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 | ||
dec214d0 TE |
139 | static int ext4_meta_trans_blocks(struct inode *inode, int lblocks, |
140 | int pextents); | |
64769240 | 141 | |
ac27a0ec DK |
142 | /* |
143 | * Test whether an inode is a fast symlink. | |
407cd7fb | 144 | * A fast symlink has its symlink data stored in ext4_inode_info->i_data. |
ac27a0ec | 145 | */ |
f348c252 | 146 | int ext4_inode_is_fast_symlink(struct inode *inode) |
ac27a0ec | 147 | { |
fc82228a AK |
148 | if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) { |
149 | int ea_blocks = EXT4_I(inode)->i_file_acl ? | |
150 | EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0; | |
151 | ||
152 | if (ext4_has_inline_data(inode)) | |
153 | return 0; | |
154 | ||
155 | return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0); | |
156 | } | |
407cd7fb TE |
157 | return S_ISLNK(inode->i_mode) && inode->i_size && |
158 | (inode->i_size < EXT4_N_BLOCKS * 4); | |
ac27a0ec DK |
159 | } |
160 | ||
ac27a0ec DK |
161 | /* |
162 | * Called at the last iput() if i_nlink is zero. | |
163 | */ | |
0930fcc1 | 164 | void ext4_evict_inode(struct inode *inode) |
ac27a0ec DK |
165 | { |
166 | handle_t *handle; | |
bc965ab3 | 167 | int err; |
65db869c JK |
168 | /* |
169 | * Credits for final inode cleanup and freeing: | |
170 | * sb + inode (ext4_orphan_del()), block bitmap, group descriptor | |
171 | * (xattr block freeing), bitmap, group descriptor (inode freeing) | |
172 | */ | |
173 | int extra_credits = 6; | |
0421a189 | 174 | struct ext4_xattr_inode_array *ea_inode_array = NULL; |
46e294ef | 175 | bool freeze_protected = false; |
ac27a0ec | 176 | |
7ff9c073 | 177 | trace_ext4_evict_inode(inode); |
2581fdc8 | 178 | |
6bc0d63d JK |
179 | if (EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL) |
180 | ext4_evict_ea_inode(inode); | |
0930fcc1 | 181 | if (inode->i_nlink) { |
91b0abe3 | 182 | truncate_inode_pages_final(&inode->i_data); |
5dc23bdd | 183 | |
0930fcc1 AV |
184 | goto no_delete; |
185 | } | |
186 | ||
e2bfb088 TT |
187 | if (is_bad_inode(inode)) |
188 | goto no_delete; | |
189 | dquot_initialize(inode); | |
907f4554 | 190 | |
678aaf48 JK |
191 | if (ext4_should_order_data(inode)) |
192 | ext4_begin_ordered_truncate(inode, 0); | |
91b0abe3 | 193 | truncate_inode_pages_final(&inode->i_data); |
ac27a0ec | 194 | |
ceff86fd JK |
195 | /* |
196 | * For inodes with journalled data, transaction commit could have | |
bc12ac98 ZY |
197 | * dirtied the inode. And for inodes with dioread_nolock, unwritten |
198 | * extents converting worker could merge extents and also have dirtied | |
199 | * the inode. Flush worker is ignoring it because of I_FREEING flag but | |
200 | * we still need to remove the inode from the writeback lists. | |
ceff86fd | 201 | */ |
bc12ac98 | 202 | if (!list_empty_careful(&inode->i_io_list)) |
ceff86fd | 203 | inode_io_list_del(inode); |
ceff86fd | 204 | |
8e8ad8a5 JK |
205 | /* |
206 | * Protect us against freezing - iput() caller didn't have to have any | |
46e294ef JK |
207 | * protection against it. When we are in a running transaction though, |
208 | * we are already protected against freezing and we cannot grab further | |
209 | * protection due to lock ordering constraints. | |
8e8ad8a5 | 210 | */ |
46e294ef JK |
211 | if (!ext4_journal_current_handle()) { |
212 | sb_start_intwrite(inode->i_sb); | |
213 | freeze_protected = true; | |
214 | } | |
e50e5129 | 215 | |
30a7eb97 TE |
216 | if (!IS_NOQUOTA(inode)) |
217 | extra_credits += EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb); | |
218 | ||
65db869c JK |
219 | /* |
220 | * Block bitmap, group descriptor, and inode are accounted in both | |
221 | * ext4_blocks_for_truncate() and extra_credits. So subtract 3. | |
222 | */ | |
30a7eb97 | 223 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, |
65db869c | 224 | ext4_blocks_for_truncate(inode) + extra_credits - 3); |
ac27a0ec | 225 | if (IS_ERR(handle)) { |
bc965ab3 | 226 | ext4_std_error(inode->i_sb, PTR_ERR(handle)); |
ac27a0ec DK |
227 | /* |
228 | * If we're going to skip the normal cleanup, we still need to | |
229 | * make sure that the in-core orphan linked list is properly | |
230 | * cleaned up. | |
231 | */ | |
617ba13b | 232 | ext4_orphan_del(NULL, inode); |
46e294ef JK |
233 | if (freeze_protected) |
234 | sb_end_intwrite(inode->i_sb); | |
ac27a0ec DK |
235 | goto no_delete; |
236 | } | |
30a7eb97 | 237 | |
ac27a0ec | 238 | if (IS_SYNC(inode)) |
0390131b | 239 | ext4_handle_sync(handle); |
407cd7fb TE |
240 | |
241 | /* | |
242 | * Set inode->i_size to 0 before calling ext4_truncate(). We need | |
243 | * special handling of symlinks here because i_size is used to | |
244 | * determine whether ext4_inode_info->i_data contains symlink data or | |
245 | * block mappings. Setting i_size to 0 will remove its fast symlink | |
246 | * status. Erase i_data so that it becomes a valid empty block map. | |
247 | */ | |
248 | if (ext4_inode_is_fast_symlink(inode)) | |
249 | memset(EXT4_I(inode)->i_data, 0, sizeof(EXT4_I(inode)->i_data)); | |
ac27a0ec | 250 | inode->i_size = 0; |
bc965ab3 TT |
251 | err = ext4_mark_inode_dirty(handle, inode); |
252 | if (err) { | |
12062ddd | 253 | ext4_warning(inode->i_sb, |
bc965ab3 TT |
254 | "couldn't mark inode dirty (err %d)", err); |
255 | goto stop_handle; | |
256 | } | |
2c98eb5e TT |
257 | if (inode->i_blocks) { |
258 | err = ext4_truncate(inode); | |
259 | if (err) { | |
54d3adbc TT |
260 | ext4_error_err(inode->i_sb, -err, |
261 | "couldn't truncate inode %lu (err %d)", | |
262 | inode->i_ino, err); | |
2c98eb5e TT |
263 | goto stop_handle; |
264 | } | |
265 | } | |
bc965ab3 | 266 | |
30a7eb97 TE |
267 | /* Remove xattr references. */ |
268 | err = ext4_xattr_delete_inode(handle, inode, &ea_inode_array, | |
269 | extra_credits); | |
270 | if (err) { | |
271 | ext4_warning(inode->i_sb, "xattr delete (err %d)", err); | |
272 | stop_handle: | |
273 | ext4_journal_stop(handle); | |
274 | ext4_orphan_del(NULL, inode); | |
46e294ef JK |
275 | if (freeze_protected) |
276 | sb_end_intwrite(inode->i_sb); | |
30a7eb97 TE |
277 | ext4_xattr_inode_array_free(ea_inode_array); |
278 | goto no_delete; | |
bc965ab3 TT |
279 | } |
280 | ||
ac27a0ec | 281 | /* |
617ba13b | 282 | * Kill off the orphan record which ext4_truncate created. |
ac27a0ec | 283 | * AKPM: I think this can be inside the above `if'. |
617ba13b | 284 | * Note that ext4_orphan_del() has to be able to cope with the |
ac27a0ec | 285 | * deletion of a non-existent orphan - this is because we don't |
617ba13b | 286 | * know if ext4_truncate() actually created an orphan record. |
ac27a0ec DK |
287 | * (Well, we could do this if we need to, but heck - it works) |
288 | */ | |
617ba13b | 289 | ext4_orphan_del(handle, inode); |
5ffff834 | 290 | EXT4_I(inode)->i_dtime = (__u32)ktime_get_real_seconds(); |
ac27a0ec DK |
291 | |
292 | /* | |
293 | * One subtle ordering requirement: if anything has gone wrong | |
294 | * (transaction abort, IO errors, whatever), then we can still | |
295 | * do these next steps (the fs will already have been marked as | |
296 | * having errors), but we can't free the inode if the mark_dirty | |
297 | * fails. | |
298 | */ | |
617ba13b | 299 | if (ext4_mark_inode_dirty(handle, inode)) |
ac27a0ec | 300 | /* If that failed, just do the required in-core inode clear. */ |
0930fcc1 | 301 | ext4_clear_inode(inode); |
ac27a0ec | 302 | else |
617ba13b MC |
303 | ext4_free_inode(handle, inode); |
304 | ext4_journal_stop(handle); | |
46e294ef JK |
305 | if (freeze_protected) |
306 | sb_end_intwrite(inode->i_sb); | |
0421a189 | 307 | ext4_xattr_inode_array_free(ea_inode_array); |
ac27a0ec DK |
308 | return; |
309 | no_delete: | |
318cdc82 ZY |
310 | /* |
311 | * Check out some where else accidentally dirty the evicting inode, | |
312 | * which may probably cause inode use-after-free issues later. | |
313 | */ | |
314 | WARN_ON_ONCE(!list_empty_careful(&inode->i_io_list)); | |
315 | ||
b21ebf14 | 316 | if (!list_empty(&EXT4_I(inode)->i_fc_list)) |
e85c81ba | 317 | ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_NOMEM, NULL); |
0930fcc1 | 318 | ext4_clear_inode(inode); /* We must guarantee clearing of inode... */ |
ac27a0ec DK |
319 | } |
320 | ||
a9e7f447 DM |
321 | #ifdef CONFIG_QUOTA |
322 | qsize_t *ext4_get_reserved_space(struct inode *inode) | |
60e58e0f | 323 | { |
a9e7f447 | 324 | return &EXT4_I(inode)->i_reserved_quota; |
60e58e0f | 325 | } |
a9e7f447 | 326 | #endif |
9d0be502 | 327 | |
0637c6f4 TT |
328 | /* |
329 | * Called with i_data_sem down, which is important since we can call | |
330 | * ext4_discard_preallocations() from here. | |
331 | */ | |
5f634d06 AK |
332 | void ext4_da_update_reserve_space(struct inode *inode, |
333 | int used, int quota_claim) | |
12219aea AK |
334 | { |
335 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0637c6f4 | 336 | struct ext4_inode_info *ei = EXT4_I(inode); |
0637c6f4 TT |
337 | |
338 | spin_lock(&ei->i_block_reservation_lock); | |
d8990240 | 339 | trace_ext4_da_update_reserve_space(inode, used, quota_claim); |
0637c6f4 | 340 | if (unlikely(used > ei->i_reserved_data_blocks)) { |
8de5c325 | 341 | ext4_warning(inode->i_sb, "%s: ino %lu, used %d " |
1084f252 | 342 | "with only %d reserved data blocks", |
0637c6f4 TT |
343 | __func__, inode->i_ino, used, |
344 | ei->i_reserved_data_blocks); | |
345 | WARN_ON(1); | |
346 | used = ei->i_reserved_data_blocks; | |
347 | } | |
12219aea | 348 | |
0637c6f4 TT |
349 | /* Update per-inode reservations */ |
350 | ei->i_reserved_data_blocks -= used; | |
71d4f7d0 | 351 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, used); |
6bc6e63f | 352 | |
f9505c72 | 353 | spin_unlock(&ei->i_block_reservation_lock); |
60e58e0f | 354 | |
72b8ab9d ES |
355 | /* Update quota subsystem for data blocks */ |
356 | if (quota_claim) | |
7b415bf6 | 357 | dquot_claim_block(inode, EXT4_C2B(sbi, used)); |
72b8ab9d | 358 | else { |
5f634d06 AK |
359 | /* |
360 | * We did fallocate with an offset that is already delayed | |
361 | * allocated. So on delayed allocated writeback we should | |
72b8ab9d | 362 | * not re-claim the quota for fallocated blocks. |
5f634d06 | 363 | */ |
7b415bf6 | 364 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, used)); |
5f634d06 | 365 | } |
d6014301 AK |
366 | |
367 | /* | |
368 | * If we have done all the pending block allocations and if | |
369 | * there aren't any writers on the inode, we can discard the | |
370 | * inode's preallocations. | |
371 | */ | |
0637c6f4 | 372 | if ((ei->i_reserved_data_blocks == 0) && |
82dd124c | 373 | !inode_is_open_for_write(inode)) |
27bc446e | 374 | ext4_discard_preallocations(inode, 0); |
12219aea AK |
375 | } |
376 | ||
e29136f8 | 377 | static int __check_block_validity(struct inode *inode, const char *func, |
c398eda0 TT |
378 | unsigned int line, |
379 | struct ext4_map_blocks *map) | |
6fd058f7 | 380 | { |
345c0dbf TT |
381 | if (ext4_has_feature_journal(inode->i_sb) && |
382 | (inode->i_ino == | |
383 | le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_journal_inum))) | |
384 | return 0; | |
ce9f24cc | 385 | if (!ext4_inode_block_valid(inode, map->m_pblk, map->m_len)) { |
c398eda0 | 386 | ext4_error_inode(inode, func, line, map->m_pblk, |
bdbd6ce0 | 387 | "lblock %lu mapped to illegal pblock %llu " |
c398eda0 | 388 | "(length %d)", (unsigned long) map->m_lblk, |
bdbd6ce0 | 389 | map->m_pblk, map->m_len); |
6a797d27 | 390 | return -EFSCORRUPTED; |
6fd058f7 TT |
391 | } |
392 | return 0; | |
393 | } | |
394 | ||
53085fac JK |
395 | int ext4_issue_zeroout(struct inode *inode, ext4_lblk_t lblk, ext4_fsblk_t pblk, |
396 | ext4_lblk_t len) | |
397 | { | |
398 | int ret; | |
399 | ||
33b4cc25 | 400 | if (IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode)) |
a7550b30 | 401 | return fscrypt_zeroout_range(inode, lblk, pblk, len); |
53085fac JK |
402 | |
403 | ret = sb_issue_zeroout(inode->i_sb, pblk, len, GFP_NOFS); | |
404 | if (ret > 0) | |
405 | ret = 0; | |
406 | ||
407 | return ret; | |
408 | } | |
409 | ||
e29136f8 | 410 | #define check_block_validity(inode, map) \ |
c398eda0 | 411 | __check_block_validity((inode), __func__, __LINE__, (map)) |
e29136f8 | 412 | |
921f266b DM |
413 | #ifdef ES_AGGRESSIVE_TEST |
414 | static void ext4_map_blocks_es_recheck(handle_t *handle, | |
415 | struct inode *inode, | |
416 | struct ext4_map_blocks *es_map, | |
417 | struct ext4_map_blocks *map, | |
418 | int flags) | |
419 | { | |
420 | int retval; | |
421 | ||
422 | map->m_flags = 0; | |
423 | /* | |
424 | * There is a race window that the result is not the same. | |
425 | * e.g. xfstests #223 when dioread_nolock enables. The reason | |
426 | * is that we lookup a block mapping in extent status tree with | |
427 | * out taking i_data_sem. So at the time the unwritten extent | |
428 | * could be converted. | |
429 | */ | |
2dcba478 | 430 | down_read(&EXT4_I(inode)->i_data_sem); |
921f266b | 431 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
9e52484c | 432 | retval = ext4_ext_map_blocks(handle, inode, map, 0); |
921f266b | 433 | } else { |
9e52484c | 434 | retval = ext4_ind_map_blocks(handle, inode, map, 0); |
921f266b | 435 | } |
2dcba478 | 436 | up_read((&EXT4_I(inode)->i_data_sem)); |
921f266b DM |
437 | |
438 | /* | |
439 | * We don't check m_len because extent will be collpased in status | |
440 | * tree. So the m_len might not equal. | |
441 | */ | |
442 | if (es_map->m_lblk != map->m_lblk || | |
443 | es_map->m_flags != map->m_flags || | |
444 | es_map->m_pblk != map->m_pblk) { | |
bdafe42a | 445 | printk("ES cache assertion failed for inode: %lu " |
921f266b DM |
446 | "es_cached ex [%d/%d/%llu/%x] != " |
447 | "found ex [%d/%d/%llu/%x] retval %d flags %x\n", | |
448 | inode->i_ino, es_map->m_lblk, es_map->m_len, | |
449 | es_map->m_pblk, es_map->m_flags, map->m_lblk, | |
450 | map->m_len, map->m_pblk, map->m_flags, | |
451 | retval, flags); | |
452 | } | |
453 | } | |
454 | #endif /* ES_AGGRESSIVE_TEST */ | |
455 | ||
f5ab0d1f | 456 | /* |
e35fd660 | 457 | * The ext4_map_blocks() function tries to look up the requested blocks, |
2b2d6d01 | 458 | * and returns if the blocks are already mapped. |
f5ab0d1f | 459 | * |
f5ab0d1f MC |
460 | * Otherwise it takes the write lock of the i_data_sem and allocate blocks |
461 | * and store the allocated blocks in the result buffer head and mark it | |
462 | * mapped. | |
463 | * | |
e35fd660 TT |
464 | * If file type is extents based, it will call ext4_ext_map_blocks(), |
465 | * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping | |
f5ab0d1f MC |
466 | * based files |
467 | * | |
facab4d9 JK |
468 | * On success, it returns the number of blocks being mapped or allocated. if |
469 | * create==0 and the blocks are pre-allocated and unwritten, the resulting @map | |
470 | * is marked as unwritten. If the create == 1, it will mark @map as mapped. | |
f5ab0d1f MC |
471 | * |
472 | * It returns 0 if plain look up failed (blocks have not been allocated), in | |
facab4d9 JK |
473 | * that case, @map is returned as unmapped but we still do fill map->m_len to |
474 | * indicate the length of a hole starting at map->m_lblk. | |
f5ab0d1f MC |
475 | * |
476 | * It returns the error in case of allocation failure. | |
477 | */ | |
e35fd660 TT |
478 | int ext4_map_blocks(handle_t *handle, struct inode *inode, |
479 | struct ext4_map_blocks *map, int flags) | |
0e855ac8 | 480 | { |
d100eef2 | 481 | struct extent_status es; |
0e855ac8 | 482 | int retval; |
b8a86845 | 483 | int ret = 0; |
921f266b DM |
484 | #ifdef ES_AGGRESSIVE_TEST |
485 | struct ext4_map_blocks orig_map; | |
486 | ||
487 | memcpy(&orig_map, map, sizeof(*map)); | |
488 | #endif | |
f5ab0d1f | 489 | |
e35fd660 | 490 | map->m_flags = 0; |
70aa1554 RH |
491 | ext_debug(inode, "flag 0x%x, max_blocks %u, logical block %lu\n", |
492 | flags, map->m_len, (unsigned long) map->m_lblk); | |
d100eef2 | 493 | |
e861b5e9 TT |
494 | /* |
495 | * ext4_map_blocks returns an int, and m_len is an unsigned int | |
496 | */ | |
497 | if (unlikely(map->m_len > INT_MAX)) | |
498 | map->m_len = INT_MAX; | |
499 | ||
4adb6ab3 KM |
500 | /* We can handle the block number less than EXT_MAX_BLOCKS */ |
501 | if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS)) | |
6a797d27 | 502 | return -EFSCORRUPTED; |
4adb6ab3 | 503 | |
d100eef2 | 504 | /* Lookup extent status tree firstly */ |
8016e29f HS |
505 | if (!(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY) && |
506 | ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) { | |
d100eef2 ZL |
507 | if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) { |
508 | map->m_pblk = ext4_es_pblock(&es) + | |
509 | map->m_lblk - es.es_lblk; | |
510 | map->m_flags |= ext4_es_is_written(&es) ? | |
511 | EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN; | |
512 | retval = es.es_len - (map->m_lblk - es.es_lblk); | |
513 | if (retval > map->m_len) | |
514 | retval = map->m_len; | |
515 | map->m_len = retval; | |
516 | } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) { | |
facab4d9 JK |
517 | map->m_pblk = 0; |
518 | retval = es.es_len - (map->m_lblk - es.es_lblk); | |
519 | if (retval > map->m_len) | |
520 | retval = map->m_len; | |
521 | map->m_len = retval; | |
d100eef2 ZL |
522 | retval = 0; |
523 | } else { | |
1e83bc81 | 524 | BUG(); |
d100eef2 | 525 | } |
9558cf14 ZY |
526 | |
527 | if (flags & EXT4_GET_BLOCKS_CACHED_NOWAIT) | |
528 | return retval; | |
921f266b DM |
529 | #ifdef ES_AGGRESSIVE_TEST |
530 | ext4_map_blocks_es_recheck(handle, inode, map, | |
531 | &orig_map, flags); | |
532 | #endif | |
d100eef2 ZL |
533 | goto found; |
534 | } | |
9558cf14 ZY |
535 | /* |
536 | * In the query cache no-wait mode, nothing we can do more if we | |
537 | * cannot find extent in the cache. | |
538 | */ | |
539 | if (flags & EXT4_GET_BLOCKS_CACHED_NOWAIT) | |
540 | return 0; | |
d100eef2 | 541 | |
4df3d265 | 542 | /* |
b920c755 TT |
543 | * Try to see if we can get the block without requesting a new |
544 | * file system block. | |
4df3d265 | 545 | */ |
2dcba478 | 546 | down_read(&EXT4_I(inode)->i_data_sem); |
12e9b892 | 547 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
9e52484c | 548 | retval = ext4_ext_map_blocks(handle, inode, map, 0); |
0e855ac8 | 549 | } else { |
9e52484c | 550 | retval = ext4_ind_map_blocks(handle, inode, map, 0); |
0e855ac8 | 551 | } |
f7fec032 | 552 | if (retval > 0) { |
3be78c73 | 553 | unsigned int status; |
f7fec032 | 554 | |
44fb851d ZL |
555 | if (unlikely(retval != map->m_len)) { |
556 | ext4_warning(inode->i_sb, | |
557 | "ES len assertion failed for inode " | |
558 | "%lu: retval %d != map->m_len %d", | |
559 | inode->i_ino, retval, map->m_len); | |
560 | WARN_ON(1); | |
921f266b | 561 | } |
921f266b | 562 | |
f7fec032 ZL |
563 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
564 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; | |
565 | if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) && | |
d2dc317d | 566 | !(status & EXTENT_STATUS_WRITTEN) && |
ad431025 EW |
567 | ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk, |
568 | map->m_lblk + map->m_len - 1)) | |
f7fec032 | 569 | status |= EXTENT_STATUS_DELAYED; |
6c120399 BL |
570 | ext4_es_insert_extent(inode, map->m_lblk, map->m_len, |
571 | map->m_pblk, status); | |
f7fec032 | 572 | } |
2dcba478 | 573 | up_read((&EXT4_I(inode)->i_data_sem)); |
f5ab0d1f | 574 | |
d100eef2 | 575 | found: |
e35fd660 | 576 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
b8a86845 | 577 | ret = check_block_validity(inode, map); |
6fd058f7 TT |
578 | if (ret != 0) |
579 | return ret; | |
580 | } | |
581 | ||
f5ab0d1f | 582 | /* If it is only a block(s) look up */ |
c2177057 | 583 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) |
f5ab0d1f MC |
584 | return retval; |
585 | ||
586 | /* | |
587 | * Returns if the blocks have already allocated | |
588 | * | |
589 | * Note that if blocks have been preallocated | |
df3ab170 | 590 | * ext4_ext_get_block() returns the create = 0 |
f5ab0d1f MC |
591 | * with buffer head unmapped. |
592 | */ | |
e35fd660 | 593 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) |
b8a86845 LC |
594 | /* |
595 | * If we need to convert extent to unwritten | |
596 | * we continue and do the actual work in | |
597 | * ext4_ext_map_blocks() | |
598 | */ | |
599 | if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) | |
600 | return retval; | |
4df3d265 | 601 | |
2a8964d6 | 602 | /* |
a25a4e1a ZL |
603 | * Here we clear m_flags because after allocating an new extent, |
604 | * it will be set again. | |
2a8964d6 | 605 | */ |
a25a4e1a | 606 | map->m_flags &= ~EXT4_MAP_FLAGS; |
2a8964d6 | 607 | |
4df3d265 | 608 | /* |
556615dc | 609 | * New blocks allocate and/or writing to unwritten extent |
f5ab0d1f | 610 | * will possibly result in updating i_data, so we take |
d91bd2c1 | 611 | * the write lock of i_data_sem, and call get_block() |
f5ab0d1f | 612 | * with create == 1 flag. |
4df3d265 | 613 | */ |
c8b459f4 | 614 | down_write(&EXT4_I(inode)->i_data_sem); |
d2a17637 | 615 | |
4df3d265 AK |
616 | /* |
617 | * We need to check for EXT4 here because migrate | |
618 | * could have changed the inode type in between | |
619 | */ | |
12e9b892 | 620 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
e35fd660 | 621 | retval = ext4_ext_map_blocks(handle, inode, map, flags); |
0e855ac8 | 622 | } else { |
e35fd660 | 623 | retval = ext4_ind_map_blocks(handle, inode, map, flags); |
267e4db9 | 624 | |
e35fd660 | 625 | if (retval > 0 && map->m_flags & EXT4_MAP_NEW) { |
267e4db9 AK |
626 | /* |
627 | * We allocated new blocks which will result in | |
628 | * i_data's format changing. Force the migrate | |
629 | * to fail by clearing migrate flags | |
630 | */ | |
19f5fb7a | 631 | ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE); |
267e4db9 | 632 | } |
5f634d06 | 633 | } |
2ac3b6e0 | 634 | |
f7fec032 | 635 | if (retval > 0) { |
3be78c73 | 636 | unsigned int status; |
f7fec032 | 637 | |
44fb851d ZL |
638 | if (unlikely(retval != map->m_len)) { |
639 | ext4_warning(inode->i_sb, | |
640 | "ES len assertion failed for inode " | |
641 | "%lu: retval %d != map->m_len %d", | |
642 | inode->i_ino, retval, map->m_len); | |
643 | WARN_ON(1); | |
921f266b | 644 | } |
921f266b | 645 | |
c86d8db3 JK |
646 | /* |
647 | * We have to zeroout blocks before inserting them into extent | |
648 | * status tree. Otherwise someone could look them up there and | |
9b623df6 JK |
649 | * use them before they are really zeroed. We also have to |
650 | * unmap metadata before zeroing as otherwise writeback can | |
651 | * overwrite zeros with stale data from block device. | |
c86d8db3 JK |
652 | */ |
653 | if (flags & EXT4_GET_BLOCKS_ZERO && | |
654 | map->m_flags & EXT4_MAP_MAPPED && | |
655 | map->m_flags & EXT4_MAP_NEW) { | |
656 | ret = ext4_issue_zeroout(inode, map->m_lblk, | |
657 | map->m_pblk, map->m_len); | |
658 | if (ret) { | |
659 | retval = ret; | |
660 | goto out_sem; | |
661 | } | |
662 | } | |
663 | ||
adb23551 ZL |
664 | /* |
665 | * If the extent has been zeroed out, we don't need to update | |
666 | * extent status tree. | |
667 | */ | |
668 | if ((flags & EXT4_GET_BLOCKS_PRE_IO) && | |
bb5835ed | 669 | ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) { |
adb23551 | 670 | if (ext4_es_is_written(&es)) |
c86d8db3 | 671 | goto out_sem; |
adb23551 | 672 | } |
f7fec032 ZL |
673 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
674 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; | |
675 | if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) && | |
d2dc317d | 676 | !(status & EXTENT_STATUS_WRITTEN) && |
ad431025 EW |
677 | ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk, |
678 | map->m_lblk + map->m_len - 1)) | |
f7fec032 | 679 | status |= EXTENT_STATUS_DELAYED; |
6c120399 BL |
680 | ext4_es_insert_extent(inode, map->m_lblk, map->m_len, |
681 | map->m_pblk, status); | |
5356f261 AK |
682 | } |
683 | ||
c86d8db3 | 684 | out_sem: |
4df3d265 | 685 | up_write((&EXT4_I(inode)->i_data_sem)); |
e35fd660 | 686 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
b8a86845 | 687 | ret = check_block_validity(inode, map); |
6fd058f7 TT |
688 | if (ret != 0) |
689 | return ret; | |
06bd3c36 JK |
690 | |
691 | /* | |
692 | * Inodes with freshly allocated blocks where contents will be | |
693 | * visible after transaction commit must be on transaction's | |
694 | * ordered data list. | |
695 | */ | |
696 | if (map->m_flags & EXT4_MAP_NEW && | |
697 | !(map->m_flags & EXT4_MAP_UNWRITTEN) && | |
698 | !(flags & EXT4_GET_BLOCKS_ZERO) && | |
02749a4c | 699 | !ext4_is_quota_file(inode) && |
06bd3c36 | 700 | ext4_should_order_data(inode)) { |
73131fbb RZ |
701 | loff_t start_byte = |
702 | (loff_t)map->m_lblk << inode->i_blkbits; | |
703 | loff_t length = (loff_t)map->m_len << inode->i_blkbits; | |
704 | ||
ee0876bc | 705 | if (flags & EXT4_GET_BLOCKS_IO_SUBMIT) |
73131fbb RZ |
706 | ret = ext4_jbd2_inode_add_wait(handle, inode, |
707 | start_byte, length); | |
ee0876bc | 708 | else |
73131fbb RZ |
709 | ret = ext4_jbd2_inode_add_write(handle, inode, |
710 | start_byte, length); | |
06bd3c36 JK |
711 | if (ret) |
712 | return ret; | |
713 | } | |
6fd058f7 | 714 | } |
5e4d0eba XY |
715 | if (retval > 0 && (map->m_flags & EXT4_MAP_UNWRITTEN || |
716 | map->m_flags & EXT4_MAP_MAPPED)) | |
717 | ext4_fc_track_range(handle, inode, map->m_lblk, | |
718 | map->m_lblk + map->m_len - 1); | |
ec8c60be | 719 | if (retval < 0) |
70aa1554 | 720 | ext_debug(inode, "failed with err %d\n", retval); |
0e855ac8 AK |
721 | return retval; |
722 | } | |
723 | ||
ed8ad838 JK |
724 | /* |
725 | * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages | |
726 | * we have to be careful as someone else may be manipulating b_state as well. | |
727 | */ | |
728 | static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags) | |
729 | { | |
730 | unsigned long old_state; | |
731 | unsigned long new_state; | |
732 | ||
733 | flags &= EXT4_MAP_FLAGS; | |
734 | ||
735 | /* Dummy buffer_head? Set non-atomically. */ | |
736 | if (!bh->b_page) { | |
737 | bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags; | |
738 | return; | |
739 | } | |
740 | /* | |
741 | * Someone else may be modifying b_state. Be careful! This is ugly but | |
742 | * once we get rid of using bh as a container for mapping information | |
743 | * to pass to / from get_block functions, this can go away. | |
744 | */ | |
3ee2a3e7 | 745 | old_state = READ_ONCE(bh->b_state); |
ed8ad838 | 746 | do { |
ed8ad838 | 747 | new_state = (old_state & ~EXT4_MAP_FLAGS) | flags; |
3ee2a3e7 | 748 | } while (unlikely(!try_cmpxchg(&bh->b_state, &old_state, new_state))); |
ed8ad838 JK |
749 | } |
750 | ||
2ed88685 TT |
751 | static int _ext4_get_block(struct inode *inode, sector_t iblock, |
752 | struct buffer_head *bh, int flags) | |
ac27a0ec | 753 | { |
2ed88685 | 754 | struct ext4_map_blocks map; |
efe70c29 | 755 | int ret = 0; |
ac27a0ec | 756 | |
46c7f254 TM |
757 | if (ext4_has_inline_data(inode)) |
758 | return -ERANGE; | |
759 | ||
2ed88685 TT |
760 | map.m_lblk = iblock; |
761 | map.m_len = bh->b_size >> inode->i_blkbits; | |
762 | ||
efe70c29 JK |
763 | ret = ext4_map_blocks(ext4_journal_current_handle(), inode, &map, |
764 | flags); | |
7fb5409d | 765 | if (ret > 0) { |
2ed88685 | 766 | map_bh(bh, inode->i_sb, map.m_pblk); |
ed8ad838 | 767 | ext4_update_bh_state(bh, map.m_flags); |
2ed88685 | 768 | bh->b_size = inode->i_sb->s_blocksize * map.m_len; |
7fb5409d | 769 | ret = 0; |
547edce3 RZ |
770 | } else if (ret == 0) { |
771 | /* hole case, need to fill in bh->b_size */ | |
772 | bh->b_size = inode->i_sb->s_blocksize * map.m_len; | |
ac27a0ec DK |
773 | } |
774 | return ret; | |
775 | } | |
776 | ||
2ed88685 TT |
777 | int ext4_get_block(struct inode *inode, sector_t iblock, |
778 | struct buffer_head *bh, int create) | |
779 | { | |
780 | return _ext4_get_block(inode, iblock, bh, | |
781 | create ? EXT4_GET_BLOCKS_CREATE : 0); | |
782 | } | |
783 | ||
705965bd JK |
784 | /* |
785 | * Get block function used when preparing for buffered write if we require | |
786 | * creating an unwritten extent if blocks haven't been allocated. The extent | |
787 | * will be converted to written after the IO is complete. | |
788 | */ | |
789 | int ext4_get_block_unwritten(struct inode *inode, sector_t iblock, | |
790 | struct buffer_head *bh_result, int create) | |
791 | { | |
792 | ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n", | |
793 | inode->i_ino, create); | |
794 | return _ext4_get_block(inode, iblock, bh_result, | |
8d5459c1 | 795 | EXT4_GET_BLOCKS_CREATE_UNWRIT_EXT); |
705965bd JK |
796 | } |
797 | ||
efe70c29 JK |
798 | /* Maximum number of blocks we map for direct IO at once. */ |
799 | #define DIO_MAX_BLOCKS 4096 | |
800 | ||
ac27a0ec DK |
801 | /* |
802 | * `handle' can be NULL if create is zero | |
803 | */ | |
617ba13b | 804 | struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode, |
c5e298ae | 805 | ext4_lblk_t block, int map_flags) |
ac27a0ec | 806 | { |
2ed88685 TT |
807 | struct ext4_map_blocks map; |
808 | struct buffer_head *bh; | |
c5e298ae | 809 | int create = map_flags & EXT4_GET_BLOCKS_CREATE; |
9558cf14 | 810 | bool nowait = map_flags & EXT4_GET_BLOCKS_CACHED_NOWAIT; |
10560082 | 811 | int err; |
ac27a0ec | 812 | |
837c23fb CX |
813 | ASSERT((EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY) |
814 | || handle != NULL || create == 0); | |
9558cf14 | 815 | ASSERT(create == 0 || !nowait); |
ac27a0ec | 816 | |
2ed88685 TT |
817 | map.m_lblk = block; |
818 | map.m_len = 1; | |
c5e298ae | 819 | err = ext4_map_blocks(handle, inode, &map, map_flags); |
ac27a0ec | 820 | |
10560082 TT |
821 | if (err == 0) |
822 | return create ? ERR_PTR(-ENOSPC) : NULL; | |
2ed88685 | 823 | if (err < 0) |
10560082 | 824 | return ERR_PTR(err); |
2ed88685 | 825 | |
9558cf14 ZY |
826 | if (nowait) |
827 | return sb_find_get_block(inode->i_sb, map.m_pblk); | |
828 | ||
2ed88685 | 829 | bh = sb_getblk(inode->i_sb, map.m_pblk); |
10560082 TT |
830 | if (unlikely(!bh)) |
831 | return ERR_PTR(-ENOMEM); | |
2ed88685 | 832 | if (map.m_flags & EXT4_MAP_NEW) { |
837c23fb CX |
833 | ASSERT(create != 0); |
834 | ASSERT((EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY) | |
835 | || (handle != NULL)); | |
ac27a0ec | 836 | |
2ed88685 TT |
837 | /* |
838 | * Now that we do not always journal data, we should | |
839 | * keep in mind whether this should always journal the | |
840 | * new buffer as metadata. For now, regular file | |
841 | * writes use ext4_get_block instead, so it's not a | |
842 | * problem. | |
843 | */ | |
844 | lock_buffer(bh); | |
845 | BUFFER_TRACE(bh, "call get_create_access"); | |
188c299e JK |
846 | err = ext4_journal_get_create_access(handle, inode->i_sb, bh, |
847 | EXT4_JTR_NONE); | |
10560082 TT |
848 | if (unlikely(err)) { |
849 | unlock_buffer(bh); | |
850 | goto errout; | |
851 | } | |
852 | if (!buffer_uptodate(bh)) { | |
2ed88685 TT |
853 | memset(bh->b_data, 0, inode->i_sb->s_blocksize); |
854 | set_buffer_uptodate(bh); | |
ac27a0ec | 855 | } |
2ed88685 TT |
856 | unlock_buffer(bh); |
857 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); | |
858 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
10560082 TT |
859 | if (unlikely(err)) |
860 | goto errout; | |
861 | } else | |
2ed88685 | 862 | BUFFER_TRACE(bh, "not a new buffer"); |
2ed88685 | 863 | return bh; |
10560082 TT |
864 | errout: |
865 | brelse(bh); | |
866 | return ERR_PTR(err); | |
ac27a0ec DK |
867 | } |
868 | ||
617ba13b | 869 | struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode, |
c5e298ae | 870 | ext4_lblk_t block, int map_flags) |
ac27a0ec | 871 | { |
af5bc92d | 872 | struct buffer_head *bh; |
2d069c08 | 873 | int ret; |
ac27a0ec | 874 | |
c5e298ae | 875 | bh = ext4_getblk(handle, inode, block, map_flags); |
1c215028 | 876 | if (IS_ERR(bh)) |
ac27a0ec | 877 | return bh; |
7963e5ac | 878 | if (!bh || ext4_buffer_uptodate(bh)) |
ac27a0ec | 879 | return bh; |
2d069c08 | 880 | |
881 | ret = ext4_read_bh_lock(bh, REQ_META | REQ_PRIO, true); | |
882 | if (ret) { | |
883 | put_bh(bh); | |
884 | return ERR_PTR(ret); | |
885 | } | |
886 | return bh; | |
ac27a0ec DK |
887 | } |
888 | ||
9699d4f9 TE |
889 | /* Read a contiguous batch of blocks. */ |
890 | int ext4_bread_batch(struct inode *inode, ext4_lblk_t block, int bh_count, | |
891 | bool wait, struct buffer_head **bhs) | |
892 | { | |
893 | int i, err; | |
