Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
ac27a0ec | 2 | /* |
617ba13b | 3 | * linux/fs/ext4/inode.c |
ac27a0ec DK |
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 | * | |
ac27a0ec DK |
16 | * 64-bit file support on 64-bit platforms by Jakub Jelinek |
17 | * (jj@sunsite.ms.mff.cuni.cz) | |
18 | * | |
617ba13b | 19 | * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000 |
ac27a0ec DK |
20 | */ |
21 | ||
ac27a0ec | 22 | #include <linux/fs.h> |
14f3db55 | 23 | #include <linux/mount.h> |
ac27a0ec | 24 | #include <linux/time.h> |
ac27a0ec DK |
25 | #include <linux/highuid.h> |
26 | #include <linux/pagemap.h> | |
c94c2acf | 27 | #include <linux/dax.h> |
ac27a0ec DK |
28 | #include <linux/quotaops.h> |
29 | #include <linux/string.h> | |
30 | #include <linux/buffer_head.h> | |
31 | #include <linux/writeback.h> | |
64769240 | 32 | #include <linux/pagevec.h> |
ac27a0ec | 33 | #include <linux/mpage.h> |
e83c1397 | 34 | #include <linux/namei.h> |
ac27a0ec DK |
35 | #include <linux/uio.h> |
36 | #include <linux/bio.h> | |
4c0425ff | 37 | #include <linux/workqueue.h> |
744692dc | 38 | #include <linux/kernel.h> |
6db26ffc | 39 | #include <linux/printk.h> |
5a0e3ad6 | 40 | #include <linux/slab.h> |
00a1a053 | 41 | #include <linux/bitops.h> |
364443cb | 42 | #include <linux/iomap.h> |
ae5e165d | 43 | #include <linux/iversion.h> |
9bffad1e | 44 | |
3dcf5451 | 45 | #include "ext4_jbd2.h" |
ac27a0ec DK |
46 | #include "xattr.h" |
47 | #include "acl.h" | |
9f125d64 | 48 | #include "truncate.h" |
ac27a0ec | 49 | |
9bffad1e TT |
50 | #include <trace/events/ext4.h> |
51 | ||
814525f4 DW |
52 | static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw, |
53 | struct ext4_inode_info *ei) | |
54 | { | |
55 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
814525f4 | 56 | __u32 csum; |
b47820ed DJ |
57 | __u16 dummy_csum = 0; |
58 | int offset = offsetof(struct ext4_inode, i_checksum_lo); | |
59 | unsigned int csum_size = sizeof(dummy_csum); | |
814525f4 | 60 | |
b47820ed DJ |
61 | csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw, offset); |
62 | csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, csum_size); | |
63 | offset += csum_size; | |
64 | csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset, | |
65 | EXT4_GOOD_OLD_INODE_SIZE - offset); | |
814525f4 | 66 | |
b47820ed DJ |
67 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
68 | offset = offsetof(struct ext4_inode, i_checksum_hi); | |
69 | csum = ext4_chksum(sbi, csum, (__u8 *)raw + | |
70 | EXT4_GOOD_OLD_INODE_SIZE, | |
71 | offset - EXT4_GOOD_OLD_INODE_SIZE); | |
72 | if (EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) { | |
73 | csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, | |
74 | csum_size); | |
75 | offset += csum_size; | |
b47820ed | 76 | } |
05ac5aa1 DJ |
77 | csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset, |
78 | EXT4_INODE_SIZE(inode->i_sb) - offset); | |
814525f4 DW |
79 | } |
80 | ||
814525f4 DW |
81 | return csum; |
82 | } | |
83 | ||
84 | static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw, | |
85 | struct ext4_inode_info *ei) | |
86 | { | |
87 | __u32 provided, calculated; | |
88 | ||
89 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != | |
90 | cpu_to_le32(EXT4_OS_LINUX) || | |
9aa5d32b | 91 | !ext4_has_metadata_csum(inode->i_sb)) |
814525f4 DW |
92 | return 1; |
93 | ||
94 | provided = le16_to_cpu(raw->i_checksum_lo); | |
95 | calculated = ext4_inode_csum(inode, raw, ei); | |
96 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && | |
97 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) | |
98 | provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16; | |
99 | else | |
100 | calculated &= 0xFFFF; | |
101 | ||
102 | return provided == calculated; | |
103 | } | |
104 | ||
8016e29f HS |
105 | void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw, |
106 | struct ext4_inode_info *ei) | |
814525f4 DW |
107 | { |
108 | __u32 csum; | |
109 | ||
110 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != | |
111 | cpu_to_le32(EXT4_OS_LINUX) || | |
9aa5d32b | 112 | !ext4_has_metadata_csum(inode->i_sb)) |
814525f4 DW |
113 | return; |
114 | ||
115 | csum = ext4_inode_csum(inode, raw, ei); | |
116 | raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF); | |
117 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && | |
118 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) | |
119 | raw->i_checksum_hi = cpu_to_le16(csum >> 16); | |
120 | } | |
121 | ||
678aaf48 JK |
122 | static inline int ext4_begin_ordered_truncate(struct inode *inode, |
123 | loff_t new_size) | |
124 | { | |
7ff9c073 | 125 | trace_ext4_begin_ordered_truncate(inode, new_size); |
8aefcd55 TT |
126 | /* |
127 | * If jinode is zero, then we never opened the file for | |
128 | * writing, so there's no need to call | |
129 | * jbd2_journal_begin_ordered_truncate() since there's no | |
130 | * outstanding writes we need to flush. | |
131 | */ | |
132 | if (!EXT4_I(inode)->jinode) | |
133 | return 0; | |
134 | return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode), | |
135 | EXT4_I(inode)->jinode, | |
136 | new_size); | |
678aaf48 JK |
137 | } |
138 | ||
cb20d518 | 139 | static int __ext4_journalled_writepage(struct page *page, unsigned int len); |
dec214d0 TE |
140 | static int ext4_meta_trans_blocks(struct inode *inode, int lblocks, |
141 | int pextents); | |
64769240 | 142 | |
ac27a0ec DK |
143 | /* |
144 | * Test whether an inode is a fast symlink. | |
407cd7fb | 145 | * A fast symlink has its symlink data stored in ext4_inode_info->i_data. |
ac27a0ec | 146 | */ |
f348c252 | 147 | int ext4_inode_is_fast_symlink(struct inode *inode) |
ac27a0ec | 148 | { |
fc82228a AK |
149 | if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) { |
150 | int ea_blocks = EXT4_I(inode)->i_file_acl ? | |
151 | EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0; | |
152 | ||
153 | if (ext4_has_inline_data(inode)) | |
154 | return 0; | |
155 | ||
156 | return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0); | |
157 | } | |
407cd7fb TE |
158 | return S_ISLNK(inode->i_mode) && inode->i_size && |
159 | (inode->i_size < EXT4_N_BLOCKS * 4); | |
ac27a0ec DK |
160 | } |
161 | ||
ac27a0ec DK |
162 | /* |
163 | * Called at the last iput() if i_nlink is zero. | |
164 | */ | |
0930fcc1 | 165 | void ext4_evict_inode(struct inode *inode) |
ac27a0ec DK |
166 | { |
167 | handle_t *handle; | |
bc965ab3 | 168 | int err; |
65db869c JK |
169 | /* |
170 | * Credits for final inode cleanup and freeing: | |
171 | * sb + inode (ext4_orphan_del()), block bitmap, group descriptor | |
172 | * (xattr block freeing), bitmap, group descriptor (inode freeing) | |
173 | */ | |
174 | int extra_credits = 6; | |
0421a189 | 175 | struct ext4_xattr_inode_array *ea_inode_array = NULL; |
46e294ef | 176 | bool freeze_protected = false; |
ac27a0ec | 177 | |
7ff9c073 | 178 | trace_ext4_evict_inode(inode); |
2581fdc8 | 179 | |
6bc0d63d JK |
180 | if (EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL) |
181 | ext4_evict_ea_inode(inode); | |
0930fcc1 | 182 | if (inode->i_nlink) { |
2d859db3 JK |
183 | /* |
184 | * When journalling data dirty buffers are tracked only in the | |
185 | * journal. So although mm thinks everything is clean and | |
186 | * ready for reaping the inode might still have some pages to | |
187 | * write in the running transaction or waiting to be | |
ccd16945 | 188 | * checkpointed. Thus calling jbd2_journal_invalidate_folio() |
2d859db3 JK |
189 | * (via truncate_inode_pages()) to discard these buffers can |
190 | * cause data loss. Also even if we did not discard these | |
191 | * buffers, we would have no way to find them after the inode | |
192 | * is reaped and thus user could see stale data if he tries to | |
193 | * read them before the transaction is checkpointed. So be | |
194 | * careful and force everything to disk here... We use | |
195 | * ei->i_datasync_tid to store the newest transaction | |
196 | * containing inode's data. | |
197 | * | |
198 | * Note that directories do not have this problem because they | |
199 | * don't use page cache. | |
200 | */ | |
6a7fd522 VN |
201 | if (inode->i_ino != EXT4_JOURNAL_INO && |
202 | ext4_should_journal_data(inode) && | |
6493792d | 203 | S_ISREG(inode->i_mode) && inode->i_data.nrpages) { |
2d859db3 JK |
204 | journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; |
205 | tid_t commit_tid = EXT4_I(inode)->i_datasync_tid; | |
206 | ||
d76a3a77 | 207 | jbd2_complete_transaction(journal, commit_tid); |
2d859db3 JK |
208 | filemap_write_and_wait(&inode->i_data); |
209 | } | |
91b0abe3 | 210 | truncate_inode_pages_final(&inode->i_data); |
5dc23bdd | 211 | |
0930fcc1 AV |
212 | goto no_delete; |
213 | } | |
214 | ||
e2bfb088 TT |
215 | if (is_bad_inode(inode)) |
216 | goto no_delete; | |
217 | dquot_initialize(inode); | |
907f4554 | 218 | |
678aaf48 JK |
219 | if (ext4_should_order_data(inode)) |
220 | ext4_begin_ordered_truncate(inode, 0); | |
91b0abe3 | 221 | truncate_inode_pages_final(&inode->i_data); |
ac27a0ec | 222 | |
ceff86fd JK |
223 | /* |
224 | * For inodes with journalled data, transaction commit could have | |
bc12ac98 ZY |
225 | * dirtied the inode. And for inodes with dioread_nolock, unwritten |
226 | * extents converting worker could merge extents and also have dirtied | |
227 | * the inode. Flush worker is ignoring it because of I_FREEING flag but | |
228 | * we still need to remove the inode from the writeback lists. | |
ceff86fd | 229 | */ |
bc12ac98 | 230 | if (!list_empty_careful(&inode->i_io_list)) |
ceff86fd | 231 | inode_io_list_del(inode); |
ceff86fd | 232 | |
8e8ad8a5 JK |
233 | /* |
234 | * Protect us against freezing - iput() caller didn't have to have any | |
46e294ef JK |
235 | * protection against it. When we are in a running transaction though, |
236 | * we are already protected against freezing and we cannot grab further | |
237 | * protection due to lock ordering constraints. | |
8e8ad8a5 | 238 | */ |
46e294ef JK |
239 | if (!ext4_journal_current_handle()) { |
240 | sb_start_intwrite(inode->i_sb); | |
241 | freeze_protected = true; | |
242 | } | |
e50e5129 | 243 | |
30a7eb97 TE |
244 | if (!IS_NOQUOTA(inode)) |
245 | extra_credits += EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb); | |
246 | ||
65db869c JK |
247 | /* |
248 | * Block bitmap, group descriptor, and inode are accounted in both | |
249 | * ext4_blocks_for_truncate() and extra_credits. So subtract 3. | |
250 | */ | |
30a7eb97 | 251 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, |
65db869c | 252 | ext4_blocks_for_truncate(inode) + extra_credits - 3); |
ac27a0ec | 253 | if (IS_ERR(handle)) { |
bc965ab3 | 254 | ext4_std_error(inode->i_sb, PTR_ERR(handle)); |
ac27a0ec DK |
255 | /* |
256 | * If we're going to skip the normal cleanup, we still need to | |
257 | * make sure that the in-core orphan linked list is properly | |
258 | * cleaned up. | |
259 | */ | |
617ba13b | 260 | ext4_orphan_del(NULL, inode); |
46e294ef JK |
261 | if (freeze_protected) |
262 | sb_end_intwrite(inode->i_sb); | |
ac27a0ec DK |
263 | goto no_delete; |
264 | } | |
30a7eb97 | 265 | |
ac27a0ec | 266 | if (IS_SYNC(inode)) |
0390131b | 267 | ext4_handle_sync(handle); |
407cd7fb TE |
268 | |
269 | /* | |
270 | * Set inode->i_size to 0 before calling ext4_truncate(). We need | |
271 | * special handling of symlinks here because i_size is used to | |
272 | * determine whether ext4_inode_info->i_data contains symlink data or | |
273 | * block mappings. Setting i_size to 0 will remove its fast symlink | |
274 | * status. Erase i_data so that it becomes a valid empty block map. | |
275 | */ | |
276 | if (ext4_inode_is_fast_symlink(inode)) | |
277 | memset(EXT4_I(inode)->i_data, 0, sizeof(EXT4_I(inode)->i_data)); | |
ac27a0ec | 278 | inode->i_size = 0; |
bc965ab3 TT |
279 | err = ext4_mark_inode_dirty(handle, inode); |
280 | if (err) { | |
12062ddd | 281 | ext4_warning(inode->i_sb, |
bc965ab3 TT |
282 | "couldn't mark inode dirty (err %d)", err); |
283 | goto stop_handle; | |
284 | } | |
2c98eb5e TT |
285 | if (inode->i_blocks) { |
286 | err = ext4_truncate(inode); | |
287 | if (err) { | |
54d3adbc TT |
288 | ext4_error_err(inode->i_sb, -err, |
289 | "couldn't truncate inode %lu (err %d)", | |
290 | inode->i_ino, err); | |
2c98eb5e TT |
291 | goto stop_handle; |
292 | } | |
293 | } | |
bc965ab3 | 294 | |
30a7eb97 TE |
295 | /* Remove xattr references. */ |
296 | err = ext4_xattr_delete_inode(handle, inode, &ea_inode_array, | |
297 | extra_credits); | |
298 | if (err) { | |
299 | ext4_warning(inode->i_sb, "xattr delete (err %d)", err); | |
300 | stop_handle: | |
301 | ext4_journal_stop(handle); | |
302 | ext4_orphan_del(NULL, inode); | |
46e294ef JK |
303 | if (freeze_protected) |
304 | sb_end_intwrite(inode->i_sb); | |
30a7eb97 TE |
305 | ext4_xattr_inode_array_free(ea_inode_array); |
306 | goto no_delete; | |
bc965ab3 TT |
307 | } |
308 | ||
ac27a0ec | 309 | /* |
617ba13b | 310 | * Kill off the orphan record which ext4_truncate created. |
ac27a0ec | 311 | * AKPM: I think this can be inside the above `if'. |
617ba13b | 312 | * Note that ext4_orphan_del() has to be able to cope with the |
ac27a0ec | 313 | * deletion of a non-existent orphan - this is because we don't |
617ba13b | 314 | * know if ext4_truncate() actually created an orphan record. |
ac27a0ec DK |
315 | * (Well, we could do this if we need to, but heck - it works) |
316 | */ | |
617ba13b | 317 | ext4_orphan_del(handle, inode); |
5ffff834 | 318 | EXT4_I(inode)->i_dtime = (__u32)ktime_get_real_seconds(); |
ac27a0ec DK |
319 | |
320 | /* | |
321 | * One subtle ordering requirement: if anything has gone wrong | |
322 | * (transaction abort, IO errors, whatever), then we can still | |
323 | * do these next steps (the fs will already have been marked as | |
324 | * having errors), but we can't free the inode if the mark_dirty | |
325 | * fails. | |
326 | */ | |
617ba13b | 327 | if (ext4_mark_inode_dirty(handle, inode)) |
ac27a0ec | 328 | /* If that failed, just do the required in-core inode clear. */ |
0930fcc1 | 329 | ext4_clear_inode(inode); |
ac27a0ec | 330 | else |
617ba13b MC |
331 | ext4_free_inode(handle, inode); |
332 | ext4_journal_stop(handle); | |
46e294ef JK |
333 | if (freeze_protected) |
334 | sb_end_intwrite(inode->i_sb); | |
0421a189 | 335 | ext4_xattr_inode_array_free(ea_inode_array); |
ac27a0ec DK |
336 | return; |
337 | no_delete: | |
318cdc82 ZY |
338 | /* |
339 | * Check out some where else accidentally dirty the evicting inode, | |
340 | * which may probably cause inode use-after-free issues later. | |
341 | */ | |
342 | WARN_ON_ONCE(!list_empty_careful(&inode->i_io_list)); | |
343 | ||
b21ebf14 | 344 | if (!list_empty(&EXT4_I(inode)->i_fc_list)) |
e85c81ba | 345 | ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_NOMEM, NULL); |
0930fcc1 | 346 | ext4_clear_inode(inode); /* We must guarantee clearing of inode... */ |
ac27a0ec DK |
347 | } |
348 | ||
a9e7f447 DM |
349 | #ifdef CONFIG_QUOTA |
350 | qsize_t *ext4_get_reserved_space(struct inode *inode) | |
60e58e0f | 351 | { |
a9e7f447 | 352 | return &EXT4_I(inode)->i_reserved_quota; |
60e58e0f | 353 | } |
a9e7f447 | 354 | #endif |
9d0be502 | 355 | |
0637c6f4 TT |
356 | /* |
357 | * Called with i_data_sem down, which is important since we can call | |
358 | * ext4_discard_preallocations() from here. | |
359 | */ | |
5f634d06 AK |
360 | void ext4_da_update_reserve_space(struct inode *inode, |
361 | int used, int quota_claim) | |
12219aea AK |
362 | { |
363 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0637c6f4 | 364 | struct ext4_inode_info *ei = EXT4_I(inode); |
0637c6f4 TT |
365 | |
366 | spin_lock(&ei->i_block_reservation_lock); | |
d8990240 | 367 | trace_ext4_da_update_reserve_space(inode, used, quota_claim); |
0637c6f4 | 368 | if (unlikely(used > ei->i_reserved_data_blocks)) { |
8de5c325 | 369 | ext4_warning(inode->i_sb, "%s: ino %lu, used %d " |
1084f252 | 370 | "with only %d reserved data blocks", |
0637c6f4 TT |
371 | __func__, inode->i_ino, used, |
372 | ei->i_reserved_data_blocks); | |
373 | WARN_ON(1); | |
374 | used = ei->i_reserved_data_blocks; | |
375 | } | |
12219aea | 376 | |
0637c6f4 TT |
377 | /* Update per-inode reservations */ |
378 | ei->i_reserved_data_blocks -= used; | |
71d4f7d0 | 379 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, used); |
6bc6e63f | 380 | |
f9505c72 | 381 | spin_unlock(&ei->i_block_reservation_lock); |
60e58e0f | 382 | |
72b8ab9d ES |
383 | /* Update quota subsystem for data blocks */ |
384 | if (quota_claim) | |
7b415bf6 | 385 | dquot_claim_block(inode, EXT4_C2B(sbi, used)); |
72b8ab9d | 386 | else { |
5f634d06 AK |
387 | /* |
388 | * We did fallocate with an offset that is already delayed | |
389 | * allocated. So on delayed allocated writeback we should | |
72b8ab9d | 390 | * not re-claim the quota for fallocated blocks. |
5f634d06 | 391 | */ |
7b415bf6 | 392 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, used)); |
5f634d06 | 393 | } |
d6014301 AK |
394 | |
395 | /* | |
396 | * If we have done all the pending block allocations and if | |
397 | * there aren't any writers on the inode, we can discard the | |
398 | * inode's preallocations. | |
399 | */ | |
0637c6f4 | 400 | if ((ei->i_reserved_data_blocks == 0) && |
82dd124c | 401 | !inode_is_open_for_write(inode)) |
27bc446e | 402 | ext4_discard_preallocations(inode, 0); |
12219aea AK |
403 | } |
404 | ||
e29136f8 | 405 | static int __check_block_validity(struct inode *inode, const char *func, |
c398eda0 TT |
406 | unsigned int line, |
407 | struct ext4_map_blocks *map) | |
6fd058f7 | 408 | { |
345c0dbf TT |
409 | if (ext4_has_feature_journal(inode->i_sb) && |
410 | (inode->i_ino == | |
411 | le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_journal_inum))) | |
412 | return 0; | |
ce9f24cc | 413 | if (!ext4_inode_block_valid(inode, map->m_pblk, map->m_len)) { |
c398eda0 | 414 | ext4_error_inode(inode, func, line, map->m_pblk, |
bdbd6ce0 | 415 | "lblock %lu mapped to illegal pblock %llu " |
c398eda0 | 416 | "(length %d)", (unsigned long) map->m_lblk, |
bdbd6ce0 | 417 | map->m_pblk, map->m_len); |
6a797d27 | 418 | return -EFSCORRUPTED; |
6fd058f7 TT |
419 | } |
420 | return 0; | |
421 | } | |
422 | ||
53085fac JK |
423 | int ext4_issue_zeroout(struct inode *inode, ext4_lblk_t lblk, ext4_fsblk_t pblk, |
424 | ext4_lblk_t len) | |
425 | { | |
426 | int ret; | |
427 | ||
33b4cc25 | 428 | if (IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode)) |
a7550b30 | 429 | return fscrypt_zeroout_range(inode, lblk, pblk, len); |
53085fac JK |
430 | |
431 | ret = sb_issue_zeroout(inode->i_sb, pblk, len, GFP_NOFS); | |
432 | if (ret > 0) | |
433 | ret = 0; | |
434 | ||
435 | return ret; | |
436 | } | |
437 | ||
e29136f8 | 438 | #define check_block_validity(inode, map) \ |
c398eda0 | 439 | __check_block_validity((inode), __func__, __LINE__, (map)) |
e29136f8 | 440 | |
921f266b DM |
441 | #ifdef ES_AGGRESSIVE_TEST |
442 | static void ext4_map_blocks_es_recheck(handle_t *handle, | |
443 | struct inode *inode, | |
444 | struct ext4_map_blocks *es_map, | |
445 | struct ext4_map_blocks *map, | |
446 | int flags) | |
447 | { | |
448 | int retval; | |
449 | ||
450 | map->m_flags = 0; | |
451 | /* | |
452 | * There is a race window that the result is not the same. | |
453 | * e.g. xfstests #223 when dioread_nolock enables. The reason | |
454 | * is that we lookup a block mapping in extent status tree with | |
455 | * out taking i_data_sem. So at the time the unwritten extent | |
456 | * could be converted. | |
457 | */ | |
2dcba478 | 458 | down_read(&EXT4_I(inode)->i_data_sem); |
921f266b | 459 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
9e52484c | 460 | retval = ext4_ext_map_blocks(handle, inode, map, 0); |
921f266b | 461 | } else { |
9e52484c | 462 | retval = ext4_ind_map_blocks(handle, inode, map, 0); |
921f266b | 463 | } |
2dcba478 | 464 | up_read((&EXT4_I(inode)->i_data_sem)); |
921f266b DM |
465 | |
466 | /* | |
467 | * We don't check m_len because extent will be collpased in status | |
468 | * tree. So the m_len might not equal. | |
469 | */ | |
470 | if (es_map->m_lblk != map->m_lblk || | |
471 | es_map->m_flags != map->m_flags || | |
472 | es_map->m_pblk != map->m_pblk) { | |
bdafe42a | 473 | printk("ES cache assertion failed for inode: %lu " |
921f266b DM |
474 | "es_cached ex [%d/%d/%llu/%x] != " |
475 | "found ex [%d/%d/%llu/%x] retval %d flags %x\n", | |
476 | inode->i_ino, es_map->m_lblk, es_map->m_len, | |
477 | es_map->m_pblk, es_map->m_flags, map->m_lblk, | |
478 | map->m_len, map->m_pblk, map->m_flags, | |
479 | retval, flags); | |
480 | } | |
481 | } | |
482 | #endif /* ES_AGGRESSIVE_TEST */ | |
483 | ||
f5ab0d1f | 484 | /* |
e35fd660 | 485 | * The ext4_map_blocks() function tries to look up the requested blocks, |
2b2d6d01 | 486 | * and returns if the blocks are already mapped. |
f5ab0d1f | 487 | * |
f5ab0d1f MC |
488 | * Otherwise it takes the write lock of the i_data_sem and allocate blocks |
489 | * and store the allocated blocks in the result buffer head and mark it | |
490 | * mapped. | |
491 | * | |
e35fd660 TT |
492 | * If file type is extents based, it will call ext4_ext_map_blocks(), |
493 | * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping | |
f5ab0d1f MC |
494 | * based files |
495 | * | |
facab4d9 JK |
496 | * On success, it returns the number of blocks being mapped or allocated. if |
497 | * create==0 and the blocks are pre-allocated and unwritten, the resulting @map | |
498 | * is marked as unwritten. If the create == 1, it will mark @map as mapped. | |
f5ab0d1f MC |
499 | * |
500 | * It returns 0 if plain look up failed (blocks have not been allocated), in | |
facab4d9 JK |
501 | * that case, @map is returned as unmapped but we still do fill map->m_len to |
502 | * indicate the length of a hole starting at map->m_lblk. | |
f5ab0d1f MC |
503 | * |
504 | * It returns the error in case of allocation failure. | |
505 | */ | |
e35fd660 TT |
506 | int ext4_map_blocks(handle_t *handle, struct inode *inode, |
507 | struct ext4_map_blocks *map, int flags) | |
0e855ac8 | 508 | { |
d100eef2 | 509 | struct extent_status es; |
0e855ac8 | 510 | int retval; |
b8a86845 | 511 | int ret = 0; |
921f266b DM |
512 | #ifdef ES_AGGRESSIVE_TEST |
513 | struct ext4_map_blocks orig_map; | |
514 | ||
515 | memcpy(&orig_map, map, sizeof(*map)); | |
516 | #endif | |
f5ab0d1f | 517 | |
e35fd660 | 518 | map->m_flags = 0; |
70aa1554 RH |
519 | ext_debug(inode, "flag 0x%x, max_blocks %u, logical block %lu\n", |
520 | flags, map->m_len, (unsigned long) map->m_lblk); | |
d100eef2 | 521 | |
e861b5e9 TT |
522 | /* |
523 | * ext4_map_blocks returns an int, and m_len is an unsigned int | |
524 | */ | |
525 | if (unlikely(map->m_len > INT_MAX)) | |
526 | map->m_len = INT_MAX; | |
527 | ||
4adb6ab3 KM |
528 | /* We can handle the block number less than EXT_MAX_BLOCKS */ |
529 | if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS)) | |
6a797d27 | 530 | return -EFSCORRUPTED; |
4adb6ab3 | 531 | |
d100eef2 | 532 | /* Lookup extent status tree firstly */ |
8016e29f HS |
533 | if (!(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY) && |
534 | ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) { | |
d100eef2 ZL |
535 | if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) { |
536 | map->m_pblk = ext4_es_pblock(&es) + | |
537 | map->m_lblk - es.es_lblk; | |
538 | map->m_flags |= ext4_es_is_written(&es) ? | |
539 | EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN; | |
540 | retval = es.es_len - (map->m_lblk - es.es_lblk); | |
541 | if (retval > map->m_len) | |
542 | retval = map->m_len; | |
543 | map->m_len = retval; | |
544 | } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) { | |
facab4d9 JK |
545 | map->m_pblk = 0; |
546 | retval = es.es_len - (map->m_lblk - es.es_lblk); | |
547 | if (retval > map->m_len) | |
548 | retval = map->m_len; | |
549 | map->m_len = retval; | |
d100eef2 ZL |
550 | retval = 0; |
551 | } else { | |
1e83bc81 | 552 | BUG(); |
d100eef2 | 553 | } |
9558cf14 ZY |
554 | |
555 | if (flags & EXT4_GET_BLOCKS_CACHED_NOWAIT) | |
556 | return retval; | |
921f266b DM |
557 | #ifdef ES_AGGRESSIVE_TEST |
558 | ext4_map_blocks_es_recheck(handle, inode, map, | |
559 | &orig_map, flags); | |
560 | #endif | |
d100eef2 ZL |
561 | goto found; |
562 | } | |
9558cf14 ZY |
563 | /* |
564 | * In the query cache no-wait mode, nothing we can do more if we | |
565 | * cannot find extent in the cache. | |
566 | */ | |
567 | if (flags & EXT4_GET_BLOCKS_CACHED_NOWAIT) | |
568 | return 0; | |
d100eef2 | 569 | |
4df3d265 | 570 | /* |
b920c755 TT |
571 | * Try to see if we can get the block without requesting a new |
572 | * file system block. | |
4df3d265 | 573 | */ |
2dcba478 | 574 | down_read(&EXT4_I(inode)->i_data_sem); |
12e9b892 | 575 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
9e52484c | 576 | retval = ext4_ext_map_blocks(handle, inode, map, 0); |
0e855ac8 | 577 | } else { |
9e52484c | 578 | retval = ext4_ind_map_blocks(handle, inode, map, 0); |
0e855ac8 | 579 | } |
f7fec032 | 580 | if (retval > 0) { |
3be78c73 | 581 | unsigned int status; |
f7fec032 | 582 | |
44fb851d ZL |
583 | if (unlikely(retval != map->m_len)) { |
584 | ext4_warning(inode->i_sb, | |
585 | "ES len assertion failed for inode " | |
586 | "%lu: retval %d != map->m_len %d", | |
587 | inode->i_ino, retval, map->m_len); | |
588 | WARN_ON(1); | |
921f266b | 589 | } |
921f266b | 590 | |
f7fec032 ZL |
591 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
592 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; | |
593 | if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) && | |
d2dc317d | 594 | !(status & EXTENT_STATUS_WRITTEN) && |
ad431025 EW |
595 | ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk, |
596 | map->m_lblk + map->m_len - 1)) | |
f7fec032 ZL |
597 | status |= EXTENT_STATUS_DELAYED; |
598 | ret = ext4_es_insert_extent(inode, map->m_lblk, | |
599 | map->m_len, map->m_pblk, status); | |
600 | if (ret < 0) | |
601 | retval = ret; | |
602 | } | |
2dcba478 | 603 | up_read((&EXT4_I(inode)->i_data_sem)); |
f5ab0d1f | 604 | |
d100eef2 | 605 | found: |
e35fd660 | 606 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
b8a86845 | 607 | ret = check_block_validity(inode, map); |
6fd058f7 TT |
608 | if (ret != 0) |
609 | return ret; | |
610 | } | |
611 | ||
f5ab0d1f | 612 | /* If it is only a block(s) look up */ |
c2177057 | 613 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) |
f5ab0d1f MC |
614 | return retval; |
615 | ||
616 | /* | |
617 | * Returns if the blocks have already allocated | |
618 | * | |
619 | * Note that if blocks have been preallocated | |
df3ab170 | 620 | * ext4_ext_get_block() returns the create = 0 |
f5ab0d1f MC |
621 | * with buffer head unmapped. |
622 | */ | |
e35fd660 | 623 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) |
b8a86845 LC |
624 | /* |
625 | * If we need to convert extent to unwritten | |
626 | * we continue and do the actual work in | |
627 | * ext4_ext_map_blocks() | |
628 | */ | |
629 | if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) | |
630 | return retval; | |
4df3d265 | 631 | |
2a8964d6 | 632 | /* |
a25a4e1a ZL |
633 | * Here we clear m_flags because after allocating an new extent, |
634 | * it will be set again. | |
2a8964d6 | 635 | */ |
a25a4e1a | 636 | map->m_flags &= ~EXT4_MAP_FLAGS; |
2a8964d6 | 637 | |
4df3d265 | 638 | /* |
556615dc | 639 | * New blocks allocate and/or writing to unwritten extent |
f5ab0d1f | 640 | * will possibly result in updating i_data, so we take |
d91bd2c1 | 641 | * the write lock of i_data_sem, and call get_block() |
f5ab0d1f | 642 | * with create == 1 flag. |
4df3d265 | 643 | */ |
c8b459f4 | 644 | down_write(&EXT4_I(inode)->i_data_sem); |
d2a17637 | 645 | |
4df3d265 AK |
646 | /* |
647 | * We need to check for EXT4 here because migrate | |
648 | * could have changed the inode type in between | |
649 | */ | |
12e9b892 | 650 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
e35fd660 | 651 | retval = ext4_ext_map_blocks(handle, inode, map, flags); |
0e855ac8 | 652 | } else { |
e35fd660 | 653 | retval = ext4_ind_map_blocks(handle, inode, map, flags); |
267e4db9 | 654 | |
e35fd660 | 655 | if (retval > 0 && map->m_flags & EXT4_MAP_NEW) { |
267e4db9 AK |
656 | /* |
657 | * We allocated new blocks which will result in | |
658 | * i_data's format changing. Force the migrate | |
659 | * to fail by clearing migrate flags | |
660 | */ | |
19f5fb7a | 661 | ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE); |
267e4db9 | 662 | } |
d2a17637 | 663 | |
5f634d06 AK |
664 | /* |
665 | * Update reserved blocks/metadata blocks after successful | |
666 | * block allocation which had been deferred till now. We don't | |
667 | * support fallocate for non extent files. So we can update | |
668 | * reserve space here. | |
669 | */ | |
670 | if ((retval > 0) && | |
1296cc85 | 671 | (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)) |
5f634d06 AK |
672 | ext4_da_update_reserve_space(inode, retval, 1); |
673 | } | |
2ac3b6e0 | 674 | |
f7fec032 | 675 | if (retval > 0) { |
3be78c73 | 676 | unsigned int status; |
f7fec032 | 677 | |
44fb851d ZL |
678 | if (unlikely(retval != map->m_len)) { |
679 | ext4_warning(inode->i_sb, | |
680 | "ES len assertion failed for inode " | |
681 | "%lu: retval %d != map->m_len %d", | |
682 | inode->i_ino, retval, map->m_len); | |
683 | WARN_ON(1); | |
921f266b | 684 | } |
921f266b | 685 | |
c86d8db3 JK |
686 | /* |
687 | * We have to zeroout blocks before inserting them into extent | |
688 | * status tree. Otherwise someone could look them up there and | |
9b623df6 JK |
689 | * use them before they are really zeroed. We also have to |
690 | * unmap metadata before zeroing as otherwise writeback can | |
691 | * overwrite zeros with stale data from block device. | |
c86d8db3 JK |
692 | */ |
693 | if (flags & EXT4_GET_BLOCKS_ZERO && | |
694 | map->m_flags & EXT4_MAP_MAPPED && | |
695 | map->m_flags & EXT4_MAP_NEW) { | |
696 | ret = ext4_issue_zeroout(inode, map->m_lblk, | |
697 | map->m_pblk, map->m_len); | |
698 | if (ret) { | |
699 | retval = ret; | |
700 | goto out_sem; | |
701 | } | |
702 | } | |
703 | ||
adb23551 ZL |
704 | /* |
705 | * If the extent has been zeroed out, we don't need to update | |
706 | * extent status tree. | |
707 | */ | |
708 | if ((flags & EXT4_GET_BLOCKS_PRE_IO) && | |
bb5835ed | 709 | ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) { |
adb23551 | 710 | if (ext4_es_is_written(&es)) |
c86d8db3 | 711 | goto out_sem; |
adb23551 | 712 | } |
f7fec032 ZL |
713 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
714 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; | |
715 | if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) && | |
d2dc317d | 716 | !(status & EXTENT_STATUS_WRITTEN) && |
ad431025 EW |
717 | ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk, |
718 | map->m_lblk + map->m_len - 1)) | |
f7fec032 ZL |
719 | status |= EXTENT_STATUS_DELAYED; |
720 | ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, | |
721 | map->m_pblk, status); | |
c86d8db3 | 722 | if (ret < 0) { |
f7fec032 | 723 | retval = ret; |
c86d8db3 JK |
724 | goto out_sem; |
725 | } | |
5356f261 AK |
726 | } |
727 | ||
c86d8db3 | 728 | out_sem: |
4df3d265 | 729 | up_write((&EXT4_I(inode)->i_data_sem)); |
e35fd660 | 730 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
b8a86845 | 731 | ret = check_block_validity(inode, map); |
6fd058f7 TT |
732 | if (ret != 0) |
733 | return ret; | |
06bd3c36 JK |
734 | |
735 | /* | |
736 | * Inodes with freshly allocated blocks where contents will be | |
737 | * visible after transaction commit must be on transaction's | |
738 | * ordered data list. | |
739 | */ | |
740 | if (map->m_flags & EXT4_MAP_NEW && | |
741 | !(map->m_flags & EXT4_MAP_UNWRITTEN) && | |
742 | !(flags & EXT4_GET_BLOCKS_ZERO) && | |
02749a4c | 743 | !ext4_is_quota_file(inode) && |
06bd3c36 | 744 | ext4_should_order_data(inode)) { |
73131fbb RZ |
745 | loff_t start_byte = |
746 | (loff_t)map->m_lblk << inode->i_blkbits; | |
747 | loff_t length = (loff_t)map->m_len << inode->i_blkbits; | |
748 | ||
ee0876bc | 749 | if (flags & EXT4_GET_BLOCKS_IO_SUBMIT) |
73131fbb RZ |
750 | ret = ext4_jbd2_inode_add_wait(handle, inode, |
751 | start_byte, length); | |
ee0876bc | 752 | else |
73131fbb RZ |
753 | ret = ext4_jbd2_inode_add_write(handle, inode, |
754 | start_byte, length); | |
06bd3c36 JK |
755 | if (ret) |
756 | return ret; | |
757 | } | |
6fd058f7 | 758 | } |
5e4d0eba XY |
759 | if (retval > 0 && (map->m_flags & EXT4_MAP_UNWRITTEN || |
760 | map->m_flags & EXT4_MAP_MAPPED)) | |
761 | ext4_fc_track_range(handle, inode, map->m_lblk, | |
762 | map->m_lblk + map->m_len - 1); | |
ec8c60be | 763 | if (retval < 0) |
70aa1554 | 764 | ext_debug(inode, "failed with err %d\n", retval); |
0e855ac8 AK |
765 | return retval; |
766 | } | |
767 | ||
ed8ad838 JK |
768 | /* |
769 | * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages | |
770 | * we have to be careful as someone else may be manipulating b_state as well. | |
771 | */ | |
772 | static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags) | |
773 | { | |
774 | unsigned long old_state; | |
775 | unsigned long new_state; | |
776 | ||
777 | flags &= EXT4_MAP_FLAGS; | |
778 | ||
779 | /* Dummy buffer_head? Set non-atomically. */ | |
780 | if (!bh->b_page) { | |
