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 | ||
a1d6cc56 AK |
51 | #define MPAGE_DA_EXTENT_TAIL 0x01 |
52 | ||
814525f4 DW |
53 | static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw, |
54 | struct ext4_inode_info *ei) | |
55 | { | |
56 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
814525f4 | 57 | __u32 csum; |
b47820ed DJ |
58 | __u16 dummy_csum = 0; |
59 | int offset = offsetof(struct ext4_inode, i_checksum_lo); | |
60 | unsigned int csum_size = sizeof(dummy_csum); | |
814525f4 | 61 | |
b47820ed DJ |
62 | csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw, offset); |
63 | csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, csum_size); | |
64 | offset += csum_size; | |
65 | csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset, | |
66 | EXT4_GOOD_OLD_INODE_SIZE - offset); | |
814525f4 | 67 | |
b47820ed DJ |
68 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
69 | offset = offsetof(struct ext4_inode, i_checksum_hi); | |
70 | csum = ext4_chksum(sbi, csum, (__u8 *)raw + | |
71 | EXT4_GOOD_OLD_INODE_SIZE, | |
72 | offset - EXT4_GOOD_OLD_INODE_SIZE); | |
73 | if (EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) { | |
74 | csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, | |
75 | csum_size); | |
76 | offset += csum_size; | |
b47820ed | 77 | } |
05ac5aa1 DJ |
78 | csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset, |
79 | EXT4_INODE_SIZE(inode->i_sb) - offset); | |
814525f4 DW |
80 | } |
81 | ||
814525f4 DW |
82 | return csum; |
83 | } | |
84 | ||
85 | static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw, | |
86 | struct ext4_inode_info *ei) | |
87 | { | |
88 | __u32 provided, calculated; | |
89 | ||
90 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != | |
91 | cpu_to_le32(EXT4_OS_LINUX) || | |
9aa5d32b | 92 | !ext4_has_metadata_csum(inode->i_sb)) |
814525f4 DW |
93 | return 1; |
94 | ||
95 | provided = le16_to_cpu(raw->i_checksum_lo); | |
96 | calculated = ext4_inode_csum(inode, raw, ei); | |
97 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && | |
98 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) | |
99 | provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16; | |
100 | else | |
101 | calculated &= 0xFFFF; | |
102 | ||
103 | return provided == calculated; | |
104 | } | |
105 | ||
106 | static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw, | |
107 | struct ext4_inode_info *ei) | |
108 | { | |
109 | __u32 csum; | |
110 | ||
111 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != | |
112 | cpu_to_le32(EXT4_OS_LINUX) || | |
9aa5d32b | 113 | !ext4_has_metadata_csum(inode->i_sb)) |
814525f4 DW |
114 | return; |
115 | ||
116 | csum = ext4_inode_csum(inode, raw, ei); | |
117 | raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF); | |
118 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && | |
119 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) | |
120 | raw->i_checksum_hi = cpu_to_le16(csum >> 16); | |
121 | } | |
122 | ||
678aaf48 JK |
123 | static inline int ext4_begin_ordered_truncate(struct inode *inode, |
124 | loff_t new_size) | |
125 | { | |
7ff9c073 | 126 | trace_ext4_begin_ordered_truncate(inode, new_size); |
8aefcd55 TT |
127 | /* |
128 | * If jinode is zero, then we never opened the file for | |
129 | * writing, so there's no need to call | |
130 | * jbd2_journal_begin_ordered_truncate() since there's no | |
131 | * outstanding writes we need to flush. | |
132 | */ | |
133 | if (!EXT4_I(inode)->jinode) | |
134 | return 0; | |
135 | return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode), | |
136 | EXT4_I(inode)->jinode, | |
137 | new_size); | |
678aaf48 JK |
138 | } |
139 | ||
d47992f8 LC |
140 | static void ext4_invalidatepage(struct page *page, unsigned int offset, |
141 | unsigned int length); | |
cb20d518 TT |
142 | static int __ext4_journalled_writepage(struct page *page, unsigned int len); |
143 | static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh); | |
dec214d0 TE |
144 | static int ext4_meta_trans_blocks(struct inode *inode, int lblocks, |
145 | int pextents); | |
64769240 | 146 | |
ac27a0ec DK |
147 | /* |
148 | * Test whether an inode is a fast symlink. | |
407cd7fb | 149 | * A fast symlink has its symlink data stored in ext4_inode_info->i_data. |
ac27a0ec | 150 | */ |
f348c252 | 151 | int ext4_inode_is_fast_symlink(struct inode *inode) |
ac27a0ec | 152 | { |
fc82228a AK |
153 | if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) { |
154 | int ea_blocks = EXT4_I(inode)->i_file_acl ? | |
155 | EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0; | |
156 | ||
157 | if (ext4_has_inline_data(inode)) | |
158 | return 0; | |
159 | ||
160 | return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0); | |
161 | } | |
407cd7fb TE |
162 | return S_ISLNK(inode->i_mode) && inode->i_size && |
163 | (inode->i_size < EXT4_N_BLOCKS * 4); | |
ac27a0ec DK |
164 | } |
165 | ||
ac27a0ec DK |
166 | /* |
167 | * Restart the transaction associated with *handle. This does a commit, | |
168 | * so before we call here everything must be consistently dirtied against | |
169 | * this transaction. | |
170 | */ | |
fa5d1113 | 171 | int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode, |
487caeef | 172 | int nblocks) |
ac27a0ec | 173 | { |
487caeef JK |
174 | int ret; |
175 | ||
176 | /* | |
e35fd660 | 177 | * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this |
487caeef JK |
178 | * moment, get_block can be called only for blocks inside i_size since |
179 | * page cache has been already dropped and writes are blocked by | |
180 | * i_mutex. So we can safely drop the i_data_sem here. | |
181 | */ | |
0390131b | 182 | BUG_ON(EXT4_JOURNAL(inode) == NULL); |
ac27a0ec | 183 | jbd_debug(2, "restarting handle %p\n", handle); |
487caeef | 184 | up_write(&EXT4_I(inode)->i_data_sem); |
8e8eaabe | 185 | ret = ext4_journal_restart(handle, nblocks); |
487caeef | 186 | down_write(&EXT4_I(inode)->i_data_sem); |
fa5d1113 | 187 | ext4_discard_preallocations(inode); |
487caeef JK |
188 | |
189 | return ret; | |
ac27a0ec DK |
190 | } |
191 | ||
192 | /* | |
193 | * Called at the last iput() if i_nlink is zero. | |
194 | */ | |
0930fcc1 | 195 | void ext4_evict_inode(struct inode *inode) |
ac27a0ec DK |
196 | { |
197 | handle_t *handle; | |
bc965ab3 | 198 | int err; |
e50e5129 | 199 | int extra_credits = 3; |
0421a189 | 200 | struct ext4_xattr_inode_array *ea_inode_array = NULL; |
ac27a0ec | 201 | |
7ff9c073 | 202 | trace_ext4_evict_inode(inode); |
2581fdc8 | 203 | |
0930fcc1 | 204 | if (inode->i_nlink) { |
2d859db3 JK |
205 | /* |
206 | * When journalling data dirty buffers are tracked only in the | |
207 | * journal. So although mm thinks everything is clean and | |
208 | * ready for reaping the inode might still have some pages to | |
209 | * write in the running transaction or waiting to be | |
210 | * checkpointed. Thus calling jbd2_journal_invalidatepage() | |
211 | * (via truncate_inode_pages()) to discard these buffers can | |
212 | * cause data loss. Also even if we did not discard these | |
213 | * buffers, we would have no way to find them after the inode | |
214 | * is reaped and thus user could see stale data if he tries to | |
215 | * read them before the transaction is checkpointed. So be | |
216 | * careful and force everything to disk here... We use | |