894 | ||
895 | for (i = 0; i < bh_count; i++) { | |
896 | bhs[i] = ext4_getblk(NULL, inode, block + i, 0 /* map_flags */); | |
897 | if (IS_ERR(bhs[i])) { | |
898 | err = PTR_ERR(bhs[i]); | |
899 | bh_count = i; | |
900 | goto out_brelse; | |
901 | } | |
902 | } | |
903 | ||
904 | for (i = 0; i < bh_count; i++) | |
905 | /* Note that NULL bhs[i] is valid because of holes. */ | |
2d069c08 | 906 | if (bhs[i] && !ext4_buffer_uptodate(bhs[i])) |
907 | ext4_read_bh_lock(bhs[i], REQ_META | REQ_PRIO, false); | |
9699d4f9 TE |
908 | |
909 | if (!wait) | |
910 | return 0; | |
911 | ||
912 | for (i = 0; i < bh_count; i++) | |
913 | if (bhs[i]) | |
914 | wait_on_buffer(bhs[i]); | |
915 | ||
916 | for (i = 0; i < bh_count; i++) { | |
917 | if (bhs[i] && !buffer_uptodate(bhs[i])) { | |
918 | err = -EIO; | |
919 | goto out_brelse; | |
920 | } | |
921 | } | |
922 | return 0; | |
923 | ||
924 | out_brelse: | |
925 | for (i = 0; i < bh_count; i++) { | |
926 | brelse(bhs[i]); | |
927 | bhs[i] = NULL; | |
928 | } | |
929 | return err; | |
930 | } | |
931 | ||
188c299e | 932 | int ext4_walk_page_buffers(handle_t *handle, struct inode *inode, |
f19d5870 TM |
933 | struct buffer_head *head, |
934 | unsigned from, | |
935 | unsigned to, | |
936 | int *partial, | |
188c299e | 937 | int (*fn)(handle_t *handle, struct inode *inode, |
f19d5870 | 938 | struct buffer_head *bh)) |
ac27a0ec DK |
939 | { |
940 | struct buffer_head *bh; | |
941 | unsigned block_start, block_end; | |
942 | unsigned blocksize = head->b_size; | |
943 | int err, ret = 0; | |
944 | struct buffer_head *next; | |
945 | ||
af5bc92d TT |
946 | for (bh = head, block_start = 0; |
947 | ret == 0 && (bh != head || !block_start); | |
de9a55b8 | 948 | block_start = block_end, bh = next) { |
ac27a0ec DK |
949 | next = bh->b_this_page; |
950 | block_end = block_start + blocksize; | |
951 | if (block_end <= from || block_start >= to) { | |
952 | if (partial && !buffer_uptodate(bh)) | |
953 | *partial = 1; | |
954 | continue; | |
955 | } | |
188c299e | 956 | err = (*fn)(handle, inode, bh); |
ac27a0ec DK |
957 | if (!ret) |
958 | ret = err; | |
959 | } | |
960 | return ret; | |
961 | } | |
962 | ||
d84c9ebd JK |
963 | /* |
964 | * Helper for handling dirtying of journalled data. We also mark the folio as | |
965 | * dirty so that writeback code knows about this page (and inode) contains | |
966 | * dirty data. ext4_writepages() then commits appropriate transaction to | |
967 | * make data stable. | |
968 | */ | |
969 | static int ext4_dirty_journalled_data(handle_t *handle, struct buffer_head *bh) | |
970 | { | |
971 | folio_mark_dirty(bh->b_folio); | |
972 | return ext4_handle_dirty_metadata(handle, NULL, bh); | |
973 | } | |
974 | ||
188c299e | 975 | int do_journal_get_write_access(handle_t *handle, struct inode *inode, |
f19d5870 | 976 | struct buffer_head *bh) |
ac27a0ec | 977 | { |
56d35a4c JK |
978 | int dirty = buffer_dirty(bh); |
979 | int ret; | |
980 | ||
ac27a0ec DK |
981 | if (!buffer_mapped(bh) || buffer_freed(bh)) |
982 | return 0; | |
56d35a4c | 983 | /* |
ebdec241 | 984 | * __block_write_begin() could have dirtied some buffers. Clean |
56d35a4c JK |
985 | * the dirty bit as jbd2_journal_get_write_access() could complain |
986 | * otherwise about fs integrity issues. Setting of the dirty bit | |
ebdec241 | 987 | * by __block_write_begin() isn't a real problem here as we clear |
56d35a4c JK |
988 | * the bit before releasing a page lock and thus writeback cannot |
989 | * ever write the buffer. | |
990 | */ | |
991 | if (dirty) | |
992 | clear_buffer_dirty(bh); | |
5d601255 | 993 | BUFFER_TRACE(bh, "get write access"); |
188c299e JK |
994 | ret = ext4_journal_get_write_access(handle, inode->i_sb, bh, |
995 | EXT4_JTR_NONE); | |
56d35a4c | 996 | if (!ret && dirty) |
d84c9ebd | 997 | ret = ext4_dirty_journalled_data(handle, bh); |
56d35a4c | 998 | return ret; |
ac27a0ec DK |
999 | } |
1000 | ||
643fa961 | 1001 | #ifdef CONFIG_FS_ENCRYPTION |
86b38c27 | 1002 | static int ext4_block_write_begin(struct folio *folio, loff_t pos, unsigned len, |
2058f83a MH |
1003 | get_block_t *get_block) |
1004 | { | |
09cbfeaf | 1005 | unsigned from = pos & (PAGE_SIZE - 1); |
2058f83a | 1006 | unsigned to = from + len; |
86b38c27 | 1007 | struct inode *inode = folio->mapping->host; |
2058f83a MH |
1008 | unsigned block_start, block_end; |
1009 | sector_t block; | |
1010 | int err = 0; | |
1011 | unsigned blocksize = inode->i_sb->s_blocksize; | |
1012 | unsigned bbits; | |
0b578f35 CR |
1013 | struct buffer_head *bh, *head, *wait[2]; |
1014 | int nr_wait = 0; | |
1015 | int i; | |
2058f83a | 1016 | |
86b38c27 | 1017 | BUG_ON(!folio_test_locked(folio)); |
09cbfeaf KS |
1018 | BUG_ON(from > PAGE_SIZE); |
1019 | BUG_ON(to > PAGE_SIZE); | |
2058f83a MH |
1020 | BUG_ON(from > to); |
1021 | ||
86b38c27 MW |
1022 | head = folio_buffers(folio); |
1023 | if (!head) { | |
1024 | create_empty_buffers(&folio->page, blocksize, 0); | |
1025 | head = folio_buffers(folio); | |
1026 | } | |
2058f83a | 1027 | bbits = ilog2(blocksize); |
86b38c27 | 1028 | block = (sector_t)folio->index << (PAGE_SHIFT - bbits); |
2058f83a MH |
1029 | |
1030 | for (bh = head, block_start = 0; bh != head || !block_start; | |
1031 | block++, block_start = block_end, bh = bh->b_this_page) { | |
1032 | block_end = block_start + blocksize; | |
1033 | if (block_end <= from || block_start >= to) { | |
86b38c27 | 1034 | if (folio_test_uptodate(folio)) { |
3cd46171 | 1035 | set_buffer_uptodate(bh); |
2058f83a MH |
1036 | } |
1037 | continue; | |
1038 | } | |
1039 | if (buffer_new(bh)) | |
1040 | clear_buffer_new(bh); | |
1041 | if (!buffer_mapped(bh)) { | |
1042 | WARN_ON(bh->b_size != blocksize); | |
1043 | err = get_block(inode, block, bh, 1); | |
1044 | if (err) | |
1045 | break; | |
1046 | if (buffer_new(bh)) { | |
86b38c27 | 1047 | if (folio_test_uptodate(folio)) { |
2058f83a MH |
1048 | clear_buffer_new(bh); |
1049 | set_buffer_uptodate(bh); | |
1050 | mark_buffer_dirty(bh); | |
1051 | continue; | |
1052 | } | |
1053 | if (block_end > to || block_start < from) | |
86b38c27 MW |
1054 | folio_zero_segments(folio, to, |
1055 | block_end, | |
1056 | block_start, from); | |
2058f83a MH |
1057 | continue; |
1058 | } | |
1059 | } | |
86b38c27 | 1060 | if (folio_test_uptodate(folio)) { |
3cd46171 | 1061 | set_buffer_uptodate(bh); |
2058f83a MH |
1062 | continue; |
1063 | } | |
1064 | if (!buffer_uptodate(bh) && !buffer_delay(bh) && | |
1065 | !buffer_unwritten(bh) && | |
1066 | (block_start < from || block_end > to)) { | |
2d069c08 | 1067 | ext4_read_bh_lock(bh, 0, false); |
0b578f35 | 1068 | wait[nr_wait++] = bh; |
2058f83a MH |
1069 | } |
1070 | } | |
1071 | /* | |
1072 | * If we issued read requests, let them complete. | |
1073 | */ | |
0b578f35 CR |
1074 | for (i = 0; i < nr_wait; i++) { |
1075 | wait_on_buffer(wait[i]); | |
1076 | if (!buffer_uptodate(wait[i])) | |
2058f83a MH |
1077 | err = -EIO; |
1078 | } | |
7e0785fc | 1079 | if (unlikely(err)) { |
4a9622f2 | 1080 | folio_zero_new_buffers(folio, from, to); |
4f74d15f | 1081 | } else if (fscrypt_inode_uses_fs_layer_crypto(inode)) { |
0b578f35 CR |
1082 | for (i = 0; i < nr_wait; i++) { |
1083 | int err2; | |
1084 | ||
86b38c27 MW |
1085 | err2 = fscrypt_decrypt_pagecache_blocks(folio, |
1086 | blocksize, bh_offset(wait[i])); | |
0b578f35 CR |
1087 | if (err2) { |
1088 | clear_buffer_uptodate(wait[i]); | |
1089 | err = err2; | |
1090 | } | |
1091 | } | |
7e0785fc CR |
1092 | } |
1093 | ||
2058f83a MH |
1094 | return err; |
1095 | } | |
1096 | #endif | |
1097 | ||
9462f770 JK |
1098 | /* |
1099 | * To preserve ordering, it is essential that the hole instantiation and | |
1100 | * the data write be encapsulated in a single transaction. We cannot | |
1101 | * close off a transaction and start a new one between the ext4_get_block() | |
1102 | * and the ext4_write_end(). So doing the jbd2_journal_start at the start of | |
1103 | * ext4_write_begin() is the right place. | |
1104 | */ | |
bfc1af65 | 1105 | static int ext4_write_begin(struct file *file, struct address_space *mapping, |
9d6b0cd7 | 1106 | loff_t pos, unsigned len, |
de9a55b8 | 1107 | struct page **pagep, void **fsdata) |
ac27a0ec | 1108 | { |
af5bc92d | 1109 | struct inode *inode = mapping->host; |
1938a150 | 1110 | int ret, needed_blocks; |
ac27a0ec DK |
1111 | handle_t *handle; |
1112 | int retries = 0; | |
4d934a5e | 1113 | struct folio *folio; |
de9a55b8 | 1114 | pgoff_t index; |
af5bc92d | 1115 | unsigned from, to; |
bfc1af65 | 1116 | |
0db1ff22 TT |
1117 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
1118 | return -EIO; | |
1119 | ||
9d6b0cd7 | 1120 | trace_ext4_write_begin(inode, pos, len); |
1938a150 AK |
1121 | /* |
1122 | * Reserve one block more for addition to orphan list in case | |
1123 | * we allocate blocks but write fails for some reason | |
1124 | */ | |
1125 | needed_blocks = ext4_writepage_trans_blocks(inode) + 1; | |
09cbfeaf KS |
1126 | index = pos >> PAGE_SHIFT; |
1127 | from = pos & (PAGE_SIZE - 1); | |
af5bc92d | 1128 | to = from + len; |
ac27a0ec | 1129 | |
f19d5870 TM |
1130 | if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { |
1131 | ret = ext4_try_to_write_inline_data(mapping, inode, pos, len, | |
832ee62d | 1132 | pagep); |
f19d5870 | 1133 | if (ret < 0) |
47564bfb TT |
1134 | return ret; |
1135 | if (ret == 1) | |
1136 | return 0; | |
f19d5870 TM |
1137 | } |
1138 | ||
47564bfb | 1139 | /* |
4d934a5e MW |
1140 | * __filemap_get_folio() can take a long time if the |
1141 | * system is thrashing due to memory pressure, or if the folio | |
47564bfb TT |
1142 | * is being written back. So grab it first before we start |
1143 | * the transaction handle. This also allows us to allocate | |
4d934a5e | 1144 | * the folio (if needed) without using GFP_NOFS. |
47564bfb TT |
1145 | */ |
1146 | retry_grab: | |
4d934a5e MW |
1147 | folio = __filemap_get_folio(mapping, index, FGP_WRITEBEGIN, |
1148 | mapping_gfp_mask(mapping)); | |
7fa8a8ee LT |
1149 | if (IS_ERR(folio)) |
1150 | return PTR_ERR(folio); | |
d1052d23 JH |
1151 | /* |
1152 | * The same as page allocation, we prealloc buffer heads before | |
1153 | * starting the handle. | |
1154 | */ | |
4d934a5e MW |
1155 | if (!folio_buffers(folio)) |
1156 | create_empty_buffers(&folio->page, inode->i_sb->s_blocksize, 0); | |
d1052d23 | 1157 | |
4d934a5e | 1158 | folio_unlock(folio); |
47564bfb TT |
1159 | |
1160 | retry_journal: | |
9924a92a | 1161 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks); |
af5bc92d | 1162 | if (IS_ERR(handle)) { |
4d934a5e | 1163 | folio_put(folio); |
47564bfb | 1164 | return PTR_ERR(handle); |
7479d2b9 | 1165 | } |
ac27a0ec | 1166 | |
4d934a5e MW |
1167 | folio_lock(folio); |
1168 | if (folio->mapping != mapping) { | |
1169 | /* The folio got truncated from under us */ | |
1170 | folio_unlock(folio); | |
1171 | folio_put(folio); | |
cf108bca | 1172 | ext4_journal_stop(handle); |
47564bfb | 1173 | goto retry_grab; |
cf108bca | 1174 | } |
4d934a5e MW |
1175 | /* In case writeback began while the folio was unlocked */ |
1176 | folio_wait_stable(folio); | |
cf108bca | 1177 | |
643fa961 | 1178 | #ifdef CONFIG_FS_ENCRYPTION |
2058f83a | 1179 | if (ext4_should_dioread_nolock(inode)) |
86b38c27 | 1180 | ret = ext4_block_write_begin(folio, pos, len, |
705965bd | 1181 | ext4_get_block_unwritten); |
2058f83a | 1182 | else |
86b38c27 | 1183 | ret = ext4_block_write_begin(folio, pos, len, ext4_get_block); |
2058f83a | 1184 | #else |
744692dc | 1185 | if (ext4_should_dioread_nolock(inode)) |
4d934a5e | 1186 | ret = __block_write_begin(&folio->page, pos, len, |
705965bd | 1187 | ext4_get_block_unwritten); |
744692dc | 1188 | else |
4d934a5e | 1189 | ret = __block_write_begin(&folio->page, pos, len, ext4_get_block); |
2058f83a | 1190 | #endif |
bfc1af65 | 1191 | if (!ret && ext4_should_journal_data(inode)) { |
188c299e | 1192 | ret = ext4_walk_page_buffers(handle, inode, |
4d934a5e MW |
1193 | folio_buffers(folio), from, to, |
1194 | NULL, do_journal_get_write_access); | |
ac27a0ec | 1195 | } |
bfc1af65 NP |
1196 | |
1197 | if (ret) { | |
c93d8f88 EB |
1198 | bool extended = (pos + len > inode->i_size) && |
1199 | !ext4_verity_in_progress(inode); | |
1200 | ||
4d934a5e | 1201 | folio_unlock(folio); |
ae4d5372 | 1202 | /* |
6e1db88d | 1203 | * __block_write_begin may have instantiated a few blocks |
ae4d5372 | 1204 | * outside i_size. Trim these off again. Don't need |
f340b3d9 | 1205 | * i_size_read because we hold i_rwsem. |
1938a150 AK |
1206 | * |
1207 | * Add inode to orphan list in case we crash before | |
1208 | * truncate finishes | |
ae4d5372 | 1209 | */ |
c93d8f88 | 1210 | if (extended && ext4_can_truncate(inode)) |
1938a150 AK |
1211 | ext4_orphan_add(handle, inode); |
1212 | ||
1213 | ext4_journal_stop(handle); | |
c93d8f88 | 1214 | if (extended) { |
b9a4207d | 1215 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1216 | /* |
ffacfa7a | 1217 | * If truncate failed early the inode might |
1938a150 AK |
1218 | * still be on the orphan list; we need to |
1219 | * make sure the inode is removed from the | |
1220 | * orphan list in that case. | |
1221 | */ | |
1222 | if (inode->i_nlink) | |
1223 | ext4_orphan_del(NULL, inode); | |
1224 | } | |
bfc1af65 | 1225 | |
47564bfb TT |
1226 | if (ret == -ENOSPC && |
1227 | ext4_should_retry_alloc(inode->i_sb, &retries)) | |
1228 | goto retry_journal; | |
4d934a5e | 1229 | folio_put(folio); |
47564bfb TT |
1230 | return ret; |
1231 | } | |
4d934a5e | 1232 | *pagep = &folio->page; |
ac27a0ec DK |
1233 | return ret; |
1234 | } | |
1235 | ||
bfc1af65 | 1236 | /* For write_end() in data=journal mode */ |
188c299e JK |
1237 | static int write_end_fn(handle_t *handle, struct inode *inode, |
1238 | struct buffer_head *bh) | |
ac27a0ec | 1239 | { |
13fca323 | 1240 | int ret; |
ac27a0ec DK |
1241 | if (!buffer_mapped(bh) || buffer_freed(bh)) |
1242 | return 0; | |
1243 | set_buffer_uptodate(bh); | |
d84c9ebd | 1244 | ret = ext4_dirty_journalled_data(handle, bh); |
13fca323 TT |
1245 | clear_buffer_meta(bh); |
1246 | clear_buffer_prio(bh); | |
1247 | return ret; | |
ac27a0ec DK |
1248 | } |
1249 | ||
eed4333f ZL |
1250 | /* |
1251 | * We need to pick up the new inode size which generic_commit_write gave us | |
1252 | * `file' can be NULL - eg, when called from page_symlink(). | |
1253 | * | |
1254 | * ext4 never places buffers on inode->i_mapping->private_list. metadata | |
1255 | * buffers are managed internally. | |
1256 | */ | |
1257 | static int ext4_write_end(struct file *file, | |
1258 | struct address_space *mapping, | |
1259 | loff_t pos, unsigned len, unsigned copied, | |
1260 | struct page *page, void *fsdata) | |
f8514083 | 1261 | { |
64fb3136 | 1262 | struct folio *folio = page_folio(page); |
f8514083 | 1263 | handle_t *handle = ext4_journal_current_handle(); |
eed4333f | 1264 | struct inode *inode = mapping->host; |
0572639f | 1265 | loff_t old_size = inode->i_size; |
eed4333f ZL |
1266 | int ret = 0, ret2; |
1267 | int i_size_changed = 0; | |
c93d8f88 | 1268 | bool verity = ext4_verity_in_progress(inode); |
eed4333f ZL |
1269 | |
1270 | trace_ext4_write_end(inode, pos, len, copied); | |
6984aef5 | 1271 | |
5c099c4f YB |
1272 | if (ext4_has_inline_data(inode) && |
1273 | ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) | |
d19500da RH |
1274 | return ext4_write_inline_data_end(inode, pos, len, copied, |
1275 | folio); | |
6984aef5 ZY |
1276 | |
1277 | copied = block_write_end(file, mapping, pos, len, copied, page, fsdata); | |
f8514083 | 1278 | /* |
64fb3136 | 1279 | * it's important to update i_size while still holding folio lock: |
f8514083 | 1280 | * page writeout could otherwise come in and zero beyond i_size. |
c93d8f88 EB |
1281 | * |
1282 | * If FS_IOC_ENABLE_VERITY is running on this inode, then Merkle tree | |
1283 | * blocks are being written past EOF, so skip the i_size update. | |
f8514083 | 1284 | */ |
c93d8f88 EB |
1285 | if (!verity) |
1286 | i_size_changed = ext4_update_inode_size(inode, pos + copied); | |
64fb3136 MW |
1287 | folio_unlock(folio); |
1288 | folio_put(folio); | |
f8514083 | 1289 | |
c93d8f88 | 1290 | if (old_size < pos && !verity) |
0572639f | 1291 | pagecache_isize_extended(inode, old_size, pos); |
f8514083 | 1292 | /* |
64fb3136 MW |
1293 | * Don't mark the inode dirty under folio lock. First, it unnecessarily |
1294 | * makes the holding time of folio lock longer. Second, it forces lock | |
1295 | * ordering of folio lock and transaction start for journaling | |
f8514083 AK |
1296 | * filesystems. |
1297 | */ | |
6984aef5 | 1298 | if (i_size_changed) |
4209ae12 | 1299 | ret = ext4_mark_inode_dirty(handle, inode); |
f8514083 | 1300 | |
c93d8f88 | 1301 | if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode)) |
f8514083 AK |
1302 | /* if we have allocated more blocks and copied |
1303 | * less. We will have blocks allocated outside | |
1304 | * inode->i_size. So truncate them | |
1305 | */ | |
1306 | ext4_orphan_add(handle, inode); | |
55ce2f64 | 1307 | |
617ba13b | 1308 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1309 | if (!ret) |
1310 | ret = ret2; | |
bfc1af65 | 1311 | |
c93d8f88 | 1312 | if (pos + len > inode->i_size && !verity) { |
b9a4207d | 1313 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1314 | /* |
ffacfa7a | 1315 | * If truncate failed early the inode might still be |
f8514083 AK |
1316 | * on the orphan list; we need to make sure the inode |
1317 | * is removed from the orphan list in that case. | |
1318 | */ | |
1319 | if (inode->i_nlink) | |
1320 | ext4_orphan_del(NULL, inode); | |
1321 | } | |
1322 | ||
bfc1af65 | 1323 | return ret ? ret : copied; |
ac27a0ec DK |
1324 | } |
1325 | ||
b90197b6 | 1326 | /* |
4a9622f2 | 1327 | * This is a private version of folio_zero_new_buffers() which doesn't |
b90197b6 | 1328 | * set the buffer to be dirty, since in data=journalled mode we need |
d84c9ebd | 1329 | * to call ext4_dirty_journalled_data() instead. |
b90197b6 | 1330 | */ |
3b136499 | 1331 | static void ext4_journalled_zero_new_buffers(handle_t *handle, |
188c299e | 1332 | struct inode *inode, |
86324a21 | 1333 | struct folio *folio, |
3b136499 | 1334 | unsigned from, unsigned to) |
b90197b6 TT |
1335 | { |
1336 | unsigned int block_start = 0, block_end; | |
1337 | struct buffer_head *head, *bh; | |
1338 | ||
86324a21 | 1339 | bh = head = folio_buffers(folio); |
b90197b6 TT |
1340 | do { |
1341 | block_end = block_start + bh->b_size; | |
1342 | if (buffer_new(bh)) { | |
1343 | if (block_end > from && block_start < to) { | |
86324a21 | 1344 | if (!folio_test_uptodate(folio)) { |
b90197b6 TT |
1345 | unsigned start, size; |
1346 | ||
1347 | start = max(from, block_start); | |
1348 | size = min(to, block_end) - start; | |
1349 | ||
86324a21 | 1350 | folio_zero_range(folio, start, size); |
188c299e | 1351 | write_end_fn(handle, inode, bh); |
b90197b6 TT |
1352 | } |
1353 | clear_buffer_new(bh); | |
1354 | } | |
1355 | } | |
1356 | block_start = block_end; | |
1357 | bh = bh->b_this_page; | |
1358 | } while (bh != head); | |
1359 | } | |
1360 | ||
bfc1af65 | 1361 | static int ext4_journalled_write_end(struct file *file, |
de9a55b8 TT |
1362 | struct address_space *mapping, |
1363 | loff_t pos, unsigned len, unsigned copied, | |
1364 | struct page *page, void *fsdata) | |
ac27a0ec | 1365 | { |
feb22b77 | 1366 | struct folio *folio = page_folio(page); |
617ba13b | 1367 | handle_t *handle = ext4_journal_current_handle(); |
bfc1af65 | 1368 | struct inode *inode = mapping->host; |
0572639f | 1369 | loff_t old_size = inode->i_size; |
ac27a0ec DK |
1370 | int ret = 0, ret2; |
1371 | int partial = 0; | |
bfc1af65 | 1372 | unsigned from, to; |
4631dbf6 | 1373 | int size_changed = 0; |
c93d8f88 | 1374 | bool verity = ext4_verity_in_progress(inode); |
ac27a0ec | 1375 | |
9bffad1e | 1376 | trace_ext4_journalled_write_end(inode, pos, len, copied); |
09cbfeaf | 1377 | from = pos & (PAGE_SIZE - 1); |
bfc1af65 NP |
1378 | to = from + len; |
1379 | ||
441c8508 CW |
1380 | BUG_ON(!ext4_handle_valid(handle)); |
1381 | ||
6984aef5 | 1382 | if (ext4_has_inline_data(inode)) |
d19500da RH |
1383 | return ext4_write_inline_data_end(inode, pos, len, copied, |
1384 | folio); | |
6984aef5 | 1385 | |
feb22b77 | 1386 | if (unlikely(copied < len) && !folio_test_uptodate(folio)) { |
3b136499 | 1387 | copied = 0; |
86324a21 MW |
1388 | ext4_journalled_zero_new_buffers(handle, inode, folio, |
1389 | from, to); | |
3b136499 JK |
1390 | } else { |
1391 | if (unlikely(copied < len)) | |
86324a21 | 1392 | ext4_journalled_zero_new_buffers(handle, inode, folio, |
3b136499 | 1393 | from + copied, to); |
feb22b77 MW |
1394 | ret = ext4_walk_page_buffers(handle, inode, |
1395 | folio_buffers(folio), | |
188c299e | 1396 | from, from + copied, &partial, |
3b136499 | 1397 | write_end_fn); |
3fdcfb66 | 1398 | if (!partial) |
feb22b77 | 1399 | folio_mark_uptodate(folio); |
3fdcfb66 | 1400 | } |
c93d8f88 EB |
1401 | if (!verity) |
1402 | size_changed = ext4_update_inode_size(inode, pos + copied); | |
2d859db3 | 1403 | EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; |
feb22b77 MW |
1404 | folio_unlock(folio); |
1405 | folio_put(folio); | |
4631dbf6 | 1406 | |
c93d8f88 | 1407 | if (old_size < pos && !verity) |
0572639f XW |
1408 | pagecache_isize_extended(inode, old_size, pos); |
1409 | ||
6984aef5 | 1410 | if (size_changed) { |
617ba13b | 1411 | ret2 = ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
1412 | if (!ret) |
1413 | ret = ret2; | |
1414 | } | |
bfc1af65 | 1415 | |
c93d8f88 | 1416 | if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode)) |
f8514083 AK |
1417 | /* if we have allocated more blocks and copied |
1418 | * less. We will have blocks allocated outside | |
1419 | * inode->i_size. So truncate them | |
1420 | */ | |
1421 | ext4_orphan_add(handle, inode); | |
1422 | ||
617ba13b | 1423 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1424 | if (!ret) |
1425 | ret = ret2; | |
c93d8f88 | 1426 | if (pos + len > inode->i_size && !verity) { |
b9a4207d | 1427 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1428 | /* |
ffacfa7a | 1429 | * If truncate failed early the inode might still be |
f8514083 AK |
1430 | * on the orphan list; we need to make sure the inode |
1431 | * is removed from the orphan list in that case. | |
1432 | */ | |
1433 | if (inode->i_nlink) | |
1434 | ext4_orphan_del(NULL, inode); | |
1435 | } | |
bfc1af65 NP |
1436 | |
1437 | return ret ? ret : copied; | |
ac27a0ec | 1438 | } |
d2a17637 | 1439 | |
9d0be502 | 1440 | /* |
c27e43a1 | 1441 | * Reserve space for a single cluster |
9d0be502 | 1442 | */ |
c27e43a1 | 1443 | static int ext4_da_reserve_space(struct inode *inode) |
d2a17637 | 1444 | { |
60e58e0f | 1445 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
0637c6f4 | 1446 | struct ext4_inode_info *ei = EXT4_I(inode); |
5dd4056d | 1447 | int ret; |
03179fe9 TT |
1448 | |
1449 | /* | |
1450 | * We will charge metadata quota at writeout time; this saves | |
1451 | * us from metadata over-estimation, though we may go over by | |
1452 | * a small amount in the end. Here we just reserve for data. | |
1453 | */ | |
1454 | ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1)); | |
1455 | if (ret) | |
1456 | return ret; | |
d2a17637 | 1457 | |
0637c6f4 | 1458 | spin_lock(&ei->i_block_reservation_lock); |
71d4f7d0 | 1459 | if (ext4_claim_free_clusters(sbi, 1, 0)) { |
03179fe9 | 1460 | spin_unlock(&ei->i_block_reservation_lock); |
03179fe9 | 1461 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1)); |
d2a17637 MC |
1462 | return -ENOSPC; |
1463 | } | |
9d0be502 | 1464 | ei->i_reserved_data_blocks++; |
c27e43a1 | 1465 | trace_ext4_da_reserve_space(inode); |
0637c6f4 | 1466 | spin_unlock(&ei->i_block_reservation_lock); |
39bc680a | 1467 | |
d2a17637 MC |
1468 | return 0; /* success */ |
1469 | } | |
1470 | ||
f456767d | 1471 | void ext4_da_release_space(struct inode *inode, int to_free) |
d2a17637 MC |
1472 | { |
1473 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0637c6f4 | 1474 | struct ext4_inode_info *ei = EXT4_I(inode); |
d2a17637 | 1475 | |
cd213226 MC |
1476 | if (!to_free) |
1477 | return; /* Nothing to release, exit */ | |
1478 | ||
d2a17637 | 1479 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); |
cd213226 | 1480 | |
5a58ec87 | 1481 | trace_ext4_da_release_space(inode, to_free); |
0637c6f4 | 1482 | if (unlikely(to_free > ei->i_reserved_data_blocks)) { |
cd213226 | 1483 | /* |
0637c6f4 TT |
1484 | * if there aren't enough reserved blocks, then the |
1485 | * counter is messed up somewhere. Since this | |
1486 | * function is called from invalidate page, it's | |
1487 | * harmless to return without any action. | |
cd213226 | 1488 | */ |
8de5c325 | 1489 | ext4_warning(inode->i_sb, "ext4_da_release_space: " |
0637c6f4 | 1490 | "ino %lu, to_free %d with only %d reserved " |
1084f252 | 1491 | "data blocks", inode->i_ino, to_free, |
0637c6f4 TT |
1492 | ei->i_reserved_data_blocks); |
1493 | WARN_ON(1); | |
1494 | to_free = ei->i_reserved_data_blocks; | |
cd213226 | 1495 | } |
0637c6f4 | 1496 | ei->i_reserved_data_blocks -= to_free; |
cd213226 | 1497 | |
72b8ab9d | 1498 | /* update fs dirty data blocks counter */ |
57042651 | 1499 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free); |
d2a17637 | 1500 | |
d2a17637 | 1501 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
60e58e0f | 1502 | |
7b415bf6 | 1503 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free)); |
d2a17637 MC |
1504 | } |
1505 | ||
64769240 AT |
1506 | /* |
1507 | * Delayed allocation stuff | |
1508 | */ | |
1509 | ||
4e7ea81d | 1510 | struct mpage_da_data { |
15648d59 | 1511 | /* These are input fields for ext4_do_writepages() */ |
4e7ea81d JK |
1512 | struct inode *inode; |
1513 | struct writeback_control *wbc; | |
15648d59 | 1514 | unsigned int can_map:1; /* Can writepages call map blocks? */ |
6b523df4 | 1515 | |
15648d59 | 1516 | /* These are internal state of ext4_do_writepages() */ |
4e7ea81d JK |
1517 | pgoff_t first_page; /* The first page to write */ |
1518 | pgoff_t next_page; /* Current page to examine */ | |
1519 | pgoff_t last_page; /* Last page to examine */ | |
791b7f08 | 1520 | /* |
4e7ea81d JK |
1521 | * Extent to map - this can be after first_page because that can be |
1522 | * fully mapped. We somewhat abuse m_flags to store whether the extent | |
1523 | * is delalloc or unwritten. | |
791b7f08 | 1524 | */ |
4e7ea81d JK |
1525 | struct ext4_map_blocks map; |
1526 | struct ext4_io_submit io_submit; /* IO submission data */ | |
dddbd6ac | 1527 | unsigned int do_map:1; |
6b8ed620 | 1528 | unsigned int scanned_until_end:1; |
1f1a55f0 | 1529 | unsigned int journalled_more_data:1; |
4e7ea81d | 1530 | }; |
64769240 | 1531 | |
4e7ea81d JK |
1532 | static void mpage_release_unused_pages(struct mpage_da_data *mpd, |
1533 | bool invalidate) | |
c4a0c46e | 1534 | { |
fb5a5be0 | 1535 | unsigned nr, i; |
c4a0c46e | 1536 | pgoff_t index, end; |
fb5a5be0 | 1537 | struct folio_batch fbatch; |
c4a0c46e AK |
1538 | struct inode *inode = mpd->inode; |
1539 | struct address_space *mapping = inode->i_mapping; | |
4e7ea81d JK |
1540 | |
1541 | /* This is necessary when next_page == 0. */ | |
1542 | if (mpd->first_page >= mpd->next_page) | |
1543 | return; | |
c4a0c46e | 1544 | |
6b8ed620 | 1545 | mpd->scanned_until_end = 0; |
c7f5938a CW |
1546 | index = mpd->first_page; |
1547 | end = mpd->next_page - 1; | |
4e7ea81d JK |
1548 | if (invalidate) { |
1549 | ext4_lblk_t start, last; | |
09cbfeaf KS |
1550 | start = index << (PAGE_SHIFT - inode->i_blkbits); |
1551 | last = end << (PAGE_SHIFT - inode->i_blkbits); | |
7f0d8e1d EW |
1552 | |
1553 | /* | |
1554 | * avoid racing with extent status tree scans made by | |
1555 | * ext4_insert_delayed_block() | |
1556 | */ | |
1557 | down_write(&EXT4_I(inode)->i_data_sem); | |
4e7ea81d | 1558 | ext4_es_remove_extent(inode, start, last - start + 1); |
7f0d8e1d | 1559 | up_write(&EXT4_I(inode)->i_data_sem); |
4e7ea81d | 1560 | } |
51865fda | 1561 | |
fb5a5be0 | 1562 | folio_batch_init(&fbatch); |
c4a0c46e | 1563 | while (index <= end) { |
fb5a5be0 MWO |
1564 | nr = filemap_get_folios(mapping, &index, end, &fbatch); |
1565 | if (nr == 0) | |
c4a0c46e | 1566 | break; |
fb5a5be0 MWO |
1567 | for (i = 0; i < nr; i++) { |
1568 | struct folio *folio = fbatch.folios[i]; | |
2b85a617 | 1569 | |
fb5a5be0 MWO |
1570 | if (folio->index < mpd->first_page) |
1571 | continue; | |
1572 | if (folio->index + folio_nr_pages(folio) - 1 > end) | |
1573 | continue; | |
7ba13abb MWO |
1574 | BUG_ON(!folio_test_locked(folio)); |
1575 | BUG_ON(folio_test_writeback(folio)); | |
4e7ea81d | 1576 | if (invalidate) { |
7ba13abb MWO |
1577 | if (folio_mapped(folio)) |
1578 | folio_clear_dirty_for_io(folio); | |
1579 | block_invalidate_folio(folio, 0, | |
1580 | folio_size(folio)); | |
1581 | folio_clear_uptodate(folio); | |
4e7ea81d | 1582 | } |
7ba13abb | 1583 | folio_unlock(folio); |
c4a0c46e | 1584 | } |
fb5a5be0 | 1585 | folio_batch_release(&fbatch); |
c4a0c46e | 1586 | } |
c4a0c46e AK |
1587 | } |
1588 | ||
df22291f AK |
1589 | static void ext4_print_free_blocks(struct inode *inode) |
1590 | { | |
1591 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
92b97816 | 1592 | struct super_block *sb = inode->i_sb; |
f78ee70d | 1593 | struct ext4_inode_info *ei = EXT4_I(inode); |
92b97816 TT |
1594 | |
1595 | ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld", | |
5dee5437 | 1596 | EXT4_C2B(EXT4_SB(inode->i_sb), |
f78ee70d | 1597 | ext4_count_free_clusters(sb))); |
92b97816 TT |
1598 | ext4_msg(sb, KERN_CRIT, "Free/Dirty block details"); |
1599 | ext4_msg(sb, KERN_CRIT, "free_blocks=%lld", | |
f78ee70d | 1600 | (long long) EXT4_C2B(EXT4_SB(sb), |
57042651 | 1601 | percpu_counter_sum(&sbi->s_freeclusters_counter))); |
92b97816 | 1602 | ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld", |
f78ee70d | 1603 | (long long) EXT4_C2B(EXT4_SB(sb), |
7b415bf6 | 1604 | percpu_counter_sum(&sbi->s_dirtyclusters_counter))); |
92b97816 TT |
1605 | ext4_msg(sb, KERN_CRIT, "Block reservation details"); |
1606 | ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u", | |
f78ee70d | 1607 | ei->i_reserved_data_blocks); |
df22291f AK |
1608 | return; |
1609 | } | |
1610 | ||
0b02f4c0 EW |
1611 | /* |
1612 | * ext4_insert_delayed_block - adds a delayed block to the extents status | |
1613 | * tree, incrementing the reserved cluster/block | |
1614 | * count or making a pending reservation | |
1615 | * where needed | |
1616 | * | |
1617 | * @inode - file containing the newly added block | |
1618 | * @lblk - logical block to be added | |
1619 | * | |
1620 | * Returns 0 on success, negative error code on failure. | |
1621 | */ | |
1622 | static int ext4_insert_delayed_block(struct inode *inode, ext4_lblk_t lblk) | |
1623 | { | |
1624 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
1625 | int ret; | |
1626 | bool allocated = false; | |
1627 | ||
1628 | /* | |
1629 | * If the cluster containing lblk is shared with a delayed, | |
1630 | * written, or unwritten extent in a bigalloc file system, it's | |
1631 | * already been accounted for and does not need to be reserved. | |
1632 | * A pending reservation must be made for the cluster if it's | |
1633 | * shared with a written or unwritten extent and doesn't already | |
1634 | * have one. Written and unwritten extents can be purged from the | |
1635 | * extents status tree if the system is under memory pressure, so | |
1636 | * it's necessary to examine the extent tree if a search of the | |
1637 | * extents status tree doesn't get a match. | |
1638 | */ | |
1639 | if (sbi->s_cluster_ratio == 1) { | |
1640 | ret = ext4_da_reserve_space(inode); | |
1641 | if (ret != 0) /* ENOSPC */ | |
8782b020 | 1642 | return ret; |
0b02f4c0 EW |
1643 | } else { /* bigalloc */ |
1644 | if (!ext4_es_scan_clu(inode, &ext4_es_is_delonly, lblk)) { | |
1645 | if (!ext4_es_scan_clu(inode, | |
1646 | &ext4_es_is_mapped, lblk)) { | |
1647 | ret = ext4_clu_mapped(inode, | |
1648 | EXT4_B2C(sbi, lblk)); | |
1649 | if (ret < 0) | |
8782b020 | 1650 | return ret; |
0b02f4c0 EW |
1651 | if (ret == 0) { |
1652 | ret = ext4_da_reserve_space(inode); | |
1653 | if (ret != 0) /* ENOSPC */ | |
8782b020 | 1654 | return ret; |
0b02f4c0 EW |
1655 | } else { |
1656 | allocated = true; | |
1657 | } | |
1658 | } else { | |
1659 | allocated = true; | |
1660 | } | |
1661 | } | |
1662 | } | |
1663 | ||
8782b020 BL |
1664 | ext4_es_insert_delayed_block(inode, lblk, allocated); |
