Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * linux/fs/ext3/inode.c |
| 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 | * |
| 15 | * Goal-directed block allocation by Stephen Tweedie |
Dave Kleikamp | e9ad562 | 2006-09-27 01:49:35 -0700 | [diff] [blame] | 16 | * (sct@redhat.com), 1993, 1998 |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 17 | * Big-endian to little-endian byte-swapping/bitmaps by |
| 18 | * David S. Miller (davem@caip.rutgers.edu), 1995 |
| 19 | * 64-bit file support on 64-bit platforms by Jakub Jelinek |
Dave Kleikamp | e9ad562 | 2006-09-27 01:49:35 -0700 | [diff] [blame] | 20 | * (jj@sunsite.ms.mff.cuni.cz) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 21 | * |
| 22 | * Assorted race fixes, rewrite of ext3_get_block() by Al Viro, 2000 |
| 23 | */ |
| 24 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 25 | #include <linux/highuid.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 26 | #include <linux/quotaops.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 27 | #include <linux/writeback.h> |
| 28 | #include <linux/mpage.h> |
Duane Griffin | b5ed311 | 2008-12-19 20:47:14 +0000 | [diff] [blame] | 29 | #include <linux/namei.h> |
Al Viro | 4613ad1 | 2012-03-29 22:30:07 -0400 | [diff] [blame] | 30 | #include "ext3.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 31 | #include "xattr.h" |
| 32 | #include "acl.h" |
| 33 | |
| 34 | static int ext3_writepage_trans_blocks(struct inode *inode); |
Jan Kara | ee3e77f | 2011-06-03 21:58:11 +0200 | [diff] [blame] | 35 | static int ext3_block_truncate_page(struct inode *inode, loff_t from); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 36 | |
| 37 | /* |
| 38 | * Test whether an inode is a fast symlink. |
| 39 | */ |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 40 | static int ext3_inode_is_fast_symlink(struct inode *inode) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 41 | { |
| 42 | int ea_blocks = EXT3_I(inode)->i_file_acl ? |
| 43 | (inode->i_sb->s_blocksize >> 9) : 0; |
| 44 | |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 45 | return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 46 | } |
| 47 | |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 48 | /* |
| 49 | * The ext3 forget function must perform a revoke if we are freeing data |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 50 | * which has been journaled. Metadata (eg. indirect blocks) must be |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 51 | * revoked in all cases. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 52 | * |
| 53 | * "bh" may be NULL: a metadata block may have been freed from memory |
| 54 | * but there may still be a record of it in the journal, and that record |
| 55 | * still needs to be revoked. |
| 56 | */ |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 57 | int ext3_forget(handle_t *handle, int is_metadata, struct inode *inode, |
Mingming Cao | 1c2bf37 | 2006-06-25 05:48:06 -0700 | [diff] [blame] | 58 | struct buffer_head *bh, ext3_fsblk_t blocknr) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 59 | { |
| 60 | int err; |
| 61 | |
| 62 | might_sleep(); |
| 63 | |
Lukas Czerner | 785c4bc | 2011-05-23 18:33:01 +0200 | [diff] [blame] | 64 | trace_ext3_forget(inode, is_metadata, blocknr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 65 | BUFFER_TRACE(bh, "enter"); |
| 66 | |
| 67 | jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, " |
| 68 | "data mode %lx\n", |
| 69 | bh, is_metadata, inode->i_mode, |
| 70 | test_opt(inode->i_sb, DATA_FLAGS)); |
| 71 | |
| 72 | /* Never use the revoke function if we are doing full data |
| 73 | * journaling: there is no need to, and a V1 superblock won't |
| 74 | * support it. Otherwise, only skip the revoke on un-journaled |
| 75 | * data blocks. */ |
| 76 | |
| 77 | if (test_opt(inode->i_sb, DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA || |
| 78 | (!is_metadata && !ext3_should_journal_data(inode))) { |
| 79 | if (bh) { |
| 80 | BUFFER_TRACE(bh, "call journal_forget"); |
| 81 | return ext3_journal_forget(handle, bh); |
| 82 | } |
| 83 | return 0; |
| 84 | } |
| 85 | |
| 86 | /* |
| 87 | * data!=journal && (is_metadata || should_journal_data(inode)) |
| 88 | */ |
| 89 | BUFFER_TRACE(bh, "call ext3_journal_revoke"); |
| 90 | err = ext3_journal_revoke(handle, blocknr, bh); |
| 91 | if (err) |
Harvey Harrison | e05b6b5 | 2008-04-28 02:16:15 -0700 | [diff] [blame] | 92 | ext3_abort(inode->i_sb, __func__, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 93 | "error %d when attempting revoke", err); |
| 94 | BUFFER_TRACE(bh, "exit"); |
| 95 | return err; |
| 96 | } |
| 97 | |
| 98 | /* |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 99 | * Work out how many blocks we need to proceed with the next chunk of a |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 100 | * truncate transaction. |
| 101 | */ |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 102 | static unsigned long blocks_for_truncate(struct inode *inode) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 103 | { |
| 104 | unsigned long needed; |
| 105 | |
| 106 | needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9); |
| 107 | |
| 108 | /* Give ourselves just enough room to cope with inodes in which |
| 109 | * i_blocks is corrupt: we've seen disk corruptions in the past |
| 110 | * which resulted in random data in an inode which looked enough |
| 111 | * like a regular file for ext3 to try to delete it. Things |
| 112 | * will go a bit crazy if that happens, but at least we should |
| 113 | * try not to panic the whole kernel. */ |
| 114 | if (needed < 2) |
| 115 | needed = 2; |
| 116 | |
| 117 | /* But we need to bound the transaction so we don't overflow the |
| 118 | * journal. */ |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 119 | if (needed > EXT3_MAX_TRANS_DATA) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 120 | needed = EXT3_MAX_TRANS_DATA; |
| 121 | |
Jan Kara | 1f54587 | 2005-06-23 22:01:04 -0700 | [diff] [blame] | 122 | return EXT3_DATA_TRANS_BLOCKS(inode->i_sb) + needed; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 123 | } |
| 124 | |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 125 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 126 | * Truncate transactions can be complex and absolutely huge. So we need to |
| 127 | * be able to restart the transaction at a conventient checkpoint to make |
| 128 | * sure we don't overflow the journal. |
| 129 | * |
| 130 | * start_transaction gets us a new handle for a truncate transaction, |
| 131 | * and extend_transaction tries to extend the existing one a bit. If |
| 132 | * extend fails, we need to propagate the failure up and restart the |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 133 | * transaction in the top-level truncate loop. --sct |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 134 | */ |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 135 | static handle_t *start_transaction(struct inode *inode) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 136 | { |
| 137 | handle_t *result; |
| 138 | |
| 139 | result = ext3_journal_start(inode, blocks_for_truncate(inode)); |
| 140 | if (!IS_ERR(result)) |
| 141 | return result; |
| 142 | |
| 143 | ext3_std_error(inode->i_sb, PTR_ERR(result)); |
| 144 | return result; |
| 145 | } |
| 146 | |
| 147 | /* |
| 148 | * Try to extend this transaction for the purposes of truncation. |
| 149 | * |
| 150 | * Returns 0 if we managed to create more room. If we can't create more |
| 151 | * room, and the transaction must be restarted we return 1. |
| 152 | */ |
| 153 | static int try_to_extend_transaction(handle_t *handle, struct inode *inode) |
| 154 | { |
| 155 | if (handle->h_buffer_credits > EXT3_RESERVE_TRANS_BLOCKS) |
| 156 | return 0; |
| 157 | if (!ext3_journal_extend(handle, blocks_for_truncate(inode))) |
| 158 | return 0; |
| 159 | return 1; |
| 160 | } |
| 161 | |
| 162 | /* |
| 163 | * Restart the transaction associated with *handle. This does a commit, |
| 164 | * so before we call here everything must be consistently dirtied against |
| 165 | * this transaction. |
| 166 | */ |
Jan Kara | 00171d3 | 2009-08-11 19:06:10 +0200 | [diff] [blame] | 167 | static int truncate_restart_transaction(handle_t *handle, struct inode *inode) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 168 | { |
Jan Kara | 00171d3 | 2009-08-11 19:06:10 +0200 | [diff] [blame] | 169 | int ret; |
| 170 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 171 | jbd_debug(2, "restarting handle %p\n", handle); |
Jan Kara | 00171d3 | 2009-08-11 19:06:10 +0200 | [diff] [blame] | 172 | /* |
| 173 | * Drop truncate_mutex to avoid deadlock with ext3_get_blocks_handle |
| 174 | * At this moment, get_block can be called only for blocks inside |
| 175 | * i_size since page cache has been already dropped and writes are |
| 176 | * blocked by i_mutex. So we can safely drop the truncate_mutex. |
| 177 | */ |
| 178 | mutex_unlock(&EXT3_I(inode)->truncate_mutex); |
| 179 | ret = ext3_journal_restart(handle, blocks_for_truncate(inode)); |
| 180 | mutex_lock(&EXT3_I(inode)->truncate_mutex); |
| 181 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 182 | } |
| 183 | |
| 184 | /* |
Al Viro | ac14a95 | 2010-06-06 07:08:19 -0400 | [diff] [blame] | 185 | * Called at inode eviction from icache |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 186 | */ |
Al Viro | ac14a95 | 2010-06-06 07:08:19 -0400 | [diff] [blame] | 187 | void ext3_evict_inode (struct inode *inode) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 188 | { |
Jan Kara | b22570d | 2011-07-23 01:21:38 +0200 | [diff] [blame] | 189 | struct ext3_inode_info *ei = EXT3_I(inode); |
Al Viro | ac14a95 | 2010-06-06 07:08:19 -0400 | [diff] [blame] | 190 | struct ext3_block_alloc_info *rsv; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 191 | handle_t *handle; |
Al Viro | ac14a95 | 2010-06-06 07:08:19 -0400 | [diff] [blame] | 192 | int want_delete = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 193 | |
Lukas Czerner | 785c4bc | 2011-05-23 18:33:01 +0200 | [diff] [blame] | 194 | trace_ext3_evict_inode(inode); |
Al Viro | ac14a95 | 2010-06-06 07:08:19 -0400 | [diff] [blame] | 195 | if (!inode->i_nlink && !is_bad_inode(inode)) { |
Christoph Hellwig | 871a293 | 2010-03-03 09:05:07 -0500 | [diff] [blame] | 196 | dquot_initialize(inode); |
Al Viro | ac14a95 | 2010-06-06 07:08:19 -0400 | [diff] [blame] | 197 | want_delete = 1; |
| 198 | } |
Christoph Hellwig | 907f455 | 2010-03-03 09:05:06 -0500 | [diff] [blame] | 199 | |
Jan Kara | b22570d | 2011-07-23 01:21:38 +0200 | [diff] [blame] | 200 | /* |
| 201 | * When journalling data dirty buffers are tracked only in the journal. |
| 202 | * So although mm thinks everything is clean and ready for reaping the |
| 203 | * inode might still have some pages to write in the running |
| 204 | * transaction or waiting to be checkpointed. Thus calling |
| 205 | * journal_invalidatepage() (via truncate_inode_pages()) to discard |
| 206 | * these buffers can cause data loss. Also even if we did not discard |
| 207 | * these buffers, we would have no way to find them after the inode |
| 208 | * is reaped and thus user could see stale data if he tries to read |
| 209 | * them before the transaction is checkpointed. So be careful and |
| 210 | * force everything to disk here... We use ei->i_datasync_tid to |
| 211 | * store the newest transaction containing inode's data. |
| 212 | * |
| 213 | * Note that directories do not have this problem because they don't |
| 214 | * use page cache. |
Dan Carpenter | bcdd0c1 | 2011-11-22 11:00:20 +0300 | [diff] [blame] | 215 | * |
| 216 | * The s_journal check handles the case when ext3_get_journal() fails |
| 217 | * and puts the journal inode. |
Jan Kara | b22570d | 2011-07-23 01:21:38 +0200 | [diff] [blame] | 218 | */ |
| 219 | if (inode->i_nlink && ext3_should_journal_data(inode) && |
Dan Carpenter | bcdd0c1 | 2011-11-22 11:00:20 +0300 | [diff] [blame] | 220 | EXT3_SB(inode->i_sb)->s_journal && |
Jan Kara | 62305be | 2013-03-20 14:39:05 +0100 | [diff] [blame] | 221 | (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) && |
| 222 | inode->i_ino != EXT3_JOURNAL_INO) { |
Jan Kara | b22570d | 2011-07-23 01:21:38 +0200 | [diff] [blame] | 223 | tid_t commit_tid = atomic_read(&ei->i_datasync_tid); |
| 224 | journal_t *journal = EXT3_SB(inode->i_sb)->s_journal; |
| 225 | |
| 226 | log_start_commit(journal, commit_tid); |
| 227 | log_wait_commit(journal, commit_tid); |
| 228 | filemap_write_and_wait(&inode->i_data); |
| 229 | } |
Mark Fasheh | fef2665 | 2005-09-09 13:01:31 -0700 | [diff] [blame] | 230 | truncate_inode_pages(&inode->i_data, 0); |
| 231 | |
Al Viro | ac14a95 | 2010-06-06 07:08:19 -0400 | [diff] [blame] | 232 | ext3_discard_reservation(inode); |
Jan Kara | b22570d | 2011-07-23 01:21:38 +0200 | [diff] [blame] | 233 | rsv = ei->i_block_alloc_info; |
| 234 | ei->i_block_alloc_info = NULL; |
Al Viro | ac14a95 | 2010-06-06 07:08:19 -0400 | [diff] [blame] | 235 | if (unlikely(rsv)) |
| 236 | kfree(rsv); |
| 237 | |
| 238 | if (!want_delete) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 239 | goto no_delete; |
| 240 | |
| 241 | handle = start_transaction(inode); |
| 242 | if (IS_ERR(handle)) { |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 243 | /* |
| 244 | * If we're going to skip the normal cleanup, we still need to |
| 245 | * make sure that the in-core orphan linked list is properly |
| 246 | * cleaned up. |
| 247 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 248 | ext3_orphan_del(NULL, inode); |
| 249 | goto no_delete; |
| 250 | } |
| 251 | |
| 252 | if (IS_SYNC(inode)) |
| 253 | handle->h_sync = 1; |
| 254 | inode->i_size = 0; |
| 255 | if (inode->i_blocks) |
| 256 | ext3_truncate(inode); |
| 257 | /* |
Jan Kara | 40680f2f | 2011-05-24 22:24:47 +0200 | [diff] [blame] | 258 | * Kill off the orphan record created when the inode lost the last |
| 259 | * link. Note that ext3_orphan_del() has to be able to cope with the |
| 260 | * deletion of a non-existent orphan - ext3_truncate() could |
| 261 | * have removed the record. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 262 | */ |
| 263 | ext3_orphan_del(handle, inode); |
Jan Kara | b22570d | 2011-07-23 01:21:38 +0200 | [diff] [blame] | 264 | ei->i_dtime = get_seconds(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 265 | |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 266 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 267 | * One subtle ordering requirement: if anything has gone wrong |
| 268 | * (transaction abort, IO errors, whatever), then we can still |
| 269 | * do these next steps (the fs will already have been marked as |
| 270 | * having errors), but we can't free the inode if the mark_dirty |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 271 | * fails. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 272 | */ |
Al Viro | ac14a95 | 2010-06-06 07:08:19 -0400 | [diff] [blame] | 273 | if (ext3_mark_inode_dirty(handle, inode)) { |
| 274 | /* If that failed, just dquot_drop() and be done with that */ |
| 275 | dquot_drop(inode); |
| 276 | end_writeback(inode); |
| 277 | } else { |
| 278 | ext3_xattr_delete_inode(handle, inode); |
| 279 | dquot_free_inode(inode); |
| 280 | dquot_drop(inode); |
| 281 | end_writeback(inode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 282 | ext3_free_inode(handle, inode); |
Al Viro | ac14a95 | 2010-06-06 07:08:19 -0400 | [diff] [blame] | 283 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 284 | ext3_journal_stop(handle); |
| 285 | return; |
| 286 | no_delete: |
Al Viro | ac14a95 | 2010-06-06 07:08:19 -0400 | [diff] [blame] | 287 | end_writeback(inode); |
| 288 | dquot_drop(inode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 289 | } |
| 290 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 291 | typedef struct { |
| 292 | __le32 *p; |
| 293 | __le32 key; |
| 294 | struct buffer_head *bh; |
| 295 | } Indirect; |
| 296 | |
| 297 | static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v) |
| 298 | { |
| 299 | p->key = *(p->p = v); |
| 300 | p->bh = bh; |
| 301 | } |
| 302 | |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 303 | static int verify_chain(Indirect *from, Indirect *to) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 304 | { |
| 305 | while (from <= to && from->key == *from->p) |
| 306 | from++; |
| 307 | return (from > to); |
| 308 | } |
| 309 | |
| 310 | /** |
| 311 | * ext3_block_to_path - parse the block number into array of offsets |
| 312 | * @inode: inode in question (we are only interested in its superblock) |
| 313 | * @i_block: block number to be parsed |
| 314 | * @offsets: array to store the offsets in |
| 315 | * @boundary: set this non-zero if the referred-to block is likely to be |
| 316 | * followed (on disk) by an indirect block. |
| 317 | * |
| 318 | * To store the locations of file's data ext3 uses a data structure common |
| 319 | * for UNIX filesystems - tree of pointers anchored in the inode, with |
| 320 | * data blocks at leaves and indirect blocks in intermediate nodes. |
| 321 | * This function translates the block number into path in that tree - |
| 322 | * return value is the path length and @offsets[n] is the offset of |
| 323 | * pointer to (n+1)th node in the nth one. If @block is out of range |
| 324 | * (negative or too large) warning is printed and zero returned. |
| 325 | * |
| 326 | * Note: function doesn't find node addresses, so no IO is needed. All |
| 327 | * we need to know is the capacity of indirect blocks (taken from the |
| 328 | * inode->i_sb). |
| 329 | */ |
| 330 | |
| 331 | /* |
| 332 | * Portability note: the last comparison (check that we fit into triple |
| 333 | * indirect block) is spelled differently, because otherwise on an |
| 334 | * architecture with 32-bit longs and 8Kb pages we might get into trouble |
| 335 | * if our filesystem had 8Kb blocks. We might use long long, but that would |
| 336 | * kill us on x86. Oh, well, at least the sign propagation does not matter - |
| 337 | * i_block would have to be negative in the very beginning, so we would not |
| 338 | * get there at all. |
| 339 | */ |
| 340 | |
| 341 | static int ext3_block_to_path(struct inode *inode, |
| 342 | long i_block, int offsets[4], int *boundary) |
| 343 | { |
| 344 | int ptrs = EXT3_ADDR_PER_BLOCK(inode->i_sb); |
| 345 | int ptrs_bits = EXT3_ADDR_PER_BLOCK_BITS(inode->i_sb); |
| 346 | const long direct_blocks = EXT3_NDIR_BLOCKS, |
| 347 | indirect_blocks = ptrs, |
| 348 | double_blocks = (1 << (ptrs_bits * 2)); |
| 349 | int n = 0; |
| 350 | int final = 0; |
| 351 | |
| 352 | if (i_block < 0) { |
| 353 | ext3_warning (inode->i_sb, "ext3_block_to_path", "block < 0"); |
| 354 | } else if (i_block < direct_blocks) { |
| 355 | offsets[n++] = i_block; |
| 356 | final = direct_blocks; |
| 357 | } else if ( (i_block -= direct_blocks) < indirect_blocks) { |
| 358 | offsets[n++] = EXT3_IND_BLOCK; |
| 359 | offsets[n++] = i_block; |
| 360 | final = ptrs; |
| 361 | } else if ((i_block -= indirect_blocks) < double_blocks) { |
| 362 | offsets[n++] = EXT3_DIND_BLOCK; |
| 363 | offsets[n++] = i_block >> ptrs_bits; |
| 364 | offsets[n++] = i_block & (ptrs - 1); |
| 365 | final = ptrs; |
| 366 | } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) { |
| 367 | offsets[n++] = EXT3_TIND_BLOCK; |
| 368 | offsets[n++] = i_block >> (ptrs_bits * 2); |
| 369 | offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1); |
| 370 | offsets[n++] = i_block & (ptrs - 1); |
| 371 | final = ptrs; |
| 372 | } else { |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 373 | ext3_warning(inode->i_sb, "ext3_block_to_path", "block > big"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 374 | } |
| 375 | if (boundary) |
Mingming Cao | 89747d3 | 2006-03-26 01:37:55 -0800 | [diff] [blame] | 376 | *boundary = final - 1 - (i_block & (ptrs - 1)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 377 | return n; |
| 378 | } |
| 379 | |
| 380 | /** |
| 381 | * ext3_get_branch - read the chain of indirect blocks leading to data |
| 382 | * @inode: inode in question |
| 383 | * @depth: depth of the chain (1 - direct pointer, etc.) |
| 384 | * @offsets: offsets of pointers in inode/indirect blocks |
| 385 | * @chain: place to store the result |
| 386 | * @err: here we store the error value |
| 387 | * |
| 388 | * Function fills the array of triples <key, p, bh> and returns %NULL |
| 389 | * if everything went OK or the pointer to the last filled triple |
| 390 | * (incomplete one) otherwise. Upon the return chain[i].key contains |
| 391 | * the number of (i+1)-th block in the chain (as it is stored in memory, |
| 392 | * i.e. little-endian 32-bit), chain[i].p contains the address of that |
| 393 | * number (it points into struct inode for i==0 and into the bh->b_data |
| 394 | * for i>0) and chain[i].bh points to the buffer_head of i-th indirect |
| 395 | * block for i>0 and NULL for i==0. In other words, it holds the block |
| 396 | * numbers of the chain, addresses they were taken from (and where we can |
| 397 | * verify that chain did not change) and buffer_heads hosting these |
| 398 | * numbers. |
| 399 | * |
| 400 | * Function stops when it stumbles upon zero pointer (absent block) |
| 401 | * (pointer to last triple returned, *@err == 0) |
| 402 | * or when it gets an IO error reading an indirect block |
| 403 | * (ditto, *@err == -EIO) |
| 404 | * or when it notices that chain had been changed while it was reading |
| 405 | * (ditto, *@err == -EAGAIN) |
| 406 | * or when it reads all @depth-1 indirect blocks successfully and finds |
| 407 | * the whole chain, all way to the data (returns %NULL, *err == 0). |
| 408 | */ |
| 409 | static Indirect *ext3_get_branch(struct inode *inode, int depth, int *offsets, |
| 410 | Indirect chain[4], int *err) |
| 411 | { |
| 412 | struct super_block *sb = inode->i_sb; |
| 413 | Indirect *p = chain; |
| 414 | struct buffer_head *bh; |
| 415 | |
| 416 | *err = 0; |
| 417 | /* i_data is not going away, no lock needed */ |
| 418 | add_chain (chain, NULL, EXT3_I(inode)->i_data + *offsets); |
| 419 | if (!p->key) |
| 420 | goto no_block; |
| 421 | while (--depth) { |
| 422 | bh = sb_bread(sb, le32_to_cpu(p->key)); |
| 423 | if (!bh) |
| 424 | goto failure; |
| 425 | /* Reader: pointers */ |
| 426 | if (!verify_chain(chain, p)) |
| 427 | goto changed; |
| 428 | add_chain(++p, bh, (__le32*)bh->b_data + *++offsets); |
| 429 | /* Reader: end */ |
| 430 | if (!p->key) |
| 431 | goto no_block; |
| 432 | } |
| 433 | return NULL; |
| 434 | |
| 435 | changed: |
| 436 | brelse(bh); |
| 437 | *err = -EAGAIN; |
| 438 | goto no_block; |
| 439 | failure: |
| 440 | *err = -EIO; |
| 441 | no_block: |
| 442 | return p; |
| 443 | } |
| 444 | |
| 445 | /** |
| 446 | * ext3_find_near - find a place for allocation with sufficient locality |
| 447 | * @inode: owner |
| 448 | * @ind: descriptor of indirect block. |
| 449 | * |
Benoit Boissinot | 1cc8dcf | 2008-04-21 22:45:55 +0000 | [diff] [blame] | 450 | * This function returns the preferred place for block allocation. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 451 | * It is used when heuristic for sequential allocation fails. |
| 452 | * Rules are: |
| 453 | * + if there is a block to the left of our position - allocate near it. |
| 454 | * + if pointer will live in indirect block - allocate near that block. |
