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 HS |
339 | if (!list_empty(&EXT4_I(inode)->i_fc_list)) |
340 | ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_NOMEM); | |
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 | } | |
a80f7fcf | 744 | ext4_fc_track_range(handle, inode, map->m_lblk, |
aa75f4d3 | 745 | map->m_lblk + map->m_len - 1); |
6fd058f7 | 746 | } |
ec8c60be RH |
747 | |
748 | if (retval < 0) | |
70aa1554 | 749 | ext_debug(inode, "failed with err %d\n", retval); |
0e855ac8 AK |
750 | return retval; |
751 | } | |
752 | ||
ed8ad838 JK |
753 | /* |
754 | * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages | |
755 | * we have to be careful as someone else may be manipulating b_state as well. | |
756 | */ | |
757 | static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags) | |
758 | { | |
759 | unsigned long old_state; | |
760 | unsigned long new_state; | |
761 | ||
762 | flags &= EXT4_MAP_FLAGS; | |
763 | ||
764 | /* Dummy buffer_head? Set non-atomically. */ | |
765 | if (!bh->b_page) { | |
766 | bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags; | |
767 | return; | |
768 | } | |
769 | /* | |
770 | * Someone else may be modifying b_state. Be careful! This is ugly but | |
771 | * once we get rid of using bh as a container for mapping information | |
772 | * to pass to / from get_block functions, this can go away. | |
773 | */ | |
774 | do { | |
775 | old_state = READ_ONCE(bh->b_state); | |
776 | new_state = (old_state & ~EXT4_MAP_FLAGS) | flags; | |
777 | } while (unlikely( | |
778 | cmpxchg(&bh->b_state, old_state, new_state) != old_state)); | |
779 | } | |
780 | ||
2ed88685 TT |
781 | static int _ext4_get_block(struct inode *inode, sector_t iblock, |
782 | struct buffer_head *bh, int flags) | |
ac27a0ec | 783 | { |
2ed88685 | 784 | struct ext4_map_blocks map; |
efe70c29 | 785 | int ret = 0; |
ac27a0ec | 786 | |
46c7f254 TM |
787 | if (ext4_has_inline_data(inode)) |
788 | return -ERANGE; | |
789 | ||
2ed88685 TT |
790 | map.m_lblk = iblock; |
791 | map.m_len = bh->b_size >> inode->i_blkbits; | |
792 | ||
efe70c29 JK |
793 | ret = ext4_map_blocks(ext4_journal_current_handle(), inode, &map, |
794 | flags); | |
7fb5409d | 795 | if (ret > 0) { |
2ed88685 | 796 | map_bh(bh, inode->i_sb, map.m_pblk); |
ed8ad838 | 797 | ext4_update_bh_state(bh, map.m_flags); |
2ed88685 | 798 | bh->b_size = inode->i_sb->s_blocksize * map.m_len; |
7fb5409d | 799 | ret = 0; |
547edce3 RZ |
800 | } else if (ret == 0) { |
801 | /* hole case, need to fill in bh->b_size */ | |
802 | bh->b_size = inode->i_sb->s_blocksize * map.m_len; | |
ac27a0ec DK |
803 | } |
804 | return ret; | |
805 | } | |
806 | ||
2ed88685 TT |
807 | int ext4_get_block(struct inode *inode, sector_t iblock, |
808 | struct buffer_head *bh, int create) | |
809 | { | |
810 | return _ext4_get_block(inode, iblock, bh, | |
811 | create ? EXT4_GET_BLOCKS_CREATE : 0); | |
812 | } | |
813 | ||
705965bd JK |
814 | /* |
815 | * Get block function used when preparing for buffered write if we require | |
816 | * creating an unwritten extent if blocks haven't been allocated. The extent | |
817 | * will be converted to written after the IO is complete. | |
818 | */ | |
819 | int ext4_get_block_unwritten(struct inode *inode, sector_t iblock, | |
820 | struct buffer_head *bh_result, int create) | |
821 | { | |
822 | ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n", | |
823 | inode->i_ino, create); | |
824 | return _ext4_get_block(inode, iblock, bh_result, | |
825 | EXT4_GET_BLOCKS_IO_CREATE_EXT); | |
826 | } | |
827 | ||
efe70c29 JK |
828 | /* Maximum number of blocks we map for direct IO at once. */ |
829 | #define DIO_MAX_BLOCKS 4096 | |
830 | ||
ac27a0ec DK |
831 | /* |
832 | * `handle' can be NULL if create is zero | |
833 | */ | |
617ba13b | 834 | struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode, |
c5e298ae | 835 | ext4_lblk_t block, int map_flags) |
ac27a0ec | 836 | { |
2ed88685 TT |
837 | struct ext4_map_blocks map; |
838 | struct buffer_head *bh; | |
c5e298ae | 839 | int create = map_flags & EXT4_GET_BLOCKS_CREATE; |
10560082 | 840 | int err; |
ac27a0ec | 841 | |
837c23fb CX |
842 | ASSERT((EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY) |
843 | || handle != NULL || create == 0); | |
ac27a0ec | 844 | |
2ed88685 TT |
845 | map.m_lblk = block; |
846 | map.m_len = 1; | |
c5e298ae | 847 | err = ext4_map_blocks(handle, inode, &map, map_flags); |
ac27a0ec | 848 | |
10560082 TT |
849 | if (err == 0) |
850 | return create ? ERR_PTR(-ENOSPC) : NULL; | |
2ed88685 | 851 | if (err < 0) |
10560082 | 852 | return ERR_PTR(err); |
2ed88685 TT |
853 | |
854 | bh = sb_getblk(inode->i_sb, map.m_pblk); | |
10560082 TT |
855 | if (unlikely(!bh)) |
856 | return ERR_PTR(-ENOMEM); | |
2ed88685 | 857 | if (map.m_flags & EXT4_MAP_NEW) { |
837c23fb CX |
858 | ASSERT(create != 0); |
859 | ASSERT((EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY) | |
860 | || (handle != NULL)); | |
ac27a0ec | 861 | |
2ed88685 TT |
862 | /* |
863 | * Now that we do not always journal data, we should | |
864 | * keep in mind whether this should always journal the | |
865 | * new buffer as metadata. For now, regular file | |
866 | * writes use ext4_get_block instead, so it's not a | |
867 | * problem. | |
868 | */ | |
869 | lock_buffer(bh); | |
870 | BUFFER_TRACE(bh, "call get_create_access"); | |
188c299e JK |
871 | err = ext4_journal_get_create_access(handle, inode->i_sb, bh, |
872 | EXT4_JTR_NONE); | |
10560082 TT |
873 | if (unlikely(err)) { |
874 | unlock_buffer(bh); | |
875 | goto errout; | |
876 | } | |
877 | if (!buffer_uptodate(bh)) { | |
2ed88685 TT |
878 | memset(bh->b_data, 0, inode->i_sb->s_blocksize); |
879 | set_buffer_uptodate(bh); | |
ac27a0ec | 880 | } |
2ed88685 TT |
881 | unlock_buffer(bh); |
882 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); | |
883 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
10560082 TT |
884 | if (unlikely(err)) |
885 | goto errout; | |
886 | } else | |
2ed88685 | 887 | BUFFER_TRACE(bh, "not a new buffer"); |
2ed88685 | 888 | return bh; |
10560082 TT |
889 | errout: |
890 | brelse(bh); | |
891 | return ERR_PTR(err); | |
ac27a0ec DK |
892 | } |
893 | ||
617ba13b | 894 | struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode, |
c5e298ae | 895 | ext4_lblk_t block, int map_flags) |
ac27a0ec | 896 | { |
af5bc92d | 897 | struct buffer_head *bh; |
2d069c08 | 898 | int ret; |
ac27a0ec | 899 | |
c5e298ae | 900 | bh = ext4_getblk(handle, inode, block, map_flags); |
1c215028 | 901 | if (IS_ERR(bh)) |
ac27a0ec | 902 | return bh; |
7963e5ac | 903 | if (!bh || ext4_buffer_uptodate(bh)) |
ac27a0ec | 904 | return bh; |
2d069c08 | 905 | |
906 | ret = ext4_read_bh_lock(bh, REQ_META | REQ_PRIO, true); | |
907 | if (ret) { | |
908 | put_bh(bh); | |
909 | return ERR_PTR(ret); | |
910 | } | |
911 | return bh; | |
ac27a0ec DK |
912 | } |
913 | ||
9699d4f9 TE |
914 | /* Read a contiguous batch of blocks. */ |
915 | int ext4_bread_batch(struct inode *inode, ext4_lblk_t block, int bh_count, | |
916 | bool wait, struct buffer_head **bhs) | |
917 | { | |
918 | int i, err; | |
919 | ||
920 | for (i = 0; i < bh_count; i++) { | |
921 | bhs[i] = ext4_getblk(NULL, inode, block + i, 0 /* map_flags */); | |
922 | if (IS_ERR(bhs[i])) { | |
923 | err = PTR_ERR(bhs[i]); | |
924 | bh_count = i; | |
925 | goto out_brelse; | |
926 | } | |
927 | } | |
928 | ||
929 | for (i = 0; i < bh_count; i++) | |
930 | /* Note that NULL bhs[i] is valid because of holes. */ | |
2d069c08 | 931 | if (bhs[i] && !ext4_buffer_uptodate(bhs[i])) |
932 | ext4_read_bh_lock(bhs[i], REQ_META | REQ_PRIO, false); | |
9699d4f9 TE |
933 | |
934 | if (!wait) | |
935 | return 0; | |
936 | ||
937 | for (i = 0; i < bh_count; i++) | |
938 | if (bhs[i]) | |
939 | wait_on_buffer(bhs[i]); | |
940 | ||
941 | for (i = 0; i < bh_count; i++) { | |
942 | if (bhs[i] && !buffer_uptodate(bhs[i])) { | |
943 | err = -EIO; | |
944 | goto out_brelse; | |
945 | } | |
946 | } | |
947 | return 0; | |
948 | ||
949 | out_brelse: | |
950 | for (i = 0; i < bh_count; i++) { | |
951 | brelse(bhs[i]); | |
952 | bhs[i] = NULL; | |
953 | } | |
954 | return err; | |
955 | } | |
956 | ||
188c299e | 957 | int ext4_walk_page_buffers(handle_t *handle, struct inode *inode, |
f19d5870 TM |
958 | struct buffer_head *head, |
959 | unsigned from, | |
960 | unsigned to, | |
961 | int *partial, | |
188c299e | 962 | int (*fn)(handle_t *handle, struct inode *inode, |
f19d5870 | 963 | struct buffer_head *bh)) |
ac27a0ec DK |
964 | { |
965 | struct buffer_head *bh; | |
966 | unsigned block_start, block_end; | |
967 | unsigned blocksize = head->b_size; | |
968 | int err, ret = 0; | |
969 | struct buffer_head *next; | |
970 | ||
af5bc92d TT |
971 | for (bh = head, block_start = 0; |
972 | ret == 0 && (bh != head || !block_start); | |
de9a55b8 | 973 | block_start = block_end, bh = next) { |
ac27a0ec DK |
974 | next = bh->b_this_page; |
975 | block_end = block_start + blocksize; | |
976 | if (block_end <= from || block_start >= to) { | |
977 | if (partial && !buffer_uptodate(bh)) | |
978 | *partial = 1; | |
979 | continue; | |
980 | } | |
188c299e | 981 | err = (*fn)(handle, inode, bh); |
ac27a0ec DK |
982 | if (!ret) |
983 | ret = err; | |
984 | } | |
985 | return ret; | |
986 | } | |
987 | ||
988 | /* | |
989 | * To preserve ordering, it is essential that the hole instantiation and | |
990 | * the data write be encapsulated in a single transaction. We cannot | |
617ba13b | 991 | * close off a transaction and start a new one between the ext4_get_block() |
dab291af | 992 | * and the commit_write(). So doing the jbd2_journal_start at the start of |
ac27a0ec DK |
993 | * prepare_write() is the right place. |
994 | * | |
36ade451 JK |
995 | * Also, this function can nest inside ext4_writepage(). In that case, we |
996 | * *know* that ext4_writepage() has generated enough buffer credits to do the | |
997 | * whole page. So we won't block on the journal in that case, which is good, | |
998 | * because the caller may be PF_MEMALLOC. | |
ac27a0ec | 999 | * |
617ba13b | 1000 | * By accident, ext4 can be reentered when a transaction is open via |
ac27a0ec DK |
1001 | * quota file writes. If we were to commit the transaction while thus |
1002 | * reentered, there can be a deadlock - we would be holding a quota | |
1003 | * lock, and the commit would never complete if another thread had a | |
1004 | * transaction open and was blocking on the quota lock - a ranking | |
1005 | * violation. | |
1006 | * | |
dab291af | 1007 | * So what we do is to rely on the fact that jbd2_journal_stop/journal_start |
ac27a0ec DK |
1008 | * will _not_ run commit under these circumstances because handle->h_ref |
1009 | * is elevated. We'll still have enough credits for the tiny quotafile | |
1010 | * write. | |
1011 | */ | |
188c299e | 1012 | int do_journal_get_write_access(handle_t *handle, struct inode *inode, |
f19d5870 | 1013 | struct buffer_head *bh) |
ac27a0ec | 1014 | { |
56d35a4c JK |
1015 | int dirty = buffer_dirty(bh); |
1016 | int ret; | |
1017 | ||
ac27a0ec DK |
1018 | if (!buffer_mapped(bh) || buffer_freed(bh)) |
1019 | return 0; | |
56d35a4c | 1020 | /* |
ebdec241 | 1021 | * __block_write_begin() could have dirtied some buffers. Clean |
56d35a4c JK |
1022 | * the dirty bit as jbd2_journal_get_write_access() could complain |
1023 | * otherwise about fs integrity issues. Setting of the dirty bit | |
ebdec241 | 1024 | * by __block_write_begin() isn't a real problem here as we clear |
56d35a4c JK |
1025 | * the bit before releasing a page lock and thus writeback cannot |
1026 | * ever write the buffer. | |
1027 | */ | |
1028 | if (dirty) | |
1029 | clear_buffer_dirty(bh); | |
5d601255 | 1030 | BUFFER_TRACE(bh, "get write access"); |
188c299e JK |
1031 | ret = ext4_journal_get_write_access(handle, inode->i_sb, bh, |
1032 | EXT4_JTR_NONE); | |
56d35a4c JK |
1033 | if (!ret && dirty) |
1034 | ret = ext4_handle_dirty_metadata(handle, NULL, bh); | |
1035 | return ret; | |
ac27a0ec DK |
1036 | } |
1037 | ||
643fa961 | 1038 | #ifdef CONFIG_FS_ENCRYPTION |
2058f83a MH |
1039 | static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len, |
1040 | get_block_t *get_block) | |
1041 | { | |
09cbfeaf | 1042 | unsigned from = pos & (PAGE_SIZE - 1); |
2058f83a MH |
1043 | unsigned to = from + len; |
1044 | struct inode *inode = page->mapping->host; | |
1045 | unsigned block_start, block_end; | |
1046 | sector_t block; | |
1047 | int err = 0; | |
1048 | unsigned blocksize = inode->i_sb->s_blocksize; | |
1049 | unsigned bbits; | |
0b578f35 CR |
1050 | struct buffer_head *bh, *head, *wait[2]; |
1051 | int nr_wait = 0; | |
1052 | int i; | |
2058f83a MH |
1053 | |
1054 | BUG_ON(!PageLocked(page)); | |
09cbfeaf KS |
1055 | BUG_ON(from > PAGE_SIZE); |
1056 | BUG_ON(to > PAGE_SIZE); | |
2058f83a MH |
1057 | BUG_ON(from > to); |
1058 | ||
1059 | if (!page_has_buffers(page)) | |
1060 | create_empty_buffers(page, blocksize, 0); | |
1061 | head = page_buffers(page); | |
1062 | bbits = ilog2(blocksize); | |
09cbfeaf | 1063 | block = (sector_t)page->index << (PAGE_SHIFT - bbits); |
2058f83a MH |
1064 | |
1065 | for (bh = head, block_start = 0; bh != head || !block_start; | |
1066 | block++, block_start = block_end, bh = bh->b_this_page) { | |
1067 | block_end = block_start + blocksize; | |
1068 | if (block_end <= from || block_start >= to) { | |
1069 | if (PageUptodate(page)) { | |
3cd46171 | 1070 | set_buffer_uptodate(bh); |
2058f83a MH |
1071 | } |
1072 | continue; | |
1073 | } | |
1074 | if (buffer_new(bh)) | |
1075 | clear_buffer_new(bh); | |
1076 | if (!buffer_mapped(bh)) { | |
1077 | WARN_ON(bh->b_size != blocksize); | |
1078 | err = get_block(inode, block, bh, 1); | |
1079 | if (err) | |
1080 | break; | |
1081 | if (buffer_new(bh)) { | |
2058f83a MH |
1082 | if (PageUptodate(page)) { |
1083 | clear_buffer_new(bh); | |
1084 | set_buffer_uptodate(bh); | |
1085 | mark_buffer_dirty(bh); | |
1086 | continue; | |
1087 | } | |
1088 | if (block_end > to || block_start < from) | |
1089 | zero_user_segments(page, to, block_end, | |
1090 | block_start, from); | |
1091 | continue; | |
1092 | } | |
1093 | } | |
1094 | if (PageUptodate(page)) { | |
3cd46171 | 1095 | set_buffer_uptodate(bh); |
2058f83a MH |
1096 | continue; |
1097 | } | |
1098 | if (!buffer_uptodate(bh) && !buffer_delay(bh) && | |
1099 | !buffer_unwritten(bh) && | |
1100 | (block_start < from || block_end > to)) { | |
2d069c08 | 1101 | ext4_read_bh_lock(bh, 0, false); |
0b578f35 | 1102 | wait[nr_wait++] = bh; |
2058f83a MH |
1103 | } |
1104 | } | |
1105 | /* | |
1106 | * If we issued read requests, let them complete. | |
1107 | */ | |
0b578f35 CR |
1108 | for (i = 0; i < nr_wait; i++) { |
1109 | wait_on_buffer(wait[i]); | |
1110 | if (!buffer_uptodate(wait[i])) | |
2058f83a MH |
1111 | err = -EIO; |
1112 | } | |
7e0785fc | 1113 | if (unlikely(err)) { |
2058f83a | 1114 | page_zero_new_buffers(page, from, to); |
4f74d15f | 1115 | } else if (fscrypt_inode_uses_fs_layer_crypto(inode)) { |
0b578f35 CR |
1116 | for (i = 0; i < nr_wait; i++) { |
1117 | int err2; | |
1118 | ||
1119 | err2 = fscrypt_decrypt_pagecache_blocks(page, blocksize, | |
1120 | bh_offset(wait[i])); | |
1121 | if (err2) { | |
1122 | clear_buffer_uptodate(wait[i]); | |
1123 | err = err2; | |
1124 | } | |
1125 | } | |
7e0785fc CR |
1126 | } |
1127 | ||
2058f83a MH |
1128 | return err; |
1129 | } | |
1130 | #endif | |
1131 | ||
bfc1af65 | 1132 | static int ext4_write_begin(struct file *file, struct address_space *mapping, |
de9a55b8 TT |
1133 | loff_t pos, unsigned len, unsigned flags, |
1134 | struct page **pagep, void **fsdata) | |
ac27a0ec | 1135 | { |
af5bc92d | 1136 | struct inode *inode = mapping->host; |
1938a150 | 1137 | int ret, needed_blocks; |
ac27a0ec DK |
1138 | handle_t *handle; |
1139 | int retries = 0; | |
af5bc92d | 1140 | struct page *page; |
de9a55b8 | 1141 | pgoff_t index; |
af5bc92d | 1142 | unsigned from, to; |
bfc1af65 | 1143 | |
0db1ff22 TT |
1144 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
1145 | return -EIO; | |
1146 | ||
9bffad1e | 1147 | trace_ext4_write_begin(inode, pos, len, flags); |
1938a150 AK |
1148 | /* |
1149 | * Reserve one block more for addition to orphan list in case | |
1150 | * we allocate blocks but write fails for some reason | |
1151 | */ | |
1152 | needed_blocks = ext4_writepage_trans_blocks(inode) + 1; | |
09cbfeaf KS |
1153 | index = pos >> PAGE_SHIFT; |
1154 | from = pos & (PAGE_SIZE - 1); | |
af5bc92d | 1155 | to = from + len; |
ac27a0ec | 1156 | |
f19d5870 TM |
1157 | if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { |
1158 | ret = ext4_try_to_write_inline_data(mapping, inode, pos, len, | |
1159 | flags, pagep); | |
1160 | if (ret < 0) | |
47564bfb TT |
1161 | return ret; |
1162 | if (ret == 1) | |
1163 | return 0; | |
f19d5870 TM |
1164 | } |
1165 | ||
47564bfb TT |
1166 | /* |
1167 | * grab_cache_page_write_begin() can take a long time if the | |
1168 | * system is thrashing due to memory pressure, or if the page | |
1169 | * is being written back. So grab it first before we start | |
1170 | * the transaction handle. This also allows us to allocate | |
1171 | * the page (if needed) without using GFP_NOFS. | |
1172 | */ | |
1173 | retry_grab: | |
1174 | page = grab_cache_page_write_begin(mapping, index, flags); | |
1175 | if (!page) | |
1176 | return -ENOMEM; | |
1177 | unlock_page(page); | |
1178 | ||
1179 | retry_journal: | |
9924a92a | 1180 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks); |
af5bc92d | 1181 | if (IS_ERR(handle)) { |
09cbfeaf | 1182 | put_page(page); |
47564bfb | 1183 | return PTR_ERR(handle); |
7479d2b9 | 1184 | } |
ac27a0ec | 1185 | |
47564bfb TT |
1186 | lock_page(page); |
1187 | if (page->mapping != mapping) { | |
1188 | /* The page got truncated from under us */ | |
1189 | unlock_page(page); | |
09cbfeaf | 1190 | put_page(page); |
cf108bca | 1191 | ext4_journal_stop(handle); |
47564bfb | 1192 | goto retry_grab; |
cf108bca | 1193 | } |
7afe5aa5 DM |
1194 | /* In case writeback began while the page was unlocked */ |
1195 | wait_for_stable_page(page); | |
cf108bca | 1196 | |
643fa961 | 1197 | #ifdef CONFIG_FS_ENCRYPTION |
2058f83a MH |
1198 | if (ext4_should_dioread_nolock(inode)) |
1199 | ret = ext4_block_write_begin(page, pos, len, | |
705965bd | 1200 | ext4_get_block_unwritten); |
2058f83a MH |
1201 | else |
1202 | ret = ext4_block_write_begin(page, pos, len, | |
1203 | ext4_get_block); | |
1204 | #else | |
744692dc | 1205 | if (ext4_should_dioread_nolock(inode)) |
705965bd JK |
1206 | ret = __block_write_begin(page, pos, len, |
1207 | ext4_get_block_unwritten); | |
744692dc | 1208 | else |
6e1db88d | 1209 | ret = __block_write_begin(page, pos, len, ext4_get_block); |
2058f83a | 1210 | #endif |
bfc1af65 | 1211 | if (!ret && ext4_should_journal_data(inode)) { |
188c299e JK |
1212 | ret = ext4_walk_page_buffers(handle, inode, |
1213 | page_buffers(page), from, to, NULL, | |
f19d5870 | 1214 | do_journal_get_write_access); |
ac27a0ec | 1215 | } |
bfc1af65 NP |
1216 | |
1217 | if (ret) { | |
c93d8f88 EB |
1218 | bool extended = (pos + len > inode->i_size) && |
1219 | !ext4_verity_in_progress(inode); | |
1220 | ||
af5bc92d | 1221 | unlock_page(page); |
ae4d5372 | 1222 | /* |
6e1db88d | 1223 | * __block_write_begin may have instantiated a few blocks |
ae4d5372 AK |
1224 | * outside i_size. Trim these off again. Don't need |
1225 | * i_size_read because we hold i_mutex. | |
1938a150 AK |
1226 | * |
1227 | * Add inode to orphan list in case we crash before | |
1228 | * truncate finishes | |
ae4d5372 | 1229 | */ |
c93d8f88 | 1230 | if (extended && ext4_can_truncate(inode)) |
1938a150 AK |
1231 | ext4_orphan_add(handle, inode); |
1232 | ||
1233 | ext4_journal_stop(handle); | |
c93d8f88 | 1234 | if (extended) { |
b9a4207d | 1235 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1236 | /* |
ffacfa7a | 1237 | * If truncate failed early the inode might |
1938a150 AK |
1238 | * still be on the orphan list; we need to |
1239 | * make sure the inode is removed from the | |
1240 | * orphan list in that case. | |
1241 | */ | |
1242 | if (inode->i_nlink) | |
1243 | ext4_orphan_del(NULL, inode); | |
1244 | } | |
bfc1af65 | 1245 | |
47564bfb TT |
1246 | if (ret == -ENOSPC && |
1247 | ext4_should_retry_alloc(inode->i_sb, &retries)) | |
1248 | goto retry_journal; | |
09cbfeaf | 1249 | put_page(page); |
47564bfb TT |
1250 | return ret; |
1251 | } | |
1252 | *pagep = page; | |
ac27a0ec DK |
1253 | return ret; |
1254 | } | |
1255 | ||
bfc1af65 | 1256 | /* For write_end() in data=journal mode */ |
188c299e JK |
1257 | static int write_end_fn(handle_t *handle, struct inode *inode, |
1258 | struct buffer_head *bh) | |
ac27a0ec | 1259 | { |
13fca323 | 1260 | int ret; |
ac27a0ec DK |
1261 | if (!buffer_mapped(bh) || buffer_freed(bh)) |
1262 | return 0; | |
1263 | set_buffer_uptodate(bh); | |
13fca323 TT |
1264 | ret = ext4_handle_dirty_metadata(handle, NULL, bh); |
1265 | clear_buffer_meta(bh); | |
1266 | clear_buffer_prio(bh); | |
1267 | return ret; | |
ac27a0ec DK |
1268 | } |
1269 | ||
eed4333f ZL |
1270 | /* |
1271 | * We need to pick up the new inode size which generic_commit_write gave us | |
1272 | * `file' can be NULL - eg, when called from page_symlink(). | |
1273 | * | |
1274 | * ext4 never places buffers on inode->i_mapping->private_list. metadata | |
1275 | * buffers are managed internally. | |
1276 | */ | |
1277 | static int ext4_write_end(struct file *file, | |
1278 | struct address_space *mapping, | |
1279 | loff_t pos, unsigned len, unsigned copied, | |
1280 | struct page *page, void *fsdata) | |
f8514083 | 1281 | { |
f8514083 | 1282 | handle_t *handle = ext4_journal_current_handle(); |
eed4333f | 1283 | struct inode *inode = mapping->host; |
0572639f | 1284 | loff_t old_size = inode->i_size; |
eed4333f ZL |
1285 | int ret = 0, ret2; |
1286 | int i_size_changed = 0; | |
362eca70 | 1287 | int inline_data = ext4_has_inline_data(inode); |
c93d8f88 | 1288 | bool verity = ext4_verity_in_progress(inode); |
eed4333f ZL |
1289 | |
1290 | trace_ext4_write_end(inode, pos, len, copied); | |
362eca70 | 1291 | if (inline_data) { |
42c832de TT |
1292 | ret = ext4_write_inline_data_end(inode, pos, len, |
1293 | copied, page); | |
eb5efbcb TT |
1294 | if (ret < 0) { |
1295 | unlock_page(page); | |
1296 | put_page(page); | |
42c832de | 1297 | goto errout; |
eb5efbcb | 1298 | } |
42c832de TT |
1299 | copied = ret; |
1300 | } else | |
f19d5870 TM |
1301 | copied = block_write_end(file, mapping, pos, |
1302 | len, copied, page, fsdata); | |
f8514083 | 1303 | /* |
4631dbf6 | 1304 | * it's important to update i_size while still holding page lock: |
f8514083 | 1305 | * page writeout could otherwise come in and zero beyond i_size. |
c93d8f88 EB |
1306 | * |
1307 | * If FS_IOC_ENABLE_VERITY is running on this inode, then Merkle tree | |
1308 | * blocks are being written past EOF, so skip the i_size update. | |
f8514083 | 1309 | */ |
c93d8f88 EB |
1310 | if (!verity) |
1311 | i_size_changed = ext4_update_inode_size(inode, pos + copied); | |
f8514083 | 1312 | unlock_page(page); |
09cbfeaf | 1313 | put_page(page); |
f8514083 | 1314 | |
c93d8f88 | 1315 | if (old_size < pos && !verity) |
0572639f | 1316 | pagecache_isize_extended(inode, old_size, pos); |
f8514083 AK |
1317 | /* |
1318 | * Don't mark the inode dirty under page lock. First, it unnecessarily | |
1319 | * makes the holding time of page lock longer. Second, it forces lock | |
1320 | * ordering of page lock and transaction start for journaling | |
1321 | * filesystems. | |
1322 | */ | |
362eca70 | 1323 | if (i_size_changed || inline_data) |
4209ae12 | 1324 | ret = ext4_mark_inode_dirty(handle, inode); |
f8514083 | 1325 | |
c93d8f88 | 1326 | if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode)) |
f8514083 AK |
1327 | /* if we have allocated more blocks and copied |
1328 | * less. We will have blocks allocated outside | |
1329 | * inode->i_size. So truncate them | |
1330 | */ | |
1331 | ext4_orphan_add(handle, inode); | |
74d553aa | 1332 | errout: |
617ba13b | 1333 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1334 | if (!ret) |
1335 | ret = ret2; | |
bfc1af65 | 1336 | |
c93d8f88 | 1337 | if (pos + len > inode->i_size && !verity) { |
b9a4207d | 1338 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1339 | /* |
ffacfa7a | 1340 | * If truncate failed early the inode might still be |
f8514083 AK |
1341 | * on the orphan list; we need to make sure the inode |
1342 | * is removed from the orphan list in that case. | |
1343 | */ | |
1344 | if (inode->i_nlink) | |
1345 | ext4_orphan_del(NULL, inode); | |
1346 | } | |
1347 | ||
bfc1af65 | 1348 | return ret ? ret : copied; |
ac27a0ec DK |
1349 | } |
1350 | ||
b90197b6 TT |
1351 | /* |
1352 | * This is a private version of page_zero_new_buffers() which doesn't | |
1353 | * set the buffer to be dirty, since in data=journalled mode we need | |
1354 | * to call ext4_handle_dirty_metadata() instead. | |
1355 | */ | |
3b136499 | 1356 | static void ext4_journalled_zero_new_buffers(handle_t *handle, |
188c299e | 1357 | struct inode *inode, |
3b136499 JK |
1358 | struct page *page, |
1359 | unsigned from, unsigned to) | |
b90197b6 TT |
1360 | { |
1361 | unsigned int block_start = 0, block_end; | |
1362 | struct buffer_head *head, *bh; | |
1363 | ||
1364 | bh = head = page_buffers(page); | |
1365 | do { | |
1366 | block_end = block_start + bh->b_size; | |
1367 | if (buffer_new(bh)) { | |
1368 | if (block_end > from && block_start < to) { | |
1369 | if (!PageUptodate(page)) { | |
1370 | unsigned start, size; | |
1371 | ||
1372 | start = max(from, block_start); | |
1373 | size = min(to, block_end) - start; | |
1374 | ||
1375 | zero_user(page, start, size); | |
188c299e | 1376 | write_end_fn(handle, inode, bh); |
b90197b6 TT |
1377 | } |
1378 | clear_buffer_new(bh); | |
1379 | } | |
1380 | } | |
1381 | block_start = block_end; | |
1382 | bh = bh->b_this_page; | |
1383 | } while (bh != head); | |
1384 | } | |
1385 | ||
bfc1af65 | 1386 | static int ext4_journalled_write_end(struct file *file, |
de9a55b8 TT |
1387 | struct address_space *mapping, |
1388 | loff_t pos, unsigned len, unsigned copied, | |
1389 | struct page *page, void *fsdata) | |
ac27a0ec | 1390 | { |
617ba13b | 1391 | handle_t *handle = ext4_journal_current_handle(); |
bfc1af65 | 1392 | struct inode *inode = mapping->host; |
0572639f | 1393 | loff_t old_size = inode->i_size; |
ac27a0ec DK |
1394 | int ret = 0, ret2; |
1395 | int partial = 0; | |
bfc1af65 | 1396 | unsigned from, to; |
4631dbf6 | 1397 | int size_changed = 0; |
362eca70 | 1398 | int inline_data = ext4_has_inline_data(inode); |
c93d8f88 | 1399 | bool verity = ext4_verity_in_progress(inode); |
ac27a0ec | 1400 | |
9bffad1e | 1401 | trace_ext4_journalled_write_end(inode, pos, len, copied); |
09cbfeaf | 1402 | from = pos & (PAGE_SIZE - 1); |
bfc1af65 NP |
1403 | to = from + len; |
1404 | ||
441c8508 CW |
1405 | BUG_ON(!ext4_handle_valid(handle)); |
1406 | ||
362eca70 | 1407 | if (inline_data) { |
eb5efbcb TT |
1408 | ret = ext4_write_inline_data_end(inode, pos, len, |
1409 | copied, page); | |
1410 | if (ret < 0) { | |
1411 | unlock_page(page); | |
1412 | put_page(page); | |
1413 | goto errout; | |
1414 | } | |
1415 | copied = ret; | |
1416 | } else if (unlikely(copied < len) && !PageUptodate(page)) { | |
3b136499 | 1417 | copied = 0; |
188c299e | 1418 | ext4_journalled_zero_new_buffers(handle, inode, page, from, to); |
3b136499 JK |
1419 | } else { |
1420 | if (unlikely(copied < len)) | |
188c299e | 1421 | ext4_journalled_zero_new_buffers(handle, inode, page, |
3b136499 | 1422 | from + copied, to); |
188c299e JK |
1423 | ret = ext4_walk_page_buffers(handle, inode, page_buffers(page), |
1424 | from, from + copied, &partial, | |
3b136499 | 1425 | write_end_fn); |
3fdcfb66 TM |
1426 | if (!partial) |
1427 | SetPageUptodate(page); | |
1428 | } | |
c93d8f88 EB |
1429 | if (!verity) |
1430 | size_changed = ext4_update_inode_size(inode, pos + copied); | |
19f5fb7a | 1431 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
2d859db3 | 1432 | EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; |
4631dbf6 | 1433 | unlock_page(page); |
09cbfeaf | 1434 | put_page(page); |
4631dbf6 | 1435 | |
c93d8f88 | 1436 | if (old_size < pos && !verity) |
0572639f XW |
1437 | pagecache_isize_extended(inode, old_size, pos); |
1438 | ||
362eca70 | 1439 | if (size_changed || inline_data) { |
617ba13b | 1440 | ret2 = ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
1441 | if (!ret) |
1442 | ret = ret2; | |
1443 | } | |
bfc1af65 | 1444 | |
c93d8f88 | 1445 | if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode)) |
f8514083 AK |
1446 | /* if we have allocated more blocks and copied |
1447 | * less. We will have blocks allocated outside | |
1448 | * inode->i_size. So truncate them | |
1449 | */ | |
1450 | ext4_orphan_add(handle, inode); | |
1451 | ||
eb5efbcb | 1452 | errout: |
617ba13b | 1453 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1454 | if (!ret) |
1455 | ret = ret2; | |
c93d8f88 | 1456 | if (pos + len > inode->i_size && !verity) { |
b9a4207d | 1457 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1458 | /* |
ffacfa7a | 1459 | * If truncate failed early the inode might still be |
