]>
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 | ||
d47992f8 LC |
139 | static void ext4_invalidatepage(struct page *page, unsigned int offset, |
140 | unsigned int length); | |
cb20d518 | 141 | static int __ext4_journalled_writepage(struct page *page, unsigned int len); |
dec214d0 TE |
142 | static int ext4_meta_trans_blocks(struct inode *inode, int lblocks, |
143 | int pextents); | |
64769240 | 144 | |
ac27a0ec DK |
145 | /* |
146 | * Test whether an inode is a fast symlink. | |
407cd7fb | 147 | * A fast symlink has its symlink data stored in ext4_inode_info->i_data. |
ac27a0ec | 148 | */ |
f348c252 | 149 | int ext4_inode_is_fast_symlink(struct inode *inode) |
ac27a0ec | 150 | { |
fc82228a AK |
151 | if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) { |
152 | int ea_blocks = EXT4_I(inode)->i_file_acl ? | |
153 | EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0; | |
154 | ||
155 | if (ext4_has_inline_data(inode)) | |
156 | return 0; | |
157 | ||
158 | return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0); | |
159 | } | |
407cd7fb TE |
160 | return S_ISLNK(inode->i_mode) && inode->i_size && |
161 | (inode->i_size < EXT4_N_BLOCKS * 4); | |
ac27a0ec DK |
162 | } |
163 | ||
ac27a0ec DK |
164 | /* |
165 | * Called at the last iput() if i_nlink is zero. | |
166 | */ | |
0930fcc1 | 167 | void ext4_evict_inode(struct inode *inode) |
ac27a0ec DK |
168 | { |
169 | handle_t *handle; | |
bc965ab3 | 170 | int err; |
65db869c JK |
171 | /* |
172 | * Credits for final inode cleanup and freeing: | |
173 | * sb + inode (ext4_orphan_del()), block bitmap, group descriptor | |
174 | * (xattr block freeing), bitmap, group descriptor (inode freeing) | |
175 | */ | |
176 | int extra_credits = 6; | |
0421a189 | 177 | struct ext4_xattr_inode_array *ea_inode_array = NULL; |
46e294ef | 178 | bool freeze_protected = false; |
ac27a0ec | 179 | |
7ff9c073 | 180 | trace_ext4_evict_inode(inode); |
2581fdc8 | 181 | |
0930fcc1 | 182 | if (inode->i_nlink) { |
2d859db3 JK |
183 | /* |
184 | * When journalling data dirty buffers are tracked only in the | |
185 | * journal. So although mm thinks everything is clean and | |
186 | * ready for reaping the inode might still have some pages to | |
187 | * write in the running transaction or waiting to be | |
188 | * checkpointed. Thus calling jbd2_journal_invalidatepage() | |
189 | * (via truncate_inode_pages()) to discard these buffers can | |
190 | * cause data loss. Also even if we did not discard these | |
191 | * buffers, we would have no way to find them after the inode | |
192 | * is reaped and thus user could see stale data if he tries to | |
193 | * read them before the transaction is checkpointed. So be | |
194 | * careful and force everything to disk here... We use | |
195 | * ei->i_datasync_tid to store the newest transaction | |
196 | * containing inode's data. | |
197 | * | |
198 | * Note that directories do not have this problem because they | |
199 | * don't use page cache. | |
200 | */ | |
6a7fd522 VN |
201 | if (inode->i_ino != EXT4_JOURNAL_INO && |
202 | ext4_should_journal_data(inode) && | |
3abb1a0f JK |
203 | (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) && |
204 | inode->i_data.nrpages) { | |
2d859db3 JK |
205 | journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; |
206 | tid_t commit_tid = EXT4_I(inode)->i_datasync_tid; | |
207 | ||
d76a3a77 | 208 | jbd2_complete_transaction(journal, commit_tid); |
2d859db3 JK |
209 | filemap_write_and_wait(&inode->i_data); |
210 | } | |
91b0abe3 | 211 | truncate_inode_pages_final(&inode->i_data); |
5dc23bdd | 212 | |
0930fcc1 AV |
213 | goto no_delete; |
214 | } | |
215 | ||
e2bfb088 TT |
216 | if (is_bad_inode(inode)) |
217 | goto no_delete; | |
218 | dquot_initialize(inode); | |
907f4554 | 219 | |
678aaf48 JK |
220 | if (ext4_should_order_data(inode)) |
221 | ext4_begin_ordered_truncate(inode, 0); | |
91b0abe3 | 222 | truncate_inode_pages_final(&inode->i_data); |
ac27a0ec | 223 | |
ceff86fd JK |
224 | /* |
225 | * For inodes with journalled data, transaction commit could have | |
226 | * dirtied the inode. Flush worker is ignoring it because of I_FREEING | |
227 | * flag but we still need to remove the inode from the writeback lists. | |
228 | */ | |
229 | if (!list_empty_careful(&inode->i_io_list)) { | |
230 | WARN_ON_ONCE(!ext4_should_journal_data(inode)); | |
231 | inode_io_list_del(inode); | |
232 | } | |
233 | ||
8e8ad8a5 JK |
234 | /* |
235 | * Protect us against freezing - iput() caller didn't have to have any | |
46e294ef JK |
236 | * protection against it. When we are in a running transaction though, |
237 | * we are already protected against freezing and we cannot grab further | |
238 | * protection due to lock ordering constraints. | |
8e8ad8a5 | 239 | */ |
46e294ef JK |
240 | if (!ext4_journal_current_handle()) { |
241 | sb_start_intwrite(inode->i_sb); | |
242 | freeze_protected = true; | |
243 | } | |
e50e5129 | 244 | |
30a7eb97 TE |
245 | if (!IS_NOQUOTA(inode)) |
246 | extra_credits += EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb); | |
247 | ||
65db869c JK |
248 | /* |
249 | * Block bitmap, group descriptor, and inode are accounted in both | |
250 | * ext4_blocks_for_truncate() and extra_credits. So subtract 3. | |
251 | */ | |
30a7eb97 | 252 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, |
65db869c | 253 | ext4_blocks_for_truncate(inode) + extra_credits - 3); |
ac27a0ec | 254 | if (IS_ERR(handle)) { |
bc965ab3 | 255 | ext4_std_error(inode->i_sb, PTR_ERR(handle)); |
ac27a0ec DK |
256 | /* |
257 | * If we're going to skip the normal cleanup, we still need to | |
258 | * make sure that the in-core orphan linked list is properly | |
259 | * cleaned up. | |
260 | */ | |
617ba13b | 261 | ext4_orphan_del(NULL, inode); |
46e294ef JK |
262 | if (freeze_protected) |
263 | sb_end_intwrite(inode->i_sb); | |
ac27a0ec DK |
264 | goto no_delete; |
265 | } | |
30a7eb97 | 266 | |
ac27a0ec | 267 | if (IS_SYNC(inode)) |
0390131b | 268 | ext4_handle_sync(handle); |
407cd7fb TE |
269 | |
270 | /* | |
271 | * Set inode->i_size to 0 before calling ext4_truncate(). We need | |
272 | * special handling of symlinks here because i_size is used to | |
273 | * determine whether ext4_inode_info->i_data contains symlink data or | |
274 | * block mappings. Setting i_size to 0 will remove its fast symlink | |
275 | * status. Erase i_data so that it becomes a valid empty block map. | |
276 | */ | |
277 | if (ext4_inode_is_fast_symlink(inode)) | |
278 | memset(EXT4_I(inode)->i_data, 0, sizeof(EXT4_I(inode)->i_data)); | |
ac27a0ec | 279 | inode->i_size = 0; |
bc965ab3 TT |
280 | err = ext4_mark_inode_dirty(handle, inode); |
281 | if (err) { | |
12062ddd | 282 | ext4_warning(inode->i_sb, |
bc965ab3 TT |
283 | "couldn't mark inode dirty (err %d)", err); |
284 | goto stop_handle; | |
285 | } | |
2c98eb5e TT |
286 | if (inode->i_blocks) { |
287 | err = ext4_truncate(inode); | |
288 | if (err) { | |
54d3adbc TT |
289 | ext4_error_err(inode->i_sb, -err, |
290 | "couldn't truncate inode %lu (err %d)", | |
291 | inode->i_ino, err); | |
2c98eb5e TT |
292 | goto stop_handle; |
293 | } | |
294 | } | |
bc965ab3 | 295 | |
30a7eb97 TE |
296 | /* Remove xattr references. */ |
297 | err = ext4_xattr_delete_inode(handle, inode, &ea_inode_array, | |
298 | extra_credits); | |
299 | if (err) { | |
300 | ext4_warning(inode->i_sb, "xattr delete (err %d)", err); | |
301 | stop_handle: | |
302 | ext4_journal_stop(handle); | |
303 | ext4_orphan_del(NULL, inode); | |
46e294ef JK |
304 | if (freeze_protected) |
305 | sb_end_intwrite(inode->i_sb); | |
30a7eb97 TE |
306 | ext4_xattr_inode_array_free(ea_inode_array); |
307 | goto no_delete; | |
bc965ab3 TT |
308 | } |
309 | ||
ac27a0ec | 310 | /* |
617ba13b | 311 | * Kill off the orphan record which ext4_truncate created. |
ac27a0ec | 312 | * AKPM: I think this can be inside the above `if'. |
617ba13b | 313 | * Note that ext4_orphan_del() has to be able to cope with the |
ac27a0ec | 314 | * deletion of a non-existent orphan - this is because we don't |
617ba13b | 315 | * know if ext4_truncate() actually created an orphan record. |
ac27a0ec DK |
316 | * (Well, we could do this if we need to, but heck - it works) |
317 | */ | |
617ba13b | 318 | ext4_orphan_del(handle, inode); |
5ffff834 | 319 | EXT4_I(inode)->i_dtime = (__u32)ktime_get_real_seconds(); |
ac27a0ec DK |
320 | |
321 | /* | |
322 | * One subtle ordering requirement: if anything has gone wrong | |
323 | * (transaction abort, IO errors, whatever), then we can still | |
324 | * do these next steps (the fs will already have been marked as | |
325 | * having errors), but we can't free the inode if the mark_dirty | |
326 | * fails. | |
327 | */ | |
617ba13b | 328 | if (ext4_mark_inode_dirty(handle, inode)) |
ac27a0ec | 329 | /* If that failed, just do the required in-core inode clear. */ |
0930fcc1 | 330 | ext4_clear_inode(inode); |
ac27a0ec | 331 | else |
617ba13b MC |
332 | ext4_free_inode(handle, inode); |
333 | ext4_journal_stop(handle); | |
46e294ef JK |
334 | if (freeze_protected) |
335 | sb_end_intwrite(inode->i_sb); | |
0421a189 | 336 | ext4_xattr_inode_array_free(ea_inode_array); |
ac27a0ec DK |
337 | return; |
338 | no_delete: | |
b21ebf14 | 339 | if (!list_empty(&EXT4_I(inode)->i_fc_list)) |
e5aed39e | 340 | ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_NOMEM, NULL); |
0930fcc1 | 341 | ext4_clear_inode(inode); /* We must guarantee clearing of inode... */ |
ac27a0ec DK |
342 | } |
343 | ||
a9e7f447 DM |
344 | #ifdef CONFIG_QUOTA |
345 | qsize_t *ext4_get_reserved_space(struct inode *inode) | |
60e58e0f | 346 | { |
a9e7f447 | 347 | return &EXT4_I(inode)->i_reserved_quota; |
60e58e0f | 348 | } |
a9e7f447 | 349 | #endif |
9d0be502 | 350 | |
0637c6f4 TT |
351 | /* |
352 | * Called with i_data_sem down, which is important since we can call | |
353 | * ext4_discard_preallocations() from here. | |
354 | */ | |
5f634d06 AK |
355 | void ext4_da_update_reserve_space(struct inode *inode, |
356 | int used, int quota_claim) | |
12219aea AK |
357 | { |
358 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0637c6f4 | 359 | struct ext4_inode_info *ei = EXT4_I(inode); |
0637c6f4 TT |
360 | |
361 | spin_lock(&ei->i_block_reservation_lock); | |
d8990240 | 362 | trace_ext4_da_update_reserve_space(inode, used, quota_claim); |
0637c6f4 | 363 | if (unlikely(used > ei->i_reserved_data_blocks)) { |
8de5c325 | 364 | ext4_warning(inode->i_sb, "%s: ino %lu, used %d " |
1084f252 | 365 | "with only %d reserved data blocks", |
0637c6f4 TT |
366 | __func__, inode->i_ino, used, |
367 | ei->i_reserved_data_blocks); | |
368 | WARN_ON(1); | |
369 | used = ei->i_reserved_data_blocks; | |
370 | } | |
12219aea | 371 | |
0637c6f4 TT |
372 | /* Update per-inode reservations */ |
373 | ei->i_reserved_data_blocks -= used; | |
71d4f7d0 | 374 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, used); |
6bc6e63f | 375 | |
f9505c72 | 376 | spin_unlock(&ei->i_block_reservation_lock); |
60e58e0f | 377 | |
72b8ab9d ES |
378 | /* Update quota subsystem for data blocks */ |
379 | if (quota_claim) | |
7b415bf6 | 380 | dquot_claim_block(inode, EXT4_C2B(sbi, used)); |
72b8ab9d | 381 | else { |
5f634d06 AK |
382 | /* |
383 | * We did fallocate with an offset that is already delayed | |
384 | * allocated. So on delayed allocated writeback we should | |
72b8ab9d | 385 | * not re-claim the quota for fallocated blocks. |
5f634d06 | 386 | */ |
7b415bf6 | 387 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, used)); |
5f634d06 | 388 | } |
d6014301 AK |
389 | |
390 | /* | |
391 | * If we have done all the pending block allocations and if | |
392 | * there aren't any writers on the inode, we can discard the | |
393 | * inode's preallocations. | |
394 | */ | |
0637c6f4 | 395 | if ((ei->i_reserved_data_blocks == 0) && |
82dd124c | 396 | !inode_is_open_for_write(inode)) |
27bc446e | 397 | ext4_discard_preallocations(inode, 0); |
12219aea AK |
398 | } |
399 | ||
e29136f8 | 400 | static int __check_block_validity(struct inode *inode, const char *func, |
c398eda0 TT |
401 | unsigned int line, |
402 | struct ext4_map_blocks *map) | |
6fd058f7 | 403 | { |
345c0dbf TT |
404 | if (ext4_has_feature_journal(inode->i_sb) && |
405 | (inode->i_ino == | |
406 | le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_journal_inum))) | |
407 | return 0; | |
ce9f24cc | 408 | if (!ext4_inode_block_valid(inode, map->m_pblk, map->m_len)) { |
c398eda0 | 409 | ext4_error_inode(inode, func, line, map->m_pblk, |
bdbd6ce0 | 410 | "lblock %lu mapped to illegal pblock %llu " |
c398eda0 | 411 | "(length %d)", (unsigned long) map->m_lblk, |
bdbd6ce0 | 412 | map->m_pblk, map->m_len); |
6a797d27 | 413 | return -EFSCORRUPTED; |
6fd058f7 TT |
414 | } |
415 | return 0; | |
416 | } | |
417 | ||
53085fac JK |
418 | int ext4_issue_zeroout(struct inode *inode, ext4_lblk_t lblk, ext4_fsblk_t pblk, |
419 | ext4_lblk_t len) | |
420 | { | |
421 | int ret; | |
422 | ||
33b4cc25 | 423 | if (IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode)) |
a7550b30 | 424 | return fscrypt_zeroout_range(inode, lblk, pblk, len); |
53085fac JK |
425 | |
426 | ret = sb_issue_zeroout(inode->i_sb, pblk, len, GFP_NOFS); | |
427 | if (ret > 0) | |
428 | ret = 0; | |
429 | ||
430 | return ret; | |
431 | } | |
432 | ||
e29136f8 | 433 | #define check_block_validity(inode, map) \ |
c398eda0 | 434 | __check_block_validity((inode), __func__, __LINE__, (map)) |
e29136f8 | 435 | |
921f266b DM |
436 | #ifdef ES_AGGRESSIVE_TEST |
437 | static void ext4_map_blocks_es_recheck(handle_t *handle, | |
438 | struct inode *inode, | |
439 | struct ext4_map_blocks *es_map, | |
440 | struct ext4_map_blocks *map, | |
441 | int flags) | |
442 | { | |
443 | int retval; | |
444 | ||
445 | map->m_flags = 0; | |
446 | /* | |
447 | * There is a race window that the result is not the same. | |
448 | * e.g. xfstests #223 when dioread_nolock enables. The reason | |
449 | * is that we lookup a block mapping in extent status tree with | |
450 | * out taking i_data_sem. So at the time the unwritten extent | |
451 | * could be converted. | |
452 | */ | |
2dcba478 | 453 | down_read(&EXT4_I(inode)->i_data_sem); |
921f266b | 454 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
9e52484c | 455 | retval = ext4_ext_map_blocks(handle, inode, map, 0); |
921f266b | 456 | } else { |
9e52484c | 457 | retval = ext4_ind_map_blocks(handle, inode, map, 0); |
921f266b | 458 | } |
2dcba478 | 459 | up_read((&EXT4_I(inode)->i_data_sem)); |
921f266b DM |
460 | |
461 | /* | |
462 | * We don't check m_len because extent will be collpased in status | |
463 | * tree. So the m_len might not equal. | |
464 | */ | |
465 | if (es_map->m_lblk != map->m_lblk || | |
466 | es_map->m_flags != map->m_flags || | |
467 | es_map->m_pblk != map->m_pblk) { | |
bdafe42a | 468 | printk("ES cache assertion failed for inode: %lu " |
921f266b DM |
469 | "es_cached ex [%d/%d/%llu/%x] != " |
470 | "found ex [%d/%d/%llu/%x] retval %d flags %x\n", | |
471 | inode->i_ino, es_map->m_lblk, es_map->m_len, | |
472 | es_map->m_pblk, es_map->m_flags, map->m_lblk, | |
473 | map->m_len, map->m_pblk, map->m_flags, | |
474 | retval, flags); | |
475 | } | |
476 | } | |
477 | #endif /* ES_AGGRESSIVE_TEST */ | |
478 | ||
f5ab0d1f | 479 | /* |
e35fd660 | 480 | * The ext4_map_blocks() function tries to look up the requested blocks, |
2b2d6d01 | 481 | * and returns if the blocks are already mapped. |
f5ab0d1f | 482 | * |
f5ab0d1f MC |
483 | * Otherwise it takes the write lock of the i_data_sem and allocate blocks |
484 | * and store the allocated blocks in the result buffer head and mark it | |
485 | * mapped. | |
486 | * | |
e35fd660 TT |
487 | * If file type is extents based, it will call ext4_ext_map_blocks(), |
488 | * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping | |
f5ab0d1f MC |
489 | * based files |
490 | * | |
facab4d9 JK |
491 | * On success, it returns the number of blocks being mapped or allocated. if |
492 | * create==0 and the blocks are pre-allocated and unwritten, the resulting @map | |
493 | * is marked as unwritten. If the create == 1, it will mark @map as mapped. | |
f5ab0d1f MC |
494 | * |
495 | * It returns 0 if plain look up failed (blocks have not been allocated), in | |
facab4d9 JK |
496 | * that case, @map is returned as unmapped but we still do fill map->m_len to |
497 | * indicate the length of a hole starting at map->m_lblk. | |
f5ab0d1f MC |
498 | * |
499 | * It returns the error in case of allocation failure. | |
500 | */ | |
e35fd660 TT |
501 | int ext4_map_blocks(handle_t *handle, struct inode *inode, |
502 | struct ext4_map_blocks *map, int flags) | |
0e855ac8 | 503 | { |
d100eef2 | 504 | struct extent_status es; |
0e855ac8 | 505 | int retval; |
b8a86845 | 506 | int ret = 0; |
921f266b DM |
507 | #ifdef ES_AGGRESSIVE_TEST |
508 | struct ext4_map_blocks orig_map; | |
509 | ||
510 | memcpy(&orig_map, map, sizeof(*map)); | |
511 | #endif | |
f5ab0d1f | 512 | |
e35fd660 | 513 | map->m_flags = 0; |
70aa1554 RH |
514 | ext_debug(inode, "flag 0x%x, max_blocks %u, logical block %lu\n", |
515 | flags, map->m_len, (unsigned long) map->m_lblk); | |
d100eef2 | 516 | |
e861b5e9 TT |
517 | /* |
518 | * ext4_map_blocks returns an int, and m_len is an unsigned int | |
519 | */ | |
520 | if (unlikely(map->m_len > INT_MAX)) | |
521 | map->m_len = INT_MAX; | |
522 | ||
4adb6ab3 KM |
523 | /* We can handle the block number less than EXT_MAX_BLOCKS */ |
524 | if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS)) | |
6a797d27 | 525 | return -EFSCORRUPTED; |
4adb6ab3 | 526 | |
d100eef2 | 527 | /* Lookup extent status tree firstly */ |
8016e29f HS |
528 | if (!(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY) && |
529 | ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) { | |
d100eef2 ZL |
530 | if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) { |
531 | map->m_pblk = ext4_es_pblock(&es) + | |
532 | map->m_lblk - es.es_lblk; | |
533 | map->m_flags |= ext4_es_is_written(&es) ? | |
534 | EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN; | |
535 | retval = es.es_len - (map->m_lblk - es.es_lblk); | |
536 | if (retval > map->m_len) | |
537 | retval = map->m_len; | |
538 | map->m_len = retval; | |
539 | } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) { | |
facab4d9 JK |
540 | map->m_pblk = 0; |
541 | retval = es.es_len - (map->m_lblk - es.es_lblk); | |
542 | if (retval > map->m_len) | |
543 | retval = map->m_len; | |
544 | map->m_len = retval; | |
d100eef2 ZL |
545 | retval = 0; |
546 | } else { | |
1e83bc81 | 547 | BUG(); |
d100eef2 | 548 | } |
921f266b DM |
549 | #ifdef ES_AGGRESSIVE_TEST |
550 | ext4_map_blocks_es_recheck(handle, inode, map, | |
551 | &orig_map, flags); | |
552 | #endif | |
d100eef2 ZL |
553 | goto found; |
554 | } | |
555 | ||
4df3d265 | 556 | /* |
b920c755 TT |
557 | * Try to see if we can get the block without requesting a new |
558 | * file system block. | |
4df3d265 | 559 | */ |
2dcba478 | 560 | down_read(&EXT4_I(inode)->i_data_sem); |
12e9b892 | 561 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
9e52484c | 562 | retval = ext4_ext_map_blocks(handle, inode, map, 0); |
0e855ac8 | 563 | } else { |
9e52484c | 564 | retval = ext4_ind_map_blocks(handle, inode, map, 0); |
0e855ac8 | 565 | } |
f7fec032 | 566 | if (retval > 0) { |
3be78c73 | 567 | unsigned int status; |
f7fec032 | 568 | |
44fb851d ZL |
569 | if (unlikely(retval != map->m_len)) { |
570 | ext4_warning(inode->i_sb, | |
571 | "ES len assertion failed for inode " | |
572 | "%lu: retval %d != map->m_len %d", | |
573 | inode->i_ino, retval, map->m_len); | |
574 | WARN_ON(1); | |
921f266b | 575 | } |
921f266b | 576 | |
f7fec032 ZL |
577 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
578 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; | |
579 | if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) && | |
d2dc317d | 580 | !(status & EXTENT_STATUS_WRITTEN) && |
ad431025 EW |
581 | ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk, |
582 | map->m_lblk + map->m_len - 1)) | |
f7fec032 ZL |
583 | status |= EXTENT_STATUS_DELAYED; |
584 | ret = ext4_es_insert_extent(inode, map->m_lblk, | |
585 | map->m_len, map->m_pblk, status); | |
586 | if (ret < 0) | |
587 | retval = ret; | |
588 | } | |
2dcba478 | 589 | up_read((&EXT4_I(inode)->i_data_sem)); |
f5ab0d1f | 590 | |
d100eef2 | 591 | found: |
e35fd660 | 592 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
b8a86845 | 593 | ret = check_block_validity(inode, map); |
6fd058f7 TT |
594 | if (ret != 0) |
595 | return ret; | |
596 | } | |
597 | ||
f5ab0d1f | 598 | /* If it is only a block(s) look up */ |
c2177057 | 599 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) |
f5ab0d1f MC |
600 | return retval; |
601 | ||
602 | /* | |
603 | * Returns if the blocks have already allocated | |
604 | * | |
605 | * Note that if blocks have been preallocated | |
df3ab170 | 606 | * ext4_ext_get_block() returns the create = 0 |
f5ab0d1f MC |
607 | * with buffer head unmapped. |
608 | */ | |
e35fd660 | 609 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) |
b8a86845 LC |
610 | /* |
611 | * If we need to convert extent to unwritten | |
612 | * we continue and do the actual work in | |
613 | * ext4_ext_map_blocks() | |
614 | */ | |
615 | if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) | |
616 | return retval; | |
4df3d265 | 617 | |
2a8964d6 | 618 | /* |
a25a4e1a ZL |
619 | * Here we clear m_flags because after allocating an new extent, |
620 | * it will be set again. | |
2a8964d6 | 621 | */ |
a25a4e1a | 622 | map->m_flags &= ~EXT4_MAP_FLAGS; |
2a8964d6 | 623 | |
4df3d265 | 624 | /* |
556615dc | 625 | * New blocks allocate and/or writing to unwritten extent |
f5ab0d1f | 626 | * will possibly result in updating i_data, so we take |
d91bd2c1 | 627 | * the write lock of i_data_sem, and call get_block() |
f5ab0d1f | 628 | * with create == 1 flag. |
4df3d265 | 629 | */ |
c8b459f4 | 630 | down_write(&EXT4_I(inode)->i_data_sem); |
d2a17637 | 631 | |
4df3d265 AK |
632 | /* |
633 | * We need to check for EXT4 here because migrate | |
634 | * could have changed the inode type in between | |
635 | */ | |
12e9b892 | 636 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
e35fd660 | 637 | retval = ext4_ext_map_blocks(handle, inode, map, flags); |
0e855ac8 | 638 | } else { |
e35fd660 | 639 | retval = ext4_ind_map_blocks(handle, inode, map, flags); |
267e4db9 | 640 | |
e35fd660 | 641 | if (retval > 0 && map->m_flags & EXT4_MAP_NEW) { |
267e4db9 AK |
642 | /* |
643 | * We allocated new blocks which will result in | |
644 | * i_data's format changing. Force the migrate | |
645 | * to fail by clearing migrate flags | |
646 | */ | |
19f5fb7a | 647 | ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE); |
267e4db9 | 648 | } |
d2a17637 | 649 | |
5f634d06 AK |
650 | /* |
651 | * Update reserved blocks/metadata blocks after successful | |
652 | * block allocation which had been deferred till now. We don't | |
653 | * support fallocate for non extent files. So we can update | |
654 | * reserve space here. | |
655 | */ | |
656 | if ((retval > 0) && | |
1296cc85 | 657 | (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)) |
5f634d06 AK |
658 | ext4_da_update_reserve_space(inode, retval, 1); |
659 | } | |
2ac3b6e0 | 660 | |
f7fec032 | 661 | if (retval > 0) { |
3be78c73 | 662 | unsigned int status; |
f7fec032 | 663 | |
44fb851d ZL |
664 | if (unlikely(retval != map->m_len)) { |
665 | ext4_warning(inode->i_sb, | |
666 | "ES len assertion failed for inode " | |
667 | "%lu: retval %d != map->m_len %d", | |
668 | inode->i_ino, retval, map->m_len); | |
669 | WARN_ON(1); | |
921f266b | 670 | } |
921f266b | 671 | |
c86d8db3 JK |
672 | /* |
673 | * We have to zeroout blocks before inserting them into extent | |
674 | * status tree. Otherwise someone could look them up there and | |
9b623df6 JK |
675 | * use them before they are really zeroed. We also have to |
676 | * unmap metadata before zeroing as otherwise writeback can | |
677 | * overwrite zeros with stale data from block device. | |
c86d8db3 JK |
678 | */ |
679 | if (flags & EXT4_GET_BLOCKS_ZERO && | |
680 | map->m_flags & EXT4_MAP_MAPPED && | |
681 | map->m_flags & EXT4_MAP_NEW) { | |
682 | ret = ext4_issue_zeroout(inode, map->m_lblk, | |
683 | map->m_pblk, map->m_len); | |
684 | if (ret) { | |
685 | retval = ret; | |
686 | goto out_sem; | |
687 | } | |
688 | } | |
689 | ||
adb23551 ZL |
690 | /* |
691 | * If the extent has been zeroed out, we don't need to update | |
692 | * extent status tree. | |
693 | */ | |
694 | if ((flags & EXT4_GET_BLOCKS_PRE_IO) && | |
bb5835ed | 695 | ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) { |
adb23551 | 696 | if (ext4_es_is_written(&es)) |
c86d8db3 | 697 | goto out_sem; |
adb23551 | 698 | } |
f7fec032 ZL |
699 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
700 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; | |
701 | if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) && | |
d2dc317d | 702 | !(status & EXTENT_STATUS_WRITTEN) && |
ad431025 EW |
703 | ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk, |
704 | map->m_lblk + map->m_len - 1)) | |
f7fec032 ZL |
705 | status |= EXTENT_STATUS_DELAYED; |
706 | ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, | |
707 | map->m_pblk, status); | |
c86d8db3 | 708 | if (ret < 0) { |
f7fec032 | 709 | retval = ret; |
c86d8db3 JK |
710 | goto out_sem; |
711 | } | |
5356f261 AK |
712 | } |
713 | ||
c86d8db3 | 714 | out_sem: |
4df3d265 | 715 | up_write((&EXT4_I(inode)->i_data_sem)); |
e35fd660 | 716 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
b8a86845 | 717 | ret = check_block_validity(inode, map); |
6fd058f7 TT |
718 | if (ret != 0) |
719 | return ret; | |
06bd3c36 JK |
720 | |
721 | /* | |
722 | * Inodes with freshly allocated blocks where contents will be | |
723 | * visible after transaction commit must be on transaction's | |
724 | * ordered data list. | |
725 | */ | |
726 | if (map->m_flags & EXT4_MAP_NEW && | |
727 | !(map->m_flags & EXT4_MAP_UNWRITTEN) && | |
728 | !(flags & EXT4_GET_BLOCKS_ZERO) && | |
02749a4c | 729 | !ext4_is_quota_file(inode) && |
06bd3c36 | 730 | ext4_should_order_data(inode)) { |
73131fbb RZ |
731 | loff_t start_byte = |
732 | (loff_t)map->m_lblk << inode->i_blkbits; | |
733 | loff_t length = (loff_t)map->m_len << inode->i_blkbits; | |
734 | ||
ee0876bc | 735 | if (flags & EXT4_GET_BLOCKS_IO_SUBMIT) |
73131fbb RZ |
736 | ret = ext4_jbd2_inode_add_wait(handle, inode, |
737 | start_byte, length); | |
ee0876bc | 738 | else |
73131fbb RZ |
739 | ret = ext4_jbd2_inode_add_write(handle, inode, |
740 | start_byte, length); | |
06bd3c36 JK |
741 | if (ret) |
742 | return ret; | |
743 | } | |
6fd058f7 | 744 | } |
f7914e07 XY |
745 | if (retval > 0 && (map->m_flags & EXT4_MAP_UNWRITTEN || |
746 | map->m_flags & EXT4_MAP_MAPPED)) | |
747 | ext4_fc_track_range(handle, inode, map->m_lblk, | |
748 | map->m_lblk + map->m_len - 1); | |
ec8c60be | 749 | if (retval < 0) |
70aa1554 | 750 | ext_debug(inode, "failed with err %d\n", retval); |
0e855ac8 AK |
751 | return retval; |
752 | } | |
753 | ||
ed8ad838 JK |
754 | /* |
755 | * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages | |
756 | * we have to be careful as someone else may be manipulating b_state as well. | |
757 | */ | |
758 | static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags) | |
759 | { | |
760 | unsigned long old_state; | |
761 | unsigned long new_state; | |
762 | ||
763 | flags &= EXT4_MAP_FLAGS; | |
764 | ||
765 | /* Dummy buffer_head? Set non-atomically. */ | |
766 | if (!bh->b_page) { | |
767 | bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags; | |
768 | return; | |
769 | } | |
770 | /* | |
771 | * Someone else may be modifying b_state. Be careful! This is ugly but | |
772 | * once we get rid of using bh as a container for mapping information | |
773 | * to pass to / from get_block functions, this can go away. | |
774 | */ | |
775 | do { | |
776 | old_state = READ_ONCE(bh->b_state); | |
777 | new_state = (old_state & ~EXT4_MAP_FLAGS) | flags; | |
778 | } while (unlikely( | |
779 | cmpxchg(&bh->b_state, old_state, new_state) != old_state)); | |
780 | } | |
781 | ||
2ed88685 TT |
782 | static int _ext4_get_block(struct inode *inode, sector_t iblock, |
783 | struct buffer_head *bh, int flags) | |
ac27a0ec | 784 | { |
2ed88685 | 785 | struct ext4_map_blocks map; |
efe70c29 | 786 | int ret = 0; |
ac27a0ec | 787 | |
46c7f254 TM |
788 | if (ext4_has_inline_data(inode)) |
789 | return -ERANGE; | |
790 | ||
2ed88685 TT |
791 | map.m_lblk = iblock; |
792 | map.m_len = bh->b_size >> inode->i_blkbits; | |
793 | ||
efe70c29 JK |
794 | ret = ext4_map_blocks(ext4_journal_current_handle(), inode, &map, |
795 | flags); | |
7fb5409d | 796 | if (ret > 0) { |
2ed88685 | 797 | map_bh(bh, inode->i_sb, map.m_pblk); |
ed8ad838 | 798 | ext4_update_bh_state(bh, map.m_flags); |
2ed88685 | 799 | bh->b_size = inode->i_sb->s_blocksize * map.m_len; |
7fb5409d | 800 | ret = 0; |
547edce3 RZ |
801 | } else if (ret == 0) { |
802 | /* hole case, need to fill in bh->b_size */ | |
803 | bh->b_size = inode->i_sb->s_blocksize * map.m_len; | |
ac27a0ec DK |
804 | } |
805 | return ret; | |
806 | } | |
807 | ||
2ed88685 TT |
808 | int ext4_get_block(struct inode *inode, sector_t iblock, |
809 | struct buffer_head *bh, int create) | |
810 | { | |
811 | return _ext4_get_block(inode, iblock, bh, | |
812 | create ? EXT4_GET_BLOCKS_CREATE : 0); | |
813 | } | |
814 | ||
705965bd JK |
815 | /* |
816 | * Get block function used when preparing for buffered write if we require | |
817 | * creating an unwritten extent if blocks haven't been allocated. The extent | |
818 | * will be converted to written after the IO is complete. | |
819 | */ | |
820 | int ext4_get_block_unwritten(struct inode *inode, sector_t iblock, | |
821 | struct buffer_head *bh_result, int create) | |
822 | { | |
823 | ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n", | |
824 | inode->i_ino, create); | |
825 | return _ext4_get_block(inode, iblock, bh_result, | |
826 | EXT4_GET_BLOCKS_IO_CREATE_EXT); | |
827 | } | |
828 | ||
efe70c29 JK |
829 | /* Maximum number of blocks we map for direct IO at once. */ |
830 | #define DIO_MAX_BLOCKS 4096 | |
831 | ||
ac27a0ec DK |
832 | /* |
833 | * `handle' can be NULL if create is zero | |
834 | */ | |
617ba13b | 835 | struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode, |
c5e298ae | 836 | ext4_lblk_t block, int map_flags) |
ac27a0ec | 837 | { |
2ed88685 TT |
838 | struct ext4_map_blocks map; |
839 | struct buffer_head *bh; | |
c5e298ae | 840 | int create = map_flags & EXT4_GET_BLOCKS_CREATE; |
10560082 | 841 | int err; |
ac27a0ec | 842 | |
837c23fb CX |
843 | ASSERT((EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY) |
844 | || handle != NULL || create == 0); | |
ac27a0ec | 845 | |
2ed88685 TT |
846 | map.m_lblk = block; |
847 | map.m_len = 1; | |
