]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame - fs/ext4/inode.c
ext4: enable block_validity by default
[mirror_ubuntu-artful-kernel.git] / fs / ext4 / inode.c
CommitLineData
ac27a0ec 1/*
617ba13b 2 * linux/fs/ext4/inode.c
ac27a0ec
DK
3 *
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
8 *
9 * from
10 *
11 * linux/fs/minix/inode.c
12 *
13 * Copyright (C) 1991, 1992 Linus Torvalds
14 *
ac27a0ec
DK
15 * 64-bit file support on 64-bit platforms by Jakub Jelinek
16 * (jj@sunsite.ms.mff.cuni.cz)
17 *
617ba13b 18 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
ac27a0ec
DK
19 */
20
ac27a0ec
DK
21#include <linux/fs.h>
22#include <linux/time.h>
dab291af 23#include <linux/jbd2.h>
ac27a0ec
DK
24#include <linux/highuid.h>
25#include <linux/pagemap.h>
26#include <linux/quotaops.h>
27#include <linux/string.h>
28#include <linux/buffer_head.h>
29#include <linux/writeback.h>
64769240 30#include <linux/pagevec.h>
ac27a0ec 31#include <linux/mpage.h>
e83c1397 32#include <linux/namei.h>
ac27a0ec
DK
33#include <linux/uio.h>
34#include <linux/bio.h>
4c0425ff 35#include <linux/workqueue.h>
744692dc 36#include <linux/kernel.h>
6db26ffc 37#include <linux/printk.h>
5a0e3ad6 38#include <linux/slab.h>
a8901d34 39#include <linux/ratelimit.h>
a27bb332 40#include <linux/aio.h>
00a1a053 41#include <linux/bitops.h>
9bffad1e 42
3dcf5451 43#include "ext4_jbd2.h"
ac27a0ec
DK
44#include "xattr.h"
45#include "acl.h"
9f125d64 46#include "truncate.h"
ac27a0ec 47
9bffad1e
TT
48#include <trace/events/ext4.h>
49
a1d6cc56
AK
50#define MPAGE_DA_EXTENT_TAIL 0x01
51
814525f4
DW
52static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw,
53 struct ext4_inode_info *ei)
54{
55 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
56 __u16 csum_lo;
57 __u16 csum_hi = 0;
58 __u32 csum;
59
171a7f21 60 csum_lo = le16_to_cpu(raw->i_checksum_lo);
814525f4
DW
61 raw->i_checksum_lo = 0;
62 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
63 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
171a7f21 64 csum_hi = le16_to_cpu(raw->i_checksum_hi);
814525f4
DW
65 raw->i_checksum_hi = 0;
66 }
67
68 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw,
69 EXT4_INODE_SIZE(inode->i_sb));
70
171a7f21 71 raw->i_checksum_lo = cpu_to_le16(csum_lo);
814525f4
DW
72 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
73 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
171a7f21 74 raw->i_checksum_hi = cpu_to_le16(csum_hi);
814525f4
DW
75
76 return csum;
77}
78
79static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
80 struct ext4_inode_info *ei)
81{
82 __u32 provided, calculated;
83
84 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
85 cpu_to_le32(EXT4_OS_LINUX) ||
86 !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
87 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
88 return 1;
89
90 provided = le16_to_cpu(raw->i_checksum_lo);
91 calculated = ext4_inode_csum(inode, raw, ei);
92 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
93 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
94 provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
95 else
96 calculated &= 0xFFFF;
97
98 return provided == calculated;
99}
100
101static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
102 struct ext4_inode_info *ei)
103{
104 __u32 csum;
105
106 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
107 cpu_to_le32(EXT4_OS_LINUX) ||
108 !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
109 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
110 return;
111
112 csum = ext4_inode_csum(inode, raw, ei);
113 raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
114 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
115 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
116 raw->i_checksum_hi = cpu_to_le16(csum >> 16);
117}
118
678aaf48
JK
119static inline int ext4_begin_ordered_truncate(struct inode *inode,
120 loff_t new_size)
121{
7ff9c073 122 trace_ext4_begin_ordered_truncate(inode, new_size);
8aefcd55
TT
123 /*
124 * If jinode is zero, then we never opened the file for
125 * writing, so there's no need to call
126 * jbd2_journal_begin_ordered_truncate() since there's no
127 * outstanding writes we need to flush.
128 */
129 if (!EXT4_I(inode)->jinode)
130 return 0;
131 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
132 EXT4_I(inode)->jinode,
133 new_size);
678aaf48
JK
134}
135
d47992f8
LC
136static void ext4_invalidatepage(struct page *page, unsigned int offset,
137 unsigned int length);
cb20d518
TT
138static int __ext4_journalled_writepage(struct page *page, unsigned int len);
139static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
fffb2739
JK
140static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
141 int pextents);
64769240 142
ac27a0ec
DK
143/*
144 * Test whether an inode is a fast symlink.
145 */
617ba13b 146static int ext4_inode_is_fast_symlink(struct inode *inode)
ac27a0ec 147{
65eddb56
YY
148 int ea_blocks = EXT4_I(inode)->i_file_acl ?
149 EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0;
ac27a0ec 150
bd9db175
ZL
151 if (ext4_has_inline_data(inode))
152 return 0;
153
ac27a0ec
DK
154 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
155}
156
ac27a0ec
DK
157/*
158 * Restart the transaction associated with *handle. This does a commit,
159 * so before we call here everything must be consistently dirtied against
160 * this transaction.
161 */
fa5d1113 162int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode,
487caeef 163 int nblocks)
ac27a0ec 164{
487caeef
JK
165 int ret;
166
167 /*
e35fd660 168 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
487caeef
JK
169 * moment, get_block can be called only for blocks inside i_size since
170 * page cache has been already dropped and writes are blocked by
171 * i_mutex. So we can safely drop the i_data_sem here.
172 */
0390131b 173 BUG_ON(EXT4_JOURNAL(inode) == NULL);
ac27a0ec 174 jbd_debug(2, "restarting handle %p\n", handle);
487caeef 175 up_write(&EXT4_I(inode)->i_data_sem);
8e8eaabe 176 ret = ext4_journal_restart(handle, nblocks);
487caeef 177 down_write(&EXT4_I(inode)->i_data_sem);
fa5d1113 178 ext4_discard_preallocations(inode);
487caeef
JK
179
180 return ret;
ac27a0ec
DK
181}
182
183/*
184 * Called at the last iput() if i_nlink is zero.
185 */
0930fcc1 186void ext4_evict_inode(struct inode *inode)
ac27a0ec
DK
187{
188 handle_t *handle;
bc965ab3 189 int err;
ac27a0ec 190
7ff9c073 191 trace_ext4_evict_inode(inode);
2581fdc8 192
0930fcc1 193 if (inode->i_nlink) {
2d859db3
JK
194 /*
195 * When journalling data dirty buffers are tracked only in the
196 * journal. So although mm thinks everything is clean and
197 * ready for reaping the inode might still have some pages to
198 * write in the running transaction or waiting to be
199 * checkpointed. Thus calling jbd2_journal_invalidatepage()
200 * (via truncate_inode_pages()) to discard these buffers can
201 * cause data loss. Also even if we did not discard these
202 * buffers, we would have no way to find them after the inode
203 * is reaped and thus user could see stale data if he tries to
204 * read them before the transaction is checkpointed. So be
205 * careful and force everything to disk here... We use
206 * ei->i_datasync_tid to store the newest transaction
207 * containing inode's data.
208 *
209 * Note that directories do not have this problem because they
210 * don't use page cache.
211 */
212 if (ext4_should_journal_data(inode) &&
2b405bfa
TT
213 (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) &&
214 inode->i_ino != EXT4_JOURNAL_INO) {
2d859db3
JK
215 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
216 tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
217
d76a3a77 218 jbd2_complete_transaction(journal, commit_tid);
2d859db3
JK
219 filemap_write_and_wait(&inode->i_data);
220 }
91b0abe3 221 truncate_inode_pages_final(&inode->i_data);
5dc23bdd
JK
222
223 WARN_ON(atomic_read(&EXT4_I(inode)->i_ioend_count));
0930fcc1
AV
224 goto no_delete;
225 }
226
907f4554 227 if (!is_bad_inode(inode))
871a2931 228 dquot_initialize(inode);
907f4554 229
678aaf48
JK
230 if (ext4_should_order_data(inode))
231 ext4_begin_ordered_truncate(inode, 0);
91b0abe3 232 truncate_inode_pages_final(&inode->i_data);
ac27a0ec 233
5dc23bdd 234 WARN_ON(atomic_read(&EXT4_I(inode)->i_ioend_count));
ac27a0ec
DK
235 if (is_bad_inode(inode))
236 goto no_delete;
237
8e8ad8a5
JK
238 /*
239 * Protect us against freezing - iput() caller didn't have to have any
240 * protection against it
241 */
242 sb_start_intwrite(inode->i_sb);
9924a92a
TT
243 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
244 ext4_blocks_for_truncate(inode)+3);
ac27a0ec 245 if (IS_ERR(handle)) {
bc965ab3 246 ext4_std_error(inode->i_sb, PTR_ERR(handle));
ac27a0ec
DK
247 /*
248 * If we're going to skip the normal cleanup, we still need to
249 * make sure that the in-core orphan linked list is properly
250 * cleaned up.
251 */
617ba13b 252 ext4_orphan_del(NULL, inode);
8e8ad8a5 253 sb_end_intwrite(inode->i_sb);
ac27a0ec
DK
254 goto no_delete;
255 }
256
257 if (IS_SYNC(inode))
0390131b 258 ext4_handle_sync(handle);
ac27a0ec 259 inode->i_size = 0;
bc965ab3
TT
260 err = ext4_mark_inode_dirty(handle, inode);
261 if (err) {
12062ddd 262 ext4_warning(inode->i_sb,
bc965ab3
TT
263 "couldn't mark inode dirty (err %d)", err);
264 goto stop_handle;
265 }
ac27a0ec 266 if (inode->i_blocks)
617ba13b 267 ext4_truncate(inode);
bc965ab3
TT
268
269 /*
270 * ext4_ext_truncate() doesn't reserve any slop when it
271 * restarts journal transactions; therefore there may not be
272 * enough credits left in the handle to remove the inode from
273 * the orphan list and set the dtime field.
274 */
0390131b 275 if (!ext4_handle_has_enough_credits(handle, 3)) {
bc965ab3
TT
276 err = ext4_journal_extend(handle, 3);
277 if (err > 0)
278 err = ext4_journal_restart(handle, 3);
279 if (err != 0) {
12062ddd 280 ext4_warning(inode->i_sb,
bc965ab3
TT
281 "couldn't extend journal (err %d)", err);
282 stop_handle:
283 ext4_journal_stop(handle);
45388219 284 ext4_orphan_del(NULL, inode);
8e8ad8a5 285 sb_end_intwrite(inode->i_sb);
bc965ab3
TT
286 goto no_delete;
287 }
288 }
289
ac27a0ec 290 /*
617ba13b 291 * Kill off the orphan record which ext4_truncate created.
ac27a0ec 292 * AKPM: I think this can be inside the above `if'.
617ba13b 293 * Note that ext4_orphan_del() has to be able to cope with the
ac27a0ec 294 * deletion of a non-existent orphan - this is because we don't
617ba13b 295 * know if ext4_truncate() actually created an orphan record.
ac27a0ec
DK
296 * (Well, we could do this if we need to, but heck - it works)
297 */
617ba13b
MC
298 ext4_orphan_del(handle, inode);
299 EXT4_I(inode)->i_dtime = get_seconds();
ac27a0ec
DK
300
301 /*
302 * One subtle ordering requirement: if anything has gone wrong
303 * (transaction abort, IO errors, whatever), then we can still
304 * do these next steps (the fs will already have been marked as
305 * having errors), but we can't free the inode if the mark_dirty
306 * fails.
307 */
617ba13b 308 if (ext4_mark_inode_dirty(handle, inode))
ac27a0ec 309 /* If that failed, just do the required in-core inode clear. */
0930fcc1 310 ext4_clear_inode(inode);
ac27a0ec 311 else
617ba13b
MC
312 ext4_free_inode(handle, inode);
313 ext4_journal_stop(handle);
8e8ad8a5 314 sb_end_intwrite(inode->i_sb);
ac27a0ec
DK
315 return;
316no_delete:
0930fcc1 317 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
ac27a0ec
DK
318}
319
a9e7f447
DM
320#ifdef CONFIG_QUOTA
321qsize_t *ext4_get_reserved_space(struct inode *inode)
60e58e0f 322{
a9e7f447 323 return &EXT4_I(inode)->i_reserved_quota;
60e58e0f 324}
a9e7f447 325#endif
9d0be502 326
0637c6f4
TT
327/*
328 * Called with i_data_sem down, which is important since we can call
329 * ext4_discard_preallocations() from here.
330 */
5f634d06
AK
331void ext4_da_update_reserve_space(struct inode *inode,
332 int used, int quota_claim)
12219aea
AK
333{
334 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
0637c6f4 335 struct ext4_inode_info *ei = EXT4_I(inode);
0637c6f4
TT
336
337 spin_lock(&ei->i_block_reservation_lock);
d8990240 338 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
0637c6f4 339 if (unlikely(used > ei->i_reserved_data_blocks)) {
8de5c325 340 ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
1084f252 341 "with only %d reserved data blocks",
0637c6f4
TT
342 __func__, inode->i_ino, used,
343 ei->i_reserved_data_blocks);
344 WARN_ON(1);
345 used = ei->i_reserved_data_blocks;
346 }
12219aea 347
0637c6f4
TT
348 /* Update per-inode reservations */
349 ei->i_reserved_data_blocks -= used;
71d4f7d0 350 percpu_counter_sub(&sbi->s_dirtyclusters_counter, used);
6bc6e63f 351
12219aea 352 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
60e58e0f 353
72b8ab9d
ES
354 /* Update quota subsystem for data blocks */
355 if (quota_claim)
7b415bf6 356 dquot_claim_block(inode, EXT4_C2B(sbi, used));
72b8ab9d 357 else {
5f634d06
AK
358 /*
359 * We did fallocate with an offset that is already delayed
360 * allocated. So on delayed allocated writeback we should
72b8ab9d 361 * not re-claim the quota for fallocated blocks.
5f634d06 362 */
7b415bf6 363 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
5f634d06 364 }
d6014301
AK
365
366 /*
367 * If we have done all the pending block allocations and if
368 * there aren't any writers on the inode, we can discard the
369 * inode's preallocations.
370 */
0637c6f4
TT
371 if ((ei->i_reserved_data_blocks == 0) &&
372 (atomic_read(&inode->i_writecount) == 0))
d6014301 373 ext4_discard_preallocations(inode);
12219aea
AK
374}
375
e29136f8 376static int __check_block_validity(struct inode *inode, const char *func,
c398eda0
TT
377 unsigned int line,
378 struct ext4_map_blocks *map)
6fd058f7 379{
24676da4
TT
380 if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
381 map->m_len)) {
c398eda0
TT
382 ext4_error_inode(inode, func, line, map->m_pblk,
383 "lblock %lu mapped to illegal pblock "
384 "(length %d)", (unsigned long) map->m_lblk,
385 map->m_len);
6fd058f7
TT
386 return -EIO;
387 }
388 return 0;
389}
390
e29136f8 391#define check_block_validity(inode, map) \
c398eda0 392 __check_block_validity((inode), __func__, __LINE__, (map))
e29136f8 393
921f266b
DM
394#ifdef ES_AGGRESSIVE_TEST
395static void ext4_map_blocks_es_recheck(handle_t *handle,
396 struct inode *inode,
397 struct ext4_map_blocks *es_map,
398 struct ext4_map_blocks *map,
399 int flags)
400{
401 int retval;
402
403 map->m_flags = 0;
404 /*
405 * There is a race window that the result is not the same.
406 * e.g. xfstests #223 when dioread_nolock enables. The reason
407 * is that we lookup a block mapping in extent status tree with
408 * out taking i_data_sem. So at the time the unwritten extent
409 * could be converted.
410 */
411 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
c8b459f4 412 down_read(&EXT4_I(inode)->i_data_sem);
921f266b
DM
413 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
414 retval = ext4_ext_map_blocks(handle, inode, map, flags &
415 EXT4_GET_BLOCKS_KEEP_SIZE);
416 } else {
417 retval = ext4_ind_map_blocks(handle, inode, map, flags &
418 EXT4_GET_BLOCKS_KEEP_SIZE);
419 }
420 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
421 up_read((&EXT4_I(inode)->i_data_sem));
422 /*
423 * Clear EXT4_MAP_FROM_CLUSTER and EXT4_MAP_BOUNDARY flag
424 * because it shouldn't be marked in es_map->m_flags.
425 */
426 map->m_flags &= ~(EXT4_MAP_FROM_CLUSTER | EXT4_MAP_BOUNDARY);
427
428 /*
429 * We don't check m_len because extent will be collpased in status
430 * tree. So the m_len might not equal.
431 */
432 if (es_map->m_lblk != map->m_lblk ||
433 es_map->m_flags != map->m_flags ||
434 es_map->m_pblk != map->m_pblk) {
bdafe42a 435 printk("ES cache assertion failed for inode: %lu "
921f266b
DM
436 "es_cached ex [%d/%d/%llu/%x] != "
437 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
438 inode->i_ino, es_map->m_lblk, es_map->m_len,
439 es_map->m_pblk, es_map->m_flags, map->m_lblk,
440 map->m_len, map->m_pblk, map->m_flags,
441 retval, flags);
442 }
443}
444#endif /* ES_AGGRESSIVE_TEST */
445
f5ab0d1f 446/*
e35fd660 447 * The ext4_map_blocks() function tries to look up the requested blocks,
2b2d6d01 448 * and returns if the blocks are already mapped.
f5ab0d1f 449 *
f5ab0d1f
MC
450 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
451 * and store the allocated blocks in the result buffer head and mark it
452 * mapped.
453 *
e35fd660
TT
454 * If file type is extents based, it will call ext4_ext_map_blocks(),
455 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
f5ab0d1f
MC
456 * based files
457 *
556615dc
LC
458 * On success, it returns the number of blocks being mapped or allocated.
459 * if create==0 and the blocks are pre-allocated and unwritten block,
f5ab0d1f
MC
460 * the result buffer head is unmapped. If the create ==1, it will make sure
461 * the buffer head is mapped.
462 *
463 * It returns 0 if plain look up failed (blocks have not been allocated), in
df3ab170 464 * that case, buffer head is unmapped
f5ab0d1f
MC
465 *
466 * It returns the error in case of allocation failure.
467 */
e35fd660
TT
468int ext4_map_blocks(handle_t *handle, struct inode *inode,
469 struct ext4_map_blocks *map, int flags)
0e855ac8 470{
d100eef2 471 struct extent_status es;
0e855ac8 472 int retval;
b8a86845 473 int ret = 0;
921f266b
DM
474#ifdef ES_AGGRESSIVE_TEST
475 struct ext4_map_blocks orig_map;
476
477 memcpy(&orig_map, map, sizeof(*map));
478#endif
f5ab0d1f 479
e35fd660
TT
480 map->m_flags = 0;
481 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
482 "logical block %lu\n", inode->i_ino, flags, map->m_len,
483 (unsigned long) map->m_lblk);
d100eef2 484
e861b5e9
TT
485 /*
486 * ext4_map_blocks returns an int, and m_len is an unsigned int
487 */
488 if (unlikely(map->m_len > INT_MAX))
489 map->m_len = INT_MAX;
490
4adb6ab3
KM
491 /* We can handle the block number less than EXT_MAX_BLOCKS */
492 if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
493 return -EIO;
494
d100eef2
ZL
495 /* Lookup extent status tree firstly */
496 if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
63b99968 497 ext4_es_lru_add(inode);
d100eef2
ZL
498 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
499 map->m_pblk = ext4_es_pblock(&es) +
500 map->m_lblk - es.es_lblk;
501 map->m_flags |= ext4_es_is_written(&es) ?
502 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
503 retval = es.es_len - (map->m_lblk - es.es_lblk);
504 if (retval > map->m_len)
505 retval = map->m_len;
506 map->m_len = retval;
507 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
508 retval = 0;
509 } else {
510 BUG_ON(1);
511 }
921f266b
DM
512#ifdef ES_AGGRESSIVE_TEST
513 ext4_map_blocks_es_recheck(handle, inode, map,
514 &orig_map, flags);
515#endif
d100eef2
ZL
516 goto found;
517 }
518
4df3d265 519 /*
b920c755
TT
520 * Try to see if we can get the block without requesting a new
521 * file system block.
4df3d265 522 */
729f52c6 523 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
c8b459f4 524 down_read(&EXT4_I(inode)->i_data_sem);
12e9b892 525 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
a4e5d88b
DM
526 retval = ext4_ext_map_blocks(handle, inode, map, flags &
527 EXT4_GET_BLOCKS_KEEP_SIZE);
0e855ac8 528 } else {
a4e5d88b
DM
529 retval = ext4_ind_map_blocks(handle, inode, map, flags &
530 EXT4_GET_BLOCKS_KEEP_SIZE);
0e855ac8 531 }
f7fec032 532 if (retval > 0) {
3be78c73 533 unsigned int status;
f7fec032 534
44fb851d
ZL
535 if (unlikely(retval != map->m_len)) {
536 ext4_warning(inode->i_sb,
537 "ES len assertion failed for inode "
538 "%lu: retval %d != map->m_len %d",
539 inode->i_ino, retval, map->m_len);
540 WARN_ON(1);
921f266b 541 }
921f266b 542
f7fec032
ZL
543 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
544 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
545 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
546 ext4_find_delalloc_range(inode, map->m_lblk,
547 map->m_lblk + map->m_len - 1))
548 status |= EXTENT_STATUS_DELAYED;
549 ret = ext4_es_insert_extent(inode, map->m_lblk,
550 map->m_len, map->m_pblk, status);
551 if (ret < 0)
552 retval = ret;
553 }
729f52c6
ZL
554 if (!(flags & EXT4_GET_BLOCKS_NO_LOCK))
555 up_read((&EXT4_I(inode)->i_data_sem));
f5ab0d1f 556
d100eef2 557found:
e35fd660 558 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
b8a86845 559 ret = check_block_validity(inode, map);
6fd058f7
TT
560 if (ret != 0)
561 return ret;
562 }
563
f5ab0d1f 564 /* If it is only a block(s) look up */
c2177057 565 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
f5ab0d1f
MC
566 return retval;
567
568 /*
569 * Returns if the blocks have already allocated
570 *
571 * Note that if blocks have been preallocated
df3ab170 572 * ext4_ext_get_block() returns the create = 0
f5ab0d1f
MC
573 * with buffer head unmapped.
574 */
e35fd660 575 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
b8a86845
LC
576 /*
577 * If we need to convert extent to unwritten
578 * we continue and do the actual work in
579 * ext4_ext_map_blocks()
580 */
581 if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
582 return retval;
4df3d265 583
2a8964d6 584 /*
a25a4e1a
ZL
585 * Here we clear m_flags because after allocating an new extent,
586 * it will be set again.
2a8964d6 587 */
a25a4e1a 588 map->m_flags &= ~EXT4_MAP_FLAGS;
2a8964d6 589
4df3d265 590 /*
556615dc 591 * New blocks allocate and/or writing to unwritten extent
f5ab0d1f
MC
592 * will possibly result in updating i_data, so we take
593 * the write lock of i_data_sem, and call get_blocks()
594 * with create == 1 flag.
4df3d265 595 */
c8b459f4 596 down_write(&EXT4_I(inode)->i_data_sem);
d2a17637
MC
597
598 /*
599 * if the caller is from delayed allocation writeout path
600 * we have already reserved fs blocks for allocation
601 * let the underlying get_block() function know to
602 * avoid double accounting
603 */
c2177057 604 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
f2321097 605 ext4_set_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
4df3d265
AK
606 /*
607 * We need to check for EXT4 here because migrate
608 * could have changed the inode type in between
609 */
12e9b892 610 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
e35fd660 611 retval = ext4_ext_map_blocks(handle, inode, map, flags);
0e855ac8 612 } else {
e35fd660 613 retval = ext4_ind_map_blocks(handle, inode, map, flags);
267e4db9 614
e35fd660 615 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
267e4db9
AK
616 /*
617 * We allocated new blocks which will result in
618 * i_data's format changing. Force the migrate
619 * to fail by clearing migrate flags
620 */
19f5fb7a 621 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
267e4db9 622 }
d2a17637 623
5f634d06
AK
624 /*
625 * Update reserved blocks/metadata blocks after successful
626 * block allocation which had been deferred till now. We don't
627 * support fallocate for non extent files. So we can update
628 * reserve space here.
629 */
630 if ((retval > 0) &&
1296cc85 631 (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
5f634d06
AK
632 ext4_da_update_reserve_space(inode, retval, 1);
633 }
f7fec032 634 if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)
f2321097 635 ext4_clear_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED);
2ac3b6e0 636
f7fec032 637 if (retval > 0) {
3be78c73 638 unsigned int status;
f7fec032 639
44fb851d
ZL
640 if (unlikely(retval != map->m_len)) {
641 ext4_warning(inode->i_sb,
642 "ES len assertion failed for inode "
643 "%lu: retval %d != map->m_len %d",
644 inode->i_ino, retval, map->m_len);
645 WARN_ON(1);
921f266b 646 }
921f266b 647
adb23551
ZL
648 /*
649 * If the extent has been zeroed out, we don't need to update
650 * extent status tree.
