1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * linux/fs/ext4/inode.c
4 *
5 * Copyright (C) 1992, 1993, 1994, 1995
6 * Remy Card (card@masi.ibp.fr)
7 * Laboratoire MASI - Institut Blaise Pascal
8 * Universite Pierre et Marie Curie (Paris VI)
9 *
10 * from
11 *
12 * linux/fs/minix/inode.c
13 *
14 * Copyright (C) 1991, 1992 Linus Torvalds
15 *
16 * 64-bit file support on 64-bit platforms by Jakub Jelinek
17 * (jj@sunsite.ms.mff.cuni.cz)
18 *
19 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
20 */
21
22 #include <linux/fs.h>
23 #include <linux/mount.h>
24 #include <linux/time.h>
25 #include <linux/highuid.h>
26 #include <linux/pagemap.h>
27 #include <linux/dax.h>
28 #include <linux/quotaops.h>
29 #include <linux/string.h>
30 #include <linux/buffer_head.h>
31 #include <linux/writeback.h>
32 #include <linux/pagevec.h>
33 #include <linux/mpage.h>
34 #include <linux/namei.h>
35 #include <linux/uio.h>
36 #include <linux/bio.h>
37 #include <linux/workqueue.h>
38 #include <linux/kernel.h>
39 #include <linux/printk.h>
40 #include <linux/slab.h>
41 #include <linux/bitops.h>
42 #include <linux/iomap.h>
43 #include <linux/iversion.h>
44
45 #include "ext4_jbd2.h"
46 #include "xattr.h"
47 #include "acl.h"
48 #include "truncate.h"
49
50 #include <trace/events/ext4.h>
51
ext4_inode_csum(struct inode * inode,struct ext4_inode * raw,struct ext4_inode_info * ei)52 static __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 __u32 csum;
57 __u16 dummy_csum = 0;
58 int offset = offsetof(struct ext4_inode, i_checksum_lo);
59 unsigned int csum_size = sizeof(dummy_csum);
60
61 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw, offset);
62 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, csum_size);
63 offset += csum_size;
64 csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
65 EXT4_GOOD_OLD_INODE_SIZE - offset);
66
67 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
68 offset = offsetof(struct ext4_inode, i_checksum_hi);
69 csum = ext4_chksum(sbi, csum, (__u8 *)raw +
70 EXT4_GOOD_OLD_INODE_SIZE,
71 offset - EXT4_GOOD_OLD_INODE_SIZE);
72 if (EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
73 csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum,
74 csum_size);
75 offset += csum_size;
76 }
77 csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
78 EXT4_INODE_SIZE(inode->i_sb) - offset);
79 }
80
81 return csum;
82 }
83
ext4_inode_csum_verify(struct inode * inode,struct ext4_inode * raw,struct ext4_inode_info * ei)84 static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
85 struct ext4_inode_info *ei)
86 {
87 __u32 provided, calculated;
88
89 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
90 cpu_to_le32(EXT4_OS_LINUX) ||
91 !ext4_has_metadata_csum(inode->i_sb))
92 return 1;
93
94 provided = le16_to_cpu(raw->i_checksum_lo);
95 calculated = ext4_inode_csum(inode, raw, ei);
96 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
97 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
98 provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
99 else
100 calculated &= 0xFFFF;
101
102 return provided == calculated;
103 }
104
ext4_inode_csum_set(struct inode * inode,struct ext4_inode * raw,struct ext4_inode_info * ei)105 void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
106 struct ext4_inode_info *ei)
107 {
108 __u32 csum;
109
110 if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
111 cpu_to_le32(EXT4_OS_LINUX) ||
112 !ext4_has_metadata_csum(inode->i_sb))
113 return;
114
115 csum = ext4_inode_csum(inode, raw, ei);
116 raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
117 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
118 EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
119 raw->i_checksum_hi = cpu_to_le16(csum >> 16);
120 }
121
ext4_begin_ordered_truncate(struct inode * inode,loff_t new_size)122 static inline int ext4_begin_ordered_truncate(struct inode *inode,
123 loff_t new_size)
124 {
125 trace_ext4_begin_ordered_truncate(inode, new_size);
126 /*
127 * If jinode is zero, then we never opened the file for
128 * writing, so there's no need to call
129 * jbd2_journal_begin_ordered_truncate() since there's no
130 * outstanding writes we need to flush.
131 */
132 if (!EXT4_I(inode)->jinode)
133 return 0;
134 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
135 EXT4_I(inode)->jinode,
136 new_size);
137 }
138
139 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
140 int pextents);
141
142 /*
143 * Test whether an inode is a fast symlink.
144 * A fast symlink has its symlink data stored in ext4_inode_info->i_data.
145 */
ext4_inode_is_fast_symlink(struct inode * inode)146 int ext4_inode_is_fast_symlink(struct inode *inode)
147 {
148 if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
149 int ea_blocks = EXT4_I(inode)->i_file_acl ?
150 EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0;
151
152 if (ext4_has_inline_data(inode))
153 return 0;
154
155 return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
156 }
157 return S_ISLNK(inode->i_mode) && inode->i_size &&
158 (inode->i_size < EXT4_N_BLOCKS * 4);
159 }
160
161 /*
162 * Called at the last iput() if i_nlink is zero.
163 */
ext4_evict_inode(struct inode * inode)164 void ext4_evict_inode(struct inode *inode)
165 {
166 handle_t *handle;
167 int err;
168 /*
169 * Credits for final inode cleanup and freeing:
170 * sb + inode (ext4_orphan_del()), block bitmap, group descriptor
171 * (xattr block freeing), bitmap, group descriptor (inode freeing)
172 */
173 int extra_credits = 6;
174 struct ext4_xattr_inode_array *ea_inode_array = NULL;
175 bool freeze_protected = false;
176
177 trace_ext4_evict_inode(inode);
178
179 if (EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)
180 ext4_evict_ea_inode(inode);
181 if (inode->i_nlink) {
182 truncate_inode_pages_final(&inode->i_data);
183
184 goto no_delete;
185 }
186
187 if (is_bad_inode(inode))
188 goto no_delete;
189 dquot_initialize(inode);
190
191 if (ext4_should_order_data(inode))
192 ext4_begin_ordered_truncate(inode, 0);
193 truncate_inode_pages_final(&inode->i_data);
194
195 /*
196 * For inodes with journalled data, transaction commit could have
197 * dirtied the inode. And for inodes with dioread_nolock, unwritten
198 * extents converting worker could merge extents and also have dirtied
199 * the inode. Flush worker is ignoring it because of I_FREEING flag but
200 * we still need to remove the inode from the writeback lists.
201 */
202 if (!list_empty_careful(&inode->i_io_list))
203 inode_io_list_del(inode);
204
205 /*
206 * Protect us against freezing - iput() caller didn't have to have any
207 * protection against it. When we are in a running transaction though,
208 * we are already protected against freezing and we cannot grab further
209 * protection due to lock ordering constraints.
210 */
211 if (!ext4_journal_current_handle()) {
212 sb_start_intwrite(inode->i_sb);
213 freeze_protected = true;
214 }
215
216 if (!IS_NOQUOTA(inode))
217 extra_credits += EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb);
218
219 /*
220 * Block bitmap, group descriptor, and inode are accounted in both
221 * ext4_blocks_for_truncate() and extra_credits. So subtract 3.
222 */
223 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
224 ext4_blocks_for_truncate(inode) + extra_credits - 3);
225 if (IS_ERR(handle)) {
226 ext4_std_error(inode->i_sb, PTR_ERR(handle));
227 /*
228 * If we're going to skip the normal cleanup, we still need to
229 * make sure that the in-core orphan linked list is properly
230 * cleaned up.
231 */
232 ext4_orphan_del(NULL, inode);
233 if (freeze_protected)
234 sb_end_intwrite(inode->i_sb);
235 goto no_delete;
236 }
237
238 if (IS_SYNC(inode))
239 ext4_handle_sync(handle);
240
241 /*
242 * Set inode->i_size to 0 before calling ext4_truncate(). We need
243 * special handling of symlinks here because i_size is used to
244 * determine whether ext4_inode_info->i_data contains symlink data or
245 * block mappings. Setting i_size to 0 will remove its fast symlink
246 * status. Erase i_data so that it becomes a valid empty block map.
247 */
248 if (ext4_inode_is_fast_symlink(inode))
249 memset(EXT4_I(inode)->i_data, 0, sizeof(EXT4_I(inode)->i_data));
250 inode->i_size = 0;
251 err = ext4_mark_inode_dirty(handle, inode);
252 if (err) {
253 ext4_warning(inode->i_sb,
254 "couldn't mark inode dirty (err %d)", err);
255 goto stop_handle;
256 }
257 if (inode->i_blocks) {
258 err = ext4_truncate(inode);
259 if (err) {
260 ext4_error_err(inode->i_sb, -err,
261 "couldn't truncate inode %lu (err %d)",
262 inode->i_ino, err);
263 goto stop_handle;
264 }
265 }
266
267 /* Remove xattr references. */
268 err = ext4_xattr_delete_inode(handle, inode, &ea_inode_array,
269 extra_credits);
270 if (err) {
271 ext4_warning(inode->i_sb, "xattr delete (err %d)", err);
272 stop_handle:
273 ext4_journal_stop(handle);
274 ext4_orphan_del(NULL, inode);
275 if (freeze_protected)
276 sb_end_intwrite(inode->i_sb);
277 ext4_xattr_inode_array_free(ea_inode_array);
278 goto no_delete;
279 }
280
281 /*
282 * Kill off the orphan record which ext4_truncate created.
283 * AKPM: I think this can be inside the above `if'.
284 * Note that ext4_orphan_del() has to be able to cope with the
285 * deletion of a non-existent orphan - this is because we don't
286 * know if ext4_truncate() actually created an orphan record.
287 * (Well, we could do this if we need to, but heck - it works)
288 */
289 ext4_orphan_del(handle, inode);
290 EXT4_I(inode)->i_dtime = (__u32)ktime_get_real_seconds();
291
292 /*
293 * One subtle ordering requirement: if anything has gone wrong
294 * (transaction abort, IO errors, whatever), then we can still
295 * do these next steps (the fs will already have been marked as
296 * having errors), but we can't free the inode if the mark_dirty
297 * fails.
298 */
299 if (ext4_mark_inode_dirty(handle, inode))
300 /* If that failed, just do the required in-core inode clear. */
301 ext4_clear_inode(inode);
302 else
303 ext4_free_inode(handle, inode);
304 ext4_journal_stop(handle);
305 if (freeze_protected)
306 sb_end_intwrite(inode->i_sb);
307 ext4_xattr_inode_array_free(ea_inode_array);
308 return;
309 no_delete:
310 /*
311 * Check out some where else accidentally dirty the evicting inode,
312 * which may probably cause inode use-after-free issues later.
313 */
314 WARN_ON_ONCE(!list_empty_careful(&inode->i_io_list));
315
316 if (!list_empty(&EXT4_I(inode)->i_fc_list))
317 ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_NOMEM, NULL);
318 ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
319 }
320
321 #ifdef CONFIG_QUOTA
ext4_get_reserved_space(struct inode * inode)322 qsize_t *ext4_get_reserved_space(struct inode *inode)
323 {
324 return &EXT4_I(inode)->i_reserved_quota;
325 }
326 #endif
327
328 /*
329 * Called with i_data_sem down, which is important since we can call
330 * ext4_discard_preallocations() from here.
331 */
ext4_da_update_reserve_space(struct inode * inode,int used,int quota_claim)332 void ext4_da_update_reserve_space(struct inode *inode,
333 int used, int quota_claim)
334 {
335 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
336 struct ext4_inode_info *ei = EXT4_I(inode);
337
338 spin_lock(&ei->i_block_reservation_lock);
339 trace_ext4_da_update_reserve_space(inode, used, quota_claim);
340 if (unlikely(used > ei->i_reserved_data_blocks)) {
341 ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
342 "with only %d reserved data blocks",
343 __func__, inode->i_ino, used,
344 ei->i_reserved_data_blocks);
345 WARN_ON(1);
346 used = ei->i_reserved_data_blocks;
347 }
348
349 /* Update per-inode reservations */
350 ei->i_reserved_data_blocks -= used;
351 percpu_counter_sub(&sbi->s_dirtyclusters_counter, used);
352
353 spin_unlock(&ei->i_block_reservation_lock);
354
355 /* Update quota subsystem for data blocks */
356 if (quota_claim)
357 dquot_claim_block(inode, EXT4_C2B(sbi, used));
358 else {
359 /*
360 * We did fallocate with an offset that is already delayed
361 * allocated. So on delayed allocated writeback we should
362 * not re-claim the quota for fallocated blocks.
363 */
364 dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
365 }
366
367 /*
368 * If we have done all the pending block allocations and if
369 * there aren't any writers on the inode, we can discard the
370 * inode's preallocations.
371 */
372 if ((ei->i_reserved_data_blocks == 0) &&
373 !inode_is_open_for_write(inode))
374 ext4_discard_preallocations(inode);
375 }
376
__check_block_validity(struct inode * inode,const char * func,unsigned int line,struct ext4_map_blocks * map)377 static int __check_block_validity(struct inode *inode, const char *func,
378 unsigned int line,
379 struct ext4_map_blocks *map)
380 {
381 if (ext4_has_feature_journal(inode->i_sb) &&
382 (inode->i_ino ==
383 le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_journal_inum)))
384 return 0;
385 if (!ext4_inode_block_valid(inode, map->m_pblk, map->m_len)) {
386 ext4_error_inode(inode, func, line, map->m_pblk,
387 "lblock %lu mapped to illegal pblock %llu "
388 "(length %d)", (unsigned long) map->m_lblk,
389 map->m_pblk, map->m_len);
390 return -EFSCORRUPTED;
391 }
392 return 0;
393 }
394
ext4_issue_zeroout(struct inode * inode,ext4_lblk_t lblk,ext4_fsblk_t pblk,ext4_lblk_t len)395 int ext4_issue_zeroout(struct inode *inode, ext4_lblk_t lblk, ext4_fsblk_t pblk,
396 ext4_lblk_t len)
397 {
398 int ret;
399
400 if (IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode))
401 return fscrypt_zeroout_range(inode, lblk, pblk, len);
402
403 ret = sb_issue_zeroout(inode->i_sb, pblk, len, GFP_NOFS);
404 if (ret > 0)
405 ret = 0;
406
407 return ret;
408 }
409
410 #define check_block_validity(inode, map) \
411 __check_block_validity((inode), __func__, __LINE__, (map))
412
413 #ifdef ES_AGGRESSIVE_TEST
ext4_map_blocks_es_recheck(handle_t * handle,struct inode * inode,struct ext4_map_blocks * es_map,struct ext4_map_blocks * map,int flags)414 static void ext4_map_blocks_es_recheck(handle_t *handle,
415 struct inode *inode,
416 struct ext4_map_blocks *es_map,
417 struct ext4_map_blocks *map,
418 int flags)
419 {
420 int retval;
421
422 map->m_flags = 0;
423 /*
424 * There is a race window that the result is not the same.
425 * e.g. xfstests #223 when dioread_nolock enables. The reason
426 * is that we lookup a block mapping in extent status tree with
427 * out taking i_data_sem. So at the time the unwritten extent
428 * could be converted.
429 */
430 down_read(&EXT4_I(inode)->i_data_sem);
431 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
432 retval = ext4_ext_map_blocks(handle, inode, map, 0);
433 } else {
434 retval = ext4_ind_map_blocks(handle, inode, map, 0);
435 }
436 up_read((&EXT4_I(inode)->i_data_sem));
437
438 /*
439 * We don't check m_len because extent will be collpased in status
440 * tree. So the m_len might not equal.
441 */
442 if (es_map->m_lblk != map->m_lblk ||
443 es_map->m_flags != map->m_flags ||
444 es_map->m_pblk != map->m_pblk) {
445 printk("ES cache assertion failed for inode: %lu "
446 "es_cached ex [%d/%d/%llu/%x] != "
447 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
448 inode->i_ino, es_map->m_lblk, es_map->m_len,
449 es_map->m_pblk, es_map->m_flags, map->m_lblk,
450 map->m_len, map->m_pblk, map->m_flags,
451 retval, flags);
452 }
453 }
454 #endif /* ES_AGGRESSIVE_TEST */
455
456 /*
457 * The ext4_map_blocks() function tries to look up the requested blocks,
458 * and returns if the blocks are already mapped.
459 *
460 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
461 * and store the allocated blocks in the result buffer head and mark it
462 * mapped.
463 *
464 * If file type is extents based, it will call ext4_ext_map_blocks(),
465 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
466 * based files
467 *
468 * On success, it returns the number of blocks being mapped or allocated.
469 * If flags doesn't contain EXT4_GET_BLOCKS_CREATE the blocks are
470 * pre-allocated and unwritten, the resulting @map is marked as unwritten.
471 * If the flags contain EXT4_GET_BLOCKS_CREATE, it will mark @map as mapped.
472 *
473 * It returns 0 if plain look up failed (blocks have not been allocated), in
474 * that case, @map is returned as unmapped but we still do fill map->m_len to
475 * indicate the length of a hole starting at map->m_lblk.
476 *
477 * It returns the error in case of allocation failure.
478 */
ext4_map_blocks(handle_t * handle,struct inode * inode,struct ext4_map_blocks * map,int flags)479 int ext4_map_blocks(handle_t *handle, struct inode *inode,
480 struct ext4_map_blocks *map, int flags)
481 {
482 struct extent_status es;
483 int retval;
484 int ret = 0;
485 #ifdef ES_AGGRESSIVE_TEST
486 struct ext4_map_blocks orig_map;
487
488 memcpy(&orig_map, map, sizeof(*map));
489 #endif
490
491 map->m_flags = 0;
492 ext_debug(inode, "flag 0x%x, max_blocks %u, logical block %lu\n",
493 flags, map->m_len, (unsigned long) map->m_lblk);
494
495 /*
496 * ext4_map_blocks returns an int, and m_len is an unsigned int
497 */
498 if (unlikely(map->m_len > INT_MAX))
499 map->m_len = INT_MAX;
500
501 /* We can handle the block number less than EXT_MAX_BLOCKS */
502 if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
503 return -EFSCORRUPTED;
504
505 /* Lookup extent status tree firstly */
506 if (!(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY) &&
507 ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) {
508 if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
509 map->m_pblk = ext4_es_pblock(&es) +
510 map->m_lblk - es.es_lblk;
511 map->m_flags |= ext4_es_is_written(&es) ?
512 EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
513 retval = es.es_len - (map->m_lblk - es.es_lblk);
514 if (retval > map->m_len)
515 retval = map->m_len;
516 map->m_len = retval;
517 } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
518 map->m_pblk = 0;
519 map->m_flags |= ext4_es_is_delayed(&es) ?
520 EXT4_MAP_DELAYED : 0;
521 retval = es.es_len - (map->m_lblk - es.es_lblk);
522 if (retval > map->m_len)
523 retval = map->m_len;
524 map->m_len = retval;
525 retval = 0;
526 } else {
527 BUG();
528 }
529
530 if (flags & EXT4_GET_BLOCKS_CACHED_NOWAIT)
531 return retval;
532 #ifdef ES_AGGRESSIVE_TEST
533 ext4_map_blocks_es_recheck(handle, inode, map,
534 &orig_map, flags);
535 #endif
536 goto found;
537 }
538 /*
539 * In the query cache no-wait mode, nothing we can do more if we
540 * cannot find extent in the cache.
541 */
542 if (flags & EXT4_GET_BLOCKS_CACHED_NOWAIT)
543 return 0;
544
545 /*
546 * Try to see if we can get the block without requesting a new
547 * file system block.
548 */
549 down_read(&EXT4_I(inode)->i_data_sem);
550 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
551 retval = ext4_ext_map_blocks(handle, inode, map, 0);
552 } else {
553 retval = ext4_ind_map_blocks(handle, inode, map, 0);
554 }
555 if (retval > 0) {
556 unsigned int status;
557
558 if (unlikely(retval != map->m_len)) {
559 ext4_warning(inode->i_sb,
560 "ES len assertion failed for inode "
561 "%lu: retval %d != map->m_len %d",
562 inode->i_ino, retval, map->m_len);
563 WARN_ON(1);
564 }
565
566 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
567 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
568 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
569 !(status & EXTENT_STATUS_WRITTEN) &&
570 ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk,
571 map->m_lblk + map->m_len - 1))
572 status |= EXTENT_STATUS_DELAYED;
573 ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
574 map->m_pblk, status);
575 }
576 up_read((&EXT4_I(inode)->i_data_sem));
577
578 found:
579 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
580 ret = check_block_validity(inode, map);
581 if (ret != 0)
582 return ret;
583 }
584
585 /* If it is only a block(s) look up */
586 if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
587 return retval;
588
589 /*
590 * Returns if the blocks have already allocated
591 *
592 * Note that if blocks have been preallocated
593 * ext4_ext_map_blocks() returns with buffer head unmapped
594 */
595 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
596 /*
597 * If we need to convert extent to unwritten
598 * we continue and do the actual work in
599 * ext4_ext_map_blocks()
600 */
601 if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
602 return retval;
603
604 /*
605 * Here we clear m_flags because after allocating an new extent,
606 * it will be set again.
607 */
608 map->m_flags &= ~EXT4_MAP_FLAGS;
609
610 /*
611 * New blocks allocate and/or writing to unwritten extent
612 * will possibly result in updating i_data, so we take
613 * the write lock of i_data_sem, and call get_block()
614 * with create == 1 flag.
615 */
616 down_write(&EXT4_I(inode)->i_data_sem);
617
618 /*
619 * We need to check for EXT4 here because migrate
620 * could have changed the inode type in between
621 */
622 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
623 retval = ext4_ext_map_blocks(handle, inode, map, flags);
624 } else {
625 retval = ext4_ind_map_blocks(handle, inode, map, flags);
626
627 if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
628 /*
629 * We allocated new blocks which will result in
630 * i_data's format changing. Force the migrate
631 * to fail by clearing migrate flags
632 */
633 ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
634 }
635 }
636
637 if (retval > 0) {
638 unsigned int status;
639
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);
646 }
647
648 /*
649 * We have to zeroout blocks before inserting them into extent
650 * status tree. Otherwise someone could look them up there and
651 * use them before they are really zeroed. We also have to
652 * unmap metadata before zeroing as otherwise writeback can
653 * overwrite zeros with stale data from block device.
654 */
655 if (flags & EXT4_GET_BLOCKS_ZERO &&
656 map->m_flags & EXT4_MAP_MAPPED &&
657 map->m_flags & EXT4_MAP_NEW) {
658 ret = ext4_issue_zeroout(inode, map->m_lblk,
659 map->m_pblk, map->m_len);
660 if (ret) {
661 retval = ret;
662 goto out_sem;
663 }
664 }
665
666 /*
667 * If the extent has been zeroed out, we don't need to update
668 * extent status tree.
669 */
670 if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
671 ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) {
672 if (ext4_es_is_written(&es))
673 goto out_sem;
674 }
675 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
676 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
677 if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
678 !(status & EXTENT_STATUS_WRITTEN) &&
679 ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk,
680 map->m_lblk + map->m_len - 1))
681 status |= EXTENT_STATUS_DELAYED;
682 ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
683 map->m_pblk, status);
684 }
685
686 out_sem:
687 up_write((&EXT4_I(inode)->i_data_sem));
688 if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
689 ret = check_block_validity(inode, map);
690 if (ret != 0)
691 return ret;
692
693 /*
694 * Inodes with freshly allocated blocks where contents will be
695 * visible after transaction commit must be on transaction's
696 * ordered data list.
697 */
698 if (map->m_flags & EXT4_MAP_NEW &&
699 !(map->m_flags & EXT4_MAP_UNWRITTEN) &&
700 !(flags & EXT4_GET_BLOCKS_ZERO) &&
701 !ext4_is_quota_file(inode) &&
702 ext4_should_order_data(inode)) {
703 loff_t start_byte =
704 (loff_t)map->m_lblk << inode->i_blkbits;
705 loff_t length = (loff_t)map->m_len << inode->i_blkbits;
706
707 if (flags & EXT4_GET_BLOCKS_IO_SUBMIT)
708 ret = ext4_jbd2_inode_add_wait(handle, inode,
709 start_byte, length);
710 else
711 ret = ext4_jbd2_inode_add_write(handle, inode,
712 start_byte, length);
713 if (ret)
714 return ret;
715 }
716 }
717 if (retval > 0 && (map->m_flags & EXT4_MAP_UNWRITTEN ||
718 map->m_flags & EXT4_MAP_MAPPED))
719 ext4_fc_track_range(handle, inode, map->m_lblk,
720 map->m_lblk + map->m_len - 1);
721 if (retval < 0)
722 ext_debug(inode, "failed with err %d\n", retval);
723 return retval;
724 }
725
726 /*
727 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
728 * we have to be careful as someone else may be manipulating b_state as well.
729 */
ext4_update_bh_state(struct buffer_head * bh,unsigned long flags)730 static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags)
731 {
732 unsigned long old_state;
733 unsigned long new_state;
734
735 flags &= EXT4_MAP_FLAGS;
736
737 /* Dummy buffer_head? Set non-atomically. */
738 if (!bh->b_page) {
739 bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags;
740 return;
741 }
742 /*
743 * Someone else may be modifying b_state. Be careful! This is ugly but
744 * once we get rid of using bh as a container for mapping information
745 * to pass to / from get_block functions, this can go away.
746 */
747 old_state = READ_ONCE(bh->b_state);
748 do {
749 new_state = (old_state & ~EXT4_MAP_FLAGS) | flags;
750 } while (unlikely(!try_cmpxchg(&bh->b_state, &old_state, new_state)));
751 }
752
_ext4_get_block(struct inode * inode,sector_t iblock,struct buffer_head * bh,int flags)753 static int _ext4_get_block(struct inode *inode, sector_t iblock,
754 struct buffer_head *bh, int flags)
755 {
756 struct ext4_map_blocks map;
757 int ret = 0;
758
759 if (ext4_has_inline_data(inode))
760 return -ERANGE;
761
762 map.m_lblk = iblock;
763 map.m_len = bh->b_size >> inode->i_blkbits;
764
765 ret = ext4_map_blocks(ext4_journal_current_handle(), inode, &map,
766 flags);
767 if (ret > 0) {
768 map_bh(bh, inode->i_sb, map.m_pblk);
769 ext4_update_bh_state(bh, map.m_flags);
770 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
771 ret = 0;
772 } else if (ret == 0) {
773 /* hole case, need to fill in bh->b_size */
774 bh->b_size = inode->i_sb->s_blocksize * map.m_len;
775 }
776 return ret;
777 }
778
ext4_get_block(struct inode * inode,sector_t iblock,struct buffer_head * bh,int create)779 int ext4_get_block(struct inode *inode, sector_t iblock,
780 struct buffer_head *bh, int create)
781 {
782 return _ext4_get_block(inode, iblock, bh,
783 create ? EXT4_GET_BLOCKS_CREATE : 0);
784 }
785
786 /*
787 * Get block function used when preparing for buffered write if we require
788 * creating an unwritten extent if blocks haven't been allocated. The extent
789 * will be converted to written after the IO is complete.
790 */
ext4_get_block_unwritten(struct inode * inode,sector_t iblock,struct buffer_head * bh_result,int create)791 int ext4_get_block_unwritten(struct inode *inode, sector_t iblock,
792 struct buffer_head *bh_result, int create)
793 {
794 int ret = 0;
795
796 ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n",
797 inode->i_ino, create);
798 ret = _ext4_get_block(inode, iblock, bh_result,
799 EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT);
800
801 /*
802 * If the buffer is marked unwritten, mark it as new to make sure it is
803 * zeroed out correctly in case of partial writes. Otherwise, there is
804 * a chance of stale data getting exposed.
805 */
806 if (ret == 0 && buffer_unwritten(bh_result))
807 set_buffer_new(bh_result);
808
809 return ret;
810 }
811
812 /* Maximum number of blocks we map for direct IO at once. */
813 #define DIO_MAX_BLOCKS 4096
814
815 /*
816 * `handle' can be NULL if create is zero
817 */
ext4_getblk(handle_t * handle,struct inode * inode,ext4_lblk_t block,int map_flags)818 struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
819 ext4_lblk_t block, int map_flags)
820 {
821 struct ext4_map_blocks map;
822 struct buffer_head *bh;
823 int create = map_flags & EXT4_GET_BLOCKS_CREATE;
824 bool nowait = map_flags & EXT4_GET_BLOCKS_CACHED_NOWAIT;
825 int err;
826
827 ASSERT((EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
828 || handle != NULL || create == 0);
829 ASSERT(create == 0 || !nowait);
830
831 map.m_lblk = block;
832 map.m_len = 1;
833 err = ext4_map_blocks(handle, inode, &map, map_flags);
834
835 if (err == 0)
836 return create ? ERR_PTR(-ENOSPC) : NULL;
837 if (err < 0)
838 return ERR_PTR(err);
839
840 if (nowait)
841 return sb_find_get_block(inode->i_sb, map.m_pblk);
842
843 bh = sb_getblk(inode->i_sb, map.m_pblk);
844 if (unlikely(!bh))
845 return ERR_PTR(-ENOMEM);
846 if (map.m_flags & EXT4_MAP_NEW) {
847 ASSERT(create != 0);
848 ASSERT((EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
849 || (handle != NULL));
850
851 /*
852 * Now that we do not always journal data, we should
853 * keep in mind whether this should always journal the
854 * new buffer as metadata. For now, regular file
855 * writes use ext4_get_block instead, so it's not a
856 * problem.