1665 | return 0; | |
0b02f4c0 EW |
1666 | } |
1667 | ||
5356f261 AK |
1668 | /* |
1669 | * This function is grabs code from the very beginning of | |
1670 | * ext4_map_blocks, but assumes that the caller is from delayed write | |
1671 | * time. This function looks up the requested blocks and sets the | |
1672 | * buffer delay bit under the protection of i_data_sem. | |
1673 | */ | |
1674 | static int ext4_da_map_blocks(struct inode *inode, sector_t iblock, | |
1675 | struct ext4_map_blocks *map, | |
1676 | struct buffer_head *bh) | |
1677 | { | |
d100eef2 | 1678 | struct extent_status es; |
5356f261 AK |
1679 | int retval; |
1680 | sector_t invalid_block = ~((sector_t) 0xffff); | |
921f266b DM |
1681 | #ifdef ES_AGGRESSIVE_TEST |
1682 | struct ext4_map_blocks orig_map; | |
1683 | ||
1684 | memcpy(&orig_map, map, sizeof(*map)); | |
1685 | #endif | |
5356f261 AK |
1686 | |
1687 | if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es)) | |
1688 | invalid_block = ~0; | |
1689 | ||
1690 | map->m_flags = 0; | |
70aa1554 | 1691 | ext_debug(inode, "max_blocks %u, logical block %lu\n", map->m_len, |
5356f261 | 1692 | (unsigned long) map->m_lblk); |
d100eef2 ZL |
1693 | |
1694 | /* Lookup extent status tree firstly */ | |
bb5835ed | 1695 | if (ext4_es_lookup_extent(inode, iblock, NULL, &es)) { |
d100eef2 ZL |
1696 | if (ext4_es_is_hole(&es)) { |
1697 | retval = 0; | |
c8b459f4 | 1698 | down_read(&EXT4_I(inode)->i_data_sem); |
d100eef2 ZL |
1699 | goto add_delayed; |
1700 | } | |
1701 | ||
1702 | /* | |
3eda41df EW |
1703 | * Delayed extent could be allocated by fallocate. |
1704 | * So we need to check it. | |
d100eef2 | 1705 | */ |
3eda41df EW |
1706 | if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) { |
1707 | map_bh(bh, inode->i_sb, invalid_block); | |
1708 | set_buffer_new(bh); | |
1709 | set_buffer_delay(bh); | |
d100eef2 ZL |
1710 | return 0; |
1711 | } | |
1712 | ||
1713 | map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk; | |
1714 | retval = es.es_len - (iblock - es.es_lblk); | |
1715 | if (retval > map->m_len) | |
1716 | retval = map->m_len; | |
1717 | map->m_len = retval; | |
1718 | if (ext4_es_is_written(&es)) | |
1719 | map->m_flags |= EXT4_MAP_MAPPED; | |
1720 | else if (ext4_es_is_unwritten(&es)) | |
1721 | map->m_flags |= EXT4_MAP_UNWRITTEN; | |
1722 | else | |
1e83bc81 | 1723 | BUG(); |
d100eef2 | 1724 | |
921f266b DM |
1725 | #ifdef ES_AGGRESSIVE_TEST |
1726 | ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0); | |
1727 | #endif | |
d100eef2 ZL |
1728 | return retval; |
1729 | } | |
1730 | ||
5356f261 AK |
1731 | /* |
1732 | * Try to see if we can get the block without requesting a new | |
1733 | * file system block. | |
1734 | */ | |
c8b459f4 | 1735 | down_read(&EXT4_I(inode)->i_data_sem); |
cbd7584e | 1736 | if (ext4_has_inline_data(inode)) |
9c3569b5 | 1737 | retval = 0; |
cbd7584e | 1738 | else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
2f8e0a7c | 1739 | retval = ext4_ext_map_blocks(NULL, inode, map, 0); |
5356f261 | 1740 | else |
2f8e0a7c | 1741 | retval = ext4_ind_map_blocks(NULL, inode, map, 0); |
5356f261 | 1742 | |
d100eef2 | 1743 | add_delayed: |
5356f261 | 1744 | if (retval == 0) { |
f7fec032 | 1745 | int ret; |
ad431025 | 1746 | |
5356f261 AK |
1747 | /* |
1748 | * XXX: __block_prepare_write() unmaps passed block, | |
1749 | * is it OK? | |
1750 | */ | |
5356f261 | 1751 | |
0b02f4c0 EW |
1752 | ret = ext4_insert_delayed_block(inode, map->m_lblk); |
1753 | if (ret != 0) { | |
f7fec032 | 1754 | retval = ret; |
51865fda | 1755 | goto out_unlock; |
f7fec032 | 1756 | } |
51865fda | 1757 | |
5356f261 AK |
1758 | map_bh(bh, inode->i_sb, invalid_block); |
1759 | set_buffer_new(bh); | |
1760 | set_buffer_delay(bh); | |
f7fec032 | 1761 | } else if (retval > 0) { |
3be78c73 | 1762 | unsigned int status; |
f7fec032 | 1763 | |
44fb851d ZL |
1764 | if (unlikely(retval != map->m_len)) { |
1765 | ext4_warning(inode->i_sb, | |
1766 | "ES len assertion failed for inode " | |
1767 | "%lu: retval %d != map->m_len %d", | |
1768 | inode->i_ino, retval, map->m_len); | |
1769 | WARN_ON(1); | |
921f266b | 1770 | } |
921f266b | 1771 | |
f7fec032 ZL |
1772 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
1773 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; | |
6c120399 BL |
1774 | ext4_es_insert_extent(inode, map->m_lblk, map->m_len, |
1775 | map->m_pblk, status); | |
5356f261 AK |
1776 | } |
1777 | ||
1778 | out_unlock: | |
1779 | up_read((&EXT4_I(inode)->i_data_sem)); | |
1780 | ||
1781 | return retval; | |
1782 | } | |
1783 | ||
64769240 | 1784 | /* |
d91bd2c1 | 1785 | * This is a special get_block_t callback which is used by |
b920c755 TT |
1786 | * ext4_da_write_begin(). It will either return mapped block or |
1787 | * reserve space for a single block. | |
29fa89d0 AK |
1788 | * |
1789 | * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set. | |
1790 | * We also have b_blocknr = -1 and b_bdev initialized properly | |
1791 | * | |
1792 | * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set. | |
1793 | * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev | |
1794 | * initialized properly. | |
64769240 | 1795 | */ |
9c3569b5 TM |
1796 | int ext4_da_get_block_prep(struct inode *inode, sector_t iblock, |
1797 | struct buffer_head *bh, int create) | |
64769240 | 1798 | { |
2ed88685 | 1799 | struct ext4_map_blocks map; |
64769240 AT |
1800 | int ret = 0; |
1801 | ||
1802 | BUG_ON(create == 0); | |
2ed88685 TT |
1803 | BUG_ON(bh->b_size != inode->i_sb->s_blocksize); |
1804 | ||
1805 | map.m_lblk = iblock; | |
1806 | map.m_len = 1; | |
64769240 AT |
1807 | |
1808 | /* | |
1809 | * first, we need to know whether the block is allocated already | |
1810 | * preallocated blocks are unmapped but should treated | |
1811 | * the same as allocated blocks. | |
1812 | */ | |
5356f261 AK |
1813 | ret = ext4_da_map_blocks(inode, iblock, &map, bh); |
1814 | if (ret <= 0) | |
2ed88685 | 1815 | return ret; |
64769240 | 1816 | |
2ed88685 | 1817 | map_bh(bh, inode->i_sb, map.m_pblk); |
ed8ad838 | 1818 | ext4_update_bh_state(bh, map.m_flags); |
2ed88685 TT |
1819 | |
1820 | if (buffer_unwritten(bh)) { | |
1821 | /* A delayed write to unwritten bh should be marked | |
1822 | * new and mapped. Mapped ensures that we don't do | |
1823 | * get_block multiple times when we write to the same | |
1824 | * offset and new ensures that we do proper zero out | |
1825 | * for partial write. | |
1826 | */ | |
1827 | set_buffer_new(bh); | |
c8205636 | 1828 | set_buffer_mapped(bh); |
2ed88685 TT |
1829 | } |
1830 | return 0; | |
64769240 | 1831 | } |
61628a3f | 1832 | |
33483b3b | 1833 | static void mpage_folio_done(struct mpage_da_data *mpd, struct folio *folio) |
eaf2ca10 | 1834 | { |
33483b3b MW |
1835 | mpd->first_page += folio_nr_pages(folio); |
1836 | folio_unlock(folio); | |
eaf2ca10 JK |
1837 | } |
1838 | ||
81a0d3e1 | 1839 | static int mpage_submit_folio(struct mpage_da_data *mpd, struct folio *folio) |
5f1132b2 | 1840 | { |
81a0d3e1 | 1841 | size_t len; |
a056bdaa | 1842 | loff_t size; |
5f1132b2 JK |
1843 | int err; |
1844 | ||
81a0d3e1 MW |
1845 | BUG_ON(folio->index != mpd->first_page); |
1846 | folio_clear_dirty_for_io(folio); | |
a056bdaa JK |
1847 | /* |
1848 | * We have to be very careful here! Nothing protects writeback path | |
1849 | * against i_size changes and the page can be writeably mapped into | |
1850 | * page tables. So an application can be growing i_size and writing | |
81a0d3e1 | 1851 | * data through mmap while writeback runs. folio_clear_dirty_for_io() |
a056bdaa | 1852 | * write-protects our page in page tables and the page cannot get |
81a0d3e1 MW |
1853 | * written to again until we release folio lock. So only after |
1854 | * folio_clear_dirty_for_io() we are safe to sample i_size for | |
e8d6062c MW |
1855 | * ext4_bio_write_folio() to zero-out tail of the written page. We rely |
1856 | * on the barrier provided by folio_test_clear_dirty() in | |
81a0d3e1 | 1857 | * folio_clear_dirty_for_io() to make sure i_size is really sampled only |
a056bdaa JK |
1858 | * after page tables are updated. |
1859 | */ | |
1860 | size = i_size_read(mpd->inode); | |
81a0d3e1 MW |
1861 | len = folio_size(folio); |
1862 | if (folio_pos(folio) + len > size && | |
c93d8f88 | 1863 | !ext4_verity_in_progress(mpd->inode)) |
09cbfeaf | 1864 | len = size & ~PAGE_MASK; |
e8d6062c | 1865 | err = ext4_bio_write_folio(&mpd->io_submit, folio, len); |
5f1132b2 JK |
1866 | if (!err) |
1867 | mpd->wbc->nr_to_write--; | |
5f1132b2 JK |
1868 | |
1869 | return err; | |
1870 | } | |
1871 | ||
6db07461 | 1872 | #define BH_FLAGS (BIT(BH_Unwritten) | BIT(BH_Delay)) |
4e7ea81d | 1873 | |
61628a3f | 1874 | /* |
fffb2739 JK |
1875 | * mballoc gives us at most this number of blocks... |
1876 | * XXX: That seems to be only a limitation of ext4_mb_normalize_request(). | |
70261f56 | 1877 | * The rest of mballoc seems to handle chunks up to full group size. |
61628a3f | 1878 | */ |
fffb2739 | 1879 | #define MAX_WRITEPAGES_EXTENT_LEN 2048 |
525f4ed8 | 1880 | |
4e7ea81d JK |
1881 | /* |
1882 | * mpage_add_bh_to_extent - try to add bh to extent of blocks to map | |
1883 | * | |
1884 | * @mpd - extent of blocks | |
1885 | * @lblk - logical number of the block in the file | |
09930042 | 1886 | * @bh - buffer head we want to add to the extent |
4e7ea81d | 1887 | * |
09930042 JK |
1888 | * The function is used to collect contig. blocks in the same state. If the |
1889 | * buffer doesn't require mapping for writeback and we haven't started the | |
1890 | * extent of buffers to map yet, the function returns 'true' immediately - the | |
1891 | * caller can write the buffer right away. Otherwise the function returns true | |
1892 | * if the block has been added to the extent, false if the block couldn't be | |
1893 | * added. | |
4e7ea81d | 1894 | */ |
09930042 JK |
1895 | static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk, |
1896 | struct buffer_head *bh) | |
4e7ea81d JK |
1897 | { |
1898 | struct ext4_map_blocks *map = &mpd->map; | |
1899 | ||
09930042 JK |
1900 | /* Buffer that doesn't need mapping for writeback? */ |
1901 | if (!buffer_dirty(bh) || !buffer_mapped(bh) || | |
1902 | (!buffer_delay(bh) && !buffer_unwritten(bh))) { | |
1903 | /* So far no extent to map => we write the buffer right away */ | |
1904 | if (map->m_len == 0) | |
1905 | return true; | |
1906 | return false; | |
1907 | } | |
4e7ea81d JK |
1908 | |
1909 | /* First block in the extent? */ | |
1910 | if (map->m_len == 0) { | |
dddbd6ac JK |
1911 | /* We cannot map unless handle is started... */ |
1912 | if (!mpd->do_map) | |
1913 | return false; | |
4e7ea81d JK |
1914 | map->m_lblk = lblk; |
1915 | map->m_len = 1; | |
09930042 JK |
1916 | map->m_flags = bh->b_state & BH_FLAGS; |
1917 | return true; | |
4e7ea81d JK |
1918 | } |
1919 | ||
09930042 JK |
1920 | /* Don't go larger than mballoc is willing to allocate */ |
1921 | if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN) | |
1922 | return false; | |
1923 | ||
4e7ea81d JK |
1924 | /* Can we merge the block to our big extent? */ |
1925 | if (lblk == map->m_lblk + map->m_len && | |
09930042 | 1926 | (bh->b_state & BH_FLAGS) == map->m_flags) { |
4e7ea81d | 1927 | map->m_len++; |
09930042 | 1928 | return true; |
4e7ea81d | 1929 | } |
09930042 | 1930 | return false; |
4e7ea81d JK |
1931 | } |
1932 | ||
5f1132b2 JK |
1933 | /* |
1934 | * mpage_process_page_bufs - submit page buffers for IO or add them to extent | |
1935 | * | |
1936 | * @mpd - extent of blocks for mapping | |
1937 | * @head - the first buffer in the page | |
1938 | * @bh - buffer we should start processing from | |
1939 | * @lblk - logical number of the block in the file corresponding to @bh | |
1940 | * | |
1941 | * Walk through page buffers from @bh upto @head (exclusive) and either submit | |
1942 | * the page for IO if all buffers in this page were mapped and there's no | |
1943 | * accumulated extent of buffers to map or add buffers in the page to the | |
1944 | * extent of buffers to map. The function returns 1 if the caller can continue | |
1945 | * by processing the next page, 0 if it should stop adding buffers to the | |
1946 | * extent to map because we cannot extend it anymore. It can also return value | |
1947 | * < 0 in case of error during IO submission. | |
1948 | */ | |
1949 | static int mpage_process_page_bufs(struct mpage_da_data *mpd, | |
1950 | struct buffer_head *head, | |
1951 | struct buffer_head *bh, | |
1952 | ext4_lblk_t lblk) | |
4e7ea81d JK |
1953 | { |
1954 | struct inode *inode = mpd->inode; | |
5f1132b2 | 1955 | int err; |
93407472 | 1956 | ext4_lblk_t blocks = (i_size_read(inode) + i_blocksize(inode) - 1) |
4e7ea81d JK |
1957 | >> inode->i_blkbits; |
1958 | ||
c93d8f88 EB |
1959 | if (ext4_verity_in_progress(inode)) |
1960 | blocks = EXT_MAX_BLOCKS; | |
1961 | ||
4e7ea81d JK |
1962 | do { |
1963 | BUG_ON(buffer_locked(bh)); | |
1964 | ||
09930042 | 1965 | if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) { |
4e7ea81d JK |
1966 | /* Found extent to map? */ |
1967 | if (mpd->map.m_len) | |
5f1132b2 | 1968 | return 0; |
dddbd6ac JK |
1969 | /* Buffer needs mapping and handle is not started? */ |
1970 | if (!mpd->do_map) | |
1971 | return 0; | |
09930042 | 1972 | /* Everything mapped so far and we hit EOF */ |
5f1132b2 | 1973 | break; |
4e7ea81d | 1974 | } |
4e7ea81d | 1975 | } while (lblk++, (bh = bh->b_this_page) != head); |
5f1132b2 JK |
1976 | /* So far everything mapped? Submit the page for IO. */ |
1977 | if (mpd->map.m_len == 0) { | |
81a0d3e1 | 1978 | err = mpage_submit_folio(mpd, head->b_folio); |
5f1132b2 JK |
1979 | if (err < 0) |
1980 | return err; | |
33483b3b | 1981 | mpage_folio_done(mpd, head->b_folio); |
5f1132b2 | 1982 | } |
6b8ed620 JK |
1983 | if (lblk >= blocks) { |
1984 | mpd->scanned_until_end = 1; | |
1985 | return 0; | |
1986 | } | |
1987 | return 1; | |
4e7ea81d JK |
1988 | } |
1989 | ||
2943fdbc | 1990 | /* |
4da2f6e3 MW |
1991 | * mpage_process_folio - update folio buffers corresponding to changed extent |
1992 | * and may submit fully mapped page for IO | |
1993 | * @mpd: description of extent to map, on return next extent to map | |
1994 | * @folio: Contains these buffers. | |
1995 | * @m_lblk: logical block mapping. | |
1996 | * @m_pblk: corresponding physical mapping. | |
1997 | * @map_bh: determines on return whether this page requires any further | |
2943fdbc | 1998 | * mapping or not. |
4da2f6e3 MW |
1999 | * |
2000 | * Scan given folio buffers corresponding to changed extent and update buffer | |
2943fdbc RH |
2001 | * state according to new extent state. |
2002 | * We map delalloc buffers to their physical location, clear unwritten bits. | |
4da2f6e3 MW |
2003 | * If the given folio is not fully mapped, we update @mpd to the next extent in |
2004 | * the given folio that needs mapping & return @map_bh as true. | |
2943fdbc | 2005 | */ |
4da2f6e3 | 2006 | static int mpage_process_folio(struct mpage_da_data *mpd, struct folio *folio, |
2943fdbc RH |
2007 | ext4_lblk_t *m_lblk, ext4_fsblk_t *m_pblk, |
2008 | bool *map_bh) | |
2009 | { | |
2010 | struct buffer_head *head, *bh; | |
2011 | ext4_io_end_t *io_end = mpd->io_submit.io_end; | |
2012 | ext4_lblk_t lblk = *m_lblk; | |
2013 | ext4_fsblk_t pblock = *m_pblk; | |
2014 | int err = 0; | |
c8cc8816 RH |
2015 | int blkbits = mpd->inode->i_blkbits; |
2016 | ssize_t io_end_size = 0; | |
2017 | struct ext4_io_end_vec *io_end_vec = ext4_last_io_end_vec(io_end); | |
2943fdbc | 2018 | |
4da2f6e3 | 2019 | bh = head = folio_buffers(folio); |
2943fdbc RH |
2020 | do { |
2021 | if (lblk < mpd->map.m_lblk) | |
2022 | continue; | |
2023 | if (lblk >= mpd->map.m_lblk + mpd->map.m_len) { | |
2024 | /* | |
2025 | * Buffer after end of mapped extent. | |
4da2f6e3 | 2026 | * Find next buffer in the folio to map. |
2943fdbc RH |
2027 | */ |
2028 | mpd->map.m_len = 0; | |
2029 | mpd->map.m_flags = 0; | |
c8cc8816 | 2030 | io_end_vec->size += io_end_size; |
2943fdbc | 2031 | |
2943fdbc RH |
2032 | err = mpage_process_page_bufs(mpd, head, bh, lblk); |
2033 | if (err > 0) | |
2034 | err = 0; | |
c8cc8816 RH |
2035 | if (!err && mpd->map.m_len && mpd->map.m_lblk > lblk) { |
2036 | io_end_vec = ext4_alloc_io_end_vec(io_end); | |
4d06bfb9 RH |
2037 | if (IS_ERR(io_end_vec)) { |
2038 | err = PTR_ERR(io_end_vec); | |
2039 | goto out; | |
2040 | } | |
d1e18b88 | 2041 | io_end_vec->offset = (loff_t)mpd->map.m_lblk << blkbits; |
c8cc8816 | 2042 | } |
2943fdbc RH |
2043 | *map_bh = true; |
2044 | goto out; | |
2045 | } | |
2046 | if (buffer_delay(bh)) { | |
2047 | clear_buffer_delay(bh); | |
2048 | bh->b_blocknr = pblock++; | |
2049 | } | |
2050 | clear_buffer_unwritten(bh); | |
c8cc8816 | 2051 | io_end_size += (1 << blkbits); |
2943fdbc | 2052 | } while (lblk++, (bh = bh->b_this_page) != head); |
c8cc8816 RH |
2053 | |
2054 | io_end_vec->size += io_end_size; | |
2943fdbc RH |
2055 | *map_bh = false; |
2056 | out: | |
2057 | *m_lblk = lblk; | |
2058 | *m_pblk = pblock; | |
2059 | return err; | |
2060 | } | |
2061 | ||
4e7ea81d JK |
2062 | /* |
2063 | * mpage_map_buffers - update buffers corresponding to changed extent and | |
2064 | * submit fully mapped pages for IO | |
2065 | * | |
2066 | * @mpd - description of extent to map, on return next extent to map | |
2067 | * | |
2068 | * Scan buffers corresponding to changed extent (we expect corresponding pages | |
2069 | * to be already locked) and update buffer state according to new extent state. | |
2070 | * We map delalloc buffers to their physical location, clear unwritten bits, | |
556615dc | 2071 | * and mark buffers as uninit when we perform writes to unwritten extents |
4e7ea81d JK |
2072 | * and do extent conversion after IO is finished. If the last page is not fully |
2073 | * mapped, we update @map to the next extent in the last page that needs | |
2074 | * mapping. Otherwise we submit the page for IO. | |
2075 | */ | |
2076 | static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd) | |
2077 | { | |
7530d093 MWO |
2078 | struct folio_batch fbatch; |
2079 | unsigned nr, i; | |
4e7ea81d | 2080 | struct inode *inode = mpd->inode; |
09cbfeaf | 2081 | int bpp_bits = PAGE_SHIFT - inode->i_blkbits; |
4e7ea81d JK |
2082 | pgoff_t start, end; |
2083 | ext4_lblk_t lblk; | |
2943fdbc | 2084 | ext4_fsblk_t pblock; |
4e7ea81d | 2085 | int err; |
2943fdbc | 2086 | bool map_bh = false; |
4e7ea81d JK |
2087 | |
2088 | start = mpd->map.m_lblk >> bpp_bits; | |
2089 | end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits; | |
2090 | lblk = start << bpp_bits; | |
2091 | pblock = mpd->map.m_pblk; | |
2092 | ||
7530d093 | 2093 | folio_batch_init(&fbatch); |
4e7ea81d | 2094 | while (start <= end) { |
7530d093 MWO |
2095 | nr = filemap_get_folios(inode->i_mapping, &start, end, &fbatch); |
2096 | if (nr == 0) | |
4e7ea81d | 2097 | break; |
7530d093 | 2098 | for (i = 0; i < nr; i++) { |
4da2f6e3 | 2099 | struct folio *folio = fbatch.folios[i]; |
4e7ea81d | 2100 | |
4da2f6e3 | 2101 | err = mpage_process_folio(mpd, folio, &lblk, &pblock, |
2943fdbc | 2102 | &map_bh); |
4e7ea81d | 2103 | /* |
2943fdbc RH |
2104 | * If map_bh is true, means page may require further bh |
2105 | * mapping, or maybe the page was submitted for IO. | |
2106 | * So we return to call further extent mapping. | |
4e7ea81d | 2107 | */ |
39c0ae16 | 2108 | if (err < 0 || map_bh) |
2943fdbc | 2109 | goto out; |
4e7ea81d | 2110 | /* Page fully mapped - let IO run! */ |
81a0d3e1 | 2111 | err = mpage_submit_folio(mpd, folio); |
2943fdbc RH |
2112 | if (err < 0) |
2113 | goto out; | |
33483b3b | 2114 | mpage_folio_done(mpd, folio); |
4e7ea81d | 2115 | } |
7530d093 | 2116 | folio_batch_release(&fbatch); |
4e7ea81d JK |
2117 | } |
2118 | /* Extent fully mapped and matches with page boundary. We are done. */ | |
2119 | mpd->map.m_len = 0; | |
2120 | mpd->map.m_flags = 0; | |
2121 | return 0; | |
2943fdbc | 2122 | out: |
7530d093 | 2123 | folio_batch_release(&fbatch); |
2943fdbc | 2124 | return err; |
4e7ea81d JK |
2125 | } |
2126 | ||
2127 | static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd) | |
2128 | { | |
2129 | struct inode *inode = mpd->inode; | |
2130 | struct ext4_map_blocks *map = &mpd->map; | |
2131 | int get_blocks_flags; | |
090f32ee | 2132 | int err, dioread_nolock; |
4e7ea81d JK |
2133 | |
2134 | trace_ext4_da_write_pages_extent(inode, map); | |
2135 | /* | |
2136 | * Call ext4_map_blocks() to allocate any delayed allocation blocks, or | |
556615dc | 2137 | * to convert an unwritten extent to be initialized (in the case |
4e7ea81d JK |
2138 | * where we have written into one or more preallocated blocks). It is |
2139 | * possible that we're going to need more metadata blocks than | |
2140 | * previously reserved. However we must not fail because we're in | |
2141 | * writeback and there is nothing we can do about it so it might result | |
2142 | * in data loss. So use reserved blocks to allocate metadata if | |
2143 | * possible. | |
2144 | * | |
754cfed6 TT |
2145 | * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if |
2146 | * the blocks in question are delalloc blocks. This indicates | |
2147 | * that the blocks and quotas has already been checked when | |
2148 | * the data was copied into the page cache. | |
4e7ea81d JK |
2149 | */ |
2150 | get_blocks_flags = EXT4_GET_BLOCKS_CREATE | | |
ee0876bc JK |
2151 | EXT4_GET_BLOCKS_METADATA_NOFAIL | |
2152 | EXT4_GET_BLOCKS_IO_SUBMIT; | |
090f32ee LC |
2153 | dioread_nolock = ext4_should_dioread_nolock(inode); |
2154 | if (dioread_nolock) | |
4e7ea81d | 2155 | get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT; |
6db07461 | 2156 | if (map->m_flags & BIT(BH_Delay)) |
4e7ea81d JK |
2157 | get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE; |
2158 | ||
2159 | err = ext4_map_blocks(handle, inode, map, get_blocks_flags); | |
2160 | if (err < 0) | |
2161 | return err; | |
090f32ee | 2162 | if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) { |
6b523df4 JK |
2163 | if (!mpd->io_submit.io_end->handle && |
2164 | ext4_handle_valid(handle)) { | |
2165 | mpd->io_submit.io_end->handle = handle->h_rsv_handle; | |
2166 | handle->h_rsv_handle = NULL; | |
2167 | } | |
3613d228 | 2168 | ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end); |
6b523df4 | 2169 | } |
4e7ea81d JK |
2170 | |
2171 | BUG_ON(map->m_len == 0); | |
4e7ea81d JK |
2172 | return 0; |
2173 | } | |
2174 | ||
2175 | /* | |
2176 | * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length | |
2177 | * mpd->len and submit pages underlying it for IO | |
2178 | * | |
2179 | * @handle - handle for journal operations | |
2180 | * @mpd - extent to map | |
7534e854 JK |
2181 | * @give_up_on_write - we set this to true iff there is a fatal error and there |
2182 | * is no hope of writing the data. The caller should discard | |
2183 | * dirty pages to avoid infinite loops. | |
4e7ea81d JK |
2184 | * |
2185 | * The function maps extent starting at mpd->lblk of length mpd->len. If it is | |
2186 | * delayed, blocks are allocated, if it is unwritten, we may need to convert | |
2187 | * them to initialized or split the described range from larger unwritten | |
2188 | * extent. Note that we need not map all the described range since allocation | |
2189 | * can return less blocks or the range is covered by more unwritten extents. We | |
2190 | * cannot map more because we are limited by reserved transaction credits. On | |
2191 | * the other hand we always make sure that the last touched page is fully | |
2192 | * mapped so that it can be written out (and thus forward progress is | |
2193 | * guaranteed). After mapping we submit all mapped pages for IO. | |
2194 | */ | |
2195 | static int mpage_map_and_submit_extent(handle_t *handle, | |
cb530541 TT |
2196 | struct mpage_da_data *mpd, |
2197 | bool *give_up_on_write) | |
4e7ea81d JK |
2198 | { |
2199 | struct inode *inode = mpd->inode; | |
2200 | struct ext4_map_blocks *map = &mpd->map; | |
2201 | int err; | |
2202 | loff_t disksize; | |
6603120e | 2203 | int progress = 0; |
c8cc8816 | 2204 | ext4_io_end_t *io_end = mpd->io_submit.io_end; |
4d06bfb9 | 2205 | struct ext4_io_end_vec *io_end_vec; |
4e7ea81d | 2206 | |
4d06bfb9 RH |
2207 | io_end_vec = ext4_alloc_io_end_vec(io_end); |
2208 | if (IS_ERR(io_end_vec)) | |
2209 | return PTR_ERR(io_end_vec); | |
c8cc8816 | 2210 | io_end_vec->offset = ((loff_t)map->m_lblk) << inode->i_blkbits; |
27d7c4ed | 2211 | do { |
4e7ea81d JK |
2212 | err = mpage_map_one_extent(handle, mpd); |
2213 | if (err < 0) { | |
2214 | struct super_block *sb = inode->i_sb; | |
2215 | ||
0db1ff22 | 2216 | if (ext4_forced_shutdown(EXT4_SB(sb)) || |
9b5f6c9b | 2217 | ext4_test_mount_flag(sb, EXT4_MF_FS_ABORTED)) |
cb530541 | 2218 | goto invalidate_dirty_pages; |
4e7ea81d | 2219 | /* |
cb530541 TT |
2220 | * Let the uper layers retry transient errors. |
2221 | * In the case of ENOSPC, if ext4_count_free_blocks() | |
2222 | * is non-zero, a commit should free up blocks. | |
4e7ea81d | 2223 | */ |
cb530541 | 2224 | if ((err == -ENOMEM) || |
6603120e DM |
2225 | (err == -ENOSPC && ext4_count_free_clusters(sb))) { |
2226 | if (progress) | |
2227 | goto update_disksize; | |
cb530541 | 2228 | return err; |
6603120e | 2229 | } |
cb530541 TT |
2230 | ext4_msg(sb, KERN_CRIT, |
2231 | "Delayed block allocation failed for " | |
2232 | "inode %lu at logical offset %llu with" | |
2233 | " max blocks %u with error %d", | |
2234 | inode->i_ino, | |
2235 | (unsigned long long)map->m_lblk, | |
2236 | (unsigned)map->m_len, -err); | |
2237 | ext4_msg(sb, KERN_CRIT, | |
2238 | "This should not happen!! Data will " | |
2239 | "be lost\n"); | |
2240 | if (err == -ENOSPC) | |
2241 | ext4_print_free_blocks(inode); | |
2242 | invalidate_dirty_pages: | |
2243 | *give_up_on_write = true; | |
4e7ea81d JK |
2244 | return err; |
2245 | } | |
6603120e | 2246 | progress = 1; |
4e7ea81d JK |
2247 | /* |
2248 | * Update buffer state, submit mapped pages, and get us new | |
2249 | * extent to map | |
2250 | */ | |
2251 | err = mpage_map_and_submit_buffers(mpd); | |
2252 | if (err < 0) | |
6603120e | 2253 | goto update_disksize; |
27d7c4ed | 2254 | } while (map->m_len); |
4e7ea81d | 2255 | |
6603120e | 2256 | update_disksize: |
622cad13 TT |
2257 | /* |
2258 | * Update on-disk size after IO is submitted. Races with | |
2259 | * truncate are avoided by checking i_size under i_data_sem. | |
2260 | */ | |
09cbfeaf | 2261 | disksize = ((loff_t)mpd->first_page) << PAGE_SHIFT; |
35df4299 | 2262 | if (disksize > READ_ONCE(EXT4_I(inode)->i_disksize)) { |
4e7ea81d | 2263 | int err2; |
622cad13 TT |
2264 | loff_t i_size; |
2265 | ||
2266 | down_write(&EXT4_I(inode)->i_data_sem); | |
2267 | i_size = i_size_read(inode); | |
2268 | if (disksize > i_size) | |
2269 | disksize = i_size; | |
2270 | if (disksize > EXT4_I(inode)->i_disksize) | |
2271 | EXT4_I(inode)->i_disksize = disksize; | |
622cad13 | 2272 | up_write(&EXT4_I(inode)->i_data_sem); |
b907f2d5 | 2273 | err2 = ext4_mark_inode_dirty(handle, inode); |
878520ac | 2274 | if (err2) { |
54d3adbc TT |
2275 | ext4_error_err(inode->i_sb, -err2, |
2276 | "Failed to mark inode %lu dirty", | |
2277 | inode->i_ino); | |
878520ac | 2278 | } |
4e7ea81d JK |
2279 | if (!err) |
2280 | err = err2; | |
2281 | } | |
2282 | return err; | |
2283 | } | |
2284 | ||
fffb2739 JK |
2285 | /* |
2286 | * Calculate the total number of credits to reserve for one writepages | |
20970ba6 | 2287 | * iteration. This is called from ext4_writepages(). We map an extent of |
70261f56 | 2288 | * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping |
fffb2739 JK |
2289 | * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN + |
2290 | * bpp - 1 blocks in bpp different extents. | |
2291 | */ | |
525f4ed8 MC |
2292 | static int ext4_da_writepages_trans_blocks(struct inode *inode) |
2293 | { | |
fffb2739 | 2294 | int bpp = ext4_journal_blocks_per_page(inode); |
525f4ed8 | 2295 | |
fffb2739 JK |
2296 | return ext4_meta_trans_blocks(inode, |
2297 | MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp); | |
525f4ed8 | 2298 | } |
61628a3f | 2299 | |
80be8c5c RH |
2300 | static int ext4_journal_folio_buffers(handle_t *handle, struct folio *folio, |
2301 | size_t len) | |
3f079114 | 2302 | { |
80be8c5c RH |
2303 | struct buffer_head *page_bufs = folio_buffers(folio); |
2304 | struct inode *inode = folio->mapping->host; | |
3f079114 JK |
2305 | int ret, err; |
2306 | ||
2307 | ret = ext4_walk_page_buffers(handle, inode, page_bufs, 0, len, | |
2308 | NULL, do_journal_get_write_access); | |
2309 | err = ext4_walk_page_buffers(handle, inode, page_bufs, 0, len, | |
2310 | NULL, write_end_fn); | |
2311 | if (ret == 0) | |
2312 | ret = err; | |
80be8c5c | 2313 | err = ext4_jbd2_inode_add_write(handle, inode, folio_pos(folio), len); |
3f079114 JK |
2314 | if (ret == 0) |
2315 | ret = err; | |
2316 | EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; | |
2317 | ||
3f079114 JK |
2318 | return ret; |
2319 | } | |
2320 | ||
2321 | static int mpage_journal_page_buffers(handle_t *handle, | |
2322 | struct mpage_da_data *mpd, | |
80be8c5c | 2323 | struct folio *folio) |
3f079114 JK |
2324 | { |
2325 | struct inode *inode = mpd->inode; | |
2326 | loff_t size = i_size_read(inode); | |
80be8c5c | 2327 | size_t len = folio_size(folio); |
3f079114 | 2328 | |
80be8c5c | 2329 | folio_clear_checked(folio); |
3f079114 JK |
2330 | mpd->wbc->nr_to_write--; |
2331 | ||
80be8c5c | 2332 | if (folio_pos(folio) + len > size && |
3f079114 | 2333 | !ext4_verity_in_progress(inode)) |
80be8c5c | 2334 | len = size - folio_pos(folio); |
3f079114 | 2335 | |
80be8c5c | 2336 | return ext4_journal_folio_buffers(handle, folio, len); |
3f079114 JK |
2337 | } |
2338 | ||
8e48dcfb | 2339 | /* |
4e7ea81d | 2340 | * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages |
de0039f6 | 2341 | * needing mapping, submit mapped pages |
4e7ea81d JK |
2342 | * |
2343 | * @mpd - where to look for pages | |
2344 | * | |
2345 | * Walk dirty pages in the mapping. If they are fully mapped, submit them for | |
de0039f6 JK |
2346 | * IO immediately. If we cannot map blocks, we submit just already mapped |
2347 | * buffers in the page for IO and keep page dirty. When we can map blocks and | |
2348 | * we find a page which isn't mapped we start accumulating extent of buffers | |
2349 | * underlying these pages that needs mapping (formed by either delayed or | |
2350 | * unwritten buffers). We also lock the pages containing these buffers. The | |
2351 | * extent found is returned in @mpd structure (starting at mpd->lblk with | |
2352 | * length mpd->len blocks). | |
4e7ea81d JK |
2353 | * |
2354 | * Note that this function can attach bios to one io_end structure which are | |
2355 | * neither logically nor physically contiguous. Although it may seem as an | |
2356 | * unnecessary complication, it is actually inevitable in blocksize < pagesize | |
2357 | * case as we need to track IO to all buffers underlying a page in one io_end. | |
8e48dcfb | 2358 | */ |
4e7ea81d | 2359 | static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd) |
8e48dcfb | 2360 | { |
4e7ea81d | 2361 | struct address_space *mapping = mpd->inode->i_mapping; |
50ead253 VMO |
2362 | struct folio_batch fbatch; |
2363 | unsigned int nr_folios; | |
4e7ea81d JK |
2364 | pgoff_t index = mpd->first_page; |
2365 | pgoff_t end = mpd->last_page; | |
10bbd235 | 2366 | xa_mark_t tag; |
4e7ea81d JK |
2367 | int i, err = 0; |
2368 | int blkbits = mpd->inode->i_blkbits; | |
2369 | ext4_lblk_t lblk; | |
2370 | struct buffer_head *head; | |
3f079114 JK |
2371 | handle_t *handle = NULL; |
2372 | int bpp = ext4_journal_blocks_per_page(mpd->inode); | |
8e48dcfb | 2373 | |
4e7ea81d | 2374 | if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages) |
5b41d924 ES |
2375 | tag = PAGECACHE_TAG_TOWRITE; |
2376 | else | |
2377 | tag = PAGECACHE_TAG_DIRTY; | |
3f079114 | 2378 | |
e6c28a26 JK |
2379 | mpd->map.m_len = 0; |
2380 | mpd->next_page = index; | |
d0ab8368 | 2381 | if (ext4_should_journal_data(mpd->inode)) { |
3f079114 JK |
2382 | handle = ext4_journal_start(mpd->inode, EXT4_HT_WRITE_PAGE, |
2383 | bpp); | |
2384 | if (IS_ERR(handle)) | |
2385 | return PTR_ERR(handle); | |
2386 | } | |
50ead253 | 2387 | folio_batch_init(&fbatch); |
4f01b02c | 2388 | while (index <= end) { |
50ead253 VMO |
2389 | nr_folios = filemap_get_folios_tag(mapping, &index, end, |
2390 | tag, &fbatch); | |
2391 | if (nr_folios == 0) | |