217 | * ei->i_datasync_tid to store the newest transaction | |
218 | * containing inode's data. | |
219 | * | |
220 | * Note that directories do not have this problem because they | |
221 | * don't use page cache. | |
222 | */ | |
6a7fd522 VN |
223 | if (inode->i_ino != EXT4_JOURNAL_INO && |
224 | ext4_should_journal_data(inode) && | |
3abb1a0f JK |
225 | (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) && |
226 | inode->i_data.nrpages) { | |
2d859db3 JK |
227 | journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; |
228 | tid_t commit_tid = EXT4_I(inode)->i_datasync_tid; | |
229 | ||
d76a3a77 | 230 | jbd2_complete_transaction(journal, commit_tid); |
2d859db3 JK |
231 | filemap_write_and_wait(&inode->i_data); |
232 | } | |
91b0abe3 | 233 | truncate_inode_pages_final(&inode->i_data); |
5dc23bdd | 234 | |
0930fcc1 AV |
235 | goto no_delete; |
236 | } | |
237 | ||
e2bfb088 TT |
238 | if (is_bad_inode(inode)) |
239 | goto no_delete; | |
240 | dquot_initialize(inode); | |
907f4554 | 241 | |
678aaf48 JK |
242 | if (ext4_should_order_data(inode)) |
243 | ext4_begin_ordered_truncate(inode, 0); | |
91b0abe3 | 244 | truncate_inode_pages_final(&inode->i_data); |
ac27a0ec | 245 | |
8e8ad8a5 JK |
246 | /* |
247 | * Protect us against freezing - iput() caller didn't have to have any | |
248 | * protection against it | |
249 | */ | |
250 | sb_start_intwrite(inode->i_sb); | |
e50e5129 | 251 | |
30a7eb97 TE |
252 | if (!IS_NOQUOTA(inode)) |
253 | extra_credits += EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb); | |
254 | ||
255 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, | |
256 | ext4_blocks_for_truncate(inode)+extra_credits); | |
ac27a0ec | 257 | if (IS_ERR(handle)) { |
bc965ab3 | 258 | ext4_std_error(inode->i_sb, PTR_ERR(handle)); |
ac27a0ec DK |
259 | /* |
260 | * If we're going to skip the normal cleanup, we still need to | |
261 | * make sure that the in-core orphan linked list is properly | |
262 | * cleaned up. | |
263 | */ | |
617ba13b | 264 | ext4_orphan_del(NULL, inode); |
8e8ad8a5 | 265 | sb_end_intwrite(inode->i_sb); |
ac27a0ec DK |
266 | goto no_delete; |
267 | } | |
30a7eb97 | 268 | |
ac27a0ec | 269 | if (IS_SYNC(inode)) |
0390131b | 270 | ext4_handle_sync(handle); |
407cd7fb TE |
271 | |
272 | /* | |
273 | * Set inode->i_size to 0 before calling ext4_truncate(). We need | |
274 | * special handling of symlinks here because i_size is used to | |
275 | * determine whether ext4_inode_info->i_data contains symlink data or | |
276 | * block mappings. Setting i_size to 0 will remove its fast symlink | |
277 | * status. Erase i_data so that it becomes a valid empty block map. | |
278 | */ | |
279 | if (ext4_inode_is_fast_symlink(inode)) | |
280 | memset(EXT4_I(inode)->i_data, 0, sizeof(EXT4_I(inode)->i_data)); | |
ac27a0ec | 281 | inode->i_size = 0; |
bc965ab3 TT |
282 | err = ext4_mark_inode_dirty(handle, inode); |
283 | if (err) { | |
12062ddd | 284 | ext4_warning(inode->i_sb, |
bc965ab3 TT |
285 | "couldn't mark inode dirty (err %d)", err); |
286 | goto stop_handle; | |
287 | } | |
2c98eb5e TT |
288 | if (inode->i_blocks) { |
289 | err = ext4_truncate(inode); | |
290 | if (err) { | |
291 | ext4_error(inode->i_sb, | |
292 | "couldn't truncate inode %lu (err %d)", | |
293 | inode->i_ino, err); | |
294 | goto stop_handle; | |
295 | } | |
296 | } | |
bc965ab3 | 297 | |
30a7eb97 TE |
298 | /* Remove xattr references. */ |
299 | err = ext4_xattr_delete_inode(handle, inode, &ea_inode_array, | |
300 | extra_credits); | |
301 | if (err) { | |
302 | ext4_warning(inode->i_sb, "xattr delete (err %d)", err); | |
303 | stop_handle: | |
304 | ext4_journal_stop(handle); | |
305 | ext4_orphan_del(NULL, inode); | |
306 | sb_end_intwrite(inode->i_sb); | |
307 | ext4_xattr_inode_array_free(ea_inode_array); | |
308 | goto no_delete; | |
bc965ab3 TT |
309 | } |
310 | ||
ac27a0ec | 311 | /* |
617ba13b | 312 | * Kill off the orphan record which ext4_truncate created. |
ac27a0ec | 313 | * AKPM: I think this can be inside the above `if'. |
617ba13b | 314 | * Note that ext4_orphan_del() has to be able to cope with the |
ac27a0ec | 315 | * deletion of a non-existent orphan - this is because we don't |
617ba13b | 316 | * know if ext4_truncate() actually created an orphan record. |
ac27a0ec DK |
317 | * (Well, we could do this if we need to, but heck - it works) |
318 | */ | |
617ba13b | 319 | ext4_orphan_del(handle, inode); |
5ffff834 | 320 | EXT4_I(inode)->i_dtime = (__u32)ktime_get_real_seconds(); |
ac27a0ec DK |
321 | |
322 | /* | |
323 | * One subtle ordering requirement: if anything has gone wrong | |
324 | * (transaction abort, IO errors, whatever), then we can still | |
325 | * do these next steps (the fs will already have been marked as | |
326 | * having errors), but we can't free the inode if the mark_dirty | |
327 | * fails. | |
328 | */ | |
617ba13b | 329 | if (ext4_mark_inode_dirty(handle, inode)) |
ac27a0ec | 330 | /* If that failed, just do the required in-core inode clear. */ |
0930fcc1 | 331 | ext4_clear_inode(inode); |
ac27a0ec | 332 | else |
617ba13b MC |
333 | ext4_free_inode(handle, inode); |
334 | ext4_journal_stop(handle); | |
8e8ad8a5 | 335 | sb_end_intwrite(inode->i_sb); |
0421a189 | 336 | ext4_xattr_inode_array_free(ea_inode_array); |
ac27a0ec DK |
337 | return; |
338 | no_delete: | |
0930fcc1 | 339 | ext4_clear_inode(inode); /* We must guarantee clearing of inode... */ |
ac27a0ec DK |
340 | } |
341 | ||
a9e7f447 DM |
342 | #ifdef CONFIG_QUOTA |
343 | qsize_t *ext4_get_reserved_space(struct inode *inode) | |
60e58e0f | 344 | { |
a9e7f447 | 345 | return &EXT4_I(inode)->i_reserved_quota; |
60e58e0f | 346 | } |
a9e7f447 | 347 | #endif |
9d0be502 | 348 | |
0637c6f4 TT |
349 | /* |
350 | * Called with i_data_sem down, which is important since we can call | |
351 | * ext4_discard_preallocations() from here. | |
352 | */ | |
5f634d06 AK |
353 | void ext4_da_update_reserve_space(struct inode *inode, |
354 | int used, int quota_claim) | |
12219aea AK |
355 | { |
356 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0637c6f4 | 357 | struct ext4_inode_info *ei = EXT4_I(inode); |
0637c6f4 TT |
358 | |
359 | spin_lock(&ei->i_block_reservation_lock); | |
d8990240 | 360 | trace_ext4_da_update_reserve_space(inode, used, quota_claim); |
0637c6f4 | 361 | if (unlikely(used > ei->i_reserved_data_blocks)) { |
8de5c325 | 362 | ext4_warning(inode->i_sb, "%s: ino %lu, used %d " |
1084f252 | 363 | "with only %d reserved data blocks", |
0637c6f4 TT |
364 | __func__, inode->i_ino, used, |
365 | ei->i_reserved_data_blocks); | |
366 | WARN_ON(1); | |
367 | used = ei->i_reserved_data_blocks; | |
368 | } | |
12219aea | 369 | |
0637c6f4 TT |
370 | /* Update per-inode reservations */ |
371 | ei->i_reserved_data_blocks -= used; | |
71d4f7d0 | 372 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, used); |
6bc6e63f | 373 | |
12219aea | 374 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
60e58e0f | 375 | |
72b8ab9d ES |
376 | /* Update quota subsystem for data blocks */ |
377 | if (quota_claim) | |
7b415bf6 | 378 | dquot_claim_block(inode, EXT4_C2B(sbi, used)); |
72b8ab9d | 379 | else { |
5f634d06 AK |
380 | /* |
381 | * We did fallocate with an offset that is already delayed | |
382 | * allocated. So on delayed allocated writeback we should | |
72b8ab9d | 383 | * not re-claim the quota for fallocated blocks. |
