Line data Source code
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 :
52 0 : static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw,
53 : struct ext4_inode_info *ei)
54 : {
55 0 : struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
56 0 : __u32 csum;
57 0 : __u16 dummy_csum = 0;
58 0 : int offset = offsetof(struct ext4_inode, i_checksum_lo);
59 0 : unsigned int csum_size = sizeof(dummy_csum);
60 :
61 0 : csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw, offset);
62 0 : csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, csum_size);
63 0 : offset += csum_size;
64 0 : csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
65 : EXT4_GOOD_OLD_INODE_SIZE - offset);
66 :
67 0 : if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
68 0 : offset = offsetof(struct ext4_inode, i_checksum_hi);
69 0 : csum = ext4_chksum(sbi, csum, (__u8 *)raw +
70 : EXT4_GOOD_OLD_INODE_SIZE,
71 : offset - EXT4_GOOD_OLD_INODE_SIZE);
72 0 : if (EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
73 0 : csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum,
74 : csum_size);
75 0 : offset += csum_size;
76 : }
77 0 : csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
78 0 : EXT4_INODE_SIZE(inode->i_sb) - offset);
79 : }
80 :
81 0 : return csum;
82 : }
83 :
84 4935 : static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
85 : struct ext4_inode_info *ei)
86 : {
87 4935 : __u32 provided, calculated;
88 :
89 4935 : if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
90 4935 : cpu_to_le32(EXT4_OS_LINUX) ||
91 4935 : !ext4_has_metadata_csum(inode->i_sb))
92 4935 : return 1;
93 :
94 0 : provided = le16_to_cpu(raw->i_checksum_lo);
95 0 : calculated = ext4_inode_csum(inode, raw, ei);
96 0 : if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
97 0 : EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
98 0 : provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
99 : else
100 0 : calculated &= 0xFFFF;
101 :
102 0 : return provided == calculated;
103 : }
104 :
105 10433 : void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
106 : struct ext4_inode_info *ei)
107 : {
108 10433 : __u32 csum;
109 :
110 10433 : if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
111 10433 : cpu_to_le32(EXT4_OS_LINUX) ||
112 10433 : !ext4_has_metadata_csum(inode->i_sb))
113 10433 : return;
114 :
115 0 : csum = ext4_inode_csum(inode, raw, ei);
116 0 : raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
117 0 : if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
118 0 : EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
119 0 : raw->i_checksum_hi = cpu_to_le16(csum >> 16);
120 : }
121 :
122 123 : static inline int ext4_begin_ordered_truncate(struct inode *inode,
123 : loff_t new_size)
124 : {
125 123 : 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 : */
132 123 : if (!EXT4_I(inode)->jinode)
133 : return 0;
134 79 : return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
135 79 : EXT4_I(inode)->jinode,
136 : new_size);
137 : }
138 :
139 : static void ext4_invalidatepage(struct page *page, unsigned int offset,
140 : unsigned int length);
141 : static int __ext4_journalled_writepage(struct page *page, unsigned int len);
142 : static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
143 : static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
144 : int pextents);
145 :
146 : /*
147 : * Test whether an inode is a fast symlink.
148 : * A fast symlink has its symlink data stored in ext4_inode_info->i_data.
149 : */
150 1849 : int ext4_inode_is_fast_symlink(struct inode *inode)
151 : {
152 1849 : if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
153 1849 : int ea_blocks = EXT4_I(inode)->i_file_acl ?
154 0 : EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0;
155 :
156 1849 : if (ext4_has_inline_data(inode))
157 : return 0;
158 :
159 1849 : return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
160 : }
161 0 : return S_ISLNK(inode->i_mode) && inode->i_size &&
162 : (inode->i_size < EXT4_N_BLOCKS * 4);
163 : }
164 :
165 : /*
166 : * Called at the last iput() if i_nlink is zero.
167 : */
168 193 : void ext4_evict_inode(struct inode *inode)
169 : {
170 193 : handle_t *handle;
171 193 : int err;
172 : /*
173 : * Credits for final inode cleanup and freeing:
174 : * sb + inode (ext4_orphan_del()), block bitmap, group descriptor
175 : * (xattr block freeing), bitmap, group descriptor (inode freeing)
176 : */
177 193 : int extra_credits = 6;
178 193 : struct ext4_xattr_inode_array *ea_inode_array = NULL;
179 193 : bool freeze_protected = false;
180 :
181 193 : trace_ext4_evict_inode(inode);
182 :
183 193 : if (inode->i_nlink) {
184 : /*
185 : * When journalling data dirty buffers are tracked only in the
186 : * journal. So although mm thinks everything is clean and
187 : * ready for reaping the inode might still have some pages to
188 : * write in the running transaction or waiting to be
189 : * checkpointed. Thus calling jbd2_journal_invalidatepage()
190 : * (via truncate_inode_pages()) to discard these buffers can
191 : * cause data loss. Also even if we did not discard these
192 : * buffers, we would have no way to find them after the inode
193 : * is reaped and thus user could see stale data if he tries to
194 : * read them before the transaction is checkpointed. So be
195 : * careful and force everything to disk here... We use
196 : * ei->i_datasync_tid to store the newest transaction
197 : * containing inode's data.
198 : *
199 : * Note that directories do not have this problem because they
200 : * don't use page cache.
201 : */
202 0 : if (inode->i_ino != EXT4_JOURNAL_INO &&
203 0 : ext4_should_journal_data(inode) &&
204 0 : (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) &&
205 0 : inode->i_data.nrpages) {
206 0 : journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
207 0 : tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
208 :
209 0 : jbd2_complete_transaction(journal, commit_tid);
210 0 : filemap_write_and_wait(&inode->i_data);
211 : }
212 0 : truncate_inode_pages_final(&inode->i_data);
213 :
214 0 : goto no_delete;
215 : }
216 :
217 193 : if (is_bad_inode(inode))
218 0 : goto no_delete;
219 193 : dquot_initialize(inode);
220 :
221 193 : if (ext4_should_order_data(inode))
222 122 : ext4_begin_ordered_truncate(inode, 0);
223 193 : truncate_inode_pages_final(&inode->i_data);
224 :
225 : /*
226 : * For inodes with journalled data, transaction commit could have
227 : * dirtied the inode. Flush worker is ignoring it because of I_FREEING
228 : * flag but we still need to remove the inode from the writeback lists.
229 : */
230 193 : if (!list_empty_careful(&inode->i_io_list)) {
231 0 : WARN_ON_ONCE(!ext4_should_journal_data(inode));
232 0 : inode_io_list_del(inode);
233 : }
234 :
235 : /*
236 : * Protect us against freezing - iput() caller didn't have to have any
237 : * protection against it. When we are in a running transaction though,
238 : * we are already protected against freezing and we cannot grab further
239 : * protection due to lock ordering constraints.
240 : */
241 193 : if (!ext4_journal_current_handle()) {
242 193 : sb_start_intwrite(inode->i_sb);
243 193 : freeze_protected = true;
244 : }
245 :
246 193 : if (!IS_NOQUOTA(inode))
247 : extra_credits += EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb);
248 :
249 : /*
250 : * Block bitmap, group descriptor, and inode are accounted in both
251 : * ext4_blocks_for_truncate() and extra_credits. So subtract 3.
252 : */
253 193 : handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
254 : ext4_blocks_for_truncate(inode) + extra_credits - 3);
255 193 : if (IS_ERR(handle)) {
256 0 : ext4_std_error(inode->i_sb, PTR_ERR(handle));
257 : /*
258 : * If we're going to skip the normal cleanup, we still need to
259 : * make sure that the in-core orphan linked list is properly
260 : * cleaned up.
261 : */
262 0 : ext4_orphan_del(NULL, inode);
263 0 : if (freeze_protected)
264 0 : sb_end_intwrite(inode->i_sb);
265 0 : goto no_delete;
266 : }
267 :
268 193 : if (IS_SYNC(inode))
269 0 : ext4_handle_sync(handle);
270 :
271 : /*
272 : * Set inode->i_size to 0 before calling ext4_truncate(). We need
273 : * special handling of symlinks here because i_size is used to
274 : * determine whether ext4_inode_info->i_data contains symlink data or
275 : * block mappings. Setting i_size to 0 will remove its fast symlink
276 : * status. Erase i_data so that it becomes a valid empty block map.
277 : */
278 193 : if (ext4_inode_is_fast_symlink(inode))
279 0 : memset(EXT4_I(inode)->i_data, 0, sizeof(EXT4_I(inode)->i_data));
280 193 : inode->i_size = 0;
281 193 : err = ext4_mark_inode_dirty(handle, inode);
282 193 : if (err) {
283 0 : ext4_warning(inode->i_sb,
284 : "couldn't mark inode dirty (err %d)", err);
285 0 : goto stop_handle;
286 : }
287 193 : if (inode->i_blocks) {
288 135 : err = ext4_truncate(inode);
289 135 : if (err) {
290 0 : ext4_error_err(inode->i_sb, -err,
291 : "couldn't truncate inode %lu (err %d)",
292 : inode->i_ino, err);
293 0 : goto stop_handle;
294 : }
295 : }
296 :
297 : /* Remove xattr references. */
298 193 : err = ext4_xattr_delete_inode(handle, inode, &ea_inode_array,
299 : extra_credits);
300 193 : if (err) {
301 0 : ext4_warning(inode->i_sb, "xattr delete (err %d)", err);
302 0 : stop_handle:
303 0 : ext4_journal_stop(handle);
304 0 : ext4_orphan_del(NULL, inode);
305 0 : if (freeze_protected)
306 0 : sb_end_intwrite(inode->i_sb);
307 0 : ext4_xattr_inode_array_free(ea_inode_array);
308 0 : goto no_delete;
309 : }
310 :
311 : /*
312 : * Kill off the orphan record which ext4_truncate created.
313 : * AKPM: I think this can be inside the above `if'.
314 : * Note that ext4_orphan_del() has to be able to cope with the
315 : * deletion of a non-existent orphan - this is because we don't
316 : * know if ext4_truncate() actually created an orphan record.
317 : * (Well, we could do this if we need to, but heck - it works)
318 : */
319 193 : ext4_orphan_del(handle, inode);
320 193 : EXT4_I(inode)->i_dtime = (__u32)ktime_get_real_seconds();
321 :
322 : /*
323 : * One subtle ordering requirement: if anything has gone wrong
324 : * (transaction abort, IO errors, whatever), then we can still
325 : * do these next steps (the fs will already have been marked as
326 : * having errors), but we can't free the inode if the mark_dirty
327 : * fails.
328 : */
329 193 : if (ext4_mark_inode_dirty(handle, inode))
330 : /* If that failed, just do the required in-core inode clear. */
331 0 : ext4_clear_inode(inode);
332 : else
333 193 : ext4_free_inode(handle, inode);
334 193 : ext4_journal_stop(handle);
335 193 : if (freeze_protected)
336 193 : sb_end_intwrite(inode->i_sb);
337 193 : ext4_xattr_inode_array_free(ea_inode_array);
338 193 : return;
339 0 : no_delete:
340 0 : if (!list_empty(&EXT4_I(inode)->i_fc_list))
341 0 : ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_NOMEM);
342 0 : ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
343 : }
344 :
345 : #ifdef CONFIG_QUOTA
346 : qsize_t *ext4_get_reserved_space(struct inode *inode)
347 : {
348 : return &EXT4_I(inode)->i_reserved_quota;
349 : }
350 : #endif
351 :
352 : /*
353 : * Called with i_data_sem down, which is important since we can call
354 : * ext4_discard_preallocations() from here.
355 : */
356 70 : void ext4_da_update_reserve_space(struct inode *inode,
357 : int used, int quota_claim)
358 : {
359 70 : struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
360 70 : struct ext4_inode_info *ei = EXT4_I(inode);
361 :
362 70 : spin_lock(&ei->i_block_reservation_lock);
363 70 : trace_ext4_da_update_reserve_space(inode, used, quota_claim);
364 70 : if (unlikely(used > ei->i_reserved_data_blocks)) {
365 0 : ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
366 : "with only %d reserved data blocks",
367 : __func__, inode->i_ino, used,
368 : ei->i_reserved_data_blocks);
369 0 : WARN_ON(1);
370 0 : used = ei->i_reserved_data_blocks;
371 : }
372 :
373 : /* Update per-inode reservations */
374 70 : ei->i_reserved_data_blocks -= used;
375 70 : percpu_counter_sub(&sbi->s_dirtyclusters_counter, used);
376 :
377 70 : spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
378 :
379 : /* Update quota subsystem for data blocks */
380 70 : if (quota_claim)
381 70 : dquot_claim_block(inode, EXT4_C2B(sbi, used));
382 : else {
383 : /*
384 : * We did fallocate with an offset that is already delayed
385 : * allocated. So on delayed allocated writeback we should
386 : * not re-claim the quota for fallocated blocks.
387 : */
388 70 : dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
389 : }
390 :
391 : /*
392 : * If we have done all the pending block allocations and if
393 : * there aren't any writers on the inode, we can discard the
394 : * inode's preallocations.
395 : */
396 70 : if ((ei->i_reserved_data_blocks == 0) &&
397 63 : !inode_is_open_for_write(inode))
398 9 : ext4_discard_preallocations(inode, 0);
399 70 : }
400 :
401 16097 : static int __check_block_validity(struct inode *inode, const char *func,
402 : unsigned int line,
403 : struct ext4_map_blocks *map)
404 : {
405 16097 : if (ext4_has_feature_journal(inode->i_sb) &&
406 16097 : (inode->i_ino ==
407 16097 : le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_journal_inum)))
408 : return 0;
409 14148 : if (!ext4_inode_block_valid(inode, map->m_pblk, map->m_len)) {
410 0 : ext4_error_inode(inode, func, line, map->m_pblk,
411 : "lblock %lu mapped to illegal pblock %llu "
412 : "(length %d)", (unsigned long) map->m_lblk,
413 : map->m_pblk, map->m_len);
414 0 : return -EFSCORRUPTED;
415 : }
416 : return 0;
417 : }
418 :
419 0 : int ext4_issue_zeroout(struct inode *inode, ext4_lblk_t lblk, ext4_fsblk_t pblk,
420 : ext4_lblk_t len)
421 : {
422 0 : int ret;
423 :
424 0 : if (IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode))
425 0 : return fscrypt_zeroout_range(inode, lblk, pblk, len);
426 :
427 0 : ret = sb_issue_zeroout(inode->i_sb, pblk, len, GFP_NOFS);
428 0 : if (ret > 0)
429 : ret = 0;
430 :
431 : return ret;
432 : }
433 :
434 : #define check_block_validity(inode, map) \
435 : __check_block_validity((inode), __func__, __LINE__, (map))
436 :
437 : #ifdef ES_AGGRESSIVE_TEST
438 : static void ext4_map_blocks_es_recheck(handle_t *handle,
439 : struct inode *inode,
440 : struct ext4_map_blocks *es_map,
441 : struct ext4_map_blocks *map,
442 : int flags)
443 : {
444 : int retval;
445 :
446 : map->m_flags = 0;
447 : /*
448 : * There is a race window that the result is not the same.
449 : * e.g. xfstests #223 when dioread_nolock enables. The reason
450 : * is that we lookup a block mapping in extent status tree with
451 : * out taking i_data_sem. So at the time the unwritten extent
452 : * could be converted.
453 : */
454 : down_read(&EXT4_I(inode)->i_data_sem);
455 : if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
456 : retval = ext4_ext_map_blocks(handle, inode, map, 0);
457 : } else {
458 : retval = ext4_ind_map_blocks(handle, inode, map, 0);
459 : }
460 : up_read((&EXT4_I(inode)->i_data_sem));
461 :
462 : /*
463 : * We don't check m_len because extent will be collpased in status
464 : * tree. So the m_len might not equal.
465 : */
466 : if (es_map->m_lblk != map->m_lblk ||
467 : es_map->m_flags != map->m_flags ||
468 : es_map->m_pblk != map->m_pblk) {
469 : printk("ES cache assertion failed for inode: %lu "
470 : "es_cached ex [%d/%d/%llu/%x] != "
471 : "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
472 : inode->i_ino, es_map->m_lblk, es_map->m_len,
473 : es_map->m_pblk, es_map->m_flags, map->m_lblk,
474 : map->m_len, map->m_pblk, map->m_flags,
475 : retval, flags);
476 : }
477 : }
478 : #endif /* ES_AGGRESSIVE_TEST */
479 :
480 : /*
481 : * The ext4_map_blocks() function tries to look up the requested blocks,
482 : * and returns if the blocks are already mapped.
483 : *
484 : * Otherwise it takes the write lock of the i_data_sem and allocate blocks
485 : * and store the allocated blocks in the result buffer head and mark it
486 : * mapped.
487 : *
488 : * If file type is extents based, it will call ext4_ext_map_blocks(),
489 : * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
490 : * based files
491 : *
492 : * On success, it returns the number of blocks being mapped or allocated. if
493 : * create==0 and the blocks are pre-allocated and unwritten, the resulting @map
494 : * is marked as unwritten. If the create == 1, it will mark @map as mapped.
495 : *
496 : * It returns 0 if plain look up failed (blocks have not been allocated), in
497 : * that case, @map is returned as unmapped but we still do fill map->m_len to
498 : * indicate the length of a hole starting at map->m_lblk.
499 : *
500 : * It returns the error in case of allocation failure.
501 : */
502 16117 : int ext4_map_blocks(handle_t *handle, struct inode *inode,
503 : struct ext4_map_blocks *map, int flags)
504 : {
505 16117 : struct extent_status es;
506 16117 : int retval;
507 16117 : int ret = 0;
508 : #ifdef ES_AGGRESSIVE_TEST
509 : struct ext4_map_blocks orig_map;
510 :
511 : memcpy(&orig_map, map, sizeof(*map));
512 : #endif
513 :
514 16117 : map->m_flags = 0;
515 16117 : ext_debug(inode, "flag 0x%x, max_blocks %u, logical block %lu\n",
516 : flags, map->m_len, (unsigned long) map->m_lblk);
517 :
518 : /*
519 : * ext4_map_blocks returns an int, and m_len is an unsigned int
520 : */
521 16117 : if (unlikely(map->m_len > INT_MAX))
522 0 : map->m_len = INT_MAX;
523 :
524 : /* We can handle the block number less than EXT_MAX_BLOCKS */
525 16117 : if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
526 : return -EFSCORRUPTED;
527 :
528 : /* Lookup extent status tree firstly */
529 32234 : if (!(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY) &&
530 16117 : ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) {
531 14918 : if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
532 14829 : map->m_pblk = ext4_es_pblock(&es) +
533 14829 : map->m_lblk - es.es_lblk;
534 14829 : map->m_flags |= ext4_es_is_written(&es) ?
535 14829 : EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
536 14829 : retval = es.es_len - (map->m_lblk - es.es_lblk);
537 14829 : if (retval > map->m_len)
538 : retval = map->m_len;
539 14829 : map->m_len = retval;
540 89 : } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
541 89 : map->m_pblk = 0;
542 89 : retval = es.es_len - (map->m_lblk - es.es_lblk);
543 89 : if (retval > map->m_len)
544 : retval = map->m_len;
545 89 : map->m_len = retval;
546 89 : retval = 0;
547 : } else {
548 0 : BUG();
549 : }
550 : #ifdef ES_AGGRESSIVE_TEST
551 : ext4_map_blocks_es_recheck(handle, inode, map,
552 : &orig_map, flags);
553 : #endif
554 14829 : goto found;
555 : }
556 :
557 : /*
558 : * Try to see if we can get the block without requesting a new
559 : * file system block.
560 : */
561 1199 : down_read(&EXT4_I(inode)->i_data_sem);
562 1199 : if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
563 1199 : retval = ext4_ext_map_blocks(handle, inode, map, 0);
564 : } else {
565 0 : retval = ext4_ind_map_blocks(handle, inode, map, 0);
566 : }
567 1199 : if (retval > 0) {
568 1026 : unsigned int status;
569 :
570 1026 : if (unlikely(retval != map->m_len)) {
571 0 : ext4_warning(inode->i_sb,
572 : "ES len assertion failed for inode "
573 : "%lu: retval %d != map->m_len %d",
574 : inode->i_ino, retval, map->m_len);
575 0 : WARN_ON(1);
576 : }
577 :
578 2052 : status = map->m_flags & EXT4_MAP_UNWRITTEN ?
579 1026 : EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
580 1026 : if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
581 1026 : !(status & EXTENT_STATUS_WRITTEN) &&
582 0 : ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk,
583 0 : map->m_lblk + map->m_len - 1))
584 0 : status |= EXTENT_STATUS_DELAYED;
585 1026 : ret = ext4_es_insert_extent(inode, map->m_lblk,
586 : map->m_len, map->m_pblk, status);
587 1026 : if (ret < 0)
588 0 : retval = ret;
589 : }
590 1199 : up_read((&EXT4_I(inode)->i_data_sem));
591 :
592 16028 : found:
593 16117 : if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
594 15785 : ret = check_block_validity(inode, map);
595 15785 : if (ret != 0)
596 : return ret;
597 : }
598 :
599 : /* If it is only a block(s) look up */
600 16117 : if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
601 : return retval;
602 :
603 : /*
604 : * Returns if the blocks have already allocated
605 : *
606 : * Note that if blocks have been preallocated
607 : * ext4_ext_get_block() returns the create = 0
608 : * with buffer head unmapped.
609 : */
610 313 : if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
611 : /*
612 : * If we need to convert extent to unwritten
613 : * we continue and do the actual work in
614 : * ext4_ext_map_blocks()
615 : */
616 0 : if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
617 : return retval;
618 :
619 : /*
620 : * Here we clear m_flags because after allocating an new extent,
621 : * it will be set again.
622 : */
623 313 : map->m_flags &= ~EXT4_MAP_FLAGS;
624 :
625 : /*
626 : * New blocks allocate and/or writing to unwritten extent
627 : * will possibly result in updating i_data, so we take
628 : * the write lock of i_data_sem, and call get_block()
629 : * with create == 1 flag.
630 : */
631 313 : down_write(&EXT4_I(inode)->i_data_sem);
632 :
633 : /*
634 : * We need to check for EXT4 here because migrate
635 : * could have changed the inode type in between
636 : */
637 313 : if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
638 313 : retval = ext4_ext_map_blocks(handle, inode, map, flags);
639 : } else {
640 0 : retval = ext4_ind_map_blocks(handle, inode, map, flags);
641 :
642 0 : if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
643 : /*
644 : * We allocated new blocks which will result in
645 : * i_data's format changing. Force the migrate
646 : * to fail by clearing migrate flags
647 : */
648 0 : ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
649 : }
650 :
651 : /*
652 : * Update reserved blocks/metadata blocks after successful
653 : * block allocation which had been deferred till now. We don't
654 : * support fallocate for non extent files. So we can update
655 : * reserve space here.
656 : */
657 0 : if ((retval > 0) &&
658 0 : (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
659 0 : ext4_da_update_reserve_space(inode, retval, 1);
660 : }
661 :
662 313 : if (retval > 0) {
663 313 : unsigned int status;
664 :
665 313 : if (unlikely(retval != map->m_len)) {
666 0 : ext4_warning(inode->i_sb,
667 : "ES len assertion failed for inode "
668 : "%lu: retval %d != map->m_len %d",
669 : inode->i_ino, retval, map->m_len);
670 0 : WARN_ON(1);
671 : }
672 :
673 : /*
674 : * We have to zeroout blocks before inserting them into extent
675 : * status tree. Otherwise someone could look them up there and
676 : * use them before they are really zeroed. We also have to
677 : * unmap metadata before zeroing as otherwise writeback can
678 : * overwrite zeros with stale data from block device.
679 : */
680 313 : if (flags & EXT4_GET_BLOCKS_ZERO &&
681 0 : map->m_flags & EXT4_MAP_MAPPED &&
682 : map->m_flags & EXT4_MAP_NEW) {
683 0 : ret = ext4_issue_zeroout(inode, map->m_lblk,
684 : map->m_pblk, map->m_len);
685 0 : if (ret) {
686 0 : retval = ret;
687 0 : goto out_sem;
688 : }
689 : }
690 :
691 : /*
692 : * If the extent has been zeroed out, we don't need to update
693 : * extent status tree.
694 : */
695 383 : if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
696 70 : ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) {
697 70 : if (ext4_es_is_written(&es))
698 0 : goto out_sem;
699 : }
700 626 : status = map->m_flags & EXT4_MAP_UNWRITTEN ?
701 313 : EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
702 313 : if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
703 243 : !(status & EXTENT_STATUS_WRITTEN) &&
704 0 : ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk,
705 0 : map->m_lblk + map->m_len - 1))
706 0 : status |= EXTENT_STATUS_DELAYED;
707 313 : ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
708 : map->m_pblk, status);
709 313 : if (ret < 0) {
710 0 : retval = ret;
711 0 : goto out_sem;
712 : }
713 : }
714 :
715 313 : out_sem:
716 313 : up_write((&EXT4_I(inode)->i_data_sem));
717 313 : if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
718 313 : ret = check_block_validity(inode, map);
719 313 : if (ret != 0)
720 : return ret;
721 :
722 : /*
723 : * Inodes with freshly allocated blocks where contents will be
724 : * visible after transaction commit must be on transaction's
725 : * ordered data list.
726 : */
727 313 : if (map->m_flags & EXT4_MAP_NEW &&
728 173 : !(map->m_flags & EXT4_MAP_UNWRITTEN) &&
729 173 : !(flags & EXT4_GET_BLOCKS_ZERO) &&
730 173 : !ext4_is_quota_file(inode) &&
731 173 : ext4_should_order_data(inode)) {
732 0 : loff_t start_byte =
733 0 : (loff_t)map->m_lblk << inode->i_blkbits;
734 0 : loff_t length = (loff_t)map->m_len << inode->i_blkbits;
735 :
736 0 : if (flags & EXT4_GET_BLOCKS_IO_SUBMIT)
737 313 : ret = ext4_jbd2_inode_add_wait(handle, inode,
738 : start_byte, length);
739 : else
740 0 : ret = ext4_jbd2_inode_add_write(handle, inode,
741 : start_byte, length);
742 0 : if (ret)
743 : return ret;
744 : }
745 313 : ext4_fc_track_range(handle, inode, map->m_lblk,
746 313 : map->m_lblk + map->m_len - 1);
747 : }
748 :
749 : if (retval < 0)
750 : ext_debug(inode, "failed with err %d\n", retval);
751 : return retval;
752 : }
753 :
754 : /*
755 : * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
756 : * we have to be careful as someone else may be manipulating b_state as well.
757 : */
758 105 : static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags)
759 : {
760 105 : unsigned long old_state;
761 105 : unsigned long new_state;
762 :
763 105 : flags &= EXT4_MAP_FLAGS;
764 :
765 : /* Dummy buffer_head? Set non-atomically. */
766 105 : if (!bh->b_page) {
767 0 : bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags;
768 0 : return;
769 : }
770 : /*
771 : * Someone else may be modifying b_state. Be careful! This is ugly but
772 : * once we get rid of using bh as a container for mapping information
773 : * to pass to / from get_block functions, this can go away.
774 : */
775 105 : do {
776 105 : old_state = READ_ONCE(bh->b_state);
777 105 : new_state = (old_state & ~EXT4_MAP_FLAGS) | flags;
778 105 : } while (unlikely(
779 : cmpxchg(&bh->b_state, old_state, new_state) != old_state));
780 : }
781 :
782 0 : static int _ext4_get_block(struct inode *inode, sector_t iblock,
783 : struct buffer_head *bh, int flags)
784 : {
785 0 : struct ext4_map_blocks map;
786 0 : int ret = 0;
787 :
788 0 : if (ext4_has_inline_data(inode))
789 : return -ERANGE;
790 :
791 0 : map.m_lblk = iblock;
792 0 : map.m_len = bh->b_size >> inode->i_blkbits;
793 :
794 0 : ret = ext4_map_blocks(ext4_journal_current_handle(), inode, &map,
795 : flags);
796 0 : if (ret > 0) {
797 0 : map_bh(bh, inode->i_sb, map.m_pblk);
798 0 : ext4_update_bh_state(bh, map.m_flags);
799 0 : bh->b_size = inode->i_sb->s_blocksize * map.m_len;
800 0 : ret = 0;
801 0 : } else if (ret == 0) {
802 : /* hole case, need to fill in bh->b_size */
803 0 : bh->b_size = inode->i_sb->s_blocksize * map.m_len;
804 : }
805 : return ret;
806 : }
807 :
808 0 : int ext4_get_block(struct inode *inode, sector_t iblock,
809 : struct buffer_head *bh, int create)
810 : {
811 0 : return _ext4_get_block(inode, iblock, bh,
812 : create ? EXT4_GET_BLOCKS_CREATE : 0);
813 : }
814 :
815 : /*
816 : * Get block function used when preparing for buffered write if we require
817 : * creating an unwritten extent if blocks haven't been allocated. The extent
818 : * will be converted to written after the IO is complete.
819 : */
820 0 : int ext4_get_block_unwritten(struct inode *inode, sector_t iblock,
821 : struct buffer_head *bh_result, int create)
822 : {
823 0 : ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n",
824 : inode->i_ino, create);
825 0 : return _ext4_get_block(inode, iblock, bh_result,
826 : EXT4_GET_BLOCKS_IO_CREATE_EXT);
827 : }
828 :
829 : /* Maximum number of blocks we map for direct IO at once. */
830 : #define DIO_MAX_BLOCKS 4096
831 :
832 : /*
833 : * `handle' can be NULL if create is zero
834 : */
835 12332 : struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
836 : ext4_lblk_t block, int map_flags)
837 : {
838 12332 : struct ext4_map_blocks map;
839 12332 : struct buffer_head *bh;
840 12332 : int create = map_flags & EXT4_GET_BLOCKS_CREATE;
841 12332 : int err;
842 :
843 12332 : ASSERT((EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
844 : || handle != NULL || create == 0);
845 :
846 12332 : map.m_lblk = block;
847 12332 : map.m_len = 1;
848 12332 : err = ext4_map_blocks(handle, inode, &map, map_flags);
849 :
850 12332 : if (err == 0)
851 0 : return create ? ERR_PTR(-ENOSPC) : NULL;
852 12332 : if (err < 0)
853 0 : return ERR_PTR(err);
854 :
855 12332 : bh = sb_getblk(inode->i_sb, map.m_pblk);
856 12332 : if (unlikely(!bh))
857 12332 : return ERR_PTR(-ENOMEM);
858 12332 : if (map.m_flags & EXT4_MAP_NEW) {
859 173 : ASSERT(create != 0);
860 173 : ASSERT((EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY)
861 : || (handle != NULL));
862 :
863 : /*
864 : * Now that we do not always journal data, we should
865 : * keep in mind whether this should always journal the
866 : * new buffer as metadata. For now, regular file
867 : * writes use ext4_get_block instead, so it's not a
868 : * problem.
