |
| 4083 | new_fl |= EXT4_SYNC_FL; |
| 4084 | if (vfs_fl & S_APPEND) |
| 4085 | new_fl |= EXT4_APPEND_FL; |
| 4086 | if (vfs_fl & S_IMMUTABLE) |
| 4087 | new_fl |= EXT4_IMMUTABLE_FL; |
| 4088 | if (vfs_fl & S_NOATIME) |
| 4089 | new_fl |= EXT4_NOATIME_FL; |
| 4090 | if (vfs_fl & S_DIRSYNC) |
| 4091 | new_fl |= EXT4_DIRSYNC_FL; |
| 4092 | } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl); |
Jan Kara | ff9ddf7 | 2007-07-18 09:24:20 -0400 | [diff] [blame] | 4093 | } |
Theodore Ts'o | de9a55b | 2009-06-14 17:45:34 -0400 | [diff] [blame] | 4094 | |
Aneesh Kumar K.V | 0fc1b45 | 2008-01-28 23:58:26 -0500 | [diff] [blame] | 4095 | static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode, |
Theodore Ts'o | de9a55b | 2009-06-14 17:45:34 -0400 | [diff] [blame] | 4096 | struct ext4_inode_info *ei) |
Aneesh Kumar K.V | 0fc1b45 | 2008-01-28 23:58:26 -0500 | [diff] [blame] | 4097 | { |
| 4098 | blkcnt_t i_blocks ; |
Aneesh Kumar K.V | 8180a56 | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 4099 | struct inode *inode = &(ei->vfs_inode); |
| 4100 | struct super_block *sb = inode->i_sb; |
Aneesh Kumar K.V | 0fc1b45 | 2008-01-28 23:58:26 -0500 | [diff] [blame] | 4101 | |
| 4102 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, |
| 4103 | EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) { |
| 4104 | /* we are using combined 48 bit field */ |
| 4105 | i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 | |
| 4106 | le32_to_cpu(raw_inode->i_blocks_lo); |
Theodore Ts'o | 07a0382 | 2010-06-14 09:54:48 -0400 | [diff] [blame] | 4107 | if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) { |
Aneesh Kumar K.V | 8180a56 | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 4108 | /* i_blocks represent file system block size */ |
| 4109 | return i_blocks << (inode->i_blkbits - 9); |
| 4110 | } else { |
| 4111 | return i_blocks; |
| 4112 | } |
Aneesh Kumar K.V | 0fc1b45 | 2008-01-28 23:58:26 -0500 | [diff] [blame] | 4113 | } else { |
| 4114 | return le32_to_cpu(raw_inode->i_blocks_lo); |
| 4115 | } |
| 4116 | } |
Jan Kara | ff9ddf7 | 2007-07-18 09:24:20 -0400 | [diff] [blame] | 4117 | |
Tao Ma | 152a7b0 | 2012-12-02 11:13:24 -0500 | [diff] [blame] | 4118 | static inline void ext4_iget_extra_inode(struct inode *inode, |
| 4119 | struct ext4_inode *raw_inode, |
| 4120 | struct ext4_inode_info *ei) |
| 4121 | { |
| 4122 | __le32 *magic = (void *)raw_inode + |
| 4123 | EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize; |
Tao Ma | 67cf5b0 | 2012-12-10 14:04:46 -0500 | [diff] [blame] | 4124 | if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) { |
Tao Ma | 152a7b0 | 2012-12-02 11:13:24 -0500 | [diff] [blame] | 4125 | ext4_set_inode_state(inode, EXT4_STATE_XATTR); |
Tao Ma | 67cf5b0 | 2012-12-10 14:04:46 -0500 | [diff] [blame] | 4126 | ext4_find_inline_data_nolock(inode); |
Tao Ma | f19d587 | 2012-12-10 14:05:51 -0500 | [diff] [blame] | 4127 | } else |
| 4128 | EXT4_I(inode)->i_inline_off = 0; |
Tao Ma | 152a7b0 | 2012-12-02 11:13:24 -0500 | [diff] [blame] | 4129 | } |
| 4130 | |
David Howells | 1d1fe1e | 2008-02-07 00:15:37 -0800 | [diff] [blame] | 4131 | struct inode *ext4_iget(struct super_block *sb, unsigned long ino) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4132 | { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 4133 | struct ext4_iloc iloc; |
| 4134 | struct ext4_inode *raw_inode; |
David Howells | 1d1fe1e | 2008-02-07 00:15:37 -0800 | [diff] [blame] | 4135 | struct ext4_inode_info *ei; |
David Howells | 1d1fe1e | 2008-02-07 00:15:37 -0800 | [diff] [blame] | 4136 | struct inode *inode; |
Jan Kara | b436b9b | 2009-12-08 23:51:10 -0500 | [diff] [blame] | 4137 | journal_t *journal = EXT4_SB(sb)->s_journal; |
David Howells | 1d1fe1e | 2008-02-07 00:15:37 -0800 | [diff] [blame] | 4138 | long ret; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4139 | int block; |
Eric W. Biederman | 08cefc7 | 2012-02-07 15:41:49 -0800 | [diff] [blame] | 4140 | uid_t i_uid; |
| 4141 | gid_t i_gid; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4142 | |
David Howells | 1d1fe1e | 2008-02-07 00:15:37 -0800 | [diff] [blame] | 4143 | inode = iget_locked(sb, ino); |
| 4144 | if (!inode) |
| 4145 | return ERR_PTR(-ENOMEM); |
| 4146 | if (!(inode->i_state & I_NEW)) |
| 4147 | return inode; |
| 4148 | |
| 4149 | ei = EXT4_I(inode); |
Peter Huewe | 7dc5761 | 2011-02-21 21:01:42 -0500 | [diff] [blame] | 4150 | iloc.bh = NULL; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4151 | |
David Howells | 1d1fe1e | 2008-02-07 00:15:37 -0800 | [diff] [blame] | 4152 | ret = __ext4_get_inode_loc(inode, &iloc, 0); |
| 4153 | if (ret < 0) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4154 | goto bad_inode; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 4155 | raw_inode = ext4_raw_inode(&iloc); |
Darrick J. Wong | 814525f | 2012-04-29 18:31:10 -0400 | [diff] [blame] | 4156 | |
| 4157 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
| 4158 | ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize); |
| 4159 | if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > |
| 4160 | EXT4_INODE_SIZE(inode->i_sb)) { |
| 4161 | EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)", |
| 4162 | EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize, |
| 4163 | EXT4_INODE_SIZE(inode->i_sb)); |
| 4164 | ret = -EIO; |
| 4165 | goto bad_inode; |
| 4166 | } |
| 4167 | } else |
| 4168 | ei->i_extra_isize = 0; |
| 4169 | |
| 4170 | /* Precompute checksum seed for inode metadata */ |
| 4171 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, |
| 4172 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) { |
| 4173 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
| 4174 | __u32 csum; |
| 4175 | __le32 inum = cpu_to_le32(inode->i_ino); |
| 4176 | __le32 gen = raw_inode->i_generation; |
| 4177 | csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum, |
| 4178 | sizeof(inum)); |
| 4179 | ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen, |
| 4180 | sizeof(gen)); |
| 4181 | } |
| 4182 | |
| 4183 | if (!ext4_inode_csum_verify(inode, raw_inode, ei)) { |
| 4184 | EXT4_ERROR_INODE(inode, "checksum invalid"); |
| 4185 | ret = -EIO; |
| 4186 | goto bad_inode; |
| 4187 | } |
| 4188 | |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4189 | inode->i_mode = le16_to_cpu(raw_inode->i_mode); |
Eric W. Biederman | 08cefc7 | 2012-02-07 15:41:49 -0800 | [diff] [blame] | 4190 | i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low); |
| 4191 | i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low); |
Theodore Ts'o | af5bc92 | 2008-09-08 22:25:24 -0400 | [diff] [blame] | 4192 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
Eric W. Biederman | 08cefc7 | 2012-02-07 15:41:49 -0800 | [diff] [blame] | 4193 | i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16; |
| 4194 | i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4195 | } |
Eric W. Biederman | 08cefc7 | 2012-02-07 15:41:49 -0800 | [diff] [blame] | 4196 | i_uid_write(inode, i_uid); |
| 4197 | i_gid_write(inode, i_gid); |
Miklos Szeredi | bfe8684 | 2011-10-28 14:13:29 +0200 | [diff] [blame] | 4198 | set_nlink(inode, le16_to_cpu(raw_inode->i_links_count)); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4199 | |
Theodore Ts'o | 353eb83 | 2011-01-10 12:18:25 -0500 | [diff] [blame] | 4200 | ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */ |
Tao Ma | 67cf5b0 | 2012-12-10 14:04:46 -0500 | [diff] [blame] | 4201 | ei->i_inline_off = 0; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4202 | ei->i_dir_start_lookup = 0; |
| 4203 | ei->i_dtime = le32_to_cpu(raw_inode->i_dtime); |
| 4204 | /* We now have enough fields to check if the inode was active or not. |
| 4205 | * This is needed because nfsd might try to access dead inodes |
| 4206 | * the test is that same one that e2fsck uses |
| 4207 | * NeilBrown 1999oct15 |
| 4208 | */ |
| 4209 | if (inode->i_nlink == 0) { |
Dr. Tilmann Bubeck | 393d1d1 | 2013-04-08 12:54:05 -0400 | [diff] [blame] | 4210 | if ((inode->i_mode == 0 || |
| 4211 | !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) && |
| 4212 | ino != EXT4_BOOT_LOADER_INO) { |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4213 | /* this inode is deleted */ |
David Howells | 1d1fe1e | 2008-02-07 00:15:37 -0800 | [diff] [blame] | 4214 | ret = -ESTALE; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4215 | goto bad_inode; |
| 4216 | } |
| 4217 | /* The only unlinked inodes we let through here have |
| 4218 | * valid i_mode and are being read by the orphan |
| 4219 | * recovery code: that's fine, we're about to complete |
Dr. Tilmann Bubeck | 393d1d1 | 2013-04-08 12:54:05 -0400 | [diff] [blame] | 4220 | * the process of deleting those. |
| 4221 | * OR it is the EXT4_BOOT_LOADER_INO which is |
| 4222 | * not initialized on a new filesystem. */ |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4223 | } |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4224 | ei->i_flags = le32_to_cpu(raw_inode->i_flags); |
Aneesh Kumar K.V | 0fc1b45 | 2008-01-28 23:58:26 -0500 | [diff] [blame] | 4225 | inode->i_blocks = ext4_inode_blocks(raw_inode, ei); |
Aneesh Kumar K.V | 7973c0c | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 4226 | ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo); |
Theodore Ts'o | a9e8174 | 2009-04-24 16:11:18 -0400 | [diff] [blame] | 4227 | if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) |
