blob: 41f8e55afcd11491c5f25bf0989f74c12b9baf69 [file] [log] [blame]
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001/*
Mingming Cao617ba132006-10-11 01:20:53 -07002 * linux/fs/ext4/inode.c
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003 *
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
8 *
9 * from
10 *
11 * linux/fs/minix/inode.c
12 *
13 * Copyright (C) 1991, 1992 Linus Torvalds
14 *
Dave Kleikampac27a0e2006-10-11 01:20:50 -070015 * 64-bit file support on 64-bit platforms by Jakub Jelinek
16 * (jj@sunsite.ms.mff.cuni.cz)
17 *
Mingming Cao617ba132006-10-11 01:20:53 -070018 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
Dave Kleikampac27a0e2006-10-11 01:20:50 -070019 */
20
Dave Kleikampac27a0e2006-10-11 01:20:50 -070021#include <linux/fs.h>
22#include <linux/time.h>
Dave Kleikampac27a0e2006-10-11 01:20:50 -070023#include <linux/highuid.h>
24#include <linux/pagemap.h>
25#include <linux/quotaops.h>
26#include <linux/string.h>
27#include <linux/buffer_head.h>
28#include <linux/writeback.h>
Alex Tomas64769242008-07-11 19:27:31 -040029#include <linux/pagevec.h>
Dave Kleikampac27a0e2006-10-11 01:20:50 -070030#include <linux/mpage.h>
Duane Griffine83c1392008-12-19 20:47:15 +000031#include <linux/namei.h>
Dave Kleikampac27a0e2006-10-11 01:20:50 -070032#include <linux/uio.h>
33#include <linux/bio.h>
Mingming Cao4c0425f2009-09-28 15:48:41 -040034#include <linux/workqueue.h>
Jiaying Zhang744692d2010-03-04 16:14:02 -050035#include <linux/kernel.h>
Andrew Morton6db26ff2011-01-12 16:59:13 -080036#include <linux/printk.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090037#include <linux/slab.h>
Theodore Ts'o00a1a052014-03-30 10:20:01 -040038#include <linux/bitops.h>
Theodore Ts'o9bffad12009-06-17 11:48:11 -040039
Christoph Hellwig3dcf5452008-04-29 18:13:32 -040040#include "ext4_jbd2.h"
Dave Kleikampac27a0e2006-10-11 01:20:50 -070041#include "xattr.h"
42#include "acl.h"
Theodore Ts'o9f125d62011-06-27 19:16:04 -040043#include "truncate.h"
Dave Kleikampac27a0e2006-10-11 01:20:50 -070044
Theodore Ts'o9bffad12009-06-17 11:48:11 -040045#include <trace/events/ext4.h>
46
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -040047#define MPAGE_DA_EXTENT_TAIL 0x01
48
Darrick J. Wong814525f2012-04-29 18:31:10 -040049static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw,
50 struct ext4_inode_info *ei)
51{
52 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
53 __u16 csum_lo;
54 __u16 csum_hi = 0;
55 __u32 csum;
56
Dmitry Monakhov171a7f22013-04-09 23:56:48 -040057 csum_lo = le16_to_cpu(raw->i_checksum_lo);
Darrick J. Wong814525f2012-04-29 18:31:10 -040058 raw->i_checksum_lo = 0;
59 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
60 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
Dmitry Monakhov171a7f22013-04-09 23:56:48 -040061 csum_hi = le16_to_cpu(raw->i_checksum_hi);
Darrick J. Wong814525f2012-04-29 18:31:10 -040062 raw->i_checksum_hi = 0;
63 }
64
65 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw,
66 EXT4_INODE_SIZE(inode->i_sb));
67
Dmitry Monakhov171a7f22013-04-09 23:56:48 -040068 raw->i_checksum_lo = cpu_to_le16(csum_lo);
Darrick J. Wong814525f2012-04-29 18:31:10 -040069 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
70 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
Dmitry Monakhov171a7f22013-04-09 23:56:48 -040071 raw->i_checksum_hi = cpu_to_le16(csum_hi);
Darrick J. Wong814525f2012-04-29 18:31:10 -040072
73 return csum;
74}
75
76static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
77 struct ext4_inode_info *ei)
78{
79 __u32 provided, calculated;
80
81 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
82 cpu_to_le32(EXT4_OS_LINUX) ||
Dmitry Monakhov9aa5d322014-10-13 03:36:16 -040083 !ext4_has_metadata_csum(inode->i_sb))
Darrick J. Wong814525f2012-04-29 18:31:10 -040084 return 1;
85
86 provided = le16_to_cpu(raw->i_checksum_lo);
87 calculated = ext4_inode_csum(inode, raw, ei);
88 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
89 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
90 provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
91 else
92 calculated &= 0xFFFF;
93
94 return provided == calculated;
95}
96
97static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
98 struct ext4_inode_info *ei)
99{
100 __u32 csum;
101
102 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
103 cpu_to_le32(EXT4_OS_LINUX) ||
Dmitry Monakhov9aa5d322014-10-13 03:36:16 -0400104 !ext4_has_metadata_csum(inode->i_sb))
Darrick J. Wong814525f2012-04-29 18:31:10 -0400105 return;
106
107 csum = ext4_inode_csum(inode, raw, ei);
108 raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
109 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
110 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
111 raw->i_checksum_hi = cpu_to_le16(csum >> 16);
112}
113
Jan Kara678aaf42008-07-11 19:27:31 -0400114static inline int ext4_begin_ordered_truncate(struct inode *inode,
115 loff_t new_size)
116{
Theodore Ts'o7ff9c072010-11-08 13:51:33 -0500117 trace_ext4_begin_ordered_truncate(inode, new_size);
Theodore Ts'o8aefcd52011-01-10 12:29:43 -0500118 /*
119 * If jinode is zero, then we never opened the file for
120 * writing, so there's no need to call
121 * jbd2_journal_begin_ordered_truncate() since there's no
122 * outstanding writes we need to flush.
123 */
124 if (!EXT4_I(inode)->jinode)
125 return 0;
126 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
127 EXT4_I(inode)->jinode,
128 new_size);
Jan Kara678aaf42008-07-11 19:27:31 -0400129}
130
Lukas Czernerd47992f2013-05-21 23:17:23 -0400131static void ext4_invalidatepage(struct page *page, unsigned int offset,
132 unsigned int length);
Theodore Ts'ocb20d512010-10-27 21:30:09 -0400133static int __ext4_journalled_writepage(struct page *page, unsigned int len);
134static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
Jan Karafffb2732013-06-04 13:01:11 -0400135static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
136 int pextents);
Alex Tomas64769242008-07-11 19:27:31 -0400137
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700138/*
139 * Test whether an inode is a fast symlink.
140 */
Theodore Ts'of348c252015-04-16 01:55:00 -0400141int ext4_inode_is_fast_symlink(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700142{
Yongqiang Yang65eddb52014-01-06 14:06:18 -0500143 int ea_blocks = EXT4_I(inode)->i_file_acl ?
144 EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700145
Zheng Liubd9db172014-06-02 10:48:22 -0400146 if (ext4_has_inline_data(inode))
147 return 0;
148
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700149 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
150}
151
152/*
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700153 * Restart the transaction associated with *handle. This does a commit,
154 * so before we call here everything must be consistently dirtied against
155 * this transaction.
156 */
Aneesh Kumar K.Vfa5d1112009-11-02 18:50:49 -0500157int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
Jan Kara487caee2009-08-17 22:17:20 -0400158 int nblocks)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700159{
Jan Kara487caee2009-08-17 22:17:20 -0400160 int ret;
161
162 /*
Theodore Ts'oe35fd662010-05-16 19:00:00 -0400163 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
Jan Kara487caee2009-08-17 22:17:20 -0400164 * moment, get_block can be called only for blocks inside i_size since
165 * page cache has been already dropped and writes are blocked by
166 * i_mutex. So we can safely drop the i_data_sem here.
167 */
Frank Mayhar03901312009-01-07 00:06:22 -0500168 BUG_ON(EXT4_JOURNAL(inode) == NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700169 jbd_debug(2, "restarting handle %p\n", handle);
Jan Kara487caee2009-08-17 22:17:20 -0400170 up_write(&EXT4_I(inode)->i_data_sem);
Amir Goldstein8e8eaab2011-02-27 23:32:12 -0500171 ret = ext4_journal_restart(handle, nblocks);
Jan Kara487caee2009-08-17 22:17:20 -0400172 down_write(&EXT4_I(inode)->i_data_sem);
Aneesh Kumar K.Vfa5d1112009-11-02 18:50:49 -0500173 ext4_discard_preallocations(inode);
Jan Kara487caee2009-08-17 22:17:20 -0400174
175 return ret;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700176}
177
178/*
179 * Called at the last iput() if i_nlink is zero.
180 */
Al Viro0930fcc2010-06-07 13:16:22 -0400181void ext4_evict_inode(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700182{
183 handle_t *handle;
Theodore Ts'obc965ab2008-08-02 21:10:38 -0400184 int err;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700185
Theodore Ts'o7ff9c072010-11-08 13:51:33 -0500186 trace_ext4_evict_inode(inode);
Jiaying Zhang2581fdc2011-08-13 12:17:13 -0400187
Al Viro0930fcc2010-06-07 13:16:22 -0400188 if (inode->i_nlink) {
Jan Kara2d859db2011-07-26 09:07:11 -0400189 /*
190 * When journalling data dirty buffers are tracked only in the
191 * journal. So although mm thinks everything is clean and
192 * ready for reaping the inode might still have some pages to
193 * write in the running transaction or waiting to be
194 * checkpointed. Thus calling jbd2_journal_invalidatepage()
195 * (via truncate_inode_pages()) to discard these buffers can
196 * cause data loss. Also even if we did not discard these
197 * buffers, we would have no way to find them after the inode
198 * is reaped and thus user could see stale data if he tries to
199 * read them before the transaction is checkpointed. So be
200 * careful and force everything to disk here... We use
201 * ei->i_datasync_tid to store the newest transaction
202 * containing inode's data.
203 *
204 * Note that directories do not have this problem because they
205 * don't use page cache.
206 */
207 if (ext4_should_journal_data(inode) &&
Theodore Ts'o2b405bf2013-03-20 09:42:11 -0400208 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) &&
209 inode->i_ino != EXT4_JOURNAL_INO) {
Jan Kara2d859db2011-07-26 09:07:11 -0400210 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
211 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
212
Theodore Ts'od76a3a772013-04-03 22:02:52 -0400213 jbd2_complete_transaction(journal, commit_tid);
Jan Kara2d859db2011-07-26 09:07:11 -0400214 filemap_write_and_wait(&inode->i_data);
215 }
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700216 truncate_inode_pages_final(&inode->i_data);
Jan Kara5dc23bd2013-06-04 14:46:12 -0400217
218 WARN_ON(atomic_read(&EXT4_I(inode)->i_ioend_count));
Al Viro0930fcc2010-06-07 13:16:22 -0400219 goto no_delete;
220 }
221
Theodore Ts'oe2bfb082014-10-05 22:47:07 -0400222 if (is_bad_inode(inode))
223 goto no_delete;
224 dquot_initialize(inode);
Christoph Hellwig907f4552010-03-03 09:05:06 -0500225
Jan Kara678aaf42008-07-11 19:27:31 -0400226 if (ext4_should_order_data(inode))
227 ext4_begin_ordered_truncate(inode, 0);
Johannes Weiner91b0abe2014-04-03 14:47:49 -0700228 truncate_inode_pages_final(&inode->i_data);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700229
Jan Kara5dc23bd2013-06-04 14:46:12 -0400230 WARN_ON(atomic_read(&EXT4_I(inode)->i_ioend_count));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700231
Jan Kara8e8ad8a2012-06-12 16:20:38 +0200232 /*
233 * Protect us against freezing - iput() caller didn't have to have any
234 * protection against it
235 */
236 sb_start_intwrite(inode->i_sb);
Theodore Ts'o9924a922013-02-08 21:59:22 -0500237 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
238 ext4_blocks_for_truncate(inode)+3);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700239 if (IS_ERR(handle)) {
Theodore Ts'obc965ab2008-08-02 21:10:38 -0400240 ext4_std_error(inode->i_sb, PTR_ERR(handle));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700241 /*
242 * If we're going to skip the normal cleanup, we still need to
243 * make sure that the in-core orphan linked list is properly
244 * cleaned up.
245 */
Mingming Cao617ba132006-10-11 01:20:53 -0700246 ext4_orphan_del(NULL, inode);
Jan Kara8e8ad8a2012-06-12 16:20:38 +0200247 sb_end_intwrite(inode->i_sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700248 goto no_delete;
249 }
250
251 if (IS_SYNC(inode))
Frank Mayhar03901312009-01-07 00:06:22 -0500252 ext4_handle_sync(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700253 inode->i_size = 0;
Theodore Ts'obc965ab2008-08-02 21:10:38 -0400254 err = ext4_mark_inode_dirty(handle, inode);
255 if (err) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500256 ext4_warning(inode->i_sb,
Theodore Ts'obc965ab2008-08-02 21:10:38 -0400257 "couldn't mark inode dirty (err %d)", err);
258 goto stop_handle;
259 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700260 if (inode->i_blocks)
Mingming Cao617ba132006-10-11 01:20:53 -0700261 ext4_truncate(inode);
Theodore Ts'obc965ab2008-08-02 21:10:38 -0400262
263 /*
264 * ext4_ext_truncate() doesn't reserve any slop when it
265 * restarts journal transactions; therefore there may not be
266 * enough credits left in the handle to remove the inode from
267 * the orphan list and set the dtime field.
268 */
Frank Mayhar03901312009-01-07 00:06:22 -0500269 if (!ext4_handle_has_enough_credits(handle, 3)) {
Theodore Ts'obc965ab2008-08-02 21:10:38 -0400270 err = ext4_journal_extend(handle, 3);
271 if (err > 0)
272 err = ext4_journal_restart(handle, 3);
273 if (err != 0) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500274 ext4_warning(inode->i_sb,
Theodore Ts'obc965ab2008-08-02 21:10:38 -0400275 "couldn't extend journal (err %d)", err);
276 stop_handle:
277 ext4_journal_stop(handle);
Theodore Ts'o45388212010-07-29 15:06:10 -0400278 ext4_orphan_del(NULL, inode);
Jan Kara8e8ad8a2012-06-12 16:20:38 +0200279 sb_end_intwrite(inode->i_sb);
Theodore Ts'obc965ab2008-08-02 21:10:38 -0400280 goto no_delete;
281 }
282 }
283
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700284 /*
Mingming Cao617ba132006-10-11 01:20:53 -0700285 * Kill off the orphan record which ext4_truncate created.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700286 * AKPM: I think this can be inside the above `if'.
Mingming Cao617ba132006-10-11 01:20:53 -0700287 * Note that ext4_orphan_del() has to be able to cope with the
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700288 * deletion of a non-existent orphan - this is because we don't
Mingming Cao617ba132006-10-11 01:20:53 -0700289 * know if ext4_truncate() actually created an orphan record.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700290 * (Well, we could do this if we need to, but heck - it works)
291 */
Mingming Cao617ba132006-10-11 01:20:53 -0700292 ext4_orphan_del(handle, inode);
293 EXT4_I(inode)->i_dtime = get_seconds();
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700294
295 /*
296 * One subtle ordering requirement: if anything has gone wrong
297 * (transaction abort, IO errors, whatever), then we can still
298 * do these next steps (the fs will already have been marked as
299 * having errors), but we can't free the inode if the mark_dirty
300 * fails.
301 */
Mingming Cao617ba132006-10-11 01:20:53 -0700302 if (ext4_mark_inode_dirty(handle, inode))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700303 /* If that failed, just do the required in-core inode clear. */
Al Viro0930fcc2010-06-07 13:16:22 -0400304 ext4_clear_inode(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700305 else
Mingming Cao617ba132006-10-11 01:20:53 -0700306 ext4_free_inode(handle, inode);
307 ext4_journal_stop(handle);
Jan Kara8e8ad8a2012-06-12 16:20:38 +0200308 sb_end_intwrite(inode->i_sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700309 return;
310no_delete:
Al Viro0930fcc2010-06-07 13:16:22 -0400311 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700312}
313
Dmitry Monakhova9e7f442009-12-14 15:21:14 +0300314#ifdef CONFIG_QUOTA
315qsize_t *ext4_get_reserved_space(struct inode *inode)
Mingming Cao60e58e02009-01-22 18:13:05 +0100316{
Dmitry Monakhova9e7f442009-12-14 15:21:14 +0300317 return &EXT4_I(inode)->i_reserved_quota;
Mingming Cao60e58e02009-01-22 18:13:05 +0100318}
Dmitry Monakhova9e7f442009-12-14 15:21:14 +0300319#endif
Theodore Ts'o9d0be502010-01-01 02:41:30 -0500320
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -0400321/*
Theodore Ts'o0637c6f2009-12-30 14:20:45 -0500322 * Called with i_data_sem down, which is important since we can call
323 * ext4_discard_preallocations() from here.
324 */
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -0500325void ext4_da_update_reserve_space(struct inode *inode,
326 int used, int quota_claim)
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -0400327{
328 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Theodore Ts'o0637c6f2009-12-30 14:20:45 -0500329 struct ext4_inode_info *ei = EXT4_I(inode);
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -0400330
Theodore Ts'o0637c6f2009-12-30 14:20:45 -0500331 spin_lock(&ei->i_block_reservation_lock);
Aditya Kalid8990242011-09-09 19:18:51 -0400332 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
Theodore Ts'o0637c6f2009-12-30 14:20:45 -0500333 if (unlikely(used > ei->i_reserved_data_blocks)) {
Theodore Ts'o8de5c322013-02-14 15:11:41 -0500334 ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
Theodore Ts'o1084f252012-03-19 23:13:43 -0400335 "with only %d reserved data blocks",
Theodore Ts'o0637c6f2009-12-30 14:20:45 -0500336 __func__, inode->i_ino, used,
337 ei->i_reserved_data_blocks);
338 WARN_ON(1);
339 used = ei->i_reserved_data_blocks;
Aneesh Kumar K.V6bc6e632008-10-10 09:39:00 -0400340 }
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -0400341
Theodore Ts'o0637c6f2009-12-30 14:20:45 -0500342 /* Update per-inode reservations */
343 ei->i_reserved_data_blocks -= used;
Theodore Ts'o71d4f7d2014-07-15 06:02:38 -0400344 percpu_counter_sub(&sbi->s_dirtyclusters_counter, used);
Theodore Ts'o0637c6f2009-12-30 14:20:45 -0500345
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -0400346 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
Mingming Cao60e58e02009-01-22 18:13:05 +0100347
Eric Sandeen72b8ab92010-05-16 11:00:00 -0400348 /* Update quota subsystem for data blocks */
349 if (quota_claim)
Aditya Kali7b415bf2011-09-09 19:04:51 -0400350 dquot_claim_block(inode, EXT4_C2B(sbi, used));
Eric Sandeen72b8ab92010-05-16 11:00:00 -0400351 else {
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -0500352 /*
353 * We did fallocate with an offset that is already delayed
354 * allocated. So on delayed allocated writeback we should
Eric Sandeen72b8ab92010-05-16 11:00:00 -0400355 * not re-claim the quota for fallocated blocks.
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -0500356 */
Aditya Kali7b415bf2011-09-09 19:04:51 -0400357 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -0500358 }
Aneesh Kumar K.Vd6014302009-03-27 22:36:43 -0400359
360 /*
361 * If we have done all the pending block allocations and if
362 * there aren't any writers on the inode, we can discard the
363 * inode's preallocations.
364 */
Theodore Ts'o0637c6f2009-12-30 14:20:45 -0500365 if ((ei->i_reserved_data_blocks == 0) &&
366 (atomic_read(&inode->i_writecount) == 0))
Aneesh Kumar K.Vd6014302009-03-27 22:36:43 -0400367 ext4_discard_preallocations(inode);
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -0400368}
369
Theodore Ts'oe29136f2010-06-29 12:54:28 -0400370static int __check_block_validity(struct inode *inode, const char *func,
Theodore Ts'oc398eda2010-07-27 11:56:40 -0400371 unsigned int line,
372 struct ext4_map_blocks *map)
Theodore Ts'o6fd058f2009-05-17 15:38:01 -0400373{
Theodore Ts'o24676da2010-05-16 21:00:00 -0400374 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
375 map->m_len)) {
Theodore Ts'oc398eda2010-07-27 11:56:40 -0400376 ext4_error_inode(inode, func, line, map->m_pblk,
377 "lblock %lu mapped to illegal pblock "
378 "(length %d)", (unsigned long) map->m_lblk,
379 map->m_len);
Theodore Ts'o6fd058f2009-05-17 15:38:01 -0400380 return -EIO;
381 }
382 return 0;
383}
384
Theodore Ts'oe29136f2010-06-29 12:54:28 -0400385#define check_block_validity(inode, map) \
Theodore Ts'oc398eda2010-07-27 11:56:40 -0400386 __check_block_validity((inode), __func__, __LINE__, (map))
Theodore Ts'oe29136f2010-06-29 12:54:28 -0400387
Dmitry Monakhov921f2662013-03-10 21:01:03 -0400388#ifdef ES_AGGRESSIVE_TEST
389static void ext4_map_blocks_es_recheck(handle_t *handle,
390 struct inode *inode,
391 struct ext4_map_blocks *es_map,
392 struct ext4_map_blocks *map,
393 int flags)
394{
395 int retval;
396
397 map->m_flags = 0;
398 /*
399 * There is a race window that the result is not the same.
400 * e.g. xfstests #223 when dioread_nolock enables. The reason
401 * is that we lookup a block mapping in extent status tree with
402 * out taking i_data_sem. So at the time the unwritten extent
403 * could be converted.
404 */
405 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
Lukas Czernerc8b459f2014-05-12 12:55:07 -0400406 down_read(&EXT4_I(inode)->i_data_sem);
Dmitry Monakhov921f2662013-03-10 21:01:03 -0400407 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
408 retval = ext4_ext_map_blocks(handle, inode, map, flags &
409 EXT4_GET_BLOCKS_KEEP_SIZE);
410 } else {
411 retval = ext4_ind_map_blocks(handle, inode, map, flags &
412 EXT4_GET_BLOCKS_KEEP_SIZE);
413 }
414 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
415 up_read((&EXT4_I(inode)->i_data_sem));
Dmitry Monakhov921f2662013-03-10 21:01:03 -0400416
417 /*
418 * We don't check m_len because extent will be collpased in status
419 * tree. So the m_len might not equal.
420 */
421 if (es_map->m_lblk != map->m_lblk ||
422 es_map->m_flags != map->m_flags ||
423 es_map->m_pblk != map->m_pblk) {
Theodore Ts'obdafe422013-07-13 00:40:31 -0400424 printk("ES cache assertion failed for inode: %lu "
Dmitry Monakhov921f2662013-03-10 21:01:03 -0400425 "es_cached ex [%d/%d/%llu/%x] != "
426 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
427 inode->i_ino, es_map->m_lblk, es_map->m_len,
428 es_map->m_pblk, es_map->m_flags, map->m_lblk,
429 map->m_len, map->m_pblk, map->m_flags,
430 retval, flags);
431 }
432}
433#endif /* ES_AGGRESSIVE_TEST */
434
Theodore Ts'o55138e02009-09-29 13:31:31 -0400435/*
Theodore Ts'oe35fd662010-05-16 19:00:00 -0400436 * The ext4_map_blocks() function tries to look up the requested blocks,
Theodore Ts'o2b2d6d02008-07-26 16:15:44 -0400437 * and returns if the blocks are already mapped.
Mingming Caof5ab0d12008-02-25 15:29:55 -0500438 *
Mingming Caof5ab0d12008-02-25 15:29:55 -0500439 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
440 * and store the allocated blocks in the result buffer head and mark it
441 * mapped.
442 *
Theodore Ts'oe35fd662010-05-16 19:00:00 -0400443 * If file type is extents based, it will call ext4_ext_map_blocks(),
444 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
Mingming Caof5ab0d12008-02-25 15:29:55 -0500445 * based files
446 *
Lukas Czerner556615d2014-04-20 23:45:47 -0400447 * On success, it returns the number of blocks being mapped or allocated.
448 * if create==0 and the blocks are pre-allocated and unwritten block,
Mingming Caof5ab0d12008-02-25 15:29:55 -0500449 * the result buffer head is unmapped. If the create ==1, it will make sure
450 * the buffer head is mapped.
451 *
452 * It returns 0 if plain look up failed (blocks have not been allocated), in
Tao Madf3ab172011-10-08 15:53:49 -0400453 * that case, buffer head is unmapped
Mingming Caof5ab0d12008-02-25 15:29:55 -0500454 *
455 * It returns the error in case of allocation failure.
456 */
Theodore Ts'oe35fd662010-05-16 19:00:00 -0400457int ext4_map_blocks(handle_t *handle, struct inode *inode,
458 struct ext4_map_blocks *map, int flags)
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -0500459{
Zheng Liud100eef2013-02-18 00:29:59 -0500460 struct extent_status es;
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -0500461 int retval;
Lukas Czernerb8a86842014-03-18 18:05:35 -0400462 int ret = 0;
Dmitry Monakhov921f2662013-03-10 21:01:03 -0400463#ifdef ES_AGGRESSIVE_TEST
464 struct ext4_map_blocks orig_map;
465
466 memcpy(&orig_map, map, sizeof(*map));
467#endif
Mingming Caof5ab0d12008-02-25 15:29:55 -0500468
Theodore Ts'oe35fd662010-05-16 19:00:00 -0400469 map->m_flags = 0;
470 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
471 "logical block %lu\n", inode->i_ino, flags, map->m_len,
472 (unsigned long) map->m_lblk);
Zheng Liud100eef2013-02-18 00:29:59 -0500473
Theodore Ts'oe861b5e2014-02-20 12:54:05 -0500474 /*
475 * ext4_map_blocks returns an int, and m_len is an unsigned int
476 */
477 if (unlikely(map->m_len > INT_MAX))
478 map->m_len = INT_MAX;
479
Kazuya Mio4adb6ab2014-04-07 10:53:28 -0400480 /* We can handle the block number less than EXT_MAX_BLOCKS */
481 if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
482 return -EIO;
483
Zheng Liud100eef2013-02-18 00:29:59 -0500484 /* Lookup extent status tree firstly */
485 if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
486 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
487 map->m_pblk = ext4_es_pblock(&es) +
488 map->m_lblk - es.es_lblk;
489 map->m_flags |= ext4_es_is_written(&es) ?
490 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
491 retval = es.es_len - (map->m_lblk - es.es_lblk);
492 if (retval > map->m_len)
493 retval = map->m_len;
494 map->m_len = retval;
495 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
496 retval = 0;
497 } else {
498 BUG_ON(1);
499 }
Dmitry Monakhov921f2662013-03-10 21:01:03 -0400500#ifdef ES_AGGRESSIVE_TEST
501 ext4_map_blocks_es_recheck(handle, inode, map,
502 &orig_map, flags);
503#endif
Zheng Liud100eef2013-02-18 00:29:59 -0500504 goto found;
505 }
506
Aneesh Kumar K.V4df3d262008-01-28 23:58:29 -0500507 /*
Theodore Ts'ob920c752009-05-14 00:54:29 -0400508 * Try to see if we can get the block without requesting a new
509 * file system block.
Aneesh Kumar K.V4df3d262008-01-28 23:58:29 -0500510 */
Zheng Liu729f52c2012-07-09 16:29:29 -0400511 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
Lukas Czernerc8b459f2014-05-12 12:55:07 -0400512 down_read(&EXT4_I(inode)->i_data_sem);
Dmitry Monakhov12e9b892010-05-16 22:00:00 -0400513 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
Dmitry Monakhova4e5d882011-10-25 08:15:12 -0400514 retval = ext4_ext_map_blocks(handle, inode, map, flags &
515 EXT4_GET_BLOCKS_KEEP_SIZE);
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -0500516 } else {
Dmitry Monakhova4e5d882011-10-25 08:15:12 -0400517 retval = ext4_ind_map_blocks(handle, inode, map, flags &
518 EXT4_GET_BLOCKS_KEEP_SIZE);
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -0500519 }
Zheng Liuf7fec032013-02-18 00:28:47 -0500520 if (retval > 0) {
Theodore Ts'o3be78c72013-08-16 21:22:41 -0400521 unsigned int status;
Zheng Liuf7fec032013-02-18 00:28:47 -0500522
Zheng Liu44fb851d2013-07-29 12:51:42 -0400523 if (unlikely(retval != map->m_len)) {
524 ext4_warning(inode->i_sb,
525 "ES len assertion failed for inode "
526 "%lu: retval %d != map->m_len %d",
527 inode->i_ino, retval, map->m_len);
528 WARN_ON(1);
Dmitry Monakhov921f2662013-03-10 21:01:03 -0400529 }
Dmitry Monakhov921f2662013-03-10 21:01:03 -0400530
Zheng Liuf7fec032013-02-18 00:28:47 -0500531 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
532 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
533 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
Lukas Czernerd2dc3172015-05-02 21:36:55 -0400534 !(status & EXTENT_STATUS_WRITTEN) &&
Zheng Liuf7fec032013-02-18 00:28:47 -0500535 ext4_find_delalloc_range(inode, map->m_lblk,
536 map->m_lblk + map->m_len - 1))
537 status |= EXTENT_STATUS_DELAYED;
538 ret = ext4_es_insert_extent(inode, map->m_lblk,
539 map->m_len, map->m_pblk, status);
540 if (ret < 0)
541 retval = ret;
542 }
Zheng Liu729f52c2012-07-09 16:29:29 -0400543 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
544 up_read((&EXT4_I(inode)->i_data_sem));
Mingming Caof5ab0d12008-02-25 15:29:55 -0500545
Zheng Liud100eef2013-02-18 00:29:59 -0500546found:
Theodore Ts'oe35fd662010-05-16 19:00:00 -0400547 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
Lukas Czernerb8a86842014-03-18 18:05:35 -0400548 ret = check_block_validity(inode, map);
Theodore Ts'o6fd058f2009-05-17 15:38:01 -0400549 if (ret != 0)
550 return ret;
551 }
552
Mingming Caof5ab0d12008-02-25 15:29:55 -0500553 /* If it is only a block(s) look up */
Theodore Ts'oc2177052009-05-14 00:58:52 -0400554 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
Aneesh Kumar K.V4df3d262008-01-28 23:58:29 -0500555 return retval;
556
557 /*
Mingming Caof5ab0d12008-02-25 15:29:55 -0500558 * Returns if the blocks have already allocated
559 *
560 * Note that if blocks have been preallocated
Tao Madf3ab172011-10-08 15:53:49 -0400561 * ext4_ext_get_block() returns the create = 0
Mingming Caof5ab0d12008-02-25 15:29:55 -0500562 * with buffer head unmapped.
563 */
Theodore Ts'oe35fd662010-05-16 19:00:00 -0400564 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
Lukas Czernerb8a86842014-03-18 18:05:35 -0400565 /*
566 * If we need to convert extent to unwritten
567 * we continue and do the actual work in
568 * ext4_ext_map_blocks()
569 */
570 if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
571 return retval;
Mingming Caof5ab0d12008-02-25 15:29:55 -0500572
573 /*
Zheng Liua25a4e12013-02-18 00:28:04 -0500574 * Here we clear m_flags because after allocating an new extent,
575 * it will be set again.
Aneesh Kumar K.V2a8964d2009-05-14 17:05:39 -0400576 */
Zheng Liua25a4e12013-02-18 00:28:04 -0500577 map->m_flags &= ~EXT4_MAP_FLAGS;
Aneesh Kumar K.V2a8964d2009-05-14 17:05:39 -0400578
579 /*
Lukas Czerner556615d2014-04-20 23:45:47 -0400580 * New blocks allocate and/or writing to unwritten extent
Mingming Caof5ab0d12008-02-25 15:29:55 -0500581 * will possibly result in updating i_data, so we take
Seunghun Leed91bd2c2014-09-01 22:15:30 -0400582 * the write lock of i_data_sem, and call get_block()
Mingming Caof5ab0d12008-02-25 15:29:55 -0500583 * with create == 1 flag.
Aneesh Kumar K.V4df3d262008-01-28 23:58:29 -0500584 */
Lukas Czernerc8b459f2014-05-12 12:55:07 -0400585 down_write(&EXT4_I(inode)->i_data_sem);
Mingming Caod2a17632008-07-14 17:52:37 -0400586
587 /*
Aneesh Kumar K.V4df3d262008-01-28 23:58:29 -0500588 * We need to check for EXT4 here because migrate
589 * could have changed the inode type in between
590 */
Dmitry Monakhov12e9b892010-05-16 22:00:00 -0400591 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
Theodore Ts'oe35fd662010-05-16 19:00:00 -0400592 retval = ext4_ext_map_blocks(handle, inode, map, flags);
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -0500593 } else {
Theodore Ts'oe35fd662010-05-16 19:00:00 -0400594 retval = ext4_ind_map_blocks(handle, inode, map, flags);
Aneesh Kumar K.V267e4db2008-04-29 08:11:12 -0400595
Theodore Ts'oe35fd662010-05-16 19:00:00 -0400596 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
Aneesh Kumar K.V267e4db2008-04-29 08:11:12 -0400597 /*
598 * We allocated new blocks which will result in
599 * i_data's format changing. Force the migrate
600 * to fail by clearing migrate flags
601 */
Theodore Ts'o19f5fb72010-01-24 14:34:07 -0500602 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
Aneesh Kumar K.V267e4db2008-04-29 08:11:12 -0400603 }
Mingming Caod2a17632008-07-14 17:52:37 -0400604
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -0500605 /*
606 * Update reserved blocks/metadata blocks after successful
607 * block allocation which had been deferred till now. We don't
608 * support fallocate for non extent files. So we can update
609 * reserve space here.
610 */
611 if ((retval > 0) &&
Aneesh Kumar K.V1296cc82010-01-15 01:27:59 -0500612 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
Aneesh Kumar K.V5f634d02010-01-25 04:00:31 -0500613 ext4_da_update_reserve_space(inode, retval, 1);
614 }
Theodore Ts'o2ac3b6e2009-05-14 13:57:08 -0400615
Zheng Liuf7fec032013-02-18 00:28:47 -0500616 if (retval > 0) {
Theodore Ts'o3be78c72013-08-16 21:22:41 -0400617 unsigned int status;
Zheng Liuf7fec032013-02-18 00:28:47 -0500618
Zheng Liu44fb851d2013-07-29 12:51:42 -0400619 if (unlikely(retval != map->m_len)) {
620 ext4_warning(inode->i_sb,
621 "ES len assertion failed for inode "
622 "%lu: retval %d != map->m_len %d",
623 inode->i_ino, retval, map->m_len);
624 WARN_ON(1);
Dmitry Monakhov921f2662013-03-10 21:01:03 -0400625 }
Dmitry Monakhov921f2662013-03-10 21:01:03 -0400626
Zheng Liuadb23552013-03-10 21:13:05 -0400627 /*
628 * If the extent has been zeroed out, we don't need to update
629 * extent status tree.
