]>
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 TT |
141 | static int __ext4_journalled_writepage(struct page *page, unsigned int len); |
142 | static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh); | |
dec214d0 TE |
143 | static int ext4_meta_trans_blocks(struct inode *inode, int lblocks, |
144 | int pextents); | |
64769240 | 145 | |
ac27a0ec DK |
146 | /* |
147 | * Test whether an inode is a fast symlink. | |
407cd7fb | 148 | * A fast symlink has its symlink data stored in ext4_inode_info->i_data. |
ac27a0ec | 149 | */ |
f348c252 | 150 | int ext4_inode_is_fast_symlink(struct inode *inode) |
ac27a0ec | 151 | { |
fc82228a AK |
152 | if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) { |
153 | int ea_blocks = EXT4_I(inode)->i_file_acl ? | |
154 | EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0; | |
155 | ||
156 | if (ext4_has_inline_data(inode)) | |
157 | return 0; | |
158 | ||
159 | return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0); | |
160 | } | |
407cd7fb TE |
161 | return S_ISLNK(inode->i_mode) && inode->i_size && |
162 | (inode->i_size < EXT4_N_BLOCKS * 4); | |
ac27a0ec DK |
163 | } |
164 | ||
ac27a0ec DK |
165 | /* |
166 | * Called at the last iput() if i_nlink is zero. | |
167 | */ | |
0930fcc1 | 168 | void ext4_evict_inode(struct inode *inode) |
ac27a0ec DK |
169 | { |
170 | handle_t *handle; | |
bc965ab3 | 171 | int err; |
65db869c JK |
172 | /* |
173 | * Credits for final inode cleanup and freeing: | |
174 | * sb + inode (ext4_orphan_del()), block bitmap, group descriptor | |
175 | * (xattr block freeing), bitmap, group descriptor (inode freeing) | |
176 | */ | |
177 | int extra_credits = 6; | |
0421a189 | 178 | struct ext4_xattr_inode_array *ea_inode_array = NULL; |
46e294ef | 179 | bool freeze_protected = false; |
ac27a0ec | 180 | |
7ff9c073 | 181 | trace_ext4_evict_inode(inode); |
2581fdc8 | 182 | |
0930fcc1 | 183 | if (inode->i_nlink) { |
2d859db3 JK |
184 | /* |
185 | * When journalling data dirty buffers are tracked only in the | |
186 | * journal. So although mm thinks everything is clean and | |
187 | * ready for reaping the inode might still have some pages to | |
188 | * write in the running transaction or waiting to be | |
189 | * checkpointed. Thus calling jbd2_journal_invalidatepage() | |
190 | * (via truncate_inode_pages()) to discard these buffers can | |
191 | * cause data loss. Also even if we did not discard these | |
192 | * buffers, we would have no way to find them after the inode | |
193 | * is reaped and thus user could see stale data if he tries to | |
194 | * read them before the transaction is checkpointed. So be | |
195 | * careful and force everything to disk here... We use | |
196 | * ei->i_datasync_tid to store the newest transaction | |
197 | * containing inode's data. | |
198 | * | |
199 | * Note that directories do not have this problem because they | |
200 | * don't use page cache. | |
201 | */ | |
6a7fd522 VN |
202 | if (inode->i_ino != EXT4_JOURNAL_INO && |
203 | ext4_should_journal_data(inode) && | |
3abb1a0f JK |
204 | (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) && |
205 | inode->i_data.nrpages) { | |
2d859db3 JK |
206 | journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; |
207 | tid_t commit_tid = EXT4_I(inode)->i_datasync_tid; | |
208 | ||
d76a3a77 | 209 | jbd2_complete_transaction(journal, commit_tid); |
2d859db3 JK |
210 | filemap_write_and_wait(&inode->i_data); |
211 | } | |
91b0abe3 | 212 | truncate_inode_pages_final(&inode->i_data); |
5dc23bdd | 213 | |
0930fcc1 AV |
214 | goto no_delete; |
215 | } | |
216 | ||
e2bfb088 TT |
217 | if (is_bad_inode(inode)) |
218 | goto no_delete; | |
219 | dquot_initialize(inode); | |
907f4554 | 220 | |
678aaf48 JK |
221 | if (ext4_should_order_data(inode)) |
222 | ext4_begin_ordered_truncate(inode, 0); | |
91b0abe3 | 223 | truncate_inode_pages_final(&inode->i_data); |
ac27a0ec | 224 | |
ceff86fd JK |
225 | /* |
226 | * For inodes with journalled data, transaction commit could have | |
227 | * dirtied the inode. Flush worker is ignoring it because of I_FREEING | |
228 | * flag but we still need to remove the inode from the writeback lists. | |
229 | */ | |
230 | if (!list_empty_careful(&inode->i_io_list)) { | |
231 | WARN_ON_ONCE(!ext4_should_journal_data(inode)); | |
232 | inode_io_list_del(inode); | |
233 | } | |
234 | ||
8e8ad8a5 JK |
235 | /* |
236 | * Protect us against freezing - iput() caller didn't have to have any | |
46e294ef JK |
237 | * protection against it. When we are in a running transaction though, |
238 | * we are already protected against freezing and we cannot grab further | |
239 | * protection due to lock ordering constraints. | |
8e8ad8a5 | 240 | */ |
46e294ef JK |
241 | if (!ext4_journal_current_handle()) { |
242 | sb_start_intwrite(inode->i_sb); | |
243 | freeze_protected = true; | |
244 | } | |
e50e5129 | 245 | |
30a7eb97 TE |
246 | if (!IS_NOQUOTA(inode)) |
247 | extra_credits += EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb); | |
248 | ||
65db869c JK |
249 | /* |
250 | * Block bitmap, group descriptor, and inode are accounted in both | |
251 | * ext4_blocks_for_truncate() and extra_credits. So subtract 3. | |
252 | */ | |
30a7eb97 | 253 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, |
65db869c | 254 | ext4_blocks_for_truncate(inode) + extra_credits - 3); |
ac27a0ec | 255 | if (IS_ERR(handle)) { |
bc965ab3 | 256 | ext4_std_error(inode->i_sb, PTR_ERR(handle)); |
ac27a0ec DK |
257 | /* |
258 | * If we're going to skip the normal cleanup, we still need to | |
259 | * make sure that the in-core orphan linked list is properly | |
260 | * cleaned up. | |
261 | */ | |
617ba13b | 262 | ext4_orphan_del(NULL, inode); |
46e294ef JK |
263 | if (freeze_protected) |
264 | sb_end_intwrite(inode->i_sb); | |
ac27a0ec DK |
265 | goto no_delete; |
266 | } | |
30a7eb97 | 267 | |
ac27a0ec | 268 | if (IS_SYNC(inode)) |
0390131b | 269 | ext4_handle_sync(handle); |
407cd7fb TE |
270 | |
271 | /* | |
272 | * Set inode->i_size to 0 before calling ext4_truncate(). We need | |
273 | * special handling of symlinks here because i_size is used to | |
274 | * determine whether ext4_inode_info->i_data contains symlink data or | |
275 | * block mappings. Setting i_size to 0 will remove its fast symlink | |
276 | * status. Erase i_data so that it becomes a valid empty block map. | |
277 | */ | |
278 | if (ext4_inode_is_fast_symlink(inode)) | |
279 | memset(EXT4_I(inode)->i_data, 0, sizeof(EXT4_I(inode)->i_data)); | |
ac27a0ec | 280 | inode->i_size = 0; |
bc965ab3 TT |
281 | err = ext4_mark_inode_dirty(handle, inode); |
282 | if (err) { | |
12062ddd | 283 | ext4_warning(inode->i_sb, |
bc965ab3 TT |
284 | "couldn't mark inode dirty (err %d)", err); |
285 | goto stop_handle; | |
286 | } | |
2c98eb5e TT |
287 | if (inode->i_blocks) { |
288 | err = ext4_truncate(inode); | |
289 | if (err) { | |
54d3adbc TT |
290 | ext4_error_err(inode->i_sb, -err, |
291 | "couldn't truncate inode %lu (err %d)", | |
292 | inode->i_ino, err); | |
2c98eb5e TT |
293 | goto stop_handle; |
294 | } | |
295 | } | |
bc965ab3 | 296 | |
30a7eb97 TE |
297 | /* Remove xattr references. */ |
298 | err = ext4_xattr_delete_inode(handle, inode, &ea_inode_array, | |
299 | extra_credits); | |
300 | if (err) { | |
301 | ext4_warning(inode->i_sb, "xattr delete (err %d)", err); | |
302 | stop_handle: | |
303 | ext4_journal_stop(handle); | |
304 | ext4_orphan_del(NULL, inode); | |
46e294ef JK |
305 | if (freeze_protected) |
306 | sb_end_intwrite(inode->i_sb); | |
30a7eb97 TE |
307 | ext4_xattr_inode_array_free(ea_inode_array); |
308 | goto no_delete; | |
bc965ab3 TT |
309 | } |
310 | ||
ac27a0ec | 311 | /* |
617ba13b | 312 | * Kill off the orphan record which ext4_truncate created. |
ac27a0ec | 313 | * AKPM: I think this can be inside the above `if'. |
617ba13b | 314 | * Note that ext4_orphan_del() has to be able to cope with the |
ac27a0ec | 315 | * deletion of a non-existent orphan - this is because we don't |
617ba13b | 316 | * know if ext4_truncate() actually created an orphan record. |
ac27a0ec DK |
317 | * (Well, we could do this if we need to, but heck - it works) |
318 | */ | |
617ba13b | 319 | ext4_orphan_del(handle, inode); |
5ffff834 | 320 | EXT4_I(inode)->i_dtime = (__u32)ktime_get_real_seconds(); |
ac27a0ec DK |
321 | |
322 | /* | |
323 | * One subtle ordering requirement: if anything has gone wrong | |
324 | * (transaction abort, IO errors, whatever), then we can still | |
325 | * do these next steps (the fs will already have been marked as | |
326 | * having errors), but we can't free the inode if the mark_dirty | |
327 | * fails. | |
328 | */ | |
617ba13b | 329 | if (ext4_mark_inode_dirty(handle, inode)) |
ac27a0ec | 330 | /* If that failed, just do the required in-core inode clear. */ |
0930fcc1 | 331 | ext4_clear_inode(inode); |
ac27a0ec | 332 | else |
617ba13b MC |
333 | ext4_free_inode(handle, inode); |
334 | ext4_journal_stop(handle); | |
46e294ef JK |
335 | if (freeze_protected) |
336 | sb_end_intwrite(inode->i_sb); | |
0421a189 | 337 | ext4_xattr_inode_array_free(ea_inode_array); |
ac27a0ec DK |
338 | return; |
339 | no_delete: | |
b21ebf14 HS |
340 | if (!list_empty(&EXT4_I(inode)->i_fc_list)) |
341 | ext4_fc_mark_ineligible(inode->i_sb, EXT4_FC_REASON_NOMEM); | |
0930fcc1 | 342 | ext4_clear_inode(inode); /* We must guarantee clearing of inode... */ |
ac27a0ec DK |
343 | } |
344 | ||
a9e7f447 DM |
345 | #ifdef CONFIG_QUOTA |
346 | qsize_t *ext4_get_reserved_space(struct inode *inode) | |
60e58e0f | 347 | { |
a9e7f447 | 348 | return &EXT4_I(inode)->i_reserved_quota; |
60e58e0f | 349 | } |
a9e7f447 | 350 | #endif |
9d0be502 | 351 | |
0637c6f4 TT |
352 | /* |
353 | * Called with i_data_sem down, which is important since we can call | |
354 | * ext4_discard_preallocations() from here. | |
355 | */ | |
5f634d06 AK |
356 | void ext4_da_update_reserve_space(struct inode *inode, |
357 | int used, int quota_claim) | |
12219aea AK |
358 | { |
359 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0637c6f4 | 360 | struct ext4_inode_info *ei = EXT4_I(inode); |
0637c6f4 TT |
361 | |
362 | spin_lock(&ei->i_block_reservation_lock); | |
d8990240 | 363 | trace_ext4_da_update_reserve_space(inode, used, quota_claim); |
0637c6f4 | 364 | if (unlikely(used > ei->i_reserved_data_blocks)) { |
8de5c325 | 365 | ext4_warning(inode->i_sb, "%s: ino %lu, used %d " |
1084f252 | 366 | "with only %d reserved data blocks", |
0637c6f4 TT |
367 | __func__, inode->i_ino, used, |
368 | ei->i_reserved_data_blocks); | |
369 | WARN_ON(1); | |
370 | used = ei->i_reserved_data_blocks; | |
371 | } | |
12219aea | 372 | |
0637c6f4 TT |
373 | /* Update per-inode reservations */ |
374 | ei->i_reserved_data_blocks -= used; | |
71d4f7d0 | 375 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, used); |
6bc6e63f | 376 | |
12219aea | 377 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
60e58e0f | 378 | |
72b8ab9d ES |
379 | /* Update quota subsystem for data blocks */ |
380 | if (quota_claim) | |
7b415bf6 | 381 | dquot_claim_block(inode, EXT4_C2B(sbi, used)); |
72b8ab9d | 382 | else { |
5f634d06 AK |
383 | /* |
384 | * We did fallocate with an offset that is already delayed | |
385 | * allocated. So on delayed allocated writeback we should | |
72b8ab9d | 386 | * not re-claim the quota for fallocated blocks. |
5f634d06 | 387 | */ |
7b415bf6 | 388 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, used)); |
5f634d06 | 389 | } |
d6014301 AK |
390 | |
391 | /* | |
392 | * If we have done all the pending block allocations and if | |
393 | * there aren't any writers on the inode, we can discard the | |
394 | * inode's preallocations. | |
395 | */ | |
0637c6f4 | 396 | if ((ei->i_reserved_data_blocks == 0) && |
82dd124c | 397 | !inode_is_open_for_write(inode)) |
27bc446e | 398 | ext4_discard_preallocations(inode, 0); |
12219aea AK |
399 | } |
400 | ||
e29136f8 | 401 | static int __check_block_validity(struct inode *inode, const char *func, |
c398eda0 TT |
402 | unsigned int line, |
403 | struct ext4_map_blocks *map) | |
6fd058f7 | 404 | { |
345c0dbf TT |
405 | if (ext4_has_feature_journal(inode->i_sb) && |
406 | (inode->i_ino == | |
407 | le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_journal_inum))) | |
408 | return 0; | |
ce9f24cc | 409 | if (!ext4_inode_block_valid(inode, map->m_pblk, map->m_len)) { |
c398eda0 | 410 | ext4_error_inode(inode, func, line, map->m_pblk, |
bdbd6ce0 | 411 | "lblock %lu mapped to illegal pblock %llu " |
c398eda0 | 412 | "(length %d)", (unsigned long) map->m_lblk, |
bdbd6ce0 | 413 | map->m_pblk, map->m_len); |
6a797d27 | 414 | return -EFSCORRUPTED; |
6fd058f7 TT |
415 | } |
416 | return 0; | |
417 | } | |
418 | ||
53085fac JK |
419 | int ext4_issue_zeroout(struct inode *inode, ext4_lblk_t lblk, ext4_fsblk_t pblk, |
420 | ext4_lblk_t len) | |
421 | { | |
422 | int ret; | |
423 | ||
33b4cc25 | 424 | if (IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode)) |
a7550b30 | 425 | return fscrypt_zeroout_range(inode, lblk, pblk, len); |
53085fac JK |
426 | |
427 | ret = sb_issue_zeroout(inode->i_sb, pblk, len, GFP_NOFS); | |
428 | if (ret > 0) | |
429 | ret = 0; | |
430 | ||
431 | return ret; | |
432 | } | |
433 | ||
e29136f8 | 434 | #define check_block_validity(inode, map) \ |
c398eda0 | 435 | __check_block_validity((inode), __func__, __LINE__, (map)) |
e29136f8 | 436 | |
921f266b DM |
437 | #ifdef ES_AGGRESSIVE_TEST |
438 | static void ext4_map_blocks_es_recheck(handle_t *handle, | |
439 | struct inode *inode, | |
440 | struct ext4_map_blocks *es_map, | |
441 | struct ext4_map_blocks *map, | |
442 | int flags) | |
443 | { | |
444 | int retval; | |
445 | ||
446 | map->m_flags = 0; | |
447 | /* | |
448 | * There is a race window that the result is not the same. | |
449 | * e.g. xfstests #223 when dioread_nolock enables. The reason | |
450 | * is that we lookup a block mapping in extent status tree with | |
451 | * out taking i_data_sem. So at the time the unwritten extent | |
452 | * could be converted. | |
453 | */ | |
2dcba478 | 454 | down_read(&EXT4_I(inode)->i_data_sem); |
921f266b | 455 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
9e52484c | 456 | retval = ext4_ext_map_blocks(handle, inode, map, 0); |
921f266b | 457 | } else { |
9e52484c | 458 | retval = ext4_ind_map_blocks(handle, inode, map, 0); |
921f266b | 459 | } |
2dcba478 | 460 | up_read((&EXT4_I(inode)->i_data_sem)); |
921f266b DM |
461 | |
462 | /* | |
463 | * We don't check m_len because extent will be collpased in status | |
464 | * tree. So the m_len might not equal. | |
465 | */ | |
466 | if (es_map->m_lblk != map->m_lblk || | |
467 | es_map->m_flags != map->m_flags || | |
468 | es_map->m_pblk != map->m_pblk) { | |
bdafe42a | 469 | printk("ES cache assertion failed for inode: %lu " |
921f266b DM |
470 | "es_cached ex [%d/%d/%llu/%x] != " |
471 | "found ex [%d/%d/%llu/%x] retval %d flags %x\n", | |
472 | inode->i_ino, es_map->m_lblk, es_map->m_len, | |
473 | es_map->m_pblk, es_map->m_flags, map->m_lblk, | |
474 | map->m_len, map->m_pblk, map->m_flags, | |
475 | retval, flags); | |
476 | } | |
477 | } | |
478 | #endif /* ES_AGGRESSIVE_TEST */ | |
479 | ||
f5ab0d1f | 480 | /* |
e35fd660 | 481 | * The ext4_map_blocks() function tries to look up the requested blocks, |
2b2d6d01 | 482 | * and returns if the blocks are already mapped. |
f5ab0d1f | 483 | * |
f5ab0d1f MC |
484 | * Otherwise it takes the write lock of the i_data_sem and allocate blocks |
485 | * and store the allocated blocks in the result buffer head and mark it | |
486 | * mapped. | |
487 | * | |
e35fd660 TT |
488 | * If file type is extents based, it will call ext4_ext_map_blocks(), |
489 | * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping | |
f5ab0d1f MC |
490 | * based files |
491 | * | |
facab4d9 JK |
492 | * On success, it returns the number of blocks being mapped or allocated. if |
493 | * create==0 and the blocks are pre-allocated and unwritten, the resulting @map | |
494 | * is marked as unwritten. If the create == 1, it will mark @map as mapped. | |
f5ab0d1f MC |
495 | * |
496 | * It returns 0 if plain look up failed (blocks have not been allocated), in | |
facab4d9 JK |
497 | * that case, @map is returned as unmapped but we still do fill map->m_len to |
498 | * indicate the length of a hole starting at map->m_lblk. | |
f5ab0d1f MC |
499 | * |
500 | * It returns the error in case of allocation failure. | |
501 | */ | |
e35fd660 TT |
502 | int ext4_map_blocks(handle_t *handle, struct inode *inode, |
503 | struct ext4_map_blocks *map, int flags) | |
0e855ac8 | 504 | { |
d100eef2 | 505 | struct extent_status es; |
0e855ac8 | 506 | int retval; |
b8a86845 | 507 | int ret = 0; |
921f266b DM |
508 | #ifdef ES_AGGRESSIVE_TEST |
509 | struct ext4_map_blocks orig_map; | |
510 | ||
511 | memcpy(&orig_map, map, sizeof(*map)); | |
512 | #endif | |
f5ab0d1f | 513 | |
e35fd660 | 514 | map->m_flags = 0; |
70aa1554 RH |
515 | ext_debug(inode, "flag 0x%x, max_blocks %u, logical block %lu\n", |
516 | flags, map->m_len, (unsigned long) map->m_lblk); | |
d100eef2 | 517 | |
e861b5e9 TT |
518 | /* |
519 | * ext4_map_blocks returns an int, and m_len is an unsigned int | |
520 | */ | |
521 | if (unlikely(map->m_len > INT_MAX)) | |
522 | map->m_len = INT_MAX; | |
523 | ||
4adb6ab3 KM |
524 | /* We can handle the block number less than EXT_MAX_BLOCKS */ |
525 | if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS)) | |
6a797d27 | 526 | return -EFSCORRUPTED; |
4adb6ab3 | 527 | |
d100eef2 | 528 | /* Lookup extent status tree firstly */ |
8016e29f HS |
529 | if (!(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY) && |
530 | ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) { | |
d100eef2 ZL |
531 | if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) { |
532 | map->m_pblk = ext4_es_pblock(&es) + | |
533 | map->m_lblk - es.es_lblk; | |
534 | map->m_flags |= ext4_es_is_written(&es) ? | |
535 | EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN; | |
536 | retval = es.es_len - (map->m_lblk - es.es_lblk); | |
537 | if (retval > map->m_len) | |
538 | retval = map->m_len; | |
539 | map->m_len = retval; | |
540 | } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) { | |
facab4d9 JK |
541 | map->m_pblk = 0; |
542 | retval = es.es_len - (map->m_lblk - es.es_lblk); | |
543 | if (retval > map->m_len) | |
544 | retval = map->m_len; | |
545 | map->m_len = retval; | |
d100eef2 ZL |
546 | retval = 0; |
547 | } else { | |
1e83bc81 | 548 | BUG(); |
d100eef2 | 549 | } |
921f266b DM |
550 | #ifdef ES_AGGRESSIVE_TEST |
551 | ext4_map_blocks_es_recheck(handle, inode, map, | |
552 | &orig_map, flags); | |
553 | #endif | |
d100eef2 ZL |
554 | goto found; |
555 | } | |
556 | ||
4df3d265 | 557 | /* |
b920c755 TT |
558 | * Try to see if we can get the block without requesting a new |
559 | * file system block. | |
4df3d265 | 560 | */ |
2dcba478 | 561 | down_read(&EXT4_I(inode)->i_data_sem); |
12e9b892 | 562 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
9e52484c | 563 | retval = ext4_ext_map_blocks(handle, inode, map, 0); |
0e855ac8 | 564 | } else { |
9e52484c | 565 | retval = ext4_ind_map_blocks(handle, inode, map, 0); |
0e855ac8 | 566 | } |
f7fec032 | 567 | if (retval > 0) { |
3be78c73 | 568 | unsigned int status; |
f7fec032 | 569 | |
44fb851d ZL |
570 | if (unlikely(retval != map->m_len)) { |
571 | ext4_warning(inode->i_sb, | |
572 | "ES len assertion failed for inode " | |
573 | "%lu: retval %d != map->m_len %d", | |
574 | inode->i_ino, retval, map->m_len); | |
575 | WARN_ON(1); | |
921f266b | 576 | } |
921f266b | 577 | |
f7fec032 ZL |
578 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
579 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; | |
580 | if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) && | |
d2dc317d | 581 | !(status & EXTENT_STATUS_WRITTEN) && |
ad431025 EW |
582 | ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk, |
583 | map->m_lblk + map->m_len - 1)) | |
f7fec032 ZL |
584 | status |= EXTENT_STATUS_DELAYED; |
585 | ret = ext4_es_insert_extent(inode, map->m_lblk, | |
586 | map->m_len, map->m_pblk, status); | |
587 | if (ret < 0) | |
588 | retval = ret; | |
589 | } | |
2dcba478 | 590 | up_read((&EXT4_I(inode)->i_data_sem)); |
f5ab0d1f | 591 | |
d100eef2 | 592 | found: |
e35fd660 | 593 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
b8a86845 | 594 | ret = check_block_validity(inode, map); |
6fd058f7 TT |
595 | if (ret != 0) |
596 | return ret; | |
597 | } | |
598 | ||
f5ab0d1f | 599 | /* If it is only a block(s) look up */ |
c2177057 | 600 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) |
f5ab0d1f MC |
601 | return retval; |
602 | ||
603 | /* | |
604 | * Returns if the blocks have already allocated | |
605 | * | |
606 | * Note that if blocks have been preallocated | |
df3ab170 | 607 | * ext4_ext_get_block() returns the create = 0 |
f5ab0d1f MC |
608 | * with buffer head unmapped. |
609 | */ | |
e35fd660 | 610 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) |
b8a86845 LC |
611 | /* |
612 | * If we need to convert extent to unwritten | |
613 | * we continue and do the actual work in | |
614 | * ext4_ext_map_blocks() | |
615 | */ | |
616 | if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) | |
617 | return retval; | |
4df3d265 | 618 | |
2a8964d6 | 619 | /* |
a25a4e1a ZL |
620 | * Here we clear m_flags because after allocating an new extent, |
621 | * it will be set again. | |
2a8964d6 | 622 | */ |
a25a4e1a | 623 | map->m_flags &= ~EXT4_MAP_FLAGS; |
2a8964d6 | 624 | |
4df3d265 | 625 | /* |
556615dc | 626 | * New blocks allocate and/or writing to unwritten extent |
f5ab0d1f | 627 | * will possibly result in updating i_data, so we take |
d91bd2c1 | 628 | * the write lock of i_data_sem, and call get_block() |
f5ab0d1f | 629 | * with create == 1 flag. |
4df3d265 | 630 | */ |
c8b459f4 | 631 | down_write(&EXT4_I(inode)->i_data_sem); |
d2a17637 | 632 | |
4df3d265 AK |
633 | /* |
634 | * We need to check for EXT4 here because migrate | |
635 | * could have changed the inode type in between | |
636 | */ | |
12e9b892 | 637 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
e35fd660 | 638 | retval = ext4_ext_map_blocks(handle, inode, map, flags); |
0e855ac8 | 639 | } else { |
e35fd660 | 640 | retval = ext4_ind_map_blocks(handle, inode, map, flags); |
267e4db9 | 641 | |
e35fd660 | 642 | if (retval > 0 && map->m_flags & EXT4_MAP_NEW) { |
267e4db9 AK |
643 | /* |
644 | * We allocated new blocks which will result in | |
645 | * i_data's format changing. Force the migrate | |
646 | * to fail by clearing migrate flags | |
647 | */ | |
19f5fb7a | 648 | ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE); |
267e4db9 | 649 | } |
d2a17637 | 650 | |
5f634d06 AK |
651 | /* |
652 | * Update reserved blocks/metadata blocks after successful | |
653 | * block allocation which had been deferred till now. We don't | |
654 | * support fallocate for non extent files. So we can update | |
655 | * reserve space here. | |
656 | */ | |
657 | if ((retval > 0) && | |
1296cc85 | 658 | (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)) |
5f634d06 AK |
659 | ext4_da_update_reserve_space(inode, retval, 1); |
660 | } | |
2ac3b6e0 | 661 | |
f7fec032 | 662 | if (retval > 0) { |
3be78c73 | 663 | unsigned int status; |
f7fec032 | 664 | |
44fb851d ZL |
665 | if (unlikely(retval != map->m_len)) { |
666 | ext4_warning(inode->i_sb, | |
667 | "ES len assertion failed for inode " | |
668 | "%lu: retval %d != map->m_len %d", | |
669 | inode->i_ino, retval, map->m_len); | |
670 | WARN_ON(1); | |
921f266b | 671 | } |
921f266b | 672 | |
c86d8db3 JK |
673 | /* |
674 | * We have to zeroout blocks before inserting them into extent | |
675 | * status tree. Otherwise someone could look them up there and | |
9b623df6 JK |
676 | * use them before they are really zeroed. We also have to |
677 | * unmap metadata before zeroing as otherwise writeback can | |
678 | * overwrite zeros with stale data from block device. | |
c86d8db3 JK |
679 | */ |
680 | if (flags & EXT4_GET_BLOCKS_ZERO && | |
681 | map->m_flags & EXT4_MAP_MAPPED && | |
682 | map->m_flags & EXT4_MAP_NEW) { | |
683 | ret = ext4_issue_zeroout(inode, map->m_lblk, | |
684 | map->m_pblk, map->m_len); | |
685 | if (ret) { | |
686 | retval = ret; | |
687 | goto out_sem; | |
688 | } | |
689 | } | |
690 | ||
adb23551 ZL |
691 | /* |
692 | * If the extent has been zeroed out, we don't need to update | |
693 | * extent status tree. | |
694 | */ | |
695 | if ((flags & EXT4_GET_BLOCKS_PRE_IO) && | |
bb5835ed | 696 | ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) { |
adb23551 | 697 | if (ext4_es_is_written(&es)) |
c86d8db3 | 698 | goto out_sem; |
adb23551 | 699 | } |
f7fec032 ZL |
700 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
701 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; | |
702 | if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) && | |
d2dc317d | 703 | !(status & EXTENT_STATUS_WRITTEN) && |
ad431025 EW |
704 | ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk, |
705 | map->m_lblk + map->m_len - 1)) | |
f7fec032 ZL |
706 | status |= EXTENT_STATUS_DELAYED; |
707 | ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, | |
708 | map->m_pblk, status); | |
c86d8db3 | 709 | if (ret < 0) { |
f7fec032 | 710 | retval = ret; |
c86d8db3 JK |
711 | goto out_sem; |
712 | } | |
5356f261 AK |
713 | } |
714 | ||
c86d8db3 | 715 | out_sem: |
4df3d265 | 716 | up_write((&EXT4_I(inode)->i_data_sem)); |
e35fd660 | 717 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
b8a86845 | 718 | ret = check_block_validity(inode, map); |
6fd058f7 TT |
719 | if (ret != 0) |
720 | return ret; | |
06bd3c36 JK |
721 | |
722 | /* | |
723 | * Inodes with freshly allocated blocks where contents will be | |
724 | * visible after transaction commit must be on transaction's | |
725 | * ordered data list. | |
726 | */ | |
727 | if (map->m_flags & EXT4_MAP_NEW && | |
728 | !(map->m_flags & EXT4_MAP_UNWRITTEN) && | |
729 | !(flags & EXT4_GET_BLOCKS_ZERO) && | |
02749a4c | 730 | !ext4_is_quota_file(inode) && |
06bd3c36 | 731 | ext4_should_order_data(inode)) { |
73131fbb RZ |
732 | loff_t start_byte = |
733 | (loff_t)map->m_lblk << inode->i_blkbits; | |
734 | loff_t length = (loff_t)map->m_len << inode->i_blkbits; | |
735 | ||
ee0876bc | 736 | if (flags & EXT4_GET_BLOCKS_IO_SUBMIT) |
73131fbb RZ |
737 | ret = ext4_jbd2_inode_add_wait(handle, inode, |
738 | start_byte, length); | |
ee0876bc | 739 | else |
73131fbb RZ |
740 | ret = ext4_jbd2_inode_add_write(handle, inode, |
741 | start_byte, length); | |
06bd3c36 JK |
742 | if (ret) |
743 | return ret; | |
744 | } | |
a80f7fcf | 745 | ext4_fc_track_range(handle, inode, map->m_lblk, |
aa75f4d3 | 746 | map->m_lblk + map->m_len - 1); |
6fd058f7 | 747 | } |
ec8c60be RH |
748 | |
749 | if (retval < 0) | |
70aa1554 | 750 | ext_debug(inode, "failed with err %d\n", retval); |
0e855ac8 AK |
751 | return retval; |
752 | } | |
753 | ||
ed8ad838 JK |
754 | /* |
755 | * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages | |
756 | * we have to be careful as someone else may be manipulating b_state as well. | |
757 | */ | |
758 | static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags) | |
759 | { | |
760 | unsigned long old_state; | |
761 | unsigned long new_state; | |
762 | ||
763 | flags &= EXT4_MAP_FLAGS; | |
764 | ||
765 | /* Dummy buffer_head? Set non-atomically. */ | |
766 | if (!bh->b_page) { | |
767 | bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags; | |
768 | return; | |
769 | } | |
770 | /* | |
771 | * Someone else may be modifying b_state. Be careful! This is ugly but | |
772 | * once we get rid of using bh as a container for mapping information | |
773 | * to pass to / from get_block functions, this can go away. | |
774 | */ | |
775 | do { | |
776 | old_state = READ_ONCE(bh->b_state); | |
777 | new_state = (old_state & ~EXT4_MAP_FLAGS) | flags; | |
778 | } while (unlikely( | |
779 | cmpxchg(&bh->b_state, old_state, new_state) != old_state)); | |
780 | } | |
781 | ||
2ed88685 TT |
782 | static int _ext4_get_block(struct inode *inode, sector_t iblock, |
783 | struct buffer_head *bh, int flags) | |
ac27a0ec | 784 | { |
2ed88685 | 785 | struct ext4_map_blocks map; |
efe70c29 | 786 | int ret = 0; |
ac27a0ec | 787 | |
46c7f254 TM |
788 | if (ext4_has_inline_data(inode)) |
789 | return -ERANGE; | |
790 | ||
2ed88685 TT |
791 | map.m_lblk = iblock; |
792 | map.m_len = bh->b_size >> inode->i_blkbits; | |
793 | ||
efe70c29 JK |
794 | ret = ext4_map_blocks(ext4_journal_current_handle(), inode, &map, |
795 | flags); | |
7fb5409d | 796 | if (ret > 0) { |
2ed88685 | 797 | map_bh(bh, inode->i_sb, map.m_pblk); |
ed8ad838 | 798 | ext4_update_bh_state(bh, map.m_flags); |
2ed88685 | 799 | bh->b_size = inode->i_sb->s_blocksize * map.m_len; |
7fb5409d | 800 | ret = 0; |
547edce3 RZ |
801 | } else if (ret == 0) { |
802 | /* hole case, need to fill in bh->b_size */ | |
803 | bh->b_size = inode->i_sb->s_blocksize * map.m_len; | |
ac27a0ec DK |
804 | } |
805 | return ret; | |
806 | } | |
807 | ||
2ed88685 TT |
808 | int ext4_get_block(struct inode *inode, sector_t iblock, |
809 | struct buffer_head *bh, int create) | |
810 | { | |
811 | return _ext4_get_block(inode, iblock, bh, | |
812 | create ? EXT4_GET_BLOCKS_CREATE : 0); | |
813 | } | |
814 | ||
705965bd JK |
815 | /* |
816 | * Get block function used when preparing for buffered write if we require | |
817 | * creating an unwritten extent if blocks haven't been allocated. The extent | |
818 | * will be converted to written after the IO is complete. | |
819 | */ | |
820 | int ext4_get_block_unwritten(struct inode *inode, sector_t iblock, | |
821 | struct buffer_head *bh_result, int create) | |
822 | { | |
823 | ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n", | |
824 | inode->i_ino, create); | |
825 | return _ext4_get_block(inode, iblock, bh_result, | |
826 | EXT4_GET_BLOCKS_IO_CREATE_EXT); | |
827 | } | |
828 | ||
efe70c29 JK |
829 | /* Maximum number of blocks we map for direct IO at once. */ |
830 | #define DIO_MAX_BLOCKS 4096 | |
831 | ||
ac27a0ec DK |
832 | /* |
833 | * `handle' can be NULL if create is zero | |
834 | */ | |
617ba13b | 835 | struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode, |
c5e298ae | 836 | ext4_lblk_t block, int map_flags) |
ac27a0ec | 837 | { |
2ed88685 TT |
838 | struct ext4_map_blocks map; |
839 | struct buffer_head *bh; | |
c5e298ae | 840 | int create = map_flags & EXT4_GET_BLOCKS_CREATE; |
10560082 | 841 | int err; |
ac27a0ec | 842 | |
837c23fb CX |
843 | ASSERT((EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY) |
844 | || handle != NULL || create == 0); | |
ac27a0ec | 845 | |
2ed88685 TT |
846 | map.m_lblk = block; |
847 | map.m_len = 1; | |
c5e298ae | 848 | err = ext4_map_blocks(handle, inode, &map, map_flags); |
ac27a0ec | 849 | |
10560082 TT |
850 | if (err == 0) |
851 | return create ? ERR_PTR(-ENOSPC) : NULL; | |
2ed88685 | 852 | if (err < 0) |
10560082 | 853 | return ERR_PTR(err); |
2ed88685 TT |
854 | |
855 | bh = sb_getblk(inode->i_sb, map.m_pblk); | |