651 */
652 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
653 ext4_es_lookup_extent(inode, map->m_lblk, &es)) {
654 if (ext4_es_is_written(&es))
655 goto has_zeroout;
656 }
f7fec032
ZL
657 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
658 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
659 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
660 ext4_find_delalloc_range(inode, map->m_lblk,
661 map->m_lblk + map->m_len - 1))
662 status |= EXTENT_STATUS_DELAYED;
663 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
664 map->m_pblk, status);
665 if (ret < 0)
666 retval = ret;
5356f261
AK
667 }
668
adb23551 669has_zeroout:
4df3d265 670 up_write((&EXT4_I(inode)->i_data_sem));
e35fd660 671 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
b8a86845 672 ret = check_block_validity(inode, map);
6fd058f7
TT
673 if (ret != 0)
674 return ret;
675 }
0e855ac8
AK
676 return retval;
677}
678
f3bd1f3f
MC
679/* Maximum number of blocks we map for direct IO at once. */
680#define DIO_MAX_BLOCKS 4096
681
2ed88685
TT
682static int _ext4_get_block(struct inode *inode, sector_t iblock,
683 struct buffer_head *bh, int flags)
ac27a0ec 684{
3e4fdaf8 685 handle_t *handle = ext4_journal_current_handle();
2ed88685 686 struct ext4_map_blocks map;
7fb5409d 687 int ret = 0, started = 0;
f3bd1f3f 688 int dio_credits;
ac27a0ec 689
46c7f254
TM
690 if (ext4_has_inline_data(inode))
691 return -ERANGE;
692
2ed88685
TT
693 map.m_lblk = iblock;
694 map.m_len = bh->b_size >> inode->i_blkbits;
695
8b0f165f 696 if (flags && !(flags & EXT4_GET_BLOCKS_NO_LOCK) && !handle) {
7fb5409d 697 /* Direct IO write... */
2ed88685
TT
698 if (map.m_len > DIO_MAX_BLOCKS)
699 map.m_len = DIO_MAX_BLOCKS;
700 dio_credits = ext4_chunk_trans_blocks(inode, map.m_len);
9924a92a
TT
701 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS,
702 dio_credits);
7fb5409d 703 if (IS_ERR(handle)) {
ac27a0ec 704 ret = PTR_ERR(handle);
2ed88685 705 return ret;
ac27a0ec 706 }
7fb5409d 707 started = 1;
ac27a0ec
DK
708 }
709
2ed88685 710 ret = ext4_map_blocks(handle, inode, &map, flags);
7fb5409d 711 if (ret > 0) {
7b7a8665
CH
712 ext4_io_end_t *io_end = ext4_inode_aio(inode);
713
2ed88685
TT
714 map_bh(bh, inode->i_sb, map.m_pblk);
715 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
7b7a8665
CH
716 if (io_end && io_end->flag & EXT4_IO_END_UNWRITTEN)
717 set_buffer_defer_completion(bh);
2ed88685 718 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
7fb5409d 719 ret = 0;
ac27a0ec 720 }
7fb5409d
JK
721 if (started)
722 ext4_journal_stop(handle);
ac27a0ec
DK
723 return ret;
724}
725
2ed88685
TT
726int ext4_get_block(struct inode *inode, sector_t iblock,
727 struct buffer_head *bh, int create)
728{
729 return _ext4_get_block(inode, iblock, bh,
730 create ? EXT4_GET_BLOCKS_CREATE : 0);
731}
732
ac27a0ec
DK
733/*
734 * `handle' can be NULL if create is zero
735 */
617ba13b 736struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
10560082 737 ext4_lblk_t block, int create)
ac27a0ec 738{
2ed88685
TT
739 struct ext4_map_blocks map;
740 struct buffer_head *bh;
10560082 741 int err;
ac27a0ec
DK
742
743 J_ASSERT(handle != NULL || create == 0);
744
2ed88685
TT
745 map.m_lblk = block;
746 map.m_len = 1;
747 err = ext4_map_blocks(handle, inode, &map,
748 create ? EXT4_GET_BLOCKS_CREATE : 0);
ac27a0ec 749
10560082
TT
750 if (err == 0)
751 return create ? ERR_PTR(-ENOSPC) : NULL;
2ed88685 752 if (err < 0)
10560082 753 return ERR_PTR(err);
2ed88685
TT
754
755 bh = sb_getblk(inode->i_sb, map.m_pblk);
10560082
TT
756 if (unlikely(!bh))
757 return ERR_PTR(-ENOMEM);
2ed88685
TT
758 if (map.m_flags & EXT4_MAP_NEW) {
759 J_ASSERT(create != 0);
760 J_ASSERT(handle != NULL);
ac27a0ec 761
2ed88685
TT
762 /*
763 * Now that we do not always journal data, we should
764 * keep in mind whether this should always journal the
765 * new buffer as metadata. For now, regular file
766 * writes use ext4_get_block instead, so it's not a
767 * problem.
768 */
769 lock_buffer(bh);
770 BUFFER_TRACE(bh, "call get_create_access");
10560082
TT
771 err = ext4_journal_get_create_access(handle, bh);
772 if (unlikely(err)) {
773 unlock_buffer(bh);
774 goto errout;
775 }
776 if (!buffer_uptodate(bh)) {
2ed88685
TT
777 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
778 set_buffer_uptodate(bh);
ac27a0ec 779 }
2ed88685
TT
780 unlock_buffer(bh);
781 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
782 err = ext4_handle_dirty_metadata(handle, inode, bh);
10560082
TT
783 if (unlikely(err))
784 goto errout;
785 } else
2ed88685 786 BUFFER_TRACE(bh, "not a new buffer");
2ed88685 787 return bh;
10560082
TT
788errout:
789 brelse(bh);
790 return ERR_PTR(err);
ac27a0ec
DK
791}
792
617ba13b 793struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
1c215028 794 ext4_lblk_t block, int create)
ac27a0ec 795{
af5bc92d 796 struct buffer_head *bh;
ac27a0ec 797
10560082 798 bh = ext4_getblk(handle, inode, block, create);
1c215028 799 if (IS_ERR(bh))
ac27a0ec 800 return bh;
1c215028 801 if (!bh || buffer_uptodate(bh))
ac27a0ec 802 return bh;
65299a3b 803 ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh);
ac27a0ec
DK
804 wait_on_buffer(bh);
805 if (buffer_uptodate(bh))
806 return bh;
807 put_bh(bh);
1c215028 808 return ERR_PTR(-EIO);
ac27a0ec
DK
809}
810
f19d5870
TM
811int ext4_walk_page_buffers(handle_t *handle,
812 struct buffer_head *head,
813 unsigned from,
814 unsigned to,
815 int *partial,
816 int (*fn)(handle_t *handle,
817 struct buffer_head *bh))
ac27a0ec
DK
818{
819 struct buffer_head *bh;
820 unsigned block_start, block_end;
821 unsigned blocksize = head->b_size;
822 int err, ret = 0;
823 struct buffer_head *next;
824
af5bc92d
TT
825 for (bh = head, block_start = 0;
826 ret == 0 && (bh != head || !block_start);
de9a55b8 827 block_start = block_end, bh = next) {
ac27a0ec
DK
828 next = bh->b_this_page;
829 block_end = block_start + blocksize;
830 if (block_end <= from || block_start >= to) {
831 if (partial && !buffer_uptodate(bh))
832 *partial = 1;
833 continue;
834 }
835 err = (*fn)(handle, bh);
836 if (!ret)
837 ret = err;
838 }
839 return ret;
840}
841
842/*
843 * To preserve ordering, it is essential that the hole instantiation and
844 * the data write be encapsulated in a single transaction. We cannot
617ba13b 845 * close off a transaction and start a new one between the ext4_get_block()
dab291af 846 * and the commit_write(). So doing the jbd2_journal_start at the start of
ac27a0ec
DK
847 * prepare_write() is the right place.
848 *
36ade451
JK
849 * Also, this function can nest inside ext4_writepage(). In that case, we
850 * *know* that ext4_writepage() has generated enough buffer credits to do the
851 * whole page. So we won't block on the journal in that case, which is good,
852 * because the caller may be PF_MEMALLOC.
ac27a0ec 853 *
617ba13b 854 * By accident, ext4 can be reentered when a transaction is open via
ac27a0ec
DK
855 * quota file writes. If we were to commit the transaction while thus
856 * reentered, there can be a deadlock - we would be holding a quota
857 * lock, and the commit would never complete if another thread had a
858 * transaction open and was blocking on the quota lock - a ranking
859 * violation.
860 *
dab291af 861 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
ac27a0ec
DK
862 * will _not_ run commit under these circumstances because handle->h_ref
863 * is elevated. We'll still have enough credits for the tiny quotafile
864 * write.
865 */
f19d5870
TM
866int do_journal_get_write_access(handle_t *handle,
867 struct buffer_head *bh)
ac27a0ec 868{
56d35a4c
JK
869 int dirty = buffer_dirty(bh);
870 int ret;
871
ac27a0ec
DK
872 if (!buffer_mapped(bh) || buffer_freed(bh))
873 return 0;
56d35a4c 874 /*
ebdec241 875 * __block_write_begin() could have dirtied some buffers. Clean
56d35a4c
JK
876 * the dirty bit as jbd2_journal_get_write_access() could complain
877 * otherwise about fs integrity issues. Setting of the dirty bit
ebdec241 878 * by __block_write_begin() isn't a real problem here as we clear
56d35a4c
JK
879 * the bit before releasing a page lock and thus writeback cannot
880 * ever write the buffer.
881 */
882 if (dirty)
883 clear_buffer_dirty(bh);
5d601255 884 BUFFER_TRACE(bh, "get write access");
56d35a4c
JK
885 ret = ext4_journal_get_write_access(handle, bh);
886 if (!ret && dirty)
887 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
888 return ret;
ac27a0ec
DK
889}
890
8b0f165f
AP
891static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
892 struct buffer_head *bh_result, int create);
bfc1af65 893static int ext4_write_begin(struct file *file, struct address_space *mapping,
de9a55b8
TT
894 loff_t pos, unsigned len, unsigned flags,
895 struct page **pagep, void **fsdata)
ac27a0ec 896{
af5bc92d 897 struct inode *inode = mapping->host;
1938a150 898 int ret, needed_blocks;
ac27a0ec
DK
899 handle_t *handle;
900 int retries = 0;
af5bc92d 901 struct page *page;
de9a55b8 902 pgoff_t index;
af5bc92d 903 unsigned from, to;
bfc1af65 904
9bffad1e 905 trace_ext4_write_begin(inode, pos, len, flags);
1938a150
AK
906 /*
907 * Reserve one block more for addition to orphan list in case
908 * we allocate blocks but write fails for some reason
909 */
910 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
de9a55b8 911 index = pos >> PAGE_CACHE_SHIFT;
af5bc92d
TT
912 from = pos & (PAGE_CACHE_SIZE - 1);
913 to = from + len;
ac27a0ec 914
f19d5870
TM
915 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
916 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
917 flags, pagep);
918 if (ret < 0)
47564bfb
TT
919 return ret;
920 if (ret == 1)
921 return 0;
f19d5870
TM
922 }
923
47564bfb
TT
924 /*
925 * grab_cache_page_write_begin() can take a long time if the
926 * system is thrashing due to memory pressure, or if the page
927 * is being written back. So grab it first before we start
928 * the transaction handle. This also allows us to allocate
929 * the page (if needed) without using GFP_NOFS.
930 */
931retry_grab:
932 page = grab_cache_page_write_begin(mapping, index, flags);
933 if (!page)
934 return -ENOMEM;
935 unlock_page(page);
936
937retry_journal:
9924a92a 938 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
af5bc92d 939 if (IS_ERR(handle)) {
47564bfb
TT
940 page_cache_release(page);
941 return PTR_ERR(handle);
7479d2b9 942 }
ac27a0ec 943
47564bfb
TT
944 lock_page(page);
945 if (page->mapping != mapping) {
946 /* The page got truncated from under us */
947 unlock_page(page);
948 page_cache_release(page);
cf108bca 949 ext4_journal_stop(handle);
47564bfb 950 goto retry_grab;
cf108bca 951 }
7afe5aa5
DM
952 /* In case writeback began while the page was unlocked */
953 wait_for_stable_page(page);
cf108bca 954
744692dc 955 if (ext4_should_dioread_nolock(inode))
6e1db88d 956 ret = __block_write_begin(page, pos, len, ext4_get_block_write);
744692dc 957 else
6e1db88d 958 ret = __block_write_begin(page, pos, len, ext4_get_block);
bfc1af65
NP
959
960 if (!ret && ext4_should_journal_data(inode)) {
f19d5870
TM
961 ret = ext4_walk_page_buffers(handle, page_buffers(page),
962 from, to, NULL,
963 do_journal_get_write_access);
ac27a0ec 964 }
bfc1af65
NP
965
966 if (ret) {
af5bc92d 967 unlock_page(page);
ae4d5372 968 /*
6e1db88d 969 * __block_write_begin may have instantiated a few blocks
ae4d5372
AK
970 * outside i_size. Trim these off again. Don't need
971 * i_size_read because we hold i_mutex.
1938a150
AK
972 *
973 * Add inode to orphan list in case we crash before
974 * truncate finishes
ae4d5372 975 */
ffacfa7a 976 if (pos + len > inode->i_size && ext4_can_truncate(inode))
1938a150
AK
977 ext4_orphan_add(handle, inode);
978
979 ext4_journal_stop(handle);
980 if (pos + len > inode->i_size) {
b9a4207d 981 ext4_truncate_failed_write(inode);
de9a55b8 982 /*
ffacfa7a 983 * If truncate failed early the inode might
1938a150
AK
984 * still be on the orphan list; we need to
985 * make sure the inode is removed from the
986 * orphan list in that case.
987 */
988 if (inode->i_nlink)
989 ext4_orphan_del(NULL, inode);
990 }
bfc1af65 991
47564bfb
TT
992 if (ret == -ENOSPC &&
993 ext4_should_retry_alloc(inode->i_sb, &retries))
994 goto retry_journal;
995 page_cache_release(page);
996 return ret;
997 }
998 *pagep = page;
ac27a0ec
DK
999 return ret;
1000}
1001
bfc1af65
NP
1002/* For write_end() in data=journal mode */
1003static int write_end_fn(handle_t *handle, struct buffer_head *bh)
ac27a0ec 1004{
13fca323 1005 int ret;
ac27a0ec
DK
1006 if (!buffer_mapped(bh) || buffer_freed(bh))
1007 return 0;
1008 set_buffer_uptodate(bh);
13fca323
TT
1009 ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1010 clear_buffer_meta(bh);
1011 clear_buffer_prio(bh);
1012 return ret;
ac27a0ec
DK
1013}
1014
eed4333f
ZL
1015/*
1016 * We need to pick up the new inode size which generic_commit_write gave us
1017 * `file' can be NULL - eg, when called from page_symlink().
1018 *
1019 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1020 * buffers are managed internally.
1021 */
1022static int ext4_write_end(struct file *file,
1023 struct address_space *mapping,
1024 loff_t pos, unsigned len, unsigned copied,
1025 struct page *page, void *fsdata)
f8514083 1026{
f8514083 1027 handle_t *handle = ext4_journal_current_handle();
eed4333f
ZL
1028 struct inode *inode = mapping->host;
1029 int ret = 0, ret2;
1030 int i_size_changed = 0;
1031
1032 trace_ext4_write_end(inode, pos, len, copied);
1033 if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE)) {
1034 ret = ext4_jbd2_file_inode(handle, inode);
1035 if (ret) {
1036 unlock_page(page);
1037 page_cache_release(page);
1038 goto errout;
1039 }
1040 }
f8514083 1041
42c832de
TT
1042 if (ext4_has_inline_data(inode)) {
1043 ret = ext4_write_inline_data_end(inode, pos, len,
1044 copied, page);
1045 if (ret < 0)
1046 goto errout;
1047 copied = ret;
1048 } else
f19d5870
TM
1049 copied = block_write_end(file, mapping, pos,
1050 len, copied, page, fsdata);
f8514083 1051 /*
4631dbf6 1052 * it's important to update i_size while still holding page lock:
f8514083
AK
1053 * page writeout could otherwise come in and zero beyond i_size.
1054 */
4631dbf6 1055 i_size_changed = ext4_update_inode_size(inode, pos + copied);
f8514083
AK
1056 unlock_page(page);
1057 page_cache_release(page);
1058
1059 /*
1060 * Don't mark the inode dirty under page lock. First, it unnecessarily
1061 * makes the holding time of page lock longer. Second, it forces lock
1062 * ordering of page lock and transaction start for journaling
1063 * filesystems.
1064 */
1065 if (i_size_changed)
1066 ext4_mark_inode_dirty(handle, inode);
1067
ffacfa7a 1068 if (pos + len > inode->i_size && ext4_can_truncate(inode))
f8514083
AK
1069 /* if we have allocated more blocks and copied
1070 * less. We will have blocks allocated outside
1071 * inode->i_size. So truncate them
1072 */
1073 ext4_orphan_add(handle, inode);
74d553aa 1074errout:
617ba13b 1075 ret2 = ext4_journal_stop(handle);
ac27a0ec
DK
1076 if (!ret)
1077 ret = ret2;
bfc1af65 1078
f8514083 1079 if (pos + len > inode->i_size) {
b9a4207d 1080 ext4_truncate_failed_write(inode);
de9a55b8 1081 /*
ffacfa7a 1082 * If truncate failed early the inode might still be
f8514083
AK
1083 * on the orphan list; we need to make sure the inode
1084 * is removed from the orphan list in that case.
1085 */
1086 if (inode->i_nlink)
1087 ext4_orphan_del(NULL, inode);
1088 }
1089
bfc1af65 1090 return ret ? ret : copied;
ac27a0ec
DK
1091}
1092
bfc1af65 1093static int ext4_journalled_write_end(struct file *file,
de9a55b8
TT
1094 struct address_space *mapping,
1095 loff_t pos, unsigned len, unsigned copied,
1096 struct page *page, void *fsdata)
ac27a0ec 1097{
617ba13b 1098 handle_t *handle = ext4_journal_current_handle();
bfc1af65 1099 struct inode *inode = mapping->host;
ac27a0ec
DK
1100 int ret = 0, ret2;
1101 int partial = 0;
bfc1af65 1102 unsigned from, to;
4631dbf6 1103 int size_changed = 0;
ac27a0ec 1104
9bffad1e 1105 trace_ext4_journalled_write_end(inode, pos, len, copied);
bfc1af65
NP
1106 from = pos & (PAGE_CACHE_SIZE - 1);
1107 to = from + len;
1108
441c8508
CW
1109 BUG_ON(!ext4_handle_valid(handle));
1110
3fdcfb66
TM
1111 if (ext4_has_inline_data(inode))
1112 copied = ext4_write_inline_data_end(inode, pos, len,
1113 copied, page);
1114 else {
1115 if (copied < len) {
1116 if (!PageUptodate(page))
1117 copied = 0;
1118 page_zero_new_buffers(page, from+copied, to);
1119 }
ac27a0ec 1120
3fdcfb66
TM
1121 ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1122 to, &partial, write_end_fn);
1123 if (!partial)
1124 SetPageUptodate(page);
1125 }
4631dbf6 1126 size_changed = ext4_update_inode_size(inode, pos + copied);
19f5fb7a 1127 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
2d859db3 1128 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
4631dbf6
DM
1129 unlock_page(page);
1130 page_cache_release(page);
1131
1132 if (size_changed) {
617ba13b 1133 ret2 = ext4_mark_inode_dirty(handle, inode);
ac27a0ec
DK
1134 if (!ret)
1135 ret = ret2;
1136 }
bfc1af65 1137
ffacfa7a 1138 if (pos + len > inode->i_size && ext4_can_truncate(inode))
f8514083
AK
1139 /* if we have allocated more blocks and copied
1140 * less. We will have blocks allocated outside
1141 * inode->i_size. So truncate them
1142 */
1143 ext4_orphan_add(handle, inode);
1144
617ba13b 1145 ret2 = ext4_journal_stop(handle);
ac27a0ec
DK
1146 if (!ret)
1147 ret = ret2;
f8514083 1148 if (pos + len > inode->i_size) {
b9a4207d 1149 ext4_truncate_failed_write(inode);
de9a55b8 1150 /*
ffacfa7a 1151 * If truncate failed early the inode might still be
f8514083
AK
1152 * on the orphan list; we need to make sure the inode
1153 * is removed from the orphan list in that case.
1154 */
1155 if (inode->i_nlink)
1156 ext4_orphan_del(NULL, inode);
1157 }
bfc1af65
NP
1158
1159 return ret ? ret : copied;
ac27a0ec 1160}
d2a17637 1161
9d0be502 1162/*
7b415bf6 1163 * Reserve a single cluster located at lblock
9d0be502 1164 */
01f49d0b 1165static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock)
d2a17637 1166{
60e58e0f 1167 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
0637c6f4 1168 struct ext4_inode_info *ei = EXT4_I(inode);
7b415bf6 1169 unsigned int md_needed;
5dd4056d 1170 int ret;
03179fe9
TT
1171
1172 /*
1173 * We will charge metadata quota at writeout time; this saves
1174 * us from metadata over-estimation, though we may go over by
1175 * a small amount in the end. Here we just reserve for data.
1176 */
1177 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1178 if (ret)
1179 return ret;
d2a17637
MC
1180
1181 /*
1182 * recalculate the amount of metadata blocks to reserve
1183 * in order to allocate nrblocks
1184 * worse case is one extent per block
1185 */
0637c6f4 1186 spin_lock(&ei->i_block_reservation_lock);
03179fe9
TT
1187 /*
1188 * ext4_calc_metadata_amount() has side effects, which we have
1189 * to be prepared undo if we fail to claim space.
1190 */
71d4f7d0
TT
1191 md_needed = 0;
1192 trace_ext4_da_reserve_space(inode, 0);
d2a17637 1193
71d4f7d0 1194 if (ext4_claim_free_clusters(sbi, 1, 0)) {
03179fe9 1195 spin_unlock(&ei->i_block_reservation_lock);
03179fe9 1196 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
d2a17637
MC
1197 return -ENOSPC;
1198 }
9d0be502 1199 ei->i_reserved_data_blocks++;
0637c6f4 1200 spin_unlock(&ei->i_block_reservation_lock);
39bc680a 1201
d2a17637
MC
1202 return 0; /* success */
1203}
1204
12219aea 1205static void ext4_da_release_space(struct inode *inode, int to_free)
d2a17637
MC
1206{
1207 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
0637c6f4 1208 struct ext4_inode_info *ei = EXT4_I(inode);
d2a17637 1209
cd213226
MC
1210 if (!to_free)
1211 return; /* Nothing to release, exit */
1212
d2a17637 1213 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
cd213226 1214
5a58ec87 1215 trace_ext4_da_release_space(inode, to_free);
0637c6f4 1216 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
cd213226 1217 /*
0637c6f4
TT
1218 * if there aren't enough reserved blocks, then the
1219 * counter is messed up somewhere. Since this
1220 * function is called from invalidate page, it's
1221 * harmless to return without any action.
cd213226 1222 */
8de5c325 1223 ext4_warning(inode->i_sb, "ext4_da_release_space: "
0637c6f4 1224 "ino %lu, to_free %d with only %d reserved "
1084f252 1225 "data blocks", inode->i_ino, to_free,
0637c6f4
TT
1226 ei->i_reserved_data_blocks);
1227 WARN_ON(1);
1228 to_free = ei->i_reserved_data_blocks;
cd213226 1229 }
0637c6f4 1230 ei->i_reserved_data_blocks -= to_free;
cd213226 1231
72b8ab9d 1232 /* update fs dirty data blocks counter */
57042651 1233 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
d2a17637 1234
d2a17637 1235 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
60e58e0f 1236
7b415bf6 1237 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
d2a17637
MC
1238}
1239
1240static void ext4_da_page_release_reservation(struct page *page,
ca99fdd2
LC
1241 unsigned int offset,
1242 unsigned int length)
d2a17637
MC
1243{
1244 int to_release = 0;
1245 struct buffer_head *head, *bh;
1246 unsigned int curr_off = 0;
7b415bf6
AK
1247 struct inode *inode = page->mapping->host;
1248 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
ca99fdd2 1249 unsigned int stop = offset + length;
7b415bf6 1250 int num_clusters;
51865fda 1251 ext4_fsblk_t lblk;
d2a17637 1252
ca99fdd2
LC
1253 BUG_ON(stop > PAGE_CACHE_SIZE || stop < length);
1254
d2a17637
MC
1255 head = page_buffers(page);
1256 bh = head;
1257 do {
1258 unsigned int next_off = curr_off + bh->b_size;
1259
ca99fdd2
LC
1260 if (next_off > stop)
1261 break;
1262
d2a17637
MC
1263 if ((offset <= curr_off) && (buffer_delay(bh))) {
1264 to_release++;
1265 clear_buffer_delay(bh);
1266 }
1267 curr_off = next_off;
1268 } while ((bh = bh->b_this_page) != head);
7b415bf6 1269
51865fda
ZL
1270 if (to_release) {
1271 lblk = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1272 ext4_es_remove_extent(inode, lblk, to_release);
1273 }
1274
7b415bf6
AK
1275 /* If we have released all the blocks belonging to a cluster, then we
1276 * need to release the reserved space for that cluster. */
1277 num_clusters = EXT4_NUM_B2C(sbi, to_release);
1278 while (num_clusters > 0) {
7b415bf6
AK
1279 lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) +
1280 ((num_clusters - 1) << sbi->s_cluster_bits);
1281 if (sbi->s_cluster_ratio == 1 ||
7d1b1fbc 1282 !ext4_find_delalloc_cluster(inode, lblk))
7b415bf6
AK
1283 ext4_da_release_space(inode, 1);
1284
1285 num_clusters--;
1286 }
d2a17637 1287}
ac27a0ec 1288
64769240
AT
1289/*
1290 * Delayed allocation stuff
1291 */
1292
4e7ea81d
JK
1293struct mpage_da_data {
1294 struct inode *inode;
1295 struct writeback_control *wbc;
6b523df4 1296
4e7ea81d
JK
1297 pgoff_t first_page; /* The first page to write */
1298 pgoff_t next_page; /* Current page to examine */
1299 pgoff_t last_page; /* Last page to examine */
791b7f08 1300 /*
4e7ea81d
JK
1301 * Extent to map - this can be after first_page because that can be
1302 * fully mapped. We somewhat abuse m_flags to store whether the extent
1303 * is delalloc or unwritten.