869 : */
870 173 : lock_buffer(bh);
871 173 : BUFFER_TRACE(bh, "call get_create_access");
872 173 : err = ext4_journal_get_create_access(handle, bh);
873 173 : if (unlikely(err)) {
874 0 : unlock_buffer(bh);
875 0 : goto errout;
876 : }
877 173 : if (!buffer_uptodate(bh)) {
878 102 : memset(bh->b_data, 0, inode->i_sb->s_blocksize);
879 102 : set_buffer_uptodate(bh);
880 : }
881 173 : unlock_buffer(bh);
882 173 : BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
883 173 : err = ext4_handle_dirty_metadata(handle, inode, bh);
884 173 : if (unlikely(err))
885 0 : goto errout;
886 : } else
887 : BUFFER_TRACE(bh, "not a new buffer");
888 : return bh;
889 0 : errout:
890 0 : brelse(bh);
891 0 : return ERR_PTR(err);
892 : }
893 :
894 8161 : struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
895 : ext4_lblk_t block, int map_flags)
896 : {
897 8161 : struct buffer_head *bh;
898 8161 : int ret;
899 :
900 8161 : bh = ext4_getblk(handle, inode, block, map_flags);
901 8161 : if (IS_ERR(bh))
902 : return bh;
903 8161 : if (!bh || ext4_buffer_uptodate(bh))
904 7694 : return bh;
905 :
906 467 : ret = ext4_read_bh_lock(bh, REQ_META | REQ_PRIO, true);
907 467 : if (ret) {
908 0 : put_bh(bh);
909 0 : return ERR_PTR(ret);
910 : }
911 : return bh;
912 : }
913 :
914 : /* Read a contiguous batch of blocks. */
915 4171 : int ext4_bread_batch(struct inode *inode, ext4_lblk_t block, int bh_count,
916 : bool wait, struct buffer_head **bhs)
917 : {
918 4171 : int i, err;
919 :
920 8342 : for (i = 0; i < bh_count; i++) {
921 4171 : bhs[i] = ext4_getblk(NULL, inode, block + i, 0 /* map_flags */);
922 4171 : if (IS_ERR(bhs[i])) {
923 0 : err = PTR_ERR(bhs[i]);
924 0 : bh_count = i;
925 0 : goto out_brelse;
926 : }
927 : }
928 :
929 8342 : for (i = 0; i < bh_count; i++)
930 : /* Note that NULL bhs[i] is valid because of holes. */
931 4171 : if (bhs[i] && !ext4_buffer_uptodate(bhs[i]))
932 105 : ext4_read_bh_lock(bhs[i], REQ_META | REQ_PRIO, false);
933 :
934 4171 : if (!wait)
935 : return 0;
936 :
937 0 : for (i = 0; i < bh_count; i++)
938 0 : if (bhs[i])
939 0 : wait_on_buffer(bhs[i]);
940 :
941 0 : for (i = 0; i < bh_count; i++) {
942 0 : if (bhs[i] && !buffer_uptodate(bhs[i])) {
943 0 : err = -EIO;
944 0 : goto out_brelse;
945 : }
946 : }
947 : return 0;
948 :
949 0 : out_brelse:
950 0 : for (i = 0; i < bh_count; i++) {
951 0 : brelse(bhs[i]);
952 0 : bhs[i] = NULL;
953 : }
954 : return err;
955 : }
956 :
957 2 : int ext4_walk_page_buffers(handle_t *handle,
958 : struct buffer_head *head,
959 : unsigned from,
960 : unsigned to,
961 : int *partial,
962 : int (*fn)(handle_t *handle,
963 : struct buffer_head *bh))
964 : {
965 2 : struct buffer_head *bh;
966 2 : unsigned block_start, block_end;
967 2 : unsigned blocksize = head->b_size;
968 2 : int err, ret = 0;
969 2 : struct buffer_head *next;
970 :
971 2 : for (bh = head, block_start = 0;
972 4 : ret == 0 && (bh != head || !block_start);
973 : block_start = block_end, bh = next) {
974 2 : next = bh->b_this_page;
975 2 : block_end = block_start + blocksize;
976 2 : if (block_end <= from || block_start >= to) {
977 0 : if (partial && !buffer_uptodate(bh))
978 0 : *partial = 1;
979 0 : continue;
980 : }
981 2 : err = (*fn)(handle, bh);
982 2 : if (!ret)
983 2 : ret = err;
984 : }
985 2 : return ret;
986 : }
987 :
988 : /*
989 : * To preserve ordering, it is essential that the hole instantiation and
990 : * the data write be encapsulated in a single transaction. We cannot
991 : * close off a transaction and start a new one between the ext4_get_block()
992 : * and the commit_write(). So doing the jbd2_journal_start at the start of
993 : * prepare_write() is the right place.
994 : *
995 : * Also, this function can nest inside ext4_writepage(). In that case, we
996 : * *know* that ext4_writepage() has generated enough buffer credits to do the
997 : * whole page. So we won't block on the journal in that case, which is good,
998 : * because the caller may be PF_MEMALLOC.
999 : *
1000 : * By accident, ext4 can be reentered when a transaction is open via
1001 : * quota file writes. If we were to commit the transaction while thus
1002 : * reentered, there can be a deadlock - we would be holding a quota
1003 : * lock, and the commit would never complete if another thread had a
1004 : * transaction open and was blocking on the quota lock - a ranking
1005 : * violation.
1006 : *
1007 : * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1008 : * will _not_ run commit under these circumstances because handle->h_ref
1009 : * is elevated. We'll still have enough credits for the tiny quotafile
1010 : * write.
1011 : */
1012 0 : int do_journal_get_write_access(handle_t *handle,
1013 : struct buffer_head *bh)
1014 : {
1015 0 : int dirty = buffer_dirty(bh);
1016 0 : int ret;
1017 :
1018 0 : if (!buffer_mapped(bh) || buffer_freed(bh))
1019 0 : return 0;
1020 : /*
1021 : * __block_write_begin() could have dirtied some buffers. Clean
1022 : * the dirty bit as jbd2_journal_get_write_access() could complain
1023 : * otherwise about fs integrity issues. Setting of the dirty bit
1024 : * by __block_write_begin() isn't a real problem here as we clear
1025 : * the bit before releasing a page lock and thus writeback cannot
1026 : * ever write the buffer.
1027 : */
1028 0 : if (dirty)
1029 0 : clear_buffer_dirty(bh);
1030 0 : BUFFER_TRACE(bh, "get write access");
1031 0 : ret = ext4_journal_get_write_access(handle, bh);
1032 0 : if (!ret && dirty)
1033 0 : ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1034 : return ret;
1035 : }
1036 :
1037 : #ifdef CONFIG_FS_ENCRYPTION
1038 : static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len,
1039 : get_block_t *get_block)
1040 : {
1041 : unsigned from = pos & (PAGE_SIZE - 1);
1042 : unsigned to = from + len;
1043 : struct inode *inode = page->mapping->host;
1044 : unsigned block_start, block_end;
1045 : sector_t block;
1046 : int err = 0;
1047 : unsigned blocksize = inode->i_sb->s_blocksize;
1048 : unsigned bbits;
1049 : struct buffer_head *bh, *head, *wait[2];
1050 : int nr_wait = 0;
1051 : int i;
1052 :
1053 : BUG_ON(!PageLocked(page));
1054 : BUG_ON(from > PAGE_SIZE);
1055 : BUG_ON(to > PAGE_SIZE);
1056 : BUG_ON(from > to);
1057 :
1058 : if (!page_has_buffers(page))
1059 : create_empty_buffers(page, blocksize, 0);
1060 : head = page_buffers(page);
1061 : bbits = ilog2(blocksize);
1062 : block = (sector_t)page->index << (PAGE_SHIFT - bbits);
1063 :
1064 : for (bh = head, block_start = 0; bh != head || !block_start;
1065 : block++, block_start = block_end, bh = bh->b_this_page) {
1066 : block_end = block_start + blocksize;
1067 : if (block_end <= from || block_start >= to) {
1068 : if (PageUptodate(page)) {
1069 : if (!buffer_uptodate(bh))
1070 : set_buffer_uptodate(bh);
1071 : }
1072 : continue;
1073 : }
1074 : if (buffer_new(bh))
1075 : clear_buffer_new(bh);
1076 : if (!buffer_mapped(bh)) {
1077 : WARN_ON(bh->b_size != blocksize);
1078 : err = get_block(inode, block, bh, 1);
1079 : if (err)
1080 : break;
1081 : if (buffer_new(bh)) {
1082 : if (PageUptodate(page)) {
1083 : clear_buffer_new(bh);
1084 : set_buffer_uptodate(bh);
1085 : mark_buffer_dirty(bh);
1086 : continue;
1087 : }
1088 : if (block_end > to || block_start < from)
1089 : zero_user_segments(page, to, block_end,
1090 : block_start, from);
1091 : continue;
1092 : }
1093 : }
1094 : if (PageUptodate(page)) {
1095 : if (!buffer_uptodate(bh))
1096 : set_buffer_uptodate(bh);
1097 : continue;
1098 : }
1099 : if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
1100 : !buffer_unwritten(bh) &&
1101 : (block_start < from || block_end > to)) {
1102 : ext4_read_bh_lock(bh, 0, false);
1103 : wait[nr_wait++] = bh;
1104 : }
1105 : }
1106 : /*
1107 : * If we issued read requests, let them complete.
1108 : */
1109 : for (i = 0; i < nr_wait; i++) {
1110 : wait_on_buffer(wait[i]);
1111 : if (!buffer_uptodate(wait[i]))
1112 : err = -EIO;
1113 : }
1114 : if (unlikely(err)) {
1115 : page_zero_new_buffers(page, from, to);
1116 : } else if (fscrypt_inode_uses_fs_layer_crypto(inode)) {
1117 : for (i = 0; i < nr_wait; i++) {
1118 : int err2;
1119 :
1120 : err2 = fscrypt_decrypt_pagecache_blocks(page, blocksize,
1121 : bh_offset(wait[i]));
1122 : if (err2) {
1123 : clear_buffer_uptodate(wait[i]);
1124 : err = err2;
1125 : }
1126 : }
1127 : }
1128 :
1129 : return err;
1130 : }
1131 : #endif
1132 :
1133 0 : static int ext4_write_begin(struct file *file, struct address_space *mapping,
1134 : loff_t pos, unsigned len, unsigned flags,
1135 : struct page **pagep, void **fsdata)
1136 : {
1137 0 : struct inode *inode = mapping->host;
1138 0 : int ret, needed_blocks;
1139 0 : handle_t *handle;
1140 0 : int retries = 0;
1141 0 : struct page *page;
1142 0 : pgoff_t index;
1143 0 : unsigned from, to;
1144 :
1145 0 : if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
1146 : return -EIO;
1147 :
1148 0 : trace_ext4_write_begin(inode, pos, len, flags);
1149 : /*
1150 : * Reserve one block more for addition to orphan list in case
1151 : * we allocate blocks but write fails for some reason
1152 : */
1153 0 : needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1154 0 : index = pos >> PAGE_SHIFT;
1155 0 : from = pos & (PAGE_SIZE - 1);
1156 0 : to = from + len;
1157 :
1158 0 : if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1159 0 : ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1160 : flags, pagep);
1161 0 : if (ret < 0)
1162 : return ret;
1163 0 : if (ret == 1)
1164 : return 0;
1165 : }
1166 :
1167 : /*
1168 : * grab_cache_page_write_begin() can take a long time if the
1169 : * system is thrashing due to memory pressure, or if the page
1170 : * is being written back. So grab it first before we start
1171 : * the transaction handle. This also allows us to allocate
1172 : * the page (if needed) without using GFP_NOFS.
1173 : */
1174 0 : retry_grab:
1175 0 : page = grab_cache_page_write_begin(mapping, index, flags);
1176 0 : if (!page)
1177 : return -ENOMEM;
1178 0 : unlock_page(page);
1179 :
1180 0 : retry_journal:
1181 0 : handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1182 0 : if (IS_ERR(handle)) {
1183 0 : put_page(page);
1184 0 : return PTR_ERR(handle);
1185 : }
1186 :
1187 0 : lock_page(page);
1188 0 : if (page->mapping != mapping) {
1189 : /* The page got truncated from under us */
1190 0 : unlock_page(page);
1191 0 : put_page(page);
1192 0 : ext4_journal_stop(handle);
1193 0 : goto retry_grab;
1194 : }
1195 : /* In case writeback began while the page was unlocked */
1196 0 : wait_for_stable_page(page);
1197 :
1198 : #ifdef CONFIG_FS_ENCRYPTION
1199 : if (ext4_should_dioread_nolock(inode))
1200 : ret = ext4_block_write_begin(page, pos, len,
1201 : ext4_get_block_unwritten);
1202 : else
1203 : ret = ext4_block_write_begin(page, pos, len,
1204 : ext4_get_block);
1205 : #else
1206 0 : if (ext4_should_dioread_nolock(inode))
1207 0 : ret = __block_write_begin(page, pos, len,
1208 : ext4_get_block_unwritten);
1209 : else
1210 0 : ret = __block_write_begin(page, pos, len, ext4_get_block);
1211 : #endif
1212 0 : if (!ret && ext4_should_journal_data(inode)) {
1213 0 : ret = ext4_walk_page_buffers(handle, page_buffers(page),
1214 : from, to, NULL,
1215 : do_journal_get_write_access);
1216 : }
1217 :
1218 0 : if (ret) {
1219 0 : bool extended = (pos + len > inode->i_size) &&
1220 0 : !ext4_verity_in_progress(inode);
1221 :
1222 0 : unlock_page(page);
1223 : /*
1224 : * __block_write_begin may have instantiated a few blocks
1225 : * outside i_size. Trim these off again. Don't need
1226 : * i_size_read because we hold i_mutex.
1227 : *
1228 : * Add inode to orphan list in case we crash before
1229 : * truncate finishes
1230 : */
1231 0 : if (extended && ext4_can_truncate(inode))
1232 0 : ext4_orphan_add(handle, inode);
1233 :
1234 0 : ext4_journal_stop(handle);
1235 0 : if (extended) {
1236 0 : ext4_truncate_failed_write(inode);
1237 : /*
1238 : * If truncate failed early the inode might
1239 : * still be on the orphan list; we need to
1240 : * make sure the inode is removed from the
1241 : * orphan list in that case.
1242 : */
1243 0 : if (inode->i_nlink)
1244 0 : ext4_orphan_del(NULL, inode);
1245 : }
1246 :
1247 0 : if (ret == -ENOSPC &&
1248 0 : ext4_should_retry_alloc(inode->i_sb, &retries))
1249 0 : goto retry_journal;
1250 0 : put_page(page);
1251 0 : return ret;
1252 : }
1253 0 : *pagep = page;
1254 0 : return ret;
1255 : }
1256 :
1257 : /* For write_end() in data=journal mode */
1258 0 : static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1259 : {
1260 0 : int ret;
1261 0 : if (!buffer_mapped(bh) || buffer_freed(bh))
1262 0 : return 0;
1263 0 : set_buffer_uptodate(bh);
1264 0 : ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1265 0 : clear_buffer_meta(bh);
1266 0 : clear_buffer_prio(bh);
1267 0 : return ret;
1268 : }
1269 :
1270 : /*
1271 : * We need to pick up the new inode size which generic_commit_write gave us
1272 : * `file' can be NULL - eg, when called from page_symlink().
1273 : *
1274 : * ext4 never places buffers on inode->i_mapping->private_list. metadata
1275 : * buffers are managed internally.
1276 : */
1277 0 : static int ext4_write_end(struct file *file,
1278 : struct address_space *mapping,
1279 : loff_t pos, unsigned len, unsigned copied,
1280 : struct page *page, void *fsdata)
1281 : {
1282 0 : handle_t *handle = ext4_journal_current_handle();
1283 0 : struct inode *inode = mapping->host;
1284 0 : loff_t old_size = inode->i_size;
1285 0 : int ret = 0, ret2;
1286 0 : int i_size_changed = 0;
1287 0 : int inline_data = ext4_has_inline_data(inode);
1288 0 : bool verity = ext4_verity_in_progress(inode);
1289 :
1290 0 : trace_ext4_write_end(inode, pos, len, copied);
1291 0 : if (inline_data) {
1292 0 : ret = ext4_write_inline_data_end(inode, pos, len,
1293 : copied, page);
1294 0 : if (ret < 0) {
1295 0 : unlock_page(page);
1296 0 : put_page(page);
1297 0 : goto errout;
1298 : }
1299 0 : copied = ret;
1300 : } else
1301 0 : copied = block_write_end(file, mapping, pos,
1302 : len, copied, page, fsdata);
1303 : /*
1304 : * it's important to update i_size while still holding page lock:
1305 : * page writeout could otherwise come in and zero beyond i_size.
1306 : *
1307 : * If FS_IOC_ENABLE_VERITY is running on this inode, then Merkle tree
1308 : * blocks are being written past EOF, so skip the i_size update.
1309 : */
1310 0 : if (!verity)
1311 0 : i_size_changed = ext4_update_inode_size(inode, pos + copied);
1312 0 : unlock_page(page);
1313 0 : put_page(page);
1314 :
1315 0 : if (old_size < pos && !verity)
1316 0 : pagecache_isize_extended(inode, old_size, pos);
1317 : /*
1318 : * Don't mark the inode dirty under page lock. First, it unnecessarily
1319 : * makes the holding time of page lock longer. Second, it forces lock
1320 : * ordering of page lock and transaction start for journaling
1321 : * filesystems.
1322 : */
1323 0 : if (i_size_changed || inline_data)
1324 0 : ret = ext4_mark_inode_dirty(handle, inode);
1325 :
1326 0 : if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode))
1327 : /* if we have allocated more blocks and copied
1328 : * less. We will have blocks allocated outside
1329 : * inode->i_size. So truncate them
1330 : */
1331 0 : ext4_orphan_add(handle, inode);
1332 0 : errout:
1333 0 : ret2 = ext4_journal_stop(handle);
1334 0 : if (!ret)
1335 0 : ret = ret2;
1336 :
1337 0 : if (pos + len > inode->i_size && !verity) {
1338 0 : ext4_truncate_failed_write(inode);
1339 : /*
1340 : * If truncate failed early the inode might still be
1341 : * on the orphan list; we need to make sure the inode
1342 : * is removed from the orphan list in that case.
1343 : */
1344 0 : if (inode->i_nlink)
1345 0 : ext4_orphan_del(NULL, inode);
1346 : }
1347 :
1348 0 : return ret ? ret : copied;
1349 : }
1350 :
1351 : /*
1352 : * This is a private version of page_zero_new_buffers() which doesn't
1353 : * set the buffer to be dirty, since in data=journalled mode we need
1354 : * to call ext4_handle_dirty_metadata() instead.
1355 : */
1356 0 : static void ext4_journalled_zero_new_buffers(handle_t *handle,
1357 : struct page *page,
1358 : unsigned from, unsigned to)
1359 : {
1360 0 : unsigned int block_start = 0, block_end;
1361 0 : struct buffer_head *head, *bh;
1362 :
1363 0 : bh = head = page_buffers(page);
1364 0 : do {
1365 0 : block_end = block_start + bh->b_size;
1366 0 : if (buffer_new(bh)) {
1367 0 : if (block_end > from && block_start < to) {
1368 0 : if (!PageUptodate(page)) {
1369 0 : unsigned start, size;
1370 :
1371 0 : start = max(from, block_start);
1372 0 : size = min(to, block_end) - start;
1373 :
1374 0 : zero_user(page, start, size);
1375 0 : write_end_fn(handle, bh);
1376 : }
1377 0 : clear_buffer_new(bh);
1378 : }
1379 : }
1380 0 : block_start = block_end;
1381 0 : bh = bh->b_this_page;
1382 0 : } while (bh != head);
1383 0 : }
1384 :
1385 0 : static int ext4_journalled_write_end(struct file *file,
1386 : struct address_space *mapping,
1387 : loff_t pos, unsigned len, unsigned copied,
1388 : struct page *page, void *fsdata)
1389 : {
1390 0 : handle_t *handle = ext4_journal_current_handle();
1391 0 : struct inode *inode = mapping->host;
1392 0 : loff_t old_size = inode->i_size;
1393 0 : int ret = 0, ret2;
1394 0 : int partial = 0;
1395 0 : unsigned from, to;
1396 0 : int size_changed = 0;
1397 0 : int inline_data = ext4_has_inline_data(inode);
1398 0 : bool verity = ext4_verity_in_progress(inode);
1399 :
1400 0 : trace_ext4_journalled_write_end(inode, pos, len, copied);
1401 0 : from = pos & (PAGE_SIZE - 1);
1402 0 : to = from + len;
1403 :
1404 0 : BUG_ON(!ext4_handle_valid(handle));
1405 :
1406 0 : if (inline_data) {
1407 0 : ret = ext4_write_inline_data_end(inode, pos, len,
1408 : copied, page);
1409 0 : if (ret < 0) {
1410 0 : unlock_page(page);
1411 0 : put_page(page);
1412 0 : goto errout;
1413 : }
1414 0 : copied = ret;
1415 0 : } else if (unlikely(copied < len) && !PageUptodate(page)) {
1416 0 : copied = 0;
1417 0 : ext4_journalled_zero_new_buffers(handle, page, from, to);
1418 : } else {
1419 0 : if (unlikely(copied < len))
1420 0 : ext4_journalled_zero_new_buffers(handle, page,
1421 : from + copied, to);
1422 0 : ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1423 : from + copied, &partial,
1424 : write_end_fn);
1425 0 : if (!partial)
1426 0 : SetPageUptodate(page);
1427 : }
1428 0 : if (!verity)
1429 0 : size_changed = ext4_update_inode_size(inode, pos + copied);
1430 0 : ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1431 0 : EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1432 0 : unlock_page(page);
1433 0 : put_page(page);
1434 :
1435 0 : if (old_size < pos && !verity)
1436 0 : pagecache_isize_extended(inode, old_size, pos);
1437 :
1438 0 : if (size_changed || inline_data) {
1439 0 : ret2 = ext4_mark_inode_dirty(handle, inode);
1440 0 : if (!ret)
1441 0 : ret = ret2;
1442 : }
1443 :
1444 0 : if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode))
1445 : /* if we have allocated more blocks and copied
1446 : * less. We will have blocks allocated outside
1447 : * inode->i_size. So truncate them
1448 : */
1449 0 : ext4_orphan_add(handle, inode);
1450 :
1451 0 : errout:
1452 0 : ret2 = ext4_journal_stop(handle);
1453 0 : if (!ret)
1454 0 : ret = ret2;
1455 0 : if (pos + len > inode->i_size && !verity) {
1456 0 : ext4_truncate_failed_write(inode);
1457 : /*
1458 : * If truncate failed early the inode might still be
1459 : * on the orphan list; we need to make sure the inode
1460 : * is removed from the orphan list in that case.
1461 : */
1462 0 : if (inode->i_nlink)
1463 0 : ext4_orphan_del(NULL, inode);
1464 : }
1465 :
1466 0 : return ret ? ret : copied;
1467 : }
1468 :
1469 : /*
1470 : * Reserve space for a single cluster
1471 : */
1472 1716 : static int ext4_da_reserve_space(struct inode *inode)
1473 : {
1474 1716 : struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1475 1716 : struct ext4_inode_info *ei = EXT4_I(inode);
1476 1716 : int ret;
1477 :
1478 : /*
1479 : * We will charge metadata quota at writeout time; this saves
1480 : * us from metadata over-estimation, though we may go over by
1481 : * a small amount in the end. Here we just reserve for data.
1482 : */
1483 1716 : ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1484 1716 : if (ret)
1485 : return ret;
1486 :
1487 1716 : spin_lock(&ei->i_block_reservation_lock);
1488 1716 : if (ext4_claim_free_clusters(sbi, 1, 0)) {
1489 0 : spin_unlock(&ei->i_block_reservation_lock);
1490 0 : dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1491 0 : return -ENOSPC;
1492 : }
1493 1716 : ei->i_reserved_data_blocks++;
1494 1716 : trace_ext4_da_reserve_space(inode);
1495 1716 : spin_unlock(&ei->i_block_reservation_lock);
1496 :
1497 1716 : return 0; /* success */
1498 : }
1499 :
1500 333 : void ext4_da_release_space(struct inode *inode, int to_free)
1501 : {
1502 333 : struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1503 333 : struct ext4_inode_info *ei = EXT4_I(inode);
1504 :
1505 333 : if (!to_free)
1506 : return; /* Nothing to release, exit */
1507 :
1508 39 : spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1509 :
1510 39 : trace_ext4_da_release_space(inode, to_free);
1511 39 : if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1512 : /*
1513 : * if there aren't enough reserved blocks, then the
1514 : * counter is messed up somewhere. Since this
1515 : * function is called from invalidate page, it's
1516 : * harmless to return without any action.
1517 : */
1518 0 : ext4_warning(inode->i_sb, "ext4_da_release_space: "
1519 : "ino %lu, to_free %d with only %d reserved "
1520 : "data blocks", inode->i_ino, to_free,
1521 : ei->i_reserved_data_blocks);
1522 0 : WARN_ON(1);
1523 0 : to_free = ei->i_reserved_data_blocks;
1524 : }
1525 39 : ei->i_reserved_data_blocks -= to_free;
1526 :
1527 : /* update fs dirty data blocks counter */
1528 39 : percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1529 :
1530 39 : spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1531 :
1532 39 : dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1533 : }
1534 :
1535 : /*
1536 : * Delayed allocation stuff
1537 : */
1538 :
1539 : struct mpage_da_data {
1540 : struct inode *inode;
1541 : struct writeback_control *wbc;
1542 :
1543 : pgoff_t first_page; /* The first page to write */
1544 : pgoff_t next_page; /* Current page to examine */
1545 : pgoff_t last_page; /* Last page to examine */
1546 : /*
1547 : * Extent to map - this can be after first_page because that can be
1548 : * fully mapped. We somewhat abuse m_flags to store whether the extent
1549 : * is delalloc or unwritten.
1550 : */
1551 : struct ext4_map_blocks map;
1552 : struct ext4_io_submit io_submit; /* IO submission data */
1553 : unsigned int do_map:1;
1554 : unsigned int scanned_until_end:1;
1555 : };
1556 :
1557 154 : static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1558 : bool invalidate)
1559 : {
1560 154 : int nr_pages, i;
1561 154 : pgoff_t index, end;
1562 154 : struct pagevec pvec;
1563 154 : struct inode *inode = mpd->inode;
1564 154 : struct address_space *mapping = inode->i_mapping;
1565 :
1566 : /* This is necessary when next_page == 0. */
1567 154 : if (mpd->first_page >= mpd->next_page)
1568 84 : return;
1569 :
1570 70 : mpd->scanned_until_end = 0;
1571 70 : index = mpd->first_page;
1572 70 : end = mpd->next_page - 1;
1573 70 : if (invalidate) {
1574 0 : ext4_lblk_t start, last;
1575 0 : start = index << (PAGE_SHIFT - inode->i_blkbits);
1576 0 : last = end << (PAGE_SHIFT - inode->i_blkbits);
1577 0 : ext4_es_remove_extent(inode, start, last - start + 1);
1578 : }
1579 :
1580 70 : pagevec_init(&pvec);
1581 175 : while (index <= end) {
1582 105 : nr_pages = pagevec_lookup_range(&pvec, mapping, &index, end);
1583 105 : if (nr_pages == 0)
1584 : break;
1585 740 : for (i = 0; i < nr_pages; i++) {
1586 635 : struct page *page = pvec.pages[i];
1587 :
1588 1270 : BUG_ON(!PageLocked(page));
1589 1270 : BUG_ON(PageWriteback(page));
1590 635 : if (invalidate) {
1591 0 : if (page_mapped(page))
1592 0 : clear_page_dirty_for_io(page);
1593 0 : block_invalidatepage(page, 0, PAGE_SIZE);
1594 0 : ClearPageUptodate(page);
1595 : }
1596 635 : unlock_page(page);
1597 : }
1598 280 : pagevec_release(&pvec);
1599 : }
1600 : }
1601 :
1602 0 : static void ext4_print_free_blocks(struct inode *inode)
1603 : {
1604 0 : struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1605 0 : struct super_block *sb = inode->i_sb;
1606 0 : struct ext4_inode_info *ei = EXT4_I(inode);
1607 :
1608 0 : ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1609 : EXT4_C2B(EXT4_SB(inode->i_sb),
1610 : ext4_count_free_clusters(sb)));
1611 0 : ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1612 0 : ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1613 : (long long) EXT4_C2B(EXT4_SB(sb),
1614 : percpu_counter_sum(&sbi->s_freeclusters_counter)));
1615 0 : ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1616 : (long long) EXT4_C2B(EXT4_SB(sb),
1617 : percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1618 0 : ext4_msg(sb, KERN_CRIT, "Block reservation details");
1619 0 : ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1620 : ei->i_reserved_data_blocks);
1621 0 : return;
1622 : }
1623 :
1624 2 : static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1625 : {
1626 4 : return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
1627 : }
1628 :
1629 : /*
1630 : * ext4_insert_delayed_block - adds a delayed block to the extents status
1631 : * tree, incrementing the reserved cluster/block
1632 : * count or making a pending reservation
1633 : * where needed
1634 : *
1635 : * @inode - file containing the newly added block
1636 : * @lblk - logical block to be added
1637 : *
1638 : * Returns 0 on success, negative error code on failure.
1639 : */
1640 1716 : static int ext4_insert_delayed_block(struct inode *inode, ext4_lblk_t lblk)
1641 : {
1642 1716 : struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1643 1716 : int ret;
1644 1716 : bool allocated = false;
1645 :
1646 : /*
1647 : * If the cluster containing lblk is shared with a delayed,
1648 : * written, or unwritten extent in a bigalloc file system, it's
1649 : * already been accounted for and does not need to be reserved.
1650 : * A pending reservation must be made for the cluster if it's
1651 : * shared with a written or unwritten extent and doesn't already
1652 : * have one. Written and unwritten extents can be purged from the
1653 : * extents status tree if the system is under memory pressure, so
1654 : * it's necessary to examine the extent tree if a search of the
1655 : * extents status tree doesn't get a match.