781 | bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags; | |
782 | return; | |
783 | } | |
784 | /* | |
785 | * Someone else may be modifying b_state. Be careful! This is ugly but | |
786 | * once we get rid of using bh as a container for mapping information | |
787 | * to pass to / from get_block functions, this can go away. | |
788 | */ | |
789 | do { | |
790 | old_state = READ_ONCE(bh->b_state); | |
791 | new_state = (old_state & ~EXT4_MAP_FLAGS) | flags; | |
792 | } while (unlikely( | |
793 | cmpxchg(&bh->b_state, old_state, new_state) != old_state)); | |
794 | } | |
795 | ||
2ed88685 TT |
796 | static int _ext4_get_block(struct inode *inode, sector_t iblock, |
797 | struct buffer_head *bh, int flags) | |
ac27a0ec | 798 | { |
2ed88685 | 799 | struct ext4_map_blocks map; |
efe70c29 | 800 | int ret = 0; |
ac27a0ec | 801 | |
46c7f254 TM |
802 | if (ext4_has_inline_data(inode)) |
803 | return -ERANGE; | |
804 | ||
2ed88685 TT |
805 | map.m_lblk = iblock; |
806 | map.m_len = bh->b_size >> inode->i_blkbits; | |
807 | ||
efe70c29 JK |
808 | ret = ext4_map_blocks(ext4_journal_current_handle(), inode, &map, |
809 | flags); | |
7fb5409d | 810 | if (ret > 0) { |
2ed88685 | 811 | map_bh(bh, inode->i_sb, map.m_pblk); |
ed8ad838 | 812 | ext4_update_bh_state(bh, map.m_flags); |
2ed88685 | 813 | bh->b_size = inode->i_sb->s_blocksize * map.m_len; |
7fb5409d | 814 | ret = 0; |
547edce3 RZ |
815 | } else if (ret == 0) { |
816 | /* hole case, need to fill in bh->b_size */ | |
817 | bh->b_size = inode->i_sb->s_blocksize * map.m_len; | |
ac27a0ec DK |
818 | } |
819 | return ret; | |
820 | } | |
821 | ||
2ed88685 TT |
822 | int ext4_get_block(struct inode *inode, sector_t iblock, |
823 | struct buffer_head *bh, int create) | |
824 | { | |
825 | return _ext4_get_block(inode, iblock, bh, | |
826 | create ? EXT4_GET_BLOCKS_CREATE : 0); | |
827 | } | |
828 | ||
705965bd JK |
829 | /* |
830 | * Get block function used when preparing for buffered write if we require | |
831 | * creating an unwritten extent if blocks haven't been allocated. The extent | |
832 | * will be converted to written after the IO is complete. | |
833 | */ | |
834 | int ext4_get_block_unwritten(struct inode *inode, sector_t iblock, | |
835 | struct buffer_head *bh_result, int create) | |
836 | { | |
837 | ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n", | |
838 | inode->i_ino, create); | |
839 | return _ext4_get_block(inode, iblock, bh_result, | |
8d5459c1 | 840 | EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT); |
705965bd JK |
841 | } |
842 | ||
efe70c29 JK |
843 | /* Maximum number of blocks we map for direct IO at once. */ |
844 | #define DIO_MAX_BLOCKS 4096 | |
845 | ||
ac27a0ec DK |
846 | /* |
847 | * `handle' can be NULL if create is zero | |
848 | */ | |
617ba13b | 849 | struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode, |
c5e298ae | 850 | ext4_lblk_t block, int map_flags) |
ac27a0ec | 851 | { |
2ed88685 TT |
852 | struct ext4_map_blocks map; |
853 | struct buffer_head *bh; | |
c5e298ae | 854 | int create = map_flags & EXT4_GET_BLOCKS_CREATE; |
9558cf14 | 855 | bool nowait = map_flags & EXT4_GET_BLOCKS_CACHED_NOWAIT; |
10560082 | 856 | int err; |
ac27a0ec | 857 | |
837c23fb CX |
858 | ASSERT((EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY) |
859 | || handle != NULL || create == 0); | |
9558cf14 | 860 | ASSERT(create == 0 || !nowait); |
ac27a0ec | 861 | |
2ed88685 TT |
862 | map.m_lblk = block; |
863 | map.m_len = 1; | |
c5e298ae | 864 | err = ext4_map_blocks(handle, inode, &map, map_flags); |
ac27a0ec | 865 | |
10560082 TT |
866 | if (err == 0) |
867 | return create ? ERR_PTR(-ENOSPC) : NULL; | |
2ed88685 | 868 | if (err < 0) |
10560082 | 869 | return ERR_PTR(err); |
2ed88685 | 870 | |
9558cf14 ZY |
871 | if (nowait) |
872 | return sb_find_get_block(inode->i_sb, map.m_pblk); | |
873 | ||
2ed88685 | 874 | bh = sb_getblk(inode->i_sb, map.m_pblk); |
10560082 TT |
875 | if (unlikely(!bh)) |
876 | return ERR_PTR(-ENOMEM); | |
2ed88685 | 877 | if (map.m_flags & EXT4_MAP_NEW) { |
837c23fb CX |
878 | ASSERT(create != 0); |
879 | ASSERT((EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY) | |
880 | || (handle != NULL)); | |
ac27a0ec | 881 | |
2ed88685 TT |
882 | /* |
883 | * Now that we do not always journal data, we should | |
884 | * keep in mind whether this should always journal the | |
885 | * new buffer as metadata. For now, regular file | |
886 | * writes use ext4_get_block instead, so it's not a | |
887 | * problem. | |
888 | */ | |
889 | lock_buffer(bh); | |
890 | BUFFER_TRACE(bh, "call get_create_access"); | |
188c299e JK |
891 | err = ext4_journal_get_create_access(handle, inode->i_sb, bh, |
892 | EXT4_JTR_NONE); | |
10560082 TT |
893 | if (unlikely(err)) { |
894 | unlock_buffer(bh); | |
895 | goto errout; | |
896 | } | |
897 | if (!buffer_uptodate(bh)) { | |
2ed88685 TT |
898 | memset(bh->b_data, 0, inode->i_sb->s_blocksize); |
899 | set_buffer_uptodate(bh); | |
ac27a0ec | 900 | } |
2ed88685 TT |
901 | unlock_buffer(bh); |
902 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); | |
903 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
10560082 TT |
904 | if (unlikely(err)) |
905 | goto errout; | |
906 | } else | |
2ed88685 | 907 | BUFFER_TRACE(bh, "not a new buffer"); |
2ed88685 | 908 | return bh; |
10560082 TT |
909 | errout: |
910 | brelse(bh); | |
911 | return ERR_PTR(err); | |
ac27a0ec DK |
912 | } |
913 | ||
617ba13b | 914 | struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode, |
c5e298ae | 915 | ext4_lblk_t block, int map_flags) |
ac27a0ec | 916 | { |
af5bc92d | 917 | struct buffer_head *bh; |
2d069c08 | 918 | int ret; |
ac27a0ec | 919 | |
c5e298ae | 920 | bh = ext4_getblk(handle, inode, block, map_flags); |
1c215028 | 921 | if (IS_ERR(bh)) |
ac27a0ec | 922 | return bh; |
7963e5ac | 923 | if (!bh || ext4_buffer_uptodate(bh)) |
ac27a0ec | 924 | return bh; |
2d069c08 | 925 | |
926 | ret = ext4_read_bh_lock(bh, REQ_META | REQ_PRIO, true); | |
927 | if (ret) { | |
928 | put_bh(bh); | |
929 | return ERR_PTR(ret); | |
930 | } | |
931 | return bh; | |
ac27a0ec DK |
932 | } |
933 | ||
9699d4f9 TE |
934 | /* Read a contiguous batch of blocks. */ |
935 | int ext4_bread_batch(struct inode *inode, ext4_lblk_t block, int bh_count, | |
936 | bool wait, struct buffer_head **bhs) | |
937 | { | |
938 | int i, err; | |
939 | ||
940 | for (i = 0; i < bh_count; i++) { | |
941 | bhs[i] = ext4_getblk(NULL, inode, block + i, 0 /* map_flags */); | |
942 | if (IS_ERR(bhs[i])) { | |
943 | err = PTR_ERR(bhs[i]); | |
944 | bh_count = i; | |
945 | goto out_brelse; | |
946 | } | |
947 | } | |
948 | ||
949 | for (i = 0; i < bh_count; i++) | |
950 | /* Note that NULL bhs[i] is valid because of holes. */ | |
2d069c08 | 951 | if (bhs[i] && !ext4_buffer_uptodate(bhs[i])) |
952 | ext4_read_bh_lock(bhs[i], REQ_META | REQ_PRIO, false); | |
9699d4f9 TE |
953 | |
954 | if (!wait) | |
955 | return 0; | |
956 | ||
957 | for (i = 0; i < bh_count; i++) | |
958 | if (bhs[i]) | |
959 | wait_on_buffer(bhs[i]); | |
960 | ||
961 | for (i = 0; i < bh_count; i++) { | |
962 | if (bhs[i] && !buffer_uptodate(bhs[i])) { | |
963 | err = -EIO; | |
964 | goto out_brelse; | |
965 | } | |
966 | } | |
967 | return 0; | |
968 | ||
969 | out_brelse: | |
970 | for (i = 0; i < bh_count; i++) { | |
971 | brelse(bhs[i]); | |
972 | bhs[i] = NULL; | |
973 | } | |
974 | return err; | |
975 | } | |
976 | ||
188c299e | 977 | int ext4_walk_page_buffers(handle_t *handle, struct inode *inode, |
f19d5870 TM |
978 | struct buffer_head *head, |
979 | unsigned from, | |
980 | unsigned to, | |
981 | int *partial, | |
188c299e | 982 | int (*fn)(handle_t *handle, struct inode *inode, |
f19d5870 | 983 | struct buffer_head *bh)) |
ac27a0ec DK |
984 | { |
985 | struct buffer_head *bh; | |
986 | unsigned block_start, block_end; | |
987 | unsigned blocksize = head->b_size; | |
988 | int err, ret = 0; | |
989 | struct buffer_head *next; | |
990 | ||
af5bc92d TT |
991 | for (bh = head, block_start = 0; |
992 | ret == 0 && (bh != head || !block_start); | |
de9a55b8 | 993 | block_start = block_end, bh = next) { |
ac27a0ec DK |
994 | next = bh->b_this_page; |
995 | block_end = block_start + blocksize; | |
996 | if (block_end <= from || block_start >= to) { | |
997 | if (partial && !buffer_uptodate(bh)) | |
998 | *partial = 1; | |
999 | continue; | |
1000 | } | |
188c299e | 1001 | err = (*fn)(handle, inode, bh); |
ac27a0ec DK |
1002 | if (!ret) |
1003 | ret = err; | |
1004 | } | |
1005 | return ret; | |
1006 | } | |
1007 | ||
1008 | /* | |
1009 | * To preserve ordering, it is essential that the hole instantiation and | |
1010 | * the data write be encapsulated in a single transaction. We cannot | |
617ba13b | 1011 | * close off a transaction and start a new one between the ext4_get_block() |
dab291af | 1012 | * and the commit_write(). So doing the jbd2_journal_start at the start of |
ac27a0ec DK |
1013 | * prepare_write() is the right place. |
1014 | * | |
36ade451 JK |
1015 | * Also, this function can nest inside ext4_writepage(). In that case, we |
1016 | * *know* that ext4_writepage() has generated enough buffer credits to do the | |
1017 | * whole page. So we won't block on the journal in that case, which is good, | |
1018 | * because the caller may be PF_MEMALLOC. | |
ac27a0ec | 1019 | * |
617ba13b | 1020 | * By accident, ext4 can be reentered when a transaction is open via |
ac27a0ec DK |
1021 | * quota file writes. If we were to commit the transaction while thus |
1022 | * reentered, there can be a deadlock - we would be holding a quota | |
1023 | * lock, and the commit would never complete if another thread had a | |
1024 | * transaction open and was blocking on the quota lock - a ranking | |
1025 | * violation. | |
1026 | * | |
dab291af | 1027 | * So what we do is to rely on the fact that jbd2_journal_stop/journal_start |
ac27a0ec DK |
1028 | * will _not_ run commit under these circumstances because handle->h_ref |
1029 | * is elevated. We'll still have enough credits for the tiny quotafile | |
1030 | * write. | |
1031 | */ | |
188c299e | 1032 | int do_journal_get_write_access(handle_t *handle, struct inode *inode, |
f19d5870 | 1033 | struct buffer_head *bh) |
ac27a0ec | 1034 | { |
56d35a4c JK |
1035 | int dirty = buffer_dirty(bh); |
1036 | int ret; | |
1037 | ||
ac27a0ec DK |
1038 | if (!buffer_mapped(bh) || buffer_freed(bh)) |
1039 | return 0; | |
56d35a4c | 1040 | /* |
ebdec241 | 1041 | * __block_write_begin() could have dirtied some buffers. Clean |
56d35a4c JK |
1042 | * the dirty bit as jbd2_journal_get_write_access() could complain |
1043 | * otherwise about fs integrity issues. Setting of the dirty bit | |
ebdec241 | 1044 | * by __block_write_begin() isn't a real problem here as we clear |
56d35a4c JK |
1045 | * the bit before releasing a page lock and thus writeback cannot |
1046 | * ever write the buffer. | |
1047 | */ | |
1048 | if (dirty) | |
1049 | clear_buffer_dirty(bh); | |
5d601255 | 1050 | BUFFER_TRACE(bh, "get write access"); |
188c299e JK |
1051 | ret = ext4_journal_get_write_access(handle, inode->i_sb, bh, |
1052 | EXT4_JTR_NONE); | |
56d35a4c JK |
1053 | if (!ret && dirty) |
1054 | ret = ext4_handle_dirty_metadata(handle, NULL, bh); | |
1055 | return ret; | |
ac27a0ec DK |
1056 | } |
1057 | ||
643fa961 | 1058 | #ifdef CONFIG_FS_ENCRYPTION |
2058f83a MH |
1059 | static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len, |
1060 | get_block_t *get_block) | |
1061 | { | |
09cbfeaf | 1062 | unsigned from = pos & (PAGE_SIZE - 1); |
2058f83a MH |
1063 | unsigned to = from + len; |
1064 | struct inode *inode = page->mapping->host; | |
1065 | unsigned block_start, block_end; | |
1066 | sector_t block; | |
1067 | int err = 0; | |
1068 | unsigned blocksize = inode->i_sb->s_blocksize; | |
1069 | unsigned bbits; | |
0b578f35 CR |
1070 | struct buffer_head *bh, *head, *wait[2]; |
1071 | int nr_wait = 0; | |
1072 | int i; | |
2058f83a MH |
1073 | |
1074 | BUG_ON(!PageLocked(page)); | |
09cbfeaf KS |
1075 | BUG_ON(from > PAGE_SIZE); |
1076 | BUG_ON(to > PAGE_SIZE); | |
2058f83a MH |
1077 | BUG_ON(from > to); |
1078 | ||
1079 | if (!page_has_buffers(page)) | |
1080 | create_empty_buffers(page, blocksize, 0); | |
1081 | head = page_buffers(page); | |
1082 | bbits = ilog2(blocksize); | |
09cbfeaf | 1083 | block = (sector_t)page->index << (PAGE_SHIFT - bbits); |
2058f83a MH |
1084 | |
1085 | for (bh = head, block_start = 0; bh != head || !block_start; | |
1086 | block++, block_start = block_end, bh = bh->b_this_page) { | |
1087 | block_end = block_start + blocksize; | |
1088 | if (block_end <= from || block_start >= to) { | |
1089 | if (PageUptodate(page)) { | |
3cd46171 | 1090 | set_buffer_uptodate(bh); |
2058f83a MH |
1091 | } |
1092 | continue; | |
1093 | } | |
1094 | if (buffer_new(bh)) | |
1095 | clear_buffer_new(bh); | |
1096 | if (!buffer_mapped(bh)) { | |
1097 | WARN_ON(bh->b_size != blocksize); | |
1098 | err = get_block(inode, block, bh, 1); | |
1099 | if (err) | |
1100 | break; | |
1101 | if (buffer_new(bh)) { | |
2058f83a MH |
1102 | if (PageUptodate(page)) { |
1103 | clear_buffer_new(bh); | |
1104 | set_buffer_uptodate(bh); | |
1105 | mark_buffer_dirty(bh); | |
1106 | continue; | |
1107 | } | |
1108 | if (block_end > to || block_start < from) | |
1109 | zero_user_segments(page, to, block_end, | |
1110 | block_start, from); | |
1111 | continue; | |
1112 | } | |
1113 | } | |
1114 | if (PageUptodate(page)) { | |
3cd46171 | 1115 | set_buffer_uptodate(bh); |
2058f83a MH |
1116 | continue; |
1117 | } | |
1118 | if (!buffer_uptodate(bh) && !buffer_delay(bh) && | |
1119 | !buffer_unwritten(bh) && | |
1120 | (block_start < from || block_end > to)) { | |
2d069c08 | 1121 | ext4_read_bh_lock(bh, 0, false); |
0b578f35 | 1122 | wait[nr_wait++] = bh; |
2058f83a MH |
1123 | } |
1124 | } | |
1125 | /* | |
1126 | * If we issued read requests, let them complete. | |
1127 | */ | |
0b578f35 CR |
1128 | for (i = 0; i < nr_wait; i++) { |
1129 | wait_on_buffer(wait[i]); | |
1130 | if (!buffer_uptodate(wait[i])) | |
2058f83a MH |
1131 | err = -EIO; |
1132 | } | |
7e0785fc | 1133 | if (unlikely(err)) { |
2058f83a | 1134 | page_zero_new_buffers(page, from, to); |
4f74d15f | 1135 | } else if (fscrypt_inode_uses_fs_layer_crypto(inode)) { |
0b578f35 CR |
1136 | for (i = 0; i < nr_wait; i++) { |
1137 | int err2; | |
1138 | ||
1139 | err2 = fscrypt_decrypt_pagecache_blocks(page, blocksize, | |
1140 | bh_offset(wait[i])); | |
1141 | if (err2) { | |
1142 | clear_buffer_uptodate(wait[i]); | |
1143 | err = err2; | |
1144 | } | |
1145 | } | |
7e0785fc CR |
1146 | } |
1147 | ||
2058f83a MH |
1148 | return err; |
1149 | } | |
1150 | #endif | |
1151 | ||
bfc1af65 | 1152 | static int ext4_write_begin(struct file *file, struct address_space *mapping, |
9d6b0cd7 | 1153 | loff_t pos, unsigned len, |
de9a55b8 | 1154 | struct page **pagep, void **fsdata) |
ac27a0ec | 1155 | { |
af5bc92d | 1156 | struct inode *inode = mapping->host; |
1938a150 | 1157 | int ret, needed_blocks; |
ac27a0ec DK |
1158 | handle_t *handle; |
1159 | int retries = 0; | |
af5bc92d | 1160 | struct page *page; |
de9a55b8 | 1161 | pgoff_t index; |
af5bc92d | 1162 | unsigned from, to; |
bfc1af65 | 1163 | |
0db1ff22 TT |
1164 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
1165 | return -EIO; | |
1166 | ||
9d6b0cd7 | 1167 | trace_ext4_write_begin(inode, pos, len); |
1938a150 AK |
1168 | /* |
1169 | * Reserve one block more for addition to orphan list in case | |
1170 | * we allocate blocks but write fails for some reason | |
1171 | */ | |
1172 | needed_blocks = ext4_writepage_trans_blocks(inode) + 1; | |
09cbfeaf KS |
1173 | index = pos >> PAGE_SHIFT; |
1174 | from = pos & (PAGE_SIZE - 1); | |
af5bc92d | 1175 | to = from + len; |
ac27a0ec | 1176 | |
f19d5870 TM |
1177 | if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { |
1178 | ret = ext4_try_to_write_inline_data(mapping, inode, pos, len, | |
832ee62d | 1179 | pagep); |
f19d5870 | 1180 | if (ret < 0) |
47564bfb TT |
1181 | return ret; |
1182 | if (ret == 1) | |
1183 | return 0; | |
f19d5870 TM |
1184 | } |
1185 | ||
47564bfb TT |
1186 | /* |
1187 | * grab_cache_page_write_begin() can take a long time if the | |
1188 | * system is thrashing due to memory pressure, or if the page | |
1189 | * is being written back. So grab it first before we start | |
1190 | * the transaction handle. This also allows us to allocate | |
1191 | * the page (if needed) without using GFP_NOFS. | |
1192 | */ | |
1193 | retry_grab: | |
b7446e7c | 1194 | page = grab_cache_page_write_begin(mapping, index); |
47564bfb TT |
1195 | if (!page) |
1196 | return -ENOMEM; | |
d1052d23 JH |
1197 | /* |
1198 | * The same as page allocation, we prealloc buffer heads before | |
1199 | * starting the handle. | |
1200 | */ | |
1201 | if (!page_has_buffers(page)) | |
1202 | create_empty_buffers(page, inode->i_sb->s_blocksize, 0); | |
1203 | ||
47564bfb TT |
1204 | unlock_page(page); |
1205 | ||
1206 | retry_journal: | |
9924a92a | 1207 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks); |
af5bc92d | 1208 | if (IS_ERR(handle)) { |
09cbfeaf | 1209 | put_page(page); |
47564bfb | 1210 | return PTR_ERR(handle); |
7479d2b9 | 1211 | } |
ac27a0ec | 1212 | |
47564bfb TT |
1213 | lock_page(page); |
1214 | if (page->mapping != mapping) { | |
1215 | /* The page got truncated from under us */ | |
1216 | unlock_page(page); | |
09cbfeaf | 1217 | put_page(page); |
cf108bca | 1218 | ext4_journal_stop(handle); |
47564bfb | 1219 | goto retry_grab; |
cf108bca | 1220 | } |
7afe5aa5 DM |
1221 | /* In case writeback began while the page was unlocked */ |
1222 | wait_for_stable_page(page); | |
cf108bca | 1223 | |
643fa961 | 1224 | #ifdef CONFIG_FS_ENCRYPTION |
2058f83a MH |
1225 | if (ext4_should_dioread_nolock(inode)) |
1226 | ret = ext4_block_write_begin(page, pos, len, | |
705965bd | 1227 | ext4_get_block_unwritten); |
2058f83a MH |
1228 | else |
1229 | ret = ext4_block_write_begin(page, pos, len, | |
1230 | ext4_get_block); | |
1231 | #else | |
744692dc | 1232 | if (ext4_should_dioread_nolock(inode)) |
705965bd JK |
1233 | ret = __block_write_begin(page, pos, len, |
1234 | ext4_get_block_unwritten); | |
744692dc | 1235 | else |
6e1db88d | 1236 | ret = __block_write_begin(page, pos, len, ext4_get_block); |
2058f83a | 1237 | #endif |
bfc1af65 | 1238 | if (!ret && ext4_should_journal_data(inode)) { |
188c299e JK |
1239 | ret = ext4_walk_page_buffers(handle, inode, |
1240 | page_buffers(page), from, to, NULL, | |
f19d5870 | 1241 | do_journal_get_write_access); |
ac27a0ec | 1242 | } |
bfc1af65 NP |
1243 | |
1244 | if (ret) { | |
c93d8f88 EB |
1245 | bool extended = (pos + len > inode->i_size) && |
1246 | !ext4_verity_in_progress(inode); | |
1247 | ||
af5bc92d | 1248 | unlock_page(page); |
ae4d5372 | 1249 | /* |
6e1db88d | 1250 | * __block_write_begin may have instantiated a few blocks |
ae4d5372 | 1251 | * outside i_size. Trim these off again. Don't need |
f340b3d9 | 1252 | * i_size_read because we hold i_rwsem. |
1938a150 AK |
1253 | * |
1254 | * Add inode to orphan list in case we crash before | |
1255 | * truncate finishes | |
ae4d5372 | 1256 | */ |
c93d8f88 | 1257 | if (extended && ext4_can_truncate(inode)) |
1938a150 AK |
1258 | ext4_orphan_add(handle, inode); |
1259 | ||
1260 | ext4_journal_stop(handle); | |
c93d8f88 | 1261 | if (extended) { |
b9a4207d | 1262 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1263 | /* |
ffacfa7a | 1264 | * If truncate failed early the inode might |
1938a150 AK |
1265 | * still be on the orphan list; we need to |
1266 | * make sure the inode is removed from the | |
1267 | * orphan list in that case. | |
1268 | */ | |
1269 | if (inode->i_nlink) | |
1270 | ext4_orphan_del(NULL, inode); | |
1271 | } | |
bfc1af65 | 1272 | |
47564bfb TT |
1273 | if (ret == -ENOSPC && |
1274 | ext4_should_retry_alloc(inode->i_sb, &retries)) | |
1275 | goto retry_journal; | |
09cbfeaf | 1276 | put_page(page); |
47564bfb TT |
1277 | return ret; |
1278 | } | |
1279 | *pagep = page; | |
ac27a0ec DK |
1280 | return ret; |
1281 | } | |
1282 | ||
bfc1af65 | 1283 | /* For write_end() in data=journal mode */ |
188c299e JK |
1284 | static int write_end_fn(handle_t *handle, struct inode *inode, |
1285 | struct buffer_head *bh) | |
ac27a0ec | 1286 | { |
13fca323 | 1287 | int ret; |
ac27a0ec DK |
1288 | if (!buffer_mapped(bh) || buffer_freed(bh)) |
1289 | return 0; | |
1290 | set_buffer_uptodate(bh); | |
13fca323 TT |
1291 | ret = ext4_handle_dirty_metadata(handle, NULL, bh); |
1292 | clear_buffer_meta(bh); | |
1293 | clear_buffer_prio(bh); | |
1294 | return ret; | |
ac27a0ec DK |
1295 | } |
1296 | ||
eed4333f ZL |
1297 | /* |
1298 | * We need to pick up the new inode size which generic_commit_write gave us | |
1299 | * `file' can be NULL - eg, when called from page_symlink(). | |
1300 | * | |
1301 | * ext4 never places buffers on inode->i_mapping->private_list. metadata | |
1302 | * buffers are managed internally. | |
1303 | */ | |
1304 | static int ext4_write_end(struct file *file, | |
1305 | struct address_space *mapping, | |
1306 | loff_t pos, unsigned len, unsigned copied, | |
1307 | struct page *page, void *fsdata) | |
f8514083 | 1308 | { |
f8514083 | 1309 | handle_t *handle = ext4_journal_current_handle(); |
eed4333f | 1310 | struct inode *inode = mapping->host; |
0572639f | 1311 | loff_t old_size = inode->i_size; |
eed4333f ZL |
1312 | int ret = 0, ret2; |
1313 | int i_size_changed = 0; | |
c93d8f88 | 1314 | bool verity = ext4_verity_in_progress(inode); |
eed4333f ZL |
1315 | |
1316 | trace_ext4_write_end(inode, pos, len, copied); | |
6984aef5 ZY |
1317 | |
1318 | if (ext4_has_inline_data(inode)) | |
1319 | return ext4_write_inline_data_end(inode, pos, len, copied, page); | |
1320 | ||
1321 | copied = block_write_end(file, mapping, pos, len, copied, page, fsdata); | |
f8514083 | 1322 | /* |
4631dbf6 | 1323 | * it's important to update i_size while still holding page lock: |
f8514083 | 1324 | * page writeout could otherwise come in and zero beyond i_size. |
c93d8f88 EB |
1325 | * |
1326 | * If FS_IOC_ENABLE_VERITY is running on this inode, then Merkle tree | |
1327 | * blocks are being written past EOF, so skip the i_size update. | |
f8514083 | 1328 | */ |
c93d8f88 EB |
1329 | if (!verity) |
1330 | i_size_changed = ext4_update_inode_size(inode, pos + copied); | |
f8514083 | 1331 | unlock_page(page); |
09cbfeaf | 1332 | put_page(page); |
f8514083 | 1333 | |
c93d8f88 | 1334 | if (old_size < pos && !verity) |
0572639f | 1335 | pagecache_isize_extended(inode, old_size, pos); |
f8514083 AK |
1336 | /* |
1337 | * Don't mark the inode dirty under page lock. First, it unnecessarily | |
1338 | * makes the holding time of page lock longer. Second, it forces lock | |
1339 | * ordering of page lock and transaction start for journaling | |
1340 | * filesystems. | |
1341 | */ | |
6984aef5 | 1342 | if (i_size_changed) |
4209ae12 | 1343 | ret = ext4_mark_inode_dirty(handle, inode); |
f8514083 | 1344 | |
c93d8f88 | 1345 | if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode)) |
f8514083 AK |
1346 | /* if we have allocated more blocks and copied |
1347 | * less. We will have blocks allocated outside | |
1348 | * inode->i_size. So truncate them | |
1349 | */ | |
1350 | ext4_orphan_add(handle, inode); | |
55ce2f64 | 1351 | |
617ba13b | 1352 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1353 | if (!ret) |
1354 | ret = ret2; | |
bfc1af65 | 1355 | |
c93d8f88 | 1356 | if (pos + len > inode->i_size && !verity) { |
b9a4207d | 1357 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1358 | /* |
ffacfa7a | 1359 | * If truncate failed early the inode might still be |
f8514083 AK |
1360 | * on the orphan list; we need to make sure the inode |
1361 | * is removed from the orphan list in that case. | |
1362 | */ | |
1363 | if (inode->i_nlink) | |
1364 | ext4_orphan_del(NULL, inode); | |
1365 | } | |
1366 | ||
bfc1af65 | 1367 | return ret ? ret : copied; |
ac27a0ec DK |
1368 | } |
1369 | ||
b90197b6 TT |
1370 | /* |
1371 | * This is a private version of page_zero_new_buffers() which doesn't | |
1372 | * set the buffer to be dirty, since in data=journalled mode we need | |
1373 | * to call ext4_handle_dirty_metadata() instead. | |
1374 | */ | |
3b136499 | 1375 | static void ext4_journalled_zero_new_buffers(handle_t *handle, |
188c299e | 1376 | struct inode *inode, |
3b136499 JK |
1377 | struct page *page, |
1378 | unsigned from, unsigned to) | |
b90197b6 TT |
1379 | { |
1380 | unsigned int block_start = 0, block_end; | |
1381 | struct buffer_head *head, *bh; | |
1382 | ||
1383 | bh = head = page_buffers(page); | |
1384 | do { | |
1385 | block_end = block_start + bh->b_size; | |
1386 | if (buffer_new(bh)) { | |
1387 | if (block_end > from && block_start < to) { | |
1388 | if (!PageUptodate(page)) { | |
1389 | unsigned start, size; | |
1390 | ||
1391 | start = max(from, block_start); | |
1392 | size = min(to, block_end) - start; | |
1393 | ||
1394 | zero_user(page, start, size); | |
188c299e | 1395 | write_end_fn(handle, inode, bh); |
b90197b6 TT |
1396 | } |
1397 | clear_buffer_new(bh); | |
1398 | } | |
1399 | } | |
1400 | block_start = block_end; | |
1401 | bh = bh->b_this_page; | |
1402 | } while (bh != head); | |
1403 | } | |
1404 | ||
bfc1af65 | 1405 | static int ext4_journalled_write_end(struct file *file, |
de9a55b8 TT |
1406 | struct address_space *mapping, |
1407 | loff_t pos, unsigned len, unsigned copied, | |
1408 | struct page *page, void *fsdata) | |
ac27a0ec | 1409 | { |
617ba13b | 1410 | handle_t *handle = ext4_journal_current_handle(); |
bfc1af65 | 1411 | struct inode *inode = mapping->host; |
0572639f | 1412 | loff_t old_size = inode->i_size; |
ac27a0ec DK |
1413 | int ret = 0, ret2; |
1414 | int partial = 0; | |
bfc1af65 | 1415 | unsigned from, to; |
4631dbf6 | 1416 | int size_changed = 0; |
c93d8f88 | 1417 | bool verity = ext4_verity_in_progress(inode); |
ac27a0ec | 1418 | |
9bffad1e | 1419 | trace_ext4_journalled_write_end(inode, pos, len, copied); |
09cbfeaf | 1420 | from = pos & (PAGE_SIZE - 1); |
bfc1af65 NP |
1421 | to = from + len; |
1422 | ||
441c8508 CW |
1423 | BUG_ON(!ext4_handle_valid(handle)); |
1424 | ||
6984aef5 ZY |
1425 | if (ext4_has_inline_data(inode)) |
1426 | return ext4_write_inline_data_end(inode, pos, len, copied, page); | |
1427 | ||
1428 | if (unlikely(copied < len) && !PageUptodate(page)) { | |
3b136499 | 1429 | copied = 0; |
188c299e | 1430 | ext4_journalled_zero_new_buffers(handle, inode, page, from, to); |
3b136499 JK |
1431 | } else { |
1432 | if (unlikely(copied < len)) | |
188c299e | 1433 | ext4_journalled_zero_new_buffers(handle, inode, page, |
3b136499 | 1434 | from + copied, to); |
188c299e JK |
1435 | ret = ext4_walk_page_buffers(handle, inode, page_buffers(page), |
1436 | from, from + copied, &partial, | |
3b136499 | 1437 | write_end_fn); |
3fdcfb66 TM |
1438 | if (!partial) |
1439 | SetPageUptodate(page); | |
1440 | } | |
c93d8f88 EB |
1441 | if (!verity) |
1442 | size_changed = ext4_update_inode_size(inode, pos + copied); | |
19f5fb7a | 1443 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
2d859db3 | 1444 | EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; |
4631dbf6 | 1445 | unlock_page(page); |
09cbfeaf | 1446 | put_page(page); |
4631dbf6 | 1447 | |
c93d8f88 | 1448 | if (old_size < pos && !verity) |
0572639f XW |
1449 | pagecache_isize_extended(inode, old_size, pos); |
1450 | ||
6984aef5 | 1451 | if (size_changed) { |
617ba13b | 1452 | ret2 = ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
1453 | if (!ret) |
1454 | ret = ret2; | |
1455 | } | |
bfc1af65 | 1456 | |
c93d8f88 | 1457 | if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode)) |
f8514083 AK |
1458 | /* if we have allocated more blocks and copied |
1459 | * less. We will have blocks allocated outside | |
1460 | * inode->i_size. So truncate them | |
1461 | */ | |
1462 | ext4_orphan_add(handle, inode); | |
1463 | ||
617ba13b | 1464 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1465 | if (!ret) |
1466 | ret = ret2; | |
c93d8f88 | 1467 | if (pos + len > inode->i_size && !verity) { |
b9a4207d | 1468 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1469 | /* |
ffacfa7a | 1470 | * If truncate failed early the inode might still be |
f8514083 AK |
1471 | * on the orphan list; we need to make sure the inode |
1472 | * is removed from the orphan list in that case. | |
1473 | */ | |
1474 | if (inode->i_nlink) | |
1475 | ext4_orphan_del(NULL, inode); | |
1476 | } | |
bfc1af65 NP |
1477 | |
1478 | return ret ? ret : copied; | |
ac27a0ec | 1479 | } |
d2a17637 | 1480 | |
9d0be502 | 1481 | /* |
c27e43a1 | 1482 | * Reserve space for a single cluster |
9d0be502 | 1483 | */ |
c27e43a1 | 1484 | static int ext4_da_reserve_space(struct inode *inode) |
d2a17637 | 1485 | { |
60e58e0f | 1486 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
0637c6f4 | 1487 | struct ext4_inode_info *ei = EXT4_I(inode); |
5dd4056d | 1488 | int ret; |
03179fe9 TT |
1489 | |
1490 | /* | |
1491 | * We will charge metadata quota at writeout time; this saves | |
1492 | * us from metadata over-estimation, though we may go over by | |
1493 | * a small amount in the end. Here we just reserve for data. | |
1494 | */ | |
1495 | ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1)); | |
1496 | if (ret) | |
1497 | return ret; | |
d2a17637 | 1498 | |
0637c6f4 | 1499 | spin_lock(&ei->i_block_reservation_lock); |
71d4f7d0 | 1500 | if (ext4_claim_free_clusters(sbi, 1, 0)) { |
03179fe9 | 1501 | spin_unlock(&ei->i_block_reservation_lock); |
03179fe9 | 1502 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1)); |
d2a17637 MC |
1503 | return -ENOSPC; |
1504 | } | |
9d0be502 | 1505 | ei->i_reserved_data_blocks++; |
c27e43a1 | 1506 | trace_ext4_da_reserve_space(inode); |
0637c6f4 | 1507 | spin_unlock(&ei->i_block_reservation_lock); |
39bc680a | 1508 | |
d2a17637 MC |
1509 | return 0; /* success */ |
1510 | } | |
1511 | ||
f456767d | 1512 | void ext4_da_release_space(struct inode *inode, int to_free) |
d2a17637 MC |
1513 | { |
1514 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0637c6f4 | 1515 | struct ext4_inode_info *ei = EXT4_I(inode); |
d2a17637 | 1516 | |
cd213226 MC |
1517 | if (!to_free) |
1518 | return; /* Nothing to release, exit */ | |
1519 | ||
d2a17637 | 1520 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); |
cd213226 | 1521 | |
5a58ec87 | 1522 | trace_ext4_da_release_space(inode, to_free); |
0637c6f4 | 1523 | if (unlikely(to_free > ei->i_reserved_data_blocks)) { |
cd213226 | 1524 | /* |
0637c6f4 TT |
1525 | * if there aren't enough reserved blocks, then the |
1526 | * counter is messed up somewhere. Since this | |
1527 | * function is called from invalidate page, it's | |
1528 | * harmless to return without any action. | |
cd213226 | 1529 | */ |
8de5c325 | 1530 | ext4_warning(inode->i_sb, "ext4_da_release_space: " |
0637c6f4 | 1531 | "ino %lu, to_free %d with only %d reserved " |
1084f252 | 1532 | "data blocks", inode->i_ino, to_free, |
0637c6f4 TT |
1533 | ei->i_reserved_data_blocks); |
1534 | WARN_ON(1); | |
1535 | to_free = ei->i_reserved_data_blocks; | |
cd213226 | 1536 | } |
0637c6f4 | 1537 | ei->i_reserved_data_blocks -= to_free; |
cd213226 | 1538 | |
72b8ab9d | 1539 | /* update fs dirty data blocks counter */ |
57042651 | 1540 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free); |
d2a17637 | 1541 | |
d2a17637 | 1542 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
60e58e0f | 1543 | |
7b415bf6 | 1544 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free)); |
d2a17637 MC |
1545 | } |
1546 | ||
64769240 AT |
1547 | /* |
1548 | * Delayed allocation stuff | |
1549 | */ | |
1550 | ||
4e7ea81d JK |
1551 | struct mpage_da_data { |
1552 | struct inode *inode; | |
1553 | struct writeback_control *wbc; | |
6b523df4 | 1554 | |
4e7ea81d JK |
1555 | pgoff_t first_page; /* The first page to write */ |
1556 | pgoff_t next_page; /* Current page to examine */ | |
1557 | pgoff_t last_page; /* Last page to examine */ | |
791b7f08 | 1558 | /* |
4e7ea81d JK |
1559 | * Extent to map - this can be after first_page because that can be |
1560 | * fully mapped. We somewhat abuse m_flags to store whether the extent | |
1561 | * is delalloc or unwritten. | |
791b7f08 | 1562 | */ |
4e7ea81d JK |
1563 | struct ext4_map_blocks map; |
1564 | struct ext4_io_submit io_submit; /* IO submission data */ | |
dddbd6ac | 1565 | unsigned int do_map:1; |
6b8ed620 | 1566 | unsigned int scanned_until_end:1; |
4e7ea81d | 1567 | }; |
64769240 | 1568 | |
4e7ea81d JK |
1569 | static void mpage_release_unused_pages(struct mpage_da_data *mpd, |
1570 | bool invalidate) | |
c4a0c46e | 1571 | { |
fb5a5be0 | 1572 | unsigned nr, i; |
c4a0c46e | 1573 | pgoff_t index, end; |
fb5a5be0 | 1574 | struct folio_batch fbatch; |
c4a0c46e AK |
1575 | struct inode *inode = mpd->inode; |
1576 | struct address_space *mapping = inode->i_mapping; | |
4e7ea81d JK |
1577 | |
1578 | /* This is necessary when next_page == 0. */ | |
1579 | if (mpd->first_page >= mpd->next_page) | |
1580 | return; | |
c4a0c46e | 1581 | |
6b8ed620 | 1582 | mpd->scanned_until_end = 0; |
c7f5938a CW |
1583 | index = mpd->first_page; |
1584 | end = mpd->next_page - 1; | |
4e7ea81d JK |
1585 | if (invalidate) { |
1586 | ext4_lblk_t start, last; | |
09cbfeaf KS |
1587 | start = index << (PAGE_SHIFT - inode->i_blkbits); |
1588 | last = end << (PAGE_SHIFT - inode->i_blkbits); | |
7f0d8e1d EW |
1589 | |
1590 | /* | |
