Branch data 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/time.h>
24 : : #include <linux/highuid.h>
25 : : #include <linux/pagemap.h>
26 : : #include <linux/dax.h>
27 : : #include <linux/quotaops.h>
28 : : #include <linux/string.h>
29 : : #include <linux/buffer_head.h>
30 : : #include <linux/writeback.h>
31 : : #include <linux/pagevec.h>
32 : : #include <linux/mpage.h>
33 : : #include <linux/namei.h>
34 : : #include <linux/uio.h>
35 : : #include <linux/bio.h>
36 : : #include <linux/workqueue.h>
37 : : #include <linux/kernel.h>
38 : : #include <linux/printk.h>
39 : : #include <linux/slab.h>
40 : : #include <linux/bitops.h>
41 : : #include <linux/iomap.h>
42 : : #include <linux/iversion.h>
43 : :
44 : : #include "ext4_jbd2.h"
45 : : #include "xattr.h"
46 : : #include "acl.h"
47 : : #include "truncate.h"
48 : :
49 : : #include <trace/events/ext4.h>
50 : :
51 : : static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw,
52 : : struct ext4_inode_info *ei)
53 : : {
54 : : struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
55 : : __u32 csum;
56 : : __u16 dummy_csum = 0;
57 : : int offset = offsetof(struct ext4_inode, i_checksum_lo);
58 : : unsigned int csum_size = sizeof(dummy_csum);
59 : :
60 : : csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw, offset);
61 : : csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, csum_size);
62 : : offset += csum_size;
63 : : csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
64 : : EXT4_GOOD_OLD_INODE_SIZE - offset);
65 : :
66 : : if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
67 : : offset = offsetof(struct ext4_inode, i_checksum_hi);
68 : : csum = ext4_chksum(sbi, csum, (__u8 *)raw +
69 : : EXT4_GOOD_OLD_INODE_SIZE,
70 : : offset - EXT4_GOOD_OLD_INODE_SIZE);
71 : : if (EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) {
72 : : csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum,
73 : : csum_size);
74 : : offset += csum_size;
75 : : }
76 : : csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset,
77 : : EXT4_INODE_SIZE(inode->i_sb) - offset);
78 : : }
79 : :
80 : : return csum;
81 : : }
82 : :
83 : 12285 : static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw,
84 : : struct ext4_inode_info *ei)
85 : : {
86 : 12285 : __u32 provided, calculated;
87 : :
88 [ + - ]: 12285 : if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
89 [ - + ]: 12285 : cpu_to_le32(EXT4_OS_LINUX) ||
90 : 12285 : !ext4_has_metadata_csum(inode->i_sb))
91 : 0 : return 1;
92 : :
93 : 12285 : provided = le16_to_cpu(raw->i_checksum_lo);
94 : 12285 : calculated = ext4_inode_csum(inode, raw, ei);
95 [ + - ]: 12285 : if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
96 [ + - ]: 12285 : EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
97 : 12285 : provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16;
98 : : else
99 : 0 : calculated &= 0xFFFF;
100 : :
101 : 12285 : return provided == calculated;
102 : : }
103 : :
104 : 110151 : static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw,
105 : : struct ext4_inode_info *ei)
106 : : {
107 : 110151 : __u32 csum;
108 : :
109 [ + - ]: 110151 : if (EXT4_SB(inode->i_sb)->s_es->s_creator_os !=
110 [ - + ]: 110151 : cpu_to_le32(EXT4_OS_LINUX) ||
111 : 110151 : !ext4_has_metadata_csum(inode->i_sb))
112 : 0 : return;
113 : :
114 : 110151 : csum = ext4_inode_csum(inode, raw, ei);
115 : 110151 : raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF);
116 [ + - ]: 110151 : if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
117 [ + - ]: 110151 : EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi))
118 : 110151 : raw->i_checksum_hi = cpu_to_le16(csum >> 16);
119 : : }
120 : :
121 : 168 : static inline int ext4_begin_ordered_truncate(struct inode *inode,
122 : : loff_t new_size)
123 : : {
124 : 168 : trace_ext4_begin_ordered_truncate(inode, new_size);
125 : : /*
126 : : * If jinode is zero, then we never opened the file for
127 : : * writing, so there's no need to call
128 : : * jbd2_journal_begin_ordered_truncate() since there's no
129 : : * outstanding writes we need to flush.
130 : : */
131 [ - + ]: 168 : if (!EXT4_I(inode)->jinode)
132 : : return 0;
133 : 0 : return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode),
134 : 0 : EXT4_I(inode)->jinode,
135 : : new_size);
136 : : }
137 : :
138 : : static void ext4_invalidatepage(struct page *page, unsigned int offset,
139 : : unsigned int length);
140 : : static int __ext4_journalled_writepage(struct page *page, unsigned int len);
141 : : static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh);
142 : : static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
143 : : int pextents);
144 : :
145 : : /*
146 : : * Test whether an inode is a fast symlink.
147 : : * A fast symlink has its symlink data stored in ext4_inode_info->i_data.
148 : : */
149 : 5544 : int ext4_inode_is_fast_symlink(struct inode *inode)
150 : : {
151 [ + - ]: 5544 : if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) {
152 [ - + ]: 5544 : int ea_blocks = EXT4_I(inode)->i_file_acl ?
153 : 0 : EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0;
154 : :
155 : 5544 : if (ext4_has_inline_data(inode))
156 : : return 0;
157 : :
158 [ + + - + ]: 5544 : return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0);
159 : : }
160 [ # # # # : 0 : return S_ISLNK(inode->i_mode) && inode->i_size &&
# # ]
161 : : (inode->i_size < EXT4_N_BLOCKS * 4);
162 : : }
163 : :
164 : : /*
165 : : * Called at the last iput() if i_nlink is zero.
166 : : */
167 : 210 : void ext4_evict_inode(struct inode *inode)
168 : : {
169 : 210 : handle_t *handle;
170 : 210 : int err;
171 : : /*
172 : : * Credits for final inode cleanup and freeing:
173 : : * sb + inode (ext4_orphan_del()), block bitmap, group descriptor
174 : : * (xattr block freeing), bitmap, group descriptor (inode freeing)
175 : : */
176 : 210 : int extra_credits = 6;
177 : 210 : struct ext4_xattr_inode_array *ea_inode_array = NULL;
178 : :
179 : 210 : trace_ext4_evict_inode(inode);
180 : :
181 [ - + ]: 210 : if (inode->i_nlink) {
182 : : /*
183 : : * When journalling data dirty buffers are tracked only in the
184 : : * journal. So although mm thinks everything is clean and
185 : : * ready for reaping the inode might still have some pages to
186 : : * write in the running transaction or waiting to be
187 : : * checkpointed. Thus calling jbd2_journal_invalidatepage()
188 : : * (via truncate_inode_pages()) to discard these buffers can
189 : : * cause data loss. Also even if we did not discard these
190 : : * buffers, we would have no way to find them after the inode
191 : : * is reaped and thus user could see stale data if he tries to
192 : : * read them before the transaction is checkpointed. So be
193 : : * careful and force everything to disk here... We use
194 : : * ei->i_datasync_tid to store the newest transaction
195 : : * containing inode's data.
196 : : *
197 : : * Note that directories do not have this problem because they
198 : : * don't use page cache.
199 : : */
200 [ # # # # ]: 0 : if (inode->i_ino != EXT4_JOURNAL_INO &&
201 : 0 : ext4_should_journal_data(inode) &&
202 [ # # ]: 0 : (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) &&
203 [ # # ]: 0 : inode->i_data.nrpages) {
204 : 0 : journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
205 : 0 : tid_t commit_tid = EXT4_I(inode)->i_datasync_tid;
206 : :
207 : 0 : jbd2_complete_transaction(journal, commit_tid);
208 : 0 : filemap_write_and_wait(&inode->i_data);
209 : : }
210 : 0 : truncate_inode_pages_final(&inode->i_data);
211 : :
212 : 0 : goto no_delete;
213 : : }
214 : :
215 [ - + ]: 210 : if (is_bad_inode(inode))
216 : 0 : goto no_delete;
217 : 210 : dquot_initialize(inode);
218 : :
219 [ + + ]: 210 : if (ext4_should_order_data(inode))
220 : 168 : ext4_begin_ordered_truncate(inode, 0);
221 : 210 : truncate_inode_pages_final(&inode->i_data);
222 : :
223 : : /*
224 : : * Protect us against freezing - iput() caller didn't have to have any
225 : : * protection against it
226 : : */
227 : 210 : sb_start_intwrite(inode->i_sb);
228 : :
229 [ + - ]: 210 : if (!IS_NOQUOTA(inode))
230 [ + - - + ]: 210 : extra_credits += EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb);
231 : :
232 : : /*
233 : : * Block bitmap, group descriptor, and inode are accounted in both
234 : : * ext4_blocks_for_truncate() and extra_credits. So subtract 3.
235 : : */
236 : 210 : handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE,
237 : : ext4_blocks_for_truncate(inode) + extra_credits - 3);
238 [ - + ]: 210 : if (IS_ERR(handle)) {
239 [ # # ]: 0 : ext4_std_error(inode->i_sb, PTR_ERR(handle));
240 : : /*
241 : : * If we're going to skip the normal cleanup, we still need to
242 : : * make sure that the in-core orphan linked list is properly
243 : : * cleaned up.
244 : : */
245 : 0 : ext4_orphan_del(NULL, inode);
246 : 0 : sb_end_intwrite(inode->i_sb);
247 : 0 : goto no_delete;
248 : : }
249 : :
250 [ + - - + ]: 210 : if (IS_SYNC(inode))
251 [ # # ]: 0 : ext4_handle_sync(handle);
252 : :
253 : : /*
254 : : * Set inode->i_size to 0 before calling ext4_truncate(). We need
255 : : * special handling of symlinks here because i_size is used to
256 : : * determine whether ext4_inode_info->i_data contains symlink data or
257 : : * block mappings. Setting i_size to 0 will remove its fast symlink
258 : : * status. Erase i_data so that it becomes a valid empty block map.
259 : : */
260 [ - + ]: 210 : if (ext4_inode_is_fast_symlink(inode))
261 : 0 : memset(EXT4_I(inode)->i_data, 0, sizeof(EXT4_I(inode)->i_data));
262 : 210 : inode->i_size = 0;
263 : 210 : err = ext4_mark_inode_dirty(handle, inode);
264 [ - + ]: 210 : if (err) {
265 : 0 : ext4_warning(inode->i_sb,
266 : : "couldn't mark inode dirty (err %d)", err);
267 : 0 : goto stop_handle;
268 : : }
269 [ + - ]: 210 : if (inode->i_blocks) {
270 : 210 : err = ext4_truncate(inode);
271 [ - + ]: 210 : if (err) {
272 : 0 : ext4_set_errno(inode->i_sb, -err);
273 : 0 : ext4_error(inode->i_sb,
274 : : "couldn't truncate inode %lu (err %d)",
275 : : inode->i_ino, err);
276 : 0 : goto stop_handle;
277 : : }
278 : : }
279 : :
280 : : /* Remove xattr references. */
281 : 210 : err = ext4_xattr_delete_inode(handle, inode, &ea_inode_array,
282 : : extra_credits);
283 [ - + ]: 210 : if (err) {
284 : 0 : ext4_warning(inode->i_sb, "xattr delete (err %d)", err);
285 : 0 : stop_handle:
286 : 0 : ext4_journal_stop(handle);
287 : 0 : ext4_orphan_del(NULL, inode);
288 : 0 : sb_end_intwrite(inode->i_sb);
289 : 0 : ext4_xattr_inode_array_free(ea_inode_array);
290 : 0 : goto no_delete;
291 : : }
292 : :
293 : : /*
294 : : * Kill off the orphan record which ext4_truncate created.
295 : : * AKPM: I think this can be inside the above `if'.
296 : : * Note that ext4_orphan_del() has to be able to cope with the
297 : : * deletion of a non-existent orphan - this is because we don't
298 : : * know if ext4_truncate() actually created an orphan record.
299 : : * (Well, we could do this if we need to, but heck - it works)
300 : : */
301 : 210 : ext4_orphan_del(handle, inode);
302 : 210 : EXT4_I(inode)->i_dtime = (__u32)ktime_get_real_seconds();
303 : :
304 : : /*
305 : : * One subtle ordering requirement: if anything has gone wrong
306 : : * (transaction abort, IO errors, whatever), then we can still
307 : : * do these next steps (the fs will already have been marked as
308 : : * having errors), but we can't free the inode if the mark_dirty
309 : : * fails.
310 : : */
311 [ - + ]: 210 : if (ext4_mark_inode_dirty(handle, inode))
312 : : /* If that failed, just do the required in-core inode clear. */
313 : 0 : ext4_clear_inode(inode);
314 : : else
315 : 210 : ext4_free_inode(handle, inode);
316 : 210 : ext4_journal_stop(handle);
317 : 210 : sb_end_intwrite(inode->i_sb);
318 : 210 : ext4_xattr_inode_array_free(ea_inode_array);
319 : 210 : return;
320 : 0 : no_delete:
321 : 0 : ext4_clear_inode(inode); /* We must guarantee clearing of inode... */
322 : : }
323 : :
324 : : #ifdef CONFIG_QUOTA
325 : 15484 : qsize_t *ext4_get_reserved_space(struct inode *inode)
326 : : {
327 : 15484 : return &EXT4_I(inode)->i_reserved_quota;
328 : : }
329 : : #endif
330 : :
331 : : /*
332 : : * Called with i_data_sem down, which is important since we can call
333 : : * ext4_discard_preallocations() from here.
334 : : */
335 : 168 : void ext4_da_update_reserve_space(struct inode *inode,
336 : : int used, int quota_claim)
337 : : {
338 : 168 : struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
339 : 168 : struct ext4_inode_info *ei = EXT4_I(inode);
340 : :
341 : 168 : spin_lock(&ei->i_block_reservation_lock);
342 : 168 : trace_ext4_da_update_reserve_space(inode, used, quota_claim);
343 [ - + ]: 168 : if (unlikely(used > ei->i_reserved_data_blocks)) {
344 : 0 : ext4_warning(inode->i_sb, "%s: ino %lu, used %d "
345 : : "with only %d reserved data blocks",
346 : : __func__, inode->i_ino, used,
347 : : ei->i_reserved_data_blocks);
348 : 0 : WARN_ON(1);
349 : 0 : used = ei->i_reserved_data_blocks;
350 : : }
351 : :
352 : : /* Update per-inode reservations */
353 : 168 : ei->i_reserved_data_blocks -= used;
354 : 168 : percpu_counter_sub(&sbi->s_dirtyclusters_counter, used);
355 : :
356 : 168 : spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
357 : :
358 : : /* Update quota subsystem for data blocks */
359 [ + - ]: 168 : if (quota_claim)
360 : 168 : dquot_claim_block(inode, EXT4_C2B(sbi, used));
361 : : else {
362 : : /*
363 : : * We did fallocate with an offset that is already delayed
364 : : * allocated. So on delayed allocated writeback we should
365 : : * not re-claim the quota for fallocated blocks.
366 : : */
367 : 0 : dquot_release_reservation_block(inode, EXT4_C2B(sbi, used));
368 : : }
369 : :
370 : : /*
371 : : * If we have done all the pending block allocations and if
372 : : * there aren't any writers on the inode, we can discard the
373 : : * inode's preallocations.
374 : : */
375 [ + - - + ]: 336 : if ((ei->i_reserved_data_blocks == 0) &&
376 : : !inode_is_open_for_write(inode))
377 : 0 : ext4_discard_preallocations(inode);
378 : 168 : }
379 : :
380 : 95702 : static int __check_block_validity(struct inode *inode, const char *func,
381 : : unsigned int line,
382 : : struct ext4_map_blocks *map)
383 : : {
384 [ + - ]: 95702 : if (ext4_has_feature_journal(inode->i_sb) &&
385 [ + + ]: 95702 : (inode->i_ino ==
386 [ + + ]: 95702 : le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_journal_inum)))
387 : : return 0;
388 [ - + ]: 93503 : if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk,
389 : : map->m_len)) {
390 : 0 : ext4_error_inode(inode, func, line, map->m_pblk,
391 : : "lblock %lu mapped to illegal pblock %llu "
392 : : "(length %d)", (unsigned long) map->m_lblk,
393 : : map->m_pblk, map->m_len);
394 : 0 : return -EFSCORRUPTED;
395 : : }
396 : : return 0;
397 : : }
398 : :
399 : 0 : int ext4_issue_zeroout(struct inode *inode, ext4_lblk_t lblk, ext4_fsblk_t pblk,
400 : : ext4_lblk_t len)
401 : : {
402 : 0 : int ret;
403 : :
404 [ # # # # ]: 0 : if (IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode))
405 : : return fscrypt_zeroout_range(inode, lblk, pblk, len);
406 : :
407 : 0 : ret = sb_issue_zeroout(inode->i_sb, pblk, len, GFP_NOFS);
408 : 0 : if (ret > 0)
409 : : ret = 0;
410 : :
411 : : return ret;
412 : : }
413 : :
414 : : #define check_block_validity(inode, map) \
415 : : __check_block_validity((inode), __func__, __LINE__, (map))
416 : :
417 : : #ifdef ES_AGGRESSIVE_TEST
418 : : static void ext4_map_blocks_es_recheck(handle_t *handle,
419 : : struct inode *inode,
420 : : struct ext4_map_blocks *es_map,
421 : : struct ext4_map_blocks *map,
422 : : int flags)
423 : : {
424 : : int retval;
425 : :
426 : : map->m_flags = 0;
427 : : /*
428 : : * There is a race window that the result is not the same.
429 : : * e.g. xfstests #223 when dioread_nolock enables. The reason
430 : : * is that we lookup a block mapping in extent status tree with
431 : : * out taking i_data_sem. So at the time the unwritten extent
432 : : * could be converted.
433 : : */
434 : : down_read(&EXT4_I(inode)->i_data_sem);
435 : : if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
436 : : retval = ext4_ext_map_blocks(handle, inode, map, flags &
437 : : EXT4_GET_BLOCKS_KEEP_SIZE);
438 : : } else {
439 : : retval = ext4_ind_map_blocks(handle, inode, map, flags &
440 : : EXT4_GET_BLOCKS_KEEP_SIZE);
441 : : }
442 : : up_read((&EXT4_I(inode)->i_data_sem));
443 : :
444 : : /*
445 : : * We don't check m_len because extent will be collpased in status
446 : : * tree. So the m_len might not equal.
447 : : */
448 : : if (es_map->m_lblk != map->m_lblk ||
449 : : es_map->m_flags != map->m_flags ||
450 : : es_map->m_pblk != map->m_pblk) {
451 : : printk("ES cache assertion failed for inode: %lu "
452 : : "es_cached ex [%d/%d/%llu/%x] != "
453 : : "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
454 : : inode->i_ino, es_map->m_lblk, es_map->m_len,
455 : : es_map->m_pblk, es_map->m_flags, map->m_lblk,
456 : : map->m_len, map->m_pblk, map->m_flags,
457 : : retval, flags);
458 : : }
459 : : }
460 : : #endif /* ES_AGGRESSIVE_TEST */
461 : :
462 : : /*
463 : : * The ext4_map_blocks() function tries to look up the requested blocks,
464 : : * and returns if the blocks are already mapped.
465 : : *
466 : : * Otherwise it takes the write lock of the i_data_sem and allocate blocks
467 : : * and store the allocated blocks in the result buffer head and mark it
468 : : * mapped.
469 : : *
470 : : * If file type is extents based, it will call ext4_ext_map_blocks(),
471 : : * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
472 : : * based files
473 : : *
474 : : * On success, it returns the number of blocks being mapped or allocated. if
475 : : * create==0 and the blocks are pre-allocated and unwritten, the resulting @map
476 : : * is marked as unwritten. If the create == 1, it will mark @map as mapped.
477 : : *
478 : : * It returns 0 if plain look up failed (blocks have not been allocated), in
479 : : * that case, @map is returned as unmapped but we still do fill map->m_len to
480 : : * indicate the length of a hole starting at map->m_lblk.
481 : : *
482 : : * It returns the error in case of allocation failure.
483 : : */
484 : 96710 : int ext4_map_blocks(handle_t *handle, struct inode *inode,
485 : : struct ext4_map_blocks *map, int flags)
486 : : {
487 : 96710 : struct extent_status es;
488 : 96710 : int retval;
489 : 96710 : int ret = 0;
490 : : #ifdef ES_AGGRESSIVE_TEST
491 : : struct ext4_map_blocks orig_map;
492 : :
493 : : memcpy(&orig_map, map, sizeof(*map));
494 : : #endif
495 : :
496 : 96710 : map->m_flags = 0;
497 : 96710 : ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
498 : : "logical block %lu\n", inode->i_ino, flags, map->m_len,
499 : : (unsigned long) map->m_lblk);
500 : :
501 : : /*
502 : : * ext4_map_blocks returns an int, and m_len is an unsigned int
503 : : */
504 [ - + ]: 96710 : if (unlikely(map->m_len > INT_MAX))
505 : 0 : map->m_len = INT_MAX;
506 : :
507 : : /* We can handle the block number less than EXT_MAX_BLOCKS */
508 [ + - ]: 96710 : if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS))
509 : : return -EFSCORRUPTED;
510 : :
511 : : /* Lookup extent status tree firstly */
512 [ + + ]: 96710 : if (ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) {
513 [ + + + + ]: 82577 : if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) {
514 [ + + ]: 81422 : map->m_pblk = ext4_es_pblock(&es) +
515 : 81422 : map->m_lblk - es.es_lblk;
516 [ + + ]: 81422 : map->m_flags |= ext4_es_is_written(&es) ?
517 [ + + ]: 81422 : EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN;
518 : 81422 : retval = es.es_len - (map->m_lblk - es.es_lblk);
519 : 81422 : if (retval > map->m_len)
520 : : retval = map->m_len;
521 : 81422 : map->m_len = retval;
522 [ + + + - ]: 1155 : } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) {
523 : 1155 : map->m_pblk = 0;
524 : 1155 : retval = es.es_len - (map->m_lblk - es.es_lblk);
525 : 1155 : if (retval > map->m_len)
526 : : retval = map->m_len;
527 : 1155 : map->m_len = retval;
528 : 1155 : retval = 0;
529 : : } else {
530 : 0 : BUG();
531 : : }
532 : : #ifdef ES_AGGRESSIVE_TEST
533 : : ext4_map_blocks_es_recheck(handle, inode, map,
534 : : &orig_map, flags);
535 : : #endif
536 : 81422 : goto found;
537 : : }
538 : :
539 : : /*
540 : : * Try to see if we can get the block without requesting a new
541 : : * file system block.
542 : : */
543 : 14133 : down_read(&EXT4_I(inode)->i_data_sem);
544 [ + - ]: 14133 : if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
545 : 14133 : retval = ext4_ext_map_blocks(handle, inode, map, flags &
546 : : EXT4_GET_BLOCKS_KEEP_SIZE);
547 : : } else {
548 : 0 : retval = ext4_ind_map_blocks(handle, inode, map, flags &
549 : : EXT4_GET_BLOCKS_KEEP_SIZE);
550 : : }
551 [ + + ]: 14133 : if (retval > 0) {
552 : 8862 : unsigned int status;
553 : :
554 [ - + ]: 8862 : if (unlikely(retval != map->m_len)) {
555 : 0 : ext4_warning(inode->i_sb,
556 : : "ES len assertion failed for inode "
557 : : "%lu: retval %d != map->m_len %d",
558 : : inode->i_ino, retval, map->m_len);
559 : 0 : WARN_ON(1);
560 : : }
561 : :
562 : 8862 : status = map->m_flags & EXT4_MAP_UNWRITTEN ?
563 [ + - ]: 8862 : EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
564 [ + - ]: 8862 : if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
565 [ - + - - ]: 8862 : !(status & EXTENT_STATUS_WRITTEN) &&
566 : 0 : ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk,
567 : 0 : map->m_lblk + map->m_len - 1))
568 : 0 : status |= EXTENT_STATUS_DELAYED;
569 : 8862 : ret = ext4_es_insert_extent(inode, map->m_lblk,
570 : : map->m_len, map->m_pblk, status);
571 [ - + ]: 8862 : if (ret < 0)
572 : 0 : retval = ret;
573 : : }
574 : 14133 : up_read((&EXT4_I(inode)->i_data_sem));
575 : :
576 : 95555 : found:
577 [ + + + + ]: 96710 : if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
578 : 90116 : ret = check_block_validity(inode, map);
579 [ + - ]: 90116 : if (ret != 0)
580 : : return ret;
581 : : }
582 : :
583 : : /* If it is only a block(s) look up */
584 [ + + ]: 96710 : if ((flags & EXT4_GET_BLOCKS_CREATE) == 0)
585 : : return retval;
586 : :
587 : : /*
588 : : * Returns if the blocks have already allocated
589 : : *
590 : : * Note that if blocks have been preallocated
591 : : * ext4_ext_get_block() returns the create = 0
592 : : * with buffer head unmapped.
593 : : */
594 [ + + - + ]: 5586 : if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED)
595 : : /*
596 : : * If we need to convert extent to unwritten
597 : : * we continue and do the actual work in
598 : : * ext4_ext_map_blocks()
599 : : */
600 [ # # ]: 0 : if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN))
601 : : return retval;
602 : :
603 : : /*
604 : : * Here we clear m_flags because after allocating an new extent,
605 : : * it will be set again.
606 : : */
607 : 5586 : map->m_flags &= ~EXT4_MAP_FLAGS;
608 : :
609 : : /*
610 : : * New blocks allocate and/or writing to unwritten extent
611 : : * will possibly result in updating i_data, so we take
612 : : * the write lock of i_data_sem, and call get_block()
613 : : * with create == 1 flag.
614 : : */
615 : 5586 : down_write(&EXT4_I(inode)->i_data_sem);
616 : :
617 : : /*
618 : : * We need to check for EXT4 here because migrate
619 : : * could have changed the inode type in between
620 : : */
621 [ + - ]: 5586 : if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
622 : 5586 : retval = ext4_ext_map_blocks(handle, inode, map, flags);
623 : : } else {
624 : 0 : retval = ext4_ind_map_blocks(handle, inode, map, flags);
625 : :
626 [ # # # # ]: 0 : if (retval > 0 && map->m_flags & EXT4_MAP_NEW) {
627 : : /*
628 : : * We allocated new blocks which will result in
629 : : * i_data's format changing. Force the migrate
630 : : * to fail by clearing migrate flags
631 : : */
632 : 0 : ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE);
633 : : }
634 : :
635 : : /*
636 : : * Update reserved blocks/metadata blocks after successful
637 : : * block allocation which had been deferred till now. We don't
638 : : * support fallocate for non extent files. So we can update
639 : : * reserve space here.
640 : : */
641 [ # # ]: 0 : if ((retval > 0) &&
642 [ # # ]: 0 : (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE))
643 : 0 : ext4_da_update_reserve_space(inode, retval, 1);
644 : : }
645 : :
646 [ - + ]: 5586 : if (retval > 0) {
647 : 5586 : unsigned int status;
648 : :
649 [ - + ]: 5586 : if (unlikely(retval != map->m_len)) {
650 : 0 : ext4_warning(inode->i_sb,
651 : : "ES len assertion failed for inode "
652 : : "%lu: retval %d != map->m_len %d",
653 : : inode->i_ino, retval, map->m_len);
654 : 0 : WARN_ON(1);
655 : : }
656 : :
657 : : /*
658 : : * We have to zeroout blocks before inserting them into extent
659 : : * status tree. Otherwise someone could look them up there and
660 : : * use them before they are really zeroed. We also have to
661 : : * unmap metadata before zeroing as otherwise writeback can
662 : : * overwrite zeros with stale data from block device.
663 : : */
664 [ - + ]: 5586 : if (flags & EXT4_GET_BLOCKS_ZERO &&
665 [ # # ]: 0 : map->m_flags & EXT4_MAP_MAPPED &&
666 : : map->m_flags & EXT4_MAP_NEW) {
667 : 0 : ret = ext4_issue_zeroout(inode, map->m_lblk,
668 : : map->m_pblk, map->m_len);
669 [ # # ]: 0 : if (ret) {
670 : 0 : retval = ret;
671 : 0 : goto out_sem;
672 : : }
673 : : }
674 : :
675 : : /*
676 : : * If the extent has been zeroed out, we don't need to update
677 : : * extent status tree.
678 : : */
679 [ + + + - ]: 5754 : if ((flags & EXT4_GET_BLOCKS_PRE_IO) &&
680 : 168 : ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) {
681 [ - + ]: 168 : if (ext4_es_is_written(&es))
682 : 0 : goto out_sem;
683 : : }
684 : 5586 : status = map->m_flags & EXT4_MAP_UNWRITTEN ?
685 [ + + ]: 5586 : EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
686 [ + + ]: 5586 : if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) &&
687 [ - + - - ]: 5418 : !(status & EXTENT_STATUS_WRITTEN) &&
688 : 0 : ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk,
689 : 0 : map->m_lblk + map->m_len - 1))
690 : 0 : status |= EXTENT_STATUS_DELAYED;
691 : 5586 : ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
692 : : map->m_pblk, status);
693 [ + - ]: 5586 : if (ret < 0) {
694 : 0 : retval = ret;
695 : 0 : goto out_sem;
696 : : }
697 : : }
698 : :
699 : 5586 : out_sem:
700 : 5586 : up_write((&EXT4_I(inode)->i_data_sem));
701 [ + - + - ]: 5586 : if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) {
702 : 5586 : ret = check_block_validity(inode, map);
703 [ + - ]: 5586 : if (ret != 0)
704 : : return ret;
705 : :
706 : : /*
707 : : * Inodes with freshly allocated blocks where contents will be
708 : : * visible after transaction commit must be on transaction's
709 : : * ordered data list.
710 : : */
711 [ + + ]: 5586 : if (map->m_flags & EXT4_MAP_NEW &&
712 : 5250 : !(map->m_flags & EXT4_MAP_UNWRITTEN) &&
713 [ + - + - ]: 10500 : !(flags & EXT4_GET_BLOCKS_ZERO) &&
714 [ - + ]: 5250 : !ext4_is_quota_file(inode) &&
715 : : ext4_should_order_data(inode)) {
716 : 0 : loff_t start_byte =
717 : 0 : (loff_t)map->m_lblk << inode->i_blkbits;
718 : 0 : loff_t length = (loff_t)map->m_len << inode->i_blkbits;
719 : :
720 [ # # ]: 0 : if (flags & EXT4_GET_BLOCKS_IO_SUBMIT)
721 [ # # ]: 0 : ret = ext4_jbd2_inode_add_wait(handle, inode,
722 : : start_byte, length);
723 : : else
724 [ # # ]: 0 : ret = ext4_jbd2_inode_add_write(handle, inode,
725 : : start_byte, length);
726 [ # # ]: 0 : if (ret)
727 : 0 : return ret;
728 : : }
729 : : }
730 : : return retval;
731 : : }
732 : :
733 : : /*
734 : : * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
735 : : * we have to be careful as someone else may be manipulating b_state as well.
736 : : */
737 : 2157 : static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags)
738 : : {
739 : 2157 : unsigned long old_state;
740 : 2157 : unsigned long new_state;
741 : :
742 : 2157 : flags &= EXT4_MAP_FLAGS;
743 : :
744 : : /* Dummy buffer_head? Set non-atomically. */
745 [ + - ]: 2157 : if (!bh->b_page) {
746 : 2157 : bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags;
747 : 2157 : return;
748 : : }
749 : : /*
750 : : * Someone else may be modifying b_state. Be careful! This is ugly but
751 : : * once we get rid of using bh as a container for mapping information
752 : : * to pass to / from get_block functions, this can go away.
753 : : */
754 : 0 : do {
755 : 0 : old_state = READ_ONCE(bh->b_state);
756 : 0 : new_state = (old_state & ~EXT4_MAP_FLAGS) | flags;
757 [ # # ]: 0 : } while (unlikely(
758 : : cmpxchg(&bh->b_state, old_state, new_state) != old_state));
759 : : }
760 : :
761 : 2157 : static int _ext4_get_block(struct inode *inode, sector_t iblock,
762 : : struct buffer_head *bh, int flags)
763 : : {
764 : 2157 : struct ext4_map_blocks map;
765 : 2157 : int ret = 0;
766 : :
767 : 2157 : if (ext4_has_inline_data(inode))
768 : : return -ERANGE;
769 : :
770 : 2157 : map.m_lblk = iblock;
771 : 2157 : map.m_len = bh->b_size >> inode->i_blkbits;
772 : :
773 : 2157 : ret = ext4_map_blocks(ext4_journal_current_handle(), inode, &map,
774 : : flags);
775 [ + - ]: 2157 : if (ret > 0) {
776 : 2157 : map_bh(bh, inode->i_sb, map.m_pblk);
777 : 2157 : ext4_update_bh_state(bh, map.m_flags);
778 : 2157 : bh->b_size = inode->i_sb->s_blocksize * map.m_len;
779 : 2157 : ret = 0;
780 [ # # ]: 0 : } else if (ret == 0) {
781 : : /* hole case, need to fill in bh->b_size */
782 : 0 : bh->b_size = inode->i_sb->s_blocksize * map.m_len;
783 : : }
784 : : return ret;
785 : : }
786 : :
787 : 2157 : int ext4_get_block(struct inode *inode, sector_t iblock,
788 : : struct buffer_head *bh, int create)
789 : : {
790 : 2157 : return _ext4_get_block(inode, iblock, bh,
791 : : create ? EXT4_GET_BLOCKS_CREATE : 0);
792 : : }
793 : :
794 : : /*
795 : : * Get block function used when preparing for buffered write if we require
796 : : * creating an unwritten extent if blocks haven't been allocated. The extent
797 : : * will be converted to written after the IO is complete.