857 */
858 lock_buffer(bh);
859 BUFFER_TRACE(bh, "call get_create_access");
860 err = ext4_journal_get_create_access(handle, inode->i_sb, bh,
861 EXT4_JTR_NONE);
862 if (unlikely(err)) {
863 unlock_buffer(bh);
864 goto errout;
865 }
866 if (!buffer_uptodate(bh)) {
867 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
868 set_buffer_uptodate(bh);
869 }
870 unlock_buffer(bh);
871 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
872 err = ext4_handle_dirty_metadata(handle, inode, bh);
873 if (unlikely(err))
874 goto errout;
875 } else
876 BUFFER_TRACE(bh, "not a new buffer");
877 return bh;
878 errout:
879 brelse(bh);
880 return ERR_PTR(err);
881 }
882
ext4_bread(handle_t * handle,struct inode * inode,ext4_lblk_t block,int map_flags)883 struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
884 ext4_lblk_t block, int map_flags)
885 {
886 struct buffer_head *bh;
887 int ret;
888
889 bh = ext4_getblk(handle, inode, block, map_flags);
890 if (IS_ERR(bh))
891 return bh;
892 if (!bh || ext4_buffer_uptodate(bh))
893 return bh;
894
895 ret = ext4_read_bh_lock(bh, REQ_META | REQ_PRIO, true);
896 if (ret) {
897 put_bh(bh);
898 return ERR_PTR(ret);
899 }
900 return bh;
901 }
902
903 /* Read a contiguous batch of blocks. */
ext4_bread_batch(struct inode * inode,ext4_lblk_t block,int bh_count,bool wait,struct buffer_head ** bhs)904 int ext4_bread_batch(struct inode *inode, ext4_lblk_t block, int bh_count,
905 bool wait, struct buffer_head **bhs)
906 {
907 int i, err;
908
909 for (i = 0; i < bh_count; i++) {
910 bhs[i] = ext4_getblk(NULL, inode, block + i, 0 /* map_flags */);
911 if (IS_ERR(bhs[i])) {
912 err = PTR_ERR(bhs[i]);
913 bh_count = i;
914 goto out_brelse;
915 }
916 }
917
918 for (i = 0; i < bh_count; i++)
919 /* Note that NULL bhs[i] is valid because of holes. */
920 if (bhs[i] && !ext4_buffer_uptodate(bhs[i]))
921 ext4_read_bh_lock(bhs[i], REQ_META | REQ_PRIO, false);
922
923 if (!wait)
924 return 0;
925
926 for (i = 0; i < bh_count; i++)
927 if (bhs[i])
928 wait_on_buffer(bhs[i]);
929
930 for (i = 0; i < bh_count; i++) {
931 if (bhs[i] && !buffer_uptodate(bhs[i])) {
932 err = -EIO;
933 goto out_brelse;
934 }
935 }
936 return 0;
937
938 out_brelse:
939 for (i = 0; i < bh_count; i++) {
940 brelse(bhs[i]);
941 bhs[i] = NULL;
942 }
943 return err;
944 }
945
ext4_walk_page_buffers(handle_t * handle,struct inode * inode,struct buffer_head * head,unsigned from,unsigned to,int * partial,int (* fn)(handle_t * handle,struct inode * inode,struct buffer_head * bh))946 int ext4_walk_page_buffers(handle_t *handle, struct inode *inode,
947 struct buffer_head *head,
948 unsigned from,
949 unsigned to,
950 int *partial,
951 int (*fn)(handle_t *handle, struct inode *inode,
952 struct buffer_head *bh))
953 {
954 struct buffer_head *bh;
955 unsigned block_start, block_end;
956 unsigned blocksize = head->b_size;
957 int err, ret = 0;
958 struct buffer_head *next;
959
960 for (bh = head, block_start = 0;
961 ret == 0 && (bh != head || !block_start);
962 block_start = block_end, bh = next) {
963 next = bh->b_this_page;
964 block_end = block_start + blocksize;
965 if (block_end <= from || block_start >= to) {
966 if (partial && !buffer_uptodate(bh))
967 *partial = 1;
968 continue;
969 }
970 err = (*fn)(handle, inode, bh);
971 if (!ret)
972 ret = err;
973 }
974 return ret;
975 }
976
977 /*
978 * Helper for handling dirtying of journalled data. We also mark the folio as
979 * dirty so that writeback code knows about this page (and inode) contains
980 * dirty data. ext4_writepages() then commits appropriate transaction to
981 * make data stable.
982 */
ext4_dirty_journalled_data(handle_t * handle,struct buffer_head * bh)983 static int ext4_dirty_journalled_data(handle_t *handle, struct buffer_head *bh)
984 {
985 folio_mark_dirty(bh->b_folio);
986 return ext4_handle_dirty_metadata(handle, NULL, bh);
987 }
988
do_journal_get_write_access(handle_t * handle,struct inode * inode,struct buffer_head * bh)989 int do_journal_get_write_access(handle_t *handle, struct inode *inode,
990 struct buffer_head *bh)
991 {
992 int dirty = buffer_dirty(bh);
993 int ret;
994
995 if (!buffer_mapped(bh) || buffer_freed(bh))
996 return 0;
997 /*
998 * __block_write_begin() could have dirtied some buffers. Clean
999 * the dirty bit as jbd2_journal_get_write_access() could complain
1000 * otherwise about fs integrity issues. Setting of the dirty bit
1001 * by __block_write_begin() isn't a real problem here as we clear
1002 * the bit before releasing a page lock and thus writeback cannot
1003 * ever write the buffer.
1004 */
1005 if (dirty)
1006 clear_buffer_dirty(bh);
1007 BUFFER_TRACE(bh, "get write access");
1008 ret = ext4_journal_get_write_access(handle, inode->i_sb, bh,
1009 EXT4_JTR_NONE);
1010 if (!ret && dirty)
1011 ret = ext4_dirty_journalled_data(handle, bh);
1012 return ret;
1013 }
1014
1015 #ifdef CONFIG_FS_ENCRYPTION
ext4_block_write_begin(struct folio * folio,loff_t pos,unsigned len,get_block_t * get_block)1016 static int ext4_block_write_begin(struct folio *folio, loff_t pos, unsigned len,
1017 get_block_t *get_block)
1018 {
1019 unsigned from = pos & (PAGE_SIZE - 1);
1020 unsigned to = from + len;
1021 struct inode *inode = folio->mapping->host;
1022 unsigned block_start, block_end;
1023 sector_t block;
1024 int err = 0;
1025 unsigned blocksize = inode->i_sb->s_blocksize;
1026 unsigned bbits;
1027 struct buffer_head *bh, *head, *wait[2];
1028 int nr_wait = 0;
1029 int i;
1030
1031 BUG_ON(!folio_test_locked(folio));
1032 BUG_ON(from > PAGE_SIZE);
1033 BUG_ON(to > PAGE_SIZE);
1034 BUG_ON(from > to);
1035
1036 head = folio_buffers(folio);
1037 if (!head)
1038 head = create_empty_buffers(folio, blocksize, 0);
1039 bbits = ilog2(blocksize);
1040 block = (sector_t)folio->index << (PAGE_SHIFT - bbits);
1041
1042 for (bh = head, block_start = 0; bh != head || !block_start;
1043 block++, block_start = block_end, bh = bh->b_this_page) {
1044 block_end = block_start + blocksize;
1045 if (block_end <= from || block_start >= to) {
1046 if (folio_test_uptodate(folio)) {
1047 set_buffer_uptodate(bh);
1048 }
1049 continue;
1050 }
1051 if (buffer_new(bh))
1052 clear_buffer_new(bh);
1053 if (!buffer_mapped(bh)) {
1054 WARN_ON(bh->b_size != blocksize);
1055 err = get_block(inode, block, bh, 1);
1056 if (err)
1057 break;
1058 if (buffer_new(bh)) {
1059 if (folio_test_uptodate(folio)) {
1060 clear_buffer_new(bh);
1061 set_buffer_uptodate(bh);
1062 mark_buffer_dirty(bh);
1063 continue;
1064 }
1065 if (block_end > to || block_start < from)
1066 folio_zero_segments(folio, to,
1067 block_end,
1068 block_start, from);
1069 continue;
1070 }
1071 }
1072 if (folio_test_uptodate(folio)) {
1073 set_buffer_uptodate(bh);
1074 continue;
1075 }
1076 if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
1077 !buffer_unwritten(bh) &&
1078 (block_start < from || block_end > to)) {
1079 ext4_read_bh_lock(bh, 0, false);
1080 wait[nr_wait++] = bh;
1081 }
1082 }
1083 /*
1084 * If we issued read requests, let them complete.
1085 */
1086 for (i = 0; i < nr_wait; i++) {
1087 wait_on_buffer(wait[i]);
1088 if (!buffer_uptodate(wait[i]))
1089 err = -EIO;
1090 }
1091 if (unlikely(err)) {
1092 folio_zero_new_buffers(folio, from, to);
1093 } else if (fscrypt_inode_uses_fs_layer_crypto(inode)) {
1094 for (i = 0; i < nr_wait; i++) {
1095 int err2;
1096
1097 err2 = fscrypt_decrypt_pagecache_blocks(folio,
1098 blocksize, bh_offset(wait[i]));
1099 if (err2) {
1100 clear_buffer_uptodate(wait[i]);
1101 err = err2;
1102 }
1103 }
1104 }
1105
1106 return err;
1107 }
1108 #endif
1109
1110 /*
1111 * To preserve ordering, it is essential that the hole instantiation and
1112 * the data write be encapsulated in a single transaction. We cannot
1113 * close off a transaction and start a new one between the ext4_get_block()
1114 * and the ext4_write_end(). So doing the jbd2_journal_start at the start of
1115 * ext4_write_begin() is the right place.
1116 */
ext4_write_begin(struct file * file,struct address_space * mapping,loff_t pos,unsigned len,struct page ** pagep,void ** fsdata)1117 static int ext4_write_begin(struct file *file, struct address_space *mapping,
1118 loff_t pos, unsigned len,
1119 struct page **pagep, void **fsdata)
1120 {
1121 struct inode *inode = mapping->host;
1122 int ret, needed_blocks;
1123 handle_t *handle;
1124 int retries = 0;
1125 struct folio *folio;
1126 pgoff_t index;
1127 unsigned from, to;
1128
1129 if (unlikely(ext4_forced_shutdown(inode->i_sb)))
1130 return -EIO;
1131
1132 trace_ext4_write_begin(inode, pos, len);
1133 /*
1134 * Reserve one block more for addition to orphan list in case
1135 * we allocate blocks but write fails for some reason
1136 */
1137 needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1138 index = pos >> PAGE_SHIFT;
1139 from = pos & (PAGE_SIZE - 1);
1140 to = from + len;
1141
1142 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1143 ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1144 pagep);
1145 if (ret < 0)
1146 return ret;
1147 if (ret == 1)
1148 return 0;
1149 }
1150
1151 /*
1152 * __filemap_get_folio() can take a long time if the
1153 * system is thrashing due to memory pressure, or if the folio
1154 * is being written back. So grab it first before we start
1155 * the transaction handle. This also allows us to allocate
1156 * the folio (if needed) without using GFP_NOFS.
1157 */
1158 retry_grab:
1159 folio = __filemap_get_folio(mapping, index, FGP_WRITEBEGIN,
1160 mapping_gfp_mask(mapping));
1161 if (IS_ERR(folio))
1162 return PTR_ERR(folio);
1163 /*
1164 * The same as page allocation, we prealloc buffer heads before
1165 * starting the handle.
1166 */
1167 if (!folio_buffers(folio))
1168 create_empty_buffers(folio, inode->i_sb->s_blocksize, 0);
1169
1170 folio_unlock(folio);
1171
1172 retry_journal:
1173 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1174 if (IS_ERR(handle)) {
1175 folio_put(folio);
1176 return PTR_ERR(handle);
1177 }
1178
1179 folio_lock(folio);
1180 if (folio->mapping != mapping) {
1181 /* The folio got truncated from under us */
1182 folio_unlock(folio);
1183 folio_put(folio);
1184 ext4_journal_stop(handle);
1185 goto retry_grab;
1186 }
1187 /* In case writeback began while the folio was unlocked */
1188 folio_wait_stable(folio);
1189
1190 #ifdef CONFIG_FS_ENCRYPTION
1191 if (ext4_should_dioread_nolock(inode))
1192 ret = ext4_block_write_begin(folio, pos, len,
1193 ext4_get_block_unwritten);
1194 else
1195 ret = ext4_block_write_begin(folio, pos, len, ext4_get_block);
1196 #else
1197 if (ext4_should_dioread_nolock(inode))
1198 ret = __block_write_begin(&folio->page, pos, len,
1199 ext4_get_block_unwritten);
1200 else
1201 ret = __block_write_begin(&folio->page, pos, len, ext4_get_block);
1202 #endif
1203 if (!ret && ext4_should_journal_data(inode)) {
1204 ret = ext4_walk_page_buffers(handle, inode,
1205 folio_buffers(folio), from, to,
1206 NULL, do_journal_get_write_access);
1207 }
1208
1209 if (ret) {
1210 bool extended = (pos + len > inode->i_size) &&
1211 !ext4_verity_in_progress(inode);
1212
1213 folio_unlock(folio);
1214 /*
1215 * __block_write_begin may have instantiated a few blocks
1216 * outside i_size. Trim these off again. Don't need
1217 * i_size_read because we hold i_rwsem.
1218 *
1219 * Add inode to orphan list in case we crash before
1220 * truncate finishes
1221 */
1222 if (extended && ext4_can_truncate(inode))
1223 ext4_orphan_add(handle, inode);
1224
1225 ext4_journal_stop(handle);
1226 if (extended) {
1227 ext4_truncate_failed_write(inode);
1228 /*
1229 * If truncate failed early the inode might
1230 * still be on the orphan list; we need to
1231 * make sure the inode is removed from the
1232 * orphan list in that case.
1233 */
1234 if (inode->i_nlink)
1235 ext4_orphan_del(NULL, inode);
1236 }
1237
1238 if (ret == -ENOSPC &&
1239 ext4_should_retry_alloc(inode->i_sb, &retries))
1240 goto retry_journal;
1241 folio_put(folio);
1242 return ret;
1243 }
1244 *pagep = &folio->page;
1245 return ret;
1246 }
1247
1248 /* For write_end() in data=journal mode */
write_end_fn(handle_t * handle,struct inode * inode,struct buffer_head * bh)1249 static int write_end_fn(handle_t *handle, struct inode *inode,
1250 struct buffer_head *bh)
1251 {
1252 int ret;
1253 if (!buffer_mapped(bh) || buffer_freed(bh))
1254 return 0;
1255 set_buffer_uptodate(bh);
1256 ret = ext4_dirty_journalled_data(handle, bh);
1257 clear_buffer_meta(bh);
1258 clear_buffer_prio(bh);
1259 return ret;
1260 }
1261
1262 /*
1263 * We need to pick up the new inode size which generic_commit_write gave us
1264 * `file' can be NULL - eg, when called from page_symlink().
1265 *
1266 * ext4 never places buffers on inode->i_mapping->i_private_list. metadata
1267 * buffers are managed internally.
1268 */
ext4_write_end(struct file * file,struct address_space * mapping,loff_t pos,unsigned len,unsigned copied,struct page * page,void * fsdata)1269 static int ext4_write_end(struct file *file,
1270 struct address_space *mapping,
1271 loff_t pos, unsigned len, unsigned copied,
1272 struct page *page, void *fsdata)
1273 {
1274 struct folio *folio = page_folio(page);
1275 handle_t *handle = ext4_journal_current_handle();
1276 struct inode *inode = mapping->host;
1277 loff_t old_size = inode->i_size;
1278 int ret = 0, ret2;
1279 int i_size_changed = 0;
1280 bool verity = ext4_verity_in_progress(inode);
1281
1282 trace_ext4_write_end(inode, pos, len, copied);
1283
1284 if (ext4_has_inline_data(inode) &&
1285 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA))
1286 return ext4_write_inline_data_end(inode, pos, len, copied,
1287 folio);
1288
1289 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
1290 /*
1291 * it's important to update i_size while still holding folio lock:
1292 * page writeout could otherwise come in and zero beyond i_size.
1293 *
1294 * If FS_IOC_ENABLE_VERITY is running on this inode, then Merkle tree
1295 * blocks are being written past EOF, so skip the i_size update.
1296 */
1297 if (!verity)
1298 i_size_changed = ext4_update_inode_size(inode, pos + copied);
1299 folio_unlock(folio);
1300 folio_put(folio);
1301
1302 if (old_size < pos && !verity)
1303 pagecache_isize_extended(inode, old_size, pos);
1304 /*
1305 * Don't mark the inode dirty under folio lock. First, it unnecessarily
1306 * makes the holding time of folio lock longer. Second, it forces lock
1307 * ordering of folio lock and transaction start for journaling
1308 * filesystems.
1309 */
1310 if (i_size_changed)
1311 ret = ext4_mark_inode_dirty(handle, inode);
1312
1313 if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode))
1314 /* if we have allocated more blocks and copied
1315 * less. We will have blocks allocated outside
1316 * inode->i_size. So truncate them
1317 */
1318 ext4_orphan_add(handle, inode);
1319
1320 ret2 = ext4_journal_stop(handle);
1321 if (!ret)
1322 ret = ret2;
1323
1324 if (pos + len > inode->i_size && !verity) {
1325 ext4_truncate_failed_write(inode);
1326 /*
1327 * If truncate failed early the inode might still be
1328 * on the orphan list; we need to make sure the inode
1329 * is removed from the orphan list in that case.
1330 */
1331 if (inode->i_nlink)
1332 ext4_orphan_del(NULL, inode);
1333 }
1334
1335 return ret ? ret : copied;
1336 }
1337
1338 /*
1339 * This is a private version of folio_zero_new_buffers() which doesn't
1340 * set the buffer to be dirty, since in data=journalled mode we need
1341 * to call ext4_dirty_journalled_data() instead.
1342 */
ext4_journalled_zero_new_buffers(handle_t * handle,struct inode * inode,struct folio * folio,unsigned from,unsigned to)1343 static void ext4_journalled_zero_new_buffers(handle_t *handle,
1344 struct inode *inode,
1345 struct folio *folio,
1346 unsigned from, unsigned to)
1347 {
1348 unsigned int block_start = 0, block_end;
1349 struct buffer_head *head, *bh;
1350
1351 bh = head = folio_buffers(folio);
1352 do {
1353 block_end = block_start + bh->b_size;
1354 if (buffer_new(bh)) {
1355 if (block_end > from && block_start < to) {
1356 if (!folio_test_uptodate(folio)) {
1357 unsigned start, size;
1358
1359 start = max(from, block_start);
1360 size = min(to, block_end) - start;
1361
1362 folio_zero_range(folio, start, size);
1363 write_end_fn(handle, inode, bh);
1364 }
1365 clear_buffer_new(bh);
1366 }
1367 }
1368 block_start = block_end;
1369 bh = bh->b_this_page;
1370 } while (bh != head);
1371 }
1372
ext4_journalled_write_end(struct file * file,struct address_space * mapping,loff_t pos,unsigned len,unsigned copied,struct page * page,void * fsdata)1373 static int ext4_journalled_write_end(struct file *file,
1374 struct address_space *mapping,
1375 loff_t pos, unsigned len, unsigned copied,
1376 struct page *page, void *fsdata)
1377 {
1378 struct folio *folio = page_folio(page);
1379 handle_t *handle = ext4_journal_current_handle();
1380 struct inode *inode = mapping->host;
1381 loff_t old_size = inode->i_size;
1382 int ret = 0, ret2;
1383 int partial = 0;
1384 unsigned from, to;
1385 int size_changed = 0;
1386 bool verity = ext4_verity_in_progress(inode);
1387
1388 trace_ext4_journalled_write_end(inode, pos, len, copied);
1389 from = pos & (PAGE_SIZE - 1);
1390 to = from + len;
1391
1392 BUG_ON(!ext4_handle_valid(handle));
1393
1394 if (ext4_has_inline_data(inode))
1395 return ext4_write_inline_data_end(inode, pos, len, copied,
1396 folio);
1397
1398 if (unlikely(copied < len) && !folio_test_uptodate(folio)) {
1399 copied = 0;
1400 ext4_journalled_zero_new_buffers(handle, inode, folio,
1401 from, to);
1402 } else {
1403 if (unlikely(copied < len))
1404 ext4_journalled_zero_new_buffers(handle, inode, folio,
1405 from + copied, to);
1406 ret = ext4_walk_page_buffers(handle, inode,
1407 folio_buffers(folio),
1408 from, from + copied, &partial,
1409 write_end_fn);
1410 if (!partial)
1411 folio_mark_uptodate(folio);
1412 }
1413 if (!verity)
1414 size_changed = ext4_update_inode_size(inode, pos + copied);
1415 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1416 folio_unlock(folio);
1417 folio_put(folio);
1418
1419 if (old_size < pos && !verity)
1420 pagecache_isize_extended(inode, old_size, pos);
1421
1422 if (size_changed) {
1423 ret2 = ext4_mark_inode_dirty(handle, inode);
1424 if (!ret)
1425 ret = ret2;
1426 }
1427
1428 if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode))
1429 /* if we have allocated more blocks and copied
1430 * less. We will have blocks allocated outside
1431 * inode->i_size. So truncate them
1432 */
1433 ext4_orphan_add(handle, inode);
1434
1435 ret2 = ext4_journal_stop(handle);
1436 if (!ret)
1437 ret = ret2;
1438 if (pos + len > inode->i_size && !verity) {
1439 ext4_truncate_failed_write(inode);
1440 /*
1441 * If truncate failed early the inode might still be
1442 * on the orphan list; we need to make sure the inode
1443 * is removed from the orphan list in that case.
1444 */
1445 if (inode->i_nlink)
1446 ext4_orphan_del(NULL, inode);
1447 }
1448
1449 return ret ? ret : copied;
1450 }
1451
1452 /*
1453 * Reserve space for a single cluster
1454 */
ext4_da_reserve_space(struct inode * inode)1455 static int ext4_da_reserve_space(struct inode *inode)
1456 {
1457 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1458 struct ext4_inode_info *ei = EXT4_I(inode);
1459 int ret;
1460
1461 /*
1462 * We will charge metadata quota at writeout time; this saves
1463 * us from metadata over-estimation, though we may go over by
1464 * a small amount in the end. Here we just reserve for data.
1465 */
1466 ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1467 if (ret)
1468 return ret;
1469
1470 spin_lock(&ei->i_block_reservation_lock);
1471 if (ext4_claim_free_clusters(sbi, 1, 0)) {
1472 spin_unlock(&ei->i_block_reservation_lock);
1473 dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1474 return -ENOSPC;
1475 }
1476 ei->i_reserved_data_blocks++;
1477 trace_ext4_da_reserve_space(inode);
1478 spin_unlock(&ei->i_block_reservation_lock);
1479
1480 return 0; /* success */
1481 }
1482
ext4_da_release_space(struct inode * inode,int to_free)1483 void ext4_da_release_space(struct inode *inode, int to_free)
1484 {
1485 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1486 struct ext4_inode_info *ei = EXT4_I(inode);
1487
1488 if (!to_free)
1489 return; /* Nothing to release, exit */
1490
1491 spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1492
1493 trace_ext4_da_release_space(inode, to_free);
1494 if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1495 /*
1496 * if there aren't enough reserved blocks, then the
1497 * counter is messed up somewhere. Since this
1498 * function is called from invalidate page, it's
1499 * harmless to return without any action.
1500 */
1501 ext4_warning(inode->i_sb, "ext4_da_release_space: "
1502 "ino %lu, to_free %d with only %d reserved "
1503 "data blocks", inode->i_ino, to_free,
1504 ei->i_reserved_data_blocks);
1505 WARN_ON(1);
1506 to_free = ei->i_reserved_data_blocks;
1507 }
1508 ei->i_reserved_data_blocks -= to_free;
1509
1510 /* update fs dirty data blocks counter */
1511 percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1512
1513 spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1514
1515 dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1516 }
1517
1518 /*
1519 * Delayed allocation stuff
1520 */
1521
1522 struct mpage_da_data {
1523 /* These are input fields for ext4_do_writepages() */
1524 struct inode *inode;
1525 struct writeback_control *wbc;
1526 unsigned int can_map:1; /* Can writepages call map blocks? */
1527
1528 /* These are internal state of ext4_do_writepages() */
1529 pgoff_t first_page; /* The first page to write */
1530 pgoff_t next_page; /* Current page to examine */
1531 pgoff_t last_page; /* Last page to examine */
1532 /*
1533 * Extent to map - this can be after first_page because that can be
1534 * fully mapped. We somewhat abuse m_flags to store whether the extent
1535 * is delalloc or unwritten.
1536 */
1537 struct ext4_map_blocks map;
1538 struct ext4_io_submit io_submit; /* IO submission data */
1539 unsigned int do_map:1;
1540 unsigned int scanned_until_end:1;
1541 unsigned int journalled_more_data:1;
1542 };
1543
mpage_release_unused_pages(struct mpage_da_data * mpd,bool invalidate)1544 static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1545 bool invalidate)
1546 {
1547 unsigned nr, i;
1548 pgoff_t index, end;
1549 struct folio_batch fbatch;
1550 struct inode *inode = mpd->inode;
1551 struct address_space *mapping = inode->i_mapping;
1552
1553 /* This is necessary when next_page == 0. */
1554 if (mpd->first_page >= mpd->next_page)
1555 return;
1556
1557 mpd->scanned_until_end = 0;
1558 index = mpd->first_page;
1559 end = mpd->next_page - 1;
1560 if (invalidate) {
1561 ext4_lblk_t start, last;
1562 start = index << (PAGE_SHIFT - inode->i_blkbits);
1563 last = end << (PAGE_SHIFT - inode->i_blkbits);
1564
1565 /*
1566 * avoid racing with extent status tree scans made by
1567 * ext4_insert_delayed_block()
1568 */
1569 down_write(&EXT4_I(inode)->i_data_sem);
1570 ext4_es_remove_extent(inode, start, last - start + 1);
1571 up_write(&EXT4_I(inode)->i_data_sem);
1572 }
1573
1574 folio_batch_init(&fbatch);
1575 while (index <= end) {
1576 nr = filemap_get_folios(mapping, &index, end, &fbatch);
1577 if (nr == 0)
1578 break;
1579 for (i = 0; i < nr; i++) {
1580 struct folio *folio = fbatch.folios[i];
1581
1582 if (folio->index < mpd->first_page)
1583 continue;
1584 if (folio_next_index(folio) - 1 > end)
1585 continue;
1586 BUG_ON(!folio_test_locked(folio));
1587 BUG_ON(folio_test_writeback(folio));
1588 if (invalidate) {
1589 if (folio_mapped(folio))
1590 folio_clear_dirty_for_io(folio);
1591 block_invalidate_folio(folio, 0,
1592 folio_size(folio));
1593 folio_clear_uptodate(folio);
1594 }
1595 folio_unlock(folio);
1596 }
1597 folio_batch_release(&fbatch);
1598 }
1599 }
1600
ext4_print_free_blocks(struct inode * inode)1601 static void ext4_print_free_blocks(struct inode *inode)
1602 {
1603 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1604 struct super_block *sb = inode->i_sb;
1605 struct ext4_inode_info *ei = EXT4_I(inode);
1606
1607 ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1608 EXT4_C2B(EXT4_SB(inode->i_sb),
1609 ext4_count_free_clusters(sb)));
1610 ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1611 ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1612 (long long) EXT4_C2B(EXT4_SB(sb),
1613 percpu_counter_sum(&sbi->s_freeclusters_counter)));
1614 ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1615 (long long) EXT4_C2B(EXT4_SB(sb),
1616 percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1617 ext4_msg(sb, KERN_CRIT, "Block reservation details");
1618 ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1619 ei->i_reserved_data_blocks);
1620 return;
1621 }
1622
1623 /*
1624 * ext4_insert_delayed_block - adds a delayed block to the extents status
1625 * tree, incrementing the reserved cluster/block
1626 * count or making a pending reservation
1627 * where needed
1628 *
1629 * @inode - file containing the newly added block
1630 * @lblk - logical block to be added
1631 *
1632 * Returns 0 on success, negative error code on failure.