10560082 TT |
856 | if (unlikely(!bh)) |
857 | return ERR_PTR(-ENOMEM); | |
2ed88685 | 858 | if (map.m_flags & EXT4_MAP_NEW) { |
837c23fb CX |
859 | ASSERT(create != 0); |
860 | ASSERT((EXT4_SB(inode->i_sb)->s_mount_state & EXT4_FC_REPLAY) | |
861 | || (handle != NULL)); | |
ac27a0ec | 862 | |
2ed88685 TT |
863 | /* |
864 | * Now that we do not always journal data, we should | |
865 | * keep in mind whether this should always journal the | |
866 | * new buffer as metadata. For now, regular file | |
867 | * writes use ext4_get_block instead, so it's not a | |
868 | * problem. | |
869 | */ | |
870 | lock_buffer(bh); | |
871 | BUFFER_TRACE(bh, "call get_create_access"); | |
10560082 TT |
872 | err = ext4_journal_get_create_access(handle, bh); |
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 | ||
f19d5870 TM |
957 | int ext4_walk_page_buffers(handle_t *handle, |
958 | struct buffer_head *head, | |
959 | unsigned from, | |
960 | unsigned to, | |
961 | int *partial, | |
962 | int (*fn)(handle_t *handle, | |
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 | } | |
981 | err = (*fn)(handle, bh); | |
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 | */ | |
f19d5870 TM |
1012 | int do_journal_get_write_access(handle_t *handle, |
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"); |
56d35a4c JK |
1031 | ret = ext4_journal_get_write_access(handle, bh); |
1032 | if (!ret && dirty) | |
1033 | ret = ext4_handle_dirty_metadata(handle, NULL, bh); | |
1034 | return ret; | |
ac27a0ec DK |
1035 | } |
1036 | ||
643fa961 | 1037 | #ifdef CONFIG_FS_ENCRYPTION |
2058f83a MH |
1038 | static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len, |
1039 | get_block_t *get_block) | |
1040 | { | |
09cbfeaf | 1041 | unsigned from = pos & (PAGE_SIZE - 1); |
2058f83a MH |
1042 | unsigned to = from + len; |
1043 | struct inode *inode = page->mapping->host; | |
1044 | unsigned block_start, block_end; | |
1045 | sector_t block; | |
1046 | int err = 0; | |
1047 | unsigned blocksize = inode->i_sb->s_blocksize; | |
1048 | unsigned bbits; | |
0b578f35 CR |
1049 | struct buffer_head *bh, *head, *wait[2]; |
1050 | int nr_wait = 0; | |
1051 | int i; | |
2058f83a MH |
1052 | |
1053 | BUG_ON(!PageLocked(page)); | |
09cbfeaf KS |
1054 | BUG_ON(from > PAGE_SIZE); |
1055 | BUG_ON(to > PAGE_SIZE); | |
2058f83a MH |
1056 | BUG_ON(from > to); |
1057 | ||
1058 | if (!page_has_buffers(page)) | |
1059 | create_empty_buffers(page, blocksize, 0); | |
1060 | head = page_buffers(page); | |
1061 | bbits = ilog2(blocksize); | |
09cbfeaf | 1062 | block = (sector_t)page->index << (PAGE_SHIFT - bbits); |
2058f83a MH |
1063 | |
1064 | for (bh = head, block_start = 0; bh != head || !block_start; | |
1065 | block++, block_start = block_end, bh = bh->b_this_page) { | |
1066 | block_end = block_start + blocksize; | |
1067 | if (block_end <= from || block_start >= to) { | |
1068 | if (PageUptodate(page)) { | |
1069 | if (!buffer_uptodate(bh)) | |
1070 | set_buffer_uptodate(bh); | |
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)) { | |
1095 | if (!buffer_uptodate(bh)) | |
1096 | set_buffer_uptodate(bh); | |
1097 | continue; | |
1098 | } | |
1099 | if (!buffer_uptodate(bh) && !buffer_delay(bh) && | |
1100 | !buffer_unwritten(bh) && | |
1101 | (block_start < from || block_end > to)) { | |
2d069c08 | 1102 | ext4_read_bh_lock(bh, 0, false); |
0b578f35 | 1103 | wait[nr_wait++] = bh; |
2058f83a MH |
1104 | } |
1105 | } | |
1106 | /* | |
1107 | * If we issued read requests, let them complete. | |
1108 | */ | |
0b578f35 CR |
1109 | for (i = 0; i < nr_wait; i++) { |
1110 | wait_on_buffer(wait[i]); | |
1111 | if (!buffer_uptodate(wait[i])) | |
2058f83a MH |
1112 | err = -EIO; |
1113 | } | |
7e0785fc | 1114 | if (unlikely(err)) { |
2058f83a | 1115 | page_zero_new_buffers(page, from, to); |
4f74d15f | 1116 | } else if (fscrypt_inode_uses_fs_layer_crypto(inode)) { |
0b578f35 CR |
1117 | for (i = 0; i < nr_wait; i++) { |
1118 | int err2; | |
1119 | ||
1120 | err2 = fscrypt_decrypt_pagecache_blocks(page, blocksize, | |
1121 | bh_offset(wait[i])); | |
1122 | if (err2) { | |
1123 | clear_buffer_uptodate(wait[i]); | |
1124 | err = err2; | |
1125 | } | |
1126 | } | |
7e0785fc CR |
1127 | } |
1128 | ||
2058f83a MH |
1129 | return err; |
1130 | } | |
1131 | #endif | |
1132 | ||
bfc1af65 | 1133 | static int ext4_write_begin(struct file *file, struct address_space *mapping, |
de9a55b8 TT |
1134 | loff_t pos, unsigned len, unsigned flags, |
1135 | struct page **pagep, void **fsdata) | |
ac27a0ec | 1136 | { |
af5bc92d | 1137 | struct inode *inode = mapping->host; |
1938a150 | 1138 | int ret, needed_blocks; |
ac27a0ec DK |
1139 | handle_t *handle; |
1140 | int retries = 0; | |
af5bc92d | 1141 | struct page *page; |
de9a55b8 | 1142 | pgoff_t index; |
af5bc92d | 1143 | unsigned from, to; |
bfc1af65 | 1144 | |
0db1ff22 TT |
1145 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
1146 | return -EIO; | |
1147 | ||
9bffad1e | 1148 | trace_ext4_write_begin(inode, pos, len, flags); |
1938a150 AK |
1149 | /* |
1150 | * Reserve one block more for addition to orphan list in case | |
1151 | * we allocate blocks but write fails for some reason | |
1152 | */ | |
1153 | needed_blocks = ext4_writepage_trans_blocks(inode) + 1; | |
09cbfeaf KS |
1154 | index = pos >> PAGE_SHIFT; |
1155 | from = pos & (PAGE_SIZE - 1); | |
af5bc92d | 1156 | to = from + len; |
ac27a0ec | 1157 | |
f19d5870 TM |
1158 | if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { |
1159 | ret = ext4_try_to_write_inline_data(mapping, inode, pos, len, | |
1160 | flags, pagep); | |
1161 | if (ret < 0) | |
47564bfb TT |
1162 | return ret; |
1163 | if (ret == 1) | |
1164 | return 0; | |
f19d5870 TM |
1165 | } |
1166 | ||
47564bfb TT |
1167 | /* |
1168 | * grab_cache_page_write_begin() can take a long time if the | |
1169 | * system is thrashing due to memory pressure, or if the page | |
1170 | * is being written back. So grab it first before we start | |
1171 | * the transaction handle. This also allows us to allocate | |
1172 | * the page (if needed) without using GFP_NOFS. | |
1173 | */ | |
1174 | retry_grab: | |
1175 | page = grab_cache_page_write_begin(mapping, index, flags); | |
1176 | if (!page) | |
1177 | return -ENOMEM; | |
1178 | unlock_page(page); | |
1179 | ||
1180 | retry_journal: | |
9924a92a | 1181 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks); |
af5bc92d | 1182 | if (IS_ERR(handle)) { |
09cbfeaf | 1183 | put_page(page); |
47564bfb | 1184 | return PTR_ERR(handle); |
7479d2b9 | 1185 | } |
ac27a0ec | 1186 | |
47564bfb TT |
1187 | lock_page(page); |
1188 | if (page->mapping != mapping) { | |
1189 | /* The page got truncated from under us */ | |
1190 | unlock_page(page); | |
09cbfeaf | 1191 | put_page(page); |
cf108bca | 1192 | ext4_journal_stop(handle); |
47564bfb | 1193 | goto retry_grab; |
cf108bca | 1194 | } |
7afe5aa5 DM |
1195 | /* In case writeback began while the page was unlocked */ |
1196 | wait_for_stable_page(page); | |
cf108bca | 1197 | |
643fa961 | 1198 | #ifdef CONFIG_FS_ENCRYPTION |
2058f83a MH |
1199 | if (ext4_should_dioread_nolock(inode)) |
1200 | ret = ext4_block_write_begin(page, pos, len, | |
705965bd | 1201 | ext4_get_block_unwritten); |
2058f83a MH |
1202 | else |
1203 | ret = ext4_block_write_begin(page, pos, len, | |
1204 | ext4_get_block); | |
1205 | #else | |
744692dc | 1206 | if (ext4_should_dioread_nolock(inode)) |
705965bd JK |
1207 | ret = __block_write_begin(page, pos, len, |
1208 | ext4_get_block_unwritten); | |
744692dc | 1209 | else |
6e1db88d | 1210 | ret = __block_write_begin(page, pos, len, ext4_get_block); |
2058f83a | 1211 | #endif |
bfc1af65 | 1212 | if (!ret && ext4_should_journal_data(inode)) { |
f19d5870 TM |
1213 | ret = ext4_walk_page_buffers(handle, page_buffers(page), |
1214 | from, to, NULL, | |
1215 | do_journal_get_write_access); | |
ac27a0ec | 1216 | } |
bfc1af65 NP |
1217 | |
1218 | if (ret) { | |
c93d8f88 EB |
1219 | bool extended = (pos + len > inode->i_size) && |
1220 | !ext4_verity_in_progress(inode); | |
1221 | ||
af5bc92d | 1222 | unlock_page(page); |
ae4d5372 | 1223 | /* |
6e1db88d | 1224 | * __block_write_begin may have instantiated a few blocks |
ae4d5372 AK |
1225 | * outside i_size. Trim these off again. Don't need |
1226 | * i_size_read because we hold i_mutex. | |
1938a150 AK |
1227 | * |
1228 | * Add inode to orphan list in case we crash before | |
1229 | * truncate finishes | |
ae4d5372 | 1230 | */ |
c93d8f88 | 1231 | if (extended && ext4_can_truncate(inode)) |
1938a150 AK |
1232 | ext4_orphan_add(handle, inode); |
1233 | ||
1234 | ext4_journal_stop(handle); | |
c93d8f88 | 1235 | if (extended) { |
b9a4207d | 1236 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1237 | /* |
ffacfa7a | 1238 | * If truncate failed early the inode might |
1938a150 AK |
1239 | * still be on the orphan list; we need to |
1240 | * make sure the inode is removed from the | |
1241 | * orphan list in that case. | |
1242 | */ | |
1243 | if (inode->i_nlink) | |
1244 | ext4_orphan_del(NULL, inode); | |
1245 | } | |
bfc1af65 | 1246 | |
47564bfb TT |
1247 | if (ret == -ENOSPC && |
1248 | ext4_should_retry_alloc(inode->i_sb, &retries)) | |
1249 | goto retry_journal; | |
09cbfeaf | 1250 | put_page(page); |
47564bfb TT |
1251 | return ret; |
1252 | } | |
1253 | *pagep = page; | |
ac27a0ec DK |
1254 | return ret; |
1255 | } | |
1256 | ||
bfc1af65 NP |
1257 | /* For write_end() in data=journal mode */ |
1258 | static int write_end_fn(handle_t *handle, 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 JK |
1356 | static void ext4_journalled_zero_new_buffers(handle_t *handle, |
1357 | struct page *page, | |
1358 | unsigned from, unsigned to) | |
b90197b6 TT |
1359 | { |
1360 | unsigned int block_start = 0, block_end; | |
1361 | struct buffer_head *head, *bh; | |
1362 | ||
1363 | bh = head = page_buffers(page); | |
1364 | do { | |
1365 | block_end = block_start + bh->b_size; | |
1366 | if (buffer_new(bh)) { | |
1367 | if (block_end > from && block_start < to) { | |
1368 | if (!PageUptodate(page)) { | |
1369 | unsigned start, size; | |
1370 | ||
1371 | start = max(from, block_start); | |
1372 | size = min(to, block_end) - start; | |
1373 | ||
1374 | zero_user(page, start, size); | |
3b136499 | 1375 | write_end_fn(handle, bh); |
b90197b6 TT |
1376 | } |
1377 | clear_buffer_new(bh); | |
1378 | } | |
1379 | } | |
1380 | block_start = block_end; | |
1381 | bh = bh->b_this_page; | |
1382 | } while (bh != head); | |
1383 | } | |
1384 | ||
bfc1af65 | 1385 | static int ext4_journalled_write_end(struct file *file, |
de9a55b8 TT |
1386 | struct address_space *mapping, |
1387 | loff_t pos, unsigned len, unsigned copied, | |
1388 | struct page *page, void *fsdata) | |
ac27a0ec | 1389 | { |
617ba13b | 1390 | handle_t *handle = ext4_journal_current_handle(); |
bfc1af65 | 1391 | struct inode *inode = mapping->host; |
0572639f | 1392 | loff_t old_size = inode->i_size; |
ac27a0ec DK |
1393 | int ret = 0, ret2; |
1394 | int partial = 0; | |
bfc1af65 | 1395 | unsigned from, to; |
4631dbf6 | 1396 | int size_changed = 0; |
362eca70 | 1397 | int inline_data = ext4_has_inline_data(inode); |
c93d8f88 | 1398 | bool verity = ext4_verity_in_progress(inode); |
ac27a0ec | 1399 | |
9bffad1e | 1400 | trace_ext4_journalled_write_end(inode, pos, len, copied); |
09cbfeaf | 1401 | from = pos & (PAGE_SIZE - 1); |
bfc1af65 NP |
1402 | to = from + len; |
1403 | ||
441c8508 CW |
1404 | BUG_ON(!ext4_handle_valid(handle)); |
1405 | ||
362eca70 | 1406 | if (inline_data) { |
eb5efbcb TT |
1407 | ret = ext4_write_inline_data_end(inode, pos, len, |
1408 | copied, page); | |
1409 | if (ret < 0) { | |
1410 | unlock_page(page); | |
1411 | put_page(page); | |
1412 | goto errout; | |
1413 | } | |
1414 | copied = ret; | |
1415 | } else if (unlikely(copied < len) && !PageUptodate(page)) { | |
3b136499 JK |
1416 | copied = 0; |
1417 | ext4_journalled_zero_new_buffers(handle, page, from, to); | |
1418 | } else { | |
1419 | if (unlikely(copied < len)) | |
1420 | ext4_journalled_zero_new_buffers(handle, page, | |
1421 | from + copied, to); | |
3fdcfb66 | 1422 | ret = ext4_walk_page_buffers(handle, page_buffers(page), from, |
3b136499 JK |
1423 | from + copied, &partial, |
1424 | write_end_fn); | |
3fdcfb66 TM |
1425 | if (!partial) |
1426 | SetPageUptodate(page); | |
1427 | } | |
c93d8f88 EB |
1428 | if (!verity) |
1429 | size_changed = ext4_update_inode_size(inode, pos + copied); | |
19f5fb7a | 1430 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
2d859db3 | 1431 | EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; |
4631dbf6 | 1432 | unlock_page(page); |
09cbfeaf | 1433 | put_page(page); |
4631dbf6 | 1434 | |
c93d8f88 | 1435 | if (old_size < pos && !verity) |
0572639f XW |
1436 | pagecache_isize_extended(inode, old_size, pos); |
1437 | ||
362eca70 | 1438 | if (size_changed || inline_data) { |
617ba13b | 1439 | ret2 = ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
1440 | if (!ret) |
1441 | ret = ret2; | |
1442 | } | |
bfc1af65 | 1443 | |
c93d8f88 | 1444 | if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode)) |
f8514083 AK |
1445 | /* if we have allocated more blocks and copied |
1446 | * less. We will have blocks allocated outside | |
1447 | * inode->i_size. So truncate them | |
1448 | */ | |
1449 | ext4_orphan_add(handle, inode); | |
1450 | ||
eb5efbcb | 1451 | errout: |
617ba13b | 1452 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1453 | if (!ret) |
1454 | ret = ret2; | |
c93d8f88 | 1455 | if (pos + len > inode->i_size && !verity) { |
b9a4207d | 1456 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1457 | /* |
ffacfa7a | 1458 | * If truncate failed early the inode might still be |
f8514083 AK |
1459 | * on the orphan list; we need to make sure the inode |
1460 | * is removed from the orphan list in that case. | |
1461 | */ | |
1462 | if (inode->i_nlink) | |
1463 | ext4_orphan_del(NULL, inode); | |
1464 | } | |
bfc1af65 NP |
1465 | |
1466 | return ret ? ret : copied; | |
ac27a0ec | 1467 | } |
d2a17637 | 1468 | |
9d0be502 | 1469 | /* |
c27e43a1 | 1470 | * Reserve space for a single cluster |
9d0be502 | 1471 | */ |
c27e43a1 | 1472 | static int ext4_da_reserve_space(struct inode *inode) |
d2a17637 | 1473 | { |
60e58e0f | 1474 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
0637c6f4 | 1475 | struct ext4_inode_info *ei = EXT4_I(inode); |
5dd4056d | 1476 | int ret; |
03179fe9 TT |
1477 | |
1478 | /* | |
1479 | * We will charge metadata quota at writeout time; this saves | |
1480 | * us from metadata over-estimation, though we may go over by | |
1481 | * a small amount in the end. Here we just reserve for data. | |
1482 | */ | |
1483 | ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1)); | |
1484 | if (ret) | |
1485 | return ret; | |
d2a17637 | 1486 | |
0637c6f4 | 1487 | spin_lock(&ei->i_block_reservation_lock); |
71d4f7d0 | 1488 | if (ext4_claim_free_clusters(sbi, 1, 0)) { |
03179fe9 | 1489 | spin_unlock(&ei->i_block_reservation_lock); |
03179fe9 | 1490 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1)); |
d2a17637 MC |
1491 | return -ENOSPC; |
1492 | } | |
9d0be502 | 1493 | ei->i_reserved_data_blocks++; |
c27e43a1 | 1494 | trace_ext4_da_reserve_space(inode); |
0637c6f4 | 1495 | spin_unlock(&ei->i_block_reservation_lock); |
39bc680a | 1496 | |
d2a17637 MC |
1497 | return 0; /* success */ |
1498 | } | |
1499 | ||
f456767d | 1500 | void ext4_da_release_space(struct inode *inode, int to_free) |
d2a17637 MC |
1501 | { |
1502 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0637c6f4 | 1503 | struct ext4_inode_info *ei = EXT4_I(inode); |
d2a17637 | 1504 | |
cd213226 MC |
1505 | if (!to_free) |
1506 | return; /* Nothing to release, exit */ | |
1507 | ||
d2a17637 | 1508 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); |
cd213226 | 1509 | |
5a58ec87 | 1510 | trace_ext4_da_release_space(inode, to_free); |
0637c6f4 | 1511 | if (unlikely(to_free > ei->i_reserved_data_blocks)) { |
cd213226 | 1512 | /* |
0637c6f4 TT |
1513 | * if there aren't enough reserved blocks, then the |
1514 | * counter is messed up somewhere. Since this | |
1515 | * function is called from invalidate page, it's | |
1516 | * harmless to return without any action. | |
cd213226 | 1517 | */ |
8de5c325 | 1518 | ext4_warning(inode->i_sb, "ext4_da_release_space: " |
0637c6f4 | 1519 | "ino %lu, to_free %d with only %d reserved " |
1084f252 | 1520 | "data blocks", inode->i_ino, to_free, |
0637c6f4 TT |
1521 | ei->i_reserved_data_blocks); |
1522 | WARN_ON(1); | |
1523 | to_free = ei->i_reserved_data_blocks; | |
cd213226 | 1524 | } |
0637c6f4 | 1525 | ei->i_reserved_data_blocks -= to_free; |
cd213226 | 1526 | |
72b8ab9d | 1527 | /* update fs dirty data blocks counter */ |
57042651 | 1528 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free); |
d2a17637 | 1529 | |
d2a17637 | 1530 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
60e58e0f | 1531 | |
7b415bf6 | 1532 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free)); |
d2a17637 MC |
1533 | } |
1534 | ||
64769240 AT |
1535 | /* |
1536 | * Delayed allocation stuff | |
1537 | */ | |
1538 | ||
4e7ea81d JK |
1539 | struct mpage_da_data { |
1540 | struct inode *inode; | |
1541 | struct writeback_control *wbc; | |
6b523df4 | 1542 | |
4e7ea81d JK |
1543 | pgoff_t first_page; /* The first page to write */ |
1544 | pgoff_t next_page; /* Current page to examine */ | |
1545 | pgoff_t last_page; /* Last page to examine */ | |
791b7f08 | 1546 | /* |
4e7ea81d JK |
1547 | * Extent to map - this can be after first_page because that can be |
1548 | * fully mapped. We somewhat abuse m_flags to store whether the extent | |
1549 | * is delalloc or unwritten. | |
791b7f08 | 1550 | */ |
4e7ea81d JK |
1551 | struct ext4_map_blocks map; |
1552 | struct ext4_io_submit io_submit; /* IO submission data */ | |
dddbd6ac | 1553 | unsigned int do_map:1; |
6b8ed620 | 1554 | unsigned int scanned_until_end:1; |
4e7ea81d | 1555 | }; |
64769240 | 1556 | |
4e7ea81d JK |
1557 | static void mpage_release_unused_pages(struct mpage_da_data *mpd, |
1558 | bool invalidate) | |
c4a0c46e AK |
1559 | { |
1560 | int nr_pages, i; | |
1561 | pgoff_t index, end; | |
1562 | struct pagevec pvec; | |
1563 | struct inode *inode = mpd->inode; | |
1564 | struct address_space *mapping = inode->i_mapping; | |
4e7ea81d JK |
1565 | |
1566 | /* This is necessary when next_page == 0. */ | |
1567 | if (mpd->first_page >= mpd->next_page) | |
1568 | return; | |
c4a0c46e | 1569 | |
6b8ed620 | 1570 | mpd->scanned_until_end = 0; |
c7f5938a CW |
1571 | index = mpd->first_page; |
1572 | end = mpd->next_page - 1; | |
4e7ea81d JK |
1573 | if (invalidate) { |
1574 | ext4_lblk_t start, last; | |
09cbfeaf KS |
1575 | start = index << (PAGE_SHIFT - inode->i_blkbits); |
1576 | last = end << (PAGE_SHIFT - inode->i_blkbits); | |
4e7ea81d JK |
1577 | ext4_es_remove_extent(inode, start, last - start + 1); |
1578 | } | |
51865fda | 1579 | |
86679820 | 1580 | pagevec_init(&pvec); |
c4a0c46e | 1581 | while (index <= end) { |
397162ff | 1582 | nr_pages = pagevec_lookup_range(&pvec, mapping, &index, end); |
c4a0c46e AK |
1583 | if (nr_pages == 0) |
1584 | break; | |
1585 | for (i = 0; i < nr_pages; i++) { | |
1586 | struct page *page = pvec.pages[i]; | |
2b85a617 | 1587 | |
c4a0c46e AK |
1588 | BUG_ON(!PageLocked(page)); |
1589 | BUG_ON(PageWriteback(page)); | |
4e7ea81d | 1590 | if (invalidate) { |
4e800c03 | 1591 | if (page_mapped(page)) |
1592 | clear_page_dirty_for_io(page); | |
09cbfeaf | 1593 | block_invalidatepage(page, 0, PAGE_SIZE); |
4e7ea81d JK |
1594 | ClearPageUptodate(page); |
1595 | } | |
c4a0c46e AK |
1596 | unlock_page(page); |
1597 | } | |
9b1d0998 | 1598 | pagevec_release(&pvec); |
c4a0c46e | 1599 | } |
c4a0c46e AK |
1600 | } |
1601 | ||
df22291f AK |
1602 | static void ext4_print_free_blocks(struct inode *inode) |
1603 | { | |
1604 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
92b97816 | 1605 | struct super_block *sb = inode->i_sb; |
f78ee70d | 1606 | struct ext4_inode_info *ei = EXT4_I(inode); |
92b97816 TT |
1607 | |
1608 | ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld", | |
5dee5437 | 1609 | EXT4_C2B(EXT4_SB(inode->i_sb), |
f78ee70d | 1610 | ext4_count_free_clusters(sb))); |
92b97816 TT |
1611 | ext4_msg(sb, KERN_CRIT, "Free/Dirty block details"); |
1612 | ext4_msg(sb, KERN_CRIT, "free_blocks=%lld", | |
f78ee70d | 1613 | (long long) EXT4_C2B(EXT4_SB(sb), |
57042651 | 1614 | percpu_counter_sum(&sbi->s_freeclusters_counter))); |
92b97816 | 1615 | ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld", |
f78ee70d | 1616 | (long long) EXT4_C2B(EXT4_SB(sb), |
7b415bf6 | 1617 | percpu_counter_sum(&sbi->s_dirtyclusters_counter))); |
92b97816 TT |
1618 | ext4_msg(sb, KERN_CRIT, "Block reservation details"); |
1619 | ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u", | |
f78ee70d | 1620 | ei->i_reserved_data_blocks); |
df22291f AK |
1621 | return; |
1622 | } | |
1623 | ||
c364b22c | 1624 | static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh) |
29fa89d0 | 1625 | { |
c364b22c | 1626 | return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh); |
29fa89d0 AK |
1627 | } |
1628 | ||
0b02f4c0 EW |
1629 | /* |
1630 | * ext4_insert_delayed_block - adds a delayed block to the extents status | |
1631 | * tree, incrementing the reserved cluster/block | |
1632 | * count or making a pending reservation | |
1633 | * where needed | |
1634 | * | |
1635 | * @inode - file containing the newly added block | |
1636 | * @lblk - logical block to be added | |
1637 | * | |
1638 | * Returns 0 on success, negative error code on failure. | |
1639 | */ | |
1640 | static int ext4_insert_delayed_block(struct inode *inode, ext4_lblk_t lblk) | |
1641 | { | |
1642 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
1643 | int ret; | |
1644 | bool allocated = false; | |
1645 | ||
1646 | /* | |
1647 | * If the cluster containing lblk is shared with a delayed, | |
1648 | * written, or unwritten extent in a bigalloc file system, it's | |
1649 | * already been accounted for and does not need to be reserved. | |
1650 | * A pending reservation must be made for the cluster if it's | |
1651 | * shared with a written or unwritten extent and doesn't already | |
1652 | * have one. Written and unwritten extents can be purged from the | |
1653 | * extents status tree if the system is under memory pressure, so | |
1654 | * it's necessary to examine the extent tree if a search of the | |
1655 | * extents status tree doesn't get a match. | |
1656 | */ | |
1657 | if (sbi->s_cluster_ratio == 1) { | |
1658 | ret = ext4_da_reserve_space(inode); | |
1659 | if (ret != 0) /* ENOSPC */ | |
1660 | goto errout; | |
1661 | } else { /* bigalloc */ | |
1662 | if (!ext4_es_scan_clu(inode, &ext4_es_is_delonly, lblk)) { | |
1663 | if (!ext4_es_scan_clu(inode, | |
1664 | &ext4_es_is_mapped, lblk)) { | |
1665 | ret = ext4_clu_mapped(inode, | |
1666 | EXT4_B2C(sbi, lblk)); | |
1667 | if (ret < 0) | |
1668 | goto errout; | |
1669 | if (ret == 0) { | |
1670 | ret = ext4_da_reserve_space(inode); | |
1671 | if (ret != 0) /* ENOSPC */ | |
1672 | goto errout; | |
1673 | } else { | |
1674 | allocated = true; | |
1675 | } | |
1676 | } else { | |
1677 | allocated = true; | |
1678 | } | |
1679 | } | |
1680 | } | |
1681 | ||
1682 | ret = ext4_es_insert_delayed_block(inode, lblk, allocated); | |
1683 | ||
1684 | errout: | |
1685 | return ret; | |
1686 | } | |
1687 | ||
5356f261 AK |
1688 | /* |
1689 | * This function is grabs code from the very beginning of | |
1690 | * ext4_map_blocks, but assumes that the caller is from delayed write | |
1691 | * time. This function looks up the requested blocks and sets the | |
1692 | * buffer delay bit under the protection of i_data_sem. | |
1693 | */ | |
1694 | static int ext4_da_map_blocks(struct inode *inode, sector_t iblock, | |
1695 | struct ext4_map_blocks *map, | |
1696 | struct buffer_head *bh) | |
1697 | { | |
d100eef2 | 1698 | struct extent_status es; |
5356f261 AK |
1699 | int retval; |
1700 | sector_t invalid_block = ~((sector_t) 0xffff); | |
921f266b DM |
1701 | #ifdef ES_AGGRESSIVE_TEST |
1702 | struct ext4_map_blocks orig_map; | |
1703 | ||
1704 | memcpy(&orig_map, map, sizeof(*map)); | |
1705 | #endif | |
5356f261 AK |
1706 | |
1707 | if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es)) | |
1708 | invalid_block = ~0; | |
1709 | ||
1710 | map->m_flags = 0; | |
70aa1554 | 1711 | ext_debug(inode, "max_blocks %u, logical block %lu\n", map->m_len, |
5356f261 | 1712 | (unsigned long) map->m_lblk); |
d100eef2 ZL |
1713 | |
1714 | /* Lookup extent status tree firstly */ | |
bb5835ed | 1715 | if (ext4_es_lookup_extent(inode, iblock, NULL, &es)) { |
d100eef2 ZL |
1716 | if (ext4_es_is_hole(&es)) { |
1717 | retval = 0; | |
c8b459f4 | 1718 | down_read(&EXT4_I(inode)->i_data_sem); |
d100eef2 ZL |
1719 | goto add_delayed; |
1720 | } | |
1721 | ||
1722 | /* | |
1723 | * Delayed extent could be allocated by fallocate. | |
1724 | * So we need to check it. | |
1725 | */ | |
1726 | if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) { | |
1727 | map_bh(bh, inode->i_sb, invalid_block); | |
1728 | set_buffer_new(bh); | |
1729 | set_buffer_delay(bh); | |
1730 | return 0; | |
1731 | } | |
1732 | ||
1733 | map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk; | |
1734 | retval = es.es_len - (iblock - es.es_lblk); | |
1735 | if (retval > map->m_len) | |
1736 | retval = map->m_len; | |
1737 | map->m_len = retval; | |
1738 | if (ext4_es_is_written(&es)) | |
1739 | map->m_flags |= EXT4_MAP_MAPPED; | |
1740 | else if (ext4_es_is_unwritten(&es)) | |
1741 | map->m_flags |= EXT4_MAP_UNWRITTEN; | |
1742 | else | |
1e83bc81 | 1743 | BUG(); |
d100eef2 | 1744 | |
921f266b DM |
1745 | #ifdef ES_AGGRESSIVE_TEST |
1746 | ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0); | |
1747 | #endif | |
d100eef2 ZL |
1748 | return retval; |
1749 | } | |
1750 | ||
5356f261 AK |
1751 | /* |
1752 | * Try to see if we can get the block without requesting a new | |
1753 | * file system block. | |
1754 | */ | |
c8b459f4 | 1755 | down_read(&EXT4_I(inode)->i_data_sem); |
cbd7584e | 1756 | if (ext4_has_inline_data(inode)) |
9c3569b5 | 1757 | retval = 0; |
cbd7584e | 1758 | else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
2f8e0a7c | 1759 | retval = ext4_ext_map_blocks(NULL, inode, map, 0); |
5356f261 | 1760 | else |
2f8e0a7c | 1761 | retval = ext4_ind_map_blocks(NULL, inode, map, 0); |
5356f261 | 1762 | |
d100eef2 | 1763 | add_delayed: |
5356f261 | 1764 | if (retval == 0) { |
f7fec032 | 1765 | int ret; |
ad431025 | 1766 | |
5356f261 AK |
1767 | /* |
1768 | * XXX: __block_prepare_write() unmaps passed block, | |
1769 | * is it OK? | |
1770 | */ | |
5356f261 | 1771 | |
0b02f4c0 EW |
1772 | ret = ext4_insert_delayed_block(inode, map->m_lblk); |
1773 | if (ret != 0) { | |
f7fec032 | 1774 | retval = ret; |
51865fda | 1775 | goto out_unlock; |
f7fec032 | 1776 | } |
51865fda | 1777 | |
5356f261 AK |
1778 | map_bh(bh, inode->i_sb, invalid_block); |
1779 | set_buffer_new(bh); | |
1780 | set_buffer_delay(bh); | |
f7fec032 ZL |
1781 | } else if (retval > 0) { |
1782 | int ret; | |
3be78c73 | 1783 | unsigned int status; |
f7fec032 | 1784 | |
44fb851d ZL |
1785 | if (unlikely(retval != map->m_len)) { |
1786 | ext4_warning(inode->i_sb, | |
1787 | "ES len assertion failed for inode " | |
1788 | "%lu: retval %d != map->m_len %d", | |
1789 | inode->i_ino, retval, map->m_len); | |
1790 | WARN_ON(1); | |
921f266b | 1791 | } |
921f266b | 1792 | |
f7fec032 ZL |
1793 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
1794 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; | |
1795 | ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, | |
1796 | map->m_pblk, status); | |
1797 | if (ret != 0) | |
1798 | retval = ret; | |
5356f261 AK |
1799 | } |
1800 | ||
1801 | out_unlock: | |
1802 | up_read((&EXT4_I(inode)->i_data_sem)); | |
1803 | ||
1804 | return retval; | |
1805 | } | |
1806 | ||
64769240 | 1807 | /* |
d91bd2c1 | 1808 | * This is a special get_block_t callback which is used by |
b920c755 TT |
1809 | * ext4_da_write_begin(). It will either return mapped block or |
1810 | * reserve space for a single block. | |
29fa89d0 AK |
1811 | * |
1812 | * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set. | |
1813 | * We also have b_blocknr = -1 and b_bdev initialized properly | |
1814 | * | |
1815 | * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set. | |
1816 | * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev | |
1817 | * initialized properly. | |
64769240 | 1818 | */ |
9c3569b5 TM |
1819 | int ext4_da_get_block_prep(struct inode *inode, sector_t iblock, |
1820 | struct buffer_head *bh, int create) | |
64769240 | 1821 | { |
2ed88685 | 1822 | struct ext4_map_blocks map; |
64769240 AT |
1823 | int ret = 0; |
1824 | ||
1825 | BUG_ON(create == 0); | |
2ed88685 TT |
1826 | BUG_ON(bh->b_size != inode->i_sb->s_blocksize); |
1827 | ||
1828 | map.m_lblk = iblock; | |
1829 | map.m_len = 1; | |
64769240 AT |
1830 | |
1831 | /* | |
1832 | * first, we need to know whether the block is allocated already | |
1833 | * preallocated blocks are unmapped but should treated | |
1834 | * the same as allocated blocks. | |
1835 | */ | |
5356f261 AK |
1836 | ret = ext4_da_map_blocks(inode, iblock, &map, bh); |
1837 | if (ret <= 0) | |
2ed88685 | 1838 | return ret; |
64769240 | 1839 | |
2ed88685 | 1840 | map_bh(bh, inode->i_sb, map.m_pblk); |
ed8ad838 | 1841 | ext4_update_bh_state(bh, map.m_flags); |
2ed88685 TT |
1842 | |
1843 | if (buffer_unwritten(bh)) { | |
1844 | /* A delayed write to unwritten bh should be marked | |
1845 | * new and mapped. Mapped ensures that we don't do | |
1846 | * get_block multiple times when we write to the same | |
1847 | * offset and new ensures that we do proper zero out | |
1848 | * for partial write. | |
1849 | */ | |
1850 | set_buffer_new(bh); | |
c8205636 | 1851 | set_buffer_mapped(bh); |
2ed88685 TT |
1852 | } |
1853 | return 0; | |
64769240 | 1854 | } |
61628a3f | 1855 | |
62e086be AK |
1856 | static int bget_one(handle_t *handle, struct buffer_head *bh) |
1857 | { | |
1858 | get_bh(bh); | |
1859 | return 0; | |
1860 | } | |
1861 | ||
1862 | static int bput_one(handle_t *handle, struct buffer_head *bh) | |
1863 | { | |
1864 | put_bh(bh); | |
1865 | return 0; | |
1866 | } | |
1867 | ||
1868 | static int __ext4_journalled_writepage(struct page *page, | |
62e086be AK |
1869 | unsigned int len) |
1870 | { | |
1871 | struct address_space *mapping = page->mapping; | |
1872 | struct inode *inode = mapping->host; | |
3fdcfb66 | 1873 | struct buffer_head *page_bufs = NULL; |
62e086be | 1874 | handle_t *handle = NULL; |
3fdcfb66 TM |
1875 | int ret = 0, err = 0; |
1876 | int inline_data = ext4_has_inline_data(inode); | |
1877 | struct buffer_head *inode_bh = NULL; | |
62e086be | 1878 | |
cb20d518 | 1879 | ClearPageChecked(page); |
3fdcfb66 TM |
1880 | |
1881 | if (inline_data) { | |
1882 | BUG_ON(page->index != 0); | |
1883 | BUG_ON(len > ext4_get_max_inline_size(inode)); | |