1591 | * avoid racing with extent status tree scans made by | |
1592 | * ext4_insert_delayed_block() | |
1593 | */ | |
1594 | down_write(&EXT4_I(inode)->i_data_sem); | |
4e7ea81d | 1595 | ext4_es_remove_extent(inode, start, last - start + 1); |
7f0d8e1d | 1596 | up_write(&EXT4_I(inode)->i_data_sem); |
4e7ea81d | 1597 | } |
51865fda | 1598 | |
fb5a5be0 | 1599 | folio_batch_init(&fbatch); |
c4a0c46e | 1600 | while (index <= end) { |
fb5a5be0 MWO |
1601 | nr = filemap_get_folios(mapping, &index, end, &fbatch); |
1602 | if (nr == 0) | |
c4a0c46e | 1603 | break; |
fb5a5be0 MWO |
1604 | for (i = 0; i < nr; i++) { |
1605 | struct folio *folio = fbatch.folios[i]; | |
2b85a617 | 1606 | |
fb5a5be0 MWO |
1607 | if (folio->index < mpd->first_page) |
1608 | continue; | |
1609 | if (folio->index + folio_nr_pages(folio) - 1 > end) | |
1610 | continue; | |
7ba13abb MWO |
1611 | BUG_ON(!folio_test_locked(folio)); |
1612 | BUG_ON(folio_test_writeback(folio)); | |
4e7ea81d | 1613 | if (invalidate) { |
7ba13abb MWO |
1614 | if (folio_mapped(folio)) |
1615 | folio_clear_dirty_for_io(folio); | |
1616 | block_invalidate_folio(folio, 0, | |
1617 | folio_size(folio)); | |
1618 | folio_clear_uptodate(folio); | |
4e7ea81d | 1619 | } |
7ba13abb | 1620 | folio_unlock(folio); |
c4a0c46e | 1621 | } |
fb5a5be0 | 1622 | folio_batch_release(&fbatch); |
c4a0c46e | 1623 | } |
c4a0c46e AK |
1624 | } |
1625 | ||
df22291f AK |
1626 | static void ext4_print_free_blocks(struct inode *inode) |
1627 | { | |
1628 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
92b97816 | 1629 | struct super_block *sb = inode->i_sb; |
f78ee70d | 1630 | struct ext4_inode_info *ei = EXT4_I(inode); |
92b97816 TT |
1631 | |
1632 | ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld", | |
5dee5437 | 1633 | EXT4_C2B(EXT4_SB(inode->i_sb), |
f78ee70d | 1634 | ext4_count_free_clusters(sb))); |
92b97816 TT |
1635 | ext4_msg(sb, KERN_CRIT, "Free/Dirty block details"); |
1636 | ext4_msg(sb, KERN_CRIT, "free_blocks=%lld", | |
f78ee70d | 1637 | (long long) EXT4_C2B(EXT4_SB(sb), |
57042651 | 1638 | percpu_counter_sum(&sbi->s_freeclusters_counter))); |
92b97816 | 1639 | ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld", |
f78ee70d | 1640 | (long long) EXT4_C2B(EXT4_SB(sb), |
7b415bf6 | 1641 | percpu_counter_sum(&sbi->s_dirtyclusters_counter))); |
92b97816 TT |
1642 | ext4_msg(sb, KERN_CRIT, "Block reservation details"); |
1643 | ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u", | |
f78ee70d | 1644 | ei->i_reserved_data_blocks); |
df22291f AK |
1645 | return; |
1646 | } | |
1647 | ||
188c299e JK |
1648 | static int ext4_bh_delay_or_unwritten(handle_t *handle, struct inode *inode, |
1649 | struct buffer_head *bh) | |
29fa89d0 | 1650 | { |
c364b22c | 1651 | return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh); |
29fa89d0 AK |
1652 | } |
1653 | ||
0b02f4c0 EW |
1654 | /* |
1655 | * ext4_insert_delayed_block - adds a delayed block to the extents status | |
1656 | * tree, incrementing the reserved cluster/block | |
1657 | * count or making a pending reservation | |
1658 | * where needed | |
1659 | * | |
1660 | * @inode - file containing the newly added block | |
1661 | * @lblk - logical block to be added | |
1662 | * | |
1663 | * Returns 0 on success, negative error code on failure. | |
1664 | */ | |
1665 | static int ext4_insert_delayed_block(struct inode *inode, ext4_lblk_t lblk) | |
1666 | { | |
1667 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
1668 | int ret; | |
1669 | bool allocated = false; | |
6fed8395 | 1670 | bool reserved = false; |
0b02f4c0 EW |
1671 | |
1672 | /* | |
1673 | * If the cluster containing lblk is shared with a delayed, | |
1674 | * written, or unwritten extent in a bigalloc file system, it's | |
1675 | * already been accounted for and does not need to be reserved. | |
1676 | * A pending reservation must be made for the cluster if it's | |
1677 | * shared with a written or unwritten extent and doesn't already | |
1678 | * have one. Written and unwritten extents can be purged from the | |
1679 | * extents status tree if the system is under memory pressure, so | |
1680 | * it's necessary to examine the extent tree if a search of the | |
1681 | * extents status tree doesn't get a match. | |
1682 | */ | |
1683 | if (sbi->s_cluster_ratio == 1) { | |
1684 | ret = ext4_da_reserve_space(inode); | |
1685 | if (ret != 0) /* ENOSPC */ | |
1686 | goto errout; | |
6fed8395 | 1687 | reserved = true; |
0b02f4c0 EW |
1688 | } else { /* bigalloc */ |
1689 | if (!ext4_es_scan_clu(inode, &ext4_es_is_delonly, lblk)) { | |
1690 | if (!ext4_es_scan_clu(inode, | |
1691 | &ext4_es_is_mapped, lblk)) { | |
1692 | ret = ext4_clu_mapped(inode, | |
1693 | EXT4_B2C(sbi, lblk)); | |
1694 | if (ret < 0) | |
1695 | goto errout; | |
1696 | if (ret == 0) { | |
1697 | ret = ext4_da_reserve_space(inode); | |
1698 | if (ret != 0) /* ENOSPC */ | |
1699 | goto errout; | |
6fed8395 | 1700 | reserved = true; |
0b02f4c0 EW |
1701 | } else { |
1702 | allocated = true; | |
1703 | } | |
1704 | } else { | |
1705 | allocated = true; | |
1706 | } | |
1707 | } | |
1708 | } | |
1709 | ||
1710 | ret = ext4_es_insert_delayed_block(inode, lblk, allocated); | |
6fed8395 JX |
1711 | if (ret && reserved) |
1712 | ext4_da_release_space(inode, 1); | |
0b02f4c0 EW |
1713 | |
1714 | errout: | |
1715 | return ret; | |
1716 | } | |
1717 | ||
5356f261 AK |
1718 | /* |
1719 | * This function is grabs code from the very beginning of | |
1720 | * ext4_map_blocks, but assumes that the caller is from delayed write | |
1721 | * time. This function looks up the requested blocks and sets the | |
1722 | * buffer delay bit under the protection of i_data_sem. | |
1723 | */ | |
1724 | static int ext4_da_map_blocks(struct inode *inode, sector_t iblock, | |
1725 | struct ext4_map_blocks *map, | |
1726 | struct buffer_head *bh) | |
1727 | { | |
d100eef2 | 1728 | struct extent_status es; |
5356f261 AK |
1729 | int retval; |
1730 | sector_t invalid_block = ~((sector_t) 0xffff); | |
921f266b DM |
1731 | #ifdef ES_AGGRESSIVE_TEST |
1732 | struct ext4_map_blocks orig_map; | |
1733 | ||
1734 | memcpy(&orig_map, map, sizeof(*map)); | |
1735 | #endif | |
5356f261 AK |
1736 | |
1737 | if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es)) | |
1738 | invalid_block = ~0; | |
1739 | ||
1740 | map->m_flags = 0; | |
70aa1554 | 1741 | ext_debug(inode, "max_blocks %u, logical block %lu\n", map->m_len, |
5356f261 | 1742 | (unsigned long) map->m_lblk); |
d100eef2 ZL |
1743 | |
1744 | /* Lookup extent status tree firstly */ | |
bb5835ed | 1745 | if (ext4_es_lookup_extent(inode, iblock, NULL, &es)) { |
d100eef2 ZL |
1746 | if (ext4_es_is_hole(&es)) { |
1747 | retval = 0; | |
c8b459f4 | 1748 | down_read(&EXT4_I(inode)->i_data_sem); |
d100eef2 ZL |
1749 | goto add_delayed; |
1750 | } | |
1751 | ||
1752 | /* | |
3eda41df EW |
1753 | * Delayed extent could be allocated by fallocate. |
1754 | * So we need to check it. | |
d100eef2 | 1755 | */ |
3eda41df EW |
1756 | if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) { |
1757 | map_bh(bh, inode->i_sb, invalid_block); | |
1758 | set_buffer_new(bh); | |
1759 | set_buffer_delay(bh); | |
d100eef2 ZL |
1760 | return 0; |
1761 | } | |
1762 | ||
1763 | map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk; | |
1764 | retval = es.es_len - (iblock - es.es_lblk); | |
1765 | if (retval > map->m_len) | |
1766 | retval = map->m_len; | |
1767 | map->m_len = retval; | |
1768 | if (ext4_es_is_written(&es)) | |
1769 | map->m_flags |= EXT4_MAP_MAPPED; | |
1770 | else if (ext4_es_is_unwritten(&es)) | |
1771 | map->m_flags |= EXT4_MAP_UNWRITTEN; | |
1772 | else | |
1e83bc81 | 1773 | BUG(); |
d100eef2 | 1774 | |
921f266b DM |
1775 | #ifdef ES_AGGRESSIVE_TEST |
1776 | ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0); | |
1777 | #endif | |
d100eef2 ZL |
1778 | return retval; |
1779 | } | |
1780 | ||
5356f261 AK |
1781 | /* |
1782 | * Try to see if we can get the block without requesting a new | |
1783 | * file system block. | |
1784 | */ | |
c8b459f4 | 1785 | down_read(&EXT4_I(inode)->i_data_sem); |
cbd7584e | 1786 | if (ext4_has_inline_data(inode)) |
9c3569b5 | 1787 | retval = 0; |
cbd7584e | 1788 | else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
2f8e0a7c | 1789 | retval = ext4_ext_map_blocks(NULL, inode, map, 0); |
5356f261 | 1790 | else |
2f8e0a7c | 1791 | retval = ext4_ind_map_blocks(NULL, inode, map, 0); |
5356f261 | 1792 | |
d100eef2 | 1793 | add_delayed: |
5356f261 | 1794 | if (retval == 0) { |
f7fec032 | 1795 | int ret; |
ad431025 | 1796 | |
5356f261 AK |
1797 | /* |
1798 | * XXX: __block_prepare_write() unmaps passed block, | |
1799 | * is it OK? | |
1800 | */ | |
5356f261 | 1801 | |
0b02f4c0 EW |
1802 | ret = ext4_insert_delayed_block(inode, map->m_lblk); |
1803 | if (ret != 0) { | |
f7fec032 | 1804 | retval = ret; |
51865fda | 1805 | goto out_unlock; |
f7fec032 | 1806 | } |
51865fda | 1807 | |
5356f261 AK |
1808 | map_bh(bh, inode->i_sb, invalid_block); |
1809 | set_buffer_new(bh); | |
1810 | set_buffer_delay(bh); | |
f7fec032 ZL |
1811 | } else if (retval > 0) { |
1812 | int ret; | |
3be78c73 | 1813 | unsigned int status; |
f7fec032 | 1814 | |
44fb851d ZL |
1815 | if (unlikely(retval != map->m_len)) { |
1816 | ext4_warning(inode->i_sb, | |
1817 | "ES len assertion failed for inode " | |
1818 | "%lu: retval %d != map->m_len %d", | |
1819 | inode->i_ino, retval, map->m_len); | |
1820 | WARN_ON(1); | |
921f266b | 1821 | } |
921f266b | 1822 | |
f7fec032 ZL |
1823 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
1824 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; | |
1825 | ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, | |
1826 | map->m_pblk, status); | |
1827 | if (ret != 0) | |
1828 | retval = ret; | |
5356f261 AK |
1829 | } |
1830 | ||
1831 | out_unlock: | |
1832 | up_read((&EXT4_I(inode)->i_data_sem)); | |
1833 | ||
1834 | return retval; | |
1835 | } | |
1836 | ||
64769240 | 1837 | /* |
d91bd2c1 | 1838 | * This is a special get_block_t callback which is used by |
b920c755 TT |
1839 | * ext4_da_write_begin(). It will either return mapped block or |
1840 | * reserve space for a single block. | |
29fa89d0 AK |
1841 | * |
1842 | * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set. | |
1843 | * We also have b_blocknr = -1 and b_bdev initialized properly | |
1844 | * | |
1845 | * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set. | |
1846 | * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev | |
1847 | * initialized properly. | |
64769240 | 1848 | */ |
9c3569b5 TM |
1849 | int ext4_da_get_block_prep(struct inode *inode, sector_t iblock, |
1850 | struct buffer_head *bh, int create) | |
64769240 | 1851 | { |
2ed88685 | 1852 | struct ext4_map_blocks map; |
64769240 AT |
1853 | int ret = 0; |
1854 | ||
1855 | BUG_ON(create == 0); | |
2ed88685 TT |
1856 | BUG_ON(bh->b_size != inode->i_sb->s_blocksize); |
1857 | ||
1858 | map.m_lblk = iblock; | |
1859 | map.m_len = 1; | |
64769240 AT |
1860 | |
1861 | /* | |
1862 | * first, we need to know whether the block is allocated already | |
1863 | * preallocated blocks are unmapped but should treated | |
1864 | * the same as allocated blocks. | |
1865 | */ | |
5356f261 AK |
1866 | ret = ext4_da_map_blocks(inode, iblock, &map, bh); |
1867 | if (ret <= 0) | |
2ed88685 | 1868 | return ret; |
64769240 | 1869 | |
2ed88685 | 1870 | map_bh(bh, inode->i_sb, map.m_pblk); |
ed8ad838 | 1871 | ext4_update_bh_state(bh, map.m_flags); |
2ed88685 TT |
1872 | |
1873 | if (buffer_unwritten(bh)) { | |
1874 | /* A delayed write to unwritten bh should be marked | |
1875 | * new and mapped. Mapped ensures that we don't do | |
1876 | * get_block multiple times when we write to the same | |
1877 | * offset and new ensures that we do proper zero out | |
1878 | * for partial write. | |
1879 | */ | |
1880 | set_buffer_new(bh); | |
c8205636 | 1881 | set_buffer_mapped(bh); |
2ed88685 TT |
1882 | } |
1883 | return 0; | |
64769240 | 1884 | } |
61628a3f | 1885 | |
62e086be | 1886 | static int __ext4_journalled_writepage(struct page *page, |
62e086be AK |
1887 | unsigned int len) |
1888 | { | |
1889 | struct address_space *mapping = page->mapping; | |
1890 | struct inode *inode = mapping->host; | |
62e086be | 1891 | handle_t *handle = NULL; |
3fdcfb66 TM |
1892 | int ret = 0, err = 0; |
1893 | int inline_data = ext4_has_inline_data(inode); | |
1894 | struct buffer_head *inode_bh = NULL; | |
5c48a7df | 1895 | loff_t size; |
62e086be | 1896 | |
cb20d518 | 1897 | ClearPageChecked(page); |
3fdcfb66 TM |
1898 | |
1899 | if (inline_data) { | |
1900 | BUG_ON(page->index != 0); | |
1901 | BUG_ON(len > ext4_get_max_inline_size(inode)); | |
1902 | inode_bh = ext4_journalled_write_inline_data(inode, len, page); | |
1903 | if (inode_bh == NULL) | |
1904 | goto out; | |
3fdcfb66 | 1905 | } |
bdf96838 TT |
1906 | /* |
1907 | * We need to release the page lock before we start the | |
1908 | * journal, so grab a reference so the page won't disappear | |
1909 | * out from under us. | |
1910 | */ | |
1911 | get_page(page); | |
62e086be AK |
1912 | unlock_page(page); |
1913 | ||
9924a92a TT |
1914 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
1915 | ext4_writepage_trans_blocks(inode)); | |
62e086be AK |
1916 | if (IS_ERR(handle)) { |
1917 | ret = PTR_ERR(handle); | |
bdf96838 TT |
1918 | put_page(page); |
1919 | goto out_no_pagelock; | |
62e086be | 1920 | } |
441c8508 CW |
1921 | BUG_ON(!ext4_handle_valid(handle)); |
1922 | ||
bdf96838 TT |
1923 | lock_page(page); |
1924 | put_page(page); | |
5c48a7df ZY |
1925 | size = i_size_read(inode); |
1926 | if (page->mapping != mapping || page_offset(page) > size) { | |
bdf96838 TT |
1927 | /* The page got truncated from under us */ |
1928 | ext4_journal_stop(handle); | |
1929 | ret = 0; | |
1930 | goto out; | |
1931 | } | |
1932 | ||
3fdcfb66 | 1933 | if (inline_data) { |
362eca70 | 1934 | ret = ext4_mark_inode_dirty(handle, inode); |
3fdcfb66 | 1935 | } else { |
5c48a7df ZY |
1936 | struct buffer_head *page_bufs = page_buffers(page); |
1937 | ||
1938 | if (page->index == size >> PAGE_SHIFT) | |
1939 | len = size & ~PAGE_MASK; | |
1940 | else | |
1941 | len = PAGE_SIZE; | |
1942 | ||
188c299e JK |
1943 | ret = ext4_walk_page_buffers(handle, inode, page_bufs, 0, len, |
1944 | NULL, do_journal_get_write_access); | |
3fdcfb66 | 1945 | |
188c299e JK |
1946 | err = ext4_walk_page_buffers(handle, inode, page_bufs, 0, len, |
1947 | NULL, write_end_fn); | |
3fdcfb66 | 1948 | } |
afb585a9 MFO |
1949 | if (ret == 0) |
1950 | ret = err; | |
b5b18160 | 1951 | err = ext4_jbd2_inode_add_write(handle, inode, page_offset(page), len); |
62e086be AK |
1952 | if (ret == 0) |
1953 | ret = err; | |
2d859db3 | 1954 | EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; |
62e086be AK |
1955 | err = ext4_journal_stop(handle); |
1956 | if (!ret) | |
1957 | ret = err; | |
1958 | ||
19f5fb7a | 1959 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
62e086be | 1960 | out: |
bdf96838 TT |
1961 | unlock_page(page); |
1962 | out_no_pagelock: | |
3fdcfb66 | 1963 | brelse(inode_bh); |
62e086be AK |
1964 | return ret; |
1965 | } | |
1966 | ||
61628a3f | 1967 | /* |
43ce1d23 AK |
1968 | * Note that we don't need to start a transaction unless we're journaling data |
1969 | * because we should have holes filled from ext4_page_mkwrite(). We even don't | |
1970 | * need to file the inode to the transaction's list in ordered mode because if | |
1971 | * we are writing back data added by write(), the inode is already there and if | |
25985edc | 1972 | * we are writing back data modified via mmap(), no one guarantees in which |
43ce1d23 AK |
1973 | * transaction the data will hit the disk. In case we are journaling data, we |
1974 | * cannot start transaction directly because transaction start ranks above page | |
1975 | * lock so we have to do some magic. | |
1976 | * | |
b920c755 | 1977 | * This function can get called via... |
20970ba6 | 1978 | * - ext4_writepages after taking page lock (have journal handle) |
b920c755 | 1979 | * - journal_submit_inode_data_buffers (no journal handle) |
f6463b0d | 1980 | * - shrink_page_list via the kswapd/direct reclaim (no journal handle) |
b920c755 | 1981 | * - grab_page_cache when doing write_begin (have journal handle) |
43ce1d23 AK |
1982 | * |
1983 | * We don't do any block allocation in this function. If we have page with | |
1984 | * multiple blocks we need to write those buffer_heads that are mapped. This | |
1985 | * is important for mmaped based write. So if we do with blocksize 1K | |
1986 | * truncate(f, 1024); | |
1987 | * a = mmap(f, 0, 4096); | |
1988 | * a[0] = 'a'; | |
1989 | * truncate(f, 4096); | |
1990 | * we have in the page first buffer_head mapped via page_mkwrite call back | |
90802ed9 | 1991 | * but other buffer_heads would be unmapped but dirty (dirty done via the |
43ce1d23 AK |
1992 | * do_wp_page). So writepage should write the first block. If we modify |
1993 | * the mmap area beyond 1024 we will again get a page_fault and the | |
1994 | * page_mkwrite callback will do the block allocation and mark the | |
1995 | * buffer_heads mapped. | |
1996 | * | |
1997 | * We redirty the page if we have any buffer_heads that is either delay or | |
1998 | * unwritten in the page. | |
1999 | * | |
2000 | * We can get recursively called as show below. | |
2001 | * | |
2002 | * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() -> | |
2003 | * ext4_writepage() | |
2004 | * | |
2005 | * But since we don't do any block allocation we should not deadlock. | |
2006 | * Page also have the dirty flag cleared so we don't get recurive page_lock. | |
61628a3f | 2007 | */ |
43ce1d23 | 2008 | static int ext4_writepage(struct page *page, |
62e086be | 2009 | struct writeback_control *wbc) |
64769240 | 2010 | { |
020df9ba | 2011 | struct folio *folio = page_folio(page); |
f8bec370 | 2012 | int ret = 0; |
61628a3f | 2013 | loff_t size; |
498e5f24 | 2014 | unsigned int len; |
744692dc | 2015 | struct buffer_head *page_bufs = NULL; |
61628a3f | 2016 | struct inode *inode = page->mapping->host; |
36ade451 | 2017 | struct ext4_io_submit io_submit; |
1c8349a1 | 2018 | bool keep_towrite = false; |
61628a3f | 2019 | |
0db1ff22 | 2020 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) { |
020df9ba MWO |
2021 | folio_invalidate(folio, 0, folio_size(folio)); |
2022 | folio_unlock(folio); | |
0db1ff22 TT |
2023 | return -EIO; |
2024 | } | |
2025 | ||
a9c667f8 | 2026 | trace_ext4_writepage(page); |
f0e6c985 | 2027 | size = i_size_read(inode); |
c93d8f88 EB |
2028 | if (page->index == size >> PAGE_SHIFT && |
2029 | !ext4_verity_in_progress(inode)) | |
09cbfeaf | 2030 | len = size & ~PAGE_MASK; |
f0e6c985 | 2031 | else |
09cbfeaf | 2032 | len = PAGE_SIZE; |
64769240 | 2033 | |
cc509574 TT |
2034 | /* Should never happen but for bugs in other kernel subsystems */ |
2035 | if (!page_has_buffers(page)) { | |
2036 | ext4_warning_inode(inode, | |
2037 | "page %lu does not have buffers attached", page->index); | |
2038 | ClearPageDirty(page); | |
2039 | unlock_page(page); | |
2040 | return 0; | |
2041 | } | |
2042 | ||
a42afc5f | 2043 | page_bufs = page_buffers(page); |
a42afc5f | 2044 | /* |
fe386132 JK |
2045 | * We cannot do block allocation or other extent handling in this |
2046 | * function. If there are buffers needing that, we have to redirty | |
2047 | * the page. But we may reach here when we do a journal commit via | |
2048 | * journal_submit_inode_data_buffers() and in that case we must write | |
2049 | * allocated buffers to achieve data=ordered mode guarantees. | |
cccd147a TT |
2050 | * |
2051 | * Also, if there is only one buffer per page (the fs block | |
2052 | * size == the page size), if one buffer needs block | |
2053 | * allocation or needs to modify the extent tree to clear the | |
2054 | * unwritten flag, we know that the page can't be written at | |
2055 | * all, so we might as well refuse the write immediately. | |
2056 | * Unfortunately if the block size != page size, we can't as | |
2057 | * easily detect this case using ext4_walk_page_buffers(), but | |
2058 | * for the extremely common case, this is an optimization that | |
2059 | * skips a useless round trip through ext4_bio_write_page(). | |
a42afc5f | 2060 | */ |
188c299e | 2061 | if (ext4_walk_page_buffers(NULL, inode, page_bufs, 0, len, NULL, |
f19d5870 | 2062 | ext4_bh_delay_or_unwritten)) { |
f8bec370 | 2063 | redirty_page_for_writepage(wbc, page); |
cccd147a | 2064 | if ((current->flags & PF_MEMALLOC) || |
09cbfeaf | 2065 | (inode->i_sb->s_blocksize == PAGE_SIZE)) { |
fe386132 JK |
2066 | /* |
2067 | * For memory cleaning there's no point in writing only | |
2068 | * some buffers. So just bail out. Warn if we came here | |
2069 | * from direct reclaim. | |
2070 | */ | |
2071 | WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) | |
2072 | == PF_MEMALLOC); | |
f0e6c985 AK |
2073 | unlock_page(page); |
2074 | return 0; | |
2075 | } | |
1c8349a1 | 2076 | keep_towrite = true; |
a42afc5f | 2077 | } |
64769240 | 2078 | |
cb20d518 | 2079 | if (PageChecked(page) && ext4_should_journal_data(inode)) |
43ce1d23 AK |
2080 | /* |
2081 | * It's mmapped pagecache. Add buffers and journal it. There | |
2082 | * doesn't seem much point in redirtying the page here. | |
2083 | */ | |
3f0ca309 | 2084 | return __ext4_journalled_writepage(page, len); |
43ce1d23 | 2085 | |
97a851ed JK |
2086 | ext4_io_submit_init(&io_submit, wbc); |
2087 | io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS); | |
2088 | if (!io_submit.io_end) { | |
2089 | redirty_page_for_writepage(wbc, page); | |
2090 | unlock_page(page); | |
2091 | return -ENOMEM; | |
2092 | } | |
be993933 | 2093 | ret = ext4_bio_write_page(&io_submit, page, len, keep_towrite); |
36ade451 | 2094 | ext4_io_submit(&io_submit); |
97a851ed JK |
2095 | /* Drop io_end reference we got from init */ |
2096 | ext4_put_io_end_defer(io_submit.io_end); | |
64769240 AT |
2097 | return ret; |
2098 | } | |
2099 | ||
5f1132b2 JK |
2100 | static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page) |
2101 | { | |
2102 | int len; | |
a056bdaa | 2103 | loff_t size; |
5f1132b2 JK |
2104 | int err; |
2105 | ||
2106 | BUG_ON(page->index != mpd->first_page); | |
a056bdaa JK |
2107 | clear_page_dirty_for_io(page); |
2108 | /* | |
2109 | * We have to be very careful here! Nothing protects writeback path | |
2110 | * against i_size changes and the page can be writeably mapped into | |
2111 | * page tables. So an application can be growing i_size and writing | |
2112 | * data through mmap while writeback runs. clear_page_dirty_for_io() | |
2113 | * write-protects our page in page tables and the page cannot get | |
2114 | * written to again until we release page lock. So only after | |
2115 | * clear_page_dirty_for_io() we are safe to sample i_size for | |
2116 | * ext4_bio_write_page() to zero-out tail of the written page. We rely | |
2117 | * on the barrier provided by TestClearPageDirty in | |
2118 | * clear_page_dirty_for_io() to make sure i_size is really sampled only | |
2119 | * after page tables are updated. | |
2120 | */ | |
2121 | size = i_size_read(mpd->inode); | |
c93d8f88 EB |
2122 | if (page->index == size >> PAGE_SHIFT && |
2123 | !ext4_verity_in_progress(mpd->inode)) | |
09cbfeaf | 2124 | len = size & ~PAGE_MASK; |
5f1132b2 | 2125 | else |
09cbfeaf | 2126 | len = PAGE_SIZE; |
be993933 | 2127 | err = ext4_bio_write_page(&mpd->io_submit, page, len, false); |
5f1132b2 JK |
2128 | if (!err) |
2129 | mpd->wbc->nr_to_write--; | |
2130 | mpd->first_page++; | |
2131 | ||
2132 | return err; | |
2133 | } | |
2134 | ||
6db07461 | 2135 | #define BH_FLAGS (BIT(BH_Unwritten) | BIT(BH_Delay)) |
4e7ea81d | 2136 | |
61628a3f | 2137 | /* |
fffb2739 JK |
2138 | * mballoc gives us at most this number of blocks... |
2139 | * XXX: That seems to be only a limitation of ext4_mb_normalize_request(). | |
70261f56 | 2140 | * The rest of mballoc seems to handle chunks up to full group size. |
61628a3f | 2141 | */ |
fffb2739 | 2142 | #define MAX_WRITEPAGES_EXTENT_LEN 2048 |
525f4ed8 | 2143 | |
4e7ea81d JK |
2144 | /* |
2145 | * mpage_add_bh_to_extent - try to add bh to extent of blocks to map | |
2146 | * | |
2147 | * @mpd - extent of blocks | |
2148 | * @lblk - logical number of the block in the file | |
09930042 | 2149 | * @bh - buffer head we want to add to the extent |
4e7ea81d | 2150 | * |
09930042 JK |
2151 | * The function is used to collect contig. blocks in the same state. If the |
2152 | * buffer doesn't require mapping for writeback and we haven't started the | |
2153 | * extent of buffers to map yet, the function returns 'true' immediately - the | |
2154 | * caller can write the buffer right away. Otherwise the function returns true | |
2155 | * if the block has been added to the extent, false if the block couldn't be | |
2156 | * added. | |
4e7ea81d | 2157 | */ |
09930042 JK |
2158 | static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk, |
2159 | struct buffer_head *bh) | |
4e7ea81d JK |
2160 | { |
2161 | struct ext4_map_blocks *map = &mpd->map; | |
2162 | ||
09930042 JK |
2163 | /* Buffer that doesn't need mapping for writeback? */ |
2164 | if (!buffer_dirty(bh) || !buffer_mapped(bh) || | |
2165 | (!buffer_delay(bh) && !buffer_unwritten(bh))) { | |
2166 | /* So far no extent to map => we write the buffer right away */ | |
2167 | if (map->m_len == 0) | |
2168 | return true; | |
2169 | return false; | |
2170 | } | |
4e7ea81d JK |
2171 | |
2172 | /* First block in the extent? */ | |
2173 | if (map->m_len == 0) { | |
dddbd6ac JK |
2174 | /* We cannot map unless handle is started... */ |
2175 | if (!mpd->do_map) | |
2176 | return false; | |
4e7ea81d JK |
2177 | map->m_lblk = lblk; |
2178 | map->m_len = 1; | |
09930042 JK |
2179 | map->m_flags = bh->b_state & BH_FLAGS; |
2180 | return true; | |
4e7ea81d JK |
2181 | } |
2182 | ||
09930042 JK |
2183 | /* Don't go larger than mballoc is willing to allocate */ |
2184 | if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN) | |
2185 | return false; | |
2186 | ||
4e7ea81d JK |
2187 | /* Can we merge the block to our big extent? */ |
2188 | if (lblk == map->m_lblk + map->m_len && | |
09930042 | 2189 | (bh->b_state & BH_FLAGS) == map->m_flags) { |
4e7ea81d | 2190 | map->m_len++; |
09930042 | 2191 | return true; |
4e7ea81d | 2192 | } |
09930042 | 2193 | return false; |
4e7ea81d JK |
2194 | } |
2195 | ||
5f1132b2 JK |
2196 | /* |
2197 | * mpage_process_page_bufs - submit page buffers for IO or add them to extent | |
2198 | * | |
2199 | * @mpd - extent of blocks for mapping | |
2200 | * @head - the first buffer in the page | |
2201 | * @bh - buffer we should start processing from | |
2202 | * @lblk - logical number of the block in the file corresponding to @bh | |
2203 | * | |
2204 | * Walk through page buffers from @bh upto @head (exclusive) and either submit | |
2205 | * the page for IO if all buffers in this page were mapped and there's no | |
2206 | * accumulated extent of buffers to map or add buffers in the page to the | |
2207 | * extent of buffers to map. The function returns 1 if the caller can continue | |
2208 | * by processing the next page, 0 if it should stop adding buffers to the | |
2209 | * extent to map because we cannot extend it anymore. It can also return value | |
2210 | * < 0 in case of error during IO submission. | |
2211 | */ | |
2212 | static int mpage_process_page_bufs(struct mpage_da_data *mpd, | |
2213 | struct buffer_head *head, | |
2214 | struct buffer_head *bh, | |
2215 | ext4_lblk_t lblk) | |
4e7ea81d JK |
2216 | { |
2217 | struct inode *inode = mpd->inode; | |
5f1132b2 | 2218 | int err; |
93407472 | 2219 | ext4_lblk_t blocks = (i_size_read(inode) + i_blocksize(inode) - 1) |
4e7ea81d JK |
2220 | >> inode->i_blkbits; |
2221 | ||
c93d8f88 EB |
2222 | if (ext4_verity_in_progress(inode)) |
2223 | blocks = EXT_MAX_BLOCKS; | |
2224 | ||
4e7ea81d JK |
2225 | do { |
2226 | BUG_ON(buffer_locked(bh)); | |
2227 | ||
09930042 | 2228 | if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) { |
4e7ea81d JK |
2229 | /* Found extent to map? */ |
2230 | if (mpd->map.m_len) | |
5f1132b2 | 2231 | return 0; |
dddbd6ac JK |
2232 | /* Buffer needs mapping and handle is not started? */ |
2233 | if (!mpd->do_map) | |
2234 | return 0; | |
09930042 | 2235 | /* Everything mapped so far and we hit EOF */ |
5f1132b2 | 2236 | break; |
4e7ea81d | 2237 | } |
4e7ea81d | 2238 | } while (lblk++, (bh = bh->b_this_page) != head); |
5f1132b2 JK |
2239 | /* So far everything mapped? Submit the page for IO. */ |
2240 | if (mpd->map.m_len == 0) { | |
2241 | err = mpage_submit_page(mpd, head->b_page); | |
2242 | if (err < 0) | |
2243 | return err; | |
2244 | } | |
6b8ed620 JK |
2245 | if (lblk >= blocks) { |
2246 | mpd->scanned_until_end = 1; | |
2247 | return 0; | |
2248 | } | |
2249 | return 1; | |
4e7ea81d JK |
2250 | } |
2251 | ||
2943fdbc RH |
2252 | /* |
2253 | * mpage_process_page - update page buffers corresponding to changed extent and | |
2254 | * may submit fully mapped page for IO | |
2255 | * | |
2256 | * @mpd - description of extent to map, on return next extent to map | |
2257 | * @m_lblk - logical block mapping. | |
2258 | * @m_pblk - corresponding physical mapping. | |
2259 | * @map_bh - determines on return whether this page requires any further | |
2260 | * mapping or not. | |
2261 | * Scan given page buffers corresponding to changed extent and update buffer | |
2262 | * state according to new extent state. | |
2263 | * We map delalloc buffers to their physical location, clear unwritten bits. | |
2264 | * If the given page is not fully mapped, we update @map to the next extent in | |
2265 | * the given page that needs mapping & return @map_bh as true. | |
2266 | */ | |
2267 | static int mpage_process_page(struct mpage_da_data *mpd, struct page *page, | |
2268 | ext4_lblk_t *m_lblk, ext4_fsblk_t *m_pblk, | |
2269 | bool *map_bh) | |
2270 | { | |
2271 | struct buffer_head *head, *bh; | |
2272 | ext4_io_end_t *io_end = mpd->io_submit.io_end; | |
2273 | ext4_lblk_t lblk = *m_lblk; | |
2274 | ext4_fsblk_t pblock = *m_pblk; | |
2275 | int err = 0; | |
c8cc8816 RH |
2276 | int blkbits = mpd->inode->i_blkbits; |
2277 | ssize_t io_end_size = 0; | |
2278 | struct ext4_io_end_vec *io_end_vec = ext4_last_io_end_vec(io_end); | |
2943fdbc RH |
2279 | |
2280 | bh = head = page_buffers(page); | |
2281 | do { | |
2282 | if (lblk < mpd->map.m_lblk) | |
2283 | continue; | |
2284 | if (lblk >= mpd->map.m_lblk + mpd->map.m_len) { | |
2285 | /* | |
2286 | * Buffer after end of mapped extent. | |
2287 | * Find next buffer in the page to map. | |
2288 | */ | |
2289 | mpd->map.m_len = 0; | |
2290 | mpd->map.m_flags = 0; | |
c8cc8816 | 2291 | io_end_vec->size += io_end_size; |
2943fdbc | 2292 | |
2943fdbc RH |
2293 | err = mpage_process_page_bufs(mpd, head, bh, lblk); |
2294 | if (err > 0) | |
2295 | err = 0; | |
c8cc8816 RH |
2296 | if (!err && mpd->map.m_len && mpd->map.m_lblk > lblk) { |
2297 | io_end_vec = ext4_alloc_io_end_vec(io_end); | |
4d06bfb9 RH |
2298 | if (IS_ERR(io_end_vec)) { |
2299 | err = PTR_ERR(io_end_vec); | |
2300 | goto out; | |
2301 | } | |
d1e18b88 | 2302 | io_end_vec->offset = (loff_t)mpd->map.m_lblk << blkbits; |
c8cc8816 | 2303 | } |
2943fdbc RH |
2304 | *map_bh = true; |
2305 | goto out; | |
2306 | } | |
2307 | if (buffer_delay(bh)) { | |
2308 | clear_buffer_delay(bh); | |
2309 | bh->b_blocknr = pblock++; | |
2310 | } | |
2311 | clear_buffer_unwritten(bh); | |
c8cc8816 | 2312 | io_end_size += (1 << blkbits); |
2943fdbc | 2313 | } while (lblk++, (bh = bh->b_this_page) != head); |
c8cc8816 RH |
2314 | |
2315 | io_end_vec->size += io_end_size; | |
2943fdbc RH |
2316 | *map_bh = false; |
2317 | out: | |
2318 | *m_lblk = lblk; | |
2319 | *m_pblk = pblock; | |
2320 | return err; | |
2321 | } | |
2322 | ||
4e7ea81d JK |
2323 | /* |
2324 | * mpage_map_buffers - update buffers corresponding to changed extent and | |
2325 | * submit fully mapped pages for IO | |
2326 | * | |
2327 | * @mpd - description of extent to map, on return next extent to map | |
2328 | * | |
2329 | * Scan buffers corresponding to changed extent (we expect corresponding pages | |
2330 | * to be already locked) and update buffer state according to new extent state. | |
2331 | * We map delalloc buffers to their physical location, clear unwritten bits, | |
556615dc | 2332 | * and mark buffers as uninit when we perform writes to unwritten extents |
4e7ea81d JK |
2333 | * and do extent conversion after IO is finished. If the last page is not fully |
2334 | * mapped, we update @map to the next extent in the last page that needs | |
2335 | * mapping. Otherwise we submit the page for IO. | |
2336 | */ | |
2337 | static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd) | |
2338 | { | |
7530d093 MWO |
2339 | struct folio_batch fbatch; |
2340 | unsigned nr, i; | |
4e7ea81d | 2341 | struct inode *inode = mpd->inode; |
09cbfeaf | 2342 | int bpp_bits = PAGE_SHIFT - inode->i_blkbits; |
4e7ea81d JK |
2343 | pgoff_t start, end; |
2344 | ext4_lblk_t lblk; | |
2943fdbc | 2345 | ext4_fsblk_t pblock; |
4e7ea81d | 2346 | int err; |
2943fdbc | 2347 | bool map_bh = false; |
4e7ea81d JK |
2348 | |
2349 | start = mpd->map.m_lblk >> bpp_bits; | |
2350 | end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits; | |
2351 | lblk = start << bpp_bits; | |
2352 | pblock = mpd->map.m_pblk; | |
2353 | ||
7530d093 | 2354 | folio_batch_init(&fbatch); |