5f634d06 | 384 | */ |
7b415bf6 | 385 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, used)); |
5f634d06 | 386 | } |
d6014301 AK |
387 | |
388 | /* | |
389 | * If we have done all the pending block allocations and if | |
390 | * there aren't any writers on the inode, we can discard the | |
391 | * inode's preallocations. | |
392 | */ | |
0637c6f4 | 393 | if ((ei->i_reserved_data_blocks == 0) && |
82dd124c | 394 | !inode_is_open_for_write(inode)) |
d6014301 | 395 | ext4_discard_preallocations(inode); |
12219aea AK |
396 | } |
397 | ||
e29136f8 | 398 | static int __check_block_validity(struct inode *inode, const char *func, |
c398eda0 TT |
399 | unsigned int line, |
400 | struct ext4_map_blocks *map) | |
6fd058f7 | 401 | { |
345c0dbf TT |
402 | if (ext4_has_feature_journal(inode->i_sb) && |
403 | (inode->i_ino == | |
404 | le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_journal_inum))) | |
405 | return 0; | |
24676da4 TT |
406 | if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk, |
407 | map->m_len)) { | |
c398eda0 | 408 | ext4_error_inode(inode, func, line, map->m_pblk, |
bdbd6ce0 | 409 | "lblock %lu mapped to illegal pblock %llu " |
c398eda0 | 410 | "(length %d)", (unsigned long) map->m_lblk, |
bdbd6ce0 | 411 | map->m_pblk, map->m_len); |
6a797d27 | 412 | return -EFSCORRUPTED; |
6fd058f7 TT |
413 | } |
414 | return 0; | |
415 | } | |
416 | ||
53085fac JK |
417 | int ext4_issue_zeroout(struct inode *inode, ext4_lblk_t lblk, ext4_fsblk_t pblk, |
418 | ext4_lblk_t len) | |
419 | { | |
420 | int ret; | |
421 | ||
592ddec7 | 422 | if (IS_ENCRYPTED(inode)) |
a7550b30 | 423 | return fscrypt_zeroout_range(inode, lblk, pblk, len); |
53085fac JK |
424 | |
425 | ret = sb_issue_zeroout(inode->i_sb, pblk, len, GFP_NOFS); | |
426 | if (ret > 0) | |
427 | ret = 0; | |
428 | ||
429 | return ret; | |
430 | } | |
431 | ||
e29136f8 | 432 | #define check_block_validity(inode, map) \ |
c398eda0 | 433 | __check_block_validity((inode), __func__, __LINE__, (map)) |
e29136f8 | 434 | |
921f266b DM |
435 | #ifdef ES_AGGRESSIVE_TEST |
436 | static void ext4_map_blocks_es_recheck(handle_t *handle, | |
437 | struct inode *inode, | |
438 | struct ext4_map_blocks *es_map, | |
439 | struct ext4_map_blocks *map, | |
440 | int flags) | |
441 | { | |
442 | int retval; | |
443 | ||
444 | map->m_flags = 0; | |
445 | /* | |
446 | * There is a race window that the result is not the same. | |
447 | * e.g. xfstests #223 when dioread_nolock enables. The reason | |
448 | * is that we lookup a block mapping in extent status tree with | |
449 | * out taking i_data_sem. So at the time the unwritten extent | |
450 | * could be converted. | |
451 | */ | |
2dcba478 | 452 | down_read(&EXT4_I(inode)->i_data_sem); |
921f266b DM |
453 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
454 | retval = ext4_ext_map_blocks(handle, inode, map, flags & | |
455 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
456 | } else { | |
457 | retval = ext4_ind_map_blocks(handle, inode, map, flags & | |
458 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
459 | } | |
2dcba478 | 460 | up_read((&EXT4_I(inode)->i_data_sem)); |
921f266b DM |
461 | |
462 | /* | |
463 | * We don't check m_len because extent will be collpased in status | |
464 | * tree. So the m_len might not equal. | |
465 | */ | |
466 | if (es_map->m_lblk != map->m_lblk || | |
467 | es_map->m_flags != map->m_flags || | |
468 | es_map->m_pblk != map->m_pblk) { | |
bdafe42a | 469 | printk("ES cache assertion failed for inode: %lu " |
921f266b DM |
470 | "es_cached ex [%d/%d/%llu/%x] != " |
471 | "found ex [%d/%d/%llu/%x] retval %d flags %x\n", | |
472 | inode->i_ino, es_map->m_lblk, es_map->m_len, | |
473 | es_map->m_pblk, es_map->m_flags, map->m_lblk, | |
474 | map->m_len, map->m_pblk, map->m_flags, | |
475 | retval, flags); | |
476 | } | |
477 | } | |
478 | #endif /* ES_AGGRESSIVE_TEST */ | |
479 | ||
f5ab0d1f | 480 | /* |
e35fd660 | 481 | * The ext4_map_blocks() function tries to look up the requested blocks, |
2b2d6d01 | 482 | * and returns if the blocks are already mapped. |
f5ab0d1f | 483 | * |
f5ab0d1f MC |
484 | * Otherwise it takes the write lock of the i_data_sem and allocate blocks |
485 | * and store the allocated blocks in the result buffer head and mark it | |
486 | * mapped. | |
487 | * | |
e35fd660 TT |
488 | * If file type is extents based, it will call ext4_ext_map_blocks(), |
489 | * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping | |
f5ab0d1f MC |
490 | * based files |
491 | * | |
facab4d9 JK |
492 | * On success, it returns the number of blocks being mapped or allocated. if |
493 | * create==0 and the blocks are pre-allocated and unwritten, the resulting @map | |
494 | * is marked as unwritten. If the create == 1, it will mark @map as mapped. | |
f5ab0d1f MC |
495 | * |
496 | * It returns 0 if plain look up failed (blocks have not been allocated), in | |
facab4d9 JK |
497 | * that case, @map is returned as unmapped but we still do fill map->m_len to |
498 | * indicate the length of a hole starting at map->m_lblk. | |
f5ab0d1f MC |
499 | * |
500 | * It returns the error in case of allocation failure. | |
501 | */ | |
e35fd660 TT |
502 | int ext4_map_blocks(handle_t *handle, struct inode *inode, |
503 | struct ext4_map_blocks *map, int flags) | |
0e855ac8 | 504 | { |
d100eef2 | 505 | struct extent_status es; |
0e855ac8 | 506 | int retval; |
b8a86845 | 507 | int ret = 0; |
921f266b DM |
508 | #ifdef ES_AGGRESSIVE_TEST |
509 | struct ext4_map_blocks orig_map; | |
510 | ||
511 | memcpy(&orig_map, map, sizeof(*map)); | |
512 | #endif | |
f5ab0d1f | 513 | |
e35fd660 TT |
514 | map->m_flags = 0; |
515 | ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u," | |
516 | "logical block %lu\n", inode->i_ino, flags, map->m_len, | |
517 | (unsigned long) map->m_lblk); | |
d100eef2 | 518 | |
e861b5e9 TT |
519 | /* |
520 | * ext4_map_blocks returns an int, and m_len is an unsigned int | |
521 | */ | |
522 | if (unlikely(map->m_len > INT_MAX)) | |
523 | map->m_len = INT_MAX; | |
524 | ||
4adb6ab3 KM |
525 | /* We can handle the block number less than EXT_MAX_BLOCKS */ |
526 | if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS)) | |
6a797d27 | 527 | return -EFSCORRUPTED; |
4adb6ab3 | 528 | |
d100eef2 ZL |
529 | /* Lookup extent status tree firstly */ |
530 | if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) { | |
531 | if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) { | |
532 | map->m_pblk = ext4_es_pblock(&es) + | |
533 | map->m_lblk - es.es_lblk; | |
534 | map->m_flags |= ext4_es_is_written(&es) ? | |
535 | EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN; | |
536 | retval = es.es_len - (map->m_lblk - es.es_lblk); | |
537 | if (retval > map->m_len) | |
538 | retval = map->m_len; | |
539 | map->m_len = retval; | |
540 | } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) { | |
facab4d9 JK |
541 | map->m_pblk = 0; |
542 | retval = es.es_len - (map->m_lblk - es.es_lblk); | |
543 | if (retval > map->m_len) | |
544 | retval = map->m_len; | |
545 | map->m_len = retval; | |
d100eef2 ZL |
546 | retval = 0; |
547 | } else { | |
1e83bc81 | 548 | BUG(); |
d100eef2 | 549 | } |
921f266b DM |
550 | #ifdef ES_AGGRESSIVE_TEST |
551 | ext4_map_blocks_es_recheck(handle, inode, map, | |
552 | &orig_map, flags); | |
553 | #endif | |