f8514083 AK |
1460 | * on the orphan list; we need to make sure the inode |
1461 | * is removed from the orphan list in that case. | |
1462 | */ | |
1463 | if (inode->i_nlink) | |
1464 | ext4_orphan_del(NULL, inode); | |
1465 | } | |
bfc1af65 NP |
1466 | |
1467 | return ret ? ret : copied; | |
ac27a0ec | 1468 | } |
d2a17637 | 1469 | |
9d0be502 | 1470 | /* |
c27e43a1 | 1471 | * Reserve space for a single cluster |
9d0be502 | 1472 | */ |
c27e43a1 | 1473 | static int ext4_da_reserve_space(struct inode *inode) |
d2a17637 | 1474 | { |
60e58e0f | 1475 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
0637c6f4 | 1476 | struct ext4_inode_info *ei = EXT4_I(inode); |
5dd4056d | 1477 | int ret; |
03179fe9 TT |
1478 | |
1479 | /* | |
1480 | * We will charge metadata quota at writeout time; this saves | |
1481 | * us from metadata over-estimation, though we may go over by | |
1482 | * a small amount in the end. Here we just reserve for data. | |
1483 | */ | |
1484 | ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1)); | |
1485 | if (ret) | |
1486 | return ret; | |
d2a17637 | 1487 | |
0637c6f4 | 1488 | spin_lock(&ei->i_block_reservation_lock); |
71d4f7d0 | 1489 | if (ext4_claim_free_clusters(sbi, 1, 0)) { |
03179fe9 | 1490 | spin_unlock(&ei->i_block_reservation_lock); |
03179fe9 | 1491 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1)); |
d2a17637 MC |
1492 | return -ENOSPC; |
1493 | } | |
9d0be502 | 1494 | ei->i_reserved_data_blocks++; |
c27e43a1 | 1495 | trace_ext4_da_reserve_space(inode); |
0637c6f4 | 1496 | spin_unlock(&ei->i_block_reservation_lock); |
39bc680a | 1497 | |
d2a17637 MC |
1498 | return 0; /* success */ |
1499 | } | |
1500 | ||
f456767d | 1501 | void ext4_da_release_space(struct inode *inode, int to_free) |
d2a17637 MC |
1502 | { |
1503 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0637c6f4 | 1504 | struct ext4_inode_info *ei = EXT4_I(inode); |
d2a17637 | 1505 | |
cd213226 MC |
1506 | if (!to_free) |
1507 | return; /* Nothing to release, exit */ | |
1508 | ||
d2a17637 | 1509 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); |
cd213226 | 1510 | |
5a58ec87 | 1511 | trace_ext4_da_release_space(inode, to_free); |
0637c6f4 | 1512 | if (unlikely(to_free > ei->i_reserved_data_blocks)) { |
cd213226 | 1513 | /* |
0637c6f4 TT |
1514 | * if there aren't enough reserved blocks, then the |
1515 | * counter is messed up somewhere. Since this | |
1516 | * function is called from invalidate page, it's | |
1517 | * harmless to return without any action. | |
cd213226 | 1518 | */ |
8de5c325 | 1519 | ext4_warning(inode->i_sb, "ext4_da_release_space: " |
0637c6f4 | 1520 | "ino %lu, to_free %d with only %d reserved " |
1084f252 | 1521 | "data blocks", inode->i_ino, to_free, |
0637c6f4 TT |
1522 | ei->i_reserved_data_blocks); |
1523 | WARN_ON(1); | |
1524 | to_free = ei->i_reserved_data_blocks; | |
cd213226 | 1525 | } |
0637c6f4 | 1526 | ei->i_reserved_data_blocks -= to_free; |
cd213226 | 1527 | |
72b8ab9d | 1528 | /* update fs dirty data blocks counter */ |
57042651 | 1529 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free); |
d2a17637 | 1530 | |
d2a17637 | 1531 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
60e58e0f | 1532 | |
7b415bf6 | 1533 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free)); |
d2a17637 MC |
1534 | } |
1535 | ||
64769240 AT |
1536 | /* |
1537 | * Delayed allocation stuff | |
1538 | */ | |
1539 | ||
4e7ea81d JK |
1540 | struct mpage_da_data { |
1541 | struct inode *inode; | |
1542 | struct writeback_control *wbc; | |
6b523df4 | 1543 | |
4e7ea81d JK |
1544 | pgoff_t first_page; /* The first page to write */ |
1545 | pgoff_t next_page; /* Current page to examine */ | |
1546 | pgoff_t last_page; /* Last page to examine */ | |
791b7f08 | 1547 | /* |
4e7ea81d JK |
1548 | * Extent to map - this can be after first_page because that can be |
1549 | * fully mapped. We somewhat abuse m_flags to store whether the extent | |
1550 | * is delalloc or unwritten. | |
791b7f08 | 1551 | */ |
4e7ea81d JK |
1552 | struct ext4_map_blocks map; |
1553 | struct ext4_io_submit io_submit; /* IO submission data */ | |
dddbd6ac | 1554 | unsigned int do_map:1; |
6b8ed620 | 1555 | unsigned int scanned_until_end:1; |
4e7ea81d | 1556 | }; |
64769240 | 1557 | |
4e7ea81d JK |
1558 | static void mpage_release_unused_pages(struct mpage_da_data *mpd, |
1559 | bool invalidate) | |
c4a0c46e AK |
1560 | { |
1561 | int nr_pages, i; | |
1562 | pgoff_t index, end; | |
1563 | struct pagevec pvec; | |
1564 | struct inode *inode = mpd->inode; | |
1565 | struct address_space *mapping = inode->i_mapping; | |
4e7ea81d JK |
1566 | |
1567 | /* This is necessary when next_page == 0. */ | |
1568 | if (mpd->first_page >= mpd->next_page) | |
1569 | return; | |
c4a0c46e | 1570 | |
6b8ed620 | 1571 | mpd->scanned_until_end = 0; |
c7f5938a CW |
1572 | index = mpd->first_page; |
1573 | end = mpd->next_page - 1; | |
4e7ea81d JK |
1574 | if (invalidate) { |
1575 | ext4_lblk_t start, last; | |
09cbfeaf KS |
1576 | start = index << (PAGE_SHIFT - inode->i_blkbits); |
1577 | last = end << (PAGE_SHIFT - inode->i_blkbits); | |
4e7ea81d JK |
1578 | ext4_es_remove_extent(inode, start, last - start + 1); |
1579 | } | |
51865fda | 1580 | |
86679820 | 1581 | pagevec_init(&pvec); |
c4a0c46e | 1582 | while (index <= end) { |
397162ff | 1583 | nr_pages = pagevec_lookup_range(&pvec, mapping, &index, end); |
c4a0c46e AK |
1584 | if (nr_pages == 0) |
1585 | break; | |
1586 | for (i = 0; i < nr_pages; i++) { | |
1587 | struct page *page = pvec.pages[i]; | |
2b85a617 | 1588 | |
c4a0c46e AK |
1589 | BUG_ON(!PageLocked(page)); |
1590 | BUG_ON(PageWriteback(page)); | |
4e7ea81d | 1591 | if (invalidate) { |
4e800c03 | 1592 | if (page_mapped(page)) |
1593 | clear_page_dirty_for_io(page); | |
09cbfeaf | 1594 | block_invalidatepage(page, 0, PAGE_SIZE); |
4e7ea81d JK |
1595 | ClearPageUptodate(page); |
1596 | } | |
c4a0c46e AK |
1597 | unlock_page(page); |
1598 | } | |
9b1d0998 | 1599 | pagevec_release(&pvec); |
c4a0c46e | 1600 | } |
c4a0c46e AK |
1601 | } |
1602 | ||
df22291f AK |
1603 | static void ext4_print_free_blocks(struct inode *inode) |
1604 | { | |
1605 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
92b97816 | 1606 | struct super_block *sb = inode->i_sb; |
f78ee70d | 1607 | struct ext4_inode_info *ei = EXT4_I(inode); |
92b97816 TT |
1608 | |
1609 | ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld", | |
5dee5437 | 1610 | EXT4_C2B(EXT4_SB(inode->i_sb), |
f78ee70d | 1611 | ext4_count_free_clusters(sb))); |
92b97816 TT |
1612 | ext4_msg(sb, KERN_CRIT, "Free/Dirty block details"); |
1613 | ext4_msg(sb, KERN_CRIT, "free_blocks=%lld", | |
f78ee70d | 1614 | (long long) EXT4_C2B(EXT4_SB(sb), |
57042651 | 1615 | percpu_counter_sum(&sbi->s_freeclusters_counter))); |
92b97816 | 1616 | ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld", |
f78ee70d | 1617 | (long long) EXT4_C2B(EXT4_SB(sb), |
7b415bf6 | 1618 | percpu_counter_sum(&sbi->s_dirtyclusters_counter))); |
92b97816 TT |
1619 | ext4_msg(sb, KERN_CRIT, "Block reservation details"); |
1620 | ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u", | |
f78ee70d | 1621 | ei->i_reserved_data_blocks); |
df22291f AK |
1622 | return; |
1623 | } | |
1624 | ||
188c299e JK |
1625 | static int ext4_bh_delay_or_unwritten(handle_t *handle, struct inode *inode, |
1626 | struct buffer_head *bh) | |
29fa89d0 | 1627 | { |
c364b22c | 1628 | return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh); |
29fa89d0 AK |
1629 | } |
1630 | ||
0b02f4c0 EW |
1631 | /* |
1632 | * ext4_insert_delayed_block - adds a delayed block to the extents status | |
1633 | * tree, incrementing the reserved cluster/block | |
1634 | * count or making a pending reservation | |
1635 | * where needed | |
1636 | * | |
1637 | * @inode - file containing the newly added block | |
1638 | * @lblk - logical block to be added | |
1639 | * | |
1640 | * Returns 0 on success, negative error code on failure. | |
1641 | */ | |
1642 | static int ext4_insert_delayed_block(struct inode *inode, ext4_lblk_t lblk) | |
1643 | { | |
1644 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
1645 | int ret; | |
1646 | bool allocated = false; | |
1647 | ||
1648 | /* | |
1649 | * If the cluster containing lblk is shared with a delayed, | |
1650 | * written, or unwritten extent in a bigalloc file system, it's | |
1651 | * already been accounted for and does not need to be reserved. | |
1652 | * A pending reservation must be made for the cluster if it's | |
1653 | * shared with a written or unwritten extent and doesn't already | |
1654 | * have one. Written and unwritten extents can be purged from the | |
1655 | * extents status tree if the system is under memory pressure, so | |
1656 | * it's necessary to examine the extent tree if a search of the | |
1657 | * extents status tree doesn't get a match. | |
1658 | */ | |
1659 | if (sbi->s_cluster_ratio == 1) { | |
1660 | ret = ext4_da_reserve_space(inode); | |
1661 | if (ret != 0) /* ENOSPC */ | |
1662 | goto errout; | |
1663 | } else { /* bigalloc */ | |
1664 | if (!ext4_es_scan_clu(inode, &ext4_es_is_delonly, lblk)) { | |
1665 | if (!ext4_es_scan_clu(inode, | |
1666 | &ext4_es_is_mapped, lblk)) { | |
1667 | ret = ext4_clu_mapped(inode, | |
1668 | EXT4_B2C(sbi, lblk)); | |
1669 | if (ret < 0) | |
1670 | goto errout; | |
1671 | if (ret == 0) { | |
1672 | ret = ext4_da_reserve_space(inode); | |
1673 | if (ret != 0) /* ENOSPC */ | |
1674 | goto errout; | |
1675 | } else { | |
1676 | allocated = true; | |
1677 | } | |
1678 | } else { | |
1679 | allocated = true; | |
1680 | } | |
1681 | } | |
1682 | } | |
1683 | ||
1684 | ret = ext4_es_insert_delayed_block(inode, lblk, allocated); | |
1685 | ||
1686 | errout: | |
1687 | return ret; | |
1688 | } | |
1689 | ||
5356f261 AK |
1690 | /* |
1691 | * This function is grabs code from the very beginning of | |
1692 | * ext4_map_blocks, but assumes that the caller is from delayed write | |
1693 | * time. This function looks up the requested blocks and sets the | |
1694 | * buffer delay bit under the protection of i_data_sem. | |
1695 | */ | |
1696 | static int ext4_da_map_blocks(struct inode *inode, sector_t iblock, | |
1697 | struct ext4_map_blocks *map, | |
1698 | struct buffer_head *bh) | |
1699 | { | |
d100eef2 | 1700 | struct extent_status es; |
5356f261 AK |
1701 | int retval; |
1702 | sector_t invalid_block = ~((sector_t) 0xffff); | |
921f266b DM |
1703 | #ifdef ES_AGGRESSIVE_TEST |
1704 | struct ext4_map_blocks orig_map; | |
1705 | ||
1706 | memcpy(&orig_map, map, sizeof(*map)); | |
1707 | #endif | |
5356f261 AK |
1708 | |
1709 | if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es)) | |
1710 | invalid_block = ~0; | |
1711 | ||
1712 | map->m_flags = 0; | |
70aa1554 | 1713 | ext_debug(inode, "max_blocks %u, logical block %lu\n", map->m_len, |
5356f261 | 1714 | (unsigned long) map->m_lblk); |
d100eef2 ZL |
1715 | |
1716 | /* Lookup extent status tree firstly */ | |
bb5835ed | 1717 | if (ext4_es_lookup_extent(inode, iblock, NULL, &es)) { |
d100eef2 ZL |
1718 | if (ext4_es_is_hole(&es)) { |
1719 | retval = 0; | |
c8b459f4 | 1720 | down_read(&EXT4_I(inode)->i_data_sem); |
d100eef2 ZL |
1721 | goto add_delayed; |
1722 | } | |
1723 | ||
1724 | /* | |
1725 | * Delayed extent could be allocated by fallocate. | |
1726 | * So we need to check it. | |
1727 | */ | |
1728 | if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) { | |
1729 | map_bh(bh, inode->i_sb, invalid_block); | |
1730 | set_buffer_new(bh); | |
1731 | set_buffer_delay(bh); | |
1732 | return 0; | |
1733 | } | |
1734 | ||
1735 | map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk; | |
1736 | retval = es.es_len - (iblock - es.es_lblk); | |
1737 | if (retval > map->m_len) | |
1738 | retval = map->m_len; | |
1739 | map->m_len = retval; | |
1740 | if (ext4_es_is_written(&es)) | |
1741 | map->m_flags |= EXT4_MAP_MAPPED; | |
1742 | else if (ext4_es_is_unwritten(&es)) | |
1743 | map->m_flags |= EXT4_MAP_UNWRITTEN; | |
1744 | else | |
1e83bc81 | 1745 | BUG(); |
d100eef2 | 1746 | |
921f266b DM |
1747 | #ifdef ES_AGGRESSIVE_TEST |
1748 | ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0); | |
1749 | #endif | |
d100eef2 ZL |
1750 | return retval; |
1751 | } | |
1752 | ||
5356f261 AK |
1753 | /* |
1754 | * Try to see if we can get the block without requesting a new | |
1755 | * file system block. | |
1756 | */ | |
c8b459f4 | 1757 | down_read(&EXT4_I(inode)->i_data_sem); |
cbd7584e | 1758 | if (ext4_has_inline_data(inode)) |
9c3569b5 | 1759 | retval = 0; |
cbd7584e | 1760 | else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
2f8e0a7c | 1761 | retval = ext4_ext_map_blocks(NULL, inode, map, 0); |
5356f261 | 1762 | else |
2f8e0a7c | 1763 | retval = ext4_ind_map_blocks(NULL, inode, map, 0); |
5356f261 | 1764 | |
d100eef2 | 1765 | add_delayed: |
5356f261 | 1766 | if (retval == 0) { |
f7fec032 | 1767 | int ret; |
ad431025 | 1768 | |
5356f261 AK |
1769 | /* |
1770 | * XXX: __block_prepare_write() unmaps passed block, | |
1771 | * is it OK? | |
1772 | */ | |
5356f261 | 1773 | |
0b02f4c0 EW |
1774 | ret = ext4_insert_delayed_block(inode, map->m_lblk); |
1775 | if (ret != 0) { | |
f7fec032 | 1776 | retval = ret; |
51865fda | 1777 | goto out_unlock; |
f7fec032 | 1778 | } |
51865fda | 1779 | |
5356f261 AK |
1780 | map_bh(bh, inode->i_sb, invalid_block); |
1781 | set_buffer_new(bh); | |
1782 | set_buffer_delay(bh); | |
f7fec032 ZL |
1783 | } else if (retval > 0) { |
1784 | int ret; | |
3be78c73 | 1785 | unsigned int status; |
f7fec032 | 1786 | |
44fb851d ZL |
1787 | if (unlikely(retval != map->m_len)) { |
1788 | ext4_warning(inode->i_sb, | |
1789 | "ES len assertion failed for inode " | |
1790 | "%lu: retval %d != map->m_len %d", | |
1791 | inode->i_ino, retval, map->m_len); | |
1792 | WARN_ON(1); | |
921f266b | 1793 | } |
921f266b | 1794 | |
f7fec032 ZL |
1795 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
1796 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; | |
1797 | ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, | |
1798 | map->m_pblk, status); | |
1799 | if (ret != 0) | |
1800 | retval = ret; | |
5356f261 AK |
1801 | } |
1802 | ||
1803 | out_unlock: | |
1804 | up_read((&EXT4_I(inode)->i_data_sem)); | |
1805 | ||
1806 | return retval; | |
1807 | } | |
1808 | ||
64769240 | 1809 | /* |
d91bd2c1 | 1810 | * This is a special get_block_t callback which is used by |
b920c755 TT |
1811 | * ext4_da_write_begin(). It will either return mapped block or |
1812 | * reserve space for a single block. | |
29fa89d0 AK |
1813 | * |
1814 | * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set. | |
1815 | * We also have b_blocknr = -1 and b_bdev initialized properly | |
1816 | * | |
1817 | * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set. | |
1818 | * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev | |
1819 | * initialized properly. | |
64769240 | 1820 | */ |
9c3569b5 TM |
1821 | int ext4_da_get_block_prep(struct inode *inode, sector_t iblock, |
1822 | struct buffer_head *bh, int create) | |
64769240 | 1823 | { |
2ed88685 | 1824 | struct ext4_map_blocks map; |
64769240 AT |
1825 | int ret = 0; |
1826 | ||
1827 | BUG_ON(create == 0); | |
2ed88685 TT |
1828 | BUG_ON(bh->b_size != inode->i_sb->s_blocksize); |
1829 | ||
1830 | map.m_lblk = iblock; | |
1831 | map.m_len = 1; | |
64769240 AT |
1832 | |
1833 | /* | |
1834 | * first, we need to know whether the block is allocated already | |
1835 | * preallocated blocks are unmapped but should treated | |
1836 | * the same as allocated blocks. | |
1837 | */ | |
5356f261 AK |
1838 | ret = ext4_da_map_blocks(inode, iblock, &map, bh); |
1839 | if (ret <= 0) | |
2ed88685 | 1840 | return ret; |
64769240 | 1841 | |
2ed88685 | 1842 | map_bh(bh, inode->i_sb, map.m_pblk); |
ed8ad838 | 1843 | ext4_update_bh_state(bh, map.m_flags); |
2ed88685 TT |
1844 | |
1845 | if (buffer_unwritten(bh)) { | |
1846 | /* A delayed write to unwritten bh should be marked | |
1847 | * new and mapped. Mapped ensures that we don't do | |
1848 | * get_block multiple times when we write to the same | |
1849 | * offset and new ensures that we do proper zero out | |
1850 | * for partial write. | |
1851 | */ | |
1852 | set_buffer_new(bh); | |
c8205636 | 1853 | set_buffer_mapped(bh); |
2ed88685 TT |
1854 | } |
1855 | return 0; | |
64769240 | 1856 | } |
61628a3f | 1857 | |
188c299e JK |
1858 | static int bget_one(handle_t *handle, struct inode *inode, |
1859 | struct buffer_head *bh) | |
62e086be AK |
1860 | { |
1861 | get_bh(bh); | |
1862 | return 0; | |
1863 | } | |
1864 | ||
188c299e JK |
1865 | static int bput_one(handle_t *handle, struct inode *inode, |
1866 | struct buffer_head *bh) | |
62e086be AK |
1867 | { |
1868 | put_bh(bh); | |
1869 | return 0; | |
1870 | } | |
1871 | ||
1872 | static int __ext4_journalled_writepage(struct page *page, | |
62e086be AK |
1873 | unsigned int len) |
1874 | { | |
1875 | struct address_space *mapping = page->mapping; | |
1876 | struct inode *inode = mapping->host; | |
3fdcfb66 | 1877 | struct buffer_head *page_bufs = NULL; |
62e086be | 1878 | handle_t *handle = NULL; |
3fdcfb66 TM |
1879 | int ret = 0, err = 0; |
1880 | int inline_data = ext4_has_inline_data(inode); | |
1881 | struct buffer_head *inode_bh = NULL; | |
62e086be | 1882 | |
cb20d518 | 1883 | ClearPageChecked(page); |
3fdcfb66 TM |
1884 | |
1885 | if (inline_data) { | |
1886 | BUG_ON(page->index != 0); | |
1887 | BUG_ON(len > ext4_get_max_inline_size(inode)); | |
1888 | inode_bh = ext4_journalled_write_inline_data(inode, len, page); | |
1889 | if (inode_bh == NULL) | |
1890 | goto out; | |
1891 | } else { | |
1892 | page_bufs = page_buffers(page); | |
1893 | if (!page_bufs) { | |
1894 | BUG(); | |
1895 | goto out; | |
1896 | } | |
188c299e | 1897 | ext4_walk_page_buffers(handle, inode, page_bufs, 0, len, |
3fdcfb66 TM |
1898 | NULL, bget_one); |
1899 | } | |
bdf96838 TT |
1900 | /* |
1901 | * We need to release the page lock before we start the | |
1902 | * journal, so grab a reference so the page won't disappear | |
1903 | * out from under us. | |
1904 | */ | |
1905 | get_page(page); | |
62e086be AK |
1906 | unlock_page(page); |
1907 | ||
9924a92a TT |
1908 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
1909 | ext4_writepage_trans_blocks(inode)); | |
62e086be AK |
1910 | if (IS_ERR(handle)) { |
1911 | ret = PTR_ERR(handle); | |
bdf96838 TT |
1912 | put_page(page); |
1913 | goto out_no_pagelock; | |
62e086be | 1914 | } |
441c8508 CW |
1915 | BUG_ON(!ext4_handle_valid(handle)); |
1916 | ||
bdf96838 TT |
1917 | lock_page(page); |
1918 | put_page(page); | |
1919 | if (page->mapping != mapping) { | |
1920 | /* The page got truncated from under us */ | |
1921 | ext4_journal_stop(handle); | |
1922 | ret = 0; | |
1923 | goto out; | |
1924 | } | |
1925 | ||
3fdcfb66 | 1926 | if (inline_data) { |
362eca70 | 1927 | ret = ext4_mark_inode_dirty(handle, inode); |
3fdcfb66 | 1928 | } else { |
188c299e JK |
1929 | ret = ext4_walk_page_buffers(handle, inode, page_bufs, 0, len, |
1930 | NULL, do_journal_get_write_access); | |
3fdcfb66 | 1931 | |
188c299e JK |
1932 | err = ext4_walk_page_buffers(handle, inode, page_bufs, 0, len, |
1933 | NULL, write_end_fn); | |
3fdcfb66 | 1934 | } |
afb585a9 MFO |
1935 | if (ret == 0) |
1936 | ret = err; | |
b5b18160 | 1937 | err = ext4_jbd2_inode_add_write(handle, inode, page_offset(page), len); |
62e086be AK |
1938 | if (ret == 0) |
1939 | ret = err; | |
2d859db3 | 1940 | EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; |
62e086be AK |
1941 | err = ext4_journal_stop(handle); |
1942 | if (!ret) | |
1943 | ret = err; | |
1944 | ||
19f5fb7a | 1945 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
62e086be | 1946 | out: |
bdf96838 TT |
1947 | unlock_page(page); |
1948 | out_no_pagelock: | |
c915fb80 | 1949 | if (!inline_data && page_bufs) |
188c299e | 1950 | ext4_walk_page_buffers(NULL, inode, page_bufs, 0, len, |
c915fb80 | 1951 | NULL, bput_one); |
3fdcfb66 | 1952 | brelse(inode_bh); |
62e086be AK |
1953 | return ret; |
1954 | } | |
1955 | ||
61628a3f | 1956 | /* |
43ce1d23 AK |
1957 | * Note that we don't need to start a transaction unless we're journaling data |
1958 | * because we should have holes filled from ext4_page_mkwrite(). We even don't | |
1959 | * need to file the inode to the transaction's list in ordered mode because if | |
1960 | * we are writing back data added by write(), the inode is already there and if | |
25985edc | 1961 | * we are writing back data modified via mmap(), no one guarantees in which |
43ce1d23 AK |
1962 | * transaction the data will hit the disk. In case we are journaling data, we |
1963 | * cannot start transaction directly because transaction start ranks above page | |
1964 | * lock so we have to do some magic. | |
1965 | * | |
b920c755 | 1966 | * This function can get called via... |
20970ba6 | 1967 | * - ext4_writepages after taking page lock (have journal handle) |
b920c755 | 1968 | * - journal_submit_inode_data_buffers (no journal handle) |
f6463b0d | 1969 | * - shrink_page_list via the kswapd/direct reclaim (no journal handle) |
b920c755 | 1970 | * - grab_page_cache when doing write_begin (have journal handle) |
43ce1d23 AK |
1971 | * |
1972 | * We don't do any block allocation in this function. If we have page with | |
1973 | * multiple blocks we need to write those buffer_heads that are mapped. This | |
1974 | * is important for mmaped based write. So if we do with blocksize 1K | |
1975 | * truncate(f, 1024); | |
1976 | * a = mmap(f, 0, 4096); | |
1977 | * a[0] = 'a'; | |
1978 | * truncate(f, 4096); | |
1979 | * we have in the page first buffer_head mapped via page_mkwrite call back | |
90802ed9 | 1980 | * but other buffer_heads would be unmapped but dirty (dirty done via the |
43ce1d23 AK |
1981 | * do_wp_page). So writepage should write the first block. If we modify |
1982 | * the mmap area beyond 1024 we will again get a page_fault and the | |
1983 | * page_mkwrite callback will do the block allocation and mark the | |
1984 | * buffer_heads mapped. | |
1985 | * | |
1986 | * We redirty the page if we have any buffer_heads that is either delay or | |
1987 | * unwritten in the page. | |
1988 | * | |
1989 | * We can get recursively called as show below. | |
1990 | * | |
1991 | * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() -> | |
1992 | * ext4_writepage() | |
1993 | * | |
1994 | * But since we don't do any block allocation we should not deadlock. | |
1995 | * Page also have the dirty flag cleared so we don't get recurive page_lock. | |
61628a3f | 1996 | */ |
43ce1d23 | 1997 | static int ext4_writepage(struct page *page, |
62e086be | 1998 | struct writeback_control *wbc) |
64769240 | 1999 | { |
f8bec370 | 2000 | int ret = 0; |
61628a3f | 2001 | loff_t size; |
498e5f24 | 2002 | unsigned int len; |
744692dc | 2003 | struct buffer_head *page_bufs = NULL; |
61628a3f | 2004 | struct inode *inode = page->mapping->host; |
36ade451 | 2005 | struct ext4_io_submit io_submit; |
1c8349a1 | 2006 | bool keep_towrite = false; |
61628a3f | 2007 | |
0db1ff22 | 2008 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) { |
c2a559bc | 2009 | inode->i_mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE); |
0db1ff22 TT |
2010 | unlock_page(page); |
2011 | return -EIO; | |
2012 | } | |
2013 | ||
a9c667f8 | 2014 | trace_ext4_writepage(page); |
f0e6c985 | 2015 | size = i_size_read(inode); |
c93d8f88 EB |
2016 | if (page->index == size >> PAGE_SHIFT && |
2017 | !ext4_verity_in_progress(inode)) | |
09cbfeaf | 2018 | len = size & ~PAGE_MASK; |
f0e6c985 | 2019 | else |
09cbfeaf | 2020 | len = PAGE_SIZE; |
64769240 | 2021 | |
a42afc5f | 2022 | page_bufs = page_buffers(page); |
a42afc5f | 2023 | /* |
fe386132 JK |
2024 | * We cannot do block allocation or other extent handling in this |
2025 | * function. If there are buffers needing that, we have to redirty | |
2026 | * the page. But we may reach here when we do a journal commit via | |
2027 | * journal_submit_inode_data_buffers() and in that case we must write | |
2028 | * allocated buffers to achieve data=ordered mode guarantees. | |
cccd147a TT |
2029 | * |
2030 | * Also, if there is only one buffer per page (the fs block | |
2031 | * size == the page size), if one buffer needs block | |
2032 | * allocation or needs to modify the extent tree to clear the | |
2033 | * unwritten flag, we know that the page can't be written at | |
2034 | * all, so we might as well refuse the write immediately. | |
2035 | * Unfortunately if the block size != page size, we can't as | |
2036 | * easily detect this case using ext4_walk_page_buffers(), but | |
2037 | * for the extremely common case, this is an optimization that | |
2038 | * skips a useless round trip through ext4_bio_write_page(). | |
a42afc5f | 2039 | */ |
188c299e | 2040 | if (ext4_walk_page_buffers(NULL, inode, page_bufs, 0, len, NULL, |
f19d5870 | 2041 | ext4_bh_delay_or_unwritten)) { |
f8bec370 | 2042 | redirty_page_for_writepage(wbc, page); |
cccd147a | 2043 | if ((current->flags & PF_MEMALLOC) || |
09cbfeaf | 2044 | (inode->i_sb->s_blocksize == PAGE_SIZE)) { |
fe386132 JK |
2045 | /* |
2046 | * For memory cleaning there's no point in writing only | |
2047 | * some buffers. So just bail out. Warn if we came here | |
2048 | * from direct reclaim. | |
2049 | */ | |
2050 | WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) | |
2051 | == PF_MEMALLOC); | |
f0e6c985 AK |
2052 | unlock_page(page); |
2053 | return 0; | |
2054 | } | |
1c8349a1 | 2055 | keep_towrite = true; |
a42afc5f | 2056 | } |
64769240 | 2057 | |
cb20d518 | 2058 | if (PageChecked(page) && ext4_should_journal_data(inode)) |
43ce1d23 AK |
2059 | /* |
2060 | * It's mmapped pagecache. Add buffers and journal it. There | |
2061 | * doesn't seem much point in redirtying the page here. | |
2062 | */ | |
3f0ca309 | 2063 | return __ext4_journalled_writepage(page, len); |
43ce1d23 | 2064 | |
97a851ed JK |
2065 | ext4_io_submit_init(&io_submit, wbc); |
2066 | io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS); | |
2067 | if (!io_submit.io_end) { | |
2068 | redirty_page_for_writepage(wbc, page); | |
2069 | unlock_page(page); | |
2070 | return -ENOMEM; | |
2071 | } | |
be993933 | 2072 | ret = ext4_bio_write_page(&io_submit, page, len, keep_towrite); |
36ade451 | 2073 | ext4_io_submit(&io_submit); |
97a851ed JK |
2074 | /* Drop io_end reference we got from init */ |
2075 | ext4_put_io_end_defer(io_submit.io_end); | |
64769240 AT |
2076 | return ret; |
2077 | } | |
2078 | ||
5f1132b2 JK |
2079 | static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page) |
2080 | { | |
2081 | int len; | |
a056bdaa | 2082 | loff_t size; |
5f1132b2 JK |
2083 | int err; |
2084 | ||
2085 | BUG_ON(page->index != mpd->first_page); | |
a056bdaa JK |
2086 | clear_page_dirty_for_io(page); |
2087 | /* | |
2088 | * We have to be very careful here! Nothing protects writeback path | |
2089 | * against i_size changes and the page can be writeably mapped into | |
2090 | * page tables. So an application can be growing i_size and writing | |
2091 | * data through mmap while writeback runs. clear_page_dirty_for_io() | |
2092 | * write-protects our page in page tables and the page cannot get | |
2093 | * written to again until we release page lock. So only after | |
2094 | * clear_page_dirty_for_io() we are safe to sample i_size for | |
2095 | * ext4_bio_write_page() to zero-out tail of the written page. We rely | |
2096 | * on the barrier provided by TestClearPageDirty in | |
2097 | * clear_page_dirty_for_io() to make sure i_size is really sampled only | |
2098 | * after page tables are updated. | |
2099 | */ | |
2100 | size = i_size_read(mpd->inode); | |
c93d8f88 EB |
2101 | if (page->index == size >> PAGE_SHIFT && |
2102 | !ext4_verity_in_progress(mpd->inode)) | |
09cbfeaf | 2103 | len = size & ~PAGE_MASK; |
5f1132b2 | 2104 | else |
09cbfeaf | 2105 | len = PAGE_SIZE; |
be993933 | 2106 | err = ext4_bio_write_page(&mpd->io_submit, page, len, false); |
5f1132b2 JK |
2107 | if (!err) |
2108 | mpd->wbc->nr_to_write--; | |
2109 | mpd->first_page++; | |
2110 | ||
2111 | return err; | |
2112 | } | |
2113 | ||
6db07461 | 2114 | #define BH_FLAGS (BIT(BH_Unwritten) | BIT(BH_Delay)) |
4e7ea81d | 2115 | |
61628a3f | 2116 | /* |
fffb2739 JK |
2117 | * mballoc gives us at most this number of blocks... |
2118 | * XXX: That seems to be only a limitation of ext4_mb_normalize_request(). | |
70261f56 | 2119 | * The rest of mballoc seems to handle chunks up to full group size. |
61628a3f | 2120 | */ |
fffb2739 | 2121 | #define MAX_WRITEPAGES_EXTENT_LEN 2048 |
525f4ed8 | 2122 | |
4e7ea81d JK |
2123 | /* |
2124 | * mpage_add_bh_to_extent - try to add bh to extent of blocks to map | |
2125 | * | |
2126 | * @mpd - extent of blocks | |
2127 | * @lblk - logical number of the block in the file | |
09930042 | 2128 | * @bh - buffer head we want to add to the extent |
4e7ea81d | 2129 | * |
09930042 JK |
2130 | * The function is used to collect contig. blocks in the same state. If the |
2131 | * buffer doesn't require mapping for writeback and we haven't started the | |
2132 | * extent of buffers to map yet, the function returns 'true' immediately - the | |
2133 | * caller can write the buffer right away. Otherwise the function returns true | |