4e7ea81d | 2355 | while (start <= end) { |
7530d093 MWO |
2356 | nr = filemap_get_folios(inode->i_mapping, &start, end, &fbatch); |
2357 | if (nr == 0) | |
4e7ea81d | 2358 | break; |
7530d093 MWO |
2359 | for (i = 0; i < nr; i++) { |
2360 | struct page *page = &fbatch.folios[i]->page; | |
4e7ea81d | 2361 | |
2943fdbc RH |
2362 | err = mpage_process_page(mpd, page, &lblk, &pblock, |
2363 | &map_bh); | |
4e7ea81d | 2364 | /* |
2943fdbc RH |
2365 | * If map_bh is true, means page may require further bh |
2366 | * mapping, or maybe the page was submitted for IO. | |
2367 | * So we return to call further extent mapping. | |
4e7ea81d | 2368 | */ |
39c0ae16 | 2369 | if (err < 0 || map_bh) |
2943fdbc | 2370 | goto out; |
4e7ea81d JK |
2371 | /* Page fully mapped - let IO run! */ |
2372 | err = mpage_submit_page(mpd, page); | |
2943fdbc RH |
2373 | if (err < 0) |
2374 | goto out; | |
4e7ea81d | 2375 | } |
7530d093 | 2376 | folio_batch_release(&fbatch); |
4e7ea81d JK |
2377 | } |
2378 | /* Extent fully mapped and matches with page boundary. We are done. */ | |
2379 | mpd->map.m_len = 0; | |
2380 | mpd->map.m_flags = 0; | |
2381 | return 0; | |
2943fdbc | 2382 | out: |
7530d093 | 2383 | folio_batch_release(&fbatch); |
2943fdbc | 2384 | return err; |
4e7ea81d JK |
2385 | } |
2386 | ||
2387 | static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd) | |
2388 | { | |
2389 | struct inode *inode = mpd->inode; | |
2390 | struct ext4_map_blocks *map = &mpd->map; | |
2391 | int get_blocks_flags; | |
090f32ee | 2392 | int err, dioread_nolock; |
4e7ea81d JK |
2393 | |
2394 | trace_ext4_da_write_pages_extent(inode, map); | |
2395 | /* | |
2396 | * Call ext4_map_blocks() to allocate any delayed allocation blocks, or | |
556615dc | 2397 | * to convert an unwritten extent to be initialized (in the case |
4e7ea81d JK |
2398 | * where we have written into one or more preallocated blocks). It is |
2399 | * possible that we're going to need more metadata blocks than | |
2400 | * previously reserved. However we must not fail because we're in | |
2401 | * writeback and there is nothing we can do about it so it might result | |
2402 | * in data loss. So use reserved blocks to allocate metadata if | |
2403 | * possible. | |
2404 | * | |
754cfed6 TT |
2405 | * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if |
2406 | * the blocks in question are delalloc blocks. This indicates | |
2407 | * that the blocks and quotas has already been checked when | |
2408 | * the data was copied into the page cache. | |
4e7ea81d JK |
2409 | */ |
2410 | get_blocks_flags = EXT4_GET_BLOCKS_CREATE | | |
ee0876bc JK |
2411 | EXT4_GET_BLOCKS_METADATA_NOFAIL | |
2412 | EXT4_GET_BLOCKS_IO_SUBMIT; | |
090f32ee LC |
2413 | dioread_nolock = ext4_should_dioread_nolock(inode); |
2414 | if (dioread_nolock) | |
4e7ea81d | 2415 | get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT; |
6db07461 | 2416 | if (map->m_flags & BIT(BH_Delay)) |
4e7ea81d JK |
2417 | get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE; |
2418 | ||
2419 | err = ext4_map_blocks(handle, inode, map, get_blocks_flags); | |
2420 | if (err < 0) | |
2421 | return err; | |
090f32ee | 2422 | if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) { |
6b523df4 JK |
2423 | if (!mpd->io_submit.io_end->handle && |
2424 | ext4_handle_valid(handle)) { | |
2425 | mpd->io_submit.io_end->handle = handle->h_rsv_handle; | |
2426 | handle->h_rsv_handle = NULL; | |
2427 | } | |
3613d228 | 2428 | ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end); |
6b523df4 | 2429 | } |
4e7ea81d JK |
2430 | |
2431 | BUG_ON(map->m_len == 0); | |
4e7ea81d JK |
2432 | return 0; |
2433 | } | |
2434 | ||
2435 | /* | |
2436 | * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length | |
2437 | * mpd->len and submit pages underlying it for IO | |
2438 | * | |
2439 | * @handle - handle for journal operations | |
2440 | * @mpd - extent to map | |
7534e854 JK |
2441 | * @give_up_on_write - we set this to true iff there is a fatal error and there |
2442 | * is no hope of writing the data. The caller should discard | |
2443 | * dirty pages to avoid infinite loops. | |
4e7ea81d JK |
2444 | * |
2445 | * The function maps extent starting at mpd->lblk of length mpd->len. If it is | |
2446 | * delayed, blocks are allocated, if it is unwritten, we may need to convert | |
2447 | * them to initialized or split the described range from larger unwritten | |
2448 | * extent. Note that we need not map all the described range since allocation | |
2449 | * can return less blocks or the range is covered by more unwritten extents. We | |
2450 | * cannot map more because we are limited by reserved transaction credits. On | |
2451 | * the other hand we always make sure that the last touched page is fully | |
2452 | * mapped so that it can be written out (and thus forward progress is | |
2453 | * guaranteed). After mapping we submit all mapped pages for IO. | |
2454 | */ | |
2455 | static int mpage_map_and_submit_extent(handle_t *handle, | |
cb530541 TT |
2456 | struct mpage_da_data *mpd, |
2457 | bool *give_up_on_write) | |
4e7ea81d JK |
2458 | { |
2459 | struct inode *inode = mpd->inode; | |
2460 | struct ext4_map_blocks *map = &mpd->map; | |
2461 | int err; | |
2462 | loff_t disksize; | |
6603120e | 2463 | int progress = 0; |
c8cc8816 | 2464 | ext4_io_end_t *io_end = mpd->io_submit.io_end; |
4d06bfb9 | 2465 | struct ext4_io_end_vec *io_end_vec; |
4e7ea81d | 2466 | |
4d06bfb9 RH |
2467 | io_end_vec = ext4_alloc_io_end_vec(io_end); |
2468 | if (IS_ERR(io_end_vec)) | |
2469 | return PTR_ERR(io_end_vec); | |
c8cc8816 | 2470 | io_end_vec->offset = ((loff_t)map->m_lblk) << inode->i_blkbits; |
27d7c4ed | 2471 | do { |
4e7ea81d JK |
2472 | err = mpage_map_one_extent(handle, mpd); |
2473 | if (err < 0) { | |
2474 | struct super_block *sb = inode->i_sb; | |
2475 | ||
0db1ff22 | 2476 | if (ext4_forced_shutdown(EXT4_SB(sb)) || |
9b5f6c9b | 2477 | ext4_test_mount_flag(sb, EXT4_MF_FS_ABORTED)) |
cb530541 | 2478 | goto invalidate_dirty_pages; |
4e7ea81d | 2479 | /* |
cb530541 TT |
2480 | * Let the uper layers retry transient errors. |
2481 | * In the case of ENOSPC, if ext4_count_free_blocks() | |
2482 | * is non-zero, a commit should free up blocks. | |
4e7ea81d | 2483 | */ |
cb530541 | 2484 | if ((err == -ENOMEM) || |
6603120e DM |
2485 | (err == -ENOSPC && ext4_count_free_clusters(sb))) { |
2486 | if (progress) | |
2487 | goto update_disksize; | |
cb530541 | 2488 | return err; |
6603120e | 2489 | } |
cb530541 TT |
2490 | ext4_msg(sb, KERN_CRIT, |
2491 | "Delayed block allocation failed for " | |
2492 | "inode %lu at logical offset %llu with" | |
2493 | " max blocks %u with error %d", | |
2494 | inode->i_ino, | |
2495 | (unsigned long long)map->m_lblk, | |
2496 | (unsigned)map->m_len, -err); | |
2497 | ext4_msg(sb, KERN_CRIT, | |
2498 | "This should not happen!! Data will " | |
2499 | "be lost\n"); | |
2500 | if (err == -ENOSPC) | |
2501 | ext4_print_free_blocks(inode); | |
2502 | invalidate_dirty_pages: | |
2503 | *give_up_on_write = true; | |
4e7ea81d JK |
2504 | return err; |
2505 | } | |
6603120e | 2506 | progress = 1; |
4e7ea81d JK |
2507 | /* |
2508 | * Update buffer state, submit mapped pages, and get us new | |
2509 | * extent to map | |
2510 | */ | |
2511 | err = mpage_map_and_submit_buffers(mpd); | |
2512 | if (err < 0) | |
6603120e | 2513 | goto update_disksize; |
27d7c4ed | 2514 | } while (map->m_len); |
4e7ea81d | 2515 | |
6603120e | 2516 | update_disksize: |
622cad13 TT |
2517 | /* |
2518 | * Update on-disk size after IO is submitted. Races with | |
2519 | * truncate are avoided by checking i_size under i_data_sem. | |
2520 | */ | |
09cbfeaf | 2521 | disksize = ((loff_t)mpd->first_page) << PAGE_SHIFT; |
35df4299 | 2522 | if (disksize > READ_ONCE(EXT4_I(inode)->i_disksize)) { |
4e7ea81d | 2523 | int err2; |
622cad13 TT |
2524 | loff_t i_size; |
2525 | ||
2526 | down_write(&EXT4_I(inode)->i_data_sem); | |
2527 | i_size = i_size_read(inode); | |
2528 | if (disksize > i_size) | |
2529 | disksize = i_size; | |
2530 | if (disksize > EXT4_I(inode)->i_disksize) | |
2531 | EXT4_I(inode)->i_disksize = disksize; | |
622cad13 | 2532 | up_write(&EXT4_I(inode)->i_data_sem); |
b907f2d5 | 2533 | err2 = ext4_mark_inode_dirty(handle, inode); |
878520ac | 2534 | if (err2) { |
54d3adbc TT |
2535 | ext4_error_err(inode->i_sb, -err2, |
2536 | "Failed to mark inode %lu dirty", | |
2537 | inode->i_ino); | |
878520ac | 2538 | } |
4e7ea81d JK |
2539 | if (!err) |
2540 | err = err2; | |
2541 | } | |
2542 | return err; | |
2543 | } | |
2544 | ||
fffb2739 JK |
2545 | /* |
2546 | * Calculate the total number of credits to reserve for one writepages | |
20970ba6 | 2547 | * iteration. This is called from ext4_writepages(). We map an extent of |
70261f56 | 2548 | * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping |
fffb2739 JK |
2549 | * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN + |
2550 | * bpp - 1 blocks in bpp different extents. | |
2551 | */ | |
525f4ed8 MC |
2552 | static int ext4_da_writepages_trans_blocks(struct inode *inode) |
2553 | { | |
fffb2739 | 2554 | int bpp = ext4_journal_blocks_per_page(inode); |
525f4ed8 | 2555 | |
fffb2739 JK |
2556 | return ext4_meta_trans_blocks(inode, |
2557 | MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp); | |
525f4ed8 | 2558 | } |
61628a3f | 2559 | |
8e48dcfb | 2560 | /* |
4e7ea81d JK |
2561 | * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages |
2562 | * and underlying extent to map | |
2563 | * | |
2564 | * @mpd - where to look for pages | |
2565 | * | |
2566 | * Walk dirty pages in the mapping. If they are fully mapped, submit them for | |
2567 | * IO immediately. When we find a page which isn't mapped we start accumulating | |
2568 | * extent of buffers underlying these pages that needs mapping (formed by | |
2569 | * either delayed or unwritten buffers). We also lock the pages containing | |
2570 | * these buffers. The extent found is returned in @mpd structure (starting at | |
2571 | * mpd->lblk with length mpd->len blocks). | |
2572 | * | |
2573 | * Note that this function can attach bios to one io_end structure which are | |
2574 | * neither logically nor physically contiguous. Although it may seem as an | |
2575 | * unnecessary complication, it is actually inevitable in blocksize < pagesize | |
2576 | * case as we need to track IO to all buffers underlying a page in one io_end. | |
8e48dcfb | 2577 | */ |
4e7ea81d | 2578 | static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd) |
8e48dcfb | 2579 | { |
4e7ea81d JK |
2580 | struct address_space *mapping = mpd->inode->i_mapping; |
2581 | struct pagevec pvec; | |
2582 | unsigned int nr_pages; | |
aeac589a | 2583 | long left = mpd->wbc->nr_to_write; |
4e7ea81d JK |
2584 | pgoff_t index = mpd->first_page; |
2585 | pgoff_t end = mpd->last_page; | |
10bbd235 | 2586 | xa_mark_t tag; |
4e7ea81d JK |
2587 | int i, err = 0; |
2588 | int blkbits = mpd->inode->i_blkbits; | |
2589 | ext4_lblk_t lblk; | |
2590 | struct buffer_head *head; | |
8e48dcfb | 2591 | |
4e7ea81d | 2592 | if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages) |
5b41d924 ES |
2593 | tag = PAGECACHE_TAG_TOWRITE; |
2594 | else | |
2595 | tag = PAGECACHE_TAG_DIRTY; | |
2596 | ||
86679820 | 2597 | pagevec_init(&pvec); |
4e7ea81d JK |
2598 | mpd->map.m_len = 0; |
2599 | mpd->next_page = index; | |
4f01b02c | 2600 | while (index <= end) { |
dc7f3e86 | 2601 | nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end, |
67fd707f | 2602 | tag); |
8e48dcfb | 2603 | if (nr_pages == 0) |
6b8ed620 | 2604 | break; |
8e48dcfb TT |
2605 | |
2606 | for (i = 0; i < nr_pages; i++) { | |
2607 | struct page *page = pvec.pages[i]; | |
2608 | ||
aeac589a ML |
2609 | /* |
2610 | * Accumulated enough dirty pages? This doesn't apply | |
2611 | * to WB_SYNC_ALL mode. For integrity sync we have to | |
2612 | * keep going because someone may be concurrently | |
2613 | * dirtying pages, and we might have synced a lot of | |
2614 | * newly appeared dirty pages, but have not synced all | |
2615 | * of the old dirty pages. | |
2616 | */ | |
2617 | if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0) | |
2618 | goto out; | |
2619 | ||
4e7ea81d JK |
2620 | /* If we can't merge this page, we are done. */ |
2621 | if (mpd->map.m_len > 0 && mpd->next_page != page->index) | |
2622 | goto out; | |
78aaced3 | 2623 | |
8e48dcfb | 2624 | lock_page(page); |
8e48dcfb | 2625 | /* |
4e7ea81d JK |
2626 | * If the page is no longer dirty, or its mapping no |
2627 | * longer corresponds to inode we are writing (which | |
2628 | * means it has been truncated or invalidated), or the | |
2629 | * page is already under writeback and we are not doing | |
2630 | * a data integrity writeback, skip the page | |
8e48dcfb | 2631 | */ |
4f01b02c TT |
2632 | if (!PageDirty(page) || |
2633 | (PageWriteback(page) && | |
4e7ea81d | 2634 | (mpd->wbc->sync_mode == WB_SYNC_NONE)) || |
4f01b02c | 2635 | unlikely(page->mapping != mapping)) { |
8e48dcfb TT |
2636 | unlock_page(page); |
2637 | continue; | |
2638 | } | |
2639 | ||
7cb1a535 | 2640 | wait_on_page_writeback(page); |
8e48dcfb | 2641 | BUG_ON(PageWriteback(page)); |
8e48dcfb | 2642 | |
cc509574 TT |
2643 | /* |
2644 | * Should never happen but for buggy code in | |
2645 | * other subsystems that call | |
2646 | * set_page_dirty() without properly warning | |
2647 | * the file system first. See [1] for more | |
2648 | * information. | |
2649 | * | |
2650 | * [1] https://lore.kernel.org/linux-mm/20180103100430.GE4911@quack2.suse.cz | |
2651 | */ | |
2652 | if (!page_has_buffers(page)) { | |
2653 | ext4_warning_inode(mpd->inode, "page %lu does not have buffers attached", page->index); | |
2654 | ClearPageDirty(page); | |
2655 | unlock_page(page); | |
2656 | continue; | |
2657 | } | |
2658 | ||
4e7ea81d | 2659 | if (mpd->map.m_len == 0) |
8eb9e5ce | 2660 | mpd->first_page = page->index; |
8eb9e5ce | 2661 | mpd->next_page = page->index + 1; |
f8bec370 | 2662 | /* Add all dirty buffers to mpd */ |
4e7ea81d | 2663 | lblk = ((ext4_lblk_t)page->index) << |
09cbfeaf | 2664 | (PAGE_SHIFT - blkbits); |
f8bec370 | 2665 | head = page_buffers(page); |
5f1132b2 JK |
2666 | err = mpage_process_page_bufs(mpd, head, head, lblk); |
2667 | if (err <= 0) | |
4e7ea81d | 2668 | goto out; |
5f1132b2 | 2669 | err = 0; |
aeac589a | 2670 | left--; |
8e48dcfb TT |
2671 | } |
2672 | pagevec_release(&pvec); | |
2673 | cond_resched(); | |
2674 | } | |
6b8ed620 | 2675 | mpd->scanned_until_end = 1; |
4f01b02c | 2676 | return 0; |
8eb9e5ce TT |
2677 | out: |
2678 | pagevec_release(&pvec); | |
4e7ea81d | 2679 | return err; |
8e48dcfb TT |
2680 | } |
2681 | ||
20970ba6 TT |
2682 | static int ext4_writepages(struct address_space *mapping, |
2683 | struct writeback_control *wbc) | |
64769240 | 2684 | { |
4e7ea81d JK |
2685 | pgoff_t writeback_index = 0; |
2686 | long nr_to_write = wbc->nr_to_write; | |
22208ded | 2687 | int range_whole = 0; |
4e7ea81d | 2688 | int cycled = 1; |
61628a3f | 2689 | handle_t *handle = NULL; |
df22291f | 2690 | struct mpage_da_data mpd; |
5e745b04 | 2691 | struct inode *inode = mapping->host; |
6b523df4 | 2692 | int needed_blocks, rsv_blocks = 0, ret = 0; |
5e745b04 | 2693 | struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb); |
1bce63d1 | 2694 | struct blk_plug plug; |
cb530541 | 2695 | bool give_up_on_write = false; |
61628a3f | 2696 | |
0db1ff22 TT |
2697 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
2698 | return -EIO; | |
2699 | ||
bbd55937 | 2700 | percpu_down_read(&sbi->s_writepages_rwsem); |
20970ba6 | 2701 | trace_ext4_writepages(inode, wbc); |
ba80b101 | 2702 | |
61628a3f MC |
2703 | /* |
2704 | * No pages to write? This is mainly a kludge to avoid starting | |
2705 | * a transaction for special inodes like journal inode on last iput() | |
2706 | * because that could violate lock ordering on umount | |
2707 | */ | |
a1d6cc56 | 2708 | if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) |
bbf023c7 | 2709 | goto out_writepages; |
2a21e37e | 2710 | |
20970ba6 | 2711 | if (ext4_should_journal_data(inode)) { |
043d20d1 | 2712 | ret = generic_writepages(mapping, wbc); |
bbf023c7 | 2713 | goto out_writepages; |
20970ba6 TT |
2714 | } |
2715 | ||
2a21e37e TT |
2716 | /* |
2717 | * If the filesystem has aborted, it is read-only, so return | |
2718 | * right away instead of dumping stack traces later on that | |
2719 | * will obscure the real source of the problem. We test | |
1751e8a6 | 2720 | * EXT4_MF_FS_ABORTED instead of sb->s_flag's SB_RDONLY because |
2a21e37e | 2721 | * the latter could be true if the filesystem is mounted |
20970ba6 | 2722 | * read-only, and in that case, ext4_writepages should |
2a21e37e TT |
2723 | * *never* be called, so if that ever happens, we would want |
2724 | * the stack trace. | |
2725 | */ | |
0db1ff22 | 2726 | if (unlikely(ext4_forced_shutdown(EXT4_SB(mapping->host->i_sb)) || |
9b5f6c9b | 2727 | ext4_test_mount_flag(inode->i_sb, EXT4_MF_FS_ABORTED))) { |
bbf023c7 ML |
2728 | ret = -EROFS; |
2729 | goto out_writepages; | |
2730 | } | |
2a21e37e | 2731 | |
4e7ea81d JK |
2732 | /* |
2733 | * If we have inline data and arrive here, it means that | |
2734 | * we will soon create the block for the 1st page, so | |
2735 | * we'd better clear the inline data here. | |
2736 | */ | |
2737 | if (ext4_has_inline_data(inode)) { | |
2738 | /* Just inode will be modified... */ | |
2739 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 1); | |
2740 | if (IS_ERR(handle)) { | |
2741 | ret = PTR_ERR(handle); | |
2742 | goto out_writepages; | |
2743 | } | |
2744 | BUG_ON(ext4_test_inode_state(inode, | |
2745 | EXT4_STATE_MAY_INLINE_DATA)); | |
2746 | ext4_destroy_inline_data(handle, inode); | |
2747 | ext4_journal_stop(handle); | |
2748 | } | |
2749 | ||
4e343231 | 2750 | if (ext4_should_dioread_nolock(inode)) { |
2751 | /* | |
2752 | * We may need to convert up to one extent per block in | |
2753 | * the page and we may dirty the inode. | |
2754 | */ | |
2755 | rsv_blocks = 1 + ext4_chunk_trans_blocks(inode, | |
2756 | PAGE_SIZE >> inode->i_blkbits); | |
2757 | } | |
2758 | ||
22208ded AK |
2759 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
2760 | range_whole = 1; | |
61628a3f | 2761 | |
2acf2c26 | 2762 | if (wbc->range_cyclic) { |
4e7ea81d JK |
2763 | writeback_index = mapping->writeback_index; |
2764 | if (writeback_index) | |
2acf2c26 | 2765 | cycled = 0; |
4e7ea81d JK |
2766 | mpd.first_page = writeback_index; |
2767 | mpd.last_page = -1; | |
5b41d924 | 2768 | } else { |
09cbfeaf KS |
2769 | mpd.first_page = wbc->range_start >> PAGE_SHIFT; |
2770 | mpd.last_page = wbc->range_end >> PAGE_SHIFT; | |
5b41d924 | 2771 | } |
a1d6cc56 | 2772 | |
4e7ea81d JK |
2773 | mpd.inode = inode; |
2774 | mpd.wbc = wbc; | |
2775 | ext4_io_submit_init(&mpd.io_submit, wbc); | |
2acf2c26 | 2776 | retry: |
6e6938b6 | 2777 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
4e7ea81d | 2778 | tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page); |
1bce63d1 | 2779 | blk_start_plug(&plug); |
dddbd6ac JK |
2780 | |
2781 | /* | |
2782 | * First writeback pages that don't need mapping - we can avoid | |
2783 | * starting a transaction unnecessarily and also avoid being blocked | |
2784 | * in the block layer on device congestion while having transaction | |
2785 | * started. | |
2786 | */ | |
2787 | mpd.do_map = 0; | |
6b8ed620 | 2788 | mpd.scanned_until_end = 0; |
dddbd6ac JK |
2789 | mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL); |
2790 | if (!mpd.io_submit.io_end) { | |
2791 | ret = -ENOMEM; | |
2792 | goto unplug; | |
2793 | } | |
2794 | ret = mpage_prepare_extent_to_map(&mpd); | |
a297b2fc XW |
2795 | /* Unlock pages we didn't use */ |
2796 | mpage_release_unused_pages(&mpd, false); | |
dddbd6ac JK |
2797 | /* Submit prepared bio */ |
2798 | ext4_io_submit(&mpd.io_submit); | |
2799 | ext4_put_io_end_defer(mpd.io_submit.io_end); | |
2800 | mpd.io_submit.io_end = NULL; | |
dddbd6ac JK |
2801 | if (ret < 0) |
2802 | goto unplug; | |
2803 | ||
6b8ed620 | 2804 | while (!mpd.scanned_until_end && wbc->nr_to_write > 0) { |
4e7ea81d JK |
2805 | /* For each extent of pages we use new io_end */ |
2806 | mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL); | |
2807 | if (!mpd.io_submit.io_end) { | |
2808 | ret = -ENOMEM; | |
2809 | break; | |
2810 | } | |
a1d6cc56 AK |
2811 | |
2812 | /* | |
4e7ea81d JK |
2813 | * We have two constraints: We find one extent to map and we |
2814 | * must always write out whole page (makes a difference when | |
2815 | * blocksize < pagesize) so that we don't block on IO when we | |
2816 | * try to write out the rest of the page. Journalled mode is | |
2817 | * not supported by delalloc. | |
a1d6cc56 AK |
2818 | */ |
2819 | BUG_ON(ext4_should_journal_data(inode)); | |
525f4ed8 | 2820 | needed_blocks = ext4_da_writepages_trans_blocks(inode); |
a1d6cc56 | 2821 | |
4e7ea81d | 2822 | /* start a new transaction */ |
6b523df4 JK |
2823 | handle = ext4_journal_start_with_reserve(inode, |
2824 | EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks); | |
61628a3f MC |
2825 | if (IS_ERR(handle)) { |
2826 | ret = PTR_ERR(handle); | |
1693918e | 2827 | ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: " |
fbe845dd | 2828 | "%ld pages, ino %lu; err %d", __func__, |
a1d6cc56 | 2829 | wbc->nr_to_write, inode->i_ino, ret); |
4e7ea81d JK |
2830 | /* Release allocated io_end */ |
2831 | ext4_put_io_end(mpd.io_submit.io_end); | |
dddbd6ac | 2832 | mpd.io_submit.io_end = NULL; |
4e7ea81d | 2833 | break; |
61628a3f | 2834 | } |
dddbd6ac | 2835 | mpd.do_map = 1; |
f63e6005 | 2836 | |
4e7ea81d JK |
2837 | trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc); |
2838 | ret = mpage_prepare_extent_to_map(&mpd); | |
6b8ed620 JK |
2839 | if (!ret && mpd.map.m_len) |
2840 | ret = mpage_map_and_submit_extent(handle, &mpd, | |
cb530541 | 2841 | &give_up_on_write); |
646caa9c JK |
2842 | /* |
2843 | * Caution: If the handle is synchronous, | |
2844 | * ext4_journal_stop() can wait for transaction commit | |
2845 | * to finish which may depend on writeback of pages to | |
2846 | * complete or on page lock to be released. In that | |
b483bb77 | 2847 | * case, we have to wait until after we have |
646caa9c JK |
2848 | * submitted all the IO, released page locks we hold, |
2849 | * and dropped io_end reference (for extent conversion | |
2850 | * to be able to complete) before stopping the handle. | |
2851 | */ | |
2852 | if (!ext4_handle_valid(handle) || handle->h_sync == 0) { | |
2853 | ext4_journal_stop(handle); | |
2854 | handle = NULL; | |
dddbd6ac | 2855 | mpd.do_map = 0; |
646caa9c | 2856 | } |
4e7ea81d | 2857 | /* Unlock pages we didn't use */ |
cb530541 | 2858 | mpage_release_unused_pages(&mpd, give_up_on_write); |
a297b2fc XW |
2859 | /* Submit prepared bio */ |
2860 | ext4_io_submit(&mpd.io_submit); | |
2861 | ||
646caa9c JK |
2862 | /* |
2863 | * Drop our io_end reference we got from init. We have | |
2864 | * to be careful and use deferred io_end finishing if | |
2865 | * we are still holding the transaction as we can | |
2866 | * release the last reference to io_end which may end | |
2867 | * up doing unwritten extent conversion. | |
2868 | */ | |
2869 | if (handle) { | |
2870 | ext4_put_io_end_defer(mpd.io_submit.io_end); | |
2871 | ext4_journal_stop(handle); | |
2872 | } else | |
2873 | ext4_put_io_end(mpd.io_submit.io_end); | |
dddbd6ac | 2874 | mpd.io_submit.io_end = NULL; |
4e7ea81d JK |
2875 | |
2876 | if (ret == -ENOSPC && sbi->s_journal) { | |
2877 | /* | |
2878 | * Commit the transaction which would | |
22208ded AK |
2879 | * free blocks released in the transaction |
2880 | * and try again | |
2881 | */ | |
df22291f | 2882 | jbd2_journal_force_commit_nested(sbi->s_journal); |
22208ded | 2883 | ret = 0; |
4e7ea81d JK |
2884 | continue; |
2885 | } | |
2886 | /* Fatal error - ENOMEM, EIO... */ | |
2887 | if (ret) | |
61628a3f | 2888 | break; |
a1d6cc56 | 2889 | } |
dddbd6ac | 2890 | unplug: |
1bce63d1 | 2891 | blk_finish_plug(&plug); |
9c12a831 | 2892 | if (!ret && !cycled && wbc->nr_to_write > 0) { |
2acf2c26 | 2893 | cycled = 1; |
4e7ea81d JK |
2894 | mpd.last_page = writeback_index - 1; |
2895 | mpd.first_page = 0; | |
2acf2c26 AK |
2896 | goto retry; |
2897 | } | |
22208ded AK |
2898 | |
2899 | /* Update index */ | |
22208ded AK |
2900 | if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) |
2901 | /* | |
4e7ea81d | 2902 | * Set the writeback_index so that range_cyclic |
22208ded AK |
2903 | * mode will write it back later |
2904 | */ | |
4e7ea81d | 2905 | mapping->writeback_index = mpd.first_page; |
a1d6cc56 | 2906 | |
61628a3f | 2907 | out_writepages: |
20970ba6 TT |
2908 | trace_ext4_writepages_result(inode, wbc, ret, |
2909 | nr_to_write - wbc->nr_to_write); | |
bbd55937 | 2910 | percpu_up_read(&sbi->s_writepages_rwsem); |
61628a3f | 2911 | return ret; |
64769240 AT |
2912 | } |
2913 | ||
5f0663bb DW |
2914 | static int ext4_dax_writepages(struct address_space *mapping, |
2915 | struct writeback_control *wbc) | |
2916 | { | |
2917 | int ret; | |
2918 | long nr_to_write = wbc->nr_to_write; | |
2919 | struct inode *inode = mapping->host; | |
2920 | struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb); | |
2921 | ||
2922 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) | |
2923 | return -EIO; | |
2924 | ||
bbd55937 | 2925 | percpu_down_read(&sbi->s_writepages_rwsem); |
5f0663bb DW |
2926 | trace_ext4_writepages(inode, wbc); |
2927 | ||
3f666c56 | 2928 | ret = dax_writeback_mapping_range(mapping, sbi->s_daxdev, wbc); |
5f0663bb DW |
2929 | trace_ext4_writepages_result(inode, wbc, ret, |
2930 | nr_to_write - wbc->nr_to_write); | |
bbd55937 | 2931 | percpu_up_read(&sbi->s_writepages_rwsem); |
5f0663bb DW |
2932 | return ret; |
2933 | } | |
2934 | ||
79f0be8d AK |
2935 | static int ext4_nonda_switch(struct super_block *sb) |
2936 | { | |
5c1ff336 | 2937 | s64 free_clusters, dirty_clusters; |
79f0be8d AK |
2938 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
2939 | ||
2940 | /* | |
2941 | * switch to non delalloc mode if we are running low | |
2942 | * on free block. The free block accounting via percpu | |
179f7ebf | 2943 | * counters can get slightly wrong with percpu_counter_batch getting |
79f0be8d AK |
2944 | * accumulated on each CPU without updating global counters |
2945 | * Delalloc need an accurate free block accounting. So switch | |
2946 | * to non delalloc when we are near to error range. | |
2947 | */ | |
5c1ff336 EW |
2948 | free_clusters = |
2949 | percpu_counter_read_positive(&sbi->s_freeclusters_counter); | |
2950 | dirty_clusters = | |
2951 | percpu_counter_read_positive(&sbi->s_dirtyclusters_counter); | |
00d4e736 TT |
2952 | /* |
2953 | * Start pushing delalloc when 1/2 of free blocks are dirty. | |
2954 | */ | |
5c1ff336 | 2955 | if (dirty_clusters && (free_clusters < 2 * dirty_clusters)) |
10ee27a0 | 2956 | try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE); |
00d4e736 | 2957 | |
5c1ff336 EW |
2958 | if (2 * free_clusters < 3 * dirty_clusters || |
2959 | free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) { | |
79f0be8d | 2960 | /* |
c8afb446 ES |
2961 | * free block count is less than 150% of dirty blocks |
2962 | * or free blocks is less than watermark | |
79f0be8d AK |
2963 | */ |
2964 | return 1; | |
2965 | } | |
2966 | return 0; | |
2967 | } | |
2968 | ||
64769240 | 2969 | static int ext4_da_write_begin(struct file *file, struct address_space *mapping, |
9d6b0cd7 | 2970 | loff_t pos, unsigned len, |
de9a55b8 | 2971 | struct page **pagep, void **fsdata) |
64769240 | 2972 | { |
72b8ab9d | 2973 | int ret, retries = 0; |
64769240 AT |
2974 | struct page *page; |
2975 | pgoff_t index; | |
64769240 | 2976 | struct inode *inode = mapping->host; |
64769240 | 2977 | |
0db1ff22 TT |
2978 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
2979 | return -EIO; | |
2980 | ||
09cbfeaf | 2981 | index = pos >> PAGE_SHIFT; |
79f0be8d | 2982 | |
6493792d | 2983 | if (ext4_nonda_switch(inode->i_sb) || ext4_verity_in_progress(inode)) { |
79f0be8d AK |
2984 | *fsdata = (void *)FALL_BACK_TO_NONDELALLOC; |
2985 | return ext4_write_begin(file, mapping, pos, | |
9d6b0cd7 | 2986 | len, pagep, fsdata); |
79f0be8d AK |
2987 | } |
2988 | *fsdata = (void *)0; | |
9d6b0cd7 | 2989 | trace_ext4_da_write_begin(inode, pos, len); |
9c3569b5 TM |
2990 | |
2991 | if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { | |
36d116e9 | 2992 | ret = ext4_da_write_inline_data_begin(mapping, inode, pos, len, |
9c3569b5 TM |
2993 | pagep, fsdata); |
2994 | if (ret < 0) | |
47564bfb TT |
2995 | return ret; |
2996 | if (ret == 1) | |
2997 | return 0; | |
9c3569b5 TM |
2998 | } |
2999 | ||
cc883236 | 3000 | retry: |
b7446e7c | 3001 | page = grab_cache_page_write_begin(mapping, index); |
47564bfb TT |
3002 | if (!page) |
3003 | return -ENOMEM; | |
47564bfb | 3004 | |
47564bfb | 3005 | /* In case writeback began while the page was unlocked */ |
7afe5aa5 | 3006 | wait_for_stable_page(page); |
64769240 | 3007 | |
643fa961 | 3008 | #ifdef CONFIG_FS_ENCRYPTION |
2058f83a MH |
3009 | ret = ext4_block_write_begin(page, pos, len, |
3010 | ext4_da_get_block_prep); | |
3011 | #else | |
6e1db88d | 3012 | ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep); |
2058f83a | 3013 | #endif |
64769240 AT |
3014 | if (ret < 0) { |
3015 | unlock_page(page); | |
cc883236 | 3016 | put_page(page); |
ae4d5372 AK |
3017 | /* |
3018 | * block_write_begin may have instantiated a few blocks | |
3019 | * outside i_size. Trim these off again. Don't need | |
cc883236 | 3020 | * i_size_read because we hold inode lock. |
ae4d5372 AK |
3021 | */ |
3022 | if (pos + len > inode->i_size) | |
b9a4207d | 3023 | ext4_truncate_failed_write(inode); |
47564bfb TT |
3024 | |
3025 | if (ret == -ENOSPC && | |
3026 | ext4_should_retry_alloc(inode->i_sb, &retries)) | |
cc883236 | 3027 | goto retry; |
47564bfb | 3028 | return ret; |
64769240 AT |
3029 | } |
3030 | ||
47564bfb | 3031 | *pagep = page; |
64769240 AT |
3032 | return ret; |
3033 | } | |
3034 | ||
632eaeab MC |
3035 | /* |
3036 | * Check if we should update i_disksize | |
3037 | * when write to the end of file but not require block allocation | |
3038 | */ | |
3039 | static int ext4_da_should_update_i_disksize(struct page *page, | |
de9a55b8 | 3040 | unsigned long offset) |
632eaeab MC |
3041 | { |
3042 | struct buffer_head *bh; | |
3043 | struct inode *inode = page->mapping->host; | |
3044 | unsigned int idx; | |
3045 | int i; | |
3046 | ||
3047 | bh = page_buffers(page); | |
3048 | idx = offset >> inode->i_blkbits; | |
3049 | ||
af5bc92d | 3050 | for (i = 0; i < idx; i++) |
632eaeab MC |
3051 | bh = bh->b_this_page; |
3052 | ||
29fa89d0 | 3053 | if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh)) |
632eaeab MC |
3054 | return 0; |
3055 | return 1; | |
3056 | } | |
3057 | ||
64769240 | 3058 | static int ext4_da_write_end(struct file *file, |
de9a55b8 TT |
3059 | struct address_space *mapping, |
3060 | loff_t pos, unsigned len, unsigned copied, | |
3061 | struct page *page, void *fsdata) | |
64769240 AT |
3062 | { |
3063 | struct inode *inode = mapping->host; | |
64769240 | 3064 | loff_t new_i_size; |
632eaeab | 3065 | unsigned long start, end; |
79f0be8d AK |
3066 | int write_mode = (int)(unsigned long)fsdata; |
3067 | ||
74d553aa TT |
3068 | if (write_mode == FALL_BACK_TO_NONDELALLOC) |
3069 | return ext4_write_end(file, mapping, pos, | |
3070 | len, copied, page, fsdata); | |
632eaeab | 3071 | |
9bffad1e | 3072 | trace_ext4_da_write_end(inode, pos, len, copied); |
6984aef5 ZY |
3073 | |
3074 | if (write_mode != CONVERT_INLINE_DATA && | |
3075 | ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) && | |
3076 | ext4_has_inline_data(inode)) | |
3077 | return ext4_write_inline_data_end(inode, pos, len, copied, page); | |
3078 | ||
09cbfeaf | 3079 | start = pos & (PAGE_SIZE - 1); |
af5bc92d | 3080 | end = start + copied - 1; |
64769240 AT |
3081 | |
3082 | /* | |
4df031ff ZY |
3083 | * Since we are holding inode lock, we are sure i_disksize <= |
3084 | * i_size. We also know that if i_disksize < i_size, there are | |
3085 | * delalloc writes pending in the range upto i_size. If the end of | |
3086 | * the current write is <= i_size, there's no need to touch | |
3087 | * i_disksize since writeback will push i_disksize upto i_size | |
3088 | * eventually. If the end of the current write is > i_size and | |
3089 | * inside an allocated block (ext4_da_should_update_i_disksize() | |
3090 | * check), we need to update i_disksize here as neither | |
3091 | * ext4_writepage() nor certain ext4_writepages() paths not | |
3092 | * allocating blocks update i_disksize. | |
3093 | * | |
3094 | * Note that we defer inode dirtying to generic_write_end() / | |
3095 | * ext4_da_write_inline_data_end(). | |
64769240 | 3096 | */ |
64769240 | 3097 | new_i_size = pos + copied; |
6984aef5 ZY |
3098 | if (copied && new_i_size > inode->i_size && |
3099 | ext4_da_should_update_i_disksize(page, end)) | |
3100 | ext4_update_i_disksize(inode, new_i_size); | |
9c3569b5 | 3101 | |
cc883236 | 3102 | return generic_write_end(file, mapping, pos, len, copied, page, fsdata); |
64769240 AT |
3103 | } |
3104 | ||
ccd2506b TT |
3105 | /* |
3106 | * Force all delayed allocation blocks to be allocated for a given inode. | |
3107 | */ | |
3108 | int ext4_alloc_da_blocks(struct inode *inode) | |
3109 | { | |