1656 : */
1657 1716 : if (sbi->s_cluster_ratio == 1) {
1658 1716 : ret = ext4_da_reserve_space(inode);
1659 1716 : if (ret != 0) /* ENOSPC */
1660 0 : goto errout;
1661 : } else { /* bigalloc */
1662 0 : if (!ext4_es_scan_clu(inode, &ext4_es_is_delonly, lblk)) {
1663 0 : if (!ext4_es_scan_clu(inode,
1664 : &ext4_es_is_mapped, lblk)) {
1665 0 : ret = ext4_clu_mapped(inode,
1666 0 : EXT4_B2C(sbi, lblk));
1667 0 : if (ret < 0)
1668 0 : goto errout;
1669 0 : if (ret == 0) {
1670 0 : ret = ext4_da_reserve_space(inode);
1671 0 : if (ret != 0) /* ENOSPC */
1672 0 : goto errout;
1673 : } else {
1674 : allocated = true;
1675 : }
1676 : } else {
1677 : allocated = true;
1678 : }
1679 : }
1680 : }
1681 :
1682 1716 : ret = ext4_es_insert_delayed_block(inode, lblk, allocated);
1683 :
1684 1716 : errout:
1685 1716 : return ret;
1686 : }
1687 :
1688 : /*
1689 : * This function is grabs code from the very beginning of
1690 : * ext4_map_blocks, but assumes that the caller is from delayed write
1691 : * time. This function looks up the requested blocks and sets the
1692 : * buffer delay bit under the protection of i_data_sem.
1693 : */
1694 1821 : static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1695 : struct ext4_map_blocks *map,
1696 : struct buffer_head *bh)
1697 : {
1698 1821 : struct extent_status es;
1699 1821 : int retval;
1700 1821 : sector_t invalid_block = ~((sector_t) 0xffff);
1701 : #ifdef ES_AGGRESSIVE_TEST
1702 : struct ext4_map_blocks orig_map;
1703 :
1704 : memcpy(&orig_map, map, sizeof(*map));
1705 : #endif
1706 :
1707 1821 : if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1708 0 : invalid_block = ~0;
1709 :
1710 1821 : map->m_flags = 0;
1711 1821 : ext_debug(inode, "max_blocks %u, logical block %lu\n", map->m_len,
1712 : (unsigned long) map->m_lblk);
1713 :
1714 : /* Lookup extent status tree firstly */
1715 1821 : if (ext4_es_lookup_extent(inode, iblock, NULL, &es)) {
1716 1354 : if (ext4_es_is_hole(&es)) {
1717 1258 : retval = 0;
1718 1258 : down_read(&EXT4_I(inode)->i_data_sem);
1719 1258 : goto add_delayed;
1720 : }
1721 :
1722 : /*
1723 : * Delayed extent could be allocated by fallocate.
1724 : * So we need to check it.
1725 : */
1726 96 : if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1727 0 : map_bh(bh, inode->i_sb, invalid_block);
1728 0 : set_buffer_new(bh);
1729 0 : set_buffer_delay(bh);
1730 0 : return 0;
1731 : }
1732 :
1733 96 : map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1734 96 : retval = es.es_len - (iblock - es.es_lblk);
1735 96 : if (retval > map->m_len)
1736 : retval = map->m_len;
1737 96 : map->m_len = retval;
1738 96 : if (ext4_es_is_written(&es))
1739 96 : map->m_flags |= EXT4_MAP_MAPPED;
1740 0 : else if (ext4_es_is_unwritten(&es))
1741 0 : map->m_flags |= EXT4_MAP_UNWRITTEN;
1742 : else
1743 0 : BUG();
1744 :
1745 : #ifdef ES_AGGRESSIVE_TEST
1746 : ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1747 : #endif
1748 96 : return retval;
1749 : }
1750 :
1751 : /*
1752 : * Try to see if we can get the block without requesting a new
1753 : * file system block.
1754 : */
1755 467 : down_read(&EXT4_I(inode)->i_data_sem);
1756 467 : if (ext4_has_inline_data(inode))
1757 : retval = 0;
1758 467 : else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1759 467 : retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1760 : else
1761 0 : retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1762 :
1763 467 : add_delayed:
1764 1725 : if (retval == 0) {
1765 1716 : int ret;
1766 :
1767 : /*
1768 : * XXX: __block_prepare_write() unmaps passed block,
1769 : * is it OK?
1770 : */
1771 :
1772 1716 : ret = ext4_insert_delayed_block(inode, map->m_lblk);
1773 1716 : if (ret != 0) {
1774 0 : retval = ret;
1775 0 : goto out_unlock;
1776 : }
1777 :
1778 1716 : map_bh(bh, inode->i_sb, invalid_block);
1779 1716 : set_buffer_new(bh);
1780 1716 : set_buffer_delay(bh);
1781 9 : } else if (retval > 0) {
1782 9 : int ret;
1783 9 : unsigned int status;
1784 :
1785 9 : if (unlikely(retval != map->m_len)) {
1786 0 : ext4_warning(inode->i_sb,
1787 : "ES len assertion failed for inode "
1788 : "%lu: retval %d != map->m_len %d",
1789 : inode->i_ino, retval, map->m_len);
1790 0 : WARN_ON(1);
1791 : }
1792 :
1793 18 : status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1794 9 : EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1795 9 : ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1796 : map->m_pblk, status);
1797 9 : if (ret != 0)
1798 0 : retval = ret;
1799 : }
1800 :
1801 9 : out_unlock:
1802 1725 : up_read((&EXT4_I(inode)->i_data_sem));
1803 :
1804 1725 : return retval;
1805 : }
1806 :
1807 : /*
1808 : * This is a special get_block_t callback which is used by
1809 : * ext4_da_write_begin(). It will either return mapped block or
1810 : * reserve space for a single block.
1811 : *
1812 : * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1813 : * We also have b_blocknr = -1 and b_bdev initialized properly
1814 : *
1815 : * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1816 : * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1817 : * initialized properly.
1818 : */
1819 1821 : int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1820 : struct buffer_head *bh, int create)
1821 : {
1822 1821 : struct ext4_map_blocks map;
1823 1821 : int ret = 0;
1824 :
1825 1821 : BUG_ON(create == 0);
1826 1821 : BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1827 :
1828 1821 : map.m_lblk = iblock;
1829 1821 : map.m_len = 1;
1830 :
1831 : /*
1832 : * first, we need to know whether the block is allocated already
1833 : * preallocated blocks are unmapped but should treated
1834 : * the same as allocated blocks.
1835 : */
1836 1821 : ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1837 1821 : if (ret <= 0)
1838 : return ret;
1839 :
1840 105 : map_bh(bh, inode->i_sb, map.m_pblk);
1841 105 : ext4_update_bh_state(bh, map.m_flags);
1842 :
1843 105 : if (buffer_unwritten(bh)) {
1844 : /* A delayed write to unwritten bh should be marked
1845 : * new and mapped. Mapped ensures that we don't do
1846 : * get_block multiple times when we write to the same
1847 : * offset and new ensures that we do proper zero out
1848 : * for partial write.
1849 : */
1850 0 : set_buffer_new(bh);
1851 0 : set_buffer_mapped(bh);
1852 : }
1853 : return 0;
1854 : }
1855 :
1856 0 : static int bget_one(handle_t *handle, struct buffer_head *bh)
1857 : {
1858 0 : get_bh(bh);
1859 0 : return 0;
1860 : }
1861 :
1862 0 : static int bput_one(handle_t *handle, struct buffer_head *bh)
1863 : {
1864 0 : put_bh(bh);
1865 0 : return 0;
1866 : }
1867 :
1868 0 : static int __ext4_journalled_writepage(struct page *page,
1869 : unsigned int len)
1870 : {
1871 0 : struct address_space *mapping = page->mapping;
1872 0 : struct inode *inode = mapping->host;
1873 0 : struct buffer_head *page_bufs = NULL;
1874 0 : handle_t *handle = NULL;
1875 0 : int ret = 0, err = 0;
1876 0 : int inline_data = ext4_has_inline_data(inode);
1877 0 : struct buffer_head *inode_bh = NULL;
1878 :
1879 0 : ClearPageChecked(page);
1880 :
1881 0 : if (inline_data) {
1882 0 : BUG_ON(page->index != 0);
1883 0 : BUG_ON(len > ext4_get_max_inline_size(inode));
1884 0 : inode_bh = ext4_journalled_write_inline_data(inode, len, page);
1885 0 : if (inode_bh == NULL)
1886 0 : goto out;
1887 : } else {
1888 0 : page_bufs = page_buffers(page);
1889 0 : if (!page_bufs) {
1890 0 : BUG();
1891 : goto out;
1892 : }
1893 0 : ext4_walk_page_buffers(handle, page_bufs, 0, len,
1894 : NULL, bget_one);
1895 : }
1896 : /*
1897 : * We need to release the page lock before we start the
1898 : * journal, so grab a reference so the page won't disappear
1899 : * out from under us.
1900 : */
1901 0 : get_page(page);
1902 0 : unlock_page(page);
1903 :
1904 0 : handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
1905 : ext4_writepage_trans_blocks(inode));
1906 0 : if (IS_ERR(handle)) {
1907 0 : ret = PTR_ERR(handle);
1908 0 : put_page(page);
1909 0 : goto out_no_pagelock;
1910 : }
1911 0 : BUG_ON(!ext4_handle_valid(handle));
1912 :
1913 0 : lock_page(page);
1914 0 : put_page(page);
1915 0 : if (page->mapping != mapping) {
1916 : /* The page got truncated from under us */
1917 0 : ext4_journal_stop(handle);
1918 0 : ret = 0;
1919 0 : goto out;
1920 : }
1921 :
1922 0 : if (inline_data) {
1923 0 : ret = ext4_mark_inode_dirty(handle, inode);
1924 : } else {
1925 0 : ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1926 : do_journal_get_write_access);
1927 :
1928 0 : err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1929 : write_end_fn);
1930 : }
1931 0 : if (ret == 0)
1932 0 : ret = err;
1933 0 : err = ext4_jbd2_inode_add_write(handle, inode, page_offset(page), len);
1934 0 : if (ret == 0)
1935 0 : ret = err;
1936 0 : EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1937 0 : err = ext4_journal_stop(handle);
1938 0 : if (!ret)
1939 0 : ret = err;
1940 :
1941 0 : if (!ext4_has_inline_data(inode))
1942 0 : ext4_walk_page_buffers(NULL, page_bufs, 0, len,
1943 : NULL, bput_one);
1944 0 : ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1945 0 : out:
1946 0 : unlock_page(page);
1947 0 : out_no_pagelock:
1948 0 : brelse(inode_bh);
1949 0 : return ret;
1950 : }
1951 :
1952 : /*
1953 : * Note that we don't need to start a transaction unless we're journaling data
1954 : * because we should have holes filled from ext4_page_mkwrite(). We even don't
1955 : * need to file the inode to the transaction's list in ordered mode because if
1956 : * we are writing back data added by write(), the inode is already there and if
1957 : * we are writing back data modified via mmap(), no one guarantees in which
1958 : * transaction the data will hit the disk. In case we are journaling data, we
1959 : * cannot start transaction directly because transaction start ranks above page
1960 : * lock so we have to do some magic.
1961 : *
1962 : * This function can get called via...
1963 : * - ext4_writepages after taking page lock (have journal handle)
1964 : * - journal_submit_inode_data_buffers (no journal handle)
1965 : * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
1966 : * - grab_page_cache when doing write_begin (have journal handle)
1967 : *
1968 : * We don't do any block allocation in this function. If we have page with
1969 : * multiple blocks we need to write those buffer_heads that are mapped. This
1970 : * is important for mmaped based write. So if we do with blocksize 1K
1971 : * truncate(f, 1024);
1972 : * a = mmap(f, 0, 4096);
1973 : * a[0] = 'a';
1974 : * truncate(f, 4096);
1975 : * we have in the page first buffer_head mapped via page_mkwrite call back
1976 : * but other buffer_heads would be unmapped but dirty (dirty done via the
1977 : * do_wp_page). So writepage should write the first block. If we modify
1978 : * the mmap area beyond 1024 we will again get a page_fault and the
1979 : * page_mkwrite callback will do the block allocation and mark the
1980 : * buffer_heads mapped.
1981 : *
1982 : * We redirty the page if we have any buffer_heads that is either delay or
1983 : * unwritten in the page.
1984 : *
1985 : * We can get recursively called as show below.
1986 : *
1987 : * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1988 : * ext4_writepage()
1989 : *
1990 : * But since we don't do any block allocation we should not deadlock.
1991 : * Page also have the dirty flag cleared so we don't get recurive page_lock.
1992 : */
1993 2 : static int ext4_writepage(struct page *page,
1994 : struct writeback_control *wbc)
1995 : {
1996 2 : int ret = 0;
1997 2 : loff_t size;
1998 2 : unsigned int len;
1999 2 : struct buffer_head *page_bufs = NULL;
2000 2 : struct inode *inode = page->mapping->host;
2001 2 : struct ext4_io_submit io_submit;
2002 2 : bool keep_towrite = false;
2003 :
2004 2 : if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) {
2005 0 : inode->i_mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE);
2006 0 : unlock_page(page);
2007 0 : return -EIO;
2008 : }
2009 :
2010 2 : trace_ext4_writepage(page);
2011 2 : size = i_size_read(inode);
2012 2 : if (page->index == size >> PAGE_SHIFT &&
2013 2 : !ext4_verity_in_progress(inode))
2014 2 : len = size & ~PAGE_MASK;
2015 : else
2016 : len = PAGE_SIZE;
2017 :
2018 2 : page_bufs = page_buffers(page);
2019 : /*
2020 : * We cannot do block allocation or other extent handling in this
2021 : * function. If there are buffers needing that, we have to redirty
2022 : * the page. But we may reach here when we do a journal commit via
2023 : * journal_submit_inode_data_buffers() and in that case we must write
2024 : * allocated buffers to achieve data=ordered mode guarantees.
2025 : *
2026 : * Also, if there is only one buffer per page (the fs block
2027 : * size == the page size), if one buffer needs block
2028 : * allocation or needs to modify the extent tree to clear the
2029 : * unwritten flag, we know that the page can't be written at
2030 : * all, so we might as well refuse the write immediately.
2031 : * Unfortunately if the block size != page size, we can't as
2032 : * easily detect this case using ext4_walk_page_buffers(), but
2033 : * for the extremely common case, this is an optimization that
2034 : * skips a useless round trip through ext4_bio_write_page().
2035 : */
2036 2 : if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2037 : ext4_bh_delay_or_unwritten)) {
2038 2 : redirty_page_for_writepage(wbc, page);
2039 2 : if ((current->flags & PF_MEMALLOC) ||
2040 2 : (inode->i_sb->s_blocksize == PAGE_SIZE)) {
2041 : /*
2042 : * For memory cleaning there's no point in writing only
2043 : * some buffers. So just bail out. Warn if we came here
2044 : * from direct reclaim.
2045 : */
2046 2 : WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
2047 : == PF_MEMALLOC);
2048 2 : unlock_page(page);
2049 2 : return 0;
2050 : }
2051 : keep_towrite = true;
2052 : }
2053 :
2054 0 : if (PageChecked(page) && ext4_should_journal_data(inode))
2055 : /*
2056 : * It's mmapped pagecache. Add buffers and journal it. There
2057 : * doesn't seem much point in redirtying the page here.
2058 : */
2059 0 : return __ext4_journalled_writepage(page, len);
2060 :
2061 0 : ext4_io_submit_init(&io_submit, wbc);
2062 0 : io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
2063 0 : if (!io_submit.io_end) {
2064 0 : redirty_page_for_writepage(wbc, page);
2065 0 : unlock_page(page);
2066 0 : return -ENOMEM;
2067 : }
2068 0 : ret = ext4_bio_write_page(&io_submit, page, len, keep_towrite);
2069 0 : ext4_io_submit(&io_submit);
2070 : /* Drop io_end reference we got from init */
2071 0 : ext4_put_io_end_defer(io_submit.io_end);
2072 0 : return ret;
2073 : }
2074 :
2075 514 : static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page)
2076 : {
2077 514 : int len;
2078 514 : loff_t size;
2079 514 : int err;
2080 :
2081 514 : BUG_ON(page->index != mpd->first_page);
2082 514 : clear_page_dirty_for_io(page);
2083 : /*
2084 : * We have to be very careful here! Nothing protects writeback path
2085 : * against i_size changes and the page can be writeably mapped into
2086 : * page tables. So an application can be growing i_size and writing
2087 : * data through mmap while writeback runs. clear_page_dirty_for_io()
2088 : * write-protects our page in page tables and the page cannot get
2089 : * written to again until we release page lock. So only after
2090 : * clear_page_dirty_for_io() we are safe to sample i_size for
2091 : * ext4_bio_write_page() to zero-out tail of the written page. We rely
2092 : * on the barrier provided by TestClearPageDirty in
2093 : * clear_page_dirty_for_io() to make sure i_size is really sampled only
2094 : * after page tables are updated.
2095 : */
2096 514 : size = i_size_read(mpd->inode);
2097 514 : if (page->index == size >> PAGE_SHIFT &&
2098 29 : !ext4_verity_in_progress(mpd->inode))
2099 29 : len = size & ~PAGE_MASK;
2100 : else
2101 : len = PAGE_SIZE;
2102 514 : err = ext4_bio_write_page(&mpd->io_submit, page, len, false);
2103 514 : if (!err)
2104 514 : mpd->wbc->nr_to_write--;
2105 514 : mpd->first_page++;
2106 :
2107 514 : return err;
2108 : }
2109 :
2110 : #define BH_FLAGS (BIT(BH_Unwritten) | BIT(BH_Delay))
2111 :
2112 : /*
2113 : * mballoc gives us at most this number of blocks...
2114 : * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
2115 : * The rest of mballoc seems to handle chunks up to full group size.
2116 : */
2117 : #define MAX_WRITEPAGES_EXTENT_LEN 2048
2118 :
2119 : /*
2120 : * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
2121 : *
2122 : * @mpd - extent of blocks
2123 : * @lblk - logical number of the block in the file
2124 : * @bh - buffer head we want to add to the extent
2125 : *
2126 : * The function is used to collect contig. blocks in the same state. If the
2127 : * buffer doesn't require mapping for writeback and we haven't started the
2128 : * extent of buffers to map yet, the function returns 'true' immediately - the
2129 : * caller can write the buffer right away. Otherwise the function returns true
2130 : * if the block has been added to the extent, false if the block couldn't be
2131 : * added.
2132 : */
2133 1149 : static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
2134 : struct buffer_head *bh)
2135 : {
2136 1149 : struct ext4_map_blocks *map = &mpd->map;
2137 :
2138 : /* Buffer that doesn't need mapping for writeback? */
2139 2298 : if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
2140 1315 : (!buffer_delay(bh) && !buffer_unwritten(bh))) {
2141 : /* So far no extent to map => we write the buffer right away */
2142 166 : if (map->m_len == 0)
2143 : return true;
2144 0 : return false;
2145 : }
2146 :
2147 : /* First block in the extent? */
2148 983 : if (map->m_len == 0) {
2149 : /* We cannot map unless handle is started... */
2150 133 : if (!mpd->do_map)
2151 : return false;
2152 70 : map->m_lblk = lblk;
2153 70 : map->m_len = 1;
2154 70 : map->m_flags = bh->b_state & BH_FLAGS;
2155 70 : return true;
2156 : }
2157 :
2158 : /* Don't go larger than mballoc is willing to allocate */
2159 850 : if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
2160 : return false;
2161 :
2162 : /* Can we merge the block to our big extent? */
2163 850 : if (lblk == map->m_lblk + map->m_len &&
2164 850 : (bh->b_state & BH_FLAGS) == map->m_flags) {
2165 850 : map->m_len++;
2166 850 : return true;
2167 : }
2168 : return false;
2169 : }
2170 :
2171 : /*
2172 : * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2173 : *
2174 : * @mpd - extent of blocks for mapping
2175 : * @head - the first buffer in the page
2176 : * @bh - buffer we should start processing from
2177 : * @lblk - logical number of the block in the file corresponding to @bh
2178 : *
2179 : * Walk through page buffers from @bh upto @head (exclusive) and either submit
2180 : * the page for IO if all buffers in this page were mapped and there's no
2181 : * accumulated extent of buffers to map or add buffers in the page to the
2182 : * extent of buffers to map. The function returns 1 if the caller can continue
2183 : * by processing the next page, 0 if it should stop adding buffers to the
2184 : * extent to map because we cannot extend it anymore. It can also return value
2185 : * < 0 in case of error during IO submission.
2186 : */
2187 1149 : static int mpage_process_page_bufs(struct mpage_da_data *mpd,
2188 : struct buffer_head *head,
2189 : struct buffer_head *bh,
2190 : ext4_lblk_t lblk)
2191 : {
2192 1149 : struct inode *inode = mpd->inode;
2193 1149 : int err;
2194 1149 : ext4_lblk_t blocks = (i_size_read(inode) + i_blocksize(inode) - 1)
2195 1149 : >> inode->i_blkbits;
2196 :
2197 1149 : if (ext4_verity_in_progress(inode))
2198 : blocks = EXT_MAX_BLOCKS;
2199 :
2200 1149 : do {
2201 1149 : BUG_ON(buffer_locked(bh));
2202 :
2203 1149 : if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
2204 : /* Found extent to map? */
2205 63 : if (mpd->map.m_len)
2206 : return 0;
2207 : /* Buffer needs mapping and handle is not started? */
2208 63 : if (!mpd->do_map)
2209 : return 0;
2210 : /* Everything mapped so far and we hit EOF */
2211 : break;
2212 : }
2213 1086 : } while (lblk++, (bh = bh->b_this_page) != head);
2214 : /* So far everything mapped? Submit the page for IO. */
2215 1086 : if (mpd->map.m_len == 0) {
2216 166 : err = mpage_submit_page(mpd, head->b_page);
2217 166 : if (err < 0)
2218 : return err;
2219 : }
2220 1086 : if (lblk >= blocks) {
2221 91 : mpd->scanned_until_end = 1;
2222 91 : return 0;
2223 : }
2224 : return 1;
2225 : }
2226 :
2227 : /*
2228 : * mpage_process_page - update page buffers corresponding to changed extent and
2229 : * may submit fully mapped page for IO
2230 : *
2231 : * @mpd - description of extent to map, on return next extent to map
2232 : * @m_lblk - logical block mapping.
2233 : * @m_pblk - corresponding physical mapping.
2234 : * @map_bh - determines on return whether this page requires any further
2235 : * mapping or not.
2236 : * Scan given page buffers corresponding to changed extent and update buffer
2237 : * state according to new extent state.
2238 : * We map delalloc buffers to their physical location, clear unwritten bits.
2239 : * If the given page is not fully mapped, we update @map to the next extent in
2240 : * the given page that needs mapping & return @map_bh as true.
2241 : */
2242 348 : static int mpage_process_page(struct mpage_da_data *mpd, struct page *page,
2243 : ext4_lblk_t *m_lblk, ext4_fsblk_t *m_pblk,
2244 : bool *map_bh)
2245 : {
2246 348 : struct buffer_head *head, *bh;
2247 348 : ext4_io_end_t *io_end = mpd->io_submit.io_end;
2248 348 : ext4_lblk_t lblk = *m_lblk;
2249 348 : ext4_fsblk_t pblock = *m_pblk;
2250 348 : int err = 0;
2251 348 : int blkbits = mpd->inode->i_blkbits;
2252 348 : ssize_t io_end_size = 0;
2253 348 : struct ext4_io_end_vec *io_end_vec = ext4_last_io_end_vec(io_end);
2254 :
2255 348 : bh = head = page_buffers(page);
2256 348 : do {
2257 348 : if (lblk < mpd->map.m_lblk)
2258 0 : continue;
2259 348 : if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2260 : /*
2261 : * Buffer after end of mapped extent.
2262 : * Find next buffer in the page to map.
2263 : */
2264 0 : mpd->map.m_len = 0;
2265 0 : mpd->map.m_flags = 0;
2266 0 : io_end_vec->size += io_end_size;
2267 0 : io_end_size = 0;
2268 :
2269 0 : err = mpage_process_page_bufs(mpd, head, bh, lblk);
2270 0 : if (err > 0)
2271 : err = 0;
2272 0 : if (!err && mpd->map.m_len && mpd->map.m_lblk > lblk) {
2273 0 : io_end_vec = ext4_alloc_io_end_vec(io_end);
2274 0 : if (IS_ERR(io_end_vec)) {
2275 0 : err = PTR_ERR(io_end_vec);
2276 0 : goto out;
2277 : }
2278 0 : io_end_vec->offset = (loff_t)mpd->map.m_lblk << blkbits;
2279 : }
2280 0 : *map_bh = true;
2281 0 : goto out;
2282 : }
2283 348 : if (buffer_delay(bh)) {
2284 348 : clear_buffer_delay(bh);
2285 348 : bh->b_blocknr = pblock++;
2286 : }
2287 348 : clear_buffer_unwritten(bh);
2288 348 : io_end_size += (1 << blkbits);
2289 348 : } while (lblk++, (bh = bh->b_this_page) != head);
2290 :
2291 348 : io_end_vec->size += io_end_size;
2292 348 : io_end_size = 0;
2293 348 : *map_bh = false;
2294 348 : out:
2295 348 : *m_lblk = lblk;
2296 348 : *m_pblk = pblock;
2297 348 : return err;
2298 : }
2299 :
2300 : /*
2301 : * mpage_map_buffers - update buffers corresponding to changed extent and
2302 : * submit fully mapped pages for IO
2303 : *
2304 : * @mpd - description of extent to map, on return next extent to map
2305 : *
2306 : * Scan buffers corresponding to changed extent (we expect corresponding pages
2307 : * to be already locked) and update buffer state according to new extent state.
2308 : * We map delalloc buffers to their physical location, clear unwritten bits,
2309 : * and mark buffers as uninit when we perform writes to unwritten extents
2310 : * and do extent conversion after IO is finished. If the last page is not fully
2311 : * mapped, we update @map to the next extent in the last page that needs
2312 : * mapping. Otherwise we submit the page for IO.
2313 : */
2314 70 : static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
2315 : {
2316 70 : struct pagevec pvec;
2317 70 : int nr_pages, i;
2318 70 : struct inode *inode = mpd->inode;
2319 70 : int bpp_bits = PAGE_SHIFT - inode->i_blkbits;
2320 70 : pgoff_t start, end;
2321 70 : ext4_lblk_t lblk;
2322 70 : ext4_fsblk_t pblock;
2323 70 : int err;
2324 70 : bool map_bh = false;
2325 :
2326 70 : start = mpd->map.m_lblk >> bpp_bits;
2327 70 : end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2328 70 : lblk = start << bpp_bits;
2329 70 : pblock = mpd->map.m_pblk;
2330 :
2331 70 : pagevec_init(&pvec);
2332 146 : while (start <= end) {
2333 76 : nr_pages = pagevec_lookup_range(&pvec, inode->i_mapping,
2334 : &start, end);
2335 76 : if (nr_pages == 0)
2336 : break;
2337 424 : for (i = 0; i < nr_pages; i++) {
2338 348 : struct page *page = pvec.pages[i];
2339 :
2340 348 : err = mpage_process_page(mpd, page, &lblk, &pblock,
2341 : &map_bh);
2342 : /*
2343 : * If map_bh is true, means page may require further bh
2344 : * mapping, or maybe the page was submitted for IO.
2345 : * So we return to call further extent mapping.
2346 : */
2347 348 : if (err < 0 || map_bh)
2348 0 : goto out;
2349 : /* Page fully mapped - let IO run! */
2350 348 : err = mpage_submit_page(mpd, page);
2351 348 : if (err < 0)
2352 0 : goto out;
2353 : }
2354 222 : pagevec_release(&pvec);
2355 : }
2356 : /* Extent fully mapped and matches with page boundary. We are done. */
2357 70 : mpd->map.m_len = 0;
2358 70 : mpd->map.m_flags = 0;
2359 70 : return 0;
2360 0 : out:
2361 0 : pagevec_release(&pvec);
2362 : return err;
2363 : }
2364 :
2365 70 : static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2366 : {
2367 70 : struct inode *inode = mpd->inode;
2368 70 : struct ext4_map_blocks *map = &mpd->map;
2369 70 : int get_blocks_flags;
2370 70 : int err, dioread_nolock;
2371 :
2372 70 : trace_ext4_da_write_pages_extent(inode, map);
2373 : /*
2374 : * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2375 : * to convert an unwritten extent to be initialized (in the case
2376 : * where we have written into one or more preallocated blocks). It is
2377 : * possible that we're going to need more metadata blocks than
2378 : * previously reserved. However we must not fail because we're in
2379 : * writeback and there is nothing we can do about it so it might result
2380 : * in data loss. So use reserved blocks to allocate metadata if
2381 : * possible.
2382 : *
2383 : * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2384 : * the blocks in question are delalloc blocks. This indicates
2385 : * that the blocks and quotas has already been checked when
2386 : * the data was copied into the page cache.
2387 : */
2388 70 : get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2389 : EXT4_GET_BLOCKS_METADATA_NOFAIL |
2390 : EXT4_GET_BLOCKS_IO_SUBMIT;
2391 70 : dioread_nolock = ext4_should_dioread_nolock(inode);
2392 70 : if (dioread_nolock)
2393 70 : get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2394 70 : if (map->m_flags & BIT(BH_Delay))
2395 70 : get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2396 :
2397 70 : err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2398 70 : if (err < 0)
2399 : return err;
2400 70 : if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2401 70 : if (!mpd->io_submit.io_end->handle &&
2402 140 : ext4_handle_valid(handle)) {
2403 70 : mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2404 70 : handle->h_rsv_handle = NULL;
2405 : }
2406 70 : ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2407 : }
2408 :
2409 70 : BUG_ON(map->m_len == 0);
2410 : return 0;
2411 : }
2412 :
2413 : /*
2414 : * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2415 : * mpd->len and submit pages underlying it for IO
2416 : *
2417 : * @handle - handle for journal operations
2418 : * @mpd - extent to map
2419 : * @give_up_on_write - we set this to true iff there is a fatal error and there
2420 : * is no hope of writing the data. The caller should discard
2421 : * dirty pages to avoid infinite loops.
2422 : *
2423 : * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2424 : * delayed, blocks are allocated, if it is unwritten, we may need to convert
2425 : * them to initialized or split the described range from larger unwritten
2426 : * extent. Note that we need not map all the described range since allocation
2427 : * can return less blocks or the range is covered by more unwritten extents. We
2428 : * cannot map more because we are limited by reserved transaction credits. On
2429 : * the other hand we always make sure that the last touched page is fully
2430 : * mapped so that it can be written out (and thus forward progress is
2431 : * guaranteed). After mapping we submit all mapped pages for IO.