791b7f08 1304 */
4e7ea81d
JK
1305 struct ext4_map_blocks map;
1306 struct ext4_io_submit io_submit; /* IO submission data */
1307};
64769240 1308
4e7ea81d
JK
1309static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1310 bool invalidate)
c4a0c46e
AK
1311{
1312 int nr_pages, i;
1313 pgoff_t index, end;
1314 struct pagevec pvec;
1315 struct inode *inode = mpd->inode;
1316 struct address_space *mapping = inode->i_mapping;
4e7ea81d
JK
1317
1318 /* This is necessary when next_page == 0. */
1319 if (mpd->first_page >= mpd->next_page)
1320 return;
c4a0c46e 1321
c7f5938a
CW
1322 index = mpd->first_page;
1323 end = mpd->next_page - 1;
4e7ea81d
JK
1324 if (invalidate) {
1325 ext4_lblk_t start, last;
1326 start = index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1327 last = end << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1328 ext4_es_remove_extent(inode, start, last - start + 1);
1329 }
51865fda 1330
66bea92c 1331 pagevec_init(&pvec, 0);
c4a0c46e
AK
1332 while (index <= end) {
1333 nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE);
1334 if (nr_pages == 0)
1335 break;
1336 for (i = 0; i < nr_pages; i++) {
1337 struct page *page = pvec.pages[i];
9b1d0998 1338 if (page->index > end)
c4a0c46e 1339 break;
c4a0c46e
AK
1340 BUG_ON(!PageLocked(page));
1341 BUG_ON(PageWriteback(page));
4e7ea81d
JK
1342 if (invalidate) {
1343 block_invalidatepage(page, 0, PAGE_CACHE_SIZE);
1344 ClearPageUptodate(page);
1345 }
c4a0c46e
AK
1346 unlock_page(page);
1347 }
9b1d0998
JK
1348 index = pvec.pages[nr_pages - 1]->index + 1;
1349 pagevec_release(&pvec);
c4a0c46e 1350 }
c4a0c46e
AK
1351}
1352
df22291f
AK
1353static void ext4_print_free_blocks(struct inode *inode)
1354{
1355 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
92b97816 1356 struct super_block *sb = inode->i_sb;
f78ee70d 1357 struct ext4_inode_info *ei = EXT4_I(inode);
92b97816
TT
1358
1359 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
5dee5437 1360 EXT4_C2B(EXT4_SB(inode->i_sb),
f78ee70d 1361 ext4_count_free_clusters(sb)));
92b97816
TT
1362 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1363 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
f78ee70d 1364 (long long) EXT4_C2B(EXT4_SB(sb),
57042651 1365 percpu_counter_sum(&sbi->s_freeclusters_counter)));
92b97816 1366 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
f78ee70d 1367 (long long) EXT4_C2B(EXT4_SB(sb),
7b415bf6 1368 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
92b97816
TT
1369 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1370 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
f78ee70d 1371 ei->i_reserved_data_blocks);
df22291f
AK
1372 return;
1373}
1374
c364b22c 1375static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
29fa89d0 1376{
c364b22c 1377 return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
29fa89d0
AK
1378}
1379
5356f261
AK
1380/*
1381 * This function is grabs code from the very beginning of
1382 * ext4_map_blocks, but assumes that the caller is from delayed write
1383 * time. This function looks up the requested blocks and sets the
1384 * buffer delay bit under the protection of i_data_sem.
1385 */
1386static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1387 struct ext4_map_blocks *map,
1388 struct buffer_head *bh)
1389{
d100eef2 1390 struct extent_status es;
5356f261
AK
1391 int retval;
1392 sector_t invalid_block = ~((sector_t) 0xffff);
921f266b
DM
1393#ifdef ES_AGGRESSIVE_TEST
1394 struct ext4_map_blocks orig_map;
1395
1396 memcpy(&orig_map, map, sizeof(*map));
1397#endif
5356f261
AK
1398
1399 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1400 invalid_block = ~0;
1401
1402 map->m_flags = 0;
1403 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1404 "logical block %lu\n", inode->i_ino, map->m_len,
1405 (unsigned long) map->m_lblk);
d100eef2
ZL
1406
1407 /* Lookup extent status tree firstly */
1408 if (ext4_es_lookup_extent(inode, iblock, &es)) {
63b99968 1409 ext4_es_lru_add(inode);
d100eef2
ZL
1410 if (ext4_es_is_hole(&es)) {
1411 retval = 0;
c8b459f4 1412 down_read(&EXT4_I(inode)->i_data_sem);
d100eef2
ZL
1413 goto add_delayed;
1414 }
1415
1416 /*
1417 * Delayed extent could be allocated by fallocate.
1418 * So we need to check it.
1419 */
1420 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1421 map_bh(bh, inode->i_sb, invalid_block);
1422 set_buffer_new(bh);
1423 set_buffer_delay(bh);
1424 return 0;
1425 }
1426
1427 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1428 retval = es.es_len - (iblock - es.es_lblk);
1429 if (retval > map->m_len)
1430 retval = map->m_len;
1431 map->m_len = retval;
1432 if (ext4_es_is_written(&es))
1433 map->m_flags |= EXT4_MAP_MAPPED;
1434 else if (ext4_es_is_unwritten(&es))
1435 map->m_flags |= EXT4_MAP_UNWRITTEN;
1436 else
1437 BUG_ON(1);
1438
921f266b
DM
1439#ifdef ES_AGGRESSIVE_TEST
1440 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1441#endif
d100eef2
ZL
1442 return retval;
1443 }
1444
5356f261
AK
1445 /*
1446 * Try to see if we can get the block without requesting a new
1447 * file system block.
1448 */
c8b459f4 1449 down_read(&EXT4_I(inode)->i_data_sem);
9c3569b5
TM
1450 if (ext4_has_inline_data(inode)) {
1451 /*
1452 * We will soon create blocks for this page, and let
1453 * us pretend as if the blocks aren't allocated yet.
1454 * In case of clusters, we have to handle the work
1455 * of mapping from cluster so that the reserved space
1456 * is calculated properly.
1457 */
1458 if ((EXT4_SB(inode->i_sb)->s_cluster_ratio > 1) &&
1459 ext4_find_delalloc_cluster(inode, map->m_lblk))
1460 map->m_flags |= EXT4_MAP_FROM_CLUSTER;
1461 retval = 0;
1462 } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
d100eef2
ZL
1463 retval = ext4_ext_map_blocks(NULL, inode, map,
1464 EXT4_GET_BLOCKS_NO_PUT_HOLE);
5356f261 1465 else
d100eef2
ZL
1466 retval = ext4_ind_map_blocks(NULL, inode, map,
1467 EXT4_GET_BLOCKS_NO_PUT_HOLE);
5356f261 1468
d100eef2 1469add_delayed:
5356f261 1470 if (retval == 0) {
f7fec032 1471 int ret;
5356f261
AK
1472 /*
1473 * XXX: __block_prepare_write() unmaps passed block,
1474 * is it OK?
1475 */
386ad67c
LC
1476 /*
1477 * If the block was allocated from previously allocated cluster,
1478 * then we don't need to reserve it again. However we still need
1479 * to reserve metadata for every block we're going to write.
1480 */
5356f261 1481 if (!(map->m_flags & EXT4_MAP_FROM_CLUSTER)) {
f7fec032
ZL
1482 ret = ext4_da_reserve_space(inode, iblock);
1483 if (ret) {
5356f261 1484 /* not enough space to reserve */
f7fec032 1485 retval = ret;
5356f261 1486 goto out_unlock;
f7fec032 1487 }
5356f261
AK
1488 }
1489
f7fec032
ZL
1490 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1491 ~0, EXTENT_STATUS_DELAYED);
1492 if (ret) {
1493 retval = ret;
51865fda 1494 goto out_unlock;
f7fec032 1495 }
51865fda 1496
5356f261
AK
1497 /* Clear EXT4_MAP_FROM_CLUSTER flag since its purpose is served
1498 * and it should not appear on the bh->b_state.
1499 */
1500 map->m_flags &= ~EXT4_MAP_FROM_CLUSTER;
1501
1502 map_bh(bh, inode->i_sb, invalid_block);
1503 set_buffer_new(bh);
1504 set_buffer_delay(bh);
f7fec032
ZL
1505 } else if (retval > 0) {
1506 int ret;
3be78c73 1507 unsigned int status;
f7fec032 1508
44fb851d
ZL
1509 if (unlikely(retval != map->m_len)) {
1510 ext4_warning(inode->i_sb,
1511 "ES len assertion failed for inode "
1512 "%lu: retval %d != map->m_len %d",
1513 inode->i_ino, retval, map->m_len);
1514 WARN_ON(1);
921f266b 1515 }
921f266b 1516
f7fec032
ZL
1517 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1518 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1519 ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1520 map->m_pblk, status);
1521 if (ret != 0)
1522 retval = ret;
5356f261
AK
1523 }
1524
1525out_unlock:
1526 up_read((&EXT4_I(inode)->i_data_sem));
1527
1528 return retval;
1529}
1530
64769240 1531/*
b920c755
TT
1532 * This is a special get_blocks_t callback which is used by
1533 * ext4_da_write_begin(). It will either return mapped block or
1534 * reserve space for a single block.
29fa89d0
AK
1535 *
1536 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1537 * We also have b_blocknr = -1 and b_bdev initialized properly
1538 *
1539 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1540 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1541 * initialized properly.
64769240 1542 */
9c3569b5
TM
1543int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1544 struct buffer_head *bh, int create)
64769240 1545{
2ed88685 1546 struct ext4_map_blocks map;
64769240
AT
1547 int ret = 0;
1548
1549 BUG_ON(create == 0);
2ed88685
TT
1550 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1551
1552 map.m_lblk = iblock;
1553 map.m_len = 1;
64769240
AT
1554
1555 /*
1556 * first, we need to know whether the block is allocated already
1557 * preallocated blocks are unmapped but should treated
1558 * the same as allocated blocks.
1559 */
5356f261
AK
1560 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1561 if (ret <= 0)
2ed88685 1562 return ret;
64769240 1563
2ed88685
TT
1564 map_bh(bh, inode->i_sb, map.m_pblk);
1565 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags;
1566
1567 if (buffer_unwritten(bh)) {
1568 /* A delayed write to unwritten bh should be marked
1569 * new and mapped. Mapped ensures that we don't do
1570 * get_block multiple times when we write to the same
1571 * offset and new ensures that we do proper zero out
1572 * for partial write.
1573 */
1574 set_buffer_new(bh);
c8205636 1575 set_buffer_mapped(bh);
2ed88685
TT
1576 }
1577 return 0;
64769240 1578}
61628a3f 1579
62e086be
AK
1580static int bget_one(handle_t *handle, struct buffer_head *bh)
1581{
1582 get_bh(bh);
1583 return 0;
1584}
1585
1586static int bput_one(handle_t *handle, struct buffer_head *bh)
1587{
1588 put_bh(bh);
1589 return 0;
1590}
1591
1592static int __ext4_journalled_writepage(struct page *page,
62e086be
AK
1593 unsigned int len)
1594{
1595 struct address_space *mapping = page->mapping;
1596 struct inode *inode = mapping->host;
3fdcfb66 1597 struct buffer_head *page_bufs = NULL;
62e086be 1598 handle_t *handle = NULL;
3fdcfb66
TM
1599 int ret = 0, err = 0;
1600 int inline_data = ext4_has_inline_data(inode);
1601 struct buffer_head *inode_bh = NULL;
62e086be 1602
cb20d518 1603 ClearPageChecked(page);
3fdcfb66
TM
1604
1605 if (inline_data) {
1606 BUG_ON(page->index != 0);
1607 BUG_ON(len > ext4_get_max_inline_size(inode));
1608 inode_bh = ext4_journalled_write_inline_data(inode, len, page);
1609 if (inode_bh == NULL)
1610 goto out;
1611 } else {
1612 page_bufs = page_buffers(page);
1613 if (!page_bufs) {
1614 BUG();
1615 goto out;
1616 }
1617 ext4_walk_page_buffers(handle, page_bufs, 0, len,
1618 NULL, bget_one);
1619 }
62e086be
AK
1620 /* As soon as we unlock the page, it can go away, but we have
1621 * references to buffers so we are safe */
1622 unlock_page(page);
1623
9924a92a
TT
1624 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
1625 ext4_writepage_trans_blocks(inode));
62e086be
AK
1626 if (IS_ERR(handle)) {
1627 ret = PTR_ERR(handle);
1628 goto out;
1629 }
1630
441c8508
CW
1631 BUG_ON(!ext4_handle_valid(handle));
1632
3fdcfb66 1633 if (inline_data) {
5d601255 1634 BUFFER_TRACE(inode_bh, "get write access");
3fdcfb66 1635 ret = ext4_journal_get_write_access(handle, inode_bh);
62e086be 1636
3fdcfb66
TM
1637 err = ext4_handle_dirty_metadata(handle, inode, inode_bh);
1638
1639 } else {
1640 ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1641 do_journal_get_write_access);
1642
1643 err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1644 write_end_fn);
1645 }
62e086be
AK
1646 if (ret == 0)
1647 ret = err;
2d859db3 1648 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
62e086be
AK
1649 err = ext4_journal_stop(handle);
1650 if (!ret)
1651 ret = err;
1652
3fdcfb66 1653 if (!ext4_has_inline_data(inode))
8c9367fd 1654 ext4_walk_page_buffers(NULL, page_bufs, 0, len,
3fdcfb66 1655 NULL, bput_one);
19f5fb7a 1656 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
62e086be 1657out:
3fdcfb66 1658 brelse(inode_bh);
62e086be
AK
1659 return ret;
1660}
1661
61628a3f 1662/*
43ce1d23
AK
1663 * Note that we don't need to start a transaction unless we're journaling data
1664 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1665 * need to file the inode to the transaction's list in ordered mode because if
1666 * we are writing back data added by write(), the inode is already there and if
25985edc 1667 * we are writing back data modified via mmap(), no one guarantees in which
43ce1d23
AK
1668 * transaction the data will hit the disk. In case we are journaling data, we
1669 * cannot start transaction directly because transaction start ranks above page
1670 * lock so we have to do some magic.
1671 *
b920c755 1672 * This function can get called via...
20970ba6 1673 * - ext4_writepages after taking page lock (have journal handle)
b920c755 1674 * - journal_submit_inode_data_buffers (no journal handle)
f6463b0d 1675 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
b920c755 1676 * - grab_page_cache when doing write_begin (have journal handle)
43ce1d23
AK
1677 *
1678 * We don't do any block allocation in this function. If we have page with
1679 * multiple blocks we need to write those buffer_heads that are mapped. This
1680 * is important for mmaped based write. So if we do with blocksize 1K
1681 * truncate(f, 1024);
1682 * a = mmap(f, 0, 4096);
1683 * a[0] = 'a';
1684 * truncate(f, 4096);
1685 * we have in the page first buffer_head mapped via page_mkwrite call back
90802ed9 1686 * but other buffer_heads would be unmapped but dirty (dirty done via the
43ce1d23
AK
1687 * do_wp_page). So writepage should write the first block. If we modify
1688 * the mmap area beyond 1024 we will again get a page_fault and the
1689 * page_mkwrite callback will do the block allocation and mark the
1690 * buffer_heads mapped.
1691 *
1692 * We redirty the page if we have any buffer_heads that is either delay or
1693 * unwritten in the page.
1694 *
1695 * We can get recursively called as show below.
1696 *
1697 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1698 * ext4_writepage()
1699 *
1700 * But since we don't do any block allocation we should not deadlock.
1701 * Page also have the dirty flag cleared so we don't get recurive page_lock.
61628a3f 1702 */
43ce1d23 1703static int ext4_writepage(struct page *page,
62e086be 1704 struct writeback_control *wbc)
64769240 1705{
f8bec370 1706 int ret = 0;
61628a3f 1707 loff_t size;
498e5f24 1708 unsigned int len;
744692dc 1709 struct buffer_head *page_bufs = NULL;
61628a3f 1710 struct inode *inode = page->mapping->host;
36ade451 1711 struct ext4_io_submit io_submit;
1c8349a1 1712 bool keep_towrite = false;
61628a3f 1713
a9c667f8 1714 trace_ext4_writepage(page);
f0e6c985
AK
1715 size = i_size_read(inode);
1716 if (page->index == size >> PAGE_CACHE_SHIFT)
1717 len = size & ~PAGE_CACHE_MASK;
1718 else
1719 len = PAGE_CACHE_SIZE;
64769240 1720
a42afc5f 1721 page_bufs = page_buffers(page);
a42afc5f 1722 /*
fe386132
JK
1723 * We cannot do block allocation or other extent handling in this
1724 * function. If there are buffers needing that, we have to redirty
1725 * the page. But we may reach here when we do a journal commit via
1726 * journal_submit_inode_data_buffers() and in that case we must write
1727 * allocated buffers to achieve data=ordered mode guarantees.
a42afc5f 1728 */
f19d5870
TM
1729 if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
1730 ext4_bh_delay_or_unwritten)) {
f8bec370 1731 redirty_page_for_writepage(wbc, page);
fe386132
JK
1732 if (current->flags & PF_MEMALLOC) {
1733 /*
1734 * For memory cleaning there's no point in writing only
1735 * some buffers. So just bail out. Warn if we came here
1736 * from direct reclaim.
1737 */
1738 WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
1739 == PF_MEMALLOC);
f0e6c985
AK
1740 unlock_page(page);
1741 return 0;
1742 }
1c8349a1 1743 keep_towrite = true;
a42afc5f 1744 }
64769240 1745
cb20d518 1746 if (PageChecked(page) && ext4_should_journal_data(inode))
43ce1d23
AK
1747 /*
1748 * It's mmapped pagecache. Add buffers and journal it. There
1749 * doesn't seem much point in redirtying the page here.
1750 */
3f0ca309 1751 return __ext4_journalled_writepage(page, len);
43ce1d23 1752
97a851ed
JK
1753 ext4_io_submit_init(&io_submit, wbc);
1754 io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
1755 if (!io_submit.io_end) {
1756 redirty_page_for_writepage(wbc, page);
1757 unlock_page(page);
1758 return -ENOMEM;
1759 }
1c8349a1 1760 ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite);
36ade451 1761 ext4_io_submit(&io_submit);
97a851ed
JK
1762 /* Drop io_end reference we got from init */
1763 ext4_put_io_end_defer(io_submit.io_end);
64769240
AT
1764 return ret;
1765}
1766
5f1132b2
JK
1767static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page)
1768{
1769 int len;
1770 loff_t size = i_size_read(mpd->inode);
1771 int err;
1772
1773 BUG_ON(page->index != mpd->first_page);
1774 if (page->index == size >> PAGE_CACHE_SHIFT)
1775 len = size & ~PAGE_CACHE_MASK;
1776 else
1777 len = PAGE_CACHE_SIZE;
1778 clear_page_dirty_for_io(page);
1c8349a1 1779 err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false);
5f1132b2
JK
1780 if (!err)
1781 mpd->wbc->nr_to_write--;
1782 mpd->first_page++;
1783
1784 return err;
1785}
1786
4e7ea81d
JK
1787#define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
1788
61628a3f 1789/*
fffb2739
JK
1790 * mballoc gives us at most this number of blocks...
1791 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
70261f56 1792 * The rest of mballoc seems to handle chunks up to full group size.
61628a3f 1793 */
fffb2739 1794#define MAX_WRITEPAGES_EXTENT_LEN 2048
525f4ed8 1795
4e7ea81d
JK
1796/*
1797 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
1798 *
1799 * @mpd - extent of blocks
1800 * @lblk - logical number of the block in the file
09930042 1801 * @bh - buffer head we want to add to the extent
4e7ea81d 1802 *
09930042
JK
1803 * The function is used to collect contig. blocks in the same state. If the
1804 * buffer doesn't require mapping for writeback and we haven't started the
1805 * extent of buffers to map yet, the function returns 'true' immediately - the
1806 * caller can write the buffer right away. Otherwise the function returns true
1807 * if the block has been added to the extent, false if the block couldn't be
1808 * added.
4e7ea81d 1809 */
09930042
JK
1810static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
1811 struct buffer_head *bh)
4e7ea81d
JK
1812{
1813 struct ext4_map_blocks *map = &mpd->map;
1814
09930042
JK
1815 /* Buffer that doesn't need mapping for writeback? */
1816 if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
1817 (!buffer_delay(bh) && !buffer_unwritten(bh))) {
1818 /* So far no extent to map => we write the buffer right away */
1819 if (map->m_len == 0)
1820 return true;
1821 return false;
1822 }
4e7ea81d
JK
1823
1824 /* First block in the extent? */
1825 if (map->m_len == 0) {
1826 map->m_lblk = lblk;
1827 map->m_len = 1;
09930042
JK
1828 map->m_flags = bh->b_state & BH_FLAGS;
1829 return true;
4e7ea81d
JK
1830 }
1831
09930042
JK
1832 /* Don't go larger than mballoc is willing to allocate */
1833 if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
1834 return false;
1835
4e7ea81d
JK
1836 /* Can we merge the block to our big extent? */
1837 if (lblk == map->m_lblk + map->m_len &&
09930042 1838 (bh->b_state & BH_FLAGS) == map->m_flags) {
4e7ea81d 1839 map->m_len++;
09930042 1840 return true;
4e7ea81d 1841 }
09930042 1842 return false;
4e7ea81d
JK
1843}
1844
5f1132b2
JK
1845/*
1846 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
1847 *
1848 * @mpd - extent of blocks for mapping
1849 * @head - the first buffer in the page
1850 * @bh - buffer we should start processing from
1851 * @lblk - logical number of the block in the file corresponding to @bh
1852 *
1853 * Walk through page buffers from @bh upto @head (exclusive) and either submit
1854 * the page for IO if all buffers in this page were mapped and there's no
1855 * accumulated extent of buffers to map or add buffers in the page to the
1856 * extent of buffers to map. The function returns 1 if the caller can continue
1857 * by processing the next page, 0 if it should stop adding buffers to the
1858 * extent to map because we cannot extend it anymore. It can also return value
1859 * < 0 in case of error during IO submission.
1860 */
1861static int mpage_process_page_bufs(struct mpage_da_data *mpd,
1862 struct buffer_head *head,
1863 struct buffer_head *bh,
1864 ext4_lblk_t lblk)
4e7ea81d
JK
1865{
1866 struct inode *inode = mpd->inode;
5f1132b2 1867 int err;
4e7ea81d
JK
1868 ext4_lblk_t blocks = (i_size_read(inode) + (1 << inode->i_blkbits) - 1)
1869 >> inode->i_blkbits;
1870
1871 do {
1872 BUG_ON(buffer_locked(bh));
1873
09930042 1874 if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
4e7ea81d
JK
1875 /* Found extent to map? */
1876 if (mpd->map.m_len)
5f1132b2 1877 return 0;
09930042 1878 /* Everything mapped so far and we hit EOF */
5f1132b2 1879 break;
4e7ea81d 1880 }
4e7ea81d 1881 } while (lblk++, (bh = bh->b_this_page) != head);
5f1132b2
JK
1882 /* So far everything mapped? Submit the page for IO. */
1883 if (mpd->map.m_len == 0) {
1884 err = mpage_submit_page(mpd, head->b_page);
1885 if (err < 0)
1886 return err;
1887 }
1888 return lblk < blocks;
4e7ea81d
JK
1889}
1890
1891/*
1892 * mpage_map_buffers - update buffers corresponding to changed extent and
1893 * submit fully mapped pages for IO
1894 *
1895 * @mpd - description of extent to map, on return next extent to map
1896 *
1897 * Scan buffers corresponding to changed extent (we expect corresponding pages
1898 * to be already locked) and update buffer state according to new extent state.
1899 * We map delalloc buffers to their physical location, clear unwritten bits,
556615dc 1900 * and mark buffers as uninit when we perform writes to unwritten extents
4e7ea81d
JK
1901 * and do extent conversion after IO is finished. If the last page is not fully
1902 * mapped, we update @map to the next extent in the last page that needs
1903 * mapping. Otherwise we submit the page for IO.
1904 */
1905static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
1906{
1907 struct pagevec pvec;
1908 int nr_pages, i;
1909 struct inode *inode = mpd->inode;
1910 struct buffer_head *head, *bh;
1911 int bpp_bits = PAGE_CACHE_SHIFT - inode->i_blkbits;
4e7ea81d
JK
1912 pgoff_t start, end;
1913 ext4_lblk_t lblk;
1914 sector_t pblock;
1915 int err;
1916
1917 start = mpd->map.m_lblk >> bpp_bits;
1918 end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
1919 lblk = start << bpp_bits;
1920 pblock = mpd->map.m_pblk;
1921
1922 pagevec_init(&pvec, 0);
1923 while (start <= end) {
1924 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, start,
1925 PAGEVEC_SIZE);
1926 if (nr_pages == 0)
1927 break;
1928 for (i = 0; i < nr_pages; i++) {
1929 struct page *page = pvec.pages[i];
1930
1931 if (page->index > end)
1932 break;
70261f56 1933 /* Up to 'end' pages must be contiguous */
4e7ea81d
JK
1934 BUG_ON(page->index != start);
1935 bh = head = page_buffers(page);
1936 do {
1937 if (lblk < mpd->map.m_lblk)
1938 continue;
1939 if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
1940 /*
1941 * Buffer after end of mapped extent.