c5e298ae | 848 | err = ext4_map_blocks(handle, inode, &map, map_flags); |
ac27a0ec | 849 | |
10560082 TT |
850 | if (err == 0) |
851 | return create ? ERR_PTR(-ENOSPC) : NULL; | |
2ed88685 | 852 | if (err < 0) |
10560082 | 853 | return ERR_PTR(err); |
2ed88685 TT |
854 | |
855 | bh = sb_getblk(inode->i_sb, map.m_pblk); | |
10560082 TT |
856 | if (unlikely(!bh)) |
857 | return ERR_PTR(-ENOMEM); | |
2ed88685 | 858 | if (map.m_flags & EXT4_MAP_NEW) { |
837c23fb CX |
859 | ASSERT(create != 0); |
860 | ASSERT((EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY) | |
861 | || (handle != NULL)); | |
ac27a0ec | 862 | |
2ed88685 TT |
863 | /* |
864 | * Now that we do not always journal data, we should | |
865 | * keep in mind whether this should always journal the | |
866 | * new buffer as metadata. For now, regular file | |
867 | * writes use ext4_get_block instead, so it's not a | |
868 | * problem. | |
869 | */ | |
870 | lock_buffer(bh); | |
871 | BUFFER_TRACE(bh, "call get_create_access"); | |
188c299e JK |
872 | err = ext4_journal_get_create_access(handle, inode->i_sb, bh, |
873 | EXT4_JTR_NONE); | |
10560082 TT |
874 | if (unlikely(err)) { |
875 | unlock_buffer(bh); | |
876 | goto errout; | |
877 | } | |
878 | if (!buffer_uptodate(bh)) { | |
2ed88685 TT |
879 | memset(bh->b_data, 0, inode->i_sb->s_blocksize); |
880 | set_buffer_uptodate(bh); | |
ac27a0ec | 881 | } |
2ed88685 TT |
882 | unlock_buffer(bh); |
883 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); | |
884 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
10560082 TT |
885 | if (unlikely(err)) |
886 | goto errout; | |
887 | } else | |
2ed88685 | 888 | BUFFER_TRACE(bh, "not a new buffer"); |
2ed88685 | 889 | return bh; |
10560082 TT |
890 | errout: |
891 | brelse(bh); | |
892 | return ERR_PTR(err); | |
ac27a0ec DK |
893 | } |
894 | ||
617ba13b | 895 | struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode, |
c5e298ae | 896 | ext4_lblk_t block, int map_flags) |
ac27a0ec | 897 | { |
af5bc92d | 898 | struct buffer_head *bh; |
2d069c08 | 899 | int ret; |
ac27a0ec | 900 | |
c5e298ae | 901 | bh = ext4_getblk(handle, inode, block, map_flags); |
1c215028 | 902 | if (IS_ERR(bh)) |
ac27a0ec | 903 | return bh; |
7963e5ac | 904 | if (!bh || ext4_buffer_uptodate(bh)) |
ac27a0ec | 905 | return bh; |
2d069c08 | 906 | |
907 | ret = ext4_read_bh_lock(bh, REQ_META | REQ_PRIO, true); | |
908 | if (ret) { | |
909 | put_bh(bh); | |
910 | return ERR_PTR(ret); | |
911 | } | |
912 | return bh; | |
ac27a0ec DK |
913 | } |
914 | ||
9699d4f9 TE |
915 | /* Read a contiguous batch of blocks. */ |
916 | int ext4_bread_batch(struct inode *inode, ext4_lblk_t block, int bh_count, | |
917 | bool wait, struct buffer_head **bhs) | |
918 | { | |
919 | int i, err; | |
920 | ||
921 | for (i = 0; i < bh_count; i++) { | |
922 | bhs[i] = ext4_getblk(NULL, inode, block + i, 0 /* map_flags */); | |
923 | if (IS_ERR(bhs[i])) { | |
924 | err = PTR_ERR(bhs[i]); | |
925 | bh_count = i; | |
926 | goto out_brelse; | |
927 | } | |
928 | } | |
929 | ||
930 | for (i = 0; i < bh_count; i++) | |
931 | /* Note that NULL bhs[i] is valid because of holes. */ | |
2d069c08 | 932 | if (bhs[i] && !ext4_buffer_uptodate(bhs[i])) |
933 | ext4_read_bh_lock(bhs[i], REQ_META | REQ_PRIO, false); | |
9699d4f9 TE |
934 | |
935 | if (!wait) | |
936 | return 0; | |
937 | ||
938 | for (i = 0; i < bh_count; i++) | |
939 | if (bhs[i]) | |
940 | wait_on_buffer(bhs[i]); | |
941 | ||
942 | for (i = 0; i < bh_count; i++) { | |
943 | if (bhs[i] && !buffer_uptodate(bhs[i])) { | |
944 | err = -EIO; | |
945 | goto out_brelse; | |
946 | } | |
947 | } | |
948 | return 0; | |
949 | ||
950 | out_brelse: | |
951 | for (i = 0; i < bh_count; i++) { | |
952 | brelse(bhs[i]); | |
953 | bhs[i] = NULL; | |
954 | } | |
955 | return err; | |
956 | } | |
957 | ||
188c299e | 958 | int ext4_walk_page_buffers(handle_t *handle, struct inode *inode, |
f19d5870 TM |
959 | struct buffer_head *head, |
960 | unsigned from, | |
961 | unsigned to, | |
962 | int *partial, | |
188c299e | 963 | int (*fn)(handle_t *handle, struct inode *inode, |
f19d5870 | 964 | struct buffer_head *bh)) |
ac27a0ec DK |
965 | { |
966 | struct buffer_head *bh; | |
967 | unsigned block_start, block_end; | |
968 | unsigned blocksize = head->b_size; | |
969 | int err, ret = 0; | |
970 | struct buffer_head *next; | |
971 | ||
af5bc92d TT |
972 | for (bh = head, block_start = 0; |
973 | ret == 0 && (bh != head || !block_start); | |
de9a55b8 | 974 | block_start = block_end, bh = next) { |
ac27a0ec DK |
975 | next = bh->b_this_page; |
976 | block_end = block_start + blocksize; | |
977 | if (block_end <= from || block_start >= to) { | |
978 | if (partial && !buffer_uptodate(bh)) | |
979 | *partial = 1; | |
980 | continue; | |
981 | } | |
188c299e | 982 | err = (*fn)(handle, inode, bh); |
ac27a0ec DK |
983 | if (!ret) |
984 | ret = err; | |
985 | } | |
986 | return ret; | |
987 | } | |
988 | ||
989 | /* | |
990 | * To preserve ordering, it is essential that the hole instantiation and | |
991 | * the data write be encapsulated in a single transaction. We cannot | |
617ba13b | 992 | * close off a transaction and start a new one between the ext4_get_block() |
dab291af | 993 | * and the commit_write(). So doing the jbd2_journal_start at the start of |
ac27a0ec DK |
994 | * prepare_write() is the right place. |
995 | * | |
36ade451 JK |
996 | * Also, this function can nest inside ext4_writepage(). In that case, we |
997 | * *know* that ext4_writepage() has generated enough buffer credits to do the | |
998 | * whole page. So we won't block on the journal in that case, which is good, | |
999 | * because the caller may be PF_MEMALLOC. | |
ac27a0ec | 1000 | * |
617ba13b | 1001 | * By accident, ext4 can be reentered when a transaction is open via |
ac27a0ec DK |
1002 | * quota file writes. If we were to commit the transaction while thus |
1003 | * reentered, there can be a deadlock - we would be holding a quota | |
1004 | * lock, and the commit would never complete if another thread had a | |
1005 | * transaction open and was blocking on the quota lock - a ranking | |
1006 | * violation. | |
1007 | * | |
dab291af | 1008 | * So what we do is to rely on the fact that jbd2_journal_stop/journal_start |
ac27a0ec DK |
1009 | * will _not_ run commit under these circumstances because handle->h_ref |
1010 | * is elevated. We'll still have enough credits for the tiny quotafile | |
1011 | * write. | |
1012 | */ | |
188c299e | 1013 | int do_journal_get_write_access(handle_t *handle, struct inode *inode, |
f19d5870 | 1014 | struct buffer_head *bh) |
ac27a0ec | 1015 | { |
56d35a4c JK |
1016 | int dirty = buffer_dirty(bh); |
1017 | int ret; | |
1018 | ||
ac27a0ec DK |
1019 | if (!buffer_mapped(bh) || buffer_freed(bh)) |
1020 | return 0; | |
56d35a4c | 1021 | /* |
ebdec241 | 1022 | * __block_write_begin() could have dirtied some buffers. Clean |
56d35a4c JK |
1023 | * the dirty bit as jbd2_journal_get_write_access() could complain |
1024 | * otherwise about fs integrity issues. Setting of the dirty bit | |
ebdec241 | 1025 | * by __block_write_begin() isn't a real problem here as we clear |
56d35a4c JK |
1026 | * the bit before releasing a page lock and thus writeback cannot |
1027 | * ever write the buffer. | |
1028 | */ | |
1029 | if (dirty) | |
1030 | clear_buffer_dirty(bh); | |
5d601255 | 1031 | BUFFER_TRACE(bh, "get write access"); |
188c299e JK |
1032 | ret = ext4_journal_get_write_access(handle, inode->i_sb, bh, |
1033 | EXT4_JTR_NONE); | |
56d35a4c JK |
1034 | if (!ret && dirty) |
1035 | ret = ext4_handle_dirty_metadata(handle, NULL, bh); | |
1036 | return ret; | |
ac27a0ec DK |
1037 | } |
1038 | ||
643fa961 | 1039 | #ifdef CONFIG_FS_ENCRYPTION |
2058f83a MH |
1040 | static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len, |
1041 | get_block_t *get_block) | |
1042 | { | |
09cbfeaf | 1043 | unsigned from = pos & (PAGE_SIZE - 1); |
2058f83a MH |
1044 | unsigned to = from + len; |
1045 | struct inode *inode = page->mapping->host; | |
1046 | unsigned block_start, block_end; | |
1047 | sector_t block; | |
1048 | int err = 0; | |
1049 | unsigned blocksize = inode->i_sb->s_blocksize; | |
1050 | unsigned bbits; | |
0b578f35 CR |
1051 | struct buffer_head *bh, *head, *wait[2]; |
1052 | int nr_wait = 0; | |
1053 | int i; | |
2058f83a MH |
1054 | |
1055 | BUG_ON(!PageLocked(page)); | |
09cbfeaf KS |
1056 | BUG_ON(from > PAGE_SIZE); |
1057 | BUG_ON(to > PAGE_SIZE); | |
2058f83a MH |
1058 | BUG_ON(from > to); |
1059 | ||
1060 | if (!page_has_buffers(page)) | |
1061 | create_empty_buffers(page, blocksize, 0); | |
1062 | head = page_buffers(page); | |
1063 | bbits = ilog2(blocksize); | |
09cbfeaf | 1064 | block = (sector_t)page->index << (PAGE_SHIFT - bbits); |
2058f83a MH |
1065 | |
1066 | for (bh = head, block_start = 0; bh != head || !block_start; | |
1067 | block++, block_start = block_end, bh = bh->b_this_page) { | |
1068 | block_end = block_start + blocksize; | |
1069 | if (block_end <= from || block_start >= to) { | |
1070 | if (PageUptodate(page)) { | |
3cd46171 | 1071 | set_buffer_uptodate(bh); |
2058f83a MH |
1072 | } |
1073 | continue; | |
1074 | } | |
1075 | if (buffer_new(bh)) | |
1076 | clear_buffer_new(bh); | |
1077 | if (!buffer_mapped(bh)) { | |
1078 | WARN_ON(bh->b_size != blocksize); | |
1079 | err = get_block(inode, block, bh, 1); | |
1080 | if (err) | |
1081 | break; | |
1082 | if (buffer_new(bh)) { | |
2058f83a MH |
1083 | if (PageUptodate(page)) { |
1084 | clear_buffer_new(bh); | |
1085 | set_buffer_uptodate(bh); | |
1086 | mark_buffer_dirty(bh); | |
1087 | continue; | |
1088 | } | |
1089 | if (block_end > to || block_start < from) | |
1090 | zero_user_segments(page, to, block_end, | |
1091 | block_start, from); | |
1092 | continue; | |
1093 | } | |
1094 | } | |
1095 | if (PageUptodate(page)) { | |
3cd46171 | 1096 | set_buffer_uptodate(bh); |
2058f83a MH |
1097 | continue; |
1098 | } | |
1099 | if (!buffer_uptodate(bh) && !buffer_delay(bh) && | |
1100 | !buffer_unwritten(bh) && | |
1101 | (block_start < from || block_end > to)) { | |
2d069c08 | 1102 | ext4_read_bh_lock(bh, 0, false); |
0b578f35 | 1103 | wait[nr_wait++] = bh; |
2058f83a MH |
1104 | } |
1105 | } | |
1106 | /* | |
1107 | * If we issued read requests, let them complete. | |
1108 | */ | |
0b578f35 CR |
1109 | for (i = 0; i < nr_wait; i++) { |
1110 | wait_on_buffer(wait[i]); | |
1111 | if (!buffer_uptodate(wait[i])) | |
2058f83a MH |
1112 | err = -EIO; |
1113 | } | |
7e0785fc | 1114 | if (unlikely(err)) { |
2058f83a | 1115 | page_zero_new_buffers(page, from, to); |
4f74d15f | 1116 | } else if (fscrypt_inode_uses_fs_layer_crypto(inode)) { |
0b578f35 CR |
1117 | for (i = 0; i < nr_wait; i++) { |
1118 | int err2; | |
1119 | ||
1120 | err2 = fscrypt_decrypt_pagecache_blocks(page, blocksize, | |
1121 | bh_offset(wait[i])); | |
1122 | if (err2) { | |
1123 | clear_buffer_uptodate(wait[i]); | |
1124 | err = err2; | |
1125 | } | |
1126 | } | |
7e0785fc CR |
1127 | } |
1128 | ||
2058f83a MH |
1129 | return err; |
1130 | } | |
1131 | #endif | |
1132 | ||
bfc1af65 | 1133 | static int ext4_write_begin(struct file *file, struct address_space *mapping, |
de9a55b8 TT |
1134 | loff_t pos, unsigned len, unsigned flags, |
1135 | struct page **pagep, void **fsdata) | |
ac27a0ec | 1136 | { |
af5bc92d | 1137 | struct inode *inode = mapping->host; |
1938a150 | 1138 | int ret, needed_blocks; |
ac27a0ec DK |
1139 | handle_t *handle; |
1140 | int retries = 0; | |
af5bc92d | 1141 | struct page *page; |
de9a55b8 | 1142 | pgoff_t index; |
af5bc92d | 1143 | unsigned from, to; |
bfc1af65 | 1144 | |
0db1ff22 TT |
1145 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
1146 | return -EIO; | |
1147 | ||
9bffad1e | 1148 | trace_ext4_write_begin(inode, pos, len, flags); |
1938a150 AK |
1149 | /* |
1150 | * Reserve one block more for addition to orphan list in case | |
1151 | * we allocate blocks but write fails for some reason | |
1152 | */ | |
1153 | needed_blocks = ext4_writepage_trans_blocks(inode) + 1; | |
09cbfeaf KS |
1154 | index = pos >> PAGE_SHIFT; |
1155 | from = pos & (PAGE_SIZE - 1); | |
af5bc92d | 1156 | to = from + len; |
ac27a0ec | 1157 | |
f19d5870 TM |
1158 | if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { |
1159 | ret = ext4_try_to_write_inline_data(mapping, inode, pos, len, | |
1160 | flags, pagep); | |
1161 | if (ret < 0) | |
47564bfb TT |
1162 | return ret; |
1163 | if (ret == 1) | |
1164 | return 0; | |
f19d5870 TM |
1165 | } |
1166 | ||
47564bfb TT |
1167 | /* |
1168 | * grab_cache_page_write_begin() can take a long time if the | |
1169 | * system is thrashing due to memory pressure, or if the page | |
1170 | * is being written back. So grab it first before we start | |
1171 | * the transaction handle. This also allows us to allocate | |
1172 | * the page (if needed) without using GFP_NOFS. | |
1173 | */ | |
1174 | retry_grab: | |
1175 | page = grab_cache_page_write_begin(mapping, index, flags); | |
1176 | if (!page) | |
1177 | return -ENOMEM; | |
1178 | unlock_page(page); | |
1179 | ||
1180 | retry_journal: | |
9924a92a | 1181 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks); |
af5bc92d | 1182 | if (IS_ERR(handle)) { |
09cbfeaf | 1183 | put_page(page); |
47564bfb | 1184 | return PTR_ERR(handle); |
7479d2b9 | 1185 | } |
ac27a0ec | 1186 | |
47564bfb TT |
1187 | lock_page(page); |
1188 | if (page->mapping != mapping) { | |
1189 | /* The page got truncated from under us */ | |
1190 | unlock_page(page); | |
09cbfeaf | 1191 | put_page(page); |
cf108bca | 1192 | ext4_journal_stop(handle); |
47564bfb | 1193 | goto retry_grab; |
cf108bca | 1194 | } |
7afe5aa5 DM |
1195 | /* In case writeback began while the page was unlocked */ |
1196 | wait_for_stable_page(page); | |
cf108bca | 1197 | |
643fa961 | 1198 | #ifdef CONFIG_FS_ENCRYPTION |
2058f83a MH |
1199 | if (ext4_should_dioread_nolock(inode)) |
1200 | ret = ext4_block_write_begin(page, pos, len, | |
705965bd | 1201 | ext4_get_block_unwritten); |
2058f83a MH |
1202 | else |
1203 | ret = ext4_block_write_begin(page, pos, len, | |
1204 | ext4_get_block); | |
1205 | #else | |
744692dc | 1206 | if (ext4_should_dioread_nolock(inode)) |
705965bd JK |
1207 | ret = __block_write_begin(page, pos, len, |
1208 | ext4_get_block_unwritten); | |
744692dc | 1209 | else |
6e1db88d | 1210 | ret = __block_write_begin(page, pos, len, ext4_get_block); |
2058f83a | 1211 | #endif |
bfc1af65 | 1212 | if (!ret && ext4_should_journal_data(inode)) { |
188c299e JK |
1213 | ret = ext4_walk_page_buffers(handle, inode, |
1214 | page_buffers(page), from, to, NULL, | |
f19d5870 | 1215 | do_journal_get_write_access); |
ac27a0ec | 1216 | } |
bfc1af65 NP |
1217 | |
1218 | if (ret) { | |
c93d8f88 EB |
1219 | bool extended = (pos + len > inode->i_size) && |
1220 | !ext4_verity_in_progress(inode); | |
1221 | ||
af5bc92d | 1222 | unlock_page(page); |
ae4d5372 | 1223 | /* |
6e1db88d | 1224 | * __block_write_begin may have instantiated a few blocks |
ae4d5372 AK |
1225 | * outside i_size. Trim these off again. Don't need |
1226 | * i_size_read because we hold i_mutex. | |
1938a150 AK |
1227 | * |
1228 | * Add inode to orphan list in case we crash before | |
1229 | * truncate finishes | |
ae4d5372 | 1230 | */ |
c93d8f88 | 1231 | if (extended && ext4_can_truncate(inode)) |
1938a150 AK |
1232 | ext4_orphan_add(handle, inode); |
1233 | ||
1234 | ext4_journal_stop(handle); | |
c93d8f88 | 1235 | if (extended) { |
b9a4207d | 1236 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1237 | /* |
ffacfa7a | 1238 | * If truncate failed early the inode might |
1938a150 AK |
1239 | * still be on the orphan list; we need to |
1240 | * make sure the inode is removed from the | |
1241 | * orphan list in that case. | |
1242 | */ | |
1243 | if (inode->i_nlink) | |
1244 | ext4_orphan_del(NULL, inode); | |
1245 | } | |
bfc1af65 | 1246 | |
47564bfb TT |
1247 | if (ret == -ENOSPC && |
1248 | ext4_should_retry_alloc(inode->i_sb, &retries)) | |
1249 | goto retry_journal; | |
09cbfeaf | 1250 | put_page(page); |
47564bfb TT |
1251 | return ret; |
1252 | } | |
1253 | *pagep = page; | |
ac27a0ec DK |
1254 | return ret; |
1255 | } | |
1256 | ||
bfc1af65 | 1257 | /* For write_end() in data=journal mode */ |
188c299e JK |
1258 | static int write_end_fn(handle_t *handle, struct inode *inode, |
1259 | struct buffer_head *bh) | |
ac27a0ec | 1260 | { |
13fca323 | 1261 | int ret; |
ac27a0ec DK |
1262 | if (!buffer_mapped(bh) || buffer_freed(bh)) |
1263 | return 0; | |
1264 | set_buffer_uptodate(bh); | |
13fca323 TT |
1265 | ret = ext4_handle_dirty_metadata(handle, NULL, bh); |
1266 | clear_buffer_meta(bh); | |
1267 | clear_buffer_prio(bh); | |
1268 | return ret; | |
ac27a0ec DK |
1269 | } |
1270 | ||
eed4333f ZL |
1271 | /* |
1272 | * We need to pick up the new inode size which generic_commit_write gave us | |
1273 | * `file' can be NULL - eg, when called from page_symlink(). | |
1274 | * | |
1275 | * ext4 never places buffers on inode->i_mapping->private_list. metadata | |
1276 | * buffers are managed internally. | |
1277 | */ | |
1278 | static int ext4_write_end(struct file *file, | |
1279 | struct address_space *mapping, | |
1280 | loff_t pos, unsigned len, unsigned copied, | |
1281 | struct page *page, void *fsdata) | |
f8514083 | 1282 | { |
f8514083 | 1283 | handle_t *handle = ext4_journal_current_handle(); |
eed4333f | 1284 | struct inode *inode = mapping->host; |
0572639f | 1285 | loff_t old_size = inode->i_size; |
eed4333f ZL |
1286 | int ret = 0, ret2; |
1287 | int i_size_changed = 0; | |
c93d8f88 | 1288 | bool verity = ext4_verity_in_progress(inode); |
eed4333f ZL |
1289 | |
1290 | trace_ext4_write_end(inode, pos, len, copied); | |
6984aef5 ZY |
1291 | |
1292 | if (ext4_has_inline_data(inode)) | |
1293 | return ext4_write_inline_data_end(inode, pos, len, copied, page); | |
1294 | ||
1295 | copied = block_write_end(file, mapping, pos, len, copied, page, fsdata); | |
f8514083 | 1296 | /* |
4631dbf6 | 1297 | * it's important to update i_size while still holding page lock: |
f8514083 | 1298 | * page writeout could otherwise come in and zero beyond i_size. |
c93d8f88 EB |
1299 | * |
1300 | * If FS_IOC_ENABLE_VERITY is running on this inode, then Merkle tree | |
1301 | * blocks are being written past EOF, so skip the i_size update. | |
f8514083 | 1302 | */ |
c93d8f88 EB |
1303 | if (!verity) |
1304 | i_size_changed = ext4_update_inode_size(inode, pos + copied); | |
f8514083 | 1305 | unlock_page(page); |
09cbfeaf | 1306 | put_page(page); |
f8514083 | 1307 | |
c93d8f88 | 1308 | if (old_size < pos && !verity) |
0572639f | 1309 | pagecache_isize_extended(inode, old_size, pos); |
f8514083 AK |
1310 | /* |
1311 | * Don't mark the inode dirty under page lock. First, it unnecessarily | |
1312 | * makes the holding time of page lock longer. Second, it forces lock | |
1313 | * ordering of page lock and transaction start for journaling | |
1314 | * filesystems. | |
1315 | */ | |
6984aef5 | 1316 | if (i_size_changed) |
4209ae12 | 1317 | ret = ext4_mark_inode_dirty(handle, inode); |
f8514083 | 1318 | |
c93d8f88 | 1319 | if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode)) |
f8514083 AK |
1320 | /* if we have allocated more blocks and copied |
1321 | * less. We will have blocks allocated outside | |
1322 | * inode->i_size. So truncate them | |
1323 | */ | |
1324 | ext4_orphan_add(handle, inode); | |
55ce2f64 | 1325 | |
617ba13b | 1326 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1327 | if (!ret) |
1328 | ret = ret2; | |
bfc1af65 | 1329 | |
c93d8f88 | 1330 | if (pos + len > inode->i_size && !verity) { |
b9a4207d | 1331 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1332 | /* |
ffacfa7a | 1333 | * If truncate failed early the inode might still be |
f8514083 AK |
1334 | * on the orphan list; we need to make sure the inode |
1335 | * is removed from the orphan list in that case. | |
1336 | */ | |
1337 | if (inode->i_nlink) | |
1338 | ext4_orphan_del(NULL, inode); | |
1339 | } | |
1340 | ||
bfc1af65 | 1341 | return ret ? ret : copied; |
ac27a0ec DK |
1342 | } |
1343 | ||
b90197b6 TT |
1344 | /* |
1345 | * This is a private version of page_zero_new_buffers() which doesn't | |
1346 | * set the buffer to be dirty, since in data=journalled mode we need | |
1347 | * to call ext4_handle_dirty_metadata() instead. | |
1348 | */ | |
3b136499 | 1349 | static void ext4_journalled_zero_new_buffers(handle_t *handle, |
188c299e | 1350 | struct inode *inode, |
3b136499 JK |
1351 | struct page *page, |
1352 | unsigned from, unsigned to) | |
b90197b6 TT |
1353 | { |
1354 | unsigned int block_start = 0, block_end; | |
1355 | struct buffer_head *head, *bh; | |
1356 | ||
1357 | bh = head = page_buffers(page); | |
1358 | do { | |
1359 | block_end = block_start + bh->b_size; | |
1360 | if (buffer_new(bh)) { | |
1361 | if (block_end > from && block_start < to) { | |
1362 | if (!PageUptodate(page)) { | |
1363 | unsigned start, size; | |
1364 | ||
1365 | start = max(from, block_start); | |
1366 | size = min(to, block_end) - start; | |
1367 | ||
1368 | zero_user(page, start, size); | |
188c299e | 1369 | write_end_fn(handle, inode, bh); |
b90197b6 TT |
1370 | } |
1371 | clear_buffer_new(bh); | |
1372 | } | |
1373 | } | |
1374 | block_start = block_end; | |
1375 | bh = bh->b_this_page; | |
1376 | } while (bh != head); | |
1377 | } | |
1378 | ||
bfc1af65 | 1379 | static int ext4_journalled_write_end(struct file *file, |
de9a55b8 TT |
1380 | struct address_space *mapping, |
1381 | loff_t pos, unsigned len, unsigned copied, | |
1382 | struct page *page, void *fsdata) | |
ac27a0ec | 1383 | { |
617ba13b | 1384 | handle_t *handle = ext4_journal_current_handle(); |
bfc1af65 | 1385 | struct inode *inode = mapping->host; |
0572639f | 1386 | loff_t old_size = inode->i_size; |
ac27a0ec DK |
1387 | int ret = 0, ret2; |
1388 | int partial = 0; | |
bfc1af65 | 1389 | unsigned from, to; |
4631dbf6 | 1390 | int size_changed = 0; |
c93d8f88 | 1391 | bool verity = ext4_verity_in_progress(inode); |
ac27a0ec | 1392 | |
9bffad1e | 1393 | trace_ext4_journalled_write_end(inode, pos, len, copied); |
09cbfeaf | 1394 | from = pos & (PAGE_SIZE - 1); |
bfc1af65 NP |
1395 | to = from + len; |
1396 | ||
441c8508 CW |
1397 | BUG_ON(!ext4_handle_valid(handle)); |
1398 | ||
6984aef5 ZY |
1399 | if (ext4_has_inline_data(inode)) |
1400 | return ext4_write_inline_data_end(inode, pos, len, copied, page); | |
1401 | ||
1402 | if (unlikely(copied < len) && !PageUptodate(page)) { | |
3b136499 | 1403 | copied = 0; |
188c299e | 1404 | ext4_journalled_zero_new_buffers(handle, inode, page, from, to); |
3b136499 JK |
1405 | } else { |
1406 | if (unlikely(copied < len)) | |
188c299e | 1407 | ext4_journalled_zero_new_buffers(handle, inode, page, |
3b136499 | 1408 | from + copied, to); |
188c299e JK |
1409 | ret = ext4_walk_page_buffers(handle, inode, page_buffers(page), |
1410 | from, from + copied, &partial, | |
3b136499 | 1411 | write_end_fn); |
3fdcfb66 TM |
1412 | if (!partial) |
1413 | SetPageUptodate(page); | |
1414 | } | |
c93d8f88 EB |
1415 | if (!verity) |
1416 | size_changed = ext4_update_inode_size(inode, pos + copied); | |
19f5fb7a | 1417 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
2d859db3 | 1418 | EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; |
4631dbf6 | 1419 | unlock_page(page); |
09cbfeaf | 1420 | put_page(page); |
4631dbf6 | 1421 | |
c93d8f88 | 1422 | if (old_size < pos && !verity) |
0572639f XW |
1423 | pagecache_isize_extended(inode, old_size, pos); |
1424 | ||
6984aef5 | 1425 | if (size_changed) { |
617ba13b | 1426 | ret2 = ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
1427 | if (!ret) |
1428 | ret = ret2; | |
1429 | } | |
bfc1af65 | 1430 | |
c93d8f88 | 1431 | if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode)) |
f8514083 AK |
1432 | /* if we have allocated more blocks and copied |
1433 | * less. We will have blocks allocated outside | |
1434 | * inode->i_size. So truncate them | |
1435 | */ | |
1436 | ext4_orphan_add(handle, inode); | |
1437 | ||
617ba13b | 1438 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1439 | if (!ret) |
1440 | ret = ret2; | |
c93d8f88 | 1441 | if (pos + len > inode->i_size && !verity) { |
b9a4207d | 1442 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1443 | /* |
ffacfa7a | 1444 | * If truncate failed early the inode might still be |
f8514083 AK |
1445 | * on the orphan list; we need to make sure the inode |
1446 | * is removed from the orphan list in that case. | |
1447 | */ | |
1448 | if (inode->i_nlink) | |
1449 | ext4_orphan_del(NULL, inode); | |
1450 | } | |
bfc1af65 NP |
1451 | |
1452 | return ret ? ret : copied; | |
ac27a0ec | 1453 | } |
d2a17637 | 1454 | |
9d0be502 | 1455 | /* |
c27e43a1 | 1456 | * Reserve space for a single cluster |
9d0be502 | 1457 | */ |
c27e43a1 | 1458 | static int ext4_da_reserve_space(struct inode *inode) |
d2a17637 | 1459 | { |
60e58e0f | 1460 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
0637c6f4 | 1461 | struct ext4_inode_info *ei = EXT4_I(inode); |
5dd4056d | 1462 | int ret; |
03179fe9 TT |
1463 | |
1464 | /* | |
1465 | * We will charge metadata quota at writeout time; this saves | |
1466 | * us from metadata over-estimation, though we may go over by | |
1467 | * a small amount in the end. Here we just reserve for data. | |
1468 | */ | |
1469 | ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1)); | |
1470 | if (ret) | |
1471 | return ret; | |
d2a17637 | 1472 | |
0637c6f4 | 1473 | spin_lock(&ei->i_block_reservation_lock); |
71d4f7d0 | 1474 | if (ext4_claim_free_clusters(sbi, 1, 0)) { |
03179fe9 | 1475 | spin_unlock(&ei->i_block_reservation_lock); |
03179fe9 | 1476 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1)); |
d2a17637 MC |
1477 | return -ENOSPC; |
1478 | } | |
9d0be502 | 1479 | ei->i_reserved_data_blocks++; |
c27e43a1 | 1480 | trace_ext4_da_reserve_space(inode); |
0637c6f4 | 1481 | spin_unlock(&ei->i_block_reservation_lock); |
39bc680a | 1482 | |
d2a17637 MC |
1483 | return 0; /* success */ |
1484 | } | |
1485 | ||
f456767d | 1486 | void ext4_da_release_space(struct inode *inode, int to_free) |
d2a17637 MC |
1487 | { |
1488 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0637c6f4 | 1489 | struct ext4_inode_info *ei = EXT4_I(inode); |
d2a17637 | 1490 | |
cd213226 MC |
1491 | if (!to_free) |
1492 | return; /* Nothing to release, exit */ | |
1493 | ||
d2a17637 | 1494 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); |
cd213226 | 1495 | |
5a58ec87 | 1496 | trace_ext4_da_release_space(inode, to_free); |
0637c6f4 | 1497 | if (unlikely(to_free > ei->i_reserved_data_blocks)) { |
cd213226 | 1498 | /* |
0637c6f4 TT |
1499 | * if there aren't enough reserved blocks, then the |
1500 | * counter is messed up somewhere. Since this | |
1501 | * function is called from invalidate page, it's | |
1502 | * harmless to return without any action. | |
cd213226 | 1503 | */ |
8de5c325 | 1504 | ext4_warning(inode->i_sb, "ext4_da_release_space: " |
0637c6f4 | 1505 | "ino %lu, to_free %d with only %d reserved " |
1084f252 | 1506 | "data blocks", inode->i_ino, to_free, |
0637c6f4 TT |
1507 | ei->i_reserved_data_blocks); |
1508 | WARN_ON(1); | |
1509 | to_free = ei->i_reserved_data_blocks; | |
cd213226 | 1510 | } |
0637c6f4 | 1511 | ei->i_reserved_data_blocks -= to_free; |
cd213226 | 1512 | |
72b8ab9d | 1513 | /* update fs dirty data blocks counter */ |
57042651 | 1514 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free); |
d2a17637 | 1515 | |
d2a17637 | 1516 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
60e58e0f | 1517 | |
7b415bf6 | 1518 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free)); |
d2a17637 MC |
1519 | } |
1520 | ||
64769240 AT |
1521 | /* |
1522 | * Delayed allocation stuff | |
1523 | */ | |
1524 | ||
4e7ea81d JK |
1525 | struct mpage_da_data { |
1526 | struct inode *inode; | |
1527 | struct writeback_control *wbc; | |
6b523df4 | 1528 | |
4e7ea81d JK |
1529 | pgoff_t first_page; /* The first page to write */ |
1530 | pgoff_t next_page; /* Current page to examine */ | |
1531 | pgoff_t last_page; /* Last page to examine */ | |
791b7f08 | 1532 | /* |
4e7ea81d JK |
1533 | * Extent to map - this can be after first_page because that can be |
1534 | * fully mapped. We somewhat abuse m_flags to store whether the extent | |
1535 | * is delalloc or unwritten. | |
791b7f08 | 1536 | */ |
4e7ea81d JK |
1537 | struct ext4_map_blocks map; |
1538 | struct ext4_io_submit io_submit; /* IO submission data */ | |
dddbd6ac | 1539 | unsigned int do_map:1; |
6b8ed620 | 1540 | unsigned int scanned_until_end:1; |
4e7ea81d | 1541 | }; |
64769240 | 1542 | |
4e7ea81d JK |
1543 | static void mpage_release_unused_pages(struct mpage_da_data *mpd, |
1544 | bool invalidate) | |
c4a0c46e AK |
1545 | { |
1546 | int nr_pages, i; | |
1547 | pgoff_t index, end; | |
1548 | struct pagevec pvec; | |
1549 | struct inode *inode = mpd->inode; | |
1550 | struct address_space *mapping = inode->i_mapping; | |
4e7ea81d JK |
1551 | |
1552 | /* This is necessary when next_page == 0. */ | |
1553 | if (mpd->first_page >= mpd->next_page) | |
1554 | return; | |
c4a0c46e | 1555 | |
6b8ed620 | 1556 | mpd->scanned_until_end = 0; |
c7f5938a CW |
1557 | index = mpd->first_page; |
1558 | end = mpd->next_page - 1; | |
4e7ea81d JK |
1559 | if (invalidate) { |
1560 | ext4_lblk_t start, last; | |
09cbfeaf KS |
1561 | start = index << (PAGE_SHIFT - inode->i_blkbits); |
1562 | last = end << (PAGE_SHIFT - inode->i_blkbits); | |
4e7ea81d JK |
1563 | ext4_es_remove_extent(inode, start, last - start + 1); |
1564 | } | |
51865fda | 1565 | |
86679820 | 1566 | pagevec_init(&pvec); |
c4a0c46e | 1567 | while (index <= end) { |
397162ff | 1568 | nr_pages = pagevec_lookup_range(&pvec, mapping, &index, end); |
c4a0c46e AK |
1569 | if (nr_pages == 0) |
1570 | break; | |
1571 | for (i = 0; i < nr_pages; i++) { | |
1572 | struct page *page = pvec.pages[i]; | |
2b85a617 | 1573 | |
c4a0c46e AK |
1574 | BUG_ON(!PageLocked(page)); |
1575 | BUG_ON(PageWriteback(page)); | |
4e7ea81d | 1576 | if (invalidate) { |
4e800c03 | 1577 | if (page_mapped(page)) |
1578 | clear_page_dirty_for_io(page); | |
09cbfeaf | 1579 | block_invalidatepage(page, 0, PAGE_SIZE); |
4e7ea81d JK |
1580 | ClearPageUptodate(page); |
1581 | } | |
c4a0c46e AK |
1582 | unlock_page(page); |
1583 | } | |
9b1d0998 | 1584 | pagevec_release(&pvec); |
c4a0c46e | 1585 | } |
c4a0c46e AK |
1586 | } |
1587 | ||
df22291f AK |
1588 | static void ext4_print_free_blocks(struct inode *inode) |
1589 | { | |
1590 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
92b97816 | 1591 | struct super_block *sb = inode->i_sb; |
f78ee70d | 1592 | struct ext4_inode_info *ei = EXT4_I(inode); |
92b97816 TT |
1593 | |
1594 | ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld", | |
5dee5437 | 1595 | EXT4_C2B(EXT4_SB(inode->i_sb), |
f78ee70d | 1596 | ext4_count_free_clusters(sb))); |
92b97816 TT |
1597 | ext4_msg(sb, KERN_CRIT, "Free/Dirty block details"); |