2432 : */
2433 70 : static int mpage_map_and_submit_extent(handle_t *handle,
2434 : struct mpage_da_data *mpd,
2435 : bool *give_up_on_write)
2436 : {
2437 70 : struct inode *inode = mpd->inode;
2438 70 : struct ext4_map_blocks *map = &mpd->map;
2439 70 : int err;
2440 70 : loff_t disksize;
2441 70 : int progress = 0;
2442 70 : ext4_io_end_t *io_end = mpd->io_submit.io_end;
2443 70 : struct ext4_io_end_vec *io_end_vec;
2444 :
2445 70 : io_end_vec = ext4_alloc_io_end_vec(io_end);
2446 70 : if (IS_ERR(io_end_vec))
2447 0 : return PTR_ERR(io_end_vec);
2448 70 : io_end_vec->offset = ((loff_t)map->m_lblk) << inode->i_blkbits;
2449 70 : do {
2450 70 : err = mpage_map_one_extent(handle, mpd);
2451 70 : if (err < 0) {
2452 0 : struct super_block *sb = inode->i_sb;
2453 :
2454 0 : if (ext4_forced_shutdown(EXT4_SB(sb)) ||
2455 0 : ext4_test_mount_flag(sb, EXT4_MF_FS_ABORTED))
2456 0 : goto invalidate_dirty_pages;
2457 : /*
2458 : * Let the uper layers retry transient errors.
2459 : * In the case of ENOSPC, if ext4_count_free_blocks()
2460 : * is non-zero, a commit should free up blocks.
2461 : */
2462 0 : if ((err == -ENOMEM) ||
2463 0 : (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2464 0 : if (progress)
2465 0 : goto update_disksize;
2466 : return err;
2467 : }
2468 0 : ext4_msg(sb, KERN_CRIT,
2469 : "Delayed block allocation failed for "
2470 : "inode %lu at logical offset %llu with"
2471 : " max blocks %u with error %d",
2472 : inode->i_ino,
2473 : (unsigned long long)map->m_lblk,
2474 : (unsigned)map->m_len, -err);
2475 0 : ext4_msg(sb, KERN_CRIT,
2476 : "This should not happen!! Data will "
2477 : "be lost\n");
2478 0 : if (err == -ENOSPC)
2479 0 : ext4_print_free_blocks(inode);
2480 0 : invalidate_dirty_pages:
2481 0 : *give_up_on_write = true;
2482 0 : return err;
2483 : }
2484 70 : progress = 1;
2485 : /*
2486 : * Update buffer state, submit mapped pages, and get us new
2487 : * extent to map
2488 : */
2489 70 : err = mpage_map_and_submit_buffers(mpd);
2490 70 : if (err < 0)
2491 0 : goto update_disksize;
2492 70 : } while (map->m_len);
2493 :
2494 70 : update_disksize:
2495 : /*
2496 : * Update on-disk size after IO is submitted. Races with
2497 : * truncate are avoided by checking i_size under i_data_sem.
2498 : */
2499 70 : disksize = ((loff_t)mpd->first_page) << PAGE_SHIFT;
2500 70 : if (disksize > READ_ONCE(EXT4_I(inode)->i_disksize)) {
2501 70 : int err2;
2502 70 : loff_t i_size;
2503 :
2504 70 : down_write(&EXT4_I(inode)->i_data_sem);
2505 70 : i_size = i_size_read(inode);
2506 70 : if (disksize > i_size)
2507 : disksize = i_size;
2508 70 : if (disksize > EXT4_I(inode)->i_disksize)
2509 69 : EXT4_I(inode)->i_disksize = disksize;
2510 70 : up_write(&EXT4_I(inode)->i_data_sem);
2511 70 : err2 = ext4_mark_inode_dirty(handle, inode);
2512 70 : if (err2) {
2513 0 : ext4_error_err(inode->i_sb, -err2,
2514 : "Failed to mark inode %lu dirty",
2515 : inode->i_ino);
2516 : }
2517 70 : if (!err)
2518 70 : err = err2;
2519 : }
2520 : return err;
2521 : }
2522 :
2523 : /*
2524 : * Calculate the total number of credits to reserve for one writepages
2525 : * iteration. This is called from ext4_writepages(). We map an extent of
2526 : * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2527 : * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2528 : * bpp - 1 blocks in bpp different extents.
2529 : */
2530 70 : static int ext4_da_writepages_trans_blocks(struct inode *inode)
2531 : {
2532 70 : int bpp = ext4_journal_blocks_per_page(inode);
2533 :
2534 70 : return ext4_meta_trans_blocks(inode,
2535 : MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2536 : }
2537 :
2538 : /*
2539 : * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2540 : * and underlying extent to map
2541 : *
2542 : * @mpd - where to look for pages
2543 : *
2544 : * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2545 : * IO immediately. When we find a page which isn't mapped we start accumulating
2546 : * extent of buffers underlying these pages that needs mapping (formed by
2547 : * either delayed or unwritten buffers). We also lock the pages containing
2548 : * these buffers. The extent found is returned in @mpd structure (starting at
2549 : * mpd->lblk with length mpd->len blocks).
2550 : *
2551 : * Note that this function can attach bios to one io_end structure which are
2552 : * neither logically nor physically contiguous. Although it may seem as an
2553 : * unnecessary complication, it is actually inevitable in blocksize < pagesize
2554 : * case as we need to track IO to all buffers underlying a page in one io_end.
2555 : */
2556 154 : static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2557 : {
2558 154 : struct address_space *mapping = mpd->inode->i_mapping;
2559 154 : struct pagevec pvec;
2560 154 : unsigned int nr_pages;
2561 154 : long left = mpd->wbc->nr_to_write;
2562 154 : pgoff_t index = mpd->first_page;
2563 154 : pgoff_t end = mpd->last_page;
2564 154 : xa_mark_t tag;
2565 154 : int i, err = 0;
2566 154 : int blkbits = mpd->inode->i_blkbits;
2567 154 : ext4_lblk_t lblk;
2568 154 : struct buffer_head *head;
2569 :
2570 154 : if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2571 : tag = PAGECACHE_TAG_TOWRITE;
2572 : else
2573 30 : tag = PAGECACHE_TAG_DIRTY;
2574 :
2575 154 : pagevec_init(&pvec);
2576 154 : mpd->map.m_len = 0;
2577 154 : mpd->next_page = index;
2578 203 : while (index <= end) {
2579 203 : nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
2580 : tag);
2581 203 : if (nr_pages == 0)
2582 : break;
2583 :
2584 1198 : for (i = 0; i < nr_pages; i++) {
2585 1149 : struct page *page = pvec.pages[i];
2586 :
2587 : /*
2588 : * Accumulated enough dirty pages? This doesn't apply
2589 : * to WB_SYNC_ALL mode. For integrity sync we have to
2590 : * keep going because someone may be concurrently
2591 : * dirtying pages, and we might have synced a lot of
2592 : * newly appeared dirty pages, but have not synced all
2593 : * of the old dirty pages.
2594 : */
2595 1149 : if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0)
2596 0 : goto out;
2597 :
2598 : /* If we can't merge this page, we are done. */
2599 1149 : if (mpd->map.m_len > 0 && mpd->next_page != page->index)
2600 0 : goto out;
2601 :
2602 1149 : lock_page(page);
2603 : /*
2604 : * If the page is no longer dirty, or its mapping no
2605 : * longer corresponds to inode we are writing (which
2606 : * means it has been truncated or invalidated), or the
2607 : * page is already under writeback and we are not doing
2608 : * a data integrity writeback, skip the page
2609 : */
2610 2298 : if (!PageDirty(page) ||
2611 1149 : (PageWriteback(page) &&
2612 0 : (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2613 1149 : unlikely(page->mapping != mapping)) {
2614 0 : unlock_page(page);
2615 0 : continue;
2616 : }
2617 :
2618 1149 : wait_on_page_writeback(page);
2619 2298 : BUG_ON(PageWriteback(page));
2620 :
2621 1149 : if (mpd->map.m_len == 0)
2622 299 : mpd->first_page = page->index;
2623 1149 : mpd->next_page = page->index + 1;
2624 : /* Add all dirty buffers to mpd */
2625 1149 : lblk = ((ext4_lblk_t)page->index) <<
2626 1149 : (PAGE_SHIFT - blkbits);
2627 1149 : head = page_buffers(page);
2628 1149 : err = mpage_process_page_bufs(mpd, head, head, lblk);
2629 1149 : if (err <= 0)
2630 154 : goto out;
2631 995 : err = 0;
2632 995 : left--;
2633 : }
2634 49 : pagevec_release(&pvec);
2635 49 : cond_resched();
2636 : }
2637 0 : mpd->scanned_until_end = 1;
2638 0 : return 0;
2639 154 : out:
2640 154 : pagevec_release(&pvec);
2641 : return err;
2642 : }
2643 :
2644 548 : static int ext4_writepages(struct address_space *mapping,
2645 : struct writeback_control *wbc)
2646 : {
2647 548 : pgoff_t writeback_index = 0;
2648 548 : long nr_to_write = wbc->nr_to_write;
2649 548 : int range_whole = 0;
2650 548 : int cycled = 1;
2651 548 : handle_t *handle = NULL;
2652 548 : struct mpage_da_data mpd;
2653 548 : struct inode *inode = mapping->host;
2654 548 : int needed_blocks, rsv_blocks = 0, ret = 0;
2655 548 : struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2656 548 : struct blk_plug plug;
2657 548 : bool give_up_on_write = false;
2658 :
2659 548 : if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2660 : return -EIO;
2661 :
2662 548 : percpu_down_read(&sbi->s_writepages_rwsem);
2663 548 : trace_ext4_writepages(inode, wbc);
2664 :
2665 : /*
2666 : * No pages to write? This is mainly a kludge to avoid starting
2667 : * a transaction for special inodes like journal inode on last iput()
2668 : * because that could violate lock ordering on umount
2669 : */
2670 548 : if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2671 464 : goto out_writepages;
2672 :
2673 84 : if (ext4_should_journal_data(inode)) {
2674 0 : ret = generic_writepages(mapping, wbc);
2675 0 : goto out_writepages;
2676 : }
2677 :
2678 : /*
2679 : * If the filesystem has aborted, it is read-only, so return
2680 : * right away instead of dumping stack traces later on that
2681 : * will obscure the real source of the problem. We test
2682 : * EXT4_MF_FS_ABORTED instead of sb->s_flag's SB_RDONLY because
2683 : * the latter could be true if the filesystem is mounted
2684 : * read-only, and in that case, ext4_writepages should
2685 : * *never* be called, so if that ever happens, we would want
2686 : * the stack trace.
2687 : */
2688 84 : if (unlikely(ext4_forced_shutdown(EXT4_SB(mapping->host->i_sb)) ||
2689 : ext4_test_mount_flag(inode->i_sb, EXT4_MF_FS_ABORTED))) {
2690 0 : ret = -EROFS;
2691 0 : goto out_writepages;
2692 : }
2693 :
2694 : /*
2695 : * If we have inline data and arrive here, it means that
2696 : * we will soon create the block for the 1st page, so
2697 : * we'd better clear the inline data here.
2698 : */
2699 84 : if (ext4_has_inline_data(inode)) {
2700 : /* Just inode will be modified... */
2701 0 : handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2702 0 : if (IS_ERR(handle)) {
2703 0 : ret = PTR_ERR(handle);
2704 0 : goto out_writepages;
2705 : }
2706 0 : BUG_ON(ext4_test_inode_state(inode,
2707 : EXT4_STATE_MAY_INLINE_DATA));
2708 0 : ext4_destroy_inline_data(handle, inode);
2709 0 : ext4_journal_stop(handle);
2710 : }
2711 :
2712 84 : if (ext4_should_dioread_nolock(inode)) {
2713 : /*
2714 : * We may need to convert up to one extent per block in
2715 : * the page and we may dirty the inode.
2716 : */
2717 84 : rsv_blocks = 1 + ext4_chunk_trans_blocks(inode,
2718 84 : PAGE_SIZE >> inode->i_blkbits);
2719 : }
2720 :
2721 84 : if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2722 57 : range_whole = 1;
2723 :
2724 84 : if (wbc->range_cyclic) {
2725 15 : writeback_index = mapping->writeback_index;
2726 15 : if (writeback_index)
2727 0 : cycled = 0;
2728 15 : mpd.first_page = writeback_index;
2729 15 : mpd.last_page = -1;
2730 : } else {
2731 69 : mpd.first_page = wbc->range_start >> PAGE_SHIFT;
2732 69 : mpd.last_page = wbc->range_end >> PAGE_SHIFT;
2733 : }
2734 :
2735 84 : mpd.inode = inode;
2736 84 : mpd.wbc = wbc;
2737 84 : ext4_io_submit_init(&mpd.io_submit, wbc);
2738 84 : retry:
2739 84 : if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2740 67 : tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page);
2741 84 : blk_start_plug(&plug);
2742 :
2743 : /*
2744 : * First writeback pages that don't need mapping - we can avoid
2745 : * starting a transaction unnecessarily and also avoid being blocked
2746 : * in the block layer on device congestion while having transaction
2747 : * started.
2748 : */
2749 84 : mpd.do_map = 0;
2750 84 : mpd.scanned_until_end = 0;
2751 84 : mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2752 84 : if (!mpd.io_submit.io_end) {
2753 0 : ret = -ENOMEM;
2754 0 : goto unplug;
2755 : }
2756 84 : ret = mpage_prepare_extent_to_map(&mpd);
2757 : /* Unlock pages we didn't use */
2758 84 : mpage_release_unused_pages(&mpd, false);
2759 : /* Submit prepared bio */
2760 84 : ext4_io_submit(&mpd.io_submit);
2761 84 : ext4_put_io_end_defer(mpd.io_submit.io_end);
2762 84 : mpd.io_submit.io_end = NULL;
2763 84 : if (ret < 0)
2764 0 : goto unplug;
2765 :
2766 154 : while (!mpd.scanned_until_end && wbc->nr_to_write > 0) {
2767 : /* For each extent of pages we use new io_end */
2768 70 : mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2769 70 : if (!mpd.io_submit.io_end) {
2770 : ret = -ENOMEM;
2771 : break;
2772 : }
2773 :
2774 : /*
2775 : * We have two constraints: We find one extent to map and we
2776 : * must always write out whole page (makes a difference when
2777 : * blocksize < pagesize) so that we don't block on IO when we
2778 : * try to write out the rest of the page. Journalled mode is
2779 : * not supported by delalloc.
2780 : */
2781 70 : BUG_ON(ext4_should_journal_data(inode));
2782 70 : needed_blocks = ext4_da_writepages_trans_blocks(inode);
2783 :
2784 : /* start a new transaction */
2785 70 : handle = ext4_journal_start_with_reserve(inode,
2786 : EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2787 70 : if (IS_ERR(handle)) {
2788 0 : ret = PTR_ERR(handle);
2789 0 : ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2790 : "%ld pages, ino %lu; err %d", __func__,
2791 : wbc->nr_to_write, inode->i_ino, ret);
2792 : /* Release allocated io_end */
2793 0 : ext4_put_io_end(mpd.io_submit.io_end);
2794 0 : mpd.io_submit.io_end = NULL;
2795 0 : break;
2796 : }
2797 70 : mpd.do_map = 1;
2798 :
2799 70 : trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc);
2800 70 : ret = mpage_prepare_extent_to_map(&mpd);
2801 70 : if (!ret && mpd.map.m_len)
2802 70 : ret = mpage_map_and_submit_extent(handle, &mpd,
2803 : &give_up_on_write);
2804 : /*
2805 : * Caution: If the handle is synchronous,
2806 : * ext4_journal_stop() can wait for transaction commit
2807 : * to finish which may depend on writeback of pages to
2808 : * complete or on page lock to be released. In that
2809 : * case, we have to wait until after we have
2810 : * submitted all the IO, released page locks we hold,
2811 : * and dropped io_end reference (for extent conversion
2812 : * to be able to complete) before stopping the handle.
2813 : */
2814 70 : if (!ext4_handle_valid(handle) || handle->h_sync == 0) {
2815 70 : ext4_journal_stop(handle);
2816 70 : handle = NULL;
2817 70 : mpd.do_map = 0;
2818 : }
2819 : /* Unlock pages we didn't use */
2820 70 : mpage_release_unused_pages(&mpd, give_up_on_write);
2821 : /* Submit prepared bio */
2822 70 : ext4_io_submit(&mpd.io_submit);
2823 :
2824 : /*
2825 : * Drop our io_end reference we got from init. We have
2826 : * to be careful and use deferred io_end finishing if
2827 : * we are still holding the transaction as we can
2828 : * release the last reference to io_end which may end
2829 : * up doing unwritten extent conversion.
2830 : */
2831 70 : if (handle) {
2832 0 : ext4_put_io_end_defer(mpd.io_submit.io_end);
2833 0 : ext4_journal_stop(handle);
2834 : } else
2835 70 : ext4_put_io_end(mpd.io_submit.io_end);
2836 70 : mpd.io_submit.io_end = NULL;
2837 :
2838 70 : if (ret == -ENOSPC && sbi->s_journal) {
2839 : /*
2840 : * Commit the transaction which would
2841 : * free blocks released in the transaction
2842 : * and try again
2843 : */
2844 0 : jbd2_journal_force_commit_nested(sbi->s_journal);
2845 0 : ret = 0;
2846 0 : continue;
2847 : }
2848 : /* Fatal error - ENOMEM, EIO... */
2849 70 : if (ret)
2850 : break;
2851 : }
2852 84 : unplug:
2853 84 : blk_finish_plug(&plug);
2854 84 : if (!ret && !cycled && wbc->nr_to_write > 0) {
2855 0 : cycled = 1;
2856 0 : mpd.last_page = writeback_index - 1;
2857 0 : mpd.first_page = 0;
2858 0 : goto retry;
2859 : }
2860 :
2861 : /* Update index */
2862 84 : if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
2863 : /*
2864 : * Set the writeback_index so that range_cyclic
2865 : * mode will write it back later
2866 : */
2867 57 : mapping->writeback_index = mpd.first_page;
2868 :
2869 27 : out_writepages:
2870 548 : trace_ext4_writepages_result(inode, wbc, ret,
2871 548 : nr_to_write - wbc->nr_to_write);
2872 548 : percpu_up_read(&sbi->s_writepages_rwsem);
2873 548 : return ret;
2874 : }
2875 :
2876 : static int ext4_dax_writepages(struct address_space *mapping,
2877 : struct writeback_control *wbc)
2878 : {
2879 : int ret;
2880 : long nr_to_write = wbc->nr_to_write;
2881 : struct inode *inode = mapping->host;
2882 : struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2883 :
2884 : if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2885 : return -EIO;
2886 :
2887 : percpu_down_read(&sbi->s_writepages_rwsem);
2888 : trace_ext4_writepages(inode, wbc);
2889 :
2890 : ret = dax_writeback_mapping_range(mapping, sbi->s_daxdev, wbc);
2891 : trace_ext4_writepages_result(inode, wbc, ret,
2892 : nr_to_write - wbc->nr_to_write);
2893 : percpu_up_read(&sbi->s_writepages_rwsem);
2894 : return ret;
2895 : }
2896 :
2897 2507 : static int ext4_nonda_switch(struct super_block *sb)
2898 : {
2899 2507 : s64 free_clusters, dirty_clusters;
2900 2507 : struct ext4_sb_info *sbi = EXT4_SB(sb);
2901 :
2902 : /*
2903 : * switch to non delalloc mode if we are running low
2904 : * on free block. The free block accounting via percpu
2905 : * counters can get slightly wrong with percpu_counter_batch getting
2906 : * accumulated on each CPU without updating global counters
2907 : * Delalloc need an accurate free block accounting. So switch
2908 : * to non delalloc when we are near to error range.
2909 : */
2910 2507 : free_clusters =
2911 2507 : percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2912 2507 : dirty_clusters =
2913 2507 : percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2914 : /*
2915 : * Start pushing delalloc when 1/2 of free blocks are dirty.
2916 : */
2917 2507 : if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
2918 0 : try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2919 :
2920 2507 : if (2 * free_clusters < 3 * dirty_clusters ||
2921 2507 : free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
2922 : /*
2923 : * free block count is less than 150% of dirty blocks
2924 : * or free blocks is less than watermark
2925 : */
2926 0 : return 1;
2927 : }
2928 : return 0;
2929 : }
2930 :
2931 : /* We always reserve for an inode update; the superblock could be there too */
2932 2203 : static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len)
2933 : {
2934 2203 : if (likely(ext4_has_feature_large_file(inode->i_sb)))
2935 : return 1;
2936 :
2937 0 : if (pos + len <= 0x7fffffffULL)
2938 0 : return 1;
2939 :
2940 : /* We might need to update the superblock to set LARGE_FILE */
2941 : return 2;
2942 : }
2943 :
2944 2203 : static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2945 : loff_t pos, unsigned len, unsigned flags,
2946 : struct page **pagep, void **fsdata)
2947 : {
2948 2203 : int ret, retries = 0;
2949 2203 : struct page *page;
2950 2203 : pgoff_t index;
2951 2203 : struct inode *inode = mapping->host;
2952 2203 : handle_t *handle;
2953 :
2954 2203 : if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2955 : return -EIO;
2956 :
2957 2203 : index = pos >> PAGE_SHIFT;
2958 :
2959 2203 : if (ext4_nonda_switch(inode->i_sb) || S_ISLNK(inode->i_mode) ||
2960 2203 : ext4_verity_in_progress(inode)) {
2961 0 : *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2962 0 : return ext4_write_begin(file, mapping, pos,
2963 : len, flags, pagep, fsdata);
2964 : }
2965 2203 : *fsdata = (void *)0;
2966 2203 : trace_ext4_da_write_begin(inode, pos, len, flags);
2967 :
2968 2203 : if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2969 0 : ret = ext4_da_write_inline_data_begin(mapping, inode,
2970 : pos, len, flags,
2971 : pagep, fsdata);
2972 0 : if (ret < 0)
2973 : return ret;
2974 0 : if (ret == 1)
2975 : return 0;
2976 : }
2977 :
2978 : /*
2979 : * grab_cache_page_write_begin() can take a long time if the
2980 : * system is thrashing due to memory pressure, or if the page
2981 : * is being written back. So grab it first before we start
2982 : * the transaction handle. This also allows us to allocate
2983 : * the page (if needed) without using GFP_NOFS.
2984 : */
2985 2203 : retry_grab:
2986 2203 : page = grab_cache_page_write_begin(mapping, index, flags);
2987 2203 : if (!page)
2988 : return -ENOMEM;
2989 2203 : unlock_page(page);
2990 :
2991 : /*
2992 : * With delayed allocation, we don't log the i_disksize update
2993 : * if there is delayed block allocation. But we still need
2994 : * to journalling the i_disksize update if writes to the end
2995 : * of file which has an already mapped buffer.
2996 : */
2997 2203 : retry_journal:
2998 2203 : handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2999 : ext4_da_write_credits(inode, pos, len));
3000 2203 : if (IS_ERR(handle)) {
3001 0 : put_page(page);
3002 0 : return PTR_ERR(handle);
3003 : }
3004 :
3005 2203 : lock_page(page);
3006 2203 : if (page->mapping != mapping) {
3007 : /* The page got truncated from under us */
3008 0 : unlock_page(page);
3009 0 : put_page(page);
3010 0 : ext4_journal_stop(handle);
3011 0 : goto retry_grab;
3012 : }
3013 : /* In case writeback began while the page was unlocked */
3014 2203 : wait_for_stable_page(page);
3015 :
3016 : #ifdef CONFIG_FS_ENCRYPTION
3017 : ret = ext4_block_write_begin(page, pos, len,
3018 : ext4_da_get_block_prep);
3019 : #else
3020 2203 : ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
3021 : #endif
3022 2203 : if (ret < 0) {
3023 0 : unlock_page(page);
3024 0 : ext4_journal_stop(handle);
3025 : /*
3026 : * block_write_begin may have instantiated a few blocks
3027 : * outside i_size. Trim these off again. Don't need
3028 : * i_size_read because we hold i_mutex.
3029 : */
3030 0 : if (pos + len > inode->i_size)
3031 0 : ext4_truncate_failed_write(inode);
3032 :
3033 0 : if (ret == -ENOSPC &&
3034 0 : ext4_should_retry_alloc(inode->i_sb, &retries))
3035 0 : goto retry_journal;
3036 :
3037 0 : put_page(page);
3038 0 : return ret;
3039 : }
3040 :
3041 2203 : *pagep = page;
3042 2203 : return ret;
3043 : }
3044 :
3045 : /*
3046 : * Check if we should update i_disksize
3047 : * when write to the end of file but not require block allocation
3048 : */
3049 2110 : static int ext4_da_should_update_i_disksize(struct page *page,
3050 : unsigned long offset)
3051 : {
3052 2110 : struct buffer_head *bh;
3053 2110 : struct inode *inode = page->mapping->host;
3054 2110 : unsigned int idx;
3055 2110 : int i;
3056 :
3057 2110 : bh = page_buffers(page);
3058 2110 : idx = offset >> inode->i_blkbits;
3059 :
3060 2110 : for (i = 0; i < idx; i++)
3061 0 : bh = bh->b_this_page;
3062 :
3063 4253 : if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
3064 2077 : return 0;
3065 : return 1;
3066 : }
3067 :
3068 2203 : static int ext4_da_write_end(struct file *file,
3069 : struct address_space *mapping,
3070 : loff_t pos, unsigned len, unsigned copied,
3071 : struct page *page, void *fsdata)
3072 : {
3073 2203 : struct inode *inode = mapping->host;
3074 2203 : int ret = 0, ret2;
3075 2203 : handle_t *handle = ext4_journal_current_handle();
3076 2203 : loff_t new_i_size;
3077 2203 : unsigned long start, end;
3078 2203 : int write_mode = (int)(unsigned long)fsdata;
3079 :
3080 2203 : if (write_mode == FALL_BACK_TO_NONDELALLOC)
3081 0 : return ext4_write_end(file, mapping, pos,
3082 : len, copied, page, fsdata);
3083 :
3084 2203 : trace_ext4_da_write_end(inode, pos, len, copied);
3085 2203 : start = pos & (PAGE_SIZE - 1);
3086 2203 : end = start + copied - 1;
3087 :
3088 : /*
3089 : * generic_write_end() will run mark_inode_dirty() if i_size
3090 : * changes. So let's piggyback the i_disksize mark_inode_dirty
3091 : * into that.
3092 : */
3093 2203 : new_i_size = pos + copied;
3094 2203 : if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
3095 4220 : if (ext4_has_inline_data(inode) ||
3096 2110 : ext4_da_should_update_i_disksize(page, end)) {
3097 33 : ext4_update_i_disksize(inode, new_i_size);
3098 : /* We need to mark inode dirty even if
3099 : * new_i_size is less that inode->i_size
3100 : * bu greater than i_disksize.(hint delalloc)
3101 : */
3102 33 : ret = ext4_mark_inode_dirty(handle, inode);
3103 : }
3104 : }
3105 :
3106 2203 : if (write_mode != CONVERT_INLINE_DATA &&
3107 2203 : ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
3108 0 : ext4_has_inline_data(inode))
3109 0 : ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
3110 : page);
3111 : else
3112 2203 : ret2 = generic_write_end(file, mapping, pos, len, copied,
3113 : page, fsdata);
3114 :
3115 2203 : copied = ret2;
3116 2203 : if (ret2 < 0)
3117 0 : ret = ret2;
3118 2203 : ret2 = ext4_journal_stop(handle);
3119 2203 : if (unlikely(ret2 && !ret))
3120 0 : ret = ret2;
3121 :
3122 2203 : return ret ? ret : copied;
3123 : }
3124 :
3125 : /*
3126 : * Force all delayed allocation blocks to be allocated for a given inode.
3127 : */
3128 32 : int ext4_alloc_da_blocks(struct inode *inode)
3129 : {
3130 32 : trace_ext4_alloc_da_blocks(inode);
3131 :
3132 32 : if (!EXT4_I(inode)->i_reserved_data_blocks)
3133 : return 0;
3134 :
3135 : /*
3136 : * We do something simple for now. The filemap_flush() will
3137 : * also start triggering a write of the data blocks, which is
3138 : * not strictly speaking necessary (and for users of
3139 : * laptop_mode, not even desirable). However, to do otherwise
3140 : * would require replicating code paths in:
3141 : *
3142 : * ext4_writepages() ->
3143 : * write_cache_pages() ---> (via passed in callback function)
3144 : * __mpage_da_writepage() -->
3145 : * mpage_add_bh_to_extent()
3146 : * mpage_da_map_blocks()
3147 : *
3148 : * The problem is that write_cache_pages(), located in
3149 : * mm/page-writeback.c, marks pages clean in preparation for
3150 : * doing I/O, which is not desirable if we're not planning on
3151 : * doing I/O at all.
3152 : *
3153 : * We could call write_cache_pages(), and then redirty all of
3154 : * the pages by calling redirty_page_for_writepage() but that
3155 : * would be ugly in the extreme. So instead we would need to
3156 : * replicate parts of the code in the above functions,
3157 : * simplifying them because we wouldn't actually intend to
3158 : * write out the pages, but rather only collect contiguous
3159 : * logical block extents, call the multi-block allocator, and
3160 : * then update the buffer heads with the block allocations.
3161 : *
3162 : * For now, though, we'll cheat by calling filemap_flush(),
3163 : * which will map the blocks, and start the I/O, but not
3164 : * actually wait for the I/O to complete.
3165 : */
3166 2 : return filemap_flush(inode->i_mapping);
3167 : }
3168 :
3169 : /*
3170 : * bmap() is special. It gets used by applications such as lilo and by
3171 : * the swapper to find the on-disk block of a specific piece of data.
3172 : *
3173 : * Naturally, this is dangerous if the block concerned is still in the
3174 : * journal. If somebody makes a swapfile on an ext4 data-journaling
3175 : * filesystem and enables swap, then they may get a nasty shock when the
3176 : * data getting swapped to that swapfile suddenly gets overwritten by
3177 : * the original zero's written out previously to the journal and
3178 : * awaiting writeback in the kernel's buffer cache.
3179 : *
3180 : * So, if we see any bmap calls here on a modified, data-journaled file,
3181 : * take extra steps to flush any blocks which might be in the cache.
3182 : */
3183 1948 : static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3184 : {
3185 1948 : struct inode *inode = mapping->host;
3186 1948 : journal_t *journal;
3187 1948 : int err;
3188 :
3189 : /*
3190 : * We can get here for an inline file via the FIBMAP ioctl
3191 : */
3192 1948 : if (ext4_has_inline_data(inode))
3193 : return 0;
3194 :
3195 1948 : if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3196 0 : test_opt(inode->i_sb, DELALLOC)) {
3197 : /*
3198 : * With delalloc we want to sync the file
3199 : * so that we can make sure we allocate
3200 : * blocks for file
3201 : */
3202 0 : filemap_write_and_wait(mapping);
3203 : }
3204 :
3205 1948 : if (EXT4_JOURNAL(inode) &&
3206 1947 : ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
3207 : /*
3208 : * This is a REALLY heavyweight approach, but the use of
3209 : * bmap on dirty files is expected to be extremely rare:
3210 : * only if we run lilo or swapon on a freshly made file
3211 : * do we expect this to happen.