1942 * Find next buffer in the page to map.
1943 */
1944 mpd->map.m_len = 0;
1945 mpd->map.m_flags = 0;
5f1132b2
JK
1946 /*
1947 * FIXME: If dioread_nolock supports
1948 * blocksize < pagesize, we need to make
1949 * sure we add size mapped so far to
1950 * io_end->size as the following call
1951 * can submit the page for IO.
1952 */
1953 err = mpage_process_page_bufs(mpd, head,
1954 bh, lblk);
4e7ea81d 1955 pagevec_release(&pvec);
5f1132b2
JK
1956 if (err > 0)
1957 err = 0;
1958 return err;
4e7ea81d
JK
1959 }
1960 if (buffer_delay(bh)) {
1961 clear_buffer_delay(bh);
1962 bh->b_blocknr = pblock++;
1963 }
4e7ea81d 1964 clear_buffer_unwritten(bh);
5f1132b2 1965 } while (lblk++, (bh = bh->b_this_page) != head);
4e7ea81d
JK
1966
1967 /*
1968 * FIXME: This is going to break if dioread_nolock
1969 * supports blocksize < pagesize as we will try to
1970 * convert potentially unmapped parts of inode.
1971 */
1972 mpd->io_submit.io_end->size += PAGE_CACHE_SIZE;
1973 /* Page fully mapped - let IO run! */
1974 err = mpage_submit_page(mpd, page);
1975 if (err < 0) {
1976 pagevec_release(&pvec);
1977 return err;
1978 }
1979 start++;
1980 }
1981 pagevec_release(&pvec);
1982 }
1983 /* Extent fully mapped and matches with page boundary. We are done. */
1984 mpd->map.m_len = 0;
1985 mpd->map.m_flags = 0;
1986 return 0;
1987}
1988
1989static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
1990{
1991 struct inode *inode = mpd->inode;
1992 struct ext4_map_blocks *map = &mpd->map;
1993 int get_blocks_flags;
090f32ee 1994 int err, dioread_nolock;
4e7ea81d
JK
1995
1996 trace_ext4_da_write_pages_extent(inode, map);
1997 /*
1998 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
556615dc 1999 * to convert an unwritten extent to be initialized (in the case
4e7ea81d
JK
2000 * where we have written into one or more preallocated blocks). It is
2001 * possible that we're going to need more metadata blocks than
2002 * previously reserved. However we must not fail because we're in
2003 * writeback and there is nothing we can do about it so it might result
2004 * in data loss. So use reserved blocks to allocate metadata if
2005 * possible.
2006 *
2007 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if the blocks
2008 * in question are delalloc blocks. This affects functions in many
2009 * different parts of the allocation call path. This flag exists
2010 * primarily because we don't want to change *many* call functions, so
2011 * ext4_map_blocks() will set the EXT4_STATE_DELALLOC_RESERVED flag
2012 * once the inode's allocation semaphore is taken.
2013 */
2014 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2015 EXT4_GET_BLOCKS_METADATA_NOFAIL;
090f32ee
LC
2016 dioread_nolock = ext4_should_dioread_nolock(inode);
2017 if (dioread_nolock)
4e7ea81d
JK
2018 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2019 if (map->m_flags & (1 << BH_Delay))
2020 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2021
2022 err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2023 if (err < 0)
2024 return err;
090f32ee 2025 if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
6b523df4
JK
2026 if (!mpd->io_submit.io_end->handle &&
2027 ext4_handle_valid(handle)) {
2028 mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2029 handle->h_rsv_handle = NULL;
2030 }
3613d228 2031 ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
6b523df4 2032 }
4e7ea81d
JK
2033
2034 BUG_ON(map->m_len == 0);
2035 if (map->m_flags & EXT4_MAP_NEW) {
2036 struct block_device *bdev = inode->i_sb->s_bdev;
2037 int i;
2038
2039 for (i = 0; i < map->m_len; i++)
2040 unmap_underlying_metadata(bdev, map->m_pblk + i);
2041 }
2042 return 0;
2043}
2044
2045/*
2046 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2047 * mpd->len and submit pages underlying it for IO
2048 *
2049 * @handle - handle for journal operations
2050 * @mpd - extent to map
7534e854
JK
2051 * @give_up_on_write - we set this to true iff there is a fatal error and there
2052 * is no hope of writing the data. The caller should discard
2053 * dirty pages to avoid infinite loops.
4e7ea81d
JK
2054 *
2055 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2056 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2057 * them to initialized or split the described range from larger unwritten
2058 * extent. Note that we need not map all the described range since allocation
2059 * can return less blocks or the range is covered by more unwritten extents. We
2060 * cannot map more because we are limited by reserved transaction credits. On
2061 * the other hand we always make sure that the last touched page is fully
2062 * mapped so that it can be written out (and thus forward progress is
2063 * guaranteed). After mapping we submit all mapped pages for IO.
2064 */
2065static int mpage_map_and_submit_extent(handle_t *handle,
cb530541
TT
2066 struct mpage_da_data *mpd,
2067 bool *give_up_on_write)
4e7ea81d
JK
2068{
2069 struct inode *inode = mpd->inode;
2070 struct ext4_map_blocks *map = &mpd->map;
2071 int err;
2072 loff_t disksize;
6603120e 2073 int progress = 0;
4e7ea81d
JK
2074
2075 mpd->io_submit.io_end->offset =
2076 ((loff_t)map->m_lblk) << inode->i_blkbits;
27d7c4ed 2077 do {
4e7ea81d
JK
2078 err = mpage_map_one_extent(handle, mpd);
2079 if (err < 0) {
2080 struct super_block *sb = inode->i_sb;
2081
cb530541
TT
2082 if (EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)
2083 goto invalidate_dirty_pages;
4e7ea81d 2084 /*
cb530541
TT
2085 * Let the uper layers retry transient errors.
2086 * In the case of ENOSPC, if ext4_count_free_blocks()
2087 * is non-zero, a commit should free up blocks.
4e7ea81d 2088 */
cb530541 2089 if ((err == -ENOMEM) ||
6603120e
DM
2090 (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2091 if (progress)
2092 goto update_disksize;
cb530541 2093 return err;
6603120e 2094 }
cb530541
TT
2095 ext4_msg(sb, KERN_CRIT,
2096 "Delayed block allocation failed for "
2097 "inode %lu at logical offset %llu with"
2098 " max blocks %u with error %d",
2099 inode->i_ino,
2100 (unsigned long long)map->m_lblk,
2101 (unsigned)map->m_len, -err);
2102 ext4_msg(sb, KERN_CRIT,
2103 "This should not happen!! Data will "
2104 "be lost\n");
2105 if (err == -ENOSPC)
2106 ext4_print_free_blocks(inode);
2107 invalidate_dirty_pages:
2108 *give_up_on_write = true;
4e7ea81d
JK
2109 return err;
2110 }
6603120e 2111 progress = 1;
4e7ea81d
JK
2112 /*
2113 * Update buffer state, submit mapped pages, and get us new
2114 * extent to map
2115 */
2116 err = mpage_map_and_submit_buffers(mpd);
2117 if (err < 0)
6603120e 2118 goto update_disksize;
27d7c4ed 2119 } while (map->m_len);
4e7ea81d 2120
6603120e 2121update_disksize:
622cad13
TT
2122 /*
2123 * Update on-disk size after IO is submitted. Races with
2124 * truncate are avoided by checking i_size under i_data_sem.
2125 */
4e7ea81d 2126 disksize = ((loff_t)mpd->first_page) << PAGE_CACHE_SHIFT;
4e7ea81d
JK
2127 if (disksize > EXT4_I(inode)->i_disksize) {
2128 int err2;
622cad13
TT
2129 loff_t i_size;
2130
2131 down_write(&EXT4_I(inode)->i_data_sem);
2132 i_size = i_size_read(inode);
2133 if (disksize > i_size)
2134 disksize = i_size;
2135 if (disksize > EXT4_I(inode)->i_disksize)
2136 EXT4_I(inode)->i_disksize = disksize;
4e7ea81d 2137 err2 = ext4_mark_inode_dirty(handle, inode);
622cad13 2138 up_write(&EXT4_I(inode)->i_data_sem);
4e7ea81d
JK
2139 if (err2)
2140 ext4_error(inode->i_sb,
2141 "Failed to mark inode %lu dirty",
2142 inode->i_ino);
2143 if (!err)
2144 err = err2;
2145 }
2146 return err;
2147}
2148
fffb2739
JK
2149/*
2150 * Calculate the total number of credits to reserve for one writepages
20970ba6 2151 * iteration. This is called from ext4_writepages(). We map an extent of
70261f56 2152 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
fffb2739
JK
2153 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2154 * bpp - 1 blocks in bpp different extents.
2155 */
525f4ed8
MC
2156static int ext4_da_writepages_trans_blocks(struct inode *inode)
2157{
fffb2739 2158 int bpp = ext4_journal_blocks_per_page(inode);
525f4ed8 2159
fffb2739
JK
2160 return ext4_meta_trans_blocks(inode,
2161 MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
525f4ed8 2162}
61628a3f 2163
8e48dcfb 2164/*
4e7ea81d
JK
2165 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2166 * and underlying extent to map
2167 *
2168 * @mpd - where to look for pages
2169 *
2170 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2171 * IO immediately. When we find a page which isn't mapped we start accumulating
2172 * extent of buffers underlying these pages that needs mapping (formed by
2173 * either delayed or unwritten buffers). We also lock the pages containing
2174 * these buffers. The extent found is returned in @mpd structure (starting at
2175 * mpd->lblk with length mpd->len blocks).
2176 *
2177 * Note that this function can attach bios to one io_end structure which are
2178 * neither logically nor physically contiguous. Although it may seem as an
2179 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2180 * case as we need to track IO to all buffers underlying a page in one io_end.
8e48dcfb 2181 */
4e7ea81d 2182static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
8e48dcfb 2183{
4e7ea81d
JK
2184 struct address_space *mapping = mpd->inode->i_mapping;
2185 struct pagevec pvec;
2186 unsigned int nr_pages;
aeac589a 2187 long left = mpd->wbc->nr_to_write;
4e7ea81d
JK
2188 pgoff_t index = mpd->first_page;
2189 pgoff_t end = mpd->last_page;
2190 int tag;
2191 int i, err = 0;
2192 int blkbits = mpd->inode->i_blkbits;
2193 ext4_lblk_t lblk;
2194 struct buffer_head *head;
8e48dcfb 2195
4e7ea81d 2196 if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
5b41d924
ES
2197 tag = PAGECACHE_TAG_TOWRITE;
2198 else
2199 tag = PAGECACHE_TAG_DIRTY;
2200
4e7ea81d
JK
2201 pagevec_init(&pvec, 0);
2202 mpd->map.m_len = 0;
2203 mpd->next_page = index;
4f01b02c 2204 while (index <= end) {
5b41d924 2205 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag,
8e48dcfb
TT
2206 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
2207 if (nr_pages == 0)
4e7ea81d 2208 goto out;
8e48dcfb
TT
2209
2210 for (i = 0; i < nr_pages; i++) {
2211 struct page *page = pvec.pages[i];
2212
2213 /*
2214 * At this point, the page may be truncated or
2215 * invalidated (changing page->mapping to NULL), or
2216 * even swizzled back from swapper_space to tmpfs file
2217 * mapping. However, page->index will not change
2218 * because we have a reference on the page.
2219 */
4f01b02c
TT
2220 if (page->index > end)
2221 goto out;
8e48dcfb 2222
aeac589a
ML
2223 /*
2224 * Accumulated enough dirty pages? This doesn't apply
2225 * to WB_SYNC_ALL mode. For integrity sync we have to
2226 * keep going because someone may be concurrently
2227 * dirtying pages, and we might have synced a lot of
2228 * newly appeared dirty pages, but have not synced all
2229 * of the old dirty pages.
2230 */
2231 if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0)
2232 goto out;
2233
4e7ea81d
JK
2234 /* If we can't merge this page, we are done. */
2235 if (mpd->map.m_len > 0 && mpd->next_page != page->index)
2236 goto out;
78aaced3 2237
8e48dcfb 2238 lock_page(page);
8e48dcfb 2239 /*
4e7ea81d
JK
2240 * If the page is no longer dirty, or its mapping no
2241 * longer corresponds to inode we are writing (which
2242 * means it has been truncated or invalidated), or the
2243 * page is already under writeback and we are not doing
2244 * a data integrity writeback, skip the page
8e48dcfb 2245 */
4f01b02c
TT
2246 if (!PageDirty(page) ||
2247 (PageWriteback(page) &&
4e7ea81d 2248 (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
4f01b02c 2249 unlikely(page->mapping != mapping)) {
8e48dcfb
TT
2250 unlock_page(page);
2251 continue;
2252 }
2253
7cb1a535 2254 wait_on_page_writeback(page);
8e48dcfb 2255 BUG_ON(PageWriteback(page));
8e48dcfb 2256
4e7ea81d 2257 if (mpd->map.m_len == 0)
8eb9e5ce 2258 mpd->first_page = page->index;
8eb9e5ce 2259 mpd->next_page = page->index + 1;
f8bec370 2260 /* Add all dirty buffers to mpd */
4e7ea81d
JK
2261 lblk = ((ext4_lblk_t)page->index) <<
2262 (PAGE_CACHE_SHIFT - blkbits);
f8bec370 2263 head = page_buffers(page);
5f1132b2
JK
2264 err = mpage_process_page_bufs(mpd, head, head, lblk);
2265 if (err <= 0)
4e7ea81d 2266 goto out;
5f1132b2 2267 err = 0;
aeac589a 2268 left--;
8e48dcfb
TT
2269 }
2270 pagevec_release(&pvec);
2271 cond_resched();
2272 }
4f01b02c 2273 return 0;
8eb9e5ce
TT
2274out:
2275 pagevec_release(&pvec);
4e7ea81d 2276 return err;
8e48dcfb
TT
2277}
2278
20970ba6
TT
2279static int __writepage(struct page *page, struct writeback_control *wbc,
2280 void *data)
2281{
2282 struct address_space *mapping = data;
2283 int ret = ext4_writepage(page, wbc);
2284 mapping_set_error(mapping, ret);
2285 return ret;
2286}
2287
2288static int ext4_writepages(struct address_space *mapping,
2289 struct writeback_control *wbc)
64769240 2290{
4e7ea81d
JK
2291 pgoff_t writeback_index = 0;
2292 long nr_to_write = wbc->nr_to_write;
22208ded 2293 int range_whole = 0;
4e7ea81d 2294 int cycled = 1;
61628a3f 2295 handle_t *handle = NULL;
df22291f 2296 struct mpage_da_data mpd;
5e745b04 2297 struct inode *inode = mapping->host;
6b523df4 2298 int needed_blocks, rsv_blocks = 0, ret = 0;
5e745b04 2299 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
4e7ea81d 2300 bool done;
1bce63d1 2301 struct blk_plug plug;
cb530541 2302 bool give_up_on_write = false;
61628a3f 2303
20970ba6 2304 trace_ext4_writepages(inode, wbc);
ba80b101 2305
61628a3f
MC
2306 /*
2307 * No pages to write? This is mainly a kludge to avoid starting
2308 * a transaction for special inodes like journal inode on last iput()
2309 * because that could violate lock ordering on umount
2310 */
a1d6cc56 2311 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
bbf023c7 2312 goto out_writepages;
2a21e37e 2313
20970ba6
TT
2314 if (ext4_should_journal_data(inode)) {
2315 struct blk_plug plug;
20970ba6
TT
2316
2317 blk_start_plug(&plug);
2318 ret = write_cache_pages(mapping, wbc, __writepage, mapping);
2319 blk_finish_plug(&plug);
bbf023c7 2320 goto out_writepages;
20970ba6
TT
2321 }
2322
2a21e37e
TT
2323 /*
2324 * If the filesystem has aborted, it is read-only, so return
2325 * right away instead of dumping stack traces later on that
2326 * will obscure the real source of the problem. We test
4ab2f15b 2327 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2a21e37e 2328 * the latter could be true if the filesystem is mounted
20970ba6 2329 * read-only, and in that case, ext4_writepages should
2a21e37e
TT
2330 * *never* be called, so if that ever happens, we would want
2331 * the stack trace.
2332 */
bbf023c7
ML
2333 if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) {
2334 ret = -EROFS;
2335 goto out_writepages;
2336 }
2a21e37e 2337
6b523df4
JK
2338 if (ext4_should_dioread_nolock(inode)) {
2339 /*
70261f56 2340 * We may need to convert up to one extent per block in
6b523df4
JK
2341 * the page and we may dirty the inode.
2342 */
2343 rsv_blocks = 1 + (PAGE_CACHE_SIZE >> inode->i_blkbits);
2344 }
2345
4e7ea81d
JK
2346 /*
2347 * If we have inline data and arrive here, it means that
2348 * we will soon create the block for the 1st page, so
2349 * we'd better clear the inline data here.
2350 */
2351 if (ext4_has_inline_data(inode)) {
2352 /* Just inode will be modified... */
2353 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2354 if (IS_ERR(handle)) {
2355 ret = PTR_ERR(handle);
2356 goto out_writepages;
2357 }
2358 BUG_ON(ext4_test_inode_state(inode,
2359 EXT4_STATE_MAY_INLINE_DATA));
2360 ext4_destroy_inline_data(handle, inode);
2361 ext4_journal_stop(handle);
2362 }
2363
22208ded
AK
2364 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2365 range_whole = 1;
61628a3f 2366
2acf2c26 2367 if (wbc->range_cyclic) {
4e7ea81d
JK
2368 writeback_index = mapping->writeback_index;
2369 if (writeback_index)
2acf2c26 2370 cycled = 0;
4e7ea81d
JK
2371 mpd.first_page = writeback_index;
2372 mpd.last_page = -1;
5b41d924 2373 } else {
4e7ea81d
JK
2374 mpd.first_page = wbc->range_start >> PAGE_CACHE_SHIFT;
2375 mpd.last_page = wbc->range_end >> PAGE_CACHE_SHIFT;
5b41d924 2376 }
a1d6cc56 2377
4e7ea81d
JK
2378 mpd.inode = inode;
2379 mpd.wbc = wbc;
2380 ext4_io_submit_init(&mpd.io_submit, wbc);
2acf2c26 2381retry:
6e6938b6 2382 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
4e7ea81d
JK
2383 tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page);
2384 done = false;
1bce63d1 2385 blk_start_plug(&plug);
4e7ea81d
JK
2386 while (!done && mpd.first_page <= mpd.last_page) {
2387 /* For each extent of pages we use new io_end */
2388 mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2389 if (!mpd.io_submit.io_end) {
2390 ret = -ENOMEM;
2391 break;
2392 }
a1d6cc56
AK
2393
2394 /*
4e7ea81d
JK
2395 * We have two constraints: We find one extent to map and we
2396 * must always write out whole page (makes a difference when
2397 * blocksize < pagesize) so that we don't block on IO when we
2398 * try to write out the rest of the page. Journalled mode is
2399 * not supported by delalloc.
a1d6cc56
AK
2400 */
2401 BUG_ON(ext4_should_journal_data(inode));
525f4ed8 2402 needed_blocks = ext4_da_writepages_trans_blocks(inode);
a1d6cc56 2403
4e7ea81d 2404 /* start a new transaction */
6b523df4
JK
2405 handle = ext4_journal_start_with_reserve(inode,
2406 EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
61628a3f
MC
2407 if (IS_ERR(handle)) {
2408 ret = PTR_ERR(handle);
1693918e 2409 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
fbe845dd 2410 "%ld pages, ino %lu; err %d", __func__,
a1d6cc56 2411 wbc->nr_to_write, inode->i_ino, ret);
4e7ea81d
JK
2412 /* Release allocated io_end */
2413 ext4_put_io_end(mpd.io_submit.io_end);
2414 break;
61628a3f 2415 }
f63e6005 2416
4e7ea81d
JK
2417 trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc);
2418 ret = mpage_prepare_extent_to_map(&mpd);
2419 if (!ret) {
2420 if (mpd.map.m_len)
cb530541
TT
2421 ret = mpage_map_and_submit_extent(handle, &mpd,
2422 &give_up_on_write);
4e7ea81d
JK
2423 else {
2424 /*
2425 * We scanned the whole range (or exhausted
2426 * nr_to_write), submitted what was mapped and
2427 * didn't find anything needing mapping. We are
2428 * done.
2429 */
2430 done = true;
2431 }
f63e6005 2432 }
61628a3f 2433 ext4_journal_stop(handle);
4e7ea81d
JK
2434 /* Submit prepared bio */
2435 ext4_io_submit(&mpd.io_submit);
2436 /* Unlock pages we didn't use */
cb530541 2437 mpage_release_unused_pages(&mpd, give_up_on_write);
4e7ea81d
JK
2438 /* Drop our io_end reference we got from init */
2439 ext4_put_io_end(mpd.io_submit.io_end);
2440
2441 if (ret == -ENOSPC && sbi->s_journal) {
2442 /*
2443 * Commit the transaction which would
22208ded
AK
2444 * free blocks released in the transaction
2445 * and try again
2446 */
df22291f 2447 jbd2_journal_force_commit_nested(sbi->s_journal);
22208ded 2448 ret = 0;
4e7ea81d
JK
2449 continue;
2450 }
2451 /* Fatal error - ENOMEM, EIO... */
2452 if (ret)
61628a3f 2453 break;
a1d6cc56 2454 }
1bce63d1 2455 blk_finish_plug(&plug);
9c12a831 2456 if (!ret && !cycled && wbc->nr_to_write > 0) {
2acf2c26 2457 cycled = 1;
4e7ea81d
JK
2458 mpd.last_page = writeback_index - 1;
2459 mpd.first_page = 0;
2acf2c26
AK
2460 goto retry;
2461 }
22208ded
AK
2462
2463 /* Update index */
22208ded
AK
2464 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2465 /*
4e7ea81d 2466 * Set the writeback_index so that range_cyclic
22208ded
AK
2467 * mode will write it back later
2468 */
4e7ea81d 2469 mapping->writeback_index = mpd.first_page;
a1d6cc56 2470
61628a3f 2471out_writepages:
20970ba6
TT
2472 trace_ext4_writepages_result(inode, wbc, ret,
2473 nr_to_write - wbc->nr_to_write);
61628a3f 2474 return ret;
64769240
AT
2475}
2476
79f0be8d
AK
2477static int ext4_nonda_switch(struct super_block *sb)
2478{
5c1ff336 2479 s64 free_clusters, dirty_clusters;
79f0be8d
AK
2480 struct ext4_sb_info *sbi = EXT4_SB(sb);
2481
2482 /*
2483 * switch to non delalloc mode if we are running low
2484 * on free block. The free block accounting via percpu
179f7ebf 2485 * counters can get slightly wrong with percpu_counter_batch getting
79f0be8d
AK
2486 * accumulated on each CPU without updating global counters
2487 * Delalloc need an accurate free block accounting. So switch
2488 * to non delalloc when we are near to error range.
2489 */
5c1ff336
EW
2490 free_clusters =
2491 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2492 dirty_clusters =
2493 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
00d4e736
TT
2494 /*
2495 * Start pushing delalloc when 1/2 of free blocks are dirty.
2496 */
5c1ff336 2497 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
10ee27a0 2498 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
00d4e736 2499
5c1ff336
EW
2500 if (2 * free_clusters < 3 * dirty_clusters ||
2501 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
79f0be8d 2502 /*
c8afb446
ES
2503 * free block count is less than 150% of dirty blocks
2504 * or free blocks is less than watermark
79f0be8d
AK
2505 */
2506 return 1;
2507 }
2508 return 0;
2509}
2510
64769240 2511static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
de9a55b8
TT
2512 loff_t pos, unsigned len, unsigned flags,
2513 struct page **pagep, void **fsdata)
64769240 2514{
72b8ab9d 2515 int ret, retries = 0;
64769240
AT
2516 struct page *page;
2517 pgoff_t index;
64769240
AT
2518 struct inode *inode = mapping->host;
2519 handle_t *handle;
2520
2521 index = pos >> PAGE_CACHE_SHIFT;
79f0be8d
AK
2522
2523 if (ext4_nonda_switch(inode->i_sb)) {
2524 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2525 return ext4_write_begin(file, mapping, pos,
2526 len, flags, pagep, fsdata);
2527 }
2528 *fsdata = (void *)0;
9bffad1e 2529 trace_ext4_da_write_begin(inode, pos, len, flags);
9c3569b5
TM
2530
2531 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2532 ret = ext4_da_write_inline_data_begin(mapping, inode,
2533 pos, len, flags,
2534 pagep, fsdata);
2535 if (ret < 0)
47564bfb
TT
2536 return ret;
2537 if (ret == 1)
2538 return 0;
9c3569b5
TM
2539 }
2540
47564bfb
TT
2541 /*
2542 * grab_cache_page_write_begin() can take a long time if the
2543 * system is thrashing due to memory pressure, or if the page
2544 * is being written back. So grab it first before we start
2545 * the transaction handle. This also allows us to allocate
2546 * the page (if needed) without using GFP_NOFS.
2547 */
2548retry_grab:
2549 page = grab_cache_page_write_begin(mapping, index, flags);
2550 if (!page)
2551 return -ENOMEM;
2552 unlock_page(page);
2553
64769240
AT
2554 /*
2555 * With delayed allocation, we don't log the i_disksize update
2556 * if there is delayed block allocation. But we still need
2557 * to journalling the i_disksize update if writes to the end
2558 * of file which has an already mapped buffer.