d100eef2 ZL |
554 | goto found; |
555 | } | |
556 | ||
4df3d265 | 557 | /* |
b920c755 TT |
558 | * Try to see if we can get the block without requesting a new |
559 | * file system block. | |
4df3d265 | 560 | */ |
2dcba478 | 561 | down_read(&EXT4_I(inode)->i_data_sem); |
12e9b892 | 562 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
a4e5d88b DM |
563 | retval = ext4_ext_map_blocks(handle, inode, map, flags & |
564 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
0e855ac8 | 565 | } else { |
a4e5d88b DM |
566 | retval = ext4_ind_map_blocks(handle, inode, map, flags & |
567 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
0e855ac8 | 568 | } |
f7fec032 | 569 | if (retval > 0) { |
3be78c73 | 570 | unsigned int status; |
f7fec032 | 571 | |
44fb851d ZL |
572 | if (unlikely(retval != map->m_len)) { |
573 | ext4_warning(inode->i_sb, | |
574 | "ES len assertion failed for inode " | |
575 | "%lu: retval %d != map->m_len %d", | |
576 | inode->i_ino, retval, map->m_len); | |
577 | WARN_ON(1); | |
921f266b | 578 | } |
921f266b | 579 | |
f7fec032 ZL |
580 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
581 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; | |
582 | if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) && | |
d2dc317d | 583 | !(status & EXTENT_STATUS_WRITTEN) && |
ad431025 EW |
584 | ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk, |
585 | map->m_lblk + map->m_len - 1)) | |
f7fec032 ZL |
586 | status |= EXTENT_STATUS_DELAYED; |
587 | ret = ext4_es_insert_extent(inode, map->m_lblk, | |
588 | map->m_len, map->m_pblk, status); | |
589 | if (ret < 0) | |
590 | retval = ret; | |
591 | } | |
2dcba478 | 592 | up_read((&EXT4_I(inode)->i_data_sem)); |
f5ab0d1f | 593 | |
d100eef2 | 594 | found: |
e35fd660 | 595 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
b8a86845 | 596 | ret = check_block_validity(inode, map); |
6fd058f7 TT |
597 | if (ret != 0) |
598 | return ret; | |
599 | } | |
600 | ||
f5ab0d1f | 601 | /* If it is only a block(s) look up */ |
c2177057 | 602 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) |
f5ab0d1f MC |
603 | return retval; |
604 | ||
605 | /* | |
606 | * Returns if the blocks have already allocated | |
607 | * | |
608 | * Note that if blocks have been preallocated | |
df3ab170 | 609 | * ext4_ext_get_block() returns the create = 0 |
f5ab0d1f MC |
610 | * with buffer head unmapped. |
611 | */ | |
e35fd660 | 612 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) |
b8a86845 LC |
613 | /* |
614 | * If we need to convert extent to unwritten | |
615 | * we continue and do the actual work in | |
616 | * ext4_ext_map_blocks() | |
617 | */ | |
618 | if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) | |
619 | return retval; | |
4df3d265 | 620 | |
2a8964d6 | 621 | /* |
a25a4e1a ZL |
622 | * Here we clear m_flags because after allocating an new extent, |
623 | * it will be set again. | |
2a8964d6 | 624 | */ |
a25a4e1a | 625 | map->m_flags &= ~EXT4_MAP_FLAGS; |
2a8964d6 | 626 | |
4df3d265 | 627 | /* |
556615dc | 628 | * New blocks allocate and/or writing to unwritten extent |
f5ab0d1f | 629 | * will possibly result in updating i_data, so we take |
d91bd2c1 | 630 | * the write lock of i_data_sem, and call get_block() |
f5ab0d1f | 631 | * with create == 1 flag. |
4df3d265 | 632 | */ |
c8b459f4 | 633 | down_write(&EXT4_I(inode)->i_data_sem); |
d2a17637 | 634 | |
4df3d265 AK |
635 | /* |
636 | * We need to check for EXT4 here because migrate | |
637 | * could have changed the inode type in between | |
638 | */ | |
12e9b892 | 639 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
e35fd660 | 640 | retval = ext4_ext_map_blocks(handle, inode, map, flags); |
0e855ac8 | 641 | } else { |
e35fd660 | 642 | retval = ext4_ind_map_blocks(handle, inode, map, flags); |
267e4db9 | 643 | |
e35fd660 | 644 | if (retval > 0 && map->m_flags & EXT4_MAP_NEW) { |
267e4db9 AK |
645 | /* |
646 | * We allocated new blocks which will result in | |
647 | * i_data's format changing. Force the migrate | |
648 | * to fail by clearing migrate flags | |
649 | */ | |
19f5fb7a | 650 | ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE); |
267e4db9 | 651 | } |
d2a17637 | 652 | |
5f634d06 AK |
653 | /* |
654 | * Update reserved blocks/metadata blocks after successful | |
655 | * block allocation which had been deferred till now. We don't | |
656 | * support fallocate for non extent files. So we can update | |
657 | * reserve space here. | |
658 | */ | |
659 | if ((retval > 0) && | |
1296cc85 | 660 | (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)) |
5f634d06 AK |
661 | ext4_da_update_reserve_space(inode, retval, 1); |
662 | } | |
2ac3b6e0 | 663 | |
f7fec032 | 664 | if (retval > 0) { |
3be78c73 | 665 | unsigned int status; |
f7fec032 | 666 | |
44fb851d ZL |
667 | if (unlikely(retval != map->m_len)) { |
668 | ext4_warning(inode->i_sb, | |
669 | "ES len assertion failed for inode " | |
670 | "%lu: retval %d != map->m_len %d", | |
671 | inode->i_ino, retval, map->m_len); | |
672 | WARN_ON(1); | |
921f266b | 673 | } |
921f266b | 674 | |
c86d8db3 JK |
675 | /* |
676 | * We have to zeroout blocks before inserting them into extent | |
677 | * status tree. Otherwise someone could look them up there and | |
9b623df6 JK |
678 | * use them before they are really zeroed. We also have to |
679 | * unmap metadata before zeroing as otherwise writeback can | |
680 | * overwrite zeros with stale data from block device. | |
c86d8db3 JK |
681 | */ |
682 | if (flags & EXT4_GET_BLOCKS_ZERO && | |
683 | map->m_flags & EXT4_MAP_MAPPED && | |
684 | map->m_flags & EXT4_MAP_NEW) { | |
685 | ret = ext4_issue_zeroout(inode, map->m_lblk, | |
686 | map->m_pblk, map->m_len); | |
687 | if (ret) { | |
688 | retval = ret; | |
689 | goto out_sem; | |
690 | } | |
691 | } | |
692 | ||
adb23551 ZL |
693 | /* |
694 | * If the extent has been zeroed out, we don't need to update | |
695 | * extent status tree. | |
696 | */ | |
697 | if ((flags & EXT4_GET_BLOCKS_PRE_IO) && | |
698 | ext4_es_lookup_extent(inode, map->m_lblk, &es)) { | |
699 | if (ext4_es_is_written(&es)) | |
c86d8db3 | 700 | goto out_sem; |
adb23551 | 701 | } |
f7fec032 ZL |
702 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
703 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; | |
704 | if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) && | |
d2dc317d | 705 | !(status & EXTENT_STATUS_WRITTEN) && |
ad431025 EW |
706 | ext4_es_scan_range(inode, &ext4_es_is_delayed, map->m_lblk, |
707 | map->m_lblk + map->m_len - 1)) | |
f7fec032 ZL |
708 | status |= EXTENT_STATUS_DELAYED; |
709 | ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, | |
710 | map->m_pblk, status); | |
c86d8db3 | 711 | if (ret < 0) { |
f7fec032 | 712 | retval = ret; |
c86d8db3 JK |
713 | goto out_sem; |
714 | } | |
5356f261 AK |
715 | } |
716 | ||
c86d8db3 | 717 | out_sem: |
4df3d265 | 718 | up_write((&EXT4_I(inode)->i_data_sem)); |
e35fd660 | 719 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
b8a86845 | 720 | ret = check_block_validity(inode, map); |
6fd058f7 TT |
721 | if (ret != 0) |
722 | return ret; | |
06bd3c36 JK |
723 | |
724 | /* | |
725 | * Inodes with freshly allocated blocks where contents will be | |
726 | * visible after transaction commit must be on transaction's | |
727 | * ordered data list. | |
728 | */ | |