fb40ba0d TT |
3110 | trace_ext4_alloc_da_blocks(inode); |
3111 | ||
71d4f7d0 | 3112 | if (!EXT4_I(inode)->i_reserved_data_blocks) |
ccd2506b TT |
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: | |
de9a55b8 | 3121 | * |
20970ba6 | 3122 | * ext4_writepages() -> |
ccd2506b TT |
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 | |
380cf090 | 3134 | * the pages by calling redirty_page_for_writepage() but that |
ccd2506b TT |
3135 | * would be ugly in the extreme. So instead we would need to |
3136 | * replicate parts of the code in the above functions, | |
25985edc | 3137 | * simplifying them because we wouldn't actually intend to |
ccd2506b TT |
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. | |
de9a55b8 | 3141 | * |
ccd2506b TT |
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 | return filemap_flush(inode->i_mapping); | |
3147 | } | |
64769240 | 3148 | |
ac27a0ec DK |
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 | |
617ba13b | 3154 | * journal. If somebody makes a swapfile on an ext4 data-journaling |
ac27a0ec DK |
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 | */ | |
617ba13b | 3163 | static sector_t ext4_bmap(struct address_space *mapping, sector_t block) |
ac27a0ec DK |
3164 | { |
3165 | struct inode *inode = mapping->host; | |
3166 | journal_t *journal; | |
51ae846c | 3167 | sector_t ret = 0; |
ac27a0ec DK |
3168 | int err; |
3169 | ||
51ae846c | 3170 | inode_lock_shared(inode); |
46c7f254 TM |
3171 | /* |
3172 | * We can get here for an inline file via the FIBMAP ioctl | |
3173 | */ | |
3174 | if (ext4_has_inline_data(inode)) | |
51ae846c | 3175 | goto out; |
46c7f254 | 3176 | |
64769240 AT |
3177 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) && |
3178 | test_opt(inode->i_sb, DELALLOC)) { | |
3179 | /* | |
3180 | * With delalloc we want to sync the file | |
3181 | * so that we can make sure we allocate | |
3182 | * blocks for file | |
3183 | */ | |
3184 | filemap_write_and_wait(mapping); | |
3185 | } | |
3186 | ||
19f5fb7a TT |
3187 | if (EXT4_JOURNAL(inode) && |
3188 | ext4_test_inode_state(inode, EXT4_STATE_JDATA)) { | |
ac27a0ec DK |
3189 | /* |
3190 | * This is a REALLY heavyweight approach, but the use of | |
3191 | * bmap on dirty files is expected to be extremely rare: | |
3192 | * only if we run lilo or swapon on a freshly made file | |
3193 | * do we expect this to happen. | |
3194 | * | |
3195 | * (bmap requires CAP_SYS_RAWIO so this does not | |
3196 | * represent an unprivileged user DOS attack --- we'd be | |
3197 | * in trouble if mortal users could trigger this path at | |
3198 | * will.) | |
3199 | * | |
617ba13b | 3200 | * NB. EXT4_STATE_JDATA is not set on files other than |
ac27a0ec DK |
3201 | * regular files. If somebody wants to bmap a directory |
3202 | * or symlink and gets confused because the buffer | |
3203 | * hasn't yet been flushed to disk, they deserve | |
3204 | * everything they get. | |
3205 | */ | |
3206 | ||
19f5fb7a | 3207 | ext4_clear_inode_state(inode, EXT4_STATE_JDATA); |
617ba13b | 3208 | journal = EXT4_JOURNAL(inode); |
dab291af | 3209 | jbd2_journal_lock_updates(journal); |
01d5d965 | 3210 | err = jbd2_journal_flush(journal, 0); |
dab291af | 3211 | jbd2_journal_unlock_updates(journal); |
ac27a0ec DK |
3212 | |
3213 | if (err) | |
51ae846c | 3214 | goto out; |
ac27a0ec DK |
3215 | } |
3216 | ||
51ae846c YB |
3217 | ret = iomap_bmap(mapping, block, &ext4_iomap_ops); |
3218 | ||
3219 | out: | |
3220 | inode_unlock_shared(inode); | |
3221 | return ret; | |
ac27a0ec DK |
3222 | } |
3223 | ||
fe5ddf6b | 3224 | static int ext4_read_folio(struct file *file, struct folio *folio) |
ac27a0ec | 3225 | { |
fe5ddf6b | 3226 | struct page *page = &folio->page; |
46c7f254 TM |
3227 | int ret = -EAGAIN; |
3228 | struct inode *inode = page->mapping->host; | |
3229 | ||
0562e0ba | 3230 | trace_ext4_readpage(page); |
46c7f254 TM |
3231 | |
3232 | if (ext4_has_inline_data(inode)) | |
3233 | ret = ext4_readpage_inline(inode, page); | |
3234 | ||
3235 | if (ret == -EAGAIN) | |
a07f624b | 3236 | return ext4_mpage_readpages(inode, NULL, page); |
46c7f254 TM |
3237 | |
3238 | return ret; | |
ac27a0ec DK |
3239 | } |
3240 | ||
6311f91f | 3241 | static void ext4_readahead(struct readahead_control *rac) |
ac27a0ec | 3242 | { |
6311f91f | 3243 | struct inode *inode = rac->mapping->host; |
46c7f254 | 3244 | |
6311f91f | 3245 | /* If the file has inline data, no need to do readahead. */ |
46c7f254 | 3246 | if (ext4_has_inline_data(inode)) |
6311f91f | 3247 | return; |
46c7f254 | 3248 | |
a07f624b | 3249 | ext4_mpage_readpages(inode, rac, NULL); |
ac27a0ec DK |
3250 | } |
3251 | ||
7ba13abb MWO |
3252 | static void ext4_invalidate_folio(struct folio *folio, size_t offset, |
3253 | size_t length) | |
ac27a0ec | 3254 | { |
ccd16945 | 3255 | trace_ext4_invalidate_folio(folio, offset, length); |
0562e0ba | 3256 | |
4520fb3c | 3257 | /* No journalling happens on data buffers when this function is used */ |
7ba13abb | 3258 | WARN_ON(folio_buffers(folio) && buffer_jbd(folio_buffers(folio))); |
4520fb3c | 3259 | |
7ba13abb | 3260 | block_invalidate_folio(folio, offset, length); |
4520fb3c JK |
3261 | } |
3262 | ||
ccd16945 MWO |
3263 | static int __ext4_journalled_invalidate_folio(struct folio *folio, |
3264 | size_t offset, size_t length) | |
4520fb3c | 3265 | { |
ccd16945 | 3266 | journal_t *journal = EXT4_JOURNAL(folio->mapping->host); |
4520fb3c | 3267 | |
ccd16945 | 3268 | trace_ext4_journalled_invalidate_folio(folio, offset, length); |
4520fb3c | 3269 | |
ac27a0ec DK |
3270 | /* |
3271 | * If it's a full truncate we just forget about the pending dirtying | |
3272 | */ | |
ccd16945 MWO |
3273 | if (offset == 0 && length == folio_size(folio)) |
3274 | folio_clear_checked(folio); | |
ac27a0ec | 3275 | |
ccd16945 | 3276 | return jbd2_journal_invalidate_folio(journal, folio, offset, length); |
53e87268 JK |
3277 | } |
3278 | ||
3279 | /* Wrapper for aops... */ | |
ccd16945 MWO |
3280 | static void ext4_journalled_invalidate_folio(struct folio *folio, |
3281 | size_t offset, | |
3282 | size_t length) | |
53e87268 | 3283 | { |
ccd16945 | 3284 | WARN_ON(__ext4_journalled_invalidate_folio(folio, offset, length) < 0); |
ac27a0ec DK |
3285 | } |
3286 | ||
3c402f15 | 3287 | static bool ext4_release_folio(struct folio *folio, gfp_t wait) |
ac27a0ec | 3288 | { |
3c402f15 | 3289 | journal_t *journal = EXT4_JOURNAL(folio->mapping->host); |
ac27a0ec | 3290 | |
3c402f15 | 3291 | trace_ext4_releasepage(&folio->page); |
0562e0ba | 3292 | |
e1c36595 | 3293 | /* Page has dirty journalled data -> cannot release */ |
3c402f15 MWO |
3294 | if (folio_test_checked(folio)) |
3295 | return false; | |
0390131b | 3296 | if (journal) |
c56a6eb0 | 3297 | return jbd2_journal_try_to_free_buffers(journal, folio); |
0390131b | 3298 | else |
68189fef | 3299 | return try_to_free_buffers(folio); |
ac27a0ec DK |
3300 | } |
3301 | ||
b8a6176c JK |
3302 | static bool ext4_inode_datasync_dirty(struct inode *inode) |
3303 | { | |
3304 | journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; | |
3305 | ||
aa75f4d3 HS |
3306 | if (journal) { |
3307 | if (jbd2_transaction_committed(journal, | |
d0520df7 AR |
3308 | EXT4_I(inode)->i_datasync_tid)) |
3309 | return false; | |
3310 | if (test_opt2(inode->i_sb, JOURNAL_FAST_COMMIT)) | |
1ceecb53 | 3311 | return !list_empty(&EXT4_I(inode)->i_fc_list); |
d0520df7 | 3312 | return true; |
aa75f4d3 HS |
3313 | } |
3314 | ||
b8a6176c JK |
3315 | /* Any metadata buffers to write? */ |
3316 | if (!list_empty(&inode->i_mapping->private_list)) | |
3317 | return true; | |
3318 | return inode->i_state & I_DIRTY_DATASYNC; | |
3319 | } | |
3320 | ||
c8fdfe29 MB |
3321 | static void ext4_set_iomap(struct inode *inode, struct iomap *iomap, |
3322 | struct ext4_map_blocks *map, loff_t offset, | |
de205114 | 3323 | loff_t length, unsigned int flags) |
364443cb | 3324 | { |
c8fdfe29 | 3325 | u8 blkbits = inode->i_blkbits; |
364443cb | 3326 | |
c8fdfe29 MB |
3327 | /* |
3328 | * Writes that span EOF might trigger an I/O size update on completion, | |
3329 | * so consider them to be dirty for the purpose of O_DSYNC, even if | |
3330 | * there is no other metadata changes being made or are pending. | |
3331 | */ | |
364443cb | 3332 | iomap->flags = 0; |
c8fdfe29 MB |
3333 | if (ext4_inode_datasync_dirty(inode) || |
3334 | offset + length > i_size_read(inode)) | |
b8a6176c | 3335 | iomap->flags |= IOMAP_F_DIRTY; |
c8fdfe29 MB |
3336 | |
3337 | if (map->m_flags & EXT4_MAP_NEW) | |
3338 | iomap->flags |= IOMAP_F_NEW; | |
3339 | ||
de205114 CH |
3340 | if (flags & IOMAP_DAX) |
3341 | iomap->dax_dev = EXT4_SB(inode->i_sb)->s_daxdev; | |
3342 | else | |
3343 | iomap->bdev = inode->i_sb->s_bdev; | |
c8fdfe29 MB |
3344 | iomap->offset = (u64) map->m_lblk << blkbits; |
3345 | iomap->length = (u64) map->m_len << blkbits; | |
364443cb | 3346 | |
6386722a RH |
3347 | if ((map->m_flags & EXT4_MAP_MAPPED) && |
3348 | !ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
3349 | iomap->flags |= IOMAP_F_MERGED; | |
3350 | ||
c8fdfe29 MB |
3351 | /* |
3352 | * Flags passed to ext4_map_blocks() for direct I/O writes can result | |
3353 | * in m_flags having both EXT4_MAP_MAPPED and EXT4_MAP_UNWRITTEN bits | |
3354 | * set. In order for any allocated unwritten extents to be converted | |
3355 | * into written extents correctly within the ->end_io() handler, we | |
3356 | * need to ensure that the iomap->type is set appropriately. Hence, the | |
3357 | * reason why we need to check whether the EXT4_MAP_UNWRITTEN bit has | |
3358 | * been set first. | |
3359 | */ | |
3360 | if (map->m_flags & EXT4_MAP_UNWRITTEN) { | |
3361 | iomap->type = IOMAP_UNWRITTEN; | |
3362 | iomap->addr = (u64) map->m_pblk << blkbits; | |
de205114 CH |
3363 | if (flags & IOMAP_DAX) |
3364 | iomap->addr += EXT4_SB(inode->i_sb)->s_dax_part_off; | |
c8fdfe29 MB |
3365 | } else if (map->m_flags & EXT4_MAP_MAPPED) { |
3366 | iomap->type = IOMAP_MAPPED; | |
3367 | iomap->addr = (u64) map->m_pblk << blkbits; | |
de205114 CH |
3368 | if (flags & IOMAP_DAX) |
3369 | iomap->addr += EXT4_SB(inode->i_sb)->s_dax_part_off; | |
364443cb | 3370 | } else { |
c8fdfe29 MB |
3371 | iomap->type = IOMAP_HOLE; |
3372 | iomap->addr = IOMAP_NULL_ADDR; | |
364443cb | 3373 | } |
364443cb JK |
3374 | } |
3375 | ||
f063db5e MB |
3376 | static int ext4_iomap_alloc(struct inode *inode, struct ext4_map_blocks *map, |
3377 | unsigned int flags) | |
776722e8 | 3378 | { |
776722e8 | 3379 | handle_t *handle; |
378f32ba MB |
3380 | u8 blkbits = inode->i_blkbits; |
3381 | int ret, dio_credits, m_flags = 0, retries = 0; | |
776722e8 | 3382 | |
776722e8 | 3383 | /* |
f063db5e MB |
3384 | * Trim the mapping request to the maximum value that we can map at |
3385 | * once for direct I/O. | |
776722e8 | 3386 | */ |
f063db5e MB |
3387 | if (map->m_len > DIO_MAX_BLOCKS) |
3388 | map->m_len = DIO_MAX_BLOCKS; | |
3389 | dio_credits = ext4_chunk_trans_blocks(inode, map->m_len); | |
776722e8 | 3390 | |
f063db5e | 3391 | retry: |
776722e8 | 3392 | /* |
f063db5e MB |
3393 | * Either we allocate blocks and then don't get an unwritten extent, so |
3394 | * in that case we have reserved enough credits. Or, the blocks are | |
3395 | * already allocated and unwritten. In that case, the extent conversion | |
3396 | * fits into the credits as well. | |
776722e8 | 3397 | */ |
f063db5e MB |
3398 | handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits); |
3399 | if (IS_ERR(handle)) | |
3400 | return PTR_ERR(handle); | |
4c0425ff | 3401 | |
378f32ba MB |
3402 | /* |
3403 | * DAX and direct I/O are the only two operations that are currently | |
3404 | * supported with IOMAP_WRITE. | |
3405 | */ | |
952da063 CH |
3406 | WARN_ON(!(flags & (IOMAP_DAX | IOMAP_DIRECT))); |
3407 | if (flags & IOMAP_DAX) | |
378f32ba MB |
3408 | m_flags = EXT4_GET_BLOCKS_CREATE_ZERO; |
3409 | /* | |
3410 | * We use i_size instead of i_disksize here because delalloc writeback | |
3411 | * can complete at any point during the I/O and subsequently push the | |
3412 | * i_disksize out to i_size. This could be beyond where direct I/O is | |
3413 | * happening and thus expose allocated blocks to direct I/O reads. | |
3414 | */ | |
d0b040f5 | 3415 | else if (((loff_t)map->m_lblk << blkbits) >= i_size_read(inode)) |
378f32ba MB |
3416 | m_flags = EXT4_GET_BLOCKS_CREATE; |
3417 | else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
3418 | m_flags = EXT4_GET_BLOCKS_IO_CREATE_EXT; | |
4b70df18 | 3419 | |
378f32ba | 3420 | ret = ext4_map_blocks(handle, inode, map, m_flags); |
8d5d02e6 | 3421 | |
74c66bcb | 3422 | /* |
378f32ba MB |
3423 | * We cannot fill holes in indirect tree based inodes as that could |
3424 | * expose stale data in the case of a crash. Use the magic error code | |
3425 | * to fallback to buffered I/O. | |
74c66bcb | 3426 | */ |
378f32ba MB |
3427 | if (!m_flags && !ret) |
3428 | ret = -ENOTBLK; | |
187372a3 | 3429 | |
f063db5e MB |
3430 | ext4_journal_stop(handle); |
3431 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) | |
3432 | goto retry; | |
3433 | ||
3434 | return ret; | |
4c0425ff | 3435 | } |
c7064ef1 | 3436 | |
f063db5e | 3437 | |
364443cb | 3438 | static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length, |
c039b997 | 3439 | unsigned flags, struct iomap *iomap, struct iomap *srcmap) |
4c0425ff | 3440 | { |
364443cb | 3441 | int ret; |
09edf4d3 MB |
3442 | struct ext4_map_blocks map; |
3443 | u8 blkbits = inode->i_blkbits; | |
729f52c6 | 3444 | |
bcd8e91f TT |
3445 | if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK) |
3446 | return -EINVAL; | |
4bd809db | 3447 | |
09edf4d3 MB |
3448 | if (WARN_ON_ONCE(ext4_has_inline_data(inode))) |
3449 | return -ERANGE; | |
4bd809db | 3450 | |
e8340395 | 3451 | /* |
09edf4d3 | 3452 | * Calculate the first and last logical blocks respectively. |
e8340395 | 3453 | */ |
09edf4d3 MB |
3454 | map.m_lblk = offset >> blkbits; |
3455 | map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits, | |
3456 | EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1; | |
e8340395 | 3457 | |
9faac62d RH |
3458 | if (flags & IOMAP_WRITE) { |
3459 | /* | |
3460 | * We check here if the blocks are already allocated, then we | |
3461 | * don't need to start a journal txn and we can directly return | |
3462 | * the mapping information. This could boost performance | |
3463 | * especially in multi-threaded overwrite requests. | |
3464 | */ | |
3465 | if (offset + length <= i_size_read(inode)) { | |
3466 | ret = ext4_map_blocks(NULL, inode, &map, 0); | |
3467 | if (ret > 0 && (map.m_flags & EXT4_MAP_MAPPED)) | |
3468 | goto out; | |
3469 | } | |
f063db5e | 3470 | ret = ext4_iomap_alloc(inode, &map, flags); |
9faac62d | 3471 | } else { |
545052e9 | 3472 | ret = ext4_map_blocks(NULL, inode, &map, 0); |
9faac62d | 3473 | } |
4bd809db | 3474 | |
f063db5e MB |
3475 | if (ret < 0) |
3476 | return ret; | |
9faac62d | 3477 | out: |
38ea50da EB |
3478 | /* |
3479 | * When inline encryption is enabled, sometimes I/O to an encrypted file | |
3480 | * has to be broken up to guarantee DUN contiguity. Handle this by | |
3481 | * limiting the length of the mapping returned. | |
3482 | */ | |
3483 | map.m_len = fscrypt_limit_io_blocks(inode, map.m_lblk, map.m_len); | |
3484 | ||
de205114 | 3485 | ext4_set_iomap(inode, iomap, &map, offset, length, flags); |
4bd809db | 3486 | |
364443cb JK |
3487 | return 0; |
3488 | } | |
8d5d02e6 | 3489 | |
8cd115bd JK |
3490 | static int ext4_iomap_overwrite_begin(struct inode *inode, loff_t offset, |
3491 | loff_t length, unsigned flags, struct iomap *iomap, | |
3492 | struct iomap *srcmap) | |
3493 | { | |
3494 | int ret; | |
3495 | ||
3496 | /* | |
3497 | * Even for writes we don't need to allocate blocks, so just pretend | |
3498 | * we are reading to save overhead of starting a transaction. | |
3499 | */ | |
3500 | flags &= ~IOMAP_WRITE; | |
3501 | ret = ext4_iomap_begin(inode, offset, length, flags, iomap, srcmap); | |
3502 | WARN_ON_ONCE(iomap->type != IOMAP_MAPPED); | |
3503 | return ret; | |
3504 | } | |
3505 | ||
776722e8 JK |
3506 | static int ext4_iomap_end(struct inode *inode, loff_t offset, loff_t length, |
3507 | ssize_t written, unsigned flags, struct iomap *iomap) | |
3508 | { | |
69c499d1 | 3509 | /* |
378f32ba MB |
3510 | * Check to see whether an error occurred while writing out the data to |
3511 | * the allocated blocks. If so, return the magic error code so that we | |
3512 | * fallback to buffered I/O and attempt to complete the remainder of | |
3513 | * the I/O. Any blocks that may have been allocated in preparation for | |
3514 | * the direct I/O will be reused during buffered I/O. | |
69c499d1 | 3515 | */ |
378f32ba MB |
3516 | if (flags & (IOMAP_WRITE | IOMAP_DIRECT) && written == 0) |
3517 | return -ENOTBLK; | |
69c499d1 | 3518 | |
569342dc | 3519 | return 0; |
776722e8 | 3520 | } |
4bd809db | 3521 | |
8ff6daa1 | 3522 | const struct iomap_ops ext4_iomap_ops = { |
364443cb | 3523 | .iomap_begin = ext4_iomap_begin, |
776722e8 | 3524 | .iomap_end = ext4_iomap_end, |
364443cb | 3525 | }; |
8d5d02e6 | 3526 | |
8cd115bd JK |
3527 | const struct iomap_ops ext4_iomap_overwrite_ops = { |
3528 | .iomap_begin = ext4_iomap_overwrite_begin, | |
3529 | .iomap_end = ext4_iomap_end, | |
3530 | }; | |
3531 | ||
09edf4d3 MB |
3532 | static bool ext4_iomap_is_delalloc(struct inode *inode, |
3533 | struct ext4_map_blocks *map) | |
3534 | { | |
3535 | struct extent_status es; | |
3536 | ext4_lblk_t offset = 0, end = map->m_lblk + map->m_len - 1; | |
914f82a3 | 3537 | |
09edf4d3 MB |
3538 | ext4_es_find_extent_range(inode, &ext4_es_is_delayed, |
3539 | map->m_lblk, end, &es); | |
914f82a3 | 3540 | |
09edf4d3 MB |
3541 | if (!es.es_len || es.es_lblk > end) |
3542 | return false; | |
914f82a3 | 3543 | |
09edf4d3 MB |
3544 | if (es.es_lblk > map->m_lblk) { |
3545 | map->m_len = es.es_lblk - map->m_lblk; | |
3546 | return false; | |
914f82a3 | 3547 | } |
914f82a3 | 3548 | |
09edf4d3 MB |
3549 | offset = map->m_lblk - es.es_lblk; |
3550 | map->m_len = es.es_len - offset; | |
914f82a3 | 3551 | |
09edf4d3 | 3552 | return true; |
4c0425ff MC |
3553 | } |
3554 | ||
09edf4d3 MB |
3555 | static int ext4_iomap_begin_report(struct inode *inode, loff_t offset, |
3556 | loff_t length, unsigned int flags, | |
3557 | struct iomap *iomap, struct iomap *srcmap) | |
4c0425ff | 3558 | { |
09edf4d3 MB |
3559 | int ret; |
3560 | bool delalloc = false; | |
3561 | struct ext4_map_blocks map; | |
3562 | u8 blkbits = inode->i_blkbits; | |
4c0425ff | 3563 | |
09edf4d3 MB |
3564 | if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK) |
3565 | return -EINVAL; | |
3566 | ||
3567 | if (ext4_has_inline_data(inode)) { | |
3568 | ret = ext4_inline_data_iomap(inode, iomap); | |
3569 | if (ret != -EAGAIN) { | |
3570 | if (ret == 0 && offset >= iomap->length) | |
3571 | ret = -ENOENT; | |
3572 | return ret; | |
3573 | } | |
3574 | } | |
2058f83a | 3575 | |
84ebd795 | 3576 | /* |
09edf4d3 | 3577 | * Calculate the first and last logical block respectively. |
84ebd795 | 3578 | */ |
09edf4d3 MB |
3579 | map.m_lblk = offset >> blkbits; |
3580 | map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits, | |
3581 | EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1; | |
84ebd795 | 3582 | |
b2c57642 RH |
3583 | /* |
3584 | * Fiemap callers may call for offset beyond s_bitmap_maxbytes. | |
3585 | * So handle it here itself instead of querying ext4_map_blocks(). | |
3586 | * Since ext4_map_blocks() will warn about it and will return | |
3587 | * -EIO error. | |
3588 | */ | |
3589 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) { | |
3590 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
3591 | ||
3592 | if (offset >= sbi->s_bitmap_maxbytes) { | |
3593 | map.m_flags = 0; | |
3594 | goto set_iomap; | |
3595 | } | |
3596 | } | |
3597 | ||
09edf4d3 MB |
3598 | ret = ext4_map_blocks(NULL, inode, &map, 0); |
3599 | if (ret < 0) | |
3600 | return ret; | |
3601 | if (ret == 0) | |
3602 | delalloc = ext4_iomap_is_delalloc(inode, &map); | |
46c7f254 | 3603 | |
b2c57642 | 3604 | set_iomap: |
de205114 | 3605 | ext4_set_iomap(inode, iomap, &map, offset, length, flags); |
09edf4d3 MB |
3606 | if (delalloc && iomap->type == IOMAP_HOLE) |
3607 | iomap->type = IOMAP_DELALLOC; | |
3608 | ||
3609 | return 0; | |
4c0425ff MC |
3610 | } |
3611 | ||
09edf4d3 MB |
3612 | const struct iomap_ops ext4_iomap_report_ops = { |
3613 | .iomap_begin = ext4_iomap_begin_report, | |
3614 | }; | |
3615 | ||
ac27a0ec | 3616 | /* |
6b1f86f8 LT |
3617 | * Whenever the folio is being dirtied, corresponding buffers should already |
3618 | * be attached to the transaction (we take care of this in ext4_page_mkwrite() | |
3619 | * and ext4_write_begin()). However we cannot move buffers to dirty transaction | |
3620 | * lists here because ->dirty_folio is called under VFS locks and the folio | |
2bb8dd40 | 3621 | * is not necessarily locked. |
ac27a0ec | 3622 | * |
187c82cb | 3623 | * We cannot just dirty the folio and leave attached buffers clean, because the |
ac27a0ec DK |
3624 | * buffers' dirty state is "definitive". We cannot just set the buffers dirty |
3625 | * or jbddirty because all the journalling code will explode. | |
3626 | * | |
187c82cb | 3627 | * So what we do is to mark the folio "pending dirty" and next time writepage |
ac27a0ec DK |
3628 | * is called, propagate that into the buffers appropriately. |
3629 | */ | |
187c82cb MWO |
3630 | static bool ext4_journalled_dirty_folio(struct address_space *mapping, |
3631 | struct folio *folio) | |
ac27a0ec | 3632 | { |
0f252336 | 3633 | WARN_ON_ONCE(!folio_buffers(folio)); |
187c82cb MWO |
3634 | folio_set_checked(folio); |
3635 | return filemap_dirty_folio(mapping, folio); | |
ac27a0ec DK |
3636 | } |
3637 | ||
e621900a | 3638 | static bool ext4_dirty_folio(struct address_space *mapping, struct folio *folio) |
6dcc693b | 3639 | { |
e621900a MWO |
3640 | WARN_ON_ONCE(!folio_test_locked(folio) && !folio_test_dirty(folio)); |
3641 | WARN_ON_ONCE(!folio_buffers(folio)); | |
3642 | return block_dirty_folio(mapping, folio); | |
6dcc693b JK |
3643 | } |
3644 | ||
0e6895ba RH |
3645 | static int ext4_iomap_swap_activate(struct swap_info_struct *sis, |
3646 | struct file *file, sector_t *span) | |
3647 | { | |
3648 | return iomap_swapfile_activate(sis, file, span, | |
3649 | &ext4_iomap_report_ops); | |
3650 | } | |
3651 | ||
74d553aa | 3652 | static const struct address_space_operations ext4_aops = { |
fe5ddf6b | 3653 | .read_folio = ext4_read_folio, |
6311f91f | 3654 | .readahead = ext4_readahead, |
43ce1d23 | 3655 | .writepage = ext4_writepage, |
20970ba6 | 3656 | .writepages = ext4_writepages, |
8ab22b9a | 3657 | .write_begin = ext4_write_begin, |
74d553aa | 3658 | .write_end = ext4_write_end, |
e621900a | 3659 | .dirty_folio = ext4_dirty_folio, |
8ab22b9a | 3660 | .bmap = ext4_bmap, |
7ba13abb | 3661 | .invalidate_folio = ext4_invalidate_folio, |
3c402f15 | 3662 | .release_folio = ext4_release_folio, |
378f32ba | 3663 | .direct_IO = noop_direct_IO, |
67235182 | 3664 | .migrate_folio = buffer_migrate_folio, |
8ab22b9a | 3665 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 3666 | .error_remove_page = generic_error_remove_page, |
0e6895ba | 3667 | .swap_activate = ext4_iomap_swap_activate, |
ac27a0ec DK |
3668 | }; |
3669 | ||
617ba13b | 3670 | static const struct address_space_operations ext4_journalled_aops = { |
fe5ddf6b | 3671 | .read_folio = ext4_read_folio, |
6311f91f | 3672 | .readahead = ext4_readahead, |
43ce1d23 | 3673 | .writepage = ext4_writepage, |
20970ba6 | 3674 | .writepages = ext4_writepages, |
8ab22b9a HH |
3675 | .write_begin = ext4_write_begin, |
3676 | .write_end = ext4_journalled_write_end, | |
187c82cb | 3677 | .dirty_folio = ext4_journalled_dirty_folio, |
8ab22b9a | 3678 | .bmap = ext4_bmap, |
ccd16945 | 3679 | .invalidate_folio = ext4_journalled_invalidate_folio, |
3c402f15 | 3680 | .release_folio = ext4_release_folio, |
378f32ba | 3681 | .direct_IO = noop_direct_IO, |
8ab22b9a | 3682 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 3683 | .error_remove_page = generic_error_remove_page, |
0e6895ba | 3684 | .swap_activate = ext4_iomap_swap_activate, |
ac27a0ec DK |
3685 | }; |
3686 | ||
64769240 | 3687 | static const struct address_space_operations ext4_da_aops = { |
fe5ddf6b | 3688 | .read_folio = ext4_read_folio, |
6311f91f | 3689 | .readahead = ext4_readahead, |
43ce1d23 | 3690 | .writepage = ext4_writepage, |
20970ba6 | 3691 | .writepages = ext4_writepages, |
8ab22b9a HH |
3692 | .write_begin = ext4_da_write_begin, |
3693 | .write_end = ext4_da_write_end, | |
e621900a | 3694 | .dirty_folio = ext4_dirty_folio, |
8ab22b9a | 3695 | .bmap = ext4_bmap, |
7ba13abb | 3696 | .invalidate_folio = ext4_invalidate_folio, |
3c402f15 | 3697 | .release_folio = ext4_release_folio, |
378f32ba | 3698 | .direct_IO = noop_direct_IO, |
67235182 | 3699 | .migrate_folio = buffer_migrate_folio, |
8ab22b9a | 3700 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 3701 | .error_remove_page = generic_error_remove_page, |
0e6895ba | 3702 | .swap_activate = ext4_iomap_swap_activate, |
64769240 AT |
3703 | }; |
3704 | ||
5f0663bb DW |
3705 | static const struct address_space_operations ext4_dax_aops = { |
3706 | .writepages = ext4_dax_writepages, | |
3707 | .direct_IO = noop_direct_IO, | |
46de8b97 | 3708 | .dirty_folio = noop_dirty_folio, |
94dbb631 | 3709 | .bmap = ext4_bmap, |
0e6895ba | 3710 | .swap_activate = ext4_iomap_swap_activate, |
5f0663bb DW |
3711 | }; |
3712 | ||
617ba13b | 3713 | void ext4_set_aops(struct inode *inode) |
ac27a0ec | 3714 | { |
3d2b1582 LC |
3715 | switch (ext4_inode_journal_mode(inode)) { |
3716 | case EXT4_INODE_ORDERED_DATA_MODE: | |
3d2b1582 | 3717 | case EXT4_INODE_WRITEBACK_DATA_MODE: |
3d2b1582 LC |
3718 | break; |
3719 | case EXT4_INODE_JOURNAL_DATA_MODE: | |
617ba13b | 3720 | inode->i_mapping->a_ops = &ext4_journalled_aops; |
74d553aa | 3721 | return; |
3d2b1582 LC |
3722 | default: |
3723 | BUG(); | |
3724 | } | |
5f0663bb DW |
3725 | if (IS_DAX(inode)) |
3726 | inode->i_mapping->a_ops = &ext4_dax_aops; | |
3727 | else if (test_opt(inode->i_sb, DELALLOC)) | |
74d553aa TT |
3728 | inode->i_mapping->a_ops = &ext4_da_aops; |
3729 | else | |
3730 | inode->i_mapping->a_ops = &ext4_aops; | |
ac27a0ec DK |
3731 | } |
3732 | ||
923ae0ff | 3733 | static int __ext4_block_zero_page_range(handle_t *handle, |
d863dc36 LC |
3734 | struct address_space *mapping, loff_t from, loff_t length) |
3735 | { | |
09cbfeaf KS |
3736 | ext4_fsblk_t index = from >> PAGE_SHIFT; |
3737 | unsigned offset = from & (PAGE_SIZE-1); | |
923ae0ff | 3738 | unsigned blocksize, pos; |
d863dc36 LC |
3739 | ext4_lblk_t iblock; |
3740 | struct inode *inode = mapping->host; | |
3741 | struct buffer_head *bh; | |
3742 | struct page *page; | |
3743 | int err = 0; | |
3744 | ||
09cbfeaf | 3745 | page = find_or_create_page(mapping, from >> PAGE_SHIFT, |
c62d2555 | 3746 | mapping_gfp_constraint(mapping, ~__GFP_FS)); |
d863dc36 LC |
3747 | if (!page) |
3748 | return -ENOMEM; | |
3749 | ||
3750 | blocksize = inode->i_sb->s_blocksize; | |
d863dc36 | 3751 | |
09cbfeaf | 3752 | iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits); |
d863dc36 LC |
3753 | |
3754 | if (!page_has_buffers(page)) | |
3755 | create_empty_buffers(page, blocksize, 0); | |
3756 | ||
3757 | /* Find the buffer that contains "offset" */ | |
3758 | bh = page_buffers(page); | |
3759 | pos = blocksize; | |
3760 | while (offset >= pos) { | |
3761 | bh = bh->b_this_page; | |
3762 | iblock++; | |
3763 | pos += blocksize; | |
3764 | } | |
d863dc36 LC |
3765 | if (buffer_freed(bh)) { |
3766 | BUFFER_TRACE(bh, "freed: skip"); | |
3767 | goto unlock; | |
3768 | } | |
d863dc36 LC |
3769 | if (!buffer_mapped(bh)) { |
3770 | BUFFER_TRACE(bh, "unmapped"); | |
3771 | ext4_get_block(inode, iblock, bh, 0); | |
3772 | /* unmapped? It's a hole - nothing to do */ | |
3773 | if (!buffer_mapped(bh)) { | |
3774 | BUFFER_TRACE(bh, "still unmapped"); | |
3775 | goto unlock; | |
3776 | } | |
3777 | } | |
3778 | ||
3779 | /* Ok, it's mapped. Make sure it's up-to-date */ | |
3780 | if (PageUptodate(page)) | |
3781 | set_buffer_uptodate(bh); | |
3782 | ||
3783 | if (!buffer_uptodate(bh)) { | |
2d069c08 | 3784 | err = ext4_read_bh_lock(bh, 0, true); |
3785 | if (err) | |
d863dc36 | 3786 | goto unlock; |
4f74d15f | 3787 | if (fscrypt_inode_uses_fs_layer_crypto(inode)) { |
c9c7429c | 3788 | /* We expect the key to be set. */ |
a7550b30 | 3789 | BUG_ON(!fscrypt_has_encryption_key(inode)); |
834f1565 EB |
3790 | err = fscrypt_decrypt_pagecache_blocks(page, blocksize, |
3791 | bh_offset(bh)); | |
3792 | if (err) { | |
3793 | clear_buffer_uptodate(bh); | |
3794 | goto unlock; | |
3795 | } | |
c9c7429c | 3796 | } |
d863dc36 | 3797 | } |
d863dc36 LC |
3798 | if (ext4_should_journal_data(inode)) { |
3799 | BUFFER_TRACE(bh, "get write access"); | |
188c299e JK |
3800 | err = ext4_journal_get_write_access(handle, inode->i_sb, bh, |
3801 | EXT4_JTR_NONE); | |
d863dc36 LC |
3802 | if (err) |
3803 | goto unlock; | |
3804 | } | |
d863dc36 | 3805 | zero_user(page, offset, length); |
d863dc36 LC |
3806 | BUFFER_TRACE(bh, "zeroed end of block"); |
3807 | ||
d863dc36 LC |
3808 | if (ext4_should_journal_data(inode)) { |
3809 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
0713ed0c | 3810 | } else { |
353eefd3 | 3811 | err = 0; |
d863dc36 | 3812 | mark_buffer_dirty(bh); |
3957ef53 | 3813 | if (ext4_should_order_data(inode)) |
73131fbb RZ |
3814 | err = ext4_jbd2_inode_add_write(handle, inode, from, |
3815 | length); | |
0713ed0c | 3816 | } |
d863dc36 LC |
3817 | |
3818 | unlock: | |
3819 | unlock_page(page); | |
09cbfeaf | 3820 | put_page(page); |
d863dc36 LC |
3821 | return err; |
3822 | } | |
3823 | ||
923ae0ff RZ |
3824 | /* |
3825 | * ext4_block_zero_page_range() zeros out a mapping of length 'length' | |
3826 | * starting from file offset 'from'. The range to be zero'd must | |
3827 | * be contained with in one block. If the specified range exceeds | |
3828 | * the end of the block it will be shortened to end of the block | |
3088e5a5 | 3829 | * that corresponds to 'from' |
923ae0ff RZ |
3830 | */ |
3831 | static int ext4_block_zero_page_range(handle_t *handle, | |
3832 | struct address_space *mapping, loff_t from, loff_t length) | |
3833 | { | |
3834 | struct inode *inode = mapping->host; | |
09cbfeaf | 3835 | unsigned offset = from & (PAGE_SIZE-1); |
923ae0ff RZ |
3836 | unsigned blocksize = inode->i_sb->s_blocksize; |
3837 | unsigned max = blocksize - (offset & (blocksize - 1)); | |
3838 | ||
3839 | /* | |
3840 | * correct length if it does not fall between | |
3841 | * 'from' and the end of the block | |
3842 | */ | |
3843 | if (length > max || length < 0) | |
3844 | length = max; | |
3845 | ||
47e69351 | 3846 | if (IS_DAX(inode)) { |
c6f40468 CH |
3847 | return dax_zero_range(inode, from, length, NULL, |
3848 | &ext4_iomap_ops); | |
47e69351 | 3849 | } |
923ae0ff RZ |
3850 | return __ext4_block_zero_page_range(handle, mapping, from, length); |
3851 | } | |
3852 | ||
94350ab5 MW |
3853 | /* |
3854 | * ext4_block_truncate_page() zeroes out a mapping from file offset `from' | |
3855 | * up to the end of the block which corresponds to `from'. | |
3856 | * This required during truncate. We need to physically zero the tail end | |
3857 | * of that block so it doesn't yield old data if the file is later grown. | |
3858 | */ | |
c197855e | 3859 | static int ext4_block_truncate_page(handle_t *handle, |
94350ab5 MW |
3860 | struct address_space *mapping, loff_t from) |
3861 | { | |
09cbfeaf | 3862 | unsigned offset = from & (PAGE_SIZE-1); |
94350ab5 MW |
3863 | unsigned length; |
3864 | unsigned blocksize; | |
3865 | struct inode *inode = mapping->host; | |
3866 | ||
0d06863f | 3867 | /* If we are processing an encrypted inode during orphan list handling */ |
592ddec7 | 3868 | if (IS_ENCRYPTED(inode) && !fscrypt_has_encryption_key(inode)) |
0d06863f TT |
3869 | return 0; |
3870 | ||
94350ab5 MW |
3871 | blocksize = inode->i_sb->s_blocksize; |
3872 | length = blocksize - (offset & (blocksize - 1)); | |
3873 | ||
3874 | return ext4_block_zero_page_range(handle, mapping, from, length); | |
3875 | } | |
3876 | ||
a87dd18c LC |
3877 | int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode, |
3878 | loff_t lstart, loff_t length) | |
3879 | { | |
3880 | struct super_block *sb = inode->i_sb; | |
3881 | struct address_space *mapping = inode->i_mapping; | |
e1be3a92 | 3882 | unsigned partial_start, partial_end; |
a87dd18c LC |
3883 | ext4_fsblk_t start, end; |
3884 | loff_t byte_end = (lstart + length - 1); | |
3885 | int err = 0; | |
3886 | ||
e1be3a92 LC |
3887 | partial_start = lstart & (sb->s_blocksize - 1); |
3888 | partial_end = byte_end & (sb->s_blocksize - 1); | |
3889 | ||
a87dd18c LC |
3890 | start = lstart >> sb->s_blocksize_bits; |
3891 | end = byte_end >> sb->s_blocksize_bits; | |
3892 | ||
3893 | /* Handle partial zero within the single block */ | |
e1be3a92 LC |
3894 | if (start == end && |
3895 | (partial_start || (partial_end != sb->s_blocksize - 1))) { | |
a87dd18c LC |
3896 | err = ext4_block_zero_page_range(handle, mapping, |
3897 | lstart, length); | |
3898 | return err; | |
3899 | } | |
3900 | /* Handle partial zero out on the start of the range */ | |
e1be3a92 | 3901 | if (partial_start) { |
a87dd18c LC |
3902 | err = ext4_block_zero_page_range(handle, mapping, |
3903 | lstart, sb->s_blocksize); | |