798 : : */
799 : 0 : int ext4_get_block_unwritten(struct inode *inode, sector_t iblock,
800 : : struct buffer_head *bh_result, int create)
801 : : {
802 : 0 : ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n",
803 : : inode->i_ino, create);
804 : 0 : return _ext4_get_block(inode, iblock, bh_result,
805 : : EXT4_GET_BLOCKS_IO_CREATE_EXT);
806 : : }
807 : :
808 : : /* Maximum number of blocks we map for direct IO at once. */
809 : : #define DIO_MAX_BLOCKS 4096
810 : :
811 : : /*
812 : : * `handle' can be NULL if create is zero
813 : : */
814 : 79121 : struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode,
815 : : ext4_lblk_t block, int map_flags)
816 : : {
817 : 79121 : struct ext4_map_blocks map;
818 : 79121 : struct buffer_head *bh;
819 : 79121 : int create = map_flags & EXT4_GET_BLOCKS_CREATE;
820 : 79121 : int err;
821 : :
822 [ - + ]: 79121 : J_ASSERT(handle != NULL || create == 0);
823 : :
824 : 79121 : map.m_lblk = block;
825 : 79121 : map.m_len = 1;
826 : 79121 : err = ext4_map_blocks(handle, inode, &map, map_flags);
827 : :
828 [ - + ]: 79121 : if (err == 0)
829 [ # # ]: 0 : return create ? ERR_PTR(-ENOSPC) : NULL;
830 [ - + ]: 79121 : if (err < 0)
831 : 0 : return ERR_PTR(err);
832 : :
833 : 79121 : bh = sb_getblk(inode->i_sb, map.m_pblk);
834 [ + - ]: 79121 : if (unlikely(!bh))
835 : : return ERR_PTR(-ENOMEM);
836 [ + + ]: 79121 : if (map.m_flags & EXT4_MAP_NEW) {
837 [ - + ]: 5250 : J_ASSERT(create != 0);
838 [ - + ]: 5250 : J_ASSERT(handle != NULL);
839 : :
840 : : /*
841 : : * Now that we do not always journal data, we should
842 : : * keep in mind whether this should always journal the
843 : : * new buffer as metadata. For now, regular file
844 : : * writes use ext4_get_block instead, so it's not a
845 : : * problem.
846 : : */
847 : 5250 : lock_buffer(bh);
848 : 5250 : BUFFER_TRACE(bh, "call get_create_access");
849 : 5250 : err = ext4_journal_get_create_access(handle, bh);
850 [ - + ]: 5250 : if (unlikely(err)) {
851 : 0 : unlock_buffer(bh);
852 : 0 : goto errout;
853 : : }
854 [ + - ]: 5250 : if (!buffer_uptodate(bh)) {
855 : 5250 : memset(bh->b_data, 0, inode->i_sb->s_blocksize);
856 : 5250 : set_buffer_uptodate(bh);
857 : : }
858 : 5250 : unlock_buffer(bh);
859 : 5250 : BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
860 : 5250 : err = ext4_handle_dirty_metadata(handle, inode, bh);
861 [ - + ]: 5250 : if (unlikely(err))
862 : 0 : goto errout;
863 : : } else
864 : : BUFFER_TRACE(bh, "not a new buffer");
865 : : return bh;
866 : 0 : errout:
867 [ # # ]: 0 : brelse(bh);
868 : 0 : return ERR_PTR(err);
869 : : }
870 : :
871 : 46570 : struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode,
872 : : ext4_lblk_t block, int map_flags)
873 : : {
874 : 46570 : struct buffer_head *bh;
875 : :
876 : 46570 : bh = ext4_getblk(handle, inode, block, map_flags);
877 [ + - ]: 46570 : if (IS_ERR(bh))
878 : : return bh;
879 [ + - + + ]: 46570 : if (!bh || ext4_buffer_uptodate(bh))
880 : 45352 : return bh;
881 : 1218 : ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &bh);
882 : 1218 : wait_on_buffer(bh);
883 [ + - ]: 1218 : if (buffer_uptodate(bh))
884 : 1218 : return bh;
885 : 0 : put_bh(bh);
886 : 0 : return ERR_PTR(-EIO);
887 : : }
888 : :
889 : : /* Read a contiguous batch of blocks. */
890 : 32551 : int ext4_bread_batch(struct inode *inode, ext4_lblk_t block, int bh_count,
891 : : bool wait, struct buffer_head **bhs)
892 : : {
893 : 32551 : int i, err;
894 : :
895 [ + + ]: 65102 : for (i = 0; i < bh_count; i++) {
896 : 32551 : bhs[i] = ext4_getblk(NULL, inode, block + i, 0 /* map_flags */);
897 [ - + ]: 32551 : if (IS_ERR(bhs[i])) {
898 : 0 : err = PTR_ERR(bhs[i]);
899 : 0 : bh_count = i;
900 : 0 : goto out_brelse;
901 : : }
902 : : }
903 : :
904 [ + + ]: 65102 : for (i = 0; i < bh_count; i++)
905 : : /* Note that NULL bhs[i] is valid because of holes. */
906 [ + - + + ]: 32551 : if (bhs[i] && !ext4_buffer_uptodate(bhs[i]))
907 : 1072 : ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1,
908 : : &bhs[i]);
909 : :
910 [ - + ]: 32551 : if (!wait)
911 : : return 0;
912 : :
913 [ # # ]: 0 : for (i = 0; i < bh_count; i++)
914 [ # # ]: 0 : if (bhs[i])
915 : 0 : wait_on_buffer(bhs[i]);
916 : :
917 [ # # ]: 0 : for (i = 0; i < bh_count; i++) {
918 [ # # # # ]: 0 : if (bhs[i] && !buffer_uptodate(bhs[i])) {
919 : 0 : err = -EIO;
920 : 0 : goto out_brelse;
921 : : }
922 : : }
923 : : return 0;
924 : :
925 : 0 : out_brelse:
926 [ # # ]: 0 : for (i = 0; i < bh_count; i++) {
927 [ # # ]: 0 : brelse(bhs[i]);
928 : 0 : bhs[i] = NULL;
929 : : }
930 : : return err;
931 : : }
932 : :
933 : 0 : int ext4_walk_page_buffers(handle_t *handle,
934 : : struct buffer_head *head,
935 : : unsigned from,
936 : : unsigned to,
937 : : int *partial,
938 : : int (*fn)(handle_t *handle,
939 : : struct buffer_head *bh))
940 : : {
941 : 0 : struct buffer_head *bh;
942 : 0 : unsigned block_start, block_end;
943 : 0 : unsigned blocksize = head->b_size;
944 : 0 : int err, ret = 0;
945 : 0 : struct buffer_head *next;
946 : :
947 : 0 : for (bh = head, block_start = 0;
948 [ # # # # ]: 0 : ret == 0 && (bh != head || !block_start);
949 : : block_start = block_end, bh = next) {
950 : 0 : next = bh->b_this_page;
951 : 0 : block_end = block_start + blocksize;
952 [ # # ]: 0 : if (block_end <= from || block_start >= to) {
953 [ # # # # ]: 0 : if (partial && !buffer_uptodate(bh))
954 : 0 : *partial = 1;
955 : 0 : continue;
956 : : }
957 : 0 : err = (*fn)(handle, bh);
958 : 0 : if (!ret)
959 : 0 : ret = err;
960 : : }
961 : 0 : return ret;
962 : : }
963 : :
964 : : /*
965 : : * To preserve ordering, it is essential that the hole instantiation and
966 : : * the data write be encapsulated in a single transaction. We cannot
967 : : * close off a transaction and start a new one between the ext4_get_block()
968 : : * and the commit_write(). So doing the jbd2_journal_start at the start of
969 : : * prepare_write() is the right place.
970 : : *
971 : : * Also, this function can nest inside ext4_writepage(). In that case, we
972 : : * *know* that ext4_writepage() has generated enough buffer credits to do the
973 : : * whole page. So we won't block on the journal in that case, which is good,
974 : : * because the caller may be PF_MEMALLOC.
975 : : *
976 : : * By accident, ext4 can be reentered when a transaction is open via
977 : : * quota file writes. If we were to commit the transaction while thus
978 : : * reentered, there can be a deadlock - we would be holding a quota
979 : : * lock, and the commit would never complete if another thread had a
980 : : * transaction open and was blocking on the quota lock - a ranking
981 : : * violation.
982 : : *
983 : : * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
984 : : * will _not_ run commit under these circumstances because handle->h_ref
985 : : * is elevated. We'll still have enough credits for the tiny quotafile
986 : : * write.
987 : : */
988 : 0 : int do_journal_get_write_access(handle_t *handle,
989 : : struct buffer_head *bh)
990 : : {
991 : 0 : int dirty = buffer_dirty(bh);
992 : 0 : int ret;
993 : :
994 [ # # # # ]: 0 : if (!buffer_mapped(bh) || buffer_freed(bh))
995 : 0 : return 0;
996 : : /*
997 : : * __block_write_begin() could have dirtied some buffers. Clean
998 : : * the dirty bit as jbd2_journal_get_write_access() could complain
999 : : * otherwise about fs integrity issues. Setting of the dirty bit
1000 : : * by __block_write_begin() isn't a real problem here as we clear
1001 : : * the bit before releasing a page lock and thus writeback cannot
1002 : : * ever write the buffer.
1003 : : */
1004 [ # # ]: 0 : if (dirty)
1005 : 0 : clear_buffer_dirty(bh);
1006 : 0 : BUFFER_TRACE(bh, "get write access");
1007 : 0 : ret = ext4_journal_get_write_access(handle, bh);
1008 [ # # ]: 0 : if (!ret && dirty)
1009 : 0 : ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1010 : : return ret;
1011 : : }
1012 : :
1013 : : #ifdef CONFIG_FS_ENCRYPTION
1014 : : static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len,
1015 : : get_block_t *get_block)
1016 : : {
1017 : : unsigned from = pos & (PAGE_SIZE - 1);
1018 : : unsigned to = from + len;
1019 : : struct inode *inode = page->mapping->host;
1020 : : unsigned block_start, block_end;
1021 : : sector_t block;
1022 : : int err = 0;
1023 : : unsigned blocksize = inode->i_sb->s_blocksize;
1024 : : unsigned bbits;
1025 : : struct buffer_head *bh, *head, *wait[2];
1026 : : int nr_wait = 0;
1027 : : int i;
1028 : :
1029 : : BUG_ON(!PageLocked(page));
1030 : : BUG_ON(from > PAGE_SIZE);
1031 : : BUG_ON(to > PAGE_SIZE);
1032 : : BUG_ON(from > to);
1033 : :
1034 : : if (!page_has_buffers(page))
1035 : : create_empty_buffers(page, blocksize, 0);
1036 : : head = page_buffers(page);
1037 : : bbits = ilog2(blocksize);
1038 : : block = (sector_t)page->index << (PAGE_SHIFT - bbits);
1039 : :
1040 : : for (bh = head, block_start = 0; bh != head || !block_start;
1041 : : block++, block_start = block_end, bh = bh->b_this_page) {
1042 : : block_end = block_start + blocksize;
1043 : : if (block_end <= from || block_start >= to) {
1044 : : if (PageUptodate(page)) {
1045 : : if (!buffer_uptodate(bh))
1046 : : set_buffer_uptodate(bh);
1047 : : }
1048 : : continue;
1049 : : }
1050 : : if (buffer_new(bh))
1051 : : clear_buffer_new(bh);
1052 : : if (!buffer_mapped(bh)) {
1053 : : WARN_ON(bh->b_size != blocksize);
1054 : : err = get_block(inode, block, bh, 1);
1055 : : if (err)
1056 : : break;
1057 : : if (buffer_new(bh)) {
1058 : : if (PageUptodate(page)) {
1059 : : clear_buffer_new(bh);
1060 : : set_buffer_uptodate(bh);
1061 : : mark_buffer_dirty(bh);
1062 : : continue;
1063 : : }
1064 : : if (block_end > to || block_start < from)
1065 : : zero_user_segments(page, to, block_end,
1066 : : block_start, from);
1067 : : continue;
1068 : : }
1069 : : }
1070 : : if (PageUptodate(page)) {
1071 : : if (!buffer_uptodate(bh))
1072 : : set_buffer_uptodate(bh);
1073 : : continue;
1074 : : }
1075 : : if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
1076 : : !buffer_unwritten(bh) &&
1077 : : (block_start < from || block_end > to)) {
1078 : : ll_rw_block(REQ_OP_READ, 0, 1, &bh);
1079 : : wait[nr_wait++] = bh;
1080 : : }
1081 : : }
1082 : : /*
1083 : : * If we issued read requests, let them complete.
1084 : : */
1085 : : for (i = 0; i < nr_wait; i++) {
1086 : : wait_on_buffer(wait[i]);
1087 : : if (!buffer_uptodate(wait[i]))
1088 : : err = -EIO;
1089 : : }
1090 : : if (unlikely(err)) {
1091 : : page_zero_new_buffers(page, from, to);
1092 : : } else if (IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode)) {
1093 : : for (i = 0; i < nr_wait; i++) {
1094 : : int err2;
1095 : :
1096 : : err2 = fscrypt_decrypt_pagecache_blocks(page, blocksize,
1097 : : bh_offset(wait[i]));
1098 : : if (err2) {
1099 : : clear_buffer_uptodate(wait[i]);
1100 : : err = err2;
1101 : : }
1102 : : }
1103 : : }
1104 : :
1105 : : return err;
1106 : : }
1107 : : #endif
1108 : :
1109 : 0 : static int ext4_write_begin(struct file *file, struct address_space *mapping,
1110 : : loff_t pos, unsigned len, unsigned flags,
1111 : : struct page **pagep, void **fsdata)
1112 : : {
1113 : 0 : struct inode *inode = mapping->host;
1114 : 0 : int ret, needed_blocks;
1115 : 0 : handle_t *handle;
1116 : 0 : int retries = 0;
1117 : 0 : struct page *page;
1118 : 0 : pgoff_t index;
1119 : 0 : unsigned from, to;
1120 : :
1121 [ # # ]: 0 : if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
1122 : : return -EIO;
1123 : :
1124 : 0 : trace_ext4_write_begin(inode, pos, len, flags);
1125 : : /*
1126 : : * Reserve one block more for addition to orphan list in case
1127 : : * we allocate blocks but write fails for some reason
1128 : : */
1129 : 0 : needed_blocks = ext4_writepage_trans_blocks(inode) + 1;
1130 : 0 : index = pos >> PAGE_SHIFT;
1131 : 0 : from = pos & (PAGE_SIZE - 1);
1132 : 0 : to = from + len;
1133 : :
1134 [ # # ]: 0 : if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
1135 : 0 : ret = ext4_try_to_write_inline_data(mapping, inode, pos, len,
1136 : : flags, pagep);
1137 [ # # ]: 0 : if (ret < 0)
1138 : : return ret;
1139 [ # # ]: 0 : if (ret == 1)
1140 : : return 0;
1141 : : }
1142 : :
1143 : : /*
1144 : : * grab_cache_page_write_begin() can take a long time if the
1145 : : * system is thrashing due to memory pressure, or if the page
1146 : : * is being written back. So grab it first before we start
1147 : : * the transaction handle. This also allows us to allocate
1148 : : * the page (if needed) without using GFP_NOFS.
1149 : : */
1150 : 0 : retry_grab:
1151 : 0 : page = grab_cache_page_write_begin(mapping, index, flags);
1152 [ # # ]: 0 : if (!page)
1153 : : return -ENOMEM;
1154 : 0 : unlock_page(page);
1155 : :
1156 : 0 : retry_journal:
1157 : 0 : handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks);
1158 [ # # ]: 0 : if (IS_ERR(handle)) {
1159 : 0 : put_page(page);
1160 : 0 : return PTR_ERR(handle);
1161 : : }
1162 : :
1163 : 0 : lock_page(page);
1164 [ # # ]: 0 : if (page->mapping != mapping) {
1165 : : /* The page got truncated from under us */
1166 : 0 : unlock_page(page);
1167 : 0 : put_page(page);
1168 : 0 : ext4_journal_stop(handle);
1169 : 0 : goto retry_grab;
1170 : : }
1171 : : /* In case writeback began while the page was unlocked */
1172 : 0 : wait_for_stable_page(page);
1173 : :
1174 : : #ifdef CONFIG_FS_ENCRYPTION
1175 : : if (ext4_should_dioread_nolock(inode))
1176 : : ret = ext4_block_write_begin(page, pos, len,
1177 : : ext4_get_block_unwritten);
1178 : : else
1179 : : ret = ext4_block_write_begin(page, pos, len,
1180 : : ext4_get_block);
1181 : : #else
1182 [ # # ]: 0 : if (ext4_should_dioread_nolock(inode))
1183 : 0 : ret = __block_write_begin(page, pos, len,
1184 : : ext4_get_block_unwritten);
1185 : : else
1186 : 0 : ret = __block_write_begin(page, pos, len, ext4_get_block);
1187 : : #endif
1188 [ # # # # ]: 0 : if (!ret && ext4_should_journal_data(inode)) {
1189 [ # # ]: 0 : ret = ext4_walk_page_buffers(handle, page_buffers(page),
1190 : : from, to, NULL,
1191 : : do_journal_get_write_access);
1192 : : }
1193 : :
1194 [ # # ]: 0 : if (ret) {
1195 [ # # ]: 0 : bool extended = (pos + len > inode->i_size) &&
1196 : : !ext4_verity_in_progress(inode);
1197 : :
1198 : 0 : unlock_page(page);
1199 : : /*
1200 : : * __block_write_begin may have instantiated a few blocks
1201 : : * outside i_size. Trim these off again. Don't need
1202 : : * i_size_read because we hold i_mutex.
1203 : : *
1204 : : * Add inode to orphan list in case we crash before
1205 : : * truncate finishes
1206 : : */
1207 [ # # # # ]: 0 : if (extended && ext4_can_truncate(inode))
1208 : 0 : ext4_orphan_add(handle, inode);
1209 : :
1210 : 0 : ext4_journal_stop(handle);
1211 [ # # ]: 0 : if (extended) {
1212 : 0 : ext4_truncate_failed_write(inode);
1213 : : /*
1214 : : * If truncate failed early the inode might
1215 : : * still be on the orphan list; we need to
1216 : : * make sure the inode is removed from the
1217 : : * orphan list in that case.
1218 : : */
1219 [ # # ]: 0 : if (inode->i_nlink)
1220 : 0 : ext4_orphan_del(NULL, inode);
1221 : : }
1222 : :
1223 [ # # # # ]: 0 : if (ret == -ENOSPC &&
1224 : 0 : ext4_should_retry_alloc(inode->i_sb, &retries))
1225 : 0 : goto retry_journal;
1226 : 0 : put_page(page);
1227 : 0 : return ret;
1228 : : }
1229 : 0 : *pagep = page;
1230 : 0 : return ret;
1231 : : }
1232 : :
1233 : : /* For write_end() in data=journal mode */
1234 : 0 : static int write_end_fn(handle_t *handle, struct buffer_head *bh)
1235 : : {
1236 : 0 : int ret;
1237 [ # # # # ]: 0 : if (!buffer_mapped(bh) || buffer_freed(bh))
1238 : 0 : return 0;
1239 : 0 : set_buffer_uptodate(bh);
1240 : 0 : ret = ext4_handle_dirty_metadata(handle, NULL, bh);
1241 : 0 : clear_buffer_meta(bh);
1242 : 0 : clear_buffer_prio(bh);
1243 : 0 : return ret;
1244 : : }
1245 : :
1246 : : /*
1247 : : * We need to pick up the new inode size which generic_commit_write gave us
1248 : : * `file' can be NULL - eg, when called from page_symlink().
1249 : : *
1250 : : * ext4 never places buffers on inode->i_mapping->private_list. metadata
1251 : : * buffers are managed internally.
1252 : : */
1253 : 0 : static int ext4_write_end(struct file *file,
1254 : : struct address_space *mapping,
1255 : : loff_t pos, unsigned len, unsigned copied,
1256 : : struct page *page, void *fsdata)
1257 : : {
1258 : 0 : handle_t *handle = ext4_journal_current_handle();
1259 : 0 : struct inode *inode = mapping->host;
1260 : 0 : loff_t old_size = inode->i_size;
1261 : 0 : int ret = 0, ret2;
1262 : 0 : int i_size_changed = 0;
1263 : 0 : int inline_data = ext4_has_inline_data(inode);
1264 : 0 : bool verity = ext4_verity_in_progress(inode);
1265 : :
1266 : 0 : trace_ext4_write_end(inode, pos, len, copied);
1267 [ # # ]: 0 : if (inline_data) {
1268 : 0 : ret = ext4_write_inline_data_end(inode, pos, len,
1269 : : copied, page);
1270 [ # # ]: 0 : if (ret < 0) {
1271 : 0 : unlock_page(page);
1272 : 0 : put_page(page);
1273 : 0 : goto errout;
1274 : : }
1275 : 0 : copied = ret;
1276 : : } else
1277 : 0 : copied = block_write_end(file, mapping, pos,
1278 : : len, copied, page, fsdata);
1279 : : /*
1280 : : * it's important to update i_size while still holding page lock:
1281 : : * page writeout could otherwise come in and zero beyond i_size.
1282 : : *
1283 : : * If FS_IOC_ENABLE_VERITY is running on this inode, then Merkle tree
1284 : : * blocks are being written past EOF, so skip the i_size update.
1285 : : */
1286 : 0 : if (!verity)
1287 [ # # ]: 0 : i_size_changed = ext4_update_inode_size(inode, pos + copied);
1288 : 0 : unlock_page(page);
1289 : 0 : put_page(page);
1290 : :
1291 [ # # ]: 0 : if (old_size < pos && !verity)
1292 : 0 : pagecache_isize_extended(inode, old_size, pos);
1293 : : /*
1294 : : * Don't mark the inode dirty under page lock. First, it unnecessarily
1295 : : * makes the holding time of page lock longer. Second, it forces lock
1296 : : * ordering of page lock and transaction start for journaling
1297 : : * filesystems.
1298 : : */
1299 [ # # ]: 0 : if (i_size_changed || inline_data)
1300 : 0 : ext4_mark_inode_dirty(handle, inode);
1301 : :
1302 [ # # # # ]: 0 : if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode))
1303 : : /* if we have allocated more blocks and copied
1304 : : * less. We will have blocks allocated outside
1305 : : * inode->i_size. So truncate them
1306 : : */
1307 : 0 : ext4_orphan_add(handle, inode);
1308 : 0 : errout:
1309 : 0 : ret2 = ext4_journal_stop(handle);
1310 [ # # ]: 0 : if (!ret)
1311 : 0 : ret = ret2;
1312 : :
1313 [ # # ]: 0 : if (pos + len > inode->i_size && !verity) {
1314 : 0 : ext4_truncate_failed_write(inode);
1315 : : /*
1316 : : * If truncate failed early the inode might still be
1317 : : * on the orphan list; we need to make sure the inode
1318 : : * is removed from the orphan list in that case.
1319 : : */
1320 [ # # ]: 0 : if (inode->i_nlink)
1321 : 0 : ext4_orphan_del(NULL, inode);
1322 : : }
1323 : :
1324 [ # # ]: 0 : return ret ? ret : copied;
1325 : : }
1326 : :
1327 : : /*
1328 : : * This is a private version of page_zero_new_buffers() which doesn't
1329 : : * set the buffer to be dirty, since in data=journalled mode we need
1330 : : * to call ext4_handle_dirty_metadata() instead.
1331 : : */
1332 : 0 : static void ext4_journalled_zero_new_buffers(handle_t *handle,
1333 : : struct page *page,
1334 : : unsigned from, unsigned to)
1335 : : {
1336 : 0 : unsigned int block_start = 0, block_end;
1337 : 0 : struct buffer_head *head, *bh;
1338 : :
1339 [ # # ]: 0 : bh = head = page_buffers(page);
1340 : 0 : do {
1341 : 0 : block_end = block_start + bh->b_size;
1342 [ # # ]: 0 : if (buffer_new(bh)) {
1343 [ # # ]: 0 : if (block_end > from && block_start < to) {
1344 [ # # ]: 0 : if (!PageUptodate(page)) {
1345 : 0 : unsigned start, size;
1346 : :
1347 : 0 : start = max(from, block_start);
1348 : 0 : size = min(to, block_end) - start;
1349 : :
1350 : 0 : zero_user(page, start, size);
1351 : 0 : write_end_fn(handle, bh);
1352 : : }
1353 : 0 : clear_buffer_new(bh);
1354 : : }
1355 : : }
1356 : 0 : block_start = block_end;
1357 : 0 : bh = bh->b_this_page;
1358 [ # # ]: 0 : } while (bh != head);
1359 : 0 : }
1360 : :
1361 : 0 : static int ext4_journalled_write_end(struct file *file,
1362 : : struct address_space *mapping,
1363 : : loff_t pos, unsigned len, unsigned copied,
1364 : : struct page *page, void *fsdata)
1365 : : {
1366 : 0 : handle_t *handle = ext4_journal_current_handle();
1367 : 0 : struct inode *inode = mapping->host;
1368 : 0 : loff_t old_size = inode->i_size;
1369 : 0 : int ret = 0, ret2;
1370 : 0 : int partial = 0;
1371 : 0 : unsigned from, to;
1372 : 0 : int size_changed = 0;
1373 : 0 : int inline_data = ext4_has_inline_data(inode);
1374 : 0 : bool verity = ext4_verity_in_progress(inode);
1375 : :
1376 : 0 : trace_ext4_journalled_write_end(inode, pos, len, copied);
1377 : 0 : from = pos & (PAGE_SIZE - 1);
1378 : 0 : to = from + len;
1379 : :
1380 [ # # # # ]: 0 : BUG_ON(!ext4_handle_valid(handle));
1381 : :
1382 [ # # ]: 0 : if (inline_data) {
1383 : 0 : ret = ext4_write_inline_data_end(inode, pos, len,
1384 : : copied, page);
1385 [ # # ]: 0 : if (ret < 0) {
1386 : 0 : unlock_page(page);
1387 : 0 : put_page(page);
1388 : 0 : goto errout;
1389 : : }
1390 : 0 : copied = ret;
1391 [ # # # # ]: 0 : } else if (unlikely(copied < len) && !PageUptodate(page)) {
1392 : 0 : copied = 0;
1393 : 0 : ext4_journalled_zero_new_buffers(handle, page, from, to);
1394 : : } else {
1395 [ # # ]: 0 : if (unlikely(copied < len))
1396 : 0 : ext4_journalled_zero_new_buffers(handle, page,
1397 : : from + copied, to);
1398 [ # # ]: 0 : ret = ext4_walk_page_buffers(handle, page_buffers(page), from,
1399 : : from + copied, &partial,
1400 : : write_end_fn);
1401 [ # # ]: 0 : if (!partial)
1402 : 0 : SetPageUptodate(page);
1403 : : }
1404 : 0 : if (!verity)
1405 [ # # ]: 0 : size_changed = ext4_update_inode_size(inode, pos + copied);
1406 : 0 : ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1407 : 0 : EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1408 : 0 : unlock_page(page);
1409 : 0 : put_page(page);
1410 : :
1411 [ # # ]: 0 : if (old_size < pos && !verity)
1412 : 0 : pagecache_isize_extended(inode, old_size, pos);
1413 : :
1414 [ # # ]: 0 : if (size_changed || inline_data) {
1415 : 0 : ret2 = ext4_mark_inode_dirty(handle, inode);
1416 [ # # ]: 0 : if (!ret)
1417 : 0 : ret = ret2;
1418 : : }
1419 : :
1420 [ # # # # ]: 0 : if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode))
1421 : : /* if we have allocated more blocks and copied
1422 : : * less. We will have blocks allocated outside
1423 : : * inode->i_size. So truncate them
1424 : : */
1425 : 0 : ext4_orphan_add(handle, inode);
1426 : :
1427 : 0 : errout:
1428 : 0 : ret2 = ext4_journal_stop(handle);
1429 [ # # ]: 0 : if (!ret)
1430 : 0 : ret = ret2;
1431 [ # # ]: 0 : if (pos + len > inode->i_size && !verity) {
1432 : 0 : ext4_truncate_failed_write(inode);
1433 : : /*
1434 : : * If truncate failed early the inode might still be
1435 : : * on the orphan list; we need to make sure the inode
1436 : : * is removed from the orphan list in that case.
1437 : : */
1438 [ # # ]: 0 : if (inode->i_nlink)
1439 : 0 : ext4_orphan_del(NULL, inode);
1440 : : }
1441 : :
1442 [ # # ]: 0 : return ret ? ret : copied;
1443 : : }
1444 : :
1445 : : /*
1446 : : * Reserve space for a single cluster
1447 : : */
1448 : 15316 : static int ext4_da_reserve_space(struct inode *inode)
1449 : : {
1450 : 15316 : struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1451 : 15316 : struct ext4_inode_info *ei = EXT4_I(inode);
1452 : 15316 : int ret;
1453 : :
1454 : : /*
1455 : : * We will charge metadata quota at writeout time; this saves
1456 : : * us from metadata over-estimation, though we may go over by
1457 : : * a small amount in the end. Here we just reserve for data.
1458 : : */
1459 : 15316 : ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1));
1460 [ + - ]: 15316 : if (ret)
1461 : : return ret;
1462 : :
1463 : 15316 : spin_lock(&ei->i_block_reservation_lock);
1464 [ - + ]: 15316 : if (ext4_claim_free_clusters(sbi, 1, 0)) {
1465 : 0 : spin_unlock(&ei->i_block_reservation_lock);
1466 : 0 : dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1));
1467 : 0 : return -ENOSPC;
1468 : : }
1469 : 15316 : ei->i_reserved_data_blocks++;
1470 : 15316 : trace_ext4_da_reserve_space(inode);
1471 : 15316 : spin_unlock(&ei->i_block_reservation_lock);
1472 : :
1473 : 15316 : return 0; /* success */
1474 : : }
1475 : :
1476 : 420 : void ext4_da_release_space(struct inode *inode, int to_free)
1477 : : {
1478 [ - + ]: 420 : struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1479 : 420 : struct ext4_inode_info *ei = EXT4_I(inode);
1480 : :
1481 [ - + ]: 420 : if (!to_free)
1482 : : return; /* Nothing to release, exit */
1483 : :
1484 : 0 : spin_lock(&EXT4_I(inode)->i_block_reservation_lock);
1485 : :
1486 : 0 : trace_ext4_da_release_space(inode, to_free);
1487 [ # # ]: 0 : if (unlikely(to_free > ei->i_reserved_data_blocks)) {
1488 : : /*
1489 : : * if there aren't enough reserved blocks, then the
1490 : : * counter is messed up somewhere. Since this
1491 : : * function is called from invalidate page, it's
1492 : : * harmless to return without any action.
1493 : : */
1494 : 0 : ext4_warning(inode->i_sb, "ext4_da_release_space: "
1495 : : "ino %lu, to_free %d with only %d reserved "
1496 : : "data blocks", inode->i_ino, to_free,
1497 : : ei->i_reserved_data_blocks);
1498 : 0 : WARN_ON(1);
1499 : 0 : to_free = ei->i_reserved_data_blocks;
1500 : : }
1501 : 0 : ei->i_reserved_data_blocks -= to_free;
1502 : :
1503 : : /* update fs dirty data blocks counter */
1504 : 0 : percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free);
1505 : :
1506 : 0 : spin_unlock(&EXT4_I(inode)->i_block_reservation_lock);
1507 : :
1508 : 0 : dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free));
1509 : : }
1510 : :
1511 : : /*
1512 : : * Delayed allocation stuff
1513 : : */
1514 : :
1515 : : struct mpage_da_data {
1516 : : struct inode *inode;
1517 : : struct writeback_control *wbc;
1518 : :
1519 : : pgoff_t first_page; /* The first page to write */
1520 : : pgoff_t next_page; /* Current page to examine */
1521 : : pgoff_t last_page; /* Last page to examine */
1522 : : /*
1523 : : * Extent to map - this can be after first_page because that can be
1524 : : * fully mapped. We somewhat abuse m_flags to store whether the extent
1525 : : * is delalloc or unwritten.
1526 : : */
1527 : : struct ext4_map_blocks map;
1528 : : struct ext4_io_submit io_submit; /* IO submission data */
1529 : : unsigned int do_map:1;
1530 : : };
1531 : :
1532 : 504 : static void mpage_release_unused_pages(struct mpage_da_data *mpd,
1533 : : bool invalidate)
1534 : : {
1535 : 504 : int nr_pages, i;
1536 : 504 : pgoff_t index, end;
1537 : 504 : struct pagevec pvec;
1538 : 504 : struct inode *inode = mpd->inode;
1539 : 504 : struct address_space *mapping = inode->i_mapping;
1540 : :
1541 : : /* This is necessary when next_page == 0. */
1542 [ + + ]: 504 : if (mpd->first_page >= mpd->next_page)
1543 : 336 : return;
1544 : :
1545 : 168 : index = mpd->first_page;
1546 : 168 : end = mpd->next_page - 1;
1547 [ - + ]: 168 : if (invalidate) {
1548 : 0 : ext4_lblk_t start, last;
1549 : 0 : start = index << (PAGE_SHIFT - inode->i_blkbits);
1550 : 0 : last = end << (PAGE_SHIFT - inode->i_blkbits);
1551 : 0 : ext4_es_remove_extent(inode, start, last - start + 1);
1552 : : }
1553 : :
1554 : 168 : pagevec_init(&pvec);
1555 [ + + ]: 336 : while (index <= end) {
1556 : 168 : nr_pages = pagevec_lookup_range(&pvec, mapping, &index, end);
1557 [ + - ]: 168 : if (nr_pages == 0)
1558 : : break;
1559 [ + + ]: 336 : for (i = 0; i < nr_pages; i++) {
1560 : 168 : struct page *page = pvec.pages[i];
1561 : :
1562 [ - + - + ]: 336 : BUG_ON(!PageLocked(page));
1563 [ - + - + ]: 336 : BUG_ON(PageWriteback(page));
1564 [ - + ]: 168 : if (invalidate) {
1565 [ # # ]: 0 : if (page_mapped(page))
1566 : 0 : clear_page_dirty_for_io(page);
1567 : 0 : block_invalidatepage(page, 0, PAGE_SIZE);
1568 [ # # ]: 0 : ClearPageUptodate(page);
1569 : : }
1570 : 168 : unlock_page(page);
1571 : : }
1572 [ + - ]: 168 : pagevec_release(&pvec);
1573 : : }
1574 : : }
1575 : :
1576 : 0 : static void ext4_print_free_blocks(struct inode *inode)
1577 : : {
1578 : 0 : struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1579 : 0 : struct super_block *sb = inode->i_sb;
1580 : 0 : struct ext4_inode_info *ei = EXT4_I(inode);
1581 : :
1582 : 0 : ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld",
1583 : : EXT4_C2B(EXT4_SB(inode->i_sb),
1584 : : ext4_count_free_clusters(sb)));
1585 : 0 : ext4_msg(sb, KERN_CRIT, "Free/Dirty block details");
1586 : 0 : ext4_msg(sb, KERN_CRIT, "free_blocks=%lld",
1587 : : (long long) EXT4_C2B(EXT4_SB(sb),
1588 : : percpu_counter_sum(&sbi->s_freeclusters_counter)));
1589 : 0 : ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld",
1590 : : (long long) EXT4_C2B(EXT4_SB(sb),
1591 : : percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
1592 : 0 : ext4_msg(sb, KERN_CRIT, "Block reservation details");
1593 : 0 : ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u",
1594 : : ei->i_reserved_data_blocks);
1595 : 0 : return;
1596 : : }
1597 : :
1598 : 0 : static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh)
1599 : : {
1600 [ # # # # : 0 : return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh);
# # ]
1601 : : }
1602 : :
1603 : : /*
1604 : : * ext4_insert_delayed_block - adds a delayed block to the extents status
1605 : : * tree, incrementing the reserved cluster/block
1606 : : * count or making a pending reservation
1607 : : * where needed
1608 : : *
1609 : : * @inode - file containing the newly added block
1610 : : * @lblk - logical block to be added
1611 : : *
1612 : : * Returns 0 on success, negative error code on failure.