1884 | inode_bh = ext4_journalled_write_inline_data(inode, len, page); | |
1885 | if (inode_bh == NULL) | |
1886 | goto out; | |
1887 | } else { | |
1888 | page_bufs = page_buffers(page); | |
1889 | if (!page_bufs) { | |
1890 | BUG(); | |
1891 | goto out; | |
1892 | } | |
1893 | ext4_walk_page_buffers(handle, page_bufs, 0, len, | |
1894 | NULL, bget_one); | |
1895 | } | |
bdf96838 TT |
1896 | /* |
1897 | * We need to release the page lock before we start the | |
1898 | * journal, so grab a reference so the page won't disappear | |
1899 | * out from under us. | |
1900 | */ | |
1901 | get_page(page); | |
62e086be AK |
1902 | unlock_page(page); |
1903 | ||
9924a92a TT |
1904 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
1905 | ext4_writepage_trans_blocks(inode)); | |
62e086be AK |
1906 | if (IS_ERR(handle)) { |
1907 | ret = PTR_ERR(handle); | |
bdf96838 TT |
1908 | put_page(page); |
1909 | goto out_no_pagelock; | |
62e086be | 1910 | } |
441c8508 CW |
1911 | BUG_ON(!ext4_handle_valid(handle)); |
1912 | ||
bdf96838 TT |
1913 | lock_page(page); |
1914 | put_page(page); | |
1915 | if (page->mapping != mapping) { | |
1916 | /* The page got truncated from under us */ | |
1917 | ext4_journal_stop(handle); | |
1918 | ret = 0; | |
1919 | goto out; | |
1920 | } | |
1921 | ||
3fdcfb66 | 1922 | if (inline_data) { |
362eca70 | 1923 | ret = ext4_mark_inode_dirty(handle, inode); |
3fdcfb66 TM |
1924 | } else { |
1925 | ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL, | |
1926 | do_journal_get_write_access); | |
1927 | ||
1928 | err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL, | |
1929 | write_end_fn); | |
1930 | } | |
afb585a9 MFO |
1931 | if (ret == 0) |
1932 | ret = err; | |
b5b18160 | 1933 | err = ext4_jbd2_inode_add_write(handle, inode, page_offset(page), len); |
62e086be AK |
1934 | if (ret == 0) |
1935 | ret = err; | |
2d859db3 | 1936 | EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; |
62e086be AK |
1937 | err = ext4_journal_stop(handle); |
1938 | if (!ret) | |
1939 | ret = err; | |
1940 | ||
19f5fb7a | 1941 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
62e086be | 1942 | out: |
bdf96838 TT |
1943 | unlock_page(page); |
1944 | out_no_pagelock: | |
c915fb80 ZZ |
1945 | if (!inline_data && page_bufs) |
1946 | ext4_walk_page_buffers(NULL, page_bufs, 0, len, | |
1947 | NULL, bput_one); | |
3fdcfb66 | 1948 | brelse(inode_bh); |
62e086be AK |
1949 | return ret; |
1950 | } | |
1951 | ||
61628a3f | 1952 | /* |
43ce1d23 AK |
1953 | * Note that we don't need to start a transaction unless we're journaling data |
1954 | * because we should have holes filled from ext4_page_mkwrite(). We even don't | |
1955 | * need to file the inode to the transaction's list in ordered mode because if | |
1956 | * we are writing back data added by write(), the inode is already there and if | |
25985edc | 1957 | * we are writing back data modified via mmap(), no one guarantees in which |
43ce1d23 AK |
1958 | * transaction the data will hit the disk. In case we are journaling data, we |
1959 | * cannot start transaction directly because transaction start ranks above page | |
1960 | * lock so we have to do some magic. | |
1961 | * | |
b920c755 | 1962 | * This function can get called via... |
20970ba6 | 1963 | * - ext4_writepages after taking page lock (have journal handle) |
b920c755 | 1964 | * - journal_submit_inode_data_buffers (no journal handle) |
f6463b0d | 1965 | * - shrink_page_list via the kswapd/direct reclaim (no journal handle) |
b920c755 | 1966 | * - grab_page_cache when doing write_begin (have journal handle) |
43ce1d23 AK |
1967 | * |
1968 | * We don't do any block allocation in this function. If we have page with | |
1969 | * multiple blocks we need to write those buffer_heads that are mapped. This | |
1970 | * is important for mmaped based write. So if we do with blocksize 1K | |
1971 | * truncate(f, 1024); | |
1972 | * a = mmap(f, 0, 4096); | |
1973 | * a[0] = 'a'; | |
1974 | * truncate(f, 4096); | |
1975 | * we have in the page first buffer_head mapped via page_mkwrite call back | |
90802ed9 | 1976 | * but other buffer_heads would be unmapped but dirty (dirty done via the |
43ce1d23 AK |
1977 | * do_wp_page). So writepage should write the first block. If we modify |
1978 | * the mmap area beyond 1024 we will again get a page_fault and the | |
1979 | * page_mkwrite callback will do the block allocation and mark the | |
1980 | * buffer_heads mapped. | |
1981 | * | |
1982 | * We redirty the page if we have any buffer_heads that is either delay or | |
1983 | * unwritten in the page. | |
1984 | * | |
1985 | * We can get recursively called as show below. | |
1986 | * | |
1987 | * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() -> | |
1988 | * ext4_writepage() | |
1989 | * | |
1990 | * But since we don't do any block allocation we should not deadlock. | |
1991 | * Page also have the dirty flag cleared so we don't get recurive page_lock. | |
61628a3f | 1992 | */ |
43ce1d23 | 1993 | static int ext4_writepage(struct page *page, |
62e086be | 1994 | struct writeback_control *wbc) |
64769240 | 1995 | { |
f8bec370 | 1996 | int ret = 0; |
61628a3f | 1997 | loff_t size; |
498e5f24 | 1998 | unsigned int len; |
744692dc | 1999 | struct buffer_head *page_bufs = NULL; |
61628a3f | 2000 | struct inode *inode = page->mapping->host; |
36ade451 | 2001 | struct ext4_io_submit io_submit; |
1c8349a1 | 2002 | bool keep_towrite = false; |
61628a3f | 2003 | |
0db1ff22 | 2004 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) { |
c2a559bc | 2005 | inode->i_mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE); |
0db1ff22 TT |
2006 | unlock_page(page); |
2007 | return -EIO; | |
2008 | } | |
2009 | ||
a9c667f8 | 2010 | trace_ext4_writepage(page); |
f0e6c985 | 2011 | size = i_size_read(inode); |
c93d8f88 EB |
2012 | if (page->index == size >> PAGE_SHIFT && |
2013 | !ext4_verity_in_progress(inode)) | |
09cbfeaf | 2014 | len = size & ~PAGE_MASK; |
f0e6c985 | 2015 | else |
09cbfeaf | 2016 | len = PAGE_SIZE; |
64769240 | 2017 | |
a42afc5f | 2018 | page_bufs = page_buffers(page); |
a42afc5f | 2019 | /* |
fe386132 JK |
2020 | * We cannot do block allocation or other extent handling in this |
2021 | * function. If there are buffers needing that, we have to redirty | |
2022 | * the page. But we may reach here when we do a journal commit via | |
2023 | * journal_submit_inode_data_buffers() and in that case we must write | |
2024 | * allocated buffers to achieve data=ordered mode guarantees. | |
cccd147a TT |
2025 | * |
2026 | * Also, if there is only one buffer per page (the fs block | |
2027 | * size == the page size), if one buffer needs block | |
2028 | * allocation or needs to modify the extent tree to clear the | |
2029 | * unwritten flag, we know that the page can't be written at | |
2030 | * all, so we might as well refuse the write immediately. | |
2031 | * Unfortunately if the block size != page size, we can't as | |
2032 | * easily detect this case using ext4_walk_page_buffers(), but | |
2033 | * for the extremely common case, this is an optimization that | |
2034 | * skips a useless round trip through ext4_bio_write_page(). | |
a42afc5f | 2035 | */ |
f19d5870 TM |
2036 | if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL, |
2037 | ext4_bh_delay_or_unwritten)) { | |
f8bec370 | 2038 | redirty_page_for_writepage(wbc, page); |
cccd147a | 2039 | if ((current->flags & PF_MEMALLOC) || |
09cbfeaf | 2040 | (inode->i_sb->s_blocksize == PAGE_SIZE)) { |
fe386132 JK |
2041 | /* |
2042 | * For memory cleaning there's no point in writing only | |
2043 | * some buffers. So just bail out. Warn if we came here | |
2044 | * from direct reclaim. | |
2045 | */ | |
2046 | WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) | |
2047 | == PF_MEMALLOC); | |
f0e6c985 AK |
2048 | unlock_page(page); |
2049 | return 0; | |
2050 | } | |
1c8349a1 | 2051 | keep_towrite = true; |
a42afc5f | 2052 | } |
64769240 | 2053 | |
cb20d518 | 2054 | if (PageChecked(page) && ext4_should_journal_data(inode)) |
43ce1d23 AK |
2055 | /* |
2056 | * It's mmapped pagecache. Add buffers and journal it. There | |
2057 | * doesn't seem much point in redirtying the page here. | |
2058 | */ | |
3f0ca309 | 2059 | return __ext4_journalled_writepage(page, len); |
43ce1d23 | 2060 | |
97a851ed JK |
2061 | ext4_io_submit_init(&io_submit, wbc); |
2062 | io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS); | |
2063 | if (!io_submit.io_end) { | |
2064 | redirty_page_for_writepage(wbc, page); | |
2065 | unlock_page(page); | |
2066 | return -ENOMEM; | |
2067 | } | |
be993933 | 2068 | ret = ext4_bio_write_page(&io_submit, page, len, keep_towrite); |
36ade451 | 2069 | ext4_io_submit(&io_submit); |
97a851ed JK |
2070 | /* Drop io_end reference we got from init */ |
2071 | ext4_put_io_end_defer(io_submit.io_end); | |
64769240 AT |
2072 | return ret; |
2073 | } | |
2074 | ||
5f1132b2 JK |
2075 | static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page) |
2076 | { | |
2077 | int len; | |
a056bdaa | 2078 | loff_t size; |
5f1132b2 JK |
2079 | int err; |
2080 | ||
2081 | BUG_ON(page->index != mpd->first_page); | |
a056bdaa JK |
2082 | clear_page_dirty_for_io(page); |
2083 | /* | |
2084 | * We have to be very careful here! Nothing protects writeback path | |
2085 | * against i_size changes and the page can be writeably mapped into | |
2086 | * page tables. So an application can be growing i_size and writing | |
2087 | * data through mmap while writeback runs. clear_page_dirty_for_io() | |
2088 | * write-protects our page in page tables and the page cannot get | |
2089 | * written to again until we release page lock. So only after | |
2090 | * clear_page_dirty_for_io() we are safe to sample i_size for | |
2091 | * ext4_bio_write_page() to zero-out tail of the written page. We rely | |
2092 | * on the barrier provided by TestClearPageDirty in | |
2093 | * clear_page_dirty_for_io() to make sure i_size is really sampled only | |
2094 | * after page tables are updated. | |
2095 | */ | |
2096 | size = i_size_read(mpd->inode); | |
c93d8f88 EB |
2097 | if (page->index == size >> PAGE_SHIFT && |
2098 | !ext4_verity_in_progress(mpd->inode)) | |
09cbfeaf | 2099 | len = size & ~PAGE_MASK; |
5f1132b2 | 2100 | else |
09cbfeaf | 2101 | len = PAGE_SIZE; |
be993933 | 2102 | err = ext4_bio_write_page(&mpd->io_submit, page, len, false); |
5f1132b2 JK |
2103 | if (!err) |
2104 | mpd->wbc->nr_to_write--; | |
2105 | mpd->first_page++; | |
2106 | ||
2107 | return err; | |
2108 | } | |
2109 | ||
6db07461 | 2110 | #define BH_FLAGS (BIT(BH_Unwritten) | BIT(BH_Delay)) |
4e7ea81d | 2111 | |
61628a3f | 2112 | /* |
fffb2739 JK |
2113 | * mballoc gives us at most this number of blocks... |
2114 | * XXX: That seems to be only a limitation of ext4_mb_normalize_request(). | |
70261f56 | 2115 | * The rest of mballoc seems to handle chunks up to full group size. |
61628a3f | 2116 | */ |
fffb2739 | 2117 | #define MAX_WRITEPAGES_EXTENT_LEN 2048 |
525f4ed8 | 2118 | |
4e7ea81d JK |
2119 | /* |
2120 | * mpage_add_bh_to_extent - try to add bh to extent of blocks to map | |
2121 | * | |
2122 | * @mpd - extent of blocks | |
2123 | * @lblk - logical number of the block in the file | |
09930042 | 2124 | * @bh - buffer head we want to add to the extent |
4e7ea81d | 2125 | * |
09930042 JK |
2126 | * The function is used to collect contig. blocks in the same state. If the |
2127 | * buffer doesn't require mapping for writeback and we haven't started the | |
2128 | * extent of buffers to map yet, the function returns 'true' immediately - the | |
2129 | * caller can write the buffer right away. Otherwise the function returns true | |
2130 | * if the block has been added to the extent, false if the block couldn't be | |
2131 | * added. | |
4e7ea81d | 2132 | */ |
09930042 JK |
2133 | static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk, |
2134 | struct buffer_head *bh) | |
4e7ea81d JK |
2135 | { |
2136 | struct ext4_map_blocks *map = &mpd->map; | |
2137 | ||
09930042 JK |
2138 | /* Buffer that doesn't need mapping for writeback? */ |
2139 | if (!buffer_dirty(bh) || !buffer_mapped(bh) || | |
2140 | (!buffer_delay(bh) && !buffer_unwritten(bh))) { | |
2141 | /* So far no extent to map => we write the buffer right away */ | |
2142 | if (map->m_len == 0) | |
2143 | return true; | |
2144 | return false; | |
2145 | } | |
4e7ea81d JK |
2146 | |
2147 | /* First block in the extent? */ | |
2148 | if (map->m_len == 0) { | |
dddbd6ac JK |
2149 | /* We cannot map unless handle is started... */ |
2150 | if (!mpd->do_map) | |
2151 | return false; | |
4e7ea81d JK |
2152 | map->m_lblk = lblk; |
2153 | map->m_len = 1; | |
09930042 JK |
2154 | map->m_flags = bh->b_state & BH_FLAGS; |
2155 | return true; | |
4e7ea81d JK |
2156 | } |
2157 | ||
09930042 JK |
2158 | /* Don't go larger than mballoc is willing to allocate */ |
2159 | if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN) | |
2160 | return false; | |
2161 | ||
4e7ea81d JK |
2162 | /* Can we merge the block to our big extent? */ |
2163 | if (lblk == map->m_lblk + map->m_len && | |
09930042 | 2164 | (bh->b_state & BH_FLAGS) == map->m_flags) { |
4e7ea81d | 2165 | map->m_len++; |
09930042 | 2166 | return true; |
4e7ea81d | 2167 | } |
09930042 | 2168 | return false; |
4e7ea81d JK |
2169 | } |
2170 | ||
5f1132b2 JK |
2171 | /* |
2172 | * mpage_process_page_bufs - submit page buffers for IO or add them to extent | |
2173 | * | |
2174 | * @mpd - extent of blocks for mapping | |
2175 | * @head - the first buffer in the page | |
2176 | * @bh - buffer we should start processing from | |
2177 | * @lblk - logical number of the block in the file corresponding to @bh | |
2178 | * | |
2179 | * Walk through page buffers from @bh upto @head (exclusive) and either submit | |
2180 | * the page for IO if all buffers in this page were mapped and there's no | |
2181 | * accumulated extent of buffers to map or add buffers in the page to the | |
2182 | * extent of buffers to map. The function returns 1 if the caller can continue | |
2183 | * by processing the next page, 0 if it should stop adding buffers to the | |
2184 | * extent to map because we cannot extend it anymore. It can also return value | |
2185 | * < 0 in case of error during IO submission. | |
2186 | */ | |
2187 | static int mpage_process_page_bufs(struct mpage_da_data *mpd, | |
2188 | struct buffer_head *head, | |
2189 | struct buffer_head *bh, | |
2190 | ext4_lblk_t lblk) | |
4e7ea81d JK |
2191 | { |
2192 | struct inode *inode = mpd->inode; | |
5f1132b2 | 2193 | int err; |
93407472 | 2194 | ext4_lblk_t blocks = (i_size_read(inode) + i_blocksize(inode) - 1) |
4e7ea81d JK |
2195 | >> inode->i_blkbits; |
2196 | ||
c93d8f88 EB |
2197 | if (ext4_verity_in_progress(inode)) |
2198 | blocks = EXT_MAX_BLOCKS; | |
2199 | ||
4e7ea81d JK |
2200 | do { |
2201 | BUG_ON(buffer_locked(bh)); | |
2202 | ||
09930042 | 2203 | if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) { |
4e7ea81d JK |
2204 | /* Found extent to map? */ |
2205 | if (mpd->map.m_len) | |
5f1132b2 | 2206 | return 0; |
dddbd6ac JK |
2207 | /* Buffer needs mapping and handle is not started? */ |
2208 | if (!mpd->do_map) | |
2209 | return 0; | |
09930042 | 2210 | /* Everything mapped so far and we hit EOF */ |
5f1132b2 | 2211 | break; |
4e7ea81d | 2212 | } |
4e7ea81d | 2213 | } while (lblk++, (bh = bh->b_this_page) != head); |
5f1132b2 JK |
2214 | /* So far everything mapped? Submit the page for IO. */ |
2215 | if (mpd->map.m_len == 0) { | |
2216 | err = mpage_submit_page(mpd, head->b_page); | |
2217 | if (err < 0) | |
2218 | return err; | |
2219 | } | |
6b8ed620 JK |
2220 | if (lblk >= blocks) { |
2221 | mpd->scanned_until_end = 1; | |
2222 | return 0; | |
2223 | } | |
2224 | return 1; | |
4e7ea81d JK |
2225 | } |
2226 | ||
2943fdbc RH |
2227 | /* |
2228 | * mpage_process_page - update page buffers corresponding to changed extent and | |
2229 | * may submit fully mapped page for IO | |
2230 | * | |
2231 | * @mpd - description of extent to map, on return next extent to map | |
2232 | * @m_lblk - logical block mapping. | |
2233 | * @m_pblk - corresponding physical mapping. | |
2234 | * @map_bh - determines on return whether this page requires any further | |
2235 | * mapping or not. | |
2236 | * Scan given page buffers corresponding to changed extent and update buffer | |
2237 | * state according to new extent state. | |
2238 | * We map delalloc buffers to their physical location, clear unwritten bits. | |
2239 | * If the given page is not fully mapped, we update @map to the next extent in | |
2240 | * the given page that needs mapping & return @map_bh as true. | |
2241 | */ | |
2242 | static int mpage_process_page(struct mpage_da_data *mpd, struct page *page, | |
2243 | ext4_lblk_t *m_lblk, ext4_fsblk_t *m_pblk, | |
2244 | bool *map_bh) | |
2245 | { | |
2246 | struct buffer_head *head, *bh; | |
2247 | ext4_io_end_t *io_end = mpd->io_submit.io_end; | |
2248 | ext4_lblk_t lblk = *m_lblk; | |
2249 | ext4_fsblk_t pblock = *m_pblk; | |
2250 | int err = 0; | |
c8cc8816 RH |
2251 | int blkbits = mpd->inode->i_blkbits; |
2252 | ssize_t io_end_size = 0; | |
2253 | struct ext4_io_end_vec *io_end_vec = ext4_last_io_end_vec(io_end); | |
2943fdbc RH |
2254 | |
2255 | bh = head = page_buffers(page); | |
2256 | do { | |
2257 | if (lblk < mpd->map.m_lblk) | |
2258 | continue; | |
2259 | if (lblk >= mpd->map.m_lblk + mpd->map.m_len) { | |
2260 | /* | |
2261 | * Buffer after end of mapped extent. | |
2262 | * Find next buffer in the page to map. | |
2263 | */ | |
2264 | mpd->map.m_len = 0; | |
2265 | mpd->map.m_flags = 0; | |
c8cc8816 RH |
2266 | io_end_vec->size += io_end_size; |
2267 | io_end_size = 0; | |
2943fdbc | 2268 | |
2943fdbc RH |
2269 | err = mpage_process_page_bufs(mpd, head, bh, lblk); |
2270 | if (err > 0) | |
2271 | err = 0; | |
c8cc8816 RH |
2272 | if (!err && mpd->map.m_len && mpd->map.m_lblk > lblk) { |
2273 | io_end_vec = ext4_alloc_io_end_vec(io_end); | |
4d06bfb9 RH |
2274 | if (IS_ERR(io_end_vec)) { |
2275 | err = PTR_ERR(io_end_vec); | |
2276 | goto out; | |
2277 | } | |
d1e18b88 | 2278 | io_end_vec->offset = (loff_t)mpd->map.m_lblk << blkbits; |
c8cc8816 | 2279 | } |
2943fdbc RH |
2280 | *map_bh = true; |
2281 | goto out; | |
2282 | } | |
2283 | if (buffer_delay(bh)) { | |
2284 | clear_buffer_delay(bh); | |
2285 | bh->b_blocknr = pblock++; | |
2286 | } | |
2287 | clear_buffer_unwritten(bh); | |
c8cc8816 | 2288 | io_end_size += (1 << blkbits); |
2943fdbc | 2289 | } while (lblk++, (bh = bh->b_this_page) != head); |
c8cc8816 RH |
2290 | |
2291 | io_end_vec->size += io_end_size; | |
2292 | io_end_size = 0; | |
2943fdbc RH |
2293 | *map_bh = false; |
2294 | out: | |
2295 | *m_lblk = lblk; | |
2296 | *m_pblk = pblock; | |
2297 | return err; | |
2298 | } | |
2299 | ||
4e7ea81d JK |
2300 | /* |
2301 | * mpage_map_buffers - update buffers corresponding to changed extent and | |
2302 | * submit fully mapped pages for IO | |
2303 | * | |
2304 | * @mpd - description of extent to map, on return next extent to map | |
2305 | * | |
2306 | * Scan buffers corresponding to changed extent (we expect corresponding pages | |
2307 | * to be already locked) and update buffer state according to new extent state. | |
2308 | * We map delalloc buffers to their physical location, clear unwritten bits, | |
556615dc | 2309 | * and mark buffers as uninit when we perform writes to unwritten extents |
4e7ea81d JK |
2310 | * and do extent conversion after IO is finished. If the last page is not fully |
2311 | * mapped, we update @map to the next extent in the last page that needs | |
2312 | * mapping. Otherwise we submit the page for IO. | |
2313 | */ | |
2314 | static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd) | |
2315 | { | |
2316 | struct pagevec pvec; | |
2317 | int nr_pages, i; | |
2318 | struct inode *inode = mpd->inode; | |
09cbfeaf | 2319 | int bpp_bits = PAGE_SHIFT - inode->i_blkbits; |
4e7ea81d JK |
2320 | pgoff_t start, end; |
2321 | ext4_lblk_t lblk; | |
2943fdbc | 2322 | ext4_fsblk_t pblock; |
4e7ea81d | 2323 | int err; |
2943fdbc | 2324 | bool map_bh = false; |
4e7ea81d JK |
2325 | |
2326 | start = mpd->map.m_lblk >> bpp_bits; | |
2327 | end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits; | |
2328 | lblk = start << bpp_bits; | |
2329 | pblock = mpd->map.m_pblk; | |
2330 | ||
86679820 | 2331 | pagevec_init(&pvec); |
4e7ea81d | 2332 | while (start <= end) { |
2b85a617 | 2333 | nr_pages = pagevec_lookup_range(&pvec, inode->i_mapping, |
397162ff | 2334 | &start, end); |
4e7ea81d JK |
2335 | if (nr_pages == 0) |
2336 | break; | |
2337 | for (i = 0; i < nr_pages; i++) { | |
2338 | struct page *page = pvec.pages[i]; | |
2339 | ||
2943fdbc RH |
2340 | err = mpage_process_page(mpd, page, &lblk, &pblock, |
2341 | &map_bh); | |
4e7ea81d | 2342 | /* |
2943fdbc RH |
2343 | * If map_bh is true, means page may require further bh |
2344 | * mapping, or maybe the page was submitted for IO. | |
2345 | * So we return to call further extent mapping. | |
4e7ea81d | 2346 | */ |
39c0ae16 | 2347 | if (err < 0 || map_bh) |
2943fdbc | 2348 | goto out; |
4e7ea81d JK |
2349 | /* Page fully mapped - let IO run! */ |
2350 | err = mpage_submit_page(mpd, page); | |
2943fdbc RH |
2351 | if (err < 0) |
2352 | goto out; | |
4e7ea81d JK |
2353 | } |
2354 | pagevec_release(&pvec); | |
2355 | } | |
2356 | /* Extent fully mapped and matches with page boundary. We are done. */ | |
2357 | mpd->map.m_len = 0; | |
2358 | mpd->map.m_flags = 0; | |
2359 | return 0; | |
2943fdbc RH |
2360 | out: |
2361 | pagevec_release(&pvec); | |
2362 | return err; | |
4e7ea81d JK |
2363 | } |
2364 | ||
2365 | static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd) | |
2366 | { | |
2367 | struct inode *inode = mpd->inode; | |
2368 | struct ext4_map_blocks *map = &mpd->map; | |
2369 | int get_blocks_flags; | |
090f32ee | 2370 | int err, dioread_nolock; |
4e7ea81d JK |
2371 | |
2372 | trace_ext4_da_write_pages_extent(inode, map); | |
2373 | /* | |
2374 | * Call ext4_map_blocks() to allocate any delayed allocation blocks, or | |
556615dc | 2375 | * to convert an unwritten extent to be initialized (in the case |
4e7ea81d JK |
2376 | * where we have written into one or more preallocated blocks). It is |
2377 | * possible that we're going to need more metadata blocks than | |
2378 | * previously reserved. However we must not fail because we're in | |
2379 | * writeback and there is nothing we can do about it so it might result | |
2380 | * in data loss. So use reserved blocks to allocate metadata if | |
2381 | * possible. | |
2382 | * | |
754cfed6 TT |
2383 | * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if |
2384 | * the blocks in question are delalloc blocks. This indicates | |
2385 | * that the blocks and quotas has already been checked when | |
2386 | * the data was copied into the page cache. | |
4e7ea81d JK |
2387 | */ |
2388 | get_blocks_flags = EXT4_GET_BLOCKS_CREATE | | |
ee0876bc JK |
2389 | EXT4_GET_BLOCKS_METADATA_NOFAIL | |
2390 | EXT4_GET_BLOCKS_IO_SUBMIT; | |
090f32ee LC |
2391 | dioread_nolock = ext4_should_dioread_nolock(inode); |
2392 | if (dioread_nolock) | |
4e7ea81d | 2393 | get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT; |
6db07461 | 2394 | if (map->m_flags & BIT(BH_Delay)) |
4e7ea81d JK |
2395 | get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE; |
2396 | ||
2397 | err = ext4_map_blocks(handle, inode, map, get_blocks_flags); | |
2398 | if (err < 0) | |
2399 | return err; | |
090f32ee | 2400 | if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) { |
6b523df4 JK |
2401 | if (!mpd->io_submit.io_end->handle && |
2402 | ext4_handle_valid(handle)) { | |
2403 | mpd->io_submit.io_end->handle = handle->h_rsv_handle; | |
2404 | handle->h_rsv_handle = NULL; | |
2405 | } | |
3613d228 | 2406 | ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end); |
6b523df4 | 2407 | } |
4e7ea81d JK |
2408 | |
2409 | BUG_ON(map->m_len == 0); | |
4e7ea81d JK |
2410 | return 0; |
2411 | } | |
2412 | ||
2413 | /* | |
2414 | * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length | |
2415 | * mpd->len and submit pages underlying it for IO | |
2416 | * | |
2417 | * @handle - handle for journal operations | |
2418 | * @mpd - extent to map | |
7534e854 JK |
2419 | * @give_up_on_write - we set this to true iff there is a fatal error and there |
2420 | * is no hope of writing the data. The caller should discard | |
2421 | * dirty pages to avoid infinite loops. | |
4e7ea81d JK |
2422 | * |
2423 | * The function maps extent starting at mpd->lblk of length mpd->len. If it is | |
2424 | * delayed, blocks are allocated, if it is unwritten, we may need to convert | |
2425 | * them to initialized or split the described range from larger unwritten | |
2426 | * extent. Note that we need not map all the described range since allocation | |
2427 | * can return less blocks or the range is covered by more unwritten extents. We | |
2428 | * cannot map more because we are limited by reserved transaction credits. On | |
2429 | * the other hand we always make sure that the last touched page is fully | |
2430 | * mapped so that it can be written out (and thus forward progress is | |
2431 | * guaranteed). After mapping we submit all mapped pages for IO. | |
2432 | */ | |
2433 | static int mpage_map_and_submit_extent(handle_t *handle, | |
cb530541 TT |
2434 | struct mpage_da_data *mpd, |
2435 | bool *give_up_on_write) | |
4e7ea81d JK |
2436 | { |
2437 | struct inode *inode = mpd->inode; | |
2438 | struct ext4_map_blocks *map = &mpd->map; | |
2439 | int err; | |
2440 | loff_t disksize; | |
6603120e | 2441 | int progress = 0; |
c8cc8816 | 2442 | ext4_io_end_t *io_end = mpd->io_submit.io_end; |
4d06bfb9 | 2443 | struct ext4_io_end_vec *io_end_vec; |
4e7ea81d | 2444 | |
4d06bfb9 RH |
2445 | io_end_vec = ext4_alloc_io_end_vec(io_end); |
2446 | if (IS_ERR(io_end_vec)) | |
2447 | return PTR_ERR(io_end_vec); | |
c8cc8816 | 2448 | io_end_vec->offset = ((loff_t)map->m_lblk) << inode->i_blkbits; |
27d7c4ed | 2449 | do { |
4e7ea81d JK |
2450 | err = mpage_map_one_extent(handle, mpd); |
2451 | if (err < 0) { | |
2452 | struct super_block *sb = inode->i_sb; | |
2453 | ||
0db1ff22 | 2454 | if (ext4_forced_shutdown(EXT4_SB(sb)) || |
9b5f6c9b | 2455 | ext4_test_mount_flag(sb, EXT4_MF_FS_ABORTED)) |
cb530541 | 2456 | goto invalidate_dirty_pages; |
4e7ea81d | 2457 | /* |
cb530541 TT |
2458 | * Let the uper layers retry transient errors. |
2459 | * In the case of ENOSPC, if ext4_count_free_blocks() | |
2460 | * is non-zero, a commit should free up blocks. | |
4e7ea81d | 2461 | */ |
cb530541 | 2462 | if ((err == -ENOMEM) || |
6603120e DM |
2463 | (err == -ENOSPC && ext4_count_free_clusters(sb))) { |
2464 | if (progress) | |
2465 | goto update_disksize; | |
cb530541 | 2466 | return err; |
6603120e | 2467 | } |
cb530541 TT |
2468 | ext4_msg(sb, KERN_CRIT, |
2469 | "Delayed block allocation failed for " | |
2470 | "inode %lu at logical offset %llu with" | |
2471 | " max blocks %u with error %d", | |
2472 | inode->i_ino, | |
2473 | (unsigned long long)map->m_lblk, | |
2474 | (unsigned)map->m_len, -err); | |
2475 | ext4_msg(sb, KERN_CRIT, | |
2476 | "This should not happen!! Data will " | |
2477 | "be lost\n"); | |
2478 | if (err == -ENOSPC) | |
2479 | ext4_print_free_blocks(inode); | |
2480 | invalidate_dirty_pages: | |
2481 | *give_up_on_write = true; | |
4e7ea81d JK |
2482 | return err; |
2483 | } | |
6603120e | 2484 | progress = 1; |
4e7ea81d JK |
2485 | /* |
2486 | * Update buffer state, submit mapped pages, and get us new | |
2487 | * extent to map | |
2488 | */ | |
2489 | err = mpage_map_and_submit_buffers(mpd); | |
2490 | if (err < 0) | |
6603120e | 2491 | goto update_disksize; |
27d7c4ed | 2492 | } while (map->m_len); |
4e7ea81d | 2493 | |
6603120e | 2494 | update_disksize: |
622cad13 TT |
2495 | /* |
2496 | * Update on-disk size after IO is submitted. Races with | |
2497 | * truncate are avoided by checking i_size under i_data_sem. | |
2498 | */ | |
09cbfeaf | 2499 | disksize = ((loff_t)mpd->first_page) << PAGE_SHIFT; |
35df4299 | 2500 | if (disksize > READ_ONCE(EXT4_I(inode)->i_disksize)) { |
4e7ea81d | 2501 | int err2; |
622cad13 TT |
2502 | loff_t i_size; |
2503 | ||
2504 | down_write(&EXT4_I(inode)->i_data_sem); | |
2505 | i_size = i_size_read(inode); | |
2506 | if (disksize > i_size) | |
2507 | disksize = i_size; | |
2508 | if (disksize > EXT4_I(inode)->i_disksize) | |
2509 | EXT4_I(inode)->i_disksize = disksize; | |
622cad13 | 2510 | up_write(&EXT4_I(inode)->i_data_sem); |
b907f2d5 | 2511 | err2 = ext4_mark_inode_dirty(handle, inode); |
878520ac | 2512 | if (err2) { |
54d3adbc TT |
2513 | ext4_error_err(inode->i_sb, -err2, |
2514 | "Failed to mark inode %lu dirty", | |
2515 | inode->i_ino); | |
878520ac | 2516 | } |
4e7ea81d JK |
2517 | if (!err) |
2518 | err = err2; | |
2519 | } | |
2520 | return err; | |
2521 | } | |
2522 | ||
fffb2739 JK |
2523 | /* |
2524 | * Calculate the total number of credits to reserve for one writepages | |
20970ba6 | 2525 | * iteration. This is called from ext4_writepages(). We map an extent of |
70261f56 | 2526 | * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping |
fffb2739 JK |
2527 | * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN + |
2528 | * bpp - 1 blocks in bpp different extents. | |
2529 | */ | |
525f4ed8 MC |
2530 | static int ext4_da_writepages_trans_blocks(struct inode *inode) |
2531 | { | |
fffb2739 | 2532 | int bpp = ext4_journal_blocks_per_page(inode); |
525f4ed8 | 2533 | |
fffb2739 JK |
2534 | return ext4_meta_trans_blocks(inode, |
2535 | MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp); | |
525f4ed8 | 2536 | } |
61628a3f | 2537 | |
8e48dcfb | 2538 | /* |
4e7ea81d JK |
2539 | * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages |
2540 | * and underlying extent to map | |
2541 | * | |
2542 | * @mpd - where to look for pages | |
2543 | * | |
2544 | * Walk dirty pages in the mapping. If they are fully mapped, submit them for | |
2545 | * IO immediately. When we find a page which isn't mapped we start accumulating | |
2546 | * extent of buffers underlying these pages that needs mapping (formed by | |
2547 | * either delayed or unwritten buffers). We also lock the pages containing | |
2548 | * these buffers. The extent found is returned in @mpd structure (starting at | |
2549 | * mpd->lblk with length mpd->len blocks). | |
2550 | * | |
2551 | * Note that this function can attach bios to one io_end structure which are | |
2552 | * neither logically nor physically contiguous. Although it may seem as an | |
2553 | * unnecessary complication, it is actually inevitable in blocksize < pagesize | |
2554 | * case as we need to track IO to all buffers underlying a page in one io_end. | |
8e48dcfb | 2555 | */ |
4e7ea81d | 2556 | static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd) |
8e48dcfb | 2557 | { |
4e7ea81d JK |
2558 | struct address_space *mapping = mpd->inode->i_mapping; |
2559 | struct pagevec pvec; | |
2560 | unsigned int nr_pages; | |
aeac589a | 2561 | long left = mpd->wbc->nr_to_write; |