| 455 | * + if pointer will live in inode - allocate in the same |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 456 | * cylinder group. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 457 | * |
| 458 | * In the latter case we colour the starting block by the callers PID to |
| 459 | * prevent it from clashing with concurrent allocations for a different inode |
| 460 | * in the same block group. The PID is used here so that functionally related |
| 461 | * files will be close-by on-disk. |
| 462 | * |
| 463 | * Caller must make sure that @ind is valid and will stay that way. |
| 464 | */ |
Mingming Cao | 43d23f9 | 2006-06-25 05:48:07 -0700 | [diff] [blame] | 465 | static ext3_fsblk_t ext3_find_near(struct inode *inode, Indirect *ind) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 466 | { |
| 467 | struct ext3_inode_info *ei = EXT3_I(inode); |
| 468 | __le32 *start = ind->bh ? (__le32*) ind->bh->b_data : ei->i_data; |
| 469 | __le32 *p; |
Mingming Cao | 43d23f9 | 2006-06-25 05:48:07 -0700 | [diff] [blame] | 470 | ext3_fsblk_t bg_start; |
| 471 | ext3_grpblk_t colour; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 472 | |
| 473 | /* Try to find previous block */ |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 474 | for (p = ind->p - 1; p >= start; p--) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 475 | if (*p) |
| 476 | return le32_to_cpu(*p); |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 477 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 478 | |
| 479 | /* No such thing, so let's try location of indirect block */ |
| 480 | if (ind->bh) |
| 481 | return ind->bh->b_blocknr; |
| 482 | |
| 483 | /* |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 484 | * It is going to be referred to from the inode itself? OK, just put it |
| 485 | * into the same cylinder group then. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 486 | */ |
Mingming Cao | 43d23f9 | 2006-06-25 05:48:07 -0700 | [diff] [blame] | 487 | bg_start = ext3_group_first_block_no(inode->i_sb, ei->i_block_group); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 488 | colour = (current->pid % 16) * |
| 489 | (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16); |
| 490 | return bg_start + colour; |
| 491 | } |
| 492 | |
| 493 | /** |
Benoit Boissinot | 1cc8dcf | 2008-04-21 22:45:55 +0000 | [diff] [blame] | 494 | * ext3_find_goal - find a preferred place for allocation. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 495 | * @inode: owner |
| 496 | * @block: block we want |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 497 | * @partial: pointer to the last triple within a chain |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 498 | * |
Benoit Boissinot | 1cc8dcf | 2008-04-21 22:45:55 +0000 | [diff] [blame] | 499 | * Normally this function find the preferred place for block allocation, |
Akinobu Mita | fb01bfd | 2008-02-06 01:40:16 -0800 | [diff] [blame] | 500 | * returns it. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 501 | */ |
| 502 | |
Mingming Cao | 43d23f9 | 2006-06-25 05:48:07 -0700 | [diff] [blame] | 503 | static ext3_fsblk_t ext3_find_goal(struct inode *inode, long block, |
Akinobu Mita | fb01bfd | 2008-02-06 01:40:16 -0800 | [diff] [blame] | 504 | Indirect *partial) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 505 | { |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 506 | struct ext3_block_alloc_info *block_i; |
| 507 | |
| 508 | block_i = EXT3_I(inode)->i_block_alloc_info; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 509 | |
| 510 | /* |
| 511 | * try the heuristic for sequential allocation, |
| 512 | * failing that at least try to get decent locality. |
| 513 | */ |
| 514 | if (block_i && (block == block_i->last_alloc_logical_block + 1) |
| 515 | && (block_i->last_alloc_physical_block != 0)) { |
Mingming Cao | fe55c45 | 2005-05-01 08:59:20 -0700 | [diff] [blame] | 516 | return block_i->last_alloc_physical_block + 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 517 | } |
| 518 | |
Mingming Cao | fe55c45 | 2005-05-01 08:59:20 -0700 | [diff] [blame] | 519 | return ext3_find_near(inode, partial); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 520 | } |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 521 | |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 522 | /** |
Namhyung Kim | a4c18ad | 2010-10-19 00:34:35 +0900 | [diff] [blame] | 523 | * ext3_blks_to_allocate - Look up the block map and count the number |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 524 | * of direct blocks need to be allocated for the given branch. |
| 525 | * |
Dave Kleikamp | e9ad562 | 2006-09-27 01:49:35 -0700 | [diff] [blame] | 526 | * @branch: chain of indirect blocks |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 527 | * @k: number of blocks need for indirect blocks |
| 528 | * @blks: number of data blocks to be mapped. |
| 529 | * @blocks_to_boundary: the offset in the indirect block |
| 530 | * |
| 531 | * return the total number of blocks to be allocate, including the |
| 532 | * direct and indirect blocks. |
| 533 | */ |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 534 | static int ext3_blks_to_allocate(Indirect *branch, int k, unsigned long blks, |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 535 | int blocks_to_boundary) |
| 536 | { |
| 537 | unsigned long count = 0; |
| 538 | |
| 539 | /* |
| 540 | * Simple case, [t,d]Indirect block(s) has not allocated yet |
| 541 | * then it's clear blocks on that path have not allocated |
| 542 | */ |
| 543 | if (k > 0) { |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 544 | /* right now we don't handle cross boundary allocation */ |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 545 | if (blks < blocks_to_boundary + 1) |
| 546 | count += blks; |
| 547 | else |
| 548 | count += blocks_to_boundary + 1; |
| 549 | return count; |
| 550 | } |
| 551 | |
| 552 | count++; |
| 553 | while (count < blks && count <= blocks_to_boundary && |
| 554 | le32_to_cpu(*(branch[0].p + count)) == 0) { |
| 555 | count++; |
| 556 | } |
| 557 | return count; |
| 558 | } |
| 559 | |
| 560 | /** |
Namhyung Kim | a4c18ad | 2010-10-19 00:34:35 +0900 | [diff] [blame] | 561 | * ext3_alloc_blocks - multiple allocate blocks needed for a branch |
| 562 | * @handle: handle for this transaction |
| 563 | * @inode: owner |
| 564 | * @goal: preferred place for allocation |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 565 | * @indirect_blks: the number of blocks need to allocate for indirect |
| 566 | * blocks |
Namhyung Kim | a4c18ad | 2010-10-19 00:34:35 +0900 | [diff] [blame] | 567 | * @blks: number of blocks need to allocated for direct blocks |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 568 | * @new_blocks: on return it will store the new block numbers for |
| 569 | * the indirect blocks(if needed) and the first direct block, |
Namhyung Kim | a4c18ad | 2010-10-19 00:34:35 +0900 | [diff] [blame] | 570 | * @err: here we store the error value |
| 571 | * |
| 572 | * return the number of direct blocks allocated |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 573 | */ |
| 574 | static int ext3_alloc_blocks(handle_t *handle, struct inode *inode, |
Mingming Cao | 43d23f9 | 2006-06-25 05:48:07 -0700 | [diff] [blame] | 575 | ext3_fsblk_t goal, int indirect_blks, int blks, |
| 576 | ext3_fsblk_t new_blocks[4], int *err) |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 577 | { |
| 578 | int target, i; |
| 579 | unsigned long count = 0; |
| 580 | int index = 0; |
Mingming Cao | 43d23f9 | 2006-06-25 05:48:07 -0700 | [diff] [blame] | 581 | ext3_fsblk_t current_block = 0; |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 582 | int ret = 0; |
| 583 | |
| 584 | /* |
| 585 | * Here we try to allocate the requested multiple blocks at once, |
| 586 | * on a best-effort basis. |
| 587 | * To build a branch, we should allocate blocks for |
| 588 | * the indirect blocks(if not allocated yet), and at least |
| 589 | * the first direct block of this branch. That's the |
| 590 | * minimum number of blocks need to allocate(required) |
| 591 | */ |
| 592 | target = blks + indirect_blks; |
| 593 | |
| 594 | while (1) { |
| 595 | count = target; |
| 596 | /* allocating blocks for indirect blocks and direct blocks */ |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 597 | current_block = ext3_new_blocks(handle,inode,goal,&count,err); |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 598 | if (*err) |
| 599 | goto failed_out; |
| 600 | |
| 601 | target -= count; |
| 602 | /* allocate blocks for indirect blocks */ |
| 603 | while (index < indirect_blks && count) { |
| 604 | new_blocks[index++] = current_block++; |
| 605 | count--; |
| 606 | } |
| 607 | |
| 608 | if (count > 0) |
| 609 | break; |
| 610 | } |
| 611 | |
| 612 | /* save the new block number for the first direct block */ |
| 613 | new_blocks[index] = current_block; |
| 614 | |
| 615 | /* total number of blocks allocated for direct blocks */ |
| 616 | ret = count; |
| 617 | *err = 0; |
| 618 | return ret; |
| 619 | failed_out: |
| 620 | for (i = 0; i <index; i++) |
| 621 | ext3_free_blocks(handle, inode, new_blocks[i], 1); |
| 622 | return ret; |
| 623 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 624 | |
| 625 | /** |
| 626 | * ext3_alloc_branch - allocate and set up a chain of blocks. |
Namhyung Kim | a4c18ad | 2010-10-19 00:34:35 +0900 | [diff] [blame] | 627 | * @handle: handle for this transaction |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 628 | * @inode: owner |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 629 | * @indirect_blks: number of allocated indirect blocks |
| 630 | * @blks: number of allocated direct blocks |
Namhyung Kim | a4c18ad | 2010-10-19 00:34:35 +0900 | [diff] [blame] | 631 | * @goal: preferred place for allocation |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 632 | * @offsets: offsets (in the blocks) to store the pointers to next. |
| 633 | * @branch: place to store the chain in. |
| 634 | * |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 635 | * This function allocates blocks, zeroes out all but the last one, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 636 | * links them into chain and (if we are synchronous) writes them to disk. |
| 637 | * In other words, it prepares a branch that can be spliced onto the |
| 638 | * inode. It stores the information about that chain in the branch[], in |
| 639 | * the same format as ext3_get_branch() would do. We are calling it after |
| 640 | * we had read the existing part of chain and partial points to the last |
| 641 | * triple of that (one with zero ->key). Upon the exit we have the same |
Glauber de Oliveira Costa | 5b11687 | 2005-10-30 15:02:48 -0800 | [diff] [blame] | 642 | * picture as after the successful ext3_get_block(), except that in one |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 643 | * place chain is disconnected - *branch->p is still zero (we did not |
| 644 | * set the last link), but branch->key contains the number that should |
| 645 | * be placed into *branch->p to fill that gap. |
| 646 | * |
| 647 | * If allocation fails we free all blocks we've allocated (and forget |
| 648 | * their buffer_heads) and return the error value the from failed |
| 649 | * ext3_alloc_block() (normally -ENOSPC). Otherwise we set the chain |
| 650 | * as described above and return 0. |
| 651 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 652 | static int ext3_alloc_branch(handle_t *handle, struct inode *inode, |
Mingming Cao | 43d23f9 | 2006-06-25 05:48:07 -0700 | [diff] [blame] | 653 | int indirect_blks, int *blks, ext3_fsblk_t goal, |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 654 | int *offsets, Indirect *branch) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 655 | { |
| 656 | int blocksize = inode->i_sb->s_blocksize; |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 657 | int i, n = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 658 | int err = 0; |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 659 | struct buffer_head *bh; |
| 660 | int num; |
Mingming Cao | 43d23f9 | 2006-06-25 05:48:07 -0700 | [diff] [blame] | 661 | ext3_fsblk_t new_blocks[4]; |
| 662 | ext3_fsblk_t current_block; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 663 | |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 664 | num = ext3_alloc_blocks(handle, inode, goal, indirect_blks, |
| 665 | *blks, new_blocks, &err); |
| 666 | if (err) |
| 667 | return err; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 668 | |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 669 | branch[0].key = cpu_to_le32(new_blocks[0]); |
| 670 | /* |
| 671 | * metadata blocks and data blocks are allocated. |
| 672 | */ |
| 673 | for (n = 1; n <= indirect_blks; n++) { |
| 674 | /* |
| 675 | * Get buffer_head for parent block, zero it out |
| 676 | * and set the pointer to new one, then send |
| 677 | * parent to disk. |
| 678 | */ |
| 679 | bh = sb_getblk(inode->i_sb, new_blocks[n-1]); |
| 680 | branch[n].bh = bh; |
| 681 | lock_buffer(bh); |
| 682 | BUFFER_TRACE(bh, "call get_create_access"); |
| 683 | err = ext3_journal_get_create_access(handle, bh); |
| 684 | if (err) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 685 | unlock_buffer(bh); |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 686 | brelse(bh); |
| 687 | goto failed; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 688 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 689 | |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 690 | memset(bh->b_data, 0, blocksize); |
| 691 | branch[n].p = (__le32 *) bh->b_data + offsets[n]; |
| 692 | branch[n].key = cpu_to_le32(new_blocks[n]); |
| 693 | *branch[n].p = branch[n].key; |
| 694 | if ( n == indirect_blks) { |
| 695 | current_block = new_blocks[n]; |
| 696 | /* |
| 697 | * End of chain, update the last new metablock of |
| 698 | * the chain to point to the new allocated |
| 699 | * data blocks numbers |
| 700 | */ |
| 701 | for (i=1; i < num; i++) |
| 702 | *(branch[n].p + i) = cpu_to_le32(++current_block); |
| 703 | } |
| 704 | BUFFER_TRACE(bh, "marking uptodate"); |
| 705 | set_buffer_uptodate(bh); |
| 706 | unlock_buffer(bh); |
| 707 | |
| 708 | BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata"); |
| 709 | err = ext3_journal_dirty_metadata(handle, bh); |
| 710 | if (err) |
| 711 | goto failed; |
| 712 | } |
| 713 | *blks = num; |
| 714 | return err; |
| 715 | failed: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 716 | /* Allocation failed, free what we already allocated */ |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 717 | for (i = 1; i <= n ; i++) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 718 | BUFFER_TRACE(branch[i].bh, "call journal_forget"); |
| 719 | ext3_journal_forget(handle, branch[i].bh); |
| 720 | } |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 721 | for (i = 0; i <indirect_blks; i++) |
| 722 | ext3_free_blocks(handle, inode, new_blocks[i], 1); |
| 723 | |
| 724 | ext3_free_blocks(handle, inode, new_blocks[i], num); |
| 725 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 726 | return err; |
| 727 | } |
| 728 | |
| 729 | /** |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 730 | * ext3_splice_branch - splice the allocated branch onto inode. |
Namhyung Kim | a4c18ad | 2010-10-19 00:34:35 +0900 | [diff] [blame] | 731 | * @handle: handle for this transaction |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 732 | * @inode: owner |
| 733 | * @block: (logical) number of block we are adding |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 734 | * @where: location of missing link |
| 735 | * @num: number of indirect blocks we are adding |
| 736 | * @blks: number of direct blocks we are adding |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 737 | * |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 738 | * This function fills the missing link and does all housekeeping needed in |
| 739 | * inode (->i_blocks, etc.). In case of success we end up with the full |
| 740 | * chain to new block and return 0. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 741 | */ |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 742 | static int ext3_splice_branch(handle_t *handle, struct inode *inode, |
| 743 | long block, Indirect *where, int num, int blks) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 744 | { |
| 745 | int i; |
| 746 | int err = 0; |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 747 | struct ext3_block_alloc_info *block_i; |
Mingming Cao | 43d23f9 | 2006-06-25 05:48:07 -0700 | [diff] [blame] | 748 | ext3_fsblk_t current_block; |
Jan Kara | fe8bc91 | 2009-10-16 19:26:15 +0200 | [diff] [blame] | 749 | struct ext3_inode_info *ei = EXT3_I(inode); |
Kazuya Mio | ac1334b | 2012-02-08 16:34:29 +0900 | [diff] [blame] | 750 | struct timespec now; |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 751 | |
Jan Kara | fe8bc91 | 2009-10-16 19:26:15 +0200 | [diff] [blame] | 752 | block_i = ei->i_block_alloc_info; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 753 | /* |
| 754 | * If we're splicing into a [td]indirect block (as opposed to the |
| 755 | * inode) then we need to get write access to the [td]indirect block |
| 756 | * before the splice. |
| 757 | */ |
| 758 | if (where->bh) { |
| 759 | BUFFER_TRACE(where->bh, "get_write_access"); |
| 760 | err = ext3_journal_get_write_access(handle, where->bh); |
| 761 | if (err) |
| 762 | goto err_out; |
| 763 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 764 | /* That's it */ |
| 765 | |
| 766 | *where->p = where->key; |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 767 | |
| 768 | /* |
| 769 | * Update the host buffer_head or inode to point to more just allocated |
| 770 | * direct blocks blocks |
| 771 | */ |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 772 | if (num == 0 && blks > 1) { |
Mingming Cao | 5dea517 | 2006-05-03 19:55:12 -0700 | [diff] [blame] | 773 | current_block = le32_to_cpu(where->key) + 1; |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 774 | for (i = 1; i < blks; i++) |
| 775 | *(where->p + i ) = cpu_to_le32(current_block++); |
| 776 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 777 | |
| 778 | /* |
| 779 | * update the most recently allocated logical & physical block |
| 780 | * in i_block_alloc_info, to assist find the proper goal block for next |
| 781 | * allocation |
| 782 | */ |
| 783 | if (block_i) { |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 784 | block_i->last_alloc_logical_block = block + blks - 1; |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 785 | block_i->last_alloc_physical_block = |
Mingming Cao | 5dea517 | 2006-05-03 19:55:12 -0700 | [diff] [blame] | 786 | le32_to_cpu(where[num].key) + blks - 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 787 | } |
| 788 | |
| 789 | /* We are done with atomic stuff, now do the rest of housekeeping */ |
Kazuya Mio | ac1334b | 2012-02-08 16:34:29 +0900 | [diff] [blame] | 790 | now = CURRENT_TIME_SEC; |
| 791 | if (!timespec_equal(&inode->i_ctime, &now) || !where->bh) { |
| 792 | inode->i_ctime = now; |
| 793 | ext3_mark_inode_dirty(handle, inode); |
| 794 | } |
Jan Kara | fe8bc91 | 2009-10-16 19:26:15 +0200 | [diff] [blame] | 795 | /* ext3_mark_inode_dirty already updated i_sync_tid */ |
| 796 | atomic_set(&ei->i_datasync_tid, handle->h_transaction->t_tid); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 797 | |
| 798 | /* had we spliced it onto indirect block? */ |
| 799 | if (where->bh) { |
| 800 | /* |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 801 | * If we spliced it onto an indirect block, we haven't |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 802 | * altered the inode. Note however that if it is being spliced |
| 803 | * onto an indirect block at the very end of the file (the |
| 804 | * file is growing) then we *will* alter the inode to reflect |
| 805 | * the new i_size. But that is not done here - it is done in |
| 806 | * generic_commit_write->__mark_inode_dirty->ext3_dirty_inode. |
| 807 | */ |
| 808 | jbd_debug(5, "splicing indirect only\n"); |
| 809 | BUFFER_TRACE(where->bh, "call ext3_journal_dirty_metadata"); |
| 810 | err = ext3_journal_dirty_metadata(handle, where->bh); |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 811 | if (err) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 812 | goto err_out; |
| 813 | } else { |
| 814 | /* |
| 815 | * OK, we spliced it into the inode itself on a direct block. |
| 816 | * Inode was dirtied above. |
| 817 | */ |
| 818 | jbd_debug(5, "splicing direct\n"); |
| 819 | } |
| 820 | return err; |
| 821 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 822 | err_out: |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 823 | for (i = 1; i <= num; i++) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 824 | BUFFER_TRACE(where[i].bh, "call journal_forget"); |
| 825 | ext3_journal_forget(handle, where[i].bh); |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 826 | ext3_free_blocks(handle,inode,le32_to_cpu(where[i-1].key),1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 827 | } |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 828 | ext3_free_blocks(handle, inode, le32_to_cpu(where[num].key), blks); |
| 829 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 830 | return err; |
| 831 | } |
| 832 | |
| 833 | /* |
| 834 | * Allocation strategy is simple: if we have to allocate something, we will |
| 835 | * have to go the whole way to leaf. So let's do it before attaching anything |
| 836 | * to tree, set linkage between the newborn blocks, write them if sync is |
| 837 | * required, recheck the path, free and repeat if check fails, otherwise |
| 838 | * set the last missing link (that will protect us from any truncate-generated |
| 839 | * removals - all blocks on the path are immune now) and possibly force the |
| 840 | * write on the parent block. |
| 841 | * That has a nice additional property: no special recovery from the failed |
| 842 | * allocations is needed - we simply release blocks and do not touch anything |
| 843 | * reachable from inode. |
| 844 | * |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 845 | * `handle' can be NULL if create == 0. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 846 | * |
| 847 | * The BKL may not be held on entry here. Be sure to take it early. |
Mingming Cao | 89747d3 | 2006-03-26 01:37:55 -0800 | [diff] [blame] | 848 | * return > 0, # of blocks mapped or allocated. |
| 849 | * return = 0, if plain lookup failed. |
| 850 | * return < 0, error case. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 851 | */ |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 852 | int ext3_get_blocks_handle(handle_t *handle, struct inode *inode, |
| 853 | sector_t iblock, unsigned long maxblocks, |
| 854 | struct buffer_head *bh_result, |
Jan Kara | 43237b5 | 2009-05-20 18:41:58 +0200 | [diff] [blame] | 855 | int create) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 856 | { |
| 857 | int err = -EIO; |
| 858 | int offsets[4]; |
| 859 | Indirect chain[4]; |
| 860 | Indirect *partial; |
Mingming Cao | 43d23f9 | 2006-06-25 05:48:07 -0700 | [diff] [blame] | 861 | ext3_fsblk_t goal; |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 862 | int indirect_blks; |
Mingming Cao | 89747d3 | 2006-03-26 01:37:55 -0800 | [diff] [blame] | 863 | int blocks_to_boundary = 0; |
| 864 | int depth; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 865 | struct ext3_inode_info *ei = EXT3_I(inode); |
