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 DK |
22 | #include <linux/fs.h> |
23 | #include <linux/time.h> | |
ac27a0ec DK |
24 | #include <linux/highuid.h> |
25 | #include <linux/pagemap.h> | |
c94c2acf | 26 | #include <linux/dax.h> |
ac27a0ec DK |
27 | #include <linux/quotaops.h> |
28 | #include <linux/string.h> | |
29 | #include <linux/buffer_head.h> | |
30 | #include <linux/writeback.h> | |
64769240 | 31 | #include <linux/pagevec.h> |
ac27a0ec | 32 | #include <linux/mpage.h> |
e83c1397 | 33 | #include <linux/namei.h> |
ac27a0ec DK |
34 | #include <linux/uio.h> |
35 | #include <linux/bio.h> | |
4c0425ff | 36 | #include <linux/workqueue.h> |
744692dc | 37 | #include <linux/kernel.h> |
6db26ffc | 38 | #include <linux/printk.h> |
5a0e3ad6 | 39 | #include <linux/slab.h> |
00a1a053 | 40 | #include <linux/bitops.h> |
364443cb | 41 | #include <linux/iomap.h> |
ae5e165d | 42 | #include <linux/iversion.h> |
9bffad1e | 43 | |
3dcf5451 | 44 | #include "ext4_jbd2.h" |
ac27a0ec DK |
45 | #include "xattr.h" |
46 | #include "acl.h" | |
9f125d64 | 47 | #include "truncate.h" |
ac27a0ec | 48 | |
9bffad1e TT |
49 | #include <trace/events/ext4.h> |
50 | ||
814525f4 DW |
51 | static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw, |
52 | struct ext4_inode_info *ei) | |
53 | { | |
54 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
814525f4 | 55 | __u32 csum; |
b47820ed DJ |
56 | __u16 dummy_csum = 0; |
57 | int offset = offsetof(struct ext4_inode, i_checksum_lo); | |
58 | unsigned int csum_size = sizeof(dummy_csum); | |
814525f4 | 59 | |
b47820ed DJ |
60 | csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw, offset); |
61 | csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, csum_size); | |
62 | offset += csum_size; | |
63 | csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset, | |
64 | EXT4_GOOD_OLD_INODE_SIZE - offset); | |
814525f4 | 65 | |
b47820ed DJ |
66 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
67 | offset = offsetof(struct ext4_inode, i_checksum_hi); | |
68 | csum = ext4_chksum(sbi, csum, (__u8 *)raw + | |
69 | EXT4_GOOD_OLD_INODE_SIZE, | |
70 | offset - EXT4_GOOD_OLD_INODE_SIZE); | |
71 | if (EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) { | |
72 | csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, | |
73 | csum_size); | |
74 | offset += csum_size; | |
b47820ed | 75 | } |
05ac5aa1 DJ |
76 | csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset, |
77 | EXT4_INODE_SIZE(inode->i_sb) - offset); | |
814525f4 DW |
78 | } |
79 | ||
814525f4 DW |
80 | return csum; |
81 | } | |
82 | ||
83 | static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw, | |
84 | struct ext4_inode_info *ei) | |
85 | { | |
86 | __u32 provided, calculated; | |
87 | ||
88 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != | |
89 | cpu_to_le32(EXT4_OS_LINUX) || | |
9aa5d32b | 90 | !ext4_has_metadata_csum(inode->i_sb)) |
814525f4 DW |
91 | return 1; |
92 | ||
93 | provided = le16_to_cpu(raw->i_checksum_lo); | |
94 | calculated = ext4_inode_csum(inode, raw, ei); | |
95 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && | |
96 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) | |
97 | provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16; | |
98 | else | |
99 | calculated &= 0xFFFF; | |
100 | ||
101 | return provided == calculated; | |
102 | } | |
103 | ||
104 | static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw, | |
105 | struct ext4_inode_info *ei) | |
106 | { | |
107 | __u32 csum; | |
108 | ||
109 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != | |
110 | cpu_to_le32(EXT4_OS_LINUX) || | |
9aa5d32b | 111 | !ext4_has_metadata_csum(inode->i_sb)) |
814525f4 DW |
112 | return; |
113 | ||
114 | csum = ext4_inode_csum(inode, raw, ei); | |
115 | raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF); | |
116 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && | |
117 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) | |
118 | raw->i_checksum_hi = cpu_to_le16(csum >> 16); | |
119 | } | |
120 | ||
678aaf48 JK |
121 | static inline int ext4_begin_ordered_truncate(struct inode *inode, |
122 | loff_t new_size) | |
123 | { | |
7ff9c073 | 124 | trace_ext4_begin_ordered_truncate(inode, new_size); |
8aefcd55 TT |
125 | /* |
126 | * If jinode is zero, then we never opened the file for | |
127 | * writing, so there's no need to call | |
128 | * jbd2_journal_begin_ordered_truncate() since there's no | |
129 | * outstanding writes we need to flush. | |
130 | */ | |
131 | if (!EXT4_I(inode)->jinode) | |
132 | return 0; | |
133 | return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode), | |
134 | EXT4_I(inode)->jinode, | |
135 | new_size); | |
678aaf48 JK |
136 | } |
137 | ||
d47992f8 LC |
138 | static void ext4_invalidatepage(struct page *page, unsigned int offset, |
139 | unsigned int length); | |
cb20d518 TT |
140 | static int __ext4_journalled_writepage(struct page *page, unsigned int len); |
141 | static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh); | |
dec214d0 TE |
142 | static int ext4_meta_trans_blocks(struct inode *inode, int lblocks, |
143 | int pextents); | |
64769240 | 144 | |
ac27a0ec DK |
145 | /* |
146 | * Test whether an inode is a fast symlink. | |
407cd7fb | 147 | * A fast symlink has its symlink data stored in ext4_inode_info->i_data. |
ac27a0ec | 148 | */ |
f348c252 | 149 | int ext4_inode_is_fast_symlink(struct inode *inode) |
ac27a0ec | 150 | { |
fc82228a AK |
151 | if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) { |
152 | int ea_blocks = EXT4_I(inode)->i_file_acl ? | |
153 | EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0; | |
154 | ||
155 | if (ext4_has_inline_data(inode)) | |
156 | return 0; | |
157 | ||
158 | return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0); | |
159 | } | |
407cd7fb TE |
160 | return S_ISLNK(inode->i_mode) && inode->i_size && |
161 | (inode->i_size < EXT4_N_BLOCKS * 4); | |
ac27a0ec DK |
162 | } |
163 | ||
ac27a0ec DK |
164 | /* |
165 | * Called at the last iput() if i_nlink is zero. | |
166 | */ | |
0930fcc1 | 167 | void ext4_evict_inode(struct inode *inode) |
ac27a0ec DK |
168 | { |
169 | handle_t *handle; | |
bc965ab3 | 170 | int err; |
65db869c JK |
171 | /* |
172 | * Credits for final inode cleanup and freeing: | |
173 | * sb + inode (ext4_orphan_del()), block bitmap, group descriptor | |
174 | * (xattr block freeing), bitmap, group descriptor (inode freeing) | |
175 | */ | |
176 | int extra_credits = 6; | |
0421a189 | 177 | struct ext4_xattr_inode_array *ea_inode_array = NULL; |
ac27a0ec | 178 | |
7ff9c073 | 179 | trace_ext4_evict_inode(inode); |
2581fdc8 | 180 | |
0930fcc1 | 181 | if (inode->i_nlink) { |
2d859db3 JK |
182 | /* |
183 | * When journalling data dirty buffers are tracked only in the | |
184 | * journal. So although mm thinks everything is clean and | |
185 | * ready for reaping the inode might still have some pages to | |
186 | * write in the running transaction or waiting to be | |
187 | * checkpointed. Thus calling jbd2_journal_invalidatepage() | |
188 | * (via truncate_inode_pages()) to discard these buffers can | |
189 | * cause data loss. Also even if we did not discard these | |
190 | * buffers, we would have no way to find them after the inode | |
191 | * is reaped and thus user could see stale data if he tries to | |
192 | * read them before the transaction is checkpointed. So be | |
193 | * careful and force everything to disk here... We use | |
194 | * ei->i_datasync_tid to store the newest transaction | |
195 | * containing inode's data. | |
196 | * | |
197 | * Note that directories do not have this problem because they | |
198 | * don't use page cache. | |
199 | */ | |
6a7fd522 VN |
200 | if (inode->i_ino != EXT4_JOURNAL_INO && |
201 | ext4_should_journal_data(inode) && | |
3abb1a0f JK |
202 | (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) && |
203 | inode->i_data.nrpages) { | |
2d859db3 JK |
204 | journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; |
205 | tid_t commit_tid = EXT4_I(inode)->i_datasync_tid; | |
206 | ||
d76a3a77 | 207 | jbd2_complete_transaction(journal, commit_tid); |
2d859db3 JK |
208 | filemap_write_and_wait(&inode->i_data); |
209 | } | |
91b0abe3 | 210 | truncate_inode_pages_final(&inode->i_data); |
5dc23bdd | 211 | |
0930fcc1 AV |
212 | goto no_delete; |
213 | } | |
214 | ||
e2bfb088 TT |
215 | if (is_bad_inode(inode)) |
216 | goto no_delete; | |
217 | dquot_initialize(inode); | |
907f4554 | 218 | |
678aaf48 JK |
219 | if (ext4_should_order_data(inode)) |
220 | ext4_begin_ordered_truncate(inode, 0); | |
91b0abe3 | 221 | truncate_inode_pages_final(&inode->i_data); |
ac27a0ec | 222 | |
8e8ad8a5 JK |
223 | /* |
224 | * Protect us against freezing - iput() caller didn't have to have any | |
225 | * protection against it | |
226 | */ | |
227 | sb_start_intwrite(inode->i_sb); | |
e50e5129 | 228 | |
30a7eb97 TE |
229 | if (!IS_NOQUOTA(inode)) |
230 | extra_credits += EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb); | |
231 | ||
65db869c JK |
232 | /* |
233 | * Block bitmap, group descriptor, and inode are accounted in both | |
234 | * ext4_blocks_for_truncate() and extra_credits. So subtract 3. | |
235 | */ | |
30a7eb97 | 236 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, |
65db869c | 237 | ext4_blocks_for_truncate(inode) + extra_credits - 3); |
ac27a0ec | 238 | if (IS_ERR(handle)) { |
bc965ab3 | 239 | ext4_std_error(inode->i_sb, PTR_ERR(handle)); |
ac27a0ec DK |
240 | /* |
241 | * If we're going to skip the normal cleanup, we still need to | |
242 | * make sure that the in-core orphan linked list is properly | |
243 | * cleaned up. | |
244 | */ | |
617ba13b | 245 | ext4_orphan_del(NULL, inode); |
8e8ad8a5 | 246 | sb_end_intwrite(inode->i_sb); |
ac27a0ec DK |
247 | goto no_delete; |
248 | } | |
30a7eb97 | 249 | |
ac27a0ec | 250 | if (IS_SYNC(inode)) |
0390131b | 251 | ext4_handle_sync(handle); |
407cd7fb TE |
252 | |
253 | /* | |
254 | * Set inode->i_size to 0 before calling ext4_truncate(). We need | |
255 | * special handling of symlinks here because i_size is used to | |
256 | * determine whether ext4_inode_info->i_data contains symlink data or | |
257 | * block mappings. Setting i_size to 0 will remove its fast symlink | |
258 | * status. Erase i_data so that it becomes a valid empty block map. | |
259 | */ | |
260 | if (ext4_inode_is_fast_symlink(inode)) | |
261 | memset(EXT4_I(inode)->i_data, 0, sizeof(EXT4_I(inode)->i_data)); | |
ac27a0ec | 262 | inode->i_size = 0; |
bc965ab3 TT |
263 | err = ext4_mark_inode_dirty(handle, inode); |
264 | if (err) { | |
12062ddd | 265 | ext4_warning(inode->i_sb, |
bc965ab3 TT |
266 | "couldn't mark inode dirty (err %d)", err); |
267 | goto stop_handle; | |
268 | } | |
2c98eb5e TT |
269 | if (inode->i_blocks) { |
270 | err = ext4_truncate(inode); | |
271 | if (err) { | |
54d3adbc TT |
272 | ext4_error_err(inode->i_sb, -err, |
273 | "couldn't truncate inode %lu (err %d)", | |
274 | inode->i_ino, err); | |
2c98eb5e TT |
275 | goto stop_handle; |
276 | } | |
277 | } | |
bc965ab3 | 278 | |
30a7eb97 TE |
279 | /* Remove xattr references. */ |
280 | err = ext4_xattr_delete_inode(handle, inode, &ea_inode_array, | |
281 | extra_credits); | |
282 | if (err) { | |
283 | ext4_warning(inode->i_sb, "xattr delete (err %d)", err); | |
284 | stop_handle: | |
285 | ext4_journal_stop(handle); | |
286 | ext4_orphan_del(NULL, inode); | |
287 | sb_end_intwrite(inode->i_sb); | |
288 | ext4_xattr_inode_array_free(ea_inode_array); | |
289 | goto no_delete; | |
bc965ab3 TT |
290 | } |
291 | ||
ac27a0ec | 292 | /* |
617ba13b | 293 | * Kill off the orphan record which ext4_truncate created. |
ac27a0ec | 294 | * AKPM: I think this can be inside the above `if'. |
617ba13b | 295 | * Note that ext4_orphan_del() has to be able to cope with the |
ac27a0ec | 296 | * deletion of a non-existent orphan - this is because we don't |
617ba13b | 297 | * know if ext4_truncate() actually created an orphan record. |
ac27a0ec DK |
298 | * (Well, we could do this if we need to, but heck - it works) |
299 | */ | |
617ba13b | 300 | ext4_orphan_del(handle, inode); |
5ffff834 | 301 | EXT4_I(inode)->i_dtime = (__u32)ktime_get_real_seconds(); |
ac27a0ec DK |
302 | |
303 | /* | |
304 | * One subtle ordering requirement: if anything has gone wrong | |
305 | * (transaction abort, IO errors, whatever), then we can still | |
306 | * do these next steps (the fs will already have been marked as | |
307 | * having errors), but we can't free the inode if the mark_dirty | |
308 | * fails. | |
309 | */ | |
617ba13b | 310 | if (ext4_mark_inode_dirty(handle, inode)) |
ac27a0ec | 311 | /* If that failed, just do the required in-core inode clear. */ |
0930fcc1 | 312 | ext4_clear_inode(inode); |
ac27a0ec | 313 | else |
617ba13b MC |
314 | ext4_free_inode(handle, inode); |
315 | ext4_journal_stop(handle); | |
8e8ad8a5 | 316 | sb_end_intwrite(inode->i_sb); |
0421a189 | 317 | ext4_xattr_inode_array_free(ea_inode_array); |
ac27a0ec DK |
318 | return; |
319 | no_delete: | |
0930fcc1 | 320 | ext4_clear_inode(inode); /* We must guarantee clearing of inode... */ |
ac27a0ec DK |
321 | } |
322 | ||
a9e7f447 DM |
323 | #ifdef CONFIG_QUOTA |
324 | qsize_t *ext4_get_reserved_space(struct inode *inode) | |
60e58e0f | 325 | { |
a9e7f447 | 326 | return &EXT4_I(inode)->i_reserved_quota; |
60e58e0f | 327 | } |
a9e7f447 | 328 | #endif |
9d0be502 | 329 | |
0637c6f4 TT |
330 | /* |
331 | * Called with i_data_sem down, which is important since we can call | |
332 | * ext4_discard_preallocations() from here. | |
333 | */ | |
5f634d06 AK |
334 | void ext4_da_update_reserve_space(struct inode *inode, |
335 | int used, int quota_claim) | |
12219aea AK |
336 | { |
337 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0637c6f4 | 338 | struct ext4_inode_info *ei = EXT4_I(inode); |
0637c6f4 TT |
339 | |
340 | spin_lock(&ei->i_block_reservation_lock); | |
d8990240 | 341 | trace_ext4_da_update_reserve_space(inode, used, quota_claim); |
0637c6f4 | 342 | if (unlikely(used > ei->i_reserved_data_blocks)) { |
8de5c325 | 343 | ext4_warning(inode->i_sb, "%s: ino %lu, used %d " |
1084f252 | 344 | "with only %d reserved data blocks", |
0637c6f4 TT |
345 | __func__, inode->i_ino, used, |
346 | ei->i_reserved_data_blocks); | |
347 | WARN_ON(1); | |
348 | used = ei->i_reserved_data_blocks; | |
349 | } | |
12219aea | 350 | |
0637c6f4 TT |
351 | /* Update per-inode reservations */ |
352 | ei->i_reserved_data_blocks -= used; | |
71d4f7d0 | 353 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, used); |
6bc6e63f | 354 | |
12219aea | 355 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
60e58e0f | 356 | |
72b8ab9d ES |
357 | /* Update quota subsystem for data blocks */ |
358 | if (quota_claim) | |
7b415bf6 | 359 | dquot_claim_block(inode, EXT4_C2B(sbi, used)); |
72b8ab9d | 360 | else { |
5f634d06 AK |
361 | /* |
362 | * We did fallocate with an offset that is already delayed | |
363 | * allocated. So on delayed allocated writeback we should | |
72b8ab9d | 364 | * not re-claim the quota for fallocated blocks. |
5f634d06 | 365 | */ |
7b415bf6 | 366 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, used)); |
5f634d06 | 367 | } |
d6014301 AK |
368 | |
369 | /* | |
370 | * If we have done all the pending block allocations and if | |
371 | * there aren't any writers on the inode, we can discard the | |
372 | * inode's preallocations. | |
373 | */ | |
0637c6f4 | 374 | if ((ei->i_reserved_data_blocks == 0) && |
82dd124c | 375 | !inode_is_open_for_write(inode)) |
d6014301 | 376 | ext4_discard_preallocations(inode); |
12219aea AK |
377 | } |
378 | ||
e29136f8 | 379 | static int __check_block_validity(struct inode *inode, const char *func, |
c398eda0 TT |
380 | unsigned int line, |
381 | struct ext4_map_blocks *map) | |
6fd058f7 | 382 | { |
345c0dbf TT |
383 | if (ext4_has_feature_journal(inode->i_sb) && |
384 | (inode->i_ino == | |
385 | le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_journal_inum))) | |
386 | return 0; | |
24676da4 TT |
387 | if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk, |
388 | map->m_len)) { | |
c398eda0 | 389 | ext4_error_inode(inode, func, line, map->m_pblk, |
bdbd6ce0 | 390 | "lblock %lu mapped to illegal pblock %llu " |
c398eda0 | 391 | "(length %d)", (unsigned long) map->m_lblk, |
bdbd6ce0 | 392 | map->m_pblk, map->m_len); |
6a797d27 | 393 | return -EFSCORRUPTED; |
6fd058f7 TT |
394 | } |
395 | return 0; | |
396 | } | |
397 | ||
53085fac JK |
398 | int ext4_issue_zeroout(struct inode *inode, ext4_lblk_t lblk, ext4_fsblk_t pblk, |
399 | ext4_lblk_t len) | |
400 | { | |
401 | int ret; | |
402 | ||
33b4cc25 | 403 | if (IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode)) |
a7550b30 | 404 | return fscrypt_zeroout_range(inode, lblk, pblk, len); |
53085fac JK |
405 | |
406 | ret = sb_issue_zeroout(inode->i_sb, pblk, len, GFP_NOFS); | |
407 | if (ret > 0) | |
408 | ret = 0; | |
409 | ||
410 | return ret; | |
411 | } | |
412 | ||
e29136f8 | 413 | #define check_block_validity(inode, map) \ |
c398eda0 | 414 | __check_block_validity((inode), __func__, __LINE__, (map)) |
e29136f8 | 415 | |
921f266b DM |
416 | #ifdef ES_AGGRESSIVE_TEST |
417 | static void ext4_map_blocks_es_recheck(handle_t *handle, | |
418 | struct inode *inode, | |
419 | struct ext4_map_blocks *es_map, | |
420 | struct ext4_map_blocks *map, | |
421 | int flags) | |
422 | { | |
423 | int retval; | |
424 | ||
425 | map->m_flags = 0; | |
426 | /* | |
427 | * There is a race window that the result is not the same. | |
428 | * e.g. xfstests #223 when dioread_nolock enables. The reason | |
429 | * is that we lookup a block mapping in extent status tree with | |
430 | * out taking i_data_sem. So at the time the unwritten extent | |
431 | * could be converted. | |
432 | */ | |
2dcba478 | 433 | down_read(&EXT4_I(inode)->i_data_sem); |
921f266b DM |
434 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
435 | retval = ext4_ext_map_blocks(handle, inode, map, flags & | |
436 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
437 | } else { | |
438 | retval = ext4_ind_map_blocks(handle, inode, map, flags & | |
439 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
440 | } | |
2dcba478 | 441 | up_read((&EXT4_I(inode)->i_data_sem)); |
921f266b DM |
442 | |
443 | /* | |
444 | * We don't check m_len because extent will be collpased in status | |
445 | * tree. So the m_len might not equal. | |
446 | */ | |
447 | if (es_map->m_lblk != map->m_lblk || | |
448 | es_map->m_flags != map->m_flags || | |
449 | es_map->m_pblk != map->m_pblk) { | |
bdafe42a | 450 | printk("ES cache assertion failed for inode: %lu " |
921f266b DM |
451 | "es_cached ex [%d/%d/%llu/%x] != " |
452 | "found ex [%d/%d/%llu/%x] retval %d flags %x\n", | |
453 | inode->i_ino, es_map->m_lblk, es_map->m_len, | |
454 | es_map->m_pblk, es_map->m_flags, map->m_lblk, | |
455 | map->m_len, map->m_pblk, map->m_flags, | |
456 | retval, flags); | |
457 | } | |
458 | } | |
459 | #endif /* ES_AGGRESSIVE_TEST */ | |
460 | ||
f5ab0d1f | 461 | /* |
e35fd660 | 462 | * The ext4_map_blocks() function tries to look up the requested blocks, |
2b2d6d01 | 463 | * and returns if the blocks are already mapped. |
f5ab0d1f | 464 | * |
f5ab0d1f MC |
465 | * Otherwise it takes the write lock of the i_data_sem and allocate blocks |
466 | * and store the allocated blocks in the result buffer head and mark it | |
467 | * mapped. | |
468 | * | |
e35fd660 TT |
469 | * If file type is extents based, it will call ext4_ext_map_blocks(), |
470 | * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping | |
f5ab0d1f MC |
471 | * based files |
472 | * | |
facab4d9 JK |
473 | * On success, it returns the number of blocks being mapped or allocated. if |
474 | * create==0 and the blocks are pre-allocated and unwritten, the resulting @map | |
475 | * is marked as unwritten. If the create == 1, it will mark @map as mapped. | |
f5ab0d1f MC |
476 | * |
477 | * It returns 0 if plain look up failed (blocks have not been allocated), in | |
facab4d9 JK |
478 | * that case, @map is returned as unmapped but we still do fill map->m_len to |
479 | * indicate the length of a hole starting at map->m_lblk. | |
f5ab0d1f MC |
480 | * |
481 | * It returns the error in case of allocation failure. | |
482 | */ | |
e35fd660 TT |
483 | int ext4_map_blocks(handle_t *handle, struct inode *inode, |
484 | struct ext4_map_blocks *map, int flags) | |
0e855ac8 | 485 | { |
d100eef2 | 486 | struct extent_status es; |
0e855ac8 | 487 | int retval; |
b8a86845 | 488 | int ret = 0; |
921f266b DM |
489 | #ifdef ES_AGGRESSIVE_TEST |
490 | struct ext4_map_blocks orig_map; | |
491 | ||
492 | memcpy(&orig_map, map, sizeof(*map)); | |
493 | #endif | |
f5ab0d1f | 494 | |
e35fd660 TT |
495 | map->m_flags = 0; |
496 | ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u," | |
497 | "logical block %lu\n", inode->i_ino, flags, map->m_len, | |
498 | (unsigned long) map->m_lblk); | |
d100eef2 | 499 | |
e861b5e9 TT |
500 | /* |
501 | * ext4_map_blocks returns an int, and m_len is an unsigned int | |
502 | */ | |
503 | if (unlikely(map->m_len > INT_MAX)) | |
504 | map->m_len = INT_MAX; | |
505 | ||
4adb6ab3 KM |
506 | /* We can handle the block number less than EXT_MAX_BLOCKS */ |
507 | if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS)) | |
6a797d27 | 508 | return -EFSCORRUPTED; |
4adb6ab3 | 509 | |
d100eef2 | 510 | /* Lookup extent status tree firstly */ |
bb5835ed | 511 | if (ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) { |
d100eef2 ZL |
512 | if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) { |
513 | map->m_pblk = ext4_es_pblock(&es) + | |
514 | map->m_lblk - es.es_lblk; | |
515 | map->m_flags |= ext4_es_is_written(&es) ? | |
516 | EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN; | |
517 | retval = es.es_len - (map->m_lblk - es.es_lblk); | |
518 | if (retval > map->m_len) | |
519 | retval = map->m_len; | |
520 | map->m_len = retval; | |
521 | } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) { | |
facab4d9 JK |
522 | map->m_pblk = 0; |
523 | retval = es.es_len - (map->m_lblk - es.es_lblk); | |
524 | if (retval > map->m_len) | |
525 | retval = map->m_len; | |
526 | map->m_len = retval; | |
d100eef2 ZL |
527 | retval = 0; |
528 | } else { | |
1e83bc81 | 529 | BUG(); |
d100eef2 | 530 | } |
921f266b DM |
531 | #ifdef ES_AGGRESSIVE_TEST |
532 | ext4_map_blocks_es_recheck(handle, inode, map, | |
533 | &orig_map, flags); | |
534 | #endif | |
d100eef2 ZL |
535 | goto found; |
536 | } | |
537 | ||
4df3d265 | 538 | /* |
b920c755 TT |
539 | * Try to see if we can get the block without requesting a new |
540 | * file system block. | |
4df3d265 | 541 | */ |
2dcba478 | 542 | down_read(&EXT4_I(inode)->i_data_sem); |
12e9b892 | 543 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
a4e5d88b DM |
544 | retval = ext4_ext_map_blocks(handle, inode, map, flags & |
545 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
0e855ac8 | 546 | } else { |
a4e5d88b DM |
547 | retval = ext4_ind_map_blocks(handle, inode, map, flags & |
548 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
0e855ac8 | 549 | } |
f7fec032 | 550 | if (retval > 0) { |
3be78c73 | 551 | unsigned int status; |
f7fec032 | 552 | |
44fb851d ZL |
553 | if (unlikely(retval != map->m_len)) { |
554 | ext4_warning(inode->i_sb, | |
555 | "ES len assertion failed for inode " | |
556 | "%lu: retval %d != map->m_len %d", | |
557 | inode->i_ino, retval, map->m_len); | |
558 | WARN_ON(1); | |
921f266b | 559 | } |
921f266b | 560 | |
f7fec032 ZL |
561 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
562 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; | |
563 | if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) && | |
d2dc317d | 564 | !(status & EXTENT_STATUS_WRITTEN) && |
ad431025 EW |
565 | ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk, |
566 | map->m_lblk + map->m_len - 1)) | |
f7fec032 ZL |
567 | status |= EXTENT_STATUS_DELAYED; |
568 | ret = ext4_es_insert_extent(inode, map->m_lblk, | |
569 | map->m_len, map->m_pblk, status); | |
570 | if (ret < 0) | |
571 | retval = ret; | |
572 | } | |
2dcba478 | 573 | up_read((&EXT4_I(inode)->i_data_sem)); |
f5ab0d1f | 574 | |
d100eef2 | 575 | found: |
e35fd660 | 576 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
b8a86845 | 577 | ret = check_block_validity(inode, map); |
6fd058f7 TT |
578 | if (ret != 0) |
579 | return ret; | |
580 | } | |
581 | ||
f5ab0d1f | 582 | /* If it is only a block(s) look up */ |
c2177057 | 583 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) |
f5ab0d1f MC |
584 | return retval; |
585 | ||
586 | /* | |
587 | * Returns if the blocks have already allocated | |
588 | * | |
589 | * Note that if blocks have been preallocated | |
df3ab170 | 590 | * ext4_ext_get_block() returns the create = 0 |
f5ab0d1f MC |
591 | * with buffer head unmapped. |
592 | */ | |
e35fd660 | 593 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) |
b8a86845 LC |
594 | /* |
595 | * If we need to convert extent to unwritten | |
596 | * we continue and do the actual work in | |
597 | * ext4_ext_map_blocks() | |
598 | */ | |
599 | if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) | |
600 | return retval; | |
4df3d265 | 601 | |
2a8964d6 | 602 | /* |
a25a4e1a ZL |
603 | * Here we clear m_flags because after allocating an new extent, |
604 | * it will be set again. | |
2a8964d6 | 605 | */ |
a25a4e1a | 606 | map->m_flags &= ~EXT4_MAP_FLAGS; |
2a8964d6 | 607 | |
4df3d265 | 608 | /* |
556615dc | 609 | * New blocks allocate and/or writing to unwritten extent |
f5ab0d1f | 610 | * will possibly result in updating i_data, so we take |
d91bd2c1 | 611 | * the write lock of i_data_sem, and call get_block() |
f5ab0d1f | 612 | * with create == 1 flag. |
4df3d265 | 613 | */ |
c8b459f4 | 614 | down_write(&EXT4_I(inode)->i_data_sem); |
d2a17637 | 615 | |
4df3d265 AK |
616 | /* |
617 | * We need to check for EXT4 here because migrate | |
618 | * could have changed the inode type in between | |
619 | */ | |
12e9b892 | 620 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
e35fd660 | 621 | retval = ext4_ext_map_blocks(handle, inode, map, flags); |
0e855ac8 | 622 | } else { |
e35fd660 | 623 | retval = ext4_ind_map_blocks(handle, inode, map, flags); |
267e4db9 | 624 | |
e35fd660 | 625 | if (retval > 0 && map->m_flags & EXT4_MAP_NEW) { |
267e4db9 AK |
626 | /* |
627 | * We allocated new blocks which will result in | |
628 | * i_data's format changing. Force the migrate | |
629 | * to fail by clearing migrate flags | |
630 | */ | |
19f5fb7a | 631 | ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE); |
267e4db9 | 632 | } |
d2a17637 | 633 | |
5f634d06 AK |
634 | /* |
635 | * Update reserved blocks/metadata blocks after successful | |
636 | * block allocation which had been deferred till now. We don't | |
637 | * support fallocate for non extent files. So we can update | |
638 | * reserve space here. | |
639 | */ | |
640 | if ((retval > 0) && | |
1296cc85 | 641 | (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)) |
5f634d06 AK |
642 | ext4_da_update_reserve_space(inode, retval, 1); |
643 | } | |
2ac3b6e0 | 644 | |
f7fec032 | 645 | if (retval > 0) { |
3be78c73 | 646 | unsigned int status; |
f7fec032 | 647 | |
44fb851d ZL |
648 | if (unlikely(retval != map->m_len)) { |
649 | ext4_warning(inode->i_sb, | |
650 | "ES len assertion failed for inode " | |
651 | "%lu: retval %d != map->m_len %d", | |
652 | inode->i_ino, retval, map->m_len); | |
653 | WARN_ON(1); | |
921f266b | 654 | } |
921f266b | 655 | |
c86d8db3 JK |
656 | /* |
657 | * We have to zeroout blocks before inserting them into extent | |
658 | * status tree. Otherwise someone could look them up there and | |
9b623df6 JK |
659 | * use them before they are really zeroed. We also have to |
660 | * unmap metadata before zeroing as otherwise writeback can | |
661 | * overwrite zeros with stale data from block device. | |
c86d8db3 JK |
662 | */ |
663 | if (flags & EXT4_GET_BLOCKS_ZERO && | |
664 | map->m_flags & EXT4_MAP_MAPPED && | |
665 | map->m_flags & EXT4_MAP_NEW) { | |
666 | ret = ext4_issue_zeroout(inode, map->m_lblk, | |
667 | map->m_pblk, map->m_len); | |
668 | if (ret) { | |
669 | retval = ret; | |
670 | goto out_sem; | |
671 | } | |
672 | } | |
673 | ||
adb23551 ZL |
674 | /* |
675 | * If the extent has been zeroed out, we don't need to update | |
676 | * extent status tree. | |
677 | */ | |
678 | if ((flags & EXT4_GET_BLOCKS_PRE_IO) && | |
bb5835ed | 679 | ext4_es_lookup_extent(inode, map->m_lblk, NULL, &es)) { |
adb23551 | 680 | if (ext4_es_is_written(&es)) |
c86d8db3 | 681 | goto out_sem; |
adb23551 | 682 | } |
f7fec032 ZL |
683 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
684 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; | |
685 | if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) && | |
d2dc317d | 686 | !(status & EXTENT_STATUS_WRITTEN) && |
ad431025 EW |
687 | ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk, |
688 | map->m_lblk + map->m_len - 1)) | |
f7fec032 ZL |
689 | status |= EXTENT_STATUS_DELAYED; |
690 | ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, | |
691 | map->m_pblk, status); | |
c86d8db3 | 692 | if (ret < 0) { |
f7fec032 | 693 | retval = ret; |
c86d8db3 JK |
694 | goto out_sem; |
695 | } | |
5356f261 AK |
696 | } |
697 | ||
c86d8db3 | 698 | out_sem: |
4df3d265 | 699 | up_write((&EXT4_I(inode)->i_data_sem)); |
e35fd660 | 700 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
b8a86845 | 701 | ret = check_block_validity(inode, map); |
6fd058f7 TT |
702 | if (ret != 0) |
703 | return ret; | |
06bd3c36 JK |
704 | |
705 | /* | |
706 | * Inodes with freshly allocated blocks where contents will be | |
707 | * visible after transaction commit must be on transaction's | |
708 | * ordered data list. | |
709 | */ | |
710 | if (map->m_flags & EXT4_MAP_NEW && | |
711 | !(map->m_flags & EXT4_MAP_UNWRITTEN) && | |
712 | !(flags & EXT4_GET_BLOCKS_ZERO) && | |
02749a4c | 713 | !ext4_is_quota_file(inode) && |
06bd3c36 | 714 | ext4_should_order_data(inode)) { |
73131fbb RZ |
715 | loff_t start_byte = |
716 | (loff_t)map->m_lblk << inode->i_blkbits; | |
717 | loff_t length = (loff_t)map->m_len << inode->i_blkbits; | |
718 | ||
ee0876bc | 719 | if (flags & EXT4_GET_BLOCKS_IO_SUBMIT) |
73131fbb RZ |
720 | ret = ext4_jbd2_inode_add_wait(handle, inode, |
721 | start_byte, length); | |
ee0876bc | 722 | else |
73131fbb RZ |
723 | ret = ext4_jbd2_inode_add_write(handle, inode, |
724 | start_byte, length); | |
06bd3c36 JK |
725 | if (ret) |
726 | return ret; | |
727 | } | |
6fd058f7 | 728 | } |
0e855ac8 AK |
729 | return retval; |
730 | } | |
731 | ||
ed8ad838 JK |
732 | /* |
733 | * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages | |
734 | * we have to be careful as someone else may be manipulating b_state as well. | |
735 | */ | |
736 | static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags) | |
737 | { | |
738 | unsigned long old_state; | |
739 | unsigned long new_state; | |
740 | ||
741 | flags &= EXT4_MAP_FLAGS; | |
742 | ||
743 | /* Dummy buffer_head? Set non-atomically. */ | |
744 | if (!bh->b_page) { | |
745 | bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags; | |
746 | return; | |
747 | } | |
748 | /* | |
749 | * Someone else may be modifying b_state. Be careful! This is ugly but | |
750 | * once we get rid of using bh as a container for mapping information | |
751 | * to pass to / from get_block functions, this can go away. | |
752 | */ | |
753 | do { | |
754 | old_state = READ_ONCE(bh->b_state); | |
755 | new_state = (old_state & ~EXT4_MAP_FLAGS) | flags; | |
756 | } while (unlikely( | |
757 | cmpxchg(&bh->b_state, old_state, new_state) != old_state)); | |
758 | } | |
759 | ||
2ed88685 TT |
760 | static int _ext4_get_block(struct inode *inode, sector_t iblock, |
761 | struct buffer_head *bh, int flags) | |
ac27a0ec | 762 | { |
2ed88685 | 763 | struct ext4_map_blocks map; |
efe70c29 | 764 | int ret = 0; |
ac27a0ec | 765 | |
46c7f254 TM |
766 | if (ext4_has_inline_data(inode)) |