1598 | ext4_msg(sb, KERN_CRIT, "free_blocks=%lld", | |
f78ee70d | 1599 | (long long) EXT4_C2B(EXT4_SB(sb), |
57042651 | 1600 | percpu_counter_sum(&sbi->s_freeclusters_counter))); |
92b97816 | 1601 | ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld", |
f78ee70d | 1602 | (long long) EXT4_C2B(EXT4_SB(sb), |
7b415bf6 | 1603 | percpu_counter_sum(&sbi->s_dirtyclusters_counter))); |
92b97816 TT |
1604 | ext4_msg(sb, KERN_CRIT, "Block reservation details"); |
1605 | ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u", | |
f78ee70d | 1606 | ei->i_reserved_data_blocks); |
df22291f AK |
1607 | return; |
1608 | } | |
1609 | ||
188c299e JK |
1610 | static int ext4_bh_delay_or_unwritten(handle_t *handle, struct inode *inode, |
1611 | struct buffer_head *bh) | |
29fa89d0 | 1612 | { |
c364b22c | 1613 | return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh); |
29fa89d0 AK |
1614 | } |
1615 | ||
0b02f4c0 EW |
1616 | /* |
1617 | * ext4_insert_delayed_block - adds a delayed block to the extents status | |
1618 | * tree, incrementing the reserved cluster/block | |
1619 | * count or making a pending reservation | |
1620 | * where needed | |
1621 | * | |
1622 | * @inode - file containing the newly added block | |
1623 | * @lblk - logical block to be added | |
1624 | * | |
1625 | * Returns 0 on success, negative error code on failure. | |
1626 | */ | |
1627 | static int ext4_insert_delayed_block(struct inode *inode, ext4_lblk_t lblk) | |
1628 | { | |
1629 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
1630 | int ret; | |
1631 | bool allocated = false; | |
6fed8395 | 1632 | bool reserved = false; |
0b02f4c0 EW |
1633 | |
1634 | /* | |
1635 | * If the cluster containing lblk is shared with a delayed, | |
1636 | * written, or unwritten extent in a bigalloc file system, it's | |
1637 | * already been accounted for and does not need to be reserved. | |
1638 | * A pending reservation must be made for the cluster if it's | |
1639 | * shared with a written or unwritten extent and doesn't already | |
1640 | * have one. Written and unwritten extents can be purged from the | |
1641 | * extents status tree if the system is under memory pressure, so | |
1642 | * it's necessary to examine the extent tree if a search of the | |
1643 | * extents status tree doesn't get a match. | |
1644 | */ | |
1645 | if (sbi->s_cluster_ratio == 1) { | |
1646 | ret = ext4_da_reserve_space(inode); | |
1647 | if (ret != 0) /* ENOSPC */ | |
1648 | goto errout; | |
6fed8395 | 1649 | reserved = true; |
0b02f4c0 EW |
1650 | } else { /* bigalloc */ |
1651 | if (!ext4_es_scan_clu(inode, &ext4_es_is_delonly, lblk)) { | |
1652 | if (!ext4_es_scan_clu(inode, | |
1653 | &ext4_es_is_mapped, lblk)) { | |
1654 | ret = ext4_clu_mapped(inode, | |
1655 | EXT4_B2C(sbi, lblk)); | |
1656 | if (ret < 0) | |
1657 | goto errout; | |
1658 | if (ret == 0) { | |
1659 | ret = ext4_da_reserve_space(inode); | |
1660 | if (ret != 0) /* ENOSPC */ | |
1661 | goto errout; | |
6fed8395 | 1662 | reserved = true; |
0b02f4c0 EW |
1663 | } else { |
1664 | allocated = true; | |
1665 | } | |
1666 | } else { | |
1667 | allocated = true; | |
1668 | } | |
1669 | } | |
1670 | } | |
1671 | ||
1672 | ret = ext4_es_insert_delayed_block(inode, lblk, allocated); | |
6fed8395 JX |
1673 | if (ret && reserved) |
1674 | ext4_da_release_space(inode, 1); | |
0b02f4c0 EW |
1675 | |
1676 | errout: | |
1677 | return ret; | |
1678 | } | |
1679 | ||
5356f261 AK |
1680 | /* |
1681 | * This function is grabs code from the very beginning of | |
1682 | * ext4_map_blocks, but assumes that the caller is from delayed write | |
1683 | * time. This function looks up the requested blocks and sets the | |
1684 | * buffer delay bit under the protection of i_data_sem. | |
1685 | */ | |
1686 | static int ext4_da_map_blocks(struct inode *inode, sector_t iblock, | |
1687 | struct ext4_map_blocks *map, | |
1688 | struct buffer_head *bh) | |
1689 | { | |
d100eef2 | 1690 | struct extent_status es; |
5356f261 AK |
1691 | int retval; |
1692 | sector_t invalid_block = ~((sector_t) 0xffff); | |
921f266b DM |
1693 | #ifdef ES_AGGRESSIVE_TEST |
1694 | struct ext4_map_blocks orig_map; | |
1695 | ||
1696 | memcpy(&orig_map, map, sizeof(*map)); | |
1697 | #endif | |
5356f261 AK |
1698 | |
1699 | if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es)) | |
1700 | invalid_block = ~0; | |
1701 | ||
1702 | map->m_flags = 0; | |
70aa1554 | 1703 | ext_debug(inode, "max_blocks %u, logical block %lu\n", map->m_len, |
5356f261 | 1704 | (unsigned long) map->m_lblk); |
d100eef2 ZL |
1705 | |
1706 | /* Lookup extent status tree firstly */ | |
bb5835ed | 1707 | if (ext4_es_lookup_extent(inode, iblock, NULL, &es)) { |
d100eef2 ZL |
1708 | if (ext4_es_is_hole(&es)) { |
1709 | retval = 0; | |
c8b459f4 | 1710 | down_read(&EXT4_I(inode)->i_data_sem); |
d100eef2 ZL |
1711 | goto add_delayed; |
1712 | } | |
1713 | ||
1714 | /* | |
713f1d3b EW |
1715 | * Delayed extent could be allocated by fallocate. |
1716 | * So we need to check it. | |
d100eef2 | 1717 | */ |
713f1d3b EW |
1718 | if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) { |
1719 | map_bh(bh, inode->i_sb, invalid_block); | |
1720 | set_buffer_new(bh); | |
1721 | set_buffer_delay(bh); | |
d100eef2 ZL |
1722 | return 0; |
1723 | } | |
1724 | ||
1725 | map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk; | |
1726 | retval = es.es_len - (iblock - es.es_lblk); | |
1727 | if (retval > map->m_len) | |
1728 | retval = map->m_len; | |
1729 | map->m_len = retval; | |
1730 | if (ext4_es_is_written(&es)) | |
1731 | map->m_flags |= EXT4_MAP_MAPPED; | |
1732 | else if (ext4_es_is_unwritten(&es)) | |
1733 | map->m_flags |= EXT4_MAP_UNWRITTEN; | |
1734 | else | |
1e83bc81 | 1735 | BUG(); |
d100eef2 | 1736 | |
921f266b DM |
1737 | #ifdef ES_AGGRESSIVE_TEST |
1738 | ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0); | |
1739 | #endif | |
d100eef2 ZL |
1740 | return retval; |
1741 | } | |
1742 | ||
5356f261 AK |
1743 | /* |
1744 | * Try to see if we can get the block without requesting a new | |
1745 | * file system block. | |
1746 | */ | |
c8b459f4 | 1747 | down_read(&EXT4_I(inode)->i_data_sem); |
cbd7584e | 1748 | if (ext4_has_inline_data(inode)) |
9c3569b5 | 1749 | retval = 0; |
cbd7584e | 1750 | else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
2f8e0a7c | 1751 | retval = ext4_ext_map_blocks(NULL, inode, map, 0); |
5356f261 | 1752 | else |
2f8e0a7c | 1753 | retval = ext4_ind_map_blocks(NULL, inode, map, 0); |
5356f261 | 1754 | |
d100eef2 | 1755 | add_delayed: |
5356f261 | 1756 | if (retval == 0) { |
f7fec032 | 1757 | int ret; |
ad431025 | 1758 | |
5356f261 AK |
1759 | /* |
1760 | * XXX: __block_prepare_write() unmaps passed block, | |
1761 | * is it OK? | |
1762 | */ | |
5356f261 | 1763 | |
0b02f4c0 EW |
1764 | ret = ext4_insert_delayed_block(inode, map->m_lblk); |
1765 | if (ret != 0) { | |
f7fec032 | 1766 | retval = ret; |
51865fda | 1767 | goto out_unlock; |
f7fec032 | 1768 | } |
51865fda | 1769 | |
5356f261 AK |
1770 | map_bh(bh, inode->i_sb, invalid_block); |
1771 | set_buffer_new(bh); | |
1772 | set_buffer_delay(bh); | |
f7fec032 ZL |
1773 | } else if (retval > 0) { |
1774 | int ret; | |
3be78c73 | 1775 | unsigned int status; |
f7fec032 | 1776 | |
44fb851d ZL |
1777 | if (unlikely(retval != map->m_len)) { |
1778 | ext4_warning(inode->i_sb, | |
1779 | "ES len assertion failed for inode " | |
1780 | "%lu: retval %d != map->m_len %d", | |
1781 | inode->i_ino, retval, map->m_len); | |
1782 | WARN_ON(1); | |
921f266b | 1783 | } |
921f266b | 1784 | |
f7fec032 ZL |
1785 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
1786 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; | |
1787 | ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, | |
1788 | map->m_pblk, status); | |
1789 | if (ret != 0) | |
1790 | retval = ret; | |
5356f261 AK |
1791 | } |
1792 | ||
1793 | out_unlock: | |
1794 | up_read((&EXT4_I(inode)->i_data_sem)); | |
1795 | ||
1796 | return retval; | |
1797 | } | |
1798 | ||
64769240 | 1799 | /* |
d91bd2c1 | 1800 | * This is a special get_block_t callback which is used by |
b920c755 TT |
1801 | * ext4_da_write_begin(). It will either return mapped block or |
1802 | * reserve space for a single block. | |
29fa89d0 AK |
1803 | * |
1804 | * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set. | |
1805 | * We also have b_blocknr = -1 and b_bdev initialized properly | |
1806 | * | |
1807 | * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set. | |
1808 | * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev | |
1809 | * initialized properly. | |
64769240 | 1810 | */ |
9c3569b5 TM |
1811 | int ext4_da_get_block_prep(struct inode *inode, sector_t iblock, |
1812 | struct buffer_head *bh, int create) | |
64769240 | 1813 | { |
2ed88685 | 1814 | struct ext4_map_blocks map; |
64769240 AT |
1815 | int ret = 0; |
1816 | ||
1817 | BUG_ON(create == 0); | |
2ed88685 TT |
1818 | BUG_ON(bh->b_size != inode->i_sb->s_blocksize); |
1819 | ||
1820 | map.m_lblk = iblock; | |
1821 | map.m_len = 1; | |
64769240 AT |
1822 | |
1823 | /* | |
1824 | * first, we need to know whether the block is allocated already | |
1825 | * preallocated blocks are unmapped but should treated | |
1826 | * the same as allocated blocks. | |
1827 | */ | |
5356f261 AK |
1828 | ret = ext4_da_map_blocks(inode, iblock, &map, bh); |
1829 | if (ret <= 0) | |
2ed88685 | 1830 | return ret; |
64769240 | 1831 | |
2ed88685 | 1832 | map_bh(bh, inode->i_sb, map.m_pblk); |
ed8ad838 | 1833 | ext4_update_bh_state(bh, map.m_flags); |
2ed88685 TT |
1834 | |
1835 | if (buffer_unwritten(bh)) { | |
1836 | /* A delayed write to unwritten bh should be marked | |
1837 | * new and mapped. Mapped ensures that we don't do | |
1838 | * get_block multiple times when we write to the same | |
1839 | * offset and new ensures that we do proper zero out | |
1840 | * for partial write. | |
1841 | */ | |
1842 | set_buffer_new(bh); | |
c8205636 | 1843 | set_buffer_mapped(bh); |
2ed88685 TT |
1844 | } |
1845 | return 0; | |
64769240 | 1846 | } |
61628a3f | 1847 | |
62e086be | 1848 | static int __ext4_journalled_writepage(struct page *page, |
62e086be AK |
1849 | unsigned int len) |
1850 | { | |
1851 | struct address_space *mapping = page->mapping; | |
1852 | struct inode *inode = mapping->host; | |
62e086be | 1853 | handle_t *handle = NULL; |
3fdcfb66 TM |
1854 | int ret = 0, err = 0; |
1855 | int inline_data = ext4_has_inline_data(inode); | |
1856 | struct buffer_head *inode_bh = NULL; | |
5ceaebae | 1857 | loff_t size; |
62e086be | 1858 | |
cb20d518 | 1859 | ClearPageChecked(page); |
3fdcfb66 TM |
1860 | |
1861 | if (inline_data) { | |
1862 | BUG_ON(page->index != 0); | |
1863 | BUG_ON(len > ext4_get_max_inline_size(inode)); | |
1864 | inode_bh = ext4_journalled_write_inline_data(inode, len, page); | |
1865 | if (inode_bh == NULL) | |
1866 | goto out; | |
3fdcfb66 | 1867 | } |
bdf96838 TT |
1868 | /* |
1869 | * We need to release the page lock before we start the | |
1870 | * journal, so grab a reference so the page won't disappear | |
1871 | * out from under us. | |
1872 | */ | |
1873 | get_page(page); | |
62e086be AK |
1874 | unlock_page(page); |
1875 | ||
9924a92a TT |
1876 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
1877 | ext4_writepage_trans_blocks(inode)); | |
62e086be AK |
1878 | if (IS_ERR(handle)) { |
1879 | ret = PTR_ERR(handle); | |
bdf96838 TT |
1880 | put_page(page); |
1881 | goto out_no_pagelock; | |
62e086be | 1882 | } |
441c8508 CW |
1883 | BUG_ON(!ext4_handle_valid(handle)); |
1884 | ||
bdf96838 TT |
1885 | lock_page(page); |
1886 | put_page(page); | |
5ceaebae ZY |
1887 | size = i_size_read(inode); |
1888 | if (page->mapping != mapping || page_offset(page) > size) { | |
bdf96838 TT |
1889 | /* The page got truncated from under us */ |
1890 | ext4_journal_stop(handle); | |
1891 | ret = 0; | |
1892 | goto out; | |
1893 | } | |
1894 | ||
3fdcfb66 | 1895 | if (inline_data) { |
362eca70 | 1896 | ret = ext4_mark_inode_dirty(handle, inode); |
3fdcfb66 | 1897 | } else { |
5ceaebae ZY |
1898 | struct buffer_head *page_bufs = page_buffers(page); |
1899 | ||
1900 | if (page->index == size >> PAGE_SHIFT) | |
1901 | len = size & ~PAGE_MASK; | |
1902 | else | |
1903 | len = PAGE_SIZE; | |
1904 | ||
188c299e JK |
1905 | ret = ext4_walk_page_buffers(handle, inode, page_bufs, 0, len, |
1906 | NULL, do_journal_get_write_access); | |
3fdcfb66 | 1907 | |
188c299e JK |
1908 | err = ext4_walk_page_buffers(handle, inode, page_bufs, 0, len, |
1909 | NULL, write_end_fn); | |
3fdcfb66 | 1910 | } |
afb585a9 MFO |
1911 | if (ret == 0) |
1912 | ret = err; | |
b5b18160 | 1913 | err = ext4_jbd2_inode_add_write(handle, inode, page_offset(page), len); |
62e086be AK |
1914 | if (ret == 0) |
1915 | ret = err; | |
2d859db3 | 1916 | EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; |
62e086be AK |
1917 | err = ext4_journal_stop(handle); |
1918 | if (!ret) | |
1919 | ret = err; | |
1920 | ||
19f5fb7a | 1921 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
62e086be | 1922 | out: |
bdf96838 TT |
1923 | unlock_page(page); |
1924 | out_no_pagelock: | |
3fdcfb66 | 1925 | brelse(inode_bh); |
62e086be AK |
1926 | return ret; |
1927 | } | |
1928 | ||
61628a3f | 1929 | /* |
43ce1d23 AK |
1930 | * Note that we don't need to start a transaction unless we're journaling data |
1931 | * because we should have holes filled from ext4_page_mkwrite(). We even don't | |
1932 | * need to file the inode to the transaction's list in ordered mode because if | |
1933 | * we are writing back data added by write(), the inode is already there and if | |
25985edc | 1934 | * we are writing back data modified via mmap(), no one guarantees in which |
43ce1d23 AK |
1935 | * transaction the data will hit the disk. In case we are journaling data, we |
1936 | * cannot start transaction directly because transaction start ranks above page | |
1937 | * lock so we have to do some magic. | |
1938 | * | |
b920c755 | 1939 | * This function can get called via... |
20970ba6 | 1940 | * - ext4_writepages after taking page lock (have journal handle) |
b920c755 | 1941 | * - journal_submit_inode_data_buffers (no journal handle) |
f6463b0d | 1942 | * - shrink_page_list via the kswapd/direct reclaim (no journal handle) |
b920c755 | 1943 | * - grab_page_cache when doing write_begin (have journal handle) |
43ce1d23 AK |
1944 | * |
1945 | * We don't do any block allocation in this function. If we have page with | |
1946 | * multiple blocks we need to write those buffer_heads that are mapped. This | |
1947 | * is important for mmaped based write. So if we do with blocksize 1K | |
1948 | * truncate(f, 1024); | |
1949 | * a = mmap(f, 0, 4096); | |
1950 | * a[0] = 'a'; | |
1951 | * truncate(f, 4096); | |
1952 | * we have in the page first buffer_head mapped via page_mkwrite call back | |
90802ed9 | 1953 | * but other buffer_heads would be unmapped but dirty (dirty done via the |
43ce1d23 AK |
1954 | * do_wp_page). So writepage should write the first block. If we modify |
1955 | * the mmap area beyond 1024 we will again get a page_fault and the | |
1956 | * page_mkwrite callback will do the block allocation and mark the | |
1957 | * buffer_heads mapped. | |
1958 | * | |
1959 | * We redirty the page if we have any buffer_heads that is either delay or | |
1960 | * unwritten in the page. | |
1961 | * | |
1962 | * We can get recursively called as show below. | |
1963 | * | |
1964 | * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() -> | |
1965 | * ext4_writepage() | |
1966 | * | |
1967 | * But since we don't do any block allocation we should not deadlock. | |
1968 | * Page also have the dirty flag cleared so we don't get recurive page_lock. | |
61628a3f | 1969 | */ |
43ce1d23 | 1970 | static int ext4_writepage(struct page *page, |
62e086be | 1971 | struct writeback_control *wbc) |
64769240 | 1972 | { |
f8bec370 | 1973 | int ret = 0; |
61628a3f | 1974 | loff_t size; |
498e5f24 | 1975 | unsigned int len; |
744692dc | 1976 | struct buffer_head *page_bufs = NULL; |
61628a3f | 1977 | struct inode *inode = page->mapping->host; |
36ade451 | 1978 | struct ext4_io_submit io_submit; |
1c8349a1 | 1979 | bool keep_towrite = false; |
61628a3f | 1980 | |
0db1ff22 | 1981 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) { |
c2a559bc | 1982 | inode->i_mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE); |
0db1ff22 TT |
1983 | unlock_page(page); |
1984 | return -EIO; | |
1985 | } | |
1986 | ||
a9c667f8 | 1987 | trace_ext4_writepage(page); |
f0e6c985 | 1988 | size = i_size_read(inode); |
c93d8f88 EB |
1989 | if (page->index == size >> PAGE_SHIFT && |
1990 | !ext4_verity_in_progress(inode)) | |
09cbfeaf | 1991 | len = size & ~PAGE_MASK; |
f0e6c985 | 1992 | else |
09cbfeaf | 1993 | len = PAGE_SIZE; |
64769240 | 1994 | |
be7cb076 TT |
1995 | /* Should never happen but for bugs in other kernel subsystems */ |
1996 | if (!page_has_buffers(page)) { | |
1997 | ext4_warning_inode(inode, | |
1998 | "page %lu does not have buffers attached", page->index); | |
1999 | ClearPageDirty(page); | |
2000 | unlock_page(page); | |
2001 | return 0; | |
2002 | } | |
2003 | ||
a42afc5f | 2004 | page_bufs = page_buffers(page); |
a42afc5f | 2005 | /* |
fe386132 JK |
2006 | * We cannot do block allocation or other extent handling in this |
2007 | * function. If there are buffers needing that, we have to redirty | |
2008 | * the page. But we may reach here when we do a journal commit via | |
2009 | * journal_submit_inode_data_buffers() and in that case we must write | |
2010 | * allocated buffers to achieve data=ordered mode guarantees. | |
cccd147a TT |
2011 | * |
2012 | * Also, if there is only one buffer per page (the fs block | |
2013 | * size == the page size), if one buffer needs block | |
2014 | * allocation or needs to modify the extent tree to clear the | |
2015 | * unwritten flag, we know that the page can't be written at | |
2016 | * all, so we might as well refuse the write immediately. | |
2017 | * Unfortunately if the block size != page size, we can't as | |
2018 | * easily detect this case using ext4_walk_page_buffers(), but | |
2019 | * for the extremely common case, this is an optimization that | |
2020 | * skips a useless round trip through ext4_bio_write_page(). | |
a42afc5f | 2021 | */ |
188c299e | 2022 | if (ext4_walk_page_buffers(NULL, inode, page_bufs, 0, len, NULL, |
f19d5870 | 2023 | ext4_bh_delay_or_unwritten)) { |
f8bec370 | 2024 | redirty_page_for_writepage(wbc, page); |
cccd147a | 2025 | if ((current->flags & PF_MEMALLOC) || |
09cbfeaf | 2026 | (inode->i_sb->s_blocksize == PAGE_SIZE)) { |
fe386132 JK |
2027 | /* |
2028 | * For memory cleaning there's no point in writing only | |
2029 | * some buffers. So just bail out. Warn if we came here | |
2030 | * from direct reclaim. | |
2031 | */ | |
2032 | WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) | |
2033 | == PF_MEMALLOC); | |
f0e6c985 AK |
2034 | unlock_page(page); |
2035 | return 0; | |
2036 | } | |
1c8349a1 | 2037 | keep_towrite = true; |
a42afc5f | 2038 | } |
64769240 | 2039 | |
cb20d518 | 2040 | if (PageChecked(page) && ext4_should_journal_data(inode)) |
43ce1d23 AK |
2041 | /* |
2042 | * It's mmapped pagecache. Add buffers and journal it. There | |
2043 | * doesn't seem much point in redirtying the page here. | |
2044 | */ | |
3f0ca309 | 2045 | return __ext4_journalled_writepage(page, len); |
43ce1d23 | 2046 | |
97a851ed JK |
2047 | ext4_io_submit_init(&io_submit, wbc); |
2048 | io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS); | |
2049 | if (!io_submit.io_end) { | |
2050 | redirty_page_for_writepage(wbc, page); | |
2051 | unlock_page(page); | |
2052 | return -ENOMEM; | |
2053 | } | |
be993933 | 2054 | ret = ext4_bio_write_page(&io_submit, page, len, keep_towrite); |
36ade451 | 2055 | ext4_io_submit(&io_submit); |
97a851ed JK |
2056 | /* Drop io_end reference we got from init */ |
2057 | ext4_put_io_end_defer(io_submit.io_end); | |
64769240 AT |
2058 | return ret; |
2059 | } | |
2060 | ||
5f1132b2 JK |
2061 | static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page) |
2062 | { | |
2063 | int len; | |
a056bdaa | 2064 | loff_t size; |
5f1132b2 JK |
2065 | int err; |
2066 | ||
2067 | BUG_ON(page->index != mpd->first_page); | |
a056bdaa JK |
2068 | clear_page_dirty_for_io(page); |
2069 | /* | |
2070 | * We have to be very careful here! Nothing protects writeback path | |
2071 | * against i_size changes and the page can be writeably mapped into | |
2072 | * page tables. So an application can be growing i_size and writing | |
2073 | * data through mmap while writeback runs. clear_page_dirty_for_io() | |
2074 | * write-protects our page in page tables and the page cannot get | |
2075 | * written to again until we release page lock. So only after | |
2076 | * clear_page_dirty_for_io() we are safe to sample i_size for | |
2077 | * ext4_bio_write_page() to zero-out tail of the written page. We rely | |
2078 | * on the barrier provided by TestClearPageDirty in | |
2079 | * clear_page_dirty_for_io() to make sure i_size is really sampled only | |
2080 | * after page tables are updated. | |
2081 | */ | |
2082 | size = i_size_read(mpd->inode); | |
c93d8f88 EB |
2083 | if (page->index == size >> PAGE_SHIFT && |
2084 | !ext4_verity_in_progress(mpd->inode)) | |
09cbfeaf | 2085 | len = size & ~PAGE_MASK; |
5f1132b2 | 2086 | else |
09cbfeaf | 2087 | len = PAGE_SIZE; |
be993933 | 2088 | err = ext4_bio_write_page(&mpd->io_submit, page, len, false); |
5f1132b2 JK |
2089 | if (!err) |
2090 | mpd->wbc->nr_to_write--; | |
2091 | mpd->first_page++; | |
2092 | ||
2093 | return err; | |
2094 | } | |
2095 | ||
6db07461 | 2096 | #define BH_FLAGS (BIT(BH_Unwritten) | BIT(BH_Delay)) |
4e7ea81d | 2097 | |
61628a3f | 2098 | /* |
fffb2739 JK |
2099 | * mballoc gives us at most this number of blocks... |
2100 | * XXX: That seems to be only a limitation of ext4_mb_normalize_request(). | |
70261f56 | 2101 | * The rest of mballoc seems to handle chunks up to full group size. |
61628a3f | 2102 | */ |
fffb2739 | 2103 | #define MAX_WRITEPAGES_EXTENT_LEN 2048 |
525f4ed8 | 2104 | |
4e7ea81d JK |
2105 | /* |
2106 | * mpage_add_bh_to_extent - try to add bh to extent of blocks to map | |
2107 | * | |
2108 | * @mpd - extent of blocks | |
2109 | * @lblk - logical number of the block in the file | |
09930042 | 2110 | * @bh - buffer head we want to add to the extent |
4e7ea81d | 2111 | * |
09930042 JK |
2112 | * The function is used to collect contig. blocks in the same state. If the |
2113 | * buffer doesn't require mapping for writeback and we haven't started the | |
2114 | * extent of buffers to map yet, the function returns 'true' immediately - the | |
2115 | * caller can write the buffer right away. Otherwise the function returns true | |
2116 | * if the block has been added to the extent, false if the block couldn't be | |
2117 | * added. | |
4e7ea81d | 2118 | */ |
09930042 JK |
2119 | static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk, |
2120 | struct buffer_head *bh) | |
4e7ea81d JK |
2121 | { |
2122 | struct ext4_map_blocks *map = &mpd->map; | |
2123 | ||
09930042 JK |
2124 | /* Buffer that doesn't need mapping for writeback? */ |
2125 | if (!buffer_dirty(bh) || !buffer_mapped(bh) || | |
2126 | (!buffer_delay(bh) && !buffer_unwritten(bh))) { | |
2127 | /* So far no extent to map => we write the buffer right away */ | |
2128 | if (map->m_len == 0) | |
2129 | return true; | |
2130 | return false; | |
2131 | } | |
4e7ea81d JK |
2132 | |
2133 | /* First block in the extent? */ | |
2134 | if (map->m_len == 0) { | |
dddbd6ac JK |
2135 | /* We cannot map unless handle is started... */ |
2136 | if (!mpd->do_map) | |
2137 | return false; | |
4e7ea81d JK |
2138 | map->m_lblk = lblk; |
2139 | map->m_len = 1; | |
09930042 JK |
2140 | map->m_flags = bh->b_state & BH_FLAGS; |
2141 | return true; | |
4e7ea81d JK |
2142 | } |
2143 | ||
09930042 JK |
2144 | /* Don't go larger than mballoc is willing to allocate */ |
2145 | if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN) | |
2146 | return false; | |
2147 | ||
4e7ea81d JK |
2148 | /* Can we merge the block to our big extent? */ |
2149 | if (lblk == map->m_lblk + map->m_len && | |
09930042 | 2150 | (bh->b_state & BH_FLAGS) == map->m_flags) { |
4e7ea81d | 2151 | map->m_len++; |
09930042 | 2152 | return true; |
4e7ea81d | 2153 | } |
09930042 | 2154 | return false; |
4e7ea81d JK |
2155 | } |
2156 | ||
5f1132b2 JK |
2157 | /* |
2158 | * mpage_process_page_bufs - submit page buffers for IO or add them to extent | |
2159 | * | |
2160 | * @mpd - extent of blocks for mapping | |
2161 | * @head - the first buffer in the page | |
2162 | * @bh - buffer we should start processing from | |
2163 | * @lblk - logical number of the block in the file corresponding to @bh | |
2164 | * | |
2165 | * Walk through page buffers from @bh upto @head (exclusive) and either submit | |
2166 | * the page for IO if all buffers in this page were mapped and there's no | |
2167 | * accumulated extent of buffers to map or add buffers in the page to the | |
2168 | * extent of buffers to map. The function returns 1 if the caller can continue | |
2169 | * by processing the next page, 0 if it should stop adding buffers to the | |
2170 | * extent to map because we cannot extend it anymore. It can also return value | |
2171 | * < 0 in case of error during IO submission. | |
2172 | */ | |
2173 | static int mpage_process_page_bufs(struct mpage_da_data *mpd, | |
2174 | struct buffer_head *head, | |
2175 | struct buffer_head *bh, | |
2176 | ext4_lblk_t lblk) | |
4e7ea81d JK |
2177 | { |
2178 | struct inode *inode = mpd->inode; | |
5f1132b2 | 2179 | int err; |
93407472 | 2180 | ext4_lblk_t blocks = (i_size_read(inode) + i_blocksize(inode) - 1) |
4e7ea81d JK |
2181 | >> inode->i_blkbits; |
2182 | ||
c93d8f88 EB |
2183 | if (ext4_verity_in_progress(inode)) |
2184 | blocks = EXT_MAX_BLOCKS; | |
2185 | ||
4e7ea81d JK |
2186 | do { |
2187 | BUG_ON(buffer_locked(bh)); | |
2188 | ||
09930042 | 2189 | if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) { |
4e7ea81d JK |
2190 | /* Found extent to map? */ |
2191 | if (mpd->map.m_len) | |
5f1132b2 | 2192 | return 0; |
dddbd6ac JK |
2193 | /* Buffer needs mapping and handle is not started? */ |
2194 | if (!mpd->do_map) | |
2195 | return 0; | |
09930042 | 2196 | /* Everything mapped so far and we hit EOF */ |
5f1132b2 | 2197 | break; |
4e7ea81d | 2198 | } |
4e7ea81d | 2199 | } while (lblk++, (bh = bh->b_this_page) != head); |
5f1132b2 JK |
2200 | /* So far everything mapped? Submit the page for IO. */ |
2201 | if (mpd->map.m_len == 0) { | |
2202 | err = mpage_submit_page(mpd, head->b_page); | |
2203 | if (err < 0) | |
2204 | return err; | |
2205 | } | |
6b8ed620 JK |
2206 | if (lblk >= blocks) { |
2207 | mpd->scanned_until_end = 1; | |
2208 | return 0; | |
2209 | } | |
2210 | return 1; | |
4e7ea81d JK |
2211 | } |
2212 | ||
2943fdbc RH |
2213 | /* |
2214 | * mpage_process_page - update page buffers corresponding to changed extent and | |
2215 | * may submit fully mapped page for IO | |
2216 | * | |
2217 | * @mpd - description of extent to map, on return next extent to map | |
2218 | * @m_lblk - logical block mapping. | |
2219 | * @m_pblk - corresponding physical mapping. | |
2220 | * @map_bh - determines on return whether this page requires any further | |
2221 | * mapping or not. | |
2222 | * Scan given page buffers corresponding to changed extent and update buffer | |
2223 | * state according to new extent state. | |
2224 | * We map delalloc buffers to their physical location, clear unwritten bits. | |
2225 | * If the given page is not fully mapped, we update @map to the next extent in | |
2226 | * the given page that needs mapping & return @map_bh as true. | |
2227 | */ | |
2228 | static int mpage_process_page(struct mpage_da_data *mpd, struct page *page, | |
2229 | ext4_lblk_t *m_lblk, ext4_fsblk_t *m_pblk, | |
2230 | bool *map_bh) | |
2231 | { | |
2232 | struct buffer_head *head, *bh; | |
2233 | ext4_io_end_t *io_end = mpd->io_submit.io_end; | |
2234 | ext4_lblk_t lblk = *m_lblk; | |
2235 | ext4_fsblk_t pblock = *m_pblk; | |
2236 | int err = 0; | |
c8cc8816 RH |
2237 | int blkbits = mpd->inode->i_blkbits; |
2238 | ssize_t io_end_size = 0; | |
2239 | struct ext4_io_end_vec *io_end_vec = ext4_last_io_end_vec(io_end); | |
2943fdbc RH |
2240 | |
2241 | bh = head = page_buffers(page); | |
2242 | do { | |
2243 | if (lblk < mpd->map.m_lblk) | |
2244 | continue; | |
2245 | if (lblk >= mpd->map.m_lblk + mpd->map.m_len) { | |
2246 | /* | |
2247 | * Buffer after end of mapped extent. | |
2248 | * Find next buffer in the page to map. | |
2249 | */ | |
2250 | mpd->map.m_len = 0; | |
2251 | mpd->map.m_flags = 0; | |
c8cc8816 RH |
2252 | io_end_vec->size += io_end_size; |
2253 | io_end_size = 0; | |
2943fdbc | 2254 | |
2943fdbc RH |
2255 | err = mpage_process_page_bufs(mpd, head, bh, lblk); |
2256 | if (err > 0) | |
2257 | err = 0; | |
c8cc8816 RH |
2258 | if (!err && mpd->map.m_len && mpd->map.m_lblk > lblk) { |
2259 | io_end_vec = ext4_alloc_io_end_vec(io_end); | |
4d06bfb9 RH |
2260 | if (IS_ERR(io_end_vec)) { |
2261 | err = PTR_ERR(io_end_vec); | |
2262 | goto out; | |
2263 | } | |
d1e18b88 | 2264 | io_end_vec->offset = (loff_t)mpd->map.m_lblk << blkbits; |
c8cc8816 | 2265 | } |
2943fdbc RH |
2266 | *map_bh = true; |
2267 | goto out; | |
2268 | } | |
2269 | if (buffer_delay(bh)) { | |
2270 | clear_buffer_delay(bh); | |
2271 | bh->b_blocknr = pblock++; | |
2272 | } | |
2273 | clear_buffer_unwritten(bh); | |
c8cc8816 | 2274 | io_end_size += (1 << blkbits); |
2943fdbc | 2275 | } while (lblk++, (bh = bh->b_this_page) != head); |
c8cc8816 RH |
2276 | |
2277 | io_end_vec->size += io_end_size; | |
2278 | io_end_size = 0; | |
2943fdbc RH |
2279 | *map_bh = false; |
2280 | out: | |
2281 | *m_lblk = lblk; | |
2282 | *m_pblk = pblock; | |
2283 | return err; | |
2284 | } | |
2285 | ||
4e7ea81d JK |
2286 | /* |
2287 | * mpage_map_buffers - update buffers corresponding to changed extent and | |
2288 | * submit fully mapped pages for IO | |
2289 | * | |
2290 | * @mpd - description of extent to map, on return next extent to map | |
2291 | * | |
2292 | * Scan buffers corresponding to changed extent (we expect corresponding pages | |
2293 | * to be already locked) and update buffer state according to new extent state. | |
2294 | * We map delalloc buffers to their physical location, clear unwritten bits, | |
556615dc | 2295 | * and mark buffers as uninit when we perform writes to unwritten extents |
4e7ea81d JK |
2296 | * and do extent conversion after IO is finished. If the last page is not fully |
2297 | * mapped, we update @map to the next extent in the last page that needs | |
2298 | * mapping. Otherwise we submit the page for IO. | |
2299 | */ | |
2300 | static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd) | |
2301 | { | |
2302 | struct pagevec pvec; | |