Badari Pulavarty | a1ddeb7 | 2006-10-11 01:21:09 -0700 | [diff] [blame] | 4228 | ei->i_file_acl |= |
| 4229 | ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32; |
Aneesh Kumar K.V | a48380f | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 4230 | inode->i_size = ext4_isize(raw_inode); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4231 | ei->i_disksize = inode->i_size; |
Dmitry Monakhov | a9e7f44 | 2009-12-14 15:21:14 +0300 | [diff] [blame] | 4232 | #ifdef CONFIG_QUOTA |
| 4233 | ei->i_reserved_quota = 0; |
| 4234 | #endif |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4235 | inode->i_generation = le32_to_cpu(raw_inode->i_generation); |
| 4236 | ei->i_block_group = iloc.block_group; |
Theodore Ts'o | a491212 | 2009-03-12 12:18:34 -0400 | [diff] [blame] | 4237 | ei->i_last_alloc_group = ~0; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4238 | /* |
| 4239 | * NOTE! The in-memory inode i_data array is in little-endian order |
| 4240 | * even on big-endian machines: we do NOT byteswap the block numbers! |
| 4241 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 4242 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4243 | ei->i_data[block] = raw_inode->i_block[block]; |
| 4244 | INIT_LIST_HEAD(&ei->i_orphan); |
| 4245 | |
Jan Kara | b436b9b | 2009-12-08 23:51:10 -0500 | [diff] [blame] | 4246 | /* |
| 4247 | * Set transaction id's of transactions that have to be committed |
| 4248 | * to finish f[data]sync. We set them to currently running transaction |
| 4249 | * as we cannot be sure that the inode or some of its metadata isn't |
| 4250 | * part of the transaction - the inode could have been reclaimed and |
| 4251 | * now it is reread from disk. |
| 4252 | */ |
| 4253 | if (journal) { |
| 4254 | transaction_t *transaction; |
| 4255 | tid_t tid; |
| 4256 | |
Theodore Ts'o | a931da6 | 2010-08-03 21:35:12 -0400 | [diff] [blame] | 4257 | read_lock(&journal->j_state_lock); |
Jan Kara | b436b9b | 2009-12-08 23:51:10 -0500 | [diff] [blame] | 4258 | if (journal->j_running_transaction) |
| 4259 | transaction = journal->j_running_transaction; |
| 4260 | else |
| 4261 | transaction = journal->j_committing_transaction; |
| 4262 | if (transaction) |
| 4263 | tid = transaction->t_tid; |
| 4264 | else |
| 4265 | tid = journal->j_commit_sequence; |
Theodore Ts'o | a931da6 | 2010-08-03 21:35:12 -0400 | [diff] [blame] | 4266 | read_unlock(&journal->j_state_lock); |
Jan Kara | b436b9b | 2009-12-08 23:51:10 -0500 | [diff] [blame] | 4267 | ei->i_sync_tid = tid; |
| 4268 | ei->i_datasync_tid = tid; |
| 4269 | } |
| 4270 | |
Eric Sandeen | 0040d98 | 2008-02-05 22:36:43 -0500 | [diff] [blame] | 4271 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4272 | if (ei->i_extra_isize == 0) { |
| 4273 | /* The extra space is currently unused. Use it. */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 4274 | ei->i_extra_isize = sizeof(struct ext4_inode) - |
| 4275 | EXT4_GOOD_OLD_INODE_SIZE; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4276 | } else { |
Tao Ma | 152a7b0 | 2012-12-02 11:13:24 -0500 | [diff] [blame] | 4277 | ext4_iget_extra_inode(inode, raw_inode, ei); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4278 | } |
Darrick J. Wong | 814525f | 2012-04-29 18:31:10 -0400 | [diff] [blame] | 4279 | } |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4280 | |
Kalpak Shah | ef7f383 | 2007-07-18 09:15:20 -0400 | [diff] [blame] | 4281 | EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode); |
| 4282 | EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode); |
| 4283 | EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode); |
| 4284 | EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode); |
| 4285 | |
Jean Noel Cordenner | 25ec56b | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 4286 | inode->i_version = le32_to_cpu(raw_inode->i_disk_version); |
| 4287 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
| 4288 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) |
| 4289 | inode->i_version |= |
| 4290 | (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32; |
| 4291 | } |
| 4292 | |
Theodore Ts'o | c4b5a61 | 2009-04-24 18:45:35 -0400 | [diff] [blame] | 4293 | ret = 0; |
Theodore Ts'o | 485c26e | 2009-04-24 13:43:20 -0400 | [diff] [blame] | 4294 | if (ei->i_file_acl && |
Theodore Ts'o | 1032988 | 2009-11-15 15:29:56 -0500 | [diff] [blame] | 4295 | !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) { |
Theodore Ts'o | 24676da | 2010-05-16 21:00:00 -0400 | [diff] [blame] | 4296 | EXT4_ERROR_INODE(inode, "bad extended attribute block %llu", |
| 4297 | ei->i_file_acl); |
Theodore Ts'o | 485c26e | 2009-04-24 13:43:20 -0400 | [diff] [blame] | 4298 | ret = -EIO; |
| 4299 | goto bad_inode; |
Tao Ma | f19d587 | 2012-12-10 14:05:51 -0500 | [diff] [blame] | 4300 | } else if (!ext4_has_inline_data(inode)) { |
| 4301 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
| 4302 | if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || |
| 4303 | (S_ISLNK(inode->i_mode) && |
| 4304 | !ext4_inode_is_fast_symlink(inode)))) |
| 4305 | /* Validate extent which is part of inode */ |
| 4306 | ret = ext4_ext_check_inode(inode); |
| 4307 | } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || |
| 4308 | (S_ISLNK(inode->i_mode) && |
| 4309 | !ext4_inode_is_fast_symlink(inode))) { |
| 4310 | /* Validate block references which are part of inode */ |
| 4311 | ret = ext4_ind_check_inode(inode); |
| 4312 | } |
Thiemo Nagel | fe2c819 | 2009-03-31 08:36:10 -0400 | [diff] [blame] | 4313 | } |
Theodore Ts'o | 567f3e9 | 2009-11-14 08:19:05 -0500 | [diff] [blame] | 4314 | if (ret) |
Theodore Ts'o | de9a55b | 2009-06-14 17:45:34 -0400 | [diff] [blame] | 4315 | goto bad_inode; |
Aneesh Kumar K.V | 7a262f7 | 2009-03-27 16:39:58 -0400 | [diff] [blame] | 4316 | |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4317 | if (S_ISREG(inode->i_mode)) { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 4318 | inode->i_op = &ext4_file_inode_operations; |
| 4319 | inode->i_fop = &ext4_file_operations; |
| 4320 | ext4_set_aops(inode); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4321 | } else if (S_ISDIR(inode->i_mode)) { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 4322 | inode->i_op = &ext4_dir_inode_operations; |
| 4323 | inode->i_fop = &ext4_dir_operations; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4324 | } else if (S_ISLNK(inode->i_mode)) { |
Duane Griffin | e83c139 | 2008-12-19 20:47:15 +0000 | [diff] [blame] | 4325 | if (ext4_inode_is_fast_symlink(inode)) { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 4326 | inode->i_op = &ext4_fast_symlink_inode_operations; |
Duane Griffin | e83c139 | 2008-12-19 20:47:15 +0000 | [diff] [blame] | 4327 | nd_terminate_link(ei->i_data, inode->i_size, |
| 4328 | sizeof(ei->i_data) - 1); |
| 4329 | } else { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 4330 | inode->i_op = &ext4_symlink_inode_operations; |
| 4331 | ext4_set_aops(inode); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4332 | } |
Theodore Ts'o | 563bdd6 | 2009-03-26 00:06:19 -0400 | [diff] [blame] | 4333 | } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) || |
| 4334 | S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 4335 | inode->i_op = &ext4_special_inode_operations; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4336 | if (raw_inode->i_block[0]) |
| 4337 | init_special_inode(inode, inode->i_mode, |
| 4338 | old_decode_dev(le32_to_cpu(raw_inode->i_block[0]))); |
| 4339 | else |
| 4340 | init_special_inode(inode, inode->i_mode, |
| 4341 | new_decode_dev(le32_to_cpu(raw_inode->i_block[1]))); |
Dr. Tilmann Bubeck | 393d1d1 | 2013-04-08 12:54:05 -0400 | [diff] [blame] | 4342 | } else if (ino == EXT4_BOOT_LOADER_INO) { |
| 4343 | make_bad_inode(inode); |
Theodore Ts'o | 563bdd6 | 2009-03-26 00:06:19 -0400 | [diff] [blame] | 4344 | } else { |
Theodore Ts'o | 563bdd6 | 2009-03-26 00:06:19 -0400 | [diff] [blame] | 4345 | ret = -EIO; |
Theodore Ts'o | 24676da | 2010-05-16 21:00:00 -0400 | [diff] [blame] | 4346 | EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode); |
Theodore Ts'o | 563bdd6 | 2009-03-26 00:06:19 -0400 | [diff] [blame] | 4347 | goto bad_inode; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4348 | } |
Theodore Ts'o | af5bc92 | 2008-09-08 22:25:24 -0400 | [diff] [blame] | 4349 | brelse(iloc.bh); |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 4350 | ext4_set_inode_flags(inode); |
David Howells | 1d1fe1e | 2008-02-07 00:15:37 -0800 | [diff] [blame] | 4351 | unlock_new_inode(inode); |
| 4352 | return inode; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4353 | |
| 4354 | bad_inode: |
Theodore Ts'o | 567f3e9 | 2009-11-14 08:19:05 -0500 | [diff] [blame] | 4355 | brelse(iloc.bh); |
David Howells | 1d1fe1e | 2008-02-07 00:15:37 -0800 | [diff] [blame] | 4356 | iget_failed(inode); |
| 4357 | return ERR_PTR(ret); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4358 | } |
| 4359 | |
Aneesh Kumar K.V | 0fc1b45 | 2008-01-28 23:58:26 -0500 | [diff] [blame] | 4360 | static int ext4_inode_blocks_set(handle_t *handle, |
| 4361 | struct ext4_inode *raw_inode, |
| 4362 | struct ext4_inode_info *ei) |
| 4363 | { |
| 4364 | struct inode *inode = &(ei->vfs_inode); |
| 4365 | u64 i_blocks = inode->i_blocks; |
| 4366 | struct super_block *sb = inode->i_sb; |