2134 | * if the block has been added to the extent, false if the block couldn't be | |
2135 | * added. | |
4e7ea81d | 2136 | */ |
09930042 JK |
2137 | static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk, |
2138 | struct buffer_head *bh) | |
4e7ea81d JK |
2139 | { |
2140 | struct ext4_map_blocks *map = &mpd->map; | |
2141 | ||
09930042 JK |
2142 | /* Buffer that doesn't need mapping for writeback? */ |
2143 | if (!buffer_dirty(bh) || !buffer_mapped(bh) || | |
2144 | (!buffer_delay(bh) && !buffer_unwritten(bh))) { | |
2145 | /* So far no extent to map => we write the buffer right away */ | |
2146 | if (map->m_len == 0) | |
2147 | return true; | |
2148 | return false; | |
2149 | } | |
4e7ea81d JK |
2150 | |
2151 | /* First block in the extent? */ | |
2152 | if (map->m_len == 0) { | |
dddbd6ac JK |
2153 | /* We cannot map unless handle is started... */ |
2154 | if (!mpd->do_map) | |
2155 | return false; | |
4e7ea81d JK |
2156 | map->m_lblk = lblk; |
2157 | map->m_len = 1; | |
09930042 JK |
2158 | map->m_flags = bh->b_state & BH_FLAGS; |
2159 | return true; | |
4e7ea81d JK |
2160 | } |
2161 | ||
09930042 JK |
2162 | /* Don't go larger than mballoc is willing to allocate */ |
2163 | if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN) | |
2164 | return false; | |
2165 | ||
4e7ea81d JK |
2166 | /* Can we merge the block to our big extent? */ |
2167 | if (lblk == map->m_lblk + map->m_len && | |
09930042 | 2168 | (bh->b_state & BH_FLAGS) == map->m_flags) { |
4e7ea81d | 2169 | map->m_len++; |
09930042 | 2170 | return true; |
4e7ea81d | 2171 | } |
09930042 | 2172 | return false; |
4e7ea81d JK |
2173 | } |
2174 | ||
5f1132b2 JK |
2175 | /* |
2176 | * mpage_process_page_bufs - submit page buffers for IO or add them to extent | |
2177 | * | |
2178 | * @mpd - extent of blocks for mapping | |
2179 | * @head - the first buffer in the page | |
2180 | * @bh - buffer we should start processing from | |
2181 | * @lblk - logical number of the block in the file corresponding to @bh | |
2182 | * | |
2183 | * Walk through page buffers from @bh upto @head (exclusive) and either submit | |
2184 | * the page for IO if all buffers in this page were mapped and there's no | |
2185 | * accumulated extent of buffers to map or add buffers in the page to the | |
2186 | * extent of buffers to map. The function returns 1 if the caller can continue | |
2187 | * by processing the next page, 0 if it should stop adding buffers to the | |
2188 | * extent to map because we cannot extend it anymore. It can also return value | |
2189 | * < 0 in case of error during IO submission. | |
2190 | */ | |
2191 | static int mpage_process_page_bufs(struct mpage_da_data *mpd, | |
2192 | struct buffer_head *head, | |
2193 | struct buffer_head *bh, | |
2194 | ext4_lblk_t lblk) | |
4e7ea81d JK |
2195 | { |
2196 | struct inode *inode = mpd->inode; | |
5f1132b2 | 2197 | int err; |
93407472 | 2198 | ext4_lblk_t blocks = (i_size_read(inode) + i_blocksize(inode) - 1) |
4e7ea81d JK |
2199 | >> inode->i_blkbits; |
2200 | ||
c93d8f88 EB |
2201 | if (ext4_verity_in_progress(inode)) |
2202 | blocks = EXT_MAX_BLOCKS; | |
2203 | ||
4e7ea81d JK |
2204 | do { |
2205 | BUG_ON(buffer_locked(bh)); | |
2206 | ||
09930042 | 2207 | if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) { |
4e7ea81d JK |
2208 | /* Found extent to map? */ |
2209 | if (mpd->map.m_len) | |
5f1132b2 | 2210 | return 0; |
dddbd6ac JK |
2211 | /* Buffer needs mapping and handle is not started? */ |
2212 | if (!mpd->do_map) | |
2213 | return 0; | |
09930042 | 2214 | /* Everything mapped so far and we hit EOF */ |
5f1132b2 | 2215 | break; |
4e7ea81d | 2216 | } |
4e7ea81d | 2217 | } while (lblk++, (bh = bh->b_this_page) != head); |
5f1132b2 JK |
2218 | /* So far everything mapped? Submit the page for IO. */ |
2219 | if (mpd->map.m_len == 0) { | |
2220 | err = mpage_submit_page(mpd, head->b_page); | |
2221 | if (err < 0) | |
2222 | return err; | |
2223 | } | |
6b8ed620 JK |
2224 | if (lblk >= blocks) { |
2225 | mpd->scanned_until_end = 1; | |
2226 | return 0; | |
2227 | } | |
2228 | return 1; | |
4e7ea81d JK |
2229 | } |
2230 | ||
2943fdbc RH |
2231 | /* |
2232 | * mpage_process_page - update page buffers corresponding to changed extent and | |
2233 | * may submit fully mapped page for IO | |
2234 | * | |
2235 | * @mpd - description of extent to map, on return next extent to map | |
2236 | * @m_lblk - logical block mapping. | |
2237 | * @m_pblk - corresponding physical mapping. | |
2238 | * @map_bh - determines on return whether this page requires any further | |
2239 | * mapping or not. | |
2240 | * Scan given page buffers corresponding to changed extent and update buffer | |
2241 | * state according to new extent state. | |
2242 | * We map delalloc buffers to their physical location, clear unwritten bits. | |
2243 | * If the given page is not fully mapped, we update @map to the next extent in | |
2244 | * the given page that needs mapping & return @map_bh as true. | |
2245 | */ | |
2246 | static int mpage_process_page(struct mpage_da_data *mpd, struct page *page, | |
2247 | ext4_lblk_t *m_lblk, ext4_fsblk_t *m_pblk, | |
2248 | bool *map_bh) | |
2249 | { | |
2250 | struct buffer_head *head, *bh; | |
2251 | ext4_io_end_t *io_end = mpd->io_submit.io_end; | |
2252 | ext4_lblk_t lblk = *m_lblk; | |
2253 | ext4_fsblk_t pblock = *m_pblk; | |
2254 | int err = 0; | |
c8cc8816 RH |
2255 | int blkbits = mpd->inode->i_blkbits; |
2256 | ssize_t io_end_size = 0; | |
2257 | struct ext4_io_end_vec *io_end_vec = ext4_last_io_end_vec(io_end); | |
2943fdbc RH |
2258 | |
2259 | bh = head = page_buffers(page); | |
2260 | do { | |
2261 | if (lblk < mpd->map.m_lblk) | |
2262 | continue; | |
2263 | if (lblk >= mpd->map.m_lblk + mpd->map.m_len) { | |
2264 | /* | |
2265 | * Buffer after end of mapped extent. | |
2266 | * Find next buffer in the page to map. | |
2267 | */ | |
2268 | mpd->map.m_len = 0; | |
2269 | mpd->map.m_flags = 0; | |
c8cc8816 RH |
2270 | io_end_vec->size += io_end_size; |
2271 | io_end_size = 0; | |
2943fdbc | 2272 | |
2943fdbc RH |
2273 | err = mpage_process_page_bufs(mpd, head, bh, lblk); |
2274 | if (err > 0) | |
2275 | err = 0; | |
c8cc8816 RH |
2276 | if (!err && mpd->map.m_len && mpd->map.m_lblk > lblk) { |
2277 | io_end_vec = ext4_alloc_io_end_vec(io_end); | |
4d06bfb9 RH |
2278 | if (IS_ERR(io_end_vec)) { |
2279 | err = PTR_ERR(io_end_vec); | |
2280 | goto out; | |
2281 | } | |
d1e18b88 | 2282 | io_end_vec->offset = (loff_t)mpd->map.m_lblk << blkbits; |
c8cc8816 | 2283 | } |
2943fdbc RH |
2284 | *map_bh = true; |
2285 | goto out; | |
2286 | } | |
2287 | if (buffer_delay(bh)) { | |
2288 | clear_buffer_delay(bh); | |
2289 | bh->b_blocknr = pblock++; | |
2290 | } | |
2291 | clear_buffer_unwritten(bh); | |
c8cc8816 | 2292 | io_end_size += (1 << blkbits); |
2943fdbc | 2293 | } while (lblk++, (bh = bh->b_this_page) != head); |
c8cc8816 RH |
2294 | |
2295 | io_end_vec->size += io_end_size; | |
2296 | io_end_size = 0; | |
2943fdbc RH |
2297 | *map_bh = false; |
2298 | out: | |
2299 | *m_lblk = lblk; | |
2300 | *m_pblk = pblock; | |
2301 | return err; | |
2302 | } | |
2303 | ||
4e7ea81d JK |
2304 | /* |
2305 | * mpage_map_buffers - update buffers corresponding to changed extent and | |
2306 | * submit fully mapped pages for IO | |
2307 | * | |
2308 | * @mpd - description of extent to map, on return next extent to map | |
2309 | * | |
2310 | * Scan buffers corresponding to changed extent (we expect corresponding pages | |
2311 | * to be already locked) and update buffer state according to new extent state. | |
2312 | * We map delalloc buffers to their physical location, clear unwritten bits, | |
556615dc | 2313 | * and mark buffers as uninit when we perform writes to unwritten extents |
4e7ea81d JK |
2314 | * and do extent conversion after IO is finished. If the last page is not fully |
2315 | * mapped, we update @map to the next extent in the last page that needs | |
2316 | * mapping. Otherwise we submit the page for IO. | |
2317 | */ | |
2318 | static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd) | |
2319 | { | |
2320 | struct pagevec pvec; | |
2321 | int nr_pages, i; | |
2322 | struct inode *inode = mpd->inode; | |
09cbfeaf | 2323 | int bpp_bits = PAGE_SHIFT - inode->i_blkbits; |
4e7ea81d JK |
2324 | pgoff_t start, end; |
2325 | ext4_lblk_t lblk; | |
2943fdbc | 2326 | ext4_fsblk_t pblock; |
4e7ea81d | 2327 | int err; |
2943fdbc | 2328 | bool map_bh = false; |
4e7ea81d JK |
2329 | |
2330 | start = mpd->map.m_lblk >> bpp_bits; | |
2331 | end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits; | |
2332 | lblk = start << bpp_bits; | |
2333 | pblock = mpd->map.m_pblk; | |
2334 | ||
86679820 | 2335 | pagevec_init(&pvec); |
4e7ea81d | 2336 | while (start <= end) { |
2b85a617 | 2337 | nr_pages = pagevec_lookup_range(&pvec, inode->i_mapping, |
397162ff | 2338 | &start, end); |
4e7ea81d JK |
2339 | if (nr_pages == 0) |
2340 | break; | |
2341 | for (i = 0; i < nr_pages; i++) { | |
2342 | struct page *page = pvec.pages[i]; | |
2343 | ||
2943fdbc RH |
2344 | err = mpage_process_page(mpd, page, &lblk, &pblock, |
2345 | &map_bh); | |
4e7ea81d | 2346 | /* |
2943fdbc RH |
2347 | * If map_bh is true, means page may require further bh |
2348 | * mapping, or maybe the page was submitted for IO. | |
2349 | * So we return to call further extent mapping. | |
4e7ea81d | 2350 | */ |
39c0ae16 | 2351 | if (err < 0 || map_bh) |
2943fdbc | 2352 | goto out; |
4e7ea81d JK |
2353 | /* Page fully mapped - let IO run! */ |
2354 | err = mpage_submit_page(mpd, page); | |
2943fdbc RH |
2355 | if (err < 0) |
2356 | goto out; | |
4e7ea81d JK |
2357 | } |
2358 | pagevec_release(&pvec); | |
2359 | } | |
2360 | /* Extent fully mapped and matches with page boundary. We are done. */ | |
2361 | mpd->map.m_len = 0; | |
2362 | mpd->map.m_flags = 0; | |
2363 | return 0; | |
2943fdbc RH |
2364 | out: |
2365 | pagevec_release(&pvec); | |
2366 | return err; | |
4e7ea81d JK |
2367 | } |
2368 | ||
2369 | static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd) | |
2370 | { | |
2371 | struct inode *inode = mpd->inode; | |
2372 | struct ext4_map_blocks *map = &mpd->map; | |
2373 | int get_blocks_flags; | |
090f32ee | 2374 | int err, dioread_nolock; |
4e7ea81d JK |
2375 | |
2376 | trace_ext4_da_write_pages_extent(inode, map); | |
2377 | /* | |
2378 | * Call ext4_map_blocks() to allocate any delayed allocation blocks, or | |
556615dc | 2379 | * to convert an unwritten extent to be initialized (in the case |
4e7ea81d JK |
2380 | * where we have written into one or more preallocated blocks). It is |
2381 | * possible that we're going to need more metadata blocks than | |
2382 | * previously reserved. However we must not fail because we're in | |
2383 | * writeback and there is nothing we can do about it so it might result | |
2384 | * in data loss. So use reserved blocks to allocate metadata if | |
2385 | * possible. | |
2386 | * | |
754cfed6 TT |
2387 | * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if |
2388 | * the blocks in question are delalloc blocks. This indicates | |
2389 | * that the blocks and quotas has already been checked when | |
2390 | * the data was copied into the page cache. | |
4e7ea81d JK |
2391 | */ |
2392 | get_blocks_flags = EXT4_GET_BLOCKS_CREATE | | |
ee0876bc JK |
2393 | EXT4_GET_BLOCKS_METADATA_NOFAIL | |
2394 | EXT4_GET_BLOCKS_IO_SUBMIT; | |
090f32ee LC |
2395 | dioread_nolock = ext4_should_dioread_nolock(inode); |
2396 | if (dioread_nolock) | |
4e7ea81d | 2397 | get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT; |
6db07461 | 2398 | if (map->m_flags & BIT(BH_Delay)) |
4e7ea81d JK |
2399 | get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE; |
2400 | ||
2401 | err = ext4_map_blocks(handle, inode, map, get_blocks_flags); | |
2402 | if (err < 0) | |
2403 | return err; | |
090f32ee | 2404 | if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) { |
6b523df4 JK |
2405 | if (!mpd->io_submit.io_end->handle && |
2406 | ext4_handle_valid(handle)) { | |
2407 | mpd->io_submit.io_end->handle = handle->h_rsv_handle; | |
2408 | handle->h_rsv_handle = NULL; | |
2409 | } | |
3613d228 | 2410 | ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end); |
6b523df4 | 2411 | } |
4e7ea81d JK |
2412 | |
2413 | BUG_ON(map->m_len == 0); | |
4e7ea81d JK |
2414 | return 0; |
2415 | } | |
2416 | ||
2417 | /* | |
2418 | * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length | |
2419 | * mpd->len and submit pages underlying it for IO | |
2420 | * | |
2421 | * @handle - handle for journal operations | |
2422 | * @mpd - extent to map | |
7534e854 JK |
2423 | * @give_up_on_write - we set this to true iff there is a fatal error and there |
2424 | * is no hope of writing the data. The caller should discard | |
2425 | * dirty pages to avoid infinite loops. | |
4e7ea81d JK |
2426 | * |
2427 | * The function maps extent starting at mpd->lblk of length mpd->len. If it is | |
2428 | * delayed, blocks are allocated, if it is unwritten, we may need to convert | |
2429 | * them to initialized or split the described range from larger unwritten | |
2430 | * extent. Note that we need not map all the described range since allocation | |
2431 | * can return less blocks or the range is covered by more unwritten extents. We | |
2432 | * cannot map more because we are limited by reserved transaction credits. On | |
2433 | * the other hand we always make sure that the last touched page is fully | |
2434 | * mapped so that it can be written out (and thus forward progress is | |
2435 | * guaranteed). After mapping we submit all mapped pages for IO. | |
2436 | */ | |
2437 | static int mpage_map_and_submit_extent(handle_t *handle, | |
cb530541 TT |
2438 | struct mpage_da_data *mpd, |
2439 | bool *give_up_on_write) | |
4e7ea81d JK |
2440 | { |
2441 | struct inode *inode = mpd->inode; | |
2442 | struct ext4_map_blocks *map = &mpd->map; | |
2443 | int err; | |
2444 | loff_t disksize; | |
6603120e | 2445 | int progress = 0; |
c8cc8816 | 2446 | ext4_io_end_t *io_end = mpd->io_submit.io_end; |
4d06bfb9 | 2447 | struct ext4_io_end_vec *io_end_vec; |
4e7ea81d | 2448 | |
4d06bfb9 RH |
2449 | io_end_vec = ext4_alloc_io_end_vec(io_end); |
2450 | if (IS_ERR(io_end_vec)) | |
2451 | return PTR_ERR(io_end_vec); | |
c8cc8816 | 2452 | io_end_vec->offset = ((loff_t)map->m_lblk) << inode->i_blkbits; |
27d7c4ed | 2453 | do { |
4e7ea81d JK |
2454 | err = mpage_map_one_extent(handle, mpd); |
2455 | if (err < 0) { | |
2456 | struct super_block *sb = inode->i_sb; | |
2457 | ||
0db1ff22 | 2458 | if (ext4_forced_shutdown(EXT4_SB(sb)) || |
9b5f6c9b | 2459 | ext4_test_mount_flag(sb, EXT4_MF_FS_ABORTED)) |
cb530541 | 2460 | goto invalidate_dirty_pages; |
4e7ea81d | 2461 | /* |
cb530541 TT |
2462 | * Let the uper layers retry transient errors. |
2463 | * In the case of ENOSPC, if ext4_count_free_blocks() | |
2464 | * is non-zero, a commit should free up blocks. | |
4e7ea81d | 2465 | */ |
cb530541 | 2466 | if ((err == -ENOMEM) || |
6603120e DM |
2467 | (err == -ENOSPC && ext4_count_free_clusters(sb))) { |
2468 | if (progress) | |
2469 | goto update_disksize; | |
cb530541 | 2470 | return err; |
6603120e | 2471 | } |
cb530541 TT |
2472 | ext4_msg(sb, KERN_CRIT, |
2473 | "Delayed block allocation failed for " | |
2474 | "inode %lu at logical offset %llu with" | |
2475 | " max blocks %u with error %d", | |
2476 | inode->i_ino, | |
2477 | (unsigned long long)map->m_lblk, | |
2478 | (unsigned)map->m_len, -err); | |
2479 | ext4_msg(sb, KERN_CRIT, | |
2480 | "This should not happen!! Data will " | |
2481 | "be lost\n"); | |
2482 | if (err == -ENOSPC) | |
2483 | ext4_print_free_blocks(inode); | |
2484 | invalidate_dirty_pages: | |
2485 | *give_up_on_write = true; | |
4e7ea81d JK |
2486 | return err; |
2487 | } | |
6603120e | 2488 | progress = 1; |
4e7ea81d JK |
2489 | /* |
2490 | * Update buffer state, submit mapped pages, and get us new | |
2491 | * extent to map | |
2492 | */ | |
2493 | err = mpage_map_and_submit_buffers(mpd); | |
2494 | if (err < 0) | |
6603120e | 2495 | goto update_disksize; |
27d7c4ed | 2496 | } while (map->m_len); |
4e7ea81d | 2497 | |
6603120e | 2498 | update_disksize: |
622cad13 TT |
2499 | /* |
2500 | * Update on-disk size after IO is submitted. Races with | |
2501 | * truncate are avoided by checking i_size under i_data_sem. | |
2502 | */ | |
09cbfeaf | 2503 | disksize = ((loff_t)mpd->first_page) << PAGE_SHIFT; |
35df4299 | 2504 | if (disksize > READ_ONCE(EXT4_I(inode)->i_disksize)) { |
4e7ea81d | 2505 | int err2; |
622cad13 TT |
2506 | loff_t i_size; |
2507 | ||
2508 | down_write(&EXT4_I(inode)->i_data_sem); | |
2509 | i_size = i_size_read(inode); | |
2510 | if (disksize > i_size) | |
2511 | disksize = i_size; | |
2512 | if (disksize > EXT4_I(inode)->i_disksize) | |
2513 | EXT4_I(inode)->i_disksize = disksize; | |
622cad13 | 2514 | up_write(&EXT4_I(inode)->i_data_sem); |
b907f2d5 | 2515 | err2 = ext4_mark_inode_dirty(handle, inode); |
878520ac | 2516 | if (err2) { |
54d3adbc TT |
2517 | ext4_error_err(inode->i_sb, -err2, |
2518 | "Failed to mark inode %lu dirty", | |
2519 | inode->i_ino); | |
878520ac | 2520 | } |
4e7ea81d JK |
2521 | if (!err) |
2522 | err = err2; | |
2523 | } | |
2524 | return err; | |
2525 | } | |
2526 | ||
fffb2739 JK |
2527 | /* |
2528 | * Calculate the total number of credits to reserve for one writepages | |
20970ba6 | 2529 | * iteration. This is called from ext4_writepages(). We map an extent of |
70261f56 | 2530 | * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping |
fffb2739 JK |
2531 | * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN + |
2532 | * bpp - 1 blocks in bpp different extents. | |
2533 | */ | |
525f4ed8 MC |
2534 | static int ext4_da_writepages_trans_blocks(struct inode *inode) |
2535 | { | |
fffb2739 | 2536 | int bpp = ext4_journal_blocks_per_page(inode); |
525f4ed8 | 2537 | |
fffb2739 JK |
2538 | return ext4_meta_trans_blocks(inode, |
2539 | MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp); | |
525f4ed8 | 2540 | } |
61628a3f | 2541 | |
8e48dcfb | 2542 | /* |
4e7ea81d JK |
2543 | * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages |
2544 | * and underlying extent to map | |
2545 | * | |
2546 | * @mpd - where to look for pages | |
2547 | * | |
2548 | * Walk dirty pages in the mapping. If they are fully mapped, submit them for | |
2549 | * IO immediately. When we find a page which isn't mapped we start accumulating | |
2550 | * extent of buffers underlying these pages that needs mapping (formed by | |
2551 | * either delayed or unwritten buffers). We also lock the pages containing | |
2552 | * these buffers. The extent found is returned in @mpd structure (starting at | |
2553 | * mpd->lblk with length mpd->len blocks). | |
2554 | * | |
2555 | * Note that this function can attach bios to one io_end structure which are | |
2556 | * neither logically nor physically contiguous. Although it may seem as an | |
2557 | * unnecessary complication, it is actually inevitable in blocksize < pagesize | |
2558 | * case as we need to track IO to all buffers underlying a page in one io_end. | |
8e48dcfb | 2559 | */ |
4e7ea81d | 2560 | static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd) |
8e48dcfb | 2561 | { |
4e7ea81d JK |
2562 | struct address_space *mapping = mpd->inode->i_mapping; |
2563 | struct pagevec pvec; | |
2564 | unsigned int nr_pages; | |
aeac589a | 2565 | long left = mpd->wbc->nr_to_write; |
4e7ea81d JK |
2566 | pgoff_t index = mpd->first_page; |
2567 | pgoff_t end = mpd->last_page; | |
10bbd235 | 2568 | xa_mark_t tag; |
4e7ea81d JK |
2569 | int i, err = 0; |
2570 | int blkbits = mpd->inode->i_blkbits; | |
2571 | ext4_lblk_t lblk; | |
2572 | struct buffer_head *head; | |
8e48dcfb | 2573 | |
4e7ea81d | 2574 | if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages) |
5b41d924 ES |
2575 | tag = PAGECACHE_TAG_TOWRITE; |
2576 | else | |
2577 | tag = PAGECACHE_TAG_DIRTY; | |
2578 | ||
86679820 | 2579 | pagevec_init(&pvec); |
4e7ea81d JK |
2580 | mpd->map.m_len = 0; |
2581 | mpd->next_page = index; | |
4f01b02c | 2582 | while (index <= end) { |
dc7f3e86 | 2583 | nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end, |
67fd707f | 2584 | tag); |
8e48dcfb | 2585 | if (nr_pages == 0) |
6b8ed620 | 2586 | break; |
8e48dcfb TT |
2587 | |
2588 | for (i = 0; i < nr_pages; i++) { | |
2589 | struct page *page = pvec.pages[i]; | |
2590 | ||
aeac589a ML |
2591 | /* |
2592 | * Accumulated enough dirty pages? This doesn't apply | |
2593 | * to WB_SYNC_ALL mode. For integrity sync we have to | |
2594 | * keep going because someone may be concurrently | |
2595 | * dirtying pages, and we might have synced a lot of | |
2596 | * newly appeared dirty pages, but have not synced all | |
2597 | * of the old dirty pages. | |
2598 | */ | |
2599 | if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0) | |
2600 | goto out; | |
2601 | ||
4e7ea81d JK |
2602 | /* If we can't merge this page, we are done. */ |
2603 | if (mpd->map.m_len > 0 && mpd->next_page != page->index) | |
2604 | goto out; | |
78aaced3 | 2605 | |
8e48dcfb | 2606 | lock_page(page); |
8e48dcfb | 2607 | /* |
4e7ea81d JK |
2608 | * If the page is no longer dirty, or its mapping no |
2609 | * longer corresponds to inode we are writing (which | |
2610 | * means it has been truncated or invalidated), or the | |
2611 | * page is already under writeback and we are not doing | |
2612 | * a data integrity writeback, skip the page | |
8e48dcfb | 2613 | */ |
4f01b02c TT |
2614 | if (!PageDirty(page) || |
2615 | (PageWriteback(page) && | |
4e7ea81d | 2616 | (mpd->wbc->sync_mode == WB_SYNC_NONE)) || |
4f01b02c | 2617 | unlikely(page->mapping != mapping)) { |
8e48dcfb TT |
2618 | unlock_page(page); |
2619 | continue; | |
2620 | } | |
2621 | ||
7cb1a535 | 2622 | wait_on_page_writeback(page); |
8e48dcfb | 2623 | BUG_ON(PageWriteback(page)); |
8e48dcfb | 2624 | |
4e7ea81d | 2625 | if (mpd->map.m_len == 0) |
8eb9e5ce | 2626 | mpd->first_page = page->index; |
8eb9e5ce | 2627 | mpd->next_page = page->index + 1; |
f8bec370 | 2628 | /* Add all dirty buffers to mpd */ |
4e7ea81d | 2629 | lblk = ((ext4_lblk_t)page->index) << |
09cbfeaf | 2630 | (PAGE_SHIFT - blkbits); |
f8bec370 | 2631 | head = page_buffers(page); |
5f1132b2 JK |
2632 | err = mpage_process_page_bufs(mpd, head, head, lblk); |
2633 | if (err <= 0) | |
4e7ea81d | 2634 | goto out; |
5f1132b2 | 2635 | err = 0; |
aeac589a | 2636 | left--; |
8e48dcfb TT |
2637 | } |
2638 | pagevec_release(&pvec); | |
2639 | cond_resched(); | |
2640 | } | |
6b8ed620 | 2641 | mpd->scanned_until_end = 1; |
4f01b02c | 2642 | return 0; |
8eb9e5ce TT |
2643 | out: |
2644 | pagevec_release(&pvec); | |
4e7ea81d | 2645 | return err; |
8e48dcfb TT |
2646 | } |
2647 | ||
20970ba6 TT |
2648 | static int ext4_writepages(struct address_space *mapping, |
2649 | struct writeback_control *wbc) | |
64769240 | 2650 | { |
4e7ea81d JK |
2651 | pgoff_t writeback_index = 0; |
2652 | long nr_to_write = wbc->nr_to_write; | |
22208ded | 2653 | int range_whole = 0; |
4e7ea81d | 2654 | int cycled = 1; |
61628a3f | 2655 | handle_t *handle = NULL; |
df22291f | 2656 | struct mpage_da_data mpd; |
5e745b04 | 2657 | struct inode *inode = mapping->host; |
6b523df4 | 2658 | int needed_blocks, rsv_blocks = 0, ret = 0; |
5e745b04 | 2659 | struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb); |
1bce63d1 | 2660 | struct blk_plug plug; |
cb530541 | 2661 | bool give_up_on_write = false; |
61628a3f | 2662 | |
0db1ff22 TT |
2663 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
2664 | return -EIO; | |
2665 | ||
bbd55937 | 2666 | percpu_down_read(&sbi->s_writepages_rwsem); |
20970ba6 | 2667 | trace_ext4_writepages(inode, wbc); |
ba80b101 | 2668 | |
61628a3f MC |
2669 | /* |
2670 | * No pages to write? This is mainly a kludge to avoid starting | |
2671 | * a transaction for special inodes like journal inode on last iput() | |
2672 | * because that could violate lock ordering on umount | |
2673 | */ | |
a1d6cc56 | 2674 | if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) |
bbf023c7 | 2675 | goto out_writepages; |
2a21e37e | 2676 | |
20970ba6 | 2677 | if (ext4_should_journal_data(inode)) { |
043d20d1 | 2678 | ret = generic_writepages(mapping, wbc); |
bbf023c7 | 2679 | goto out_writepages; |
20970ba6 TT |
2680 | } |
2681 | ||
2a21e37e TT |
2682 | /* |
2683 | * If the filesystem has aborted, it is read-only, so return | |
2684 | * right away instead of dumping stack traces later on that | |
2685 | * will obscure the real source of the problem. We test | |
1751e8a6 | 2686 | * EXT4_MF_FS_ABORTED instead of sb->s_flag's SB_RDONLY because |
2a21e37e | 2687 | * the latter could be true if the filesystem is mounted |
20970ba6 | 2688 | * read-only, and in that case, ext4_writepages should |
2a21e37e TT |
2689 | * *never* be called, so if that ever happens, we would want |
2690 | * the stack trace. | |
2691 | */ | |
0db1ff22 | 2692 | if (unlikely(ext4_forced_shutdown(EXT4_SB(mapping->host->i_sb)) || |
9b5f6c9b | 2693 | ext4_test_mount_flag(inode->i_sb, EXT4_MF_FS_ABORTED))) { |
bbf023c7 ML |
2694 | ret = -EROFS; |
2695 | goto out_writepages; | |
2696 | } | |
2a21e37e | 2697 | |
4e7ea81d JK |
2698 | /* |
2699 | * If we have inline data and arrive here, it means that | |
2700 | * we will soon create the block for the 1st page, so | |
2701 | * we'd better clear the inline data here. | |
2702 | */ | |
2703 | if (ext4_has_inline_data(inode)) { | |
2704 | /* Just inode will be modified... */ | |
2705 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 1); | |
2706 | if (IS_ERR(handle)) { | |
2707 | ret = PTR_ERR(handle); | |
2708 | goto out_writepages; | |
2709 | } | |
2710 | BUG_ON(ext4_test_inode_state(inode, | |
2711 | EXT4_STATE_MAY_INLINE_DATA)); | |
2712 | ext4_destroy_inline_data(handle, inode); | |
2713 | ext4_journal_stop(handle); | |
2714 | } | |
2715 | ||
4e343231 | 2716 | if (ext4_should_dioread_nolock(inode)) { |
2717 | /* | |
2718 | * We may need to convert up to one extent per block in | |
2719 | * the page and we may dirty the inode. | |
2720 | */ | |
2721 | rsv_blocks = 1 + ext4_chunk_trans_blocks(inode, | |
2722 | PAGE_SIZE >> inode->i_blkbits); | |
2723 | } | |
2724 | ||
22208ded AK |
2725 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
2726 | range_whole = 1; | |
61628a3f | 2727 | |
2acf2c26 | 2728 | if (wbc->range_cyclic) { |
4e7ea81d JK |
2729 | writeback_index = mapping->writeback_index; |
2730 | if (writeback_index) | |
2acf2c26 | 2731 | cycled = 0; |
4e7ea81d JK |
2732 | mpd.first_page = writeback_index; |
2733 | mpd.last_page = -1; | |
5b41d924 | 2734 | } else { |
09cbfeaf KS |
2735 | mpd.first_page = wbc->range_start >> PAGE_SHIFT; |
2736 | mpd.last_page = wbc->range_end >> PAGE_SHIFT; | |
5b41d924 | 2737 | } |
a1d6cc56 | 2738 | |
4e7ea81d JK |
2739 | mpd.inode = inode; |
2740 | mpd.wbc = wbc; | |
2741 | ext4_io_submit_init(&mpd.io_submit, wbc); | |
2acf2c26 | 2742 | retry: |
6e6938b6 | 2743 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
4e7ea81d | 2744 | tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page); |
1bce63d1 | 2745 | blk_start_plug(&plug); |
dddbd6ac JK |
2746 | |
2747 | /* | |
2748 | * First writeback pages that don't need mapping - we can avoid | |
2749 | * starting a transaction unnecessarily and also avoid being blocked | |
2750 | * in the block layer on device congestion while having transaction | |
2751 | * started. | |
2752 | */ | |
2753 | mpd.do_map = 0; | |
6b8ed620 | 2754 | mpd.scanned_until_end = 0; |
dddbd6ac JK |
2755 | mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL); |
2756 | if (!mpd.io_submit.io_end) { | |
2757 | ret = -ENOMEM; | |
2758 | goto unplug; | |
2759 | } | |
2760 | ret = mpage_prepare_extent_to_map(&mpd); | |
a297b2fc XW |
2761 | /* Unlock pages we didn't use */ |
2762 | mpage_release_unused_pages(&mpd, false); | |
dddbd6ac JK |
2763 | /* Submit prepared bio */ |
2764 | ext4_io_submit(&mpd.io_submit); | |
2765 | ext4_put_io_end_defer(mpd.io_submit.io_end); | |
2766 | mpd.io_submit.io_end = NULL; | |
dddbd6ac JK |
2767 | if (ret < 0) |
2768 | goto unplug; | |
2769 | ||
6b8ed620 | 2770 | while (!mpd.scanned_until_end && wbc->nr_to_write > 0) { |
4e7ea81d JK |
2771 | /* For each extent of pages we use new io_end */ |
2772 | mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL); | |
2773 | if (!mpd.io_submit.io_end) { | |
2774 | ret = -ENOMEM; | |
2775 | break; | |
2776 | } | |
a1d6cc56 AK |
2777 | |
2778 | /* | |
4e7ea81d JK |
2779 | * We have two constraints: We find one extent to map and we |
2780 | * must always write out whole page (makes a difference when | |
2781 | * blocksize < pagesize) so that we don't block on IO when we | |
2782 | * try to write out the rest of the page. Journalled mode is | |
2783 | * not supported by delalloc. | |
a1d6cc56 AK |
2784 | */ |
2785 | BUG_ON(ext4_should_journal_data(inode)); | |
525f4ed8 | 2786 | needed_blocks = ext4_da_writepages_trans_blocks(inode); |
a1d6cc56 | 2787 | |
4e7ea81d | 2788 | /* start a new transaction */ |
6b523df4 JK |
2789 | handle = ext4_journal_start_with_reserve(inode, |
2790 | EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks); | |
61628a3f MC |
2791 | if (IS_ERR(handle)) { |
2792 | ret = PTR_ERR(handle); | |