1613 : : */
1614 : 15316 : static int ext4_insert_delayed_block(struct inode *inode, ext4_lblk_t lblk)
1615 : : {
1616 [ + - ]: 15316 : struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
1617 : 15316 : int ret;
1618 : 15316 : bool allocated = false;
1619 : :
1620 : : /*
1621 : : * If the cluster containing lblk is shared with a delayed,
1622 : : * written, or unwritten extent in a bigalloc file system, it's
1623 : : * already been accounted for and does not need to be reserved.
1624 : : * A pending reservation must be made for the cluster if it's
1625 : : * shared with a written or unwritten extent and doesn't already
1626 : : * have one. Written and unwritten extents can be purged from the
1627 : : * extents status tree if the system is under memory pressure, so
1628 : : * it's necessary to examine the extent tree if a search of the
1629 : : * extents status tree doesn't get a match.
1630 : : */
1631 [ + - ]: 15316 : if (sbi->s_cluster_ratio == 1) {
1632 : 15316 : ret = ext4_da_reserve_space(inode);
1633 [ - + ]: 15316 : if (ret != 0) /* ENOSPC */
1634 : 0 : goto errout;
1635 : : } else { /* bigalloc */
1636 [ # # ]: 0 : if (!ext4_es_scan_clu(inode, &ext4_es_is_delonly, lblk)) {
1637 [ # # ]: 0 : if (!ext4_es_scan_clu(inode,
1638 : : &ext4_es_is_mapped, lblk)) {
1639 : 0 : ret = ext4_clu_mapped(inode,
1640 : 0 : EXT4_B2C(sbi, lblk));
1641 [ # # ]: 0 : if (ret < 0)
1642 : 0 : goto errout;
1643 [ # # ]: 0 : if (ret == 0) {
1644 : 0 : ret = ext4_da_reserve_space(inode);
1645 [ # # ]: 0 : if (ret != 0) /* ENOSPC */
1646 : 0 : goto errout;
1647 : : } else {
1648 : : allocated = true;
1649 : : }
1650 : : } else {
1651 : : allocated = true;
1652 : : }
1653 : : }
1654 : : }
1655 : :
1656 : 15316 : ret = ext4_es_insert_delayed_block(inode, lblk, allocated);
1657 : :
1658 : 15316 : errout:
1659 : 15316 : return ret;
1660 : : }
1661 : :
1662 : : /*
1663 : : * This function is grabs code from the very beginning of
1664 : : * ext4_map_blocks, but assumes that the caller is from delayed write
1665 : : * time. This function looks up the requested blocks and sets the
1666 : : * buffer delay bit under the protection of i_data_sem.
1667 : : */
1668 : 15316 : static int ext4_da_map_blocks(struct inode *inode, sector_t iblock,
1669 : : struct ext4_map_blocks *map,
1670 : : struct buffer_head *bh)
1671 : : {
1672 : 15316 : struct extent_status es;
1673 : 15316 : int retval;
1674 : 15316 : sector_t invalid_block = ~((sector_t) 0xffff);
1675 : : #ifdef ES_AGGRESSIVE_TEST
1676 : : struct ext4_map_blocks orig_map;
1677 : :
1678 : : memcpy(&orig_map, map, sizeof(*map));
1679 : : #endif
1680 : :
1681 [ - + ]: 15316 : if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es))
1682 : 0 : invalid_block = ~0;
1683 : :
1684 : 15316 : map->m_flags = 0;
1685 : 15316 : ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1686 : : "logical block %lu\n", inode->i_ino, map->m_len,
1687 : : (unsigned long) map->m_lblk);
1688 : :
1689 : : /* Lookup extent status tree firstly */
1690 [ + + ]: 15316 : if (ext4_es_lookup_extent(inode, iblock, NULL, &es)) {
1691 [ + - ]: 6118 : if (ext4_es_is_hole(&es)) {
1692 : 6118 : retval = 0;
1693 : 6118 : down_read(&EXT4_I(inode)->i_data_sem);
1694 : 6118 : goto add_delayed;
1695 : : }
1696 : :
1697 : : /*
1698 : : * Delayed extent could be allocated by fallocate.
1699 : : * So we need to check it.
1700 : : */
1701 [ # # # # ]: 0 : if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) {
1702 : 0 : map_bh(bh, inode->i_sb, invalid_block);
1703 : 0 : set_buffer_new(bh);
1704 : 0 : set_buffer_delay(bh);
1705 : 0 : return 0;
1706 : : }
1707 : :
1708 [ # # ]: 0 : map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk;
1709 : 0 : retval = es.es_len - (iblock - es.es_lblk);
1710 : 0 : if (retval > map->m_len)
1711 : : retval = map->m_len;
1712 : 0 : map->m_len = retval;
1713 [ # # ]: 0 : if (ext4_es_is_written(&es))
1714 : 0 : map->m_flags |= EXT4_MAP_MAPPED;
1715 [ # # ]: 0 : else if (ext4_es_is_unwritten(&es))
1716 : 0 : map->m_flags |= EXT4_MAP_UNWRITTEN;
1717 : : else
1718 : 0 : BUG();
1719 : :
1720 : : #ifdef ES_AGGRESSIVE_TEST
1721 : : ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0);
1722 : : #endif
1723 : 0 : return retval;
1724 : : }
1725 : :
1726 : : /*
1727 : : * Try to see if we can get the block without requesting a new
1728 : : * file system block.
1729 : : */
1730 : 9198 : down_read(&EXT4_I(inode)->i_data_sem);
1731 : 9198 : if (ext4_has_inline_data(inode))
1732 : : retval = 0;
1733 [ + - ]: 9198 : else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
1734 : 9198 : retval = ext4_ext_map_blocks(NULL, inode, map, 0);
1735 : : else
1736 : 0 : retval = ext4_ind_map_blocks(NULL, inode, map, 0);
1737 : :
1738 : 9198 : add_delayed:
1739 [ + - ]: 15316 : if (retval == 0) {
1740 : 15316 : int ret;
1741 : :
1742 : : /*
1743 : : * XXX: __block_prepare_write() unmaps passed block,
1744 : : * is it OK?
1745 : : */
1746 : :
1747 : 15316 : ret = ext4_insert_delayed_block(inode, map->m_lblk);
1748 [ - + ]: 15316 : if (ret != 0) {
1749 : 0 : retval = ret;
1750 : 0 : goto out_unlock;
1751 : : }
1752 : :
1753 : 15316 : map_bh(bh, inode->i_sb, invalid_block);
1754 : 15316 : set_buffer_new(bh);
1755 : 15316 : set_buffer_delay(bh);
1756 [ # # ]: 0 : } else if (retval > 0) {
1757 : 0 : int ret;
1758 : 0 : unsigned int status;
1759 : :
1760 [ # # ]: 0 : if (unlikely(retval != map->m_len)) {
1761 : 0 : ext4_warning(inode->i_sb,
1762 : : "ES len assertion failed for inode "
1763 : : "%lu: retval %d != map->m_len %d",
1764 : : inode->i_ino, retval, map->m_len);
1765 : 0 : WARN_ON(1);
1766 : : }
1767 : :
1768 : 0 : status = map->m_flags & EXT4_MAP_UNWRITTEN ?
1769 [ # # ]: 0 : EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN;
1770 : 0 : ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len,
1771 : : map->m_pblk, status);
1772 [ # # ]: 0 : if (ret != 0)
1773 : 0 : retval = ret;
1774 : : }
1775 : :
1776 : 0 : out_unlock:
1777 : 15316 : up_read((&EXT4_I(inode)->i_data_sem));
1778 : :
1779 : 15316 : return retval;
1780 : : }
1781 : :
1782 : : /*
1783 : : * This is a special get_block_t callback which is used by
1784 : : * ext4_da_write_begin(). It will either return mapped block or
1785 : : * reserve space for a single block.
1786 : : *
1787 : : * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1788 : : * We also have b_blocknr = -1 and b_bdev initialized properly
1789 : : *
1790 : : * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1791 : : * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1792 : : * initialized properly.
1793 : : */
1794 : 15316 : int ext4_da_get_block_prep(struct inode *inode, sector_t iblock,
1795 : : struct buffer_head *bh, int create)
1796 : : {
1797 : 15316 : struct ext4_map_blocks map;
1798 : 15316 : int ret = 0;
1799 : :
1800 [ - + ]: 15316 : BUG_ON(create == 0);
1801 [ - + ]: 15316 : BUG_ON(bh->b_size != inode->i_sb->s_blocksize);
1802 : :
1803 : 15316 : map.m_lblk = iblock;
1804 : 15316 : map.m_len = 1;
1805 : :
1806 : : /*
1807 : : * first, we need to know whether the block is allocated already
1808 : : * preallocated blocks are unmapped but should treated
1809 : : * the same as allocated blocks.
1810 : : */
1811 : 15316 : ret = ext4_da_map_blocks(inode, iblock, &map, bh);
1812 [ - + ]: 15316 : if (ret <= 0)
1813 : : return ret;
1814 : :
1815 : 0 : map_bh(bh, inode->i_sb, map.m_pblk);
1816 : 0 : ext4_update_bh_state(bh, map.m_flags);
1817 : :
1818 [ # # ]: 0 : if (buffer_unwritten(bh)) {
1819 : : /* A delayed write to unwritten bh should be marked
1820 : : * new and mapped. Mapped ensures that we don't do
1821 : : * get_block multiple times when we write to the same
1822 : : * offset and new ensures that we do proper zero out
1823 : : * for partial write.
1824 : : */
1825 : 0 : set_buffer_new(bh);
1826 : 0 : set_buffer_mapped(bh);
1827 : : }
1828 : : return 0;
1829 : : }
1830 : :
1831 : 0 : static int bget_one(handle_t *handle, struct buffer_head *bh)
1832 : : {
1833 : 0 : get_bh(bh);
1834 : 0 : return 0;
1835 : : }
1836 : :
1837 : 0 : static int bput_one(handle_t *handle, struct buffer_head *bh)
1838 : : {
1839 : 0 : put_bh(bh);
1840 : 0 : return 0;
1841 : : }
1842 : :
1843 : 0 : static int __ext4_journalled_writepage(struct page *page,
1844 : : unsigned int len)
1845 : : {
1846 : 0 : struct address_space *mapping = page->mapping;
1847 : 0 : struct inode *inode = mapping->host;
1848 : 0 : struct buffer_head *page_bufs = NULL;
1849 : 0 : handle_t *handle = NULL;
1850 : 0 : int ret = 0, err = 0;
1851 : 0 : int inline_data = ext4_has_inline_data(inode);
1852 : 0 : struct buffer_head *inode_bh = NULL;
1853 : :
1854 : 0 : ClearPageChecked(page);
1855 : :
1856 [ # # ]: 0 : if (inline_data) {
1857 [ # # ]: 0 : BUG_ON(page->index != 0);
1858 [ # # ]: 0 : BUG_ON(len > ext4_get_max_inline_size(inode));
1859 : 0 : inode_bh = ext4_journalled_write_inline_data(inode, len, page);
1860 [ # # ]: 0 : if (inode_bh == NULL)
1861 : 0 : goto out;
1862 : : } else {
1863 [ # # ]: 0 : page_bufs = page_buffers(page);
1864 [ # # ]: 0 : if (!page_bufs) {
1865 : 0 : BUG();
1866 : : goto out;
1867 : : }
1868 : 0 : ext4_walk_page_buffers(handle, page_bufs, 0, len,
1869 : : NULL, bget_one);
1870 : : }
1871 : : /*
1872 : : * We need to release the page lock before we start the
1873 : : * journal, so grab a reference so the page won't disappear
1874 : : * out from under us.
1875 : : */
1876 [ # # ]: 0 : get_page(page);
1877 : 0 : unlock_page(page);
1878 : :
1879 : 0 : handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
1880 : : ext4_writepage_trans_blocks(inode));
1881 [ # # ]: 0 : if (IS_ERR(handle)) {
1882 : 0 : ret = PTR_ERR(handle);
1883 : 0 : put_page(page);
1884 : 0 : goto out_no_pagelock;
1885 : : }
1886 [ # # # # ]: 0 : BUG_ON(!ext4_handle_valid(handle));
1887 : :
1888 : 0 : lock_page(page);
1889 : 0 : put_page(page);
1890 [ # # ]: 0 : if (page->mapping != mapping) {
1891 : : /* The page got truncated from under us */
1892 : 0 : ext4_journal_stop(handle);
1893 : 0 : ret = 0;
1894 : 0 : goto out;
1895 : : }
1896 : :
1897 [ # # ]: 0 : if (inline_data) {
1898 : 0 : ret = ext4_mark_inode_dirty(handle, inode);
1899 : : } else {
1900 : 0 : ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1901 : : do_journal_get_write_access);
1902 : :
1903 : 0 : err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL,
1904 : : write_end_fn);
1905 : : }
1906 [ # # ]: 0 : if (ret == 0)
1907 : 0 : ret = err;
1908 : 0 : EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid;
1909 : 0 : err = ext4_journal_stop(handle);
1910 [ # # ]: 0 : if (!ret)
1911 : 0 : ret = err;
1912 : :
1913 : 0 : if (!ext4_has_inline_data(inode))
1914 : 0 : ext4_walk_page_buffers(NULL, page_bufs, 0, len,
1915 : : NULL, bput_one);
1916 : 0 : ext4_set_inode_state(inode, EXT4_STATE_JDATA);
1917 : 0 : out:
1918 : 0 : unlock_page(page);
1919 : 0 : out_no_pagelock:
1920 [ # # ]: 0 : brelse(inode_bh);
1921 : 0 : return ret;
1922 : : }
1923 : :
1924 : : /*
1925 : : * Note that we don't need to start a transaction unless we're journaling data
1926 : : * because we should have holes filled from ext4_page_mkwrite(). We even don't
1927 : : * need to file the inode to the transaction's list in ordered mode because if
1928 : : * we are writing back data added by write(), the inode is already there and if
1929 : : * we are writing back data modified via mmap(), no one guarantees in which
1930 : : * transaction the data will hit the disk. In case we are journaling data, we
1931 : : * cannot start transaction directly because transaction start ranks above page
1932 : : * lock so we have to do some magic.
1933 : : *
1934 : : * This function can get called via...
1935 : : * - ext4_writepages after taking page lock (have journal handle)
1936 : : * - journal_submit_inode_data_buffers (no journal handle)
1937 : : * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
1938 : : * - grab_page_cache when doing write_begin (have journal handle)
1939 : : *
1940 : : * We don't do any block allocation in this function. If we have page with
1941 : : * multiple blocks we need to write those buffer_heads that are mapped. This
1942 : : * is important for mmaped based write. So if we do with blocksize 1K
1943 : : * truncate(f, 1024);
1944 : : * a = mmap(f, 0, 4096);
1945 : : * a[0] = 'a';
1946 : : * truncate(f, 4096);
1947 : : * we have in the page first buffer_head mapped via page_mkwrite call back
1948 : : * but other buffer_heads would be unmapped but dirty (dirty done via the
1949 : : * do_wp_page). So writepage should write the first block. If we modify
1950 : : * the mmap area beyond 1024 we will again get a page_fault and the
1951 : : * page_mkwrite callback will do the block allocation and mark the
1952 : : * buffer_heads mapped.
1953 : : *
1954 : : * We redirty the page if we have any buffer_heads that is either delay or
1955 : : * unwritten in the page.
1956 : : *
1957 : : * We can get recursively called as show below.
1958 : : *
1959 : : * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1960 : : * ext4_writepage()
1961 : : *
1962 : : * But since we don't do any block allocation we should not deadlock.
1963 : : * Page also have the dirty flag cleared so we don't get recurive page_lock.
1964 : : */
1965 : 0 : static int ext4_writepage(struct page *page,
1966 : : struct writeback_control *wbc)
1967 : : {
1968 : 0 : int ret = 0;
1969 : 0 : loff_t size;
1970 : 0 : unsigned int len;
1971 : 0 : struct buffer_head *page_bufs = NULL;
1972 : 0 : struct inode *inode = page->mapping->host;
1973 : 0 : struct ext4_io_submit io_submit;
1974 : 0 : bool keep_towrite = false;
1975 : :
1976 [ # # ]: 0 : if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) {
1977 : 0 : ext4_invalidatepage(page, 0, PAGE_SIZE);
1978 : 0 : unlock_page(page);
1979 : 0 : return -EIO;
1980 : : }
1981 : :
1982 : 0 : trace_ext4_writepage(page);
1983 [ # # ]: 0 : size = i_size_read(inode);
1984 [ # # ]: 0 : if (page->index == size >> PAGE_SHIFT &&
1985 : : !ext4_verity_in_progress(inode))
1986 : 0 : len = size & ~PAGE_MASK;
1987 : : else
1988 : : len = PAGE_SIZE;
1989 : :
1990 [ # # ]: 0 : page_bufs = page_buffers(page);
1991 : : /*
1992 : : * We cannot do block allocation or other extent handling in this
1993 : : * function. If there are buffers needing that, we have to redirty
1994 : : * the page. But we may reach here when we do a journal commit via
1995 : : * journal_submit_inode_data_buffers() and in that case we must write
1996 : : * allocated buffers to achieve data=ordered mode guarantees.
1997 : : *
1998 : : * Also, if there is only one buffer per page (the fs block
1999 : : * size == the page size), if one buffer needs block
2000 : : * allocation or needs to modify the extent tree to clear the
2001 : : * unwritten flag, we know that the page can't be written at
2002 : : * all, so we might as well refuse the write immediately.
2003 : : * Unfortunately if the block size != page size, we can't as
2004 : : * easily detect this case using ext4_walk_page_buffers(), but
2005 : : * for the extremely common case, this is an optimization that
2006 : : * skips a useless round trip through ext4_bio_write_page().
2007 : : */
2008 [ # # ]: 0 : if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL,
2009 : : ext4_bh_delay_or_unwritten)) {
2010 : 0 : redirty_page_for_writepage(wbc, page);
2011 [ # # ]: 0 : if ((current->flags & PF_MEMALLOC) ||
2012 [ # # ]: 0 : (inode->i_sb->s_blocksize == PAGE_SIZE)) {
2013 : : /*
2014 : : * For memory cleaning there's no point in writing only
2015 : : * some buffers. So just bail out. Warn if we came here
2016 : : * from direct reclaim.
2017 : : */
2018 [ # # ]: 0 : WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD))
2019 : : == PF_MEMALLOC);
2020 : 0 : unlock_page(page);
2021 : 0 : return 0;
2022 : : }
2023 : : keep_towrite = true;
2024 : : }
2025 : :
2026 [ # # # # ]: 0 : if (PageChecked(page) && ext4_should_journal_data(inode))
2027 : : /*
2028 : : * It's mmapped pagecache. Add buffers and journal it. There
2029 : : * doesn't seem much point in redirtying the page here.
2030 : : */
2031 : 0 : return __ext4_journalled_writepage(page, len);
2032 : :
2033 : 0 : ext4_io_submit_init(&io_submit, wbc);
2034 : 0 : io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS);
2035 [ # # ]: 0 : if (!io_submit.io_end) {
2036 : 0 : redirty_page_for_writepage(wbc, page);
2037 : 0 : unlock_page(page);
2038 : 0 : return -ENOMEM;
2039 : : }
2040 : 0 : ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite);
2041 : 0 : ext4_io_submit(&io_submit);
2042 : : /* Drop io_end reference we got from init */
2043 : 0 : ext4_put_io_end_defer(io_submit.io_end);
2044 : 0 : return ret;
2045 : : }
2046 : :
2047 : 168 : static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page)
2048 : : {
2049 : 168 : int len;
2050 : 168 : loff_t size;
2051 : 168 : int err;
2052 : :
2053 [ - + ]: 168 : BUG_ON(page->index != mpd->first_page);
2054 : 168 : clear_page_dirty_for_io(page);
2055 : : /*
2056 : : * We have to be very careful here! Nothing protects writeback path
2057 : : * against i_size changes and the page can be writeably mapped into
2058 : : * page tables. So an application can be growing i_size and writing
2059 : : * data through mmap while writeback runs. clear_page_dirty_for_io()
2060 : : * write-protects our page in page tables and the page cannot get
2061 : : * written to again until we release page lock. So only after
2062 : : * clear_page_dirty_for_io() we are safe to sample i_size for
2063 : : * ext4_bio_write_page() to zero-out tail of the written page. We rely
2064 : : * on the barrier provided by TestClearPageDirty in
2065 : : * clear_page_dirty_for_io() to make sure i_size is really sampled only
2066 : : * after page tables are updated.
2067 : : */
2068 [ + - ]: 168 : size = i_size_read(mpd->inode);
2069 [ + - ]: 168 : if (page->index == size >> PAGE_SHIFT &&
2070 : : !ext4_verity_in_progress(mpd->inode))
2071 : 168 : len = size & ~PAGE_MASK;
2072 : : else
2073 : : len = PAGE_SIZE;
2074 : 168 : err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false);
2075 [ + - ]: 168 : if (!err)
2076 : 168 : mpd->wbc->nr_to_write--;
2077 : 168 : mpd->first_page++;
2078 : :
2079 : 168 : return err;
2080 : : }
2081 : :
2082 : : #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
2083 : :
2084 : : /*
2085 : : * mballoc gives us at most this number of blocks...
2086 : : * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
2087 : : * The rest of mballoc seems to handle chunks up to full group size.
2088 : : */
2089 : : #define MAX_WRITEPAGES_EXTENT_LEN 2048
2090 : :
2091 : : /*
2092 : : * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
2093 : : *
2094 : : * @mpd - extent of blocks
2095 : : * @lblk - logical number of the block in the file
2096 : : * @bh - buffer head we want to add to the extent
2097 : : *
2098 : : * The function is used to collect contig. blocks in the same state. If the
2099 : : * buffer doesn't require mapping for writeback and we haven't started the
2100 : : * extent of buffers to map yet, the function returns 'true' immediately - the
2101 : : * caller can write the buffer right away. Otherwise the function returns true
2102 : : * if the block has been added to the extent, false if the block couldn't be
2103 : : * added.
2104 : : */
2105 : 336 : static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk,
2106 : : struct buffer_head *bh)
2107 : : {
2108 : 336 : struct ext4_map_blocks *map = &mpd->map;
2109 : :
2110 : : /* Buffer that doesn't need mapping for writeback? */
2111 [ + - + - : 1008 : if (!buffer_dirty(bh) || !buffer_mapped(bh) ||
- + ]
2112 [ # # ]: 0 : (!buffer_delay(bh) && !buffer_unwritten(bh))) {
2113 : : /* So far no extent to map => we write the buffer right away */
2114 [ # # ]: 0 : if (map->m_len == 0)
2115 : : return true;
2116 : 0 : return false;
2117 : : }
2118 : :
2119 : : /* First block in the extent? */
2120 [ + - ]: 336 : if (map->m_len == 0) {
2121 : : /* We cannot map unless handle is started... */
2122 [ + + ]: 336 : if (!mpd->do_map)
2123 : : return false;
2124 : 168 : map->m_lblk = lblk;
2125 : 168 : map->m_len = 1;
2126 : 168 : map->m_flags = bh->b_state & BH_FLAGS;
2127 : 168 : return true;
2128 : : }
2129 : :
2130 : : /* Don't go larger than mballoc is willing to allocate */
2131 [ # # ]: 0 : if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN)
2132 : : return false;
2133 : :
2134 : : /* Can we merge the block to our big extent? */
2135 [ # # ]: 0 : if (lblk == map->m_lblk + map->m_len &&
2136 [ # # ]: 0 : (bh->b_state & BH_FLAGS) == map->m_flags) {
2137 : 0 : map->m_len++;
2138 : 0 : return true;
2139 : : }
2140 : : return false;
2141 : : }
2142 : :
2143 : : /*
2144 : : * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2145 : : *
2146 : : * @mpd - extent of blocks for mapping
2147 : : * @head - the first buffer in the page
2148 : : * @bh - buffer we should start processing from
2149 : : * @lblk - logical number of the block in the file corresponding to @bh
2150 : : *
2151 : : * Walk through page buffers from @bh upto @head (exclusive) and either submit
2152 : : * the page for IO if all buffers in this page were mapped and there's no
2153 : : * accumulated extent of buffers to map or add buffers in the page to the
2154 : : * extent of buffers to map. The function returns 1 if the caller can continue
2155 : : * by processing the next page, 0 if it should stop adding buffers to the
2156 : : * extent to map because we cannot extend it anymore. It can also return value
2157 : : * < 0 in case of error during IO submission.
2158 : : */
2159 : 336 : static int mpage_process_page_bufs(struct mpage_da_data *mpd,
2160 : : struct buffer_head *head,
2161 : : struct buffer_head *bh,
2162 : : ext4_lblk_t lblk)
2163 : : {
2164 : 336 : struct inode *inode = mpd->inode;
2165 : 336 : int err;
2166 : 336 : ext4_lblk_t blocks = (i_size_read(inode) + i_blocksize(inode) - 1)
2167 : 336 : >> inode->i_blkbits;
2168 : :
2169 : 336 : if (ext4_verity_in_progress(inode))
2170 : : blocks = EXT_MAX_BLOCKS;
2171 : :
2172 : 336 : do {
2173 [ - + ]: 336 : BUG_ON(buffer_locked(bh));
2174 : :
2175 [ + - + + ]: 336 : if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) {
2176 : : /* Found extent to map? */
2177 [ + - ]: 168 : if (mpd->map.m_len)
2178 : : return 0;
2179 : : /* Buffer needs mapping and handle is not started? */
2180 [ - + ]: 168 : if (!mpd->do_map)
2181 : : return 0;
2182 : : /* Everything mapped so far and we hit EOF */
2183 : : break;
2184 : : }
2185 [ - + ]: 168 : } while (lblk++, (bh = bh->b_this_page) != head);
2186 : : /* So far everything mapped? Submit the page for IO. */
2187 [ - + ]: 168 : if (mpd->map.m_len == 0) {
2188 : 0 : err = mpage_submit_page(mpd, head->b_page);
2189 [ # # ]: 0 : if (err < 0)
2190 : : return err;
2191 : : }
2192 : 168 : return lblk < blocks;
2193 : : }
2194 : :
2195 : : /*
2196 : : * mpage_process_page - update page buffers corresponding to changed extent and
2197 : : * may submit fully mapped page for IO
2198 : : *
2199 : : * @mpd - description of extent to map, on return next extent to map
2200 : : * @m_lblk - logical block mapping.
2201 : : * @m_pblk - corresponding physical mapping.
2202 : : * @map_bh - determines on return whether this page requires any further
2203 : : * mapping or not.
2204 : : * Scan given page buffers corresponding to changed extent and update buffer
2205 : : * state according to new extent state.
2206 : : * We map delalloc buffers to their physical location, clear unwritten bits.
2207 : : * If the given page is not fully mapped, we update @map to the next extent in
2208 : : * the given page that needs mapping & return @map_bh as true.
2209 : : */
2210 : 168 : static int mpage_process_page(struct mpage_da_data *mpd, struct page *page,
2211 : : ext4_lblk_t *m_lblk, ext4_fsblk_t *m_pblk,
2212 : : bool *map_bh)
2213 : : {
2214 : 168 : struct buffer_head *head, *bh;
2215 : 168 : ext4_io_end_t *io_end = mpd->io_submit.io_end;
2216 : 168 : ext4_lblk_t lblk = *m_lblk;
2217 : 168 : ext4_fsblk_t pblock = *m_pblk;
2218 : 168 : int err = 0;
2219 : 168 : int blkbits = mpd->inode->i_blkbits;
2220 : 168 : ssize_t io_end_size = 0;
2221 : 168 : struct ext4_io_end_vec *io_end_vec = ext4_last_io_end_vec(io_end);
2222 : :
2223 [ - + ]: 168 : bh = head = page_buffers(page);
2224 : 168 : do {
2225 [ - + ]: 168 : if (lblk < mpd->map.m_lblk)
2226 : 0 : continue;
2227 [ - + ]: 168 : if (lblk >= mpd->map.m_lblk + mpd->map.m_len) {
2228 : : /*
2229 : : * Buffer after end of mapped extent.
2230 : : * Find next buffer in the page to map.
2231 : : */
2232 : 0 : mpd->map.m_len = 0;
2233 : 0 : mpd->map.m_flags = 0;
2234 : 0 : io_end_vec->size += io_end_size;
2235 : 0 : io_end_size = 0;
2236 : :
2237 : 0 : err = mpage_process_page_bufs(mpd, head, bh, lblk);
2238 [ # # ]: 0 : if (err > 0)
2239 : : err = 0;
2240 [ # # # # : 0 : if (!err && mpd->map.m_len && mpd->map.m_lblk > lblk) {
# # ]
2241 : 0 : io_end_vec = ext4_alloc_io_end_vec(io_end);
2242 [ # # ]: 0 : if (IS_ERR(io_end_vec)) {
2243 : 0 : err = PTR_ERR(io_end_vec);
2244 : 0 : goto out;
2245 : : }
2246 : 0 : io_end_vec->offset = mpd->map.m_lblk << blkbits;
2247 : : }
2248 : 0 : *map_bh = true;
2249 : 0 : goto out;
2250 : : }
2251 [ + - ]: 168 : if (buffer_delay(bh)) {
2252 : 168 : clear_buffer_delay(bh);
2253 : 168 : bh->b_blocknr = pblock++;
2254 : : }
2255 : 168 : clear_buffer_unwritten(bh);
2256 : 168 : io_end_size += (1 << blkbits);
2257 [ - + ]: 168 : } while (lblk++, (bh = bh->b_this_page) != head);
2258 : :
2259 : 168 : io_end_vec->size += io_end_size;
2260 : 168 : io_end_size = 0;
2261 : 168 : *map_bh = false;
2262 : 168 : out:
2263 : 168 : *m_lblk = lblk;
2264 : 168 : *m_pblk = pblock;
2265 : 168 : return err;
2266 : : }
2267 : :
2268 : : /*
2269 : : * mpage_map_buffers - update buffers corresponding to changed extent and
2270 : : * submit fully mapped pages for IO
2271 : : *
2272 : : * @mpd - description of extent to map, on return next extent to map
2273 : : *
2274 : : * Scan buffers corresponding to changed extent (we expect corresponding pages
2275 : : * to be already locked) and update buffer state according to new extent state.
2276 : : * We map delalloc buffers to their physical location, clear unwritten bits,
2277 : : * and mark buffers as uninit when we perform writes to unwritten extents
2278 : : * and do extent conversion after IO is finished. If the last page is not fully
2279 : : * mapped, we update @map to the next extent in the last page that needs
2280 : : * mapping. Otherwise we submit the page for IO.
2281 : : */
2282 : 168 : static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd)
2283 : : {
2284 : 168 : struct pagevec pvec;
2285 : 168 : int nr_pages, i;
2286 : 168 : struct inode *inode = mpd->inode;
2287 : 168 : int bpp_bits = PAGE_SHIFT - inode->i_blkbits;
2288 : 168 : pgoff_t start, end;
2289 : 168 : ext4_lblk_t lblk;
2290 : 168 : ext4_fsblk_t pblock;
2291 : 168 : int err;
2292 : 168 : bool map_bh = false;
2293 : :
2294 : 168 : start = mpd->map.m_lblk >> bpp_bits;
2295 : 168 : end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits;
2296 : 168 : lblk = start << bpp_bits;
2297 : 168 : pblock = mpd->map.m_pblk;
2298 : :
2299 : 168 : pagevec_init(&pvec);
2300 [ + + ]: 336 : while (start <= end) {
2301 : 168 : nr_pages = pagevec_lookup_range(&pvec, inode->i_mapping,
2302 : : &start, end);
2303 [ + - ]: 168 : if (nr_pages == 0)
2304 : : break;
2305 [ + + ]: 336 : for (i = 0; i < nr_pages; i++) {
2306 : 168 : struct page *page = pvec.pages[i];
2307 : :
2308 : 168 : err = mpage_process_page(mpd, page, &lblk, &pblock,
2309 : : &map_bh);
2310 : : /*
2311 : : * If map_bh is true, means page may require further bh
2312 : : * mapping, or maybe the page was submitted for IO.
2313 : : * So we return to call further extent mapping.
2314 : : */
2315 [ + - - + ]: 168 : if (err < 0 || map_bh == true)
2316 : 0 : goto out;
2317 : : /* Page fully mapped - let IO run! */
2318 : 168 : err = mpage_submit_page(mpd, page);
2319 [ - + ]: 168 : if (err < 0)
2320 : 0 : goto out;
2321 : : }
2322 [ - + ]: 168 : pagevec_release(&pvec);
2323 : : }
2324 : : /* Extent fully mapped and matches with page boundary. We are done. */
2325 : 168 : mpd->map.m_len = 0;
2326 : 168 : mpd->map.m_flags = 0;
2327 : 168 : return 0;
2328 : 0 : out:
2329 [ # # ]: 0 : pagevec_release(&pvec);
2330 : : return err;
2331 : : }
2332 : :
2333 : 168 : static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd)
2334 : : {
2335 : 168 : struct inode *inode = mpd->inode;
2336 : 168 : struct ext4_map_blocks *map = &mpd->map;
2337 : 168 : int get_blocks_flags;
2338 : 168 : int err, dioread_nolock;
2339 : :
2340 : 168 : trace_ext4_da_write_pages_extent(inode, map);
2341 : : /*
2342 : : * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2343 : : * to convert an unwritten extent to be initialized (in the case
2344 : : * where we have written into one or more preallocated blocks). It is
2345 : : * possible that we're going to need more metadata blocks than
2346 : : * previously reserved. However we must not fail because we're in
2347 : : * writeback and there is nothing we can do about it so it might result
2348 : : * in data loss. So use reserved blocks to allocate metadata if
2349 : : * possible.