Aneesh Kumar K.V | 0fc1b45 | 2008-01-28 23:58:26 -0500 | [diff] [blame] | 4367 | |
| 4368 | if (i_blocks <= ~0U) { |
| 4369 | /* |
Anatol Pomozov | 4907cb7 | 2012-09-01 10:31:09 -0700 | [diff] [blame] | 4370 | * i_blocks can be represented in a 32 bit variable |
Aneesh Kumar K.V | 0fc1b45 | 2008-01-28 23:58:26 -0500 | [diff] [blame] | 4371 | * as multiple of 512 bytes |
| 4372 | */ |
Aneesh Kumar K.V | 8180a56 | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 4373 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
Aneesh Kumar K.V | 0fc1b45 | 2008-01-28 23:58:26 -0500 | [diff] [blame] | 4374 | raw_inode->i_blocks_high = 0; |
Dmitry Monakhov | 84a8dce | 2010-06-05 11:51:27 -0400 | [diff] [blame] | 4375 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
Theodore Ts'o | f287a1a | 2008-10-16 22:50:48 -0400 | [diff] [blame] | 4376 | return 0; |
| 4377 | } |
| 4378 | if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) |
| 4379 | return -EFBIG; |
| 4380 | |
| 4381 | if (i_blocks <= 0xffffffffffffULL) { |
Aneesh Kumar K.V | 0fc1b45 | 2008-01-28 23:58:26 -0500 | [diff] [blame] | 4382 | /* |
| 4383 | * i_blocks can be represented in a 48 bit variable |
| 4384 | * as multiple of 512 bytes |
| 4385 | */ |
Aneesh Kumar K.V | 8180a56 | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 4386 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
Aneesh Kumar K.V | 0fc1b45 | 2008-01-28 23:58:26 -0500 | [diff] [blame] | 4387 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); |
Dmitry Monakhov | 84a8dce | 2010-06-05 11:51:27 -0400 | [diff] [blame] | 4388 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
Aneesh Kumar K.V | 0fc1b45 | 2008-01-28 23:58:26 -0500 | [diff] [blame] | 4389 | } else { |
Dmitry Monakhov | 84a8dce | 2010-06-05 11:51:27 -0400 | [diff] [blame] | 4390 | ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
Aneesh Kumar K.V | 8180a56 | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 4391 | /* i_block is stored in file system block size */ |
| 4392 | i_blocks = i_blocks >> (inode->i_blkbits - 9); |
| 4393 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
| 4394 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); |
Aneesh Kumar K.V | 0fc1b45 | 2008-01-28 23:58:26 -0500 | [diff] [blame] | 4395 | } |
Theodore Ts'o | f287a1a | 2008-10-16 22:50:48 -0400 | [diff] [blame] | 4396 | return 0; |
Aneesh Kumar K.V | 0fc1b45 | 2008-01-28 23:58:26 -0500 | [diff] [blame] | 4397 | } |
| 4398 | |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4399 | /* |
| 4400 | * Post the struct inode info into an on-disk inode location in the |
| 4401 | * buffer-cache. This gobbles the caller's reference to the |
| 4402 | * buffer_head in the inode location struct. |
| 4403 | * |
| 4404 | * The caller must have write access to iloc->bh. |
| 4405 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 4406 | static int ext4_do_update_inode(handle_t *handle, |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4407 | struct inode *inode, |
Frank Mayhar | 830156c | 2009-09-29 10:07:47 -0400 | [diff] [blame] | 4408 | struct ext4_iloc *iloc) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4409 | { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 4410 | struct ext4_inode *raw_inode = ext4_raw_inode(iloc); |
| 4411 | struct ext4_inode_info *ei = EXT4_I(inode); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4412 | struct buffer_head *bh = iloc->bh; |
| 4413 | int err = 0, rc, block; |
Jan Kara | b71fc07 | 2012-09-26 21:52:20 -0400 | [diff] [blame] | 4414 | int need_datasync = 0; |
Eric W. Biederman | 08cefc7 | 2012-02-07 15:41:49 -0800 | [diff] [blame] | 4415 | uid_t i_uid; |
| 4416 | gid_t i_gid; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4417 | |
| 4418 | /* For fields not not tracking in the in-memory inode, |
| 4419 | * initialise them to zero for new inodes. */ |
Theodore Ts'o | 19f5fb7 | 2010-01-24 14:34:07 -0500 | [diff] [blame] | 4420 | if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 4421 | memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4422 | |
Jan Kara | ff9ddf7 | 2007-07-18 09:24:20 -0400 | [diff] [blame] | 4423 | ext4_get_inode_flags(ei); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4424 | raw_inode->i_mode = cpu_to_le16(inode->i_mode); |
Eric W. Biederman | 08cefc7 | 2012-02-07 15:41:49 -0800 | [diff] [blame] | 4425 | i_uid = i_uid_read(inode); |
| 4426 | i_gid = i_gid_read(inode); |
Theodore Ts'o | af5bc92 | 2008-09-08 22:25:24 -0400 | [diff] [blame] | 4427 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
Eric W. Biederman | 08cefc7 | 2012-02-07 15:41:49 -0800 | [diff] [blame] | 4428 | raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid)); |
| 4429 | raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid)); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4430 | /* |
| 4431 | * Fix up interoperability with old kernels. Otherwise, old inodes get |
| 4432 | * re-used with the upper 16 bits of the uid/gid intact |
| 4433 | */ |
Theodore Ts'o | af5bc92 | 2008-09-08 22:25:24 -0400 | [diff] [blame] | 4434 | if (!ei->i_dtime) { |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4435 | raw_inode->i_uid_high = |
Eric W. Biederman | 08cefc7 | 2012-02-07 15:41:49 -0800 | [diff] [blame] | 4436 | cpu_to_le16(high_16_bits(i_uid)); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4437 | raw_inode->i_gid_high = |
Eric W. Biederman | 08cefc7 | 2012-02-07 15:41:49 -0800 | [diff] [blame] | 4438 | cpu_to_le16(high_16_bits(i_gid)); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4439 | } else { |
| 4440 | raw_inode->i_uid_high = 0; |
| 4441 | raw_inode->i_gid_high = 0; |
| 4442 | } |
| 4443 | } else { |
Eric W. Biederman | 08cefc7 | 2012-02-07 15:41:49 -0800 | [diff] [blame] | 4444 | raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid)); |
| 4445 | raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid)); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4446 | raw_inode->i_uid_high = 0; |
| 4447 | raw_inode->i_gid_high = 0; |
| 4448 | } |
| 4449 | raw_inode->i_links_count = cpu_to_le16(inode->i_nlink); |
Kalpak Shah | ef7f383 | 2007-07-18 09:15:20 -0400 | [diff] [blame] | 4450 | |
| 4451 | EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode); |
| 4452 | EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode); |
| 4453 | EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode); |
| 4454 | EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode); |
| 4455 | |
Aneesh Kumar K.V | 0fc1b45 | 2008-01-28 23:58:26 -0500 | [diff] [blame] | 4456 | if (ext4_inode_blocks_set(handle, raw_inode, ei)) |
| 4457 | goto out_brelse; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4458 | raw_inode->i_dtime = cpu_to_le32(ei->i_dtime); |
Theodore Ts'o | 353eb83 | 2011-01-10 12:18:25 -0500 | [diff] [blame] | 4459 | raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF); |
Mingming Cao | 9b8f1f0 | 2006-10-11 01:21:13 -0700 | [diff] [blame] | 4460 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != |
| 4461 | cpu_to_le32(EXT4_OS_HURD)) |
Badari Pulavarty | a1ddeb7 | 2006-10-11 01:21:09 -0700 | [diff] [blame] | 4462 | raw_inode->i_file_acl_high = |
| 4463 | cpu_to_le16(ei->i_file_acl >> 32); |
Aneesh Kumar K.V | 7973c0c | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 4464 | raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl); |
Jan Kara | b71fc07 | 2012-09-26 21:52:20 -0400 | [diff] [blame] | 4465 | if (ei->i_disksize != ext4_isize(raw_inode)) { |
| 4466 | ext4_isize_set(raw_inode, ei->i_disksize); |
| 4467 | need_datasync = 1; |
| 4468 | } |
Aneesh Kumar K.V | a48380f | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 4469 | if (ei->i_disksize > 0x7fffffffULL) { |
| 4470 | struct super_block *sb = inode->i_sb; |
| 4471 | if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, |
| 4472 | EXT4_FEATURE_RO_COMPAT_LARGE_FILE) || |
| 4473 | EXT4_SB(sb)->s_es->s_rev_level == |
| 4474 | cpu_to_le32(EXT4_GOOD_OLD_REV)) { |
| 4475 | /* If this is the first large file |
| 4476 | * created, add a flag to the superblock. |
| 4477 | */ |
| 4478 | err = ext4_journal_get_write_access(handle, |
| 4479 | EXT4_SB(sb)->s_sbh); |
| 4480 | if (err) |
| 4481 | goto out_brelse; |
| 4482 | ext4_update_dynamic_rev(sb); |
| 4483 | EXT4_SET_RO_COMPAT_FEATURE(sb, |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 4484 | EXT4_FEATURE_RO_COMPAT_LARGE_FILE); |
Frank Mayhar | 0390131 | 2009-01-07 00:06:22 -0500 | [diff] [blame] | 4485 | ext4_handle_sync(handle); |
Artem Bityutskiy | b50924c | 2012-07-22 20:37:31 -0400 | [diff] [blame] | 4486 | err = ext4_handle_dirty_super(handle, sb); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4487 | } |
| 4488 | } |
| 4489 | raw_inode->i_generation = cpu_to_le32(inode->i_generation); |
| 4490 | if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) { |
| 4491 | if (old_valid_dev(inode->i_rdev)) { |
| 4492 | raw_inode->i_block[0] = |
| 4493 | cpu_to_le32(old_encode_dev(inode->i_rdev)); |
| 4494 | raw_inode->i_block[1] = 0; |
| 4495 | } else { |
| 4496 | raw_inode->i_block[0] = 0; |
| 4497 | raw_inode->i_block[1] = |
| 4498 | cpu_to_le32(new_encode_dev(inode->i_rdev)); |
| 4499 | raw_inode->i_block[2] = 0; |
| 4500 | } |
Tao Ma | f19d587 | 2012-12-10 14:05:51 -0500 | [diff] [blame] | 4501 | } else if (!ext4_has_inline_data(inode)) { |