3904 | if (err) | |
3905 | return err; | |
3906 | } | |
3907 | /* Handle partial zero out on the end of the range */ | |
e1be3a92 | 3908 | if (partial_end != sb->s_blocksize - 1) |
a87dd18c | 3909 | err = ext4_block_zero_page_range(handle, mapping, |
e1be3a92 LC |
3910 | byte_end - partial_end, |
3911 | partial_end + 1); | |
a87dd18c LC |
3912 | return err; |
3913 | } | |
3914 | ||
91ef4caf DG |
3915 | int ext4_can_truncate(struct inode *inode) |
3916 | { | |
91ef4caf DG |
3917 | if (S_ISREG(inode->i_mode)) |
3918 | return 1; | |
3919 | if (S_ISDIR(inode->i_mode)) | |
3920 | return 1; | |
3921 | if (S_ISLNK(inode->i_mode)) | |
3922 | return !ext4_inode_is_fast_symlink(inode); | |
3923 | return 0; | |
3924 | } | |
3925 | ||
01127848 JK |
3926 | /* |
3927 | * We have to make sure i_disksize gets properly updated before we truncate | |
3928 | * page cache due to hole punching or zero range. Otherwise i_disksize update | |
3929 | * can get lost as it may have been postponed to submission of writeback but | |
3930 | * that will never happen after we truncate page cache. | |
3931 | */ | |
3932 | int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset, | |
3933 | loff_t len) | |
3934 | { | |
3935 | handle_t *handle; | |
4209ae12 HS |
3936 | int ret; |
3937 | ||
01127848 JK |
3938 | loff_t size = i_size_read(inode); |
3939 | ||
5955102c | 3940 | WARN_ON(!inode_is_locked(inode)); |
01127848 JK |
3941 | if (offset > size || offset + len < size) |
3942 | return 0; | |
3943 | ||
3944 | if (EXT4_I(inode)->i_disksize >= size) | |
3945 | return 0; | |
3946 | ||
3947 | handle = ext4_journal_start(inode, EXT4_HT_MISC, 1); | |
3948 | if (IS_ERR(handle)) | |
3949 | return PTR_ERR(handle); | |
3950 | ext4_update_i_disksize(inode, size); | |
4209ae12 | 3951 | ret = ext4_mark_inode_dirty(handle, inode); |
01127848 JK |
3952 | ext4_journal_stop(handle); |
3953 | ||
4209ae12 | 3954 | return ret; |
01127848 JK |
3955 | } |
3956 | ||
d4f5258e | 3957 | static void ext4_wait_dax_page(struct inode *inode) |
430657b6 | 3958 | { |
d4f5258e | 3959 | filemap_invalidate_unlock(inode->i_mapping); |
430657b6 | 3960 | schedule(); |
d4f5258e | 3961 | filemap_invalidate_lock(inode->i_mapping); |
430657b6 RZ |
3962 | } |
3963 | ||
3964 | int ext4_break_layouts(struct inode *inode) | |
3965 | { | |
430657b6 | 3966 | struct page *page; |
430657b6 RZ |
3967 | int error; |
3968 | ||
d4f5258e | 3969 | if (WARN_ON_ONCE(!rwsem_is_locked(&inode->i_mapping->invalidate_lock))) |
430657b6 RZ |
3970 | return -EINVAL; |
3971 | ||
3972 | do { | |
430657b6 RZ |
3973 | page = dax_layout_busy_page(inode->i_mapping); |
3974 | if (!page) | |
3975 | return 0; | |
3976 | ||
3977 | error = ___wait_var_event(&page->_refcount, | |
3978 | atomic_read(&page->_refcount) == 1, | |
3979 | TASK_INTERRUPTIBLE, 0, 0, | |
d4f5258e | 3980 | ext4_wait_dax_page(inode)); |
b1f38217 | 3981 | } while (error == 0); |
430657b6 RZ |
3982 | |
3983 | return error; | |
3984 | } | |
3985 | ||
a4bb6b64 | 3986 | /* |
cca32b7e | 3987 | * ext4_punch_hole: punches a hole in a file by releasing the blocks |
a4bb6b64 AH |
3988 | * associated with the given offset and length |
3989 | * | |
3990 | * @inode: File inode | |
3991 | * @offset: The offset where the hole will begin | |
3992 | * @len: The length of the hole | |
3993 | * | |
4907cb7b | 3994 | * Returns: 0 on success or negative on failure |
a4bb6b64 AH |
3995 | */ |
3996 | ||
ad5cd4f4 | 3997 | int ext4_punch_hole(struct file *file, loff_t offset, loff_t length) |
a4bb6b64 | 3998 | { |
ad5cd4f4 | 3999 | struct inode *inode = file_inode(file); |
26a4c0c6 TT |
4000 | struct super_block *sb = inode->i_sb; |
4001 | ext4_lblk_t first_block, stop_block; | |
4002 | struct address_space *mapping = inode->i_mapping; | |
2da37622 TS |
4003 | loff_t first_block_offset, last_block_offset, max_length; |
4004 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
26a4c0c6 TT |
4005 | handle_t *handle; |
4006 | unsigned int credits; | |
4209ae12 | 4007 | int ret = 0, ret2 = 0; |
26a4c0c6 | 4008 | |
b8a86845 | 4009 | trace_ext4_punch_hole(inode, offset, length, 0); |
aaddea81 | 4010 | |
26a4c0c6 TT |
4011 | /* |
4012 | * Write out all dirty pages to avoid race conditions | |
4013 | * Then release them. | |
4014 | */ | |
cca32b7e | 4015 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { |
26a4c0c6 TT |
4016 | ret = filemap_write_and_wait_range(mapping, offset, |
4017 | offset + length - 1); | |
4018 | if (ret) | |
4019 | return ret; | |
4020 | } | |
4021 | ||
5955102c | 4022 | inode_lock(inode); |
9ef06cec | 4023 | |
26a4c0c6 TT |
4024 | /* No need to punch hole beyond i_size */ |
4025 | if (offset >= inode->i_size) | |
4026 | goto out_mutex; | |
4027 | ||
4028 | /* | |
4029 | * If the hole extends beyond i_size, set the hole | |
4030 | * to end after the page that contains i_size | |
4031 | */ | |
4032 | if (offset + length > inode->i_size) { | |
4033 | length = inode->i_size + | |
09cbfeaf | 4034 | PAGE_SIZE - (inode->i_size & (PAGE_SIZE - 1)) - |
26a4c0c6 TT |
4035 | offset; |
4036 | } | |
4037 | ||
2da37622 TS |
4038 | /* |
4039 | * For punch hole the length + offset needs to be within one block | |
4040 | * before last range. Adjust the length if it goes beyond that limit. | |
4041 | */ | |
4042 | max_length = sbi->s_bitmap_maxbytes - inode->i_sb->s_blocksize; | |
4043 | if (offset + length > max_length) | |
4044 | length = max_length - offset; | |
4045 | ||
a361293f JK |
4046 | if (offset & (sb->s_blocksize - 1) || |
4047 | (offset + length) & (sb->s_blocksize - 1)) { | |
4048 | /* | |
4049 | * Attach jinode to inode for jbd2 if we do any zeroing of | |
4050 | * partial block | |
4051 | */ | |
4052 | ret = ext4_inode_attach_jinode(inode); | |
4053 | if (ret < 0) | |
4054 | goto out_mutex; | |
4055 | ||
4056 | } | |
4057 | ||
f340b3d9 | 4058 | /* Wait all existing dio workers, newcomers will block on i_rwsem */ |
ea3d7209 JK |
4059 | inode_dio_wait(inode); |
4060 | ||
ad5cd4f4 DW |
4061 | ret = file_modified(file); |
4062 | if (ret) | |
4063 | goto out_mutex; | |
4064 | ||
ea3d7209 JK |
4065 | /* |
4066 | * Prevent page faults from reinstantiating pages we have released from | |
4067 | * page cache. | |
4068 | */ | |
d4f5258e | 4069 | filemap_invalidate_lock(mapping); |
430657b6 RZ |
4070 | |
4071 | ret = ext4_break_layouts(inode); | |
4072 | if (ret) | |
4073 | goto out_dio; | |
4074 | ||
a87dd18c LC |
4075 | first_block_offset = round_up(offset, sb->s_blocksize); |
4076 | last_block_offset = round_down((offset + length), sb->s_blocksize) - 1; | |
26a4c0c6 | 4077 | |
a87dd18c | 4078 | /* Now release the pages and zero block aligned part of pages*/ |
01127848 JK |
4079 | if (last_block_offset > first_block_offset) { |
4080 | ret = ext4_update_disksize_before_punch(inode, offset, length); | |
4081 | if (ret) | |
4082 | goto out_dio; | |
a87dd18c LC |
4083 | truncate_pagecache_range(inode, first_block_offset, |
4084 | last_block_offset); | |
01127848 | 4085 | } |
26a4c0c6 TT |
4086 | |
4087 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
4088 | credits = ext4_writepage_trans_blocks(inode); | |
4089 | else | |
4090 | credits = ext4_blocks_for_truncate(inode); | |
4091 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); | |
4092 | if (IS_ERR(handle)) { | |
4093 | ret = PTR_ERR(handle); | |
4094 | ext4_std_error(sb, ret); | |
4095 | goto out_dio; | |
4096 | } | |
4097 | ||
a87dd18c LC |
4098 | ret = ext4_zero_partial_blocks(handle, inode, offset, |
4099 | length); | |
4100 | if (ret) | |
4101 | goto out_stop; | |
26a4c0c6 TT |
4102 | |
4103 | first_block = (offset + sb->s_blocksize - 1) >> | |
4104 | EXT4_BLOCK_SIZE_BITS(sb); | |
4105 | stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb); | |
4106 | ||
eee597ac LC |
4107 | /* If there are blocks to remove, do it */ |
4108 | if (stop_block > first_block) { | |
26a4c0c6 | 4109 | |
eee597ac | 4110 | down_write(&EXT4_I(inode)->i_data_sem); |
27bc446e | 4111 | ext4_discard_preallocations(inode, 0); |
26a4c0c6 | 4112 | |
eee597ac LC |
4113 | ret = ext4_es_remove_extent(inode, first_block, |
4114 | stop_block - first_block); | |
4115 | if (ret) { | |
4116 | up_write(&EXT4_I(inode)->i_data_sem); | |
4117 | goto out_stop; | |
4118 | } | |
26a4c0c6 | 4119 | |
eee597ac LC |
4120 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
4121 | ret = ext4_ext_remove_space(inode, first_block, | |
4122 | stop_block - 1); | |
4123 | else | |
4124 | ret = ext4_ind_remove_space(handle, inode, first_block, | |
4125 | stop_block); | |
26a4c0c6 | 4126 | |
eee597ac LC |
4127 | up_write(&EXT4_I(inode)->i_data_sem); |
4128 | } | |
a80f7fcf | 4129 | ext4_fc_track_range(handle, inode, first_block, stop_block); |
26a4c0c6 TT |
4130 | if (IS_SYNC(inode)) |
4131 | ext4_handle_sync(handle); | |
e251f9bc | 4132 | |
eeca7ea1 | 4133 | inode->i_mtime = inode->i_ctime = current_time(inode); |
4209ae12 HS |
4134 | ret2 = ext4_mark_inode_dirty(handle, inode); |
4135 | if (unlikely(ret2)) | |
4136 | ret = ret2; | |
67a7d5f5 JK |
4137 | if (ret >= 0) |
4138 | ext4_update_inode_fsync_trans(handle, inode, 1); | |
26a4c0c6 TT |
4139 | out_stop: |
4140 | ext4_journal_stop(handle); | |
4141 | out_dio: | |
d4f5258e | 4142 | filemap_invalidate_unlock(mapping); |
26a4c0c6 | 4143 | out_mutex: |
5955102c | 4144 | inode_unlock(inode); |
26a4c0c6 | 4145 | return ret; |
a4bb6b64 AH |
4146 | } |
4147 | ||
a361293f JK |
4148 | int ext4_inode_attach_jinode(struct inode *inode) |
4149 | { | |
4150 | struct ext4_inode_info *ei = EXT4_I(inode); | |
4151 | struct jbd2_inode *jinode; | |
4152 | ||
4153 | if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal) | |
4154 | return 0; | |
4155 | ||
4156 | jinode = jbd2_alloc_inode(GFP_KERNEL); | |
4157 | spin_lock(&inode->i_lock); | |
4158 | if (!ei->jinode) { | |
4159 | if (!jinode) { | |
4160 | spin_unlock(&inode->i_lock); | |
4161 | return -ENOMEM; | |
4162 | } | |
4163 | ei->jinode = jinode; | |
4164 | jbd2_journal_init_jbd_inode(ei->jinode, inode); | |
4165 | jinode = NULL; | |
4166 | } | |
4167 | spin_unlock(&inode->i_lock); | |
4168 | if (unlikely(jinode != NULL)) | |
4169 | jbd2_free_inode(jinode); | |
4170 | return 0; | |
4171 | } | |
4172 | ||
ac27a0ec | 4173 | /* |
617ba13b | 4174 | * ext4_truncate() |
ac27a0ec | 4175 | * |
617ba13b MC |
4176 | * We block out ext4_get_block() block instantiations across the entire |
4177 | * transaction, and VFS/VM ensures that ext4_truncate() cannot run | |
ac27a0ec DK |
4178 | * simultaneously on behalf of the same inode. |
4179 | * | |
42b2aa86 | 4180 | * As we work through the truncate and commit bits of it to the journal there |
ac27a0ec DK |
4181 | * is one core, guiding principle: the file's tree must always be consistent on |
4182 | * disk. We must be able to restart the truncate after a crash. | |
4183 | * | |
4184 | * The file's tree may be transiently inconsistent in memory (although it | |
4185 | * probably isn't), but whenever we close off and commit a journal transaction, | |
4186 | * the contents of (the filesystem + the journal) must be consistent and | |
4187 | * restartable. It's pretty simple, really: bottom up, right to left (although | |
4188 | * left-to-right works OK too). | |
4189 | * | |
4190 | * Note that at recovery time, journal replay occurs *before* the restart of | |
4191 | * truncate against the orphan inode list. | |
4192 | * | |
4193 | * The committed inode has the new, desired i_size (which is the same as | |
617ba13b | 4194 | * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see |
ac27a0ec | 4195 | * that this inode's truncate did not complete and it will again call |
617ba13b MC |
4196 | * ext4_truncate() to have another go. So there will be instantiated blocks |
4197 | * to the right of the truncation point in a crashed ext4 filesystem. But | |
ac27a0ec | 4198 | * that's fine - as long as they are linked from the inode, the post-crash |
617ba13b | 4199 | * ext4_truncate() run will find them and release them. |
ac27a0ec | 4200 | */ |
2c98eb5e | 4201 | int ext4_truncate(struct inode *inode) |
ac27a0ec | 4202 | { |
819c4920 TT |
4203 | struct ext4_inode_info *ei = EXT4_I(inode); |
4204 | unsigned int credits; | |
4209ae12 | 4205 | int err = 0, err2; |
819c4920 TT |
4206 | handle_t *handle; |
4207 | struct address_space *mapping = inode->i_mapping; | |
819c4920 | 4208 | |
19b5ef61 TT |
4209 | /* |
4210 | * There is a possibility that we're either freeing the inode | |
e04027e8 | 4211 | * or it's a completely new inode. In those cases we might not |
f340b3d9 | 4212 | * have i_rwsem locked because it's not necessary. |
19b5ef61 TT |
4213 | */ |
4214 | if (!(inode->i_state & (I_NEW|I_FREEING))) | |
5955102c | 4215 | WARN_ON(!inode_is_locked(inode)); |
0562e0ba JZ |
4216 | trace_ext4_truncate_enter(inode); |
4217 | ||
91ef4caf | 4218 | if (!ext4_can_truncate(inode)) |
9a5d265f | 4219 | goto out_trace; |
ac27a0ec | 4220 | |
5534fb5b | 4221 | if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC)) |
19f5fb7a | 4222 | ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE); |
7d8f9f7d | 4223 | |
aef1c851 TM |
4224 | if (ext4_has_inline_data(inode)) { |
4225 | int has_inline = 1; | |
4226 | ||
01daf945 | 4227 | err = ext4_inline_data_truncate(inode, &has_inline); |
9a5d265f | 4228 | if (err || has_inline) |
4229 | goto out_trace; | |
aef1c851 TM |
4230 | } |
4231 | ||
a361293f JK |
4232 | /* If we zero-out tail of the page, we have to create jinode for jbd2 */ |
4233 | if (inode->i_size & (inode->i_sb->s_blocksize - 1)) { | |
a71248b1 BL |
4234 | err = ext4_inode_attach_jinode(inode); |
4235 | if (err) | |
9a5d265f | 4236 | goto out_trace; |
a361293f JK |
4237 | } |
4238 | ||
819c4920 TT |
4239 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
4240 | credits = ext4_writepage_trans_blocks(inode); | |
4241 | else | |
4242 | credits = ext4_blocks_for_truncate(inode); | |
4243 | ||
4244 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); | |
9a5d265f | 4245 | if (IS_ERR(handle)) { |
4246 | err = PTR_ERR(handle); | |
4247 | goto out_trace; | |
4248 | } | |
819c4920 | 4249 | |
eb3544c6 LC |
4250 | if (inode->i_size & (inode->i_sb->s_blocksize - 1)) |
4251 | ext4_block_truncate_page(handle, mapping, inode->i_size); | |
819c4920 TT |
4252 | |
4253 | /* | |
4254 | * We add the inode to the orphan list, so that if this | |
4255 | * truncate spans multiple transactions, and we crash, we will | |
4256 | * resume the truncate when the filesystem recovers. It also | |
4257 | * marks the inode dirty, to catch the new size. | |
4258 | * | |
4259 | * Implication: the file must always be in a sane, consistent | |
4260 | * truncatable state while each transaction commits. | |
4261 | */ | |
2c98eb5e TT |
4262 | err = ext4_orphan_add(handle, inode); |
4263 | if (err) | |
819c4920 TT |
4264 | goto out_stop; |
4265 | ||
4266 | down_write(&EXT4_I(inode)->i_data_sem); | |
4267 | ||
27bc446e | 4268 | ext4_discard_preallocations(inode, 0); |
819c4920 | 4269 | |
ff9893dc | 4270 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
d0abb36d | 4271 | err = ext4_ext_truncate(handle, inode); |
ff9893dc | 4272 | else |
819c4920 TT |
4273 | ext4_ind_truncate(handle, inode); |
4274 | ||
4275 | up_write(&ei->i_data_sem); | |
d0abb36d TT |
4276 | if (err) |
4277 | goto out_stop; | |
819c4920 TT |
4278 | |
4279 | if (IS_SYNC(inode)) | |
4280 | ext4_handle_sync(handle); | |
4281 | ||
4282 | out_stop: | |
4283 | /* | |
4284 | * If this was a simple ftruncate() and the file will remain alive, | |
4285 | * then we need to clear up the orphan record which we created above. | |
4286 | * However, if this was a real unlink then we were called by | |
58d86a50 | 4287 | * ext4_evict_inode(), and we allow that function to clean up the |
819c4920 TT |
4288 | * orphan info for us. |
4289 | */ | |
4290 | if (inode->i_nlink) | |
4291 | ext4_orphan_del(handle, inode); | |
4292 | ||
eeca7ea1 | 4293 | inode->i_mtime = inode->i_ctime = current_time(inode); |
4209ae12 HS |
4294 | err2 = ext4_mark_inode_dirty(handle, inode); |
4295 | if (unlikely(err2 && !err)) | |
4296 | err = err2; | |
819c4920 | 4297 | ext4_journal_stop(handle); |
ac27a0ec | 4298 | |
9a5d265f | 4299 | out_trace: |
0562e0ba | 4300 | trace_ext4_truncate_exit(inode); |
2c98eb5e | 4301 | return err; |
ac27a0ec DK |
4302 | } |
4303 | ||
9a1bf32c ZY |
4304 | static inline u64 ext4_inode_peek_iversion(const struct inode *inode) |
4305 | { | |
4306 | if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) | |
4307 | return inode_peek_iversion_raw(inode); | |
4308 | else | |
4309 | return inode_peek_iversion(inode); | |
4310 | } | |
4311 | ||
4312 | static int ext4_inode_blocks_set(struct ext4_inode *raw_inode, | |
4313 | struct ext4_inode_info *ei) | |
4314 | { | |
4315 | struct inode *inode = &(ei->vfs_inode); | |
4316 | u64 i_blocks = READ_ONCE(inode->i_blocks); | |
4317 | struct super_block *sb = inode->i_sb; | |
4318 | ||
4319 | if (i_blocks <= ~0U) { | |
4320 | /* | |
4321 | * i_blocks can be represented in a 32 bit variable | |
4322 | * as multiple of 512 bytes | |
4323 | */ | |
4324 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); | |
4325 | raw_inode->i_blocks_high = 0; | |
4326 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); | |
4327 | return 0; | |
4328 | } | |
4329 | ||
4330 | /* | |
4331 | * This should never happen since sb->s_maxbytes should not have | |
4332 | * allowed this, sb->s_maxbytes was set according to the huge_file | |
4333 | * feature in ext4_fill_super(). | |
4334 | */ | |
4335 | if (!ext4_has_feature_huge_file(sb)) | |
4336 | return -EFSCORRUPTED; | |
4337 | ||
4338 | if (i_blocks <= 0xffffffffffffULL) { | |
4339 | /* | |
4340 | * i_blocks can be represented in a 48 bit variable | |
4341 | * as multiple of 512 bytes | |
4342 | */ | |
4343 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); | |
4344 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); | |
4345 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); | |
4346 | } else { | |
4347 | ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE); | |
4348 | /* i_block is stored in file system block size */ | |
4349 | i_blocks = i_blocks >> (inode->i_blkbits - 9); | |
4350 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); | |
4351 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); | |
4352 | } | |
4353 | return 0; | |
4354 | } | |
4355 | ||
4356 | static int ext4_fill_raw_inode(struct inode *inode, struct ext4_inode *raw_inode) | |
4357 | { | |
4358 | struct ext4_inode_info *ei = EXT4_I(inode); | |
4359 | uid_t i_uid; | |
4360 | gid_t i_gid; | |
4361 | projid_t i_projid; | |
4362 | int block; | |
4363 | int err; | |
4364 | ||
4365 | err = ext4_inode_blocks_set(raw_inode, ei); | |
4366 | ||
4367 | raw_inode->i_mode = cpu_to_le16(inode->i_mode); | |
4368 | i_uid = i_uid_read(inode); | |
4369 | i_gid = i_gid_read(inode); | |
4370 | i_projid = from_kprojid(&init_user_ns, ei->i_projid); | |
4371 | if (!(test_opt(inode->i_sb, NO_UID32))) { | |
4372 | raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid)); | |
4373 | raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid)); | |
4374 | /* | |
4375 | * Fix up interoperability with old kernels. Otherwise, | |
4376 | * old inodes get re-used with the upper 16 bits of the | |
4377 | * uid/gid intact. | |
4378 | */ | |
4379 | if (ei->i_dtime && list_empty(&ei->i_orphan)) { | |
4380 | raw_inode->i_uid_high = 0; | |
4381 | raw_inode->i_gid_high = 0; | |
4382 | } else { | |
4383 | raw_inode->i_uid_high = | |
4384 | cpu_to_le16(high_16_bits(i_uid)); | |
4385 | raw_inode->i_gid_high = | |
4386 | cpu_to_le16(high_16_bits(i_gid)); | |
4387 | } | |
4388 | } else { | |
4389 | raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid)); | |
4390 | raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid)); | |
4391 | raw_inode->i_uid_high = 0; | |
4392 | raw_inode->i_gid_high = 0; | |
4393 | } | |
4394 | raw_inode->i_links_count = cpu_to_le16(inode->i_nlink); | |
4395 | ||
4396 | EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode); | |
4397 | EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode); | |
4398 | EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode); | |
4399 | EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode); | |
4400 | ||
4401 | raw_inode->i_dtime = cpu_to_le32(ei->i_dtime); | |
4402 | raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF); | |
4403 | if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) | |
4404 | raw_inode->i_file_acl_high = | |
4405 | cpu_to_le16(ei->i_file_acl >> 32); | |
4406 | raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl); | |
4407 | ext4_isize_set(raw_inode, ei->i_disksize); | |
4408 | ||
4409 | raw_inode->i_generation = cpu_to_le32(inode->i_generation); | |
4410 | if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) { | |
4411 | if (old_valid_dev(inode->i_rdev)) { | |
4412 | raw_inode->i_block[0] = | |
4413 | cpu_to_le32(old_encode_dev(inode->i_rdev)); | |
4414 | raw_inode->i_block[1] = 0; | |
4415 | } else { | |
4416 | raw_inode->i_block[0] = 0; | |
4417 | raw_inode->i_block[1] = | |
4418 | cpu_to_le32(new_encode_dev(inode->i_rdev)); | |
4419 | raw_inode->i_block[2] = 0; | |
4420 | } | |
4421 | } else if (!ext4_has_inline_data(inode)) { | |
4422 | for (block = 0; block < EXT4_N_BLOCKS; block++) | |
4423 | raw_inode->i_block[block] = ei->i_data[block]; | |
4424 | } | |
4425 | ||
4426 | if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) { | |
4427 | u64 ivers = ext4_inode_peek_iversion(inode); | |
4428 | ||
4429 | raw_inode->i_disk_version = cpu_to_le32(ivers); | |
4430 | if (ei->i_extra_isize) { | |
4431 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
4432 | raw_inode->i_version_hi = | |
4433 | cpu_to_le32(ivers >> 32); | |
4434 | raw_inode->i_extra_isize = | |
4435 | cpu_to_le16(ei->i_extra_isize); | |
4436 | } | |
4437 | } | |
4438 | ||
4439 | if (i_projid != EXT4_DEF_PROJID && | |
4440 | !ext4_has_feature_project(inode->i_sb)) | |
4441 | err = err ?: -EFSCORRUPTED; | |
4442 | ||
4443 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && | |
4444 | EXT4_FITS_IN_INODE(raw_inode, ei, i_projid)) | |
4445 | raw_inode->i_projid = cpu_to_le32(i_projid); | |
4446 | ||
4447 | ext4_inode_csum_set(inode, raw_inode, ei); | |
4448 | return err; | |
4449 | } | |
4450 | ||
ac27a0ec | 4451 | /* |
617ba13b | 4452 | * ext4_get_inode_loc returns with an extra refcount against the inode's |
de01f484 ZY |
4453 | * underlying buffer_head on success. If we pass 'inode' and it does not |
4454 | * have in-inode xattr, we have all inode data in memory that is needed | |
4455 | * to recreate the on-disk version of this inode. | |
ac27a0ec | 4456 | */ |
8016e29f | 4457 | static int __ext4_get_inode_loc(struct super_block *sb, unsigned long ino, |
de01f484 | 4458 | struct inode *inode, struct ext4_iloc *iloc, |
8016e29f | 4459 | ext4_fsblk_t *ret_block) |
ac27a0ec | 4460 | { |
240799cd TT |
4461 | struct ext4_group_desc *gdp; |
4462 | struct buffer_head *bh; | |
240799cd | 4463 | ext4_fsblk_t block; |
02f03c42 | 4464 | struct blk_plug plug; |
240799cd TT |
4465 | int inodes_per_block, inode_offset; |
4466 | ||
3a06d778 | 4467 | iloc->bh = NULL; |
8016e29f HS |
4468 | if (ino < EXT4_ROOT_INO || |
4469 | ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count)) | |
6a797d27 | 4470 | return -EFSCORRUPTED; |
ac27a0ec | 4471 | |
8016e29f | 4472 | iloc->block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb); |
240799cd TT |
4473 | gdp = ext4_get_group_desc(sb, iloc->block_group, NULL); |
4474 | if (!gdp) | |
ac27a0ec DK |
4475 | return -EIO; |
4476 | ||
240799cd TT |
4477 | /* |
4478 | * Figure out the offset within the block group inode table | |
4479 | */ | |
00d09882 | 4480 | inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; |
8016e29f | 4481 | inode_offset = ((ino - 1) % |
240799cd | 4482 | EXT4_INODES_PER_GROUP(sb)); |
240799cd TT |
4483 | iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb); |
4484 | ||
eee22187 BL |
4485 | block = ext4_inode_table(sb, gdp); |
4486 | if ((block <= le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) || | |
4487 | (block >= ext4_blocks_count(EXT4_SB(sb)->s_es))) { | |
4488 | ext4_error(sb, "Invalid inode table block %llu in " | |
4489 | "block_group %u", block, iloc->block_group); | |
4490 | return -EFSCORRUPTED; | |
4491 | } | |
4492 | block += (inode_offset / inodes_per_block); | |
4493 | ||
240799cd | 4494 | bh = sb_getblk(sb, block); |
aebf0243 | 4495 | if (unlikely(!bh)) |
860d21e2 | 4496 | return -ENOMEM; |
8e33fadf ZY |
4497 | if (ext4_buffer_uptodate(bh)) |
4498 | goto has_buffer; | |
9c83a923 | 4499 | |
8e33fadf | 4500 | lock_buffer(bh); |
f2c77973 ZY |
4501 | if (ext4_buffer_uptodate(bh)) { |
4502 | /* Someone brought it uptodate while we waited */ | |
4503 | unlock_buffer(bh); | |
4504 | goto has_buffer; | |
4505 | } | |
4506 | ||
8e33fadf ZY |
4507 | /* |
4508 | * If we have all information of the inode in memory and this | |
4509 | * is the only valid inode in the block, we need not read the | |
4510 | * block. | |
4511 | */ | |
de01f484 | 4512 | if (inode && !ext4_test_inode_state(inode, EXT4_STATE_XATTR)) { |
8e33fadf ZY |
4513 | struct buffer_head *bitmap_bh; |
4514 | int i, start; | |
ac27a0ec | 4515 | |
8e33fadf | 4516 | start = inode_offset & ~(inodes_per_block - 1); |
ac27a0ec | 4517 | |
8e33fadf ZY |
4518 | /* Is the inode bitmap in cache? */ |
4519 | bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp)); | |
4520 | if (unlikely(!bitmap_bh)) | |
4521 | goto make_io; | |
ac27a0ec | 4522 | |
8e33fadf ZY |
4523 | /* |
4524 | * If the inode bitmap isn't in cache then the | |
4525 | * optimisation may end up performing two reads instead | |
4526 | * of one, so skip it. | |
4527 | */ | |
4528 | if (!buffer_uptodate(bitmap_bh)) { | |
ac27a0ec | 4529 | brelse(bitmap_bh); |
8e33fadf | 4530 | goto make_io; |
ac27a0ec | 4531 | } |
8e33fadf ZY |
4532 | for (i = start; i < start + inodes_per_block; i++) { |
4533 | if (i == inode_offset) | |
4534 | continue; | |
4535 | if (ext4_test_bit(i, bitmap_bh->b_data)) | |
4536 | break; | |
ac27a0ec | 4537 | } |
8e33fadf ZY |
4538 | brelse(bitmap_bh); |
4539 | if (i == start + inodes_per_block) { | |
de01f484 ZY |
4540 | struct ext4_inode *raw_inode = |
4541 | (struct ext4_inode *) (bh->b_data + iloc->offset); | |
4542 | ||
8e33fadf ZY |
4543 | /* all other inodes are free, so skip I/O */ |
4544 | memset(bh->b_data, 0, bh->b_size); | |
de01f484 ZY |
4545 | if (!ext4_test_inode_state(inode, EXT4_STATE_NEW)) |
4546 | ext4_fill_raw_inode(inode, raw_inode); | |
8e33fadf ZY |
4547 | set_buffer_uptodate(bh); |
4548 | unlock_buffer(bh); | |
4549 | goto has_buffer; | |
4550 | } | |
4551 | } | |
ac27a0ec DK |
4552 | |
4553 | make_io: | |
8e33fadf ZY |
4554 | /* |
4555 | * If we need to do any I/O, try to pre-readahead extra | |
4556 | * blocks from the inode table. | |
4557 | */ | |
4558 | blk_start_plug(&plug); | |
4559 | if (EXT4_SB(sb)->s_inode_readahead_blks) { | |
4560 | ext4_fsblk_t b, end, table; | |
4561 | unsigned num; | |
4562 | __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks; | |
4563 | ||
4564 | table = ext4_inode_table(sb, gdp); | |
4565 | /* s_inode_readahead_blks is always a power of 2 */ | |
4566 | b = block & ~((ext4_fsblk_t) ra_blks - 1); | |
4567 | if (table > b) | |
4568 | b = table; | |
4569 | end = b + ra_blks; | |
4570 | num = EXT4_INODES_PER_GROUP(sb); | |
4571 | if (ext4_has_group_desc_csum(sb)) | |
4572 | num -= ext4_itable_unused_count(sb, gdp); | |
4573 | table += num / inodes_per_block; | |
4574 | if (end > table) | |
4575 | end = table; | |
4576 | while (b <= end) | |
4577 | ext4_sb_breadahead_unmovable(sb, b++); | |
4578 | } | |
240799cd | 4579 | |
8e33fadf ZY |
4580 | /* |
4581 | * There are other valid inodes in the buffer, this inode | |
4582 | * has in-inode xattrs, or we don't have this inode in memory. | |
4583 | * Read the block from disk. | |
4584 | */ | |
4585 | trace_ext4_load_inode(sb, ino); | |
4586 | ext4_read_bh_nowait(bh, REQ_META | REQ_PRIO, NULL); | |
4587 | blk_finish_plug(&plug); | |
4588 | wait_on_buffer(bh); | |
4589 | ext4_simulate_fail_bh(sb, bh, EXT4_SIM_INODE_EIO); | |
4590 | if (!buffer_uptodate(bh)) { | |
4591 | if (ret_block) | |
4592 | *ret_block = block; | |
4593 | brelse(bh); | |
4594 | return -EIO; | |
ac27a0ec DK |
4595 | } |
4596 | has_buffer: | |
4597 | iloc->bh = bh; | |
4598 | return 0; | |
4599 | } | |
4600 | ||
8016e29f HS |
4601 | static int __ext4_get_inode_loc_noinmem(struct inode *inode, |
4602 | struct ext4_iloc *iloc) | |
4603 | { | |
c27c29c6 | 4604 | ext4_fsblk_t err_blk = 0; |
8016e29f HS |
4605 | int ret; |
4606 | ||
de01f484 | 4607 | ret = __ext4_get_inode_loc(inode->i_sb, inode->i_ino, NULL, iloc, |
8016e29f HS |
4608 | &err_blk); |
4609 | ||
4610 | if (ret == -EIO) | |
4611 | ext4_error_inode_block(inode, err_blk, EIO, | |
4612 | "unable to read itable block"); | |
4613 | ||
4614 | return ret; | |
4615 | } | |
4616 | ||
617ba13b | 4617 | int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec | 4618 | { |
c27c29c6 | 4619 | ext4_fsblk_t err_blk = 0; |
8016e29f HS |
4620 | int ret; |
4621 | ||
de01f484 ZY |
4622 | ret = __ext4_get_inode_loc(inode->i_sb, inode->i_ino, inode, iloc, |
4623 | &err_blk); | |
8016e29f HS |
4624 | |
4625 | if (ret == -EIO) | |
4626 | ext4_error_inode_block(inode, err_blk, EIO, | |
4627 | "unable to read itable block"); | |
4628 | ||
4629 | return ret; | |
4630 | } | |
4631 | ||
4632 | ||
4633 | int ext4_get_fc_inode_loc(struct super_block *sb, unsigned long ino, | |
4634 | struct ext4_iloc *iloc) | |
4635 | { | |
de01f484 | 4636 | return __ext4_get_inode_loc(sb, ino, NULL, iloc, NULL); |
ac27a0ec DK |
4637 | } |
4638 | ||
a8ab6d38 | 4639 | static bool ext4_should_enable_dax(struct inode *inode) |
6642586b | 4640 | { |
a8ab6d38 IW |
4641 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
4642 | ||
9cb20f94 | 4643 | if (test_opt2(inode->i_sb, DAX_NEVER)) |
6642586b RZ |
4644 | return false; |
4645 | if (!S_ISREG(inode->i_mode)) | |
4646 | return false; | |
4647 | if (ext4_should_journal_data(inode)) | |
4648 | return false; | |
4649 | if (ext4_has_inline_data(inode)) | |
4650 | return false; | |
592ddec7 | 4651 | if (ext4_test_inode_flag(inode, EXT4_INODE_ENCRYPT)) |
6642586b | 4652 | return false; |
c93d8f88 EB |
4653 | if (ext4_test_inode_flag(inode, EXT4_INODE_VERITY)) |
4654 | return false; | |
a8ab6d38 IW |
4655 | if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags)) |
4656 | return false; | |
4657 | if (test_opt(inode->i_sb, DAX_ALWAYS)) | |
4658 | return true; | |
4659 | ||
b383a73f | 4660 | return ext4_test_inode_flag(inode, EXT4_INODE_DAX); |
6642586b RZ |
4661 | } |
4662 | ||
043546e4 | 4663 | void ext4_set_inode_flags(struct inode *inode, bool init) |
ac27a0ec | 4664 | { |
617ba13b | 4665 | unsigned int flags = EXT4_I(inode)->i_flags; |
00a1a053 | 4666 | unsigned int new_fl = 0; |
ac27a0ec | 4667 | |
043546e4 IW |
4668 | WARN_ON_ONCE(IS_DAX(inode) && init); |
4669 | ||
617ba13b | 4670 | if (flags & EXT4_SYNC_FL) |
00a1a053 | 4671 | new_fl |= S_SYNC; |
617ba13b | 4672 | if (flags & EXT4_APPEND_FL) |
00a1a053 | 4673 | new_fl |= S_APPEND; |
617ba13b | 4674 | if (flags & EXT4_IMMUTABLE_FL) |
00a1a053 | 4675 | new_fl |= S_IMMUTABLE; |
617ba13b | 4676 | if (flags & EXT4_NOATIME_FL) |
00a1a053 | 4677 | new_fl |= S_NOATIME; |
617ba13b | 4678 | if (flags & EXT4_DIRSYNC_FL) |
00a1a053 | 4679 | new_fl |= S_DIRSYNC; |
043546e4 IW |
4680 | |
4681 | /* Because of the way inode_set_flags() works we must preserve S_DAX | |
4682 | * here if already set. */ | |
4683 | new_fl |= (inode->i_flags & S_DAX); | |
4684 | if (init && ext4_should_enable_dax(inode)) | |
923ae0ff | 4685 | new_fl |= S_DAX; |
043546e4 | 4686 | |
2ee6a576 EB |
4687 | if (flags & EXT4_ENCRYPT_FL) |
4688 | new_fl |= S_ENCRYPTED; | |
b886ee3e GKB |
4689 | if (flags & EXT4_CASEFOLD_FL) |
4690 | new_fl |= S_CASEFOLD; | |
c93d8f88 EB |
4691 | if (flags & EXT4_VERITY_FL) |
4692 | new_fl |= S_VERITY; | |
5f16f322 | 4693 | inode_set_flags(inode, new_fl, |
2ee6a576 | 4694 | S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX| |
c93d8f88 | 4695 | S_ENCRYPTED|S_CASEFOLD|S_VERITY); |
ac27a0ec DK |
4696 | } |
4697 | ||
0fc1b451 | 4698 | static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode, |
de9a55b8 | 4699 | struct ext4_inode_info *ei) |
0fc1b451 AK |
4700 | { |
4701 | blkcnt_t i_blocks ; | |
8180a562 AK |
4702 | struct inode *inode = &(ei->vfs_inode); |
4703 | struct super_block *sb = inode->i_sb; | |
0fc1b451 | 4704 | |
e2b911c5 | 4705 | if (ext4_has_feature_huge_file(sb)) { |
0fc1b451 AK |
4706 | /* we are using combined 48 bit field */ |
4707 | i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 | | |
4708 | le32_to_cpu(raw_inode->i_blocks_lo); | |
07a03824 | 4709 | if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) { |
8180a562 AK |
4710 | /* i_blocks represent file system block size */ |
4711 | return i_blocks << (inode->i_blkbits - 9); | |
4712 | } else { | |
4713 | return i_blocks; | |
4714 | } | |
0fc1b451 AK |
4715 | } else { |
4716 | return le32_to_cpu(raw_inode->i_blocks_lo); | |
4717 | } | |
4718 | } | |
ff9ddf7e | 4719 | |
eb9b5f01 | 4720 | static inline int ext4_iget_extra_inode(struct inode *inode, |
152a7b0a TM |
4721 | struct ext4_inode *raw_inode, |
4722 | struct ext4_inode_info *ei) | |
4723 | { | |
4724 | __le32 *magic = (void *)raw_inode + | |
4725 | EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize; | |