6b8ed620 | 2392 | break; |
8e48dcfb | 2393 | |
50ead253 VMO |
2394 | for (i = 0; i < nr_folios; i++) { |
2395 | struct folio *folio = fbatch.folios[i]; | |
8e48dcfb | 2396 | |
aeac589a ML |
2397 | /* |
2398 | * Accumulated enough dirty pages? This doesn't apply | |
2399 | * to WB_SYNC_ALL mode. For integrity sync we have to | |
2400 | * keep going because someone may be concurrently | |
2401 | * dirtying pages, and we might have synced a lot of | |
2402 | * newly appeared dirty pages, but have not synced all | |
2403 | * of the old dirty pages. | |
2404 | */ | |
c8e8e16d JK |
2405 | if (mpd->wbc->sync_mode == WB_SYNC_NONE && |
2406 | mpd->wbc->nr_to_write <= | |
2407 | mpd->map.m_len >> (PAGE_SHIFT - blkbits)) | |
aeac589a ML |
2408 | goto out; |
2409 | ||
4e7ea81d | 2410 | /* If we can't merge this page, we are done. */ |
50ead253 | 2411 | if (mpd->map.m_len > 0 && mpd->next_page != folio->index) |
4e7ea81d | 2412 | goto out; |
78aaced3 | 2413 | |
3f079114 JK |
2414 | if (handle) { |
2415 | err = ext4_journal_ensure_credits(handle, bpp, | |
2416 | 0); | |
2417 | if (err < 0) | |
2418 | goto out; | |
2419 | } | |
2420 | ||
50ead253 | 2421 | folio_lock(folio); |
8e48dcfb | 2422 | /* |
4e7ea81d JK |
2423 | * If the page is no longer dirty, or its mapping no |
2424 | * longer corresponds to inode we are writing (which | |
2425 | * means it has been truncated or invalidated), or the | |
2426 | * page is already under writeback and we are not doing | |
2427 | * a data integrity writeback, skip the page | |
8e48dcfb | 2428 | */ |
50ead253 VMO |
2429 | if (!folio_test_dirty(folio) || |
2430 | (folio_test_writeback(folio) && | |
4e7ea81d | 2431 | (mpd->wbc->sync_mode == WB_SYNC_NONE)) || |
50ead253 VMO |
2432 | unlikely(folio->mapping != mapping)) { |
2433 | folio_unlock(folio); | |
8e48dcfb TT |
2434 | continue; |
2435 | } | |
2436 | ||
50ead253 VMO |
2437 | folio_wait_writeback(folio); |
2438 | BUG_ON(folio_test_writeback(folio)); | |
8e48dcfb | 2439 | |
cc509574 TT |
2440 | /* |
2441 | * Should never happen but for buggy code in | |
2442 | * other subsystems that call | |
2443 | * set_page_dirty() without properly warning | |
2444 | * the file system first. See [1] for more | |
2445 | * information. | |
2446 | * | |
2447 | * [1] https://lore.kernel.org/linux-mm/20180103100430.GE4911@quack2.suse.cz | |
2448 | */ | |
50ead253 VMO |
2449 | if (!folio_buffers(folio)) { |
2450 | ext4_warning_inode(mpd->inode, "page %lu does not have buffers attached", folio->index); | |
2451 | folio_clear_dirty(folio); | |
2452 | folio_unlock(folio); | |
cc509574 TT |
2453 | continue; |
2454 | } | |
2455 | ||
4e7ea81d | 2456 | if (mpd->map.m_len == 0) |
50ead253 VMO |
2457 | mpd->first_page = folio->index; |
2458 | mpd->next_page = folio->index + folio_nr_pages(folio); | |
de0039f6 | 2459 | /* |
3f079114 JK |
2460 | * Writeout when we cannot modify metadata is simple. |
2461 | * Just submit the page. For data=journal mode we | |
2462 | * first handle writeout of the page for checkpoint and | |
2463 | * only after that handle delayed page dirtying. This | |
ab382539 JK |
2464 | * makes sure current data is checkpointed to the final |
2465 | * location before possibly journalling it again which | |
2466 | * is desirable when the page is frequently dirtied | |
2467 | * through a pin. | |
de0039f6 JK |
2468 | */ |
2469 | if (!mpd->can_map) { | |
5e1bdea6 JK |
2470 | err = mpage_submit_folio(mpd, folio); |
2471 | if (err < 0) | |
2472 | goto out; | |
3f079114 | 2473 | /* Pending dirtying of journalled data? */ |
81a0d3e1 | 2474 | if (folio_test_checked(folio)) { |
3f079114 | 2475 | err = mpage_journal_page_buffers(handle, |
80be8c5c | 2476 | mpd, folio); |
3f079114 JK |
2477 | if (err < 0) |
2478 | goto out; | |
1f1a55f0 | 2479 | mpd->journalled_more_data = 1; |
3f079114 | 2480 | } |
33483b3b | 2481 | mpage_folio_done(mpd, folio); |
de0039f6 JK |
2482 | } else { |
2483 | /* Add all dirty buffers to mpd */ | |
50ead253 | 2484 | lblk = ((ext4_lblk_t)folio->index) << |
de0039f6 | 2485 | (PAGE_SHIFT - blkbits); |
50ead253 | 2486 | head = folio_buffers(folio); |
de0039f6 | 2487 | err = mpage_process_page_bufs(mpd, head, head, |
50ead253 | 2488 | lblk); |
de0039f6 JK |
2489 | if (err <= 0) |
2490 | goto out; | |
2491 | err = 0; | |
2492 | } | |
8e48dcfb | 2493 | } |
50ead253 | 2494 | folio_batch_release(&fbatch); |
8e48dcfb TT |
2495 | cond_resched(); |
2496 | } | |
6b8ed620 | 2497 | mpd->scanned_until_end = 1; |
3f079114 JK |
2498 | if (handle) |
2499 | ext4_journal_stop(handle); | |
4f01b02c | 2500 | return 0; |
8eb9e5ce | 2501 | out: |
50ead253 | 2502 | folio_batch_release(&fbatch); |
3f079114 JK |
2503 | if (handle) |
2504 | ext4_journal_stop(handle); | |
4e7ea81d | 2505 | return err; |
8e48dcfb TT |
2506 | } |
2507 | ||
15648d59 | 2508 | static int ext4_do_writepages(struct mpage_da_data *mpd) |
64769240 | 2509 | { |
15648d59 | 2510 | struct writeback_control *wbc = mpd->wbc; |
4e7ea81d JK |
2511 | pgoff_t writeback_index = 0; |
2512 | long nr_to_write = wbc->nr_to_write; | |
22208ded | 2513 | int range_whole = 0; |
4e7ea81d | 2514 | int cycled = 1; |
61628a3f | 2515 | handle_t *handle = NULL; |
15648d59 JK |
2516 | struct inode *inode = mpd->inode; |
2517 | struct address_space *mapping = inode->i_mapping; | |
6b523df4 | 2518 | int needed_blocks, rsv_blocks = 0, ret = 0; |
5e745b04 | 2519 | struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb); |
1bce63d1 | 2520 | struct blk_plug plug; |
cb530541 | 2521 | bool give_up_on_write = false; |
61628a3f | 2522 | |
20970ba6 | 2523 | trace_ext4_writepages(inode, wbc); |
ba80b101 | 2524 | |
61628a3f MC |
2525 | /* |
2526 | * No pages to write? This is mainly a kludge to avoid starting | |
2527 | * a transaction for special inodes like journal inode on last iput() | |
2528 | * because that could violate lock ordering on umount | |
2529 | */ | |
a1d6cc56 | 2530 | if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) |
bbf023c7 | 2531 | goto out_writepages; |
2a21e37e TT |
2532 | |
2533 | /* | |
2534 | * If the filesystem has aborted, it is read-only, so return | |
2535 | * right away instead of dumping stack traces later on that | |
2536 | * will obscure the real source of the problem. We test | |
1751e8a6 | 2537 | * EXT4_MF_FS_ABORTED instead of sb->s_flag's SB_RDONLY because |
2a21e37e | 2538 | * the latter could be true if the filesystem is mounted |
20970ba6 | 2539 | * read-only, and in that case, ext4_writepages should |
2a21e37e TT |
2540 | * *never* be called, so if that ever happens, we would want |
2541 | * the stack trace. | |
2542 | */ | |
0db1ff22 | 2543 | if (unlikely(ext4_forced_shutdown(EXT4_SB(mapping->host->i_sb)) || |
9b5f6c9b | 2544 | ext4_test_mount_flag(inode->i_sb, EXT4_MF_FS_ABORTED))) { |
bbf023c7 ML |
2545 | ret = -EROFS; |
2546 | goto out_writepages; | |
2547 | } | |
2a21e37e | 2548 | |
4e7ea81d JK |
2549 | /* |
2550 | * If we have inline data and arrive here, it means that | |
2551 | * we will soon create the block for the 1st page, so | |
2552 | * we'd better clear the inline data here. | |
2553 | */ | |
2554 | if (ext4_has_inline_data(inode)) { | |
2555 | /* Just inode will be modified... */ | |
2556 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 1); | |
2557 | if (IS_ERR(handle)) { | |
2558 | ret = PTR_ERR(handle); | |
2559 | goto out_writepages; | |
2560 | } | |
2561 | BUG_ON(ext4_test_inode_state(inode, | |
2562 | EXT4_STATE_MAY_INLINE_DATA)); | |
2563 | ext4_destroy_inline_data(handle, inode); | |
2564 | ext4_journal_stop(handle); | |
2565 | } | |
2566 | ||
3f079114 JK |
2567 | /* |
2568 | * data=journal mode does not do delalloc so we just need to writeout / | |
1f1a55f0 JK |
2569 | * journal already mapped buffers. On the other hand we need to commit |
2570 | * transaction to make data stable. We expect all the data to be | |
2571 | * already in the journal (the only exception are DMA pinned pages | |
2572 | * dirtied behind our back) so we commit transaction here and run the | |
2573 | * writeback loop to checkpoint them. The checkpointing is not actually | |
2574 | * necessary to make data persistent *but* quite a few places (extent | |
2575 | * shifting operations, fsverity, ...) depend on being able to drop | |
2576 | * pagecache pages after calling filemap_write_and_wait() and for that | |
2577 | * checkpointing needs to happen. | |
3f079114 | 2578 | */ |
1f1a55f0 | 2579 | if (ext4_should_journal_data(inode)) { |
3f079114 | 2580 | mpd->can_map = 0; |
1f1a55f0 JK |
2581 | if (wbc->sync_mode == WB_SYNC_ALL) |
2582 | ext4_fc_commit(sbi->s_journal, | |
2583 | EXT4_I(inode)->i_datasync_tid); | |
2584 | } | |
2585 | mpd->journalled_more_data = 0; | |
3f079114 | 2586 | |
4e343231 | 2587 | if (ext4_should_dioread_nolock(inode)) { |
2588 | /* | |
2589 | * We may need to convert up to one extent per block in | |
2590 | * the page and we may dirty the inode. | |
2591 | */ | |
2592 | rsv_blocks = 1 + ext4_chunk_trans_blocks(inode, | |
2593 | PAGE_SIZE >> inode->i_blkbits); | |
2594 | } | |
2595 | ||
22208ded AK |
2596 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
2597 | range_whole = 1; | |
61628a3f | 2598 | |
2acf2c26 | 2599 | if (wbc->range_cyclic) { |
4e7ea81d JK |
2600 | writeback_index = mapping->writeback_index; |
2601 | if (writeback_index) | |
2acf2c26 | 2602 | cycled = 0; |
15648d59 JK |
2603 | mpd->first_page = writeback_index; |
2604 | mpd->last_page = -1; | |
5b41d924 | 2605 | } else { |
15648d59 JK |
2606 | mpd->first_page = wbc->range_start >> PAGE_SHIFT; |
2607 | mpd->last_page = wbc->range_end >> PAGE_SHIFT; | |
5b41d924 | 2608 | } |
a1d6cc56 | 2609 | |
15648d59 | 2610 | ext4_io_submit_init(&mpd->io_submit, wbc); |
2acf2c26 | 2611 | retry: |
6e6938b6 | 2612 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
15648d59 JK |
2613 | tag_pages_for_writeback(mapping, mpd->first_page, |
2614 | mpd->last_page); | |
1bce63d1 | 2615 | blk_start_plug(&plug); |
dddbd6ac JK |
2616 | |
2617 | /* | |
2618 | * First writeback pages that don't need mapping - we can avoid | |
2619 | * starting a transaction unnecessarily and also avoid being blocked | |
2620 | * in the block layer on device congestion while having transaction | |
2621 | * started. | |
2622 | */ | |
15648d59 JK |
2623 | mpd->do_map = 0; |
2624 | mpd->scanned_until_end = 0; | |
2625 | mpd->io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL); | |
2626 | if (!mpd->io_submit.io_end) { | |
dddbd6ac JK |
2627 | ret = -ENOMEM; |
2628 | goto unplug; | |
2629 | } | |
15648d59 | 2630 | ret = mpage_prepare_extent_to_map(mpd); |
a297b2fc | 2631 | /* Unlock pages we didn't use */ |
15648d59 | 2632 | mpage_release_unused_pages(mpd, false); |
dddbd6ac | 2633 | /* Submit prepared bio */ |
15648d59 JK |
2634 | ext4_io_submit(&mpd->io_submit); |
2635 | ext4_put_io_end_defer(mpd->io_submit.io_end); | |
2636 | mpd->io_submit.io_end = NULL; | |
dddbd6ac JK |
2637 | if (ret < 0) |
2638 | goto unplug; | |
2639 | ||
15648d59 | 2640 | while (!mpd->scanned_until_end && wbc->nr_to_write > 0) { |
4e7ea81d | 2641 | /* For each extent of pages we use new io_end */ |
15648d59 JK |
2642 | mpd->io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL); |
2643 | if (!mpd->io_submit.io_end) { | |
4e7ea81d JK |
2644 | ret = -ENOMEM; |
2645 | break; | |
2646 | } | |
a1d6cc56 | 2647 | |
de0039f6 | 2648 | WARN_ON_ONCE(!mpd->can_map); |
a1d6cc56 | 2649 | /* |
4e7ea81d JK |
2650 | * We have two constraints: We find one extent to map and we |
2651 | * must always write out whole page (makes a difference when | |
2652 | * blocksize < pagesize) so that we don't block on IO when we | |
2653 | * try to write out the rest of the page. Journalled mode is | |
2654 | * not supported by delalloc. | |
a1d6cc56 AK |
2655 | */ |
2656 | BUG_ON(ext4_should_journal_data(inode)); | |
525f4ed8 | 2657 | needed_blocks = ext4_da_writepages_trans_blocks(inode); |
a1d6cc56 | 2658 | |
4e7ea81d | 2659 | /* start a new transaction */ |
6b523df4 JK |
2660 | handle = ext4_journal_start_with_reserve(inode, |
2661 | EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks); | |
61628a3f MC |
2662 | if (IS_ERR(handle)) { |
2663 | ret = PTR_ERR(handle); | |
1693918e | 2664 | ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: " |
fbe845dd | 2665 | "%ld pages, ino %lu; err %d", __func__, |
a1d6cc56 | 2666 | wbc->nr_to_write, inode->i_ino, ret); |
4e7ea81d | 2667 | /* Release allocated io_end */ |
15648d59 JK |
2668 | ext4_put_io_end(mpd->io_submit.io_end); |
2669 | mpd->io_submit.io_end = NULL; | |
4e7ea81d | 2670 | break; |
61628a3f | 2671 | } |
15648d59 | 2672 | mpd->do_map = 1; |
f63e6005 | 2673 | |
15648d59 JK |
2674 | trace_ext4_da_write_pages(inode, mpd->first_page, wbc); |
2675 | ret = mpage_prepare_extent_to_map(mpd); | |
2676 | if (!ret && mpd->map.m_len) | |
2677 | ret = mpage_map_and_submit_extent(handle, mpd, | |
cb530541 | 2678 | &give_up_on_write); |
646caa9c JK |
2679 | /* |
2680 | * Caution: If the handle is synchronous, | |
2681 | * ext4_journal_stop() can wait for transaction commit | |
2682 | * to finish which may depend on writeback of pages to | |
2683 | * complete or on page lock to be released. In that | |
b483bb77 | 2684 | * case, we have to wait until after we have |
646caa9c JK |
2685 | * submitted all the IO, released page locks we hold, |
2686 | * and dropped io_end reference (for extent conversion | |
2687 | * to be able to complete) before stopping the handle. | |
2688 | */ | |
2689 | if (!ext4_handle_valid(handle) || handle->h_sync == 0) { | |
2690 | ext4_journal_stop(handle); | |
2691 | handle = NULL; | |
15648d59 | 2692 | mpd->do_map = 0; |
646caa9c | 2693 | } |
4e7ea81d | 2694 | /* Unlock pages we didn't use */ |
15648d59 | 2695 | mpage_release_unused_pages(mpd, give_up_on_write); |
a297b2fc | 2696 | /* Submit prepared bio */ |
15648d59 | 2697 | ext4_io_submit(&mpd->io_submit); |
a297b2fc | 2698 | |
646caa9c JK |
2699 | /* |
2700 | * Drop our io_end reference we got from init. We have | |
2701 | * to be careful and use deferred io_end finishing if | |
2702 | * we are still holding the transaction as we can | |
2703 | * release the last reference to io_end which may end | |
2704 | * up doing unwritten extent conversion. | |
2705 | */ | |
2706 | if (handle) { | |
15648d59 | 2707 | ext4_put_io_end_defer(mpd->io_submit.io_end); |
646caa9c JK |
2708 | ext4_journal_stop(handle); |
2709 | } else | |
15648d59 JK |
2710 | ext4_put_io_end(mpd->io_submit.io_end); |
2711 | mpd->io_submit.io_end = NULL; | |
4e7ea81d JK |
2712 | |
2713 | if (ret == -ENOSPC && sbi->s_journal) { | |
2714 | /* | |
2715 | * Commit the transaction which would | |
22208ded AK |
2716 | * free blocks released in the transaction |
2717 | * and try again | |
2718 | */ | |
df22291f | 2719 | jbd2_journal_force_commit_nested(sbi->s_journal); |
22208ded | 2720 | ret = 0; |
4e7ea81d JK |
2721 | continue; |
2722 | } | |
2723 | /* Fatal error - ENOMEM, EIO... */ | |
2724 | if (ret) | |
61628a3f | 2725 | break; |
a1d6cc56 | 2726 | } |
dddbd6ac | 2727 | unplug: |
1bce63d1 | 2728 | blk_finish_plug(&plug); |
9c12a831 | 2729 | if (!ret && !cycled && wbc->nr_to_write > 0) { |
2acf2c26 | 2730 | cycled = 1; |
15648d59 JK |
2731 | mpd->last_page = writeback_index - 1; |
2732 | mpd->first_page = 0; | |
2acf2c26 AK |
2733 | goto retry; |
2734 | } | |
22208ded AK |
2735 | |
2736 | /* Update index */ | |
22208ded AK |
2737 | if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) |
2738 | /* | |
4e7ea81d | 2739 | * Set the writeback_index so that range_cyclic |
22208ded AK |
2740 | * mode will write it back later |
2741 | */ | |
15648d59 | 2742 | mapping->writeback_index = mpd->first_page; |
a1d6cc56 | 2743 | |
61628a3f | 2744 | out_writepages: |
20970ba6 TT |
2745 | trace_ext4_writepages_result(inode, wbc, ret, |
2746 | nr_to_write - wbc->nr_to_write); | |
61628a3f | 2747 | return ret; |
64769240 AT |
2748 | } |
2749 | ||
15648d59 JK |
2750 | static int ext4_writepages(struct address_space *mapping, |
2751 | struct writeback_control *wbc) | |
2752 | { | |
29bc9cea | 2753 | struct super_block *sb = mapping->host->i_sb; |
15648d59 JK |
2754 | struct mpage_da_data mpd = { |
2755 | .inode = mapping->host, | |
2756 | .wbc = wbc, | |
2757 | .can_map = 1, | |
2758 | }; | |
29bc9cea | 2759 | int ret; |
00d873c1 | 2760 | int alloc_ctx; |
29bc9cea JK |
2761 | |
2762 | if (unlikely(ext4_forced_shutdown(EXT4_SB(sb)))) | |
2763 | return -EIO; | |
15648d59 | 2764 | |
00d873c1 | 2765 | alloc_ctx = ext4_writepages_down_read(sb); |
29bc9cea | 2766 | ret = ext4_do_writepages(&mpd); |
1f1a55f0 JK |
2767 | /* |
2768 | * For data=journal writeback we could have come across pages marked | |
2769 | * for delayed dirtying (PageChecked) which were just added to the | |
2770 | * running transaction. Try once more to get them to stable storage. | |
2771 | */ | |
2772 | if (!ret && mpd.journalled_more_data) | |
2773 | ret = ext4_do_writepages(&mpd); | |
00d873c1 | 2774 | ext4_writepages_up_read(sb, alloc_ctx); |
29bc9cea JK |
2775 | |
2776 | return ret; | |
15648d59 JK |
2777 | } |
2778 | ||
59205c8d JK |
2779 | int ext4_normal_submit_inode_data_buffers(struct jbd2_inode *jinode) |
2780 | { | |
2781 | struct writeback_control wbc = { | |
2782 | .sync_mode = WB_SYNC_ALL, | |
2783 | .nr_to_write = LONG_MAX, | |
2784 | .range_start = jinode->i_dirty_start, | |
2785 | .range_end = jinode->i_dirty_end, | |
2786 | }; | |
2787 | struct mpage_da_data mpd = { | |
2788 | .inode = jinode->i_vfs_inode, | |
2789 | .wbc = &wbc, | |
2790 | .can_map = 0, | |
2791 | }; | |
2792 | return ext4_do_writepages(&mpd); | |
2793 | } | |
2794 | ||
5f0663bb DW |
2795 | static int ext4_dax_writepages(struct address_space *mapping, |
2796 | struct writeback_control *wbc) | |
2797 | { | |
2798 | int ret; | |
2799 | long nr_to_write = wbc->nr_to_write; | |
2800 | struct inode *inode = mapping->host; | |
2801 | struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb); | |
00d873c1 | 2802 | int alloc_ctx; |
5f0663bb DW |
2803 | |
2804 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) | |
2805 | return -EIO; | |
2806 | ||
00d873c1 | 2807 | alloc_ctx = ext4_writepages_down_read(inode->i_sb); |
5f0663bb DW |
2808 | trace_ext4_writepages(inode, wbc); |
2809 | ||
3f666c56 | 2810 | ret = dax_writeback_mapping_range(mapping, sbi->s_daxdev, wbc); |
5f0663bb DW |
2811 | trace_ext4_writepages_result(inode, wbc, ret, |
2812 | nr_to_write - wbc->nr_to_write); | |
00d873c1 | 2813 | ext4_writepages_up_read(inode->i_sb, alloc_ctx); |
5f0663bb DW |
2814 | return ret; |
2815 | } | |
2816 | ||
79f0be8d AK |
2817 | static int ext4_nonda_switch(struct super_block *sb) |
2818 | { | |
5c1ff336 | 2819 | s64 free_clusters, dirty_clusters; |
79f0be8d AK |
2820 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
2821 | ||
2822 | /* | |
2823 | * switch to non delalloc mode if we are running low | |
2824 | * on free block. The free block accounting via percpu | |
179f7ebf | 2825 | * counters can get slightly wrong with percpu_counter_batch getting |
79f0be8d AK |
2826 | * accumulated on each CPU without updating global counters |
2827 | * Delalloc need an accurate free block accounting. So switch | |
2828 | * to non delalloc when we are near to error range. | |
2829 | */ | |
5c1ff336 EW |
2830 | free_clusters = |
2831 | percpu_counter_read_positive(&sbi->s_freeclusters_counter); | |
2832 | dirty_clusters = | |
2833 | percpu_counter_read_positive(&sbi->s_dirtyclusters_counter); | |
00d4e736 TT |
2834 | /* |
2835 | * Start pushing delalloc when 1/2 of free blocks are dirty. | |
2836 | */ | |
5c1ff336 | 2837 | if (dirty_clusters && (free_clusters < 2 * dirty_clusters)) |
10ee27a0 | 2838 | try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE); |
00d4e736 | 2839 | |
5c1ff336 EW |
2840 | if (2 * free_clusters < 3 * dirty_clusters || |
2841 | free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) { | |
79f0be8d | 2842 | /* |
c8afb446 ES |
2843 | * free block count is less than 150% of dirty blocks |
2844 | * or free blocks is less than watermark | |
79f0be8d AK |
2845 | */ |
2846 | return 1; | |
2847 | } | |
2848 | return 0; | |
2849 | } | |
2850 | ||
64769240 | 2851 | static int ext4_da_write_begin(struct file *file, struct address_space *mapping, |
9d6b0cd7 | 2852 | loff_t pos, unsigned len, |
de9a55b8 | 2853 | struct page **pagep, void **fsdata) |
64769240 | 2854 | { |
72b8ab9d | 2855 | int ret, retries = 0; |
0b5a2543 | 2856 | struct folio *folio; |
64769240 | 2857 | pgoff_t index; |
64769240 | 2858 | struct inode *inode = mapping->host; |
64769240 | 2859 | |
0db1ff22 TT |
2860 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
2861 | return -EIO; | |
2862 | ||
09cbfeaf | 2863 | index = pos >> PAGE_SHIFT; |
79f0be8d | 2864 | |
6493792d | 2865 | if (ext4_nonda_switch(inode->i_sb) || ext4_verity_in_progress(inode)) { |
79f0be8d AK |
2866 | *fsdata = (void *)FALL_BACK_TO_NONDELALLOC; |
2867 | return ext4_write_begin(file, mapping, pos, | |
9d6b0cd7 | 2868 | len, pagep, fsdata); |
79f0be8d AK |
2869 | } |
2870 | *fsdata = (void *)0; | |
9d6b0cd7 | 2871 | trace_ext4_da_write_begin(inode, pos, len); |
9c3569b5 TM |
2872 | |
2873 | if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { | |
36d116e9 | 2874 | ret = ext4_da_write_inline_data_begin(mapping, inode, pos, len, |
9c3569b5 TM |
2875 | pagep, fsdata); |
2876 | if (ret < 0) | |
47564bfb TT |
2877 | return ret; |
2878 | if (ret == 1) | |
2879 | return 0; | |
9c3569b5 TM |
2880 | } |
2881 | ||
cc883236 | 2882 | retry: |
0b5a2543 MW |
2883 | folio = __filemap_get_folio(mapping, index, FGP_WRITEBEGIN, |
2884 | mapping_gfp_mask(mapping)); | |
7fa8a8ee LT |
2885 | if (IS_ERR(folio)) |
2886 | return PTR_ERR(folio); | |
47564bfb | 2887 | |
0b5a2543 MW |
2888 | /* In case writeback began while the folio was unlocked */ |
2889 | folio_wait_stable(folio); | |
64769240 | 2890 | |
643fa961 | 2891 | #ifdef CONFIG_FS_ENCRYPTION |
86b38c27 | 2892 | ret = ext4_block_write_begin(folio, pos, len, ext4_da_get_block_prep); |
2058f83a | 2893 | #else |
0b5a2543 | 2894 | ret = __block_write_begin(&folio->page, pos, len, ext4_da_get_block_prep); |
2058f83a | 2895 | #endif |
64769240 | 2896 | if (ret < 0) { |
0b5a2543 MW |
2897 | folio_unlock(folio); |
2898 | folio_put(folio); | |
ae4d5372 AK |
2899 | /* |
2900 | * block_write_begin may have instantiated a few blocks | |
2901 | * outside i_size. Trim these off again. Don't need | |
cc883236 | 2902 | * i_size_read because we hold inode lock. |
ae4d5372 AK |
2903 | */ |
2904 | if (pos + len > inode->i_size) | |
b9a4207d | 2905 | ext4_truncate_failed_write(inode); |
47564bfb TT |
2906 | |
2907 | if (ret == -ENOSPC && | |
2908 | ext4_should_retry_alloc(inode->i_sb, &retries)) | |
cc883236 | 2909 | goto retry; |
47564bfb | 2910 | return ret; |
64769240 AT |
2911 | } |
2912 | ||
0b5a2543 | 2913 | *pagep = &folio->page; |
64769240 AT |
2914 | return ret; |
2915 | } | |
2916 | ||
632eaeab MC |
2917 | /* |
2918 | * Check if we should update i_disksize | |
2919 | * when write to the end of file but not require block allocation | |
2920 | */ | |
d19500da | 2921 | static int ext4_da_should_update_i_disksize(struct folio *folio, |
de9a55b8 | 2922 | unsigned long offset) |
632eaeab MC |
2923 | { |
2924 | struct buffer_head *bh; | |
d19500da | 2925 | struct inode *inode = folio->mapping->host; |
632eaeab MC |
2926 | unsigned int idx; |
2927 | int i; | |
2928 | ||
d19500da | 2929 | bh = folio_buffers(folio); |
632eaeab MC |
2930 | idx = offset >> inode->i_blkbits; |
2931 | ||
af5bc92d | 2932 | for (i = 0; i < idx; i++) |
632eaeab MC |
2933 | bh = bh->b_this_page; |
2934 | ||
29fa89d0 | 2935 | if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh)) |
632eaeab MC |
2936 | return 0; |
2937 | return 1; | |
2938 | } | |
2939 | ||
64769240 | 2940 | static int ext4_da_write_end(struct file *file, |
de9a55b8 TT |
2941 | struct address_space *mapping, |
2942 | loff_t pos, unsigned len, unsigned copied, | |
2943 | struct page *page, void *fsdata) | |
64769240 AT |
2944 | { |
2945 | struct inode *inode = mapping->host; | |
64769240 | 2946 | loff_t new_i_size; |
632eaeab | 2947 | unsigned long start, end; |
79f0be8d | 2948 | int write_mode = (int)(unsigned long)fsdata; |
d19500da | 2949 | struct folio *folio = page_folio(page); |
79f0be8d | 2950 | |
74d553aa TT |
2951 | if (write_mode == FALL_BACK_TO_NONDELALLOC) |
2952 | return ext4_write_end(file, mapping, pos, | |
d19500da | 2953 | len, copied, &folio->page, fsdata); |
632eaeab | 2954 | |
9bffad1e | 2955 | trace_ext4_da_write_end(inode, pos, len, copied); |
6984aef5 ZY |
2956 | |
2957 | if (write_mode != CONVERT_INLINE_DATA && | |
2958 | ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) && | |
2959 | ext4_has_inline_data(inode)) | |
d19500da RH |
2960 | return ext4_write_inline_data_end(inode, pos, len, copied, |
2961 | folio); | |
6984aef5 | 2962 | |
1dedde69 ZC |
2963 | if (unlikely(copied < len) && !PageUptodate(page)) |
2964 | copied = 0; | |
2965 | ||
09cbfeaf | 2966 | start = pos & (PAGE_SIZE - 1); |
af5bc92d | 2967 | end = start + copied - 1; |
64769240 AT |
2968 | |
2969 | /* | |
4df031ff ZY |
2970 | * Since we are holding inode lock, we are sure i_disksize <= |
2971 | * i_size. We also know that if i_disksize < i_size, there are | |
2972 | * delalloc writes pending in the range upto i_size. If the end of | |
2973 | * the current write is <= i_size, there's no need to touch | |
2974 | * i_disksize since writeback will push i_disksize upto i_size | |
2975 | * eventually. If the end of the current write is > i_size and | |
2976 | * inside an allocated block (ext4_da_should_update_i_disksize() | |
3f079114 JK |
2977 | * check), we need to update i_disksize here as certain |
2978 | * ext4_writepages() paths not allocating blocks update i_disksize. | |
4df031ff ZY |
2979 | * |
2980 | * Note that we defer inode dirtying to generic_write_end() / | |
2981 | * ext4_da_write_inline_data_end(). | |
64769240 | 2982 | */ |
64769240 | 2983 | new_i_size = pos + copied; |
6984aef5 | 2984 | if (copied && new_i_size > inode->i_size && |
d19500da | 2985 | ext4_da_should_update_i_disksize(folio, end)) |
6984aef5 | 2986 | ext4_update_i_disksize(inode, new_i_size); |
9c3569b5 | 2987 | |
d19500da RH |
2988 | return generic_write_end(file, mapping, pos, len, copied, &folio->page, |
2989 | fsdata); | |
64769240 AT |
2990 | } |
2991 | ||
ccd2506b TT |
2992 | /* |
2993 | * Force all delayed allocation blocks to be allocated for a given inode. | |
2994 | */ | |
2995 | int ext4_alloc_da_blocks(struct inode *inode) | |
2996 | { | |
fb40ba0d TT |
2997 | trace_ext4_alloc_da_blocks(inode); |
2998 | ||
71d4f7d0 | 2999 | if (!EXT4_I(inode)->i_reserved_data_blocks) |
ccd2506b TT |
3000 | return 0; |
3001 | ||
3002 | /* | |
3003 | * We do something simple for now. The filemap_flush() will | |
3004 | * also start triggering a write of the data blocks, which is | |
3005 | * not strictly speaking necessary (and for users of | |
3006 | * laptop_mode, not even desirable). However, to do otherwise | |
3007 | * would require replicating code paths in: | |
de9a55b8 | 3008 | * |
20970ba6 | 3009 | * ext4_writepages() -> |
ccd2506b TT |
3010 | * write_cache_pages() ---> (via passed in callback function) |
3011 | * __mpage_da_writepage() --> | |
3012 | * mpage_add_bh_to_extent() | |
3013 | * mpage_da_map_blocks() | |
3014 | * | |
3015 | * The problem is that write_cache_pages(), located in | |
3016 | * mm/page-writeback.c, marks pages clean in preparation for | |
3017 | * doing I/O, which is not desirable if we're not planning on | |
3018 | * doing I/O at all. | |
3019 | * | |
3020 | * We could call write_cache_pages(), and then redirty all of | |
380cf090 | 3021 | * the pages by calling redirty_page_for_writepage() but that |
ccd2506b TT |
3022 | * would be ugly in the extreme. So instead we would need to |
3023 | * replicate parts of the code in the above functions, | |
25985edc | 3024 | * simplifying them because we wouldn't actually intend to |
ccd2506b TT |
3025 | * write out the pages, but rather only collect contiguous |
3026 | * logical block extents, call the multi-block allocator, and | |
3027 | * then update the buffer heads with the block allocations. | |
de9a55b8 | 3028 | * |
ccd2506b TT |
3029 | * For now, though, we'll cheat by calling filemap_flush(), |
3030 | * which will map the blocks, and start the I/O, but not | |
3031 | * actually wait for the I/O to complete. | |
3032 | */ | |
3033 | return filemap_flush(inode->i_mapping); | |
3034 | } | |
64769240 | 3035 | |
ac27a0ec DK |
3036 | /* |
3037 | * bmap() is special. It gets used by applications such as lilo and by | |
3038 | * the swapper to find the on-disk block of a specific piece of data. | |
3039 | * | |
3040 | * Naturally, this is dangerous if the block concerned is still in the | |
617ba13b | 3041 | * journal. If somebody makes a swapfile on an ext4 data-journaling |
ac27a0ec DK |
3042 | * filesystem and enables swap, then they may get a nasty shock when the |
3043 | * data getting swapped to that swapfile suddenly gets overwritten by | |
3044 | * the original zero's written out previously to the journal and | |
3045 | * awaiting writeback in the kernel's buffer cache. | |
3046 | * | |
3047 | * So, if we see any bmap calls here on a modified, data-journaled file, | |
3048 | * take extra steps to flush any blocks which might be in the cache. | |
3049 | */ | |
617ba13b | 3050 | static sector_t ext4_bmap(struct address_space *mapping, sector_t block) |
ac27a0ec DK |
3051 | { |
3052 | struct inode *inode = mapping->host; | |
51ae846c | 3053 | sector_t ret = 0; |
ac27a0ec | 3054 | |
51ae846c | 3055 | inode_lock_shared(inode); |
46c7f254 TM |
3056 | /* |
3057 | * We can get here for an inline file via the FIBMAP ioctl | |
3058 | */ | |
3059 | if (ext4_has_inline_data(inode)) | |
51ae846c | 3060 | goto out; |
46c7f254 | 3061 | |
64769240 | 3062 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) && |
951cafa6 JK |
3063 | (test_opt(inode->i_sb, DELALLOC) || |
3064 | ext4_should_journal_data(inode))) { | |
64769240 | 3065 | /* |
951cafa6 JK |
3066 | * With delalloc or journalled data we want to sync the file so |
3067 | * that we can make sure we allocate blocks for file and data | |
3068 | * is in place for the user to see it | |
64769240 AT |
3069 | */ |
3070 | filemap_write_and_wait(mapping); | |
3071 | } | |
3072 | ||
51ae846c YB |
3073 | ret = iomap_bmap(mapping, block, &ext4_iomap_ops); |
3074 | ||
3075 | out: | |
3076 | inode_unlock_shared(inode); | |
3077 | return ret; | |
ac27a0ec DK |
3078 | } |
3079 | ||
fe5ddf6b | 3080 | static int ext4_read_folio(struct file *file, struct folio *folio) |
ac27a0ec | 3081 | { |
46c7f254 | 3082 | int ret = -EAGAIN; |
c0be8e6f | 3083 | struct inode *inode = folio->mapping->host; |
46c7f254 | 3084 | |
36c9b450 | 3085 | trace_ext4_read_folio(inode, folio); |
46c7f254 TM |
3086 | |
3087 | if (ext4_has_inline_data(inode)) | |
3edde93e | 3088 | ret = ext4_readpage_inline(inode, folio); |
46c7f254 TM |
3089 | |
3090 | if (ret == -EAGAIN) | |
c0be8e6f | 3091 | return ext4_mpage_readpages(inode, NULL, folio); |
46c7f254 TM |
3092 | |
3093 | return ret; | |
ac27a0ec DK |
3094 | } |
3095 | ||
6311f91f | 3096 | static void ext4_readahead(struct readahead_control *rac) |
ac27a0ec | 3097 | { |
6311f91f | 3098 | struct inode *inode = rac->mapping->host; |
46c7f254 | 3099 | |
6311f91f | 3100 | /* If the file has inline data, no need to do readahead. */ |
46c7f254 | 3101 | if (ext4_has_inline_data(inode)) |
6311f91f | 3102 | return; |
46c7f254 | 3103 | |
a07f624b | 3104 | ext4_mpage_readpages(inode, rac, NULL); |
ac27a0ec DK |
3105 | } |
3106 | ||
7ba13abb MWO |
3107 | static void ext4_invalidate_folio(struct folio *folio, size_t offset, |
3108 | size_t length) | |
ac27a0ec | 3109 | { |
ccd16945 | 3110 | trace_ext4_invalidate_folio(folio, offset, length); |
0562e0ba | 3111 | |
4520fb3c | 3112 | /* No journalling happens on data buffers when this function is used */ |
7ba13abb | 3113 | WARN_ON(folio_buffers(folio) && buffer_jbd(folio_buffers(folio))); |
4520fb3c | 3114 | |
7ba13abb | 3115 | block_invalidate_folio(folio, offset, length); |
4520fb3c JK |
3116 | } |
3117 | ||
ccd16945 MWO |
3118 | static int __ext4_journalled_invalidate_folio(struct folio *folio, |
3119 | size_t offset, size_t length) | |
4520fb3c | 3120 | { |
ccd16945 | 3121 | journal_t *journal = EXT4_JOURNAL(folio->mapping->host); |
4520fb3c | 3122 | |