Theodore Ts'o | de9a55b | 2009-06-14 17:45:34 -0400 | [diff] [blame] | 4502 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
| 4503 | raw_inode->i_block[block] = ei->i_data[block]; |
Tao Ma | f19d587 | 2012-12-10 14:05:51 -0500 | [diff] [blame] | 4504 | } |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4505 | |
Jean Noel Cordenner | 25ec56b | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 4506 | raw_inode->i_disk_version = cpu_to_le32(inode->i_version); |
| 4507 | if (ei->i_extra_isize) { |
| 4508 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) |
| 4509 | raw_inode->i_version_hi = |
| 4510 | cpu_to_le32(inode->i_version >> 32); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4511 | raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize); |
Jean Noel Cordenner | 25ec56b | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 4512 | } |
| 4513 | |
Darrick J. Wong | 814525f | 2012-04-29 18:31:10 -0400 | [diff] [blame] | 4514 | ext4_inode_csum_set(inode, raw_inode, ei); |
| 4515 | |
Frank Mayhar | 830156c | 2009-09-29 10:07:47 -0400 | [diff] [blame] | 4516 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
Curt Wohlgemuth | 73b50c1 | 2010-02-16 15:06:29 -0500 | [diff] [blame] | 4517 | rc = ext4_handle_dirty_metadata(handle, NULL, bh); |
Frank Mayhar | 830156c | 2009-09-29 10:07:47 -0400 | [diff] [blame] | 4518 | if (!err) |
| 4519 | err = rc; |
Theodore Ts'o | 19f5fb7 | 2010-01-24 14:34:07 -0500 | [diff] [blame] | 4520 | ext4_clear_inode_state(inode, EXT4_STATE_NEW); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4521 | |
Jan Kara | b71fc07 | 2012-09-26 21:52:20 -0400 | [diff] [blame] | 4522 | ext4_update_inode_fsync_trans(handle, inode, need_datasync); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4523 | out_brelse: |
Theodore Ts'o | af5bc92 | 2008-09-08 22:25:24 -0400 | [diff] [blame] | 4524 | brelse(bh); |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 4525 | ext4_std_error(inode->i_sb, err); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4526 | return err; |
| 4527 | } |
| 4528 | |
| 4529 | /* |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 4530 | * ext4_write_inode() |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4531 | * |
| 4532 | * We are called from a few places: |
| 4533 | * |
| 4534 | * - Within generic_file_write() for O_SYNC files. |
| 4535 | * Here, there will be no transaction running. We wait for any running |
Anatol Pomozov | 4907cb7 | 2012-09-01 10:31:09 -0700 | [diff] [blame] | 4536 | * transaction to commit. |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4537 | * |
| 4538 | * - Within sys_sync(), kupdate and such. |
| 4539 | * We wait on commit, if tol to. |
| 4540 | * |
| 4541 | * - Within prune_icache() (PF_MEMALLOC == true) |
| 4542 | * Here we simply return. We can't afford to block kswapd on the |
| 4543 | * journal commit. |
| 4544 | * |
| 4545 | * In all cases it is actually safe for us to return without doing anything, |
| 4546 | * because the inode has been copied into a raw inode buffer in |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 4547 | * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4548 | * knfsd. |
| 4549 | * |
| 4550 | * Note that we are absolutely dependent upon all inode dirtiers doing the |
| 4551 | * right thing: they *must* call mark_inode_dirty() after dirtying info in |
| 4552 | * which we are interested. |
| 4553 | * |
| 4554 | * It would be a bug for them to not do this. The code: |
| 4555 | * |
| 4556 | * mark_inode_dirty(inode) |
| 4557 | * stuff(); |
| 4558 | * inode->i_size = expr; |
| 4559 | * |
| 4560 | * is in error because a kswapd-driven write_inode() could occur while |
| 4561 | * `stuff()' is running, and the new i_size will be lost. Plus the inode |
| 4562 | * will no longer be on the superblock's dirty inode list. |
| 4563 | */ |
Christoph Hellwig | a9185b4 | 2010-03-05 09:21:37 +0100 | [diff] [blame] | 4564 | int ext4_write_inode(struct inode *inode, struct writeback_control *wbc) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4565 | { |
Frank Mayhar | 91ac6f4 | 2009-09-09 22:33:47 -0400 | [diff] [blame] | 4566 | int err; |
| 4567 | |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4568 | if (current->flags & PF_MEMALLOC) |
| 4569 | return 0; |
| 4570 | |
Frank Mayhar | 91ac6f4 | 2009-09-09 22:33:47 -0400 | [diff] [blame] | 4571 | if (EXT4_SB(inode->i_sb)->s_journal) { |
| 4572 | if (ext4_journal_current_handle()) { |
| 4573 | jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n"); |
| 4574 | dump_stack(); |
| 4575 | return -EIO; |
| 4576 | } |
| 4577 | |
Christoph Hellwig | a9185b4 | 2010-03-05 09:21:37 +0100 | [diff] [blame] | 4578 | if (wbc->sync_mode != WB_SYNC_ALL) |
Frank Mayhar | 91ac6f4 | 2009-09-09 22:33:47 -0400 | [diff] [blame] | 4579 | return 0; |
| 4580 | |
| 4581 | err = ext4_force_commit(inode->i_sb); |
| 4582 | } else { |
| 4583 | struct ext4_iloc iloc; |
| 4584 | |
Curt Wohlgemuth | 8b472d7 | 2010-04-03 16:45:06 -0400 | [diff] [blame] | 4585 | err = __ext4_get_inode_loc(inode, &iloc, 0); |
Frank Mayhar | 91ac6f4 | 2009-09-09 22:33:47 -0400 | [diff] [blame] | 4586 | if (err) |
| 4587 | return err; |
Christoph Hellwig | a9185b4 | 2010-03-05 09:21:37 +0100 | [diff] [blame] | 4588 | if (wbc->sync_mode == WB_SYNC_ALL) |
Frank Mayhar | 830156c | 2009-09-29 10:07:47 -0400 | [diff] [blame] | 4589 | sync_dirty_buffer(iloc.bh); |
| 4590 | if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) { |
Theodore Ts'o | c398eda | 2010-07-27 11:56:40 -0400 | [diff] [blame] | 4591 | EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr, |
| 4592 | "IO error syncing inode"); |
Frank Mayhar | 830156c | 2009-09-29 10:07:47 -0400 | [diff] [blame] | 4593 | err = -EIO; |
| 4594 | } |
Curt Wohlgemuth | fd2dd9f | 2010-04-03 17:44:16 -0400 | [diff] [blame] | 4595 | brelse(iloc.bh); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4596 | } |
Frank Mayhar | 91ac6f4 | 2009-09-09 22:33:47 -0400 | [diff] [blame] | 4597 | return err; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4598 | } |
| 4599 | |
| 4600 | /* |
Jan Kara | 53e8726 | 2012-12-25 13:29:52 -0500 | [diff] [blame] | 4601 | * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate |
| 4602 | * buffers that are attached to a page stradding i_size and are undergoing |
| 4603 | * commit. In that case we have to wait for commit to finish and try again. |
| 4604 | */ |
| 4605 | static void ext4_wait_for_tail_page_commit(struct inode *inode) |
| 4606 | { |
| 4607 | struct page *page; |
| 4608 | unsigned offset; |
| 4609 | journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; |
| 4610 | tid_t commit_tid = 0; |
| 4611 | int ret; |
| 4612 | |
| 4613 | offset = inode->i_size & (PAGE_CACHE_SIZE - 1); |
| 4614 | /* |
| 4615 | * All buffers in the last page remain valid? Then there's nothing to |
| 4616 | * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE == |
| 4617 | * blocksize case |
| 4618 | */ |
| 4619 | if (offset > PAGE_CACHE_SIZE - (1 << inode->i_blkbits)) |
| 4620 | return; |
| 4621 | while (1) { |
| 4622 | page = find_lock_page(inode->i_mapping, |
| 4623 | inode->i_size >> PAGE_CACHE_SHIFT); |
| 4624 | if (!page) |
| 4625 | return; |
| 4626 | ret = __ext4_journalled_invalidatepage(page, offset); |
| 4627 | unlock_page(page); |
| 4628 | page_cache_release(page); |
| 4629 | if (ret != -EBUSY) |
| 4630 | return; |
| 4631 | commit_tid = 0; |
| 4632 | read_lock(&journal->j_state_lock); |
| 4633 | if (journal->j_committing_transaction) |
| 4634 | commit_tid = journal->j_committing_transaction->t_tid; |
| 4635 | read_unlock(&journal->j_state_lock); |
| 4636 | if (commit_tid) |
| 4637 | jbd2_log_wait_commit(journal, commit_tid); |
| 4638 | } |
| 4639 | } |
| 4640 | |
| 4641 | /* |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 4642 | * ext4_setattr() |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4643 | * |
| 4644 | * Called from notify_change. |
| 4645 | * |
| 4646 | * We want to trap VFS attempts to truncate the file as soon as |
| 4647 | * possible. In particular, we want to make sure that when the VFS |
| 4648 | * shrinks i_size, we put the inode on the orphan list and modify |
| 4649 | * i_disksize immediately, so that during the subsequent flushing of |
| 4650 | * dirty pages and freeing of disk blocks, we can guarantee that any |
| 4651 | * commit will leave the blocks being flushed in an unused state on |
| 4652 | * disk. (On recovery, the inode will get truncated and the blocks will |
| 4653 | * be freed, so we have a strong guarantee that no future commit will |
| 4654 | * leave these blocks visible to the user.) |
| 4655 | * |
Jan Kara | 678aaf4 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 4656 | * Another thing we have to assure is that if we are in ordered mode |
| 4657 | * and inode is still attached to the committing transaction, we must |
| 4658 | * we start writeout of all the dirty pages which are being truncated. |
| 4659 | * This way we are sure that all the data written in the previous |
| 4660 | * transaction are already on disk (truncate waits for pages under |
| 4661 | * writeback). |
| 4662 | * |
| 4663 | * Called with inode->i_mutex down. |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4664 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 4665 | int ext4_setattr(struct dentry *dentry, struct iattr *attr) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4666 | { |