1693918e | 2793 | ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: " |
fbe845dd | 2794 | "%ld pages, ino %lu; err %d", __func__, |
a1d6cc56 | 2795 | wbc->nr_to_write, inode->i_ino, ret); |
4e7ea81d JK |
2796 | /* Release allocated io_end */ |
2797 | ext4_put_io_end(mpd.io_submit.io_end); | |
dddbd6ac | 2798 | mpd.io_submit.io_end = NULL; |
4e7ea81d | 2799 | break; |
61628a3f | 2800 | } |
dddbd6ac | 2801 | mpd.do_map = 1; |
f63e6005 | 2802 | |
4e7ea81d JK |
2803 | trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc); |
2804 | ret = mpage_prepare_extent_to_map(&mpd); | |
6b8ed620 JK |
2805 | if (!ret && mpd.map.m_len) |
2806 | ret = mpage_map_and_submit_extent(handle, &mpd, | |
cb530541 | 2807 | &give_up_on_write); |
646caa9c JK |
2808 | /* |
2809 | * Caution: If the handle is synchronous, | |
2810 | * ext4_journal_stop() can wait for transaction commit | |
2811 | * to finish which may depend on writeback of pages to | |
2812 | * complete or on page lock to be released. In that | |
b483bb77 | 2813 | * case, we have to wait until after we have |
646caa9c JK |
2814 | * submitted all the IO, released page locks we hold, |
2815 | * and dropped io_end reference (for extent conversion | |
2816 | * to be able to complete) before stopping the handle. | |
2817 | */ | |
2818 | if (!ext4_handle_valid(handle) || handle->h_sync == 0) { | |
2819 | ext4_journal_stop(handle); | |
2820 | handle = NULL; | |
dddbd6ac | 2821 | mpd.do_map = 0; |
646caa9c | 2822 | } |
4e7ea81d | 2823 | /* Unlock pages we didn't use */ |
cb530541 | 2824 | mpage_release_unused_pages(&mpd, give_up_on_write); |
a297b2fc XW |
2825 | /* Submit prepared bio */ |
2826 | ext4_io_submit(&mpd.io_submit); | |
2827 | ||
646caa9c JK |
2828 | /* |
2829 | * Drop our io_end reference we got from init. We have | |
2830 | * to be careful and use deferred io_end finishing if | |
2831 | * we are still holding the transaction as we can | |
2832 | * release the last reference to io_end which may end | |
2833 | * up doing unwritten extent conversion. | |
2834 | */ | |
2835 | if (handle) { | |
2836 | ext4_put_io_end_defer(mpd.io_submit.io_end); | |
2837 | ext4_journal_stop(handle); | |
2838 | } else | |
2839 | ext4_put_io_end(mpd.io_submit.io_end); | |
dddbd6ac | 2840 | mpd.io_submit.io_end = NULL; |
4e7ea81d JK |
2841 | |
2842 | if (ret == -ENOSPC && sbi->s_journal) { | |
2843 | /* | |
2844 | * Commit the transaction which would | |
22208ded AK |
2845 | * free blocks released in the transaction |
2846 | * and try again | |
2847 | */ | |
df22291f | 2848 | jbd2_journal_force_commit_nested(sbi->s_journal); |
22208ded | 2849 | ret = 0; |
4e7ea81d JK |
2850 | continue; |
2851 | } | |
2852 | /* Fatal error - ENOMEM, EIO... */ | |
2853 | if (ret) | |
61628a3f | 2854 | break; |
a1d6cc56 | 2855 | } |
dddbd6ac | 2856 | unplug: |
1bce63d1 | 2857 | blk_finish_plug(&plug); |
9c12a831 | 2858 | if (!ret && !cycled && wbc->nr_to_write > 0) { |
2acf2c26 | 2859 | cycled = 1; |
4e7ea81d JK |
2860 | mpd.last_page = writeback_index - 1; |
2861 | mpd.first_page = 0; | |
2acf2c26 AK |
2862 | goto retry; |
2863 | } | |
22208ded AK |
2864 | |
2865 | /* Update index */ | |
22208ded AK |
2866 | if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) |
2867 | /* | |
4e7ea81d | 2868 | * Set the writeback_index so that range_cyclic |
22208ded AK |
2869 | * mode will write it back later |
2870 | */ | |
4e7ea81d | 2871 | mapping->writeback_index = mpd.first_page; |
a1d6cc56 | 2872 | |
61628a3f | 2873 | out_writepages: |
20970ba6 TT |
2874 | trace_ext4_writepages_result(inode, wbc, ret, |
2875 | nr_to_write - wbc->nr_to_write); | |
bbd55937 | 2876 | percpu_up_read(&sbi->s_writepages_rwsem); |
61628a3f | 2877 | return ret; |
64769240 AT |
2878 | } |
2879 | ||
5f0663bb DW |
2880 | static int ext4_dax_writepages(struct address_space *mapping, |
2881 | struct writeback_control *wbc) | |
2882 | { | |
2883 | int ret; | |
2884 | long nr_to_write = wbc->nr_to_write; | |
2885 | struct inode *inode = mapping->host; | |
2886 | struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb); | |
2887 | ||
2888 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) | |
2889 | return -EIO; | |
2890 | ||
bbd55937 | 2891 | percpu_down_read(&sbi->s_writepages_rwsem); |
5f0663bb DW |
2892 | trace_ext4_writepages(inode, wbc); |
2893 | ||
3f666c56 | 2894 | ret = dax_writeback_mapping_range(mapping, sbi->s_daxdev, wbc); |
5f0663bb DW |
2895 | trace_ext4_writepages_result(inode, wbc, ret, |
2896 | nr_to_write - wbc->nr_to_write); | |
bbd55937 | 2897 | percpu_up_read(&sbi->s_writepages_rwsem); |
5f0663bb DW |
2898 | return ret; |
2899 | } | |
2900 | ||
79f0be8d AK |
2901 | static int ext4_nonda_switch(struct super_block *sb) |
2902 | { | |
5c1ff336 | 2903 | s64 free_clusters, dirty_clusters; |
79f0be8d AK |
2904 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
2905 | ||
2906 | /* | |
2907 | * switch to non delalloc mode if we are running low | |
2908 | * on free block. The free block accounting via percpu | |
179f7ebf | 2909 | * counters can get slightly wrong with percpu_counter_batch getting |
79f0be8d AK |
2910 | * accumulated on each CPU without updating global counters |
2911 | * Delalloc need an accurate free block accounting. So switch | |
2912 | * to non delalloc when we are near to error range. | |
2913 | */ | |
5c1ff336 EW |
2914 | free_clusters = |
2915 | percpu_counter_read_positive(&sbi->s_freeclusters_counter); | |
2916 | dirty_clusters = | |
2917 | percpu_counter_read_positive(&sbi->s_dirtyclusters_counter); | |
00d4e736 TT |
2918 | /* |
2919 | * Start pushing delalloc when 1/2 of free blocks are dirty. | |
2920 | */ | |
5c1ff336 | 2921 | if (dirty_clusters && (free_clusters < 2 * dirty_clusters)) |
10ee27a0 | 2922 | try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE); |
00d4e736 | 2923 | |
5c1ff336 EW |
2924 | if (2 * free_clusters < 3 * dirty_clusters || |
2925 | free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) { | |
79f0be8d | 2926 | /* |
c8afb446 ES |
2927 | * free block count is less than 150% of dirty blocks |
2928 | * or free blocks is less than watermark | |
79f0be8d AK |
2929 | */ |
2930 | return 1; | |
2931 | } | |
2932 | return 0; | |
2933 | } | |
2934 | ||
0ff8947f ES |
2935 | /* We always reserve for an inode update; the superblock could be there too */ |
2936 | static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len) | |
2937 | { | |
e2b911c5 | 2938 | if (likely(ext4_has_feature_large_file(inode->i_sb))) |
0ff8947f ES |
2939 | return 1; |
2940 | ||
2941 | if (pos + len <= 0x7fffffffULL) | |
2942 | return 1; | |
2943 | ||
2944 | /* We might need to update the superblock to set LARGE_FILE */ | |
2945 | return 2; | |
2946 | } | |
2947 | ||
64769240 | 2948 | static int ext4_da_write_begin(struct file *file, struct address_space *mapping, |
de9a55b8 TT |
2949 | loff_t pos, unsigned len, unsigned flags, |
2950 | struct page **pagep, void **fsdata) | |
64769240 | 2951 | { |
72b8ab9d | 2952 | int ret, retries = 0; |
64769240 AT |
2953 | struct page *page; |
2954 | pgoff_t index; | |
64769240 AT |
2955 | struct inode *inode = mapping->host; |
2956 | handle_t *handle; | |
2957 | ||
0db1ff22 TT |
2958 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
2959 | return -EIO; | |
2960 | ||
09cbfeaf | 2961 | index = pos >> PAGE_SHIFT; |
79f0be8d | 2962 | |
c93d8f88 EB |
2963 | if (ext4_nonda_switch(inode->i_sb) || S_ISLNK(inode->i_mode) || |
2964 | ext4_verity_in_progress(inode)) { | |
79f0be8d AK |
2965 | *fsdata = (void *)FALL_BACK_TO_NONDELALLOC; |
2966 | return ext4_write_begin(file, mapping, pos, | |
2967 | len, flags, pagep, fsdata); | |
2968 | } | |
2969 | *fsdata = (void *)0; | |
9bffad1e | 2970 | trace_ext4_da_write_begin(inode, pos, len, flags); |
9c3569b5 TM |
2971 | |
2972 | if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { | |
2973 | ret = ext4_da_write_inline_data_begin(mapping, inode, | |
2974 | pos, len, flags, | |
2975 | pagep, fsdata); | |
2976 | if (ret < 0) | |
47564bfb TT |
2977 | return ret; |
2978 | if (ret == 1) | |
2979 | return 0; | |
9c3569b5 TM |
2980 | } |
2981 | ||
47564bfb TT |
2982 | /* |
2983 | * grab_cache_page_write_begin() can take a long time if the | |
2984 | * system is thrashing due to memory pressure, or if the page | |
2985 | * is being written back. So grab it first before we start | |
2986 | * the transaction handle. This also allows us to allocate | |
2987 | * the page (if needed) without using GFP_NOFS. | |
2988 | */ | |
2989 | retry_grab: | |
2990 | page = grab_cache_page_write_begin(mapping, index, flags); | |
2991 | if (!page) | |
2992 | return -ENOMEM; | |
2993 | unlock_page(page); | |
2994 | ||
64769240 AT |
2995 | /* |
2996 | * With delayed allocation, we don't log the i_disksize update | |
2997 | * if there is delayed block allocation. But we still need | |
2998 | * to journalling the i_disksize update if writes to the end | |
2999 | * of file which has an already mapped buffer. | |
3000 | */ | |
47564bfb | 3001 | retry_journal: |
0ff8947f ES |
3002 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
3003 | ext4_da_write_credits(inode, pos, len)); | |
64769240 | 3004 | if (IS_ERR(handle)) { |
09cbfeaf | 3005 | put_page(page); |
47564bfb | 3006 | return PTR_ERR(handle); |
64769240 AT |
3007 | } |
3008 | ||
47564bfb TT |
3009 | lock_page(page); |
3010 | if (page->mapping != mapping) { | |
3011 | /* The page got truncated from under us */ | |
3012 | unlock_page(page); | |
09cbfeaf | 3013 | put_page(page); |
d5a0d4f7 | 3014 | ext4_journal_stop(handle); |
47564bfb | 3015 | goto retry_grab; |
d5a0d4f7 | 3016 | } |
47564bfb | 3017 | /* In case writeback began while the page was unlocked */ |
7afe5aa5 | 3018 | wait_for_stable_page(page); |
64769240 | 3019 | |
643fa961 | 3020 | #ifdef CONFIG_FS_ENCRYPTION |
2058f83a MH |
3021 | ret = ext4_block_write_begin(page, pos, len, |
3022 | ext4_da_get_block_prep); | |
3023 | #else | |
6e1db88d | 3024 | ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep); |
2058f83a | 3025 | #endif |
64769240 AT |
3026 | if (ret < 0) { |
3027 | unlock_page(page); | |
3028 | ext4_journal_stop(handle); | |
ae4d5372 AK |
3029 | /* |
3030 | * block_write_begin may have instantiated a few blocks | |
3031 | * outside i_size. Trim these off again. Don't need | |
3032 | * i_size_read because we hold i_mutex. | |
3033 | */ | |
3034 | if (pos + len > inode->i_size) | |
b9a4207d | 3035 | ext4_truncate_failed_write(inode); |
47564bfb TT |
3036 | |
3037 | if (ret == -ENOSPC && | |
3038 | ext4_should_retry_alloc(inode->i_sb, &retries)) | |
3039 | goto retry_journal; | |
3040 | ||
09cbfeaf | 3041 | put_page(page); |
47564bfb | 3042 | return ret; |
64769240 AT |
3043 | } |
3044 | ||
47564bfb | 3045 | *pagep = page; |
64769240 AT |
3046 | return ret; |
3047 | } | |
3048 | ||
632eaeab MC |
3049 | /* |
3050 | * Check if we should update i_disksize | |
3051 | * when write to the end of file but not require block allocation | |
3052 | */ | |
3053 | static int ext4_da_should_update_i_disksize(struct page *page, | |
de9a55b8 | 3054 | unsigned long offset) |
632eaeab MC |
3055 | { |
3056 | struct buffer_head *bh; | |
3057 | struct inode *inode = page->mapping->host; | |
3058 | unsigned int idx; | |
3059 | int i; | |
3060 | ||
3061 | bh = page_buffers(page); | |
3062 | idx = offset >> inode->i_blkbits; | |
3063 | ||
af5bc92d | 3064 | for (i = 0; i < idx; i++) |
632eaeab MC |
3065 | bh = bh->b_this_page; |
3066 | ||
29fa89d0 | 3067 | if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh)) |
632eaeab MC |
3068 | return 0; |
3069 | return 1; | |
3070 | } | |
3071 | ||
64769240 | 3072 | static int ext4_da_write_end(struct file *file, |
de9a55b8 TT |
3073 | struct address_space *mapping, |
3074 | loff_t pos, unsigned len, unsigned copied, | |
3075 | struct page *page, void *fsdata) | |
64769240 AT |
3076 | { |
3077 | struct inode *inode = mapping->host; | |
3078 | int ret = 0, ret2; | |
3079 | handle_t *handle = ext4_journal_current_handle(); | |
3080 | loff_t new_i_size; | |
632eaeab | 3081 | unsigned long start, end; |
79f0be8d AK |
3082 | int write_mode = (int)(unsigned long)fsdata; |
3083 | ||
74d553aa TT |
3084 | if (write_mode == FALL_BACK_TO_NONDELALLOC) |
3085 | return ext4_write_end(file, mapping, pos, | |
3086 | len, copied, page, fsdata); | |
632eaeab | 3087 | |
9bffad1e | 3088 | trace_ext4_da_write_end(inode, pos, len, copied); |
09cbfeaf | 3089 | start = pos & (PAGE_SIZE - 1); |
af5bc92d | 3090 | end = start + copied - 1; |
64769240 AT |
3091 | |
3092 | /* | |
3093 | * generic_write_end() will run mark_inode_dirty() if i_size | |
3094 | * changes. So let's piggyback the i_disksize mark_inode_dirty | |
3095 | * into that. | |
3096 | */ | |
64769240 | 3097 | new_i_size = pos + copied; |
ea51d132 | 3098 | if (copied && new_i_size > EXT4_I(inode)->i_disksize) { |
9c3569b5 TM |
3099 | if (ext4_has_inline_data(inode) || |
3100 | ext4_da_should_update_i_disksize(page, end)) { | |
ee124d27 | 3101 | ext4_update_i_disksize(inode, new_i_size); |
cf17fea6 AK |
3102 | /* We need to mark inode dirty even if |
3103 | * new_i_size is less that inode->i_size | |
3104 | * bu greater than i_disksize.(hint delalloc) | |
3105 | */ | |
4209ae12 | 3106 | ret = ext4_mark_inode_dirty(handle, inode); |
64769240 | 3107 | } |
632eaeab | 3108 | } |
9c3569b5 TM |
3109 | |
3110 | if (write_mode != CONVERT_INLINE_DATA && | |
3111 | ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) && | |
3112 | ext4_has_inline_data(inode)) | |
3113 | ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied, | |
3114 | page); | |
3115 | else | |
3116 | ret2 = generic_write_end(file, mapping, pos, len, copied, | |
64769240 | 3117 | page, fsdata); |
9c3569b5 | 3118 | |
64769240 AT |
3119 | copied = ret2; |
3120 | if (ret2 < 0) | |
3121 | ret = ret2; | |
3122 | ret2 = ext4_journal_stop(handle); | |
4209ae12 | 3123 | if (unlikely(ret2 && !ret)) |
64769240 AT |
3124 | ret = ret2; |
3125 | ||
3126 | return ret ? ret : copied; | |
3127 | } | |
3128 | ||
ccd2506b TT |
3129 | /* |
3130 | * Force all delayed allocation blocks to be allocated for a given inode. | |
3131 | */ | |
3132 | int ext4_alloc_da_blocks(struct inode *inode) | |
3133 | { | |
fb40ba0d TT |
3134 | trace_ext4_alloc_da_blocks(inode); |
3135 | ||
71d4f7d0 | 3136 | if (!EXT4_I(inode)->i_reserved_data_blocks) |
ccd2506b TT |
3137 | return 0; |
3138 | ||
3139 | /* | |
3140 | * We do something simple for now. The filemap_flush() will | |
3141 | * also start triggering a write of the data blocks, which is | |
3142 | * not strictly speaking necessary (and for users of | |
3143 | * laptop_mode, not even desirable). However, to do otherwise | |
3144 | * would require replicating code paths in: | |
de9a55b8 | 3145 | * |
20970ba6 | 3146 | * ext4_writepages() -> |
ccd2506b TT |
3147 | * write_cache_pages() ---> (via passed in callback function) |
3148 | * __mpage_da_writepage() --> | |
3149 | * mpage_add_bh_to_extent() | |
3150 | * mpage_da_map_blocks() | |
3151 | * | |
3152 | * The problem is that write_cache_pages(), located in | |
3153 | * mm/page-writeback.c, marks pages clean in preparation for | |
3154 | * doing I/O, which is not desirable if we're not planning on | |
3155 | * doing I/O at all. | |
3156 | * | |
3157 | * We could call write_cache_pages(), and then redirty all of | |
380cf090 | 3158 | * the pages by calling redirty_page_for_writepage() but that |
ccd2506b TT |
3159 | * would be ugly in the extreme. So instead we would need to |
3160 | * replicate parts of the code in the above functions, | |
25985edc | 3161 | * simplifying them because we wouldn't actually intend to |
ccd2506b TT |
3162 | * write out the pages, but rather only collect contiguous |
3163 | * logical block extents, call the multi-block allocator, and | |
3164 | * then update the buffer heads with the block allocations. | |
de9a55b8 | 3165 | * |
ccd2506b TT |
3166 | * For now, though, we'll cheat by calling filemap_flush(), |
3167 | * which will map the blocks, and start the I/O, but not | |
3168 | * actually wait for the I/O to complete. | |
3169 | */ | |
3170 | return filemap_flush(inode->i_mapping); | |
3171 | } | |
64769240 | 3172 | |
ac27a0ec DK |
3173 | /* |
3174 | * bmap() is special. It gets used by applications such as lilo and by | |
3175 | * the swapper to find the on-disk block of a specific piece of data. | |
3176 | * | |
3177 | * Naturally, this is dangerous if the block concerned is still in the | |
617ba13b | 3178 | * journal. If somebody makes a swapfile on an ext4 data-journaling |
ac27a0ec DK |
3179 | * filesystem and enables swap, then they may get a nasty shock when the |
3180 | * data getting swapped to that swapfile suddenly gets overwritten by | |
3181 | * the original zero's written out previously to the journal and | |
3182 | * awaiting writeback in the kernel's buffer cache. | |
3183 | * | |
3184 | * So, if we see any bmap calls here on a modified, data-journaled file, | |
3185 | * take extra steps to flush any blocks which might be in the cache. | |
3186 | */ | |
617ba13b | 3187 | static sector_t ext4_bmap(struct address_space *mapping, sector_t block) |
ac27a0ec DK |
3188 | { |
3189 | struct inode *inode = mapping->host; | |
3190 | journal_t *journal; | |
3191 | int err; | |
3192 | ||
46c7f254 TM |
3193 | /* |
3194 | * We can get here for an inline file via the FIBMAP ioctl | |
3195 | */ | |
3196 | if (ext4_has_inline_data(inode)) | |
3197 | return 0; | |
3198 | ||
64769240 AT |
3199 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) && |
3200 | test_opt(inode->i_sb, DELALLOC)) { | |
3201 | /* | |
3202 | * With delalloc we want to sync the file | |
3203 | * so that we can make sure we allocate | |
3204 | * blocks for file | |
3205 | */ | |
3206 | filemap_write_and_wait(mapping); | |
3207 | } | |
3208 | ||
19f5fb7a TT |
3209 | if (EXT4_JOURNAL(inode) && |
3210 | ext4_test_inode_state(inode, EXT4_STATE_JDATA)) { | |
ac27a0ec DK |
3211 | /* |
3212 | * This is a REALLY heavyweight approach, but the use of | |
3213 | * bmap on dirty files is expected to be extremely rare: | |
3214 | * only if we run lilo or swapon on a freshly made file | |
3215 | * do we expect this to happen. | |
3216 | * | |
3217 | * (bmap requires CAP_SYS_RAWIO so this does not | |
3218 | * represent an unprivileged user DOS attack --- we'd be | |
3219 | * in trouble if mortal users could trigger this path at | |
3220 | * will.) | |
3221 | * | |
617ba13b | 3222 | * NB. EXT4_STATE_JDATA is not set on files other than |
ac27a0ec DK |
3223 | * regular files. If somebody wants to bmap a directory |
3224 | * or symlink and gets confused because the buffer | |
3225 | * hasn't yet been flushed to disk, they deserve | |
3226 | * everything they get. | |
3227 | */ | |
3228 | ||
19f5fb7a | 3229 | ext4_clear_inode_state(inode, EXT4_STATE_JDATA); |
617ba13b | 3230 | journal = EXT4_JOURNAL(inode); |
dab291af | 3231 | jbd2_journal_lock_updates(journal); |
01d5d965 | 3232 | err = jbd2_journal_flush(journal, 0); |
dab291af | 3233 | jbd2_journal_unlock_updates(journal); |
ac27a0ec DK |
3234 | |
3235 | if (err) | |
3236 | return 0; | |
3237 | } | |
3238 | ||
ac58e4fb | 3239 | return iomap_bmap(mapping, block, &ext4_iomap_ops); |
ac27a0ec DK |
3240 | } |
3241 | ||
617ba13b | 3242 | static int ext4_readpage(struct file *file, struct page *page) |
ac27a0ec | 3243 | { |
46c7f254 TM |
3244 | int ret = -EAGAIN; |
3245 | struct inode *inode = page->mapping->host; | |
3246 | ||
0562e0ba | 3247 | trace_ext4_readpage(page); |
46c7f254 TM |
3248 | |
3249 | if (ext4_has_inline_data(inode)) | |
3250 | ret = ext4_readpage_inline(inode, page); | |
3251 | ||
3252 | if (ret == -EAGAIN) | |
a07f624b | 3253 | return ext4_mpage_readpages(inode, NULL, page); |
46c7f254 TM |
3254 | |
3255 | return ret; | |
ac27a0ec DK |
3256 | } |
3257 | ||
6311f91f | 3258 | static void ext4_readahead(struct readahead_control *rac) |
ac27a0ec | 3259 | { |
6311f91f | 3260 | struct inode *inode = rac->mapping->host; |
46c7f254 | 3261 | |
6311f91f | 3262 | /* If the file has inline data, no need to do readahead. */ |
46c7f254 | 3263 | if (ext4_has_inline_data(inode)) |
6311f91f | 3264 | return; |
46c7f254 | 3265 | |
a07f624b | 3266 | ext4_mpage_readpages(inode, rac, NULL); |
ac27a0ec DK |
3267 | } |
3268 | ||
d47992f8 LC |
3269 | static void ext4_invalidatepage(struct page *page, unsigned int offset, |
3270 | unsigned int length) | |
ac27a0ec | 3271 | { |
ca99fdd2 | 3272 | trace_ext4_invalidatepage(page, offset, length); |
0562e0ba | 3273 | |
4520fb3c JK |
3274 | /* No journalling happens on data buffers when this function is used */ |
3275 | WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page))); | |
3276 | ||
ca99fdd2 | 3277 | block_invalidatepage(page, offset, length); |
4520fb3c JK |
3278 | } |
3279 | ||
53e87268 | 3280 | static int __ext4_journalled_invalidatepage(struct page *page, |
ca99fdd2 LC |
3281 | unsigned int offset, |
3282 | unsigned int length) | |
4520fb3c JK |
3283 | { |
3284 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); | |
3285 | ||
ca99fdd2 | 3286 | trace_ext4_journalled_invalidatepage(page, offset, length); |
4520fb3c | 3287 | |
ac27a0ec DK |
3288 | /* |
3289 | * If it's a full truncate we just forget about the pending dirtying | |
3290 | */ | |
09cbfeaf | 3291 | if (offset == 0 && length == PAGE_SIZE) |
ac27a0ec DK |
3292 | ClearPageChecked(page); |
3293 | ||
ca99fdd2 | 3294 | return jbd2_journal_invalidatepage(journal, page, offset, length); |
53e87268 JK |
3295 | } |
3296 | ||
3297 | /* Wrapper for aops... */ | |
3298 | static void ext4_journalled_invalidatepage(struct page *page, | |
d47992f8 LC |
3299 | unsigned int offset, |
3300 | unsigned int length) | |
53e87268 | 3301 | { |
ca99fdd2 | 3302 | WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0); |
ac27a0ec DK |
3303 | } |
3304 | ||
617ba13b | 3305 | static int ext4_releasepage(struct page *page, gfp_t wait) |
ac27a0ec | 3306 | { |
617ba13b | 3307 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); |
ac27a0ec | 3308 | |
0562e0ba JZ |
3309 | trace_ext4_releasepage(page); |
3310 | ||
e1c36595 JK |
3311 | /* Page has dirty journalled data -> cannot release */ |
3312 | if (PageChecked(page)) | |
ac27a0ec | 3313 | return 0; |
0390131b | 3314 | if (journal) |
529a781e | 3315 | return jbd2_journal_try_to_free_buffers(journal, page); |
0390131b FM |
3316 | else |
3317 | return try_to_free_buffers(page); | |
ac27a0ec DK |
3318 | } |
3319 | ||
b8a6176c JK |
3320 | static bool ext4_inode_datasync_dirty(struct inode *inode) |
3321 | { | |
3322 | journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; | |
3323 | ||
aa75f4d3 HS |
3324 | if (journal) { |
3325 | if (jbd2_transaction_committed(journal, | |
d0520df7 AR |
3326 | EXT4_I(inode)->i_datasync_tid)) |
3327 | return false; | |
3328 | if (test_opt2(inode->i_sb, JOURNAL_FAST_COMMIT)) | |
1ceecb53 | 3329 | return !list_empty(&EXT4_I(inode)->i_fc_list); |
d0520df7 | 3330 | return true; |
aa75f4d3 HS |
3331 | } |
3332 | ||
b8a6176c JK |
3333 | /* Any metadata buffers to write? */ |
3334 | if (!list_empty(&inode->i_mapping->private_list)) | |
3335 | return true; | |
3336 | return inode->i_state & I_DIRTY_DATASYNC; | |
3337 | } | |
3338 | ||
c8fdfe29 MB |
3339 | static void ext4_set_iomap(struct inode *inode, struct iomap *iomap, |
3340 | struct ext4_map_blocks *map, loff_t offset, | |
3341 | loff_t length) | |
364443cb | 3342 | { |
c8fdfe29 | 3343 | u8 blkbits = inode->i_blkbits; |
364443cb | 3344 | |
c8fdfe29 MB |
3345 | /* |
3346 | * Writes that span EOF might trigger an I/O size update on completion, | |
3347 | * so consider them to be dirty for the purpose of O_DSYNC, even if | |
3348 | * there is no other metadata changes being made or are pending. | |
3349 | */ | |
364443cb | 3350 | iomap->flags = 0; |
c8fdfe29 MB |
3351 | if (ext4_inode_datasync_dirty(inode) || |
3352 | offset + length > i_size_read(inode)) | |
b8a6176c | 3353 | iomap->flags |= IOMAP_F_DIRTY; |
c8fdfe29 MB |
3354 | |
3355 | if (map->m_flags & EXT4_MAP_NEW) | |
3356 | iomap->flags |= IOMAP_F_NEW; | |
3357 | ||
5e405595 | 3358 | iomap->bdev = inode->i_sb->s_bdev; |
c8fdfe29 MB |
3359 | iomap->dax_dev = EXT4_SB(inode->i_sb)->s_daxdev; |
3360 | iomap->offset = (u64) map->m_lblk << blkbits; | |
3361 | iomap->length = (u64) map->m_len << blkbits; | |
364443cb | 3362 | |
6386722a RH |
3363 | if ((map->m_flags & EXT4_MAP_MAPPED) && |
3364 | !ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
3365 | iomap->flags |= IOMAP_F_MERGED; | |
3366 | ||
c8fdfe29 MB |
3367 | /* |
3368 | * Flags passed to ext4_map_blocks() for direct I/O writes can result | |
3369 | * in m_flags having both EXT4_MAP_MAPPED and EXT4_MAP_UNWRITTEN bits | |
3370 | * set. In order for any allocated unwritten extents to be converted | |
3371 | * into written extents correctly within the ->end_io() handler, we | |
3372 | * need to ensure that the iomap->type is set appropriately. Hence, the | |
3373 | * reason why we need to check whether the EXT4_MAP_UNWRITTEN bit has | |
3374 | * been set first. | |
3375 | */ | |
3376 | if (map->m_flags & EXT4_MAP_UNWRITTEN) { | |
3377 | iomap->type = IOMAP_UNWRITTEN; | |
3378 | iomap->addr = (u64) map->m_pblk << blkbits; | |
3379 | } else if (map->m_flags & EXT4_MAP_MAPPED) { | |
3380 | iomap->type = IOMAP_MAPPED; | |
3381 | iomap->addr = (u64) map->m_pblk << blkbits; | |
364443cb | 3382 | } else { |
c8fdfe29 MB |
3383 | iomap->type = IOMAP_HOLE; |
3384 | iomap->addr = IOMAP_NULL_ADDR; | |
364443cb | 3385 | } |
364443cb JK |
3386 | } |
3387 | ||
f063db5e MB |
3388 | static int ext4_iomap_alloc(struct inode *inode, struct ext4_map_blocks *map, |
3389 | unsigned int flags) | |
776722e8 | 3390 | { |
776722e8 | 3391 | handle_t *handle; |
378f32ba MB |
3392 | u8 blkbits = inode->i_blkbits; |
3393 | int ret, dio_credits, m_flags = 0, retries = 0; | |
776722e8 | 3394 | |
776722e8 | 3395 | /* |
f063db5e MB |
3396 | * Trim the mapping request to the maximum value that we can map at |
3397 | * once for direct I/O. | |
776722e8 | 3398 | */ |
f063db5e MB |
3399 | if (map->m_len > DIO_MAX_BLOCKS) |
3400 | map->m_len = DIO_MAX_BLOCKS; | |
3401 | dio_credits = ext4_chunk_trans_blocks(inode, map->m_len); | |
776722e8 | 3402 | |
f063db5e | 3403 | retry: |
776722e8 | 3404 | /* |
f063db5e MB |
3405 | * Either we allocate blocks and then don't get an unwritten extent, so |
3406 | * in that case we have reserved enough credits. Or, the blocks are | |
3407 | * already allocated and unwritten. In that case, the extent conversion | |
3408 | * fits into the credits as well. | |
776722e8 | 3409 | */ |
f063db5e MB |
3410 | handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits); |
3411 | if (IS_ERR(handle)) | |
3412 | return PTR_ERR(handle); | |
4c0425ff | 3413 | |
378f32ba MB |
3414 | /* |
3415 | * DAX and direct I/O are the only two operations that are currently | |
3416 | * supported with IOMAP_WRITE. | |
3417 | */ | |
3418 | WARN_ON(!IS_DAX(inode) && !(flags & IOMAP_DIRECT)); | |
3419 | if (IS_DAX(inode)) | |
3420 | m_flags = EXT4_GET_BLOCKS_CREATE_ZERO; | |
3421 | /* | |
3422 | * We use i_size instead of i_disksize here because delalloc writeback | |
3423 | * can complete at any point during the I/O and subsequently push the | |
3424 | * i_disksize out to i_size. This could be beyond where direct I/O is | |
3425 | * happening and thus expose allocated blocks to direct I/O reads. | |
3426 | */ | |
d0b040f5 | 3427 | else if (((loff_t)map->m_lblk << blkbits) >= i_size_read(inode)) |
378f32ba MB |
3428 | m_flags = EXT4_GET_BLOCKS_CREATE; |
3429 | else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
3430 | m_flags = EXT4_GET_BLOCKS_IO_CREATE_EXT; | |
4b70df18 | 3431 | |
378f32ba | 3432 | ret = ext4_map_blocks(handle, inode, map, m_flags); |
8d5d02e6 | 3433 | |
74c66bcb | 3434 | /* |
378f32ba MB |
3435 | * We cannot fill holes in indirect tree based inodes as that could |
3436 | * expose stale data in the case of a crash. Use the magic error code | |
3437 | * to fallback to buffered I/O. | |
74c66bcb | 3438 | */ |
378f32ba MB |
3439 | if (!m_flags && !ret) |
3440 | ret = -ENOTBLK; | |
187372a3 | 3441 | |
f063db5e MB |
3442 | ext4_journal_stop(handle); |
3443 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) | |
3444 | goto retry; | |
3445 | ||
3446 | return ret; | |
4c0425ff | 3447 | } |
c7064ef1 | 3448 | |
f063db5e | 3449 | |
364443cb | 3450 | static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length, |
c039b997 | 3451 | unsigned flags, struct iomap *iomap, struct iomap *srcmap) |
4c0425ff | 3452 | { |
364443cb | 3453 | int ret; |
09edf4d3 MB |
3454 | struct ext4_map_blocks map; |
3455 | u8 blkbits = inode->i_blkbits; | |
729f52c6 | 3456 | |
bcd8e91f TT |
3457 | if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK) |
3458 | return -EINVAL; | |
4bd809db | 3459 | |
09edf4d3 MB |
3460 | if (WARN_ON_ONCE(ext4_has_inline_data(inode))) |
3461 | return -ERANGE; | |
4bd809db | 3462 | |
e8340395 | 3463 | /* |
09edf4d3 | 3464 | * Calculate the first and last logical blocks respectively. |
e8340395 | 3465 | */ |
09edf4d3 MB |
3466 | map.m_lblk = offset >> blkbits; |
3467 | map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits, | |
3468 | EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1; | |