2350 : : *
2351 : : * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2352 : : * the blocks in question are delalloc blocks. This indicates
2353 : : * that the blocks and quotas has already been checked when
2354 : : * the data was copied into the page cache.
2355 : : */
2356 : 168 : get_blocks_flags = EXT4_GET_BLOCKS_CREATE |
2357 : : EXT4_GET_BLOCKS_METADATA_NOFAIL |
2358 : : EXT4_GET_BLOCKS_IO_SUBMIT;
2359 : 168 : dioread_nolock = ext4_should_dioread_nolock(inode);
2360 [ + - ]: 168 : if (dioread_nolock)
2361 : 168 : get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT;
2362 [ + - ]: 168 : if (map->m_flags & (1 << BH_Delay))
2363 : 168 : get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE;
2364 : :
2365 : 168 : err = ext4_map_blocks(handle, inode, map, get_blocks_flags);
2366 [ + - ]: 168 : if (err < 0)
2367 : : return err;
2368 [ + - + - ]: 168 : if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) {
2369 [ + - ]: 168 : if (!mpd->io_submit.io_end->handle &&
2370 : : ext4_handle_valid(handle)) {
2371 : 168 : mpd->io_submit.io_end->handle = handle->h_rsv_handle;
2372 : 168 : handle->h_rsv_handle = NULL;
2373 : : }
2374 [ + - ]: 168 : ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end);
2375 : : }
2376 : :
2377 [ - + ]: 168 : BUG_ON(map->m_len == 0);
2378 : : return 0;
2379 : : }
2380 : :
2381 : : /*
2382 : : * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2383 : : * mpd->len and submit pages underlying it for IO
2384 : : *
2385 : : * @handle - handle for journal operations
2386 : : * @mpd - extent to map
2387 : : * @give_up_on_write - we set this to true iff there is a fatal error and there
2388 : : * is no hope of writing the data. The caller should discard
2389 : : * dirty pages to avoid infinite loops.
2390 : : *
2391 : : * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2392 : : * delayed, blocks are allocated, if it is unwritten, we may need to convert
2393 : : * them to initialized or split the described range from larger unwritten
2394 : : * extent. Note that we need not map all the described range since allocation
2395 : : * can return less blocks or the range is covered by more unwritten extents. We
2396 : : * cannot map more because we are limited by reserved transaction credits. On
2397 : : * the other hand we always make sure that the last touched page is fully
2398 : : * mapped so that it can be written out (and thus forward progress is
2399 : : * guaranteed). After mapping we submit all mapped pages for IO.
2400 : : */
2401 : 168 : static int mpage_map_and_submit_extent(handle_t *handle,
2402 : : struct mpage_da_data *mpd,
2403 : : bool *give_up_on_write)
2404 : : {
2405 : 168 : struct inode *inode = mpd->inode;
2406 : 168 : struct ext4_map_blocks *map = &mpd->map;
2407 : 168 : int err;
2408 : 168 : loff_t disksize;
2409 : 168 : int progress = 0;
2410 : 168 : ext4_io_end_t *io_end = mpd->io_submit.io_end;
2411 : 168 : struct ext4_io_end_vec *io_end_vec;
2412 : :
2413 : 168 : io_end_vec = ext4_alloc_io_end_vec(io_end);
2414 [ - + ]: 168 : if (IS_ERR(io_end_vec))
2415 : 0 : return PTR_ERR(io_end_vec);
2416 : 168 : io_end_vec->offset = ((loff_t)map->m_lblk) << inode->i_blkbits;
2417 : 168 : do {
2418 : 168 : err = mpage_map_one_extent(handle, mpd);
2419 [ - + ]: 168 : if (err < 0) {
2420 : 0 : struct super_block *sb = inode->i_sb;
2421 : :
2422 [ # # # # ]: 0 : if (ext4_forced_shutdown(EXT4_SB(sb)) ||
2423 [ # # ]: 0 : EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)
2424 : 0 : goto invalidate_dirty_pages;
2425 : : /*
2426 : : * Let the uper layers retry transient errors.
2427 : : * In the case of ENOSPC, if ext4_count_free_blocks()
2428 : : * is non-zero, a commit should free up blocks.
2429 : : */
2430 [ # # # # ]: 0 : if ((err == -ENOMEM) ||
2431 [ # # ]: 0 : (err == -ENOSPC && ext4_count_free_clusters(sb))) {
2432 [ # # ]: 0 : if (progress)
2433 : 0 : goto update_disksize;
2434 : : return err;
2435 : : }
2436 : 0 : ext4_msg(sb, KERN_CRIT,
2437 : : "Delayed block allocation failed for "
2438 : : "inode %lu at logical offset %llu with"
2439 : : " max blocks %u with error %d",
2440 : : inode->i_ino,
2441 : : (unsigned long long)map->m_lblk,
2442 : : (unsigned)map->m_len, -err);
2443 : 0 : ext4_msg(sb, KERN_CRIT,
2444 : : "This should not happen!! Data will "
2445 : : "be lost\n");
2446 [ # # ]: 0 : if (err == -ENOSPC)
2447 : 0 : ext4_print_free_blocks(inode);
2448 : 0 : invalidate_dirty_pages:
2449 : 0 : *give_up_on_write = true;
2450 : 0 : return err;
2451 : : }
2452 : 168 : progress = 1;
2453 : : /*
2454 : : * Update buffer state, submit mapped pages, and get us new
2455 : : * extent to map
2456 : : */
2457 : 168 : err = mpage_map_and_submit_buffers(mpd);
2458 [ - + ]: 168 : if (err < 0)
2459 : 0 : goto update_disksize;
2460 [ - + ]: 168 : } while (map->m_len);
2461 : :
2462 : 168 : update_disksize:
2463 : : /*
2464 : : * Update on-disk size after IO is submitted. Races with
2465 : : * truncate are avoided by checking i_size under i_data_sem.
2466 : : */
2467 : 168 : disksize = ((loff_t)mpd->first_page) << PAGE_SHIFT;
2468 [ + - ]: 168 : if (disksize > READ_ONCE(EXT4_I(inode)->i_disksize)) {
2469 : 168 : int err2;
2470 : 168 : loff_t i_size;
2471 : :
2472 : 168 : down_write(&EXT4_I(inode)->i_data_sem);
2473 [ + - ]: 168 : i_size = i_size_read(inode);
2474 : 168 : if (disksize > i_size)
2475 : : disksize = i_size;
2476 [ + - ]: 168 : if (disksize > EXT4_I(inode)->i_disksize)
2477 : 168 : EXT4_I(inode)->i_disksize = disksize;
2478 : 168 : up_write(&EXT4_I(inode)->i_data_sem);
2479 : 168 : err2 = ext4_mark_inode_dirty(handle, inode);
2480 [ - + ]: 168 : if (err2) {
2481 : 0 : ext4_set_errno(inode->i_sb, -err2);
2482 : 0 : ext4_error(inode->i_sb,
2483 : : "Failed to mark inode %lu dirty",
2484 : : inode->i_ino);
2485 : : }
2486 [ + - ]: 168 : if (!err)
2487 : 168 : err = err2;
2488 : : }
2489 : : return err;
2490 : : }
2491 : :
2492 : : /*
2493 : : * Calculate the total number of credits to reserve for one writepages
2494 : : * iteration. This is called from ext4_writepages(). We map an extent of
2495 : : * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2496 : : * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2497 : : * bpp - 1 blocks in bpp different extents.
2498 : : */
2499 : 336 : static int ext4_da_writepages_trans_blocks(struct inode *inode)
2500 : : {
2501 [ + - ]: 336 : int bpp = ext4_journal_blocks_per_page(inode);
2502 : :
2503 : 336 : return ext4_meta_trans_blocks(inode,
2504 : : MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp);
2505 : : }
2506 : :
2507 : : /*
2508 : : * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2509 : : * and underlying extent to map
2510 : : *
2511 : : * @mpd - where to look for pages
2512 : : *
2513 : : * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2514 : : * IO immediately. When we find a page which isn't mapped we start accumulating
2515 : : * extent of buffers underlying these pages that needs mapping (formed by
2516 : : * either delayed or unwritten buffers). We also lock the pages containing
2517 : : * these buffers. The extent found is returned in @mpd structure (starting at
2518 : : * mpd->lblk with length mpd->len blocks).
2519 : : *
2520 : : * Note that this function can attach bios to one io_end structure which are
2521 : : * neither logically nor physically contiguous. Although it may seem as an
2522 : : * unnecessary complication, it is actually inevitable in blocksize < pagesize
2523 : : * case as we need to track IO to all buffers underlying a page in one io_end.
2524 : : */
2525 : 504 : static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
2526 : : {
2527 : 504 : struct address_space *mapping = mpd->inode->i_mapping;
2528 : 504 : struct pagevec pvec;
2529 : 504 : unsigned int nr_pages;
2530 : 504 : long left = mpd->wbc->nr_to_write;
2531 : 504 : pgoff_t index = mpd->first_page;
2532 : 504 : pgoff_t end = mpd->last_page;
2533 : 504 : xa_mark_t tag;
2534 : 504 : int i, err = 0;
2535 : 504 : int blkbits = mpd->inode->i_blkbits;
2536 : 504 : ext4_lblk_t lblk;
2537 : 504 : struct buffer_head *head;
2538 : :
2539 [ + + + - ]: 504 : if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages)
2540 : : tag = PAGECACHE_TAG_TOWRITE;
2541 : : else
2542 : 63 : tag = PAGECACHE_TAG_DIRTY;
2543 : :
2544 : 504 : pagevec_init(&pvec);
2545 : 504 : mpd->map.m_len = 0;
2546 : 504 : mpd->next_page = index;
2547 [ + - ]: 504 : while (index <= end) {
2548 : 504 : nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end,
2549 : : tag);
2550 [ + + ]: 504 : if (nr_pages == 0)
2551 : 168 : goto out;
2552 : :
2553 [ + - ]: 336 : for (i = 0; i < nr_pages; i++) {
2554 : 336 : struct page *page = pvec.pages[i];
2555 : :
2556 : : /*
2557 : : * Accumulated enough dirty pages? This doesn't apply
2558 : : * to WB_SYNC_ALL mode. For integrity sync we have to
2559 : : * keep going because someone may be concurrently
2560 : : * dirtying pages, and we might have synced a lot of
2561 : : * newly appeared dirty pages, but have not synced all
2562 : : * of the old dirty pages.
2563 : : */
2564 [ + + - + ]: 336 : if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0)
2565 : 0 : goto out;
2566 : :
2567 : : /* If we can't merge this page, we are done. */
2568 [ - + - - ]: 336 : if (mpd->map.m_len > 0 && mpd->next_page != page->index)
2569 : 0 : goto out;
2570 : :
2571 : 336 : lock_page(page);
2572 : : /*
2573 : : * If the page is no longer dirty, or its mapping no
2574 : : * longer corresponds to inode we are writing (which
2575 : : * means it has been truncated or invalidated), or the
2576 : : * page is already under writeback and we are not doing
2577 : : * a data integrity writeback, skip the page
2578 : : */
2579 [ - + + - : 1008 : if (!PageDirty(page) ||
- + ]
2580 : 0 : (PageWriteback(page) &&
2581 [ # # ]: 0 : (mpd->wbc->sync_mode == WB_SYNC_NONE)) ||
2582 [ - + ]: 336 : unlikely(page->mapping != mapping)) {
2583 : 0 : unlock_page(page);
2584 : 0 : continue;
2585 : : }
2586 : :
2587 : 336 : wait_on_page_writeback(page);
2588 [ - + - + ]: 672 : BUG_ON(PageWriteback(page));
2589 : :
2590 [ + - ]: 336 : if (mpd->map.m_len == 0)
2591 : 336 : mpd->first_page = page->index;
2592 : 336 : mpd->next_page = page->index + 1;
2593 : : /* Add all dirty buffers to mpd */
2594 : 336 : lblk = ((ext4_lblk_t)page->index) <<
2595 : 336 : (PAGE_SHIFT - blkbits);
2596 [ - + ]: 336 : head = page_buffers(page);
2597 : 336 : err = mpage_process_page_bufs(mpd, head, head, lblk);
2598 [ + - ]: 336 : if (err <= 0)
2599 : 336 : goto out;
2600 : 0 : err = 0;
2601 : 0 : left--;
2602 : : }
2603 [ # # ]: 0 : pagevec_release(&pvec);
2604 : 0 : cond_resched();
2605 : : }
2606 : : return 0;
2607 : 504 : out:
2608 [ + + ]: 504 : pagevec_release(&pvec);
2609 : : return err;
2610 : : }
2611 : :
2612 : 168 : static int ext4_writepages(struct address_space *mapping,
2613 : : struct writeback_control *wbc)
2614 : : {
2615 : 168 : pgoff_t writeback_index = 0;
2616 : 168 : long nr_to_write = wbc->nr_to_write;
2617 : 168 : int range_whole = 0;
2618 : 168 : int cycled = 1;
2619 : 168 : handle_t *handle = NULL;
2620 : 168 : struct mpage_da_data mpd;
2621 : 168 : struct inode *inode = mapping->host;
2622 : 168 : int needed_blocks, rsv_blocks = 0, ret = 0;
2623 : 168 : struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2624 : 168 : bool done;
2625 : 168 : struct blk_plug plug;
2626 : 168 : bool give_up_on_write = false;
2627 : :
2628 [ + - ]: 168 : if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2629 : : return -EIO;
2630 : :
2631 : 168 : percpu_down_read(&sbi->s_writepages_rwsem);
2632 : 168 : trace_ext4_writepages(inode, wbc);
2633 : :
2634 : : /*
2635 : : * No pages to write? This is mainly a kludge to avoid starting
2636 : : * a transaction for special inodes like journal inode on last iput()
2637 : : * because that could violate lock ordering on umount
2638 : : */
2639 [ + - - + ]: 168 : if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
2640 : 0 : goto out_writepages;
2641 : :
2642 [ - + ]: 168 : if (ext4_should_journal_data(inode)) {
2643 : 0 : ret = generic_writepages(mapping, wbc);
2644 : 0 : goto out_writepages;
2645 : : }
2646 : :
2647 : : /*
2648 : : * If the filesystem has aborted, it is read-only, so return
2649 : : * right away instead of dumping stack traces later on that
2650 : : * will obscure the real source of the problem. We test
2651 : : * EXT4_MF_FS_ABORTED instead of sb->s_flag's SB_RDONLY because
2652 : : * the latter could be true if the filesystem is mounted
2653 : : * read-only, and in that case, ext4_writepages should
2654 : : * *never* be called, so if that ever happens, we would want
2655 : : * the stack trace.
2656 : : */
2657 [ + - - + ]: 168 : if (unlikely(ext4_forced_shutdown(EXT4_SB(mapping->host->i_sb)) ||
2658 : : sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) {
2659 : 0 : ret = -EROFS;
2660 : 0 : goto out_writepages;
2661 : : }
2662 : :
2663 : : /*
2664 : : * If we have inline data and arrive here, it means that
2665 : : * we will soon create the block for the 1st page, so
2666 : : * we'd better clear the inline data here.
2667 : : */
2668 : 168 : if (ext4_has_inline_data(inode)) {
2669 : : /* Just inode will be modified... */
2670 : 0 : handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
2671 [ # # ]: 0 : if (IS_ERR(handle)) {
2672 : 0 : ret = PTR_ERR(handle);
2673 : 0 : goto out_writepages;
2674 : : }
2675 [ # # ]: 0 : BUG_ON(ext4_test_inode_state(inode,
2676 : : EXT4_STATE_MAY_INLINE_DATA));
2677 : 0 : ext4_destroy_inline_data(handle, inode);
2678 : 0 : ext4_journal_stop(handle);
2679 : : }
2680 : :
2681 [ + - ]: 168 : if (ext4_should_dioread_nolock(inode)) {
2682 : : /*
2683 : : * We may need to convert up to one extent per block in
2684 : : * the page and we may dirty the inode.
2685 : : */
2686 : 168 : rsv_blocks = 1 + ext4_chunk_trans_blocks(inode,
2687 : 168 : PAGE_SIZE >> inode->i_blkbits);
2688 : : }
2689 : :
2690 [ + - + - ]: 168 : if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
2691 : 168 : range_whole = 1;
2692 : :
2693 [ - + ]: 168 : if (wbc->range_cyclic) {
2694 : 0 : writeback_index = mapping->writeback_index;
2695 [ # # ]: 0 : if (writeback_index)
2696 : 0 : cycled = 0;
2697 : 0 : mpd.first_page = writeback_index;
2698 : 0 : mpd.last_page = -1;
2699 : : } else {
2700 : 168 : mpd.first_page = wbc->range_start >> PAGE_SHIFT;
2701 : 168 : mpd.last_page = wbc->range_end >> PAGE_SHIFT;
2702 : : }
2703 : :
2704 : 168 : mpd.inode = inode;
2705 : 168 : mpd.wbc = wbc;
2706 : 168 : ext4_io_submit_init(&mpd.io_submit, wbc);
2707 : 168 : retry:
2708 [ + + - + ]: 168 : if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
2709 : 147 : tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page);
2710 : 168 : done = false;
2711 : 168 : blk_start_plug(&plug);
2712 : :
2713 : : /*
2714 : : * First writeback pages that don't need mapping - we can avoid
2715 : : * starting a transaction unnecessarily and also avoid being blocked
2716 : : * in the block layer on device congestion while having transaction
2717 : : * started.
2718 : : */
2719 : 168 : mpd.do_map = 0;
2720 : 168 : mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2721 [ - + ]: 168 : if (!mpd.io_submit.io_end) {
2722 : 0 : ret = -ENOMEM;
2723 : 0 : goto unplug;
2724 : : }
2725 : 168 : ret = mpage_prepare_extent_to_map(&mpd);
2726 : : /* Unlock pages we didn't use */
2727 : 168 : mpage_release_unused_pages(&mpd, false);
2728 : : /* Submit prepared bio */
2729 : 168 : ext4_io_submit(&mpd.io_submit);
2730 : 168 : ext4_put_io_end_defer(mpd.io_submit.io_end);
2731 : 168 : mpd.io_submit.io_end = NULL;
2732 [ - + ]: 168 : if (ret < 0)
2733 : 0 : goto unplug;
2734 : :
2735 [ + + + - ]: 504 : while (!done && mpd.first_page <= mpd.last_page) {
2736 : : /* For each extent of pages we use new io_end */
2737 : 336 : mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL);
2738 [ + - ]: 336 : if (!mpd.io_submit.io_end) {
2739 : : ret = -ENOMEM;
2740 : : break;
2741 : : }
2742 : :
2743 : : /*
2744 : : * We have two constraints: We find one extent to map and we
2745 : : * must always write out whole page (makes a difference when
2746 : : * blocksize < pagesize) so that we don't block on IO when we
2747 : : * try to write out the rest of the page. Journalled mode is
2748 : : * not supported by delalloc.
2749 : : */
2750 [ - + ]: 336 : BUG_ON(ext4_should_journal_data(inode));
2751 : 336 : needed_blocks = ext4_da_writepages_trans_blocks(inode);
2752 : :
2753 : : /* start a new transaction */
2754 : 336 : handle = ext4_journal_start_with_reserve(inode,
2755 : : EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks);
2756 [ - + ]: 336 : if (IS_ERR(handle)) {
2757 : 0 : ret = PTR_ERR(handle);
2758 : 0 : ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: "
2759 : : "%ld pages, ino %lu; err %d", __func__,
2760 : : wbc->nr_to_write, inode->i_ino, ret);
2761 : : /* Release allocated io_end */
2762 : 0 : ext4_put_io_end(mpd.io_submit.io_end);
2763 : 0 : mpd.io_submit.io_end = NULL;
2764 : 0 : break;
2765 : : }
2766 : 336 : mpd.do_map = 1;
2767 : :
2768 : 336 : trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc);
2769 : 336 : ret = mpage_prepare_extent_to_map(&mpd);
2770 [ + - ]: 336 : if (!ret) {
2771 [ + + ]: 336 : if (mpd.map.m_len)
2772 : 168 : ret = mpage_map_and_submit_extent(handle, &mpd,
2773 : : &give_up_on_write);
2774 : : else {
2775 : : /*
2776 : : * We scanned the whole range (or exhausted
2777 : : * nr_to_write), submitted what was mapped and
2778 : : * didn't find anything needing mapping. We are
2779 : : * done.
2780 : : */
2781 : : done = true;
2782 : : }
2783 : : }
2784 : : /*
2785 : : * Caution: If the handle is synchronous,
2786 : : * ext4_journal_stop() can wait for transaction commit
2787 : : * to finish which may depend on writeback of pages to
2788 : : * complete or on page lock to be released. In that
2789 : : * case, we have to wait until after after we have
2790 : : * submitted all the IO, released page locks we hold,
2791 : : * and dropped io_end reference (for extent conversion
2792 : : * to be able to complete) before stopping the handle.
2793 : : */
2794 [ + - + - ]: 336 : if (!ext4_handle_valid(handle) || handle->h_sync == 0) {
2795 : 336 : ext4_journal_stop(handle);
2796 : 336 : handle = NULL;
2797 : 336 : mpd.do_map = 0;
2798 : : }
2799 : : /* Unlock pages we didn't use */
2800 : 336 : mpage_release_unused_pages(&mpd, give_up_on_write);
2801 : : /* Submit prepared bio */
2802 : 336 : ext4_io_submit(&mpd.io_submit);
2803 : :
2804 : : /*
2805 : : * Drop our io_end reference we got from init. We have
2806 : : * to be careful and use deferred io_end finishing if
2807 : : * we are still holding the transaction as we can
2808 : : * release the last reference to io_end which may end
2809 : : * up doing unwritten extent conversion.
2810 : : */
2811 [ - + ]: 336 : if (handle) {
2812 : 0 : ext4_put_io_end_defer(mpd.io_submit.io_end);
2813 : 0 : ext4_journal_stop(handle);
2814 : : } else
2815 : 336 : ext4_put_io_end(mpd.io_submit.io_end);
2816 : 336 : mpd.io_submit.io_end = NULL;
2817 : :
2818 [ - + - - ]: 336 : if (ret == -ENOSPC && sbi->s_journal) {
2819 : : /*
2820 : : * Commit the transaction which would
2821 : : * free blocks released in the transaction
2822 : : * and try again
2823 : : */
2824 : 0 : jbd2_journal_force_commit_nested(sbi->s_journal);
2825 : 0 : ret = 0;
2826 : 0 : continue;
2827 : : }
2828 : : /* Fatal error - ENOMEM, EIO... */
2829 [ - + ]: 336 : if (ret)
2830 : : break;
2831 : : }
2832 : 168 : unplug:
2833 : 168 : blk_finish_plug(&plug);
2834 [ - + - - ]: 168 : if (!ret && !cycled && wbc->nr_to_write > 0) {
2835 : 0 : cycled = 1;
2836 : 0 : mpd.last_page = writeback_index - 1;
2837 : 0 : mpd.first_page = 0;
2838 : 0 : goto retry;
2839 : : }
2840 : :
2841 : : /* Update index */
2842 [ + - - + : 168 : if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
- + ]
2843 : : /*
2844 : : * Set the writeback_index so that range_cyclic
2845 : : * mode will write it back later
2846 : : */
2847 : 168 : mapping->writeback_index = mpd.first_page;
2848 : :
2849 : 0 : out_writepages:
2850 : 168 : trace_ext4_writepages_result(inode, wbc, ret,
2851 : 168 : nr_to_write - wbc->nr_to_write);
2852 : 168 : percpu_up_read(&sbi->s_writepages_rwsem);
2853 : 168 : return ret;
2854 : : }
2855 : :
2856 : : static int ext4_dax_writepages(struct address_space *mapping,
2857 : : struct writeback_control *wbc)
2858 : : {
2859 : : int ret;
2860 : : long nr_to_write = wbc->nr_to_write;
2861 : : struct inode *inode = mapping->host;
2862 : : struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb);
2863 : :
2864 : : if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2865 : : return -EIO;
2866 : :
2867 : : percpu_down_read(&sbi->s_writepages_rwsem);
2868 : : trace_ext4_writepages(inode, wbc);
2869 : :
2870 : : ret = dax_writeback_mapping_range(mapping, sbi->s_daxdev, wbc);
2871 : : trace_ext4_writepages_result(inode, wbc, ret,
2872 : : nr_to_write - wbc->nr_to_write);
2873 : : percpu_up_read(&sbi->s_writepages_rwsem);
2874 : : return ret;
2875 : : }
2876 : :
2877 : 18183 : static int ext4_nonda_switch(struct super_block *sb)
2878 : : {
2879 : 18183 : s64 free_clusters, dirty_clusters;
2880 : 18183 : struct ext4_sb_info *sbi = EXT4_SB(sb);
2881 : :
2882 : : /*
2883 : : * switch to non delalloc mode if we are running low
2884 : : * on free block. The free block accounting via percpu
2885 : : * counters can get slightly wrong with percpu_counter_batch getting
2886 : : * accumulated on each CPU without updating global counters
2887 : : * Delalloc need an accurate free block accounting. So switch
2888 : : * to non delalloc when we are near to error range.
2889 : : */
2890 : 18183 : free_clusters =
2891 : 18183 : percpu_counter_read_positive(&sbi->s_freeclusters_counter);
2892 : 18183 : dirty_clusters =
2893 : 18183 : percpu_counter_read_positive(&sbi->s_dirtyclusters_counter);
2894 : : /*
2895 : : * Start pushing delalloc when 1/2 of free blocks are dirty.
2896 : : */
2897 [ + + - + ]: 18183 : if (dirty_clusters && (free_clusters < 2 * dirty_clusters))
2898 : 0 : try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE);
2899 : :
2900 [ + - ]: 18183 : if (2 * free_clusters < 3 * dirty_clusters ||
2901 [ - + ]: 18183 : free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) {
2902 : : /*
2903 : : * free block count is less than 150% of dirty blocks
2904 : : * or free blocks is less than watermark
2905 : : */
2906 : 0 : return 1;
2907 : : }
2908 : : return 0;
2909 : : }
2910 : :
2911 : : /* We always reserve for an inode update; the superblock could be there too */
2912 : 18183 : static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len)
2913 : : {
2914 [ - + ]: 18183 : if (likely(ext4_has_feature_large_file(inode->i_sb)))
2915 : : return 1;
2916 : :
2917 [ # # ]: 0 : if (pos + len <= 0x7fffffffULL)
2918 : 0 : return 1;
2919 : :
2920 : : /* We might need to update the superblock to set LARGE_FILE */
2921 : : return 2;
2922 : : }
2923 : :
2924 : 18183 : static int ext4_da_write_begin(struct file *file, struct address_space *mapping,
2925 : : loff_t pos, unsigned len, unsigned flags,
2926 : : struct page **pagep, void **fsdata)
2927 : : {
2928 : 18183 : int ret, retries = 0;
2929 : 18183 : struct page *page;
2930 : 18183 : pgoff_t index;
2931 : 18183 : struct inode *inode = mapping->host;
2932 : 18183 : handle_t *handle;
2933 : :
2934 [ + - ]: 18183 : if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
2935 : : return -EIO;
2936 : :
2937 : 18183 : index = pos >> PAGE_SHIFT;
2938 : :
2939 [ + - - + ]: 18183 : if (ext4_nonda_switch(inode->i_sb) || S_ISLNK(inode->i_mode) ||
2940 : : ext4_verity_in_progress(inode)) {
2941 : 0 : *fsdata = (void *)FALL_BACK_TO_NONDELALLOC;
2942 : 0 : return ext4_write_begin(file, mapping, pos,
2943 : : len, flags, pagep, fsdata);
2944 : : }
2945 : 18183 : *fsdata = (void *)0;
2946 : 18183 : trace_ext4_da_write_begin(inode, pos, len, flags);
2947 : :
2948 [ + - ]: 18183 : if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) {
2949 : 0 : ret = ext4_da_write_inline_data_begin(mapping, inode,
2950 : : pos, len, flags,
2951 : : pagep, fsdata);
2952 [ # # ]: 0 : if (ret < 0)
2953 : : return ret;
2954 [ # # ]: 0 : if (ret == 1)
2955 : : return 0;
2956 : : }
2957 : :
2958 : : /*
2959 : : * grab_cache_page_write_begin() can take a long time if the
2960 : : * system is thrashing due to memory pressure, or if the page
2961 : : * is being written back. So grab it first before we start
2962 : : * the transaction handle. This also allows us to allocate
2963 : : * the page (if needed) without using GFP_NOFS.
2964 : : */
2965 : 18183 : retry_grab:
2966 : 18183 : page = grab_cache_page_write_begin(mapping, index, flags);
2967 [ + - ]: 18183 : if (!page)
2968 : : return -ENOMEM;
2969 : 18183 : unlock_page(page);
2970 : :
2971 : : /*
2972 : : * With delayed allocation, we don't log the i_disksize update
2973 : : * if there is delayed block allocation. But we still need
2974 : : * to journalling the i_disksize update if writes to the end
2975 : : * of file which has an already mapped buffer.
2976 : : */
2977 : 18183 : retry_journal:
2978 [ - + ]: 18183 : handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
2979 : : ext4_da_write_credits(inode, pos, len));
2980 [ - + ]: 18183 : if (IS_ERR(handle)) {
2981 : 0 : put_page(page);
2982 : 0 : return PTR_ERR(handle);
2983 : : }
2984 : :
2985 : 18183 : lock_page(page);
2986 [ - + ]: 18183 : if (page->mapping != mapping) {
2987 : : /* The page got truncated from under us */
2988 : 0 : unlock_page(page);
2989 : 0 : put_page(page);
2990 : 0 : ext4_journal_stop(handle);
2991 : 0 : goto retry_grab;
2992 : : }
2993 : : /* In case writeback began while the page was unlocked */
2994 : 18183 : wait_for_stable_page(page);
2995 : :
2996 : : #ifdef CONFIG_FS_ENCRYPTION
2997 : : ret = ext4_block_write_begin(page, pos, len,
2998 : : ext4_da_get_block_prep);
2999 : : #else
3000 : 18183 : ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep);
3001 : : #endif
3002 [ - + ]: 18183 : if (ret < 0) {
3003 : 0 : unlock_page(page);
3004 : 0 : ext4_journal_stop(handle);
3005 : : /*
3006 : : * block_write_begin may have instantiated a few blocks
3007 : : * outside i_size. Trim these off again. Don't need
3008 : : * i_size_read because we hold i_mutex.
3009 : : */
3010 [ # # ]: 0 : if (pos + len > inode->i_size)
3011 : 0 : ext4_truncate_failed_write(inode);
3012 : :
3013 [ # # # # ]: 0 : if (ret == -ENOSPC &&
3014 : 0 : ext4_should_retry_alloc(inode->i_sb, &retries))
3015 : 0 : goto retry_journal;
3016 : :
3017 : 0 : put_page(page);
3018 : 0 : return ret;
3019 : : }
3020 : :
3021 : 18183 : *pagep = page;
3022 : 18183 : return ret;
3023 : : }
3024 : :
3025 : : /*
3026 : : * Check if we should update i_disksize
3027 : : * when write to the end of file but not require block allocation
3028 : : */
3029 : 18183 : static int ext4_da_should_update_i_disksize(struct page *page,
3030 : : unsigned long offset)
3031 : : {
3032 : 18183 : struct buffer_head *bh;
3033 : 18183 : struct inode *inode = page->mapping->host;
3034 : 18183 : unsigned int idx;
3035 : 18183 : int i;
3036 : :
3037 [ - + ]: 18183 : bh = page_buffers(page);
3038 : 18183 : idx = offset >> inode->i_blkbits;
3039 : :
3040 [ - + ]: 18183 : for (i = 0; i < idx; i++)
3041 : 0 : bh = bh->b_this_page;
3042 : :
3043 [ + - - + : 36366 : if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh))
- - ]
3044 : 18183 : return 0;
3045 : : return 1;
3046 : : }
3047 : :
3048 : 18183 : static int ext4_da_write_end(struct file *file,
3049 : : struct address_space *mapping,
3050 : : loff_t pos, unsigned len, unsigned copied,
3051 : : struct page *page, void *fsdata)
3052 : : {
3053 : 18183 : struct inode *inode = mapping->host;
3054 : 18183 : int ret = 0, ret2;
3055 [ - + ]: 18183 : handle_t *handle = ext4_journal_current_handle();
3056 : 18183 : loff_t new_i_size;
3057 : 18183 : unsigned long start, end;
3058 : 18183 : int write_mode = (int)(unsigned long)fsdata;
3059 : :
3060 [ - + ]: 18183 : if (write_mode == FALL_BACK_TO_NONDELALLOC)
3061 : 0 : return ext4_write_end(file, mapping, pos,
3062 : : len, copied, page, fsdata);
3063 : :
3064 : 18183 : trace_ext4_da_write_end(inode, pos, len, copied);
3065 : 18183 : start = pos & (PAGE_SIZE - 1);
3066 : 18183 : end = start + copied - 1;
3067 : :
3068 : : /*
3069 : : * generic_write_end() will run mark_inode_dirty() if i_size
3070 : : * changes. So let's piggyback the i_disksize mark_inode_dirty
3071 : : * into that.
3072 : : */
3073 : 18183 : new_i_size = pos + copied;
3074 [ + - + - ]: 18183 : if (copied && new_i_size > EXT4_I(inode)->i_disksize) {
3075 [ - + ]: 36366 : if (ext4_has_inline_data(inode) ||
3076 : 18183 : ext4_da_should_update_i_disksize(page, end)) {
3077 : 0 : ext4_update_i_disksize(inode, new_i_size);
3078 : : /* We need to mark inode dirty even if
3079 : : * new_i_size is less that inode->i_size
3080 : : * bu greater than i_disksize.(hint delalloc)
3081 : : */
3082 : 0 : ext4_mark_inode_dirty(handle, inode);
3083 : : }
3084 : : }
3085 : :
3086 [ + - - + ]: 36366 : if (write_mode != CONVERT_INLINE_DATA &&
3087 : : ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) &&
3088 : : ext4_has_inline_data(inode))
3089 : 0 : ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied,
3090 : : page);
3091 : : else
3092 : 18183 : ret2 = generic_write_end(file, mapping, pos, len, copied,
3093 : : page, fsdata);
3094 : :
3095 : 18183 : copied = ret2;
3096 : 18183 : if (ret2 < 0)
3097 : : ret = ret2;
3098 : 18183 : ret2 = ext4_journal_stop(handle);
3099 [ + - ]: 18183 : if (!ret)
3100 : 18183 : ret = ret2;
3101 : :
3102 [ + - ]: 18183 : return ret ? ret : copied;
3103 : : }
3104 : :
3105 : : /*
3106 : : * Force all delayed allocation blocks to be allocated for a given inode.