Mingming Cao | 89747d3 | 2006-03-26 01:37:55 -0800 | [diff] [blame] | 866 | int count = 0; |
Mingming Cao | 43d23f9 | 2006-06-25 05:48:07 -0700 | [diff] [blame] | 867 | ext3_fsblk_t first_block = 0; |
Mingming Cao | 89747d3 | 2006-03-26 01:37:55 -0800 | [diff] [blame] | 868 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 869 | |
Lukas Czerner | 785c4bc | 2011-05-23 18:33:01 +0200 | [diff] [blame] | 870 | trace_ext3_get_blocks_enter(inode, iblock, maxblocks, create); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 871 | J_ASSERT(handle != NULL || create == 0); |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 872 | depth = ext3_block_to_path(inode,iblock,offsets,&blocks_to_boundary); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 873 | |
| 874 | if (depth == 0) |
| 875 | goto out; |
| 876 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 877 | partial = ext3_get_branch(inode, depth, offsets, chain, &err); |
| 878 | |
| 879 | /* Simplest case - block found, no allocation needed */ |
| 880 | if (!partial) { |
Mingming Cao | 5dea517 | 2006-05-03 19:55:12 -0700 | [diff] [blame] | 881 | first_block = le32_to_cpu(chain[depth - 1].key); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 882 | clear_buffer_new(bh_result); |
Mingming Cao | 89747d3 | 2006-03-26 01:37:55 -0800 | [diff] [blame] | 883 | count++; |
| 884 | /*map more blocks*/ |
| 885 | while (count < maxblocks && count <= blocks_to_boundary) { |
Mingming Cao | 43d23f9 | 2006-06-25 05:48:07 -0700 | [diff] [blame] | 886 | ext3_fsblk_t blk; |
Mingming Cao | 5dea517 | 2006-05-03 19:55:12 -0700 | [diff] [blame] | 887 | |
Jan Kara | e8ef7aa | 2009-06-17 16:26:23 -0700 | [diff] [blame] | 888 | if (!verify_chain(chain, chain + depth - 1)) { |
Mingming Cao | 89747d3 | 2006-03-26 01:37:55 -0800 | [diff] [blame] | 889 | /* |
| 890 | * Indirect block might be removed by |
| 891 | * truncate while we were reading it. |
| 892 | * Handling of that case: forget what we've |
| 893 | * got now. Flag the err as EAGAIN, so it |
| 894 | * will reread. |
| 895 | */ |
| 896 | err = -EAGAIN; |
| 897 | count = 0; |
| 898 | break; |
| 899 | } |
Mingming Cao | 5dea517 | 2006-05-03 19:55:12 -0700 | [diff] [blame] | 900 | blk = le32_to_cpu(*(chain[depth-1].p + count)); |
| 901 | |
| 902 | if (blk == first_block + count) |
Mingming Cao | 89747d3 | 2006-03-26 01:37:55 -0800 | [diff] [blame] | 903 | count++; |
| 904 | else |
| 905 | break; |
| 906 | } |
| 907 | if (err != -EAGAIN) |
| 908 | goto got_it; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 909 | } |
| 910 | |
| 911 | /* Next simple case - plain lookup or failed read of indirect block */ |
Mingming Cao | fe55c45 | 2005-05-01 08:59:20 -0700 | [diff] [blame] | 912 | if (!create || err == -EIO) |
| 913 | goto cleanup; |
| 914 | |
Jan Kara | 40680f2f | 2011-05-24 22:24:47 +0200 | [diff] [blame] | 915 | /* |
| 916 | * Block out ext3_truncate while we alter the tree |
| 917 | */ |
Arjan van de Ven | 9746151 | 2006-03-23 03:00:42 -0800 | [diff] [blame] | 918 | mutex_lock(&ei->truncate_mutex); |
Mingming Cao | fe55c45 | 2005-05-01 08:59:20 -0700 | [diff] [blame] | 919 | |
| 920 | /* |
| 921 | * If the indirect block is missing while we are reading |
| 922 | * the chain(ext3_get_branch() returns -EAGAIN err), or |
| 923 | * if the chain has been changed after we grab the semaphore, |
| 924 | * (either because another process truncated this branch, or |
| 925 | * another get_block allocated this branch) re-grab the chain to see if |
| 926 | * the request block has been allocated or not. |
| 927 | * |
| 928 | * Since we already block the truncate/other get_block |
| 929 | * at this point, we will have the current copy of the chain when we |
| 930 | * splice the branch into the tree. |
| 931 | */ |
| 932 | if (err == -EAGAIN || !verify_chain(chain, partial)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 933 | while (partial > chain) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 934 | brelse(partial->bh); |
| 935 | partial--; |
| 936 | } |
Mingming Cao | fe55c45 | 2005-05-01 08:59:20 -0700 | [diff] [blame] | 937 | partial = ext3_get_branch(inode, depth, offsets, chain, &err); |
| 938 | if (!partial) { |
Mingming Cao | 89747d3 | 2006-03-26 01:37:55 -0800 | [diff] [blame] | 939 | count++; |
Arjan van de Ven | 9746151 | 2006-03-23 03:00:42 -0800 | [diff] [blame] | 940 | mutex_unlock(&ei->truncate_mutex); |
Mingming Cao | fe55c45 | 2005-05-01 08:59:20 -0700 | [diff] [blame] | 941 | if (err) |
| 942 | goto cleanup; |
| 943 | clear_buffer_new(bh_result); |
| 944 | goto got_it; |
| 945 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 946 | } |
| 947 | |
| 948 | /* |
Mingming Cao | fe55c45 | 2005-05-01 08:59:20 -0700 | [diff] [blame] | 949 | * Okay, we need to do block allocation. Lazily initialize the block |
| 950 | * allocation info here if necessary |
| 951 | */ |
| 952 | if (S_ISREG(inode->i_mode) && (!ei->i_block_alloc_info)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 953 | ext3_init_block_alloc_info(inode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 954 | |
Akinobu Mita | fb01bfd | 2008-02-06 01:40:16 -0800 | [diff] [blame] | 955 | goal = ext3_find_goal(inode, iblock, partial); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 956 | |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 957 | /* the number of blocks need to allocate for [d,t]indirect blocks */ |
| 958 | indirect_blks = (chain + depth) - partial - 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 959 | |
| 960 | /* |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 961 | * Next look up the indirect map to count the totoal number of |
| 962 | * direct blocks to allocate for this branch. |
| 963 | */ |
| 964 | count = ext3_blks_to_allocate(partial, indirect_blks, |
| 965 | maxblocks, blocks_to_boundary); |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 966 | err = ext3_alloc_branch(handle, inode, indirect_blks, &count, goal, |
Mingming Cao | fe55c45 | 2005-05-01 08:59:20 -0700 | [diff] [blame] | 967 | offsets + (partial - chain), partial); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 968 | |
Mingming Cao | fe55c45 | 2005-05-01 08:59:20 -0700 | [diff] [blame] | 969 | /* |
| 970 | * The ext3_splice_branch call will free and forget any buffers |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 971 | * on the new chain if there is a failure, but that risks using |
| 972 | * up transaction credits, especially for bitmaps where the |
| 973 | * credits cannot be returned. Can we handle this somehow? We |
Mingming Cao | fe55c45 | 2005-05-01 08:59:20 -0700 | [diff] [blame] | 974 | * may need to return -EAGAIN upwards in the worst case. --sct |
| 975 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 976 | if (!err) |
Mingming Cao | b47b247 | 2006-03-26 01:37:56 -0800 | [diff] [blame] | 977 | err = ext3_splice_branch(handle, inode, iblock, |
| 978 | partial, indirect_blks, count); |
Arjan van de Ven | 9746151 | 2006-03-23 03:00:42 -0800 | [diff] [blame] | 979 | mutex_unlock(&ei->truncate_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 980 | if (err) |
| 981 | goto cleanup; |
| 982 | |
| 983 | set_buffer_new(bh_result); |
Mingming Cao | fe55c45 | 2005-05-01 08:59:20 -0700 | [diff] [blame] | 984 | got_it: |
| 985 | map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key)); |
Suparna Bhattacharya | 20acaa1 | 2006-09-16 12:15:58 -0700 | [diff] [blame] | 986 | if (count > blocks_to_boundary) |
Mingming Cao | fe55c45 | 2005-05-01 08:59:20 -0700 | [diff] [blame] | 987 | set_buffer_boundary(bh_result); |
Mingming Cao | 89747d3 | 2006-03-26 01:37:55 -0800 | [diff] [blame] | 988 | err = count; |
Mingming Cao | fe55c45 | 2005-05-01 08:59:20 -0700 | [diff] [blame] | 989 | /* Clean up and exit */ |
| 990 | partial = chain + depth - 1; /* the whole chain */ |
| 991 | cleanup: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 992 | while (partial > chain) { |
Mingming Cao | fe55c45 | 2005-05-01 08:59:20 -0700 | [diff] [blame] | 993 | BUFFER_TRACE(partial->bh, "call brelse"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 994 | brelse(partial->bh); |
| 995 | partial--; |
| 996 | } |
Mingming Cao | fe55c45 | 2005-05-01 08:59:20 -0700 | [diff] [blame] | 997 | BUFFER_TRACE(bh_result, "returned"); |
| 998 | out: |
Lukas Czerner | 785c4bc | 2011-05-23 18:33:01 +0200 | [diff] [blame] | 999 | trace_ext3_get_blocks_exit(inode, iblock, |
| 1000 | depth ? le32_to_cpu(chain[depth-1].key) : 0, |
| 1001 | count, err); |
Mingming Cao | fe55c45 | 2005-05-01 08:59:20 -0700 | [diff] [blame] | 1002 | return err; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1003 | } |
| 1004 | |
Jan Kara | bd1939d | 2008-02-06 01:40:21 -0800 | [diff] [blame] | 1005 | /* Maximum number of blocks we map for direct IO at once. */ |
| 1006 | #define DIO_MAX_BLOCKS 4096 |
| 1007 | /* |
| 1008 | * Number of credits we need for writing DIO_MAX_BLOCKS: |
| 1009 | * We need sb + group descriptor + bitmap + inode -> 4 |
| 1010 | * For B blocks with A block pointers per block we need: |
| 1011 | * 1 (triple ind.) + (B/A/A + 2) (doubly ind.) + (B/A + 2) (indirect). |
| 1012 | * If we plug in 4096 for B and 256 for A (for 1KB block size), we get 25. |
| 1013 | */ |
| 1014 | #define DIO_CREDITS 25 |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1015 | |
Badari Pulavarty | f91a2ad | 2006-03-26 01:38:04 -0800 | [diff] [blame] | 1016 | static int ext3_get_block(struct inode *inode, sector_t iblock, |
| 1017 | struct buffer_head *bh_result, int create) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1018 | { |
Dmitriy Monakhov | 3e4fdaf | 2007-02-10 01:46:35 -0800 | [diff] [blame] | 1019 | handle_t *handle = ext3_journal_current_handle(); |
Jan Kara | bd1939d | 2008-02-06 01:40:21 -0800 | [diff] [blame] | 1020 | int ret = 0, started = 0; |
Badari Pulavarty | 1d8fa7a | 2006-03-26 01:38:02 -0800 | [diff] [blame] | 1021 | unsigned max_blocks = bh_result->b_size >> inode->i_blkbits; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1022 | |
Jan Kara | bd1939d | 2008-02-06 01:40:21 -0800 | [diff] [blame] | 1023 | if (create && !handle) { /* Direct IO write... */ |
| 1024 | if (max_blocks > DIO_MAX_BLOCKS) |
| 1025 | max_blocks = DIO_MAX_BLOCKS; |
| 1026 | handle = ext3_journal_start(inode, DIO_CREDITS + |
Dmitry Monakhov | c459001 | 2009-12-09 03:05:30 +0300 | [diff] [blame] | 1027 | EXT3_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb)); |
Jan Kara | bd1939d | 2008-02-06 01:40:21 -0800 | [diff] [blame] | 1028 | if (IS_ERR(handle)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1029 | ret = PTR_ERR(handle); |
Jan Kara | bd1939d | 2008-02-06 01:40:21 -0800 | [diff] [blame] | 1030 | goto out; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1031 | } |
Jan Kara | bd1939d | 2008-02-06 01:40:21 -0800 | [diff] [blame] | 1032 | started = 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1033 | } |
| 1034 | |
Jan Kara | bd1939d | 2008-02-06 01:40:21 -0800 | [diff] [blame] | 1035 | ret = ext3_get_blocks_handle(handle, inode, iblock, |
Jan Kara | 43237b5 | 2009-05-20 18:41:58 +0200 | [diff] [blame] | 1036 | max_blocks, bh_result, create); |
Jan Kara | bd1939d | 2008-02-06 01:40:21 -0800 | [diff] [blame] | 1037 | if (ret > 0) { |
| 1038 | bh_result->b_size = (ret << inode->i_blkbits); |
| 1039 | ret = 0; |
Mingming Cao | 89747d3 | 2006-03-26 01:37:55 -0800 | [diff] [blame] | 1040 | } |
Jan Kara | bd1939d | 2008-02-06 01:40:21 -0800 | [diff] [blame] | 1041 | if (started) |
| 1042 | ext3_journal_stop(handle); |
| 1043 | out: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1044 | return ret; |
| 1045 | } |
| 1046 | |
Josef Bacik | 68c9d70 | 2008-10-03 17:32:43 -0400 | [diff] [blame] | 1047 | int ext3_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, |
| 1048 | u64 start, u64 len) |
| 1049 | { |
| 1050 | return generic_block_fiemap(inode, fieinfo, start, len, |
| 1051 | ext3_get_block); |
| 1052 | } |
| 1053 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1054 | /* |
| 1055 | * `handle' can be NULL if create is zero |
| 1056 | */ |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 1057 | struct buffer_head *ext3_getblk(handle_t *handle, struct inode *inode, |
| 1058 | long block, int create, int *errp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1059 | { |
| 1060 | struct buffer_head dummy; |
| 1061 | int fatal = 0, err; |
| 1062 | |
| 1063 | J_ASSERT(handle != NULL || create == 0); |
| 1064 | |
| 1065 | dummy.b_state = 0; |
| 1066 | dummy.b_blocknr = -1000; |
| 1067 | buffer_trace_init(&dummy.b_history); |
Mingming Cao | 89747d3 | 2006-03-26 01:37:55 -0800 | [diff] [blame] | 1068 | err = ext3_get_blocks_handle(handle, inode, block, 1, |
Jan Kara | 43237b5 | 2009-05-20 18:41:58 +0200 | [diff] [blame] | 1069 | &dummy, create); |
Badari Pulavarty | 3665d0e | 2006-09-08 09:48:21 -0700 | [diff] [blame] | 1070 | /* |
| 1071 | * ext3_get_blocks_handle() returns number of blocks |
| 1072 | * mapped. 0 in case of a HOLE. |
| 1073 | */ |
| 1074 | if (err > 0) { |
| 1075 | if (err > 1) |
| 1076 | WARN_ON(1); |
Mingming Cao | 89747d3 | 2006-03-26 01:37:55 -0800 | [diff] [blame] | 1077 | err = 0; |
Mingming Cao | 89747d3 | 2006-03-26 01:37:55 -0800 | [diff] [blame] | 1078 | } |
| 1079 | *errp = err; |
| 1080 | if (!err && buffer_mapped(&dummy)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1081 | struct buffer_head *bh; |
| 1082 | bh = sb_getblk(inode->i_sb, dummy.b_blocknr); |
Glauber de Oliveira Costa | 2973dfd | 2005-10-30 15:03:05 -0800 | [diff] [blame] | 1083 | if (!bh) { |
| 1084 | *errp = -EIO; |
| 1085 | goto err; |
| 1086 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1087 | if (buffer_new(&dummy)) { |
| 1088 | J_ASSERT(create != 0); |
Stephen Hemminger | c80544d | 2007-10-18 03:07:05 -0700 | [diff] [blame] | 1089 | J_ASSERT(handle != NULL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1090 | |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 1091 | /* |
| 1092 | * Now that we do not always journal data, we should |
| 1093 | * keep in mind whether this should always journal the |
| 1094 | * new buffer as metadata. For now, regular file |
| 1095 | * writes use ext3_get_block instead, so it's not a |
| 1096 | * problem. |
| 1097 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1098 | lock_buffer(bh); |
| 1099 | BUFFER_TRACE(bh, "call get_create_access"); |
| 1100 | fatal = ext3_journal_get_create_access(handle, bh); |
| 1101 | if (!fatal && !buffer_uptodate(bh)) { |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 1102 | memset(bh->b_data,0,inode->i_sb->s_blocksize); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1103 | set_buffer_uptodate(bh); |
| 1104 | } |
| 1105 | unlock_buffer(bh); |
| 1106 | BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata"); |
| 1107 | err = ext3_journal_dirty_metadata(handle, bh); |
| 1108 | if (!fatal) |
| 1109 | fatal = err; |
| 1110 | } else { |
| 1111 | BUFFER_TRACE(bh, "not a new buffer"); |
| 1112 | } |
| 1113 | if (fatal) { |
| 1114 | *errp = fatal; |
| 1115 | brelse(bh); |
| 1116 | bh = NULL; |
| 1117 | } |
| 1118 | return bh; |
| 1119 | } |
Glauber de Oliveira Costa | 2973dfd | 2005-10-30 15:03:05 -0800 | [diff] [blame] | 1120 | err: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1121 | return NULL; |
| 1122 | } |
| 1123 | |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 1124 | struct buffer_head *ext3_bread(handle_t *handle, struct inode *inode, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1125 | int block, int create, int *err) |
| 1126 | { |
| 1127 | struct buffer_head * bh; |
| 1128 | |
| 1129 | bh = ext3_getblk(handle, inode, block, create, err); |
| 1130 | if (!bh) |
| 1131 | return bh; |
Zheng Liu | d03e129 | 2011-12-05 15:55:11 +0800 | [diff] [blame] | 1132 | if (bh_uptodate_or_lock(bh)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1133 | return bh; |
Zheng Liu | d03e129 | 2011-12-05 15:55:11 +0800 | [diff] [blame] | 1134 | get_bh(bh); |
| 1135 | bh->b_end_io = end_buffer_read_sync; |
| 1136 | submit_bh(READ | REQ_META | REQ_PRIO, bh); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1137 | wait_on_buffer(bh); |
| 1138 | if (buffer_uptodate(bh)) |
| 1139 | return bh; |
| 1140 | put_bh(bh); |
| 1141 | *err = -EIO; |
| 1142 | return NULL; |
| 1143 | } |
| 1144 | |
| 1145 | static int walk_page_buffers( handle_t *handle, |
| 1146 | struct buffer_head *head, |
| 1147 | unsigned from, |
| 1148 | unsigned to, |
| 1149 | int *partial, |
| 1150 | int (*fn)( handle_t *handle, |
| 1151 | struct buffer_head *bh)) |
| 1152 | { |
| 1153 | struct buffer_head *bh; |
| 1154 | unsigned block_start, block_end; |
| 1155 | unsigned blocksize = head->b_size; |
| 1156 | int err, ret = 0; |
| 1157 | struct buffer_head *next; |
| 1158 | |
| 1159 | for ( bh = head, block_start = 0; |
| 1160 | ret == 0 && (bh != head || !block_start); |
Dave Kleikamp | e9ad562 | 2006-09-27 01:49:35 -0700 | [diff] [blame] | 1161 | block_start = block_end, bh = next) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1162 | { |
| 1163 | next = bh->b_this_page; |
| 1164 | block_end = block_start + blocksize; |
| 1165 | if (block_end <= from || block_start >= to) { |
| 1166 | if (partial && !buffer_uptodate(bh)) |
| 1167 | *partial = 1; |
| 1168 | continue; |
| 1169 | } |
| 1170 | err = (*fn)(handle, bh); |
| 1171 | if (!ret) |
| 1172 | ret = err; |
| 1173 | } |
| 1174 | return ret; |
| 1175 | } |
| 1176 | |
| 1177 | /* |
| 1178 | * To preserve ordering, it is essential that the hole instantiation and |
| 1179 | * the data write be encapsulated in a single transaction. We cannot |
| 1180 | * close off a transaction and start a new one between the ext3_get_block() |
| 1181 | * and the commit_write(). So doing the journal_start at the start of |
| 1182 | * prepare_write() is the right place. |
| 1183 | * |
| 1184 | * Also, this function can nest inside ext3_writepage() -> |
| 1185 | * block_write_full_page(). In that case, we *know* that ext3_writepage() |
| 1186 | * has generated enough buffer credits to do the whole page. So we won't |
| 1187 | * block on the journal in that case, which is good, because the caller may |
| 1188 | * be PF_MEMALLOC. |
| 1189 | * |
| 1190 | * By accident, ext3 can be reentered when a transaction is open via |
| 1191 | * quota file writes. If we were to commit the transaction while thus |
| 1192 | * reentered, there can be a deadlock - we would be holding a quota |
| 1193 | * lock, and the commit would never complete if another thread had a |
| 1194 | * transaction open and was blocking on the quota lock - a ranking |
| 1195 | * violation. |
| 1196 | * |
| 1197 | * So what we do is to rely on the fact that journal_stop/journal_start |
| 1198 | * will _not_ run commit under these circumstances because handle->h_ref |
| 1199 | * is elevated. We'll still have enough credits for the tiny quotafile |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 1200 | * write. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1201 | */ |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 1202 | static int do_journal_get_write_access(handle_t *handle, |
| 1203 | struct buffer_head *bh) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1204 | { |
Jan Kara | 5f11e6a | 2010-08-05 12:38:26 +0200 | [diff] [blame] | 1205 | int dirty = buffer_dirty(bh); |
| 1206 | int ret; |
| 1207 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1208 | if (!buffer_mapped(bh) || buffer_freed(bh)) |
| 1209 | return 0; |
Jan Kara | 5f11e6a | 2010-08-05 12:38:26 +0200 | [diff] [blame] | 1210 | /* |
| 1211 | * __block_prepare_write() could have dirtied some buffers. Clean |
| 1212 | * the dirty bit as jbd2_journal_get_write_access() could complain |
| 1213 | * otherwise about fs integrity issues. Setting of the dirty bit |
| 1214 | * by __block_prepare_write() isn't a real problem here as we clear |
| 1215 | * the bit before releasing a page lock and thus writeback cannot |
| 1216 | * ever write the buffer. |
| 1217 | */ |
| 1218 | if (dirty) |
| 1219 | clear_buffer_dirty(bh); |
| 1220 | ret = ext3_journal_get_write_access(handle, bh); |
| 1221 | if (!ret && dirty) |
| 1222 | ret = ext3_journal_dirty_metadata(handle, bh); |
| 1223 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1224 | } |
| 1225 | |
Jan Kara | 68eb3db | 2009-12-01 16:53:06 +0100 | [diff] [blame] | 1226 | /* |
| 1227 | * Truncate blocks that were not used by write. We have to truncate the |
| 1228 | * pagecache as well so that corresponding buffers get properly unmapped. |
| 1229 | */ |
| 1230 | static void ext3_truncate_failed_write(struct inode *inode) |
| 1231 | { |
| 1232 | truncate_inode_pages(inode->i_mapping, inode->i_size); |
| 1233 | ext3_truncate(inode); |
| 1234 | } |
| 1235 | |
Jan Kara | ee3e77f | 2011-06-03 21:58:11 +0200 | [diff] [blame] | 1236 | /* |
| 1237 | * Truncate blocks that were not used by direct IO write. We have to zero out |
| 1238 | * the last file block as well because direct IO might have written to it. |
| 1239 | */ |
| 1240 | static void ext3_truncate_failed_direct_write(struct inode *inode) |
| 1241 | { |
| 1242 | ext3_block_truncate_page(inode, inode->i_size); |
| 1243 | ext3_truncate(inode); |
| 1244 | } |
| 1245 | |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1246 | static int ext3_write_begin(struct file *file, struct address_space *mapping, |
| 1247 | loff_t pos, unsigned len, unsigned flags, |
| 1248 | struct page **pagep, void **fsdata) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1249 | { |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1250 | struct inode *inode = mapping->host; |
Jan Kara | 695f6ae | 2009-04-02 16:57:17 -0700 | [diff] [blame] | 1251 | int ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1252 | handle_t *handle; |
| 1253 | int retries = 0; |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1254 | struct page *page; |
| 1255 | pgoff_t index; |
| 1256 | unsigned from, to; |
Jan Kara | 695f6ae | 2009-04-02 16:57:17 -0700 | [diff] [blame] | 1257 | /* Reserve one block more for addition to orphan list in case |
| 1258 | * we allocate blocks but write fails for some reason */ |
| 1259 | int needed_blocks = ext3_writepage_trans_blocks(inode) + 1; |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1260 | |
Lukas Czerner | 785c4bc | 2011-05-23 18:33:01 +0200 | [diff] [blame] | 1261 | trace_ext3_write_begin(inode, pos, len, flags); |
| 1262 | |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1263 | index = pos >> PAGE_CACHE_SHIFT; |
| 1264 | from = pos & (PAGE_CACHE_SIZE - 1); |
| 1265 | to = from + len; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1266 | |
| 1267 | retry: |
Nick Piggin | 54566b2 | 2009-01-04 12:00:53 -0800 | [diff] [blame] | 1268 | page = grab_cache_page_write_begin(mapping, index, flags); |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1269 | if (!page) |
| 1270 | return -ENOMEM; |
| 1271 | *pagep = page; |
| 1272 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1273 | handle = ext3_journal_start(inode, needed_blocks); |
Andrew Morton | 1aa9b4b | 2007-04-01 23:49:43 -0700 | [diff] [blame] | 1274 | if (IS_ERR(handle)) { |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1275 | unlock_page(page); |
| 1276 | page_cache_release(page); |
Andrew Morton | 1aa9b4b | 2007-04-01 23:49:43 -0700 | [diff] [blame] | 1277 | ret = PTR_ERR(handle); |
| 1278 | goto out; |
| 1279 | } |
Christoph Hellwig | 6e1db88 | 2010-06-04 11:29:57 +0200 | [diff] [blame] | 1280 | ret = __block_write_begin(page, pos, len, ext3_get_block); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1281 | if (ret) |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1282 | goto write_begin_failed; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1283 | |