3212 : *
3213 : * (bmap requires CAP_SYS_RAWIO so this does not
3214 : * represent an unprivileged user DOS attack --- we'd be
3215 : * in trouble if mortal users could trigger this path at
3216 : * will.)
3217 : *
3218 : * NB. EXT4_STATE_JDATA is not set on files other than
3219 : * regular files. If somebody wants to bmap a directory
3220 : * or symlink and gets confused because the buffer
3221 : * hasn't yet been flushed to disk, they deserve
3222 : * everything they get.
3223 : */
3224 :
3225 0 : ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
3226 0 : journal = EXT4_JOURNAL(inode);
3227 0 : jbd2_journal_lock_updates(journal);
3228 0 : err = jbd2_journal_flush(journal);
3229 0 : jbd2_journal_unlock_updates(journal);
3230 :
3231 0 : if (err)
3232 : return 0;
3233 : }
3234 :
3235 1948 : return iomap_bmap(mapping, block, &ext4_iomap_ops);
3236 : }
3237 :
3238 28 : static int ext4_readpage(struct file *file, struct page *page)
3239 : {
3240 28 : int ret = -EAGAIN;
3241 28 : struct inode *inode = page->mapping->host;
3242 :
3243 28 : trace_ext4_readpage(page);
3244 :
3245 28 : if (ext4_has_inline_data(inode))
3246 0 : ret = ext4_readpage_inline(inode, page);
3247 :
3248 0 : if (ret == -EAGAIN)
3249 28 : return ext4_mpage_readpages(inode, NULL, page);
3250 :
3251 : return ret;
3252 : }
3253 :
3254 1528 : static void ext4_readahead(struct readahead_control *rac)
3255 : {
3256 1528 : struct inode *inode = rac->mapping->host;
3257 :
3258 : /* If the file has inline data, no need to do readahead. */
3259 1528 : if (ext4_has_inline_data(inode))
3260 : return;
3261 :
3262 1528 : ext4_mpage_readpages(inode, rac, NULL);
3263 : }
3264 :
3265 498 : static void ext4_invalidatepage(struct page *page, unsigned int offset,
3266 : unsigned int length)
3267 : {
3268 498 : trace_ext4_invalidatepage(page, offset, length);
3269 :
3270 : /* No journalling happens on data buffers when this function is used */
3271 996 : WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
3272 :
3273 498 : block_invalidatepage(page, offset, length);
3274 498 : }
3275 :
3276 0 : static int __ext4_journalled_invalidatepage(struct page *page,
3277 : unsigned int offset,
3278 : unsigned int length)
3279 : {
3280 0 : journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3281 :
3282 0 : trace_ext4_journalled_invalidatepage(page, offset, length);
3283 :
3284 : /*
3285 : * If it's a full truncate we just forget about the pending dirtying
3286 : */
3287 0 : if (offset == 0 && length == PAGE_SIZE)
3288 0 : ClearPageChecked(page);
3289 :
3290 0 : return jbd2_journal_invalidatepage(journal, page, offset, length);
3291 : }
3292 :
3293 : /* Wrapper for aops... */
3294 0 : static void ext4_journalled_invalidatepage(struct page *page,
3295 : unsigned int offset,
3296 : unsigned int length)
3297 : {
3298 0 : WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0);
3299 0 : }
3300 :
3301 496 : static int ext4_releasepage(struct page *page, gfp_t wait)
3302 : {
3303 496 : journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3304 :
3305 496 : trace_ext4_releasepage(page);
3306 :
3307 : /* Page has dirty journalled data -> cannot release */
3308 496 : if (PageChecked(page))
3309 : return 0;
3310 496 : if (journal)
3311 496 : return jbd2_journal_try_to_free_buffers(journal, page);
3312 : else
3313 0 : return try_to_free_buffers(page);
3314 : }
3315 :
3316 1948 : static bool ext4_inode_datasync_dirty(struct inode *inode)
3317 : {
3318 1948 : journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
3319 :
3320 1948 : if (journal) {
3321 1947 : if (jbd2_transaction_committed(journal,
3322 1947 : EXT4_I(inode)->i_datasync_tid))
3323 : return false;
3324 0 : if (test_opt2(inode->i_sb, JOURNAL_FAST_COMMIT))
3325 0 : return !list_empty(&EXT4_I(inode)->i_fc_list);
3326 : return true;
3327 : }
3328 :
3329 : /* Any metadata buffers to write? */
3330 1 : if (!list_empty(&inode->i_mapping->private_list))
3331 : return true;
3332 1 : return inode->i_state & I_DIRTY_DATASYNC;
3333 : }
3334 :
3335 1948 : static void ext4_set_iomap(struct inode *inode, struct iomap *iomap,
3336 : struct ext4_map_blocks *map, loff_t offset,
3337 : loff_t length)
3338 : {
3339 1948 : u8 blkbits = inode->i_blkbits;
3340 :
3341 : /*
3342 : * Writes that span EOF might trigger an I/O size update on completion,
3343 : * so consider them to be dirty for the purpose of O_DSYNC, even if
3344 : * there is no other metadata changes being made or are pending.
3345 : */
3346 1948 : iomap->flags = 0;
3347 1948 : if (ext4_inode_datasync_dirty(inode) ||
3348 1948 : offset + length > i_size_read(inode))
3349 0 : iomap->flags |= IOMAP_F_DIRTY;
3350 :
3351 1948 : if (map->m_flags & EXT4_MAP_NEW)
3352 0 : iomap->flags |= IOMAP_F_NEW;
3353 :
3354 1948 : iomap->bdev = inode->i_sb->s_bdev;
3355 1948 : iomap->dax_dev = EXT4_SB(inode->i_sb)->s_daxdev;
3356 1948 : iomap->offset = (u64) map->m_lblk << blkbits;
3357 1948 : iomap->length = (u64) map->m_len << blkbits;
3358 :
3359 1948 : if ((map->m_flags & EXT4_MAP_MAPPED) &&
3360 1948 : !ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3361 0 : iomap->flags |= IOMAP_F_MERGED;
3362 :
3363 : /*
3364 : * Flags passed to ext4_map_blocks() for direct I/O writes can result
3365 : * in m_flags having both EXT4_MAP_MAPPED and EXT4_MAP_UNWRITTEN bits
3366 : * set. In order for any allocated unwritten extents to be converted
3367 : * into written extents correctly within the ->end_io() handler, we
3368 : * need to ensure that the iomap->type is set appropriately. Hence, the
3369 : * reason why we need to check whether the EXT4_MAP_UNWRITTEN bit has
3370 : * been set first.
3371 : */
3372 1948 : if (map->m_flags & EXT4_MAP_UNWRITTEN) {
3373 0 : iomap->type = IOMAP_UNWRITTEN;
3374 0 : iomap->addr = (u64) map->m_pblk << blkbits;
3375 1948 : } else if (map->m_flags & EXT4_MAP_MAPPED) {
3376 1948 : iomap->type = IOMAP_MAPPED;
3377 1948 : iomap->addr = (u64) map->m_pblk << blkbits;
3378 : } else {
3379 0 : iomap->type = IOMAP_HOLE;
3380 0 : iomap->addr = IOMAP_NULL_ADDR;
3381 : }
3382 1948 : }
3383 :
3384 0 : static int ext4_iomap_alloc(struct inode *inode, struct ext4_map_blocks *map,
3385 : unsigned int flags)
3386 : {
3387 0 : handle_t *handle;
3388 0 : u8 blkbits = inode->i_blkbits;
3389 0 : int ret, dio_credits, m_flags = 0, retries = 0;
3390 :
3391 : /*
3392 : * Trim the mapping request to the maximum value that we can map at
3393 : * once for direct I/O.
3394 : */
3395 0 : if (map->m_len > DIO_MAX_BLOCKS)
3396 0 : map->m_len = DIO_MAX_BLOCKS;
3397 0 : dio_credits = ext4_chunk_trans_blocks(inode, map->m_len);
3398 :
3399 0 : retry:
3400 : /*
3401 : * Either we allocate blocks and then don't get an unwritten extent, so
3402 : * in that case we have reserved enough credits. Or, the blocks are
3403 : * already allocated and unwritten. In that case, the extent conversion
3404 : * fits into the credits as well.
3405 : */
3406 0 : handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits);
3407 0 : if (IS_ERR(handle))
3408 0 : return PTR_ERR(handle);
3409 :
3410 : /*
3411 : * DAX and direct I/O are the only two operations that are currently
3412 : * supported with IOMAP_WRITE.
3413 : */
3414 0 : WARN_ON(!IS_DAX(inode) && !(flags & IOMAP_DIRECT));
3415 0 : if (IS_DAX(inode))
3416 : m_flags = EXT4_GET_BLOCKS_CREATE_ZERO;
3417 : /*
3418 : * We use i_size instead of i_disksize here because delalloc writeback
3419 : * can complete at any point during the I/O and subsequently push the
3420 : * i_disksize out to i_size. This could be beyond where direct I/O is
3421 : * happening and thus expose allocated blocks to direct I/O reads.
3422 : */
3423 0 : else if ((map->m_lblk * (1 << blkbits)) >= i_size_read(inode))
3424 : m_flags = EXT4_GET_BLOCKS_CREATE;
3425 0 : else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3426 0 : m_flags = EXT4_GET_BLOCKS_IO_CREATE_EXT;
3427 :
3428 0 : ret = ext4_map_blocks(handle, inode, map, m_flags);
3429 :
3430 : /*
3431 : * We cannot fill holes in indirect tree based inodes as that could
3432 : * expose stale data in the case of a crash. Use the magic error code
3433 : * to fallback to buffered I/O.
3434 : */
3435 0 : if (!m_flags && !ret)
3436 0 : ret = -ENOTBLK;
3437 :
3438 0 : ext4_journal_stop(handle);
3439 0 : if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3440 0 : goto retry;
3441 :
3442 : return ret;
3443 : }
3444 :
3445 :
3446 1948 : static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
3447 : unsigned flags, struct iomap *iomap, struct iomap *srcmap)
3448 : {
3449 1948 : int ret;
3450 1948 : struct ext4_map_blocks map;
3451 1948 : u8 blkbits = inode->i_blkbits;
3452 :
3453 1948 : if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
3454 : return -EINVAL;
3455 :
3456 1948 : if (WARN_ON_ONCE(ext4_has_inline_data(inode)))
3457 : return -ERANGE;
3458 :
3459 : /*
3460 : * Calculate the first and last logical blocks respectively.
3461 : */
3462 1948 : map.m_lblk = offset >> blkbits;
3463 1948 : map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits,
3464 1948 : EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1;
3465 :
3466 1948 : if (flags & IOMAP_WRITE) {
3467 : /*
3468 : * We check here if the blocks are already allocated, then we
3469 : * don't need to start a journal txn and we can directly return
3470 : * the mapping information. This could boost performance
3471 : * especially in multi-threaded overwrite requests.
3472 : */
3473 0 : if (offset + length <= i_size_read(inode)) {
3474 0 : ret = ext4_map_blocks(NULL, inode, &map, 0);
3475 0 : if (ret > 0 && (map.m_flags & EXT4_MAP_MAPPED))
3476 0 : goto out;
3477 : }
3478 0 : ret = ext4_iomap_alloc(inode, &map, flags);
3479 : } else {
3480 1948 : ret = ext4_map_blocks(NULL, inode, &map, 0);
3481 : }
3482 :
3483 1948 : if (ret < 0)
3484 : return ret;
3485 1948 : out:
3486 1948 : ext4_set_iomap(inode, iomap, &map, offset, length);
3487 :
3488 1948 : return 0;
3489 : }
3490 :
3491 0 : static int ext4_iomap_overwrite_begin(struct inode *inode, loff_t offset,
3492 : loff_t length, unsigned flags, struct iomap *iomap,
3493 : struct iomap *srcmap)
3494 : {
3495 0 : int ret;
3496 :
3497 : /*
3498 : * Even for writes we don't need to allocate blocks, so just pretend
3499 : * we are reading to save overhead of starting a transaction.
3500 : */
3501 0 : flags &= ~IOMAP_WRITE;
3502 0 : ret = ext4_iomap_begin(inode, offset, length, flags, iomap, srcmap);
3503 0 : WARN_ON_ONCE(iomap->type != IOMAP_MAPPED);
3504 0 : return ret;
3505 : }
3506 :
3507 1948 : static int ext4_iomap_end(struct inode *inode, loff_t offset, loff_t length,
3508 : ssize_t written, unsigned flags, struct iomap *iomap)
3509 : {
3510 : /*
3511 : * Check to see whether an error occurred while writing out the data to
3512 : * the allocated blocks. If so, return the magic error code so that we
3513 : * fallback to buffered I/O and attempt to complete the remainder of
3514 : * the I/O. Any blocks that may have been allocated in preparation for
3515 : * the direct I/O will be reused during buffered I/O.
3516 : */
3517 1948 : if (flags & (IOMAP_WRITE | IOMAP_DIRECT) && written == 0)
3518 0 : return -ENOTBLK;
3519 :
3520 : return 0;
3521 : }
3522 :
3523 : const struct iomap_ops ext4_iomap_ops = {
3524 : .iomap_begin = ext4_iomap_begin,
3525 : .iomap_end = ext4_iomap_end,
3526 : };
3527 :
3528 : const struct iomap_ops ext4_iomap_overwrite_ops = {
3529 : .iomap_begin = ext4_iomap_overwrite_begin,
3530 : .iomap_end = ext4_iomap_end,
3531 : };
3532 :
3533 0 : static bool ext4_iomap_is_delalloc(struct inode *inode,
3534 : struct ext4_map_blocks *map)
3535 : {
3536 0 : struct extent_status es;
3537 0 : ext4_lblk_t offset = 0, end = map->m_lblk + map->m_len - 1;
3538 :
3539 0 : ext4_es_find_extent_range(inode, &ext4_es_is_delayed,
3540 : map->m_lblk, end, &es);
3541 :
3542 0 : if (!es.es_len || es.es_lblk > end)
3543 : return false;
3544 :
3545 0 : if (es.es_lblk > map->m_lblk) {
3546 0 : map->m_len = es.es_lblk - map->m_lblk;
3547 0 : return false;
3548 : }
3549 :
3550 0 : offset = map->m_lblk - es.es_lblk;
3551 0 : map->m_len = es.es_len - offset;
3552 :
3553 0 : return true;
3554 : }
3555 :
3556 0 : static int ext4_iomap_begin_report(struct inode *inode, loff_t offset,
3557 : loff_t length, unsigned int flags,
3558 : struct iomap *iomap, struct iomap *srcmap)
3559 : {
3560 0 : int ret;
3561 0 : bool delalloc = false;
3562 0 : struct ext4_map_blocks map;
3563 0 : u8 blkbits = inode->i_blkbits;
3564 :
3565 0 : if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
3566 : return -EINVAL;
3567 :
3568 0 : if (ext4_has_inline_data(inode)) {
3569 0 : ret = ext4_inline_data_iomap(inode, iomap);
3570 0 : if (ret != -EAGAIN) {
3571 0 : if (ret == 0 && offset >= iomap->length)
3572 0 : ret = -ENOENT;
3573 0 : return ret;
3574 : }
3575 : }
3576 :
3577 : /*
3578 : * Calculate the first and last logical block respectively.
3579 : */
3580 0 : map.m_lblk = offset >> blkbits;
3581 0 : map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits,
3582 0 : EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1;
3583 :
3584 : /*
3585 : * Fiemap callers may call for offset beyond s_bitmap_maxbytes.
3586 : * So handle it here itself instead of querying ext4_map_blocks().
3587 : * Since ext4_map_blocks() will warn about it and will return
3588 : * -EIO error.
3589 : */
3590 0 : if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
3591 0 : struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
3592 :
3593 0 : if (offset >= sbi->s_bitmap_maxbytes) {
3594 0 : map.m_flags = 0;
3595 0 : goto set_iomap;
3596 : }
3597 : }
3598 :
3599 0 : ret = ext4_map_blocks(NULL, inode, &map, 0);
3600 0 : if (ret < 0)
3601 : return ret;
3602 0 : if (ret == 0)
3603 0 : delalloc = ext4_iomap_is_delalloc(inode, &map);
3604 :
3605 0 : set_iomap:
3606 0 : ext4_set_iomap(inode, iomap, &map, offset, length);
3607 0 : if (delalloc && iomap->type == IOMAP_HOLE)
3608 0 : iomap->type = IOMAP_DELALLOC;
3609 :
3610 : return 0;
3611 : }
3612 :
3613 : const struct iomap_ops ext4_iomap_report_ops = {
3614 : .iomap_begin = ext4_iomap_begin_report,
3615 : };
3616 :
3617 : /*
3618 : * Pages can be marked dirty completely asynchronously from ext4's journalling
3619 : * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3620 : * much here because ->set_page_dirty is called under VFS locks. The page is
3621 : * not necessarily locked.
3622 : *
3623 : * We cannot just dirty the page and leave attached buffers clean, because the
3624 : * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3625 : * or jbddirty because all the journalling code will explode.
3626 : *
3627 : * So what we do is to mark the page "pending dirty" and next time writepage
3628 : * is called, propagate that into the buffers appropriately.
3629 : */
3630 0 : static int ext4_journalled_set_page_dirty(struct page *page)
3631 : {
3632 0 : SetPageChecked(page);
3633 0 : return __set_page_dirty_nobuffers(page);
3634 : }
3635 :
3636 912 : static int ext4_set_page_dirty(struct page *page)
3637 : {
3638 2296 : WARN_ON_ONCE(!PageLocked(page) && !PageDirty(page));
3639 912 : WARN_ON_ONCE(!page_has_buffers(page));
3640 912 : return __set_page_dirty_buffers(page);
3641 : }
3642 :
3643 0 : static int ext4_iomap_swap_activate(struct swap_info_struct *sis,
3644 : struct file *file, sector_t *span)
3645 : {
3646 0 : return iomap_swapfile_activate(sis, file, span,
3647 : &ext4_iomap_report_ops);
3648 : }
3649 :
3650 : static const struct address_space_operations ext4_aops = {
3651 : .readpage = ext4_readpage,
3652 : .readahead = ext4_readahead,
3653 : .writepage = ext4_writepage,
3654 : .writepages = ext4_writepages,
3655 : .write_begin = ext4_write_begin,
3656 : .write_end = ext4_write_end,
3657 : .set_page_dirty = ext4_set_page_dirty,
3658 : .bmap = ext4_bmap,
3659 : .invalidatepage = ext4_invalidatepage,
3660 : .releasepage = ext4_releasepage,
3661 : .direct_IO = noop_direct_IO,
3662 : .migratepage = buffer_migrate_page,
3663 : .is_partially_uptodate = block_is_partially_uptodate,
3664 : .error_remove_page = generic_error_remove_page,
3665 : .swap_activate = ext4_iomap_swap_activate,
3666 : };
3667 :
3668 : static const struct address_space_operations ext4_journalled_aops = {
3669 : .readpage = ext4_readpage,
3670 : .readahead = ext4_readahead,
3671 : .writepage = ext4_writepage,
3672 : .writepages = ext4_writepages,
3673 : .write_begin = ext4_write_begin,
3674 : .write_end = ext4_journalled_write_end,
3675 : .set_page_dirty = ext4_journalled_set_page_dirty,
3676 : .bmap = ext4_bmap,
3677 : .invalidatepage = ext4_journalled_invalidatepage,
3678 : .releasepage = ext4_releasepage,
3679 : .direct_IO = noop_direct_IO,
3680 : .is_partially_uptodate = block_is_partially_uptodate,
3681 : .error_remove_page = generic_error_remove_page,
3682 : .swap_activate = ext4_iomap_swap_activate,
3683 : };
3684 :
3685 : static const struct address_space_operations ext4_da_aops = {
3686 : .readpage = ext4_readpage,
3687 : .readahead = ext4_readahead,
3688 : .writepage = ext4_writepage,
3689 : .writepages = ext4_writepages,
3690 : .write_begin = ext4_da_write_begin,
3691 : .write_end = ext4_da_write_end,
3692 : .set_page_dirty = ext4_set_page_dirty,
3693 : .bmap = ext4_bmap,
3694 : .invalidatepage = ext4_invalidatepage,
3695 : .releasepage = ext4_releasepage,
3696 : .direct_IO = noop_direct_IO,
3697 : .migratepage = buffer_migrate_page,
3698 : .is_partially_uptodate = block_is_partially_uptodate,
3699 : .error_remove_page = generic_error_remove_page,
3700 : .swap_activate = ext4_iomap_swap_activate,
3701 : };
3702 :
3703 : static const struct address_space_operations ext4_dax_aops = {
3704 : .writepages = ext4_dax_writepages,
3705 : .direct_IO = noop_direct_IO,
3706 : .set_page_dirty = noop_set_page_dirty,
3707 : .bmap = ext4_bmap,
3708 : .invalidatepage = noop_invalidatepage,
3709 : .swap_activate = ext4_iomap_swap_activate,
3710 : };
3711 :
3712 4123 : void ext4_set_aops(struct inode *inode)
3713 : {
3714 4123 : switch (ext4_inode_journal_mode(inode)) {
3715 : case EXT4_INODE_ORDERED_DATA_MODE:
3716 : case EXT4_INODE_WRITEBACK_DATA_MODE:
3717 4123 : break;
3718 0 : case EXT4_INODE_JOURNAL_DATA_MODE:
3719 0 : inode->i_mapping->a_ops = &ext4_journalled_aops;
3720 0 : return;
3721 0 : default:
3722 0 : BUG();
3723 : }
3724 4123 : if (IS_DAX(inode))
3725 : inode->i_mapping->a_ops = &ext4_dax_aops;
3726 4123 : else if (test_opt(inode->i_sb, DELALLOC))
3727 4123 : inode->i_mapping->a_ops = &ext4_da_aops;
3728 : else
3729 0 : inode->i_mapping->a_ops = &ext4_aops;
3730 : }
3731 :
3732 2 : static int __ext4_block_zero_page_range(handle_t *handle,
3733 : struct address_space *mapping, loff_t from, loff_t length)
3734 : {
3735 2 : ext4_fsblk_t index = from >> PAGE_SHIFT;
3736 2 : unsigned offset = from & (PAGE_SIZE-1);
3737 2 : unsigned blocksize, pos;
3738 2 : ext4_lblk_t iblock;
3739 2 : struct inode *inode = mapping->host;
3740 2 : struct buffer_head *bh;
3741 2 : struct page *page;
3742 2 : int err = 0;
3743 :
3744 2 : page = find_or_create_page(mapping, from >> PAGE_SHIFT,
3745 : mapping_gfp_constraint(mapping, ~__GFP_FS));
3746 2 : if (!page)
3747 : return -ENOMEM;
3748 :
3749 2 : blocksize = inode->i_sb->s_blocksize;
3750 :
3751 2 : iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
3752 :
3753 2 : if (!page_has_buffers(page))
3754 0 : create_empty_buffers(page, blocksize, 0);
3755 :
3756 : /* Find the buffer that contains "offset" */
3757 2 : bh = page_buffers(page);
3758 2 : pos = blocksize;
3759 2 : while (offset >= pos) {
3760 0 : bh = bh->b_this_page;
3761 0 : iblock++;
3762 0 : pos += blocksize;
3763 : }
3764 2 : if (buffer_freed(bh)) {
3765 0 : BUFFER_TRACE(bh, "freed: skip");
3766 0 : goto unlock;
3767 : }
3768 2 : if (!buffer_mapped(bh)) {
3769 0 : BUFFER_TRACE(bh, "unmapped");
3770 0 : ext4_get_block(inode, iblock, bh, 0);
3771 : /* unmapped? It's a hole - nothing to do */
3772 0 : if (!buffer_mapped(bh)) {
3773 0 : BUFFER_TRACE(bh, "still unmapped");
3774 0 : goto unlock;
3775 : }
3776 : }
3777 :
3778 : /* Ok, it's mapped. Make sure it's up-to-date */
3779 2 : if (PageUptodate(page))
3780 2 : set_buffer_uptodate(bh);
3781 :
3782 2 : if (!buffer_uptodate(bh)) {
3783 0 : err = ext4_read_bh_lock(bh, 0, true);
3784 0 : if (err)
3785 0 : goto unlock;
3786 2 : if (fscrypt_inode_uses_fs_layer_crypto(inode)) {
3787 : /* We expect the key to be set. */
3788 : BUG_ON(!fscrypt_has_encryption_key(inode));
3789 : err = fscrypt_decrypt_pagecache_blocks(page, blocksize,
3790 : bh_offset(bh));
3791 : if (err) {
3792 : clear_buffer_uptodate(bh);
3793 : goto unlock;
3794 : }
3795 : }
3796 : }
3797 2 : if (ext4_should_journal_data(inode)) {
3798 0 : BUFFER_TRACE(bh, "get write access");
3799 0 : err = ext4_journal_get_write_access(handle, bh);
3800 0 : if (err)
3801 0 : goto unlock;
3802 : }
3803 2 : zero_user(page, offset, length);
3804 2 : BUFFER_TRACE(bh, "zeroed end of block");
3805 :
3806 2 : if (ext4_should_journal_data(inode)) {
3807 0 : err = ext4_handle_dirty_metadata(handle, inode, bh);
3808 : } else {
3809 2 : err = 0;
3810 2 : mark_buffer_dirty(bh);
3811 2 : if (ext4_should_order_data(inode))
3812 2 : err = ext4_jbd2_inode_add_write(handle, inode, from,
3813 : length);
3814 : }
3815 :
3816 2 : unlock:
3817 2 : unlock_page(page);
3818 2 : put_page(page);
3819 2 : return err;
3820 : }
3821 :
3822 : /*
3823 : * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3824 : * starting from file offset 'from'. The range to be zero'd must
3825 : * be contained with in one block. If the specified range exceeds
3826 : * the end of the block it will be shortened to end of the block
3827 : * that cooresponds to 'from'
3828 : */
3829 2 : static int ext4_block_zero_page_range(handle_t *handle,
3830 : struct address_space *mapping, loff_t from, loff_t length)
3831 : {
3832 2 : struct inode *inode = mapping->host;
3833 2 : unsigned offset = from & (PAGE_SIZE-1);
3834 2 : unsigned blocksize = inode->i_sb->s_blocksize;
3835 2 : unsigned max = blocksize - (offset & (blocksize - 1));
3836 :
3837 : /*
3838 : * correct length if it does not fall between
3839 : * 'from' and the end of the block
3840 : */
3841 2 : if (length > max || length < 0)
3842 0 : length = max;
3843 :
3844 2 : if (IS_DAX(inode)) {
3845 : return iomap_zero_range(inode, from, length, NULL,
3846 : &ext4_iomap_ops);
3847 : }
3848 2 : return __ext4_block_zero_page_range(handle, mapping, from, length);
3849 : }
3850 :
3851 : /*
3852 : * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3853 : * up to the end of the block which corresponds to `from'.
3854 : * This required during truncate. We need to physically zero the tail end
3855 : * of that block so it doesn't yield old data if the file is later grown.
3856 : */
3857 2 : static int ext4_block_truncate_page(handle_t *handle,
3858 : struct address_space *mapping, loff_t from)
3859 : {
3860 2 : unsigned offset = from & (PAGE_SIZE-1);
3861 2 : unsigned length;
3862 2 : unsigned blocksize;
3863 2 : struct inode *inode = mapping->host;
3864 :
3865 : /* If we are processing an encrypted inode during orphan list handling */
3866 2 : if (IS_ENCRYPTED(inode) && !fscrypt_has_encryption_key(inode))
3867 : return 0;
3868 :
3869 2 : blocksize = inode->i_sb->s_blocksize;
3870 2 : length = blocksize - (offset & (blocksize - 1));
3871 :
3872 2 : return ext4_block_zero_page_range(handle, mapping, from, length);
3873 : }
3874 :
3875 0 : int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
3876 : loff_t lstart, loff_t length)
3877 : {
3878 0 : struct super_block *sb = inode->i_sb;
3879 0 : struct address_space *mapping = inode->i_mapping;
3880 0 : unsigned partial_start, partial_end;
3881 0 : ext4_fsblk_t start, end;
3882 0 : loff_t byte_end = (lstart + length - 1);
3883 0 : int err = 0;
3884 :
3885 0 : partial_start = lstart & (sb->s_blocksize - 1);
3886 0 : partial_end = byte_end & (sb->s_blocksize - 1);
3887 :
3888 0 : start = lstart >> sb->s_blocksize_bits;
3889 0 : end = byte_end >> sb->s_blocksize_bits;
3890 :
3891 : /* Handle partial zero within the single block */
3892 0 : if (start == end &&
3893 0 : (partial_start || (partial_end != sb->s_blocksize - 1))) {
3894 0 : err = ext4_block_zero_page_range(handle, mapping,
3895 : lstart, length);
3896 0 : return err;
3897 : }
3898 : /* Handle partial zero out on the start of the range */
3899 0 : if (partial_start) {
3900 0 : err = ext4_block_zero_page_range(handle, mapping,
3901 : lstart, sb->s_blocksize);
3902 0 : if (err)
3903 : return err;
3904 : }
3905 : /* Handle partial zero out on the end of the range */
3906 0 : if (partial_end != sb->s_blocksize - 1)
3907 0 : err = ext4_block_zero_page_range(handle, mapping,
3908 : byte_end - partial_end,
3909 0 : partial_end + 1);
3910 : return err;
3911 : }
3912 :
3913 140 : int ext4_can_truncate(struct inode *inode)
3914 : {
3915 140 : if (S_ISREG(inode->i_mode))
3916 : return 1;
3917 71 : if (S_ISDIR(inode->i_mode))
3918 : return 1;
3919 0 : if (S_ISLNK(inode->i_mode))
3920 0 : return !ext4_inode_is_fast_symlink(inode);
3921 : return 0;
3922 : }
3923 :
3924 : /*
3925 : * We have to make sure i_disksize gets properly updated before we truncate
3926 : * page cache due to hole punching or zero range. Otherwise i_disksize update
3927 : * can get lost as it may have been postponed to submission of writeback but
3928 : * that will never happen after we truncate page cache.