3107 : : */
3108 : 168 : int ext4_alloc_da_blocks(struct inode *inode)
3109 : : {
3110 : 168 : trace_ext4_alloc_da_blocks(inode);
3111 : :
3112 [ + + ]: 168 : if (!EXT4_I(inode)->i_reserved_data_blocks)
3113 : : return 0;
3114 : :
3115 : : /*
3116 : : * We do something simple for now. The filemap_flush() will
3117 : : * also start triggering a write of the data blocks, which is
3118 : : * not strictly speaking necessary (and for users of
3119 : : * laptop_mode, not even desirable). However, to do otherwise
3120 : : * would require replicating code paths in:
3121 : : *
3122 : : * ext4_writepages() ->
3123 : : * write_cache_pages() ---> (via passed in callback function)
3124 : : * __mpage_da_writepage() -->
3125 : : * mpage_add_bh_to_extent()
3126 : : * mpage_da_map_blocks()
3127 : : *
3128 : : * The problem is that write_cache_pages(), located in
3129 : : * mm/page-writeback.c, marks pages clean in preparation for
3130 : : * doing I/O, which is not desirable if we're not planning on
3131 : : * doing I/O at all.
3132 : : *
3133 : : * We could call write_cache_pages(), and then redirty all of
3134 : : * the pages by calling redirty_page_for_writepage() but that
3135 : : * would be ugly in the extreme. So instead we would need to
3136 : : * replicate parts of the code in the above functions,
3137 : : * simplifying them because we wouldn't actually intend to
3138 : : * write out the pages, but rather only collect contiguous
3139 : : * logical block extents, call the multi-block allocator, and
3140 : : * then update the buffer heads with the block allocations.
3141 : : *
3142 : : * For now, though, we'll cheat by calling filemap_flush(),
3143 : : * which will map the blocks, and start the I/O, but not
3144 : : * actually wait for the I/O to complete.
3145 : : */
3146 : 21 : return filemap_flush(inode->i_mapping);
3147 : : }
3148 : :
3149 : : /*
3150 : : * bmap() is special. It gets used by applications such as lilo and by
3151 : : * the swapper to find the on-disk block of a specific piece of data.
3152 : : *
3153 : : * Naturally, this is dangerous if the block concerned is still in the
3154 : : * journal. If somebody makes a swapfile on an ext4 data-journaling
3155 : : * filesystem and enables swap, then they may get a nasty shock when the
3156 : : * data getting swapped to that swapfile suddenly gets overwritten by
3157 : : * the original zero's written out previously to the journal and
3158 : : * awaiting writeback in the kernel's buffer cache.
3159 : : *
3160 : : * So, if we see any bmap calls here on a modified, data-journaled file,
3161 : : * take extra steps to flush any blocks which might be in the cache.
3162 : : */
3163 : 2157 : static sector_t ext4_bmap(struct address_space *mapping, sector_t block)
3164 : : {
3165 : 2157 : struct inode *inode = mapping->host;
3166 : 2157 : journal_t *journal;
3167 : 2157 : int err;
3168 : :
3169 : : /*
3170 : : * We can get here for an inline file via the FIBMAP ioctl
3171 : : */
3172 : 2157 : if (ext4_has_inline_data(inode))
3173 : : return 0;
3174 : :
3175 [ - + - - ]: 2157 : if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
3176 [ # # ]: 0 : test_opt(inode->i_sb, DELALLOC)) {
3177 : : /*
3178 : : * With delalloc we want to sync the file
3179 : : * so that we can make sure we allocate
3180 : : * blocks for file
3181 : : */
3182 : 0 : filemap_write_and_wait(mapping);
3183 : : }
3184 : :
3185 [ + + - + ]: 4272 : if (EXT4_JOURNAL(inode) &&
3186 : : ext4_test_inode_state(inode, EXT4_STATE_JDATA)) {
3187 : : /*
3188 : : * This is a REALLY heavyweight approach, but the use of
3189 : : * bmap on dirty files is expected to be extremely rare:
3190 : : * only if we run lilo or swapon on a freshly made file
3191 : : * do we expect this to happen.
3192 : : *
3193 : : * (bmap requires CAP_SYS_RAWIO so this does not
3194 : : * represent an unprivileged user DOS attack --- we'd be
3195 : : * in trouble if mortal users could trigger this path at
3196 : : * will.)
3197 : : *
3198 : : * NB. EXT4_STATE_JDATA is not set on files other than
3199 : : * regular files. If somebody wants to bmap a directory
3200 : : * or symlink and gets confused because the buffer
3201 : : * hasn't yet been flushed to disk, they deserve
3202 : : * everything they get.
3203 : : */
3204 : :
3205 : 0 : ext4_clear_inode_state(inode, EXT4_STATE_JDATA);
3206 : 0 : journal = EXT4_JOURNAL(inode);
3207 : 0 : jbd2_journal_lock_updates(journal);
3208 : 0 : err = jbd2_journal_flush(journal);
3209 : 0 : jbd2_journal_unlock_updates(journal);
3210 : :
3211 [ # # ]: 0 : if (err)
3212 : : return 0;
3213 : : }
3214 : :
3215 : 2157 : return generic_block_bmap(mapping, block, ext4_get_block);
3216 : : }
3217 : :
3218 : 21 : static int ext4_readpage(struct file *file, struct page *page)
3219 : : {
3220 : 21 : int ret = -EAGAIN;
3221 : 21 : struct inode *inode = page->mapping->host;
3222 : :
3223 : 21 : trace_ext4_readpage(page);
3224 : :
3225 : 21 : if (ext4_has_inline_data(inode))
3226 : 0 : ret = ext4_readpage_inline(inode, page);
3227 : :
3228 [ + - ]: 21 : if (ret == -EAGAIN)
3229 : 21 : return ext4_mpage_readpages(page->mapping, NULL, page, 1,
3230 : : false);
3231 : :
3232 : : return ret;
3233 : : }
3234 : :
3235 : : static int
3236 : 13921 : ext4_readpages(struct file *file, struct address_space *mapping,
3237 : : struct list_head *pages, unsigned nr_pages)
3238 : : {
3239 : 13921 : struct inode *inode = mapping->host;
3240 : :
3241 : : /* If the file has inline data, no need to do readpages. */
3242 : 13921 : if (ext4_has_inline_data(inode))
3243 : : return 0;
3244 : :
3245 : 13921 : return ext4_mpage_readpages(mapping, pages, NULL, nr_pages, true);
3246 : : }
3247 : :
3248 : 0 : static void ext4_invalidatepage(struct page *page, unsigned int offset,
3249 : : unsigned int length)
3250 : : {
3251 : 0 : trace_ext4_invalidatepage(page, offset, length);
3252 : :
3253 : : /* No journalling happens on data buffers when this function is used */
3254 [ # # # # : 0 : WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page)));
# # # # ]
3255 : :
3256 : 0 : block_invalidatepage(page, offset, length);
3257 : 0 : }
3258 : :
3259 : 0 : static int __ext4_journalled_invalidatepage(struct page *page,
3260 : : unsigned int offset,
3261 : : unsigned int length)
3262 : : {
3263 : 0 : journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3264 : :
3265 : 0 : trace_ext4_journalled_invalidatepage(page, offset, length);
3266 : :
3267 : : /*
3268 : : * If it's a full truncate we just forget about the pending dirtying
3269 : : */
3270 [ # # ]: 0 : if (offset == 0 && length == PAGE_SIZE)
3271 : 0 : ClearPageChecked(page);
3272 : :
3273 : 0 : return jbd2_journal_invalidatepage(journal, page, offset, length);
3274 : : }
3275 : :
3276 : : /* Wrapper for aops... */
3277 : 0 : static void ext4_journalled_invalidatepage(struct page *page,
3278 : : unsigned int offset,
3279 : : unsigned int length)
3280 : : {
3281 [ # # ]: 0 : WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0);
3282 : 0 : }
3283 : :
3284 : 0 : static int ext4_releasepage(struct page *page, gfp_t wait)
3285 : : {
3286 : 0 : journal_t *journal = EXT4_JOURNAL(page->mapping->host);
3287 : :
3288 : 0 : trace_ext4_releasepage(page);
3289 : :
3290 : : /* Page has dirty journalled data -> cannot release */
3291 [ # # ]: 0 : if (PageChecked(page))
3292 : : return 0;
3293 [ # # ]: 0 : if (journal)
3294 : 0 : return jbd2_journal_try_to_free_buffers(journal, page, wait);
3295 : : else
3296 : 0 : return try_to_free_buffers(page);
3297 : : }
3298 : :
3299 : 0 : static bool ext4_inode_datasync_dirty(struct inode *inode)
3300 : : {
3301 [ # # ]: 0 : journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
3302 : :
3303 [ # # ]: 0 : if (journal)
3304 : 0 : return !jbd2_transaction_committed(journal,
3305 : 0 : EXT4_I(inode)->i_datasync_tid);
3306 : : /* Any metadata buffers to write? */
3307 [ # # ]: 0 : if (!list_empty(&inode->i_mapping->private_list))
3308 : : return true;
3309 : 0 : return inode->i_state & I_DIRTY_DATASYNC;
3310 : : }
3311 : :
3312 : 0 : static void ext4_set_iomap(struct inode *inode, struct iomap *iomap,
3313 : : struct ext4_map_blocks *map, loff_t offset,
3314 : : loff_t length)
3315 : : {
3316 : 0 : u8 blkbits = inode->i_blkbits;
3317 : :
3318 : : /*
3319 : : * Writes that span EOF might trigger an I/O size update on completion,
3320 : : * so consider them to be dirty for the purpose of O_DSYNC, even if
3321 : : * there is no other metadata changes being made or are pending.
3322 : : */
3323 : 0 : iomap->flags = 0;
3324 [ # # # # ]: 0 : if (ext4_inode_datasync_dirty(inode) ||
3325 [ # # ]: 0 : offset + length > i_size_read(inode))
3326 : 0 : iomap->flags |= IOMAP_F_DIRTY;
3327 : :
3328 [ # # ]: 0 : if (map->m_flags & EXT4_MAP_NEW)
3329 : 0 : iomap->flags |= IOMAP_F_NEW;
3330 : :
3331 : 0 : iomap->bdev = inode->i_sb->s_bdev;
3332 [ # # ]: 0 : iomap->dax_dev = EXT4_SB(inode->i_sb)->s_daxdev;
3333 : 0 : iomap->offset = (u64) map->m_lblk << blkbits;
3334 : 0 : iomap->length = (u64) map->m_len << blkbits;
3335 : :
3336 : : /*
3337 : : * Flags passed to ext4_map_blocks() for direct I/O writes can result
3338 : : * in m_flags having both EXT4_MAP_MAPPED and EXT4_MAP_UNWRITTEN bits
3339 : : * set. In order for any allocated unwritten extents to be converted
3340 : : * into written extents correctly within the ->end_io() handler, we
3341 : : * need to ensure that the iomap->type is set appropriately. Hence, the
3342 : : * reason why we need to check whether the EXT4_MAP_UNWRITTEN bit has
3343 : : * been set first.
3344 : : */
3345 [ # # ]: 0 : if (map->m_flags & EXT4_MAP_UNWRITTEN) {
3346 : 0 : iomap->type = IOMAP_UNWRITTEN;
3347 : 0 : iomap->addr = (u64) map->m_pblk << blkbits;
3348 [ # # ]: 0 : } else if (map->m_flags & EXT4_MAP_MAPPED) {
3349 : 0 : iomap->type = IOMAP_MAPPED;
3350 : 0 : iomap->addr = (u64) map->m_pblk << blkbits;
3351 : : } else {
3352 : 0 : iomap->type = IOMAP_HOLE;
3353 : 0 : iomap->addr = IOMAP_NULL_ADDR;
3354 : : }
3355 : 0 : }
3356 : :
3357 : 0 : static int ext4_iomap_alloc(struct inode *inode, struct ext4_map_blocks *map,
3358 : : unsigned int flags)
3359 : : {
3360 : 0 : handle_t *handle;
3361 : 0 : u8 blkbits = inode->i_blkbits;
3362 : 0 : int ret, dio_credits, m_flags = 0, retries = 0;
3363 : :
3364 : : /*
3365 : : * Trim the mapping request to the maximum value that we can map at
3366 : : * once for direct I/O.
3367 : : */
3368 [ # # ]: 0 : if (map->m_len > DIO_MAX_BLOCKS)
3369 : 0 : map->m_len = DIO_MAX_BLOCKS;
3370 : 0 : dio_credits = ext4_chunk_trans_blocks(inode, map->m_len);
3371 : :
3372 : 0 : retry:
3373 : : /*
3374 : : * Either we allocate blocks and then don't get an unwritten extent, so
3375 : : * in that case we have reserved enough credits. Or, the blocks are
3376 : : * already allocated and unwritten. In that case, the extent conversion
3377 : : * fits into the credits as well.
3378 : : */
3379 : 0 : handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits);
3380 [ # # ]: 0 : if (IS_ERR(handle))
3381 : 0 : return PTR_ERR(handle);
3382 : :
3383 : : /*
3384 : : * DAX and direct I/O are the only two operations that are currently
3385 : : * supported with IOMAP_WRITE.
3386 : : */
3387 [ # # ]: 0 : WARN_ON(!IS_DAX(inode) && !(flags & IOMAP_DIRECT));
3388 : 0 : if (IS_DAX(inode))
3389 : : m_flags = EXT4_GET_BLOCKS_CREATE_ZERO;
3390 : : /*
3391 : : * We use i_size instead of i_disksize here because delalloc writeback
3392 : : * can complete at any point during the I/O and subsequently push the
3393 : : * i_disksize out to i_size. This could be beyond where direct I/O is
3394 : : * happening and thus expose allocated blocks to direct I/O reads.
3395 : : */
3396 [ # # ]: 0 : else if ((map->m_lblk * (1 << blkbits)) >= i_size_read(inode))
3397 : : m_flags = EXT4_GET_BLOCKS_CREATE;
3398 [ # # ]: 0 : else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
3399 : 0 : m_flags = EXT4_GET_BLOCKS_IO_CREATE_EXT;
3400 : :
3401 : 0 : ret = ext4_map_blocks(handle, inode, map, m_flags);
3402 : :
3403 : : /*
3404 : : * We cannot fill holes in indirect tree based inodes as that could
3405 : : * expose stale data in the case of a crash. Use the magic error code
3406 : : * to fallback to buffered I/O.
3407 : : */
3408 [ # # ]: 0 : if (!m_flags && !ret)
3409 : 0 : ret = -ENOTBLK;
3410 : :
3411 : 0 : ext4_journal_stop(handle);
3412 [ # # # # ]: 0 : if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
3413 : 0 : goto retry;
3414 : :
3415 : : return ret;
3416 : : }
3417 : :
3418 : :
3419 : 0 : static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
3420 : : unsigned flags, struct iomap *iomap, struct iomap *srcmap)
3421 : : {
3422 : 0 : int ret;
3423 : 0 : struct ext4_map_blocks map;
3424 : 0 : u8 blkbits = inode->i_blkbits;
3425 : :
3426 [ # # ]: 0 : if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
3427 : : return -EINVAL;
3428 : :
3429 [ # # # # ]: 0 : if (WARN_ON_ONCE(ext4_has_inline_data(inode)))
3430 : : return -ERANGE;
3431 : :
3432 : : /*
3433 : : * Calculate the first and last logical blocks respectively.
3434 : : */
3435 : 0 : map.m_lblk = offset >> blkbits;
3436 : 0 : map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits,
3437 : 0 : EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1;
3438 : :
3439 [ # # ]: 0 : if (flags & IOMAP_WRITE)
3440 : 0 : ret = ext4_iomap_alloc(inode, &map, flags);
3441 : : else
3442 : 0 : ret = ext4_map_blocks(NULL, inode, &map, 0);
3443 : :
3444 [ # # ]: 0 : if (ret < 0)
3445 : : return ret;
3446 : :
3447 : 0 : ext4_set_iomap(inode, iomap, &map, offset, length);
3448 : :
3449 : 0 : return 0;
3450 : : }
3451 : :
3452 : 0 : static int ext4_iomap_overwrite_begin(struct inode *inode, loff_t offset,
3453 : : loff_t length, unsigned flags, struct iomap *iomap,
3454 : : struct iomap *srcmap)
3455 : : {
3456 : 0 : int ret;
3457 : :
3458 : : /*
3459 : : * Even for writes we don't need to allocate blocks, so just pretend
3460 : : * we are reading to save overhead of starting a transaction.
3461 : : */
3462 : 0 : flags &= ~IOMAP_WRITE;
3463 : 0 : ret = ext4_iomap_begin(inode, offset, length, flags, iomap, srcmap);
3464 [ # # ]: 0 : WARN_ON_ONCE(iomap->type != IOMAP_MAPPED);
3465 : 0 : return ret;
3466 : : }
3467 : :
3468 : 0 : static int ext4_iomap_end(struct inode *inode, loff_t offset, loff_t length,
3469 : : ssize_t written, unsigned flags, struct iomap *iomap)
3470 : : {
3471 : : /*
3472 : : * Check to see whether an error occurred while writing out the data to
3473 : : * the allocated blocks. If so, return the magic error code so that we
3474 : : * fallback to buffered I/O and attempt to complete the remainder of
3475 : : * the I/O. Any blocks that may have been allocated in preparation for
3476 : : * the direct I/O will be reused during buffered I/O.
3477 : : */
3478 [ # # # # ]: 0 : if (flags & (IOMAP_WRITE | IOMAP_DIRECT) && written == 0)
3479 : 0 : return -ENOTBLK;
3480 : :
3481 : : return 0;
3482 : : }
3483 : :
3484 : : const struct iomap_ops ext4_iomap_ops = {
3485 : : .iomap_begin = ext4_iomap_begin,
3486 : : .iomap_end = ext4_iomap_end,
3487 : : };
3488 : :
3489 : : const struct iomap_ops ext4_iomap_overwrite_ops = {
3490 : : .iomap_begin = ext4_iomap_overwrite_begin,
3491 : : .iomap_end = ext4_iomap_end,
3492 : : };
3493 : :
3494 : : static bool ext4_iomap_is_delalloc(struct inode *inode,
3495 : : struct ext4_map_blocks *map)
3496 : : {
3497 : : struct extent_status es;
3498 : : ext4_lblk_t offset = 0, end = map->m_lblk + map->m_len - 1;
3499 : :
3500 : : ext4_es_find_extent_range(inode, &ext4_es_is_delayed,
3501 : : map->m_lblk, end, &es);
3502 : :
3503 : : if (!es.es_len || es.es_lblk > end)
3504 : : return false;
3505 : :
3506 : : if (es.es_lblk > map->m_lblk) {
3507 : : map->m_len = es.es_lblk - map->m_lblk;
3508 : : return false;
3509 : : }
3510 : :
3511 : : offset = map->m_lblk - es.es_lblk;
3512 : : map->m_len = es.es_len - offset;
3513 : :
3514 : : return true;
3515 : : }
3516 : :
3517 : 0 : static int ext4_iomap_begin_report(struct inode *inode, loff_t offset,
3518 : : loff_t length, unsigned int flags,
3519 : : struct iomap *iomap, struct iomap *srcmap)
3520 : : {
3521 : 0 : int ret;
3522 : 0 : bool delalloc = false;
3523 : 0 : struct ext4_map_blocks map;
3524 : 0 : u8 blkbits = inode->i_blkbits;
3525 : :
3526 [ # # ]: 0 : if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK)
3527 : : return -EINVAL;
3528 : :
3529 : 0 : if (ext4_has_inline_data(inode)) {
3530 : 0 : ret = ext4_inline_data_iomap(inode, iomap);
3531 [ # # ]: 0 : if (ret != -EAGAIN) {
3532 [ # # # # ]: 0 : if (ret == 0 && offset >= iomap->length)
3533 : 0 : ret = -ENOENT;
3534 : 0 : return ret;
3535 : : }
3536 : : }
3537 : :
3538 : : /*
3539 : : * Calculate the first and last logical block respectively.
3540 : : */
3541 : 0 : map.m_lblk = offset >> blkbits;
3542 : 0 : map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits,
3543 : 0 : EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1;
3544 : :
3545 : 0 : ret = ext4_map_blocks(NULL, inode, &map, 0);
3546 [ # # ]: 0 : if (ret < 0)
3547 : : return ret;
3548 [ # # ]: 0 : if (ret == 0)
3549 : 0 : delalloc = ext4_iomap_is_delalloc(inode, &map);
3550 : :
3551 : 0 : ext4_set_iomap(inode, iomap, &map, offset, length);
3552 [ # # # # ]: 0 : if (delalloc && iomap->type == IOMAP_HOLE)
3553 : 0 : iomap->type = IOMAP_DELALLOC;
3554 : :
3555 : : return 0;
3556 : : }
3557 : :
3558 : : const struct iomap_ops ext4_iomap_report_ops = {
3559 : : .iomap_begin = ext4_iomap_begin_report,
3560 : : };
3561 : :
3562 : : /*
3563 : : * Pages can be marked dirty completely asynchronously from ext4's journalling
3564 : : * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3565 : : * much here because ->set_page_dirty is called under VFS locks. The page is
3566 : : * not necessarily locked.
3567 : : *
3568 : : * We cannot just dirty the page and leave attached buffers clean, because the
3569 : : * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3570 : : * or jbddirty because all the journalling code will explode.
3571 : : *
3572 : : * So what we do is to mark the page "pending dirty" and next time writepage
3573 : : * is called, propagate that into the buffers appropriately.
3574 : : */
3575 : 0 : static int ext4_journalled_set_page_dirty(struct page *page)
3576 : : {
3577 : 0 : SetPageChecked(page);
3578 : 0 : return __set_page_dirty_nobuffers(page);
3579 : : }
3580 : :
3581 : 0 : static int ext4_set_page_dirty(struct page *page)
3582 : : {
3583 [ # # # # : 0 : WARN_ON_ONCE(!PageLocked(page) && !PageDirty(page));
# # # # ]
3584 [ # # ]: 0 : WARN_ON_ONCE(!page_has_buffers(page));
3585 : 0 : return __set_page_dirty_buffers(page);
3586 : : }
3587 : :
3588 : : static const struct address_space_operations ext4_aops = {
3589 : : .readpage = ext4_readpage,
3590 : : .readpages = ext4_readpages,
3591 : : .writepage = ext4_writepage,
3592 : : .writepages = ext4_writepages,
3593 : : .write_begin = ext4_write_begin,
3594 : : .write_end = ext4_write_end,
3595 : : .set_page_dirty = ext4_set_page_dirty,
3596 : : .bmap = ext4_bmap,
3597 : : .invalidatepage = ext4_invalidatepage,
3598 : : .releasepage = ext4_releasepage,
3599 : : .direct_IO = noop_direct_IO,
3600 : : .migratepage = buffer_migrate_page,
3601 : : .is_partially_uptodate = block_is_partially_uptodate,
3602 : : .error_remove_page = generic_error_remove_page,
3603 : : };
3604 : :
3605 : : static const struct address_space_operations ext4_journalled_aops = {
3606 : : .readpage = ext4_readpage,
3607 : : .readpages = ext4_readpages,
3608 : : .writepage = ext4_writepage,
3609 : : .writepages = ext4_writepages,
3610 : : .write_begin = ext4_write_begin,
3611 : : .write_end = ext4_journalled_write_end,
3612 : : .set_page_dirty = ext4_journalled_set_page_dirty,
3613 : : .bmap = ext4_bmap,
3614 : : .invalidatepage = ext4_journalled_invalidatepage,
3615 : : .releasepage = ext4_releasepage,
3616 : : .direct_IO = noop_direct_IO,
3617 : : .is_partially_uptodate = block_is_partially_uptodate,
3618 : : .error_remove_page = generic_error_remove_page,
3619 : : };
3620 : :
3621 : : static const struct address_space_operations ext4_da_aops = {
3622 : : .readpage = ext4_readpage,
3623 : : .readpages = ext4_readpages,
3624 : : .writepage = ext4_writepage,
3625 : : .writepages = ext4_writepages,
3626 : : .write_begin = ext4_da_write_begin,
3627 : : .write_end = ext4_da_write_end,
3628 : : .set_page_dirty = ext4_set_page_dirty,
3629 : : .bmap = ext4_bmap,
3630 : : .invalidatepage = ext4_invalidatepage,
3631 : : .releasepage = ext4_releasepage,
3632 : : .direct_IO = noop_direct_IO,
3633 : : .migratepage = buffer_migrate_page,
3634 : : .is_partially_uptodate = block_is_partially_uptodate,
3635 : : .error_remove_page = generic_error_remove_page,
3636 : : };
3637 : :
3638 : : static const struct address_space_operations ext4_dax_aops = {
3639 : : .writepages = ext4_dax_writepages,
3640 : : .direct_IO = noop_direct_IO,
3641 : : .set_page_dirty = noop_set_page_dirty,
3642 : : .bmap = ext4_bmap,
3643 : : .invalidatepage = noop_invalidatepage,
3644 : : };
3645 : :
3646 : 16611 : void ext4_set_aops(struct inode *inode)
3647 : : {
3648 [ - - + ]: 16611 : switch (ext4_inode_journal_mode(inode)) {
3649 : : case EXT4_INODE_ORDERED_DATA_MODE:
3650 : : case EXT4_INODE_WRITEBACK_DATA_MODE:
3651 : 16611 : break;
3652 : 0 : case EXT4_INODE_JOURNAL_DATA_MODE:
3653 : 0 : inode->i_mapping->a_ops = &ext4_journalled_aops;
3654 : 0 : return;
3655 : 0 : default:
3656 : 0 : BUG();
3657 : : }
3658 : 16611 : if (IS_DAX(inode))
3659 : : inode->i_mapping->a_ops = &ext4_dax_aops;
3660 [ + - ]: 16611 : else if (test_opt(inode->i_sb, DELALLOC))
3661 : 16611 : inode->i_mapping->a_ops = &ext4_da_aops;
3662 : : else
3663 : 0 : inode->i_mapping->a_ops = &ext4_aops;
3664 : : }
3665 : :
3666 : 0 : static int __ext4_block_zero_page_range(handle_t *handle,
3667 : : struct address_space *mapping, loff_t from, loff_t length)
3668 : : {
3669 : 0 : ext4_fsblk_t index = from >> PAGE_SHIFT;
3670 : 0 : unsigned offset = from & (PAGE_SIZE-1);
3671 : 0 : unsigned blocksize, pos;
3672 : 0 : ext4_lblk_t iblock;
3673 : 0 : struct inode *inode = mapping->host;
3674 : 0 : struct buffer_head *bh;
3675 : 0 : struct page *page;
3676 : 0 : int err = 0;
3677 : :
3678 : 0 : page = find_or_create_page(mapping, from >> PAGE_SHIFT,
3679 : : mapping_gfp_constraint(mapping, ~__GFP_FS));
3680 [ # # ]: 0 : if (!page)
3681 : : return -ENOMEM;
3682 : :
3683 : 0 : blocksize = inode->i_sb->s_blocksize;
3684 : :
3685 : 0 : iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
3686 : :
3687 [ # # ]: 0 : if (!page_has_buffers(page))
3688 : 0 : create_empty_buffers(page, blocksize, 0);
3689 : :
3690 : : /* Find the buffer that contains "offset" */
3691 [ # # ]: 0 : bh = page_buffers(page);
3692 : 0 : pos = blocksize;
3693 [ # # ]: 0 : while (offset >= pos) {
3694 : 0 : bh = bh->b_this_page;
3695 : 0 : iblock++;
3696 : 0 : pos += blocksize;
3697 : : }
3698 [ # # ]: 0 : if (buffer_freed(bh)) {
3699 : 0 : BUFFER_TRACE(bh, "freed: skip");
3700 : 0 : goto unlock;
3701 : : }
3702 [ # # ]: 0 : if (!buffer_mapped(bh)) {
3703 : 0 : BUFFER_TRACE(bh, "unmapped");
3704 : 0 : ext4_get_block(inode, iblock, bh, 0);
3705 : : /* unmapped? It's a hole - nothing to do */
3706 [ # # ]: 0 : if (!buffer_mapped(bh)) {
3707 : 0 : BUFFER_TRACE(bh, "still unmapped");
3708 : 0 : goto unlock;
3709 : : }
3710 : : }
3711 : :
3712 : : /* Ok, it's mapped. Make sure it's up-to-date */
3713 [ # # ]: 0 : if (PageUptodate(page))
3714 : 0 : set_buffer_uptodate(bh);
3715 : :
3716 [ # # ]: 0 : if (!buffer_uptodate(bh)) {
3717 : 0 : err = -EIO;
3718 : 0 : ll_rw_block(REQ_OP_READ, 0, 1, &bh);
3719 : 0 : wait_on_buffer(bh);
3720 : : /* Uhhuh. Read error. Complain and punt. */
3721 [ # # ]: 0 : if (!buffer_uptodate(bh))
3722 : 0 : goto unlock;
3723 [ # # # # ]: 0 : if (S_ISREG(inode->i_mode) && IS_ENCRYPTED(inode)) {
3724 : : /* We expect the key to be set. */
3725 : 0 : BUG_ON(!fscrypt_has_encryption_key(inode));
3726 : : err = fscrypt_decrypt_pagecache_blocks(page, blocksize,
3727 : : bh_offset(bh));
3728 : : if (err) {
3729 : : clear_buffer_uptodate(bh);
3730 : : goto unlock;
3731 : : }
3732 : : }
3733 : : }
3734 [ # # ]: 0 : if (ext4_should_journal_data(inode)) {
3735 : 0 : BUFFER_TRACE(bh, "get write access");
3736 : 0 : err = ext4_journal_get_write_access(handle, bh);
3737 [ # # ]: 0 : if (err)
3738 : 0 : goto unlock;
3739 : : }
3740 : 0 : zero_user(page, offset, length);
3741 : 0 : BUFFER_TRACE(bh, "zeroed end of block");
3742 : :
3743 [ # # ]: 0 : if (ext4_should_journal_data(inode)) {
3744 : 0 : err = ext4_handle_dirty_metadata(handle, inode, bh);
3745 : : } else {
3746 : 0 : err = 0;
3747 : 0 : mark_buffer_dirty(bh);
3748 [ # # ]: 0 : if (ext4_should_order_data(inode))
3749 [ # # ]: 0 : err = ext4_jbd2_inode_add_write(handle, inode, from,
3750 : : length);
3751 : : }
3752 : :
3753 : 0 : unlock:
3754 : 0 : unlock_page(page);
3755 : 0 : put_page(page);
3756 : 0 : return err;
3757 : : }
3758 : :
3759 : : /*
3760 : : * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3761 : : * starting from file offset 'from'. The range to be zero'd must
3762 : : * be contained with in one block. If the specified range exceeds
3763 : : * the end of the block it will be shortened to end of the block
3764 : : * that cooresponds to 'from'
3765 : : */
3766 : 0 : static int ext4_block_zero_page_range(handle_t *handle,
3767 : : struct address_space *mapping, loff_t from, loff_t length)
3768 : : {
3769 : 0 : struct inode *inode = mapping->host;
3770 : 0 : unsigned offset = from & (PAGE_SIZE-1);
3771 : 0 : unsigned blocksize = inode->i_sb->s_blocksize;
3772 : 0 : unsigned max = blocksize - (offset & (blocksize - 1));
3773 : :
3774 : : /*
3775 : : * correct length if it does not fall between
3776 : : * 'from' and the end of the block
3777 : : */
3778 [ # # # # ]: 0 : if (length > max || length < 0)
3779 : 0 : length = max;
3780 : :
3781 : 0 : if (IS_DAX(inode)) {
3782 : : return iomap_zero_range(inode, from, length, NULL,
3783 : : &ext4_iomap_ops);
3784 : : }
3785 : 0 : return __ext4_block_zero_page_range(handle, mapping, from, length);
3786 : : }
3787 : :
3788 : : /*
3789 : : * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3790 : : * up to the end of the block which corresponds to `from'.
3791 : : * This required during truncate. We need to physically zero the tail end
3792 : : * of that block so it doesn't yield old data if the file is later grown.
3793 : : */
3794 : 0 : static int ext4_block_truncate_page(handle_t *handle,
3795 : : struct address_space *mapping, loff_t from)
3796 : : {
3797 : 0 : unsigned offset = from & (PAGE_SIZE-1);
3798 : 0 : unsigned length;
3799 : 0 : unsigned blocksize;
3800 : 0 : struct inode *inode = mapping->host;
3801 : :
3802 : : /* If we are processing an encrypted inode during orphan list handling */
3803 [ # # ]: 0 : if (IS_ENCRYPTED(inode) && !fscrypt_has_encryption_key(inode))
3804 : : return 0;
3805 : :
3806 : 0 : blocksize = inode->i_sb->s_blocksize;
3807 : 0 : length = blocksize - (offset & (blocksize - 1));
3808 : :
3809 : 0 : return ext4_block_zero_page_range(handle, mapping, from, length);
3810 : : }
3811 : :
3812 : 0 : int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode,
3813 : : loff_t lstart, loff_t length)
3814 : : {
3815 : 0 : struct super_block *sb = inode->i_sb;
3816 : 0 : struct address_space *mapping = inode->i_mapping;
3817 : 0 : unsigned partial_start, partial_end;
3818 : 0 : ext4_fsblk_t start, end;
3819 : 0 : loff_t byte_end = (lstart + length - 1);
3820 : 0 : int err = 0;
3821 : :
3822 : 0 : partial_start = lstart & (sb->s_blocksize - 1);
3823 : 0 : partial_end = byte_end & (sb->s_blocksize - 1);
3824 : :
3825 : 0 : start = lstart >> sb->s_blocksize_bits;
3826 : 0 : end = byte_end >> sb->s_blocksize_bits;
3827 : :
3828 : : /* Handle partial zero within the single block */
3829 [ # # # # ]: 0 : if (start == end &&
3830 [ # # ]: 0 : (partial_start || (partial_end != sb->s_blocksize - 1))) {
3831 : 0 : err = ext4_block_zero_page_range(handle, mapping,
3832 : : lstart, length);
3833 : 0 : return err;
3834 : : }
3835 : : /* Handle partial zero out on the start of the range */
3836 [ # # ]: 0 : if (partial_start) {
3837 : 0 : err = ext4_block_zero_page_range(handle, mapping,
3838 : : lstart, sb->s_blocksize);
3839 [ # # ]: 0 : if (err)
3840 : : return err;
3841 : : }
3842 : : /* Handle partial zero out on the end of the range */
3843 [ # # ]: 0 : if (partial_end != sb->s_blocksize - 1)
3844 : 0 : err = ext4_block_zero_page_range(handle, mapping,
3845 : : byte_end - partial_end,
3846 : 0 : partial_end + 1);
3847 : : return err;
3848 : : }
3849 : :
3850 : 210 : int ext4_can_truncate(struct inode *inode)
3851 : : {
3852 [ + + ]: 210 : if (S_ISREG(inode->i_mode))
3853 : : return 1;
3854 [ - + ]: 42 : if (S_ISDIR(inode->i_mode))
3855 : : return 1;
3856 [ # # ]: 0 : if (S_ISLNK(inode->i_mode))
3857 : 0 : return !ext4_inode_is_fast_symlink(inode);
3858 : : return 0;
3859 : : }
3860 : :
3861 : : /*
3862 : : * We have to make sure i_disksize gets properly updated before we truncate
3863 : : * page cache due to hole punching or zero range. Otherwise i_disksize update
3864 : : * can get lost as it may have been postponed to submission of writeback but
3865 : : * that will never happen after we truncate page cache.