ccd16945 | 3123 | trace_ext4_journalled_invalidate_folio(folio, offset, length); |
4520fb3c | 3124 | |
ac27a0ec DK |
3125 | /* |
3126 | * If it's a full truncate we just forget about the pending dirtying | |
3127 | */ | |
ccd16945 MWO |
3128 | if (offset == 0 && length == folio_size(folio)) |
3129 | folio_clear_checked(folio); | |
ac27a0ec | 3130 | |
ccd16945 | 3131 | return jbd2_journal_invalidate_folio(journal, folio, offset, length); |
53e87268 JK |
3132 | } |
3133 | ||
3134 | /* Wrapper for aops... */ | |
ccd16945 MWO |
3135 | static void ext4_journalled_invalidate_folio(struct folio *folio, |
3136 | size_t offset, | |
3137 | size_t length) | |
53e87268 | 3138 | { |
ccd16945 | 3139 | WARN_ON(__ext4_journalled_invalidate_folio(folio, offset, length) < 0); |
ac27a0ec DK |
3140 | } |
3141 | ||
3c402f15 | 3142 | static bool ext4_release_folio(struct folio *folio, gfp_t wait) |
ac27a0ec | 3143 | { |
36c9b450 RH |
3144 | struct inode *inode = folio->mapping->host; |
3145 | journal_t *journal = EXT4_JOURNAL(inode); | |
ac27a0ec | 3146 | |
36c9b450 | 3147 | trace_ext4_release_folio(inode, folio); |
0562e0ba | 3148 | |
e1c36595 | 3149 | /* Page has dirty journalled data -> cannot release */ |
3c402f15 MWO |
3150 | if (folio_test_checked(folio)) |
3151 | return false; | |
0390131b | 3152 | if (journal) |
c56a6eb0 | 3153 | return jbd2_journal_try_to_free_buffers(journal, folio); |
0390131b | 3154 | else |
68189fef | 3155 | return try_to_free_buffers(folio); |
ac27a0ec DK |
3156 | } |
3157 | ||
b8a6176c JK |
3158 | static bool ext4_inode_datasync_dirty(struct inode *inode) |
3159 | { | |
3160 | journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; | |
3161 | ||
aa75f4d3 HS |
3162 | if (journal) { |
3163 | if (jbd2_transaction_committed(journal, | |
d0520df7 AR |
3164 | EXT4_I(inode)->i_datasync_tid)) |
3165 | return false; | |
3166 | if (test_opt2(inode->i_sb, JOURNAL_FAST_COMMIT)) | |
1ceecb53 | 3167 | return !list_empty(&EXT4_I(inode)->i_fc_list); |
d0520df7 | 3168 | return true; |
aa75f4d3 HS |
3169 | } |
3170 | ||
b8a6176c JK |
3171 | /* Any metadata buffers to write? */ |
3172 | if (!list_empty(&inode->i_mapping->private_list)) | |
3173 | return true; | |
3174 | return inode->i_state & I_DIRTY_DATASYNC; | |
3175 | } | |
3176 | ||
c8fdfe29 MB |
3177 | static void ext4_set_iomap(struct inode *inode, struct iomap *iomap, |
3178 | struct ext4_map_blocks *map, loff_t offset, | |
de205114 | 3179 | loff_t length, unsigned int flags) |
364443cb | 3180 | { |
c8fdfe29 | 3181 | u8 blkbits = inode->i_blkbits; |
364443cb | 3182 | |
c8fdfe29 MB |
3183 | /* |
3184 | * Writes that span EOF might trigger an I/O size update on completion, | |
3185 | * so consider them to be dirty for the purpose of O_DSYNC, even if | |
3186 | * there is no other metadata changes being made or are pending. | |
3187 | */ | |
364443cb | 3188 | iomap->flags = 0; |
c8fdfe29 MB |
3189 | if (ext4_inode_datasync_dirty(inode) || |
3190 | offset + length > i_size_read(inode)) | |
b8a6176c | 3191 | iomap->flags |= IOMAP_F_DIRTY; |
c8fdfe29 MB |
3192 | |
3193 | if (map->m_flags & EXT4_MAP_NEW) | |
3194 | iomap->flags |= IOMAP_F_NEW; | |
3195 | ||
de205114 CH |
3196 | if (flags & IOMAP_DAX) |
3197 | iomap->dax_dev = EXT4_SB(inode->i_sb)->s_daxdev; | |
3198 | else | |
3199 | iomap->bdev = inode->i_sb->s_bdev; | |
c8fdfe29 MB |
3200 | iomap->offset = (u64) map->m_lblk << blkbits; |
3201 | iomap->length = (u64) map->m_len << blkbits; | |
364443cb | 3202 | |
6386722a RH |
3203 | if ((map->m_flags & EXT4_MAP_MAPPED) && |
3204 | !ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
3205 | iomap->flags |= IOMAP_F_MERGED; | |
3206 | ||
c8fdfe29 MB |
3207 | /* |
3208 | * Flags passed to ext4_map_blocks() for direct I/O writes can result | |
3209 | * in m_flags having both EXT4_MAP_MAPPED and EXT4_MAP_UNWRITTEN bits | |
3210 | * set. In order for any allocated unwritten extents to be converted | |
3211 | * into written extents correctly within the ->end_io() handler, we | |
3212 | * need to ensure that the iomap->type is set appropriately. Hence, the | |
3213 | * reason why we need to check whether the EXT4_MAP_UNWRITTEN bit has | |
3214 | * been set first. | |
3215 | */ | |
3216 | if (map->m_flags & EXT4_MAP_UNWRITTEN) { | |
3217 | iomap->type = IOMAP_UNWRITTEN; | |
3218 | iomap->addr = (u64) map->m_pblk << blkbits; | |
de205114 CH |
3219 | if (flags & IOMAP_DAX) |
3220 | iomap->addr += EXT4_SB(inode->i_sb)->s_dax_part_off; | |
c8fdfe29 MB |
3221 | } else if (map->m_flags & EXT4_MAP_MAPPED) { |
3222 | iomap->type = IOMAP_MAPPED; | |
3223 | iomap->addr = (u64) map->m_pblk << blkbits; | |
de205114 CH |
3224 | if (flags & IOMAP_DAX) |
3225 | iomap->addr += EXT4_SB(inode->i_sb)->s_dax_part_off; | |
364443cb | 3226 | } else { |
c8fdfe29 MB |
3227 | iomap->type = IOMAP_HOLE; |
3228 | iomap->addr = IOMAP_NULL_ADDR; | |
364443cb | 3229 | } |
364443cb JK |
3230 | } |
3231 | ||
f063db5e MB |
3232 | static int ext4_iomap_alloc(struct inode *inode, struct ext4_map_blocks *map, |
3233 | unsigned int flags) | |
776722e8 | 3234 | { |
776722e8 | 3235 | handle_t *handle; |
378f32ba MB |
3236 | u8 blkbits = inode->i_blkbits; |
3237 | int ret, dio_credits, m_flags = 0, retries = 0; | |
776722e8 | 3238 | |
776722e8 | 3239 | /* |
f063db5e MB |
3240 | * Trim the mapping request to the maximum value that we can map at |
3241 | * once for direct I/O. | |
776722e8 | 3242 | */ |
f063db5e MB |
3243 | if (map->m_len > DIO_MAX_BLOCKS) |
3244 | map->m_len = DIO_MAX_BLOCKS; | |
3245 | dio_credits = ext4_chunk_trans_blocks(inode, map->m_len); | |
776722e8 | 3246 | |
f063db5e | 3247 | retry: |
776722e8 | 3248 | /* |
f063db5e MB |
3249 | * Either we allocate blocks and then don't get an unwritten extent, so |
3250 | * in that case we have reserved enough credits. Or, the blocks are | |
3251 | * already allocated and unwritten. In that case, the extent conversion | |
3252 | * fits into the credits as well. | |
776722e8 | 3253 | */ |
f063db5e MB |
3254 | handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits); |
3255 | if (IS_ERR(handle)) | |
3256 | return PTR_ERR(handle); | |
4c0425ff | 3257 | |
378f32ba MB |
3258 | /* |
3259 | * DAX and direct I/O are the only two operations that are currently | |
3260 | * supported with IOMAP_WRITE. | |
3261 | */ | |
952da063 CH |
3262 | WARN_ON(!(flags & (IOMAP_DAX | IOMAP_DIRECT))); |
3263 | if (flags & IOMAP_DAX) | |
378f32ba MB |
3264 | m_flags = EXT4_GET_BLOCKS_CREATE_ZERO; |
3265 | /* | |
3266 | * We use i_size instead of i_disksize here because delalloc writeback | |
3267 | * can complete at any point during the I/O and subsequently push the | |
3268 | * i_disksize out to i_size. This could be beyond where direct I/O is | |
3269 | * happening and thus expose allocated blocks to direct I/O reads. | |
3270 | */ | |
d0b040f5 | 3271 | else if (((loff_t)map->m_lblk << blkbits) >= i_size_read(inode)) |
378f32ba MB |
3272 | m_flags = EXT4_GET_BLOCKS_CREATE; |
3273 | else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
3274 | m_flags = EXT4_GET_BLOCKS_IO_CREATE_EXT; | |
4b70df18 | 3275 | |
378f32ba | 3276 | ret = ext4_map_blocks(handle, inode, map, m_flags); |
8d5d02e6 | 3277 | |
74c66bcb | 3278 | /* |
378f32ba MB |
3279 | * We cannot fill holes in indirect tree based inodes as that could |
3280 | * expose stale data in the case of a crash. Use the magic error code | |
3281 | * to fallback to buffered I/O. | |
74c66bcb | 3282 | */ |
378f32ba MB |
3283 | if (!m_flags && !ret) |
3284 | ret = -ENOTBLK; | |
187372a3 | 3285 | |
f063db5e MB |
3286 | ext4_journal_stop(handle); |
3287 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) | |
3288 | goto retry; | |
3289 | ||
3290 | return ret; | |
4c0425ff | 3291 | } |
c7064ef1 | 3292 | |
f063db5e | 3293 | |
364443cb | 3294 | static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length, |
c039b997 | 3295 | unsigned flags, struct iomap *iomap, struct iomap *srcmap) |
4c0425ff | 3296 | { |
364443cb | 3297 | int ret; |
09edf4d3 MB |
3298 | struct ext4_map_blocks map; |
3299 | u8 blkbits = inode->i_blkbits; | |
729f52c6 | 3300 | |
bcd8e91f TT |
3301 | if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK) |
3302 | return -EINVAL; | |
4bd809db | 3303 | |
09edf4d3 MB |
3304 | if (WARN_ON_ONCE(ext4_has_inline_data(inode))) |
3305 | return -ERANGE; | |
4bd809db | 3306 | |
e8340395 | 3307 | /* |
09edf4d3 | 3308 | * Calculate the first and last logical blocks respectively. |
e8340395 | 3309 | */ |
09edf4d3 MB |
3310 | map.m_lblk = offset >> blkbits; |
3311 | map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits, | |
3312 | EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1; | |
e8340395 | 3313 | |
9faac62d RH |
3314 | if (flags & IOMAP_WRITE) { |
3315 | /* | |
3316 | * We check here if the blocks are already allocated, then we | |
3317 | * don't need to start a journal txn and we can directly return | |
3318 | * the mapping information. This could boost performance | |
3319 | * especially in multi-threaded overwrite requests. | |
3320 | */ | |
3321 | if (offset + length <= i_size_read(inode)) { | |
3322 | ret = ext4_map_blocks(NULL, inode, &map, 0); | |
3323 | if (ret > 0 && (map.m_flags & EXT4_MAP_MAPPED)) | |
3324 | goto out; | |
3325 | } | |
f063db5e | 3326 | ret = ext4_iomap_alloc(inode, &map, flags); |
9faac62d | 3327 | } else { |
545052e9 | 3328 | ret = ext4_map_blocks(NULL, inode, &map, 0); |
9faac62d | 3329 | } |
4bd809db | 3330 | |
f063db5e MB |
3331 | if (ret < 0) |
3332 | return ret; | |
9faac62d | 3333 | out: |
38ea50da EB |
3334 | /* |
3335 | * When inline encryption is enabled, sometimes I/O to an encrypted file | |
3336 | * has to be broken up to guarantee DUN contiguity. Handle this by | |
3337 | * limiting the length of the mapping returned. | |
3338 | */ | |
3339 | map.m_len = fscrypt_limit_io_blocks(inode, map.m_lblk, map.m_len); | |
3340 | ||
de205114 | 3341 | ext4_set_iomap(inode, iomap, &map, offset, length, flags); |
4bd809db | 3342 | |
364443cb JK |
3343 | return 0; |
3344 | } | |
8d5d02e6 | 3345 | |
8cd115bd JK |
3346 | static int ext4_iomap_overwrite_begin(struct inode *inode, loff_t offset, |
3347 | loff_t length, unsigned flags, struct iomap *iomap, | |
3348 | struct iomap *srcmap) | |
3349 | { | |
3350 | int ret; | |
3351 | ||
3352 | /* | |
3353 | * Even for writes we don't need to allocate blocks, so just pretend | |
3354 | * we are reading to save overhead of starting a transaction. | |
3355 | */ | |
3356 | flags &= ~IOMAP_WRITE; | |
3357 | ret = ext4_iomap_begin(inode, offset, length, flags, iomap, srcmap); | |
fa83c34e | 3358 | WARN_ON_ONCE(!ret && iomap->type != IOMAP_MAPPED); |
8cd115bd JK |
3359 | return ret; |
3360 | } | |
3361 | ||
776722e8 JK |
3362 | static int ext4_iomap_end(struct inode *inode, loff_t offset, loff_t length, |
3363 | ssize_t written, unsigned flags, struct iomap *iomap) | |
3364 | { | |
69c499d1 | 3365 | /* |
378f32ba MB |
3366 | * Check to see whether an error occurred while writing out the data to |
3367 | * the allocated blocks. If so, return the magic error code so that we | |
3368 | * fallback to buffered I/O and attempt to complete the remainder of | |
3369 | * the I/O. Any blocks that may have been allocated in preparation for | |
3370 | * the direct I/O will be reused during buffered I/O. | |
69c499d1 | 3371 | */ |
378f32ba MB |
3372 | if (flags & (IOMAP_WRITE | IOMAP_DIRECT) && written == 0) |
3373 | return -ENOTBLK; | |
69c499d1 | 3374 | |
569342dc | 3375 | return 0; |
776722e8 | 3376 | } |
4bd809db | 3377 | |
8ff6daa1 | 3378 | const struct iomap_ops ext4_iomap_ops = { |
364443cb | 3379 | .iomap_begin = ext4_iomap_begin, |
776722e8 | 3380 | .iomap_end = ext4_iomap_end, |
364443cb | 3381 | }; |
8d5d02e6 | 3382 | |
8cd115bd JK |
3383 | const struct iomap_ops ext4_iomap_overwrite_ops = { |
3384 | .iomap_begin = ext4_iomap_overwrite_begin, | |
3385 | .iomap_end = ext4_iomap_end, | |
3386 | }; | |
3387 | ||
09edf4d3 MB |
3388 | static bool ext4_iomap_is_delalloc(struct inode *inode, |
3389 | struct ext4_map_blocks *map) | |
3390 | { | |
3391 | struct extent_status es; | |
3392 | ext4_lblk_t offset = 0, end = map->m_lblk + map->m_len - 1; | |
914f82a3 | 3393 | |
09edf4d3 MB |
3394 | ext4_es_find_extent_range(inode, &ext4_es_is_delayed, |
3395 | map->m_lblk, end, &es); | |
914f82a3 | 3396 | |
09edf4d3 MB |
3397 | if (!es.es_len || es.es_lblk > end) |
3398 | return false; | |
914f82a3 | 3399 | |
09edf4d3 MB |
3400 | if (es.es_lblk > map->m_lblk) { |
3401 | map->m_len = es.es_lblk - map->m_lblk; | |
3402 | return false; | |
914f82a3 | 3403 | } |
914f82a3 | 3404 | |
09edf4d3 MB |
3405 | offset = map->m_lblk - es.es_lblk; |
3406 | map->m_len = es.es_len - offset; | |
914f82a3 | 3407 | |
09edf4d3 | 3408 | return true; |
4c0425ff MC |
3409 | } |
3410 | ||
09edf4d3 MB |
3411 | static int ext4_iomap_begin_report(struct inode *inode, loff_t offset, |
3412 | loff_t length, unsigned int flags, | |
3413 | struct iomap *iomap, struct iomap *srcmap) | |
4c0425ff | 3414 | { |
09edf4d3 MB |
3415 | int ret; |
3416 | bool delalloc = false; | |
3417 | struct ext4_map_blocks map; | |
3418 | u8 blkbits = inode->i_blkbits; | |
4c0425ff | 3419 | |
09edf4d3 MB |
3420 | if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK) |
3421 | return -EINVAL; | |
3422 | ||
3423 | if (ext4_has_inline_data(inode)) { | |
3424 | ret = ext4_inline_data_iomap(inode, iomap); | |
3425 | if (ret != -EAGAIN) { | |
3426 | if (ret == 0 && offset >= iomap->length) | |
3427 | ret = -ENOENT; | |
3428 | return ret; | |
3429 | } | |
3430 | } | |
2058f83a | 3431 | |
84ebd795 | 3432 | /* |
09edf4d3 | 3433 | * Calculate the first and last logical block respectively. |
84ebd795 | 3434 | */ |
09edf4d3 MB |
3435 | map.m_lblk = offset >> blkbits; |
3436 | map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits, | |
3437 | EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1; | |
84ebd795 | 3438 | |
b2c57642 RH |
3439 | /* |
3440 | * Fiemap callers may call for offset beyond s_bitmap_maxbytes. | |
3441 | * So handle it here itself instead of querying ext4_map_blocks(). | |
3442 | * Since ext4_map_blocks() will warn about it and will return | |
3443 | * -EIO error. | |
3444 | */ | |
3445 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) { | |
3446 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
3447 | ||
3448 | if (offset >= sbi->s_bitmap_maxbytes) { | |
3449 | map.m_flags = 0; | |
3450 | goto set_iomap; | |
3451 | } | |
3452 | } | |
3453 | ||
09edf4d3 MB |
3454 | ret = ext4_map_blocks(NULL, inode, &map, 0); |
3455 | if (ret < 0) | |
3456 | return ret; | |
3457 | if (ret == 0) | |
3458 | delalloc = ext4_iomap_is_delalloc(inode, &map); | |
46c7f254 | 3459 | |
b2c57642 | 3460 | set_iomap: |
de205114 | 3461 | ext4_set_iomap(inode, iomap, &map, offset, length, flags); |
09edf4d3 MB |
3462 | if (delalloc && iomap->type == IOMAP_HOLE) |
3463 | iomap->type = IOMAP_DELALLOC; | |
3464 | ||
3465 | return 0; | |
4c0425ff MC |
3466 | } |
3467 | ||
09edf4d3 MB |
3468 | const struct iomap_ops ext4_iomap_report_ops = { |
3469 | .iomap_begin = ext4_iomap_begin_report, | |
3470 | }; | |
3471 | ||
ac27a0ec | 3472 | /* |
3f5d3063 JK |
3473 | * For data=journal mode, folio should be marked dirty only when it was |
3474 | * writeably mapped. When that happens, it was already attached to the | |
3475 | * transaction and marked as jbddirty (we take care of this in | |
3476 | * ext4_page_mkwrite()). On transaction commit, we writeprotect page mappings | |
3477 | * so we should have nothing to do here, except for the case when someone | |
3478 | * had the page pinned and dirtied the page through this pin (e.g. by doing | |
3479 | * direct IO to it). In that case we'd need to attach buffers here to the | |
3480 | * transaction but we cannot due to lock ordering. We cannot just dirty the | |
3481 | * folio and leave attached buffers clean, because the buffers' dirty state is | |
3482 | * "definitive". We cannot just set the buffers dirty or jbddirty because all | |
3483 | * the journalling code will explode. So what we do is to mark the folio | |
3484 | * "pending dirty" and next time ext4_writepages() is called, attach buffers | |
3485 | * to the transaction appropriately. | |
ac27a0ec | 3486 | */ |
187c82cb MWO |
3487 | static bool ext4_journalled_dirty_folio(struct address_space *mapping, |
3488 | struct folio *folio) | |
ac27a0ec | 3489 | { |
0f252336 | 3490 | WARN_ON_ONCE(!folio_buffers(folio)); |
3f5d3063 JK |
3491 | if (folio_maybe_dma_pinned(folio)) |
3492 | folio_set_checked(folio); | |
187c82cb | 3493 | return filemap_dirty_folio(mapping, folio); |
ac27a0ec DK |
3494 | } |
3495 | ||
e621900a | 3496 | static bool ext4_dirty_folio(struct address_space *mapping, struct folio *folio) |
6dcc693b | 3497 | { |
e621900a MWO |
3498 | WARN_ON_ONCE(!folio_test_locked(folio) && !folio_test_dirty(folio)); |
3499 | WARN_ON_ONCE(!folio_buffers(folio)); | |
3500 | return block_dirty_folio(mapping, folio); | |
6dcc693b JK |
3501 | } |
3502 | ||
0e6895ba RH |
3503 | static int ext4_iomap_swap_activate(struct swap_info_struct *sis, |
3504 | struct file *file, sector_t *span) | |
3505 | { | |
3506 | return iomap_swapfile_activate(sis, file, span, | |
3507 | &ext4_iomap_report_ops); | |
3508 | } | |
3509 | ||
74d553aa | 3510 | static const struct address_space_operations ext4_aops = { |
fe5ddf6b | 3511 | .read_folio = ext4_read_folio, |
6311f91f | 3512 | .readahead = ext4_readahead, |
20970ba6 | 3513 | .writepages = ext4_writepages, |
8ab22b9a | 3514 | .write_begin = ext4_write_begin, |
74d553aa | 3515 | .write_end = ext4_write_end, |
e621900a | 3516 | .dirty_folio = ext4_dirty_folio, |
8ab22b9a | 3517 | .bmap = ext4_bmap, |
7ba13abb | 3518 | .invalidate_folio = ext4_invalidate_folio, |
3c402f15 | 3519 | .release_folio = ext4_release_folio, |
378f32ba | 3520 | .direct_IO = noop_direct_IO, |
67235182 | 3521 | .migrate_folio = buffer_migrate_folio, |
8ab22b9a | 3522 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 3523 | .error_remove_page = generic_error_remove_page, |
0e6895ba | 3524 | .swap_activate = ext4_iomap_swap_activate, |
ac27a0ec DK |
3525 | }; |
3526 | ||
617ba13b | 3527 | static const struct address_space_operations ext4_journalled_aops = { |
fe5ddf6b | 3528 | .read_folio = ext4_read_folio, |
6311f91f | 3529 | .readahead = ext4_readahead, |
20970ba6 | 3530 | .writepages = ext4_writepages, |
8ab22b9a HH |
3531 | .write_begin = ext4_write_begin, |
3532 | .write_end = ext4_journalled_write_end, | |
187c82cb | 3533 | .dirty_folio = ext4_journalled_dirty_folio, |
8ab22b9a | 3534 | .bmap = ext4_bmap, |
ccd16945 | 3535 | .invalidate_folio = ext4_journalled_invalidate_folio, |
3c402f15 | 3536 | .release_folio = ext4_release_folio, |
378f32ba | 3537 | .direct_IO = noop_direct_IO, |
dae99960 | 3538 | .migrate_folio = buffer_migrate_folio_norefs, |
8ab22b9a | 3539 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 3540 | .error_remove_page = generic_error_remove_page, |
0e6895ba | 3541 | .swap_activate = ext4_iomap_swap_activate, |
ac27a0ec DK |
3542 | }; |
3543 | ||
64769240 | 3544 | static const struct address_space_operations ext4_da_aops = { |
fe5ddf6b | 3545 | .read_folio = ext4_read_folio, |
6311f91f | 3546 | .readahead = ext4_readahead, |
20970ba6 | 3547 | .writepages = ext4_writepages, |
8ab22b9a HH |
3548 | .write_begin = ext4_da_write_begin, |
3549 | .write_end = ext4_da_write_end, | |
e621900a | 3550 | .dirty_folio = ext4_dirty_folio, |
8ab22b9a | 3551 | .bmap = ext4_bmap, |
7ba13abb | 3552 | .invalidate_folio = ext4_invalidate_folio, |
3c402f15 | 3553 | .release_folio = ext4_release_folio, |
378f32ba | 3554 | .direct_IO = noop_direct_IO, |
67235182 | 3555 | .migrate_folio = buffer_migrate_folio, |
8ab22b9a | 3556 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 3557 | .error_remove_page = generic_error_remove_page, |
0e6895ba | 3558 | .swap_activate = ext4_iomap_swap_activate, |
64769240 AT |
3559 | }; |
3560 | ||
5f0663bb DW |
3561 | static const struct address_space_operations ext4_dax_aops = { |
3562 | .writepages = ext4_dax_writepages, | |
3563 | .direct_IO = noop_direct_IO, | |
46de8b97 | 3564 | .dirty_folio = noop_dirty_folio, |
94dbb631 | 3565 | .bmap = ext4_bmap, |
0e6895ba | 3566 | .swap_activate = ext4_iomap_swap_activate, |
5f0663bb DW |
3567 | }; |
3568 | ||
617ba13b | 3569 | void ext4_set_aops(struct inode *inode) |
ac27a0ec | 3570 | { |
3d2b1582 LC |
3571 | switch (ext4_inode_journal_mode(inode)) { |
3572 | case EXT4_INODE_ORDERED_DATA_MODE: | |
3d2b1582 | 3573 | case EXT4_INODE_WRITEBACK_DATA_MODE: |
3d2b1582 LC |
3574 | break; |
3575 | case EXT4_INODE_JOURNAL_DATA_MODE: | |
617ba13b | 3576 | inode->i_mapping->a_ops = &ext4_journalled_aops; |
74d553aa | 3577 | return; |
3d2b1582 LC |
3578 | default: |
3579 | BUG(); | |
3580 | } | |
5f0663bb DW |
3581 | if (IS_DAX(inode)) |
3582 | inode->i_mapping->a_ops = &ext4_dax_aops; | |
3583 | else if (test_opt(inode->i_sb, DELALLOC)) | |
74d553aa TT |
3584 | inode->i_mapping->a_ops = &ext4_da_aops; |
3585 | else | |
3586 | inode->i_mapping->a_ops = &ext4_aops; | |
ac27a0ec DK |
3587 | } |
3588 | ||
923ae0ff | 3589 | static int __ext4_block_zero_page_range(handle_t *handle, |
d863dc36 LC |
3590 | struct address_space *mapping, loff_t from, loff_t length) |
3591 | { | |
09cbfeaf KS |
3592 | ext4_fsblk_t index = from >> PAGE_SHIFT; |
3593 | unsigned offset = from & (PAGE_SIZE-1); | |
923ae0ff | 3594 | unsigned blocksize, pos; |
d863dc36 LC |
3595 | ext4_lblk_t iblock; |
3596 | struct inode *inode = mapping->host; | |
3597 | struct buffer_head *bh; | |
9d3973de | 3598 | struct folio *folio; |
d863dc36 LC |
3599 | int err = 0; |
3600 | ||
9d3973de MW |
3601 | folio = __filemap_get_folio(mapping, from >> PAGE_SHIFT, |
3602 | FGP_LOCK | FGP_ACCESSED | FGP_CREAT, | |
3603 | mapping_gfp_constraint(mapping, ~__GFP_FS)); | |
7fa8a8ee LT |
3604 | if (IS_ERR(folio)) |
3605 | return PTR_ERR(folio); | |
d863dc36 LC |
3606 | |
3607 | blocksize = inode->i_sb->s_blocksize; | |
d863dc36 | 3608 | |
09cbfeaf | 3609 | iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits); |
d863dc36 | 3610 | |
9d3973de MW |
3611 | bh = folio_buffers(folio); |
3612 | if (!bh) { | |
3613 | create_empty_buffers(&folio->page, blocksize, 0); | |
3614 | bh = folio_buffers(folio); | |
3615 | } | |
d863dc36 LC |
3616 | |
3617 | /* Find the buffer that contains "offset" */ | |
d863dc36 LC |
3618 | pos = blocksize; |
3619 | while (offset >= pos) { | |
3620 | bh = bh->b_this_page; | |
3621 | iblock++; | |
3622 | pos += blocksize; | |
3623 | } | |
d863dc36 LC |
3624 | if (buffer_freed(bh)) { |
3625 | BUFFER_TRACE(bh, "freed: skip"); | |
3626 | goto unlock; | |
3627 | } | |
d863dc36 LC |
3628 | if (!buffer_mapped(bh)) { |
3629 | BUFFER_TRACE(bh, "unmapped"); | |
3630 | ext4_get_block(inode, iblock, bh, 0); | |
3631 | /* unmapped? It's a hole - nothing to do */ | |
3632 | if (!buffer_mapped(bh)) { | |
3633 | BUFFER_TRACE(bh, "still unmapped"); | |
3634 | goto unlock; | |
3635 | } | |
3636 | } | |
3637 | ||
3638 | /* Ok, it's mapped. Make sure it's up-to-date */ | |
9d3973de | 3639 | if (folio_test_uptodate(folio)) |
d863dc36 LC |
3640 | set_buffer_uptodate(bh); |
3641 | ||
3642 | if (!buffer_uptodate(bh)) { | |
2d069c08 | 3643 | err = ext4_read_bh_lock(bh, 0, true); |
3644 | if (err) | |
d863dc36 | 3645 | goto unlock; |
4f74d15f | 3646 | if (fscrypt_inode_uses_fs_layer_crypto(inode)) { |
c9c7429c | 3647 | /* We expect the key to be set. */ |
a7550b30 | 3648 | BUG_ON(!fscrypt_has_encryption_key(inode)); |
9d3973de | 3649 | err = fscrypt_decrypt_pagecache_blocks(folio, |
51e4e315 | 3650 | blocksize, |
834f1565 EB |
3651 | bh_offset(bh)); |
3652 | if (err) { | |
3653 | clear_buffer_uptodate(bh); | |
3654 | goto unlock; | |
3655 | } | |
c9c7429c | 3656 | } |
d863dc36 | 3657 | } |
d863dc36 LC |
3658 | if (ext4_should_journal_data(inode)) { |
3659 | BUFFER_TRACE(bh, "get write access"); | |
188c299e JK |
3660 | err = ext4_journal_get_write_access(handle, inode->i_sb, bh, |
3661 | EXT4_JTR_NONE); | |
d863dc36 LC |
3662 | if (err) |
3663 | goto unlock; | |
3664 | } | |
9d3973de | 3665 | folio_zero_range(folio, offset, length); |
d863dc36 LC |
3666 | BUFFER_TRACE(bh, "zeroed end of block"); |
3667 | ||
d863dc36 | 3668 | if (ext4_should_journal_data(inode)) { |
d84c9ebd | 3669 | err = ext4_dirty_journalled_data(handle, bh); |
0713ed0c | 3670 | } else { |
353eefd3 | 3671 | err = 0; |
d863dc36 | 3672 | mark_buffer_dirty(bh); |
3957ef53 | 3673 | if (ext4_should_order_data(inode)) |
73131fbb RZ |
3674 | err = ext4_jbd2_inode_add_write(handle, inode, from, |
3675 | length); | |
0713ed0c | 3676 | } |
d863dc36 LC |
3677 | |
3678 | unlock: | |
9d3973de MW |
3679 | folio_unlock(folio); |
3680 | folio_put(folio); | |
d863dc36 LC |
3681 | return err; |
3682 | } | |
3683 | ||
923ae0ff RZ |
3684 | /* |
3685 | * ext4_block_zero_page_range() zeros out a mapping of length 'length' | |
3686 | * starting from file offset 'from'. The range to be zero'd must | |
3687 | * be contained with in one block. If the specified range exceeds | |
3688 | * the end of the block it will be shortened to end of the block | |
3088e5a5 | 3689 | * that corresponds to 'from' |
923ae0ff RZ |
3690 | */ |
3691 | static int ext4_block_zero_page_range(handle_t *handle, | |
3692 | struct address_space *mapping, loff_t from, loff_t length) | |
3693 | { | |
3694 | struct inode *inode = mapping->host; | |
09cbfeaf | 3695 | unsigned offset = from & (PAGE_SIZE-1); |
923ae0ff RZ |
3696 | unsigned blocksize = inode->i_sb->s_blocksize; |
3697 | unsigned max = blocksize - (offset & (blocksize - 1)); | |
3698 | ||
3699 | /* | |
3700 | * correct length if it does not fall between | |
3701 | * 'from' and the end of the block | |
3702 | */ | |
3703 | if (length > max || length < 0) | |
3704 | length = max; | |
3705 | ||
47e69351 | 3706 | if (IS_DAX(inode)) { |
c6f40468 CH |
3707 | return dax_zero_range(inode, from, length, NULL, |
3708 | &ext4_iomap_ops); | |
47e69351 | 3709 | } |
923ae0ff RZ |
3710 | return __ext4_block_zero_page_range(handle, mapping, from, length); |
3711 | } | |
3712 | ||
94350ab5 MW |
3713 | /* |
3714 | * ext4_block_truncate_page() zeroes out a mapping from file offset `from' | |
3715 | * up to the end of the block which corresponds to `from'. | |
3716 | * This required during truncate. We need to physically zero the tail end | |
3717 | * of that block so it doesn't yield old data if the file is later grown. | |
3718 | */ | |
c197855e | 3719 | static int ext4_block_truncate_page(handle_t *handle, |
94350ab5 MW |
3720 | struct address_space *mapping, loff_t from) |
3721 | { | |
09cbfeaf | 3722 | unsigned offset = from & (PAGE_SIZE-1); |
94350ab5 MW |
3723 | unsigned length; |
3724 | unsigned blocksize; | |
3725 | struct inode *inode = mapping->host; | |
3726 | ||
0d06863f | 3727 | /* If we are processing an encrypted inode during orphan list handling */ |
592ddec7 | 3728 | if (IS_ENCRYPTED(inode) && !fscrypt_has_encryption_key(inode)) |
0d06863f TT |
3729 | return 0; |
3730 | ||
94350ab5 MW |
3731 | blocksize = inode->i_sb->s_blocksize; |
3732 | length = blocksize - (offset & (blocksize - 1)); | |
3733 | ||
3734 | return ext4_block_zero_page_range(handle, mapping, from, length); | |
3735 | } | |
3736 | ||
a87dd18c LC |
3737 | int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode, |
3738 | loff_t lstart, loff_t length) | |
3739 | { | |
3740 | struct super_block *sb = inode->i_sb; | |
3741 | struct address_space *mapping = inode->i_mapping; | |
e1be3a92 | 3742 | unsigned partial_start, partial_end; |
a87dd18c LC |
3743 | ext4_fsblk_t start, end; |
3744 | loff_t byte_end = (lstart + length - 1); | |
3745 | int err = 0; | |
3746 | ||
e1be3a92 LC |
3747 | partial_start = lstart & (sb->s_blocksize - 1); |
3748 | partial_end = byte_end & (sb->s_blocksize - 1); | |
3749 | ||
a87dd18c LC |
3750 | start = lstart >> sb->s_blocksize_bits; |
3751 | end = byte_end >> sb->s_blocksize_bits; | |
3752 | ||
3753 | /* Handle partial zero within the single block */ | |
e1be3a92 LC |
3754 | if (start == end && |
3755 | (partial_start || (partial_end != sb->s_blocksize - 1))) { | |
a87dd18c LC |
3756 | err = ext4_block_zero_page_range(handle, mapping, |
3757 | lstart, length); | |
3758 | return err; | |
3759 | } | |
3760 | /* Handle partial zero out on the start of the range */ | |
e1be3a92 | 3761 | if (partial_start) { |
a87dd18c LC |
3762 | err = ext4_block_zero_page_range(handle, mapping, |
3763 | lstart, sb->s_blocksize); | |
3764 | if (err) | |
3765 | return err; | |
3766 | } | |
3767 | /* Handle partial zero out on the end of the range */ | |
e1be3a92 | 3768 | if (partial_end != sb->s_blocksize - 1) |
a87dd18c | 3769 | err = ext4_block_zero_page_range(handle, mapping, |
e1be3a92 LC |
3770 | byte_end - partial_end, |
3771 | partial_end + 1); | |
a87dd18c LC |
3772 | return err; |
3773 | } | |
3774 | ||
91ef4caf DG |
3775 | int ext4_can_truncate(struct inode *inode) |
3776 | { | |
91ef4caf DG |
3777 | if (S_ISREG(inode->i_mode)) |
3778 | return 1; | |
3779 | if (S_ISDIR(inode->i_mode)) | |
3780 | return 1; | |
3781 | if (S_ISLNK(inode->i_mode)) | |
3782 | return !ext4_inode_is_fast_symlink(inode); | |
3783 | return 0; | |
3784 | } | |
3785 | ||
01127848 JK |
3786 | /* |
3787 | * We have to make sure i_disksize gets properly updated before we truncate | |
3788 | * page cache due to hole punching or zero range. Otherwise i_disksize update | |
3789 | * can get lost as it may have been postponed to submission of writeback but | |
3790 | * that will never happen after we truncate page cache. | |
3791 | */ | |
3792 | int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset, | |
3793 | loff_t len) | |
3794 | { | |
3795 | handle_t *handle; | |
4209ae12 HS |
3796 | int ret; |
3797 | ||
01127848 JK |
3798 | loff_t size = i_size_read(inode); |
3799 | ||
5955102c | 3800 | WARN_ON(!inode_is_locked(inode)); |
01127848 JK |
3801 | if (offset > size || offset + len < size) |
3802 | return 0; | |
3803 | ||
3804 | if (EXT4_I(inode)->i_disksize >= size) | |
3805 | return 0; | |
3806 | ||
3807 | handle = ext4_journal_start(inode, EXT4_HT_MISC, 1); | |
3808 | if (IS_ERR(handle)) | |
3809 | return PTR_ERR(handle); | |
3810 | ext4_update_i_disksize(inode, size); | |
4209ae12 | 3811 | ret = ext4_mark_inode_dirty(handle, inode); |
01127848 JK |
3812 | ext4_journal_stop(handle); |
3813 | ||
4209ae12 | 3814 | return ret; |
01127848 JK |
3815 | } |
3816 | ||
d4f5258e | 3817 | static void ext4_wait_dax_page(struct inode *inode) |
430657b6 | 3818 | { |
d4f5258e | 3819 | filemap_invalidate_unlock(inode->i_mapping); |
430657b6 | 3820 | schedule(); |
d4f5258e | 3821 | filemap_invalidate_lock(inode->i_mapping); |
430657b6 RZ |
3822 | } |
3823 | ||
3824 | int ext4_break_layouts(struct inode *inode) | |
3825 | { | |
430657b6 | 3826 | struct page *page; |
430657b6 RZ |
3827 | int error; |
3828 | ||
d4f5258e | 3829 | if (WARN_ON_ONCE(!rwsem_is_locked(&inode->i_mapping->invalidate_lock))) |
430657b6 RZ |
3830 | return -EINVAL; |
3831 | ||
3832 | do { | |
430657b6 RZ |
3833 | page = dax_layout_busy_page(inode->i_mapping); |
3834 | if (!page) | |
3835 | return 0; | |
3836 | ||
3837 | error = ___wait_var_event(&page->_refcount, | |
3838 | atomic_read(&page->_refcount) == 1, | |
3839 | TASK_INTERRUPTIBLE, 0, 0, | |
d4f5258e | 3840 | ext4_wait_dax_page(inode)); |
b1f38217 | 3841 | } while (error == 0); |
430657b6 RZ |
3842 | |
3843 | return error; | |
3844 | } | |
3845 | ||
a4bb6b64 | 3846 | /* |
cca32b7e | 3847 | * ext4_punch_hole: punches a hole in a file by releasing the blocks |
a4bb6b64 AH |
3848 | * associated with the given offset and length |
3849 | * | |
3850 | * @inode: File inode | |
3851 | * @offset: The offset where the hole will begin | |
3852 | * @len: The length of the hole | |
3853 | * | |
4907cb7b | 3854 | * Returns: 0 on success or negative on failure |
a4bb6b64 AH |
3855 | */ |
3856 | ||
ad5cd4f4 | 3857 | int ext4_punch_hole(struct file *file, loff_t offset, loff_t length) |
a4bb6b64 | 3858 | { |
ad5cd4f4 | 3859 | struct inode *inode = file_inode(file); |
26a4c0c6 TT |
3860 | struct super_block *sb = inode->i_sb; |
3861 | ext4_lblk_t first_block, stop_block; | |
3862 | struct address_space *mapping = inode->i_mapping; | |
2da37622 TS |
3863 | loff_t first_block_offset, last_block_offset, max_length; |
3864 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
26a4c0c6 TT |
3865 | handle_t *handle; |
3866 | unsigned int credits; | |
4209ae12 | 3867 | int ret = 0, ret2 = 0; |