630 */
631 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
632 ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
633 if (ext4_es_is_written(&es))
634 goto has_zeroout;
635 }
Zheng Liuf7fec032013-02-18 00:28:47 -0500636 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
637 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
638 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
Lukas Czernerd2dc3172015-05-02 21:36:55 -0400639 !(status & EXTENT_STATUS_WRITTEN) &&
Zheng Liuf7fec032013-02-18 00:28:47 -0500640 ext4_find_delalloc_range(inode, map->m_lblk,
641 map->m_lblk + map->m_len - 1))
642 status |= EXTENT_STATUS_DELAYED;
643 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
644 map->m_pblk, status);
645 if (ret < 0)
646 retval = ret;
Aditya Kali5356f262011-09-09 19:20:51 -0400647 }
648
Zheng Liuadb23552013-03-10 21:13:05 -0400649has_zeroout:
Aneesh Kumar K.V4df3d262008-01-28 23:58:29 -0500650 up_write((&EXT4_I(inode)->i_data_sem));
Theodore Ts'oe35fd662010-05-16 19:00:00 -0400651 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
Lukas Czernerb8a86842014-03-18 18:05:35 -0400652 ret = check_block_validity(inode, map);
Theodore Ts'o6fd058f2009-05-17 15:38:01 -0400653 if (ret != 0)
654 return ret;
655 }
Aneesh Kumar K.V0e855ac2008-01-28 23:58:26 -0500656 return retval;
657}
658
Mingming Caof3bd1f32008-08-19 22:16:03 -0400659/* Maximum number of blocks we map for direct IO at once. */
660#define DIO_MAX_BLOCKS 4096
661
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400662static int _ext4_get_block(struct inode *inode, sector_t iblock,
663 struct buffer_head *bh, int flags)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700664{
Dmitriy Monakhov3e4fdaf2007-02-10 01:46:35 -0800665 handle_t *handle = ext4_journal_current_handle();
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400666 struct ext4_map_blocks map;
Jan Kara7fb54092008-02-10 01:08:38 -0500667 int ret = 0, started = 0;
Mingming Caof3bd1f32008-08-19 22:16:03 -0400668 int dio_credits;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700669
Tao Ma46c7f252012-12-10 14:04:52 -0500670 if (ext4_has_inline_data(inode))
671 return -ERANGE;
672
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400673 map.m_lblk = iblock;
674 map.m_len = bh->b_size >> inode->i_blkbits;
675
Anatol Pomozov8b0f1652012-11-08 15:07:16 -0500676 if (flags && !(flags & EXT4_GET_BLOCKS_NO_LOCK) && !handle) {
Jan Kara7fb54092008-02-10 01:08:38 -0500677 /* Direct IO write... */
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400678 if (map.m_len > DIO_MAX_BLOCKS)
679 map.m_len = DIO_MAX_BLOCKS;
680 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
Theodore Ts'o9924a922013-02-08 21:59:22 -0500681 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
682 dio_credits);
Jan Kara7fb54092008-02-10 01:08:38 -0500683 if (IS_ERR(handle)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700684 ret = PTR_ERR(handle);
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400685 return ret;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700686 }
Jan Kara7fb54092008-02-10 01:08:38 -0500687 started = 1;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700688 }
689
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400690 ret = ext4_map_blocks(handle, inode, &map, flags);
Jan Kara7fb54092008-02-10 01:08:38 -0500691 if (ret > 0) {
Christoph Hellwig7b7a8662013-09-04 15:04:39 +0200692 ext4_io_end_t *io_end = ext4_inode_aio(inode);
693
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400694 map_bh(bh, inode->i_sb, map.m_pblk);
695 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
Dave Chinnere842f292015-06-04 09:18:18 +1000696 if (IS_DAX(inode) && buffer_unwritten(bh)) {
697 /*
698 * dgc: I suspect unwritten conversion on ext4+DAX is
699 * fundamentally broken here when there are concurrent
700 * read/write in progress on this inode.
701 */
702 WARN_ON_ONCE(io_end);
Ross Zwisler923ae0f2015-02-16 15:59:38 -0800703 bh->b_assoc_map = inode->i_mapping;
704 bh->b_private = (void *)(unsigned long)iblock;
Ross Zwisler923ae0f2015-02-16 15:59:38 -0800705 }
Christoph Hellwig7b7a8662013-09-04 15:04:39 +0200706 if (io_end && io_end->flag & EXT4_IO_END_UNWRITTEN)
707 set_buffer_defer_completion(bh);
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400708 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
Jan Kara7fb54092008-02-10 01:08:38 -0500709 ret = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700710 }
Jan Kara7fb54092008-02-10 01:08:38 -0500711 if (started)
712 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700713 return ret;
714}
715
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400716int ext4_get_block(struct inode *inode, sector_t iblock,
717 struct buffer_head *bh, int create)
718{
719 return _ext4_get_block(inode, iblock, bh,
720 create ? EXT4_GET_BLOCKS_CREATE : 0);
721}
722
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700723/*
724 * `handle' can be NULL if create is zero
725 */
Mingming Cao617ba132006-10-11 01:20:53 -0700726struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
Theodore Ts'oc5e298a2015-06-21 01:25:29 -0400727 ext4_lblk_t block, int map_flags)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700728{
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400729 struct ext4_map_blocks map;
730 struct buffer_head *bh;
Theodore Ts'oc5e298a2015-06-21 01:25:29 -0400731 int create = map_flags & EXT4_GET_BLOCKS_CREATE;
Theodore Ts'o10560082014-08-29 20:51:32 -0400732 int err;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700733
734 J_ASSERT(handle != NULL || create == 0);
735
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400736 map.m_lblk = block;
737 map.m_len = 1;
Theodore Ts'oc5e298a2015-06-21 01:25:29 -0400738 err = ext4_map_blocks(handle, inode, &map, map_flags);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700739
Theodore Ts'o10560082014-08-29 20:51:32 -0400740 if (err == 0)
741 return create ? ERR_PTR(-ENOSPC) : NULL;
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400742 if (err < 0)
Theodore Ts'o10560082014-08-29 20:51:32 -0400743 return ERR_PTR(err);
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400744
745 bh = sb_getblk(inode->i_sb, map.m_pblk);
Theodore Ts'o10560082014-08-29 20:51:32 -0400746 if (unlikely(!bh))
747 return ERR_PTR(-ENOMEM);
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400748 if (map.m_flags & EXT4_MAP_NEW) {
749 J_ASSERT(create != 0);
750 J_ASSERT(handle != NULL);
751
752 /*
753 * Now that we do not always journal data, we should
754 * keep in mind whether this should always journal the
755 * new buffer as metadata. For now, regular file
756 * writes use ext4_get_block instead, so it's not a
757 * problem.
758 */
759 lock_buffer(bh);
760 BUFFER_TRACE(bh, "call get_create_access");
Theodore Ts'o10560082014-08-29 20:51:32 -0400761 err = ext4_journal_get_create_access(handle, bh);
762 if (unlikely(err)) {
763 unlock_buffer(bh);
764 goto errout;
765 }
766 if (!buffer_uptodate(bh)) {
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400767 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
768 set_buffer_uptodate(bh);
769 }
770 unlock_buffer(bh);
771 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
772 err = ext4_handle_dirty_metadata(handle, inode, bh);
Theodore Ts'o10560082014-08-29 20:51:32 -0400773 if (unlikely(err))
774 goto errout;
775 } else
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400776 BUFFER_TRACE(bh, "not a new buffer");
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400777 return bh;
Theodore Ts'o10560082014-08-29 20:51:32 -0400778errout:
779 brelse(bh);
780 return ERR_PTR(err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700781}
782
Mingming Cao617ba132006-10-11 01:20:53 -0700783struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
Theodore Ts'oc5e298a2015-06-21 01:25:29 -0400784 ext4_lblk_t block, int map_flags)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700785{
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400786 struct buffer_head *bh;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700787
Theodore Ts'oc5e298a2015-06-21 01:25:29 -0400788 bh = ext4_getblk(handle, inode, block, map_flags);
Theodore Ts'o1c215022014-08-29 20:52:15 -0400789 if (IS_ERR(bh))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700790 return bh;
Theodore Ts'o1c215022014-08-29 20:52:15 -0400791 if (!bh || buffer_uptodate(bh))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700792 return bh;
Christoph Hellwig65299a32011-08-23 14:50:29 +0200793 ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700794 wait_on_buffer(bh);
795 if (buffer_uptodate(bh))
796 return bh;
797 put_bh(bh);
Theodore Ts'o1c215022014-08-29 20:52:15 -0400798 return ERR_PTR(-EIO);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700799}
800
Tao Maf19d5872012-12-10 14:05:51 -0500801int ext4_walk_page_buffers(handle_t *handle,
802 struct buffer_head *head,
803 unsigned from,
804 unsigned to,
805 int *partial,
806 int (*fn)(handle_t *handle,
807 struct buffer_head *bh))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700808{
809 struct buffer_head *bh;
810 unsigned block_start, block_end;
811 unsigned blocksize = head->b_size;
812 int err, ret = 0;
813 struct buffer_head *next;
814
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400815 for (bh = head, block_start = 0;
816 ret == 0 && (bh != head || !block_start);
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400817 block_start = block_end, bh = next) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700818 next = bh->b_this_page;
819 block_end = block_start + blocksize;
820 if (block_end <= from || block_start >= to) {
821 if (partial && !buffer_uptodate(bh))
822 *partial = 1;
823 continue;
824 }
825 err = (*fn)(handle, bh);
826 if (!ret)
827 ret = err;
828 }
829 return ret;
830}
831
832/*
833 * To preserve ordering, it is essential that the hole instantiation and
834 * the data write be encapsulated in a single transaction. We cannot
Mingming Cao617ba132006-10-11 01:20:53 -0700835 * close off a transaction and start a new one between the ext4_get_block()
Mingming Caodab291a2006-10-11 01:21:01 -0700836 * and the commit_write(). So doing the jbd2_journal_start at the start of
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700837 * prepare_write() is the right place.
838 *
Jan Kara36ade452013-01-28 09:30:52 -0500839 * Also, this function can nest inside ext4_writepage(). In that case, we
840 * *know* that ext4_writepage() has generated enough buffer credits to do the
841 * whole page. So we won't block on the journal in that case, which is good,
842 * because the caller may be PF_MEMALLOC.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700843 *
Mingming Cao617ba132006-10-11 01:20:53 -0700844 * By accident, ext4 can be reentered when a transaction is open via
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700845 * quota file writes. If we were to commit the transaction while thus
846 * reentered, there can be a deadlock - we would be holding a quota
847 * lock, and the commit would never complete if another thread had a
848 * transaction open and was blocking on the quota lock - a ranking
849 * violation.
850 *
Mingming Caodab291a2006-10-11 01:21:01 -0700851 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700852 * will _not_ run commit under these circumstances because handle->h_ref
853 * is elevated. We'll still have enough credits for the tiny quotafile
854 * write.
855 */
Tao Maf19d5872012-12-10 14:05:51 -0500856int do_journal_get_write_access(handle_t *handle,
857 struct buffer_head *bh)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700858{
Jan Kara56d35a42010-08-05 14:41:42 -0400859 int dirty = buffer_dirty(bh);
860 int ret;
861
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700862 if (!buffer_mapped(bh) || buffer_freed(bh))
863 return 0;
Jan Kara56d35a42010-08-05 14:41:42 -0400864 /*
Christoph Hellwigebdec242010-10-06 10:47:23 +0200865 * __block_write_begin() could have dirtied some buffers. Clean
Jan Kara56d35a42010-08-05 14:41:42 -0400866 * the dirty bit as jbd2_journal_get_write_access() could complain
867 * otherwise about fs integrity issues. Setting of the dirty bit
Christoph Hellwigebdec242010-10-06 10:47:23 +0200868 * by __block_write_begin() isn't a real problem here as we clear
Jan Kara56d35a42010-08-05 14:41:42 -0400869 * the bit before releasing a page lock and thus writeback cannot
870 * ever write the buffer.
871 */
872 if (dirty)
873 clear_buffer_dirty(bh);
liang xie5d601252014-05-12 22:06:43 -0400874 BUFFER_TRACE(bh, "get write access");
Jan Kara56d35a42010-08-05 14:41:42 -0400875 ret = ext4_journal_get_write_access(handle, bh);
876 if (!ret && dirty)
877 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
878 return ret;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700879}
880
Anatol Pomozov8b0f1652012-11-08 15:07:16 -0500881static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
882 struct buffer_head *bh_result, int create);
Michael Halcrow2058f832015-04-12 00:55:10 -0400883
884#ifdef CONFIG_EXT4_FS_ENCRYPTION
885static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len,
886 get_block_t *get_block)
887{
888 unsigned from = pos & (PAGE_CACHE_SIZE - 1);
889 unsigned to = from + len;
890 struct inode *inode = page->mapping->host;
891 unsigned block_start, block_end;
892 sector_t block;
893 int err = 0;
894 unsigned blocksize = inode->i_sb->s_blocksize;
895 unsigned bbits;
896 struct buffer_head *bh, *head, *wait[2], **wait_bh = wait;
897 bool decrypt = false;
898
899 BUG_ON(!PageLocked(page));
900 BUG_ON(from > PAGE_CACHE_SIZE);
901 BUG_ON(to > PAGE_CACHE_SIZE);
902 BUG_ON(from > to);
903
904 if (!page_has_buffers(page))
905 create_empty_buffers(page, blocksize, 0);
906 head = page_buffers(page);
907 bbits = ilog2(blocksize);
908 block = (sector_t)page->index << (PAGE_CACHE_SHIFT - bbits);
909
910 for (bh = head, block_start = 0; bh != head || !block_start;
911 block++, block_start = block_end, bh = bh->b_this_page) {
912 block_end = block_start + blocksize;
913 if (block_end <= from || block_start >= to) {
914 if (PageUptodate(page)) {
915 if (!buffer_uptodate(bh))
916 set_buffer_uptodate(bh);
917 }
918 continue;
919 }
920 if (buffer_new(bh))
921 clear_buffer_new(bh);
922 if (!buffer_mapped(bh)) {
923 WARN_ON(bh->b_size != blocksize);
924 err = get_block(inode, block, bh, 1);
925 if (err)
926 break;
927 if (buffer_new(bh)) {
928 unmap_underlying_metadata(bh->b_bdev,
929 bh->b_blocknr);
930 if (PageUptodate(page)) {
931 clear_buffer_new(bh);
932 set_buffer_uptodate(bh);
933 mark_buffer_dirty(bh);
934 continue;
935 }
936 if (block_end > to || block_start < from)
937 zero_user_segments(page, to, block_end,
938 block_start, from);
939 continue;
940 }
941 }
942 if (PageUptodate(page)) {
943 if (!buffer_uptodate(bh))
944 set_buffer_uptodate(bh);
945 continue;
946 }
947 if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
948 !buffer_unwritten(bh) &&
949 (block_start < from || block_end > to)) {
950 ll_rw_block(READ, 1, &bh);
951 *wait_bh++ = bh;
952 decrypt = ext4_encrypted_inode(inode) &&
953 S_ISREG(inode->i_mode);
954 }
955 }
956 /*
957 * If we issued read requests, let them complete.
958 */
959 while (wait_bh > wait) {
960 wait_on_buffer(*--wait_bh);
961 if (!buffer_uptodate(*wait_bh))
962 err = -EIO;
963 }
964 if (unlikely(err))
965 page_zero_new_buffers(page, from, to);
966 else if (decrypt)
967 err = ext4_decrypt_one(inode, page);
968 return err;
969}
970#endif
971
Nick Pigginbfc1af62007-10-16 01:25:05 -0700972static int ext4_write_begin(struct file *file, struct address_space *mapping,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400973 loff_t pos, unsigned len, unsigned flags,
974 struct page **pagep, void **fsdata)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700975{
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400976 struct inode *inode = mapping->host;
Aneesh Kumar K.V1938a152009-06-05 01:00:26 -0400977 int ret, needed_blocks;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700978 handle_t *handle;
979 int retries = 0;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400980 struct page *page;
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400981 pgoff_t index;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400982 unsigned from, to;
Nick Pigginbfc1af62007-10-16 01:25:05 -0700983
Theodore Ts'o9bffad12009-06-17 11:48:11 -0400984 trace_ext4_write_begin(inode, pos, len, flags);
Aneesh Kumar K.V1938a152009-06-05 01:00:26 -0400985 /*
986 * Reserve one block more for addition to orphan list in case
987 * we allocate blocks but write fails for some reason
988 */
989 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
Theodore Ts'ode9a55b2009-06-14 17:45:34 -0400990 index = pos >> PAGE_CACHE_SHIFT;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400991 from = pos & (PAGE_CACHE_SIZE - 1);
992 to = from + len;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700993
Tao Maf19d5872012-12-10 14:05:51 -0500994 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
995 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
996 flags, pagep);
997 if (ret < 0)
Theodore Ts'o47564bf2013-02-09 09:24:14 -0500998 return ret;
999 if (ret == 1)
1000 return 0;
Tao Maf19d5872012-12-10 14:05:51 -05001001 }
1002
Theodore Ts'o47564bf2013-02-09 09:24:14 -05001003 /*
1004 * grab_cache_page_write_begin() can take a long time if the
1005 * system is thrashing due to memory pressure, or if the page
1006 * is being written back. So grab it first before we start
1007 * the transaction handle. This also allows us to allocate
1008 * the page (if needed) without using GFP_NOFS.
1009 */
1010retry_grab:
Nick Piggin54566b22009-01-04 12:00:53 -08001011 page = grab_cache_page_write_begin(mapping, index, flags);
Theodore Ts'o47564bf2013-02-09 09:24:14 -05001012 if (!page)
1013 return -ENOMEM;
1014 unlock_page(page);
1015
1016retry_journal:
Theodore Ts'o9924a922013-02-08 21:59:22 -05001017 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001018 if (IS_ERR(handle)) {
Theodore Ts'o47564bf2013-02-09 09:24:14 -05001019 page_cache_release(page);
1020 return PTR_ERR(handle);
Jan Karacf108bc2008-07-11 19:27:31 -04001021 }
Tao Maf19d5872012-12-10 14:05:51 -05001022
Theodore Ts'o47564bf2013-02-09 09:24:14 -05001023 lock_page(page);
1024 if (page->mapping != mapping) {
1025 /* The page got truncated from under us */
1026 unlock_page(page);
1027 page_cache_release(page);
Jan Karacf108bc2008-07-11 19:27:31 -04001028 ext4_journal_stop(handle);
Theodore Ts'o47564bf2013-02-09 09:24:14 -05001029 goto retry_grab;
Jan Karacf108bc2008-07-11 19:27:31 -04001030 }
Dmitry Monakhov7afe5aa2013-08-28 14:30:47 -04001031 /* In case writeback began while the page was unlocked */
1032 wait_for_stable_page(page);
Jan Karacf108bc2008-07-11 19:27:31 -04001033
Michael Halcrow2058f832015-04-12 00:55:10 -04001034#ifdef CONFIG_EXT4_FS_ENCRYPTION
1035 if (ext4_should_dioread_nolock(inode))
1036 ret = ext4_block_write_begin(page, pos, len,
1037 ext4_get_block_write);
1038 else
1039 ret = ext4_block_write_begin(page, pos, len,
1040 ext4_get_block);
1041#else
Jiaying Zhang744692d2010-03-04 16:14:02 -05001042 if (ext4_should_dioread_nolock(inode))
Christoph Hellwig6e1db882010-06-04 11:29:57 +02001043 ret = __block_write_begin(page, pos, len, ext4_get_block_write);
Jiaying Zhang744692d2010-03-04 16:14:02 -05001044 else
Christoph Hellwig6e1db882010-06-04 11:29:57 +02001045 ret = __block_write_begin(page, pos, len, ext4_get_block);
Michael Halcrow2058f832015-04-12 00:55:10 -04001046#endif
Nick Pigginbfc1af62007-10-16 01:25:05 -07001047 if (!ret && ext4_should_journal_data(inode)) {
Tao Maf19d5872012-12-10 14:05:51 -05001048 ret = ext4_walk_page_buffers(handle, page_buffers(page),
1049 from, to, NULL,
1050 do_journal_get_write_access);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001051 }
Nick Pigginbfc1af62007-10-16 01:25:05 -07001052
1053 if (ret) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001054 unlock_page(page);
Aneesh Kumar K.Vae4d5372008-09-13 13:10:25 -04001055 /*
Christoph Hellwig6e1db882010-06-04 11:29:57 +02001056 * __block_write_begin may have instantiated a few blocks
Aneesh Kumar K.Vae4d5372008-09-13 13:10:25 -04001057 * outside i_size. Trim these off again. Don't need
1058 * i_size_read because we hold i_mutex.
Aneesh Kumar K.V1938a152009-06-05 01:00:26 -04001059 *
1060 * Add inode to orphan list in case we crash before
1061 * truncate finishes
Aneesh Kumar K.Vae4d5372008-09-13 13:10:25 -04001062 */
Jan Karaffacfa72009-07-13 16:22:22 -04001063 if (pos + len > inode->i_size && ext4_can_truncate(inode))
Aneesh Kumar K.V1938a152009-06-05 01:00:26 -04001064 ext4_orphan_add(handle, inode);
1065
1066 ext4_journal_stop(handle);
1067 if (pos + len > inode->i_size) {
Jan Karab9a42072009-12-08 21:24:33 -05001068 ext4_truncate_failed_write(inode);
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001069 /*
Jan Karaffacfa72009-07-13 16:22:22 -04001070 * If truncate failed early the inode might
Aneesh Kumar K.V1938a152009-06-05 01:00:26 -04001071 * still be on the orphan list; we need to
1072 * make sure the inode is removed from the
1073 * orphan list in that case.
1074 */
1075 if (inode->i_nlink)
1076 ext4_orphan_del(NULL, inode);
1077 }
Nick Pigginbfc1af62007-10-16 01:25:05 -07001078
Theodore Ts'o47564bf2013-02-09 09:24:14 -05001079 if (ret == -ENOSPC &&
1080 ext4_should_retry_alloc(inode->i_sb, &retries))
1081 goto retry_journal;
1082 page_cache_release(page);
1083 return ret;
1084 }
1085 *pagep = page;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001086 return ret;
1087}
1088
Nick Pigginbfc1af62007-10-16 01:25:05 -07001089/* For write_end() in data=journal mode */
1090static int write_end_fn(handle_t *handle, struct buffer_head *bh)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001091{
Theodore Ts'o13fca322013-04-21 16:45:54 -04001092 int ret;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001093 if (!buffer_mapped(bh) || buffer_freed(bh))
1094 return 0;
1095 set_buffer_uptodate(bh);
Theodore Ts'o13fca322013-04-21 16:45:54 -04001096 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1097 clear_buffer_meta(bh);
1098 clear_buffer_prio(bh);
1099 return ret;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001100}
1101
Zheng Liueed43332013-04-03 12:41:17 -04001102/*
1103 * We need to pick up the new inode size which generic_commit_write gave us
1104 * `file' can be NULL - eg, when called from page_symlink().
1105 *
1106 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1107 * buffers are managed internally.
1108 */
1109static int ext4_write_end(struct file *file,
1110 struct address_space *mapping,
1111 loff_t pos, unsigned len, unsigned copied,
1112 struct page *page, void *fsdata)
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001113{
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001114 handle_t *handle = ext4_journal_current_handle();
Zheng Liueed43332013-04-03 12:41:17 -04001115 struct inode *inode = mapping->host;
Xiaoguang Wang05726392015-02-12 23:00:17 -05001116 loff_t old_size = inode->i_size;
Zheng Liueed43332013-04-03 12:41:17 -04001117 int ret = 0, ret2;
1118 int i_size_changed = 0;
1119
1120 trace_ext4_write_end(inode, pos, len, copied);
1121 if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE)) {
1122 ret = ext4_jbd2_file_inode(handle, inode);
1123 if (ret) {
1124 unlock_page(page);
1125 page_cache_release(page);
1126 goto errout;
1127 }
1128 }
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001129
Theodore Ts'o42c832d2013-07-01 08:12:39 -04001130 if (ext4_has_inline_data(inode)) {
1131 ret = ext4_write_inline_data_end(inode, pos, len,
1132 copied, page);
1133 if (ret < 0)
1134 goto errout;
1135 copied = ret;
1136 } else
Tao Maf19d5872012-12-10 14:05:51 -05001137 copied = block_write_end(file, mapping, pos,
1138 len, copied, page, fsdata);
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001139 /*
Dmitry Monakhov4631dbf2014-08-23 17:48:28 -04001140 * it's important to update i_size while still holding page lock:
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001141 * page writeout could otherwise come in and zero beyond i_size.
1142 */
Dmitry Monakhov4631dbf2014-08-23 17:48:28 -04001143 i_size_changed = ext4_update_inode_size(inode, pos + copied);
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001144 unlock_page(page);
1145 page_cache_release(page);
1146
Xiaoguang Wang05726392015-02-12 23:00:17 -05001147 if (old_size < pos)
1148 pagecache_isize_extended(inode, old_size, pos);
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001149 /*
1150 * Don't mark the inode dirty under page lock. First, it unnecessarily
1151 * makes the holding time of page lock longer. Second, it forces lock
1152 * ordering of page lock and transaction start for journaling
1153 * filesystems.
1154 */
1155 if (i_size_changed)
1156 ext4_mark_inode_dirty(handle, inode);
1157
Jan Karaffacfa72009-07-13 16:22:22 -04001158 if (pos + len > inode->i_size && ext4_can_truncate(inode))
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001159 /* if we have allocated more blocks and copied
1160 * less. We will have blocks allocated outside
1161 * inode->i_size. So truncate them
1162 */
1163 ext4_orphan_add(handle, inode);
Theodore Ts'o74d553a2013-04-03 12:39:17 -04001164errout:
Mingming Cao617ba132006-10-11 01:20:53 -07001165 ret2 = ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001166 if (!ret)
1167 ret = ret2;
Nick Pigginbfc1af62007-10-16 01:25:05 -07001168
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001169 if (pos + len > inode->i_size) {
Jan Karab9a42072009-12-08 21:24:33 -05001170 ext4_truncate_failed_write(inode);
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001171 /*
Jan Karaffacfa72009-07-13 16:22:22 -04001172 * If truncate failed early the inode might still be
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001173 * on the orphan list; we need to make sure the inode
1174 * is removed from the orphan list in that case.
1175 */
1176 if (inode->i_nlink)
1177 ext4_orphan_del(NULL, inode);
1178 }
1179
Nick Pigginbfc1af62007-10-16 01:25:05 -07001180 return ret ? ret : copied;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001181}
1182
Nick Pigginbfc1af62007-10-16 01:25:05 -07001183static int ext4_journalled_write_end(struct file *file,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001184 struct address_space *mapping,
1185 loff_t pos, unsigned len, unsigned copied,
1186 struct page *page, void *fsdata)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001187{
Mingming Cao617ba132006-10-11 01:20:53 -07001188 handle_t *handle = ext4_journal_current_handle();
Nick Pigginbfc1af62007-10-16 01:25:05 -07001189 struct inode *inode = mapping->host;
Xiaoguang Wang05726392015-02-12 23:00:17 -05001190 loff_t old_size = inode->i_size;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001191 int ret = 0, ret2;
1192 int partial = 0;
Nick Pigginbfc1af62007-10-16 01:25:05 -07001193 unsigned from, to;
Dmitry Monakhov4631dbf2014-08-23 17:48:28 -04001194 int size_changed = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001195
Theodore Ts'o9bffad12009-06-17 11:48:11 -04001196 trace_ext4_journalled_write_end(inode, pos, len, copied);
Nick Pigginbfc1af62007-10-16 01:25:05 -07001197 from = pos & (PAGE_CACHE_SIZE - 1);
1198 to = from + len;
1199
Curt Wohlgemuth441c8502011-08-13 11:25:18 -04001200 BUG_ON(!ext4_handle_valid(handle));
1201
Tao Ma3fdcfb62012-12-10 14:05:57 -05001202 if (ext4_has_inline_data(inode))
1203 copied = ext4_write_inline_data_end(inode, pos, len,
1204 copied, page);
1205 else {
1206 if (copied < len) {
1207 if (!PageUptodate(page))
1208 copied = 0;
1209 page_zero_new_buffers(page, from+copied, to);
1210 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001211
Tao Ma3fdcfb62012-12-10 14:05:57 -05001212 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1213 to, &partial, write_end_fn);
1214 if (!partial)
1215 SetPageUptodate(page);
1216 }
Dmitry Monakhov4631dbf2014-08-23 17:48:28 -04001217 size_changed = ext4_update_inode_size(inode, pos + copied);
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05001218 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
Jan Kara2d859db2011-07-26 09:07:11 -04001219 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
Dmitry Monakhov4631dbf2014-08-23 17:48:28 -04001220 unlock_page(page);
1221 page_cache_release(page);
1222
Xiaoguang Wang05726392015-02-12 23:00:17 -05001223 if (old_size < pos)
1224 pagecache_isize_extended(inode, old_size, pos);
1225
Dmitry Monakhov4631dbf2014-08-23 17:48:28 -04001226 if (size_changed) {
Mingming Cao617ba132006-10-11 01:20:53 -07001227 ret2 = ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001228 if (!ret)
1229 ret = ret2;
1230 }
Nick Pigginbfc1af62007-10-16 01:25:05 -07001231
Jan Karaffacfa72009-07-13 16:22:22 -04001232 if (pos + len > inode->i_size && ext4_can_truncate(inode))
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001233 /* if we have allocated more blocks and copied
1234 * less. We will have blocks allocated outside
1235 * inode->i_size. So truncate them
1236 */
1237 ext4_orphan_add(handle, inode);
1238
Mingming Cao617ba132006-10-11 01:20:53 -07001239 ret2 = ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001240 if (!ret)
1241 ret = ret2;
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001242 if (pos + len > inode->i_size) {
Jan Karab9a42072009-12-08 21:24:33 -05001243 ext4_truncate_failed_write(inode);
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04001244 /*
Jan Karaffacfa72009-07-13 16:22:22 -04001245 * If truncate failed early the inode might still be
Aneesh Kumar K.Vf8514082009-06-05 00:56:49 -04001246 * on the orphan list; we need to make sure the inode
1247 * is removed from the orphan list in that case.
1248 */
1249 if (inode->i_nlink)
1250 ext4_orphan_del(NULL, inode);
1251 }
Nick Pigginbfc1af62007-10-16 01:25:05 -07001252
1253 return ret ? ret : copied;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001254}
Mingming Caod2a17632008-07-14 17:52:37 -04001255
Theodore Ts'o9d0be502010-01-01 02:41:30 -05001256/*
Eric Whitneyc27e43a2015-06-21 21:37:05 -04001257 * Reserve space for a single cluster
Theodore Ts'o9d0be502010-01-01 02:41:30 -05001258 */
Eric Whitneyc27e43a2015-06-21 21:37:05 -04001259static int ext4_da_reserve_space(struct inode *inode)
Mingming Caod2a17632008-07-14 17:52:37 -04001260{
Mingming Cao60e58e02009-01-22 18:13:05 +01001261 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Theodore Ts'o0637c6f2009-12-30 14:20:45 -05001262 struct ext4_inode_info *ei = EXT4_I(inode);
Christoph Hellwig5dd40562010-03-03 09:05:00 -05001263 int ret;
Mingming Caod2a17632008-07-14 17:52:37 -04001264
Mingming Cao60e58e02009-01-22 18:13:05 +01001265 /*
Eric Sandeen72b8ab92010-05-16 11:00:00 -04001266 * We will charge metadata quota at writeout time; this saves
1267 * us from metadata over-estimation, though we may go over by
1268 * a small amount in the end. Here we just reserve for data.
Mingming Cao60e58e02009-01-22 18:13:05 +01001269 */
Aditya Kali7b415bf2011-09-09 19:04:51 -04001270 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
Christoph Hellwig5dd40562010-03-03 09:05:00 -05001271 if (ret)
1272 return ret;
Theodore Ts'o03179fe2012-07-23 00:00:20 -04001273
Theodore Ts'o03179fe2012-07-23 00:00:20 -04001274 spin_lock(&ei->i_block_reservation_lock);
Theodore Ts'o71d4f7d2014-07-15 06:02:38 -04001275 if (ext4_claim_free_clusters(sbi, 1, 0)) {
Theodore Ts'o03179fe2012-07-23 00:00:20 -04001276 spin_unlock(&ei->i_block_reservation_lock);
Theodore Ts'o03179fe2012-07-23 00:00:20 -04001277 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
Mingming Caod2a17632008-07-14 17:52:37 -04001278 return -ENOSPC;
1279 }
Theodore Ts'o9d0be502010-01-01 02:41:30 -05001280 ei->i_reserved_data_blocks++;
Eric Whitneyc27e43a2015-06-21 21:37:05 -04001281 trace_ext4_da_reserve_space(inode);
Theodore Ts'o0637c6f2009-12-30 14:20:45 -05001282 spin_unlock(&ei->i_block_reservation_lock);
Dmitry Monakhov39bc6802009-12-10 16:36:27 +00001283
Mingming Caod2a17632008-07-14 17:52:37 -04001284 return 0; /* success */
1285}
1286
Aneesh Kumar K.V12219ae2008-07-17 16:12:08 -04001287static void ext4_da_release_space(struct inode *inode, int to_free)
Mingming Caod2a17632008-07-14 17:52:37 -04001288{
1289 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Theodore Ts'o0637c6f2009-12-30 14:20:45 -05001290 struct ext4_inode_info *ei = EXT4_I(inode);
Mingming Caod2a17632008-07-14 17:52:37 -04001291
Mingming Caocd213222008-08-19 22:16:59 -04001292 if (!to_free)
1293 return; /* Nothing to release, exit */
1294
Mingming Caod2a17632008-07-14 17:52:37 -04001295 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
Mingming Caocd213222008-08-19 22:16:59 -04001296
Li Zefan5a58ec82010-05-17 02:00:00 -04001297 trace_ext4_da_release_space(inode, to_free);
Theodore Ts'o0637c6f2009-12-30 14:20:45 -05001298 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
Mingming Caocd213222008-08-19 22:16:59 -04001299 /*
Theodore Ts'o0637c6f2009-12-30 14:20:45 -05001300 * if there aren't enough reserved blocks, then the
1301 * counter is messed up somewhere. Since this
1302 * function is called from invalidate page, it's
1303 * harmless to return without any action.