3866 : : */
3867 : 0 : int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset,
3868 : : loff_t len)
3869 : : {
3870 : 0 : handle_t *handle;
3871 : 0 : loff_t size = i_size_read(inode);
3872 : :
3873 [ # # ]: 0 : WARN_ON(!inode_is_locked(inode));
3874 [ # # # # ]: 0 : if (offset > size || offset + len < size)
3875 : : return 0;
3876 : :
3877 [ # # ]: 0 : if (EXT4_I(inode)->i_disksize >= size)
3878 : : return 0;
3879 : :
3880 : 0 : handle = ext4_journal_start(inode, EXT4_HT_MISC, 1);
3881 [ # # ]: 0 : if (IS_ERR(handle))
3882 : 0 : return PTR_ERR(handle);
3883 : 0 : ext4_update_i_disksize(inode, size);
3884 : 0 : ext4_mark_inode_dirty(handle, inode);
3885 : 0 : ext4_journal_stop(handle);
3886 : :
3887 : 0 : return 0;
3888 : : }
3889 : :
3890 : : static void ext4_wait_dax_page(struct ext4_inode_info *ei)
3891 : : {
3892 : : up_write(&ei->i_mmap_sem);
3893 : : schedule();
3894 : : down_write(&ei->i_mmap_sem);
3895 : : }
3896 : :
3897 : 0 : int ext4_break_layouts(struct inode *inode)
3898 : : {
3899 : 0 : struct ext4_inode_info *ei = EXT4_I(inode);
3900 : 0 : struct page *page;
3901 : 0 : int error;
3902 : :
3903 [ # # # # ]: 0 : if (WARN_ON_ONCE(!rwsem_is_locked(&ei->i_mmap_sem)))
3904 : : return -EINVAL;
3905 : :
3906 : 0 : do {
3907 : 0 : page = dax_layout_busy_page(inode->i_mapping);
3908 : 0 : if (!page)
3909 : 0 : return 0;
3910 : :
3911 : : error = ___wait_var_event(&page->_refcount,
3912 : : atomic_read(&page->_refcount) == 1,
3913 : : TASK_INTERRUPTIBLE, 0, 0,
3914 : : ext4_wait_dax_page(ei));
3915 : : } while (error == 0);
3916 : :
3917 : : return error;
3918 : : }
3919 : :
3920 : : /*
3921 : : * ext4_punch_hole: punches a hole in a file by releasing the blocks
3922 : : * associated with the given offset and length
3923 : : *
3924 : : * @inode: File inode
3925 : : * @offset: The offset where the hole will begin
3926 : : * @len: The length of the hole
3927 : : *
3928 : : * Returns: 0 on success or negative on failure
3929 : : */
3930 : :
3931 : 0 : int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length)
3932 : : {
3933 : 0 : struct super_block *sb = inode->i_sb;
3934 : 0 : ext4_lblk_t first_block, stop_block;
3935 : 0 : struct address_space *mapping = inode->i_mapping;
3936 : 0 : loff_t first_block_offset, last_block_offset;
3937 : 0 : handle_t *handle;
3938 : 0 : unsigned int credits;
3939 : 0 : int ret = 0;
3940 : :
3941 : 0 : trace_ext4_punch_hole(inode, offset, length, 0);
3942 : :
3943 : 0 : ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA);
3944 : 0 : if (ext4_has_inline_data(inode)) {
3945 : 0 : down_write(&EXT4_I(inode)->i_mmap_sem);
3946 : 0 : ret = ext4_convert_inline_data(inode);
3947 : 0 : up_write(&EXT4_I(inode)->i_mmap_sem);
3948 [ # # ]: 0 : if (ret)
3949 : : return ret;
3950 : : }
3951 : :
3952 : : /*
3953 : : * Write out all dirty pages to avoid race conditions
3954 : : * Then release them.
3955 : : */
3956 [ # # ]: 0 : if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
3957 : 0 : ret = filemap_write_and_wait_range(mapping, offset,
3958 : 0 : offset + length - 1);
3959 [ # # ]: 0 : if (ret)
3960 : : return ret;
3961 : : }
3962 : :
3963 : 0 : inode_lock(inode);
3964 : :
3965 : : /* No need to punch hole beyond i_size */
3966 [ # # ]: 0 : if (offset >= inode->i_size)
3967 : 0 : goto out_mutex;
3968 : :
3969 : : /*
3970 : : * If the hole extends beyond i_size, set the hole
3971 : : * to end after the page that contains i_size
3972 : : */
3973 [ # # ]: 0 : if (offset + length > inode->i_size) {
3974 : 0 : length = inode->i_size +
3975 : 0 : PAGE_SIZE - (inode->i_size & (PAGE_SIZE - 1)) -
3976 : : offset;
3977 : : }
3978 : :
3979 [ # # ]: 0 : if (offset & (sb->s_blocksize - 1) ||
3980 [ # # ]: 0 : (offset + length) & (sb->s_blocksize - 1)) {
3981 : : /*
3982 : : * Attach jinode to inode for jbd2 if we do any zeroing of
3983 : : * partial block
3984 : : */
3985 : 0 : ret = ext4_inode_attach_jinode(inode);
3986 [ # # ]: 0 : if (ret < 0)
3987 : 0 : goto out_mutex;
3988 : :
3989 : : }
3990 : :
3991 : : /* Wait all existing dio workers, newcomers will block on i_mutex */
3992 : 0 : inode_dio_wait(inode);
3993 : :
3994 : : /*
3995 : : * Prevent page faults from reinstantiating pages we have released from
3996 : : * page cache.
3997 : : */
3998 : 0 : down_write(&EXT4_I(inode)->i_mmap_sem);
3999 : :
4000 : 0 : ret = ext4_break_layouts(inode);
4001 [ # # ]: 0 : if (ret)
4002 : 0 : goto out_dio;
4003 : :
4004 : 0 : first_block_offset = round_up(offset, sb->s_blocksize);
4005 : 0 : last_block_offset = round_down((offset + length), sb->s_blocksize) - 1;
4006 : :
4007 : : /* Now release the pages and zero block aligned part of pages*/
4008 [ # # ]: 0 : if (last_block_offset > first_block_offset) {
4009 : 0 : ret = ext4_update_disksize_before_punch(inode, offset, length);
4010 [ # # ]: 0 : if (ret)
4011 : 0 : goto out_dio;
4012 : 0 : truncate_pagecache_range(inode, first_block_offset,
4013 : : last_block_offset);
4014 : : }
4015 : :
4016 [ # # ]: 0 : if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4017 : 0 : credits = ext4_writepage_trans_blocks(inode);
4018 : : else
4019 : 0 : credits = ext4_blocks_for_truncate(inode);
4020 : 0 : handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4021 [ # # ]: 0 : if (IS_ERR(handle)) {
4022 [ # # ]: 0 : ret = PTR_ERR(handle);
4023 [ # # ]: 0 : ext4_std_error(sb, ret);
4024 : 0 : goto out_dio;
4025 : : }
4026 : :
4027 : 0 : ret = ext4_zero_partial_blocks(handle, inode, offset,
4028 : : length);
4029 [ # # ]: 0 : if (ret)
4030 : 0 : goto out_stop;
4031 : :
4032 : 0 : first_block = (offset + sb->s_blocksize - 1) >>
4033 : 0 : EXT4_BLOCK_SIZE_BITS(sb);
4034 : 0 : stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb);
4035 : :
4036 : : /* If there are blocks to remove, do it */
4037 [ # # ]: 0 : if (stop_block > first_block) {
4038 : :
4039 : 0 : down_write(&EXT4_I(inode)->i_data_sem);
4040 : 0 : ext4_discard_preallocations(inode);
4041 : :
4042 : 0 : ret = ext4_es_remove_extent(inode, first_block,
4043 : : stop_block - first_block);
4044 [ # # ]: 0 : if (ret) {
4045 : 0 : up_write(&EXT4_I(inode)->i_data_sem);
4046 : 0 : goto out_stop;
4047 : : }
4048 : :
4049 [ # # ]: 0 : if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4050 : 0 : ret = ext4_ext_remove_space(inode, first_block,
4051 : : stop_block - 1);
4052 : : else
4053 : 0 : ret = ext4_ind_remove_space(handle, inode, first_block,
4054 : : stop_block);
4055 : :
4056 : 0 : up_write(&EXT4_I(inode)->i_data_sem);
4057 : : }
4058 [ # # # # ]: 0 : if (IS_SYNC(inode))
4059 [ # # ]: 0 : ext4_handle_sync(handle);
4060 : :
4061 : 0 : inode->i_mtime = inode->i_ctime = current_time(inode);
4062 : 0 : ext4_mark_inode_dirty(handle, inode);
4063 [ # # ]: 0 : if (ret >= 0)
4064 [ # # ]: 0 : ext4_update_inode_fsync_trans(handle, inode, 1);
4065 : 0 : out_stop:
4066 : 0 : ext4_journal_stop(handle);
4067 : 0 : out_dio:
4068 : 0 : up_write(&EXT4_I(inode)->i_mmap_sem);
4069 : 0 : out_mutex:
4070 : 0 : inode_unlock(inode);
4071 : 0 : return ret;
4072 : : }
4073 : :
4074 : 9429 : int ext4_inode_attach_jinode(struct inode *inode)
4075 : : {
4076 : 9429 : struct ext4_inode_info *ei = EXT4_I(inode);
4077 : 9429 : struct jbd2_inode *jinode;
4078 : :
4079 [ + + + - ]: 9429 : if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal)
4080 : : return 0;
4081 : :
4082 : 9303 : jinode = jbd2_alloc_inode(GFP_KERNEL);
4083 : 9303 : spin_lock(&inode->i_lock);
4084 [ + - ]: 9303 : if (!ei->jinode) {
4085 [ - + ]: 9303 : if (!jinode) {
4086 : 0 : spin_unlock(&inode->i_lock);
4087 : 0 : return -ENOMEM;
4088 : : }
4089 : 9303 : ei->jinode = jinode;
4090 : 9303 : jbd2_journal_init_jbd_inode(ei->jinode, inode);
4091 : 9303 : jinode = NULL;
4092 : : }
4093 : 9303 : spin_unlock(&inode->i_lock);
4094 [ - + ]: 9303 : if (unlikely(jinode != NULL))
4095 : 0 : jbd2_free_inode(jinode);
4096 : : return 0;
4097 : : }
4098 : :
4099 : : /*
4100 : : * ext4_truncate()
4101 : : *
4102 : : * We block out ext4_get_block() block instantiations across the entire
4103 : : * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4104 : : * simultaneously on behalf of the same inode.
4105 : : *
4106 : : * As we work through the truncate and commit bits of it to the journal there
4107 : : * is one core, guiding principle: the file's tree must always be consistent on
4108 : : * disk. We must be able to restart the truncate after a crash.
4109 : : *
4110 : : * The file's tree may be transiently inconsistent in memory (although it
4111 : : * probably isn't), but whenever we close off and commit a journal transaction,
4112 : : * the contents of (the filesystem + the journal) must be consistent and
4113 : : * restartable. It's pretty simple, really: bottom up, right to left (although
4114 : : * left-to-right works OK too).
4115 : : *
4116 : : * Note that at recovery time, journal replay occurs *before* the restart of
4117 : : * truncate against the orphan inode list.
4118 : : *
4119 : : * The committed inode has the new, desired i_size (which is the same as
4120 : : * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4121 : : * that this inode's truncate did not complete and it will again call
4122 : : * ext4_truncate() to have another go. So there will be instantiated blocks
4123 : : * to the right of the truncation point in a crashed ext4 filesystem. But
4124 : : * that's fine - as long as they are linked from the inode, the post-crash
4125 : : * ext4_truncate() run will find them and release them.
4126 : : */
4127 : 210 : int ext4_truncate(struct inode *inode)
4128 : : {
4129 : 210 : struct ext4_inode_info *ei = EXT4_I(inode);
4130 : 210 : unsigned int credits;
4131 : 210 : int err = 0;
4132 : 210 : handle_t *handle;
4133 : 210 : struct address_space *mapping = inode->i_mapping;
4134 : :
4135 : : /*
4136 : : * There is a possibility that we're either freeing the inode
4137 : : * or it's a completely new inode. In those cases we might not
4138 : : * have i_mutex locked because it's not necessary.
4139 : : */
4140 [ - + ]: 210 : if (!(inode->i_state & (I_NEW|I_FREEING)))
4141 [ # # ]: 0 : WARN_ON(!inode_is_locked(inode));
4142 : 210 : trace_ext4_truncate_enter(inode);
4143 : :
4144 [ + - ]: 210 : if (!ext4_can_truncate(inode))
4145 : : return 0;
4146 : :
4147 : 210 : ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS);
4148 : :
4149 [ + - + - ]: 210 : if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC))
4150 : 210 : ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
4151 : :
4152 : 210 : if (ext4_has_inline_data(inode)) {
4153 : 0 : int has_inline = 1;
4154 : :
4155 : 0 : err = ext4_inline_data_truncate(inode, &has_inline);
4156 [ # # ]: 0 : if (err)
4157 : 0 : return err;
4158 [ # # ]: 0 : if (has_inline)
4159 : : return 0;
4160 : : }
4161 : :
4162 : : /* If we zero-out tail of the page, we have to create jinode for jbd2 */
4163 [ - + ]: 210 : if (inode->i_size & (inode->i_sb->s_blocksize - 1)) {
4164 [ # # ]: 0 : if (ext4_inode_attach_jinode(inode) < 0)
4165 : : return 0;
4166 : : }
4167 : :
4168 [ + - ]: 210 : if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4169 : 210 : credits = ext4_writepage_trans_blocks(inode);
4170 : : else
4171 : 0 : credits = ext4_blocks_for_truncate(inode);
4172 : :
4173 : 210 : handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits);
4174 [ - + ]: 210 : if (IS_ERR(handle))
4175 : 0 : return PTR_ERR(handle);
4176 : :
4177 [ - + ]: 210 : if (inode->i_size & (inode->i_sb->s_blocksize - 1))
4178 : 0 : ext4_block_truncate_page(handle, mapping, inode->i_size);
4179 : :
4180 : : /*
4181 : : * We add the inode to the orphan list, so that if this
4182 : : * truncate spans multiple transactions, and we crash, we will
4183 : : * resume the truncate when the filesystem recovers. It also
4184 : : * marks the inode dirty, to catch the new size.
4185 : : *
4186 : : * Implication: the file must always be in a sane, consistent
4187 : : * truncatable state while each transaction commits.
4188 : : */
4189 : 210 : err = ext4_orphan_add(handle, inode);
4190 [ - + ]: 210 : if (err)
4191 : 0 : goto out_stop;
4192 : :
4193 : 210 : down_write(&EXT4_I(inode)->i_data_sem);
4194 : :
4195 : 210 : ext4_discard_preallocations(inode);
4196 : :
4197 [ + - ]: 210 : if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4198 : 210 : err = ext4_ext_truncate(handle, inode);
4199 : : else
4200 : 0 : ext4_ind_truncate(handle, inode);
4201 : :
4202 : 210 : up_write(&ei->i_data_sem);
4203 [ - + ]: 210 : if (err)
4204 : 0 : goto out_stop;
4205 : :
4206 [ + - + - ]: 210 : if (IS_SYNC(inode))
4207 [ # # ]: 0 : ext4_handle_sync(handle);
4208 : :
4209 : 210 : out_stop:
4210 : : /*
4211 : : * If this was a simple ftruncate() and the file will remain alive,
4212 : : * then we need to clear up the orphan record which we created above.
4213 : : * However, if this was a real unlink then we were called by
4214 : : * ext4_evict_inode(), and we allow that function to clean up the
4215 : : * orphan info for us.
4216 : : */
4217 [ - + ]: 210 : if (inode->i_nlink)
4218 : 0 : ext4_orphan_del(handle, inode);
4219 : :
4220 : 210 : inode->i_mtime = inode->i_ctime = current_time(inode);
4221 : 210 : ext4_mark_inode_dirty(handle, inode);
4222 : 210 : ext4_journal_stop(handle);
4223 : :
4224 : 210 : trace_ext4_truncate_exit(inode);
4225 : 210 : return err;
4226 : : }
4227 : :
4228 : : /*
4229 : : * ext4_get_inode_loc returns with an extra refcount against the inode's
4230 : : * underlying buffer_head on success. If 'in_mem' is true, we have all
4231 : : * data in memory that is needed to recreate the on-disk version of this
4232 : : * inode.
4233 : : */
4234 : 122478 : static int __ext4_get_inode_loc(struct inode *inode,
4235 : : struct ext4_iloc *iloc, int in_mem)
4236 : : {
4237 : 122478 : struct ext4_group_desc *gdp;
4238 : 122478 : struct buffer_head *bh;
4239 : 122478 : struct super_block *sb = inode->i_sb;
4240 : 122478 : ext4_fsblk_t block;
4241 : 122478 : struct blk_plug plug;
4242 : 122478 : int inodes_per_block, inode_offset;
4243 : :
4244 : 122478 : iloc->bh = NULL;
4245 [ + - + - ]: 122478 : if (inode->i_ino < EXT4_ROOT_INO ||
4246 [ + - ]: 122478 : inode->i_ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
4247 : : return -EFSCORRUPTED;
4248 : :
4249 : 122478 : iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb);
4250 : 122478 : gdp = ext4_get_group_desc(sb, iloc->block_group, NULL);
4251 [ + - ]: 122478 : if (!gdp)
4252 : : return -EIO;
4253 : :
4254 : : /*
4255 : : * Figure out the offset within the block group inode table
4256 : : */
4257 : 122478 : inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4258 : 122478 : inode_offset = ((inode->i_ino - 1) %
4259 : 122478 : EXT4_INODES_PER_GROUP(sb));
4260 : 122478 : block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block);
4261 : 122478 : iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb);
4262 : :
4263 : 122478 : bh = sb_getblk(sb, block);
4264 [ + - ]: 122478 : if (unlikely(!bh))
4265 : : return -ENOMEM;
4266 : 122478 : if (ext4_simulate_fail(sb, EXT4_SIM_INODE_EIO))
4267 : : goto simulate_eio;
4268 [ + + ]: 122478 : if (!buffer_uptodate(bh)) {
4269 : 1261 : lock_buffer(bh);
4270 : :
4271 : : /*
4272 : : * If the buffer has the write error flag, we have failed
4273 : : * to write out another inode in the same block. In this
4274 : : * case, we don't have to read the block because we may
4275 : : * read the old inode data successfully.
4276 : : */
4277 [ - + - - ]: 1261 : if (buffer_write_io_error(bh) && !buffer_uptodate(bh))
4278 : 0 : set_buffer_uptodate(bh);
4279 : :
4280 [ + + ]: 1261 : if (buffer_uptodate(bh)) {
4281 : : /* someone brought it uptodate while we waited */
4282 : 1 : unlock_buffer(bh);
4283 : 1 : goto has_buffer;
4284 : : }
4285 : :
4286 : : /*
4287 : : * If we have all information of the inode in memory and this
4288 : : * is the only valid inode in the block, we need not read the
4289 : : * block.
4290 : : */
4291 [ + + ]: 1260 : if (in_mem) {
4292 : 903 : struct buffer_head *bitmap_bh;
4293 : 903 : int i, start;
4294 : :
4295 : 903 : start = inode_offset & ~(inodes_per_block - 1);
4296 : :
4297 : : /* Is the inode bitmap in cache? */
4298 : 903 : bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp));
4299 [ - + ]: 903 : if (unlikely(!bitmap_bh))
4300 : 0 : goto make_io;
4301 : :
4302 : : /*
4303 : : * If the inode bitmap isn't in cache then the
4304 : : * optimisation may end up performing two reads instead
4305 : : * of one, so skip it.
4306 : : */
4307 [ - + ]: 903 : if (!buffer_uptodate(bitmap_bh)) {
4308 [ # # ]: 0 : brelse(bitmap_bh);
4309 : 0 : goto make_io;
4310 : : }
4311 [ + + ]: 15351 : for (i = start; i < start + inodes_per_block; i++) {
4312 [ + + ]: 14448 : if (i == inode_offset)
4313 : 903 : continue;
4314 [ + - ]: 13545 : if (ext4_test_bit(i, bitmap_bh->b_data))
4315 : : break;
4316 : : }
4317 [ + - ]: 903 : brelse(bitmap_bh);
4318 [ - + ]: 903 : if (i == start + inodes_per_block) {
4319 : : /* all other inodes are free, so skip I/O */
4320 : 903 : memset(bh->b_data, 0, bh->b_size);
4321 : 903 : set_buffer_uptodate(bh);
4322 : 903 : unlock_buffer(bh);
4323 : 903 : goto has_buffer;
4324 : : }
4325 : : }
4326 : :
4327 : 357 : make_io:
4328 : : /*
4329 : : * If we need to do any I/O, try to pre-readahead extra
4330 : : * blocks from the inode table.
4331 : : */
4332 : 357 : blk_start_plug(&plug);
4333 [ + - ]: 357 : if (EXT4_SB(sb)->s_inode_readahead_blks) {
4334 : 357 : ext4_fsblk_t b, end, table;
4335 : 357 : unsigned num;
4336 : 357 : __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks;
4337 : :
4338 : 357 : table = ext4_inode_table(sb, gdp);
4339 : : /* s_inode_readahead_blks is always a power of 2 */
4340 : 357 : b = block & ~((ext4_fsblk_t) ra_blks - 1);
4341 : 357 : if (table > b)
4342 : : b = table;
4343 : 357 : end = b + ra_blks;
4344 : 357 : num = EXT4_INODES_PER_GROUP(sb);
4345 [ + - ]: 357 : if (ext4_has_group_desc_csum(sb))
4346 : 357 : num -= ext4_itable_unused_count(sb, gdp);
4347 : 357 : table += num / inodes_per_block;
4348 : 357 : if (end > table)
4349 : : end = table;
4350 [ + + ]: 10794 : while (b <= end)
4351 : 10437 : sb_breadahead(sb, b++);
4352 : : }
4353 : :
4354 : : /*
4355 : : * There are other valid inodes in the buffer, this inode
4356 : : * has in-inode xattrs, or we don't have this inode in memory.
4357 : : * Read the block from disk.
4358 : : */
4359 : 357 : trace_ext4_load_inode(inode);
4360 : 357 : get_bh(bh);
4361 : 357 : bh->b_end_io = end_buffer_read_sync;
4362 : 357 : submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh);
4363 : 357 : blk_finish_plug(&plug);
4364 : 357 : wait_on_buffer(bh);
4365 [ + - ]: 357 : if (!buffer_uptodate(bh)) {
4366 : 0 : simulate_eio:
4367 : 0 : ext4_set_errno(inode->i_sb, EIO);
4368 : 0 : EXT4_ERROR_INODE_BLOCK(inode, block,
4369 : : "unable to read itable block");
4370 [ # # ]: 0 : brelse(bh);
4371 : 0 : return -EIO;
4372 : : }
4373 : : }
4374 : 121574 : has_buffer:
4375 : 122478 : iloc->bh = bh;
4376 : 122478 : return 0;
4377 : : }
4378 : :
4379 : 110193 : int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc)
4380 : : {
4381 : : /* We have all inode data except xattrs in memory here. */
4382 : 110193 : return __ext4_get_inode_loc(inode, iloc,
4383 : : !ext4_test_inode_state(inode, EXT4_STATE_XATTR));
4384 : : }
4385 : :
4386 : 26817 : static bool ext4_should_use_dax(struct inode *inode)
4387 : : {
4388 : 26817 : if (!test_opt(inode->i_sb, DAX))
4389 : 26817 : return false;
4390 : : if (!S_ISREG(inode->i_mode))
4391 : : return false;
4392 : : if (ext4_should_journal_data(inode))
4393 : : return false;
4394 : : if (ext4_has_inline_data(inode))
4395 : : return false;
4396 : : if (ext4_test_inode_flag(inode, EXT4_INODE_ENCRYPT))
4397 : : return false;
4398 : : if (ext4_test_inode_flag(inode, EXT4_INODE_VERITY))
4399 : : return false;
4400 : : return true;
4401 : : }
4402 : :
4403 : 26817 : void ext4_set_inode_flags(struct inode *inode)
4404 : : {
4405 : 26817 : unsigned int flags = EXT4_I(inode)->i_flags;
4406 : 26817 : unsigned int new_fl = 0;
4407 : :
4408 [ - + ]: 26817 : if (flags & EXT4_SYNC_FL)
4409 : 0 : new_fl |= S_SYNC;
4410 [ - + ]: 26817 : if (flags & EXT4_APPEND_FL)
4411 : 0 : new_fl |= S_APPEND;
4412 [ - + ]: 26817 : if (flags & EXT4_IMMUTABLE_FL)
4413 : 0 : new_fl |= S_IMMUTABLE;
4414 [ - + ]: 26817 : if (flags & EXT4_NOATIME_FL)
4415 : 0 : new_fl |= S_NOATIME;
4416 [ - + ]: 26817 : if (flags & EXT4_DIRSYNC_FL)
4417 : 0 : new_fl |= S_DIRSYNC;
4418 [ - + ]: 26817 : if (ext4_should_use_dax(inode))
4419 : : new_fl |= S_DAX;
4420 [ - + ]: 26817 : if (flags & EXT4_ENCRYPT_FL)
4421 : 0 : new_fl |= S_ENCRYPTED;
4422 [ - + ]: 26817 : if (flags & EXT4_CASEFOLD_FL)
4423 : 0 : new_fl |= S_CASEFOLD;
4424 [ - + ]: 26817 : if (flags & EXT4_VERITY_FL)
4425 : 0 : new_fl |= S_VERITY;
4426 : 26817 : inode_set_flags(inode, new_fl,
4427 : : S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX|
4428 : : S_ENCRYPTED|S_CASEFOLD|S_VERITY);
4429 : 26817 : }
4430 : :
4431 : : static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode,
4432 : : struct ext4_inode_info *ei)
4433 : : {
4434 : : blkcnt_t i_blocks ;
4435 : : struct inode *inode = &(ei->vfs_inode);
4436 : : struct super_block *sb = inode->i_sb;
4437 : :
4438 : : if (ext4_has_feature_huge_file(sb)) {
4439 : : /* we are using combined 48 bit field */
4440 : : i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 |
4441 : : le32_to_cpu(raw_inode->i_blocks_lo);
4442 : : if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) {
4443 : : /* i_blocks represent file system block size */
4444 : : return i_blocks << (inode->i_blkbits - 9);
4445 : : } else {
4446 : : return i_blocks;
4447 : : }
4448 : : } else {
4449 : : return le32_to_cpu(raw_inode->i_blocks_lo);
4450 : : }
4451 : : }
4452 : :
4453 : : static inline int ext4_iget_extra_inode(struct inode *inode,
4454 : : struct ext4_inode *raw_inode,
4455 : : struct ext4_inode_info *ei)
4456 : : {
4457 : : __le32 *magic = (void *)raw_inode +
4458 : : EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize;
4459 : :
4460 : : if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize + sizeof(__le32) <=
4461 : : EXT4_INODE_SIZE(inode->i_sb) &&
4462 : : *magic == cpu_to_le32(EXT4_XATTR_MAGIC)) {
4463 : : ext4_set_inode_state(inode, EXT4_STATE_XATTR);
4464 : : return ext4_find_inline_data_nolock(inode);
4465 : : } else
4466 : : EXT4_I(inode)->i_inline_off = 0;
4467 : : return 0;
4468 : : }
4469 : :
4470 : 0 : int ext4_get_projid(struct inode *inode, kprojid_t *projid)
4471 : : {
4472 [ # # ]: 0 : if (!ext4_has_feature_project(inode->i_sb))
4473 : : return -EOPNOTSUPP;
4474 : 0 : *projid = EXT4_I(inode)->i_projid;
4475 : 0 : return 0;
4476 : : }
4477 : :
4478 : : /*
4479 : : * ext4 has self-managed i_version for ea inodes, it stores the lower 32bit of
4480 : : * refcount in i_version, so use raw values if inode has EXT4_EA_INODE_FL flag
4481 : : * set.
4482 : : */
4483 : 12285 : static inline void ext4_inode_set_iversion_queried(struct inode *inode, u64 val)
4484 : : {
4485 [ - + ]: 12285 : if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4486 : 0 : inode_set_iversion_raw(inode, val);
4487 : : else
4488 : 12285 : inode_set_iversion_queried(inode, val);
4489 : 12285 : }
4490 : 110151 : static inline u64 ext4_inode_peek_iversion(const struct inode *inode)
4491 : : {
4492 [ - + ]: 110151 : if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL))
4493 : 0 : return inode_peek_iversion_raw(inode);
4494 : : else
4495 : 110151 : return inode_peek_iversion(inode);
4496 : : }
4497 : :
4498 : 13733 : struct inode *__ext4_iget(struct super_block *sb, unsigned long ino,
4499 : : ext4_iget_flags flags, const char *function,
4500 : : unsigned int line)
4501 : : {
4502 : 13733 : struct ext4_iloc iloc;
4503 : 13733 : struct ext4_inode *raw_inode;
4504 : 13733 : struct ext4_inode_info *ei;
4505 : 13733 : struct inode *inode;
4506 [ + + ]: 13733 : journal_t *journal = EXT4_SB(sb)->s_journal;
4507 : 13733 : long ret;
4508 : 13733 : loff_t size;
4509 : 13733 : int block;
4510 : 13733 : uid_t i_uid;
4511 : 13733 : gid_t i_gid;
4512 : 13733 : projid_t i_projid;
4513 : :
4514 [ + + ]: 13733 : if ((!(flags & EXT4_IGET_SPECIAL) &&
4515 [ - + - - : 13733 : (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)) ||
+ - ]
4516 : 13733 : (ino < EXT4_ROOT_INO) ||
4517 [ - + ]: 13733 : (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))) {
4518 [ # # ]: 0 : if (flags & EXT4_IGET_HANDLE)
4519 : : return ERR_PTR(-ESTALE);
4520 : 0 : __ext4_error(sb, function, line,
4521 : : "inode #%lu: comm %s: iget: illegal inode #",
4522 : 0 : ino, current->comm);
4523 : 0 : return ERR_PTR(-EFSCORRUPTED);
4524 : : }
4525 : :
4526 : 13733 : inode = iget_locked(sb, ino);
4527 [ + - ]: 13733 : if (!inode)
4528 : : return ERR_PTR(-ENOMEM);
4529 [ + + ]: 13733 : if (!(inode->i_state & I_NEW))
4530 : : return inode;
4531 : :
4532 : 12285 : ei = EXT4_I(inode);
4533 : 12285 : iloc.bh = NULL;
4534 : :
4535 : 12285 : ret = __ext4_get_inode_loc(inode, &iloc, 0);
4536 [ - + ]: 12285 : if (ret < 0)
4537 : 0 : goto bad_inode;
4538 [ + + ]: 12285 : raw_inode = ext4_raw_inode(&iloc);
4539 : :
4540 [ + + - + ]: 12285 : if ((ino == EXT4_ROOT_INO) && (raw_inode->i_links_count == 0)) {
4541 : 0 : ext4_error_inode(inode, function, line, 0,
4542 : : "iget: root inode unallocated");
4543 : 0 : ret = -EFSCORRUPTED;
4544 : 0 : goto bad_inode;
4545 : : }
4546 : :
4547 [ - + ]: 12285 : if ((flags & EXT4_IGET_HANDLE) &&
4548 [ # # # # ]: 0 : (raw_inode->i_links_count == 0) && (raw_inode->i_mode == 0)) {
4549 : 0 : ret = -ESTALE;
4550 : 0 : goto bad_inode;
4551 : : }
4552 : :
4553 [ + - ]: 12285 : if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4554 : 12285 : ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize);
4555 : 12285 : if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize >
4556 [ + - - + ]: 12285 : EXT4_INODE_SIZE(inode->i_sb) ||
4557 : : (ei->i_extra_isize & 3)) {
4558 : 0 : ext4_error_inode(inode, function, line, 0,
4559 : : "iget: bad extra_isize %u "
4560 : : "(inode size %u)",
4561 : : ei->i_extra_isize,
4562 : : EXT4_INODE_SIZE(inode->i_sb));
4563 : 0 : ret = -EFSCORRUPTED;
4564 : 0 : goto bad_inode;
4565 : : }
4566 : : } else
4567 : 0 : ei->i_extra_isize = 0;
4568 : :
4569 : : /* Precompute checksum seed for inode metadata */
4570 [ + - ]: 12285 : if (ext4_has_metadata_csum(sb)) {
4571 : 12285 : struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
4572 : 12285 : __u32 csum;
4573 : 12285 : __le32 inum = cpu_to_le32(inode->i_ino);
4574 : 12285 : __le32 gen = raw_inode->i_generation;
4575 : 12285 : csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum,
4576 : : sizeof(inum));
4577 : 12285 : ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen,
4578 : : sizeof(gen));
4579 : : }
4580 : :
4581 [ - + ]: 12285 : if (!ext4_inode_csum_verify(inode, raw_inode, ei) ||
4582 : : ext4_simulate_fail(sb, EXT4_SIM_INODE_CRC)) {
4583 : 0 : ext4_set_errno(inode->i_sb, EFSBADCRC);
4584 : 0 : ext4_error_inode(inode, function, line, 0,
4585 : : "iget: checksum invalid");
4586 : 0 : ret = -EFSBADCRC;
4587 : 0 : goto bad_inode;
4588 : : }
4589 : :
4590 : 12285 : inode->i_mode = le16_to_cpu(raw_inode->i_mode);
4591 : 12285 : i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
4592 : 12285 : i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
4593 [ - + ]: 12285 : if (ext4_has_feature_project(sb) &&
4594 [ # # ]: 0 : EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE &&
4595 [ # # ]: 0 : EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
4596 : 0 : i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid);
4597 : : else
4598 : : i_projid = EXT4_DEF_PROJID;
4599 : :
4600 [ + - ]: 12285 : if (!(test_opt(inode->i_sb, NO_UID32))) {
4601 : 12285 : i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
4602 : 12285 : i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
4603 : : }
4604 : 12285 : i_uid_write(inode, i_uid);
4605 : 12285 : i_gid_write(inode, i_gid);
4606 : 12285 : ei->i_projid = make_kprojid(&init_user_ns, i_projid);
4607 : 12285 : set_nlink(inode, le16_to_cpu(raw_inode->i_links_count));
4608 : :
4609 [ - + ]: 12285 : ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */
4610 : 12285 : ei->i_inline_off = 0;
4611 : 12285 : ei->i_dir_start_lookup = 0;
4612 : 12285 : ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
4613 : : /* We now have enough fields to check if the inode was active or not.