| 1284 | if (ext3_should_journal_data(inode)) { |
| 1285 | ret = walk_page_buffers(handle, page_buffers(page), |
| 1286 | from, to, NULL, do_journal_get_write_access); |
| 1287 | } |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1288 | write_begin_failed: |
| 1289 | if (ret) { |
Aneesh Kumar K.V | 5ec8b75 | 2008-10-18 20:28:00 -0700 | [diff] [blame] | 1290 | /* |
| 1291 | * block_write_begin may have instantiated a few blocks |
| 1292 | * outside i_size. Trim these off again. Don't need |
| 1293 | * i_size_read because we hold i_mutex. |
Jan Kara | 695f6ae | 2009-04-02 16:57:17 -0700 | [diff] [blame] | 1294 | * |
| 1295 | * Add inode to orphan list in case we crash before truncate |
Jan Kara | 9eaaa2d | 2009-07-13 20:26:52 +0200 | [diff] [blame] | 1296 | * finishes. Do this only if ext3_can_truncate() agrees so |
| 1297 | * that orphan processing code is happy. |
Aneesh Kumar K.V | 5ec8b75 | 2008-10-18 20:28:00 -0700 | [diff] [blame] | 1298 | */ |
Jan Kara | 9eaaa2d | 2009-07-13 20:26:52 +0200 | [diff] [blame] | 1299 | if (pos + len > inode->i_size && ext3_can_truncate(inode)) |
Jan Kara | 695f6ae | 2009-04-02 16:57:17 -0700 | [diff] [blame] | 1300 | ext3_orphan_add(handle, inode); |
| 1301 | ext3_journal_stop(handle); |
| 1302 | unlock_page(page); |
| 1303 | page_cache_release(page); |
| 1304 | if (pos + len > inode->i_size) |
Jan Kara | 68eb3db | 2009-12-01 16:53:06 +0100 | [diff] [blame] | 1305 | ext3_truncate_failed_write(inode); |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1306 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1307 | if (ret == -ENOSPC && ext3_should_retry_alloc(inode->i_sb, &retries)) |
| 1308 | goto retry; |
Andrew Morton | 1aa9b4b | 2007-04-01 23:49:43 -0700 | [diff] [blame] | 1309 | out: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1310 | return ret; |
| 1311 | } |
| 1312 | |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1313 | |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 1314 | int ext3_journal_dirty_data(handle_t *handle, struct buffer_head *bh) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1315 | { |
| 1316 | int err = journal_dirty_data(handle, bh); |
| 1317 | if (err) |
Harvey Harrison | e05b6b5 | 2008-04-28 02:16:15 -0700 | [diff] [blame] | 1318 | ext3_journal_abort_handle(__func__, __func__, |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1319 | bh, handle, err); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1320 | return err; |
| 1321 | } |
| 1322 | |
Jan Kara | 695f6ae | 2009-04-02 16:57:17 -0700 | [diff] [blame] | 1323 | /* For ordered writepage and write_end functions */ |
| 1324 | static int journal_dirty_data_fn(handle_t *handle, struct buffer_head *bh) |
| 1325 | { |
| 1326 | /* |
| 1327 | * Write could have mapped the buffer but it didn't copy the data in |
| 1328 | * yet. So avoid filing such buffer into a transaction. |
| 1329 | */ |
| 1330 | if (buffer_mapped(bh) && buffer_uptodate(bh)) |
| 1331 | return ext3_journal_dirty_data(handle, bh); |
| 1332 | return 0; |
| 1333 | } |
| 1334 | |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1335 | /* For write_end() in data=journal mode */ |
| 1336 | static int write_end_fn(handle_t *handle, struct buffer_head *bh) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1337 | { |
| 1338 | if (!buffer_mapped(bh) || buffer_freed(bh)) |
| 1339 | return 0; |
| 1340 | set_buffer_uptodate(bh); |
| 1341 | return ext3_journal_dirty_metadata(handle, bh); |
| 1342 | } |
| 1343 | |
| 1344 | /* |
Jan Kara | 695f6ae | 2009-04-02 16:57:17 -0700 | [diff] [blame] | 1345 | * This is nasty and subtle: ext3_write_begin() could have allocated blocks |
| 1346 | * for the whole page but later we failed to copy the data in. Update inode |
| 1347 | * size according to what we managed to copy. The rest is going to be |
| 1348 | * truncated in write_end function. |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1349 | */ |
Jan Kara | 695f6ae | 2009-04-02 16:57:17 -0700 | [diff] [blame] | 1350 | static void update_file_sizes(struct inode *inode, loff_t pos, unsigned copied) |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1351 | { |
Jan Kara | 695f6ae | 2009-04-02 16:57:17 -0700 | [diff] [blame] | 1352 | /* What matters to us is i_disksize. We don't write i_size anywhere */ |
| 1353 | if (pos + copied > inode->i_size) |
| 1354 | i_size_write(inode, pos + copied); |
| 1355 | if (pos + copied > EXT3_I(inode)->i_disksize) { |
| 1356 | EXT3_I(inode)->i_disksize = pos + copied; |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1357 | mark_inode_dirty(inode); |
| 1358 | } |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1359 | } |
| 1360 | |
| 1361 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1362 | * We need to pick up the new inode size which generic_commit_write gave us |
| 1363 | * `file' can be NULL - eg, when called from page_symlink(). |
| 1364 | * |
| 1365 | * ext3 never places buffers on inode->i_mapping->private_list. metadata |
| 1366 | * buffers are managed internally. |
| 1367 | */ |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1368 | static int ext3_ordered_write_end(struct file *file, |
| 1369 | struct address_space *mapping, |
| 1370 | loff_t pos, unsigned len, unsigned copied, |
| 1371 | struct page *page, void *fsdata) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1372 | { |
| 1373 | handle_t *handle = ext3_journal_current_handle(); |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1374 | struct inode *inode = file->f_mapping->host; |
| 1375 | unsigned from, to; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1376 | int ret = 0, ret2; |
| 1377 | |
Lukas Czerner | 785c4bc | 2011-05-23 18:33:01 +0200 | [diff] [blame] | 1378 | trace_ext3_ordered_write_end(inode, pos, len, copied); |
Jan Kara | 695f6ae | 2009-04-02 16:57:17 -0700 | [diff] [blame] | 1379 | copied = block_write_end(file, mapping, pos, len, copied, page, fsdata); |
| 1380 | |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1381 | from = pos & (PAGE_CACHE_SIZE - 1); |
Jan Kara | 695f6ae | 2009-04-02 16:57:17 -0700 | [diff] [blame] | 1382 | to = from + copied; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1383 | ret = walk_page_buffers(handle, page_buffers(page), |
Jan Kara | 695f6ae | 2009-04-02 16:57:17 -0700 | [diff] [blame] | 1384 | from, to, NULL, journal_dirty_data_fn); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1385 | |
Jan Kara | 695f6ae | 2009-04-02 16:57:17 -0700 | [diff] [blame] | 1386 | if (ret == 0) |
| 1387 | update_file_sizes(inode, pos, copied); |
| 1388 | /* |
| 1389 | * There may be allocated blocks outside of i_size because |
| 1390 | * we failed to copy some data. Prepare for truncate. |
| 1391 | */ |
Jan Kara | 9eaaa2d | 2009-07-13 20:26:52 +0200 | [diff] [blame] | 1392 | if (pos + len > inode->i_size && ext3_can_truncate(inode)) |
Jan Kara | 695f6ae | 2009-04-02 16:57:17 -0700 | [diff] [blame] | 1393 | ext3_orphan_add(handle, inode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1394 | ret2 = ext3_journal_stop(handle); |
| 1395 | if (!ret) |
| 1396 | ret = ret2; |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1397 | unlock_page(page); |
| 1398 | page_cache_release(page); |
| 1399 | |
Jan Kara | 695f6ae | 2009-04-02 16:57:17 -0700 | [diff] [blame] | 1400 | if (pos + len > inode->i_size) |
Jan Kara | 68eb3db | 2009-12-01 16:53:06 +0100 | [diff] [blame] | 1401 | ext3_truncate_failed_write(inode); |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1402 | return ret ? ret : copied; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1403 | } |
| 1404 | |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1405 | static int ext3_writeback_write_end(struct file *file, |
| 1406 | struct address_space *mapping, |
| 1407 | loff_t pos, unsigned len, unsigned copied, |
| 1408 | struct page *page, void *fsdata) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1409 | { |
| 1410 | handle_t *handle = ext3_journal_current_handle(); |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1411 | struct inode *inode = file->f_mapping->host; |
Jan Kara | 695f6ae | 2009-04-02 16:57:17 -0700 | [diff] [blame] | 1412 | int ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1413 | |
Lukas Czerner | 785c4bc | 2011-05-23 18:33:01 +0200 | [diff] [blame] | 1414 | trace_ext3_writeback_write_end(inode, pos, len, copied); |
Jan Kara | 695f6ae | 2009-04-02 16:57:17 -0700 | [diff] [blame] | 1415 | copied = block_write_end(file, mapping, pos, len, copied, page, fsdata); |
| 1416 | update_file_sizes(inode, pos, copied); |
| 1417 | /* |
| 1418 | * There may be allocated blocks outside of i_size because |
| 1419 | * we failed to copy some data. Prepare for truncate. |
| 1420 | */ |
Jan Kara | 9eaaa2d | 2009-07-13 20:26:52 +0200 | [diff] [blame] | 1421 | if (pos + len > inode->i_size && ext3_can_truncate(inode)) |
Jan Kara | 695f6ae | 2009-04-02 16:57:17 -0700 | [diff] [blame] | 1422 | ext3_orphan_add(handle, inode); |
| 1423 | ret = ext3_journal_stop(handle); |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1424 | unlock_page(page); |
| 1425 | page_cache_release(page); |
| 1426 | |
Jan Kara | 695f6ae | 2009-04-02 16:57:17 -0700 | [diff] [blame] | 1427 | if (pos + len > inode->i_size) |
Jan Kara | 68eb3db | 2009-12-01 16:53:06 +0100 | [diff] [blame] | 1428 | ext3_truncate_failed_write(inode); |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1429 | return ret ? ret : copied; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1430 | } |
| 1431 | |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1432 | static int ext3_journalled_write_end(struct file *file, |
| 1433 | struct address_space *mapping, |
| 1434 | loff_t pos, unsigned len, unsigned copied, |
| 1435 | struct page *page, void *fsdata) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1436 | { |
| 1437 | handle_t *handle = ext3_journal_current_handle(); |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1438 | struct inode *inode = mapping->host; |
Jan Kara | b22570d | 2011-07-23 01:21:38 +0200 | [diff] [blame] | 1439 | struct ext3_inode_info *ei = EXT3_I(inode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1440 | int ret = 0, ret2; |
| 1441 | int partial = 0; |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1442 | unsigned from, to; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1443 | |
Lukas Czerner | 785c4bc | 2011-05-23 18:33:01 +0200 | [diff] [blame] | 1444 | trace_ext3_journalled_write_end(inode, pos, len, copied); |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1445 | from = pos & (PAGE_CACHE_SIZE - 1); |
| 1446 | to = from + len; |
| 1447 | |
| 1448 | if (copied < len) { |
| 1449 | if (!PageUptodate(page)) |
| 1450 | copied = 0; |
Jan Kara | 695f6ae | 2009-04-02 16:57:17 -0700 | [diff] [blame] | 1451 | page_zero_new_buffers(page, from + copied, to); |
| 1452 | to = from + copied; |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1453 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1454 | |
| 1455 | ret = walk_page_buffers(handle, page_buffers(page), from, |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1456 | to, &partial, write_end_fn); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1457 | if (!partial) |
| 1458 | SetPageUptodate(page); |
Jan Kara | 695f6ae | 2009-04-02 16:57:17 -0700 | [diff] [blame] | 1459 | |
| 1460 | if (pos + copied > inode->i_size) |
| 1461 | i_size_write(inode, pos + copied); |
| 1462 | /* |
| 1463 | * There may be allocated blocks outside of i_size because |
| 1464 | * we failed to copy some data. Prepare for truncate. |
| 1465 | */ |
Jan Kara | 9eaaa2d | 2009-07-13 20:26:52 +0200 | [diff] [blame] | 1466 | if (pos + len > inode->i_size && ext3_can_truncate(inode)) |
Jan Kara | 695f6ae | 2009-04-02 16:57:17 -0700 | [diff] [blame] | 1467 | ext3_orphan_add(handle, inode); |
Jan Kara | 9df9393 | 2010-01-06 21:58:48 +0100 | [diff] [blame] | 1468 | ext3_set_inode_state(inode, EXT3_STATE_JDATA); |
Jan Kara | b22570d | 2011-07-23 01:21:38 +0200 | [diff] [blame] | 1469 | atomic_set(&ei->i_datasync_tid, handle->h_transaction->t_tid); |
| 1470 | if (inode->i_size > ei->i_disksize) { |
| 1471 | ei->i_disksize = inode->i_size; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1472 | ret2 = ext3_mark_inode_dirty(handle, inode); |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 1473 | if (!ret) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1474 | ret = ret2; |
| 1475 | } |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1476 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1477 | ret2 = ext3_journal_stop(handle); |
| 1478 | if (!ret) |
| 1479 | ret = ret2; |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1480 | unlock_page(page); |
| 1481 | page_cache_release(page); |
| 1482 | |
Jan Kara | 695f6ae | 2009-04-02 16:57:17 -0700 | [diff] [blame] | 1483 | if (pos + len > inode->i_size) |
Jan Kara | 68eb3db | 2009-12-01 16:53:06 +0100 | [diff] [blame] | 1484 | ext3_truncate_failed_write(inode); |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1485 | return ret ? ret : copied; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1486 | } |
| 1487 | |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 1488 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1489 | * bmap() is special. It gets used by applications such as lilo and by |
| 1490 | * the swapper to find the on-disk block of a specific piece of data. |
| 1491 | * |
| 1492 | * Naturally, this is dangerous if the block concerned is still in the |
| 1493 | * journal. If somebody makes a swapfile on an ext3 data-journaling |
| 1494 | * filesystem and enables swap, then they may get a nasty shock when the |
| 1495 | * data getting swapped to that swapfile suddenly gets overwritten by |
| 1496 | * the original zero's written out previously to the journal and |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 1497 | * awaiting writeback in the kernel's buffer cache. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1498 | * |
| 1499 | * So, if we see any bmap calls here on a modified, data-journaled file, |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 1500 | * take extra steps to flush any blocks which might be in the cache. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1501 | */ |
| 1502 | static sector_t ext3_bmap(struct address_space *mapping, sector_t block) |
| 1503 | { |
| 1504 | struct inode *inode = mapping->host; |
| 1505 | journal_t *journal; |
| 1506 | int err; |
| 1507 | |
Jan Kara | 9df9393 | 2010-01-06 21:58:48 +0100 | [diff] [blame] | 1508 | if (ext3_test_inode_state(inode, EXT3_STATE_JDATA)) { |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 1509 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1510 | * This is a REALLY heavyweight approach, but the use of |
| 1511 | * bmap on dirty files is expected to be extremely rare: |
| 1512 | * only if we run lilo or swapon on a freshly made file |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 1513 | * do we expect this to happen. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1514 | * |
| 1515 | * (bmap requires CAP_SYS_RAWIO so this does not |
| 1516 | * represent an unprivileged user DOS attack --- we'd be |
| 1517 | * in trouble if mortal users could trigger this path at |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 1518 | * will.) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1519 | * |
| 1520 | * NB. EXT3_STATE_JDATA is not set on files other than |
| 1521 | * regular files. If somebody wants to bmap a directory |
| 1522 | * or symlink and gets confused because the buffer |
| 1523 | * hasn't yet been flushed to disk, they deserve |
| 1524 | * everything they get. |
| 1525 | */ |
| 1526 | |
Jan Kara | 9df9393 | 2010-01-06 21:58:48 +0100 | [diff] [blame] | 1527 | ext3_clear_inode_state(inode, EXT3_STATE_JDATA); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1528 | journal = EXT3_JOURNAL(inode); |
| 1529 | journal_lock_updates(journal); |
| 1530 | err = journal_flush(journal); |
| 1531 | journal_unlock_updates(journal); |
| 1532 | |
| 1533 | if (err) |
| 1534 | return 0; |
| 1535 | } |
| 1536 | |
| 1537 | return generic_block_bmap(mapping,block,ext3_get_block); |
| 1538 | } |
| 1539 | |
| 1540 | static int bget_one(handle_t *handle, struct buffer_head *bh) |
| 1541 | { |
| 1542 | get_bh(bh); |
| 1543 | return 0; |
| 1544 | } |
| 1545 | |
| 1546 | static int bput_one(handle_t *handle, struct buffer_head *bh) |
| 1547 | { |
| 1548 | put_bh(bh); |
| 1549 | return 0; |
| 1550 | } |
| 1551 | |
Jan Kara | 9e80d40 | 2009-03-26 13:08:04 +0100 | [diff] [blame] | 1552 | static int buffer_unmapped(handle_t *handle, struct buffer_head *bh) |
| 1553 | { |
| 1554 | return !buffer_mapped(bh); |
| 1555 | } |
Jan Kara | 695f6ae | 2009-04-02 16:57:17 -0700 | [diff] [blame] | 1556 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1557 | /* |
| 1558 | * Note that we always start a transaction even if we're not journalling |
| 1559 | * data. This is to preserve ordering: any hole instantiation within |
| 1560 | * __block_write_full_page -> ext3_get_block() should be journalled |
| 1561 | * along with the data so we don't crash and then get metadata which |
| 1562 | * refers to old data. |
| 1563 | * |
| 1564 | * In all journalling modes block_write_full_page() will start the I/O. |
| 1565 | * |
| 1566 | * Problem: |
| 1567 | * |
| 1568 | * ext3_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() -> |
| 1569 | * ext3_writepage() |
| 1570 | * |
| 1571 | * Similar for: |
| 1572 | * |
| 1573 | * ext3_file_write() -> generic_file_write() -> __alloc_pages() -> ... |
| 1574 | * |
| 1575 | * Same applies to ext3_get_block(). We will deadlock on various things like |
Arjan van de Ven | 9746151 | 2006-03-23 03:00:42 -0800 | [diff] [blame] | 1576 | * lock_journal and i_truncate_mutex. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1577 | * |
| 1578 | * Setting PF_MEMALLOC here doesn't work - too many internal memory |
| 1579 | * allocations fail. |
| 1580 | * |
| 1581 | * 16May01: If we're reentered then journal_current_handle() will be |
| 1582 | * non-zero. We simply *return*. |
| 1583 | * |
| 1584 | * 1 July 2001: @@@ FIXME: |
| 1585 | * In journalled data mode, a data buffer may be metadata against the |
| 1586 | * current transaction. But the same file is part of a shared mapping |
| 1587 | * and someone does a writepage() on it. |
| 1588 | * |
| 1589 | * We will move the buffer onto the async_data list, but *after* it has |
| 1590 | * been dirtied. So there's a small window where we have dirty data on |
| 1591 | * BJ_Metadata. |
| 1592 | * |
| 1593 | * Note that this only applies to the last partial page in the file. The |
| 1594 | * bit which block_write_full_page() uses prepare/commit for. (That's |
| 1595 | * broken code anyway: it's wrong for msync()). |
| 1596 | * |
| 1597 | * It's a rare case: affects the final partial page, for journalled data |
| 1598 | * where the file is subject to bith write() and writepage() in the same |
| 1599 | * transction. To fix it we'll need a custom block_write_full_page(). |
| 1600 | * We'll probably need that anyway for journalling writepage() output. |
| 1601 | * |
| 1602 | * We don't honour synchronous mounts for writepage(). That would be |
| 1603 | * disastrous. Any write() or metadata operation will sync the fs for |
| 1604 | * us. |
| 1605 | * |
| 1606 | * AKPM2: if all the page's buffers are mapped to disk and !data=journal, |
| 1607 | * we don't need to open a transaction here. |
| 1608 | */ |
| 1609 | static int ext3_ordered_writepage(struct page *page, |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 1610 | struct writeback_control *wbc) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1611 | { |
| 1612 | struct inode *inode = page->mapping->host; |
| 1613 | struct buffer_head *page_bufs; |
| 1614 | handle_t *handle = NULL; |
| 1615 | int ret = 0; |
| 1616 | int err; |
| 1617 | |
| 1618 | J_ASSERT(PageLocked(page)); |
Jan Kara | 33c104d | 2011-12-22 16:49:05 +0100 | [diff] [blame] | 1619 | /* |
| 1620 | * We don't want to warn for emergency remount. The condition is |
| 1621 | * ordered to avoid dereferencing inode->i_sb in non-error case to |
| 1622 | * avoid slow-downs. |
| 1623 | */ |
| 1624 | WARN_ON_ONCE(IS_RDONLY(inode) && |
| 1625 | !(EXT3_SB(inode->i_sb)->s_mount_state & EXT3_ERROR_FS)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1626 | |
| 1627 | /* |
| 1628 | * We give up here if we're reentered, because it might be for a |
| 1629 | * different filesystem. |
| 1630 | */ |
| 1631 | if (ext3_journal_current_handle()) |
| 1632 | goto out_fail; |
| 1633 | |
Lukas Czerner | 785c4bc | 2011-05-23 18:33:01 +0200 | [diff] [blame] | 1634 | trace_ext3_ordered_writepage(page); |
Jan Kara | 9e80d40 | 2009-03-26 13:08:04 +0100 | [diff] [blame] | 1635 | if (!page_has_buffers(page)) { |
| 1636 | create_empty_buffers(page, inode->i_sb->s_blocksize, |
| 1637 | (1 << BH_Dirty)|(1 << BH_Uptodate)); |
Jan Kara | 430db32 | 2009-04-07 18:25:01 -0400 | [diff] [blame] | 1638 | page_bufs = page_buffers(page); |
| 1639 | } else { |
| 1640 | page_bufs = page_buffers(page); |
| 1641 | if (!walk_page_buffers(NULL, page_bufs, 0, PAGE_CACHE_SIZE, |
| 1642 | NULL, buffer_unmapped)) { |
| 1643 | /* Provide NULL get_block() to catch bugs if buffers |
| 1644 | * weren't really mapped */ |
| 1645 | return block_write_full_page(page, NULL, wbc); |
| 1646 | } |
Jan Kara | 9e80d40 | 2009-03-26 13:08:04 +0100 | [diff] [blame] | 1647 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1648 | handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode)); |
| 1649 | |
| 1650 | if (IS_ERR(handle)) { |
| 1651 | ret = PTR_ERR(handle); |
| 1652 | goto out_fail; |
| 1653 | } |
| 1654 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1655 | walk_page_buffers(handle, page_bufs, 0, |
| 1656 | PAGE_CACHE_SIZE, NULL, bget_one); |
| 1657 | |
| 1658 | ret = block_write_full_page(page, ext3_get_block, wbc); |
| 1659 | |
| 1660 | /* |
| 1661 | * The page can become unlocked at any point now, and |
| 1662 | * truncate can then come in and change things. So we |
| 1663 | * can't touch *page from now on. But *page_bufs is |
| 1664 | * safe due to elevated refcount. |
| 1665 | */ |
| 1666 | |
| 1667 | /* |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 1668 | * And attach them to the current transaction. But only if |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1669 | * block_write_full_page() succeeded. Otherwise they are unmapped, |
| 1670 | * and generally junk. |
| 1671 | */ |
| 1672 | if (ret == 0) { |
| 1673 | err = walk_page_buffers(handle, page_bufs, 0, PAGE_CACHE_SIZE, |
| 1674 | NULL, journal_dirty_data_fn); |
| 1675 | if (!ret) |
| 1676 | ret = err; |
| 1677 | } |
| 1678 | walk_page_buffers(handle, page_bufs, 0, |
| 1679 | PAGE_CACHE_SIZE, NULL, bput_one); |
| 1680 | err = ext3_journal_stop(handle); |
| 1681 | if (!ret) |
| 1682 | ret = err; |
| 1683 | return ret; |
| 1684 | |
| 1685 | out_fail: |
| 1686 | redirty_page_for_writepage(wbc, page); |
| 1687 | unlock_page(page); |
| 1688 | return ret; |
| 1689 | } |
| 1690 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1691 | static int ext3_writeback_writepage(struct page *page, |
| 1692 | struct writeback_control *wbc) |
| 1693 | { |
| 1694 | struct inode *inode = page->mapping->host; |
| 1695 | handle_t *handle = NULL; |
| 1696 | int ret = 0; |
| 1697 | int err; |
| 1698 | |
Dmitry Monakhov | 49792c8 | 2010-03-02 15:51:02 +0300 | [diff] [blame] | 1699 | J_ASSERT(PageLocked(page)); |