2303 | int nr_pages, i; | |
2304 | struct inode *inode = mpd->inode; | |
09cbfeaf | 2305 | int bpp_bits = PAGE_SHIFT - inode->i_blkbits; |
4e7ea81d JK |
2306 | pgoff_t start, end; |
2307 | ext4_lblk_t lblk; | |
2943fdbc | 2308 | ext4_fsblk_t pblock; |
4e7ea81d | 2309 | int err; |
2943fdbc | 2310 | bool map_bh = false; |
4e7ea81d JK |
2311 | |
2312 | start = mpd->map.m_lblk >> bpp_bits; | |
2313 | end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits; | |
2314 | lblk = start << bpp_bits; | |
2315 | pblock = mpd->map.m_pblk; | |
2316 | ||
86679820 | 2317 | pagevec_init(&pvec); |
4e7ea81d | 2318 | while (start <= end) { |
2b85a617 | 2319 | nr_pages = pagevec_lookup_range(&pvec, inode->i_mapping, |
397162ff | 2320 | &start, end); |
4e7ea81d JK |
2321 | if (nr_pages == 0) |
2322 | break; | |
2323 | for (i = 0; i < nr_pages; i++) { | |
2324 | struct page *page = pvec.pages[i]; | |
2325 | ||
2943fdbc RH |
2326 | err = mpage_process_page(mpd, page, &lblk, &pblock, |
2327 | &map_bh); | |
4e7ea81d | 2328 | /* |
2943fdbc RH |
2329 | * If map_bh is true, means page may require further bh |
2330 | * mapping, or maybe the page was submitted for IO. | |
2331 | * So we return to call further extent mapping. | |
4e7ea81d | 2332 | */ |
39c0ae16 | 2333 | if (err < 0 || map_bh) |
2943fdbc | 2334 | goto out; |
4e7ea81d JK |
2335 | /* Page fully mapped - let IO run! */ |
2336 | err = mpage_submit_page(mpd, page); | |
2943fdbc RH |
2337 | if (err < 0) |
2338 | goto out; | |
4e7ea81d JK |
2339 | } |
2340 | pagevec_release(&pvec); | |
2341 | } | |
2342 | /* Extent fully mapped and matches with page boundary. We are done. */ | |
2343 | mpd->map.m_len = 0; | |
2344 | mpd->map.m_flags = 0; | |
2345 | return 0; | |
2943fdbc RH |
2346 | out: |
2347 | pagevec_release(&pvec); | |
2348 | return err; | |
4e7ea81d JK |
2349 | } |
2350 | ||
2351 | static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd) | |
2352 | { | |
2353 | struct inode *inode = mpd->inode; | |
2354 | struct ext4_map_blocks *map = &mpd->map; | |
2355 | int get_blocks_flags; | |
090f32ee | 2356 | int err, dioread_nolock; |
4e7ea81d JK |
2357 | |
2358 | trace_ext4_da_write_pages_extent(inode, map); | |
2359 | /* | |
2360 | * Call ext4_map_blocks() to allocate any delayed allocation blocks, or | |
556615dc | 2361 | * to convert an unwritten extent to be initialized (in the case |
4e7ea81d JK |
2362 | * where we have written into one or more preallocated blocks). It is |
2363 | * possible that we're going to need more metadata blocks than | |
2364 | * previously reserved. However we must not fail because we're in | |
2365 | * writeback and there is nothing we can do about it so it might result | |
2366 | * in data loss. So use reserved blocks to allocate metadata if | |
2367 | * possible. | |
2368 | * | |
754cfed6 TT |
2369 | * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if |
2370 | * the blocks in question are delalloc blocks. This indicates | |
2371 | * that the blocks and quotas has already been checked when | |
2372 | * the data was copied into the page cache. | |
4e7ea81d JK |
2373 | */ |
2374 | get_blocks_flags = EXT4_GET_BLOCKS_CREATE | | |
ee0876bc JK |
2375 | EXT4_GET_BLOCKS_METADATA_NOFAIL | |
2376 | EXT4_GET_BLOCKS_IO_SUBMIT; | |
090f32ee LC |
2377 | dioread_nolock = ext4_should_dioread_nolock(inode); |
2378 | if (dioread_nolock) | |
4e7ea81d | 2379 | get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT; |
6db07461 | 2380 | if (map->m_flags & BIT(BH_Delay)) |
4e7ea81d JK |
2381 | get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE; |
2382 | ||
2383 | err = ext4_map_blocks(handle, inode, map, get_blocks_flags); | |
2384 | if (err < 0) | |
2385 | return err; | |
090f32ee | 2386 | if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) { |
6b523df4 JK |
2387 | if (!mpd->io_submit.io_end->handle && |
2388 | ext4_handle_valid(handle)) { | |
2389 | mpd->io_submit.io_end->handle = handle->h_rsv_handle; | |
2390 | handle->h_rsv_handle = NULL; | |
2391 | } | |
3613d228 | 2392 | ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end); |
6b523df4 | 2393 | } |
4e7ea81d JK |
2394 | |
2395 | BUG_ON(map->m_len == 0); | |
4e7ea81d JK |
2396 | return 0; |
2397 | } | |
2398 | ||
2399 | /* | |
2400 | * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length | |
2401 | * mpd->len and submit pages underlying it for IO | |
2402 | * | |
2403 | * @handle - handle for journal operations | |
2404 | * @mpd - extent to map | |
7534e854 JK |
2405 | * @give_up_on_write - we set this to true iff there is a fatal error and there |
2406 | * is no hope of writing the data. The caller should discard | |
2407 | * dirty pages to avoid infinite loops. | |
4e7ea81d JK |
2408 | * |
2409 | * The function maps extent starting at mpd->lblk of length mpd->len. If it is | |
2410 | * delayed, blocks are allocated, if it is unwritten, we may need to convert | |
2411 | * them to initialized or split the described range from larger unwritten | |
2412 | * extent. Note that we need not map all the described range since allocation | |
2413 | * can return less blocks or the range is covered by more unwritten extents. We | |
2414 | * cannot map more because we are limited by reserved transaction credits. On | |
2415 | * the other hand we always make sure that the last touched page is fully | |
2416 | * mapped so that it can be written out (and thus forward progress is | |
2417 | * guaranteed). After mapping we submit all mapped pages for IO. | |
2418 | */ | |
2419 | static int mpage_map_and_submit_extent(handle_t *handle, | |
cb530541 TT |
2420 | struct mpage_da_data *mpd, |
2421 | bool *give_up_on_write) | |
4e7ea81d JK |
2422 | { |
2423 | struct inode *inode = mpd->inode; | |
2424 | struct ext4_map_blocks *map = &mpd->map; | |
2425 | int err; | |
2426 | loff_t disksize; | |
6603120e | 2427 | int progress = 0; |
c8cc8816 | 2428 | ext4_io_end_t *io_end = mpd->io_submit.io_end; |
4d06bfb9 | 2429 | struct ext4_io_end_vec *io_end_vec; |
4e7ea81d | 2430 | |
4d06bfb9 RH |
2431 | io_end_vec = ext4_alloc_io_end_vec(io_end); |
2432 | if (IS_ERR(io_end_vec)) | |
2433 | return PTR_ERR(io_end_vec); | |
c8cc8816 | 2434 | io_end_vec->offset = ((loff_t)map->m_lblk) << inode->i_blkbits; |
27d7c4ed | 2435 | do { |
4e7ea81d JK |
2436 | err = mpage_map_one_extent(handle, mpd); |
2437 | if (err < 0) { | |
2438 | struct super_block *sb = inode->i_sb; | |
2439 | ||
0db1ff22 | 2440 | if (ext4_forced_shutdown(EXT4_SB(sb)) || |
9b5f6c9b | 2441 | ext4_test_mount_flag(sb, EXT4_MF_FS_ABORTED)) |
cb530541 | 2442 | goto invalidate_dirty_pages; |
4e7ea81d | 2443 | /* |
cb530541 TT |
2444 | * Let the uper layers retry transient errors. |
2445 | * In the case of ENOSPC, if ext4_count_free_blocks() | |
2446 | * is non-zero, a commit should free up blocks. | |
4e7ea81d | 2447 | */ |
cb530541 | 2448 | if ((err == -ENOMEM) || |
6603120e DM |
2449 | (err == -ENOSPC && ext4_count_free_clusters(sb))) { |
2450 | if (progress) | |
2451 | goto update_disksize; | |
cb530541 | 2452 | return err; |
6603120e | 2453 | } |
cb530541 TT |
2454 | ext4_msg(sb, KERN_CRIT, |
2455 | "Delayed block allocation failed for " | |
2456 | "inode %lu at logical offset %llu with" | |
2457 | " max blocks %u with error %d", | |
2458 | inode->i_ino, | |
2459 | (unsigned long long)map->m_lblk, | |
2460 | (unsigned)map->m_len, -err); | |
2461 | ext4_msg(sb, KERN_CRIT, | |
2462 | "This should not happen!! Data will " | |
2463 | "be lost\n"); | |
2464 | if (err == -ENOSPC) | |
2465 | ext4_print_free_blocks(inode); | |
2466 | invalidate_dirty_pages: | |
2467 | *give_up_on_write = true; | |
4e7ea81d JK |
2468 | return err; |
2469 | } | |
6603120e | 2470 | progress = 1; |
4e7ea81d JK |
2471 | /* |
2472 | * Update buffer state, submit mapped pages, and get us new | |
2473 | * extent to map | |
2474 | */ | |
2475 | err = mpage_map_and_submit_buffers(mpd); | |
2476 | if (err < 0) | |
6603120e | 2477 | goto update_disksize; |
27d7c4ed | 2478 | } while (map->m_len); |
4e7ea81d | 2479 | |
6603120e | 2480 | update_disksize: |
622cad13 TT |
2481 | /* |
2482 | * Update on-disk size after IO is submitted. Races with | |
2483 | * truncate are avoided by checking i_size under i_data_sem. | |
2484 | */ | |
09cbfeaf | 2485 | disksize = ((loff_t)mpd->first_page) << PAGE_SHIFT; |
35df4299 | 2486 | if (disksize > READ_ONCE(EXT4_I(inode)->i_disksize)) { |
4e7ea81d | 2487 | int err2; |
622cad13 TT |
2488 | loff_t i_size; |
2489 | ||
2490 | down_write(&EXT4_I(inode)->i_data_sem); | |
2491 | i_size = i_size_read(inode); | |
2492 | if (disksize > i_size) | |
2493 | disksize = i_size; | |
2494 | if (disksize > EXT4_I(inode)->i_disksize) | |
2495 | EXT4_I(inode)->i_disksize = disksize; | |
622cad13 | 2496 | up_write(&EXT4_I(inode)->i_data_sem); |
b907f2d5 | 2497 | err2 = ext4_mark_inode_dirty(handle, inode); |
878520ac | 2498 | if (err2) { |
54d3adbc TT |
2499 | ext4_error_err(inode->i_sb, -err2, |
2500 | "Failed to mark inode %lu dirty", | |
2501 | inode->i_ino); | |
878520ac | 2502 | } |
4e7ea81d JK |
2503 | if (!err) |
2504 | err = err2; | |
2505 | } | |
2506 | return err; | |
2507 | } | |
2508 | ||
fffb2739 JK |
2509 | /* |
2510 | * Calculate the total number of credits to reserve for one writepages | |
20970ba6 | 2511 | * iteration. This is called from ext4_writepages(). We map an extent of |
70261f56 | 2512 | * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping |
fffb2739 JK |
2513 | * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN + |
2514 | * bpp - 1 blocks in bpp different extents. | |
2515 | */ | |
525f4ed8 MC |
2516 | static int ext4_da_writepages_trans_blocks(struct inode *inode) |
2517 | { | |
fffb2739 | 2518 | int bpp = ext4_journal_blocks_per_page(inode); |
525f4ed8 | 2519 | |
fffb2739 JK |
2520 | return ext4_meta_trans_blocks(inode, |
2521 | MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp); | |
525f4ed8 | 2522 | } |
61628a3f | 2523 | |
8e48dcfb | 2524 | /* |
4e7ea81d JK |
2525 | * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages |
2526 | * and underlying extent to map | |
2527 | * | |
2528 | * @mpd - where to look for pages | |
2529 | * | |
2530 | * Walk dirty pages in the mapping. If they are fully mapped, submit them for | |
2531 | * IO immediately. When we find a page which isn't mapped we start accumulating | |
2532 | * extent of buffers underlying these pages that needs mapping (formed by | |
2533 | * either delayed or unwritten buffers). We also lock the pages containing | |
2534 | * these buffers. The extent found is returned in @mpd structure (starting at | |
2535 | * mpd->lblk with length mpd->len blocks). | |
2536 | * | |
2537 | * Note that this function can attach bios to one io_end structure which are | |
2538 | * neither logically nor physically contiguous. Although it may seem as an | |
2539 | * unnecessary complication, it is actually inevitable in blocksize < pagesize | |
2540 | * case as we need to track IO to all buffers underlying a page in one io_end. | |
8e48dcfb | 2541 | */ |
4e7ea81d | 2542 | static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd) |
8e48dcfb | 2543 | { |
4e7ea81d JK |
2544 | struct address_space *mapping = mpd->inode->i_mapping; |
2545 | struct pagevec pvec; | |
2546 | unsigned int nr_pages; | |
aeac589a | 2547 | long left = mpd->wbc->nr_to_write; |
4e7ea81d JK |
2548 | pgoff_t index = mpd->first_page; |
2549 | pgoff_t end = mpd->last_page; | |
10bbd235 | 2550 | xa_mark_t tag; |
4e7ea81d JK |
2551 | int i, err = 0; |
2552 | int blkbits = mpd->inode->i_blkbits; | |
2553 | ext4_lblk_t lblk; | |
2554 | struct buffer_head *head; | |
8e48dcfb | 2555 | |
4e7ea81d | 2556 | if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages) |
5b41d924 ES |
2557 | tag = PAGECACHE_TAG_TOWRITE; |
2558 | else | |
2559 | tag = PAGECACHE_TAG_DIRTY; | |
2560 | ||
86679820 | 2561 | pagevec_init(&pvec); |
4e7ea81d JK |
2562 | mpd->map.m_len = 0; |
2563 | mpd->next_page = index; | |
4f01b02c | 2564 | while (index <= end) { |
dc7f3e86 | 2565 | nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end, |
67fd707f | 2566 | tag); |
8e48dcfb | 2567 | if (nr_pages == 0) |
6b8ed620 | 2568 | break; |
8e48dcfb TT |
2569 | |
2570 | for (i = 0; i < nr_pages; i++) { | |
2571 | struct page *page = pvec.pages[i]; | |
2572 | ||
aeac589a ML |
2573 | /* |
2574 | * Accumulated enough dirty pages? This doesn't apply | |
2575 | * to WB_SYNC_ALL mode. For integrity sync we have to | |
2576 | * keep going because someone may be concurrently | |
2577 | * dirtying pages, and we might have synced a lot of | |
2578 | * newly appeared dirty pages, but have not synced all | |
2579 | * of the old dirty pages. | |
2580 | */ | |
2581 | if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0) | |
2582 | goto out; | |
2583 | ||
4e7ea81d JK |
2584 | /* If we can't merge this page, we are done. */ |
2585 | if (mpd->map.m_len > 0 && mpd->next_page != page->index) | |
2586 | goto out; | |
78aaced3 | 2587 | |
8e48dcfb | 2588 | lock_page(page); |
8e48dcfb | 2589 | /* |
4e7ea81d JK |
2590 | * If the page is no longer dirty, or its mapping no |
2591 | * longer corresponds to inode we are writing (which | |
2592 | * means it has been truncated or invalidated), or the | |
2593 | * page is already under writeback and we are not doing | |
2594 | * a data integrity writeback, skip the page | |
8e48dcfb | 2595 | */ |
4f01b02c TT |
2596 | if (!PageDirty(page) || |
2597 | (PageWriteback(page) && | |
4e7ea81d | 2598 | (mpd->wbc->sync_mode == WB_SYNC_NONE)) || |
4f01b02c | 2599 | unlikely(page->mapping != mapping)) { |
8e48dcfb TT |
2600 | unlock_page(page); |
2601 | continue; | |
2602 | } | |
2603 | ||
7cb1a535 | 2604 | wait_on_page_writeback(page); |
8e48dcfb | 2605 | BUG_ON(PageWriteback(page)); |
8e48dcfb | 2606 | |
be7cb076 TT |
2607 | /* |
2608 | * Should never happen but for buggy code in | |
2609 | * other subsystems that call | |
2610 | * set_page_dirty() without properly warning | |
2611 | * the file system first. See [1] for more | |
2612 | * information. | |
2613 | * | |
2614 | * [1] https://lore.kernel.org/linux-mm/20180103100430.GE4911@quack2.suse.cz | |
2615 | */ | |
2616 | if (!page_has_buffers(page)) { | |
2617 | ext4_warning_inode(mpd->inode, "page %lu does not have buffers attached", page->index); | |
2618 | ClearPageDirty(page); | |
2619 | unlock_page(page); | |
2620 | continue; | |
2621 | } | |
2622 | ||
4e7ea81d | 2623 | if (mpd->map.m_len == 0) |
8eb9e5ce | 2624 | mpd->first_page = page->index; |
8eb9e5ce | 2625 | mpd->next_page = page->index + 1; |
f8bec370 | 2626 | /* Add all dirty buffers to mpd */ |
4e7ea81d | 2627 | lblk = ((ext4_lblk_t)page->index) << |
09cbfeaf | 2628 | (PAGE_SHIFT - blkbits); |
f8bec370 | 2629 | head = page_buffers(page); |
5f1132b2 JK |
2630 | err = mpage_process_page_bufs(mpd, head, head, lblk); |
2631 | if (err <= 0) | |
4e7ea81d | 2632 | goto out; |
5f1132b2 | 2633 | err = 0; |
aeac589a | 2634 | left--; |
8e48dcfb TT |
2635 | } |
2636 | pagevec_release(&pvec); | |
2637 | cond_resched(); | |
2638 | } | |
6b8ed620 | 2639 | mpd->scanned_until_end = 1; |
4f01b02c | 2640 | return 0; |
8eb9e5ce TT |
2641 | out: |
2642 | pagevec_release(&pvec); | |
4e7ea81d | 2643 | return err; |
8e48dcfb TT |
2644 | } |
2645 | ||
20970ba6 TT |
2646 | static int ext4_writepages(struct address_space *mapping, |
2647 | struct writeback_control *wbc) | |
64769240 | 2648 | { |
4e7ea81d JK |
2649 | pgoff_t writeback_index = 0; |
2650 | long nr_to_write = wbc->nr_to_write; | |
22208ded | 2651 | int range_whole = 0; |
4e7ea81d | 2652 | int cycled = 1; |
61628a3f | 2653 | handle_t *handle = NULL; |
df22291f | 2654 | struct mpage_da_data mpd; |
5e745b04 | 2655 | struct inode *inode = mapping->host; |
6b523df4 | 2656 | int needed_blocks, rsv_blocks = 0, ret = 0; |
5e745b04 | 2657 | struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb); |
1bce63d1 | 2658 | struct blk_plug plug; |
cb530541 | 2659 | bool give_up_on_write = false; |
61628a3f | 2660 | |
0db1ff22 TT |
2661 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
2662 | return -EIO; | |
2663 | ||
bbd55937 | 2664 | percpu_down_read(&sbi->s_writepages_rwsem); |
20970ba6 | 2665 | trace_ext4_writepages(inode, wbc); |
ba80b101 | 2666 | |
61628a3f MC |
2667 | /* |
2668 | * No pages to write? This is mainly a kludge to avoid starting | |
2669 | * a transaction for special inodes like journal inode on last iput() | |
2670 | * because that could violate lock ordering on umount | |
2671 | */ | |
a1d6cc56 | 2672 | if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) |
bbf023c7 | 2673 | goto out_writepages; |
2a21e37e | 2674 | |
20970ba6 | 2675 | if (ext4_should_journal_data(inode)) { |
043d20d1 | 2676 | ret = generic_writepages(mapping, wbc); |
bbf023c7 | 2677 | goto out_writepages; |
20970ba6 TT |
2678 | } |
2679 | ||
2a21e37e TT |
2680 | /* |
2681 | * If the filesystem has aborted, it is read-only, so return | |
2682 | * right away instead of dumping stack traces later on that | |
2683 | * will obscure the real source of the problem. We test | |
1751e8a6 | 2684 | * EXT4_MF_FS_ABORTED instead of sb->s_flag's SB_RDONLY because |
2a21e37e | 2685 | * the latter could be true if the filesystem is mounted |
20970ba6 | 2686 | * read-only, and in that case, ext4_writepages should |
2a21e37e TT |
2687 | * *never* be called, so if that ever happens, we would want |
2688 | * the stack trace. | |
2689 | */ | |
0db1ff22 | 2690 | if (unlikely(ext4_forced_shutdown(EXT4_SB(mapping->host->i_sb)) || |
9b5f6c9b | 2691 | ext4_test_mount_flag(inode->i_sb, EXT4_MF_FS_ABORTED))) { |
bbf023c7 ML |
2692 | ret = -EROFS; |
2693 | goto out_writepages; | |
2694 | } | |
2a21e37e | 2695 | |
4e7ea81d JK |
2696 | /* |
2697 | * If we have inline data and arrive here, it means that | |
2698 | * we will soon create the block for the 1st page, so | |
2699 | * we'd better clear the inline data here. | |
2700 | */ | |
2701 | if (ext4_has_inline_data(inode)) { | |
2702 | /* Just inode will be modified... */ | |
2703 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 1); | |
2704 | if (IS_ERR(handle)) { | |
2705 | ret = PTR_ERR(handle); | |
2706 | goto out_writepages; | |
2707 | } | |
2708 | BUG_ON(ext4_test_inode_state(inode, | |
2709 | EXT4_STATE_MAY_INLINE_DATA)); | |
2710 | ext4_destroy_inline_data(handle, inode); | |
2711 | ext4_journal_stop(handle); | |
2712 | } | |
2713 | ||
4e343231 | 2714 | if (ext4_should_dioread_nolock(inode)) { |
2715 | /* | |
2716 | * We may need to convert up to one extent per block in | |
2717 | * the page and we may dirty the inode. | |
2718 | */ | |
2719 | rsv_blocks = 1 + ext4_chunk_trans_blocks(inode, | |
2720 | PAGE_SIZE >> inode->i_blkbits); | |
2721 | } | |
2722 | ||
22208ded AK |
2723 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
2724 | range_whole = 1; | |
61628a3f | 2725 | |
2acf2c26 | 2726 | if (wbc->range_cyclic) { |
4e7ea81d JK |
2727 | writeback_index = mapping->writeback_index; |
2728 | if (writeback_index) | |
2acf2c26 | 2729 | cycled = 0; |
4e7ea81d JK |
2730 | mpd.first_page = writeback_index; |
2731 | mpd.last_page = -1; | |
5b41d924 | 2732 | } else { |
09cbfeaf KS |
2733 | mpd.first_page = wbc->range_start >> PAGE_SHIFT; |
2734 | mpd.last_page = wbc->range_end >> PAGE_SHIFT; | |
5b41d924 | 2735 | } |
a1d6cc56 | 2736 | |
4e7ea81d JK |
2737 | mpd.inode = inode; |
2738 | mpd.wbc = wbc; | |
2739 | ext4_io_submit_init(&mpd.io_submit, wbc); | |
2acf2c26 | 2740 | retry: |
6e6938b6 | 2741 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
4e7ea81d | 2742 | tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page); |
1bce63d1 | 2743 | blk_start_plug(&plug); |
dddbd6ac JK |
2744 | |
2745 | /* | |
2746 | * First writeback pages that don't need mapping - we can avoid | |
2747 | * starting a transaction unnecessarily and also avoid being blocked | |
2748 | * in the block layer on device congestion while having transaction | |
2749 | * started. | |
2750 | */ | |
2751 | mpd.do_map = 0; | |
6b8ed620 | 2752 | mpd.scanned_until_end = 0; |
dddbd6ac JK |
2753 | mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL); |
2754 | if (!mpd.io_submit.io_end) { | |
2755 | ret = -ENOMEM; | |
2756 | goto unplug; | |
2757 | } | |
2758 | ret = mpage_prepare_extent_to_map(&mpd); | |
a297b2fc XW |
2759 | /* Unlock pages we didn't use */ |
2760 | mpage_release_unused_pages(&mpd, false); | |
dddbd6ac JK |
2761 | /* Submit prepared bio */ |
2762 | ext4_io_submit(&mpd.io_submit); | |
2763 | ext4_put_io_end_defer(mpd.io_submit.io_end); | |
2764 | mpd.io_submit.io_end = NULL; | |
dddbd6ac JK |
2765 | if (ret < 0) |
2766 | goto unplug; | |
2767 | ||
6b8ed620 | 2768 | while (!mpd.scanned_until_end && wbc->nr_to_write > 0) { |
4e7ea81d JK |
2769 | /* For each extent of pages we use new io_end */ |
2770 | mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL); | |
2771 | if (!mpd.io_submit.io_end) { | |
2772 | ret = -ENOMEM; | |
2773 | break; | |
2774 | } | |
a1d6cc56 AK |
2775 | |
2776 | /* | |
4e7ea81d JK |
2777 | * We have two constraints: We find one extent to map and we |
2778 | * must always write out whole page (makes a difference when | |
2779 | * blocksize < pagesize) so that we don't block on IO when we | |
2780 | * try to write out the rest of the page. Journalled mode is | |
2781 | * not supported by delalloc. | |
a1d6cc56 AK |
2782 | */ |
2783 | BUG_ON(ext4_should_journal_data(inode)); | |
525f4ed8 | 2784 | needed_blocks = ext4_da_writepages_trans_blocks(inode); |
a1d6cc56 | 2785 | |
4e7ea81d | 2786 | /* start a new transaction */ |
6b523df4 JK |
2787 | handle = ext4_journal_start_with_reserve(inode, |
2788 | EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks); | |
61628a3f MC |
2789 | if (IS_ERR(handle)) { |
2790 | ret = PTR_ERR(handle); | |
1693918e | 2791 | ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: " |
fbe845dd | 2792 | "%ld pages, ino %lu; err %d", __func__, |
a1d6cc56 | 2793 | wbc->nr_to_write, inode->i_ino, ret); |
4e7ea81d JK |
2794 | /* Release allocated io_end */ |
2795 | ext4_put_io_end(mpd.io_submit.io_end); | |
dddbd6ac | 2796 | mpd.io_submit.io_end = NULL; |
4e7ea81d | 2797 | break; |
61628a3f | 2798 | } |
dddbd6ac | 2799 | mpd.do_map = 1; |
f63e6005 | 2800 | |
4e7ea81d JK |
2801 | trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc); |
2802 | ret = mpage_prepare_extent_to_map(&mpd); | |
6b8ed620 JK |
2803 | if (!ret && mpd.map.m_len) |
2804 | ret = mpage_map_and_submit_extent(handle, &mpd, | |
cb530541 | 2805 | &give_up_on_write); |
646caa9c JK |
2806 | /* |
2807 | * Caution: If the handle is synchronous, | |
2808 | * ext4_journal_stop() can wait for transaction commit | |
2809 | * to finish which may depend on writeback of pages to | |
2810 | * complete or on page lock to be released. In that | |
b483bb77 | 2811 | * case, we have to wait until after we have |
646caa9c JK |
2812 | * submitted all the IO, released page locks we hold, |
2813 | * and dropped io_end reference (for extent conversion | |
2814 | * to be able to complete) before stopping the handle. | |
2815 | */ | |
2816 | if (!ext4_handle_valid(handle) || handle->h_sync == 0) { | |
2817 | ext4_journal_stop(handle); | |
2818 | handle = NULL; | |
dddbd6ac | 2819 | mpd.do_map = 0; |
646caa9c | 2820 | } |
4e7ea81d | 2821 | /* Unlock pages we didn't use */ |
cb530541 | 2822 | mpage_release_unused_pages(&mpd, give_up_on_write); |
a297b2fc XW |
2823 | /* Submit prepared bio */ |
2824 | ext4_io_submit(&mpd.io_submit); | |
2825 | ||
646caa9c JK |
2826 | /* |
2827 | * Drop our io_end reference we got from init. We have | |
2828 | * to be careful and use deferred io_end finishing if | |
2829 | * we are still holding the transaction as we can | |
2830 | * release the last reference to io_end which may end | |
2831 | * up doing unwritten extent conversion. | |
2832 | */ | |
2833 | if (handle) { | |
2834 | ext4_put_io_end_defer(mpd.io_submit.io_end); | |
2835 | ext4_journal_stop(handle); | |
2836 | } else | |
2837 | ext4_put_io_end(mpd.io_submit.io_end); | |
dddbd6ac | 2838 | mpd.io_submit.io_end = NULL; |
4e7ea81d JK |
2839 | |
2840 | if (ret == -ENOSPC && sbi->s_journal) { | |
2841 | /* | |
2842 | * Commit the transaction which would | |
22208ded AK |
2843 | * free blocks released in the transaction |
2844 | * and try again | |
2845 | */ | |
df22291f | 2846 | jbd2_journal_force_commit_nested(sbi->s_journal); |
22208ded | 2847 | ret = 0; |
4e7ea81d JK |
2848 | continue; |
2849 | } | |
2850 | /* Fatal error - ENOMEM, EIO... */ | |
2851 | if (ret) | |
61628a3f | 2852 | break; |
a1d6cc56 | 2853 | } |
dddbd6ac | 2854 | unplug: |
1bce63d1 | 2855 | blk_finish_plug(&plug); |
9c12a831 | 2856 | if (!ret && !cycled && wbc->nr_to_write > 0) { |
2acf2c26 | 2857 | cycled = 1; |
4e7ea81d JK |
2858 | mpd.last_page = writeback_index - 1; |
2859 | mpd.first_page = 0; | |
2acf2c26 AK |
2860 | goto retry; |
2861 | } | |
22208ded AK |
2862 | |
2863 | /* Update index */ | |
22208ded AK |
2864 | if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) |
2865 | /* | |
4e7ea81d | 2866 | * Set the writeback_index so that range_cyclic |
22208ded AK |
2867 | * mode will write it back later |
2868 | */ | |
4e7ea81d | 2869 | mapping->writeback_index = mpd.first_page; |
a1d6cc56 | 2870 | |
61628a3f | 2871 | out_writepages: |
20970ba6 TT |
2872 | trace_ext4_writepages_result(inode, wbc, ret, |
2873 | nr_to_write - wbc->nr_to_write); | |
bbd55937 | 2874 | percpu_up_read(&sbi->s_writepages_rwsem); |
61628a3f | 2875 | return ret; |
64769240 AT |
2876 | } |
2877 | ||
5f0663bb DW |
2878 | static int ext4_dax_writepages(struct address_space *mapping, |
2879 | struct writeback_control *wbc) | |
2880 | { | |
2881 | int ret; | |
2882 | long nr_to_write = wbc->nr_to_write; | |
2883 | struct inode *inode = mapping->host; | |
2884 | struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb); | |
2885 | ||
2886 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) | |
2887 | return -EIO; | |
2888 | ||
bbd55937 | 2889 | percpu_down_read(&sbi->s_writepages_rwsem); |
5f0663bb DW |
2890 | trace_ext4_writepages(inode, wbc); |
2891 | ||
3f666c56 | 2892 | ret = dax_writeback_mapping_range(mapping, sbi->s_daxdev, wbc); |
5f0663bb DW |
2893 | trace_ext4_writepages_result(inode, wbc, ret, |
2894 | nr_to_write - wbc->nr_to_write); | |
bbd55937 | 2895 | percpu_up_read(&sbi->s_writepages_rwsem); |
5f0663bb DW |
2896 | return ret; |
2897 | } | |
2898 | ||
79f0be8d AK |
2899 | static int ext4_nonda_switch(struct super_block *sb) |
2900 | { | |
5c1ff336 | 2901 | s64 free_clusters, dirty_clusters; |
79f0be8d AK |
2902 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
2903 | ||
2904 | /* | |
2905 | * switch to non delalloc mode if we are running low | |
2906 | * on free block. The free block accounting via percpu | |
179f7ebf | 2907 | * counters can get slightly wrong with percpu_counter_batch getting |
79f0be8d AK |
2908 | * accumulated on each CPU without updating global counters |
2909 | * Delalloc need an accurate free block accounting. So switch | |
2910 | * to non delalloc when we are near to error range. | |
2911 | */ | |
5c1ff336 EW |
2912 | free_clusters = |
2913 | percpu_counter_read_positive(&sbi->s_freeclusters_counter); | |
2914 | dirty_clusters = | |
2915 | percpu_counter_read_positive(&sbi->s_dirtyclusters_counter); | |
00d4e736 TT |
2916 | /* |
2917 | * Start pushing delalloc when 1/2 of free blocks are dirty. | |
2918 | */ | |
5c1ff336 | 2919 | if (dirty_clusters && (free_clusters < 2 * dirty_clusters)) |
10ee27a0 | 2920 | try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE); |
00d4e736 | 2921 | |
5c1ff336 EW |
2922 | if (2 * free_clusters < 3 * dirty_clusters || |
2923 | free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) { | |
79f0be8d | 2924 | /* |
c8afb446 ES |
2925 | * free block count is less than 150% of dirty blocks |
2926 | * or free blocks is less than watermark | |
79f0be8d AK |
2927 | */ |
2928 | return 1; | |
2929 | } | |
2930 | return 0; | |
2931 | } | |
2932 | ||
64769240 | 2933 | static int ext4_da_write_begin(struct file *file, struct address_space *mapping, |
de9a55b8 TT |
2934 | loff_t pos, unsigned len, unsigned flags, |
2935 | struct page **pagep, void **fsdata) | |
64769240 | 2936 | { |
72b8ab9d | 2937 | int ret, retries = 0; |
64769240 AT |
2938 | struct page *page; |
2939 | pgoff_t index; | |
64769240 | 2940 | struct inode *inode = mapping->host; |
64769240 | 2941 | |
0db1ff22 TT |
2942 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
2943 | return -EIO; | |
2944 | ||
09cbfeaf | 2945 | index = pos >> PAGE_SHIFT; |
79f0be8d | 2946 | |
c93d8f88 EB |
2947 | if (ext4_nonda_switch(inode->i_sb) || S_ISLNK(inode->i_mode) || |
2948 | ext4_verity_in_progress(inode)) { | |
79f0be8d AK |
2949 | *fsdata = (void *)FALL_BACK_TO_NONDELALLOC; |
2950 | return ext4_write_begin(file, mapping, pos, | |
2951 | len, flags, pagep, fsdata); | |
2952 | } | |
2953 | *fsdata = (void *)0; | |
9bffad1e | 2954 | trace_ext4_da_write_begin(inode, pos, len, flags); |
9c3569b5 TM |
2955 | |
2956 | if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { | |
2957 | ret = ext4_da_write_inline_data_begin(mapping, inode, | |
2958 | pos, len, flags, | |
2959 | pagep, fsdata); | |
2960 | if (ret < 0) | |
47564bfb TT |
2961 | return ret; |
2962 | if (ret == 1) | |
2963 | return 0; | |
9c3569b5 TM |
2964 | } |
2965 | ||
cc883236 | 2966 | retry: |
47564bfb TT |
2967 | page = grab_cache_page_write_begin(mapping, index, flags); |
2968 | if (!page) | |
2969 | return -ENOMEM; | |
47564bfb | 2970 | |
47564bfb | 2971 | /* In case writeback began while the page was unlocked */ |
7afe5aa5 | 2972 | wait_for_stable_page(page); |
64769240 | 2973 | |
643fa961 | 2974 | #ifdef CONFIG_FS_ENCRYPTION |
2058f83a MH |
2975 | ret = ext4_block_write_begin(page, pos, len, |
2976 | ext4_da_get_block_prep); | |
2977 | #else | |
6e1db88d | 2978 | ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep); |
2058f83a | 2979 | #endif |
64769240 AT |
2980 | if (ret < 0) { |
2981 | unlock_page(page); | |
cc883236 | 2982 | put_page(page); |
ae4d5372 AK |
2983 | /* |
2984 | * block_write_begin may have instantiated a few blocks | |
2985 | * outside i_size. Trim these off again. Don't need | |
cc883236 | 2986 | * i_size_read because we hold inode lock. |
ae4d5372 AK |
2987 | */ |
2988 | if (pos + len > inode->i_size) | |
b9a4207d | 2989 | ext4_truncate_failed_write(inode); |
47564bfb TT |
2990 | |
2991 | if (ret == -ENOSPC && | |
2992 | ext4_should_retry_alloc(inode->i_sb, &retries)) | |
cc883236 | 2993 | goto retry; |
47564bfb | 2994 | return ret; |
64769240 AT |
2995 | } |
2996 | ||
47564bfb | 2997 | *pagep = page; |
64769240 AT |
2998 | return ret; |
2999 | } | |
3000 | ||
632eaeab MC |
3001 | /* |
3002 | * Check if we should update i_disksize | |
3003 | * when write to the end of file but not require block allocation | |
3004 | */ | |
3005 | static int ext4_da_should_update_i_disksize(struct page *page, | |