Mingming Caocd213222008-08-19 22:16:59 -04001304 */
Theodore Ts'o8de5c322013-02-14 15:11:41 -05001305 ext4_warning(inode->i_sb, "ext4_da_release_space: "
Theodore Ts'o0637c6f2009-12-30 14:20:45 -05001306 "ino %lu, to_free %d with only %d reserved "
Theodore Ts'o1084f252012-03-19 23:13:43 -04001307 "data blocks", inode->i_ino, to_free,
Theodore Ts'o0637c6f2009-12-30 14:20:45 -05001308 ei->i_reserved_data_blocks);
1309 WARN_ON(1);
1310 to_free = ei->i_reserved_data_blocks;
1311 }
1312 ei->i_reserved_data_blocks -= to_free;
1313
Eric Sandeen72b8ab92010-05-16 11:00:00 -04001314 /* update fs dirty data blocks counter */
Theodore Ts'o57042652011-09-09 18:56:51 -04001315 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
Mingming Caod2a17632008-07-14 17:52:37 -04001316
Mingming Caod2a17632008-07-14 17:52:37 -04001317 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
Mingming Cao60e58e02009-01-22 18:13:05 +01001318
Aditya Kali7b415bf2011-09-09 19:04:51 -04001319 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
Mingming Caod2a17632008-07-14 17:52:37 -04001320}
1321
1322static void ext4_da_page_release_reservation(struct page *page,
Lukas Czernerca99fdd2013-05-21 23:25:01 -04001323 unsigned int offset,
1324 unsigned int length)
Mingming Caod2a17632008-07-14 17:52:37 -04001325{
1326 int to_release = 0;
1327 struct buffer_head *head, *bh;
1328 unsigned int curr_off = 0;
Aditya Kali7b415bf2011-09-09 19:04:51 -04001329 struct inode *inode = page->mapping->host;
1330 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Lukas Czernerca99fdd2013-05-21 23:25:01 -04001331 unsigned int stop = offset + length;
Aditya Kali7b415bf2011-09-09 19:04:51 -04001332 int num_clusters;
Zheng Liu51865fd2012-11-08 21:57:32 -05001333 ext4_fsblk_t lblk;
Mingming Caod2a17632008-07-14 17:52:37 -04001334
Lukas Czernerca99fdd2013-05-21 23:25:01 -04001335 BUG_ON(stop > PAGE_CACHE_SIZE || stop < length);
1336
Mingming Caod2a17632008-07-14 17:52:37 -04001337 head = page_buffers(page);
1338 bh = head;
1339 do {
1340 unsigned int next_off = curr_off + bh->b_size;
1341
Lukas Czernerca99fdd2013-05-21 23:25:01 -04001342 if (next_off > stop)
1343 break;
1344
Mingming Caod2a17632008-07-14 17:52:37 -04001345 if ((offset <= curr_off) && (buffer_delay(bh))) {
1346 to_release++;
1347 clear_buffer_delay(bh);
1348 }
1349 curr_off = next_off;
1350 } while ((bh = bh->b_this_page) != head);
Aditya Kali7b415bf2011-09-09 19:04:51 -04001351
Zheng Liu51865fd2012-11-08 21:57:32 -05001352 if (to_release) {
1353 lblk = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1354 ext4_es_remove_extent(inode, lblk, to_release);
1355 }
1356
Aditya Kali7b415bf2011-09-09 19:04:51 -04001357 /* If we have released all the blocks belonging to a cluster, then we
1358 * need to release the reserved space for that cluster. */
1359 num_clusters = EXT4_NUM_B2C(sbi, to_release);
1360 while (num_clusters > 0) {
Aditya Kali7b415bf2011-09-09 19:04:51 -04001361 lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) +
1362 ((num_clusters - 1) << sbi->s_cluster_bits);
1363 if (sbi->s_cluster_ratio == 1 ||
Zheng Liu7d1b1fb2012-11-08 21:57:35 -05001364 !ext4_find_delalloc_cluster(inode, lblk))
Aditya Kali7b415bf2011-09-09 19:04:51 -04001365 ext4_da_release_space(inode, 1);
1366
1367 num_clusters--;
1368 }
Mingming Caod2a17632008-07-14 17:52:37 -04001369}
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001370
1371/*
Alex Tomas64769242008-07-11 19:27:31 -04001372 * Delayed allocation stuff
1373 */
1374
Jan Kara4e7ea812013-06-04 13:17:40 -04001375struct mpage_da_data {
1376 struct inode *inode;
1377 struct writeback_control *wbc;
Jan Kara6b523df2013-06-04 13:21:11 -04001378
Jan Kara4e7ea812013-06-04 13:17:40 -04001379 pgoff_t first_page; /* The first page to write */
1380 pgoff_t next_page; /* Current page to examine */
1381 pgoff_t last_page; /* Last page to examine */
Aneesh Kumar K.V791b7f02009-01-05 21:50:43 -05001382 /*
Jan Kara4e7ea812013-06-04 13:17:40 -04001383 * Extent to map - this can be after first_page because that can be
1384 * fully mapped. We somewhat abuse m_flags to store whether the extent
1385 * is delalloc or unwritten.
Aneesh Kumar K.V791b7f02009-01-05 21:50:43 -05001386 */
Jan Kara4e7ea812013-06-04 13:17:40 -04001387 struct ext4_map_blocks map;
1388 struct ext4_io_submit io_submit; /* IO submission data */
1389};
Alex Tomas64769242008-07-11 19:27:31 -04001390
Jan Kara4e7ea812013-06-04 13:17:40 -04001391static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1392 bool invalidate)
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04001393{
1394 int nr_pages, i;
1395 pgoff_t index, end;
1396 struct pagevec pvec;
1397 struct inode *inode = mpd->inode;
1398 struct address_space *mapping = inode->i_mapping;
Jan Kara4e7ea812013-06-04 13:17:40 -04001399
1400 /* This is necessary when next_page == 0. */
1401 if (mpd->first_page >= mpd->next_page)
1402 return;
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04001403
Curt Wohlgemuthc7f59382011-02-26 12:27:52 -05001404 index = mpd->first_page;
1405 end = mpd->next_page - 1;
Jan Kara4e7ea812013-06-04 13:17:40 -04001406 if (invalidate) {
1407 ext4_lblk_t start, last;
1408 start = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1409 last = end << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1410 ext4_es_remove_extent(inode, start, last - start + 1);
1411 }
Zheng Liu51865fd2012-11-08 21:57:32 -05001412
Eric Sandeen66bea922012-11-14 22:22:05 -05001413 pagevec_init(&pvec, 0);
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04001414 while (index <= end) {
1415 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1416 if (nr_pages == 0)
1417 break;
1418 for (i = 0; i < nr_pages; i++) {
1419 struct page *page = pvec.pages[i];
Jan Kara9b1d09982010-03-03 16:19:32 -05001420 if (page->index > end)
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04001421 break;
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04001422 BUG_ON(!PageLocked(page));
1423 BUG_ON(PageWriteback(page));
Jan Kara4e7ea812013-06-04 13:17:40 -04001424 if (invalidate) {
1425 block_invalidatepage(page, 0, PAGE_CACHE_SIZE);
1426 ClearPageUptodate(page);
1427 }
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04001428 unlock_page(page);
1429 }
Jan Kara9b1d09982010-03-03 16:19:32 -05001430 index = pvec.pages[nr_pages - 1]->index + 1;
1431 pagevec_release(&pvec);
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04001432 }
Aneesh Kumar K.Vc4a0c462008-08-19 21:08:18 -04001433}
1434
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04001435static void ext4_print_free_blocks(struct inode *inode)
1436{
1437 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
Theodore Ts'o92b97812012-03-19 23:41:49 -04001438 struct super_block *sb = inode->i_sb;
Lukas Czernerf78ee70d2013-04-03 23:33:30 -04001439 struct ext4_inode_info *ei = EXT4_I(inode);
Theodore Ts'o92b97812012-03-19 23:41:49 -04001440
1441 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
Theodore Ts'o5dee5432011-09-09 19:10:51 -04001442 EXT4_C2B(EXT4_SB(inode->i_sb),
Lukas Czernerf78ee70d2013-04-03 23:33:30 -04001443 ext4_count_free_clusters(sb)));
Theodore Ts'o92b97812012-03-19 23:41:49 -04001444 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1445 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
Lukas Czernerf78ee70d2013-04-03 23:33:30 -04001446 (long long) EXT4_C2B(EXT4_SB(sb),
Theodore Ts'o57042652011-09-09 18:56:51 -04001447 percpu_counter_sum(&sbi->s_freeclusters_counter)));
Theodore Ts'o92b97812012-03-19 23:41:49 -04001448 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
Lukas Czernerf78ee70d2013-04-03 23:33:30 -04001449 (long long) EXT4_C2B(EXT4_SB(sb),
Aditya Kali7b415bf2011-09-09 19:04:51 -04001450 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
Theodore Ts'o92b97812012-03-19 23:41:49 -04001451 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1452 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
Lukas Czernerf78ee70d2013-04-03 23:33:30 -04001453 ei->i_reserved_data_blocks);
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04001454 return;
1455}
1456
Aneesh Kumar K.Vc364b222009-06-14 17:57:10 -04001457static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
Aneesh Kumar K.V29fa89d2009-05-12 16:30:27 -04001458{
Aneesh Kumar K.Vc364b222009-06-14 17:57:10 -04001459 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
Aneesh Kumar K.V29fa89d2009-05-12 16:30:27 -04001460}
1461
Alex Tomas64769242008-07-11 19:27:31 -04001462/*
Aditya Kali5356f262011-09-09 19:20:51 -04001463 * This function is grabs code from the very beginning of
1464 * ext4_map_blocks, but assumes that the caller is from delayed write
1465 * time. This function looks up the requested blocks and sets the
1466 * buffer delay bit under the protection of i_data_sem.
1467 */
1468static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1469 struct ext4_map_blocks *map,
1470 struct buffer_head *bh)
1471{
Zheng Liud100eef2013-02-18 00:29:59 -05001472 struct extent_status es;
Aditya Kali5356f262011-09-09 19:20:51 -04001473 int retval;
1474 sector_t invalid_block = ~((sector_t) 0xffff);
Dmitry Monakhov921f2662013-03-10 21:01:03 -04001475#ifdef ES_AGGRESSIVE_TEST
1476 struct ext4_map_blocks orig_map;
1477
1478 memcpy(&orig_map, map, sizeof(*map));
1479#endif
Aditya Kali5356f262011-09-09 19:20:51 -04001480
1481 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1482 invalid_block = ~0;
1483
1484 map->m_flags = 0;
1485 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1486 "logical block %lu\n", inode->i_ino, map->m_len,
1487 (unsigned long) map->m_lblk);
Zheng Liud100eef2013-02-18 00:29:59 -05001488
1489 /* Lookup extent status tree firstly */
1490 if (ext4_es_lookup_extent(inode, iblock, &es)) {
Zheng Liud100eef2013-02-18 00:29:59 -05001491 if (ext4_es_is_hole(&es)) {
1492 retval = 0;
Lukas Czernerc8b459f2014-05-12 12:55:07 -04001493 down_read(&EXT4_I(inode)->i_data_sem);
Zheng Liud100eef2013-02-18 00:29:59 -05001494 goto add_delayed;
1495 }
1496
1497 /*
1498 * Delayed extent could be allocated by fallocate.
1499 * So we need to check it.
1500 */
1501 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1502 map_bh(bh, inode->i_sb, invalid_block);
1503 set_buffer_new(bh);
1504 set_buffer_delay(bh);
1505 return 0;
1506 }
1507
1508 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1509 retval = es.es_len - (iblock - es.es_lblk);
1510 if (retval > map->m_len)
1511 retval = map->m_len;
1512 map->m_len = retval;
1513 if (ext4_es_is_written(&es))
1514 map->m_flags |= EXT4_MAP_MAPPED;
1515 else if (ext4_es_is_unwritten(&es))
1516 map->m_flags |= EXT4_MAP_UNWRITTEN;
1517 else
1518 BUG_ON(1);
1519
Dmitry Monakhov921f2662013-03-10 21:01:03 -04001520#ifdef ES_AGGRESSIVE_TEST
1521 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1522#endif
Zheng Liud100eef2013-02-18 00:29:59 -05001523 return retval;
1524 }
1525
Aditya Kali5356f262011-09-09 19:20:51 -04001526 /*
1527 * Try to see if we can get the block without requesting a new
1528 * file system block.
1529 */
Lukas Czernerc8b459f2014-05-12 12:55:07 -04001530 down_read(&EXT4_I(inode)->i_data_sem);
Jan Karacbd75842014-11-25 11:41:49 -05001531 if (ext4_has_inline_data(inode))
Tao Ma9c3569b2012-12-10 14:05:57 -05001532 retval = 0;
Jan Karacbd75842014-11-25 11:41:49 -05001533 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
Zheng Liu2f8e0a72014-11-25 11:44:37 -05001534 retval = ext4_ext_map_blocks(NULL, inode, map, 0);
Aditya Kali5356f262011-09-09 19:20:51 -04001535 else
Zheng Liu2f8e0a72014-11-25 11:44:37 -05001536 retval = ext4_ind_map_blocks(NULL, inode, map, 0);
Aditya Kali5356f262011-09-09 19:20:51 -04001537
Zheng Liud100eef2013-02-18 00:29:59 -05001538add_delayed:
Aditya Kali5356f262011-09-09 19:20:51 -04001539 if (retval == 0) {
Zheng Liuf7fec032013-02-18 00:28:47 -05001540 int ret;
Aditya Kali5356f262011-09-09 19:20:51 -04001541 /*
1542 * XXX: __block_prepare_write() unmaps passed block,
1543 * is it OK?
1544 */
Lukas Czerner386ad672013-03-10 22:50:00 -04001545 /*
1546 * If the block was allocated from previously allocated cluster,
1547 * then we don't need to reserve it again. However we still need
1548 * to reserve metadata for every block we're going to write.
1549 */
Eric Whitneyc27e43a2015-06-21 21:37:05 -04001550 if (EXT4_SB(inode->i_sb)->s_cluster_ratio == 1 ||
Jan Karacbd75842014-11-25 11:41:49 -05001551 !ext4_find_delalloc_cluster(inode, map->m_lblk)) {
Eric Whitneyc27e43a2015-06-21 21:37:05 -04001552 ret = ext4_da_reserve_space(inode);
Zheng Liuf7fec032013-02-18 00:28:47 -05001553 if (ret) {
Aditya Kali5356f262011-09-09 19:20:51 -04001554 /* not enough space to reserve */
Zheng Liuf7fec032013-02-18 00:28:47 -05001555 retval = ret;
Aditya Kali5356f262011-09-09 19:20:51 -04001556 goto out_unlock;
Zheng Liuf7fec032013-02-18 00:28:47 -05001557 }
Aditya Kali5356f262011-09-09 19:20:51 -04001558 }
1559
Zheng Liuf7fec032013-02-18 00:28:47 -05001560 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1561 ~0, EXTENT_STATUS_DELAYED);
1562 if (ret) {
1563 retval = ret;
Zheng Liu51865fd2012-11-08 21:57:32 -05001564 goto out_unlock;
Zheng Liuf7fec032013-02-18 00:28:47 -05001565 }
Zheng Liu51865fd2012-11-08 21:57:32 -05001566
Aditya Kali5356f262011-09-09 19:20:51 -04001567 map_bh(bh, inode->i_sb, invalid_block);
1568 set_buffer_new(bh);
1569 set_buffer_delay(bh);
Zheng Liuf7fec032013-02-18 00:28:47 -05001570 } else if (retval > 0) {
1571 int ret;
Theodore Ts'o3be78c72013-08-16 21:22:41 -04001572 unsigned int status;
Zheng Liuf7fec032013-02-18 00:28:47 -05001573
Zheng Liu44fb851d2013-07-29 12:51:42 -04001574 if (unlikely(retval != map->m_len)) {
1575 ext4_warning(inode->i_sb,
1576 "ES len assertion failed for inode "
1577 "%lu: retval %d != map->m_len %d",
1578 inode->i_ino, retval, map->m_len);
1579 WARN_ON(1);
Dmitry Monakhov921f2662013-03-10 21:01:03 -04001580 }
Dmitry Monakhov921f2662013-03-10 21:01:03 -04001581
Zheng Liuf7fec032013-02-18 00:28:47 -05001582 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1583 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1584 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1585 map->m_pblk, status);
1586 if (ret != 0)
1587 retval = ret;
Aditya Kali5356f262011-09-09 19:20:51 -04001588 }
1589
1590out_unlock:
1591 up_read((&EXT4_I(inode)->i_data_sem));
1592
1593 return retval;
1594}
1595
1596/*
Seunghun Leed91bd2c2014-09-01 22:15:30 -04001597 * This is a special get_block_t callback which is used by
Theodore Ts'ob920c752009-05-14 00:54:29 -04001598 * ext4_da_write_begin(). It will either return mapped block or
1599 * reserve space for a single block.
Aneesh Kumar K.V29fa89d2009-05-12 16:30:27 -04001600 *
1601 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1602 * We also have b_blocknr = -1 and b_bdev initialized properly
1603 *
1604 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1605 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1606 * initialized properly.
Alex Tomas64769242008-07-11 19:27:31 -04001607 */
Tao Ma9c3569b2012-12-10 14:05:57 -05001608int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1609 struct buffer_head *bh, int create)
Alex Tomas64769242008-07-11 19:27:31 -04001610{
Theodore Ts'o2ed88682010-05-16 20:00:00 -04001611 struct ext4_map_blocks map;
Alex Tomas64769242008-07-11 19:27:31 -04001612 int ret = 0;
1613
1614 BUG_ON(create == 0);
Theodore Ts'o2ed88682010-05-16 20:00:00 -04001615 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1616
1617 map.m_lblk = iblock;
1618 map.m_len = 1;
Alex Tomas64769242008-07-11 19:27:31 -04001619
1620 /*
1621 * first, we need to know whether the block is allocated already
1622 * preallocated blocks are unmapped but should treated
1623 * the same as allocated blocks.
1624 */
Aditya Kali5356f262011-09-09 19:20:51 -04001625 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1626 if (ret <= 0)
Theodore Ts'o2ed88682010-05-16 20:00:00 -04001627 return ret;
Alex Tomas64769242008-07-11 19:27:31 -04001628
Theodore Ts'o2ed88682010-05-16 20:00:00 -04001629 map_bh(bh, inode->i_sb, map.m_pblk);
1630 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
1631
1632 if (buffer_unwritten(bh)) {
1633 /* A delayed write to unwritten bh should be marked
1634 * new and mapped. Mapped ensures that we don't do
1635 * get_block multiple times when we write to the same
1636 * offset and new ensures that we do proper zero out
1637 * for partial write.
1638 */
1639 set_buffer_new(bh);
Theodore Ts'oc8205632011-04-10 22:30:07 -04001640 set_buffer_mapped(bh);
Theodore Ts'o2ed88682010-05-16 20:00:00 -04001641 }
1642 return 0;
Alex Tomas64769242008-07-11 19:27:31 -04001643}
Mingming Cao61628a32008-07-11 19:27:31 -04001644
Aneesh Kumar K.V62e086b2009-06-14 17:59:34 -04001645static int bget_one(handle_t *handle, struct buffer_head *bh)
1646{
1647 get_bh(bh);
1648 return 0;
1649}
1650
1651static int bput_one(handle_t *handle, struct buffer_head *bh)
1652{
1653 put_bh(bh);
1654 return 0;
1655}
1656
1657static int __ext4_journalled_writepage(struct page *page,
Aneesh Kumar K.V62e086b2009-06-14 17:59:34 -04001658 unsigned int len)
1659{
1660 struct address_space *mapping = page->mapping;
1661 struct inode *inode = mapping->host;
Tao Ma3fdcfb62012-12-10 14:05:57 -05001662 struct buffer_head *page_bufs = NULL;
Aneesh Kumar K.V62e086b2009-06-14 17:59:34 -04001663 handle_t *handle = NULL;
Tao Ma3fdcfb62012-12-10 14:05:57 -05001664 int ret = 0, err = 0;
1665 int inline_data = ext4_has_inline_data(inode);
1666 struct buffer_head *inode_bh = NULL;
Aneesh Kumar K.V62e086b2009-06-14 17:59:34 -04001667
Theodore Ts'ocb20d512010-10-27 21:30:09 -04001668 ClearPageChecked(page);
Tao Ma3fdcfb62012-12-10 14:05:57 -05001669
1670 if (inline_data) {
1671 BUG_ON(page->index != 0);
1672 BUG_ON(len > ext4_get_max_inline_size(inode));
1673 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
1674 if (inode_bh == NULL)
1675 goto out;
1676 } else {
1677 page_bufs = page_buffers(page);
1678 if (!page_bufs) {
1679 BUG();
1680 goto out;
1681 }
1682 ext4_walk_page_buffers(handle, page_bufs, 0, len,
1683 NULL, bget_one);
1684 }
Theodore Ts'obdf96832015-06-12 23:45:33 -04001685 /*
1686 * We need to release the page lock before we start the
1687 * journal, so grab a reference so the page won't disappear
1688 * out from under us.
1689 */
1690 get_page(page);
Aneesh Kumar K.V62e086b2009-06-14 17:59:34 -04001691 unlock_page(page);
1692
Theodore Ts'o9924a922013-02-08 21:59:22 -05001693 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
1694 ext4_writepage_trans_blocks(inode));
Aneesh Kumar K.V62e086b2009-06-14 17:59:34 -04001695 if (IS_ERR(handle)) {
1696 ret = PTR_ERR(handle);
Theodore Ts'obdf96832015-06-12 23:45:33 -04001697 put_page(page);
1698 goto out_no_pagelock;
1699 }
1700 BUG_ON(!ext4_handle_valid(handle));
1701
1702 lock_page(page);
1703 put_page(page);
1704 if (page->mapping != mapping) {
1705 /* The page got truncated from under us */
1706 ext4_journal_stop(handle);
1707 ret = 0;
Aneesh Kumar K.V62e086b2009-06-14 17:59:34 -04001708 goto out;
1709 }
1710
Tao Ma3fdcfb62012-12-10 14:05:57 -05001711 if (inline_data) {
liang xie5d601252014-05-12 22:06:43 -04001712 BUFFER_TRACE(inode_bh, "get write access");
Tao Ma3fdcfb62012-12-10 14:05:57 -05001713 ret = ext4_journal_get_write_access(handle, inode_bh);
Aneesh Kumar K.V62e086b2009-06-14 17:59:34 -04001714
Tao Ma3fdcfb62012-12-10 14:05:57 -05001715 err = ext4_handle_dirty_metadata(handle, inode, inode_bh);
1716
1717 } else {
1718 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1719 do_journal_get_write_access);
1720
1721 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1722 write_end_fn);
1723 }
Aneesh Kumar K.V62e086b2009-06-14 17:59:34 -04001724 if (ret == 0)
1725 ret = err;
Jan Kara2d859db2011-07-26 09:07:11 -04001726 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
Aneesh Kumar K.V62e086b2009-06-14 17:59:34 -04001727 err = ext4_journal_stop(handle);
1728 if (!ret)
1729 ret = err;
1730
Tao Ma3fdcfb62012-12-10 14:05:57 -05001731 if (!ext4_has_inline_data(inode))
Theodore Ts'o8c9367f2014-01-07 13:08:03 -05001732 ext4_walk_page_buffers(NULL, page_bufs, 0, len,
Tao Ma3fdcfb62012-12-10 14:05:57 -05001733 NULL, bput_one);
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05001734 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
Aneesh Kumar K.V62e086b2009-06-14 17:59:34 -04001735out:
Theodore Ts'obdf96832015-06-12 23:45:33 -04001736 unlock_page(page);
1737out_no_pagelock:
Tao Ma3fdcfb62012-12-10 14:05:57 -05001738 brelse(inode_bh);
Aneesh Kumar K.V62e086b2009-06-14 17:59:34 -04001739 return ret;
1740}
1741
Mingming Cao61628a32008-07-11 19:27:31 -04001742/*
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04001743 * Note that we don't need to start a transaction unless we're journaling data
1744 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1745 * need to file the inode to the transaction's list in ordered mode because if
1746 * we are writing back data added by write(), the inode is already there and if
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001747 * we are writing back data modified via mmap(), no one guarantees in which
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04001748 * transaction the data will hit the disk. In case we are journaling data, we
1749 * cannot start transaction directly because transaction start ranks above page
1750 * lock so we have to do some magic.
1751 *
Theodore Ts'ob920c752009-05-14 00:54:29 -04001752 * This function can get called via...
Theodore Ts'o20970ba2013-06-06 14:00:46 -04001753 * - ext4_writepages after taking page lock (have journal handle)
Theodore Ts'ob920c752009-05-14 00:54:29 -04001754 * - journal_submit_inode_data_buffers (no journal handle)
Artem Bityutskiyf6463b02012-07-25 18:12:04 +03001755 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
Theodore Ts'ob920c752009-05-14 00:54:29 -04001756 * - grab_page_cache when doing write_begin (have journal handle)
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04001757 *
1758 * We don't do any block allocation in this function. If we have page with
1759 * multiple blocks we need to write those buffer_heads that are mapped. This
1760 * is important for mmaped based write. So if we do with blocksize 1K
1761 * truncate(f, 1024);
1762 * a = mmap(f, 0, 4096);
1763 * a[0] = 'a';
1764 * truncate(f, 4096);
1765 * we have in the page first buffer_head mapped via page_mkwrite call back
Paul Bolle90802ed2011-12-05 13:00:34 +01001766 * but other buffer_heads would be unmapped but dirty (dirty done via the
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04001767 * do_wp_page). So writepage should write the first block. If we modify
1768 * the mmap area beyond 1024 we will again get a page_fault and the
1769 * page_mkwrite callback will do the block allocation and mark the
1770 * buffer_heads mapped.
1771 *
1772 * We redirty the page if we have any buffer_heads that is either delay or
1773 * unwritten in the page.
1774 *
1775 * We can get recursively called as show below.
1776 *
1777 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1778 * ext4_writepage()
1779 *
1780 * But since we don't do any block allocation we should not deadlock.
1781 * Page also have the dirty flag cleared so we don't get recurive page_lock.
Mingming Cao61628a32008-07-11 19:27:31 -04001782 */
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04001783static int ext4_writepage(struct page *page,
Aneesh Kumar K.V62e086b2009-06-14 17:59:34 -04001784 struct writeback_control *wbc)
Alex Tomas64769242008-07-11 19:27:31 -04001785{
Jan Karaf8bec372013-01-28 12:55:08 -05001786 int ret = 0;
Mingming Cao61628a32008-07-11 19:27:31 -04001787 loff_t size;
Theodore Ts'o498e5f22008-11-05 00:14:04 -05001788 unsigned int len;
Jiaying Zhang744692d2010-03-04 16:14:02 -05001789 struct buffer_head *page_bufs = NULL;
Mingming Cao61628a32008-07-11 19:27:31 -04001790 struct inode *inode = page->mapping->host;
Jan Kara36ade452013-01-28 09:30:52 -05001791 struct ext4_io_submit io_submit;
Namjae Jeon1c8349a2014-05-12 08:12:25 -04001792 bool keep_towrite = false;
Alex Tomas64769242008-07-11 19:27:31 -04001793
Lukas Czernera9c667f2011-06-06 09:51:52 -04001794 trace_ext4_writepage(page);
Aneesh Kumar K.Vf0e6c982008-07-11 19:27:31 -04001795 size = i_size_read(inode);
1796 if (page->index == size >> PAGE_CACHE_SHIFT)
1797 len = size & ~PAGE_CACHE_MASK;
1798 else
1799 len = PAGE_CACHE_SIZE;
Alex Tomas64769242008-07-11 19:27:31 -04001800
Theodore Ts'oa42afc52010-10-27 21:30:09 -04001801 page_bufs = page_buffers(page);
Aneesh Kumar K.Vc364b222009-06-14 17:57:10 -04001802 /*
Jan Karafe386132013-01-28 21:06:42 -05001803 * We cannot do block allocation or other extent handling in this
1804 * function. If there are buffers needing that, we have to redirty
1805 * the page. But we may reach here when we do a journal commit via
1806 * journal_submit_inode_data_buffers() and in that case we must write
1807 * allocated buffers to achieve data=ordered mode guarantees.
Aneesh Kumar K.Vcd1aac32008-07-11 19:27:31 -04001808 */
Tao Maf19d5872012-12-10 14:05:51 -05001809 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
1810 ext4_bh_delay_or_unwritten)) {
Jan Karaf8bec372013-01-28 12:55:08 -05001811 redirty_page_for_writepage(wbc, page);
Jan Karafe386132013-01-28 21:06:42 -05001812 if (current->flags & PF_MEMALLOC) {
1813 /*
1814 * For memory cleaning there's no point in writing only
1815 * some buffers. So just bail out. Warn if we came here
1816 * from direct reclaim.
1817 */
1818 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
1819 == PF_MEMALLOC);
Aneesh Kumar K.Vf0e6c982008-07-11 19:27:31 -04001820 unlock_page(page);
1821 return 0;
1822 }
Namjae Jeon1c8349a2014-05-12 08:12:25 -04001823 keep_towrite = true;
Theodore Ts'oa42afc52010-10-27 21:30:09 -04001824 }
Alex Tomas64769242008-07-11 19:27:31 -04001825
Theodore Ts'ocb20d512010-10-27 21:30:09 -04001826 if (PageChecked(page) && ext4_should_journal_data(inode))
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04001827 /*
1828 * It's mmapped pagecache. Add buffers and journal it. There
1829 * doesn't seem much point in redirtying the page here.
1830 */
Wu Fengguang3f0ca302009-11-24 11:15:44 -05001831 return __ext4_journalled_writepage(page, len);
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04001832
Jan Kara97a851e2013-06-04 11:58:58 -04001833 ext4_io_submit_init(&io_submit, wbc);
1834 io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
1835 if (!io_submit.io_end) {
1836 redirty_page_for_writepage(wbc, page);
1837 unlock_page(page);
1838 return -ENOMEM;
1839 }
Namjae Jeon1c8349a2014-05-12 08:12:25 -04001840 ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite);
Jan Kara36ade452013-01-28 09:30:52 -05001841 ext4_io_submit(&io_submit);
Jan Kara97a851e2013-06-04 11:58:58 -04001842 /* Drop io_end reference we got from init */
1843 ext4_put_io_end_defer(io_submit.io_end);
Alex Tomas64769242008-07-11 19:27:31 -04001844 return ret;
1845}
1846
Jan Kara5f1132b2013-08-17 10:02:33 -04001847static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page)
1848{
1849 int len;
1850 loff_t size = i_size_read(mpd->inode);
1851 int err;
1852
1853 BUG_ON(page->index != mpd->first_page);
1854 if (page->index == size >> PAGE_CACHE_SHIFT)
1855 len = size & ~PAGE_CACHE_MASK;
1856 else
1857 len = PAGE_CACHE_SIZE;
1858 clear_page_dirty_for_io(page);
Namjae Jeon1c8349a2014-05-12 08:12:25 -04001859 err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false);
Jan Kara5f1132b2013-08-17 10:02:33 -04001860 if (!err)
1861 mpd->wbc->nr_to_write--;
1862 mpd->first_page++;
1863
1864 return err;
1865}
1866
Jan Kara4e7ea812013-06-04 13:17:40 -04001867#define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
1868
Mingming Cao61628a32008-07-11 19:27:31 -04001869/*
Jan Karafffb2732013-06-04 13:01:11 -04001870 * mballoc gives us at most this number of blocks...
1871 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
Anatol Pomozov70261f52013-08-28 14:40:12 -04001872 * The rest of mballoc seems to handle chunks up to full group size.
Mingming Cao61628a32008-07-11 19:27:31 -04001873 */
Jan Karafffb2732013-06-04 13:01:11 -04001874#define MAX_WRITEPAGES_EXTENT_LEN 2048
Mingming Cao525f4ed2008-08-19 22:15:58 -04001875
Jan Karafffb2732013-06-04 13:01:11 -04001876/*
Jan Kara4e7ea812013-06-04 13:17:40 -04001877 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
1878 *
1879 * @mpd - extent of blocks
1880 * @lblk - logical number of the block in the file
Jan Kara09930042013-08-17 09:57:56 -04001881 * @bh - buffer head we want to add to the extent
Jan Kara4e7ea812013-06-04 13:17:40 -04001882 *
Jan Kara09930042013-08-17 09:57:56 -04001883 * The function is used to collect contig. blocks in the same state. If the
1884 * buffer doesn't require mapping for writeback and we haven't started the
1885 * extent of buffers to map yet, the function returns 'true' immediately - the
1886 * caller can write the buffer right away. Otherwise the function returns true
1887 * if the block has been added to the extent, false if the block couldn't be
1888 * added.
Jan Kara4e7ea812013-06-04 13:17:40 -04001889 */
Jan Kara09930042013-08-17 09:57:56 -04001890static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
1891 struct buffer_head *bh)
Jan Kara4e7ea812013-06-04 13:17:40 -04001892{
1893 struct ext4_map_blocks *map = &mpd->map;
1894
Jan Kara09930042013-08-17 09:57:56 -04001895 /* Buffer that doesn't need mapping for writeback? */
1896 if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
1897 (!buffer_delay(bh) && !buffer_unwritten(bh))) {
1898 /* So far no extent to map => we write the buffer right away */
1899 if (map->m_len == 0)
1900 return true;
1901 return false;
1902 }
Jan Kara4e7ea812013-06-04 13:17:40 -04001903
1904 /* First block in the extent? */
1905 if (map->m_len == 0) {
1906 map->m_lblk = lblk;
1907 map->m_len = 1;
Jan Kara09930042013-08-17 09:57:56 -04001908 map->m_flags = bh->b_state & BH_FLAGS;
1909 return true;
Jan Kara4e7ea812013-06-04 13:17:40 -04001910 }
1911
Jan Kara09930042013-08-17 09:57:56 -04001912 /* Don't go larger than mballoc is willing to allocate */
1913 if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
1914 return false;
1915
Jan Kara4e7ea812013-06-04 13:17:40 -04001916 /* Can we merge the block to our big extent? */
1917 if (lblk == map->m_lblk + map->m_len &&
Jan Kara09930042013-08-17 09:57:56 -04001918 (bh->b_state & BH_FLAGS) == map->m_flags) {
Jan Kara4e7ea812013-06-04 13:17:40 -04001919 map->m_len++;
Jan Kara09930042013-08-17 09:57:56 -04001920 return true;
Jan Kara4e7ea812013-06-04 13:17:40 -04001921 }
Jan Kara09930042013-08-17 09:57:56 -04001922 return false;
Jan Kara4e7ea812013-06-04 13:17:40 -04001923}
1924
Jan Kara5f1132b2013-08-17 10:02:33 -04001925/*
1926 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
1927 *
1928 * @mpd - extent of blocks for mapping
1929 * @head - the first buffer in the page
1930 * @bh - buffer we should start processing from
1931 * @lblk - logical number of the block in the file corresponding to @bh
1932 *
1933 * Walk through page buffers from @bh upto @head (exclusive) and either submit
1934 * the page for IO if all buffers in this page were mapped and there's no
1935 * accumulated extent of buffers to map or add buffers in the page to the
1936 * extent of buffers to map. The function returns 1 if the caller can continue
1937 * by processing the next page, 0 if it should stop adding buffers to the
1938 * extent to map because we cannot extend it anymore. It can also return value
1939 * < 0 in case of error during IO submission.