3929 : */
3930 0 : int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset,
3931 : loff_t len)
3932 : {
3933 0 : handle_t *handle;
3934 0 : int ret;
3935 :
3936 0 : loff_t size = i_size_read(inode);
3937 :
3938 0 : WARN_ON(!inode_is_locked(inode));
3939 0 : if (offset > size || offset + len < size)
3940 : return 0;
3941 :
3942 0 : if (EXT4_I(inode)->i_disksize >= size)
3943 : return 0;
3944 :
3945 0 : handle = ext4_journal_start(inode, EXT4_HT_MISC, 1);
3946 0 : if (IS_ERR(handle))
3947 0 : return PTR_ERR(handle);
3948 0 : ext4_update_i_disksize(inode, size);
3949 0 : ret = ext4_mark_inode_dirty(handle, inode);
3950 0 : ext4_journal_stop(handle);
3951 :
3952 0 : return ret;
3953 : }
3954 :
3955 : static void ext4_wait_dax_page(struct ext4_inode_info *ei)
3956 : {
3957 : up_write(&ei->i_mmap_sem);
3958 : schedule();
3959 : down_write(&ei->i_mmap_sem);
3960 : }
3961 :
3962 5 : int ext4_break_layouts(struct inode *inode)
3963 : {
3964 5 : struct ext4_inode_info *ei = EXT4_I(inode);
3965 5 : struct page *page;
3966 5 : int error;
3967 :
3968 5 : if (WARN_ON_ONCE(!rwsem_is_locked(&ei->i_mmap_sem)))
3969 : return -EINVAL;
3970 :
3971 5 : do {
3972 5 : page = dax_layout_busy_page(inode->i_mapping);
3973 5 : if (!page)
3974 5 : return 0;
3975 :
3976 : error = ___wait_var_event(&page->_refcount,
3977 : atomic_read(&page->_refcount) == 1,
3978 : TASK_INTERRUPTIBLE, 0, 0,
3979 : ext4_wait_dax_page(ei));
3980 : } while (error == 0);
3981 :
3982 : return error;
3983 : }
3984 :
3985 : /*
3986 : * ext4_punch_hole: punches a hole in a file by releasing the blocks
3987 : * associated with the given offset and length
3988 : *
3989 : * @inode: File inode
3990 : * @offset: The offset where the hole will begin
3991 : * @len: The length of the hole
3992 : *
3993 : * Returns: 0 on success or negative on failure
3994 : */
3995 :
3996 0 : int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length)
3997 : {
3998 0 : struct super_block *sb = inode->i_sb;
3999 0 : ext4_lblk_t first_block, stop_block;
4000 0 : struct address_space *mapping = inode->i_mapping;
4001 0 : loff_t first_block_offset, last_block_offset;
4002 0 : handle_t *handle;
4003 0 : unsigned int credits;
4004 0 : int ret = 0, ret2 = 0;
4005 :
4006 0 : trace_ext4_punch_hole(inode, offset, length, 0);
4007 :
4008 0 : ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
4009 0 : if (ext4_has_inline_data(inode)) {
4010 0 : down_write(&EXT4_I(inode)->i_mmap_sem);
4011 0 : ret = ext4_convert_inline_data(inode);
4012 0 : up_write(&EXT4_I(inode)->i_mmap_sem);
4013 0 : if (ret)
4014 : return ret;
4015 : }
4016 :
4017 : /*
4018 : * Write out all dirty pages to avoid race conditions
4019 : * Then release them.
4020 : */
4021 0 : if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
4022 0 : ret = filemap_write_and_wait_range(mapping, offset,
4023 0 : offset + length - 1);
4024 0 : if (ret)
4025 : return ret;
4026 : }
4027 :
4028 0 : inode_lock(inode);
4029 :
4030 : /* No need to punch hole beyond i_size */
4031 0 : if (offset >= inode->i_size)
4032 0 : goto out_mutex;
4033 :
4034 : /*
4035 : * If the hole extends beyond i_size, set the hole
4036 : * to end after the page that contains i_size
4037 : */
4038 0 : if (offset + length > inode->i_size) {
4039 0 : length = inode->i_size +
4040 0 : PAGE_SIZE - (inode->i_size & (PAGE_SIZE - 1)) -
4041 : offset;
4042 : }
4043 :
4044 0 : if (offset & (sb->s_blocksize - 1) ||
4045 0 : (offset + length) & (sb->s_blocksize - 1)) {
4046 : /*
4047 : * Attach jinode to inode for jbd2 if we do any zeroing of
4048 : * partial block
4049 : */
4050 0 : ret = ext4_inode_attach_jinode(inode);
4051 0 : if (ret < 0)
4052 0 : goto out_mutex;
4053 :
4054 : }
4055 :
4056 : /* Wait all existing dio workers, newcomers will block on i_mutex */
4057 0 : inode_dio_wait(inode);
4058 :
4059 : /*
4060 : * Prevent page faults from reinstantiating pages we have released from
4061 : * page cache.
4062 : */
4063 0 : down_write(&EXT4_I(inode)->i_mmap_sem);
4064 :
4065 0 : ret = ext4_break_layouts(inode);
4066 0 : if (ret)
4067 0 : goto out_dio;
4068 :
4069 0 : first_block_offset = round_up(offset, sb->s_blocksize);
4070 0 : last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
4071 :
4072 : /* Now release the pages and zero block aligned part of pages*/
4073 0 : if (last_block_offset > first_block_offset) {
4074 0 : ret = ext4_update_disksize_before_punch(inode, offset, length);
4075 0 : if (ret)
4076 0 : goto out_dio;
4077 0 : truncate_pagecache_range(inode, first_block_offset,
4078 : last_block_offset);
4079 : }
4080 :
4081 0 : if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4082 0 : credits = ext4_writepage_trans_blocks(inode);
4083 : else
4084 0 : credits = ext4_blocks_for_truncate(inode);
4085 0 : handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4086 0 : if (IS_ERR(handle)) {
4087 0 : ret = PTR_ERR(handle);
4088 0 : ext4_std_error(sb, ret);
4089 0 : goto out_dio;
4090 : }
4091 :
4092 0 : ret = ext4_zero_partial_blocks(handle, inode, offset,
4093 : length);
4094 0 : if (ret)
4095 0 : goto out_stop;
4096 :
4097 0 : first_block = (offset + sb->s_blocksize - 1) >>
4098 0 : EXT4_BLOCK_SIZE_BITS(sb);
4099 0 : stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
4100 :
4101 : /* If there are blocks to remove, do it */
4102 0 : if (stop_block > first_block) {
4103 :
4104 0 : down_write(&EXT4_I(inode)->i_data_sem);
4105 0 : ext4_discard_preallocations(inode, 0);
4106 :
4107 0 : ret = ext4_es_remove_extent(inode, first_block,
4108 : stop_block - first_block);
4109 0 : if (ret) {
4110 0 : up_write(&EXT4_I(inode)->i_data_sem);
4111 0 : goto out_stop;
4112 : }
4113 :
4114 0 : if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4115 0 : ret = ext4_ext_remove_space(inode, first_block,
4116 : stop_block - 1);
4117 : else
4118 0 : ret = ext4_ind_remove_space(handle, inode, first_block,
4119 : stop_block);
4120 :
4121 0 : up_write(&EXT4_I(inode)->i_data_sem);
4122 : }
4123 0 : ext4_fc_track_range(handle, inode, first_block, stop_block);
4124 0 : if (IS_SYNC(inode))
4125 0 : ext4_handle_sync(handle);
4126 :
4127 0 : inode->i_mtime = inode->i_ctime = current_time(inode);
4128 0 : ret2 = ext4_mark_inode_dirty(handle, inode);
4129 0 : if (unlikely(ret2))
4130 0 : ret = ret2;
4131 0 : if (ret >= 0)
4132 0 : ext4_update_inode_fsync_trans(handle, inode, 1);
4133 0 : out_stop:
4134 0 : ext4_journal_stop(handle);
4135 0 : out_dio:
4136 0 : up_write(&EXT4_I(inode)->i_mmap_sem);
4137 0 : out_mutex:
4138 0 : inode_unlock(inode);
4139 0 : return ret;
4140 : }
4141 :
4142 536 : int ext4_inode_attach_jinode(struct inode *inode)
4143 : {
4144 536 : struct ext4_inode_info *ei = EXT4_I(inode);
4145 536 : struct jbd2_inode *jinode;
4146 :
4147 536 : if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
4148 : return 0;
4149 :
4150 515 : jinode = jbd2_alloc_inode(GFP_KERNEL);
4151 515 : spin_lock(&inode->i_lock);
4152 515 : if (!ei->jinode) {
4153 515 : if (!jinode) {
4154 0 : spin_unlock(&inode->i_lock);
4155 0 : return -ENOMEM;
4156 : }
4157 515 : ei->jinode = jinode;
4158 515 : jbd2_journal_init_jbd_inode(ei->jinode, inode);
4159 515 : jinode = NULL;
4160 : }
4161 515 : spin_unlock(&inode->i_lock);
4162 515 : if (unlikely(jinode != NULL))
4163 0 : jbd2_free_inode(jinode);
4164 : return 0;
4165 : }
4166 :
4167 : /*
4168 : * ext4_truncate()
4169 : *
4170 : * We block out ext4_get_block() block instantiations across the entire
4171 : * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4172 : * simultaneously on behalf of the same inode.
4173 : *
4174 : * As we work through the truncate and commit bits of it to the journal there
4175 : * is one core, guiding principle: the file's tree must always be consistent on
4176 : * disk. We must be able to restart the truncate after a crash.
4177 : *
4178 : * The file's tree may be transiently inconsistent in memory (although it
4179 : * probably isn't), but whenever we close off and commit a journal transaction,
4180 : * the contents of (the filesystem + the journal) must be consistent and
4181 : * restartable. It's pretty simple, really: bottom up, right to left (although
4182 : * left-to-right works OK too).
4183 : *
4184 : * Note that at recovery time, journal replay occurs *before* the restart of
4185 : * truncate against the orphan inode list.
4186 : *
4187 : * The committed inode has the new, desired i_size (which is the same as
4188 : * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4189 : * that this inode's truncate did not complete and it will again call
4190 : * ext4_truncate() to have another go. So there will be instantiated blocks
4191 : * to the right of the truncation point in a crashed ext4 filesystem. But
4192 : * that's fine - as long as they are linked from the inode, the post-crash
4193 : * ext4_truncate() run will find them and release them.
4194 : */
4195 140 : int ext4_truncate(struct inode *inode)
4196 : {
4197 140 : struct ext4_inode_info *ei = EXT4_I(inode);
4198 140 : unsigned int credits;
4199 140 : int err = 0, err2;
4200 140 : handle_t *handle;
4201 140 : struct address_space *mapping = inode->i_mapping;
4202 :
4203 : /*
4204 : * There is a possibility that we're either freeing the inode
4205 : * or it's a completely new inode. In those cases we might not
4206 : * have i_mutex locked because it's not necessary.
4207 : */
4208 140 : if (!(inode->i_state & (I_NEW|I_FREEING)))
4209 5 : WARN_ON(!inode_is_locked(inode));
4210 140 : trace_ext4_truncate_enter(inode);
4211 :
4212 140 : if (!ext4_can_truncate(inode))
4213 0 : goto out_trace;
4214 :
4215 140 : if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4216 138 : ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
4217 :
4218 140 : if (ext4_has_inline_data(inode)) {
4219 0 : int has_inline = 1;
4220 :
4221 0 : err = ext4_inline_data_truncate(inode, &has_inline);
4222 0 : if (err || has_inline)
4223 0 : goto out_trace;
4224 : }
4225 :
4226 : /* If we zero-out tail of the page, we have to create jinode for jbd2 */
4227 140 : if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
4228 2 : if (ext4_inode_attach_jinode(inode) < 0)
4229 0 : goto out_trace;
4230 : }
4231 :
4232 140 : if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4233 140 : credits = ext4_writepage_trans_blocks(inode);
4234 : else
4235 0 : credits = ext4_blocks_for_truncate(inode);
4236 :
4237 140 : handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4238 140 : if (IS_ERR(handle)) {
4239 0 : err = PTR_ERR(handle);
4240 0 : goto out_trace;
4241 : }
4242 :
4243 140 : if (inode->i_size & (inode->i_sb->s_blocksize - 1))
4244 2 : ext4_block_truncate_page(handle, mapping, inode->i_size);
4245 :
4246 : /*
4247 : * We add the inode to the orphan list, so that if this
4248 : * truncate spans multiple transactions, and we crash, we will
4249 : * resume the truncate when the filesystem recovers. It also
4250 : * marks the inode dirty, to catch the new size.
4251 : *
4252 : * Implication: the file must always be in a sane, consistent
4253 : * truncatable state while each transaction commits.
4254 : */
4255 140 : err = ext4_orphan_add(handle, inode);
4256 140 : if (err)
4257 0 : goto out_stop;
4258 :
4259 140 : down_write(&EXT4_I(inode)->i_data_sem);
4260 :
4261 140 : ext4_discard_preallocations(inode, 0);
4262 :
4263 140 : if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4264 140 : err = ext4_ext_truncate(handle, inode);
4265 : else
4266 0 : ext4_ind_truncate(handle, inode);
4267 :
4268 140 : up_write(&ei->i_data_sem);
4269 140 : if (err)
4270 0 : goto out_stop;
4271 :
4272 140 : if (IS_SYNC(inode))
4273 0 : ext4_handle_sync(handle);
4274 :
4275 140 : out_stop:
4276 : /*
4277 : * If this was a simple ftruncate() and the file will remain alive,
4278 : * then we need to clear up the orphan record which we created above.
4279 : * However, if this was a real unlink then we were called by
4280 : * ext4_evict_inode(), and we allow that function to clean up the
4281 : * orphan info for us.
4282 : */
4283 140 : if (inode->i_nlink)
4284 5 : ext4_orphan_del(handle, inode);
4285 :
4286 140 : inode->i_mtime = inode->i_ctime = current_time(inode);
4287 140 : err2 = ext4_mark_inode_dirty(handle, inode);
4288 140 : if (unlikely(err2 && !err))
4289 0 : err = err2;
4290 140 : ext4_journal_stop(handle);
4291 :
4292 140 : out_trace:
4293 140 : trace_ext4_truncate_exit(inode);
4294 140 : return err;
4295 : }
4296 :
4297 : /*
4298 : * ext4_get_inode_loc returns with an extra refcount against the inode's
4299 : * underlying buffer_head on success. If 'in_mem' is true, we have all
4300 : * data in memory that is needed to recreate the on-disk version of this
4301 : * inode.
4302 : */
4303 15367 : static int __ext4_get_inode_loc(struct super_block *sb, unsigned long ino,
4304 : struct ext4_iloc *iloc, int in_mem,
4305 : ext4_fsblk_t *ret_block)
4306 : {
4307 15367 : struct ext4_group_desc *gdp;
4308 15367 : struct buffer_head *bh;
4309 15367 : ext4_fsblk_t block;
4310 15367 : struct blk_plug plug;
4311 15367 : int inodes_per_block, inode_offset;
4312 :
4313 15367 : iloc->bh = NULL;
4314 15367 : if (ino < EXT4_ROOT_INO ||
4315 15367 : ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
4316 : return -EFSCORRUPTED;
4317 :
4318 15367 : iloc->block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb);
4319 15367 : gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4320 15367 : if (!gdp)
4321 : return -EIO;
4322 :
4323 : /*
4324 : * Figure out the offset within the block group inode table
4325 : */
4326 15367 : inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4327 15367 : inode_offset = ((ino - 1) %
4328 15367 : EXT4_INODES_PER_GROUP(sb));
4329 15367 : block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
4330 15367 : iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4331 :
4332 15367 : bh = sb_getblk(sb, block);
4333 15368 : if (unlikely(!bh))
4334 : return -ENOMEM;
4335 15368 : if (ext4_simulate_fail(sb, EXT4_SIM_INODE_EIO))
4336 : goto simulate_eio;
4337 15368 : if (!buffer_uptodate(bh)) {
4338 103 : lock_buffer(bh);
4339 :
4340 103 : if (ext4_buffer_uptodate(bh)) {
4341 : /* someone brought it uptodate while we waited */
4342 0 : unlock_buffer(bh);
4343 0 : goto has_buffer;
4344 : }
4345 :
4346 : /*
4347 : * If we have all information of the inode in memory and this
4348 : * is the only valid inode in the block, we need not read the
4349 : * block.
4350 : */
4351 103 : if (in_mem) {
4352 1 : struct buffer_head *bitmap_bh;
4353 1 : int i, start;
4354 :
4355 1 : start = inode_offset & ~(inodes_per_block - 1);
4356 :
4357 : /* Is the inode bitmap in cache? */
4358 1 : bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4359 1 : if (unlikely(!bitmap_bh))
4360 0 : goto make_io;
4361 :
4362 : /*
4363 : * If the inode bitmap isn't in cache then the
4364 : * optimisation may end up performing two reads instead
4365 : * of one, so skip it.
4366 : */
4367 1 : if (!buffer_uptodate(bitmap_bh)) {
4368 0 : brelse(bitmap_bh);
4369 0 : goto make_io;
4370 : }
4371 17 : for (i = start; i < start + inodes_per_block; i++) {
4372 16 : if (i == inode_offset)
4373 1 : continue;
4374 15 : if (ext4_test_bit(i, bitmap_bh->b_data))
4375 : break;
4376 : }
4377 1 : brelse(bitmap_bh);
4378 1 : if (i == start + inodes_per_block) {
4379 : /* all other inodes are free, so skip I/O */
4380 1 : memset(bh->b_data, 0, bh->b_size);
4381 1 : set_buffer_uptodate(bh);
4382 1 : unlock_buffer(bh);
4383 1 : goto has_buffer;
4384 : }
4385 : }
4386 :
4387 102 : make_io:
4388 : /*
4389 : * If we need to do any I/O, try to pre-readahead extra
4390 : * blocks from the inode table.
4391 : */
4392 102 : blk_start_plug(&plug);
4393 102 : if (EXT4_SB(sb)->s_inode_readahead_blks) {
4394 102 : ext4_fsblk_t b, end, table;
4395 102 : unsigned num;
4396 102 : __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
4397 :
4398 102 : table = ext4_inode_table(sb, gdp);
4399 : /* s_inode_readahead_blks is always a power of 2 */
4400 102 : b = block & ~((ext4_fsblk_t) ra_blks - 1);
4401 102 : if (table > b)
4402 : b = table;
4403 102 : end = b + ra_blks;
4404 102 : num = EXT4_INODES_PER_GROUP(sb);
4405 102 : if (ext4_has_group_desc_csum(sb))
4406 102 : num -= ext4_itable_unused_count(sb, gdp);
4407 102 : table += num / inodes_per_block;
4408 102 : if (end > table)
4409 : end = table;
4410 3468 : while (b <= end)
4411 3366 : ext4_sb_breadahead_unmovable(sb, b++);
4412 : }
4413 :
4414 : /*
4415 : * There are other valid inodes in the buffer, this inode
4416 : * has in-inode xattrs, or we don't have this inode in memory.
4417 : * Read the block from disk.
4418 : */
4419 102 : trace_ext4_load_inode(sb, ino);
4420 102 : ext4_read_bh_nowait(bh, REQ_META | REQ_PRIO, NULL);
4421 102 : blk_finish_plug(&plug);
4422 102 : wait_on_buffer(bh);
4423 102 : if (!buffer_uptodate(bh)) {
4424 0 : simulate_eio:
4425 0 : if (ret_block)
4426 0 : *ret_block = block;
4427 0 : brelse(bh);
4428 0 : return -EIO;
4429 : }
4430 : }
4431 15367 : has_buffer:
4432 15368 : iloc->bh = bh;
4433 15368 : return 0;
4434 : }
4435 :
4436 4935 : static int __ext4_get_inode_loc_noinmem(struct inode *inode,
4437 : struct ext4_iloc *iloc)
4438 : {
4439 4935 : ext4_fsblk_t err_blk;
4440 4935 : int ret;
4441 :
4442 4935 : ret = __ext4_get_inode_loc(inode->i_sb, inode->i_ino, iloc, 0,
4443 : &err_blk);
4444 :
4445 4935 : if (ret == -EIO)
4446 0 : ext4_error_inode_block(inode, err_blk, EIO,
4447 : "unable to read itable block");
4448 :
4449 4935 : return ret;
4450 : }
4451 :
4452 10432 : int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4453 : {
4454 10432 : ext4_fsblk_t err_blk;
4455 10432 : int ret;
4456 :
4457 : /* We have all inode data except xattrs in memory here. */
4458 10432 : ret = __ext4_get_inode_loc(inode->i_sb, inode->i_ino, iloc,
4459 10432 : !ext4_test_inode_state(inode, EXT4_STATE_XATTR), &err_blk);
4460 :
4461 10433 : if (ret == -EIO)
4462 0 : ext4_error_inode_block(inode, err_blk, EIO,
4463 : "unable to read itable block");
4464 :
4465 10433 : return ret;
4466 : }
4467 :
4468 :
4469 0 : int ext4_get_fc_inode_loc(struct super_block *sb, unsigned long ino,
4470 : struct ext4_iloc *iloc)
4471 : {
4472 0 : return __ext4_get_inode_loc(sb, ino, iloc, 0, NULL);
4473 : }
4474 :
4475 5566 : static bool ext4_should_enable_dax(struct inode *inode)
4476 : {
4477 5566 : struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4478 :
4479 5566 : if (test_opt2(inode->i_sb, DAX_NEVER))
4480 : return false;
4481 5566 : if (!S_ISREG(inode->i_mode))
4482 : return false;
4483 4123 : if (ext4_should_journal_data(inode))
4484 : return false;
4485 4123 : if (ext4_has_inline_data(inode))
4486 : return false;
4487 4123 : if (ext4_test_inode_flag(inode, EXT4_INODE_ENCRYPT))
4488 : return false;
4489 4123 : if (ext4_test_inode_flag(inode, EXT4_INODE_VERITY))
4490 : return false;
4491 4123 : if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags))
4492 : return false;
4493 0 : if (test_opt(inode->i_sb, DAX_ALWAYS))
4494 : return true;
4495 :
4496 0 : return ext4_test_inode_flag(inode, EXT4_INODE_DAX);
4497 : }
4498 :
4499 5566 : void ext4_set_inode_flags(struct inode *inode, bool init)
4500 : {
4501 5566 : unsigned int flags = EXT4_I(inode)->i_flags;
4502 5566 : unsigned int new_fl = 0;
4503 :
4504 5566 : WARN_ON_ONCE(IS_DAX(inode) && init);
4505 :
4506 5566 : if (flags & EXT4_SYNC_FL)
4507 0 : new_fl |= S_SYNC;
4508 5566 : if (flags & EXT4_APPEND_FL)
4509 0 : new_fl |= S_APPEND;
4510 5566 : if (flags & EXT4_IMMUTABLE_FL)
4511 0 : new_fl |= S_IMMUTABLE;
4512 5566 : if (flags & EXT4_NOATIME_FL)
4513 0 : new_fl |= S_NOATIME;
4514 5566 : if (flags & EXT4_DIRSYNC_FL)
4515 0 : new_fl |= S_DIRSYNC;
4516 :
4517 : /* Because of the way inode_set_flags() works we must preserve S_DAX
4518 : * here if already set. */
4519 5566 : new_fl |= (inode->i_flags & S_DAX);
4520 5566 : if (init && ext4_should_enable_dax(inode))
4521 : new_fl |= S_DAX;
4522 :
4523 5566 : if (flags & EXT4_ENCRYPT_FL)
4524 0 : new_fl |= S_ENCRYPTED;
4525 5566 : if (flags & EXT4_CASEFOLD_FL)
4526 0 : new_fl |= S_CASEFOLD;
4527 5566 : if (flags & EXT4_VERITY_FL)
4528 0 : new_fl |= S_VERITY;
4529 5566 : inode_set_flags(inode, new_fl,
4530 : S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX|
4531 : S_ENCRYPTED|S_CASEFOLD|S_VERITY);
4532 5566 : }
4533 :
4534 4935 : static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4535 : struct ext4_inode_info *ei)
4536 : {
4537 4935 : blkcnt_t i_blocks ;
4538 4935 : struct inode *inode = &(ei->vfs_inode);
4539 4935 : struct super_block *sb = inode->i_sb;
4540 :
4541 4935 : if (ext4_has_feature_huge_file(sb)) {
4542 : /* we are using combined 48 bit field */
4543 4935 : i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4544 4935 : le32_to_cpu(raw_inode->i_blocks_lo);
4545 4935 : if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4546 : /* i_blocks represent file system block size */
4547 0 : return i_blocks << (inode->i_blkbits - 9);
4548 : } else {
4549 : return i_blocks;
4550 : }
4551 : } else {
4552 0 : return le32_to_cpu(raw_inode->i_blocks_lo);
4553 : }
4554 : }
4555 :
4556 4935 : static inline int ext4_iget_extra_inode(struct inode *inode,
4557 : struct ext4_inode *raw_inode,
4558 : struct ext4_inode_info *ei)
4559 : {
4560 4935 : __le32 *magic = (void *)raw_inode +
4561 4935 : EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4562 :
4563 4935 : if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize + sizeof(__le32) <=
4564 4935 : EXT4_INODE_SIZE(inode->i_sb) &&
4565 4935 : *magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4566 0 : ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4567 0 : return ext4_find_inline_data_nolock(inode);
4568 : } else
4569 4935 : EXT4_I(inode)->i_inline_off = 0;
4570 4935 : return 0;
4571 : }
4572 :
4573 0 : int ext4_get_projid(struct inode *inode, kprojid_t *projid)
4574 : {
4575 0 : if (!ext4_has_feature_project(inode->i_sb))
4576 : return -EOPNOTSUPP;
4577 0 : *projid = EXT4_I(inode)->i_projid;
4578 0 : return 0;
4579 : }
4580 :
4581 : /*
4582 : * ext4 has self-managed i_version for ea inodes, it stores the lower 32bit of
4583 : * refcount in i_version, so use raw values if inode has EXT4_EA_INODE_FL flag
4584 : * set.
4585 : */
4586 4935 : static inline void ext4_inode_set_iversion_queried(struct inode *inode, u64 val)
4587 : {
4588 4935 : if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4589 0 : inode_set_iversion_raw(inode, val);
4590 : else
4591 4935 : inode_set_iversion_queried(inode, val);
4592 4935 : }
4593 10433 : static inline u64 ext4_inode_peek_iversion(const struct inode *inode)
4594 : {
4595 10433 : if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4596 0 : return inode_peek_iversion_raw(inode);
4597 : else
4598 10433 : return inode_peek_iversion(inode);
4599 : }
4600 :
4601 5020 : struct inode *__ext4_iget(struct super_block *sb, unsigned long ino,
4602 : ext4_iget_flags flags, const char *function,
4603 : unsigned int line)
4604 : {
4605 5020 : struct ext4_iloc iloc;
4606 5020 : struct ext4_inode *raw_inode;
4607 5020 : struct ext4_inode_info *ei;
4608 5020 : struct inode *inode;
4609 5020 : journal_t *journal = EXT4_SB(sb)->s_journal;
4610 5020 : long ret;
4611 5020 : loff_t size;
4612 5020 : int block;
4613 5020 : uid_t i_uid;
4614 5020 : gid_t i_gid;
4615 5020 : projid_t i_projid;
4616 :
4617 5020 : if ((!(flags & EXT4_IGET_SPECIAL) &&
4618 5020 : (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)) ||
4619 5020 : (ino < EXT4_ROOT_INO) ||
4620 5020 : (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))) {
4621 0 : if (flags & EXT4_IGET_HANDLE)
4622 5020 : return ERR_PTR(-ESTALE);
4623 0 : __ext4_error(sb, function, line, false, EFSCORRUPTED, 0,
4624 : "inode #%lu: comm %s: iget: illegal inode #",
4625 0 : ino, current->comm);
4626 0 : return ERR_PTR(-EFSCORRUPTED);
4627 : }
4628 :
4629 5020 : inode = iget_locked(sb, ino);
4630 5020 : if (!inode)
4631 5020 : return ERR_PTR(-ENOMEM);
4632 5020 : if (!(inode->i_state & I_NEW))
4633 : return inode;
4634 :
4635 4935 : ei = EXT4_I(inode);
4636 4935 : iloc.bh = NULL;
4637 :
4638 4935 : ret = __ext4_get_inode_loc_noinmem(inode, &iloc);
4639 4935 : if (ret < 0)
4640 0 : goto bad_inode;
4641 4935 : raw_inode = ext4_raw_inode(&iloc);
4642 :
4643 4935 : if ((ino == EXT4_ROOT_INO) && (raw_inode->i_links_count == 0)) {
4644 0 : ext4_error_inode(inode, function, line, 0,
4645 : "iget: root inode unallocated");
4646 0 : ret = -EFSCORRUPTED;
4647 0 : goto bad_inode;
4648 : }
4649 :
4650 4935 : if ((flags & EXT4_IGET_HANDLE) &&
4651 0 : (raw_inode->i_links_count == 0) && (raw_inode->i_mode == 0)) {
4652 0 : ret = -ESTALE;
4653 0 : goto bad_inode;
4654 : }
4655 :
4656 4935 : if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4657 4935 : ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4658 4935 : if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4659 4935 : EXT4_INODE_SIZE(inode->i_sb) ||
4660 : (ei->i_extra_isize & 3)) {
4661 0 : ext4_error_inode(inode, function, line, 0,
4662 : "iget: bad extra_isize %u "
4663 : "(inode size %u)",
4664 : ei->i_extra_isize,
4665 : EXT4_INODE_SIZE(inode->i_sb));
4666 0 : ret = -EFSCORRUPTED;
4667 0 : goto bad_inode;
4668 : }
4669 : } else
4670 0 : ei->i_extra_isize = 0;
4671 :
4672 : /* Precompute checksum seed for inode metadata */
4673 4935 : if (ext4_has_metadata_csum(sb)) {
4674 0 : struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4675 0 : __u32 csum;
4676 0 : __le32 inum = cpu_to_le32(inode->i_ino);
4677 0 : __le32 gen = raw_inode->i_generation;
4678 0 : csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4679 : sizeof(inum));
4680 0 : ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4681 : sizeof(gen));
4682 : }
4683 :
4684 4935 : if ((!ext4_inode_csum_verify(inode, raw_inode, ei) ||
4685 4935 : ext4_simulate_fail(sb, EXT4_SIM_INODE_CRC)) &&
4686 0 : (!(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY))) {
4687 0 : ext4_error_inode_err(inode, function, line, 0,
4688 : EFSBADCRC, "iget: checksum invalid");
4689 0 : ret = -EFSBADCRC;
4690 0 : goto bad_inode;
4691 : }
4692 :
4693 4935 : inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4694 4935 : i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4695 4935 : i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4696 4935 : if (ext4_has_feature_project(sb) &&
4697 0 : EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4698 0 : EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4699 0 : i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid);
4700 : else
4701 : i_projid = EXT4_DEF_PROJID;
4702 :
4703 4935 : if (!(test_opt(inode->i_sb, NO_UID32))) {
4704 4935 : i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4705 4935 : i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4706 : }
4707 4935 : i_uid_write(inode, i_uid);
4708 4935 : i_gid_write(inode, i_gid);
4709 4935 : ei->i_projid = make_kprojid(&init_user_ns, i_projid);
4710 4935 : set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4711 :
4712 4935 : ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4713 4935 : ei->i_inline_off = 0;
4714 4935 : ei->i_dir_start_lookup = 0;
4715 4935 : ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4716 : /* We now have enough fields to check if the inode was active or not.