767 | return -ERANGE; | |
768 | ||
2ed88685 TT |
769 | map.m_lblk = iblock; |
770 | map.m_len = bh->b_size >> inode->i_blkbits; | |
771 | ||
efe70c29 JK |
772 | ret = ext4_map_blocks(ext4_journal_current_handle(), inode, &map, |
773 | flags); | |
7fb5409d | 774 | if (ret > 0) { |
2ed88685 | 775 | map_bh(bh, inode->i_sb, map.m_pblk); |
ed8ad838 | 776 | ext4_update_bh_state(bh, map.m_flags); |
2ed88685 | 777 | bh->b_size = inode->i_sb->s_blocksize * map.m_len; |
7fb5409d | 778 | ret = 0; |
547edce3 RZ |
779 | } else if (ret == 0) { |
780 | /* hole case, need to fill in bh->b_size */ | |
781 | bh->b_size = inode->i_sb->s_blocksize * map.m_len; | |
ac27a0ec DK |
782 | } |
783 | return ret; | |
784 | } | |
785 | ||
2ed88685 TT |
786 | int ext4_get_block(struct inode *inode, sector_t iblock, |
787 | struct buffer_head *bh, int create) | |
788 | { | |
789 | return _ext4_get_block(inode, iblock, bh, | |
790 | create ? EXT4_GET_BLOCKS_CREATE : 0); | |
791 | } | |
792 | ||
705965bd JK |
793 | /* |
794 | * Get block function used when preparing for buffered write if we require | |
795 | * creating an unwritten extent if blocks haven't been allocated. The extent | |
796 | * will be converted to written after the IO is complete. | |
797 | */ | |
798 | int ext4_get_block_unwritten(struct inode *inode, sector_t iblock, | |
799 | struct buffer_head *bh_result, int create) | |
800 | { | |
801 | ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n", | |
802 | inode->i_ino, create); | |
803 | return _ext4_get_block(inode, iblock, bh_result, | |
804 | EXT4_GET_BLOCKS_IO_CREATE_EXT); | |
805 | } | |
806 | ||
efe70c29 JK |
807 | /* Maximum number of blocks we map for direct IO at once. */ |
808 | #define DIO_MAX_BLOCKS 4096 | |
809 | ||
ac27a0ec DK |
810 | /* |
811 | * `handle' can be NULL if create is zero | |
812 | */ | |
617ba13b | 813 | struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode, |
c5e298ae | 814 | ext4_lblk_t block, int map_flags) |
ac27a0ec | 815 | { |
2ed88685 TT |
816 | struct ext4_map_blocks map; |
817 | struct buffer_head *bh; | |
c5e298ae | 818 | int create = map_flags & EXT4_GET_BLOCKS_CREATE; |
10560082 | 819 | int err; |
ac27a0ec DK |
820 | |
821 | J_ASSERT(handle != NULL || create == 0); | |
822 | ||
2ed88685 TT |
823 | map.m_lblk = block; |
824 | map.m_len = 1; | |
c5e298ae | 825 | err = ext4_map_blocks(handle, inode, &map, map_flags); |
ac27a0ec | 826 | |
10560082 TT |
827 | if (err == 0) |
828 | return create ? ERR_PTR(-ENOSPC) : NULL; | |
2ed88685 | 829 | if (err < 0) |
10560082 | 830 | return ERR_PTR(err); |
2ed88685 TT |
831 | |
832 | bh = sb_getblk(inode->i_sb, map.m_pblk); | |
10560082 TT |
833 | if (unlikely(!bh)) |
834 | return ERR_PTR(-ENOMEM); | |
2ed88685 TT |
835 | if (map.m_flags & EXT4_MAP_NEW) { |
836 | J_ASSERT(create != 0); | |
837 | J_ASSERT(handle != NULL); | |
ac27a0ec | 838 | |
2ed88685 TT |
839 | /* |
840 | * Now that we do not always journal data, we should | |
841 | * keep in mind whether this should always journal the | |
842 | * new buffer as metadata. For now, regular file | |
843 | * writes use ext4_get_block instead, so it's not a | |
844 | * problem. | |
845 | */ | |
846 | lock_buffer(bh); | |
847 | BUFFER_TRACE(bh, "call get_create_access"); | |
10560082 TT |
848 | err = ext4_journal_get_create_access(handle, bh); |
849 | if (unlikely(err)) { | |
850 | unlock_buffer(bh); | |
851 | goto errout; | |
852 | } | |
853 | if (!buffer_uptodate(bh)) { | |
2ed88685 TT |
854 | memset(bh->b_data, 0, inode->i_sb->s_blocksize); |
855 | set_buffer_uptodate(bh); | |
ac27a0ec | 856 | } |
2ed88685 TT |
857 | unlock_buffer(bh); |
858 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); | |
859 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
10560082 TT |
860 | if (unlikely(err)) |
861 | goto errout; | |
862 | } else | |
2ed88685 | 863 | BUFFER_TRACE(bh, "not a new buffer"); |
2ed88685 | 864 | return bh; |
10560082 TT |
865 | errout: |
866 | brelse(bh); | |
867 | return ERR_PTR(err); | |
ac27a0ec DK |
868 | } |
869 | ||
617ba13b | 870 | struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode, |
c5e298ae | 871 | ext4_lblk_t block, int map_flags) |
ac27a0ec | 872 | { |
af5bc92d | 873 | struct buffer_head *bh; |
ac27a0ec | 874 | |
c5e298ae | 875 | bh = ext4_getblk(handle, inode, block, map_flags); |
1c215028 | 876 | if (IS_ERR(bh)) |
ac27a0ec | 877 | return bh; |
7963e5ac | 878 | if (!bh || ext4_buffer_uptodate(bh)) |
ac27a0ec | 879 | return bh; |
dfec8a14 | 880 | ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &bh); |
ac27a0ec DK |
881 | wait_on_buffer(bh); |
882 | if (buffer_uptodate(bh)) | |
883 | return bh; | |
884 | put_bh(bh); | |
1c215028 | 885 | return ERR_PTR(-EIO); |
ac27a0ec DK |
886 | } |
887 | ||
9699d4f9 TE |
888 | /* Read a contiguous batch of blocks. */ |
889 | int ext4_bread_batch(struct inode *inode, ext4_lblk_t block, int bh_count, | |
890 | bool wait, struct buffer_head **bhs) | |
891 | { | |
892 | int i, err; | |
893 | ||
894 | for (i = 0; i < bh_count; i++) { | |
895 | bhs[i] = ext4_getblk(NULL, inode, block + i, 0 /* map_flags */); | |
896 | if (IS_ERR(bhs[i])) { | |
897 | err = PTR_ERR(bhs[i]); | |
898 | bh_count = i; | |
899 | goto out_brelse; | |
900 | } | |
901 | } | |
902 | ||
903 | for (i = 0; i < bh_count; i++) | |
904 | /* Note that NULL bhs[i] is valid because of holes. */ | |
7963e5ac | 905 | if (bhs[i] && !ext4_buffer_uptodate(bhs[i])) |
9699d4f9 TE |
906 | ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, |
907 | &bhs[i]); | |
908 | ||
909 | if (!wait) | |
910 | return 0; | |
911 | ||
912 | for (i = 0; i < bh_count; i++) | |
913 | if (bhs[i]) | |
914 | wait_on_buffer(bhs[i]); | |
915 | ||
916 | for (i = 0; i < bh_count; i++) { | |
917 | if (bhs[i] && !buffer_uptodate(bhs[i])) { | |
918 | err = -EIO; | |
919 | goto out_brelse; | |
920 | } | |
921 | } | |
922 | return 0; | |
923 | ||
924 | out_brelse: | |
925 | for (i = 0; i < bh_count; i++) { | |
926 | brelse(bhs[i]); | |
927 | bhs[i] = NULL; | |
928 | } | |
929 | return err; | |
930 | } | |
931 | ||
f19d5870 TM |
932 | int ext4_walk_page_buffers(handle_t *handle, |
933 | struct buffer_head *head, | |
934 | unsigned from, | |
935 | unsigned to, | |
936 | int *partial, | |
937 | int (*fn)(handle_t *handle, | |
938 | struct buffer_head *bh)) | |
ac27a0ec DK |
939 | { |
940 | struct buffer_head *bh; | |
941 | unsigned block_start, block_end; | |
942 | unsigned blocksize = head->b_size; | |
943 | int err, ret = 0; | |
944 | struct buffer_head *next; | |
945 | ||
af5bc92d TT |
946 | for (bh = head, block_start = 0; |
947 | ret == 0 && (bh != head || !block_start); | |
de9a55b8 | 948 | block_start = block_end, bh = next) { |
ac27a0ec DK |
949 | next = bh->b_this_page; |
950 | block_end = block_start + blocksize; | |
951 | if (block_end <= from || block_start >= to) { | |
952 | if (partial && !buffer_uptodate(bh)) | |
953 | *partial = 1; | |
954 | continue; | |
955 | } | |
956 | err = (*fn)(handle, bh); | |
957 | if (!ret) | |
958 | ret = err; | |
959 | } | |
960 | return ret; | |
961 | } | |
962 | ||
963 | /* | |
964 | * To preserve ordering, it is essential that the hole instantiation and | |
965 | * the data write be encapsulated in a single transaction. We cannot | |
617ba13b | 966 | * close off a transaction and start a new one between the ext4_get_block() |
dab291af | 967 | * and the commit_write(). So doing the jbd2_journal_start at the start of |
ac27a0ec DK |
968 | * prepare_write() is the right place. |
969 | * | |
36ade451 JK |
970 | * Also, this function can nest inside ext4_writepage(). In that case, we |
971 | * *know* that ext4_writepage() has generated enough buffer credits to do the | |
972 | * whole page. So we won't block on the journal in that case, which is good, | |
973 | * because the caller may be PF_MEMALLOC. | |
ac27a0ec | 974 | * |
617ba13b | 975 | * By accident, ext4 can be reentered when a transaction is open via |
ac27a0ec DK |
976 | * quota file writes. If we were to commit the transaction while thus |
977 | * reentered, there can be a deadlock - we would be holding a quota | |
978 | * lock, and the commit would never complete if another thread had a | |
979 | * transaction open and was blocking on the quota lock - a ranking | |
980 | * violation. | |
981 | * | |
dab291af | 982 | * So what we do is to rely on the fact that jbd2_journal_stop/journal_start |
ac27a0ec DK |
983 | * will _not_ run commit under these circumstances because handle->h_ref |
984 | * is elevated. We'll still have enough credits for the tiny quotafile | |
985 | * write. | |
986 | */ | |
f19d5870 TM |
987 | int do_journal_get_write_access(handle_t *handle, |
988 | struct buffer_head *bh) | |
ac27a0ec | 989 | { |
56d35a4c JK |
990 | int dirty = buffer_dirty(bh); |
991 | int ret; | |
992 | ||
ac27a0ec DK |
993 | if (!buffer_mapped(bh) || buffer_freed(bh)) |
994 | return 0; | |
56d35a4c | 995 | /* |
ebdec241 | 996 | * __block_write_begin() could have dirtied some buffers. Clean |
56d35a4c JK |
997 | * the dirty bit as jbd2_journal_get_write_access() could complain |
998 | * otherwise about fs integrity issues. Setting of the dirty bit | |
ebdec241 | 999 | * by __block_write_begin() isn't a real problem here as we clear |
56d35a4c JK |
1000 | * the bit before releasing a page lock and thus writeback cannot |
1001 | * ever write the buffer. | |
1002 | */ | |
1003 | if (dirty) | |
1004 | clear_buffer_dirty(bh); | |
5d601255 | 1005 | BUFFER_TRACE(bh, "get write access"); |
56d35a4c JK |
1006 | ret = ext4_journal_get_write_access(handle, bh); |
1007 | if (!ret && dirty) | |
1008 | ret = ext4_handle_dirty_metadata(handle, NULL, bh); | |
1009 | return ret; | |
ac27a0ec DK |
1010 | } |
1011 | ||
643fa961 | 1012 | #ifdef CONFIG_FS_ENCRYPTION |
2058f83a MH |
1013 | static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len, |
1014 | get_block_t *get_block) | |
1015 | { | |
09cbfeaf | 1016 | unsigned from = pos & (PAGE_SIZE - 1); |
2058f83a MH |
1017 | unsigned to = from + len; |
1018 | struct inode *inode = page->mapping->host; | |
1019 | unsigned block_start, block_end; | |
1020 | sector_t block; | |
1021 | int err = 0; | |
1022 | unsigned blocksize = inode->i_sb->s_blocksize; | |
1023 | unsigned bbits; | |
0b578f35 CR |
1024 | struct buffer_head *bh, *head, *wait[2]; |
1025 | int nr_wait = 0; | |
1026 | int i; | |
2058f83a MH |
1027 | |
1028 | BUG_ON(!PageLocked(page)); | |
09cbfeaf KS |
1029 | BUG_ON(from > PAGE_SIZE); |
1030 | BUG_ON(to > PAGE_SIZE); | |
2058f83a MH |
1031 | BUG_ON(from > to); |
1032 | ||
1033 | if (!page_has_buffers(page)) | |
1034 | create_empty_buffers(page, blocksize, 0); | |
1035 | head = page_buffers(page); | |
1036 | bbits = ilog2(blocksize); | |
09cbfeaf | 1037 | block = (sector_t)page->index << (PAGE_SHIFT - bbits); |
2058f83a MH |
1038 | |
1039 | for (bh = head, block_start = 0; bh != head || !block_start; | |
1040 | block++, block_start = block_end, bh = bh->b_this_page) { | |
1041 | block_end = block_start + blocksize; | |
1042 | if (block_end <= from || block_start >= to) { | |
1043 | if (PageUptodate(page)) { | |
1044 | if (!buffer_uptodate(bh)) | |
1045 | set_buffer_uptodate(bh); | |
1046 | } | |
1047 | continue; | |
1048 | } | |
1049 | if (buffer_new(bh)) | |
1050 | clear_buffer_new(bh); | |
1051 | if (!buffer_mapped(bh)) { | |
1052 | WARN_ON(bh->b_size != blocksize); | |
1053 | err = get_block(inode, block, bh, 1); | |
1054 | if (err) | |
1055 | break; | |
1056 | if (buffer_new(bh)) { | |
2058f83a MH |
1057 | if (PageUptodate(page)) { |
1058 | clear_buffer_new(bh); | |
1059 | set_buffer_uptodate(bh); | |
1060 | mark_buffer_dirty(bh); | |
1061 | continue; | |
1062 | } | |
1063 | if (block_end > to || block_start < from) | |
1064 | zero_user_segments(page, to, block_end, | |
1065 | block_start, from); | |
1066 | continue; | |
1067 | } | |
1068 | } | |
1069 | if (PageUptodate(page)) { | |
1070 | if (!buffer_uptodate(bh)) | |
1071 | set_buffer_uptodate(bh); | |
1072 | continue; | |
1073 | } | |
1074 | if (!buffer_uptodate(bh) && !buffer_delay(bh) && | |
1075 | !buffer_unwritten(bh) && | |
1076 | (block_start < from || block_end > to)) { | |
dfec8a14 | 1077 | ll_rw_block(REQ_OP_READ, 0, 1, &bh); |
0b578f35 | 1078 | wait[nr_wait++] = bh; |
2058f83a MH |
1079 | } |
1080 | } | |
1081 | /* | |
1082 | * If we issued read requests, let them complete. | |
1083 | */ | |
0b578f35 CR |
1084 | for (i = 0; i < nr_wait; i++) { |
1085 | wait_on_buffer(wait[i]); | |
1086 | if (!buffer_uptodate(wait[i])) | |
2058f83a MH |
1087 | err = -EIO; |
1088 | } | |
7e0785fc | 1089 | if (unlikely(err)) { |
2058f83a | 1090 | page_zero_new_buffers(page, from, to); |
0b578f35 CR |
1091 | } else if (IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode)) { |
1092 | for (i = 0; i < nr_wait; i++) { | |
1093 | int err2; | |
1094 | ||
1095 | err2 = fscrypt_decrypt_pagecache_blocks(page, blocksize, | |
1096 | bh_offset(wait[i])); | |
1097 | if (err2) { | |
1098 | clear_buffer_uptodate(wait[i]); | |
1099 | err = err2; | |
1100 | } | |
1101 | } | |
7e0785fc CR |
1102 | } |
1103 | ||
2058f83a MH |
1104 | return err; |
1105 | } | |
1106 | #endif | |
1107 | ||
bfc1af65 | 1108 | static int ext4_write_begin(struct file *file, struct address_space *mapping, |
de9a55b8 TT |
1109 | loff_t pos, unsigned len, unsigned flags, |
1110 | struct page **pagep, void **fsdata) | |
ac27a0ec | 1111 | { |
af5bc92d | 1112 | struct inode *inode = mapping->host; |
1938a150 | 1113 | int ret, needed_blocks; |
ac27a0ec DK |
1114 | handle_t *handle; |
1115 | int retries = 0; | |
af5bc92d | 1116 | struct page *page; |
de9a55b8 | 1117 | pgoff_t index; |
af5bc92d | 1118 | unsigned from, to; |
bfc1af65 | 1119 | |
0db1ff22 TT |
1120 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
1121 | return -EIO; | |
1122 | ||
9bffad1e | 1123 | trace_ext4_write_begin(inode, pos, len, flags); |
1938a150 AK |
1124 | /* |
1125 | * Reserve one block more for addition to orphan list in case | |
1126 | * we allocate blocks but write fails for some reason | |
1127 | */ | |
1128 | needed_blocks = ext4_writepage_trans_blocks(inode) + 1; | |
09cbfeaf KS |
1129 | index = pos >> PAGE_SHIFT; |
1130 | from = pos & (PAGE_SIZE - 1); | |
af5bc92d | 1131 | to = from + len; |
ac27a0ec | 1132 | |
f19d5870 TM |
1133 | if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { |
1134 | ret = ext4_try_to_write_inline_data(mapping, inode, pos, len, | |
1135 | flags, pagep); | |
1136 | if (ret < 0) | |
47564bfb TT |
1137 | return ret; |
1138 | if (ret == 1) | |
1139 | return 0; | |
f19d5870 TM |
1140 | } |
1141 | ||
47564bfb TT |
1142 | /* |
1143 | * grab_cache_page_write_begin() can take a long time if the | |
1144 | * system is thrashing due to memory pressure, or if the page | |
1145 | * is being written back. So grab it first before we start | |
1146 | * the transaction handle. This also allows us to allocate | |
1147 | * the page (if needed) without using GFP_NOFS. | |
1148 | */ | |
1149 | retry_grab: | |
1150 | page = grab_cache_page_write_begin(mapping, index, flags); | |
1151 | if (!page) | |
1152 | return -ENOMEM; | |
1153 | unlock_page(page); | |
1154 | ||
1155 | retry_journal: | |
9924a92a | 1156 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks); |
af5bc92d | 1157 | if (IS_ERR(handle)) { |
09cbfeaf | 1158 | put_page(page); |
47564bfb | 1159 | return PTR_ERR(handle); |
7479d2b9 | 1160 | } |
ac27a0ec | 1161 | |
47564bfb TT |
1162 | lock_page(page); |
1163 | if (page->mapping != mapping) { | |
1164 | /* The page got truncated from under us */ | |
1165 | unlock_page(page); | |
09cbfeaf | 1166 | put_page(page); |
cf108bca | 1167 | ext4_journal_stop(handle); |
47564bfb | 1168 | goto retry_grab; |
cf108bca | 1169 | } |
7afe5aa5 DM |
1170 | /* In case writeback began while the page was unlocked */ |
1171 | wait_for_stable_page(page); | |
cf108bca | 1172 | |
643fa961 | 1173 | #ifdef CONFIG_FS_ENCRYPTION |
2058f83a MH |
1174 | if (ext4_should_dioread_nolock(inode)) |
1175 | ret = ext4_block_write_begin(page, pos, len, | |
705965bd | 1176 | ext4_get_block_unwritten); |
2058f83a MH |
1177 | else |
1178 | ret = ext4_block_write_begin(page, pos, len, | |
1179 | ext4_get_block); | |
1180 | #else | |
744692dc | 1181 | if (ext4_should_dioread_nolock(inode)) |
705965bd JK |
1182 | ret = __block_write_begin(page, pos, len, |
1183 | ext4_get_block_unwritten); | |
744692dc | 1184 | else |
6e1db88d | 1185 | ret = __block_write_begin(page, pos, len, ext4_get_block); |
2058f83a | 1186 | #endif |
bfc1af65 | 1187 | if (!ret && ext4_should_journal_data(inode)) { |
f19d5870 TM |
1188 | ret = ext4_walk_page_buffers(handle, page_buffers(page), |
1189 | from, to, NULL, | |
1190 | do_journal_get_write_access); | |
ac27a0ec | 1191 | } |
bfc1af65 NP |
1192 | |
1193 | if (ret) { | |
c93d8f88 EB |
1194 | bool extended = (pos + len > inode->i_size) && |
1195 | !ext4_verity_in_progress(inode); | |
1196 | ||
af5bc92d | 1197 | unlock_page(page); |
ae4d5372 | 1198 | /* |
6e1db88d | 1199 | * __block_write_begin may have instantiated a few blocks |
ae4d5372 AK |
1200 | * outside i_size. Trim these off again. Don't need |
1201 | * i_size_read because we hold i_mutex. | |
1938a150 AK |
1202 | * |
1203 | * Add inode to orphan list in case we crash before | |
1204 | * truncate finishes | |
ae4d5372 | 1205 | */ |
c93d8f88 | 1206 | if (extended && ext4_can_truncate(inode)) |
1938a150 AK |
1207 | ext4_orphan_add(handle, inode); |
1208 | ||
1209 | ext4_journal_stop(handle); | |
c93d8f88 | 1210 | if (extended) { |
b9a4207d | 1211 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1212 | /* |
ffacfa7a | 1213 | * If truncate failed early the inode might |
1938a150 AK |
1214 | * still be on the orphan list; we need to |
1215 | * make sure the inode is removed from the | |
1216 | * orphan list in that case. | |
1217 | */ | |
1218 | if (inode->i_nlink) | |
1219 | ext4_orphan_del(NULL, inode); | |
1220 | } | |
bfc1af65 | 1221 | |
47564bfb TT |
1222 | if (ret == -ENOSPC && |
1223 | ext4_should_retry_alloc(inode->i_sb, &retries)) | |
1224 | goto retry_journal; | |
09cbfeaf | 1225 | put_page(page); |
47564bfb TT |
1226 | return ret; |
1227 | } | |
1228 | *pagep = page; | |
ac27a0ec DK |
1229 | return ret; |
1230 | } | |
1231 | ||
bfc1af65 NP |
1232 | /* For write_end() in data=journal mode */ |
1233 | static int write_end_fn(handle_t *handle, struct buffer_head *bh) | |
ac27a0ec | 1234 | { |
13fca323 | 1235 | int ret; |
ac27a0ec DK |
1236 | if (!buffer_mapped(bh) || buffer_freed(bh)) |
1237 | return 0; | |
1238 | set_buffer_uptodate(bh); | |
13fca323 TT |
1239 | ret = ext4_handle_dirty_metadata(handle, NULL, bh); |
1240 | clear_buffer_meta(bh); | |
1241 | clear_buffer_prio(bh); | |
1242 | return ret; | |
ac27a0ec DK |
1243 | } |
1244 | ||
eed4333f ZL |
1245 | /* |
1246 | * We need to pick up the new inode size which generic_commit_write gave us | |
1247 | * `file' can be NULL - eg, when called from page_symlink(). | |
1248 | * | |
1249 | * ext4 never places buffers on inode->i_mapping->private_list. metadata | |
1250 | * buffers are managed internally. | |
1251 | */ | |
1252 | static int ext4_write_end(struct file *file, | |
1253 | struct address_space *mapping, | |
1254 | loff_t pos, unsigned len, unsigned copied, | |
1255 | struct page *page, void *fsdata) | |
f8514083 | 1256 | { |
f8514083 | 1257 | handle_t *handle = ext4_journal_current_handle(); |
eed4333f | 1258 | struct inode *inode = mapping->host; |
0572639f | 1259 | loff_t old_size = inode->i_size; |
eed4333f ZL |
1260 | int ret = 0, ret2; |
1261 | int i_size_changed = 0; | |
362eca70 | 1262 | int inline_data = ext4_has_inline_data(inode); |
c93d8f88 | 1263 | bool verity = ext4_verity_in_progress(inode); |
eed4333f ZL |
1264 | |
1265 | trace_ext4_write_end(inode, pos, len, copied); | |
362eca70 | 1266 | if (inline_data) { |
42c832de TT |
1267 | ret = ext4_write_inline_data_end(inode, pos, len, |
1268 | copied, page); | |
eb5efbcb TT |
1269 | if (ret < 0) { |
1270 | unlock_page(page); | |
1271 | put_page(page); | |
42c832de | 1272 | goto errout; |
eb5efbcb | 1273 | } |
42c832de TT |
1274 | copied = ret; |
1275 | } else | |
f19d5870 TM |
1276 | copied = block_write_end(file, mapping, pos, |
1277 | len, copied, page, fsdata); | |
f8514083 | 1278 | /* |
4631dbf6 | 1279 | * it's important to update i_size while still holding page lock: |
f8514083 | 1280 | * page writeout could otherwise come in and zero beyond i_size. |
c93d8f88 EB |
1281 | * |
1282 | * If FS_IOC_ENABLE_VERITY is running on this inode, then Merkle tree | |
1283 | * blocks are being written past EOF, so skip the i_size update. | |
f8514083 | 1284 | */ |
c93d8f88 EB |
1285 | if (!verity) |
1286 | i_size_changed = ext4_update_inode_size(inode, pos + copied); | |
f8514083 | 1287 | unlock_page(page); |
09cbfeaf | 1288 | put_page(page); |
f8514083 | 1289 | |
c93d8f88 | 1290 | if (old_size < pos && !verity) |
0572639f | 1291 | pagecache_isize_extended(inode, old_size, pos); |
f8514083 AK |
1292 | /* |
1293 | * Don't mark the inode dirty under page lock. First, it unnecessarily | |
1294 | * makes the holding time of page lock longer. Second, it forces lock | |
1295 | * ordering of page lock and transaction start for journaling | |
1296 | * filesystems. | |
1297 | */ | |
362eca70 | 1298 | if (i_size_changed || inline_data) |
f8514083 AK |
1299 | ext4_mark_inode_dirty(handle, inode); |
1300 | ||
c93d8f88 | 1301 | if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode)) |
f8514083 AK |
1302 | /* if we have allocated more blocks and copied |
1303 | * less. We will have blocks allocated outside | |
1304 | * inode->i_size. So truncate them | |
1305 | */ | |
1306 | ext4_orphan_add(handle, inode); | |
74d553aa | 1307 | errout: |
617ba13b | 1308 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1309 | if (!ret) |
1310 | ret = ret2; | |
bfc1af65 | 1311 | |
c93d8f88 | 1312 | if (pos + len > inode->i_size && !verity) { |
b9a4207d | 1313 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1314 | /* |
ffacfa7a | 1315 | * If truncate failed early the inode might still be |
f8514083 AK |
1316 | * on the orphan list; we need to make sure the inode |
1317 | * is removed from the orphan list in that case. | |
1318 | */ | |
1319 | if (inode->i_nlink) | |
1320 | ext4_orphan_del(NULL, inode); | |
1321 | } | |
1322 | ||
bfc1af65 | 1323 | return ret ? ret : copied; |
ac27a0ec DK |
1324 | } |
1325 | ||
b90197b6 TT |
1326 | /* |
1327 | * This is a private version of page_zero_new_buffers() which doesn't | |
1328 | * set the buffer to be dirty, since in data=journalled mode we need | |
1329 | * to call ext4_handle_dirty_metadata() instead. | |
1330 | */ | |
3b136499 JK |
1331 | static void ext4_journalled_zero_new_buffers(handle_t *handle, |
1332 | struct page *page, | |
1333 | unsigned from, unsigned to) | |
b90197b6 TT |
1334 | { |
1335 | unsigned int block_start = 0, block_end; | |
1336 | struct buffer_head *head, *bh; | |
1337 | ||
1338 | bh = head = page_buffers(page); | |
1339 | do { | |
1340 | block_end = block_start + bh->b_size; | |
1341 | if (buffer_new(bh)) { | |
1342 | if (block_end > from && block_start < to) { | |
1343 | if (!PageUptodate(page)) { | |
1344 | unsigned start, size; | |
1345 | ||
1346 | start = max(from, block_start); | |
1347 | size = min(to, block_end) - start; | |
1348 | ||
1349 | zero_user(page, start, size); | |
3b136499 | 1350 | write_end_fn(handle, bh); |
b90197b6 TT |
1351 | } |
1352 | clear_buffer_new(bh); | |
1353 | } | |
1354 | } | |
1355 | block_start = block_end; | |
1356 | bh = bh->b_this_page; | |
1357 | } while (bh != head); | |
1358 | } | |
1359 | ||
bfc1af65 | 1360 | static int ext4_journalled_write_end(struct file *file, |
de9a55b8 TT |
1361 | struct address_space *mapping, |
1362 | loff_t pos, unsigned len, unsigned copied, | |
1363 | struct page *page, void *fsdata) | |
ac27a0ec | 1364 | { |
617ba13b | 1365 | handle_t *handle = ext4_journal_current_handle(); |
bfc1af65 | 1366 | struct inode *inode = mapping->host; |
0572639f | 1367 | loff_t old_size = inode->i_size; |
ac27a0ec DK |
1368 | int ret = 0, ret2; |
1369 | int partial = 0; | |
bfc1af65 | 1370 | unsigned from, to; |
4631dbf6 | 1371 | int size_changed = 0; |
362eca70 | 1372 | int inline_data = ext4_has_inline_data(inode); |
c93d8f88 | 1373 | bool verity = ext4_verity_in_progress(inode); |
ac27a0ec | 1374 | |
9bffad1e | 1375 | trace_ext4_journalled_write_end(inode, pos, len, copied); |
09cbfeaf | 1376 | from = pos & (PAGE_SIZE - 1); |
bfc1af65 NP |
1377 | to = from + len; |
1378 | ||
441c8508 CW |
1379 | BUG_ON(!ext4_handle_valid(handle)); |
1380 | ||
362eca70 | 1381 | if (inline_data) { |
eb5efbcb TT |
1382 | ret = ext4_write_inline_data_end(inode, pos, len, |
1383 | copied, page); | |
1384 | if (ret < 0) { | |
1385 | unlock_page(page); | |
1386 | put_page(page); | |
1387 | goto errout; | |
1388 | } | |
1389 | copied = ret; | |
1390 | } else if (unlikely(copied < len) && !PageUptodate(page)) { | |
3b136499 JK |
1391 | copied = 0; |
1392 | ext4_journalled_zero_new_buffers(handle, page, from, to); | |
1393 | } else { | |
1394 | if (unlikely(copied < len)) | |
1395 | ext4_journalled_zero_new_buffers(handle, page, | |
1396 | from + copied, to); | |
3fdcfb66 | 1397 | ret = ext4_walk_page_buffers(handle, page_buffers(page), from, |
3b136499 JK |
1398 | from + copied, &partial, |
1399 | write_end_fn); | |
3fdcfb66 TM |
1400 | if (!partial) |
1401 | SetPageUptodate(page); | |
1402 | } | |
c93d8f88 EB |
1403 | if (!verity) |
1404 | size_changed = ext4_update_inode_size(inode, pos + copied); | |
19f5fb7a | 1405 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
2d859db3 | 1406 | EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; |
4631dbf6 | 1407 | unlock_page(page); |
09cbfeaf | 1408 | put_page(page); |
4631dbf6 | 1409 | |
c93d8f88 | 1410 | if (old_size < pos && !verity) |
0572639f XW |
1411 | pagecache_isize_extended(inode, old_size, pos); |
1412 | ||
362eca70 | 1413 | if (size_changed || inline_data) { |
617ba13b | 1414 | ret2 = ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
1415 | if (!ret) |
1416 | ret = ret2; | |
1417 | } | |
bfc1af65 | 1418 | |
c93d8f88 | 1419 | if (pos + len > inode->i_size && !verity && ext4_can_truncate(inode)) |
f8514083 AK |
1420 | /* if we have allocated more blocks and copied |
1421 | * less. We will have blocks allocated outside | |
1422 | * inode->i_size. So truncate them | |
1423 | */ | |
1424 | ext4_orphan_add(handle, inode); | |
1425 | ||
eb5efbcb | 1426 | errout: |
617ba13b | 1427 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1428 | if (!ret) |
1429 | ret = ret2; | |
c93d8f88 | 1430 | if (pos + len > inode->i_size && !verity) { |
b9a4207d | 1431 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1432 | /* |
ffacfa7a | 1433 | * If truncate failed early the inode might still be |
f8514083 AK |
1434 | * on the orphan list; we need to make sure the inode |
1435 | * is removed from the orphan list in that case. | |
1436 | */ | |
1437 | if (inode->i_nlink) | |
1438 | ext4_orphan_del(NULL, inode); | |
1439 | } | |
bfc1af65 NP |
1440 | |
1441 | return ret ? ret : copied; | |
ac27a0ec | 1442 | } |
d2a17637 | 1443 | |
9d0be502 | 1444 | /* |
c27e43a1 | 1445 | * Reserve space for a single cluster |
9d0be502 | 1446 | */ |
c27e43a1 | 1447 | static int ext4_da_reserve_space(struct inode *inode) |
d2a17637 | 1448 | { |
60e58e0f | 1449 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
0637c6f4 | 1450 | struct ext4_inode_info *ei = EXT4_I(inode); |
5dd4056d | 1451 | int ret; |
03179fe9 TT |
1452 | |
1453 | /* | |
1454 | * We will charge metadata quota at writeout time; this saves | |
1455 | * us from metadata over-estimation, though we may go over by | |
1456 | * a small amount in the end. Here we just reserve for data. | |
1457 | */ | |
1458 | ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1)); | |
1459 | if (ret) | |
1460 | return ret; | |
d2a17637 | 1461 | |
0637c6f4 | 1462 | spin_lock(&ei->i_block_reservation_lock); |
71d4f7d0 | 1463 | if (ext4_claim_free_clusters(sbi, 1, 0)) { |
03179fe9 | 1464 | spin_unlock(&ei->i_block_reservation_lock); |
03179fe9 | 1465 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1)); |
d2a17637 MC |
1466 | return -ENOSPC; |
1467 | } | |
9d0be502 | 1468 | ei->i_reserved_data_blocks++; |
c27e43a1 | 1469 | trace_ext4_da_reserve_space(inode); |
0637c6f4 | 1470 | spin_unlock(&ei->i_block_reservation_lock); |
39bc680a | 1471 | |
d2a17637 MC |
1472 | return 0; /* success */ |
1473 | } | |
1474 | ||
f456767d | 1475 | void ext4_da_release_space(struct inode *inode, int to_free) |
d2a17637 MC |
1476 | { |
1477 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0637c6f4 | 1478 | struct ext4_inode_info *ei = EXT4_I(inode); |
d2a17637 | 1479 | |
cd213226 MC |
1480 | if (!to_free) |
1481 | return; /* Nothing to release, exit */ | |
1482 | ||
d2a17637 | 1483 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); |
cd213226 | 1484 | |
5a58ec87 | 1485 | trace_ext4_da_release_space(inode, to_free); |
0637c6f4 | 1486 | if (unlikely(to_free > ei->i_reserved_data_blocks)) { |
cd213226 | 1487 | /* |
0637c6f4 TT |
1488 | * if there aren't enough reserved blocks, then the |
1489 | * counter is messed up somewhere. Since this | |
1490 | * function is called from invalidate page, it's | |
1491 | * harmless to return without any action. | |
cd213226 | 1492 | */ |
8de5c325 | 1493 | ext4_warning(inode->i_sb, "ext4_da_release_space: " |
0637c6f4 | 1494 | "ino %lu, to_free %d with only %d reserved " |
1084f252 | 1495 | "data blocks", inode->i_ino, to_free, |
0637c6f4 TT |
1496 | ei->i_reserved_data_blocks); |
1497 | WARN_ON(1); | |
1498 | to_free = ei->i_reserved_data_blocks; | |
cd213226 | 1499 | } |
0637c6f4 | 1500 | ei->i_reserved_data_blocks -= to_free; |
cd213226 | 1501 | |
72b8ab9d | 1502 | /* update fs dirty data blocks counter */ |
57042651 | 1503 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free); |
d2a17637 | 1504 | |
d2a17637 | 1505 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
60e58e0f | 1506 | |
7b415bf6 | 1507 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free)); |
d2a17637 MC |
1508 | } |
1509 | ||
64769240 AT |
1510 | /* |
1511 | * Delayed allocation stuff | |
1512 | */ | |
1513 | ||
4e7ea81d JK |
1514 | struct mpage_da_data { |
1515 | struct inode *inode; | |
1516 | struct writeback_control *wbc; | |
6b523df4 | 1517 | |
4e7ea81d JK |
1518 | pgoff_t first_page; /* The first page to write */ |
1519 | pgoff_t next_page; /* Current page to examine */ | |
1520 | pgoff_t last_page; /* Last page to examine */ | |
791b7f08 | 1521 | /* |
4e7ea81d JK |
1522 | * Extent to map - this can be after first_page because that can be |
1523 | * fully mapped. We somewhat abuse m_flags to store whether the extent | |
1524 | * is delalloc or unwritten. | |
791b7f08 | 1525 | */ |
4e7ea81d JK |
1526 | struct ext4_map_blocks map; |
1527 | struct ext4_io_submit io_submit; /* IO submission data */ | |
dddbd6ac | 1528 | unsigned int do_map:1; |
4e7ea81d | 1529 | }; |
64769240 | 1530 | |
4e7ea81d JK |
1531 | static void mpage_release_unused_pages(struct mpage_da_data *mpd, |
1532 | bool invalidate) | |
c4a0c46e AK |
1533 | { |
1534 | int nr_pages, i; | |
1535 | pgoff_t index, end; | |
1536 | struct pagevec pvec; | |
1537 | struct inode *inode = mpd->inode; | |
1538 | struct address_space *mapping = inode->i_mapping; | |
4e7ea81d JK |
1539 | |
1540 | /* This is necessary when next_page == 0. */ | |
1541 | if (mpd->first_page >= mpd->next_page) | |
1542 | return; | |
c4a0c46e | 1543 | |
c7f5938a CW |
1544 | index = mpd->first_page; |
1545 | end = mpd->next_page - 1; | |
4e7ea81d JK |
1546 | if (invalidate) { |
1547 | ext4_lblk_t start, last; | |
09cbfeaf KS |
1548 | start = index << (PAGE_SHIFT - inode->i_blkbits); |
1549 | last = end << (PAGE_SHIFT - inode->i_blkbits); | |
4e7ea81d JK |
1550 | ext4_es_remove_extent(inode, start, last - start + 1); |
1551 | } | |
51865fda | 1552 | |
86679820 | 1553 | pagevec_init(&pvec); |
c4a0c46e | 1554 | while (index <= end) { |