1940 */
1941static int mpage_process_page_bufs(struct mpage_da_data *mpd,
1942 struct buffer_head *head,
1943 struct buffer_head *bh,
1944 ext4_lblk_t lblk)
Jan Kara4e7ea812013-06-04 13:17:40 -04001945{
1946 struct inode *inode = mpd->inode;
Jan Kara5f1132b2013-08-17 10:02:33 -04001947 int err;
Jan Kara4e7ea812013-06-04 13:17:40 -04001948 ext4_lblk_t blocks = (i_size_read(inode) + (1 << inode->i_blkbits) - 1)
1949 >> inode->i_blkbits;
1950
1951 do {
1952 BUG_ON(buffer_locked(bh));
1953
Jan Kara09930042013-08-17 09:57:56 -04001954 if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
Jan Kara4e7ea812013-06-04 13:17:40 -04001955 /* Found extent to map? */
1956 if (mpd->map.m_len)
Jan Kara5f1132b2013-08-17 10:02:33 -04001957 return 0;
Jan Kara09930042013-08-17 09:57:56 -04001958 /* Everything mapped so far and we hit EOF */
Jan Kara5f1132b2013-08-17 10:02:33 -04001959 break;
Jan Kara4e7ea812013-06-04 13:17:40 -04001960 }
Jan Kara4e7ea812013-06-04 13:17:40 -04001961 } while (lblk++, (bh = bh->b_this_page) != head);
Jan Kara5f1132b2013-08-17 10:02:33 -04001962 /* So far everything mapped? Submit the page for IO. */
1963 if (mpd->map.m_len == 0) {
1964 err = mpage_submit_page(mpd, head->b_page);
1965 if (err < 0)
1966 return err;
1967 }
1968 return lblk < blocks;
Jan Kara4e7ea812013-06-04 13:17:40 -04001969}
1970
1971/*
1972 * mpage_map_buffers - update buffers corresponding to changed extent and
1973 * submit fully mapped pages for IO
1974 *
1975 * @mpd - description of extent to map, on return next extent to map
1976 *
1977 * Scan buffers corresponding to changed extent (we expect corresponding pages
1978 * to be already locked) and update buffer state according to new extent state.
1979 * We map delalloc buffers to their physical location, clear unwritten bits,
Lukas Czerner556615d2014-04-20 23:45:47 -04001980 * and mark buffers as uninit when we perform writes to unwritten extents
Jan Kara4e7ea812013-06-04 13:17:40 -04001981 * and do extent conversion after IO is finished. If the last page is not fully
1982 * mapped, we update @map to the next extent in the last page that needs
1983 * mapping. Otherwise we submit the page for IO.
1984 */
1985static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
1986{
1987 struct pagevec pvec;
1988 int nr_pages, i;
1989 struct inode *inode = mpd->inode;
1990 struct buffer_head *head, *bh;
1991 int bpp_bits = PAGE_CACHE_SHIFT - inode->i_blkbits;
Jan Kara4e7ea812013-06-04 13:17:40 -04001992 pgoff_t start, end;
1993 ext4_lblk_t lblk;
1994 sector_t pblock;
1995 int err;
1996
1997 start = mpd->map.m_lblk >> bpp_bits;
1998 end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
1999 lblk = start << bpp_bits;
2000 pblock = mpd->map.m_pblk;
2001
2002 pagevec_init(&pvec, 0);
2003 while (start <= end) {
2004 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, start,
2005 PAGEVEC_SIZE);
2006 if (nr_pages == 0)
2007 break;
2008 for (i = 0; i < nr_pages; i++) {
2009 struct page *page = pvec.pages[i];
2010
2011 if (page->index > end)
2012 break;
Anatol Pomozov70261f52013-08-28 14:40:12 -04002013 /* Up to 'end' pages must be contiguous */
Jan Kara4e7ea812013-06-04 13:17:40 -04002014 BUG_ON(page->index != start);
2015 bh = head = page_buffers(page);
2016 do {
2017 if (lblk < mpd->map.m_lblk)
2018 continue;
2019 if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2020 /*
2021 * Buffer after end of mapped extent.
2022 * Find next buffer in the page to map.
2023 */
2024 mpd->map.m_len = 0;
2025 mpd->map.m_flags = 0;
Jan Kara5f1132b2013-08-17 10:02:33 -04002026 /*
2027 * FIXME: If dioread_nolock supports
2028 * blocksize < pagesize, we need to make
2029 * sure we add size mapped so far to
2030 * io_end->size as the following call
2031 * can submit the page for IO.
2032 */
2033 err = mpage_process_page_bufs(mpd, head,
2034 bh, lblk);
Jan Kara4e7ea812013-06-04 13:17:40 -04002035 pagevec_release(&pvec);
Jan Kara5f1132b2013-08-17 10:02:33 -04002036 if (err > 0)
2037 err = 0;
2038 return err;
Jan Kara4e7ea812013-06-04 13:17:40 -04002039 }
2040 if (buffer_delay(bh)) {
2041 clear_buffer_delay(bh);
2042 bh->b_blocknr = pblock++;
2043 }
Jan Kara4e7ea812013-06-04 13:17:40 -04002044 clear_buffer_unwritten(bh);
Jan Kara5f1132b2013-08-17 10:02:33 -04002045 } while (lblk++, (bh = bh->b_this_page) != head);
Jan Kara4e7ea812013-06-04 13:17:40 -04002046
2047 /*
2048 * FIXME: This is going to break if dioread_nolock
2049 * supports blocksize < pagesize as we will try to
2050 * convert potentially unmapped parts of inode.
2051 */
2052 mpd->io_submit.io_end->size += PAGE_CACHE_SIZE;
2053 /* Page fully mapped - let IO run! */
2054 err = mpage_submit_page(mpd, page);
2055 if (err < 0) {
2056 pagevec_release(&pvec);
2057 return err;
2058 }
2059 start++;
2060 }
2061 pagevec_release(&pvec);
2062 }
2063 /* Extent fully mapped and matches with page boundary. We are done. */
2064 mpd->map.m_len = 0;
2065 mpd->map.m_flags = 0;
2066 return 0;
2067}
2068
2069static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2070{
2071 struct inode *inode = mpd->inode;
2072 struct ext4_map_blocks *map = &mpd->map;
2073 int get_blocks_flags;
Lukas Czerner090f32e2014-04-20 23:44:47 -04002074 int err, dioread_nolock;
Jan Kara4e7ea812013-06-04 13:17:40 -04002075
2076 trace_ext4_da_write_pages_extent(inode, map);
2077 /*
2078 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
Lukas Czerner556615d2014-04-20 23:45:47 -04002079 * to convert an unwritten extent to be initialized (in the case
Jan Kara4e7ea812013-06-04 13:17:40 -04002080 * where we have written into one or more preallocated blocks). It is
2081 * possible that we're going to need more metadata blocks than
2082 * previously reserved. However we must not fail because we're in
2083 * writeback and there is nothing we can do about it so it might result
2084 * in data loss. So use reserved blocks to allocate metadata if
2085 * possible.
2086 *
Theodore Ts'o754cfed2014-09-04 18:08:22 -04002087 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2088 * the blocks in question are delalloc blocks. This indicates
2089 * that the blocks and quotas has already been checked when
2090 * the data was copied into the page cache.
Jan Kara4e7ea812013-06-04 13:17:40 -04002091 */
2092 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2093 EXT4_GET_BLOCKS_METADATA_NOFAIL;
Lukas Czerner090f32e2014-04-20 23:44:47 -04002094 dioread_nolock = ext4_should_dioread_nolock(inode);
2095 if (dioread_nolock)
Jan Kara4e7ea812013-06-04 13:17:40 -04002096 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2097 if (map->m_flags & (1 << BH_Delay))
2098 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2099
2100 err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2101 if (err < 0)
2102 return err;
Lukas Czerner090f32e2014-04-20 23:44:47 -04002103 if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
Jan Kara6b523df2013-06-04 13:21:11 -04002104 if (!mpd->io_submit.io_end->handle &&
2105 ext4_handle_valid(handle)) {
2106 mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2107 handle->h_rsv_handle = NULL;
2108 }
Jan Kara3613d222013-06-04 13:19:34 -04002109 ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
Jan Kara6b523df2013-06-04 13:21:11 -04002110 }
Jan Kara4e7ea812013-06-04 13:17:40 -04002111
2112 BUG_ON(map->m_len == 0);
2113 if (map->m_flags & EXT4_MAP_NEW) {
2114 struct block_device *bdev = inode->i_sb->s_bdev;
2115 int i;
2116
2117 for (i = 0; i < map->m_len; i++)
2118 unmap_underlying_metadata(bdev, map->m_pblk + i);
2119 }
2120 return 0;
2121}
2122
2123/*
2124 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2125 * mpd->len and submit pages underlying it for IO
2126 *
2127 * @handle - handle for journal operations
2128 * @mpd - extent to map
Jan Kara7534e852013-10-16 08:26:08 -04002129 * @give_up_on_write - we set this to true iff there is a fatal error and there
2130 * is no hope of writing the data. The caller should discard
2131 * dirty pages to avoid infinite loops.
Jan Kara4e7ea812013-06-04 13:17:40 -04002132 *
2133 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2134 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2135 * them to initialized or split the described range from larger unwritten
2136 * extent. Note that we need not map all the described range since allocation
2137 * can return less blocks or the range is covered by more unwritten extents. We
2138 * cannot map more because we are limited by reserved transaction credits. On
2139 * the other hand we always make sure that the last touched page is fully
2140 * mapped so that it can be written out (and thus forward progress is
2141 * guaranteed). After mapping we submit all mapped pages for IO.
2142 */
2143static int mpage_map_and_submit_extent(handle_t *handle,
Theodore Ts'ocb530542013-07-01 08:12:40 -04002144 struct mpage_da_data *mpd,
2145 bool *give_up_on_write)
Jan Kara4e7ea812013-06-04 13:17:40 -04002146{
2147 struct inode *inode = mpd->inode;
2148 struct ext4_map_blocks *map = &mpd->map;
2149 int err;
2150 loff_t disksize;
Dmitry Monakhov66031202014-08-27 18:40:03 -04002151 int progress = 0;
Jan Kara4e7ea812013-06-04 13:17:40 -04002152
2153 mpd->io_submit.io_end->offset =
2154 ((loff_t)map->m_lblk) << inode->i_blkbits;
Jan Kara27d7c4e2013-07-05 21:57:22 -04002155 do {
Jan Kara4e7ea812013-06-04 13:17:40 -04002156 err = mpage_map_one_extent(handle, mpd);
2157 if (err < 0) {
2158 struct super_block *sb = inode->i_sb;
2159
Theodore Ts'ocb530542013-07-01 08:12:40 -04002160 if (EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)
2161 goto invalidate_dirty_pages;
Jan Kara4e7ea812013-06-04 13:17:40 -04002162 /*
Theodore Ts'ocb530542013-07-01 08:12:40 -04002163 * Let the uper layers retry transient errors.
2164 * In the case of ENOSPC, if ext4_count_free_blocks()
2165 * is non-zero, a commit should free up blocks.
Jan Kara4e7ea812013-06-04 13:17:40 -04002166 */
Theodore Ts'ocb530542013-07-01 08:12:40 -04002167 if ((err == -ENOMEM) ||
Dmitry Monakhov66031202014-08-27 18:40:03 -04002168 (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2169 if (progress)
2170 goto update_disksize;
Theodore Ts'ocb530542013-07-01 08:12:40 -04002171 return err;
Dmitry Monakhov66031202014-08-27 18:40:03 -04002172 }
Theodore Ts'ocb530542013-07-01 08:12:40 -04002173 ext4_msg(sb, KERN_CRIT,
2174 "Delayed block allocation failed for "
2175 "inode %lu at logical offset %llu with"
2176 " max blocks %u with error %d",
2177 inode->i_ino,
2178 (unsigned long long)map->m_lblk,
2179 (unsigned)map->m_len, -err);
2180 ext4_msg(sb, KERN_CRIT,
2181 "This should not happen!! Data will "
2182 "be lost\n");
2183 if (err == -ENOSPC)
2184 ext4_print_free_blocks(inode);
2185 invalidate_dirty_pages:
2186 *give_up_on_write = true;
Jan Kara4e7ea812013-06-04 13:17:40 -04002187 return err;
2188 }
Dmitry Monakhov66031202014-08-27 18:40:03 -04002189 progress = 1;
Jan Kara4e7ea812013-06-04 13:17:40 -04002190 /*
2191 * Update buffer state, submit mapped pages, and get us new
2192 * extent to map
2193 */
2194 err = mpage_map_and_submit_buffers(mpd);
2195 if (err < 0)
Dmitry Monakhov66031202014-08-27 18:40:03 -04002196 goto update_disksize;
Jan Kara27d7c4e2013-07-05 21:57:22 -04002197 } while (map->m_len);
Jan Kara4e7ea812013-06-04 13:17:40 -04002198
Dmitry Monakhov66031202014-08-27 18:40:03 -04002199update_disksize:
Theodore Ts'o622cad12014-04-11 10:35:17 -04002200 /*
2201 * Update on-disk size after IO is submitted. Races with
2202 * truncate are avoided by checking i_size under i_data_sem.
2203 */
Jan Kara4e7ea812013-06-04 13:17:40 -04002204 disksize = ((loff_t)mpd->first_page) << PAGE_CACHE_SHIFT;
Jan Kara4e7ea812013-06-04 13:17:40 -04002205 if (disksize > EXT4_I(inode)->i_disksize) {
2206 int err2;
Theodore Ts'o622cad12014-04-11 10:35:17 -04002207 loff_t i_size;
Jan Kara4e7ea812013-06-04 13:17:40 -04002208
Theodore Ts'o622cad12014-04-11 10:35:17 -04002209 down_write(&EXT4_I(inode)->i_data_sem);
2210 i_size = i_size_read(inode);
2211 if (disksize > i_size)
2212 disksize = i_size;
2213 if (disksize > EXT4_I(inode)->i_disksize)
2214 EXT4_I(inode)->i_disksize = disksize;
Jan Kara4e7ea812013-06-04 13:17:40 -04002215 err2 = ext4_mark_inode_dirty(handle, inode);
Theodore Ts'o622cad12014-04-11 10:35:17 -04002216 up_write(&EXT4_I(inode)->i_data_sem);
Jan Kara4e7ea812013-06-04 13:17:40 -04002217 if (err2)
2218 ext4_error(inode->i_sb,
2219 "Failed to mark inode %lu dirty",
2220 inode->i_ino);
2221 if (!err)
2222 err = err2;
2223 }
2224 return err;
2225}
2226
2227/*
Jan Karafffb2732013-06-04 13:01:11 -04002228 * Calculate the total number of credits to reserve for one writepages
Theodore Ts'o20970ba2013-06-06 14:00:46 -04002229 * iteration. This is called from ext4_writepages(). We map an extent of
Anatol Pomozov70261f52013-08-28 14:40:12 -04002230 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
Jan Karafffb2732013-06-04 13:01:11 -04002231 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2232 * bpp - 1 blocks in bpp different extents.
2233 */
Mingming Cao525f4ed2008-08-19 22:15:58 -04002234static int ext4_da_writepages_trans_blocks(struct inode *inode)
2235{
Jan Karafffb2732013-06-04 13:01:11 -04002236 int bpp = ext4_journal_blocks_per_page(inode);
Mingming Cao525f4ed2008-08-19 22:15:58 -04002237
Jan Karafffb2732013-06-04 13:01:11 -04002238 return ext4_meta_trans_blocks(inode,
2239 MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
Mingming Cao525f4ed2008-08-19 22:15:58 -04002240}
Mingming Cao61628a32008-07-11 19:27:31 -04002241
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002242/*
Jan Kara4e7ea812013-06-04 13:17:40 -04002243 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2244 * and underlying extent to map
2245 *
2246 * @mpd - where to look for pages
2247 *
2248 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2249 * IO immediately. When we find a page which isn't mapped we start accumulating
2250 * extent of buffers underlying these pages that needs mapping (formed by
2251 * either delayed or unwritten buffers). We also lock the pages containing
2252 * these buffers. The extent found is returned in @mpd structure (starting at
2253 * mpd->lblk with length mpd->len blocks).
2254 *
2255 * Note that this function can attach bios to one io_end structure which are
2256 * neither logically nor physically contiguous. Although it may seem as an
2257 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2258 * case as we need to track IO to all buffers underlying a page in one io_end.
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002259 */
Jan Kara4e7ea812013-06-04 13:17:40 -04002260static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002261{
Jan Kara4e7ea812013-06-04 13:17:40 -04002262 struct address_space *mapping = mpd->inode->i_mapping;
2263 struct pagevec pvec;
2264 unsigned int nr_pages;
Ming Leiaeac5892013-10-17 18:56:16 -04002265 long left = mpd->wbc->nr_to_write;
Jan Kara4e7ea812013-06-04 13:17:40 -04002266 pgoff_t index = mpd->first_page;
2267 pgoff_t end = mpd->last_page;
2268 int tag;
2269 int i, err = 0;
2270 int blkbits = mpd->inode->i_blkbits;
2271 ext4_lblk_t lblk;
2272 struct buffer_head *head;
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002273
Jan Kara4e7ea812013-06-04 13:17:40 -04002274 if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
Eric Sandeen5b41d922010-10-27 21:30:13 -04002275 tag = PAGECACHE_TAG_TOWRITE;
2276 else
2277 tag = PAGECACHE_TAG_DIRTY;
2278
Jan Kara4e7ea812013-06-04 13:17:40 -04002279 pagevec_init(&pvec, 0);
2280 mpd->map.m_len = 0;
2281 mpd->next_page = index;
Theodore Ts'o4f01b022011-02-26 14:07:37 -05002282 while (index <= end) {
Eric Sandeen5b41d922010-10-27 21:30:13 -04002283 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002284 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2285 if (nr_pages == 0)
Jan Kara4e7ea812013-06-04 13:17:40 -04002286 goto out;
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002287
2288 for (i = 0; i < nr_pages; i++) {
2289 struct page *page = pvec.pages[i];
2290
2291 /*
2292 * At this point, the page may be truncated or
2293 * invalidated (changing page->mapping to NULL), or
2294 * even swizzled back from swapper_space to tmpfs file
2295 * mapping. However, page->index will not change
2296 * because we have a reference on the page.
2297 */
Theodore Ts'o4f01b022011-02-26 14:07:37 -05002298 if (page->index > end)
2299 goto out;
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002300
Ming Leiaeac5892013-10-17 18:56:16 -04002301 /*
2302 * Accumulated enough dirty pages? This doesn't apply
2303 * to WB_SYNC_ALL mode. For integrity sync we have to
2304 * keep going because someone may be concurrently
2305 * dirtying pages, and we might have synced a lot of
2306 * newly appeared dirty pages, but have not synced all
2307 * of the old dirty pages.
2308 */
2309 if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0)
2310 goto out;
2311
Jan Kara4e7ea812013-06-04 13:17:40 -04002312 /* If we can't merge this page, we are done. */
2313 if (mpd->map.m_len > 0 && mpd->next_page != page->index)
2314 goto out;
Theodore Ts'o78aaced2011-02-26 14:09:14 -05002315
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002316 lock_page(page);
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002317 /*
Jan Kara4e7ea812013-06-04 13:17:40 -04002318 * If the page is no longer dirty, or its mapping no
2319 * longer corresponds to inode we are writing (which
2320 * means it has been truncated or invalidated), or the
2321 * page is already under writeback and we are not doing
2322 * a data integrity writeback, skip the page
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002323 */
Theodore Ts'o4f01b022011-02-26 14:07:37 -05002324 if (!PageDirty(page) ||
2325 (PageWriteback(page) &&
Jan Kara4e7ea812013-06-04 13:17:40 -04002326 (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
Theodore Ts'o4f01b022011-02-26 14:07:37 -05002327 unlikely(page->mapping != mapping)) {
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002328 unlock_page(page);
2329 continue;
2330 }
2331
Darrick J. Wong7cb1a532011-05-18 13:53:20 -04002332 wait_on_page_writeback(page);
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002333 BUG_ON(PageWriteback(page));
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002334
Jan Kara4e7ea812013-06-04 13:17:40 -04002335 if (mpd->map.m_len == 0)
Theodore Ts'o8eb9e5c2011-02-26 14:07:31 -05002336 mpd->first_page = page->index;
Theodore Ts'o8eb9e5c2011-02-26 14:07:31 -05002337 mpd->next_page = page->index + 1;
Jan Karaf8bec372013-01-28 12:55:08 -05002338 /* Add all dirty buffers to mpd */
Jan Kara4e7ea812013-06-04 13:17:40 -04002339 lblk = ((ext4_lblk_t)page->index) <<
2340 (PAGE_CACHE_SHIFT - blkbits);
Jan Karaf8bec372013-01-28 12:55:08 -05002341 head = page_buffers(page);
Jan Kara5f1132b2013-08-17 10:02:33 -04002342 err = mpage_process_page_bufs(mpd, head, head, lblk);
2343 if (err <= 0)
Jan Kara4e7ea812013-06-04 13:17:40 -04002344 goto out;
Jan Kara5f1132b2013-08-17 10:02:33 -04002345 err = 0;
Ming Leiaeac5892013-10-17 18:56:16 -04002346 left--;
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002347 }
2348 pagevec_release(&pvec);
2349 cond_resched();
2350 }
Theodore Ts'o4f01b022011-02-26 14:07:37 -05002351 return 0;
Theodore Ts'o8eb9e5c2011-02-26 14:07:31 -05002352out:
2353 pagevec_release(&pvec);
Jan Kara4e7ea812013-06-04 13:17:40 -04002354 return err;
Theodore Ts'o8e48dcf2010-05-16 18:00:00 -04002355}
2356
Theodore Ts'o20970ba2013-06-06 14:00:46 -04002357static int __writepage(struct page *page, struct writeback_control *wbc,
2358 void *data)
2359{
2360 struct address_space *mapping = data;
2361 int ret = ext4_writepage(page, wbc);
2362 mapping_set_error(mapping, ret);
2363 return ret;
2364}
2365
2366static int ext4_writepages(struct address_space *mapping,
2367 struct writeback_control *wbc)
Alex Tomas64769242008-07-11 19:27:31 -04002368{
Jan Kara4e7ea812013-06-04 13:17:40 -04002369 pgoff_t writeback_index = 0;
2370 long nr_to_write = wbc->nr_to_write;
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002371 int range_whole = 0;
Jan Kara4e7ea812013-06-04 13:17:40 -04002372 int cycled = 1;
Mingming Cao61628a32008-07-11 19:27:31 -04002373 handle_t *handle = NULL;
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04002374 struct mpage_da_data mpd;
Aneesh Kumar K.V5e745b02008-08-18 18:00:57 -04002375 struct inode *inode = mapping->host;
Jan Kara6b523df2013-06-04 13:21:11 -04002376 int needed_blocks, rsv_blocks = 0, ret = 0;
Aneesh Kumar K.V5e745b02008-08-18 18:00:57 -04002377 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
Jan Kara4e7ea812013-06-04 13:17:40 -04002378 bool done;
Shaohua Li1bce63d2011-10-18 10:55:51 -04002379 struct blk_plug plug;
Theodore Ts'ocb530542013-07-01 08:12:40 -04002380 bool give_up_on_write = false;
Mingming Cao61628a32008-07-11 19:27:31 -04002381
Theodore Ts'o20970ba2013-06-06 14:00:46 -04002382 trace_ext4_writepages(inode, wbc);
Theodore Ts'oba80b102009-01-03 20:03:21 -05002383
Mingming Cao61628a32008-07-11 19:27:31 -04002384 /*
2385 * No pages to write? This is mainly a kludge to avoid starting
2386 * a transaction for special inodes like journal inode on last iput()
2387 * because that could violate lock ordering on umount
2388 */
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002389 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
Ming Leibbf023c2013-10-30 07:27:16 -04002390 goto out_writepages;
Theodore Ts'o2a21e372008-11-05 09:22:24 -05002391
Theodore Ts'o20970ba2013-06-06 14:00:46 -04002392 if (ext4_should_journal_data(inode)) {
2393 struct blk_plug plug;
Theodore Ts'o20970ba2013-06-06 14:00:46 -04002394
2395 blk_start_plug(&plug);
2396 ret = write_cache_pages(mapping, wbc, __writepage, mapping);
2397 blk_finish_plug(&plug);
Ming Leibbf023c2013-10-30 07:27:16 -04002398 goto out_writepages;
Theodore Ts'o20970ba2013-06-06 14:00:46 -04002399 }
2400
Theodore Ts'o2a21e372008-11-05 09:22:24 -05002401 /*
2402 * If the filesystem has aborted, it is read-only, so return
2403 * right away instead of dumping stack traces later on that
2404 * will obscure the real source of the problem. We test
Theodore Ts'o4ab2f152009-06-13 10:09:36 -04002405 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
Theodore Ts'o2a21e372008-11-05 09:22:24 -05002406 * the latter could be true if the filesystem is mounted
Theodore Ts'o20970ba2013-06-06 14:00:46 -04002407 * read-only, and in that case, ext4_writepages should
Theodore Ts'o2a21e372008-11-05 09:22:24 -05002408 * *never* be called, so if that ever happens, we would want
2409 * the stack trace.
2410 */
Ming Leibbf023c2013-10-30 07:27:16 -04002411 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) {
2412 ret = -EROFS;
2413 goto out_writepages;
2414 }
Theodore Ts'o2a21e372008-11-05 09:22:24 -05002415
Jan Kara6b523df2013-06-04 13:21:11 -04002416 if (ext4_should_dioread_nolock(inode)) {
2417 /*
Anatol Pomozov70261f52013-08-28 14:40:12 -04002418 * We may need to convert up to one extent per block in
Jan Kara6b523df2013-06-04 13:21:11 -04002419 * the page and we may dirty the inode.
2420 */
2421 rsv_blocks = 1 + (PAGE_CACHE_SIZE >> inode->i_blkbits);
2422 }
2423
Jan Kara4e7ea812013-06-04 13:17:40 -04002424 /*
2425 * If we have inline data and arrive here, it means that
2426 * we will soon create the block for the 1st page, so
2427 * we'd better clear the inline data here.
2428 */
2429 if (ext4_has_inline_data(inode)) {
2430 /* Just inode will be modified... */
2431 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2432 if (IS_ERR(handle)) {
2433 ret = PTR_ERR(handle);
2434 goto out_writepages;
2435 }
2436 BUG_ON(ext4_test_inode_state(inode,
2437 EXT4_STATE_MAY_INLINE_DATA));
2438 ext4_destroy_inline_data(handle, inode);
2439 ext4_journal_stop(handle);
2440 }
2441
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002442 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2443 range_whole = 1;
Mingming Cao61628a32008-07-11 19:27:31 -04002444
Aneesh Kumar K.V2acf2c22009-02-14 10:42:58 -05002445 if (wbc->range_cyclic) {
Jan Kara4e7ea812013-06-04 13:17:40 -04002446 writeback_index = mapping->writeback_index;
2447 if (writeback_index)
Aneesh Kumar K.V2acf2c22009-02-14 10:42:58 -05002448 cycled = 0;
Jan Kara4e7ea812013-06-04 13:17:40 -04002449 mpd.first_page = writeback_index;
2450 mpd.last_page = -1;
Eric Sandeen5b41d922010-10-27 21:30:13 -04002451 } else {
Jan Kara4e7ea812013-06-04 13:17:40 -04002452 mpd.first_page = wbc->range_start >> PAGE_CACHE_SHIFT;
2453 mpd.last_page = wbc->range_end >> PAGE_CACHE_SHIFT;
Eric Sandeen5b41d922010-10-27 21:30:13 -04002454 }
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002455
Jan Kara4e7ea812013-06-04 13:17:40 -04002456 mpd.inode = inode;
2457 mpd.wbc = wbc;
2458 ext4_io_submit_init(&mpd.io_submit, wbc);
Aneesh Kumar K.V2acf2c22009-02-14 10:42:58 -05002459retry:
Wu Fengguang6e6938b2010-06-06 10:38:15 -06002460 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
Jan Kara4e7ea812013-06-04 13:17:40 -04002461 tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page);
2462 done = false;
Shaohua Li1bce63d2011-10-18 10:55:51 -04002463 blk_start_plug(&plug);
Jan Kara4e7ea812013-06-04 13:17:40 -04002464 while (!done && mpd.first_page <= mpd.last_page) {
2465 /* For each extent of pages we use new io_end */
2466 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2467 if (!mpd.io_submit.io_end) {
2468 ret = -ENOMEM;
2469 break;
2470 }
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002471
2472 /*
Jan Kara4e7ea812013-06-04 13:17:40 -04002473 * We have two constraints: We find one extent to map and we
2474 * must always write out whole page (makes a difference when
2475 * blocksize < pagesize) so that we don't block on IO when we
2476 * try to write out the rest of the page. Journalled mode is
2477 * not supported by delalloc.
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002478 */
2479 BUG_ON(ext4_should_journal_data(inode));
Mingming Cao525f4ed2008-08-19 22:15:58 -04002480 needed_blocks = ext4_da_writepages_trans_blocks(inode);
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002481
Jan Kara4e7ea812013-06-04 13:17:40 -04002482 /* start a new transaction */
Jan Kara6b523df2013-06-04 13:21:11 -04002483 handle = ext4_journal_start_with_reserve(inode,
2484 EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
Mingming Cao61628a32008-07-11 19:27:31 -04002485 if (IS_ERR(handle)) {
2486 ret = PTR_ERR(handle);
Theodore Ts'o16939182009-09-26 17:43:59 -04002487 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
Curt Wohlgemuthfbe845d2010-05-16 13:00:00 -04002488 "%ld pages, ino %lu; err %d", __func__,
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002489 wbc->nr_to_write, inode->i_ino, ret);
Jan Kara4e7ea812013-06-04 13:17:40 -04002490 /* Release allocated io_end */
2491 ext4_put_io_end(mpd.io_submit.io_end);
2492 break;
Mingming Cao61628a32008-07-11 19:27:31 -04002493 }
Theodore Ts'of63e6002009-02-23 16:42:39 -05002494
Jan Kara4e7ea812013-06-04 13:17:40 -04002495 trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc);
2496 ret = mpage_prepare_extent_to_map(&mpd);
2497 if (!ret) {
2498 if (mpd.map.m_len)
Theodore Ts'ocb530542013-07-01 08:12:40 -04002499 ret = mpage_map_and_submit_extent(handle, &mpd,
2500 &give_up_on_write);
Jan Kara4e7ea812013-06-04 13:17:40 -04002501 else {
2502 /*
2503 * We scanned the whole range (or exhausted
2504 * nr_to_write), submitted what was mapped and
2505 * didn't find anything needing mapping. We are
2506 * done.
2507 */
2508 done = true;
2509 }
Theodore Ts'of63e6002009-02-23 16:42:39 -05002510 }
Mingming Cao61628a32008-07-11 19:27:31 -04002511 ext4_journal_stop(handle);
Jan Kara4e7ea812013-06-04 13:17:40 -04002512 /* Submit prepared bio */
2513 ext4_io_submit(&mpd.io_submit);
2514 /* Unlock pages we didn't use */
Theodore Ts'ocb530542013-07-01 08:12:40 -04002515 mpage_release_unused_pages(&mpd, give_up_on_write);
Jan Kara4e7ea812013-06-04 13:17:40 -04002516 /* Drop our io_end reference we got from init */
2517 ext4_put_io_end(mpd.io_submit.io_end);
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04002518
Jan Kara4e7ea812013-06-04 13:17:40 -04002519 if (ret == -ENOSPC && sbi->s_journal) {
2520 /*
2521 * Commit the transaction which would
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002522 * free blocks released in the transaction
2523 * and try again
2524 */
Aneesh Kumar K.Vdf222912008-09-08 23:05:34 -04002525 jbd2_journal_force_commit_nested(sbi->s_journal);
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002526 ret = 0;
Jan Kara4e7ea812013-06-04 13:17:40 -04002527 continue;
2528 }
2529 /* Fatal error - ENOMEM, EIO... */
2530 if (ret)
Mingming Cao61628a32008-07-11 19:27:31 -04002531 break;
Mingming Cao61628a32008-07-11 19:27:31 -04002532 }
Shaohua Li1bce63d2011-10-18 10:55:51 -04002533 blk_finish_plug(&plug);
Jan Kara9c12a832013-09-16 08:24:26 -04002534 if (!ret && !cycled && wbc->nr_to_write > 0) {
Aneesh Kumar K.V2acf2c22009-02-14 10:42:58 -05002535 cycled = 1;
Jan Kara4e7ea812013-06-04 13:17:40 -04002536 mpd.last_page = writeback_index - 1;
2537 mpd.first_page = 0;
Aneesh Kumar K.V2acf2c22009-02-14 10:42:58 -05002538 goto retry;
2539 }
Mingming Cao61628a32008-07-11 19:27:31 -04002540
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002541 /* Update index */
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002542 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2543 /*
Jan Kara4e7ea812013-06-04 13:17:40 -04002544 * Set the writeback_index so that range_cyclic
Aneesh Kumar K.V22208de2008-10-16 10:10:36 -04002545 * mode will write it back later
2546 */
Jan Kara4e7ea812013-06-04 13:17:40 -04002547 mapping->writeback_index = mpd.first_page;
Aneesh Kumar K.Va1d6cc52008-08-19 21:55:02 -04002548
Mingming Cao61628a32008-07-11 19:27:31 -04002549out_writepages:
Theodore Ts'o20970ba2013-06-06 14:00:46 -04002550 trace_ext4_writepages_result(inode, wbc, ret,
2551 nr_to_write - wbc->nr_to_write);
Mingming Cao61628a32008-07-11 19:27:31 -04002552 return ret;
Alex Tomas64769242008-07-11 19:27:31 -04002553}
2554
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04002555static int ext4_nonda_switch(struct super_block *sb)
2556{
Eric Whitney5c1ff332013-04-09 09:27:31 -04002557 s64 free_clusters, dirty_clusters;
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04002558 struct ext4_sb_info *sbi = EXT4_SB(sb);
2559
2560 /*
2561 * switch to non delalloc mode if we are running low
2562 * on free block. The free block accounting via percpu
Eric Dumazet179f7eb2009-01-06 14:41:04 -08002563 * counters can get slightly wrong with percpu_counter_batch getting
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04002564 * accumulated on each CPU without updating global counters
2565 * Delalloc need an accurate free block accounting. So switch
2566 * to non delalloc when we are near to error range.