Jan Kara | 33c104d | 2011-12-22 16:49:05 +0100 | [diff] [blame] | 1700 | /* |
| 1701 | * We don't want to warn for emergency remount. The condition is |
| 1702 | * ordered to avoid dereferencing inode->i_sb in non-error case to |
| 1703 | * avoid slow-downs. |
| 1704 | */ |
| 1705 | WARN_ON_ONCE(IS_RDONLY(inode) && |
| 1706 | !(EXT3_SB(inode->i_sb)->s_mount_state & EXT3_ERROR_FS)); |
Dmitry Monakhov | 49792c8 | 2010-03-02 15:51:02 +0300 | [diff] [blame] | 1707 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1708 | if (ext3_journal_current_handle()) |
| 1709 | goto out_fail; |
| 1710 | |
Lukas Czerner | 785c4bc | 2011-05-23 18:33:01 +0200 | [diff] [blame] | 1711 | trace_ext3_writeback_writepage(page); |
Jan Kara | 430db32 | 2009-04-07 18:25:01 -0400 | [diff] [blame] | 1712 | if (page_has_buffers(page)) { |
| 1713 | if (!walk_page_buffers(NULL, page_buffers(page), 0, |
| 1714 | PAGE_CACHE_SIZE, NULL, buffer_unmapped)) { |
| 1715 | /* Provide NULL get_block() to catch bugs if buffers |
| 1716 | * weren't really mapped */ |
| 1717 | return block_write_full_page(page, NULL, wbc); |
| 1718 | } |
| 1719 | } |
| 1720 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1721 | handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode)); |
| 1722 | if (IS_ERR(handle)) { |
| 1723 | ret = PTR_ERR(handle); |
| 1724 | goto out_fail; |
| 1725 | } |
| 1726 | |
Christoph Hellwig | 4c4d390 | 2010-06-07 10:20:39 +0200 | [diff] [blame] | 1727 | ret = block_write_full_page(page, ext3_get_block, wbc); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1728 | |
| 1729 | err = ext3_journal_stop(handle); |
| 1730 | if (!ret) |
| 1731 | ret = err; |
| 1732 | return ret; |
| 1733 | |
| 1734 | out_fail: |
| 1735 | redirty_page_for_writepage(wbc, page); |
| 1736 | unlock_page(page); |
| 1737 | return ret; |
| 1738 | } |
| 1739 | |
| 1740 | static int ext3_journalled_writepage(struct page *page, |
| 1741 | struct writeback_control *wbc) |
| 1742 | { |
| 1743 | struct inode *inode = page->mapping->host; |
| 1744 | handle_t *handle = NULL; |
| 1745 | int ret = 0; |
| 1746 | int err; |
| 1747 | |
Dmitry Monakhov | 49792c8 | 2010-03-02 15:51:02 +0300 | [diff] [blame] | 1748 | J_ASSERT(PageLocked(page)); |
Jan Kara | 33c104d | 2011-12-22 16:49:05 +0100 | [diff] [blame] | 1749 | /* |
| 1750 | * We don't want to warn for emergency remount. The condition is |
| 1751 | * ordered to avoid dereferencing inode->i_sb in non-error case to |
| 1752 | * avoid slow-downs. |
| 1753 | */ |
| 1754 | WARN_ON_ONCE(IS_RDONLY(inode) && |
| 1755 | !(EXT3_SB(inode->i_sb)->s_mount_state & EXT3_ERROR_FS)); |
Dmitry Monakhov | 49792c8 | 2010-03-02 15:51:02 +0300 | [diff] [blame] | 1756 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1757 | if (ext3_journal_current_handle()) |
| 1758 | goto no_write; |
| 1759 | |
Lukas Czerner | 785c4bc | 2011-05-23 18:33:01 +0200 | [diff] [blame] | 1760 | trace_ext3_journalled_writepage(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1761 | handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode)); |
| 1762 | if (IS_ERR(handle)) { |
| 1763 | ret = PTR_ERR(handle); |
| 1764 | goto no_write; |
| 1765 | } |
| 1766 | |
| 1767 | if (!page_has_buffers(page) || PageChecked(page)) { |
| 1768 | /* |
| 1769 | * It's mmapped pagecache. Add buffers and journal it. There |
| 1770 | * doesn't seem much point in redirtying the page here. |
| 1771 | */ |
| 1772 | ClearPageChecked(page); |
Christoph Hellwig | ebdec24 | 2010-10-06 10:47:23 +0200 | [diff] [blame] | 1773 | ret = __block_write_begin(page, 0, PAGE_CACHE_SIZE, |
| 1774 | ext3_get_block); |
Denis Lunev | ab4eb43 | 2005-11-13 16:07:17 -0800 | [diff] [blame] | 1775 | if (ret != 0) { |
| 1776 | ext3_journal_stop(handle); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1777 | goto out_unlock; |
Denis Lunev | ab4eb43 | 2005-11-13 16:07:17 -0800 | [diff] [blame] | 1778 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1779 | ret = walk_page_buffers(handle, page_buffers(page), 0, |
| 1780 | PAGE_CACHE_SIZE, NULL, do_journal_get_write_access); |
| 1781 | |
| 1782 | err = walk_page_buffers(handle, page_buffers(page), 0, |
Nick Piggin | f4fc66a | 2007-10-16 01:25:05 -0700 | [diff] [blame] | 1783 | PAGE_CACHE_SIZE, NULL, write_end_fn); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1784 | if (ret == 0) |
| 1785 | ret = err; |
Jan Kara | 9df9393 | 2010-01-06 21:58:48 +0100 | [diff] [blame] | 1786 | ext3_set_inode_state(inode, EXT3_STATE_JDATA); |
Jan Kara | b22570d | 2011-07-23 01:21:38 +0200 | [diff] [blame] | 1787 | atomic_set(&EXT3_I(inode)->i_datasync_tid, |
| 1788 | handle->h_transaction->t_tid); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1789 | unlock_page(page); |
| 1790 | } else { |
| 1791 | /* |
| 1792 | * It may be a page full of checkpoint-mode buffers. We don't |
| 1793 | * really know unless we go poke around in the buffer_heads. |
| 1794 | * But block_write_full_page will do the right thing. |
| 1795 | */ |
| 1796 | ret = block_write_full_page(page, ext3_get_block, wbc); |
| 1797 | } |
| 1798 | err = ext3_journal_stop(handle); |
| 1799 | if (!ret) |
| 1800 | ret = err; |
| 1801 | out: |
| 1802 | return ret; |
| 1803 | |
| 1804 | no_write: |
| 1805 | redirty_page_for_writepage(wbc, page); |
| 1806 | out_unlock: |
| 1807 | unlock_page(page); |
| 1808 | goto out; |
| 1809 | } |
| 1810 | |
| 1811 | static int ext3_readpage(struct file *file, struct page *page) |
| 1812 | { |
Lukas Czerner | 785c4bc | 2011-05-23 18:33:01 +0200 | [diff] [blame] | 1813 | trace_ext3_readpage(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1814 | return mpage_readpage(page, ext3_get_block); |
| 1815 | } |
| 1816 | |
| 1817 | static int |
| 1818 | ext3_readpages(struct file *file, struct address_space *mapping, |
| 1819 | struct list_head *pages, unsigned nr_pages) |
| 1820 | { |
| 1821 | return mpage_readpages(mapping, pages, nr_pages, ext3_get_block); |
| 1822 | } |
| 1823 | |
NeilBrown | 2ff28e2 | 2006-03-26 01:37:18 -0800 | [diff] [blame] | 1824 | static void ext3_invalidatepage(struct page *page, unsigned long offset) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1825 | { |
| 1826 | journal_t *journal = EXT3_JOURNAL(page->mapping->host); |
| 1827 | |
Lukas Czerner | 785c4bc | 2011-05-23 18:33:01 +0200 | [diff] [blame] | 1828 | trace_ext3_invalidatepage(page, offset); |
| 1829 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1830 | /* |
| 1831 | * If it's a full truncate we just forget about the pending dirtying |
| 1832 | */ |
| 1833 | if (offset == 0) |
| 1834 | ClearPageChecked(page); |
| 1835 | |
NeilBrown | 2ff28e2 | 2006-03-26 01:37:18 -0800 | [diff] [blame] | 1836 | journal_invalidatepage(journal, page, offset); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1837 | } |
| 1838 | |
Al Viro | 27496a8 | 2005-10-21 03:20:48 -0400 | [diff] [blame] | 1839 | static int ext3_releasepage(struct page *page, gfp_t wait) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1840 | { |
| 1841 | journal_t *journal = EXT3_JOURNAL(page->mapping->host); |
| 1842 | |
Lukas Czerner | 785c4bc | 2011-05-23 18:33:01 +0200 | [diff] [blame] | 1843 | trace_ext3_releasepage(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1844 | WARN_ON(PageChecked(page)); |
| 1845 | if (!page_has_buffers(page)) |
| 1846 | return 0; |
| 1847 | return journal_try_to_free_buffers(journal, page, wait); |
| 1848 | } |
| 1849 | |
| 1850 | /* |
| 1851 | * If the O_DIRECT write will extend the file then add this inode to the |
| 1852 | * orphan list. So recovery will truncate it back to the original size |
| 1853 | * if the machine crashes during the write. |
| 1854 | * |
| 1855 | * If the O_DIRECT write is intantiating holes inside i_size and the machine |
Jan Kara | bd1939d | 2008-02-06 01:40:21 -0800 | [diff] [blame] | 1856 | * crashes then stale disk data _may_ be exposed inside the file. But current |
| 1857 | * VFS code falls back into buffered path in that case so we are safe. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1858 | */ |
| 1859 | static ssize_t ext3_direct_IO(int rw, struct kiocb *iocb, |
| 1860 | const struct iovec *iov, loff_t offset, |
| 1861 | unsigned long nr_segs) |
| 1862 | { |
| 1863 | struct file *file = iocb->ki_filp; |
| 1864 | struct inode *inode = file->f_mapping->host; |
| 1865 | struct ext3_inode_info *ei = EXT3_I(inode); |
Jan Kara | bd1939d | 2008-02-06 01:40:21 -0800 | [diff] [blame] | 1866 | handle_t *handle; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1867 | ssize_t ret; |
| 1868 | int orphan = 0; |
| 1869 | size_t count = iov_length(iov, nr_segs); |
Eric Sandeen | ea0174a | 2009-10-12 21:34:27 -0500 | [diff] [blame] | 1870 | int retries = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1871 | |
Lukas Czerner | 785c4bc | 2011-05-23 18:33:01 +0200 | [diff] [blame] | 1872 | trace_ext3_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw); |
| 1873 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1874 | if (rw == WRITE) { |
| 1875 | loff_t final_size = offset + count; |
| 1876 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1877 | if (final_size > inode->i_size) { |
Jan Kara | bd1939d | 2008-02-06 01:40:21 -0800 | [diff] [blame] | 1878 | /* Credits for sb + inode write */ |
| 1879 | handle = ext3_journal_start(inode, 2); |
| 1880 | if (IS_ERR(handle)) { |
| 1881 | ret = PTR_ERR(handle); |
| 1882 | goto out; |
| 1883 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1884 | ret = ext3_orphan_add(handle, inode); |
Jan Kara | bd1939d | 2008-02-06 01:40:21 -0800 | [diff] [blame] | 1885 | if (ret) { |
| 1886 | ext3_journal_stop(handle); |
| 1887 | goto out; |
| 1888 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1889 | orphan = 1; |
| 1890 | ei->i_disksize = inode->i_size; |
Jan Kara | bd1939d | 2008-02-06 01:40:21 -0800 | [diff] [blame] | 1891 | ext3_journal_stop(handle); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1892 | } |
| 1893 | } |
| 1894 | |
Eric Sandeen | ea0174a | 2009-10-12 21:34:27 -0500 | [diff] [blame] | 1895 | retry: |
Christoph Hellwig | aacfc19 | 2011-06-24 14:29:47 -0400 | [diff] [blame] | 1896 | ret = blockdev_direct_IO(rw, iocb, inode, iov, offset, nr_segs, |
| 1897 | ext3_get_block); |
Christoph Hellwig | eafdc7d | 2010-06-04 11:29:53 +0200 | [diff] [blame] | 1898 | /* |
| 1899 | * In case of error extending write may have instantiated a few |
| 1900 | * blocks outside i_size. Trim these off again. |
| 1901 | */ |
| 1902 | if (unlikely((rw & WRITE) && ret < 0)) { |
| 1903 | loff_t isize = i_size_read(inode); |
| 1904 | loff_t end = offset + iov_length(iov, nr_segs); |
| 1905 | |
| 1906 | if (end > isize) |
Jan Kara | ee3e77f | 2011-06-03 21:58:11 +0200 | [diff] [blame] | 1907 | ext3_truncate_failed_direct_write(inode); |
Christoph Hellwig | eafdc7d | 2010-06-04 11:29:53 +0200 | [diff] [blame] | 1908 | } |
Eric Sandeen | ea0174a | 2009-10-12 21:34:27 -0500 | [diff] [blame] | 1909 | if (ret == -ENOSPC && ext3_should_retry_alloc(inode->i_sb, &retries)) |
| 1910 | goto retry; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1911 | |
Jan Kara | bd1939d | 2008-02-06 01:40:21 -0800 | [diff] [blame] | 1912 | if (orphan) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1913 | int err; |
| 1914 | |
Jan Kara | bd1939d | 2008-02-06 01:40:21 -0800 | [diff] [blame] | 1915 | /* Credits for sb + inode write */ |
| 1916 | handle = ext3_journal_start(inode, 2); |
| 1917 | if (IS_ERR(handle)) { |
| 1918 | /* This is really bad luck. We've written the data |
Jan Kara | 7eb4969 | 2010-03-01 14:02:37 +0100 | [diff] [blame] | 1919 | * but cannot extend i_size. Truncate allocated blocks |
| 1920 | * and pretend the write failed... */ |
Jan Kara | ee3e77f | 2011-06-03 21:58:11 +0200 | [diff] [blame] | 1921 | ext3_truncate_failed_direct_write(inode); |
Jan Kara | bd1939d | 2008-02-06 01:40:21 -0800 | [diff] [blame] | 1922 | ret = PTR_ERR(handle); |
| 1923 | goto out; |
| 1924 | } |
| 1925 | if (inode->i_nlink) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1926 | ext3_orphan_del(handle, inode); |
Jan Kara | bd1939d | 2008-02-06 01:40:21 -0800 | [diff] [blame] | 1927 | if (ret > 0) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1928 | loff_t end = offset + ret; |
| 1929 | if (end > inode->i_size) { |
| 1930 | ei->i_disksize = end; |
| 1931 | i_size_write(inode, end); |
| 1932 | /* |
| 1933 | * We're going to return a positive `ret' |
| 1934 | * here due to non-zero-length I/O, so there's |
| 1935 | * no way of reporting error returns from |
| 1936 | * ext3_mark_inode_dirty() to userspace. So |
| 1937 | * ignore it. |
| 1938 | */ |
| 1939 | ext3_mark_inode_dirty(handle, inode); |
| 1940 | } |
| 1941 | } |
| 1942 | err = ext3_journal_stop(handle); |
| 1943 | if (ret == 0) |
| 1944 | ret = err; |
| 1945 | } |
| 1946 | out: |
Lukas Czerner | 785c4bc | 2011-05-23 18:33:01 +0200 | [diff] [blame] | 1947 | trace_ext3_direct_IO_exit(inode, offset, |
| 1948 | iov_length(iov, nr_segs), rw, ret); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1949 | return ret; |
| 1950 | } |
| 1951 | |
| 1952 | /* |
| 1953 | * Pages can be marked dirty completely asynchronously from ext3's journalling |
| 1954 | * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do |
| 1955 | * much here because ->set_page_dirty is called under VFS locks. The page is |
| 1956 | * not necessarily locked. |
| 1957 | * |
| 1958 | * We cannot just dirty the page and leave attached buffers clean, because the |
| 1959 | * buffers' dirty state is "definitive". We cannot just set the buffers dirty |
| 1960 | * or jbddirty because all the journalling code will explode. |
| 1961 | * |
| 1962 | * So what we do is to mark the page "pending dirty" and next time writepage |
| 1963 | * is called, propagate that into the buffers appropriately. |
| 1964 | */ |
| 1965 | static int ext3_journalled_set_page_dirty(struct page *page) |
| 1966 | { |
| 1967 | SetPageChecked(page); |
| 1968 | return __set_page_dirty_nobuffers(page); |
| 1969 | } |
| 1970 | |
Christoph Hellwig | f5e54d6 | 2006-06-28 04:26:44 -0700 | [diff] [blame] | 1971 | static const struct address_space_operations ext3_ordered_aops = { |
Hisashi Hifumi | 8ab22b9 | 2008-07-28 15:46:36 -0700 | [diff] [blame] | 1972 | .readpage = ext3_readpage, |
| 1973 | .readpages = ext3_readpages, |
| 1974 | .writepage = ext3_ordered_writepage, |
Hisashi Hifumi | 8ab22b9 | 2008-07-28 15:46:36 -0700 | [diff] [blame] | 1975 | .write_begin = ext3_write_begin, |
| 1976 | .write_end = ext3_ordered_write_end, |
| 1977 | .bmap = ext3_bmap, |
| 1978 | .invalidatepage = ext3_invalidatepage, |
| 1979 | .releasepage = ext3_releasepage, |
| 1980 | .direct_IO = ext3_direct_IO, |
| 1981 | .migratepage = buffer_migrate_page, |
| 1982 | .is_partially_uptodate = block_is_partially_uptodate, |
Andi Kleen | aa261f5 | 2009-09-16 11:50:16 +0200 | [diff] [blame] | 1983 | .error_remove_page = generic_error_remove_page, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1984 | }; |
| 1985 | |
Christoph Hellwig | f5e54d6 | 2006-06-28 04:26:44 -0700 | [diff] [blame] | 1986 | static const struct address_space_operations ext3_writeback_aops = { |
Hisashi Hifumi | 8ab22b9 | 2008-07-28 15:46:36 -0700 | [diff] [blame] | 1987 | .readpage = ext3_readpage, |
| 1988 | .readpages = ext3_readpages, |
| 1989 | .writepage = ext3_writeback_writepage, |
Hisashi Hifumi | 8ab22b9 | 2008-07-28 15:46:36 -0700 | [diff] [blame] | 1990 | .write_begin = ext3_write_begin, |
| 1991 | .write_end = ext3_writeback_write_end, |
| 1992 | .bmap = ext3_bmap, |
| 1993 | .invalidatepage = ext3_invalidatepage, |
| 1994 | .releasepage = ext3_releasepage, |
| 1995 | .direct_IO = ext3_direct_IO, |
| 1996 | .migratepage = buffer_migrate_page, |
| 1997 | .is_partially_uptodate = block_is_partially_uptodate, |
Andi Kleen | aa261f5 | 2009-09-16 11:50:16 +0200 | [diff] [blame] | 1998 | .error_remove_page = generic_error_remove_page, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1999 | }; |
| 2000 | |
Christoph Hellwig | f5e54d6 | 2006-06-28 04:26:44 -0700 | [diff] [blame] | 2001 | static const struct address_space_operations ext3_journalled_aops = { |
Hisashi Hifumi | 8ab22b9 | 2008-07-28 15:46:36 -0700 | [diff] [blame] | 2002 | .readpage = ext3_readpage, |
| 2003 | .readpages = ext3_readpages, |
| 2004 | .writepage = ext3_journalled_writepage, |
Hisashi Hifumi | 8ab22b9 | 2008-07-28 15:46:36 -0700 | [diff] [blame] | 2005 | .write_begin = ext3_write_begin, |
| 2006 | .write_end = ext3_journalled_write_end, |
| 2007 | .set_page_dirty = ext3_journalled_set_page_dirty, |
| 2008 | .bmap = ext3_bmap, |
| 2009 | .invalidatepage = ext3_invalidatepage, |
| 2010 | .releasepage = ext3_releasepage, |
| 2011 | .is_partially_uptodate = block_is_partially_uptodate, |
Andi Kleen | aa261f5 | 2009-09-16 11:50:16 +0200 | [diff] [blame] | 2012 | .error_remove_page = generic_error_remove_page, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2013 | }; |
| 2014 | |
| 2015 | void ext3_set_aops(struct inode *inode) |
| 2016 | { |
| 2017 | if (ext3_should_order_data(inode)) |
| 2018 | inode->i_mapping->a_ops = &ext3_ordered_aops; |
| 2019 | else if (ext3_should_writeback_data(inode)) |
| 2020 | inode->i_mapping->a_ops = &ext3_writeback_aops; |
| 2021 | else |
| 2022 | inode->i_mapping->a_ops = &ext3_journalled_aops; |
| 2023 | } |
| 2024 | |
| 2025 | /* |
| 2026 | * ext3_block_truncate_page() zeroes out a mapping from file offset `from' |
| 2027 | * up to the end of the block which corresponds to `from'. |
| 2028 | * This required during truncate. We need to physically zero the tail end |
| 2029 | * of that block so it doesn't yield old data if the file is later grown. |
| 2030 | */ |
Jan Kara | ee3e77f | 2011-06-03 21:58:11 +0200 | [diff] [blame] | 2031 | static int ext3_block_truncate_page(struct inode *inode, loff_t from) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2032 | { |
Mingming Cao | 43d23f9 | 2006-06-25 05:48:07 -0700 | [diff] [blame] | 2033 | ext3_fsblk_t index = from >> PAGE_CACHE_SHIFT; |
Jan Kara | ee3e77f | 2011-06-03 21:58:11 +0200 | [diff] [blame] | 2034 | unsigned offset = from & (PAGE_CACHE_SIZE - 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2035 | unsigned blocksize, iblock, length, pos; |
Jan Kara | ee3e77f | 2011-06-03 21:58:11 +0200 | [diff] [blame] | 2036 | struct page *page; |
| 2037 | handle_t *handle = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2038 | struct buffer_head *bh; |
| 2039 | int err = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2040 | |
Jan Kara | ee3e77f | 2011-06-03 21:58:11 +0200 | [diff] [blame] | 2041 | /* Truncated on block boundary - nothing to do */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2042 | blocksize = inode->i_sb->s_blocksize; |
Jan Kara | ee3e77f | 2011-06-03 21:58:11 +0200 | [diff] [blame] | 2043 | if ((from & (blocksize - 1)) == 0) |
| 2044 | return 0; |
| 2045 | |
| 2046 | page = grab_cache_page(inode->i_mapping, index); |
| 2047 | if (!page) |
| 2048 | return -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2049 | length = blocksize - (offset & (blocksize - 1)); |
| 2050 | iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits); |
| 2051 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2052 | if (!page_has_buffers(page)) |
| 2053 | create_empty_buffers(page, blocksize, 0); |
| 2054 | |
| 2055 | /* Find the buffer that contains "offset" */ |
| 2056 | bh = page_buffers(page); |
| 2057 | pos = blocksize; |
| 2058 | while (offset >= pos) { |
| 2059 | bh = bh->b_this_page; |
| 2060 | iblock++; |
| 2061 | pos += blocksize; |
| 2062 | } |
| 2063 | |
| 2064 | err = 0; |
| 2065 | if (buffer_freed(bh)) { |
| 2066 | BUFFER_TRACE(bh, "freed: skip"); |
| 2067 | goto unlock; |
| 2068 | } |
| 2069 | |
| 2070 | if (!buffer_mapped(bh)) { |
| 2071 | BUFFER_TRACE(bh, "unmapped"); |
| 2072 | ext3_get_block(inode, iblock, bh, 0); |
| 2073 | /* unmapped? It's a hole - nothing to do */ |
| 2074 | if (!buffer_mapped(bh)) { |
| 2075 | BUFFER_TRACE(bh, "still unmapped"); |
| 2076 | goto unlock; |
| 2077 | } |
| 2078 | } |
| 2079 | |
| 2080 | /* Ok, it's mapped. Make sure it's up-to-date */ |
| 2081 | if (PageUptodate(page)) |
| 2082 | set_buffer_uptodate(bh); |
| 2083 | |
Zheng Liu | d03e129 | 2011-12-05 15:55:11 +0800 | [diff] [blame] | 2084 | if (!bh_uptodate_or_lock(bh)) { |
| 2085 | err = bh_submit_read(bh); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2086 | /* Uhhuh. Read error. Complain and punt. */ |
Zheng Liu | d03e129 | 2011-12-05 15:55:11 +0800 | [diff] [blame] | 2087 | if (err) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2088 | goto unlock; |
| 2089 | } |
| 2090 | |
Jan Kara | ee3e77f | 2011-06-03 21:58:11 +0200 | [diff] [blame] | 2091 | /* data=writeback mode doesn't need transaction to zero-out data */ |
| 2092 | if (!ext3_should_writeback_data(inode)) { |
| 2093 | /* We journal at most one block */ |
| 2094 | handle = ext3_journal_start(inode, 1); |
| 2095 | if (IS_ERR(handle)) { |
| 2096 | clear_highpage(page); |
| 2097 | flush_dcache_page(page); |
| 2098 | err = PTR_ERR(handle); |
| 2099 | goto unlock; |
| 2100 | } |
| 2101 | } |
| 2102 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2103 | if (ext3_should_journal_data(inode)) { |
| 2104 | BUFFER_TRACE(bh, "get write access"); |
| 2105 | err = ext3_journal_get_write_access(handle, bh); |
| 2106 | if (err) |
Jan Kara | ee3e77f | 2011-06-03 21:58:11 +0200 | [diff] [blame] | 2107 | goto stop; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2108 | } |
| 2109 | |
Christoph Lameter | eebd2aa | 2008-02-04 22:28:29 -0800 | [diff] [blame] | 2110 | zero_user(page, offset, length); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2111 | BUFFER_TRACE(bh, "zeroed end of block"); |
| 2112 | |
| 2113 | err = 0; |
| 2114 | if (ext3_should_journal_data(inode)) { |
| 2115 | err = ext3_journal_dirty_metadata(handle, bh); |
| 2116 | } else { |
| 2117 | if (ext3_should_order_data(inode)) |
| 2118 | err = ext3_journal_dirty_data(handle, bh); |
| 2119 | mark_buffer_dirty(bh); |
| 2120 | } |
Jan Kara | ee3e77f | 2011-06-03 21:58:11 +0200 | [diff] [blame] | 2121 | stop: |
| 2122 | if (handle) |
| 2123 | ext3_journal_stop(handle); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2124 | |
| 2125 | unlock: |
| 2126 | unlock_page(page); |
| 2127 | page_cache_release(page); |
| 2128 | return err; |
| 2129 | } |
| 2130 | |
| 2131 | /* |
| 2132 | * Probably it should be a library function... search for first non-zero word |
| 2133 | * or memcmp with zero_page, whatever is better for particular architecture. |
| 2134 | * Linus? |
| 2135 | */ |
| 2136 | static inline int all_zeroes(__le32 *p, __le32 *q) |
| 2137 | { |
| 2138 | while (p < q) |
| 2139 | if (*p++) |
| 2140 | return 0; |
| 2141 | return 1; |
| 2142 | } |
| 2143 | |
| 2144 | /** |
| 2145 | * ext3_find_shared - find the indirect blocks for partial truncation. |
| 2146 | * @inode: inode in question |
| 2147 | * @depth: depth of the affected branch |
| 2148 | * @offsets: offsets of pointers in that branch (see ext3_block_to_path) |
| 2149 | * @chain: place to store the pointers to partial indirect blocks |
| 2150 | * @top: place to the (detached) top of branch |
| 2151 | * |
| 2152 | * This is a helper function used by ext3_truncate(). |
| 2153 | * |
| 2154 | * When we do truncate() we may have to clean the ends of several |
| 2155 | * indirect blocks but leave the blocks themselves alive. Block is |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 2156 | * partially truncated if some data below the new i_size is referred |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2157 | * from it (and it is on the path to the first completely truncated |