4717 : * This is needed because nfsd might try to access dead inodes
4718 : * the test is that same one that e2fsck uses
4719 : * NeilBrown 1999oct15
4720 : */
4721 4935 : if (inode->i_nlink == 0) {
4722 0 : if ((inode->i_mode == 0 ||
4723 0 : !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4724 : ino != EXT4_BOOT_LOADER_INO) {
4725 : /* this inode is deleted */
4726 0 : ret = -ESTALE;
4727 0 : goto bad_inode;
4728 : }
4729 : /* The only unlinked inodes we let through here have
4730 : * valid i_mode and are being read by the orphan
4731 : * recovery code: that's fine, we're about to complete
4732 : * the process of deleting those.
4733 : * OR it is the EXT4_BOOT_LOADER_INO which is
4734 : * not initialized on a new filesystem. */
4735 : }
4736 4935 : ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4737 4935 : ext4_set_inode_flags(inode, true);
4738 4935 : inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4739 4935 : ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4740 4935 : if (ext4_has_feature_64bit(sb))
4741 0 : ei->i_file_acl |=
4742 0 : ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4743 4935 : inode->i_size = ext4_isize(sb, raw_inode);
4744 4935 : if ((size = i_size_read(inode)) < 0) {
4745 0 : ext4_error_inode(inode, function, line, 0,
4746 : "iget: bad i_size value: %lld", size);
4747 0 : ret = -EFSCORRUPTED;
4748 0 : goto bad_inode;
4749 : }
4750 : /*
4751 : * If dir_index is not enabled but there's dir with INDEX flag set,
4752 : * we'd normally treat htree data as empty space. But with metadata
4753 : * checksumming that corrupts checksums so forbid that.
4754 : */
4755 4935 : if (!ext4_has_feature_dir_index(sb) && ext4_has_metadata_csum(sb) &&
4756 0 : ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) {
4757 0 : ext4_error_inode(inode, function, line, 0,
4758 : "iget: Dir with htree data on filesystem without dir_index feature.");
4759 0 : ret = -EFSCORRUPTED;
4760 0 : goto bad_inode;
4761 : }
4762 4935 : ei->i_disksize = inode->i_size;
4763 : #ifdef CONFIG_QUOTA
4764 : ei->i_reserved_quota = 0;
4765 : #endif
4766 4935 : inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4767 4935 : ei->i_block_group = iloc.block_group;
4768 4935 : ei->i_last_alloc_group = ~0;
4769 : /*
4770 : * NOTE! The in-memory inode i_data array is in little-endian order
4771 : * even on big-endian machines: we do NOT byteswap the block numbers!
4772 : */
4773 78960 : for (block = 0; block < EXT4_N_BLOCKS; block++)
4774 74025 : ei->i_data[block] = raw_inode->i_block[block];
4775 4935 : INIT_LIST_HEAD(&ei->i_orphan);
4776 4935 : ext4_fc_init_inode(&ei->vfs_inode);
4777 :
4778 : /*
4779 : * Set transaction id's of transactions that have to be committed
4780 : * to finish f[data]sync. We set them to currently running transaction
4781 : * as we cannot be sure that the inode or some of its metadata isn't
4782 : * part of the transaction - the inode could have been reclaimed and
4783 : * now it is reread from disk.
4784 : */
4785 4935 : if (journal) {
4786 4934 : transaction_t *transaction;
4787 4934 : tid_t tid;
4788 :
4789 4934 : read_lock(&journal->j_state_lock);
4790 4934 : if (journal->j_running_transaction)
4791 : transaction = journal->j_running_transaction;
4792 : else
4793 521 : transaction = journal->j_committing_transaction;
4794 4934 : if (transaction)
4795 4449 : tid = transaction->t_tid;
4796 : else
4797 485 : tid = journal->j_commit_sequence;
4798 4934 : read_unlock(&journal->j_state_lock);
4799 4934 : ei->i_sync_tid = tid;
4800 4934 : ei->i_datasync_tid = tid;
4801 : }
4802 :
4803 4935 : if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4804 4935 : if (ei->i_extra_isize == 0) {
4805 : /* The extra space is currently unused. Use it. */
4806 0 : BUILD_BUG_ON(sizeof(struct ext4_inode) & 3);
4807 0 : ei->i_extra_isize = sizeof(struct ext4_inode) -
4808 : EXT4_GOOD_OLD_INODE_SIZE;
4809 : } else {
4810 4935 : ret = ext4_iget_extra_inode(inode, raw_inode, ei);
4811 4935 : if (ret)
4812 0 : goto bad_inode;
4813 : }
4814 : }
4815 :
4816 4935 : EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4817 4935 : EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4818 4935 : EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4819 4935 : EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
4820 :
4821 4935 : if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4822 4935 : u64 ivers = le32_to_cpu(raw_inode->i_disk_version);
4823 :
4824 4935 : if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4825 4935 : if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4826 4935 : ivers |=
4827 4935 : (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4828 : }
4829 4935 : ext4_inode_set_iversion_queried(inode, ivers);
4830 : }
4831 :
4832 4935 : ret = 0;
4833 4935 : if (ei->i_file_acl &&
4834 0 : !ext4_inode_block_valid(inode, ei->i_file_acl, 1)) {
4835 0 : ext4_error_inode(inode, function, line, 0,
4836 : "iget: bad extended attribute block %llu",
4837 : ei->i_file_acl);
4838 0 : ret = -EFSCORRUPTED;
4839 0 : goto bad_inode;
4840 4935 : } else if (!ext4_has_inline_data(inode)) {
4841 : /* validate the block references in the inode */
4842 4935 : if (!(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY) &&
4843 4935 : (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4844 828 : (S_ISLNK(inode->i_mode) &&
4845 828 : !ext4_inode_is_fast_symlink(inode)))) {
4846 4107 : if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4847 4107 : ret = ext4_ext_check_inode(inode);
4848 : else
4849 0 : ret = ext4_ind_check_inode(inode);
4850 : }
4851 : }
4852 4107 : if (ret)
4853 0 : goto bad_inode;
4854 :
4855 4935 : if (S_ISREG(inode->i_mode)) {
4856 3665 : inode->i_op = &ext4_file_inode_operations;
4857 3665 : inode->i_fop = &ext4_file_operations;
4858 3665 : ext4_set_aops(inode);
4859 1270 : } else if (S_ISDIR(inode->i_mode)) {
4860 442 : inode->i_op = &ext4_dir_inode_operations;
4861 442 : inode->i_fop = &ext4_dir_operations;
4862 828 : } else if (S_ISLNK(inode->i_mode)) {
4863 : /* VFS does not allow setting these so must be corruption */
4864 828 : if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) {
4865 0 : ext4_error_inode(inode, function, line, 0,
4866 : "iget: immutable or append flags "
4867 : "not allowed on symlinks");
4868 0 : ret = -EFSCORRUPTED;
4869 0 : goto bad_inode;
4870 : }
4871 828 : if (IS_ENCRYPTED(inode)) {
4872 0 : inode->i_op = &ext4_encrypted_symlink_inode_operations;
4873 0 : ext4_set_aops(inode);
4874 828 : } else if (ext4_inode_is_fast_symlink(inode)) {
4875 828 : inode->i_link = (char *)ei->i_data;
4876 828 : inode->i_op = &ext4_fast_symlink_inode_operations;
4877 828 : nd_terminate_link(ei->i_data, inode->i_size,
4878 : sizeof(ei->i_data) - 1);
4879 : } else {
4880 0 : inode->i_op = &ext4_symlink_inode_operations;
4881 0 : ext4_set_aops(inode);
4882 : }
4883 828 : inode_nohighmem(inode);
4884 0 : } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4885 0 : S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4886 0 : inode->i_op = &ext4_special_inode_operations;
4887 0 : if (raw_inode->i_block[0])
4888 0 : init_special_inode(inode, inode->i_mode,
4889 : old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4890 : else
4891 0 : init_special_inode(inode, inode->i_mode,
4892 0 : new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4893 0 : } else if (ino == EXT4_BOOT_LOADER_INO) {
4894 0 : make_bad_inode(inode);
4895 : } else {
4896 0 : ret = -EFSCORRUPTED;
4897 0 : ext4_error_inode(inode, function, line, 0,
4898 : "iget: bogus i_mode (%o)", inode->i_mode);
4899 0 : goto bad_inode;
4900 : }
4901 4935 : if (IS_CASEFOLDED(inode) && !ext4_has_feature_casefold(inode->i_sb))
4902 0 : ext4_error_inode(inode, function, line, 0,
4903 : "casefold flag without casefold feature");
4904 4935 : brelse(iloc.bh);
4905 :
4906 4935 : unlock_new_inode(inode);
4907 4935 : return inode;
4908 :
4909 0 : bad_inode:
4910 0 : brelse(iloc.bh);
4911 0 : iget_failed(inode);
4912 0 : return ERR_PTR(ret);
4913 : }
4914 :
4915 10433 : static int ext4_inode_blocks_set(handle_t *handle,
4916 : struct ext4_inode *raw_inode,
4917 : struct ext4_inode_info *ei)
4918 : {
4919 10433 : struct inode *inode = &(ei->vfs_inode);
4920 10433 : u64 i_blocks = READ_ONCE(inode->i_blocks);
4921 10433 : struct super_block *sb = inode->i_sb;
4922 :
4923 10433 : if (i_blocks <= ~0U) {
4924 : /*
4925 : * i_blocks can be represented in a 32 bit variable
4926 : * as multiple of 512 bytes
4927 : */
4928 10433 : raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4929 10433 : raw_inode->i_blocks_high = 0;
4930 10433 : ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4931 10433 : return 0;
4932 : }
4933 0 : if (!ext4_has_feature_huge_file(sb))
4934 : return -EFBIG;
4935 :
4936 0 : if (i_blocks <= 0xffffffffffffULL) {
4937 : /*
4938 : * i_blocks can be represented in a 48 bit variable
4939 : * as multiple of 512 bytes
4940 : */
4941 0 : raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4942 0 : raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4943 0 : ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4944 : } else {
4945 0 : ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4946 : /* i_block is stored in file system block size */
4947 0 : i_blocks = i_blocks >> (inode->i_blkbits - 9);
4948 0 : raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4949 0 : raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4950 : }
4951 : return 0;
4952 : }
4953 :
4954 0 : static void __ext4_update_other_inode_time(struct super_block *sb,
4955 : unsigned long orig_ino,
4956 : unsigned long ino,
4957 : struct ext4_inode *raw_inode)
4958 : {
4959 0 : struct inode *inode;
4960 :
4961 0 : inode = find_inode_by_ino_rcu(sb, ino);
4962 0 : if (!inode)
4963 : return;
4964 :
4965 0 : if (!inode_is_dirtytime_only(inode))
4966 : return;
4967 :
4968 0 : spin_lock(&inode->i_lock);
4969 0 : if (inode_is_dirtytime_only(inode)) {
4970 0 : struct ext4_inode_info *ei = EXT4_I(inode);
4971 :
4972 0 : inode->i_state &= ~I_DIRTY_TIME;
4973 0 : spin_unlock(&inode->i_lock);
4974 :
4975 0 : spin_lock(&ei->i_raw_lock);
4976 0 : EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4977 0 : EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4978 0 : EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4979 0 : ext4_inode_csum_set(inode, raw_inode, ei);
4980 0 : spin_unlock(&ei->i_raw_lock);
4981 0 : trace_ext4_other_inode_update_time(inode, orig_ino);
4982 0 : return;
4983 : }
4984 0 : spin_unlock(&inode->i_lock);
4985 : }
4986 :
4987 : /*
4988 : * Opportunistically update the other time fields for other inodes in
4989 : * the same inode table block.
4990 : */
4991 0 : static void ext4_update_other_inodes_time(struct super_block *sb,
4992 : unsigned long orig_ino, char *buf)
4993 : {
4994 0 : unsigned long ino;
4995 0 : int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4996 0 : int inode_size = EXT4_INODE_SIZE(sb);
4997 :
4998 : /*
4999 : * Calculate the first inode in the inode table block. Inode
5000 : * numbers are one-based. That is, the first inode in a block
5001 : * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
5002 : */
5003 0 : ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1;
5004 0 : rcu_read_lock();
5005 0 : for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) {
5006 0 : if (ino == orig_ino)
5007 0 : continue;
5008 0 : __ext4_update_other_inode_time(sb, orig_ino, ino,
5009 : (struct ext4_inode *)buf);
5010 : }
5011 0 : rcu_read_unlock();
5012 0 : }
5013 :
5014 : /*
5015 : * Post the struct inode info into an on-disk inode location in the
5016 : * buffer-cache. This gobbles the caller's reference to the
5017 : * buffer_head in the inode location struct.
5018 : *
5019 : * The caller must have write access to iloc->bh.
5020 : */
5021 10433 : static int ext4_do_update_inode(handle_t *handle,
5022 : struct inode *inode,
5023 : struct ext4_iloc *iloc)
5024 : {
5025 10433 : struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
5026 10433 : struct ext4_inode_info *ei = EXT4_I(inode);
5027 10433 : struct buffer_head *bh = iloc->bh;
5028 10433 : struct super_block *sb = inode->i_sb;
5029 10433 : int err = 0, rc, block;
5030 10433 : int need_datasync = 0, set_large_file = 0;
5031 10433 : uid_t i_uid;
5032 10433 : gid_t i_gid;
5033 10433 : projid_t i_projid;
5034 :
5035 10433 : spin_lock(&ei->i_raw_lock);
5036 :
5037 : /* For fields not tracked in the in-memory inode,
5038 : * initialise them to zero for new inodes. */
5039 10433 : if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
5040 631 : memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
5041 :
5042 10433 : err = ext4_inode_blocks_set(handle, raw_inode, ei);
5043 10433 : if (err) {
5044 0 : spin_unlock(&ei->i_raw_lock);
5045 0 : goto out_brelse;
5046 : }
5047 :
5048 10433 : raw_inode->i_mode = cpu_to_le16(inode->i_mode);
5049 10433 : i_uid = i_uid_read(inode);
5050 10433 : i_gid = i_gid_read(inode);
5051 10433 : i_projid = from_kprojid(&init_user_ns, ei->i_projid);
5052 10433 : if (!(test_opt(inode->i_sb, NO_UID32))) {
5053 10433 : raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
5054 10433 : raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
5055 : /*
5056 : * Fix up interoperability with old kernels. Otherwise, old inodes get
5057 : * re-used with the upper 16 bits of the uid/gid intact
5058 : */
5059 10433 : if (ei->i_dtime && list_empty(&ei->i_orphan)) {
5060 193 : raw_inode->i_uid_high = 0;
5061 193 : raw_inode->i_gid_high = 0;
5062 : } else {
5063 10240 : raw_inode->i_uid_high =
5064 10240 : cpu_to_le16(high_16_bits(i_uid));
5065 10240 : raw_inode->i_gid_high =
5066 10240 : cpu_to_le16(high_16_bits(i_gid));
5067 : }
5068 : } else {
5069 0 : raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
5070 0 : raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
5071 0 : raw_inode->i_uid_high = 0;
5072 0 : raw_inode->i_gid_high = 0;
5073 : }
5074 10433 : raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
5075 :
5076 10433 : EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
5077 10433 : EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
5078 10433 : EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
5079 10433 : EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
5080 :
5081 10433 : raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
5082 10433 : raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
5083 10433 : if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
5084 10433 : raw_inode->i_file_acl_high =
5085 10433 : cpu_to_le16(ei->i_file_acl >> 32);
5086 10433 : raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
5087 10433 : if (READ_ONCE(ei->i_disksize) != ext4_isize(inode->i_sb, raw_inode)) {
5088 413 : ext4_isize_set(raw_inode, ei->i_disksize);
5089 413 : need_datasync = 1;
5090 : }
5091 10433 : if (ei->i_disksize > 0x7fffffffULL) {
5092 0 : if (!ext4_has_feature_large_file(sb) ||
5093 0 : EXT4_SB(sb)->s_es->s_rev_level ==
5094 : cpu_to_le32(EXT4_GOOD_OLD_REV))
5095 : set_large_file = 1;
5096 : }
5097 10433 : raw_inode->i_generation = cpu_to_le32(inode->i_generation);
5098 10433 : if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
5099 0 : if (old_valid_dev(inode->i_rdev)) {
5100 0 : raw_inode->i_block[0] =
5101 0 : cpu_to_le32(old_encode_dev(inode->i_rdev));
5102 0 : raw_inode->i_block[1] = 0;
5103 : } else {
5104 0 : raw_inode->i_block[0] = 0;
5105 0 : raw_inode->i_block[1] =
5106 0 : cpu_to_le32(new_encode_dev(inode->i_rdev));
5107 0 : raw_inode->i_block[2] = 0;
5108 : }
5109 10433 : } else if (!ext4_has_inline_data(inode)) {
5110 166928 : for (block = 0; block < EXT4_N_BLOCKS; block++)
5111 156495 : raw_inode->i_block[block] = ei->i_data[block];
5112 : }
5113 :
5114 10433 : if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
5115 10433 : u64 ivers = ext4_inode_peek_iversion(inode);
5116 :
5117 10433 : raw_inode->i_disk_version = cpu_to_le32(ivers);
5118 10433 : if (ei->i_extra_isize) {
5119 10433 : if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5120 10433 : raw_inode->i_version_hi =
5121 10433 : cpu_to_le32(ivers >> 32);
5122 10433 : raw_inode->i_extra_isize =
5123 10433 : cpu_to_le16(ei->i_extra_isize);
5124 : }
5125 : }
5126 :
5127 10433 : BUG_ON(!ext4_has_feature_project(inode->i_sb) &&
5128 : i_projid != EXT4_DEF_PROJID);
5129 :
5130 10433 : if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
5131 10433 : EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
5132 10433 : raw_inode->i_projid = cpu_to_le32(i_projid);
5133 :
5134 10433 : ext4_inode_csum_set(inode, raw_inode, ei);
5135 10433 : spin_unlock(&ei->i_raw_lock);
5136 10433 : if (inode->i_sb->s_flags & SB_LAZYTIME)
5137 0 : ext4_update_other_inodes_time(inode->i_sb, inode->i_ino,
5138 : bh->b_data);
5139 :
5140 10433 : BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
5141 10433 : rc = ext4_handle_dirty_metadata(handle, NULL, bh);
5142 10433 : if (!err)
5143 10433 : err = rc;
5144 10433 : ext4_clear_inode_state(inode, EXT4_STATE_NEW);
5145 10433 : if (set_large_file) {
5146 0 : BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
5147 0 : err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
5148 0 : if (err)
5149 0 : goto out_brelse;
5150 0 : lock_buffer(EXT4_SB(sb)->s_sbh);
5151 0 : ext4_set_feature_large_file(sb);
5152 0 : ext4_superblock_csum_set(sb);
5153 0 : unlock_buffer(EXT4_SB(sb)->s_sbh);
5154 0 : ext4_handle_sync(handle);
5155 0 : err = ext4_handle_dirty_metadata(handle, NULL,
5156 : EXT4_SB(sb)->s_sbh);
5157 : }
5158 10433 : ext4_update_inode_fsync_trans(handle, inode, need_datasync);
5159 10433 : out_brelse:
5160 10433 : brelse(bh);
5161 10433 : ext4_std_error(inode->i_sb, err);
5162 10433 : return err;
5163 : }
5164 :
5165 : /*
5166 : * ext4_write_inode()
5167 : *
5168 : * We are called from a few places:
5169 : *
5170 : * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
5171 : * Here, there will be no transaction running. We wait for any running
5172 : * transaction to commit.
5173 : *
5174 : * - Within flush work (sys_sync(), kupdate and such).
5175 : * We wait on commit, if told to.
5176 : *
5177 : * - Within iput_final() -> write_inode_now()
5178 : * We wait on commit, if told to.
5179 : *
5180 : * In all cases it is actually safe for us to return without doing anything,
5181 : * because the inode has been copied into a raw inode buffer in
5182 : * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
5183 : * writeback.
5184 : *
5185 : * Note that we are absolutely dependent upon all inode dirtiers doing the
5186 : * right thing: they *must* call mark_inode_dirty() after dirtying info in
5187 : * which we are interested.
5188 : *
5189 : * It would be a bug for them to not do this. The code:
5190 : *
5191 : * mark_inode_dirty(inode)
5192 : * stuff();
5193 : * inode->i_size = expr;
5194 : *
5195 : * is in error because write_inode() could occur while `stuff()' is running,
5196 : * and the new i_size will be lost. Plus the inode will no longer be on the
5197 : * superblock's dirty inode list.
5198 : */
5199 1464 : int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
5200 : {
5201 1464 : int err;
5202 :
5203 1464 : if (WARN_ON_ONCE(current->flags & PF_MEMALLOC) ||
5204 1464 : sb_rdonly(inode->i_sb))
5205 : return 0;
5206 :
5207 1464 : if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5208 : return -EIO;
5209 :
5210 1464 : if (EXT4_SB(inode->i_sb)->s_journal) {
5211 1464 : if (ext4_journal_current_handle()) {
5212 0 : jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5213 0 : dump_stack();
5214 0 : return -EIO;
5215 : }
5216 :
5217 : /*
5218 : * No need to force transaction in WB_SYNC_NONE mode. Also
5219 : * ext4_sync_fs() will force the commit after everything is
5220 : * written.
5221 : */
5222 1464 : if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
5223 : return 0;
5224 :
5225 0 : err = ext4_fc_commit(EXT4_SB(inode->i_sb)->s_journal,
5226 0 : EXT4_I(inode)->i_sync_tid);
5227 : } else {
5228 0 : struct ext4_iloc iloc;
5229 :
5230 0 : err = __ext4_get_inode_loc_noinmem(inode, &iloc);
5231 0 : if (err)
5232 0 : return err;
5233 : /*
5234 : * sync(2) will flush the whole buffer cache. No need to do
5235 : * it here separately for each inode.
5236 : */
5237 0 : if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
5238 0 : sync_dirty_buffer(iloc.bh);
5239 0 : if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5240 0 : ext4_error_inode_block(inode, iloc.bh->b_blocknr, EIO,
5241 : "IO error syncing inode");
5242 0 : err = -EIO;
5243 : }
5244 0 : brelse(iloc.bh);
5245 : }
5246 : return err;
5247 : }
5248 :
5249 : /*
5250 : * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
5251 : * buffers that are attached to a page stradding i_size and are undergoing
5252 : * commit. In that case we have to wait for commit to finish and try again.
5253 : */
5254 0 : static void ext4_wait_for_tail_page_commit(struct inode *inode)
5255 : {
5256 0 : struct page *page;
5257 0 : unsigned offset;
5258 0 : journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
5259 0 : tid_t commit_tid = 0;
5260 0 : int ret;
5261 :
5262 0 : offset = inode->i_size & (PAGE_SIZE - 1);
5263 : /*
5264 : * If the page is fully truncated, we don't need to wait for any commit
5265 : * (and we even should not as __ext4_journalled_invalidatepage() may
5266 : * strip all buffers from the page but keep the page dirty which can then
5267 : * confuse e.g. concurrent ext4_writepage() seeing dirty page without
5268 : * buffers). Also we don't need to wait for any commit if all buffers in
5269 : * the page remain valid. This is most beneficial for the common case of
5270 : * blocksize == PAGESIZE.
5271 : */
5272 0 : if (!offset || offset > (PAGE_SIZE - i_blocksize(inode)))
5273 : return;
5274 0 : while (1) {
5275 0 : page = find_lock_page(inode->i_mapping,
5276 0 : inode->i_size >> PAGE_SHIFT);
5277 0 : if (!page)
5278 : return;
5279 0 : ret = __ext4_journalled_invalidatepage(page, offset,
5280 : PAGE_SIZE - offset);
5281 0 : unlock_page(page);
5282 0 : put_page(page);
5283 0 : if (ret != -EBUSY)
5284 : return;
5285 0 : commit_tid = 0;
5286 0 : read_lock(&journal->j_state_lock);
5287 0 : if (journal->j_committing_transaction)
5288 0 : commit_tid = journal->j_committing_transaction->t_tid;
5289 0 : read_unlock(&journal->j_state_lock);
5290 0 : if (commit_tid)
5291 0 : jbd2_log_wait_commit(journal, commit_tid);
5292 : }
5293 : }
5294 :
5295 : /*
5296 : * ext4_setattr()
5297 : *
5298 : * Called from notify_change.
5299 : *
5300 : * We want to trap VFS attempts to truncate the file as soon as
5301 : * possible. In particular, we want to make sure that when the VFS
5302 : * shrinks i_size, we put the inode on the orphan list and modify
5303 : * i_disksize immediately, so that during the subsequent flushing of
5304 : * dirty pages and freeing of disk blocks, we can guarantee that any
5305 : * commit will leave the blocks being flushed in an unused state on
5306 : * disk. (On recovery, the inode will get truncated and the blocks will
5307 : * be freed, so we have a strong guarantee that no future commit will
5308 : * leave these blocks visible to the user.)
5309 : *
5310 : * Another thing we have to assure is that if we are in ordered mode
5311 : * and inode is still attached to the committing transaction, we must
5312 : * we start writeout of all the dirty pages which are being truncated.
5313 : * This way we are sure that all the data written in the previous
5314 : * transaction are already on disk (truncate waits for pages under
5315 : * writeback).
5316 : *
5317 : * Called with inode->i_mutex down.
5318 : */
5319 303 : int ext4_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
5320 : struct iattr *attr)
5321 : {
5322 303 : struct inode *inode = d_inode(dentry);
5323 303 : int error, rc = 0;
5324 303 : int orphan = 0;
5325 303 : const unsigned int ia_valid = attr->ia_valid;
5326 :
5327 303 : if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5328 : return -EIO;
5329 :
5330 303 : if (unlikely(IS_IMMUTABLE(inode)))
5331 : return -EPERM;
5332 :
5333 303 : if (unlikely(IS_APPEND(inode) &&
5334 : (ia_valid & (ATTR_MODE | ATTR_UID |
5335 : ATTR_GID | ATTR_TIMES_SET))))
5336 : return -EPERM;
5337 :
5338 303 : error = setattr_prepare(mnt_userns, dentry, attr);
5339 303 : if (error)
5340 : return error;
5341 :
5342 303 : error = fscrypt_prepare_setattr(dentry, attr);
5343 303 : if (error)
5344 : return error;
5345 :
5346 303 : error = fsverity_prepare_setattr(dentry, attr);
5347 303 : if (error)
5348 : return error;
5349 :
5350 303 : if (is_quota_modification(inode, attr)) {
5351 303 : error = dquot_initialize(inode);
5352 : if (error)
5353 : return error;
5354 : }
5355 303 : ext4_fc_start_update(inode);
5356 303 : if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
5357 301 : (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
5358 23 : handle_t *handle;
5359 :
5360 : /* (user+group)*(old+new) structure, inode write (sb,
5361 : * inode block, ? - but truncate inode update has it) */
5362 23 : handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5363 : (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
5364 : EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
5365 23 : if (IS_ERR(handle)) {
5366 0 : error = PTR_ERR(handle);
5367 0 : goto err_out;
5368 : }
5369 :
5370 : /* dquot_transfer() calls back ext4_get_inode_usage() which
5371 : * counts xattr inode references.
5372 : */
5373 23 : down_read(&EXT4_I(inode)->xattr_sem);
5374 23 : error = dquot_transfer(inode, attr);
5375 23 : up_read(&EXT4_I(inode)->xattr_sem);
5376 :
5377 23 : if (error) {
5378 : ext4_journal_stop(handle);
5379 : ext4_fc_stop_update(inode);
5380 : return error;
5381 : }
5382 : /* Update corresponding info in inode so that everything is in
5383 : * one transaction */
5384 23 : if (attr->ia_valid & ATTR_UID)
5385 23 : inode->i_uid = attr->ia_uid;
5386 23 : if (attr->ia_valid & ATTR_GID)
5387 23 : inode->i_gid = attr->ia_gid;
5388 23 : error = ext4_mark_inode_dirty(handle, inode);
5389 23 : ext4_journal_stop(handle);
5390 23 : if (unlikely(error))
5391 : return error;
5392 : }
5393 :
5394 303 : if (attr->ia_valid & ATTR_SIZE) {
5395 5 : handle_t *handle;
5396 5 : loff_t oldsize = inode->i_size;
5397 5 : int shrink = (attr->ia_size < inode->i_size);
5398 :
5399 5 : if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
5400 0 : struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5401 :
5402 0 : if (attr->ia_size > sbi->s_bitmap_maxbytes) {
5403 0 : ext4_fc_stop_update(inode);
5404 0 : return -EFBIG;
5405 : }
5406 : }
5407 5 : if (!S_ISREG(inode->i_mode)) {
5408 0 : ext4_fc_stop_update(inode);
5409 0 : return -EINVAL;
5410 : }
5411 :
5412 5 : if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
5413 0 : inode_inc_iversion(inode);
5414 :
5415 5 : if (shrink) {
5416 1 : if (ext4_should_order_data(inode)) {
5417 1 : error = ext4_begin_ordered_truncate(inode,
5418 : attr->ia_size);
5419 1 : if (error)
5420 0 : goto err_out;
5421 : }
5422 : /*
5423 : * Blocks are going to be removed from the inode. Wait
5424 : * for dio in flight.