4614 : : * This is needed because nfsd might try to access dead inodes
4615 : : * the test is that same one that e2fsck uses
4616 : : * NeilBrown 1999oct15
4617 : : */
4618 [ - + ]: 12285 : if (inode->i_nlink == 0) {
4619 [ # # # # ]: 0 : if ((inode->i_mode == 0 ||
4620 [ # # # # ]: 0 : !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) &&
4621 : : ino != EXT4_BOOT_LOADER_INO) {
4622 : : /* this inode is deleted */
4623 : 0 : ret = -ESTALE;
4624 : 0 : goto bad_inode;
4625 : : }
4626 : : /* The only unlinked inodes we let through here have
4627 : : * valid i_mode and are being read by the orphan
4628 : : * recovery code: that's fine, we're about to complete
4629 : : * the process of deleting those.
4630 : : * OR it is the EXT4_BOOT_LOADER_INO which is
4631 : : * not initialized on a new filesystem. */
4632 : : }
4633 : 12285 : ei->i_flags = le32_to_cpu(raw_inode->i_flags);
4634 : 12285 : ext4_set_inode_flags(inode);
4635 : 12285 : inode->i_blocks = ext4_inode_blocks(raw_inode, ei);
4636 : 12285 : ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo);
4637 [ + - ]: 12285 : if (ext4_has_feature_64bit(sb))
4638 : 12285 : ei->i_file_acl |=
4639 : 12285 : ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
4640 [ + - ]: 12285 : inode->i_size = ext4_isize(sb, raw_inode);
4641 [ - + ]: 12285 : if ((size = i_size_read(inode)) < 0) {
4642 : 0 : ext4_error_inode(inode, function, line, 0,
4643 : : "iget: bad i_size value: %lld", size);
4644 : 0 : ret = -EFSCORRUPTED;
4645 : 0 : goto bad_inode;
4646 : : }
4647 : : /*
4648 : : * If dir_index is not enabled but there's dir with INDEX flag set,
4649 : : * we'd normally treat htree data as empty space. But with metadata
4650 : : * checksumming that corrupts checksums so forbid that.
4651 : : */
4652 [ - + - - : 12285 : if (!ext4_has_feature_dir_index(sb) && ext4_has_metadata_csum(sb) &&
- - ]
4653 : : ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) {
4654 : 0 : ext4_error_inode(inode, function, line, 0,
4655 : : "iget: Dir with htree data on filesystem without dir_index feature.");
4656 : 0 : ret = -EFSCORRUPTED;
4657 : 0 : goto bad_inode;
4658 : : }
4659 : 12285 : ei->i_disksize = inode->i_size;
4660 : : #ifdef CONFIG_QUOTA
4661 : 12285 : ei->i_reserved_quota = 0;
4662 : : #endif
4663 : 12285 : inode->i_generation = le32_to_cpu(raw_inode->i_generation);
4664 : 12285 : ei->i_block_group = iloc.block_group;
4665 : 12285 : ei->i_last_alloc_group = ~0;
4666 : : /*
4667 : : * NOTE! The in-memory inode i_data array is in little-endian order
4668 : : * even on big-endian machines: we do NOT byteswap the block numbers!
4669 : : */
4670 [ + + ]: 196560 : for (block = 0; block < EXT4_N_BLOCKS; block++)
4671 : 184275 : ei->i_data[block] = raw_inode->i_block[block];
4672 [ + + ]: 12285 : INIT_LIST_HEAD(&ei->i_orphan);
4673 : :
4674 : : /*
4675 : : * Set transaction id's of transactions that have to be committed
4676 : : * to finish f[data]sync. We set them to currently running transaction
4677 : : * as we cannot be sure that the inode or some of its metadata isn't
4678 : : * part of the transaction - the inode could have been reclaimed and
4679 : : * now it is reread from disk.
4680 : : */
4681 [ + + ]: 12285 : if (journal) {
4682 : 12243 : transaction_t *transaction;
4683 : 12243 : tid_t tid;
4684 : :
4685 : 12243 : read_lock(&journal->j_state_lock);
4686 [ + + ]: 12243 : if (journal->j_running_transaction)
4687 : : transaction = journal->j_running_transaction;
4688 : : else
4689 : 7477 : transaction = journal->j_committing_transaction;
4690 [ + + ]: 12243 : if (transaction)
4691 : 4766 : tid = transaction->t_tid;
4692 : : else
4693 : 7477 : tid = journal->j_commit_sequence;
4694 : 12243 : read_unlock(&journal->j_state_lock);
4695 : 12243 : ei->i_sync_tid = tid;
4696 : 12243 : ei->i_datasync_tid = tid;
4697 : : }
4698 : :
4699 [ + - ]: 12285 : if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4700 [ - + ]: 12285 : if (ei->i_extra_isize == 0) {
4701 : : /* The extra space is currently unused. Use it. */
4702 : 0 : BUILD_BUG_ON(sizeof(struct ext4_inode) & 3);
4703 : 0 : ei->i_extra_isize = sizeof(struct ext4_inode) -
4704 : : EXT4_GOOD_OLD_INODE_SIZE;
4705 : : } else {
4706 : 12285 : ret = ext4_iget_extra_inode(inode, raw_inode, ei);
4707 [ - + ]: 12285 : if (ret)
4708 : 0 : goto bad_inode;
4709 : : }
4710 : : }
4711 : :
4712 [ + - - + ]: 12285 : EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode);
4713 [ + - - + ]: 12285 : EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode);
4714 [ + - - + ]: 12285 : EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode);
4715 [ + - + - : 12285 : EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode);
- + ]
4716 : :
4717 [ + - ]: 12285 : if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
4718 : 12285 : u64 ivers = le32_to_cpu(raw_inode->i_disk_version);
4719 : :
4720 [ + - ]: 12285 : if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) {
4721 [ + - ]: 12285 : if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
4722 : 12285 : ivers |=
4723 : 12285 : (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32;
4724 : : }
4725 : 12285 : ext4_inode_set_iversion_queried(inode, ivers);
4726 : : }
4727 : :
4728 : 12285 : ret = 0;
4729 [ - + - - ]: 12285 : if (ei->i_file_acl &&
4730 : 0 : !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) {
4731 : 0 : ext4_error_inode(inode, function, line, 0,
4732 : : "iget: bad extended attribute block %llu",
4733 : : ei->i_file_acl);
4734 : 0 : ret = -EFSCORRUPTED;
4735 : 0 : goto bad_inode;
4736 : 12285 : } else if (!ext4_has_inline_data(inode)) {
4737 : : /* validate the block references in the inode */
4738 [ + + + - ]: 12285 : if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
4739 [ - + ]: 2667 : (S_ISLNK(inode->i_mode) &&
4740 : 2667 : !ext4_inode_is_fast_symlink(inode))) {
4741 [ + - ]: 9618 : if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))
4742 : 9618 : ret = ext4_ext_check_inode(inode);
4743 : : else
4744 : 0 : ret = ext4_ind_check_inode(inode);
4745 : : }
4746 : : }
4747 [ - + ]: 9618 : if (ret)
4748 : 0 : goto bad_inode;
4749 : :
4750 [ + + ]: 12285 : if (S_ISREG(inode->i_mode)) {
4751 : 7350 : inode->i_op = &ext4_file_inode_operations;
4752 : 7350 : inode->i_fop = &ext4_file_operations;
4753 : 7350 : ext4_set_aops(inode);
4754 [ + + ]: 4935 : } else if (S_ISDIR(inode->i_mode)) {
4755 : 2268 : inode->i_op = &ext4_dir_inode_operations;
4756 : 2268 : inode->i_fop = &ext4_dir_operations;
4757 [ + - ]: 2667 : } else if (S_ISLNK(inode->i_mode)) {
4758 : : /* VFS does not allow setting these so must be corruption */
4759 [ - + ]: 2667 : if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) {
4760 : 0 : ext4_error_inode(inode, function, line, 0,
4761 : : "iget: immutable or append flags "
4762 : : "not allowed on symlinks");
4763 : 0 : ret = -EFSCORRUPTED;
4764 : 0 : goto bad_inode;
4765 : : }
4766 [ - + ]: 2667 : if (IS_ENCRYPTED(inode)) {
4767 : 0 : inode->i_op = &ext4_encrypted_symlink_inode_operations;
4768 : 0 : ext4_set_aops(inode);
4769 [ + - ]: 2667 : } else if (ext4_inode_is_fast_symlink(inode)) {
4770 : 2667 : inode->i_link = (char *)ei->i_data;
4771 : 2667 : inode->i_op = &ext4_fast_symlink_inode_operations;
4772 : 2667 : nd_terminate_link(ei->i_data, inode->i_size,
4773 : : sizeof(ei->i_data) - 1);
4774 : : } else {
4775 : 0 : inode->i_op = &ext4_symlink_inode_operations;
4776 : 0 : ext4_set_aops(inode);
4777 : : }
4778 : 2667 : inode_nohighmem(inode);
4779 [ # # # # ]: 0 : } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
4780 [ # # ]: 0 : S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
4781 : 0 : inode->i_op = &ext4_special_inode_operations;
4782 [ # # ]: 0 : if (raw_inode->i_block[0])
4783 : 0 : init_special_inode(inode, inode->i_mode,
4784 : : old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
4785 : : else
4786 : 0 : init_special_inode(inode, inode->i_mode,
4787 : 0 : new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
4788 [ # # ]: 0 : } else if (ino == EXT4_BOOT_LOADER_INO) {
4789 : 0 : make_bad_inode(inode);
4790 : : } else {
4791 : 0 : ret = -EFSCORRUPTED;
4792 : 0 : ext4_error_inode(inode, function, line, 0,
4793 : : "iget: bogus i_mode (%o)", inode->i_mode);
4794 : 0 : goto bad_inode;
4795 : : }
4796 [ - + - - ]: 12285 : if (IS_CASEFOLDED(inode) && !ext4_has_feature_casefold(inode->i_sb))
4797 : 0 : ext4_error_inode(inode, function, line, 0,
4798 : : "casefold flag without casefold feature");
4799 [ + - ]: 12285 : brelse(iloc.bh);
4800 : :
4801 : 12285 : unlock_new_inode(inode);
4802 : 12285 : return inode;
4803 : :
4804 : 0 : bad_inode:
4805 [ # # ]: 0 : brelse(iloc.bh);
4806 : 0 : iget_failed(inode);
4807 : 0 : return ERR_PTR(ret);
4808 : : }
4809 : :
4810 : : static int ext4_inode_blocks_set(handle_t *handle,
4811 : : struct ext4_inode *raw_inode,
4812 : : struct ext4_inode_info *ei)
4813 : : {
4814 : : struct inode *inode = &(ei->vfs_inode);
4815 : : u64 i_blocks = inode->i_blocks;
4816 : : struct super_block *sb = inode->i_sb;
4817 : :
4818 : : if (i_blocks <= ~0U) {
4819 : : /*
4820 : : * i_blocks can be represented in a 32 bit variable
4821 : : * as multiple of 512 bytes
4822 : : */
4823 : : raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4824 : : raw_inode->i_blocks_high = 0;
4825 : : ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4826 : : return 0;
4827 : : }
4828 : : if (!ext4_has_feature_huge_file(sb))
4829 : : return -EFBIG;
4830 : :
4831 : : if (i_blocks <= 0xffffffffffffULL) {
4832 : : /*
4833 : : * i_blocks can be represented in a 48 bit variable
4834 : : * as multiple of 512 bytes
4835 : : */
4836 : : raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4837 : : raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4838 : : ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4839 : : } else {
4840 : : ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE);
4841 : : /* i_block is stored in file system block size */
4842 : : i_blocks = i_blocks >> (inode->i_blkbits - 9);
4843 : : raw_inode->i_blocks_lo = cpu_to_le32(i_blocks);
4844 : : raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32);
4845 : : }
4846 : : return 0;
4847 : : }
4848 : :
4849 : : struct other_inode {
4850 : : unsigned long orig_ino;
4851 : : struct ext4_inode *raw_inode;
4852 : : };
4853 : :
4854 : 0 : static int other_inode_match(struct inode * inode, unsigned long ino,
4855 : : void *data)
4856 : : {
4857 : 0 : struct other_inode *oi = (struct other_inode *) data;
4858 : :
4859 [ # # ]: 0 : if ((inode->i_ino != ino) ||
4860 : : (inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
4861 [ # # ]: 0 : I_DIRTY_INODE)) ||
4862 : : ((inode->i_state & I_DIRTY_TIME) == 0))
4863 : : return 0;
4864 : 0 : spin_lock(&inode->i_lock);
4865 : 0 : if (((inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW |
4866 [ # # ]: 0 : I_DIRTY_INODE)) == 0) &&
4867 : : (inode->i_state & I_DIRTY_TIME)) {
4868 : 0 : struct ext4_inode_info *ei = EXT4_I(inode);
4869 : :
4870 : 0 : inode->i_state &= ~(I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED);
4871 : 0 : spin_unlock(&inode->i_lock);
4872 : :
4873 : 0 : spin_lock(&ei->i_raw_lock);
4874 [ # # ]: 0 : EXT4_INODE_SET_XTIME(i_ctime, inode, oi->raw_inode);
4875 [ # # ]: 0 : EXT4_INODE_SET_XTIME(i_mtime, inode, oi->raw_inode);
4876 [ # # ]: 0 : EXT4_INODE_SET_XTIME(i_atime, inode, oi->raw_inode);
4877 : 0 : ext4_inode_csum_set(inode, oi->raw_inode, ei);
4878 : 0 : spin_unlock(&ei->i_raw_lock);
4879 : 0 : trace_ext4_other_inode_update_time(inode, oi->orig_ino);
4880 : 0 : return -1;
4881 : : }
4882 : 0 : spin_unlock(&inode->i_lock);
4883 : 0 : return -1;
4884 : : }
4885 : :
4886 : : /*
4887 : : * Opportunistically update the other time fields for other inodes in
4888 : : * the same inode table block.
4889 : : */
4890 : 0 : static void ext4_update_other_inodes_time(struct super_block *sb,
4891 : : unsigned long orig_ino, char *buf)
4892 : : {
4893 : 0 : struct other_inode oi;
4894 : 0 : unsigned long ino;
4895 : 0 : int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block;
4896 : 0 : int inode_size = EXT4_INODE_SIZE(sb);
4897 : :
4898 : 0 : oi.orig_ino = orig_ino;
4899 : : /*
4900 : : * Calculate the first inode in the inode table block. Inode
4901 : : * numbers are one-based. That is, the first inode in a block
4902 : : * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
4903 : : */
4904 : 0 : ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1;
4905 [ # # ]: 0 : for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) {
4906 [ # # ]: 0 : if (ino == orig_ino)
4907 : 0 : continue;
4908 : 0 : oi.raw_inode = (struct ext4_inode *) buf;
4909 : 0 : (void) find_inode_nowait(sb, ino, other_inode_match, &oi);
4910 : : }
4911 : 0 : }
4912 : :
4913 : : /*
4914 : : * Post the struct inode info into an on-disk inode location in the
4915 : : * buffer-cache. This gobbles the caller's reference to the
4916 : : * buffer_head in the inode location struct.
4917 : : *
4918 : : * The caller must have write access to iloc->bh.
4919 : : */
4920 : 110151 : static int ext4_do_update_inode(handle_t *handle,
4921 : : struct inode *inode,
4922 : : struct ext4_iloc *iloc)
4923 : : {
4924 : 110151 : struct ext4_inode *raw_inode = ext4_raw_inode(iloc);
4925 : 110151 : struct ext4_inode_info *ei = EXT4_I(inode);
4926 : 110151 : struct buffer_head *bh = iloc->bh;
4927 : 110151 : struct super_block *sb = inode->i_sb;
4928 : 110151 : int err = 0, rc, block;
4929 : 110151 : int need_datasync = 0, set_large_file = 0;
4930 : 110151 : uid_t i_uid;
4931 : 110151 : gid_t i_gid;
4932 : 110151 : projid_t i_projid;
4933 : :
4934 : 110151 : spin_lock(&ei->i_raw_lock);
4935 : :
4936 : : /* For fields not tracked in the in-memory inode,
4937 : : * initialise them to zero for new inodes. */
4938 [ + + ]: 110151 : if (ext4_test_inode_state(inode, EXT4_STATE_NEW))
4939 : 14532 : memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size);
4940 : :
4941 : 110151 : raw_inode->i_mode = cpu_to_le16(inode->i_mode);
4942 [ + - ]: 110151 : i_uid = i_uid_read(inode);
4943 [ + - ]: 110151 : i_gid = i_gid_read(inode);
4944 [ + - ]: 110151 : i_projid = from_kprojid(&init_user_ns, ei->i_projid);
4945 [ + - ]: 110151 : if (!(test_opt(inode->i_sb, NO_UID32))) {
4946 : 110151 : raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid));
4947 : 110151 : raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid));
4948 : : /*
4949 : : * Fix up interoperability with old kernels. Otherwise, old inodes get
4950 : : * re-used with the upper 16 bits of the uid/gid intact
4951 : : */
4952 [ + + + - ]: 110151 : if (ei->i_dtime && list_empty(&ei->i_orphan)) {
4953 : 210 : raw_inode->i_uid_high = 0;
4954 : 210 : raw_inode->i_gid_high = 0;
4955 : : } else {
4956 : 109941 : raw_inode->i_uid_high =
4957 : 109941 : cpu_to_le16(high_16_bits(i_uid));
4958 : 109941 : raw_inode->i_gid_high =
4959 : 109941 : cpu_to_le16(high_16_bits(i_gid));
4960 : : }
4961 : : } else {
4962 [ # # ]: 0 : raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid));
4963 [ # # ]: 0 : raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid));
4964 : 0 : raw_inode->i_uid_high = 0;
4965 : 0 : raw_inode->i_gid_high = 0;
4966 : : }
4967 : 110151 : raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
4968 : :
4969 [ + - ]: 110151 : EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode);
4970 [ + - ]: 110151 : EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode);
4971 [ + - ]: 110151 : EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode);
4972 [ + - + - ]: 110151 : EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode);
4973 : :
4974 : 110151 : err = ext4_inode_blocks_set(handle, raw_inode, ei);
4975 [ - + ]: 110151 : if (err) {
4976 : 0 : spin_unlock(&ei->i_raw_lock);
4977 : 0 : goto out_brelse;
4978 : : }
4979 : 110151 : raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
4980 : 110151 : raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF);
4981 [ + - ]: 110151 : if (likely(!test_opt2(inode->i_sb, HURD_COMPAT)))
4982 : 110151 : raw_inode->i_file_acl_high =
4983 : 110151 : cpu_to_le16(ei->i_file_acl >> 32);
4984 : 110151 : raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl);
4985 [ + - + + ]: 220302 : if (ei->i_disksize != ext4_isize(inode->i_sb, raw_inode)) {
4986 : 5649 : ext4_isize_set(raw_inode, ei->i_disksize);
4987 : 5649 : need_datasync = 1;
4988 : : }
4989 [ - + ]: 110151 : if (ei->i_disksize > 0x7fffffffULL) {
4990 [ # # ]: 0 : if (!ext4_has_feature_large_file(sb) ||
4991 [ # # ]: 0 : EXT4_SB(sb)->s_es->s_rev_level ==
4992 : : cpu_to_le32(EXT4_GOOD_OLD_REV))
4993 : : set_large_file = 1;
4994 : : }
4995 : 110151 : raw_inode->i_generation = cpu_to_le32(inode->i_generation);
4996 [ - + ]: 110151 : if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
4997 [ # # # # ]: 0 : if (old_valid_dev(inode->i_rdev)) {
4998 : 0 : raw_inode->i_block[0] =
4999 : 0 : cpu_to_le32(old_encode_dev(inode->i_rdev));
5000 : 0 : raw_inode->i_block[1] = 0;
5001 : : } else {
5002 : 0 : raw_inode->i_block[0] = 0;
5003 : 0 : raw_inode->i_block[1] =
5004 : 0 : cpu_to_le32(new_encode_dev(inode->i_rdev));
5005 : 0 : raw_inode->i_block[2] = 0;
5006 : : }
5007 : 110151 : } else if (!ext4_has_inline_data(inode)) {
5008 [ + + ]: 1762416 : for (block = 0; block < EXT4_N_BLOCKS; block++)
5009 : 1652265 : raw_inode->i_block[block] = ei->i_data[block];
5010 : : }
5011 : :
5012 [ + - ]: 110151 : if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) {
5013 : 110151 : u64 ivers = ext4_inode_peek_iversion(inode);
5014 : :
5015 : 110151 : raw_inode->i_disk_version = cpu_to_le32(ivers);
5016 [ + - ]: 110151 : if (ei->i_extra_isize) {
5017 [ + - ]: 110151 : if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi))
5018 : 110151 : raw_inode->i_version_hi =
5019 : 110151 : cpu_to_le32(ivers >> 32);
5020 : 110151 : raw_inode->i_extra_isize =
5021 : 110151 : cpu_to_le16(ei->i_extra_isize);
5022 : : }
5023 : : }
5024 : :
5025 [ + - - + ]: 110151 : BUG_ON(!ext4_has_feature_project(inode->i_sb) &&
5026 : : i_projid != EXT4_DEF_PROJID);
5027 : :
5028 [ + - ]: 110151 : if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE &&
5029 [ + - ]: 110151 : EXT4_FITS_IN_INODE(raw_inode, ei, i_projid))
5030 : 110151 : raw_inode->i_projid = cpu_to_le32(i_projid);
5031 : :
5032 : 110151 : ext4_inode_csum_set(inode, raw_inode, ei);
5033 : 110151 : spin_unlock(&ei->i_raw_lock);
5034 [ - + ]: 110151 : if (inode->i_sb->s_flags & SB_LAZYTIME)
5035 : 0 : ext4_update_other_inodes_time(inode->i_sb, inode->i_ino,
5036 : : bh->b_data);
5037 : :
5038 : 110151 : BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
5039 : 110151 : rc = ext4_handle_dirty_metadata(handle, NULL, bh);
5040 : 110151 : if (!err)
5041 : 110151 : err = rc;
5042 : 110151 : ext4_clear_inode_state(inode, EXT4_STATE_NEW);
5043 [ - + ]: 110151 : if (set_large_file) {
5044 : 0 : BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access");
5045 : 0 : err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
5046 [ # # ]: 0 : if (err)
5047 : 0 : goto out_brelse;
5048 : 0 : ext4_set_feature_large_file(sb);
5049 [ # # ]: 0 : ext4_handle_sync(handle);
5050 : 0 : err = ext4_handle_dirty_super(handle, sb);
5051 : : }
5052 : 110151 : ext4_update_inode_fsync_trans(handle, inode, need_datasync);
5053 : 110151 : out_brelse:
5054 [ + - ]: 110151 : brelse(bh);
5055 [ - + ]: 110151 : ext4_std_error(inode->i_sb, err);
5056 : 110151 : return err;
5057 : : }
5058 : :
5059 : : /*
5060 : : * ext4_write_inode()
5061 : : *
5062 : : * We are called from a few places:
5063 : : *
5064 : : * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
5065 : : * Here, there will be no transaction running. We wait for any running
5066 : : * transaction to commit.
5067 : : *
5068 : : * - Within flush work (sys_sync(), kupdate and such).
5069 : : * We wait on commit, if told to.
5070 : : *
5071 : : * - Within iput_final() -> write_inode_now()
5072 : : * We wait on commit, if told to.
5073 : : *
5074 : : * In all cases it is actually safe for us to return without doing anything,
5075 : : * because the inode has been copied into a raw inode buffer in
5076 : : * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
5077 : : * writeback.
5078 : : *
5079 : : * Note that we are absolutely dependent upon all inode dirtiers doing the
5080 : : * right thing: they *must* call mark_inode_dirty() after dirtying info in
5081 : : * which we are interested.
5082 : : *
5083 : : * It would be a bug for them to not do this. The code:
5084 : : *
5085 : : * mark_inode_dirty(inode)
5086 : : * stuff();
5087 : : * inode->i_size = expr;
5088 : : *
5089 : : * is in error because write_inode() could occur while `stuff()' is running,
5090 : : * and the new i_size will be lost. Plus the inode will no longer be on the
5091 : : * superblock's dirty inode list.
5092 : : */
5093 : 0 : int ext4_write_inode(struct inode *inode, struct writeback_control *wbc)
5094 : : {
5095 : 0 : int err;
5096 : :
5097 [ # # # # : 0 : if (WARN_ON_ONCE(current->flags & PF_MEMALLOC) ||
# # ]
5098 [ # # ]: 0 : sb_rdonly(inode->i_sb))
5099 : : return 0;
5100 : :
5101 [ # # ]: 0 : if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5102 : : return -EIO;
5103 : :
5104 [ # # ]: 0 : if (EXT4_SB(inode->i_sb)->s_journal) {
5105 [ # # ]: 0 : if (ext4_journal_current_handle()) {
5106 : 0 : jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5107 : 0 : dump_stack();
5108 : 0 : return -EIO;
5109 : : }
5110 : :
5111 : : /*
5112 : : * No need to force transaction in WB_SYNC_NONE mode. Also
5113 : : * ext4_sync_fs() will force the commit after everything is
5114 : : * written.
5115 : : */
5116 [ # # ]: 0 : if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync)
5117 : : return 0;
5118 : :
5119 : 0 : err = jbd2_complete_transaction(EXT4_SB(inode->i_sb)->s_journal,
5120 : 0 : EXT4_I(inode)->i_sync_tid);
5121 : : } else {
5122 : 0 : struct ext4_iloc iloc;
5123 : :
5124 : 0 : err = __ext4_get_inode_loc(inode, &iloc, 0);
5125 [ # # ]: 0 : if (err)
5126 : 0 : return err;
5127 : : /*
5128 : : * sync(2) will flush the whole buffer cache. No need to do
5129 : : * it here separately for each inode.
5130 : : */
5131 [ # # ]: 0 : if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)
5132 : 0 : sync_dirty_buffer(iloc.bh);
5133 [ # # # # ]: 0 : if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) {
5134 : 0 : ext4_set_errno(inode->i_sb, EIO);
5135 : 0 : EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr,
5136 : : "IO error syncing inode");
5137 : 0 : err = -EIO;
5138 : : }
5139 [ # # ]: 0 : brelse(iloc.bh);
5140 : : }
5141 : : return err;
5142 : : }
5143 : :
5144 : : /*
5145 : : * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
5146 : : * buffers that are attached to a page stradding i_size and are undergoing
5147 : : * commit. In that case we have to wait for commit to finish and try again.
5148 : : */
5149 : 0 : static void ext4_wait_for_tail_page_commit(struct inode *inode)
5150 : : {
5151 : 0 : struct page *page;
5152 : 0 : unsigned offset;
5153 [ # # ]: 0 : journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
5154 : 0 : tid_t commit_tid = 0;
5155 : 0 : int ret;
5156 : :
5157 : 0 : offset = inode->i_size & (PAGE_SIZE - 1);
5158 : : /*
5159 : : * If the page is fully truncated, we don't need to wait for any commit
5160 : : * (and we even should not as __ext4_journalled_invalidatepage() may
5161 : : * strip all buffers from the page but keep the page dirty which can then
5162 : : * confuse e.g. concurrent ext4_writepage() seeing dirty page without
5163 : : * buffers). Also we don't need to wait for any commit if all buffers in
5164 : : * the page remain valid. This is most beneficial for the common case of
5165 : : * blocksize == PAGESIZE.
5166 : : */
5167 [ # # # # ]: 0 : if (!offset || offset > (PAGE_SIZE - i_blocksize(inode)))
5168 : : return;
5169 : 0 : while (1) {
5170 : 0 : page = find_lock_page(inode->i_mapping,
5171 : 0 : inode->i_size >> PAGE_SHIFT);
5172 [ # # ]: 0 : if (!page)
5173 : : return;
5174 : 0 : ret = __ext4_journalled_invalidatepage(page, offset,
5175 : : PAGE_SIZE - offset);
5176 : 0 : unlock_page(page);
5177 : 0 : put_page(page);
5178 [ # # ]: 0 : if (ret != -EBUSY)
5179 : : return;
5180 : 0 : commit_tid = 0;
5181 : 0 : read_lock(&journal->j_state_lock);
5182 [ # # ]: 0 : if (journal->j_committing_transaction)
5183 : 0 : commit_tid = journal->j_committing_transaction->t_tid;
5184 : 0 : read_unlock(&journal->j_state_lock);
5185 [ # # ]: 0 : if (commit_tid)
5186 : 0 : jbd2_log_wait_commit(journal, commit_tid);
5187 : : }
5188 : : }
5189 : :
5190 : : /*
5191 : : * ext4_setattr()
5192 : : *
5193 : : * Called from notify_change.
5194 : : *
5195 : : * We want to trap VFS attempts to truncate the file as soon as
5196 : : * possible. In particular, we want to make sure that when the VFS
5197 : : * shrinks i_size, we put the inode on the orphan list and modify
5198 : : * i_disksize immediately, so that during the subsequent flushing of
5199 : : * dirty pages and freeing of disk blocks, we can guarantee that any
5200 : : * commit will leave the blocks being flushed in an unused state on
5201 : : * disk. (On recovery, the inode will get truncated and the blocks will
5202 : : * be freed, so we have a strong guarantee that no future commit will
5203 : : * leave these blocks visible to the user.)
5204 : : *
5205 : : * Another thing we have to assure is that if we are in ordered mode
5206 : : * and inode is still attached to the committing transaction, we must
5207 : : * we start writeout of all the dirty pages which are being truncated.
5208 : : * This way we are sure that all the data written in the previous
5209 : : * transaction are already on disk (truncate waits for pages under
5210 : : * writeback).
5211 : : *
5212 : : * Called with inode->i_mutex down.
5213 : : */
5214 : 777 : int ext4_setattr(struct dentry *dentry, struct iattr *attr)
5215 : : {
5216 : 777 : struct inode *inode = d_inode(dentry);
5217 : 777 : int error, rc = 0;
5218 : 777 : int orphan = 0;
5219 : 777 : const unsigned int ia_valid = attr->ia_valid;
5220 : :
5221 [ + - ]: 777 : if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5222 : : return -EIO;
5223 : :
5224 [ + - ]: 777 : if (unlikely(IS_IMMUTABLE(inode)))
5225 : : return -EPERM;
5226 : :
5227 [ - + - - ]: 777 : if (unlikely(IS_APPEND(inode) &&
5228 : : (ia_valid & (ATTR_MODE | ATTR_UID |
5229 : : ATTR_GID | ATTR_TIMES_SET))))
5230 : : return -EPERM;
5231 : :
5232 : 777 : error = setattr_prepare(dentry, attr);
5233 [ + - ]: 777 : if (error)
5234 : : return error;
5235 : :
5236 [ - + ]: 777 : error = fscrypt_prepare_setattr(dentry, attr);
5237 : : if (error)
5238 : : return error;
5239 : :
5240 [ + - ]: 777 : error = fsverity_prepare_setattr(dentry, attr);
5241 : 777 : if (error)
5242 : : return error;
5243 : :
5244 [ + - + + ]: 1554 : if (is_quota_modification(inode, attr)) {
5245 : 231 : error = dquot_initialize(inode);
5246 [ + - ]: 231 : if (error)
5247 : : return error;
5248 : : }
5249 [ + + + + ]: 777 : if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
5250 [ + + + + ]: 756 : (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
5251 : 231 : handle_t *handle;
5252 : :
5253 : : /* (user+group)*(old+new) structure, inode write (sb,
5254 : : * inode block, ? - but truncate inode update has it) */
5255 [ + - - + : 231 : handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
- - + - -
+ ]
5256 : : (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) +
5257 : : EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3);
5258 [ - + ]: 231 : if (IS_ERR(handle)) {
5259 : 0 : error = PTR_ERR(handle);
5260 : 0 : goto err_out;
5261 : : }
5262 : :
5263 : : /* dquot_transfer() calls back ext4_get_inode_usage() which
5264 : : * counts xattr inode references.