1633 */
ext4_insert_delayed_block(struct inode * inode,ext4_lblk_t lblk)1634 static int ext4_insert_delayed_block(struct inode *inode, ext4_lblk_t lblk)
1635 {
1636 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1637 int ret;
1638 bool allocated = false;
1639
1640 /*
1641 * If the cluster containing lblk is shared with a delayed,
1642 * written, or unwritten extent in a bigalloc file system, it's
1643 * already been accounted for and does not need to be reserved.
1644 * A pending reservation must be made for the cluster if it's
1645 * shared with a written or unwritten extent and doesn't already
1646 * have one. Written and unwritten extents can be purged from the
1647 * extents status tree if the system is under memory pressure, so
1648 * it's necessary to examine the extent tree if a search of the
1649 * extents status tree doesn't get a match.
1650 */
1651 if (sbi->s_cluster_ratio == 1) {
1652 ret = ext4_da_reserve_space(inode);
1653 if (ret != 0) /* ENOSPC */
1654 return ret;
1655 } else { /* bigalloc */
1656 if (!ext4_es_scan_clu(inode, &ext4_es_is_delonly, lblk)) {
1657 if (!ext4_es_scan_clu(inode,
1658 &ext4_es_is_mapped, lblk)) {
1659 ret = ext4_clu_mapped(inode,
1660 EXT4_B2C(sbi, lblk));
1661 if (ret < 0)
1662 return ret;
1663 if (ret == 0) {
1664 ret = ext4_da_reserve_space(inode);
1665 if (ret != 0) /* ENOSPC */
1666 return ret;
1667 } else {
1668 allocated = true;
1669 }
1670 } else {
1671 allocated = true;
1672 }
1673 }
1674 }
1675
1676 ext4_es_insert_delayed_block(inode, lblk, allocated);
1677 return 0;
1678 }
1679
1680 /*
1681 * This function is grabs code from the very beginning of
1682 * ext4_map_blocks, but assumes that the caller is from delayed write
1683 * time. This function looks up the requested blocks and sets the
1684 * buffer delay bit under the protection of i_data_sem.
1685 */
ext4_da_map_blocks(struct inode * inode,sector_t iblock,struct ext4_map_blocks * map,struct buffer_head * bh)1686 static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1687 struct ext4_map_blocks *map,
1688 struct buffer_head *bh)
1689 {
1690 struct extent_status es;
1691 int retval;
1692 sector_t invalid_block = ~((sector_t) 0xffff);
1693 #ifdef ES_AGGRESSIVE_TEST
1694 struct ext4_map_blocks orig_map;
1695
1696 memcpy(&orig_map, map, sizeof(*map));
1697 #endif
1698
1699 if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1700 invalid_block = ~0;
1701
1702 map->m_flags = 0;
1703 ext_debug(inode, "max_blocks %u, logical block %lu\n", map->m_len,
1704 (unsigned long) map->m_lblk);
1705
1706 /* Lookup extent status tree firstly */
1707 if (ext4_es_lookup_extent(inode, iblock, NULL, &es)) {
1708 if (ext4_es_is_hole(&es))
1709 goto add_delayed;
1710
1711 /*
1712 * Delayed extent could be allocated by fallocate.
1713 * So we need to check it.
1714 */
1715 if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1716 map_bh(bh, inode->i_sb, invalid_block);
1717 set_buffer_new(bh);
1718 set_buffer_delay(bh);
1719 return 0;
1720 }
1721
1722 map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1723 retval = es.es_len - (iblock - es.es_lblk);
1724 if (retval > map->m_len)
1725 retval = map->m_len;
1726 map->m_len = retval;
1727 if (ext4_es_is_written(&es))
1728 map->m_flags |= EXT4_MAP_MAPPED;
1729 else if (ext4_es_is_unwritten(&es))
1730 map->m_flags |= EXT4_MAP_UNWRITTEN;
1731 else
1732 BUG();
1733
1734 #ifdef ES_AGGRESSIVE_TEST
1735 ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1736 #endif
1737 return retval;
1738 }
1739
1740 /*
1741 * Try to see if we can get the block without requesting a new
1742 * file system block.
1743 */
1744 down_read(&EXT4_I(inode)->i_data_sem);
1745 if (ext4_has_inline_data(inode))
1746 retval = 0;
1747 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1748 retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1749 else
1750 retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1751 if (retval < 0) {
1752 up_read(&EXT4_I(inode)->i_data_sem);
1753 return retval;
1754 }
1755 if (retval > 0) {
1756 unsigned int status;
1757
1758 if (unlikely(retval != map->m_len)) {
1759 ext4_warning(inode->i_sb,
1760 "ES len assertion failed for inode "
1761 "%lu: retval %d != map->m_len %d",
1762 inode->i_ino, retval, map->m_len);
1763 WARN_ON(1);
1764 }
1765
1766 status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1767 EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1768 ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1769 map->m_pblk, status);
1770 up_read(&EXT4_I(inode)->i_data_sem);
1771 return retval;
1772 }
1773 up_read(&EXT4_I(inode)->i_data_sem);
1774
1775 add_delayed:
1776 down_write(&EXT4_I(inode)->i_data_sem);
1777 retval = ext4_insert_delayed_block(inode, map->m_lblk);
1778 up_write(&EXT4_I(inode)->i_data_sem);
1779 if (retval)
1780 return retval;
1781
1782 map_bh(bh, inode->i_sb, invalid_block);
1783 set_buffer_new(bh);
1784 set_buffer_delay(bh);
1785 return retval;
1786 }
1787
1788 /*
1789 * This is a special get_block_t callback which is used by
1790 * ext4_da_write_begin(). It will either return mapped block or
1791 * reserve space for a single block.
1792 *
1793 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1794 * We also have b_blocknr = -1 and b_bdev initialized properly
1795 *
1796 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1797 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1798 * initialized properly.
1799 */
ext4_da_get_block_prep(struct inode * inode,sector_t iblock,struct buffer_head * bh,int create)1800 int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1801 struct buffer_head *bh, int create)
1802 {
1803 struct ext4_map_blocks map;
1804 int ret = 0;
1805
1806 BUG_ON(create == 0);
1807 BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1808
1809 map.m_lblk = iblock;
1810 map.m_len = 1;
1811
1812 /*
1813 * first, we need to know whether the block is allocated already
1814 * preallocated blocks are unmapped but should treated
1815 * the same as allocated blocks.
1816 */
1817 ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1818 if (ret <= 0)
1819 return ret;
1820
1821 map_bh(bh, inode->i_sb, map.m_pblk);
1822 ext4_update_bh_state(bh, map.m_flags);
1823
1824 if (buffer_unwritten(bh)) {
1825 /* A delayed write to unwritten bh should be marked
1826 * new and mapped. Mapped ensures that we don't do
1827 * get_block multiple times when we write to the same
1828 * offset and new ensures that we do proper zero out
1829 * for partial write.
1830 */
1831 set_buffer_new(bh);
1832 set_buffer_mapped(bh);
1833 }
1834 return 0;
1835 }
1836
mpage_folio_done(struct mpage_da_data * mpd,struct folio * folio)1837 static void mpage_folio_done(struct mpage_da_data *mpd, struct folio *folio)
1838 {
1839 mpd->first_page += folio_nr_pages(folio);
1840 folio_unlock(folio);
1841 }
1842
mpage_submit_folio(struct mpage_da_data * mpd,struct folio * folio)1843 static int mpage_submit_folio(struct mpage_da_data *mpd, struct folio *folio)
1844 {
1845 size_t len;
1846 loff_t size;
1847 int err;
1848
1849 BUG_ON(folio->index != mpd->first_page);
1850 folio_clear_dirty_for_io(folio);
1851 /*
1852 * We have to be very careful here! Nothing protects writeback path
1853 * against i_size changes and the page can be writeably mapped into
1854 * page tables. So an application can be growing i_size and writing
1855 * data through mmap while writeback runs. folio_clear_dirty_for_io()
1856 * write-protects our page in page tables and the page cannot get
1857 * written to again until we release folio lock. So only after
1858 * folio_clear_dirty_for_io() we are safe to sample i_size for
1859 * ext4_bio_write_folio() to zero-out tail of the written page. We rely
1860 * on the barrier provided by folio_test_clear_dirty() in
1861 * folio_clear_dirty_for_io() to make sure i_size is really sampled only
1862 * after page tables are updated.
1863 */
1864 size = i_size_read(mpd->inode);
1865 len = folio_size(folio);
1866 if (folio_pos(folio) + len > size &&
1867 !ext4_verity_in_progress(mpd->inode))
1868 len = size & (len - 1);
1869 err = ext4_bio_write_folio(&mpd->io_submit, folio, len);
1870 if (!err)
1871 mpd->wbc->nr_to_write--;
1872
1873 return err;
1874 }
1875
1876 #define BH_FLAGS (BIT(BH_Unwritten) | BIT(BH_Delay))
1877
1878 /*
1879 * mballoc gives us at most this number of blocks...
1880 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
1881 * The rest of mballoc seems to handle chunks up to full group size.
1882 */
1883 #define MAX_WRITEPAGES_EXTENT_LEN 2048
1884
1885 /*
1886 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
1887 *
1888 * @mpd - extent of blocks
1889 * @lblk - logical number of the block in the file
1890 * @bh - buffer head we want to add to the extent
1891 *
1892 * The function is used to collect contig. blocks in the same state. If the
1893 * buffer doesn't require mapping for writeback and we haven't started the
1894 * extent of buffers to map yet, the function returns 'true' immediately - the
1895 * caller can write the buffer right away. Otherwise the function returns true
1896 * if the block has been added to the extent, false if the block couldn't be
1897 * added.
1898 */
mpage_add_bh_to_extent(struct mpage_da_data * mpd,ext4_lblk_t lblk,struct buffer_head * bh)1899 static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
1900 struct buffer_head *bh)
1901 {
1902 struct ext4_map_blocks *map = &mpd->map;
1903
1904 /* Buffer that doesn't need mapping for writeback? */
1905 if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
1906 (!buffer_delay(bh) && !buffer_unwritten(bh))) {
1907 /* So far no extent to map => we write the buffer right away */
1908 if (map->m_len == 0)
1909 return true;
1910 return false;
1911 }
1912
1913 /* First block in the extent? */
1914 if (map->m_len == 0) {
1915 /* We cannot map unless handle is started... */
1916 if (!mpd->do_map)
1917 return false;
1918 map->m_lblk = lblk;
1919 map->m_len = 1;
1920 map->m_flags = bh->b_state & BH_FLAGS;
1921 return true;
1922 }
1923
1924 /* Don't go larger than mballoc is willing to allocate */
1925 if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
1926 return false;
1927
1928 /* Can we merge the block to our big extent? */
1929 if (lblk == map->m_lblk + map->m_len &&
1930 (bh->b_state & BH_FLAGS) == map->m_flags) {
1931 map->m_len++;
1932 return true;
1933 }
1934 return false;
1935 }
1936
1937 /*
1938 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
1939 *
1940 * @mpd - extent of blocks for mapping
1941 * @head - the first buffer in the page
1942 * @bh - buffer we should start processing from
1943 * @lblk - logical number of the block in the file corresponding to @bh
1944 *
1945 * Walk through page buffers from @bh upto @head (exclusive) and either submit
1946 * the page for IO if all buffers in this page were mapped and there's no
1947 * accumulated extent of buffers to map or add buffers in the page to the
1948 * extent of buffers to map. The function returns 1 if the caller can continue
1949 * by processing the next page, 0 if it should stop adding buffers to the
1950 * extent to map because we cannot extend it anymore. It can also return value
1951 * < 0 in case of error during IO submission.
1952 */
mpage_process_page_bufs(struct mpage_da_data * mpd,struct buffer_head * head,struct buffer_head * bh,ext4_lblk_t lblk)1953 static int mpage_process_page_bufs(struct mpage_da_data *mpd,
1954 struct buffer_head *head,
1955 struct buffer_head *bh,
1956 ext4_lblk_t lblk)
1957 {
1958 struct inode *inode = mpd->inode;
1959 int err;
1960 ext4_lblk_t blocks = (i_size_read(inode) + i_blocksize(inode) - 1)
1961 >> inode->i_blkbits;
1962
1963 if (ext4_verity_in_progress(inode))
1964 blocks = EXT_MAX_BLOCKS;
1965
1966 do {
1967 BUG_ON(buffer_locked(bh));
1968
1969 if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
1970 /* Found extent to map? */
1971 if (mpd->map.m_len)
1972 return 0;
1973 /* Buffer needs mapping and handle is not started? */
1974 if (!mpd->do_map)
1975 return 0;
1976 /* Everything mapped so far and we hit EOF */
1977 break;
1978 }
1979 } while (lblk++, (bh = bh->b_this_page) != head);
1980 /* So far everything mapped? Submit the page for IO. */
1981 if (mpd->map.m_len == 0) {
1982 err = mpage_submit_folio(mpd, head->b_folio);
1983 if (err < 0)
1984 return err;
1985 mpage_folio_done(mpd, head->b_folio);
1986 }
1987 if (lblk >= blocks) {
1988 mpd->scanned_until_end = 1;
1989 return 0;
1990 }
1991 return 1;
1992 }
1993
1994 /*
1995 * mpage_process_folio - update folio buffers corresponding to changed extent
1996 * and may submit fully mapped page for IO
1997 * @mpd: description of extent to map, on return next extent to map
1998 * @folio: Contains these buffers.
1999 * @m_lblk: logical block mapping.
2000 * @m_pblk: corresponding physical mapping.
2001 * @map_bh: determines on return whether this page requires any further
2002 * mapping or not.
2003 *
2004 * Scan given folio buffers corresponding to changed extent and update buffer
2005 * state according to new extent state.
2006 * We map delalloc buffers to their physical location, clear unwritten bits.
2007 * If the given folio is not fully mapped, we update @mpd to the next extent in
2008 * the given folio that needs mapping & return @map_bh as true.
2009 */
mpage_process_folio(struct mpage_da_data * mpd,struct folio * folio,ext4_lblk_t * m_lblk,ext4_fsblk_t * m_pblk,bool * map_bh)2010 static int mpage_process_folio(struct mpage_da_data *mpd, struct folio *folio,
2011 ext4_lblk_t *m_lblk, ext4_fsblk_t *m_pblk,
2012 bool *map_bh)
2013 {
2014 struct buffer_head *head, *bh;
2015 ext4_io_end_t *io_end = mpd->io_submit.io_end;
2016 ext4_lblk_t lblk = *m_lblk;
2017 ext4_fsblk_t pblock = *m_pblk;
2018 int err = 0;
2019 int blkbits = mpd->inode->i_blkbits;
2020 ssize_t io_end_size = 0;
2021 struct ext4_io_end_vec *io_end_vec = ext4_last_io_end_vec(io_end);
2022
2023 bh = head = folio_buffers(folio);
2024 do {
2025 if (lblk < mpd->map.m_lblk)
2026 continue;
2027 if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2028 /*
2029 * Buffer after end of mapped extent.
2030 * Find next buffer in the folio to map.
2031 */
2032 mpd->map.m_len = 0;
2033 mpd->map.m_flags = 0;
2034 io_end_vec->size += io_end_size;
2035
2036 err = mpage_process_page_bufs(mpd, head, bh, lblk);
2037 if (err > 0)
2038 err = 0;
2039 if (!err && mpd->map.m_len && mpd->map.m_lblk > lblk) {
2040 io_end_vec = ext4_alloc_io_end_vec(io_end);
2041 if (IS_ERR(io_end_vec)) {
2042 err = PTR_ERR(io_end_vec);
2043 goto out;
2044 }
2045 io_end_vec->offset = (loff_t)mpd->map.m_lblk << blkbits;
2046 }
2047 *map_bh = true;
2048 goto out;
2049 }
2050 if (buffer_delay(bh)) {
2051 clear_buffer_delay(bh);
2052 bh->b_blocknr = pblock++;
2053 }
2054 clear_buffer_unwritten(bh);
2055 io_end_size += (1 << blkbits);
2056 } while (lblk++, (bh = bh->b_this_page) != head);
2057
2058 io_end_vec->size += io_end_size;
2059 *map_bh = false;
2060 out:
2061 *m_lblk = lblk;
2062 *m_pblk = pblock;
2063 return err;
2064 }
2065
2066 /*
2067 * mpage_map_buffers - update buffers corresponding to changed extent and
2068 * submit fully mapped pages for IO
2069 *
2070 * @mpd - description of extent to map, on return next extent to map
2071 *
2072 * Scan buffers corresponding to changed extent (we expect corresponding pages
2073 * to be already locked) and update buffer state according to new extent state.
2074 * We map delalloc buffers to their physical location, clear unwritten bits,
2075 * and mark buffers as uninit when we perform writes to unwritten extents
2076 * and do extent conversion after IO is finished. If the last page is not fully
2077 * mapped, we update @map to the next extent in the last page that needs
2078 * mapping. Otherwise we submit the page for IO.
2079 */
mpage_map_and_submit_buffers(struct mpage_da_data * mpd)2080 static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
2081 {
2082 struct folio_batch fbatch;
2083 unsigned nr, i;
2084 struct inode *inode = mpd->inode;
2085 int bpp_bits = PAGE_SHIFT - inode->i_blkbits;
2086 pgoff_t start, end;
2087 ext4_lblk_t lblk;
2088 ext4_fsblk_t pblock;
2089 int err;
2090 bool map_bh = false;
2091
2092 start = mpd->map.m_lblk >> bpp_bits;
2093 end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2094 lblk = start << bpp_bits;
2095 pblock = mpd->map.m_pblk;
2096
2097 folio_batch_init(&fbatch);
2098 while (start <= end) {
2099 nr = filemap_get_folios(inode->i_mapping, &start, end, &fbatch);
2100 if (nr == 0)
2101 break;
2102 for (i = 0; i < nr; i++) {
2103 struct folio *folio = fbatch.folios[i];
2104
2105 err = mpage_process_folio(mpd, folio, &lblk, &pblock,
2106 &map_bh);
2107 /*
2108 * If map_bh is true, means page may require further bh
2109 * mapping, or maybe the page was submitted for IO.
2110 * So we return to call further extent mapping.
2111 */
2112 if (err < 0 || map_bh)
2113 goto out;
2114 /* Page fully mapped - let IO run! */
2115 err = mpage_submit_folio(mpd, folio);
2116 if (err < 0)
2117 goto out;
2118 mpage_folio_done(mpd, folio);
2119 }
2120 folio_batch_release(&fbatch);
2121 }
2122 /* Extent fully mapped and matches with page boundary. We are done. */
2123 mpd->map.m_len = 0;
2124 mpd->map.m_flags = 0;
2125 return 0;
2126 out:
2127 folio_batch_release(&fbatch);
2128 return err;
2129 }
2130
mpage_map_one_extent(handle_t * handle,struct mpage_da_data * mpd)2131 static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2132 {
2133 struct inode *inode = mpd->inode;
2134 struct ext4_map_blocks *map = &mpd->map;
2135 int get_blocks_flags;
2136 int err, dioread_nolock;
2137
2138 trace_ext4_da_write_pages_extent(inode, map);
2139 /*
2140 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2141 * to convert an unwritten extent to be initialized (in the case
2142 * where we have written into one or more preallocated blocks). It is
2143 * possible that we're going to need more metadata blocks than
2144 * previously reserved. However we must not fail because we're in
2145 * writeback and there is nothing we can do about it so it might result
2146 * in data loss. So use reserved blocks to allocate metadata if
2147 * possible.
2148 *
2149 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2150 * the blocks in question are delalloc blocks. This indicates
2151 * that the blocks and quotas has already been checked when
2152 * the data was copied into the page cache.
2153 */
2154 get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2155 EXT4_GET_BLOCKS_METADATA_NOFAIL |
2156 EXT4_GET_BLOCKS_IO_SUBMIT;
2157 dioread_nolock = ext4_should_dioread_nolock(inode);
2158 if (dioread_nolock)
2159 get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2160 if (map->m_flags & BIT(BH_Delay))
2161 get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2162
2163 err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2164 if (err < 0)
2165 return err;
2166 if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2167 if (!mpd->io_submit.io_end->handle &&
2168 ext4_handle_valid(handle)) {
2169 mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2170 handle->h_rsv_handle = NULL;
2171 }
2172 ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2173 }
2174
2175 BUG_ON(map->m_len == 0);
2176 return 0;
2177 }
2178
2179 /*
2180 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2181 * mpd->len and submit pages underlying it for IO
2182 *
2183 * @handle - handle for journal operations
2184 * @mpd - extent to map
2185 * @give_up_on_write - we set this to true iff there is a fatal error and there
2186 * is no hope of writing the data. The caller should discard
2187 * dirty pages to avoid infinite loops.
2188 *
2189 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2190 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2191 * them to initialized or split the described range from larger unwritten
2192 * extent. Note that we need not map all the described range since allocation
2193 * can return less blocks or the range is covered by more unwritten extents. We
2194 * cannot map more because we are limited by reserved transaction credits. On
2195 * the other hand we always make sure that the last touched page is fully
2196 * mapped so that it can be written out (and thus forward progress is
2197 * guaranteed). After mapping we submit all mapped pages for IO.
2198 */
mpage_map_and_submit_extent(handle_t * handle,struct mpage_da_data * mpd,bool * give_up_on_write)2199 static int mpage_map_and_submit_extent(handle_t *handle,
2200 struct mpage_da_data *mpd,
2201 bool *give_up_on_write)
2202 {
2203 struct inode *inode = mpd->inode;
2204 struct ext4_map_blocks *map = &mpd->map;
2205 int err;
2206 loff_t disksize;
2207 int progress = 0;
2208 ext4_io_end_t *io_end = mpd->io_submit.io_end;
2209 struct ext4_io_end_vec *io_end_vec;
2210
2211 io_end_vec = ext4_alloc_io_end_vec(io_end);
2212 if (IS_ERR(io_end_vec))
2213 return PTR_ERR(io_end_vec);
2214 io_end_vec->offset = ((loff_t)map->m_lblk) << inode->i_blkbits;
2215 do {
2216 err = mpage_map_one_extent(handle, mpd);
2217 if (err < 0) {
2218 struct super_block *sb = inode->i_sb;
2219
2220 if (ext4_forced_shutdown(sb))
2221 goto invalidate_dirty_pages;
2222 /*
2223 * Let the uper layers retry transient errors.
2224 * In the case of ENOSPC, if ext4_count_free_blocks()
2225 * is non-zero, a commit should free up blocks.
2226 */
2227 if ((err == -ENOMEM) ||
2228 (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2229 if (progress)
2230 goto update_disksize;
2231 return err;
2232 }
2233 ext4_msg(sb, KERN_CRIT,
2234 "Delayed block allocation failed for "
2235 "inode %lu at logical offset %llu with"
2236 " max blocks %u with error %d",
2237 inode->i_ino,
2238 (unsigned long long)map->m_lblk,
2239 (unsigned)map->m_len, -err);
2240 ext4_msg(sb, KERN_CRIT,
2241 "This should not happen!! Data will "
2242 "be lost\n");
2243 if (err == -ENOSPC)
2244 ext4_print_free_blocks(inode);
2245 invalidate_dirty_pages:
2246 *give_up_on_write = true;
2247 return err;
2248 }
2249 progress = 1;
2250 /*
2251 * Update buffer state, submit mapped pages, and get us new
2252 * extent to map
2253 */
2254 err = mpage_map_and_submit_buffers(mpd);
2255 if (err < 0)
2256 goto update_disksize;
2257 } while (map->m_len);
2258
2259 update_disksize:
2260 /*
2261 * Update on-disk size after IO is submitted. Races with
2262 * truncate are avoided by checking i_size under i_data_sem.
2263 */
2264 disksize = ((loff_t)mpd->first_page) << PAGE_SHIFT;
2265 if (disksize > READ_ONCE(EXT4_I(inode)->i_disksize)) {
2266 int err2;
2267 loff_t i_size;
2268
2269 down_write(&EXT4_I(inode)->i_data_sem);
2270 i_size = i_size_read(inode);
2271 if (disksize > i_size)
2272 disksize = i_size;
2273 if (disksize > EXT4_I(inode)->i_disksize)
2274 EXT4_I(inode)->i_disksize = disksize;
2275 up_write(&EXT4_I(inode)->i_data_sem);
2276 err2 = ext4_mark_inode_dirty(handle, inode);
2277 if (err2) {
2278 ext4_error_err(inode->i_sb, -err2,
2279 "Failed to mark inode %lu dirty",
2280 inode->i_ino);
2281 }
2282 if (!err)
2283 err = err2;
2284 }
2285 return err;
2286 }
2287
2288 /*
2289 * Calculate the total number of credits to reserve for one writepages
2290 * iteration. This is called from ext4_writepages(). We map an extent of
2291 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2292 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2293 * bpp - 1 blocks in bpp different extents.
2294 */
ext4_da_writepages_trans_blocks(struct inode * inode)2295 static int ext4_da_writepages_trans_blocks(struct inode *inode)
2296 {
2297 int bpp = ext4_journal_blocks_per_page(inode);
2298
2299 return ext4_meta_trans_blocks(inode,
2300 MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2301 }
2302
ext4_journal_folio_buffers(handle_t * handle,struct folio * folio,size_t len)2303 static int ext4_journal_folio_buffers(handle_t *handle, struct folio *folio,
2304 size_t len)
2305 {
2306 struct buffer_head *page_bufs = folio_buffers(folio);
2307 struct inode *inode = folio->mapping->host;
2308 int ret, err;
2309
2310 ret = ext4_walk_page_buffers(handle, inode, page_bufs, 0, len,
2311 NULL, do_journal_get_write_access);
2312 err = ext4_walk_page_buffers(handle, inode, page_bufs, 0, len,
2313 NULL, write_end_fn);
2314 if (ret == 0)
2315 ret = err;
2316 err = ext4_jbd2_inode_add_write(handle, inode, folio_pos(folio), len);
2317 if (ret == 0)
2318 ret = err;
2319 EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
2320
2321 return ret;
2322 }
2323
mpage_journal_page_buffers(handle_t * handle,struct mpage_da_data * mpd,struct folio * folio)2324 static int mpage_journal_page_buffers(handle_t *handle,
2325 struct mpage_da_data *mpd,
2326 struct folio *folio)
2327 {
2328 struct inode *inode = mpd->inode;
2329 loff_t size = i_size_read(inode);
2330 size_t len = folio_size(folio);
2331
2332 folio_clear_checked(folio);
2333 mpd->wbc->nr_to_write--;
2334
2335 if (folio_pos(folio) + len > size &&
2336 !ext4_verity_in_progress(inode))
2337 len = size & (len - 1);
2338
2339 return ext4_journal_folio_buffers(handle, folio, len);
2340 }
2341
2342 /*
2343 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2344 * needing mapping, submit mapped pages
2345 *
2346 * @mpd - where to look for pages
2347 *
2348 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2349 * IO immediately. If we cannot map blocks, we submit just already mapped
2350 * buffers in the page for IO and keep page dirty. When we can map blocks and
2351 * we find a page which isn't mapped we start accumulating extent of buffers
2352 * underlying these pages that needs mapping (formed by either delayed or
2353 * unwritten buffers). We also lock the pages containing these buffers. The
2354 * extent found is returned in @mpd structure (starting at mpd->lblk with
2355 * length mpd->len blocks).
2356 *
2357 * Note that this function can attach bios to one io_end structure which are
2358 * neither logically nor physically contiguous. Although it may seem as an
2359 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2360 * case as we need to track IO to all buffers underlying a page in one io_end.
2361 */
mpage_prepare_extent_to_map(struct mpage_da_data * mpd)2362 static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2363 {
2364 struct address_space *mapping = mpd->inode->i_mapping;
2365 struct folio_batch fbatch;
2366 unsigned int nr_folios;
2367 pgoff_t index = mpd->first_page;
2368 pgoff_t end = mpd->last_page;
2369 xa_mark_t tag;
2370 int i, err = 0;
2371 int blkbits = mpd->inode->i_blkbits;
2372 ext4_lblk_t lblk;
2373 struct buffer_head *head;
2374 handle_t *handle = NULL;
2375 int bpp = ext4_journal_blocks_per_page(mpd->inode);
2376
2377 if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2378 tag = PAGECACHE_TAG_TOWRITE;
2379 else
2380 tag = PAGECACHE_TAG_DIRTY;
2381
2382 mpd->map.m_len = 0;
2383 mpd->next_page = index;
2384 if (ext4_should_journal_data(mpd->inode)) {
2385 handle = ext4_journal_start(mpd->inode, EXT4_HT_WRITE_PAGE,
2386 bpp);
2387 if (IS_ERR(handle))
2388 return PTR_ERR(handle);
2389 }
2390 folio_batch_init(&fbatch);
2391 while (index <= end) {
2392 nr_folios = filemap_get_folios_tag(mapping, &index, end,
2393 tag, &fbatch);
2394 if (nr_folios == 0)
2395 break;
2396
2397 for (i = 0; i < nr_folios; i++) {
2398 struct folio *folio = fbatch.folios[i];
2399
2400 /*
2401 * Accumulated enough dirty pages? This doesn't apply
2402 * to WB_SYNC_ALL mode. For integrity sync we have to
2403 * keep going because someone may be concurrently
2404 * dirtying pages, and we might have synced a lot of
2405 * newly appeared dirty pages, but have not synced all
2406 * of the old dirty pages.