de9a55b8 | 3006 | unsigned long offset) |
632eaeab MC |
3007 | { |
3008 | struct buffer_head *bh; | |
3009 | struct inode *inode = page->mapping->host; | |
3010 | unsigned int idx; | |
3011 | int i; | |
3012 | ||
3013 | bh = page_buffers(page); | |
3014 | idx = offset >> inode->i_blkbits; | |
3015 | ||
af5bc92d | 3016 | for (i = 0; i < idx; i++) |
632eaeab MC |
3017 | bh = bh->b_this_page; |
3018 | ||
29fa89d0 | 3019 | if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh)) |
632eaeab MC |
3020 | return 0; |
3021 | return 1; | |
3022 | } | |
3023 | ||
64769240 | 3024 | static int ext4_da_write_end(struct file *file, |
de9a55b8 TT |
3025 | struct address_space *mapping, |
3026 | loff_t pos, unsigned len, unsigned copied, | |
3027 | struct page *page, void *fsdata) | |
64769240 AT |
3028 | { |
3029 | struct inode *inode = mapping->host; | |
64769240 | 3030 | loff_t new_i_size; |
632eaeab | 3031 | unsigned long start, end; |
79f0be8d AK |
3032 | int write_mode = (int)(unsigned long)fsdata; |
3033 | ||
74d553aa TT |
3034 | if (write_mode == FALL_BACK_TO_NONDELALLOC) |
3035 | return ext4_write_end(file, mapping, pos, | |
3036 | len, copied, page, fsdata); | |
632eaeab | 3037 | |
9bffad1e | 3038 | trace_ext4_da_write_end(inode, pos, len, copied); |
6984aef5 ZY |
3039 | |
3040 | if (write_mode != CONVERT_INLINE_DATA && | |
3041 | ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) && | |
3042 | ext4_has_inline_data(inode)) | |
3043 | return ext4_write_inline_data_end(inode, pos, len, copied, page); | |
3044 | ||
09cbfeaf | 3045 | start = pos & (PAGE_SIZE - 1); |
af5bc92d | 3046 | end = start + copied - 1; |
64769240 AT |
3047 | |
3048 | /* | |
4df031ff ZY |
3049 | * Since we are holding inode lock, we are sure i_disksize <= |
3050 | * i_size. We also know that if i_disksize < i_size, there are | |
3051 | * delalloc writes pending in the range upto i_size. If the end of | |
3052 | * the current write is <= i_size, there's no need to touch | |
3053 | * i_disksize since writeback will push i_disksize upto i_size | |
3054 | * eventually. If the end of the current write is > i_size and | |
3055 | * inside an allocated block (ext4_da_should_update_i_disksize() | |
3056 | * check), we need to update i_disksize here as neither | |
3057 | * ext4_writepage() nor certain ext4_writepages() paths not | |
3058 | * allocating blocks update i_disksize. | |
3059 | * | |
3060 | * Note that we defer inode dirtying to generic_write_end() / | |
3061 | * ext4_da_write_inline_data_end(). | |
64769240 | 3062 | */ |
64769240 | 3063 | new_i_size = pos + copied; |
6984aef5 ZY |
3064 | if (copied && new_i_size > inode->i_size && |
3065 | ext4_da_should_update_i_disksize(page, end)) | |
3066 | ext4_update_i_disksize(inode, new_i_size); | |
9c3569b5 | 3067 | |
cc883236 | 3068 | return generic_write_end(file, mapping, pos, len, copied, page, fsdata); |
64769240 AT |
3069 | } |
3070 | ||
ccd2506b TT |
3071 | /* |
3072 | * Force all delayed allocation blocks to be allocated for a given inode. | |
3073 | */ | |
3074 | int ext4_alloc_da_blocks(struct inode *inode) | |
3075 | { | |
fb40ba0d TT |
3076 | trace_ext4_alloc_da_blocks(inode); |
3077 | ||
71d4f7d0 | 3078 | if (!EXT4_I(inode)->i_reserved_data_blocks) |
ccd2506b TT |
3079 | return 0; |
3080 | ||
3081 | /* | |
3082 | * We do something simple for now. The filemap_flush() will | |
3083 | * also start triggering a write of the data blocks, which is | |
3084 | * not strictly speaking necessary (and for users of | |
3085 | * laptop_mode, not even desirable). However, to do otherwise | |
3086 | * would require replicating code paths in: | |
de9a55b8 | 3087 | * |
20970ba6 | 3088 | * ext4_writepages() -> |
ccd2506b TT |
3089 | * write_cache_pages() ---> (via passed in callback function) |
3090 | * __mpage_da_writepage() --> | |
3091 | * mpage_add_bh_to_extent() | |
3092 | * mpage_da_map_blocks() | |
3093 | * | |
3094 | * The problem is that write_cache_pages(), located in | |
3095 | * mm/page-writeback.c, marks pages clean in preparation for | |
3096 | * doing I/O, which is not desirable if we're not planning on | |
3097 | * doing I/O at all. | |
3098 | * | |
3099 | * We could call write_cache_pages(), and then redirty all of | |
380cf090 | 3100 | * the pages by calling redirty_page_for_writepage() but that |
ccd2506b TT |
3101 | * would be ugly in the extreme. So instead we would need to |
3102 | * replicate parts of the code in the above functions, | |
25985edc | 3103 | * simplifying them because we wouldn't actually intend to |
ccd2506b TT |
3104 | * write out the pages, but rather only collect contiguous |
3105 | * logical block extents, call the multi-block allocator, and | |
3106 | * then update the buffer heads with the block allocations. | |
de9a55b8 | 3107 | * |
ccd2506b TT |
3108 | * For now, though, we'll cheat by calling filemap_flush(), |
3109 | * which will map the blocks, and start the I/O, but not | |
3110 | * actually wait for the I/O to complete. | |
3111 | */ | |
3112 | return filemap_flush(inode->i_mapping); | |
3113 | } | |
64769240 | 3114 | |
ac27a0ec DK |
3115 | /* |
3116 | * bmap() is special. It gets used by applications such as lilo and by | |
3117 | * the swapper to find the on-disk block of a specific piece of data. | |
3118 | * | |
3119 | * Naturally, this is dangerous if the block concerned is still in the | |
617ba13b | 3120 | * journal. If somebody makes a swapfile on an ext4 data-journaling |
ac27a0ec DK |
3121 | * filesystem and enables swap, then they may get a nasty shock when the |
3122 | * data getting swapped to that swapfile suddenly gets overwritten by | |
3123 | * the original zero's written out previously to the journal and | |
3124 | * awaiting writeback in the kernel's buffer cache. | |
3125 | * | |
3126 | * So, if we see any bmap calls here on a modified, data-journaled file, | |
3127 | * take extra steps to flush any blocks which might be in the cache. | |
3128 | */ | |
617ba13b | 3129 | static sector_t ext4_bmap(struct address_space *mapping, sector_t block) |
ac27a0ec DK |
3130 | { |
3131 | struct inode *inode = mapping->host; | |
3132 | journal_t *journal; | |
3133 | int err; | |
3134 | ||
46c7f254 TM |
3135 | /* |
3136 | * We can get here for an inline file via the FIBMAP ioctl | |
3137 | */ | |
3138 | if (ext4_has_inline_data(inode)) | |
3139 | return 0; | |
3140 | ||
64769240 AT |
3141 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) && |
3142 | test_opt(inode->i_sb, DELALLOC)) { | |
3143 | /* | |
3144 | * With delalloc we want to sync the file | |
3145 | * so that we can make sure we allocate | |
3146 | * blocks for file | |
3147 | */ | |
3148 | filemap_write_and_wait(mapping); | |
3149 | } | |
3150 | ||
19f5fb7a TT |
3151 | if (EXT4_JOURNAL(inode) && |
3152 | ext4_test_inode_state(inode, EXT4_STATE_JDATA)) { | |
ac27a0ec DK |
3153 | /* |
3154 | * This is a REALLY heavyweight approach, but the use of | |
3155 | * bmap on dirty files is expected to be extremely rare: | |
3156 | * only if we run lilo or swapon on a freshly made file | |
3157 | * do we expect this to happen. | |
3158 | * | |
3159 | * (bmap requires CAP_SYS_RAWIO so this does not | |
3160 | * represent an unprivileged user DOS attack --- we'd be | |
3161 | * in trouble if mortal users could trigger this path at | |
3162 | * will.) | |
3163 | * | |
617ba13b | 3164 | * NB. EXT4_STATE_JDATA is not set on files other than |
ac27a0ec DK |
3165 | * regular files. If somebody wants to bmap a directory |
3166 | * or symlink and gets confused because the buffer | |
3167 | * hasn't yet been flushed to disk, they deserve | |
3168 | * everything they get. | |
3169 | */ | |
3170 | ||
19f5fb7a | 3171 | ext4_clear_inode_state(inode, EXT4_STATE_JDATA); |
617ba13b | 3172 | journal = EXT4_JOURNAL(inode); |
dab291af | 3173 | jbd2_journal_lock_updates(journal); |
01d5d965 | 3174 | err = jbd2_journal_flush(journal, 0); |
dab291af | 3175 | jbd2_journal_unlock_updates(journal); |
ac27a0ec DK |
3176 | |
3177 | if (err) | |
3178 | return 0; | |
3179 | } | |
3180 | ||
ac58e4fb | 3181 | return iomap_bmap(mapping, block, &ext4_iomap_ops); |
ac27a0ec DK |
3182 | } |
3183 | ||
617ba13b | 3184 | static int ext4_readpage(struct file *file, struct page *page) |
ac27a0ec | 3185 | { |
46c7f254 TM |
3186 | int ret = -EAGAIN; |
3187 | struct inode *inode = page->mapping->host; | |
3188 | ||
0562e0ba | 3189 | trace_ext4_readpage(page); |
46c7f254 TM |
3190 | |
3191 | if (ext4_has_inline_data(inode)) | |
3192 | ret = ext4_readpage_inline(inode, page); | |
3193 | ||
3194 | if (ret == -EAGAIN) | |
a07f624b | 3195 | return ext4_mpage_readpages(inode, NULL, page); |
46c7f254 TM |
3196 | |
3197 | return ret; | |
ac27a0ec DK |
3198 | } |
3199 | ||
6311f91f | 3200 | static void ext4_readahead(struct readahead_control *rac) |
ac27a0ec | 3201 | { |
6311f91f | 3202 | struct inode *inode = rac->mapping->host; |
46c7f254 | 3203 | |
6311f91f | 3204 | /* If the file has inline data, no need to do readahead. */ |
46c7f254 | 3205 | if (ext4_has_inline_data(inode)) |
6311f91f | 3206 | return; |
46c7f254 | 3207 | |
a07f624b | 3208 | ext4_mpage_readpages(inode, rac, NULL); |
ac27a0ec DK |
3209 | } |
3210 | ||
d47992f8 LC |
3211 | static void ext4_invalidatepage(struct page *page, unsigned int offset, |
3212 | unsigned int length) | |
ac27a0ec | 3213 | { |
ca99fdd2 | 3214 | trace_ext4_invalidatepage(page, offset, length); |
0562e0ba | 3215 | |
4520fb3c JK |
3216 | /* No journalling happens on data buffers when this function is used */ |
3217 | WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page))); | |
3218 | ||
ca99fdd2 | 3219 | block_invalidatepage(page, offset, length); |
4520fb3c JK |
3220 | } |
3221 | ||
53e87268 | 3222 | static int __ext4_journalled_invalidatepage(struct page *page, |
ca99fdd2 LC |
3223 | unsigned int offset, |
3224 | unsigned int length) | |
4520fb3c JK |
3225 | { |
3226 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); | |
3227 | ||
ca99fdd2 | 3228 | trace_ext4_journalled_invalidatepage(page, offset, length); |
4520fb3c | 3229 | |
ac27a0ec DK |
3230 | /* |
3231 | * If it's a full truncate we just forget about the pending dirtying | |
3232 | */ | |
09cbfeaf | 3233 | if (offset == 0 && length == PAGE_SIZE) |
ac27a0ec DK |
3234 | ClearPageChecked(page); |
3235 | ||
ca99fdd2 | 3236 | return jbd2_journal_invalidatepage(journal, page, offset, length); |
53e87268 JK |
3237 | } |
3238 | ||
3239 | /* Wrapper for aops... */ | |
3240 | static void ext4_journalled_invalidatepage(struct page *page, | |
d47992f8 LC |
3241 | unsigned int offset, |
3242 | unsigned int length) | |
53e87268 | 3243 | { |
ca99fdd2 | 3244 | WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0); |
ac27a0ec DK |
3245 | } |
3246 | ||
617ba13b | 3247 | static int ext4_releasepage(struct page *page, gfp_t wait) |
ac27a0ec | 3248 | { |
617ba13b | 3249 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); |
ac27a0ec | 3250 | |
0562e0ba JZ |
3251 | trace_ext4_releasepage(page); |
3252 | ||
e1c36595 JK |
3253 | /* Page has dirty journalled data -> cannot release */ |
3254 | if (PageChecked(page)) | |
ac27a0ec | 3255 | return 0; |
0390131b | 3256 | if (journal) |
529a781e | 3257 | return jbd2_journal_try_to_free_buffers(journal, page); |
0390131b FM |
3258 | else |
3259 | return try_to_free_buffers(page); | |
ac27a0ec DK |
3260 | } |
3261 | ||
b8a6176c JK |
3262 | static bool ext4_inode_datasync_dirty(struct inode *inode) |
3263 | { | |
3264 | journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; | |
3265 | ||
aa75f4d3 HS |
3266 | if (journal) { |
3267 | if (jbd2_transaction_committed(journal, | |
d0520df7 AR |
3268 | EXT4_I(inode)->i_datasync_tid)) |
3269 | return false; | |
3270 | if (test_opt2(inode->i_sb, JOURNAL_FAST_COMMIT)) | |
1ceecb53 | 3271 | return !list_empty(&EXT4_I(inode)->i_fc_list); |
d0520df7 | 3272 | return true; |
aa75f4d3 HS |
3273 | } |
3274 | ||
b8a6176c JK |
3275 | /* Any metadata buffers to write? */ |
3276 | if (!list_empty(&inode->i_mapping->private_list)) | |
3277 | return true; | |
3278 | return inode->i_state & I_DIRTY_DATASYNC; | |
3279 | } | |
3280 | ||
c8fdfe29 MB |
3281 | static void ext4_set_iomap(struct inode *inode, struct iomap *iomap, |
3282 | struct ext4_map_blocks *map, loff_t offset, | |
3283 | loff_t length) | |
364443cb | 3284 | { |
c8fdfe29 | 3285 | u8 blkbits = inode->i_blkbits; |
364443cb | 3286 | |
c8fdfe29 MB |
3287 | /* |
3288 | * Writes that span EOF might trigger an I/O size update on completion, | |
3289 | * so consider them to be dirty for the purpose of O_DSYNC, even if | |
3290 | * there is no other metadata changes being made or are pending. | |
3291 | */ | |
364443cb | 3292 | iomap->flags = 0; |
c8fdfe29 MB |
3293 | if (ext4_inode_datasync_dirty(inode) || |
3294 | offset + length > i_size_read(inode)) | |
b8a6176c | 3295 | iomap->flags |= IOMAP_F_DIRTY; |
c8fdfe29 MB |
3296 | |
3297 | if (map->m_flags & EXT4_MAP_NEW) | |
3298 | iomap->flags |= IOMAP_F_NEW; | |
3299 | ||
5e405595 | 3300 | iomap->bdev = inode->i_sb->s_bdev; |
c8fdfe29 MB |
3301 | iomap->dax_dev = EXT4_SB(inode->i_sb)->s_daxdev; |
3302 | iomap->offset = (u64) map->m_lblk << blkbits; | |
3303 | iomap->length = (u64) map->m_len << blkbits; | |
364443cb | 3304 | |
6386722a RH |
3305 | if ((map->m_flags & EXT4_MAP_MAPPED) && |
3306 | !ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
3307 | iomap->flags |= IOMAP_F_MERGED; | |
3308 | ||
c8fdfe29 MB |
3309 | /* |
3310 | * Flags passed to ext4_map_blocks() for direct I/O writes can result | |
3311 | * in m_flags having both EXT4_MAP_MAPPED and EXT4_MAP_UNWRITTEN bits | |
3312 | * set. In order for any allocated unwritten extents to be converted | |
3313 | * into written extents correctly within the ->end_io() handler, we | |
3314 | * need to ensure that the iomap->type is set appropriately. Hence, the | |
3315 | * reason why we need to check whether the EXT4_MAP_UNWRITTEN bit has | |
3316 | * been set first. | |
3317 | */ | |
3318 | if (map->m_flags & EXT4_MAP_UNWRITTEN) { | |
3319 | iomap->type = IOMAP_UNWRITTEN; | |
3320 | iomap->addr = (u64) map->m_pblk << blkbits; | |
3321 | } else if (map->m_flags & EXT4_MAP_MAPPED) { | |
3322 | iomap->type = IOMAP_MAPPED; | |
3323 | iomap->addr = (u64) map->m_pblk << blkbits; | |
364443cb | 3324 | } else { |
c8fdfe29 MB |
3325 | iomap->type = IOMAP_HOLE; |
3326 | iomap->addr = IOMAP_NULL_ADDR; | |
364443cb | 3327 | } |
364443cb JK |
3328 | } |
3329 | ||
f063db5e MB |
3330 | static int ext4_iomap_alloc(struct inode *inode, struct ext4_map_blocks *map, |
3331 | unsigned int flags) | |
776722e8 | 3332 | { |
776722e8 | 3333 | handle_t *handle; |
378f32ba MB |
3334 | u8 blkbits = inode->i_blkbits; |
3335 | int ret, dio_credits, m_flags = 0, retries = 0; | |
776722e8 | 3336 | |
776722e8 | 3337 | /* |
f063db5e MB |
3338 | * Trim the mapping request to the maximum value that we can map at |
3339 | * once for direct I/O. | |
776722e8 | 3340 | */ |
f063db5e MB |
3341 | if (map->m_len > DIO_MAX_BLOCKS) |
3342 | map->m_len = DIO_MAX_BLOCKS; | |
3343 | dio_credits = ext4_chunk_trans_blocks(inode, map->m_len); | |
776722e8 | 3344 | |
f063db5e | 3345 | retry: |
776722e8 | 3346 | /* |
f063db5e MB |
3347 | * Either we allocate blocks and then don't get an unwritten extent, so |
3348 | * in that case we have reserved enough credits. Or, the blocks are | |
3349 | * already allocated and unwritten. In that case, the extent conversion | |
3350 | * fits into the credits as well. | |
776722e8 | 3351 | */ |
f063db5e MB |
3352 | handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits); |
3353 | if (IS_ERR(handle)) | |
3354 | return PTR_ERR(handle); | |
4c0425ff | 3355 | |
378f32ba MB |
3356 | /* |
3357 | * DAX and direct I/O are the only two operations that are currently | |
3358 | * supported with IOMAP_WRITE. | |
3359 | */ | |
3360 | WARN_ON(!IS_DAX(inode) && !(flags & IOMAP_DIRECT)); | |
3361 | if (IS_DAX(inode)) | |
3362 | m_flags = EXT4_GET_BLOCKS_CREATE_ZERO; | |
3363 | /* | |
3364 | * We use i_size instead of i_disksize here because delalloc writeback | |
3365 | * can complete at any point during the I/O and subsequently push the | |
3366 | * i_disksize out to i_size. This could be beyond where direct I/O is | |
3367 | * happening and thus expose allocated blocks to direct I/O reads. | |
3368 | */ | |
d0b040f5 | 3369 | else if (((loff_t)map->m_lblk << blkbits) >= i_size_read(inode)) |
378f32ba MB |
3370 | m_flags = EXT4_GET_BLOCKS_CREATE; |
3371 | else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
3372 | m_flags = EXT4_GET_BLOCKS_IO_CREATE_EXT; | |
4b70df18 | 3373 | |
378f32ba | 3374 | ret = ext4_map_blocks(handle, inode, map, m_flags); |
8d5d02e6 | 3375 | |
74c66bcb | 3376 | /* |
378f32ba MB |
3377 | * We cannot fill holes in indirect tree based inodes as that could |
3378 | * expose stale data in the case of a crash. Use the magic error code | |
3379 | * to fallback to buffered I/O. | |
74c66bcb | 3380 | */ |
378f32ba MB |
3381 | if (!m_flags && !ret) |
3382 | ret = -ENOTBLK; | |
187372a3 | 3383 | |
f063db5e MB |
3384 | ext4_journal_stop(handle); |
3385 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) | |
3386 | goto retry; | |
3387 | ||
3388 | return ret; | |
4c0425ff | 3389 | } |
c7064ef1 | 3390 | |
f063db5e | 3391 | |
364443cb | 3392 | static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length, |
c039b997 | 3393 | unsigned flags, struct iomap *iomap, struct iomap *srcmap) |
4c0425ff | 3394 | { |
364443cb | 3395 | int ret; |
09edf4d3 MB |
3396 | struct ext4_map_blocks map; |
3397 | u8 blkbits = inode->i_blkbits; | |
729f52c6 | 3398 | |
bcd8e91f TT |
3399 | if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK) |
3400 | return -EINVAL; | |
4bd809db | 3401 | |
09edf4d3 MB |
3402 | if (WARN_ON_ONCE(ext4_has_inline_data(inode))) |
3403 | return -ERANGE; | |
4bd809db | 3404 | |
e8340395 | 3405 | /* |
09edf4d3 | 3406 | * Calculate the first and last logical blocks respectively. |
e8340395 | 3407 | */ |
09edf4d3 MB |
3408 | map.m_lblk = offset >> blkbits; |
3409 | map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits, | |
3410 | EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1; | |
e8340395 | 3411 | |
9faac62d RH |
3412 | if (flags & IOMAP_WRITE) { |
3413 | /* | |
3414 | * We check here if the blocks are already allocated, then we | |
3415 | * don't need to start a journal txn and we can directly return | |
3416 | * the mapping information. This could boost performance | |
3417 | * especially in multi-threaded overwrite requests. | |
3418 | */ | |
3419 | if (offset + length <= i_size_read(inode)) { | |
3420 | ret = ext4_map_blocks(NULL, inode, &map, 0); | |
3421 | if (ret > 0 && (map.m_flags & EXT4_MAP_MAPPED)) | |
3422 | goto out; | |
3423 | } | |
f063db5e | 3424 | ret = ext4_iomap_alloc(inode, &map, flags); |
9faac62d | 3425 | } else { |
545052e9 | 3426 | ret = ext4_map_blocks(NULL, inode, &map, 0); |
9faac62d | 3427 | } |
4bd809db | 3428 | |
f063db5e MB |
3429 | if (ret < 0) |
3430 | return ret; | |
9faac62d | 3431 | out: |
c8fdfe29 | 3432 | ext4_set_iomap(inode, iomap, &map, offset, length); |
4bd809db | 3433 | |
364443cb JK |
3434 | return 0; |
3435 | } | |
8d5d02e6 | 3436 | |
8cd115bd JK |
3437 | static int ext4_iomap_overwrite_begin(struct inode *inode, loff_t offset, |
3438 | loff_t length, unsigned flags, struct iomap *iomap, | |
3439 | struct iomap *srcmap) | |
3440 | { | |
3441 | int ret; | |
3442 | ||
3443 | /* | |
3444 | * Even for writes we don't need to allocate blocks, so just pretend | |
3445 | * we are reading to save overhead of starting a transaction. | |
3446 | */ | |
3447 | flags &= ~IOMAP_WRITE; | |
3448 | ret = ext4_iomap_begin(inode, offset, length, flags, iomap, srcmap); | |
3449 | WARN_ON_ONCE(iomap->type != IOMAP_MAPPED); | |
3450 | return ret; | |
3451 | } | |
3452 | ||
776722e8 JK |
3453 | static int ext4_iomap_end(struct inode *inode, loff_t offset, loff_t length, |
3454 | ssize_t written, unsigned flags, struct iomap *iomap) | |
3455 | { | |
69c499d1 | 3456 | /* |
378f32ba MB |
3457 | * Check to see whether an error occurred while writing out the data to |
3458 | * the allocated blocks. If so, return the magic error code so that we | |
3459 | * fallback to buffered I/O and attempt to complete the remainder of | |
3460 | * the I/O. Any blocks that may have been allocated in preparation for | |
3461 | * the direct I/O will be reused during buffered I/O. | |
69c499d1 | 3462 | */ |
378f32ba MB |
3463 | if (flags & (IOMAP_WRITE | IOMAP_DIRECT) && written == 0) |
3464 | return -ENOTBLK; | |
69c499d1 | 3465 | |
569342dc | 3466 | return 0; |
776722e8 | 3467 | } |
4bd809db | 3468 | |
8ff6daa1 | 3469 | const struct iomap_ops ext4_iomap_ops = { |
364443cb | 3470 | .iomap_begin = ext4_iomap_begin, |
776722e8 | 3471 | .iomap_end = ext4_iomap_end, |
364443cb | 3472 | }; |
8d5d02e6 | 3473 | |
8cd115bd JK |
3474 | const struct iomap_ops ext4_iomap_overwrite_ops = { |
3475 | .iomap_begin = ext4_iomap_overwrite_begin, | |
3476 | .iomap_end = ext4_iomap_end, | |
3477 | }; | |
3478 | ||
09edf4d3 MB |
3479 | static bool ext4_iomap_is_delalloc(struct inode *inode, |
3480 | struct ext4_map_blocks *map) | |
3481 | { | |
3482 | struct extent_status es; | |
3483 | ext4_lblk_t offset = 0, end = map->m_lblk + map->m_len - 1; | |
914f82a3 | 3484 | |
09edf4d3 MB |
3485 | ext4_es_find_extent_range(inode, &ext4_es_is_delayed, |
3486 | map->m_lblk, end, &es); | |
914f82a3 | 3487 | |
09edf4d3 MB |
3488 | if (!es.es_len || es.es_lblk > end) |
3489 | return false; | |
914f82a3 | 3490 | |
09edf4d3 MB |
3491 | if (es.es_lblk > map->m_lblk) { |
3492 | map->m_len = es.es_lblk - map->m_lblk; | |
3493 | return false; | |
914f82a3 | 3494 | } |
914f82a3 | 3495 | |
09edf4d3 MB |
3496 | offset = map->m_lblk - es.es_lblk; |
3497 | map->m_len = es.es_len - offset; | |
914f82a3 | 3498 | |
09edf4d3 | 3499 | return true; |
4c0425ff MC |
3500 | } |
3501 | ||
09edf4d3 MB |
3502 | static int ext4_iomap_begin_report(struct inode *inode, loff_t offset, |
3503 | loff_t length, unsigned int flags, | |
3504 | struct iomap *iomap, struct iomap *srcmap) | |
4c0425ff | 3505 | { |
09edf4d3 MB |
3506 | int ret; |
3507 | bool delalloc = false; | |
3508 | struct ext4_map_blocks map; | |
3509 | u8 blkbits = inode->i_blkbits; | |
4c0425ff | 3510 | |
09edf4d3 MB |
3511 | if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK) |
3512 | return -EINVAL; | |
3513 | ||
3514 | if (ext4_has_inline_data(inode)) { | |
3515 | ret = ext4_inline_data_iomap(inode, iomap); | |
3516 | if (ret != -EAGAIN) { | |
3517 | if (ret == 0 && offset >= iomap->length) | |
3518 | ret = -ENOENT; | |
3519 | return ret; | |
3520 | } | |
3521 | } | |
2058f83a | 3522 | |
84ebd795 | 3523 | /* |
09edf4d3 | 3524 | * Calculate the first and last logical block respectively. |
84ebd795 | 3525 | */ |
09edf4d3 MB |
3526 | map.m_lblk = offset >> blkbits; |
3527 | map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits, | |
3528 | EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1; | |
84ebd795 | 3529 | |
b2c57642 RH |
3530 | /* |
3531 | * Fiemap callers may call for offset beyond s_bitmap_maxbytes. | |
3532 | * So handle it here itself instead of querying ext4_map_blocks(). | |
3533 | * Since ext4_map_blocks() will warn about it and will return | |
3534 | * -EIO error. | |
3535 | */ | |
3536 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) { | |
3537 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
3538 | ||
3539 | if (offset >= sbi->s_bitmap_maxbytes) { | |
3540 | map.m_flags = 0; | |
3541 | goto set_iomap; | |
3542 | } | |
3543 | } | |
3544 | ||
09edf4d3 MB |
3545 | ret = ext4_map_blocks(NULL, inode, &map, 0); |
3546 | if (ret < 0) | |
3547 | return ret; | |
3548 | if (ret == 0) | |
3549 | delalloc = ext4_iomap_is_delalloc(inode, &map); | |
46c7f254 | 3550 | |
b2c57642 | 3551 | set_iomap: |
09edf4d3 MB |
3552 | ext4_set_iomap(inode, iomap, &map, offset, length); |
3553 | if (delalloc && iomap->type == IOMAP_HOLE) | |
3554 | iomap->type = IOMAP_DELALLOC; | |
3555 | ||
3556 | return 0; | |
4c0425ff MC |
3557 | } |
3558 | ||
09edf4d3 MB |
3559 | const struct iomap_ops ext4_iomap_report_ops = { |
3560 | .iomap_begin = ext4_iomap_begin_report, | |
3561 | }; | |
3562 | ||
ac27a0ec | 3563 | /* |
617ba13b | 3564 | * Pages can be marked dirty completely asynchronously from ext4's journalling |
ac27a0ec DK |
3565 | * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do |
3566 | * much here because ->set_page_dirty is called under VFS locks. The page is | |
3567 | * not necessarily locked. | |
3568 | * | |
3569 | * We cannot just dirty the page and leave attached buffers clean, because the | |
3570 | * buffers' dirty state is "definitive". We cannot just set the buffers dirty | |
3571 | * or jbddirty because all the journalling code will explode. | |
3572 | * | |
3573 | * So what we do is to mark the page "pending dirty" and next time writepage | |
3574 | * is called, propagate that into the buffers appropriately. | |
3575 | */ | |
617ba13b | 3576 | static int ext4_journalled_set_page_dirty(struct page *page) |
ac27a0ec DK |
3577 | { |
3578 | SetPageChecked(page); | |
3579 | return __set_page_dirty_nobuffers(page); | |
3580 | } | |
3581 | ||
6dcc693b JK |
3582 | static int ext4_set_page_dirty(struct page *page) |
3583 | { | |
3584 | WARN_ON_ONCE(!PageLocked(page) && !PageDirty(page)); | |
3585 | WARN_ON_ONCE(!page_has_buffers(page)); | |
3586 | return __set_page_dirty_buffers(page); | |
3587 | } | |
3588 | ||
0e6895ba RH |
3589 | static int ext4_iomap_swap_activate(struct swap_info_struct *sis, |
3590 | struct file *file, sector_t *span) | |
3591 | { | |
3592 | return iomap_swapfile_activate(sis, file, span, | |
3593 | &ext4_iomap_report_ops); | |
3594 | } | |
3595 | ||
74d553aa | 3596 | static const struct address_space_operations ext4_aops = { |
8ab22b9a | 3597 | .readpage = ext4_readpage, |
6311f91f | 3598 | .readahead = ext4_readahead, |
43ce1d23 | 3599 | .writepage = ext4_writepage, |
20970ba6 | 3600 | .writepages = ext4_writepages, |
8ab22b9a | 3601 | .write_begin = ext4_write_begin, |
74d553aa | 3602 | .write_end = ext4_write_end, |
6dcc693b | 3603 | .set_page_dirty = ext4_set_page_dirty, |
8ab22b9a HH |
3604 | .bmap = ext4_bmap, |
3605 | .invalidatepage = ext4_invalidatepage, | |
3606 | .releasepage = ext4_releasepage, | |
378f32ba | 3607 | .direct_IO = noop_direct_IO, |
8ab22b9a HH |
3608 | .migratepage = buffer_migrate_page, |
3609 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 3610 | .error_remove_page = generic_error_remove_page, |
0e6895ba | 3611 | .swap_activate = ext4_iomap_swap_activate, |
ac27a0ec DK |
3612 | }; |
3613 | ||
617ba13b | 3614 | static const struct address_space_operations ext4_journalled_aops = { |
8ab22b9a | 3615 | .readpage = ext4_readpage, |
6311f91f | 3616 | .readahead = ext4_readahead, |
43ce1d23 | 3617 | .writepage = ext4_writepage, |
20970ba6 | 3618 | .writepages = ext4_writepages, |
8ab22b9a HH |
3619 | .write_begin = ext4_write_begin, |
3620 | .write_end = ext4_journalled_write_end, | |
3621 | .set_page_dirty = ext4_journalled_set_page_dirty, | |
3622 | .bmap = ext4_bmap, | |
4520fb3c | 3623 | .invalidatepage = ext4_journalled_invalidatepage, |
8ab22b9a | 3624 | .releasepage = ext4_releasepage, |
378f32ba | 3625 | .direct_IO = noop_direct_IO, |
8ab22b9a | 3626 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 3627 | .error_remove_page = generic_error_remove_page, |
0e6895ba | 3628 | .swap_activate = ext4_iomap_swap_activate, |
ac27a0ec DK |
3629 | }; |
3630 | ||
64769240 | 3631 | static const struct address_space_operations ext4_da_aops = { |
8ab22b9a | 3632 | .readpage = ext4_readpage, |
6311f91f | 3633 | .readahead = ext4_readahead, |
43ce1d23 | 3634 | .writepage = ext4_writepage, |
20970ba6 | 3635 | .writepages = ext4_writepages, |
8ab22b9a HH |
3636 | .write_begin = ext4_da_write_begin, |
3637 | .write_end = ext4_da_write_end, | |
6dcc693b | 3638 | .set_page_dirty = ext4_set_page_dirty, |
8ab22b9a | 3639 | .bmap = ext4_bmap, |
8fcc3a58 | 3640 | .invalidatepage = ext4_invalidatepage, |
8ab22b9a | 3641 | .releasepage = ext4_releasepage, |
378f32ba | 3642 | .direct_IO = noop_direct_IO, |
8ab22b9a HH |
3643 | .migratepage = buffer_migrate_page, |
3644 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 3645 | .error_remove_page = generic_error_remove_page, |
0e6895ba | 3646 | .swap_activate = ext4_iomap_swap_activate, |
64769240 AT |
3647 | }; |
3648 | ||
5f0663bb DW |
3649 | static const struct address_space_operations ext4_dax_aops = { |
3650 | .writepages = ext4_dax_writepages, | |
3651 | .direct_IO = noop_direct_IO, | |
b82a96c9 | 3652 | .set_page_dirty = __set_page_dirty_no_writeback, |
94dbb631 | 3653 | .bmap = ext4_bmap, |
5f0663bb | 3654 | .invalidatepage = noop_invalidatepage, |
0e6895ba | 3655 | .swap_activate = ext4_iomap_swap_activate, |
5f0663bb DW |
3656 | }; |
3657 | ||
617ba13b | 3658 | void ext4_set_aops(struct inode *inode) |
ac27a0ec | 3659 | { |
3d2b1582 LC |
3660 | switch (ext4_inode_journal_mode(inode)) { |
3661 | case EXT4_INODE_ORDERED_DATA_MODE: | |
3d2b1582 | 3662 | case EXT4_INODE_WRITEBACK_DATA_MODE: |
3d2b1582 LC |
3663 | break; |
3664 | case EXT4_INODE_JOURNAL_DATA_MODE: | |
617ba13b | 3665 | inode->i_mapping->a_ops = &ext4_journalled_aops; |
74d553aa | 3666 | return; |
3d2b1582 LC |
3667 | default: |
3668 | BUG(); | |
3669 | } | |
5f0663bb DW |
3670 | if (IS_DAX(inode)) |
3671 | inode->i_mapping->a_ops = &ext4_dax_aops; | |
3672 | else if (test_opt(inode->i_sb, DELALLOC)) | |
74d553aa TT |
3673 | inode->i_mapping->a_ops = &ext4_da_aops; |
3674 | else | |
3675 | inode->i_mapping->a_ops = &ext4_aops; | |
ac27a0ec DK |
3676 | } |
3677 | ||
923ae0ff | 3678 | static int __ext4_block_zero_page_range(handle_t *handle, |
d863dc36 LC |
3679 | struct address_space *mapping, loff_t from, loff_t length) |
3680 | { | |
09cbfeaf KS |
3681 | ext4_fsblk_t index = from >> PAGE_SHIFT; |
3682 | unsigned offset = from & (PAGE_SIZE-1); | |
923ae0ff | 3683 | unsigned blocksize, pos; |
d863dc36 LC |
3684 | ext4_lblk_t iblock; |
3685 | struct inode *inode = mapping->host; | |
3686 | struct buffer_head *bh; | |
3687 | struct page *page; | |
3688 | int err = 0; | |
3689 | ||
09cbfeaf | 3690 | page = find_or_create_page(mapping, from >> PAGE_SHIFT, |
c62d2555 | 3691 | mapping_gfp_constraint(mapping, ~__GFP_FS)); |
d863dc36 LC |
3692 | if (!page) |
3693 | return -ENOMEM; | |
3694 | ||
3695 | blocksize = inode->i_sb->s_blocksize; | |
d863dc36 | 3696 | |
09cbfeaf | 3697 | iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits); |
d863dc36 LC |
3698 | |
3699 | if (!page_has_buffers(page)) | |
3700 | create_empty_buffers(page, blocksize, 0); | |
3701 | ||
3702 | /* Find the buffer that contains "offset" */ | |
3703 | bh = page_buffers(page); | |
3704 | pos = blocksize; | |
3705 | while (offset >= pos) { | |
3706 | bh = bh->b_this_page; | |
3707 | iblock++; | |
3708 | pos += blocksize; | |