| 4667 | struct inode *inode = dentry->d_inode; |
| 4668 | int error, rc = 0; |
Dmitry Monakhov | 3d287de | 2010-10-27 22:08:46 -0400 | [diff] [blame] | 4669 | int orphan = 0; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4670 | const unsigned int ia_valid = attr->ia_valid; |
| 4671 | |
| 4672 | error = inode_change_ok(inode, attr); |
| 4673 | if (error) |
| 4674 | return error; |
| 4675 | |
Dmitry Monakhov | 1275562 | 2010-04-08 22:04:20 +0400 | [diff] [blame] | 4676 | if (is_quota_modification(inode, attr)) |
Christoph Hellwig | 871a293 | 2010-03-03 09:05:07 -0500 | [diff] [blame] | 4677 | dquot_initialize(inode); |
Eric W. Biederman | 08cefc7 | 2012-02-07 15:41:49 -0800 | [diff] [blame] | 4678 | if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) || |
| 4679 | (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) { |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4680 | handle_t *handle; |
| 4681 | |
| 4682 | /* (user+group)*(old+new) structure, inode write (sb, |
| 4683 | * inode block, ? - but truncate inode update has it) */ |
Theodore Ts'o | 9924a92 | 2013-02-08 21:59:22 -0500 | [diff] [blame] | 4684 | handle = ext4_journal_start(inode, EXT4_HT_QUOTA, |
| 4685 | (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) + |
| 4686 | EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4687 | if (IS_ERR(handle)) { |
| 4688 | error = PTR_ERR(handle); |
| 4689 | goto err_out; |
| 4690 | } |
Christoph Hellwig | b43fa82 | 2010-03-03 09:05:03 -0500 | [diff] [blame] | 4691 | error = dquot_transfer(inode, attr); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4692 | if (error) { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 4693 | ext4_journal_stop(handle); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4694 | return error; |
| 4695 | } |
| 4696 | /* Update corresponding info in inode so that everything is in |
| 4697 | * one transaction */ |
| 4698 | if (attr->ia_valid & ATTR_UID) |
| 4699 | inode->i_uid = attr->ia_uid; |
| 4700 | if (attr->ia_valid & ATTR_GID) |
| 4701 | inode->i_gid = attr->ia_gid; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 4702 | error = ext4_mark_inode_dirty(handle, inode); |
| 4703 | ext4_journal_stop(handle); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4704 | } |
| 4705 | |
Eric Sandeen | e2b4657 | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 4706 | if (attr->ia_valid & ATTR_SIZE) { |
Christoph Hellwig | 562c72a | 2011-06-24 14:29:45 -0400 | [diff] [blame] | 4707 | |
Dmitry Monakhov | 12e9b89 | 2010-05-16 22:00:00 -0400 | [diff] [blame] | 4708 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) { |
Eric Sandeen | e2b4657 | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 4709 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
| 4710 | |
Theodore Ts'o | 0c095c7 | 2010-07-27 11:56:06 -0400 | [diff] [blame] | 4711 | if (attr->ia_size > sbi->s_bitmap_maxbytes) |
| 4712 | return -EFBIG; |
Eric Sandeen | e2b4657 | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 4713 | } |
| 4714 | } |
| 4715 | |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4716 | if (S_ISREG(inode->i_mode) && |
Jiaying Zhang | c8d46e4 | 2010-02-24 09:52:53 -0500 | [diff] [blame] | 4717 | attr->ia_valid & ATTR_SIZE && |
Theodore Ts'o | 072bd7e | 2011-05-23 15:13:02 -0400 | [diff] [blame] | 4718 | (attr->ia_size < inode->i_size)) { |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4719 | handle_t *handle; |
| 4720 | |
Theodore Ts'o | 9924a92 | 2013-02-08 21:59:22 -0500 | [diff] [blame] | 4721 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 3); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4722 | if (IS_ERR(handle)) { |
| 4723 | error = PTR_ERR(handle); |
| 4724 | goto err_out; |
| 4725 | } |
Dmitry Monakhov | 3d287de | 2010-10-27 22:08:46 -0400 | [diff] [blame] | 4726 | if (ext4_handle_valid(handle)) { |
| 4727 | error = ext4_orphan_add(handle, inode); |
| 4728 | orphan = 1; |
| 4729 | } |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 4730 | EXT4_I(inode)->i_disksize = attr->ia_size; |
| 4731 | rc = ext4_mark_inode_dirty(handle, inode); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4732 | if (!error) |
| 4733 | error = rc; |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 4734 | ext4_journal_stop(handle); |
Jan Kara | 678aaf4 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 4735 | |
| 4736 | if (ext4_should_order_data(inode)) { |
| 4737 | error = ext4_begin_ordered_truncate(inode, |
| 4738 | attr->ia_size); |
| 4739 | if (error) { |
| 4740 | /* Do as much error cleanup as possible */ |
Theodore Ts'o | 9924a92 | 2013-02-08 21:59:22 -0500 | [diff] [blame] | 4741 | handle = ext4_journal_start(inode, |
| 4742 | EXT4_HT_INODE, 3); |
Jan Kara | 678aaf4 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 4743 | if (IS_ERR(handle)) { |
| 4744 | ext4_orphan_del(NULL, inode); |
| 4745 | goto err_out; |
| 4746 | } |
| 4747 | ext4_orphan_del(handle, inode); |
Dmitry Monakhov | 3d287de | 2010-10-27 22:08:46 -0400 | [diff] [blame] | 4748 | orphan = 0; |
Jan Kara | 678aaf4 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 4749 | ext4_journal_stop(handle); |
| 4750 | goto err_out; |
| 4751 | } |
| 4752 | } |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4753 | } |
| 4754 | |
Theodore Ts'o | 072bd7e | 2011-05-23 15:13:02 -0400 | [diff] [blame] | 4755 | if (attr->ia_valid & ATTR_SIZE) { |
Jan Kara | 53e8726 | 2012-12-25 13:29:52 -0500 | [diff] [blame] | 4756 | if (attr->ia_size != inode->i_size) { |
| 4757 | loff_t oldsize = inode->i_size; |
| 4758 | |
| 4759 | i_size_write(inode, attr->ia_size); |
| 4760 | /* |
| 4761 | * Blocks are going to be removed from the inode. Wait |
| 4762 | * for dio in flight. Temporarily disable |
| 4763 | * dioread_nolock to prevent livelock. |
| 4764 | */ |
Dmitry Monakhov | 1b65007 | 2012-09-29 00:56:15 -0400 | [diff] [blame] | 4765 | if (orphan) { |
Jan Kara | 53e8726 | 2012-12-25 13:29:52 -0500 | [diff] [blame] | 4766 | if (!ext4_should_journal_data(inode)) { |
| 4767 | ext4_inode_block_unlocked_dio(inode); |
| 4768 | inode_dio_wait(inode); |
| 4769 | ext4_inode_resume_unlocked_dio(inode); |
| 4770 | } else |
| 4771 | ext4_wait_for_tail_page_commit(inode); |
Dmitry Monakhov | 1b65007 | 2012-09-29 00:56:15 -0400 | [diff] [blame] | 4772 | } |
Jan Kara | 53e8726 | 2012-12-25 13:29:52 -0500 | [diff] [blame] | 4773 | /* |
| 4774 | * Truncate pagecache after we've waited for commit |
| 4775 | * in data=journal mode to make pages freeable. |
| 4776 | */ |
| 4777 | truncate_pagecache(inode, oldsize, inode->i_size); |
Dmitry Monakhov | 1c9114f | 2012-09-29 00:55:23 -0400 | [diff] [blame] | 4778 | } |
Lukas Czerner | afcff5d | 2012-03-21 21:47:55 -0400 | [diff] [blame] | 4779 | ext4_truncate(inode); |
Theodore Ts'o | 072bd7e | 2011-05-23 15:13:02 -0400 | [diff] [blame] | 4780 | } |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4781 | |
Christoph Hellwig | 1025774 | 2010-06-04 11:30:02 +0200 | [diff] [blame] | 4782 | if (!rc) { |
| 4783 | setattr_copy(inode, attr); |
| 4784 | mark_inode_dirty(inode); |
| 4785 | } |
| 4786 | |
| 4787 | /* |
| 4788 | * If the call to ext4_truncate failed to get a transaction handle at |
| 4789 | * all, we need to clean up the in-core orphan list manually. |
| 4790 | */ |
Dmitry Monakhov | 3d287de | 2010-10-27 22:08:46 -0400 | [diff] [blame] | 4791 | if (orphan && inode->i_nlink) |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 4792 | ext4_orphan_del(NULL, inode); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4793 | |
| 4794 | if (!rc && (ia_valid & ATTR_MODE)) |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 4795 | rc = ext4_acl_chmod(inode); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4796 | |
| 4797 | err_out: |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 4798 | ext4_std_error(inode->i_sb, error); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4799 | if (!error) |
| 4800 | error = rc; |
| 4801 | return error; |
| 4802 | } |
| 4803 | |
Mingming Cao | 3e3398a | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 4804 | int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry, |
| 4805 | struct kstat *stat) |
| 4806 | { |
| 4807 | struct inode *inode; |
| 4808 | unsigned long delalloc_blocks; |
| 4809 | |
| 4810 | inode = dentry->d_inode; |
| 4811 | generic_fillattr(inode, stat); |
| 4812 | |
| 4813 | /* |
| 4814 | * We can't update i_blocks if the block allocation is delayed |
| 4815 | * otherwise in the case of system crash before the real block |
| 4816 | * allocation is done, we will have i_blocks inconsistent with |
| 4817 | * on-disk file blocks. |
| 4818 | * We always keep i_blocks updated together with real |
| 4819 | * allocation. But to not confuse with user, stat |
| 4820 | * will return the blocks that include the delayed allocation |
| 4821 | * blocks for this file. |
| 4822 | */ |
Tao Ma | 9660755 | 2012-05-31 22:54:16 -0400 | [diff] [blame] | 4823 | delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb), |
| 4824 | EXT4_I(inode)->i_reserved_data_blocks); |
Mingming Cao | 3e3398a | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 4825 | |
| 4826 | stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9; |
| 4827 | return 0; |
| 4828 | } |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4829 | |