| 2158 | * data block, indeed). We have to free the top of that path along |
| 2159 | * with everything to the right of the path. Since no allocation |
| 2160 | * past the truncation point is possible until ext3_truncate() |
| 2161 | * finishes, we may safely do the latter, but top of branch may |
| 2162 | * require special attention - pageout below the truncation point |
| 2163 | * might try to populate it. |
| 2164 | * |
| 2165 | * We atomically detach the top of branch from the tree, store the |
| 2166 | * block number of its root in *@top, pointers to buffer_heads of |
| 2167 | * partially truncated blocks - in @chain[].bh and pointers to |
| 2168 | * their last elements that should not be removed - in |
| 2169 | * @chain[].p. Return value is the pointer to last filled element |
| 2170 | * of @chain. |
| 2171 | * |
| 2172 | * The work left to caller to do the actual freeing of subtrees: |
| 2173 | * a) free the subtree starting from *@top |
| 2174 | * b) free the subtrees whose roots are stored in |
| 2175 | * (@chain[i].p+1 .. end of @chain[i].bh->b_data) |
| 2176 | * c) free the subtrees growing from the inode past the @chain[0]. |
| 2177 | * (no partially truncated stuff there). */ |
| 2178 | |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 2179 | static Indirect *ext3_find_shared(struct inode *inode, int depth, |
| 2180 | int offsets[4], Indirect chain[4], __le32 *top) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2181 | { |
| 2182 | Indirect *partial, *p; |
| 2183 | int k, err; |
| 2184 | |
| 2185 | *top = 0; |
Uwe Kleine-König | bf48aab | 2009-10-28 20:11:03 +0100 | [diff] [blame] | 2186 | /* Make k index the deepest non-null offset + 1 */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2187 | for (k = depth; k > 1 && !offsets[k-1]; k--) |
| 2188 | ; |
| 2189 | partial = ext3_get_branch(inode, k, offsets, chain, &err); |
| 2190 | /* Writer: pointers */ |
| 2191 | if (!partial) |
| 2192 | partial = chain + k-1; |
| 2193 | /* |
| 2194 | * If the branch acquired continuation since we've looked at it - |
| 2195 | * fine, it should all survive and (new) top doesn't belong to us. |
| 2196 | */ |
| 2197 | if (!partial->key && *partial->p) |
| 2198 | /* Writer: end */ |
| 2199 | goto no_top; |
| 2200 | for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--) |
| 2201 | ; |
| 2202 | /* |
| 2203 | * OK, we've found the last block that must survive. The rest of our |
| 2204 | * branch should be detached before unlocking. However, if that rest |
| 2205 | * of branch is all ours and does not grow immediately from the inode |
| 2206 | * it's easier to cheat and just decrement partial->p. |
| 2207 | */ |
| 2208 | if (p == chain + k - 1 && p > chain) { |
| 2209 | p->p--; |
| 2210 | } else { |
| 2211 | *top = *p->p; |
| 2212 | /* Nope, don't do this in ext3. Must leave the tree intact */ |
| 2213 | #if 0 |
| 2214 | *p->p = 0; |
| 2215 | #endif |
| 2216 | } |
| 2217 | /* Writer: end */ |
| 2218 | |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 2219 | while(partial > p) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2220 | brelse(partial->bh); |
| 2221 | partial--; |
| 2222 | } |
| 2223 | no_top: |
| 2224 | return partial; |
| 2225 | } |
| 2226 | |
| 2227 | /* |
| 2228 | * Zero a number of block pointers in either an inode or an indirect block. |
| 2229 | * If we restart the transaction we must again get write access to the |
| 2230 | * indirect block for further modification. |
| 2231 | * |
| 2232 | * We release `count' blocks on disk, but (last - first) may be greater |
| 2233 | * than `count' because there can be holes in there. |
| 2234 | */ |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 2235 | static void ext3_clear_blocks(handle_t *handle, struct inode *inode, |
Mingming Cao | 43d23f9 | 2006-06-25 05:48:07 -0700 | [diff] [blame] | 2236 | struct buffer_head *bh, ext3_fsblk_t block_to_free, |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 2237 | unsigned long count, __le32 *first, __le32 *last) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2238 | { |
| 2239 | __le32 *p; |
| 2240 | if (try_to_extend_transaction(handle, inode)) { |
| 2241 | if (bh) { |
| 2242 | BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata"); |
Namhyung Kim | 156e743 | 2010-11-25 01:53:13 +0900 | [diff] [blame] | 2243 | if (ext3_journal_dirty_metadata(handle, bh)) |
| 2244 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2245 | } |
| 2246 | ext3_mark_inode_dirty(handle, inode); |
Jan Kara | 00171d3 | 2009-08-11 19:06:10 +0200 | [diff] [blame] | 2247 | truncate_restart_transaction(handle, inode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2248 | if (bh) { |
| 2249 | BUFFER_TRACE(bh, "retaking write access"); |
Namhyung Kim | 156e743 | 2010-11-25 01:53:13 +0900 | [diff] [blame] | 2250 | if (ext3_journal_get_write_access(handle, bh)) |
| 2251 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2252 | } |
| 2253 | } |
| 2254 | |
| 2255 | /* |
| 2256 | * Any buffers which are on the journal will be in memory. We find |
| 2257 | * them on the hash table so journal_revoke() will run journal_forget() |
| 2258 | * on them. We've already detached each block from the file, so |
| 2259 | * bforget() in journal_forget() should be safe. |
| 2260 | * |
| 2261 | * AKPM: turn on bforget in journal_forget()!!! |
| 2262 | */ |
| 2263 | for (p = first; p < last; p++) { |
| 2264 | u32 nr = le32_to_cpu(*p); |
| 2265 | if (nr) { |
| 2266 | struct buffer_head *bh; |
| 2267 | |
| 2268 | *p = 0; |
| 2269 | bh = sb_find_get_block(inode->i_sb, nr); |
| 2270 | ext3_forget(handle, 0, inode, bh, nr); |
| 2271 | } |
| 2272 | } |
| 2273 | |
| 2274 | ext3_free_blocks(handle, inode, block_to_free, count); |
| 2275 | } |
| 2276 | |
| 2277 | /** |
| 2278 | * ext3_free_data - free a list of data blocks |
| 2279 | * @handle: handle for this transaction |
| 2280 | * @inode: inode we are dealing with |
| 2281 | * @this_bh: indirect buffer_head which contains *@first and *@last |
| 2282 | * @first: array of block numbers |
| 2283 | * @last: points immediately past the end of array |
| 2284 | * |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 2285 | * We are freeing all blocks referred from that array (numbers are stored as |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2286 | * little-endian 32-bit) and updating @inode->i_blocks appropriately. |
| 2287 | * |
| 2288 | * We accumulate contiguous runs of blocks to free. Conveniently, if these |
| 2289 | * blocks are contiguous then releasing them at one time will only affect one |
| 2290 | * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't |
| 2291 | * actually use a lot of journal space. |
| 2292 | * |
| 2293 | * @this_bh will be %NULL if @first and @last point into the inode's direct |
| 2294 | * block pointers. |
| 2295 | */ |
| 2296 | static void ext3_free_data(handle_t *handle, struct inode *inode, |
| 2297 | struct buffer_head *this_bh, |
| 2298 | __le32 *first, __le32 *last) |
| 2299 | { |
Mingming Cao | 43d23f9 | 2006-06-25 05:48:07 -0700 | [diff] [blame] | 2300 | ext3_fsblk_t block_to_free = 0; /* Starting block # of a run */ |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 2301 | unsigned long count = 0; /* Number of blocks in the run */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2302 | __le32 *block_to_free_p = NULL; /* Pointer into inode/ind |
| 2303 | corresponding to |
| 2304 | block_to_free */ |
Mingming Cao | 43d23f9 | 2006-06-25 05:48:07 -0700 | [diff] [blame] | 2305 | ext3_fsblk_t nr; /* Current block # */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2306 | __le32 *p; /* Pointer into inode/ind |
| 2307 | for current block */ |
| 2308 | int err; |
| 2309 | |
| 2310 | if (this_bh) { /* For indirect block */ |
| 2311 | BUFFER_TRACE(this_bh, "get_write_access"); |
| 2312 | err = ext3_journal_get_write_access(handle, this_bh); |
| 2313 | /* Important: if we can't update the indirect pointers |
| 2314 | * to the blocks, we can't free them. */ |
| 2315 | if (err) |
| 2316 | return; |
| 2317 | } |
| 2318 | |
| 2319 | for (p = first; p < last; p++) { |
| 2320 | nr = le32_to_cpu(*p); |
| 2321 | if (nr) { |
| 2322 | /* accumulate blocks to free if they're contiguous */ |
| 2323 | if (count == 0) { |
| 2324 | block_to_free = nr; |
| 2325 | block_to_free_p = p; |
| 2326 | count = 1; |
| 2327 | } else if (nr == block_to_free + count) { |
| 2328 | count++; |
| 2329 | } else { |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 2330 | ext3_clear_blocks(handle, inode, this_bh, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2331 | block_to_free, |
| 2332 | count, block_to_free_p, p); |
| 2333 | block_to_free = nr; |
| 2334 | block_to_free_p = p; |
| 2335 | count = 1; |
| 2336 | } |
| 2337 | } |
| 2338 | } |
| 2339 | |
| 2340 | if (count > 0) |
| 2341 | ext3_clear_blocks(handle, inode, this_bh, block_to_free, |
| 2342 | count, block_to_free_p, p); |
| 2343 | |
| 2344 | if (this_bh) { |
| 2345 | BUFFER_TRACE(this_bh, "call ext3_journal_dirty_metadata"); |
Duane Griffin | 3ccc316 | 2008-07-25 01:46:26 -0700 | [diff] [blame] | 2346 | |
| 2347 | /* |
| 2348 | * The buffer head should have an attached journal head at this |
| 2349 | * point. However, if the data is corrupted and an indirect |
| 2350 | * block pointed to itself, it would have been detached when |
| 2351 | * the block was cleared. Check for this instead of OOPSing. |
| 2352 | */ |
| 2353 | if (bh2jh(this_bh)) |
| 2354 | ext3_journal_dirty_metadata(handle, this_bh); |
| 2355 | else |
| 2356 | ext3_error(inode->i_sb, "ext3_free_data", |
| 2357 | "circular indirect block detected, " |
| 2358 | "inode=%lu, block=%llu", |
| 2359 | inode->i_ino, |
| 2360 | (unsigned long long)this_bh->b_blocknr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2361 | } |
| 2362 | } |
| 2363 | |
| 2364 | /** |
| 2365 | * ext3_free_branches - free an array of branches |
| 2366 | * @handle: JBD handle for this transaction |
| 2367 | * @inode: inode we are dealing with |
| 2368 | * @parent_bh: the buffer_head which contains *@first and *@last |
| 2369 | * @first: array of block numbers |
| 2370 | * @last: pointer immediately past the end of array |
| 2371 | * @depth: depth of the branches to free |
| 2372 | * |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 2373 | * We are freeing all blocks referred from these branches (numbers are |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2374 | * stored as little-endian 32-bit) and updating @inode->i_blocks |
| 2375 | * appropriately. |
| 2376 | */ |
| 2377 | static void ext3_free_branches(handle_t *handle, struct inode *inode, |
| 2378 | struct buffer_head *parent_bh, |
| 2379 | __le32 *first, __le32 *last, int depth) |
| 2380 | { |
Mingming Cao | 43d23f9 | 2006-06-25 05:48:07 -0700 | [diff] [blame] | 2381 | ext3_fsblk_t nr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2382 | __le32 *p; |
| 2383 | |
| 2384 | if (is_handle_aborted(handle)) |
| 2385 | return; |
| 2386 | |
| 2387 | if (depth--) { |
| 2388 | struct buffer_head *bh; |
| 2389 | int addr_per_block = EXT3_ADDR_PER_BLOCK(inode->i_sb); |
| 2390 | p = last; |
| 2391 | while (--p >= first) { |
| 2392 | nr = le32_to_cpu(*p); |
| 2393 | if (!nr) |
| 2394 | continue; /* A hole */ |
| 2395 | |
| 2396 | /* Go read the buffer for the next level down */ |
| 2397 | bh = sb_bread(inode->i_sb, nr); |
| 2398 | |
| 2399 | /* |
| 2400 | * A read failure? Report error and clear slot |
| 2401 | * (should be rare). |
| 2402 | */ |
| 2403 | if (!bh) { |
| 2404 | ext3_error(inode->i_sb, "ext3_free_branches", |
Eric Sandeen | eee194e | 2006-09-27 01:49:30 -0700 | [diff] [blame] | 2405 | "Read failure, inode=%lu, block="E3FSBLK, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2406 | inode->i_ino, nr); |
| 2407 | continue; |
| 2408 | } |
| 2409 | |
| 2410 | /* This zaps the entire block. Bottom up. */ |
| 2411 | BUFFER_TRACE(bh, "free child branches"); |
| 2412 | ext3_free_branches(handle, inode, bh, |
| 2413 | (__le32*)bh->b_data, |
| 2414 | (__le32*)bh->b_data + addr_per_block, |
| 2415 | depth); |
| 2416 | |
| 2417 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2418 | * Everything below this this pointer has been |
| 2419 | * released. Now let this top-of-subtree go. |
| 2420 | * |
| 2421 | * We want the freeing of this indirect block to be |
| 2422 | * atomic in the journal with the updating of the |
| 2423 | * bitmap block which owns it. So make some room in |
| 2424 | * the journal. |
| 2425 | * |
| 2426 | * We zero the parent pointer *after* freeing its |
| 2427 | * pointee in the bitmaps, so if extend_transaction() |
| 2428 | * for some reason fails to put the bitmap changes and |
| 2429 | * the release into the same transaction, recovery |
| 2430 | * will merely complain about releasing a free block, |
| 2431 | * rather than leaking blocks. |
| 2432 | */ |
| 2433 | if (is_handle_aborted(handle)) |
| 2434 | return; |
| 2435 | if (try_to_extend_transaction(handle, inode)) { |
| 2436 | ext3_mark_inode_dirty(handle, inode); |
Jan Kara | 00171d3 | 2009-08-11 19:06:10 +0200 | [diff] [blame] | 2437 | truncate_restart_transaction(handle, inode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2438 | } |
| 2439 | |
Jan Kara | f25f624 | 2010-07-12 21:04:31 +0200 | [diff] [blame] | 2440 | /* |
| 2441 | * We've probably journalled the indirect block several |
| 2442 | * times during the truncate. But it's no longer |
| 2443 | * needed and we now drop it from the transaction via |
| 2444 | * journal_revoke(). |
| 2445 | * |
| 2446 | * That's easy if it's exclusively part of this |
| 2447 | * transaction. But if it's part of the committing |
| 2448 | * transaction then journal_forget() will simply |
| 2449 | * brelse() it. That means that if the underlying |
| 2450 | * block is reallocated in ext3_get_block(), |
| 2451 | * unmap_underlying_metadata() will find this block |
| 2452 | * and will try to get rid of it. damn, damn. Thus |
| 2453 | * we don't allow a block to be reallocated until |
| 2454 | * a transaction freeing it has fully committed. |
| 2455 | * |
| 2456 | * We also have to make sure journal replay after a |
| 2457 | * crash does not overwrite non-journaled data blocks |
| 2458 | * with old metadata when the block got reallocated for |
| 2459 | * data. Thus we have to store a revoke record for a |
| 2460 | * block in the same transaction in which we free the |
| 2461 | * block. |
| 2462 | */ |
| 2463 | ext3_forget(handle, 1, inode, bh, bh->b_blocknr); |
| 2464 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2465 | ext3_free_blocks(handle, inode, nr, 1); |
| 2466 | |
| 2467 | if (parent_bh) { |
| 2468 | /* |
| 2469 | * The block which we have just freed is |
| 2470 | * pointed to by an indirect block: journal it |
| 2471 | */ |
| 2472 | BUFFER_TRACE(parent_bh, "get_write_access"); |
| 2473 | if (!ext3_journal_get_write_access(handle, |
| 2474 | parent_bh)){ |
| 2475 | *p = 0; |
| 2476 | BUFFER_TRACE(parent_bh, |
| 2477 | "call ext3_journal_dirty_metadata"); |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 2478 | ext3_journal_dirty_metadata(handle, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2479 | parent_bh); |
| 2480 | } |
| 2481 | } |
| 2482 | } |
| 2483 | } else { |
| 2484 | /* We have reached the bottom of the tree. */ |
| 2485 | BUFFER_TRACE(parent_bh, "free data blocks"); |
| 2486 | ext3_free_data(handle, inode, parent_bh, first, last); |
| 2487 | } |
| 2488 | } |
| 2489 | |
Duane Griffin | ae76dd9 | 2008-07-25 01:46:23 -0700 | [diff] [blame] | 2490 | int ext3_can_truncate(struct inode *inode) |
| 2491 | { |
Duane Griffin | ae76dd9 | 2008-07-25 01:46:23 -0700 | [diff] [blame] | 2492 | if (S_ISREG(inode->i_mode)) |
| 2493 | return 1; |
| 2494 | if (S_ISDIR(inode->i_mode)) |
| 2495 | return 1; |
| 2496 | if (S_ISLNK(inode->i_mode)) |
| 2497 | return !ext3_inode_is_fast_symlink(inode); |
| 2498 | return 0; |
| 2499 | } |
| 2500 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2501 | /* |
| 2502 | * ext3_truncate() |
| 2503 | * |
| 2504 | * We block out ext3_get_block() block instantiations across the entire |
| 2505 | * transaction, and VFS/VM ensures that ext3_truncate() cannot run |
| 2506 | * simultaneously on behalf of the same inode. |
| 2507 | * |
Justin P. Mattock | 42b2aa8 | 2011-11-28 20:31:00 -0800 | [diff] [blame] | 2508 | * As we work through the truncate and commit bits of it to the journal there |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2509 | * is one core, guiding principle: the file's tree must always be consistent on |
| 2510 | * disk. We must be able to restart the truncate after a crash. |
| 2511 | * |
| 2512 | * The file's tree may be transiently inconsistent in memory (although it |
| 2513 | * probably isn't), but whenever we close off and commit a journal transaction, |
| 2514 | * the contents of (the filesystem + the journal) must be consistent and |
| 2515 | * restartable. It's pretty simple, really: bottom up, right to left (although |
| 2516 | * left-to-right works OK too). |
| 2517 | * |
| 2518 | * Note that at recovery time, journal replay occurs *before* the restart of |
| 2519 | * truncate against the orphan inode list. |
| 2520 | * |
| 2521 | * The committed inode has the new, desired i_size (which is the same as |
| 2522 | * i_disksize in this case). After a crash, ext3_orphan_cleanup() will see |
| 2523 | * that this inode's truncate did not complete and it will again call |
| 2524 | * ext3_truncate() to have another go. So there will be instantiated blocks |
| 2525 | * to the right of the truncation point in a crashed ext3 filesystem. But |
| 2526 | * that's fine - as long as they are linked from the inode, the post-crash |
| 2527 | * ext3_truncate() run will find them and release them. |
| 2528 | */ |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 2529 | void ext3_truncate(struct inode *inode) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2530 | { |
| 2531 | handle_t *handle; |
| 2532 | struct ext3_inode_info *ei = EXT3_I(inode); |
| 2533 | __le32 *i_data = ei->i_data; |
| 2534 | int addr_per_block = EXT3_ADDR_PER_BLOCK(inode->i_sb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2535 | int offsets[4]; |
| 2536 | Indirect chain[4]; |
| 2537 | Indirect *partial; |
| 2538 | __le32 nr = 0; |
| 2539 | int n; |
| 2540 | long last_block; |
| 2541 | unsigned blocksize = inode->i_sb->s_blocksize; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2542 | |
Lukas Czerner | 785c4bc | 2011-05-23 18:33:01 +0200 | [diff] [blame] | 2543 | trace_ext3_truncate_enter(inode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2544 | |
Duane Griffin | ae76dd9 | 2008-07-25 01:46:23 -0700 | [diff] [blame] | 2545 | if (!ext3_can_truncate(inode)) |
Jan Kara | ef43618 | 2009-06-17 16:26:24 -0700 | [diff] [blame] | 2546 | goto out_notrans; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2547 | |
Theodore Ts'o | f7ab34e | 2009-04-03 01:34:35 -0400 | [diff] [blame] | 2548 | if (inode->i_size == 0 && ext3_should_writeback_data(inode)) |
Jan Kara | 9df9393 | 2010-01-06 21:58:48 +0100 | [diff] [blame] | 2549 | ext3_set_inode_state(inode, EXT3_STATE_FLUSH_ON_CLOSE); |
Theodore Ts'o | f7ab34e | 2009-04-03 01:34:35 -0400 | [diff] [blame] | 2550 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2551 | handle = start_transaction(inode); |
Jan Kara | ee3e77f | 2011-06-03 21:58:11 +0200 | [diff] [blame] | 2552 | if (IS_ERR(handle)) |
Jan Kara | ef43618 | 2009-06-17 16:26:24 -0700 | [diff] [blame] | 2553 | goto out_notrans; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2554 | |
| 2555 | last_block = (inode->i_size + blocksize-1) |
| 2556 | >> EXT3_BLOCK_SIZE_BITS(inode->i_sb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2557 | n = ext3_block_to_path(inode, last_block, offsets, NULL); |
| 2558 | if (n == 0) |
| 2559 | goto out_stop; /* error */ |
| 2560 | |
| 2561 | /* |
| 2562 | * OK. This truncate is going to happen. We add the inode to the |
| 2563 | * orphan list, so that if this truncate spans multiple transactions, |
| 2564 | * and we crash, we will resume the truncate when the filesystem |
| 2565 | * recovers. It also marks the inode dirty, to catch the new size. |
| 2566 | * |
| 2567 | * Implication: the file must always be in a sane, consistent |
| 2568 | * truncatable state while each transaction commits. |
| 2569 | */ |
| 2570 | if (ext3_orphan_add(handle, inode)) |
| 2571 | goto out_stop; |
| 2572 | |
| 2573 | /* |
| 2574 | * The orphan list entry will now protect us from any crash which |
| 2575 | * occurs before the truncate completes, so it is now safe to propagate |
| 2576 | * the new, shorter inode size (held for now in i_size) into the |
| 2577 | * on-disk inode. We do this via i_disksize, which is the value which |
| 2578 | * ext3 *really* writes onto the disk inode. |
| 2579 | */ |
| 2580 | ei->i_disksize = inode->i_size; |
| 2581 | |
| 2582 | /* |
| 2583 | * From here we block out all ext3_get_block() callers who want to |
| 2584 | * modify the block allocation tree. |
| 2585 | */ |
Arjan van de Ven | 9746151 | 2006-03-23 03:00:42 -0800 | [diff] [blame] | 2586 | mutex_lock(&ei->truncate_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2587 | |
| 2588 | if (n == 1) { /* direct blocks */ |
| 2589 | ext3_free_data(handle, inode, NULL, i_data+offsets[0], |
| 2590 | i_data + EXT3_NDIR_BLOCKS); |
| 2591 | goto do_indirects; |
| 2592 | } |
| 2593 | |
| 2594 | partial = ext3_find_shared(inode, n, offsets, chain, &nr); |
| 2595 | /* Kill the top of shared branch (not detached) */ |
| 2596 | if (nr) { |
| 2597 | if (partial == chain) { |
| 2598 | /* Shared branch grows from the inode */ |
| 2599 | ext3_free_branches(handle, inode, NULL, |
| 2600 | &nr, &nr+1, (chain+n-1) - partial); |
| 2601 | *partial->p = 0; |
| 2602 | /* |
| 2603 | * We mark the inode dirty prior to restart, |
| 2604 | * and prior to stop. No need for it here. |
| 2605 | */ |
| 2606 | } else { |
| 2607 | /* Shared branch grows from an indirect block */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2608 | ext3_free_branches(handle, inode, partial->bh, |
| 2609 | partial->p, |
| 2610 | partial->p+1, (chain+n-1) - partial); |
| 2611 | } |
| 2612 | } |
| 2613 | /* Clear the ends of indirect blocks on the shared branch */ |
| 2614 | while (partial > chain) { |
| 2615 | ext3_free_branches(handle, inode, partial->bh, partial->p + 1, |
| 2616 | (__le32*)partial->bh->b_data+addr_per_block, |
| 2617 | (chain+n-1) - partial); |
| 2618 | BUFFER_TRACE(partial->bh, "call brelse"); |
| 2619 | brelse (partial->bh); |
| 2620 | partial--; |
| 2621 | } |
| 2622 | do_indirects: |
| 2623 | /* Kill the remaining (whole) subtrees */ |
| 2624 | switch (offsets[0]) { |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 2625 | default: |
| 2626 | nr = i_data[EXT3_IND_BLOCK]; |
| 2627 | if (nr) { |
| 2628 | ext3_free_branches(handle, inode, NULL, &nr, &nr+1, 1); |
| 2629 | i_data[EXT3_IND_BLOCK] = 0; |
| 2630 | } |
| 2631 | case EXT3_IND_BLOCK: |
| 2632 | nr = i_data[EXT3_DIND_BLOCK]; |
| 2633 | if (nr) { |
| 2634 | ext3_free_branches(handle, inode, NULL, &nr, &nr+1, 2); |
| 2635 | i_data[EXT3_DIND_BLOCK] = 0; |
| 2636 | } |
| 2637 | case EXT3_DIND_BLOCK: |
| 2638 | nr = i_data[EXT3_TIND_BLOCK]; |
| 2639 | if (nr) { |
| 2640 | ext3_free_branches(handle, inode, NULL, &nr, &nr+1, 3); |
| 2641 | i_data[EXT3_TIND_BLOCK] = 0; |
| 2642 | } |
| 2643 | case EXT3_TIND_BLOCK: |
| 2644 | ; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2645 | } |