729 | if (map->m_flags & EXT4_MAP_NEW && | |
730 | !(map->m_flags & EXT4_MAP_UNWRITTEN) && | |
731 | !(flags & EXT4_GET_BLOCKS_ZERO) && | |
02749a4c | 732 | !ext4_is_quota_file(inode) && |
06bd3c36 | 733 | ext4_should_order_data(inode)) { |
73131fbb RZ |
734 | loff_t start_byte = |
735 | (loff_t)map->m_lblk << inode->i_blkbits; | |
736 | loff_t length = (loff_t)map->m_len << inode->i_blkbits; | |
737 | ||
ee0876bc | 738 | if (flags & EXT4_GET_BLOCKS_IO_SUBMIT) |
73131fbb RZ |
739 | ret = ext4_jbd2_inode_add_wait(handle, inode, |
740 | start_byte, length); | |
ee0876bc | 741 | else |
73131fbb RZ |
742 | ret = ext4_jbd2_inode_add_write(handle, inode, |
743 | start_byte, length); | |
06bd3c36 JK |
744 | if (ret) |
745 | return ret; | |
746 | } | |
6fd058f7 | 747 | } |
0e855ac8 AK |
748 | return retval; |
749 | } | |
750 | ||
ed8ad838 JK |
751 | /* |
752 | * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages | |
753 | * we have to be careful as someone else may be manipulating b_state as well. | |
754 | */ | |
755 | static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags) | |
756 | { | |
757 | unsigned long old_state; | |
758 | unsigned long new_state; | |
759 | ||
760 | flags &= EXT4_MAP_FLAGS; | |
761 | ||
762 | /* Dummy buffer_head? Set non-atomically. */ | |
763 | if (!bh->b_page) { | |
764 | bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags; | |
765 | return; | |
766 | } | |
767 | /* | |
768 | * Someone else may be modifying b_state. Be careful! This is ugly but | |
769 | * once we get rid of using bh as a container for mapping information | |
770 | * to pass to / from get_block functions, this can go away. | |
771 | */ | |
772 | do { | |
773 | old_state = READ_ONCE(bh->b_state); | |
774 | new_state = (old_state & ~EXT4_MAP_FLAGS) | flags; | |
775 | } while (unlikely( | |
776 | cmpxchg(&bh->b_state, old_state, new_state) != old_state)); | |
777 | } | |
778 | ||
2ed88685 TT |
779 | static int _ext4_get_block(struct inode *inode, sector_t iblock, |
780 | struct buffer_head *bh, int flags) | |
ac27a0ec | 781 | { |
2ed88685 | 782 | struct ext4_map_blocks map; |
efe70c29 | 783 | int ret = 0; |
ac27a0ec | 784 | |
46c7f254 TM |
785 | if (ext4_has_inline_data(inode)) |
786 | return -ERANGE; | |
787 | ||
2ed88685 TT |
788 | map.m_lblk = iblock; |
789 | map.m_len = bh->b_size >> inode->i_blkbits; | |
790 | ||
efe70c29 JK |
791 | ret = ext4_map_blocks(ext4_journal_current_handle(), inode, &map, |
792 | flags); | |
7fb5409d | 793 | if (ret > 0) { |
2ed88685 | 794 | map_bh(bh, inode->i_sb, map.m_pblk); |
ed8ad838 | 795 | ext4_update_bh_state(bh, map.m_flags); |
2ed88685 | 796 | bh->b_size = inode->i_sb->s_blocksize * map.m_len; |
7fb5409d | 797 | ret = 0; |
547edce3 RZ |
798 | } else if (ret == 0) { |
799 | /* hole case, need to fill in bh->b_size */ | |
800 | bh->b_size = inode->i_sb->s_blocksize * map.m_len; | |
ac27a0ec DK |
801 | } |
802 | return ret; | |
803 | } | |
804 | ||
2ed88685 TT |
805 | int ext4_get_block(struct inode *inode, sector_t iblock, |
806 | struct buffer_head *bh, int create) | |
807 | { | |
808 | return _ext4_get_block(inode, iblock, bh, | |
809 | create ? EXT4_GET_BLOCKS_CREATE : 0); | |
810 | } | |
811 | ||
705965bd JK |
812 | /* |
813 | * Get block function used when preparing for buffered write if we require | |
814 | * creating an unwritten extent if blocks haven't been allocated. The extent | |
815 | * will be converted to written after the IO is complete. | |
816 | */ | |
817 | int ext4_get_block_unwritten(struct inode *inode, sector_t iblock, | |
818 | struct buffer_head *bh_result, int create) | |
819 | { | |
820 | ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n", | |
821 | inode->i_ino, create); | |
822 | return _ext4_get_block(inode, iblock, bh_result, | |
823 | EXT4_GET_BLOCKS_IO_CREATE_EXT); | |
824 | } | |
825 | ||
efe70c29 JK |
826 | /* Maximum number of blocks we map for direct IO at once. */ |
827 | #define DIO_MAX_BLOCKS 4096 | |
828 | ||
e84dfbe2 JK |
829 | /* |
830 | * Get blocks function for the cases that need to start a transaction - | |
831 | * generally difference cases of direct IO and DAX IO. It also handles retries | |
832 | * in case of ENOSPC. | |
833 | */ | |
834 | static int ext4_get_block_trans(struct inode *inode, sector_t iblock, | |
835 | struct buffer_head *bh_result, int flags) | |
efe70c29 JK |
836 | { |
837 | int dio_credits; | |
e84dfbe2 JK |
838 | handle_t *handle; |
839 | int retries = 0; | |
840 | int ret; | |
efe70c29 JK |
841 | |
842 | /* Trim mapping request to maximum we can map at once for DIO */ | |
843 | if (bh_result->b_size >> inode->i_blkbits > DIO_MAX_BLOCKS) | |
844 | bh_result->b_size = DIO_MAX_BLOCKS << inode->i_blkbits; | |
845 | dio_credits = ext4_chunk_trans_blocks(inode, | |
846 | bh_result->b_size >> inode->i_blkbits); | |
e84dfbe2 JK |
847 | retry: |
848 | handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits); | |
849 | if (IS_ERR(handle)) | |
850 | return PTR_ERR(handle); | |
851 | ||
852 | ret = _ext4_get_block(inode, iblock, bh_result, flags); | |
853 | ext4_journal_stop(handle); | |
854 | ||
855 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) | |
856 | goto retry; | |
857 | return ret; | |
efe70c29 JK |
858 | } |
859 | ||
705965bd JK |
860 | /* Get block function for DIO reads and writes to inodes without extents */ |
861 | int ext4_dio_get_block(struct inode *inode, sector_t iblock, | |
862 | struct buffer_head *bh, int create) | |
863 | { | |
efe70c29 JK |
864 | /* We don't expect handle for direct IO */ |
865 | WARN_ON_ONCE(ext4_journal_current_handle()); | |
866 | ||
e84dfbe2 JK |
867 | if (!create) |
868 | return _ext4_get_block(inode, iblock, bh, 0); | |
869 | return ext4_get_block_trans(inode, iblock, bh, EXT4_GET_BLOCKS_CREATE); | |
705965bd JK |
870 | } |
871 | ||
872 | /* | |
109811c2 | 873 | * Get block function for AIO DIO writes when we create unwritten extent if |
705965bd JK |
874 | * blocks are not allocated yet. The extent will be converted to written |
875 | * after IO is complete. | |
876 | */ | |
109811c2 JK |
877 | static int ext4_dio_get_block_unwritten_async(struct inode *inode, |
878 | sector_t iblock, struct buffer_head *bh_result, int create) | |
705965bd | 879 | { |
efe70c29 JK |
880 | int ret; |
881 | ||
efe70c29 JK |
882 | /* We don't expect handle for direct IO */ |
883 | WARN_ON_ONCE(ext4_journal_current_handle()); | |
884 | ||
e84dfbe2 JK |
885 | ret = ext4_get_block_trans(inode, iblock, bh_result, |
886 | EXT4_GET_BLOCKS_IO_CREATE_EXT); | |
efe70c29 | 887 | |
109811c2 JK |
888 | /* |
889 | * When doing DIO using unwritten extents, we need io_end to convert | |
890 | * unwritten extents to written on IO completion. We allocate io_end | |
891 | * once we spot unwritten extent and store it in b_private. Generic | |
892 | * DIO code keeps b_private set and furthermore passes the value to | |
893 | * our completion callback in 'private' argument. | |
894 | */ | |
895 | if (!ret && buffer_unwritten(bh_result)) { | |
896 | if (!bh_result->b_private) { | |
897 | ext4_io_end_t *io_end; | |
898 | ||
899 | io_end = ext4_init_io_end(inode, GFP_KERNEL); | |
900 | if (!io_end) | |
901 | return -ENOMEM; | |
902 | bh_result->b_private = io_end; | |
903 | ext4_set_io_unwritten_flag(inode, io_end); | |
904 | } | |
efe70c29 | 905 | set_buffer_defer_completion(bh_result); |
efe70c29 JK |