3709 | } | |
d863dc36 LC |
3710 | if (buffer_freed(bh)) { |
3711 | BUFFER_TRACE(bh, "freed: skip"); | |
3712 | goto unlock; | |
3713 | } | |
d863dc36 LC |
3714 | if (!buffer_mapped(bh)) { |
3715 | BUFFER_TRACE(bh, "unmapped"); | |
3716 | ext4_get_block(inode, iblock, bh, 0); | |
3717 | /* unmapped? It's a hole - nothing to do */ | |
3718 | if (!buffer_mapped(bh)) { | |
3719 | BUFFER_TRACE(bh, "still unmapped"); | |
3720 | goto unlock; | |
3721 | } | |
3722 | } | |
3723 | ||
3724 | /* Ok, it's mapped. Make sure it's up-to-date */ | |
3725 | if (PageUptodate(page)) | |
3726 | set_buffer_uptodate(bh); | |
3727 | ||
3728 | if (!buffer_uptodate(bh)) { | |
2d069c08 | 3729 | err = ext4_read_bh_lock(bh, 0, true); |
3730 | if (err) | |
d863dc36 | 3731 | goto unlock; |
4f74d15f | 3732 | if (fscrypt_inode_uses_fs_layer_crypto(inode)) { |
c9c7429c | 3733 | /* We expect the key to be set. */ |
a7550b30 | 3734 | BUG_ON(!fscrypt_has_encryption_key(inode)); |
834f1565 EB |
3735 | err = fscrypt_decrypt_pagecache_blocks(page, blocksize, |
3736 | bh_offset(bh)); | |
3737 | if (err) { | |
3738 | clear_buffer_uptodate(bh); | |
3739 | goto unlock; | |
3740 | } | |
c9c7429c | 3741 | } |
d863dc36 | 3742 | } |
d863dc36 LC |
3743 | if (ext4_should_journal_data(inode)) { |
3744 | BUFFER_TRACE(bh, "get write access"); | |
188c299e JK |
3745 | err = ext4_journal_get_write_access(handle, inode->i_sb, bh, |
3746 | EXT4_JTR_NONE); | |
d863dc36 LC |
3747 | if (err) |
3748 | goto unlock; | |
3749 | } | |
d863dc36 | 3750 | zero_user(page, offset, length); |
d863dc36 LC |
3751 | BUFFER_TRACE(bh, "zeroed end of block"); |
3752 | ||
d863dc36 LC |
3753 | if (ext4_should_journal_data(inode)) { |
3754 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
0713ed0c | 3755 | } else { |
353eefd3 | 3756 | err = 0; |
d863dc36 | 3757 | mark_buffer_dirty(bh); |
3957ef53 | 3758 | if (ext4_should_order_data(inode)) |
73131fbb RZ |
3759 | err = ext4_jbd2_inode_add_write(handle, inode, from, |
3760 | length); | |
0713ed0c | 3761 | } |
d863dc36 LC |
3762 | |
3763 | unlock: | |
3764 | unlock_page(page); | |
09cbfeaf | 3765 | put_page(page); |
d863dc36 LC |
3766 | return err; |
3767 | } | |
3768 | ||
923ae0ff RZ |
3769 | /* |
3770 | * ext4_block_zero_page_range() zeros out a mapping of length 'length' | |
3771 | * starting from file offset 'from'. The range to be zero'd must | |
3772 | * be contained with in one block. If the specified range exceeds | |
3773 | * the end of the block it will be shortened to end of the block | |
3088e5a5 | 3774 | * that corresponds to 'from' |
923ae0ff RZ |
3775 | */ |
3776 | static int ext4_block_zero_page_range(handle_t *handle, | |
3777 | struct address_space *mapping, loff_t from, loff_t length) | |
3778 | { | |
3779 | struct inode *inode = mapping->host; | |
09cbfeaf | 3780 | unsigned offset = from & (PAGE_SIZE-1); |
923ae0ff RZ |
3781 | unsigned blocksize = inode->i_sb->s_blocksize; |
3782 | unsigned max = blocksize - (offset & (blocksize - 1)); | |
3783 | ||
3784 | /* | |
3785 | * correct length if it does not fall between | |
3786 | * 'from' and the end of the block | |
3787 | */ | |
3788 | if (length > max || length < 0) | |
3789 | length = max; | |
3790 | ||
47e69351 JK |
3791 | if (IS_DAX(inode)) { |
3792 | return iomap_zero_range(inode, from, length, NULL, | |
3793 | &ext4_iomap_ops); | |
3794 | } | |
923ae0ff RZ |
3795 | return __ext4_block_zero_page_range(handle, mapping, from, length); |
3796 | } | |
3797 | ||
94350ab5 MW |
3798 | /* |
3799 | * ext4_block_truncate_page() zeroes out a mapping from file offset `from' | |
3800 | * up to the end of the block which corresponds to `from'. | |
3801 | * This required during truncate. We need to physically zero the tail end | |
3802 | * of that block so it doesn't yield old data if the file is later grown. | |
3803 | */ | |
c197855e | 3804 | static int ext4_block_truncate_page(handle_t *handle, |
94350ab5 MW |
3805 | struct address_space *mapping, loff_t from) |
3806 | { | |
09cbfeaf | 3807 | unsigned offset = from & (PAGE_SIZE-1); |
94350ab5 MW |
3808 | unsigned length; |
3809 | unsigned blocksize; | |
3810 | struct inode *inode = mapping->host; | |
3811 | ||
0d06863f | 3812 | /* If we are processing an encrypted inode during orphan list handling */ |
592ddec7 | 3813 | if (IS_ENCRYPTED(inode) && !fscrypt_has_encryption_key(inode)) |
0d06863f TT |
3814 | return 0; |
3815 | ||
94350ab5 MW |
3816 | blocksize = inode->i_sb->s_blocksize; |
3817 | length = blocksize - (offset & (blocksize - 1)); | |
3818 | ||
3819 | return ext4_block_zero_page_range(handle, mapping, from, length); | |
3820 | } | |
3821 | ||
a87dd18c LC |
3822 | int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode, |
3823 | loff_t lstart, loff_t length) | |
3824 | { | |
3825 | struct super_block *sb = inode->i_sb; | |
3826 | struct address_space *mapping = inode->i_mapping; | |
e1be3a92 | 3827 | unsigned partial_start, partial_end; |
a87dd18c LC |
3828 | ext4_fsblk_t start, end; |
3829 | loff_t byte_end = (lstart + length - 1); | |
3830 | int err = 0; | |
3831 | ||
e1be3a92 LC |
3832 | partial_start = lstart & (sb->s_blocksize - 1); |
3833 | partial_end = byte_end & (sb->s_blocksize - 1); | |
3834 | ||
a87dd18c LC |
3835 | start = lstart >> sb->s_blocksize_bits; |
3836 | end = byte_end >> sb->s_blocksize_bits; | |
3837 | ||
3838 | /* Handle partial zero within the single block */ | |
e1be3a92 LC |
3839 | if (start == end && |
3840 | (partial_start || (partial_end != sb->s_blocksize - 1))) { | |
a87dd18c LC |
3841 | err = ext4_block_zero_page_range(handle, mapping, |
3842 | lstart, length); | |
3843 | return err; | |
3844 | } | |
3845 | /* Handle partial zero out on the start of the range */ | |
e1be3a92 | 3846 | if (partial_start) { |
a87dd18c LC |
3847 | err = ext4_block_zero_page_range(handle, mapping, |
3848 | lstart, sb->s_blocksize); | |
3849 | if (err) | |
3850 | return err; | |
3851 | } | |
3852 | /* Handle partial zero out on the end of the range */ | |
e1be3a92 | 3853 | if (partial_end != sb->s_blocksize - 1) |
a87dd18c | 3854 | err = ext4_block_zero_page_range(handle, mapping, |
e1be3a92 LC |
3855 | byte_end - partial_end, |
3856 | partial_end + 1); | |
a87dd18c LC |
3857 | return err; |
3858 | } | |
3859 | ||
91ef4caf DG |
3860 | int ext4_can_truncate(struct inode *inode) |
3861 | { | |
91ef4caf DG |
3862 | if (S_ISREG(inode->i_mode)) |
3863 | return 1; | |
3864 | if (S_ISDIR(inode->i_mode)) | |
3865 | return 1; | |
3866 | if (S_ISLNK(inode->i_mode)) | |
3867 | return !ext4_inode_is_fast_symlink(inode); | |
3868 | return 0; | |
3869 | } | |
3870 | ||
01127848 JK |
3871 | /* |
3872 | * We have to make sure i_disksize gets properly updated before we truncate | |
3873 | * page cache due to hole punching or zero range. Otherwise i_disksize update | |
3874 | * can get lost as it may have been postponed to submission of writeback but | |
3875 | * that will never happen after we truncate page cache. | |
3876 | */ | |
3877 | int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset, | |
3878 | loff_t len) | |
3879 | { | |
3880 | handle_t *handle; | |
4209ae12 HS |
3881 | int ret; |
3882 | ||
01127848 JK |
3883 | loff_t size = i_size_read(inode); |
3884 | ||
5955102c | 3885 | WARN_ON(!inode_is_locked(inode)); |
01127848 JK |
3886 | if (offset > size || offset + len < size) |
3887 | return 0; | |
3888 | ||
3889 | if (EXT4_I(inode)->i_disksize >= size) | |
3890 | return 0; | |
3891 | ||
3892 | handle = ext4_journal_start(inode, EXT4_HT_MISC, 1); | |
3893 | if (IS_ERR(handle)) | |
3894 | return PTR_ERR(handle); | |
3895 | ext4_update_i_disksize(inode, size); | |
4209ae12 | 3896 | ret = ext4_mark_inode_dirty(handle, inode); |
01127848 JK |
3897 | ext4_journal_stop(handle); |
3898 | ||
4209ae12 | 3899 | return ret; |
01127848 JK |
3900 | } |
3901 | ||
d4f5258e | 3902 | static void ext4_wait_dax_page(struct inode *inode) |
430657b6 | 3903 | { |
d4f5258e | 3904 | filemap_invalidate_unlock(inode->i_mapping); |
430657b6 | 3905 | schedule(); |
d4f5258e | 3906 | filemap_invalidate_lock(inode->i_mapping); |
430657b6 RZ |
3907 | } |
3908 | ||
3909 | int ext4_break_layouts(struct inode *inode) | |
3910 | { | |
430657b6 | 3911 | struct page *page; |
430657b6 RZ |
3912 | int error; |
3913 | ||
d4f5258e | 3914 | if (WARN_ON_ONCE(!rwsem_is_locked(&inode->i_mapping->invalidate_lock))) |
430657b6 RZ |
3915 | return -EINVAL; |
3916 | ||
3917 | do { | |
430657b6 RZ |
3918 | page = dax_layout_busy_page(inode->i_mapping); |
3919 | if (!page) | |
3920 | return 0; | |
3921 | ||
3922 | error = ___wait_var_event(&page->_refcount, | |
3923 | atomic_read(&page->_refcount) == 1, | |
3924 | TASK_INTERRUPTIBLE, 0, 0, | |
d4f5258e | 3925 | ext4_wait_dax_page(inode)); |
b1f38217 | 3926 | } while (error == 0); |
430657b6 RZ |
3927 | |
3928 | return error; | |
3929 | } | |
3930 | ||
a4bb6b64 | 3931 | /* |
cca32b7e | 3932 | * ext4_punch_hole: punches a hole in a file by releasing the blocks |
a4bb6b64 AH |
3933 | * associated with the given offset and length |
3934 | * | |
3935 | * @inode: File inode | |
3936 | * @offset: The offset where the hole will begin | |
3937 | * @len: The length of the hole | |
3938 | * | |
4907cb7b | 3939 | * Returns: 0 on success or negative on failure |
a4bb6b64 AH |
3940 | */ |
3941 | ||
aeb2817a | 3942 | int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length) |
a4bb6b64 | 3943 | { |
26a4c0c6 TT |
3944 | struct super_block *sb = inode->i_sb; |
3945 | ext4_lblk_t first_block, stop_block; | |
3946 | struct address_space *mapping = inode->i_mapping; | |
5f430795 TS |
3947 | loff_t first_block_offset, last_block_offset, max_length; |
3948 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
26a4c0c6 TT |
3949 | handle_t *handle; |
3950 | unsigned int credits; | |
4209ae12 | 3951 | int ret = 0, ret2 = 0; |
26a4c0c6 | 3952 | |
b8a86845 | 3953 | trace_ext4_punch_hole(inode, offset, length, 0); |
aaddea81 | 3954 | |
c1e8220b TT |
3955 | ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA); |
3956 | if (ext4_has_inline_data(inode)) { | |
d4f5258e | 3957 | filemap_invalidate_lock(mapping); |
c1e8220b | 3958 | ret = ext4_convert_inline_data(inode); |
d4f5258e | 3959 | filemap_invalidate_unlock(mapping); |
c1e8220b TT |
3960 | if (ret) |
3961 | return ret; | |
3962 | } | |
3963 | ||
26a4c0c6 TT |
3964 | /* |
3965 | * Write out all dirty pages to avoid race conditions | |
3966 | * Then release them. | |
3967 | */ | |
cca32b7e | 3968 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { |
26a4c0c6 TT |
3969 | ret = filemap_write_and_wait_range(mapping, offset, |
3970 | offset + length - 1); | |
3971 | if (ret) | |
3972 | return ret; | |
3973 | } | |
3974 | ||
5955102c | 3975 | inode_lock(inode); |
9ef06cec | 3976 | |
26a4c0c6 TT |
3977 | /* No need to punch hole beyond i_size */ |
3978 | if (offset >= inode->i_size) | |
3979 | goto out_mutex; | |
3980 | ||
3981 | /* | |
3982 | * If the hole extends beyond i_size, set the hole | |
3983 | * to end after the page that contains i_size | |
3984 | */ | |
3985 | if (offset + length > inode->i_size) { | |
3986 | length = inode->i_size + | |
09cbfeaf | 3987 | PAGE_SIZE - (inode->i_size & (PAGE_SIZE - 1)) - |
26a4c0c6 TT |
3988 | offset; |
3989 | } | |
3990 | ||
5f430795 TS |
3991 | /* |
3992 | * For punch hole the length + offset needs to be within one block | |
3993 | * before last range. Adjust the length if it goes beyond that limit. | |
3994 | */ | |
3995 | max_length = sbi->s_bitmap_maxbytes - inode->i_sb->s_blocksize; | |
3996 | if (offset + length > max_length) | |
3997 | length = max_length - offset; | |
3998 | ||
a361293f JK |
3999 | if (offset & (sb->s_blocksize - 1) || |
4000 | (offset + length) & (sb->s_blocksize - 1)) { | |
4001 | /* | |
4002 | * Attach jinode to inode for jbd2 if we do any zeroing of | |
4003 | * partial block | |
4004 | */ | |
4005 | ret = ext4_inode_attach_jinode(inode); | |
4006 | if (ret < 0) | |
4007 | goto out_mutex; | |
4008 | ||
4009 | } | |
4010 | ||
ea3d7209 | 4011 | /* Wait all existing dio workers, newcomers will block on i_mutex */ |
ea3d7209 JK |
4012 | inode_dio_wait(inode); |
4013 | ||
4014 | /* | |
4015 | * Prevent page faults from reinstantiating pages we have released from | |
4016 | * page cache. | |
4017 | */ | |
d4f5258e | 4018 | filemap_invalidate_lock(mapping); |
430657b6 RZ |
4019 | |
4020 | ret = ext4_break_layouts(inode); | |
4021 | if (ret) | |
4022 | goto out_dio; | |
4023 | ||
a87dd18c LC |
4024 | first_block_offset = round_up(offset, sb->s_blocksize); |
4025 | last_block_offset = round_down((offset + length), sb->s_blocksize) - 1; | |
26a4c0c6 | 4026 | |
a87dd18c | 4027 | /* Now release the pages and zero block aligned part of pages*/ |
01127848 JK |
4028 | if (last_block_offset > first_block_offset) { |
4029 | ret = ext4_update_disksize_before_punch(inode, offset, length); | |
4030 | if (ret) | |
4031 | goto out_dio; | |
a87dd18c LC |
4032 | truncate_pagecache_range(inode, first_block_offset, |
4033 | last_block_offset); | |
01127848 | 4034 | } |
26a4c0c6 TT |
4035 | |
4036 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
4037 | credits = ext4_writepage_trans_blocks(inode); | |
4038 | else | |
4039 | credits = ext4_blocks_for_truncate(inode); | |
4040 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); | |
4041 | if (IS_ERR(handle)) { | |
4042 | ret = PTR_ERR(handle); | |
4043 | ext4_std_error(sb, ret); | |
4044 | goto out_dio; | |
4045 | } | |
4046 | ||
a87dd18c LC |
4047 | ret = ext4_zero_partial_blocks(handle, inode, offset, |
4048 | length); | |
4049 | if (ret) | |
4050 | goto out_stop; | |
26a4c0c6 TT |
4051 | |
4052 | first_block = (offset + sb->s_blocksize - 1) >> | |
4053 | EXT4_BLOCK_SIZE_BITS(sb); | |
4054 | stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb); | |
4055 | ||
eee597ac LC |
4056 | /* If there are blocks to remove, do it */ |
4057 | if (stop_block > first_block) { | |
26a4c0c6 | 4058 | |
eee597ac | 4059 | down_write(&EXT4_I(inode)->i_data_sem); |
27bc446e | 4060 | ext4_discard_preallocations(inode, 0); |
26a4c0c6 | 4061 | |
eee597ac LC |
4062 | ret = ext4_es_remove_extent(inode, first_block, |
4063 | stop_block - first_block); | |
4064 | if (ret) { | |
4065 | up_write(&EXT4_I(inode)->i_data_sem); | |
4066 | goto out_stop; | |
4067 | } | |
26a4c0c6 | 4068 | |
eee597ac LC |
4069 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
4070 | ret = ext4_ext_remove_space(inode, first_block, | |
4071 | stop_block - 1); | |
4072 | else | |
4073 | ret = ext4_ind_remove_space(handle, inode, first_block, | |
4074 | stop_block); | |
26a4c0c6 | 4075 | |
eee597ac LC |
4076 | up_write(&EXT4_I(inode)->i_data_sem); |
4077 | } | |
a80f7fcf | 4078 | ext4_fc_track_range(handle, inode, first_block, stop_block); |
26a4c0c6 TT |
4079 | if (IS_SYNC(inode)) |
4080 | ext4_handle_sync(handle); | |
e251f9bc | 4081 | |
eeca7ea1 | 4082 | inode->i_mtime = inode->i_ctime = current_time(inode); |
4209ae12 HS |
4083 | ret2 = ext4_mark_inode_dirty(handle, inode); |
4084 | if (unlikely(ret2)) | |
4085 | ret = ret2; | |
67a7d5f5 JK |
4086 | if (ret >= 0) |
4087 | ext4_update_inode_fsync_trans(handle, inode, 1); | |
26a4c0c6 TT |
4088 | out_stop: |
4089 | ext4_journal_stop(handle); | |
4090 | out_dio: | |
d4f5258e | 4091 | filemap_invalidate_unlock(mapping); |
26a4c0c6 | 4092 | out_mutex: |
5955102c | 4093 | inode_unlock(inode); |
26a4c0c6 | 4094 | return ret; |
a4bb6b64 AH |
4095 | } |
4096 | ||
a361293f JK |
4097 | int ext4_inode_attach_jinode(struct inode *inode) |
4098 | { | |
4099 | struct ext4_inode_info *ei = EXT4_I(inode); | |
4100 | struct jbd2_inode *jinode; | |
4101 | ||
4102 | if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal) | |
4103 | return 0; | |
4104 | ||
4105 | jinode = jbd2_alloc_inode(GFP_KERNEL); | |
4106 | spin_lock(&inode->i_lock); | |
4107 | if (!ei->jinode) { | |
4108 | if (!jinode) { | |
4109 | spin_unlock(&inode->i_lock); | |
4110 | return -ENOMEM; | |
4111 | } | |
4112 | ei->jinode = jinode; | |
4113 | jbd2_journal_init_jbd_inode(ei->jinode, inode); | |
4114 | jinode = NULL; | |
4115 | } | |
4116 | spin_unlock(&inode->i_lock); | |
4117 | if (unlikely(jinode != NULL)) | |
4118 | jbd2_free_inode(jinode); | |
4119 | return 0; | |
4120 | } | |
4121 | ||
ac27a0ec | 4122 | /* |
617ba13b | 4123 | * ext4_truncate() |
ac27a0ec | 4124 | * |
617ba13b MC |
4125 | * We block out ext4_get_block() block instantiations across the entire |
4126 | * transaction, and VFS/VM ensures that ext4_truncate() cannot run | |
ac27a0ec DK |
4127 | * simultaneously on behalf of the same inode. |
4128 | * | |
42b2aa86 | 4129 | * As we work through the truncate and commit bits of it to the journal there |
ac27a0ec DK |
4130 | * is one core, guiding principle: the file's tree must always be consistent on |
4131 | * disk. We must be able to restart the truncate after a crash. | |
4132 | * | |
4133 | * The file's tree may be transiently inconsistent in memory (although it | |
4134 | * probably isn't), but whenever we close off and commit a journal transaction, | |
4135 | * the contents of (the filesystem + the journal) must be consistent and | |
4136 | * restartable. It's pretty simple, really: bottom up, right to left (although | |
4137 | * left-to-right works OK too). | |
4138 | * | |
4139 | * Note that at recovery time, journal replay occurs *before* the restart of | |
4140 | * truncate against the orphan inode list. | |
4141 | * | |
4142 | * The committed inode has the new, desired i_size (which is the same as | |
617ba13b | 4143 | * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see |
ac27a0ec | 4144 | * that this inode's truncate did not complete and it will again call |
617ba13b MC |
4145 | * ext4_truncate() to have another go. So there will be instantiated blocks |
4146 | * to the right of the truncation point in a crashed ext4 filesystem. But | |
ac27a0ec | 4147 | * that's fine - as long as they are linked from the inode, the post-crash |
617ba13b | 4148 | * ext4_truncate() run will find them and release them. |
ac27a0ec | 4149 | */ |
2c98eb5e | 4150 | int ext4_truncate(struct inode *inode) |
ac27a0ec | 4151 | { |
819c4920 TT |
4152 | struct ext4_inode_info *ei = EXT4_I(inode); |
4153 | unsigned int credits; | |
4209ae12 | 4154 | int err = 0, err2; |
819c4920 TT |
4155 | handle_t *handle; |
4156 | struct address_space *mapping = inode->i_mapping; | |
819c4920 | 4157 | |
19b5ef61 TT |
4158 | /* |
4159 | * There is a possibility that we're either freeing the inode | |
e04027e8 | 4160 | * or it's a completely new inode. In those cases we might not |
19b5ef61 TT |
4161 | * have i_mutex locked because it's not necessary. |
4162 | */ | |
4163 | if (!(inode->i_state & (I_NEW|I_FREEING))) | |
5955102c | 4164 | WARN_ON(!inode_is_locked(inode)); |
0562e0ba JZ |
4165 | trace_ext4_truncate_enter(inode); |
4166 | ||
91ef4caf | 4167 | if (!ext4_can_truncate(inode)) |
9a5d265f | 4168 | goto out_trace; |
ac27a0ec | 4169 | |
5534fb5b | 4170 | if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC)) |
19f5fb7a | 4171 | ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE); |
7d8f9f7d | 4172 | |
aef1c851 TM |
4173 | if (ext4_has_inline_data(inode)) { |
4174 | int has_inline = 1; | |
4175 | ||
01daf945 | 4176 | err = ext4_inline_data_truncate(inode, &has_inline); |
9a5d265f | 4177 | if (err || has_inline) |
4178 | goto out_trace; | |
aef1c851 TM |
4179 | } |
4180 | ||
a361293f JK |
4181 | /* If we zero-out tail of the page, we have to create jinode for jbd2 */ |
4182 | if (inode->i_size & (inode->i_sb->s_blocksize - 1)) { | |
4183 | if (ext4_inode_attach_jinode(inode) < 0) | |
9a5d265f | 4184 | goto out_trace; |
a361293f JK |
4185 | } |
4186 | ||
819c4920 TT |
4187 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
4188 | credits = ext4_writepage_trans_blocks(inode); | |
4189 | else | |
4190 | credits = ext4_blocks_for_truncate(inode); | |
4191 | ||
4192 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); | |
9a5d265f | 4193 | if (IS_ERR(handle)) { |
4194 | err = PTR_ERR(handle); | |
4195 | goto out_trace; | |
4196 | } | |
819c4920 | 4197 | |
eb3544c6 LC |
4198 | if (inode->i_size & (inode->i_sb->s_blocksize - 1)) |
4199 | ext4_block_truncate_page(handle, mapping, inode->i_size); | |
819c4920 TT |
4200 | |
4201 | /* | |
4202 | * We add the inode to the orphan list, so that if this | |
4203 | * truncate spans multiple transactions, and we crash, we will | |
4204 | * resume the truncate when the filesystem recovers. It also | |
4205 | * marks the inode dirty, to catch the new size. | |
4206 | * | |
4207 | * Implication: the file must always be in a sane, consistent | |
4208 | * truncatable state while each transaction commits. | |
4209 | */ | |
2c98eb5e TT |
4210 | err = ext4_orphan_add(handle, inode); |
4211 | if (err) | |
819c4920 TT |
4212 | goto out_stop; |
4213 | ||
4214 | down_write(&EXT4_I(inode)->i_data_sem); | |
4215 | ||
27bc446e | 4216 | ext4_discard_preallocations(inode, 0); |
819c4920 | 4217 | |
ff9893dc | 4218 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
d0abb36d | 4219 | err = ext4_ext_truncate(handle, inode); |
ff9893dc | 4220 | else |
819c4920 TT |
4221 | ext4_ind_truncate(handle, inode); |
4222 | ||
4223 | up_write(&ei->i_data_sem); | |
d0abb36d TT |
4224 | if (err) |
4225 | goto out_stop; | |
819c4920 TT |
4226 | |
4227 | if (IS_SYNC(inode)) | |
4228 | ext4_handle_sync(handle); | |
4229 | ||
4230 | out_stop: | |
4231 | /* | |
4232 | * If this was a simple ftruncate() and the file will remain alive, | |
4233 | * then we need to clear up the orphan record which we created above. | |
4234 | * However, if this was a real unlink then we were called by | |
58d86a50 | 4235 | * ext4_evict_inode(), and we allow that function to clean up the |
819c4920 TT |
4236 | * orphan info for us. |
4237 | */ | |
4238 | if (inode->i_nlink) | |
4239 | ext4_orphan_del(handle, inode); | |
4240 | ||
eeca7ea1 | 4241 | inode->i_mtime = inode->i_ctime = current_time(inode); |
4209ae12 HS |
4242 | err2 = ext4_mark_inode_dirty(handle, inode); |
4243 | if (unlikely(err2 && !err)) | |
4244 | err = err2; | |
819c4920 | 4245 | ext4_journal_stop(handle); |
ac27a0ec | 4246 | |
9a5d265f | 4247 | out_trace: |
0562e0ba | 4248 | trace_ext4_truncate_exit(inode); |
2c98eb5e | 4249 | return err; |
ac27a0ec DK |
4250 | } |
4251 | ||
ac27a0ec | 4252 | /* |
617ba13b | 4253 | * ext4_get_inode_loc returns with an extra refcount against the inode's |
ac27a0ec DK |
4254 | * underlying buffer_head on success. If 'in_mem' is true, we have all |
4255 | * data in memory that is needed to recreate the on-disk version of this | |
4256 | * inode. | |
4257 | */ | |
8016e29f HS |
4258 | static int __ext4_get_inode_loc(struct super_block *sb, unsigned long ino, |
4259 | struct ext4_iloc *iloc, int in_mem, | |
4260 | ext4_fsblk_t *ret_block) | |
ac27a0ec | 4261 | { |
240799cd TT |
4262 | struct ext4_group_desc *gdp; |
4263 | struct buffer_head *bh; | |
240799cd | 4264 | ext4_fsblk_t block; |
02f03c42 | 4265 | struct blk_plug plug; |
240799cd TT |
4266 | int inodes_per_block, inode_offset; |
4267 | ||
3a06d778 | 4268 | iloc->bh = NULL; |
8016e29f HS |
4269 | if (ino < EXT4_ROOT_INO || |
4270 | ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count)) | |
6a797d27 | 4271 | return -EFSCORRUPTED; |
ac27a0ec | 4272 | |
8016e29f | 4273 | iloc->block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb); |
240799cd TT |
4274 | gdp = ext4_get_group_desc(sb, iloc->block_group, NULL); |
4275 | if (!gdp) | |
ac27a0ec DK |
4276 | return -EIO; |
4277 | ||
240799cd TT |
4278 | /* |
4279 | * Figure out the offset within the block group inode table | |
4280 | */ | |
00d09882 | 4281 | inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; |
8016e29f | 4282 | inode_offset = ((ino - 1) % |
240799cd TT |
4283 | EXT4_INODES_PER_GROUP(sb)); |
4284 | block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block); | |
4285 | iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb); | |
4286 | ||
4287 | bh = sb_getblk(sb, block); | |
aebf0243 | 4288 | if (unlikely(!bh)) |
860d21e2 | 4289 | return -ENOMEM; |
8e33fadf ZY |
4290 | if (ext4_buffer_uptodate(bh)) |
4291 | goto has_buffer; | |
9c83a923 | 4292 | |
8e33fadf | 4293 | lock_buffer(bh); |
f2c77973 ZY |
4294 | if (ext4_buffer_uptodate(bh)) { |
4295 | /* Someone brought it uptodate while we waited */ | |
4296 | unlock_buffer(bh); | |
4297 | goto has_buffer; | |
4298 | } | |
4299 | ||
8e33fadf ZY |
4300 | /* |
4301 | * If we have all information of the inode in memory and this | |
4302 | * is the only valid inode in the block, we need not read the | |
4303 | * block. | |
4304 | */ | |
4305 | if (in_mem) { | |
4306 | struct buffer_head *bitmap_bh; | |
4307 | int i, start; | |
ac27a0ec | 4308 | |
8e33fadf | 4309 | start = inode_offset & ~(inodes_per_block - 1); |
ac27a0ec | 4310 | |
8e33fadf ZY |
4311 | /* Is the inode bitmap in cache? */ |
4312 | bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp)); | |
4313 | if (unlikely(!bitmap_bh)) | |
4314 | goto make_io; | |
ac27a0ec | 4315 | |
8e33fadf ZY |
4316 | /* |
4317 | * If the inode bitmap isn't in cache then the | |
4318 | * optimisation may end up performing two reads instead | |
4319 | * of one, so skip it. | |
4320 | */ | |
4321 | if (!buffer_uptodate(bitmap_bh)) { | |
ac27a0ec | 4322 | brelse(bitmap_bh); |
8e33fadf | 4323 | goto make_io; |
ac27a0ec | 4324 | } |
8e33fadf ZY |
4325 | for (i = start; i < start + inodes_per_block; i++) { |
4326 | if (i == inode_offset) | |
4327 | continue; | |
4328 | if (ext4_test_bit(i, bitmap_bh->b_data)) | |
4329 | break; | |
ac27a0ec | 4330 | } |
8e33fadf ZY |
4331 | brelse(bitmap_bh); |
4332 | if (i == start + inodes_per_block) { | |
4333 | /* all other inodes are free, so skip I/O */ | |
4334 | memset(bh->b_data, 0, bh->b_size); | |
4335 | set_buffer_uptodate(bh); | |
4336 | unlock_buffer(bh); | |
4337 | goto has_buffer; | |
4338 | } | |
4339 | } | |
ac27a0ec DK |
4340 | |
4341 | make_io: | |
8e33fadf ZY |
4342 | /* |
4343 | * If we need to do any I/O, try to pre-readahead extra | |
4344 | * blocks from the inode table. | |
4345 | */ | |
4346 | blk_start_plug(&plug); | |
4347 | if (EXT4_SB(sb)->s_inode_readahead_blks) { | |
4348 | ext4_fsblk_t b, end, table; | |
4349 | unsigned num; | |
4350 | __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks; | |
4351 | ||
4352 | table = ext4_inode_table(sb, gdp); | |
4353 | /* s_inode_readahead_blks is always a power of 2 */ | |
4354 | b = block & ~((ext4_fsblk_t) ra_blks - 1); | |
4355 | if (table > b) | |
4356 | b = table; | |
4357 | end = b + ra_blks; | |
4358 | num = EXT4_INODES_PER_GROUP(sb); | |
4359 | if (ext4_has_group_desc_csum(sb)) | |
4360 | num -= ext4_itable_unused_count(sb, gdp); | |
4361 | table += num / inodes_per_block; | |
4362 | if (end > table) | |
4363 | end = table; | |
4364 | while (b <= end) | |
4365 | ext4_sb_breadahead_unmovable(sb, b++); | |
4366 | } | |
240799cd | 4367 | |
8e33fadf ZY |
4368 | /* |
4369 | * There are other valid inodes in the buffer, this inode | |
4370 | * has in-inode xattrs, or we don't have this inode in memory. | |
4371 | * Read the block from disk. | |
4372 | */ | |
4373 | trace_ext4_load_inode(sb, ino); | |
4374 | ext4_read_bh_nowait(bh, REQ_META | REQ_PRIO, NULL); | |
4375 | blk_finish_plug(&plug); | |
4376 | wait_on_buffer(bh); | |
4377 | ext4_simulate_fail_bh(sb, bh, EXT4_SIM_INODE_EIO); | |
4378 | if (!buffer_uptodate(bh)) { | |
4379 | if (ret_block) | |
4380 | *ret_block = block; | |
4381 | brelse(bh); | |
4382 | return -EIO; | |
ac27a0ec DK |
4383 | } |
4384 | has_buffer: | |
4385 | iloc->bh = bh; | |
4386 | return 0; | |
4387 | } | |
4388 | ||
8016e29f HS |
4389 | static int __ext4_get_inode_loc_noinmem(struct inode *inode, |
4390 | struct ext4_iloc *iloc) | |
4391 | { | |
36108e9e | 4392 | ext4_fsblk_t err_blk = 0; |
8016e29f HS |
4393 | int ret; |
4394 | ||
4395 | ret = __ext4_get_inode_loc(inode->i_sb, inode->i_ino, iloc, 0, | |
4396 | &err_blk); | |
4397 | ||
4398 | if (ret == -EIO) | |
4399 | ext4_error_inode_block(inode, err_blk, EIO, | |
4400 | "unable to read itable block"); | |
4401 | ||
4402 | return ret; | |
4403 | } | |
4404 | ||
617ba13b | 4405 | int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec | 4406 | { |
36108e9e | 4407 | ext4_fsblk_t err_blk = 0; |
8016e29f HS |
4408 | int ret; |
4409 | ||
ac27a0ec | 4410 | /* We have all inode data except xattrs in memory here. */ |
8016e29f HS |
4411 | ret = __ext4_get_inode_loc(inode->i_sb, inode->i_ino, iloc, |
4412 | !ext4_test_inode_state(inode, EXT4_STATE_XATTR), &err_blk); | |
4413 | ||
4414 | if (ret == -EIO) | |
4415 | ext4_error_inode_block(inode, err_blk, EIO, | |
4416 | "unable to read itable block"); | |
4417 | ||
4418 | return ret; | |
4419 | } | |
4420 | ||
4421 | ||
4422 | int ext4_get_fc_inode_loc(struct super_block *sb, unsigned long ino, | |
4423 | struct ext4_iloc *iloc) | |
4424 | { | |
4425 | return __ext4_get_inode_loc(sb, ino, iloc, 0, NULL); | |
ac27a0ec DK |
4426 | } |
4427 | ||
a8ab6d38 | 4428 | static bool ext4_should_enable_dax(struct inode *inode) |
6642586b | 4429 | { |
a8ab6d38 IW |
4430 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
4431 | ||
9cb20f94 | 4432 | if (test_opt2(inode->i_sb, DAX_NEVER)) |
6642586b RZ |
4433 | return false; |
4434 | if (!S_ISREG(inode->i_mode)) | |
4435 | return false; | |
4436 | if (ext4_should_journal_data(inode)) | |
4437 | return false; | |
4438 | if (ext4_has_inline_data(inode)) | |
4439 | return false; | |
592ddec7 | 4440 | if (ext4_test_inode_flag(inode, EXT4_INODE_ENCRYPT)) |
6642586b | 4441 | return false; |
c93d8f88 EB |
4442 | if (ext4_test_inode_flag(inode, EXT4_INODE_VERITY)) |
4443 | return false; | |
a8ab6d38 IW |
4444 | if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags)) |
4445 | return false; | |
4446 | if (test_opt(inode->i_sb, DAX_ALWAYS)) | |
4447 | return true; | |
4448 | ||
b383a73f | 4449 | return ext4_test_inode_flag(inode, EXT4_INODE_DAX); |
6642586b RZ |
4450 | } |
4451 | ||
043546e4 | 4452 | void ext4_set_inode_flags(struct inode *inode, bool init) |
ac27a0ec | 4453 | { |
617ba13b | 4454 | unsigned int flags = EXT4_I(inode)->i_flags; |
00a1a053 | 4455 | unsigned int new_fl = 0; |
ac27a0ec | 4456 | |
043546e4 IW |
4457 | WARN_ON_ONCE(IS_DAX(inode) && init); |
4458 | ||
617ba13b | 4459 | if (flags & EXT4_SYNC_FL) |
00a1a053 | 4460 | new_fl |= S_SYNC; |
617ba13b | 4461 | if (flags & EXT4_APPEND_FL) |
00a1a053 | 4462 | new_fl |= S_APPEND; |
617ba13b | 4463 | if (flags & EXT4_IMMUTABLE_FL) |
00a1a053 | 4464 | new_fl |= S_IMMUTABLE; |
617ba13b | 4465 | if (flags & EXT4_NOATIME_FL) |
00a1a053 | 4466 | new_fl |= S_NOATIME; |
617ba13b | 4467 | if (flags & EXT4_DIRSYNC_FL) |
00a1a053 | 4468 | new_fl |= S_DIRSYNC; |
043546e4 IW |
4469 | |
4470 | /* Because of the way inode_set_flags() works we must preserve S_DAX | |
4471 | * here if already set. */ | |
4472 | new_fl |= (inode->i_flags & S_DAX); | |