4e7ea81d JK |
2562 | pgoff_t index = mpd->first_page; |
2563 | pgoff_t end = mpd->last_page; | |
10bbd235 | 2564 | xa_mark_t tag; |
4e7ea81d JK |
2565 | int i, err = 0; |
2566 | int blkbits = mpd->inode->i_blkbits; | |
2567 | ext4_lblk_t lblk; | |
2568 | struct buffer_head *head; | |
8e48dcfb | 2569 | |
4e7ea81d | 2570 | if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages) |
5b41d924 ES |
2571 | tag = PAGECACHE_TAG_TOWRITE; |
2572 | else | |
2573 | tag = PAGECACHE_TAG_DIRTY; | |
2574 | ||
86679820 | 2575 | pagevec_init(&pvec); |
4e7ea81d JK |
2576 | mpd->map.m_len = 0; |
2577 | mpd->next_page = index; | |
4f01b02c | 2578 | while (index <= end) { |
dc7f3e86 | 2579 | nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end, |
67fd707f | 2580 | tag); |
8e48dcfb | 2581 | if (nr_pages == 0) |
6b8ed620 | 2582 | break; |
8e48dcfb TT |
2583 | |
2584 | for (i = 0; i < nr_pages; i++) { | |
2585 | struct page *page = pvec.pages[i]; | |
2586 | ||
aeac589a ML |
2587 | /* |
2588 | * Accumulated enough dirty pages? This doesn't apply | |
2589 | * to WB_SYNC_ALL mode. For integrity sync we have to | |
2590 | * keep going because someone may be concurrently | |
2591 | * dirtying pages, and we might have synced a lot of | |
2592 | * newly appeared dirty pages, but have not synced all | |
2593 | * of the old dirty pages. | |
2594 | */ | |
2595 | if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0) | |
2596 | goto out; | |
2597 | ||
4e7ea81d JK |
2598 | /* If we can't merge this page, we are done. */ |
2599 | if (mpd->map.m_len > 0 && mpd->next_page != page->index) | |
2600 | goto out; | |
78aaced3 | 2601 | |
8e48dcfb | 2602 | lock_page(page); |
8e48dcfb | 2603 | /* |
4e7ea81d JK |
2604 | * If the page is no longer dirty, or its mapping no |
2605 | * longer corresponds to inode we are writing (which | |
2606 | * means it has been truncated or invalidated), or the | |
2607 | * page is already under writeback and we are not doing | |
2608 | * a data integrity writeback, skip the page | |
8e48dcfb | 2609 | */ |
4f01b02c TT |
2610 | if (!PageDirty(page) || |
2611 | (PageWriteback(page) && | |
4e7ea81d | 2612 | (mpd->wbc->sync_mode == WB_SYNC_NONE)) || |
4f01b02c | 2613 | unlikely(page->mapping != mapping)) { |
8e48dcfb TT |
2614 | unlock_page(page); |
2615 | continue; | |
2616 | } | |
2617 | ||
7cb1a535 | 2618 | wait_on_page_writeback(page); |
8e48dcfb | 2619 | BUG_ON(PageWriteback(page)); |
8e48dcfb | 2620 | |
4e7ea81d | 2621 | if (mpd->map.m_len == 0) |
8eb9e5ce | 2622 | mpd->first_page = page->index; |
8eb9e5ce | 2623 | mpd->next_page = page->index + 1; |
f8bec370 | 2624 | /* Add all dirty buffers to mpd */ |
4e7ea81d | 2625 | lblk = ((ext4_lblk_t)page->index) << |
09cbfeaf | 2626 | (PAGE_SHIFT - blkbits); |
f8bec370 | 2627 | head = page_buffers(page); |
5f1132b2 JK |
2628 | err = mpage_process_page_bufs(mpd, head, head, lblk); |
2629 | if (err <= 0) | |
4e7ea81d | 2630 | goto out; |
5f1132b2 | 2631 | err = 0; |
aeac589a | 2632 | left--; |
8e48dcfb TT |
2633 | } |
2634 | pagevec_release(&pvec); | |
2635 | cond_resched(); | |
2636 | } | |
6b8ed620 | 2637 | mpd->scanned_until_end = 1; |
4f01b02c | 2638 | return 0; |
8eb9e5ce TT |
2639 | out: |
2640 | pagevec_release(&pvec); | |
4e7ea81d | 2641 | return err; |
8e48dcfb TT |
2642 | } |
2643 | ||
20970ba6 TT |
2644 | static int ext4_writepages(struct address_space *mapping, |
2645 | struct writeback_control *wbc) | |
64769240 | 2646 | { |
4e7ea81d JK |
2647 | pgoff_t writeback_index = 0; |
2648 | long nr_to_write = wbc->nr_to_write; | |
22208ded | 2649 | int range_whole = 0; |
4e7ea81d | 2650 | int cycled = 1; |
61628a3f | 2651 | handle_t *handle = NULL; |
df22291f | 2652 | struct mpage_da_data mpd; |
5e745b04 | 2653 | struct inode *inode = mapping->host; |
6b523df4 | 2654 | int needed_blocks, rsv_blocks = 0, ret = 0; |
5e745b04 | 2655 | struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb); |
1bce63d1 | 2656 | struct blk_plug plug; |
cb530541 | 2657 | bool give_up_on_write = false; |
61628a3f | 2658 | |
0db1ff22 TT |
2659 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
2660 | return -EIO; | |
2661 | ||
bbd55937 | 2662 | percpu_down_read(&sbi->s_writepages_rwsem); |
20970ba6 | 2663 | trace_ext4_writepages(inode, wbc); |
ba80b101 | 2664 | |
61628a3f MC |
2665 | /* |
2666 | * No pages to write? This is mainly a kludge to avoid starting | |
2667 | * a transaction for special inodes like journal inode on last iput() | |
2668 | * because that could violate lock ordering on umount | |
2669 | */ | |
a1d6cc56 | 2670 | if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) |
bbf023c7 | 2671 | goto out_writepages; |
2a21e37e | 2672 | |
20970ba6 | 2673 | if (ext4_should_journal_data(inode)) { |
043d20d1 | 2674 | ret = generic_writepages(mapping, wbc); |
bbf023c7 | 2675 | goto out_writepages; |
20970ba6 TT |
2676 | } |
2677 | ||
2a21e37e TT |
2678 | /* |
2679 | * If the filesystem has aborted, it is read-only, so return | |
2680 | * right away instead of dumping stack traces later on that | |
2681 | * will obscure the real source of the problem. We test | |
1751e8a6 | 2682 | * EXT4_MF_FS_ABORTED instead of sb->s_flag's SB_RDONLY because |
2a21e37e | 2683 | * the latter could be true if the filesystem is mounted |
20970ba6 | 2684 | * read-only, and in that case, ext4_writepages should |
2a21e37e TT |
2685 | * *never* be called, so if that ever happens, we would want |
2686 | * the stack trace. | |
2687 | */ | |
0db1ff22 | 2688 | if (unlikely(ext4_forced_shutdown(EXT4_SB(mapping->host->i_sb)) || |
9b5f6c9b | 2689 | ext4_test_mount_flag(inode->i_sb, EXT4_MF_FS_ABORTED))) { |
bbf023c7 ML |
2690 | ret = -EROFS; |
2691 | goto out_writepages; | |
2692 | } | |
2a21e37e | 2693 | |
4e7ea81d JK |
2694 | /* |
2695 | * If we have inline data and arrive here, it means that | |
2696 | * we will soon create the block for the 1st page, so | |
2697 | * we'd better clear the inline data here. | |
2698 | */ | |
2699 | if (ext4_has_inline_data(inode)) { | |
2700 | /* Just inode will be modified... */ | |
2701 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 1); | |
2702 | if (IS_ERR(handle)) { | |
2703 | ret = PTR_ERR(handle); | |
2704 | goto out_writepages; | |
2705 | } | |
2706 | BUG_ON(ext4_test_inode_state(inode, | |
2707 | EXT4_STATE_MAY_INLINE_DATA)); | |
2708 | ext4_destroy_inline_data(handle, inode); | |
2709 | ext4_journal_stop(handle); | |
2710 | } | |
2711 | ||
4e343231 | 2712 | if (ext4_should_dioread_nolock(inode)) { |
2713 | /* | |
2714 | * We may need to convert up to one extent per block in | |
2715 | * the page and we may dirty the inode. | |
2716 | */ | |
2717 | rsv_blocks = 1 + ext4_chunk_trans_blocks(inode, | |
2718 | PAGE_SIZE >> inode->i_blkbits); | |
2719 | } | |
2720 | ||
22208ded AK |
2721 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
2722 | range_whole = 1; | |
61628a3f | 2723 | |
2acf2c26 | 2724 | if (wbc->range_cyclic) { |
4e7ea81d JK |
2725 | writeback_index = mapping->writeback_index; |
2726 | if (writeback_index) | |
2acf2c26 | 2727 | cycled = 0; |
4e7ea81d JK |
2728 | mpd.first_page = writeback_index; |
2729 | mpd.last_page = -1; | |
5b41d924 | 2730 | } else { |
09cbfeaf KS |
2731 | mpd.first_page = wbc->range_start >> PAGE_SHIFT; |
2732 | mpd.last_page = wbc->range_end >> PAGE_SHIFT; | |
5b41d924 | 2733 | } |
a1d6cc56 | 2734 | |
4e7ea81d JK |
2735 | mpd.inode = inode; |
2736 | mpd.wbc = wbc; | |
2737 | ext4_io_submit_init(&mpd.io_submit, wbc); | |
2acf2c26 | 2738 | retry: |
6e6938b6 | 2739 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
4e7ea81d | 2740 | tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page); |
1bce63d1 | 2741 | blk_start_plug(&plug); |
dddbd6ac JK |
2742 | |
2743 | /* | |
2744 | * First writeback pages that don't need mapping - we can avoid | |
2745 | * starting a transaction unnecessarily and also avoid being blocked | |
2746 | * in the block layer on device congestion while having transaction | |
2747 | * started. | |
2748 | */ | |
2749 | mpd.do_map = 0; | |
6b8ed620 | 2750 | mpd.scanned_until_end = 0; |
dddbd6ac JK |
2751 | mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL); |
2752 | if (!mpd.io_submit.io_end) { | |
2753 | ret = -ENOMEM; | |
2754 | goto unplug; | |
2755 | } | |
2756 | ret = mpage_prepare_extent_to_map(&mpd); | |
a297b2fc XW |
2757 | /* Unlock pages we didn't use */ |
2758 | mpage_release_unused_pages(&mpd, false); | |
dddbd6ac JK |
2759 | /* Submit prepared bio */ |
2760 | ext4_io_submit(&mpd.io_submit); | |
2761 | ext4_put_io_end_defer(mpd.io_submit.io_end); | |
2762 | mpd.io_submit.io_end = NULL; | |
dddbd6ac JK |
2763 | if (ret < 0) |
2764 | goto unplug; | |
2765 | ||
6b8ed620 | 2766 | while (!mpd.scanned_until_end && wbc->nr_to_write > 0) { |
4e7ea81d JK |
2767 | /* For each extent of pages we use new io_end */ |
2768 | mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL); | |
2769 | if (!mpd.io_submit.io_end) { | |
2770 | ret = -ENOMEM; | |
2771 | break; | |
2772 | } | |
a1d6cc56 AK |
2773 | |
2774 | /* | |
4e7ea81d JK |
2775 | * We have two constraints: We find one extent to map and we |
2776 | * must always write out whole page (makes a difference when | |
2777 | * blocksize < pagesize) so that we don't block on IO when we | |
2778 | * try to write out the rest of the page. Journalled mode is | |
2779 | * not supported by delalloc. | |
a1d6cc56 AK |
2780 | */ |
2781 | BUG_ON(ext4_should_journal_data(inode)); | |
525f4ed8 | 2782 | needed_blocks = ext4_da_writepages_trans_blocks(inode); |
a1d6cc56 | 2783 | |
4e7ea81d | 2784 | /* start a new transaction */ |
6b523df4 JK |
2785 | handle = ext4_journal_start_with_reserve(inode, |
2786 | EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks); | |
61628a3f MC |
2787 | if (IS_ERR(handle)) { |
2788 | ret = PTR_ERR(handle); | |
1693918e | 2789 | ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: " |
fbe845dd | 2790 | "%ld pages, ino %lu; err %d", __func__, |
a1d6cc56 | 2791 | wbc->nr_to_write, inode->i_ino, ret); |
4e7ea81d JK |
2792 | /* Release allocated io_end */ |
2793 | ext4_put_io_end(mpd.io_submit.io_end); | |
dddbd6ac | 2794 | mpd.io_submit.io_end = NULL; |
4e7ea81d | 2795 | break; |
61628a3f | 2796 | } |
dddbd6ac | 2797 | mpd.do_map = 1; |
f63e6005 | 2798 | |
4e7ea81d JK |
2799 | trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc); |
2800 | ret = mpage_prepare_extent_to_map(&mpd); | |
6b8ed620 JK |
2801 | if (!ret && mpd.map.m_len) |
2802 | ret = mpage_map_and_submit_extent(handle, &mpd, | |
cb530541 | 2803 | &give_up_on_write); |
646caa9c JK |
2804 | /* |
2805 | * Caution: If the handle is synchronous, | |
2806 | * ext4_journal_stop() can wait for transaction commit | |
2807 | * to finish which may depend on writeback of pages to | |
2808 | * complete or on page lock to be released. In that | |
b483bb77 | 2809 | * case, we have to wait until after we have |
646caa9c JK |
2810 | * submitted all the IO, released page locks we hold, |
2811 | * and dropped io_end reference (for extent conversion | |
2812 | * to be able to complete) before stopping the handle. | |
2813 | */ | |
2814 | if (!ext4_handle_valid(handle) || handle->h_sync == 0) { | |
2815 | ext4_journal_stop(handle); | |
2816 | handle = NULL; | |
dddbd6ac | 2817 | mpd.do_map = 0; |
646caa9c | 2818 | } |
4e7ea81d | 2819 | /* Unlock pages we didn't use */ |
cb530541 | 2820 | mpage_release_unused_pages(&mpd, give_up_on_write); |
a297b2fc XW |
2821 | /* Submit prepared bio */ |
2822 | ext4_io_submit(&mpd.io_submit); | |
2823 | ||
646caa9c JK |
2824 | /* |
2825 | * Drop our io_end reference we got from init. We have | |
2826 | * to be careful and use deferred io_end finishing if | |
2827 | * we are still holding the transaction as we can | |
2828 | * release the last reference to io_end which may end | |
2829 | * up doing unwritten extent conversion. | |
2830 | */ | |
2831 | if (handle) { | |
2832 | ext4_put_io_end_defer(mpd.io_submit.io_end); | |
2833 | ext4_journal_stop(handle); | |
2834 | } else | |
2835 | ext4_put_io_end(mpd.io_submit.io_end); | |
dddbd6ac | 2836 | mpd.io_submit.io_end = NULL; |
4e7ea81d JK |
2837 | |
2838 | if (ret == -ENOSPC && sbi->s_journal) { | |
2839 | /* | |
2840 | * Commit the transaction which would | |
22208ded AK |
2841 | * free blocks released in the transaction |
2842 | * and try again | |
2843 | */ | |
df22291f | 2844 | jbd2_journal_force_commit_nested(sbi->s_journal); |
22208ded | 2845 | ret = 0; |
4e7ea81d JK |
2846 | continue; |
2847 | } | |
2848 | /* Fatal error - ENOMEM, EIO... */ | |
2849 | if (ret) | |
61628a3f | 2850 | break; |
a1d6cc56 | 2851 | } |
dddbd6ac | 2852 | unplug: |
1bce63d1 | 2853 | blk_finish_plug(&plug); |
9c12a831 | 2854 | if (!ret && !cycled && wbc->nr_to_write > 0) { |
2acf2c26 | 2855 | cycled = 1; |
4e7ea81d JK |
2856 | mpd.last_page = writeback_index - 1; |
2857 | mpd.first_page = 0; | |
2acf2c26 AK |
2858 | goto retry; |
2859 | } | |
22208ded AK |
2860 | |
2861 | /* Update index */ | |
22208ded AK |
2862 | if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) |
2863 | /* | |
4e7ea81d | 2864 | * Set the writeback_index so that range_cyclic |
22208ded AK |
2865 | * mode will write it back later |
2866 | */ | |
4e7ea81d | 2867 | mapping->writeback_index = mpd.first_page; |
a1d6cc56 | 2868 | |
61628a3f | 2869 | out_writepages: |
20970ba6 TT |
2870 | trace_ext4_writepages_result(inode, wbc, ret, |
2871 | nr_to_write - wbc->nr_to_write); | |
bbd55937 | 2872 | percpu_up_read(&sbi->s_writepages_rwsem); |
61628a3f | 2873 | return ret; |
64769240 AT |
2874 | } |
2875 | ||
5f0663bb DW |
2876 | static int ext4_dax_writepages(struct address_space *mapping, |
2877 | struct writeback_control *wbc) | |
2878 | { | |
2879 | int ret; | |
2880 | long nr_to_write = wbc->nr_to_write; | |
2881 | struct inode *inode = mapping->host; | |
2882 | struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb); | |
2883 | ||
2884 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) | |
2885 | return -EIO; | |
2886 | ||
bbd55937 | 2887 | percpu_down_read(&sbi->s_writepages_rwsem); |
5f0663bb DW |
2888 | trace_ext4_writepages(inode, wbc); |
2889 | ||
3f666c56 | 2890 | ret = dax_writeback_mapping_range(mapping, sbi->s_daxdev, wbc); |
5f0663bb DW |
2891 | trace_ext4_writepages_result(inode, wbc, ret, |
2892 | nr_to_write - wbc->nr_to_write); | |
bbd55937 | 2893 | percpu_up_read(&sbi->s_writepages_rwsem); |
5f0663bb DW |
2894 | return ret; |
2895 | } | |
2896 | ||
79f0be8d AK |
2897 | static int ext4_nonda_switch(struct super_block *sb) |
2898 | { | |
5c1ff336 | 2899 | s64 free_clusters, dirty_clusters; |
79f0be8d AK |
2900 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
2901 | ||
2902 | /* | |
2903 | * switch to non delalloc mode if we are running low | |
2904 | * on free block. The free block accounting via percpu | |
179f7ebf | 2905 | * counters can get slightly wrong with percpu_counter_batch getting |
79f0be8d AK |
2906 | * accumulated on each CPU without updating global counters |
2907 | * Delalloc need an accurate free block accounting. So switch | |
2908 | * to non delalloc when we are near to error range. | |
2909 | */ | |
5c1ff336 EW |
2910 | free_clusters = |
2911 | percpu_counter_read_positive(&sbi->s_freeclusters_counter); | |
2912 | dirty_clusters = | |
2913 | percpu_counter_read_positive(&sbi->s_dirtyclusters_counter); | |
00d4e736 TT |
2914 | /* |
2915 | * Start pushing delalloc when 1/2 of free blocks are dirty. | |
2916 | */ | |
5c1ff336 | 2917 | if (dirty_clusters && (free_clusters < 2 * dirty_clusters)) |
10ee27a0 | 2918 | try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE); |
00d4e736 | 2919 | |
5c1ff336 EW |
2920 | if (2 * free_clusters < 3 * dirty_clusters || |
2921 | free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) { | |
79f0be8d | 2922 | /* |
c8afb446 ES |
2923 | * free block count is less than 150% of dirty blocks |
2924 | * or free blocks is less than watermark | |
79f0be8d AK |
2925 | */ |
2926 | return 1; | |
2927 | } | |
2928 | return 0; | |
2929 | } | |
2930 | ||
0ff8947f ES |
2931 | /* We always reserve for an inode update; the superblock could be there too */ |
2932 | static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len) | |
2933 | { | |
e2b911c5 | 2934 | if (likely(ext4_has_feature_large_file(inode->i_sb))) |
0ff8947f ES |
2935 | return 1; |
2936 | ||
2937 | if (pos + len <= 0x7fffffffULL) | |
2938 | return 1; | |
2939 | ||
2940 | /* We might need to update the superblock to set LARGE_FILE */ | |
2941 | return 2; | |
2942 | } | |
2943 | ||
64769240 | 2944 | static int ext4_da_write_begin(struct file *file, struct address_space *mapping, |
de9a55b8 TT |
2945 | loff_t pos, unsigned len, unsigned flags, |
2946 | struct page **pagep, void **fsdata) | |
64769240 | 2947 | { |
72b8ab9d | 2948 | int ret, retries = 0; |
64769240 AT |
2949 | struct page *page; |
2950 | pgoff_t index; | |
64769240 AT |
2951 | struct inode *inode = mapping->host; |
2952 | handle_t *handle; | |
2953 | ||
0db1ff22 TT |
2954 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
2955 | return -EIO; | |
2956 | ||
09cbfeaf | 2957 | index = pos >> PAGE_SHIFT; |
79f0be8d | 2958 | |
c93d8f88 EB |
2959 | if (ext4_nonda_switch(inode->i_sb) || S_ISLNK(inode->i_mode) || |
2960 | ext4_verity_in_progress(inode)) { | |
79f0be8d AK |
2961 | *fsdata = (void *)FALL_BACK_TO_NONDELALLOC; |
2962 | return ext4_write_begin(file, mapping, pos, | |
2963 | len, flags, pagep, fsdata); | |
2964 | } | |
2965 | *fsdata = (void *)0; | |
9bffad1e | 2966 | trace_ext4_da_write_begin(inode, pos, len, flags); |
9c3569b5 TM |
2967 | |
2968 | if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { | |
2969 | ret = ext4_da_write_inline_data_begin(mapping, inode, | |
2970 | pos, len, flags, | |
2971 | pagep, fsdata); | |
2972 | if (ret < 0) | |
47564bfb TT |
2973 | return ret; |
2974 | if (ret == 1) | |
2975 | return 0; | |
9c3569b5 TM |
2976 | } |
2977 | ||
47564bfb TT |
2978 | /* |
2979 | * grab_cache_page_write_begin() can take a long time if the | |
2980 | * system is thrashing due to memory pressure, or if the page | |
2981 | * is being written back. So grab it first before we start | |
2982 | * the transaction handle. This also allows us to allocate | |
2983 | * the page (if needed) without using GFP_NOFS. | |
2984 | */ | |
2985 | retry_grab: | |
2986 | page = grab_cache_page_write_begin(mapping, index, flags); | |
2987 | if (!page) | |
2988 | return -ENOMEM; | |
2989 | unlock_page(page); | |
2990 | ||
64769240 AT |
2991 | /* |
2992 | * With delayed allocation, we don't log the i_disksize update | |
2993 | * if there is delayed block allocation. But we still need | |
2994 | * to journalling the i_disksize update if writes to the end | |
2995 | * of file which has an already mapped buffer. | |
2996 | */ | |
47564bfb | 2997 | retry_journal: |
0ff8947f ES |
2998 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
2999 | ext4_da_write_credits(inode, pos, len)); | |
64769240 | 3000 | if (IS_ERR(handle)) { |
09cbfeaf | 3001 | put_page(page); |
47564bfb | 3002 | return PTR_ERR(handle); |
64769240 AT |
3003 | } |
3004 | ||
47564bfb TT |
3005 | lock_page(page); |
3006 | if (page->mapping != mapping) { | |
3007 | /* The page got truncated from under us */ | |
3008 | unlock_page(page); | |
09cbfeaf | 3009 | put_page(page); |
d5a0d4f7 | 3010 | ext4_journal_stop(handle); |
47564bfb | 3011 | goto retry_grab; |
d5a0d4f7 | 3012 | } |
47564bfb | 3013 | /* In case writeback began while the page was unlocked */ |
7afe5aa5 | 3014 | wait_for_stable_page(page); |
64769240 | 3015 | |
643fa961 | 3016 | #ifdef CONFIG_FS_ENCRYPTION |
2058f83a MH |
3017 | ret = ext4_block_write_begin(page, pos, len, |
3018 | ext4_da_get_block_prep); | |
3019 | #else | |
6e1db88d | 3020 | ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep); |
2058f83a | 3021 | #endif |
64769240 AT |
3022 | if (ret < 0) { |
3023 | unlock_page(page); | |
3024 | ext4_journal_stop(handle); | |
ae4d5372 AK |
3025 | /* |
3026 | * block_write_begin may have instantiated a few blocks | |
3027 | * outside i_size. Trim these off again. Don't need | |
3028 | * i_size_read because we hold i_mutex. | |
3029 | */ | |
3030 | if (pos + len > inode->i_size) | |
b9a4207d | 3031 | ext4_truncate_failed_write(inode); |
47564bfb TT |
3032 | |
3033 | if (ret == -ENOSPC && | |
3034 | ext4_should_retry_alloc(inode->i_sb, &retries)) | |
3035 | goto retry_journal; | |
3036 | ||
09cbfeaf | 3037 | put_page(page); |
47564bfb | 3038 | return ret; |
64769240 AT |
3039 | } |
3040 | ||
47564bfb | 3041 | *pagep = page; |
64769240 AT |
3042 | return ret; |
3043 | } | |
3044 | ||
632eaeab MC |
3045 | /* |
3046 | * Check if we should update i_disksize | |
3047 | * when write to the end of file but not require block allocation | |
3048 | */ | |
3049 | static int ext4_da_should_update_i_disksize(struct page *page, | |
de9a55b8 | 3050 | unsigned long offset) |
632eaeab MC |
3051 | { |
3052 | struct buffer_head *bh; | |
3053 | struct inode *inode = page->mapping->host; | |
3054 | unsigned int idx; | |
3055 | int i; | |
3056 | ||
3057 | bh = page_buffers(page); | |
3058 | idx = offset >> inode->i_blkbits; | |
3059 | ||
af5bc92d | 3060 | for (i = 0; i < idx; i++) |
632eaeab MC |
3061 | bh = bh->b_this_page; |
3062 | ||
29fa89d0 | 3063 | if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh)) |
632eaeab MC |
3064 | return 0; |
3065 | return 1; | |
3066 | } | |
3067 | ||
64769240 | 3068 | static int ext4_da_write_end(struct file *file, |
de9a55b8 TT |
3069 | struct address_space *mapping, |
3070 | loff_t pos, unsigned len, unsigned copied, | |
3071 | struct page *page, void *fsdata) | |
64769240 AT |
3072 | { |
3073 | struct inode *inode = mapping->host; | |
3074 | int ret = 0, ret2; | |
3075 | handle_t *handle = ext4_journal_current_handle(); | |
3076 | loff_t new_i_size; | |
632eaeab | 3077 | unsigned long start, end; |
79f0be8d AK |
3078 | int write_mode = (int)(unsigned long)fsdata; |
3079 | ||
74d553aa TT |
3080 | if (write_mode == FALL_BACK_TO_NONDELALLOC) |
3081 | return ext4_write_end(file, mapping, pos, | |
3082 | len, copied, page, fsdata); | |
632eaeab | 3083 | |
9bffad1e | 3084 | trace_ext4_da_write_end(inode, pos, len, copied); |
09cbfeaf | 3085 | start = pos & (PAGE_SIZE - 1); |
af5bc92d | 3086 | end = start + copied - 1; |
64769240 AT |
3087 | |
3088 | /* | |
3089 | * generic_write_end() will run mark_inode_dirty() if i_size | |
3090 | * changes. So let's piggyback the i_disksize mark_inode_dirty | |
3091 | * into that. | |
3092 | */ | |
64769240 | 3093 | new_i_size = pos + copied; |
ea51d132 | 3094 | if (copied && new_i_size > EXT4_I(inode)->i_disksize) { |
9c3569b5 TM |
3095 | if (ext4_has_inline_data(inode) || |
3096 | ext4_da_should_update_i_disksize(page, end)) { | |
ee124d27 | 3097 | ext4_update_i_disksize(inode, new_i_size); |
cf17fea6 AK |
3098 | /* We need to mark inode dirty even if |
3099 | * new_i_size is less that inode->i_size | |
3100 | * bu greater than i_disksize.(hint delalloc) | |
3101 | */ | |
4209ae12 | 3102 | ret = ext4_mark_inode_dirty(handle, inode); |
64769240 | 3103 | } |
632eaeab | 3104 | } |
9c3569b5 TM |
3105 | |
3106 | if (write_mode != CONVERT_INLINE_DATA && | |
3107 | ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) && | |
3108 | ext4_has_inline_data(inode)) | |
3109 | ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied, | |
3110 | page); | |
3111 | else | |
3112 | ret2 = generic_write_end(file, mapping, pos, len, copied, | |
64769240 | 3113 | page, fsdata); |
9c3569b5 | 3114 | |
64769240 AT |
3115 | copied = ret2; |
3116 | if (ret2 < 0) | |
3117 | ret = ret2; | |
3118 | ret2 = ext4_journal_stop(handle); | |
4209ae12 | 3119 | if (unlikely(ret2 && !ret)) |
64769240 AT |
3120 | ret = ret2; |
3121 | ||
3122 | return ret ? ret : copied; | |
3123 | } | |
3124 | ||
ccd2506b TT |
3125 | /* |
3126 | * Force all delayed allocation blocks to be allocated for a given inode. | |
3127 | */ | |
3128 | int ext4_alloc_da_blocks(struct inode *inode) | |
3129 | { | |
fb40ba0d TT |
3130 | trace_ext4_alloc_da_blocks(inode); |
3131 | ||
71d4f7d0 | 3132 | if (!EXT4_I(inode)->i_reserved_data_blocks) |
ccd2506b TT |
3133 | return 0; |
3134 | ||
3135 | /* | |
3136 | * We do something simple for now. The filemap_flush() will | |
3137 | * also start triggering a write of the data blocks, which is | |
3138 | * not strictly speaking necessary (and for users of | |
3139 | * laptop_mode, not even desirable). However, to do otherwise | |
3140 | * would require replicating code paths in: | |
de9a55b8 | 3141 | * |
20970ba6 | 3142 | * ext4_writepages() -> |
ccd2506b TT |
3143 | * write_cache_pages() ---> (via passed in callback function) |
3144 | * __mpage_da_writepage() --> | |
3145 | * mpage_add_bh_to_extent() | |
3146 | * mpage_da_map_blocks() | |
3147 | * | |
3148 | * The problem is that write_cache_pages(), located in | |
3149 | * mm/page-writeback.c, marks pages clean in preparation for | |
3150 | * doing I/O, which is not desirable if we're not planning on | |
3151 | * doing I/O at all. | |
3152 | * | |
3153 | * We could call write_cache_pages(), and then redirty all of | |
380cf090 | 3154 | * the pages by calling redirty_page_for_writepage() but that |
ccd2506b TT |
3155 | * would be ugly in the extreme. So instead we would need to |
3156 | * replicate parts of the code in the above functions, | |
25985edc | 3157 | * simplifying them because we wouldn't actually intend to |
ccd2506b TT |
3158 | * write out the pages, but rather only collect contiguous |
3159 | * logical block extents, call the multi-block allocator, and | |
3160 | * then update the buffer heads with the block allocations. | |
de9a55b8 | 3161 | * |
ccd2506b TT |
3162 | * For now, though, we'll cheat by calling filemap_flush(), |
3163 | * which will map the blocks, and start the I/O, but not | |
3164 | * actually wait for the I/O to complete. | |
3165 | */ | |
3166 | return filemap_flush(inode->i_mapping); | |
3167 | } | |
64769240 | 3168 | |
ac27a0ec DK |
3169 | /* |
3170 | * bmap() is special. It gets used by applications such as lilo and by | |
3171 | * the swapper to find the on-disk block of a specific piece of data. | |
3172 | * | |
3173 | * Naturally, this is dangerous if the block concerned is still in the | |
617ba13b | 3174 | * journal. If somebody makes a swapfile on an ext4 data-journaling |
ac27a0ec DK |
3175 | * filesystem and enables swap, then they may get a nasty shock when the |
3176 | * data getting swapped to that swapfile suddenly gets overwritten by | |
3177 | * the original zero's written out previously to the journal and | |
3178 | * awaiting writeback in the kernel's buffer cache. | |
3179 | * | |
3180 | * So, if we see any bmap calls here on a modified, data-journaled file, | |
3181 | * take extra steps to flush any blocks which might be in the cache. | |
3182 | */ | |
617ba13b | 3183 | static sector_t ext4_bmap(struct address_space *mapping, sector_t block) |
ac27a0ec DK |
3184 | { |
3185 | struct inode *inode = mapping->host; | |
3186 | journal_t *journal; | |
3187 | int err; | |
3188 | ||
46c7f254 TM |
3189 | /* |
3190 | * We can get here for an inline file via the FIBMAP ioctl | |
3191 | */ | |
3192 | if (ext4_has_inline_data(inode)) | |
3193 | return 0; | |
3194 | ||
64769240 AT |
3195 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) && |
3196 | test_opt(inode->i_sb, DELALLOC)) { | |
3197 | /* | |
3198 | * With delalloc we want to sync the file | |
3199 | * so that we can make sure we allocate | |
3200 | * blocks for file | |
3201 | */ | |
3202 | filemap_write_and_wait(mapping); | |
3203 | } | |
3204 | ||
19f5fb7a TT |
3205 | if (EXT4_JOURNAL(inode) && |
3206 | ext4_test_inode_state(inode, EXT4_STATE_JDATA)) { | |
ac27a0ec DK |
3207 | /* |
3208 | * This is a REALLY heavyweight approach, but the use of | |
3209 | * bmap on dirty files is expected to be extremely rare: | |
3210 | * only if we run lilo or swapon on a freshly made file | |
3211 | * do we expect this to happen. | |
3212 | * | |
3213 | * (bmap requires CAP_SYS_RAWIO so this does not | |
3214 | * represent an unprivileged user DOS attack --- we'd be | |
3215 | * in trouble if mortal users could trigger this path at | |
3216 | * will.) | |
3217 | * | |
617ba13b | 3218 | * NB. EXT4_STATE_JDATA is not set on files other than |
ac27a0ec DK |
3219 | * regular files. If somebody wants to bmap a directory |
3220 | * or symlink and gets confused because the buffer | |
3221 | * hasn't yet been flushed to disk, they deserve | |
3222 | * everything they get. | |
3223 | */ | |
3224 | ||
19f5fb7a | 3225 | ext4_clear_inode_state(inode, EXT4_STATE_JDATA); |
617ba13b | 3226 | journal = EXT4_JOURNAL(inode); |
dab291af MC |
3227 | jbd2_journal_lock_updates(journal); |
3228 | err = jbd2_journal_flush(journal); | |
3229 | jbd2_journal_unlock_updates(journal); | |
ac27a0ec DK |
3230 | |
3231 | if (err) | |
3232 | return 0; | |
3233 | } | |
3234 | ||
ac58e4fb | 3235 | return iomap_bmap(mapping, block, &ext4_iomap_ops); |
ac27a0ec DK |
3236 | } |
3237 | ||
617ba13b | 3238 | static int ext4_readpage(struct file *file, struct page *page) |
ac27a0ec | 3239 | { |
46c7f254 TM |
3240 | int ret = -EAGAIN; |
3241 | struct inode *inode = page->mapping->host; | |
3242 | ||
0562e0ba | 3243 | trace_ext4_readpage(page); |
46c7f254 TM |
3244 | |
3245 | if (ext4_has_inline_data(inode)) | |
3246 | ret = ext4_readpage_inline(inode, page); | |
3247 | ||
3248 | if (ret == -EAGAIN) | |
a07f624b | 3249 | return ext4_mpage_readpages(inode, NULL, page); |
46c7f254 TM |
3250 | |
3251 | return ret; | |
ac27a0ec DK |
3252 | } |
3253 | ||
6311f91f | 3254 | static void ext4_readahead(struct readahead_control *rac) |
ac27a0ec | 3255 | { |
6311f91f | 3256 | struct inode *inode = rac->mapping->host; |
46c7f254 | 3257 | |
6311f91f | 3258 | /* If the file has inline data, no need to do readahead. */ |
46c7f254 | 3259 | if (ext4_has_inline_data(inode)) |
6311f91f | 3260 | return; |
46c7f254 | 3261 | |
a07f624b | 3262 | ext4_mpage_readpages(inode, rac, NULL); |
ac27a0ec DK |
3263 | } |
3264 | ||
d47992f8 LC |
3265 | static void ext4_invalidatepage(struct page *page, unsigned int offset, |
3266 | unsigned int length) | |
ac27a0ec | 3267 | { |
ca99fdd2 | 3268 | trace_ext4_invalidatepage(page, offset, length); |
0562e0ba | 3269 | |
4520fb3c JK |
3270 | /* No journalling happens on data buffers when this function is used */ |
3271 | WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page))); | |
3272 | ||
ca99fdd2 | 3273 | block_invalidatepage(page, offset, length); |
4520fb3c JK |
3274 | } |
3275 | ||
53e87268 | 3276 | static int __ext4_journalled_invalidatepage(struct page *page, |
ca99fdd2 LC |
3277 | unsigned int offset, |
3278 | unsigned int length) | |
4520fb3c JK |
3279 | { |
3280 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); | |
3281 | ||
ca99fdd2 | 3282 | trace_ext4_journalled_invalidatepage(page, offset, length); |
4520fb3c | 3283 | |
ac27a0ec DK |
3284 | /* |
3285 | * If it's a full truncate we just forget about the pending dirtying | |
3286 | */ | |