2559 */
47564bfb 2560retry_journal:
9924a92a 2561 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, 1);
64769240 2562 if (IS_ERR(handle)) {
47564bfb
TT
2563 page_cache_release(page);
2564 return PTR_ERR(handle);
64769240
AT
2565 }
2566
47564bfb
TT
2567 lock_page(page);
2568 if (page->mapping != mapping) {
2569 /* The page got truncated from under us */
2570 unlock_page(page);
2571 page_cache_release(page);
d5a0d4f7 2572 ext4_journal_stop(handle);
47564bfb 2573 goto retry_grab;
d5a0d4f7 2574 }
47564bfb 2575 /* In case writeback began while the page was unlocked */
7afe5aa5 2576 wait_for_stable_page(page);
64769240 2577
6e1db88d 2578 ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
64769240
AT
2579 if (ret < 0) {
2580 unlock_page(page);
2581 ext4_journal_stop(handle);
ae4d5372
AK
2582 /*
2583 * block_write_begin may have instantiated a few blocks
2584 * outside i_size. Trim these off again. Don't need
2585 * i_size_read because we hold i_mutex.
2586 */
2587 if (pos + len > inode->i_size)
b9a4207d 2588 ext4_truncate_failed_write(inode);
47564bfb
TT
2589
2590 if (ret == -ENOSPC &&
2591 ext4_should_retry_alloc(inode->i_sb, &retries))
2592 goto retry_journal;
2593
2594 page_cache_release(page);
2595 return ret;
64769240
AT
2596 }
2597
47564bfb 2598 *pagep = page;
64769240
AT
2599 return ret;
2600}
2601
632eaeab
MC
2602/*
2603 * Check if we should update i_disksize
2604 * when write to the end of file but not require block allocation
2605 */
2606static int ext4_da_should_update_i_disksize(struct page *page,
de9a55b8 2607 unsigned long offset)
632eaeab
MC
2608{
2609 struct buffer_head *bh;
2610 struct inode *inode = page->mapping->host;
2611 unsigned int idx;
2612 int i;
2613
2614 bh = page_buffers(page);
2615 idx = offset >> inode->i_blkbits;
2616
af5bc92d 2617 for (i = 0; i < idx; i++)
632eaeab
MC
2618 bh = bh->b_this_page;
2619
29fa89d0 2620 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
632eaeab
MC
2621 return 0;
2622 return 1;
2623}
2624
64769240 2625static int ext4_da_write_end(struct file *file,
de9a55b8
TT
2626 struct address_space *mapping,
2627 loff_t pos, unsigned len, unsigned copied,
2628 struct page *page, void *fsdata)
64769240
AT
2629{
2630 struct inode *inode = mapping->host;
2631 int ret = 0, ret2;
2632 handle_t *handle = ext4_journal_current_handle();
2633 loff_t new_i_size;
632eaeab 2634 unsigned long start, end;
79f0be8d
AK
2635 int write_mode = (int)(unsigned long)fsdata;
2636
74d553aa
TT
2637 if (write_mode == FALL_BACK_TO_NONDELALLOC)
2638 return ext4_write_end(file, mapping, pos,
2639 len, copied, page, fsdata);
632eaeab 2640
9bffad1e 2641 trace_ext4_da_write_end(inode, pos, len, copied);
632eaeab 2642 start = pos & (PAGE_CACHE_SIZE - 1);
af5bc92d 2643 end = start + copied - 1;
64769240
AT
2644
2645 /*
2646 * generic_write_end() will run mark_inode_dirty() if i_size
2647 * changes. So let's piggyback the i_disksize mark_inode_dirty
2648 * into that.
2649 */
64769240 2650 new_i_size = pos + copied;
ea51d132 2651 if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
9c3569b5
TM
2652 if (ext4_has_inline_data(inode) ||
2653 ext4_da_should_update_i_disksize(page, end)) {
ee124d27 2654 ext4_update_i_disksize(inode, new_i_size);
cf17fea6
AK
2655 /* We need to mark inode dirty even if
2656 * new_i_size is less that inode->i_size
2657 * bu greater than i_disksize.(hint delalloc)
2658 */
2659 ext4_mark_inode_dirty(handle, inode);
64769240 2660 }
632eaeab 2661 }
9c3569b5
TM
2662
2663 if (write_mode != CONVERT_INLINE_DATA &&
2664 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
2665 ext4_has_inline_data(inode))
2666 ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
2667 page);
2668 else
2669 ret2 = generic_write_end(file, mapping, pos, len, copied,
64769240 2670 page, fsdata);
9c3569b5 2671
64769240
AT
2672 copied = ret2;
2673 if (ret2 < 0)
2674 ret = ret2;
2675 ret2 = ext4_journal_stop(handle);
2676 if (!ret)
2677 ret = ret2;
2678
2679 return ret ? ret : copied;
2680}
2681
d47992f8
LC
2682static void ext4_da_invalidatepage(struct page *page, unsigned int offset,
2683 unsigned int length)
64769240 2684{
64769240
AT
2685 /*
2686 * Drop reserved blocks
2687 */
2688 BUG_ON(!PageLocked(page));
2689 if (!page_has_buffers(page))
2690 goto out;
2691
ca99fdd2 2692 ext4_da_page_release_reservation(page, offset, length);
64769240
AT
2693
2694out:
d47992f8 2695 ext4_invalidatepage(page, offset, length);
64769240
AT
2696
2697 return;
2698}
2699
ccd2506b
TT
2700/*
2701 * Force all delayed allocation blocks to be allocated for a given inode.
2702 */
2703int ext4_alloc_da_blocks(struct inode *inode)
2704{
fb40ba0d
TT
2705 trace_ext4_alloc_da_blocks(inode);
2706
71d4f7d0 2707 if (!EXT4_I(inode)->i_reserved_data_blocks)
ccd2506b
TT
2708 return 0;
2709
2710 /*
2711 * We do something simple for now. The filemap_flush() will
2712 * also start triggering a write of the data blocks, which is
2713 * not strictly speaking necessary (and for users of
2714 * laptop_mode, not even desirable). However, to do otherwise
2715 * would require replicating code paths in:
de9a55b8 2716 *
20970ba6 2717 * ext4_writepages() ->
ccd2506b
TT
2718 * write_cache_pages() ---> (via passed in callback function)
2719 * __mpage_da_writepage() -->
2720 * mpage_add_bh_to_extent()
2721 * mpage_da_map_blocks()
2722 *
2723 * The problem is that write_cache_pages(), located in
2724 * mm/page-writeback.c, marks pages clean in preparation for
2725 * doing I/O, which is not desirable if we're not planning on
2726 * doing I/O at all.
2727 *
2728 * We could call write_cache_pages(), and then redirty all of
380cf090 2729 * the pages by calling redirty_page_for_writepage() but that
ccd2506b
TT
2730 * would be ugly in the extreme. So instead we would need to
2731 * replicate parts of the code in the above functions,
25985edc 2732 * simplifying them because we wouldn't actually intend to
ccd2506b
TT
2733 * write out the pages, but rather only collect contiguous
2734 * logical block extents, call the multi-block allocator, and
2735 * then update the buffer heads with the block allocations.
de9a55b8 2736 *
ccd2506b
TT
2737 * For now, though, we'll cheat by calling filemap_flush(),
2738 * which will map the blocks, and start the I/O, but not
2739 * actually wait for the I/O to complete.
2740 */
2741 return filemap_flush(inode->i_mapping);
2742}
64769240 2743
ac27a0ec
DK
2744/*
2745 * bmap() is special. It gets used by applications such as lilo and by
2746 * the swapper to find the on-disk block of a specific piece of data.
2747 *
2748 * Naturally, this is dangerous if the block concerned is still in the
617ba13b 2749 * journal. If somebody makes a swapfile on an ext4 data-journaling
ac27a0ec
DK
2750 * filesystem and enables swap, then they may get a nasty shock when the
2751 * data getting swapped to that swapfile suddenly gets overwritten by
2752 * the original zero's written out previously to the journal and
2753 * awaiting writeback in the kernel's buffer cache.
2754 *
2755 * So, if we see any bmap calls here on a modified, data-journaled file,
2756 * take extra steps to flush any blocks which might be in the cache.
2757 */
617ba13b 2758static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
ac27a0ec
DK
2759{
2760 struct inode *inode = mapping->host;
2761 journal_t *journal;
2762 int err;
2763
46c7f254
TM
2764 /*
2765 * We can get here for an inline file via the FIBMAP ioctl
2766 */
2767 if (ext4_has_inline_data(inode))
2768 return 0;
2769
64769240
AT
2770 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
2771 test_opt(inode->i_sb, DELALLOC)) {
2772 /*
2773 * With delalloc we want to sync the file
2774 * so that we can make sure we allocate
2775 * blocks for file
2776 */
2777 filemap_write_and_wait(mapping);
2778 }
2779
19f5fb7a
TT
2780 if (EXT4_JOURNAL(inode) &&
2781 ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
ac27a0ec
DK
2782 /*
2783 * This is a REALLY heavyweight approach, but the use of
2784 * bmap on dirty files is expected to be extremely rare:
2785 * only if we run lilo or swapon on a freshly made file
2786 * do we expect this to happen.
2787 *
2788 * (bmap requires CAP_SYS_RAWIO so this does not
2789 * represent an unprivileged user DOS attack --- we'd be
2790 * in trouble if mortal users could trigger this path at
2791 * will.)
2792 *
617ba13b 2793 * NB. EXT4_STATE_JDATA is not set on files other than
ac27a0ec
DK
2794 * regular files. If somebody wants to bmap a directory
2795 * or symlink and gets confused because the buffer
2796 * hasn't yet been flushed to disk, they deserve
2797 * everything they get.
2798 */
2799
19f5fb7a 2800 ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
617ba13b 2801 journal = EXT4_JOURNAL(inode);
dab291af
MC
2802 jbd2_journal_lock_updates(journal);
2803 err = jbd2_journal_flush(journal);
2804 jbd2_journal_unlock_updates(journal);
ac27a0ec
DK
2805
2806 if (err)
2807 return 0;
2808 }
2809
af5bc92d 2810 return generic_block_bmap(mapping, block, ext4_get_block);
ac27a0ec
DK
2811}
2812
617ba13b 2813static int ext4_readpage(struct file *file, struct page *page)
ac27a0ec 2814{
46c7f254
TM
2815 int ret = -EAGAIN;
2816 struct inode *inode = page->mapping->host;
2817
0562e0ba 2818 trace_ext4_readpage(page);
46c7f254
TM
2819
2820 if (ext4_has_inline_data(inode))
2821 ret = ext4_readpage_inline(inode, page);
2822
2823 if (ret == -EAGAIN)
2824 return mpage_readpage(page, ext4_get_block);
2825
2826 return ret;
ac27a0ec
DK
2827}
2828
2829static int
617ba13b 2830ext4_readpages(struct file *file, struct address_space *mapping,
ac27a0ec
DK
2831 struct list_head *pages, unsigned nr_pages)
2832{
46c7f254
TM
2833 struct inode *inode = mapping->host;
2834
2835 /* If the file has inline data, no need to do readpages. */
2836 if (ext4_has_inline_data(inode))
2837 return 0;
2838
617ba13b 2839 return mpage_readpages(mapping, pages, nr_pages, ext4_get_block);
ac27a0ec
DK
2840}
2841
d47992f8
LC
2842static void ext4_invalidatepage(struct page *page, unsigned int offset,
2843 unsigned int length)
ac27a0ec 2844{
ca99fdd2 2845 trace_ext4_invalidatepage(page, offset, length);
0562e0ba 2846
4520fb3c
JK
2847 /* No journalling happens on data buffers when this function is used */
2848 WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
2849
ca99fdd2 2850 block_invalidatepage(page, offset, length);
4520fb3c
JK
2851}
2852
53e87268 2853static int __ext4_journalled_invalidatepage(struct page *page,
ca99fdd2
LC
2854 unsigned int offset,
2855 unsigned int length)
4520fb3c
JK
2856{
2857 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
2858
ca99fdd2 2859 trace_ext4_journalled_invalidatepage(page, offset, length);
4520fb3c 2860
ac27a0ec
DK
2861 /*
2862 * If it's a full truncate we just forget about the pending dirtying
2863 */
ca99fdd2 2864 if (offset == 0 && length == PAGE_CACHE_SIZE)
ac27a0ec
DK
2865 ClearPageChecked(page);
2866
ca99fdd2 2867 return jbd2_journal_invalidatepage(journal, page, offset, length);
53e87268
JK
2868}
2869
2870/* Wrapper for aops... */
2871static void ext4_journalled_invalidatepage(struct page *page,
d47992f8
LC
2872 unsigned int offset,
2873 unsigned int length)
53e87268 2874{
ca99fdd2 2875 WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0);
ac27a0ec
DK
2876}
2877
617ba13b 2878static int ext4_releasepage(struct page *page, gfp_t wait)
ac27a0ec 2879{
617ba13b 2880 journal_t *journal = EXT4_JOURNAL(page->mapping->host);
ac27a0ec 2881
0562e0ba
JZ
2882 trace_ext4_releasepage(page);
2883
e1c36595
JK
2884 /* Page has dirty journalled data -> cannot release */
2885 if (PageChecked(page))
ac27a0ec 2886 return 0;
0390131b
FM
2887 if (journal)
2888 return jbd2_journal_try_to_free_buffers(journal, page, wait);
2889 else
2890 return try_to_free_buffers(page);
ac27a0ec
DK
2891}
2892
2ed88685
TT
2893/*
2894 * ext4_get_block used when preparing for a DIO write or buffer write.
2895 * We allocate an uinitialized extent if blocks haven't been allocated.
2896 * The extent will be converted to initialized after the IO is complete.
2897 */
f19d5870 2898int ext4_get_block_write(struct inode *inode, sector_t iblock,
4c0425ff
MC
2899 struct buffer_head *bh_result, int create)
2900{
c7064ef1 2901 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
8d5d02e6 2902 inode->i_ino, create);
2ed88685
TT
2903 return _ext4_get_block(inode, iblock, bh_result,
2904 EXT4_GET_BLOCKS_IO_CREATE_EXT);
4c0425ff
MC
2905}
2906
729f52c6 2907static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock,
8b0f165f 2908 struct buffer_head *bh_result, int create)
729f52c6 2909{
8b0f165f
AP
2910 ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n",
2911 inode->i_ino, create);
2912 return _ext4_get_block(inode, iblock, bh_result,
2913 EXT4_GET_BLOCKS_NO_LOCK);
729f52c6
ZL
2914}
2915
4c0425ff 2916static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset,
7b7a8665 2917 ssize_t size, void *private)
4c0425ff
MC
2918{
2919 ext4_io_end_t *io_end = iocb->private;
4c0425ff 2920
97a851ed 2921 /* if not async direct IO just return */
7b7a8665 2922 if (!io_end)
97a851ed 2923 return;
4b70df18 2924
88635ca2 2925 ext_debug("ext4_end_io_dio(): io_end 0x%p "
ace36ad4 2926 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
8d5d02e6
MC
2927 iocb->private, io_end->inode->i_ino, iocb, offset,
2928 size);
8d5d02e6 2929
b5a7e970 2930 iocb->private = NULL;
4c0425ff
MC
2931 io_end->offset = offset;
2932 io_end->size = size;
7b7a8665 2933 ext4_put_io_end(io_end);
4c0425ff 2934}
c7064ef1 2935
4c0425ff
MC
2936/*
2937 * For ext4 extent files, ext4 will do direct-io write to holes,
2938 * preallocated extents, and those write extend the file, no need to
2939 * fall back to buffered IO.
2940 *
556615dc 2941 * For holes, we fallocate those blocks, mark them as unwritten
69c499d1 2942 * If those blocks were preallocated, we mark sure they are split, but
556615dc 2943 * still keep the range to write as unwritten.
4c0425ff 2944 *
69c499d1 2945 * The unwritten extents will be converted to written when DIO is completed.
8d5d02e6 2946 * For async direct IO, since the IO may still pending when return, we
25985edc 2947 * set up an end_io call back function, which will do the conversion
8d5d02e6 2948 * when async direct IO completed.
4c0425ff
MC
2949 *
2950 * If the O_DIRECT write will extend the file then add this inode to the
2951 * orphan list. So recovery will truncate it back to the original size
2952 * if the machine crashes during the write.
2953 *
2954 */
2955static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb,
16b1f05d 2956 struct iov_iter *iter, loff_t offset)
4c0425ff
MC
2957{
2958 struct file *file = iocb->ki_filp;
2959 struct inode *inode = file->f_mapping->host;
2960 ssize_t ret;
a6cbcd4a 2961 size_t count = iov_iter_count(iter);
69c499d1
TT
2962 int overwrite = 0;
2963 get_block_t *get_block_func = NULL;
2964 int dio_flags = 0;
4c0425ff 2965 loff_t final_size = offset + count;
97a851ed 2966 ext4_io_end_t *io_end = NULL;
729f52c6 2967
69c499d1
TT
2968 /* Use the old path for reads and writes beyond i_size. */
2969 if (rw != WRITE || final_size > inode->i_size)
16b1f05d 2970 return ext4_ind_direct_IO(rw, iocb, iter, offset);
4bd809db 2971
69c499d1 2972 BUG_ON(iocb->private == NULL);
4bd809db 2973
e8340395
JK
2974 /*
2975 * Make all waiters for direct IO properly wait also for extent
2976 * conversion. This also disallows race between truncate() and
2977 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
2978 */
2979 if (rw == WRITE)
2980 atomic_inc(&inode->i_dio_count);
2981
69c499d1
TT
2982 /* If we do a overwrite dio, i_mutex locking can be released */
2983 overwrite = *((int *)iocb->private);
4bd809db 2984
69c499d1 2985 if (overwrite) {
69c499d1
TT
2986 down_read(&EXT4_I(inode)->i_data_sem);
2987 mutex_unlock(&inode->i_mutex);
2988 }
8d5d02e6 2989
69c499d1
TT
2990 /*
2991 * We could direct write to holes and fallocate.
2992 *
2993 * Allocated blocks to fill the hole are marked as
556615dc 2994 * unwritten to prevent parallel buffered read to expose
69c499d1
TT
2995 * the stale data before DIO complete the data IO.
2996 *
2997 * As to previously fallocated extents, ext4 get_block will
2998 * just simply mark the buffer mapped but still keep the
556615dc 2999 * extents unwritten.
69c499d1
TT
3000 *
3001 * For non AIO case, we will convert those unwritten extents
3002 * to written after return back from blockdev_direct_IO.
3003 *
3004 * For async DIO, the conversion needs to be deferred when the
3005 * IO is completed. The ext4 end_io callback function will be
3006 * called to take care of the conversion work. Here for async
3007 * case, we allocate an io_end structure to hook to the iocb.
3008 */
3009 iocb->private = NULL;
3010 ext4_inode_aio_set(inode, NULL);
3011 if (!is_sync_kiocb(iocb)) {
97a851ed 3012 io_end = ext4_init_io_end(inode, GFP_NOFS);
69c499d1
TT
3013 if (!io_end) {
3014 ret = -ENOMEM;
3015 goto retake_lock;
8b0f165f 3016 }
97a851ed
JK
3017 /*
3018 * Grab reference for DIO. Will be dropped in ext4_end_io_dio()
3019 */
3020 iocb->private = ext4_get_io_end(io_end);
8d5d02e6 3021 /*
69c499d1
TT
3022 * we save the io structure for current async direct
3023 * IO, so that later ext4_map_blocks() could flag the
3024 * io structure whether there is a unwritten extents
3025 * needs to be converted when IO is completed.
8d5d02e6 3026 */
69c499d1
TT
3027 ext4_inode_aio_set(inode, io_end);
3028 }
4bd809db 3029
69c499d1
TT
3030 if (overwrite) {
3031 get_block_func = ext4_get_block_write_nolock;
3032 } else {
3033 get_block_func = ext4_get_block_write;
3034 dio_flags = DIO_LOCKING;
3035 }
3036 ret = __blockdev_direct_IO(rw, iocb, inode,
31b14039
AV
3037 inode->i_sb->s_bdev, iter,
3038 offset,
69c499d1
TT
3039 get_block_func,
3040 ext4_end_io_dio,
3041 NULL,
3042 dio_flags);
3043
69c499d1 3044 /*
97a851ed
JK
3045 * Put our reference to io_end. This can free the io_end structure e.g.
3046 * in sync IO case or in case of error. It can even perform extent
3047 * conversion if all bios we submitted finished before we got here.
3048 * Note that in that case iocb->private can be already set to NULL
3049 * here.
69c499d1 3050 */
97a851ed
JK
3051 if (io_end) {
3052 ext4_inode_aio_set(inode, NULL);
3053 ext4_put_io_end(io_end);
3054 /*
3055 * When no IO was submitted ext4_end_io_dio() was not
3056 * called so we have to put iocb's reference.
3057 */
3058 if (ret <= 0 && ret != -EIOCBQUEUED && iocb->private) {
3059 WARN_ON(iocb->private != io_end);
3060 WARN_ON(io_end->flag & EXT4_IO_END_UNWRITTEN);
97a851ed
JK
3061 ext4_put_io_end(io_end);
3062 iocb->private = NULL;
3063 }
3064 }
3065 if (ret > 0 && !overwrite && ext4_test_inode_state(inode,
69c499d1
TT
3066 EXT4_STATE_DIO_UNWRITTEN)) {
3067 int err;
3068 /*
3069 * for non AIO case, since the IO is already
3070 * completed, we could do the conversion right here
3071 */
6b523df4 3072 err = ext4_convert_unwritten_extents(NULL, inode,
69c499d1
TT
3073 offset, ret);
3074 if (err < 0)
3075 ret = err;
3076 ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN);
3077 }
4bd809db 3078
69c499d1 3079retake_lock:
e8340395
JK
3080 if (rw == WRITE)
3081 inode_dio_done(inode);
69c499d1
TT
3082 /* take i_mutex locking again if we do a ovewrite dio */
3083 if (overwrite) {
69c499d1
TT
3084 up_read(&EXT4_I(inode)->i_data_sem);
3085 mutex_lock(&inode->i_mutex);
4c0425ff 3086 }
8d5d02e6 3087
69c499d1 3088 return ret;
4c0425ff
MC
3089}
3090
3091static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb,
d8d3d94b 3092 struct iov_iter *iter, loff_t offset)
4c0425ff
MC
3093{
3094 struct file *file = iocb->ki_filp;
3095 struct inode *inode = file->f_mapping->host;
a6cbcd4a 3096 size_t count = iov_iter_count(iter);
0562e0ba 3097 ssize_t ret;
4c0425ff 3098
84ebd795
TT
3099 /*
3100 * If we are doing data journalling we don't support O_DIRECT
3101 */
3102 if (ext4_should_journal_data(inode))
3103 return 0;
3104
46c7f254
TM
3105 /* Let buffer I/O handle the inline data case. */
3106 if (ext4_has_inline_data(inode))
3107 return 0;
3108
a6cbcd4a 3109 trace_ext4_direct_IO_enter(inode, offset, count, rw);
12e9b892 3110 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
16b1f05d 3111 ret = ext4_ext_direct_IO(rw, iocb, iter, offset);
0562e0ba 3112 else
16b1f05d 3113 ret = ext4_ind_direct_IO(rw, iocb, iter, offset);
a6cbcd4a 3114 trace_ext4_direct_IO_exit(inode, offset, count, rw, ret);
0562e0ba 3115 return ret;
4c0425ff
MC
3116}
3117
ac27a0ec 3118/*
617ba13b 3119 * Pages can be marked dirty completely asynchronously from ext4's journalling
ac27a0ec
DK
3120 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3121 * much here because ->set_page_dirty is called under VFS locks. The page is
3122 * not necessarily locked.
3123 *
3124 * We cannot just dirty the page and leave attached buffers clean, because the
3125 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3126 * or jbddirty because all the journalling code will explode.
3127 *
3128 * So what we do is to mark the page "pending dirty" and next time writepage
3129 * is called, propagate that into the buffers appropriately.