26a4c0c6 | 3868 | |
b8a86845 | 3869 | trace_ext4_punch_hole(inode, offset, length, 0); |
aaddea81 | 3870 | |
26a4c0c6 TT |
3871 | /* |
3872 | * Write out all dirty pages to avoid race conditions | |
3873 | * Then release them. | |
3874 | */ | |
cca32b7e | 3875 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { |
26a4c0c6 TT |
3876 | ret = filemap_write_and_wait_range(mapping, offset, |
3877 | offset + length - 1); | |
3878 | if (ret) | |
3879 | return ret; | |
3880 | } | |
3881 | ||
5955102c | 3882 | inode_lock(inode); |
9ef06cec | 3883 | |
26a4c0c6 TT |
3884 | /* No need to punch hole beyond i_size */ |
3885 | if (offset >= inode->i_size) | |
3886 | goto out_mutex; | |
3887 | ||
3888 | /* | |
3889 | * If the hole extends beyond i_size, set the hole | |
3890 | * to end after the page that contains i_size | |
3891 | */ | |
3892 | if (offset + length > inode->i_size) { | |
3893 | length = inode->i_size + | |
09cbfeaf | 3894 | PAGE_SIZE - (inode->i_size & (PAGE_SIZE - 1)) - |
26a4c0c6 TT |
3895 | offset; |
3896 | } | |
3897 | ||
2da37622 TS |
3898 | /* |
3899 | * For punch hole the length + offset needs to be within one block | |
3900 | * before last range. Adjust the length if it goes beyond that limit. | |
3901 | */ | |
3902 | max_length = sbi->s_bitmap_maxbytes - inode->i_sb->s_blocksize; | |
3903 | if (offset + length > max_length) | |
3904 | length = max_length - offset; | |
3905 | ||
a361293f JK |
3906 | if (offset & (sb->s_blocksize - 1) || |
3907 | (offset + length) & (sb->s_blocksize - 1)) { | |
3908 | /* | |
3909 | * Attach jinode to inode for jbd2 if we do any zeroing of | |
3910 | * partial block | |
3911 | */ | |
3912 | ret = ext4_inode_attach_jinode(inode); | |
3913 | if (ret < 0) | |
3914 | goto out_mutex; | |
3915 | ||
3916 | } | |
3917 | ||
f340b3d9 | 3918 | /* Wait all existing dio workers, newcomers will block on i_rwsem */ |
ea3d7209 JK |
3919 | inode_dio_wait(inode); |
3920 | ||
ad5cd4f4 DW |
3921 | ret = file_modified(file); |
3922 | if (ret) | |
3923 | goto out_mutex; | |
3924 | ||
ea3d7209 JK |
3925 | /* |
3926 | * Prevent page faults from reinstantiating pages we have released from | |
3927 | * page cache. | |
3928 | */ | |
d4f5258e | 3929 | filemap_invalidate_lock(mapping); |
430657b6 RZ |
3930 | |
3931 | ret = ext4_break_layouts(inode); | |
3932 | if (ret) | |
3933 | goto out_dio; | |
3934 | ||
a87dd18c LC |
3935 | first_block_offset = round_up(offset, sb->s_blocksize); |
3936 | last_block_offset = round_down((offset + length), sb->s_blocksize) - 1; | |
26a4c0c6 | 3937 | |
a87dd18c | 3938 | /* Now release the pages and zero block aligned part of pages*/ |
01127848 JK |
3939 | if (last_block_offset > first_block_offset) { |
3940 | ret = ext4_update_disksize_before_punch(inode, offset, length); | |
3941 | if (ret) | |
3942 | goto out_dio; | |
a87dd18c LC |
3943 | truncate_pagecache_range(inode, first_block_offset, |
3944 | last_block_offset); | |
01127848 | 3945 | } |
26a4c0c6 TT |
3946 | |
3947 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
3948 | credits = ext4_writepage_trans_blocks(inode); | |
3949 | else | |
3950 | credits = ext4_blocks_for_truncate(inode); | |
3951 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); | |
3952 | if (IS_ERR(handle)) { | |
3953 | ret = PTR_ERR(handle); | |
3954 | ext4_std_error(sb, ret); | |
3955 | goto out_dio; | |
3956 | } | |
3957 | ||
a87dd18c LC |
3958 | ret = ext4_zero_partial_blocks(handle, inode, offset, |
3959 | length); | |
3960 | if (ret) | |
3961 | goto out_stop; | |
26a4c0c6 TT |
3962 | |
3963 | first_block = (offset + sb->s_blocksize - 1) >> | |
3964 | EXT4_BLOCK_SIZE_BITS(sb); | |
3965 | stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb); | |
3966 | ||
eee597ac LC |
3967 | /* If there are blocks to remove, do it */ |
3968 | if (stop_block > first_block) { | |
26a4c0c6 | 3969 | |
eee597ac | 3970 | down_write(&EXT4_I(inode)->i_data_sem); |
27bc446e | 3971 | ext4_discard_preallocations(inode, 0); |
26a4c0c6 | 3972 | |
ed5d285b BL |
3973 | ext4_es_remove_extent(inode, first_block, |
3974 | stop_block - first_block); | |
26a4c0c6 | 3975 | |
eee597ac LC |
3976 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
3977 | ret = ext4_ext_remove_space(inode, first_block, | |
3978 | stop_block - 1); | |
3979 | else | |
3980 | ret = ext4_ind_remove_space(handle, inode, first_block, | |
3981 | stop_block); | |
26a4c0c6 | 3982 | |
eee597ac LC |
3983 | up_write(&EXT4_I(inode)->i_data_sem); |
3984 | } | |
a80f7fcf | 3985 | ext4_fc_track_range(handle, inode, first_block, stop_block); |
26a4c0c6 TT |
3986 | if (IS_SYNC(inode)) |
3987 | ext4_handle_sync(handle); | |
e251f9bc | 3988 | |
eeca7ea1 | 3989 | inode->i_mtime = inode->i_ctime = current_time(inode); |
4209ae12 HS |
3990 | ret2 = ext4_mark_inode_dirty(handle, inode); |
3991 | if (unlikely(ret2)) | |
3992 | ret = ret2; | |
67a7d5f5 JK |
3993 | if (ret >= 0) |
3994 | ext4_update_inode_fsync_trans(handle, inode, 1); | |
26a4c0c6 TT |
3995 | out_stop: |
3996 | ext4_journal_stop(handle); | |
3997 | out_dio: | |
d4f5258e | 3998 | filemap_invalidate_unlock(mapping); |
26a4c0c6 | 3999 | out_mutex: |
5955102c | 4000 | inode_unlock(inode); |
26a4c0c6 | 4001 | return ret; |
a4bb6b64 AH |
4002 | } |
4003 | ||
a361293f JK |
4004 | int ext4_inode_attach_jinode(struct inode *inode) |
4005 | { | |
4006 | struct ext4_inode_info *ei = EXT4_I(inode); | |
4007 | struct jbd2_inode *jinode; | |
4008 | ||
4009 | if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal) | |
4010 | return 0; | |
4011 | ||
4012 | jinode = jbd2_alloc_inode(GFP_KERNEL); | |
4013 | spin_lock(&inode->i_lock); | |
4014 | if (!ei->jinode) { | |
4015 | if (!jinode) { | |
4016 | spin_unlock(&inode->i_lock); | |
4017 | return -ENOMEM; | |
4018 | } | |
4019 | ei->jinode = jinode; | |
4020 | jbd2_journal_init_jbd_inode(ei->jinode, inode); | |
4021 | jinode = NULL; | |
4022 | } | |
4023 | spin_unlock(&inode->i_lock); | |
4024 | if (unlikely(jinode != NULL)) | |
4025 | jbd2_free_inode(jinode); | |
4026 | return 0; | |
4027 | } | |
4028 | ||
ac27a0ec | 4029 | /* |
617ba13b | 4030 | * ext4_truncate() |
ac27a0ec | 4031 | * |
617ba13b MC |
4032 | * We block out ext4_get_block() block instantiations across the entire |
4033 | * transaction, and VFS/VM ensures that ext4_truncate() cannot run | |
ac27a0ec DK |
4034 | * simultaneously on behalf of the same inode. |
4035 | * | |
42b2aa86 | 4036 | * As we work through the truncate and commit bits of it to the journal there |
ac27a0ec DK |
4037 | * is one core, guiding principle: the file's tree must always be consistent on |
4038 | * disk. We must be able to restart the truncate after a crash. | |
4039 | * | |
4040 | * The file's tree may be transiently inconsistent in memory (although it | |
4041 | * probably isn't), but whenever we close off and commit a journal transaction, | |
4042 | * the contents of (the filesystem + the journal) must be consistent and | |
4043 | * restartable. It's pretty simple, really: bottom up, right to left (although | |
4044 | * left-to-right works OK too). | |
4045 | * | |
4046 | * Note that at recovery time, journal replay occurs *before* the restart of | |
4047 | * truncate against the orphan inode list. | |
4048 | * | |
4049 | * The committed inode has the new, desired i_size (which is the same as | |
617ba13b | 4050 | * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see |
ac27a0ec | 4051 | * that this inode's truncate did not complete and it will again call |
617ba13b MC |
4052 | * ext4_truncate() to have another go. So there will be instantiated blocks |
4053 | * to the right of the truncation point in a crashed ext4 filesystem. But | |
ac27a0ec | 4054 | * that's fine - as long as they are linked from the inode, the post-crash |
617ba13b | 4055 | * ext4_truncate() run will find them and release them. |
ac27a0ec | 4056 | */ |
2c98eb5e | 4057 | int ext4_truncate(struct inode *inode) |
ac27a0ec | 4058 | { |
819c4920 TT |
4059 | struct ext4_inode_info *ei = EXT4_I(inode); |
4060 | unsigned int credits; | |
4209ae12 | 4061 | int err = 0, err2; |
819c4920 TT |
4062 | handle_t *handle; |
4063 | struct address_space *mapping = inode->i_mapping; | |
819c4920 | 4064 | |
19b5ef61 TT |
4065 | /* |
4066 | * There is a possibility that we're either freeing the inode | |
e04027e8 | 4067 | * or it's a completely new inode. In those cases we might not |
f340b3d9 | 4068 | * have i_rwsem locked because it's not necessary. |
19b5ef61 TT |
4069 | */ |
4070 | if (!(inode->i_state & (I_NEW|I_FREEING))) | |
5955102c | 4071 | WARN_ON(!inode_is_locked(inode)); |
0562e0ba JZ |
4072 | trace_ext4_truncate_enter(inode); |
4073 | ||
91ef4caf | 4074 | if (!ext4_can_truncate(inode)) |
9a5d265f | 4075 | goto out_trace; |
ac27a0ec | 4076 | |
5534fb5b | 4077 | if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC)) |
19f5fb7a | 4078 | ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE); |
7d8f9f7d | 4079 | |
aef1c851 TM |
4080 | if (ext4_has_inline_data(inode)) { |
4081 | int has_inline = 1; | |
4082 | ||
01daf945 | 4083 | err = ext4_inline_data_truncate(inode, &has_inline); |
9a5d265f | 4084 | if (err || has_inline) |
4085 | goto out_trace; | |
aef1c851 TM |
4086 | } |
4087 | ||
a361293f JK |
4088 | /* If we zero-out tail of the page, we have to create jinode for jbd2 */ |
4089 | if (inode->i_size & (inode->i_sb->s_blocksize - 1)) { | |
a71248b1 BL |
4090 | err = ext4_inode_attach_jinode(inode); |
4091 | if (err) | |
9a5d265f | 4092 | goto out_trace; |
a361293f JK |
4093 | } |
4094 | ||
819c4920 TT |
4095 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
4096 | credits = ext4_writepage_trans_blocks(inode); | |
4097 | else | |
4098 | credits = ext4_blocks_for_truncate(inode); | |
4099 | ||
4100 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); | |
9a5d265f | 4101 | if (IS_ERR(handle)) { |
4102 | err = PTR_ERR(handle); | |
4103 | goto out_trace; | |
4104 | } | |
819c4920 | 4105 | |
eb3544c6 LC |
4106 | if (inode->i_size & (inode->i_sb->s_blocksize - 1)) |
4107 | ext4_block_truncate_page(handle, mapping, inode->i_size); | |
819c4920 TT |
4108 | |
4109 | /* | |
4110 | * We add the inode to the orphan list, so that if this | |
4111 | * truncate spans multiple transactions, and we crash, we will | |
4112 | * resume the truncate when the filesystem recovers. It also | |
4113 | * marks the inode dirty, to catch the new size. | |
4114 | * | |
4115 | * Implication: the file must always be in a sane, consistent | |
4116 | * truncatable state while each transaction commits. | |
4117 | */ | |
2c98eb5e TT |
4118 | err = ext4_orphan_add(handle, inode); |
4119 | if (err) | |
819c4920 TT |
4120 | goto out_stop; |
4121 | ||
4122 | down_write(&EXT4_I(inode)->i_data_sem); | |
4123 | ||
27bc446e | 4124 | ext4_discard_preallocations(inode, 0); |
819c4920 | 4125 | |
ff9893dc | 4126 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
d0abb36d | 4127 | err = ext4_ext_truncate(handle, inode); |
ff9893dc | 4128 | else |
819c4920 TT |
4129 | ext4_ind_truncate(handle, inode); |
4130 | ||
4131 | up_write(&ei->i_data_sem); | |
d0abb36d TT |
4132 | if (err) |
4133 | goto out_stop; | |
819c4920 TT |
4134 | |
4135 | if (IS_SYNC(inode)) | |
4136 | ext4_handle_sync(handle); | |
4137 | ||
4138 | out_stop: | |
4139 | /* | |
4140 | * If this was a simple ftruncate() and the file will remain alive, | |
4141 | * then we need to clear up the orphan record which we created above. | |
4142 | * However, if this was a real unlink then we were called by | |
58d86a50 | 4143 | * ext4_evict_inode(), and we allow that function to clean up the |
819c4920 TT |
4144 | * orphan info for us. |
4145 | */ | |
4146 | if (inode->i_nlink) | |
4147 | ext4_orphan_del(handle, inode); | |
4148 | ||
eeca7ea1 | 4149 | inode->i_mtime = inode->i_ctime = current_time(inode); |
4209ae12 HS |
4150 | err2 = ext4_mark_inode_dirty(handle, inode); |
4151 | if (unlikely(err2 && !err)) | |
4152 | err = err2; | |
819c4920 | 4153 | ext4_journal_stop(handle); |
ac27a0ec | 4154 | |
9a5d265f | 4155 | out_trace: |
0562e0ba | 4156 | trace_ext4_truncate_exit(inode); |
2c98eb5e | 4157 | return err; |
ac27a0ec DK |
4158 | } |
4159 | ||
9a1bf32c ZY |
4160 | static inline u64 ext4_inode_peek_iversion(const struct inode *inode) |
4161 | { | |
4162 | if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) | |
4163 | return inode_peek_iversion_raw(inode); | |
4164 | else | |
4165 | return inode_peek_iversion(inode); | |
4166 | } | |
4167 | ||
4168 | static int ext4_inode_blocks_set(struct ext4_inode *raw_inode, | |
4169 | struct ext4_inode_info *ei) | |
4170 | { | |
4171 | struct inode *inode = &(ei->vfs_inode); | |
4172 | u64 i_blocks = READ_ONCE(inode->i_blocks); | |
4173 | struct super_block *sb = inode->i_sb; | |
4174 | ||
4175 | if (i_blocks <= ~0U) { | |
4176 | /* | |
4177 | * i_blocks can be represented in a 32 bit variable | |
4178 | * as multiple of 512 bytes | |
4179 | */ | |
4180 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); | |
4181 | raw_inode->i_blocks_high = 0; | |
4182 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); | |
4183 | return 0; | |
4184 | } | |
4185 | ||
4186 | /* | |
4187 | * This should never happen since sb->s_maxbytes should not have | |
4188 | * allowed this, sb->s_maxbytes was set according to the huge_file | |
4189 | * feature in ext4_fill_super(). | |
4190 | */ | |
4191 | if (!ext4_has_feature_huge_file(sb)) | |
4192 | return -EFSCORRUPTED; | |
4193 | ||
4194 | if (i_blocks <= 0xffffffffffffULL) { | |
4195 | /* | |
4196 | * i_blocks can be represented in a 48 bit variable | |
4197 | * as multiple of 512 bytes | |
4198 | */ | |
4199 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); | |
4200 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); | |
4201 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); | |
4202 | } else { | |
4203 | ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE); | |
4204 | /* i_block is stored in file system block size */ | |
4205 | i_blocks = i_blocks >> (inode->i_blkbits - 9); | |
4206 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); | |
4207 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); | |
4208 | } | |
4209 | return 0; | |
4210 | } | |
4211 | ||
4212 | static int ext4_fill_raw_inode(struct inode *inode, struct ext4_inode *raw_inode) | |
4213 | { | |
4214 | struct ext4_inode_info *ei = EXT4_I(inode); | |
4215 | uid_t i_uid; | |
4216 | gid_t i_gid; | |
4217 | projid_t i_projid; | |
4218 | int block; | |
4219 | int err; | |
4220 | ||
4221 | err = ext4_inode_blocks_set(raw_inode, ei); | |
4222 | ||
4223 | raw_inode->i_mode = cpu_to_le16(inode->i_mode); | |
4224 | i_uid = i_uid_read(inode); | |
4225 | i_gid = i_gid_read(inode); | |
4226 | i_projid = from_kprojid(&init_user_ns, ei->i_projid); | |
4227 | if (!(test_opt(inode->i_sb, NO_UID32))) { | |
4228 | raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid)); | |
4229 | raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid)); | |
4230 | /* | |
4231 | * Fix up interoperability with old kernels. Otherwise, | |
4232 | * old inodes get re-used with the upper 16 bits of the | |
4233 | * uid/gid intact. | |
4234 | */ | |
4235 | if (ei->i_dtime && list_empty(&ei->i_orphan)) { | |
4236 | raw_inode->i_uid_high = 0; | |
4237 | raw_inode->i_gid_high = 0; | |
4238 | } else { | |
4239 | raw_inode->i_uid_high = | |
4240 | cpu_to_le16(high_16_bits(i_uid)); | |
4241 | raw_inode->i_gid_high = | |
4242 | cpu_to_le16(high_16_bits(i_gid)); | |
4243 | } | |
4244 | } else { | |
4245 | raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid)); | |
4246 | raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid)); | |
4247 | raw_inode->i_uid_high = 0; | |
4248 | raw_inode->i_gid_high = 0; | |
4249 | } | |
4250 | raw_inode->i_links_count = cpu_to_le16(inode->i_nlink); | |
4251 | ||
4252 | EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode); | |
4253 | EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode); | |
4254 | EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode); | |
4255 | EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode); | |
4256 | ||
4257 | raw_inode->i_dtime = cpu_to_le32(ei->i_dtime); | |
4258 | raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF); | |
4259 | if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) | |
4260 | raw_inode->i_file_acl_high = | |
4261 | cpu_to_le16(ei->i_file_acl >> 32); | |
4262 | raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl); | |
4263 | ext4_isize_set(raw_inode, ei->i_disksize); | |
4264 | ||
4265 | raw_inode->i_generation = cpu_to_le32(inode->i_generation); | |
4266 | if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) { | |
4267 | if (old_valid_dev(inode->i_rdev)) { | |
4268 | raw_inode->i_block[0] = | |
4269 | cpu_to_le32(old_encode_dev(inode->i_rdev)); | |
4270 | raw_inode->i_block[1] = 0; | |
4271 | } else { | |
4272 | raw_inode->i_block[0] = 0; | |
4273 | raw_inode->i_block[1] = | |
4274 | cpu_to_le32(new_encode_dev(inode->i_rdev)); | |
4275 | raw_inode->i_block[2] = 0; | |
4276 | } | |
4277 | } else if (!ext4_has_inline_data(inode)) { | |
4278 | for (block = 0; block < EXT4_N_BLOCKS; block++) | |
4279 | raw_inode->i_block[block] = ei->i_data[block]; | |
4280 | } | |
4281 | ||
4282 | if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) { | |
4283 | u64 ivers = ext4_inode_peek_iversion(inode); | |
4284 | ||
4285 | raw_inode->i_disk_version = cpu_to_le32(ivers); | |
4286 | if (ei->i_extra_isize) { | |
4287 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
4288 | raw_inode->i_version_hi = | |
4289 | cpu_to_le32(ivers >> 32); | |
4290 | raw_inode->i_extra_isize = | |
4291 | cpu_to_le16(ei->i_extra_isize); | |
4292 | } | |
4293 | } | |
4294 | ||
4295 | if (i_projid != EXT4_DEF_PROJID && | |
4296 | !ext4_has_feature_project(inode->i_sb)) | |
4297 | err = err ?: -EFSCORRUPTED; | |
4298 | ||
4299 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && | |
4300 | EXT4_FITS_IN_INODE(raw_inode, ei, i_projid)) | |
4301 | raw_inode->i_projid = cpu_to_le32(i_projid); | |
4302 | ||
4303 | ext4_inode_csum_set(inode, raw_inode, ei); | |
4304 | return err; | |
4305 | } | |
4306 | ||
ac27a0ec | 4307 | /* |
617ba13b | 4308 | * ext4_get_inode_loc returns with an extra refcount against the inode's |
de01f484 ZY |
4309 | * underlying buffer_head on success. If we pass 'inode' and it does not |
4310 | * have in-inode xattr, we have all inode data in memory that is needed | |
4311 | * to recreate the on-disk version of this inode. | |
ac27a0ec | 4312 | */ |
8016e29f | 4313 | static int __ext4_get_inode_loc(struct super_block *sb, unsigned long ino, |
de01f484 | 4314 | struct inode *inode, struct ext4_iloc *iloc, |
8016e29f | 4315 | ext4_fsblk_t *ret_block) |
ac27a0ec | 4316 | { |
240799cd TT |
4317 | struct ext4_group_desc *gdp; |
4318 | struct buffer_head *bh; | |
240799cd | 4319 | ext4_fsblk_t block; |
02f03c42 | 4320 | struct blk_plug plug; |
240799cd TT |
4321 | int inodes_per_block, inode_offset; |
4322 | ||
3a06d778 | 4323 | iloc->bh = NULL; |
8016e29f HS |
4324 | if (ino < EXT4_ROOT_INO || |
4325 | ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count)) | |
6a797d27 | 4326 | return -EFSCORRUPTED; |
ac27a0ec | 4327 | |
8016e29f | 4328 | iloc->block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb); |
240799cd TT |
4329 | gdp = ext4_get_group_desc(sb, iloc->block_group, NULL); |
4330 | if (!gdp) | |
ac27a0ec DK |
4331 | return -EIO; |
4332 | ||
240799cd TT |
4333 | /* |
4334 | * Figure out the offset within the block group inode table | |
4335 | */ | |
00d09882 | 4336 | inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; |
8016e29f | 4337 | inode_offset = ((ino - 1) % |
240799cd | 4338 | EXT4_INODES_PER_GROUP(sb)); |
240799cd TT |
4339 | iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb); |
4340 | ||
eee22187 BL |
4341 | block = ext4_inode_table(sb, gdp); |
4342 | if ((block <= le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block)) || | |
4343 | (block >= ext4_blocks_count(EXT4_SB(sb)->s_es))) { | |
4344 | ext4_error(sb, "Invalid inode table block %llu in " | |
4345 | "block_group %u", block, iloc->block_group); | |
4346 | return -EFSCORRUPTED; | |
4347 | } | |
4348 | block += (inode_offset / inodes_per_block); | |
4349 | ||
240799cd | 4350 | bh = sb_getblk(sb, block); |
aebf0243 | 4351 | if (unlikely(!bh)) |
860d21e2 | 4352 | return -ENOMEM; |
8e33fadf ZY |
4353 | if (ext4_buffer_uptodate(bh)) |
4354 | goto has_buffer; | |
9c83a923 | 4355 | |
8e33fadf | 4356 | lock_buffer(bh); |
f2c77973 ZY |
4357 | if (ext4_buffer_uptodate(bh)) { |
4358 | /* Someone brought it uptodate while we waited */ | |
4359 | unlock_buffer(bh); | |
4360 | goto has_buffer; | |
4361 | } | |
4362 | ||
8e33fadf ZY |
4363 | /* |
4364 | * If we have all information of the inode in memory and this | |
4365 | * is the only valid inode in the block, we need not read the | |
4366 | * block. | |
4367 | */ | |
de01f484 | 4368 | if (inode && !ext4_test_inode_state(inode, EXT4_STATE_XATTR)) { |
8e33fadf ZY |
4369 | struct buffer_head *bitmap_bh; |
4370 | int i, start; | |
ac27a0ec | 4371 | |
8e33fadf | 4372 | start = inode_offset & ~(inodes_per_block - 1); |
ac27a0ec | 4373 | |
8e33fadf ZY |
4374 | /* Is the inode bitmap in cache? */ |
4375 | bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp)); | |
4376 | if (unlikely(!bitmap_bh)) | |
4377 | goto make_io; | |
ac27a0ec | 4378 | |
8e33fadf ZY |
4379 | /* |
4380 | * If the inode bitmap isn't in cache then the | |
4381 | * optimisation may end up performing two reads instead | |
4382 | * of one, so skip it. | |
4383 | */ | |
4384 | if (!buffer_uptodate(bitmap_bh)) { | |
ac27a0ec | 4385 | brelse(bitmap_bh); |
8e33fadf | 4386 | goto make_io; |
ac27a0ec | 4387 | } |
8e33fadf ZY |
4388 | for (i = start; i < start + inodes_per_block; i++) { |
4389 | if (i == inode_offset) | |
4390 | continue; | |
4391 | if (ext4_test_bit(i, bitmap_bh->b_data)) | |
4392 | break; | |
ac27a0ec | 4393 | } |
8e33fadf ZY |
4394 | brelse(bitmap_bh); |
4395 | if (i == start + inodes_per_block) { | |
de01f484 ZY |
4396 | struct ext4_inode *raw_inode = |
4397 | (struct ext4_inode *) (bh->b_data + iloc->offset); | |
4398 | ||
8e33fadf ZY |
4399 | /* all other inodes are free, so skip I/O */ |
4400 | memset(bh->b_data, 0, bh->b_size); | |
de01f484 ZY |
4401 | if (!ext4_test_inode_state(inode, EXT4_STATE_NEW)) |
4402 | ext4_fill_raw_inode(inode, raw_inode); | |
8e33fadf ZY |
4403 | set_buffer_uptodate(bh); |
4404 | unlock_buffer(bh); | |
4405 | goto has_buffer; | |
4406 | } | |
4407 | } | |
ac27a0ec DK |
4408 | |
4409 | make_io: | |
8e33fadf ZY |
4410 | /* |
4411 | * If we need to do any I/O, try to pre-readahead extra | |
4412 | * blocks from the inode table. | |
4413 | */ | |
4414 | blk_start_plug(&plug); | |
4415 | if (EXT4_SB(sb)->s_inode_readahead_blks) { | |
4416 | ext4_fsblk_t b, end, table; | |
4417 | unsigned num; | |
4418 | __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks; | |
4419 | ||
4420 | table = ext4_inode_table(sb, gdp); | |
4421 | /* s_inode_readahead_blks is always a power of 2 */ | |
4422 | b = block & ~((ext4_fsblk_t) ra_blks - 1); | |
4423 | if (table > b) | |
4424 | b = table; | |
4425 | end = b + ra_blks; | |
4426 | num = EXT4_INODES_PER_GROUP(sb); | |
4427 | if (ext4_has_group_desc_csum(sb)) | |
4428 | num -= ext4_itable_unused_count(sb, gdp); | |
4429 | table += num / inodes_per_block; | |
4430 | if (end > table) | |
4431 | end = table; | |
4432 | while (b <= end) | |
4433 | ext4_sb_breadahead_unmovable(sb, b++); | |
4434 | } | |
240799cd | 4435 | |
8e33fadf ZY |
4436 | /* |
4437 | * There are other valid inodes in the buffer, this inode | |
4438 | * has in-inode xattrs, or we don't have this inode in memory. | |
4439 | * Read the block from disk. | |
4440 | */ | |
4441 | trace_ext4_load_inode(sb, ino); | |
4442 | ext4_read_bh_nowait(bh, REQ_META | REQ_PRIO, NULL); | |
4443 | blk_finish_plug(&plug); | |
4444 | wait_on_buffer(bh); | |
4445 | ext4_simulate_fail_bh(sb, bh, EXT4_SIM_INODE_EIO); | |
4446 | if (!buffer_uptodate(bh)) { | |
4447 | if (ret_block) | |
4448 | *ret_block = block; | |
4449 | brelse(bh); | |
4450 | return -EIO; | |
ac27a0ec DK |
4451 | } |
4452 | has_buffer: | |
4453 | iloc->bh = bh; | |
4454 | return 0; | |
4455 | } | |
4456 | ||
8016e29f HS |
4457 | static int __ext4_get_inode_loc_noinmem(struct inode *inode, |
4458 | struct ext4_iloc *iloc) | |
4459 | { | |
c27c29c6 | 4460 | ext4_fsblk_t err_blk = 0; |
8016e29f HS |
4461 | int ret; |
4462 | ||
de01f484 | 4463 | ret = __ext4_get_inode_loc(inode->i_sb, inode->i_ino, NULL, iloc, |
8016e29f HS |
4464 | &err_blk); |
4465 | ||
4466 | if (ret == -EIO) | |
4467 | ext4_error_inode_block(inode, err_blk, EIO, | |
4468 | "unable to read itable block"); | |
4469 | ||
4470 | return ret; | |
4471 | } | |
4472 | ||
617ba13b | 4473 | int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec | 4474 | { |
c27c29c6 | 4475 | ext4_fsblk_t err_blk = 0; |
8016e29f HS |
4476 | int ret; |
4477 | ||
de01f484 ZY |
4478 | ret = __ext4_get_inode_loc(inode->i_sb, inode->i_ino, inode, iloc, |
4479 | &err_blk); | |
8016e29f HS |
4480 | |
4481 | if (ret == -EIO) | |
4482 | ext4_error_inode_block(inode, err_blk, EIO, | |
4483 | "unable to read itable block"); | |
4484 | ||
4485 | return ret; | |
4486 | } | |
4487 | ||
4488 | ||
4489 | int ext4_get_fc_inode_loc(struct super_block *sb, unsigned long ino, | |
4490 | struct ext4_iloc *iloc) | |
4491 | { | |
de01f484 | 4492 | return __ext4_get_inode_loc(sb, ino, NULL, iloc, NULL); |
ac27a0ec DK |
4493 | } |
4494 | ||
a8ab6d38 | 4495 | static bool ext4_should_enable_dax(struct inode *inode) |
6642586b | 4496 | { |
a8ab6d38 IW |
4497 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
4498 | ||
9cb20f94 | 4499 | if (test_opt2(inode->i_sb, DAX_NEVER)) |
6642586b RZ |
4500 | return false; |
4501 | if (!S_ISREG(inode->i_mode)) | |
4502 | return false; | |
4503 | if (ext4_should_journal_data(inode)) | |
4504 | return false; | |
4505 | if (ext4_has_inline_data(inode)) | |
4506 | return false; | |
592ddec7 | 4507 | if (ext4_test_inode_flag(inode, EXT4_INODE_ENCRYPT)) |
6642586b | 4508 | return false; |
c93d8f88 EB |
4509 | if (ext4_test_inode_flag(inode, EXT4_INODE_VERITY)) |
4510 | return false; | |
a8ab6d38 IW |
4511 | if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags)) |
4512 | return false; | |
4513 | if (test_opt(inode->i_sb, DAX_ALWAYS)) | |
4514 | return true; | |
4515 | ||
b383a73f | 4516 | return ext4_test_inode_flag(inode, EXT4_INODE_DAX); |
6642586b RZ |
4517 | } |
4518 | ||
043546e4 | 4519 | void ext4_set_inode_flags(struct inode *inode, bool init) |
ac27a0ec | 4520 | { |
617ba13b | 4521 | unsigned int flags = EXT4_I(inode)->i_flags; |
00a1a053 | 4522 | unsigned int new_fl = 0; |
ac27a0ec | 4523 | |
043546e4 IW |
4524 | WARN_ON_ONCE(IS_DAX(inode) && init); |
4525 | ||
617ba13b | 4526 | if (flags & EXT4_SYNC_FL) |
00a1a053 | 4527 | new_fl |= S_SYNC; |
617ba13b | 4528 | if (flags & EXT4_APPEND_FL) |
00a1a053 | 4529 | new_fl |= S_APPEND; |
617ba13b | 4530 | if (flags & EXT4_IMMUTABLE_FL) |
00a1a053 | 4531 | new_fl |= S_IMMUTABLE; |
617ba13b | 4532 | if (flags & EXT4_NOATIME_FL) |
00a1a053 | 4533 | new_fl |= S_NOATIME; |
617ba13b | 4534 | if (flags & EXT4_DIRSYNC_FL) |
00a1a053 | 4535 | new_fl |= S_DIRSYNC; |
043546e4 IW |
4536 | |
4537 | /* Because of the way inode_set_flags() works we must preserve S_DAX | |
4538 | * here if already set. */ | |
4539 | new_fl |= (inode->i_flags & S_DAX); | |
4540 | if (init && ext4_should_enable_dax(inode)) | |
923ae0ff | 4541 | new_fl |= S_DAX; |
043546e4 | 4542 | |
2ee6a576 EB |
4543 | if (flags & EXT4_ENCRYPT_FL) |
4544 | new_fl |= S_ENCRYPTED; | |
b886ee3e GKB |
4545 | if (flags & EXT4_CASEFOLD_FL) |
4546 | new_fl |= S_CASEFOLD; | |
c93d8f88 EB |
4547 | if (flags & EXT4_VERITY_FL) |
4548 | new_fl |= S_VERITY; | |
5f16f322 | 4549 | inode_set_flags(inode, new_fl, |
2ee6a576 | 4550 | S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX| |
c93d8f88 | 4551 | S_ENCRYPTED|S_CASEFOLD|S_VERITY); |
ac27a0ec DK |
4552 | } |
4553 | ||
0fc1b451 | 4554 | static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode, |
de9a55b8 | 4555 | struct ext4_inode_info *ei) |
0fc1b451 AK |
4556 | { |
4557 | blkcnt_t i_blocks ; | |
8180a562 AK |
4558 | struct inode *inode = &(ei->vfs_inode); |
4559 | struct super_block *sb = inode->i_sb; | |
0fc1b451 | 4560 | |
e2b911c5 | 4561 | if (ext4_has_feature_huge_file(sb)) { |
0fc1b451 AK |
4562 | /* we are using combined 48 bit field */ |
4563 | i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 | | |
4564 | le32_to_cpu(raw_inode->i_blocks_lo); | |
07a03824 | 4565 | if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) { |
8180a562 AK |
4566 | /* i_blocks represent file system block size */ |
4567 | return i_blocks << (inode->i_blkbits - 9); | |
4568 | } else { | |
4569 | return i_blocks; | |
4570 | } | |
0fc1b451 AK |
4571 | } else { |
4572 | return le32_to_cpu(raw_inode->i_blocks_lo); | |
4573 | } | |
4574 | } | |
ff9ddf7e | 4575 | |
eb9b5f01 | 4576 | static inline int ext4_iget_extra_inode(struct inode *inode, |
152a7b0a TM |
4577 | struct ext4_inode *raw_inode, |
4578 | struct ext4_inode_info *ei) | |
4579 | { | |
4580 | __le32 *magic = (void *)raw_inode + | |
4581 | EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize; | |
eb9b5f01 | 4582 | |
fd7e672e | 4583 | if (EXT4_INODE_HAS_XATTR_SPACE(inode) && |
290ab230 | 4584 | *magic == cpu_to_le32(EXT4_XATTR_MAGIC)) { |
1dcdce59 YB |
4585 | int err; |
4586 | ||
152a7b0a | 4587 | ext4_set_inode_state(inode, EXT4_STATE_XATTR); |
1dcdce59 YB |
4588 | err = ext4_find_inline_data_nolock(inode); |
4589 | if (!err && ext4_has_inline_data(inode)) | |
4590 | ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA); | |
4591 | return err; | |
f19d5870 TM |
4592 | } else |
4593 | EXT4_I(inode)->i_inline_off = 0; | |
eb9b5f01 | 4594 | return 0; |
152a7b0a TM |
4595 | } |
4596 | ||
040cb378 LX |
4597 | int ext4_get_projid(struct inode *inode, kprojid_t *projid) |
4598 | { | |
0b7b7779 | 4599 | if (!ext4_has_feature_project(inode->i_sb)) |
040cb378 LX |
4600 | return -EOPNOTSUPP; |
4601 | *projid = EXT4_I(inode)->i_projid; | |
4602 | return 0; | |
4603 | } | |
4604 | ||
e254d1af EG |
4605 | /* |
4606 | * ext4 has self-managed i_version for ea inodes, it stores the lower 32bit of | |
4607 | * refcount in i_version, so use raw values if inode has EXT4_EA_INODE_FL flag | |
4608 | * set. | |
4609 | */ | |
4610 | static inline void ext4_inode_set_iversion_queried(struct inode *inode, u64 val) | |
4611 | { | |
4612 | if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) | |
4613 | inode_set_iversion_raw(inode, val); | |
4614 | else | |
4615 | inode_set_iversion_queried(inode, val); | |
4616 | } | |
e254d1af | 4617 | |
b3e6bcb9 TT |
4618 | static const char *check_igot_inode(struct inode *inode, ext4_iget_flags flags) |
4619 | ||
4620 | { | |
4621 | if (flags & EXT4_IGET_EA_INODE) { | |
4622 | if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) | |
4623 | return "missing EA_INODE flag"; | |
2bc7e7c1 TT |
4624 | if (ext4_test_inode_state(inode, EXT4_STATE_XATTR) || |
4625 | EXT4_I(inode)->i_file_acl) | |
4626 | return "ea_inode with extended attributes"; | |
b3e6bcb9 TT |
4627 | } else { |
4628 | if ((EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) | |
4629 | return "unexpected EA_INODE flag"; | |
4630 | } | |
4631 | if (is_bad_inode(inode) && !(flags & EXT4_IGET_BAD)) | |
4632 | return "unexpected bad inode w/o EXT4_IGET_BAD"; | |
4633 | return NULL; | |
4634 | } | |
4635 | ||
8a363970 TT |
4636 | struct inode *__ext4_iget(struct super_block *sb, unsigned long ino, |
4637 | ext4_iget_flags flags, const char *function, | |
4638 | unsigned int line) | |
ac27a0ec | 4639 | { |
617ba13b MC |
4640 | struct ext4_iloc iloc; |
4641 | struct ext4_inode *raw_inode; | |
1d1fe1ee | 4642 | struct ext4_inode_info *ei; |
bd2c38cf | 4643 | struct ext4_super_block *es = EXT4_SB(sb)->s_es; |
1d1fe1ee | 4644 | struct inode *inode; |
b3e6bcb9 | 4645 | const char *err_str; |
b436b9be | 4646 | journal_t *journal = EXT4_SB(sb)->s_journal; |
1d1fe1ee | 4647 | long ret; |
7e6e1ef4 | 4648 | loff_t size; |
ac27a0ec | 4649 | int block; |
08cefc7a EB |
4650 | uid_t i_uid; |
4651 | gid_t i_gid; | |
040cb378 | 4652 | projid_t i_projid; |
ac27a0ec | 4653 | |
191ce178 | 4654 | if ((!(flags & EXT4_IGET_SPECIAL) && |
bd2c38cf JK |
4655 | ((ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO) || |