5265 : : */
5266 : 231 : down_read(&EXT4_I(inode)->xattr_sem);
5267 : 231 : error = dquot_transfer(inode, attr);
5268 : 231 : up_read(&EXT4_I(inode)->xattr_sem);
5269 : :
5270 [ - + ]: 231 : if (error) {
5271 : 0 : ext4_journal_stop(handle);
5272 : 0 : return error;
5273 : : }
5274 : : /* Update corresponding info in inode so that everything is in
5275 : : * one transaction */
5276 [ + - ]: 231 : if (attr->ia_valid & ATTR_UID)
5277 : 231 : inode->i_uid = attr->ia_uid;
5278 [ + - ]: 231 : if (attr->ia_valid & ATTR_GID)
5279 : 231 : inode->i_gid = attr->ia_gid;
5280 : 231 : error = ext4_mark_inode_dirty(handle, inode);
5281 : 231 : ext4_journal_stop(handle);
5282 : : }
5283 : :
5284 [ - + ]: 777 : if (attr->ia_valid & ATTR_SIZE) {
5285 : 0 : handle_t *handle;
5286 : 0 : loff_t oldsize = inode->i_size;
5287 : 0 : int shrink = (attr->ia_size < inode->i_size);
5288 : :
5289 [ # # ]: 0 : if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
5290 [ # # ]: 0 : struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5291 : :
5292 [ # # ]: 0 : if (attr->ia_size > sbi->s_bitmap_maxbytes)
5293 : : return -EFBIG;
5294 : : }
5295 [ # # ]: 0 : if (!S_ISREG(inode->i_mode))
5296 : : return -EINVAL;
5297 : :
5298 [ # # # # ]: 0 : if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
5299 : 0 : inode_inc_iversion(inode);
5300 : :
5301 [ # # ]: 0 : if (shrink) {
5302 [ # # ]: 0 : if (ext4_should_order_data(inode)) {
5303 : 0 : error = ext4_begin_ordered_truncate(inode,
5304 : : attr->ia_size);
5305 [ # # ]: 0 : if (error)
5306 : 0 : goto err_out;
5307 : : }
5308 : : /*
5309 : : * Blocks are going to be removed from the inode. Wait
5310 : : * for dio in flight.
5311 : : */
5312 : 0 : inode_dio_wait(inode);
5313 : : }
5314 : :
5315 : 0 : down_write(&EXT4_I(inode)->i_mmap_sem);
5316 : :
5317 : 0 : rc = ext4_break_layouts(inode);
5318 [ # # ]: 0 : if (rc) {
5319 : 0 : up_write(&EXT4_I(inode)->i_mmap_sem);
5320 : 0 : return rc;
5321 : : }
5322 : :
5323 [ # # ]: 0 : if (attr->ia_size != inode->i_size) {
5324 : 0 : handle = ext4_journal_start(inode, EXT4_HT_INODE, 3);
5325 [ # # ]: 0 : if (IS_ERR(handle)) {
5326 : 0 : error = PTR_ERR(handle);
5327 : 0 : goto out_mmap_sem;
5328 : : }
5329 [ # # # # ]: 0 : if (ext4_handle_valid(handle) && shrink) {
5330 : 0 : error = ext4_orphan_add(handle, inode);
5331 : 0 : orphan = 1;
5332 : : }
5333 : : /*
5334 : : * Update c/mtime on truncate up, ext4_truncate() will
5335 : : * update c/mtime in shrink case below
5336 : : */
5337 [ # # ]: 0 : if (!shrink) {
5338 : 0 : inode->i_mtime = current_time(inode);
5339 : 0 : inode->i_ctime = inode->i_mtime;
5340 : : }
5341 : 0 : down_write(&EXT4_I(inode)->i_data_sem);
5342 : 0 : EXT4_I(inode)->i_disksize = attr->ia_size;
5343 : 0 : rc = ext4_mark_inode_dirty(handle, inode);
5344 [ # # ]: 0 : if (!error)
5345 : 0 : error = rc;
5346 : : /*
5347 : : * We have to update i_size under i_data_sem together
5348 : : * with i_disksize to avoid races with writeback code
5349 : : * running ext4_wb_update_i_disksize().
5350 : : */
5351 [ # # ]: 0 : if (!error)
5352 : 0 : i_size_write(inode, attr->ia_size);
5353 : 0 : up_write(&EXT4_I(inode)->i_data_sem);
5354 : 0 : ext4_journal_stop(handle);
5355 [ # # ]: 0 : if (error)
5356 : 0 : goto out_mmap_sem;
5357 [ # # ]: 0 : if (!shrink) {
5358 : 0 : pagecache_isize_extended(inode, oldsize,
5359 : : inode->i_size);
5360 [ # # ]: 0 : } else if (ext4_should_journal_data(inode)) {
5361 : 0 : ext4_wait_for_tail_page_commit(inode);
5362 : : }
5363 : : }
5364 : :
5365 : : /*
5366 : : * Truncate pagecache after we've waited for commit
5367 : : * in data=journal mode to make pages freeable.
5368 : : */
5369 : 0 : truncate_pagecache(inode, inode->i_size);
5370 : : /*
5371 : : * Call ext4_truncate() even if i_size didn't change to
5372 : : * truncate possible preallocated blocks.
5373 : : */
5374 [ # # ]: 0 : if (attr->ia_size <= oldsize) {
5375 : 0 : rc = ext4_truncate(inode);
5376 [ # # ]: 0 : if (rc)
5377 : 0 : error = rc;
5378 : : }
5379 : 0 : out_mmap_sem:
5380 : 0 : up_write(&EXT4_I(inode)->i_mmap_sem);
5381 : : }
5382 : :
5383 [ + - ]: 777 : if (!error) {
5384 : 777 : setattr_copy(inode, attr);
5385 : 777 : mark_inode_dirty(inode);
5386 : : }
5387 : :
5388 : : /*
5389 : : * If the call to ext4_truncate failed to get a transaction handle at
5390 : : * all, we need to clean up the in-core orphan list manually.
5391 : : */
5392 [ - + - - ]: 777 : if (orphan && inode->i_nlink)
5393 : 0 : ext4_orphan_del(NULL, inode);
5394 : :
5395 [ - + + + ]: 777 : if (!error && (ia_valid & ATTR_MODE))
5396 : 252 : rc = posix_acl_chmod(inode, inode->i_mode);
5397 : :
5398 : 525 : err_out:
5399 [ - + ]: 777 : ext4_std_error(inode->i_sb, error);
5400 [ + - ]: 777 : if (!error)
5401 : 777 : error = rc;
5402 : : return error;
5403 : : }
5404 : :
5405 : 485331 : int ext4_getattr(const struct path *path, struct kstat *stat,
5406 : : u32 request_mask, unsigned int query_flags)
5407 : : {
5408 [ - + ]: 485331 : struct inode *inode = d_inode(path->dentry);
5409 : 485331 : struct ext4_inode *raw_inode;
5410 : 485331 : struct ext4_inode_info *ei = EXT4_I(inode);
5411 : 485331 : unsigned int flags;
5412 : :
5413 [ - + ]: 485331 : if ((request_mask & STATX_BTIME) &&
5414 [ # # ]: 0 : EXT4_FITS_IN_INODE(raw_inode, ei, i_crtime)) {
5415 : 0 : stat->result_mask |= STATX_BTIME;
5416 : 0 : stat->btime.tv_sec = ei->i_crtime.tv_sec;
5417 : 0 : stat->btime.tv_nsec = ei->i_crtime.tv_nsec;
5418 : : }
5419 : :
5420 : 485331 : flags = ei->i_flags & EXT4_FL_USER_VISIBLE;
5421 [ - + ]: 485331 : if (flags & EXT4_APPEND_FL)
5422 : 0 : stat->attributes |= STATX_ATTR_APPEND;
5423 [ - + ]: 485331 : if (flags & EXT4_COMPR_FL)
5424 : 0 : stat->attributes |= STATX_ATTR_COMPRESSED;
5425 [ - + ]: 485331 : if (flags & EXT4_ENCRYPT_FL)
5426 : 0 : stat->attributes |= STATX_ATTR_ENCRYPTED;
5427 [ - + ]: 485331 : if (flags & EXT4_IMMUTABLE_FL)
5428 : 0 : stat->attributes |= STATX_ATTR_IMMUTABLE;
5429 [ - + ]: 485331 : if (flags & EXT4_NODUMP_FL)
5430 : 0 : stat->attributes |= STATX_ATTR_NODUMP;
5431 [ - + ]: 485331 : if (flags & EXT4_VERITY_FL)
5432 : 0 : stat->attributes |= STATX_ATTR_VERITY;
5433 : :
5434 : 485331 : stat->attributes_mask |= (STATX_ATTR_APPEND |
5435 : : STATX_ATTR_COMPRESSED |
5436 : : STATX_ATTR_ENCRYPTED |
5437 : : STATX_ATTR_IMMUTABLE |
5438 : : STATX_ATTR_NODUMP |
5439 : : STATX_ATTR_VERITY);
5440 : :
5441 : 485331 : generic_fillattr(inode, stat);
5442 : 485331 : return 0;
5443 : : }
5444 : :
5445 : 331254 : int ext4_file_getattr(const struct path *path, struct kstat *stat,
5446 : : u32 request_mask, unsigned int query_flags)
5447 : : {
5448 : 331254 : struct inode *inode = d_inode(path->dentry);
5449 : 331254 : u64 delalloc_blocks;
5450 : :
5451 : 331254 : ext4_getattr(path, stat, request_mask, query_flags);
5452 : :
5453 : : /*
5454 : : * If there is inline data in the inode, the inode will normally not
5455 : : * have data blocks allocated (it may have an external xattr block).
5456 : : * Report at least one sector for such files, so tools like tar, rsync,
5457 : : * others don't incorrectly think the file is completely sparse.
5458 : : */
5459 [ - + ]: 662508 : if (unlikely(ext4_has_inline_data(inode)))
5460 : 0 : stat->blocks += (stat->size + 511) >> 9;
5461 : :
5462 : : /*
5463 : : * We can't update i_blocks if the block allocation is delayed
5464 : : * otherwise in the case of system crash before the real block
5465 : : * allocation is done, we will have i_blocks inconsistent with
5466 : : * on-disk file blocks.
5467 : : * We always keep i_blocks updated together with real
5468 : : * allocation. But to not confuse with user, stat
5469 : : * will return the blocks that include the delayed allocation
5470 : : * blocks for this file.
5471 : : */
5472 : 331254 : delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb),
5473 : : EXT4_I(inode)->i_reserved_data_blocks);
5474 : 331254 : stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9);
5475 : 331254 : return 0;
5476 : : }
5477 : :
5478 : 714 : static int ext4_index_trans_blocks(struct inode *inode, int lblocks,
5479 : : int pextents)
5480 : : {
5481 [ - + ]: 714 : if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
5482 : 0 : return ext4_ind_trans_blocks(inode, lblocks);
5483 : 714 : return ext4_ext_index_trans_blocks(inode, pextents);
5484 : : }
5485 : :
5486 : : /*
5487 : : * Account for index blocks, block groups bitmaps and block group
5488 : : * descriptor blocks if modify datablocks and index blocks
5489 : : * worse case, the indexs blocks spread over different block groups
5490 : : *
5491 : : * If datablocks are discontiguous, they are possible to spread over
5492 : : * different block groups too. If they are contiguous, with flexbg,
5493 : : * they could still across block group boundary.
5494 : : *
5495 : : * Also account for superblock, inode, quota and xattr blocks
5496 : : */
5497 : 714 : static int ext4_meta_trans_blocks(struct inode *inode, int lblocks,
5498 : : int pextents)
5499 : : {
5500 : 714 : ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb);
5501 : 714 : int gdpblocks;
5502 : 714 : int idxblocks;
5503 : 714 : int ret = 0;
5504 : :
5505 : : /*
5506 : : * How many index blocks need to touch to map @lblocks logical blocks
5507 : : * to @pextents physical extents?
5508 : : */
5509 : 714 : idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents);
5510 : :
5511 : 714 : ret = idxblocks;
5512 : :
5513 : : /*
5514 : : * Now let's see how many group bitmaps and group descriptors need
5515 : : * to account
5516 : : */
5517 : 714 : groups = idxblocks + pextents;
5518 : 714 : gdpblocks = groups;
5519 : 714 : if (groups > ngroups)
5520 : : groups = ngroups;
5521 [ - + ]: 714 : if (groups > EXT4_SB(inode->i_sb)->s_gdb_count)
5522 : 0 : gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count;
5523 : :
5524 : : /* bitmaps and block group descriptor blocks */
5525 : 714 : ret += groups + gdpblocks;
5526 : :
5527 : : /* Blocks for super block, inode, quota and xattr blocks */
5528 [ + - - + ]: 714 : ret += EXT4_META_TRANS_BLOCKS(inode->i_sb);
5529 : :
5530 : 714 : return ret;
5531 : : }
5532 : :
5533 : : /*
5534 : : * Calculate the total number of credits to reserve to fit
5535 : : * the modification of a single pages into a single transaction,
5536 : : * which may include multiple chunks of block allocations.
5537 : : *
5538 : : * This could be called via ext4_write_begin()
5539 : : *
5540 : : * We need to consider the worse case, when
5541 : : * one new block per extent.
5542 : : */
5543 : 210 : int ext4_writepage_trans_blocks(struct inode *inode)
5544 : : {
5545 [ + - ]: 210 : int bpp = ext4_journal_blocks_per_page(inode);
5546 : 210 : int ret;
5547 : :
5548 : 210 : ret = ext4_meta_trans_blocks(inode, bpp, bpp);
5549 : :
5550 : : /* Account for data blocks for journalled mode */
5551 [ + + ]: 210 : if (ext4_should_journal_data(inode))
5552 : 42 : ret += bpp;
5553 : 210 : return ret;
5554 : : }
5555 : :
5556 : : /*
5557 : : * Calculate the journal credits for a chunk of data modification.
5558 : : *
5559 : : * This is called from DIO, fallocate or whoever calling
5560 : : * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5561 : : *
5562 : : * journal buffers for data blocks are not included here, as DIO
5563 : : * and fallocate do no need to journal data buffers.
5564 : : */
5565 : 168 : int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks)
5566 : : {
5567 : 168 : return ext4_meta_trans_blocks(inode, nrblocks, 1);
5568 : : }
5569 : :
5570 : : /*
5571 : : * The caller must have previously called ext4_reserve_inode_write().
5572 : : * Give this, we know that the caller already has write access to iloc->bh.
5573 : : */
5574 : 110151 : int ext4_mark_iloc_dirty(handle_t *handle,
5575 : : struct inode *inode, struct ext4_iloc *iloc)
5576 : : {
5577 : 110151 : int err = 0;
5578 : :
5579 [ - + ]: 110151 : if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) {
5580 : 0 : put_bh(iloc->bh);
5581 : 0 : return -EIO;
5582 : : }
5583 [ - + ]: 110151 : if (IS_I_VERSION(inode))
5584 : 0 : inode_inc_iversion(inode);
5585 : :
5586 : : /* the do_update_inode consumes one bh->b_count */
5587 : 110151 : get_bh(iloc->bh);
5588 : :
5589 : : /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5590 : 110151 : err = ext4_do_update_inode(handle, inode, iloc);
5591 : 110151 : put_bh(iloc->bh);
5592 : 110151 : return err;
5593 : : }
5594 : :
5595 : : /*
5596 : : * On success, We end up with an outstanding reference count against
5597 : : * iloc->bh. This _must_ be cleaned up later.
5598 : : */
5599 : :
5600 : : int
5601 : 110193 : ext4_reserve_inode_write(handle_t *handle, struct inode *inode,
5602 : : struct ext4_iloc *iloc)
5603 : : {
5604 : 110193 : int err;
5605 : :
5606 [ + - ]: 110193 : if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
5607 : : return -EIO;
5608 : :
5609 : 110193 : err = ext4_get_inode_loc(inode, iloc);
5610 [ + - ]: 110193 : if (!err) {
5611 : 110193 : BUFFER_TRACE(iloc->bh, "get_write_access");
5612 : 110193 : err = ext4_journal_get_write_access(handle, iloc->bh);
5613 [ - + ]: 110193 : if (err) {
5614 [ # # ]: 0 : brelse(iloc->bh);
5615 : 0 : iloc->bh = NULL;
5616 : : }
5617 : : }
5618 [ - + ]: 110193 : ext4_std_error(inode->i_sb, err);
5619 : : return err;
5620 : : }
5621 : :
5622 : 0 : static int __ext4_expand_extra_isize(struct inode *inode,
5623 : : unsigned int new_extra_isize,
5624 : : struct ext4_iloc *iloc,
5625 : : handle_t *handle, int *no_expand)
5626 : : {
5627 : 0 : struct ext4_inode *raw_inode;
5628 : 0 : struct ext4_xattr_ibody_header *header;
5629 [ # # ]: 0 : unsigned int inode_size = EXT4_INODE_SIZE(inode->i_sb);
5630 : 0 : struct ext4_inode_info *ei = EXT4_I(inode);
5631 : 0 : int error;
5632 : :
5633 : : /* this was checked at iget time, but double check for good measure */
5634 [ # # # # ]: 0 : if ((EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > inode_size) ||
5635 : : (ei->i_extra_isize & 3)) {
5636 : 0 : EXT4_ERROR_INODE(inode, "bad extra_isize %u (inode size %u)",
5637 : : ei->i_extra_isize,
5638 : : EXT4_INODE_SIZE(inode->i_sb));
5639 : 0 : return -EFSCORRUPTED;
5640 : : }
5641 [ # # # # ]: 0 : if ((new_extra_isize < ei->i_extra_isize) ||
5642 : 0 : (new_extra_isize < 4) ||
5643 [ # # ]: 0 : (new_extra_isize > inode_size - EXT4_GOOD_OLD_INODE_SIZE))
5644 : : return -EINVAL; /* Should never happen */
5645 : :
5646 : 0 : raw_inode = ext4_raw_inode(iloc);
5647 : :
5648 : 0 : header = IHDR(inode, raw_inode);
5649 : :
5650 : : /* No extended attributes present */
5651 [ # # ]: 0 : if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) ||
5652 [ # # ]: 0 : header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) {
5653 : 0 : memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE +
5654 : 0 : EXT4_I(inode)->i_extra_isize, 0,
5655 : 0 : new_extra_isize - EXT4_I(inode)->i_extra_isize);
5656 : 0 : EXT4_I(inode)->i_extra_isize = new_extra_isize;
5657 : 0 : return 0;
5658 : : }
5659 : :
5660 : : /* try to expand with EAs present */
5661 : 0 : error = ext4_expand_extra_isize_ea(inode, new_extra_isize,
5662 : : raw_inode, handle);
5663 [ # # ]: 0 : if (error) {
5664 : : /*
5665 : : * Inode size expansion failed; don't try again
5666 : : */
5667 : 0 : *no_expand = 1;
5668 : : }
5669 : :
5670 : : return error;
5671 : : }
5672 : :
5673 : : /*
5674 : : * Expand an inode by new_extra_isize bytes.
5675 : : * Returns 0 on success or negative error number on failure.
5676 : : */
5677 : 0 : static int ext4_try_to_expand_extra_isize(struct inode *inode,
5678 : : unsigned int new_extra_isize,
5679 : : struct ext4_iloc iloc,
5680 : : handle_t *handle)
5681 : : {
5682 : 0 : int no_expand;
5683 : 0 : int error;
5684 : :
5685 [ # # ]: 0 : if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND))
5686 : : return -EOVERFLOW;
5687 : :
5688 : : /*
5689 : : * In nojournal mode, we can immediately attempt to expand
5690 : : * the inode. When journaled, we first need to obtain extra
5691 : : * buffer credits since we may write into the EA block
5692 : : * with this same handle. If journal_extend fails, then it will
5693 : : * only result in a minor loss of functionality for that inode.
5694 : : * If this is felt to be critical, then e2fsck should be run to
5695 : : * force a large enough s_min_extra_isize.
5696 : : */
5697 [ # # ]: 0 : if (ext4_journal_extend(handle,
5698 [ # # # # : 0 : EXT4_DATA_TRANS_BLOCKS(inode->i_sb), 0) != 0)
# # # # ]
5699 : : return -ENOSPC;
5700 : :
5701 [ # # ]: 0 : if (ext4_write_trylock_xattr(inode, &no_expand) == 0)
5702 : : return -EBUSY;
5703 : :
5704 : 0 : error = __ext4_expand_extra_isize(inode, new_extra_isize, &iloc,
5705 : : handle, &no_expand);
5706 [ # # ]: 0 : ext4_write_unlock_xattr(inode, &no_expand);
5707 : :
5708 : 0 : return error;
5709 : : }
5710 : :
5711 : 0 : int ext4_expand_extra_isize(struct inode *inode,
5712 : : unsigned int new_extra_isize,
5713 : : struct ext4_iloc *iloc)
5714 : : {
5715 : 0 : handle_t *handle;
5716 : 0 : int no_expand;
5717 : 0 : int error, rc;
5718 : :
5719 [ # # ]: 0 : if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) {
5720 [ # # ]: 0 : brelse(iloc->bh);
5721 : 0 : return -EOVERFLOW;
5722 : : }
5723 : :
5724 [ # # # # : 0 : handle = ext4_journal_start(inode, EXT4_HT_INODE,
# # ]
5725 : : EXT4_DATA_TRANS_BLOCKS(inode->i_sb));
5726 [ # # ]: 0 : if (IS_ERR(handle)) {
5727 [ # # ]: 0 : error = PTR_ERR(handle);
5728 [ # # ]: 0 : brelse(iloc->bh);
5729 : 0 : return error;
5730 : : }
5731 : :
5732 : 0 : ext4_write_lock_xattr(inode, &no_expand);
5733 : :
5734 : 0 : BUFFER_TRACE(iloc->bh, "get_write_access");
5735 : 0 : error = ext4_journal_get_write_access(handle, iloc->bh);
5736 [ # # ]: 0 : if (error) {
5737 [ # # ]: 0 : brelse(iloc->bh);
5738 : 0 : goto out_unlock;
5739 : : }
5740 : :
5741 : 0 : error = __ext4_expand_extra_isize(inode, new_extra_isize, iloc,
5742 : : handle, &no_expand);
5743 : :
5744 : 0 : rc = ext4_mark_iloc_dirty(handle, inode, iloc);
5745 [ # # ]: 0 : if (!error)
5746 : 0 : error = rc;
5747 : :
5748 : 0 : out_unlock:
5749 [ # # ]: 0 : ext4_write_unlock_xattr(inode, &no_expand);
5750 : 0 : ext4_journal_stop(handle);
5751 : 0 : return error;
5752 : : }
5753 : :
5754 : : /*
5755 : : * What we do here is to mark the in-core inode as clean with respect to inode
5756 : : * dirtiness (it may still be data-dirty).
5757 : : * This means that the in-core inode may be reaped by prune_icache
5758 : : * without having to perform any I/O. This is a very good thing,
5759 : : * because *any* task may call prune_icache - even ones which
5760 : : * have a transaction open against a different journal.
5761 : : *
5762 : : * Is this cheating? Not really. Sure, we haven't written the
5763 : : * inode out, but prune_icache isn't a user-visible syncing function.
5764 : : * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5765 : : * we start and wait on commits.
5766 : : */
5767 : 109731 : int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode)
5768 : : {
5769 : 109731 : struct ext4_iloc iloc;
5770 : 109731 : struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5771 : 109731 : int err;
5772 : :
5773 : 109731 : might_sleep();
5774 : 109731 : trace_ext4_mark_inode_dirty(inode, _RET_IP_);
5775 : 109731 : err = ext4_reserve_inode_write(handle, inode, &iloc);
5776 [ + - ]: 109731 : if (err)
5777 : : return err;
5778 : :
5779 [ - + ]: 109731 : if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize)
5780 : 0 : ext4_try_to_expand_extra_isize(inode, sbi->s_want_extra_isize,
5781 : : iloc, handle);
5782 : :
5783 : 109731 : return ext4_mark_iloc_dirty(handle, inode, &iloc);
5784 : : }
5785 : :
5786 : : /*
5787 : : * ext4_dirty_inode() is called from __mark_inode_dirty()
5788 : : *
5789 : : * We're really interested in the case where a file is being extended.
5790 : : * i_size has been changed by generic_commit_write() and we thus need
5791 : : * to include the updated inode in the current transaction.
5792 : : *
5793 : : * Also, dquot_alloc_block() will always dirty the inode when blocks
5794 : : * are allocated to the file.
5795 : : *
5796 : : * If the inode is marked synchronous, we don't honour that here - doing
5797 : : * so would cause a commit on atime updates, which we don't bother doing.
5798 : : * We handle synchronous inodes at the highest possible level.
5799 : : *
5800 : : * If only the I_DIRTY_TIME flag is set, we can skip everything. If
5801 : : * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need
5802 : : * to copy into the on-disk inode structure are the timestamp files.
5803 : : */
5804 : 38184 : void ext4_dirty_inode(struct inode *inode, int flags)
5805 : : {
5806 : 38184 : handle_t *handle;
5807 : :
5808 [ + - ]: 38184 : if (flags == I_DIRTY_TIME)
5809 : : return;
5810 : 38184 : handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
5811 [ - + ]: 38184 : if (IS_ERR(handle))
5812 : 0 : goto out;
5813 : :
5814 : 38184 : ext4_mark_inode_dirty(handle, inode);
5815 : :
5816 : 38184 : ext4_journal_stop(handle);
5817 : 38184 : out:
5818 : : return;
5819 : : }
5820 : :
5821 : 0 : int ext4_change_inode_journal_flag(struct inode *inode, int val)
5822 : : {
5823 : 0 : journal_t *journal;
5824 : 0 : handle_t *handle;
5825 : 0 : int err;
5826 [ # # ]: 0 : struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
5827 : :
5828 : : /*
5829 : : * We have to be very careful here: changing a data block's
5830 : : * journaling status dynamically is dangerous. If we write a
5831 : : * data block to the journal, change the status and then delete
5832 : : * that block, we risk forgetting to revoke the old log record
5833 : : * from the journal and so a subsequent replay can corrupt data.
5834 : : * So, first we make sure that the journal is empty and that
5835 : : * nobody is changing anything.
5836 : : */
5837 : :
5838 : 0 : journal = EXT4_JOURNAL(inode);
5839 [ # # ]: 0 : if (!journal)
5840 : : return 0;
5841 [ # # ]: 0 : if (is_journal_aborted(journal))
5842 : : return -EROFS;
5843 : :
5844 : : /* Wait for all existing dio workers */
5845 : 0 : inode_dio_wait(inode);
5846 : :
5847 : : /*
5848 : : * Before flushing the journal and switching inode's aops, we have
5849 : : * to flush all dirty data the inode has. There can be outstanding
5850 : : * delayed allocations, there can be unwritten extents created by
5851 : : * fallocate or buffered writes in dioread_nolock mode covered by
5852 : : * dirty data which can be converted only after flushing the dirty
5853 : : * data (and journalled aops don't know how to handle these cases).
5854 : : */
5855 [ # # ]: 0 : if (val) {
5856 : 0 : down_write(&EXT4_I(inode)->i_mmap_sem);
5857 : 0 : err = filemap_write_and_wait(inode->i_mapping);
5858 [ # # ]: 0 : if (err < 0) {
5859 : 0 : up_write(&EXT4_I(inode)->i_mmap_sem);
5860 : 0 : return err;
5861 : : }
5862 : : }
5863 : :
5864 : 0 : percpu_down_write(&sbi->s_writepages_rwsem);
5865 : 0 : jbd2_journal_lock_updates(journal);
5866 : :
5867 : : /*
5868 : : * OK, there are no updates running now, and all cached data is
5869 : : * synced to disk. We are now in a completely consistent state
5870 : : * which doesn't have anything in the journal, and we know that
5871 : : * no filesystem updates are running, so it is safe to modify
5872 : : * the inode's in-core data-journaling state flag now.
5873 : : */
5874 : :
5875 [ # # ]: 0 : if (val)
5876 : 0 : ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5877 : : else {
5878 : 0 : err = jbd2_journal_flush(journal);
5879 [ # # ]: 0 : if (err < 0) {
5880 : 0 : jbd2_journal_unlock_updates(journal);
5881 : 0 : percpu_up_write(&sbi->s_writepages_rwsem);
5882 : 0 : return err;
5883 : : }
5884 : 0 : ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);
5885 : : }
5886 : 0 : ext4_set_aops(inode);
5887 : :
5888 : 0 : jbd2_journal_unlock_updates(journal);
5889 : 0 : percpu_up_write(&sbi->s_writepages_rwsem);
5890 : :
5891 [ # # ]: 0 : if (val)
5892 : 0 : up_write(&EXT4_I(inode)->i_mmap_sem);
5893 : :
5894 : : /* Finally we can mark the inode as dirty. */
5895 : :
5896 : 0 : handle = ext4_journal_start(inode, EXT4_HT_INODE, 1);
5897 [ # # ]: 0 : if (IS_ERR(handle))
5898 : 0 : return PTR_ERR(handle);
5899 : :
5900 : 0 : err = ext4_mark_inode_dirty(handle, inode);
5901 [ # # ]: 0 : ext4_handle_sync(handle);
5902 : 0 : ext4_journal_stop(handle);
5903 [ # # ]: 0 : ext4_std_error(inode->i_sb, err);
5904 : :
5905 : : return err;
5906 : : }
5907 : :
5908 : 0 : static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh)
5909 : : {
5910 : 0 : return !buffer_mapped(bh);
5911 : : }
5912 : :
5913 : 0 : vm_fault_t ext4_page_mkwrite(struct vm_fault *vmf)
5914 : : {
5915 : 0 : struct vm_area_struct *vma = vmf->vma;
5916 : 0 : struct page *page = vmf->page;
5917 : 0 : loff_t size;
5918 : 0 : unsigned long len;
5919 : 0 : int err;
5920 : 0 : vm_fault_t ret;
5921 : 0 : struct file *file = vma->vm_file;
5922 [ # # ]: 0 : struct inode *inode = file_inode(file);
5923 : 0 : struct address_space *mapping = inode->i_mapping;
5924 : 0 : handle_t *handle;
5925 : 0 : get_block_t *get_block;
5926 : 0 : int retries = 0;
5927 : :
5928 [ # # ]: 0 : if (unlikely(IS_IMMUTABLE(inode)))
5929 : : return VM_FAULT_SIGBUS;
5930 : :
5931 : 0 : sb_start_pagefault(inode->i_sb);
5932 : 0 : file_update_time(vma->vm_file);
5933 : :
5934 : 0 : down_read(&EXT4_I(inode)->i_mmap_sem);
5935 : :
5936 : 0 : err = ext4_convert_inline_data(inode);
5937 [ # # ]: 0 : if (err)
5938 : 0 : goto out_ret;
5939 : :
5940 : : /* Delalloc case is easy... */
5941 [ # # # # ]: 0 : if (test_opt(inode->i_sb, DELALLOC) &&
5942 [ # # ]: 0 : !ext4_should_journal_data(inode) &&
5943 : 0 : !ext4_nonda_switch(inode->i_sb)) {
5944 : 0 : do {
5945 : 0 : err = block_page_mkwrite(vma, vmf,
5946 : : ext4_da_get_block_prep);
5947 [ # # ]: 0 : } while (err == -ENOSPC &&
5948 [ # # ]: 0 : ext4_should_retry_alloc(inode->i_sb, &retries));
5949 : 0 : goto out_ret;
5950 : : }
5951 : :
5952 : 0 : lock_page(page);
5953 [ # # ]: 0 : size = i_size_read(inode);
5954 : : /* Page got truncated from under us? */
5955 [ # # # # ]: 0 : if (page->mapping != mapping || page_offset(page) > size) {
5956 : 0 : unlock_page(page);
5957 : 0 : ret = VM_FAULT_NOPAGE;
5958 : 0 : goto out;
5959 : : }
5960 : :
5961 [ # # ]: 0 : if (page->index == size >> PAGE_SHIFT)
5962 : 0 : len = size & ~PAGE_MASK;
5963 : : else
5964 : : len = PAGE_SIZE;
5965 : : /*
5966 : : * Return if we have all the buffers mapped. This avoids the need to do
5967 : : * journal_start/journal_stop which can block and take a long time
5968 : : */
5969 [ # # ]: 0 : if (page_has_buffers(page)) {
5970 [ # # # # ]: 0 : if (!ext4_walk_page_buffers(NULL, page_buffers(page),
5971 : : 0, len, NULL,
5972 : : ext4_bh_unmapped)) {
5973 : : /* Wait so that we don't change page under IO */
5974 : 0 : wait_for_stable_page(page);
5975 : 0 : ret = VM_FAULT_LOCKED;
5976 : 0 : goto out;
5977 : : }
5978 : : }
5979 : 0 : unlock_page(page);
5980 : : /* OK, we need to fill the hole... */
5981 [ # # ]: 0 : if (ext4_should_dioread_nolock(inode))
5982 : : get_block = ext4_get_block_unwritten;
5983 : : else
5984 : 0 : get_block = ext4_get_block;
5985 : 0 : retry_alloc:
5986 : 0 : handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE,
5987 : : ext4_writepage_trans_blocks(inode));
5988 [ # # ]: 0 : if (IS_ERR(handle)) {
5989 : 0 : ret = VM_FAULT_SIGBUS;
5990 : 0 : goto out;
5991 : : }
5992 : 0 : err = block_page_mkwrite(vma, vmf, get_block);
5993 [ # # # # ]: 0 : if (!err && ext4_should_journal_data(inode)) {
5994 [ # # # # ]: 0 : if (ext4_walk_page_buffers(handle, page_buffers(page), 0,
5995 : : PAGE_SIZE, NULL, do_journal_get_write_access)) {
5996 : 0 : unlock_page(page);
5997 : 0 : ret = VM_FAULT_SIGBUS;
5998 : 0 : ext4_journal_stop(handle);
5999 : 0 : goto out;
6000 : : }
6001 : 0 : ext4_set_inode_state(inode, EXT4_STATE_JDATA);
6002 : : }
6003 : 0 : ext4_journal_stop(handle);
6004 [ # # # # ]: 0 : if (err == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries))
6005 : 0 : goto retry_alloc;
6006 : 0 : out_ret:
6007 [ # # ]: 0 : ret = block_page_mkwrite_return(err);
6008 : 0 : out:
6009 : 0 : up_read(&EXT4_I(inode)->i_mmap_sem);
6010 : 0 : sb_end_pagefault(inode->i_sb);
6011 : 0 : return ret;
6012 : : }
6013 : :
6014 : 214047 : vm_fault_t ext4_filemap_fault(struct vm_fault *vmf)
6015 : : {
6016 : 214047 : struct inode *inode = file_inode(vmf->vma->vm_file);
6017 : 214047 : vm_fault_t ret;
6018 : :
6019 : 214047 : down_read(&EXT4_I(inode)->i_mmap_sem);
6020 : 214047 : ret = filemap_fault(vmf);
6021 : 214047 : up_read(&EXT4_I(inode)->i_mmap_sem);
6022 : :
6023 : 214047 : return ret;
6024 : : }
|