eb9b5f01 | 4726 | |
fd7e672e | 4727 | if (EXT4_INODE_HAS_XATTR_SPACE(inode) && |
290ab230 | 4728 | *magic == cpu_to_le32(EXT4_XATTR_MAGIC)) { |
152a7b0a | 4729 | ext4_set_inode_state(inode, EXT4_STATE_XATTR); |
eb9b5f01 | 4730 | return ext4_find_inline_data_nolock(inode); |
f19d5870 TM |
4731 | } else |
4732 | EXT4_I(inode)->i_inline_off = 0; | |
eb9b5f01 | 4733 | return 0; |
152a7b0a TM |
4734 | } |
4735 | ||
040cb378 LX |
4736 | int ext4_get_projid(struct inode *inode, kprojid_t *projid) |
4737 | { | |
0b7b7779 | 4738 | if (!ext4_has_feature_project(inode->i_sb)) |
040cb378 LX |
4739 | return -EOPNOTSUPP; |
4740 | *projid = EXT4_I(inode)->i_projid; | |
4741 | return 0; | |
4742 | } | |
4743 | ||
e254d1af EG |
4744 | /* |
4745 | * ext4 has self-managed i_version for ea inodes, it stores the lower 32bit of | |
4746 | * refcount in i_version, so use raw values if inode has EXT4_EA_INODE_FL flag | |
4747 | * set. | |
4748 | */ | |
4749 | static inline void ext4_inode_set_iversion_queried(struct inode *inode, u64 val) | |
4750 | { | |
4751 | if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) | |
4752 | inode_set_iversion_raw(inode, val); | |
4753 | else | |
4754 | inode_set_iversion_queried(inode, val); | |
4755 | } | |
e254d1af | 4756 | |
8a363970 TT |
4757 | struct inode *__ext4_iget(struct super_block *sb, unsigned long ino, |
4758 | ext4_iget_flags flags, const char *function, | |
4759 | unsigned int line) | |
ac27a0ec | 4760 | { |
617ba13b MC |
4761 | struct ext4_iloc iloc; |
4762 | struct ext4_inode *raw_inode; | |
1d1fe1ee | 4763 | struct ext4_inode_info *ei; |
bd2c38cf | 4764 | struct ext4_super_block *es = EXT4_SB(sb)->s_es; |
1d1fe1ee | 4765 | struct inode *inode; |
b436b9be | 4766 | journal_t *journal = EXT4_SB(sb)->s_journal; |
1d1fe1ee | 4767 | long ret; |
7e6e1ef4 | 4768 | loff_t size; |
ac27a0ec | 4769 | int block; |
08cefc7a EB |
4770 | uid_t i_uid; |
4771 | gid_t i_gid; | |
040cb378 | 4772 | projid_t i_projid; |
ac27a0ec | 4773 | |
191ce178 | 4774 | if ((!(flags & EXT4_IGET_SPECIAL) && |
bd2c38cf JK |
4775 | ((ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO) || |
4776 | ino == le32_to_cpu(es->s_usr_quota_inum) || | |
4777 | ino == le32_to_cpu(es->s_grp_quota_inum) || | |
02f310fc JK |
4778 | ino == le32_to_cpu(es->s_prj_quota_inum) || |
4779 | ino == le32_to_cpu(es->s_orphan_file_inum))) || | |
8a363970 | 4780 | (ino < EXT4_ROOT_INO) || |
bd2c38cf | 4781 | (ino > le32_to_cpu(es->s_inodes_count))) { |
8a363970 TT |
4782 | if (flags & EXT4_IGET_HANDLE) |
4783 | return ERR_PTR(-ESTALE); | |
014c9caa | 4784 | __ext4_error(sb, function, line, false, EFSCORRUPTED, 0, |
8a363970 TT |
4785 | "inode #%lu: comm %s: iget: illegal inode #", |
4786 | ino, current->comm); | |
4787 | return ERR_PTR(-EFSCORRUPTED); | |
4788 | } | |
4789 | ||
1d1fe1ee DH |
4790 | inode = iget_locked(sb, ino); |
4791 | if (!inode) | |
4792 | return ERR_PTR(-ENOMEM); | |
4793 | if (!(inode->i_state & I_NEW)) | |
4794 | return inode; | |
4795 | ||
4796 | ei = EXT4_I(inode); | |
7dc57615 | 4797 | iloc.bh = NULL; |
ac27a0ec | 4798 | |
8016e29f | 4799 | ret = __ext4_get_inode_loc_noinmem(inode, &iloc); |
1d1fe1ee | 4800 | if (ret < 0) |
ac27a0ec | 4801 | goto bad_inode; |
617ba13b | 4802 | raw_inode = ext4_raw_inode(&iloc); |
814525f4 | 4803 | |
8e4b5eae | 4804 | if ((ino == EXT4_ROOT_INO) && (raw_inode->i_links_count == 0)) { |
8a363970 TT |
4805 | ext4_error_inode(inode, function, line, 0, |
4806 | "iget: root inode unallocated"); | |
8e4b5eae TT |
4807 | ret = -EFSCORRUPTED; |
4808 | goto bad_inode; | |
4809 | } | |
4810 | ||
8a363970 TT |
4811 | if ((flags & EXT4_IGET_HANDLE) && |
4812 | (raw_inode->i_links_count == 0) && (raw_inode->i_mode == 0)) { | |
4813 | ret = -ESTALE; | |
4814 | goto bad_inode; | |
4815 | } | |
4816 | ||
814525f4 DW |
4817 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
4818 | ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize); | |
4819 | if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > | |
2dc8d9e1 EB |
4820 | EXT4_INODE_SIZE(inode->i_sb) || |
4821 | (ei->i_extra_isize & 3)) { | |
8a363970 TT |
4822 | ext4_error_inode(inode, function, line, 0, |
4823 | "iget: bad extra_isize %u " | |
4824 | "(inode size %u)", | |
2dc8d9e1 EB |
4825 | ei->i_extra_isize, |
4826 | EXT4_INODE_SIZE(inode->i_sb)); | |
6a797d27 | 4827 | ret = -EFSCORRUPTED; |
814525f4 DW |
4828 | goto bad_inode; |
4829 | } | |
4830 | } else | |
4831 | ei->i_extra_isize = 0; | |
4832 | ||
4833 | /* Precompute checksum seed for inode metadata */ | |
9aa5d32b | 4834 | if (ext4_has_metadata_csum(sb)) { |
814525f4 DW |
4835 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
4836 | __u32 csum; | |
4837 | __le32 inum = cpu_to_le32(inode->i_ino); | |
4838 | __le32 gen = raw_inode->i_generation; | |
4839 | csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum, | |
4840 | sizeof(inum)); | |
4841 | ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen, | |
4842 | sizeof(gen)); | |
4843 | } | |
4844 | ||
8016e29f HS |
4845 | if ((!ext4_inode_csum_verify(inode, raw_inode, ei) || |
4846 | ext4_simulate_fail(sb, EXT4_SIM_INODE_CRC)) && | |
4847 | (!(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY))) { | |
4848 | ext4_error_inode_err(inode, function, line, 0, | |
4849 | EFSBADCRC, "iget: checksum invalid"); | |
6a797d27 | 4850 | ret = -EFSBADCRC; |
814525f4 DW |
4851 | goto bad_inode; |
4852 | } | |
4853 | ||
ac27a0ec | 4854 | inode->i_mode = le16_to_cpu(raw_inode->i_mode); |
08cefc7a EB |
4855 | i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low); |
4856 | i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low); | |
0b7b7779 | 4857 | if (ext4_has_feature_project(sb) && |
040cb378 LX |
4858 | EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE && |
4859 | EXT4_FITS_IN_INODE(raw_inode, ei, i_projid)) | |
4860 | i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid); | |
4861 | else | |
4862 | i_projid = EXT4_DEF_PROJID; | |
4863 | ||
af5bc92d | 4864 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
08cefc7a EB |
4865 | i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16; |
4866 | i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16; | |
ac27a0ec | 4867 | } |
08cefc7a EB |
4868 | i_uid_write(inode, i_uid); |
4869 | i_gid_write(inode, i_gid); | |
040cb378 | 4870 | ei->i_projid = make_kprojid(&init_user_ns, i_projid); |
bfe86848 | 4871 | set_nlink(inode, le16_to_cpu(raw_inode->i_links_count)); |
ac27a0ec | 4872 | |
353eb83c | 4873 | ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */ |
67cf5b09 | 4874 | ei->i_inline_off = 0; |
ac27a0ec DK |
4875 | ei->i_dir_start_lookup = 0; |
4876 | ei->i_dtime = le32_to_cpu(raw_inode->i_dtime); | |
4877 | /* We now have enough fields to check if the inode was active or not. | |
4878 | * This is needed because nfsd might try to access dead inodes | |
4879 | * the test is that same one that e2fsck uses | |
4880 | * NeilBrown 1999oct15 | |
4881 | */ | |
4882 | if (inode->i_nlink == 0) { | |
393d1d1d DTB |
4883 | if ((inode->i_mode == 0 || |
4884 | !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) && | |
4885 | ino != EXT4_BOOT_LOADER_INO) { | |
ac27a0ec | 4886 | /* this inode is deleted */ |
1d1fe1ee | 4887 | ret = -ESTALE; |
ac27a0ec DK |
4888 | goto bad_inode; |
4889 | } | |
4890 | /* The only unlinked inodes we let through here have | |
4891 | * valid i_mode and are being read by the orphan | |
4892 | * recovery code: that's fine, we're about to complete | |
393d1d1d DTB |
4893 | * the process of deleting those. |
4894 | * OR it is the EXT4_BOOT_LOADER_INO which is | |
4895 | * not initialized on a new filesystem. */ | |
ac27a0ec | 4896 | } |
ac27a0ec | 4897 | ei->i_flags = le32_to_cpu(raw_inode->i_flags); |
043546e4 | 4898 | ext4_set_inode_flags(inode, true); |
0fc1b451 | 4899 | inode->i_blocks = ext4_inode_blocks(raw_inode, ei); |
7973c0c1 | 4900 | ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo); |
e2b911c5 | 4901 | if (ext4_has_feature_64bit(sb)) |
a1ddeb7e BP |
4902 | ei->i_file_acl |= |
4903 | ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32; | |
e08ac99f | 4904 | inode->i_size = ext4_isize(sb, raw_inode); |
7e6e1ef4 | 4905 | if ((size = i_size_read(inode)) < 0) { |
8a363970 TT |
4906 | ext4_error_inode(inode, function, line, 0, |
4907 | "iget: bad i_size value: %lld", size); | |
7e6e1ef4 DW |
4908 | ret = -EFSCORRUPTED; |
4909 | goto bad_inode; | |
4910 | } | |
48a34311 JK |
4911 | /* |
4912 | * If dir_index is not enabled but there's dir with INDEX flag set, | |
4913 | * we'd normally treat htree data as empty space. But with metadata | |
4914 | * checksumming that corrupts checksums so forbid that. | |
4915 | */ | |
4916 | if (!ext4_has_feature_dir_index(sb) && ext4_has_metadata_csum(sb) && | |
4917 | ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) { | |
4918 | ext4_error_inode(inode, function, line, 0, | |
4919 | "iget: Dir with htree data on filesystem without dir_index feature."); | |
4920 | ret = -EFSCORRUPTED; | |
4921 | goto bad_inode; | |
4922 | } | |
ac27a0ec | 4923 | ei->i_disksize = inode->i_size; |
a9e7f447 DM |
4924 | #ifdef CONFIG_QUOTA |
4925 | ei->i_reserved_quota = 0; | |
4926 | #endif | |
ac27a0ec DK |
4927 | inode->i_generation = le32_to_cpu(raw_inode->i_generation); |
4928 | ei->i_block_group = iloc.block_group; | |
a4912123 | 4929 | ei->i_last_alloc_group = ~0; |
ac27a0ec DK |
4930 | /* |
4931 | * NOTE! The in-memory inode i_data array is in little-endian order | |
4932 | * even on big-endian machines: we do NOT byteswap the block numbers! | |
4933 | */ | |
617ba13b | 4934 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
ac27a0ec DK |
4935 | ei->i_data[block] = raw_inode->i_block[block]; |
4936 | INIT_LIST_HEAD(&ei->i_orphan); | |
aa75f4d3 | 4937 | ext4_fc_init_inode(&ei->vfs_inode); |
ac27a0ec | 4938 | |
b436b9be JK |
4939 | /* |
4940 | * Set transaction id's of transactions that have to be committed | |
4941 | * to finish f[data]sync. We set them to currently running transaction | |
4942 | * as we cannot be sure that the inode or some of its metadata isn't | |
4943 | * part of the transaction - the inode could have been reclaimed and | |
4944 | * now it is reread from disk. | |
4945 | */ | |
4946 | if (journal) { | |
4947 | transaction_t *transaction; | |
4948 | tid_t tid; | |
4949 | ||
a931da6a | 4950 | read_lock(&journal->j_state_lock); |
b436b9be JK |
4951 | if (journal->j_running_transaction) |
4952 | transaction = journal->j_running_transaction; | |
4953 | else | |
4954 | transaction = journal->j_committing_transaction; | |
4955 | if (transaction) | |
4956 | tid = transaction->t_tid; | |
4957 | else | |
4958 | tid = journal->j_commit_sequence; | |
a931da6a | 4959 | read_unlock(&journal->j_state_lock); |
b436b9be JK |
4960 | ei->i_sync_tid = tid; |
4961 | ei->i_datasync_tid = tid; | |
4962 | } | |
4963 | ||
0040d987 | 4964 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
ac27a0ec DK |
4965 | if (ei->i_extra_isize == 0) { |
4966 | /* The extra space is currently unused. Use it. */ | |
2dc8d9e1 | 4967 | BUILD_BUG_ON(sizeof(struct ext4_inode) & 3); |
617ba13b MC |
4968 | ei->i_extra_isize = sizeof(struct ext4_inode) - |
4969 | EXT4_GOOD_OLD_INODE_SIZE; | |
ac27a0ec | 4970 | } else { |
eb9b5f01 TT |
4971 | ret = ext4_iget_extra_inode(inode, raw_inode, ei); |
4972 | if (ret) | |
4973 | goto bad_inode; | |
ac27a0ec | 4974 | } |
814525f4 | 4975 | } |
ac27a0ec | 4976 | |
ef7f3835 KS |
4977 | EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode); |
4978 | EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode); | |
4979 | EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode); | |
4980 | EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode); | |
4981 | ||
ed3654eb | 4982 | if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) { |
ee73f9a5 JL |
4983 | u64 ivers = le32_to_cpu(raw_inode->i_disk_version); |
4984 | ||
c4f65706 TT |
4985 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
4986 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
ee73f9a5 | 4987 | ivers |= |
c4f65706 TT |
4988 | (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32; |
4989 | } | |
e254d1af | 4990 | ext4_inode_set_iversion_queried(inode, ivers); |
25ec56b5 JNC |
4991 | } |
4992 | ||
c4b5a614 | 4993 | ret = 0; |
485c26ec | 4994 | if (ei->i_file_acl && |
ce9f24cc | 4995 | !ext4_inode_block_valid(inode, ei->i_file_acl, 1)) { |
8a363970 TT |
4996 | ext4_error_inode(inode, function, line, 0, |
4997 | "iget: bad extended attribute block %llu", | |
24676da4 | 4998 | ei->i_file_acl); |
6a797d27 | 4999 | ret = -EFSCORRUPTED; |
485c26ec | 5000 | goto bad_inode; |
f19d5870 | 5001 | } else if (!ext4_has_inline_data(inode)) { |
bc716523 | 5002 | /* validate the block references in the inode */ |
8016e29f HS |
5003 | if (!(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY) && |
5004 | (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || | |
5005 | (S_ISLNK(inode->i_mode) && | |
5006 | !ext4_inode_is_fast_symlink(inode)))) { | |
bc716523 | 5007 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
f19d5870 | 5008 | ret = ext4_ext_check_inode(inode); |
bc716523 LS |
5009 | else |
5010 | ret = ext4_ind_check_inode(inode); | |
f19d5870 | 5011 | } |
fe2c8191 | 5012 | } |
567f3e9a | 5013 | if (ret) |
de9a55b8 | 5014 | goto bad_inode; |
7a262f7c | 5015 | |
ac27a0ec | 5016 | if (S_ISREG(inode->i_mode)) { |
617ba13b | 5017 | inode->i_op = &ext4_file_inode_operations; |
be64f884 | 5018 | inode->i_fop = &ext4_file_operations; |
617ba13b | 5019 | ext4_set_aops(inode); |
ac27a0ec | 5020 | } else if (S_ISDIR(inode->i_mode)) { |
617ba13b MC |
5021 | inode->i_op = &ext4_dir_inode_operations; |
5022 | inode->i_fop = &ext4_dir_operations; | |
ac27a0ec | 5023 | } else if (S_ISLNK(inode->i_mode)) { |
6390d33b LR |
5024 | /* VFS does not allow setting these so must be corruption */ |
5025 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) { | |
8a363970 TT |
5026 | ext4_error_inode(inode, function, line, 0, |
5027 | "iget: immutable or append flags " | |
5028 | "not allowed on symlinks"); | |
6390d33b LR |
5029 | ret = -EFSCORRUPTED; |
5030 | goto bad_inode; | |
5031 | } | |
592ddec7 | 5032 | if (IS_ENCRYPTED(inode)) { |
a7a67e8a | 5033 | inode->i_op = &ext4_encrypted_symlink_inode_operations; |
a7a67e8a | 5034 | } else if (ext4_inode_is_fast_symlink(inode)) { |
75e7566b | 5035 | inode->i_link = (char *)ei->i_data; |
617ba13b | 5036 | inode->i_op = &ext4_fast_symlink_inode_operations; |
e83c1397 DG |
5037 | nd_terminate_link(ei->i_data, inode->i_size, |
5038 | sizeof(ei->i_data) - 1); | |
5039 | } else { | |
617ba13b | 5040 | inode->i_op = &ext4_symlink_inode_operations; |
ac27a0ec | 5041 | } |
563bdd61 TT |
5042 | } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) || |
5043 | S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) { | |
617ba13b | 5044 | inode->i_op = &ext4_special_inode_operations; |
ac27a0ec DK |
5045 | if (raw_inode->i_block[0]) |
5046 | init_special_inode(inode, inode->i_mode, | |
5047 | old_decode_dev(le32_to_cpu(raw_inode->i_block[0]))); | |
5048 | else | |
5049 | init_special_inode(inode, inode->i_mode, | |
5050 | new_decode_dev(le32_to_cpu(raw_inode->i_block[1]))); | |
393d1d1d DTB |
5051 | } else if (ino == EXT4_BOOT_LOADER_INO) { |
5052 | make_bad_inode(inode); | |
563bdd61 | 5053 | } else { |
6a797d27 | 5054 | ret = -EFSCORRUPTED; |
8a363970 TT |
5055 | ext4_error_inode(inode, function, line, 0, |
5056 | "iget: bogus i_mode (%o)", inode->i_mode); | |
563bdd61 | 5057 | goto bad_inode; |
ac27a0ec | 5058 | } |
6456ca65 TT |
5059 | if (IS_CASEFOLDED(inode) && !ext4_has_feature_casefold(inode->i_sb)) |
5060 | ext4_error_inode(inode, function, line, 0, | |
5061 | "casefold flag without casefold feature"); | |
63b1e9bc BL |
5062 | if (is_bad_inode(inode) && !(flags & EXT4_IGET_BAD)) { |
5063 | ext4_error_inode(inode, function, line, 0, | |
5064 | "bad inode without EXT4_IGET_BAD flag"); | |
5065 | ret = -EUCLEAN; | |
5066 | goto bad_inode; | |
5067 | } | |
dec214d0 | 5068 | |
63b1e9bc | 5069 | brelse(iloc.bh); |
1d1fe1ee DH |
5070 | unlock_new_inode(inode); |
5071 | return inode; | |
ac27a0ec DK |
5072 | |
5073 | bad_inode: | |
567f3e9a | 5074 | brelse(iloc.bh); |
1d1fe1ee DH |
5075 | iget_failed(inode); |
5076 | return ERR_PTR(ret); | |
ac27a0ec DK |
5077 | } |
5078 | ||
3f19b2ab DH |
5079 | static void __ext4_update_other_inode_time(struct super_block *sb, |
5080 | unsigned long orig_ino, | |
5081 | unsigned long ino, | |
5082 | struct ext4_inode *raw_inode) | |
a26f4992 | 5083 | { |
3f19b2ab DH |
5084 | struct inode *inode; |
5085 | ||
5086 | inode = find_inode_by_ino_rcu(sb, ino); | |
5087 | if (!inode) | |
5088 | return; | |
a26f4992 | 5089 | |
ed296c6c | 5090 | if (!inode_is_dirtytime_only(inode)) |
3f19b2ab DH |
5091 | return; |
5092 | ||
a26f4992 | 5093 | spin_lock(&inode->i_lock); |
ed296c6c | 5094 | if (inode_is_dirtytime_only(inode)) { |
a26f4992 TT |
5095 | struct ext4_inode_info *ei = EXT4_I(inode); |
5096 | ||
5fcd5750 | 5097 | inode->i_state &= ~I_DIRTY_TIME; |
a26f4992 TT |
5098 | spin_unlock(&inode->i_lock); |
5099 | ||
5100 | spin_lock(&ei->i_raw_lock); | |
3f19b2ab DH |
5101 | EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode); |
5102 | EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode); | |
5103 | EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode); | |
5104 | ext4_inode_csum_set(inode, raw_inode, ei); | |
a26f4992 | 5105 | spin_unlock(&ei->i_raw_lock); |
3f19b2ab DH |
5106 | trace_ext4_other_inode_update_time(inode, orig_ino); |
5107 | return; | |
a26f4992 TT |
5108 | } |
5109 | spin_unlock(&inode->i_lock); | |
a26f4992 TT |
5110 | } |
5111 | ||
5112 | /* | |
5113 | * Opportunistically update the other time fields for other inodes in | |
5114 | * the same inode table block. | |
5115 | */ | |
5116 | static void ext4_update_other_inodes_time(struct super_block *sb, | |
5117 | unsigned long orig_ino, char *buf) | |
5118 | { | |
a26f4992 TT |
5119 | unsigned long ino; |
5120 | int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; | |
5121 | int inode_size = EXT4_INODE_SIZE(sb); | |
5122 | ||
0f0ff9a9 TT |
5123 | /* |
5124 | * Calculate the first inode in the inode table block. Inode | |
5125 | * numbers are one-based. That is, the first inode in a block | |
5126 | * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1). | |
5127 | */ | |
5128 | ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1; | |
3f19b2ab | 5129 | rcu_read_lock(); |
a26f4992 TT |
5130 | for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) { |
5131 | if (ino == orig_ino) | |
5132 | continue; | |
3f19b2ab DH |
5133 | __ext4_update_other_inode_time(sb, orig_ino, ino, |
5134 | (struct ext4_inode *)buf); | |
a26f4992 | 5135 | } |
3f19b2ab | 5136 | rcu_read_unlock(); |
a26f4992 TT |
5137 | } |
5138 | ||
664bd38b ZY |
5139 | /* |
5140 | * Post the struct inode info into an on-disk inode location in the | |
5141 | * buffer-cache. This gobbles the caller's reference to the | |
5142 | * buffer_head in the inode location struct. | |
5143 | * | |
5144 | * The caller must have write access to iloc->bh. | |
5145 | */ | |
5146 | static int ext4_do_update_inode(handle_t *handle, | |
5147 | struct inode *inode, | |
5148 | struct ext4_iloc *iloc) | |
5149 | { | |
5150 | struct ext4_inode *raw_inode = ext4_raw_inode(iloc); | |
5151 | struct ext4_inode_info *ei = EXT4_I(inode); | |
5152 | struct buffer_head *bh = iloc->bh; | |
5153 | struct super_block *sb = inode->i_sb; | |
5154 | int err; | |
5155 | int need_datasync = 0, set_large_file = 0; | |
5156 | ||
5157 | spin_lock(&ei->i_raw_lock); | |
5158 | ||
5159 | /* | |
5160 | * For fields not tracked in the in-memory inode, initialise them | |
5161 | * to zero for new inodes. | |
5162 | */ | |
5163 | if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) | |
5164 | memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size); | |
5165 | ||
5166 | if (READ_ONCE(ei->i_disksize) != ext4_isize(inode->i_sb, raw_inode)) | |
5167 | need_datasync = 1; | |
5168 | if (ei->i_disksize > 0x7fffffffULL) { | |
5169 | if (!ext4_has_feature_large_file(sb) || | |
5170 | EXT4_SB(sb)->s_es->s_rev_level == cpu_to_le32(EXT4_GOOD_OLD_REV)) | |
5171 | set_large_file = 1; | |
5172 | } | |
5173 | ||
5174 | err = ext4_fill_raw_inode(inode, raw_inode); | |
202ee5df | 5175 | spin_unlock(&ei->i_raw_lock); |
baaae979 ZY |
5176 | if (err) { |
5177 | EXT4_ERROR_INODE(inode, "corrupted inode contents"); | |
5178 | goto out_brelse; | |
5179 | } | |
5180 | ||
1751e8a6 | 5181 | if (inode->i_sb->s_flags & SB_LAZYTIME) |
a26f4992 TT |
5182 | ext4_update_other_inodes_time(inode->i_sb, inode->i_ino, |
5183 | bh->b_data); | |
202ee5df | 5184 | |
830156c7 | 5185 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
7d8bd3c7 SL |
5186 | err = ext4_handle_dirty_metadata(handle, NULL, bh); |
5187 | if (err) | |
baaae979 | 5188 | goto out_error; |
19f5fb7a | 5189 | ext4_clear_inode_state(inode, EXT4_STATE_NEW); |
202ee5df | 5190 | if (set_large_file) { |
5d601255 | 5191 | BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access"); |
188c299e JK |
5192 | err = ext4_journal_get_write_access(handle, sb, |
5193 | EXT4_SB(sb)->s_sbh, | |
5194 | EXT4_JTR_NONE); | |
202ee5df | 5195 | if (err) |
baaae979 | 5196 | goto out_error; |
05c2c00f | 5197 | lock_buffer(EXT4_SB(sb)->s_sbh); |
e2b911c5 | 5198 | ext4_set_feature_large_file(sb); |
05c2c00f JK |
5199 | ext4_superblock_csum_set(sb); |
5200 | unlock_buffer(EXT4_SB(sb)->s_sbh); | |
202ee5df | 5201 | ext4_handle_sync(handle); |
a3f5cf14 JK |
5202 | err = ext4_handle_dirty_metadata(handle, NULL, |
5203 | EXT4_SB(sb)->s_sbh); | |
202ee5df | 5204 | } |
b71fc079 | 5205 | ext4_update_inode_fsync_trans(handle, inode, need_datasync); |
baaae979 ZY |
5206 | out_error: |
5207 | ext4_std_error(inode->i_sb, err); | |
ac27a0ec | 5208 | out_brelse: |
af5bc92d | 5209 | brelse(bh); |
ac27a0ec DK |
5210 | return err; |
5211 | } | |
5212 | ||
5213 | /* | |
617ba13b | 5214 | * ext4_write_inode() |
ac27a0ec DK |
5215 | * |
5216 | * We are called from a few places: | |
5217 | * | |
87f7e416 | 5218 | * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files. |
ac27a0ec | 5219 | * Here, there will be no transaction running. We wait for any running |
4907cb7b | 5220 | * transaction to commit. |
ac27a0ec | 5221 | * |
87f7e416 TT |
5222 | * - Within flush work (sys_sync(), kupdate and such). |
5223 | * We wait on commit, if told to. | |
ac27a0ec | 5224 | * |
87f7e416 TT |
5225 | * - Within iput_final() -> write_inode_now() |
5226 | * We wait on commit, if told to. | |
ac27a0ec DK |
5227 | * |
5228 | * In all cases it is actually safe for us to return without doing anything, | |
5229 | * because the inode has been copied into a raw inode buffer in | |
87f7e416 TT |
5230 | * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL |
5231 | * writeback. | |
ac27a0ec DK |
5232 | * |
5233 | * Note that we are absolutely dependent upon all inode dirtiers doing the | |
5234 | * right thing: they *must* call mark_inode_dirty() after dirtying info in | |
5235 | * which we are interested. | |
5236 | * | |
5237 | * It would be a bug for them to not do this. The code: | |
5238 | * | |
5239 | * mark_inode_dirty(inode) | |
5240 | * stuff(); | |
5241 | * inode->i_size = expr; | |
5242 | * | |
87f7e416 TT |
5243 | * is in error because write_inode() could occur while `stuff()' is running, |
5244 | * and the new i_size will be lost. Plus the inode will no longer be on the | |
5245 | * superblock's dirty inode list. | |
ac27a0ec | 5246 | */ |
a9185b41 | 5247 | int ext4_write_inode(struct inode *inode, struct writeback_control *wbc) |
ac27a0ec | 5248 | { |
91ac6f43 FM |
5249 | int err; |
5250 | ||
18f2c4fc TT |
5251 | if (WARN_ON_ONCE(current->flags & PF_MEMALLOC) || |
5252 | sb_rdonly(inode->i_sb)) | |
ac27a0ec DK |
5253 | return 0; |
5254 | ||
18f2c4fc TT |
5255 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
5256 | return -EIO; | |
5257 | ||
91ac6f43 FM |
5258 | if (EXT4_SB(inode->i_sb)->s_journal) { |
5259 | if (ext4_journal_current_handle()) { | |
4978c659 | 5260 | ext4_debug("called recursively, non-PF_MEMALLOC!\n"); |
91ac6f43 FM |
5261 | dump_stack(); |
5262 | return -EIO; | |
5263 | } | |
ac27a0ec | 5264 | |
10542c22 JK |
5265 | /* |
5266 | * No need to force transaction in WB_SYNC_NONE mode. Also | |
5267 | * ext4_sync_fs() will force the commit after everything is | |
5268 | * written. | |
5269 | */ | |
5270 | if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync) | |
91ac6f43 FM |
5271 | return 0; |
5272 | ||
aa75f4d3 | 5273 | err = ext4_fc_commit(EXT4_SB(inode->i_sb)->s_journal, |
18f2c4fc | 5274 | EXT4_I(inode)->i_sync_tid); |
91ac6f43 FM |
5275 | } else { |
5276 | struct ext4_iloc iloc; | |
ac27a0ec | 5277 | |
8016e29f | 5278 | err = __ext4_get_inode_loc_noinmem(inode, &iloc); |
91ac6f43 FM |
5279 | if (err) |
5280 | return err; | |
10542c22 JK |
5281 | /* |
5282 | * sync(2) will flush the whole buffer cache. No need to do | |
5283 | * it here separately for each inode. | |
5284 | */ | |
5285 | if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) | |
830156c7 FM |
5286 | sync_dirty_buffer(iloc.bh); |
5287 | if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) { | |
54d3adbc TT |
5288 | ext4_error_inode_block(inode, iloc.bh->b_blocknr, EIO, |
5289 | "IO error syncing inode"); | |
830156c7 FM |
5290 | err = -EIO; |
5291 | } | |
fd2dd9fb | 5292 | brelse(iloc.bh); |
91ac6f43 FM |
5293 | } |
5294 | return err; | |
ac27a0ec DK |
5295 | } |
5296 | ||
53e87268 | 5297 | /* |
ccd16945 MWO |
5298 | * In data=journal mode ext4_journalled_invalidate_folio() may fail to invalidate |
5299 | * buffers that are attached to a folio straddling i_size and are undergoing | |
53e87268 JK |
5300 | * commit. In that case we have to wait for commit to finish and try again. |
5301 | */ | |
5302 | static void ext4_wait_for_tail_page_commit(struct inode *inode) | |
5303 | { | |
53e87268 JK |
5304 | unsigned offset; |
5305 | journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; | |
5306 | tid_t commit_tid = 0; | |
5307 | int ret; | |
5308 | ||
09cbfeaf | 5309 | offset = inode->i_size & (PAGE_SIZE - 1); |
53e87268 | 5310 | /* |
ccd16945 MWO |
5311 | * If the folio is fully truncated, we don't need to wait for any commit |
5312 | * (and we even should not as __ext4_journalled_invalidate_folio() may | |
5313 | * strip all buffers from the folio but keep the folio dirty which can then | |
5314 | * confuse e.g. concurrent ext4_writepage() seeing dirty folio without | |
565333a1 | 5315 | * buffers). Also we don't need to wait for any commit if all buffers in |
ccd16945 | 5316 | * the folio remain valid. This is most beneficial for the common case of |
565333a1 | 5317 | * blocksize == PAGESIZE. |
53e87268 | 5318 | */ |
565333a1 | 5319 | if (!offset || offset > (PAGE_SIZE - i_blocksize(inode))) |
53e87268 JK |
5320 | return; |
5321 | while (1) { | |
ccd16945 | 5322 | struct folio *folio = filemap_lock_folio(inode->i_mapping, |
09cbfeaf | 5323 | inode->i_size >> PAGE_SHIFT); |
ccd16945 | 5324 | if (!folio) |
53e87268 | 5325 | return; |
ccd16945 MWO |
5326 | ret = __ext4_journalled_invalidate_folio(folio, offset, |
5327 | folio_size(folio) - offset); | |
5328 | folio_unlock(folio); | |
5329 | folio_put(folio); | |
53e87268 JK |
5330 | if (ret != -EBUSY) |
5331 | return; | |
5332 | commit_tid = 0; | |
5333 | read_lock(&journal->j_state_lock); | |
5334 | if (journal->j_committing_transaction) | |
5335 | commit_tid = journal->j_committing_transaction->t_tid; | |
5336 | read_unlock(&journal->j_state_lock); | |
5337 | if (commit_tid) | |
5338 | jbd2_log_wait_commit(journal, commit_tid); | |
5339 | } | |
5340 | } | |
5341 | ||
ac27a0ec | 5342 | /* |
617ba13b | 5343 | * ext4_setattr() |
ac27a0ec DK |
5344 | * |
5345 | * Called from notify_change. | |
5346 | * | |
5347 | * We want to trap VFS attempts to truncate the file as soon as | |
5348 | * possible. In particular, we want to make sure that when the VFS | |
5349 | * shrinks i_size, we put the inode on the orphan list and modify | |
5350 | * i_disksize immediately, so that during the subsequent flushing of | |
5351 | * dirty pages and freeing of disk blocks, we can guarantee that any | |
5352 | * commit will leave the blocks being flushed in an unused state on | |
5353 | * disk. (On recovery, the inode will get truncated and the blocks will | |
5354 | * be freed, so we have a strong guarantee that no future commit will | |
5355 | * leave these blocks visible to the user.) | |
5356 | * | |
678aaf48 JK |
5357 | * Another thing we have to assure is that if we are in ordered mode |
5358 | * and inode is still attached to the committing transaction, we must | |
5359 | * we start writeout of all the dirty pages which are being truncated. | |
5360 | * This way we are sure that all the data written in the previous | |
5361 | * transaction are already on disk (truncate waits for pages under | |
5362 | * writeback). | |
5363 | * | |
f340b3d9 | 5364 | * Called with inode->i_rwsem down. |
ac27a0ec | 5365 | */ |
549c7297 CB |
5366 | int ext4_setattr(struct user_namespace *mnt_userns, struct dentry *dentry, |
5367 | struct iattr *attr) | |
ac27a0ec | 5368 | { |
2b0143b5 | 5369 | struct inode *inode = d_inode(dentry); |
ac27a0ec | 5370 | int error, rc = 0; |
3d287de3 | 5371 | int orphan = 0; |
ac27a0ec | 5372 | const unsigned int ia_valid = attr->ia_valid; |
a642c2c0 | 5373 | bool inc_ivers = true; |
ac27a0ec | 5374 | |
0db1ff22 TT |
5375 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
5376 | return -EIO; | |
5377 | ||
02b016ca TT |
5378 | if (unlikely(IS_IMMUTABLE(inode))) |
5379 | return -EPERM; | |
5380 | ||
5381 | if (unlikely(IS_APPEND(inode) && | |
5382 | (ia_valid & (ATTR_MODE | ATTR_UID | | |
5383 | ATTR_GID | ATTR_TIMES_SET)))) | |
5384 | return -EPERM; | |
5385 | ||
14f3db55 | 5386 | error = setattr_prepare(mnt_userns, dentry, attr); |
ac27a0ec DK |
5387 | if (error) |
5388 | return error; | |
5389 | ||
3ce2b8dd EB |
5390 | error = fscrypt_prepare_setattr(dentry, attr); |
5391 | if (error) | |
5392 | return error; | |
5393 | ||
c93d8f88 EB |
5394 | error = fsverity_prepare_setattr(dentry, attr); |
5395 | if (error) | |
5396 | return error; | |
5397 | ||
b27c82e1 | 5398 | if (is_quota_modification(mnt_userns, inode, attr)) { |
a7cdadee JK |
5399 | error = dquot_initialize(inode); |
5400 | if (error) | |
5401 | return error; | |
5402 | } | |
2729cfdc | 5403 | |
b27c82e1 CB |
5404 | if (i_uid_needs_update(mnt_userns, attr, inode) || |
5405 | i_gid_needs_update(mnt_userns, attr, inode)) { | |
ac27a0ec DK |
5406 | handle_t *handle; |
5407 | ||
5408 | /* (user+group)*(old+new) structure, inode write (sb, | |
5409 | * inode block, ? - but truncate inode update has it) */ | |
9924a92a TT |
5410 | handle = ext4_journal_start(inode, EXT4_HT_QUOTA, |
5411 | (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) + | |
5412 | EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3); | |
ac27a0ec DK |
5413 | if (IS_ERR(handle)) { |
5414 | error = PTR_ERR(handle); | |
5415 | goto err_out; | |
5416 | } | |
7a9ca53a TE |
5417 | |
5418 | /* dquot_transfer() calls back ext4_get_inode_usage() which | |
5419 | * counts xattr inode references. | |
5420 | */ | |
5421 | down_read(&EXT4_I(inode)->xattr_sem); | |
b27c82e1 | 5422 | error = dquot_transfer(mnt_userns, inode, attr); |
7a9ca53a TE |
5423 | up_read(&EXT4_I(inode)->xattr_sem); |
5424 | ||
ac27a0ec | 5425 | if (error) { |
617ba13b | 5426 | ext4_journal_stop(handle); |
ac27a0ec DK |
5427 | return error; |
5428 | } | |
5429 | /* Update corresponding info in inode so that everything is in | |
5430 | * one transaction */ | |
b27c82e1 CB |
5431 | i_uid_update(mnt_userns, attr, inode); |
5432 | i_gid_update(mnt_userns, attr, inode); | |
617ba13b MC |
5433 | error = ext4_mark_inode_dirty(handle, inode); |
5434 | ext4_journal_stop(handle); | |
512c15ef | 5435 | if (unlikely(error)) { |
4209ae12 | 5436 | return error; |
512c15ef | 5437 | } |
ac27a0ec DK |
5438 | } |
5439 | ||
3da40c7b | 5440 | if (attr->ia_valid & ATTR_SIZE) { |
5208386c | 5441 | handle_t *handle; |
3da40c7b | 5442 | loff_t oldsize = inode->i_size; |
f4534c9f | 5443 | loff_t old_disksize; |
b9c1c267 | 5444 | int shrink = (attr->ia_size < inode->i_size); |
562c72aa | 5445 | |
12e9b892 | 5446 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) { |
e2b46574 ES |
5447 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
5448 | ||
aa75f4d3 | 5449 | if (attr->ia_size > sbi->s_bitmap_maxbytes) { |
0c095c7f | 5450 | return -EFBIG; |
aa75f4d3 | 5451 | } |
e2b46574 | 5452 | } |
aa75f4d3 | 5453 | if (!S_ISREG(inode->i_mode)) { |
3da40c7b | 5454 | return -EINVAL; |
aa75f4d3 | 5455 | } |
dff6efc3 | 5456 | |
a642c2c0 JL |
5457 | if (attr->ia_size == inode->i_size) |
5458 | inc_ivers = false; | |
dff6efc3 | 5459 | |
b9c1c267 JK |
5460 | if (shrink) { |
5461 | if (ext4_should_order_data(inode)) { | |
5462 | error = ext4_begin_ordered_truncate(inode, | |
678aaf48 | 5463 | attr->ia_size); |
b9c1c267 JK |
5464 | if (error) |
5465 | goto err_out; | |
5466 | } | |
5467 | /* | |
5468 | * Blocks are going to be removed from the inode. Wait | |
5469 | * for dio in flight. | |
5470 | */ | |
5471 | inode_dio_wait(inode); | |
5472 | } | |
5473 | ||
d4f5258e | 5474 | filemap_invalidate_lock(inode->i_mapping); |
b9c1c267 JK |
5475 | |
5476 | rc = ext4_break_layouts(inode); | |