397162ff | 1555 | nr_pages = pagevec_lookup_range(&pvec, mapping, &index, end); |
c4a0c46e AK |
1556 | if (nr_pages == 0) |
1557 | break; | |
1558 | for (i = 0; i < nr_pages; i++) { | |
1559 | struct page *page = pvec.pages[i]; | |
2b85a617 | 1560 | |
c4a0c46e AK |
1561 | BUG_ON(!PageLocked(page)); |
1562 | BUG_ON(PageWriteback(page)); | |
4e7ea81d | 1563 | if (invalidate) { |
4e800c03 | 1564 | if (page_mapped(page)) |
1565 | clear_page_dirty_for_io(page); | |
09cbfeaf | 1566 | block_invalidatepage(page, 0, PAGE_SIZE); |
4e7ea81d JK |
1567 | ClearPageUptodate(page); |
1568 | } | |
c4a0c46e AK |
1569 | unlock_page(page); |
1570 | } | |
9b1d0998 | 1571 | pagevec_release(&pvec); |
c4a0c46e | 1572 | } |
c4a0c46e AK |
1573 | } |
1574 | ||
df22291f AK |
1575 | static void ext4_print_free_blocks(struct inode *inode) |
1576 | { | |
1577 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
92b97816 | 1578 | struct super_block *sb = inode->i_sb; |
f78ee70d | 1579 | struct ext4_inode_info *ei = EXT4_I(inode); |
92b97816 TT |
1580 | |
1581 | ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld", | |
5dee5437 | 1582 | EXT4_C2B(EXT4_SB(inode->i_sb), |
f78ee70d | 1583 | ext4_count_free_clusters(sb))); |
92b97816 TT |
1584 | ext4_msg(sb, KERN_CRIT, "Free/Dirty block details"); |
1585 | ext4_msg(sb, KERN_CRIT, "free_blocks=%lld", | |
f78ee70d | 1586 | (long long) EXT4_C2B(EXT4_SB(sb), |
57042651 | 1587 | percpu_counter_sum(&sbi->s_freeclusters_counter))); |
92b97816 | 1588 | ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld", |
f78ee70d | 1589 | (long long) EXT4_C2B(EXT4_SB(sb), |
7b415bf6 | 1590 | percpu_counter_sum(&sbi->s_dirtyclusters_counter))); |
92b97816 TT |
1591 | ext4_msg(sb, KERN_CRIT, "Block reservation details"); |
1592 | ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u", | |
f78ee70d | 1593 | ei->i_reserved_data_blocks); |
df22291f AK |
1594 | return; |
1595 | } | |
1596 | ||
c364b22c | 1597 | static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh) |
29fa89d0 | 1598 | { |
c364b22c | 1599 | return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh); |
29fa89d0 AK |
1600 | } |
1601 | ||
0b02f4c0 EW |
1602 | /* |
1603 | * ext4_insert_delayed_block - adds a delayed block to the extents status | |
1604 | * tree, incrementing the reserved cluster/block | |
1605 | * count or making a pending reservation | |
1606 | * where needed | |
1607 | * | |
1608 | * @inode - file containing the newly added block | |
1609 | * @lblk - logical block to be added | |
1610 | * | |
1611 | * Returns 0 on success, negative error code on failure. | |
1612 | */ | |
1613 | static int ext4_insert_delayed_block(struct inode *inode, ext4_lblk_t lblk) | |
1614 | { | |
1615 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
1616 | int ret; | |
1617 | bool allocated = false; | |
1618 | ||
1619 | /* | |
1620 | * If the cluster containing lblk is shared with a delayed, | |
1621 | * written, or unwritten extent in a bigalloc file system, it's | |
1622 | * already been accounted for and does not need to be reserved. | |
1623 | * A pending reservation must be made for the cluster if it's | |
1624 | * shared with a written or unwritten extent and doesn't already | |
1625 | * have one. Written and unwritten extents can be purged from the | |
1626 | * extents status tree if the system is under memory pressure, so | |
1627 | * it's necessary to examine the extent tree if a search of the | |
1628 | * extents status tree doesn't get a match. | |
1629 | */ | |
1630 | if (sbi->s_cluster_ratio == 1) { | |
1631 | ret = ext4_da_reserve_space(inode); | |
1632 | if (ret != 0) /* ENOSPC */ | |
1633 | goto errout; | |
1634 | } else { /* bigalloc */ | |
1635 | if (!ext4_es_scan_clu(inode, &ext4_es_is_delonly, lblk)) { | |
1636 | if (!ext4_es_scan_clu(inode, | |
1637 | &ext4_es_is_mapped, lblk)) { | |
1638 | ret = ext4_clu_mapped(inode, | |
1639 | EXT4_B2C(sbi, lblk)); | |
1640 | if (ret < 0) | |
1641 | goto errout; | |
1642 | if (ret == 0) { | |
1643 | ret = ext4_da_reserve_space(inode); | |
1644 | if (ret != 0) /* ENOSPC */ | |
1645 | goto errout; | |
1646 | } else { | |
1647 | allocated = true; | |
1648 | } | |
1649 | } else { | |
1650 | allocated = true; | |
1651 | } | |
1652 | } | |
1653 | } | |
1654 | ||
1655 | ret = ext4_es_insert_delayed_block(inode, lblk, allocated); | |
1656 | ||
1657 | errout: | |
1658 | return ret; | |
1659 | } | |
1660 | ||
5356f261 AK |
1661 | /* |
1662 | * This function is grabs code from the very beginning of | |
1663 | * ext4_map_blocks, but assumes that the caller is from delayed write | |
1664 | * time. This function looks up the requested blocks and sets the | |
1665 | * buffer delay bit under the protection of i_data_sem. | |
1666 | */ | |
1667 | static int ext4_da_map_blocks(struct inode *inode, sector_t iblock, | |
1668 | struct ext4_map_blocks *map, | |
1669 | struct buffer_head *bh) | |
1670 | { | |
d100eef2 | 1671 | struct extent_status es; |
5356f261 AK |
1672 | int retval; |
1673 | sector_t invalid_block = ~((sector_t) 0xffff); | |
921f266b DM |
1674 | #ifdef ES_AGGRESSIVE_TEST |
1675 | struct ext4_map_blocks orig_map; | |
1676 | ||
1677 | memcpy(&orig_map, map, sizeof(*map)); | |
1678 | #endif | |
5356f261 AK |
1679 | |
1680 | if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es)) | |
1681 | invalid_block = ~0; | |
1682 | ||
1683 | map->m_flags = 0; | |
1684 | ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u," | |
1685 | "logical block %lu\n", inode->i_ino, map->m_len, | |
1686 | (unsigned long) map->m_lblk); | |
d100eef2 ZL |
1687 | |
1688 | /* Lookup extent status tree firstly */ | |
bb5835ed | 1689 | if (ext4_es_lookup_extent(inode, iblock, NULL, &es)) { |
d100eef2 ZL |
1690 | if (ext4_es_is_hole(&es)) { |
1691 | retval = 0; | |
c8b459f4 | 1692 | down_read(&EXT4_I(inode)->i_data_sem); |
d100eef2 ZL |
1693 | goto add_delayed; |
1694 | } | |
1695 | ||
1696 | /* | |
1697 | * Delayed extent could be allocated by fallocate. | |
1698 | * So we need to check it. | |
1699 | */ | |
1700 | if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) { | |
1701 | map_bh(bh, inode->i_sb, invalid_block); | |
1702 | set_buffer_new(bh); | |
1703 | set_buffer_delay(bh); | |
1704 | return 0; | |
1705 | } | |
1706 | ||
1707 | map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk; | |
1708 | retval = es.es_len - (iblock - es.es_lblk); | |
1709 | if (retval > map->m_len) | |
1710 | retval = map->m_len; | |
1711 | map->m_len = retval; | |
1712 | if (ext4_es_is_written(&es)) | |
1713 | map->m_flags |= EXT4_MAP_MAPPED; | |
1714 | else if (ext4_es_is_unwritten(&es)) | |
1715 | map->m_flags |= EXT4_MAP_UNWRITTEN; | |
1716 | else | |
1e83bc81 | 1717 | BUG(); |
d100eef2 | 1718 | |
921f266b DM |
1719 | #ifdef ES_AGGRESSIVE_TEST |
1720 | ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0); | |
1721 | #endif | |
d100eef2 ZL |
1722 | return retval; |
1723 | } | |
1724 | ||
5356f261 AK |
1725 | /* |
1726 | * Try to see if we can get the block without requesting a new | |
1727 | * file system block. | |
1728 | */ | |
c8b459f4 | 1729 | down_read(&EXT4_I(inode)->i_data_sem); |
cbd7584e | 1730 | if (ext4_has_inline_data(inode)) |
9c3569b5 | 1731 | retval = 0; |
cbd7584e | 1732 | else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
2f8e0a7c | 1733 | retval = ext4_ext_map_blocks(NULL, inode, map, 0); |
5356f261 | 1734 | else |
2f8e0a7c | 1735 | retval = ext4_ind_map_blocks(NULL, inode, map, 0); |
5356f261 | 1736 | |
d100eef2 | 1737 | add_delayed: |
5356f261 | 1738 | if (retval == 0) { |
f7fec032 | 1739 | int ret; |
ad431025 | 1740 | |
5356f261 AK |
1741 | /* |
1742 | * XXX: __block_prepare_write() unmaps passed block, | |
1743 | * is it OK? | |
1744 | */ | |
5356f261 | 1745 | |
0b02f4c0 EW |
1746 | ret = ext4_insert_delayed_block(inode, map->m_lblk); |
1747 | if (ret != 0) { | |
f7fec032 | 1748 | retval = ret; |
51865fda | 1749 | goto out_unlock; |
f7fec032 | 1750 | } |
51865fda | 1751 | |
5356f261 AK |
1752 | map_bh(bh, inode->i_sb, invalid_block); |
1753 | set_buffer_new(bh); | |
1754 | set_buffer_delay(bh); | |
f7fec032 ZL |
1755 | } else if (retval > 0) { |
1756 | int ret; | |
3be78c73 | 1757 | unsigned int status; |
f7fec032 | 1758 | |
44fb851d ZL |
1759 | if (unlikely(retval != map->m_len)) { |
1760 | ext4_warning(inode->i_sb, | |
1761 | "ES len assertion failed for inode " | |
1762 | "%lu: retval %d != map->m_len %d", | |
1763 | inode->i_ino, retval, map->m_len); | |
1764 | WARN_ON(1); | |
921f266b | 1765 | } |
921f266b | 1766 | |
f7fec032 ZL |
1767 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
1768 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; | |
1769 | ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, | |
1770 | map->m_pblk, status); | |
1771 | if (ret != 0) | |
1772 | retval = ret; | |
5356f261 AK |
1773 | } |
1774 | ||
1775 | out_unlock: | |
1776 | up_read((&EXT4_I(inode)->i_data_sem)); | |
1777 | ||
1778 | return retval; | |
1779 | } | |
1780 | ||
64769240 | 1781 | /* |
d91bd2c1 | 1782 | * This is a special get_block_t callback which is used by |
b920c755 TT |
1783 | * ext4_da_write_begin(). It will either return mapped block or |
1784 | * reserve space for a single block. | |
29fa89d0 AK |
1785 | * |
1786 | * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set. | |
1787 | * We also have b_blocknr = -1 and b_bdev initialized properly | |
1788 | * | |
1789 | * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set. | |
1790 | * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev | |
1791 | * initialized properly. | |
64769240 | 1792 | */ |
9c3569b5 TM |
1793 | int ext4_da_get_block_prep(struct inode *inode, sector_t iblock, |
1794 | struct buffer_head *bh, int create) | |
64769240 | 1795 | { |
2ed88685 | 1796 | struct ext4_map_blocks map; |
64769240 AT |
1797 | int ret = 0; |
1798 | ||
1799 | BUG_ON(create == 0); | |
2ed88685 TT |
1800 | BUG_ON(bh->b_size != inode->i_sb->s_blocksize); |
1801 | ||
1802 | map.m_lblk = iblock; | |
1803 | map.m_len = 1; | |
64769240 AT |
1804 | |
1805 | /* | |
1806 | * first, we need to know whether the block is allocated already | |
1807 | * preallocated blocks are unmapped but should treated | |
1808 | * the same as allocated blocks. | |
1809 | */ | |
5356f261 AK |
1810 | ret = ext4_da_map_blocks(inode, iblock, &map, bh); |
1811 | if (ret <= 0) | |
2ed88685 | 1812 | return ret; |
64769240 | 1813 | |
2ed88685 | 1814 | map_bh(bh, inode->i_sb, map.m_pblk); |
ed8ad838 | 1815 | ext4_update_bh_state(bh, map.m_flags); |
2ed88685 TT |
1816 | |
1817 | if (buffer_unwritten(bh)) { | |
1818 | /* A delayed write to unwritten bh should be marked | |
1819 | * new and mapped. Mapped ensures that we don't do | |
1820 | * get_block multiple times when we write to the same | |
1821 | * offset and new ensures that we do proper zero out | |
1822 | * for partial write. | |
1823 | */ | |
1824 | set_buffer_new(bh); | |
c8205636 | 1825 | set_buffer_mapped(bh); |
2ed88685 TT |
1826 | } |
1827 | return 0; | |
64769240 | 1828 | } |
61628a3f | 1829 | |
62e086be AK |
1830 | static int bget_one(handle_t *handle, struct buffer_head *bh) |
1831 | { | |
1832 | get_bh(bh); | |
1833 | return 0; | |
1834 | } | |
1835 | ||
1836 | static int bput_one(handle_t *handle, struct buffer_head *bh) | |
1837 | { | |
1838 | put_bh(bh); | |
1839 | return 0; | |
1840 | } | |
1841 | ||
1842 | static int __ext4_journalled_writepage(struct page *page, | |
62e086be AK |
1843 | unsigned int len) |
1844 | { | |
1845 | struct address_space *mapping = page->mapping; | |
1846 | struct inode *inode = mapping->host; | |
3fdcfb66 | 1847 | struct buffer_head *page_bufs = NULL; |
62e086be | 1848 | handle_t *handle = NULL; |
3fdcfb66 TM |
1849 | int ret = 0, err = 0; |
1850 | int inline_data = ext4_has_inline_data(inode); | |
1851 | struct buffer_head *inode_bh = NULL; | |
62e086be | 1852 | |
cb20d518 | 1853 | ClearPageChecked(page); |
3fdcfb66 TM |
1854 | |
1855 | if (inline_data) { | |
1856 | BUG_ON(page->index != 0); | |
1857 | BUG_ON(len > ext4_get_max_inline_size(inode)); | |
1858 | inode_bh = ext4_journalled_write_inline_data(inode, len, page); | |
1859 | if (inode_bh == NULL) | |
1860 | goto out; | |
1861 | } else { | |
1862 | page_bufs = page_buffers(page); | |
1863 | if (!page_bufs) { | |
1864 | BUG(); | |
1865 | goto out; | |
1866 | } | |
1867 | ext4_walk_page_buffers(handle, page_bufs, 0, len, | |
1868 | NULL, bget_one); | |
1869 | } | |
bdf96838 TT |
1870 | /* |
1871 | * We need to release the page lock before we start the | |
1872 | * journal, so grab a reference so the page won't disappear | |
1873 | * out from under us. | |
1874 | */ | |
1875 | get_page(page); | |
62e086be AK |
1876 | unlock_page(page); |
1877 | ||
9924a92a TT |
1878 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
1879 | ext4_writepage_trans_blocks(inode)); | |
62e086be AK |
1880 | if (IS_ERR(handle)) { |
1881 | ret = PTR_ERR(handle); | |
bdf96838 TT |
1882 | put_page(page); |
1883 | goto out_no_pagelock; | |
62e086be | 1884 | } |
441c8508 CW |
1885 | BUG_ON(!ext4_handle_valid(handle)); |
1886 | ||
bdf96838 TT |
1887 | lock_page(page); |
1888 | put_page(page); | |
1889 | if (page->mapping != mapping) { | |
1890 | /* The page got truncated from under us */ | |
1891 | ext4_journal_stop(handle); | |
1892 | ret = 0; | |
1893 | goto out; | |
1894 | } | |
1895 | ||
3fdcfb66 | 1896 | if (inline_data) { |
362eca70 | 1897 | ret = ext4_mark_inode_dirty(handle, inode); |
3fdcfb66 TM |
1898 | } else { |
1899 | ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL, | |
1900 | do_journal_get_write_access); | |
1901 | ||
1902 | err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL, | |
1903 | write_end_fn); | |
1904 | } | |
62e086be AK |
1905 | if (ret == 0) |
1906 | ret = err; | |
2d859db3 | 1907 | EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; |
62e086be AK |
1908 | err = ext4_journal_stop(handle); |
1909 | if (!ret) | |
1910 | ret = err; | |
1911 | ||
3fdcfb66 | 1912 | if (!ext4_has_inline_data(inode)) |
8c9367fd | 1913 | ext4_walk_page_buffers(NULL, page_bufs, 0, len, |
3fdcfb66 | 1914 | NULL, bput_one); |
19f5fb7a | 1915 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
62e086be | 1916 | out: |
bdf96838 TT |
1917 | unlock_page(page); |
1918 | out_no_pagelock: | |
3fdcfb66 | 1919 | brelse(inode_bh); |
62e086be AK |
1920 | return ret; |
1921 | } | |
1922 | ||
61628a3f | 1923 | /* |
43ce1d23 AK |
1924 | * Note that we don't need to start a transaction unless we're journaling data |
1925 | * because we should have holes filled from ext4_page_mkwrite(). We even don't | |
1926 | * need to file the inode to the transaction's list in ordered mode because if | |
1927 | * we are writing back data added by write(), the inode is already there and if | |
25985edc | 1928 | * we are writing back data modified via mmap(), no one guarantees in which |
43ce1d23 AK |
1929 | * transaction the data will hit the disk. In case we are journaling data, we |
1930 | * cannot start transaction directly because transaction start ranks above page | |
1931 | * lock so we have to do some magic. | |
1932 | * | |
b920c755 | 1933 | * This function can get called via... |
20970ba6 | 1934 | * - ext4_writepages after taking page lock (have journal handle) |
b920c755 | 1935 | * - journal_submit_inode_data_buffers (no journal handle) |
f6463b0d | 1936 | * - shrink_page_list via the kswapd/direct reclaim (no journal handle) |
b920c755 | 1937 | * - grab_page_cache when doing write_begin (have journal handle) |
43ce1d23 AK |
1938 | * |
1939 | * We don't do any block allocation in this function. If we have page with | |
1940 | * multiple blocks we need to write those buffer_heads that are mapped. This | |
1941 | * is important for mmaped based write. So if we do with blocksize 1K | |
1942 | * truncate(f, 1024); | |
1943 | * a = mmap(f, 0, 4096); | |
1944 | * a[0] = 'a'; | |
1945 | * truncate(f, 4096); | |
1946 | * we have in the page first buffer_head mapped via page_mkwrite call back | |
90802ed9 | 1947 | * but other buffer_heads would be unmapped but dirty (dirty done via the |
43ce1d23 AK |
1948 | * do_wp_page). So writepage should write the first block. If we modify |
1949 | * the mmap area beyond 1024 we will again get a page_fault and the | |
1950 | * page_mkwrite callback will do the block allocation and mark the | |
1951 | * buffer_heads mapped. | |
1952 | * | |
1953 | * We redirty the page if we have any buffer_heads that is either delay or | |
1954 | * unwritten in the page. | |
1955 | * | |
1956 | * We can get recursively called as show below. | |
1957 | * | |
1958 | * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() -> | |
1959 | * ext4_writepage() | |
1960 | * | |
1961 | * But since we don't do any block allocation we should not deadlock. | |
1962 | * Page also have the dirty flag cleared so we don't get recurive page_lock. | |
61628a3f | 1963 | */ |
43ce1d23 | 1964 | static int ext4_writepage(struct page *page, |
62e086be | 1965 | struct writeback_control *wbc) |
64769240 | 1966 | { |
f8bec370 | 1967 | int ret = 0; |
61628a3f | 1968 | loff_t size; |
498e5f24 | 1969 | unsigned int len; |
744692dc | 1970 | struct buffer_head *page_bufs = NULL; |
61628a3f | 1971 | struct inode *inode = page->mapping->host; |
36ade451 | 1972 | struct ext4_io_submit io_submit; |
1c8349a1 | 1973 | bool keep_towrite = false; |
61628a3f | 1974 | |
0db1ff22 | 1975 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) { |
c2a559bc | 1976 | inode->i_mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE); |
0db1ff22 TT |
1977 | unlock_page(page); |
1978 | return -EIO; | |
1979 | } | |
1980 | ||
a9c667f8 | 1981 | trace_ext4_writepage(page); |
f0e6c985 | 1982 | size = i_size_read(inode); |
c93d8f88 EB |
1983 | if (page->index == size >> PAGE_SHIFT && |
1984 | !ext4_verity_in_progress(inode)) | |
09cbfeaf | 1985 | len = size & ~PAGE_MASK; |
f0e6c985 | 1986 | else |
09cbfeaf | 1987 | len = PAGE_SIZE; |
64769240 | 1988 | |
a42afc5f | 1989 | page_bufs = page_buffers(page); |
a42afc5f | 1990 | /* |
fe386132 JK |
1991 | * We cannot do block allocation or other extent handling in this |
1992 | * function. If there are buffers needing that, we have to redirty | |
1993 | * the page. But we may reach here when we do a journal commit via | |
1994 | * journal_submit_inode_data_buffers() and in that case we must write | |
1995 | * allocated buffers to achieve data=ordered mode guarantees. | |
cccd147a TT |
1996 | * |
1997 | * Also, if there is only one buffer per page (the fs block | |
1998 | * size == the page size), if one buffer needs block | |
1999 | * allocation or needs to modify the extent tree to clear the | |
2000 | * unwritten flag, we know that the page can't be written at | |
2001 | * all, so we might as well refuse the write immediately. | |
2002 | * Unfortunately if the block size != page size, we can't as | |
2003 | * easily detect this case using ext4_walk_page_buffers(), but | |
2004 | * for the extremely common case, this is an optimization that | |
2005 | * skips a useless round trip through ext4_bio_write_page(). | |
a42afc5f | 2006 | */ |
f19d5870 TM |
2007 | if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL, |
2008 | ext4_bh_delay_or_unwritten)) { | |
f8bec370 | 2009 | redirty_page_for_writepage(wbc, page); |
cccd147a | 2010 | if ((current->flags & PF_MEMALLOC) || |
09cbfeaf | 2011 | (inode->i_sb->s_blocksize == PAGE_SIZE)) { |
fe386132 JK |
2012 | /* |
2013 | * For memory cleaning there's no point in writing only | |
2014 | * some buffers. So just bail out. Warn if we came here | |
2015 | * from direct reclaim. | |
2016 | */ | |
2017 | WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) | |
2018 | == PF_MEMALLOC); | |
f0e6c985 AK |
2019 | unlock_page(page); |
2020 | return 0; | |
2021 | } | |
1c8349a1 | 2022 | keep_towrite = true; |
a42afc5f | 2023 | } |
64769240 | 2024 | |
cb20d518 | 2025 | if (PageChecked(page) && ext4_should_journal_data(inode)) |
43ce1d23 AK |
2026 | /* |
2027 | * It's mmapped pagecache. Add buffers and journal it. There | |
2028 | * doesn't seem much point in redirtying the page here. | |
2029 | */ | |
3f0ca309 | 2030 | return __ext4_journalled_writepage(page, len); |
43ce1d23 | 2031 | |
97a851ed JK |
2032 | ext4_io_submit_init(&io_submit, wbc); |
2033 | io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS); | |
2034 | if (!io_submit.io_end) { | |
2035 | redirty_page_for_writepage(wbc, page); | |
2036 | unlock_page(page); | |
2037 | return -ENOMEM; | |
2038 | } | |
1c8349a1 | 2039 | ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite); |
36ade451 | 2040 | ext4_io_submit(&io_submit); |
97a851ed JK |
2041 | /* Drop io_end reference we got from init */ |
2042 | ext4_put_io_end_defer(io_submit.io_end); | |
64769240 AT |
2043 | return ret; |
2044 | } | |
2045 | ||
5f1132b2 JK |
2046 | static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page) |
2047 | { | |
2048 | int len; | |
a056bdaa | 2049 | loff_t size; |
5f1132b2 JK |
2050 | int err; |
2051 | ||
2052 | BUG_ON(page->index != mpd->first_page); | |
a056bdaa JK |
2053 | clear_page_dirty_for_io(page); |
2054 | /* | |
2055 | * We have to be very careful here! Nothing protects writeback path | |
2056 | * against i_size changes and the page can be writeably mapped into | |
2057 | * page tables. So an application can be growing i_size and writing | |
2058 | * data through mmap while writeback runs. clear_page_dirty_for_io() | |
2059 | * write-protects our page in page tables and the page cannot get | |
2060 | * written to again until we release page lock. So only after | |
2061 | * clear_page_dirty_for_io() we are safe to sample i_size for | |
2062 | * ext4_bio_write_page() to zero-out tail of the written page. We rely | |
2063 | * on the barrier provided by TestClearPageDirty in | |
2064 | * clear_page_dirty_for_io() to make sure i_size is really sampled only | |
2065 | * after page tables are updated. | |
2066 | */ | |
2067 | size = i_size_read(mpd->inode); | |
c93d8f88 EB |
2068 | if (page->index == size >> PAGE_SHIFT && |
2069 | !ext4_verity_in_progress(mpd->inode)) | |
09cbfeaf | 2070 | len = size & ~PAGE_MASK; |
5f1132b2 | 2071 | else |
09cbfeaf | 2072 | len = PAGE_SIZE; |
1c8349a1 | 2073 | err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false); |
5f1132b2 JK |
2074 | if (!err) |
2075 | mpd->wbc->nr_to_write--; | |
2076 | mpd->first_page++; | |
2077 | ||
2078 | return err; | |
2079 | } | |
2080 | ||
4e7ea81d JK |
2081 | #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay)) |
2082 | ||
61628a3f | 2083 | /* |
fffb2739 JK |
2084 | * mballoc gives us at most this number of blocks... |
2085 | * XXX: That seems to be only a limitation of ext4_mb_normalize_request(). | |
70261f56 | 2086 | * The rest of mballoc seems to handle chunks up to full group size. |
61628a3f | 2087 | */ |
fffb2739 | 2088 | #define MAX_WRITEPAGES_EXTENT_LEN 2048 |
525f4ed8 | 2089 | |
4e7ea81d JK |
2090 | /* |
2091 | * mpage_add_bh_to_extent - try to add bh to extent of blocks to map | |
2092 | * | |
2093 | * @mpd - extent of blocks | |
2094 | * @lblk - logical number of the block in the file | |
09930042 | 2095 | * @bh - buffer head we want to add to the extent |
4e7ea81d | 2096 | * |
09930042 JK |
2097 | * The function is used to collect contig. blocks in the same state. If the |
2098 | * buffer doesn't require mapping for writeback and we haven't started the | |
2099 | * extent of buffers to map yet, the function returns 'true' immediately - the | |
2100 | * caller can write the buffer right away. Otherwise the function returns true | |
2101 | * if the block has been added to the extent, false if the block couldn't be | |
2102 | * added. | |
4e7ea81d | 2103 | */ |
09930042 JK |
2104 | static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk, |
2105 | struct buffer_head *bh) | |
4e7ea81d JK |
2106 | { |
2107 | struct ext4_map_blocks *map = &mpd->map; | |
2108 | ||
09930042 JK |
2109 | /* Buffer that doesn't need mapping for writeback? */ |
2110 | if (!buffer_dirty(bh) || !buffer_mapped(bh) || | |
2111 | (!buffer_delay(bh) && !buffer_unwritten(bh))) { | |
2112 | /* So far no extent to map => we write the buffer right away */ | |
2113 | if (map->m_len == 0) | |
2114 | return true; | |
2115 | return false; | |
2116 | } | |
4e7ea81d JK |
2117 | |
2118 | /* First block in the extent? */ | |
2119 | if (map->m_len == 0) { | |
dddbd6ac JK |
2120 | /* We cannot map unless handle is started... */ |
2121 | if (!mpd->do_map) | |
2122 | return false; | |
4e7ea81d JK |
2123 | map->m_lblk = lblk; |
2124 | map->m_len = 1; | |
09930042 JK |
2125 | map->m_flags = bh->b_state & BH_FLAGS; |
2126 | return true; | |
4e7ea81d JK |
2127 | } |
2128 | ||
09930042 JK |
2129 | /* Don't go larger than mballoc is willing to allocate */ |
2130 | if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN) | |
2131 | return false; | |
2132 | ||
4e7ea81d JK |
2133 | /* Can we merge the block to our big extent? */ |
2134 | if (lblk == map->m_lblk + map->m_len && | |
09930042 | 2135 | (bh->b_state & BH_FLAGS) == map->m_flags) { |
4e7ea81d | 2136 | map->m_len++; |
09930042 | 2137 | return true; |
4e7ea81d | 2138 | } |
09930042 | 2139 | return false; |
4e7ea81d JK |
2140 | } |
2141 | ||
5f1132b2 JK |
2142 | /* |
2143 | * mpage_process_page_bufs - submit page buffers for IO or add them to extent | |
2144 | * | |
2145 | * @mpd - extent of blocks for mapping | |
2146 | * @head - the first buffer in the page | |
2147 | * @bh - buffer we should start processing from | |
2148 | * @lblk - logical number of the block in the file corresponding to @bh | |
2149 | * | |
2150 | * Walk through page buffers from @bh upto @head (exclusive) and either submit | |
2151 | * the page for IO if all buffers in this page were mapped and there's no | |
2152 | * accumulated extent of buffers to map or add buffers in the page to the | |
2153 | * extent of buffers to map. The function returns 1 if the caller can continue | |
2154 | * by processing the next page, 0 if it should stop adding buffers to the | |
2155 | * extent to map because we cannot extend it anymore. It can also return value | |
2156 | * < 0 in case of error during IO submission. | |
2157 | */ | |
2158 | static int mpage_process_page_bufs(struct mpage_da_data *mpd, | |
2159 | struct buffer_head *head, | |
2160 | struct buffer_head *bh, | |
2161 | ext4_lblk_t lblk) | |
4e7ea81d JK |
2162 | { |
2163 | struct inode *inode = mpd->inode; | |
5f1132b2 | 2164 | int err; |
93407472 | 2165 | ext4_lblk_t blocks = (i_size_read(inode) + i_blocksize(inode) - 1) |
4e7ea81d JK |
2166 | >> inode->i_blkbits; |
2167 | ||
c93d8f88 EB |
2168 | if (ext4_verity_in_progress(inode)) |
2169 | blocks = EXT_MAX_BLOCKS; | |
2170 | ||
4e7ea81d JK |
2171 | do { |
2172 | BUG_ON(buffer_locked(bh)); | |
2173 | ||
09930042 | 2174 | if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) { |
4e7ea81d JK |
2175 | /* Found extent to map? */ |
2176 | if (mpd->map.m_len) | |
5f1132b2 | 2177 | return 0; |
dddbd6ac JK |
2178 | /* Buffer needs mapping and handle is not started? */ |
2179 | if (!mpd->do_map) | |
2180 | return 0; | |
09930042 | 2181 | /* Everything mapped so far and we hit EOF */ |
5f1132b2 | 2182 | break; |
4e7ea81d | 2183 | } |
4e7ea81d | 2184 | } while (lblk++, (bh = bh->b_this_page) != head); |
5f1132b2 JK |
2185 | /* So far everything mapped? Submit the page for IO. */ |
2186 | if (mpd->map.m_len == 0) { | |
2187 | err = mpage_submit_page(mpd, head->b_page); | |
2188 | if (err < 0) | |
2189 | return err; | |
2190 | } | |
2191 | return lblk < blocks; | |
4e7ea81d JK |
2192 | } |
2193 | ||
2943fdbc RH |
2194 | /* |
2195 | * mpage_process_page - update page buffers corresponding to changed extent and | |
2196 | * may submit fully mapped page for IO | |
2197 | * | |
2198 | * @mpd - description of extent to map, on return next extent to map | |
2199 | * @m_lblk - logical block mapping. | |
2200 | * @m_pblk - corresponding physical mapping. | |
2201 | * @map_bh - determines on return whether this page requires any further | |
2202 | * mapping or not. | |
2203 | * Scan given page buffers corresponding to changed extent and update buffer | |
2204 | * state according to new extent state. | |
2205 | * We map delalloc buffers to their physical location, clear unwritten bits. | |
2206 | * If the given page is not fully mapped, we update @map to the next extent in | |
2207 | * the given page that needs mapping & return @map_bh as true. | |
2208 | */ | |
2209 | static int mpage_process_page(struct mpage_da_data *mpd, struct page *page, | |
2210 | ext4_lblk_t *m_lblk, ext4_fsblk_t *m_pblk, | |
2211 | bool *map_bh) | |
2212 | { | |
2213 | struct buffer_head *head, *bh; | |
2214 | ext4_io_end_t *io_end = mpd->io_submit.io_end; | |
2215 | ext4_lblk_t lblk = *m_lblk; | |
2216 | ext4_fsblk_t pblock = *m_pblk; | |
2217 | int err = 0; | |
c8cc8816 RH |
2218 | int blkbits = mpd->inode->i_blkbits; |
2219 | ssize_t io_end_size = 0; | |
2220 | struct ext4_io_end_vec *io_end_vec = ext4_last_io_end_vec(io_end); | |
2943fdbc RH |
2221 | |
2222 | bh = head = page_buffers(page); | |
2223 | do { | |
2224 | if (lblk < mpd->map.m_lblk) | |
2225 | continue; | |
2226 | if (lblk >= mpd->map.m_lblk + mpd->map.m_len) { | |
2227 | /* | |
2228 | * Buffer after end of mapped extent. | |
2229 | * Find next buffer in the page to map. | |
2230 | */ | |
2231 | mpd->map.m_len = 0; | |
2232 | mpd->map.m_flags = 0; | |
c8cc8816 RH |
2233 | io_end_vec->size += io_end_size; |
2234 | io_end_size = 0; | |
2943fdbc | 2235 | |
2943fdbc RH |
2236 | err = mpage_process_page_bufs(mpd, head, bh, lblk); |
2237 | if (err > 0) | |
2238 | err = 0; | |
c8cc8816 RH |
2239 | if (!err && mpd->map.m_len && mpd->map.m_lblk > lblk) { |
2240 | io_end_vec = ext4_alloc_io_end_vec(io_end); | |
4d06bfb9 RH |
2241 | if (IS_ERR(io_end_vec)) { |
2242 | err = PTR_ERR(io_end_vec); | |
2243 | goto out; | |
2244 | } | |
c8cc8816 RH |
2245 | io_end_vec->offset = mpd->map.m_lblk << blkbits; |
2246 | } | |
2943fdbc RH |
2247 | *map_bh = true; |
2248 | goto out; | |
2249 | } | |
2250 | if (buffer_delay(bh)) { | |
2251 | clear_buffer_delay(bh); | |
2252 | bh->b_blocknr = pblock++; | |
2253 | } | |
2254 | clear_buffer_unwritten(bh); | |
c8cc8816 | 2255 | io_end_size += (1 << blkbits); |
2943fdbc | 2256 | } while (lblk++, (bh = bh->b_this_page) != head); |
c8cc8816 RH |
2257 | |
2258 | io_end_vec->size += io_end_size; | |
2259 | io_end_size = 0; | |
2943fdbc RH |
2260 | *map_bh = false; |
2261 | out: | |
2262 | *m_lblk = lblk; | |
2263 | *m_pblk = pblock; | |
2264 | return err; | |
2265 | } | |
2266 | ||
4e7ea81d JK |
2267 | /* |
2268 | * mpage_map_buffers - update buffers corresponding to changed extent and | |
2269 | * submit fully mapped pages for IO | |
2270 | * | |
2271 | * @mpd - description of extent to map, on return next extent to map | |
2272 | * | |
2273 | * Scan buffers corresponding to changed extent (we expect corresponding pages | |
2274 | * to be already locked) and update buffer state according to new extent state. | |
2275 | * We map delalloc buffers to their physical location, clear unwritten bits, | |
556615dc | 2276 | * and mark buffers as uninit when we perform writes to unwritten extents |
4e7ea81d JK |
2277 | * and do extent conversion after IO is finished. If the last page is not fully |
2278 | * mapped, we update @map to the next extent in the last page that needs | |
2279 | * mapping. Otherwise we submit the page for IO. | |
2280 | */ | |
2281 | static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd) | |
2282 | { | |
2283 | struct pagevec pvec; | |
2284 | int nr_pages, i; | |
2285 | struct inode *inode = mpd->inode; | |
09cbfeaf | 2286 | int bpp_bits = PAGE_SHIFT - inode->i_blkbits; |
4e7ea81d JK |
2287 | pgoff_t start, end; |
2288 | ext4_lblk_t lblk; | |
2943fdbc | 2289 | ext4_fsblk_t pblock; |
4e7ea81d | 2290 | int err; |
2943fdbc | 2291 | bool map_bh = false; |
4e7ea81d JK |
2292 | |
2293 | start = mpd->map.m_lblk >> bpp_bits; | |
2294 | end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits; | |