e8340395 | 3469 | |
9faac62d RH |
3470 | if (flags & IOMAP_WRITE) { |
3471 | /* | |
3472 | * We check here if the blocks are already allocated, then we | |
3473 | * don't need to start a journal txn and we can directly return | |
3474 | * the mapping information. This could boost performance | |
3475 | * especially in multi-threaded overwrite requests. | |
3476 | */ | |
3477 | if (offset + length <= i_size_read(inode)) { | |
3478 | ret = ext4_map_blocks(NULL, inode, &map, 0); | |
3479 | if (ret > 0 && (map.m_flags & EXT4_MAP_MAPPED)) | |
3480 | goto out; | |
3481 | } | |
f063db5e | 3482 | ret = ext4_iomap_alloc(inode, &map, flags); |
9faac62d | 3483 | } else { |
545052e9 | 3484 | ret = ext4_map_blocks(NULL, inode, &map, 0); |
9faac62d | 3485 | } |
4bd809db | 3486 | |
f063db5e MB |
3487 | if (ret < 0) |
3488 | return ret; | |
9faac62d | 3489 | out: |
c8fdfe29 | 3490 | ext4_set_iomap(inode, iomap, &map, offset, length); |
4bd809db | 3491 | |
364443cb JK |
3492 | return 0; |
3493 | } | |
8d5d02e6 | 3494 | |
8cd115bd JK |
3495 | static int ext4_iomap_overwrite_begin(struct inode *inode, loff_t offset, |
3496 | loff_t length, unsigned flags, struct iomap *iomap, | |
3497 | struct iomap *srcmap) | |
3498 | { | |
3499 | int ret; | |
3500 | ||
3501 | /* | |
3502 | * Even for writes we don't need to allocate blocks, so just pretend | |
3503 | * we are reading to save overhead of starting a transaction. | |
3504 | */ | |
3505 | flags &= ~IOMAP_WRITE; | |
3506 | ret = ext4_iomap_begin(inode, offset, length, flags, iomap, srcmap); | |
3507 | WARN_ON_ONCE(iomap->type != IOMAP_MAPPED); | |
3508 | return ret; | |
3509 | } | |
3510 | ||
776722e8 JK |
3511 | static int ext4_iomap_end(struct inode *inode, loff_t offset, loff_t length, |
3512 | ssize_t written, unsigned flags, struct iomap *iomap) | |
3513 | { | |
69c499d1 | 3514 | /* |
378f32ba MB |
3515 | * Check to see whether an error occurred while writing out the data to |
3516 | * the allocated blocks. If so, return the magic error code so that we | |
3517 | * fallback to buffered I/O and attempt to complete the remainder of | |
3518 | * the I/O. Any blocks that may have been allocated in preparation for | |
3519 | * the direct I/O will be reused during buffered I/O. | |
69c499d1 | 3520 | */ |
378f32ba MB |
3521 | if (flags & (IOMAP_WRITE | IOMAP_DIRECT) && written == 0) |
3522 | return -ENOTBLK; | |
69c499d1 | 3523 | |
569342dc | 3524 | return 0; |
776722e8 | 3525 | } |
4bd809db | 3526 | |
8ff6daa1 | 3527 | const struct iomap_ops ext4_iomap_ops = { |
364443cb | 3528 | .iomap_begin = ext4_iomap_begin, |
776722e8 | 3529 | .iomap_end = ext4_iomap_end, |
364443cb | 3530 | }; |
8d5d02e6 | 3531 | |
8cd115bd JK |
3532 | const struct iomap_ops ext4_iomap_overwrite_ops = { |
3533 | .iomap_begin = ext4_iomap_overwrite_begin, | |
3534 | .iomap_end = ext4_iomap_end, | |
3535 | }; | |
3536 | ||
09edf4d3 MB |
3537 | static bool ext4_iomap_is_delalloc(struct inode *inode, |
3538 | struct ext4_map_blocks *map) | |
3539 | { | |
3540 | struct extent_status es; | |
3541 | ext4_lblk_t offset = 0, end = map->m_lblk + map->m_len - 1; | |
914f82a3 | 3542 | |
09edf4d3 MB |
3543 | ext4_es_find_extent_range(inode, &ext4_es_is_delayed, |
3544 | map->m_lblk, end, &es); | |
914f82a3 | 3545 | |
09edf4d3 MB |
3546 | if (!es.es_len || es.es_lblk > end) |
3547 | return false; | |
914f82a3 | 3548 | |
09edf4d3 MB |
3549 | if (es.es_lblk > map->m_lblk) { |
3550 | map->m_len = es.es_lblk - map->m_lblk; | |
3551 | return false; | |
914f82a3 | 3552 | } |
914f82a3 | 3553 | |
09edf4d3 MB |
3554 | offset = map->m_lblk - es.es_lblk; |
3555 | map->m_len = es.es_len - offset; | |
914f82a3 | 3556 | |
09edf4d3 | 3557 | return true; |
4c0425ff MC |
3558 | } |
3559 | ||
09edf4d3 MB |
3560 | static int ext4_iomap_begin_report(struct inode *inode, loff_t offset, |
3561 | loff_t length, unsigned int flags, | |
3562 | struct iomap *iomap, struct iomap *srcmap) | |
4c0425ff | 3563 | { |
09edf4d3 MB |
3564 | int ret; |
3565 | bool delalloc = false; | |
3566 | struct ext4_map_blocks map; | |
3567 | u8 blkbits = inode->i_blkbits; | |
4c0425ff | 3568 | |
09edf4d3 MB |
3569 | if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK) |
3570 | return -EINVAL; | |
3571 | ||
3572 | if (ext4_has_inline_data(inode)) { | |
3573 | ret = ext4_inline_data_iomap(inode, iomap); | |
3574 | if (ret != -EAGAIN) { | |
3575 | if (ret == 0 && offset >= iomap->length) | |
3576 | ret = -ENOENT; | |
3577 | return ret; | |
3578 | } | |
3579 | } | |
2058f83a | 3580 | |
84ebd795 | 3581 | /* |
09edf4d3 | 3582 | * Calculate the first and last logical block respectively. |
84ebd795 | 3583 | */ |
09edf4d3 MB |
3584 | map.m_lblk = offset >> blkbits; |
3585 | map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits, | |
3586 | EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1; | |
84ebd795 | 3587 | |
b2c57642 RH |
3588 | /* |
3589 | * Fiemap callers may call for offset beyond s_bitmap_maxbytes. | |
3590 | * So handle it here itself instead of querying ext4_map_blocks(). | |
3591 | * Since ext4_map_blocks() will warn about it and will return | |
3592 | * -EIO error. | |
3593 | */ | |
3594 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) { | |
3595 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
3596 | ||
3597 | if (offset >= sbi->s_bitmap_maxbytes) { | |
3598 | map.m_flags = 0; | |
3599 | goto set_iomap; | |
3600 | } | |
3601 | } | |
3602 | ||
09edf4d3 MB |
3603 | ret = ext4_map_blocks(NULL, inode, &map, 0); |
3604 | if (ret < 0) | |
3605 | return ret; | |
3606 | if (ret == 0) | |
3607 | delalloc = ext4_iomap_is_delalloc(inode, &map); | |
46c7f254 | 3608 | |
b2c57642 | 3609 | set_iomap: |
09edf4d3 MB |
3610 | ext4_set_iomap(inode, iomap, &map, offset, length); |
3611 | if (delalloc && iomap->type == IOMAP_HOLE) | |
3612 | iomap->type = IOMAP_DELALLOC; | |
3613 | ||
3614 | return 0; | |
4c0425ff MC |
3615 | } |
3616 | ||
09edf4d3 MB |
3617 | const struct iomap_ops ext4_iomap_report_ops = { |
3618 | .iomap_begin = ext4_iomap_begin_report, | |
3619 | }; | |
3620 | ||
ac27a0ec | 3621 | /* |
617ba13b | 3622 | * Pages can be marked dirty completely asynchronously from ext4's journalling |
ac27a0ec DK |
3623 | * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do |
3624 | * much here because ->set_page_dirty is called under VFS locks. The page is | |
3625 | * not necessarily locked. | |
3626 | * | |
3627 | * We cannot just dirty the page and leave attached buffers clean, because the | |
3628 | * buffers' dirty state is "definitive". We cannot just set the buffers dirty | |
3629 | * or jbddirty because all the journalling code will explode. | |
3630 | * | |
3631 | * So what we do is to mark the page "pending dirty" and next time writepage | |
3632 | * is called, propagate that into the buffers appropriately. | |
3633 | */ | |
617ba13b | 3634 | static int ext4_journalled_set_page_dirty(struct page *page) |
ac27a0ec DK |
3635 | { |
3636 | SetPageChecked(page); | |
3637 | return __set_page_dirty_nobuffers(page); | |
3638 | } | |
3639 | ||
6dcc693b JK |
3640 | static int ext4_set_page_dirty(struct page *page) |
3641 | { | |
3642 | WARN_ON_ONCE(!PageLocked(page) && !PageDirty(page)); | |
3643 | WARN_ON_ONCE(!page_has_buffers(page)); | |
3644 | return __set_page_dirty_buffers(page); | |
3645 | } | |
3646 | ||
0e6895ba RH |
3647 | static int ext4_iomap_swap_activate(struct swap_info_struct *sis, |
3648 | struct file *file, sector_t *span) | |
3649 | { | |
3650 | return iomap_swapfile_activate(sis, file, span, | |
3651 | &ext4_iomap_report_ops); | |
3652 | } | |
3653 | ||
74d553aa | 3654 | static const struct address_space_operations ext4_aops = { |
8ab22b9a | 3655 | .readpage = ext4_readpage, |
6311f91f | 3656 | .readahead = ext4_readahead, |
43ce1d23 | 3657 | .writepage = ext4_writepage, |
20970ba6 | 3658 | .writepages = ext4_writepages, |
8ab22b9a | 3659 | .write_begin = ext4_write_begin, |
74d553aa | 3660 | .write_end = ext4_write_end, |
6dcc693b | 3661 | .set_page_dirty = ext4_set_page_dirty, |
8ab22b9a HH |
3662 | .bmap = ext4_bmap, |
3663 | .invalidatepage = ext4_invalidatepage, | |
3664 | .releasepage = ext4_releasepage, | |
378f32ba | 3665 | .direct_IO = noop_direct_IO, |
8ab22b9a HH |
3666 | .migratepage = buffer_migrate_page, |
3667 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 3668 | .error_remove_page = generic_error_remove_page, |
0e6895ba | 3669 | .swap_activate = ext4_iomap_swap_activate, |
ac27a0ec DK |
3670 | }; |
3671 | ||
617ba13b | 3672 | static const struct address_space_operations ext4_journalled_aops = { |
8ab22b9a | 3673 | .readpage = ext4_readpage, |
6311f91f | 3674 | .readahead = ext4_readahead, |
43ce1d23 | 3675 | .writepage = ext4_writepage, |
20970ba6 | 3676 | .writepages = ext4_writepages, |
8ab22b9a HH |
3677 | .write_begin = ext4_write_begin, |
3678 | .write_end = ext4_journalled_write_end, | |
3679 | .set_page_dirty = ext4_journalled_set_page_dirty, | |
3680 | .bmap = ext4_bmap, | |
4520fb3c | 3681 | .invalidatepage = ext4_journalled_invalidatepage, |
8ab22b9a | 3682 | .releasepage = ext4_releasepage, |
378f32ba | 3683 | .direct_IO = noop_direct_IO, |
8ab22b9a | 3684 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 3685 | .error_remove_page = generic_error_remove_page, |
0e6895ba | 3686 | .swap_activate = ext4_iomap_swap_activate, |
ac27a0ec DK |
3687 | }; |
3688 | ||
64769240 | 3689 | static const struct address_space_operations ext4_da_aops = { |
8ab22b9a | 3690 | .readpage = ext4_readpage, |
6311f91f | 3691 | .readahead = ext4_readahead, |
43ce1d23 | 3692 | .writepage = ext4_writepage, |
20970ba6 | 3693 | .writepages = ext4_writepages, |
8ab22b9a HH |
3694 | .write_begin = ext4_da_write_begin, |
3695 | .write_end = ext4_da_write_end, | |
6dcc693b | 3696 | .set_page_dirty = ext4_set_page_dirty, |
8ab22b9a | 3697 | .bmap = ext4_bmap, |
8fcc3a58 | 3698 | .invalidatepage = ext4_invalidatepage, |
8ab22b9a | 3699 | .releasepage = ext4_releasepage, |
378f32ba | 3700 | .direct_IO = noop_direct_IO, |
8ab22b9a HH |
3701 | .migratepage = buffer_migrate_page, |
3702 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 3703 | .error_remove_page = generic_error_remove_page, |
0e6895ba | 3704 | .swap_activate = ext4_iomap_swap_activate, |
64769240 AT |
3705 | }; |
3706 | ||
5f0663bb DW |
3707 | static const struct address_space_operations ext4_dax_aops = { |
3708 | .writepages = ext4_dax_writepages, | |
3709 | .direct_IO = noop_direct_IO, | |
b82a96c9 | 3710 | .set_page_dirty = __set_page_dirty_no_writeback, |
94dbb631 | 3711 | .bmap = ext4_bmap, |
5f0663bb | 3712 | .invalidatepage = noop_invalidatepage, |
0e6895ba | 3713 | .swap_activate = ext4_iomap_swap_activate, |
5f0663bb DW |
3714 | }; |
3715 | ||
617ba13b | 3716 | void ext4_set_aops(struct inode *inode) |
ac27a0ec | 3717 | { |
3d2b1582 LC |
3718 | switch (ext4_inode_journal_mode(inode)) { |
3719 | case EXT4_INODE_ORDERED_DATA_MODE: | |
3d2b1582 | 3720 | case EXT4_INODE_WRITEBACK_DATA_MODE: |
3d2b1582 LC |
3721 | break; |
3722 | case EXT4_INODE_JOURNAL_DATA_MODE: | |
617ba13b | 3723 | inode->i_mapping->a_ops = &ext4_journalled_aops; |
74d553aa | 3724 | return; |
3d2b1582 LC |
3725 | default: |
3726 | BUG(); | |
3727 | } | |
5f0663bb DW |
3728 | if (IS_DAX(inode)) |
3729 | inode->i_mapping->a_ops = &ext4_dax_aops; | |
3730 | else if (test_opt(inode->i_sb, DELALLOC)) | |
74d553aa TT |
3731 | inode->i_mapping->a_ops = &ext4_da_aops; |
3732 | else | |
3733 | inode->i_mapping->a_ops = &ext4_aops; | |
ac27a0ec DK |
3734 | } |
3735 | ||
923ae0ff | 3736 | static int __ext4_block_zero_page_range(handle_t *handle, |
d863dc36 LC |
3737 | struct address_space *mapping, loff_t from, loff_t length) |
3738 | { | |
09cbfeaf KS |
3739 | ext4_fsblk_t index = from >> PAGE_SHIFT; |
3740 | unsigned offset = from & (PAGE_SIZE-1); | |
923ae0ff | 3741 | unsigned blocksize, pos; |
d863dc36 LC |
3742 | ext4_lblk_t iblock; |
3743 | struct inode *inode = mapping->host; | |
3744 | struct buffer_head *bh; | |
3745 | struct page *page; | |
3746 | int err = 0; | |
3747 | ||
09cbfeaf | 3748 | page = find_or_create_page(mapping, from >> PAGE_SHIFT, |
c62d2555 | 3749 | mapping_gfp_constraint(mapping, ~__GFP_FS)); |
d863dc36 LC |
3750 | if (!page) |
3751 | return -ENOMEM; | |
3752 | ||
3753 | blocksize = inode->i_sb->s_blocksize; | |
d863dc36 | 3754 | |
09cbfeaf | 3755 | iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits); |
d863dc36 LC |
3756 | |
3757 | if (!page_has_buffers(page)) | |
3758 | create_empty_buffers(page, blocksize, 0); | |
3759 | ||
3760 | /* Find the buffer that contains "offset" */ | |
3761 | bh = page_buffers(page); | |
3762 | pos = blocksize; | |
3763 | while (offset >= pos) { | |
3764 | bh = bh->b_this_page; | |
3765 | iblock++; | |
3766 | pos += blocksize; | |
3767 | } | |
d863dc36 LC |
3768 | if (buffer_freed(bh)) { |
3769 | BUFFER_TRACE(bh, "freed: skip"); | |
3770 | goto unlock; | |
3771 | } | |
d863dc36 LC |
3772 | if (!buffer_mapped(bh)) { |
3773 | BUFFER_TRACE(bh, "unmapped"); | |
3774 | ext4_get_block(inode, iblock, bh, 0); | |
3775 | /* unmapped? It's a hole - nothing to do */ | |
3776 | if (!buffer_mapped(bh)) { | |
3777 | BUFFER_TRACE(bh, "still unmapped"); | |
3778 | goto unlock; | |
3779 | } | |
3780 | } | |
3781 | ||
3782 | /* Ok, it's mapped. Make sure it's up-to-date */ | |
3783 | if (PageUptodate(page)) | |
3784 | set_buffer_uptodate(bh); | |
3785 | ||
3786 | if (!buffer_uptodate(bh)) { | |
2d069c08 | 3787 | err = ext4_read_bh_lock(bh, 0, true); |
3788 | if (err) | |
d863dc36 | 3789 | goto unlock; |
4f74d15f | 3790 | if (fscrypt_inode_uses_fs_layer_crypto(inode)) { |
c9c7429c | 3791 | /* We expect the key to be set. */ |
a7550b30 | 3792 | BUG_ON(!fscrypt_has_encryption_key(inode)); |
834f1565 EB |
3793 | err = fscrypt_decrypt_pagecache_blocks(page, blocksize, |
3794 | bh_offset(bh)); | |
3795 | if (err) { | |
3796 | clear_buffer_uptodate(bh); | |
3797 | goto unlock; | |
3798 | } | |
c9c7429c | 3799 | } |
d863dc36 | 3800 | } |
d863dc36 LC |
3801 | if (ext4_should_journal_data(inode)) { |
3802 | BUFFER_TRACE(bh, "get write access"); | |
188c299e JK |
3803 | err = ext4_journal_get_write_access(handle, inode->i_sb, bh, |
3804 | EXT4_JTR_NONE); | |
d863dc36 LC |
3805 | if (err) |
3806 | goto unlock; | |
3807 | } | |
d863dc36 | 3808 | zero_user(page, offset, length); |
d863dc36 LC |
3809 | BUFFER_TRACE(bh, "zeroed end of block"); |
3810 | ||
d863dc36 LC |
3811 | if (ext4_should_journal_data(inode)) { |
3812 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
0713ed0c | 3813 | } else { |
353eefd3 | 3814 | err = 0; |
d863dc36 | 3815 | mark_buffer_dirty(bh); |
3957ef53 | 3816 | if (ext4_should_order_data(inode)) |
73131fbb RZ |
3817 | err = ext4_jbd2_inode_add_write(handle, inode, from, |
3818 | length); | |
0713ed0c | 3819 | } |
d863dc36 LC |
3820 | |
3821 | unlock: | |
3822 | unlock_page(page); | |
09cbfeaf | 3823 | put_page(page); |
d863dc36 LC |
3824 | return err; |
3825 | } | |
3826 | ||
923ae0ff RZ |
3827 | /* |
3828 | * ext4_block_zero_page_range() zeros out a mapping of length 'length' | |
3829 | * starting from file offset 'from'. The range to be zero'd must | |
3830 | * be contained with in one block. If the specified range exceeds | |
3831 | * the end of the block it will be shortened to end of the block | |
3088e5a5 | 3832 | * that corresponds to 'from' |
923ae0ff RZ |
3833 | */ |
3834 | static int ext4_block_zero_page_range(handle_t *handle, | |
3835 | struct address_space *mapping, loff_t from, loff_t length) | |
3836 | { | |
3837 | struct inode *inode = mapping->host; | |
09cbfeaf | 3838 | unsigned offset = from & (PAGE_SIZE-1); |
923ae0ff RZ |
3839 | unsigned blocksize = inode->i_sb->s_blocksize; |
3840 | unsigned max = blocksize - (offset & (blocksize - 1)); | |
3841 | ||
3842 | /* | |
3843 | * correct length if it does not fall between | |
3844 | * 'from' and the end of the block | |
3845 | */ | |
3846 | if (length > max || length < 0) | |
3847 | length = max; | |
3848 | ||
47e69351 JK |
3849 | if (IS_DAX(inode)) { |
3850 | return iomap_zero_range(inode, from, length, NULL, | |
3851 | &ext4_iomap_ops); | |
3852 | } | |
923ae0ff RZ |
3853 | return __ext4_block_zero_page_range(handle, mapping, from, length); |
3854 | } | |
3855 | ||
94350ab5 MW |
3856 | /* |
3857 | * ext4_block_truncate_page() zeroes out a mapping from file offset `from' | |
3858 | * up to the end of the block which corresponds to `from'. | |
3859 | * This required during truncate. We need to physically zero the tail end | |
3860 | * of that block so it doesn't yield old data if the file is later grown. | |
3861 | */ | |
c197855e | 3862 | static int ext4_block_truncate_page(handle_t *handle, |
94350ab5 MW |
3863 | struct address_space *mapping, loff_t from) |
3864 | { | |
09cbfeaf | 3865 | unsigned offset = from & (PAGE_SIZE-1); |
94350ab5 MW |
3866 | unsigned length; |
3867 | unsigned blocksize; | |
3868 | struct inode *inode = mapping->host; | |
3869 | ||
0d06863f | 3870 | /* If we are processing an encrypted inode during orphan list handling */ |
592ddec7 | 3871 | if (IS_ENCRYPTED(inode) && !fscrypt_has_encryption_key(inode)) |
0d06863f TT |
3872 | return 0; |
3873 | ||
94350ab5 MW |
3874 | blocksize = inode->i_sb->s_blocksize; |
3875 | length = blocksize - (offset & (blocksize - 1)); | |
3876 | ||
3877 | return ext4_block_zero_page_range(handle, mapping, from, length); | |
3878 | } | |
3879 | ||
a87dd18c LC |
3880 | int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode, |
3881 | loff_t lstart, loff_t length) | |
3882 | { | |
3883 | struct super_block *sb = inode->i_sb; | |
3884 | struct address_space *mapping = inode->i_mapping; | |
e1be3a92 | 3885 | unsigned partial_start, partial_end; |
a87dd18c LC |
3886 | ext4_fsblk_t start, end; |
3887 | loff_t byte_end = (lstart + length - 1); | |
3888 | int err = 0; | |
3889 | ||
e1be3a92 LC |
3890 | partial_start = lstart & (sb->s_blocksize - 1); |
3891 | partial_end = byte_end & (sb->s_blocksize - 1); | |
3892 | ||
a87dd18c LC |
3893 | start = lstart >> sb->s_blocksize_bits; |
3894 | end = byte_end >> sb->s_blocksize_bits; | |
3895 | ||
3896 | /* Handle partial zero within the single block */ | |
e1be3a92 LC |
3897 | if (start == end && |
3898 | (partial_start || (partial_end != sb->s_blocksize - 1))) { | |
a87dd18c LC |
3899 | err = ext4_block_zero_page_range(handle, mapping, |
3900 | lstart, length); | |
3901 | return err; | |
3902 | } | |
3903 | /* Handle partial zero out on the start of the range */ | |
e1be3a92 | 3904 | if (partial_start) { |
a87dd18c LC |
3905 | err = ext4_block_zero_page_range(handle, mapping, |
3906 | lstart, sb->s_blocksize); | |
3907 | if (err) | |
3908 | return err; | |
3909 | } | |
3910 | /* Handle partial zero out on the end of the range */ | |
e1be3a92 | 3911 | if (partial_end != sb->s_blocksize - 1) |
a87dd18c | 3912 | err = ext4_block_zero_page_range(handle, mapping, |
e1be3a92 LC |
3913 | byte_end - partial_end, |
3914 | partial_end + 1); | |
a87dd18c LC |
3915 | return err; |
3916 | } | |
3917 | ||
91ef4caf DG |
3918 | int ext4_can_truncate(struct inode *inode) |
3919 | { | |
91ef4caf DG |
3920 | if (S_ISREG(inode->i_mode)) |
3921 | return 1; | |
3922 | if (S_ISDIR(inode->i_mode)) | |
3923 | return 1; | |
3924 | if (S_ISLNK(inode->i_mode)) | |
3925 | return !ext4_inode_is_fast_symlink(inode); | |
3926 | return 0; | |
3927 | } | |
3928 | ||
01127848 JK |
3929 | /* |
3930 | * We have to make sure i_disksize gets properly updated before we truncate | |
3931 | * page cache due to hole punching or zero range. Otherwise i_disksize update | |
3932 | * can get lost as it may have been postponed to submission of writeback but | |
3933 | * that will never happen after we truncate page cache. | |
3934 | */ | |
3935 | int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset, | |
3936 | loff_t len) | |
3937 | { | |
3938 | handle_t *handle; | |
4209ae12 HS |
3939 | int ret; |
3940 | ||
01127848 JK |
3941 | loff_t size = i_size_read(inode); |
3942 | ||
5955102c | 3943 | WARN_ON(!inode_is_locked(inode)); |
01127848 JK |
3944 | if (offset > size || offset + len < size) |
3945 | return 0; | |
3946 | ||
3947 | if (EXT4_I(inode)->i_disksize >= size) | |
3948 | return 0; | |
3949 | ||
3950 | handle = ext4_journal_start(inode, EXT4_HT_MISC, 1); | |
3951 | if (IS_ERR(handle)) | |
3952 | return PTR_ERR(handle); | |
3953 | ext4_update_i_disksize(inode, size); | |
4209ae12 | 3954 | ret = ext4_mark_inode_dirty(handle, inode); |
01127848 JK |
3955 | ext4_journal_stop(handle); |
3956 | ||
4209ae12 | 3957 | return ret; |
01127848 JK |
3958 | } |
3959 | ||
b1f38217 | 3960 | static void ext4_wait_dax_page(struct ext4_inode_info *ei) |
430657b6 | 3961 | { |
430657b6 RZ |
3962 | up_write(&ei->i_mmap_sem); |
3963 | schedule(); | |
3964 | down_write(&ei->i_mmap_sem); | |
3965 | } | |
3966 | ||
3967 | int ext4_break_layouts(struct inode *inode) | |
3968 | { | |
3969 | struct ext4_inode_info *ei = EXT4_I(inode); | |
3970 | struct page *page; | |
430657b6 RZ |
3971 | int error; |
3972 | ||
3973 | if (WARN_ON_ONCE(!rwsem_is_locked(&ei->i_mmap_sem))) | |
3974 | return -EINVAL; | |
3975 | ||
3976 | do { | |
430657b6 RZ |
3977 | page = dax_layout_busy_page(inode->i_mapping); |
3978 | if (!page) | |
3979 | return 0; | |
3980 | ||
3981 | error = ___wait_var_event(&page->_refcount, | |
3982 | atomic_read(&page->_refcount) == 1, | |
3983 | TASK_INTERRUPTIBLE, 0, 0, | |
b1f38217 RZ |
3984 | ext4_wait_dax_page(ei)); |
3985 | } while (error == 0); | |
430657b6 RZ |
3986 | |
3987 | return error; | |
3988 | } | |
3989 | ||
a4bb6b64 | 3990 | /* |
cca32b7e | 3991 | * ext4_punch_hole: punches a hole in a file by releasing the blocks |
a4bb6b64 AH |
3992 | * associated with the given offset and length |
3993 | * | |
3994 | * @inode: File inode | |
3995 | * @offset: The offset where the hole will begin | |
3996 | * @len: The length of the hole | |
3997 | * | |
4907cb7b | 3998 | * Returns: 0 on success or negative on failure |
a4bb6b64 AH |
3999 | */ |
4000 | ||
aeb2817a | 4001 | int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length) |
a4bb6b64 | 4002 | { |
26a4c0c6 TT |
4003 | struct super_block *sb = inode->i_sb; |
4004 | ext4_lblk_t first_block, stop_block; | |
4005 | struct address_space *mapping = inode->i_mapping; | |
a87dd18c | 4006 | loff_t first_block_offset, last_block_offset; |
26a4c0c6 TT |
4007 | handle_t *handle; |
4008 | unsigned int credits; | |
4209ae12 | 4009 | int ret = 0, ret2 = 0; |
26a4c0c6 | 4010 | |
b8a86845 | 4011 | trace_ext4_punch_hole(inode, offset, length, 0); |
aaddea81 | 4012 | |
c1e8220b TT |
4013 | ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA); |
4014 | if (ext4_has_inline_data(inode)) { | |
4015 | down_write(&EXT4_I(inode)->i_mmap_sem); | |
4016 | ret = ext4_convert_inline_data(inode); | |
4017 | up_write(&EXT4_I(inode)->i_mmap_sem); | |
4018 | if (ret) | |
4019 | return ret; | |
4020 | } | |
4021 | ||
26a4c0c6 TT |
4022 | /* |
4023 | * Write out all dirty pages to avoid race conditions | |
4024 | * Then release them. | |
4025 | */ | |
cca32b7e | 4026 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { |
26a4c0c6 TT |
4027 | ret = filemap_write_and_wait_range(mapping, offset, |
4028 | offset + length - 1); | |
4029 | if (ret) | |
4030 | return ret; | |
4031 | } | |
4032 | ||
5955102c | 4033 | inode_lock(inode); |
9ef06cec | 4034 | |
26a4c0c6 TT |
4035 | /* No need to punch hole beyond i_size */ |
4036 | if (offset >= inode->i_size) | |
4037 | goto out_mutex; | |
4038 | ||
4039 | /* | |
4040 | * If the hole extends beyond i_size, set the hole | |
4041 | * to end after the page that contains i_size | |
4042 | */ | |
4043 | if (offset + length > inode->i_size) { | |
4044 | length = inode->i_size + | |
09cbfeaf | 4045 | PAGE_SIZE - (inode->i_size & (PAGE_SIZE - 1)) - |
26a4c0c6 TT |
4046 | offset; |
4047 | } | |
4048 | ||
a361293f JK |
4049 | if (offset & (sb->s_blocksize - 1) || |
4050 | (offset + length) & (sb->s_blocksize - 1)) { | |
4051 | /* | |
4052 | * Attach jinode to inode for jbd2 if we do any zeroing of | |
4053 | * partial block | |
4054 | */ | |
4055 | ret = ext4_inode_attach_jinode(inode); | |
4056 | if (ret < 0) | |
4057 | goto out_mutex; | |
4058 | ||
4059 | } | |
4060 | ||
ea3d7209 | 4061 | /* Wait all existing dio workers, newcomers will block on i_mutex */ |
ea3d7209 JK |
4062 | inode_dio_wait(inode); |
4063 | ||
4064 | /* | |
4065 | * Prevent page faults from reinstantiating pages we have released from | |
4066 | * page cache. | |
4067 | */ | |
4068 | down_write(&EXT4_I(inode)->i_mmap_sem); | |
430657b6 RZ |
4069 | |
4070 | ret = ext4_break_layouts(inode); | |
4071 | if (ret) | |
4072 | goto out_dio; | |
4073 | ||
a87dd18c LC |
4074 | first_block_offset = round_up(offset, sb->s_blocksize); |
4075 | last_block_offset = round_down((offset + length), sb->s_blocksize) - 1; | |
26a4c0c6 | 4076 | |
a87dd18c | 4077 | /* Now release the pages and zero block aligned part of pages*/ |
01127848 JK |
4078 | if (last_block_offset > first_block_offset) { |
4079 | ret = ext4_update_disksize_before_punch(inode, offset, length); | |
4080 | if (ret) | |
4081 | goto out_dio; | |
a87dd18c LC |
4082 | truncate_pagecache_range(inode, first_block_offset, |
4083 | last_block_offset); | |
01127848 | 4084 | } |
26a4c0c6 TT |
4085 | |
4086 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
4087 | credits = ext4_writepage_trans_blocks(inode); | |
4088 | else | |
4089 | credits = ext4_blocks_for_truncate(inode); | |
4090 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); | |
4091 | if (IS_ERR(handle)) { | |
4092 | ret = PTR_ERR(handle); | |
4093 | ext4_std_error(sb, ret); | |
4094 | goto out_dio; | |
4095 | } | |
4096 | ||
a87dd18c LC |
4097 | ret = ext4_zero_partial_blocks(handle, inode, offset, |
4098 | length); | |
4099 | if (ret) | |
4100 | goto out_stop; | |
26a4c0c6 TT |
4101 | |
4102 | first_block = (offset + sb->s_blocksize - 1) >> | |
4103 | EXT4_BLOCK_SIZE_BITS(sb); | |
4104 | stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb); | |
4105 | ||
eee597ac LC |
4106 | /* If there are blocks to remove, do it */ |
4107 | if (stop_block > first_block) { | |
26a4c0c6 | 4108 | |
eee597ac | 4109 | down_write(&EXT4_I(inode)->i_data_sem); |
27bc446e | 4110 | ext4_discard_preallocations(inode, 0); |
26a4c0c6 | 4111 | |
eee597ac LC |
4112 | ret = ext4_es_remove_extent(inode, first_block, |
4113 | stop_block - first_block); | |
4114 | if (ret) { | |
4115 | up_write(&EXT4_I(inode)->i_data_sem); | |
4116 | goto out_stop; | |
4117 | } | |
26a4c0c6 | 4118 | |
eee597ac LC |
4119 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
4120 | ret = ext4_ext_remove_space(inode, first_block, | |
4121 | stop_block - 1); | |
4122 | else | |
4123 | ret = ext4_ind_remove_space(handle, inode, first_block, | |
4124 | stop_block); | |
26a4c0c6 | 4125 | |
eee597ac LC |
4126 | up_write(&EXT4_I(inode)->i_data_sem); |
4127 | } | |
a80f7fcf | 4128 | ext4_fc_track_range(handle, inode, first_block, stop_block); |
26a4c0c6 TT |
4129 | if (IS_SYNC(inode)) |
4130 | ext4_handle_sync(handle); | |
e251f9bc | 4131 | |
eeca7ea1 | 4132 | inode->i_mtime = inode->i_ctime = current_time(inode); |
4209ae12 HS |
4133 | ret2 = ext4_mark_inode_dirty(handle, inode); |
4134 | if (unlikely(ret2)) | |
4135 | ret = ret2; | |
67a7d5f5 JK |
4136 | if (ret >= 0) |
4137 | ext4_update_inode_fsync_trans(handle, inode, 1); | |
26a4c0c6 TT |
4138 | out_stop: |
4139 | ext4_journal_stop(handle); | |
4140 | out_dio: | |
ea3d7209 | 4141 | up_write(&EXT4_I(inode)->i_mmap_sem); |
26a4c0c6 | 4142 | out_mutex: |
5955102c | 4143 | inode_unlock(inode); |
26a4c0c6 | 4144 | return ret; |
a4bb6b64 AH |
4145 | } |
4146 | ||
a361293f JK |
4147 | int ext4_inode_attach_jinode(struct inode *inode) |
4148 | { | |
4149 | struct ext4_inode_info *ei = EXT4_I(inode); | |
4150 | struct jbd2_inode *jinode; | |
4151 | ||
4152 | if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal) | |
4153 | return 0; | |
4154 | ||
4155 | jinode = jbd2_alloc_inode(GFP_KERNEL); | |
4156 | spin_lock(&inode->i_lock); | |
4157 | if (!ei->jinode) { | |
4158 | if (!jinode) { | |
4159 | spin_unlock(&inode->i_lock); | |
4160 | return -ENOMEM; | |
4161 | } | |
4162 | ei->jinode = jinode; | |
4163 | jbd2_journal_init_jbd_inode(ei->jinode, inode); | |
4164 | jinode = NULL; | |
4165 | } | |
4166 | spin_unlock(&inode->i_lock); | |
4167 | if (unlikely(jinode != NULL)) | |
4168 | jbd2_free_inode(jinode); | |
4169 | return 0; | |
4170 | } | |
4171 | ||
ac27a0ec | 4172 | /* |
617ba13b | 4173 | * ext4_truncate() |
ac27a0ec | 4174 | * |
617ba13b MC |
4175 | * We block out ext4_get_block() block instantiations across the entire |
4176 | * transaction, and VFS/VM ensures that ext4_truncate() cannot run | |
ac27a0ec DK |
4177 | * simultaneously on behalf of the same inode. |
4178 | * | |
42b2aa86 | 4179 | * As we work through the truncate and commit bits of it to the journal there |
ac27a0ec DK |
4180 | * is one core, guiding principle: the file's tree must always be consistent on |
4181 | * disk. We must be able to restart the truncate after a crash. | |
4182 | * | |
4183 | * The file's tree may be transiently inconsistent in memory (although it | |
4184 | * probably isn't), but whenever we close off and commit a journal transaction, | |