4473 | if (init && ext4_should_enable_dax(inode)) | |
923ae0ff | 4474 | new_fl |= S_DAX; |
043546e4 | 4475 | |
2ee6a576 EB |
4476 | if (flags & EXT4_ENCRYPT_FL) |
4477 | new_fl |= S_ENCRYPTED; | |
b886ee3e GKB |
4478 | if (flags & EXT4_CASEFOLD_FL) |
4479 | new_fl |= S_CASEFOLD; | |
c93d8f88 EB |
4480 | if (flags & EXT4_VERITY_FL) |
4481 | new_fl |= S_VERITY; | |
5f16f322 | 4482 | inode_set_flags(inode, new_fl, |
2ee6a576 | 4483 | S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX| |
c93d8f88 | 4484 | S_ENCRYPTED|S_CASEFOLD|S_VERITY); |
ac27a0ec DK |
4485 | } |
4486 | ||
0fc1b451 | 4487 | static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode, |
de9a55b8 | 4488 | struct ext4_inode_info *ei) |
0fc1b451 AK |
4489 | { |
4490 | blkcnt_t i_blocks ; | |
8180a562 AK |
4491 | struct inode *inode = &(ei->vfs_inode); |
4492 | struct super_block *sb = inode->i_sb; | |
0fc1b451 | 4493 | |
e2b911c5 | 4494 | if (ext4_has_feature_huge_file(sb)) { |
0fc1b451 AK |
4495 | /* we are using combined 48 bit field */ |
4496 | i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 | | |
4497 | le32_to_cpu(raw_inode->i_blocks_lo); | |
07a03824 | 4498 | if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) { |
8180a562 AK |
4499 | /* i_blocks represent file system block size */ |
4500 | return i_blocks << (inode->i_blkbits - 9); | |
4501 | } else { | |
4502 | return i_blocks; | |
4503 | } | |
0fc1b451 AK |
4504 | } else { |
4505 | return le32_to_cpu(raw_inode->i_blocks_lo); | |
4506 | } | |
4507 | } | |
ff9ddf7e | 4508 | |
eb9b5f01 | 4509 | static inline int ext4_iget_extra_inode(struct inode *inode, |
152a7b0a TM |
4510 | struct ext4_inode *raw_inode, |
4511 | struct ext4_inode_info *ei) | |
4512 | { | |
4513 | __le32 *magic = (void *)raw_inode + | |
4514 | EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize; | |
eb9b5f01 | 4515 | |
290ab230 EB |
4516 | if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize + sizeof(__le32) <= |
4517 | EXT4_INODE_SIZE(inode->i_sb) && | |
4518 | *magic == cpu_to_le32(EXT4_XATTR_MAGIC)) { | |
152a7b0a | 4519 | ext4_set_inode_state(inode, EXT4_STATE_XATTR); |
eb9b5f01 | 4520 | return ext4_find_inline_data_nolock(inode); |
f19d5870 TM |
4521 | } else |
4522 | EXT4_I(inode)->i_inline_off = 0; | |
eb9b5f01 | 4523 | return 0; |
152a7b0a TM |
4524 | } |
4525 | ||
040cb378 LX |
4526 | int ext4_get_projid(struct inode *inode, kprojid_t *projid) |
4527 | { | |
0b7b7779 | 4528 | if (!ext4_has_feature_project(inode->i_sb)) |
040cb378 LX |
4529 | return -EOPNOTSUPP; |
4530 | *projid = EXT4_I(inode)->i_projid; | |
4531 | return 0; | |
4532 | } | |
4533 | ||
e254d1af EG |
4534 | /* |
4535 | * ext4 has self-managed i_version for ea inodes, it stores the lower 32bit of | |
4536 | * refcount in i_version, so use raw values if inode has EXT4_EA_INODE_FL flag | |
4537 | * set. | |
4538 | */ | |
4539 | static inline void ext4_inode_set_iversion_queried(struct inode *inode, u64 val) | |
4540 | { | |
4541 | if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) | |
4542 | inode_set_iversion_raw(inode, val); | |
4543 | else | |
4544 | inode_set_iversion_queried(inode, val); | |
4545 | } | |
4546 | static inline u64 ext4_inode_peek_iversion(const struct inode *inode) | |
4547 | { | |
4548 | if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) | |
4549 | return inode_peek_iversion_raw(inode); | |
4550 | else | |
4551 | return inode_peek_iversion(inode); | |
4552 | } | |
4553 | ||
8a363970 TT |
4554 | struct inode *__ext4_iget(struct super_block *sb, unsigned long ino, |
4555 | ext4_iget_flags flags, const char *function, | |
4556 | unsigned int line) | |
ac27a0ec | 4557 | { |
617ba13b MC |
4558 | struct ext4_iloc iloc; |
4559 | struct ext4_inode *raw_inode; | |
1d1fe1ee | 4560 | struct ext4_inode_info *ei; |
bd2c38cf | 4561 | struct ext4_super_block *es = EXT4_SB(sb)->s_es; |
1d1fe1ee | 4562 | struct inode *inode; |
b436b9be | 4563 | journal_t *journal = EXT4_SB(sb)->s_journal; |
1d1fe1ee | 4564 | long ret; |
7e6e1ef4 | 4565 | loff_t size; |
ac27a0ec | 4566 | int block; |
08cefc7a EB |
4567 | uid_t i_uid; |
4568 | gid_t i_gid; | |
040cb378 | 4569 | projid_t i_projid; |
ac27a0ec | 4570 | |
191ce178 | 4571 | if ((!(flags & EXT4_IGET_SPECIAL) && |
bd2c38cf JK |
4572 | ((ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO) || |
4573 | ino == le32_to_cpu(es->s_usr_quota_inum) || | |
4574 | ino == le32_to_cpu(es->s_grp_quota_inum) || | |
02f310fc JK |
4575 | ino == le32_to_cpu(es->s_prj_quota_inum) || |
4576 | ino == le32_to_cpu(es->s_orphan_file_inum))) || | |
8a363970 | 4577 | (ino < EXT4_ROOT_INO) || |
bd2c38cf | 4578 | (ino > le32_to_cpu(es->s_inodes_count))) { |
8a363970 TT |
4579 | if (flags & EXT4_IGET_HANDLE) |
4580 | return ERR_PTR(-ESTALE); | |
014c9caa | 4581 | __ext4_error(sb, function, line, false, EFSCORRUPTED, 0, |
8a363970 TT |
4582 | "inode #%lu: comm %s: iget: illegal inode #", |
4583 | ino, current->comm); | |
4584 | return ERR_PTR(-EFSCORRUPTED); | |
4585 | } | |
4586 | ||
1d1fe1ee DH |
4587 | inode = iget_locked(sb, ino); |
4588 | if (!inode) | |
4589 | return ERR_PTR(-ENOMEM); | |
4590 | if (!(inode->i_state & I_NEW)) | |
4591 | return inode; | |
4592 | ||
4593 | ei = EXT4_I(inode); | |
7dc57615 | 4594 | iloc.bh = NULL; |
ac27a0ec | 4595 | |
8016e29f | 4596 | ret = __ext4_get_inode_loc_noinmem(inode, &iloc); |
1d1fe1ee | 4597 | if (ret < 0) |
ac27a0ec | 4598 | goto bad_inode; |
617ba13b | 4599 | raw_inode = ext4_raw_inode(&iloc); |
814525f4 | 4600 | |
8e4b5eae | 4601 | if ((ino == EXT4_ROOT_INO) && (raw_inode->i_links_count == 0)) { |
8a363970 TT |
4602 | ext4_error_inode(inode, function, line, 0, |
4603 | "iget: root inode unallocated"); | |
8e4b5eae TT |
4604 | ret = -EFSCORRUPTED; |
4605 | goto bad_inode; | |
4606 | } | |
4607 | ||
8a363970 TT |
4608 | if ((flags & EXT4_IGET_HANDLE) && |
4609 | (raw_inode->i_links_count == 0) && (raw_inode->i_mode == 0)) { | |
4610 | ret = -ESTALE; | |
4611 | goto bad_inode; | |
4612 | } | |
4613 | ||
814525f4 DW |
4614 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
4615 | ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize); | |
4616 | if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > | |
2dc8d9e1 EB |
4617 | EXT4_INODE_SIZE(inode->i_sb) || |
4618 | (ei->i_extra_isize & 3)) { | |
8a363970 TT |
4619 | ext4_error_inode(inode, function, line, 0, |
4620 | "iget: bad extra_isize %u " | |
4621 | "(inode size %u)", | |
2dc8d9e1 EB |
4622 | ei->i_extra_isize, |
4623 | EXT4_INODE_SIZE(inode->i_sb)); | |
6a797d27 | 4624 | ret = -EFSCORRUPTED; |
814525f4 DW |
4625 | goto bad_inode; |
4626 | } | |
4627 | } else | |
4628 | ei->i_extra_isize = 0; | |
4629 | ||
4630 | /* Precompute checksum seed for inode metadata */ | |
9aa5d32b | 4631 | if (ext4_has_metadata_csum(sb)) { |
814525f4 DW |
4632 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
4633 | __u32 csum; | |
4634 | __le32 inum = cpu_to_le32(inode->i_ino); | |
4635 | __le32 gen = raw_inode->i_generation; | |
4636 | csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum, | |
4637 | sizeof(inum)); | |
4638 | ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen, | |
4639 | sizeof(gen)); | |
4640 | } | |
4641 | ||
8016e29f HS |
4642 | if ((!ext4_inode_csum_verify(inode, raw_inode, ei) || |
4643 | ext4_simulate_fail(sb, EXT4_SIM_INODE_CRC)) && | |
4644 | (!(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY))) { | |
4645 | ext4_error_inode_err(inode, function, line, 0, | |
4646 | EFSBADCRC, "iget: checksum invalid"); | |
6a797d27 | 4647 | ret = -EFSBADCRC; |
814525f4 DW |
4648 | goto bad_inode; |
4649 | } | |
4650 | ||
ac27a0ec | 4651 | inode->i_mode = le16_to_cpu(raw_inode->i_mode); |
08cefc7a EB |
4652 | i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low); |
4653 | i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low); | |
0b7b7779 | 4654 | if (ext4_has_feature_project(sb) && |
040cb378 LX |
4655 | EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE && |
4656 | EXT4_FITS_IN_INODE(raw_inode, ei, i_projid)) | |
4657 | i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid); | |
4658 | else | |
4659 | i_projid = EXT4_DEF_PROJID; | |
4660 | ||
af5bc92d | 4661 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
08cefc7a EB |
4662 | i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16; |
4663 | i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16; | |
ac27a0ec | 4664 | } |
08cefc7a EB |
4665 | i_uid_write(inode, i_uid); |
4666 | i_gid_write(inode, i_gid); | |
040cb378 | 4667 | ei->i_projid = make_kprojid(&init_user_ns, i_projid); |
bfe86848 | 4668 | set_nlink(inode, le16_to_cpu(raw_inode->i_links_count)); |
ac27a0ec | 4669 | |
353eb83c | 4670 | ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */ |
67cf5b09 | 4671 | ei->i_inline_off = 0; |
ac27a0ec DK |
4672 | ei->i_dir_start_lookup = 0; |
4673 | ei->i_dtime = le32_to_cpu(raw_inode->i_dtime); | |
4674 | /* We now have enough fields to check if the inode was active or not. | |
4675 | * This is needed because nfsd might try to access dead inodes | |
4676 | * the test is that same one that e2fsck uses | |
4677 | * NeilBrown 1999oct15 | |
4678 | */ | |
4679 | if (inode->i_nlink == 0) { | |
393d1d1d DTB |
4680 | if ((inode->i_mode == 0 || |
4681 | !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) && | |
4682 | ino != EXT4_BOOT_LOADER_INO) { | |
ac27a0ec | 4683 | /* this inode is deleted */ |
1d1fe1ee | 4684 | ret = -ESTALE; |
ac27a0ec DK |
4685 | goto bad_inode; |
4686 | } | |
4687 | /* The only unlinked inodes we let through here have | |
4688 | * valid i_mode and are being read by the orphan | |
4689 | * recovery code: that's fine, we're about to complete | |
393d1d1d DTB |
4690 | * the process of deleting those. |
4691 | * OR it is the EXT4_BOOT_LOADER_INO which is | |
4692 | * not initialized on a new filesystem. */ | |
ac27a0ec | 4693 | } |
ac27a0ec | 4694 | ei->i_flags = le32_to_cpu(raw_inode->i_flags); |
043546e4 | 4695 | ext4_set_inode_flags(inode, true); |
0fc1b451 | 4696 | inode->i_blocks = ext4_inode_blocks(raw_inode, ei); |
7973c0c1 | 4697 | ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo); |
e2b911c5 | 4698 | if (ext4_has_feature_64bit(sb)) |
a1ddeb7e BP |
4699 | ei->i_file_acl |= |
4700 | ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32; | |
e08ac99f | 4701 | inode->i_size = ext4_isize(sb, raw_inode); |
7e6e1ef4 | 4702 | if ((size = i_size_read(inode)) < 0) { |
8a363970 TT |
4703 | ext4_error_inode(inode, function, line, 0, |
4704 | "iget: bad i_size value: %lld", size); | |
7e6e1ef4 DW |
4705 | ret = -EFSCORRUPTED; |
4706 | goto bad_inode; | |
4707 | } | |
48a34311 JK |
4708 | /* |
4709 | * If dir_index is not enabled but there's dir with INDEX flag set, | |
4710 | * we'd normally treat htree data as empty space. But with metadata | |
4711 | * checksumming that corrupts checksums so forbid that. | |
4712 | */ | |
4713 | if (!ext4_has_feature_dir_index(sb) && ext4_has_metadata_csum(sb) && | |
4714 | ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) { | |
4715 | ext4_error_inode(inode, function, line, 0, | |
4716 | "iget: Dir with htree data on filesystem without dir_index feature."); | |
4717 | ret = -EFSCORRUPTED; | |
4718 | goto bad_inode; | |
4719 | } | |
ac27a0ec | 4720 | ei->i_disksize = inode->i_size; |
a9e7f447 DM |
4721 | #ifdef CONFIG_QUOTA |
4722 | ei->i_reserved_quota = 0; | |
4723 | #endif | |
ac27a0ec DK |
4724 | inode->i_generation = le32_to_cpu(raw_inode->i_generation); |
4725 | ei->i_block_group = iloc.block_group; | |
a4912123 | 4726 | ei->i_last_alloc_group = ~0; |
ac27a0ec DK |
4727 | /* |
4728 | * NOTE! The in-memory inode i_data array is in little-endian order | |
4729 | * even on big-endian machines: we do NOT byteswap the block numbers! | |
4730 | */ | |
617ba13b | 4731 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
ac27a0ec DK |
4732 | ei->i_data[block] = raw_inode->i_block[block]; |
4733 | INIT_LIST_HEAD(&ei->i_orphan); | |
aa75f4d3 | 4734 | ext4_fc_init_inode(&ei->vfs_inode); |
ac27a0ec | 4735 | |
b436b9be JK |
4736 | /* |
4737 | * Set transaction id's of transactions that have to be committed | |
4738 | * to finish f[data]sync. We set them to currently running transaction | |
4739 | * as we cannot be sure that the inode or some of its metadata isn't | |
4740 | * part of the transaction - the inode could have been reclaimed and | |
4741 | * now it is reread from disk. | |
4742 | */ | |
4743 | if (journal) { | |
4744 | transaction_t *transaction; | |
4745 | tid_t tid; | |
4746 | ||
a931da6a | 4747 | read_lock(&journal->j_state_lock); |
b436b9be JK |
4748 | if (journal->j_running_transaction) |
4749 | transaction = journal->j_running_transaction; | |
4750 | else | |
4751 | transaction = journal->j_committing_transaction; | |
4752 | if (transaction) | |
4753 | tid = transaction->t_tid; | |
4754 | else | |
4755 | tid = journal->j_commit_sequence; | |
a931da6a | 4756 | read_unlock(&journal->j_state_lock); |
b436b9be JK |
4757 | ei->i_sync_tid = tid; |
4758 | ei->i_datasync_tid = tid; | |
4759 | } | |
4760 | ||
0040d987 | 4761 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
ac27a0ec DK |
4762 | if (ei->i_extra_isize == 0) { |
4763 | /* The extra space is currently unused. Use it. */ | |
2dc8d9e1 | 4764 | BUILD_BUG_ON(sizeof(struct ext4_inode) & 3); |
617ba13b MC |
4765 | ei->i_extra_isize = sizeof(struct ext4_inode) - |
4766 | EXT4_GOOD_OLD_INODE_SIZE; | |
ac27a0ec | 4767 | } else { |
eb9b5f01 TT |
4768 | ret = ext4_iget_extra_inode(inode, raw_inode, ei); |
4769 | if (ret) | |
4770 | goto bad_inode; | |
ac27a0ec | 4771 | } |
814525f4 | 4772 | } |
ac27a0ec | 4773 | |
ef7f3835 KS |
4774 | EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode); |
4775 | EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode); | |
4776 | EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode); | |
4777 | EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode); | |
4778 | ||
ed3654eb | 4779 | if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) { |
ee73f9a5 JL |
4780 | u64 ivers = le32_to_cpu(raw_inode->i_disk_version); |
4781 | ||
c4f65706 TT |
4782 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
4783 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
ee73f9a5 | 4784 | ivers |= |
c4f65706 TT |
4785 | (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32; |
4786 | } | |
e254d1af | 4787 | ext4_inode_set_iversion_queried(inode, ivers); |
25ec56b5 JNC |
4788 | } |
4789 | ||
c4b5a614 | 4790 | ret = 0; |
485c26ec | 4791 | if (ei->i_file_acl && |
ce9f24cc | 4792 | !ext4_inode_block_valid(inode, ei->i_file_acl, 1)) { |
8a363970 TT |
4793 | ext4_error_inode(inode, function, line, 0, |
4794 | "iget: bad extended attribute block %llu", | |
24676da4 | 4795 | ei->i_file_acl); |
6a797d27 | 4796 | ret = -EFSCORRUPTED; |
485c26ec | 4797 | goto bad_inode; |
f19d5870 | 4798 | } else if (!ext4_has_inline_data(inode)) { |
bc716523 | 4799 | /* validate the block references in the inode */ |
8016e29f HS |
4800 | if (!(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY) && |
4801 | (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || | |
4802 | (S_ISLNK(inode->i_mode) && | |
4803 | !ext4_inode_is_fast_symlink(inode)))) { | |
bc716523 | 4804 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
f19d5870 | 4805 | ret = ext4_ext_check_inode(inode); |
bc716523 LS |
4806 | else |
4807 | ret = ext4_ind_check_inode(inode); | |
f19d5870 | 4808 | } |
fe2c8191 | 4809 | } |
567f3e9a | 4810 | if (ret) |
de9a55b8 | 4811 | goto bad_inode; |
7a262f7c | 4812 | |
ac27a0ec | 4813 | if (S_ISREG(inode->i_mode)) { |
617ba13b | 4814 | inode->i_op = &ext4_file_inode_operations; |
be64f884 | 4815 | inode->i_fop = &ext4_file_operations; |
617ba13b | 4816 | ext4_set_aops(inode); |
ac27a0ec | 4817 | } else if (S_ISDIR(inode->i_mode)) { |
617ba13b MC |
4818 | inode->i_op = &ext4_dir_inode_operations; |
4819 | inode->i_fop = &ext4_dir_operations; | |
ac27a0ec | 4820 | } else if (S_ISLNK(inode->i_mode)) { |
6390d33b LR |
4821 | /* VFS does not allow setting these so must be corruption */ |
4822 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) { | |
8a363970 TT |
4823 | ext4_error_inode(inode, function, line, 0, |
4824 | "iget: immutable or append flags " | |
4825 | "not allowed on symlinks"); | |
6390d33b LR |
4826 | ret = -EFSCORRUPTED; |
4827 | goto bad_inode; | |
4828 | } | |
592ddec7 | 4829 | if (IS_ENCRYPTED(inode)) { |
a7a67e8a AV |
4830 | inode->i_op = &ext4_encrypted_symlink_inode_operations; |
4831 | ext4_set_aops(inode); | |
4832 | } else if (ext4_inode_is_fast_symlink(inode)) { | |
75e7566b | 4833 | inode->i_link = (char *)ei->i_data; |
617ba13b | 4834 | inode->i_op = &ext4_fast_symlink_inode_operations; |
e83c1397 DG |
4835 | nd_terminate_link(ei->i_data, inode->i_size, |
4836 | sizeof(ei->i_data) - 1); | |
4837 | } else { | |
617ba13b MC |
4838 | inode->i_op = &ext4_symlink_inode_operations; |
4839 | ext4_set_aops(inode); | |
ac27a0ec | 4840 | } |
21fc61c7 | 4841 | inode_nohighmem(inode); |
563bdd61 TT |
4842 | } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) || |
4843 | S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) { | |
617ba13b | 4844 | inode->i_op = &ext4_special_inode_operations; |
ac27a0ec DK |
4845 | if (raw_inode->i_block[0]) |
4846 | init_special_inode(inode, inode->i_mode, | |
4847 | old_decode_dev(le32_to_cpu(raw_inode->i_block[0]))); | |
4848 | else | |
4849 | init_special_inode(inode, inode->i_mode, | |
4850 | new_decode_dev(le32_to_cpu(raw_inode->i_block[1]))); | |
393d1d1d DTB |
4851 | } else if (ino == EXT4_BOOT_LOADER_INO) { |
4852 | make_bad_inode(inode); | |
563bdd61 | 4853 | } else { |
6a797d27 | 4854 | ret = -EFSCORRUPTED; |
8a363970 TT |
4855 | ext4_error_inode(inode, function, line, 0, |
4856 | "iget: bogus i_mode (%o)", inode->i_mode); | |
563bdd61 | 4857 | goto bad_inode; |
ac27a0ec | 4858 | } |
6456ca65 TT |
4859 | if (IS_CASEFOLDED(inode) && !ext4_has_feature_casefold(inode->i_sb)) |
4860 | ext4_error_inode(inode, function, line, 0, | |
4861 | "casefold flag without casefold feature"); | |
af5bc92d | 4862 | brelse(iloc.bh); |
dec214d0 | 4863 | |
1d1fe1ee DH |
4864 | unlock_new_inode(inode); |
4865 | return inode; | |
ac27a0ec DK |
4866 | |
4867 | bad_inode: | |
567f3e9a | 4868 | brelse(iloc.bh); |
1d1fe1ee DH |
4869 | iget_failed(inode); |
4870 | return ERR_PTR(ret); | |
ac27a0ec DK |
4871 | } |
4872 | ||
0fc1b451 AK |
4873 | static int ext4_inode_blocks_set(handle_t *handle, |
4874 | struct ext4_inode *raw_inode, | |
4875 | struct ext4_inode_info *ei) | |
4876 | { | |
4877 | struct inode *inode = &(ei->vfs_inode); | |
28936b62 | 4878 | u64 i_blocks = READ_ONCE(inode->i_blocks); |
0fc1b451 | 4879 | struct super_block *sb = inode->i_sb; |
0fc1b451 AK |
4880 | |
4881 | if (i_blocks <= ~0U) { | |
4882 | /* | |
4907cb7b | 4883 | * i_blocks can be represented in a 32 bit variable |
0fc1b451 AK |
4884 | * as multiple of 512 bytes |
4885 | */ | |
8180a562 | 4886 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 4887 | raw_inode->i_blocks_high = 0; |
84a8dce2 | 4888 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
f287a1a5 TT |
4889 | return 0; |
4890 | } | |
baaae979 ZY |
4891 | |
4892 | /* | |
4893 | * This should never happen since sb->s_maxbytes should not have | |
4894 | * allowed this, sb->s_maxbytes was set according to the huge_file | |
4895 | * feature in ext4_fill_super(). | |
4896 | */ | |
e2b911c5 | 4897 | if (!ext4_has_feature_huge_file(sb)) |
baaae979 | 4898 | return -EFSCORRUPTED; |
f287a1a5 TT |
4899 | |
4900 | if (i_blocks <= 0xffffffffffffULL) { | |
0fc1b451 AK |
4901 | /* |
4902 | * i_blocks can be represented in a 48 bit variable | |
4903 | * as multiple of 512 bytes | |
4904 | */ | |
8180a562 | 4905 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 4906 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); |
84a8dce2 | 4907 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
0fc1b451 | 4908 | } else { |
84a8dce2 | 4909 | ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
8180a562 AK |
4910 | /* i_block is stored in file system block size */ |
4911 | i_blocks = i_blocks >> (inode->i_blkbits - 9); | |
4912 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); | |
4913 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); | |
0fc1b451 | 4914 | } |
f287a1a5 | 4915 | return 0; |
0fc1b451 AK |
4916 | } |
4917 | ||
3f19b2ab DH |
4918 | static void __ext4_update_other_inode_time(struct super_block *sb, |
4919 | unsigned long orig_ino, | |
4920 | unsigned long ino, | |
4921 | struct ext4_inode *raw_inode) | |
a26f4992 | 4922 | { |
3f19b2ab DH |
4923 | struct inode *inode; |
4924 | ||
4925 | inode = find_inode_by_ino_rcu(sb, ino); | |
4926 | if (!inode) | |
4927 | return; | |
a26f4992 | 4928 | |
ed296c6c | 4929 | if (!inode_is_dirtytime_only(inode)) |
3f19b2ab DH |
4930 | return; |
4931 | ||
a26f4992 | 4932 | spin_lock(&inode->i_lock); |
ed296c6c | 4933 | if (inode_is_dirtytime_only(inode)) { |
a26f4992 TT |
4934 | struct ext4_inode_info *ei = EXT4_I(inode); |
4935 | ||
5fcd5750 | 4936 | inode->i_state &= ~I_DIRTY_TIME; |
a26f4992 TT |
4937 | spin_unlock(&inode->i_lock); |
4938 | ||
4939 | spin_lock(&ei->i_raw_lock); | |
3f19b2ab DH |
4940 | EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode); |
4941 | EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode); | |
4942 | EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode); | |
4943 | ext4_inode_csum_set(inode, raw_inode, ei); | |
a26f4992 | 4944 | spin_unlock(&ei->i_raw_lock); |
3f19b2ab DH |
4945 | trace_ext4_other_inode_update_time(inode, orig_ino); |
4946 | return; | |
a26f4992 TT |
4947 | } |
4948 | spin_unlock(&inode->i_lock); | |
a26f4992 TT |
4949 | } |
4950 | ||
4951 | /* | |
4952 | * Opportunistically update the other time fields for other inodes in | |
4953 | * the same inode table block. | |
4954 | */ | |
4955 | static void ext4_update_other_inodes_time(struct super_block *sb, | |
4956 | unsigned long orig_ino, char *buf) | |
4957 | { | |
a26f4992 TT |
4958 | unsigned long ino; |
4959 | int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; | |
4960 | int inode_size = EXT4_INODE_SIZE(sb); | |
4961 | ||
0f0ff9a9 TT |
4962 | /* |
4963 | * Calculate the first inode in the inode table block. Inode | |
4964 | * numbers are one-based. That is, the first inode in a block | |
4965 | * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1). | |
4966 | */ | |
4967 | ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1; | |
3f19b2ab | 4968 | rcu_read_lock(); |
a26f4992 TT |
4969 | for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) { |
4970 | if (ino == orig_ino) | |
4971 | continue; | |
3f19b2ab DH |
4972 | __ext4_update_other_inode_time(sb, orig_ino, ino, |
4973 | (struct ext4_inode *)buf); | |
a26f4992 | 4974 | } |
3f19b2ab | 4975 | rcu_read_unlock(); |
a26f4992 TT |
4976 | } |
4977 | ||
ac27a0ec DK |
4978 | /* |
4979 | * Post the struct inode info into an on-disk inode location in the | |
4980 | * buffer-cache. This gobbles the caller's reference to the | |
4981 | * buffer_head in the inode location struct. | |
4982 | * | |
4983 | * The caller must have write access to iloc->bh. | |
4984 | */ | |
617ba13b | 4985 | static int ext4_do_update_inode(handle_t *handle, |
ac27a0ec | 4986 | struct inode *inode, |
830156c7 | 4987 | struct ext4_iloc *iloc) |
ac27a0ec | 4988 | { |
617ba13b MC |
4989 | struct ext4_inode *raw_inode = ext4_raw_inode(iloc); |
4990 | struct ext4_inode_info *ei = EXT4_I(inode); | |
ac27a0ec | 4991 | struct buffer_head *bh = iloc->bh; |
202ee5df | 4992 | struct super_block *sb = inode->i_sb; |
7d8bd3c7 | 4993 | int err = 0, block; |
202ee5df | 4994 | int need_datasync = 0, set_large_file = 0; |
08cefc7a EB |
4995 | uid_t i_uid; |
4996 | gid_t i_gid; | |
040cb378 | 4997 | projid_t i_projid; |
ac27a0ec | 4998 | |
202ee5df TT |
4999 | spin_lock(&ei->i_raw_lock); |
5000 | ||
baaae979 ZY |
5001 | /* |
5002 | * For fields not tracked in the in-memory inode, initialise them | |
5003 | * to zero for new inodes. | |
5004 | */ | |
19f5fb7a | 5005 | if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) |
617ba13b | 5006 | memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size); |
ac27a0ec | 5007 | |
13221811 | 5008 | err = ext4_inode_blocks_set(handle, raw_inode, ei); |
13221811 | 5009 | |
ac27a0ec | 5010 | raw_inode->i_mode = cpu_to_le16(inode->i_mode); |
08cefc7a EB |
5011 | i_uid = i_uid_read(inode); |
5012 | i_gid = i_gid_read(inode); | |
040cb378 | 5013 | i_projid = from_kprojid(&init_user_ns, ei->i_projid); |
af5bc92d | 5014 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
08cefc7a EB |
5015 | raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid)); |
5016 | raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid)); | |
baaae979 ZY |
5017 | /* |
5018 | * Fix up interoperability with old kernels. Otherwise, | |
5019 | * old inodes get re-used with the upper 16 bits of the | |
5020 | * uid/gid intact. | |
5021 | */ | |
93e3b4e6 DJ |
5022 | if (ei->i_dtime && list_empty(&ei->i_orphan)) { |
5023 | raw_inode->i_uid_high = 0; | |
5024 | raw_inode->i_gid_high = 0; | |
5025 | } else { | |
ac27a0ec | 5026 | raw_inode->i_uid_high = |
08cefc7a | 5027 | cpu_to_le16(high_16_bits(i_uid)); |
ac27a0ec | 5028 | raw_inode->i_gid_high = |
08cefc7a | 5029 | cpu_to_le16(high_16_bits(i_gid)); |
ac27a0ec DK |
5030 | } |
5031 | } else { | |
08cefc7a EB |
5032 | raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid)); |
5033 | raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid)); | |
ac27a0ec DK |
5034 | raw_inode->i_uid_high = 0; |
5035 | raw_inode->i_gid_high = 0; | |
5036 | } | |
5037 | raw_inode->i_links_count = cpu_to_le16(inode->i_nlink); | |
ef7f3835 KS |
5038 | |
5039 | EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode); | |
5040 | EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode); | |
5041 | EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode); | |
5042 | EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode); | |
5043 | ||
ac27a0ec | 5044 | raw_inode->i_dtime = cpu_to_le32(ei->i_dtime); |
353eb83c | 5045 | raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF); |
ed3654eb | 5046 | if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) |
a1ddeb7e BP |
5047 | raw_inode->i_file_acl_high = |
5048 | cpu_to_le16(ei->i_file_acl >> 32); | |
7973c0c1 | 5049 | raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl); |
dce8e237 | 5050 | if (READ_ONCE(ei->i_disksize) != ext4_isize(inode->i_sb, raw_inode)) { |
b71fc079 JK |
5051 | ext4_isize_set(raw_inode, ei->i_disksize); |
5052 | need_datasync = 1; | |
5053 | } | |
a48380f7 | 5054 | if (ei->i_disksize > 0x7fffffffULL) { |
e2b911c5 | 5055 | if (!ext4_has_feature_large_file(sb) || |
a48380f7 | 5056 | EXT4_SB(sb)->s_es->s_rev_level == |
202ee5df TT |
5057 | cpu_to_le32(EXT4_GOOD_OLD_REV)) |
5058 | set_large_file = 1; | |
ac27a0ec DK |
5059 | } |
5060 | raw_inode->i_generation = cpu_to_le32(inode->i_generation); | |
5061 | if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) { | |
5062 | if (old_valid_dev(inode->i_rdev)) { | |
5063 | raw_inode->i_block[0] = | |
5064 | cpu_to_le32(old_encode_dev(inode->i_rdev)); | |
5065 | raw_inode->i_block[1] = 0; | |
5066 | } else { | |
5067 | raw_inode->i_block[0] = 0; | |
5068 | raw_inode->i_block[1] = | |
5069 | cpu_to_le32(new_encode_dev(inode->i_rdev)); | |
5070 | raw_inode->i_block[2] = 0; | |
5071 | } | |
f19d5870 | 5072 | } else if (!ext4_has_inline_data(inode)) { |
de9a55b8 TT |
5073 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
5074 | raw_inode->i_block[block] = ei->i_data[block]; | |
f19d5870 | 5075 | } |
ac27a0ec | 5076 | |
ed3654eb | 5077 | if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) { |
e254d1af | 5078 | u64 ivers = ext4_inode_peek_iversion(inode); |
ee73f9a5 JL |
5079 | |
5080 | raw_inode->i_disk_version = cpu_to_le32(ivers); | |
c4f65706 TT |
5081 | if (ei->i_extra_isize) { |
5082 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
5083 | raw_inode->i_version_hi = | |
ee73f9a5 | 5084 | cpu_to_le32(ivers >> 32); |
c4f65706 TT |
5085 | raw_inode->i_extra_isize = |
5086 | cpu_to_le16(ei->i_extra_isize); | |
5087 | } | |
25ec56b5 | 5088 | } |
040cb378 | 5089 | |
baaae979 ZY |
5090 | if (i_projid != EXT4_DEF_PROJID && |
5091 | !ext4_has_feature_project(inode->i_sb)) | |
5092 | err = err ?: -EFSCORRUPTED; | |
040cb378 LX |
5093 | |
5094 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && | |
5095 | EXT4_FITS_IN_INODE(raw_inode, ei, i_projid)) | |
5096 | raw_inode->i_projid = cpu_to_le32(i_projid); | |
5097 | ||
814525f4 | 5098 | ext4_inode_csum_set(inode, raw_inode, ei); |
202ee5df | 5099 | spin_unlock(&ei->i_raw_lock); |
baaae979 ZY |
5100 | if (err) { |
5101 | EXT4_ERROR_INODE(inode, "corrupted inode contents"); | |
5102 | goto out_brelse; | |
5103 | } | |
5104 | ||
1751e8a6 | 5105 | if (inode->i_sb->s_flags & SB_LAZYTIME) |
a26f4992 TT |
5106 | ext4_update_other_inodes_time(inode->i_sb, inode->i_ino, |
5107 | bh->b_data); | |
202ee5df | 5108 | |
830156c7 | 5109 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
7d8bd3c7 SL |
5110 | err = ext4_handle_dirty_metadata(handle, NULL, bh); |
5111 | if (err) | |
baaae979 | 5112 | goto out_error; |
19f5fb7a | 5113 | ext4_clear_inode_state(inode, EXT4_STATE_NEW); |
202ee5df | 5114 | if (set_large_file) { |
5d601255 | 5115 | BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access"); |
188c299e JK |
5116 | err = ext4_journal_get_write_access(handle, sb, |
5117 | EXT4_SB(sb)->s_sbh, | |
5118 | EXT4_JTR_NONE); | |
202ee5df | 5119 | if (err) |
baaae979 | 5120 | goto out_error; |
05c2c00f | 5121 | lock_buffer(EXT4_SB(sb)->s_sbh); |
e2b911c5 | 5122 | ext4_set_feature_large_file(sb); |
05c2c00f JK |
5123 | ext4_superblock_csum_set(sb); |
5124 | unlock_buffer(EXT4_SB(sb)->s_sbh); | |
202ee5df | 5125 | ext4_handle_sync(handle); |
a3f5cf14 JK |
5126 | err = ext4_handle_dirty_metadata(handle, NULL, |
5127 | EXT4_SB(sb)->s_sbh); | |
202ee5df | 5128 | } |
b71fc079 | 5129 | ext4_update_inode_fsync_trans(handle, inode, need_datasync); |
baaae979 ZY |
5130 | out_error: |
5131 | ext4_std_error(inode->i_sb, err); | |
ac27a0ec | 5132 | out_brelse: |
af5bc92d | 5133 | brelse(bh); |
ac27a0ec DK |
5134 | return err; |
5135 | } | |
5136 | ||
5137 | /* | |
617ba13b | 5138 | * ext4_write_inode() |
ac27a0ec DK |
5139 | * |
5140 | * We are called from a few places: | |
5141 | * | |
87f7e416 | 5142 | * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files. |
ac27a0ec | 5143 | * Here, there will be no transaction running. We wait for any running |
4907cb7b | 5144 | * transaction to commit. |
ac27a0ec | 5145 | * |
87f7e416 TT |
5146 | * - Within flush work (sys_sync(), kupdate and such). |
5147 | * We wait on commit, if told to. | |
ac27a0ec | 5148 | * |
87f7e416 TT |
5149 | * - Within iput_final() -> write_inode_now() |
5150 | * We wait on commit, if told to. | |