2295 | lblk = start << bpp_bits; | |
2296 | pblock = mpd->map.m_pblk; | |
2297 | ||
86679820 | 2298 | pagevec_init(&pvec); |
4e7ea81d | 2299 | while (start <= end) { |
2b85a617 | 2300 | nr_pages = pagevec_lookup_range(&pvec, inode->i_mapping, |
397162ff | 2301 | &start, end); |
4e7ea81d JK |
2302 | if (nr_pages == 0) |
2303 | break; | |
2304 | for (i = 0; i < nr_pages; i++) { | |
2305 | struct page *page = pvec.pages[i]; | |
2306 | ||
2943fdbc RH |
2307 | err = mpage_process_page(mpd, page, &lblk, &pblock, |
2308 | &map_bh); | |
4e7ea81d | 2309 | /* |
2943fdbc RH |
2310 | * If map_bh is true, means page may require further bh |
2311 | * mapping, or maybe the page was submitted for IO. | |
2312 | * So we return to call further extent mapping. | |
4e7ea81d | 2313 | */ |
2943fdbc RH |
2314 | if (err < 0 || map_bh == true) |
2315 | goto out; | |
4e7ea81d JK |
2316 | /* Page fully mapped - let IO run! */ |
2317 | err = mpage_submit_page(mpd, page); | |
2943fdbc RH |
2318 | if (err < 0) |
2319 | goto out; | |
4e7ea81d JK |
2320 | } |
2321 | pagevec_release(&pvec); | |
2322 | } | |
2323 | /* Extent fully mapped and matches with page boundary. We are done. */ | |
2324 | mpd->map.m_len = 0; | |
2325 | mpd->map.m_flags = 0; | |
2326 | return 0; | |
2943fdbc RH |
2327 | out: |
2328 | pagevec_release(&pvec); | |
2329 | return err; | |
4e7ea81d JK |
2330 | } |
2331 | ||
2332 | static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd) | |
2333 | { | |
2334 | struct inode *inode = mpd->inode; | |
2335 | struct ext4_map_blocks *map = &mpd->map; | |
2336 | int get_blocks_flags; | |
090f32ee | 2337 | int err, dioread_nolock; |
4e7ea81d JK |
2338 | |
2339 | trace_ext4_da_write_pages_extent(inode, map); | |
2340 | /* | |
2341 | * Call ext4_map_blocks() to allocate any delayed allocation blocks, or | |
556615dc | 2342 | * to convert an unwritten extent to be initialized (in the case |
4e7ea81d JK |
2343 | * where we have written into one or more preallocated blocks). It is |
2344 | * possible that we're going to need more metadata blocks than | |
2345 | * previously reserved. However we must not fail because we're in | |
2346 | * writeback and there is nothing we can do about it so it might result | |
2347 | * in data loss. So use reserved blocks to allocate metadata if | |
2348 | * possible. | |
2349 | * | |
754cfed6 TT |
2350 | * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if |
2351 | * the blocks in question are delalloc blocks. This indicates | |
2352 | * that the blocks and quotas has already been checked when | |
2353 | * the data was copied into the page cache. | |
4e7ea81d JK |
2354 | */ |
2355 | get_blocks_flags = EXT4_GET_BLOCKS_CREATE | | |
ee0876bc JK |
2356 | EXT4_GET_BLOCKS_METADATA_NOFAIL | |
2357 | EXT4_GET_BLOCKS_IO_SUBMIT; | |
090f32ee LC |
2358 | dioread_nolock = ext4_should_dioread_nolock(inode); |
2359 | if (dioread_nolock) | |
4e7ea81d JK |
2360 | get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT; |
2361 | if (map->m_flags & (1 << BH_Delay)) | |
2362 | get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE; | |
2363 | ||
2364 | err = ext4_map_blocks(handle, inode, map, get_blocks_flags); | |
2365 | if (err < 0) | |
2366 | return err; | |
090f32ee | 2367 | if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) { |
6b523df4 JK |
2368 | if (!mpd->io_submit.io_end->handle && |
2369 | ext4_handle_valid(handle)) { | |
2370 | mpd->io_submit.io_end->handle = handle->h_rsv_handle; | |
2371 | handle->h_rsv_handle = NULL; | |
2372 | } | |
3613d228 | 2373 | ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end); |
6b523df4 | 2374 | } |
4e7ea81d JK |
2375 | |
2376 | BUG_ON(map->m_len == 0); | |
4e7ea81d JK |
2377 | return 0; |
2378 | } | |
2379 | ||
2380 | /* | |
2381 | * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length | |
2382 | * mpd->len and submit pages underlying it for IO | |
2383 | * | |
2384 | * @handle - handle for journal operations | |
2385 | * @mpd - extent to map | |
7534e854 JK |
2386 | * @give_up_on_write - we set this to true iff there is a fatal error and there |
2387 | * is no hope of writing the data. The caller should discard | |
2388 | * dirty pages to avoid infinite loops. | |
4e7ea81d JK |
2389 | * |
2390 | * The function maps extent starting at mpd->lblk of length mpd->len. If it is | |
2391 | * delayed, blocks are allocated, if it is unwritten, we may need to convert | |
2392 | * them to initialized or split the described range from larger unwritten | |
2393 | * extent. Note that we need not map all the described range since allocation | |
2394 | * can return less blocks or the range is covered by more unwritten extents. We | |
2395 | * cannot map more because we are limited by reserved transaction credits. On | |
2396 | * the other hand we always make sure that the last touched page is fully | |
2397 | * mapped so that it can be written out (and thus forward progress is | |
2398 | * guaranteed). After mapping we submit all mapped pages for IO. | |
2399 | */ | |
2400 | static int mpage_map_and_submit_extent(handle_t *handle, | |
cb530541 TT |
2401 | struct mpage_da_data *mpd, |
2402 | bool *give_up_on_write) | |
4e7ea81d JK |
2403 | { |
2404 | struct inode *inode = mpd->inode; | |
2405 | struct ext4_map_blocks *map = &mpd->map; | |
2406 | int err; | |
2407 | loff_t disksize; | |
6603120e | 2408 | int progress = 0; |
c8cc8816 | 2409 | ext4_io_end_t *io_end = mpd->io_submit.io_end; |
4d06bfb9 | 2410 | struct ext4_io_end_vec *io_end_vec; |
4e7ea81d | 2411 | |
4d06bfb9 RH |
2412 | io_end_vec = ext4_alloc_io_end_vec(io_end); |
2413 | if (IS_ERR(io_end_vec)) | |
2414 | return PTR_ERR(io_end_vec); | |
c8cc8816 | 2415 | io_end_vec->offset = ((loff_t)map->m_lblk) << inode->i_blkbits; |
27d7c4ed | 2416 | do { |
4e7ea81d JK |
2417 | err = mpage_map_one_extent(handle, mpd); |
2418 | if (err < 0) { | |
2419 | struct super_block *sb = inode->i_sb; | |
2420 | ||
0db1ff22 TT |
2421 | if (ext4_forced_shutdown(EXT4_SB(sb)) || |
2422 | EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED) | |
cb530541 | 2423 | goto invalidate_dirty_pages; |
4e7ea81d | 2424 | /* |
cb530541 TT |
2425 | * Let the uper layers retry transient errors. |
2426 | * In the case of ENOSPC, if ext4_count_free_blocks() | |
2427 | * is non-zero, a commit should free up blocks. | |
4e7ea81d | 2428 | */ |
cb530541 | 2429 | if ((err == -ENOMEM) || |
6603120e DM |
2430 | (err == -ENOSPC && ext4_count_free_clusters(sb))) { |
2431 | if (progress) | |
2432 | goto update_disksize; | |
cb530541 | 2433 | return err; |
6603120e | 2434 | } |
cb530541 TT |
2435 | ext4_msg(sb, KERN_CRIT, |
2436 | "Delayed block allocation failed for " | |
2437 | "inode %lu at logical offset %llu with" | |
2438 | " max blocks %u with error %d", | |
2439 | inode->i_ino, | |
2440 | (unsigned long long)map->m_lblk, | |
2441 | (unsigned)map->m_len, -err); | |
2442 | ext4_msg(sb, KERN_CRIT, | |
2443 | "This should not happen!! Data will " | |
2444 | "be lost\n"); | |
2445 | if (err == -ENOSPC) | |
2446 | ext4_print_free_blocks(inode); | |
2447 | invalidate_dirty_pages: | |
2448 | *give_up_on_write = true; | |
4e7ea81d JK |
2449 | return err; |
2450 | } | |
6603120e | 2451 | progress = 1; |
4e7ea81d JK |
2452 | /* |
2453 | * Update buffer state, submit mapped pages, and get us new | |
2454 | * extent to map | |
2455 | */ | |
2456 | err = mpage_map_and_submit_buffers(mpd); | |
2457 | if (err < 0) | |
6603120e | 2458 | goto update_disksize; |
27d7c4ed | 2459 | } while (map->m_len); |
4e7ea81d | 2460 | |
6603120e | 2461 | update_disksize: |
622cad13 TT |
2462 | /* |
2463 | * Update on-disk size after IO is submitted. Races with | |
2464 | * truncate are avoided by checking i_size under i_data_sem. | |
2465 | */ | |
09cbfeaf | 2466 | disksize = ((loff_t)mpd->first_page) << PAGE_SHIFT; |
35df4299 | 2467 | if (disksize > READ_ONCE(EXT4_I(inode)->i_disksize)) { |
4e7ea81d | 2468 | int err2; |
622cad13 TT |
2469 | loff_t i_size; |
2470 | ||
2471 | down_write(&EXT4_I(inode)->i_data_sem); | |
2472 | i_size = i_size_read(inode); | |
2473 | if (disksize > i_size) | |
2474 | disksize = i_size; | |
2475 | if (disksize > EXT4_I(inode)->i_disksize) | |
2476 | EXT4_I(inode)->i_disksize = disksize; | |
622cad13 | 2477 | up_write(&EXT4_I(inode)->i_data_sem); |
b907f2d5 | 2478 | err2 = ext4_mark_inode_dirty(handle, inode); |
878520ac | 2479 | if (err2) { |
54d3adbc TT |
2480 | ext4_error_err(inode->i_sb, -err2, |
2481 | "Failed to mark inode %lu dirty", | |
2482 | inode->i_ino); | |
878520ac | 2483 | } |
4e7ea81d JK |
2484 | if (!err) |
2485 | err = err2; | |
2486 | } | |
2487 | return err; | |
2488 | } | |
2489 | ||
fffb2739 JK |
2490 | /* |
2491 | * Calculate the total number of credits to reserve for one writepages | |
20970ba6 | 2492 | * iteration. This is called from ext4_writepages(). We map an extent of |
70261f56 | 2493 | * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping |
fffb2739 JK |
2494 | * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN + |
2495 | * bpp - 1 blocks in bpp different extents. | |
2496 | */ | |
525f4ed8 MC |
2497 | static int ext4_da_writepages_trans_blocks(struct inode *inode) |
2498 | { | |
fffb2739 | 2499 | int bpp = ext4_journal_blocks_per_page(inode); |
525f4ed8 | 2500 | |
fffb2739 JK |
2501 | return ext4_meta_trans_blocks(inode, |
2502 | MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp); | |
525f4ed8 | 2503 | } |
61628a3f | 2504 | |
8e48dcfb | 2505 | /* |
4e7ea81d JK |
2506 | * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages |
2507 | * and underlying extent to map | |
2508 | * | |
2509 | * @mpd - where to look for pages | |
2510 | * | |
2511 | * Walk dirty pages in the mapping. If they are fully mapped, submit them for | |
2512 | * IO immediately. When we find a page which isn't mapped we start accumulating | |
2513 | * extent of buffers underlying these pages that needs mapping (formed by | |
2514 | * either delayed or unwritten buffers). We also lock the pages containing | |
2515 | * these buffers. The extent found is returned in @mpd structure (starting at | |
2516 | * mpd->lblk with length mpd->len blocks). | |
2517 | * | |
2518 | * Note that this function can attach bios to one io_end structure which are | |
2519 | * neither logically nor physically contiguous. Although it may seem as an | |
2520 | * unnecessary complication, it is actually inevitable in blocksize < pagesize | |
2521 | * case as we need to track IO to all buffers underlying a page in one io_end. | |
8e48dcfb | 2522 | */ |
4e7ea81d | 2523 | static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd) |
8e48dcfb | 2524 | { |
4e7ea81d JK |
2525 | struct address_space *mapping = mpd->inode->i_mapping; |
2526 | struct pagevec pvec; | |
2527 | unsigned int nr_pages; | |
aeac589a | 2528 | long left = mpd->wbc->nr_to_write; |
4e7ea81d JK |
2529 | pgoff_t index = mpd->first_page; |
2530 | pgoff_t end = mpd->last_page; | |
10bbd235 | 2531 | xa_mark_t tag; |
4e7ea81d JK |
2532 | int i, err = 0; |
2533 | int blkbits = mpd->inode->i_blkbits; | |
2534 | ext4_lblk_t lblk; | |
2535 | struct buffer_head *head; | |
8e48dcfb | 2536 | |
4e7ea81d | 2537 | if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages) |
5b41d924 ES |
2538 | tag = PAGECACHE_TAG_TOWRITE; |
2539 | else | |
2540 | tag = PAGECACHE_TAG_DIRTY; | |
2541 | ||
86679820 | 2542 | pagevec_init(&pvec); |
4e7ea81d JK |
2543 | mpd->map.m_len = 0; |
2544 | mpd->next_page = index; | |
4f01b02c | 2545 | while (index <= end) { |
dc7f3e86 | 2546 | nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end, |
67fd707f | 2547 | tag); |
8e48dcfb | 2548 | if (nr_pages == 0) |
4e7ea81d | 2549 | goto out; |
8e48dcfb TT |
2550 | |
2551 | for (i = 0; i < nr_pages; i++) { | |
2552 | struct page *page = pvec.pages[i]; | |
2553 | ||
aeac589a ML |
2554 | /* |
2555 | * Accumulated enough dirty pages? This doesn't apply | |
2556 | * to WB_SYNC_ALL mode. For integrity sync we have to | |
2557 | * keep going because someone may be concurrently | |
2558 | * dirtying pages, and we might have synced a lot of | |
2559 | * newly appeared dirty pages, but have not synced all | |
2560 | * of the old dirty pages. | |
2561 | */ | |
2562 | if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0) | |
2563 | goto out; | |
2564 | ||
4e7ea81d JK |
2565 | /* If we can't merge this page, we are done. */ |
2566 | if (mpd->map.m_len > 0 && mpd->next_page != page->index) | |
2567 | goto out; | |
78aaced3 | 2568 | |
8e48dcfb | 2569 | lock_page(page); |
8e48dcfb | 2570 | /* |
4e7ea81d JK |
2571 | * If the page is no longer dirty, or its mapping no |
2572 | * longer corresponds to inode we are writing (which | |
2573 | * means it has been truncated or invalidated), or the | |
2574 | * page is already under writeback and we are not doing | |
2575 | * a data integrity writeback, skip the page | |
8e48dcfb | 2576 | */ |
4f01b02c TT |
2577 | if (!PageDirty(page) || |
2578 | (PageWriteback(page) && | |
4e7ea81d | 2579 | (mpd->wbc->sync_mode == WB_SYNC_NONE)) || |
4f01b02c | 2580 | unlikely(page->mapping != mapping)) { |
8e48dcfb TT |
2581 | unlock_page(page); |
2582 | continue; | |
2583 | } | |
2584 | ||
7cb1a535 | 2585 | wait_on_page_writeback(page); |
8e48dcfb | 2586 | BUG_ON(PageWriteback(page)); |
8e48dcfb | 2587 | |
4e7ea81d | 2588 | if (mpd->map.m_len == 0) |
8eb9e5ce | 2589 | mpd->first_page = page->index; |
8eb9e5ce | 2590 | mpd->next_page = page->index + 1; |
f8bec370 | 2591 | /* Add all dirty buffers to mpd */ |
4e7ea81d | 2592 | lblk = ((ext4_lblk_t)page->index) << |
09cbfeaf | 2593 | (PAGE_SHIFT - blkbits); |
f8bec370 | 2594 | head = page_buffers(page); |
5f1132b2 JK |
2595 | err = mpage_process_page_bufs(mpd, head, head, lblk); |
2596 | if (err <= 0) | |
4e7ea81d | 2597 | goto out; |
5f1132b2 | 2598 | err = 0; |
aeac589a | 2599 | left--; |
8e48dcfb TT |
2600 | } |
2601 | pagevec_release(&pvec); | |
2602 | cond_resched(); | |
2603 | } | |
4f01b02c | 2604 | return 0; |
8eb9e5ce TT |
2605 | out: |
2606 | pagevec_release(&pvec); | |
4e7ea81d | 2607 | return err; |
8e48dcfb TT |
2608 | } |
2609 | ||
20970ba6 TT |
2610 | static int ext4_writepages(struct address_space *mapping, |
2611 | struct writeback_control *wbc) | |
64769240 | 2612 | { |
4e7ea81d JK |
2613 | pgoff_t writeback_index = 0; |
2614 | long nr_to_write = wbc->nr_to_write; | |
22208ded | 2615 | int range_whole = 0; |
4e7ea81d | 2616 | int cycled = 1; |
61628a3f | 2617 | handle_t *handle = NULL; |
df22291f | 2618 | struct mpage_da_data mpd; |
5e745b04 | 2619 | struct inode *inode = mapping->host; |
6b523df4 | 2620 | int needed_blocks, rsv_blocks = 0, ret = 0; |
5e745b04 | 2621 | struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb); |
4e7ea81d | 2622 | bool done; |
1bce63d1 | 2623 | struct blk_plug plug; |
cb530541 | 2624 | bool give_up_on_write = false; |
61628a3f | 2625 | |
0db1ff22 TT |
2626 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
2627 | return -EIO; | |
2628 | ||
bbd55937 | 2629 | percpu_down_read(&sbi->s_writepages_rwsem); |
20970ba6 | 2630 | trace_ext4_writepages(inode, wbc); |
ba80b101 | 2631 | |
61628a3f MC |
2632 | /* |
2633 | * No pages to write? This is mainly a kludge to avoid starting | |
2634 | * a transaction for special inodes like journal inode on last iput() | |
2635 | * because that could violate lock ordering on umount | |
2636 | */ | |
a1d6cc56 | 2637 | if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) |
bbf023c7 | 2638 | goto out_writepages; |
2a21e37e | 2639 | |
20970ba6 | 2640 | if (ext4_should_journal_data(inode)) { |
043d20d1 | 2641 | ret = generic_writepages(mapping, wbc); |
bbf023c7 | 2642 | goto out_writepages; |
20970ba6 TT |
2643 | } |
2644 | ||
2a21e37e TT |
2645 | /* |
2646 | * If the filesystem has aborted, it is read-only, so return | |
2647 | * right away instead of dumping stack traces later on that | |
2648 | * will obscure the real source of the problem. We test | |
1751e8a6 | 2649 | * EXT4_MF_FS_ABORTED instead of sb->s_flag's SB_RDONLY because |
2a21e37e | 2650 | * the latter could be true if the filesystem is mounted |
20970ba6 | 2651 | * read-only, and in that case, ext4_writepages should |
2a21e37e TT |
2652 | * *never* be called, so if that ever happens, we would want |
2653 | * the stack trace. | |
2654 | */ | |
0db1ff22 TT |
2655 | if (unlikely(ext4_forced_shutdown(EXT4_SB(mapping->host->i_sb)) || |
2656 | sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) { | |
bbf023c7 ML |
2657 | ret = -EROFS; |
2658 | goto out_writepages; | |
2659 | } | |
2a21e37e | 2660 | |
4e7ea81d JK |
2661 | /* |
2662 | * If we have inline data and arrive here, it means that | |
2663 | * we will soon create the block for the 1st page, so | |
2664 | * we'd better clear the inline data here. | |
2665 | */ | |
2666 | if (ext4_has_inline_data(inode)) { | |
2667 | /* Just inode will be modified... */ | |
2668 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 1); | |
2669 | if (IS_ERR(handle)) { | |
2670 | ret = PTR_ERR(handle); | |
2671 | goto out_writepages; | |
2672 | } | |
2673 | BUG_ON(ext4_test_inode_state(inode, | |
2674 | EXT4_STATE_MAY_INLINE_DATA)); | |
2675 | ext4_destroy_inline_data(handle, inode); | |
2676 | ext4_journal_stop(handle); | |
2677 | } | |
2678 | ||
4e343231 | 2679 | if (ext4_should_dioread_nolock(inode)) { |
2680 | /* | |
2681 | * We may need to convert up to one extent per block in | |
2682 | * the page and we may dirty the inode. | |
2683 | */ | |
2684 | rsv_blocks = 1 + ext4_chunk_trans_blocks(inode, | |
2685 | PAGE_SIZE >> inode->i_blkbits); | |
2686 | } | |
2687 | ||
22208ded AK |
2688 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
2689 | range_whole = 1; | |
61628a3f | 2690 | |
2acf2c26 | 2691 | if (wbc->range_cyclic) { |
4e7ea81d JK |
2692 | writeback_index = mapping->writeback_index; |
2693 | if (writeback_index) | |
2acf2c26 | 2694 | cycled = 0; |
4e7ea81d JK |
2695 | mpd.first_page = writeback_index; |
2696 | mpd.last_page = -1; | |
5b41d924 | 2697 | } else { |
09cbfeaf KS |
2698 | mpd.first_page = wbc->range_start >> PAGE_SHIFT; |
2699 | mpd.last_page = wbc->range_end >> PAGE_SHIFT; | |
5b41d924 | 2700 | } |
a1d6cc56 | 2701 | |
4e7ea81d JK |
2702 | mpd.inode = inode; |
2703 | mpd.wbc = wbc; | |
2704 | ext4_io_submit_init(&mpd.io_submit, wbc); | |
2acf2c26 | 2705 | retry: |
6e6938b6 | 2706 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
4e7ea81d JK |
2707 | tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page); |
2708 | done = false; | |
1bce63d1 | 2709 | blk_start_plug(&plug); |
dddbd6ac JK |
2710 | |
2711 | /* | |
2712 | * First writeback pages that don't need mapping - we can avoid | |
2713 | * starting a transaction unnecessarily and also avoid being blocked | |
2714 | * in the block layer on device congestion while having transaction | |
2715 | * started. | |
2716 | */ | |
2717 | mpd.do_map = 0; | |
2718 | mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL); | |
2719 | if (!mpd.io_submit.io_end) { | |
2720 | ret = -ENOMEM; | |
2721 | goto unplug; | |
2722 | } | |
2723 | ret = mpage_prepare_extent_to_map(&mpd); | |
a297b2fc XW |
2724 | /* Unlock pages we didn't use */ |
2725 | mpage_release_unused_pages(&mpd, false); | |
dddbd6ac JK |
2726 | /* Submit prepared bio */ |
2727 | ext4_io_submit(&mpd.io_submit); | |
2728 | ext4_put_io_end_defer(mpd.io_submit.io_end); | |
2729 | mpd.io_submit.io_end = NULL; | |
dddbd6ac JK |
2730 | if (ret < 0) |
2731 | goto unplug; | |
2732 | ||
4e7ea81d JK |
2733 | while (!done && mpd.first_page <= mpd.last_page) { |
2734 | /* For each extent of pages we use new io_end */ | |
2735 | mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL); | |
2736 | if (!mpd.io_submit.io_end) { | |
2737 | ret = -ENOMEM; | |
2738 | break; | |
2739 | } | |
a1d6cc56 AK |
2740 | |
2741 | /* | |
4e7ea81d JK |
2742 | * We have two constraints: We find one extent to map and we |
2743 | * must always write out whole page (makes a difference when | |
2744 | * blocksize < pagesize) so that we don't block on IO when we | |
2745 | * try to write out the rest of the page. Journalled mode is | |
2746 | * not supported by delalloc. | |
a1d6cc56 AK |
2747 | */ |
2748 | BUG_ON(ext4_should_journal_data(inode)); | |
525f4ed8 | 2749 | needed_blocks = ext4_da_writepages_trans_blocks(inode); |
a1d6cc56 | 2750 | |
4e7ea81d | 2751 | /* start a new transaction */ |
6b523df4 JK |
2752 | handle = ext4_journal_start_with_reserve(inode, |
2753 | EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks); | |
61628a3f MC |
2754 | if (IS_ERR(handle)) { |
2755 | ret = PTR_ERR(handle); | |
1693918e | 2756 | ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: " |
fbe845dd | 2757 | "%ld pages, ino %lu; err %d", __func__, |
a1d6cc56 | 2758 | wbc->nr_to_write, inode->i_ino, ret); |
4e7ea81d JK |
2759 | /* Release allocated io_end */ |
2760 | ext4_put_io_end(mpd.io_submit.io_end); | |
dddbd6ac | 2761 | mpd.io_submit.io_end = NULL; |
4e7ea81d | 2762 | break; |
61628a3f | 2763 | } |
dddbd6ac | 2764 | mpd.do_map = 1; |
f63e6005 | 2765 | |
4e7ea81d JK |
2766 | trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc); |
2767 | ret = mpage_prepare_extent_to_map(&mpd); | |
2768 | if (!ret) { | |
2769 | if (mpd.map.m_len) | |
cb530541 TT |
2770 | ret = mpage_map_and_submit_extent(handle, &mpd, |
2771 | &give_up_on_write); | |
4e7ea81d JK |
2772 | else { |
2773 | /* | |
2774 | * We scanned the whole range (or exhausted | |
2775 | * nr_to_write), submitted what was mapped and | |
2776 | * didn't find anything needing mapping. We are | |
2777 | * done. | |
2778 | */ | |
2779 | done = true; | |
2780 | } | |
f63e6005 | 2781 | } |
646caa9c JK |
2782 | /* |
2783 | * Caution: If the handle is synchronous, | |
2784 | * ext4_journal_stop() can wait for transaction commit | |
2785 | * to finish which may depend on writeback of pages to | |
2786 | * complete or on page lock to be released. In that | |
2787 | * case, we have to wait until after after we have | |
2788 | * submitted all the IO, released page locks we hold, | |
2789 | * and dropped io_end reference (for extent conversion | |
2790 | * to be able to complete) before stopping the handle. | |
2791 | */ | |
2792 | if (!ext4_handle_valid(handle) || handle->h_sync == 0) { | |
2793 | ext4_journal_stop(handle); | |
2794 | handle = NULL; | |
dddbd6ac | 2795 | mpd.do_map = 0; |
646caa9c | 2796 | } |
4e7ea81d | 2797 | /* Unlock pages we didn't use */ |
cb530541 | 2798 | mpage_release_unused_pages(&mpd, give_up_on_write); |
a297b2fc XW |
2799 | /* Submit prepared bio */ |
2800 | ext4_io_submit(&mpd.io_submit); | |
2801 | ||
646caa9c JK |
2802 | /* |
2803 | * Drop our io_end reference we got from init. We have | |
2804 | * to be careful and use deferred io_end finishing if | |
2805 | * we are still holding the transaction as we can | |
2806 | * release the last reference to io_end which may end | |
2807 | * up doing unwritten extent conversion. | |
2808 | */ | |
2809 | if (handle) { | |
2810 | ext4_put_io_end_defer(mpd.io_submit.io_end); | |
2811 | ext4_journal_stop(handle); | |
2812 | } else | |
2813 | ext4_put_io_end(mpd.io_submit.io_end); | |
dddbd6ac | 2814 | mpd.io_submit.io_end = NULL; |
4e7ea81d JK |
2815 | |
2816 | if (ret == -ENOSPC && sbi->s_journal) { | |
2817 | /* | |
2818 | * Commit the transaction which would | |
22208ded AK |
2819 | * free blocks released in the transaction |
2820 | * and try again | |
2821 | */ | |
df22291f | 2822 | jbd2_journal_force_commit_nested(sbi->s_journal); |
22208ded | 2823 | ret = 0; |
4e7ea81d JK |
2824 | continue; |
2825 | } | |
2826 | /* Fatal error - ENOMEM, EIO... */ | |
2827 | if (ret) | |
61628a3f | 2828 | break; |
a1d6cc56 | 2829 | } |
dddbd6ac | 2830 | unplug: |
1bce63d1 | 2831 | blk_finish_plug(&plug); |
9c12a831 | 2832 | if (!ret && !cycled && wbc->nr_to_write > 0) { |
2acf2c26 | 2833 | cycled = 1; |
4e7ea81d JK |
2834 | mpd.last_page = writeback_index - 1; |
2835 | mpd.first_page = 0; | |
2acf2c26 AK |
2836 | goto retry; |
2837 | } | |
22208ded AK |
2838 | |
2839 | /* Update index */ | |
22208ded AK |
2840 | if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) |
2841 | /* | |
4e7ea81d | 2842 | * Set the writeback_index so that range_cyclic |
22208ded AK |
2843 | * mode will write it back later |
2844 | */ | |
4e7ea81d | 2845 | mapping->writeback_index = mpd.first_page; |
a1d6cc56 | 2846 | |
61628a3f | 2847 | out_writepages: |
20970ba6 TT |
2848 | trace_ext4_writepages_result(inode, wbc, ret, |
2849 | nr_to_write - wbc->nr_to_write); | |
bbd55937 | 2850 | percpu_up_read(&sbi->s_writepages_rwsem); |
61628a3f | 2851 | return ret; |
64769240 AT |
2852 | } |
2853 | ||
5f0663bb DW |
2854 | static int ext4_dax_writepages(struct address_space *mapping, |
2855 | struct writeback_control *wbc) | |
2856 | { | |
2857 | int ret; | |
2858 | long nr_to_write = wbc->nr_to_write; | |
2859 | struct inode *inode = mapping->host; | |
2860 | struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb); | |
2861 | ||
2862 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) | |
2863 | return -EIO; | |
2864 | ||
bbd55937 | 2865 | percpu_down_read(&sbi->s_writepages_rwsem); |
5f0663bb DW |
2866 | trace_ext4_writepages(inode, wbc); |
2867 | ||
3f666c56 | 2868 | ret = dax_writeback_mapping_range(mapping, sbi->s_daxdev, wbc); |
5f0663bb DW |
2869 | trace_ext4_writepages_result(inode, wbc, ret, |
2870 | nr_to_write - wbc->nr_to_write); | |
bbd55937 | 2871 | percpu_up_read(&sbi->s_writepages_rwsem); |
5f0663bb DW |
2872 | return ret; |
2873 | } | |
2874 | ||
79f0be8d AK |
2875 | static int ext4_nonda_switch(struct super_block *sb) |
2876 | { | |
5c1ff336 | 2877 | s64 free_clusters, dirty_clusters; |
79f0be8d AK |
2878 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
2879 | ||
2880 | /* | |
2881 | * switch to non delalloc mode if we are running low | |
2882 | * on free block. The free block accounting via percpu | |
179f7ebf | 2883 | * counters can get slightly wrong with percpu_counter_batch getting |
79f0be8d AK |
2884 | * accumulated on each CPU without updating global counters |
2885 | * Delalloc need an accurate free block accounting. So switch | |
2886 | * to non delalloc when we are near to error range. | |
2887 | */ | |
5c1ff336 EW |
2888 | free_clusters = |
2889 | percpu_counter_read_positive(&sbi->s_freeclusters_counter); | |
2890 | dirty_clusters = | |
2891 | percpu_counter_read_positive(&sbi->s_dirtyclusters_counter); | |
00d4e736 TT |
2892 | /* |
2893 | * Start pushing delalloc when 1/2 of free blocks are dirty. | |
2894 | */ | |
5c1ff336 | 2895 | if (dirty_clusters && (free_clusters < 2 * dirty_clusters)) |
10ee27a0 | 2896 | try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE); |
00d4e736 | 2897 | |
5c1ff336 EW |
2898 | if (2 * free_clusters < 3 * dirty_clusters || |
2899 | free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) { | |
79f0be8d | 2900 | /* |
c8afb446 ES |
2901 | * free block count is less than 150% of dirty blocks |
2902 | * or free blocks is less than watermark | |
79f0be8d AK |
2903 | */ |
2904 | return 1; | |
2905 | } | |
2906 | return 0; | |
2907 | } | |
2908 | ||
0ff8947f ES |
2909 | /* We always reserve for an inode update; the superblock could be there too */ |
2910 | static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len) | |
2911 | { | |
e2b911c5 | 2912 | if (likely(ext4_has_feature_large_file(inode->i_sb))) |
0ff8947f ES |
2913 | return 1; |
2914 | ||
2915 | if (pos + len <= 0x7fffffffULL) | |
2916 | return 1; | |
2917 | ||
2918 | /* We might need to update the superblock to set LARGE_FILE */ | |
2919 | return 2; | |
2920 | } | |
2921 | ||
64769240 | 2922 | static int ext4_da_write_begin(struct file *file, struct address_space *mapping, |
de9a55b8 TT |
2923 | loff_t pos, unsigned len, unsigned flags, |
2924 | struct page **pagep, void **fsdata) | |
64769240 | 2925 | { |
72b8ab9d | 2926 | int ret, retries = 0; |
64769240 AT |
2927 | struct page *page; |
2928 | pgoff_t index; | |
64769240 AT |
2929 | struct inode *inode = mapping->host; |
2930 | handle_t *handle; | |
2931 | ||
0db1ff22 TT |
2932 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
2933 | return -EIO; | |
2934 | ||
09cbfeaf | 2935 | index = pos >> PAGE_SHIFT; |
79f0be8d | 2936 | |
c93d8f88 EB |
2937 | if (ext4_nonda_switch(inode->i_sb) || S_ISLNK(inode->i_mode) || |
2938 | ext4_verity_in_progress(inode)) { | |
79f0be8d AK |
2939 | *fsdata = (void *)FALL_BACK_TO_NONDELALLOC; |
2940 | return ext4_write_begin(file, mapping, pos, | |
2941 | len, flags, pagep, fsdata); | |
2942 | } | |
2943 | *fsdata = (void *)0; | |
9bffad1e | 2944 | trace_ext4_da_write_begin(inode, pos, len, flags); |
9c3569b5 TM |
2945 | |
2946 | if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { | |
2947 | ret = ext4_da_write_inline_data_begin(mapping, inode, | |
2948 | pos, len, flags, | |
2949 | pagep, fsdata); | |
2950 | if (ret < 0) | |
47564bfb TT |
2951 | return ret; |
2952 | if (ret == 1) | |
2953 | return 0; | |
9c3569b5 TM |
2954 | } |
2955 | ||
47564bfb TT |
2956 | /* |
2957 | * grab_cache_page_write_begin() can take a long time if the | |
2958 | * system is thrashing due to memory pressure, or if the page | |
2959 | * is being written back. So grab it first before we start | |
2960 | * the transaction handle. This also allows us to allocate | |
2961 | * the page (if needed) without using GFP_NOFS. | |
2962 | */ | |
2963 | retry_grab: | |
2964 | page = grab_cache_page_write_begin(mapping, index, flags); | |
2965 | if (!page) | |
2966 | return -ENOMEM; | |
2967 | unlock_page(page); | |
2968 | ||
64769240 AT |
2969 | /* |
2970 | * With delayed allocation, we don't log the i_disksize update | |
2971 | * if there is delayed block allocation. But we still need | |
2972 | * to journalling the i_disksize update if writes to the end | |
2973 | * of file which has an already mapped buffer. | |
2974 | */ | |
47564bfb | 2975 | retry_journal: |
0ff8947f ES |
2976 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
2977 | ext4_da_write_credits(inode, pos, len)); | |
64769240 | 2978 | if (IS_ERR(handle)) { |
09cbfeaf | 2979 | put_page(page); |
47564bfb | 2980 | return PTR_ERR(handle); |
64769240 AT |
2981 | } |
2982 | ||
47564bfb TT |
2983 | lock_page(page); |
2984 | if (page->mapping != mapping) { | |
2985 | /* The page got truncated from under us */ | |
2986 | unlock_page(page); | |
09cbfeaf | 2987 | put_page(page); |
d5a0d4f7 | 2988 | ext4_journal_stop(handle); |
47564bfb | 2989 | goto retry_grab; |
d5a0d4f7 | 2990 | } |
47564bfb | 2991 | /* In case writeback began while the page was unlocked */ |
7afe5aa5 | 2992 | wait_for_stable_page(page); |
64769240 | 2993 | |
643fa961 | 2994 | #ifdef CONFIG_FS_ENCRYPTION |
2058f83a MH |
2995 | ret = ext4_block_write_begin(page, pos, len, |
2996 | ext4_da_get_block_prep); | |
2997 | #else | |
6e1db88d | 2998 | ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep); |
2058f83a | 2999 | #endif |
64769240 AT |
3000 | if (ret < 0) { |
3001 | unlock_page(page); | |
3002 | ext4_journal_stop(handle); | |
ae4d5372 AK |
3003 | /* |
3004 | * block_write_begin may have instantiated a few blocks | |
3005 | * outside i_size. Trim these off again. Don't need | |
3006 | * i_size_read because we hold i_mutex. | |
3007 | */ | |
3008 | if (pos + len > inode->i_size) | |
b9a4207d | 3009 | ext4_truncate_failed_write(inode); |
47564bfb TT |
3010 | |
3011 | if (ret == -ENOSPC && | |
3012 | ext4_should_retry_alloc(inode->i_sb, &retries)) | |
3013 | goto retry_journal; | |
3014 | ||
09cbfeaf | 3015 | put_page(page); |
47564bfb | 3016 | return ret; |
64769240 AT |
3017 | } |
3018 | ||
47564bfb | 3019 | *pagep = page; |
64769240 AT |
3020 | return ret; |
3021 | } | |
3022 | ||
632eaeab MC |
3023 | /* |
3024 | * Check if we should update i_disksize | |
3025 | * when write to the end of file but not require block allocation | |
3026 | */ | |
3027 | static int ext4_da_should_update_i_disksize(struct page *page, | |
de9a55b8 | 3028 | unsigned long offset) |
632eaeab MC |
3029 | { |
3030 | struct buffer_head *bh; | |