Mingming Cao | a02908f | 2008-08-19 22:16:07 -0400 | [diff] [blame] | 4830 | static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk) |
| 4831 | { |
Dmitry Monakhov | 12e9b89 | 2010-05-16 22:00:00 -0400 | [diff] [blame] | 4832 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) |
Amir Goldstein | 8bb2b24 | 2011-06-27 17:10:28 -0400 | [diff] [blame] | 4833 | return ext4_ind_trans_blocks(inode, nrblocks, chunk); |
Theodore Ts'o | ac51d83 | 2008-11-06 16:49:36 -0500 | [diff] [blame] | 4834 | return ext4_ext_index_trans_blocks(inode, nrblocks, chunk); |
Mingming Cao | a02908f | 2008-08-19 22:16:07 -0400 | [diff] [blame] | 4835 | } |
Theodore Ts'o | ac51d83 | 2008-11-06 16:49:36 -0500 | [diff] [blame] | 4836 | |
Mingming Cao | a02908f | 2008-08-19 22:16:07 -0400 | [diff] [blame] | 4837 | /* |
| 4838 | * Account for index blocks, block groups bitmaps and block group |
| 4839 | * descriptor blocks if modify datablocks and index blocks |
| 4840 | * worse case, the indexs blocks spread over different block groups |
| 4841 | * |
| 4842 | * If datablocks are discontiguous, they are possible to spread over |
Anatol Pomozov | 4907cb7 | 2012-09-01 10:31:09 -0700 | [diff] [blame] | 4843 | * different block groups too. If they are contiguous, with flexbg, |
Mingming Cao | a02908f | 2008-08-19 22:16:07 -0400 | [diff] [blame] | 4844 | * they could still across block group boundary. |
| 4845 | * |
| 4846 | * Also account for superblock, inode, quota and xattr blocks |
| 4847 | */ |
Theodore Ts'o | 1f109d5 | 2010-10-27 21:30:14 -0400 | [diff] [blame] | 4848 | static int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk) |
Mingming Cao | a02908f | 2008-08-19 22:16:07 -0400 | [diff] [blame] | 4849 | { |
Theodore Ts'o | 8df9675 | 2009-05-01 08:50:38 -0400 | [diff] [blame] | 4850 | ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb); |
| 4851 | int gdpblocks; |
Mingming Cao | a02908f | 2008-08-19 22:16:07 -0400 | [diff] [blame] | 4852 | int idxblocks; |
| 4853 | int ret = 0; |
| 4854 | |
| 4855 | /* |
| 4856 | * How many index blocks need to touch to modify nrblocks? |
| 4857 | * The "Chunk" flag indicating whether the nrblocks is |
| 4858 | * physically contiguous on disk |
| 4859 | * |
| 4860 | * For Direct IO and fallocate, they calls get_block to allocate |
| 4861 | * one single extent at a time, so they could set the "Chunk" flag |
| 4862 | */ |
| 4863 | idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk); |
| 4864 | |
| 4865 | ret = idxblocks; |
| 4866 | |
| 4867 | /* |
| 4868 | * Now let's see how many group bitmaps and group descriptors need |
| 4869 | * to account |
| 4870 | */ |
| 4871 | groups = idxblocks; |
| 4872 | if (chunk) |
| 4873 | groups += 1; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4874 | else |
Mingming Cao | a02908f | 2008-08-19 22:16:07 -0400 | [diff] [blame] | 4875 | groups += nrblocks; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4876 | |
Mingming Cao | a02908f | 2008-08-19 22:16:07 -0400 | [diff] [blame] | 4877 | gdpblocks = groups; |
Theodore Ts'o | 8df9675 | 2009-05-01 08:50:38 -0400 | [diff] [blame] | 4878 | if (groups > ngroups) |
| 4879 | groups = ngroups; |
Mingming Cao | a02908f | 2008-08-19 22:16:07 -0400 | [diff] [blame] | 4880 | if (groups > EXT4_SB(inode->i_sb)->s_gdb_count) |
| 4881 | gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count; |
| 4882 | |
| 4883 | /* bitmaps and block group descriptor blocks */ |
| 4884 | ret += groups + gdpblocks; |
| 4885 | |
| 4886 | /* Blocks for super block, inode, quota and xattr blocks */ |
| 4887 | ret += EXT4_META_TRANS_BLOCKS(inode->i_sb); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4888 | |
| 4889 | return ret; |
| 4890 | } |
| 4891 | |
| 4892 | /* |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 4893 | * Calculate the total number of credits to reserve to fit |
Mingming Cao | f3bd1f3 | 2008-08-19 22:16:03 -0400 | [diff] [blame] | 4894 | * the modification of a single pages into a single transaction, |
| 4895 | * which may include multiple chunks of block allocations. |
Mingming Cao | a02908f | 2008-08-19 22:16:07 -0400 | [diff] [blame] | 4896 | * |
Mingming Cao | 525f4ed | 2008-08-19 22:15:58 -0400 | [diff] [blame] | 4897 | * This could be called via ext4_write_begin() |
Mingming Cao | a02908f | 2008-08-19 22:16:07 -0400 | [diff] [blame] | 4898 | * |
Mingming Cao | 525f4ed | 2008-08-19 22:15:58 -0400 | [diff] [blame] | 4899 | * We need to consider the worse case, when |
Mingming Cao | a02908f | 2008-08-19 22:16:07 -0400 | [diff] [blame] | 4900 | * one new block per extent. |
Mingming Cao | a02908f | 2008-08-19 22:16:07 -0400 | [diff] [blame] | 4901 | */ |
| 4902 | int ext4_writepage_trans_blocks(struct inode *inode) |
| 4903 | { |
| 4904 | int bpp = ext4_journal_blocks_per_page(inode); |
| 4905 | int ret; |
| 4906 | |
| 4907 | ret = ext4_meta_trans_blocks(inode, bpp, 0); |
| 4908 | |
| 4909 | /* Account for data blocks for journalled mode */ |
| 4910 | if (ext4_should_journal_data(inode)) |
| 4911 | ret += bpp; |
| 4912 | return ret; |
| 4913 | } |
Mingming Cao | f3bd1f3 | 2008-08-19 22:16:03 -0400 | [diff] [blame] | 4914 | |
| 4915 | /* |
| 4916 | * Calculate the journal credits for a chunk of data modification. |
| 4917 | * |
| 4918 | * This is called from DIO, fallocate or whoever calling |
Eric Sandeen | 79e8303 | 2010-07-27 11:56:07 -0400 | [diff] [blame] | 4919 | * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks. |
Mingming Cao | f3bd1f3 | 2008-08-19 22:16:03 -0400 | [diff] [blame] | 4920 | * |
| 4921 | * journal buffers for data blocks are not included here, as DIO |
| 4922 | * and fallocate do no need to journal data buffers. |
| 4923 | */ |
| 4924 | int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks) |
| 4925 | { |
| 4926 | return ext4_meta_trans_blocks(inode, nrblocks, 1); |
| 4927 | } |
| 4928 | |
Mingming Cao | a02908f | 2008-08-19 22:16:07 -0400 | [diff] [blame] | 4929 | /* |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 4930 | * The caller must have previously called ext4_reserve_inode_write(). |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4931 | * Give this, we know that the caller already has write access to iloc->bh. |
| 4932 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 4933 | int ext4_mark_iloc_dirty(handle_t *handle, |
Theodore Ts'o | de9a55b | 2009-06-14 17:45:34 -0400 | [diff] [blame] | 4934 | struct inode *inode, struct ext4_iloc *iloc) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4935 | { |
| 4936 | int err = 0; |
| 4937 | |
Theodore Ts'o | c64db50 | 2012-03-02 12:23:11 -0500 | [diff] [blame] | 4938 | if (IS_I_VERSION(inode)) |
Jean Noel Cordenner | 25ec56b | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 4939 | inode_inc_iversion(inode); |
| 4940 | |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4941 | /* the do_update_inode consumes one bh->b_count */ |
| 4942 | get_bh(iloc->bh); |
| 4943 | |
Mingming Cao | dab291a | 2006-10-11 01:21:01 -0700 | [diff] [blame] | 4944 | /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */ |
Frank Mayhar | 830156c | 2009-09-29 10:07:47 -0400 | [diff] [blame] | 4945 | err = ext4_do_update_inode(handle, inode, iloc); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4946 | put_bh(iloc->bh); |
| 4947 | return err; |
| 4948 | } |
| 4949 | |
| 4950 | /* |
| 4951 | * On success, We end up with an outstanding reference count against |
| 4952 | * iloc->bh. This _must_ be cleaned up later. |
| 4953 | */ |
| 4954 | |
| 4955 | int |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 4956 | ext4_reserve_inode_write(handle_t *handle, struct inode *inode, |
| 4957 | struct ext4_iloc *iloc) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4958 | { |
Frank Mayhar | 0390131 | 2009-01-07 00:06:22 -0500 | [diff] [blame] | 4959 | int err; |
| 4960 | |
| 4961 | err = ext4_get_inode_loc(inode, iloc); |
| 4962 | if (!err) { |
| 4963 | BUFFER_TRACE(iloc->bh, "get_write_access"); |
| 4964 | err = ext4_journal_get_write_access(handle, iloc->bh); |
| 4965 | if (err) { |
| 4966 | brelse(iloc->bh); |
| 4967 | iloc->bh = NULL; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4968 | } |
| 4969 | } |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 4970 | ext4_std_error(inode->i_sb, err); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 4971 | return err; |
| 4972 | } |
| 4973 | |
| 4974 | /* |
Kalpak Shah | 6dd4ee7 | 2007-07-18 09:19:57 -0400 | [diff] [blame] | 4975 | * Expand an inode by new_extra_isize bytes. |
| 4976 | * Returns 0 on success or negative error number on failure. |
| 4977 | */ |
Aneesh Kumar K.V | 1d03ec9 | 2008-01-28 23:58:27 -0500 | [diff] [blame] | 4978 | static int ext4_expand_extra_isize(struct inode *inode, |
| 4979 | unsigned int new_extra_isize, |
| 4980 | struct ext4_iloc iloc, |
| 4981 | handle_t *handle) |
Kalpak Shah | 6dd4ee7 | 2007-07-18 09:19:57 -0400 | [diff] [blame] | 4982 | { |
| 4983 | struct ext4_inode *raw_inode; |
| 4984 | struct ext4_xattr_ibody_header *header; |
Kalpak Shah | 6dd4ee7 | 2007-07-18 09:19:57 -0400 | [diff] [blame] | 4985 | |
| 4986 | if (EXT4_I(inode)->i_extra_isize >= new_extra_isize) |
| 4987 | return 0; |
| 4988 | |
| 4989 | raw_inode = ext4_raw_inode(&iloc); |
| 4990 | |
| 4991 | header = IHDR(inode, raw_inode); |
Kalpak Shah | 6dd4ee7 | 2007-07-18 09:19:57 -0400 | [diff] [blame] | 4992 | |
| 4993 | /* No extended attributes present */ |