5477 | if (rc) { | |
d4f5258e | 5478 | filemap_invalidate_unlock(inode->i_mapping); |
aa75f4d3 | 5479 | goto err_out; |
3da40c7b | 5480 | } |
b9c1c267 | 5481 | |
3da40c7b | 5482 | if (attr->ia_size != inode->i_size) { |
5208386c JK |
5483 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 3); |
5484 | if (IS_ERR(handle)) { | |
5485 | error = PTR_ERR(handle); | |
b9c1c267 | 5486 | goto out_mmap_sem; |
5208386c | 5487 | } |
3da40c7b | 5488 | if (ext4_handle_valid(handle) && shrink) { |
5208386c JK |
5489 | error = ext4_orphan_add(handle, inode); |
5490 | orphan = 1; | |
5491 | } | |
911af577 EG |
5492 | /* |
5493 | * Update c/mtime on truncate up, ext4_truncate() will | |
5494 | * update c/mtime in shrink case below | |
5495 | */ | |
5496 | if (!shrink) { | |
eeca7ea1 | 5497 | inode->i_mtime = current_time(inode); |
911af577 EG |
5498 | inode->i_ctime = inode->i_mtime; |
5499 | } | |
aa75f4d3 HS |
5500 | |
5501 | if (shrink) | |
a80f7fcf | 5502 | ext4_fc_track_range(handle, inode, |
aa75f4d3 HS |
5503 | (attr->ia_size > 0 ? attr->ia_size - 1 : 0) >> |
5504 | inode->i_sb->s_blocksize_bits, | |
9725958b | 5505 | EXT_MAX_BLOCKS - 1); |
aa75f4d3 HS |
5506 | else |
5507 | ext4_fc_track_range( | |
a80f7fcf | 5508 | handle, inode, |
aa75f4d3 HS |
5509 | (oldsize > 0 ? oldsize - 1 : oldsize) >> |
5510 | inode->i_sb->s_blocksize_bits, | |
5511 | (attr->ia_size > 0 ? attr->ia_size - 1 : 0) >> | |
5512 | inode->i_sb->s_blocksize_bits); | |
5513 | ||
90e775b7 | 5514 | down_write(&EXT4_I(inode)->i_data_sem); |
f4534c9f | 5515 | old_disksize = EXT4_I(inode)->i_disksize; |
5208386c JK |
5516 | EXT4_I(inode)->i_disksize = attr->ia_size; |
5517 | rc = ext4_mark_inode_dirty(handle, inode); | |
5518 | if (!error) | |
5519 | error = rc; | |
90e775b7 JK |
5520 | /* |
5521 | * We have to update i_size under i_data_sem together | |
5522 | * with i_disksize to avoid races with writeback code | |
5523 | * running ext4_wb_update_i_disksize(). | |
5524 | */ | |
5525 | if (!error) | |
5526 | i_size_write(inode, attr->ia_size); | |
f4534c9f YB |
5527 | else |
5528 | EXT4_I(inode)->i_disksize = old_disksize; | |
90e775b7 | 5529 | up_write(&EXT4_I(inode)->i_data_sem); |
5208386c | 5530 | ext4_journal_stop(handle); |
b9c1c267 JK |
5531 | if (error) |
5532 | goto out_mmap_sem; | |
5533 | if (!shrink) { | |
5534 | pagecache_isize_extended(inode, oldsize, | |
5535 | inode->i_size); | |
5536 | } else if (ext4_should_journal_data(inode)) { | |
5537 | ext4_wait_for_tail_page_commit(inode); | |
678aaf48 | 5538 | } |
d6320cbf | 5539 | } |
430657b6 | 5540 | |
5208386c JK |
5541 | /* |
5542 | * Truncate pagecache after we've waited for commit | |
5543 | * in data=journal mode to make pages freeable. | |
5544 | */ | |
923ae0ff | 5545 | truncate_pagecache(inode, inode->i_size); |
b9c1c267 JK |
5546 | /* |
5547 | * Call ext4_truncate() even if i_size didn't change to | |
5548 | * truncate possible preallocated blocks. | |
5549 | */ | |
5550 | if (attr->ia_size <= oldsize) { | |
2c98eb5e TT |
5551 | rc = ext4_truncate(inode); |
5552 | if (rc) | |
5553 | error = rc; | |
5554 | } | |
b9c1c267 | 5555 | out_mmap_sem: |
d4f5258e | 5556 | filemap_invalidate_unlock(inode->i_mapping); |
072bd7ea | 5557 | } |
ac27a0ec | 5558 | |
2c98eb5e | 5559 | if (!error) { |
a642c2c0 JL |
5560 | if (inc_ivers) |
5561 | inode_inc_iversion(inode); | |
14f3db55 | 5562 | setattr_copy(mnt_userns, inode, attr); |
1025774c CH |
5563 | mark_inode_dirty(inode); |
5564 | } | |
5565 | ||
5566 | /* | |
5567 | * If the call to ext4_truncate failed to get a transaction handle at | |
5568 | * all, we need to clean up the in-core orphan list manually. | |
5569 | */ | |
3d287de3 | 5570 | if (orphan && inode->i_nlink) |
617ba13b | 5571 | ext4_orphan_del(NULL, inode); |
ac27a0ec | 5572 | |
2c98eb5e | 5573 | if (!error && (ia_valid & ATTR_MODE)) |
14f3db55 | 5574 | rc = posix_acl_chmod(mnt_userns, inode, inode->i_mode); |
ac27a0ec DK |
5575 | |
5576 | err_out: | |
aa75f4d3 HS |
5577 | if (error) |
5578 | ext4_std_error(inode->i_sb, error); | |
ac27a0ec DK |
5579 | if (!error) |
5580 | error = rc; | |
5581 | return error; | |
5582 | } | |
5583 | ||
8434ef1d EB |
5584 | u32 ext4_dio_alignment(struct inode *inode) |
5585 | { | |
5586 | if (fsverity_active(inode)) | |
5587 | return 0; | |
5588 | if (ext4_should_journal_data(inode)) | |
5589 | return 0; | |
5590 | if (ext4_has_inline_data(inode)) | |
5591 | return 0; | |
5592 | if (IS_ENCRYPTED(inode)) { | |
5593 | if (!fscrypt_dio_supported(inode)) | |
5594 | return 0; | |
5595 | return i_blocksize(inode); | |
5596 | } | |
5597 | return 1; /* use the iomap defaults */ | |
5598 | } | |
5599 | ||
549c7297 CB |
5600 | int ext4_getattr(struct user_namespace *mnt_userns, const struct path *path, |
5601 | struct kstat *stat, u32 request_mask, unsigned int query_flags) | |
3e3398a0 | 5602 | { |
99652ea5 DH |
5603 | struct inode *inode = d_inode(path->dentry); |
5604 | struct ext4_inode *raw_inode; | |
5605 | struct ext4_inode_info *ei = EXT4_I(inode); | |
5606 | unsigned int flags; | |
5607 | ||
d4c5e960 TT |
5608 | if ((request_mask & STATX_BTIME) && |
5609 | EXT4_FITS_IN_INODE(raw_inode, ei, i_crtime)) { | |
99652ea5 DH |
5610 | stat->result_mask |= STATX_BTIME; |
5611 | stat->btime.tv_sec = ei->i_crtime.tv_sec; | |
5612 | stat->btime.tv_nsec = ei->i_crtime.tv_nsec; | |
5613 | } | |
5614 | ||
8434ef1d EB |
5615 | /* |
5616 | * Return the DIO alignment restrictions if requested. We only return | |
5617 | * this information when requested, since on encrypted files it might | |
5618 | * take a fair bit of work to get if the file wasn't opened recently. | |
5619 | */ | |
5620 | if ((request_mask & STATX_DIOALIGN) && S_ISREG(inode->i_mode)) { | |
5621 | u32 dio_align = ext4_dio_alignment(inode); | |
5622 | ||
5623 | stat->result_mask |= STATX_DIOALIGN; | |
5624 | if (dio_align == 1) { | |
5625 | struct block_device *bdev = inode->i_sb->s_bdev; | |
5626 | ||
5627 | /* iomap defaults */ | |
5628 | stat->dio_mem_align = bdev_dma_alignment(bdev) + 1; | |
5629 | stat->dio_offset_align = bdev_logical_block_size(bdev); | |
5630 | } else { | |
5631 | stat->dio_mem_align = dio_align; | |
5632 | stat->dio_offset_align = dio_align; | |
5633 | } | |
5634 | } | |
5635 | ||
99652ea5 DH |
5636 | flags = ei->i_flags & EXT4_FL_USER_VISIBLE; |
5637 | if (flags & EXT4_APPEND_FL) | |
5638 | stat->attributes |= STATX_ATTR_APPEND; | |
5639 | if (flags & EXT4_COMPR_FL) | |
5640 | stat->attributes |= STATX_ATTR_COMPRESSED; | |
5641 | if (flags & EXT4_ENCRYPT_FL) | |
5642 | stat->attributes |= STATX_ATTR_ENCRYPTED; | |
5643 | if (flags & EXT4_IMMUTABLE_FL) | |
5644 | stat->attributes |= STATX_ATTR_IMMUTABLE; | |
5645 | if (flags & EXT4_NODUMP_FL) | |
5646 | stat->attributes |= STATX_ATTR_NODUMP; | |
1f607195 EB |
5647 | if (flags & EXT4_VERITY_FL) |
5648 | stat->attributes |= STATX_ATTR_VERITY; | |
3e3398a0 | 5649 | |
3209f68b DH |
5650 | stat->attributes_mask |= (STATX_ATTR_APPEND | |
5651 | STATX_ATTR_COMPRESSED | | |
5652 | STATX_ATTR_ENCRYPTED | | |
5653 | STATX_ATTR_IMMUTABLE | | |
1f607195 EB |
5654 | STATX_ATTR_NODUMP | |
5655 | STATX_ATTR_VERITY); | |
3209f68b | 5656 | |
14f3db55 | 5657 | generic_fillattr(mnt_userns, inode, stat); |
99652ea5 DH |
5658 | return 0; |
5659 | } | |
5660 | ||
549c7297 CB |
5661 | int ext4_file_getattr(struct user_namespace *mnt_userns, |
5662 | const struct path *path, struct kstat *stat, | |
99652ea5 DH |
5663 | u32 request_mask, unsigned int query_flags) |
5664 | { | |
5665 | struct inode *inode = d_inode(path->dentry); | |
5666 | u64 delalloc_blocks; | |
5667 | ||
14f3db55 | 5668 | ext4_getattr(mnt_userns, path, stat, request_mask, query_flags); |
3e3398a0 | 5669 | |
9206c561 AD |
5670 | /* |
5671 | * If there is inline data in the inode, the inode will normally not | |
5672 | * have data blocks allocated (it may have an external xattr block). | |
5673 | * Report at least one sector for such files, so tools like tar, rsync, | |
d67d64f4 | 5674 | * others don't incorrectly think the file is completely sparse. |
9206c561 AD |
5675 | */ |
5676 | if (unlikely(ext4_has_inline_data(inode))) | |
5677 | stat->blocks += (stat->size + 511) >> 9; | |
5678 | ||
3e3398a0 MC |
5679 | /* |
5680 | * We can't update i_blocks if the block allocation is delayed | |
5681 | * otherwise in the case of system crash before the real block | |
5682 | * allocation is done, we will have i_blocks inconsistent with | |
5683 | * on-disk file blocks. | |
5684 | * We always keep i_blocks updated together with real | |
5685 | * allocation. But to not confuse with user, stat | |
5686 | * will return the blocks that include the delayed allocation | |
5687 | * blocks for this file. | |
5688 | */ | |
96607551 | 5689 | delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb), |
9206c561 AD |
5690 | EXT4_I(inode)->i_reserved_data_blocks); |
5691 | stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9); | |
3e3398a0 MC |
5692 | return 0; |
5693 | } | |
ac27a0ec | 5694 | |
fffb2739 JK |
5695 | static int ext4_index_trans_blocks(struct inode *inode, int lblocks, |
5696 | int pextents) | |
a02908f1 | 5697 | { |
12e9b892 | 5698 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) |
fffb2739 JK |
5699 | return ext4_ind_trans_blocks(inode, lblocks); |
5700 | return ext4_ext_index_trans_blocks(inode, pextents); | |
a02908f1 | 5701 | } |
ac51d837 | 5702 | |
ac27a0ec | 5703 | /* |
a02908f1 MC |
5704 | * Account for index blocks, block groups bitmaps and block group |
5705 | * descriptor blocks if modify datablocks and index blocks | |
5706 | * worse case, the indexs blocks spread over different block groups | |
ac27a0ec | 5707 | * |
a02908f1 | 5708 | * If datablocks are discontiguous, they are possible to spread over |
4907cb7b | 5709 | * different block groups too. If they are contiguous, with flexbg, |
a02908f1 | 5710 | * they could still across block group boundary. |
ac27a0ec | 5711 | * |
a02908f1 MC |
5712 | * Also account for superblock, inode, quota and xattr blocks |
5713 | */ | |
dec214d0 | 5714 | static int ext4_meta_trans_blocks(struct inode *inode, int lblocks, |
fffb2739 | 5715 | int pextents) |
a02908f1 | 5716 | { |
8df9675f TT |
5717 | ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb); |
5718 | int gdpblocks; | |
a02908f1 MC |
5719 | int idxblocks; |
5720 | int ret = 0; | |
5721 | ||
5722 | /* | |
fffb2739 JK |
5723 | * How many index blocks need to touch to map @lblocks logical blocks |
5724 | * to @pextents physical extents? | |
a02908f1 | 5725 | */ |
fffb2739 | 5726 | idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents); |
a02908f1 MC |
5727 | |
5728 | ret = idxblocks; | |
5729 | ||
5730 | /* | |
5731 | * Now let's see how many group bitmaps and group descriptors need | |
5732 | * to account | |
5733 | */ | |
fffb2739 | 5734 | groups = idxblocks + pextents; |
a02908f1 | 5735 | gdpblocks = groups; |
8df9675f TT |
5736 | if (groups > ngroups) |
5737 | groups = ngroups; | |
a02908f1 MC |
5738 | if (groups > EXT4_SB(inode->i_sb)->s_gdb_count) |
5739 | gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count; | |
5740 | ||
5741 | /* bitmaps and block group descriptor blocks */ | |
5742 | ret += groups + gdpblocks; | |
5743 | ||
5744 | /* Blocks for super block, inode, quota and xattr blocks */ | |
5745 | ret += EXT4_META_TRANS_BLOCKS(inode->i_sb); | |
5746 | ||
5747 | return ret; | |
5748 | } | |
5749 | ||
5750 | /* | |
25985edc | 5751 | * Calculate the total number of credits to reserve to fit |
f3bd1f3f MC |
5752 | * the modification of a single pages into a single transaction, |
5753 | * which may include multiple chunks of block allocations. | |
ac27a0ec | 5754 | * |
525f4ed8 | 5755 | * This could be called via ext4_write_begin() |
ac27a0ec | 5756 | * |
525f4ed8 | 5757 | * We need to consider the worse case, when |
a02908f1 | 5758 | * one new block per extent. |
ac27a0ec | 5759 | */ |
a86c6181 | 5760 | int ext4_writepage_trans_blocks(struct inode *inode) |
ac27a0ec | 5761 | { |
617ba13b | 5762 | int bpp = ext4_journal_blocks_per_page(inode); |
ac27a0ec DK |
5763 | int ret; |
5764 | ||
fffb2739 | 5765 | ret = ext4_meta_trans_blocks(inode, bpp, bpp); |
a86c6181 | 5766 | |
a02908f1 | 5767 | /* Account for data blocks for journalled mode */ |
617ba13b | 5768 | if (ext4_should_journal_data(inode)) |
a02908f1 | 5769 | ret += bpp; |
ac27a0ec DK |
5770 | return ret; |
5771 | } | |
f3bd1f3f MC |
5772 | |
5773 | /* | |
5774 | * Calculate the journal credits for a chunk of data modification. | |
5775 | * | |
5776 | * This is called from DIO, fallocate or whoever calling | |
79e83036 | 5777 | * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks. |
f3bd1f3f MC |
5778 | * |
5779 | * journal buffers for data blocks are not included here, as DIO | |
5780 | * and fallocate do no need to journal data buffers. | |
5781 | */ | |
5782 | int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks) | |
5783 | { | |
5784 | return ext4_meta_trans_blocks(inode, nrblocks, 1); | |
5785 | } | |
5786 | ||
ac27a0ec | 5787 | /* |
617ba13b | 5788 | * The caller must have previously called ext4_reserve_inode_write(). |
ac27a0ec DK |
5789 | * Give this, we know that the caller already has write access to iloc->bh. |
5790 | */ | |
617ba13b | 5791 | int ext4_mark_iloc_dirty(handle_t *handle, |
de9a55b8 | 5792 | struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec DK |
5793 | { |
5794 | int err = 0; | |
5795 | ||
a6758309 VA |
5796 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) { |
5797 | put_bh(iloc->bh); | |
0db1ff22 | 5798 | return -EIO; |
a6758309 | 5799 | } |
a80f7fcf | 5800 | ext4_fc_track_inode(handle, inode); |
aa75f4d3 | 5801 | |
ac27a0ec DK |
5802 | /* the do_update_inode consumes one bh->b_count */ |
5803 | get_bh(iloc->bh); | |
5804 | ||
dab291af | 5805 | /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */ |
830156c7 | 5806 | err = ext4_do_update_inode(handle, inode, iloc); |
ac27a0ec DK |
5807 | put_bh(iloc->bh); |
5808 | return err; | |
5809 | } | |
5810 | ||
5811 | /* | |
5812 | * On success, We end up with an outstanding reference count against | |
5813 | * iloc->bh. This _must_ be cleaned up later. | |
5814 | */ | |
5815 | ||
5816 | int | |
617ba13b MC |
5817 | ext4_reserve_inode_write(handle_t *handle, struct inode *inode, |
5818 | struct ext4_iloc *iloc) | |
ac27a0ec | 5819 | { |
0390131b FM |
5820 | int err; |
5821 | ||
0db1ff22 TT |
5822 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
5823 | return -EIO; | |
5824 | ||
0390131b FM |
5825 | err = ext4_get_inode_loc(inode, iloc); |
5826 | if (!err) { | |
5827 | BUFFER_TRACE(iloc->bh, "get_write_access"); | |
188c299e JK |
5828 | err = ext4_journal_get_write_access(handle, inode->i_sb, |
5829 | iloc->bh, EXT4_JTR_NONE); | |
0390131b FM |
5830 | if (err) { |
5831 | brelse(iloc->bh); | |
5832 | iloc->bh = NULL; | |
ac27a0ec DK |
5833 | } |
5834 | } | |
617ba13b | 5835 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
5836 | return err; |
5837 | } | |
5838 | ||
c03b45b8 MX |
5839 | static int __ext4_expand_extra_isize(struct inode *inode, |
5840 | unsigned int new_extra_isize, | |
5841 | struct ext4_iloc *iloc, | |
5842 | handle_t *handle, int *no_expand) | |
5843 | { | |
5844 | struct ext4_inode *raw_inode; | |
5845 | struct ext4_xattr_ibody_header *header; | |
4ea99936 TT |
5846 | unsigned int inode_size = EXT4_INODE_SIZE(inode->i_sb); |
5847 | struct ext4_inode_info *ei = EXT4_I(inode); | |
c03b45b8 MX |
5848 | int error; |
5849 | ||
4ea99936 TT |
5850 | /* this was checked at iget time, but double check for good measure */ |
5851 | if ((EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > inode_size) || | |
5852 | (ei->i_extra_isize & 3)) { | |
5853 | EXT4_ERROR_INODE(inode, "bad extra_isize %u (inode size %u)", | |
5854 | ei->i_extra_isize, | |
5855 | EXT4_INODE_SIZE(inode->i_sb)); | |
5856 | return -EFSCORRUPTED; | |
5857 | } | |
5858 | if ((new_extra_isize < ei->i_extra_isize) || | |
5859 | (new_extra_isize < 4) || | |
5860 | (new_extra_isize > inode_size - EXT4_GOOD_OLD_INODE_SIZE)) | |
5861 | return -EINVAL; /* Should never happen */ | |
5862 | ||
c03b45b8 MX |
5863 | raw_inode = ext4_raw_inode(iloc); |
5864 | ||
5865 | header = IHDR(inode, raw_inode); | |
5866 | ||
5867 | /* No extended attributes present */ | |
5868 | if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) || | |
5869 | header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) { | |
5870 | memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE + | |
5871 | EXT4_I(inode)->i_extra_isize, 0, | |
5872 | new_extra_isize - EXT4_I(inode)->i_extra_isize); | |
5873 | EXT4_I(inode)->i_extra_isize = new_extra_isize; | |
5874 | return 0; | |
5875 | } | |
5876 | ||
5877 | /* try to expand with EAs present */ | |
5878 | error = ext4_expand_extra_isize_ea(inode, new_extra_isize, | |
5879 | raw_inode, handle); | |
5880 | if (error) { | |
5881 | /* | |
5882 | * Inode size expansion failed; don't try again | |
5883 | */ | |
5884 | *no_expand = 1; | |
5885 | } | |
5886 | ||
5887 | return error; | |
5888 | } | |
5889 | ||
6dd4ee7c KS |
5890 | /* |
5891 | * Expand an inode by new_extra_isize bytes. | |
5892 | * Returns 0 on success or negative error number on failure. | |
5893 | */ | |
cf0a5e81 MX |
5894 | static int ext4_try_to_expand_extra_isize(struct inode *inode, |
5895 | unsigned int new_extra_isize, | |
5896 | struct ext4_iloc iloc, | |
5897 | handle_t *handle) | |
6dd4ee7c | 5898 | { |
3b10fdc6 MX |
5899 | int no_expand; |
5900 | int error; | |
6dd4ee7c | 5901 | |
cf0a5e81 MX |
5902 | if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) |
5903 | return -EOVERFLOW; | |
5904 | ||
5905 | /* | |
5906 | * In nojournal mode, we can immediately attempt to expand | |
5907 | * the inode. When journaled, we first need to obtain extra | |
5908 | * buffer credits since we may write into the EA block | |
5909 | * with this same handle. If journal_extend fails, then it will | |
5910 | * only result in a minor loss of functionality for that inode. | |
5911 | * If this is felt to be critical, then e2fsck should be run to | |
5912 | * force a large enough s_min_extra_isize. | |
5913 | */ | |
6cb367c2 | 5914 | if (ext4_journal_extend(handle, |
83448bdf | 5915 | EXT4_DATA_TRANS_BLOCKS(inode->i_sb), 0) != 0) |
cf0a5e81 | 5916 | return -ENOSPC; |
6dd4ee7c | 5917 | |
3b10fdc6 | 5918 | if (ext4_write_trylock_xattr(inode, &no_expand) == 0) |
cf0a5e81 | 5919 | return -EBUSY; |
3b10fdc6 | 5920 | |
c03b45b8 MX |
5921 | error = __ext4_expand_extra_isize(inode, new_extra_isize, &iloc, |
5922 | handle, &no_expand); | |
5923 | ext4_write_unlock_xattr(inode, &no_expand); | |
6dd4ee7c | 5924 | |
c03b45b8 MX |
5925 | return error; |
5926 | } | |
6dd4ee7c | 5927 | |
c03b45b8 MX |
5928 | int ext4_expand_extra_isize(struct inode *inode, |
5929 | unsigned int new_extra_isize, | |
5930 | struct ext4_iloc *iloc) | |
5931 | { | |
5932 | handle_t *handle; | |
5933 | int no_expand; | |
5934 | int error, rc; | |
5935 | ||
5936 | if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) { | |
5937 | brelse(iloc->bh); | |
5938 | return -EOVERFLOW; | |
6dd4ee7c KS |
5939 | } |
5940 | ||
c03b45b8 MX |
5941 | handle = ext4_journal_start(inode, EXT4_HT_INODE, |
5942 | EXT4_DATA_TRANS_BLOCKS(inode->i_sb)); | |
5943 | if (IS_ERR(handle)) { | |
5944 | error = PTR_ERR(handle); | |
5945 | brelse(iloc->bh); | |
5946 | return error; | |
5947 | } | |
5948 | ||
5949 | ext4_write_lock_xattr(inode, &no_expand); | |
5950 | ||
ddccb6db | 5951 | BUFFER_TRACE(iloc->bh, "get_write_access"); |
188c299e JK |
5952 | error = ext4_journal_get_write_access(handle, inode->i_sb, iloc->bh, |
5953 | EXT4_JTR_NONE); | |
3b10fdc6 | 5954 | if (error) { |
c03b45b8 | 5955 | brelse(iloc->bh); |
7f420d64 | 5956 | goto out_unlock; |
3b10fdc6 | 5957 | } |
cf0a5e81 | 5958 | |
c03b45b8 MX |
5959 | error = __ext4_expand_extra_isize(inode, new_extra_isize, iloc, |
5960 | handle, &no_expand); | |
5961 | ||
5962 | rc = ext4_mark_iloc_dirty(handle, inode, iloc); | |
5963 | if (!error) | |
5964 | error = rc; | |
5965 | ||
7f420d64 | 5966 | out_unlock: |
c03b45b8 | 5967 | ext4_write_unlock_xattr(inode, &no_expand); |
c03b45b8 | 5968 | ext4_journal_stop(handle); |
3b10fdc6 | 5969 | return error; |
6dd4ee7c KS |
5970 | } |
5971 | ||
ac27a0ec DK |
5972 | /* |
5973 | * What we do here is to mark the in-core inode as clean with respect to inode | |
5974 | * dirtiness (it may still be data-dirty). | |
5975 | * This means that the in-core inode may be reaped by prune_icache | |
5976 | * without having to perform any I/O. This is a very good thing, | |
5977 | * because *any* task may call prune_icache - even ones which | |
5978 | * have a transaction open against a different journal. | |
5979 | * | |
5980 | * Is this cheating? Not really. Sure, we haven't written the | |
5981 | * inode out, but prune_icache isn't a user-visible syncing function. | |
5982 | * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync) | |
5983 | * we start and wait on commits. | |
ac27a0ec | 5984 | */ |
4209ae12 HS |
5985 | int __ext4_mark_inode_dirty(handle_t *handle, struct inode *inode, |
5986 | const char *func, unsigned int line) | |
ac27a0ec | 5987 | { |
617ba13b | 5988 | struct ext4_iloc iloc; |
6dd4ee7c | 5989 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
cf0a5e81 | 5990 | int err; |
ac27a0ec DK |
5991 | |
5992 | might_sleep(); | |
7ff9c073 | 5993 | trace_ext4_mark_inode_dirty(inode, _RET_IP_); |
617ba13b | 5994 | err = ext4_reserve_inode_write(handle, inode, &iloc); |
5e1021f2 | 5995 | if (err) |
4209ae12 | 5996 | goto out; |
cf0a5e81 MX |
5997 | |
5998 | if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize) | |
5999 | ext4_try_to_expand_extra_isize(inode, sbi->s_want_extra_isize, | |
6000 | iloc, handle); | |
6001 | ||
4209ae12 HS |
6002 | err = ext4_mark_iloc_dirty(handle, inode, &iloc); |
6003 | out: | |
6004 | if (unlikely(err)) | |
6005 | ext4_error_inode_err(inode, func, line, 0, err, | |
6006 | "mark_inode_dirty error"); | |
6007 | return err; | |
ac27a0ec DK |
6008 | } |
6009 | ||
6010 | /* | |
617ba13b | 6011 | * ext4_dirty_inode() is called from __mark_inode_dirty() |
ac27a0ec DK |
6012 | * |
6013 | * We're really interested in the case where a file is being extended. | |
6014 | * i_size has been changed by generic_commit_write() and we thus need | |
6015 | * to include the updated inode in the current transaction. | |
6016 | * | |
5dd4056d | 6017 | * Also, dquot_alloc_block() will always dirty the inode when blocks |
ac27a0ec DK |
6018 | * are allocated to the file. |
6019 | * | |
6020 | * If the inode is marked synchronous, we don't honour that here - doing | |
6021 | * so would cause a commit on atime updates, which we don't bother doing. | |
6022 | * We handle synchronous inodes at the highest possible level. | |
6023 | */ | |
aa385729 | 6024 | void ext4_dirty_inode(struct inode *inode, int flags) |
ac27a0ec | 6025 | { |
ac27a0ec DK |
6026 | handle_t *handle; |
6027 | ||
9924a92a | 6028 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 2); |
ac27a0ec | 6029 | if (IS_ERR(handle)) |
e2728c56 | 6030 | return; |
f3dc272f | 6031 | ext4_mark_inode_dirty(handle, inode); |
617ba13b | 6032 | ext4_journal_stop(handle); |
ac27a0ec DK |
6033 | } |
6034 | ||
617ba13b | 6035 | int ext4_change_inode_journal_flag(struct inode *inode, int val) |
ac27a0ec DK |
6036 | { |
6037 | journal_t *journal; | |
6038 | handle_t *handle; | |
6039 | int err; | |
c8585c6f | 6040 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
ac27a0ec DK |
6041 | |
6042 | /* | |
6043 | * We have to be very careful here: changing a data block's | |
6044 | * journaling status dynamically is dangerous. If we write a | |
6045 | * data block to the journal, change the status and then delete | |
6046 | * that block, we risk forgetting to revoke the old log record | |
6047 | * from the journal and so a subsequent replay can corrupt data. | |
6048 | * So, first we make sure that the journal is empty and that | |
6049 | * nobody is changing anything. | |
6050 | */ | |
6051 | ||
617ba13b | 6052 | journal = EXT4_JOURNAL(inode); |
0390131b FM |
6053 | if (!journal) |
6054 | return 0; | |
d699594d | 6055 | if (is_journal_aborted(journal)) |
ac27a0ec DK |
6056 | return -EROFS; |
6057 | ||
17335dcc | 6058 | /* Wait for all existing dio workers */ |
17335dcc DM |
6059 | inode_dio_wait(inode); |
6060 | ||
4c546592 DJ |
6061 | /* |
6062 | * Before flushing the journal and switching inode's aops, we have | |
6063 | * to flush all dirty data the inode has. There can be outstanding | |
6064 | * delayed allocations, there can be unwritten extents created by | |
6065 | * fallocate or buffered writes in dioread_nolock mode covered by | |
6066 | * dirty data which can be converted only after flushing the dirty | |
6067 | * data (and journalled aops don't know how to handle these cases). | |
6068 | */ | |
6069 | if (val) { | |
d4f5258e | 6070 | filemap_invalidate_lock(inode->i_mapping); |
4c546592 DJ |
6071 | err = filemap_write_and_wait(inode->i_mapping); |
6072 | if (err < 0) { | |
d4f5258e | 6073 | filemap_invalidate_unlock(inode->i_mapping); |
4c546592 DJ |
6074 | return err; |
6075 | } | |
6076 | } | |
6077 | ||
bbd55937 | 6078 | percpu_down_write(&sbi->s_writepages_rwsem); |
dab291af | 6079 | jbd2_journal_lock_updates(journal); |
ac27a0ec DK |
6080 | |
6081 | /* | |
6082 | * OK, there are no updates running now, and all cached data is | |
6083 | * synced to disk. We are now in a completely consistent state | |
6084 | * which doesn't have anything in the journal, and we know that | |
6085 | * no filesystem updates are running, so it is safe to modify | |
6086 | * the inode's in-core data-journaling state flag now. | |
6087 | */ | |
6088 | ||
6089 | if (val) | |
12e9b892 | 6090 | ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
5872ddaa | 6091 | else { |
01d5d965 | 6092 | err = jbd2_journal_flush(journal, 0); |
4f879ca6 JK |
6093 | if (err < 0) { |
6094 | jbd2_journal_unlock_updates(journal); | |
bbd55937 | 6095 | percpu_up_write(&sbi->s_writepages_rwsem); |
4f879ca6 JK |
6096 | return err; |
6097 | } | |
12e9b892 | 6098 | ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
5872ddaa | 6099 | } |
617ba13b | 6100 | ext4_set_aops(inode); |
ac27a0ec | 6101 | |
dab291af | 6102 | jbd2_journal_unlock_updates(journal); |
bbd55937 | 6103 | percpu_up_write(&sbi->s_writepages_rwsem); |
c8585c6f | 6104 | |
4c546592 | 6105 | if (val) |
d4f5258e | 6106 | filemap_invalidate_unlock(inode->i_mapping); |
ac27a0ec DK |
6107 | |
6108 | /* Finally we can mark the inode as dirty. */ | |
6109 | ||
9924a92a | 6110 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 1); |
ac27a0ec DK |
6111 | if (IS_ERR(handle)) |
6112 | return PTR_ERR(handle); | |
6113 | ||
aa75f4d3 | 6114 | ext4_fc_mark_ineligible(inode->i_sb, |
e85c81ba | 6115 | EXT4_FC_REASON_JOURNAL_FLAG_CHANGE, handle); |
617ba13b | 6116 | err = ext4_mark_inode_dirty(handle, inode); |
0390131b | 6117 | ext4_handle_sync(handle); |
617ba13b MC |
6118 | ext4_journal_stop(handle); |
6119 | ext4_std_error(inode->i_sb, err); | |
ac27a0ec DK |
6120 | |
6121 | return err; | |
6122 | } | |
2e9ee850 | 6123 | |
188c299e JK |
6124 | static int ext4_bh_unmapped(handle_t *handle, struct inode *inode, |
6125 | struct buffer_head *bh) | |
2e9ee850 AK |
6126 | { |
6127 | return !buffer_mapped(bh); | |
6128 | } | |
6129 | ||
401b25aa | 6130 | vm_fault_t ext4_page_mkwrite(struct vm_fault *vmf) |
2e9ee850 | 6131 | { |
11bac800 | 6132 | struct vm_area_struct *vma = vmf->vma; |
c2ec175c | 6133 | struct page *page = vmf->page; |
2e9ee850 AK |
6134 | loff_t size; |
6135 | unsigned long len; | |
401b25aa SJ |
6136 | int err; |
6137 | vm_fault_t ret; | |
2e9ee850 | 6138 | struct file *file = vma->vm_file; |
496ad9aa | 6139 | struct inode *inode = file_inode(file); |
2e9ee850 | 6140 | struct address_space *mapping = inode->i_mapping; |
9ea7df53 JK |
6141 | handle_t *handle; |
6142 | get_block_t *get_block; | |
6143 | int retries = 0; | |
2e9ee850 | 6144 | |
02b016ca TT |
6145 | if (unlikely(IS_IMMUTABLE(inode))) |
6146 | return VM_FAULT_SIGBUS; | |
6147 | ||
8e8ad8a5 | 6148 | sb_start_pagefault(inode->i_sb); |
041bbb6d | 6149 | file_update_time(vma->vm_file); |
ea3d7209 | 6150 | |
d4f5258e | 6151 | filemap_invalidate_lock_shared(mapping); |
7b4cc978 | 6152 | |
401b25aa SJ |
6153 | err = ext4_convert_inline_data(inode); |
6154 | if (err) | |
7b4cc978 EB |
6155 | goto out_ret; |
6156 | ||
64a9f144 MFO |
6157 | /* |
6158 | * On data journalling we skip straight to the transaction handle: | |
6159 | * there's no delalloc; page truncated will be checked later; the | |
6160 | * early return w/ all buffers mapped (calculates size/len) can't | |
6161 | * be used; and there's no dioread_nolock, so only ext4_get_block. | |
6162 | */ | |
6163 | if (ext4_should_journal_data(inode)) | |
6164 | goto retry_alloc; | |
6165 | ||
9ea7df53 JK |
6166 | /* Delalloc case is easy... */ |
6167 | if (test_opt(inode->i_sb, DELALLOC) && | |
9ea7df53 JK |
6168 | !ext4_nonda_switch(inode->i_sb)) { |
6169 | do { | |
401b25aa | 6170 | err = block_page_mkwrite(vma, vmf, |
9ea7df53 | 6171 | ext4_da_get_block_prep); |
401b25aa | 6172 | } while (err == -ENOSPC && |
9ea7df53 JK |
6173 | ext4_should_retry_alloc(inode->i_sb, &retries)); |
6174 | goto out_ret; | |
2e9ee850 | 6175 | } |
0e499890 DW |
6176 | |
6177 | lock_page(page); | |
9ea7df53 JK |
6178 | size = i_size_read(inode); |
6179 | /* Page got truncated from under us? */ | |
6180 | if (page->mapping != mapping || page_offset(page) > size) { | |
6181 | unlock_page(page); | |
6182 | ret = VM_FAULT_NOPAGE; | |
6183 | goto out; | |
0e499890 | 6184 | } |
2e9ee850 | 6185 | |
09cbfeaf KS |
6186 | if (page->index == size >> PAGE_SHIFT) |
6187 | len = size & ~PAGE_MASK; | |
2e9ee850 | 6188 | else |
09cbfeaf | 6189 | len = PAGE_SIZE; |
a827eaff | 6190 | /* |
9ea7df53 JK |
6191 | * Return if we have all the buffers mapped. This avoids the need to do |
6192 | * journal_start/journal_stop which can block and take a long time | |
64a9f144 MFO |
6193 | * |
6194 | * This cannot be done for data journalling, as we have to add the | |
6195 | * inode to the transaction's list to writeprotect pages on commit. | |
a827eaff | 6196 | */ |
2e9ee850 | 6197 | if (page_has_buffers(page)) { |
188c299e | 6198 | if (!ext4_walk_page_buffers(NULL, inode, page_buffers(page), |
f19d5870 TM |
6199 | 0, len, NULL, |
6200 | ext4_bh_unmapped)) { | |
9ea7df53 | 6201 | /* Wait so that we don't change page under IO */ |
1d1d1a76 | 6202 | wait_for_stable_page(page); |
9ea7df53 JK |
6203 | ret = VM_FAULT_LOCKED; |
6204 | goto out; | |
a827eaff | 6205 | } |
2e9ee850 | 6206 | } |
a827eaff | 6207 | unlock_page(page); |
9ea7df53 JK |
6208 | /* OK, we need to fill the hole... */ |
6209 | if (ext4_should_dioread_nolock(inode)) | |
705965bd | 6210 | get_block = ext4_get_block_unwritten; |
9ea7df53 JK |
6211 | else |
6212 | get_block = ext4_get_block; | |
6213 | retry_alloc: | |
9924a92a TT |
6214 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
6215 | ext4_writepage_trans_blocks(inode)); | |
9ea7df53 | 6216 | if (IS_ERR(handle)) { |
c2ec175c | 6217 | ret = VM_FAULT_SIGBUS; |
9ea7df53 JK |
6218 | goto out; |
6219 | } | |
64a9f144 MFO |
6220 | /* |
6221 | * Data journalling can't use block_page_mkwrite() because it | |
6222 | * will set_buffer_dirty() before do_journal_get_write_access() | |
6223 | * thus might hit warning messages for dirty metadata buffers. | |
6224 | */ | |
6225 | if (!ext4_should_journal_data(inode)) { | |
6226 | err = block_page_mkwrite(vma, vmf, get_block); | |
6227 | } else { | |
6228 | lock_page(page); | |
6229 | size = i_size_read(inode); | |
6230 | /* Page got truncated from under us? */ | |
6231 | if (page->mapping != mapping || page_offset(page) > size) { | |
64a9f144 | 6232 | ret = VM_FAULT_NOPAGE; |
afb585a9 | 6233 | goto out_error; |
9ea7df53 | 6234 | } |
64a9f144 MFO |
6235 | |
6236 | if (page->index == size >> PAGE_SHIFT) | |
6237 | len = size & ~PAGE_MASK; | |
6238 | else | |
6239 | len = PAGE_SIZE; | |
6240 | ||
6241 | err = __block_write_begin(page, 0, len, ext4_get_block); | |
6242 | if (!err) { | |
afb585a9 | 6243 | ret = VM_FAULT_SIGBUS; |
188c299e JK |
6244 | if (ext4_walk_page_buffers(handle, inode, |
6245 | page_buffers(page), 0, len, NULL, | |
6246 | do_journal_get_write_access)) | |
afb585a9 | 6247 | goto out_error; |
188c299e JK |
6248 | if (ext4_walk_page_buffers(handle, inode, |
6249 | page_buffers(page), 0, len, NULL, | |
6250 | write_end_fn)) | |
afb585a9 | 6251 | goto out_error; |
b5b18160 JK |
6252 | if (ext4_jbd2_inode_add_write(handle, inode, |
6253 | page_offset(page), len)) | |
afb585a9 | 6254 | goto out_error; |
64a9f144 MFO |
6255 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
6256 | } else { | |
6257 | unlock_page(page); | |
6258 | } | |
9ea7df53 JK |
6259 | } |
6260 | ext4_journal_stop(handle); | |
401b25aa | 6261 | if (err == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) |
9ea7df53 JK |
6262 | goto retry_alloc; |
6263 | out_ret: | |
401b25aa | 6264 | ret = block_page_mkwrite_return(err); |
9ea7df53 | 6265 | out: |
d4f5258e | 6266 | filemap_invalidate_unlock_shared(mapping); |
8e8ad8a5 | 6267 | sb_end_pagefault(inode->i_sb); |
2e9ee850 | 6268 | return ret; |
afb585a9 MFO |
6269 | out_error: |
6270 | unlock_page(page); | |
6271 | ext4_journal_stop(handle); | |
6272 | goto out; | |
2e9ee850 | 6273 | } |