| 2646 | |
| 2647 | ext3_discard_reservation(inode); |
| 2648 | |
Arjan van de Ven | 9746151 | 2006-03-23 03:00:42 -0800 | [diff] [blame] | 2649 | mutex_unlock(&ei->truncate_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2650 | inode->i_mtime = inode->i_ctime = CURRENT_TIME_SEC; |
| 2651 | ext3_mark_inode_dirty(handle, inode); |
| 2652 | |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 2653 | /* |
| 2654 | * In a multi-transaction truncate, we only make the final transaction |
| 2655 | * synchronous |
| 2656 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2657 | if (IS_SYNC(inode)) |
| 2658 | handle->h_sync = 1; |
| 2659 | out_stop: |
| 2660 | /* |
| 2661 | * If this was a simple ftruncate(), and the file will remain alive |
| 2662 | * then we need to clear up the orphan record which we created above. |
| 2663 | * However, if this was a real unlink then we were called by |
Al Viro | ac14a95 | 2010-06-06 07:08:19 -0400 | [diff] [blame] | 2664 | * ext3_evict_inode(), and we allow that function to clean up the |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2665 | * orphan info for us. |
| 2666 | */ |
| 2667 | if (inode->i_nlink) |
| 2668 | ext3_orphan_del(handle, inode); |
| 2669 | |
| 2670 | ext3_journal_stop(handle); |
Lukas Czerner | 785c4bc | 2011-05-23 18:33:01 +0200 | [diff] [blame] | 2671 | trace_ext3_truncate_exit(inode); |
Jan Kara | ef43618 | 2009-06-17 16:26:24 -0700 | [diff] [blame] | 2672 | return; |
| 2673 | out_notrans: |
| 2674 | /* |
| 2675 | * Delete the inode from orphan list so that it doesn't stay there |
| 2676 | * forever and trigger assertion on umount. |
| 2677 | */ |
| 2678 | if (inode->i_nlink) |
| 2679 | ext3_orphan_del(NULL, inode); |
Lukas Czerner | 785c4bc | 2011-05-23 18:33:01 +0200 | [diff] [blame] | 2680 | trace_ext3_truncate_exit(inode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2681 | } |
| 2682 | |
Mingming Cao | 43d23f9 | 2006-06-25 05:48:07 -0700 | [diff] [blame] | 2683 | static ext3_fsblk_t ext3_get_inode_block(struct super_block *sb, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2684 | unsigned long ino, struct ext3_iloc *iloc) |
| 2685 | { |
Akinobu Mita | e0e369a | 2008-04-28 02:16:08 -0700 | [diff] [blame] | 2686 | unsigned long block_group; |
Mingming Cao | 43d23f9 | 2006-06-25 05:48:07 -0700 | [diff] [blame] | 2687 | unsigned long offset; |
| 2688 | ext3_fsblk_t block; |
Akinobu Mita | e0e369a | 2008-04-28 02:16:08 -0700 | [diff] [blame] | 2689 | struct ext3_group_desc *gdp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2690 | |
Neil Brown | 2ccb48e | 2006-07-30 03:03:01 -0700 | [diff] [blame] | 2691 | if (!ext3_valid_inum(sb, ino)) { |
| 2692 | /* |
| 2693 | * This error is already checked for in namei.c unless we are |
| 2694 | * looking at an NFS filehandle, in which case no error |
| 2695 | * report is needed |
| 2696 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2697 | return 0; |
| 2698 | } |
Neil Brown | 2ccb48e | 2006-07-30 03:03:01 -0700 | [diff] [blame] | 2699 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2700 | block_group = (ino - 1) / EXT3_INODES_PER_GROUP(sb); |
Akinobu Mita | e0e369a | 2008-04-28 02:16:08 -0700 | [diff] [blame] | 2701 | gdp = ext3_get_group_desc(sb, block_group, NULL); |
| 2702 | if (!gdp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2703 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2704 | /* |
| 2705 | * Figure out the offset within the block group inode table |
| 2706 | */ |
| 2707 | offset = ((ino - 1) % EXT3_INODES_PER_GROUP(sb)) * |
| 2708 | EXT3_INODE_SIZE(sb); |
Akinobu Mita | e0e369a | 2008-04-28 02:16:08 -0700 | [diff] [blame] | 2709 | block = le32_to_cpu(gdp->bg_inode_table) + |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2710 | (offset >> EXT3_BLOCK_SIZE_BITS(sb)); |
| 2711 | |
| 2712 | iloc->block_group = block_group; |
| 2713 | iloc->offset = offset & (EXT3_BLOCK_SIZE(sb) - 1); |
| 2714 | return block; |
| 2715 | } |
| 2716 | |
| 2717 | /* |
| 2718 | * ext3_get_inode_loc returns with an extra refcount against the inode's |
| 2719 | * underlying buffer_head on success. If 'in_mem' is true, we have all |
| 2720 | * data in memory that is needed to recreate the on-disk version of this |
| 2721 | * inode. |
| 2722 | */ |
| 2723 | static int __ext3_get_inode_loc(struct inode *inode, |
| 2724 | struct ext3_iloc *iloc, int in_mem) |
| 2725 | { |
Mingming Cao | 43d23f9 | 2006-06-25 05:48:07 -0700 | [diff] [blame] | 2726 | ext3_fsblk_t block; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2727 | struct buffer_head *bh; |
| 2728 | |
| 2729 | block = ext3_get_inode_block(inode->i_sb, inode->i_ino, iloc); |
| 2730 | if (!block) |
| 2731 | return -EIO; |
| 2732 | |
| 2733 | bh = sb_getblk(inode->i_sb, block); |
| 2734 | if (!bh) { |
| 2735 | ext3_error (inode->i_sb, "ext3_get_inode_loc", |
| 2736 | "unable to read inode block - " |
Mingming Cao | 43d23f9 | 2006-06-25 05:48:07 -0700 | [diff] [blame] | 2737 | "inode=%lu, block="E3FSBLK, |
| 2738 | inode->i_ino, block); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2739 | return -EIO; |
| 2740 | } |
| 2741 | if (!buffer_uptodate(bh)) { |
| 2742 | lock_buffer(bh); |
Hidehiro Kawai | 95450f5 | 2008-07-25 01:46:24 -0700 | [diff] [blame] | 2743 | |
| 2744 | /* |
| 2745 | * If the buffer has the write error flag, we have failed |
| 2746 | * to write out another inode in the same block. In this |
| 2747 | * case, we don't have to read the block because we may |
| 2748 | * read the old inode data successfully. |
| 2749 | */ |
| 2750 | if (buffer_write_io_error(bh) && !buffer_uptodate(bh)) |
| 2751 | set_buffer_uptodate(bh); |
| 2752 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2753 | if (buffer_uptodate(bh)) { |
| 2754 | /* someone brought it uptodate while we waited */ |
| 2755 | unlock_buffer(bh); |
| 2756 | goto has_buffer; |
| 2757 | } |
| 2758 | |
| 2759 | /* |
| 2760 | * If we have all information of the inode in memory and this |
| 2761 | * is the only valid inode in the block, we need not read the |
| 2762 | * block. |
| 2763 | */ |
| 2764 | if (in_mem) { |
| 2765 | struct buffer_head *bitmap_bh; |
| 2766 | struct ext3_group_desc *desc; |
| 2767 | int inodes_per_buffer; |
| 2768 | int inode_offset, i; |
| 2769 | int block_group; |
| 2770 | int start; |
| 2771 | |
| 2772 | block_group = (inode->i_ino - 1) / |
| 2773 | EXT3_INODES_PER_GROUP(inode->i_sb); |
| 2774 | inodes_per_buffer = bh->b_size / |
| 2775 | EXT3_INODE_SIZE(inode->i_sb); |
| 2776 | inode_offset = ((inode->i_ino - 1) % |
| 2777 | EXT3_INODES_PER_GROUP(inode->i_sb)); |
| 2778 | start = inode_offset & ~(inodes_per_buffer - 1); |
| 2779 | |
| 2780 | /* Is the inode bitmap in cache? */ |
| 2781 | desc = ext3_get_group_desc(inode->i_sb, |
| 2782 | block_group, NULL); |
| 2783 | if (!desc) |
| 2784 | goto make_io; |
| 2785 | |
| 2786 | bitmap_bh = sb_getblk(inode->i_sb, |
| 2787 | le32_to_cpu(desc->bg_inode_bitmap)); |
| 2788 | if (!bitmap_bh) |
| 2789 | goto make_io; |
| 2790 | |
| 2791 | /* |
| 2792 | * If the inode bitmap isn't in cache then the |
| 2793 | * optimisation may end up performing two reads instead |
| 2794 | * of one, so skip it. |
| 2795 | */ |
| 2796 | if (!buffer_uptodate(bitmap_bh)) { |
| 2797 | brelse(bitmap_bh); |
| 2798 | goto make_io; |
| 2799 | } |
| 2800 | for (i = start; i < start + inodes_per_buffer; i++) { |
| 2801 | if (i == inode_offset) |
| 2802 | continue; |
| 2803 | if (ext3_test_bit(i, bitmap_bh->b_data)) |
| 2804 | break; |
| 2805 | } |
| 2806 | brelse(bitmap_bh); |
| 2807 | if (i == start + inodes_per_buffer) { |
| 2808 | /* all other inodes are free, so skip I/O */ |
| 2809 | memset(bh->b_data, 0, bh->b_size); |
| 2810 | set_buffer_uptodate(bh); |
| 2811 | unlock_buffer(bh); |
| 2812 | goto has_buffer; |
| 2813 | } |
| 2814 | } |
| 2815 | |
| 2816 | make_io: |
| 2817 | /* |
| 2818 | * There are other valid inodes in the buffer, this inode |
| 2819 | * has in-inode xattrs, or we don't have this inode in memory. |
| 2820 | * Read the block from disk. |
| 2821 | */ |
Lukas Czerner | 785c4bc | 2011-05-23 18:33:01 +0200 | [diff] [blame] | 2822 | trace_ext3_load_inode(inode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2823 | get_bh(bh); |
| 2824 | bh->b_end_io = end_buffer_read_sync; |
Christoph Hellwig | 65299a3 | 2011-08-23 14:50:29 +0200 | [diff] [blame] | 2825 | submit_bh(READ | REQ_META | REQ_PRIO, bh); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2826 | wait_on_buffer(bh); |
| 2827 | if (!buffer_uptodate(bh)) { |
| 2828 | ext3_error(inode->i_sb, "ext3_get_inode_loc", |
| 2829 | "unable to read inode block - " |
Mingming Cao | 43d23f9 | 2006-06-25 05:48:07 -0700 | [diff] [blame] | 2830 | "inode=%lu, block="E3FSBLK, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2831 | inode->i_ino, block); |
| 2832 | brelse(bh); |
| 2833 | return -EIO; |
| 2834 | } |
| 2835 | } |
| 2836 | has_buffer: |
| 2837 | iloc->bh = bh; |
| 2838 | return 0; |
| 2839 | } |
| 2840 | |
| 2841 | int ext3_get_inode_loc(struct inode *inode, struct ext3_iloc *iloc) |
| 2842 | { |
| 2843 | /* We have all inode data except xattrs in memory here. */ |
| 2844 | return __ext3_get_inode_loc(inode, iloc, |
Jan Kara | 9df9393 | 2010-01-06 21:58:48 +0100 | [diff] [blame] | 2845 | !ext3_test_inode_state(inode, EXT3_STATE_XATTR)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2846 | } |
| 2847 | |
| 2848 | void ext3_set_inode_flags(struct inode *inode) |
| 2849 | { |
| 2850 | unsigned int flags = EXT3_I(inode)->i_flags; |
| 2851 | |
| 2852 | inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC); |
| 2853 | if (flags & EXT3_SYNC_FL) |
| 2854 | inode->i_flags |= S_SYNC; |
| 2855 | if (flags & EXT3_APPEND_FL) |
| 2856 | inode->i_flags |= S_APPEND; |
| 2857 | if (flags & EXT3_IMMUTABLE_FL) |
| 2858 | inode->i_flags |= S_IMMUTABLE; |
| 2859 | if (flags & EXT3_NOATIME_FL) |
| 2860 | inode->i_flags |= S_NOATIME; |
| 2861 | if (flags & EXT3_DIRSYNC_FL) |
| 2862 | inode->i_flags |= S_DIRSYNC; |
| 2863 | } |
| 2864 | |
Jan Kara | 28be5ab | 2007-05-08 00:30:33 -0700 | [diff] [blame] | 2865 | /* Propagate flags from i_flags to EXT3_I(inode)->i_flags */ |
| 2866 | void ext3_get_inode_flags(struct ext3_inode_info *ei) |
| 2867 | { |
| 2868 | unsigned int flags = ei->vfs_inode.i_flags; |
| 2869 | |
| 2870 | ei->i_flags &= ~(EXT3_SYNC_FL|EXT3_APPEND_FL| |
| 2871 | EXT3_IMMUTABLE_FL|EXT3_NOATIME_FL|EXT3_DIRSYNC_FL); |
| 2872 | if (flags & S_SYNC) |
| 2873 | ei->i_flags |= EXT3_SYNC_FL; |
| 2874 | if (flags & S_APPEND) |
| 2875 | ei->i_flags |= EXT3_APPEND_FL; |
| 2876 | if (flags & S_IMMUTABLE) |
| 2877 | ei->i_flags |= EXT3_IMMUTABLE_FL; |
| 2878 | if (flags & S_NOATIME) |
| 2879 | ei->i_flags |= EXT3_NOATIME_FL; |
| 2880 | if (flags & S_DIRSYNC) |
| 2881 | ei->i_flags |= EXT3_DIRSYNC_FL; |
| 2882 | } |
| 2883 | |
David Howells | 473043d | 2008-02-07 00:15:36 -0800 | [diff] [blame] | 2884 | struct inode *ext3_iget(struct super_block *sb, unsigned long ino) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2885 | { |
| 2886 | struct ext3_iloc iloc; |
| 2887 | struct ext3_inode *raw_inode; |
David Howells | 473043d | 2008-02-07 00:15:36 -0800 | [diff] [blame] | 2888 | struct ext3_inode_info *ei; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2889 | struct buffer_head *bh; |
David Howells | 473043d | 2008-02-07 00:15:36 -0800 | [diff] [blame] | 2890 | struct inode *inode; |
Jan Kara | fe8bc91 | 2009-10-16 19:26:15 +0200 | [diff] [blame] | 2891 | journal_t *journal = EXT3_SB(sb)->s_journal; |
| 2892 | transaction_t *transaction; |
David Howells | 473043d | 2008-02-07 00:15:36 -0800 | [diff] [blame] | 2893 | long ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2894 | int block; |
| 2895 | |
David Howells | 473043d | 2008-02-07 00:15:36 -0800 | [diff] [blame] | 2896 | inode = iget_locked(sb, ino); |
| 2897 | if (!inode) |
| 2898 | return ERR_PTR(-ENOMEM); |
| 2899 | if (!(inode->i_state & I_NEW)) |
| 2900 | return inode; |
| 2901 | |
| 2902 | ei = EXT3_I(inode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2903 | ei->i_block_alloc_info = NULL; |
| 2904 | |
David Howells | 473043d | 2008-02-07 00:15:36 -0800 | [diff] [blame] | 2905 | ret = __ext3_get_inode_loc(inode, &iloc, 0); |
| 2906 | if (ret < 0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2907 | goto bad_inode; |
| 2908 | bh = iloc.bh; |
| 2909 | raw_inode = ext3_raw_inode(&iloc); |
| 2910 | inode->i_mode = le16_to_cpu(raw_inode->i_mode); |
| 2911 | inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low); |
| 2912 | inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low); |
| 2913 | if(!(test_opt (inode->i_sb, NO_UID32))) { |
| 2914 | inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16; |
| 2915 | inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16; |
| 2916 | } |
Miklos Szeredi | bfe8684 | 2011-10-28 14:13:29 +0200 | [diff] [blame] | 2917 | set_nlink(inode, le16_to_cpu(raw_inode->i_links_count)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2918 | inode->i_size = le32_to_cpu(raw_inode->i_size); |
Markus Rechberger | 4d7bf11 | 2007-05-08 00:23:39 -0700 | [diff] [blame] | 2919 | inode->i_atime.tv_sec = (signed)le32_to_cpu(raw_inode->i_atime); |
| 2920 | inode->i_ctime.tv_sec = (signed)le32_to_cpu(raw_inode->i_ctime); |
| 2921 | inode->i_mtime.tv_sec = (signed)le32_to_cpu(raw_inode->i_mtime); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2922 | inode->i_atime.tv_nsec = inode->i_ctime.tv_nsec = inode->i_mtime.tv_nsec = 0; |
| 2923 | |
Linus Torvalds | de32982 | 2010-03-29 14:30:19 -0700 | [diff] [blame] | 2924 | ei->i_state_flags = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2925 | ei->i_dir_start_lookup = 0; |
| 2926 | ei->i_dtime = le32_to_cpu(raw_inode->i_dtime); |
| 2927 | /* We now have enough fields to check if the inode was active or not. |
| 2928 | * This is needed because nfsd might try to access dead inodes |
| 2929 | * the test is that same one that e2fsck uses |
| 2930 | * NeilBrown 1999oct15 |
| 2931 | */ |
| 2932 | if (inode->i_nlink == 0) { |
| 2933 | if (inode->i_mode == 0 || |
| 2934 | !(EXT3_SB(inode->i_sb)->s_mount_state & EXT3_ORPHAN_FS)) { |
| 2935 | /* this inode is deleted */ |
| 2936 | brelse (bh); |
David Howells | 473043d | 2008-02-07 00:15:36 -0800 | [diff] [blame] | 2937 | ret = -ESTALE; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2938 | goto bad_inode; |
| 2939 | } |
| 2940 | /* The only unlinked inodes we let through here have |
| 2941 | * valid i_mode and are being read by the orphan |
| 2942 | * recovery code: that's fine, we're about to complete |
| 2943 | * the process of deleting those. */ |
| 2944 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2945 | inode->i_blocks = le32_to_cpu(raw_inode->i_blocks); |
| 2946 | ei->i_flags = le32_to_cpu(raw_inode->i_flags); |
| 2947 | #ifdef EXT3_FRAGMENTS |
| 2948 | ei->i_faddr = le32_to_cpu(raw_inode->i_faddr); |
| 2949 | ei->i_frag_no = raw_inode->i_frag; |
| 2950 | ei->i_frag_size = raw_inode->i_fsize; |
| 2951 | #endif |
| 2952 | ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl); |
| 2953 | if (!S_ISREG(inode->i_mode)) { |
| 2954 | ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl); |
| 2955 | } else { |
| 2956 | inode->i_size |= |
| 2957 | ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32; |
| 2958 | } |
| 2959 | ei->i_disksize = inode->i_size; |
| 2960 | inode->i_generation = le32_to_cpu(raw_inode->i_generation); |
| 2961 | ei->i_block_group = iloc.block_group; |
| 2962 | /* |
| 2963 | * NOTE! The in-memory inode i_data array is in little-endian order |
| 2964 | * even on big-endian machines: we do NOT byteswap the block numbers! |
| 2965 | */ |
| 2966 | for (block = 0; block < EXT3_N_BLOCKS; block++) |
| 2967 | ei->i_data[block] = raw_inode->i_block[block]; |
| 2968 | INIT_LIST_HEAD(&ei->i_orphan); |
| 2969 | |
Jan Kara | fe8bc91 | 2009-10-16 19:26:15 +0200 | [diff] [blame] | 2970 | /* |
| 2971 | * Set transaction id's of transactions that have to be committed |
| 2972 | * to finish f[data]sync. We set them to currently running transaction |
| 2973 | * as we cannot be sure that the inode or some of its metadata isn't |
| 2974 | * part of the transaction - the inode could have been reclaimed and |
| 2975 | * now it is reread from disk. |
| 2976 | */ |
| 2977 | if (journal) { |
| 2978 | tid_t tid; |
| 2979 | |
| 2980 | spin_lock(&journal->j_state_lock); |
| 2981 | if (journal->j_running_transaction) |
| 2982 | transaction = journal->j_running_transaction; |
| 2983 | else |
| 2984 | transaction = journal->j_committing_transaction; |
| 2985 | if (transaction) |
| 2986 | tid = transaction->t_tid; |
| 2987 | else |
| 2988 | tid = journal->j_commit_sequence; |
| 2989 | spin_unlock(&journal->j_state_lock); |
| 2990 | atomic_set(&ei->i_sync_tid, tid); |
| 2991 | atomic_set(&ei->i_datasync_tid, tid); |
| 2992 | } |
| 2993 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2994 | if (inode->i_ino >= EXT3_FIRST_INO(inode->i_sb) + 1 && |
| 2995 | EXT3_INODE_SIZE(inode->i_sb) > EXT3_GOOD_OLD_INODE_SIZE) { |
| 2996 | /* |
| 2997 | * When mke2fs creates big inodes it does not zero out |
| 2998 | * the unused bytes above EXT3_GOOD_OLD_INODE_SIZE, |
| 2999 | * so ignore those first few inodes. |
| 3000 | */ |
| 3001 | ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize); |
| 3002 | if (EXT3_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > |
Kirill Korotaev | e4a10a3 | 2007-06-23 17:16:48 -0700 | [diff] [blame] | 3003 | EXT3_INODE_SIZE(inode->i_sb)) { |
| 3004 | brelse (bh); |
David Howells | 473043d | 2008-02-07 00:15:36 -0800 | [diff] [blame] | 3005 | ret = -EIO; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3006 | goto bad_inode; |
Kirill Korotaev | e4a10a3 | 2007-06-23 17:16:48 -0700 | [diff] [blame] | 3007 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3008 | if (ei->i_extra_isize == 0) { |
| 3009 | /* The extra space is currently unused. Use it. */ |
| 3010 | ei->i_extra_isize = sizeof(struct ext3_inode) - |
| 3011 | EXT3_GOOD_OLD_INODE_SIZE; |
| 3012 | } else { |
| 3013 | __le32 *magic = (void *)raw_inode + |
| 3014 | EXT3_GOOD_OLD_INODE_SIZE + |
| 3015 | ei->i_extra_isize; |
| 3016 | if (*magic == cpu_to_le32(EXT3_XATTR_MAGIC)) |
Jan Kara | 9df9393 | 2010-01-06 21:58:48 +0100 | [diff] [blame] | 3017 | ext3_set_inode_state(inode, EXT3_STATE_XATTR); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3018 | } |
| 3019 | } else |
| 3020 | ei->i_extra_isize = 0; |
| 3021 | |
| 3022 | if (S_ISREG(inode->i_mode)) { |
| 3023 | inode->i_op = &ext3_file_inode_operations; |
| 3024 | inode->i_fop = &ext3_file_operations; |
| 3025 | ext3_set_aops(inode); |
| 3026 | } else if (S_ISDIR(inode->i_mode)) { |
| 3027 | inode->i_op = &ext3_dir_inode_operations; |
| 3028 | inode->i_fop = &ext3_dir_operations; |
| 3029 | } else if (S_ISLNK(inode->i_mode)) { |
Duane Griffin | b5ed311 | 2008-12-19 20:47:14 +0000 | [diff] [blame] | 3030 | if (ext3_inode_is_fast_symlink(inode)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3031 | inode->i_op = &ext3_fast_symlink_inode_operations; |
Duane Griffin | b5ed311 | 2008-12-19 20:47:14 +0000 | [diff] [blame] | 3032 | nd_terminate_link(ei->i_data, inode->i_size, |
| 3033 | sizeof(ei->i_data) - 1); |
| 3034 | } else { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3035 | inode->i_op = &ext3_symlink_inode_operations; |
| 3036 | ext3_set_aops(inode); |
| 3037 | } |
| 3038 | } else { |
| 3039 | inode->i_op = &ext3_special_inode_operations; |
| 3040 | if (raw_inode->i_block[0]) |
| 3041 | init_special_inode(inode, inode->i_mode, |
| 3042 | old_decode_dev(le32_to_cpu(raw_inode->i_block[0]))); |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 3043 | else |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3044 | init_special_inode(inode, inode->i_mode, |
| 3045 | new_decode_dev(le32_to_cpu(raw_inode->i_block[1]))); |
| 3046 | } |
| 3047 | brelse (iloc.bh); |
| 3048 | ext3_set_inode_flags(inode); |
David Howells | 473043d | 2008-02-07 00:15:36 -0800 | [diff] [blame] | 3049 | unlock_new_inode(inode); |
| 3050 | return inode; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3051 | |
| 3052 | bad_inode: |
David Howells | 473043d | 2008-02-07 00:15:36 -0800 | [diff] [blame] | 3053 | iget_failed(inode); |
| 3054 | return ERR_PTR(ret); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3055 | } |
| 3056 | |
| 3057 | /* |
| 3058 | * Post the struct inode info into an on-disk inode location in the |
| 3059 | * buffer-cache. This gobbles the caller's reference to the |
| 3060 | * buffer_head in the inode location struct. |
| 3061 | * |
| 3062 | * The caller must have write access to iloc->bh. |
| 3063 | */ |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 3064 | static int ext3_do_update_inode(handle_t *handle, |
| 3065 | struct inode *inode, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3066 | struct ext3_iloc *iloc) |
| 3067 | { |
| 3068 | struct ext3_inode *raw_inode = ext3_raw_inode(iloc); |
| 3069 | struct ext3_inode_info *ei = EXT3_I(inode); |
| 3070 | struct buffer_head *bh = iloc->bh; |
| 3071 | int err = 0, rc, block; |
Jan Kara | 66afca2 | 2012-09-03 16:50:42 +0200 | [diff] [blame] | 3072 | int need_datasync = 0; |
| 3073 | __le32 disksize; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3074 | |
Chris Mason | 4f003fd | 2009-09-08 00:22:14 +0200 | [diff] [blame] | 3075 | again: |
| 3076 | /* we can't allow multiple procs in here at once, its a bit racey */ |
| 3077 | lock_buffer(bh); |
| 3078 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3079 | /* For fields not not tracking in the in-memory inode, |
| 3080 | * initialise them to zero for new inodes. */ |
Jan Kara | 9df9393 | 2010-01-06 21:58:48 +0100 | [diff] [blame] | 3081 | if (ext3_test_inode_state(inode, EXT3_STATE_NEW)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3082 | memset(raw_inode, 0, EXT3_SB(inode->i_sb)->s_inode_size); |
| 3083 | |
Jan Kara | 28be5ab | 2007-05-08 00:30:33 -0700 | [diff] [blame] | 3084 | ext3_get_inode_flags(ei); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3085 | raw_inode->i_mode = cpu_to_le16(inode->i_mode); |
| 3086 | if(!(test_opt(inode->i_sb, NO_UID32))) { |
| 3087 | raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid)); |
| 3088 | raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid)); |
| 3089 | /* |
| 3090 | * Fix up interoperability with old kernels. Otherwise, old inodes get |
| 3091 | * re-used with the upper 16 bits of the uid/gid intact |
| 3092 | */ |
| 3093 | if(!ei->i_dtime) { |
| 3094 | raw_inode->i_uid_high = |
| 3095 | cpu_to_le16(high_16_bits(inode->i_uid)); |
| 3096 | raw_inode->i_gid_high = |
| 3097 | cpu_to_le16(high_16_bits(inode->i_gid)); |
| 3098 | } else { |
| 3099 | raw_inode->i_uid_high = 0; |
| 3100 | raw_inode->i_gid_high = 0; |
| 3101 | } |
| 3102 | } else { |
| 3103 | raw_inode->i_uid_low = |
| 3104 | cpu_to_le16(fs_high2lowuid(inode->i_uid)); |
| 3105 | raw_inode->i_gid_low = |
| 3106 | cpu_to_le16(fs_high2lowgid(inode->i_gid)); |
| 3107 | raw_inode->i_uid_high = 0; |
| 3108 | raw_inode->i_gid_high = 0; |
| 3109 | } |
| 3110 | raw_inode->i_links_count = cpu_to_le16(inode->i_nlink); |
Jan Kara | 66afca2 | 2012-09-03 16:50:42 +0200 | [diff] [blame] | 3111 | disksize = cpu_to_le32(ei->i_disksize); |
| 3112 | if (disksize != raw_inode->i_size) { |
| 3113 | need_datasync = 1; |
| 3114 | raw_inode->i_size = disksize; |
| 3115 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3116 | raw_inode->i_atime = cpu_to_le32(inode->i_atime.tv_sec); |
| 3117 | raw_inode->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec); |