2407 */
2408 if (mpd->wbc->sync_mode == WB_SYNC_NONE &&
2409 mpd->wbc->nr_to_write <=
2410 mpd->map.m_len >> (PAGE_SHIFT - blkbits))
2411 goto out;
2412
2413 /* If we can't merge this page, we are done. */
2414 if (mpd->map.m_len > 0 && mpd->next_page != folio->index)
2415 goto out;
2416
2417 if (handle) {
2418 err = ext4_journal_ensure_credits(handle, bpp,
2419 0);
2420 if (err < 0)
2421 goto out;
2422 }
2423
2424 folio_lock(folio);
2425 /*
2426 * If the page is no longer dirty, or its mapping no
2427 * longer corresponds to inode we are writing (which
2428 * means it has been truncated or invalidated), or the
2429 * page is already under writeback and we are not doing
2430 * a data integrity writeback, skip the page
2431 */
2432 if (!folio_test_dirty(folio) ||
2433 (folio_test_writeback(folio) &&
2434 (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2435 unlikely(folio->mapping != mapping)) {
2436 folio_unlock(folio);
2437 continue;
2438 }
2439
2440 folio_wait_writeback(folio);
2441 BUG_ON(folio_test_writeback(folio));
2442
2443 /*
2444 * Should never happen but for buggy code in
2445 * other subsystems that call
2446 * set_page_dirty() without properly warning
2447 * the file system first. See [1] for more
2448 * information.
2449 *
2450 * [1] https://lore.kernel.org/linux-mm/20180103100430.GE4911@quack2.suse.cz
2451 */
2452 if (!folio_buffers(folio)) {
2453 ext4_warning_inode(mpd->inode, "page %lu does not have buffers attached", folio->index);
2454 folio_clear_dirty(folio);
2455 folio_unlock(folio);
2456 continue;
2457 }
2458
2459 if (mpd->map.m_len == 0)
2460 mpd->first_page = folio->index;
2461 mpd->next_page = folio_next_index(folio);
2462 /*
2463 * Writeout when we cannot modify metadata is simple.
2464 * Just submit the page. For data=journal mode we
2465 * first handle writeout of the page for checkpoint and
2466 * only after that handle delayed page dirtying. This
2467 * makes sure current data is checkpointed to the final
2468 * location before possibly journalling it again which
2469 * is desirable when the page is frequently dirtied
2470 * through a pin.
2471 */
2472 if (!mpd->can_map) {
2473 err = mpage_submit_folio(mpd, folio);
2474 if (err < 0)
2475 goto out;
2476 /* Pending dirtying of journalled data? */
2477 if (folio_test_checked(folio)) {
2478 err = mpage_journal_page_buffers(handle,
2479 mpd, folio);
2480 if (err < 0)
2481 goto out;
2482 mpd->journalled_more_data = 1;
2483 }
2484 mpage_folio_done(mpd, folio);
2485 } else {
2486 /* Add all dirty buffers to mpd */
2487 lblk = ((ext4_lblk_t)folio->index) <<
2488 (PAGE_SHIFT - blkbits);
2489 head = folio_buffers(folio);
2490 err = mpage_process_page_bufs(mpd, head, head,
2491 lblk);
2492 if (err <= 0)
2493 goto out;
2494 err = 0;
2495 }
2496 }
2497 folio_batch_release(&fbatch);
2498 cond_resched();
2499 }
2500 mpd->scanned_until_end = 1;
2501 if (handle)
2502 ext4_journal_stop(handle);
2503 return 0;
2504 out:
2505 folio_batch_release(&fbatch);
2506 if (handle)
2507 ext4_journal_stop(handle);
2508 return err;
2509 }
2510
ext4_do_writepages(struct mpage_da_data * mpd)2511 static int ext4_do_writepages(struct mpage_da_data *mpd)
2512 {
2513 struct writeback_control *wbc = mpd->wbc;
2514 pgoff_t writeback_index = 0;
2515 long nr_to_write = wbc->nr_to_write;
2516 int range_whole = 0;
2517 int cycled = 1;
2518 handle_t *handle = NULL;
2519 struct inode *inode = mpd->inode;
2520 struct address_space *mapping = inode->i_mapping;
2521 int needed_blocks, rsv_blocks = 0, ret = 0;
2522 struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2523 struct blk_plug plug;
2524 bool give_up_on_write = false;
2525
2526 trace_ext4_writepages(inode, wbc);
2527
2528 /*
2529 * No pages to write? This is mainly a kludge to avoid starting
2530 * a transaction for special inodes like journal inode on last iput()
2531 * because that could violate lock ordering on umount
2532 */
2533 if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2534 goto out_writepages;
2535
2536 /*
2537 * If the filesystem has aborted, it is read-only, so return
2538 * right away instead of dumping stack traces later on that
2539 * will obscure the real source of the problem. We test
2540 * fs shutdown state instead of sb->s_flag's SB_RDONLY because
2541 * the latter could be true if the filesystem is mounted
2542 * read-only, and in that case, ext4_writepages should
2543 * *never* be called, so if that ever happens, we would want
2544 * the stack trace.
2545 */
2546 if (unlikely(ext4_forced_shutdown(mapping->host->i_sb))) {
2547 ret = -EROFS;
2548 goto out_writepages;
2549 }
2550
2551 /*
2552 * If we have inline data and arrive here, it means that
2553 * we will soon create the block for the 1st page, so
2554 * we'd better clear the inline data here.
2555 */
2556 if (ext4_has_inline_data(inode)) {
2557 /* Just inode will be modified... */
2558 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2559 if (IS_ERR(handle)) {
2560 ret = PTR_ERR(handle);
2561 goto out_writepages;
2562 }
2563 BUG_ON(ext4_test_inode_state(inode,
2564 EXT4_STATE_MAY_INLINE_DATA));
2565 ext4_destroy_inline_data(handle, inode);
2566 ext4_journal_stop(handle);
2567 }
2568
2569 /*
2570 * data=journal mode does not do delalloc so we just need to writeout /
2571 * journal already mapped buffers. On the other hand we need to commit
2572 * transaction to make data stable. We expect all the data to be
2573 * already in the journal (the only exception are DMA pinned pages
2574 * dirtied behind our back) so we commit transaction here and run the
2575 * writeback loop to checkpoint them. The checkpointing is not actually
2576 * necessary to make data persistent *but* quite a few places (extent
2577 * shifting operations, fsverity, ...) depend on being able to drop
2578 * pagecache pages after calling filemap_write_and_wait() and for that
2579 * checkpointing needs to happen.
2580 */
2581 if (ext4_should_journal_data(inode)) {
2582 mpd->can_map = 0;
2583 if (wbc->sync_mode == WB_SYNC_ALL)
2584 ext4_fc_commit(sbi->s_journal,
2585 EXT4_I(inode)->i_datasync_tid);
2586 }
2587 mpd->journalled_more_data = 0;
2588
2589 if (ext4_should_dioread_nolock(inode)) {
2590 /*
2591 * We may need to convert up to one extent per block in
2592 * the page and we may dirty the inode.
2593 */
2594 rsv_blocks = 1 + ext4_chunk_trans_blocks(inode,
2595 PAGE_SIZE >> inode->i_blkbits);
2596 }
2597
2598 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2599 range_whole = 1;
2600
2601 if (wbc->range_cyclic) {
2602 writeback_index = mapping->writeback_index;
2603 if (writeback_index)
2604 cycled = 0;
2605 mpd->first_page = writeback_index;
2606 mpd->last_page = -1;
2607 } else {
2608 mpd->first_page = wbc->range_start >> PAGE_SHIFT;
2609 mpd->last_page = wbc->range_end >> PAGE_SHIFT;
2610 }
2611
2612 ext4_io_submit_init(&mpd->io_submit, wbc);
2613 retry:
2614 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2615 tag_pages_for_writeback(mapping, mpd->first_page,
2616 mpd->last_page);
2617 blk_start_plug(&plug);
2618
2619 /*
2620 * First writeback pages that don't need mapping - we can avoid
2621 * starting a transaction unnecessarily and also avoid being blocked
2622 * in the block layer on device congestion while having transaction
2623 * started.
2624 */
2625 mpd->do_map = 0;
2626 mpd->scanned_until_end = 0;
2627 mpd->io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2628 if (!mpd->io_submit.io_end) {
2629 ret = -ENOMEM;
2630 goto unplug;
2631 }
2632 ret = mpage_prepare_extent_to_map(mpd);
2633 /* Unlock pages we didn't use */
2634 mpage_release_unused_pages(mpd, false);
2635 /* Submit prepared bio */
2636 ext4_io_submit(&mpd->io_submit);
2637 ext4_put_io_end_defer(mpd->io_submit.io_end);
2638 mpd->io_submit.io_end = NULL;
2639 if (ret < 0)
2640 goto unplug;
2641
2642 while (!mpd->scanned_until_end && wbc->nr_to_write > 0) {
2643 /* For each extent of pages we use new io_end */
2644 mpd->io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2645 if (!mpd->io_submit.io_end) {
2646 ret = -ENOMEM;
2647 break;
2648 }
2649
2650 WARN_ON_ONCE(!mpd->can_map);
2651 /*
2652 * We have two constraints: We find one extent to map and we
2653 * must always write out whole page (makes a difference when
2654 * blocksize < pagesize) so that we don't block on IO when we
2655 * try to write out the rest of the page. Journalled mode is
2656 * not supported by delalloc.
2657 */
2658 BUG_ON(ext4_should_journal_data(inode));
2659 needed_blocks = ext4_da_writepages_trans_blocks(inode);
2660
2661 /* start a new transaction */
2662 handle = ext4_journal_start_with_reserve(inode,
2663 EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2664 if (IS_ERR(handle)) {
2665 ret = PTR_ERR(handle);
2666 ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2667 "%ld pages, ino %lu; err %d", __func__,
2668 wbc->nr_to_write, inode->i_ino, ret);
2669 /* Release allocated io_end */
2670 ext4_put_io_end(mpd->io_submit.io_end);
2671 mpd->io_submit.io_end = NULL;
2672 break;
2673 }
2674 mpd->do_map = 1;
2675
2676 trace_ext4_da_write_pages(inode, mpd->first_page, wbc);
2677 ret = mpage_prepare_extent_to_map(mpd);
2678 if (!ret && mpd->map.m_len)
2679 ret = mpage_map_and_submit_extent(handle, mpd,
2680 &give_up_on_write);
2681 /*
2682 * Caution: If the handle is synchronous,
2683 * ext4_journal_stop() can wait for transaction commit
2684 * to finish which may depend on writeback of pages to
2685 * complete or on page lock to be released. In that
2686 * case, we have to wait until after we have
2687 * submitted all the IO, released page locks we hold,
2688 * and dropped io_end reference (for extent conversion
2689 * to be able to complete) before stopping the handle.
2690 */
2691 if (!ext4_handle_valid(handle) || handle->h_sync == 0) {
2692 ext4_journal_stop(handle);
2693 handle = NULL;
2694 mpd->do_map = 0;
2695 }
2696 /* Unlock pages we didn't use */
2697 mpage_release_unused_pages(mpd, give_up_on_write);
2698 /* Submit prepared bio */
2699 ext4_io_submit(&mpd->io_submit);
2700
2701 /*
2702 * Drop our io_end reference we got from init. We have
2703 * to be careful and use deferred io_end finishing if
2704 * we are still holding the transaction as we can
2705 * release the last reference to io_end which may end
2706 * up doing unwritten extent conversion.
2707 */
2708 if (handle) {
2709 ext4_put_io_end_defer(mpd->io_submit.io_end);
2710 ext4_journal_stop(handle);
2711 } else
2712 ext4_put_io_end(mpd->io_submit.io_end);
2713 mpd->io_submit.io_end = NULL;
2714
2715 if (ret == -ENOSPC && sbi->s_journal) {
2716 /*
2717 * Commit the transaction which would
2718 * free blocks released in the transaction
2719 * and try again
2720 */
2721 jbd2_journal_force_commit_nested(sbi->s_journal);
2722 ret = 0;
2723 continue;
2724 }
2725 /* Fatal error - ENOMEM, EIO... */
2726 if (ret)
2727 break;
2728 }
2729 unplug:
2730 blk_finish_plug(&plug);
2731 if (!ret && !cycled && wbc->nr_to_write > 0) {
2732 cycled = 1;
2733 mpd->last_page = writeback_index - 1;
2734 mpd->first_page = 0;
2735 goto retry;
2736 }
2737
2738 /* Update index */
2739 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2740 /*
2741 * Set the writeback_index so that range_cyclic
2742 * mode will write it back later
2743 */
2744 mapping->writeback_index = mpd->first_page;
2745
2746 out_writepages:
2747 trace_ext4_writepages_result(inode, wbc, ret,
2748 nr_to_write - wbc->nr_to_write);
2749 return ret;
2750 }
2751
ext4_writepages(struct address_space * mapping,struct writeback_control * wbc)2752 static int ext4_writepages(struct address_space *mapping,
2753 struct writeback_control *wbc)
2754 {
2755 struct super_block *sb = mapping->host->i_sb;
2756 struct mpage_da_data mpd = {
2757 .inode = mapping->host,
2758 .wbc = wbc,
2759 .can_map = 1,
2760 };
2761 int ret;
2762 int alloc_ctx;
2763
2764 if (unlikely(ext4_forced_shutdown(sb)))
2765 return -EIO;
2766
2767 alloc_ctx = ext4_writepages_down_read(sb);
2768 ret = ext4_do_writepages(&mpd);
2769 /*
2770 * For data=journal writeback we could have come across pages marked
2771 * for delayed dirtying (PageChecked) which were just added to the
2772 * running transaction. Try once more to get them to stable storage.
2773 */
2774 if (!ret && mpd.journalled_more_data)
2775 ret = ext4_do_writepages(&mpd);
2776 ext4_writepages_up_read(sb, alloc_ctx);
2777
2778 return ret;
2779 }
2780
ext4_normal_submit_inode_data_buffers(struct jbd2_inode * jinode)2781 int ext4_normal_submit_inode_data_buffers(struct jbd2_inode *jinode)
2782 {
2783 struct writeback_control wbc = {
2784 .sync_mode = WB_SYNC_ALL,
2785 .nr_to_write = LONG_MAX,
2786 .range_start = jinode->i_dirty_start,
2787 .range_end = jinode->i_dirty_end,
2788 };
2789 struct mpage_da_data mpd = {
2790 .inode = jinode->i_vfs_inode,
2791 .wbc = &wbc,
2792 .can_map = 0,
2793 };
2794 return ext4_do_writepages(&mpd);
2795 }
2796
ext4_dax_writepages(struct address_space * mapping,struct writeback_control * wbc)2797 static int ext4_dax_writepages(struct address_space *mapping,
2798 struct writeback_control *wbc)
2799 {
2800 int ret;
2801 long nr_to_write = wbc->nr_to_write;
2802 struct inode *inode = mapping->host;
2803 int alloc_ctx;
2804
2805 if (unlikely(ext4_forced_shutdown(inode->i_sb)))
2806 return -EIO;
2807
2808 alloc_ctx = ext4_writepages_down_read(inode->i_sb);
2809 trace_ext4_writepages(inode, wbc);
2810
2811 ret = dax_writeback_mapping_range(mapping,
2812 EXT4_SB(inode->i_sb)->s_daxdev, wbc);
2813 trace_ext4_writepages_result(inode, wbc, ret,
2814 nr_to_write - wbc->nr_to_write);
2815 ext4_writepages_up_read(inode->i_sb, alloc_ctx);
2816 return ret;
2817 }
2818
ext4_nonda_switch(struct super_block * sb)2819 static int ext4_nonda_switch(struct super_block *sb)
2820 {
2821 s64 free_clusters, dirty_clusters;
2822 struct ext4_sb_info *sbi = EXT4_SB(sb);
2823
2824 /*
2825 * switch to non delalloc mode if we are running low
2826 * on free block. The free block accounting via percpu
2827 * counters can get slightly wrong with percpu_counter_batch getting
2828 * accumulated on each CPU without updating global counters
2829 * Delalloc need an accurate free block accounting. So switch
2830 * to non delalloc when we are near to error range.
2831 */
2832 free_clusters =
2833 percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2834 dirty_clusters =
2835 percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2836 /*
2837 * Start pushing delalloc when 1/2 of free blocks are dirty.
2838 */
2839 if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
2840 try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2841
2842 if (2 * free_clusters < 3 * dirty_clusters ||
2843 free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
2844 /*
2845 * free block count is less than 150% of dirty blocks
2846 * or free blocks is less than watermark
2847 */
2848 return 1;
2849 }
2850 return 0;
2851 }
2852
ext4_da_write_begin(struct file * file,struct address_space * mapping,loff_t pos,unsigned len,struct page ** pagep,void ** fsdata)2853 static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2854 loff_t pos, unsigned len,
2855 struct page **pagep, void **fsdata)
2856 {
2857 int ret, retries = 0;
2858 struct folio *folio;
2859 pgoff_t index;
2860 struct inode *inode = mapping->host;
2861
2862 if (unlikely(ext4_forced_shutdown(inode->i_sb)))
2863 return -EIO;
2864
2865 index = pos >> PAGE_SHIFT;
2866
2867 if (ext4_nonda_switch(inode->i_sb) || ext4_verity_in_progress(inode)) {
2868 *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2869 return ext4_write_begin(file, mapping, pos,
2870 len, pagep, fsdata);
2871 }
2872 *fsdata = (void *)0;
2873 trace_ext4_da_write_begin(inode, pos, len);
2874
2875 if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2876 ret = ext4_da_write_inline_data_begin(mapping, inode, pos, len,
2877 pagep, fsdata);
2878 if (ret < 0)
2879 return ret;
2880 if (ret == 1)
2881 return 0;
2882 }
2883
2884 retry:
2885 folio = __filemap_get_folio(mapping, index, FGP_WRITEBEGIN,
2886 mapping_gfp_mask(mapping));
2887 if (IS_ERR(folio))
2888 return PTR_ERR(folio);
2889
2890 #ifdef CONFIG_FS_ENCRYPTION
2891 ret = ext4_block_write_begin(folio, pos, len, ext4_da_get_block_prep);
2892 #else
2893 ret = __block_write_begin(&folio->page, pos, len, ext4_da_get_block_prep);
2894 #endif
2895 if (ret < 0) {
2896 folio_unlock(folio);
2897 folio_put(folio);
2898 /*
2899 * block_write_begin may have instantiated a few blocks
2900 * outside i_size. Trim these off again. Don't need
2901 * i_size_read because we hold inode lock.
2902 */
2903 if (pos + len > inode->i_size)
2904 ext4_truncate_failed_write(inode);
2905
2906 if (ret == -ENOSPC &&
2907 ext4_should_retry_alloc(inode->i_sb, &retries))
2908 goto retry;
2909 return ret;
2910 }
2911
2912 *pagep = &folio->page;
2913 return ret;
2914 }
2915
2916 /*
2917 * Check if we should update i_disksize
2918 * when write to the end of file but not require block allocation
2919 */
ext4_da_should_update_i_disksize(struct folio * folio,unsigned long offset)2920 static int ext4_da_should_update_i_disksize(struct folio *folio,
2921 unsigned long offset)
2922 {
2923 struct buffer_head *bh;
2924 struct inode *inode = folio->mapping->host;
2925 unsigned int idx;
2926 int i;
2927
2928 bh = folio_buffers(folio);
2929 idx = offset >> inode->i_blkbits;
2930
2931 for (i = 0; i < idx; i++)
2932 bh = bh->b_this_page;
2933
2934 if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
2935 return 0;
2936 return 1;
2937 }
2938
ext4_da_do_write_end(struct address_space * mapping,loff_t pos,unsigned len,unsigned copied,struct folio * folio)2939 static int ext4_da_do_write_end(struct address_space *mapping,
2940 loff_t pos, unsigned len, unsigned copied,
2941 struct folio *folio)
2942 {
2943 struct inode *inode = mapping->host;
2944 loff_t old_size = inode->i_size;
2945 bool disksize_changed = false;
2946 loff_t new_i_size;
2947
2948 /*
2949 * block_write_end() will mark the inode as dirty with I_DIRTY_PAGES
2950 * flag, which all that's needed to trigger page writeback.
2951 */
2952 copied = block_write_end(NULL, mapping, pos, len, copied,
2953 &folio->page, NULL);
2954 new_i_size = pos + copied;
2955
2956 /*
2957 * It's important to update i_size while still holding folio lock,
2958 * because folio writeout could otherwise come in and zero beyond
2959 * i_size.
2960 *
2961 * Since we are holding inode lock, we are sure i_disksize <=
2962 * i_size. We also know that if i_disksize < i_size, there are
2963 * delalloc writes pending in the range up to i_size. If the end of
2964 * the current write is <= i_size, there's no need to touch
2965 * i_disksize since writeback will push i_disksize up to i_size
2966 * eventually. If the end of the current write is > i_size and
2967 * inside an allocated block which ext4_da_should_update_i_disksize()
2968 * checked, we need to update i_disksize here as certain
2969 * ext4_writepages() paths not allocating blocks and update i_disksize.
2970 */
2971 if (new_i_size > inode->i_size) {
2972 unsigned long end;
2973
2974 i_size_write(inode, new_i_size);
2975 end = (new_i_size - 1) & (PAGE_SIZE - 1);
2976 if (copied && ext4_da_should_update_i_disksize(folio, end)) {
2977 ext4_update_i_disksize(inode, new_i_size);
2978 disksize_changed = true;
2979 }
2980 }
2981
2982 folio_unlock(folio);
2983 folio_put(folio);
2984
2985 if (old_size < pos)
2986 pagecache_isize_extended(inode, old_size, pos);
2987
2988 if (disksize_changed) {
2989 handle_t *handle;
2990
2991 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
2992 if (IS_ERR(handle))
2993 return PTR_ERR(handle);
2994 ext4_mark_inode_dirty(handle, inode);
2995 ext4_journal_stop(handle);
2996 }
2997
2998 return copied;
2999 }
3000
ext4_da_write_end(struct file * file,struct address_space * mapping,loff_t pos,unsigned len,unsigned copied,struct page * page,void * fsdata)3001 static int ext4_da_write_end(struct file *file,
3002 struct address_space *mapping,
3003 loff_t pos, unsigned len, unsigned copied,
3004 struct page *page, void *fsdata)
3005 {
3006 struct inode *inode = mapping->host;
3007 int write_mode = (int)(unsigned long)fsdata;
3008 struct folio *folio = page_folio(page);
3009
3010 if (write_mode == FALL_BACK_TO_NONDELALLOC)
3011 return ext4_write_end(file, mapping, pos,
3012 len, copied, &folio->page, fsdata);
3013
3014 trace_ext4_da_write_end(inode, pos, len, copied);
3015
3016 if (write_mode != CONVERT_INLINE_DATA &&
3017 ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
3018 ext4_has_inline_data(inode))
3019 return ext4_write_inline_data_end(inode, pos, len, copied,
3020 folio);
3021
3022 if (unlikely(copied < len) && !folio_test_uptodate(folio))
3023 copied = 0;
3024
3025 return ext4_da_do_write_end(mapping, pos, len, copied, folio);
3026 }
3027
3028 /*
3029 * Force all delayed allocation blocks to be allocated for a given inode.
3030 */
ext4_alloc_da_blocks(struct inode * inode)3031 int ext4_alloc_da_blocks(struct inode *inode)
3032 {
3033 trace_ext4_alloc_da_blocks(inode);
3034
3035 if (!EXT4_I(inode)->i_reserved_data_blocks)
3036 return 0;
3037
3038 /*
3039 * We do something simple for now. The filemap_flush() will
3040 * also start triggering a write of the data blocks, which is
3041 * not strictly speaking necessary (and for users of
3042 * laptop_mode, not even desirable). However, to do otherwise
3043 * would require replicating code paths in:
3044 *
3045 * ext4_writepages() ->
3046 * write_cache_pages() ---> (via passed in callback function)
3047 * __mpage_da_writepage() -->
3048 * mpage_add_bh_to_extent()
3049 * mpage_da_map_blocks()
3050 *
3051 * The problem is that write_cache_pages(), located in
3052 * mm/page-writeback.c, marks pages clean in preparation for
3053 * doing I/O, which is not desirable if we're not planning on
3054 * doing I/O at all.
3055 *
3056 * We could call write_cache_pages(), and then redirty all of
3057 * the pages by calling redirty_page_for_writepage() but that
3058 * would be ugly in the extreme. So instead we would need to
3059 * replicate parts of the code in the above functions,
3060 * simplifying them because we wouldn't actually intend to
3061 * write out the pages, but rather only collect contiguous
3062 * logical block extents, call the multi-block allocator, and
3063 * then update the buffer heads with the block allocations.
3064 *
3065 * For now, though, we'll cheat by calling filemap_flush(),
3066 * which will map the blocks, and start the I/O, but not
3067 * actually wait for the I/O to complete.
3068 */
3069 return filemap_flush(inode->i_mapping);
3070 }
3071
3072 /*
3073 * bmap() is special. It gets used by applications such as lilo and by
3074 * the swapper to find the on-disk block of a specific piece of data.
3075 *
3076 * Naturally, this is dangerous if the block concerned is still in the
3077 * journal. If somebody makes a swapfile on an ext4 data-journaling
3078 * filesystem and enables swap, then they may get a nasty shock when the
3079 * data getting swapped to that swapfile suddenly gets overwritten by
3080 * the original zero's written out previously to the journal and
3081 * awaiting writeback in the kernel's buffer cache.
3082 *
3083 * So, if we see any bmap calls here on a modified, data-journaled file,
3084 * take extra steps to flush any blocks which might be in the cache.