3031 | struct inode *inode = page->mapping->host; | |
3032 | unsigned int idx; | |
3033 | int i; | |
3034 | ||
3035 | bh = page_buffers(page); | |
3036 | idx = offset >> inode->i_blkbits; | |
3037 | ||
af5bc92d | 3038 | for (i = 0; i < idx; i++) |
632eaeab MC |
3039 | bh = bh->b_this_page; |
3040 | ||
29fa89d0 | 3041 | if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh)) |
632eaeab MC |
3042 | return 0; |
3043 | return 1; | |
3044 | } | |
3045 | ||
64769240 | 3046 | static int ext4_da_write_end(struct file *file, |
de9a55b8 TT |
3047 | struct address_space *mapping, |
3048 | loff_t pos, unsigned len, unsigned copied, | |
3049 | struct page *page, void *fsdata) | |
64769240 AT |
3050 | { |
3051 | struct inode *inode = mapping->host; | |
3052 | int ret = 0, ret2; | |
3053 | handle_t *handle = ext4_journal_current_handle(); | |
3054 | loff_t new_i_size; | |
632eaeab | 3055 | unsigned long start, end; |
79f0be8d AK |
3056 | int write_mode = (int)(unsigned long)fsdata; |
3057 | ||
74d553aa TT |
3058 | if (write_mode == FALL_BACK_TO_NONDELALLOC) |
3059 | return ext4_write_end(file, mapping, pos, | |
3060 | len, copied, page, fsdata); | |
632eaeab | 3061 | |
9bffad1e | 3062 | trace_ext4_da_write_end(inode, pos, len, copied); |
09cbfeaf | 3063 | start = pos & (PAGE_SIZE - 1); |
af5bc92d | 3064 | end = start + copied - 1; |
64769240 AT |
3065 | |
3066 | /* | |
3067 | * generic_write_end() will run mark_inode_dirty() if i_size | |
3068 | * changes. So let's piggyback the i_disksize mark_inode_dirty | |
3069 | * into that. | |
3070 | */ | |
64769240 | 3071 | new_i_size = pos + copied; |
ea51d132 | 3072 | if (copied && new_i_size > EXT4_I(inode)->i_disksize) { |
9c3569b5 TM |
3073 | if (ext4_has_inline_data(inode) || |
3074 | ext4_da_should_update_i_disksize(page, end)) { | |
ee124d27 | 3075 | ext4_update_i_disksize(inode, new_i_size); |
cf17fea6 AK |
3076 | /* We need to mark inode dirty even if |
3077 | * new_i_size is less that inode->i_size | |
3078 | * bu greater than i_disksize.(hint delalloc) | |
3079 | */ | |
3080 | ext4_mark_inode_dirty(handle, inode); | |
64769240 | 3081 | } |
632eaeab | 3082 | } |
9c3569b5 TM |
3083 | |
3084 | if (write_mode != CONVERT_INLINE_DATA && | |
3085 | ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) && | |
3086 | ext4_has_inline_data(inode)) | |
3087 | ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied, | |
3088 | page); | |
3089 | else | |
3090 | ret2 = generic_write_end(file, mapping, pos, len, copied, | |
64769240 | 3091 | page, fsdata); |
9c3569b5 | 3092 | |
64769240 AT |
3093 | copied = ret2; |
3094 | if (ret2 < 0) | |
3095 | ret = ret2; | |
3096 | ret2 = ext4_journal_stop(handle); | |
3097 | if (!ret) | |
3098 | ret = ret2; | |
3099 | ||
3100 | return ret ? ret : copied; | |
3101 | } | |
3102 | ||
ccd2506b TT |
3103 | /* |
3104 | * Force all delayed allocation blocks to be allocated for a given inode. | |
3105 | */ | |
3106 | int ext4_alloc_da_blocks(struct inode *inode) | |
3107 | { | |
fb40ba0d TT |
3108 | trace_ext4_alloc_da_blocks(inode); |
3109 | ||
71d4f7d0 | 3110 | if (!EXT4_I(inode)->i_reserved_data_blocks) |
ccd2506b TT |
3111 | return 0; |
3112 | ||
3113 | /* | |
3114 | * We do something simple for now. The filemap_flush() will | |
3115 | * also start triggering a write of the data blocks, which is | |
3116 | * not strictly speaking necessary (and for users of | |
3117 | * laptop_mode, not even desirable). However, to do otherwise | |
3118 | * would require replicating code paths in: | |
de9a55b8 | 3119 | * |
20970ba6 | 3120 | * ext4_writepages() -> |
ccd2506b TT |
3121 | * write_cache_pages() ---> (via passed in callback function) |
3122 | * __mpage_da_writepage() --> | |
3123 | * mpage_add_bh_to_extent() | |
3124 | * mpage_da_map_blocks() | |
3125 | * | |
3126 | * The problem is that write_cache_pages(), located in | |
3127 | * mm/page-writeback.c, marks pages clean in preparation for | |
3128 | * doing I/O, which is not desirable if we're not planning on | |
3129 | * doing I/O at all. | |
3130 | * | |
3131 | * We could call write_cache_pages(), and then redirty all of | |
380cf090 | 3132 | * the pages by calling redirty_page_for_writepage() but that |
ccd2506b TT |
3133 | * would be ugly in the extreme. So instead we would need to |
3134 | * replicate parts of the code in the above functions, | |
25985edc | 3135 | * simplifying them because we wouldn't actually intend to |
ccd2506b TT |
3136 | * write out the pages, but rather only collect contiguous |
3137 | * logical block extents, call the multi-block allocator, and | |
3138 | * then update the buffer heads with the block allocations. | |
de9a55b8 | 3139 | * |
ccd2506b TT |
3140 | * For now, though, we'll cheat by calling filemap_flush(), |
3141 | * which will map the blocks, and start the I/O, but not | |
3142 | * actually wait for the I/O to complete. | |
3143 | */ | |
3144 | return filemap_flush(inode->i_mapping); | |
3145 | } | |
64769240 | 3146 | |
ac27a0ec DK |
3147 | /* |
3148 | * bmap() is special. It gets used by applications such as lilo and by | |
3149 | * the swapper to find the on-disk block of a specific piece of data. | |
3150 | * | |
3151 | * Naturally, this is dangerous if the block concerned is still in the | |
617ba13b | 3152 | * journal. If somebody makes a swapfile on an ext4 data-journaling |
ac27a0ec DK |
3153 | * filesystem and enables swap, then they may get a nasty shock when the |
3154 | * data getting swapped to that swapfile suddenly gets overwritten by | |
3155 | * the original zero's written out previously to the journal and | |
3156 | * awaiting writeback in the kernel's buffer cache. | |
3157 | * | |
3158 | * So, if we see any bmap calls here on a modified, data-journaled file, | |
3159 | * take extra steps to flush any blocks which might be in the cache. | |
3160 | */ | |
617ba13b | 3161 | static sector_t ext4_bmap(struct address_space *mapping, sector_t block) |
ac27a0ec DK |
3162 | { |
3163 | struct inode *inode = mapping->host; | |
3164 | journal_t *journal; | |
3165 | int err; | |
3166 | ||
46c7f254 TM |
3167 | /* |
3168 | * We can get here for an inline file via the FIBMAP ioctl | |
3169 | */ | |
3170 | if (ext4_has_inline_data(inode)) | |
3171 | return 0; | |
3172 | ||
64769240 AT |
3173 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) && |
3174 | test_opt(inode->i_sb, DELALLOC)) { | |
3175 | /* | |
3176 | * With delalloc we want to sync the file | |
3177 | * so that we can make sure we allocate | |
3178 | * blocks for file | |
3179 | */ | |
3180 | filemap_write_and_wait(mapping); | |
3181 | } | |
3182 | ||
19f5fb7a TT |
3183 | if (EXT4_JOURNAL(inode) && |
3184 | ext4_test_inode_state(inode, EXT4_STATE_JDATA)) { | |
ac27a0ec DK |
3185 | /* |
3186 | * This is a REALLY heavyweight approach, but the use of | |
3187 | * bmap on dirty files is expected to be extremely rare: | |
3188 | * only if we run lilo or swapon on a freshly made file | |
3189 | * do we expect this to happen. | |
3190 | * | |
3191 | * (bmap requires CAP_SYS_RAWIO so this does not | |
3192 | * represent an unprivileged user DOS attack --- we'd be | |
3193 | * in trouble if mortal users could trigger this path at | |
3194 | * will.) | |
3195 | * | |
617ba13b | 3196 | * NB. EXT4_STATE_JDATA is not set on files other than |
ac27a0ec DK |
3197 | * regular files. If somebody wants to bmap a directory |
3198 | * or symlink and gets confused because the buffer | |
3199 | * hasn't yet been flushed to disk, they deserve | |
3200 | * everything they get. | |
3201 | */ | |
3202 | ||
19f5fb7a | 3203 | ext4_clear_inode_state(inode, EXT4_STATE_JDATA); |
617ba13b | 3204 | journal = EXT4_JOURNAL(inode); |
dab291af MC |
3205 | jbd2_journal_lock_updates(journal); |
3206 | err = jbd2_journal_flush(journal); | |
3207 | jbd2_journal_unlock_updates(journal); | |
ac27a0ec DK |
3208 | |
3209 | if (err) | |
3210 | return 0; | |
3211 | } | |
3212 | ||
ac58e4fb | 3213 | return iomap_bmap(mapping, block, &ext4_iomap_ops); |
ac27a0ec DK |
3214 | } |
3215 | ||
617ba13b | 3216 | static int ext4_readpage(struct file *file, struct page *page) |
ac27a0ec | 3217 | { |
46c7f254 TM |
3218 | int ret = -EAGAIN; |
3219 | struct inode *inode = page->mapping->host; | |
3220 | ||
0562e0ba | 3221 | trace_ext4_readpage(page); |
46c7f254 TM |
3222 | |
3223 | if (ext4_has_inline_data(inode)) | |
3224 | ret = ext4_readpage_inline(inode, page); | |
3225 | ||
3226 | if (ret == -EAGAIN) | |
a07f624b | 3227 | return ext4_mpage_readpages(inode, NULL, page); |
46c7f254 TM |
3228 | |
3229 | return ret; | |
ac27a0ec DK |
3230 | } |
3231 | ||
6311f91f | 3232 | static void ext4_readahead(struct readahead_control *rac) |
ac27a0ec | 3233 | { |
6311f91f | 3234 | struct inode *inode = rac->mapping->host; |
46c7f254 | 3235 | |
6311f91f | 3236 | /* If the file has inline data, no need to do readahead. */ |
46c7f254 | 3237 | if (ext4_has_inline_data(inode)) |
6311f91f | 3238 | return; |
46c7f254 | 3239 | |
a07f624b | 3240 | ext4_mpage_readpages(inode, rac, NULL); |
ac27a0ec DK |
3241 | } |
3242 | ||
d47992f8 LC |
3243 | static void ext4_invalidatepage(struct page *page, unsigned int offset, |
3244 | unsigned int length) | |
ac27a0ec | 3245 | { |
ca99fdd2 | 3246 | trace_ext4_invalidatepage(page, offset, length); |
0562e0ba | 3247 | |
4520fb3c JK |
3248 | /* No journalling happens on data buffers when this function is used */ |
3249 | WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page))); | |
3250 | ||
ca99fdd2 | 3251 | block_invalidatepage(page, offset, length); |
4520fb3c JK |
3252 | } |
3253 | ||
53e87268 | 3254 | static int __ext4_journalled_invalidatepage(struct page *page, |
ca99fdd2 LC |
3255 | unsigned int offset, |
3256 | unsigned int length) | |
4520fb3c JK |
3257 | { |
3258 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); | |
3259 | ||
ca99fdd2 | 3260 | trace_ext4_journalled_invalidatepage(page, offset, length); |
4520fb3c | 3261 | |
ac27a0ec DK |
3262 | /* |
3263 | * If it's a full truncate we just forget about the pending dirtying | |
3264 | */ | |
09cbfeaf | 3265 | if (offset == 0 && length == PAGE_SIZE) |
ac27a0ec DK |
3266 | ClearPageChecked(page); |
3267 | ||
ca99fdd2 | 3268 | return jbd2_journal_invalidatepage(journal, page, offset, length); |
53e87268 JK |
3269 | } |
3270 | ||
3271 | /* Wrapper for aops... */ | |
3272 | static void ext4_journalled_invalidatepage(struct page *page, | |
d47992f8 LC |
3273 | unsigned int offset, |
3274 | unsigned int length) | |
53e87268 | 3275 | { |
ca99fdd2 | 3276 | WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0); |
ac27a0ec DK |
3277 | } |
3278 | ||
617ba13b | 3279 | static int ext4_releasepage(struct page *page, gfp_t wait) |
ac27a0ec | 3280 | { |
617ba13b | 3281 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); |
ac27a0ec | 3282 | |
0562e0ba JZ |
3283 | trace_ext4_releasepage(page); |
3284 | ||
e1c36595 JK |
3285 | /* Page has dirty journalled data -> cannot release */ |
3286 | if (PageChecked(page)) | |
ac27a0ec | 3287 | return 0; |
0390131b FM |
3288 | if (journal) |
3289 | return jbd2_journal_try_to_free_buffers(journal, page, wait); | |
3290 | else | |
3291 | return try_to_free_buffers(page); | |
ac27a0ec DK |
3292 | } |
3293 | ||
b8a6176c JK |
3294 | static bool ext4_inode_datasync_dirty(struct inode *inode) |
3295 | { | |
3296 | journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; | |
3297 | ||
3298 | if (journal) | |
3299 | return !jbd2_transaction_committed(journal, | |
3300 | EXT4_I(inode)->i_datasync_tid); | |
3301 | /* Any metadata buffers to write? */ | |
3302 | if (!list_empty(&inode->i_mapping->private_list)) | |
3303 | return true; | |
3304 | return inode->i_state & I_DIRTY_DATASYNC; | |
3305 | } | |
3306 | ||
c8fdfe29 MB |
3307 | static void ext4_set_iomap(struct inode *inode, struct iomap *iomap, |
3308 | struct ext4_map_blocks *map, loff_t offset, | |
3309 | loff_t length) | |
364443cb | 3310 | { |
c8fdfe29 | 3311 | u8 blkbits = inode->i_blkbits; |
364443cb | 3312 | |
c8fdfe29 MB |
3313 | /* |
3314 | * Writes that span EOF might trigger an I/O size update on completion, | |
3315 | * so consider them to be dirty for the purpose of O_DSYNC, even if | |
3316 | * there is no other metadata changes being made or are pending. | |
3317 | */ | |
364443cb | 3318 | iomap->flags = 0; |
c8fdfe29 MB |
3319 | if (ext4_inode_datasync_dirty(inode) || |
3320 | offset + length > i_size_read(inode)) | |
b8a6176c | 3321 | iomap->flags |= IOMAP_F_DIRTY; |
c8fdfe29 MB |
3322 | |
3323 | if (map->m_flags & EXT4_MAP_NEW) | |
3324 | iomap->flags |= IOMAP_F_NEW; | |
3325 | ||
5e405595 | 3326 | iomap->bdev = inode->i_sb->s_bdev; |
c8fdfe29 MB |
3327 | iomap->dax_dev = EXT4_SB(inode->i_sb)->s_daxdev; |
3328 | iomap->offset = (u64) map->m_lblk << blkbits; | |
3329 | iomap->length = (u64) map->m_len << blkbits; | |
364443cb | 3330 | |
6386722a RH |
3331 | if ((map->m_flags & EXT4_MAP_MAPPED) && |
3332 | !ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
3333 | iomap->flags |= IOMAP_F_MERGED; | |
3334 | ||
c8fdfe29 MB |
3335 | /* |
3336 | * Flags passed to ext4_map_blocks() for direct I/O writes can result | |
3337 | * in m_flags having both EXT4_MAP_MAPPED and EXT4_MAP_UNWRITTEN bits | |
3338 | * set. In order for any allocated unwritten extents to be converted | |
3339 | * into written extents correctly within the ->end_io() handler, we | |
3340 | * need to ensure that the iomap->type is set appropriately. Hence, the | |
3341 | * reason why we need to check whether the EXT4_MAP_UNWRITTEN bit has | |
3342 | * been set first. | |
3343 | */ | |
3344 | if (map->m_flags & EXT4_MAP_UNWRITTEN) { | |
3345 | iomap->type = IOMAP_UNWRITTEN; | |
3346 | iomap->addr = (u64) map->m_pblk << blkbits; | |
3347 | } else if (map->m_flags & EXT4_MAP_MAPPED) { | |
3348 | iomap->type = IOMAP_MAPPED; | |
3349 | iomap->addr = (u64) map->m_pblk << blkbits; | |
364443cb | 3350 | } else { |
c8fdfe29 MB |
3351 | iomap->type = IOMAP_HOLE; |
3352 | iomap->addr = IOMAP_NULL_ADDR; | |
364443cb | 3353 | } |
364443cb JK |
3354 | } |
3355 | ||
f063db5e MB |
3356 | static int ext4_iomap_alloc(struct inode *inode, struct ext4_map_blocks *map, |
3357 | unsigned int flags) | |
776722e8 | 3358 | { |
776722e8 | 3359 | handle_t *handle; |
378f32ba MB |
3360 | u8 blkbits = inode->i_blkbits; |
3361 | int ret, dio_credits, m_flags = 0, retries = 0; | |
776722e8 | 3362 | |
776722e8 | 3363 | /* |
f063db5e MB |
3364 | * Trim the mapping request to the maximum value that we can map at |
3365 | * once for direct I/O. | |
776722e8 | 3366 | */ |
f063db5e MB |
3367 | if (map->m_len > DIO_MAX_BLOCKS) |
3368 | map->m_len = DIO_MAX_BLOCKS; | |
3369 | dio_credits = ext4_chunk_trans_blocks(inode, map->m_len); | |
776722e8 | 3370 | |
f063db5e | 3371 | retry: |
776722e8 | 3372 | /* |
f063db5e MB |
3373 | * Either we allocate blocks and then don't get an unwritten extent, so |
3374 | * in that case we have reserved enough credits. Or, the blocks are | |
3375 | * already allocated and unwritten. In that case, the extent conversion | |
3376 | * fits into the credits as well. | |
776722e8 | 3377 | */ |
f063db5e MB |
3378 | handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits); |
3379 | if (IS_ERR(handle)) | |
3380 | return PTR_ERR(handle); | |
4c0425ff | 3381 | |
378f32ba MB |
3382 | /* |
3383 | * DAX and direct I/O are the only two operations that are currently | |
3384 | * supported with IOMAP_WRITE. | |
3385 | */ | |
3386 | WARN_ON(!IS_DAX(inode) && !(flags & IOMAP_DIRECT)); | |
3387 | if (IS_DAX(inode)) | |
3388 | m_flags = EXT4_GET_BLOCKS_CREATE_ZERO; | |
3389 | /* | |
3390 | * We use i_size instead of i_disksize here because delalloc writeback | |
3391 | * can complete at any point during the I/O and subsequently push the | |
3392 | * i_disksize out to i_size. This could be beyond where direct I/O is | |
3393 | * happening and thus expose allocated blocks to direct I/O reads. | |
3394 | */ | |
3395 | else if ((map->m_lblk * (1 << blkbits)) >= i_size_read(inode)) | |
3396 | m_flags = EXT4_GET_BLOCKS_CREATE; | |
3397 | else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
3398 | m_flags = EXT4_GET_BLOCKS_IO_CREATE_EXT; | |
4b70df18 | 3399 | |
378f32ba | 3400 | ret = ext4_map_blocks(handle, inode, map, m_flags); |
8d5d02e6 | 3401 | |
74c66bcb | 3402 | /* |
378f32ba MB |
3403 | * We cannot fill holes in indirect tree based inodes as that could |
3404 | * expose stale data in the case of a crash. Use the magic error code | |
3405 | * to fallback to buffered I/O. | |
74c66bcb | 3406 | */ |
378f32ba MB |
3407 | if (!m_flags && !ret) |
3408 | ret = -ENOTBLK; | |
187372a3 | 3409 | |
f063db5e MB |
3410 | ext4_journal_stop(handle); |
3411 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) | |
3412 | goto retry; | |
3413 | ||
3414 | return ret; | |
4c0425ff | 3415 | } |
c7064ef1 | 3416 | |
f063db5e | 3417 | |
364443cb | 3418 | static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length, |
c039b997 | 3419 | unsigned flags, struct iomap *iomap, struct iomap *srcmap) |
4c0425ff | 3420 | { |
364443cb | 3421 | int ret; |
09edf4d3 MB |
3422 | struct ext4_map_blocks map; |
3423 | u8 blkbits = inode->i_blkbits; | |
729f52c6 | 3424 | |
bcd8e91f TT |
3425 | if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK) |
3426 | return -EINVAL; | |
4bd809db | 3427 | |
09edf4d3 MB |
3428 | if (WARN_ON_ONCE(ext4_has_inline_data(inode))) |
3429 | return -ERANGE; | |
4bd809db | 3430 | |
e8340395 | 3431 | /* |
09edf4d3 | 3432 | * Calculate the first and last logical blocks respectively. |
e8340395 | 3433 | */ |
09edf4d3 MB |
3434 | map.m_lblk = offset >> blkbits; |
3435 | map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits, | |
3436 | EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1; | |
e8340395 | 3437 | |
09edf4d3 | 3438 | if (flags & IOMAP_WRITE) |
f063db5e | 3439 | ret = ext4_iomap_alloc(inode, &map, flags); |
09edf4d3 | 3440 | else |
545052e9 | 3441 | ret = ext4_map_blocks(NULL, inode, &map, 0); |
4bd809db | 3442 | |
f063db5e MB |
3443 | if (ret < 0) |
3444 | return ret; | |
e8340395 | 3445 | |
c8fdfe29 | 3446 | ext4_set_iomap(inode, iomap, &map, offset, length); |
4bd809db | 3447 | |
364443cb JK |
3448 | return 0; |
3449 | } | |
8d5d02e6 | 3450 | |
8cd115bd JK |
3451 | static int ext4_iomap_overwrite_begin(struct inode *inode, loff_t offset, |
3452 | loff_t length, unsigned flags, struct iomap *iomap, | |
3453 | struct iomap *srcmap) | |
3454 | { | |
3455 | int ret; | |
3456 | ||
3457 | /* | |
3458 | * Even for writes we don't need to allocate blocks, so just pretend | |
3459 | * we are reading to save overhead of starting a transaction. | |
3460 | */ | |
3461 | flags &= ~IOMAP_WRITE; | |
3462 | ret = ext4_iomap_begin(inode, offset, length, flags, iomap, srcmap); | |
3463 | WARN_ON_ONCE(iomap->type != IOMAP_MAPPED); | |
3464 | return ret; | |
3465 | } | |
3466 | ||
776722e8 JK |
3467 | static int ext4_iomap_end(struct inode *inode, loff_t offset, loff_t length, |
3468 | ssize_t written, unsigned flags, struct iomap *iomap) | |
3469 | { | |
69c499d1 | 3470 | /* |
378f32ba MB |
3471 | * Check to see whether an error occurred while writing out the data to |
3472 | * the allocated blocks. If so, return the magic error code so that we | |
3473 | * fallback to buffered I/O and attempt to complete the remainder of | |
3474 | * the I/O. Any blocks that may have been allocated in preparation for | |
3475 | * the direct I/O will be reused during buffered I/O. | |
69c499d1 | 3476 | */ |
378f32ba MB |
3477 | if (flags & (IOMAP_WRITE | IOMAP_DIRECT) && written == 0) |
3478 | return -ENOTBLK; | |
69c499d1 | 3479 | |
569342dc | 3480 | return 0; |
776722e8 | 3481 | } |
4bd809db | 3482 | |
8ff6daa1 | 3483 | const struct iomap_ops ext4_iomap_ops = { |
364443cb | 3484 | .iomap_begin = ext4_iomap_begin, |
776722e8 | 3485 | .iomap_end = ext4_iomap_end, |
364443cb | 3486 | }; |
8d5d02e6 | 3487 | |
8cd115bd JK |
3488 | const struct iomap_ops ext4_iomap_overwrite_ops = { |
3489 | .iomap_begin = ext4_iomap_overwrite_begin, | |
3490 | .iomap_end = ext4_iomap_end, | |
3491 | }; | |
3492 | ||
09edf4d3 MB |
3493 | static bool ext4_iomap_is_delalloc(struct inode *inode, |
3494 | struct ext4_map_blocks *map) | |
3495 | { | |
3496 | struct extent_status es; | |
3497 | ext4_lblk_t offset = 0, end = map->m_lblk + map->m_len - 1; | |
914f82a3 | 3498 | |
09edf4d3 MB |
3499 | ext4_es_find_extent_range(inode, &ext4_es_is_delayed, |
3500 | map->m_lblk, end, &es); | |
914f82a3 | 3501 | |
09edf4d3 MB |
3502 | if (!es.es_len || es.es_lblk > end) |
3503 | return false; | |
914f82a3 | 3504 | |
09edf4d3 MB |
3505 | if (es.es_lblk > map->m_lblk) { |
3506 | map->m_len = es.es_lblk - map->m_lblk; | |
3507 | return false; | |
914f82a3 | 3508 | } |
914f82a3 | 3509 | |
09edf4d3 MB |
3510 | offset = map->m_lblk - es.es_lblk; |
3511 | map->m_len = es.es_len - offset; | |
914f82a3 | 3512 | |
09edf4d3 | 3513 | return true; |
4c0425ff MC |
3514 | } |
3515 | ||
09edf4d3 MB |
3516 | static int ext4_iomap_begin_report(struct inode *inode, loff_t offset, |
3517 | loff_t length, unsigned int flags, | |
3518 | struct iomap *iomap, struct iomap *srcmap) | |
4c0425ff | 3519 | { |
09edf4d3 MB |
3520 | int ret; |
3521 | bool delalloc = false; | |
3522 | struct ext4_map_blocks map; | |
3523 | u8 blkbits = inode->i_blkbits; | |
4c0425ff | 3524 | |
09edf4d3 MB |
3525 | if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK) |
3526 | return -EINVAL; | |
3527 | ||
3528 | if (ext4_has_inline_data(inode)) { | |
3529 | ret = ext4_inline_data_iomap(inode, iomap); | |
3530 | if (ret != -EAGAIN) { | |
3531 | if (ret == 0 && offset >= iomap->length) | |
3532 | ret = -ENOENT; | |
3533 | return ret; | |
3534 | } | |
3535 | } | |
2058f83a | 3536 | |
84ebd795 | 3537 | /* |
09edf4d3 | 3538 | * Calculate the first and last logical block respectively. |
84ebd795 | 3539 | */ |
09edf4d3 MB |
3540 | map.m_lblk = offset >> blkbits; |
3541 | map.m_len = min_t(loff_t, (offset + length - 1) >> blkbits, | |
3542 | EXT4_MAX_LOGICAL_BLOCK) - map.m_lblk + 1; | |
84ebd795 | 3543 | |
b2c57642 RH |
3544 | /* |
3545 | * Fiemap callers may call for offset beyond s_bitmap_maxbytes. | |
3546 | * So handle it here itself instead of querying ext4_map_blocks(). | |
3547 | * Since ext4_map_blocks() will warn about it and will return | |
3548 | * -EIO error. | |
3549 | */ | |
3550 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) { | |
3551 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
3552 | ||
3553 | if (offset >= sbi->s_bitmap_maxbytes) { | |
3554 | map.m_flags = 0; | |
3555 | goto set_iomap; | |
3556 | } | |
3557 | } | |
3558 | ||
09edf4d3 MB |
3559 | ret = ext4_map_blocks(NULL, inode, &map, 0); |
3560 | if (ret < 0) | |
3561 | return ret; | |
3562 | if (ret == 0) | |
3563 | delalloc = ext4_iomap_is_delalloc(inode, &map); | |
46c7f254 | 3564 | |
b2c57642 | 3565 | set_iomap: |
09edf4d3 MB |
3566 | ext4_set_iomap(inode, iomap, &map, offset, length); |
3567 | if (delalloc && iomap->type == IOMAP_HOLE) | |
3568 | iomap->type = IOMAP_DELALLOC; | |
3569 | ||
3570 | return 0; | |
4c0425ff MC |
3571 | } |
3572 | ||
09edf4d3 MB |
3573 | const struct iomap_ops ext4_iomap_report_ops = { |
3574 | .iomap_begin = ext4_iomap_begin_report, | |
3575 | }; | |
3576 | ||
ac27a0ec | 3577 | /* |
617ba13b | 3578 | * Pages can be marked dirty completely asynchronously from ext4's journalling |
ac27a0ec DK |
3579 | * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do |
3580 | * much here because ->set_page_dirty is called under VFS locks. The page is | |
3581 | * not necessarily locked. | |
3582 | * | |
3583 | * We cannot just dirty the page and leave attached buffers clean, because the | |
3584 | * buffers' dirty state is "definitive". We cannot just set the buffers dirty | |
3585 | * or jbddirty because all the journalling code will explode. | |
3586 | * | |
3587 | * So what we do is to mark the page "pending dirty" and next time writepage | |
3588 | * is called, propagate that into the buffers appropriately. | |
3589 | */ | |
617ba13b | 3590 | static int ext4_journalled_set_page_dirty(struct page *page) |
ac27a0ec DK |
3591 | { |
3592 | SetPageChecked(page); | |
3593 | return __set_page_dirty_nobuffers(page); | |
3594 | } | |
3595 | ||
6dcc693b JK |
3596 | static int ext4_set_page_dirty(struct page *page) |
3597 | { | |
3598 | WARN_ON_ONCE(!PageLocked(page) && !PageDirty(page)); | |
3599 | WARN_ON_ONCE(!page_has_buffers(page)); | |
3600 | return __set_page_dirty_buffers(page); | |
3601 | } | |
3602 | ||
74d553aa | 3603 | static const struct address_space_operations ext4_aops = { |
8ab22b9a | 3604 | .readpage = ext4_readpage, |
6311f91f | 3605 | .readahead = ext4_readahead, |
43ce1d23 | 3606 | .writepage = ext4_writepage, |
20970ba6 | 3607 | .writepages = ext4_writepages, |
8ab22b9a | 3608 | .write_begin = ext4_write_begin, |
74d553aa | 3609 | .write_end = ext4_write_end, |
6dcc693b | 3610 | .set_page_dirty = ext4_set_page_dirty, |
8ab22b9a HH |
3611 | .bmap = ext4_bmap, |
3612 | .invalidatepage = ext4_invalidatepage, | |
3613 | .releasepage = ext4_releasepage, | |
378f32ba | 3614 | .direct_IO = noop_direct_IO, |
8ab22b9a HH |
3615 | .migratepage = buffer_migrate_page, |
3616 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 3617 | .error_remove_page = generic_error_remove_page, |
ac27a0ec DK |
3618 | }; |
3619 | ||
617ba13b | 3620 | static const struct address_space_operations ext4_journalled_aops = { |
8ab22b9a | 3621 | .readpage = ext4_readpage, |
6311f91f | 3622 | .readahead = ext4_readahead, |
43ce1d23 | 3623 | .writepage = ext4_writepage, |
20970ba6 | 3624 | .writepages = ext4_writepages, |
8ab22b9a HH |
3625 | .write_begin = ext4_write_begin, |
3626 | .write_end = ext4_journalled_write_end, | |
3627 | .set_page_dirty = ext4_journalled_set_page_dirty, | |
3628 | .bmap = ext4_bmap, | |
4520fb3c | 3629 | .invalidatepage = ext4_journalled_invalidatepage, |
8ab22b9a | 3630 | .releasepage = ext4_releasepage, |
378f32ba | 3631 | .direct_IO = noop_direct_IO, |
8ab22b9a | 3632 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 3633 | .error_remove_page = generic_error_remove_page, |
ac27a0ec DK |
3634 | }; |
3635 | ||
64769240 | 3636 | static const struct address_space_operations ext4_da_aops = { |
8ab22b9a | 3637 | .readpage = ext4_readpage, |
6311f91f | 3638 | .readahead = ext4_readahead, |
43ce1d23 | 3639 | .writepage = ext4_writepage, |
20970ba6 | 3640 | .writepages = ext4_writepages, |
8ab22b9a HH |
3641 | .write_begin = ext4_da_write_begin, |
3642 | .write_end = ext4_da_write_end, | |
6dcc693b | 3643 | .set_page_dirty = ext4_set_page_dirty, |
8ab22b9a | 3644 | .bmap = ext4_bmap, |
8fcc3a58 | 3645 | .invalidatepage = ext4_invalidatepage, |
8ab22b9a | 3646 | .releasepage = ext4_releasepage, |
378f32ba | 3647 | .direct_IO = noop_direct_IO, |
8ab22b9a HH |
3648 | .migratepage = buffer_migrate_page, |
3649 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 3650 | .error_remove_page = generic_error_remove_page, |
64769240 AT |
3651 | }; |
3652 | ||
5f0663bb DW |
3653 | static const struct address_space_operations ext4_dax_aops = { |
3654 | .writepages = ext4_dax_writepages, | |
3655 | .direct_IO = noop_direct_IO, | |
3656 | .set_page_dirty = noop_set_page_dirty, | |
94dbb631 | 3657 | .bmap = ext4_bmap, |
5f0663bb DW |
3658 | .invalidatepage = noop_invalidatepage, |
3659 | }; | |
3660 | ||
617ba13b | 3661 | void ext4_set_aops(struct inode *inode) |
ac27a0ec | 3662 | { |
3d2b1582 LC |
3663 | switch (ext4_inode_journal_mode(inode)) { |
3664 | case EXT4_INODE_ORDERED_DATA_MODE: | |
3d2b1582 | 3665 | case EXT4_INODE_WRITEBACK_DATA_MODE: |
3d2b1582 LC |
3666 | break; |
3667 | case EXT4_INODE_JOURNAL_DATA_MODE: | |
617ba13b | 3668 | inode->i_mapping->a_ops = &ext4_journalled_aops; |
74d553aa | 3669 | return; |
3d2b1582 LC |
3670 | default: |
3671 | BUG(); | |
3672 | } | |
5f0663bb DW |
3673 | if (IS_DAX(inode)) |
3674 | inode->i_mapping->a_ops = &ext4_dax_aops; | |
3675 | else if (test_opt(inode->i_sb, DELALLOC)) | |
74d553aa TT |
3676 | inode->i_mapping->a_ops = &ext4_da_aops; |
3677 | else | |
3678 | inode->i_mapping->a_ops = &ext4_aops; | |
ac27a0ec DK |
3679 | } |
3680 | ||
923ae0ff | 3681 | static int __ext4_block_zero_page_range(handle_t *handle, |
d863dc36 LC |
3682 | struct address_space *mapping, loff_t from, loff_t length) |
3683 | { | |
09cbfeaf KS |
3684 | ext4_fsblk_t index = from >> PAGE_SHIFT; |
3685 | unsigned offset = from & (PAGE_SIZE-1); | |
923ae0ff | 3686 | unsigned blocksize, pos; |
d863dc36 LC |
3687 | ext4_lblk_t iblock; |
3688 | struct inode *inode = mapping->host; | |
3689 | struct buffer_head *bh; | |
3690 | struct page *page; | |
3691 | int err = 0; | |
3692 | ||
09cbfeaf | 3693 | page = find_or_create_page(mapping, from >> PAGE_SHIFT, |
c62d2555 | 3694 | mapping_gfp_constraint(mapping, ~__GFP_FS)); |
d863dc36 LC |
3695 | if (!page) |
3696 | return -ENOMEM; | |
3697 | ||
3698 | blocksize = inode->i_sb->s_blocksize; | |
d863dc36 | 3699 | |
09cbfeaf | 3700 | iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits); |
d863dc36 LC |
3701 | |
3702 | if (!page_has_buffers(page)) | |
3703 | create_empty_buffers(page, blocksize, 0); | |
3704 | ||
3705 | /* Find the buffer that contains "offset" */ | |
3706 | bh = page_buffers(page); | |
3707 | pos = blocksize; | |
3708 | while (offset >= pos) { | |
3709 | bh = bh->b_this_page; | |
3710 | iblock++; | |
3711 | pos += blocksize; | |
3712 | } | |
d863dc36 LC |
3713 | if (buffer_freed(bh)) { |
3714 | BUFFER_TRACE(bh, "freed: skip"); | |
3715 | goto unlock; | |
3716 | } | |
d863dc36 LC |
3717 | if (!buffer_mapped(bh)) { |
3718 | BUFFER_TRACE(bh, "unmapped"); | |
3719 | ext4_get_block(inode, iblock, bh, 0); | |
3720 | /* unmapped? It's a hole - nothing to do */ | |
3721 | if (!buffer_mapped(bh)) { | |
3722 | BUFFER_TRACE(bh, "still unmapped"); | |
3723 | goto unlock; | |
3724 | } | |
3725 | } | |
3726 | ||
3727 | /* Ok, it's mapped. Make sure it's up-to-date */ | |
3728 | if (PageUptodate(page)) | |
3729 | set_buffer_uptodate(bh); | |
3730 | ||
3731 | if (!buffer_uptodate(bh)) { | |
3732 | err = -EIO; | |
dfec8a14 | 3733 | ll_rw_block(REQ_OP_READ, 0, 1, &bh); |
d863dc36 LC |
3734 | wait_on_buffer(bh); |
3735 | /* Uhhuh. Read error. Complain and punt. */ | |
3736 | if (!buffer_uptodate(bh)) | |
3737 | goto unlock; | |
592ddec7 | 3738 | if (S_ISREG(inode->i_mode) && IS_ENCRYPTED(inode)) { |
c9c7429c | 3739 | /* We expect the key to be set. */ |
a7550b30 | 3740 | BUG_ON(!fscrypt_has_encryption_key(inode)); |
834f1565 EB |
3741 | err = fscrypt_decrypt_pagecache_blocks(page, blocksize, |
3742 | bh_offset(bh)); | |
3743 | if (err) { | |
3744 | clear_buffer_uptodate(bh); | |
3745 | goto unlock; | |
3746 | } | |
c9c7429c | 3747 | } |
d863dc36 | 3748 | } |
d863dc36 LC |
3749 | if (ext4_should_journal_data(inode)) { |
3750 | BUFFER_TRACE(bh, "get write access"); | |
3751 | err = ext4_journal_get_write_access(handle, bh); | |
3752 | if (err) | |
3753 | goto unlock; | |
3754 | } | |
d863dc36 | 3755 | zero_user(page, offset, length); |
d863dc36 LC |
3756 | BUFFER_TRACE(bh, "zeroed end of block"); |
3757 | ||
d863dc36 LC |
3758 | if (ext4_should_journal_data(inode)) { |
3759 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
0713ed0c | 3760 | } else { |
353eefd3 | 3761 | err = 0; |
d863dc36 | 3762 | mark_buffer_dirty(bh); |
3957ef53 | 3763 | if (ext4_should_order_data(inode)) |
73131fbb RZ |
3764 | err = ext4_jbd2_inode_add_write(handle, inode, from, |
3765 | length); | |
0713ed0c | 3766 | } |
d863dc36 LC |
3767 | |
3768 | unlock: | |
3769 | unlock_page(page); | |
09cbfeaf | 3770 | put_page(page); |
d863dc36 LC |
3771 | return err; |
3772 | } | |
3773 | ||
923ae0ff RZ |
3774 | /* |
3775 | * ext4_block_zero_page_range() zeros out a mapping of length 'length' | |
3776 | * starting from file offset 'from'. The range to be zero'd must | |
3777 | * be contained with in one block. If the specified range exceeds | |
3778 | * the end of the block it will be shortened to end of the block | |
3779 | * that cooresponds to 'from' | |