906 | } |
907 | ||
908 | return ret; | |
705965bd JK |
909 | } |
910 | ||
109811c2 JK |
911 | /* |
912 | * Get block function for non-AIO DIO writes when we create unwritten extent if | |
913 | * blocks are not allocated yet. The extent will be converted to written | |
1e21196c | 914 | * after IO is complete by ext4_direct_IO_write(). |
109811c2 JK |
915 | */ |
916 | static int ext4_dio_get_block_unwritten_sync(struct inode *inode, | |
917 | sector_t iblock, struct buffer_head *bh_result, int create) | |
918 | { | |
109811c2 JK |
919 | int ret; |
920 | ||
921 | /* We don't expect handle for direct IO */ | |
922 | WARN_ON_ONCE(ext4_journal_current_handle()); | |
923 | ||
e84dfbe2 JK |
924 | ret = ext4_get_block_trans(inode, iblock, bh_result, |
925 | EXT4_GET_BLOCKS_IO_CREATE_EXT); | |
109811c2 JK |
926 | |
927 | /* | |
928 | * Mark inode as having pending DIO writes to unwritten extents. | |
1e21196c | 929 | * ext4_direct_IO_write() checks this flag and converts extents to |
109811c2 JK |
930 | * written. |
931 | */ | |
932 | if (!ret && buffer_unwritten(bh_result)) | |
933 | ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN); | |
934 | ||
935 | return ret; | |
936 | } | |
937 | ||
705965bd JK |
938 | static int ext4_dio_get_block_overwrite(struct inode *inode, sector_t iblock, |
939 | struct buffer_head *bh_result, int create) | |
940 | { | |
941 | int ret; | |
942 | ||
943 | ext4_debug("ext4_dio_get_block_overwrite: inode %lu, create flag %d\n", | |
944 | inode->i_ino, create); | |
efe70c29 JK |
945 | /* We don't expect handle for direct IO */ |
946 | WARN_ON_ONCE(ext4_journal_current_handle()); | |
947 | ||
705965bd JK |
948 | ret = _ext4_get_block(inode, iblock, bh_result, 0); |
949 | /* | |
950 | * Blocks should have been preallocated! ext4_file_write_iter() checks | |
951 | * that. | |
952 | */ | |
efe70c29 | 953 | WARN_ON_ONCE(!buffer_mapped(bh_result) || buffer_unwritten(bh_result)); |
705965bd JK |
954 | |
955 | return ret; | |
956 | } | |
957 | ||
958 | ||
ac27a0ec DK |
959 | /* |
960 | * `handle' can be NULL if create is zero | |
961 | */ | |
617ba13b | 962 | struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode, |
c5e298ae | 963 | ext4_lblk_t block, int map_flags) |
ac27a0ec | 964 | { |
2ed88685 TT |
965 | struct ext4_map_blocks map; |
966 | struct buffer_head *bh; | |
c5e298ae | 967 | int create = map_flags & EXT4_GET_BLOCKS_CREATE; |
10560082 | 968 | int err; |
ac27a0ec DK |
969 | |
970 | J_ASSERT(handle != NULL || create == 0); | |
971 | ||
2ed88685 TT |
972 | map.m_lblk = block; |
973 | map.m_len = 1; | |
c5e298ae | 974 | err = ext4_map_blocks(handle, inode, &map, map_flags); |
ac27a0ec | 975 | |
10560082 TT |
976 | if (err == 0) |
977 | return create ? ERR_PTR(-ENOSPC) : NULL; | |
2ed88685 | 978 | if (err < 0) |
10560082 | 979 | return ERR_PTR(err); |
2ed88685 TT |
980 | |
981 | bh = sb_getblk(inode->i_sb, map.m_pblk); | |
10560082 TT |
982 | if (unlikely(!bh)) |
983 | return ERR_PTR(-ENOMEM); | |
2ed88685 TT |
984 | if (map.m_flags & EXT4_MAP_NEW) { |
985 | J_ASSERT(create != 0); | |
986 | J_ASSERT(handle != NULL); | |
ac27a0ec | 987 | |
2ed88685 TT |
988 | /* |
989 | * Now that we do not always journal data, we should | |
990 | * keep in mind whether this should always journal the | |
991 | * new buffer as metadata. For now, regular file | |
992 | * writes use ext4_get_block instead, so it's not a | |
993 | * problem. | |
994 | */ | |
995 | lock_buffer(bh); | |
996 | BUFFER_TRACE(bh, "call get_create_access"); | |
10560082 TT |
997 | err = ext4_journal_get_create_access(handle, bh); |
998 | if (unlikely(err)) { | |
999 | unlock_buffer(bh); | |
1000 | goto errout; | |
1001 | } | |
1002 | if (!buffer_uptodate(bh)) { | |
2ed88685 TT |
1003 | memset(bh->b_data, 0, inode->i_sb->s_blocksize); |
1004 | set_buffer_uptodate(bh); | |
ac27a0ec | 1005 | } |
2ed88685 TT |
1006 | unlock_buffer(bh); |
1007 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); | |
1008 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
10560082 TT |
1009 | if (unlikely(err)) |
1010 | goto errout; | |
1011 | } else | |
2ed88685 | 1012 | BUFFER_TRACE(bh, "not a new buffer"); |
2ed88685 | 1013 | return bh; |
10560082 TT |
1014 | errout: |
1015 | brelse(bh); | |
1016 | return ERR_PTR(err); | |
ac27a0ec DK |
1017 | } |
1018 | ||
617ba13b | 1019 | struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode, |
c5e298ae | 1020 | ext4_lblk_t block, int map_flags) |
ac27a0ec | 1021 | { |
af5bc92d | 1022 | struct buffer_head *bh; |
ac27a0ec | 1023 | |
c5e298ae | 1024 | bh = ext4_getblk(handle, inode, block, map_flags); |
1c215028 | 1025 | if (IS_ERR(bh)) |
ac27a0ec | 1026 | return bh; |
1c215028 | 1027 | if (!bh || buffer_uptodate(bh)) |
ac27a0ec | 1028 | return bh; |
dfec8a14 | 1029 | ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &bh); |
ac27a0ec DK |
1030 | wait_on_buffer(bh); |
1031 | if (buffer_uptodate(bh)) | |
1032 | return bh; | |
1033 | put_bh(bh); | |
1c215028 | 1034 | return ERR_PTR(-EIO); |
ac27a0ec DK |
1035 | } |
1036 | ||
9699d4f9 TE |
1037 | /* Read a contiguous batch of blocks. */ |
1038 | int ext4_bread_batch(struct inode *inode, ext4_lblk_t block, int bh_count, | |
1039 | bool wait, struct buffer_head **bhs) | |
1040 | { | |
1041 | int i, err; | |
1042 | ||
1043 | for (i = 0; i < bh_count; i++) { | |
1044 | bhs[i] = ext4_getblk(NULL, inode, block + i, 0 /* map_flags */); | |
1045 | if (IS_ERR(bhs[i])) { | |
1046 | err = PTR_ERR(bhs[i]); | |
1047 | bh_count = i; | |
1048 | goto out_brelse; | |
1049 | } | |
1050 | } | |
1051 | ||
1052 | for (i = 0; i < bh_count; i++) | |
1053 | /* Note that NULL bhs[i] is valid because of holes. */ | |
1054 | if (bhs[i] && !buffer_uptodate(bhs[i])) | |
1055 | ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, | |
1056 | &bhs[i]); | |
1057 | ||
1058 | if (!wait) | |
1059 | return 0; | |
1060 | ||
1061 | for (i = 0; i < bh_count; i++) | |
1062 | if (bhs[i]) | |
1063 | wait_on_buffer(bhs[i]); | |
1064 | ||
1065 | for (i = 0; i < bh_count; i++) { | |
1066 | if (bhs[i] && !buffer_uptodate(bhs[i])) { | |
1067 | err = -EIO; | |
1068 | goto out_brelse; | |
1069 | } | |
1070 | } | |
1071 | return 0; | |
1072 | ||
1073 | out_brelse: | |
1074 | for (i = 0; i < bh_count; i++) { | |
1075 | brelse(bhs[i]); | |
1076 | bhs[i] = NULL; | |
1077 | } | |
1078 | return err; | |
1079 | } | |
1080 | ||
f19d5870 TM |
1081 | int ext4_walk_page_buffers(handle_t *handle, |
1082 | struct buffer_head *head, | |
1083 | unsigned from, | |
1084 | unsigned to, | |
1085 | int *partial, | |
1086 | int (*fn)(handle_t *handle, | |
1087 | struct buffer_head *bh)) | |
ac27a0ec DK |
1088 | { |
1089 | struct buffer_head *bh; | |
1090 | unsigned block_start, block_end; | |
1091 | unsigned blocksize = head->b_size; | |
1092 | int err, ret = 0; | |
1093 | struct buffer_head *next; | |
1094 | ||
af5bc92d TT |
1095 | for (bh = head, block_start = 0; |
1096 | ret == 0 && (bh != head || !block_start); | |
de9a55b8 | 1097 | block_start = block_end, bh = next) { |
ac27a0ec DK |
1098 | next = bh->b_this_page; |
1099 | block_end = block_start + blocksize; | |
1100 | if (block_end <= from || block_start >= to) { | |
1101 | if (partial && !buffer_uptodate(bh)) | |
1102 | *partial = 1; | |
1103 | continue; | |
1104 | } | |
1105 | err = (*fn)(handle, bh); | |
1106 | if (!ret) | |
1107 | ret = err; | |