3085 */
ext4_bmap(struct address_space * mapping,sector_t block)3086 static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3087 {
3088 struct inode *inode = mapping->host;
3089 sector_t ret = 0;
3090
3091 inode_lock_shared(inode);
3092 /*
3093 * We can get here for an inline file via the FIBMAP ioctl
3094 */
3095 if (ext4_has_inline_data(inode))
3096 goto out;
3097
3098 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3099 (test_opt(inode->i_sb, DELALLOC) ||
3100 ext4_should_journal_data(inode))) {
3101 /*
3102 * With delalloc or journalled data we want to sync the file so
3103 * that we can make sure we allocate blocks for file and data
3104 * is in place for the user to see it
3105 */
3106 filemap_write_and_wait(mapping);
3107 }
3108
3109 ret = iomap_bmap(mapping, block, &ext4_iomap_ops);
3110
3111 out:
3112 inode_unlock_shared(inode);
3113 return ret;
3114 }
3115
ext4_read_folio(struct file * file,struct folio * folio)3116 static int ext4_read_folio(struct file *file, struct folio *folio)
3117 {
3118 int ret = -EAGAIN;
3119 struct inode *inode = folio->mapping->host;
3120
3121 trace_ext4_read_folio(inode, folio);
3122
3123 if (ext4_has_inline_data(inode))
3124 ret = ext4_readpage_inline(inode, folio);
3125
3126 if (ret == -EAGAIN)
3127 return ext4_mpage_readpages(inode, NULL, folio);
3128
3129 return ret;
3130 }
3131
ext4_readahead(struct readahead_control * rac)3132 static void ext4_readahead(struct readahead_control *rac)
3133 {
3134 struct inode *inode = rac->mapping->host;
3135
3136 /* If the file has inline data, no need to do readahead. */
3137 if (ext4_has_inline_data(inode))
3138 return;
3139
3140 ext4_mpage_readpages(inode, rac, NULL);
3141 }
3142
ext4_invalidate_folio(struct folio * folio,size_t offset,size_t length)3143 static void ext4_invalidate_folio(struct folio *folio, size_t offset,
3144 size_t length)
3145 {
3146 trace_ext4_invalidate_folio(folio, offset, length);
3147
3148 /* No journalling happens on data buffers when this function is used */
3149 WARN_ON(folio_buffers(folio) && buffer_jbd(folio_buffers(folio)));
3150
3151 block_invalidate_folio(folio, offset, length);
3152 }
3153
__ext4_journalled_invalidate_folio(struct folio * folio,size_t offset,size_t length)3154 static int __ext4_journalled_invalidate_folio(struct folio *folio,
3155 size_t offset, size_t length)
3156 {
3157 journal_t *journal = EXT4_JOURNAL(folio->mapping->host);
3158
3159 trace_ext4_journalled_invalidate_folio(folio, offset, length);
3160
3161 /*
3162 * If it's a full truncate we just forget about the pending dirtying
3163 */
3164 if (offset == 0 && length == folio_size(folio))
3165 folio_clear_checked(folio);
3166
3167 return jbd2_journal_invalidate_folio(journal, folio, offset, length);
3168 }
3169
3170 /* Wrapper for aops... */
ext4_journalled_invalidate_folio(struct folio * folio,size_t offset,size_t length)3171 static void ext4_journalled_invalidate_folio(struct folio *folio,
3172 size_t offset,
3173 size_t length)
3174 {
3175 WARN_ON(__ext4_journalled_invalidate_folio(folio, offset, length) < 0);
3176 }
3177
ext4_release_folio(struct folio * folio,gfp_t wait)3178 static bool ext4_release_folio(struct folio *folio, gfp_t wait)
3179 {
3180 struct inode *inode = folio->mapping->host;
3181 journal_t *journal = EXT4_JOURNAL(inode);
3182
3183 trace_ext4_release_folio(inode, folio);
3184
3185 /* Page has dirty journalled data -> cannot release */
3186 if (folio_test_checked(folio))
3187 return false;
3188 if (journal)
3189 return jbd2_journal_try_to_free_buffers(journal, folio);
3190 else
3191 return try_to_free_buffers(folio);
3192 }
3193
ext4_inode_datasync_dirty(struct inode * inode)3194 static bool ext4_inode_datasync_dirty(struct inode *inode)
3195 {
3196 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
3197
3198 if (journal) {
3199 if (jbd2_transaction_committed(journal,
3200 EXT4_I(inode)->i_datasync_tid))
3201 return false;
3202 if (test_opt2(inode->i_sb, JOURNAL_FAST_COMMIT))
3203 return !list_empty(&EXT4_I(inode)->i_fc_list);
3204 return true;
3205 }
3206
3207 /* Any metadata buffers to write? */
3208 if (!list_empty(&inode->i_mapping->i_private_list))
3209 return true;
3210 return inode->i_state & I_DIRTY_DATASYNC;
3211 }
3212
ext4_set_iomap(struct inode * inode,struct iomap * iomap,struct ext4_map_blocks * map,loff_t offset,loff_t length,unsigned int flags)3213 static void ext4_set_iomap(struct inode *inode, struct iomap *iomap,
3214 struct ext4_map_blocks *map, loff_t offset,
3215 loff_t length, unsigned int flags)
3216 {
3217 u8 blkbits = inode->i_blkbits;
3218
3219 /*
3220 * Writes that span EOF might trigger an I/O size update on completion,
3221 * so consider them to be dirty for the purpose of O_DSYNC, even if
3222 * there is no other metadata changes being made or are pending.
3223 */
3224 iomap->flags = 0;
3225 if (ext4_inode_datasync_dirty(inode) ||
3226 offset + length > i_size_read(inode))
3227 iomap->flags |= IOMAP_F_DIRTY;
3228
3229 if (map->m_flags & EXT4_MAP_NEW)
3230 iomap->flags |= IOMAP_F_NEW;
3231
3232 if (flags & IOMAP_DAX)
3233 iomap->dax_dev = EXT4_SB(inode->i_sb)->s_daxdev;
3234 else
3235 iomap->bdev = inode->i_sb->s_bdev;
3236 iomap->offset = (u64) map->m_lblk << blkbits;
3237 iomap->length = (u64) map->m_len << blkbits;
3238
3239 if ((map->m_flags & EXT4_MAP_MAPPED) &&
3240 !ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3241 iomap->flags |= IOMAP_F_MERGED;
3242
3243 /*
3244 * Flags passed to ext4_map_blocks() for direct I/O writes can result
3245 * in m_flags having both EXT4_MAP_MAPPED and EXT4_MAP_UNWRITTEN bits
3246 * set. In order for any allocated unwritten extents to be converted
3247 * into written extents correctly within the ->end_io() handler, we
3248 * need to ensure that the iomap->type is set appropriately. Hence, the
3249 * reason why we need to check whether the EXT4_MAP_UNWRITTEN bit has
3250 * been set first.
3251 */
3252 if (map->m_flags & EXT4_MAP_UNWRITTEN) {
3253 iomap->type = IOMAP_UNWRITTEN;
3254 iomap->addr = (u64) map->m_pblk << blkbits;
3255 if (flags & IOMAP_DAX)
3256 iomap->addr += EXT4_SB(inode->i_sb)->s_dax_part_off;
3257 } else if (map->m_flags & EXT4_MAP_MAPPED) {
3258 iomap->type = IOMAP_MAPPED;
3259 iomap->addr = (u64) map->m_pblk << blkbits;
3260 if (flags & IOMAP_DAX)
3261 iomap->addr += EXT4_SB(inode->i_sb)->s_dax_part_off;
3262 } else if (map->m_flags & EXT4_MAP_DELAYED) {
3263 iomap->type = IOMAP_DELALLOC;
3264 iomap->addr = IOMAP_NULL_ADDR;
3265 } else {
3266 iomap->type = IOMAP_HOLE;
3267 iomap->addr = IOMAP_NULL_ADDR;
3268 }
3269 }
3270
ext4_iomap_alloc(struct inode * inode,struct ext4_map_blocks * map,unsigned int flags)3271 static int ext4_iomap_alloc(struct inode *inode, struct ext4_map_blocks *map,
3272 unsigned int flags)
3273 {
3274 handle_t *handle;
3275 u8 blkbits = inode->i_blkbits;
3276 int ret, dio_credits, m_flags = 0, retries = 0;
3277
3278 /*
3279 * Trim the mapping request to the maximum value that we can map at
3280 * once for direct I/O.
3281 */
3282 if (map->m_len > DIO_MAX_BLOCKS)
3283 map->m_len = DIO_MAX_BLOCKS;
3284 dio_credits = ext4_chunk_trans_blocks(inode, map->m_len);
3285
3286 retry:
3287 /*
3288 * Either we allocate blocks and then don't get an unwritten extent, so
3289 * in that case we have reserved enough credits. Or, the blocks are
3290 * already allocated and unwritten. In that case, the extent conversion
3291 * fits into the credits as well.
3292 */
3293 handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits);
3294 if (IS_ERR(handle))
3295 return PTR_ERR(handle);
3296
3297 /*
3298 * DAX and direct I/O are the only two operations that are currently
3299 * supported with IOMAP_WRITE.
3300 */
3301 WARN_ON(!(flags & (IOMAP_DAX | IOMAP_DIRECT)));
3302 if (flags & IOMAP_DAX)
3303 m_flags = EXT4_GET_BLOCKS_CREATE_ZERO;
3304 /*
3305 * We use i_size instead of i_disksize here because delalloc writeback
3306 * can complete at any point during the I/O and subsequently push the
3307 * i_disksize out to i_size. This could be beyond where direct I/O is
3308 * happening and thus expose allocated blocks to direct I/O reads.
3309 */
3310 else if (((loff_t)map->m_lblk << blkbits) >= i_size_read(inode))
3311 m_flags = EXT4_GET_BLOCKS_CREATE;
3312 else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3313 m_flags = EXT4_GET_BLOCKS_IO_CREATE_EXT;
3314
3315 ret = ext4_map_blocks(handle, inode, map, m_flags);
3316
3317 /*
3318 * We cannot fill holes in indirect tree based inodes as that could
3319 * expose stale data in the case of a crash. Use the magic error code
3320 * to fallback to buffered I/O.
3321 */
3322 if (!m_flags && !ret)
3323 ret = -ENOTBLK;
3324
3325 ext4_journal_stop(handle);
3326 if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3327 goto retry;
3328
3329 return ret;
3330 }
3331
3332
ext4_iomap_begin(struct inode * inode,loff_t offset,loff_t length,unsigned flags,struct iomap * iomap,struct iomap * srcmap)3333 static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
3334 unsigned flags, struct iomap *iomap, struct iomap *srcmap)
3335 {
3336 int ret;
3337 struct ext4_map_blocks map;
3338 u8 blkbits = inode->i_blkbits;
3339
3340 if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
3341 return -EINVAL;
3342
3343 if (WARN_ON_ONCE(ext4_has_inline_data(inode)))
3344 return -ERANGE;
3345
3346 /*
3347 * Calculate the first and last logical blocks respectively.
3348 */
3349 map.m_lblk = offset >> blkbits;
3350 map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits,
3351 EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1;
3352
3353 if (flags & IOMAP_WRITE) {
3354 /*
3355 * We check here if the blocks are already allocated, then we
3356 * don't need to start a journal txn and we can directly return
3357 * the mapping information. This could boost performance
3358 * especially in multi-threaded overwrite requests.
3359 */
3360 if (offset + length <= i_size_read(inode)) {
3361 ret = ext4_map_blocks(NULL, inode, &map, 0);
3362 if (ret > 0 && (map.m_flags & EXT4_MAP_MAPPED))
3363 goto out;
3364 }
3365 ret = ext4_iomap_alloc(inode, &map, flags);
3366 } else {
3367 ret = ext4_map_blocks(NULL, inode, &map, 0);
3368 }
3369
3370 if (ret < 0)
3371 return ret;
3372 out:
3373 /*
3374 * When inline encryption is enabled, sometimes I/O to an encrypted file
3375 * has to be broken up to guarantee DUN contiguity. Handle this by
3376 * limiting the length of the mapping returned.
3377 */
3378 map.m_len = fscrypt_limit_io_blocks(inode, map.m_lblk, map.m_len);
3379
3380 ext4_set_iomap(inode, iomap, &map, offset, length, flags);
3381
3382 return 0;
3383 }
3384
ext4_iomap_overwrite_begin(struct inode * inode,loff_t offset,loff_t length,unsigned flags,struct iomap * iomap,struct iomap * srcmap)3385 static int ext4_iomap_overwrite_begin(struct inode *inode, loff_t offset,
3386 loff_t length, unsigned flags, struct iomap *iomap,
3387 struct iomap *srcmap)
3388 {
3389 int ret;
3390
3391 /*
3392 * Even for writes we don't need to allocate blocks, so just pretend
3393 * we are reading to save overhead of starting a transaction.
3394 */
3395 flags &= ~IOMAP_WRITE;
3396 ret = ext4_iomap_begin(inode, offset, length, flags, iomap, srcmap);
3397 WARN_ON_ONCE(!ret && iomap->type != IOMAP_MAPPED);
3398 return ret;
3399 }
3400
ext4_iomap_end(struct inode * inode,loff_t offset,loff_t length,ssize_t written,unsigned flags,struct iomap * iomap)3401 static int ext4_iomap_end(struct inode *inode, loff_t offset, loff_t length,
3402 ssize_t written, unsigned flags, struct iomap *iomap)
3403 {
3404 /*
3405 * Check to see whether an error occurred while writing out the data to
3406 * the allocated blocks. If so, return the magic error code so that we
3407 * fallback to buffered I/O and attempt to complete the remainder of
3408 * the I/O. Any blocks that may have been allocated in preparation for
3409 * the direct I/O will be reused during buffered I/O.
3410 */
3411 if (flags & (IOMAP_WRITE | IOMAP_DIRECT) && written == 0)
3412 return -ENOTBLK;
3413
3414 return 0;
3415 }
3416
3417 const struct iomap_ops ext4_iomap_ops = {
3418 .iomap_begin = ext4_iomap_begin,
3419 .iomap_end = ext4_iomap_end,
3420 };
3421
3422 const struct iomap_ops ext4_iomap_overwrite_ops = {
3423 .iomap_begin = ext4_iomap_overwrite_begin,
3424 .iomap_end = ext4_iomap_end,
3425 };
3426
ext4_iomap_begin_report(struct inode * inode,loff_t offset,loff_t length,unsigned int flags,struct iomap * iomap,struct iomap * srcmap)3427 static int ext4_iomap_begin_report(struct inode *inode, loff_t offset,
3428 loff_t length, unsigned int flags,
3429 struct iomap *iomap, struct iomap *srcmap)
3430 {
3431 int ret;
3432 struct ext4_map_blocks map;
3433 u8 blkbits = inode->i_blkbits;
3434
3435 if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
3436 return -EINVAL;
3437
3438 if (ext4_has_inline_data(inode)) {
3439 ret = ext4_inline_data_iomap(inode, iomap);
3440 if (ret != -EAGAIN) {
3441 if (ret == 0 && offset >= iomap->length)
3442 ret = -ENOENT;
3443 return ret;
3444 }
3445 }
3446
3447 /*
3448 * Calculate the first and last logical block respectively.
3449 */
3450 map.m_lblk = offset >> blkbits;
3451 map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits,
3452 EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1;
3453
3454 /*
3455 * Fiemap callers may call for offset beyond s_bitmap_maxbytes.
3456 * So handle it here itself instead of querying ext4_map_blocks().
3457 * Since ext4_map_blocks() will warn about it and will return
3458 * -EIO error.
3459 */
3460 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
3461 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3462
3463 if (offset >= sbi->s_bitmap_maxbytes) {
3464 map.m_flags = 0;
3465 goto set_iomap;
3466 }
3467 }
3468
3469 ret = ext4_map_blocks(NULL, inode, &map, 0);
3470 if (ret < 0)
3471 return ret;
3472 set_iomap:
3473 ext4_set_iomap(inode, iomap, &map, offset, length, flags);
3474
3475 return 0;
3476 }
3477
3478 const struct iomap_ops ext4_iomap_report_ops = {
3479 .iomap_begin = ext4_iomap_begin_report,
3480 };
3481
3482 /*
3483 * For data=journal mode, folio should be marked dirty only when it was
3484 * writeably mapped. When that happens, it was already attached to the
3485 * transaction and marked as jbddirty (we take care of this in
3486 * ext4_page_mkwrite()). On transaction commit, we writeprotect page mappings
3487 * so we should have nothing to do here, except for the case when someone
3488 * had the page pinned and dirtied the page through this pin (e.g. by doing
3489 * direct IO to it). In that case we'd need to attach buffers here to the
3490 * transaction but we cannot due to lock ordering. We cannot just dirty the
3491 * folio and leave attached buffers clean, because the buffers' dirty state is
3492 * "definitive". We cannot just set the buffers dirty or jbddirty because all
3493 * the journalling code will explode. So what we do is to mark the folio
3494 * "pending dirty" and next time ext4_writepages() is called, attach buffers
3495 * to the transaction appropriately.
3496 */
ext4_journalled_dirty_folio(struct address_space * mapping,struct folio * folio)3497 static bool ext4_journalled_dirty_folio(struct address_space *mapping,
3498 struct folio *folio)
3499 {
3500 WARN_ON_ONCE(!folio_buffers(folio));
3501 if (folio_maybe_dma_pinned(folio))
3502 folio_set_checked(folio);
3503 return filemap_dirty_folio(mapping, folio);
3504 }
3505
ext4_dirty_folio(struct address_space * mapping,struct folio * folio)3506 static bool ext4_dirty_folio(struct address_space *mapping, struct folio *folio)
3507 {
3508 WARN_ON_ONCE(!folio_test_locked(folio) && !folio_test_dirty(folio));
3509 WARN_ON_ONCE(!folio_buffers(folio));
3510 return block_dirty_folio(mapping, folio);
3511 }
3512
ext4_iomap_swap_activate(struct swap_info_struct * sis,struct file * file,sector_t * span)3513 static int ext4_iomap_swap_activate(struct swap_info_struct *sis,
3514 struct file *file, sector_t *span)
3515 {
3516 return iomap_swapfile_activate(sis, file, span,
3517 &ext4_iomap_report_ops);
3518 }
3519
3520 static const struct address_space_operations ext4_aops = {
3521 .read_folio = ext4_read_folio,
3522 .readahead = ext4_readahead,
3523 .writepages = ext4_writepages,
3524 .write_begin = ext4_write_begin,
3525 .write_end = ext4_write_end,
3526 .dirty_folio = ext4_dirty_folio,
3527 .bmap = ext4_bmap,
3528 .invalidate_folio = ext4_invalidate_folio,
3529 .release_folio = ext4_release_folio,
3530 .migrate_folio = buffer_migrate_folio,
3531 .is_partially_uptodate = block_is_partially_uptodate,
3532 .error_remove_folio = generic_error_remove_folio,
3533 .swap_activate = ext4_iomap_swap_activate,
3534 };
3535
3536 static const struct address_space_operations ext4_journalled_aops = {
3537 .read_folio = ext4_read_folio,
3538 .readahead = ext4_readahead,
3539 .writepages = ext4_writepages,
3540 .write_begin = ext4_write_begin,
3541 .write_end = ext4_journalled_write_end,
3542 .dirty_folio = ext4_journalled_dirty_folio,
3543 .bmap = ext4_bmap,
3544 .invalidate_folio = ext4_journalled_invalidate_folio,
3545 .release_folio = ext4_release_folio,
3546 .migrate_folio = buffer_migrate_folio_norefs,
3547 .is_partially_uptodate = block_is_partially_uptodate,
3548 .error_remove_folio = generic_error_remove_folio,
3549 .swap_activate = ext4_iomap_swap_activate,
3550 };
3551
3552 static const struct address_space_operations ext4_da_aops = {
3553 .read_folio = ext4_read_folio,
3554 .readahead = ext4_readahead,
3555 .writepages = ext4_writepages,
3556 .write_begin = ext4_da_write_begin,
3557 .write_end = ext4_da_write_end,
3558 .dirty_folio = ext4_dirty_folio,
3559 .bmap = ext4_bmap,
3560 .invalidate_folio = ext4_invalidate_folio,
3561 .release_folio = ext4_release_folio,
3562 .migrate_folio = buffer_migrate_folio,
3563 .is_partially_uptodate = block_is_partially_uptodate,
3564 .error_remove_folio = generic_error_remove_folio,
3565 .swap_activate = ext4_iomap_swap_activate,
3566 };
3567
3568 static const struct address_space_operations ext4_dax_aops = {
3569 .writepages = ext4_dax_writepages,
3570 .dirty_folio = noop_dirty_folio,
3571 .bmap = ext4_bmap,
3572 .swap_activate = ext4_iomap_swap_activate,
3573 };
3574
ext4_set_aops(struct inode * inode)3575 void ext4_set_aops(struct inode *inode)
3576 {
3577 switch (ext4_inode_journal_mode(inode)) {
3578 case EXT4_INODE_ORDERED_DATA_MODE:
3579 case EXT4_INODE_WRITEBACK_DATA_MODE:
3580 break;
3581 case EXT4_INODE_JOURNAL_DATA_MODE:
3582 inode->i_mapping->a_ops = &ext4_journalled_aops;
3583 return;
3584 default:
3585 BUG();
3586 }
3587 if (IS_DAX(inode))
3588 inode->i_mapping->a_ops = &ext4_dax_aops;
3589 else if (test_opt(inode->i_sb, DELALLOC))
3590 inode->i_mapping->a_ops = &ext4_da_aops;
3591 else
3592 inode->i_mapping->a_ops = &ext4_aops;
3593 }
3594
3595 /*
3596 * Here we can't skip an unwritten buffer even though it usually reads zero
3597 * because it might have data in pagecache (eg, if called from ext4_zero_range,
3598 * ext4_punch_hole, etc) which needs to be properly zeroed out. Otherwise a
3599 * racing writeback can come later and flush the stale pagecache to disk.
3600 */
__ext4_block_zero_page_range(handle_t * handle,struct address_space * mapping,loff_t from,loff_t length)3601 static int __ext4_block_zero_page_range(handle_t *handle,
3602 struct address_space *mapping, loff_t from, loff_t length)
3603 {
3604 ext4_fsblk_t index = from >> PAGE_SHIFT;
3605 unsigned offset = from & (PAGE_SIZE-1);
3606 unsigned blocksize, pos;
3607 ext4_lblk_t iblock;
3608 struct inode *inode = mapping->host;
3609 struct buffer_head *bh;
3610 struct folio *folio;
3611 int err = 0;
3612
3613 folio = __filemap_get_folio(mapping, from >> PAGE_SHIFT,
3614 FGP_LOCK | FGP_ACCESSED | FGP_CREAT,
3615 mapping_gfp_constraint(mapping, ~__GFP_FS));
3616 if (IS_ERR(folio))
3617 return PTR_ERR(folio);
3618
3619 blocksize = inode->i_sb->s_blocksize;
3620
3621 iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
3622
3623 bh = folio_buffers(folio);
3624 if (!bh)
3625 bh = create_empty_buffers(folio, blocksize, 0);
3626
3627 /* Find the buffer that contains "offset" */
3628 pos = blocksize;
3629 while (offset >= pos) {
3630 bh = bh->b_this_page;
3631 iblock++;
3632 pos += blocksize;
3633 }
3634 if (buffer_freed(bh)) {
3635 BUFFER_TRACE(bh, "freed: skip");
3636 goto unlock;
3637 }
3638 if (!buffer_mapped(bh)) {
3639 BUFFER_TRACE(bh, "unmapped");
3640 ext4_get_block(inode, iblock, bh, 0);
3641 /* unmapped? It's a hole - nothing to do */
3642 if (!buffer_mapped(bh)) {
3643 BUFFER_TRACE(bh, "still unmapped");
3644 goto unlock;
3645 }
3646 }
3647
3648 /* Ok, it's mapped. Make sure it's up-to-date */
3649 if (folio_test_uptodate(folio))
3650 set_buffer_uptodate(bh);
3651
3652 if (!buffer_uptodate(bh)) {
3653 err = ext4_read_bh_lock(bh, 0, true);
3654 if (err)
3655 goto unlock;
3656 if (fscrypt_inode_uses_fs_layer_crypto(inode)) {
3657 /* We expect the key to be set. */
3658 BUG_ON(!fscrypt_has_encryption_key(inode));
3659 err = fscrypt_decrypt_pagecache_blocks(folio,
3660 blocksize,
3661 bh_offset(bh));
3662 if (err) {
3663 clear_buffer_uptodate(bh);
3664 goto unlock;
3665 }
3666 }
3667 }
3668 if (ext4_should_journal_data(inode)) {
3669 BUFFER_TRACE(bh, "get write access");
3670 err = ext4_journal_get_write_access(handle, inode->i_sb, bh,
3671 EXT4_JTR_NONE);
3672 if (err)
3673 goto unlock;
3674 }
3675 folio_zero_range(folio, offset, length);
3676 BUFFER_TRACE(bh, "zeroed end of block");
3677
3678 if (ext4_should_journal_data(inode)) {
3679 err = ext4_dirty_journalled_data(handle, bh);
3680 } else {
3681 err = 0;
3682 mark_buffer_dirty(bh);
3683 if (ext4_should_order_data(inode))
3684 err = ext4_jbd2_inode_add_write(handle, inode, from,
3685 length);
3686 }
3687
3688 unlock:
3689 folio_unlock(folio);
3690 folio_put(folio);
3691 return err;
3692 }
3693
3694 /*
3695 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3696 * starting from file offset 'from'. The range to be zero'd must
3697 * be contained with in one block. If the specified range exceeds
3698 * the end of the block it will be shortened to end of the block
3699 * that corresponds to 'from'
3700 */
ext4_block_zero_page_range(handle_t * handle,struct address_space * mapping,loff_t from,loff_t length)3701 static int ext4_block_zero_page_range(handle_t *handle,
3702 struct address_space *mapping, loff_t from, loff_t length)
3703 {
3704 struct inode *inode = mapping->host;
3705 unsigned offset = from & (PAGE_SIZE-1);
3706 unsigned blocksize = inode->i_sb->s_blocksize;
3707 unsigned max = blocksize - (offset & (blocksize - 1));
3708
3709 /*
3710 * correct length if it does not fall between
3711 * 'from' and the end of the block
3712 */
3713 if (length > max || length < 0)
3714 length = max;
3715
3716 if (IS_DAX(inode)) {
3717 return dax_zero_range(inode, from, length, NULL,
3718 &ext4_iomap_ops);
3719 }
3720 return __ext4_block_zero_page_range(handle, mapping, from, length);
3721 }
3722
3723 /*
3724 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3725 * up to the end of the block which corresponds to `from'.
3726 * This required during truncate. We need to physically zero the tail end
3727 * of that block so it doesn't yield old data if the file is later grown.
3728 */
ext4_block_truncate_page(handle_t * handle,struct address_space * mapping,loff_t from)3729 static int ext4_block_truncate_page(handle_t *handle,
3730 struct address_space *mapping, loff_t from)
3731 {
3732 unsigned offset = from & (PAGE_SIZE-1);
3733 unsigned length;
3734 unsigned blocksize;
3735 struct inode *inode = mapping->host;
3736
3737 /* If we are processing an encrypted inode during orphan list handling */
3738 if (IS_ENCRYPTED(inode) && !fscrypt_has_encryption_key(inode))
3739 return 0;
3740
3741 blocksize = inode->i_sb->s_blocksize;
3742 length = blocksize - (offset & (blocksize - 1));
3743
3744 return ext4_block_zero_page_range(handle, mapping, from, length);
3745 }
3746
ext4_zero_partial_blocks(handle_t * handle,struct inode * inode,loff_t lstart,loff_t length)3747 int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
3748 loff_t lstart, loff_t length)
3749 {
3750 struct super_block *sb = inode->i_sb;
3751 struct address_space *mapping = inode->i_mapping;
3752 unsigned partial_start, partial_end;
3753 ext4_fsblk_t start, end;
3754 loff_t byte_end = (lstart + length - 1);
3755 int err = 0;
3756
3757 partial_start = lstart & (sb->s_blocksize - 1);
3758 partial_end = byte_end & (sb->s_blocksize - 1);
3759
3760 start = lstart >> sb->s_blocksize_bits;
3761 end = byte_end >> sb->s_blocksize_bits;
3762
3763 /* Handle partial zero within the single block */
3764 if (start == end &&
3765 (partial_start || (partial_end != sb->s_blocksize - 1))) {
3766 err = ext4_block_zero_page_range(handle, mapping,
3767 lstart, length);
3768 return err;
3769 }
3770 /* Handle partial zero out on the start of the range */
3771 if (partial_start) {
3772 err = ext4_block_zero_page_range(handle, mapping,
3773 lstart, sb->s_blocksize);
3774 if (err)
3775 return err;
3776 }
3777 /* Handle partial zero out on the end of the range */
3778 if (partial_end != sb->s_blocksize - 1)
3779 err = ext4_block_zero_page_range(handle, mapping,
3780 byte_end - partial_end,
3781 partial_end + 1);
3782 return err;
3783 }
3784
ext4_can_truncate(struct inode * inode)3785 int ext4_can_truncate(struct inode *inode)
3786 {
3787 if (S_ISREG(inode->i_mode))
3788 return 1;
3789 if (S_ISDIR(inode->i_mode))
3790 return 1;
3791 if (S_ISLNK(inode->i_mode))
3792 return !ext4_inode_is_fast_symlink(inode);
3793 return 0;
3794 }
3795
3796 /*
3797 * We have to make sure i_disksize gets properly updated before we truncate
3798 * page cache due to hole punching or zero range. Otherwise i_disksize update
3799 * can get lost as it may have been postponed to submission of writeback but
3800 * that will never happen after we truncate page cache.
3801 */
ext4_update_disksize_before_punch(struct inode * inode,loff_t offset,loff_t len)3802 int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset,
3803 loff_t len)
3804 {
3805 handle_t *handle;
3806 int ret;
3807
3808 loff_t size = i_size_read(inode);
3809
3810 WARN_ON(!inode_is_locked(inode));
3811 if (offset > size || offset + len < size)
3812 return 0;
3813
3814 if (EXT4_I(inode)->i_disksize >= size)
3815 return 0;
3816
3817 handle = ext4_journal_start(inode, EXT4_HT_MISC, 1);
3818 if (IS_ERR(handle))
3819 return PTR_ERR(handle);
3820 ext4_update_i_disksize(inode, size);
3821 ret = ext4_mark_inode_dirty(handle, inode);
3822 ext4_journal_stop(handle);
3823
3824 return ret;
3825 }
3826
ext4_wait_dax_page(struct inode * inode)3827 static void ext4_wait_dax_page(struct inode *inode)
3828 {
3829 filemap_invalidate_unlock(inode->i_mapping);
3830 schedule();
3831 filemap_invalidate_lock(inode->i_mapping);
3832 }
3833
ext4_break_layouts(struct inode * inode)3834 int ext4_break_layouts(struct inode *inode)
3835 {
3836 struct page *page;
3837 int error;
3838
3839 if (WARN_ON_ONCE(!rwsem_is_locked(&inode->i_mapping->invalidate_lock)))
3840 return -EINVAL;
3841
3842 do {
3843 page = dax_layout_busy_page(inode->i_mapping);
3844 if (!page)
3845 return 0;
3846
3847 error = ___wait_var_event(&page->_refcount,
3848 atomic_read(&page->_refcount) == 1,
3849 TASK_INTERRUPTIBLE, 0, 0,
3850 ext4_wait_dax_page(inode));
3851 } while (error == 0);
3852
3853 return error;
3854 }
3855
3856 /*
3857 * ext4_punch_hole: punches a hole in a file by releasing the blocks
3858 * associated with the given offset and length
3859 *
3860 * @inode: File inode
3861 * @offset: The offset where the hole will begin
3862 * @len: The length of the hole
3863 *
3864 * Returns: 0 on success or negative on failure
3865 */
3866
ext4_punch_hole(struct file * file,loff_t offset,loff_t length)3867 int ext4_punch_hole(struct file *file, loff_t offset, loff_t length)
3868 {
3869 struct inode *inode = file_inode(file);
3870 struct super_block *sb = inode->i_sb;
3871 ext4_lblk_t first_block, stop_block;
3872 struct address_space *mapping = inode->i_mapping;
3873 loff_t first_block_offset, last_block_offset, max_length;
3874 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3875 handle_t *handle;
3876 unsigned int credits;
3877 int ret = 0, ret2 = 0;
3878
3879 trace_ext4_punch_hole(inode, offset, length, 0);
3880
3881 /*
3882 * Write out all dirty pages to avoid race conditions
3883 * Then release them.