| 3118 | raw_inode->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec); |
| 3119 | raw_inode->i_blocks = cpu_to_le32(inode->i_blocks); |
| 3120 | raw_inode->i_dtime = cpu_to_le32(ei->i_dtime); |
| 3121 | raw_inode->i_flags = cpu_to_le32(ei->i_flags); |
| 3122 | #ifdef EXT3_FRAGMENTS |
| 3123 | raw_inode->i_faddr = cpu_to_le32(ei->i_faddr); |
| 3124 | raw_inode->i_frag = ei->i_frag_no; |
| 3125 | raw_inode->i_fsize = ei->i_frag_size; |
| 3126 | #endif |
| 3127 | raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl); |
| 3128 | if (!S_ISREG(inode->i_mode)) { |
| 3129 | raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl); |
| 3130 | } else { |
Jan Kara | 66afca2 | 2012-09-03 16:50:42 +0200 | [diff] [blame] | 3131 | disksize = cpu_to_le32(ei->i_disksize >> 32); |
| 3132 | if (disksize != raw_inode->i_size_high) { |
| 3133 | raw_inode->i_size_high = disksize; |
| 3134 | need_datasync = 1; |
| 3135 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3136 | if (ei->i_disksize > 0x7fffffffULL) { |
| 3137 | struct super_block *sb = inode->i_sb; |
| 3138 | if (!EXT3_HAS_RO_COMPAT_FEATURE(sb, |
| 3139 | EXT3_FEATURE_RO_COMPAT_LARGE_FILE) || |
| 3140 | EXT3_SB(sb)->s_es->s_rev_level == |
| 3141 | cpu_to_le32(EXT3_GOOD_OLD_REV)) { |
| 3142 | /* If this is the first large file |
| 3143 | * created, add a flag to the superblock. |
| 3144 | */ |
Chris Mason | 4f003fd | 2009-09-08 00:22:14 +0200 | [diff] [blame] | 3145 | unlock_buffer(bh); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3146 | err = ext3_journal_get_write_access(handle, |
| 3147 | EXT3_SB(sb)->s_sbh); |
| 3148 | if (err) |
| 3149 | goto out_brelse; |
Chris Mason | 4f003fd | 2009-09-08 00:22:14 +0200 | [diff] [blame] | 3150 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3151 | ext3_update_dynamic_rev(sb); |
| 3152 | EXT3_SET_RO_COMPAT_FEATURE(sb, |
| 3153 | EXT3_FEATURE_RO_COMPAT_LARGE_FILE); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3154 | handle->h_sync = 1; |
| 3155 | err = ext3_journal_dirty_metadata(handle, |
| 3156 | EXT3_SB(sb)->s_sbh); |
Chris Mason | 4f003fd | 2009-09-08 00:22:14 +0200 | [diff] [blame] | 3157 | /* get our lock and start over */ |
| 3158 | goto again; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3159 | } |
| 3160 | } |
| 3161 | } |
| 3162 | raw_inode->i_generation = cpu_to_le32(inode->i_generation); |
| 3163 | if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) { |
| 3164 | if (old_valid_dev(inode->i_rdev)) { |
| 3165 | raw_inode->i_block[0] = |
| 3166 | cpu_to_le32(old_encode_dev(inode->i_rdev)); |
| 3167 | raw_inode->i_block[1] = 0; |
| 3168 | } else { |
| 3169 | raw_inode->i_block[0] = 0; |
| 3170 | raw_inode->i_block[1] = |
| 3171 | cpu_to_le32(new_encode_dev(inode->i_rdev)); |
| 3172 | raw_inode->i_block[2] = 0; |
| 3173 | } |
| 3174 | } else for (block = 0; block < EXT3_N_BLOCKS; block++) |
| 3175 | raw_inode->i_block[block] = ei->i_data[block]; |
| 3176 | |
Andreas Gruenbacher | ff87b37 | 2005-07-07 17:57:00 -0700 | [diff] [blame] | 3177 | if (ei->i_extra_isize) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3178 | raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize); |
| 3179 | |
| 3180 | BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata"); |
Chris Mason | 4f003fd | 2009-09-08 00:22:14 +0200 | [diff] [blame] | 3181 | unlock_buffer(bh); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3182 | rc = ext3_journal_dirty_metadata(handle, bh); |
| 3183 | if (!err) |
| 3184 | err = rc; |
Jan Kara | 9df9393 | 2010-01-06 21:58:48 +0100 | [diff] [blame] | 3185 | ext3_clear_inode_state(inode, EXT3_STATE_NEW); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3186 | |
Jan Kara | fe8bc91 | 2009-10-16 19:26:15 +0200 | [diff] [blame] | 3187 | atomic_set(&ei->i_sync_tid, handle->h_transaction->t_tid); |
Jan Kara | 66afca2 | 2012-09-03 16:50:42 +0200 | [diff] [blame] | 3188 | if (need_datasync) |
| 3189 | atomic_set(&ei->i_datasync_tid, handle->h_transaction->t_tid); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3190 | out_brelse: |
| 3191 | brelse (bh); |
| 3192 | ext3_std_error(inode->i_sb, err); |
| 3193 | return err; |
| 3194 | } |
| 3195 | |
| 3196 | /* |
| 3197 | * ext3_write_inode() |
| 3198 | * |
| 3199 | * We are called from a few places: |
| 3200 | * |
| 3201 | * - Within generic_file_write() for O_SYNC files. |
| 3202 | * Here, there will be no transaction running. We wait for any running |
| 3203 | * trasnaction to commit. |
| 3204 | * |
| 3205 | * - Within sys_sync(), kupdate and such. |
| 3206 | * We wait on commit, if tol to. |
| 3207 | * |
| 3208 | * - Within prune_icache() (PF_MEMALLOC == true) |
| 3209 | * Here we simply return. We can't afford to block kswapd on the |
| 3210 | * journal commit. |
| 3211 | * |
| 3212 | * In all cases it is actually safe for us to return without doing anything, |
| 3213 | * because the inode has been copied into a raw inode buffer in |
| 3214 | * ext3_mark_inode_dirty(). This is a correctness thing for O_SYNC and for |
| 3215 | * knfsd. |
| 3216 | * |
| 3217 | * Note that we are absolutely dependent upon all inode dirtiers doing the |
| 3218 | * right thing: they *must* call mark_inode_dirty() after dirtying info in |
| 3219 | * which we are interested. |
| 3220 | * |
| 3221 | * It would be a bug for them to not do this. The code: |
| 3222 | * |
| 3223 | * mark_inode_dirty(inode) |
| 3224 | * stuff(); |
| 3225 | * inode->i_size = expr; |
| 3226 | * |
| 3227 | * is in error because a kswapd-driven write_inode() could occur while |
| 3228 | * `stuff()' is running, and the new i_size will be lost. Plus the inode |
| 3229 | * will no longer be on the superblock's dirty inode list. |
| 3230 | */ |
Christoph Hellwig | a9185b4 | 2010-03-05 09:21:37 +0100 | [diff] [blame] | 3231 | int ext3_write_inode(struct inode *inode, struct writeback_control *wbc) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3232 | { |
| 3233 | if (current->flags & PF_MEMALLOC) |
| 3234 | return 0; |
| 3235 | |
| 3236 | if (ext3_journal_current_handle()) { |
Jose R. Santos | 9ad163a | 2007-10-18 23:39:23 -0700 | [diff] [blame] | 3237 | jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3238 | dump_stack(); |
| 3239 | return -EIO; |
| 3240 | } |
| 3241 | |
Christoph Hellwig | a9185b4 | 2010-03-05 09:21:37 +0100 | [diff] [blame] | 3242 | if (wbc->sync_mode != WB_SYNC_ALL) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3243 | return 0; |
| 3244 | |
| 3245 | return ext3_force_commit(inode->i_sb); |
| 3246 | } |
| 3247 | |
| 3248 | /* |
| 3249 | * ext3_setattr() |
| 3250 | * |
| 3251 | * Called from notify_change. |
| 3252 | * |
| 3253 | * We want to trap VFS attempts to truncate the file as soon as |
| 3254 | * possible. In particular, we want to make sure that when the VFS |
| 3255 | * shrinks i_size, we put the inode on the orphan list and modify |
| 3256 | * i_disksize immediately, so that during the subsequent flushing of |
| 3257 | * dirty pages and freeing of disk blocks, we can guarantee that any |
| 3258 | * commit will leave the blocks being flushed in an unused state on |
| 3259 | * disk. (On recovery, the inode will get truncated and the blocks will |
| 3260 | * be freed, so we have a strong guarantee that no future commit will |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 3261 | * leave these blocks visible to the user.) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3262 | * |
| 3263 | * Called with inode->sem down. |
| 3264 | */ |
| 3265 | int ext3_setattr(struct dentry *dentry, struct iattr *attr) |
| 3266 | { |
| 3267 | struct inode *inode = dentry->d_inode; |
| 3268 | int error, rc = 0; |
| 3269 | const unsigned int ia_valid = attr->ia_valid; |
| 3270 | |
| 3271 | error = inode_change_ok(inode, attr); |
| 3272 | if (error) |
| 3273 | return error; |
| 3274 | |
Dmitry Monakhov | 1275562 | 2010-04-08 22:04:20 +0400 | [diff] [blame] | 3275 | if (is_quota_modification(inode, attr)) |
Christoph Hellwig | 871a293 | 2010-03-03 09:05:07 -0500 | [diff] [blame] | 3276 | dquot_initialize(inode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3277 | if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) || |
| 3278 | (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) { |
| 3279 | handle_t *handle; |
| 3280 | |
| 3281 | /* (user+group)*(old+new) structure, inode write (sb, |
| 3282 | * inode block, ? - but truncate inode update has it) */ |
Dmitry Monakhov | c459001 | 2009-12-09 03:05:30 +0300 | [diff] [blame] | 3283 | handle = ext3_journal_start(inode, EXT3_MAXQUOTAS_INIT_BLOCKS(inode->i_sb)+ |
| 3284 | EXT3_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)+3); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3285 | if (IS_ERR(handle)) { |
| 3286 | error = PTR_ERR(handle); |
| 3287 | goto err_out; |
| 3288 | } |
Christoph Hellwig | b43fa82 | 2010-03-03 09:05:03 -0500 | [diff] [blame] | 3289 | error = dquot_transfer(inode, attr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3290 | if (error) { |
| 3291 | ext3_journal_stop(handle); |
| 3292 | return error; |
| 3293 | } |
| 3294 | /* Update corresponding info in inode so that everything is in |
| 3295 | * one transaction */ |
| 3296 | if (attr->ia_valid & ATTR_UID) |
| 3297 | inode->i_uid = attr->ia_uid; |
| 3298 | if (attr->ia_valid & ATTR_GID) |
| 3299 | inode->i_gid = attr->ia_gid; |
| 3300 | error = ext3_mark_inode_dirty(handle, inode); |
| 3301 | ext3_journal_stop(handle); |
| 3302 | } |
| 3303 | |
Christoph Hellwig | 562c72a | 2011-06-24 14:29:45 -0400 | [diff] [blame] | 3304 | if (attr->ia_valid & ATTR_SIZE) |
| 3305 | inode_dio_wait(inode); |
| 3306 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3307 | if (S_ISREG(inode->i_mode) && |
| 3308 | attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) { |
| 3309 | handle_t *handle; |
| 3310 | |
| 3311 | handle = ext3_journal_start(inode, 3); |
| 3312 | if (IS_ERR(handle)) { |
| 3313 | error = PTR_ERR(handle); |
| 3314 | goto err_out; |
| 3315 | } |
| 3316 | |
| 3317 | error = ext3_orphan_add(handle, inode); |
Jan Kara | ee3e77f | 2011-06-03 21:58:11 +0200 | [diff] [blame] | 3318 | if (error) { |
| 3319 | ext3_journal_stop(handle); |
| 3320 | goto err_out; |
| 3321 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3322 | EXT3_I(inode)->i_disksize = attr->ia_size; |
Jan Kara | ee3e77f | 2011-06-03 21:58:11 +0200 | [diff] [blame] | 3323 | error = ext3_mark_inode_dirty(handle, inode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3324 | ext3_journal_stop(handle); |
Jan Kara | ee3e77f | 2011-06-03 21:58:11 +0200 | [diff] [blame] | 3325 | if (error) { |
| 3326 | /* Some hard fs error must have happened. Bail out. */ |
| 3327 | ext3_orphan_del(NULL, inode); |
| 3328 | goto err_out; |
| 3329 | } |
| 3330 | rc = ext3_block_truncate_page(inode, attr->ia_size); |
| 3331 | if (rc) { |
| 3332 | /* Cleanup orphan list and exit */ |
| 3333 | handle = ext3_journal_start(inode, 3); |
| 3334 | if (IS_ERR(handle)) { |
| 3335 | ext3_orphan_del(NULL, inode); |
| 3336 | goto err_out; |
| 3337 | } |
| 3338 | ext3_orphan_del(handle, inode); |
| 3339 | ext3_journal_stop(handle); |
| 3340 | goto err_out; |
| 3341 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3342 | } |
| 3343 | |
Christoph Hellwig | 1025774 | 2010-06-04 11:30:02 +0200 | [diff] [blame] | 3344 | if ((attr->ia_valid & ATTR_SIZE) && |
| 3345 | attr->ia_size != i_size_read(inode)) { |
Jan Kara | 40680f2f | 2011-05-24 22:24:47 +0200 | [diff] [blame] | 3346 | truncate_setsize(inode, attr->ia_size); |
| 3347 | ext3_truncate(inode); |
Christoph Hellwig | 1025774 | 2010-06-04 11:30:02 +0200 | [diff] [blame] | 3348 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3349 | |
Christoph Hellwig | 1025774 | 2010-06-04 11:30:02 +0200 | [diff] [blame] | 3350 | setattr_copy(inode, attr); |
| 3351 | mark_inode_dirty(inode); |
| 3352 | |
| 3353 | if (ia_valid & ATTR_MODE) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3354 | rc = ext3_acl_chmod(inode); |
| 3355 | |
| 3356 | err_out: |
| 3357 | ext3_std_error(inode->i_sb, error); |
| 3358 | if (!error) |
| 3359 | error = rc; |
| 3360 | return error; |
| 3361 | } |
| 3362 | |
| 3363 | |
| 3364 | /* |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 3365 | * How many blocks doth make a writepage()? |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3366 | * |
| 3367 | * With N blocks per page, it may be: |
| 3368 | * N data blocks |
| 3369 | * 2 indirect block |
| 3370 | * 2 dindirect |
| 3371 | * 1 tindirect |
| 3372 | * N+5 bitmap blocks (from the above) |
| 3373 | * N+5 group descriptor summary blocks |
| 3374 | * 1 inode block |
| 3375 | * 1 superblock. |
| 3376 | * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quote files |
| 3377 | * |
| 3378 | * 3 * (N + 5) + 2 + 2 * EXT3_SINGLEDATA_TRANS_BLOCKS |
| 3379 | * |
| 3380 | * With ordered or writeback data it's the same, less the N data blocks. |
| 3381 | * |
| 3382 | * If the inode's direct blocks can hold an integral number of pages then a |
| 3383 | * page cannot straddle two indirect blocks, and we can only touch one indirect |
| 3384 | * and dindirect block, and the "5" above becomes "3". |
| 3385 | * |
| 3386 | * This still overestimates under most circumstances. If we were to pass the |
| 3387 | * start and end offsets in here as well we could do block_to_path() on each |
| 3388 | * block and work out the exact number of indirects which are touched. Pah. |
| 3389 | */ |
| 3390 | |
| 3391 | static int ext3_writepage_trans_blocks(struct inode *inode) |
| 3392 | { |
| 3393 | int bpp = ext3_journal_blocks_per_page(inode); |
| 3394 | int indirects = (EXT3_NDIR_BLOCKS % bpp) ? 5 : 3; |
| 3395 | int ret; |
| 3396 | |
| 3397 | if (ext3_should_journal_data(inode)) |
| 3398 | ret = 3 * (bpp + indirects) + 2; |
| 3399 | else |
Yongqiang Yang | 523334b | 2011-03-24 08:48:39 +0800 | [diff] [blame] | 3400 | ret = 2 * (bpp + indirects) + indirects + 2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3401 | |
| 3402 | #ifdef CONFIG_QUOTA |
Christoph Hellwig | 871a293 | 2010-03-03 09:05:07 -0500 | [diff] [blame] | 3403 | /* We know that structure was already allocated during dquot_initialize so |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3404 | * we will be updating only the data blocks + inodes */ |
Dmitry Monakhov | c459001 | 2009-12-09 03:05:30 +0300 | [diff] [blame] | 3405 | ret += EXT3_MAXQUOTAS_TRANS_BLOCKS(inode->i_sb); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3406 | #endif |
| 3407 | |
| 3408 | return ret; |
| 3409 | } |
| 3410 | |
| 3411 | /* |
| 3412 | * The caller must have previously called ext3_reserve_inode_write(). |
| 3413 | * Give this, we know that the caller already has write access to iloc->bh. |
| 3414 | */ |
| 3415 | int ext3_mark_iloc_dirty(handle_t *handle, |
| 3416 | struct inode *inode, struct ext3_iloc *iloc) |
| 3417 | { |
| 3418 | int err = 0; |
| 3419 | |
| 3420 | /* the do_update_inode consumes one bh->b_count */ |
| 3421 | get_bh(iloc->bh); |
| 3422 | |
| 3423 | /* ext3_do_update_inode() does journal_dirty_metadata */ |
| 3424 | err = ext3_do_update_inode(handle, inode, iloc); |
| 3425 | put_bh(iloc->bh); |
| 3426 | return err; |
| 3427 | } |
| 3428 | |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 3429 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3430 | * On success, We end up with an outstanding reference count against |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 3431 | * iloc->bh. This _must_ be cleaned up later. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3432 | */ |
| 3433 | |
| 3434 | int |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 3435 | ext3_reserve_inode_write(handle_t *handle, struct inode *inode, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3436 | struct ext3_iloc *iloc) |
| 3437 | { |
| 3438 | int err = 0; |
| 3439 | if (handle) { |
| 3440 | err = ext3_get_inode_loc(inode, iloc); |
| 3441 | if (!err) { |
| 3442 | BUFFER_TRACE(iloc->bh, "get_write_access"); |
| 3443 | err = ext3_journal_get_write_access(handle, iloc->bh); |
| 3444 | if (err) { |
| 3445 | brelse(iloc->bh); |
| 3446 | iloc->bh = NULL; |
| 3447 | } |
| 3448 | } |
| 3449 | } |
| 3450 | ext3_std_error(inode->i_sb, err); |
| 3451 | return err; |
| 3452 | } |
| 3453 | |
| 3454 | /* |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 3455 | * What we do here is to mark the in-core inode as clean with respect to inode |
| 3456 | * dirtiness (it may still be data-dirty). |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3457 | * This means that the in-core inode may be reaped by prune_icache |
| 3458 | * without having to perform any I/O. This is a very good thing, |
| 3459 | * because *any* task may call prune_icache - even ones which |
| 3460 | * have a transaction open against a different journal. |
| 3461 | * |
| 3462 | * Is this cheating? Not really. Sure, we haven't written the |
| 3463 | * inode out, but prune_icache isn't a user-visible syncing function. |
| 3464 | * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync) |
| 3465 | * we start and wait on commits. |
| 3466 | * |
| 3467 | * Is this efficient/effective? Well, we're being nice to the system |
| 3468 | * by cleaning up our inodes proactively so they can be reaped |
| 3469 | * without I/O. But we are potentially leaving up to five seconds' |
| 3470 | * worth of inodes floating about which prune_icache wants us to |
| 3471 | * write out. One way to fix that would be to get prune_icache() |
| 3472 | * to do a write_super() to free up some memory. It has the desired |
| 3473 | * effect. |
| 3474 | */ |
| 3475 | int ext3_mark_inode_dirty(handle_t *handle, struct inode *inode) |
| 3476 | { |
| 3477 | struct ext3_iloc iloc; |
| 3478 | int err; |
| 3479 | |
| 3480 | might_sleep(); |
Lukas Czerner | 785c4bc | 2011-05-23 18:33:01 +0200 | [diff] [blame] | 3481 | trace_ext3_mark_inode_dirty(inode, _RET_IP_); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3482 | err = ext3_reserve_inode_write(handle, inode, &iloc); |
| 3483 | if (!err) |
| 3484 | err = ext3_mark_iloc_dirty(handle, inode, &iloc); |
| 3485 | return err; |
| 3486 | } |
| 3487 | |
| 3488 | /* |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 3489 | * ext3_dirty_inode() is called from __mark_inode_dirty() |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3490 | * |
| 3491 | * We're really interested in the case where a file is being extended. |
| 3492 | * i_size has been changed by generic_commit_write() and we thus need |
| 3493 | * to include the updated inode in the current transaction. |
| 3494 | * |
Christoph Hellwig | 5dd4056 | 2010-03-03 09:05:00 -0500 | [diff] [blame] | 3495 | * Also, dquot_alloc_space() will always dirty the inode when blocks |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3496 | * are allocated to the file. |
| 3497 | * |
| 3498 | * If the inode is marked synchronous, we don't honour that here - doing |
| 3499 | * so would cause a commit on atime updates, which we don't bother doing. |
| 3500 | * We handle synchronous inodes at the highest possible level. |
| 3501 | */ |
Christoph Hellwig | aa38572 | 2011-05-27 06:53:02 -0400 | [diff] [blame] | 3502 | void ext3_dirty_inode(struct inode *inode, int flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3503 | { |
| 3504 | handle_t *current_handle = ext3_journal_current_handle(); |
| 3505 | handle_t *handle; |
| 3506 | |
| 3507 | handle = ext3_journal_start(inode, 2); |
| 3508 | if (IS_ERR(handle)) |
| 3509 | goto out; |
| 3510 | if (current_handle && |
| 3511 | current_handle->h_transaction != handle->h_transaction) { |
| 3512 | /* This task has a transaction open against a different fs */ |
| 3513 | printk(KERN_EMERG "%s: transactions do not match!\n", |
Harvey Harrison | e05b6b5 | 2008-04-28 02:16:15 -0700 | [diff] [blame] | 3514 | __func__); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3515 | } else { |
| 3516 | jbd_debug(5, "marking dirty. outer handle=%p\n", |
| 3517 | current_handle); |
| 3518 | ext3_mark_inode_dirty(handle, inode); |
| 3519 | } |
| 3520 | ext3_journal_stop(handle); |
| 3521 | out: |
| 3522 | return; |
| 3523 | } |
| 3524 | |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 3525 | #if 0 |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 3526 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3527 | * Bind an inode's backing buffer_head into this transaction, to prevent |
| 3528 | * it from being flushed to disk early. Unlike |
| 3529 | * ext3_reserve_inode_write, this leaves behind no bh reference and |
| 3530 | * returns no iloc structure, so the caller needs to repeat the iloc |
| 3531 | * lookup to mark the inode dirty later. |
| 3532 | */ |
Andrew Morton | d6859bf | 2006-03-26 01:38:03 -0800 | [diff] [blame] | 3533 | static int ext3_pin_inode(handle_t *handle, struct inode *inode) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3534 | { |
| 3535 | struct ext3_iloc iloc; |
| 3536 | |
| 3537 | int err = 0; |
| 3538 | if (handle) { |
| 3539 | err = ext3_get_inode_loc(inode, &iloc); |
| 3540 | if (!err) { |
| 3541 | BUFFER_TRACE(iloc.bh, "get_write_access"); |
| 3542 | err = journal_get_write_access(handle, iloc.bh); |
| 3543 | if (!err) |
Mingming Cao | ae6ddcc | 2006-09-27 01:49:27 -0700 | [diff] [blame] | 3544 | err = ext3_journal_dirty_metadata(handle, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3545 | iloc.bh); |
| 3546 | brelse(iloc.bh); |
| 3547 | } |
| 3548 | } |
| 3549 | ext3_std_error(inode->i_sb, err); |
| 3550 | return err; |
| 3551 | } |
| 3552 | #endif |
| 3553 | |
| 3554 | int ext3_change_inode_journal_flag(struct inode *inode, int val) |
| 3555 | { |
| 3556 | journal_t *journal; |
| 3557 | handle_t *handle; |
| 3558 | int err; |
| 3559 | |
| 3560 | /* |
| 3561 | * We have to be very careful here: changing a data block's |
| 3562 | * journaling status dynamically is dangerous. If we write a |
| 3563 | * data block to the journal, change the status and then delete |
| 3564 | * that block, we risk forgetting to revoke the old log record |
| 3565 | * from the journal and so a subsequent replay can corrupt data. |
| 3566 | * So, first we make sure that the journal is empty and that |
| 3567 | * nobody is changing anything. |
| 3568 | */ |
| 3569 | |
| 3570 | journal = EXT3_JOURNAL(inode); |
Dave Hansen | e3a68e3 | 2007-07-15 23:41:14 -0700 | [diff] [blame] | 3571 | if (is_journal_aborted(journal)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3572 | return -EROFS; |
| 3573 | |
| 3574 | journal_lock_updates(journal); |
| 3575 | journal_flush(journal); |
| 3576 | |
| 3577 | /* |
| 3578 | * OK, there are no updates running now, and all cached data is |
| 3579 | * synced to disk. We are now in a completely consistent state |
| 3580 | * which doesn't have anything in the journal, and we know that |
| 3581 | * no filesystem updates are running, so it is safe to modify |
| 3582 | * the inode's in-core data-journaling state flag now. |
| 3583 | */ |
| 3584 | |
| 3585 | if (val) |
| 3586 | EXT3_I(inode)->i_flags |= EXT3_JOURNAL_DATA_FL; |
| 3587 | else |
| 3588 | EXT3_I(inode)->i_flags &= ~EXT3_JOURNAL_DATA_FL; |
| 3589 | ext3_set_aops(inode); |
| 3590 | |
| 3591 | journal_unlock_updates(journal); |
| 3592 | |
| 3593 | /* Finally we can mark the inode as dirty. */ |
| 3594 | |
| 3595 | handle = ext3_journal_start(inode, 1); |
| 3596 | if (IS_ERR(handle)) |
| 3597 | return PTR_ERR(handle); |
| 3598 | |
| 3599 | err = ext3_mark_inode_dirty(handle, inode); |
| 3600 | handle->h_sync = 1; |
| 3601 | ext3_journal_stop(handle); |
| 3602 | ext3_std_error(inode->i_sb, err); |
| 3603 | |
| 3604 | return err; |
| 3605 | } |