3130 */
617ba13b 3131static int ext4_journalled_set_page_dirty(struct page *page)
ac27a0ec
DK
3132{
3133 SetPageChecked(page);
3134 return __set_page_dirty_nobuffers(page);
3135}
3136
74d553aa 3137static const struct address_space_operations ext4_aops = {
8ab22b9a
HH
3138 .readpage = ext4_readpage,
3139 .readpages = ext4_readpages,
43ce1d23 3140 .writepage = ext4_writepage,
20970ba6 3141 .writepages = ext4_writepages,
8ab22b9a 3142 .write_begin = ext4_write_begin,
74d553aa 3143 .write_end = ext4_write_end,
8ab22b9a
HH
3144 .bmap = ext4_bmap,
3145 .invalidatepage = ext4_invalidatepage,
3146 .releasepage = ext4_releasepage,
3147 .direct_IO = ext4_direct_IO,
3148 .migratepage = buffer_migrate_page,
3149 .is_partially_uptodate = block_is_partially_uptodate,
aa261f54 3150 .error_remove_page = generic_error_remove_page,
ac27a0ec
DK
3151};
3152
617ba13b 3153static const struct address_space_operations ext4_journalled_aops = {
8ab22b9a
HH
3154 .readpage = ext4_readpage,
3155 .readpages = ext4_readpages,
43ce1d23 3156 .writepage = ext4_writepage,
20970ba6 3157 .writepages = ext4_writepages,
8ab22b9a
HH
3158 .write_begin = ext4_write_begin,
3159 .write_end = ext4_journalled_write_end,
3160 .set_page_dirty = ext4_journalled_set_page_dirty,
3161 .bmap = ext4_bmap,
4520fb3c 3162 .invalidatepage = ext4_journalled_invalidatepage,
8ab22b9a 3163 .releasepage = ext4_releasepage,
84ebd795 3164 .direct_IO = ext4_direct_IO,
8ab22b9a 3165 .is_partially_uptodate = block_is_partially_uptodate,
aa261f54 3166 .error_remove_page = generic_error_remove_page,
ac27a0ec
DK
3167};
3168
64769240 3169static const struct address_space_operations ext4_da_aops = {
8ab22b9a
HH
3170 .readpage = ext4_readpage,
3171 .readpages = ext4_readpages,
43ce1d23 3172 .writepage = ext4_writepage,
20970ba6 3173 .writepages = ext4_writepages,
8ab22b9a
HH
3174 .write_begin = ext4_da_write_begin,
3175 .write_end = ext4_da_write_end,
3176 .bmap = ext4_bmap,
3177 .invalidatepage = ext4_da_invalidatepage,
3178 .releasepage = ext4_releasepage,
3179 .direct_IO = ext4_direct_IO,
3180 .migratepage = buffer_migrate_page,
3181 .is_partially_uptodate = block_is_partially_uptodate,
aa261f54 3182 .error_remove_page = generic_error_remove_page,
64769240
AT
3183};
3184
617ba13b 3185void ext4_set_aops(struct inode *inode)
ac27a0ec 3186{
3d2b1582
LC
3187 switch (ext4_inode_journal_mode(inode)) {
3188 case EXT4_INODE_ORDERED_DATA_MODE:
74d553aa 3189 ext4_set_inode_state(inode, EXT4_STATE_ORDERED_MODE);
3d2b1582
LC
3190 break;
3191 case EXT4_INODE_WRITEBACK_DATA_MODE:
74d553aa 3192 ext4_clear_inode_state(inode, EXT4_STATE_ORDERED_MODE);
3d2b1582
LC
3193 break;
3194 case EXT4_INODE_JOURNAL_DATA_MODE:
617ba13b 3195 inode->i_mapping->a_ops = &ext4_journalled_aops;
74d553aa 3196 return;
3d2b1582
LC
3197 default:
3198 BUG();
3199 }
74d553aa
TT
3200 if (test_opt(inode->i_sb, DELALLOC))
3201 inode->i_mapping->a_ops = &ext4_da_aops;
3202 else
3203 inode->i_mapping->a_ops = &ext4_aops;
ac27a0ec
DK
3204}
3205
d863dc36
LC
3206/*
3207 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3208 * starting from file offset 'from'. The range to be zero'd must
3209 * be contained with in one block. If the specified range exceeds
3210 * the end of the block it will be shortened to end of the block
3211 * that cooresponds to 'from'
3212 */
94350ab5 3213static int ext4_block_zero_page_range(handle_t *handle,
d863dc36
LC
3214 struct address_space *mapping, loff_t from, loff_t length)
3215{
3216 ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT;
3217 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3218 unsigned blocksize, max, pos;
3219 ext4_lblk_t iblock;
3220 struct inode *inode = mapping->host;
3221 struct buffer_head *bh;
3222 struct page *page;
3223 int err = 0;
3224
3225 page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT,
3226 mapping_gfp_mask(mapping) & ~__GFP_FS);
3227 if (!page)
3228 return -ENOMEM;
3229
3230 blocksize = inode->i_sb->s_blocksize;
3231 max = blocksize - (offset & (blocksize - 1));
3232
3233 /*
3234 * correct length if it does not fall between
3235 * 'from' and the end of the block
3236 */
3237 if (length > max || length < 0)
3238 length = max;
3239
3240 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
3241
3242 if (!page_has_buffers(page))
3243 create_empty_buffers(page, blocksize, 0);
3244
3245 /* Find the buffer that contains "offset" */
3246 bh = page_buffers(page);
3247 pos = blocksize;
3248 while (offset >= pos) {
3249 bh = bh->b_this_page;
3250 iblock++;
3251 pos += blocksize;
3252 }
d863dc36
LC
3253 if (buffer_freed(bh)) {
3254 BUFFER_TRACE(bh, "freed: skip");
3255 goto unlock;
3256 }
d863dc36
LC
3257 if (!buffer_mapped(bh)) {
3258 BUFFER_TRACE(bh, "unmapped");
3259 ext4_get_block(inode, iblock, bh, 0);
3260 /* unmapped? It's a hole - nothing to do */
3261 if (!buffer_mapped(bh)) {
3262 BUFFER_TRACE(bh, "still unmapped");
3263 goto unlock;
3264 }
3265 }
3266
3267 /* Ok, it's mapped. Make sure it's up-to-date */
3268 if (PageUptodate(page))
3269 set_buffer_uptodate(bh);
3270
3271 if (!buffer_uptodate(bh)) {
3272 err = -EIO;
3273 ll_rw_block(READ, 1, &bh);
3274 wait_on_buffer(bh);
3275 /* Uhhuh. Read error. Complain and punt. */
3276 if (!buffer_uptodate(bh))
3277 goto unlock;
3278 }
d863dc36
LC
3279 if (ext4_should_journal_data(inode)) {
3280 BUFFER_TRACE(bh, "get write access");
3281 err = ext4_journal_get_write_access(handle, bh);
3282 if (err)
3283 goto unlock;
3284 }
d863dc36 3285 zero_user(page, offset, length);
d863dc36
LC
3286 BUFFER_TRACE(bh, "zeroed end of block");
3287
d863dc36
LC
3288 if (ext4_should_journal_data(inode)) {
3289 err = ext4_handle_dirty_metadata(handle, inode, bh);
0713ed0c 3290 } else {
353eefd3 3291 err = 0;
d863dc36 3292 mark_buffer_dirty(bh);
0713ed0c
LC
3293 if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE))
3294 err = ext4_jbd2_file_inode(handle, inode);
3295 }
d863dc36
LC
3296
3297unlock:
3298 unlock_page(page);
3299 page_cache_release(page);
3300 return err;
3301}
3302
94350ab5
MW
3303/*
3304 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3305 * up to the end of the block which corresponds to `from'.
3306 * This required during truncate. We need to physically zero the tail end
3307 * of that block so it doesn't yield old data if the file is later grown.
3308 */
c197855e 3309static int ext4_block_truncate_page(handle_t *handle,
94350ab5
MW
3310 struct address_space *mapping, loff_t from)
3311{
3312 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3313 unsigned length;
3314 unsigned blocksize;
3315 struct inode *inode = mapping->host;
3316
3317 blocksize = inode->i_sb->s_blocksize;
3318 length = blocksize - (offset & (blocksize - 1));
3319
3320 return ext4_block_zero_page_range(handle, mapping, from, length);
3321}
3322
a87dd18c
LC
3323int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
3324 loff_t lstart, loff_t length)
3325{
3326 struct super_block *sb = inode->i_sb;
3327 struct address_space *mapping = inode->i_mapping;
e1be3a92 3328 unsigned partial_start, partial_end;
a87dd18c
LC
3329 ext4_fsblk_t start, end;
3330 loff_t byte_end = (lstart + length - 1);
3331 int err = 0;
3332
e1be3a92
LC
3333 partial_start = lstart & (sb->s_blocksize - 1);
3334 partial_end = byte_end & (sb->s_blocksize - 1);
3335
a87dd18c
LC
3336 start = lstart >> sb->s_blocksize_bits;
3337 end = byte_end >> sb->s_blocksize_bits;
3338
3339 /* Handle partial zero within the single block */
e1be3a92
LC
3340 if (start == end &&
3341 (partial_start || (partial_end != sb->s_blocksize - 1))) {
a87dd18c
LC
3342 err = ext4_block_zero_page_range(handle, mapping,
3343 lstart, length);
3344 return err;
3345 }
3346 /* Handle partial zero out on the start of the range */
e1be3a92 3347 if (partial_start) {
a87dd18c
LC
3348 err = ext4_block_zero_page_range(handle, mapping,
3349 lstart, sb->s_blocksize);
3350 if (err)
3351 return err;
3352 }
3353 /* Handle partial zero out on the end of the range */
e1be3a92 3354 if (partial_end != sb->s_blocksize - 1)
a87dd18c 3355 err = ext4_block_zero_page_range(handle, mapping,
e1be3a92
LC
3356 byte_end - partial_end,
3357 partial_end + 1);
a87dd18c
LC
3358 return err;
3359}
3360
91ef4caf
DG
3361int ext4_can_truncate(struct inode *inode)
3362{
91ef4caf
DG
3363 if (S_ISREG(inode->i_mode))
3364 return 1;
3365 if (S_ISDIR(inode->i_mode))
3366 return 1;
3367 if (S_ISLNK(inode->i_mode))
3368 return !ext4_inode_is_fast_symlink(inode);
3369 return 0;
3370}
3371
a4bb6b64
AH
3372/*
3373 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3374 * associated with the given offset and length
3375 *
3376 * @inode: File inode
3377 * @offset: The offset where the hole will begin
3378 * @len: The length of the hole
3379 *
4907cb7b 3380 * Returns: 0 on success or negative on failure
a4bb6b64
AH
3381 */
3382
aeb2817a 3383int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length)
a4bb6b64 3384{
26a4c0c6
TT
3385 struct super_block *sb = inode->i_sb;
3386 ext4_lblk_t first_block, stop_block;
3387 struct address_space *mapping = inode->i_mapping;
a87dd18c 3388 loff_t first_block_offset, last_block_offset;
26a4c0c6
TT
3389 handle_t *handle;
3390 unsigned int credits;
3391 int ret = 0;
3392
a4bb6b64 3393 if (!S_ISREG(inode->i_mode))
73355192 3394 return -EOPNOTSUPP;
a4bb6b64 3395
b8a86845 3396 trace_ext4_punch_hole(inode, offset, length, 0);
aaddea81 3397
26a4c0c6
TT
3398 /*
3399 * Write out all dirty pages to avoid race conditions
3400 * Then release them.
3401 */
3402 if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
3403 ret = filemap_write_and_wait_range(mapping, offset,
3404 offset + length - 1);
3405 if (ret)
3406 return ret;
3407 }
3408
3409 mutex_lock(&inode->i_mutex);
9ef06cec 3410
26a4c0c6
TT
3411 /* No need to punch hole beyond i_size */
3412 if (offset >= inode->i_size)
3413 goto out_mutex;
3414
3415 /*
3416 * If the hole extends beyond i_size, set the hole
3417 * to end after the page that contains i_size
3418 */
3419 if (offset + length > inode->i_size) {
3420 length = inode->i_size +
3421 PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) -
3422 offset;
3423 }
3424
a361293f
JK
3425 if (offset & (sb->s_blocksize - 1) ||
3426 (offset + length) & (sb->s_blocksize - 1)) {
3427 /*
3428 * Attach jinode to inode for jbd2 if we do any zeroing of
3429 * partial block
3430 */
3431 ret = ext4_inode_attach_jinode(inode);
3432 if (ret < 0)
3433 goto out_mutex;
3434
3435 }
3436
a87dd18c
LC
3437 first_block_offset = round_up(offset, sb->s_blocksize);
3438 last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
26a4c0c6 3439
a87dd18c
LC
3440 /* Now release the pages and zero block aligned part of pages*/
3441 if (last_block_offset > first_block_offset)
3442 truncate_pagecache_range(inode, first_block_offset,
3443 last_block_offset);
26a4c0c6
TT
3444
3445 /* Wait all existing dio workers, newcomers will block on i_mutex */
3446 ext4_inode_block_unlocked_dio(inode);
26a4c0c6
TT
3447 inode_dio_wait(inode);
3448
3449 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3450 credits = ext4_writepage_trans_blocks(inode);
3451 else
3452 credits = ext4_blocks_for_truncate(inode);
3453 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3454 if (IS_ERR(handle)) {
3455 ret = PTR_ERR(handle);
3456 ext4_std_error(sb, ret);
3457 goto out_dio;
3458 }
3459
a87dd18c
LC
3460 ret = ext4_zero_partial_blocks(handle, inode, offset,
3461 length);
3462 if (ret)
3463 goto out_stop;
26a4c0c6
TT
3464
3465 first_block = (offset + sb->s_blocksize - 1) >>
3466 EXT4_BLOCK_SIZE_BITS(sb);
3467 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
3468
3469 /* If there are no blocks to remove, return now */
3470 if (first_block >= stop_block)
3471 goto out_stop;
3472
3473 down_write(&EXT4_I(inode)->i_data_sem);
3474 ext4_discard_preallocations(inode);
3475
3476 ret = ext4_es_remove_extent(inode, first_block,
3477 stop_block - first_block);
3478 if (ret) {
3479 up_write(&EXT4_I(inode)->i_data_sem);
3480 goto out_stop;
3481 }
3482
3483 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3484 ret = ext4_ext_remove_space(inode, first_block,
3485 stop_block - 1);
3486 else
4f579ae7 3487 ret = ext4_ind_remove_space(handle, inode, first_block,
26a4c0c6
TT
3488 stop_block);
3489
819c4920 3490 up_write(&EXT4_I(inode)->i_data_sem);
26a4c0c6
TT
3491 if (IS_SYNC(inode))
3492 ext4_handle_sync(handle);
e251f9bc
MP
3493
3494 /* Now release the pages again to reduce race window */
3495 if (last_block_offset > first_block_offset)
3496 truncate_pagecache_range(inode, first_block_offset,
3497 last_block_offset);
3498
26a4c0c6
TT
3499 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3500 ext4_mark_inode_dirty(handle, inode);
3501out_stop:
3502 ext4_journal_stop(handle);
3503out_dio:
3504 ext4_inode_resume_unlocked_dio(inode);
3505out_mutex:
3506 mutex_unlock(&inode->i_mutex);
3507 return ret;
a4bb6b64
AH
3508}
3509
a361293f
JK
3510int ext4_inode_attach_jinode(struct inode *inode)
3511{
3512 struct ext4_inode_info *ei = EXT4_I(inode);
3513 struct jbd2_inode *jinode;
3514
3515 if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
3516 return 0;
3517
3518 jinode = jbd2_alloc_inode(GFP_KERNEL);
3519 spin_lock(&inode->i_lock);
3520 if (!ei->jinode) {
3521 if (!jinode) {
3522 spin_unlock(&inode->i_lock);
3523 return -ENOMEM;
3524 }
3525 ei->jinode = jinode;
3526 jbd2_journal_init_jbd_inode(ei->jinode, inode);
3527 jinode = NULL;
3528 }
3529 spin_unlock(&inode->i_lock);
3530 if (unlikely(jinode != NULL))
3531 jbd2_free_inode(jinode);
3532 return 0;
3533}
3534
ac27a0ec 3535/*
617ba13b 3536 * ext4_truncate()
ac27a0ec 3537 *
617ba13b
MC
3538 * We block out ext4_get_block() block instantiations across the entire
3539 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
ac27a0ec
DK
3540 * simultaneously on behalf of the same inode.
3541 *
42b2aa86 3542 * As we work through the truncate and commit bits of it to the journal there
ac27a0ec
DK
3543 * is one core, guiding principle: the file's tree must always be consistent on
3544 * disk. We must be able to restart the truncate after a crash.
3545 *
3546 * The file's tree may be transiently inconsistent in memory (although it
3547 * probably isn't), but whenever we close off and commit a journal transaction,
3548 * the contents of (the filesystem + the journal) must be consistent and
3549 * restartable. It's pretty simple, really: bottom up, right to left (although
3550 * left-to-right works OK too).
3551 *
3552 * Note that at recovery time, journal replay occurs *before* the restart of
3553 * truncate against the orphan inode list.
3554 *
3555 * The committed inode has the new, desired i_size (which is the same as
617ba13b 3556 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
ac27a0ec 3557 * that this inode's truncate did not complete and it will again call
617ba13b
MC
3558 * ext4_truncate() to have another go. So there will be instantiated blocks
3559 * to the right of the truncation point in a crashed ext4 filesystem. But
ac27a0ec 3560 * that's fine - as long as they are linked from the inode, the post-crash
617ba13b 3561 * ext4_truncate() run will find them and release them.
ac27a0ec 3562 */
617ba13b 3563void ext4_truncate(struct inode *inode)
ac27a0ec 3564{
819c4920
TT
3565 struct ext4_inode_info *ei = EXT4_I(inode);
3566 unsigned int credits;
3567 handle_t *handle;
3568 struct address_space *mapping = inode->i_mapping;
819c4920 3569
19b5ef61
TT
3570 /*
3571 * There is a possibility that we're either freeing the inode
e04027e8 3572 * or it's a completely new inode. In those cases we might not
19b5ef61
TT
3573 * have i_mutex locked because it's not necessary.
3574 */
3575 if (!(inode->i_state & (I_NEW|I_FREEING)))
3576 WARN_ON(!mutex_is_locked(&inode->i_mutex));
0562e0ba
JZ
3577 trace_ext4_truncate_enter(inode);
3578
91ef4caf 3579 if (!ext4_can_truncate(inode))
ac27a0ec
DK
3580 return;
3581
12e9b892 3582 ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
c8d46e41 3583
5534fb5b 3584 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
19f5fb7a 3585 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
7d8f9f7d 3586
aef1c851
TM
3587 if (ext4_has_inline_data(inode)) {
3588 int has_inline = 1;
3589
3590 ext4_inline_data_truncate(inode, &has_inline);
3591 if (has_inline)
3592 return;
3593 }
3594
a361293f
JK
3595 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
3596 if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
3597 if (ext4_inode_attach_jinode(inode) < 0)
3598 return;
3599 }
3600
819c4920
TT
3601 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3602 credits = ext4_writepage_trans_blocks(inode);
3603 else
3604 credits = ext4_blocks_for_truncate(inode);
3605
3606 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3607 if (IS_ERR(handle)) {
3608 ext4_std_error(inode->i_sb, PTR_ERR(handle));
3609 return;
3610 }
3611
eb3544c6
LC
3612 if (inode->i_size & (inode->i_sb->s_blocksize - 1))
3613 ext4_block_truncate_page(handle, mapping, inode->i_size);
819c4920
TT
3614
3615 /*
3616 * We add the inode to the orphan list, so that if this
3617 * truncate spans multiple transactions, and we crash, we will
3618 * resume the truncate when the filesystem recovers. It also
3619 * marks the inode dirty, to catch the new size.
3620 *
3621 * Implication: the file must always be in a sane, consistent
3622 * truncatable state while each transaction commits.
3623 */
3624 if (ext4_orphan_add(handle, inode))
3625 goto out_stop;
3626
3627 down_write(&EXT4_I(inode)->i_data_sem);
3628
3629 ext4_discard_preallocations(inode);
3630
ff9893dc 3631 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
819c4920 3632 ext4_ext_truncate(handle, inode);
ff9893dc 3633 else
819c4920
TT
3634 ext4_ind_truncate(handle, inode);
3635
3636 up_write(&ei->i_data_sem);
3637
3638 if (IS_SYNC(inode))
3639 ext4_handle_sync(handle);
3640
3641out_stop:
3642 /*
3643 * If this was a simple ftruncate() and the file will remain alive,
3644 * then we need to clear up the orphan record which we created above.
3645 * However, if this was a real unlink then we were called by
3646 * ext4_delete_inode(), and we allow that function to clean up the
3647 * orphan info for us.
3648 */
3649 if (inode->i_nlink)
3650 ext4_orphan_del(handle, inode);
3651
3652 inode->i_mtime = inode->i_ctime = ext4_current_time(inode);
3653 ext4_mark_inode_dirty(handle, inode);
3654 ext4_journal_stop(handle);
ac27a0ec 3655
0562e0ba 3656 trace_ext4_truncate_exit(inode);
ac27a0ec
DK
3657}
3658
ac27a0ec 3659/*
617ba13b 3660 * ext4_get_inode_loc returns with an extra refcount against the inode's
ac27a0ec
DK
3661 * underlying buffer_head on success. If 'in_mem' is true, we have all
3662 * data in memory that is needed to recreate the on-disk version of this
3663 * inode.
3664 */
617ba13b
MC
3665static int __ext4_get_inode_loc(struct inode *inode,
3666 struct ext4_iloc *iloc, int in_mem)
ac27a0ec 3667{
240799cd
TT
3668 struct ext4_group_desc *gdp;
3669 struct buffer_head *bh;
3670 struct super_block *sb = inode->i_sb;
3671 ext4_fsblk_t block;
3672 int inodes_per_block, inode_offset;
3673
3a06d778 3674 iloc->bh = NULL;
240799cd
TT
3675 if (!ext4_valid_inum(sb, inode->i_ino))
3676 return -EIO;
ac27a0ec 3677
240799cd
TT
3678 iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
3679 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
3680 if (!gdp)
ac27a0ec
DK
3681 return -EIO;
3682
240799cd
TT
3683 /*
3684 * Figure out the offset within the block group inode table
3685 */
00d09882 3686 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
240799cd
TT
3687 inode_offset = ((inode->i_ino - 1) %
3688 EXT4_INODES_PER_GROUP(sb));
3689 block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
3690 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
3691
3692 bh = sb_getblk(sb, block);
aebf0243 3693 if (unlikely(!bh))
860d21e2 3694 return -ENOMEM;
ac27a0ec
DK
3695 if (!buffer_uptodate(bh)) {
3696 lock_buffer(bh);
9c83a923
HK
3697
3698 /*
3699 * If the buffer has the write error flag, we have failed
3700 * to write out another inode in the same block. In this
3701 * case, we don't have to read the block because we may
3702 * read the old inode data successfully.
3703 */
3704 if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
3705 set_buffer_uptodate(bh);
3706
ac27a0ec
DK
3707 if (buffer_uptodate(bh)) {
3708 /* someone brought it uptodate while we waited */
3709 unlock_buffer(bh);
3710 goto has_buffer;
3711 }
3712
3713 /*
3714 * If we have all information of the inode in memory and this
3715 * is the only valid inode in the block, we need not read the
3716 * block.
3717 */
3718 if (in_mem) {
3719 struct buffer_head *bitmap_bh;
240799cd 3720 int i, start;
ac27a0ec 3721
240799cd 3722 start = inode_offset & ~(inodes_per_block - 1);
ac27a0ec 3723
240799cd
TT
3724 /* Is the inode bitmap in cache? */
3725 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
aebf0243 3726 if (unlikely(!bitmap_bh))
ac27a0ec
DK
3727 goto make_io;
3728
3729 /*
3730 * If the inode bitmap isn't in cache then the
3731 * optimisation may end up performing two reads instead
3732 * of one, so skip it.
3733 */
3734 if (!buffer_uptodate(bitmap_bh)) {
3735 brelse(bitmap_bh);
3736 goto make_io;
3737 }
240799cd 3738 for (i = start; i < start + inodes_per_block; i++) {
ac27a0ec
DK
3739 if (i == inode_offset)
3740 continue;
617ba13b 3741 if (ext4_test_bit(i, bitmap_bh->b_data))
ac27a0ec
DK
3742 break;
3743 }
3744 brelse(bitmap_bh);
240799cd 3745 if (i == start + inodes_per_block) {
ac27a0ec
DK
3746 /* all other inodes are free, so skip I/O */
3747 memset(bh->b_data, 0, bh->b_size);
3748 set_buffer_uptodate(bh);
3749 unlock_buffer(bh);
3750 goto has_buffer;
3751 }
3752 }
3753
3754make_io:
240799cd
TT
3755 /*
3756 * If we need to do any I/O, try to pre-readahead extra
3757 * blocks from the inode table.
3758 */
3759 if (EXT4_SB(sb)->s_inode_readahead_blks) {
3760 ext4_fsblk_t b, end, table;
3761 unsigned num;
0d606e2c 3762 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
240799cd
TT
3763
3764 table = ext4_inode_table(sb, gdp);
b713a5ec 3765 /* s_inode_readahead_blks is always a power of 2 */
0d606e2c 3766 b = block & ~((ext4_fsblk_t) ra_blks - 1);
240799cd
TT
3767 if (table > b)
3768 b = table;
0d606e2c 3769 end = b + ra_blks;
240799cd 3770 num = EXT4_INODES_PER_GROUP(sb);
feb0ab32 3771 if (ext4_has_group_desc_csum(sb))
560671a0 3772 num -= ext4_itable_unused_count(sb, gdp);
240799cd
TT
3773 table += num / inodes_per_block;
3774 if (end > table)
3775 end = table;
3776 while (b <= end)
3777 sb_breadahead(sb, b++);
3778 }
3779
ac27a0ec
DK
3780 /*
3781 * There are other valid inodes in the buffer, this inode
3782 * has in-inode xattrs, or we don't have this inode in memory.
3783 * Read the block from disk.