4185 | * the contents of (the filesystem + the journal) must be consistent and | |
4186 | * restartable. It's pretty simple, really: bottom up, right to left (although | |
4187 | * left-to-right works OK too). | |
4188 | * | |
4189 | * Note that at recovery time, journal replay occurs *before* the restart of | |
4190 | * truncate against the orphan inode list. | |
4191 | * | |
4192 | * The committed inode has the new, desired i_size (which is the same as | |
617ba13b | 4193 | * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see |
ac27a0ec | 4194 | * that this inode's truncate did not complete and it will again call |
617ba13b MC |
4195 | * ext4_truncate() to have another go. So there will be instantiated blocks |
4196 | * to the right of the truncation point in a crashed ext4 filesystem. But | |
ac27a0ec | 4197 | * that's fine - as long as they are linked from the inode, the post-crash |
617ba13b | 4198 | * ext4_truncate() run will find them and release them. |
ac27a0ec | 4199 | */ |
2c98eb5e | 4200 | int ext4_truncate(struct inode *inode) |
ac27a0ec | 4201 | { |
819c4920 TT |
4202 | struct ext4_inode_info *ei = EXT4_I(inode); |
4203 | unsigned int credits; | |
4209ae12 | 4204 | int err = 0, err2; |
819c4920 TT |
4205 | handle_t *handle; |
4206 | struct address_space *mapping = inode->i_mapping; | |
819c4920 | 4207 | |
19b5ef61 TT |
4208 | /* |
4209 | * There is a possibility that we're either freeing the inode | |
e04027e8 | 4210 | * or it's a completely new inode. In those cases we might not |
19b5ef61 TT |
4211 | * have i_mutex locked because it's not necessary. |
4212 | */ | |
4213 | if (!(inode->i_state & (I_NEW|I_FREEING))) | |
5955102c | 4214 | WARN_ON(!inode_is_locked(inode)); |
0562e0ba JZ |
4215 | trace_ext4_truncate_enter(inode); |
4216 | ||
91ef4caf | 4217 | if (!ext4_can_truncate(inode)) |
9a5d265f | 4218 | goto out_trace; |
ac27a0ec | 4219 | |
5534fb5b | 4220 | if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC)) |
19f5fb7a | 4221 | ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE); |
7d8f9f7d | 4222 | |
aef1c851 TM |
4223 | if (ext4_has_inline_data(inode)) { |
4224 | int has_inline = 1; | |
4225 | ||
01daf945 | 4226 | err = ext4_inline_data_truncate(inode, &has_inline); |
9a5d265f | 4227 | if (err || has_inline) |
4228 | goto out_trace; | |
aef1c851 TM |
4229 | } |
4230 | ||
a361293f JK |
4231 | /* If we zero-out tail of the page, we have to create jinode for jbd2 */ |
4232 | if (inode->i_size & (inode->i_sb->s_blocksize - 1)) { | |
4233 | if (ext4_inode_attach_jinode(inode) < 0) | |
9a5d265f | 4234 | goto out_trace; |
a361293f JK |
4235 | } |
4236 | ||
819c4920 TT |
4237 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
4238 | credits = ext4_writepage_trans_blocks(inode); | |
4239 | else | |
4240 | credits = ext4_blocks_for_truncate(inode); | |
4241 | ||
4242 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); | |
9a5d265f | 4243 | if (IS_ERR(handle)) { |
4244 | err = PTR_ERR(handle); | |
4245 | goto out_trace; | |
4246 | } | |
819c4920 | 4247 | |
eb3544c6 LC |
4248 | if (inode->i_size & (inode->i_sb->s_blocksize - 1)) |
4249 | ext4_block_truncate_page(handle, mapping, inode->i_size); | |
819c4920 TT |
4250 | |
4251 | /* | |
4252 | * We add the inode to the orphan list, so that if this | |
4253 | * truncate spans multiple transactions, and we crash, we will | |
4254 | * resume the truncate when the filesystem recovers. It also | |
4255 | * marks the inode dirty, to catch the new size. | |
4256 | * | |
4257 | * Implication: the file must always be in a sane, consistent | |
4258 | * truncatable state while each transaction commits. | |
4259 | */ | |
2c98eb5e TT |
4260 | err = ext4_orphan_add(handle, inode); |
4261 | if (err) | |
819c4920 TT |
4262 | goto out_stop; |
4263 | ||
4264 | down_write(&EXT4_I(inode)->i_data_sem); | |
4265 | ||
27bc446e | 4266 | ext4_discard_preallocations(inode, 0); |
819c4920 | 4267 | |
ff9893dc | 4268 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
d0abb36d | 4269 | err = ext4_ext_truncate(handle, inode); |
ff9893dc | 4270 | else |
819c4920 TT |
4271 | ext4_ind_truncate(handle, inode); |
4272 | ||
4273 | up_write(&ei->i_data_sem); | |
d0abb36d TT |
4274 | if (err) |
4275 | goto out_stop; | |
819c4920 TT |
4276 | |
4277 | if (IS_SYNC(inode)) | |
4278 | ext4_handle_sync(handle); | |
4279 | ||
4280 | out_stop: | |
4281 | /* | |
4282 | * If this was a simple ftruncate() and the file will remain alive, | |
4283 | * then we need to clear up the orphan record which we created above. | |
4284 | * However, if this was a real unlink then we were called by | |
58d86a50 | 4285 | * ext4_evict_inode(), and we allow that function to clean up the |
819c4920 TT |
4286 | * orphan info for us. |
4287 | */ | |
4288 | if (inode->i_nlink) | |
4289 | ext4_orphan_del(handle, inode); | |
4290 | ||
eeca7ea1 | 4291 | inode->i_mtime = inode->i_ctime = current_time(inode); |
4209ae12 HS |
4292 | err2 = ext4_mark_inode_dirty(handle, inode); |
4293 | if (unlikely(err2 && !err)) | |
4294 | err = err2; | |
819c4920 | 4295 | ext4_journal_stop(handle); |
ac27a0ec | 4296 | |
9a5d265f | 4297 | out_trace: |
0562e0ba | 4298 | trace_ext4_truncate_exit(inode); |
2c98eb5e | 4299 | return err; |
ac27a0ec DK |
4300 | } |
4301 | ||
ac27a0ec | 4302 | /* |
617ba13b | 4303 | * ext4_get_inode_loc returns with an extra refcount against the inode's |
ac27a0ec DK |
4304 | * underlying buffer_head on success. If 'in_mem' is true, we have all |
4305 | * data in memory that is needed to recreate the on-disk version of this | |
4306 | * inode. | |
4307 | */ | |
8016e29f HS |
4308 | static int __ext4_get_inode_loc(struct super_block *sb, unsigned long ino, |
4309 | struct ext4_iloc *iloc, int in_mem, | |
4310 | ext4_fsblk_t *ret_block) | |
ac27a0ec | 4311 | { |
240799cd TT |
4312 | struct ext4_group_desc *gdp; |
4313 | struct buffer_head *bh; | |
240799cd | 4314 | ext4_fsblk_t block; |
02f03c42 | 4315 | struct blk_plug plug; |
240799cd TT |
4316 | int inodes_per_block, inode_offset; |
4317 | ||
3a06d778 | 4318 | iloc->bh = NULL; |
8016e29f HS |
4319 | if (ino < EXT4_ROOT_INO || |
4320 | ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count)) | |
6a797d27 | 4321 | return -EFSCORRUPTED; |
ac27a0ec | 4322 | |
8016e29f | 4323 | iloc->block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb); |
240799cd TT |
4324 | gdp = ext4_get_group_desc(sb, iloc->block_group, NULL); |
4325 | if (!gdp) | |
ac27a0ec DK |
4326 | return -EIO; |
4327 | ||
240799cd TT |
4328 | /* |
4329 | * Figure out the offset within the block group inode table | |
4330 | */ | |
00d09882 | 4331 | inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; |
8016e29f | 4332 | inode_offset = ((ino - 1) % |
240799cd TT |
4333 | EXT4_INODES_PER_GROUP(sb)); |
4334 | block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block); | |
4335 | iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb); | |
4336 | ||
4337 | bh = sb_getblk(sb, block); | |
aebf0243 | 4338 | if (unlikely(!bh)) |
860d21e2 | 4339 | return -ENOMEM; |
ac27a0ec DK |
4340 | if (!buffer_uptodate(bh)) { |
4341 | lock_buffer(bh); | |
9c83a923 | 4342 | |
60c776e5 | 4343 | if (ext4_buffer_uptodate(bh)) { |
ac27a0ec DK |
4344 | /* someone brought it uptodate while we waited */ |
4345 | unlock_buffer(bh); | |
4346 | goto has_buffer; | |
4347 | } | |
4348 | ||
4349 | /* | |
4350 | * If we have all information of the inode in memory and this | |
4351 | * is the only valid inode in the block, we need not read the | |
4352 | * block. | |
4353 | */ | |
4354 | if (in_mem) { | |
4355 | struct buffer_head *bitmap_bh; | |
240799cd | 4356 | int i, start; |
ac27a0ec | 4357 | |
240799cd | 4358 | start = inode_offset & ~(inodes_per_block - 1); |
ac27a0ec | 4359 | |
240799cd TT |
4360 | /* Is the inode bitmap in cache? */ |
4361 | bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp)); | |
aebf0243 | 4362 | if (unlikely(!bitmap_bh)) |
ac27a0ec DK |
4363 | goto make_io; |
4364 | ||
4365 | /* | |
4366 | * If the inode bitmap isn't in cache then the | |
4367 | * optimisation may end up performing two reads instead | |
4368 | * of one, so skip it. | |
4369 | */ | |
4370 | if (!buffer_uptodate(bitmap_bh)) { | |
4371 | brelse(bitmap_bh); | |
4372 | goto make_io; | |
4373 | } | |
240799cd | 4374 | for (i = start; i < start + inodes_per_block; i++) { |
ac27a0ec DK |
4375 | if (i == inode_offset) |
4376 | continue; | |
617ba13b | 4377 | if (ext4_test_bit(i, bitmap_bh->b_data)) |
ac27a0ec DK |
4378 | break; |
4379 | } | |
4380 | brelse(bitmap_bh); | |
240799cd | 4381 | if (i == start + inodes_per_block) { |
ac27a0ec DK |
4382 | /* all other inodes are free, so skip I/O */ |
4383 | memset(bh->b_data, 0, bh->b_size); | |
4384 | set_buffer_uptodate(bh); | |
4385 | unlock_buffer(bh); | |
4386 | goto has_buffer; | |
4387 | } | |
4388 | } | |
4389 | ||
4390 | make_io: | |
240799cd TT |
4391 | /* |
4392 | * If we need to do any I/O, try to pre-readahead extra | |
4393 | * blocks from the inode table. | |
4394 | */ | |
02f03c42 | 4395 | blk_start_plug(&plug); |
240799cd TT |
4396 | if (EXT4_SB(sb)->s_inode_readahead_blks) { |
4397 | ext4_fsblk_t b, end, table; | |
4398 | unsigned num; | |
0d606e2c | 4399 | __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks; |
240799cd TT |
4400 | |
4401 | table = ext4_inode_table(sb, gdp); | |
b713a5ec | 4402 | /* s_inode_readahead_blks is always a power of 2 */ |
0d606e2c | 4403 | b = block & ~((ext4_fsblk_t) ra_blks - 1); |
240799cd TT |
4404 | if (table > b) |
4405 | b = table; | |
0d606e2c | 4406 | end = b + ra_blks; |
240799cd | 4407 | num = EXT4_INODES_PER_GROUP(sb); |
feb0ab32 | 4408 | if (ext4_has_group_desc_csum(sb)) |
560671a0 | 4409 | num -= ext4_itable_unused_count(sb, gdp); |
240799cd TT |
4410 | table += num / inodes_per_block; |
4411 | if (end > table) | |
4412 | end = table; | |
4413 | while (b <= end) | |
5df1d412 | 4414 | ext4_sb_breadahead_unmovable(sb, b++); |
240799cd TT |
4415 | } |
4416 | ||
ac27a0ec DK |
4417 | /* |
4418 | * There are other valid inodes in the buffer, this inode | |
4419 | * has in-inode xattrs, or we don't have this inode in memory. | |
4420 | * Read the block from disk. | |
4421 | */ | |
8016e29f | 4422 | trace_ext4_load_inode(sb, ino); |
2d069c08 | 4423 | ext4_read_bh_nowait(bh, REQ_META | REQ_PRIO, NULL); |
02f03c42 | 4424 | blk_finish_plug(&plug); |
ac27a0ec | 4425 | wait_on_buffer(bh); |
0904c9ae | 4426 | ext4_simulate_fail_bh(sb, bh, EXT4_SIM_INODE_EIO); |
ac27a0ec | 4427 | if (!buffer_uptodate(bh)) { |
8016e29f HS |
4428 | if (ret_block) |
4429 | *ret_block = block; | |
ac27a0ec DK |
4430 | brelse(bh); |
4431 | return -EIO; | |
4432 | } | |
4433 | } | |
4434 | has_buffer: | |
4435 | iloc->bh = bh; | |
4436 | return 0; | |
4437 | } | |
4438 | ||
8016e29f HS |
4439 | static int __ext4_get_inode_loc_noinmem(struct inode *inode, |
4440 | struct ext4_iloc *iloc) | |
4441 | { | |
4442 | ext4_fsblk_t err_blk; | |
4443 | int ret; | |
4444 | ||
4445 | ret = __ext4_get_inode_loc(inode->i_sb, inode->i_ino, iloc, 0, | |
4446 | &err_blk); | |
4447 | ||
4448 | if (ret == -EIO) | |
4449 | ext4_error_inode_block(inode, err_blk, EIO, | |
4450 | "unable to read itable block"); | |
4451 | ||
4452 | return ret; | |
4453 | } | |
4454 | ||
617ba13b | 4455 | int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec | 4456 | { |
8016e29f HS |
4457 | ext4_fsblk_t err_blk; |
4458 | int ret; | |
4459 | ||
ac27a0ec | 4460 | /* We have all inode data except xattrs in memory here. */ |
8016e29f HS |
4461 | ret = __ext4_get_inode_loc(inode->i_sb, inode->i_ino, iloc, |
4462 | !ext4_test_inode_state(inode, EXT4_STATE_XATTR), &err_blk); | |
4463 | ||
4464 | if (ret == -EIO) | |
4465 | ext4_error_inode_block(inode, err_blk, EIO, | |
4466 | "unable to read itable block"); | |
4467 | ||
4468 | return ret; | |
4469 | } | |
4470 | ||
4471 | ||
4472 | int ext4_get_fc_inode_loc(struct super_block *sb, unsigned long ino, | |
4473 | struct ext4_iloc *iloc) | |
4474 | { | |
4475 | return __ext4_get_inode_loc(sb, ino, iloc, 0, NULL); | |
ac27a0ec DK |
4476 | } |
4477 | ||
a8ab6d38 | 4478 | static bool ext4_should_enable_dax(struct inode *inode) |
6642586b | 4479 | { |
a8ab6d38 IW |
4480 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
4481 | ||
9cb20f94 | 4482 | if (test_opt2(inode->i_sb, DAX_NEVER)) |
6642586b RZ |
4483 | return false; |
4484 | if (!S_ISREG(inode->i_mode)) | |
4485 | return false; | |
4486 | if (ext4_should_journal_data(inode)) | |
4487 | return false; | |
4488 | if (ext4_has_inline_data(inode)) | |
4489 | return false; | |
592ddec7 | 4490 | if (ext4_test_inode_flag(inode, EXT4_INODE_ENCRYPT)) |
6642586b | 4491 | return false; |
c93d8f88 EB |
4492 | if (ext4_test_inode_flag(inode, EXT4_INODE_VERITY)) |
4493 | return false; | |
a8ab6d38 IW |
4494 | if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags)) |
4495 | return false; | |
4496 | if (test_opt(inode->i_sb, DAX_ALWAYS)) | |
4497 | return true; | |
4498 | ||
b383a73f | 4499 | return ext4_test_inode_flag(inode, EXT4_INODE_DAX); |
6642586b RZ |
4500 | } |
4501 | ||
043546e4 | 4502 | void ext4_set_inode_flags(struct inode *inode, bool init) |
ac27a0ec | 4503 | { |
617ba13b | 4504 | unsigned int flags = EXT4_I(inode)->i_flags; |
00a1a053 | 4505 | unsigned int new_fl = 0; |
ac27a0ec | 4506 | |
043546e4 IW |
4507 | WARN_ON_ONCE(IS_DAX(inode) && init); |
4508 | ||
617ba13b | 4509 | if (flags & EXT4_SYNC_FL) |
00a1a053 | 4510 | new_fl |= S_SYNC; |
617ba13b | 4511 | if (flags & EXT4_APPEND_FL) |
00a1a053 | 4512 | new_fl |= S_APPEND; |
617ba13b | 4513 | if (flags & EXT4_IMMUTABLE_FL) |
00a1a053 | 4514 | new_fl |= S_IMMUTABLE; |
617ba13b | 4515 | if (flags & EXT4_NOATIME_FL) |
00a1a053 | 4516 | new_fl |= S_NOATIME; |
617ba13b | 4517 | if (flags & EXT4_DIRSYNC_FL) |
00a1a053 | 4518 | new_fl |= S_DIRSYNC; |
043546e4 IW |
4519 | |
4520 | /* Because of the way inode_set_flags() works we must preserve S_DAX | |
4521 | * here if already set. */ | |
4522 | new_fl |= (inode->i_flags & S_DAX); | |
4523 | if (init && ext4_should_enable_dax(inode)) | |
923ae0ff | 4524 | new_fl |= S_DAX; |
043546e4 | 4525 | |
2ee6a576 EB |
4526 | if (flags & EXT4_ENCRYPT_FL) |
4527 | new_fl |= S_ENCRYPTED; | |
b886ee3e GKB |
4528 | if (flags & EXT4_CASEFOLD_FL) |
4529 | new_fl |= S_CASEFOLD; | |
c93d8f88 EB |
4530 | if (flags & EXT4_VERITY_FL) |
4531 | new_fl |= S_VERITY; | |
5f16f322 | 4532 | inode_set_flags(inode, new_fl, |
2ee6a576 | 4533 | S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX| |
c93d8f88 | 4534 | S_ENCRYPTED|S_CASEFOLD|S_VERITY); |
ac27a0ec DK |
4535 | } |
4536 | ||
0fc1b451 | 4537 | static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode, |
de9a55b8 | 4538 | struct ext4_inode_info *ei) |
0fc1b451 AK |
4539 | { |
4540 | blkcnt_t i_blocks ; | |
8180a562 AK |
4541 | struct inode *inode = &(ei->vfs_inode); |
4542 | struct super_block *sb = inode->i_sb; | |
0fc1b451 | 4543 | |
e2b911c5 | 4544 | if (ext4_has_feature_huge_file(sb)) { |
0fc1b451 AK |
4545 | /* we are using combined 48 bit field */ |
4546 | i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 | | |
4547 | le32_to_cpu(raw_inode->i_blocks_lo); | |
07a03824 | 4548 | if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) { |
8180a562 AK |
4549 | /* i_blocks represent file system block size */ |
4550 | return i_blocks << (inode->i_blkbits - 9); | |
4551 | } else { | |
4552 | return i_blocks; | |
4553 | } | |
0fc1b451 AK |
4554 | } else { |
4555 | return le32_to_cpu(raw_inode->i_blocks_lo); | |
4556 | } | |
4557 | } | |
ff9ddf7e | 4558 | |
eb9b5f01 | 4559 | static inline int ext4_iget_extra_inode(struct inode *inode, |
152a7b0a TM |
4560 | struct ext4_inode *raw_inode, |
4561 | struct ext4_inode_info *ei) | |
4562 | { | |
4563 | __le32 *magic = (void *)raw_inode + | |
4564 | EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize; | |
eb9b5f01 | 4565 | |
290ab230 EB |
4566 | if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize + sizeof(__le32) <= |
4567 | EXT4_INODE_SIZE(inode->i_sb) && | |
4568 | *magic == cpu_to_le32(EXT4_XATTR_MAGIC)) { | |
152a7b0a | 4569 | ext4_set_inode_state(inode, EXT4_STATE_XATTR); |
eb9b5f01 | 4570 | return ext4_find_inline_data_nolock(inode); |
f19d5870 TM |
4571 | } else |
4572 | EXT4_I(inode)->i_inline_off = 0; | |
eb9b5f01 | 4573 | return 0; |
152a7b0a TM |
4574 | } |
4575 | ||
040cb378 LX |
4576 | int ext4_get_projid(struct inode *inode, kprojid_t *projid) |
4577 | { | |
0b7b7779 | 4578 | if (!ext4_has_feature_project(inode->i_sb)) |
040cb378 LX |
4579 | return -EOPNOTSUPP; |
4580 | *projid = EXT4_I(inode)->i_projid; | |
4581 | return 0; | |
4582 | } | |
4583 | ||
e254d1af EG |
4584 | /* |
4585 | * ext4 has self-managed i_version for ea inodes, it stores the lower 32bit of | |
4586 | * refcount in i_version, so use raw values if inode has EXT4_EA_INODE_FL flag | |
4587 | * set. | |
4588 | */ | |
4589 | static inline void ext4_inode_set_iversion_queried(struct inode *inode, u64 val) | |
4590 | { | |
4591 | if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) | |
4592 | inode_set_iversion_raw(inode, val); | |
4593 | else | |
4594 | inode_set_iversion_queried(inode, val); | |
4595 | } | |
4596 | static inline u64 ext4_inode_peek_iversion(const struct inode *inode) | |
4597 | { | |
4598 | if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) | |
4599 | return inode_peek_iversion_raw(inode); | |
4600 | else | |
4601 | return inode_peek_iversion(inode); | |
4602 | } | |
4603 | ||
8a363970 TT |
4604 | struct inode *__ext4_iget(struct super_block *sb, unsigned long ino, |
4605 | ext4_iget_flags flags, const char *function, | |
4606 | unsigned int line) | |
ac27a0ec | 4607 | { |
617ba13b MC |
4608 | struct ext4_iloc iloc; |
4609 | struct ext4_inode *raw_inode; | |
1d1fe1ee | 4610 | struct ext4_inode_info *ei; |
bd2c38cf | 4611 | struct ext4_super_block *es = EXT4_SB(sb)->s_es; |
1d1fe1ee | 4612 | struct inode *inode; |
b436b9be | 4613 | journal_t *journal = EXT4_SB(sb)->s_journal; |
1d1fe1ee | 4614 | long ret; |
7e6e1ef4 | 4615 | loff_t size; |
ac27a0ec | 4616 | int block; |
08cefc7a EB |
4617 | uid_t i_uid; |
4618 | gid_t i_gid; | |
040cb378 | 4619 | projid_t i_projid; |
ac27a0ec | 4620 | |
191ce178 | 4621 | if ((!(flags & EXT4_IGET_SPECIAL) && |
bd2c38cf JK |
4622 | ((ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO) || |
4623 | ino == le32_to_cpu(es->s_usr_quota_inum) || | |
4624 | ino == le32_to_cpu(es->s_grp_quota_inum) || | |
02f310fc JK |
4625 | ino == le32_to_cpu(es->s_prj_quota_inum) || |
4626 | ino == le32_to_cpu(es->s_orphan_file_inum))) || | |
8a363970 | 4627 | (ino < EXT4_ROOT_INO) || |
bd2c38cf | 4628 | (ino > le32_to_cpu(es->s_inodes_count))) { |
8a363970 TT |
4629 | if (flags & EXT4_IGET_HANDLE) |
4630 | return ERR_PTR(-ESTALE); | |
014c9caa | 4631 | __ext4_error(sb, function, line, false, EFSCORRUPTED, 0, |
8a363970 TT |
4632 | "inode #%lu: comm %s: iget: illegal inode #", |
4633 | ino, current->comm); | |
4634 | return ERR_PTR(-EFSCORRUPTED); | |
4635 | } | |
4636 | ||
1d1fe1ee DH |
4637 | inode = iget_locked(sb, ino); |
4638 | if (!inode) | |
4639 | return ERR_PTR(-ENOMEM); | |
4640 | if (!(inode->i_state & I_NEW)) | |
4641 | return inode; | |
4642 | ||
4643 | ei = EXT4_I(inode); | |
7dc57615 | 4644 | iloc.bh = NULL; |
ac27a0ec | 4645 | |
8016e29f | 4646 | ret = __ext4_get_inode_loc_noinmem(inode, &iloc); |
1d1fe1ee | 4647 | if (ret < 0) |
ac27a0ec | 4648 | goto bad_inode; |
617ba13b | 4649 | raw_inode = ext4_raw_inode(&iloc); |
814525f4 | 4650 | |
8e4b5eae | 4651 | if ((ino == EXT4_ROOT_INO) && (raw_inode->i_links_count == 0)) { |
8a363970 TT |
4652 | ext4_error_inode(inode, function, line, 0, |
4653 | "iget: root inode unallocated"); | |
8e4b5eae TT |
4654 | ret = -EFSCORRUPTED; |
4655 | goto bad_inode; | |
4656 | } | |
4657 | ||
8a363970 TT |
4658 | if ((flags & EXT4_IGET_HANDLE) && |
4659 | (raw_inode->i_links_count == 0) && (raw_inode->i_mode == 0)) { | |
4660 | ret = -ESTALE; | |
4661 | goto bad_inode; | |
4662 | } | |
4663 | ||
814525f4 DW |
4664 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
4665 | ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize); | |
4666 | if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > | |
2dc8d9e1 EB |
4667 | EXT4_INODE_SIZE(inode->i_sb) || |
4668 | (ei->i_extra_isize & 3)) { | |
8a363970 TT |
4669 | ext4_error_inode(inode, function, line, 0, |
4670 | "iget: bad extra_isize %u " | |
4671 | "(inode size %u)", | |
2dc8d9e1 EB |
4672 | ei->i_extra_isize, |
4673 | EXT4_INODE_SIZE(inode->i_sb)); | |
6a797d27 | 4674 | ret = -EFSCORRUPTED; |
814525f4 DW |
4675 | goto bad_inode; |
4676 | } | |
4677 | } else | |
4678 | ei->i_extra_isize = 0; | |
4679 | ||
4680 | /* Precompute checksum seed for inode metadata */ | |
9aa5d32b | 4681 | if (ext4_has_metadata_csum(sb)) { |
814525f4 DW |
4682 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
4683 | __u32 csum; | |
4684 | __le32 inum = cpu_to_le32(inode->i_ino); | |
4685 | __le32 gen = raw_inode->i_generation; | |
4686 | csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum, | |
4687 | sizeof(inum)); | |
4688 | ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen, | |
4689 | sizeof(gen)); | |
4690 | } | |
4691 | ||
8016e29f HS |
4692 | if ((!ext4_inode_csum_verify(inode, raw_inode, ei) || |
4693 | ext4_simulate_fail(sb, EXT4_SIM_INODE_CRC)) && | |
4694 | (!(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY))) { | |
4695 | ext4_error_inode_err(inode, function, line, 0, | |
4696 | EFSBADCRC, "iget: checksum invalid"); | |
6a797d27 | 4697 | ret = -EFSBADCRC; |
814525f4 DW |
4698 | goto bad_inode; |
4699 | } | |
4700 | ||
ac27a0ec | 4701 | inode->i_mode = le16_to_cpu(raw_inode->i_mode); |
08cefc7a EB |
4702 | i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low); |
4703 | i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low); | |
0b7b7779 | 4704 | if (ext4_has_feature_project(sb) && |
040cb378 LX |
4705 | EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE && |
4706 | EXT4_FITS_IN_INODE(raw_inode, ei, i_projid)) | |
4707 | i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid); | |
4708 | else | |
4709 | i_projid = EXT4_DEF_PROJID; | |
4710 | ||
af5bc92d | 4711 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
08cefc7a EB |
4712 | i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16; |
4713 | i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16; | |
ac27a0ec | 4714 | } |
08cefc7a EB |
4715 | i_uid_write(inode, i_uid); |
4716 | i_gid_write(inode, i_gid); | |
040cb378 | 4717 | ei->i_projid = make_kprojid(&init_user_ns, i_projid); |
bfe86848 | 4718 | set_nlink(inode, le16_to_cpu(raw_inode->i_links_count)); |
ac27a0ec | 4719 | |
353eb83c | 4720 | ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */ |
67cf5b09 | 4721 | ei->i_inline_off = 0; |
ac27a0ec DK |
4722 | ei->i_dir_start_lookup = 0; |
4723 | ei->i_dtime = le32_to_cpu(raw_inode->i_dtime); | |
4724 | /* We now have enough fields to check if the inode was active or not. | |
4725 | * This is needed because nfsd might try to access dead inodes | |
4726 | * the test is that same one that e2fsck uses | |
4727 | * NeilBrown 1999oct15 | |
4728 | */ | |
4729 | if (inode->i_nlink == 0) { | |
393d1d1d DTB |
4730 | if ((inode->i_mode == 0 || |
4731 | !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) && | |
4732 | ino != EXT4_BOOT_LOADER_INO) { | |
ac27a0ec | 4733 | /* this inode is deleted */ |
1d1fe1ee | 4734 | ret = -ESTALE; |
ac27a0ec DK |
4735 | goto bad_inode; |
4736 | } | |
4737 | /* The only unlinked inodes we let through here have | |
4738 | * valid i_mode and are being read by the orphan | |
4739 | * recovery code: that's fine, we're about to complete | |
393d1d1d DTB |
4740 | * the process of deleting those. |
4741 | * OR it is the EXT4_BOOT_LOADER_INO which is | |
4742 | * not initialized on a new filesystem. */ | |
ac27a0ec | 4743 | } |
ac27a0ec | 4744 | ei->i_flags = le32_to_cpu(raw_inode->i_flags); |
043546e4 | 4745 | ext4_set_inode_flags(inode, true); |
0fc1b451 | 4746 | inode->i_blocks = ext4_inode_blocks(raw_inode, ei); |
7973c0c1 | 4747 | ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo); |
e2b911c5 | 4748 | if (ext4_has_feature_64bit(sb)) |
a1ddeb7e BP |
4749 | ei->i_file_acl |= |
4750 | ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32; | |
e08ac99f | 4751 | inode->i_size = ext4_isize(sb, raw_inode); |
7e6e1ef4 | 4752 | if ((size = i_size_read(inode)) < 0) { |
8a363970 TT |
4753 | ext4_error_inode(inode, function, line, 0, |
4754 | "iget: bad i_size value: %lld", size); | |
7e6e1ef4 DW |
4755 | ret = -EFSCORRUPTED; |
4756 | goto bad_inode; | |
4757 | } | |
48a34311 JK |
4758 | /* |
4759 | * If dir_index is not enabled but there's dir with INDEX flag set, | |
4760 | * we'd normally treat htree data as empty space. But with metadata | |
4761 | * checksumming that corrupts checksums so forbid that. | |
4762 | */ | |
4763 | if (!ext4_has_feature_dir_index(sb) && ext4_has_metadata_csum(sb) && | |
4764 | ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) { | |
4765 | ext4_error_inode(inode, function, line, 0, | |
4766 | "iget: Dir with htree data on filesystem without dir_index feature."); | |
4767 | ret = -EFSCORRUPTED; | |
4768 | goto bad_inode; | |
4769 | } | |
ac27a0ec | 4770 | ei->i_disksize = inode->i_size; |
a9e7f447 DM |
4771 | #ifdef CONFIG_QUOTA |
4772 | ei->i_reserved_quota = 0; | |
4773 | #endif | |
ac27a0ec DK |
4774 | inode->i_generation = le32_to_cpu(raw_inode->i_generation); |
4775 | ei->i_block_group = iloc.block_group; | |
a4912123 | 4776 | ei->i_last_alloc_group = ~0; |
ac27a0ec DK |
4777 | /* |
4778 | * NOTE! The in-memory inode i_data array is in little-endian order | |
4779 | * even on big-endian machines: we do NOT byteswap the block numbers! | |
4780 | */ | |
617ba13b | 4781 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
ac27a0ec DK |
4782 | ei->i_data[block] = raw_inode->i_block[block]; |
4783 | INIT_LIST_HEAD(&ei->i_orphan); | |
aa75f4d3 | 4784 | ext4_fc_init_inode(&ei->vfs_inode); |
ac27a0ec | 4785 | |
b436b9be JK |
4786 | /* |
4787 | * Set transaction id's of transactions that have to be committed | |
4788 | * to finish f[data]sync. We set them to currently running transaction | |
4789 | * as we cannot be sure that the inode or some of its metadata isn't | |
4790 | * part of the transaction - the inode could have been reclaimed and | |
4791 | * now it is reread from disk. | |
4792 | */ | |
4793 | if (journal) { | |
4794 | transaction_t *transaction; | |
4795 | tid_t tid; | |
4796 | ||
a931da6a | 4797 | read_lock(&journal->j_state_lock); |
b436b9be JK |
4798 | if (journal->j_running_transaction) |
4799 | transaction = journal->j_running_transaction; | |
4800 | else | |
4801 | transaction = journal->j_committing_transaction; | |
4802 | if (transaction) | |
4803 | tid = transaction->t_tid; | |
4804 | else | |
4805 | tid = journal->j_commit_sequence; | |
a931da6a | 4806 | read_unlock(&journal->j_state_lock); |
b436b9be JK |
4807 | ei->i_sync_tid = tid; |
4808 | ei->i_datasync_tid = tid; | |
4809 | } | |
4810 | ||
0040d987 | 4811 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
ac27a0ec DK |
4812 | if (ei->i_extra_isize == 0) { |
4813 | /* The extra space is currently unused. Use it. */ | |
2dc8d9e1 | 4814 | BUILD_BUG_ON(sizeof(struct ext4_inode) & 3); |
617ba13b MC |
4815 | ei->i_extra_isize = sizeof(struct ext4_inode) - |
4816 | EXT4_GOOD_OLD_INODE_SIZE; | |
ac27a0ec | 4817 | } else { |
eb9b5f01 TT |
4818 | ret = ext4_iget_extra_inode(inode, raw_inode, ei); |
4819 | if (ret) | |
4820 | goto bad_inode; | |
ac27a0ec | 4821 | } |
814525f4 | 4822 | } |
ac27a0ec | 4823 | |
ef7f3835 KS |
4824 | EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode); |
4825 | EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode); | |
4826 | EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode); | |
4827 | EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode); | |
4828 | ||
ed3654eb | 4829 | if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) { |
ee73f9a5 JL |
4830 | u64 ivers = le32_to_cpu(raw_inode->i_disk_version); |
4831 | ||
c4f65706 TT |
4832 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
4833 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
ee73f9a5 | 4834 | ivers |= |
c4f65706 TT |
4835 | (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32; |
4836 | } | |
e254d1af | 4837 | ext4_inode_set_iversion_queried(inode, ivers); |
25ec56b5 JNC |
4838 | } |
4839 | ||
c4b5a614 | 4840 | ret = 0; |
485c26ec | 4841 | if (ei->i_file_acl && |
ce9f24cc | 4842 | !ext4_inode_block_valid(inode, ei->i_file_acl, 1)) { |
8a363970 TT |
4843 | ext4_error_inode(inode, function, line, 0, |
4844 | "iget: bad extended attribute block %llu", | |
24676da4 | 4845 | ei->i_file_acl); |
6a797d27 | 4846 | ret = -EFSCORRUPTED; |
485c26ec | 4847 | goto bad_inode; |
f19d5870 | 4848 | } else if (!ext4_has_inline_data(inode)) { |
bc716523 | 4849 | /* validate the block references in the inode */ |
8016e29f HS |
4850 | if (!(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY) && |
4851 | (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || | |
4852 | (S_ISLNK(inode->i_mode) && | |
4853 | !ext4_inode_is_fast_symlink(inode)))) { | |