3780 | */ | |
3781 | static int ext4_block_zero_page_range(handle_t *handle, | |
3782 | struct address_space *mapping, loff_t from, loff_t length) | |
3783 | { | |
3784 | struct inode *inode = mapping->host; | |
09cbfeaf | 3785 | unsigned offset = from & (PAGE_SIZE-1); |
923ae0ff RZ |
3786 | unsigned blocksize = inode->i_sb->s_blocksize; |
3787 | unsigned max = blocksize - (offset & (blocksize - 1)); | |
3788 | ||
3789 | /* | |
3790 | * correct length if it does not fall between | |
3791 | * 'from' and the end of the block | |
3792 | */ | |
3793 | if (length > max || length < 0) | |
3794 | length = max; | |
3795 | ||
47e69351 JK |
3796 | if (IS_DAX(inode)) { |
3797 | return iomap_zero_range(inode, from, length, NULL, | |
3798 | &ext4_iomap_ops); | |
3799 | } | |
923ae0ff RZ |
3800 | return __ext4_block_zero_page_range(handle, mapping, from, length); |
3801 | } | |
3802 | ||
94350ab5 MW |
3803 | /* |
3804 | * ext4_block_truncate_page() zeroes out a mapping from file offset `from' | |
3805 | * up to the end of the block which corresponds to `from'. | |
3806 | * This required during truncate. We need to physically zero the tail end | |
3807 | * of that block so it doesn't yield old data if the file is later grown. | |
3808 | */ | |
c197855e | 3809 | static int ext4_block_truncate_page(handle_t *handle, |
94350ab5 MW |
3810 | struct address_space *mapping, loff_t from) |
3811 | { | |
09cbfeaf | 3812 | unsigned offset = from & (PAGE_SIZE-1); |
94350ab5 MW |
3813 | unsigned length; |
3814 | unsigned blocksize; | |
3815 | struct inode *inode = mapping->host; | |
3816 | ||
0d06863f | 3817 | /* If we are processing an encrypted inode during orphan list handling */ |
592ddec7 | 3818 | if (IS_ENCRYPTED(inode) && !fscrypt_has_encryption_key(inode)) |
0d06863f TT |
3819 | return 0; |
3820 | ||
94350ab5 MW |
3821 | blocksize = inode->i_sb->s_blocksize; |
3822 | length = blocksize - (offset & (blocksize - 1)); | |
3823 | ||
3824 | return ext4_block_zero_page_range(handle, mapping, from, length); | |
3825 | } | |
3826 | ||
a87dd18c LC |
3827 | int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode, |
3828 | loff_t lstart, loff_t length) | |
3829 | { | |
3830 | struct super_block *sb = inode->i_sb; | |
3831 | struct address_space *mapping = inode->i_mapping; | |
e1be3a92 | 3832 | unsigned partial_start, partial_end; |
a87dd18c LC |
3833 | ext4_fsblk_t start, end; |
3834 | loff_t byte_end = (lstart + length - 1); | |
3835 | int err = 0; | |
3836 | ||
e1be3a92 LC |
3837 | partial_start = lstart & (sb->s_blocksize - 1); |
3838 | partial_end = byte_end & (sb->s_blocksize - 1); | |
3839 | ||
a87dd18c LC |
3840 | start = lstart >> sb->s_blocksize_bits; |
3841 | end = byte_end >> sb->s_blocksize_bits; | |
3842 | ||
3843 | /* Handle partial zero within the single block */ | |
e1be3a92 LC |
3844 | if (start == end && |
3845 | (partial_start || (partial_end != sb->s_blocksize - 1))) { | |
a87dd18c LC |
3846 | err = ext4_block_zero_page_range(handle, mapping, |
3847 | lstart, length); | |
3848 | return err; | |
3849 | } | |
3850 | /* Handle partial zero out on the start of the range */ | |
e1be3a92 | 3851 | if (partial_start) { |
a87dd18c LC |
3852 | err = ext4_block_zero_page_range(handle, mapping, |
3853 | lstart, sb->s_blocksize); | |
3854 | if (err) | |
3855 | return err; | |
3856 | } | |
3857 | /* Handle partial zero out on the end of the range */ | |
e1be3a92 | 3858 | if (partial_end != sb->s_blocksize - 1) |
a87dd18c | 3859 | err = ext4_block_zero_page_range(handle, mapping, |
e1be3a92 LC |
3860 | byte_end - partial_end, |
3861 | partial_end + 1); | |
a87dd18c LC |
3862 | return err; |
3863 | } | |
3864 | ||
91ef4caf DG |
3865 | int ext4_can_truncate(struct inode *inode) |
3866 | { | |
91ef4caf DG |
3867 | if (S_ISREG(inode->i_mode)) |
3868 | return 1; | |
3869 | if (S_ISDIR(inode->i_mode)) | |
3870 | return 1; | |
3871 | if (S_ISLNK(inode->i_mode)) | |
3872 | return !ext4_inode_is_fast_symlink(inode); | |
3873 | return 0; | |
3874 | } | |
3875 | ||
01127848 JK |
3876 | /* |
3877 | * We have to make sure i_disksize gets properly updated before we truncate | |
3878 | * page cache due to hole punching or zero range. Otherwise i_disksize update | |
3879 | * can get lost as it may have been postponed to submission of writeback but | |
3880 | * that will never happen after we truncate page cache. | |
3881 | */ | |
3882 | int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset, | |
3883 | loff_t len) | |
3884 | { | |
3885 | handle_t *handle; | |
3886 | loff_t size = i_size_read(inode); | |
3887 | ||
5955102c | 3888 | WARN_ON(!inode_is_locked(inode)); |
01127848 JK |
3889 | if (offset > size || offset + len < size) |
3890 | return 0; | |
3891 | ||
3892 | if (EXT4_I(inode)->i_disksize >= size) | |
3893 | return 0; | |
3894 | ||
3895 | handle = ext4_journal_start(inode, EXT4_HT_MISC, 1); | |
3896 | if (IS_ERR(handle)) | |
3897 | return PTR_ERR(handle); | |
3898 | ext4_update_i_disksize(inode, size); | |
3899 | ext4_mark_inode_dirty(handle, inode); | |
3900 | ext4_journal_stop(handle); | |
3901 | ||
3902 | return 0; | |
3903 | } | |
3904 | ||
b1f38217 | 3905 | static void ext4_wait_dax_page(struct ext4_inode_info *ei) |
430657b6 | 3906 | { |
430657b6 RZ |
3907 | up_write(&ei->i_mmap_sem); |
3908 | schedule(); | |
3909 | down_write(&ei->i_mmap_sem); | |
3910 | } | |
3911 | ||
3912 | int ext4_break_layouts(struct inode *inode) | |
3913 | { | |
3914 | struct ext4_inode_info *ei = EXT4_I(inode); | |
3915 | struct page *page; | |
430657b6 RZ |
3916 | int error; |
3917 | ||
3918 | if (WARN_ON_ONCE(!rwsem_is_locked(&ei->i_mmap_sem))) | |
3919 | return -EINVAL; | |
3920 | ||
3921 | do { | |
430657b6 RZ |
3922 | page = dax_layout_busy_page(inode->i_mapping); |
3923 | if (!page) | |
3924 | return 0; | |
3925 | ||
3926 | error = ___wait_var_event(&page->_refcount, | |
3927 | atomic_read(&page->_refcount) == 1, | |
3928 | TASK_INTERRUPTIBLE, 0, 0, | |
b1f38217 RZ |
3929 | ext4_wait_dax_page(ei)); |
3930 | } while (error == 0); | |
430657b6 RZ |
3931 | |
3932 | return error; | |
3933 | } | |
3934 | ||
a4bb6b64 | 3935 | /* |
cca32b7e | 3936 | * ext4_punch_hole: punches a hole in a file by releasing the blocks |
a4bb6b64 AH |
3937 | * associated with the given offset and length |
3938 | * | |
3939 | * @inode: File inode | |
3940 | * @offset: The offset where the hole will begin | |
3941 | * @len: The length of the hole | |
3942 | * | |
4907cb7b | 3943 | * Returns: 0 on success or negative on failure |
a4bb6b64 AH |
3944 | */ |
3945 | ||
aeb2817a | 3946 | int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length) |
a4bb6b64 | 3947 | { |
26a4c0c6 TT |
3948 | struct super_block *sb = inode->i_sb; |
3949 | ext4_lblk_t first_block, stop_block; | |
3950 | struct address_space *mapping = inode->i_mapping; | |
a87dd18c | 3951 | loff_t first_block_offset, last_block_offset; |
26a4c0c6 TT |
3952 | handle_t *handle; |
3953 | unsigned int credits; | |
3954 | int ret = 0; | |
3955 | ||
b8a86845 | 3956 | trace_ext4_punch_hole(inode, offset, length, 0); |
aaddea81 | 3957 | |
c1e8220b TT |
3958 | ext4_clear_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA); |
3959 | if (ext4_has_inline_data(inode)) { | |
3960 | down_write(&EXT4_I(inode)->i_mmap_sem); | |
3961 | ret = ext4_convert_inline_data(inode); | |
3962 | up_write(&EXT4_I(inode)->i_mmap_sem); | |
3963 | if (ret) | |
3964 | return ret; | |
3965 | } | |
3966 | ||
26a4c0c6 TT |
3967 | /* |
3968 | * Write out all dirty pages to avoid race conditions | |
3969 | * Then release them. | |
3970 | */ | |
cca32b7e | 3971 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { |
26a4c0c6 TT |
3972 | ret = filemap_write_and_wait_range(mapping, offset, |
3973 | offset + length - 1); | |
3974 | if (ret) | |
3975 | return ret; | |
3976 | } | |
3977 | ||
5955102c | 3978 | inode_lock(inode); |
9ef06cec | 3979 | |
26a4c0c6 TT |
3980 | /* No need to punch hole beyond i_size */ |
3981 | if (offset >= inode->i_size) | |
3982 | goto out_mutex; | |
3983 | ||
3984 | /* | |
3985 | * If the hole extends beyond i_size, set the hole | |
3986 | * to end after the page that contains i_size | |
3987 | */ | |
3988 | if (offset + length > inode->i_size) { | |
3989 | length = inode->i_size + | |
09cbfeaf | 3990 | PAGE_SIZE - (inode->i_size & (PAGE_SIZE - 1)) - |
26a4c0c6 TT |
3991 | offset; |
3992 | } | |
3993 | ||
a361293f JK |
3994 | if (offset & (sb->s_blocksize - 1) || |
3995 | (offset + length) & (sb->s_blocksize - 1)) { | |
3996 | /* | |
3997 | * Attach jinode to inode for jbd2 if we do any zeroing of | |
3998 | * partial block | |
3999 | */ | |
4000 | ret = ext4_inode_attach_jinode(inode); | |
4001 | if (ret < 0) | |
4002 | goto out_mutex; | |
4003 | ||
4004 | } | |
4005 | ||
ea3d7209 | 4006 | /* Wait all existing dio workers, newcomers will block on i_mutex */ |
ea3d7209 JK |
4007 | inode_dio_wait(inode); |
4008 | ||
4009 | /* | |
4010 | * Prevent page faults from reinstantiating pages we have released from | |
4011 | * page cache. | |
4012 | */ | |
4013 | down_write(&EXT4_I(inode)->i_mmap_sem); | |
430657b6 RZ |
4014 | |
4015 | ret = ext4_break_layouts(inode); | |
4016 | if (ret) | |
4017 | goto out_dio; | |
4018 | ||
a87dd18c LC |
4019 | first_block_offset = round_up(offset, sb->s_blocksize); |
4020 | last_block_offset = round_down((offset + length), sb->s_blocksize) - 1; | |
26a4c0c6 | 4021 | |
a87dd18c | 4022 | /* Now release the pages and zero block aligned part of pages*/ |
01127848 JK |
4023 | if (last_block_offset > first_block_offset) { |
4024 | ret = ext4_update_disksize_before_punch(inode, offset, length); | |
4025 | if (ret) | |
4026 | goto out_dio; | |
a87dd18c LC |
4027 | truncate_pagecache_range(inode, first_block_offset, |
4028 | last_block_offset); | |
01127848 | 4029 | } |
26a4c0c6 TT |
4030 | |
4031 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
4032 | credits = ext4_writepage_trans_blocks(inode); | |
4033 | else | |
4034 | credits = ext4_blocks_for_truncate(inode); | |
4035 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); | |
4036 | if (IS_ERR(handle)) { | |
4037 | ret = PTR_ERR(handle); | |
4038 | ext4_std_error(sb, ret); | |
4039 | goto out_dio; | |
4040 | } | |
4041 | ||
a87dd18c LC |
4042 | ret = ext4_zero_partial_blocks(handle, inode, offset, |
4043 | length); | |
4044 | if (ret) | |
4045 | goto out_stop; | |
26a4c0c6 TT |
4046 | |
4047 | first_block = (offset + sb->s_blocksize - 1) >> | |
4048 | EXT4_BLOCK_SIZE_BITS(sb); | |
4049 | stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb); | |
4050 | ||
eee597ac LC |
4051 | /* If there are blocks to remove, do it */ |
4052 | if (stop_block > first_block) { | |
26a4c0c6 | 4053 | |
eee597ac LC |
4054 | down_write(&EXT4_I(inode)->i_data_sem); |
4055 | ext4_discard_preallocations(inode); | |
26a4c0c6 | 4056 | |
eee597ac LC |
4057 | ret = ext4_es_remove_extent(inode, first_block, |
4058 | stop_block - first_block); | |
4059 | if (ret) { | |
4060 | up_write(&EXT4_I(inode)->i_data_sem); | |
4061 | goto out_stop; | |
4062 | } | |
26a4c0c6 | 4063 | |
eee597ac LC |
4064 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
4065 | ret = ext4_ext_remove_space(inode, first_block, | |
4066 | stop_block - 1); | |
4067 | else | |
4068 | ret = ext4_ind_remove_space(handle, inode, first_block, | |
4069 | stop_block); | |
26a4c0c6 | 4070 | |
eee597ac LC |
4071 | up_write(&EXT4_I(inode)->i_data_sem); |
4072 | } | |
26a4c0c6 TT |
4073 | if (IS_SYNC(inode)) |
4074 | ext4_handle_sync(handle); | |
e251f9bc | 4075 | |
eeca7ea1 | 4076 | inode->i_mtime = inode->i_ctime = current_time(inode); |
26a4c0c6 | 4077 | ext4_mark_inode_dirty(handle, inode); |
67a7d5f5 JK |
4078 | if (ret >= 0) |
4079 | ext4_update_inode_fsync_trans(handle, inode, 1); | |
26a4c0c6 TT |
4080 | out_stop: |
4081 | ext4_journal_stop(handle); | |
4082 | out_dio: | |
ea3d7209 | 4083 | up_write(&EXT4_I(inode)->i_mmap_sem); |
26a4c0c6 | 4084 | out_mutex: |
5955102c | 4085 | inode_unlock(inode); |
26a4c0c6 | 4086 | return ret; |
a4bb6b64 AH |
4087 | } |
4088 | ||
a361293f JK |
4089 | int ext4_inode_attach_jinode(struct inode *inode) |
4090 | { | |
4091 | struct ext4_inode_info *ei = EXT4_I(inode); | |
4092 | struct jbd2_inode *jinode; | |
4093 | ||
4094 | if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal) | |
4095 | return 0; | |
4096 | ||
4097 | jinode = jbd2_alloc_inode(GFP_KERNEL); | |
4098 | spin_lock(&inode->i_lock); | |
4099 | if (!ei->jinode) { | |
4100 | if (!jinode) { | |
4101 | spin_unlock(&inode->i_lock); | |
4102 | return -ENOMEM; | |
4103 | } | |
4104 | ei->jinode = jinode; | |
4105 | jbd2_journal_init_jbd_inode(ei->jinode, inode); | |
4106 | jinode = NULL; | |
4107 | } | |
4108 | spin_unlock(&inode->i_lock); | |
4109 | if (unlikely(jinode != NULL)) | |
4110 | jbd2_free_inode(jinode); | |
4111 | return 0; | |
4112 | } | |
4113 | ||
ac27a0ec | 4114 | /* |
617ba13b | 4115 | * ext4_truncate() |
ac27a0ec | 4116 | * |
617ba13b MC |
4117 | * We block out ext4_get_block() block instantiations across the entire |
4118 | * transaction, and VFS/VM ensures that ext4_truncate() cannot run | |
ac27a0ec DK |
4119 | * simultaneously on behalf of the same inode. |
4120 | * | |
42b2aa86 | 4121 | * As we work through the truncate and commit bits of it to the journal there |
ac27a0ec DK |
4122 | * is one core, guiding principle: the file's tree must always be consistent on |
4123 | * disk. We must be able to restart the truncate after a crash. | |
4124 | * | |
4125 | * The file's tree may be transiently inconsistent in memory (although it | |
4126 | * probably isn't), but whenever we close off and commit a journal transaction, | |
4127 | * the contents of (the filesystem + the journal) must be consistent and | |
4128 | * restartable. It's pretty simple, really: bottom up, right to left (although | |
4129 | * left-to-right works OK too). | |
4130 | * | |
4131 | * Note that at recovery time, journal replay occurs *before* the restart of | |
4132 | * truncate against the orphan inode list. | |
4133 | * | |
4134 | * The committed inode has the new, desired i_size (which is the same as | |
617ba13b | 4135 | * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see |
ac27a0ec | 4136 | * that this inode's truncate did not complete and it will again call |
617ba13b MC |
4137 | * ext4_truncate() to have another go. So there will be instantiated blocks |
4138 | * to the right of the truncation point in a crashed ext4 filesystem. But | |
ac27a0ec | 4139 | * that's fine - as long as they are linked from the inode, the post-crash |
617ba13b | 4140 | * ext4_truncate() run will find them and release them. |
ac27a0ec | 4141 | */ |
2c98eb5e | 4142 | int ext4_truncate(struct inode *inode) |
ac27a0ec | 4143 | { |
819c4920 TT |
4144 | struct ext4_inode_info *ei = EXT4_I(inode); |
4145 | unsigned int credits; | |
2c98eb5e | 4146 | int err = 0; |
819c4920 TT |
4147 | handle_t *handle; |
4148 | struct address_space *mapping = inode->i_mapping; | |
819c4920 | 4149 | |
19b5ef61 TT |
4150 | /* |
4151 | * There is a possibility that we're either freeing the inode | |
e04027e8 | 4152 | * or it's a completely new inode. In those cases we might not |
19b5ef61 TT |
4153 | * have i_mutex locked because it's not necessary. |
4154 | */ | |
4155 | if (!(inode->i_state & (I_NEW|I_FREEING))) | |
5955102c | 4156 | WARN_ON(!inode_is_locked(inode)); |
0562e0ba JZ |
4157 | trace_ext4_truncate_enter(inode); |
4158 | ||
91ef4caf | 4159 | if (!ext4_can_truncate(inode)) |
2c98eb5e | 4160 | return 0; |
ac27a0ec | 4161 | |
5534fb5b | 4162 | if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC)) |
19f5fb7a | 4163 | ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE); |
7d8f9f7d | 4164 | |
aef1c851 TM |
4165 | if (ext4_has_inline_data(inode)) { |
4166 | int has_inline = 1; | |
4167 | ||
01daf945 TT |
4168 | err = ext4_inline_data_truncate(inode, &has_inline); |
4169 | if (err) | |
4170 | return err; | |
aef1c851 | 4171 | if (has_inline) |
2c98eb5e | 4172 | return 0; |
aef1c851 TM |
4173 | } |
4174 | ||
a361293f JK |
4175 | /* If we zero-out tail of the page, we have to create jinode for jbd2 */ |
4176 | if (inode->i_size & (inode->i_sb->s_blocksize - 1)) { | |
4177 | if (ext4_inode_attach_jinode(inode) < 0) | |
2c98eb5e | 4178 | return 0; |
a361293f JK |
4179 | } |
4180 | ||
819c4920 TT |
4181 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
4182 | credits = ext4_writepage_trans_blocks(inode); | |
4183 | else | |
4184 | credits = ext4_blocks_for_truncate(inode); | |
4185 | ||
4186 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); | |
2c98eb5e TT |
4187 | if (IS_ERR(handle)) |
4188 | return PTR_ERR(handle); | |
819c4920 | 4189 | |
eb3544c6 LC |
4190 | if (inode->i_size & (inode->i_sb->s_blocksize - 1)) |
4191 | ext4_block_truncate_page(handle, mapping, inode->i_size); | |
819c4920 TT |
4192 | |
4193 | /* | |
4194 | * We add the inode to the orphan list, so that if this | |
4195 | * truncate spans multiple transactions, and we crash, we will | |
4196 | * resume the truncate when the filesystem recovers. It also | |
4197 | * marks the inode dirty, to catch the new size. | |
4198 | * | |
4199 | * Implication: the file must always be in a sane, consistent | |
4200 | * truncatable state while each transaction commits. | |
4201 | */ | |
2c98eb5e TT |
4202 | err = ext4_orphan_add(handle, inode); |
4203 | if (err) | |
819c4920 TT |
4204 | goto out_stop; |
4205 | ||
4206 | down_write(&EXT4_I(inode)->i_data_sem); | |
4207 | ||
4208 | ext4_discard_preallocations(inode); | |
4209 | ||
ff9893dc | 4210 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
d0abb36d | 4211 | err = ext4_ext_truncate(handle, inode); |
ff9893dc | 4212 | else |
819c4920 TT |
4213 | ext4_ind_truncate(handle, inode); |
4214 | ||
4215 | up_write(&ei->i_data_sem); | |
d0abb36d TT |
4216 | if (err) |
4217 | goto out_stop; | |
819c4920 TT |
4218 | |
4219 | if (IS_SYNC(inode)) | |
4220 | ext4_handle_sync(handle); | |
4221 | ||
4222 | out_stop: | |
4223 | /* | |
4224 | * If this was a simple ftruncate() and the file will remain alive, | |
4225 | * then we need to clear up the orphan record which we created above. | |
4226 | * However, if this was a real unlink then we were called by | |
58d86a50 | 4227 | * ext4_evict_inode(), and we allow that function to clean up the |
819c4920 TT |
4228 | * orphan info for us. |
4229 | */ | |
4230 | if (inode->i_nlink) | |
4231 | ext4_orphan_del(handle, inode); | |
4232 | ||
eeca7ea1 | 4233 | inode->i_mtime = inode->i_ctime = current_time(inode); |
819c4920 TT |
4234 | ext4_mark_inode_dirty(handle, inode); |
4235 | ext4_journal_stop(handle); | |
ac27a0ec | 4236 | |
0562e0ba | 4237 | trace_ext4_truncate_exit(inode); |
2c98eb5e | 4238 | return err; |
ac27a0ec DK |
4239 | } |
4240 | ||
ac27a0ec | 4241 | /* |
617ba13b | 4242 | * ext4_get_inode_loc returns with an extra refcount against the inode's |
ac27a0ec DK |
4243 | * underlying buffer_head on success. If 'in_mem' is true, we have all |
4244 | * data in memory that is needed to recreate the on-disk version of this | |
4245 | * inode. | |
4246 | */ | |
617ba13b MC |
4247 | static int __ext4_get_inode_loc(struct inode *inode, |
4248 | struct ext4_iloc *iloc, int in_mem) | |
ac27a0ec | 4249 | { |
240799cd TT |
4250 | struct ext4_group_desc *gdp; |
4251 | struct buffer_head *bh; | |
4252 | struct super_block *sb = inode->i_sb; | |
4253 | ext4_fsblk_t block; | |
02f03c42 | 4254 | struct blk_plug plug; |
240799cd TT |
4255 | int inodes_per_block, inode_offset; |
4256 | ||
3a06d778 | 4257 | iloc->bh = NULL; |
c37e9e01 TT |
4258 | if (inode->i_ino < EXT4_ROOT_INO || |
4259 | inode->i_ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count)) | |
6a797d27 | 4260 | return -EFSCORRUPTED; |
ac27a0ec | 4261 | |
240799cd TT |
4262 | iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb); |
4263 | gdp = ext4_get_group_desc(sb, iloc->block_group, NULL); | |
4264 | if (!gdp) | |
ac27a0ec DK |
4265 | return -EIO; |
4266 | ||
240799cd TT |
4267 | /* |
4268 | * Figure out the offset within the block group inode table | |
4269 | */ | |
00d09882 | 4270 | inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; |
240799cd TT |
4271 | inode_offset = ((inode->i_ino - 1) % |
4272 | EXT4_INODES_PER_GROUP(sb)); | |
4273 | block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block); | |
4274 | iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb); | |
4275 | ||
4276 | bh = sb_getblk(sb, block); | |
aebf0243 | 4277 | if (unlikely(!bh)) |
860d21e2 | 4278 | return -ENOMEM; |
46f870d6 TT |
4279 | if (ext4_simulate_fail(sb, EXT4_SIM_INODE_EIO)) |
4280 | goto simulate_eio; | |
ac27a0ec DK |
4281 | if (!buffer_uptodate(bh)) { |
4282 | lock_buffer(bh); | |
9c83a923 HK |
4283 | |
4284 | /* | |
4285 | * If the buffer has the write error flag, we have failed | |
4286 | * to write out another inode in the same block. In this | |
4287 | * case, we don't have to read the block because we may | |
4288 | * read the old inode data successfully. | |
4289 | */ | |
4290 | if (buffer_write_io_error(bh) && !buffer_uptodate(bh)) | |
4291 | set_buffer_uptodate(bh); | |
4292 | ||
ac27a0ec DK |
4293 | if (buffer_uptodate(bh)) { |
4294 | /* someone brought it uptodate while we waited */ | |
4295 | unlock_buffer(bh); | |
4296 | goto has_buffer; | |
4297 | } | |
4298 | ||
4299 | /* | |
4300 | * If we have all information of the inode in memory and this | |
4301 | * is the only valid inode in the block, we need not read the | |
4302 | * block. | |
4303 | */ | |
4304 | if (in_mem) { | |
4305 | struct buffer_head *bitmap_bh; | |
240799cd | 4306 | int i, start; |
ac27a0ec | 4307 | |
240799cd | 4308 | start = inode_offset & ~(inodes_per_block - 1); |
ac27a0ec | 4309 | |
240799cd TT |
4310 | /* Is the inode bitmap in cache? */ |
4311 | bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp)); | |
aebf0243 | 4312 | if (unlikely(!bitmap_bh)) |
ac27a0ec DK |
4313 | goto make_io; |
4314 | ||
4315 | /* | |
4316 | * If the inode bitmap isn't in cache then the | |
4317 | * optimisation may end up performing two reads instead | |
4318 | * of one, so skip it. | |
4319 | */ | |
4320 | if (!buffer_uptodate(bitmap_bh)) { | |
4321 | brelse(bitmap_bh); | |
4322 | goto make_io; | |
4323 | } | |
240799cd | 4324 | for (i = start; i < start + inodes_per_block; i++) { |
ac27a0ec DK |
4325 | if (i == inode_offset) |
4326 | continue; | |
617ba13b | 4327 | if (ext4_test_bit(i, bitmap_bh->b_data)) |
ac27a0ec DK |
4328 | break; |
4329 | } | |
4330 | brelse(bitmap_bh); | |
240799cd | 4331 | if (i == start + inodes_per_block) { |
ac27a0ec DK |
4332 | /* all other inodes are free, so skip I/O */ |
4333 | memset(bh->b_data, 0, bh->b_size); | |
4334 | set_buffer_uptodate(bh); | |
4335 | unlock_buffer(bh); | |
4336 | goto has_buffer; | |
4337 | } | |
4338 | } | |
4339 | ||
4340 | make_io: | |
240799cd TT |
4341 | /* |
4342 | * If we need to do any I/O, try to pre-readahead extra | |
4343 | * blocks from the inode table. | |
4344 | */ | |
02f03c42 | 4345 | blk_start_plug(&plug); |
240799cd TT |
4346 | if (EXT4_SB(sb)->s_inode_readahead_blks) { |
4347 | ext4_fsblk_t b, end, table; | |
4348 | unsigned num; | |
0d606e2c | 4349 | __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks; |
240799cd TT |
4350 | |
4351 | table = ext4_inode_table(sb, gdp); | |
b713a5ec | 4352 | /* s_inode_readahead_blks is always a power of 2 */ |
0d606e2c | 4353 | b = block & ~((ext4_fsblk_t) ra_blks - 1); |
240799cd TT |
4354 | if (table > b) |
4355 | b = table; | |
0d606e2c | 4356 | end = b + ra_blks; |
240799cd | 4357 | num = EXT4_INODES_PER_GROUP(sb); |
feb0ab32 | 4358 | if (ext4_has_group_desc_csum(sb)) |
560671a0 | 4359 | num -= ext4_itable_unused_count(sb, gdp); |
240799cd TT |
4360 | table += num / inodes_per_block; |
4361 | if (end > table) | |
4362 | end = table; | |
4363 | while (b <= end) | |
d87f6392 | 4364 | sb_breadahead_unmovable(sb, b++); |
240799cd TT |
4365 | } |
4366 | ||
ac27a0ec DK |
4367 | /* |
4368 | * There are other valid inodes in the buffer, this inode | |
4369 | * has in-inode xattrs, or we don't have this inode in memory. | |
4370 | * Read the block from disk. | |
4371 | */ | |
0562e0ba | 4372 | trace_ext4_load_inode(inode); |
ac27a0ec DK |
4373 | get_bh(bh); |
4374 | bh->b_end_io = end_buffer_read_sync; | |
2a222ca9 | 4375 | submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh); |
02f03c42 | 4376 | blk_finish_plug(&plug); |
ac27a0ec DK |
4377 | wait_on_buffer(bh); |
4378 | if (!buffer_uptodate(bh)) { | |
46f870d6 | 4379 | simulate_eio: |
54d3adbc | 4380 | ext4_error_inode_block(inode, block, EIO, |
c398eda0 | 4381 | "unable to read itable block"); |
ac27a0ec DK |
4382 | brelse(bh); |
4383 | return -EIO; | |
4384 | } | |
4385 | } | |
4386 | has_buffer: | |
4387 | iloc->bh = bh; | |
4388 | return 0; | |
4389 | } | |
4390 | ||
617ba13b | 4391 | int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec DK |
4392 | { |
4393 | /* We have all inode data except xattrs in memory here. */ | |
617ba13b | 4394 | return __ext4_get_inode_loc(inode, iloc, |
19f5fb7a | 4395 | !ext4_test_inode_state(inode, EXT4_STATE_XATTR)); |
ac27a0ec DK |
4396 | } |
4397 | ||
6642586b RZ |
4398 | static bool ext4_should_use_dax(struct inode *inode) |
4399 | { | |
4400 | if (!test_opt(inode->i_sb, DAX)) | |
4401 | return false; | |
4402 | if (!S_ISREG(inode->i_mode)) | |
4403 | return false; | |
4404 | if (ext4_should_journal_data(inode)) | |
4405 | return false; | |
4406 | if (ext4_has_inline_data(inode)) | |
4407 | return false; | |
592ddec7 | 4408 | if (ext4_test_inode_flag(inode, EXT4_INODE_ENCRYPT)) |
6642586b | 4409 | return false; |
c93d8f88 EB |
4410 | if (ext4_test_inode_flag(inode, EXT4_INODE_VERITY)) |
4411 | return false; | |
6642586b RZ |
4412 | return true; |
4413 | } | |
4414 | ||
617ba13b | 4415 | void ext4_set_inode_flags(struct inode *inode) |
ac27a0ec | 4416 | { |
617ba13b | 4417 | unsigned int flags = EXT4_I(inode)->i_flags; |
00a1a053 | 4418 | unsigned int new_fl = 0; |
ac27a0ec | 4419 | |
617ba13b | 4420 | if (flags & EXT4_SYNC_FL) |
00a1a053 | 4421 | new_fl |= S_SYNC; |
617ba13b | 4422 | if (flags & EXT4_APPEND_FL) |
00a1a053 | 4423 | new_fl |= S_APPEND; |
617ba13b | 4424 | if (flags & EXT4_IMMUTABLE_FL) |
00a1a053 | 4425 | new_fl |= S_IMMUTABLE; |
617ba13b | 4426 | if (flags & EXT4_NOATIME_FL) |
00a1a053 | 4427 | new_fl |= S_NOATIME; |
617ba13b | 4428 | if (flags & EXT4_DIRSYNC_FL) |
00a1a053 | 4429 | new_fl |= S_DIRSYNC; |
6642586b | 4430 | if (ext4_should_use_dax(inode)) |
923ae0ff | 4431 | new_fl |= S_DAX; |
2ee6a576 EB |
4432 | if (flags & EXT4_ENCRYPT_FL) |
4433 | new_fl |= S_ENCRYPTED; | |
b886ee3e GKB |
4434 | if (flags & EXT4_CASEFOLD_FL) |
4435 | new_fl |= S_CASEFOLD; | |
c93d8f88 EB |
4436 | if (flags & EXT4_VERITY_FL) |
4437 | new_fl |= S_VERITY; | |
5f16f322 | 4438 | inode_set_flags(inode, new_fl, |
2ee6a576 | 4439 | S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX| |
c93d8f88 | 4440 | S_ENCRYPTED|S_CASEFOLD|S_VERITY); |
ac27a0ec DK |
4441 | } |
4442 | ||
0fc1b451 | 4443 | static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode, |
de9a55b8 | 4444 | struct ext4_inode_info *ei) |
0fc1b451 AK |
4445 | { |
4446 | blkcnt_t i_blocks ; | |
8180a562 AK |
4447 | struct inode *inode = &(ei->vfs_inode); |
4448 | struct super_block *sb = inode->i_sb; | |
0fc1b451 | 4449 | |
e2b911c5 | 4450 | if (ext4_has_feature_huge_file(sb)) { |
0fc1b451 AK |
4451 | /* we are using combined 48 bit field */ |
4452 | i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 | | |
4453 | le32_to_cpu(raw_inode->i_blocks_lo); | |
07a03824 | 4454 | if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) { |
8180a562 AK |
4455 | /* i_blocks represent file system block size */ |
4456 | return i_blocks << (inode->i_blkbits - 9); | |
4457 | } else { | |
4458 | return i_blocks; | |
4459 | } | |
0fc1b451 AK |
4460 | } else { |
4461 | return le32_to_cpu(raw_inode->i_blocks_lo); | |
4462 | } | |
4463 | } | |
ff9ddf7e | 4464 | |
eb9b5f01 | 4465 | static inline int ext4_iget_extra_inode(struct inode *inode, |
152a7b0a TM |
4466 | struct ext4_inode *raw_inode, |
4467 | struct ext4_inode_info *ei) | |
4468 | { | |
4469 | __le32 *magic = (void *)raw_inode + | |
4470 | EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize; | |
eb9b5f01 | 4471 | |
290ab230 EB |
4472 | if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize + sizeof(__le32) <= |
4473 | EXT4_INODE_SIZE(inode->i_sb) && | |
4474 | *magic == cpu_to_le32(EXT4_XATTR_MAGIC)) { | |
152a7b0a | 4475 | ext4_set_inode_state(inode, EXT4_STATE_XATTR); |
eb9b5f01 | 4476 | return ext4_find_inline_data_nolock(inode); |
f19d5870 TM |
4477 | } else |
4478 | EXT4_I(inode)->i_inline_off = 0; | |
eb9b5f01 | 4479 | return 0; |
152a7b0a TM |
4480 | } |
4481 | ||
040cb378 LX |
4482 | int ext4_get_projid(struct inode *inode, kprojid_t *projid) |
4483 | { | |
0b7b7779 | 4484 | if (!ext4_has_feature_project(inode->i_sb)) |
040cb378 LX |
4485 | return -EOPNOTSUPP; |
4486 | *projid = EXT4_I(inode)->i_projid; | |
4487 | return 0; | |
4488 | } | |
4489 | ||
e254d1af EG |
4490 | /* |
4491 | * ext4 has self-managed i_version for ea inodes, it stores the lower 32bit of | |
4492 | * refcount in i_version, so use raw values if inode has EXT4_EA_INODE_FL flag | |
4493 | * set. | |
4494 | */ | |
4495 | static inline void ext4_inode_set_iversion_queried(struct inode *inode, u64 val) | |
4496 | { | |
4497 | if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) | |
4498 | inode_set_iversion_raw(inode, val); | |
4499 | else | |
4500 | inode_set_iversion_queried(inode, val); | |
4501 | } | |
4502 | static inline u64 ext4_inode_peek_iversion(const struct inode *inode) | |
4503 | { | |
4504 | if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) | |
4505 | return inode_peek_iversion_raw(inode); | |
4506 | else | |
4507 | return inode_peek_iversion(inode); | |
4508 | } | |
4509 | ||
8a363970 TT |
4510 | struct inode *__ext4_iget(struct super_block *sb, unsigned long ino, |
4511 | ext4_iget_flags flags, const char *function, | |
4512 | unsigned int line) | |
ac27a0ec | 4513 | { |
617ba13b MC |
4514 | struct ext4_iloc iloc; |
4515 | struct ext4_inode *raw_inode; | |
1d1fe1ee | 4516 | struct ext4_inode_info *ei; |
1d1fe1ee | 4517 | struct inode *inode; |
b436b9be | 4518 | journal_t *journal = EXT4_SB(sb)->s_journal; |
1d1fe1ee | 4519 | long ret; |
7e6e1ef4 | 4520 | loff_t size; |
ac27a0ec | 4521 | int block; |
08cefc7a EB |
4522 | uid_t i_uid; |
4523 | gid_t i_gid; | |
040cb378 | 4524 | projid_t i_projid; |
ac27a0ec | 4525 | |
191ce178 | 4526 | if ((!(flags & EXT4_IGET_SPECIAL) && |
8a363970 TT |
4527 | (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)) || |
4528 | (ino < EXT4_ROOT_INO) || | |
4529 | (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))) { | |
4530 | if (flags & EXT4_IGET_HANDLE) | |
4531 | return ERR_PTR(-ESTALE); | |
54d3adbc | 4532 | __ext4_error(sb, function, line, EFSCORRUPTED, 0, |
8a363970 TT |
4533 | "inode #%lu: comm %s: iget: illegal inode #", |
4534 | ino, current->comm); | |
4535 | return ERR_PTR(-EFSCORRUPTED); | |
4536 | } | |
4537 | ||
1d1fe1ee DH |
4538 | inode = iget_locked(sb, ino); |
4539 | if (!inode) | |
4540 | return ERR_PTR(-ENOMEM); | |
4541 | if (!(inode->i_state & I_NEW)) | |
4542 | return inode; | |
4543 | ||
4544 | ei = EXT4_I(inode); | |
7dc57615 | 4545 | iloc.bh = NULL; |
ac27a0ec | 4546 | |
1d1fe1ee DH |
4547 | ret = __ext4_get_inode_loc(inode, &iloc, 0); |
4548 | if (ret < 0) | |
ac27a0ec | 4549 | goto bad_inode; |
617ba13b | 4550 | raw_inode = ext4_raw_inode(&iloc); |
814525f4 | 4551 | |
8e4b5eae | 4552 | if ((ino == EXT4_ROOT_INO) && (raw_inode->i_links_count == 0)) { |
8a363970 TT |
4553 | ext4_error_inode(inode, function, line, 0, |
4554 | "iget: root inode unallocated"); | |
8e4b5eae TT |
4555 | ret = -EFSCORRUPTED; |
4556 | goto bad_inode; | |
4557 | } | |
4558 | ||
8a363970 TT |
4559 | if ((flags & EXT4_IGET_HANDLE) && |
4560 | (raw_inode->i_links_count == 0) && (raw_inode->i_mode == 0)) { | |
4561 | ret = -ESTALE; | |