4656 | ino == le32_to_cpu(es->s_usr_quota_inum) || | |
4657 | ino == le32_to_cpu(es->s_grp_quota_inum) || | |
02f310fc JK |
4658 | ino == le32_to_cpu(es->s_prj_quota_inum) || |
4659 | ino == le32_to_cpu(es->s_orphan_file_inum))) || | |
8a363970 | 4660 | (ino < EXT4_ROOT_INO) || |
bd2c38cf | 4661 | (ino > le32_to_cpu(es->s_inodes_count))) { |
8a363970 TT |
4662 | if (flags & EXT4_IGET_HANDLE) |
4663 | return ERR_PTR(-ESTALE); | |
014c9caa | 4664 | __ext4_error(sb, function, line, false, EFSCORRUPTED, 0, |
8a363970 TT |
4665 | "inode #%lu: comm %s: iget: illegal inode #", |
4666 | ino, current->comm); | |
4667 | return ERR_PTR(-EFSCORRUPTED); | |
4668 | } | |
4669 | ||
1d1fe1ee DH |
4670 | inode = iget_locked(sb, ino); |
4671 | if (!inode) | |
4672 | return ERR_PTR(-ENOMEM); | |
b3e6bcb9 TT |
4673 | if (!(inode->i_state & I_NEW)) { |
4674 | if ((err_str = check_igot_inode(inode, flags)) != NULL) { | |
4675 | ext4_error_inode(inode, function, line, 0, err_str); | |
4676 | iput(inode); | |
4677 | return ERR_PTR(-EFSCORRUPTED); | |
4678 | } | |
1d1fe1ee | 4679 | return inode; |
b3e6bcb9 | 4680 | } |
1d1fe1ee DH |
4681 | |
4682 | ei = EXT4_I(inode); | |
7dc57615 | 4683 | iloc.bh = NULL; |
ac27a0ec | 4684 | |
8016e29f | 4685 | ret = __ext4_get_inode_loc_noinmem(inode, &iloc); |
1d1fe1ee | 4686 | if (ret < 0) |
ac27a0ec | 4687 | goto bad_inode; |
617ba13b | 4688 | raw_inode = ext4_raw_inode(&iloc); |
814525f4 | 4689 | |
8a363970 TT |
4690 | if ((flags & EXT4_IGET_HANDLE) && |
4691 | (raw_inode->i_links_count == 0) && (raw_inode->i_mode == 0)) { | |
4692 | ret = -ESTALE; | |
4693 | goto bad_inode; | |
4694 | } | |
4695 | ||
814525f4 DW |
4696 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
4697 | ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize); | |
4698 | if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > | |
2dc8d9e1 EB |
4699 | EXT4_INODE_SIZE(inode->i_sb) || |
4700 | (ei->i_extra_isize & 3)) { | |
8a363970 TT |
4701 | ext4_error_inode(inode, function, line, 0, |
4702 | "iget: bad extra_isize %u " | |
4703 | "(inode size %u)", | |
2dc8d9e1 EB |
4704 | ei->i_extra_isize, |
4705 | EXT4_INODE_SIZE(inode->i_sb)); | |
6a797d27 | 4706 | ret = -EFSCORRUPTED; |
814525f4 DW |
4707 | goto bad_inode; |
4708 | } | |
4709 | } else | |
4710 | ei->i_extra_isize = 0; | |
4711 | ||
4712 | /* Precompute checksum seed for inode metadata */ | |
9aa5d32b | 4713 | if (ext4_has_metadata_csum(sb)) { |
814525f4 DW |
4714 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
4715 | __u32 csum; | |
4716 | __le32 inum = cpu_to_le32(inode->i_ino); | |
4717 | __le32 gen = raw_inode->i_generation; | |
4718 | csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum, | |
4719 | sizeof(inum)); | |
4720 | ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen, | |
4721 | sizeof(gen)); | |
4722 | } | |
4723 | ||
8016e29f HS |
4724 | if ((!ext4_inode_csum_verify(inode, raw_inode, ei) || |
4725 | ext4_simulate_fail(sb, EXT4_SIM_INODE_CRC)) && | |
4726 | (!(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY))) { | |
4727 | ext4_error_inode_err(inode, function, line, 0, | |
4728 | EFSBADCRC, "iget: checksum invalid"); | |
6a797d27 | 4729 | ret = -EFSBADCRC; |
814525f4 DW |
4730 | goto bad_inode; |
4731 | } | |
4732 | ||
ac27a0ec | 4733 | inode->i_mode = le16_to_cpu(raw_inode->i_mode); |
08cefc7a EB |
4734 | i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low); |
4735 | i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low); | |
0b7b7779 | 4736 | if (ext4_has_feature_project(sb) && |
040cb378 LX |
4737 | EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE && |
4738 | EXT4_FITS_IN_INODE(raw_inode, ei, i_projid)) | |
4739 | i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid); | |
4740 | else | |
4741 | i_projid = EXT4_DEF_PROJID; | |
4742 | ||
af5bc92d | 4743 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
08cefc7a EB |
4744 | i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16; |
4745 | i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16; | |
ac27a0ec | 4746 | } |
08cefc7a EB |
4747 | i_uid_write(inode, i_uid); |
4748 | i_gid_write(inode, i_gid); | |
040cb378 | 4749 | ei->i_projid = make_kprojid(&init_user_ns, i_projid); |
bfe86848 | 4750 | set_nlink(inode, le16_to_cpu(raw_inode->i_links_count)); |
ac27a0ec | 4751 | |
353eb83c | 4752 | ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */ |
67cf5b09 | 4753 | ei->i_inline_off = 0; |
ac27a0ec DK |
4754 | ei->i_dir_start_lookup = 0; |
4755 | ei->i_dtime = le32_to_cpu(raw_inode->i_dtime); | |
4756 | /* We now have enough fields to check if the inode was active or not. | |
4757 | * This is needed because nfsd might try to access dead inodes | |
4758 | * the test is that same one that e2fsck uses | |
4759 | * NeilBrown 1999oct15 | |
4760 | */ | |
4761 | if (inode->i_nlink == 0) { | |
5cd74028 | 4762 | if ((inode->i_mode == 0 || flags & EXT4_IGET_SPECIAL || |
393d1d1d DTB |
4763 | !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) && |
4764 | ino != EXT4_BOOT_LOADER_INO) { | |
5cd74028 BL |
4765 | /* this inode is deleted or unallocated */ |
4766 | if (flags & EXT4_IGET_SPECIAL) { | |
4767 | ext4_error_inode(inode, function, line, 0, | |
4768 | "iget: special inode unallocated"); | |
4769 | ret = -EFSCORRUPTED; | |
4770 | } else | |
4771 | ret = -ESTALE; | |
ac27a0ec DK |
4772 | goto bad_inode; |
4773 | } | |
4774 | /* The only unlinked inodes we let through here have | |
4775 | * valid i_mode and are being read by the orphan | |
4776 | * recovery code: that's fine, we're about to complete | |
393d1d1d DTB |
4777 | * the process of deleting those. |
4778 | * OR it is the EXT4_BOOT_LOADER_INO which is | |
4779 | * not initialized on a new filesystem. */ | |
ac27a0ec | 4780 | } |
ac27a0ec | 4781 | ei->i_flags = le32_to_cpu(raw_inode->i_flags); |
043546e4 | 4782 | ext4_set_inode_flags(inode, true); |
0fc1b451 | 4783 | inode->i_blocks = ext4_inode_blocks(raw_inode, ei); |
7973c0c1 | 4784 | ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo); |
e2b911c5 | 4785 | if (ext4_has_feature_64bit(sb)) |
a1ddeb7e BP |
4786 | ei->i_file_acl |= |
4787 | ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32; | |
e08ac99f | 4788 | inode->i_size = ext4_isize(sb, raw_inode); |
7e6e1ef4 | 4789 | if ((size = i_size_read(inode)) < 0) { |
8a363970 TT |
4790 | ext4_error_inode(inode, function, line, 0, |
4791 | "iget: bad i_size value: %lld", size); | |
7e6e1ef4 DW |
4792 | ret = -EFSCORRUPTED; |
4793 | goto bad_inode; | |
4794 | } | |
48a34311 JK |
4795 | /* |
4796 | * If dir_index is not enabled but there's dir with INDEX flag set, | |
4797 | * we'd normally treat htree data as empty space. But with metadata | |
4798 | * checksumming that corrupts checksums so forbid that. | |
4799 | */ | |
4800 | if (!ext4_has_feature_dir_index(sb) && ext4_has_metadata_csum(sb) && | |
4801 | ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) { | |
4802 | ext4_error_inode(inode, function, line, 0, | |
4803 | "iget: Dir with htree data on filesystem without dir_index feature."); | |
4804 | ret = -EFSCORRUPTED; | |
4805 | goto bad_inode; | |
4806 | } | |
ac27a0ec | 4807 | ei->i_disksize = inode->i_size; |
a9e7f447 DM |
4808 | #ifdef CONFIG_QUOTA |
4809 | ei->i_reserved_quota = 0; | |
4810 | #endif | |
ac27a0ec DK |
4811 | inode->i_generation = le32_to_cpu(raw_inode->i_generation); |
4812 | ei->i_block_group = iloc.block_group; | |
a4912123 | 4813 | ei->i_last_alloc_group = ~0; |
ac27a0ec DK |
4814 | /* |
4815 | * NOTE! The in-memory inode i_data array is in little-endian order | |
4816 | * even on big-endian machines: we do NOT byteswap the block numbers! | |
4817 | */ | |
617ba13b | 4818 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
ac27a0ec DK |
4819 | ei->i_data[block] = raw_inode->i_block[block]; |
4820 | INIT_LIST_HEAD(&ei->i_orphan); | |
aa75f4d3 | 4821 | ext4_fc_init_inode(&ei->vfs_inode); |
ac27a0ec | 4822 | |
b436b9be JK |
4823 | /* |
4824 | * Set transaction id's of transactions that have to be committed | |
4825 | * to finish f[data]sync. We set them to currently running transaction | |
4826 | * as we cannot be sure that the inode or some of its metadata isn't | |
4827 | * part of the transaction - the inode could have been reclaimed and | |
4828 | * now it is reread from disk. | |
4829 | */ | |
4830 | if (journal) { | |
4831 | transaction_t *transaction; | |
4832 | tid_t tid; | |
4833 | ||
a931da6a | 4834 | read_lock(&journal->j_state_lock); |
b436b9be JK |
4835 | if (journal->j_running_transaction) |
4836 | transaction = journal->j_running_transaction; | |
4837 | else | |
4838 | transaction = journal->j_committing_transaction; | |
4839 | if (transaction) | |
4840 | tid = transaction->t_tid; | |
4841 | else | |
4842 | tid = journal->j_commit_sequence; | |
a931da6a | 4843 | read_unlock(&journal->j_state_lock); |
b436b9be JK |
4844 | ei->i_sync_tid = tid; |
4845 | ei->i_datasync_tid = tid; | |
4846 | } | |
4847 | ||
0040d987 | 4848 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
ac27a0ec DK |
4849 | if (ei->i_extra_isize == 0) { |
4850 | /* The extra space is currently unused. Use it. */ | |
2dc8d9e1 | 4851 | BUILD_BUG_ON(sizeof(struct ext4_inode) & 3); |
617ba13b MC |
4852 | ei->i_extra_isize = sizeof(struct ext4_inode) - |
4853 | EXT4_GOOD_OLD_INODE_SIZE; | |
ac27a0ec | 4854 | } else { |
eb9b5f01 TT |
4855 | ret = ext4_iget_extra_inode(inode, raw_inode, ei); |
4856 | if (ret) | |
4857 | goto bad_inode; | |
ac27a0ec | 4858 | } |
814525f4 | 4859 | } |
ac27a0ec | 4860 | |
ef7f3835 KS |
4861 | EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode); |
4862 | EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode); | |
4863 | EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode); | |
4864 | EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode); | |
4865 | ||
ed3654eb | 4866 | if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) { |
ee73f9a5 JL |
4867 | u64 ivers = le32_to_cpu(raw_inode->i_disk_version); |
4868 | ||
c4f65706 TT |
4869 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
4870 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
ee73f9a5 | 4871 | ivers |= |
c4f65706 TT |
4872 | (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32; |
4873 | } | |
e254d1af | 4874 | ext4_inode_set_iversion_queried(inode, ivers); |
25ec56b5 JNC |
4875 | } |
4876 | ||
c4b5a614 | 4877 | ret = 0; |
485c26ec | 4878 | if (ei->i_file_acl && |
ce9f24cc | 4879 | !ext4_inode_block_valid(inode, ei->i_file_acl, 1)) { |
8a363970 TT |
4880 | ext4_error_inode(inode, function, line, 0, |
4881 | "iget: bad extended attribute block %llu", | |
24676da4 | 4882 | ei->i_file_acl); |
6a797d27 | 4883 | ret = -EFSCORRUPTED; |
485c26ec | 4884 | goto bad_inode; |
f19d5870 | 4885 | } else if (!ext4_has_inline_data(inode)) { |
bc716523 | 4886 | /* validate the block references in the inode */ |
8016e29f HS |
4887 | if (!(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY) && |
4888 | (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || | |
4889 | (S_ISLNK(inode->i_mode) && | |
4890 | !ext4_inode_is_fast_symlink(inode)))) { | |
bc716523 | 4891 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
f19d5870 | 4892 | ret = ext4_ext_check_inode(inode); |
bc716523 LS |
4893 | else |
4894 | ret = ext4_ind_check_inode(inode); | |
f19d5870 | 4895 | } |
fe2c8191 | 4896 | } |
567f3e9a | 4897 | if (ret) |
de9a55b8 | 4898 | goto bad_inode; |
7a262f7c | 4899 | |
ac27a0ec | 4900 | if (S_ISREG(inode->i_mode)) { |
617ba13b | 4901 | inode->i_op = &ext4_file_inode_operations; |
be64f884 | 4902 | inode->i_fop = &ext4_file_operations; |
617ba13b | 4903 | ext4_set_aops(inode); |
ac27a0ec | 4904 | } else if (S_ISDIR(inode->i_mode)) { |
617ba13b MC |
4905 | inode->i_op = &ext4_dir_inode_operations; |
4906 | inode->i_fop = &ext4_dir_operations; | |
ac27a0ec | 4907 | } else if (S_ISLNK(inode->i_mode)) { |
6390d33b LR |
4908 | /* VFS does not allow setting these so must be corruption */ |
4909 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) { | |
8a363970 TT |
4910 | ext4_error_inode(inode, function, line, 0, |
4911 | "iget: immutable or append flags " | |
4912 | "not allowed on symlinks"); | |
6390d33b LR |
4913 | ret = -EFSCORRUPTED; |
4914 | goto bad_inode; | |
4915 | } | |
592ddec7 | 4916 | if (IS_ENCRYPTED(inode)) { |
a7a67e8a | 4917 | inode->i_op = &ext4_encrypted_symlink_inode_operations; |
a7a67e8a | 4918 | } else if (ext4_inode_is_fast_symlink(inode)) { |
75e7566b | 4919 | inode->i_link = (char *)ei->i_data; |
617ba13b | 4920 | inode->i_op = &ext4_fast_symlink_inode_operations; |
e83c1397 DG |
4921 | nd_terminate_link(ei->i_data, inode->i_size, |
4922 | sizeof(ei->i_data) - 1); | |
4923 | } else { | |
617ba13b | 4924 | inode->i_op = &ext4_symlink_inode_operations; |
ac27a0ec | 4925 | } |
563bdd61 TT |
4926 | } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) || |
4927 | S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) { | |
617ba13b | 4928 | inode->i_op = &ext4_special_inode_operations; |
ac27a0ec DK |
4929 | if (raw_inode->i_block[0]) |
4930 | init_special_inode(inode, inode->i_mode, | |
4931 | old_decode_dev(le32_to_cpu(raw_inode->i_block[0]))); | |
4932 | else | |
4933 | init_special_inode(inode, inode->i_mode, | |
4934 | new_decode_dev(le32_to_cpu(raw_inode->i_block[1]))); | |
393d1d1d DTB |
4935 | } else if (ino == EXT4_BOOT_LOADER_INO) { |
4936 | make_bad_inode(inode); | |
563bdd61 | 4937 | } else { |
6a797d27 | 4938 | ret = -EFSCORRUPTED; |
8a363970 TT |
4939 | ext4_error_inode(inode, function, line, 0, |
4940 | "iget: bogus i_mode (%o)", inode->i_mode); | |
563bdd61 | 4941 | goto bad_inode; |
ac27a0ec | 4942 | } |
6456ca65 TT |
4943 | if (IS_CASEFOLDED(inode) && !ext4_has_feature_casefold(inode->i_sb)) |
4944 | ext4_error_inode(inode, function, line, 0, | |
4945 | "casefold flag without casefold feature"); | |
b3e6bcb9 TT |
4946 | if ((err_str = check_igot_inode(inode, flags)) != NULL) { |
4947 | ext4_error_inode(inode, function, line, 0, err_str); | |
4948 | ret = -EFSCORRUPTED; | |
63b1e9bc BL |
4949 | goto bad_inode; |
4950 | } | |
dec214d0 | 4951 | |
63b1e9bc | 4952 | brelse(iloc.bh); |
1d1fe1ee DH |
4953 | unlock_new_inode(inode); |
4954 | return inode; | |
ac27a0ec DK |
4955 | |
4956 | bad_inode: | |
567f3e9a | 4957 | brelse(iloc.bh); |
1d1fe1ee DH |
4958 | iget_failed(inode); |
4959 | return ERR_PTR(ret); | |
ac27a0ec DK |
4960 | } |
4961 | ||
3f19b2ab DH |
4962 | static void __ext4_update_other_inode_time(struct super_block *sb, |
4963 | unsigned long orig_ino, | |
4964 | unsigned long ino, | |
4965 | struct ext4_inode *raw_inode) | |
a26f4992 | 4966 | { |
3f19b2ab DH |
4967 | struct inode *inode; |
4968 | ||
4969 | inode = find_inode_by_ino_rcu(sb, ino); | |
4970 | if (!inode) | |
4971 | return; | |
a26f4992 | 4972 | |
ed296c6c | 4973 | if (!inode_is_dirtytime_only(inode)) |
3f19b2ab DH |
4974 | return; |
4975 | ||
a26f4992 | 4976 | spin_lock(&inode->i_lock); |
ed296c6c | 4977 | if (inode_is_dirtytime_only(inode)) { |
a26f4992 TT |
4978 | struct ext4_inode_info *ei = EXT4_I(inode); |
4979 | ||
5fcd5750 | 4980 | inode->i_state &= ~I_DIRTY_TIME; |
a26f4992 TT |
4981 | spin_unlock(&inode->i_lock); |
4982 | ||
4983 | spin_lock(&ei->i_raw_lock); | |
3f19b2ab DH |
4984 | EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode); |
4985 | EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode); | |
4986 | EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode); | |
4987 | ext4_inode_csum_set(inode, raw_inode, ei); | |
a26f4992 | 4988 | spin_unlock(&ei->i_raw_lock); |
3f19b2ab DH |
4989 | trace_ext4_other_inode_update_time(inode, orig_ino); |
4990 | return; | |
a26f4992 TT |
4991 | } |
4992 | spin_unlock(&inode->i_lock); | |
a26f4992 TT |
4993 | } |
4994 | ||
4995 | /* | |
4996 | * Opportunistically update the other time fields for other inodes in | |
4997 | * the same inode table block. | |
4998 | */ | |
4999 | static void ext4_update_other_inodes_time(struct super_block *sb, | |
5000 | unsigned long orig_ino, char *buf) | |
5001 | { | |
a26f4992 TT |
5002 | unsigned long ino; |
5003 | int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; | |
5004 | int inode_size = EXT4_INODE_SIZE(sb); | |
5005 | ||
0f0ff9a9 TT |
5006 | /* |
5007 | * Calculate the first inode in the inode table block. Inode | |
5008 | * numbers are one-based. That is, the first inode in a block | |
5009 | * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1). | |
5010 | */ | |
5011 | ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1; | |
3f19b2ab | 5012 | rcu_read_lock(); |
a26f4992 TT |
5013 | for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) { |
5014 | if (ino == orig_ino) | |
5015 | continue; | |
3f19b2ab DH |
5016 | __ext4_update_other_inode_time(sb, orig_ino, ino, |
5017 | (struct ext4_inode *)buf); | |
a26f4992 | 5018 | } |
3f19b2ab | 5019 | rcu_read_unlock(); |
a26f4992 TT |
5020 | } |
5021 | ||
664bd38b ZY |
5022 | /* |
5023 | * Post the struct inode info into an on-disk inode location in the | |
5024 | * buffer-cache. This gobbles the caller's reference to the | |
5025 | * buffer_head in the inode location struct. | |
5026 | * | |
5027 | * The caller must have write access to iloc->bh. | |
5028 | */ | |
5029 | static int ext4_do_update_inode(handle_t *handle, | |
5030 | struct inode *inode, | |
5031 | struct ext4_iloc *iloc) | |
5032 | { | |
5033 | struct ext4_inode *raw_inode = ext4_raw_inode(iloc); | |
5034 | struct ext4_inode_info *ei = EXT4_I(inode); | |
5035 | struct buffer_head *bh = iloc->bh; | |
5036 | struct super_block *sb = inode->i_sb; | |
5037 | int err; | |
5038 | int need_datasync = 0, set_large_file = 0; | |
5039 | ||
5040 | spin_lock(&ei->i_raw_lock); | |
5041 | ||
5042 | /* | |
5043 | * For fields not tracked in the in-memory inode, initialise them | |
5044 | * to zero for new inodes. | |
5045 | */ | |
5046 | if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) | |
5047 | memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size); | |
5048 | ||
5049 | if (READ_ONCE(ei->i_disksize) != ext4_isize(inode->i_sb, raw_inode)) | |
5050 | need_datasync = 1; | |
5051 | if (ei->i_disksize > 0x7fffffffULL) { | |
5052 | if (!ext4_has_feature_large_file(sb) || | |
5053 | EXT4_SB(sb)->s_es->s_rev_level == cpu_to_le32(EXT4_GOOD_OLD_REV)) | |
5054 | set_large_file = 1; | |
5055 | } | |
5056 | ||
5057 | err = ext4_fill_raw_inode(inode, raw_inode); | |
202ee5df | 5058 | spin_unlock(&ei->i_raw_lock); |
baaae979 ZY |
5059 | if (err) { |
5060 | EXT4_ERROR_INODE(inode, "corrupted inode contents"); | |
5061 | goto out_brelse; | |
5062 | } | |
5063 | ||
1751e8a6 | 5064 | if (inode->i_sb->s_flags & SB_LAZYTIME) |
a26f4992 TT |
5065 | ext4_update_other_inodes_time(inode->i_sb, inode->i_ino, |
5066 | bh->b_data); | |
202ee5df | 5067 | |
830156c7 | 5068 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
7d8bd3c7 SL |
5069 | err = ext4_handle_dirty_metadata(handle, NULL, bh); |
5070 | if (err) | |
baaae979 | 5071 | goto out_error; |
19f5fb7a | 5072 | ext4_clear_inode_state(inode, EXT4_STATE_NEW); |
202ee5df | 5073 | if (set_large_file) { |
5d601255 | 5074 | BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access"); |
188c299e JK |
5075 | err = ext4_journal_get_write_access(handle, sb, |
5076 | EXT4_SB(sb)->s_sbh, | |
5077 | EXT4_JTR_NONE); | |
202ee5df | 5078 | if (err) |
baaae979 | 5079 | goto out_error; |
05c2c00f | 5080 | lock_buffer(EXT4_SB(sb)->s_sbh); |
e2b911c5 | 5081 | ext4_set_feature_large_file(sb); |
05c2c00f JK |
5082 | ext4_superblock_csum_set(sb); |
5083 | unlock_buffer(EXT4_SB(sb)->s_sbh); | |
202ee5df | 5084 | ext4_handle_sync(handle); |
a3f5cf14 JK |
5085 | err = ext4_handle_dirty_metadata(handle, NULL, |
5086 | EXT4_SB(sb)->s_sbh); | |
202ee5df | 5087 | } |
b71fc079 | 5088 | ext4_update_inode_fsync_trans(handle, inode, need_datasync); |
baaae979 ZY |
5089 | out_error: |
5090 | ext4_std_error(inode->i_sb, err); | |
ac27a0ec | 5091 | out_brelse: |
af5bc92d | 5092 | brelse(bh); |
ac27a0ec DK |
5093 | return err; |
5094 | } | |
5095 | ||
5096 | /* | |
617ba13b | 5097 | * ext4_write_inode() |
ac27a0ec DK |
5098 | * |
5099 | * We are called from a few places: | |
5100 | * | |
87f7e416 | 5101 | * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files. |
ac27a0ec | 5102 | * Here, there will be no transaction running. We wait for any running |
4907cb7b | 5103 | * transaction to commit. |
ac27a0ec | 5104 | * |
87f7e416 TT |
5105 | * - Within flush work (sys_sync(), kupdate and such). |
5106 | * We wait on commit, if told to. | |
ac27a0ec | 5107 | * |
87f7e416 TT |
5108 | * - Within iput_final() -> write_inode_now() |
5109 | * We wait on commit, if told to. | |
ac27a0ec DK |
5110 | * |
5111 | * In all cases it is actually safe for us to return without doing anything, | |
5112 | * because the inode has been copied into a raw inode buffer in | |
87f7e416 TT |
5113 | * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL |
5114 | * writeback. | |
ac27a0ec DK |
5115 | * |
5116 | * Note that we are absolutely dependent upon all inode dirtiers doing the | |
5117 | * right thing: they *must* call mark_inode_dirty() after dirtying info in | |
5118 | * which we are interested. | |
5119 | * | |
5120 | * It would be a bug for them to not do this. The code: | |
5121 | * | |
5122 | * mark_inode_dirty(inode) | |
5123 | * stuff(); | |
5124 | * inode->i_size = expr; | |
5125 | * | |
87f7e416 TT |
5126 | * is in error because write_inode() could occur while `stuff()' is running, |
5127 | * and the new i_size will be lost. Plus the inode will no longer be on the | |
5128 | * superblock's dirty inode list. | |
ac27a0ec | 5129 | */ |
a9185b41 | 5130 | int ext4_write_inode(struct inode *inode, struct writeback_control *wbc) |
ac27a0ec | 5131 | { |
91ac6f43 FM |
5132 | int err; |
5133 | ||
18f2c4fc TT |
5134 | if (WARN_ON_ONCE(current->flags & PF_MEMALLOC) || |
5135 | sb_rdonly(inode->i_sb)) | |
ac27a0ec DK |
5136 | return 0; |
5137 | ||
18f2c4fc TT |
5138 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
5139 | return -EIO; | |
5140 | ||
91ac6f43 FM |
5141 | if (EXT4_SB(inode->i_sb)->s_journal) { |
5142 | if (ext4_journal_current_handle()) { | |
4978c659 | 5143 | ext4_debug("called recursively, non-PF_MEMALLOC!\n"); |
91ac6f43 FM |
5144 | dump_stack(); |
5145 | return -EIO; | |
5146 | } | |
ac27a0ec | 5147 | |
10542c22 JK |
5148 | /* |
5149 | * No need to force transaction in WB_SYNC_NONE mode. Also | |
5150 | * ext4_sync_fs() will force the commit after everything is | |
5151 | * written. | |
5152 | */ | |
5153 | if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync) | |
91ac6f43 FM |
5154 | return 0; |
5155 | ||
aa75f4d3 | 5156 | err = ext4_fc_commit(EXT4_SB(inode->i_sb)->s_journal, |
18f2c4fc | 5157 | EXT4_I(inode)->i_sync_tid); |
91ac6f43 FM |
5158 | } else { |
5159 | struct ext4_iloc iloc; | |
ac27a0ec | 5160 | |
8016e29f | 5161 | err = __ext4_get_inode_loc_noinmem(inode, &iloc); |
91ac6f43 FM |
5162 | if (err) |
5163 | return err; | |
10542c22 JK |
5164 | /* |
5165 | * sync(2) will flush the whole buffer cache. No need to do | |
5166 | * it here separately for each inode. | |
5167 | */ | |
5168 | if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) | |
830156c7 FM |
5169 | sync_dirty_buffer(iloc.bh); |
5170 | if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) { | |
54d3adbc TT |
5171 | ext4_error_inode_block(inode, iloc.bh->b_blocknr, EIO, |
5172 | "IO error syncing inode"); | |
830156c7 FM |
5173 | err = -EIO; |
5174 | } | |
fd2dd9fb | 5175 | brelse(iloc.bh); |
91ac6f43 FM |
5176 | } |
5177 | return err; | |
ac27a0ec DK |
5178 | } |
5179 | ||
53e87268 | 5180 | /* |
ccd16945 MWO |
5181 | * In data=journal mode ext4_journalled_invalidate_folio() may fail to invalidate |
5182 | * buffers that are attached to a folio straddling i_size and are undergoing | |
53e87268 JK |
5183 | * commit. In that case we have to wait for commit to finish and try again. |
5184 | */ | |
5185 | static void ext4_wait_for_tail_page_commit(struct inode *inode) | |
5186 | { | |
53e87268 JK |
5187 | unsigned offset; |
5188 | journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; | |
5189 | tid_t commit_tid = 0; | |
5190 | int ret; | |
5191 | ||
09cbfeaf | 5192 | offset = inode->i_size & (PAGE_SIZE - 1); |
53e87268 | 5193 | /* |
ccd16945 MWO |
5194 | * If the folio is fully truncated, we don't need to wait for any commit |
5195 | * (and we even should not as __ext4_journalled_invalidate_folio() may | |
5196 | * strip all buffers from the folio but keep the folio dirty which can then | |
3f079114 | 5197 | * confuse e.g. concurrent ext4_writepages() seeing dirty folio without |
565333a1 | 5198 | * buffers). Also we don't need to wait for any commit if all buffers in |
ccd16945 | 5199 | * the folio remain valid. This is most beneficial for the common case of |
565333a1 | 5200 | * blocksize == PAGESIZE. |
53e87268 | 5201 | */ |
565333a1 | 5202 | if (!offset || offset > (PAGE_SIZE - i_blocksize(inode))) |
53e87268 JK |
5203 | return; |
5204 | while (1) { | |
ccd16945 | 5205 | struct folio *folio = filemap_lock_folio(inode->i_mapping, |
09cbfeaf | 5206 | inode->i_size >> PAGE_SHIFT); |
66dabbb6 | 5207 | if (IS_ERR(folio)) |
53e87268 | 5208 | return; |
ccd16945 MWO |
5209 | ret = __ext4_journalled_invalidate_folio(folio, offset, |
5210 | folio_size(folio) - offset); | |
5211 | folio_unlock(folio); | |
5212 | folio_put(folio); | |
53e87268 JK |
5213 | if (ret != -EBUSY) |
5214 | return; | |
5215 | commit_tid = 0; | |
5216 | read_lock(&journal->j_state_lock); | |
5217 | if (journal->j_committing_transaction) | |
5218 | commit_tid = journal->j_committing_transaction->t_tid; | |
5219 | read_unlock(&journal->j_state_lock); | |
5220 | if (commit_tid) | |
5221 | jbd2_log_wait_commit(journal, commit_tid); | |
5222 | } | |
5223 | } | |
5224 | ||
ac27a0ec | 5225 | /* |
617ba13b | 5226 | * ext4_setattr() |
ac27a0ec DK |
5227 | * |
5228 | * Called from notify_change. | |
5229 | * | |
5230 | * We want to trap VFS attempts to truncate the file as soon as | |
5231 | * possible. In particular, we want to make sure that when the VFS | |
5232 | * shrinks i_size, we put the inode on the orphan list and modify | |
5233 | * i_disksize immediately, so that during the subsequent flushing of | |
5234 | * dirty pages and freeing of disk blocks, we can guarantee that any | |
5235 | * commit will leave the blocks being flushed in an unused state on | |
5236 | * disk. (On recovery, the inode will get truncated and the blocks will | |
5237 | * be freed, so we have a strong guarantee that no future commit will | |
5238 | * leave these blocks visible to the user.) | |
5239 | * | |
678aaf48 JK |
5240 | * Another thing we have to assure is that if we are in ordered mode |
5241 | * and inode is still attached to the committing transaction, we must | |
5242 | * we start writeout of all the dirty pages which are being truncated. | |
5243 | * This way we are sure that all the data written in the previous | |
5244 | * transaction are already on disk (truncate waits for pages under | |
5245 | * writeback). | |
5246 | * | |
f340b3d9 | 5247 | * Called with inode->i_rwsem down. |
ac27a0ec | 5248 | */ |
c1632a0f | 5249 | int ext4_setattr(struct mnt_idmap *idmap, struct dentry *dentry, |
549c7297 | 5250 | struct iattr *attr) |
ac27a0ec | 5251 | { |
2b0143b5 | 5252 | struct inode *inode = d_inode(dentry); |
ac27a0ec | 5253 | int error, rc = 0; |
3d287de3 | 5254 | int orphan = 0; |
ac27a0ec | 5255 | const unsigned int ia_valid = attr->ia_valid; |
a642c2c0 | 5256 | bool inc_ivers = true; |
ac27a0ec | 5257 | |
0db1ff22 TT |
5258 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
5259 | return -EIO; | |
5260 | ||
02b016ca TT |
5261 | if (unlikely(IS_IMMUTABLE(inode))) |
5262 | return -EPERM; | |
5263 | ||
5264 | if (unlikely(IS_APPEND(inode) && | |
5265 | (ia_valid & (ATTR_MODE | ATTR_UID | | |
5266 | ATTR_GID | ATTR_TIMES_SET)))) | |
5267 | return -EPERM; | |
5268 | ||
c1632a0f | 5269 | error = setattr_prepare(idmap, dentry, attr); |
ac27a0ec DK |
5270 | if (error) |
5271 | return error; | |
5272 | ||
3ce2b8dd EB |
5273 | error = fscrypt_prepare_setattr(dentry, attr); |
5274 | if (error) | |
5275 | return error; | |
5276 | ||
c93d8f88 EB |
5277 | error = fsverity_prepare_setattr(dentry, attr); |
5278 | if (error) | |
5279 | return error; | |
5280 | ||
f861646a | 5281 | if (is_quota_modification(idmap, inode, attr)) { |
a7cdadee JK |
5282 | error = dquot_initialize(inode); |
5283 | if (error) | |
5284 | return error; | |
5285 | } | |
2729cfdc | 5286 | |
0dbe12f2 CB |
5287 | if (i_uid_needs_update(idmap, attr, inode) || |
5288 | i_gid_needs_update(idmap, attr, inode)) { | |
ac27a0ec DK |
5289 | handle_t *handle; |
5290 | ||
5291 | /* (user+group)*(old+new) structure, inode write (sb, | |
5292 | * inode block, ? - but truncate inode update has it) */ | |
9924a92a TT |
5293 | handle = ext4_journal_start(inode, EXT4_HT_QUOTA, |
5294 | (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) + | |
5295 | EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3); | |
ac27a0ec DK |
5296 | if (IS_ERR(handle)) { |
5297 | error = PTR_ERR(handle); | |
5298 | goto err_out; | |
5299 | } | |
7a9ca53a TE |
5300 | |
5301 | /* dquot_transfer() calls back ext4_get_inode_usage() which | |
5302 | * counts xattr inode references. | |
5303 | */ | |
5304 | down_read(&EXT4_I(inode)->xattr_sem); | |
f861646a | 5305 | error = dquot_transfer(idmap, inode, attr); |
7a9ca53a TE |
5306 | up_read(&EXT4_I(inode)->xattr_sem); |
5307 | ||
ac27a0ec | 5308 | if (error) { |
617ba13b | 5309 | ext4_journal_stop(handle); |
ac27a0ec DK |
5310 | return error; |
5311 | } | |
5312 | /* Update corresponding info in inode so that everything is in | |
5313 | * one transaction */ | |
0dbe12f2 CB |
5314 | i_uid_update(idmap, attr, inode); |
5315 | i_gid_update(idmap, attr, inode); | |
617ba13b MC |
5316 | error = ext4_mark_inode_dirty(handle, inode); |
5317 | ext4_journal_stop(handle); | |
512c15ef | 5318 | if (unlikely(error)) { |
4209ae12 | 5319 | return error; |
512c15ef | 5320 | } |
ac27a0ec DK |
5321 | } |
5322 | ||
3da40c7b | 5323 | if (attr->ia_valid & ATTR_SIZE) { |
5208386c | 5324 | handle_t *handle; |
3da40c7b | 5325 | loff_t oldsize = inode->i_size; |
f4534c9f | 5326 | loff_t old_disksize; |
b9c1c267 | 5327 | int shrink = (attr->ia_size < inode->i_size); |
562c72aa | 5328 | |
12e9b892 | 5329 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) { |
e2b46574 ES |
5330 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
5331 | ||
aa75f4d3 | 5332 | if (attr->ia_size > sbi->s_bitmap_maxbytes) { |
0c095c7f | 5333 | return -EFBIG; |
aa75f4d3 | 5334 | } |
e2b46574 | 5335 | } |
aa75f4d3 | 5336 | if (!S_ISREG(inode->i_mode)) { |
3da40c7b | 5337 | return -EINVAL; |
aa75f4d3 | 5338 | } |
dff6efc3 | 5339 | |
a642c2c0 JL |
5340 | if (attr->ia_size == inode->i_size) |
5341 | inc_ivers = false; | |
dff6efc3 | 5342 | |
b9c1c267 JK |
5343 | if (shrink) { |
5344 | if (ext4_should_order_data(inode)) { | |
5345 | error = ext4_begin_ordered_truncate(inode, | |
678aaf48 | 5346 | attr->ia_size); |
b9c1c267 JK |
5347 | if (error) |
5348 | goto err_out; | |
5349 | } | |
5350 | /* | |
5351 | * Blocks are going to be removed from the inode. Wait | |
5352 | * for dio in flight. | |
5353 | */ | |
5354 | inode_dio_wait(inode); | |
5355 | } | |
5356 | ||
d4f5258e | 5357 | filemap_invalidate_lock(inode->i_mapping); |
b9c1c267 JK |
5358 | |
5359 | rc = ext4_break_layouts(inode); | |
5360 | if (rc) { | |
d4f5258e | 5361 | filemap_invalidate_unlock(inode->i_mapping); |
aa75f4d3 | 5362 | goto err_out; |
3da40c7b | 5363 | } |
b9c1c267 | 5364 | |
3da40c7b | 5365 | if (attr->ia_size != inode->i_size) { |
5208386c JK |
5366 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 3); |
5367 | if (IS_ERR(handle)) { | |
5368 | error = PTR_ERR(handle); | |
b9c1c267 | 5369 | goto out_mmap_sem; |
5208386c | 5370 | } |
3da40c7b | 5371 | if (ext4_handle_valid(handle) && shrink) { |
5208386c JK |
5372 | error = ext4_orphan_add(handle, inode); |
5373 | orphan = 1; | |
5374 | } | |
911af577 EG |
5375 | /* |
5376 | * Update c/mtime on truncate up, ext4_truncate() will | |
5377 | * update c/mtime in shrink case below | |
5378 | */ | |
5379 | if (!shrink) { | |
eeca7ea1 | 5380 | inode->i_mtime = current_time(inode); |
911af577 EG |
5381 | inode->i_ctime = inode->i_mtime; |
5382 | } | |
aa75f4d3 HS |
5383 | |
5384 | if (shrink) | |