2567 */
Eric Whitney5c1ff332013-04-09 09:27:31 -04002568 free_clusters =
2569 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2570 dirty_clusters =
2571 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
Theodore Ts'o00d4e732012-09-19 22:42:36 -04002572 /*
2573 * Start pushing delalloc when 1/2 of free blocks are dirty.
2574 */
Eric Whitney5c1ff332013-04-09 09:27:31 -04002575 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
Miao Xie10ee27a2013-01-10 13:47:57 +08002576 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
Theodore Ts'o00d4e732012-09-19 22:42:36 -04002577
Eric Whitney5c1ff332013-04-09 09:27:31 -04002578 if (2 * free_clusters < 3 * dirty_clusters ||
2579 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04002580 /*
Eric Sandeenc8afb442009-12-23 07:58:12 -05002581 * free block count is less than 150% of dirty blocks
2582 * or free blocks is less than watermark
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04002583 */
2584 return 1;
2585 }
2586 return 0;
2587}
2588
Eric Sandeen0ff89472014-10-11 19:51:17 -04002589/* We always reserve for an inode update; the superblock could be there too */
2590static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len)
2591{
2592 if (likely(EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
2593 EXT4_FEATURE_RO_COMPAT_LARGE_FILE)))
2594 return 1;
2595
2596 if (pos + len <= 0x7fffffffULL)
2597 return 1;
2598
2599 /* We might need to update the superblock to set LARGE_FILE */
2600 return 2;
2601}
2602
Alex Tomas64769242008-07-11 19:27:31 -04002603static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04002604 loff_t pos, unsigned len, unsigned flags,
2605 struct page **pagep, void **fsdata)
Alex Tomas64769242008-07-11 19:27:31 -04002606{
Eric Sandeen72b8ab92010-05-16 11:00:00 -04002607 int ret, retries = 0;
Alex Tomas64769242008-07-11 19:27:31 -04002608 struct page *page;
2609 pgoff_t index;
Alex Tomas64769242008-07-11 19:27:31 -04002610 struct inode *inode = mapping->host;
2611 handle_t *handle;
2612
2613 index = pos >> PAGE_CACHE_SHIFT;
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04002614
2615 if (ext4_nonda_switch(inode->i_sb)) {
2616 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2617 return ext4_write_begin(file, mapping, pos,
2618 len, flags, pagep, fsdata);
2619 }
2620 *fsdata = (void *)0;
Theodore Ts'o9bffad12009-06-17 11:48:11 -04002621 trace_ext4_da_write_begin(inode, pos, len, flags);
Tao Ma9c3569b2012-12-10 14:05:57 -05002622
2623 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2624 ret = ext4_da_write_inline_data_begin(mapping, inode,
2625 pos, len, flags,
2626 pagep, fsdata);
2627 if (ret < 0)
Theodore Ts'o47564bf2013-02-09 09:24:14 -05002628 return ret;
2629 if (ret == 1)
2630 return 0;
Tao Ma9c3569b2012-12-10 14:05:57 -05002631 }
2632
Theodore Ts'o47564bf2013-02-09 09:24:14 -05002633 /*
2634 * grab_cache_page_write_begin() can take a long time if the
2635 * system is thrashing due to memory pressure, or if the page
2636 * is being written back. So grab it first before we start
2637 * the transaction handle. This also allows us to allocate
2638 * the page (if needed) without using GFP_NOFS.
2639 */
2640retry_grab:
2641 page = grab_cache_page_write_begin(mapping, index, flags);
2642 if (!page)
2643 return -ENOMEM;
2644 unlock_page(page);
2645
Alex Tomas64769242008-07-11 19:27:31 -04002646 /*
2647 * With delayed allocation, we don't log the i_disksize update
2648 * if there is delayed block allocation. But we still need
2649 * to journalling the i_disksize update if writes to the end
2650 * of file which has an already mapped buffer.
2651 */
Theodore Ts'o47564bf2013-02-09 09:24:14 -05002652retry_journal:
Eric Sandeen0ff89472014-10-11 19:51:17 -04002653 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2654 ext4_da_write_credits(inode, pos, len));
Alex Tomas64769242008-07-11 19:27:31 -04002655 if (IS_ERR(handle)) {
Theodore Ts'o47564bf2013-02-09 09:24:14 -05002656 page_cache_release(page);
2657 return PTR_ERR(handle);
Alex Tomas64769242008-07-11 19:27:31 -04002658 }
2659
Theodore Ts'o47564bf2013-02-09 09:24:14 -05002660 lock_page(page);
2661 if (page->mapping != mapping) {
2662 /* The page got truncated from under us */
2663 unlock_page(page);
2664 page_cache_release(page);
Eric Sandeend5a0d4f2008-08-02 18:51:06 -04002665 ext4_journal_stop(handle);
Theodore Ts'o47564bf2013-02-09 09:24:14 -05002666 goto retry_grab;
Eric Sandeend5a0d4f2008-08-02 18:51:06 -04002667 }
Theodore Ts'o47564bf2013-02-09 09:24:14 -05002668 /* In case writeback began while the page was unlocked */
Dmitry Monakhov7afe5aa2013-08-28 14:30:47 -04002669 wait_for_stable_page(page);
Alex Tomas64769242008-07-11 19:27:31 -04002670
Michael Halcrow2058f832015-04-12 00:55:10 -04002671#ifdef CONFIG_EXT4_FS_ENCRYPTION
2672 ret = ext4_block_write_begin(page, pos, len,
2673 ext4_da_get_block_prep);
2674#else
Christoph Hellwig6e1db882010-06-04 11:29:57 +02002675 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
Michael Halcrow2058f832015-04-12 00:55:10 -04002676#endif
Alex Tomas64769242008-07-11 19:27:31 -04002677 if (ret < 0) {
2678 unlock_page(page);
2679 ext4_journal_stop(handle);
Aneesh Kumar K.Vae4d5372008-09-13 13:10:25 -04002680 /*
2681 * block_write_begin may have instantiated a few blocks
2682 * outside i_size. Trim these off again. Don't need
2683 * i_size_read because we hold i_mutex.
2684 */
2685 if (pos + len > inode->i_size)
Jan Karab9a42072009-12-08 21:24:33 -05002686 ext4_truncate_failed_write(inode);
Theodore Ts'o47564bf2013-02-09 09:24:14 -05002687
2688 if (ret == -ENOSPC &&
2689 ext4_should_retry_alloc(inode->i_sb, &retries))
2690 goto retry_journal;
2691
2692 page_cache_release(page);
2693 return ret;
Alex Tomas64769242008-07-11 19:27:31 -04002694 }
2695
Theodore Ts'o47564bf2013-02-09 09:24:14 -05002696 *pagep = page;
Alex Tomas64769242008-07-11 19:27:31 -04002697 return ret;
2698}
2699
Mingming Cao632eaea2008-07-11 19:27:31 -04002700/*
2701 * Check if we should update i_disksize
2702 * when write to the end of file but not require block allocation
2703 */
2704static int ext4_da_should_update_i_disksize(struct page *page,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04002705 unsigned long offset)
Mingming Cao632eaea2008-07-11 19:27:31 -04002706{
2707 struct buffer_head *bh;
2708 struct inode *inode = page->mapping->host;
2709 unsigned int idx;
2710 int i;
2711
2712 bh = page_buffers(page);
2713 idx = offset >> inode->i_blkbits;
2714
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002715 for (i = 0; i < idx; i++)
Mingming Cao632eaea2008-07-11 19:27:31 -04002716 bh = bh->b_this_page;
2717
Aneesh Kumar K.V29fa89d2009-05-12 16:30:27 -04002718 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
Mingming Cao632eaea2008-07-11 19:27:31 -04002719 return 0;
2720 return 1;
2721}
2722
Alex Tomas64769242008-07-11 19:27:31 -04002723static int ext4_da_write_end(struct file *file,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04002724 struct address_space *mapping,
2725 loff_t pos, unsigned len, unsigned copied,
2726 struct page *page, void *fsdata)
Alex Tomas64769242008-07-11 19:27:31 -04002727{
2728 struct inode *inode = mapping->host;
2729 int ret = 0, ret2;
2730 handle_t *handle = ext4_journal_current_handle();
2731 loff_t new_i_size;
Mingming Cao632eaea2008-07-11 19:27:31 -04002732 unsigned long start, end;
Aneesh Kumar K.V79f0be82008-10-08 23:13:30 -04002733 int write_mode = (int)(unsigned long)fsdata;
2734
Theodore Ts'o74d553a2013-04-03 12:39:17 -04002735 if (write_mode == FALL_BACK_TO_NONDELALLOC)
2736 return ext4_write_end(file, mapping, pos,
2737 len, copied, page, fsdata);
Mingming Cao632eaea2008-07-11 19:27:31 -04002738
Theodore Ts'o9bffad12009-06-17 11:48:11 -04002739 trace_ext4_da_write_end(inode, pos, len, copied);
Mingming Cao632eaea2008-07-11 19:27:31 -04002740 start = pos & (PAGE_CACHE_SIZE - 1);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002741 end = start + copied - 1;
Alex Tomas64769242008-07-11 19:27:31 -04002742
2743 /*
2744 * generic_write_end() will run mark_inode_dirty() if i_size
2745 * changes. So let's piggyback the i_disksize mark_inode_dirty
2746 * into that.
2747 */
Alex Tomas64769242008-07-11 19:27:31 -04002748 new_i_size = pos + copied;
Andrea Arcangeliea51d132011-12-13 21:41:15 -05002749 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
Tao Ma9c3569b2012-12-10 14:05:57 -05002750 if (ext4_has_inline_data(inode) ||
2751 ext4_da_should_update_i_disksize(page, end)) {
Dmitry Monakhovee124d22014-08-30 23:34:06 -04002752 ext4_update_i_disksize(inode, new_i_size);
Aneesh Kumar K.Vcf17fea2008-09-13 13:06:18 -04002753 /* We need to mark inode dirty even if
2754 * new_i_size is less that inode->i_size
2755 * bu greater than i_disksize.(hint delalloc)
2756 */
2757 ext4_mark_inode_dirty(handle, inode);
Alex Tomas64769242008-07-11 19:27:31 -04002758 }
Mingming Cao632eaea2008-07-11 19:27:31 -04002759 }
Tao Ma9c3569b2012-12-10 14:05:57 -05002760
2761 if (write_mode != CONVERT_INLINE_DATA &&
2762 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
2763 ext4_has_inline_data(inode))
2764 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
2765 page);
2766 else
2767 ret2 = generic_write_end(file, mapping, pos, len, copied,
Alex Tomas64769242008-07-11 19:27:31 -04002768 page, fsdata);
Tao Ma9c3569b2012-12-10 14:05:57 -05002769
Alex Tomas64769242008-07-11 19:27:31 -04002770 copied = ret2;
2771 if (ret2 < 0)
2772 ret = ret2;
2773 ret2 = ext4_journal_stop(handle);
2774 if (!ret)
2775 ret = ret2;
2776
2777 return ret ? ret : copied;
2778}
2779
Lukas Czernerd47992f2013-05-21 23:17:23 -04002780static void ext4_da_invalidatepage(struct page *page, unsigned int offset,
2781 unsigned int length)
Alex Tomas64769242008-07-11 19:27:31 -04002782{
Alex Tomas64769242008-07-11 19:27:31 -04002783 /*
2784 * Drop reserved blocks
2785 */
2786 BUG_ON(!PageLocked(page));
2787 if (!page_has_buffers(page))
2788 goto out;
2789
Lukas Czernerca99fdd2013-05-21 23:25:01 -04002790 ext4_da_page_release_reservation(page, offset, length);
Alex Tomas64769242008-07-11 19:27:31 -04002791
2792out:
Lukas Czernerd47992f2013-05-21 23:17:23 -04002793 ext4_invalidatepage(page, offset, length);
Alex Tomas64769242008-07-11 19:27:31 -04002794
2795 return;
2796}
2797
Theodore Ts'occd25062009-02-26 01:04:07 -05002798/*
2799 * Force all delayed allocation blocks to be allocated for a given inode.
2800 */
2801int ext4_alloc_da_blocks(struct inode *inode)
2802{
Theodore Ts'ofb40ba02009-09-16 19:30:40 -04002803 trace_ext4_alloc_da_blocks(inode);
2804
Theodore Ts'o71d4f7d2014-07-15 06:02:38 -04002805 if (!EXT4_I(inode)->i_reserved_data_blocks)
Theodore Ts'occd25062009-02-26 01:04:07 -05002806 return 0;
2807
2808 /*
2809 * We do something simple for now. The filemap_flush() will
2810 * also start triggering a write of the data blocks, which is
2811 * not strictly speaking necessary (and for users of
2812 * laptop_mode, not even desirable). However, to do otherwise
2813 * would require replicating code paths in:
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04002814 *
Theodore Ts'o20970ba2013-06-06 14:00:46 -04002815 * ext4_writepages() ->
Theodore Ts'occd25062009-02-26 01:04:07 -05002816 * write_cache_pages() ---> (via passed in callback function)
2817 * __mpage_da_writepage() -->
2818 * mpage_add_bh_to_extent()
2819 * mpage_da_map_blocks()
2820 *
2821 * The problem is that write_cache_pages(), located in
2822 * mm/page-writeback.c, marks pages clean in preparation for
2823 * doing I/O, which is not desirable if we're not planning on
2824 * doing I/O at all.
2825 *
2826 * We could call write_cache_pages(), and then redirty all of
Wu Fengguang380cf092010-11-11 19:23:29 +08002827 * the pages by calling redirty_page_for_writepage() but that
Theodore Ts'occd25062009-02-26 01:04:07 -05002828 * would be ugly in the extreme. So instead we would need to
2829 * replicate parts of the code in the above functions,
Lucas De Marchi25985ed2011-03-30 22:57:33 -03002830 * simplifying them because we wouldn't actually intend to
Theodore Ts'occd25062009-02-26 01:04:07 -05002831 * write out the pages, but rather only collect contiguous
2832 * logical block extents, call the multi-block allocator, and
2833 * then update the buffer heads with the block allocations.
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04002834 *
Theodore Ts'occd25062009-02-26 01:04:07 -05002835 * For now, though, we'll cheat by calling filemap_flush(),
2836 * which will map the blocks, and start the I/O, but not
2837 * actually wait for the I/O to complete.
2838 */
2839 return filemap_flush(inode->i_mapping);
2840}
Alex Tomas64769242008-07-11 19:27:31 -04002841
2842/*
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002843 * bmap() is special. It gets used by applications such as lilo and by
2844 * the swapper to find the on-disk block of a specific piece of data.
2845 *
2846 * Naturally, this is dangerous if the block concerned is still in the
Mingming Cao617ba132006-10-11 01:20:53 -07002847 * journal. If somebody makes a swapfile on an ext4 data-journaling
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002848 * filesystem and enables swap, then they may get a nasty shock when the
2849 * data getting swapped to that swapfile suddenly gets overwritten by
2850 * the original zero's written out previously to the journal and
2851 * awaiting writeback in the kernel's buffer cache.
2852 *
2853 * So, if we see any bmap calls here on a modified, data-journaled file,
2854 * take extra steps to flush any blocks which might be in the cache.
2855 */
Mingming Cao617ba132006-10-11 01:20:53 -07002856static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002857{
2858 struct inode *inode = mapping->host;
2859 journal_t *journal;
2860 int err;
2861
Tao Ma46c7f252012-12-10 14:04:52 -05002862 /*
2863 * We can get here for an inline file via the FIBMAP ioctl
2864 */
2865 if (ext4_has_inline_data(inode))
2866 return 0;
2867
Alex Tomas64769242008-07-11 19:27:31 -04002868 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2869 test_opt(inode->i_sb, DELALLOC)) {
2870 /*
2871 * With delalloc we want to sync the file
2872 * so that we can make sure we allocate
2873 * blocks for file
2874 */
2875 filemap_write_and_wait(mapping);
2876 }
2877
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05002878 if (EXT4_JOURNAL(inode) &&
2879 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002880 /*
2881 * This is a REALLY heavyweight approach, but the use of
2882 * bmap on dirty files is expected to be extremely rare:
2883 * only if we run lilo or swapon on a freshly made file
2884 * do we expect this to happen.
2885 *
2886 * (bmap requires CAP_SYS_RAWIO so this does not
2887 * represent an unprivileged user DOS attack --- we'd be
2888 * in trouble if mortal users could trigger this path at
2889 * will.)
2890 *
Mingming Cao617ba132006-10-11 01:20:53 -07002891 * NB. EXT4_STATE_JDATA is not set on files other than
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002892 * regular files. If somebody wants to bmap a directory
2893 * or symlink and gets confused because the buffer
2894 * hasn't yet been flushed to disk, they deserve
2895 * everything they get.
2896 */
2897
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05002898 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
Mingming Cao617ba132006-10-11 01:20:53 -07002899 journal = EXT4_JOURNAL(inode);
Mingming Caodab291a2006-10-11 01:21:01 -07002900 jbd2_journal_lock_updates(journal);
2901 err = jbd2_journal_flush(journal);
2902 jbd2_journal_unlock_updates(journal);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002903
2904 if (err)
2905 return 0;
2906 }
2907
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002908 return generic_block_bmap(mapping, block, ext4_get_block);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002909}
2910
Mingming Cao617ba132006-10-11 01:20:53 -07002911static int ext4_readpage(struct file *file, struct page *page)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002912{
Tao Ma46c7f252012-12-10 14:04:52 -05002913 int ret = -EAGAIN;
2914 struct inode *inode = page->mapping->host;
2915
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04002916 trace_ext4_readpage(page);
Tao Ma46c7f252012-12-10 14:04:52 -05002917
2918 if (ext4_has_inline_data(inode))
2919 ret = ext4_readpage_inline(inode, page);
2920
2921 if (ret == -EAGAIN)
Theodore Ts'of64e02f2015-04-08 00:00:32 -04002922 return ext4_mpage_readpages(page->mapping, NULL, page, 1);
Tao Ma46c7f252012-12-10 14:04:52 -05002923
2924 return ret;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002925}
2926
2927static int
Mingming Cao617ba132006-10-11 01:20:53 -07002928ext4_readpages(struct file *file, struct address_space *mapping,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002929 struct list_head *pages, unsigned nr_pages)
2930{
Tao Ma46c7f252012-12-10 14:04:52 -05002931 struct inode *inode = mapping->host;
2932
2933 /* If the file has inline data, no need to do readpages. */
2934 if (ext4_has_inline_data(inode))
2935 return 0;
2936
Theodore Ts'of64e02f2015-04-08 00:00:32 -04002937 return ext4_mpage_readpages(mapping, pages, NULL, nr_pages);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002938}
2939
Lukas Czernerd47992f2013-05-21 23:17:23 -04002940static void ext4_invalidatepage(struct page *page, unsigned int offset,
2941 unsigned int length)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002942{
Lukas Czernerca99fdd2013-05-21 23:25:01 -04002943 trace_ext4_invalidatepage(page, offset, length);
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04002944
Jan Kara4520fb32012-12-25 13:28:54 -05002945 /* No journalling happens on data buffers when this function is used */
2946 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
2947
Lukas Czernerca99fdd2013-05-21 23:25:01 -04002948 block_invalidatepage(page, offset, length);
Jan Kara4520fb32012-12-25 13:28:54 -05002949}
2950
Jan Kara53e87262012-12-25 13:29:52 -05002951static int __ext4_journalled_invalidatepage(struct page *page,
Lukas Czernerca99fdd2013-05-21 23:25:01 -04002952 unsigned int offset,
2953 unsigned int length)
Jan Kara4520fb32012-12-25 13:28:54 -05002954{
2955 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2956
Lukas Czernerca99fdd2013-05-21 23:25:01 -04002957 trace_ext4_journalled_invalidatepage(page, offset, length);
Jan Kara4520fb32012-12-25 13:28:54 -05002958
Jiaying Zhang744692d2010-03-04 16:14:02 -05002959 /*
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002960 * If it's a full truncate we just forget about the pending dirtying
2961 */
Lukas Czernerca99fdd2013-05-21 23:25:01 -04002962 if (offset == 0 && length == PAGE_CACHE_SIZE)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002963 ClearPageChecked(page);
2964
Lukas Czernerca99fdd2013-05-21 23:25:01 -04002965 return jbd2_journal_invalidatepage(journal, page, offset, length);
Jan Kara53e87262012-12-25 13:29:52 -05002966}
2967
2968/* Wrapper for aops... */
2969static void ext4_journalled_invalidatepage(struct page *page,
Lukas Czernerd47992f2013-05-21 23:17:23 -04002970 unsigned int offset,
2971 unsigned int length)
Jan Kara53e87262012-12-25 13:29:52 -05002972{
Lukas Czernerca99fdd2013-05-21 23:25:01 -04002973 WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002974}
2975
Mingming Cao617ba132006-10-11 01:20:53 -07002976static int ext4_releasepage(struct page *page, gfp_t wait)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002977{
Mingming Cao617ba132006-10-11 01:20:53 -07002978 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002979
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04002980 trace_ext4_releasepage(page);
2981
Jan Karae1c36592013-03-10 22:19:00 -04002982 /* Page has dirty journalled data -> cannot release */
2983 if (PageChecked(page))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002984 return 0;
Frank Mayhar03901312009-01-07 00:06:22 -05002985 if (journal)
2986 return jbd2_journal_try_to_free_buffers(journal, page, wait);
2987 else
2988 return try_to_free_buffers(page);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002989}
2990
2991/*
Theodore Ts'o2ed88682010-05-16 20:00:00 -04002992 * ext4_get_block used when preparing for a DIO write or buffer write.
2993 * We allocate an uinitialized extent if blocks haven't been allocated.
2994 * The extent will be converted to initialized after the IO is complete.
2995 */
Tao Maf19d5872012-12-10 14:05:51 -05002996int ext4_get_block_write(struct inode *inode, sector_t iblock,
Mingming Cao4c0425f2009-09-28 15:48:41 -04002997 struct buffer_head *bh_result, int create)
2998{
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05002999 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003000 inode->i_ino, create);
Theodore Ts'o2ed88682010-05-16 20:00:00 -04003001 return _ext4_get_block(inode, iblock, bh_result,
3002 EXT4_GET_BLOCKS_IO_CREATE_EXT);
Mingming Cao4c0425f2009-09-28 15:48:41 -04003003}
3004
Zheng Liu729f52c2012-07-09 16:29:29 -04003005static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
Anatol Pomozov8b0f1652012-11-08 15:07:16 -05003006 struct buffer_head *bh_result, int create)
Zheng Liu729f52c2012-07-09 16:29:29 -04003007{
Anatol Pomozov8b0f1652012-11-08 15:07:16 -05003008 ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n",
3009 inode->i_ino, create);
3010 return _ext4_get_block(inode, iblock, bh_result,
3011 EXT4_GET_BLOCKS_NO_LOCK);
Zheng Liu729f52c2012-07-09 16:29:29 -04003012}
3013
Mingming Cao4c0425f2009-09-28 15:48:41 -04003014static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
Christoph Hellwig7b7a8662013-09-04 15:04:39 +02003015 ssize_t size, void *private)
Mingming Cao4c0425f2009-09-28 15:48:41 -04003016{
3017 ext4_io_end_t *io_end = iocb->private;
Mingming Cao4c0425f2009-09-28 15:48:41 -04003018
Jan Kara97a851e2013-06-04 11:58:58 -04003019 /* if not async direct IO just return */
Christoph Hellwig7b7a8662013-09-04 15:04:39 +02003020 if (!io_end)
Jan Kara97a851e2013-06-04 11:58:58 -04003021 return;
Mingming4b70df12009-11-03 14:44:54 -05003022
Zheng Liu88635ca2011-12-28 19:00:25 -05003023 ext_debug("ext4_end_io_dio(): io_end 0x%p "
Joe Perchesace36ad2012-03-19 23:11:43 -04003024 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003025 iocb->private, io_end->inode->i_ino, iocb, offset,
3026 size);
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003027
Theodore Ts'ob5a7e972011-12-12 10:53:02 -05003028 iocb->private = NULL;
Mingming Cao4c0425f2009-09-28 15:48:41 -04003029 io_end->offset = offset;
3030 io_end->size = size;
Christoph Hellwig7b7a8662013-09-04 15:04:39 +02003031 ext4_put_io_end(io_end);
Mingming Cao4c0425f2009-09-28 15:48:41 -04003032}
Jiaying Zhangc7064ef2010-03-02 13:28:44 -05003033
Mingming Cao4c0425f2009-09-28 15:48:41 -04003034/*
3035 * For ext4 extent files, ext4 will do direct-io write to holes,
3036 * preallocated extents, and those write extend the file, no need to
3037 * fall back to buffered IO.
3038 *
Lukas Czerner556615d2014-04-20 23:45:47 -04003039 * For holes, we fallocate those blocks, mark them as unwritten
Theodore Ts'o69c499d2012-11-29 21:13:48 -05003040 * If those blocks were preallocated, we mark sure they are split, but
Lukas Czerner556615d2014-04-20 23:45:47 -04003041 * still keep the range to write as unwritten.
Mingming Cao4c0425f2009-09-28 15:48:41 -04003042 *
Theodore Ts'o69c499d2012-11-29 21:13:48 -05003043 * The unwritten extents will be converted to written when DIO is completed.
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003044 * For async direct IO, since the IO may still pending when return, we
Lucas De Marchi25985ed2011-03-30 22:57:33 -03003045 * set up an end_io call back function, which will do the conversion
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003046 * when async direct IO completed.
Mingming Cao4c0425f2009-09-28 15:48:41 -04003047 *
3048 * If the O_DIRECT write will extend the file then add this inode to the
3049 * orphan list. So recovery will truncate it back to the original size
3050 * if the machine crashes during the write.
3051 *
3052 */
Omar Sandoval6f673762015-03-16 04:33:52 -07003053static ssize_t ext4_ext_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
3054 loff_t offset)
Mingming Cao4c0425f2009-09-28 15:48:41 -04003055{
3056 struct file *file = iocb->ki_filp;
3057 struct inode *inode = file->f_mapping->host;
3058 ssize_t ret;
Al Viroa6cbcd42014-03-04 22:38:00 -05003059 size_t count = iov_iter_count(iter);
Theodore Ts'o69c499d2012-11-29 21:13:48 -05003060 int overwrite = 0;
3061 get_block_t *get_block_func = NULL;
3062 int dio_flags = 0;
Mingming Cao4c0425f2009-09-28 15:48:41 -04003063 loff_t final_size = offset + count;
Jan Kara97a851e2013-06-04 11:58:58 -04003064 ext4_io_end_t *io_end = NULL;
Zheng Liu729f52c2012-07-09 16:29:29 -04003065
Theodore Ts'o69c499d2012-11-29 21:13:48 -05003066 /* Use the old path for reads and writes beyond i_size. */
Omar Sandoval6f673762015-03-16 04:33:52 -07003067 if (iov_iter_rw(iter) != WRITE || final_size > inode->i_size)
3068 return ext4_ind_direct_IO(iocb, iter, offset);
Zheng Liu4bd809d2012-07-22 20:19:31 -04003069
Theodore Ts'o69c499d2012-11-29 21:13:48 -05003070 BUG_ON(iocb->private == NULL);
Zheng Liu4bd809d2012-07-22 20:19:31 -04003071
Jan Karae8340392013-06-04 14:27:38 -04003072 /*
3073 * Make all waiters for direct IO properly wait also for extent
3074 * conversion. This also disallows race between truncate() and
3075 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
3076 */
Omar Sandoval6f673762015-03-16 04:33:52 -07003077 if (iov_iter_rw(iter) == WRITE)
Jens Axboefe0f07d2015-04-15 17:05:48 -06003078 inode_dio_begin(inode);
Jan Karae8340392013-06-04 14:27:38 -04003079
Theodore Ts'o69c499d2012-11-29 21:13:48 -05003080 /* If we do a overwrite dio, i_mutex locking can be released */
3081 overwrite = *((int *)iocb->private);
Zheng Liu4bd809d2012-07-22 20:19:31 -04003082
Theodore Ts'o69c499d2012-11-29 21:13:48 -05003083 if (overwrite) {
Theodore Ts'o69c499d2012-11-29 21:13:48 -05003084 down_read(&EXT4_I(inode)->i_data_sem);
3085 mutex_unlock(&inode->i_mutex);
Mingming Cao4c0425f2009-09-28 15:48:41 -04003086 }
Mingming Cao8d5d02e2009-09-28 15:48:29 -04003087
Theodore Ts'o69c499d2012-11-29 21:13:48 -05003088 /*
3089 * We could direct write to holes and fallocate.
3090 *
3091 * Allocated blocks to fill the hole are marked as
Lukas Czerner556615d2014-04-20 23:45:47 -04003092 * unwritten to prevent parallel buffered read to expose
Theodore Ts'o69c499d2012-11-29 21:13:48 -05003093 * the stale data before DIO complete the data IO.
3094 *
3095 * As to previously fallocated extents, ext4 get_block will
3096 * just simply mark the buffer mapped but still keep the
Lukas Czerner556615d2014-04-20 23:45:47 -04003097 * extents unwritten.
Theodore Ts'o69c499d2012-11-29 21:13:48 -05003098 *
3099 * For non AIO case, we will convert those unwritten extents
3100 * to written after return back from blockdev_direct_IO.
3101 *
3102 * For async DIO, the conversion needs to be deferred when the
3103 * IO is completed. The ext4 end_io callback function will be
3104 * called to take care of the conversion work. Here for async
3105 * case, we allocate an io_end structure to hook to the iocb.
3106 */
3107 iocb->private = NULL;
3108 ext4_inode_aio_set(inode, NULL);
3109 if (!is_sync_kiocb(iocb)) {
Jan Kara97a851e2013-06-04 11:58:58 -04003110 io_end = ext4_init_io_end(inode, GFP_NOFS);
Theodore Ts'o69c499d2012-11-29 21:13:48 -05003111 if (!io_end) {
3112 ret = -ENOMEM;
3113 goto retake_lock;
3114 }
Jan Kara97a851e2013-06-04 11:58:58 -04003115 /*
3116 * Grab reference for DIO. Will be dropped in ext4_end_io_dio()
3117 */
3118 iocb->private = ext4_get_io_end(io_end);
Theodore Ts'o69c499d2012-11-29 21:13:48 -05003119 /*
3120 * we save the io structure for current async direct
3121 * IO, so that later ext4_map_blocks() could flag the
3122 * io structure whether there is a unwritten extents
3123 * needs to be converted when IO is completed.
3124 */
3125 ext4_inode_aio_set(inode, io_end);
3126 }
3127
3128 if (overwrite) {
3129 get_block_func = ext4_get_block_write_nolock;
3130 } else {
3131 get_block_func = ext4_get_block_write;
3132 dio_flags = DIO_LOCKING;
3133 }
Michael Halcrow2058f832015-04-12 00:55:10 -04003134#ifdef CONFIG_EXT4_FS_ENCRYPTION
3135 BUG_ON(ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode));
3136#endif
Ross Zwisler923ae0f2015-02-16 15:59:38 -08003137 if (IS_DAX(inode))
Omar Sandovala95cd632015-03-16 04:33:51 -07003138 ret = dax_do_io(iocb, inode, iter, offset, get_block_func,
Ross Zwisler923ae0f2015-02-16 15:59:38 -08003139 ext4_end_io_dio, dio_flags);
3140 else
Omar Sandoval17f8c842015-03-16 04:33:50 -07003141 ret = __blockdev_direct_IO(iocb, inode,
Ross Zwisler923ae0f2015-02-16 15:59:38 -08003142 inode->i_sb->s_bdev, iter, offset,
3143 get_block_func,
3144 ext4_end_io_dio, NULL, dio_flags);
Theodore Ts'o69c499d2012-11-29 21:13:48 -05003145
Theodore Ts'oa5499842013-05-11 19:07:42 -04003146 /*
Jan Kara97a851e2013-06-04 11:58:58 -04003147 * Put our reference to io_end. This can free the io_end structure e.g.
3148 * in sync IO case or in case of error. It can even perform extent
3149 * conversion if all bios we submitted finished before we got here.
3150 * Note that in that case iocb->private can be already set to NULL
3151 * here.
Theodore Ts'oa5499842013-05-11 19:07:42 -04003152 */
Jan Kara97a851e2013-06-04 11:58:58 -04003153 if (io_end) {
3154 ext4_inode_aio_set(inode, NULL);
3155 ext4_put_io_end(io_end);
3156 /*
3157 * When no IO was submitted ext4_end_io_dio() was not
3158 * called so we have to put iocb's reference.