Theodore Ts'o | 19f5fb7 | 2010-01-24 14:34:07 -0500 | [diff] [blame] | 4994 | if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) || |
| 4995 | header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) { |
Kalpak Shah | 6dd4ee7 | 2007-07-18 09:19:57 -0400 | [diff] [blame] | 4996 | memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0, |
| 4997 | new_extra_isize); |
| 4998 | EXT4_I(inode)->i_extra_isize = new_extra_isize; |
| 4999 | return 0; |
| 5000 | } |
| 5001 | |
| 5002 | /* try to expand with EAs present */ |
| 5003 | return ext4_expand_extra_isize_ea(inode, new_extra_isize, |
| 5004 | raw_inode, handle); |
| 5005 | } |
| 5006 | |
| 5007 | /* |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 5008 | * What we do here is to mark the in-core inode as clean with respect to inode |
| 5009 | * dirtiness (it may still be data-dirty). |
| 5010 | * This means that the in-core inode may be reaped by prune_icache |
| 5011 | * without having to perform any I/O. This is a very good thing, |
| 5012 | * because *any* task may call prune_icache - even ones which |
| 5013 | * have a transaction open against a different journal. |
| 5014 | * |
| 5015 | * Is this cheating? Not really. Sure, we haven't written the |
| 5016 | * inode out, but prune_icache isn't a user-visible syncing function. |
| 5017 | * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync) |
| 5018 | * we start and wait on commits. |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 5019 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 5020 | int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 5021 | { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 5022 | struct ext4_iloc iloc; |
Kalpak Shah | 6dd4ee7 | 2007-07-18 09:19:57 -0400 | [diff] [blame] | 5023 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
| 5024 | static unsigned int mnt_count; |
| 5025 | int err, ret; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 5026 | |
| 5027 | might_sleep(); |
Theodore Ts'o | 7ff9c07 | 2010-11-08 13:51:33 -0500 | [diff] [blame] | 5028 | trace_ext4_mark_inode_dirty(inode, _RET_IP_); |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 5029 | err = ext4_reserve_inode_write(handle, inode, &iloc); |
Frank Mayhar | 0390131 | 2009-01-07 00:06:22 -0500 | [diff] [blame] | 5030 | if (ext4_handle_valid(handle) && |
| 5031 | EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize && |
Theodore Ts'o | 19f5fb7 | 2010-01-24 14:34:07 -0500 | [diff] [blame] | 5032 | !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) { |
Kalpak Shah | 6dd4ee7 | 2007-07-18 09:19:57 -0400 | [diff] [blame] | 5033 | /* |
| 5034 | * We need extra buffer credits since we may write into EA block |
| 5035 | * with this same handle. If journal_extend fails, then it will |
| 5036 | * only result in a minor loss of functionality for that inode. |
| 5037 | * If this is felt to be critical, then e2fsck should be run to |
| 5038 | * force a large enough s_min_extra_isize. |
| 5039 | */ |
| 5040 | if ((jbd2_journal_extend(handle, |
| 5041 | EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) { |
| 5042 | ret = ext4_expand_extra_isize(inode, |
| 5043 | sbi->s_want_extra_isize, |
| 5044 | iloc, handle); |
| 5045 | if (ret) { |
Theodore Ts'o | 19f5fb7 | 2010-01-24 14:34:07 -0500 | [diff] [blame] | 5046 | ext4_set_inode_state(inode, |
| 5047 | EXT4_STATE_NO_EXPAND); |
Aneesh Kumar K.V | c1bddad | 2007-10-16 18:38:25 -0400 | [diff] [blame] | 5048 | if (mnt_count != |
| 5049 | le16_to_cpu(sbi->s_es->s_mnt_count)) { |
Eric Sandeen | 12062dd | 2010-02-15 14:19:27 -0500 | [diff] [blame] | 5050 | ext4_warning(inode->i_sb, |
Kalpak Shah | 6dd4ee7 | 2007-07-18 09:19:57 -0400 | [diff] [blame] | 5051 | "Unable to expand inode %lu. Delete" |
| 5052 | " some EAs or run e2fsck.", |
| 5053 | inode->i_ino); |
Aneesh Kumar K.V | c1bddad | 2007-10-16 18:38:25 -0400 | [diff] [blame] | 5054 | mnt_count = |
| 5055 | le16_to_cpu(sbi->s_es->s_mnt_count); |
Kalpak Shah | 6dd4ee7 | 2007-07-18 09:19:57 -0400 | [diff] [blame] | 5056 | } |
| 5057 | } |
| 5058 | } |
| 5059 | } |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 5060 | if (!err) |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 5061 | err = ext4_mark_iloc_dirty(handle, inode, &iloc); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 5062 | return err; |
| 5063 | } |
| 5064 | |
| 5065 | /* |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 5066 | * ext4_dirty_inode() is called from __mark_inode_dirty() |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 5067 | * |
| 5068 | * We're really interested in the case where a file is being extended. |
| 5069 | * i_size has been changed by generic_commit_write() and we thus need |
| 5070 | * to include the updated inode in the current transaction. |
| 5071 | * |
Christoph Hellwig | 5dd4056 | 2010-03-03 09:05:00 -0500 | [diff] [blame] | 5072 | * Also, dquot_alloc_block() will always dirty the inode when blocks |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 5073 | * are allocated to the file. |
| 5074 | * |
| 5075 | * If the inode is marked synchronous, we don't honour that here - doing |
| 5076 | * so would cause a commit on atime updates, which we don't bother doing. |
| 5077 | * We handle synchronous inodes at the highest possible level. |
| 5078 | */ |
Christoph Hellwig | aa38572 | 2011-05-27 06:53:02 -0400 | [diff] [blame] | 5079 | void ext4_dirty_inode(struct inode *inode, int flags) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 5080 | { |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 5081 | handle_t *handle; |
| 5082 | |
Theodore Ts'o | 9924a92 | 2013-02-08 21:59:22 -0500 | [diff] [blame] | 5083 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 2); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 5084 | if (IS_ERR(handle)) |
| 5085 | goto out; |
Curt Wohlgemuth | f3dc272 | 2009-09-29 16:06:01 -0400 | [diff] [blame] | 5086 | |
Curt Wohlgemuth | f3dc272 | 2009-09-29 16:06:01 -0400 | [diff] [blame] | 5087 | ext4_mark_inode_dirty(handle, inode); |
| 5088 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 5089 | ext4_journal_stop(handle); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 5090 | out: |
| 5091 | return; |
| 5092 | } |
| 5093 | |
| 5094 | #if 0 |
| 5095 | /* |
| 5096 | * Bind an inode's backing buffer_head into this transaction, to prevent |
| 5097 | * it from being flushed to disk early. Unlike |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 5098 | * ext4_reserve_inode_write, this leaves behind no bh reference and |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 5099 | * returns no iloc structure, so the caller needs to repeat the iloc |
| 5100 | * lookup to mark the inode dirty later. |
| 5101 | */ |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 5102 | static int ext4_pin_inode(handle_t *handle, struct inode *inode) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 5103 | { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 5104 | struct ext4_iloc iloc; |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 5105 | |
| 5106 | int err = 0; |
| 5107 | if (handle) { |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 5108 | err = ext4_get_inode_loc(inode, &iloc); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 5109 | if (!err) { |
| 5110 | BUFFER_TRACE(iloc.bh, "get_write_access"); |
Mingming Cao | dab291a | 2006-10-11 01:21:01 -0700 | [diff] [blame] | 5111 | err = jbd2_journal_get_write_access(handle, iloc.bh); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 5112 | if (!err) |
Frank Mayhar | 0390131 | 2009-01-07 00:06:22 -0500 | [diff] [blame] | 5113 | err = ext4_handle_dirty_metadata(handle, |
Curt Wohlgemuth | 73b50c1 | 2010-02-16 15:06:29 -0500 | [diff] [blame] | 5114 | NULL, |
Frank Mayhar | 0390131 | 2009-01-07 00:06:22 -0500 | [diff] [blame] | 5115 | iloc.bh); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 5116 | brelse(iloc.bh); |
| 5117 | } |
| 5118 | } |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 5119 | ext4_std_error(inode->i_sb, err); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 5120 | return err; |
| 5121 | } |
| 5122 | #endif |
| 5123 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 5124 | int ext4_change_inode_journal_flag(struct inode *inode, int val) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 5125 | { |
| 5126 | journal_t *journal; |
| 5127 | handle_t *handle; |
| 5128 | int err; |
| 5129 | |
| 5130 | /* |
| 5131 | * We have to be very careful here: changing a data block's |
| 5132 | * journaling status dynamically is dangerous. If we write a |
| 5133 | * data block to the journal, change the status and then delete |
| 5134 | * that block, we risk forgetting to revoke the old log record |
| 5135 | * from the journal and so a subsequent replay can corrupt data. |
| 5136 | * So, first we make sure that the journal is empty and that |
| 5137 | * nobody is changing anything. |
| 5138 | */ |
| 5139 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 5140 | journal = EXT4_JOURNAL(inode); |
Frank Mayhar | 0390131 | 2009-01-07 00:06:22 -0500 | [diff] [blame] | 5141 | if (!journal) |
| 5142 | return 0; |
Dave Hansen | d699594 | 2007-07-18 08:33:51 -0400 | [diff] [blame] | 5143 | if (is_journal_aborted(journal)) |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 5144 | return -EROFS; |