a80f7fcf | 5385 | ext4_fc_track_range(handle, inode, |
aa75f4d3 HS |
5386 | (attr->ia_size > 0 ? attr->ia_size - 1 : 0) >> |
5387 | inode->i_sb->s_blocksize_bits, | |
9725958b | 5388 | EXT_MAX_BLOCKS - 1); |
aa75f4d3 HS |
5389 | else |
5390 | ext4_fc_track_range( | |
a80f7fcf | 5391 | handle, inode, |
aa75f4d3 HS |
5392 | (oldsize > 0 ? oldsize - 1 : oldsize) >> |
5393 | inode->i_sb->s_blocksize_bits, | |
5394 | (attr->ia_size > 0 ? attr->ia_size - 1 : 0) >> | |
5395 | inode->i_sb->s_blocksize_bits); | |
5396 | ||
90e775b7 | 5397 | down_write(&EXT4_I(inode)->i_data_sem); |
f4534c9f | 5398 | old_disksize = EXT4_I(inode)->i_disksize; |
5208386c JK |
5399 | EXT4_I(inode)->i_disksize = attr->ia_size; |
5400 | rc = ext4_mark_inode_dirty(handle, inode); | |
5401 | if (!error) | |
5402 | error = rc; | |
90e775b7 JK |
5403 | /* |
5404 | * We have to update i_size under i_data_sem together | |
5405 | * with i_disksize to avoid races with writeback code | |
5406 | * running ext4_wb_update_i_disksize(). | |
5407 | */ | |
5408 | if (!error) | |
5409 | i_size_write(inode, attr->ia_size); | |
f4534c9f YB |
5410 | else |
5411 | EXT4_I(inode)->i_disksize = old_disksize; | |
90e775b7 | 5412 | up_write(&EXT4_I(inode)->i_data_sem); |
5208386c | 5413 | ext4_journal_stop(handle); |
b9c1c267 JK |
5414 | if (error) |
5415 | goto out_mmap_sem; | |
5416 | if (!shrink) { | |
5417 | pagecache_isize_extended(inode, oldsize, | |
5418 | inode->i_size); | |
5419 | } else if (ext4_should_journal_data(inode)) { | |
5420 | ext4_wait_for_tail_page_commit(inode); | |
678aaf48 | 5421 | } |
d6320cbf | 5422 | } |
430657b6 | 5423 | |
5208386c JK |
5424 | /* |
5425 | * Truncate pagecache after we've waited for commit | |
5426 | * in data=journal mode to make pages freeable. | |
5427 | */ | |
923ae0ff | 5428 | truncate_pagecache(inode, inode->i_size); |
b9c1c267 JK |
5429 | /* |
5430 | * Call ext4_truncate() even if i_size didn't change to | |
5431 | * truncate possible preallocated blocks. | |
5432 | */ | |
5433 | if (attr->ia_size <= oldsize) { | |
2c98eb5e TT |
5434 | rc = ext4_truncate(inode); |
5435 | if (rc) | |
5436 | error = rc; | |
5437 | } | |
b9c1c267 | 5438 | out_mmap_sem: |
d4f5258e | 5439 | filemap_invalidate_unlock(inode->i_mapping); |
072bd7ea | 5440 | } |
ac27a0ec | 5441 | |
2c98eb5e | 5442 | if (!error) { |
a642c2c0 JL |
5443 | if (inc_ivers) |
5444 | inode_inc_iversion(inode); | |
c1632a0f | 5445 | setattr_copy(idmap, inode, attr); |
1025774c CH |
5446 | mark_inode_dirty(inode); |
5447 | } | |
5448 | ||
5449 | /* | |
5450 | * If the call to ext4_truncate failed to get a transaction handle at | |
5451 | * all, we need to clean up the in-core orphan list manually. | |
5452 | */ | |
3d287de3 | 5453 | if (orphan && inode->i_nlink) |
617ba13b | 5454 | ext4_orphan_del(NULL, inode); |
ac27a0ec | 5455 | |
2c98eb5e | 5456 | if (!error && (ia_valid & ATTR_MODE)) |
13e83a49 | 5457 | rc = posix_acl_chmod(idmap, dentry, inode->i_mode); |
ac27a0ec DK |
5458 | |
5459 | err_out: | |
aa75f4d3 HS |
5460 | if (error) |
5461 | ext4_std_error(inode->i_sb, error); | |
ac27a0ec DK |
5462 | if (!error) |
5463 | error = rc; | |
5464 | return error; | |
5465 | } | |
5466 | ||
8434ef1d EB |
5467 | u32 ext4_dio_alignment(struct inode *inode) |
5468 | { | |
5469 | if (fsverity_active(inode)) | |
5470 | return 0; | |
5471 | if (ext4_should_journal_data(inode)) | |
5472 | return 0; | |
5473 | if (ext4_has_inline_data(inode)) | |
5474 | return 0; | |
5475 | if (IS_ENCRYPTED(inode)) { | |
5476 | if (!fscrypt_dio_supported(inode)) | |
5477 | return 0; | |
5478 | return i_blocksize(inode); | |
5479 | } | |
5480 | return 1; /* use the iomap defaults */ | |
5481 | } | |
5482 | ||
b74d24f7 | 5483 | int ext4_getattr(struct mnt_idmap *idmap, const struct path *path, |
549c7297 | 5484 | struct kstat *stat, u32 request_mask, unsigned int query_flags) |
3e3398a0 | 5485 | { |
99652ea5 DH |
5486 | struct inode *inode = d_inode(path->dentry); |
5487 | struct ext4_inode *raw_inode; | |
5488 | struct ext4_inode_info *ei = EXT4_I(inode); | |
5489 | unsigned int flags; | |
5490 | ||
d4c5e960 TT |
5491 | if ((request_mask & STATX_BTIME) && |
5492 | EXT4_FITS_IN_INODE(raw_inode, ei, i_crtime)) { | |
99652ea5 DH |
5493 | stat->result_mask |= STATX_BTIME; |
5494 | stat->btime.tv_sec = ei->i_crtime.tv_sec; | |
5495 | stat->btime.tv_nsec = ei->i_crtime.tv_nsec; | |
5496 | } | |
5497 | ||
8434ef1d EB |
5498 | /* |
5499 | * Return the DIO alignment restrictions if requested. We only return | |
5500 | * this information when requested, since on encrypted files it might | |
5501 | * take a fair bit of work to get if the file wasn't opened recently. | |
5502 | */ | |
5503 | if ((request_mask & STATX_DIOALIGN) && S_ISREG(inode->i_mode)) { | |
5504 | u32 dio_align = ext4_dio_alignment(inode); | |
5505 | ||
5506 | stat->result_mask |= STATX_DIOALIGN; | |
5507 | if (dio_align == 1) { | |
5508 | struct block_device *bdev = inode->i_sb->s_bdev; | |
5509 | ||
5510 | /* iomap defaults */ | |
5511 | stat->dio_mem_align = bdev_dma_alignment(bdev) + 1; | |
5512 | stat->dio_offset_align = bdev_logical_block_size(bdev); | |
5513 | } else { | |
5514 | stat->dio_mem_align = dio_align; | |
5515 | stat->dio_offset_align = dio_align; | |
5516 | } | |
5517 | } | |
5518 | ||
99652ea5 DH |
5519 | flags = ei->i_flags & EXT4_FL_USER_VISIBLE; |
5520 | if (flags & EXT4_APPEND_FL) | |
5521 | stat->attributes |= STATX_ATTR_APPEND; | |
5522 | if (flags & EXT4_COMPR_FL) | |
5523 | stat->attributes |= STATX_ATTR_COMPRESSED; | |
5524 | if (flags & EXT4_ENCRYPT_FL) | |
5525 | stat->attributes |= STATX_ATTR_ENCRYPTED; | |
5526 | if (flags & EXT4_IMMUTABLE_FL) | |
5527 | stat->attributes |= STATX_ATTR_IMMUTABLE; | |
5528 | if (flags & EXT4_NODUMP_FL) | |
5529 | stat->attributes |= STATX_ATTR_NODUMP; | |
1f607195 EB |
5530 | if (flags & EXT4_VERITY_FL) |
5531 | stat->attributes |= STATX_ATTR_VERITY; | |
3e3398a0 | 5532 | |
3209f68b DH |
5533 | stat->attributes_mask |= (STATX_ATTR_APPEND | |
5534 | STATX_ATTR_COMPRESSED | | |
5535 | STATX_ATTR_ENCRYPTED | | |
5536 | STATX_ATTR_IMMUTABLE | | |
1f607195 EB |
5537 | STATX_ATTR_NODUMP | |
5538 | STATX_ATTR_VERITY); | |
3209f68b | 5539 | |
b74d24f7 | 5540 | generic_fillattr(idmap, inode, stat); |
99652ea5 DH |
5541 | return 0; |
5542 | } | |
5543 | ||
b74d24f7 | 5544 | int ext4_file_getattr(struct mnt_idmap *idmap, |
549c7297 | 5545 | const struct path *path, struct kstat *stat, |
99652ea5 DH |
5546 | u32 request_mask, unsigned int query_flags) |
5547 | { | |
5548 | struct inode *inode = d_inode(path->dentry); | |
5549 | u64 delalloc_blocks; | |
5550 | ||
b74d24f7 | 5551 | ext4_getattr(idmap, path, stat, request_mask, query_flags); |
3e3398a0 | 5552 | |
9206c561 AD |
5553 | /* |
5554 | * If there is inline data in the inode, the inode will normally not | |
5555 | * have data blocks allocated (it may have an external xattr block). | |
5556 | * Report at least one sector for such files, so tools like tar, rsync, | |
d67d64f4 | 5557 | * others don't incorrectly think the file is completely sparse. |
9206c561 AD |
5558 | */ |
5559 | if (unlikely(ext4_has_inline_data(inode))) | |
5560 | stat->blocks += (stat->size + 511) >> 9; | |
5561 | ||
3e3398a0 MC |
5562 | /* |
5563 | * We can't update i_blocks if the block allocation is delayed | |
5564 | * otherwise in the case of system crash before the real block | |
5565 | * allocation is done, we will have i_blocks inconsistent with | |
5566 | * on-disk file blocks. | |
5567 | * We always keep i_blocks updated together with real | |
5568 | * allocation. But to not confuse with user, stat | |
5569 | * will return the blocks that include the delayed allocation | |
5570 | * blocks for this file. | |
5571 | */ | |
96607551 | 5572 | delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb), |
9206c561 AD |
5573 | EXT4_I(inode)->i_reserved_data_blocks); |
5574 | stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9); | |
3e3398a0 MC |
5575 | return 0; |
5576 | } | |
ac27a0ec | 5577 | |
fffb2739 JK |
5578 | static int ext4_index_trans_blocks(struct inode *inode, int lblocks, |
5579 | int pextents) | |
a02908f1 | 5580 | { |
12e9b892 | 5581 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) |
fffb2739 JK |
5582 | return ext4_ind_trans_blocks(inode, lblocks); |
5583 | return ext4_ext_index_trans_blocks(inode, pextents); | |
a02908f1 | 5584 | } |
ac51d837 | 5585 | |
ac27a0ec | 5586 | /* |
a02908f1 MC |
5587 | * Account for index blocks, block groups bitmaps and block group |
5588 | * descriptor blocks if modify datablocks and index blocks | |
5589 | * worse case, the indexs blocks spread over different block groups | |
ac27a0ec | 5590 | * |
a02908f1 | 5591 | * If datablocks are discontiguous, they are possible to spread over |
4907cb7b | 5592 | * different block groups too. If they are contiguous, with flexbg, |
a02908f1 | 5593 | * they could still across block group boundary. |
ac27a0ec | 5594 | * |
a02908f1 MC |
5595 | * Also account for superblock, inode, quota and xattr blocks |
5596 | */ | |
dec214d0 | 5597 | static int ext4_meta_trans_blocks(struct inode *inode, int lblocks, |
fffb2739 | 5598 | int pextents) |
a02908f1 | 5599 | { |
8df9675f TT |
5600 | ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb); |
5601 | int gdpblocks; | |
a02908f1 | 5602 | int idxblocks; |
7fc51f92 | 5603 | int ret; |
a02908f1 MC |
5604 | |
5605 | /* | |
fffb2739 JK |
5606 | * How many index blocks need to touch to map @lblocks logical blocks |
5607 | * to @pextents physical extents? | |
a02908f1 | 5608 | */ |
fffb2739 | 5609 | idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents); |
a02908f1 MC |
5610 | |
5611 | ret = idxblocks; | |
5612 | ||
5613 | /* | |
5614 | * Now let's see how many group bitmaps and group descriptors need | |
5615 | * to account | |
5616 | */ | |
fffb2739 | 5617 | groups = idxblocks + pextents; |
a02908f1 | 5618 | gdpblocks = groups; |
8df9675f TT |
5619 | if (groups > ngroups) |
5620 | groups = ngroups; | |
a02908f1 MC |
5621 | if (groups > EXT4_SB(inode->i_sb)->s_gdb_count) |
5622 | gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count; | |
5623 | ||
5624 | /* bitmaps and block group descriptor blocks */ | |
5625 | ret += groups + gdpblocks; | |
5626 | ||
5627 | /* Blocks for super block, inode, quota and xattr blocks */ | |
5628 | ret += EXT4_META_TRANS_BLOCKS(inode->i_sb); | |
5629 | ||
5630 | return ret; | |
5631 | } | |
5632 | ||
5633 | /* | |
25985edc | 5634 | * Calculate the total number of credits to reserve to fit |
f3bd1f3f MC |
5635 | * the modification of a single pages into a single transaction, |
5636 | * which may include multiple chunks of block allocations. | |
ac27a0ec | 5637 | * |
525f4ed8 | 5638 | * This could be called via ext4_write_begin() |
ac27a0ec | 5639 | * |
525f4ed8 | 5640 | * We need to consider the worse case, when |
a02908f1 | 5641 | * one new block per extent. |
ac27a0ec | 5642 | */ |
a86c6181 | 5643 | int ext4_writepage_trans_blocks(struct inode *inode) |
ac27a0ec | 5644 | { |
617ba13b | 5645 | int bpp = ext4_journal_blocks_per_page(inode); |
ac27a0ec DK |
5646 | int ret; |
5647 | ||
fffb2739 | 5648 | ret = ext4_meta_trans_blocks(inode, bpp, bpp); |
a86c6181 | 5649 | |
a02908f1 | 5650 | /* Account for data blocks for journalled mode */ |
617ba13b | 5651 | if (ext4_should_journal_data(inode)) |
a02908f1 | 5652 | ret += bpp; |
ac27a0ec DK |
5653 | return ret; |
5654 | } | |
f3bd1f3f MC |
5655 | |
5656 | /* | |
5657 | * Calculate the journal credits for a chunk of data modification. | |
5658 | * | |
5659 | * This is called from DIO, fallocate or whoever calling | |
79e83036 | 5660 | * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks. |
f3bd1f3f MC |
5661 | * |
5662 | * journal buffers for data blocks are not included here, as DIO | |
5663 | * and fallocate do no need to journal data buffers. | |
5664 | */ | |
5665 | int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks) | |
5666 | { | |
5667 | return ext4_meta_trans_blocks(inode, nrblocks, 1); | |
5668 | } | |
5669 | ||
ac27a0ec | 5670 | /* |
617ba13b | 5671 | * The caller must have previously called ext4_reserve_inode_write(). |
ac27a0ec DK |
5672 | * Give this, we know that the caller already has write access to iloc->bh. |
5673 | */ | |
617ba13b | 5674 | int ext4_mark_iloc_dirty(handle_t *handle, |
de9a55b8 | 5675 | struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec DK |
5676 | { |
5677 | int err = 0; | |
5678 | ||
a6758309 VA |
5679 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) { |
5680 | put_bh(iloc->bh); | |
0db1ff22 | 5681 | return -EIO; |
a6758309 | 5682 | } |
a80f7fcf | 5683 | ext4_fc_track_inode(handle, inode); |
aa75f4d3 | 5684 | |
ac27a0ec DK |
5685 | /* the do_update_inode consumes one bh->b_count */ |
5686 | get_bh(iloc->bh); | |
5687 | ||
dab291af | 5688 | /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */ |
830156c7 | 5689 | err = ext4_do_update_inode(handle, inode, iloc); |
ac27a0ec DK |
5690 | put_bh(iloc->bh); |
5691 | return err; | |
5692 | } | |
5693 | ||
5694 | /* | |
5695 | * On success, We end up with an outstanding reference count against | |
5696 | * iloc->bh. This _must_ be cleaned up later. | |
5697 | */ | |
5698 | ||
5699 | int | |
617ba13b MC |
5700 | ext4_reserve_inode_write(handle_t *handle, struct inode *inode, |
5701 | struct ext4_iloc *iloc) | |
ac27a0ec | 5702 | { |
0390131b FM |
5703 | int err; |
5704 | ||
0db1ff22 TT |
5705 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
5706 | return -EIO; | |
5707 | ||
0390131b FM |
5708 | err = ext4_get_inode_loc(inode, iloc); |
5709 | if (!err) { | |
5710 | BUFFER_TRACE(iloc->bh, "get_write_access"); | |
188c299e JK |
5711 | err = ext4_journal_get_write_access(handle, inode->i_sb, |
5712 | iloc->bh, EXT4_JTR_NONE); | |
0390131b FM |
5713 | if (err) { |
5714 | brelse(iloc->bh); | |
5715 | iloc->bh = NULL; | |
ac27a0ec DK |
5716 | } |
5717 | } | |
617ba13b | 5718 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
5719 | return err; |
5720 | } | |
5721 | ||
c03b45b8 MX |
5722 | static int __ext4_expand_extra_isize(struct inode *inode, |
5723 | unsigned int new_extra_isize, | |
5724 | struct ext4_iloc *iloc, | |
5725 | handle_t *handle, int *no_expand) | |
5726 | { | |
5727 | struct ext4_inode *raw_inode; | |
5728 | struct ext4_xattr_ibody_header *header; | |
4ea99936 TT |
5729 | unsigned int inode_size = EXT4_INODE_SIZE(inode->i_sb); |
5730 | struct ext4_inode_info *ei = EXT4_I(inode); | |
c03b45b8 MX |
5731 | int error; |
5732 | ||
4ea99936 TT |
5733 | /* this was checked at iget time, but double check for good measure */ |
5734 | if ((EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > inode_size) || | |
5735 | (ei->i_extra_isize & 3)) { | |
5736 | EXT4_ERROR_INODE(inode, "bad extra_isize %u (inode size %u)", | |
5737 | ei->i_extra_isize, | |
5738 | EXT4_INODE_SIZE(inode->i_sb)); | |
5739 | return -EFSCORRUPTED; | |
5740 | } | |
5741 | if ((new_extra_isize < ei->i_extra_isize) || | |
5742 | (new_extra_isize < 4) || | |
5743 | (new_extra_isize > inode_size - EXT4_GOOD_OLD_INODE_SIZE)) | |
5744 | return -EINVAL; /* Should never happen */ | |
5745 | ||
c03b45b8 MX |
5746 | raw_inode = ext4_raw_inode(iloc); |
5747 | ||
5748 | header = IHDR(inode, raw_inode); | |
5749 | ||
5750 | /* No extended attributes present */ | |
5751 | if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) || | |
5752 | header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) { | |
5753 | memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE + | |
5754 | EXT4_I(inode)->i_extra_isize, 0, | |
5755 | new_extra_isize - EXT4_I(inode)->i_extra_isize); | |
5756 | EXT4_I(inode)->i_extra_isize = new_extra_isize; | |
5757 | return 0; | |
5758 | } | |
5759 | ||
8994d113 JK |
5760 | /* |
5761 | * We may need to allocate external xattr block so we need quotas | |
5762 | * initialized. Here we can be called with various locks held so we | |
5763 | * cannot affort to initialize quotas ourselves. So just bail. | |
5764 | */ | |
5765 | if (dquot_initialize_needed(inode)) | |
5766 | return -EAGAIN; | |
5767 | ||
c03b45b8 MX |
5768 | /* try to expand with EAs present */ |
5769 | error = ext4_expand_extra_isize_ea(inode, new_extra_isize, | |
5770 | raw_inode, handle); | |
5771 | if (error) { | |
5772 | /* | |
5773 | * Inode size expansion failed; don't try again | |
5774 | */ | |
5775 | *no_expand = 1; | |
5776 | } | |
5777 | ||
5778 | return error; | |
5779 | } | |
5780 | ||
6dd4ee7c KS |
5781 | /* |
5782 | * Expand an inode by new_extra_isize bytes. | |
5783 | * Returns 0 on success or negative error number on failure. | |
5784 | */ | |
cf0a5e81 MX |
5785 | static int ext4_try_to_expand_extra_isize(struct inode *inode, |
5786 | unsigned int new_extra_isize, | |
5787 | struct ext4_iloc iloc, | |
5788 | handle_t *handle) | |
6dd4ee7c | 5789 | { |
3b10fdc6 MX |
5790 | int no_expand; |
5791 | int error; | |
6dd4ee7c | 5792 | |
cf0a5e81 MX |
5793 | if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) |
5794 | return -EOVERFLOW; | |
5795 | ||
5796 | /* | |
5797 | * In nojournal mode, we can immediately attempt to expand | |
5798 | * the inode. When journaled, we first need to obtain extra | |
5799 | * buffer credits since we may write into the EA block | |
5800 | * with this same handle. If journal_extend fails, then it will | |
5801 | * only result in a minor loss of functionality for that inode. | |
5802 | * If this is felt to be critical, then e2fsck should be run to | |
5803 | * force a large enough s_min_extra_isize. | |
5804 | */ | |
6cb367c2 | 5805 | if (ext4_journal_extend(handle, |
83448bdf | 5806 | EXT4_DATA_TRANS_BLOCKS(inode->i_sb), 0) != 0) |
cf0a5e81 | 5807 | return -ENOSPC; |
6dd4ee7c | 5808 | |
3b10fdc6 | 5809 | if (ext4_write_trylock_xattr(inode, &no_expand) == 0) |
cf0a5e81 | 5810 | return -EBUSY; |
3b10fdc6 | 5811 | |
c03b45b8 MX |
5812 | error = __ext4_expand_extra_isize(inode, new_extra_isize, &iloc, |
5813 | handle, &no_expand); | |
5814 | ext4_write_unlock_xattr(inode, &no_expand); | |
6dd4ee7c | 5815 | |
c03b45b8 MX |
5816 | return error; |
5817 | } | |
6dd4ee7c | 5818 | |
c03b45b8 MX |
5819 | int ext4_expand_extra_isize(struct inode *inode, |
5820 | unsigned int new_extra_isize, | |
5821 | struct ext4_iloc *iloc) | |
5822 | { | |
5823 | handle_t *handle; | |
5824 | int no_expand; | |
5825 | int error, rc; | |
5826 | ||
5827 | if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) { | |
5828 | brelse(iloc->bh); | |
5829 | return -EOVERFLOW; | |
6dd4ee7c KS |
5830 | } |
5831 | ||
c03b45b8 MX |
5832 | handle = ext4_journal_start(inode, EXT4_HT_INODE, |
5833 | EXT4_DATA_TRANS_BLOCKS(inode->i_sb)); | |
5834 | if (IS_ERR(handle)) { | |
5835 | error = PTR_ERR(handle); | |
5836 | brelse(iloc->bh); | |
5837 | return error; | |
5838 | } | |
5839 | ||
5840 | ext4_write_lock_xattr(inode, &no_expand); | |
5841 | ||
ddccb6db | 5842 | BUFFER_TRACE(iloc->bh, "get_write_access"); |
188c299e JK |
5843 | error = ext4_journal_get_write_access(handle, inode->i_sb, iloc->bh, |
5844 | EXT4_JTR_NONE); | |
3b10fdc6 | 5845 | if (error) { |
c03b45b8 | 5846 | brelse(iloc->bh); |
7f420d64 | 5847 | goto out_unlock; |
3b10fdc6 | 5848 | } |
cf0a5e81 | 5849 | |
c03b45b8 MX |
5850 | error = __ext4_expand_extra_isize(inode, new_extra_isize, iloc, |
5851 | handle, &no_expand); | |
5852 | ||
5853 | rc = ext4_mark_iloc_dirty(handle, inode, iloc); | |
5854 | if (!error) | |
5855 | error = rc; | |
5856 | ||
7f420d64 | 5857 | out_unlock: |
c03b45b8 | 5858 | ext4_write_unlock_xattr(inode, &no_expand); |
c03b45b8 | 5859 | ext4_journal_stop(handle); |
3b10fdc6 | 5860 | return error; |
6dd4ee7c KS |
5861 | } |
5862 | ||
ac27a0ec DK |
5863 | /* |
5864 | * What we do here is to mark the in-core inode as clean with respect to inode | |
5865 | * dirtiness (it may still be data-dirty). | |
5866 | * This means that the in-core inode may be reaped by prune_icache | |
5867 | * without having to perform any I/O. This is a very good thing, | |
5868 | * because *any* task may call prune_icache - even ones which | |
5869 | * have a transaction open against a different journal. | |
5870 | * | |
5871 | * Is this cheating? Not really. Sure, we haven't written the | |
5872 | * inode out, but prune_icache isn't a user-visible syncing function. | |
5873 | * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync) | |
5874 | * we start and wait on commits. | |
ac27a0ec | 5875 | */ |
4209ae12 HS |
5876 | int __ext4_mark_inode_dirty(handle_t *handle, struct inode *inode, |
5877 | const char *func, unsigned int line) | |
ac27a0ec | 5878 | { |
617ba13b | 5879 | struct ext4_iloc iloc; |
6dd4ee7c | 5880 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
cf0a5e81 | 5881 | int err; |
ac27a0ec DK |
5882 | |
5883 | might_sleep(); | |
7ff9c073 | 5884 | trace_ext4_mark_inode_dirty(inode, _RET_IP_); |
617ba13b | 5885 | err = ext4_reserve_inode_write(handle, inode, &iloc); |
5e1021f2 | 5886 | if (err) |
4209ae12 | 5887 | goto out; |
cf0a5e81 MX |
5888 | |
5889 | if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize) | |
5890 | ext4_try_to_expand_extra_isize(inode, sbi->s_want_extra_isize, | |
5891 | iloc, handle); | |
5892 | ||
4209ae12 HS |
5893 | err = ext4_mark_iloc_dirty(handle, inode, &iloc); |
5894 | out: | |
5895 | if (unlikely(err)) | |
5896 | ext4_error_inode_err(inode, func, line, 0, err, | |
5897 | "mark_inode_dirty error"); | |
5898 | return err; | |
ac27a0ec DK |
5899 | } |
5900 | ||
5901 | /* | |
617ba13b | 5902 | * ext4_dirty_inode() is called from __mark_inode_dirty() |
ac27a0ec DK |
5903 | * |
5904 | * We're really interested in the case where a file is being extended. | |
5905 | * i_size has been changed by generic_commit_write() and we thus need | |
5906 | * to include the updated inode in the current transaction. | |
5907 | * | |
5dd4056d | 5908 | * Also, dquot_alloc_block() will always dirty the inode when blocks |
ac27a0ec DK |
5909 | * are allocated to the file. |
5910 | * | |
5911 | * If the inode is marked synchronous, we don't honour that here - doing | |
5912 | * so would cause a commit on atime updates, which we don't bother doing. | |
5913 | * We handle synchronous inodes at the highest possible level. | |
5914 | */ | |
aa385729 | 5915 | void ext4_dirty_inode(struct inode *inode, int flags) |
ac27a0ec | 5916 | { |
ac27a0ec DK |
5917 | handle_t *handle; |
5918 | ||
9924a92a | 5919 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 2); |
ac27a0ec | 5920 | if (IS_ERR(handle)) |
e2728c56 | 5921 | return; |
f3dc272f | 5922 | ext4_mark_inode_dirty(handle, inode); |
617ba13b | 5923 | ext4_journal_stop(handle); |
ac27a0ec DK |
5924 | } |
5925 | ||
617ba13b | 5926 | int ext4_change_inode_journal_flag(struct inode *inode, int val) |
ac27a0ec DK |
5927 | { |
5928 | journal_t *journal; | |
5929 | handle_t *handle; | |
5930 | int err; | |
00d873c1 | 5931 | int alloc_ctx; |
ac27a0ec DK |
5932 | |
5933 | /* | |
5934 | * We have to be very careful here: changing a data block's | |
5935 | * journaling status dynamically is dangerous. If we write a | |
5936 | * data block to the journal, change the status and then delete | |
5937 | * that block, we risk forgetting to revoke the old log record | |
5938 | * from the journal and so a subsequent replay can corrupt data. | |
5939 | * So, first we make sure that the journal is empty and that | |
5940 | * nobody is changing anything. | |
5941 | */ | |
5942 | ||
617ba13b | 5943 | journal = EXT4_JOURNAL(inode); |
0390131b FM |
5944 | if (!journal) |
5945 | return 0; | |
d699594d | 5946 | if (is_journal_aborted(journal)) |
ac27a0ec DK |
5947 | return -EROFS; |
5948 | ||
17335dcc | 5949 | /* Wait for all existing dio workers */ |
17335dcc DM |
5950 | inode_dio_wait(inode); |
5951 | ||
4c546592 DJ |
5952 | /* |
5953 | * Before flushing the journal and switching inode's aops, we have | |
5954 | * to flush all dirty data the inode has. There can be outstanding | |
5955 | * delayed allocations, there can be unwritten extents created by | |
5956 | * fallocate or buffered writes in dioread_nolock mode covered by | |
5957 | * dirty data which can be converted only after flushing the dirty | |
5958 | * data (and journalled aops don't know how to handle these cases). | |
5959 | */ | |
5960 | if (val) { | |
d4f5258e | 5961 | filemap_invalidate_lock(inode->i_mapping); |
4c546592 DJ |
5962 | err = filemap_write_and_wait(inode->i_mapping); |
5963 | if (err < 0) { | |
d4f5258e | 5964 | filemap_invalidate_unlock(inode->i_mapping); |
4c546592 DJ |
5965 | return err; |
5966 | } | |
5967 | } | |
5968 | ||
00d873c1 | 5969 | alloc_ctx = ext4_writepages_down_write(inode->i_sb); |
dab291af | 5970 | jbd2_journal_lock_updates(journal); |
ac27a0ec DK |
5971 | |
5972 | /* | |
5973 | * OK, there are no updates running now, and all cached data is | |
5974 | * synced to disk. We are now in a completely consistent state | |
5975 | * which doesn't have anything in the journal, and we know that | |
5976 | * no filesystem updates are running, so it is safe to modify | |
5977 | * the inode's in-core data-journaling state flag now. | |
5978 | */ | |
5979 | ||
5980 | if (val) | |
12e9b892 | 5981 | ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
5872ddaa | 5982 | else { |
01d5d965 | 5983 | err = jbd2_journal_flush(journal, 0); |
4f879ca6 JK |
5984 | if (err < 0) { |
5985 | jbd2_journal_unlock_updates(journal); | |
00d873c1 | 5986 | ext4_writepages_up_write(inode->i_sb, alloc_ctx); |
4f879ca6 JK |
5987 | return err; |
5988 | } | |
12e9b892 | 5989 | ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
5872ddaa | 5990 | } |
617ba13b | 5991 | ext4_set_aops(inode); |
ac27a0ec | 5992 | |
dab291af | 5993 | jbd2_journal_unlock_updates(journal); |
00d873c1 | 5994 | ext4_writepages_up_write(inode->i_sb, alloc_ctx); |
c8585c6f | 5995 | |
4c546592 | 5996 | if (val) |
d4f5258e | 5997 | filemap_invalidate_unlock(inode->i_mapping); |
ac27a0ec DK |
5998 | |
5999 | /* Finally we can mark the inode as dirty. */ | |
6000 | ||
9924a92a | 6001 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 1); |
ac27a0ec DK |
6002 | if (IS_ERR(handle)) |
6003 | return PTR_ERR(handle); | |
6004 | ||
aa75f4d3 | 6005 | ext4_fc_mark_ineligible(inode->i_sb, |
e85c81ba | 6006 | EXT4_FC_REASON_JOURNAL_FLAG_CHANGE, handle); |
617ba13b | 6007 | err = ext4_mark_inode_dirty(handle, inode); |
0390131b | 6008 | ext4_handle_sync(handle); |
617ba13b MC |
6009 | ext4_journal_stop(handle); |
6010 | ext4_std_error(inode->i_sb, err); | |
ac27a0ec DK |
6011 | |
6012 | return err; | |
6013 | } | |
2e9ee850 | 6014 | |
188c299e JK |
6015 | static int ext4_bh_unmapped(handle_t *handle, struct inode *inode, |
6016 | struct buffer_head *bh) | |
2e9ee850 AK |
6017 | { |
6018 | return !buffer_mapped(bh); | |
6019 | } | |
6020 | ||
401b25aa | 6021 | vm_fault_t ext4_page_mkwrite(struct vm_fault *vmf) |
2e9ee850 | 6022 | { |
11bac800 | 6023 | struct vm_area_struct *vma = vmf->vma; |
9ea0e45b | 6024 | struct folio *folio = page_folio(vmf->page); |
2e9ee850 AK |
6025 | loff_t size; |
6026 | unsigned long len; | |
401b25aa SJ |
6027 | int err; |
6028 | vm_fault_t ret; | |
2e9ee850 | 6029 | struct file *file = vma->vm_file; |
496ad9aa | 6030 | struct inode *inode = file_inode(file); |
2e9ee850 | 6031 | struct address_space *mapping = inode->i_mapping; |
9ea7df53 JK |
6032 | handle_t *handle; |
6033 | get_block_t *get_block; | |
6034 | int retries = 0; | |
2e9ee850 | 6035 | |
02b016ca TT |
6036 | if (unlikely(IS_IMMUTABLE(inode))) |
6037 | return VM_FAULT_SIGBUS; | |
6038 | ||
8e8ad8a5 | 6039 | sb_start_pagefault(inode->i_sb); |
041bbb6d | 6040 | file_update_time(vma->vm_file); |
ea3d7209 | 6041 | |
d4f5258e | 6042 | filemap_invalidate_lock_shared(mapping); |
7b4cc978 | 6043 | |
401b25aa SJ |
6044 | err = ext4_convert_inline_data(inode); |
6045 | if (err) | |
7b4cc978 EB |
6046 | goto out_ret; |
6047 | ||
64a9f144 MFO |
6048 | /* |
6049 | * On data journalling we skip straight to the transaction handle: | |
6050 | * there's no delalloc; page truncated will be checked later; the | |
6051 | * early return w/ all buffers mapped (calculates size/len) can't | |
6052 | * be used; and there's no dioread_nolock, so only ext4_get_block. | |
6053 | */ | |
6054 | if (ext4_should_journal_data(inode)) | |
6055 | goto retry_alloc; | |
6056 | ||
9ea7df53 JK |
6057 | /* Delalloc case is easy... */ |
6058 | if (test_opt(inode->i_sb, DELALLOC) && | |
9ea7df53 JK |
6059 | !ext4_nonda_switch(inode->i_sb)) { |
6060 | do { | |
401b25aa | 6061 | err = block_page_mkwrite(vma, vmf, |
9ea7df53 | 6062 | ext4_da_get_block_prep); |
401b25aa | 6063 | } while (err == -ENOSPC && |
9ea7df53 JK |
6064 | ext4_should_retry_alloc(inode->i_sb, &retries)); |
6065 | goto out_ret; | |
2e9ee850 | 6066 | } |
0e499890 | 6067 | |
9ea0e45b | 6068 | folio_lock(folio); |
9ea7df53 JK |
6069 | size = i_size_read(inode); |
6070 | /* Page got truncated from under us? */ | |
9ea0e45b MW |
6071 | if (folio->mapping != mapping || folio_pos(folio) > size) { |
6072 | folio_unlock(folio); | |
9ea7df53 JK |
6073 | ret = VM_FAULT_NOPAGE; |
6074 | goto out; | |
0e499890 | 6075 | } |
2e9ee850 | 6076 | |
9ea0e45b MW |
6077 | len = folio_size(folio); |
6078 | if (folio_pos(folio) + len > size) | |
6079 | len = size - folio_pos(folio); | |
a827eaff | 6080 | /* |
9ea7df53 JK |
6081 | * Return if we have all the buffers mapped. This avoids the need to do |
6082 | * journal_start/journal_stop which can block and take a long time | |
64a9f144 MFO |
6083 | * |
6084 | * This cannot be done for data journalling, as we have to add the | |
6085 | * inode to the transaction's list to writeprotect pages on commit. | |
a827eaff | 6086 | */ |
9ea0e45b MW |
6087 | if (folio_buffers(folio)) { |
6088 | if (!ext4_walk_page_buffers(NULL, inode, folio_buffers(folio), | |
f19d5870 TM |
6089 | 0, len, NULL, |
6090 | ext4_bh_unmapped)) { | |
9ea7df53 | 6091 | /* Wait so that we don't change page under IO */ |
9ea0e45b | 6092 | folio_wait_stable(folio); |
9ea7df53 JK |
6093 | ret = VM_FAULT_LOCKED; |
6094 | goto out; | |
a827eaff | 6095 | } |
2e9ee850 | 6096 | } |
9ea0e45b | 6097 | folio_unlock(folio); |
9ea7df53 JK |
6098 | /* OK, we need to fill the hole... */ |
6099 | if (ext4_should_dioread_nolock(inode)) | |
705965bd | 6100 | get_block = ext4_get_block_unwritten; |
9ea7df53 JK |
6101 | else |
6102 | get_block = ext4_get_block; | |
6103 | retry_alloc: | |
9924a92a TT |
6104 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
6105 | ext4_writepage_trans_blocks(inode)); | |
9ea7df53 | 6106 | if (IS_ERR(handle)) { |
c2ec175c | 6107 | ret = VM_FAULT_SIGBUS; |
9ea7df53 JK |
6108 | goto out; |
6109 | } | |
64a9f144 MFO |
6110 | /* |
6111 | * Data journalling can't use block_page_mkwrite() because it | |
6112 | * will set_buffer_dirty() before do_journal_get_write_access() | |
6113 | * thus might hit warning messages for dirty metadata buffers. | |
6114 | */ | |
6115 | if (!ext4_should_journal_data(inode)) { | |
6116 | err = block_page_mkwrite(vma, vmf, get_block); | |
6117 | } else { | |
9ea0e45b | 6118 | folio_lock(folio); |
64a9f144 MFO |
6119 | size = i_size_read(inode); |
6120 | /* Page got truncated from under us? */ | |
9ea0e45b | 6121 | if (folio->mapping != mapping || folio_pos(folio) > size) { |
64a9f144 | 6122 | ret = VM_FAULT_NOPAGE; |
afb585a9 | 6123 | goto out_error; |
9ea7df53 | 6124 | } |
64a9f144 | 6125 | |
9ea0e45b MW |
6126 | len = folio_size(folio); |
6127 | if (folio_pos(folio) + len > size) | |
6128 | len = size - folio_pos(folio); | |
64a9f144 | 6129 | |
9ea0e45b | 6130 | err = __block_write_begin(&folio->page, 0, len, ext4_get_block); |
64a9f144 | 6131 | if (!err) { |
afb585a9 | 6132 | ret = VM_FAULT_SIGBUS; |
80be8c5c | 6133 | if (ext4_journal_folio_buffers(handle, folio, len)) |
afb585a9 | 6134 | goto out_error; |
64a9f144 | 6135 | } else { |
9ea0e45b | 6136 | folio_unlock(folio); |
64a9f144 | 6137 | } |
9ea7df53 JK |
6138 | } |
6139 | ext4_journal_stop(handle); | |
401b25aa | 6140 | if (err == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) |
9ea7df53 JK |
6141 | goto retry_alloc; |
6142 | out_ret: | |
401b25aa | 6143 | ret = block_page_mkwrite_return(err); |
9ea7df53 | 6144 | out: |
d4f5258e | 6145 | filemap_invalidate_unlock_shared(mapping); |
8e8ad8a5 | 6146 | sb_end_pagefault(inode->i_sb); |
2e9ee850 | 6147 | return ret; |
afb585a9 | 6148 | out_error: |
9ea0e45b | 6149 | folio_unlock(folio); |
afb585a9 MFO |
6150 | ext4_journal_stop(handle); |
6151 | goto out; | |
2e9ee850 | 6152 | } |