3159 */
3160 if (ret <= 0 && ret != -EIOCBQUEUED && iocb->private) {
3161 WARN_ON(iocb->private != io_end);
3162 WARN_ON(io_end->flag & EXT4_IO_END_UNWRITTEN);
Jan Kara97a851e2013-06-04 11:58:58 -04003163 ext4_put_io_end(io_end);
3164 iocb->private = NULL;
3165 }
3166 }
3167 if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
Theodore Ts'o69c499d2012-11-29 21:13:48 -05003168 EXT4_STATE_DIO_UNWRITTEN)) {
3169 int err;
3170 /*
3171 * for non AIO case, since the IO is already
3172 * completed, we could do the conversion right here
3173 */
Jan Kara6b523df2013-06-04 13:21:11 -04003174 err = ext4_convert_unwritten_extents(NULL, inode,
Theodore Ts'o69c499d2012-11-29 21:13:48 -05003175 offset, ret);
3176 if (err < 0)
3177 ret = err;
3178 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3179 }
3180
3181retake_lock:
Omar Sandoval6f673762015-03-16 04:33:52 -07003182 if (iov_iter_rw(iter) == WRITE)
Jens Axboefe0f07d2015-04-15 17:05:48 -06003183 inode_dio_end(inode);
Theodore Ts'o69c499d2012-11-29 21:13:48 -05003184 /* take i_mutex locking again if we do a ovewrite dio */
3185 if (overwrite) {
Theodore Ts'o69c499d2012-11-29 21:13:48 -05003186 up_read(&EXT4_I(inode)->i_data_sem);
3187 mutex_lock(&inode->i_mutex);
3188 }
3189
3190 return ret;
Mingming Cao4c0425f2009-09-28 15:48:41 -04003191}
3192
Omar Sandoval22c61862015-03-16 04:33:53 -07003193static ssize_t ext4_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
3194 loff_t offset)
Mingming Cao4c0425f2009-09-28 15:48:41 -04003195{
3196 struct file *file = iocb->ki_filp;
3197 struct inode *inode = file->f_mapping->host;
Al Viroa6cbcd42014-03-04 22:38:00 -05003198 size_t count = iov_iter_count(iter);
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04003199 ssize_t ret;
Mingming Cao4c0425f2009-09-28 15:48:41 -04003200
Michael Halcrow2058f832015-04-12 00:55:10 -04003201#ifdef CONFIG_EXT4_FS_ENCRYPTION
3202 if (ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode))
3203 return 0;
3204#endif
3205
Theodore Ts'o84ebd792011-08-31 11:56:51 -04003206 /*
3207 * If we are doing data journalling we don't support O_DIRECT
3208 */
3209 if (ext4_should_journal_data(inode))
3210 return 0;
3211
Tao Ma46c7f252012-12-10 14:04:52 -05003212 /* Let buffer I/O handle the inline data case. */
3213 if (ext4_has_inline_data(inode))
3214 return 0;
3215
Omar Sandoval6f673762015-03-16 04:33:52 -07003216 trace_ext4_direct_IO_enter(inode, offset, count, iov_iter_rw(iter));
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04003217 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
Omar Sandoval6f673762015-03-16 04:33:52 -07003218 ret = ext4_ext_direct_IO(iocb, iter, offset);
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04003219 else
Omar Sandoval6f673762015-03-16 04:33:52 -07003220 ret = ext4_ind_direct_IO(iocb, iter, offset);
3221 trace_ext4_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), ret);
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04003222 return ret;
Mingming Cao4c0425f2009-09-28 15:48:41 -04003223}
3224
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003225/*
Mingming Cao617ba132006-10-11 01:20:53 -07003226 * Pages can be marked dirty completely asynchronously from ext4's journalling
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003227 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3228 * much here because ->set_page_dirty is called under VFS locks. The page is
3229 * not necessarily locked.
3230 *
3231 * We cannot just dirty the page and leave attached buffers clean, because the
3232 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3233 * or jbddirty because all the journalling code will explode.
3234 *
3235 * So what we do is to mark the page "pending dirty" and next time writepage
3236 * is called, propagate that into the buffers appropriately.
3237 */
Mingming Cao617ba132006-10-11 01:20:53 -07003238static int ext4_journalled_set_page_dirty(struct page *page)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003239{
3240 SetPageChecked(page);
3241 return __set_page_dirty_nobuffers(page);
3242}
3243
Theodore Ts'o74d553a2013-04-03 12:39:17 -04003244static const struct address_space_operations ext4_aops = {
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003245 .readpage = ext4_readpage,
3246 .readpages = ext4_readpages,
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04003247 .writepage = ext4_writepage,
Theodore Ts'o20970ba2013-06-06 14:00:46 -04003248 .writepages = ext4_writepages,
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003249 .write_begin = ext4_write_begin,
Theodore Ts'o74d553a2013-04-03 12:39:17 -04003250 .write_end = ext4_write_end,
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003251 .bmap = ext4_bmap,
3252 .invalidatepage = ext4_invalidatepage,
3253 .releasepage = ext4_releasepage,
3254 .direct_IO = ext4_direct_IO,
3255 .migratepage = buffer_migrate_page,
3256 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02003257 .error_remove_page = generic_error_remove_page,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003258};
3259
Mingming Cao617ba132006-10-11 01:20:53 -07003260static const struct address_space_operations ext4_journalled_aops = {
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003261 .readpage = ext4_readpage,
3262 .readpages = ext4_readpages,
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04003263 .writepage = ext4_writepage,
Theodore Ts'o20970ba2013-06-06 14:00:46 -04003264 .writepages = ext4_writepages,
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003265 .write_begin = ext4_write_begin,
3266 .write_end = ext4_journalled_write_end,
3267 .set_page_dirty = ext4_journalled_set_page_dirty,
3268 .bmap = ext4_bmap,
Jan Kara4520fb32012-12-25 13:28:54 -05003269 .invalidatepage = ext4_journalled_invalidatepage,
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003270 .releasepage = ext4_releasepage,
Theodore Ts'o84ebd792011-08-31 11:56:51 -04003271 .direct_IO = ext4_direct_IO,
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003272 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02003273 .error_remove_page = generic_error_remove_page,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003274};
3275
Alex Tomas64769242008-07-11 19:27:31 -04003276static const struct address_space_operations ext4_da_aops = {
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003277 .readpage = ext4_readpage,
3278 .readpages = ext4_readpages,
Aneesh Kumar K.V43ce1d22009-06-14 17:58:45 -04003279 .writepage = ext4_writepage,
Theodore Ts'o20970ba2013-06-06 14:00:46 -04003280 .writepages = ext4_writepages,
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07003281 .write_begin = ext4_da_write_begin,
3282 .write_end = ext4_da_write_end,
3283 .bmap = ext4_bmap,
3284 .invalidatepage = ext4_da_invalidatepage,
3285 .releasepage = ext4_releasepage,
3286 .direct_IO = ext4_direct_IO,
3287 .migratepage = buffer_migrate_page,
3288 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02003289 .error_remove_page = generic_error_remove_page,
Alex Tomas64769242008-07-11 19:27:31 -04003290};
3291
Mingming Cao617ba132006-10-11 01:20:53 -07003292void ext4_set_aops(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003293{
Lukas Czerner3d2b1582012-02-20 17:53:00 -05003294 switch (ext4_inode_journal_mode(inode)) {
3295 case EXT4_INODE_ORDERED_DATA_MODE:
Theodore Ts'o74d553a2013-04-03 12:39:17 -04003296 ext4_set_inode_state(inode, EXT4_STATE_ORDERED_MODE);
Lukas Czerner3d2b1582012-02-20 17:53:00 -05003297 break;
3298 case EXT4_INODE_WRITEBACK_DATA_MODE:
Theodore Ts'o74d553a2013-04-03 12:39:17 -04003299 ext4_clear_inode_state(inode, EXT4_STATE_ORDERED_MODE);
Lukas Czerner3d2b1582012-02-20 17:53:00 -05003300 break;
3301 case EXT4_INODE_JOURNAL_DATA_MODE:
Mingming Cao617ba132006-10-11 01:20:53 -07003302 inode->i_mapping->a_ops = &ext4_journalled_aops;
Theodore Ts'o74d553a2013-04-03 12:39:17 -04003303 return;
Lukas Czerner3d2b1582012-02-20 17:53:00 -05003304 default:
3305 BUG();
3306 }
Theodore Ts'o74d553a2013-04-03 12:39:17 -04003307 if (test_opt(inode->i_sb, DELALLOC))
3308 inode->i_mapping->a_ops = &ext4_da_aops;
3309 else
3310 inode->i_mapping->a_ops = &ext4_aops;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003311}
3312
Ross Zwisler923ae0f2015-02-16 15:59:38 -08003313static int __ext4_block_zero_page_range(handle_t *handle,
Lukas Czernerd863dc32013-05-27 23:32:35 -04003314 struct address_space *mapping, loff_t from, loff_t length)
3315{
3316 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3317 unsigned offset = from & (PAGE_CACHE_SIZE-1);
Ross Zwisler923ae0f2015-02-16 15:59:38 -08003318 unsigned blocksize, pos;
Lukas Czernerd863dc32013-05-27 23:32:35 -04003319 ext4_lblk_t iblock;
3320 struct inode *inode = mapping->host;
3321 struct buffer_head *bh;
3322 struct page *page;
3323 int err = 0;
3324
3325 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3326 mapping_gfp_mask(mapping) & ~__GFP_FS);
3327 if (!page)
3328 return -ENOMEM;
3329
3330 blocksize = inode->i_sb->s_blocksize;
Lukas Czernerd863dc32013-05-27 23:32:35 -04003331
3332 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3333
3334 if (!page_has_buffers(page))
3335 create_empty_buffers(page, blocksize, 0);
3336
3337 /* Find the buffer that contains "offset" */
3338 bh = page_buffers(page);
3339 pos = blocksize;
3340 while (offset >= pos) {
3341 bh = bh->b_this_page;
3342 iblock++;
3343 pos += blocksize;
3344 }
Lukas Czernerd863dc32013-05-27 23:32:35 -04003345 if (buffer_freed(bh)) {
3346 BUFFER_TRACE(bh, "freed: skip");
3347 goto unlock;
3348 }
Lukas Czernerd863dc32013-05-27 23:32:35 -04003349 if (!buffer_mapped(bh)) {
3350 BUFFER_TRACE(bh, "unmapped");
3351 ext4_get_block(inode, iblock, bh, 0);
3352 /* unmapped? It's a hole - nothing to do */
3353 if (!buffer_mapped(bh)) {
3354 BUFFER_TRACE(bh, "still unmapped");
3355 goto unlock;
3356 }
3357 }
3358
3359 /* Ok, it's mapped. Make sure it's up-to-date */
3360 if (PageUptodate(page))
3361 set_buffer_uptodate(bh);
3362
3363 if (!buffer_uptodate(bh)) {
3364 err = -EIO;
3365 ll_rw_block(READ, 1, &bh);
3366 wait_on_buffer(bh);
3367 /* Uhhuh. Read error. Complain and punt. */
3368 if (!buffer_uptodate(bh))
3369 goto unlock;
Michael Halcrowc9c74292015-04-12 00:56:10 -04003370 if (S_ISREG(inode->i_mode) &&
3371 ext4_encrypted_inode(inode)) {
3372 /* We expect the key to be set. */
3373 BUG_ON(!ext4_has_encryption_key(inode));
3374 BUG_ON(blocksize != PAGE_CACHE_SIZE);
3375 WARN_ON_ONCE(ext4_decrypt_one(inode, page));
3376 }
Lukas Czernerd863dc32013-05-27 23:32:35 -04003377 }
Lukas Czernerd863dc32013-05-27 23:32:35 -04003378 if (ext4_should_journal_data(inode)) {
3379 BUFFER_TRACE(bh, "get write access");
3380 err = ext4_journal_get_write_access(handle, bh);
3381 if (err)
3382 goto unlock;
3383 }
Lukas Czernerd863dc32013-05-27 23:32:35 -04003384 zero_user(page, offset, length);
Lukas Czernerd863dc32013-05-27 23:32:35 -04003385 BUFFER_TRACE(bh, "zeroed end of block");
3386
Lukas Czernerd863dc32013-05-27 23:32:35 -04003387 if (ext4_should_journal_data(inode)) {
3388 err = ext4_handle_dirty_metadata(handle, inode, bh);
Lukas Czerner0713ed02013-05-27 23:32:35 -04003389 } else {
jon ernst353eefd2013-07-01 08:12:39 -04003390 err = 0;
Lukas Czernerd863dc32013-05-27 23:32:35 -04003391 mark_buffer_dirty(bh);
Lukas Czerner0713ed02013-05-27 23:32:35 -04003392 if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE))
3393 err = ext4_jbd2_file_inode(handle, inode);
3394 }
Lukas Czernerd863dc32013-05-27 23:32:35 -04003395
3396unlock:
3397 unlock_page(page);
3398 page_cache_release(page);
3399 return err;
3400}
3401
Matthew Wilcox94350ab2014-03-24 15:09:16 -04003402/*
Ross Zwisler923ae0f2015-02-16 15:59:38 -08003403 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3404 * starting from file offset 'from'. The range to be zero'd must
3405 * be contained with in one block. If the specified range exceeds
3406 * the end of the block it will be shortened to end of the block
3407 * that cooresponds to 'from'
3408 */
3409static int ext4_block_zero_page_range(handle_t *handle,
3410 struct address_space *mapping, loff_t from, loff_t length)
3411{
3412 struct inode *inode = mapping->host;
3413 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3414 unsigned blocksize = inode->i_sb->s_blocksize;
3415 unsigned max = blocksize - (offset & (blocksize - 1));
3416
3417 /*
3418 * correct length if it does not fall between
3419 * 'from' and the end of the block
3420 */
3421 if (length > max || length < 0)
3422 length = max;
3423
3424 if (IS_DAX(inode))
3425 return dax_zero_page_range(inode, from, length, ext4_get_block);
3426 return __ext4_block_zero_page_range(handle, mapping, from, length);
3427}
3428
3429/*
Matthew Wilcox94350ab2014-03-24 15:09:16 -04003430 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3431 * up to the end of the block which corresponds to `from'.
3432 * This required during truncate. We need to physically zero the tail end
3433 * of that block so it doesn't yield old data if the file is later grown.
3434 */
Stephen Hemmingerc1978552014-05-12 10:50:23 -04003435static int ext4_block_truncate_page(handle_t *handle,
Matthew Wilcox94350ab2014-03-24 15:09:16 -04003436 struct address_space *mapping, loff_t from)
3437{
3438 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3439 unsigned length;
3440 unsigned blocksize;
3441 struct inode *inode = mapping->host;
3442
3443 blocksize = inode->i_sb->s_blocksize;
3444 length = blocksize - (offset & (blocksize - 1));
3445
3446 return ext4_block_zero_page_range(handle, mapping, from, length);
3447}
3448
Lukas Czernera87dd182013-05-27 23:32:35 -04003449int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
3450 loff_t lstart, loff_t length)
3451{
3452 struct super_block *sb = inode->i_sb;
3453 struct address_space *mapping = inode->i_mapping;
Lukas Czernere1be3a92013-07-01 08:12:39 -04003454 unsigned partial_start, partial_end;
Lukas Czernera87dd182013-05-27 23:32:35 -04003455 ext4_fsblk_t start, end;
3456 loff_t byte_end = (lstart + length - 1);
3457 int err = 0;
3458
Lukas Czernere1be3a92013-07-01 08:12:39 -04003459 partial_start = lstart & (sb->s_blocksize - 1);
3460 partial_end = byte_end & (sb->s_blocksize - 1);
3461
Lukas Czernera87dd182013-05-27 23:32:35 -04003462 start = lstart >> sb->s_blocksize_bits;
3463 end = byte_end >> sb->s_blocksize_bits;
3464
3465 /* Handle partial zero within the single block */
Lukas Czernere1be3a92013-07-01 08:12:39 -04003466 if (start == end &&
3467 (partial_start || (partial_end != sb->s_blocksize - 1))) {
Lukas Czernera87dd182013-05-27 23:32:35 -04003468 err = ext4_block_zero_page_range(handle, mapping,
3469 lstart, length);
3470 return err;
3471 }
3472 /* Handle partial zero out on the start of the range */
Lukas Czernere1be3a92013-07-01 08:12:39 -04003473 if (partial_start) {
Lukas Czernera87dd182013-05-27 23:32:35 -04003474 err = ext4_block_zero_page_range(handle, mapping,
3475 lstart, sb->s_blocksize);
3476 if (err)
3477 return err;
3478 }
3479 /* Handle partial zero out on the end of the range */
Lukas Czernere1be3a92013-07-01 08:12:39 -04003480 if (partial_end != sb->s_blocksize - 1)
Lukas Czernera87dd182013-05-27 23:32:35 -04003481 err = ext4_block_zero_page_range(handle, mapping,
Lukas Czernere1be3a92013-07-01 08:12:39 -04003482 byte_end - partial_end,
3483 partial_end + 1);
Lukas Czernera87dd182013-05-27 23:32:35 -04003484 return err;
3485}
3486
Duane Griffin91ef4ca2008-07-11 19:27:31 -04003487int ext4_can_truncate(struct inode *inode)
3488{
Duane Griffin91ef4ca2008-07-11 19:27:31 -04003489 if (S_ISREG(inode->i_mode))
3490 return 1;
3491 if (S_ISDIR(inode->i_mode))
3492 return 1;
3493 if (S_ISLNK(inode->i_mode))
3494 return !ext4_inode_is_fast_symlink(inode);
3495 return 0;
3496}
3497
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003498/*
Allison Hendersona4bb6b62011-05-25 07:41:50 -04003499 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3500 * associated with the given offset and length
3501 *
3502 * @inode: File inode
3503 * @offset: The offset where the hole will begin
3504 * @len: The length of the hole
3505 *
Anatol Pomozov4907cb72012-09-01 10:31:09 -07003506 * Returns: 0 on success or negative on failure
Allison Hendersona4bb6b62011-05-25 07:41:50 -04003507 */
3508
Ashish Sangwanaeb28172013-07-01 08:12:38 -04003509int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length)
Allison Hendersona4bb6b62011-05-25 07:41:50 -04003510{
Theodore Ts'o26a4c0c2013-04-03 12:45:17 -04003511 struct super_block *sb = inode->i_sb;
3512 ext4_lblk_t first_block, stop_block;
3513 struct address_space *mapping = inode->i_mapping;
Lukas Czernera87dd182013-05-27 23:32:35 -04003514 loff_t first_block_offset, last_block_offset;
Theodore Ts'o26a4c0c2013-04-03 12:45:17 -04003515 handle_t *handle;
3516 unsigned int credits;
3517 int ret = 0;
3518
Allison Hendersona4bb6b62011-05-25 07:41:50 -04003519 if (!S_ISREG(inode->i_mode))
Allison Henderson73355192012-03-21 22:23:31 -04003520 return -EOPNOTSUPP;
Allison Hendersona4bb6b62011-05-25 07:41:50 -04003521
Lukas Czernerb8a86842014-03-18 18:05:35 -04003522 trace_ext4_punch_hole(inode, offset, length, 0);
Zheng Liuaaddea82013-01-16 20:21:26 -05003523
Theodore Ts'o26a4c0c2013-04-03 12:45:17 -04003524 /*
3525 * Write out all dirty pages to avoid race conditions
3526 * Then release them.
3527 */
3528 if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
3529 ret = filemap_write_and_wait_range(mapping, offset,
3530 offset + length - 1);
3531 if (ret)
3532 return ret;
3533 }
3534
3535 mutex_lock(&inode->i_mutex);
Lukas Czerner9ef06ce2014-04-12 09:47:00 -04003536
Theodore Ts'o26a4c0c2013-04-03 12:45:17 -04003537 /* No need to punch hole beyond i_size */
3538 if (offset >= inode->i_size)
3539 goto out_mutex;
3540
3541 /*
3542 * If the hole extends beyond i_size, set the hole
3543 * to end after the page that contains i_size
3544 */
3545 if (offset + length > inode->i_size) {
3546 length = inode->i_size +
3547 PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) -
3548 offset;
3549 }
3550
Jan Karaa3612932013-08-16 21:19:41 -04003551 if (offset & (sb->s_blocksize - 1) ||
3552 (offset + length) & (sb->s_blocksize - 1)) {
3553 /*
3554 * Attach jinode to inode for jbd2 if we do any zeroing of
3555 * partial block
3556 */
3557 ret = ext4_inode_attach_jinode(inode);
3558 if (ret < 0)
3559 goto out_mutex;
3560
3561 }
3562
Lukas Czernera87dd182013-05-27 23:32:35 -04003563 first_block_offset = round_up(offset, sb->s_blocksize);
3564 last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
Theodore Ts'o26a4c0c2013-04-03 12:45:17 -04003565
Lukas Czernera87dd182013-05-27 23:32:35 -04003566 /* Now release the pages and zero block aligned part of pages*/
3567 if (last_block_offset > first_block_offset)
3568 truncate_pagecache_range(inode, first_block_offset,
3569 last_block_offset);
Theodore Ts'o26a4c0c2013-04-03 12:45:17 -04003570
3571 /* Wait all existing dio workers, newcomers will block on i_mutex */
3572 ext4_inode_block_unlocked_dio(inode);
Theodore Ts'o26a4c0c2013-04-03 12:45:17 -04003573 inode_dio_wait(inode);
3574
3575 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3576 credits = ext4_writepage_trans_blocks(inode);
3577 else
3578 credits = ext4_blocks_for_truncate(inode);
3579 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3580 if (IS_ERR(handle)) {
3581 ret = PTR_ERR(handle);
3582 ext4_std_error(sb, ret);
3583 goto out_dio;
3584 }
3585
Lukas Czernera87dd182013-05-27 23:32:35 -04003586 ret = ext4_zero_partial_blocks(handle, inode, offset,
3587 length);
3588 if (ret)
3589 goto out_stop;
Theodore Ts'o26a4c0c2013-04-03 12:45:17 -04003590
3591 first_block = (offset + sb->s_blocksize - 1) >>
3592 EXT4_BLOCK_SIZE_BITS(sb);
3593 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
3594
3595 /* If there are no blocks to remove, return now */
3596 if (first_block >= stop_block)
3597 goto out_stop;
3598
3599 down_write(&EXT4_I(inode)->i_data_sem);
3600 ext4_discard_preallocations(inode);
3601
3602 ret = ext4_es_remove_extent(inode, first_block,
3603 stop_block - first_block);
3604 if (ret) {
3605 up_write(&EXT4_I(inode)->i_data_sem);
3606 goto out_stop;
3607 }
3608
3609 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3610 ret = ext4_ext_remove_space(inode, first_block,
3611 stop_block - 1);
3612 else
Lukas Czerner4f579ae2014-07-15 06:03:38 -04003613 ret = ext4_ind_remove_space(handle, inode, first_block,
Theodore Ts'o26a4c0c2013-04-03 12:45:17 -04003614 stop_block);
3615
Theodore Ts'o819c4922013-04-03 12:47:17 -04003616 up_write(&EXT4_I(inode)->i_data_sem);
Theodore Ts'o26a4c0c2013-04-03 12:45:17 -04003617 if (IS_SYNC(inode))
3618 ext4_handle_sync(handle);
Maxim Patlasove251f9b2014-02-20 16:58:05 -05003619
3620 /* Now release the pages again to reduce race window */
3621 if (last_block_offset > first_block_offset)
3622 truncate_pagecache_range(inode, first_block_offset,
3623 last_block_offset);
3624
Theodore Ts'o26a4c0c2013-04-03 12:45:17 -04003625 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3626 ext4_mark_inode_dirty(handle, inode);
3627out_stop:
3628 ext4_journal_stop(handle);
3629out_dio:
3630 ext4_inode_resume_unlocked_dio(inode);
3631out_mutex:
3632 mutex_unlock(&inode->i_mutex);
3633 return ret;
Allison Hendersona4bb6b62011-05-25 07:41:50 -04003634}
3635
Jan Karaa3612932013-08-16 21:19:41 -04003636int ext4_inode_attach_jinode(struct inode *inode)
3637{
3638 struct ext4_inode_info *ei = EXT4_I(inode);
3639 struct jbd2_inode *jinode;
3640
3641 if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
3642 return 0;
3643
3644 jinode = jbd2_alloc_inode(GFP_KERNEL);
3645 spin_lock(&inode->i_lock);
3646 if (!ei->jinode) {
3647 if (!jinode) {
3648 spin_unlock(&inode->i_lock);
3649 return -ENOMEM;
3650 }
3651 ei->jinode = jinode;
3652 jbd2_journal_init_jbd_inode(ei->jinode, inode);
3653 jinode = NULL;
3654 }
3655 spin_unlock(&inode->i_lock);
3656 if (unlikely(jinode != NULL))
3657 jbd2_free_inode(jinode);
3658 return 0;
3659}
3660
Allison Hendersona4bb6b62011-05-25 07:41:50 -04003661/*
Mingming Cao617ba132006-10-11 01:20:53 -07003662 * ext4_truncate()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003663 *
Mingming Cao617ba132006-10-11 01:20:53 -07003664 * We block out ext4_get_block() block instantiations across the entire
3665 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003666 * simultaneously on behalf of the same inode.
3667 *
Justin P. Mattock42b2aa82011-11-28 20:31:00 -08003668 * As we work through the truncate and commit bits of it to the journal there
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003669 * is one core, guiding principle: the file's tree must always be consistent on
3670 * disk. We must be able to restart the truncate after a crash.
3671 *
3672 * The file's tree may be transiently inconsistent in memory (although it
3673 * probably isn't), but whenever we close off and commit a journal transaction,
3674 * the contents of (the filesystem + the journal) must be consistent and
3675 * restartable. It's pretty simple, really: bottom up, right to left (although
3676 * left-to-right works OK too).
3677 *
3678 * Note that at recovery time, journal replay occurs *before* the restart of
3679 * truncate against the orphan inode list.
3680 *
3681 * The committed inode has the new, desired i_size (which is the same as
Mingming Cao617ba132006-10-11 01:20:53 -07003682 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003683 * that this inode's truncate did not complete and it will again call
Mingming Cao617ba132006-10-11 01:20:53 -07003684 * ext4_truncate() to have another go. So there will be instantiated blocks
3685 * to the right of the truncation point in a crashed ext4 filesystem. But
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003686 * that's fine - as long as they are linked from the inode, the post-crash
Mingming Cao617ba132006-10-11 01:20:53 -07003687 * ext4_truncate() run will find them and release them.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003688 */
Mingming Cao617ba132006-10-11 01:20:53 -07003689void ext4_truncate(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003690{
Theodore Ts'o819c4922013-04-03 12:47:17 -04003691 struct ext4_inode_info *ei = EXT4_I(inode);
3692 unsigned int credits;
3693 handle_t *handle;
3694 struct address_space *mapping = inode->i_mapping;
Theodore Ts'o819c4922013-04-03 12:47:17 -04003695
Theodore Ts'o19b5ef62013-04-03 21:58:52 -04003696 /*
3697 * There is a possibility that we're either freeing the inode
Matthew Wilcoxe04027e2014-03-24 15:15:07 -04003698 * or it's a completely new inode. In those cases we might not
Theodore Ts'o19b5ef62013-04-03 21:58:52 -04003699 * have i_mutex locked because it's not necessary.
3700 */
3701 if (!(inode->i_state & (I_NEW|I_FREEING)))
3702 WARN_ON(!mutex_is_locked(&inode->i_mutex));
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04003703 trace_ext4_truncate_enter(inode);
3704
Duane Griffin91ef4ca2008-07-11 19:27:31 -04003705 if (!ext4_can_truncate(inode))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003706 return;
3707
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04003708 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
Jiaying Zhangc8d46e42010-02-24 09:52:53 -05003709
Theodore Ts'o5534fb52009-09-17 09:34:16 -04003710 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05003711 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
Theodore Ts'o7d8f9f72009-02-24 08:21:14 -05003712
Tao Maaef1c852012-12-10 14:06:02 -05003713 if (ext4_has_inline_data(inode)) {
3714 int has_inline = 1;
3715
3716 ext4_inline_data_truncate(inode, &has_inline);
3717 if (has_inline)
3718 return;
3719 }
3720
Jan Karaa3612932013-08-16 21:19:41 -04003721 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
3722 if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
3723 if (ext4_inode_attach_jinode(inode) < 0)
3724 return;
3725 }
3726
Amir Goldsteinff9893d2011-06-27 16:36:31 -04003727 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
Theodore Ts'o819c4922013-04-03 12:47:17 -04003728 credits = ext4_writepage_trans_blocks(inode);
Amir Goldsteinff9893d2011-06-27 16:36:31 -04003729 else
Theodore Ts'o819c4922013-04-03 12:47:17 -04003730 credits = ext4_blocks_for_truncate(inode);
3731
3732 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3733 if (IS_ERR(handle)) {
3734 ext4_std_error(inode->i_sb, PTR_ERR(handle));
3735 return;
3736 }
3737
Lukas Czernereb3544c2013-05-27 23:32:35 -04003738 if (inode->i_size & (inode->i_sb->s_blocksize - 1))
3739 ext4_block_truncate_page(handle, mapping, inode->i_size);
Theodore Ts'o819c4922013-04-03 12:47:17 -04003740
3741 /*
3742 * We add the inode to the orphan list, so that if this
3743 * truncate spans multiple transactions, and we crash, we will
3744 * resume the truncate when the filesystem recovers. It also
3745 * marks the inode dirty, to catch the new size.
3746 *
3747 * Implication: the file must always be in a sane, consistent
3748 * truncatable state while each transaction commits.
3749 */
3750 if (ext4_orphan_add(handle, inode))
3751 goto out_stop;
3752
3753 down_write(&EXT4_I(inode)->i_data_sem);
3754
3755 ext4_discard_preallocations(inode);
3756
3757 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3758 ext4_ext_truncate(handle, inode);
3759 else
3760 ext4_ind_truncate(handle, inode);
3761
3762 up_write(&ei->i_data_sem);
3763
3764 if (IS_SYNC(inode))
3765 ext4_handle_sync(handle);
3766
3767out_stop:
3768 /*
3769 * If this was a simple ftruncate() and the file will remain alive,
3770 * then we need to clear up the orphan record which we created above.
3771 * However, if this was a real unlink then we were called by
Wang Shilong58d86a52014-11-25 16:17:29 -05003772 * ext4_evict_inode(), and we allow that function to clean up the
Theodore Ts'o819c4922013-04-03 12:47:17 -04003773 * orphan info for us.
3774 */
3775 if (inode->i_nlink)
3776 ext4_orphan_del(handle, inode);
3777
3778 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3779 ext4_mark_inode_dirty(handle, inode);
3780 ext4_journal_stop(handle);
Alex Tomasa86c6182006-10-11 01:21:03 -07003781
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04003782 trace_ext4_truncate_exit(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003783}
3784
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003785/*
Mingming Cao617ba132006-10-11 01:20:53 -07003786 * ext4_get_inode_loc returns with an extra refcount against the inode's
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003787 * underlying buffer_head on success. If 'in_mem' is true, we have all
3788 * data in memory that is needed to recreate the on-disk version of this
3789 * inode.
3790 */
Mingming Cao617ba132006-10-11 01:20:53 -07003791static int __ext4_get_inode_loc(struct inode *inode,
3792 struct ext4_iloc *iloc, int in_mem)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003793{
Theodore Ts'o240799c2008-10-09 23:53:47 -04003794 struct ext4_group_desc *gdp;
3795 struct buffer_head *bh;
3796 struct super_block *sb = inode->i_sb;
3797 ext4_fsblk_t block;
3798 int inodes_per_block, inode_offset;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003799
Aneesh Kumar K.V3a06d772008-11-22 15:04:59 -05003800 iloc->bh = NULL;
Theodore Ts'o240799c2008-10-09 23:53:47 -04003801 if (!ext4_valid_inum(sb, inode->i_ino))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003802 return -EIO;
3803
Theodore Ts'o240799c2008-10-09 23:53:47 -04003804 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
3805 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
3806 if (!gdp)
3807 return -EIO;
3808
3809 /*
3810 * Figure out the offset within the block group inode table
3811 */
Tao Ma00d09882011-05-09 10:26:41 -04003812 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
Theodore Ts'o240799c2008-10-09 23:53:47 -04003813 inode_offset = ((inode->i_ino - 1) %
3814 EXT4_INODES_PER_GROUP(sb));
3815 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
3816 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
3817
3818 bh = sb_getblk(sb, block);
Wang Shilongaebf0242013-01-12 16:28:47 -05003819 if (unlikely(!bh))
Theodore Ts'o860d21e2013-01-12 16:19:36 -05003820 return -ENOMEM;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003821 if (!buffer_uptodate(bh)) {
3822 lock_buffer(bh);
Hidehiro Kawai9c83a922008-07-26 16:39:26 -04003823
3824 /*
3825 * If the buffer has the write error flag, we have failed
3826 * to write out another inode in the same block. In this
3827 * case, we don't have to read the block because we may
3828 * read the old inode data successfully.
3829 */
3830 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
3831 set_buffer_uptodate(bh);
3832
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003833 if (buffer_uptodate(bh)) {
3834 /* someone brought it uptodate while we waited */
3835 unlock_buffer(bh);
3836 goto has_buffer;
3837 }
3838
3839 /*
3840 * If we have all information of the inode in memory and this
3841 * is the only valid inode in the block, we need not read the
3842 * block.
3843 */
3844 if (in_mem) {
3845 struct buffer_head *bitmap_bh;
Theodore Ts'o240799c2008-10-09 23:53:47 -04003846 int i, start;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003847
Theodore Ts'o240799c2008-10-09 23:53:47 -04003848 start = inode_offset & ~(inodes_per_block - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003849
3850 /* Is the inode bitmap in cache? */
Theodore Ts'o240799c2008-10-09 23:53:47 -04003851 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
Wang Shilongaebf0242013-01-12 16:28:47 -05003852 if (unlikely(!bitmap_bh))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003853 goto make_io;
3854
3855 /*
3856 * If the inode bitmap isn't in cache then the
3857 * optimisation may end up performing two reads instead
3858 * of one, so skip it.
3859 */
3860 if (!buffer_uptodate(bitmap_bh)) {
3861 brelse(bitmap_bh);
3862 goto make_io;
3863 }
Theodore Ts'o240799c2008-10-09 23:53:47 -04003864 for (i = start; i < start + inodes_per_block; i++) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003865 if (i == inode_offset)
3866 continue;
Mingming Cao617ba132006-10-11 01:20:53 -07003867 if (ext4_test_bit(i, bitmap_bh->b_data))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003868 break;
3869 }
3870 brelse(bitmap_bh);
Theodore Ts'o240799c2008-10-09 23:53:47 -04003871 if (i == start + inodes_per_block) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003872 /* all other inodes are free, so skip I/O */
3873 memset(bh->b_data, 0, bh->b_size);
3874 set_buffer_uptodate(bh);
3875 unlock_buffer(bh);
3876 goto has_buffer;
3877 }
3878 }
3879
3880make_io:
3881 /*
Theodore Ts'o240799c2008-10-09 23:53:47 -04003882 * If we need to do any I/O, try to pre-readahead extra
3883 * blocks from the inode table.