bc716523 | 4854 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
f19d5870 | 4855 | ret = ext4_ext_check_inode(inode); |
bc716523 LS |
4856 | else |
4857 | ret = ext4_ind_check_inode(inode); | |
f19d5870 | 4858 | } |
fe2c8191 | 4859 | } |
567f3e9a | 4860 | if (ret) |
de9a55b8 | 4861 | goto bad_inode; |
7a262f7c | 4862 | |
ac27a0ec | 4863 | if (S_ISREG(inode->i_mode)) { |
617ba13b | 4864 | inode->i_op = &ext4_file_inode_operations; |
be64f884 | 4865 | inode->i_fop = &ext4_file_operations; |
617ba13b | 4866 | ext4_set_aops(inode); |
ac27a0ec | 4867 | } else if (S_ISDIR(inode->i_mode)) { |
617ba13b MC |
4868 | inode->i_op = &ext4_dir_inode_operations; |
4869 | inode->i_fop = &ext4_dir_operations; | |
ac27a0ec | 4870 | } else if (S_ISLNK(inode->i_mode)) { |
6390d33b LR |
4871 | /* VFS does not allow setting these so must be corruption */ |
4872 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) { | |
8a363970 TT |
4873 | ext4_error_inode(inode, function, line, 0, |
4874 | "iget: immutable or append flags " | |
4875 | "not allowed on symlinks"); | |
6390d33b LR |
4876 | ret = -EFSCORRUPTED; |
4877 | goto bad_inode; | |
4878 | } | |
592ddec7 | 4879 | if (IS_ENCRYPTED(inode)) { |
a7a67e8a AV |
4880 | inode->i_op = &ext4_encrypted_symlink_inode_operations; |
4881 | ext4_set_aops(inode); | |
4882 | } else if (ext4_inode_is_fast_symlink(inode)) { | |
75e7566b | 4883 | inode->i_link = (char *)ei->i_data; |
617ba13b | 4884 | inode->i_op = &ext4_fast_symlink_inode_operations; |
e83c1397 DG |
4885 | nd_terminate_link(ei->i_data, inode->i_size, |
4886 | sizeof(ei->i_data) - 1); | |
4887 | } else { | |
617ba13b MC |
4888 | inode->i_op = &ext4_symlink_inode_operations; |
4889 | ext4_set_aops(inode); | |
ac27a0ec | 4890 | } |
21fc61c7 | 4891 | inode_nohighmem(inode); |
563bdd61 TT |
4892 | } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) || |
4893 | S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) { | |
617ba13b | 4894 | inode->i_op = &ext4_special_inode_operations; |
ac27a0ec DK |
4895 | if (raw_inode->i_block[0]) |
4896 | init_special_inode(inode, inode->i_mode, | |
4897 | old_decode_dev(le32_to_cpu(raw_inode->i_block[0]))); | |
4898 | else | |
4899 | init_special_inode(inode, inode->i_mode, | |
4900 | new_decode_dev(le32_to_cpu(raw_inode->i_block[1]))); | |
393d1d1d DTB |
4901 | } else if (ino == EXT4_BOOT_LOADER_INO) { |
4902 | make_bad_inode(inode); | |
563bdd61 | 4903 | } else { |
6a797d27 | 4904 | ret = -EFSCORRUPTED; |
8a363970 TT |
4905 | ext4_error_inode(inode, function, line, 0, |
4906 | "iget: bogus i_mode (%o)", inode->i_mode); | |
563bdd61 | 4907 | goto bad_inode; |
ac27a0ec | 4908 | } |
6456ca65 TT |
4909 | if (IS_CASEFOLDED(inode) && !ext4_has_feature_casefold(inode->i_sb)) |
4910 | ext4_error_inode(inode, function, line, 0, | |
4911 | "casefold flag without casefold feature"); | |
af5bc92d | 4912 | brelse(iloc.bh); |
dec214d0 | 4913 | |
1d1fe1ee DH |
4914 | unlock_new_inode(inode); |
4915 | return inode; | |
ac27a0ec DK |
4916 | |
4917 | bad_inode: | |
567f3e9a | 4918 | brelse(iloc.bh); |
1d1fe1ee DH |
4919 | iget_failed(inode); |
4920 | return ERR_PTR(ret); | |
ac27a0ec DK |
4921 | } |
4922 | ||
0fc1b451 AK |
4923 | static int ext4_inode_blocks_set(handle_t *handle, |
4924 | struct ext4_inode *raw_inode, | |
4925 | struct ext4_inode_info *ei) | |
4926 | { | |
4927 | struct inode *inode = &(ei->vfs_inode); | |
28936b62 | 4928 | u64 i_blocks = READ_ONCE(inode->i_blocks); |
0fc1b451 | 4929 | struct super_block *sb = inode->i_sb; |
0fc1b451 AK |
4930 | |
4931 | if (i_blocks <= ~0U) { | |
4932 | /* | |
4907cb7b | 4933 | * i_blocks can be represented in a 32 bit variable |
0fc1b451 AK |
4934 | * as multiple of 512 bytes |
4935 | */ | |
8180a562 | 4936 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 4937 | raw_inode->i_blocks_high = 0; |
84a8dce2 | 4938 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
f287a1a5 TT |
4939 | return 0; |
4940 | } | |
e2b911c5 | 4941 | if (!ext4_has_feature_huge_file(sb)) |
f287a1a5 TT |
4942 | return -EFBIG; |
4943 | ||
4944 | if (i_blocks <= 0xffffffffffffULL) { | |
0fc1b451 AK |
4945 | /* |
4946 | * i_blocks can be represented in a 48 bit variable | |
4947 | * as multiple of 512 bytes | |
4948 | */ | |
8180a562 | 4949 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 4950 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); |
84a8dce2 | 4951 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
0fc1b451 | 4952 | } else { |
84a8dce2 | 4953 | ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
8180a562 AK |
4954 | /* i_block is stored in file system block size */ |
4955 | i_blocks = i_blocks >> (inode->i_blkbits - 9); | |
4956 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); | |
4957 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); | |
0fc1b451 | 4958 | } |
f287a1a5 | 4959 | return 0; |
0fc1b451 AK |
4960 | } |
4961 | ||
3f19b2ab DH |
4962 | static void __ext4_update_other_inode_time(struct super_block *sb, |
4963 | unsigned long orig_ino, | |
4964 | unsigned long ino, | |
4965 | struct ext4_inode *raw_inode) | |
a26f4992 | 4966 | { |
3f19b2ab DH |
4967 | struct inode *inode; |
4968 | ||
4969 | inode = find_inode_by_ino_rcu(sb, ino); | |
4970 | if (!inode) | |
4971 | return; | |
a26f4992 | 4972 | |
ed296c6c | 4973 | if (!inode_is_dirtytime_only(inode)) |
3f19b2ab DH |
4974 | return; |
4975 | ||
a26f4992 | 4976 | spin_lock(&inode->i_lock); |
ed296c6c | 4977 | if (inode_is_dirtytime_only(inode)) { |
a26f4992 TT |
4978 | struct ext4_inode_info *ei = EXT4_I(inode); |
4979 | ||
5fcd5750 | 4980 | inode->i_state &= ~I_DIRTY_TIME; |
a26f4992 TT |
4981 | spin_unlock(&inode->i_lock); |
4982 | ||
4983 | spin_lock(&ei->i_raw_lock); | |
3f19b2ab DH |
4984 | EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode); |
4985 | EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode); | |
4986 | EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode); | |
4987 | ext4_inode_csum_set(inode, raw_inode, ei); | |
a26f4992 | 4988 | spin_unlock(&ei->i_raw_lock); |
3f19b2ab DH |
4989 | trace_ext4_other_inode_update_time(inode, orig_ino); |
4990 | return; | |
a26f4992 TT |
4991 | } |
4992 | spin_unlock(&inode->i_lock); | |
a26f4992 TT |
4993 | } |
4994 | ||
4995 | /* | |
4996 | * Opportunistically update the other time fields for other inodes in | |
4997 | * the same inode table block. | |
4998 | */ | |
4999 | static void ext4_update_other_inodes_time(struct super_block *sb, | |
5000 | unsigned long orig_ino, char *buf) | |
5001 | { | |
a26f4992 TT |
5002 | unsigned long ino; |
5003 | int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; | |
5004 | int inode_size = EXT4_INODE_SIZE(sb); | |
5005 | ||
0f0ff9a9 TT |
5006 | /* |
5007 | * Calculate the first inode in the inode table block. Inode | |
5008 | * numbers are one-based. That is, the first inode in a block | |
5009 | * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1). | |
5010 | */ | |
5011 | ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1; | |
3f19b2ab | 5012 | rcu_read_lock(); |
a26f4992 TT |
5013 | for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) { |
5014 | if (ino == orig_ino) | |
5015 | continue; | |
3f19b2ab DH |
5016 | __ext4_update_other_inode_time(sb, orig_ino, ino, |
5017 | (struct ext4_inode *)buf); | |
a26f4992 | 5018 | } |
3f19b2ab | 5019 | rcu_read_unlock(); |
a26f4992 TT |
5020 | } |
5021 | ||
ac27a0ec DK |
5022 | /* |
5023 | * Post the struct inode info into an on-disk inode location in the | |
5024 | * buffer-cache. This gobbles the caller's reference to the | |
5025 | * buffer_head in the inode location struct. | |
5026 | * | |
5027 | * The caller must have write access to iloc->bh. | |
5028 | */ | |
617ba13b | 5029 | static int ext4_do_update_inode(handle_t *handle, |
ac27a0ec | 5030 | struct inode *inode, |
830156c7 | 5031 | struct ext4_iloc *iloc) |
ac27a0ec | 5032 | { |
617ba13b MC |
5033 | struct ext4_inode *raw_inode = ext4_raw_inode(iloc); |
5034 | struct ext4_inode_info *ei = EXT4_I(inode); | |
ac27a0ec | 5035 | struct buffer_head *bh = iloc->bh; |
202ee5df | 5036 | struct super_block *sb = inode->i_sb; |
7d8bd3c7 | 5037 | int err = 0, block; |
202ee5df | 5038 | int need_datasync = 0, set_large_file = 0; |
08cefc7a EB |
5039 | uid_t i_uid; |
5040 | gid_t i_gid; | |
040cb378 | 5041 | projid_t i_projid; |
ac27a0ec | 5042 | |
202ee5df TT |
5043 | spin_lock(&ei->i_raw_lock); |
5044 | ||
5045 | /* For fields not tracked in the in-memory inode, | |
ac27a0ec | 5046 | * initialise them to zero for new inodes. */ |
19f5fb7a | 5047 | if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) |
617ba13b | 5048 | memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size); |
ac27a0ec | 5049 | |
13221811 LM |
5050 | err = ext4_inode_blocks_set(handle, raw_inode, ei); |
5051 | if (err) { | |
5052 | spin_unlock(&ei->i_raw_lock); | |
5053 | goto out_brelse; | |
5054 | } | |
5055 | ||
ac27a0ec | 5056 | raw_inode->i_mode = cpu_to_le16(inode->i_mode); |
08cefc7a EB |
5057 | i_uid = i_uid_read(inode); |
5058 | i_gid = i_gid_read(inode); | |
040cb378 | 5059 | i_projid = from_kprojid(&init_user_ns, ei->i_projid); |
af5bc92d | 5060 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
08cefc7a EB |
5061 | raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid)); |
5062 | raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid)); | |
ac27a0ec DK |
5063 | /* |
5064 | * Fix up interoperability with old kernels. Otherwise, old inodes get | |
5065 | * re-used with the upper 16 bits of the uid/gid intact | |
5066 | */ | |
93e3b4e6 DJ |
5067 | if (ei->i_dtime && list_empty(&ei->i_orphan)) { |
5068 | raw_inode->i_uid_high = 0; | |
5069 | raw_inode->i_gid_high = 0; | |
5070 | } else { | |
ac27a0ec | 5071 | raw_inode->i_uid_high = |
08cefc7a | 5072 | cpu_to_le16(high_16_bits(i_uid)); |
ac27a0ec | 5073 | raw_inode->i_gid_high = |
08cefc7a | 5074 | cpu_to_le16(high_16_bits(i_gid)); |
ac27a0ec DK |
5075 | } |
5076 | } else { | |
08cefc7a EB |
5077 | raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid)); |
5078 | raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid)); | |
ac27a0ec DK |
5079 | raw_inode->i_uid_high = 0; |
5080 | raw_inode->i_gid_high = 0; | |
5081 | } | |
5082 | raw_inode->i_links_count = cpu_to_le16(inode->i_nlink); | |
ef7f3835 KS |
5083 | |
5084 | EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode); | |
5085 | EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode); | |
5086 | EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode); | |
5087 | EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode); | |
5088 | ||
ac27a0ec | 5089 | raw_inode->i_dtime = cpu_to_le32(ei->i_dtime); |
353eb83c | 5090 | raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF); |
ed3654eb | 5091 | if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) |
a1ddeb7e BP |
5092 | raw_inode->i_file_acl_high = |
5093 | cpu_to_le16(ei->i_file_acl >> 32); | |
7973c0c1 | 5094 | raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl); |
dce8e237 | 5095 | if (READ_ONCE(ei->i_disksize) != ext4_isize(inode->i_sb, raw_inode)) { |
b71fc079 JK |
5096 | ext4_isize_set(raw_inode, ei->i_disksize); |
5097 | need_datasync = 1; | |
5098 | } | |
a48380f7 | 5099 | if (ei->i_disksize > 0x7fffffffULL) { |
e2b911c5 | 5100 | if (!ext4_has_feature_large_file(sb) || |
a48380f7 | 5101 | EXT4_SB(sb)->s_es->s_rev_level == |
202ee5df TT |
5102 | cpu_to_le32(EXT4_GOOD_OLD_REV)) |
5103 | set_large_file = 1; | |
ac27a0ec DK |
5104 | } |
5105 | raw_inode->i_generation = cpu_to_le32(inode->i_generation); | |
5106 | if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) { | |
5107 | if (old_valid_dev(inode->i_rdev)) { | |
5108 | raw_inode->i_block[0] = | |
5109 | cpu_to_le32(old_encode_dev(inode->i_rdev)); | |
5110 | raw_inode->i_block[1] = 0; | |
5111 | } else { | |
5112 | raw_inode->i_block[0] = 0; | |
5113 | raw_inode->i_block[1] = | |
5114 | cpu_to_le32(new_encode_dev(inode->i_rdev)); | |
5115 | raw_inode->i_block[2] = 0; | |
5116 | } | |
f19d5870 | 5117 | } else if (!ext4_has_inline_data(inode)) { |
de9a55b8 TT |
5118 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
5119 | raw_inode->i_block[block] = ei->i_data[block]; | |
f19d5870 | 5120 | } |
ac27a0ec | 5121 | |
ed3654eb | 5122 | if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) { |
e254d1af | 5123 | u64 ivers = ext4_inode_peek_iversion(inode); |
ee73f9a5 JL |
5124 | |
5125 | raw_inode->i_disk_version = cpu_to_le32(ivers); | |
c4f65706 TT |
5126 | if (ei->i_extra_isize) { |
5127 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
5128 | raw_inode->i_version_hi = | |
ee73f9a5 | 5129 | cpu_to_le32(ivers >> 32); |
c4f65706 TT |
5130 | raw_inode->i_extra_isize = |
5131 | cpu_to_le16(ei->i_extra_isize); | |
5132 | } | |
25ec56b5 | 5133 | } |
040cb378 | 5134 | |
0b7b7779 | 5135 | BUG_ON(!ext4_has_feature_project(inode->i_sb) && |
040cb378 LX |
5136 | i_projid != EXT4_DEF_PROJID); |
5137 | ||
5138 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && | |
5139 | EXT4_FITS_IN_INODE(raw_inode, ei, i_projid)) | |
5140 | raw_inode->i_projid = cpu_to_le32(i_projid); | |
5141 | ||
814525f4 | 5142 | ext4_inode_csum_set(inode, raw_inode, ei); |
202ee5df | 5143 | spin_unlock(&ei->i_raw_lock); |
1751e8a6 | 5144 | if (inode->i_sb->s_flags & SB_LAZYTIME) |
a26f4992 TT |
5145 | ext4_update_other_inodes_time(inode->i_sb, inode->i_ino, |
5146 | bh->b_data); | |
202ee5df | 5147 | |
830156c7 | 5148 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
7d8bd3c7 SL |
5149 | err = ext4_handle_dirty_metadata(handle, NULL, bh); |
5150 | if (err) | |
5151 | goto out_brelse; | |
19f5fb7a | 5152 | ext4_clear_inode_state(inode, EXT4_STATE_NEW); |
202ee5df | 5153 | if (set_large_file) { |
5d601255 | 5154 | BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access"); |
188c299e JK |
5155 | err = ext4_journal_get_write_access(handle, sb, |
5156 | EXT4_SB(sb)->s_sbh, | |
5157 | EXT4_JTR_NONE); | |
202ee5df TT |
5158 | if (err) |
5159 | goto out_brelse; | |
05c2c00f | 5160 | lock_buffer(EXT4_SB(sb)->s_sbh); |
e2b911c5 | 5161 | ext4_set_feature_large_file(sb); |
05c2c00f JK |
5162 | ext4_superblock_csum_set(sb); |
5163 | unlock_buffer(EXT4_SB(sb)->s_sbh); | |
202ee5df | 5164 | ext4_handle_sync(handle); |
a3f5cf14 JK |
5165 | err = ext4_handle_dirty_metadata(handle, NULL, |
5166 | EXT4_SB(sb)->s_sbh); | |
202ee5df | 5167 | } |
b71fc079 | 5168 | ext4_update_inode_fsync_trans(handle, inode, need_datasync); |
ac27a0ec | 5169 | out_brelse: |
af5bc92d | 5170 | brelse(bh); |
617ba13b | 5171 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
5172 | return err; |
5173 | } | |
5174 | ||
5175 | /* | |
617ba13b | 5176 | * ext4_write_inode() |
ac27a0ec DK |
5177 | * |
5178 | * We are called from a few places: | |
5179 | * | |
87f7e416 | 5180 | * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files. |
ac27a0ec | 5181 | * Here, there will be no transaction running. We wait for any running |
4907cb7b | 5182 | * transaction to commit. |
ac27a0ec | 5183 | * |
87f7e416 TT |
5184 | * - Within flush work (sys_sync(), kupdate and such). |
5185 | * We wait on commit, if told to. | |
ac27a0ec | 5186 | * |
87f7e416 TT |
5187 | * - Within iput_final() -> write_inode_now() |
5188 | * We wait on commit, if told to. | |
ac27a0ec DK |
5189 | * |
5190 | * In all cases it is actually safe for us to return without doing anything, | |
5191 | * because the inode has been copied into a raw inode buffer in | |
87f7e416 TT |
5192 | * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL |
5193 | * writeback. | |
ac27a0ec DK |
5194 | * |
5195 | * Note that we are absolutely dependent upon all inode dirtiers doing the | |
5196 | * right thing: they *must* call mark_inode_dirty() after dirtying info in | |
5197 | * which we are interested. | |
5198 | * | |
5199 | * It would be a bug for them to not do this. The code: | |
5200 | * | |
5201 | * mark_inode_dirty(inode) | |
5202 | * stuff(); | |
5203 | * inode->i_size = expr; | |
5204 | * | |
87f7e416 TT |
5205 | * is in error because write_inode() could occur while `stuff()' is running, |
5206 | * and the new i_size will be lost. Plus the inode will no longer be on the | |
5207 | * superblock's dirty inode list. | |
ac27a0ec | 5208 | */ |
a9185b41 | 5209 | int ext4_write_inode(struct inode *inode, struct writeback_control *wbc) |
ac27a0ec | 5210 | { |
91ac6f43 FM |
5211 | int err; |
5212 | ||
18f2c4fc TT |
5213 | if (WARN_ON_ONCE(current->flags & PF_MEMALLOC) || |
5214 | sb_rdonly(inode->i_sb)) | |
ac27a0ec DK |
5215 | return 0; |
5216 | ||
18f2c4fc TT |
5217 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
5218 | return -EIO; | |
5219 | ||
91ac6f43 FM |
5220 | if (EXT4_SB(inode->i_sb)->s_journal) { |
5221 | if (ext4_journal_current_handle()) { | |
5222 | jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n"); | |
5223 | dump_stack(); | |
5224 | return -EIO; | |
5225 | } | |
ac27a0ec | 5226 | |
10542c22 JK |
5227 | /* |
5228 | * No need to force transaction in WB_SYNC_NONE mode. Also | |
5229 | * ext4_sync_fs() will force the commit after everything is | |
5230 | * written. | |
5231 | */ | |
5232 | if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync) | |
91ac6f43 FM |
5233 | return 0; |
5234 | ||
aa75f4d3 | 5235 | err = ext4_fc_commit(EXT4_SB(inode->i_sb)->s_journal, |
18f2c4fc | 5236 | EXT4_I(inode)->i_sync_tid); |
91ac6f43 FM |
5237 | } else { |
5238 | struct ext4_iloc iloc; | |
ac27a0ec | 5239 | |
8016e29f | 5240 | err = __ext4_get_inode_loc_noinmem(inode, &iloc); |
91ac6f43 FM |
5241 | if (err) |
5242 | return err; | |
10542c22 JK |
5243 | /* |
5244 | * sync(2) will flush the whole buffer cache. No need to do | |
5245 | * it here separately for each inode. | |
5246 | */ | |
5247 | if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) | |
830156c7 FM |
5248 | sync_dirty_buffer(iloc.bh); |
5249 | if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) { | |
54d3adbc TT |
5250 | ext4_error_inode_block(inode, iloc.bh->b_blocknr, EIO, |
5251 | "IO error syncing inode"); | |
830156c7 FM |
5252 | err = -EIO; |
5253 | } | |
fd2dd9fb | 5254 | brelse(iloc.bh); |
91ac6f43 FM |
5255 | } |
5256 | return err; | |
ac27a0ec DK |
5257 | } |
5258 | ||
53e87268 JK |
5259 | /* |
5260 | * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate | |
5261 | * buffers that are attached to a page stradding i_size and are undergoing | |
5262 | * commit. In that case we have to wait for commit to finish and try again. | |
5263 | */ | |
5264 | static void ext4_wait_for_tail_page_commit(struct inode *inode) | |
5265 | { | |
5266 | struct page *page; | |
5267 | unsigned offset; | |
5268 | journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; | |
5269 | tid_t commit_tid = 0; | |
5270 | int ret; | |
5271 | ||
09cbfeaf | 5272 | offset = inode->i_size & (PAGE_SIZE - 1); |
53e87268 | 5273 | /* |
565333a1 | 5274 | * If the page is fully truncated, we don't need to wait for any commit |
5275 | * (and we even should not as __ext4_journalled_invalidatepage() may | |
5276 | * strip all buffers from the page but keep the page dirty which can then | |
5277 | * confuse e.g. concurrent ext4_writepage() seeing dirty page without | |
5278 | * buffers). Also we don't need to wait for any commit if all buffers in | |
5279 | * the page remain valid. This is most beneficial for the common case of | |
5280 | * blocksize == PAGESIZE. | |
53e87268 | 5281 | */ |
565333a1 | 5282 | if (!offset || offset > (PAGE_SIZE - i_blocksize(inode))) |
53e87268 JK |
5283 | return; |
5284 | while (1) { | |
5285 | page = find_lock_page(inode->i_mapping, | |
09cbfeaf | 5286 | inode->i_size >> PAGE_SHIFT); |
53e87268 JK |
5287 | if (!page) |
5288 | return; | |
ca99fdd2 | 5289 | ret = __ext4_journalled_invalidatepage(page, offset, |
09cbfeaf | 5290 | PAGE_SIZE - offset); |
53e87268 | 5291 | unlock_page(page); |
09cbfeaf | 5292 | put_page(page); |
53e87268 JK |
5293 | if (ret != -EBUSY) |
5294 | return; | |
5295 | commit_tid = 0; | |
5296 | read_lock(&journal->j_state_lock); | |
5297 | if (journal->j_committing_transaction) | |
5298 | commit_tid = journal->j_committing_transaction->t_tid; | |
5299 | read_unlock(&journal->j_state_lock); | |
5300 | if (commit_tid) | |
5301 | jbd2_log_wait_commit(journal, commit_tid); | |
5302 | } | |
5303 | } | |
5304 | ||
ac27a0ec | 5305 | /* |
617ba13b | 5306 | * ext4_setattr() |
ac27a0ec DK |
5307 | * |
5308 | * Called from notify_change. | |
5309 | * | |
5310 | * We want to trap VFS attempts to truncate the file as soon as | |
5311 | * possible. In particular, we want to make sure that when the VFS | |
5312 | * shrinks i_size, we put the inode on the orphan list and modify | |
5313 | * i_disksize immediately, so that during the subsequent flushing of | |
5314 | * dirty pages and freeing of disk blocks, we can guarantee that any | |
5315 | * commit will leave the blocks being flushed in an unused state on | |
5316 | * disk. (On recovery, the inode will get truncated and the blocks will | |
5317 | * be freed, so we have a strong guarantee that no future commit will | |
5318 | * leave these blocks visible to the user.) | |
5319 | * | |
678aaf48 JK |
5320 | * Another thing we have to assure is that if we are in ordered mode |
5321 | * and inode is still attached to the committing transaction, we must | |
5322 | * we start writeout of all the dirty pages which are being truncated. | |
5323 | * This way we are sure that all the data written in the previous | |
5324 | * transaction are already on disk (truncate waits for pages under | |
5325 | * writeback). | |
5326 | * | |
5327 | * Called with inode->i_mutex down. | |
ac27a0ec | 5328 | */ |
549c7297 CB |
5329 | int ext4_setattr(struct user_namespace *mnt_userns, struct dentry *dentry, |
5330 | struct iattr *attr) | |
ac27a0ec | 5331 | { |
2b0143b5 | 5332 | struct inode *inode = d_inode(dentry); |
ac27a0ec | 5333 | int error, rc = 0; |
3d287de3 | 5334 | int orphan = 0; |
ac27a0ec DK |
5335 | const unsigned int ia_valid = attr->ia_valid; |
5336 | ||
0db1ff22 TT |
5337 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
5338 | return -EIO; | |
5339 | ||
02b016ca TT |
5340 | if (unlikely(IS_IMMUTABLE(inode))) |
5341 | return -EPERM; | |
5342 | ||
5343 | if (unlikely(IS_APPEND(inode) && | |
5344 | (ia_valid & (ATTR_MODE | ATTR_UID | | |
5345 | ATTR_GID | ATTR_TIMES_SET)))) | |
5346 | return -EPERM; | |
5347 | ||
14f3db55 | 5348 | error = setattr_prepare(mnt_userns, dentry, attr); |
ac27a0ec DK |
5349 | if (error) |
5350 | return error; | |
5351 | ||
3ce2b8dd EB |
5352 | error = fscrypt_prepare_setattr(dentry, attr); |
5353 | if (error) | |
5354 | return error; | |
5355 | ||
c93d8f88 EB |
5356 | error = fsverity_prepare_setattr(dentry, attr); |
5357 | if (error) | |
5358 | return error; | |
5359 | ||
a7cdadee JK |
5360 | if (is_quota_modification(inode, attr)) { |
5361 | error = dquot_initialize(inode); | |
5362 | if (error) | |
5363 | return error; | |
5364 | } | |
aa75f4d3 | 5365 | ext4_fc_start_update(inode); |
08cefc7a EB |
5366 | if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) || |
5367 | (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) { | |
ac27a0ec DK |
5368 | handle_t *handle; |
5369 | ||
5370 | /* (user+group)*(old+new) structure, inode write (sb, | |
5371 | * inode block, ? - but truncate inode update has it) */ | |
9924a92a TT |
5372 | handle = ext4_journal_start(inode, EXT4_HT_QUOTA, |
5373 | (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) + | |
5374 | EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3); | |
ac27a0ec DK |
5375 | if (IS_ERR(handle)) { |
5376 | error = PTR_ERR(handle); | |
5377 | goto err_out; | |
5378 | } | |
7a9ca53a TE |
5379 | |
5380 | /* dquot_transfer() calls back ext4_get_inode_usage() which | |
5381 | * counts xattr inode references. | |
5382 | */ | |
5383 | down_read(&EXT4_I(inode)->xattr_sem); | |
b43fa828 | 5384 | error = dquot_transfer(inode, attr); |
7a9ca53a TE |
5385 | up_read(&EXT4_I(inode)->xattr_sem); |
5386 | ||
ac27a0ec | 5387 | if (error) { |
617ba13b | 5388 | ext4_journal_stop(handle); |
aa75f4d3 | 5389 | ext4_fc_stop_update(inode); |
ac27a0ec DK |
5390 | return error; |
5391 | } | |
5392 | /* Update corresponding info in inode so that everything is in | |
5393 | * one transaction */ | |
5394 | if (attr->ia_valid & ATTR_UID) | |
5395 | inode->i_uid = attr->ia_uid; | |
5396 | if (attr->ia_valid & ATTR_GID) | |
5397 | inode->i_gid = attr->ia_gid; | |
617ba13b MC |
5398 | error = ext4_mark_inode_dirty(handle, inode); |
5399 | ext4_journal_stop(handle); | |
512c15ef PB |
5400 | if (unlikely(error)) { |
5401 | ext4_fc_stop_update(inode); | |
4209ae12 | 5402 | return error; |
512c15ef | 5403 | } |
ac27a0ec DK |
5404 | } |
5405 | ||
3da40c7b | 5406 | if (attr->ia_valid & ATTR_SIZE) { |
5208386c | 5407 | handle_t *handle; |
3da40c7b | 5408 | loff_t oldsize = inode->i_size; |
b9c1c267 | 5409 | int shrink = (attr->ia_size < inode->i_size); |
562c72aa | 5410 | |
12e9b892 | 5411 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) { |
e2b46574 ES |
5412 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
5413 | ||
aa75f4d3 HS |
5414 | if (attr->ia_size > sbi->s_bitmap_maxbytes) { |
5415 | ext4_fc_stop_update(inode); | |
0c095c7f | 5416 | return -EFBIG; |
aa75f4d3 | 5417 | } |
e2b46574 | 5418 | } |
aa75f4d3 HS |
5419 | if (!S_ISREG(inode->i_mode)) { |
5420 | ext4_fc_stop_update(inode); | |
3da40c7b | 5421 | return -EINVAL; |
aa75f4d3 | 5422 | } |
dff6efc3 CH |
5423 | |
5424 | if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size) | |
5425 | inode_inc_iversion(inode); | |
5426 | ||
b9c1c267 JK |
5427 | if (shrink) { |
5428 | if (ext4_should_order_data(inode)) { | |
5429 | error = ext4_begin_ordered_truncate(inode, | |
678aaf48 | 5430 | attr->ia_size); |
b9c1c267 JK |
5431 | if (error) |
5432 | goto err_out; | |
5433 | } | |
5434 | /* | |
5435 | * Blocks are going to be removed from the inode. Wait | |
5436 | * for dio in flight. | |
5437 | */ | |
5438 | inode_dio_wait(inode); | |
5439 | } | |
5440 | ||
5441 | down_write(&EXT4_I(inode)->i_mmap_sem); | |
5442 | ||
5443 | rc = ext4_break_layouts(inode); | |
5444 | if (rc) { | |
5445 | up_write(&EXT4_I(inode)->i_mmap_sem); | |
aa75f4d3 | 5446 | goto err_out; |
3da40c7b | 5447 | } |
b9c1c267 | 5448 | |
3da40c7b | 5449 | if (attr->ia_size != inode->i_size) { |
5208386c JK |
5450 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 3); |
5451 | if (IS_ERR(handle)) { | |
5452 | error = PTR_ERR(handle); | |
b9c1c267 | 5453 | goto out_mmap_sem; |
5208386c | 5454 | } |
3da40c7b | 5455 | if (ext4_handle_valid(handle) && shrink) { |
5208386c JK |
5456 | error = ext4_orphan_add(handle, inode); |
5457 | orphan = 1; | |
5458 | } | |
911af577 EG |
5459 | /* |
5460 | * Update c/mtime on truncate up, ext4_truncate() will | |
5461 | * update c/mtime in shrink case below | |
5462 | */ | |
5463 | if (!shrink) { | |
eeca7ea1 | 5464 | inode->i_mtime = current_time(inode); |
911af577 EG |
5465 | inode->i_ctime = inode->i_mtime; |
5466 | } | |
aa75f4d3 HS |
5467 | |
5468 | if (shrink) | |
a80f7fcf | 5469 | ext4_fc_track_range(handle, inode, |
aa75f4d3 HS |
5470 | (attr->ia_size > 0 ? attr->ia_size - 1 : 0) >> |
5471 | inode->i_sb->s_blocksize_bits, | |
5472 | (oldsize > 0 ? oldsize - 1 : 0) >> | |
5473 | inode->i_sb->s_blocksize_bits); | |
5474 | else | |
5475 | ext4_fc_track_range( | |
a80f7fcf | 5476 | handle, inode, |
aa75f4d3 HS |
5477 | (oldsize > 0 ? oldsize - 1 : oldsize) >> |
5478 | inode->i_sb->s_blocksize_bits, | |
5479 | (attr->ia_size > 0 ? attr->ia_size - 1 : 0) >> | |
5480 | inode->i_sb->s_blocksize_bits); | |
5481 | ||
90e775b7 | 5482 | down_write(&EXT4_I(inode)->i_data_sem); |
5208386c JK |
5483 | EXT4_I(inode)->i_disksize = attr->ia_size; |
5484 | rc = ext4_mark_inode_dirty(handle, inode); | |
5485 | if (!error) | |
5486 | error = rc; | |
90e775b7 JK |
5487 | /* |
5488 | * We have to update i_size under i_data_sem together | |
5489 | * with i_disksize to avoid races with writeback code | |
5490 | * running ext4_wb_update_i_disksize(). | |
5491 | */ | |
5492 | if (!error) | |
5493 | i_size_write(inode, attr->ia_size); | |
5494 | up_write(&EXT4_I(inode)->i_data_sem); | |
5208386c | 5495 | ext4_journal_stop(handle); |
b9c1c267 JK |
5496 | if (error) |
5497 | goto out_mmap_sem; | |
5498 | if (!shrink) { | |
5499 | pagecache_isize_extended(inode, oldsize, | |
5500 | inode->i_size); | |
5501 | } else if (ext4_should_journal_data(inode)) { | |
5502 | ext4_wait_for_tail_page_commit(inode); | |
678aaf48 | 5503 | } |
d6320cbf | 5504 | } |
430657b6 | 5505 | |
5208386c JK |
5506 | /* |
5507 | * Truncate pagecache after we've waited for commit | |
5508 | * in data=journal mode to make pages freeable. | |
5509 | */ | |
923ae0ff | 5510 | truncate_pagecache(inode, inode->i_size); |
b9c1c267 JK |
5511 | /* |
5512 | * Call ext4_truncate() even if i_size didn't change to | |
5513 | * truncate possible preallocated blocks. | |
5514 | */ | |
5515 | if (attr->ia_size <= oldsize) { | |
2c98eb5e TT |
5516 | rc = ext4_truncate(inode); |
5517 | if (rc) | |
5518 | error = rc; | |
5519 | } | |
b9c1c267 | 5520 | out_mmap_sem: |
ea3d7209 | 5521 | up_write(&EXT4_I(inode)->i_mmap_sem); |
072bd7ea | 5522 | } |
ac27a0ec | 5523 | |