09cbfeaf | 3287 | if (offset == 0 && length == PAGE_SIZE) |
ac27a0ec DK |
3288 | ClearPageChecked(page); |
3289 | ||
ca99fdd2 | 3290 | return jbd2_journal_invalidatepage(journal, page, offset, length); |
53e87268 JK |
3291 | } |
3292 | ||
3293 | /* Wrapper for aops... */ | |
3294 | static void ext4_journalled_invalidatepage(struct page *page, | |
d47992f8 LC |
3295 | unsigned int offset, |
3296 | unsigned int length) | |
53e87268 | 3297 | { |
ca99fdd2 | 3298 | WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0); |
ac27a0ec DK |
3299 | } |
3300 | ||
617ba13b | 3301 | static int ext4_releasepage(struct page *page, gfp_t wait) |
ac27a0ec | 3302 | { |
617ba13b | 3303 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); |
ac27a0ec | 3304 | |
0562e0ba JZ |
3305 | trace_ext4_releasepage(page); |
3306 | ||
e1c36595 JK |
3307 | /* Page has dirty journalled data -> cannot release */ |
3308 | if (PageChecked(page)) | |
ac27a0ec | 3309 | return 0; |
0390131b | 3310 | if (journal) |
529a781e | 3311 | return jbd2_journal_try_to_free_buffers(journal, page); |
0390131b FM |
3312 | else |
3313 | return try_to_free_buffers(page); | |
ac27a0ec DK |
3314 | } |
3315 | ||
b8a6176c JK |
3316 | static bool ext4_inode_datasync_dirty(struct inode *inode) |
3317 | { | |
3318 | journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; | |
3319 | ||
aa75f4d3 HS |
3320 | if (journal) { |
3321 | if (jbd2_transaction_committed(journal, | |
d0520df7 AR |
3322 | EXT4_I(inode)->i_datasync_tid)) |
3323 | return false; | |
3324 | if (test_opt2(inode->i_sb, JOURNAL_FAST_COMMIT)) | |
1ceecb53 | 3325 | return !list_empty(&EXT4_I(inode)->i_fc_list); |
d0520df7 | 3326 | return true; |
aa75f4d3 HS |
3327 | } |
3328 | ||
b8a6176c JK |
3329 | /* Any metadata buffers to write? */ |
3330 | if (!list_empty(&inode->i_mapping->private_list)) | |
3331 | return true; | |
3332 | return inode->i_state & I_DIRTY_DATASYNC; | |
3333 | } | |
3334 | ||
c8fdfe29 MB |
3335 | static void ext4_set_iomap(struct inode *inode, struct iomap *iomap, |
3336 | struct ext4_map_blocks *map, loff_t offset, | |
3337 | loff_t length) | |
364443cb | 3338 | { |
c8fdfe29 | 3339 | u8 blkbits = inode->i_blkbits; |
364443cb | 3340 | |
c8fdfe29 MB |
3341 | /* |
3342 | * Writes that span EOF might trigger an I/O size update on completion, | |
3343 | * so consider them to be dirty for the purpose of O_DSYNC, even if | |
3344 | * there is no other metadata changes being made or are pending. | |
3345 | */ | |
364443cb | 3346 | iomap->flags = 0; |
c8fdfe29 MB |
3347 | if (ext4_inode_datasync_dirty(inode) || |
3348 | offset + length > i_size_read(inode)) | |
b8a6176c | 3349 | iomap->flags |= IOMAP_F_DIRTY; |
c8fdfe29 MB |
3350 | |
3351 | if (map->m_flags & EXT4_MAP_NEW) | |
3352 | iomap->flags |= IOMAP_F_NEW; | |
3353 | ||
5e405595 | 3354 | iomap->bdev = inode->i_sb->s_bdev; |
c8fdfe29 MB |
3355 | iomap->dax_dev = EXT4_SB(inode->i_sb)->s_daxdev; |
3356 | iomap->offset = (u64) map->m_lblk << blkbits; | |
3357 | iomap->length = (u64) map->m_len << blkbits; | |
364443cb | 3358 | |
6386722a RH |
3359 | if ((map->m_flags & EXT4_MAP_MAPPED) && |
3360 | !ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
3361 | iomap->flags |= IOMAP_F_MERGED; | |
3362 | ||
c8fdfe29 MB |
3363 | /* |
3364 | * Flags passed to ext4_map_blocks() for direct I/O writes can result | |
3365 | * in m_flags having both EXT4_MAP_MAPPED and EXT4_MAP_UNWRITTEN bits | |
3366 | * set. In order for any allocated unwritten extents to be converted | |
3367 | * into written extents correctly within the ->end_io() handler, we | |
3368 | * need to ensure that the iomap->type is set appropriately. Hence, the | |
3369 | * reason why we need to check whether the EXT4_MAP_UNWRITTEN bit has | |
3370 | * been set first. | |
3371 | */ | |
3372 | if (map->m_flags & EXT4_MAP_UNWRITTEN) { | |
3373 | iomap->type = IOMAP_UNWRITTEN; | |
3374 | iomap->addr = (u64) map->m_pblk << blkbits; | |
3375 | } else if (map->m_flags & EXT4_MAP_MAPPED) { | |
3376 | iomap->type = IOMAP_MAPPED; | |
3377 | iomap->addr = (u64) map->m_pblk << blkbits; | |
364443cb | 3378 | } else { |
c8fdfe29 MB |
3379 | iomap->type = IOMAP_HOLE; |
3380 | iomap->addr = IOMAP_NULL_ADDR; | |
364443cb | 3381 | } |
364443cb JK |
3382 | } |
3383 | ||
f063db5e MB |
3384 | static int ext4_iomap_alloc(struct inode *inode, struct ext4_map_blocks *map, |
3385 | unsigned int flags) | |
776722e8 | 3386 | { |
776722e8 | 3387 | handle_t *handle; |
378f32ba MB |
3388 | u8 blkbits = inode->i_blkbits; |
3389 | int ret, dio_credits, m_flags = 0, retries = 0; | |
776722e8 | 3390 | |
776722e8 | 3391 | /* |
f063db5e MB |
3392 | * Trim the mapping request to the maximum value that we can map at |
3393 | * once for direct I/O. | |
776722e8 | 3394 | */ |
f063db5e MB |
3395 | if (map->m_len > DIO_MAX_BLOCKS) |
3396 | map->m_len = DIO_MAX_BLOCKS; | |
3397 | dio_credits = ext4_chunk_trans_blocks(inode, map->m_len); | |
776722e8 | 3398 | |
f063db5e | 3399 | retry: |
776722e8 | 3400 | /* |
f063db5e MB |
3401 | * Either we allocate blocks and then don't get an unwritten extent, so |
3402 | * in that case we have reserved enough credits. Or, the blocks are | |
3403 | * already allocated and unwritten. In that case, the extent conversion | |
3404 | * fits into the credits as well. | |
776722e8 | 3405 | */ |
f063db5e MB |
3406 | handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits); |
3407 | if (IS_ERR(handle)) | |
3408 | return PTR_ERR(handle); | |
4c0425ff | 3409 | |
378f32ba MB |
3410 | /* |
3411 | * DAX and direct I/O are the only two operations that are currently | |
3412 | * supported with IOMAP_WRITE. | |
3413 | */ | |
3414 | WARN_ON(!IS_DAX(inode) && !(flags & IOMAP_DIRECT)); | |
3415 | if (IS_DAX(inode)) | |
3416 | m_flags = EXT4_GET_BLOCKS_CREATE_ZERO; | |
3417 | /* | |
3418 | * We use i_size instead of i_disksize here because delalloc writeback | |
3419 | * can complete at any point during the I/O and subsequently push the | |
3420 | * i_disksize out to i_size. This could be beyond where direct I/O is | |
3421 | * happening and thus expose allocated blocks to direct I/O reads. | |
3422 | */ | |
3423 | else if ((map->m_lblk * (1 << blkbits)) >= i_size_read(inode)) | |
3424 | m_flags = EXT4_GET_BLOCKS_CREATE; | |
3425 | else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
3426 | m_flags = EXT4_GET_BLOCKS_IO_CREATE_EXT; | |
4b70df18 | 3427 | |
378f32ba | 3428 | ret = ext4_map_blocks(handle, inode, map, m_flags); |
8d5d02e6 | 3429 | |
74c66bcb | 3430 | /* |
378f32ba MB |
3431 | * We cannot fill holes in indirect tree based inodes as that could |
3432 | * expose stale data in the case of a crash. Use the magic error code | |
3433 | * to fallback to buffered I/O. | |
74c66bcb | 3434 | */ |
378f32ba MB |
3435 | if (!m_flags && !ret) |
3436 | ret = -ENOTBLK; | |
187372a3 | 3437 | |
f063db5e MB |
3438 | ext4_journal_stop(handle); |
3439 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) | |
3440 | goto retry; | |
3441 | ||
3442 | return ret; | |
4c0425ff | 3443 | } |
c7064ef1 | 3444 | |
f063db5e | 3445 | |
364443cb | 3446 | static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length, |
c039b997 | 3447 | unsigned flags, struct iomap *iomap, struct iomap *srcmap) |
4c0425ff | 3448 | { |
364443cb | 3449 | int ret; |
09edf4d3 MB |
3450 | struct ext4_map_blocks map; |
3451 | u8 blkbits = inode->i_blkbits; | |
729f52c6 | 3452 | |
bcd8e91f TT |
3453 | if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK) |
3454 | return -EINVAL; | |
4bd809db | 3455 | |
09edf4d3 MB |
3456 | if (WARN_ON_ONCE(ext4_has_inline_data(inode))) |
3457 | return -ERANGE; | |
4bd809db | 3458 | |
e8340395 | 3459 | /* |
09edf4d3 | 3460 | * Calculate the first and last logical blocks respectively. |
e8340395 | 3461 | */ |
09edf4d3 MB |
3462 | map.m_lblk = offset >> blkbits; |
3463 | map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits, | |
3464 | EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1; | |
e8340395 | 3465 | |
9faac62d RH |
3466 | if (flags & IOMAP_WRITE) { |
3467 | /* | |
3468 | * We check here if the blocks are already allocated, then we | |
3469 | * don't need to start a journal txn and we can directly return | |
3470 | * the mapping information. This could boost performance | |
3471 | * especially in multi-threaded overwrite requests. | |
3472 | */ | |
3473 | if (offset + length <= i_size_read(inode)) { | |
3474 | ret = ext4_map_blocks(NULL, inode, &map, 0); | |
3475 | if (ret > 0 && (map.m_flags & EXT4_MAP_MAPPED)) | |
3476 | goto out; | |
3477 | } | |
f063db5e | 3478 | ret = ext4_iomap_alloc(inode, &map, flags); |
9faac62d | 3479 | } else { |
545052e9 | 3480 | ret = ext4_map_blocks(NULL, inode, &map, 0); |
9faac62d | 3481 | } |
4bd809db | 3482 | |
f063db5e MB |
3483 | if (ret < 0) |
3484 | return ret; | |
9faac62d | 3485 | out: |
c8fdfe29 | 3486 | ext4_set_iomap(inode, iomap, &map, offset, length); |
4bd809db | 3487 | |
364443cb JK |
3488 | return 0; |
3489 | } | |
8d5d02e6 | 3490 | |
8cd115bd JK |
3491 | static int ext4_iomap_overwrite_begin(struct inode *inode, loff_t offset, |
3492 | loff_t length, unsigned flags, struct iomap *iomap, | |
3493 | struct iomap *srcmap) | |
3494 | { | |
3495 | int ret; | |
3496 | ||
3497 | /* | |
3498 | * Even for writes we don't need to allocate blocks, so just pretend | |
3499 | * we are reading to save overhead of starting a transaction. | |
3500 | */ | |
3501 | flags &= ~IOMAP_WRITE; | |
3502 | ret = ext4_iomap_begin(inode, offset, length, flags, iomap, srcmap); | |
3503 | WARN_ON_ONCE(iomap->type != IOMAP_MAPPED); | |
3504 | return ret; | |
3505 | } | |
3506 | ||
776722e8 JK |
3507 | static int ext4_iomap_end(struct inode *inode, loff_t offset, loff_t length, |
3508 | ssize_t written, unsigned flags, struct iomap *iomap) | |
3509 | { | |
69c499d1 | 3510 | /* |
378f32ba MB |
3511 | * Check to see whether an error occurred while writing out the data to |
3512 | * the allocated blocks. If so, return the magic error code so that we | |
3513 | * fallback to buffered I/O and attempt to complete the remainder of | |
3514 | * the I/O. Any blocks that may have been allocated in preparation for | |
3515 | * the direct I/O will be reused during buffered I/O. | |
69c499d1 | 3516 | */ |
378f32ba MB |
3517 | if (flags & (IOMAP_WRITE | IOMAP_DIRECT) && written == 0) |
3518 | return -ENOTBLK; | |
69c499d1 | 3519 | |
569342dc | 3520 | return 0; |
776722e8 | 3521 | } |
4bd809db | 3522 | |
8ff6daa1 | 3523 | const struct iomap_ops ext4_iomap_ops = { |
364443cb | 3524 | .iomap_begin = ext4_iomap_begin, |
776722e8 | 3525 | .iomap_end = ext4_iomap_end, |
364443cb | 3526 | }; |
8d5d02e6 | 3527 | |
8cd115bd JK |
3528 | const struct iomap_ops ext4_iomap_overwrite_ops = { |
3529 | .iomap_begin = ext4_iomap_overwrite_begin, | |
3530 | .iomap_end = ext4_iomap_end, | |
3531 | }; | |
3532 | ||
09edf4d3 MB |
3533 | static bool ext4_iomap_is_delalloc(struct inode *inode, |
3534 | struct ext4_map_blocks *map) | |
3535 | { | |
3536 | struct extent_status es; | |
3537 | ext4_lblk_t offset = 0, end = map->m_lblk + map->m_len - 1; | |
914f82a3 | 3538 | |
09edf4d3 MB |
3539 | ext4_es_find_extent_range(inode, &ext4_es_is_delayed, |
3540 | map->m_lblk, end, &es); | |
914f82a3 | 3541 | |
09edf4d3 MB |
3542 | if (!es.es_len || es.es_lblk > end) |
3543 | return false; | |
914f82a3 | 3544 | |
09edf4d3 MB |
3545 | if (es.es_lblk > map->m_lblk) { |
3546 | map->m_len = es.es_lblk - map->m_lblk; | |
3547 | return false; | |
914f82a3 | 3548 | } |
914f82a3 | 3549 | |
09edf4d3 MB |
3550 | offset = map->m_lblk - es.es_lblk; |
3551 | map->m_len = es.es_len - offset; | |
914f82a3 | 3552 | |
09edf4d3 | 3553 | return true; |
4c0425ff MC |
3554 | } |
3555 | ||
09edf4d3 MB |
3556 | static int ext4_iomap_begin_report(struct inode *inode, loff_t offset, |
3557 | loff_t length, unsigned int flags, | |
3558 | struct iomap *iomap, struct iomap *srcmap) | |
4c0425ff | 3559 | { |
09edf4d3 MB |
3560 | int ret; |
3561 | bool delalloc = false; | |
3562 | struct ext4_map_blocks map; | |
3563 | u8 blkbits = inode->i_blkbits; | |
4c0425ff | 3564 | |
09edf4d3 MB |
3565 | if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK) |
3566 | return -EINVAL; | |
3567 | ||
3568 | if (ext4_has_inline_data(inode)) { | |
3569 | ret = ext4_inline_data_iomap(inode, iomap); | |
3570 | if (ret != -EAGAIN) { | |
3571 | if (ret == 0 && offset >= iomap->length) | |
3572 | ret = -ENOENT; | |
3573 | return ret; | |
3574 | } | |
3575 | } | |
2058f83a | 3576 | |
84ebd795 | 3577 | /* |
09edf4d3 | 3578 | * Calculate the first and last logical block respectively. |
84ebd795 | 3579 | */ |
09edf4d3 MB |
3580 | map.m_lblk = offset >> blkbits; |
3581 | map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits, | |
3582 | EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1; | |
84ebd795 | 3583 | |
b2c57642 RH |
3584 | /* |
3585 | * Fiemap callers may call for offset beyond s_bitmap_maxbytes. | |
3586 | * So handle it here itself instead of querying ext4_map_blocks(). | |
3587 | * Since ext4_map_blocks() will warn about it and will return | |
3588 | * -EIO error. | |
3589 | */ | |
3590 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) { | |
3591 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
3592 | ||
3593 | if (offset >= sbi->s_bitmap_maxbytes) { | |
3594 | map.m_flags = 0; | |
3595 | goto set_iomap; | |
3596 | } | |
3597 | } | |
3598 | ||
09edf4d3 MB |
3599 | ret = ext4_map_blocks(NULL, inode, &map, 0); |
3600 | if (ret < 0) | |
3601 | return ret; | |
3602 | if (ret == 0) | |
3603 | delalloc = ext4_iomap_is_delalloc(inode, &map); | |
46c7f254 | 3604 | |
b2c57642 | 3605 | set_iomap: |
09edf4d3 MB |
3606 | ext4_set_iomap(inode, iomap, &map, offset, length); |
3607 | if (delalloc && iomap->type == IOMAP_HOLE) | |
3608 | iomap->type = IOMAP_DELALLOC; | |
3609 | ||
3610 | return 0; | |
4c0425ff MC |
3611 | } |
3612 | ||
09edf4d3 MB |
3613 | const struct iomap_ops ext4_iomap_report_ops = { |
3614 | .iomap_begin = ext4_iomap_begin_report, | |
3615 | }; | |
3616 | ||
ac27a0ec | 3617 | /* |
617ba13b | 3618 | * Pages can be marked dirty completely asynchronously from ext4's journalling |
ac27a0ec DK |
3619 | * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do |
3620 | * much here because ->set_page_dirty is called under VFS locks. The page is | |
3621 | * not necessarily locked. | |
3622 | * | |
3623 | * We cannot just dirty the page and leave attached buffers clean, because the | |
3624 | * buffers' dirty state is "definitive". We cannot just set the buffers dirty | |
3625 | * or jbddirty because all the journalling code will explode. | |
3626 | * | |
3627 | * So what we do is to mark the page "pending dirty" and next time writepage | |
3628 | * is called, propagate that into the buffers appropriately. | |
3629 | */ | |
617ba13b | 3630 | static int ext4_journalled_set_page_dirty(struct page *page) |
ac27a0ec DK |
3631 | { |
3632 | SetPageChecked(page); | |
3633 | return __set_page_dirty_nobuffers(page); | |
3634 | } | |
3635 | ||
6dcc693b JK |
3636 | static int ext4_set_page_dirty(struct page *page) |
3637 | { | |
3638 | WARN_ON_ONCE(!PageLocked(page) && !PageDirty(page)); | |
3639 | WARN_ON_ONCE(!page_has_buffers(page)); | |
3640 | return __set_page_dirty_buffers(page); | |
3641 | } | |
3642 | ||
0e6895ba RH |
3643 | static int ext4_iomap_swap_activate(struct swap_info_struct *sis, |
3644 | struct file *file, sector_t *span) | |
3645 | { | |
3646 | return iomap_swapfile_activate(sis, file, span, | |
3647 | &ext4_iomap_report_ops); | |
3648 | } | |
3649 | ||
74d553aa | 3650 | static const struct address_space_operations ext4_aops = { |
8ab22b9a | 3651 | .readpage = ext4_readpage, |
6311f91f | 3652 | .readahead = ext4_readahead, |
43ce1d23 | 3653 | .writepage = ext4_writepage, |
20970ba6 | 3654 | .writepages = ext4_writepages, |
8ab22b9a | 3655 | .write_begin = ext4_write_begin, |
74d553aa | 3656 | .write_end = ext4_write_end, |
6dcc693b | 3657 | .set_page_dirty = ext4_set_page_dirty, |
8ab22b9a HH |
3658 | .bmap = ext4_bmap, |
3659 | .invalidatepage = ext4_invalidatepage, | |
3660 | .releasepage = ext4_releasepage, | |
378f32ba | 3661 | .direct_IO = noop_direct_IO, |
8ab22b9a HH |
3662 | .migratepage = buffer_migrate_page, |
3663 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 3664 | .error_remove_page = generic_error_remove_page, |
0e6895ba | 3665 | .swap_activate = ext4_iomap_swap_activate, |
ac27a0ec DK |
3666 | }; |
3667 | ||
617ba13b | 3668 | static const struct address_space_operations ext4_journalled_aops = { |
8ab22b9a | 3669 | .readpage = ext4_readpage, |
6311f91f | 3670 | .readahead = ext4_readahead, |
43ce1d23 | 3671 | .writepage = ext4_writepage, |
20970ba6 | 3672 | .writepages = ext4_writepages, |
8ab22b9a HH |
3673 | .write_begin = ext4_write_begin, |
3674 | .write_end = ext4_journalled_write_end, | |
3675 | .set_page_dirty = ext4_journalled_set_page_dirty, | |
3676 | .bmap = ext4_bmap, | |
4520fb3c | 3677 | .invalidatepage = ext4_journalled_invalidatepage, |
8ab22b9a | 3678 | .releasepage = ext4_releasepage, |
378f32ba | 3679 | .direct_IO = noop_direct_IO, |
8ab22b9a | 3680 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 3681 | .error_remove_page = generic_error_remove_page, |
0e6895ba | 3682 | .swap_activate = ext4_iomap_swap_activate, |
ac27a0ec DK |
3683 | }; |
3684 | ||
64769240 | 3685 | static const struct address_space_operations ext4_da_aops = { |
8ab22b9a | 3686 | .readpage = ext4_readpage, |
6311f91f | 3687 | .readahead = ext4_readahead, |
43ce1d23 | 3688 | .writepage = ext4_writepage, |
20970ba6 | 3689 | .writepages = ext4_writepages, |
8ab22b9a HH |
3690 | .write_begin = ext4_da_write_begin, |
3691 | .write_end = ext4_da_write_end, | |
6dcc693b | 3692 | .set_page_dirty = ext4_set_page_dirty, |
8ab22b9a | 3693 | .bmap = ext4_bmap, |
8fcc3a58 | 3694 | .invalidatepage = ext4_invalidatepage, |
8ab22b9a | 3695 | .releasepage = ext4_releasepage, |
378f32ba | 3696 | .direct_IO = noop_direct_IO, |
8ab22b9a HH |
3697 | .migratepage = buffer_migrate_page, |
3698 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 3699 | .error_remove_page = generic_error_remove_page, |
0e6895ba | 3700 | .swap_activate = ext4_iomap_swap_activate, |
64769240 AT |
3701 | }; |
3702 | ||
5f0663bb DW |
3703 | static const struct address_space_operations ext4_dax_aops = { |
3704 | .writepages = ext4_dax_writepages, | |
3705 | .direct_IO = noop_direct_IO, | |
3706 | .set_page_dirty = noop_set_page_dirty, | |
94dbb631 | 3707 | .bmap = ext4_bmap, |
5f0663bb | 3708 | .invalidatepage = noop_invalidatepage, |
0e6895ba | 3709 | .swap_activate = ext4_iomap_swap_activate, |
5f0663bb DW |
3710 | }; |
3711 | ||
617ba13b | 3712 | void ext4_set_aops(struct inode *inode) |
ac27a0ec | 3713 | { |
3d2b1582 LC |
3714 | switch (ext4_inode_journal_mode(inode)) { |
3715 | case EXT4_INODE_ORDERED_DATA_MODE: | |
3d2b1582 | 3716 | case EXT4_INODE_WRITEBACK_DATA_MODE: |
3d2b1582 LC |
3717 | break; |
3718 | case EXT4_INODE_JOURNAL_DATA_MODE: | |
617ba13b | 3719 | inode->i_mapping->a_ops = &ext4_journalled_aops; |
74d553aa | 3720 | return; |
3d2b1582 LC |
3721 | default: |
3722 | BUG(); | |
3723 | } | |
5f0663bb DW |
3724 | if (IS_DAX(inode)) |
3725 | inode->i_mapping->a_ops = &ext4_dax_aops; | |
3726 | else if (test_opt(inode->i_sb, DELALLOC)) | |
74d553aa TT |
3727 | inode->i_mapping->a_ops = &ext4_da_aops; |
3728 | else | |
3729 | inode->i_mapping->a_ops = &ext4_aops; | |
ac27a0ec DK |
3730 | } |
3731 | ||
923ae0ff | 3732 | static int __ext4_block_zero_page_range(handle_t *handle, |
d863dc36 LC |
3733 | struct address_space *mapping, loff_t from, loff_t length) |
3734 | { | |
09cbfeaf KS |
3735 | ext4_fsblk_t index = from >> PAGE_SHIFT; |
3736 | unsigned offset = from & (PAGE_SIZE-1); | |
923ae0ff | 3737 | unsigned blocksize, pos; |
d863dc36 LC |
3738 | ext4_lblk_t iblock; |
3739 | struct inode *inode = mapping->host; | |
3740 | struct buffer_head *bh; | |
3741 | struct page *page; | |
3742 | int err = 0; | |
3743 | ||
09cbfeaf | 3744 | page = find_or_create_page(mapping, from >> PAGE_SHIFT, |
c62d2555 | 3745 | mapping_gfp_constraint(mapping, ~__GFP_FS)); |
d863dc36 LC |
3746 | if (!page) |
3747 | return -ENOMEM; | |
3748 | ||
3749 | blocksize = inode->i_sb->s_blocksize; | |
d863dc36 | 3750 | |
09cbfeaf | 3751 | iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits); |
d863dc36 LC |
3752 | |
3753 | if (!page_has_buffers(page)) | |
3754 | create_empty_buffers(page, blocksize, 0); | |
3755 | ||
3756 | /* Find the buffer that contains "offset" */ | |
3757 | bh = page_buffers(page); | |
3758 | pos = blocksize; | |
3759 | while (offset >= pos) { | |
3760 | bh = bh->b_this_page; | |
3761 | iblock++; | |
3762 | pos += blocksize; | |
3763 | } | |
d863dc36 LC |
3764 | if (buffer_freed(bh)) { |
3765 | BUFFER_TRACE(bh, "freed: skip"); | |
3766 | goto unlock; | |
3767 | } | |
d863dc36 LC |
3768 | if (!buffer_mapped(bh)) { |
3769 | BUFFER_TRACE(bh, "unmapped"); | |
3770 | ext4_get_block(inode, iblock, bh, 0); | |
3771 | /* unmapped? It's a hole - nothing to do */ | |
3772 | if (!buffer_mapped(bh)) { | |
3773 | BUFFER_TRACE(bh, "still unmapped"); | |
3774 | goto unlock; | |
3775 | } | |
3776 | } | |
3777 | ||
3778 | /* Ok, it's mapped. Make sure it's up-to-date */ | |
3779 | if (PageUptodate(page)) | |
3780 | set_buffer_uptodate(bh); | |
3781 | ||
3782 | if (!buffer_uptodate(bh)) { | |
2d069c08 | 3783 | err = ext4_read_bh_lock(bh, 0, true); |
3784 | if (err) | |
d863dc36 | 3785 | goto unlock; |
4f74d15f | 3786 | if (fscrypt_inode_uses_fs_layer_crypto(inode)) { |
c9c7429c | 3787 | /* We expect the key to be set. */ |
a7550b30 | 3788 | BUG_ON(!fscrypt_has_encryption_key(inode)); |
834f1565 EB |
3789 | err = fscrypt_decrypt_pagecache_blocks(page, blocksize, |
3790 | bh_offset(bh)); | |
3791 | if (err) { | |
3792 | clear_buffer_uptodate(bh); | |
3793 | goto unlock; | |
3794 | } | |
c9c7429c | 3795 | } |
d863dc36 | 3796 | } |
d863dc36 LC |
3797 | if (ext4_should_journal_data(inode)) { |
3798 | BUFFER_TRACE(bh, "get write access"); | |
3799 | err = ext4_journal_get_write_access(handle, bh); | |
3800 | if (err) | |
3801 | goto unlock; | |
3802 | } | |
d863dc36 | 3803 | zero_user(page, offset, length); |
d863dc36 LC |
3804 | BUFFER_TRACE(bh, "zeroed end of block"); |
3805 | ||
d863dc36 LC |
3806 | if (ext4_should_journal_data(inode)) { |
3807 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
0713ed0c | 3808 | } else { |
353eefd3 | 3809 | err = 0; |
d863dc36 | 3810 | mark_buffer_dirty(bh); |
3957ef53 | 3811 | if (ext4_should_order_data(inode)) |
73131fbb RZ |
3812 | err = ext4_jbd2_inode_add_write(handle, inode, from, |
3813 | length); | |
0713ed0c | 3814 | } |
d863dc36 LC |
3815 | |
3816 | unlock: | |
3817 | unlock_page(page); | |
09cbfeaf | 3818 | put_page(page); |
d863dc36 LC |
3819 | return err; |
3820 | } | |
3821 | ||
923ae0ff RZ |
3822 | /* |
3823 | * ext4_block_zero_page_range() zeros out a mapping of length 'length' | |
3824 | * starting from file offset 'from'. The range to be zero'd must | |
3825 | * be contained with in one block. If the specified range exceeds | |
3826 | * the end of the block it will be shortened to end of the block | |
3827 | * that cooresponds to 'from' | |
3828 | */ | |
3829 | static int ext4_block_zero_page_range(handle_t *handle, | |
3830 | struct address_space *mapping, loff_t from, loff_t length) | |
3831 | { | |
3832 | struct inode *inode = mapping->host; | |
09cbfeaf | 3833 | unsigned offset = from & (PAGE_SIZE-1); |
923ae0ff RZ |
3834 | unsigned blocksize = inode->i_sb->s_blocksize; |
3835 | unsigned max = blocksize - (offset & (blocksize - 1)); | |
3836 | ||
3837 | /* | |
3838 | * correct length if it does not fall between | |
3839 | * 'from' and the end of the block | |
3840 | */ | |
3841 | if (length > max || length < 0) | |
3842 | length = max; | |
3843 | ||
47e69351 JK |
3844 | if (IS_DAX(inode)) { |
3845 | return iomap_zero_range(inode, from, length, NULL, | |
3846 | &ext4_iomap_ops); | |
3847 | } | |
923ae0ff RZ |
3848 | return __ext4_block_zero_page_range(handle, mapping, from, length); |
3849 | } | |
3850 | ||
94350ab5 MW |
3851 | /* |
3852 | * ext4_block_truncate_page() zeroes out a mapping from file offset `from' | |
3853 | * up to the end of the block which corresponds to `from'. | |
3854 | * This required during truncate. We need to physically zero the tail end | |
3855 | * of that block so it doesn't yield old data if the file is later grown. | |
3856 | */ | |
c197855e | 3857 | static int ext4_block_truncate_page(handle_t *handle, |
94350ab5 MW |
3858 | struct address_space *mapping, loff_t from) |
3859 | { | |
09cbfeaf | 3860 | unsigned offset = from & (PAGE_SIZE-1); |
94350ab5 MW |
3861 | unsigned length; |
3862 | unsigned blocksize; | |
3863 | struct inode *inode = mapping->host; | |
3864 | ||
0d06863f | 3865 | /* If we are processing an encrypted inode during orphan list handling */ |
592ddec7 | 3866 | if (IS_ENCRYPTED(inode) && !fscrypt_has_encryption_key(inode)) |
0d06863f TT |
3867 | return 0; |
3868 | ||
94350ab5 MW |
3869 | blocksize = inode->i_sb->s_blocksize; |
3870 | length = blocksize - (offset & (blocksize - 1)); | |
3871 | ||
3872 | return ext4_block_zero_page_range(handle, mapping, from, length); | |
3873 | } | |
3874 | ||
a87dd18c LC |
3875 | int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode, |
3876 | loff_t lstart, loff_t length) | |
3877 | { | |
3878 | struct super_block *sb = inode->i_sb; | |
3879 | struct address_space *mapping = inode->i_mapping; | |
e1be3a92 | 3880 | unsigned partial_start, partial_end; |
a87dd18c LC |
3881 | ext4_fsblk_t start, end; |
3882 | loff_t byte_end = (lstart + length - 1); | |
3883 | int err = 0; | |
3884 | ||
e1be3a92 LC |
3885 | partial_start = lstart & (sb->s_blocksize - 1); |
3886 | partial_end = byte_end & (sb->s_blocksize - 1); | |
3887 | ||
a87dd18c LC |
3888 | start = lstart >> sb->s_blocksize_bits; |
3889 | end = byte_end >> sb->s_blocksize_bits; | |
3890 | ||
3891 | /* Handle partial zero within the single block */ | |
e1be3a92 LC |
3892 | if (start == end && |
3893 | (partial_start || (partial_end != sb->s_blocksize - 1))) { | |
a87dd18c LC |
3894 | err = ext4_block_zero_page_range(handle, mapping, |
3895 | lstart, length); | |
3896 | return err; | |
3897 | } | |
3898 | /* Handle partial zero out on the start of the range */ | |
e1be3a92 | 3899 | if (partial_start) { |
a87dd18c LC |
3900 | err = ext4_block_zero_page_range(handle, mapping, |
3901 | lstart, sb->s_blocksize); | |
3902 | if (err) | |
3903 | return err; | |
3904 | } | |
3905 | /* Handle partial zero out on the end of the range */ | |
e1be3a92 | 3906 | if (partial_end != sb->s_blocksize - 1) |
a87dd18c | 3907 | err = ext4_block_zero_page_range(handle, mapping, |
e1be3a92 LC |
3908 | byte_end - partial_end, |
3909 | partial_end + 1); | |
a87dd18c LC |
3910 | return err; |
3911 | } | |
3912 | ||
91ef4caf DG |
3913 | int ext4_can_truncate(struct inode *inode) |
3914 | { | |
91ef4caf DG |
3915 | if (S_ISREG(inode->i_mode)) |
3916 | return 1; | |
3917 | if (S_ISDIR(inode->i_mode)) | |
3918 | return 1; | |
3919 | if (S_ISLNK(inode->i_mode)) | |
3920 | return !ext4_inode_is_fast_symlink(inode); | |
3921 | return 0; | |
3922 | } | |
3923 | ||
01127848 JK |
3924 | /* |
3925 | * We have to make sure i_disksize gets properly updated before we truncate | |
3926 | * page cache due to hole punching or zero range. Otherwise i_disksize update | |
3927 | * can get lost as it may have been postponed to submission of writeback but | |
3928 | * that will never happen after we truncate page cache. | |
3929 | */ | |
3930 | int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset, | |
3931 | loff_t len) | |
3932 | { | |
3933 | handle_t *handle; | |
4209ae12 HS |
3934 | int ret; |
3935 | ||
01127848 JK |
3936 | loff_t size = i_size_read(inode); |
3937 | ||
5955102c | 3938 | WARN_ON(!inode_is_locked(inode)); |
01127848 JK |
3939 | if (offset > size || offset + len < size) |
3940 | return 0; | |
3941 | ||
3942 | if (EXT4_I(inode)->i_disksize >= size) | |
3943 | return 0; | |
3944 | ||
3945 | handle = ext4_journal_start(inode, EXT4_HT_MISC, 1); | |
3946 | if (IS_ERR(handle)) | |
3947 | return PTR_ERR(handle); | |
3948 | ext4_update_i_disksize(inode, size); | |
4209ae12 | 3949 | ret = ext4_mark_inode_dirty(handle, inode); |
01127848 JK |
3950 | ext4_journal_stop(handle); |
3951 | ||
4209ae12 | 3952 | return ret; |
01127848 JK |
3953 | } |
3954 | ||
b1f38217 | 3955 | static void ext4_wait_dax_page(struct ext4_inode_info *ei) |
430657b6 | 3956 | { |
430657b6 RZ |
3957 | up_write(&ei->i_mmap_sem); |
3958 | schedule(); | |
3959 | down_write(&ei->i_mmap_sem); | |
3960 | } | |
3961 | ||
3962 | int ext4_break_layouts(struct inode *inode) | |
3963 | { | |
3964 | struct ext4_inode_info *ei = EXT4_I(inode); | |
3965 | struct page *page; | |
430657b6 RZ |
3966 | int error; |
3967 | ||
3968 | if (WARN_ON_ONCE(!rwsem_is_locked(&ei->i_mmap_sem))) | |
3969 | return -EINVAL; | |
3970 | ||
3971 | do { | |
430657b6 RZ |
3972 | page = dax_layout_busy_page(inode->i_mapping); |
3973 | if (!page) | |
3974 | return 0; | |
3975 | ||
3976 | error = ___wait_var_event(&page->_refcount, | |
3977 | atomic_read(&page->_refcount) == 1, | |
3978 | TASK_INTERRUPTIBLE, 0, 0, | |
b1f38217 RZ |
3979 | ext4_wait_dax_page(ei)); |
3980 | } while (error == 0); | |
430657b6 RZ |
3981 | |
3982 | return error; | |
3983 | } | |
3984 | ||
a4bb6b64 | 3985 | /* |
cca32b7e | 3986 | * ext4_punch_hole: punches a hole in a file by releasing the blocks |
a4bb6b64 AH |
3987 | * associated with the given offset and length |
3988 | * | |
3989 | * @inode: File inode | |
3990 | * @offset: The offset where the hole will begin | |
3991 | * @len: The length of the hole | |
3992 | * | |
4907cb7b | 3993 | * Returns: 0 on success or negative on failure |
a4bb6b64 AH |
3994 | */ |
3995 | ||
aeb2817a | 3996 | int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length) |
a4bb6b64 | 3997 | { |
26a4c0c6 TT |
3998 | struct super_block *sb = inode->i_sb; |
3999 | ext4_lblk_t first_block, stop_block; | |
4000 | struct address_space *mapping = inode->i_mapping; | |
a87dd18c | 4001 | loff_t first_block_offset, last_block_offset; |
26a4c0c6 TT |
4002 | handle_t *handle; |
4003 | unsigned int credits; | |
4209ae12 | 4004 | int ret = 0, ret2 = 0; |
26a4c0c6 | 4005 | |
b8a86845 | 4006 | trace_ext4_punch_hole(inode, offset, length, 0); |
aaddea81 | 4007 | |
c1e8220b TT |
4008 | ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA); |
4009 | if (ext4_has_inline_data(inode)) { | |
4010 | down_write(&EXT4_I(inode)->i_mmap_sem); | |
4011 | ret = ext4_convert_inline_data(inode); | |
4012 | up_write(&EXT4_I(inode)->i_mmap_sem); | |
4013 | if (ret) | |
4014 | return ret; | |
4015 | } | |
4016 | ||
26a4c0c6 TT |
4017 | /* |
4018 | * Write out all dirty pages to avoid race conditions | |
4019 | * Then release them. | |
4020 | */ | |
cca32b7e | 4021 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { |
26a4c0c6 TT |
4022 | ret = filemap_write_and_wait_range(mapping, offset, |
4023 | offset + length - 1); | |