3884 */
3885 if (EXT4_SB(sb)->s_inode_readahead_blks) {
3886 ext4_fsblk_t b, end, table;
3887 unsigned num;
Theodore Ts'o0d606e22013-04-23 08:59:35 -04003888 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
Theodore Ts'o240799c2008-10-09 23:53:47 -04003889
3890 table = ext4_inode_table(sb, gdp);
Theodore Ts'ob713a5e2009-03-31 09:11:14 -04003891 /* s_inode_readahead_blks is always a power of 2 */
Theodore Ts'o0d606e22013-04-23 08:59:35 -04003892 b = block & ~((ext4_fsblk_t) ra_blks - 1);
Theodore Ts'o240799c2008-10-09 23:53:47 -04003893 if (table > b)
3894 b = table;
Theodore Ts'o0d606e22013-04-23 08:59:35 -04003895 end = b + ra_blks;
Theodore Ts'o240799c2008-10-09 23:53:47 -04003896 num = EXT4_INODES_PER_GROUP(sb);
Darrick J. Wongfeb0ab32012-04-29 18:45:10 -04003897 if (ext4_has_group_desc_csum(sb))
Aneesh Kumar K.V560671a2009-01-05 22:20:24 -05003898 num -= ext4_itable_unused_count(sb, gdp);
Theodore Ts'o240799c2008-10-09 23:53:47 -04003899 table += num / inodes_per_block;
3900 if (end > table)
3901 end = table;
3902 while (b <= end)
3903 sb_breadahead(sb, b++);
3904 }
3905
3906 /*
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003907 * There are other valid inodes in the buffer, this inode
3908 * has in-inode xattrs, or we don't have this inode in memory.
3909 * Read the block from disk.
3910 */
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04003911 trace_ext4_load_inode(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003912 get_bh(bh);
3913 bh->b_end_io = end_buffer_read_sync;
Christoph Hellwig65299a32011-08-23 14:50:29 +02003914 submit_bh(READ | REQ_META | REQ_PRIO, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003915 wait_on_buffer(bh);
3916 if (!buffer_uptodate(bh)) {
Theodore Ts'oc398eda2010-07-27 11:56:40 -04003917 EXT4_ERROR_INODE_BLOCK(inode, block,
3918 "unable to read itable block");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003919 brelse(bh);
3920 return -EIO;
3921 }
3922 }
3923has_buffer:
3924 iloc->bh = bh;
3925 return 0;
3926}
3927
Mingming Cao617ba132006-10-11 01:20:53 -07003928int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003929{
3930 /* We have all inode data except xattrs in memory here. */
Mingming Cao617ba132006-10-11 01:20:53 -07003931 return __ext4_get_inode_loc(inode, iloc,
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05003932 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003933}
3934
Mingming Cao617ba132006-10-11 01:20:53 -07003935void ext4_set_inode_flags(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003936{
Mingming Cao617ba132006-10-11 01:20:53 -07003937 unsigned int flags = EXT4_I(inode)->i_flags;
Theodore Ts'o00a1a052014-03-30 10:20:01 -04003938 unsigned int new_fl = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003939
Mingming Cao617ba132006-10-11 01:20:53 -07003940 if (flags & EXT4_SYNC_FL)
Theodore Ts'o00a1a052014-03-30 10:20:01 -04003941 new_fl |= S_SYNC;
Mingming Cao617ba132006-10-11 01:20:53 -07003942 if (flags & EXT4_APPEND_FL)
Theodore Ts'o00a1a052014-03-30 10:20:01 -04003943 new_fl |= S_APPEND;
Mingming Cao617ba132006-10-11 01:20:53 -07003944 if (flags & EXT4_IMMUTABLE_FL)
Theodore Ts'o00a1a052014-03-30 10:20:01 -04003945 new_fl |= S_IMMUTABLE;
Mingming Cao617ba132006-10-11 01:20:53 -07003946 if (flags & EXT4_NOATIME_FL)
Theodore Ts'o00a1a052014-03-30 10:20:01 -04003947 new_fl |= S_NOATIME;
Mingming Cao617ba132006-10-11 01:20:53 -07003948 if (flags & EXT4_DIRSYNC_FL)
Theodore Ts'o00a1a052014-03-30 10:20:01 -04003949 new_fl |= S_DIRSYNC;
Ross Zwisler923ae0f2015-02-16 15:59:38 -08003950 if (test_opt(inode->i_sb, DAX))
3951 new_fl |= S_DAX;
Theodore Ts'o5f16f322014-03-24 14:43:12 -04003952 inode_set_flags(inode, new_fl,
Ross Zwisler923ae0f2015-02-16 15:59:38 -08003953 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003954}
3955
Jan Karaff9ddf72007-07-18 09:24:20 -04003956/* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3957void ext4_get_inode_flags(struct ext4_inode_info *ei)
3958{
Dmitry Monakhov84a8dce2010-06-05 11:51:27 -04003959 unsigned int vfs_fl;
3960 unsigned long old_fl, new_fl;
Jan Karaff9ddf72007-07-18 09:24:20 -04003961
Dmitry Monakhov84a8dce2010-06-05 11:51:27 -04003962 do {
3963 vfs_fl = ei->vfs_inode.i_flags;
3964 old_fl = ei->i_flags;
3965 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
3966 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
3967 EXT4_DIRSYNC_FL);
3968 if (vfs_fl & S_SYNC)
3969 new_fl |= EXT4_SYNC_FL;
3970 if (vfs_fl & S_APPEND)
3971 new_fl |= EXT4_APPEND_FL;
3972 if (vfs_fl & S_IMMUTABLE)
3973 new_fl |= EXT4_IMMUTABLE_FL;
3974 if (vfs_fl & S_NOATIME)
3975 new_fl |= EXT4_NOATIME_FL;
3976 if (vfs_fl & S_DIRSYNC)
3977 new_fl |= EXT4_DIRSYNC_FL;
3978 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
Jan Karaff9ddf72007-07-18 09:24:20 -04003979}
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04003980
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05003981static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04003982 struct ext4_inode_info *ei)
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05003983{
3984 blkcnt_t i_blocks ;
Aneesh Kumar K.V8180a562008-01-28 23:58:27 -05003985 struct inode *inode = &(ei->vfs_inode);
3986 struct super_block *sb = inode->i_sb;
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05003987
3988 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3989 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
3990 /* we are using combined 48 bit field */
3991 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
3992 le32_to_cpu(raw_inode->i_blocks_lo);
Theodore Ts'o07a03822010-06-14 09:54:48 -04003993 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
Aneesh Kumar K.V8180a562008-01-28 23:58:27 -05003994 /* i_blocks represent file system block size */
3995 return i_blocks << (inode->i_blkbits - 9);
3996 } else {
3997 return i_blocks;
3998 }
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05003999 } else {
4000 return le32_to_cpu(raw_inode->i_blocks_lo);
4001 }
4002}
Jan Karaff9ddf72007-07-18 09:24:20 -04004003
Tao Ma152a7b02012-12-02 11:13:24 -05004004static inline void ext4_iget_extra_inode(struct inode *inode,
4005 struct ext4_inode *raw_inode,
4006 struct ext4_inode_info *ei)
4007{
4008 __le32 *magic = (void *)raw_inode +
4009 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
Tao Ma67cf5b02012-12-10 14:04:46 -05004010 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
Tao Ma152a7b02012-12-02 11:13:24 -05004011 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
Tao Ma67cf5b02012-12-10 14:04:46 -05004012 ext4_find_inline_data_nolock(inode);
Tao Maf19d5872012-12-10 14:05:51 -05004013 } else
4014 EXT4_I(inode)->i_inline_off = 0;
Tao Ma152a7b02012-12-02 11:13:24 -05004015}
4016
David Howells1d1fe1e2008-02-07 00:15:37 -08004017struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004018{
Mingming Cao617ba132006-10-11 01:20:53 -07004019 struct ext4_iloc iloc;
4020 struct ext4_inode *raw_inode;
David Howells1d1fe1e2008-02-07 00:15:37 -08004021 struct ext4_inode_info *ei;
David Howells1d1fe1e2008-02-07 00:15:37 -08004022 struct inode *inode;
Jan Karab436b9b2009-12-08 23:51:10 -05004023 journal_t *journal = EXT4_SB(sb)->s_journal;
David Howells1d1fe1e2008-02-07 00:15:37 -08004024 long ret;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004025 int block;
Eric W. Biederman08cefc72012-02-07 15:41:49 -08004026 uid_t i_uid;
4027 gid_t i_gid;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004028
David Howells1d1fe1e2008-02-07 00:15:37 -08004029 inode = iget_locked(sb, ino);
4030 if (!inode)
4031 return ERR_PTR(-ENOMEM);
4032 if (!(inode->i_state & I_NEW))
4033 return inode;
4034
4035 ei = EXT4_I(inode);
Peter Huewe7dc57612011-02-21 21:01:42 -05004036 iloc.bh = NULL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004037
David Howells1d1fe1e2008-02-07 00:15:37 -08004038 ret = __ext4_get_inode_loc(inode, &iloc, 0);
4039 if (ret < 0)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004040 goto bad_inode;
Mingming Cao617ba132006-10-11 01:20:53 -07004041 raw_inode = ext4_raw_inode(&iloc);
Darrick J. Wong814525f2012-04-29 18:31:10 -04004042
4043 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4044 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4045 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4046 EXT4_INODE_SIZE(inode->i_sb)) {
4047 EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)",
4048 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize,
4049 EXT4_INODE_SIZE(inode->i_sb));
4050 ret = -EIO;
4051 goto bad_inode;
4052 }
4053 } else
4054 ei->i_extra_isize = 0;
4055
4056 /* Precompute checksum seed for inode metadata */
Dmitry Monakhov9aa5d322014-10-13 03:36:16 -04004057 if (ext4_has_metadata_csum(sb)) {
Darrick J. Wong814525f2012-04-29 18:31:10 -04004058 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4059 __u32 csum;
4060 __le32 inum = cpu_to_le32(inode->i_ino);
4061 __le32 gen = raw_inode->i_generation;
4062 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4063 sizeof(inum));
4064 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4065 sizeof(gen));
4066 }
4067
4068 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
4069 EXT4_ERROR_INODE(inode, "checksum invalid");
4070 ret = -EIO;
4071 goto bad_inode;
4072 }
4073
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004074 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
Eric W. Biederman08cefc72012-02-07 15:41:49 -08004075 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4076 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04004077 if (!(test_opt(inode->i_sb, NO_UID32))) {
Eric W. Biederman08cefc72012-02-07 15:41:49 -08004078 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4079 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004080 }
Eric W. Biederman08cefc72012-02-07 15:41:49 -08004081 i_uid_write(inode, i_uid);
4082 i_gid_write(inode, i_gid);
Miklos Szeredibfe86842011-10-28 14:13:29 +02004083 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004084
Theodore Ts'o353eb832011-01-10 12:18:25 -05004085 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
Tao Ma67cf5b02012-12-10 14:04:46 -05004086 ei->i_inline_off = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004087 ei->i_dir_start_lookup = 0;
4088 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4089 /* We now have enough fields to check if the inode was active or not.
4090 * This is needed because nfsd might try to access dead inodes
4091 * the test is that same one that e2fsck uses
4092 * NeilBrown 1999oct15
4093 */
4094 if (inode->i_nlink == 0) {
Dr. Tilmann Bubeck393d1d12013-04-08 12:54:05 -04004095 if ((inode->i_mode == 0 ||
4096 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4097 ino != EXT4_BOOT_LOADER_INO) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004098 /* this inode is deleted */
David Howells1d1fe1e2008-02-07 00:15:37 -08004099 ret = -ESTALE;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004100 goto bad_inode;
4101 }
4102 /* The only unlinked inodes we let through here have
4103 * valid i_mode and are being read by the orphan
4104 * recovery code: that's fine, we're about to complete
Dr. Tilmann Bubeck393d1d12013-04-08 12:54:05 -04004105 * the process of deleting those.
4106 * OR it is the EXT4_BOOT_LOADER_INO which is
4107 * not initialized on a new filesystem. */
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004108 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004109 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004110 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
Aneesh Kumar K.V7973c0c2008-01-28 23:58:27 -05004111 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
Theodore Ts'oa9e81742009-04-24 16:11:18 -04004112 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
Badari Pulavartya1ddeb72006-10-11 01:21:09 -07004113 ei->i_file_acl |=
4114 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
Aneesh Kumar K.Va48380f2008-01-28 23:58:27 -05004115 inode->i_size = ext4_isize(raw_inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004116 ei->i_disksize = inode->i_size;
Dmitry Monakhova9e7f442009-12-14 15:21:14 +03004117#ifdef CONFIG_QUOTA
4118 ei->i_reserved_quota = 0;
4119#endif
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004120 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4121 ei->i_block_group = iloc.block_group;
Theodore Ts'oa4912122009-03-12 12:18:34 -04004122 ei->i_last_alloc_group = ~0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004123 /*
4124 * NOTE! The in-memory inode i_data array is in little-endian order
4125 * even on big-endian machines: we do NOT byteswap the block numbers!
4126 */
Mingming Cao617ba132006-10-11 01:20:53 -07004127 for (block = 0; block < EXT4_N_BLOCKS; block++)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004128 ei->i_data[block] = raw_inode->i_block[block];
4129 INIT_LIST_HEAD(&ei->i_orphan);
4130
Jan Karab436b9b2009-12-08 23:51:10 -05004131 /*
4132 * Set transaction id's of transactions that have to be committed
4133 * to finish f[data]sync. We set them to currently running transaction
4134 * as we cannot be sure that the inode or some of its metadata isn't
4135 * part of the transaction - the inode could have been reclaimed and
4136 * now it is reread from disk.
4137 */
4138 if (journal) {
4139 transaction_t *transaction;
4140 tid_t tid;
4141
Theodore Ts'oa931da62010-08-03 21:35:12 -04004142 read_lock(&journal->j_state_lock);
Jan Karab436b9b2009-12-08 23:51:10 -05004143 if (journal->j_running_transaction)
4144 transaction = journal->j_running_transaction;
4145 else
4146 transaction = journal->j_committing_transaction;
4147 if (transaction)
4148 tid = transaction->t_tid;
4149 else
4150 tid = journal->j_commit_sequence;
Theodore Ts'oa931da62010-08-03 21:35:12 -04004151 read_unlock(&journal->j_state_lock);
Jan Karab436b9b2009-12-08 23:51:10 -05004152 ei->i_sync_tid = tid;
4153 ei->i_datasync_tid = tid;
4154 }
4155
Eric Sandeen0040d982008-02-05 22:36:43 -05004156 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004157 if (ei->i_extra_isize == 0) {
4158 /* The extra space is currently unused. Use it. */
Mingming Cao617ba132006-10-11 01:20:53 -07004159 ei->i_extra_isize = sizeof(struct ext4_inode) -
4160 EXT4_GOOD_OLD_INODE_SIZE;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004161 } else {
Tao Ma152a7b02012-12-02 11:13:24 -05004162 ext4_iget_extra_inode(inode, raw_inode, ei);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004163 }
Darrick J. Wong814525f2012-04-29 18:31:10 -04004164 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004165
Kalpak Shahef7f3832007-07-18 09:15:20 -04004166 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4167 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4168 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4169 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4170
Theodore Ts'oed3654e2014-03-24 14:09:06 -04004171 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
Theodore Ts'oc4f65702014-03-20 00:32:57 -04004172 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4173 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4174 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4175 inode->i_version |=
4176 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4177 }
Jean Noel Cordenner25ec56b2008-01-28 23:58:27 -05004178 }
4179
Theodore Ts'oc4b5a612009-04-24 18:45:35 -04004180 ret = 0;
Theodore Ts'o485c26e2009-04-24 13:43:20 -04004181 if (ei->i_file_acl &&
Theodore Ts'o10329882009-11-15 15:29:56 -05004182 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
Theodore Ts'o24676da2010-05-16 21:00:00 -04004183 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
4184 ei->i_file_acl);
Theodore Ts'o485c26e2009-04-24 13:43:20 -04004185 ret = -EIO;
4186 goto bad_inode;
Tao Maf19d5872012-12-10 14:05:51 -05004187 } else if (!ext4_has_inline_data(inode)) {
4188 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4189 if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4190 (S_ISLNK(inode->i_mode) &&
4191 !ext4_inode_is_fast_symlink(inode))))
4192 /* Validate extent which is part of inode */
4193 ret = ext4_ext_check_inode(inode);
4194 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4195 (S_ISLNK(inode->i_mode) &&
4196 !ext4_inode_is_fast_symlink(inode))) {
4197 /* Validate block references which are part of inode */
4198 ret = ext4_ind_check_inode(inode);
4199 }
Thiemo Nagelfe2c8192009-03-31 08:36:10 -04004200 }
Theodore Ts'o567f3e92009-11-14 08:19:05 -05004201 if (ret)
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04004202 goto bad_inode;
Aneesh Kumar K.V7a262f72009-03-27 16:39:58 -04004203
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004204 if (S_ISREG(inode->i_mode)) {
Mingming Cao617ba132006-10-11 01:20:53 -07004205 inode->i_op = &ext4_file_inode_operations;
Boaz Harroshbe64f882015-04-15 16:15:17 -07004206 inode->i_fop = &ext4_file_operations;
Mingming Cao617ba132006-10-11 01:20:53 -07004207 ext4_set_aops(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004208 } else if (S_ISDIR(inode->i_mode)) {
Mingming Cao617ba132006-10-11 01:20:53 -07004209 inode->i_op = &ext4_dir_inode_operations;
4210 inode->i_fop = &ext4_dir_operations;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004211 } else if (S_ISLNK(inode->i_mode)) {
Al Viroa7a67e82015-04-27 17:51:30 -04004212 if (ext4_encrypted_inode(inode)) {
4213 inode->i_op = &ext4_encrypted_symlink_inode_operations;
4214 ext4_set_aops(inode);
4215 } else if (ext4_inode_is_fast_symlink(inode)) {
Al Viro75e75662015-05-02 10:13:58 -04004216 inode->i_link = (char *)ei->i_data;
Mingming Cao617ba132006-10-11 01:20:53 -07004217 inode->i_op = &ext4_fast_symlink_inode_operations;
Duane Griffine83c1392008-12-19 20:47:15 +00004218 nd_terminate_link(ei->i_data, inode->i_size,
4219 sizeof(ei->i_data) - 1);
4220 } else {
Mingming Cao617ba132006-10-11 01:20:53 -07004221 inode->i_op = &ext4_symlink_inode_operations;
4222 ext4_set_aops(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004223 }
Theodore Ts'o563bdd62009-03-26 00:06:19 -04004224 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4225 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
Mingming Cao617ba132006-10-11 01:20:53 -07004226 inode->i_op = &ext4_special_inode_operations;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004227 if (raw_inode->i_block[0])
4228 init_special_inode(inode, inode->i_mode,
4229 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4230 else
4231 init_special_inode(inode, inode->i_mode,
4232 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
Dr. Tilmann Bubeck393d1d12013-04-08 12:54:05 -04004233 } else if (ino == EXT4_BOOT_LOADER_INO) {
4234 make_bad_inode(inode);
Theodore Ts'o563bdd62009-03-26 00:06:19 -04004235 } else {
Theodore Ts'o563bdd62009-03-26 00:06:19 -04004236 ret = -EIO;
Theodore Ts'o24676da2010-05-16 21:00:00 -04004237 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
Theodore Ts'o563bdd62009-03-26 00:06:19 -04004238 goto bad_inode;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004239 }
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04004240 brelse(iloc.bh);
Mingming Cao617ba132006-10-11 01:20:53 -07004241 ext4_set_inode_flags(inode);
David Howells1d1fe1e2008-02-07 00:15:37 -08004242 unlock_new_inode(inode);
4243 return inode;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004244
4245bad_inode:
Theodore Ts'o567f3e92009-11-14 08:19:05 -05004246 brelse(iloc.bh);
David Howells1d1fe1e2008-02-07 00:15:37 -08004247 iget_failed(inode);
4248 return ERR_PTR(ret);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004249}
4250
Theodore Ts'of4bb2982014-10-05 22:56:00 -04004251struct inode *ext4_iget_normal(struct super_block *sb, unsigned long ino)
4252{
4253 if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
4254 return ERR_PTR(-EIO);
4255 return ext4_iget(sb, ino);
4256}
4257
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004258static int ext4_inode_blocks_set(handle_t *handle,
4259 struct ext4_inode *raw_inode,
4260 struct ext4_inode_info *ei)
4261{
4262 struct inode *inode = &(ei->vfs_inode);
4263 u64 i_blocks = inode->i_blocks;
4264 struct super_block *sb = inode->i_sb;
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004265
4266 if (i_blocks <= ~0U) {
4267 /*
Anatol Pomozov4907cb72012-09-01 10:31:09 -07004268 * i_blocks can be represented in a 32 bit variable
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004269 * as multiple of 512 bytes
4270 */
Aneesh Kumar K.V8180a562008-01-28 23:58:27 -05004271 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004272 raw_inode->i_blocks_high = 0;
Dmitry Monakhov84a8dce2010-06-05 11:51:27 -04004273 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
Theodore Ts'of287a1a2008-10-16 22:50:48 -04004274 return 0;
4275 }
4276 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
4277 return -EFBIG;
4278
4279 if (i_blocks <= 0xffffffffffffULL) {
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004280 /*
4281 * i_blocks can be represented in a 48 bit variable
4282 * as multiple of 512 bytes
4283 */
Aneesh Kumar K.V8180a562008-01-28 23:58:27 -05004284 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004285 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
Dmitry Monakhov84a8dce2010-06-05 11:51:27 -04004286 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004287 } else {
Dmitry Monakhov84a8dce2010-06-05 11:51:27 -04004288 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
Aneesh Kumar K.V8180a562008-01-28 23:58:27 -05004289 /* i_block is stored in file system block size */
4290 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4291 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4292 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004293 }
Theodore Ts'of287a1a2008-10-16 22:50:48 -04004294 return 0;
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004295}
4296
Theodore Ts'oa26f4992015-02-02 00:37:02 -05004297struct other_inode {
4298 unsigned long orig_ino;
4299 struct ext4_inode *raw_inode;
4300};
4301
4302static int other_inode_match(struct inode * inode, unsigned long ino,
4303 void *data)
4304{
4305 struct other_inode *oi = (struct other_inode *) data;
4306
4307 if ((inode->i_ino != ino) ||
4308 (inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
4309 I_DIRTY_SYNC | I_DIRTY_DATASYNC)) ||
4310 ((inode->i_state & I_DIRTY_TIME) == 0))
4311 return 0;
4312 spin_lock(&inode->i_lock);
4313 if (((inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
4314 I_DIRTY_SYNC | I_DIRTY_DATASYNC)) == 0) &&
4315 (inode->i_state & I_DIRTY_TIME)) {
4316 struct ext4_inode_info *ei = EXT4_I(inode);
4317
4318 inode->i_state &= ~(I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED);
4319 spin_unlock(&inode->i_lock);
4320
4321 spin_lock(&ei->i_raw_lock);
4322 EXT4_INODE_SET_XTIME(i_ctime, inode, oi->raw_inode);
4323 EXT4_INODE_SET_XTIME(i_mtime, inode, oi->raw_inode);
4324 EXT4_INODE_SET_XTIME(i_atime, inode, oi->raw_inode);
4325 ext4_inode_csum_set(inode, oi->raw_inode, ei);
4326 spin_unlock(&ei->i_raw_lock);
4327 trace_ext4_other_inode_update_time(inode, oi->orig_ino);
4328 return -1;
4329 }
4330 spin_unlock(&inode->i_lock);
4331 return -1;
4332}
4333
4334/*
4335 * Opportunistically update the other time fields for other inodes in
4336 * the same inode table block.
4337 */
4338static void ext4_update_other_inodes_time(struct super_block *sb,
4339 unsigned long orig_ino, char *buf)
4340{
4341 struct other_inode oi;
4342 unsigned long ino;
4343 int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4344 int inode_size = EXT4_INODE_SIZE(sb);
4345
4346 oi.orig_ino = orig_ino;
Theodore Ts'o8f4d8552015-05-14 18:19:01 -04004347 ino = (orig_ino & ~(inodes_per_block - 1)) + 1;
Theodore Ts'oa26f4992015-02-02 00:37:02 -05004348 for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) {
4349 if (ino == orig_ino)
4350 continue;
4351 oi.raw_inode = (struct ext4_inode *) buf;
4352 (void) find_inode_nowait(sb, ino, other_inode_match, &oi);
4353 }
4354}
4355
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004356/*
4357 * Post the struct inode info into an on-disk inode location in the
4358 * buffer-cache. This gobbles the caller's reference to the
4359 * buffer_head in the inode location struct.
4360 *
4361 * The caller must have write access to iloc->bh.
4362 */
Mingming Cao617ba132006-10-11 01:20:53 -07004363static int ext4_do_update_inode(handle_t *handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004364 struct inode *inode,
Frank Mayhar830156c2009-09-29 10:07:47 -04004365 struct ext4_iloc *iloc)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004366{
Mingming Cao617ba132006-10-11 01:20:53 -07004367 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4368 struct ext4_inode_info *ei = EXT4_I(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004369 struct buffer_head *bh = iloc->bh;
Theodore Ts'o202ee5d2014-04-21 14:37:55 -04004370 struct super_block *sb = inode->i_sb;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004371 int err = 0, rc, block;
Theodore Ts'o202ee5d2014-04-21 14:37:55 -04004372 int need_datasync = 0, set_large_file = 0;
Eric W. Biederman08cefc72012-02-07 15:41:49 -08004373 uid_t i_uid;
4374 gid_t i_gid;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004375
Theodore Ts'o202ee5d2014-04-21 14:37:55 -04004376 spin_lock(&ei->i_raw_lock);
4377
4378 /* For fields not tracked in the in-memory inode,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004379 * initialise them to zero for new inodes. */
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05004380 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
Mingming Cao617ba132006-10-11 01:20:53 -07004381 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004382
Jan Karaff9ddf72007-07-18 09:24:20 -04004383 ext4_get_inode_flags(ei);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004384 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
Eric W. Biederman08cefc72012-02-07 15:41:49 -08004385 i_uid = i_uid_read(inode);
4386 i_gid = i_gid_read(inode);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04004387 if (!(test_opt(inode->i_sb, NO_UID32))) {
Eric W. Biederman08cefc72012-02-07 15:41:49 -08004388 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4389 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004390/*
4391 * Fix up interoperability with old kernels. Otherwise, old inodes get
4392 * re-used with the upper 16 bits of the uid/gid intact
4393 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04004394 if (!ei->i_dtime) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004395 raw_inode->i_uid_high =
Eric W. Biederman08cefc72012-02-07 15:41:49 -08004396 cpu_to_le16(high_16_bits(i_uid));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004397 raw_inode->i_gid_high =
Eric W. Biederman08cefc72012-02-07 15:41:49 -08004398 cpu_to_le16(high_16_bits(i_gid));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004399 } else {
4400 raw_inode->i_uid_high = 0;
4401 raw_inode->i_gid_high = 0;
4402 }
4403 } else {
Eric W. Biederman08cefc72012-02-07 15:41:49 -08004404 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4405 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004406 raw_inode->i_uid_high = 0;
4407 raw_inode->i_gid_high = 0;
4408 }
4409 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
Kalpak Shahef7f3832007-07-18 09:15:20 -04004410
4411 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4412 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4413 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4414 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4415
Li Xibce92d52014-10-01 22:11:06 -04004416 err = ext4_inode_blocks_set(handle, raw_inode, ei);
4417 if (err) {
Theodore Ts'o202ee5d2014-04-21 14:37:55 -04004418 spin_unlock(&ei->i_raw_lock);
Aneesh Kumar K.V0fc1b452008-01-28 23:58:26 -05004419 goto out_brelse;
Theodore Ts'o202ee5d2014-04-21 14:37:55 -04004420 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004421 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
Theodore Ts'o353eb832011-01-10 12:18:25 -05004422 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
Theodore Ts'oed3654e2014-03-24 14:09:06 -04004423 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
Badari Pulavartya1ddeb72006-10-11 01:21:09 -07004424 raw_inode->i_file_acl_high =
4425 cpu_to_le16(ei->i_file_acl >> 32);
Aneesh Kumar K.V7973c0c2008-01-28 23:58:27 -05004426 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
Jan Karab71fc072012-09-26 21:52:20 -04004427 if (ei->i_disksize != ext4_isize(raw_inode)) {
4428 ext4_isize_set(raw_inode, ei->i_disksize);
4429 need_datasync = 1;
4430 }
Aneesh Kumar K.Va48380f2008-01-28 23:58:27 -05004431 if (ei->i_disksize > 0x7fffffffULL) {
Aneesh Kumar K.Va48380f2008-01-28 23:58:27 -05004432 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4433 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4434 EXT4_SB(sb)->s_es->s_rev_level ==
Theodore Ts'o202ee5d2014-04-21 14:37:55 -04004435 cpu_to_le32(EXT4_GOOD_OLD_REV))
4436 set_large_file = 1;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004437 }
4438 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4439 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4440 if (old_valid_dev(inode->i_rdev)) {
4441 raw_inode->i_block[0] =
4442 cpu_to_le32(old_encode_dev(inode->i_rdev));
4443 raw_inode->i_block[1] = 0;
4444 } else {
4445 raw_inode->i_block[0] = 0;
4446 raw_inode->i_block[1] =
4447 cpu_to_le32(new_encode_dev(inode->i_rdev));
4448 raw_inode->i_block[2] = 0;
4449 }
Tao Maf19d5872012-12-10 14:05:51 -05004450 } else if (!ext4_has_inline_data(inode)) {
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04004451 for (block = 0; block < EXT4_N_BLOCKS; block++)
4452 raw_inode->i_block[block] = ei->i_data[block];
Tao Maf19d5872012-12-10 14:05:51 -05004453 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004454
Theodore Ts'oed3654e2014-03-24 14:09:06 -04004455 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
Theodore Ts'oc4f65702014-03-20 00:32:57 -04004456 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4457 if (ei->i_extra_isize) {
4458 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4459 raw_inode->i_version_hi =
4460 cpu_to_le32(inode->i_version >> 32);
4461 raw_inode->i_extra_isize =
4462 cpu_to_le16(ei->i_extra_isize);
4463 }
Jean Noel Cordenner25ec56b2008-01-28 23:58:27 -05004464 }
Darrick J. Wong814525f2012-04-29 18:31:10 -04004465 ext4_inode_csum_set(inode, raw_inode, ei);
Theodore Ts'o202ee5d2014-04-21 14:37:55 -04004466 spin_unlock(&ei->i_raw_lock);
Theodore Ts'oa26f4992015-02-02 00:37:02 -05004467 if (inode->i_sb->s_flags & MS_LAZYTIME)
4468 ext4_update_other_inodes_time(inode->i_sb, inode->i_ino,
4469 bh->b_data);
Theodore Ts'o202ee5d2014-04-21 14:37:55 -04004470
Frank Mayhar830156c2009-09-29 10:07:47 -04004471 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
Curt Wohlgemuth73b50c12010-02-16 15:06:29 -05004472 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
Frank Mayhar830156c2009-09-29 10:07:47 -04004473 if (!err)
4474 err = rc;
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05004475 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
Theodore Ts'o202ee5d2014-04-21 14:37:55 -04004476 if (set_large_file) {
liang xie5d601252014-05-12 22:06:43 -04004477 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
Theodore Ts'o202ee5d2014-04-21 14:37:55 -04004478 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
4479 if (err)
4480 goto out_brelse;
4481 ext4_update_dynamic_rev(sb);
4482 EXT4_SET_RO_COMPAT_FEATURE(sb,
4483 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4484 ext4_handle_sync(handle);
4485 err = ext4_handle_dirty_super(handle, sb);
4486 }
Jan Karab71fc072012-09-26 21:52:20 -04004487 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004488out_brelse:
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04004489 brelse(bh);
Mingming Cao617ba132006-10-11 01:20:53 -07004490 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004491 return err;
4492}
4493
4494/*
Mingming Cao617ba132006-10-11 01:20:53 -07004495 * ext4_write_inode()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004496 *
4497 * We are called from a few places:
4498 *
Theodore Ts'o87f7e412014-04-08 11:38:28 -04004499 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004500 * Here, there will be no transaction running. We wait for any running
Anatol Pomozov4907cb72012-09-01 10:31:09 -07004501 * transaction to commit.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004502 *
Theodore Ts'o87f7e412014-04-08 11:38:28 -04004503 * - Within flush work (sys_sync(), kupdate and such).
4504 * We wait on commit, if told to.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004505 *
Theodore Ts'o87f7e412014-04-08 11:38:28 -04004506 * - Within iput_final() -> write_inode_now()
4507 * We wait on commit, if told to.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004508 *
4509 * In all cases it is actually safe for us to return without doing anything,
4510 * because the inode has been copied into a raw inode buffer in
Theodore Ts'o87f7e412014-04-08 11:38:28 -04004511 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
4512 * writeback.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004513 *
4514 * Note that we are absolutely dependent upon all inode dirtiers doing the
4515 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4516 * which we are interested.
4517 *
4518 * It would be a bug for them to not do this. The code:
4519 *
4520 * mark_inode_dirty(inode)
4521 * stuff();
4522 * inode->i_size = expr;
4523 *
Theodore Ts'o87f7e412014-04-08 11:38:28 -04004524 * is in error because write_inode() could occur while `stuff()' is running,
4525 * and the new i_size will be lost. Plus the inode will no longer be on the
4526 * superblock's dirty inode list.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004527 */
Christoph Hellwiga9185b42010-03-05 09:21:37 +01004528int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004529{
Frank Mayhar91ac6f42009-09-09 22:33:47 -04004530 int err;
4531
Theodore Ts'o87f7e412014-04-08 11:38:28 -04004532 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004533 return 0;
4534
Frank Mayhar91ac6f42009-09-09 22:33:47 -04004535 if (EXT4_SB(inode->i_sb)->s_journal) {
4536 if (ext4_journal_current_handle()) {
4537 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4538 dump_stack();
4539 return -EIO;
4540 }
4541
Jan Kara10542c22014-03-04 10:50:50 -05004542 /*
4543 * No need to force transaction in WB_SYNC_NONE mode. Also
4544 * ext4_sync_fs() will force the commit after everything is
4545 * written.
4546 */
4547 if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
Frank Mayhar91ac6f42009-09-09 22:33:47 -04004548 return 0;
4549
4550 err = ext4_force_commit(inode->i_sb);
4551 } else {
4552 struct ext4_iloc iloc;
4553
Curt Wohlgemuth8b472d72010-04-03 16:45:06 -04004554 err = __ext4_get_inode_loc(inode, &iloc, 0);
Frank Mayhar91ac6f42009-09-09 22:33:47 -04004555 if (err)
4556 return err;
Jan Kara10542c22014-03-04 10:50:50 -05004557 /*
4558 * sync(2) will flush the whole buffer cache. No need to do
4559 * it here separately for each inode.