3784 */
0562e0ba 3785 trace_ext4_load_inode(inode);
ac27a0ec
DK
3786 get_bh(bh);
3787 bh->b_end_io = end_buffer_read_sync;
65299a3b 3788 submit_bh(READ | REQ_META | REQ_PRIO, bh);
ac27a0ec
DK
3789 wait_on_buffer(bh);
3790 if (!buffer_uptodate(bh)) {
c398eda0
TT
3791 EXT4_ERROR_INODE_BLOCK(inode, block,
3792 "unable to read itable block");
ac27a0ec
DK
3793 brelse(bh);
3794 return -EIO;
3795 }
3796 }
3797has_buffer:
3798 iloc->bh = bh;
3799 return 0;
3800}
3801
617ba13b 3802int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
ac27a0ec
DK
3803{
3804 /* We have all inode data except xattrs in memory here. */
617ba13b 3805 return __ext4_get_inode_loc(inode, iloc,
19f5fb7a 3806 !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
ac27a0ec
DK
3807}
3808
617ba13b 3809void ext4_set_inode_flags(struct inode *inode)
ac27a0ec 3810{
617ba13b 3811 unsigned int flags = EXT4_I(inode)->i_flags;
00a1a053 3812 unsigned int new_fl = 0;
ac27a0ec 3813
617ba13b 3814 if (flags & EXT4_SYNC_FL)
00a1a053 3815 new_fl |= S_SYNC;
617ba13b 3816 if (flags & EXT4_APPEND_FL)
00a1a053 3817 new_fl |= S_APPEND;
617ba13b 3818 if (flags & EXT4_IMMUTABLE_FL)
00a1a053 3819 new_fl |= S_IMMUTABLE;
617ba13b 3820 if (flags & EXT4_NOATIME_FL)
00a1a053 3821 new_fl |= S_NOATIME;
617ba13b 3822 if (flags & EXT4_DIRSYNC_FL)
00a1a053 3823 new_fl |= S_DIRSYNC;
5f16f322
TT
3824 inode_set_flags(inode, new_fl,
3825 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
ac27a0ec
DK
3826}
3827
ff9ddf7e
JK
3828/* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3829void ext4_get_inode_flags(struct ext4_inode_info *ei)
3830{
84a8dce2
DM
3831 unsigned int vfs_fl;
3832 unsigned long old_fl, new_fl;
3833
3834 do {
3835 vfs_fl = ei->vfs_inode.i_flags;
3836 old_fl = ei->i_flags;
3837 new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL|
3838 EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL|
3839 EXT4_DIRSYNC_FL);
3840 if (vfs_fl & S_SYNC)
3841 new_fl |= EXT4_SYNC_FL;
3842 if (vfs_fl & S_APPEND)
3843 new_fl |= EXT4_APPEND_FL;
3844 if (vfs_fl & S_IMMUTABLE)
3845 new_fl |= EXT4_IMMUTABLE_FL;
3846 if (vfs_fl & S_NOATIME)
3847 new_fl |= EXT4_NOATIME_FL;
3848 if (vfs_fl & S_DIRSYNC)
3849 new_fl |= EXT4_DIRSYNC_FL;
3850 } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl);
ff9ddf7e 3851}
de9a55b8 3852
0fc1b451 3853static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
de9a55b8 3854 struct ext4_inode_info *ei)
0fc1b451
AK
3855{
3856 blkcnt_t i_blocks ;
8180a562
AK
3857 struct inode *inode = &(ei->vfs_inode);
3858 struct super_block *sb = inode->i_sb;
0fc1b451
AK
3859
3860 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3861 EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
3862 /* we are using combined 48 bit field */
3863 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
3864 le32_to_cpu(raw_inode->i_blocks_lo);
07a03824 3865 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
8180a562
AK
3866 /* i_blocks represent file system block size */
3867 return i_blocks << (inode->i_blkbits - 9);
3868 } else {
3869 return i_blocks;
3870 }
0fc1b451
AK
3871 } else {
3872 return le32_to_cpu(raw_inode->i_blocks_lo);
3873 }
3874}
ff9ddf7e 3875
152a7b0a
TM
3876static inline void ext4_iget_extra_inode(struct inode *inode,
3877 struct ext4_inode *raw_inode,
3878 struct ext4_inode_info *ei)
3879{
3880 __le32 *magic = (void *)raw_inode +
3881 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
67cf5b09 3882 if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
152a7b0a 3883 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
67cf5b09 3884 ext4_find_inline_data_nolock(inode);
f19d5870
TM
3885 } else
3886 EXT4_I(inode)->i_inline_off = 0;
152a7b0a
TM
3887}
3888
1d1fe1ee 3889struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
ac27a0ec 3890{
617ba13b
MC
3891 struct ext4_iloc iloc;
3892 struct ext4_inode *raw_inode;
1d1fe1ee 3893 struct ext4_inode_info *ei;
1d1fe1ee 3894 struct inode *inode;
b436b9be 3895 journal_t *journal = EXT4_SB(sb)->s_journal;
1d1fe1ee 3896 long ret;
ac27a0ec 3897 int block;
08cefc7a
EB
3898 uid_t i_uid;
3899 gid_t i_gid;
ac27a0ec 3900
1d1fe1ee
DH
3901 inode = iget_locked(sb, ino);
3902 if (!inode)
3903 return ERR_PTR(-ENOMEM);
3904 if (!(inode->i_state & I_NEW))
3905 return inode;
3906
3907 ei = EXT4_I(inode);
7dc57615 3908 iloc.bh = NULL;
ac27a0ec 3909
1d1fe1ee
DH
3910 ret = __ext4_get_inode_loc(inode, &iloc, 0);
3911 if (ret < 0)
ac27a0ec 3912 goto bad_inode;
617ba13b 3913 raw_inode = ext4_raw_inode(&iloc);
814525f4
DW
3914
3915 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
3916 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
3917 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
3918 EXT4_INODE_SIZE(inode->i_sb)) {
3919 EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)",
3920 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize,
3921 EXT4_INODE_SIZE(inode->i_sb));
3922 ret = -EIO;
3923 goto bad_inode;
3924 }
3925 } else
3926 ei->i_extra_isize = 0;
3927
3928 /* Precompute checksum seed for inode metadata */
3929 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3930 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3931 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3932 __u32 csum;
3933 __le32 inum = cpu_to_le32(inode->i_ino);
3934 __le32 gen = raw_inode->i_generation;
3935 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
3936 sizeof(inum));
3937 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
3938 sizeof(gen));
3939 }
3940
3941 if (!ext4_inode_csum_verify(inode, raw_inode, ei)) {
3942 EXT4_ERROR_INODE(inode, "checksum invalid");
3943 ret = -EIO;
3944 goto bad_inode;
3945 }
3946
ac27a0ec 3947 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
08cefc7a
EB
3948 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
3949 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
af5bc92d 3950 if (!(test_opt(inode->i_sb, NO_UID32))) {
08cefc7a
EB
3951 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
3952 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
ac27a0ec 3953 }
08cefc7a
EB
3954 i_uid_write(inode, i_uid);
3955 i_gid_write(inode, i_gid);
bfe86848 3956 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
ac27a0ec 3957
353eb83c 3958 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
67cf5b09 3959 ei->i_inline_off = 0;
ac27a0ec
DK
3960 ei->i_dir_start_lookup = 0;
3961 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
3962 /* We now have enough fields to check if the inode was active or not.
3963 * This is needed because nfsd might try to access dead inodes
3964 * the test is that same one that e2fsck uses
3965 * NeilBrown 1999oct15
3966 */
3967 if (inode->i_nlink == 0) {
393d1d1d
DTB
3968 if ((inode->i_mode == 0 ||
3969 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
3970 ino != EXT4_BOOT_LOADER_INO) {
ac27a0ec 3971 /* this inode is deleted */
1d1fe1ee 3972 ret = -ESTALE;
ac27a0ec
DK
3973 goto bad_inode;
3974 }
3975 /* The only unlinked inodes we let through here have
3976 * valid i_mode and are being read by the orphan
3977 * recovery code: that's fine, we're about to complete
393d1d1d
DTB
3978 * the process of deleting those.
3979 * OR it is the EXT4_BOOT_LOADER_INO which is
3980 * not initialized on a new filesystem. */
ac27a0ec 3981 }
ac27a0ec 3982 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
0fc1b451 3983 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
7973c0c1 3984 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
a9e81742 3985 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
a1ddeb7e
BP
3986 ei->i_file_acl |=
3987 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
a48380f7 3988 inode->i_size = ext4_isize(raw_inode);
ac27a0ec 3989 ei->i_disksize = inode->i_size;
a9e7f447
DM
3990#ifdef CONFIG_QUOTA
3991 ei->i_reserved_quota = 0;
3992#endif
ac27a0ec
DK
3993 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
3994 ei->i_block_group = iloc.block_group;
a4912123 3995 ei->i_last_alloc_group = ~0;
ac27a0ec
DK
3996 /*
3997 * NOTE! The in-memory inode i_data array is in little-endian order
3998 * even on big-endian machines: we do NOT byteswap the block numbers!
3999 */
617ba13b 4000 for (block = 0; block < EXT4_N_BLOCKS; block++)
ac27a0ec
DK
4001 ei->i_data[block] = raw_inode->i_block[block];
4002 INIT_LIST_HEAD(&ei->i_orphan);
4003
b436b9be
JK
4004 /*
4005 * Set transaction id's of transactions that have to be committed
4006 * to finish f[data]sync. We set them to currently running transaction
4007 * as we cannot be sure that the inode or some of its metadata isn't
4008 * part of the transaction - the inode could have been reclaimed and
4009 * now it is reread from disk.
4010 */
4011 if (journal) {
4012 transaction_t *transaction;
4013 tid_t tid;
4014
a931da6a 4015 read_lock(&journal->j_state_lock);
b436b9be
JK
4016 if (journal->j_running_transaction)
4017 transaction = journal->j_running_transaction;
4018 else
4019 transaction = journal->j_committing_transaction;
4020 if (transaction)
4021 tid = transaction->t_tid;
4022 else
4023 tid = journal->j_commit_sequence;
a931da6a 4024 read_unlock(&journal->j_state_lock);
b436b9be
JK
4025 ei->i_sync_tid = tid;
4026 ei->i_datasync_tid = tid;
4027 }
4028
0040d987 4029 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
ac27a0ec
DK
4030 if (ei->i_extra_isize == 0) {
4031 /* The extra space is currently unused. Use it. */
617ba13b
MC
4032 ei->i_extra_isize = sizeof(struct ext4_inode) -
4033 EXT4_GOOD_OLD_INODE_SIZE;
ac27a0ec 4034 } else {
152a7b0a 4035 ext4_iget_extra_inode(inode, raw_inode, ei);
ac27a0ec 4036 }
814525f4 4037 }
ac27a0ec 4038
ef7f3835
KS
4039 EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4040 EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4041 EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4042 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4043
ed3654eb 4044 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
c4f65706
TT
4045 inode->i_version = le32_to_cpu(raw_inode->i_disk_version);
4046 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4047 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4048 inode->i_version |=
4049 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4050 }
25ec56b5
JNC
4051 }
4052
c4b5a614 4053 ret = 0;
485c26ec 4054 if (ei->i_file_acl &&
1032988c 4055 !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
24676da4
TT
4056 EXT4_ERROR_INODE(inode, "bad extended attribute block %llu",
4057 ei->i_file_acl);
485c26ec
TT
4058 ret = -EIO;
4059 goto bad_inode;
f19d5870
TM
4060 } else if (!ext4_has_inline_data(inode)) {
4061 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
4062 if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4063 (S_ISLNK(inode->i_mode) &&
4064 !ext4_inode_is_fast_symlink(inode))))
4065 /* Validate extent which is part of inode */
4066 ret = ext4_ext_check_inode(inode);
4067 } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4068 (S_ISLNK(inode->i_mode) &&
4069 !ext4_inode_is_fast_symlink(inode))) {
4070 /* Validate block references which are part of inode */
4071 ret = ext4_ind_check_inode(inode);
4072 }
fe2c8191 4073 }
567f3e9a 4074 if (ret)
de9a55b8 4075 goto bad_inode;
7a262f7c 4076
ac27a0ec 4077 if (S_ISREG(inode->i_mode)) {
617ba13b
MC
4078 inode->i_op = &ext4_file_inode_operations;
4079 inode->i_fop = &ext4_file_operations;
4080 ext4_set_aops(inode);
ac27a0ec 4081 } else if (S_ISDIR(inode->i_mode)) {
617ba13b
MC
4082 inode->i_op = &ext4_dir_inode_operations;
4083 inode->i_fop = &ext4_dir_operations;
ac27a0ec 4084 } else if (S_ISLNK(inode->i_mode)) {
e83c1397 4085 if (ext4_inode_is_fast_symlink(inode)) {
617ba13b 4086 inode->i_op = &ext4_fast_symlink_inode_operations;
e83c1397
DG
4087 nd_terminate_link(ei->i_data, inode->i_size,
4088 sizeof(ei->i_data) - 1);
4089 } else {
617ba13b
MC
4090 inode->i_op = &ext4_symlink_inode_operations;
4091 ext4_set_aops(inode);
ac27a0ec 4092 }
563bdd61
TT
4093 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4094 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
617ba13b 4095 inode->i_op = &ext4_special_inode_operations;
ac27a0ec
DK
4096 if (raw_inode->i_block[0])
4097 init_special_inode(inode, inode->i_mode,
4098 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4099 else
4100 init_special_inode(inode, inode->i_mode,
4101 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
393d1d1d
DTB
4102 } else if (ino == EXT4_BOOT_LOADER_INO) {
4103 make_bad_inode(inode);
563bdd61 4104 } else {
563bdd61 4105 ret = -EIO;
24676da4 4106 EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode);
563bdd61 4107 goto bad_inode;
ac27a0ec 4108 }
af5bc92d 4109 brelse(iloc.bh);
617ba13b 4110 ext4_set_inode_flags(inode);
1d1fe1ee
DH
4111 unlock_new_inode(inode);
4112 return inode;
ac27a0ec
DK
4113
4114bad_inode:
567f3e9a 4115 brelse(iloc.bh);
1d1fe1ee
DH
4116 iget_failed(inode);
4117 return ERR_PTR(ret);
ac27a0ec
DK
4118}
4119
0fc1b451
AK
4120static int ext4_inode_blocks_set(handle_t *handle,
4121 struct ext4_inode *raw_inode,
4122 struct ext4_inode_info *ei)
4123{
4124 struct inode *inode = &(ei->vfs_inode);
4125 u64 i_blocks = inode->i_blocks;
4126 struct super_block *sb = inode->i_sb;
0fc1b451
AK
4127
4128 if (i_blocks <= ~0U) {
4129 /*
4907cb7b 4130 * i_blocks can be represented in a 32 bit variable
0fc1b451
AK
4131 * as multiple of 512 bytes
4132 */
8180a562 4133 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
0fc1b451 4134 raw_inode->i_blocks_high = 0;
84a8dce2 4135 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
f287a1a5
TT
4136 return 0;
4137 }
4138 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE))
4139 return -EFBIG;
4140
4141 if (i_blocks <= 0xffffffffffffULL) {
0fc1b451
AK
4142 /*
4143 * i_blocks can be represented in a 48 bit variable
4144 * as multiple of 512 bytes
4145 */
8180a562 4146 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
0fc1b451 4147 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
84a8dce2 4148 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
0fc1b451 4149 } else {
84a8dce2 4150 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
8180a562
AK
4151 /* i_block is stored in file system block size */
4152 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4153 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4154 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
0fc1b451 4155 }
f287a1a5 4156 return 0;
0fc1b451
AK
4157}
4158
ac27a0ec
DK
4159/*
4160 * Post the struct inode info into an on-disk inode location in the
4161 * buffer-cache. This gobbles the caller's reference to the
4162 * buffer_head in the inode location struct.
4163 *
4164 * The caller must have write access to iloc->bh.
4165 */
617ba13b 4166static int ext4_do_update_inode(handle_t *handle,
ac27a0ec 4167 struct inode *inode,
830156c7 4168 struct ext4_iloc *iloc)
ac27a0ec 4169{
617ba13b
MC
4170 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4171 struct ext4_inode_info *ei = EXT4_I(inode);
ac27a0ec 4172 struct buffer_head *bh = iloc->bh;
202ee5df 4173 struct super_block *sb = inode->i_sb;
ac27a0ec 4174 int err = 0, rc, block;
202ee5df 4175 int need_datasync = 0, set_large_file = 0;
08cefc7a
EB
4176 uid_t i_uid;
4177 gid_t i_gid;
ac27a0ec 4178
202ee5df
TT
4179 spin_lock(&ei->i_raw_lock);
4180
4181 /* For fields not tracked in the in-memory inode,
ac27a0ec 4182 * initialise them to zero for new inodes. */
19f5fb7a 4183 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
617ba13b 4184 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
ac27a0ec 4185
ff9ddf7e 4186 ext4_get_inode_flags(ei);
ac27a0ec 4187 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
08cefc7a
EB
4188 i_uid = i_uid_read(inode);
4189 i_gid = i_gid_read(inode);
af5bc92d 4190 if (!(test_opt(inode->i_sb, NO_UID32))) {
08cefc7a
EB
4191 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4192 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
ac27a0ec
DK
4193/*
4194 * Fix up interoperability with old kernels. Otherwise, old inodes get
4195 * re-used with the upper 16 bits of the uid/gid intact
4196 */
af5bc92d 4197 if (!ei->i_dtime) {
ac27a0ec 4198 raw_inode->i_uid_high =
08cefc7a 4199 cpu_to_le16(high_16_bits(i_uid));
ac27a0ec 4200 raw_inode->i_gid_high =
08cefc7a 4201 cpu_to_le16(high_16_bits(i_gid));
ac27a0ec
DK
4202 } else {
4203 raw_inode->i_uid_high = 0;
4204 raw_inode->i_gid_high = 0;
4205 }
4206 } else {
08cefc7a
EB
4207 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4208 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
ac27a0ec
DK
4209 raw_inode->i_uid_high = 0;
4210 raw_inode->i_gid_high = 0;
4211 }
4212 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
ef7f3835
KS
4213
4214 EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4215 EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4216 EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4217 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4218
202ee5df
TT
4219 if (ext4_inode_blocks_set(handle, raw_inode, ei)) {
4220 spin_unlock(&ei->i_raw_lock);
0fc1b451 4221 goto out_brelse;
202ee5df 4222 }
ac27a0ec 4223 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
353eb83c 4224 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
ed3654eb 4225 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
a1ddeb7e
BP
4226 raw_inode->i_file_acl_high =
4227 cpu_to_le16(ei->i_file_acl >> 32);
7973c0c1 4228 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
b71fc079
JK
4229 if (ei->i_disksize != ext4_isize(raw_inode)) {
4230 ext4_isize_set(raw_inode, ei->i_disksize);
4231 need_datasync = 1;
4232 }
a48380f7 4233 if (ei->i_disksize > 0x7fffffffULL) {
a48380f7
AK
4234 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
4235 EXT4_FEATURE_RO_COMPAT_LARGE_FILE) ||
4236 EXT4_SB(sb)->s_es->s_rev_level ==
202ee5df
TT
4237 cpu_to_le32(EXT4_GOOD_OLD_REV))
4238 set_large_file = 1;
ac27a0ec
DK
4239 }
4240 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4241 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4242 if (old_valid_dev(inode->i_rdev)) {
4243 raw_inode->i_block[0] =
4244 cpu_to_le32(old_encode_dev(inode->i_rdev));
4245 raw_inode->i_block[1] = 0;
4246 } else {
4247 raw_inode->i_block[0] = 0;
4248 raw_inode->i_block[1] =
4249 cpu_to_le32(new_encode_dev(inode->i_rdev));
4250 raw_inode->i_block[2] = 0;
4251 }
f19d5870 4252 } else if (!ext4_has_inline_data(inode)) {
de9a55b8
TT
4253 for (block = 0; block < EXT4_N_BLOCKS; block++)
4254 raw_inode->i_block[block] = ei->i_data[block];
f19d5870 4255 }
ac27a0ec 4256
ed3654eb 4257 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
c4f65706
TT
4258 raw_inode->i_disk_version = cpu_to_le32(inode->i_version);
4259 if (ei->i_extra_isize) {
4260 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4261 raw_inode->i_version_hi =
4262 cpu_to_le32(inode->i_version >> 32);
4263 raw_inode->i_extra_isize =
4264 cpu_to_le16(ei->i_extra_isize);
4265 }
25ec56b5
JNC
4266 }
4267
814525f4
DW
4268 ext4_inode_csum_set(inode, raw_inode, ei);
4269
202ee5df
TT
4270 spin_unlock(&ei->i_raw_lock);
4271
830156c7 4272 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
73b50c1c 4273 rc = ext4_handle_dirty_metadata(handle, NULL, bh);
830156c7
FM
4274 if (!err)
4275 err = rc;
19f5fb7a 4276 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
202ee5df 4277 if (set_large_file) {
5d601255 4278 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
202ee5df
TT
4279 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
4280 if (err)
4281 goto out_brelse;
4282 ext4_update_dynamic_rev(sb);
4283 EXT4_SET_RO_COMPAT_FEATURE(sb,
4284 EXT4_FEATURE_RO_COMPAT_LARGE_FILE);
4285 ext4_handle_sync(handle);
4286 err = ext4_handle_dirty_super(handle, sb);
4287 }
b71fc079 4288 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
ac27a0ec 4289out_brelse:
af5bc92d 4290 brelse(bh);
617ba13b 4291 ext4_std_error(inode->i_sb, err);
ac27a0ec
DK
4292 return err;
4293}
4294
4295/*
617ba13b 4296 * ext4_write_inode()
ac27a0ec
DK
4297 *
4298 * We are called from a few places:
4299 *
87f7e416 4300 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
ac27a0ec 4301 * Here, there will be no transaction running. We wait for any running
4907cb7b 4302 * transaction to commit.
ac27a0ec 4303 *
87f7e416
TT
4304 * - Within flush work (sys_sync(), kupdate and such).
4305 * We wait on commit, if told to.
ac27a0ec 4306 *
87f7e416
TT
4307 * - Within iput_final() -> write_inode_now()
4308 * We wait on commit, if told to.
ac27a0ec
DK
4309 *
4310 * In all cases it is actually safe for us to return without doing anything,
4311 * because the inode has been copied into a raw inode buffer in
87f7e416
TT
4312 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
4313 * writeback.
ac27a0ec
DK
4314 *
4315 * Note that we are absolutely dependent upon all inode dirtiers doing the
4316 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4317 * which we are interested.
4318 *
4319 * It would be a bug for them to not do this. The code:
4320 *
4321 * mark_inode_dirty(inode)
4322 * stuff();
4323 * inode->i_size = expr;
4324 *
87f7e416
TT
4325 * is in error because write_inode() could occur while `stuff()' is running,
4326 * and the new i_size will be lost. Plus the inode will no longer be on the
4327 * superblock's dirty inode list.
ac27a0ec 4328 */
a9185b41 4329int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
ac27a0ec 4330{
91ac6f43
FM
4331 int err;
4332
87f7e416 4333 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
ac27a0ec
DK
4334 return 0;
4335
91ac6f43
FM
4336 if (EXT4_SB(inode->i_sb)->s_journal) {
4337 if (ext4_journal_current_handle()) {
4338 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4339 dump_stack();
4340 return -EIO;
4341 }
ac27a0ec 4342
10542c22
JK
4343 /*
4344 * No need to force transaction in WB_SYNC_NONE mode. Also
4345 * ext4_sync_fs() will force the commit after everything is
4346 * written.
4347 */
4348 if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
91ac6f43
FM
4349 return 0;
4350
4351 err = ext4_force_commit(inode->i_sb);
4352 } else {
4353 struct ext4_iloc iloc;
ac27a0ec 4354
8b472d73 4355 err = __ext4_get_inode_loc(inode, &iloc, 0);
91ac6f43
FM
4356 if (err)
4357 return err;
10542c22
JK
4358 /*
4359 * sync(2) will flush the whole buffer cache. No need to do
4360 * it here separately for each inode.
4361 */
4362 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
830156c7
FM
4363 sync_dirty_buffer(iloc.bh);
4364 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
c398eda0
TT
4365 EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
4366 "IO error syncing inode");
830156c7
FM
4367 err = -EIO;
4368 }
fd2dd9fb 4369 brelse(iloc.bh);
91ac6f43
FM
4370 }
4371 return err;
ac27a0ec
DK
4372}
4373
53e87268
JK
4374/*
4375 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4376 * buffers that are attached to a page stradding i_size and are undergoing
4377 * commit. In that case we have to wait for commit to finish and try again.
4378 */
4379static void ext4_wait_for_tail_page_commit(struct inode *inode)
4380{
4381 struct page *page;
4382 unsigned offset;
4383 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
4384 tid_t commit_tid = 0;
4385 int ret;
4386
4387 offset = inode->i_size & (PAGE_CACHE_SIZE - 1);
4388 /*
4389 * All buffers in the last page remain valid? Then there's nothing to
4390 * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE ==
4391 * blocksize case
4392 */
4393 if (offset > PAGE_CACHE_SIZE - (1 << inode->i_blkbits))
4394 return;
4395 while (1) {
4396 page = find_lock_page(inode->i_mapping,
4397 inode->i_size >> PAGE_CACHE_SHIFT);
4398 if (!page)
4399 return;
ca99fdd2
LC
4400 ret = __ext4_journalled_invalidatepage(page, offset,
4401 PAGE_CACHE_SIZE - offset);
53e87268
JK
4402 unlock_page(page);
4403 page_cache_release(page);
4404 if (ret != -EBUSY)
4405 return;
4406 commit_tid = 0;
4407 read_lock(&journal->j_state_lock);
4408 if (journal->j_committing_transaction)
4409 commit_tid = journal->j_committing_transaction->t_tid;
4410 read_unlock(&journal->j_state_lock);
4411 if (commit_tid)
4412 jbd2_log_wait_commit(journal, commit_tid);
4413 }
4414}
4415
ac27a0ec 4416/*
617ba13b 4417 * ext4_setattr()
ac27a0ec
DK
4418 *
4419 * Called from notify_change.
4420 *
4421 * We want to trap VFS attempts to truncate the file as soon as
4422 * possible. In particular, we want to make sure that when the VFS
4423 * shrinks i_size, we put the inode on the orphan list and modify
4424 * i_disksize immediately, so that during the subsequent flushing of
4425 * dirty pages and freeing of disk blocks, we can guarantee that any
4426 * commit will leave the blocks being flushed in an unused state on
4427 * disk. (On recovery, the inode will get truncated and the blocks will
4428 * be freed, so we have a strong guarantee that no future commit will
4429 * leave these blocks visible to the user.)