4562 | goto bad_inode; | |
4563 | } | |
4564 | ||
814525f4 DW |
4565 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
4566 | ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize); | |
4567 | if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > | |
2dc8d9e1 EB |
4568 | EXT4_INODE_SIZE(inode->i_sb) || |
4569 | (ei->i_extra_isize & 3)) { | |
8a363970 TT |
4570 | ext4_error_inode(inode, function, line, 0, |
4571 | "iget: bad extra_isize %u " | |
4572 | "(inode size %u)", | |
2dc8d9e1 EB |
4573 | ei->i_extra_isize, |
4574 | EXT4_INODE_SIZE(inode->i_sb)); | |
6a797d27 | 4575 | ret = -EFSCORRUPTED; |
814525f4 DW |
4576 | goto bad_inode; |
4577 | } | |
4578 | } else | |
4579 | ei->i_extra_isize = 0; | |
4580 | ||
4581 | /* Precompute checksum seed for inode metadata */ | |
9aa5d32b | 4582 | if (ext4_has_metadata_csum(sb)) { |
814525f4 DW |
4583 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
4584 | __u32 csum; | |
4585 | __le32 inum = cpu_to_le32(inode->i_ino); | |
4586 | __le32 gen = raw_inode->i_generation; | |
4587 | csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum, | |
4588 | sizeof(inum)); | |
4589 | ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen, | |
4590 | sizeof(gen)); | |
4591 | } | |
4592 | ||
46f870d6 TT |
4593 | if (!ext4_inode_csum_verify(inode, raw_inode, ei) || |
4594 | ext4_simulate_fail(sb, EXT4_SIM_INODE_CRC)) { | |
54d3adbc TT |
4595 | ext4_error_inode_err(inode, function, line, 0, EFSBADCRC, |
4596 | "iget: checksum invalid"); | |
6a797d27 | 4597 | ret = -EFSBADCRC; |
814525f4 DW |
4598 | goto bad_inode; |
4599 | } | |
4600 | ||
ac27a0ec | 4601 | inode->i_mode = le16_to_cpu(raw_inode->i_mode); |
08cefc7a EB |
4602 | i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low); |
4603 | i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low); | |
0b7b7779 | 4604 | if (ext4_has_feature_project(sb) && |
040cb378 LX |
4605 | EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE && |
4606 | EXT4_FITS_IN_INODE(raw_inode, ei, i_projid)) | |
4607 | i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid); | |
4608 | else | |
4609 | i_projid = EXT4_DEF_PROJID; | |
4610 | ||
af5bc92d | 4611 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
08cefc7a EB |
4612 | i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16; |
4613 | i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16; | |
ac27a0ec | 4614 | } |
08cefc7a EB |
4615 | i_uid_write(inode, i_uid); |
4616 | i_gid_write(inode, i_gid); | |
040cb378 | 4617 | ei->i_projid = make_kprojid(&init_user_ns, i_projid); |
bfe86848 | 4618 | set_nlink(inode, le16_to_cpu(raw_inode->i_links_count)); |
ac27a0ec | 4619 | |
353eb83c | 4620 | ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */ |
67cf5b09 | 4621 | ei->i_inline_off = 0; |
ac27a0ec DK |
4622 | ei->i_dir_start_lookup = 0; |
4623 | ei->i_dtime = le32_to_cpu(raw_inode->i_dtime); | |
4624 | /* We now have enough fields to check if the inode was active or not. | |
4625 | * This is needed because nfsd might try to access dead inodes | |
4626 | * the test is that same one that e2fsck uses | |
4627 | * NeilBrown 1999oct15 | |
4628 | */ | |
4629 | if (inode->i_nlink == 0) { | |
393d1d1d DTB |
4630 | if ((inode->i_mode == 0 || |
4631 | !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) && | |
4632 | ino != EXT4_BOOT_LOADER_INO) { | |
ac27a0ec | 4633 | /* this inode is deleted */ |
1d1fe1ee | 4634 | ret = -ESTALE; |
ac27a0ec DK |
4635 | goto bad_inode; |
4636 | } | |
4637 | /* The only unlinked inodes we let through here have | |
4638 | * valid i_mode and are being read by the orphan | |
4639 | * recovery code: that's fine, we're about to complete | |
393d1d1d DTB |
4640 | * the process of deleting those. |
4641 | * OR it is the EXT4_BOOT_LOADER_INO which is | |
4642 | * not initialized on a new filesystem. */ | |
ac27a0ec | 4643 | } |
ac27a0ec | 4644 | ei->i_flags = le32_to_cpu(raw_inode->i_flags); |
cce6c9f7 | 4645 | ext4_set_inode_flags(inode); |
0fc1b451 | 4646 | inode->i_blocks = ext4_inode_blocks(raw_inode, ei); |
7973c0c1 | 4647 | ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo); |
e2b911c5 | 4648 | if (ext4_has_feature_64bit(sb)) |
a1ddeb7e BP |
4649 | ei->i_file_acl |= |
4650 | ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32; | |
e08ac99f | 4651 | inode->i_size = ext4_isize(sb, raw_inode); |
7e6e1ef4 | 4652 | if ((size = i_size_read(inode)) < 0) { |
8a363970 TT |
4653 | ext4_error_inode(inode, function, line, 0, |
4654 | "iget: bad i_size value: %lld", size); | |
7e6e1ef4 DW |
4655 | ret = -EFSCORRUPTED; |
4656 | goto bad_inode; | |
4657 | } | |
48a34311 JK |
4658 | /* |
4659 | * If dir_index is not enabled but there's dir with INDEX flag set, | |
4660 | * we'd normally treat htree data as empty space. But with metadata | |
4661 | * checksumming that corrupts checksums so forbid that. | |
4662 | */ | |
4663 | if (!ext4_has_feature_dir_index(sb) && ext4_has_metadata_csum(sb) && | |
4664 | ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) { | |
4665 | ext4_error_inode(inode, function, line, 0, | |
4666 | "iget: Dir with htree data on filesystem without dir_index feature."); | |
4667 | ret = -EFSCORRUPTED; | |
4668 | goto bad_inode; | |
4669 | } | |
ac27a0ec | 4670 | ei->i_disksize = inode->i_size; |
a9e7f447 DM |
4671 | #ifdef CONFIG_QUOTA |
4672 | ei->i_reserved_quota = 0; | |
4673 | #endif | |
ac27a0ec DK |
4674 | inode->i_generation = le32_to_cpu(raw_inode->i_generation); |
4675 | ei->i_block_group = iloc.block_group; | |
a4912123 | 4676 | ei->i_last_alloc_group = ~0; |
ac27a0ec DK |
4677 | /* |
4678 | * NOTE! The in-memory inode i_data array is in little-endian order | |
4679 | * even on big-endian machines: we do NOT byteswap the block numbers! | |
4680 | */ | |
617ba13b | 4681 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
ac27a0ec DK |
4682 | ei->i_data[block] = raw_inode->i_block[block]; |
4683 | INIT_LIST_HEAD(&ei->i_orphan); | |
4684 | ||
b436b9be JK |
4685 | /* |
4686 | * Set transaction id's of transactions that have to be committed | |
4687 | * to finish f[data]sync. We set them to currently running transaction | |
4688 | * as we cannot be sure that the inode or some of its metadata isn't | |
4689 | * part of the transaction - the inode could have been reclaimed and | |
4690 | * now it is reread from disk. | |
4691 | */ | |
4692 | if (journal) { | |
4693 | transaction_t *transaction; | |
4694 | tid_t tid; | |
4695 | ||
a931da6a | 4696 | read_lock(&journal->j_state_lock); |
b436b9be JK |
4697 | if (journal->j_running_transaction) |
4698 | transaction = journal->j_running_transaction; | |
4699 | else | |
4700 | transaction = journal->j_committing_transaction; | |
4701 | if (transaction) | |
4702 | tid = transaction->t_tid; | |
4703 | else | |
4704 | tid = journal->j_commit_sequence; | |
a931da6a | 4705 | read_unlock(&journal->j_state_lock); |
b436b9be JK |
4706 | ei->i_sync_tid = tid; |
4707 | ei->i_datasync_tid = tid; | |
4708 | } | |
4709 | ||
0040d987 | 4710 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
ac27a0ec DK |
4711 | if (ei->i_extra_isize == 0) { |
4712 | /* The extra space is currently unused. Use it. */ | |
2dc8d9e1 | 4713 | BUILD_BUG_ON(sizeof(struct ext4_inode) & 3); |
617ba13b MC |
4714 | ei->i_extra_isize = sizeof(struct ext4_inode) - |
4715 | EXT4_GOOD_OLD_INODE_SIZE; | |
ac27a0ec | 4716 | } else { |
eb9b5f01 TT |
4717 | ret = ext4_iget_extra_inode(inode, raw_inode, ei); |
4718 | if (ret) | |
4719 | goto bad_inode; | |
ac27a0ec | 4720 | } |
814525f4 | 4721 | } |
ac27a0ec | 4722 | |
ef7f3835 KS |
4723 | EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode); |
4724 | EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode); | |
4725 | EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode); | |
4726 | EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode); | |
4727 | ||
ed3654eb | 4728 | if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) { |
ee73f9a5 JL |
4729 | u64 ivers = le32_to_cpu(raw_inode->i_disk_version); |
4730 | ||
c4f65706 TT |
4731 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
4732 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
ee73f9a5 | 4733 | ivers |= |
c4f65706 TT |
4734 | (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32; |
4735 | } | |
e254d1af | 4736 | ext4_inode_set_iversion_queried(inode, ivers); |
25ec56b5 JNC |
4737 | } |
4738 | ||
c4b5a614 | 4739 | ret = 0; |
485c26ec | 4740 | if (ei->i_file_acl && |
1032988c | 4741 | !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) { |
8a363970 TT |
4742 | ext4_error_inode(inode, function, line, 0, |
4743 | "iget: bad extended attribute block %llu", | |
24676da4 | 4744 | ei->i_file_acl); |
6a797d27 | 4745 | ret = -EFSCORRUPTED; |
485c26ec | 4746 | goto bad_inode; |
f19d5870 | 4747 | } else if (!ext4_has_inline_data(inode)) { |
bc716523 LS |
4748 | /* validate the block references in the inode */ |
4749 | if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || | |
4750 | (S_ISLNK(inode->i_mode) && | |
4751 | !ext4_inode_is_fast_symlink(inode))) { | |
4752 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
f19d5870 | 4753 | ret = ext4_ext_check_inode(inode); |
bc716523 LS |
4754 | else |
4755 | ret = ext4_ind_check_inode(inode); | |
f19d5870 | 4756 | } |
fe2c8191 | 4757 | } |
567f3e9a | 4758 | if (ret) |
de9a55b8 | 4759 | goto bad_inode; |
7a262f7c | 4760 | |
ac27a0ec | 4761 | if (S_ISREG(inode->i_mode)) { |
617ba13b | 4762 | inode->i_op = &ext4_file_inode_operations; |
be64f884 | 4763 | inode->i_fop = &ext4_file_operations; |
617ba13b | 4764 | ext4_set_aops(inode); |
ac27a0ec | 4765 | } else if (S_ISDIR(inode->i_mode)) { |
617ba13b MC |
4766 | inode->i_op = &ext4_dir_inode_operations; |
4767 | inode->i_fop = &ext4_dir_operations; | |
ac27a0ec | 4768 | } else if (S_ISLNK(inode->i_mode)) { |
6390d33b LR |
4769 | /* VFS does not allow setting these so must be corruption */ |
4770 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) { | |
8a363970 TT |
4771 | ext4_error_inode(inode, function, line, 0, |
4772 | "iget: immutable or append flags " | |
4773 | "not allowed on symlinks"); | |
6390d33b LR |
4774 | ret = -EFSCORRUPTED; |
4775 | goto bad_inode; | |
4776 | } | |
592ddec7 | 4777 | if (IS_ENCRYPTED(inode)) { |
a7a67e8a AV |
4778 | inode->i_op = &ext4_encrypted_symlink_inode_operations; |
4779 | ext4_set_aops(inode); | |
4780 | } else if (ext4_inode_is_fast_symlink(inode)) { | |
75e7566b | 4781 | inode->i_link = (char *)ei->i_data; |
617ba13b | 4782 | inode->i_op = &ext4_fast_symlink_inode_operations; |
e83c1397 DG |
4783 | nd_terminate_link(ei->i_data, inode->i_size, |
4784 | sizeof(ei->i_data) - 1); | |
4785 | } else { | |
617ba13b MC |
4786 | inode->i_op = &ext4_symlink_inode_operations; |
4787 | ext4_set_aops(inode); | |
ac27a0ec | 4788 | } |
21fc61c7 | 4789 | inode_nohighmem(inode); |
563bdd61 TT |
4790 | } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) || |
4791 | S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) { | |
617ba13b | 4792 | inode->i_op = &ext4_special_inode_operations; |
ac27a0ec DK |
4793 | if (raw_inode->i_block[0]) |
4794 | init_special_inode(inode, inode->i_mode, | |
4795 | old_decode_dev(le32_to_cpu(raw_inode->i_block[0]))); | |
4796 | else | |
4797 | init_special_inode(inode, inode->i_mode, | |
4798 | new_decode_dev(le32_to_cpu(raw_inode->i_block[1]))); | |
393d1d1d DTB |
4799 | } else if (ino == EXT4_BOOT_LOADER_INO) { |
4800 | make_bad_inode(inode); | |
563bdd61 | 4801 | } else { |
6a797d27 | 4802 | ret = -EFSCORRUPTED; |
8a363970 TT |
4803 | ext4_error_inode(inode, function, line, 0, |
4804 | "iget: bogus i_mode (%o)", inode->i_mode); | |
563bdd61 | 4805 | goto bad_inode; |
ac27a0ec | 4806 | } |
6456ca65 TT |
4807 | if (IS_CASEFOLDED(inode) && !ext4_has_feature_casefold(inode->i_sb)) |
4808 | ext4_error_inode(inode, function, line, 0, | |
4809 | "casefold flag without casefold feature"); | |
af5bc92d | 4810 | brelse(iloc.bh); |
dec214d0 | 4811 | |
1d1fe1ee DH |
4812 | unlock_new_inode(inode); |
4813 | return inode; | |
ac27a0ec DK |
4814 | |
4815 | bad_inode: | |
567f3e9a | 4816 | brelse(iloc.bh); |
1d1fe1ee DH |
4817 | iget_failed(inode); |
4818 | return ERR_PTR(ret); | |
ac27a0ec DK |
4819 | } |
4820 | ||
0fc1b451 AK |
4821 | static int ext4_inode_blocks_set(handle_t *handle, |
4822 | struct ext4_inode *raw_inode, | |
4823 | struct ext4_inode_info *ei) | |
4824 | { | |
4825 | struct inode *inode = &(ei->vfs_inode); | |
28936b62 | 4826 | u64 i_blocks = READ_ONCE(inode->i_blocks); |
0fc1b451 | 4827 | struct super_block *sb = inode->i_sb; |
0fc1b451 AK |
4828 | |
4829 | if (i_blocks <= ~0U) { | |
4830 | /* | |
4907cb7b | 4831 | * i_blocks can be represented in a 32 bit variable |
0fc1b451 AK |
4832 | * as multiple of 512 bytes |
4833 | */ | |
8180a562 | 4834 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 4835 | raw_inode->i_blocks_high = 0; |
84a8dce2 | 4836 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
f287a1a5 TT |
4837 | return 0; |
4838 | } | |
e2b911c5 | 4839 | if (!ext4_has_feature_huge_file(sb)) |
f287a1a5 TT |
4840 | return -EFBIG; |
4841 | ||
4842 | if (i_blocks <= 0xffffffffffffULL) { | |
0fc1b451 AK |
4843 | /* |
4844 | * i_blocks can be represented in a 48 bit variable | |
4845 | * as multiple of 512 bytes | |
4846 | */ | |
8180a562 | 4847 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 4848 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); |
84a8dce2 | 4849 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
0fc1b451 | 4850 | } else { |
84a8dce2 | 4851 | ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
8180a562 AK |
4852 | /* i_block is stored in file system block size */ |
4853 | i_blocks = i_blocks >> (inode->i_blkbits - 9); | |
4854 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); | |
4855 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); | |
0fc1b451 | 4856 | } |
f287a1a5 | 4857 | return 0; |
0fc1b451 AK |
4858 | } |
4859 | ||
a26f4992 TT |
4860 | struct other_inode { |
4861 | unsigned long orig_ino; | |
4862 | struct ext4_inode *raw_inode; | |
4863 | }; | |
4864 | ||
4865 | static int other_inode_match(struct inode * inode, unsigned long ino, | |
4866 | void *data) | |
4867 | { | |
4868 | struct other_inode *oi = (struct other_inode *) data; | |
4869 | ||
4870 | if ((inode->i_ino != ino) || | |
4871 | (inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW | | |
0e11f644 | 4872 | I_DIRTY_INODE)) || |
a26f4992 TT |
4873 | ((inode->i_state & I_DIRTY_TIME) == 0)) |
4874 | return 0; | |
4875 | spin_lock(&inode->i_lock); | |
4876 | if (((inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW | | |
0e11f644 | 4877 | I_DIRTY_INODE)) == 0) && |
a26f4992 TT |
4878 | (inode->i_state & I_DIRTY_TIME)) { |
4879 | struct ext4_inode_info *ei = EXT4_I(inode); | |
4880 | ||
4881 | inode->i_state &= ~(I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED); | |
4882 | spin_unlock(&inode->i_lock); | |
4883 | ||
4884 | spin_lock(&ei->i_raw_lock); | |
4885 | EXT4_INODE_SET_XTIME(i_ctime, inode, oi->raw_inode); | |
4886 | EXT4_INODE_SET_XTIME(i_mtime, inode, oi->raw_inode); | |
4887 | EXT4_INODE_SET_XTIME(i_atime, inode, oi->raw_inode); | |
4888 | ext4_inode_csum_set(inode, oi->raw_inode, ei); | |
4889 | spin_unlock(&ei->i_raw_lock); | |
4890 | trace_ext4_other_inode_update_time(inode, oi->orig_ino); | |
4891 | return -1; | |
4892 | } | |
4893 | spin_unlock(&inode->i_lock); | |
4894 | return -1; | |
4895 | } | |
4896 | ||
4897 | /* | |
4898 | * Opportunistically update the other time fields for other inodes in | |
4899 | * the same inode table block. | |
4900 | */ | |
4901 | static void ext4_update_other_inodes_time(struct super_block *sb, | |
4902 | unsigned long orig_ino, char *buf) | |
4903 | { | |
4904 | struct other_inode oi; | |
4905 | unsigned long ino; | |
4906 | int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; | |
4907 | int inode_size = EXT4_INODE_SIZE(sb); | |
4908 | ||
4909 | oi.orig_ino = orig_ino; | |
0f0ff9a9 TT |
4910 | /* |
4911 | * Calculate the first inode in the inode table block. Inode | |
4912 | * numbers are one-based. That is, the first inode in a block | |
4913 | * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1). | |
4914 | */ | |
4915 | ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1; | |
a26f4992 TT |
4916 | for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) { |
4917 | if (ino == orig_ino) | |
4918 | continue; | |
4919 | oi.raw_inode = (struct ext4_inode *) buf; | |
4920 | (void) find_inode_nowait(sb, ino, other_inode_match, &oi); | |
4921 | } | |
4922 | } | |
4923 | ||
ac27a0ec DK |
4924 | /* |
4925 | * Post the struct inode info into an on-disk inode location in the | |
4926 | * buffer-cache. This gobbles the caller's reference to the | |
4927 | * buffer_head in the inode location struct. | |
4928 | * | |
4929 | * The caller must have write access to iloc->bh. | |
4930 | */ | |
617ba13b | 4931 | static int ext4_do_update_inode(handle_t *handle, |
ac27a0ec | 4932 | struct inode *inode, |
830156c7 | 4933 | struct ext4_iloc *iloc) |
ac27a0ec | 4934 | { |
617ba13b MC |
4935 | struct ext4_inode *raw_inode = ext4_raw_inode(iloc); |
4936 | struct ext4_inode_info *ei = EXT4_I(inode); | |
ac27a0ec | 4937 | struct buffer_head *bh = iloc->bh; |
202ee5df | 4938 | struct super_block *sb = inode->i_sb; |
ac27a0ec | 4939 | int err = 0, rc, block; |
202ee5df | 4940 | int need_datasync = 0, set_large_file = 0; |
08cefc7a EB |
4941 | uid_t i_uid; |
4942 | gid_t i_gid; | |
040cb378 | 4943 | projid_t i_projid; |
ac27a0ec | 4944 | |
202ee5df TT |
4945 | spin_lock(&ei->i_raw_lock); |
4946 | ||
4947 | /* For fields not tracked in the in-memory inode, | |
ac27a0ec | 4948 | * initialise them to zero for new inodes. */ |
19f5fb7a | 4949 | if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) |
617ba13b | 4950 | memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size); |
ac27a0ec DK |
4951 | |
4952 | raw_inode->i_mode = cpu_to_le16(inode->i_mode); | |
08cefc7a EB |
4953 | i_uid = i_uid_read(inode); |
4954 | i_gid = i_gid_read(inode); | |
040cb378 | 4955 | i_projid = from_kprojid(&init_user_ns, ei->i_projid); |
af5bc92d | 4956 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
08cefc7a EB |
4957 | raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid)); |
4958 | raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid)); | |
ac27a0ec DK |
4959 | /* |
4960 | * Fix up interoperability with old kernels. Otherwise, old inodes get | |
4961 | * re-used with the upper 16 bits of the uid/gid intact | |
4962 | */ | |
93e3b4e6 DJ |
4963 | if (ei->i_dtime && list_empty(&ei->i_orphan)) { |
4964 | raw_inode->i_uid_high = 0; | |
4965 | raw_inode->i_gid_high = 0; | |
4966 | } else { | |
ac27a0ec | 4967 | raw_inode->i_uid_high = |
08cefc7a | 4968 | cpu_to_le16(high_16_bits(i_uid)); |
ac27a0ec | 4969 | raw_inode->i_gid_high = |
08cefc7a | 4970 | cpu_to_le16(high_16_bits(i_gid)); |
ac27a0ec DK |
4971 | } |
4972 | } else { | |
08cefc7a EB |
4973 | raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid)); |
4974 | raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid)); | |
ac27a0ec DK |
4975 | raw_inode->i_uid_high = 0; |
4976 | raw_inode->i_gid_high = 0; | |
4977 | } | |
4978 | raw_inode->i_links_count = cpu_to_le16(inode->i_nlink); | |
ef7f3835 KS |
4979 | |
4980 | EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode); | |
4981 | EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode); | |
4982 | EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode); | |
4983 | EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode); | |
4984 | ||
bce92d56 LX |
4985 | err = ext4_inode_blocks_set(handle, raw_inode, ei); |
4986 | if (err) { | |
202ee5df | 4987 | spin_unlock(&ei->i_raw_lock); |
0fc1b451 | 4988 | goto out_brelse; |
202ee5df | 4989 | } |
ac27a0ec | 4990 | raw_inode->i_dtime = cpu_to_le32(ei->i_dtime); |
353eb83c | 4991 | raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF); |
ed3654eb | 4992 | if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) |
a1ddeb7e BP |
4993 | raw_inode->i_file_acl_high = |
4994 | cpu_to_le16(ei->i_file_acl >> 32); | |
7973c0c1 | 4995 | raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl); |
dce8e237 | 4996 | if (READ_ONCE(ei->i_disksize) != ext4_isize(inode->i_sb, raw_inode)) { |
b71fc079 JK |
4997 | ext4_isize_set(raw_inode, ei->i_disksize); |
4998 | need_datasync = 1; | |
4999 | } | |
a48380f7 | 5000 | if (ei->i_disksize > 0x7fffffffULL) { |
e2b911c5 | 5001 | if (!ext4_has_feature_large_file(sb) || |
a48380f7 | 5002 | EXT4_SB(sb)->s_es->s_rev_level == |
202ee5df TT |
5003 | cpu_to_le32(EXT4_GOOD_OLD_REV)) |
5004 | set_large_file = 1; | |
ac27a0ec DK |
5005 | } |
5006 | raw_inode->i_generation = cpu_to_le32(inode->i_generation); | |
5007 | if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) { | |
5008 | if (old_valid_dev(inode->i_rdev)) { | |
5009 | raw_inode->i_block[0] = | |
5010 | cpu_to_le32(old_encode_dev(inode->i_rdev)); | |
5011 | raw_inode->i_block[1] = 0; | |
5012 | } else { | |
5013 | raw_inode->i_block[0] = 0; | |
5014 | raw_inode->i_block[1] = | |
5015 | cpu_to_le32(new_encode_dev(inode->i_rdev)); | |
5016 | raw_inode->i_block[2] = 0; | |
5017 | } | |
f19d5870 | 5018 | } else if (!ext4_has_inline_data(inode)) { |
de9a55b8 TT |
5019 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
5020 | raw_inode->i_block[block] = ei->i_data[block]; | |
f19d5870 | 5021 | } |
ac27a0ec | 5022 | |
ed3654eb | 5023 | if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) { |
e254d1af | 5024 | u64 ivers = ext4_inode_peek_iversion(inode); |
ee73f9a5 JL |
5025 | |
5026 | raw_inode->i_disk_version = cpu_to_le32(ivers); | |
c4f65706 TT |
5027 | if (ei->i_extra_isize) { |
5028 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
5029 | raw_inode->i_version_hi = | |
ee73f9a5 | 5030 | cpu_to_le32(ivers >> 32); |
c4f65706 TT |
5031 | raw_inode->i_extra_isize = |
5032 | cpu_to_le16(ei->i_extra_isize); | |
5033 | } | |
25ec56b5 | 5034 | } |
040cb378 | 5035 | |
0b7b7779 | 5036 | BUG_ON(!ext4_has_feature_project(inode->i_sb) && |
040cb378 LX |
5037 | i_projid != EXT4_DEF_PROJID); |
5038 | ||
5039 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && | |
5040 | EXT4_FITS_IN_INODE(raw_inode, ei, i_projid)) | |
5041 | raw_inode->i_projid = cpu_to_le32(i_projid); | |
5042 | ||
814525f4 | 5043 | ext4_inode_csum_set(inode, raw_inode, ei); |
202ee5df | 5044 | spin_unlock(&ei->i_raw_lock); |
1751e8a6 | 5045 | if (inode->i_sb->s_flags & SB_LAZYTIME) |
a26f4992 TT |
5046 | ext4_update_other_inodes_time(inode->i_sb, inode->i_ino, |
5047 | bh->b_data); | |
202ee5df | 5048 | |
830156c7 | 5049 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
73b50c1c | 5050 | rc = ext4_handle_dirty_metadata(handle, NULL, bh); |
830156c7 FM |
5051 | if (!err) |
5052 | err = rc; | |
19f5fb7a | 5053 | ext4_clear_inode_state(inode, EXT4_STATE_NEW); |
202ee5df | 5054 | if (set_large_file) { |
5d601255 | 5055 | BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access"); |
202ee5df TT |
5056 | err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh); |
5057 | if (err) | |
5058 | goto out_brelse; | |
e2b911c5 | 5059 | ext4_set_feature_large_file(sb); |
202ee5df TT |
5060 | ext4_handle_sync(handle); |
5061 | err = ext4_handle_dirty_super(handle, sb); | |
5062 | } | |
b71fc079 | 5063 | ext4_update_inode_fsync_trans(handle, inode, need_datasync); |
ac27a0ec | 5064 | out_brelse: |
af5bc92d | 5065 | brelse(bh); |
617ba13b | 5066 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
5067 | return err; |
5068 | } | |
5069 | ||
5070 | /* | |
617ba13b | 5071 | * ext4_write_inode() |
ac27a0ec DK |
5072 | * |
5073 | * We are called from a few places: | |
5074 | * | |
87f7e416 | 5075 | * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files. |
ac27a0ec | 5076 | * Here, there will be no transaction running. We wait for any running |
4907cb7b | 5077 | * transaction to commit. |
ac27a0ec | 5078 | * |
87f7e416 TT |
5079 | * - Within flush work (sys_sync(), kupdate and such). |
5080 | * We wait on commit, if told to. | |
ac27a0ec | 5081 | * |
87f7e416 TT |
5082 | * - Within iput_final() -> write_inode_now() |
5083 | * We wait on commit, if told to. | |
ac27a0ec DK |
5084 | * |
5085 | * In all cases it is actually safe for us to return without doing anything, | |
5086 | * because the inode has been copied into a raw inode buffer in | |
87f7e416 TT |
5087 | * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL |
5088 | * writeback. | |
ac27a0ec DK |
5089 | * |
5090 | * Note that we are absolutely dependent upon all inode dirtiers doing the | |
5091 | * right thing: they *must* call mark_inode_dirty() after dirtying info in | |
5092 | * which we are interested. | |
5093 | * | |
5094 | * It would be a bug for them to not do this. The code: | |
5095 | * | |
5096 | * mark_inode_dirty(inode) | |
5097 | * stuff(); | |
5098 | * inode->i_size = expr; | |
5099 | * | |
87f7e416 TT |
5100 | * is in error because write_inode() could occur while `stuff()' is running, |
5101 | * and the new i_size will be lost. Plus the inode will no longer be on the | |
5102 | * superblock's dirty inode list. | |
ac27a0ec | 5103 | */ |
a9185b41 | 5104 | int ext4_write_inode(struct inode *inode, struct writeback_control *wbc) |
ac27a0ec | 5105 | { |
91ac6f43 FM |
5106 | int err; |
5107 | ||
18f2c4fc TT |
5108 | if (WARN_ON_ONCE(current->flags & PF_MEMALLOC) || |
5109 | sb_rdonly(inode->i_sb)) | |
ac27a0ec DK |
5110 | return 0; |
5111 | ||
18f2c4fc TT |
5112 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
5113 | return -EIO; | |
5114 | ||
91ac6f43 FM |
5115 | if (EXT4_SB(inode->i_sb)->s_journal) { |
5116 | if (ext4_journal_current_handle()) { | |
5117 | jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n"); | |
5118 | dump_stack(); | |
5119 | return -EIO; | |
5120 | } | |
ac27a0ec | 5121 | |
10542c22 JK |
5122 | /* |
5123 | * No need to force transaction in WB_SYNC_NONE mode. Also | |
5124 | * ext4_sync_fs() will force the commit after everything is | |
5125 | * written. | |
5126 | */ | |
5127 | if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync) | |
91ac6f43 FM |
5128 | return 0; |
5129 | ||
18f2c4fc TT |
5130 | err = jbd2_complete_transaction(EXT4_SB(inode->i_sb)->s_journal, |
5131 | EXT4_I(inode)->i_sync_tid); | |
91ac6f43 FM |
5132 | } else { |
5133 | struct ext4_iloc iloc; | |
ac27a0ec | 5134 | |
8b472d73 | 5135 | err = __ext4_get_inode_loc(inode, &iloc, 0); |
91ac6f43 FM |
5136 | if (err) |
5137 | return err; | |
10542c22 JK |
5138 | /* |
5139 | * sync(2) will flush the whole buffer cache. No need to do | |
5140 | * it here separately for each inode. | |
5141 | */ | |
5142 | if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) | |
830156c7 FM |
5143 | sync_dirty_buffer(iloc.bh); |
5144 | if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) { | |
54d3adbc TT |
5145 | ext4_error_inode_block(inode, iloc.bh->b_blocknr, EIO, |
5146 | "IO error syncing inode"); | |
830156c7 FM |
5147 | err = -EIO; |
5148 | } | |
fd2dd9fb | 5149 | brelse(iloc.bh); |
91ac6f43 FM |
5150 | } |
5151 | return err; | |
ac27a0ec DK |
5152 | } |
5153 | ||
53e87268 JK |
5154 | /* |
5155 | * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate | |
5156 | * buffers that are attached to a page stradding i_size and are undergoing | |
5157 | * commit. In that case we have to wait for commit to finish and try again. | |
5158 | */ | |
5159 | static void ext4_wait_for_tail_page_commit(struct inode *inode) | |
5160 | { | |
5161 | struct page *page; | |
5162 | unsigned offset; | |
5163 | journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; | |
5164 | tid_t commit_tid = 0; | |
5165 | int ret; | |
5166 | ||
09cbfeaf | 5167 | offset = inode->i_size & (PAGE_SIZE - 1); |
53e87268 | 5168 | /* |
565333a1 | 5169 | * If the page is fully truncated, we don't need to wait for any commit |
5170 | * (and we even should not as __ext4_journalled_invalidatepage() may | |
5171 | * strip all buffers from the page but keep the page dirty which can then | |
5172 | * confuse e.g. concurrent ext4_writepage() seeing dirty page without | |
5173 | * buffers). Also we don't need to wait for any commit if all buffers in | |
5174 | * the page remain valid. This is most beneficial for the common case of | |
5175 | * blocksize == PAGESIZE. | |
53e87268 | 5176 | */ |
565333a1 | 5177 | if (!offset || offset > (PAGE_SIZE - i_blocksize(inode))) |
53e87268 JK |
5178 | return; |
5179 | while (1) { | |
5180 | page = find_lock_page(inode->i_mapping, | |
09cbfeaf | 5181 | inode->i_size >> PAGE_SHIFT); |
53e87268 JK |
5182 | if (!page) |
5183 | return; | |
ca99fdd2 | 5184 | ret = __ext4_journalled_invalidatepage(page, offset, |
09cbfeaf | 5185 | PAGE_SIZE - offset); |
53e87268 | 5186 | unlock_page(page); |
09cbfeaf | 5187 | put_page(page); |
53e87268 JK |
5188 | if (ret != -EBUSY) |
5189 | return; | |
5190 | commit_tid = 0; | |
5191 | read_lock(&journal->j_state_lock); | |
5192 | if (journal->j_committing_transaction) | |
5193 | commit_tid = journal->j_committing_transaction->t_tid; | |
5194 | read_unlock(&journal->j_state_lock); | |
5195 | if (commit_tid) | |
5196 | jbd2_log_wait_commit(journal, commit_tid); | |
5197 | } | |
5198 | } | |
5199 | ||
ac27a0ec | 5200 | /* |
617ba13b | 5201 | * ext4_setattr() |
ac27a0ec DK |
5202 | * |
5203 | * Called from notify_change. | |
5204 | * | |
5205 | * We want to trap VFS attempts to truncate the file as soon as | |
5206 | * possible. In particular, we want to make sure that when the VFS | |
5207 | * shrinks i_size, we put the inode on the orphan list and modify | |
5208 | * i_disksize immediately, so that during the subsequent flushing of | |
5209 | * dirty pages and freeing of disk blocks, we can guarantee that any | |
5210 | * commit will leave the blocks being flushed in an unused state on | |
5211 | * disk. (On recovery, the inode will get truncated and the blocks will | |
5212 | * be freed, so we have a strong guarantee that no future commit will | |
5213 | * leave these blocks visible to the user.) | |
5214 | * | |
678aaf48 JK |
5215 | * Another thing we have to assure is that if we are in ordered mode |
5216 | * and inode is still attached to the committing transaction, we must | |
5217 | * we start writeout of all the dirty pages which are being truncated. | |
5218 | * This way we are sure that all the data written in the previous | |
5219 | * transaction are already on disk (truncate waits for pages under | |
5220 | * writeback). | |
5221 | * | |
5222 | * Called with inode->i_mutex down. | |
ac27a0ec | 5223 | */ |
617ba13b | 5224 | int ext4_setattr(struct dentry *dentry, struct iattr *attr) |
ac27a0ec | 5225 | { |
2b0143b5 | 5226 | struct inode *inode = d_inode(dentry); |
ac27a0ec | 5227 | int error, rc = 0; |
3d287de3 | 5228 | int orphan = 0; |
ac27a0ec DK |
5229 | const unsigned int ia_valid = attr->ia_valid; |
5230 | ||
0db1ff22 TT |
5231 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
5232 | return -EIO; | |
5233 | ||
02b016ca TT |
5234 | if (unlikely(IS_IMMUTABLE(inode))) |
5235 | return -EPERM; | |
5236 | ||
5237 | if (unlikely(IS_APPEND(inode) && | |
5238 | (ia_valid & (ATTR_MODE | ATTR_UID | | |
5239 | ATTR_GID | ATTR_TIMES_SET)))) | |
5240 | return -EPERM; | |
5241 | ||
31051c85 | 5242 | error = setattr_prepare(dentry, attr); |
ac27a0ec DK |
5243 | if (error) |
5244 | return error; | |
5245 | ||
3ce2b8dd EB |
5246 | error = fscrypt_prepare_setattr(dentry, attr); |
5247 | if (error) | |
5248 | return error; | |
5249 | ||
c93d8f88 EB |
5250 | error = fsverity_prepare_setattr(dentry, attr); |
5251 | if (error) | |
5252 | return error; | |
5253 | ||
a7cdadee JK |
5254 | if (is_quota_modification(inode, attr)) { |
5255 | error = dquot_initialize(inode); | |
5256 | if (error) | |
5257 | return error; | |
5258 | } | |
08cefc7a EB |
5259 | if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) || |
5260 | (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) { | |
ac27a0ec DK |
5261 | handle_t *handle; |
5262 | ||