4560 */
4561 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
Frank Mayhar830156c2009-09-29 10:07:47 -04004562 sync_dirty_buffer(iloc.bh);
4563 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
Theodore Ts'oc398eda2010-07-27 11:56:40 -04004564 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
4565 "IO error syncing inode");
Frank Mayhar830156c2009-09-29 10:07:47 -04004566 err = -EIO;
4567 }
Curt Wohlgemuthfd2dd9f2010-04-03 17:44:16 -04004568 brelse(iloc.bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004569 }
Frank Mayhar91ac6f42009-09-09 22:33:47 -04004570 return err;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004571}
4572
4573/*
Jan Kara53e87262012-12-25 13:29:52 -05004574 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4575 * buffers that are attached to a page stradding i_size and are undergoing
4576 * commit. In that case we have to wait for commit to finish and try again.
4577 */
4578static void ext4_wait_for_tail_page_commit(struct inode *inode)
4579{
4580 struct page *page;
4581 unsigned offset;
4582 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
4583 tid_t commit_tid = 0;
4584 int ret;
4585
4586 offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
4587 /*
4588 * All buffers in the last page remain valid? Then there's nothing to
4589 * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE ==
4590 * blocksize case
4591 */
4592 if (offset > PAGE_CACHE_SIZE - (1 << inode->i_blkbits))
4593 return;
4594 while (1) {
4595 page = find_lock_page(inode->i_mapping,
4596 inode->i_size >> PAGE_CACHE_SHIFT);
4597 if (!page)
4598 return;
Lukas Czernerca99fdd2013-05-21 23:25:01 -04004599 ret = __ext4_journalled_invalidatepage(page, offset,
4600 PAGE_CACHE_SIZE - offset);
Jan Kara53e87262012-12-25 13:29:52 -05004601 unlock_page(page);
4602 page_cache_release(page);
4603 if (ret != -EBUSY)
4604 return;
4605 commit_tid = 0;
4606 read_lock(&journal->j_state_lock);
4607 if (journal->j_committing_transaction)
4608 commit_tid = journal->j_committing_transaction->t_tid;
4609 read_unlock(&journal->j_state_lock);
4610 if (commit_tid)
4611 jbd2_log_wait_commit(journal, commit_tid);
4612 }
4613}
4614
4615/*
Mingming Cao617ba132006-10-11 01:20:53 -07004616 * ext4_setattr()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004617 *
4618 * Called from notify_change.
4619 *
4620 * We want to trap VFS attempts to truncate the file as soon as
4621 * possible. In particular, we want to make sure that when the VFS
4622 * shrinks i_size, we put the inode on the orphan list and modify
4623 * i_disksize immediately, so that during the subsequent flushing of
4624 * dirty pages and freeing of disk blocks, we can guarantee that any
4625 * commit will leave the blocks being flushed in an unused state on
4626 * disk. (On recovery, the inode will get truncated and the blocks will
4627 * be freed, so we have a strong guarantee that no future commit will
4628 * leave these blocks visible to the user.)
4629 *
Jan Kara678aaf42008-07-11 19:27:31 -04004630 * Another thing we have to assure is that if we are in ordered mode
4631 * and inode is still attached to the committing transaction, we must
4632 * we start writeout of all the dirty pages which are being truncated.
4633 * This way we are sure that all the data written in the previous
4634 * transaction are already on disk (truncate waits for pages under
4635 * writeback).
4636 *
4637 * Called with inode->i_mutex down.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004638 */
Mingming Cao617ba132006-10-11 01:20:53 -07004639int ext4_setattr(struct dentry *dentry, struct iattr *attr)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004640{
David Howells2b0143b2015-03-17 22:25:59 +00004641 struct inode *inode = d_inode(dentry);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004642 int error, rc = 0;
Dmitry Monakhov3d287de2010-10-27 22:08:46 -04004643 int orphan = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004644 const unsigned int ia_valid = attr->ia_valid;
4645
4646 error = inode_change_ok(inode, attr);
4647 if (error)
4648 return error;
4649
Dmitry Monakhov12755622010-04-08 22:04:20 +04004650 if (is_quota_modification(inode, attr))
Christoph Hellwig871a2932010-03-03 09:05:07 -05004651 dquot_initialize(inode);
Eric W. Biederman08cefc72012-02-07 15:41:49 -08004652 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
4653 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004654 handle_t *handle;
4655
4656 /* (user+group)*(old+new) structure, inode write (sb,
4657 * inode block, ? - but truncate inode update has it) */
Theodore Ts'o9924a922013-02-08 21:59:22 -05004658 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
4659 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
4660 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004661 if (IS_ERR(handle)) {
4662 error = PTR_ERR(handle);
4663 goto err_out;
4664 }
Christoph Hellwigb43fa822010-03-03 09:05:03 -05004665 error = dquot_transfer(inode, attr);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004666 if (error) {
Mingming Cao617ba132006-10-11 01:20:53 -07004667 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004668 return error;
4669 }
4670 /* Update corresponding info in inode so that everything is in
4671 * one transaction */
4672 if (attr->ia_valid & ATTR_UID)
4673 inode->i_uid = attr->ia_uid;
4674 if (attr->ia_valid & ATTR_GID)
4675 inode->i_gid = attr->ia_gid;
Mingming Cao617ba132006-10-11 01:20:53 -07004676 error = ext4_mark_inode_dirty(handle, inode);
4677 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004678 }
4679
Josef Bacik3da40c72015-06-22 00:31:26 -04004680 if (attr->ia_valid & ATTR_SIZE) {
Jan Kara52083862013-08-17 10:07:17 -04004681 handle_t *handle;
Josef Bacik3da40c72015-06-22 00:31:26 -04004682 loff_t oldsize = inode->i_size;
4683 int shrink = (attr->ia_size <= inode->i_size);
Christoph Hellwig562c72a2011-06-24 14:29:45 -04004684
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04004685 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
Eric Sandeene2b46572008-01-28 23:58:27 -05004686 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4687
Theodore Ts'o0c095c72010-07-27 11:56:06 -04004688 if (attr->ia_size > sbi->s_bitmap_maxbytes)
4689 return -EFBIG;
Eric Sandeene2b46572008-01-28 23:58:27 -05004690 }
Josef Bacik3da40c72015-06-22 00:31:26 -04004691 if (!S_ISREG(inode->i_mode))
4692 return -EINVAL;
Christoph Hellwigdff6efc2013-11-19 07:17:07 -08004693
4694 if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
4695 inode_inc_iversion(inode);
4696
Josef Bacik3da40c72015-06-22 00:31:26 -04004697 if (ext4_should_order_data(inode) &&
Jan Kara52083862013-08-17 10:07:17 -04004698 (attr->ia_size < inode->i_size)) {
Josef Bacik3da40c72015-06-22 00:31:26 -04004699 error = ext4_begin_ordered_truncate(inode,
Jan Kara678aaf42008-07-11 19:27:31 -04004700 attr->ia_size);
Josef Bacik3da40c72015-06-22 00:31:26 -04004701 if (error)
4702 goto err_out;
4703 }
4704 if (attr->ia_size != inode->i_size) {
Jan Kara52083862013-08-17 10:07:17 -04004705 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
4706 if (IS_ERR(handle)) {
4707 error = PTR_ERR(handle);
4708 goto err_out;
4709 }
Josef Bacik3da40c72015-06-22 00:31:26 -04004710 if (ext4_handle_valid(handle) && shrink) {
Jan Kara52083862013-08-17 10:07:17 -04004711 error = ext4_orphan_add(handle, inode);
4712 orphan = 1;
4713 }
Jan Kara90e775b2013-08-17 10:09:31 -04004714 down_write(&EXT4_I(inode)->i_data_sem);
Jan Kara52083862013-08-17 10:07:17 -04004715 EXT4_I(inode)->i_disksize = attr->ia_size;
4716 rc = ext4_mark_inode_dirty(handle, inode);
4717 if (!error)
4718 error = rc;
Jan Kara90e775b2013-08-17 10:09:31 -04004719 /*
4720 * We have to update i_size under i_data_sem together
4721 * with i_disksize to avoid races with writeback code
4722 * running ext4_wb_update_i_disksize().
4723 */
4724 if (!error)
4725 i_size_write(inode, attr->ia_size);
4726 up_write(&EXT4_I(inode)->i_data_sem);
Jan Kara52083862013-08-17 10:07:17 -04004727 ext4_journal_stop(handle);
4728 if (error) {
Josef Bacik3da40c72015-06-22 00:31:26 -04004729 if (orphan)
4730 ext4_orphan_del(NULL, inode);
Jan Kara678aaf42008-07-11 19:27:31 -04004731 goto err_out;
4732 }
Jan Karad6320cb2014-10-01 21:49:46 -04004733 }
Josef Bacik3da40c72015-06-22 00:31:26 -04004734 if (!shrink)
4735 pagecache_isize_extended(inode, oldsize, inode->i_size);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004736
Jan Kara52083862013-08-17 10:07:17 -04004737 /*
4738 * Blocks are going to be removed from the inode. Wait
4739 * for dio in flight. Temporarily disable
4740 * dioread_nolock to prevent livelock.
4741 */
4742 if (orphan) {
4743 if (!ext4_should_journal_data(inode)) {
4744 ext4_inode_block_unlocked_dio(inode);
4745 inode_dio_wait(inode);
4746 ext4_inode_resume_unlocked_dio(inode);
4747 } else
4748 ext4_wait_for_tail_page_commit(inode);
Dmitry Monakhov1c9114f2012-09-29 00:55:23 -04004749 }
Jan Kara52083862013-08-17 10:07:17 -04004750 /*
4751 * Truncate pagecache after we've waited for commit
4752 * in data=journal mode to make pages freeable.
4753 */
Ross Zwisler923ae0f2015-02-16 15:59:38 -08004754 truncate_pagecache(inode, inode->i_size);
Josef Bacik3da40c72015-06-22 00:31:26 -04004755 if (shrink)
4756 ext4_truncate(inode);
Theodore Ts'o072bd7e2011-05-23 15:13:02 -04004757 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004758
Christoph Hellwig10257742010-06-04 11:30:02 +02004759 if (!rc) {
4760 setattr_copy(inode, attr);
4761 mark_inode_dirty(inode);
4762 }
4763
4764 /*
4765 * If the call to ext4_truncate failed to get a transaction handle at
4766 * all, we need to clean up the in-core orphan list manually.
4767 */
Dmitry Monakhov3d287de2010-10-27 22:08:46 -04004768 if (orphan && inode->i_nlink)
Mingming Cao617ba132006-10-11 01:20:53 -07004769 ext4_orphan_del(NULL, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004770
4771 if (!rc && (ia_valid & ATTR_MODE))
Christoph Hellwig64e178a2013-12-20 05:16:44 -08004772 rc = posix_acl_chmod(inode, inode->i_mode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004773
4774err_out:
Mingming Cao617ba132006-10-11 01:20:53 -07004775 ext4_std_error(inode->i_sb, error);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004776 if (!error)
4777 error = rc;
4778 return error;
4779}
4780
Mingming Cao3e3398a2008-07-11 19:27:31 -04004781int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4782 struct kstat *stat)
4783{
4784 struct inode *inode;
Jan Kara8af8eec2013-05-31 19:39:56 -04004785 unsigned long long delalloc_blocks;
Mingming Cao3e3398a2008-07-11 19:27:31 -04004786
David Howells2b0143b2015-03-17 22:25:59 +00004787 inode = d_inode(dentry);
Mingming Cao3e3398a2008-07-11 19:27:31 -04004788 generic_fillattr(inode, stat);
4789
4790 /*
Andreas Dilger9206c562013-11-11 22:38:12 -05004791 * If there is inline data in the inode, the inode will normally not
4792 * have data blocks allocated (it may have an external xattr block).
4793 * Report at least one sector for such files, so tools like tar, rsync,
4794 * others doen't incorrectly think the file is completely sparse.
4795 */
4796 if (unlikely(ext4_has_inline_data(inode)))
4797 stat->blocks += (stat->size + 511) >> 9;
4798
4799 /*
Mingming Cao3e3398a2008-07-11 19:27:31 -04004800 * We can't update i_blocks if the block allocation is delayed
4801 * otherwise in the case of system crash before the real block
4802 * allocation is done, we will have i_blocks inconsistent with
4803 * on-disk file blocks.
4804 * We always keep i_blocks updated together with real
4805 * allocation. But to not confuse with user, stat
4806 * will return the blocks that include the delayed allocation
4807 * blocks for this file.
4808 */
Tao Ma96607552012-05-31 22:54:16 -04004809 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
Andreas Dilger9206c562013-11-11 22:38:12 -05004810 EXT4_I(inode)->i_reserved_data_blocks);
4811 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
Mingming Cao3e3398a2008-07-11 19:27:31 -04004812 return 0;
4813}
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004814
Jan Karafffb2732013-06-04 13:01:11 -04004815static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
4816 int pextents)
Mingming Caoa02908f2008-08-19 22:16:07 -04004817{
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04004818 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
Jan Karafffb2732013-06-04 13:01:11 -04004819 return ext4_ind_trans_blocks(inode, lblocks);
4820 return ext4_ext_index_trans_blocks(inode, pextents);
Mingming Caoa02908f2008-08-19 22:16:07 -04004821}
Theodore Ts'oac51d832008-11-06 16:49:36 -05004822
Mingming Caoa02908f2008-08-19 22:16:07 -04004823/*
4824 * Account for index blocks, block groups bitmaps and block group
4825 * descriptor blocks if modify datablocks and index blocks
4826 * worse case, the indexs blocks spread over different block groups
4827 *
4828 * If datablocks are discontiguous, they are possible to spread over
Anatol Pomozov4907cb72012-09-01 10:31:09 -07004829 * different block groups too. If they are contiguous, with flexbg,
Mingming Caoa02908f2008-08-19 22:16:07 -04004830 * they could still across block group boundary.
4831 *
4832 * Also account for superblock, inode, quota and xattr blocks
4833 */
Jan Karafffb2732013-06-04 13:01:11 -04004834static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
4835 int pextents)
Mingming Caoa02908f2008-08-19 22:16:07 -04004836{
Theodore Ts'o8df96752009-05-01 08:50:38 -04004837 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
4838 int gdpblocks;
Mingming Caoa02908f2008-08-19 22:16:07 -04004839 int idxblocks;
4840 int ret = 0;
4841
4842 /*
Jan Karafffb2732013-06-04 13:01:11 -04004843 * How many index blocks need to touch to map @lblocks logical blocks
4844 * to @pextents physical extents?
Mingming Caoa02908f2008-08-19 22:16:07 -04004845 */
Jan Karafffb2732013-06-04 13:01:11 -04004846 idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
Mingming Caoa02908f2008-08-19 22:16:07 -04004847
4848 ret = idxblocks;
4849
4850 /*
4851 * Now let's see how many group bitmaps and group descriptors need
4852 * to account
4853 */
Jan Karafffb2732013-06-04 13:01:11 -04004854 groups = idxblocks + pextents;
Mingming Caoa02908f2008-08-19 22:16:07 -04004855 gdpblocks = groups;
Theodore Ts'o8df96752009-05-01 08:50:38 -04004856 if (groups > ngroups)
4857 groups = ngroups;
Mingming Caoa02908f2008-08-19 22:16:07 -04004858 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4859 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4860
4861 /* bitmaps and block group descriptor blocks */
4862 ret += groups + gdpblocks;
4863
4864 /* Blocks for super block, inode, quota and xattr blocks */
4865 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004866
4867 return ret;
4868}
4869
4870/*
Lucas De Marchi25985ed2011-03-30 22:57:33 -03004871 * Calculate the total number of credits to reserve to fit
Mingming Caof3bd1f32008-08-19 22:16:03 -04004872 * the modification of a single pages into a single transaction,
4873 * which may include multiple chunks of block allocations.
Mingming Caoa02908f2008-08-19 22:16:07 -04004874 *
Mingming Cao525f4ed2008-08-19 22:15:58 -04004875 * This could be called via ext4_write_begin()
Mingming Caoa02908f2008-08-19 22:16:07 -04004876 *
Mingming Cao525f4ed2008-08-19 22:15:58 -04004877 * We need to consider the worse case, when
Mingming Caoa02908f2008-08-19 22:16:07 -04004878 * one new block per extent.
Mingming Caoa02908f2008-08-19 22:16:07 -04004879 */
4880int ext4_writepage_trans_blocks(struct inode *inode)
4881{
4882 int bpp = ext4_journal_blocks_per_page(inode);
4883 int ret;
4884
Jan Karafffb2732013-06-04 13:01:11 -04004885 ret = ext4_meta_trans_blocks(inode, bpp, bpp);
Mingming Caoa02908f2008-08-19 22:16:07 -04004886
4887 /* Account for data blocks for journalled mode */
4888 if (ext4_should_journal_data(inode))
4889 ret += bpp;
4890 return ret;
4891}
Mingming Caof3bd1f32008-08-19 22:16:03 -04004892
4893/*
4894 * Calculate the journal credits for a chunk of data modification.
4895 *
4896 * This is called from DIO, fallocate or whoever calling
Eric Sandeen79e83032010-07-27 11:56:07 -04004897 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
Mingming Caof3bd1f32008-08-19 22:16:03 -04004898 *
4899 * journal buffers for data blocks are not included here, as DIO
4900 * and fallocate do no need to journal data buffers.
4901 */
4902int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4903{
4904 return ext4_meta_trans_blocks(inode, nrblocks, 1);
4905}
4906
Mingming Caoa02908f2008-08-19 22:16:07 -04004907/*
Mingming Cao617ba132006-10-11 01:20:53 -07004908 * The caller must have previously called ext4_reserve_inode_write().
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004909 * Give this, we know that the caller already has write access to iloc->bh.
4910 */
Mingming Cao617ba132006-10-11 01:20:53 -07004911int ext4_mark_iloc_dirty(handle_t *handle,
Theodore Ts'ode9a55b2009-06-14 17:45:34 -04004912 struct inode *inode, struct ext4_iloc *iloc)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004913{
4914 int err = 0;
4915
Theodore Ts'oc64db502012-03-02 12:23:11 -05004916 if (IS_I_VERSION(inode))
Jean Noel Cordenner25ec56b2008-01-28 23:58:27 -05004917 inode_inc_iversion(inode);
4918
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004919 /* the do_update_inode consumes one bh->b_count */
4920 get_bh(iloc->bh);
4921
Mingming Caodab291a2006-10-11 01:21:01 -07004922 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
Frank Mayhar830156c2009-09-29 10:07:47 -04004923 err = ext4_do_update_inode(handle, inode, iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004924 put_bh(iloc->bh);
4925 return err;
4926}
4927
4928/*
4929 * On success, We end up with an outstanding reference count against
4930 * iloc->bh. This _must_ be cleaned up later.
4931 */
4932
4933int
Mingming Cao617ba132006-10-11 01:20:53 -07004934ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4935 struct ext4_iloc *iloc)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004936{
Frank Mayhar03901312009-01-07 00:06:22 -05004937 int err;
4938
4939 err = ext4_get_inode_loc(inode, iloc);
4940 if (!err) {
4941 BUFFER_TRACE(iloc->bh, "get_write_access");
4942 err = ext4_journal_get_write_access(handle, iloc->bh);
4943 if (err) {
4944 brelse(iloc->bh);
4945 iloc->bh = NULL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004946 }
4947 }
Mingming Cao617ba132006-10-11 01:20:53 -07004948 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004949 return err;
4950}
4951
4952/*
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04004953 * Expand an inode by new_extra_isize bytes.
4954 * Returns 0 on success or negative error number on failure.
4955 */
Aneesh Kumar K.V1d03ec92008-01-28 23:58:27 -05004956static int ext4_expand_extra_isize(struct inode *inode,
4957 unsigned int new_extra_isize,
4958 struct ext4_iloc iloc,
4959 handle_t *handle)
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04004960{
4961 struct ext4_inode *raw_inode;
4962 struct ext4_xattr_ibody_header *header;
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04004963
4964 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4965 return 0;
4966
4967 raw_inode = ext4_raw_inode(&iloc);
4968
4969 header = IHDR(inode, raw_inode);
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04004970
4971 /* No extended attributes present */
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05004972 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
4973 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04004974 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
4975 new_extra_isize);
4976 EXT4_I(inode)->i_extra_isize = new_extra_isize;
4977 return 0;
4978 }
4979
4980 /* try to expand with EAs present */
4981 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
4982 raw_inode, handle);
4983}
4984
4985/*
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004986 * What we do here is to mark the in-core inode as clean with respect to inode
4987 * dirtiness (it may still be data-dirty).
4988 * This means that the in-core inode may be reaped by prune_icache
4989 * without having to perform any I/O. This is a very good thing,
4990 * because *any* task may call prune_icache - even ones which
4991 * have a transaction open against a different journal.
4992 *
4993 * Is this cheating? Not really. Sure, we haven't written the
4994 * inode out, but prune_icache isn't a user-visible syncing function.
4995 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4996 * we start and wait on commits.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004997 */
Mingming Cao617ba132006-10-11 01:20:53 -07004998int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07004999{
Mingming Cao617ba132006-10-11 01:20:53 -07005000 struct ext4_iloc iloc;
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04005001 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5002 static unsigned int mnt_count;
5003 int err, ret;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005004
5005 might_sleep();
Theodore Ts'o7ff9c072010-11-08 13:51:33 -05005006 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
Mingming Cao617ba132006-10-11 01:20:53 -07005007 err = ext4_reserve_inode_write(handle, inode, &iloc);
Frank Mayhar03901312009-01-07 00:06:22 -05005008 if (ext4_handle_valid(handle) &&
5009 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05005010 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04005011 /*
5012 * We need extra buffer credits since we may write into EA block
5013 * with this same handle. If journal_extend fails, then it will
5014 * only result in a minor loss of functionality for that inode.
5015 * If this is felt to be critical, then e2fsck should be run to
5016 * force a large enough s_min_extra_isize.
5017 */
5018 if ((jbd2_journal_extend(handle,
5019 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
5020 ret = ext4_expand_extra_isize(inode,
5021 sbi->s_want_extra_isize,
5022 iloc, handle);
5023 if (ret) {
Theodore Ts'o19f5fb72010-01-24 14:34:07 -05005024 ext4_set_inode_state(inode,
5025 EXT4_STATE_NO_EXPAND);
Aneesh Kumar K.Vc1bddad2007-10-16 18:38:25 -04005026 if (mnt_count !=
5027 le16_to_cpu(sbi->s_es->s_mnt_count)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05005028 ext4_warning(inode->i_sb,
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04005029 "Unable to expand inode %lu. Delete"
5030 " some EAs or run e2fsck.",
5031 inode->i_ino);
Aneesh Kumar K.Vc1bddad2007-10-16 18:38:25 -04005032 mnt_count =
5033 le16_to_cpu(sbi->s_es->s_mnt_count);
Kalpak Shah6dd4ee72007-07-18 09:19:57 -04005034 }
5035 }
5036 }
5037 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005038 if (!err)
Mingming Cao617ba132006-10-11 01:20:53 -07005039 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005040 return err;
5041}
5042
5043/*
Mingming Cao617ba132006-10-11 01:20:53 -07005044 * ext4_dirty_inode() is called from __mark_inode_dirty()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005045 *
5046 * We're really interested in the case where a file is being extended.
5047 * i_size has been changed by generic_commit_write() and we thus need
5048 * to include the updated inode in the current transaction.
5049 *
Christoph Hellwig5dd40562010-03-03 09:05:00 -05005050 * Also, dquot_alloc_block() will always dirty the inode when blocks
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005051 * are allocated to the file.
5052 *
5053 * If the inode is marked synchronous, we don't honour that here - doing
5054 * so would cause a commit on atime updates, which we don't bother doing.
5055 * We handle synchronous inodes at the highest possible level.
Theodore Ts'o0ae45f62015-02-02 00:37:00 -05005056 *
5057 * If only the I_DIRTY_TIME flag is set, we can skip everything. If
5058 * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need
5059 * to copy into the on-disk inode structure are the timestamp files.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005060 */
Christoph Hellwigaa385722011-05-27 06:53:02 -04005061void ext4_dirty_inode(struct inode *inode, int flags)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005062{
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005063 handle_t *handle;
5064
Theodore Ts'o0ae45f62015-02-02 00:37:00 -05005065 if (flags == I_DIRTY_TIME)
5066 return;
Theodore Ts'o9924a922013-02-08 21:59:22 -05005067 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005068 if (IS_ERR(handle))
5069 goto out;
Curt Wohlgemuthf3dc2722009-09-29 16:06:01 -04005070
Curt Wohlgemuthf3dc2722009-09-29 16:06:01 -04005071 ext4_mark_inode_dirty(handle, inode);
5072
Mingming Cao617ba132006-10-11 01:20:53 -07005073 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005074out:
5075 return;
5076}
5077
5078#if 0
5079/*
5080 * Bind an inode's backing buffer_head into this transaction, to prevent
5081 * it from being flushed to disk early. Unlike
Mingming Cao617ba132006-10-11 01:20:53 -07005082 * ext4_reserve_inode_write, this leaves behind no bh reference and
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005083 * returns no iloc structure, so the caller needs to repeat the iloc
5084 * lookup to mark the inode dirty later.
5085 */
Mingming Cao617ba132006-10-11 01:20:53 -07005086static int ext4_pin_inode(handle_t *handle, struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005087{
Mingming Cao617ba132006-10-11 01:20:53 -07005088 struct ext4_iloc iloc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005089
5090 int err = 0;
5091 if (handle) {
Mingming Cao617ba132006-10-11 01:20:53 -07005092 err = ext4_get_inode_loc(inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005093 if (!err) {
5094 BUFFER_TRACE(iloc.bh, "get_write_access");
Mingming Caodab291a2006-10-11 01:21:01 -07005095 err = jbd2_journal_get_write_access(handle, iloc.bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005096 if (!err)
Frank Mayhar03901312009-01-07 00:06:22 -05005097 err = ext4_handle_dirty_metadata(handle,
Curt Wohlgemuth73b50c12010-02-16 15:06:29 -05005098 NULL,
Frank Mayhar03901312009-01-07 00:06:22 -05005099 iloc.bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005100 brelse(iloc.bh);
5101 }
5102 }
Mingming Cao617ba132006-10-11 01:20:53 -07005103 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005104 return err;
5105}
5106#endif
5107
Mingming Cao617ba132006-10-11 01:20:53 -07005108int ext4_change_inode_journal_flag(struct inode *inode, int val)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005109{
5110 journal_t *journal;
5111 handle_t *handle;
5112 int err;
5113
5114 /*
5115 * We have to be very careful here: changing a data block's
5116 * journaling status dynamically is dangerous. If we write a
5117 * data block to the journal, change the status and then delete
5118 * that block, we risk forgetting to revoke the old log record
5119 * from the journal and so a subsequent replay can corrupt data.
5120 * So, first we make sure that the journal is empty and that
5121 * nobody is changing anything.
5122 */
5123
Mingming Cao617ba132006-10-11 01:20:53 -07005124 journal = EXT4_JOURNAL(inode);
Frank Mayhar03901312009-01-07 00:06:22 -05005125 if (!journal)
5126 return 0;
Dave Hansend6995942007-07-18 08:33:51 -04005127 if (is_journal_aborted(journal))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005128 return -EROFS;
Yongqiang Yang2aff57b2011-12-28 12:02:13 -05005129 /* We have to allocate physical blocks for delalloc blocks
5130 * before flushing journal. otherwise delalloc blocks can not
5131 * be allocated any more. even more truncate on delalloc blocks
5132 * could trigger BUG by flushing delalloc blocks in journal.
5133 * There is no delalloc block in non-journal data mode.
5134 */
5135 if (val && test_opt(inode->i_sb, DELALLOC)) {
5136 err = ext4_alloc_da_blocks(inode);
5137 if (err < 0)
5138 return err;
5139 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005140
Dmitry Monakhov17335dc2012-09-29 00:41:21 -04005141 /* Wait for all existing dio workers */
5142 ext4_inode_block_unlocked_dio(inode);
5143 inode_dio_wait(inode);
5144
Mingming Caodab291a2006-10-11 01:21:01 -07005145 jbd2_journal_lock_updates(journal);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005146
5147 /*
5148 * OK, there are no updates running now, and all cached data is
5149 * synced to disk. We are now in a completely consistent state
5150 * which doesn't have anything in the journal, and we know that
5151 * no filesystem updates are running, so it is safe to modify
5152 * the inode's in-core data-journaling state flag now.
5153 */
5154
5155 if (val)
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04005156 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
Yongqiang Yang5872dda2011-12-28 13:55:51 -05005157 else {
Jan Kara4f879ca2014-10-30 10:53:17 -04005158 err = jbd2_journal_flush(journal);
5159 if (err < 0) {
5160 jbd2_journal_unlock_updates(journal);
5161 ext4_inode_resume_unlocked_dio(inode);
5162 return err;
5163 }
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04005164 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
Yongqiang Yang5872dda2011-12-28 13:55:51 -05005165 }
Mingming Cao617ba132006-10-11 01:20:53 -07005166 ext4_set_aops(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005167
Mingming Caodab291a2006-10-11 01:21:01 -07005168 jbd2_journal_unlock_updates(journal);
Dmitry Monakhov17335dc2012-09-29 00:41:21 -04005169 ext4_inode_resume_unlocked_dio(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005170
5171 /* Finally we can mark the inode as dirty. */
5172
Theodore Ts'o9924a922013-02-08 21:59:22 -05005173 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005174 if (IS_ERR(handle))
5175 return PTR_ERR(handle);
5176
Mingming Cao617ba132006-10-11 01:20:53 -07005177 err = ext4_mark_inode_dirty(handle, inode);
Frank Mayhar03901312009-01-07 00:06:22 -05005178 ext4_handle_sync(handle);
Mingming Cao617ba132006-10-11 01:20:53 -07005179 ext4_journal_stop(handle);
5180 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07005181
5182 return err;
5183}
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005184
5185static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5186{
5187 return !buffer_mapped(bh);
5188}
5189
Nick Pigginc2ec1752009-03-31 15:23:21 -07005190int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005191{
Nick Pigginc2ec1752009-03-31 15:23:21 -07005192 struct page *page = vmf->page;
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005193 loff_t size;
5194 unsigned long len;
Jan Kara9ea7df52011-06-24 14:29:41 -04005195 int ret;
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005196 struct file *file = vma->vm_file;
Al Viro496ad9a2013-01-23 17:07:38 -05005197 struct inode *inode = file_inode(file);
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005198 struct address_space *mapping = inode->i_mapping;
Jan Kara9ea7df52011-06-24 14:29:41 -04005199 handle_t *handle;
5200 get_block_t *get_block;
5201 int retries = 0;
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005202
Jan Kara8e8ad8a2012-06-12 16:20:38 +02005203 sb_start_pagefault(inode->i_sb);
Theodore Ts'o041bbb6d2012-09-30 23:04:56 -04005204 file_update_time(vma->vm_file);
Jan Kara9ea7df52011-06-24 14:29:41 -04005205 /* Delalloc case is easy... */
5206 if (test_opt(inode->i_sb, DELALLOC) &&
5207 !ext4_should_journal_data(inode) &&
5208 !ext4_nonda_switch(inode->i_sb)) {
5209 do {
5210 ret = __block_page_mkwrite(vma, vmf,
5211 ext4_da_get_block_prep);
5212 } while (ret == -ENOSPC &&
5213 ext4_should_retry_alloc(inode->i_sb, &retries));
5214 goto out_ret;
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005215 }
Darrick J. Wong0e499892011-05-18 13:55:20 -04005216
5217 lock_page(page);
Jan Kara9ea7df52011-06-24 14:29:41 -04005218 size = i_size_read(inode);
5219 /* Page got truncated from under us? */
5220 if (page->mapping != mapping || page_offset(page) > size) {
5221 unlock_page(page);
5222 ret = VM_FAULT_NOPAGE;
5223 goto out;
Darrick J. Wong0e499892011-05-18 13:55:20 -04005224 }
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005225
5226 if (page->index == size >> PAGE_CACHE_SHIFT)
5227 len = size & ~PAGE_CACHE_MASK;
5228 else
5229 len = PAGE_CACHE_SIZE;
Aneesh Kumar K.Va827eaf2009-09-09 22:36:03 -04005230 /*
Jan Kara9ea7df52011-06-24 14:29:41 -04005231 * Return if we have all the buffers mapped. This avoids the need to do
5232 * journal_start/journal_stop which can block and take a long time
Aneesh Kumar K.Va827eaf2009-09-09 22:36:03 -04005233 */
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005234 if (page_has_buffers(page)) {
Tao Maf19d5872012-12-10 14:05:51 -05005235 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
5236 0, len, NULL,
5237 ext4_bh_unmapped)) {
Jan Kara9ea7df52011-06-24 14:29:41 -04005238 /* Wait so that we don't change page under IO */
Darrick J. Wong1d1d1a72013-02-21 16:42:51 -08005239 wait_for_stable_page(page);
Jan Kara9ea7df52011-06-24 14:29:41 -04005240 ret = VM_FAULT_LOCKED;
5241 goto out;
Aneesh Kumar K.Va827eaf2009-09-09 22:36:03 -04005242 }
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005243 }
Aneesh Kumar K.Va827eaf2009-09-09 22:36:03 -04005244 unlock_page(page);
Jan Kara9ea7df52011-06-24 14:29:41 -04005245 /* OK, we need to fill the hole... */
5246 if (ext4_should_dioread_nolock(inode))
5247 get_block = ext4_get_block_write;
5248 else
5249 get_block = ext4_get_block;
5250retry_alloc:
Theodore Ts'o9924a922013-02-08 21:59:22 -05005251 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
5252 ext4_writepage_trans_blocks(inode));
Jan Kara9ea7df52011-06-24 14:29:41 -04005253 if (IS_ERR(handle)) {
Nick Pigginc2ec1752009-03-31 15:23:21 -07005254 ret = VM_FAULT_SIGBUS;
Jan Kara9ea7df52011-06-24 14:29:41 -04005255 goto out;
5256 }
5257 ret = __block_page_mkwrite(vma, vmf, get_block);
5258 if (!ret && ext4_should_journal_data(inode)) {
Tao Maf19d5872012-12-10 14:05:51 -05005259 if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
Jan Kara9ea7df52011-06-24 14:29:41 -04005260 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) {
5261 unlock_page(page);
5262 ret = VM_FAULT_SIGBUS;
Yongqiang Yangfcbb5512011-10-26 05:00:19 -04005263 ext4_journal_stop(handle);
Jan Kara9ea7df52011-06-24 14:29:41 -04005264 goto out;
5265 }
5266 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
5267 }
5268 ext4_journal_stop(handle);
5269 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
5270 goto retry_alloc;
5271out_ret:
5272 ret = block_page_mkwrite_return(ret);
5273out:
Jan Kara8e8ad8a2012-06-12 16:20:38 +02005274 sb_end_pagefault(inode->i_sb);
Aneesh Kumar K.V2e9ee852008-07-11 19:27:31 -04005275 return ret;
5276}