Yongqiang Yang | 2aff57b | 2011-12-28 12:02:13 -0500 | [diff] [blame] | 5145 | /* We have to allocate physical blocks for delalloc blocks |
| 5146 | * before flushing journal. otherwise delalloc blocks can not |
| 5147 | * be allocated any more. even more truncate on delalloc blocks |
| 5148 | * could trigger BUG by flushing delalloc blocks in journal. |
| 5149 | * There is no delalloc block in non-journal data mode. |
| 5150 | */ |
| 5151 | if (val && test_opt(inode->i_sb, DELALLOC)) { |
| 5152 | err = ext4_alloc_da_blocks(inode); |
| 5153 | if (err < 0) |
| 5154 | return err; |
| 5155 | } |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 5156 | |
Dmitry Monakhov | 17335dc | 2012-09-29 00:41:21 -0400 | [diff] [blame] | 5157 | /* Wait for all existing dio workers */ |
| 5158 | ext4_inode_block_unlocked_dio(inode); |
| 5159 | inode_dio_wait(inode); |
| 5160 | |
Mingming Cao | dab291a | 2006-10-11 01:21:01 -0700 | [diff] [blame] | 5161 | jbd2_journal_lock_updates(journal); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 5162 | |
| 5163 | /* |
| 5164 | * OK, there are no updates running now, and all cached data is |
| 5165 | * synced to disk. We are now in a completely consistent state |
| 5166 | * which doesn't have anything in the journal, and we know that |
| 5167 | * no filesystem updates are running, so it is safe to modify |
| 5168 | * the inode's in-core data-journaling state flag now. |
| 5169 | */ |
| 5170 | |
| 5171 | if (val) |
Dmitry Monakhov | 12e9b89 | 2010-05-16 22:00:00 -0400 | [diff] [blame] | 5172 | ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
Yongqiang Yang | 5872dda | 2011-12-28 13:55:51 -0500 | [diff] [blame] | 5173 | else { |
| 5174 | jbd2_journal_flush(journal); |
Dmitry Monakhov | 12e9b89 | 2010-05-16 22:00:00 -0400 | [diff] [blame] | 5175 | ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
Yongqiang Yang | 5872dda | 2011-12-28 13:55:51 -0500 | [diff] [blame] | 5176 | } |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 5177 | ext4_set_aops(inode); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 5178 | |
Mingming Cao | dab291a | 2006-10-11 01:21:01 -0700 | [diff] [blame] | 5179 | jbd2_journal_unlock_updates(journal); |
Dmitry Monakhov | 17335dc | 2012-09-29 00:41:21 -0400 | [diff] [blame] | 5180 | ext4_inode_resume_unlocked_dio(inode); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 5181 | |
| 5182 | /* Finally we can mark the inode as dirty. */ |
| 5183 | |
Theodore Ts'o | 9924a92 | 2013-02-08 21:59:22 -0500 | [diff] [blame] | 5184 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 1); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 5185 | if (IS_ERR(handle)) |
| 5186 | return PTR_ERR(handle); |
| 5187 | |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 5188 | err = ext4_mark_inode_dirty(handle, inode); |
Frank Mayhar | 0390131 | 2009-01-07 00:06:22 -0500 | [diff] [blame] | 5189 | ext4_handle_sync(handle); |
Mingming Cao | 617ba13 | 2006-10-11 01:20:53 -0700 | [diff] [blame] | 5190 | ext4_journal_stop(handle); |
| 5191 | ext4_std_error(inode->i_sb, err); |
Dave Kleikamp | ac27a0e | 2006-10-11 01:20:50 -0700 | [diff] [blame] | 5192 | |
| 5193 | return err; |
| 5194 | } |
Aneesh Kumar K.V | 2e9ee85 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 5195 | |
| 5196 | static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh) |
| 5197 | { |
| 5198 | return !buffer_mapped(bh); |
| 5199 | } |
| 5200 | |
Nick Piggin | c2ec175 | 2009-03-31 15:23:21 -0700 | [diff] [blame] | 5201 | int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
Aneesh Kumar K.V | 2e9ee85 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 5202 | { |
Nick Piggin | c2ec175 | 2009-03-31 15:23:21 -0700 | [diff] [blame] | 5203 | struct page *page = vmf->page; |
Aneesh Kumar K.V | 2e9ee85 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 5204 | loff_t size; |
| 5205 | unsigned long len; |
Jan Kara | 9ea7df5 | 2011-06-24 14:29:41 -0400 | [diff] [blame] | 5206 | int ret; |
Aneesh Kumar K.V | 2e9ee85 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 5207 | struct file *file = vma->vm_file; |
Al Viro | 496ad9a | 2013-01-23 17:07:38 -0500 | [diff] [blame] | 5208 | struct inode *inode = file_inode(file); |
Aneesh Kumar K.V | 2e9ee85 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 5209 | struct address_space *mapping = inode->i_mapping; |
Jan Kara | 9ea7df5 | 2011-06-24 14:29:41 -0400 | [diff] [blame] | 5210 | handle_t *handle; |
| 5211 | get_block_t *get_block; |
| 5212 | int retries = 0; |
Aneesh Kumar K.V | 2e9ee85 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 5213 | |
Jan Kara | 8e8ad8a | 2012-06-12 16:20:38 +0200 | [diff] [blame] | 5214 | sb_start_pagefault(inode->i_sb); |
Theodore Ts'o | 041bbb6d | 2012-09-30 23:04:56 -0400 | [diff] [blame] | 5215 | file_update_time(vma->vm_file); |
Jan Kara | 9ea7df5 | 2011-06-24 14:29:41 -0400 | [diff] [blame] | 5216 | /* Delalloc case is easy... */ |
| 5217 | if (test_opt(inode->i_sb, DELALLOC) && |
| 5218 | !ext4_should_journal_data(inode) && |
| 5219 | !ext4_nonda_switch(inode->i_sb)) { |
| 5220 | do { |
| 5221 | ret = __block_page_mkwrite(vma, vmf, |
| 5222 | ext4_da_get_block_prep); |
| 5223 | } while (ret == -ENOSPC && |
| 5224 | ext4_should_retry_alloc(inode->i_sb, &retries)); |
| 5225 | goto out_ret; |
Aneesh Kumar K.V | 2e9ee85 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 5226 | } |
Darrick J. Wong | 0e49989 | 2011-05-18 13:55:20 -0400 | [diff] [blame] | 5227 | |
| 5228 | lock_page(page); |
Jan Kara | 9ea7df5 | 2011-06-24 14:29:41 -0400 | [diff] [blame] | 5229 | size = i_size_read(inode); |
| 5230 | /* Page got truncated from under us? */ |
| 5231 | if (page->mapping != mapping || page_offset(page) > size) { |
| 5232 | unlock_page(page); |
| 5233 | ret = VM_FAULT_NOPAGE; |
| 5234 | goto out; |
Darrick J. Wong | 0e49989 | 2011-05-18 13:55:20 -0400 | [diff] [blame] | 5235 | } |
Aneesh Kumar K.V | 2e9ee85 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 5236 | |
| 5237 | if (page->index == size >> PAGE_CACHE_SHIFT) |
| 5238 | len = size & ~PAGE_CACHE_MASK; |
| 5239 | else |
| 5240 | len = PAGE_CACHE_SIZE; |
Aneesh Kumar K.V | a827eaf | 2009-09-09 22:36:03 -0400 | [diff] [blame] | 5241 | /* |
Jan Kara | 9ea7df5 | 2011-06-24 14:29:41 -0400 | [diff] [blame] | 5242 | * Return if we have all the buffers mapped. This avoids the need to do |
| 5243 | * journal_start/journal_stop which can block and take a long time |
Aneesh Kumar K.V | a827eaf | 2009-09-09 22:36:03 -0400 | [diff] [blame] | 5244 | */ |
Aneesh Kumar K.V | 2e9ee85 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 5245 | if (page_has_buffers(page)) { |
Tao Ma | f19d587 | 2012-12-10 14:05:51 -0500 | [diff] [blame] | 5246 | if (!ext4_walk_page_buffers(NULL, page_buffers(page), |
| 5247 | 0, len, NULL, |
| 5248 | ext4_bh_unmapped)) { |
Jan Kara | 9ea7df5 | 2011-06-24 14:29:41 -0400 | [diff] [blame] | 5249 | /* Wait so that we don't change page under IO */ |
Darrick J. Wong | 1d1d1a7 | 2013-02-21 16:42:51 -0800 | [diff] [blame] | 5250 | wait_for_stable_page(page); |
Jan Kara | 9ea7df5 | 2011-06-24 14:29:41 -0400 | [diff] [blame] | 5251 | ret = VM_FAULT_LOCKED; |
| 5252 | goto out; |
Aneesh Kumar K.V | a827eaf | 2009-09-09 22:36:03 -0400 | [diff] [blame] | 5253 | } |
Aneesh Kumar K.V | 2e9ee85 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 5254 | } |
Aneesh Kumar K.V | a827eaf | 2009-09-09 22:36:03 -0400 | [diff] [blame] | 5255 | unlock_page(page); |
Jan Kara | 9ea7df5 | 2011-06-24 14:29:41 -0400 | [diff] [blame] | 5256 | /* OK, we need to fill the hole... */ |
| 5257 | if (ext4_should_dioread_nolock(inode)) |
| 5258 | get_block = ext4_get_block_write; |
| 5259 | else |
| 5260 | get_block = ext4_get_block; |
| 5261 | retry_alloc: |
Theodore Ts'o | 9924a92 | 2013-02-08 21:59:22 -0500 | [diff] [blame] | 5262 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
| 5263 | ext4_writepage_trans_blocks(inode)); |
Jan Kara | 9ea7df5 | 2011-06-24 14:29:41 -0400 | [diff] [blame] | 5264 | if (IS_ERR(handle)) { |
Nick Piggin | c2ec175 | 2009-03-31 15:23:21 -0700 | [diff] [blame] | 5265 | ret = VM_FAULT_SIGBUS; |
Jan Kara | 9ea7df5 | 2011-06-24 14:29:41 -0400 | [diff] [blame] | 5266 | goto out; |
| 5267 | } |
| 5268 | ret = __block_page_mkwrite(vma, vmf, get_block); |
| 5269 | if (!ret && ext4_should_journal_data(inode)) { |
Tao Ma | f19d587 | 2012-12-10 14:05:51 -0500 | [diff] [blame] | 5270 | if (ext4_walk_page_buffers(handle, page_buffers(page), 0, |
Jan Kara | 9ea7df5 | 2011-06-24 14:29:41 -0400 | [diff] [blame] | 5271 | PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) { |
| 5272 | unlock_page(page); |
| 5273 | ret = VM_FAULT_SIGBUS; |
Yongqiang Yang | fcbb551 | 2011-10-26 05:00:19 -0400 | [diff] [blame] | 5274 | ext4_journal_stop(handle); |
Jan Kara | 9ea7df5 | 2011-06-24 14:29:41 -0400 | [diff] [blame] | 5275 | goto out; |
| 5276 | } |
| 5277 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
| 5278 | } |
| 5279 | ext4_journal_stop(handle); |
| 5280 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) |
| 5281 | goto retry_alloc; |
| 5282 | out_ret: |
| 5283 | ret = block_page_mkwrite_return(ret); |
| 5284 | out: |
Jan Kara | 8e8ad8a | 2012-06-12 16:20:38 +0200 | [diff] [blame] | 5285 | sb_end_pagefault(inode->i_sb); |
Aneesh Kumar K.V | 2e9ee85 | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 5286 | return ret; |
| 5287 | } |