2c98eb5e | 5524 | if (!error) { |
14f3db55 | 5525 | setattr_copy(mnt_userns, inode, attr); |
1025774c CH |
5526 | mark_inode_dirty(inode); |
5527 | } | |
5528 | ||
5529 | /* | |
5530 | * If the call to ext4_truncate failed to get a transaction handle at | |
5531 | * all, we need to clean up the in-core orphan list manually. | |
5532 | */ | |
3d287de3 | 5533 | if (orphan && inode->i_nlink) |
617ba13b | 5534 | ext4_orphan_del(NULL, inode); |
ac27a0ec | 5535 | |
2c98eb5e | 5536 | if (!error && (ia_valid & ATTR_MODE)) |
14f3db55 | 5537 | rc = posix_acl_chmod(mnt_userns, inode, inode->i_mode); |
ac27a0ec DK |
5538 | |
5539 | err_out: | |
aa75f4d3 HS |
5540 | if (error) |
5541 | ext4_std_error(inode->i_sb, error); | |
ac27a0ec DK |
5542 | if (!error) |
5543 | error = rc; | |
aa75f4d3 | 5544 | ext4_fc_stop_update(inode); |
ac27a0ec DK |
5545 | return error; |
5546 | } | |
5547 | ||
549c7297 CB |
5548 | int ext4_getattr(struct user_namespace *mnt_userns, const struct path *path, |
5549 | struct kstat *stat, u32 request_mask, unsigned int query_flags) | |
3e3398a0 | 5550 | { |
99652ea5 DH |
5551 | struct inode *inode = d_inode(path->dentry); |
5552 | struct ext4_inode *raw_inode; | |
5553 | struct ext4_inode_info *ei = EXT4_I(inode); | |
5554 | unsigned int flags; | |
5555 | ||
d4c5e960 TT |
5556 | if ((request_mask & STATX_BTIME) && |
5557 | EXT4_FITS_IN_INODE(raw_inode, ei, i_crtime)) { | |
99652ea5 DH |
5558 | stat->result_mask |= STATX_BTIME; |
5559 | stat->btime.tv_sec = ei->i_crtime.tv_sec; | |
5560 | stat->btime.tv_nsec = ei->i_crtime.tv_nsec; | |
5561 | } | |
5562 | ||
5563 | flags = ei->i_flags & EXT4_FL_USER_VISIBLE; | |
5564 | if (flags & EXT4_APPEND_FL) | |
5565 | stat->attributes |= STATX_ATTR_APPEND; | |
5566 | if (flags & EXT4_COMPR_FL) | |
5567 | stat->attributes |= STATX_ATTR_COMPRESSED; | |
5568 | if (flags & EXT4_ENCRYPT_FL) | |
5569 | stat->attributes |= STATX_ATTR_ENCRYPTED; | |
5570 | if (flags & EXT4_IMMUTABLE_FL) | |
5571 | stat->attributes |= STATX_ATTR_IMMUTABLE; | |
5572 | if (flags & EXT4_NODUMP_FL) | |
5573 | stat->attributes |= STATX_ATTR_NODUMP; | |
1f607195 EB |
5574 | if (flags & EXT4_VERITY_FL) |
5575 | stat->attributes |= STATX_ATTR_VERITY; | |
3e3398a0 | 5576 | |
3209f68b DH |
5577 | stat->attributes_mask |= (STATX_ATTR_APPEND | |
5578 | STATX_ATTR_COMPRESSED | | |
5579 | STATX_ATTR_ENCRYPTED | | |
5580 | STATX_ATTR_IMMUTABLE | | |
1f607195 EB |
5581 | STATX_ATTR_NODUMP | |
5582 | STATX_ATTR_VERITY); | |
3209f68b | 5583 | |
14f3db55 | 5584 | generic_fillattr(mnt_userns, inode, stat); |
99652ea5 DH |
5585 | return 0; |
5586 | } | |
5587 | ||
549c7297 CB |
5588 | int ext4_file_getattr(struct user_namespace *mnt_userns, |
5589 | const struct path *path, struct kstat *stat, | |
99652ea5 DH |
5590 | u32 request_mask, unsigned int query_flags) |
5591 | { | |
5592 | struct inode *inode = d_inode(path->dentry); | |
5593 | u64 delalloc_blocks; | |
5594 | ||
14f3db55 | 5595 | ext4_getattr(mnt_userns, path, stat, request_mask, query_flags); |
3e3398a0 | 5596 | |
9206c561 AD |
5597 | /* |
5598 | * If there is inline data in the inode, the inode will normally not | |
5599 | * have data blocks allocated (it may have an external xattr block). | |
5600 | * Report at least one sector for such files, so tools like tar, rsync, | |
d67d64f4 | 5601 | * others don't incorrectly think the file is completely sparse. |
9206c561 AD |
5602 | */ |
5603 | if (unlikely(ext4_has_inline_data(inode))) | |
5604 | stat->blocks += (stat->size + 511) >> 9; | |
5605 | ||
3e3398a0 MC |
5606 | /* |
5607 | * We can't update i_blocks if the block allocation is delayed | |
5608 | * otherwise in the case of system crash before the real block | |
5609 | * allocation is done, we will have i_blocks inconsistent with | |
5610 | * on-disk file blocks. | |
5611 | * We always keep i_blocks updated together with real | |
5612 | * allocation. But to not confuse with user, stat | |
5613 | * will return the blocks that include the delayed allocation | |
5614 | * blocks for this file. | |
5615 | */ | |
96607551 | 5616 | delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb), |
9206c561 AD |
5617 | EXT4_I(inode)->i_reserved_data_blocks); |
5618 | stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9); | |
3e3398a0 MC |
5619 | return 0; |
5620 | } | |
ac27a0ec | 5621 | |
fffb2739 JK |
5622 | static int ext4_index_trans_blocks(struct inode *inode, int lblocks, |
5623 | int pextents) | |
a02908f1 | 5624 | { |
12e9b892 | 5625 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) |
fffb2739 JK |
5626 | return ext4_ind_trans_blocks(inode, lblocks); |
5627 | return ext4_ext_index_trans_blocks(inode, pextents); | |
a02908f1 | 5628 | } |
ac51d837 | 5629 | |
ac27a0ec | 5630 | /* |
a02908f1 MC |
5631 | * Account for index blocks, block groups bitmaps and block group |
5632 | * descriptor blocks if modify datablocks and index blocks | |
5633 | * worse case, the indexs blocks spread over different block groups | |
ac27a0ec | 5634 | * |
a02908f1 | 5635 | * If datablocks are discontiguous, they are possible to spread over |
4907cb7b | 5636 | * different block groups too. If they are contiguous, with flexbg, |
a02908f1 | 5637 | * they could still across block group boundary. |
ac27a0ec | 5638 | * |
a02908f1 MC |
5639 | * Also account for superblock, inode, quota and xattr blocks |
5640 | */ | |
dec214d0 | 5641 | static int ext4_meta_trans_blocks(struct inode *inode, int lblocks, |
fffb2739 | 5642 | int pextents) |
a02908f1 | 5643 | { |
8df9675f TT |
5644 | ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb); |
5645 | int gdpblocks; | |
a02908f1 MC |
5646 | int idxblocks; |
5647 | int ret = 0; | |
5648 | ||
5649 | /* | |
fffb2739 JK |
5650 | * How many index blocks need to touch to map @lblocks logical blocks |
5651 | * to @pextents physical extents? | |
a02908f1 | 5652 | */ |
fffb2739 | 5653 | idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents); |
a02908f1 MC |
5654 | |
5655 | ret = idxblocks; | |
5656 | ||
5657 | /* | |
5658 | * Now let's see how many group bitmaps and group descriptors need | |
5659 | * to account | |
5660 | */ | |
fffb2739 | 5661 | groups = idxblocks + pextents; |
a02908f1 | 5662 | gdpblocks = groups; |
8df9675f TT |
5663 | if (groups > ngroups) |
5664 | groups = ngroups; | |
a02908f1 MC |
5665 | if (groups > EXT4_SB(inode->i_sb)->s_gdb_count) |
5666 | gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count; | |
5667 | ||
5668 | /* bitmaps and block group descriptor blocks */ | |
5669 | ret += groups + gdpblocks; | |
5670 | ||
5671 | /* Blocks for super block, inode, quota and xattr blocks */ | |
5672 | ret += EXT4_META_TRANS_BLOCKS(inode->i_sb); | |
5673 | ||
5674 | return ret; | |
5675 | } | |
5676 | ||
5677 | /* | |
25985edc | 5678 | * Calculate the total number of credits to reserve to fit |
f3bd1f3f MC |
5679 | * the modification of a single pages into a single transaction, |
5680 | * which may include multiple chunks of block allocations. | |
ac27a0ec | 5681 | * |
525f4ed8 | 5682 | * This could be called via ext4_write_begin() |
ac27a0ec | 5683 | * |
525f4ed8 | 5684 | * We need to consider the worse case, when |
a02908f1 | 5685 | * one new block per extent. |
ac27a0ec | 5686 | */ |
a86c6181 | 5687 | int ext4_writepage_trans_blocks(struct inode *inode) |
ac27a0ec | 5688 | { |
617ba13b | 5689 | int bpp = ext4_journal_blocks_per_page(inode); |
ac27a0ec DK |
5690 | int ret; |
5691 | ||
fffb2739 | 5692 | ret = ext4_meta_trans_blocks(inode, bpp, bpp); |
a86c6181 | 5693 | |
a02908f1 | 5694 | /* Account for data blocks for journalled mode */ |
617ba13b | 5695 | if (ext4_should_journal_data(inode)) |
a02908f1 | 5696 | ret += bpp; |
ac27a0ec DK |
5697 | return ret; |
5698 | } | |
f3bd1f3f MC |
5699 | |
5700 | /* | |
5701 | * Calculate the journal credits for a chunk of data modification. | |
5702 | * | |
5703 | * This is called from DIO, fallocate or whoever calling | |
79e83036 | 5704 | * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks. |
f3bd1f3f MC |
5705 | * |
5706 | * journal buffers for data blocks are not included here, as DIO | |
5707 | * and fallocate do no need to journal data buffers. | |
5708 | */ | |
5709 | int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks) | |
5710 | { | |
5711 | return ext4_meta_trans_blocks(inode, nrblocks, 1); | |
5712 | } | |
5713 | ||
ac27a0ec | 5714 | /* |
617ba13b | 5715 | * The caller must have previously called ext4_reserve_inode_write(). |
ac27a0ec DK |
5716 | * Give this, we know that the caller already has write access to iloc->bh. |
5717 | */ | |
617ba13b | 5718 | int ext4_mark_iloc_dirty(handle_t *handle, |
de9a55b8 | 5719 | struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec DK |
5720 | { |
5721 | int err = 0; | |
5722 | ||
a6758309 VA |
5723 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) { |
5724 | put_bh(iloc->bh); | |
0db1ff22 | 5725 | return -EIO; |
a6758309 | 5726 | } |
a80f7fcf | 5727 | ext4_fc_track_inode(handle, inode); |
aa75f4d3 | 5728 | |
c64db50e | 5729 | if (IS_I_VERSION(inode)) |
25ec56b5 JNC |
5730 | inode_inc_iversion(inode); |
5731 | ||
ac27a0ec DK |
5732 | /* the do_update_inode consumes one bh->b_count */ |
5733 | get_bh(iloc->bh); | |
5734 | ||
dab291af | 5735 | /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */ |
830156c7 | 5736 | err = ext4_do_update_inode(handle, inode, iloc); |
ac27a0ec DK |
5737 | put_bh(iloc->bh); |
5738 | return err; | |
5739 | } | |
5740 | ||
5741 | /* | |
5742 | * On success, We end up with an outstanding reference count against | |
5743 | * iloc->bh. This _must_ be cleaned up later. | |
5744 | */ | |
5745 | ||
5746 | int | |
617ba13b MC |
5747 | ext4_reserve_inode_write(handle_t *handle, struct inode *inode, |
5748 | struct ext4_iloc *iloc) | |
ac27a0ec | 5749 | { |
0390131b FM |
5750 | int err; |
5751 | ||
0db1ff22 TT |
5752 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
5753 | return -EIO; | |
5754 | ||
0390131b FM |
5755 | err = ext4_get_inode_loc(inode, iloc); |
5756 | if (!err) { | |
5757 | BUFFER_TRACE(iloc->bh, "get_write_access"); | |
188c299e JK |
5758 | err = ext4_journal_get_write_access(handle, inode->i_sb, |
5759 | iloc->bh, EXT4_JTR_NONE); | |
0390131b FM |
5760 | if (err) { |
5761 | brelse(iloc->bh); | |
5762 | iloc->bh = NULL; | |
ac27a0ec DK |
5763 | } |
5764 | } | |
617ba13b | 5765 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
5766 | return err; |
5767 | } | |
5768 | ||
c03b45b8 MX |
5769 | static int __ext4_expand_extra_isize(struct inode *inode, |
5770 | unsigned int new_extra_isize, | |
5771 | struct ext4_iloc *iloc, | |
5772 | handle_t *handle, int *no_expand) | |
5773 | { | |
5774 | struct ext4_inode *raw_inode; | |
5775 | struct ext4_xattr_ibody_header *header; | |
4ea99936 TT |
5776 | unsigned int inode_size = EXT4_INODE_SIZE(inode->i_sb); |
5777 | struct ext4_inode_info *ei = EXT4_I(inode); | |
c03b45b8 MX |
5778 | int error; |
5779 | ||
4ea99936 TT |
5780 | /* this was checked at iget time, but double check for good measure */ |
5781 | if ((EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > inode_size) || | |
5782 | (ei->i_extra_isize & 3)) { | |
5783 | EXT4_ERROR_INODE(inode, "bad extra_isize %u (inode size %u)", | |
5784 | ei->i_extra_isize, | |
5785 | EXT4_INODE_SIZE(inode->i_sb)); | |
5786 | return -EFSCORRUPTED; | |
5787 | } | |
5788 | if ((new_extra_isize < ei->i_extra_isize) || | |
5789 | (new_extra_isize < 4) || | |
5790 | (new_extra_isize > inode_size - EXT4_GOOD_OLD_INODE_SIZE)) | |
5791 | return -EINVAL; /* Should never happen */ | |
5792 | ||
c03b45b8 MX |
5793 | raw_inode = ext4_raw_inode(iloc); |
5794 | ||
5795 | header = IHDR(inode, raw_inode); | |
5796 | ||
5797 | /* No extended attributes present */ | |
5798 | if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) || | |
5799 | header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) { | |
5800 | memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE + | |
5801 | EXT4_I(inode)->i_extra_isize, 0, | |
5802 | new_extra_isize - EXT4_I(inode)->i_extra_isize); | |
5803 | EXT4_I(inode)->i_extra_isize = new_extra_isize; | |
5804 | return 0; | |
5805 | } | |
5806 | ||
5807 | /* try to expand with EAs present */ | |
5808 | error = ext4_expand_extra_isize_ea(inode, new_extra_isize, | |
5809 | raw_inode, handle); | |
5810 | if (error) { | |
5811 | /* | |
5812 | * Inode size expansion failed; don't try again | |
5813 | */ | |
5814 | *no_expand = 1; | |
5815 | } | |
5816 | ||
5817 | return error; | |
5818 | } | |
5819 | ||
6dd4ee7c KS |
5820 | /* |
5821 | * Expand an inode by new_extra_isize bytes. | |
5822 | * Returns 0 on success or negative error number on failure. | |
5823 | */ | |
cf0a5e81 MX |
5824 | static int ext4_try_to_expand_extra_isize(struct inode *inode, |
5825 | unsigned int new_extra_isize, | |
5826 | struct ext4_iloc iloc, | |
5827 | handle_t *handle) | |
6dd4ee7c | 5828 | { |
3b10fdc6 MX |
5829 | int no_expand; |
5830 | int error; | |
6dd4ee7c | 5831 | |
cf0a5e81 MX |
5832 | if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) |
5833 | return -EOVERFLOW; | |
5834 | ||
5835 | /* | |
5836 | * In nojournal mode, we can immediately attempt to expand | |
5837 | * the inode. When journaled, we first need to obtain extra | |
5838 | * buffer credits since we may write into the EA block | |
5839 | * with this same handle. If journal_extend fails, then it will | |
5840 | * only result in a minor loss of functionality for that inode. | |
5841 | * If this is felt to be critical, then e2fsck should be run to | |
5842 | * force a large enough s_min_extra_isize. | |
5843 | */ | |
6cb367c2 | 5844 | if (ext4_journal_extend(handle, |
83448bdf | 5845 | EXT4_DATA_TRANS_BLOCKS(inode->i_sb), 0) != 0) |
cf0a5e81 | 5846 | return -ENOSPC; |
6dd4ee7c | 5847 | |
3b10fdc6 | 5848 | if (ext4_write_trylock_xattr(inode, &no_expand) == 0) |
cf0a5e81 | 5849 | return -EBUSY; |
3b10fdc6 | 5850 | |
c03b45b8 MX |
5851 | error = __ext4_expand_extra_isize(inode, new_extra_isize, &iloc, |
5852 | handle, &no_expand); | |
5853 | ext4_write_unlock_xattr(inode, &no_expand); | |
6dd4ee7c | 5854 | |
c03b45b8 MX |
5855 | return error; |
5856 | } | |
6dd4ee7c | 5857 | |
c03b45b8 MX |
5858 | int ext4_expand_extra_isize(struct inode *inode, |
5859 | unsigned int new_extra_isize, | |
5860 | struct ext4_iloc *iloc) | |
5861 | { | |
5862 | handle_t *handle; | |
5863 | int no_expand; | |
5864 | int error, rc; | |
5865 | ||
5866 | if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) { | |
5867 | brelse(iloc->bh); | |
5868 | return -EOVERFLOW; | |
6dd4ee7c KS |
5869 | } |
5870 | ||
c03b45b8 MX |
5871 | handle = ext4_journal_start(inode, EXT4_HT_INODE, |
5872 | EXT4_DATA_TRANS_BLOCKS(inode->i_sb)); | |
5873 | if (IS_ERR(handle)) { | |
5874 | error = PTR_ERR(handle); | |
5875 | brelse(iloc->bh); | |
5876 | return error; | |
5877 | } | |
5878 | ||
5879 | ext4_write_lock_xattr(inode, &no_expand); | |
5880 | ||
ddccb6db | 5881 | BUFFER_TRACE(iloc->bh, "get_write_access"); |
188c299e JK |
5882 | error = ext4_journal_get_write_access(handle, inode->i_sb, iloc->bh, |
5883 | EXT4_JTR_NONE); | |
3b10fdc6 | 5884 | if (error) { |
c03b45b8 | 5885 | brelse(iloc->bh); |
7f420d64 | 5886 | goto out_unlock; |
3b10fdc6 | 5887 | } |
cf0a5e81 | 5888 | |
c03b45b8 MX |
5889 | error = __ext4_expand_extra_isize(inode, new_extra_isize, iloc, |
5890 | handle, &no_expand); | |
5891 | ||
5892 | rc = ext4_mark_iloc_dirty(handle, inode, iloc); | |
5893 | if (!error) | |
5894 | error = rc; | |
5895 | ||
7f420d64 | 5896 | out_unlock: |
c03b45b8 | 5897 | ext4_write_unlock_xattr(inode, &no_expand); |
c03b45b8 | 5898 | ext4_journal_stop(handle); |
3b10fdc6 | 5899 | return error; |
6dd4ee7c KS |
5900 | } |
5901 | ||
ac27a0ec DK |
5902 | /* |
5903 | * What we do here is to mark the in-core inode as clean with respect to inode | |
5904 | * dirtiness (it may still be data-dirty). | |
5905 | * This means that the in-core inode may be reaped by prune_icache | |
5906 | * without having to perform any I/O. This is a very good thing, | |
5907 | * because *any* task may call prune_icache - even ones which | |
5908 | * have a transaction open against a different journal. | |
5909 | * | |
5910 | * Is this cheating? Not really. Sure, we haven't written the | |
5911 | * inode out, but prune_icache isn't a user-visible syncing function. | |
5912 | * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync) | |
5913 | * we start and wait on commits. | |
ac27a0ec | 5914 | */ |
4209ae12 HS |
5915 | int __ext4_mark_inode_dirty(handle_t *handle, struct inode *inode, |
5916 | const char *func, unsigned int line) | |
ac27a0ec | 5917 | { |
617ba13b | 5918 | struct ext4_iloc iloc; |
6dd4ee7c | 5919 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
cf0a5e81 | 5920 | int err; |
ac27a0ec DK |
5921 | |
5922 | might_sleep(); | |
7ff9c073 | 5923 | trace_ext4_mark_inode_dirty(inode, _RET_IP_); |
617ba13b | 5924 | err = ext4_reserve_inode_write(handle, inode, &iloc); |
5e1021f2 | 5925 | if (err) |
4209ae12 | 5926 | goto out; |
cf0a5e81 MX |
5927 | |
5928 | if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize) | |
5929 | ext4_try_to_expand_extra_isize(inode, sbi->s_want_extra_isize, | |
5930 | iloc, handle); | |
5931 | ||
4209ae12 HS |
5932 | err = ext4_mark_iloc_dirty(handle, inode, &iloc); |
5933 | out: | |
5934 | if (unlikely(err)) | |
5935 | ext4_error_inode_err(inode, func, line, 0, err, | |
5936 | "mark_inode_dirty error"); | |
5937 | return err; | |
ac27a0ec DK |
5938 | } |
5939 | ||
5940 | /* | |
617ba13b | 5941 | * ext4_dirty_inode() is called from __mark_inode_dirty() |
ac27a0ec DK |
5942 | * |
5943 | * We're really interested in the case where a file is being extended. | |
5944 | * i_size has been changed by generic_commit_write() and we thus need | |
5945 | * to include the updated inode in the current transaction. | |
5946 | * | |
5dd4056d | 5947 | * Also, dquot_alloc_block() will always dirty the inode when blocks |
ac27a0ec DK |
5948 | * are allocated to the file. |
5949 | * | |
5950 | * If the inode is marked synchronous, we don't honour that here - doing | |
5951 | * so would cause a commit on atime updates, which we don't bother doing. | |
5952 | * We handle synchronous inodes at the highest possible level. | |
5953 | */ | |
aa385729 | 5954 | void ext4_dirty_inode(struct inode *inode, int flags) |
ac27a0ec | 5955 | { |
ac27a0ec DK |
5956 | handle_t *handle; |
5957 | ||
9924a92a | 5958 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 2); |
ac27a0ec | 5959 | if (IS_ERR(handle)) |
e2728c56 | 5960 | return; |
f3dc272f | 5961 | ext4_mark_inode_dirty(handle, inode); |
617ba13b | 5962 | ext4_journal_stop(handle); |
ac27a0ec DK |
5963 | } |
5964 | ||
617ba13b | 5965 | int ext4_change_inode_journal_flag(struct inode *inode, int val) |
ac27a0ec DK |
5966 | { |
5967 | journal_t *journal; | |
5968 | handle_t *handle; | |
5969 | int err; | |
c8585c6f | 5970 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
ac27a0ec DK |
5971 | |
5972 | /* | |
5973 | * We have to be very careful here: changing a data block's | |
5974 | * journaling status dynamically is dangerous. If we write a | |
5975 | * data block to the journal, change the status and then delete | |
5976 | * that block, we risk forgetting to revoke the old log record | |
5977 | * from the journal and so a subsequent replay can corrupt data. | |
5978 | * So, first we make sure that the journal is empty and that | |
5979 | * nobody is changing anything. | |
5980 | */ | |
5981 | ||
617ba13b | 5982 | journal = EXT4_JOURNAL(inode); |
0390131b FM |
5983 | if (!journal) |
5984 | return 0; | |
d699594d | 5985 | if (is_journal_aborted(journal)) |
ac27a0ec DK |
5986 | return -EROFS; |
5987 | ||
17335dcc | 5988 | /* Wait for all existing dio workers */ |
17335dcc DM |
5989 | inode_dio_wait(inode); |
5990 | ||
4c546592 DJ |
5991 | /* |
5992 | * Before flushing the journal and switching inode's aops, we have | |
5993 | * to flush all dirty data the inode has. There can be outstanding | |
5994 | * delayed allocations, there can be unwritten extents created by | |
5995 | * fallocate or buffered writes in dioread_nolock mode covered by | |
5996 | * dirty data which can be converted only after flushing the dirty | |
5997 | * data (and journalled aops don't know how to handle these cases). | |
5998 | */ | |
5999 | if (val) { | |
6000 | down_write(&EXT4_I(inode)->i_mmap_sem); | |
6001 | err = filemap_write_and_wait(inode->i_mapping); | |
6002 | if (err < 0) { | |
6003 | up_write(&EXT4_I(inode)->i_mmap_sem); | |
4c546592 DJ |
6004 | return err; |
6005 | } | |
6006 | } | |
6007 | ||
bbd55937 | 6008 | percpu_down_write(&sbi->s_writepages_rwsem); |
dab291af | 6009 | jbd2_journal_lock_updates(journal); |
ac27a0ec DK |
6010 | |
6011 | /* | |
6012 | * OK, there are no updates running now, and all cached data is | |
6013 | * synced to disk. We are now in a completely consistent state | |
6014 | * which doesn't have anything in the journal, and we know that | |
6015 | * no filesystem updates are running, so it is safe to modify | |
6016 | * the inode's in-core data-journaling state flag now. | |
6017 | */ | |
6018 | ||
6019 | if (val) | |
12e9b892 | 6020 | ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
5872ddaa | 6021 | else { |
01d5d965 | 6022 | err = jbd2_journal_flush(journal, 0); |
4f879ca6 JK |
6023 | if (err < 0) { |
6024 | jbd2_journal_unlock_updates(journal); | |
bbd55937 | 6025 | percpu_up_write(&sbi->s_writepages_rwsem); |
4f879ca6 JK |
6026 | return err; |
6027 | } | |
12e9b892 | 6028 | ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
5872ddaa | 6029 | } |
617ba13b | 6030 | ext4_set_aops(inode); |
ac27a0ec | 6031 | |
dab291af | 6032 | jbd2_journal_unlock_updates(journal); |
bbd55937 | 6033 | percpu_up_write(&sbi->s_writepages_rwsem); |
c8585c6f | 6034 | |
4c546592 DJ |
6035 | if (val) |
6036 | up_write(&EXT4_I(inode)->i_mmap_sem); | |
ac27a0ec DK |
6037 | |
6038 | /* Finally we can mark the inode as dirty. */ | |
6039 | ||
9924a92a | 6040 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 1); |
ac27a0ec DK |
6041 | if (IS_ERR(handle)) |
6042 | return PTR_ERR(handle); | |
6043 | ||
aa75f4d3 HS |
6044 | ext4_fc_mark_ineligible(inode->i_sb, |
6045 | EXT4_FC_REASON_JOURNAL_FLAG_CHANGE); | |
617ba13b | 6046 | err = ext4_mark_inode_dirty(handle, inode); |
0390131b | 6047 | ext4_handle_sync(handle); |
617ba13b MC |
6048 | ext4_journal_stop(handle); |
6049 | ext4_std_error(inode->i_sb, err); | |
ac27a0ec DK |
6050 | |
6051 | return err; | |
6052 | } | |
2e9ee850 | 6053 | |
188c299e JK |
6054 | static int ext4_bh_unmapped(handle_t *handle, struct inode *inode, |
6055 | struct buffer_head *bh) | |
2e9ee850 AK |
6056 | { |
6057 | return !buffer_mapped(bh); | |
6058 | } | |
6059 | ||
401b25aa | 6060 | vm_fault_t ext4_page_mkwrite(struct vm_fault *vmf) |
2e9ee850 | 6061 | { |
11bac800 | 6062 | struct vm_area_struct *vma = vmf->vma; |
c2ec175c | 6063 | struct page *page = vmf->page; |
2e9ee850 AK |
6064 | loff_t size; |
6065 | unsigned long len; | |
401b25aa SJ |
6066 | int err; |
6067 | vm_fault_t ret; | |
2e9ee850 | 6068 | struct file *file = vma->vm_file; |
496ad9aa | 6069 | struct inode *inode = file_inode(file); |
2e9ee850 | 6070 | struct address_space *mapping = inode->i_mapping; |
9ea7df53 JK |
6071 | handle_t *handle; |
6072 | get_block_t *get_block; | |
6073 | int retries = 0; | |
2e9ee850 | 6074 | |
02b016ca TT |
6075 | if (unlikely(IS_IMMUTABLE(inode))) |
6076 | return VM_FAULT_SIGBUS; | |
6077 | ||
8e8ad8a5 | 6078 | sb_start_pagefault(inode->i_sb); |
041bbb6d | 6079 | file_update_time(vma->vm_file); |
ea3d7209 JK |
6080 | |
6081 | down_read(&EXT4_I(inode)->i_mmap_sem); | |
7b4cc978 | 6082 | |
401b25aa SJ |
6083 | err = ext4_convert_inline_data(inode); |
6084 | if (err) | |
7b4cc978 EB |
6085 | goto out_ret; |
6086 | ||
64a9f144 MFO |
6087 | /* |
6088 | * On data journalling we skip straight to the transaction handle: | |
6089 | * there's no delalloc; page truncated will be checked later; the | |
6090 | * early return w/ all buffers mapped (calculates size/len) can't | |
6091 | * be used; and there's no dioread_nolock, so only ext4_get_block. | |
6092 | */ | |
6093 | if (ext4_should_journal_data(inode)) | |
6094 | goto retry_alloc; | |
6095 | ||
9ea7df53 JK |
6096 | /* Delalloc case is easy... */ |
6097 | if (test_opt(inode->i_sb, DELALLOC) && | |
9ea7df53 JK |
6098 | !ext4_nonda_switch(inode->i_sb)) { |
6099 | do { | |
401b25aa | 6100 | err = block_page_mkwrite(vma, vmf, |
9ea7df53 | 6101 | ext4_da_get_block_prep); |
401b25aa | 6102 | } while (err == -ENOSPC && |
9ea7df53 JK |
6103 | ext4_should_retry_alloc(inode->i_sb, &retries)); |
6104 | goto out_ret; | |
2e9ee850 | 6105 | } |
0e499890 DW |
6106 | |
6107 | lock_page(page); | |
9ea7df53 JK |
6108 | size = i_size_read(inode); |
6109 | /* Page got truncated from under us? */ | |
6110 | if (page->mapping != mapping || page_offset(page) > size) { | |
6111 | unlock_page(page); | |
6112 | ret = VM_FAULT_NOPAGE; | |
6113 | goto out; | |
0e499890 | 6114 | } |
2e9ee850 | 6115 | |
09cbfeaf KS |
6116 | if (page->index == size >> PAGE_SHIFT) |
6117 | len = size & ~PAGE_MASK; | |
2e9ee850 | 6118 | else |
09cbfeaf | 6119 | len = PAGE_SIZE; |
a827eaff | 6120 | /* |
9ea7df53 JK |
6121 | * Return if we have all the buffers mapped. This avoids the need to do |
6122 | * journal_start/journal_stop which can block and take a long time | |
64a9f144 MFO |
6123 | * |
6124 | * This cannot be done for data journalling, as we have to add the | |
6125 | * inode to the transaction's list to writeprotect pages on commit. | |
a827eaff | 6126 | */ |
2e9ee850 | 6127 | if (page_has_buffers(page)) { |
188c299e | 6128 | if (!ext4_walk_page_buffers(NULL, inode, page_buffers(page), |
f19d5870 TM |
6129 | 0, len, NULL, |
6130 | ext4_bh_unmapped)) { | |
9ea7df53 | 6131 | /* Wait so that we don't change page under IO */ |
1d1d1a76 | 6132 | wait_for_stable_page(page); |
9ea7df53 JK |
6133 | ret = VM_FAULT_LOCKED; |
6134 | goto out; | |
a827eaff | 6135 | } |
2e9ee850 | 6136 | } |
a827eaff | 6137 | unlock_page(page); |
9ea7df53 JK |
6138 | /* OK, we need to fill the hole... */ |
6139 | if (ext4_should_dioread_nolock(inode)) | |
705965bd | 6140 | get_block = ext4_get_block_unwritten; |
9ea7df53 JK |
6141 | else |
6142 | get_block = ext4_get_block; | |
6143 | retry_alloc: | |
9924a92a TT |
6144 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
6145 | ext4_writepage_trans_blocks(inode)); | |
9ea7df53 | 6146 | if (IS_ERR(handle)) { |
c2ec175c | 6147 | ret = VM_FAULT_SIGBUS; |
9ea7df53 JK |
6148 | goto out; |
6149 | } | |
64a9f144 MFO |
6150 | /* |
6151 | * Data journalling can't use block_page_mkwrite() because it | |
6152 | * will set_buffer_dirty() before do_journal_get_write_access() | |
6153 | * thus might hit warning messages for dirty metadata buffers. | |
6154 | */ | |
6155 | if (!ext4_should_journal_data(inode)) { | |
6156 | err = block_page_mkwrite(vma, vmf, get_block); | |
6157 | } else { | |
6158 | lock_page(page); | |
6159 | size = i_size_read(inode); | |
6160 | /* Page got truncated from under us? */ | |
6161 | if (page->mapping != mapping || page_offset(page) > size) { | |
64a9f144 | 6162 | ret = VM_FAULT_NOPAGE; |
afb585a9 | 6163 | goto out_error; |
9ea7df53 | 6164 | } |
64a9f144 MFO |
6165 | |
6166 | if (page->index == size >> PAGE_SHIFT) | |
6167 | len = size & ~PAGE_MASK; | |
6168 | else | |
6169 | len = PAGE_SIZE; | |
6170 | ||
6171 | err = __block_write_begin(page, 0, len, ext4_get_block); | |
6172 | if (!err) { | |
afb585a9 | 6173 | ret = VM_FAULT_SIGBUS; |
188c299e JK |
6174 | if (ext4_walk_page_buffers(handle, inode, |
6175 | page_buffers(page), 0, len, NULL, | |
6176 | do_journal_get_write_access)) | |
afb585a9 | 6177 | goto out_error; |
188c299e JK |
6178 | if (ext4_walk_page_buffers(handle, inode, |
6179 | page_buffers(page), 0, len, NULL, | |
6180 | write_end_fn)) | |
afb585a9 | 6181 | goto out_error; |
b5b18160 JK |
6182 | if (ext4_jbd2_inode_add_write(handle, inode, |
6183 | page_offset(page), len)) | |
afb585a9 | 6184 | goto out_error; |
64a9f144 MFO |
6185 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
6186 | } else { | |
6187 | unlock_page(page); | |
6188 | } | |
9ea7df53 JK |
6189 | } |
6190 | ext4_journal_stop(handle); | |
401b25aa | 6191 | if (err == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) |
9ea7df53 JK |
6192 | goto retry_alloc; |
6193 | out_ret: | |
401b25aa | 6194 | ret = block_page_mkwrite_return(err); |
9ea7df53 | 6195 | out: |
ea3d7209 | 6196 | up_read(&EXT4_I(inode)->i_mmap_sem); |
8e8ad8a5 | 6197 | sb_end_pagefault(inode->i_sb); |
2e9ee850 | 6198 | return ret; |
afb585a9 MFO |
6199 | out_error: |
6200 | unlock_page(page); | |
6201 | ext4_journal_stop(handle); | |
6202 | goto out; | |
2e9ee850 | 6203 | } |
ea3d7209 | 6204 | |
401b25aa | 6205 | vm_fault_t ext4_filemap_fault(struct vm_fault *vmf) |
ea3d7209 | 6206 | { |
11bac800 | 6207 | struct inode *inode = file_inode(vmf->vma->vm_file); |
401b25aa | 6208 | vm_fault_t ret; |
ea3d7209 JK |
6209 | |
6210 | down_read(&EXT4_I(inode)->i_mmap_sem); | |
401b25aa | 6211 | ret = filemap_fault(vmf); |
ea3d7209 JK |
6212 | up_read(&EXT4_I(inode)->i_mmap_sem); |
6213 | ||
401b25aa | 6214 | return ret; |
ea3d7209 | 6215 | } |