3884 */
3885 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
3886 ret = filemap_write_and_wait_range(mapping, offset,
3887 offset + length - 1);
3888 if (ret)
3889 return ret;
3890 }
3891
3892 inode_lock(inode);
3893
3894 /* No need to punch hole beyond i_size */
3895 if (offset >= inode->i_size)
3896 goto out_mutex;
3897
3898 /*
3899 * If the hole extends beyond i_size, set the hole
3900 * to end after the page that contains i_size
3901 */
3902 if (offset + length > inode->i_size) {
3903 length = inode->i_size +
3904 PAGE_SIZE - (inode->i_size & (PAGE_SIZE - 1)) -
3905 offset;
3906 }
3907
3908 /*
3909 * For punch hole the length + offset needs to be within one block
3910 * before last range. Adjust the length if it goes beyond that limit.
3911 */
3912 max_length = sbi->s_bitmap_maxbytes - inode->i_sb->s_blocksize;
3913 if (offset + length > max_length)
3914 length = max_length - offset;
3915
3916 if (offset & (sb->s_blocksize - 1) ||
3917 (offset + length) & (sb->s_blocksize - 1)) {
3918 /*
3919 * Attach jinode to inode for jbd2 if we do any zeroing of
3920 * partial block
3921 */
3922 ret = ext4_inode_attach_jinode(inode);
3923 if (ret < 0)
3924 goto out_mutex;
3925
3926 }
3927
3928 /* Wait all existing dio workers, newcomers will block on i_rwsem */
3929 inode_dio_wait(inode);
3930
3931 ret = file_modified(file);
3932 if (ret)
3933 goto out_mutex;
3934
3935 /*
3936 * Prevent page faults from reinstantiating pages we have released from
3937 * page cache.
3938 */
3939 filemap_invalidate_lock(mapping);
3940
3941 ret = ext4_break_layouts(inode);
3942 if (ret)
3943 goto out_dio;
3944
3945 first_block_offset = round_up(offset, sb->s_blocksize);
3946 last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
3947
3948 /* Now release the pages and zero block aligned part of pages*/
3949 if (last_block_offset > first_block_offset) {
3950 ret = ext4_update_disksize_before_punch(inode, offset, length);
3951 if (ret)
3952 goto out_dio;
3953 truncate_pagecache_range(inode, first_block_offset,
3954 last_block_offset);
3955 }
3956
3957 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3958 credits = ext4_writepage_trans_blocks(inode);
3959 else
3960 credits = ext4_blocks_for_truncate(inode);
3961 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
3962 if (IS_ERR(handle)) {
3963 ret = PTR_ERR(handle);
3964 ext4_std_error(sb, ret);
3965 goto out_dio;
3966 }
3967
3968 ret = ext4_zero_partial_blocks(handle, inode, offset,
3969 length);
3970 if (ret)
3971 goto out_stop;
3972
3973 first_block = (offset + sb->s_blocksize - 1) >>
3974 EXT4_BLOCK_SIZE_BITS(sb);
3975 stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
3976
3977 /* If there are blocks to remove, do it */
3978 if (stop_block > first_block) {
3979 ext4_lblk_t hole_len = stop_block - first_block;
3980
3981 down_write(&EXT4_I(inode)->i_data_sem);
3982 ext4_discard_preallocations(inode);
3983
3984 ext4_es_remove_extent(inode, first_block, hole_len);
3985
3986 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3987 ret = ext4_ext_remove_space(inode, first_block,
3988 stop_block - 1);
3989 else
3990 ret = ext4_ind_remove_space(handle, inode, first_block,
3991 stop_block);
3992
3993 ext4_es_insert_extent(inode, first_block, hole_len, ~0,
3994 EXTENT_STATUS_HOLE);
3995 up_write(&EXT4_I(inode)->i_data_sem);
3996 }
3997 ext4_fc_track_range(handle, inode, first_block, stop_block);
3998 if (IS_SYNC(inode))
3999 ext4_handle_sync(handle);
4000
4001 inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
4002 ret2 = ext4_mark_inode_dirty(handle, inode);
4003 if (unlikely(ret2))
4004 ret = ret2;
4005 if (ret >= 0)
4006 ext4_update_inode_fsync_trans(handle, inode, 1);
4007 out_stop:
4008 ext4_journal_stop(handle);
4009 out_dio:
4010 filemap_invalidate_unlock(mapping);
4011 out_mutex:
4012 inode_unlock(inode);
4013 return ret;
4014 }
4015
ext4_inode_attach_jinode(struct inode * inode)4016 int ext4_inode_attach_jinode(struct inode *inode)
4017 {
4018 struct ext4_inode_info *ei = EXT4_I(inode);
4019 struct jbd2_inode *jinode;
4020
4021 if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
4022 return 0;
4023
4024 jinode = jbd2_alloc_inode(GFP_KERNEL);
4025 spin_lock(&inode->i_lock);
4026 if (!ei->jinode) {
4027 if (!jinode) {
4028 spin_unlock(&inode->i_lock);
4029 return -ENOMEM;
4030 }
4031 ei->jinode = jinode;
4032 jbd2_journal_init_jbd_inode(ei->jinode, inode);
4033 jinode = NULL;
4034 }
4035 spin_unlock(&inode->i_lock);
4036 if (unlikely(jinode != NULL))
4037 jbd2_free_inode(jinode);
4038 return 0;
4039 }
4040
4041 /*
4042 * ext4_truncate()
4043 *
4044 * We block out ext4_get_block() block instantiations across the entire
4045 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4046 * simultaneously on behalf of the same inode.
4047 *
4048 * As we work through the truncate and commit bits of it to the journal there
4049 * is one core, guiding principle: the file's tree must always be consistent on
4050 * disk. We must be able to restart the truncate after a crash.
4051 *
4052 * The file's tree may be transiently inconsistent in memory (although it
4053 * probably isn't), but whenever we close off and commit a journal transaction,
4054 * the contents of (the filesystem + the journal) must be consistent and
4055 * restartable. It's pretty simple, really: bottom up, right to left (although
4056 * left-to-right works OK too).
4057 *
4058 * Note that at recovery time, journal replay occurs *before* the restart of
4059 * truncate against the orphan inode list.
4060 *
4061 * The committed inode has the new, desired i_size (which is the same as
4062 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4063 * that this inode's truncate did not complete and it will again call
4064 * ext4_truncate() to have another go. So there will be instantiated blocks
4065 * to the right of the truncation point in a crashed ext4 filesystem. But
4066 * that's fine - as long as they are linked from the inode, the post-crash
4067 * ext4_truncate() run will find them and release them.
4068 */
ext4_truncate(struct inode * inode)4069 int ext4_truncate(struct inode *inode)
4070 {
4071 struct ext4_inode_info *ei = EXT4_I(inode);
4072 unsigned int credits;
4073 int err = 0, err2;
4074 handle_t *handle;
4075 struct address_space *mapping = inode->i_mapping;
4076
4077 /*
4078 * There is a possibility that we're either freeing the inode
4079 * or it's a completely new inode. In those cases we might not
4080 * have i_rwsem locked because it's not necessary.
4081 */
4082 if (!(inode->i_state & (I_NEW|I_FREEING)))
4083 WARN_ON(!inode_is_locked(inode));
4084 trace_ext4_truncate_enter(inode);
4085
4086 if (!ext4_can_truncate(inode))
4087 goto out_trace;
4088
4089 if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4090 ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
4091
4092 if (ext4_has_inline_data(inode)) {
4093 int has_inline = 1;
4094
4095 err = ext4_inline_data_truncate(inode, &has_inline);
4096 if (err || has_inline)
4097 goto out_trace;
4098 }
4099
4100 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
4101 if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
4102 err = ext4_inode_attach_jinode(inode);
4103 if (err)
4104 goto out_trace;
4105 }
4106
4107 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4108 credits = ext4_writepage_trans_blocks(inode);
4109 else
4110 credits = ext4_blocks_for_truncate(inode);
4111
4112 handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4113 if (IS_ERR(handle)) {
4114 err = PTR_ERR(handle);
4115 goto out_trace;
4116 }
4117
4118 if (inode->i_size & (inode->i_sb->s_blocksize - 1))
4119 ext4_block_truncate_page(handle, mapping, inode->i_size);
4120
4121 /*
4122 * We add the inode to the orphan list, so that if this
4123 * truncate spans multiple transactions, and we crash, we will
4124 * resume the truncate when the filesystem recovers. It also
4125 * marks the inode dirty, to catch the new size.
4126 *
4127 * Implication: the file must always be in a sane, consistent
4128 * truncatable state while each transaction commits.
4129 */
4130 err = ext4_orphan_add(handle, inode);
4131 if (err)
4132 goto out_stop;
4133
4134 down_write(&EXT4_I(inode)->i_data_sem);
4135
4136 ext4_discard_preallocations(inode);
4137
4138 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4139 err = ext4_ext_truncate(handle, inode);
4140 else
4141 ext4_ind_truncate(handle, inode);
4142
4143 up_write(&ei->i_data_sem);
4144 if (err)
4145 goto out_stop;
4146
4147 if (IS_SYNC(inode))
4148 ext4_handle_sync(handle);
4149
4150 out_stop:
4151 /*
4152 * If this was a simple ftruncate() and the file will remain alive,
4153 * then we need to clear up the orphan record which we created above.
4154 * However, if this was a real unlink then we were called by
4155 * ext4_evict_inode(), and we allow that function to clean up the
4156 * orphan info for us.
4157 */
4158 if (inode->i_nlink)
4159 ext4_orphan_del(handle, inode);
4160
4161 inode_set_mtime_to_ts(inode, inode_set_ctime_current(inode));
4162 err2 = ext4_mark_inode_dirty(handle, inode);
4163 if (unlikely(err2 && !err))
4164 err = err2;
4165 ext4_journal_stop(handle);
4166
4167 out_trace:
4168 trace_ext4_truncate_exit(inode);
4169 return err;
4170 }
4171
ext4_inode_peek_iversion(const struct inode * inode)4172 static inline u64 ext4_inode_peek_iversion(const struct inode *inode)
4173 {
4174 if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4175 return inode_peek_iversion_raw(inode);
4176 else
4177 return inode_peek_iversion(inode);
4178 }
4179
ext4_inode_blocks_set(struct ext4_inode * raw_inode,struct ext4_inode_info * ei)4180 static int ext4_inode_blocks_set(struct ext4_inode *raw_inode,
4181 struct ext4_inode_info *ei)
4182 {
4183 struct inode *inode = &(ei->vfs_inode);
4184 u64 i_blocks = READ_ONCE(inode->i_blocks);
4185 struct super_block *sb = inode->i_sb;
4186
4187 if (i_blocks <= ~0U) {
4188 /*
4189 * i_blocks can be represented in a 32 bit variable
4190 * as multiple of 512 bytes
4191 */
4192 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4193 raw_inode->i_blocks_high = 0;
4194 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4195 return 0;
4196 }
4197
4198 /*
4199 * This should never happen since sb->s_maxbytes should not have
4200 * allowed this, sb->s_maxbytes was set according to the huge_file
4201 * feature in ext4_fill_super().
4202 */
4203 if (!ext4_has_feature_huge_file(sb))
4204 return -EFSCORRUPTED;
4205
4206 if (i_blocks <= 0xffffffffffffULL) {
4207 /*
4208 * i_blocks can be represented in a 48 bit variable
4209 * as multiple of 512 bytes
4210 */
4211 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4212 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4213 ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4214 } else {
4215 ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4216 /* i_block is stored in file system block size */
4217 i_blocks = i_blocks >> (inode->i_blkbits - 9);
4218 raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4219 raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4220 }
4221 return 0;
4222 }
4223
ext4_fill_raw_inode(struct inode * inode,struct ext4_inode * raw_inode)4224 static int ext4_fill_raw_inode(struct inode *inode, struct ext4_inode *raw_inode)
4225 {
4226 struct ext4_inode_info *ei = EXT4_I(inode);
4227 uid_t i_uid;
4228 gid_t i_gid;
4229 projid_t i_projid;
4230 int block;
4231 int err;
4232
4233 err = ext4_inode_blocks_set(raw_inode, ei);
4234
4235 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4236 i_uid = i_uid_read(inode);
4237 i_gid = i_gid_read(inode);
4238 i_projid = from_kprojid(&init_user_ns, ei->i_projid);
4239 if (!(test_opt(inode->i_sb, NO_UID32))) {
4240 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4241 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4242 /*
4243 * Fix up interoperability with old kernels. Otherwise,
4244 * old inodes get re-used with the upper 16 bits of the
4245 * uid/gid intact.
4246 */
4247 if (ei->i_dtime && list_empty(&ei->i_orphan)) {
4248 raw_inode->i_uid_high = 0;
4249 raw_inode->i_gid_high = 0;
4250 } else {
4251 raw_inode->i_uid_high =
4252 cpu_to_le16(high_16_bits(i_uid));
4253 raw_inode->i_gid_high =
4254 cpu_to_le16(high_16_bits(i_gid));
4255 }
4256 } else {
4257 raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4258 raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4259 raw_inode->i_uid_high = 0;
4260 raw_inode->i_gid_high = 0;
4261 }
4262 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4263
4264 EXT4_INODE_SET_CTIME(inode, raw_inode);
4265 EXT4_INODE_SET_MTIME(inode, raw_inode);
4266 EXT4_INODE_SET_ATIME(inode, raw_inode);
4267 EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4268
4269 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4270 raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4271 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
4272 raw_inode->i_file_acl_high =
4273 cpu_to_le16(ei->i_file_acl >> 32);
4274 raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4275 ext4_isize_set(raw_inode, ei->i_disksize);
4276
4277 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4278 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4279 if (old_valid_dev(inode->i_rdev)) {
4280 raw_inode->i_block[0] =
4281 cpu_to_le32(old_encode_dev(inode->i_rdev));
4282 raw_inode->i_block[1] = 0;
4283 } else {
4284 raw_inode->i_block[0] = 0;
4285 raw_inode->i_block[1] =
4286 cpu_to_le32(new_encode_dev(inode->i_rdev));
4287 raw_inode->i_block[2] = 0;
4288 }
4289 } else if (!ext4_has_inline_data(inode)) {
4290 for (block = 0; block < EXT4_N_BLOCKS; block++)
4291 raw_inode->i_block[block] = ei->i_data[block];
4292 }
4293
4294 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4295 u64 ivers = ext4_inode_peek_iversion(inode);
4296
4297 raw_inode->i_disk_version = cpu_to_le32(ivers);
4298 if (ei->i_extra_isize) {
4299 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4300 raw_inode->i_version_hi =
4301 cpu_to_le32(ivers >> 32);
4302 raw_inode->i_extra_isize =
4303 cpu_to_le16(ei->i_extra_isize);
4304 }
4305 }
4306
4307 if (i_projid != EXT4_DEF_PROJID &&
4308 !ext4_has_feature_project(inode->i_sb))
4309 err = err ?: -EFSCORRUPTED;
4310
4311 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4312 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4313 raw_inode->i_projid = cpu_to_le32(i_projid);
4314
4315 ext4_inode_csum_set(inode, raw_inode, ei);
4316 return err;
4317 }
4318
4319 /*
4320 * ext4_get_inode_loc returns with an extra refcount against the inode's
4321 * underlying buffer_head on success. If we pass 'inode' and it does not
4322 * have in-inode xattr, we have all inode data in memory that is needed
4323 * to recreate the on-disk version of this inode.
4324 */
__ext4_get_inode_loc(struct super_block * sb,unsigned long ino,struct inode * inode,struct ext4_iloc * iloc,ext4_fsblk_t * ret_block)4325 static int __ext4_get_inode_loc(struct super_block *sb, unsigned long ino,
4326 struct inode *inode, struct ext4_iloc *iloc,
4327 ext4_fsblk_t *ret_block)
4328 {
4329 struct ext4_group_desc *gdp;
4330 struct buffer_head *bh;
4331 ext4_fsblk_t block;
4332 struct blk_plug plug;
4333 int inodes_per_block, inode_offset;
4334
4335 iloc->bh = NULL;
4336 if (ino < EXT4_ROOT_INO ||
4337 ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
4338 return -EFSCORRUPTED;
4339
4340 iloc->block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
4341 gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4342 if (!gdp)
4343 return -EIO;
4344
4345 /*
4346 * Figure out the offset within the block group inode table
4347 */
4348 inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4349 inode_offset = ((ino - 1) %
4350 EXT4_INODES_PER_GROUP(sb));
4351 iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4352
4353 block = ext4_inode_table(sb, gdp);
4354 if ((block <= le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) ||
4355 (block >= ext4_blocks_count(EXT4_SB(sb)->s_es))) {
4356 ext4_error(sb, "Invalid inode table block %llu in "
4357 "block_group %u", block, iloc->block_group);
4358 return -EFSCORRUPTED;
4359 }
4360 block += (inode_offset / inodes_per_block);
4361
4362 bh = sb_getblk(sb, block);
4363 if (unlikely(!bh))
4364 return -ENOMEM;
4365 if (ext4_buffer_uptodate(bh))
4366 goto has_buffer;
4367
4368 lock_buffer(bh);
4369 if (ext4_buffer_uptodate(bh)) {
4370 /* Someone brought it uptodate while we waited */
4371 unlock_buffer(bh);
4372 goto has_buffer;
4373 }
4374
4375 /*
4376 * If we have all information of the inode in memory and this
4377 * is the only valid inode in the block, we need not read the
4378 * block.
4379 */
4380 if (inode && !ext4_test_inode_state(inode, EXT4_STATE_XATTR)) {
4381 struct buffer_head *bitmap_bh;
4382 int i, start;
4383
4384 start = inode_offset & ~(inodes_per_block - 1);
4385
4386 /* Is the inode bitmap in cache? */
4387 bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4388 if (unlikely(!bitmap_bh))
4389 goto make_io;
4390
4391 /*
4392 * If the inode bitmap isn't in cache then the
4393 * optimisation may end up performing two reads instead
4394 * of one, so skip it.
4395 */
4396 if (!buffer_uptodate(bitmap_bh)) {
4397 brelse(bitmap_bh);
4398 goto make_io;
4399 }
4400 for (i = start; i < start + inodes_per_block; i++) {
4401 if (i == inode_offset)
4402 continue;
4403 if (ext4_test_bit(i, bitmap_bh->b_data))
4404 break;
4405 }
4406 brelse(bitmap_bh);
4407 if (i == start + inodes_per_block) {
4408 struct ext4_inode *raw_inode =
4409 (struct ext4_inode *) (bh->b_data + iloc->offset);
4410
4411 /* all other inodes are free, so skip I/O */
4412 memset(bh->b_data, 0, bh->b_size);
4413 if (!ext4_test_inode_state(inode, EXT4_STATE_NEW))
4414 ext4_fill_raw_inode(inode, raw_inode);
4415 set_buffer_uptodate(bh);
4416 unlock_buffer(bh);
4417 goto has_buffer;
4418 }
4419 }
4420
4421 make_io:
4422 /*
4423 * If we need to do any I/O, try to pre-readahead extra
4424 * blocks from the inode table.
4425 */
4426 blk_start_plug(&plug);
4427 if (EXT4_SB(sb)->s_inode_readahead_blks) {
4428 ext4_fsblk_t b, end, table;
4429 unsigned num;
4430 __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
4431
4432 table = ext4_inode_table(sb, gdp);
4433 /* s_inode_readahead_blks is always a power of 2 */
4434 b = block & ~((ext4_fsblk_t) ra_blks - 1);
4435 if (table > b)
4436 b = table;
4437 end = b + ra_blks;
4438 num = EXT4_INODES_PER_GROUP(sb);
4439 if (ext4_has_group_desc_csum(sb))
4440 num -= ext4_itable_unused_count(sb, gdp);
4441 table += num / inodes_per_block;
4442 if (end > table)
4443 end = table;
4444 while (b <= end)
4445 ext4_sb_breadahead_unmovable(sb, b++);
4446 }
4447
4448 /*
4449 * There are other valid inodes in the buffer, this inode
4450 * has in-inode xattrs, or we don't have this inode in memory.
4451 * Read the block from disk.
4452 */
4453 trace_ext4_load_inode(sb, ino);
4454 ext4_read_bh_nowait(bh, REQ_META | REQ_PRIO, NULL);
4455 blk_finish_plug(&plug);
4456 wait_on_buffer(bh);
4457 ext4_simulate_fail_bh(sb, bh, EXT4_SIM_INODE_EIO);
4458 if (!buffer_uptodate(bh)) {
4459 if (ret_block)
4460 *ret_block = block;
4461 brelse(bh);
4462 return -EIO;
4463 }
4464 has_buffer:
4465 iloc->bh = bh;
4466 return 0;
4467 }
4468
__ext4_get_inode_loc_noinmem(struct inode * inode,struct ext4_iloc * iloc)4469 static int __ext4_get_inode_loc_noinmem(struct inode *inode,
4470 struct ext4_iloc *iloc)
4471 {
4472 ext4_fsblk_t err_blk = 0;
4473 int ret;
4474
4475 ret = __ext4_get_inode_loc(inode->i_sb, inode->i_ino, NULL, iloc,
4476 &err_blk);
4477
4478 if (ret == -EIO)
4479 ext4_error_inode_block(inode, err_blk, EIO,
4480 "unable to read itable block");
4481
4482 return ret;
4483 }
4484
ext4_get_inode_loc(struct inode * inode,struct ext4_iloc * iloc)4485 int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4486 {
4487 ext4_fsblk_t err_blk = 0;
4488 int ret;
4489
4490 ret = __ext4_get_inode_loc(inode->i_sb, inode->i_ino, inode, iloc,
4491 &err_blk);
4492
4493 if (ret == -EIO)
4494 ext4_error_inode_block(inode, err_blk, EIO,
4495 "unable to read itable block");
4496
4497 return ret;
4498 }
4499
4500
ext4_get_fc_inode_loc(struct super_block * sb,unsigned long ino,struct ext4_iloc * iloc)4501 int ext4_get_fc_inode_loc(struct super_block *sb, unsigned long ino,
4502 struct ext4_iloc *iloc)
4503 {
4504 return __ext4_get_inode_loc(sb, ino, NULL, iloc, NULL);
4505 }
4506
ext4_should_enable_dax(struct inode * inode)4507 static bool ext4_should_enable_dax(struct inode *inode)
4508 {
4509 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4510
4511 if (test_opt2(inode->i_sb, DAX_NEVER))
4512 return false;
4513 if (!S_ISREG(inode->i_mode))
4514 return false;
4515 if (ext4_should_journal_data(inode))
4516 return false;
4517 if (ext4_has_inline_data(inode))
4518 return false;
4519 if (ext4_test_inode_flag(inode, EXT4_INODE_ENCRYPT))
4520 return false;
4521 if (ext4_test_inode_flag(inode, EXT4_INODE_VERITY))
4522 return false;
4523 if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags))
4524 return false;
4525 if (test_opt(inode->i_sb, DAX_ALWAYS))
4526 return true;
4527
4528 return ext4_test_inode_flag(inode, EXT4_INODE_DAX);
4529 }
4530
ext4_set_inode_flags(struct inode * inode,bool init)4531 void ext4_set_inode_flags(struct inode *inode, bool init)
4532 {
4533 unsigned int flags = EXT4_I(inode)->i_flags;
4534 unsigned int new_fl = 0;
4535
4536 WARN_ON_ONCE(IS_DAX(inode) && init);
4537
4538 if (flags & EXT4_SYNC_FL)
4539 new_fl |= S_SYNC;
4540 if (flags & EXT4_APPEND_FL)
4541 new_fl |= S_APPEND;
4542 if (flags & EXT4_IMMUTABLE_FL)
4543 new_fl |= S_IMMUTABLE;
4544 if (flags & EXT4_NOATIME_FL)
4545 new_fl |= S_NOATIME;
4546 if (flags & EXT4_DIRSYNC_FL)
4547 new_fl |= S_DIRSYNC;
4548
4549 /* Because of the way inode_set_flags() works we must preserve S_DAX
4550 * here if already set. */
4551 new_fl |= (inode->i_flags & S_DAX);
4552 if (init && ext4_should_enable_dax(inode))
4553 new_fl |= S_DAX;
4554
4555 if (flags & EXT4_ENCRYPT_FL)
4556 new_fl |= S_ENCRYPTED;
4557 if (flags & EXT4_CASEFOLD_FL)
4558 new_fl |= S_CASEFOLD;
4559 if (flags & EXT4_VERITY_FL)
4560 new_fl |= S_VERITY;
4561 inode_set_flags(inode, new_fl,
4562 S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX|
4563 S_ENCRYPTED|S_CASEFOLD|S_VERITY);
4564 }
4565
ext4_inode_blocks(struct ext4_inode * raw_inode,struct ext4_inode_info * ei)4566 static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4567 struct ext4_inode_info *ei)
4568 {
4569 blkcnt_t i_blocks ;
4570 struct inode *inode = &(ei->vfs_inode);
4571 struct super_block *sb = inode->i_sb;
4572
4573 if (ext4_has_feature_huge_file(sb)) {
4574 /* we are using combined 48 bit field */
4575 i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4576 le32_to_cpu(raw_inode->i_blocks_lo);
4577 if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4578 /* i_blocks represent file system block size */
4579 return i_blocks << (inode->i_blkbits - 9);
4580 } else {
4581 return i_blocks;
4582 }
4583 } else {
4584 return le32_to_cpu(raw_inode->i_blocks_lo);
4585 }
4586 }
4587
ext4_iget_extra_inode(struct inode * inode,struct ext4_inode * raw_inode,struct ext4_inode_info * ei)4588 static inline int ext4_iget_extra_inode(struct inode *inode,
4589 struct ext4_inode *raw_inode,
4590 struct ext4_inode_info *ei)
4591 {
4592 __le32 *magic = (void *)raw_inode +
4593 EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4594
4595 if (EXT4_INODE_HAS_XATTR_SPACE(inode) &&
4596 *magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4597 int err;
4598
4599 ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4600 err = ext4_find_inline_data_nolock(inode);
4601 if (!err && ext4_has_inline_data(inode))
4602 ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
4603 return err;
4604 } else
4605 EXT4_I(inode)->i_inline_off = 0;
4606 return 0;
4607 }
4608
ext4_get_projid(struct inode * inode,kprojid_t * projid)4609 int ext4_get_projid(struct inode *inode, kprojid_t *projid)
4610 {
4611 if (!ext4_has_feature_project(inode->i_sb))
4612 return -EOPNOTSUPP;
4613 *projid = EXT4_I(inode)->i_projid;
4614 return 0;
4615 }
4616
4617 /*
4618 * ext4 has self-managed i_version for ea inodes, it stores the lower 32bit of
4619 * refcount in i_version, so use raw values if inode has EXT4_EA_INODE_FL flag
4620 * set.