4024 | if (ret) | |
4025 | return ret; | |
4026 | } | |
4027 | ||
5955102c | 4028 | inode_lock(inode); |
9ef06cec | 4029 | |
26a4c0c6 TT |
4030 | /* No need to punch hole beyond i_size */ |
4031 | if (offset >= inode->i_size) | |
4032 | goto out_mutex; | |
4033 | ||
4034 | /* | |
4035 | * If the hole extends beyond i_size, set the hole | |
4036 | * to end after the page that contains i_size | |
4037 | */ | |
4038 | if (offset + length > inode->i_size) { | |
4039 | length = inode->i_size + | |
09cbfeaf | 4040 | PAGE_SIZE - (inode->i_size & (PAGE_SIZE - 1)) - |
26a4c0c6 TT |
4041 | offset; |
4042 | } | |
4043 | ||
a361293f JK |
4044 | if (offset & (sb->s_blocksize - 1) || |
4045 | (offset + length) & (sb->s_blocksize - 1)) { | |
4046 | /* | |
4047 | * Attach jinode to inode for jbd2 if we do any zeroing of | |
4048 | * partial block | |
4049 | */ | |
4050 | ret = ext4_inode_attach_jinode(inode); | |
4051 | if (ret < 0) | |
4052 | goto out_mutex; | |
4053 | ||
4054 | } | |
4055 | ||
ea3d7209 | 4056 | /* Wait all existing dio workers, newcomers will block on i_mutex */ |
ea3d7209 JK |
4057 | inode_dio_wait(inode); |
4058 | ||
4059 | /* | |
4060 | * Prevent page faults from reinstantiating pages we have released from | |
4061 | * page cache. | |
4062 | */ | |
4063 | down_write(&EXT4_I(inode)->i_mmap_sem); | |
430657b6 RZ |
4064 | |
4065 | ret = ext4_break_layouts(inode); | |
4066 | if (ret) | |
4067 | goto out_dio; | |
4068 | ||
a87dd18c LC |
4069 | first_block_offset = round_up(offset, sb->s_blocksize); |
4070 | last_block_offset = round_down((offset + length), sb->s_blocksize) - 1; | |
26a4c0c6 | 4071 | |
a87dd18c | 4072 | /* Now release the pages and zero block aligned part of pages*/ |
01127848 JK |
4073 | if (last_block_offset > first_block_offset) { |
4074 | ret = ext4_update_disksize_before_punch(inode, offset, length); | |
4075 | if (ret) | |
4076 | goto out_dio; | |
a87dd18c LC |
4077 | truncate_pagecache_range(inode, first_block_offset, |
4078 | last_block_offset); | |
01127848 | 4079 | } |
26a4c0c6 TT |
4080 | |
4081 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
4082 | credits = ext4_writepage_trans_blocks(inode); | |
4083 | else | |
4084 | credits = ext4_blocks_for_truncate(inode); | |
4085 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); | |
4086 | if (IS_ERR(handle)) { | |
4087 | ret = PTR_ERR(handle); | |
4088 | ext4_std_error(sb, ret); | |
4089 | goto out_dio; | |
4090 | } | |
4091 | ||
a87dd18c LC |
4092 | ret = ext4_zero_partial_blocks(handle, inode, offset, |
4093 | length); | |
4094 | if (ret) | |
4095 | goto out_stop; | |
26a4c0c6 TT |
4096 | |
4097 | first_block = (offset + sb->s_blocksize - 1) >> | |
4098 | EXT4_BLOCK_SIZE_BITS(sb); | |
4099 | stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb); | |
4100 | ||
eee597ac LC |
4101 | /* If there are blocks to remove, do it */ |
4102 | if (stop_block > first_block) { | |
26a4c0c6 | 4103 | |
eee597ac | 4104 | down_write(&EXT4_I(inode)->i_data_sem); |
27bc446e | 4105 | ext4_discard_preallocations(inode, 0); |
26a4c0c6 | 4106 | |
eee597ac LC |
4107 | ret = ext4_es_remove_extent(inode, first_block, |
4108 | stop_block - first_block); | |
4109 | if (ret) { | |
4110 | up_write(&EXT4_I(inode)->i_data_sem); | |
4111 | goto out_stop; | |
4112 | } | |
26a4c0c6 | 4113 | |
eee597ac LC |
4114 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
4115 | ret = ext4_ext_remove_space(inode, first_block, | |
4116 | stop_block - 1); | |
4117 | else | |
4118 | ret = ext4_ind_remove_space(handle, inode, first_block, | |
4119 | stop_block); | |
26a4c0c6 | 4120 | |
eee597ac LC |
4121 | up_write(&EXT4_I(inode)->i_data_sem); |
4122 | } | |
a80f7fcf | 4123 | ext4_fc_track_range(handle, inode, first_block, stop_block); |
26a4c0c6 TT |
4124 | if (IS_SYNC(inode)) |
4125 | ext4_handle_sync(handle); | |
e251f9bc | 4126 | |
eeca7ea1 | 4127 | inode->i_mtime = inode->i_ctime = current_time(inode); |
4209ae12 HS |
4128 | ret2 = ext4_mark_inode_dirty(handle, inode); |
4129 | if (unlikely(ret2)) | |
4130 | ret = ret2; | |
67a7d5f5 JK |
4131 | if (ret >= 0) |
4132 | ext4_update_inode_fsync_trans(handle, inode, 1); | |
26a4c0c6 TT |
4133 | out_stop: |
4134 | ext4_journal_stop(handle); | |
4135 | out_dio: | |
ea3d7209 | 4136 | up_write(&EXT4_I(inode)->i_mmap_sem); |
26a4c0c6 | 4137 | out_mutex: |
5955102c | 4138 | inode_unlock(inode); |
26a4c0c6 | 4139 | return ret; |
a4bb6b64 AH |
4140 | } |
4141 | ||
a361293f JK |
4142 | int ext4_inode_attach_jinode(struct inode *inode) |
4143 | { | |
4144 | struct ext4_inode_info *ei = EXT4_I(inode); | |
4145 | struct jbd2_inode *jinode; | |
4146 | ||
4147 | if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal) | |
4148 | return 0; | |
4149 | ||
4150 | jinode = jbd2_alloc_inode(GFP_KERNEL); | |
4151 | spin_lock(&inode->i_lock); | |
4152 | if (!ei->jinode) { | |
4153 | if (!jinode) { | |
4154 | spin_unlock(&inode->i_lock); | |
4155 | return -ENOMEM; | |
4156 | } | |
4157 | ei->jinode = jinode; | |
4158 | jbd2_journal_init_jbd_inode(ei->jinode, inode); | |
4159 | jinode = NULL; | |
4160 | } | |
4161 | spin_unlock(&inode->i_lock); | |
4162 | if (unlikely(jinode != NULL)) | |
4163 | jbd2_free_inode(jinode); | |
4164 | return 0; | |
4165 | } | |
4166 | ||
ac27a0ec | 4167 | /* |
617ba13b | 4168 | * ext4_truncate() |
ac27a0ec | 4169 | * |
617ba13b MC |
4170 | * We block out ext4_get_block() block instantiations across the entire |
4171 | * transaction, and VFS/VM ensures that ext4_truncate() cannot run | |
ac27a0ec DK |
4172 | * simultaneously on behalf of the same inode. |
4173 | * | |
42b2aa86 | 4174 | * As we work through the truncate and commit bits of it to the journal there |
ac27a0ec DK |
4175 | * is one core, guiding principle: the file's tree must always be consistent on |
4176 | * disk. We must be able to restart the truncate after a crash. | |
4177 | * | |
4178 | * The file's tree may be transiently inconsistent in memory (although it | |
4179 | * probably isn't), but whenever we close off and commit a journal transaction, | |
4180 | * the contents of (the filesystem + the journal) must be consistent and | |
4181 | * restartable. It's pretty simple, really: bottom up, right to left (although | |
4182 | * left-to-right works OK too). | |
4183 | * | |
4184 | * Note that at recovery time, journal replay occurs *before* the restart of | |
4185 | * truncate against the orphan inode list. | |
4186 | * | |
4187 | * The committed inode has the new, desired i_size (which is the same as | |
617ba13b | 4188 | * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see |
ac27a0ec | 4189 | * that this inode's truncate did not complete and it will again call |
617ba13b MC |
4190 | * ext4_truncate() to have another go. So there will be instantiated blocks |
4191 | * to the right of the truncation point in a crashed ext4 filesystem. But | |
ac27a0ec | 4192 | * that's fine - as long as they are linked from the inode, the post-crash |
617ba13b | 4193 | * ext4_truncate() run will find them and release them. |
ac27a0ec | 4194 | */ |
2c98eb5e | 4195 | int ext4_truncate(struct inode *inode) |
ac27a0ec | 4196 | { |
819c4920 TT |
4197 | struct ext4_inode_info *ei = EXT4_I(inode); |
4198 | unsigned int credits; | |
4209ae12 | 4199 | int err = 0, err2; |
819c4920 TT |
4200 | handle_t *handle; |
4201 | struct address_space *mapping = inode->i_mapping; | |
819c4920 | 4202 | |
19b5ef61 TT |
4203 | /* |
4204 | * There is a possibility that we're either freeing the inode | |
e04027e8 | 4205 | * or it's a completely new inode. In those cases we might not |
19b5ef61 TT |
4206 | * have i_mutex locked because it's not necessary. |
4207 | */ | |
4208 | if (!(inode->i_state & (I_NEW|I_FREEING))) | |
5955102c | 4209 | WARN_ON(!inode_is_locked(inode)); |
0562e0ba JZ |
4210 | trace_ext4_truncate_enter(inode); |
4211 | ||
91ef4caf | 4212 | if (!ext4_can_truncate(inode)) |
9a5d265f | 4213 | goto out_trace; |
ac27a0ec | 4214 | |
5534fb5b | 4215 | if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC)) |
19f5fb7a | 4216 | ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE); |
7d8f9f7d | 4217 | |
aef1c851 TM |
4218 | if (ext4_has_inline_data(inode)) { |
4219 | int has_inline = 1; | |
4220 | ||
01daf945 | 4221 | err = ext4_inline_data_truncate(inode, &has_inline); |
9a5d265f | 4222 | if (err || has_inline) |
4223 | goto out_trace; | |
aef1c851 TM |
4224 | } |
4225 | ||
a361293f JK |
4226 | /* If we zero-out tail of the page, we have to create jinode for jbd2 */ |
4227 | if (inode->i_size & (inode->i_sb->s_blocksize - 1)) { | |
4228 | if (ext4_inode_attach_jinode(inode) < 0) | |
9a5d265f | 4229 | goto out_trace; |
a361293f JK |
4230 | } |
4231 | ||
819c4920 TT |
4232 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
4233 | credits = ext4_writepage_trans_blocks(inode); | |
4234 | else | |
4235 | credits = ext4_blocks_for_truncate(inode); | |
4236 | ||
4237 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); | |
9a5d265f | 4238 | if (IS_ERR(handle)) { |
4239 | err = PTR_ERR(handle); | |
4240 | goto out_trace; | |
4241 | } | |
819c4920 | 4242 | |
eb3544c6 LC |
4243 | if (inode->i_size & (inode->i_sb->s_blocksize - 1)) |
4244 | ext4_block_truncate_page(handle, mapping, inode->i_size); | |
819c4920 TT |
4245 | |
4246 | /* | |
4247 | * We add the inode to the orphan list, so that if this | |
4248 | * truncate spans multiple transactions, and we crash, we will | |
4249 | * resume the truncate when the filesystem recovers. It also | |
4250 | * marks the inode dirty, to catch the new size. | |
4251 | * | |
4252 | * Implication: the file must always be in a sane, consistent | |
4253 | * truncatable state while each transaction commits. | |
4254 | */ | |
2c98eb5e TT |
4255 | err = ext4_orphan_add(handle, inode); |
4256 | if (err) | |
819c4920 TT |
4257 | goto out_stop; |
4258 | ||
4259 | down_write(&EXT4_I(inode)->i_data_sem); | |
4260 | ||
27bc446e | 4261 | ext4_discard_preallocations(inode, 0); |
819c4920 | 4262 | |
ff9893dc | 4263 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
d0abb36d | 4264 | err = ext4_ext_truncate(handle, inode); |
ff9893dc | 4265 | else |
819c4920 TT |
4266 | ext4_ind_truncate(handle, inode); |
4267 | ||
4268 | up_write(&ei->i_data_sem); | |
d0abb36d TT |
4269 | if (err) |
4270 | goto out_stop; | |
819c4920 TT |
4271 | |
4272 | if (IS_SYNC(inode)) | |
4273 | ext4_handle_sync(handle); | |
4274 | ||
4275 | out_stop: | |
4276 | /* | |
4277 | * If this was a simple ftruncate() and the file will remain alive, | |
4278 | * then we need to clear up the orphan record which we created above. | |
4279 | * However, if this was a real unlink then we were called by | |
58d86a50 | 4280 | * ext4_evict_inode(), and we allow that function to clean up the |
819c4920 TT |
4281 | * orphan info for us. |
4282 | */ | |
4283 | if (inode->i_nlink) | |
4284 | ext4_orphan_del(handle, inode); | |
4285 | ||
eeca7ea1 | 4286 | inode->i_mtime = inode->i_ctime = current_time(inode); |
4209ae12 HS |
4287 | err2 = ext4_mark_inode_dirty(handle, inode); |
4288 | if (unlikely(err2 && !err)) | |
4289 | err = err2; | |
819c4920 | 4290 | ext4_journal_stop(handle); |
ac27a0ec | 4291 | |
9a5d265f | 4292 | out_trace: |
0562e0ba | 4293 | trace_ext4_truncate_exit(inode); |
2c98eb5e | 4294 | return err; |
ac27a0ec DK |
4295 | } |
4296 | ||
ac27a0ec | 4297 | /* |
617ba13b | 4298 | * ext4_get_inode_loc returns with an extra refcount against the inode's |
ac27a0ec DK |
4299 | * underlying buffer_head on success. If 'in_mem' is true, we have all |
4300 | * data in memory that is needed to recreate the on-disk version of this | |
4301 | * inode. | |
4302 | */ | |
8016e29f HS |
4303 | static int __ext4_get_inode_loc(struct super_block *sb, unsigned long ino, |
4304 | struct ext4_iloc *iloc, int in_mem, | |
4305 | ext4_fsblk_t *ret_block) | |
ac27a0ec | 4306 | { |
240799cd TT |
4307 | struct ext4_group_desc *gdp; |
4308 | struct buffer_head *bh; | |
240799cd | 4309 | ext4_fsblk_t block; |
02f03c42 | 4310 | struct blk_plug plug; |
240799cd TT |
4311 | int inodes_per_block, inode_offset; |
4312 | ||
3a06d778 | 4313 | iloc->bh = NULL; |
8016e29f HS |
4314 | if (ino < EXT4_ROOT_INO || |
4315 | ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count)) | |
6a797d27 | 4316 | return -EFSCORRUPTED; |
ac27a0ec | 4317 | |
8016e29f | 4318 | iloc->block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb); |
240799cd TT |
4319 | gdp = ext4_get_group_desc(sb, iloc->block_group, NULL); |
4320 | if (!gdp) | |
ac27a0ec DK |
4321 | return -EIO; |
4322 | ||
240799cd TT |
4323 | /* |
4324 | * Figure out the offset within the block group inode table | |
4325 | */ | |
00d09882 | 4326 | inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; |
8016e29f | 4327 | inode_offset = ((ino - 1) % |
240799cd TT |
4328 | EXT4_INODES_PER_GROUP(sb)); |
4329 | block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block); | |
4330 | iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb); | |
4331 | ||
4332 | bh = sb_getblk(sb, block); | |
aebf0243 | 4333 | if (unlikely(!bh)) |
860d21e2 | 4334 | return -ENOMEM; |
46f870d6 TT |
4335 | if (ext4_simulate_fail(sb, EXT4_SIM_INODE_EIO)) |
4336 | goto simulate_eio; | |
ac27a0ec DK |
4337 | if (!buffer_uptodate(bh)) { |
4338 | lock_buffer(bh); | |
9c83a923 | 4339 | |
60c776e5 | 4340 | if (ext4_buffer_uptodate(bh)) { |
ac27a0ec DK |
4341 | /* someone brought it uptodate while we waited */ |
4342 | unlock_buffer(bh); | |
4343 | goto has_buffer; | |
4344 | } | |
4345 | ||
4346 | /* | |
4347 | * If we have all information of the inode in memory and this | |
4348 | * is the only valid inode in the block, we need not read the | |
4349 | * block. | |
4350 | */ | |
4351 | if (in_mem) { | |
4352 | struct buffer_head *bitmap_bh; | |
240799cd | 4353 | int i, start; |
ac27a0ec | 4354 | |
240799cd | 4355 | start = inode_offset & ~(inodes_per_block - 1); |
ac27a0ec | 4356 | |
240799cd TT |
4357 | /* Is the inode bitmap in cache? */ |
4358 | bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp)); | |
aebf0243 | 4359 | if (unlikely(!bitmap_bh)) |
ac27a0ec DK |
4360 | goto make_io; |
4361 | ||
4362 | /* | |
4363 | * If the inode bitmap isn't in cache then the | |
4364 | * optimisation may end up performing two reads instead | |
4365 | * of one, so skip it. | |
4366 | */ | |
4367 | if (!buffer_uptodate(bitmap_bh)) { | |
4368 | brelse(bitmap_bh); | |
4369 | goto make_io; | |
4370 | } | |
240799cd | 4371 | for (i = start; i < start + inodes_per_block; i++) { |
ac27a0ec DK |
4372 | if (i == inode_offset) |
4373 | continue; | |
617ba13b | 4374 | if (ext4_test_bit(i, bitmap_bh->b_data)) |
ac27a0ec DK |
4375 | break; |
4376 | } | |
4377 | brelse(bitmap_bh); | |
240799cd | 4378 | if (i == start + inodes_per_block) { |
ac27a0ec DK |
4379 | /* all other inodes are free, so skip I/O */ |
4380 | memset(bh->b_data, 0, bh->b_size); | |
4381 | set_buffer_uptodate(bh); | |
4382 | unlock_buffer(bh); | |
4383 | goto has_buffer; | |
4384 | } | |
4385 | } | |
4386 | ||
4387 | make_io: | |
240799cd TT |
4388 | /* |
4389 | * If we need to do any I/O, try to pre-readahead extra | |
4390 | * blocks from the inode table. | |
4391 | */ | |
02f03c42 | 4392 | blk_start_plug(&plug); |
240799cd TT |
4393 | if (EXT4_SB(sb)->s_inode_readahead_blks) { |
4394 | ext4_fsblk_t b, end, table; | |
4395 | unsigned num; | |
0d606e2c | 4396 | __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks; |
240799cd TT |
4397 | |
4398 | table = ext4_inode_table(sb, gdp); | |
b713a5ec | 4399 | /* s_inode_readahead_blks is always a power of 2 */ |
0d606e2c | 4400 | b = block & ~((ext4_fsblk_t) ra_blks - 1); |
240799cd TT |
4401 | if (table > b) |
4402 | b = table; | |
0d606e2c | 4403 | end = b + ra_blks; |
240799cd | 4404 | num = EXT4_INODES_PER_GROUP(sb); |
feb0ab32 | 4405 | if (ext4_has_group_desc_csum(sb)) |
560671a0 | 4406 | num -= ext4_itable_unused_count(sb, gdp); |
240799cd TT |
4407 | table += num / inodes_per_block; |
4408 | if (end > table) | |
4409 | end = table; | |
4410 | while (b <= end) | |
5df1d412 | 4411 | ext4_sb_breadahead_unmovable(sb, b++); |
240799cd TT |
4412 | } |
4413 | ||
ac27a0ec DK |
4414 | /* |
4415 | * There are other valid inodes in the buffer, this inode | |
4416 | * has in-inode xattrs, or we don't have this inode in memory. | |
4417 | * Read the block from disk. | |
4418 | */ | |
8016e29f | 4419 | trace_ext4_load_inode(sb, ino); |
2d069c08 | 4420 | ext4_read_bh_nowait(bh, REQ_META | REQ_PRIO, NULL); |
02f03c42 | 4421 | blk_finish_plug(&plug); |
ac27a0ec DK |
4422 | wait_on_buffer(bh); |
4423 | if (!buffer_uptodate(bh)) { | |
46f870d6 | 4424 | simulate_eio: |
8016e29f HS |
4425 | if (ret_block) |
4426 | *ret_block = block; | |
ac27a0ec DK |
4427 | brelse(bh); |
4428 | return -EIO; | |
4429 | } | |
4430 | } | |
4431 | has_buffer: | |
4432 | iloc->bh = bh; | |
4433 | return 0; | |
4434 | } | |
4435 | ||
8016e29f HS |
4436 | static int __ext4_get_inode_loc_noinmem(struct inode *inode, |
4437 | struct ext4_iloc *iloc) | |
4438 | { | |
4439 | ext4_fsblk_t err_blk; | |
4440 | int ret; | |
4441 | ||
4442 | ret = __ext4_get_inode_loc(inode->i_sb, inode->i_ino, iloc, 0, | |
4443 | &err_blk); | |
4444 | ||
4445 | if (ret == -EIO) | |
4446 | ext4_error_inode_block(inode, err_blk, EIO, | |
4447 | "unable to read itable block"); | |
4448 | ||
4449 | return ret; | |
4450 | } | |
4451 | ||
617ba13b | 4452 | int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec | 4453 | { |
8016e29f HS |
4454 | ext4_fsblk_t err_blk; |
4455 | int ret; | |
4456 | ||
ac27a0ec | 4457 | /* We have all inode data except xattrs in memory here. */ |
8016e29f HS |
4458 | ret = __ext4_get_inode_loc(inode->i_sb, inode->i_ino, iloc, |
4459 | !ext4_test_inode_state(inode, EXT4_STATE_XATTR), &err_blk); | |
4460 | ||
4461 | if (ret == -EIO) | |
4462 | ext4_error_inode_block(inode, err_blk, EIO, | |
4463 | "unable to read itable block"); | |
4464 | ||
4465 | return ret; | |
4466 | } | |
4467 | ||
4468 | ||
4469 | int ext4_get_fc_inode_loc(struct super_block *sb, unsigned long ino, | |
4470 | struct ext4_iloc *iloc) | |
4471 | { | |
4472 | return __ext4_get_inode_loc(sb, ino, iloc, 0, NULL); | |
ac27a0ec DK |
4473 | } |
4474 | ||
a8ab6d38 | 4475 | static bool ext4_should_enable_dax(struct inode *inode) |
6642586b | 4476 | { |
a8ab6d38 IW |
4477 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
4478 | ||
9cb20f94 | 4479 | if (test_opt2(inode->i_sb, DAX_NEVER)) |
6642586b RZ |
4480 | return false; |
4481 | if (!S_ISREG(inode->i_mode)) | |
4482 | return false; | |
4483 | if (ext4_should_journal_data(inode)) | |
4484 | return false; | |
4485 | if (ext4_has_inline_data(inode)) | |
4486 | return false; | |
592ddec7 | 4487 | if (ext4_test_inode_flag(inode, EXT4_INODE_ENCRYPT)) |
6642586b | 4488 | return false; |
c93d8f88 EB |
4489 | if (ext4_test_inode_flag(inode, EXT4_INODE_VERITY)) |
4490 | return false; | |
a8ab6d38 IW |
4491 | if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX, &sbi->s_ext4_flags)) |
4492 | return false; | |
4493 | if (test_opt(inode->i_sb, DAX_ALWAYS)) | |
4494 | return true; | |
4495 | ||
b383a73f | 4496 | return ext4_test_inode_flag(inode, EXT4_INODE_DAX); |
6642586b RZ |
4497 | } |
4498 | ||
043546e4 | 4499 | void ext4_set_inode_flags(struct inode *inode, bool init) |
ac27a0ec | 4500 | { |
617ba13b | 4501 | unsigned int flags = EXT4_I(inode)->i_flags; |
00a1a053 | 4502 | unsigned int new_fl = 0; |
ac27a0ec | 4503 | |
043546e4 IW |
4504 | WARN_ON_ONCE(IS_DAX(inode) && init); |
4505 | ||
617ba13b | 4506 | if (flags & EXT4_SYNC_FL) |
00a1a053 | 4507 | new_fl |= S_SYNC; |
617ba13b | 4508 | if (flags & EXT4_APPEND_FL) |
00a1a053 | 4509 | new_fl |= S_APPEND; |
617ba13b | 4510 | if (flags & EXT4_IMMUTABLE_FL) |
00a1a053 | 4511 | new_fl |= S_IMMUTABLE; |
617ba13b | 4512 | if (flags & EXT4_NOATIME_FL) |
00a1a053 | 4513 | new_fl |= S_NOATIME; |
617ba13b | 4514 | if (flags & EXT4_DIRSYNC_FL) |
00a1a053 | 4515 | new_fl |= S_DIRSYNC; |
043546e4 IW |
4516 | |
4517 | /* Because of the way inode_set_flags() works we must preserve S_DAX | |
4518 | * here if already set. */ | |
4519 | new_fl |= (inode->i_flags & S_DAX); | |
4520 | if (init && ext4_should_enable_dax(inode)) | |
923ae0ff | 4521 | new_fl |= S_DAX; |
043546e4 | 4522 | |
2ee6a576 EB |
4523 | if (flags & EXT4_ENCRYPT_FL) |
4524 | new_fl |= S_ENCRYPTED; | |
b886ee3e GKB |
4525 | if (flags & EXT4_CASEFOLD_FL) |
4526 | new_fl |= S_CASEFOLD; | |
c93d8f88 EB |
4527 | if (flags & EXT4_VERITY_FL) |
4528 | new_fl |= S_VERITY; | |
5f16f322 | 4529 | inode_set_flags(inode, new_fl, |
2ee6a576 | 4530 | S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX| |
c93d8f88 | 4531 | S_ENCRYPTED|S_CASEFOLD|S_VERITY); |
ac27a0ec DK |
4532 | } |
4533 | ||
0fc1b451 | 4534 | static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode, |
de9a55b8 | 4535 | struct ext4_inode_info *ei) |
0fc1b451 AK |
4536 | { |
4537 | blkcnt_t i_blocks ; | |
8180a562 AK |
4538 | struct inode *inode = &(ei->vfs_inode); |
4539 | struct super_block *sb = inode->i_sb; | |
0fc1b451 | 4540 | |
e2b911c5 | 4541 | if (ext4_has_feature_huge_file(sb)) { |
0fc1b451 AK |
4542 | /* we are using combined 48 bit field */ |
4543 | i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 | | |
4544 | le32_to_cpu(raw_inode->i_blocks_lo); | |
07a03824 | 4545 | if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) { |
8180a562 AK |
4546 | /* i_blocks represent file system block size */ |
4547 | return i_blocks << (inode->i_blkbits - 9); | |
4548 | } else { | |
4549 | return i_blocks; | |
4550 | } | |
0fc1b451 AK |
4551 | } else { |
4552 | return le32_to_cpu(raw_inode->i_blocks_lo); | |
4553 | } | |
4554 | } | |
ff9ddf7e | 4555 | |
eb9b5f01 | 4556 | static inline int ext4_iget_extra_inode(struct inode *inode, |
152a7b0a TM |
4557 | struct ext4_inode *raw_inode, |
4558 | struct ext4_inode_info *ei) | |
4559 | { | |
4560 | __le32 *magic = (void *)raw_inode + | |
4561 | EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize; | |
eb9b5f01 | 4562 | |
290ab230 EB |
4563 | if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize + sizeof(__le32) <= |
4564 | EXT4_INODE_SIZE(inode->i_sb) && | |
4565 | *magic == cpu_to_le32(EXT4_XATTR_MAGIC)) { | |
152a7b0a | 4566 | ext4_set_inode_state(inode, EXT4_STATE_XATTR); |
eb9b5f01 | 4567 | return ext4_find_inline_data_nolock(inode); |
f19d5870 TM |
4568 | } else |
4569 | EXT4_I(inode)->i_inline_off = 0; | |
eb9b5f01 | 4570 | return 0; |
152a7b0a TM |
4571 | } |
4572 | ||
040cb378 LX |
4573 | int ext4_get_projid(struct inode *inode, kprojid_t *projid) |
4574 | { | |
0b7b7779 | 4575 | if (!ext4_has_feature_project(inode->i_sb)) |
040cb378 LX |
4576 | return -EOPNOTSUPP; |
4577 | *projid = EXT4_I(inode)->i_projid; | |
4578 | return 0; | |
4579 | } | |
4580 | ||
e254d1af EG |
4581 | /* |
4582 | * ext4 has self-managed i_version for ea inodes, it stores the lower 32bit of | |
4583 | * refcount in i_version, so use raw values if inode has EXT4_EA_INODE_FL flag | |
4584 | * set. | |
4585 | */ | |
4586 | static inline void ext4_inode_set_iversion_queried(struct inode *inode, u64 val) | |
4587 | { | |
4588 | if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) | |
4589 | inode_set_iversion_raw(inode, val); | |
4590 | else | |
4591 | inode_set_iversion_queried(inode, val); | |
4592 | } | |
4593 | static inline u64 ext4_inode_peek_iversion(const struct inode *inode) | |
4594 | { | |
4595 | if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) | |
4596 | return inode_peek_iversion_raw(inode); | |
4597 | else | |
4598 | return inode_peek_iversion(inode); | |
4599 | } | |
4600 | ||
8a363970 TT |
4601 | struct inode *__ext4_iget(struct super_block *sb, unsigned long ino, |
4602 | ext4_iget_flags flags, const char *function, | |
4603 | unsigned int line) | |
ac27a0ec | 4604 | { |
617ba13b MC |
4605 | struct ext4_iloc iloc; |
4606 | struct ext4_inode *raw_inode; | |
1d1fe1ee | 4607 | struct ext4_inode_info *ei; |
1d1fe1ee | 4608 | struct inode *inode; |
b436b9be | 4609 | journal_t *journal = EXT4_SB(sb)->s_journal; |
1d1fe1ee | 4610 | long ret; |
7e6e1ef4 | 4611 | loff_t size; |
ac27a0ec | 4612 | int block; |
08cefc7a EB |
4613 | uid_t i_uid; |
4614 | gid_t i_gid; | |
040cb378 | 4615 | projid_t i_projid; |
ac27a0ec | 4616 | |
191ce178 | 4617 | if ((!(flags & EXT4_IGET_SPECIAL) && |
8a363970 TT |
4618 | (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)) || |
4619 | (ino < EXT4_ROOT_INO) || | |
4620 | (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))) { | |
4621 | if (flags & EXT4_IGET_HANDLE) | |
4622 | return ERR_PTR(-ESTALE); | |
014c9caa | 4623 | __ext4_error(sb, function, line, false, EFSCORRUPTED, 0, |
8a363970 TT |
4624 | "inode #%lu: comm %s: iget: illegal inode #", |
4625 | ino, current->comm); | |
4626 | return ERR_PTR(-EFSCORRUPTED); | |
4627 | } | |
4628 | ||
1d1fe1ee DH |
4629 | inode = iget_locked(sb, ino); |
4630 | if (!inode) | |
4631 | return ERR_PTR(-ENOMEM); | |
4632 | if (!(inode->i_state & I_NEW)) | |
4633 | return inode; | |
4634 | ||
4635 | ei = EXT4_I(inode); | |
7dc57615 | 4636 | iloc.bh = NULL; |
ac27a0ec | 4637 | |
8016e29f | 4638 | ret = __ext4_get_inode_loc_noinmem(inode, &iloc); |
1d1fe1ee | 4639 | if (ret < 0) |
ac27a0ec | 4640 | goto bad_inode; |
617ba13b | 4641 | raw_inode = ext4_raw_inode(&iloc); |
814525f4 | 4642 | |
8e4b5eae | 4643 | if ((ino == EXT4_ROOT_INO) && (raw_inode->i_links_count == 0)) { |
8a363970 TT |
4644 | ext4_error_inode(inode, function, line, 0, |
4645 | "iget: root inode unallocated"); | |
8e4b5eae TT |
4646 | ret = -EFSCORRUPTED; |
4647 | goto bad_inode; | |
4648 | } | |
4649 | ||
8a363970 TT |
4650 | if ((flags & EXT4_IGET_HANDLE) && |
4651 | (raw_inode->i_links_count == 0) && (raw_inode->i_mode == 0)) { | |
4652 | ret = -ESTALE; | |
4653 | goto bad_inode; | |
4654 | } | |
4655 | ||
814525f4 DW |
4656 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
4657 | ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize); | |
4658 | if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > | |
2dc8d9e1 EB |
4659 | EXT4_INODE_SIZE(inode->i_sb) || |
4660 | (ei->i_extra_isize & 3)) { | |
8a363970 TT |
4661 | ext4_error_inode(inode, function, line, 0, |
4662 | "iget: bad extra_isize %u " | |
4663 | "(inode size %u)", | |
2dc8d9e1 EB |
4664 | ei->i_extra_isize, |
4665 | EXT4_INODE_SIZE(inode->i_sb)); | |
6a797d27 | 4666 | ret = -EFSCORRUPTED; |
814525f4 DW |
4667 | goto bad_inode; |
4668 | } | |
4669 | } else | |
4670 | ei->i_extra_isize = 0; | |
4671 | ||
4672 | /* Precompute checksum seed for inode metadata */ | |
9aa5d32b | 4673 | if (ext4_has_metadata_csum(sb)) { |
814525f4 DW |
4674 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
4675 | __u32 csum; | |
4676 | __le32 inum = cpu_to_le32(inode->i_ino); | |
4677 | __le32 gen = raw_inode->i_generation; | |
4678 | csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum, | |
4679 | sizeof(inum)); | |
4680 | ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen, | |
4681 | sizeof(gen)); | |
4682 | } | |
4683 | ||
8016e29f HS |
4684 | if ((!ext4_inode_csum_verify(inode, raw_inode, ei) || |
4685 | ext4_simulate_fail(sb, EXT4_SIM_INODE_CRC)) && | |
4686 | (!(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY))) { | |
4687 | ext4_error_inode_err(inode, function, line, 0, | |
4688 | EFSBADCRC, "iget: checksum invalid"); | |
6a797d27 | 4689 | ret = -EFSBADCRC; |
814525f4 DW |
4690 | goto bad_inode; |
4691 | } | |
4692 | ||
ac27a0ec | 4693 | inode->i_mode = le16_to_cpu(raw_inode->i_mode); |
08cefc7a EB |
4694 | i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low); |
4695 | i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low); | |
0b7b7779 | 4696 | if (ext4_has_feature_project(sb) && |
040cb378 LX |
4697 | EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE && |
4698 | EXT4_FITS_IN_INODE(raw_inode, ei, i_projid)) | |
4699 | i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid); | |
4700 | else | |
4701 | i_projid = EXT4_DEF_PROJID; | |
4702 | ||
af5bc92d | 4703 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
08cefc7a EB |
4704 | i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16; |
4705 | i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16; | |
ac27a0ec | 4706 | } |
08cefc7a EB |
4707 | i_uid_write(inode, i_uid); |
4708 | i_gid_write(inode, i_gid); | |
040cb378 | 4709 | ei->i_projid = make_kprojid(&init_user_ns, i_projid); |
bfe86848 | 4710 | set_nlink(inode, le16_to_cpu(raw_inode->i_links_count)); |
ac27a0ec | 4711 | |
353eb83c | 4712 | ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */ |
67cf5b09 | 4713 | ei->i_inline_off = 0; |
ac27a0ec DK |
4714 | ei->i_dir_start_lookup = 0; |
4715 | ei->i_dtime = le32_to_cpu(raw_inode->i_dtime); | |
4716 | /* We now have enough fields to check if the inode was active or not. | |
4717 | * This is needed because nfsd might try to access dead inodes | |
4718 | * the test is that same one that e2fsck uses | |
4719 | * NeilBrown 1999oct15 | |
4720 | */ | |
4721 | if (inode->i_nlink == 0) { | |
393d1d1d DTB |
4722 | if ((inode->i_mode == 0 || |
4723 | !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) && | |
4724 | ino != EXT4_BOOT_LOADER_INO) { | |
ac27a0ec | 4725 | /* this inode is deleted */ |
1d1fe1ee | 4726 | ret = -ESTALE; |
ac27a0ec DK |
4727 | goto bad_inode; |
4728 | } | |
4729 | /* The only unlinked inodes we let through here have | |
4730 | * valid i_mode and are being read by the orphan | |
4731 | * recovery code: that's fine, we're about to complete | |
393d1d1d DTB |
4732 | * the process of deleting those. |
4733 | * OR it is the EXT4_BOOT_LOADER_INO which is | |
4734 | * not initialized on a new filesystem. */ | |
ac27a0ec | 4735 | } |
ac27a0ec | 4736 | ei->i_flags = le32_to_cpu(raw_inode->i_flags); |
043546e4 | 4737 | ext4_set_inode_flags(inode, true); |
0fc1b451 | 4738 | inode->i_blocks = ext4_inode_blocks(raw_inode, ei); |
7973c0c1 | 4739 | ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo); |
e2b911c5 | 4740 | if (ext4_has_feature_64bit(sb)) |
a1ddeb7e BP |
4741 | ei->i_file_acl |= |
4742 | ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32; | |
e08ac99f | 4743 | inode->i_size = ext4_isize(sb, raw_inode); |
7e6e1ef4 | 4744 | if ((size = i_size_read(inode)) < 0) { |
8a363970 TT |
4745 | ext4_error_inode(inode, function, line, 0, |
4746 | "iget: bad i_size value: %lld", size); | |
7e6e1ef4 DW |
4747 | ret = -EFSCORRUPTED; |
4748 | goto bad_inode; | |
4749 | } | |
48a34311 JK |
4750 | /* |
4751 | * If dir_index is not enabled but there's dir with INDEX flag set, | |
4752 | * we'd normally treat htree data as empty space. But with metadata | |
4753 | * checksumming that corrupts checksums so forbid that. | |
4754 | */ | |
4755 | if (!ext4_has_feature_dir_index(sb) && ext4_has_metadata_csum(sb) && | |
4756 | ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) { | |
4757 | ext4_error_inode(inode, function, line, 0, | |
4758 | "iget: Dir with htree data on filesystem without dir_index feature."); | |
4759 | ret = -EFSCORRUPTED; | |
4760 | goto bad_inode; | |
4761 | } | |