ac27a0ec DK |
5151 | * |
5152 | * In all cases it is actually safe for us to return without doing anything, | |
5153 | * because the inode has been copied into a raw inode buffer in | |
87f7e416 TT |
5154 | * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL |
5155 | * writeback. | |
ac27a0ec DK |
5156 | * |
5157 | * Note that we are absolutely dependent upon all inode dirtiers doing the | |
5158 | * right thing: they *must* call mark_inode_dirty() after dirtying info in | |
5159 | * which we are interested. | |
5160 | * | |
5161 | * It would be a bug for them to not do this. The code: | |
5162 | * | |
5163 | * mark_inode_dirty(inode) | |
5164 | * stuff(); | |
5165 | * inode->i_size = expr; | |
5166 | * | |
87f7e416 TT |
5167 | * is in error because write_inode() could occur while `stuff()' is running, |
5168 | * and the new i_size will be lost. Plus the inode will no longer be on the | |
5169 | * superblock's dirty inode list. | |
ac27a0ec | 5170 | */ |
a9185b41 | 5171 | int ext4_write_inode(struct inode *inode, struct writeback_control *wbc) |
ac27a0ec | 5172 | { |
91ac6f43 FM |
5173 | int err; |
5174 | ||
18f2c4fc TT |
5175 | if (WARN_ON_ONCE(current->flags & PF_MEMALLOC) || |
5176 | sb_rdonly(inode->i_sb)) | |
ac27a0ec DK |
5177 | return 0; |
5178 | ||
18f2c4fc TT |
5179 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
5180 | return -EIO; | |
5181 | ||
91ac6f43 FM |
5182 | if (EXT4_SB(inode->i_sb)->s_journal) { |
5183 | if (ext4_journal_current_handle()) { | |
5184 | jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n"); | |
5185 | dump_stack(); | |
5186 | return -EIO; | |
5187 | } | |
ac27a0ec | 5188 | |
10542c22 JK |
5189 | /* |
5190 | * No need to force transaction in WB_SYNC_NONE mode. Also | |
5191 | * ext4_sync_fs() will force the commit after everything is | |
5192 | * written. | |
5193 | */ | |
5194 | if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync) | |
91ac6f43 FM |
5195 | return 0; |
5196 | ||
aa75f4d3 | 5197 | err = ext4_fc_commit(EXT4_SB(inode->i_sb)->s_journal, |
18f2c4fc | 5198 | EXT4_I(inode)->i_sync_tid); |
91ac6f43 FM |
5199 | } else { |
5200 | struct ext4_iloc iloc; | |
ac27a0ec | 5201 | |
8016e29f | 5202 | err = __ext4_get_inode_loc_noinmem(inode, &iloc); |
91ac6f43 FM |
5203 | if (err) |
5204 | return err; | |
10542c22 JK |
5205 | /* |
5206 | * sync(2) will flush the whole buffer cache. No need to do | |
5207 | * it here separately for each inode. | |
5208 | */ | |
5209 | if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) | |
830156c7 FM |
5210 | sync_dirty_buffer(iloc.bh); |
5211 | if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) { | |
54d3adbc TT |
5212 | ext4_error_inode_block(inode, iloc.bh->b_blocknr, EIO, |
5213 | "IO error syncing inode"); | |
830156c7 FM |
5214 | err = -EIO; |
5215 | } | |
fd2dd9fb | 5216 | brelse(iloc.bh); |
91ac6f43 FM |
5217 | } |
5218 | return err; | |
ac27a0ec DK |
5219 | } |
5220 | ||
53e87268 JK |
5221 | /* |
5222 | * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate | |
5223 | * buffers that are attached to a page stradding i_size and are undergoing | |
5224 | * commit. In that case we have to wait for commit to finish and try again. | |
5225 | */ | |
5226 | static void ext4_wait_for_tail_page_commit(struct inode *inode) | |
5227 | { | |
5228 | struct page *page; | |
5229 | unsigned offset; | |
5230 | journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; | |
5231 | tid_t commit_tid = 0; | |
5232 | int ret; | |
5233 | ||
09cbfeaf | 5234 | offset = inode->i_size & (PAGE_SIZE - 1); |
53e87268 | 5235 | /* |
565333a1 | 5236 | * If the page is fully truncated, we don't need to wait for any commit |
5237 | * (and we even should not as __ext4_journalled_invalidatepage() may | |
5238 | * strip all buffers from the page but keep the page dirty which can then | |
5239 | * confuse e.g. concurrent ext4_writepage() seeing dirty page without | |
5240 | * buffers). Also we don't need to wait for any commit if all buffers in | |
5241 | * the page remain valid. This is most beneficial for the common case of | |
5242 | * blocksize == PAGESIZE. | |
53e87268 | 5243 | */ |
565333a1 | 5244 | if (!offset || offset > (PAGE_SIZE - i_blocksize(inode))) |
53e87268 JK |
5245 | return; |
5246 | while (1) { | |
5247 | page = find_lock_page(inode->i_mapping, | |
09cbfeaf | 5248 | inode->i_size >> PAGE_SHIFT); |
53e87268 JK |
5249 | if (!page) |
5250 | return; | |
ca99fdd2 | 5251 | ret = __ext4_journalled_invalidatepage(page, offset, |
09cbfeaf | 5252 | PAGE_SIZE - offset); |
53e87268 | 5253 | unlock_page(page); |
09cbfeaf | 5254 | put_page(page); |
53e87268 JK |
5255 | if (ret != -EBUSY) |
5256 | return; | |
5257 | commit_tid = 0; | |
5258 | read_lock(&journal->j_state_lock); | |
5259 | if (journal->j_committing_transaction) | |
5260 | commit_tid = journal->j_committing_transaction->t_tid; | |
5261 | read_unlock(&journal->j_state_lock); | |
5262 | if (commit_tid) | |
5263 | jbd2_log_wait_commit(journal, commit_tid); | |
5264 | } | |
5265 | } | |
5266 | ||
ac27a0ec | 5267 | /* |
617ba13b | 5268 | * ext4_setattr() |
ac27a0ec DK |
5269 | * |
5270 | * Called from notify_change. | |
5271 | * | |
5272 | * We want to trap VFS attempts to truncate the file as soon as | |
5273 | * possible. In particular, we want to make sure that when the VFS | |
5274 | * shrinks i_size, we put the inode on the orphan list and modify | |
5275 | * i_disksize immediately, so that during the subsequent flushing of | |
5276 | * dirty pages and freeing of disk blocks, we can guarantee that any | |
5277 | * commit will leave the blocks being flushed in an unused state on | |
5278 | * disk. (On recovery, the inode will get truncated and the blocks will | |
5279 | * be freed, so we have a strong guarantee that no future commit will | |
5280 | * leave these blocks visible to the user.) | |
5281 | * | |
678aaf48 JK |
5282 | * Another thing we have to assure is that if we are in ordered mode |
5283 | * and inode is still attached to the committing transaction, we must | |
5284 | * we start writeout of all the dirty pages which are being truncated. | |
5285 | * This way we are sure that all the data written in the previous | |
5286 | * transaction are already on disk (truncate waits for pages under | |
5287 | * writeback). | |
5288 | * | |
5289 | * Called with inode->i_mutex down. | |
ac27a0ec | 5290 | */ |
549c7297 CB |
5291 | int ext4_setattr(struct user_namespace *mnt_userns, struct dentry *dentry, |
5292 | struct iattr *attr) | |
ac27a0ec | 5293 | { |
2b0143b5 | 5294 | struct inode *inode = d_inode(dentry); |
ac27a0ec | 5295 | int error, rc = 0; |
3d287de3 | 5296 | int orphan = 0; |
ac27a0ec DK |
5297 | const unsigned int ia_valid = attr->ia_valid; |
5298 | ||
0db1ff22 TT |
5299 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
5300 | return -EIO; | |
5301 | ||
02b016ca TT |
5302 | if (unlikely(IS_IMMUTABLE(inode))) |
5303 | return -EPERM; | |
5304 | ||
5305 | if (unlikely(IS_APPEND(inode) && | |
5306 | (ia_valid & (ATTR_MODE | ATTR_UID | | |
5307 | ATTR_GID | ATTR_TIMES_SET)))) | |
5308 | return -EPERM; | |
5309 | ||
14f3db55 | 5310 | error = setattr_prepare(mnt_userns, dentry, attr); |
ac27a0ec DK |
5311 | if (error) |
5312 | return error; | |
5313 | ||
3ce2b8dd EB |
5314 | error = fscrypt_prepare_setattr(dentry, attr); |
5315 | if (error) | |
5316 | return error; | |
5317 | ||
c93d8f88 EB |
5318 | error = fsverity_prepare_setattr(dentry, attr); |
5319 | if (error) | |
5320 | return error; | |
5321 | ||
a7cdadee JK |
5322 | if (is_quota_modification(inode, attr)) { |
5323 | error = dquot_initialize(inode); | |
5324 | if (error) | |
5325 | return error; | |
5326 | } | |
aa75f4d3 | 5327 | ext4_fc_start_update(inode); |
08cefc7a EB |
5328 | if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) || |
5329 | (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) { | |
ac27a0ec DK |
5330 | handle_t *handle; |
5331 | ||
5332 | /* (user+group)*(old+new) structure, inode write (sb, | |
5333 | * inode block, ? - but truncate inode update has it) */ | |
9924a92a TT |
5334 | handle = ext4_journal_start(inode, EXT4_HT_QUOTA, |
5335 | (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) + | |
5336 | EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3); | |
ac27a0ec DK |
5337 | if (IS_ERR(handle)) { |
5338 | error = PTR_ERR(handle); | |
5339 | goto err_out; | |
5340 | } | |
7a9ca53a TE |
5341 | |
5342 | /* dquot_transfer() calls back ext4_get_inode_usage() which | |
5343 | * counts xattr inode references. | |
5344 | */ | |
5345 | down_read(&EXT4_I(inode)->xattr_sem); | |
b43fa828 | 5346 | error = dquot_transfer(inode, attr); |
7a9ca53a TE |
5347 | up_read(&EXT4_I(inode)->xattr_sem); |
5348 | ||
ac27a0ec | 5349 | if (error) { |
617ba13b | 5350 | ext4_journal_stop(handle); |
aa75f4d3 | 5351 | ext4_fc_stop_update(inode); |
ac27a0ec DK |
5352 | return error; |
5353 | } | |
5354 | /* Update corresponding info in inode so that everything is in | |
5355 | * one transaction */ | |
5356 | if (attr->ia_valid & ATTR_UID) | |
5357 | inode->i_uid = attr->ia_uid; | |
5358 | if (attr->ia_valid & ATTR_GID) | |
5359 | inode->i_gid = attr->ia_gid; | |
617ba13b MC |
5360 | error = ext4_mark_inode_dirty(handle, inode); |
5361 | ext4_journal_stop(handle); | |
512c15ef PB |
5362 | if (unlikely(error)) { |
5363 | ext4_fc_stop_update(inode); | |
4209ae12 | 5364 | return error; |
512c15ef | 5365 | } |
ac27a0ec DK |
5366 | } |
5367 | ||
3da40c7b | 5368 | if (attr->ia_valid & ATTR_SIZE) { |
5208386c | 5369 | handle_t *handle; |
3da40c7b | 5370 | loff_t oldsize = inode->i_size; |
b9c1c267 | 5371 | int shrink = (attr->ia_size < inode->i_size); |
562c72aa | 5372 | |
12e9b892 | 5373 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) { |
e2b46574 ES |
5374 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
5375 | ||
aa75f4d3 HS |
5376 | if (attr->ia_size > sbi->s_bitmap_maxbytes) { |
5377 | ext4_fc_stop_update(inode); | |
0c095c7f | 5378 | return -EFBIG; |
aa75f4d3 | 5379 | } |
e2b46574 | 5380 | } |
aa75f4d3 HS |
5381 | if (!S_ISREG(inode->i_mode)) { |
5382 | ext4_fc_stop_update(inode); | |
3da40c7b | 5383 | return -EINVAL; |
aa75f4d3 | 5384 | } |
dff6efc3 CH |
5385 | |
5386 | if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size) | |
5387 | inode_inc_iversion(inode); | |
5388 | ||
b9c1c267 JK |
5389 | if (shrink) { |
5390 | if (ext4_should_order_data(inode)) { | |
5391 | error = ext4_begin_ordered_truncate(inode, | |
678aaf48 | 5392 | attr->ia_size); |
b9c1c267 JK |
5393 | if (error) |
5394 | goto err_out; | |
5395 | } | |
5396 | /* | |
5397 | * Blocks are going to be removed from the inode. Wait | |
5398 | * for dio in flight. | |
5399 | */ | |
5400 | inode_dio_wait(inode); | |
5401 | } | |
5402 | ||
d4f5258e | 5403 | filemap_invalidate_lock(inode->i_mapping); |
b9c1c267 JK |
5404 | |
5405 | rc = ext4_break_layouts(inode); | |
5406 | if (rc) { | |
d4f5258e | 5407 | filemap_invalidate_unlock(inode->i_mapping); |
aa75f4d3 | 5408 | goto err_out; |
3da40c7b | 5409 | } |
b9c1c267 | 5410 | |
3da40c7b | 5411 | if (attr->ia_size != inode->i_size) { |
5208386c JK |
5412 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 3); |
5413 | if (IS_ERR(handle)) { | |
5414 | error = PTR_ERR(handle); | |
b9c1c267 | 5415 | goto out_mmap_sem; |
5208386c | 5416 | } |
3da40c7b | 5417 | if (ext4_handle_valid(handle) && shrink) { |
5208386c JK |
5418 | error = ext4_orphan_add(handle, inode); |
5419 | orphan = 1; | |
5420 | } | |
911af577 EG |
5421 | /* |
5422 | * Update c/mtime on truncate up, ext4_truncate() will | |
5423 | * update c/mtime in shrink case below | |
5424 | */ | |
5425 | if (!shrink) { | |
eeca7ea1 | 5426 | inode->i_mtime = current_time(inode); |
911af577 EG |
5427 | inode->i_ctime = inode->i_mtime; |
5428 | } | |
aa75f4d3 HS |
5429 | |
5430 | if (shrink) | |
a80f7fcf | 5431 | ext4_fc_track_range(handle, inode, |
aa75f4d3 HS |
5432 | (attr->ia_size > 0 ? attr->ia_size - 1 : 0) >> |
5433 | inode->i_sb->s_blocksize_bits, | |
a6b43324 | 5434 | EXT_MAX_BLOCKS - 1); |
aa75f4d3 HS |
5435 | else |
5436 | ext4_fc_track_range( | |
a80f7fcf | 5437 | handle, inode, |
aa75f4d3 HS |
5438 | (oldsize > 0 ? oldsize - 1 : oldsize) >> |
5439 | inode->i_sb->s_blocksize_bits, | |
5440 | (attr->ia_size > 0 ? attr->ia_size - 1 : 0) >> | |
5441 | inode->i_sb->s_blocksize_bits); | |
5442 | ||
90e775b7 | 5443 | down_write(&EXT4_I(inode)->i_data_sem); |
5208386c JK |
5444 | EXT4_I(inode)->i_disksize = attr->ia_size; |
5445 | rc = ext4_mark_inode_dirty(handle, inode); | |
5446 | if (!error) | |
5447 | error = rc; | |
90e775b7 JK |
5448 | /* |
5449 | * We have to update i_size under i_data_sem together | |
5450 | * with i_disksize to avoid races with writeback code | |
5451 | * running ext4_wb_update_i_disksize(). | |
5452 | */ | |
5453 | if (!error) | |
5454 | i_size_write(inode, attr->ia_size); | |
5455 | up_write(&EXT4_I(inode)->i_data_sem); | |
5208386c | 5456 | ext4_journal_stop(handle); |
b9c1c267 JK |
5457 | if (error) |
5458 | goto out_mmap_sem; | |
5459 | if (!shrink) { | |
5460 | pagecache_isize_extended(inode, oldsize, | |
5461 | inode->i_size); | |
5462 | } else if (ext4_should_journal_data(inode)) { | |
5463 | ext4_wait_for_tail_page_commit(inode); | |
678aaf48 | 5464 | } |
d6320cbf | 5465 | } |
430657b6 | 5466 | |
5208386c JK |
5467 | /* |
5468 | * Truncate pagecache after we've waited for commit | |
5469 | * in data=journal mode to make pages freeable. | |
5470 | */ | |
923ae0ff | 5471 | truncate_pagecache(inode, inode->i_size); |
b9c1c267 JK |
5472 | /* |
5473 | * Call ext4_truncate() even if i_size didn't change to | |
5474 | * truncate possible preallocated blocks. | |
5475 | */ | |
5476 | if (attr->ia_size <= oldsize) { | |
2c98eb5e TT |
5477 | rc = ext4_truncate(inode); |
5478 | if (rc) | |
5479 | error = rc; | |
5480 | } | |
b9c1c267 | 5481 | out_mmap_sem: |
d4f5258e | 5482 | filemap_invalidate_unlock(inode->i_mapping); |
072bd7ea | 5483 | } |
ac27a0ec | 5484 | |
2c98eb5e | 5485 | if (!error) { |
14f3db55 | 5486 | setattr_copy(mnt_userns, inode, attr); |
1025774c CH |
5487 | mark_inode_dirty(inode); |
5488 | } | |
5489 | ||
5490 | /* | |
5491 | * If the call to ext4_truncate failed to get a transaction handle at | |
5492 | * all, we need to clean up the in-core orphan list manually. | |
5493 | */ | |
3d287de3 | 5494 | if (orphan && inode->i_nlink) |
617ba13b | 5495 | ext4_orphan_del(NULL, inode); |
ac27a0ec | 5496 | |
2c98eb5e | 5497 | if (!error && (ia_valid & ATTR_MODE)) |
14f3db55 | 5498 | rc = posix_acl_chmod(mnt_userns, inode, inode->i_mode); |
ac27a0ec DK |
5499 | |
5500 | err_out: | |
aa75f4d3 HS |
5501 | if (error) |
5502 | ext4_std_error(inode->i_sb, error); | |
ac27a0ec DK |
5503 | if (!error) |
5504 | error = rc; | |
aa75f4d3 | 5505 | ext4_fc_stop_update(inode); |
ac27a0ec DK |
5506 | return error; |
5507 | } | |
5508 | ||
549c7297 CB |
5509 | int ext4_getattr(struct user_namespace *mnt_userns, const struct path *path, |
5510 | struct kstat *stat, u32 request_mask, unsigned int query_flags) | |
3e3398a0 | 5511 | { |
99652ea5 DH |
5512 | struct inode *inode = d_inode(path->dentry); |
5513 | struct ext4_inode *raw_inode; | |
5514 | struct ext4_inode_info *ei = EXT4_I(inode); | |
5515 | unsigned int flags; | |
5516 | ||
d4c5e960 TT |
5517 | if ((request_mask & STATX_BTIME) && |
5518 | EXT4_FITS_IN_INODE(raw_inode, ei, i_crtime)) { | |
99652ea5 DH |
5519 | stat->result_mask |= STATX_BTIME; |
5520 | stat->btime.tv_sec = ei->i_crtime.tv_sec; | |
5521 | stat->btime.tv_nsec = ei->i_crtime.tv_nsec; | |
5522 | } | |
5523 | ||
5524 | flags = ei->i_flags & EXT4_FL_USER_VISIBLE; | |
5525 | if (flags & EXT4_APPEND_FL) | |
5526 | stat->attributes |= STATX_ATTR_APPEND; | |
5527 | if (flags & EXT4_COMPR_FL) | |
5528 | stat->attributes |= STATX_ATTR_COMPRESSED; | |
5529 | if (flags & EXT4_ENCRYPT_FL) | |
5530 | stat->attributes |= STATX_ATTR_ENCRYPTED; | |
5531 | if (flags & EXT4_IMMUTABLE_FL) | |
5532 | stat->attributes |= STATX_ATTR_IMMUTABLE; | |
5533 | if (flags & EXT4_NODUMP_FL) | |
5534 | stat->attributes |= STATX_ATTR_NODUMP; | |
1f607195 EB |
5535 | if (flags & EXT4_VERITY_FL) |
5536 | stat->attributes |= STATX_ATTR_VERITY; | |
3e3398a0 | 5537 | |
3209f68b DH |
5538 | stat->attributes_mask |= (STATX_ATTR_APPEND | |
5539 | STATX_ATTR_COMPRESSED | | |
5540 | STATX_ATTR_ENCRYPTED | | |
5541 | STATX_ATTR_IMMUTABLE | | |
1f607195 EB |
5542 | STATX_ATTR_NODUMP | |
5543 | STATX_ATTR_VERITY); | |
3209f68b | 5544 | |
14f3db55 | 5545 | generic_fillattr(mnt_userns, inode, stat); |
99652ea5 DH |
5546 | return 0; |
5547 | } | |
5548 | ||
549c7297 CB |
5549 | int ext4_file_getattr(struct user_namespace *mnt_userns, |
5550 | const struct path *path, struct kstat *stat, | |
99652ea5 DH |
5551 | u32 request_mask, unsigned int query_flags) |
5552 | { | |
5553 | struct inode *inode = d_inode(path->dentry); | |
5554 | u64 delalloc_blocks; | |
5555 | ||
14f3db55 | 5556 | ext4_getattr(mnt_userns, path, stat, request_mask, query_flags); |
3e3398a0 | 5557 | |
9206c561 AD |
5558 | /* |
5559 | * If there is inline data in the inode, the inode will normally not | |
5560 | * have data blocks allocated (it may have an external xattr block). | |
5561 | * Report at least one sector for such files, so tools like tar, rsync, | |
d67d64f4 | 5562 | * others don't incorrectly think the file is completely sparse. |
9206c561 AD |
5563 | */ |
5564 | if (unlikely(ext4_has_inline_data(inode))) | |
5565 | stat->blocks += (stat->size + 511) >> 9; | |
5566 | ||
3e3398a0 MC |
5567 | /* |
5568 | * We can't update i_blocks if the block allocation is delayed | |
5569 | * otherwise in the case of system crash before the real block | |
5570 | * allocation is done, we will have i_blocks inconsistent with | |
5571 | * on-disk file blocks. | |
5572 | * We always keep i_blocks updated together with real | |
5573 | * allocation. But to not confuse with user, stat | |
5574 | * will return the blocks that include the delayed allocation | |
5575 | * blocks for this file. | |
5576 | */ | |
96607551 | 5577 | delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb), |
9206c561 AD |
5578 | EXT4_I(inode)->i_reserved_data_blocks); |
5579 | stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9); | |
3e3398a0 MC |
5580 | return 0; |
5581 | } | |
ac27a0ec | 5582 | |
fffb2739 JK |
5583 | static int ext4_index_trans_blocks(struct inode *inode, int lblocks, |
5584 | int pextents) | |
a02908f1 | 5585 | { |
12e9b892 | 5586 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) |
fffb2739 JK |
5587 | return ext4_ind_trans_blocks(inode, lblocks); |
5588 | return ext4_ext_index_trans_blocks(inode, pextents); | |
a02908f1 | 5589 | } |
ac51d837 | 5590 | |
ac27a0ec | 5591 | /* |
a02908f1 MC |
5592 | * Account for index blocks, block groups bitmaps and block group |
5593 | * descriptor blocks if modify datablocks and index blocks | |
5594 | * worse case, the indexs blocks spread over different block groups | |
ac27a0ec | 5595 | * |
a02908f1 | 5596 | * If datablocks are discontiguous, they are possible to spread over |
4907cb7b | 5597 | * different block groups too. If they are contiguous, with flexbg, |
a02908f1 | 5598 | * they could still across block group boundary. |
ac27a0ec | 5599 | * |
a02908f1 MC |
5600 | * Also account for superblock, inode, quota and xattr blocks |
5601 | */ | |
dec214d0 | 5602 | static int ext4_meta_trans_blocks(struct inode *inode, int lblocks, |
fffb2739 | 5603 | int pextents) |
a02908f1 | 5604 | { |
8df9675f TT |
5605 | ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb); |
5606 | int gdpblocks; | |
a02908f1 MC |
5607 | int idxblocks; |
5608 | int ret = 0; | |
5609 | ||
5610 | /* | |
fffb2739 JK |
5611 | * How many index blocks need to touch to map @lblocks logical blocks |
5612 | * to @pextents physical extents? | |
a02908f1 | 5613 | */ |
fffb2739 | 5614 | idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents); |
a02908f1 MC |
5615 | |
5616 | ret = idxblocks; | |
5617 | ||
5618 | /* | |
5619 | * Now let's see how many group bitmaps and group descriptors need | |
5620 | * to account | |
5621 | */ | |
fffb2739 | 5622 | groups = idxblocks + pextents; |
a02908f1 | 5623 | gdpblocks = groups; |
8df9675f TT |
5624 | if (groups > ngroups) |
5625 | groups = ngroups; | |
a02908f1 MC |
5626 | if (groups > EXT4_SB(inode->i_sb)->s_gdb_count) |
5627 | gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count; | |
5628 | ||
5629 | /* bitmaps and block group descriptor blocks */ | |
5630 | ret += groups + gdpblocks; | |
5631 | ||
5632 | /* Blocks for super block, inode, quota and xattr blocks */ | |
5633 | ret += EXT4_META_TRANS_BLOCKS(inode->i_sb); | |
5634 | ||
5635 | return ret; | |
5636 | } | |
5637 | ||
5638 | /* | |
25985edc | 5639 | * Calculate the total number of credits to reserve to fit |
f3bd1f3f MC |
5640 | * the modification of a single pages into a single transaction, |
5641 | * which may include multiple chunks of block allocations. | |
ac27a0ec | 5642 | * |
525f4ed8 | 5643 | * This could be called via ext4_write_begin() |
ac27a0ec | 5644 | * |
525f4ed8 | 5645 | * We need to consider the worse case, when |
a02908f1 | 5646 | * one new block per extent. |
ac27a0ec | 5647 | */ |
a86c6181 | 5648 | int ext4_writepage_trans_blocks(struct inode *inode) |
ac27a0ec | 5649 | { |
617ba13b | 5650 | int bpp = ext4_journal_blocks_per_page(inode); |
ac27a0ec DK |
5651 | int ret; |
5652 | ||
fffb2739 | 5653 | ret = ext4_meta_trans_blocks(inode, bpp, bpp); |
a86c6181 | 5654 | |
a02908f1 | 5655 | /* Account for data blocks for journalled mode */ |
617ba13b | 5656 | if (ext4_should_journal_data(inode)) |
a02908f1 | 5657 | ret += bpp; |
ac27a0ec DK |
5658 | return ret; |
5659 | } | |
f3bd1f3f MC |
5660 | |
5661 | /* | |
5662 | * Calculate the journal credits for a chunk of data modification. | |
5663 | * | |
5664 | * This is called from DIO, fallocate or whoever calling | |
79e83036 | 5665 | * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks. |
f3bd1f3f MC |
5666 | * |
5667 | * journal buffers for data blocks are not included here, as DIO | |
5668 | * and fallocate do no need to journal data buffers. | |
5669 | */ | |
5670 | int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks) | |
5671 | { | |
5672 | return ext4_meta_trans_blocks(inode, nrblocks, 1); | |
5673 | } | |
5674 | ||
ac27a0ec | 5675 | /* |
617ba13b | 5676 | * The caller must have previously called ext4_reserve_inode_write(). |
ac27a0ec DK |
5677 | * Give this, we know that the caller already has write access to iloc->bh. |
5678 | */ | |
617ba13b | 5679 | int ext4_mark_iloc_dirty(handle_t *handle, |
de9a55b8 | 5680 | struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec DK |
5681 | { |
5682 | int err = 0; | |
5683 | ||
a6758309 VA |
5684 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) { |
5685 | put_bh(iloc->bh); | |
0db1ff22 | 5686 | return -EIO; |
a6758309 | 5687 | } |
a80f7fcf | 5688 | ext4_fc_track_inode(handle, inode); |
aa75f4d3 | 5689 | |
c64db50e | 5690 | if (IS_I_VERSION(inode)) |
25ec56b5 JNC |
5691 | inode_inc_iversion(inode); |
5692 | ||
ac27a0ec DK |
5693 | /* the do_update_inode consumes one bh->b_count */ |
5694 | get_bh(iloc->bh); | |
5695 | ||
dab291af | 5696 | /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */ |
830156c7 | 5697 | err = ext4_do_update_inode(handle, inode, iloc); |
ac27a0ec DK |
5698 | put_bh(iloc->bh); |
5699 | return err; | |
5700 | } | |
5701 | ||
5702 | /* | |
5703 | * On success, We end up with an outstanding reference count against | |
5704 | * iloc->bh. This _must_ be cleaned up later. | |
5705 | */ | |
5706 | ||
5707 | int | |
617ba13b MC |
5708 | ext4_reserve_inode_write(handle_t *handle, struct inode *inode, |
5709 | struct ext4_iloc *iloc) | |
ac27a0ec | 5710 | { |
0390131b FM |
5711 | int err; |
5712 | ||
0db1ff22 TT |
5713 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
5714 | return -EIO; | |
5715 | ||
0390131b FM |
5716 | err = ext4_get_inode_loc(inode, iloc); |
5717 | if (!err) { | |
5718 | BUFFER_TRACE(iloc->bh, "get_write_access"); | |
188c299e JK |
5719 | err = ext4_journal_get_write_access(handle, inode->i_sb, |
5720 | iloc->bh, EXT4_JTR_NONE); | |
0390131b FM |
5721 | if (err) { |
5722 | brelse(iloc->bh); | |
5723 | iloc->bh = NULL; | |
ac27a0ec DK |
5724 | } |
5725 | } | |
617ba13b | 5726 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
5727 | return err; |
5728 | } | |
5729 | ||
c03b45b8 MX |
5730 | static int __ext4_expand_extra_isize(struct inode *inode, |
5731 | unsigned int new_extra_isize, | |
5732 | struct ext4_iloc *iloc, | |
5733 | handle_t *handle, int *no_expand) | |
5734 | { | |
5735 | struct ext4_inode *raw_inode; | |
5736 | struct ext4_xattr_ibody_header *header; | |
4ea99936 TT |
5737 | unsigned int inode_size = EXT4_INODE_SIZE(inode->i_sb); |
5738 | struct ext4_inode_info *ei = EXT4_I(inode); | |
c03b45b8 MX |
5739 | int error; |
5740 | ||
4ea99936 TT |
5741 | /* this was checked at iget time, but double check for good measure */ |
5742 | if ((EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > inode_size) || | |
5743 | (ei->i_extra_isize & 3)) { | |
5744 | EXT4_ERROR_INODE(inode, "bad extra_isize %u (inode size %u)", | |
5745 | ei->i_extra_isize, | |
5746 | EXT4_INODE_SIZE(inode->i_sb)); | |
5747 | return -EFSCORRUPTED; | |
5748 | } | |
5749 | if ((new_extra_isize < ei->i_extra_isize) || | |
5750 | (new_extra_isize < 4) || | |
5751 | (new_extra_isize > inode_size - EXT4_GOOD_OLD_INODE_SIZE)) | |
5752 | return -EINVAL; /* Should never happen */ | |
5753 | ||
c03b45b8 MX |
5754 | raw_inode = ext4_raw_inode(iloc); |
5755 | ||
5756 | header = IHDR(inode, raw_inode); | |
5757 | ||
5758 | /* No extended attributes present */ | |
5759 | if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) || | |
5760 | header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) { | |
5761 | memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE + | |
5762 | EXT4_I(inode)->i_extra_isize, 0, | |
5763 | new_extra_isize - EXT4_I(inode)->i_extra_isize); | |
5764 | EXT4_I(inode)->i_extra_isize = new_extra_isize; | |
5765 | return 0; | |
5766 | } | |
5767 | ||
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; | |
c8585c6f | 5931 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
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 | ||
bbd55937 | 5969 | percpu_down_write(&sbi->s_writepages_rwsem); |
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); | |
bbd55937 | 5986 | percpu_up_write(&sbi->s_writepages_rwsem); |
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); |
bbd55937 | 5994 | percpu_up_write(&sbi->s_writepages_rwsem); |
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, |
e5aed39e | 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; |
c2ec175c | 6024 | struct page *page = 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 DW |
6067 | |
6068 | lock_page(page); | |
9ea7df53 JK |
6069 | size = i_size_read(inode); |
6070 | /* Page got truncated from under us? */ | |
6071 | if (page->mapping != mapping || page_offset(page) > size) { | |
6072 | unlock_page(page); | |
6073 | ret = VM_FAULT_NOPAGE; | |
6074 | goto out; | |
0e499890 | 6075 | } |
2e9ee850 | 6076 | |
09cbfeaf KS |
6077 | if (page->index == size >> PAGE_SHIFT) |
6078 | len = size & ~PAGE_MASK; | |
2e9ee850 | 6079 | else |
09cbfeaf | 6080 | len = PAGE_SIZE; |
a827eaff | 6081 | /* |
9ea7df53 JK |
6082 | * Return if we have all the buffers mapped. This avoids the need to do |
6083 | * journal_start/journal_stop which can block and take a long time | |
64a9f144 MFO |
6084 | * |
6085 | * This cannot be done for data journalling, as we have to add the | |
6086 | * inode to the transaction's list to writeprotect pages on commit. | |
a827eaff | 6087 | */ |
2e9ee850 | 6088 | if (page_has_buffers(page)) { |
188c299e | 6089 | if (!ext4_walk_page_buffers(NULL, inode, page_buffers(page), |
f19d5870 TM |
6090 | 0, len, NULL, |
6091 | ext4_bh_unmapped)) { | |
9ea7df53 | 6092 | /* Wait so that we don't change page under IO */ |
1d1d1a76 | 6093 | wait_for_stable_page(page); |
9ea7df53 JK |
6094 | ret = VM_FAULT_LOCKED; |
6095 | goto out; | |
a827eaff | 6096 | } |
2e9ee850 | 6097 | } |
a827eaff | 6098 | unlock_page(page); |
9ea7df53 JK |
6099 | /* OK, we need to fill the hole... */ |
6100 | if (ext4_should_dioread_nolock(inode)) | |
705965bd | 6101 | get_block = ext4_get_block_unwritten; |
9ea7df53 JK |
6102 | else |
6103 | get_block = ext4_get_block; | |
6104 | retry_alloc: | |
9924a92a TT |
6105 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
6106 | ext4_writepage_trans_blocks(inode)); | |
9ea7df53 | 6107 | if (IS_ERR(handle)) { |
c2ec175c | 6108 | ret = VM_FAULT_SIGBUS; |
9ea7df53 JK |
6109 | goto out; |
6110 | } | |
64a9f144 MFO |
6111 | /* |
6112 | * Data journalling can't use block_page_mkwrite() because it | |
6113 | * will set_buffer_dirty() before do_journal_get_write_access() | |
6114 | * thus might hit warning messages for dirty metadata buffers. | |
6115 | */ | |
6116 | if (!ext4_should_journal_data(inode)) { | |
6117 | err = block_page_mkwrite(vma, vmf, get_block); | |
6118 | } else { | |
6119 | lock_page(page); | |
6120 | size = i_size_read(inode); | |
6121 | /* Page got truncated from under us? */ | |
6122 | if (page->mapping != mapping || page_offset(page) > size) { | |
64a9f144 | 6123 | ret = VM_FAULT_NOPAGE; |
afb585a9 | 6124 | goto out_error; |
9ea7df53 | 6125 | } |
64a9f144 MFO |
6126 | |
6127 | if (page->index == size >> PAGE_SHIFT) | |
6128 | len = size & ~PAGE_MASK; | |
6129 | else | |
6130 | len = PAGE_SIZE; | |
6131 | ||
6132 | err = __block_write_begin(page, 0, len, ext4_get_block); | |
6133 | if (!err) { | |
afb585a9 | 6134 | ret = VM_FAULT_SIGBUS; |
188c299e JK |
6135 | if (ext4_walk_page_buffers(handle, inode, |
6136 | page_buffers(page), 0, len, NULL, | |
6137 | do_journal_get_write_access)) | |
afb585a9 | 6138 | goto out_error; |
188c299e JK |
6139 | if (ext4_walk_page_buffers(handle, inode, |
6140 | page_buffers(page), 0, len, NULL, | |
6141 | write_end_fn)) | |
afb585a9 | 6142 | goto out_error; |
b5b18160 JK |
6143 | if (ext4_jbd2_inode_add_write(handle, inode, |
6144 | page_offset(page), len)) | |
afb585a9 | 6145 | goto out_error; |
64a9f144 MFO |
6146 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
6147 | } else { | |
6148 | unlock_page(page); | |
6149 | } | |
9ea7df53 JK |
6150 | } |
6151 | ext4_journal_stop(handle); | |
401b25aa | 6152 | if (err == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) |
9ea7df53 JK |
6153 | goto retry_alloc; |
6154 | out_ret: | |
401b25aa | 6155 | ret = block_page_mkwrite_return(err); |
9ea7df53 | 6156 | out: |
d4f5258e | 6157 | filemap_invalidate_unlock_shared(mapping); |
8e8ad8a5 | 6158 | sb_end_pagefault(inode->i_sb); |
2e9ee850 | 6159 | return ret; |
afb585a9 MFO |
6160 | out_error: |
6161 | unlock_page(page); | |
6162 | ext4_journal_stop(handle); | |
6163 | goto out; | |
2e9ee850 | 6164 | } |