4621 */
ext4_inode_set_iversion_queried(struct inode * inode,u64 val)4622 static inline void ext4_inode_set_iversion_queried(struct inode *inode, u64 val)
4623 {
4624 if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4625 inode_set_iversion_raw(inode, val);
4626 else
4627 inode_set_iversion_queried(inode, val);
4628 }
4629
check_igot_inode(struct inode * inode,ext4_iget_flags flags)4630 static const char *check_igot_inode(struct inode *inode, ext4_iget_flags flags)
4631
4632 {
4633 if (flags & EXT4_IGET_EA_INODE) {
4634 if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4635 return "missing EA_INODE flag";
4636 if (ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
4637 EXT4_I(inode)->i_file_acl)
4638 return "ea_inode with extended attributes";
4639 } else {
4640 if ((EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4641 return "unexpected EA_INODE flag";
4642 }
4643 if (is_bad_inode(inode) && !(flags & EXT4_IGET_BAD))
4644 return "unexpected bad inode w/o EXT4_IGET_BAD";
4645 return NULL;
4646 }
4647
__ext4_iget(struct super_block * sb,unsigned long ino,ext4_iget_flags flags,const char * function,unsigned int line)4648 struct inode *__ext4_iget(struct super_block *sb, unsigned long ino,
4649 ext4_iget_flags flags, const char *function,
4650 unsigned int line)
4651 {
4652 struct ext4_iloc iloc;
4653 struct ext4_inode *raw_inode;
4654 struct ext4_inode_info *ei;
4655 struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4656 struct inode *inode;
4657 const char *err_str;
4658 journal_t *journal = EXT4_SB(sb)->s_journal;
4659 long ret;
4660 loff_t size;
4661 int block;
4662 uid_t i_uid;
4663 gid_t i_gid;
4664 projid_t i_projid;
4665
4666 if ((!(flags & EXT4_IGET_SPECIAL) &&
4667 ((ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO) ||
4668 ino == le32_to_cpu(es->s_usr_quota_inum) ||
4669 ino == le32_to_cpu(es->s_grp_quota_inum) ||
4670 ino == le32_to_cpu(es->s_prj_quota_inum) ||
4671 ino == le32_to_cpu(es->s_orphan_file_inum))) ||
4672 (ino < EXT4_ROOT_INO) ||
4673 (ino > le32_to_cpu(es->s_inodes_count))) {
4674 if (flags & EXT4_IGET_HANDLE)
4675 return ERR_PTR(-ESTALE);
4676 __ext4_error(sb, function, line, false, EFSCORRUPTED, 0,
4677 "inode #%lu: comm %s: iget: illegal inode #",
4678 ino, current->comm);
4679 return ERR_PTR(-EFSCORRUPTED);
4680 }
4681
4682 inode = iget_locked(sb, ino);
4683 if (!inode)
4684 return ERR_PTR(-ENOMEM);
4685 if (!(inode->i_state & I_NEW)) {
4686 if ((err_str = check_igot_inode(inode, flags)) != NULL) {
4687 ext4_error_inode(inode, function, line, 0, err_str);
4688 iput(inode);
4689 return ERR_PTR(-EFSCORRUPTED);
4690 }
4691 return inode;
4692 }
4693
4694 ei = EXT4_I(inode);
4695 iloc.bh = NULL;
4696
4697 ret = __ext4_get_inode_loc_noinmem(inode, &iloc);
4698 if (ret < 0)
4699 goto bad_inode;
4700 raw_inode = ext4_raw_inode(&iloc);
4701
4702 if ((flags & EXT4_IGET_HANDLE) &&
4703 (raw_inode->i_links_count == 0) && (raw_inode->i_mode == 0)) {
4704 ret = -ESTALE;
4705 goto bad_inode;
4706 }
4707
4708 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4709 ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4710 if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4711 EXT4_INODE_SIZE(inode->i_sb) ||
4712 (ei->i_extra_isize & 3)) {
4713 ext4_error_inode(inode, function, line, 0,
4714 "iget: bad extra_isize %u "
4715 "(inode size %u)",
4716 ei->i_extra_isize,
4717 EXT4_INODE_SIZE(inode->i_sb));
4718 ret = -EFSCORRUPTED;
4719 goto bad_inode;
4720 }
4721 } else
4722 ei->i_extra_isize = 0;
4723
4724 /* Precompute checksum seed for inode metadata */
4725 if (ext4_has_metadata_csum(sb)) {
4726 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4727 __u32 csum;
4728 __le32 inum = cpu_to_le32(inode->i_ino);
4729 __le32 gen = raw_inode->i_generation;
4730 csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4731 sizeof(inum));
4732 ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4733 sizeof(gen));
4734 }
4735
4736 if ((!ext4_inode_csum_verify(inode, raw_inode, ei) ||
4737 ext4_simulate_fail(sb, EXT4_SIM_INODE_CRC)) &&
4738 (!(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY))) {
4739 ext4_error_inode_err(inode, function, line, 0,
4740 EFSBADCRC, "iget: checksum invalid");
4741 ret = -EFSBADCRC;
4742 goto bad_inode;
4743 }
4744
4745 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4746 i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4747 i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4748 if (ext4_has_feature_project(sb) &&
4749 EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4750 EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4751 i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid);
4752 else
4753 i_projid = EXT4_DEF_PROJID;
4754
4755 if (!(test_opt(inode->i_sb, NO_UID32))) {
4756 i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4757 i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4758 }
4759 i_uid_write(inode, i_uid);
4760 i_gid_write(inode, i_gid);
4761 ei->i_projid = make_kprojid(&init_user_ns, i_projid);
4762 set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4763
4764 ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4765 ei->i_inline_off = 0;
4766 ei->i_dir_start_lookup = 0;
4767 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4768 /* We now have enough fields to check if the inode was active or not.
4769 * This is needed because nfsd might try to access dead inodes
4770 * the test is that same one that e2fsck uses
4771 * NeilBrown 1999oct15
4772 */
4773 if (inode->i_nlink == 0) {
4774 if ((inode->i_mode == 0 || flags & EXT4_IGET_SPECIAL ||
4775 !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4776 ino != EXT4_BOOT_LOADER_INO) {
4777 /* this inode is deleted or unallocated */
4778 if (flags & EXT4_IGET_SPECIAL) {
4779 ext4_error_inode(inode, function, line, 0,
4780 "iget: special inode unallocated");
4781 ret = -EFSCORRUPTED;
4782 } else
4783 ret = -ESTALE;
4784 goto bad_inode;
4785 }
4786 /* The only unlinked inodes we let through here have
4787 * valid i_mode and are being read by the orphan
4788 * recovery code: that's fine, we're about to complete
4789 * the process of deleting those.
4790 * OR it is the EXT4_BOOT_LOADER_INO which is
4791 * not initialized on a new filesystem. */
4792 }
4793 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4794 ext4_set_inode_flags(inode, true);
4795 inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4796 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4797 if (ext4_has_feature_64bit(sb))
4798 ei->i_file_acl |=
4799 ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4800 inode->i_size = ext4_isize(sb, raw_inode);
4801 if ((size = i_size_read(inode)) < 0) {
4802 ext4_error_inode(inode, function, line, 0,
4803 "iget: bad i_size value: %lld", size);
4804 ret = -EFSCORRUPTED;
4805 goto bad_inode;
4806 }
4807 /*
4808 * If dir_index is not enabled but there's dir with INDEX flag set,
4809 * we'd normally treat htree data as empty space. But with metadata
4810 * checksumming that corrupts checksums so forbid that.
4811 */
4812 if (!ext4_has_feature_dir_index(sb) && ext4_has_metadata_csum(sb) &&
4813 ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) {
4814 ext4_error_inode(inode, function, line, 0,
4815 "iget: Dir with htree data on filesystem without dir_index feature.");
4816 ret = -EFSCORRUPTED;
4817 goto bad_inode;
4818 }
4819 ei->i_disksize = inode->i_size;
4820 #ifdef CONFIG_QUOTA
4821 ei->i_reserved_quota = 0;
4822 #endif
4823 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4824 ei->i_block_group = iloc.block_group;
4825 ei->i_last_alloc_group = ~0;
4826 /*
4827 * NOTE! The in-memory inode i_data array is in little-endian order
4828 * even on big-endian machines: we do NOT byteswap the block numbers!
4829 */
4830 for (block = 0; block < EXT4_N_BLOCKS; block++)
4831 ei->i_data[block] = raw_inode->i_block[block];
4832 INIT_LIST_HEAD(&ei->i_orphan);
4833 ext4_fc_init_inode(&ei->vfs_inode);
4834
4835 /*
4836 * Set transaction id's of transactions that have to be committed
4837 * to finish f[data]sync. We set them to currently running transaction
4838 * as we cannot be sure that the inode or some of its metadata isn't
4839 * part of the transaction - the inode could have been reclaimed and
4840 * now it is reread from disk.
4841 */
4842 if (journal) {
4843 transaction_t *transaction;
4844 tid_t tid;
4845
4846 read_lock(&journal->j_state_lock);
4847 if (journal->j_running_transaction)
4848 transaction = journal->j_running_transaction;
4849 else
4850 transaction = journal->j_committing_transaction;
4851 if (transaction)
4852 tid = transaction->t_tid;
4853 else
4854 tid = journal->j_commit_sequence;
4855 read_unlock(&journal->j_state_lock);
4856 ei->i_sync_tid = tid;
4857 ei->i_datasync_tid = tid;
4858 }
4859
4860 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4861 if (ei->i_extra_isize == 0) {
4862 /* The extra space is currently unused. Use it. */
4863 BUILD_BUG_ON(sizeof(struct ext4_inode) & 3);
4864 ei->i_extra_isize = sizeof(struct ext4_inode) -
4865 EXT4_GOOD_OLD_INODE_SIZE;
4866 } else {
4867 ret = ext4_iget_extra_inode(inode, raw_inode, ei);
4868 if (ret)
4869 goto bad_inode;
4870 }
4871 }
4872
4873 EXT4_INODE_GET_CTIME(inode, raw_inode);
4874 EXT4_INODE_GET_ATIME(inode, raw_inode);
4875 EXT4_INODE_GET_MTIME(inode, raw_inode);
4876 EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4877
4878 if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4879 u64 ivers = le32_to_cpu(raw_inode->i_disk_version);
4880
4881 if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4882 if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4883 ivers |=
4884 (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4885 }
4886 ext4_inode_set_iversion_queried(inode, ivers);
4887 }
4888
4889 ret = 0;
4890 if (ei->i_file_acl &&
4891 !ext4_inode_block_valid(inode, ei->i_file_acl, 1)) {
4892 ext4_error_inode(inode, function, line, 0,
4893 "iget: bad extended attribute block %llu",
4894 ei->i_file_acl);
4895 ret = -EFSCORRUPTED;
4896 goto bad_inode;
4897 } else if (!ext4_has_inline_data(inode)) {
4898 /* validate the block references in the inode */
4899 if (!(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY) &&
4900 (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4901 (S_ISLNK(inode->i_mode) &&
4902 !ext4_inode_is_fast_symlink(inode)))) {
4903 if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4904 ret = ext4_ext_check_inode(inode);
4905 else
4906 ret = ext4_ind_check_inode(inode);
4907 }
4908 }
4909 if (ret)
4910 goto bad_inode;
4911
4912 if (S_ISREG(inode->i_mode)) {
4913 inode->i_op = &ext4_file_inode_operations;
4914 inode->i_fop = &ext4_file_operations;
4915 ext4_set_aops(inode);
4916 } else if (S_ISDIR(inode->i_mode)) {
4917 inode->i_op = &ext4_dir_inode_operations;
4918 inode->i_fop = &ext4_dir_operations;
4919 } else if (S_ISLNK(inode->i_mode)) {
4920 /* VFS does not allow setting these so must be corruption */
4921 if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) {
4922 ext4_error_inode(inode, function, line, 0,
4923 "iget: immutable or append flags "
4924 "not allowed on symlinks");
4925 ret = -EFSCORRUPTED;
4926 goto bad_inode;
4927 }
4928 if (IS_ENCRYPTED(inode)) {
4929 inode->i_op = &ext4_encrypted_symlink_inode_operations;
4930 } else if (ext4_inode_is_fast_symlink(inode)) {
4931 inode->i_link = (char *)ei->i_data;
4932 inode->i_op = &ext4_fast_symlink_inode_operations;
4933 nd_terminate_link(ei->i_data, inode->i_size,
4934 sizeof(ei->i_data) - 1);
4935 } else {
4936 inode->i_op = &ext4_symlink_inode_operations;
4937 }
4938 } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4939 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4940 inode->i_op = &ext4_special_inode_operations;
4941 if (raw_inode->i_block[0])
4942 init_special_inode(inode, inode->i_mode,
4943 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4944 else
4945 init_special_inode(inode, inode->i_mode,
4946 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4947 } else if (ino == EXT4_BOOT_LOADER_INO) {
4948 make_bad_inode(inode);
4949 } else {
4950 ret = -EFSCORRUPTED;
4951 ext4_error_inode(inode, function, line, 0,
4952 "iget: bogus i_mode (%o)", inode->i_mode);
4953 goto bad_inode;
4954 }
4955 if (IS_CASEFOLDED(inode) && !ext4_has_feature_casefold(inode->i_sb)) {
4956 ext4_error_inode(inode, function, line, 0,
4957 "casefold flag without casefold feature");
4958 ret = -EFSCORRUPTED;
4959 goto bad_inode;
4960 }
4961 if ((err_str = check_igot_inode(inode, flags)) != NULL) {
4962 ext4_error_inode(inode, function, line, 0, err_str);
4963 ret = -EFSCORRUPTED;
4964 goto bad_inode;
4965 }
4966
4967 brelse(iloc.bh);
4968 unlock_new_inode(inode);
4969 return inode;
4970
4971 bad_inode:
4972 brelse(iloc.bh);
4973 iget_failed(inode);
4974 return ERR_PTR(ret);
4975 }
4976
__ext4_update_other_inode_time(struct super_block * sb,unsigned long orig_ino,unsigned long ino,struct ext4_inode * raw_inode)4977 static void __ext4_update_other_inode_time(struct super_block *sb,
4978 unsigned long orig_ino,
4979 unsigned long ino,
4980 struct ext4_inode *raw_inode)
4981 {
4982 struct inode *inode;
4983
4984 inode = find_inode_by_ino_rcu(sb, ino);
4985 if (!inode)
4986 return;
4987
4988 if (!inode_is_dirtytime_only(inode))
4989 return;
4990
4991 spin_lock(&inode->i_lock);
4992 if (inode_is_dirtytime_only(inode)) {
4993 struct ext4_inode_info *ei = EXT4_I(inode);
4994
4995 inode->i_state &= ~I_DIRTY_TIME;
4996 spin_unlock(&inode->i_lock);
4997
4998 spin_lock(&ei->i_raw_lock);
4999 EXT4_INODE_SET_CTIME(inode, raw_inode);
5000 EXT4_INODE_SET_MTIME(inode, raw_inode);
5001 EXT4_INODE_SET_ATIME(inode, raw_inode);
5002 ext4_inode_csum_set(inode, raw_inode, ei);
5003 spin_unlock(&ei->i_raw_lock);
5004 trace_ext4_other_inode_update_time(inode, orig_ino);
5005 return;
5006 }
5007 spin_unlock(&inode->i_lock);
5008 }
5009
5010 /*
5011 * Opportunistically update the other time fields for other inodes in
5012 * the same inode table block.
5013 */
ext4_update_other_inodes_time(struct super_block * sb,unsigned long orig_ino,char * buf)5014 static void ext4_update_other_inodes_time(struct super_block *sb,
5015 unsigned long orig_ino, char *buf)
5016 {
5017 unsigned long ino;
5018 int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
5019 int inode_size = EXT4_INODE_SIZE(sb);
5020
5021 /*
5022 * Calculate the first inode in the inode table block. Inode
5023 * numbers are one-based. That is, the first inode in a block
5024 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
5025 */
5026 ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1;
5027 rcu_read_lock();
5028 for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) {
5029 if (ino == orig_ino)
5030 continue;
5031 __ext4_update_other_inode_time(sb, orig_ino, ino,
5032 (struct ext4_inode *)buf);
5033 }
5034 rcu_read_unlock();
5035 }
5036
5037 /*
5038 * Post the struct inode info into an on-disk inode location in the
5039 * buffer-cache. This gobbles the caller's reference to the
5040 * buffer_head in the inode location struct.
5041 *
5042 * The caller must have write access to iloc->bh.
5043 */
ext4_do_update_inode(handle_t * handle,struct inode * inode,struct ext4_iloc * iloc)5044 static int ext4_do_update_inode(handle_t *handle,
5045 struct inode *inode,
5046 struct ext4_iloc *iloc)
5047 {
5048 struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
5049 struct ext4_inode_info *ei = EXT4_I(inode);
5050 struct buffer_head *bh = iloc->bh;
5051 struct super_block *sb = inode->i_sb;
5052 int err;
5053 int need_datasync = 0, set_large_file = 0;
5054
5055 spin_lock(&ei->i_raw_lock);
5056
5057 /*
5058 * For fields not tracked in the in-memory inode, initialise them
5059 * to zero for new inodes.
5060 */
5061 if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
5062 memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
5063
5064 if (READ_ONCE(ei->i_disksize) != ext4_isize(inode->i_sb, raw_inode))
5065 need_datasync = 1;
5066 if (ei->i_disksize > 0x7fffffffULL) {
5067 if (!ext4_has_feature_large_file(sb) ||
5068 EXT4_SB(sb)->s_es->s_rev_level == cpu_to_le32(EXT4_GOOD_OLD_REV))
5069 set_large_file = 1;
5070 }
5071
5072 err = ext4_fill_raw_inode(inode, raw_inode);
5073 spin_unlock(&ei->i_raw_lock);
5074 if (err) {
5075 EXT4_ERROR_INODE(inode, "corrupted inode contents");
5076 goto out_brelse;
5077 }
5078
5079 if (inode->i_sb->s_flags & SB_LAZYTIME)
5080 ext4_update_other_inodes_time(inode->i_sb, inode->i_ino,
5081 bh->b_data);
5082
5083 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
5084 err = ext4_handle_dirty_metadata(handle, NULL, bh);
5085 if (err)
5086 goto out_error;
5087 ext4_clear_inode_state(inode, EXT4_STATE_NEW);
5088 if (set_large_file) {
5089 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
5090 err = ext4_journal_get_write_access(handle, sb,
5091 EXT4_SB(sb)->s_sbh,
5092 EXT4_JTR_NONE);
5093 if (err)
5094 goto out_error;
5095 lock_buffer(EXT4_SB(sb)->s_sbh);
5096 ext4_set_feature_large_file(sb);
5097 ext4_superblock_csum_set(sb);
5098 unlock_buffer(EXT4_SB(sb)->s_sbh);
5099 ext4_handle_sync(handle);
5100 err = ext4_handle_dirty_metadata(handle, NULL,
5101 EXT4_SB(sb)->s_sbh);
5102 }
5103 ext4_update_inode_fsync_trans(handle, inode, need_datasync);
5104 out_error:
5105 ext4_std_error(inode->i_sb, err);
5106 out_brelse:
5107 brelse(bh);
5108 return err;
5109 }
5110
5111 /*
5112 * ext4_write_inode()
5113 *
5114 * We are called from a few places:
5115 *
5116 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
5117 * Here, there will be no transaction running. We wait for any running
5118 * transaction to commit.
5119 *
5120 * - Within flush work (sys_sync(), kupdate and such).
5121 * We wait on commit, if told to.
5122 *
5123 * - Within iput_final() -> write_inode_now()
5124 * We wait on commit, if told to.
5125 *
5126 * In all cases it is actually safe for us to return without doing anything,
5127 * because the inode has been copied into a raw inode buffer in
5128 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
5129 * writeback.
5130 *
5131 * Note that we are absolutely dependent upon all inode dirtiers doing the
5132 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5133 * which we are interested.
5134 *
5135 * It would be a bug for them to not do this. The code:
5136 *
5137 * mark_inode_dirty(inode)
5138 * stuff();
5139 * inode->i_size = expr;
5140 *
5141 * is in error because write_inode() could occur while `stuff()' is running,
5142 * and the new i_size will be lost. Plus the inode will no longer be on the
5143 * superblock's dirty inode list.
5144 */
ext4_write_inode(struct inode * inode,struct writeback_control * wbc)5145 int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
5146 {
5147 int err;
5148
5149 if (WARN_ON_ONCE(current->flags & PF_MEMALLOC))
5150 return 0;
5151
5152 if (unlikely(ext4_forced_shutdown(inode->i_sb)))
5153 return -EIO;
5154
5155 if (EXT4_SB(inode->i_sb)->s_journal) {
5156 if (ext4_journal_current_handle()) {
5157 ext4_debug("called recursively, non-PF_MEMALLOC!\n");
5158 dump_stack();
5159 return -EIO;
5160 }
5161
5162 /*
5163 * No need to force transaction in WB_SYNC_NONE mode. Also
5164 * ext4_sync_fs() will force the commit after everything is
5165 * written.
5166 */
5167 if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
5168 return 0;
5169
5170 err = ext4_fc_commit(EXT4_SB(inode->i_sb)->s_journal,
5171 EXT4_I(inode)->i_sync_tid);
5172 } else {
5173 struct ext4_iloc iloc;
5174
5175 err = __ext4_get_inode_loc_noinmem(inode, &iloc);
5176 if (err)
5177 return err;
5178 /*
5179 * sync(2) will flush the whole buffer cache. No need to do
5180 * it here separately for each inode.
5181 */
5182 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
5183 sync_dirty_buffer(iloc.bh);
5184 if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5185 ext4_error_inode_block(inode, iloc.bh->b_blocknr, EIO,
5186 "IO error syncing inode");
5187 err = -EIO;
5188 }
5189 brelse(iloc.bh);
5190 }
5191 return err;
5192 }
5193
5194 /*
5195 * In data=journal mode ext4_journalled_invalidate_folio() may fail to invalidate
5196 * buffers that are attached to a folio straddling i_size and are undergoing
5197 * commit. In that case we have to wait for commit to finish and try again.
5198 */
ext4_wait_for_tail_page_commit(struct inode * inode)5199 static void ext4_wait_for_tail_page_commit(struct inode *inode)
5200 {
5201 unsigned offset;
5202 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
5203 tid_t commit_tid = 0;
5204 int ret;
5205
5206 offset = inode->i_size & (PAGE_SIZE - 1);
5207 /*
5208 * If the folio is fully truncated, we don't need to wait for any commit
5209 * (and we even should not as __ext4_journalled_invalidate_folio() may
5210 * strip all buffers from the folio but keep the folio dirty which can then
5211 * confuse e.g. concurrent ext4_writepages() seeing dirty folio without
5212 * buffers). Also we don't need to wait for any commit if all buffers in
5213 * the folio remain valid. This is most beneficial for the common case of
5214 * blocksize == PAGESIZE.
5215 */
5216 if (!offset || offset > (PAGE_SIZE - i_blocksize(inode)))
5217 return;
5218 while (1) {
5219 struct folio *folio = filemap_lock_folio(inode->i_mapping,
5220 inode->i_size >> PAGE_SHIFT);
5221 if (IS_ERR(folio))
5222 return;
5223 ret = __ext4_journalled_invalidate_folio(folio, offset,
5224 folio_size(folio) - offset);
5225 folio_unlock(folio);
5226 folio_put(folio);
5227 if (ret != -EBUSY)
5228 return;
5229 commit_tid = 0;
5230 read_lock(&journal->j_state_lock);
5231 if (journal->j_committing_transaction)
5232 commit_tid = journal->j_committing_transaction->t_tid;
5233 read_unlock(&journal->j_state_lock);
5234 if (commit_tid)
5235 jbd2_log_wait_commit(journal, commit_tid);
5236 }
5237 }
5238
5239 /*
5240 * ext4_setattr()
5241 *
5242 * Called from notify_change.
5243 *
5244 * We want to trap VFS attempts to truncate the file as soon as
5245 * possible. In particular, we want to make sure that when the VFS
5246 * shrinks i_size, we put the inode on the orphan list and modify
5247 * i_disksize immediately, so that during the subsequent flushing of
5248 * dirty pages and freeing of disk blocks, we can guarantee that any
5249 * commit will leave the blocks being flushed in an unused state on
5250 * disk. (On recovery, the inode will get truncated and the blocks will
5251 * be freed, so we have a strong guarantee that no future commit will
5252 * leave these blocks visible to the user.)
5253 *
5254 * Another thing we have to assure is that if we are in ordered mode
5255 * and inode is still attached to the committing transaction, we must
5256 * we start writeout of all the dirty pages which are being truncated.
5257 * This way we are sure that all the data written in the previous
5258 * transaction are already on disk (truncate waits for pages under
5259 * writeback).
5260 *
5261 * Called with inode->i_rwsem down.
5262 */
ext4_setattr(struct mnt_idmap * idmap,struct dentry * dentry,struct iattr * attr)5263 int ext4_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
5264 struct iattr *attr)
5265 {
5266 struct inode *inode = d_inode(dentry);
5267 int error, rc = 0;
5268 int orphan = 0;
5269 const unsigned int ia_valid = attr->ia_valid;
5270 bool inc_ivers = true;
5271
5272 if (unlikely(ext4_forced_shutdown(inode->i_sb)))
5273 return -EIO;
5274
5275 if (unlikely(IS_IMMUTABLE(inode)))
5276 return -EPERM;
5277
5278 if (unlikely(IS_APPEND(inode) &&
5279 (ia_valid & (ATTR_MODE | ATTR_UID |
5280 ATTR_GID | ATTR_TIMES_SET))))
5281 return -EPERM;
5282
5283 error = setattr_prepare(idmap, dentry, attr);
5284 if (error)
5285 return error;
5286
5287 error = fscrypt_prepare_setattr(dentry, attr);
5288 if (error)
5289 return error;
5290
5291 error = fsverity_prepare_setattr(dentry, attr);
5292 if (error)
5293 return error;
5294
5295 if (is_quota_modification(idmap, inode, attr)) {
5296 error = dquot_initialize(inode);
5297 if (error)
5298 return error;
5299 }
5300
5301 if (i_uid_needs_update(idmap, attr, inode) ||
5302 i_gid_needs_update(idmap, attr, inode)) {
5303 handle_t *handle;
5304
5305 /* (user+group)*(old+new) structure, inode write (sb,
5306 * inode block, ? - but truncate inode update has it) */
5307 handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5308 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
5309 EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
5310 if (IS_ERR(handle)) {
5311 error = PTR_ERR(handle);
5312 goto err_out;
5313 }
5314
5315 /* dquot_transfer() calls back ext4_get_inode_usage() which
5316 * counts xattr inode references.
5317 */
5318 down_read(&EXT4_I(inode)->xattr_sem);
5319 error = dquot_transfer(idmap, inode, attr);
5320 up_read(&EXT4_I(inode)->xattr_sem);
5321
5322 if (error) {
5323 ext4_journal_stop(handle);
5324 return error;
5325 }
5326 /* Update corresponding info in inode so that everything is in
5327 * one transaction */
5328 i_uid_update(idmap, attr, inode);
5329 i_gid_update(idmap, attr, inode);
5330 error = ext4_mark_inode_dirty(handle, inode);
5331 ext4_journal_stop(handle);
5332 if (unlikely(error)) {
5333 return error;
5334 }
5335 }
5336
5337 if (attr->ia_valid & ATTR_SIZE) {
5338 handle_t *handle;
5339 loff_t oldsize = inode->i_size;
5340 loff_t old_disksize;
5341 int shrink = (attr->ia_size < inode->i_size);
5342
5343 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
5344 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5345
5346 if (attr->ia_size > sbi->s_bitmap_maxbytes) {
5347 return -EFBIG;
5348 }
5349 }
5350 if (!S_ISREG(inode->i_mode)) {
5351 return -EINVAL;
5352 }
5353
5354 if (attr->ia_size == inode->i_size)
5355 inc_ivers = false;
5356
5357 if (shrink) {
5358 if (ext4_should_order_data(inode)) {
5359 error = ext4_begin_ordered_truncate(inode,
5360 attr->ia_size);
5361 if (error)
5362 goto err_out;
5363 }
5364 /*
5365 * Blocks are going to be removed from the inode. Wait
5366 * for dio in flight.
5367 */
5368 inode_dio_wait(inode);
5369 }
5370
5371 filemap_invalidate_lock(inode->i_mapping);
5372
5373 rc = ext4_break_layouts(inode);
5374 if (rc) {
5375 filemap_invalidate_unlock(inode->i_mapping);
5376 goto err_out;
5377 }
5378
5379 if (attr->ia_size != inode->i_size) {
5380 handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
5381 if (IS_ERR(handle)) {
5382 error = PTR_ERR(handle);
5383 goto out_mmap_sem;
5384 }
5385 if (ext4_handle_valid(handle) && shrink) {
5386 error = ext4_orphan_add(handle, inode);
5387 orphan = 1;
5388 }
5389 /*
5390 * Update c/mtime on truncate up, ext4_truncate() will
5391 * update c/mtime in shrink case below
5392 */
5393 if (!shrink)
5394 inode_set_mtime_to_ts(inode,
5395 inode_set_ctime_current(inode));
5396
5397 if (shrink)
5398 ext4_fc_track_range(handle, inode,
5399 (attr->ia_size > 0 ? attr->ia_size - 1 : 0) >>
5400 inode->i_sb->s_blocksize_bits,
5401 EXT_MAX_BLOCKS - 1);
5402 else
5403 ext4_fc_track_range(
5404 handle, inode,
5405 (oldsize > 0 ? oldsize - 1 : oldsize) >>
5406 inode->i_sb->s_blocksize_bits,
5407 (attr->ia_size > 0 ? attr->ia_size - 1 : 0) >>
5408 inode->i_sb->s_blocksize_bits);
5409
5410 down_write(&EXT4_I(inode)->i_data_sem);
5411 old_disksize = EXT4_I(inode)->i_disksize;
5412 EXT4_I(inode)->i_disksize = attr->ia_size;
5413 rc = ext4_mark_inode_dirty(handle, inode);
5414 if (!error)
5415 error = rc;
5416 /*
5417 * We have to update i_size under i_data_sem together
5418 * with i_disksize to avoid races with writeback code
5419 * running ext4_wb_update_i_disksize().