5425 : */
5426 1 : inode_dio_wait(inode);
5427 : }
5428 :
5429 5 : down_write(&EXT4_I(inode)->i_mmap_sem);
5430 :
5431 5 : rc = ext4_break_layouts(inode);
5432 5 : if (rc) {
5433 0 : up_write(&EXT4_I(inode)->i_mmap_sem);
5434 0 : goto err_out;
5435 : }
5436 :
5437 5 : if (attr->ia_size != inode->i_size) {
5438 1 : handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
5439 1 : if (IS_ERR(handle)) {
5440 0 : error = PTR_ERR(handle);
5441 0 : goto out_mmap_sem;
5442 : }
5443 1 : if (ext4_handle_valid(handle) && shrink) {
5444 1 : error = ext4_orphan_add(handle, inode);
5445 1 : orphan = 1;
5446 : }
5447 : /*
5448 : * Update c/mtime on truncate up, ext4_truncate() will
5449 : * update c/mtime in shrink case below
5450 : */
5451 1 : if (!shrink) {
5452 0 : inode->i_mtime = current_time(inode);
5453 0 : inode->i_ctime = inode->i_mtime;
5454 : }
5455 :
5456 1 : if (shrink)
5457 1 : ext4_fc_track_range(handle, inode,
5458 1 : (attr->ia_size > 0 ? attr->ia_size - 1 : 0) >>
5459 : inode->i_sb->s_blocksize_bits,
5460 1 : (oldsize > 0 ? oldsize - 1 : 0) >>
5461 1 : inode->i_sb->s_blocksize_bits);
5462 : else
5463 0 : ext4_fc_track_range(
5464 : handle, inode,
5465 0 : (oldsize > 0 ? oldsize - 1 : oldsize) >>
5466 : inode->i_sb->s_blocksize_bits,
5467 0 : (attr->ia_size > 0 ? attr->ia_size - 1 : 0) >>
5468 0 : inode->i_sb->s_blocksize_bits);
5469 :
5470 1 : down_write(&EXT4_I(inode)->i_data_sem);
5471 1 : EXT4_I(inode)->i_disksize = attr->ia_size;
5472 1 : rc = ext4_mark_inode_dirty(handle, inode);
5473 1 : if (!error)
5474 1 : error = rc;
5475 : /*
5476 : * We have to update i_size under i_data_sem together
5477 : * with i_disksize to avoid races with writeback code
5478 : * running ext4_wb_update_i_disksize().
5479 : */
5480 1 : if (!error)
5481 1 : i_size_write(inode, attr->ia_size);
5482 1 : up_write(&EXT4_I(inode)->i_data_sem);
5483 1 : ext4_journal_stop(handle);
5484 1 : if (error)
5485 0 : goto out_mmap_sem;
5486 1 : if (!shrink) {
5487 0 : pagecache_isize_extended(inode, oldsize,
5488 : inode->i_size);
5489 1 : } else if (ext4_should_journal_data(inode)) {
5490 0 : ext4_wait_for_tail_page_commit(inode);
5491 : }
5492 : }
5493 :
5494 : /*
5495 : * Truncate pagecache after we've waited for commit
5496 : * in data=journal mode to make pages freeable.
5497 : */
5498 5 : truncate_pagecache(inode, inode->i_size);
5499 : /*
5500 : * Call ext4_truncate() even if i_size didn't change to
5501 : * truncate possible preallocated blocks.
5502 : */
5503 5 : if (attr->ia_size <= oldsize) {
5504 5 : rc = ext4_truncate(inode);
5505 5 : if (rc)
5506 0 : error = rc;
5507 : }
5508 5 : out_mmap_sem:
5509 5 : up_write(&EXT4_I(inode)->i_mmap_sem);
5510 : }
5511 :
5512 303 : if (!error) {
5513 303 : setattr_copy(mnt_userns, inode, attr);
5514 303 : mark_inode_dirty(inode);
5515 : }
5516 :
5517 : /*
5518 : * If the call to ext4_truncate failed to get a transaction handle at
5519 : * all, we need to clean up the in-core orphan list manually.
5520 : */
5521 303 : if (orphan && inode->i_nlink)
5522 1 : ext4_orphan_del(NULL, inode);
5523 :
5524 303 : if (!error && (ia_valid & ATTR_MODE))
5525 143 : rc = posix_acl_chmod(mnt_userns, inode, inode->i_mode);
5526 :
5527 160 : err_out:
5528 303 : if (error)
5529 0 : ext4_std_error(inode->i_sb, error);
5530 303 : if (!error)
5531 303 : error = rc;
5532 303 : ext4_fc_stop_update(inode);
5533 303 : return error;
5534 : }
5535 :
5536 25288 : int ext4_getattr(struct user_namespace *mnt_userns, const struct path *path,
5537 : struct kstat *stat, u32 request_mask, unsigned int query_flags)
5538 : {
5539 25288 : struct inode *inode = d_inode(path->dentry);
5540 25288 : struct ext4_inode *raw_inode;
5541 25288 : struct ext4_inode_info *ei = EXT4_I(inode);
5542 25288 : unsigned int flags;
5543 :
5544 25288 : if ((request_mask & STATX_BTIME) &&
5545 0 : EXT4_FITS_IN_INODE(raw_inode, ei, i_crtime)) {
5546 0 : stat->result_mask |= STATX_BTIME;
5547 0 : stat->btime.tv_sec = ei->i_crtime.tv_sec;
5548 0 : stat->btime.tv_nsec = ei->i_crtime.tv_nsec;
5549 : }
5550 :
5551 25288 : flags = ei->i_flags & EXT4_FL_USER_VISIBLE;
5552 25288 : if (flags & EXT4_APPEND_FL)
5553 0 : stat->attributes |= STATX_ATTR_APPEND;
5554 25288 : if (flags & EXT4_COMPR_FL)
5555 0 : stat->attributes |= STATX_ATTR_COMPRESSED;
5556 25288 : if (flags & EXT4_ENCRYPT_FL)
5557 0 : stat->attributes |= STATX_ATTR_ENCRYPTED;
5558 25288 : if (flags & EXT4_IMMUTABLE_FL)
5559 0 : stat->attributes |= STATX_ATTR_IMMUTABLE;
5560 25288 : if (flags & EXT4_NODUMP_FL)
5561 0 : stat->attributes |= STATX_ATTR_NODUMP;
5562 25288 : if (flags & EXT4_VERITY_FL)
5563 0 : stat->attributes |= STATX_ATTR_VERITY;
5564 :
5565 25288 : stat->attributes_mask |= (STATX_ATTR_APPEND |
5566 : STATX_ATTR_COMPRESSED |
5567 : STATX_ATTR_ENCRYPTED |
5568 : STATX_ATTR_IMMUTABLE |
5569 : STATX_ATTR_NODUMP |
5570 : STATX_ATTR_VERITY);
5571 :
5572 25288 : generic_fillattr(mnt_userns, inode, stat);
5573 25284 : return 0;
5574 : }
5575 :
5576 14723 : int ext4_file_getattr(struct user_namespace *mnt_userns,
5577 : const struct path *path, struct kstat *stat,
5578 : u32 request_mask, unsigned int query_flags)
5579 : {
5580 14723 : struct inode *inode = d_inode(path->dentry);
5581 14723 : u64 delalloc_blocks;
5582 :
5583 14723 : ext4_getattr(mnt_userns, path, stat, request_mask, query_flags);
5584 :
5585 : /*
5586 : * If there is inline data in the inode, the inode will normally not
5587 : * have data blocks allocated (it may have an external xattr block).
5588 : * Report at least one sector for such files, so tools like tar, rsync,
5589 : * others don't incorrectly think the file is completely sparse.
5590 : */
5591 14723 : if (unlikely(ext4_has_inline_data(inode)))
5592 0 : stat->blocks += (stat->size + 511) >> 9;
5593 :
5594 : /*
5595 : * We can't update i_blocks if the block allocation is delayed
5596 : * otherwise in the case of system crash before the real block
5597 : * allocation is done, we will have i_blocks inconsistent with
5598 : * on-disk file blocks.
5599 : * We always keep i_blocks updated together with real
5600 : * allocation. But to not confuse with user, stat
5601 : * will return the blocks that include the delayed allocation
5602 : * blocks for this file.
5603 : */
5604 14722 : delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
5605 : EXT4_I(inode)->i_reserved_data_blocks);
5606 14722 : stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
5607 14722 : return 0;
5608 : }
5609 :
5610 294 : static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
5611 : int pextents)
5612 : {
5613 294 : if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5614 0 : return ext4_ind_trans_blocks(inode, lblocks);
5615 294 : return ext4_ext_index_trans_blocks(inode, pextents);
5616 : }
5617 :
5618 : /*
5619 : * Account for index blocks, block groups bitmaps and block group
5620 : * descriptor blocks if modify datablocks and index blocks
5621 : * worse case, the indexs blocks spread over different block groups
5622 : *
5623 : * If datablocks are discontiguous, they are possible to spread over
5624 : * different block groups too. If they are contiguous, with flexbg,
5625 : * they could still across block group boundary.
5626 : *
5627 : * Also account for superblock, inode, quota and xattr blocks
5628 : */
5629 294 : static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
5630 : int pextents)
5631 : {
5632 294 : ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5633 294 : int gdpblocks;
5634 294 : int idxblocks;
5635 294 : int ret = 0;
5636 :
5637 : /*
5638 : * How many index blocks need to touch to map @lblocks logical blocks
5639 : * to @pextents physical extents?
5640 : */
5641 294 : idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
5642 :
5643 294 : ret = idxblocks;
5644 :
5645 : /*
5646 : * Now let's see how many group bitmaps and group descriptors need
5647 : * to account
5648 : */
5649 294 : groups = idxblocks + pextents;
5650 294 : gdpblocks = groups;
5651 294 : if (groups > ngroups)
5652 : groups = ngroups;
5653 294 : if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5654 15 : gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5655 :
5656 : /* bitmaps and block group descriptor blocks */
5657 294 : ret += groups + gdpblocks;
5658 :
5659 : /* Blocks for super block, inode, quota and xattr blocks */
5660 294 : ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5661 :
5662 294 : return ret;
5663 : }
5664 :
5665 : /*
5666 : * Calculate the total number of credits to reserve to fit
5667 : * the modification of a single pages into a single transaction,
5668 : * which may include multiple chunks of block allocations.
5669 : *
5670 : * This could be called via ext4_write_begin()
5671 : *
5672 : * We need to consider the worse case, when
5673 : * one new block per extent.
5674 : */
5675 140 : int ext4_writepage_trans_blocks(struct inode *inode)
5676 : {
5677 140 : int bpp = ext4_journal_blocks_per_page(inode);
5678 140 : int ret;
5679 :
5680 140 : ret = ext4_meta_trans_blocks(inode, bpp, bpp);
5681 :
5682 : /* Account for data blocks for journalled mode */
5683 140 : if (ext4_should_journal_data(inode))
5684 71 : ret += bpp;
5685 140 : return ret;
5686 : }
5687 :
5688 : /*
5689 : * Calculate the journal credits for a chunk of data modification.
5690 : *
5691 : * This is called from DIO, fallocate or whoever calling
5692 : * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5693 : *
5694 : * journal buffers for data blocks are not included here, as DIO
5695 : * and fallocate do no need to journal data buffers.
5696 : */
5697 84 : int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5698 : {
5699 84 : return ext4_meta_trans_blocks(inode, nrblocks, 1);
5700 : }
5701 :
5702 : /*
5703 : * The caller must have previously called ext4_reserve_inode_write().
5704 : * Give this, we know that the caller already has write access to iloc->bh.
5705 : */
5706 10433 : int ext4_mark_iloc_dirty(handle_t *handle,
5707 : struct inode *inode, struct ext4_iloc *iloc)
5708 : {
5709 10433 : int err = 0;
5710 :
5711 10433 : if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) {
5712 0 : put_bh(iloc->bh);
5713 0 : return -EIO;
5714 : }
5715 10433 : ext4_fc_track_inode(handle, inode);
5716 :
5717 10433 : if (IS_I_VERSION(inode))
5718 0 : inode_inc_iversion(inode);
5719 :
5720 : /* the do_update_inode consumes one bh->b_count */
5721 10433 : get_bh(iloc->bh);
5722 :
5723 : /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5724 10433 : err = ext4_do_update_inode(handle, inode, iloc);
5725 10433 : put_bh(iloc->bh);
5726 10433 : return err;
5727 : }
5728 :
5729 : /*
5730 : * On success, We end up with an outstanding reference count against
5731 : * iloc->bh. This _must_ be cleaned up later.
5732 : */
5733 :
5734 : int
5735 10433 : ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5736 : struct ext4_iloc *iloc)
5737 : {
5738 10433 : int err;
5739 :
5740 10433 : if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5741 : return -EIO;
5742 :
5743 10432 : err = ext4_get_inode_loc(inode, iloc);
5744 10433 : if (!err) {
5745 10433 : BUFFER_TRACE(iloc->bh, "get_write_access");
5746 10433 : err = ext4_journal_get_write_access(handle, iloc->bh);
5747 10433 : if (err) {
5748 0 : brelse(iloc->bh);
5749 0 : iloc->bh = NULL;
5750 : }
5751 : }
5752 10433 : ext4_std_error(inode->i_sb, err);
5753 : return err;
5754 : }
5755 :
5756 0 : static int __ext4_expand_extra_isize(struct inode *inode,
5757 : unsigned int new_extra_isize,
5758 : struct ext4_iloc *iloc,
5759 : handle_t *handle, int *no_expand)
5760 : {
5761 0 : struct ext4_inode *raw_inode;
5762 0 : struct ext4_xattr_ibody_header *header;
5763 0 : unsigned int inode_size = EXT4_INODE_SIZE(inode->i_sb);
5764 0 : struct ext4_inode_info *ei = EXT4_I(inode);
5765 0 : int error;
5766 :
5767 : /* this was checked at iget time, but double check for good measure */
5768 0 : if ((EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > inode_size) ||
5769 : (ei->i_extra_isize & 3)) {
5770 0 : EXT4_ERROR_INODE(inode, "bad extra_isize %u (inode size %u)",
5771 : ei->i_extra_isize,
5772 : EXT4_INODE_SIZE(inode->i_sb));
5773 0 : return -EFSCORRUPTED;
5774 : }
5775 0 : if ((new_extra_isize < ei->i_extra_isize) ||
5776 0 : (new_extra_isize < 4) ||
5777 0 : (new_extra_isize > inode_size - EXT4_GOOD_OLD_INODE_SIZE))
5778 : return -EINVAL; /* Should never happen */
5779 :
5780 0 : raw_inode = ext4_raw_inode(iloc);
5781 :
5782 0 : header = IHDR(inode, raw_inode);
5783 :
5784 : /* No extended attributes present */
5785 0 : if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5786 0 : header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5787 0 : memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE +
5788 0 : EXT4_I(inode)->i_extra_isize, 0,
5789 0 : new_extra_isize - EXT4_I(inode)->i_extra_isize);
5790 0 : EXT4_I(inode)->i_extra_isize = new_extra_isize;
5791 0 : return 0;
5792 : }
5793 :
5794 : /* try to expand with EAs present */
5795 0 : error = ext4_expand_extra_isize_ea(inode, new_extra_isize,
5796 : raw_inode, handle);
5797 0 : if (error) {
5798 : /*
5799 : * Inode size expansion failed; don't try again
5800 : */
5801 0 : *no_expand = 1;
5802 : }
5803 :
5804 : return error;
5805 : }
5806 :
5807 : /*
5808 : * Expand an inode by new_extra_isize bytes.
5809 : * Returns 0 on success or negative error number on failure.
5810 : */
5811 0 : static int ext4_try_to_expand_extra_isize(struct inode *inode,
5812 : unsigned int new_extra_isize,
5813 : struct ext4_iloc iloc,
5814 : handle_t *handle)
5815 : {
5816 0 : int no_expand;
5817 0 : int error;
5818 :
5819 0 : if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND))
5820 : return -EOVERFLOW;
5821 :
5822 : /*
5823 : * In nojournal mode, we can immediately attempt to expand
5824 : * the inode. When journaled, we first need to obtain extra
5825 : * buffer credits since we may write into the EA block
5826 : * with this same handle. If journal_extend fails, then it will
5827 : * only result in a minor loss of functionality for that inode.
5828 : * If this is felt to be critical, then e2fsck should be run to
5829 : * force a large enough s_min_extra_isize.
5830 : */
5831 0 : if (ext4_journal_extend(handle,
5832 0 : EXT4_DATA_TRANS_BLOCKS(inode->i_sb), 0) != 0)
5833 : return -ENOSPC;
5834 :
5835 0 : if (ext4_write_trylock_xattr(inode, &no_expand) == 0)
5836 : return -EBUSY;
5837 :
5838 0 : error = __ext4_expand_extra_isize(inode, new_extra_isize, &iloc,
5839 : handle, &no_expand);
5840 0 : ext4_write_unlock_xattr(inode, &no_expand);
5841 :
5842 0 : return error;
5843 : }
5844 :
5845 0 : int ext4_expand_extra_isize(struct inode *inode,
5846 : unsigned int new_extra_isize,
5847 : struct ext4_iloc *iloc)
5848 : {
5849 0 : handle_t *handle;
5850 0 : int no_expand;
5851 0 : int error, rc;
5852 :
5853 0 : if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5854 0 : brelse(iloc->bh);
5855 0 : return -EOVERFLOW;
5856 : }
5857 :
5858 0 : handle = ext4_journal_start(inode, EXT4_HT_INODE,
5859 : EXT4_DATA_TRANS_BLOCKS(inode->i_sb));
5860 0 : if (IS_ERR(handle)) {
5861 0 : error = PTR_ERR(handle);
5862 0 : brelse(iloc->bh);
5863 0 : return error;
5864 : }
5865 :
5866 0 : ext4_write_lock_xattr(inode, &no_expand);
5867 :
5868 0 : BUFFER_TRACE(iloc->bh, "get_write_access");
5869 0 : error = ext4_journal_get_write_access(handle, iloc->bh);
5870 0 : if (error) {
5871 0 : brelse(iloc->bh);
5872 0 : goto out_unlock;
5873 : }
5874 :
5875 0 : error = __ext4_expand_extra_isize(inode, new_extra_isize, iloc,
5876 : handle, &no_expand);
5877 :
5878 0 : rc = ext4_mark_iloc_dirty(handle, inode, iloc);
5879 0 : if (!error)
5880 0 : error = rc;
5881 :
5882 0 : out_unlock:
5883 0 : ext4_write_unlock_xattr(inode, &no_expand);
5884 0 : ext4_journal_stop(handle);
5885 0 : return error;
5886 : }
5887 :
5888 : /*
5889 : * What we do here is to mark the in-core inode as clean with respect to inode
5890 : * dirtiness (it may still be data-dirty).
5891 : * This means that the in-core inode may be reaped by prune_icache
5892 : * without having to perform any I/O. This is a very good thing,
5893 : * because *any* task may call prune_icache - even ones which
5894 : * have a transaction open against a different journal.
5895 : *
5896 : * Is this cheating? Not really. Sure, we haven't written the
5897 : * inode out, but prune_icache isn't a user-visible syncing function.
5898 : * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5899 : * we start and wait on commits.
5900 : */
5901 10036 : int __ext4_mark_inode_dirty(handle_t *handle, struct inode *inode,
5902 : const char *func, unsigned int line)
5903 : {
5904 10036 : struct ext4_iloc iloc;
5905 10036 : struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5906 10036 : int err;
5907 :
5908 10036 : might_sleep();
5909 10037 : trace_ext4_mark_inode_dirty(inode, _RET_IP_);
5910 10037 : err = ext4_reserve_inode_write(handle, inode, &iloc);
5911 10037 : if (err)
5912 0 : goto out;
5913 :
5914 10037 : if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize)
5915 0 : ext4_try_to_expand_extra_isize(inode, sbi->s_want_extra_isize,
5916 : iloc, handle);
5917 :
5918 10037 : err = ext4_mark_iloc_dirty(handle, inode, &iloc);
5919 10037 : out:
5920 10037 : if (unlikely(err))
5921 0 : ext4_error_inode_err(inode, func, line, 0, err,
5922 : "mark_inode_dirty error");
5923 10037 : return err;
5924 : }
5925 :
5926 : /*
5927 : * ext4_dirty_inode() is called from __mark_inode_dirty()
5928 : *
5929 : * We're really interested in the case where a file is being extended.
5930 : * i_size has been changed by generic_commit_write() and we thus need
5931 : * to include the updated inode in the current transaction.
5932 : *
5933 : * Also, dquot_alloc_block() will always dirty the inode when blocks
5934 : * are allocated to the file.
5935 : *
5936 : * If the inode is marked synchronous, we don't honour that here - doing
5937 : * so would cause a commit on atime updates, which we don't bother doing.
5938 : * We handle synchronous inodes at the highest possible level.
5939 : */
5940 5241 : void ext4_dirty_inode(struct inode *inode, int flags)
5941 : {
5942 5241 : handle_t *handle;
5943 :
5944 5241 : handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
5945 5241 : if (IS_ERR(handle))
5946 : return;
5947 5241 : ext4_mark_inode_dirty(handle, inode);
5948 5241 : ext4_journal_stop(handle);
5949 : }
5950 :
5951 0 : int ext4_change_inode_journal_flag(struct inode *inode, int val)
5952 : {
5953 0 : journal_t *journal;
5954 0 : handle_t *handle;
5955 0 : int err;
5956 0 : struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5957 :
5958 : /*
5959 : * We have to be very careful here: changing a data block's
5960 : * journaling status dynamically is dangerous. If we write a
5961 : * data block to the journal, change the status and then delete
5962 : * that block, we risk forgetting to revoke the old log record
5963 : * from the journal and so a subsequent replay can corrupt data.
5964 : * So, first we make sure that the journal is empty and that
5965 : * nobody is changing anything.
5966 : */
5967 :
5968 0 : journal = EXT4_JOURNAL(inode);
5969 0 : if (!journal)
5970 : return 0;
5971 0 : if (is_journal_aborted(journal))
5972 : return -EROFS;
5973 :
5974 : /* Wait for all existing dio workers */
5975 0 : inode_dio_wait(inode);
5976 :
5977 : /*
5978 : * Before flushing the journal and switching inode's aops, we have
5979 : * to flush all dirty data the inode has. There can be outstanding
5980 : * delayed allocations, there can be unwritten extents created by
5981 : * fallocate or buffered writes in dioread_nolock mode covered by
5982 : * dirty data which can be converted only after flushing the dirty
5983 : * data (and journalled aops don't know how to handle these cases).
5984 : */
5985 0 : if (val) {
5986 0 : down_write(&EXT4_I(inode)->i_mmap_sem);
5987 0 : err = filemap_write_and_wait(inode->i_mapping);
5988 0 : if (err < 0) {
5989 0 : up_write(&EXT4_I(inode)->i_mmap_sem);
5990 0 : return err;
5991 : }
5992 : }
5993 :
5994 0 : percpu_down_write(&sbi->s_writepages_rwsem);
5995 0 : jbd2_journal_lock_updates(journal);
5996 :
5997 : /*
5998 : * OK, there are no updates running now, and all cached data is
5999 : * synced to disk. We are now in a completely consistent state
6000 : * which doesn't have anything in the journal, and we know that
6001 : * no filesystem updates are running, so it is safe to modify
6002 : * the inode's in-core data-journaling state flag now.
6003 : */
6004 :
6005 0 : if (val)
6006 0 : ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
6007 : else {
6008 0 : err = jbd2_journal_flush(journal);
6009 0 : if (err < 0) {
6010 0 : jbd2_journal_unlock_updates(journal);
6011 0 : percpu_up_write(&sbi->s_writepages_rwsem);
6012 0 : return err;
6013 : }
6014 0 : ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
6015 : }
6016 0 : ext4_set_aops(inode);
6017 :
6018 0 : jbd2_journal_unlock_updates(journal);
6019 0 : percpu_up_write(&sbi->s_writepages_rwsem);
6020 :
6021 0 : if (val)
6022 0 : up_write(&EXT4_I(inode)->i_mmap_sem);
6023 :
6024 : /* Finally we can mark the inode as dirty. */
6025 :
6026 0 : handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
6027 0 : if (IS_ERR(handle))
6028 0 : return PTR_ERR(handle);
6029 :
6030 0 : ext4_fc_mark_ineligible(inode->i_sb,
6031 : EXT4_FC_REASON_JOURNAL_FLAG_CHANGE);
6032 0 : err = ext4_mark_inode_dirty(handle, inode);
6033 0 : ext4_handle_sync(handle);
6034 0 : ext4_journal_stop(handle);
6035 0 : ext4_std_error(inode->i_sb, err);
6036 :
6037 : return err;
6038 : }
6039 :
6040 0 : static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
6041 : {
6042 0 : return !buffer_mapped(bh);
6043 : }
6044 :
6045 304 : vm_fault_t ext4_page_mkwrite(struct vm_fault *vmf)
6046 : {
6047 304 : struct vm_area_struct *vma = vmf->vma;
6048 304 : struct page *page = vmf->page;
6049 304 : loff_t size;
6050 304 : unsigned long len;
6051 304 : int err;
6052 304 : vm_fault_t ret;
6053 304 : struct file *file = vma->vm_file;
6054 304 : struct inode *inode = file_inode(file);
6055 304 : struct address_space *mapping = inode->i_mapping;
6056 304 : handle_t *handle;
6057 304 : get_block_t *get_block;
6058 304 : int retries = 0;
6059 :
6060 304 : if (unlikely(IS_IMMUTABLE(inode)))
6061 : return VM_FAULT_SIGBUS;
6062 :
6063 304 : sb_start_pagefault(inode->i_sb);
6064 304 : file_update_time(vma->vm_file);
6065 :
6066 304 : down_read(&EXT4_I(inode)->i_mmap_sem);
6067 :
6068 304 : err = ext4_convert_inline_data(inode);
6069 304 : if (err)
6070 0 : goto out_ret;
6071 :
6072 : /*
6073 : * On data journalling we skip straight to the transaction handle:
6074 : * there's no delalloc; page truncated will be checked later; the
6075 : * early return w/ all buffers mapped (calculates size/len) can't
6076 : * be used; and there's no dioread_nolock, so only ext4_get_block.
6077 : */
6078 304 : if (ext4_should_journal_data(inode))
6079 0 : goto retry_alloc;
6080 :
6081 : /* Delalloc case is easy... */
6082 608 : if (test_opt(inode->i_sb, DELALLOC) &&
6083 304 : !ext4_nonda_switch(inode->i_sb)) {
6084 304 : do {
6085 304 : err = block_page_mkwrite(vma, vmf,
6086 : ext4_da_get_block_prep);
6087 0 : } while (err == -ENOSPC &&
6088 304 : ext4_should_retry_alloc(inode->i_sb, &retries));
6089 304 : goto out_ret;
6090 : }
6091 :
6092 0 : lock_page(page);
6093 0 : size = i_size_read(inode);
6094 : /* Page got truncated from under us? */
6095 0 : if (page->mapping != mapping || page_offset(page) > size) {
6096 0 : unlock_page(page);
6097 0 : ret = VM_FAULT_NOPAGE;
6098 0 : goto out;
6099 : }
6100 :
6101 0 : if (page->index == size >> PAGE_SHIFT)
6102 0 : len = size & ~PAGE_MASK;
6103 : else
6104 : len = PAGE_SIZE;
6105 : /*
6106 : * Return if we have all the buffers mapped. This avoids the need to do
6107 : * journal_start/journal_stop which can block and take a long time
6108 : *
6109 : * This cannot be done for data journalling, as we have to add the
6110 : * inode to the transaction's list to writeprotect pages on commit.
6111 : */
6112 0 : if (page_has_buffers(page)) {
6113 0 : if (!ext4_walk_page_buffers(NULL, page_buffers(page),
6114 : 0, len, NULL,
6115 : ext4_bh_unmapped)) {
6116 : /* Wait so that we don't change page under IO */
6117 0 : wait_for_stable_page(page);
6118 0 : ret = VM_FAULT_LOCKED;
6119 0 : goto out;
6120 : }
6121 : }
6122 0 : unlock_page(page);
6123 : /* OK, we need to fill the hole... */
6124 0 : if (ext4_should_dioread_nolock(inode))
6125 0 : get_block = ext4_get_block_unwritten;
6126 : else
6127 0 : get_block = ext4_get_block;
6128 0 : retry_alloc:
6129 0 : handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
6130 : ext4_writepage_trans_blocks(inode));
6131 0 : if (IS_ERR(handle)) {
6132 0 : ret = VM_FAULT_SIGBUS;
6133 0 : goto out;
6134 : }
6135 : /*
6136 : * Data journalling can't use block_page_mkwrite() because it
6137 : * will set_buffer_dirty() before do_journal_get_write_access()
6138 : * thus might hit warning messages for dirty metadata buffers.
6139 : */
6140 0 : if (!ext4_should_journal_data(inode)) {
6141 0 : err = block_page_mkwrite(vma, vmf, get_block);
6142 : } else {
6143 0 : lock_page(page);
6144 0 : size = i_size_read(inode);
6145 : /* Page got truncated from under us? */
6146 0 : if (page->mapping != mapping || page_offset(page) > size) {
6147 0 : ret = VM_FAULT_NOPAGE;
6148 0 : goto out_error;
6149 : }
6150 :
6151 0 : if (page->index == size >> PAGE_SHIFT)
6152 0 : len = size & ~PAGE_MASK;
6153 : else
6154 : len = PAGE_SIZE;
6155 :
6156 0 : err = __block_write_begin(page, 0, len, ext4_get_block);
6157 0 : if (!err) {
6158 0 : ret = VM_FAULT_SIGBUS;
6159 0 : if (ext4_walk_page_buffers(handle, page_buffers(page),
6160 : 0, len, NULL, do_journal_get_write_access))
6161 0 : goto out_error;
6162 0 : if (ext4_walk_page_buffers(handle, page_buffers(page),
6163 : 0, len, NULL, write_end_fn))
6164 0 : goto out_error;
6165 0 : if (ext4_jbd2_inode_add_write(handle, inode,
6166 : page_offset(page), len))
6167 0 : goto out_error;
6168 0 : ext4_set_inode_state(inode, EXT4_STATE_JDATA);
6169 : } else {
6170 0 : unlock_page(page);
6171 : }
6172 : }
6173 0 : ext4_journal_stop(handle);
6174 0 : if (err == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
6175 0 : goto retry_alloc;
6176 0 : out_ret:
6177 304 : ret = block_page_mkwrite_return(err);
6178 304 : out:
6179 304 : up_read(&EXT4_I(inode)->i_mmap_sem);
6180 304 : sb_end_pagefault(inode->i_sb);
6181 304 : return ret;
6182 0 : out_error:
6183 0 : unlock_page(page);
6184 0 : ext4_journal_stop(handle);
6185 0 : goto out;
6186 : }
6187 :
6188 11551 : vm_fault_t ext4_filemap_fault(struct vm_fault *vmf)
6189 : {
6190 11551 : struct inode *inode = file_inode(vmf->vma->vm_file);
6191 11551 : vm_fault_t ret;
6192 :
6193 11551 : down_read(&EXT4_I(inode)->i_mmap_sem);
6194 11551 : ret = filemap_fault(vmf);
6195 11551 : up_read(&EXT4_I(inode)->i_mmap_sem);
6196 :
6197 11551 : return ret;
6198 : }
|