ac27a0ec | 4762 | ei->i_disksize = inode->i_size; |
a9e7f447 DM |
4763 | #ifdef CONFIG_QUOTA |
4764 | ei->i_reserved_quota = 0; | |
4765 | #endif | |
ac27a0ec DK |
4766 | inode->i_generation = le32_to_cpu(raw_inode->i_generation); |
4767 | ei->i_block_group = iloc.block_group; | |
a4912123 | 4768 | ei->i_last_alloc_group = ~0; |
ac27a0ec DK |
4769 | /* |
4770 | * NOTE! The in-memory inode i_data array is in little-endian order | |
4771 | * even on big-endian machines: we do NOT byteswap the block numbers! | |
4772 | */ | |
617ba13b | 4773 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
ac27a0ec DK |
4774 | ei->i_data[block] = raw_inode->i_block[block]; |
4775 | INIT_LIST_HEAD(&ei->i_orphan); | |
aa75f4d3 | 4776 | ext4_fc_init_inode(&ei->vfs_inode); |
ac27a0ec | 4777 | |
b436b9be JK |
4778 | /* |
4779 | * Set transaction id's of transactions that have to be committed | |
4780 | * to finish f[data]sync. We set them to currently running transaction | |
4781 | * as we cannot be sure that the inode or some of its metadata isn't | |
4782 | * part of the transaction - the inode could have been reclaimed and | |
4783 | * now it is reread from disk. | |
4784 | */ | |
4785 | if (journal) { | |
4786 | transaction_t *transaction; | |
4787 | tid_t tid; | |
4788 | ||
a931da6a | 4789 | read_lock(&journal->j_state_lock); |
b436b9be JK |
4790 | if (journal->j_running_transaction) |
4791 | transaction = journal->j_running_transaction; | |
4792 | else | |
4793 | transaction = journal->j_committing_transaction; | |
4794 | if (transaction) | |
4795 | tid = transaction->t_tid; | |
4796 | else | |
4797 | tid = journal->j_commit_sequence; | |
a931da6a | 4798 | read_unlock(&journal->j_state_lock); |
b436b9be JK |
4799 | ei->i_sync_tid = tid; |
4800 | ei->i_datasync_tid = tid; | |
4801 | } | |
4802 | ||
0040d987 | 4803 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
ac27a0ec DK |
4804 | if (ei->i_extra_isize == 0) { |
4805 | /* The extra space is currently unused. Use it. */ | |
2dc8d9e1 | 4806 | BUILD_BUG_ON(sizeof(struct ext4_inode) & 3); |
617ba13b MC |
4807 | ei->i_extra_isize = sizeof(struct ext4_inode) - |
4808 | EXT4_GOOD_OLD_INODE_SIZE; | |
ac27a0ec | 4809 | } else { |
eb9b5f01 TT |
4810 | ret = ext4_iget_extra_inode(inode, raw_inode, ei); |
4811 | if (ret) | |
4812 | goto bad_inode; | |
ac27a0ec | 4813 | } |
814525f4 | 4814 | } |
ac27a0ec | 4815 | |
ef7f3835 KS |
4816 | EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode); |
4817 | EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode); | |
4818 | EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode); | |
4819 | EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode); | |
4820 | ||
ed3654eb | 4821 | if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) { |
ee73f9a5 JL |
4822 | u64 ivers = le32_to_cpu(raw_inode->i_disk_version); |
4823 | ||
c4f65706 TT |
4824 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
4825 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
ee73f9a5 | 4826 | ivers |= |
c4f65706 TT |
4827 | (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32; |
4828 | } | |
e254d1af | 4829 | ext4_inode_set_iversion_queried(inode, ivers); |
25ec56b5 JNC |
4830 | } |
4831 | ||
c4b5a614 | 4832 | ret = 0; |
485c26ec | 4833 | if (ei->i_file_acl && |
ce9f24cc | 4834 | !ext4_inode_block_valid(inode, ei->i_file_acl, 1)) { |
8a363970 TT |
4835 | ext4_error_inode(inode, function, line, 0, |
4836 | "iget: bad extended attribute block %llu", | |
24676da4 | 4837 | ei->i_file_acl); |
6a797d27 | 4838 | ret = -EFSCORRUPTED; |
485c26ec | 4839 | goto bad_inode; |
f19d5870 | 4840 | } else if (!ext4_has_inline_data(inode)) { |
bc716523 | 4841 | /* validate the block references in the inode */ |
8016e29f HS |
4842 | if (!(EXT4_SB(sb)->s_mount_state & EXT4_FC_REPLAY) && |
4843 | (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || | |
4844 | (S_ISLNK(inode->i_mode) && | |
4845 | !ext4_inode_is_fast_symlink(inode)))) { | |
bc716523 | 4846 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
f19d5870 | 4847 | ret = ext4_ext_check_inode(inode); |
bc716523 LS |
4848 | else |
4849 | ret = ext4_ind_check_inode(inode); | |
f19d5870 | 4850 | } |
fe2c8191 | 4851 | } |
567f3e9a | 4852 | if (ret) |
de9a55b8 | 4853 | goto bad_inode; |
7a262f7c | 4854 | |
ac27a0ec | 4855 | if (S_ISREG(inode->i_mode)) { |
617ba13b | 4856 | inode->i_op = &ext4_file_inode_operations; |
be64f884 | 4857 | inode->i_fop = &ext4_file_operations; |
617ba13b | 4858 | ext4_set_aops(inode); |
ac27a0ec | 4859 | } else if (S_ISDIR(inode->i_mode)) { |
617ba13b MC |
4860 | inode->i_op = &ext4_dir_inode_operations; |
4861 | inode->i_fop = &ext4_dir_operations; | |
ac27a0ec | 4862 | } else if (S_ISLNK(inode->i_mode)) { |
6390d33b LR |
4863 | /* VFS does not allow setting these so must be corruption */ |
4864 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) { | |
8a363970 TT |
4865 | ext4_error_inode(inode, function, line, 0, |
4866 | "iget: immutable or append flags " | |
4867 | "not allowed on symlinks"); | |
6390d33b LR |
4868 | ret = -EFSCORRUPTED; |
4869 | goto bad_inode; | |
4870 | } | |
592ddec7 | 4871 | if (IS_ENCRYPTED(inode)) { |
a7a67e8a AV |
4872 | inode->i_op = &ext4_encrypted_symlink_inode_operations; |
4873 | ext4_set_aops(inode); | |
4874 | } else if (ext4_inode_is_fast_symlink(inode)) { | |
75e7566b | 4875 | inode->i_link = (char *)ei->i_data; |
617ba13b | 4876 | inode->i_op = &ext4_fast_symlink_inode_operations; |
e83c1397 DG |
4877 | nd_terminate_link(ei->i_data, inode->i_size, |
4878 | sizeof(ei->i_data) - 1); | |
4879 | } else { | |
617ba13b MC |
4880 | inode->i_op = &ext4_symlink_inode_operations; |
4881 | ext4_set_aops(inode); | |
ac27a0ec | 4882 | } |
21fc61c7 | 4883 | inode_nohighmem(inode); |
563bdd61 TT |
4884 | } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) || |
4885 | S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) { | |
617ba13b | 4886 | inode->i_op = &ext4_special_inode_operations; |
ac27a0ec DK |
4887 | if (raw_inode->i_block[0]) |
4888 | init_special_inode(inode, inode->i_mode, | |
4889 | old_decode_dev(le32_to_cpu(raw_inode->i_block[0]))); | |
4890 | else | |
4891 | init_special_inode(inode, inode->i_mode, | |
4892 | new_decode_dev(le32_to_cpu(raw_inode->i_block[1]))); | |
393d1d1d DTB |
4893 | } else if (ino == EXT4_BOOT_LOADER_INO) { |
4894 | make_bad_inode(inode); | |
563bdd61 | 4895 | } else { |
6a797d27 | 4896 | ret = -EFSCORRUPTED; |
8a363970 TT |
4897 | ext4_error_inode(inode, function, line, 0, |
4898 | "iget: bogus i_mode (%o)", inode->i_mode); | |
563bdd61 | 4899 | goto bad_inode; |
ac27a0ec | 4900 | } |
6456ca65 TT |
4901 | if (IS_CASEFOLDED(inode) && !ext4_has_feature_casefold(inode->i_sb)) |
4902 | ext4_error_inode(inode, function, line, 0, | |
4903 | "casefold flag without casefold feature"); | |
af5bc92d | 4904 | brelse(iloc.bh); |
dec214d0 | 4905 | |
1d1fe1ee DH |
4906 | unlock_new_inode(inode); |
4907 | return inode; | |
ac27a0ec DK |
4908 | |
4909 | bad_inode: | |
567f3e9a | 4910 | brelse(iloc.bh); |
1d1fe1ee DH |
4911 | iget_failed(inode); |
4912 | return ERR_PTR(ret); | |
ac27a0ec DK |
4913 | } |
4914 | ||
0fc1b451 AK |
4915 | static int ext4_inode_blocks_set(handle_t *handle, |
4916 | struct ext4_inode *raw_inode, | |
4917 | struct ext4_inode_info *ei) | |
4918 | { | |
4919 | struct inode *inode = &(ei->vfs_inode); | |
28936b62 | 4920 | u64 i_blocks = READ_ONCE(inode->i_blocks); |
0fc1b451 | 4921 | struct super_block *sb = inode->i_sb; |
0fc1b451 AK |
4922 | |
4923 | if (i_blocks <= ~0U) { | |
4924 | /* | |
4907cb7b | 4925 | * i_blocks can be represented in a 32 bit variable |
0fc1b451 AK |
4926 | * as multiple of 512 bytes |
4927 | */ | |
8180a562 | 4928 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 4929 | raw_inode->i_blocks_high = 0; |
84a8dce2 | 4930 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
f287a1a5 TT |
4931 | return 0; |
4932 | } | |
e2b911c5 | 4933 | if (!ext4_has_feature_huge_file(sb)) |
f287a1a5 TT |
4934 | return -EFBIG; |
4935 | ||
4936 | if (i_blocks <= 0xffffffffffffULL) { | |
0fc1b451 AK |
4937 | /* |
4938 | * i_blocks can be represented in a 48 bit variable | |
4939 | * as multiple of 512 bytes | |
4940 | */ | |
8180a562 | 4941 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 4942 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); |
84a8dce2 | 4943 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
0fc1b451 | 4944 | } else { |
84a8dce2 | 4945 | ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
8180a562 AK |
4946 | /* i_block is stored in file system block size */ |
4947 | i_blocks = i_blocks >> (inode->i_blkbits - 9); | |
4948 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); | |
4949 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); | |
0fc1b451 | 4950 | } |
f287a1a5 | 4951 | return 0; |
0fc1b451 AK |
4952 | } |
4953 | ||
3f19b2ab DH |
4954 | static void __ext4_update_other_inode_time(struct super_block *sb, |
4955 | unsigned long orig_ino, | |
4956 | unsigned long ino, | |
4957 | struct ext4_inode *raw_inode) | |
a26f4992 | 4958 | { |
3f19b2ab DH |
4959 | struct inode *inode; |
4960 | ||
4961 | inode = find_inode_by_ino_rcu(sb, ino); | |
4962 | if (!inode) | |
4963 | return; | |
a26f4992 | 4964 | |
ed296c6c | 4965 | if (!inode_is_dirtytime_only(inode)) |
3f19b2ab DH |
4966 | return; |
4967 | ||
a26f4992 | 4968 | spin_lock(&inode->i_lock); |
ed296c6c | 4969 | if (inode_is_dirtytime_only(inode)) { |
a26f4992 TT |
4970 | struct ext4_inode_info *ei = EXT4_I(inode); |
4971 | ||
5fcd5750 | 4972 | inode->i_state &= ~I_DIRTY_TIME; |
a26f4992 TT |
4973 | spin_unlock(&inode->i_lock); |
4974 | ||
4975 | spin_lock(&ei->i_raw_lock); | |
3f19b2ab DH |
4976 | EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode); |
4977 | EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode); | |
4978 | EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode); | |
4979 | ext4_inode_csum_set(inode, raw_inode, ei); | |
a26f4992 | 4980 | spin_unlock(&ei->i_raw_lock); |
3f19b2ab DH |
4981 | trace_ext4_other_inode_update_time(inode, orig_ino); |
4982 | return; | |
a26f4992 TT |
4983 | } |
4984 | spin_unlock(&inode->i_lock); | |
a26f4992 TT |
4985 | } |
4986 | ||
4987 | /* | |
4988 | * Opportunistically update the other time fields for other inodes in | |
4989 | * the same inode table block. | |
4990 | */ | |
4991 | static void ext4_update_other_inodes_time(struct super_block *sb, | |
4992 | unsigned long orig_ino, char *buf) | |
4993 | { | |
a26f4992 TT |
4994 | unsigned long ino; |
4995 | int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; | |
4996 | int inode_size = EXT4_INODE_SIZE(sb); | |
4997 | ||
0f0ff9a9 TT |
4998 | /* |
4999 | * Calculate the first inode in the inode table block. Inode | |
5000 | * numbers are one-based. That is, the first inode in a block | |
5001 | * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1). | |
5002 | */ | |
5003 | ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1; | |
3f19b2ab | 5004 | rcu_read_lock(); |
a26f4992 TT |
5005 | for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) { |
5006 | if (ino == orig_ino) | |
5007 | continue; | |
3f19b2ab DH |
5008 | __ext4_update_other_inode_time(sb, orig_ino, ino, |
5009 | (struct ext4_inode *)buf); | |
a26f4992 | 5010 | } |
3f19b2ab | 5011 | rcu_read_unlock(); |
a26f4992 TT |
5012 | } |
5013 | ||
ac27a0ec DK |
5014 | /* |
5015 | * Post the struct inode info into an on-disk inode location in the | |
5016 | * buffer-cache. This gobbles the caller's reference to the | |
5017 | * buffer_head in the inode location struct. | |
5018 | * | |
5019 | * The caller must have write access to iloc->bh. | |
5020 | */ | |
617ba13b | 5021 | static int ext4_do_update_inode(handle_t *handle, |
ac27a0ec | 5022 | struct inode *inode, |
830156c7 | 5023 | struct ext4_iloc *iloc) |
ac27a0ec | 5024 | { |
617ba13b MC |
5025 | struct ext4_inode *raw_inode = ext4_raw_inode(iloc); |
5026 | struct ext4_inode_info *ei = EXT4_I(inode); | |
ac27a0ec | 5027 | struct buffer_head *bh = iloc->bh; |
202ee5df | 5028 | struct super_block *sb = inode->i_sb; |
7d8bd3c7 | 5029 | int err = 0, block; |
202ee5df | 5030 | int need_datasync = 0, set_large_file = 0; |
08cefc7a EB |
5031 | uid_t i_uid; |
5032 | gid_t i_gid; | |
040cb378 | 5033 | projid_t i_projid; |
ac27a0ec | 5034 | |
202ee5df TT |
5035 | spin_lock(&ei->i_raw_lock); |
5036 | ||
5037 | /* For fields not tracked in the in-memory inode, | |
ac27a0ec | 5038 | * initialise them to zero for new inodes. */ |
19f5fb7a | 5039 | if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) |
617ba13b | 5040 | memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size); |
ac27a0ec | 5041 | |
13221811 LM |
5042 | err = ext4_inode_blocks_set(handle, raw_inode, ei); |
5043 | if (err) { | |
5044 | spin_unlock(&ei->i_raw_lock); | |
5045 | goto out_brelse; | |
5046 | } | |
5047 | ||
ac27a0ec | 5048 | raw_inode->i_mode = cpu_to_le16(inode->i_mode); |
08cefc7a EB |
5049 | i_uid = i_uid_read(inode); |
5050 | i_gid = i_gid_read(inode); | |
040cb378 | 5051 | i_projid = from_kprojid(&init_user_ns, ei->i_projid); |
af5bc92d | 5052 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
08cefc7a EB |
5053 | raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid)); |
5054 | raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid)); | |
ac27a0ec DK |
5055 | /* |
5056 | * Fix up interoperability with old kernels. Otherwise, old inodes get | |
5057 | * re-used with the upper 16 bits of the uid/gid intact | |
5058 | */ | |
93e3b4e6 DJ |
5059 | if (ei->i_dtime && list_empty(&ei->i_orphan)) { |
5060 | raw_inode->i_uid_high = 0; | |
5061 | raw_inode->i_gid_high = 0; | |
5062 | } else { | |
ac27a0ec | 5063 | raw_inode->i_uid_high = |
08cefc7a | 5064 | cpu_to_le16(high_16_bits(i_uid)); |
ac27a0ec | 5065 | raw_inode->i_gid_high = |
08cefc7a | 5066 | cpu_to_le16(high_16_bits(i_gid)); |
ac27a0ec DK |
5067 | } |
5068 | } else { | |
08cefc7a EB |
5069 | raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid)); |
5070 | raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid)); | |
ac27a0ec DK |
5071 | raw_inode->i_uid_high = 0; |
5072 | raw_inode->i_gid_high = 0; | |
5073 | } | |
5074 | raw_inode->i_links_count = cpu_to_le16(inode->i_nlink); | |
ef7f3835 KS |
5075 | |
5076 | EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode); | |
5077 | EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode); | |
5078 | EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode); | |
5079 | EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode); | |
5080 | ||
ac27a0ec | 5081 | raw_inode->i_dtime = cpu_to_le32(ei->i_dtime); |
353eb83c | 5082 | raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF); |
ed3654eb | 5083 | if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) |
a1ddeb7e BP |
5084 | raw_inode->i_file_acl_high = |
5085 | cpu_to_le16(ei->i_file_acl >> 32); | |
7973c0c1 | 5086 | raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl); |
dce8e237 | 5087 | if (READ_ONCE(ei->i_disksize) != ext4_isize(inode->i_sb, raw_inode)) { |
b71fc079 JK |
5088 | ext4_isize_set(raw_inode, ei->i_disksize); |
5089 | need_datasync = 1; | |
5090 | } | |
a48380f7 | 5091 | if (ei->i_disksize > 0x7fffffffULL) { |
e2b911c5 | 5092 | if (!ext4_has_feature_large_file(sb) || |
a48380f7 | 5093 | EXT4_SB(sb)->s_es->s_rev_level == |
202ee5df TT |
5094 | cpu_to_le32(EXT4_GOOD_OLD_REV)) |
5095 | set_large_file = 1; | |
ac27a0ec DK |
5096 | } |
5097 | raw_inode->i_generation = cpu_to_le32(inode->i_generation); | |
5098 | if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) { | |
5099 | if (old_valid_dev(inode->i_rdev)) { | |
5100 | raw_inode->i_block[0] = | |
5101 | cpu_to_le32(old_encode_dev(inode->i_rdev)); | |
5102 | raw_inode->i_block[1] = 0; | |
5103 | } else { | |
5104 | raw_inode->i_block[0] = 0; | |
5105 | raw_inode->i_block[1] = | |
5106 | cpu_to_le32(new_encode_dev(inode->i_rdev)); | |
5107 | raw_inode->i_block[2] = 0; | |
5108 | } | |
f19d5870 | 5109 | } else if (!ext4_has_inline_data(inode)) { |
de9a55b8 TT |
5110 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
5111 | raw_inode->i_block[block] = ei->i_data[block]; | |
f19d5870 | 5112 | } |
ac27a0ec | 5113 | |
ed3654eb | 5114 | if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) { |
e254d1af | 5115 | u64 ivers = ext4_inode_peek_iversion(inode); |
ee73f9a5 JL |
5116 | |
5117 | raw_inode->i_disk_version = cpu_to_le32(ivers); | |
c4f65706 TT |
5118 | if (ei->i_extra_isize) { |
5119 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
5120 | raw_inode->i_version_hi = | |
ee73f9a5 | 5121 | cpu_to_le32(ivers >> 32); |
c4f65706 TT |
5122 | raw_inode->i_extra_isize = |
5123 | cpu_to_le16(ei->i_extra_isize); | |
5124 | } | |
25ec56b5 | 5125 | } |
040cb378 | 5126 | |
0b7b7779 | 5127 | BUG_ON(!ext4_has_feature_project(inode->i_sb) && |
040cb378 LX |
5128 | i_projid != EXT4_DEF_PROJID); |
5129 | ||
5130 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && | |
5131 | EXT4_FITS_IN_INODE(raw_inode, ei, i_projid)) | |
5132 | raw_inode->i_projid = cpu_to_le32(i_projid); | |
5133 | ||
814525f4 | 5134 | ext4_inode_csum_set(inode, raw_inode, ei); |
202ee5df | 5135 | spin_unlock(&ei->i_raw_lock); |
1751e8a6 | 5136 | if (inode->i_sb->s_flags & SB_LAZYTIME) |
a26f4992 TT |
5137 | ext4_update_other_inodes_time(inode->i_sb, inode->i_ino, |
5138 | bh->b_data); | |
202ee5df | 5139 | |
830156c7 | 5140 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
7d8bd3c7 SL |
5141 | err = ext4_handle_dirty_metadata(handle, NULL, bh); |
5142 | if (err) | |
5143 | goto out_brelse; | |
19f5fb7a | 5144 | ext4_clear_inode_state(inode, EXT4_STATE_NEW); |
202ee5df | 5145 | if (set_large_file) { |
5d601255 | 5146 | BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access"); |
202ee5df TT |
5147 | err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh); |
5148 | if (err) | |
5149 | goto out_brelse; | |
05c2c00f | 5150 | lock_buffer(EXT4_SB(sb)->s_sbh); |
e2b911c5 | 5151 | ext4_set_feature_large_file(sb); |
05c2c00f JK |
5152 | ext4_superblock_csum_set(sb); |
5153 | unlock_buffer(EXT4_SB(sb)->s_sbh); | |
202ee5df | 5154 | ext4_handle_sync(handle); |
a3f5cf14 JK |
5155 | err = ext4_handle_dirty_metadata(handle, NULL, |
5156 | EXT4_SB(sb)->s_sbh); | |
202ee5df | 5157 | } |
b71fc079 | 5158 | ext4_update_inode_fsync_trans(handle, inode, need_datasync); |
ac27a0ec | 5159 | out_brelse: |
af5bc92d | 5160 | brelse(bh); |
617ba13b | 5161 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
5162 | return err; |
5163 | } | |
5164 | ||
5165 | /* | |
617ba13b | 5166 | * ext4_write_inode() |
ac27a0ec DK |
5167 | * |
5168 | * We are called from a few places: | |
5169 | * | |
87f7e416 | 5170 | * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files. |
ac27a0ec | 5171 | * Here, there will be no transaction running. We wait for any running |
4907cb7b | 5172 | * transaction to commit. |
ac27a0ec | 5173 | * |
87f7e416 TT |
5174 | * - Within flush work (sys_sync(), kupdate and such). |
5175 | * We wait on commit, if told to. | |
ac27a0ec | 5176 | * |
87f7e416 TT |
5177 | * - Within iput_final() -> write_inode_now() |
5178 | * We wait on commit, if told to. | |
ac27a0ec DK |
5179 | * |
5180 | * In all cases it is actually safe for us to return without doing anything, | |
5181 | * because the inode has been copied into a raw inode buffer in | |
87f7e416 TT |
5182 | * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL |
5183 | * writeback. | |
ac27a0ec DK |
5184 | * |
5185 | * Note that we are absolutely dependent upon all inode dirtiers doing the | |
5186 | * right thing: they *must* call mark_inode_dirty() after dirtying info in | |
5187 | * which we are interested. | |
5188 | * | |
5189 | * It would be a bug for them to not do this. The code: | |
5190 | * | |
5191 | * mark_inode_dirty(inode) | |
5192 | * stuff(); | |
5193 | * inode->i_size = expr; | |
5194 | * | |
87f7e416 TT |
5195 | * is in error because write_inode() could occur while `stuff()' is running, |
5196 | * and the new i_size will be lost. Plus the inode will no longer be on the | |
5197 | * superblock's dirty inode list. | |
ac27a0ec | 5198 | */ |
a9185b41 | 5199 | int ext4_write_inode(struct inode *inode, struct writeback_control *wbc) |
ac27a0ec | 5200 | { |
91ac6f43 FM |
5201 | int err; |
5202 | ||
18f2c4fc TT |
5203 | if (WARN_ON_ONCE(current->flags & PF_MEMALLOC) || |
5204 | sb_rdonly(inode->i_sb)) | |
ac27a0ec DK |
5205 | return 0; |
5206 | ||
18f2c4fc TT |
5207 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
5208 | return -EIO; | |
5209 | ||
91ac6f43 FM |
5210 | if (EXT4_SB(inode->i_sb)->s_journal) { |
5211 | if (ext4_journal_current_handle()) { | |
5212 | jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n"); | |
5213 | dump_stack(); | |
5214 | return -EIO; | |
5215 | } | |
ac27a0ec | 5216 | |
10542c22 JK |
5217 | /* |
5218 | * No need to force transaction in WB_SYNC_NONE mode. Also | |
5219 | * ext4_sync_fs() will force the commit after everything is | |
5220 | * written. | |
5221 | */ | |
5222 | if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync) | |
91ac6f43 FM |
5223 | return 0; |
5224 | ||
aa75f4d3 | 5225 | err = ext4_fc_commit(EXT4_SB(inode->i_sb)->s_journal, |
18f2c4fc | 5226 | EXT4_I(inode)->i_sync_tid); |
91ac6f43 FM |
5227 | } else { |
5228 | struct ext4_iloc iloc; | |
ac27a0ec | 5229 | |
8016e29f | 5230 | err = __ext4_get_inode_loc_noinmem(inode, &iloc); |
91ac6f43 FM |
5231 | if (err) |
5232 | return err; | |
10542c22 JK |
5233 | /* |
5234 | * sync(2) will flush the whole buffer cache. No need to do | |
5235 | * it here separately for each inode. | |
5236 | */ | |
5237 | if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) | |
830156c7 FM |
5238 | sync_dirty_buffer(iloc.bh); |
5239 | if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) { | |
54d3adbc TT |
5240 | ext4_error_inode_block(inode, iloc.bh->b_blocknr, EIO, |
5241 | "IO error syncing inode"); | |
830156c7 FM |
5242 | err = -EIO; |
5243 | } | |
fd2dd9fb | 5244 | brelse(iloc.bh); |
91ac6f43 FM |
5245 | } |
5246 | return err; | |
ac27a0ec DK |
5247 | } |
5248 | ||
53e87268 JK |
5249 | /* |
5250 | * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate | |
5251 | * buffers that are attached to a page stradding i_size and are undergoing | |
5252 | * commit. In that case we have to wait for commit to finish and try again. | |
5253 | */ | |
5254 | static void ext4_wait_for_tail_page_commit(struct inode *inode) | |
5255 | { | |
5256 | struct page *page; | |
5257 | unsigned offset; | |
5258 | journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; | |
5259 | tid_t commit_tid = 0; | |
5260 | int ret; | |
5261 | ||
09cbfeaf | 5262 | offset = inode->i_size & (PAGE_SIZE - 1); |
53e87268 | 5263 | /* |
565333a1 | 5264 | * If the page is fully truncated, we don't need to wait for any commit |
5265 | * (and we even should not as __ext4_journalled_invalidatepage() may | |
5266 | * strip all buffers from the page but keep the page dirty which can then | |
5267 | * confuse e.g. concurrent ext4_writepage() seeing dirty page without | |
5268 | * buffers). Also we don't need to wait for any commit if all buffers in | |
5269 | * the page remain valid. This is most beneficial for the common case of | |
5270 | * blocksize == PAGESIZE. | |
53e87268 | 5271 | */ |
565333a1 | 5272 | if (!offset || offset > (PAGE_SIZE - i_blocksize(inode))) |
53e87268 JK |
5273 | return; |
5274 | while (1) { | |
5275 | page = find_lock_page(inode->i_mapping, | |
09cbfeaf | 5276 | inode->i_size >> PAGE_SHIFT); |
53e87268 JK |
5277 | if (!page) |
5278 | return; | |
ca99fdd2 | 5279 | ret = __ext4_journalled_invalidatepage(page, offset, |
09cbfeaf | 5280 | PAGE_SIZE - offset); |
53e87268 | 5281 | unlock_page(page); |
09cbfeaf | 5282 | put_page(page); |
53e87268 JK |
5283 | if (ret != -EBUSY) |
5284 | return; | |
5285 | commit_tid = 0; | |
5286 | read_lock(&journal->j_state_lock); | |
5287 | if (journal->j_committing_transaction) | |
5288 | commit_tid = journal->j_committing_transaction->t_tid; | |
5289 | read_unlock(&journal->j_state_lock); | |
5290 | if (commit_tid) | |
5291 | jbd2_log_wait_commit(journal, commit_tid); | |
5292 | } | |
5293 | } | |
5294 | ||
ac27a0ec | 5295 | /* |
617ba13b | 5296 | * ext4_setattr() |
ac27a0ec DK |
5297 | * |
5298 | * Called from notify_change. | |
5299 | * | |
5300 | * We want to trap VFS attempts to truncate the file as soon as | |
5301 | * possible. In particular, we want to make sure that when the VFS | |
5302 | * shrinks i_size, we put the inode on the orphan list and modify | |
5303 | * i_disksize immediately, so that during the subsequent flushing of | |
5304 | * dirty pages and freeing of disk blocks, we can guarantee that any | |
5305 | * commit will leave the blocks being flushed in an unused state on | |
5306 | * disk. (On recovery, the inode will get truncated and the blocks will | |
5307 | * be freed, so we have a strong guarantee that no future commit will | |
5308 | * leave these blocks visible to the user.) | |
5309 | * | |
678aaf48 JK |
5310 | * Another thing we have to assure is that if we are in ordered mode |
5311 | * and inode is still attached to the committing transaction, we must | |
5312 | * we start writeout of all the dirty pages which are being truncated. | |
5313 | * This way we are sure that all the data written in the previous | |
5314 | * transaction are already on disk (truncate waits for pages under | |
5315 | * writeback). | |
5316 | * | |
5317 | * Called with inode->i_mutex down. | |
ac27a0ec | 5318 | */ |
549c7297 CB |
5319 | int ext4_setattr(struct user_namespace *mnt_userns, struct dentry *dentry, |
5320 | struct iattr *attr) | |
ac27a0ec | 5321 | { |
2b0143b5 | 5322 | struct inode *inode = d_inode(dentry); |
ac27a0ec | 5323 | int error, rc = 0; |
3d287de3 | 5324 | int orphan = 0; |
ac27a0ec DK |
5325 | const unsigned int ia_valid = attr->ia_valid; |
5326 | ||
0db1ff22 TT |
5327 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
5328 | return -EIO; | |
5329 | ||
02b016ca TT |
5330 | if (unlikely(IS_IMMUTABLE(inode))) |
5331 | return -EPERM; | |
5332 | ||
5333 | if (unlikely(IS_APPEND(inode) && | |
5334 | (ia_valid & (ATTR_MODE | ATTR_UID | | |
5335 | ATTR_GID | ATTR_TIMES_SET)))) | |
5336 | return -EPERM; | |
5337 | ||
14f3db55 | 5338 | error = setattr_prepare(mnt_userns, dentry, attr); |
ac27a0ec DK |
5339 | if (error) |
5340 | return error; | |
5341 | ||
3ce2b8dd EB |
5342 | error = fscrypt_prepare_setattr(dentry, attr); |
5343 | if (error) | |
5344 | return error; | |
5345 | ||
c93d8f88 EB |
5346 | error = fsverity_prepare_setattr(dentry, attr); |
5347 | if (error) | |
5348 | return error; | |
5349 | ||
a7cdadee JK |
5350 | if (is_quota_modification(inode, attr)) { |
5351 | error = dquot_initialize(inode); | |
5352 | if (error) | |
5353 | return error; | |
5354 | } | |
aa75f4d3 | 5355 | ext4_fc_start_update(inode); |
08cefc7a EB |
5356 | if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) || |
5357 | (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) { | |
ac27a0ec DK |
5358 | handle_t *handle; |
5359 | ||
5360 | /* (user+group)*(old+new) structure, inode write (sb, | |
5361 | * inode block, ? - but truncate inode update has it) */ | |
9924a92a TT |
5362 | handle = ext4_journal_start(inode, EXT4_HT_QUOTA, |
5363 | (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) + | |
5364 | EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3); | |
ac27a0ec DK |
5365 | if (IS_ERR(handle)) { |
5366 | error = PTR_ERR(handle); | |
5367 | goto err_out; | |
5368 | } | |
7a9ca53a TE |
5369 | |
5370 | /* dquot_transfer() calls back ext4_get_inode_usage() which | |
5371 | * counts xattr inode references. | |
5372 | */ | |
5373 | down_read(&EXT4_I(inode)->xattr_sem); | |
b43fa828 | 5374 | error = dquot_transfer(inode, attr); |
7a9ca53a TE |
5375 | up_read(&EXT4_I(inode)->xattr_sem); |
5376 | ||
ac27a0ec | 5377 | if (error) { |
617ba13b | 5378 | ext4_journal_stop(handle); |
aa75f4d3 | 5379 | ext4_fc_stop_update(inode); |
ac27a0ec DK |
5380 | return error; |
5381 | } | |
5382 | /* Update corresponding info in inode so that everything is in | |
5383 | * one transaction */ | |
5384 | if (attr->ia_valid & ATTR_UID) | |
5385 | inode->i_uid = attr->ia_uid; | |
5386 | if (attr->ia_valid & ATTR_GID) | |
5387 | inode->i_gid = attr->ia_gid; | |
617ba13b MC |
5388 | error = ext4_mark_inode_dirty(handle, inode); |
5389 | ext4_journal_stop(handle); | |
512c15ef PB |
5390 | if (unlikely(error)) { |
5391 | ext4_fc_stop_update(inode); | |
4209ae12 | 5392 | return error; |
512c15ef | 5393 | } |
ac27a0ec DK |
5394 | } |
5395 | ||
3da40c7b | 5396 | if (attr->ia_valid & ATTR_SIZE) { |
5208386c | 5397 | handle_t *handle; |
3da40c7b | 5398 | loff_t oldsize = inode->i_size; |
b9c1c267 | 5399 | int shrink = (attr->ia_size < inode->i_size); |
562c72aa | 5400 | |
12e9b892 | 5401 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) { |
e2b46574 ES |
5402 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
5403 | ||
aa75f4d3 HS |
5404 | if (attr->ia_size > sbi->s_bitmap_maxbytes) { |
5405 | ext4_fc_stop_update(inode); | |
0c095c7f | 5406 | return -EFBIG; |
aa75f4d3 | 5407 | } |
e2b46574 | 5408 | } |
aa75f4d3 HS |
5409 | if (!S_ISREG(inode->i_mode)) { |
5410 | ext4_fc_stop_update(inode); | |
3da40c7b | 5411 | return -EINVAL; |
aa75f4d3 | 5412 | } |
dff6efc3 CH |
5413 | |
5414 | if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size) | |
5415 | inode_inc_iversion(inode); | |
5416 | ||
b9c1c267 JK |
5417 | if (shrink) { |
5418 | if (ext4_should_order_data(inode)) { | |
5419 | error = ext4_begin_ordered_truncate(inode, | |
678aaf48 | 5420 | attr->ia_size); |
b9c1c267 JK |
5421 | if (error) |
5422 | goto err_out; | |
5423 | } | |
5424 | /* | |
5425 | * Blocks are going to be removed from the inode. Wait | |
5426 | * for dio in flight. | |
5427 | */ | |
5428 | inode_dio_wait(inode); | |
5429 | } | |
5430 | ||
5431 | down_write(&EXT4_I(inode)->i_mmap_sem); | |
5432 | ||
5433 | rc = ext4_break_layouts(inode); | |
5434 | if (rc) { | |
5435 | up_write(&EXT4_I(inode)->i_mmap_sem); | |
aa75f4d3 | 5436 | goto err_out; |
3da40c7b | 5437 | } |
b9c1c267 | 5438 | |
3da40c7b | 5439 | if (attr->ia_size != inode->i_size) { |
5208386c JK |
5440 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 3); |
5441 | if (IS_ERR(handle)) { | |
5442 | error = PTR_ERR(handle); | |
b9c1c267 | 5443 | goto out_mmap_sem; |
5208386c | 5444 | } |
3da40c7b | 5445 | if (ext4_handle_valid(handle) && shrink) { |
5208386c JK |
5446 | error = ext4_orphan_add(handle, inode); |
5447 | orphan = 1; | |
5448 | } | |
911af577 EG |
5449 | /* |
5450 | * Update c/mtime on truncate up, ext4_truncate() will | |
5451 | * update c/mtime in shrink case below | |
5452 | */ | |
5453 | if (!shrink) { | |
eeca7ea1 | 5454 | inode->i_mtime = current_time(inode); |
911af577 EG |
5455 | inode->i_ctime = inode->i_mtime; |
5456 | } | |
aa75f4d3 HS |
5457 | |
5458 | if (shrink) | |
a80f7fcf | 5459 | ext4_fc_track_range(handle, inode, |
aa75f4d3 HS |
5460 | (attr->ia_size > 0 ? attr->ia_size - 1 : 0) >> |