5420 */
5421 if (!error)
5422 i_size_write(inode, attr->ia_size);
5423 else
5424 EXT4_I(inode)->i_disksize = old_disksize;
5425 up_write(&EXT4_I(inode)->i_data_sem);
5426 ext4_journal_stop(handle);
5427 if (error)
5428 goto out_mmap_sem;
5429 if (!shrink) {
5430 pagecache_isize_extended(inode, oldsize,
5431 inode->i_size);
5432 } else if (ext4_should_journal_data(inode)) {
5433 ext4_wait_for_tail_page_commit(inode);
5434 }
5435 }
5436
5437 /*
5438 * Truncate pagecache after we've waited for commit
5439 * in data=journal mode to make pages freeable.
5440 */
5441 truncate_pagecache(inode, inode->i_size);
5442 /*
5443 * Call ext4_truncate() even if i_size didn't change to
5444 * truncate possible preallocated blocks.
5445 */
5446 if (attr->ia_size <= oldsize) {
5447 rc = ext4_truncate(inode);
5448 if (rc)
5449 error = rc;
5450 }
5451 out_mmap_sem:
5452 filemap_invalidate_unlock(inode->i_mapping);
5453 }
5454
5455 if (!error) {
5456 if (inc_ivers)
5457 inode_inc_iversion(inode);
5458 setattr_copy(idmap, inode, attr);
5459 mark_inode_dirty(inode);
5460 }
5461
5462 /*
5463 * If the call to ext4_truncate failed to get a transaction handle at
5464 * all, we need to clean up the in-core orphan list manually.
5465 */
5466 if (orphan && inode->i_nlink)
5467 ext4_orphan_del(NULL, inode);
5468
5469 if (!error && (ia_valid & ATTR_MODE))
5470 rc = posix_acl_chmod(idmap, dentry, inode->i_mode);
5471
5472 err_out:
5473 if (error)
5474 ext4_std_error(inode->i_sb, error);
5475 if (!error)
5476 error = rc;
5477 return error;
5478 }
5479
ext4_dio_alignment(struct inode * inode)5480 u32 ext4_dio_alignment(struct inode *inode)
5481 {
5482 if (fsverity_active(inode))
5483 return 0;
5484 if (ext4_should_journal_data(inode))
5485 return 0;
5486 if (ext4_has_inline_data(inode))
5487 return 0;
5488 if (IS_ENCRYPTED(inode)) {
5489 if (!fscrypt_dio_supported(inode))
5490 return 0;
5491 return i_blocksize(inode);
5492 }
5493 return 1; /* use the iomap defaults */
5494 }
5495
ext4_getattr(struct mnt_idmap * idmap,const struct path * path,struct kstat * stat,u32 request_mask,unsigned int query_flags)5496 int ext4_getattr(struct mnt_idmap *idmap, const struct path *path,
5497 struct kstat *stat, u32 request_mask, unsigned int query_flags)
5498 {
5499 struct inode *inode = d_inode(path->dentry);
5500 struct ext4_inode *raw_inode;
5501 struct ext4_inode_info *ei = EXT4_I(inode);
5502 unsigned int flags;
5503
5504 if ((request_mask & STATX_BTIME) &&
5505 EXT4_FITS_IN_INODE(raw_inode, ei, i_crtime)) {
5506 stat->result_mask |= STATX_BTIME;
5507 stat->btime.tv_sec = ei->i_crtime.tv_sec;
5508 stat->btime.tv_nsec = ei->i_crtime.tv_nsec;
5509 }
5510
5511 /*
5512 * Return the DIO alignment restrictions if requested. We only return
5513 * this information when requested, since on encrypted files it might
5514 * take a fair bit of work to get if the file wasn't opened recently.
5515 */
5516 if ((request_mask & STATX_DIOALIGN) && S_ISREG(inode->i_mode)) {
5517 u32 dio_align = ext4_dio_alignment(inode);
5518
5519 stat->result_mask |= STATX_DIOALIGN;
5520 if (dio_align == 1) {
5521 struct block_device *bdev = inode->i_sb->s_bdev;
5522
5523 /* iomap defaults */
5524 stat->dio_mem_align = bdev_dma_alignment(bdev) + 1;
5525 stat->dio_offset_align = bdev_logical_block_size(bdev);
5526 } else {
5527 stat->dio_mem_align = dio_align;
5528 stat->dio_offset_align = dio_align;
5529 }
5530 }
5531
5532 flags = ei->i_flags & EXT4_FL_USER_VISIBLE;
5533 if (flags & EXT4_APPEND_FL)
5534 stat->attributes |= STATX_ATTR_APPEND;
5535 if (flags & EXT4_COMPR_FL)
5536 stat->attributes |= STATX_ATTR_COMPRESSED;
5537 if (flags & EXT4_ENCRYPT_FL)
5538 stat->attributes |= STATX_ATTR_ENCRYPTED;
5539 if (flags & EXT4_IMMUTABLE_FL)
5540 stat->attributes |= STATX_ATTR_IMMUTABLE;
5541 if (flags & EXT4_NODUMP_FL)
5542 stat->attributes |= STATX_ATTR_NODUMP;
5543 if (flags & EXT4_VERITY_FL)
5544 stat->attributes |= STATX_ATTR_VERITY;
5545
5546 stat->attributes_mask |= (STATX_ATTR_APPEND |
5547 STATX_ATTR_COMPRESSED |
5548 STATX_ATTR_ENCRYPTED |
5549 STATX_ATTR_IMMUTABLE |
5550 STATX_ATTR_NODUMP |
5551 STATX_ATTR_VERITY);
5552
5553 generic_fillattr(idmap, request_mask, inode, stat);
5554 return 0;
5555 }
5556
ext4_file_getattr(struct mnt_idmap * idmap,const struct path * path,struct kstat * stat,u32 request_mask,unsigned int query_flags)5557 int ext4_file_getattr(struct mnt_idmap *idmap,
5558 const struct path *path, struct kstat *stat,
5559 u32 request_mask, unsigned int query_flags)
5560 {
5561 struct inode *inode = d_inode(path->dentry);
5562 u64 delalloc_blocks;
5563
5564 ext4_getattr(idmap, path, stat, request_mask, query_flags);
5565
5566 /*
5567 * If there is inline data in the inode, the inode will normally not
5568 * have data blocks allocated (it may have an external xattr block).
5569 * Report at least one sector for such files, so tools like tar, rsync,
5570 * others don't incorrectly think the file is completely sparse.
5571 */
5572 if (unlikely(ext4_has_inline_data(inode)))
5573 stat->blocks += (stat->size + 511) >> 9;
5574
5575 /*
5576 * We can't update i_blocks if the block allocation is delayed
5577 * otherwise in the case of system crash before the real block
5578 * allocation is done, we will have i_blocks inconsistent with
5579 * on-disk file blocks.
5580 * We always keep i_blocks updated together with real
5581 * allocation. But to not confuse with user, stat
5582 * will return the blocks that include the delayed allocation
5583 * blocks for this file.
5584 */
5585 delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
5586 EXT4_I(inode)->i_reserved_data_blocks);
5587 stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
5588 return 0;
5589 }
5590
ext4_index_trans_blocks(struct inode * inode,int lblocks,int pextents)5591 static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
5592 int pextents)
5593 {
5594 if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5595 return ext4_ind_trans_blocks(inode, lblocks);
5596 return ext4_ext_index_trans_blocks(inode, pextents);
5597 }
5598
5599 /*
5600 * Account for index blocks, block groups bitmaps and block group
5601 * descriptor blocks if modify datablocks and index blocks
5602 * worse case, the indexs blocks spread over different block groups
5603 *
5604 * If datablocks are discontiguous, they are possible to spread over
5605 * different block groups too. If they are contiguous, with flexbg,
5606 * they could still across block group boundary.
5607 *
5608 * Also account for superblock, inode, quota and xattr blocks
5609 */
ext4_meta_trans_blocks(struct inode * inode,int lblocks,int pextents)5610 static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
5611 int pextents)
5612 {
5613 ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5614 int gdpblocks;
5615 int idxblocks;
5616 int ret;
5617
5618 /*
5619 * How many index blocks need to touch to map @lblocks logical blocks
5620 * to @pextents physical extents?
5621 */
5622 idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
5623
5624 ret = idxblocks;
5625
5626 /*
5627 * Now let's see how many group bitmaps and group descriptors need
5628 * to account
5629 */
5630 groups = idxblocks + pextents;
5631 gdpblocks = groups;
5632 if (groups > ngroups)
5633 groups = ngroups;
5634 if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5635 gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5636
5637 /* bitmaps and block group descriptor blocks */
5638 ret += groups + gdpblocks;
5639
5640 /* Blocks for super block, inode, quota and xattr blocks */
5641 ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5642
5643 return ret;
5644 }
5645
5646 /*
5647 * Calculate the total number of credits to reserve to fit
5648 * the modification of a single pages into a single transaction,
5649 * which may include multiple chunks of block allocations.
5650 *
5651 * This could be called via ext4_write_begin()
5652 *
5653 * We need to consider the worse case, when
5654 * one new block per extent.
5655 */
ext4_writepage_trans_blocks(struct inode * inode)5656 int ext4_writepage_trans_blocks(struct inode *inode)
5657 {
5658 int bpp = ext4_journal_blocks_per_page(inode);
5659 int ret;
5660
5661 ret = ext4_meta_trans_blocks(inode, bpp, bpp);
5662
5663 /* Account for data blocks for journalled mode */
5664 if (ext4_should_journal_data(inode))
5665 ret += bpp;
5666 return ret;
5667 }
5668
5669 /*
5670 * Calculate the journal credits for a chunk of data modification.
5671 *
5672 * This is called from DIO, fallocate or whoever calling
5673 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5674 *
5675 * journal buffers for data blocks are not included here, as DIO
5676 * and fallocate do no need to journal data buffers.
5677 */
ext4_chunk_trans_blocks(struct inode * inode,int nrblocks)5678 int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5679 {
5680 return ext4_meta_trans_blocks(inode, nrblocks, 1);
5681 }
5682
5683 /*
5684 * The caller must have previously called ext4_reserve_inode_write().
5685 * Give this, we know that the caller already has write access to iloc->bh.
5686 */
ext4_mark_iloc_dirty(handle_t * handle,struct inode * inode,struct ext4_iloc * iloc)5687 int ext4_mark_iloc_dirty(handle_t *handle,
5688 struct inode *inode, struct ext4_iloc *iloc)
5689 {
5690 int err = 0;
5691
5692 if (unlikely(ext4_forced_shutdown(inode->i_sb))) {
5693 put_bh(iloc->bh);
5694 return -EIO;
5695 }
5696 ext4_fc_track_inode(handle, inode);
5697
5698 /* the do_update_inode consumes one bh->b_count */
5699 get_bh(iloc->bh);
5700
5701 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5702 err = ext4_do_update_inode(handle, inode, iloc);
5703 put_bh(iloc->bh);
5704 return err;
5705 }
5706
5707 /*
5708 * On success, We end up with an outstanding reference count against
5709 * iloc->bh. This _must_ be cleaned up later.
5710 */
5711
5712 int
ext4_reserve_inode_write(handle_t * handle,struct inode * inode,struct ext4_iloc * iloc)5713 ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5714 struct ext4_iloc *iloc)
5715 {
5716 int err;
5717
5718 if (unlikely(ext4_forced_shutdown(inode->i_sb)))
5719 return -EIO;
5720
5721 err = ext4_get_inode_loc(inode, iloc);
5722 if (!err) {
5723 BUFFER_TRACE(iloc->bh, "get_write_access");
5724 err = ext4_journal_get_write_access(handle, inode->i_sb,
5725 iloc->bh, EXT4_JTR_NONE);
5726 if (err) {
5727 brelse(iloc->bh);
5728 iloc->bh = NULL;
5729 }
5730 }
5731 ext4_std_error(inode->i_sb, err);
5732 return err;
5733 }
5734
__ext4_expand_extra_isize(struct inode * inode,unsigned int new_extra_isize,struct ext4_iloc * iloc,handle_t * handle,int * no_expand)5735 static int __ext4_expand_extra_isize(struct inode *inode,
5736 unsigned int new_extra_isize,
5737 struct ext4_iloc *iloc,
5738 handle_t *handle, int *no_expand)
5739 {
5740 struct ext4_inode *raw_inode;
5741 struct ext4_xattr_ibody_header *header;
5742 unsigned int inode_size = EXT4_INODE_SIZE(inode->i_sb);
5743 struct ext4_inode_info *ei = EXT4_I(inode);
5744 int error;
5745
5746 /* this was checked at iget time, but double check for good measure */
5747 if ((EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > inode_size) ||
5748 (ei->i_extra_isize & 3)) {
5749 EXT4_ERROR_INODE(inode, "bad extra_isize %u (inode size %u)",
5750 ei->i_extra_isize,
5751 EXT4_INODE_SIZE(inode->i_sb));
5752 return -EFSCORRUPTED;
5753 }
5754 if ((new_extra_isize < ei->i_extra_isize) ||
5755 (new_extra_isize < 4) ||
5756 (new_extra_isize > inode_size - EXT4_GOOD_OLD_INODE_SIZE))
5757 return -EINVAL; /* Should never happen */
5758
5759 raw_inode = ext4_raw_inode(iloc);
5760
5761 header = IHDR(inode, raw_inode);
5762
5763 /* No extended attributes present */
5764 if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5765 header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5766 memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE +
5767 EXT4_I(inode)->i_extra_isize, 0,
5768 new_extra_isize - EXT4_I(inode)->i_extra_isize);
5769 EXT4_I(inode)->i_extra_isize = new_extra_isize;
5770 return 0;
5771 }
5772
5773 /*
5774 * We may need to allocate external xattr block so we need quotas
5775 * initialized. Here we can be called with various locks held so we
5776 * cannot affort to initialize quotas ourselves. So just bail.
5777 */
5778 if (dquot_initialize_needed(inode))
5779 return -EAGAIN;
5780
5781 /* try to expand with EAs present */
5782 error = ext4_expand_extra_isize_ea(inode, new_extra_isize,
5783 raw_inode, handle);
5784 if (error) {
5785 /*
5786 * Inode size expansion failed; don't try again
5787 */
5788 *no_expand = 1;
5789 }
5790
5791 return error;
5792 }
5793
5794 /*
5795 * Expand an inode by new_extra_isize bytes.
5796 * Returns 0 on success or negative error number on failure.
5797 */
ext4_try_to_expand_extra_isize(struct inode * inode,unsigned int new_extra_isize,struct ext4_iloc iloc,handle_t * handle)5798 static int ext4_try_to_expand_extra_isize(struct inode *inode,
5799 unsigned int new_extra_isize,
5800 struct ext4_iloc iloc,
5801 handle_t *handle)
5802 {
5803 int no_expand;
5804 int error;
5805
5806 if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND))
5807 return -EOVERFLOW;
5808
5809 /*
5810 * In nojournal mode, we can immediately attempt to expand
5811 * the inode. When journaled, we first need to obtain extra
5812 * buffer credits since we may write into the EA block
5813 * with this same handle. If journal_extend fails, then it will
5814 * only result in a minor loss of functionality for that inode.
5815 * If this is felt to be critical, then e2fsck should be run to
5816 * force a large enough s_min_extra_isize.
5817 */
5818 if (ext4_journal_extend(handle,
5819 EXT4_DATA_TRANS_BLOCKS(inode->i_sb), 0) != 0)
5820 return -ENOSPC;
5821
5822 if (ext4_write_trylock_xattr(inode, &no_expand) == 0)
5823 return -EBUSY;
5824
5825 error = __ext4_expand_extra_isize(inode, new_extra_isize, &iloc,
5826 handle, &no_expand);
5827 ext4_write_unlock_xattr(inode, &no_expand);
5828
5829 return error;
5830 }
5831
ext4_expand_extra_isize(struct inode * inode,unsigned int new_extra_isize,struct ext4_iloc * iloc)5832 int ext4_expand_extra_isize(struct inode *inode,
5833 unsigned int new_extra_isize,
5834 struct ext4_iloc *iloc)
5835 {
5836 handle_t *handle;
5837 int no_expand;
5838 int error, rc;
5839
5840 if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5841 brelse(iloc->bh);
5842 return -EOVERFLOW;
5843 }
5844
5845 handle = ext4_journal_start(inode, EXT4_HT_INODE,
5846 EXT4_DATA_TRANS_BLOCKS(inode->i_sb));
5847 if (IS_ERR(handle)) {
5848 error = PTR_ERR(handle);
5849 brelse(iloc->bh);
5850 return error;
5851 }
5852
5853 ext4_write_lock_xattr(inode, &no_expand);
5854
5855 BUFFER_TRACE(iloc->bh, "get_write_access");
5856 error = ext4_journal_get_write_access(handle, inode->i_sb, iloc->bh,
5857 EXT4_JTR_NONE);
5858 if (error) {
5859 brelse(iloc->bh);
5860 goto out_unlock;
5861 }
5862
5863 error = __ext4_expand_extra_isize(inode, new_extra_isize, iloc,
5864 handle, &no_expand);
5865
5866 rc = ext4_mark_iloc_dirty(handle, inode, iloc);
5867 if (!error)
5868 error = rc;
5869
5870 out_unlock:
5871 ext4_write_unlock_xattr(inode, &no_expand);
5872 ext4_journal_stop(handle);
5873 return error;
5874 }
5875
5876 /*
5877 * What we do here is to mark the in-core inode as clean with respect to inode
5878 * dirtiness (it may still be data-dirty).
5879 * This means that the in-core inode may be reaped by prune_icache
5880 * without having to perform any I/O. This is a very good thing,
5881 * because *any* task may call prune_icache - even ones which
5882 * have a transaction open against a different journal.
5883 *
5884 * Is this cheating? Not really. Sure, we haven't written the
5885 * inode out, but prune_icache isn't a user-visible syncing function.
5886 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5887 * we start and wait on commits.
5888 */
__ext4_mark_inode_dirty(handle_t * handle,struct inode * inode,const char * func,unsigned int line)5889 int __ext4_mark_inode_dirty(handle_t *handle, struct inode *inode,
5890 const char *func, unsigned int line)
5891 {
5892 struct ext4_iloc iloc;
5893 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5894 int err;
5895
5896 might_sleep();
5897 trace_ext4_mark_inode_dirty(inode, _RET_IP_);
5898 err = ext4_reserve_inode_write(handle, inode, &iloc);
5899 if (err)
5900 goto out;
5901
5902 if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize)
5903 ext4_try_to_expand_extra_isize(inode, sbi->s_want_extra_isize,
5904 iloc, handle);
5905
5906 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
5907 out:
5908 if (unlikely(err))
5909 ext4_error_inode_err(inode, func, line, 0, err,
5910 "mark_inode_dirty error");
5911 return err;
5912 }
5913
5914 /*
5915 * ext4_dirty_inode() is called from __mark_inode_dirty()
5916 *
5917 * We're really interested in the case where a file is being extended.
5918 * i_size has been changed by generic_commit_write() and we thus need
5919 * to include the updated inode in the current transaction.
5920 *
5921 * Also, dquot_alloc_block() will always dirty the inode when blocks
5922 * are allocated to the file.
5923 *
5924 * If the inode is marked synchronous, we don't honour that here - doing
5925 * so would cause a commit on atime updates, which we don't bother doing.
5926 * We handle synchronous inodes at the highest possible level.
5927 */
ext4_dirty_inode(struct inode * inode,int flags)5928 void ext4_dirty_inode(struct inode *inode, int flags)
5929 {
5930 handle_t *handle;
5931
5932 handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
5933 if (IS_ERR(handle))
5934 return;
5935 ext4_mark_inode_dirty(handle, inode);
5936 ext4_journal_stop(handle);
5937 }
5938
ext4_change_inode_journal_flag(struct inode * inode,int val)5939 int ext4_change_inode_journal_flag(struct inode *inode, int val)
5940 {
5941 journal_t *journal;
5942 handle_t *handle;
5943 int err;
5944 int alloc_ctx;
5945
5946 /*
5947 * We have to be very careful here: changing a data block's
5948 * journaling status dynamically is dangerous. If we write a
5949 * data block to the journal, change the status and then delete
5950 * that block, we risk forgetting to revoke the old log record
5951 * from the journal and so a subsequent replay can corrupt data.
5952 * So, first we make sure that the journal is empty and that
5953 * nobody is changing anything.
5954 */
5955
5956 journal = EXT4_JOURNAL(inode);
5957 if (!journal)
5958 return 0;
5959 if (is_journal_aborted(journal))
5960 return -EROFS;
5961
5962 /* Wait for all existing dio workers */
5963 inode_dio_wait(inode);
5964
5965 /*
5966 * Before flushing the journal and switching inode's aops, we have
5967 * to flush all dirty data the inode has. There can be outstanding
5968 * delayed allocations, there can be unwritten extents created by
5969 * fallocate or buffered writes in dioread_nolock mode covered by
5970 * dirty data which can be converted only after flushing the dirty
5971 * data (and journalled aops don't know how to handle these cases).
5972 */
5973 if (val) {
5974 filemap_invalidate_lock(inode->i_mapping);
5975 err = filemap_write_and_wait(inode->i_mapping);
5976 if (err < 0) {
5977 filemap_invalidate_unlock(inode->i_mapping);
5978 return err;
5979 }
5980 }
5981
5982 alloc_ctx = ext4_writepages_down_write(inode->i_sb);
5983 jbd2_journal_lock_updates(journal);
5984
5985 /*
5986 * OK, there are no updates running now, and all cached data is
5987 * synced to disk. We are now in a completely consistent state
5988 * which doesn't have anything in the journal, and we know that
5989 * no filesystem updates are running, so it is safe to modify
5990 * the inode's in-core data-journaling state flag now.
5991 */
5992
5993 if (val)
5994 ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5995 else {
5996 err = jbd2_journal_flush(journal, 0);
5997 if (err < 0) {
5998 jbd2_journal_unlock_updates(journal);
5999 ext4_writepages_up_write(inode->i_sb, alloc_ctx);
6000 return err;
6001 }
6002 ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
6003 }
6004 ext4_set_aops(inode);
6005
6006 jbd2_journal_unlock_updates(journal);
6007 ext4_writepages_up_write(inode->i_sb, alloc_ctx);
6008
6009 if (val)
6010 filemap_invalidate_unlock(inode->i_mapping);
6011
6012 /* Finally we can mark the inode as dirty. */
6013
6014 handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
6015 if (IS_ERR(handle))
6016 return PTR_ERR(handle);
6017
6018 ext4_fc_mark_ineligible(inode->i_sb,
6019 EXT4_FC_REASON_JOURNAL_FLAG_CHANGE, handle);
6020 err = ext4_mark_inode_dirty(handle, inode);
6021 ext4_handle_sync(handle);
6022 ext4_journal_stop(handle);
6023 ext4_std_error(inode->i_sb, err);
6024
6025 return err;
6026 }
6027
ext4_bh_unmapped(handle_t * handle,struct inode * inode,struct buffer_head * bh)6028 static int ext4_bh_unmapped(handle_t *handle, struct inode *inode,
6029 struct buffer_head *bh)
6030 {
6031 return !buffer_mapped(bh);
6032 }
6033
ext4_page_mkwrite(struct vm_fault * vmf)6034 vm_fault_t ext4_page_mkwrite(struct vm_fault *vmf)
6035 {
6036 struct vm_area_struct *vma = vmf->vma;
6037 struct folio *folio = page_folio(vmf->page);
6038 loff_t size;
6039 unsigned long len;
6040 int err;
6041 vm_fault_t ret;
6042 struct file *file = vma->vm_file;
6043 struct inode *inode = file_inode(file);
6044 struct address_space *mapping = inode->i_mapping;
6045 handle_t *handle;
6046 get_block_t *get_block;
6047 int retries = 0;
6048
6049 if (unlikely(IS_IMMUTABLE(inode)))
6050 return VM_FAULT_SIGBUS;
6051
6052 sb_start_pagefault(inode->i_sb);
6053 file_update_time(vma->vm_file);
6054
6055 filemap_invalidate_lock_shared(mapping);
6056
6057 err = ext4_convert_inline_data(inode);
6058 if (err)
6059 goto out_ret;
6060
6061 /*
6062 * On data journalling we skip straight to the transaction handle:
6063 * there's no delalloc; page truncated will be checked later; the
6064 * early return w/ all buffers mapped (calculates size/len) can't
6065 * be used; and there's no dioread_nolock, so only ext4_get_block.
6066 */
6067 if (ext4_should_journal_data(inode))
6068 goto retry_alloc;
6069
6070 /* Delalloc case is easy... */
6071 if (test_opt(inode->i_sb, DELALLOC) &&
6072 !ext4_nonda_switch(inode->i_sb)) {
6073 do {
6074 err = block_page_mkwrite(vma, vmf,
6075 ext4_da_get_block_prep);
6076 } while (err == -ENOSPC &&
6077 ext4_should_retry_alloc(inode->i_sb, &retries));
6078 goto out_ret;
6079 }
6080
6081 folio_lock(folio);
6082 size = i_size_read(inode);
6083 /* Page got truncated from under us? */
6084 if (folio->mapping != mapping || folio_pos(folio) > size) {
6085 folio_unlock(folio);
6086 ret = VM_FAULT_NOPAGE;
6087 goto out;
6088 }
6089
6090 len = folio_size(folio);
6091 if (folio_pos(folio) + len > size)
6092 len = size - folio_pos(folio);
6093 /*
6094 * Return if we have all the buffers mapped. This avoids the need to do
6095 * journal_start/journal_stop which can block and take a long time
6096 *
6097 * This cannot be done for data journalling, as we have to add the
6098 * inode to the transaction's list to writeprotect pages on commit.
6099 */
6100 if (folio_buffers(folio)) {
6101 if (!ext4_walk_page_buffers(NULL, inode, folio_buffers(folio),
6102 0, len, NULL,
6103 ext4_bh_unmapped)) {
6104 /* Wait so that we don't change page under IO */
6105 folio_wait_stable(folio);
6106 ret = VM_FAULT_LOCKED;
6107 goto out;
6108 }
6109 }
6110 folio_unlock(folio);
6111 /* OK, we need to fill the hole... */
6112 if (ext4_should_dioread_nolock(inode))
6113 get_block = ext4_get_block_unwritten;
6114 else
6115 get_block = ext4_get_block;
6116 retry_alloc:
6117 handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
6118 ext4_writepage_trans_blocks(inode));
6119 if (IS_ERR(handle)) {
6120 ret = VM_FAULT_SIGBUS;
6121 goto out;
6122 }
6123 /*
6124 * Data journalling can't use block_page_mkwrite() because it
6125 * will set_buffer_dirty() before do_journal_get_write_access()
6126 * thus might hit warning messages for dirty metadata buffers.
6127 */
6128 if (!ext4_should_journal_data(inode)) {
6129 err = block_page_mkwrite(vma, vmf, get_block);
6130 } else {
6131 folio_lock(folio);
6132 size = i_size_read(inode);
6133 /* Page got truncated from under us? */
6134 if (folio->mapping != mapping || folio_pos(folio) > size) {
6135 ret = VM_FAULT_NOPAGE;
6136 goto out_error;
6137 }
6138
6139 len = folio_size(folio);
6140 if (folio_pos(folio) + len > size)
6141 len = size - folio_pos(folio);
6142
6143 err = __block_write_begin(&folio->page, 0, len, ext4_get_block);
6144 if (!err) {
6145 ret = VM_FAULT_SIGBUS;
6146 if (ext4_journal_folio_buffers(handle, folio, len))
6147 goto out_error;
6148 } else {
6149 folio_unlock(folio);
6150 }
6151 }
6152 ext4_journal_stop(handle);
6153 if (err == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
6154 goto retry_alloc;
6155 out_ret:
6156 ret = vmf_fs_error(err);
6157 out:
6158 filemap_invalidate_unlock_shared(mapping);
6159 sb_end_pagefault(inode->i_sb);
6160 return ret;
6161 out_error:
6162 folio_unlock(folio);
6163 ext4_journal_stop(handle);
6164 goto out;
6165 }
6166