5263 | /* (user+group)*(old+new) structure, inode write (sb, | |
5264 | * inode block, ? - but truncate inode update has it) */ | |
9924a92a TT |
5265 | handle = ext4_journal_start(inode, EXT4_HT_QUOTA, |
5266 | (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) + | |
5267 | EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3); | |
ac27a0ec DK |
5268 | if (IS_ERR(handle)) { |
5269 | error = PTR_ERR(handle); | |
5270 | goto err_out; | |
5271 | } | |
7a9ca53a TE |
5272 | |
5273 | /* dquot_transfer() calls back ext4_get_inode_usage() which | |
5274 | * counts xattr inode references. | |
5275 | */ | |
5276 | down_read(&EXT4_I(inode)->xattr_sem); | |
b43fa828 | 5277 | error = dquot_transfer(inode, attr); |
7a9ca53a TE |
5278 | up_read(&EXT4_I(inode)->xattr_sem); |
5279 | ||
ac27a0ec | 5280 | if (error) { |
617ba13b | 5281 | ext4_journal_stop(handle); |
ac27a0ec DK |
5282 | return error; |
5283 | } | |
5284 | /* Update corresponding info in inode so that everything is in | |
5285 | * one transaction */ | |
5286 | if (attr->ia_valid & ATTR_UID) | |
5287 | inode->i_uid = attr->ia_uid; | |
5288 | if (attr->ia_valid & ATTR_GID) | |
5289 | inode->i_gid = attr->ia_gid; | |
617ba13b MC |
5290 | error = ext4_mark_inode_dirty(handle, inode); |
5291 | ext4_journal_stop(handle); | |
ac27a0ec DK |
5292 | } |
5293 | ||
3da40c7b | 5294 | if (attr->ia_valid & ATTR_SIZE) { |
5208386c | 5295 | handle_t *handle; |
3da40c7b | 5296 | loff_t oldsize = inode->i_size; |
b9c1c267 | 5297 | int shrink = (attr->ia_size < inode->i_size); |
562c72aa | 5298 | |
12e9b892 | 5299 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) { |
e2b46574 ES |
5300 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
5301 | ||
0c095c7f TT |
5302 | if (attr->ia_size > sbi->s_bitmap_maxbytes) |
5303 | return -EFBIG; | |
e2b46574 | 5304 | } |
3da40c7b JB |
5305 | if (!S_ISREG(inode->i_mode)) |
5306 | return -EINVAL; | |
dff6efc3 CH |
5307 | |
5308 | if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size) | |
5309 | inode_inc_iversion(inode); | |
5310 | ||
b9c1c267 JK |
5311 | if (shrink) { |
5312 | if (ext4_should_order_data(inode)) { | |
5313 | error = ext4_begin_ordered_truncate(inode, | |
678aaf48 | 5314 | attr->ia_size); |
b9c1c267 JK |
5315 | if (error) |
5316 | goto err_out; | |
5317 | } | |
5318 | /* | |
5319 | * Blocks are going to be removed from the inode. Wait | |
5320 | * for dio in flight. | |
5321 | */ | |
5322 | inode_dio_wait(inode); | |
5323 | } | |
5324 | ||
5325 | down_write(&EXT4_I(inode)->i_mmap_sem); | |
5326 | ||
5327 | rc = ext4_break_layouts(inode); | |
5328 | if (rc) { | |
5329 | up_write(&EXT4_I(inode)->i_mmap_sem); | |
5330 | return rc; | |
3da40c7b | 5331 | } |
b9c1c267 | 5332 | |
3da40c7b | 5333 | if (attr->ia_size != inode->i_size) { |
5208386c JK |
5334 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 3); |
5335 | if (IS_ERR(handle)) { | |
5336 | error = PTR_ERR(handle); | |
b9c1c267 | 5337 | goto out_mmap_sem; |
5208386c | 5338 | } |
3da40c7b | 5339 | if (ext4_handle_valid(handle) && shrink) { |
5208386c JK |
5340 | error = ext4_orphan_add(handle, inode); |
5341 | orphan = 1; | |
5342 | } | |
911af577 EG |
5343 | /* |
5344 | * Update c/mtime on truncate up, ext4_truncate() will | |
5345 | * update c/mtime in shrink case below | |
5346 | */ | |
5347 | if (!shrink) { | |
eeca7ea1 | 5348 | inode->i_mtime = current_time(inode); |
911af577 EG |
5349 | inode->i_ctime = inode->i_mtime; |
5350 | } | |
90e775b7 | 5351 | down_write(&EXT4_I(inode)->i_data_sem); |
5208386c JK |
5352 | EXT4_I(inode)->i_disksize = attr->ia_size; |
5353 | rc = ext4_mark_inode_dirty(handle, inode); | |
5354 | if (!error) | |
5355 | error = rc; | |
90e775b7 JK |
5356 | /* |
5357 | * We have to update i_size under i_data_sem together | |
5358 | * with i_disksize to avoid races with writeback code | |
5359 | * running ext4_wb_update_i_disksize(). | |
5360 | */ | |
5361 | if (!error) | |
5362 | i_size_write(inode, attr->ia_size); | |
5363 | up_write(&EXT4_I(inode)->i_data_sem); | |
5208386c | 5364 | ext4_journal_stop(handle); |
b9c1c267 JK |
5365 | if (error) |
5366 | goto out_mmap_sem; | |
5367 | if (!shrink) { | |
5368 | pagecache_isize_extended(inode, oldsize, | |
5369 | inode->i_size); | |
5370 | } else if (ext4_should_journal_data(inode)) { | |
5371 | ext4_wait_for_tail_page_commit(inode); | |
678aaf48 | 5372 | } |
d6320cbf | 5373 | } |
430657b6 | 5374 | |
5208386c JK |
5375 | /* |
5376 | * Truncate pagecache after we've waited for commit | |
5377 | * in data=journal mode to make pages freeable. | |
5378 | */ | |
923ae0ff | 5379 | truncate_pagecache(inode, inode->i_size); |
b9c1c267 JK |
5380 | /* |
5381 | * Call ext4_truncate() even if i_size didn't change to | |
5382 | * truncate possible preallocated blocks. | |
5383 | */ | |
5384 | if (attr->ia_size <= oldsize) { | |
2c98eb5e TT |
5385 | rc = ext4_truncate(inode); |
5386 | if (rc) | |
5387 | error = rc; | |
5388 | } | |
b9c1c267 | 5389 | out_mmap_sem: |
ea3d7209 | 5390 | up_write(&EXT4_I(inode)->i_mmap_sem); |
072bd7ea | 5391 | } |
ac27a0ec | 5392 | |
2c98eb5e | 5393 | if (!error) { |
1025774c CH |
5394 | setattr_copy(inode, attr); |
5395 | mark_inode_dirty(inode); | |
5396 | } | |
5397 | ||
5398 | /* | |
5399 | * If the call to ext4_truncate failed to get a transaction handle at | |
5400 | * all, we need to clean up the in-core orphan list manually. | |
5401 | */ | |
3d287de3 | 5402 | if (orphan && inode->i_nlink) |
617ba13b | 5403 | ext4_orphan_del(NULL, inode); |
ac27a0ec | 5404 | |
2c98eb5e | 5405 | if (!error && (ia_valid & ATTR_MODE)) |
64e178a7 | 5406 | rc = posix_acl_chmod(inode, inode->i_mode); |
ac27a0ec DK |
5407 | |
5408 | err_out: | |
617ba13b | 5409 | ext4_std_error(inode->i_sb, error); |
ac27a0ec DK |
5410 | if (!error) |
5411 | error = rc; | |
5412 | return error; | |
5413 | } | |
5414 | ||
a528d35e DH |
5415 | int ext4_getattr(const struct path *path, struct kstat *stat, |
5416 | u32 request_mask, unsigned int query_flags) | |
3e3398a0 | 5417 | { |
99652ea5 DH |
5418 | struct inode *inode = d_inode(path->dentry); |
5419 | struct ext4_inode *raw_inode; | |
5420 | struct ext4_inode_info *ei = EXT4_I(inode); | |
5421 | unsigned int flags; | |
5422 | ||
d4c5e960 TT |
5423 | if ((request_mask & STATX_BTIME) && |
5424 | EXT4_FITS_IN_INODE(raw_inode, ei, i_crtime)) { | |
99652ea5 DH |
5425 | stat->result_mask |= STATX_BTIME; |
5426 | stat->btime.tv_sec = ei->i_crtime.tv_sec; | |
5427 | stat->btime.tv_nsec = ei->i_crtime.tv_nsec; | |
5428 | } | |
5429 | ||
5430 | flags = ei->i_flags & EXT4_FL_USER_VISIBLE; | |
5431 | if (flags & EXT4_APPEND_FL) | |
5432 | stat->attributes |= STATX_ATTR_APPEND; | |
5433 | if (flags & EXT4_COMPR_FL) | |
5434 | stat->attributes |= STATX_ATTR_COMPRESSED; | |
5435 | if (flags & EXT4_ENCRYPT_FL) | |
5436 | stat->attributes |= STATX_ATTR_ENCRYPTED; | |
5437 | if (flags & EXT4_IMMUTABLE_FL) | |
5438 | stat->attributes |= STATX_ATTR_IMMUTABLE; | |
5439 | if (flags & EXT4_NODUMP_FL) | |
5440 | stat->attributes |= STATX_ATTR_NODUMP; | |
1f607195 EB |
5441 | if (flags & EXT4_VERITY_FL) |
5442 | stat->attributes |= STATX_ATTR_VERITY; | |
3e3398a0 | 5443 | |
3209f68b DH |
5444 | stat->attributes_mask |= (STATX_ATTR_APPEND | |
5445 | STATX_ATTR_COMPRESSED | | |
5446 | STATX_ATTR_ENCRYPTED | | |
5447 | STATX_ATTR_IMMUTABLE | | |
1f607195 EB |
5448 | STATX_ATTR_NODUMP | |
5449 | STATX_ATTR_VERITY); | |
3209f68b | 5450 | |
3e3398a0 | 5451 | generic_fillattr(inode, stat); |
99652ea5 DH |
5452 | return 0; |
5453 | } | |
5454 | ||
5455 | int ext4_file_getattr(const struct path *path, struct kstat *stat, | |
5456 | u32 request_mask, unsigned int query_flags) | |
5457 | { | |
5458 | struct inode *inode = d_inode(path->dentry); | |
5459 | u64 delalloc_blocks; | |
5460 | ||
5461 | ext4_getattr(path, stat, request_mask, query_flags); | |
3e3398a0 | 5462 | |
9206c561 AD |
5463 | /* |
5464 | * If there is inline data in the inode, the inode will normally not | |
5465 | * have data blocks allocated (it may have an external xattr block). | |
5466 | * Report at least one sector for such files, so tools like tar, rsync, | |
d67d64f4 | 5467 | * others don't incorrectly think the file is completely sparse. |
9206c561 AD |
5468 | */ |
5469 | if (unlikely(ext4_has_inline_data(inode))) | |
5470 | stat->blocks += (stat->size + 511) >> 9; | |
5471 | ||
3e3398a0 MC |
5472 | /* |
5473 | * We can't update i_blocks if the block allocation is delayed | |
5474 | * otherwise in the case of system crash before the real block | |
5475 | * allocation is done, we will have i_blocks inconsistent with | |
5476 | * on-disk file blocks. | |
5477 | * We always keep i_blocks updated together with real | |
5478 | * allocation. But to not confuse with user, stat | |
5479 | * will return the blocks that include the delayed allocation | |
5480 | * blocks for this file. | |
5481 | */ | |
96607551 | 5482 | delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb), |
9206c561 AD |
5483 | EXT4_I(inode)->i_reserved_data_blocks); |
5484 | stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9); | |
3e3398a0 MC |
5485 | return 0; |
5486 | } | |
ac27a0ec | 5487 | |
fffb2739 JK |
5488 | static int ext4_index_trans_blocks(struct inode *inode, int lblocks, |
5489 | int pextents) | |
a02908f1 | 5490 | { |
12e9b892 | 5491 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) |
fffb2739 JK |
5492 | return ext4_ind_trans_blocks(inode, lblocks); |
5493 | return ext4_ext_index_trans_blocks(inode, pextents); | |
a02908f1 | 5494 | } |
ac51d837 | 5495 | |
ac27a0ec | 5496 | /* |
a02908f1 MC |
5497 | * Account for index blocks, block groups bitmaps and block group |
5498 | * descriptor blocks if modify datablocks and index blocks | |
5499 | * worse case, the indexs blocks spread over different block groups | |
ac27a0ec | 5500 | * |
a02908f1 | 5501 | * If datablocks are discontiguous, they are possible to spread over |
4907cb7b | 5502 | * different block groups too. If they are contiguous, with flexbg, |
a02908f1 | 5503 | * they could still across block group boundary. |
ac27a0ec | 5504 | * |
a02908f1 MC |
5505 | * Also account for superblock, inode, quota and xattr blocks |
5506 | */ | |
dec214d0 | 5507 | static int ext4_meta_trans_blocks(struct inode *inode, int lblocks, |
fffb2739 | 5508 | int pextents) |
a02908f1 | 5509 | { |
8df9675f TT |
5510 | ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb); |
5511 | int gdpblocks; | |
a02908f1 MC |
5512 | int idxblocks; |
5513 | int ret = 0; | |
5514 | ||
5515 | /* | |
fffb2739 JK |
5516 | * How many index blocks need to touch to map @lblocks logical blocks |
5517 | * to @pextents physical extents? | |
a02908f1 | 5518 | */ |
fffb2739 | 5519 | idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents); |
a02908f1 MC |
5520 | |
5521 | ret = idxblocks; | |
5522 | ||
5523 | /* | |
5524 | * Now let's see how many group bitmaps and group descriptors need | |
5525 | * to account | |
5526 | */ | |
fffb2739 | 5527 | groups = idxblocks + pextents; |
a02908f1 | 5528 | gdpblocks = groups; |
8df9675f TT |
5529 | if (groups > ngroups) |
5530 | groups = ngroups; | |
a02908f1 MC |
5531 | if (groups > EXT4_SB(inode->i_sb)->s_gdb_count) |
5532 | gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count; | |
5533 | ||
5534 | /* bitmaps and block group descriptor blocks */ | |
5535 | ret += groups + gdpblocks; | |
5536 | ||
5537 | /* Blocks for super block, inode, quota and xattr blocks */ | |
5538 | ret += EXT4_META_TRANS_BLOCKS(inode->i_sb); | |
5539 | ||
5540 | return ret; | |
5541 | } | |
5542 | ||
5543 | /* | |
25985edc | 5544 | * Calculate the total number of credits to reserve to fit |
f3bd1f3f MC |
5545 | * the modification of a single pages into a single transaction, |
5546 | * which may include multiple chunks of block allocations. | |
ac27a0ec | 5547 | * |
525f4ed8 | 5548 | * This could be called via ext4_write_begin() |
ac27a0ec | 5549 | * |
525f4ed8 | 5550 | * We need to consider the worse case, when |
a02908f1 | 5551 | * one new block per extent. |
ac27a0ec | 5552 | */ |
a86c6181 | 5553 | int ext4_writepage_trans_blocks(struct inode *inode) |
ac27a0ec | 5554 | { |
617ba13b | 5555 | int bpp = ext4_journal_blocks_per_page(inode); |
ac27a0ec DK |
5556 | int ret; |
5557 | ||
fffb2739 | 5558 | ret = ext4_meta_trans_blocks(inode, bpp, bpp); |
a86c6181 | 5559 | |
a02908f1 | 5560 | /* Account for data blocks for journalled mode */ |
617ba13b | 5561 | if (ext4_should_journal_data(inode)) |
a02908f1 | 5562 | ret += bpp; |
ac27a0ec DK |
5563 | return ret; |
5564 | } | |
f3bd1f3f MC |
5565 | |
5566 | /* | |
5567 | * Calculate the journal credits for a chunk of data modification. | |
5568 | * | |
5569 | * This is called from DIO, fallocate or whoever calling | |
79e83036 | 5570 | * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks. |
f3bd1f3f MC |
5571 | * |
5572 | * journal buffers for data blocks are not included here, as DIO | |
5573 | * and fallocate do no need to journal data buffers. | |
5574 | */ | |
5575 | int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks) | |
5576 | { | |
5577 | return ext4_meta_trans_blocks(inode, nrblocks, 1); | |
5578 | } | |
5579 | ||
ac27a0ec | 5580 | /* |
617ba13b | 5581 | * The caller must have previously called ext4_reserve_inode_write(). |
ac27a0ec DK |
5582 | * Give this, we know that the caller already has write access to iloc->bh. |
5583 | */ | |
617ba13b | 5584 | int ext4_mark_iloc_dirty(handle_t *handle, |
de9a55b8 | 5585 | struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec DK |
5586 | { |
5587 | int err = 0; | |
5588 | ||
a6758309 VA |
5589 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) { |
5590 | put_bh(iloc->bh); | |
0db1ff22 | 5591 | return -EIO; |
a6758309 | 5592 | } |
c64db50e | 5593 | if (IS_I_VERSION(inode)) |
25ec56b5 JNC |
5594 | inode_inc_iversion(inode); |
5595 | ||
ac27a0ec DK |
5596 | /* the do_update_inode consumes one bh->b_count */ |
5597 | get_bh(iloc->bh); | |
5598 | ||
dab291af | 5599 | /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */ |
830156c7 | 5600 | err = ext4_do_update_inode(handle, inode, iloc); |
ac27a0ec DK |
5601 | put_bh(iloc->bh); |
5602 | return err; | |
5603 | } | |
5604 | ||
5605 | /* | |
5606 | * On success, We end up with an outstanding reference count against | |
5607 | * iloc->bh. This _must_ be cleaned up later. | |
5608 | */ | |
5609 | ||
5610 | int | |
617ba13b MC |
5611 | ext4_reserve_inode_write(handle_t *handle, struct inode *inode, |
5612 | struct ext4_iloc *iloc) | |
ac27a0ec | 5613 | { |
0390131b FM |
5614 | int err; |
5615 | ||
0db1ff22 TT |
5616 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
5617 | return -EIO; | |
5618 | ||
0390131b FM |
5619 | err = ext4_get_inode_loc(inode, iloc); |
5620 | if (!err) { | |
5621 | BUFFER_TRACE(iloc->bh, "get_write_access"); | |
5622 | err = ext4_journal_get_write_access(handle, iloc->bh); | |
5623 | if (err) { | |
5624 | brelse(iloc->bh); | |
5625 | iloc->bh = NULL; | |
ac27a0ec DK |
5626 | } |
5627 | } | |
617ba13b | 5628 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
5629 | return err; |
5630 | } | |
5631 | ||
c03b45b8 MX |
5632 | static int __ext4_expand_extra_isize(struct inode *inode, |
5633 | unsigned int new_extra_isize, | |
5634 | struct ext4_iloc *iloc, | |
5635 | handle_t *handle, int *no_expand) | |
5636 | { | |
5637 | struct ext4_inode *raw_inode; | |
5638 | struct ext4_xattr_ibody_header *header; | |
4ea99936 TT |
5639 | unsigned int inode_size = EXT4_INODE_SIZE(inode->i_sb); |
5640 | struct ext4_inode_info *ei = EXT4_I(inode); | |
c03b45b8 MX |
5641 | int error; |
5642 | ||
4ea99936 TT |
5643 | /* this was checked at iget time, but double check for good measure */ |
5644 | if ((EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > inode_size) || | |
5645 | (ei->i_extra_isize & 3)) { | |
5646 | EXT4_ERROR_INODE(inode, "bad extra_isize %u (inode size %u)", | |
5647 | ei->i_extra_isize, | |
5648 | EXT4_INODE_SIZE(inode->i_sb)); | |
5649 | return -EFSCORRUPTED; | |
5650 | } | |
5651 | if ((new_extra_isize < ei->i_extra_isize) || | |
5652 | (new_extra_isize < 4) || | |
5653 | (new_extra_isize > inode_size - EXT4_GOOD_OLD_INODE_SIZE)) | |
5654 | return -EINVAL; /* Should never happen */ | |
5655 | ||
c03b45b8 MX |
5656 | raw_inode = ext4_raw_inode(iloc); |
5657 | ||
5658 | header = IHDR(inode, raw_inode); | |
5659 | ||
5660 | /* No extended attributes present */ | |
5661 | if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) || | |
5662 | header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) { | |
5663 | memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE + | |
5664 | EXT4_I(inode)->i_extra_isize, 0, | |
5665 | new_extra_isize - EXT4_I(inode)->i_extra_isize); | |
5666 | EXT4_I(inode)->i_extra_isize = new_extra_isize; | |
5667 | return 0; | |
5668 | } | |
5669 | ||
5670 | /* try to expand with EAs present */ | |
5671 | error = ext4_expand_extra_isize_ea(inode, new_extra_isize, | |
5672 | raw_inode, handle); | |
5673 | if (error) { | |
5674 | /* | |
5675 | * Inode size expansion failed; don't try again | |
5676 | */ | |
5677 | *no_expand = 1; | |
5678 | } | |
5679 | ||
5680 | return error; | |
5681 | } | |
5682 | ||
6dd4ee7c KS |
5683 | /* |
5684 | * Expand an inode by new_extra_isize bytes. | |
5685 | * Returns 0 on success or negative error number on failure. | |
5686 | */ | |
cf0a5e81 MX |
5687 | static int ext4_try_to_expand_extra_isize(struct inode *inode, |
5688 | unsigned int new_extra_isize, | |
5689 | struct ext4_iloc iloc, | |
5690 | handle_t *handle) | |
6dd4ee7c | 5691 | { |
3b10fdc6 MX |
5692 | int no_expand; |
5693 | int error; | |
6dd4ee7c | 5694 | |
cf0a5e81 MX |
5695 | if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) |
5696 | return -EOVERFLOW; | |
5697 | ||
5698 | /* | |
5699 | * In nojournal mode, we can immediately attempt to expand | |
5700 | * the inode. When journaled, we first need to obtain extra | |
5701 | * buffer credits since we may write into the EA block | |
5702 | * with this same handle. If journal_extend fails, then it will | |
5703 | * only result in a minor loss of functionality for that inode. | |
5704 | * If this is felt to be critical, then e2fsck should be run to | |
5705 | * force a large enough s_min_extra_isize. | |
5706 | */ | |
6cb367c2 | 5707 | if (ext4_journal_extend(handle, |
83448bdf | 5708 | EXT4_DATA_TRANS_BLOCKS(inode->i_sb), 0) != 0) |
cf0a5e81 | 5709 | return -ENOSPC; |
6dd4ee7c | 5710 | |
3b10fdc6 | 5711 | if (ext4_write_trylock_xattr(inode, &no_expand) == 0) |
cf0a5e81 | 5712 | return -EBUSY; |
3b10fdc6 | 5713 | |
c03b45b8 MX |
5714 | error = __ext4_expand_extra_isize(inode, new_extra_isize, &iloc, |
5715 | handle, &no_expand); | |
5716 | ext4_write_unlock_xattr(inode, &no_expand); | |
6dd4ee7c | 5717 | |
c03b45b8 MX |
5718 | return error; |
5719 | } | |
6dd4ee7c | 5720 | |
c03b45b8 MX |
5721 | int ext4_expand_extra_isize(struct inode *inode, |
5722 | unsigned int new_extra_isize, | |
5723 | struct ext4_iloc *iloc) | |
5724 | { | |
5725 | handle_t *handle; | |
5726 | int no_expand; | |
5727 | int error, rc; | |
5728 | ||
5729 | if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) { | |
5730 | brelse(iloc->bh); | |
5731 | return -EOVERFLOW; | |
6dd4ee7c KS |
5732 | } |
5733 | ||
c03b45b8 MX |
5734 | handle = ext4_journal_start(inode, EXT4_HT_INODE, |
5735 | EXT4_DATA_TRANS_BLOCKS(inode->i_sb)); | |
5736 | if (IS_ERR(handle)) { | |
5737 | error = PTR_ERR(handle); | |
5738 | brelse(iloc->bh); | |
5739 | return error; | |
5740 | } | |
5741 | ||
5742 | ext4_write_lock_xattr(inode, &no_expand); | |
5743 | ||
ddccb6db | 5744 | BUFFER_TRACE(iloc->bh, "get_write_access"); |
c03b45b8 | 5745 | error = ext4_journal_get_write_access(handle, iloc->bh); |
3b10fdc6 | 5746 | if (error) { |
c03b45b8 | 5747 | brelse(iloc->bh); |
7f420d64 | 5748 | goto out_unlock; |
3b10fdc6 | 5749 | } |
cf0a5e81 | 5750 | |
c03b45b8 MX |
5751 | error = __ext4_expand_extra_isize(inode, new_extra_isize, iloc, |
5752 | handle, &no_expand); | |
5753 | ||
5754 | rc = ext4_mark_iloc_dirty(handle, inode, iloc); | |
5755 | if (!error) | |
5756 | error = rc; | |
5757 | ||
7f420d64 | 5758 | out_unlock: |
c03b45b8 | 5759 | ext4_write_unlock_xattr(inode, &no_expand); |
c03b45b8 | 5760 | ext4_journal_stop(handle); |
3b10fdc6 | 5761 | return error; |
6dd4ee7c KS |
5762 | } |
5763 | ||
ac27a0ec DK |
5764 | /* |
5765 | * What we do here is to mark the in-core inode as clean with respect to inode | |
5766 | * dirtiness (it may still be data-dirty). | |
5767 | * This means that the in-core inode may be reaped by prune_icache | |
5768 | * without having to perform any I/O. This is a very good thing, | |
5769 | * because *any* task may call prune_icache - even ones which | |
5770 | * have a transaction open against a different journal. | |
5771 | * | |
5772 | * Is this cheating? Not really. Sure, we haven't written the | |
5773 | * inode out, but prune_icache isn't a user-visible syncing function. | |
5774 | * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync) | |
5775 | * we start and wait on commits. | |
ac27a0ec | 5776 | */ |
617ba13b | 5777 | int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode) |
ac27a0ec | 5778 | { |
617ba13b | 5779 | struct ext4_iloc iloc; |
6dd4ee7c | 5780 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
cf0a5e81 | 5781 | int err; |
ac27a0ec DK |
5782 | |
5783 | might_sleep(); | |
7ff9c073 | 5784 | trace_ext4_mark_inode_dirty(inode, _RET_IP_); |
617ba13b | 5785 | err = ext4_reserve_inode_write(handle, inode, &iloc); |
5e1021f2 EG |
5786 | if (err) |
5787 | return err; | |
cf0a5e81 MX |
5788 | |
5789 | if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize) | |
5790 | ext4_try_to_expand_extra_isize(inode, sbi->s_want_extra_isize, | |
5791 | iloc, handle); | |
5792 | ||
5e1021f2 | 5793 | return ext4_mark_iloc_dirty(handle, inode, &iloc); |
ac27a0ec DK |
5794 | } |
5795 | ||
5796 | /* | |
617ba13b | 5797 | * ext4_dirty_inode() is called from __mark_inode_dirty() |
ac27a0ec DK |
5798 | * |
5799 | * We're really interested in the case where a file is being extended. | |
5800 | * i_size has been changed by generic_commit_write() and we thus need | |
5801 | * to include the updated inode in the current transaction. | |
5802 | * | |
5dd4056d | 5803 | * Also, dquot_alloc_block() will always dirty the inode when blocks |
ac27a0ec DK |
5804 | * are allocated to the file. |
5805 | * | |
5806 | * If the inode is marked synchronous, we don't honour that here - doing | |
5807 | * so would cause a commit on atime updates, which we don't bother doing. | |
5808 | * We handle synchronous inodes at the highest possible level. | |
0ae45f63 TT |
5809 | * |
5810 | * If only the I_DIRTY_TIME flag is set, we can skip everything. If | |
5811 | * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need | |
5812 | * to copy into the on-disk inode structure are the timestamp files. | |
ac27a0ec | 5813 | */ |
aa385729 | 5814 | void ext4_dirty_inode(struct inode *inode, int flags) |
ac27a0ec | 5815 | { |
ac27a0ec DK |
5816 | handle_t *handle; |
5817 | ||
0ae45f63 TT |
5818 | if (flags == I_DIRTY_TIME) |
5819 | return; | |
9924a92a | 5820 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 2); |
ac27a0ec DK |
5821 | if (IS_ERR(handle)) |
5822 | goto out; | |
f3dc272f | 5823 | |
f3dc272f CW |
5824 | ext4_mark_inode_dirty(handle, inode); |
5825 | ||
617ba13b | 5826 | ext4_journal_stop(handle); |
ac27a0ec DK |
5827 | out: |
5828 | return; | |
5829 | } | |
5830 | ||
617ba13b | 5831 | int ext4_change_inode_journal_flag(struct inode *inode, int val) |
ac27a0ec DK |
5832 | { |
5833 | journal_t *journal; | |
5834 | handle_t *handle; | |
5835 | int err; | |
c8585c6f | 5836 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
ac27a0ec DK |
5837 | |
5838 | /* | |
5839 | * We have to be very careful here: changing a data block's | |
5840 | * journaling status dynamically is dangerous. If we write a | |
5841 | * data block to the journal, change the status and then delete | |
5842 | * that block, we risk forgetting to revoke the old log record | |
5843 | * from the journal and so a subsequent replay can corrupt data. | |
5844 | * So, first we make sure that the journal is empty and that | |
5845 | * nobody is changing anything. | |
5846 | */ | |
5847 | ||
617ba13b | 5848 | journal = EXT4_JOURNAL(inode); |
0390131b FM |
5849 | if (!journal) |
5850 | return 0; | |
d699594d | 5851 | if (is_journal_aborted(journal)) |
ac27a0ec DK |
5852 | return -EROFS; |
5853 | ||
17335dcc | 5854 | /* Wait for all existing dio workers */ |
17335dcc DM |
5855 | inode_dio_wait(inode); |
5856 | ||
4c546592 DJ |
5857 | /* |
5858 | * Before flushing the journal and switching inode's aops, we have | |
5859 | * to flush all dirty data the inode has. There can be outstanding | |
5860 | * delayed allocations, there can be unwritten extents created by | |
5861 | * fallocate or buffered writes in dioread_nolock mode covered by | |
5862 | * dirty data which can be converted only after flushing the dirty | |
5863 | * data (and journalled aops don't know how to handle these cases). | |
5864 | */ | |
5865 | if (val) { | |
5866 | down_write(&EXT4_I(inode)->i_mmap_sem); | |
5867 | err = filemap_write_and_wait(inode->i_mapping); | |
5868 | if (err < 0) { | |
5869 | up_write(&EXT4_I(inode)->i_mmap_sem); | |
4c546592 DJ |
5870 | return err; |
5871 | } | |
5872 | } | |
5873 | ||
bbd55937 | 5874 | percpu_down_write(&sbi->s_writepages_rwsem); |
dab291af | 5875 | jbd2_journal_lock_updates(journal); |
ac27a0ec DK |
5876 | |
5877 | /* | |
5878 | * OK, there are no updates running now, and all cached data is | |
5879 | * synced to disk. We are now in a completely consistent state | |
5880 | * which doesn't have anything in the journal, and we know that | |
5881 | * no filesystem updates are running, so it is safe to modify | |
5882 | * the inode's in-core data-journaling state flag now. | |
5883 | */ | |
5884 | ||
5885 | if (val) | |
12e9b892 | 5886 | ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
5872ddaa | 5887 | else { |
4f879ca6 JK |
5888 | err = jbd2_journal_flush(journal); |
5889 | if (err < 0) { | |
5890 | jbd2_journal_unlock_updates(journal); | |
bbd55937 | 5891 | percpu_up_write(&sbi->s_writepages_rwsem); |
4f879ca6 JK |
5892 | return err; |
5893 | } | |
12e9b892 | 5894 | ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
5872ddaa | 5895 | } |
617ba13b | 5896 | ext4_set_aops(inode); |
ac27a0ec | 5897 | |
dab291af | 5898 | jbd2_journal_unlock_updates(journal); |
bbd55937 | 5899 | percpu_up_write(&sbi->s_writepages_rwsem); |
c8585c6f | 5900 | |
4c546592 DJ |
5901 | if (val) |
5902 | up_write(&EXT4_I(inode)->i_mmap_sem); | |
ac27a0ec DK |
5903 | |
5904 | /* Finally we can mark the inode as dirty. */ | |
5905 | ||
9924a92a | 5906 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 1); |
ac27a0ec DK |
5907 | if (IS_ERR(handle)) |
5908 | return PTR_ERR(handle); | |
5909 | ||
617ba13b | 5910 | err = ext4_mark_inode_dirty(handle, inode); |
0390131b | 5911 | ext4_handle_sync(handle); |
617ba13b MC |
5912 | ext4_journal_stop(handle); |
5913 | ext4_std_error(inode->i_sb, err); | |
ac27a0ec DK |
5914 | |
5915 | return err; | |
5916 | } | |
2e9ee850 AK |
5917 | |
5918 | static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh) | |
5919 | { | |
5920 | return !buffer_mapped(bh); | |
5921 | } | |
5922 | ||
401b25aa | 5923 | vm_fault_t ext4_page_mkwrite(struct vm_fault *vmf) |
2e9ee850 | 5924 | { |
11bac800 | 5925 | struct vm_area_struct *vma = vmf->vma; |
c2ec175c | 5926 | struct page *page = vmf->page; |
2e9ee850 AK |
5927 | loff_t size; |
5928 | unsigned long len; | |
401b25aa SJ |
5929 | int err; |
5930 | vm_fault_t ret; | |
2e9ee850 | 5931 | struct file *file = vma->vm_file; |
496ad9aa | 5932 | struct inode *inode = file_inode(file); |
2e9ee850 | 5933 | struct address_space *mapping = inode->i_mapping; |
9ea7df53 JK |
5934 | handle_t *handle; |
5935 | get_block_t *get_block; | |
5936 | int retries = 0; | |
2e9ee850 | 5937 | |
02b016ca TT |
5938 | if (unlikely(IS_IMMUTABLE(inode))) |
5939 | return VM_FAULT_SIGBUS; | |
5940 | ||
8e8ad8a5 | 5941 | sb_start_pagefault(inode->i_sb); |
041bbb6d | 5942 | file_update_time(vma->vm_file); |
ea3d7209 JK |
5943 | |
5944 | down_read(&EXT4_I(inode)->i_mmap_sem); | |
7b4cc978 | 5945 | |
401b25aa SJ |
5946 | err = ext4_convert_inline_data(inode); |
5947 | if (err) | |
7b4cc978 EB |
5948 | goto out_ret; |
5949 | ||
9ea7df53 JK |
5950 | /* Delalloc case is easy... */ |
5951 | if (test_opt(inode->i_sb, DELALLOC) && | |
5952 | !ext4_should_journal_data(inode) && | |
5953 | !ext4_nonda_switch(inode->i_sb)) { | |
5954 | do { | |
401b25aa | 5955 | err = block_page_mkwrite(vma, vmf, |
9ea7df53 | 5956 | ext4_da_get_block_prep); |
401b25aa | 5957 | } while (err == -ENOSPC && |
9ea7df53 JK |
5958 | ext4_should_retry_alloc(inode->i_sb, &retries)); |
5959 | goto out_ret; | |
2e9ee850 | 5960 | } |
0e499890 DW |
5961 | |
5962 | lock_page(page); | |
9ea7df53 JK |
5963 | size = i_size_read(inode); |
5964 | /* Page got truncated from under us? */ | |
5965 | if (page->mapping != mapping || page_offset(page) > size) { | |
5966 | unlock_page(page); | |
5967 | ret = VM_FAULT_NOPAGE; | |
5968 | goto out; | |
0e499890 | 5969 | } |
2e9ee850 | 5970 | |
09cbfeaf KS |
5971 | if (page->index == size >> PAGE_SHIFT) |
5972 | len = size & ~PAGE_MASK; | |
2e9ee850 | 5973 | else |
09cbfeaf | 5974 | len = PAGE_SIZE; |
a827eaff | 5975 | /* |
9ea7df53 JK |
5976 | * Return if we have all the buffers mapped. This avoids the need to do |
5977 | * journal_start/journal_stop which can block and take a long time | |
a827eaff | 5978 | */ |
2e9ee850 | 5979 | if (page_has_buffers(page)) { |
f19d5870 TM |
5980 | if (!ext4_walk_page_buffers(NULL, page_buffers(page), |
5981 | 0, len, NULL, | |
5982 | ext4_bh_unmapped)) { | |
9ea7df53 | 5983 | /* Wait so that we don't change page under IO */ |
1d1d1a76 | 5984 | wait_for_stable_page(page); |
9ea7df53 JK |
5985 | ret = VM_FAULT_LOCKED; |
5986 | goto out; | |
a827eaff | 5987 | } |
2e9ee850 | 5988 | } |
a827eaff | 5989 | unlock_page(page); |
9ea7df53 JK |
5990 | /* OK, we need to fill the hole... */ |
5991 | if (ext4_should_dioread_nolock(inode)) | |
705965bd | 5992 | get_block = ext4_get_block_unwritten; |
9ea7df53 JK |
5993 | else |
5994 | get_block = ext4_get_block; | |
5995 | retry_alloc: | |
9924a92a TT |
5996 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
5997 | ext4_writepage_trans_blocks(inode)); | |
9ea7df53 | 5998 | if (IS_ERR(handle)) { |
c2ec175c | 5999 | ret = VM_FAULT_SIGBUS; |
9ea7df53 JK |
6000 | goto out; |
6001 | } | |
401b25aa SJ |
6002 | err = block_page_mkwrite(vma, vmf, get_block); |
6003 | if (!err && ext4_should_journal_data(inode)) { | |
f19d5870 | 6004 | if (ext4_walk_page_buffers(handle, page_buffers(page), 0, |
09cbfeaf | 6005 | PAGE_SIZE, NULL, do_journal_get_write_access)) { |
9ea7df53 JK |
6006 | unlock_page(page); |
6007 | ret = VM_FAULT_SIGBUS; | |
fcbb5515 | 6008 | ext4_journal_stop(handle); |
9ea7df53 JK |
6009 | goto out; |
6010 | } | |
6011 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); | |
6012 | } | |
6013 | ext4_journal_stop(handle); | |
401b25aa | 6014 | if (err == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) |
9ea7df53 JK |
6015 | goto retry_alloc; |
6016 | out_ret: | |
401b25aa | 6017 | ret = block_page_mkwrite_return(err); |
9ea7df53 | 6018 | out: |
ea3d7209 | 6019 | up_read(&EXT4_I(inode)->i_mmap_sem); |
8e8ad8a5 | 6020 | sb_end_pagefault(inode->i_sb); |
2e9ee850 AK |
6021 | return ret; |
6022 | } | |
ea3d7209 | 6023 | |
401b25aa | 6024 | vm_fault_t ext4_filemap_fault(struct vm_fault *vmf) |
ea3d7209 | 6025 | { |
11bac800 | 6026 | struct inode *inode = file_inode(vmf->vma->vm_file); |
401b25aa | 6027 | vm_fault_t ret; |
ea3d7209 JK |
6028 | |
6029 | down_read(&EXT4_I(inode)->i_mmap_sem); | |
401b25aa | 6030 | ret = filemap_fault(vmf); |
ea3d7209 JK |
6031 | up_read(&EXT4_I(inode)->i_mmap_sem); |
6032 | ||
401b25aa | 6033 | return ret; |
ea3d7209 | 6034 | } |