5461 | inode->i_sb->s_blocksize_bits, | |
5462 | (oldsize > 0 ? oldsize - 1 : 0) >> | |
5463 | inode->i_sb->s_blocksize_bits); | |
5464 | else | |
5465 | ext4_fc_track_range( | |
a80f7fcf | 5466 | handle, inode, |
aa75f4d3 HS |
5467 | (oldsize > 0 ? oldsize - 1 : oldsize) >> |
5468 | inode->i_sb->s_blocksize_bits, | |
5469 | (attr->ia_size > 0 ? attr->ia_size - 1 : 0) >> | |
5470 | inode->i_sb->s_blocksize_bits); | |
5471 | ||
90e775b7 | 5472 | down_write(&EXT4_I(inode)->i_data_sem); |
5208386c JK |
5473 | EXT4_I(inode)->i_disksize = attr->ia_size; |
5474 | rc = ext4_mark_inode_dirty(handle, inode); | |
5475 | if (!error) | |
5476 | error = rc; | |
90e775b7 JK |
5477 | /* |
5478 | * We have to update i_size under i_data_sem together | |
5479 | * with i_disksize to avoid races with writeback code | |
5480 | * running ext4_wb_update_i_disksize(). | |
5481 | */ | |
5482 | if (!error) | |
5483 | i_size_write(inode, attr->ia_size); | |
5484 | up_write(&EXT4_I(inode)->i_data_sem); | |
5208386c | 5485 | ext4_journal_stop(handle); |
b9c1c267 JK |
5486 | if (error) |
5487 | goto out_mmap_sem; | |
5488 | if (!shrink) { | |
5489 | pagecache_isize_extended(inode, oldsize, | |
5490 | inode->i_size); | |
5491 | } else if (ext4_should_journal_data(inode)) { | |
5492 | ext4_wait_for_tail_page_commit(inode); | |
678aaf48 | 5493 | } |
d6320cbf | 5494 | } |
430657b6 | 5495 | |
5208386c JK |
5496 | /* |
5497 | * Truncate pagecache after we've waited for commit | |
5498 | * in data=journal mode to make pages freeable. | |
5499 | */ | |
923ae0ff | 5500 | truncate_pagecache(inode, inode->i_size); |
b9c1c267 JK |
5501 | /* |
5502 | * Call ext4_truncate() even if i_size didn't change to | |
5503 | * truncate possible preallocated blocks. | |
5504 | */ | |
5505 | if (attr->ia_size <= oldsize) { | |
2c98eb5e TT |
5506 | rc = ext4_truncate(inode); |
5507 | if (rc) | |
5508 | error = rc; | |
5509 | } | |
b9c1c267 | 5510 | out_mmap_sem: |
ea3d7209 | 5511 | up_write(&EXT4_I(inode)->i_mmap_sem); |
072bd7ea | 5512 | } |
ac27a0ec | 5513 | |
2c98eb5e | 5514 | if (!error) { |
14f3db55 | 5515 | setattr_copy(mnt_userns, inode, attr); |
1025774c CH |
5516 | mark_inode_dirty(inode); |
5517 | } | |
5518 | ||
5519 | /* | |
5520 | * If the call to ext4_truncate failed to get a transaction handle at | |
5521 | * all, we need to clean up the in-core orphan list manually. | |
5522 | */ | |
3d287de3 | 5523 | if (orphan && inode->i_nlink) |
617ba13b | 5524 | ext4_orphan_del(NULL, inode); |
ac27a0ec | 5525 | |
2c98eb5e | 5526 | if (!error && (ia_valid & ATTR_MODE)) |
14f3db55 | 5527 | rc = posix_acl_chmod(mnt_userns, inode, inode->i_mode); |
ac27a0ec DK |
5528 | |
5529 | err_out: | |
aa75f4d3 HS |
5530 | if (error) |
5531 | ext4_std_error(inode->i_sb, error); | |
ac27a0ec DK |
5532 | if (!error) |
5533 | error = rc; | |
aa75f4d3 | 5534 | ext4_fc_stop_update(inode); |
ac27a0ec DK |
5535 | return error; |
5536 | } | |
5537 | ||
549c7297 CB |
5538 | int ext4_getattr(struct user_namespace *mnt_userns, const struct path *path, |
5539 | struct kstat *stat, u32 request_mask, unsigned int query_flags) | |
3e3398a0 | 5540 | { |
99652ea5 DH |
5541 | struct inode *inode = d_inode(path->dentry); |
5542 | struct ext4_inode *raw_inode; | |
5543 | struct ext4_inode_info *ei = EXT4_I(inode); | |
5544 | unsigned int flags; | |
5545 | ||
d4c5e960 TT |
5546 | if ((request_mask & STATX_BTIME) && |
5547 | EXT4_FITS_IN_INODE(raw_inode, ei, i_crtime)) { | |
99652ea5 DH |
5548 | stat->result_mask |= STATX_BTIME; |
5549 | stat->btime.tv_sec = ei->i_crtime.tv_sec; | |
5550 | stat->btime.tv_nsec = ei->i_crtime.tv_nsec; | |
5551 | } | |
5552 | ||
5553 | flags = ei->i_flags & EXT4_FL_USER_VISIBLE; | |
5554 | if (flags & EXT4_APPEND_FL) | |
5555 | stat->attributes |= STATX_ATTR_APPEND; | |
5556 | if (flags & EXT4_COMPR_FL) | |
5557 | stat->attributes |= STATX_ATTR_COMPRESSED; | |
5558 | if (flags & EXT4_ENCRYPT_FL) | |
5559 | stat->attributes |= STATX_ATTR_ENCRYPTED; | |
5560 | if (flags & EXT4_IMMUTABLE_FL) | |
5561 | stat->attributes |= STATX_ATTR_IMMUTABLE; | |
5562 | if (flags & EXT4_NODUMP_FL) | |
5563 | stat->attributes |= STATX_ATTR_NODUMP; | |
1f607195 EB |
5564 | if (flags & EXT4_VERITY_FL) |
5565 | stat->attributes |= STATX_ATTR_VERITY; | |
3e3398a0 | 5566 | |
3209f68b DH |
5567 | stat->attributes_mask |= (STATX_ATTR_APPEND | |
5568 | STATX_ATTR_COMPRESSED | | |
5569 | STATX_ATTR_ENCRYPTED | | |
5570 | STATX_ATTR_IMMUTABLE | | |
1f607195 EB |
5571 | STATX_ATTR_NODUMP | |
5572 | STATX_ATTR_VERITY); | |
3209f68b | 5573 | |
14f3db55 | 5574 | generic_fillattr(mnt_userns, inode, stat); |
99652ea5 DH |
5575 | return 0; |
5576 | } | |
5577 | ||
549c7297 CB |
5578 | int ext4_file_getattr(struct user_namespace *mnt_userns, |
5579 | const struct path *path, struct kstat *stat, | |
99652ea5 DH |
5580 | u32 request_mask, unsigned int query_flags) |
5581 | { | |
5582 | struct inode *inode = d_inode(path->dentry); | |
5583 | u64 delalloc_blocks; | |
5584 | ||
14f3db55 | 5585 | ext4_getattr(mnt_userns, path, stat, request_mask, query_flags); |
3e3398a0 | 5586 | |
9206c561 AD |
5587 | /* |
5588 | * If there is inline data in the inode, the inode will normally not | |
5589 | * have data blocks allocated (it may have an external xattr block). | |
5590 | * Report at least one sector for such files, so tools like tar, rsync, | |
d67d64f4 | 5591 | * others don't incorrectly think the file is completely sparse. |
9206c561 AD |
5592 | */ |
5593 | if (unlikely(ext4_has_inline_data(inode))) | |
5594 | stat->blocks += (stat->size + 511) >> 9; | |
5595 | ||
3e3398a0 MC |
5596 | /* |
5597 | * We can't update i_blocks if the block allocation is delayed | |
5598 | * otherwise in the case of system crash before the real block | |
5599 | * allocation is done, we will have i_blocks inconsistent with | |
5600 | * on-disk file blocks. | |
5601 | * We always keep i_blocks updated together with real | |
5602 | * allocation. But to not confuse with user, stat | |
5603 | * will return the blocks that include the delayed allocation | |
5604 | * blocks for this file. | |
5605 | */ | |
96607551 | 5606 | delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb), |
9206c561 AD |
5607 | EXT4_I(inode)->i_reserved_data_blocks); |
5608 | stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9); | |
3e3398a0 MC |
5609 | return 0; |
5610 | } | |
ac27a0ec | 5611 | |
fffb2739 JK |
5612 | static int ext4_index_trans_blocks(struct inode *inode, int lblocks, |
5613 | int pextents) | |
a02908f1 | 5614 | { |
12e9b892 | 5615 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) |
fffb2739 JK |
5616 | return ext4_ind_trans_blocks(inode, lblocks); |
5617 | return ext4_ext_index_trans_blocks(inode, pextents); | |
a02908f1 | 5618 | } |
ac51d837 | 5619 | |
ac27a0ec | 5620 | /* |
a02908f1 MC |
5621 | * Account for index blocks, block groups bitmaps and block group |
5622 | * descriptor blocks if modify datablocks and index blocks | |
5623 | * worse case, the indexs blocks spread over different block groups | |
ac27a0ec | 5624 | * |
a02908f1 | 5625 | * If datablocks are discontiguous, they are possible to spread over |
4907cb7b | 5626 | * different block groups too. If they are contiguous, with flexbg, |
a02908f1 | 5627 | * they could still across block group boundary. |
ac27a0ec | 5628 | * |
a02908f1 MC |
5629 | * Also account for superblock, inode, quota and xattr blocks |
5630 | */ | |
dec214d0 | 5631 | static int ext4_meta_trans_blocks(struct inode *inode, int lblocks, |
fffb2739 | 5632 | int pextents) |
a02908f1 | 5633 | { |
8df9675f TT |
5634 | ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb); |
5635 | int gdpblocks; | |
a02908f1 MC |
5636 | int idxblocks; |
5637 | int ret = 0; | |
5638 | ||
5639 | /* | |
fffb2739 JK |
5640 | * How many index blocks need to touch to map @lblocks logical blocks |
5641 | * to @pextents physical extents? | |
a02908f1 | 5642 | */ |
fffb2739 | 5643 | idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents); |
a02908f1 MC |
5644 | |
5645 | ret = idxblocks; | |
5646 | ||
5647 | /* | |
5648 | * Now let's see how many group bitmaps and group descriptors need | |
5649 | * to account | |
5650 | */ | |
fffb2739 | 5651 | groups = idxblocks + pextents; |
a02908f1 | 5652 | gdpblocks = groups; |
8df9675f TT |
5653 | if (groups > ngroups) |
5654 | groups = ngroups; | |
a02908f1 MC |
5655 | if (groups > EXT4_SB(inode->i_sb)->s_gdb_count) |
5656 | gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count; | |
5657 | ||
5658 | /* bitmaps and block group descriptor blocks */ | |
5659 | ret += groups + gdpblocks; | |
5660 | ||
5661 | /* Blocks for super block, inode, quota and xattr blocks */ | |
5662 | ret += EXT4_META_TRANS_BLOCKS(inode->i_sb); | |
5663 | ||
5664 | return ret; | |
5665 | } | |
5666 | ||
5667 | /* | |
25985edc | 5668 | * Calculate the total number of credits to reserve to fit |
f3bd1f3f MC |
5669 | * the modification of a single pages into a single transaction, |
5670 | * which may include multiple chunks of block allocations. | |
ac27a0ec | 5671 | * |
525f4ed8 | 5672 | * This could be called via ext4_write_begin() |
ac27a0ec | 5673 | * |
525f4ed8 | 5674 | * We need to consider the worse case, when |
a02908f1 | 5675 | * one new block per extent. |
ac27a0ec | 5676 | */ |
a86c6181 | 5677 | int ext4_writepage_trans_blocks(struct inode *inode) |
ac27a0ec | 5678 | { |
617ba13b | 5679 | int bpp = ext4_journal_blocks_per_page(inode); |
ac27a0ec DK |
5680 | int ret; |
5681 | ||
fffb2739 | 5682 | ret = ext4_meta_trans_blocks(inode, bpp, bpp); |
a86c6181 | 5683 | |
a02908f1 | 5684 | /* Account for data blocks for journalled mode */ |
617ba13b | 5685 | if (ext4_should_journal_data(inode)) |
a02908f1 | 5686 | ret += bpp; |
ac27a0ec DK |
5687 | return ret; |
5688 | } | |
f3bd1f3f MC |
5689 | |
5690 | /* | |
5691 | * Calculate the journal credits for a chunk of data modification. | |
5692 | * | |
5693 | * This is called from DIO, fallocate or whoever calling | |
79e83036 | 5694 | * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks. |
f3bd1f3f MC |
5695 | * |
5696 | * journal buffers for data blocks are not included here, as DIO | |
5697 | * and fallocate do no need to journal data buffers. | |
5698 | */ | |
5699 | int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks) | |
5700 | { | |
5701 | return ext4_meta_trans_blocks(inode, nrblocks, 1); | |
5702 | } | |
5703 | ||
ac27a0ec | 5704 | /* |
617ba13b | 5705 | * The caller must have previously called ext4_reserve_inode_write(). |
ac27a0ec DK |
5706 | * Give this, we know that the caller already has write access to iloc->bh. |
5707 | */ | |
617ba13b | 5708 | int ext4_mark_iloc_dirty(handle_t *handle, |
de9a55b8 | 5709 | struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec DK |
5710 | { |
5711 | int err = 0; | |
5712 | ||
a6758309 VA |
5713 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) { |
5714 | put_bh(iloc->bh); | |
0db1ff22 | 5715 | return -EIO; |
a6758309 | 5716 | } |
a80f7fcf | 5717 | ext4_fc_track_inode(handle, inode); |
aa75f4d3 | 5718 | |
c64db50e | 5719 | if (IS_I_VERSION(inode)) |
25ec56b5 JNC |
5720 | inode_inc_iversion(inode); |
5721 | ||
ac27a0ec DK |
5722 | /* the do_update_inode consumes one bh->b_count */ |
5723 | get_bh(iloc->bh); | |
5724 | ||
dab291af | 5725 | /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */ |
830156c7 | 5726 | err = ext4_do_update_inode(handle, inode, iloc); |
ac27a0ec DK |
5727 | put_bh(iloc->bh); |
5728 | return err; | |
5729 | } | |
5730 | ||
5731 | /* | |
5732 | * On success, We end up with an outstanding reference count against | |
5733 | * iloc->bh. This _must_ be cleaned up later. | |
5734 | */ | |
5735 | ||
5736 | int | |
617ba13b MC |
5737 | ext4_reserve_inode_write(handle_t *handle, struct inode *inode, |
5738 | struct ext4_iloc *iloc) | |
ac27a0ec | 5739 | { |
0390131b FM |
5740 | int err; |
5741 | ||
0db1ff22 TT |
5742 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
5743 | return -EIO; | |
5744 | ||
0390131b FM |
5745 | err = ext4_get_inode_loc(inode, iloc); |
5746 | if (!err) { | |
5747 | BUFFER_TRACE(iloc->bh, "get_write_access"); | |
5748 | err = ext4_journal_get_write_access(handle, iloc->bh); | |
5749 | if (err) { | |
5750 | brelse(iloc->bh); | |
5751 | iloc->bh = NULL; | |
ac27a0ec DK |
5752 | } |
5753 | } | |
617ba13b | 5754 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
5755 | return err; |
5756 | } | |
5757 | ||
c03b45b8 MX |
5758 | static int __ext4_expand_extra_isize(struct inode *inode, |
5759 | unsigned int new_extra_isize, | |
5760 | struct ext4_iloc *iloc, | |
5761 | handle_t *handle, int *no_expand) | |
5762 | { | |
5763 | struct ext4_inode *raw_inode; | |
5764 | struct ext4_xattr_ibody_header *header; | |
4ea99936 TT |
5765 | unsigned int inode_size = EXT4_INODE_SIZE(inode->i_sb); |
5766 | struct ext4_inode_info *ei = EXT4_I(inode); | |
c03b45b8 MX |
5767 | int error; |
5768 | ||
4ea99936 TT |
5769 | /* this was checked at iget time, but double check for good measure */ |
5770 | if ((EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > inode_size) || | |
5771 | (ei->i_extra_isize & 3)) { | |
5772 | EXT4_ERROR_INODE(inode, "bad extra_isize %u (inode size %u)", | |
5773 | ei->i_extra_isize, | |
5774 | EXT4_INODE_SIZE(inode->i_sb)); | |
5775 | return -EFSCORRUPTED; | |
5776 | } | |
5777 | if ((new_extra_isize < ei->i_extra_isize) || | |
5778 | (new_extra_isize < 4) || | |
5779 | (new_extra_isize > inode_size - EXT4_GOOD_OLD_INODE_SIZE)) | |
5780 | return -EINVAL; /* Should never happen */ | |
5781 | ||
c03b45b8 MX |
5782 | raw_inode = ext4_raw_inode(iloc); |
5783 | ||
5784 | header = IHDR(inode, raw_inode); | |
5785 | ||
5786 | /* No extended attributes present */ | |
5787 | if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) || | |
5788 | header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) { | |
5789 | memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE + | |
5790 | EXT4_I(inode)->i_extra_isize, 0, | |
5791 | new_extra_isize - EXT4_I(inode)->i_extra_isize); | |
5792 | EXT4_I(inode)->i_extra_isize = new_extra_isize; | |
5793 | return 0; | |
5794 | } | |
5795 | ||
5796 | /* try to expand with EAs present */ | |
5797 | error = ext4_expand_extra_isize_ea(inode, new_extra_isize, | |
5798 | raw_inode, handle); | |
5799 | if (error) { | |
5800 | /* | |
5801 | * Inode size expansion failed; don't try again | |
5802 | */ | |
5803 | *no_expand = 1; | |
5804 | } | |
5805 | ||
5806 | return error; | |
5807 | } | |
5808 | ||
6dd4ee7c KS |
5809 | /* |
5810 | * Expand an inode by new_extra_isize bytes. | |
5811 | * Returns 0 on success or negative error number on failure. | |
5812 | */ | |
cf0a5e81 MX |
5813 | static int ext4_try_to_expand_extra_isize(struct inode *inode, |
5814 | unsigned int new_extra_isize, | |
5815 | struct ext4_iloc iloc, | |
5816 | handle_t *handle) | |
6dd4ee7c | 5817 | { |
3b10fdc6 MX |
5818 | int no_expand; |
5819 | int error; | |
6dd4ee7c | 5820 | |
cf0a5e81 MX |
5821 | if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) |
5822 | return -EOVERFLOW; | |
5823 | ||
5824 | /* | |
5825 | * In nojournal mode, we can immediately attempt to expand | |
5826 | * the inode. When journaled, we first need to obtain extra | |
5827 | * buffer credits since we may write into the EA block | |
5828 | * with this same handle. If journal_extend fails, then it will | |
5829 | * only result in a minor loss of functionality for that inode. | |
5830 | * If this is felt to be critical, then e2fsck should be run to | |
5831 | * force a large enough s_min_extra_isize. | |
5832 | */ | |
6cb367c2 | 5833 | if (ext4_journal_extend(handle, |
83448bdf | 5834 | EXT4_DATA_TRANS_BLOCKS(inode->i_sb), 0) != 0) |
cf0a5e81 | 5835 | return -ENOSPC; |
6dd4ee7c | 5836 | |
3b10fdc6 | 5837 | if (ext4_write_trylock_xattr(inode, &no_expand) == 0) |
cf0a5e81 | 5838 | return -EBUSY; |
3b10fdc6 | 5839 | |
c03b45b8 MX |
5840 | error = __ext4_expand_extra_isize(inode, new_extra_isize, &iloc, |
5841 | handle, &no_expand); | |
5842 | ext4_write_unlock_xattr(inode, &no_expand); | |
6dd4ee7c | 5843 | |
c03b45b8 MX |
5844 | return error; |
5845 | } | |
6dd4ee7c | 5846 | |
c03b45b8 MX |
5847 | int ext4_expand_extra_isize(struct inode *inode, |
5848 | unsigned int new_extra_isize, | |
5849 | struct ext4_iloc *iloc) | |
5850 | { | |
5851 | handle_t *handle; | |
5852 | int no_expand; | |
5853 | int error, rc; | |
5854 | ||
5855 | if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) { | |
5856 | brelse(iloc->bh); | |
5857 | return -EOVERFLOW; | |
6dd4ee7c KS |
5858 | } |
5859 | ||
c03b45b8 MX |
5860 | handle = ext4_journal_start(inode, EXT4_HT_INODE, |
5861 | EXT4_DATA_TRANS_BLOCKS(inode->i_sb)); | |
5862 | if (IS_ERR(handle)) { | |
5863 | error = PTR_ERR(handle); | |
5864 | brelse(iloc->bh); | |
5865 | return error; | |
5866 | } | |
5867 | ||
5868 | ext4_write_lock_xattr(inode, &no_expand); | |
5869 | ||
ddccb6db | 5870 | BUFFER_TRACE(iloc->bh, "get_write_access"); |
c03b45b8 | 5871 | error = ext4_journal_get_write_access(handle, iloc->bh); |
3b10fdc6 | 5872 | if (error) { |
c03b45b8 | 5873 | brelse(iloc->bh); |
7f420d64 | 5874 | goto out_unlock; |
3b10fdc6 | 5875 | } |
cf0a5e81 | 5876 | |
c03b45b8 MX |
5877 | error = __ext4_expand_extra_isize(inode, new_extra_isize, iloc, |
5878 | handle, &no_expand); | |
5879 | ||
5880 | rc = ext4_mark_iloc_dirty(handle, inode, iloc); | |
5881 | if (!error) | |
5882 | error = rc; | |
5883 | ||
7f420d64 | 5884 | out_unlock: |
c03b45b8 | 5885 | ext4_write_unlock_xattr(inode, &no_expand); |
c03b45b8 | 5886 | ext4_journal_stop(handle); |
3b10fdc6 | 5887 | return error; |
6dd4ee7c KS |
5888 | } |
5889 | ||
ac27a0ec DK |
5890 | /* |
5891 | * What we do here is to mark the in-core inode as clean with respect to inode | |
5892 | * dirtiness (it may still be data-dirty). | |
5893 | * This means that the in-core inode may be reaped by prune_icache | |
5894 | * without having to perform any I/O. This is a very good thing, | |
5895 | * because *any* task may call prune_icache - even ones which | |
5896 | * have a transaction open against a different journal. | |
5897 | * | |
5898 | * Is this cheating? Not really. Sure, we haven't written the | |
5899 | * inode out, but prune_icache isn't a user-visible syncing function. | |
5900 | * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync) | |
5901 | * we start and wait on commits. | |
ac27a0ec | 5902 | */ |
4209ae12 HS |
5903 | int __ext4_mark_inode_dirty(handle_t *handle, struct inode *inode, |
5904 | const char *func, unsigned int line) | |
ac27a0ec | 5905 | { |
617ba13b | 5906 | struct ext4_iloc iloc; |
6dd4ee7c | 5907 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
cf0a5e81 | 5908 | int err; |
ac27a0ec DK |
5909 | |
5910 | might_sleep(); | |
7ff9c073 | 5911 | trace_ext4_mark_inode_dirty(inode, _RET_IP_); |
617ba13b | 5912 | err = ext4_reserve_inode_write(handle, inode, &iloc); |
5e1021f2 | 5913 | if (err) |
4209ae12 | 5914 | goto out; |
cf0a5e81 MX |
5915 | |
5916 | if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize) | |
5917 | ext4_try_to_expand_extra_isize(inode, sbi->s_want_extra_isize, | |
5918 | iloc, handle); | |
5919 | ||
4209ae12 HS |
5920 | err = ext4_mark_iloc_dirty(handle, inode, &iloc); |
5921 | out: | |
5922 | if (unlikely(err)) | |
5923 | ext4_error_inode_err(inode, func, line, 0, err, | |
5924 | "mark_inode_dirty error"); | |
5925 | return err; | |
ac27a0ec DK |
5926 | } |
5927 | ||
5928 | /* | |
617ba13b | 5929 | * ext4_dirty_inode() is called from __mark_inode_dirty() |
ac27a0ec DK |
5930 | * |
5931 | * We're really interested in the case where a file is being extended. | |
5932 | * i_size has been changed by generic_commit_write() and we thus need | |
5933 | * to include the updated inode in the current transaction. | |
5934 | * | |
5dd4056d | 5935 | * Also, dquot_alloc_block() will always dirty the inode when blocks |
ac27a0ec DK |
5936 | * are allocated to the file. |
5937 | * | |
5938 | * If the inode is marked synchronous, we don't honour that here - doing | |
5939 | * so would cause a commit on atime updates, which we don't bother doing. | |
5940 | * We handle synchronous inodes at the highest possible level. | |
5941 | */ | |
aa385729 | 5942 | void ext4_dirty_inode(struct inode *inode, int flags) |
ac27a0ec | 5943 | { |
ac27a0ec DK |
5944 | handle_t *handle; |
5945 | ||
9924a92a | 5946 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 2); |
ac27a0ec | 5947 | if (IS_ERR(handle)) |
e2728c56 | 5948 | return; |
f3dc272f | 5949 | ext4_mark_inode_dirty(handle, inode); |
617ba13b | 5950 | ext4_journal_stop(handle); |
ac27a0ec DK |
5951 | } |
5952 | ||
617ba13b | 5953 | int ext4_change_inode_journal_flag(struct inode *inode, int val) |
ac27a0ec DK |
5954 | { |
5955 | journal_t *journal; | |
5956 | handle_t *handle; | |
5957 | int err; | |
c8585c6f | 5958 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
ac27a0ec DK |
5959 | |
5960 | /* | |
5961 | * We have to be very careful here: changing a data block's | |
5962 | * journaling status dynamically is dangerous. If we write a | |
5963 | * data block to the journal, change the status and then delete | |
5964 | * that block, we risk forgetting to revoke the old log record | |
5965 | * from the journal and so a subsequent replay can corrupt data. | |
5966 | * So, first we make sure that the journal is empty and that | |
5967 | * nobody is changing anything. | |
5968 | */ | |
5969 | ||
617ba13b | 5970 | journal = EXT4_JOURNAL(inode); |
0390131b FM |
5971 | if (!journal) |
5972 | return 0; | |
d699594d | 5973 | if (is_journal_aborted(journal)) |
ac27a0ec DK |
5974 | return -EROFS; |
5975 | ||
17335dcc | 5976 | /* Wait for all existing dio workers */ |
17335dcc DM |
5977 | inode_dio_wait(inode); |
5978 | ||
4c546592 DJ |
5979 | /* |
5980 | * Before flushing the journal and switching inode's aops, we have | |
5981 | * to flush all dirty data the inode has. There can be outstanding | |
5982 | * delayed allocations, there can be unwritten extents created by | |
5983 | * fallocate or buffered writes in dioread_nolock mode covered by | |
5984 | * dirty data which can be converted only after flushing the dirty | |
5985 | * data (and journalled aops don't know how to handle these cases). | |
5986 | */ | |
5987 | if (val) { | |
5988 | down_write(&EXT4_I(inode)->i_mmap_sem); | |
5989 | err = filemap_write_and_wait(inode->i_mapping); | |
5990 | if (err < 0) { | |
5991 | up_write(&EXT4_I(inode)->i_mmap_sem); | |
4c546592 DJ |
5992 | return err; |
5993 | } | |
5994 | } | |
5995 | ||
bbd55937 | 5996 | percpu_down_write(&sbi->s_writepages_rwsem); |
dab291af | 5997 | jbd2_journal_lock_updates(journal); |
ac27a0ec DK |
5998 | |
5999 | /* | |
6000 | * OK, there are no updates running now, and all cached data is | |
6001 | * synced to disk. We are now in a completely consistent state | |
6002 | * which doesn't have anything in the journal, and we know that | |
6003 | * no filesystem updates are running, so it is safe to modify | |
6004 | * the inode's in-core data-journaling state flag now. | |
6005 | */ | |
6006 | ||
6007 | if (val) | |
12e9b892 | 6008 | ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
5872ddaa | 6009 | else { |
4f879ca6 JK |
6010 | err = jbd2_journal_flush(journal); |
6011 | if (err < 0) { | |
6012 | jbd2_journal_unlock_updates(journal); | |
bbd55937 | 6013 | percpu_up_write(&sbi->s_writepages_rwsem); |
4f879ca6 JK |
6014 | return err; |
6015 | } | |
12e9b892 | 6016 | ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
5872ddaa | 6017 | } |
617ba13b | 6018 | ext4_set_aops(inode); |
ac27a0ec | 6019 | |
dab291af | 6020 | jbd2_journal_unlock_updates(journal); |
bbd55937 | 6021 | percpu_up_write(&sbi->s_writepages_rwsem); |
c8585c6f | 6022 | |
4c546592 DJ |
6023 | if (val) |
6024 | up_write(&EXT4_I(inode)->i_mmap_sem); | |
ac27a0ec DK |
6025 | |
6026 | /* Finally we can mark the inode as dirty. */ | |
6027 | ||
9924a92a | 6028 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 1); |
ac27a0ec DK |
6029 | if (IS_ERR(handle)) |
6030 | return PTR_ERR(handle); | |
6031 | ||
aa75f4d3 HS |
6032 | ext4_fc_mark_ineligible(inode->i_sb, |
6033 | EXT4_FC_REASON_JOURNAL_FLAG_CHANGE); | |
617ba13b | 6034 | err = ext4_mark_inode_dirty(handle, inode); |
0390131b | 6035 | ext4_handle_sync(handle); |
617ba13b MC |
6036 | ext4_journal_stop(handle); |
6037 | ext4_std_error(inode->i_sb, err); | |
ac27a0ec DK |
6038 | |
6039 | return err; | |
6040 | } | |
2e9ee850 AK |
6041 | |
6042 | static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh) | |
6043 | { | |
6044 | return !buffer_mapped(bh); | |
6045 | } | |
6046 | ||
401b25aa | 6047 | vm_fault_t ext4_page_mkwrite(struct vm_fault *vmf) |
2e9ee850 | 6048 | { |
11bac800 | 6049 | struct vm_area_struct *vma = vmf->vma; |
c2ec175c | 6050 | struct page *page = vmf->page; |
2e9ee850 AK |
6051 | loff_t size; |
6052 | unsigned long len; | |
401b25aa SJ |
6053 | int err; |
6054 | vm_fault_t ret; | |
2e9ee850 | 6055 | struct file *file = vma->vm_file; |
496ad9aa | 6056 | struct inode *inode = file_inode(file); |
2e9ee850 | 6057 | struct address_space *mapping = inode->i_mapping; |
9ea7df53 JK |
6058 | handle_t *handle; |
6059 | get_block_t *get_block; | |
6060 | int retries = 0; | |
2e9ee850 | 6061 | |
02b016ca TT |
6062 | if (unlikely(IS_IMMUTABLE(inode))) |
6063 | return VM_FAULT_SIGBUS; | |
6064 | ||
8e8ad8a5 | 6065 | sb_start_pagefault(inode->i_sb); |
041bbb6d | 6066 | file_update_time(vma->vm_file); |
ea3d7209 JK |
6067 | |
6068 | down_read(&EXT4_I(inode)->i_mmap_sem); | |
7b4cc978 | 6069 | |
401b25aa SJ |
6070 | err = ext4_convert_inline_data(inode); |
6071 | if (err) | |
7b4cc978 EB |
6072 | goto out_ret; |
6073 | ||
64a9f144 MFO |
6074 | /* |
6075 | * On data journalling we skip straight to the transaction handle: | |
6076 | * there's no delalloc; page truncated will be checked later; the | |
6077 | * early return w/ all buffers mapped (calculates size/len) can't | |
6078 | * be used; and there's no dioread_nolock, so only ext4_get_block. | |
6079 | */ | |
6080 | if (ext4_should_journal_data(inode)) | |
6081 | goto retry_alloc; | |
6082 | ||
9ea7df53 JK |
6083 | /* Delalloc case is easy... */ |
6084 | if (test_opt(inode->i_sb, DELALLOC) && | |
9ea7df53 JK |
6085 | !ext4_nonda_switch(inode->i_sb)) { |
6086 | do { | |
401b25aa | 6087 | err = block_page_mkwrite(vma, vmf, |
9ea7df53 | 6088 | ext4_da_get_block_prep); |
401b25aa | 6089 | } while (err == -ENOSPC && |
9ea7df53 JK |
6090 | ext4_should_retry_alloc(inode->i_sb, &retries)); |
6091 | goto out_ret; | |
2e9ee850 | 6092 | } |
0e499890 DW |
6093 | |
6094 | lock_page(page); | |
9ea7df53 JK |
6095 | size = i_size_read(inode); |
6096 | /* Page got truncated from under us? */ | |
6097 | if (page->mapping != mapping || page_offset(page) > size) { | |
6098 | unlock_page(page); | |
6099 | ret = VM_FAULT_NOPAGE; | |
6100 | goto out; | |
0e499890 | 6101 | } |
2e9ee850 | 6102 | |
09cbfeaf KS |
6103 | if (page->index == size >> PAGE_SHIFT) |
6104 | len = size & ~PAGE_MASK; | |
2e9ee850 | 6105 | else |
09cbfeaf | 6106 | len = PAGE_SIZE; |
a827eaff | 6107 | /* |
9ea7df53 JK |
6108 | * Return if we have all the buffers mapped. This avoids the need to do |
6109 | * journal_start/journal_stop which can block and take a long time | |
64a9f144 MFO |
6110 | * |
6111 | * This cannot be done for data journalling, as we have to add the | |
6112 | * inode to the transaction's list to writeprotect pages on commit. | |
a827eaff | 6113 | */ |
2e9ee850 | 6114 | if (page_has_buffers(page)) { |
f19d5870 TM |
6115 | if (!ext4_walk_page_buffers(NULL, page_buffers(page), |
6116 | 0, len, NULL, | |
6117 | ext4_bh_unmapped)) { | |
9ea7df53 | 6118 | /* Wait so that we don't change page under IO */ |
1d1d1a76 | 6119 | wait_for_stable_page(page); |
9ea7df53 JK |
6120 | ret = VM_FAULT_LOCKED; |
6121 | goto out; | |
a827eaff | 6122 | } |
2e9ee850 | 6123 | } |
a827eaff | 6124 | unlock_page(page); |
9ea7df53 JK |
6125 | /* OK, we need to fill the hole... */ |
6126 | if (ext4_should_dioread_nolock(inode)) | |
705965bd | 6127 | get_block = ext4_get_block_unwritten; |
9ea7df53 JK |
6128 | else |
6129 | get_block = ext4_get_block; | |
6130 | retry_alloc: | |
9924a92a TT |
6131 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
6132 | ext4_writepage_trans_blocks(inode)); | |
9ea7df53 | 6133 | if (IS_ERR(handle)) { |
c2ec175c | 6134 | ret = VM_FAULT_SIGBUS; |
9ea7df53 JK |
6135 | goto out; |
6136 | } | |
64a9f144 MFO |
6137 | /* |
6138 | * Data journalling can't use block_page_mkwrite() because it | |
6139 | * will set_buffer_dirty() before do_journal_get_write_access() | |
6140 | * thus might hit warning messages for dirty metadata buffers. | |
6141 | */ | |
6142 | if (!ext4_should_journal_data(inode)) { | |
6143 | err = block_page_mkwrite(vma, vmf, get_block); | |
6144 | } else { | |
6145 | lock_page(page); | |
6146 | size = i_size_read(inode); | |
6147 | /* Page got truncated from under us? */ | |
6148 | if (page->mapping != mapping || page_offset(page) > size) { | |
64a9f144 | 6149 | ret = VM_FAULT_NOPAGE; |
afb585a9 | 6150 | goto out_error; |
9ea7df53 | 6151 | } |
64a9f144 MFO |
6152 | |
6153 | if (page->index == size >> PAGE_SHIFT) | |
6154 | len = size & ~PAGE_MASK; | |
6155 | else | |
6156 | len = PAGE_SIZE; | |
6157 | ||
6158 | err = __block_write_begin(page, 0, len, ext4_get_block); | |
6159 | if (!err) { | |
afb585a9 | 6160 | ret = VM_FAULT_SIGBUS; |
64a9f144 | 6161 | if (ext4_walk_page_buffers(handle, page_buffers(page), |
afb585a9 MFO |
6162 | 0, len, NULL, do_journal_get_write_access)) |
6163 | goto out_error; | |
6164 | if (ext4_walk_page_buffers(handle, page_buffers(page), | |
6165 | 0, len, NULL, write_end_fn)) | |
6166 | goto out_error; | |
b5b18160 JK |
6167 | if (ext4_jbd2_inode_add_write(handle, inode, |
6168 | page_offset(page), len)) | |
afb585a9 | 6169 | goto out_error; |
64a9f144 MFO |
6170 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
6171 | } else { | |
6172 | unlock_page(page); | |
6173 | } | |
9ea7df53 JK |
6174 | } |
6175 | ext4_journal_stop(handle); | |
401b25aa | 6176 | if (err == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) |
9ea7df53 JK |
6177 | goto retry_alloc; |
6178 | out_ret: | |
401b25aa | 6179 | ret = block_page_mkwrite_return(err); |
9ea7df53 | 6180 | out: |
ea3d7209 | 6181 | up_read(&EXT4_I(inode)->i_mmap_sem); |
8e8ad8a5 | 6182 | sb_end_pagefault(inode->i_sb); |
2e9ee850 | 6183 | return ret; |
afb585a9 MFO |
6184 | out_error: |
6185 | unlock_page(page); | |
6186 | ext4_journal_stop(handle); | |
6187 | goto out; | |
2e9ee850 | 6188 | } |
ea3d7209 | 6189 | |
401b25aa | 6190 | vm_fault_t ext4_filemap_fault(struct vm_fault *vmf) |
ea3d7209 | 6191 | { |
11bac800 | 6192 | struct inode *inode = file_inode(vmf->vma->vm_file); |
401b25aa | 6193 | vm_fault_t ret; |
ea3d7209 JK |
6194 | |
6195 | down_read(&EXT4_I(inode)->i_mmap_sem); | |
401b25aa | 6196 | ret = filemap_fault(vmf); |
ea3d7209 JK |
6197 | up_read(&EXT4_I(inode)->i_mmap_sem); |
6198 | ||
401b25aa | 6199 | return ret; |
ea3d7209 | 6200 | } |