4430 *
678aaf48
JK
4431 * Another thing we have to assure is that if we are in ordered mode
4432 * and inode is still attached to the committing transaction, we must
4433 * we start writeout of all the dirty pages which are being truncated.
4434 * This way we are sure that all the data written in the previous
4435 * transaction are already on disk (truncate waits for pages under
4436 * writeback).
4437 *
4438 * Called with inode->i_mutex down.
ac27a0ec 4439 */
617ba13b 4440int ext4_setattr(struct dentry *dentry, struct iattr *attr)
ac27a0ec
DK
4441{
4442 struct inode *inode = dentry->d_inode;
4443 int error, rc = 0;
3d287de3 4444 int orphan = 0;
ac27a0ec
DK
4445 const unsigned int ia_valid = attr->ia_valid;
4446
4447 error = inode_change_ok(inode, attr);
4448 if (error)
4449 return error;
4450
12755627 4451 if (is_quota_modification(inode, attr))
871a2931 4452 dquot_initialize(inode);
08cefc7a
EB
4453 if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
4454 (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
ac27a0ec
DK
4455 handle_t *handle;
4456
4457 /* (user+group)*(old+new) structure, inode write (sb,
4458 * inode block, ? - but truncate inode update has it) */
9924a92a
TT
4459 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
4460 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
4461 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
ac27a0ec
DK
4462 if (IS_ERR(handle)) {
4463 error = PTR_ERR(handle);
4464 goto err_out;
4465 }
b43fa828 4466 error = dquot_transfer(inode, attr);
ac27a0ec 4467 if (error) {
617ba13b 4468 ext4_journal_stop(handle);
ac27a0ec
DK
4469 return error;
4470 }
4471 /* Update corresponding info in inode so that everything is in
4472 * one transaction */
4473 if (attr->ia_valid & ATTR_UID)
4474 inode->i_uid = attr->ia_uid;
4475 if (attr->ia_valid & ATTR_GID)
4476 inode->i_gid = attr->ia_gid;
617ba13b
MC
4477 error = ext4_mark_inode_dirty(handle, inode);
4478 ext4_journal_stop(handle);
ac27a0ec
DK
4479 }
4480
5208386c
JK
4481 if (attr->ia_valid & ATTR_SIZE && attr->ia_size != inode->i_size) {
4482 handle_t *handle;
562c72aa 4483
12e9b892 4484 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
e2b46574
ES
4485 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4486
0c095c7f
TT
4487 if (attr->ia_size > sbi->s_bitmap_maxbytes)
4488 return -EFBIG;
e2b46574 4489 }
dff6efc3
CH
4490
4491 if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
4492 inode_inc_iversion(inode);
4493
5208386c
JK
4494 if (S_ISREG(inode->i_mode) &&
4495 (attr->ia_size < inode->i_size)) {
4496 if (ext4_should_order_data(inode)) {
4497 error = ext4_begin_ordered_truncate(inode,
678aaf48 4498 attr->ia_size);
5208386c 4499 if (error)
678aaf48 4500 goto err_out;
5208386c
JK
4501 }
4502 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
4503 if (IS_ERR(handle)) {
4504 error = PTR_ERR(handle);
4505 goto err_out;
4506 }
4507 if (ext4_handle_valid(handle)) {
4508 error = ext4_orphan_add(handle, inode);
4509 orphan = 1;
4510 }
90e775b7 4511 down_write(&EXT4_I(inode)->i_data_sem);
5208386c
JK
4512 EXT4_I(inode)->i_disksize = attr->ia_size;
4513 rc = ext4_mark_inode_dirty(handle, inode);
4514 if (!error)
4515 error = rc;
90e775b7
JK
4516 /*
4517 * We have to update i_size under i_data_sem together
4518 * with i_disksize to avoid races with writeback code
4519 * running ext4_wb_update_i_disksize().
4520 */
4521 if (!error)
4522 i_size_write(inode, attr->ia_size);
4523 up_write(&EXT4_I(inode)->i_data_sem);
5208386c
JK
4524 ext4_journal_stop(handle);
4525 if (error) {
4526 ext4_orphan_del(NULL, inode);
678aaf48
JK
4527 goto err_out;
4528 }
90e775b7
JK
4529 } else
4530 i_size_write(inode, attr->ia_size);
53e87268 4531
5208386c
JK
4532 /*
4533 * Blocks are going to be removed from the inode. Wait
4534 * for dio in flight. Temporarily disable
4535 * dioread_nolock to prevent livelock.
4536 */
4537 if (orphan) {
4538 if (!ext4_should_journal_data(inode)) {
4539 ext4_inode_block_unlocked_dio(inode);
4540 inode_dio_wait(inode);
4541 ext4_inode_resume_unlocked_dio(inode);
4542 } else
4543 ext4_wait_for_tail_page_commit(inode);
1c9114f9 4544 }
5208386c
JK
4545 /*
4546 * Truncate pagecache after we've waited for commit
4547 * in data=journal mode to make pages freeable.
4548 */
7caef267 4549 truncate_pagecache(inode, inode->i_size);
072bd7ea 4550 }
5208386c
JK
4551 /*
4552 * We want to call ext4_truncate() even if attr->ia_size ==
4553 * inode->i_size for cases like truncation of fallocated space
4554 */
4555 if (attr->ia_valid & ATTR_SIZE)
4556 ext4_truncate(inode);
ac27a0ec 4557
1025774c
CH
4558 if (!rc) {
4559 setattr_copy(inode, attr);
4560 mark_inode_dirty(inode);
4561 }
4562
4563 /*
4564 * If the call to ext4_truncate failed to get a transaction handle at
4565 * all, we need to clean up the in-core orphan list manually.
4566 */
3d287de3 4567 if (orphan && inode->i_nlink)
617ba13b 4568 ext4_orphan_del(NULL, inode);
ac27a0ec
DK
4569
4570 if (!rc && (ia_valid & ATTR_MODE))
64e178a7 4571 rc = posix_acl_chmod(inode, inode->i_mode);
ac27a0ec
DK
4572
4573err_out:
617ba13b 4574 ext4_std_error(inode->i_sb, error);
ac27a0ec
DK
4575 if (!error)
4576 error = rc;
4577 return error;
4578}
4579
3e3398a0
MC
4580int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry,
4581 struct kstat *stat)
4582{
4583 struct inode *inode;
8af8eecc 4584 unsigned long long delalloc_blocks;
3e3398a0
MC
4585
4586 inode = dentry->d_inode;
4587 generic_fillattr(inode, stat);
4588
9206c561
AD
4589 /*
4590 * If there is inline data in the inode, the inode will normally not
4591 * have data blocks allocated (it may have an external xattr block).
4592 * Report at least one sector for such files, so tools like tar, rsync,
4593 * others doen't incorrectly think the file is completely sparse.
4594 */
4595 if (unlikely(ext4_has_inline_data(inode)))
4596 stat->blocks += (stat->size + 511) >> 9;
4597
3e3398a0
MC
4598 /*
4599 * We can't update i_blocks if the block allocation is delayed
4600 * otherwise in the case of system crash before the real block
4601 * allocation is done, we will have i_blocks inconsistent with
4602 * on-disk file blocks.
4603 * We always keep i_blocks updated together with real
4604 * allocation. But to not confuse with user, stat
4605 * will return the blocks that include the delayed allocation
4606 * blocks for this file.
4607 */
96607551 4608 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
9206c561
AD
4609 EXT4_I(inode)->i_reserved_data_blocks);
4610 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
3e3398a0
MC
4611 return 0;
4612}
ac27a0ec 4613
fffb2739
JK
4614static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
4615 int pextents)
a02908f1 4616{
12e9b892 4617 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
fffb2739
JK
4618 return ext4_ind_trans_blocks(inode, lblocks);
4619 return ext4_ext_index_trans_blocks(inode, pextents);
a02908f1 4620}
ac51d837 4621
ac27a0ec 4622/*
a02908f1
MC
4623 * Account for index blocks, block groups bitmaps and block group
4624 * descriptor blocks if modify datablocks and index blocks
4625 * worse case, the indexs blocks spread over different block groups
ac27a0ec 4626 *
a02908f1 4627 * If datablocks are discontiguous, they are possible to spread over
4907cb7b 4628 * different block groups too. If they are contiguous, with flexbg,
a02908f1 4629 * they could still across block group boundary.
ac27a0ec 4630 *
a02908f1
MC
4631 * Also account for superblock, inode, quota and xattr blocks
4632 */
fffb2739
JK
4633static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
4634 int pextents)
a02908f1 4635{
8df9675f
TT
4636 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
4637 int gdpblocks;
a02908f1
MC
4638 int idxblocks;
4639 int ret = 0;
4640
4641 /*
fffb2739
JK
4642 * How many index blocks need to touch to map @lblocks logical blocks
4643 * to @pextents physical extents?
a02908f1 4644 */
fffb2739 4645 idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
a02908f1
MC
4646
4647 ret = idxblocks;
4648
4649 /*
4650 * Now let's see how many group bitmaps and group descriptors need
4651 * to account
4652 */
fffb2739 4653 groups = idxblocks + pextents;
a02908f1 4654 gdpblocks = groups;
8df9675f
TT
4655 if (groups > ngroups)
4656 groups = ngroups;
a02908f1
MC
4657 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
4658 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
4659
4660 /* bitmaps and block group descriptor blocks */
4661 ret += groups + gdpblocks;
4662
4663 /* Blocks for super block, inode, quota and xattr blocks */
4664 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
4665
4666 return ret;
4667}
4668
4669/*
25985edc 4670 * Calculate the total number of credits to reserve to fit
f3bd1f3f
MC
4671 * the modification of a single pages into a single transaction,
4672 * which may include multiple chunks of block allocations.
ac27a0ec 4673 *
525f4ed8 4674 * This could be called via ext4_write_begin()
ac27a0ec 4675 *
525f4ed8 4676 * We need to consider the worse case, when
a02908f1 4677 * one new block per extent.
ac27a0ec 4678 */
a86c6181 4679int ext4_writepage_trans_blocks(struct inode *inode)
ac27a0ec 4680{
617ba13b 4681 int bpp = ext4_journal_blocks_per_page(inode);
ac27a0ec
DK
4682 int ret;
4683
fffb2739 4684 ret = ext4_meta_trans_blocks(inode, bpp, bpp);
a86c6181 4685
a02908f1 4686 /* Account for data blocks for journalled mode */
617ba13b 4687 if (ext4_should_journal_data(inode))
a02908f1 4688 ret += bpp;
ac27a0ec
DK
4689 return ret;
4690}
f3bd1f3f
MC
4691
4692/*
4693 * Calculate the journal credits for a chunk of data modification.
4694 *
4695 * This is called from DIO, fallocate or whoever calling
79e83036 4696 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
f3bd1f3f
MC
4697 *
4698 * journal buffers for data blocks are not included here, as DIO
4699 * and fallocate do no need to journal data buffers.
4700 */
4701int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
4702{
4703 return ext4_meta_trans_blocks(inode, nrblocks, 1);
4704}
4705
ac27a0ec 4706/*
617ba13b 4707 * The caller must have previously called ext4_reserve_inode_write().
ac27a0ec
DK
4708 * Give this, we know that the caller already has write access to iloc->bh.
4709 */
617ba13b 4710int ext4_mark_iloc_dirty(handle_t *handle,
de9a55b8 4711 struct inode *inode, struct ext4_iloc *iloc)
ac27a0ec
DK
4712{
4713 int err = 0;
4714
c64db50e 4715 if (IS_I_VERSION(inode))
25ec56b5
JNC
4716 inode_inc_iversion(inode);
4717
ac27a0ec
DK
4718 /* the do_update_inode consumes one bh->b_count */
4719 get_bh(iloc->bh);
4720
dab291af 4721 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
830156c7 4722 err = ext4_do_update_inode(handle, inode, iloc);
ac27a0ec
DK
4723 put_bh(iloc->bh);
4724 return err;
4725}
4726
4727/*
4728 * On success, We end up with an outstanding reference count against
4729 * iloc->bh. This _must_ be cleaned up later.
4730 */
4731
4732int
617ba13b
MC
4733ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
4734 struct ext4_iloc *iloc)
ac27a0ec 4735{
0390131b
FM
4736 int err;
4737
4738 err = ext4_get_inode_loc(inode, iloc);
4739 if (!err) {
4740 BUFFER_TRACE(iloc->bh, "get_write_access");
4741 err = ext4_journal_get_write_access(handle, iloc->bh);
4742 if (err) {
4743 brelse(iloc->bh);
4744 iloc->bh = NULL;
ac27a0ec
DK
4745 }
4746 }
617ba13b 4747 ext4_std_error(inode->i_sb, err);
ac27a0ec
DK
4748 return err;
4749}
4750
6dd4ee7c
KS
4751/*
4752 * Expand an inode by new_extra_isize bytes.
4753 * Returns 0 on success or negative error number on failure.
4754 */
1d03ec98
AK
4755static int ext4_expand_extra_isize(struct inode *inode,
4756 unsigned int new_extra_isize,
4757 struct ext4_iloc iloc,
4758 handle_t *handle)
6dd4ee7c
KS
4759{
4760 struct ext4_inode *raw_inode;
4761 struct ext4_xattr_ibody_header *header;
6dd4ee7c
KS
4762
4763 if (EXT4_I(inode)->i_extra_isize >= new_extra_isize)
4764 return 0;
4765
4766 raw_inode = ext4_raw_inode(&iloc);
4767
4768 header = IHDR(inode, raw_inode);
6dd4ee7c
KS
4769
4770 /* No extended attributes present */
19f5fb7a
TT
4771 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
4772 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
6dd4ee7c
KS
4773 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0,
4774 new_extra_isize);
4775 EXT4_I(inode)->i_extra_isize = new_extra_isize;
4776 return 0;
4777 }
4778
4779 /* try to expand with EAs present */
4780 return ext4_expand_extra_isize_ea(inode, new_extra_isize,
4781 raw_inode, handle);
4782}
4783
ac27a0ec
DK
4784/*
4785 * What we do here is to mark the in-core inode as clean with respect to inode
4786 * dirtiness (it may still be data-dirty).
4787 * This means that the in-core inode may be reaped by prune_icache
4788 * without having to perform any I/O. This is a very good thing,
4789 * because *any* task may call prune_icache - even ones which
4790 * have a transaction open against a different journal.
4791 *
4792 * Is this cheating? Not really. Sure, we haven't written the
4793 * inode out, but prune_icache isn't a user-visible syncing function.
4794 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4795 * we start and wait on commits.
ac27a0ec 4796 */
617ba13b 4797int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
ac27a0ec 4798{
617ba13b 4799 struct ext4_iloc iloc;
6dd4ee7c
KS
4800 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4801 static unsigned int mnt_count;
4802 int err, ret;
ac27a0ec
DK
4803
4804 might_sleep();
7ff9c073 4805 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
617ba13b 4806 err = ext4_reserve_inode_write(handle, inode, &iloc);
0390131b
FM
4807 if (ext4_handle_valid(handle) &&
4808 EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize &&
19f5fb7a 4809 !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
6dd4ee7c
KS
4810 /*
4811 * We need extra buffer credits since we may write into EA block
4812 * with this same handle. If journal_extend fails, then it will
4813 * only result in a minor loss of functionality for that inode.
4814 * If this is felt to be critical, then e2fsck should be run to
4815 * force a large enough s_min_extra_isize.
4816 */
4817 if ((jbd2_journal_extend(handle,
4818 EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) {
4819 ret = ext4_expand_extra_isize(inode,
4820 sbi->s_want_extra_isize,
4821 iloc, handle);
4822 if (ret) {
19f5fb7a
TT
4823 ext4_set_inode_state(inode,
4824 EXT4_STATE_NO_EXPAND);
c1bddad9
AK
4825 if (mnt_count !=
4826 le16_to_cpu(sbi->s_es->s_mnt_count)) {
12062ddd 4827 ext4_warning(inode->i_sb,
6dd4ee7c
KS
4828 "Unable to expand inode %lu. Delete"
4829 " some EAs or run e2fsck.",
4830 inode->i_ino);
c1bddad9
AK
4831 mnt_count =
4832 le16_to_cpu(sbi->s_es->s_mnt_count);
6dd4ee7c
KS
4833 }
4834 }
4835 }
4836 }
ac27a0ec 4837 if (!err)
617ba13b 4838 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
ac27a0ec
DK
4839 return err;
4840}
4841
4842/*
617ba13b 4843 * ext4_dirty_inode() is called from __mark_inode_dirty()
ac27a0ec
DK
4844 *
4845 * We're really interested in the case where a file is being extended.
4846 * i_size has been changed by generic_commit_write() and we thus need
4847 * to include the updated inode in the current transaction.
4848 *
5dd4056d 4849 * Also, dquot_alloc_block() will always dirty the inode when blocks
ac27a0ec
DK
4850 * are allocated to the file.
4851 *
4852 * If the inode is marked synchronous, we don't honour that here - doing
4853 * so would cause a commit on atime updates, which we don't bother doing.
4854 * We handle synchronous inodes at the highest possible level.
4855 */
aa385729 4856void ext4_dirty_inode(struct inode *inode, int flags)
ac27a0ec 4857{
ac27a0ec
DK
4858 handle_t *handle;
4859
9924a92a 4860 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
ac27a0ec
DK
4861 if (IS_ERR(handle))
4862 goto out;
f3dc272f 4863
f3dc272f
CW
4864 ext4_mark_inode_dirty(handle, inode);
4865
617ba13b 4866 ext4_journal_stop(handle);
ac27a0ec
DK
4867out:
4868 return;
4869}
4870
4871#if 0
4872/*
4873 * Bind an inode's backing buffer_head into this transaction, to prevent
4874 * it from being flushed to disk early. Unlike
617ba13b 4875 * ext4_reserve_inode_write, this leaves behind no bh reference and
ac27a0ec
DK
4876 * returns no iloc structure, so the caller needs to repeat the iloc
4877 * lookup to mark the inode dirty later.
4878 */
617ba13b 4879static int ext4_pin_inode(handle_t *handle, struct inode *inode)
ac27a0ec 4880{
617ba13b 4881 struct ext4_iloc iloc;
ac27a0ec
DK
4882
4883 int err = 0;
4884 if (handle) {
617ba13b 4885 err = ext4_get_inode_loc(inode, &iloc);
ac27a0ec
DK
4886 if (!err) {
4887 BUFFER_TRACE(iloc.bh, "get_write_access");
dab291af 4888 err = jbd2_journal_get_write_access(handle, iloc.bh);
ac27a0ec 4889 if (!err)
0390131b 4890 err = ext4_handle_dirty_metadata(handle,
73b50c1c 4891 NULL,
0390131b 4892 iloc.bh);
ac27a0ec
DK
4893 brelse(iloc.bh);
4894 }
4895 }
617ba13b 4896 ext4_std_error(inode->i_sb, err);
ac27a0ec
DK
4897 return err;
4898}
4899#endif
4900
617ba13b 4901int ext4_change_inode_journal_flag(struct inode *inode, int val)
ac27a0ec
DK
4902{
4903 journal_t *journal;
4904 handle_t *handle;
4905 int err;
4906
4907 /*
4908 * We have to be very careful here: changing a data block's
4909 * journaling status dynamically is dangerous. If we write a
4910 * data block to the journal, change the status and then delete
4911 * that block, we risk forgetting to revoke the old log record
4912 * from the journal and so a subsequent replay can corrupt data.
4913 * So, first we make sure that the journal is empty and that
4914 * nobody is changing anything.
4915 */
4916
617ba13b 4917 journal = EXT4_JOURNAL(inode);
0390131b
FM
4918 if (!journal)
4919 return 0;
d699594d 4920 if (is_journal_aborted(journal))
ac27a0ec 4921 return -EROFS;
2aff57b0
YY
4922 /* We have to allocate physical blocks for delalloc blocks
4923 * before flushing journal. otherwise delalloc blocks can not
4924 * be allocated any more. even more truncate on delalloc blocks
4925 * could trigger BUG by flushing delalloc blocks in journal.
4926 * There is no delalloc block in non-journal data mode.
4927 */
4928 if (val && test_opt(inode->i_sb, DELALLOC)) {
4929 err = ext4_alloc_da_blocks(inode);
4930 if (err < 0)
4931 return err;
4932 }
ac27a0ec 4933
17335dcc
DM
4934 /* Wait for all existing dio workers */
4935 ext4_inode_block_unlocked_dio(inode);
4936 inode_dio_wait(inode);
4937
dab291af 4938 jbd2_journal_lock_updates(journal);
ac27a0ec
DK
4939
4940 /*
4941 * OK, there are no updates running now, and all cached data is
4942 * synced to disk. We are now in a completely consistent state
4943 * which doesn't have anything in the journal, and we know that
4944 * no filesystem updates are running, so it is safe to modify
4945 * the inode's in-core data-journaling state flag now.
4946 */
4947
4948 if (val)
12e9b892 4949 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5872ddaa
YY
4950 else {
4951 jbd2_journal_flush(journal);
12e9b892 4952 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5872ddaa 4953 }
617ba13b 4954 ext4_set_aops(inode);
ac27a0ec 4955
dab291af 4956 jbd2_journal_unlock_updates(journal);
17335dcc 4957 ext4_inode_resume_unlocked_dio(inode);
ac27a0ec
DK
4958
4959 /* Finally we can mark the inode as dirty. */
4960
9924a92a 4961 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
ac27a0ec
DK
4962 if (IS_ERR(handle))
4963 return PTR_ERR(handle);
4964
617ba13b 4965 err = ext4_mark_inode_dirty(handle, inode);
0390131b 4966 ext4_handle_sync(handle);
617ba13b
MC
4967 ext4_journal_stop(handle);
4968 ext4_std_error(inode->i_sb, err);
ac27a0ec
DK
4969
4970 return err;
4971}
2e9ee850
AK
4972
4973static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
4974{
4975 return !buffer_mapped(bh);
4976}
4977
c2ec175c 4978int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
2e9ee850 4979{
c2ec175c 4980 struct page *page = vmf->page;
2e9ee850
AK
4981 loff_t size;
4982 unsigned long len;
9ea7df53 4983 int ret;
2e9ee850 4984 struct file *file = vma->vm_file;
496ad9aa 4985 struct inode *inode = file_inode(file);
2e9ee850 4986 struct address_space *mapping = inode->i_mapping;
9ea7df53
JK
4987 handle_t *handle;
4988 get_block_t *get_block;
4989 int retries = 0;
2e9ee850 4990
8e8ad8a5 4991 sb_start_pagefault(inode->i_sb);
041bbb6d 4992 file_update_time(vma->vm_file);
9ea7df53
JK
4993 /* Delalloc case is easy... */
4994 if (test_opt(inode->i_sb, DELALLOC) &&
4995 !ext4_should_journal_data(inode) &&
4996 !ext4_nonda_switch(inode->i_sb)) {
4997 do {
4998 ret = __block_page_mkwrite(vma, vmf,
4999 ext4_da_get_block_prep);
5000 } while (ret == -ENOSPC &&
5001 ext4_should_retry_alloc(inode->i_sb, &retries));
5002 goto out_ret;
2e9ee850 5003 }
0e499890
DW
5004
5005 lock_page(page);
9ea7df53
JK
5006 size = i_size_read(inode);
5007 /* Page got truncated from under us? */
5008 if (page->mapping != mapping || page_offset(page) > size) {
5009 unlock_page(page);
5010 ret = VM_FAULT_NOPAGE;
5011 goto out;
0e499890 5012 }
2e9ee850
AK
5013
5014 if (page->index == size >> PAGE_CACHE_SHIFT)
5015 len = size & ~PAGE_CACHE_MASK;
5016 else
5017 len = PAGE_CACHE_SIZE;
a827eaff 5018 /*
9ea7df53
JK
5019 * Return if we have all the buffers mapped. This avoids the need to do
5020 * journal_start/journal_stop which can block and take a long time
a827eaff 5021 */
2e9ee850 5022 if (page_has_buffers(page)) {
f19d5870
TM
5023 if (!ext4_walk_page_buffers(NULL, page_buffers(page),
5024 0, len, NULL,
5025 ext4_bh_unmapped)) {
9ea7df53 5026 /* Wait so that we don't change page under IO */
1d1d1a76 5027 wait_for_stable_page(page);
9ea7df53
JK
5028 ret = VM_FAULT_LOCKED;
5029 goto out;
a827eaff 5030 }
2e9ee850 5031 }
a827eaff 5032 unlock_page(page);
9ea7df53
JK
5033 /* OK, we need to fill the hole... */
5034 if (ext4_should_dioread_nolock(inode))
5035 get_block = ext4_get_block_write;
5036 else
5037 get_block = ext4_get_block;
5038retry_alloc:
9924a92a
TT
5039 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
5040 ext4_writepage_trans_blocks(inode));
9ea7df53 5041 if (IS_ERR(handle)) {
c2ec175c 5042 ret = VM_FAULT_SIGBUS;
9ea7df53
JK
5043 goto out;
5044 }
5045 ret = __block_page_mkwrite(vma, vmf, get_block);
5046 if (!ret && ext4_should_journal_data(inode)) {
f19d5870 5047 if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
9ea7df53
JK
5048 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) {
5049 unlock_page(page);
5050 ret = VM_FAULT_SIGBUS;
fcbb5515 5051 ext4_journal_stop(handle);
9ea7df53
JK
5052 goto out;
5053 }
5054 ext4_set_inode_state(inode, EXT4_STATE_JDATA);
5055 }
5056 ext4_journal_stop(handle);
5057 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
5058 goto retry_alloc;
5059out_ret:
5060 ret = block_page_mkwrite_return(ret);
5061out:
8e8ad8a5 5062 sb_end_pagefault(inode->i_sb);
2e9ee850
AK
5063 return ret;
5064}