1108 | } | |
1109 | return ret; | |
1110 | } | |
1111 | ||
1112 | /* | |
1113 | * To preserve ordering, it is essential that the hole instantiation and | |
1114 | * the data write be encapsulated in a single transaction. We cannot | |
617ba13b | 1115 | * close off a transaction and start a new one between the ext4_get_block() |
dab291af | 1116 | * and the commit_write(). So doing the jbd2_journal_start at the start of |
ac27a0ec DK |
1117 | * prepare_write() is the right place. |
1118 | * | |
36ade451 JK |
1119 | * Also, this function can nest inside ext4_writepage(). In that case, we |
1120 | * *know* that ext4_writepage() has generated enough buffer credits to do the | |
1121 | * whole page. So we won't block on the journal in that case, which is good, | |
1122 | * because the caller may be PF_MEMALLOC. | |
ac27a0ec | 1123 | * |
617ba13b | 1124 | * By accident, ext4 can be reentered when a transaction is open via |
ac27a0ec DK |
1125 | * quota file writes. If we were to commit the transaction while thus |
1126 | * reentered, there can be a deadlock - we would be holding a quota | |
1127 | * lock, and the commit would never complete if another thread had a | |
1128 | * transaction open and was blocking on the quota lock - a ranking | |
1129 | * violation. | |
1130 | * | |
dab291af | 1131 | * So what we do is to rely on the fact that jbd2_journal_stop/journal_start |
ac27a0ec DK |
1132 | * will _not_ run commit under these circumstances because handle->h_ref |
1133 | * is elevated. We'll still have enough credits for the tiny quotafile | |
1134 | * write. | |
1135 | */ | |
f19d5870 TM |
1136 | int do_journal_get_write_access(handle_t *handle, |
1137 | struct buffer_head *bh) | |
ac27a0ec | 1138 | { |
56d35a4c JK |
1139 | int dirty = buffer_dirty(bh); |
1140 | int ret; | |
1141 | ||
ac27a0ec DK |
1142 | if (!buffer_mapped(bh) || buffer_freed(bh)) |
1143 | return 0; | |
56d35a4c | 1144 | /* |
ebdec241 | 1145 | * __block_write_begin() could have dirtied some buffers. Clean |
56d35a4c JK |
1146 | * the dirty bit as jbd2_journal_get_write_access() could complain |
1147 | * otherwise about fs integrity issues. Setting of the dirty bit | |
ebdec241 | 1148 | * by __block_write_begin() isn't a real problem here as we clear |
56d35a4c JK |
1149 | * the bit before releasing a page lock and thus writeback cannot |
1150 | * ever write the buffer. | |
1151 | */ | |
1152 | if (dirty) | |
1153 | clear_buffer_dirty(bh); | |
5d601255 | 1154 | BUFFER_TRACE(bh, "get write access"); |
56d35a4c JK |
1155 | ret = ext4_journal_get_write_access(handle, bh); |
1156 | if (!ret && dirty) | |
1157 | ret = ext4_handle_dirty_metadata(handle, NULL, bh); | |
1158 | return ret; | |
ac27a0ec DK |
1159 | } |
1160 | ||
643fa961 | 1161 | #ifdef CONFIG_FS_ENCRYPTION |
2058f83a MH |
1162 | static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len, |
1163 | get_block_t *get_block) | |
1164 | { | |
09cbfeaf | 1165 | unsigned from = pos & (PAGE_SIZE - 1); |
2058f83a MH |
1166 | unsigned to = from + len; |
1167 | struct inode *inode = page->mapping->host; | |
1168 | unsigned block_start, block_end; | |
1169 | sector_t block; | |
1170 | int err = 0; | |
1171 | unsigned blocksize = inode->i_sb->s_blocksize; | |
1172 | unsigned bbits; | |
0b578f35 CR |
1173 | struct buffer_head *bh, *head, *wait[2]; |
1174 | int nr_wait = 0; | |
1175 | int i; | |
2058f83a MH |
1176 | |
1177 | BUG_ON(!PageLocked(page)); | |
09cbfeaf KS |
1178 | BUG_ON(from > PAGE_SIZE); |
1179 | BUG_ON(to > PAGE_SIZE); | |
2058f83a MH |
1180 | BUG_ON(from > to); |
1181 | ||
1182 | if (!page_has_buffers(page)) | |
1183 | create_empty_buffers(page, blocksize, 0); | |
1184 | head = page_buffers(page); | |
1185 | bbits = ilog2(blocksize); | |
09cbfeaf | 1186 | block = (sector_t)page->index << (PAGE_SHIFT - bbits); |
2058f83a MH |
1187 | |
1188 | for (bh = head, block_start = 0; bh != head || !block_start; | |
1189 | block++, block_start = block_end, bh = bh->b_this_page) { | |
1190 | block_end = block_start + blocksize; | |
1191 | if (block_end <= from || block_start >= to) { | |
1192 | if (PageUptodate(page)) { | |
1193 | if (!buffer_uptodate(bh)) | |
1194 | set_buffer_uptodate(bh); | |
1195 | } | |
1196 | continue; | |
1197 | } | |
1198 | if (buffer_new(bh)) | |
1199 | clear_buffer_new(bh); | |
1200 | if (!buffer_mapped(bh)) { | |
1201 | WARN_ON(bh->b_size != blocksize); | |
1202 | err = get_block(inode, block, bh, 1); | |
1203 | if (err) | |
1204 | break; | |
1205 | if (buffer_new(bh)) { | |
2058f83a MH |
1206 | if (PageUptodate(page)) { |
1207 | clear_buffer_new(bh); | |
1208 | set_buffer_uptodate(bh); | |
1209 | mark_buffer_dirty(bh); | |
1210 | continue; | |
1211 | } | |
1212 | if (block_end > to || block_start < from) | |
1213 | zero_user_segments(page, to, block_end, | |
1214 | block_start, from); | |
1215 | continue; | |
1216 | } | |
1217 | } | |
1218 | if (PageUptodate(page)) { | |
1219 | if (!buffer_uptodate(bh)) | |
1220 | set_buffer_uptodate(bh); | |
1221 | continue; | |
1222 | } | |
1223 | if (!buffer_uptodate(bh) && !buffer_delay(bh) && | |
1224 | !buffer_unwritten(bh) && | |
1225 | (block_start < from || block_end > to)) { | |
dfec8a14 | 1226 | ll_rw_block(REQ_OP_READ, 0, 1, &bh); |
0b578f35 | 1227 | wait[nr_wait++] = bh; |
2058f83a MH |
1228 | } |
1229 | } | |
1230 | /* | |
1231 | * If we issued read requests, let them complete. | |
1232 | */ | |
0b578f35 CR |
1233 | for (i = 0; i < nr_wait; i++) { |
1234 | wait_on_buffer(wait[i]); | |
1235 | if (!buffer_uptodate(wait[i])) | |
2058f83a MH |
1236 | err = -EIO; |
1237 | } | |
7e0785fc | 1238 | if (unlikely(err)) { |
2058f83a | 1239 | page_zero_new_buffers(page, from, to); |
0b578f35 CR |
1240 | } else if (IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode)) { |
1241 | for (i = 0; i < nr_wait; i++) { | |
1242 | int err2; | |
1243 | ||
1244 | err2 = fscrypt_decrypt_pagecache_blocks(page, blocksize, | |
1245 | bh_offset(wait[i])); | |
1246 | if (err2) { | |
1247 | clear_buffer_uptodate(wait[i]); | |
1248 | err = err2; | |
1249 | } | |
1250 | } | |
7e0785fc CR |
1251 | } |
1252 | ||
2058f83a MH |
1253 | return err; |
1254 | } | |
1255 | #endif | |
1256 | ||
bfc1af65 | 1257 | static int ext4_write_begin(struct file *file, struct address_space *mapping, |
de9a55b8 TT |
1258 | loff_t pos, unsigned len, unsigned flags, |
1259 | struct page **pagep, void **fsdata) | |
ac27a0ec | 1260 | { |
af5bc92d | 1261 | struct inode *inode = mapping->host; |
1938a150 | 1262 | int ret, needed_blocks; |
ac27a0ec DK |
1263 | handle_t *handle; |
1264 | int retries = 0; | |
af5bc92d | 1265 | struct page *page; |
de9a55b8 | 1266 | pgoff_t index; |
af5bc92d | 1267 | unsigned from, to; |
bfc1af65 | 1268 | |
0db1ff22 TT |
1269 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
1270 | return -EIO; | |
1271 | ||
9bffad1e | 1272 | trace_ext4_write_begin(inode, pos, len, flags); |
1938a150 AK |
1273 | /* |
1274 | * Reserve one block more for addition to orphan list in case | |
1275 | * we allocate blocks but write fails for some reason | |
1276 | */ | |
1277 | needed_blocks = ext4_writepage_trans_blocks(inode) + 1; | |
09cbfeaf KS |
1278 | index = pos >> PAGE_SHIFT; |
1279 | from = pos & (PAGE_SIZE - 1); | |
af5bc92d | 1280 | to = from + len; |
ac27a0ec | 1281 | |
f19d5870 TM |
1282 | if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { |
1283 | ret = ext4_try_to_write_inline_data(mapping, inode, pos, len, | |
1284 | flags, pagep); | |
1285 | if (ret < 0) | |
47564bfb TT |
1286 | return ret; |
1287 | if (ret == 1) | |
1288 | return 0; | |
f19d5870 TM |
1289 | } |
1290 | ||
47564bfb TT |
1291 | /* |
1292 | * grab_cache_page_write_begin() can take a long time if the | |
1293 | * system is thrashing due to memory pressure, or if the page | |
1294 | * is being written back. So grab it first before we start | |
1295 | * the transaction handle. This also allows us to allocate | |
1296 | * the page (if needed) without using GFP_NOFS. | |
1297 | */ | |
1298 | retry_grab: | |
1299 | page = grab_cache_page_write_begin(mapping, index, flags); | |
1300 | if (!page) | |
1301 | return -ENOMEM; | |
1302 | unlock_page(page); | |
1303 | ||
1304 | retry_journal: | |
9924a92a | 1305 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks); |
af5bc92d | 1306 | if (IS_ERR(handle)) { |
09cbfeaf | 1307 | put_page(page); |
47564bfb | 1308 | return PTR_ERR(handle); |
7479d2b9 | 1309 | } |
ac27a0ec | 1310 | |
47564bfb TT |
1311 | lock_page(page); |
1312 | if (page->mapping != mapping) { | |
1313 | /* The page got truncated from under us */ | |
1314 | unlock_page(page); | |
09cbfeaf | 1315 | put_page(page); |
cf108bca | 1316 | ext4_journal_stop(handle); |
47564bfb | 1317 | goto retry_grab; |
cf108bca | 1318 | } |
7afe5aa5 DM |
1319 | /* In case writeback began while the page was unlocked */ |
1320 | wait_for_stable_page(page); | |
cf108bca | 1321 | |
643fa961 | 1322 | #ifdef CONFIG_FS_ENCRYPTION |
2058f83a MH |
1323 | if (ext4_should_dioread_nolock(inode)) |
1324 | ret = ext4_block_write_begin(page, pos, len, | |
705965bd | 1325 | ext4_get_block_unwritten); |
2058f83a MH |
1326 | else |
1327 | ret = ext4_block_write_begin(page, pos, len, | |
1328 | ext4_get_block); | |
1329 | #else | |
744692dc | 1330 | if (ext4_should_dioread_nolock(inode)) |
705965bd JK |
1331 | ret = __block_write_begin(page, pos, len, |
1332 | ext4_get_block_unwritten); | |
744692dc | 1333 | else |
6e1db88d | 1334 | ret = __block_write_begin(page, pos, len, ext4_get_block); |
2058f83a | 1335 | #endif |
bfc1af65 | 1336 | if (!ret && ext4_should_journal_data(inode)) { |
f19d5870 TM |
1337 | ret = ext4_walk_page_buffers(handle, page_buffers(page), |
1338 | from, to, NULL, | |
1339 | do_journal_get_write_access); | |
ac27a0ec | 1340 | } |
bfc1af65 NP |
1341 | |
1342 | if (ret) { | |
af5bc92d | 1343 | unlock_page(page); |
ae4d5372 | 1344 | /* |
6e1db88d | 1345 | * __block_write_begin may have instantiated a few blocks |
ae4d5372 AK |
1346 | * outside i_size. Trim these off again. Don't need |
1347 | * i_size_read because we hold i_mutex. | |
1938a150 AK |
1348 | * |
1349 | * Add inode to orphan list in case we crash before | |
1350 | * truncate finishes | |
ae4d5372 | 1351 | */ |
ffacfa7a | 1352 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
1938a150 AK |
1353 | ext4_orphan_add(handle, inode); |
1354 | ||
1355 | ext4_journal_stop(handle); | |
1356 | if (pos + len > inode->i_size) { | |
b9a4207d | 1357 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1358 | /* |
ffacfa7a | 1359 | * If truncate failed early the inode might |
1938a150 AK |
1360 | * still be on the orphan list; we need to |
1361 | * make sure the inode is removed from the | |
1362 | * orphan list in that case. | |
1363 | */ | |
1364 | if (inode->i_nlink) | |
1365 | ext4_orphan_del(NULL, inode); | |
1366 | } | |
bfc1af65 | 1367 | |
47564bfb TT |
1368 | if (ret == -ENOSPC && |
1369 | ext4_should_retry_alloc(inode->i_sb, &retries)) | |
1370 | goto retry_journal; | |
09cbfeaf | 1371 | put_page(page); |
47564bfb TT |
1372 | return ret; |
1373 | } | |
1374 | *pagep = page; | |
ac27a0ec DK |
1375 | return ret; |
1376 | } | |
1377 | ||
bfc1af65 NP |
1378 | /* For write_end() in data=journal mode */ |
1379 | static int write_end_fn(handle_t *handle, struct buffer_head *bh) | |
ac27a0ec | 1380 | { |
13fca323 | 1381 | int ret; |
ac27a0ec DK |
1382 | if (!buffer_mapped(bh) || buffer_freed(bh)) |
1383 | return 0; | |
1384 | set_buffer_uptodate(bh); | |
13fca323 TT |
1385 | ret = ext4_handle_dirty_metadata(handle, NULL, bh); |
1386 | clear_buffer_meta(bh); | |
1387 | clear_buffer_prio(bh); | |
1388 | return ret; | |
ac27a0ec DK |
1389 | } |
1390 | ||
eed4333f ZL |
1391 | /* |
1392 | * We need to pick up the new inode size which generic_commit_write gave us | |
1393 | * `file' can be NULL - eg, when called from page_symlink(). | |
1394 | * | |
1395 | * ext4 never places buffers on inode->i_mapping->private_list. metadata | |
1396 | * buffers are managed internally. | |
1397 | */ | |
1398 | static int ext4_write_end(struct file *file, | |
1399 | struct address_space *mapping, | |
1400 | loff_t pos, unsigned len, unsigned copied, | |
1401 | struct page *page, void *fsdata) | |
f8514083 | 1402 | { |
f8514083 | 1403 | handle_t *handle = ext4_journal_current_handle(); |
eed4333f | 1404 | struct inode *inode = mapping->host; |
0572639f | 1405 | loff_t old_size = inode->i_size; |
eed4333f ZL |
1406 | int ret = 0, ret2; |
1407 | int i_size_changed = 0; | |
362eca70 | 1408 | int inline_data = ext4_has_inline_data(inode); |
eed4333f ZL |
1409 | |
1410 | trace_ext4_write_end(inode, pos, len, copied); | |
362eca70 | 1411 | if (inline_data) { |
42c832de TT |
1412 | ret = ext4_write_inline_data_end(inode, pos, len, |
1413 | copied, page); | |
eb5efbcb TT |
1414 | if (ret < 0) { |
1415 | unlock_page(page); | |
1416 | put_page(page); | |
42c832de | 1417 | goto errout; |
eb5efbcb | 1418 | } |
42c832de TT |
1419 | copied = ret; |
1420 | } else | |
f19d5870 TM |
1421 | copied = block_write_end(file, mapping, pos, |
1422 | len, copied, page, fsdata); | |
f8514083 | 1423 | /* |
4631dbf6 | 1424 | * it's important to update i_size while still holding page lock: |
f8514083 AK |
1425 | * page writeout could otherwise come in and zero beyond i_size. |
1426 | */ | |
4631dbf6 | 1427 | i_size_changed = ext4_update_inode_size(inode, pos + copied); |
f8514083 | 1428 | unlock_page(page); |
09cbfeaf | 1429 | put_page(page); |
f8514083 | 1430 | |
0572639f XW |
1431 | if (old_size < pos) |
1432 | pagecache_isize_extended(inode, old_size, pos); | |
f8514083 AK |
1433 | /* |
1434 | * Don't mark the inode dirty under page lock. First, it unnecessarily | |
1435 | * makes the holding time of page lock longer. Second, it forces lock | |
1436 | * ordering of page lock and transaction start for journaling | |
1437 | * filesystems. | |
1438 | */ | |
362eca70 | 1439 | if (i_size_changed || inline_data) |
f8514083 AK |
1440 | ext4_mark_inode_dirty(handle, inode); |
1441 | ||
ffacfa7a | 1442 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
f8514083 AK |
1443 | /* if we have allocated more blocks and copied |
1444 | * less. We will have blocks allocated outside | |
1445 | * inode->i_size. So truncate them | |
1446 | */ | |
1447 | ext4_orphan_add(handle, inode); | |
74d553aa | 1448 | errout: |
617ba13b | 1449 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1450 | if (!ret) |
1451 | ret = ret2; | |
bfc1af65 | 1452 | |
f8514083 | 1453 | if (pos + len > inode->i_size) { |
b9a4207d | 1454 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1455 | /* |
ffacfa7a | 1456 | * If truncate failed early the inode might still be |
f8514083 AK |
1457 | * on the orphan list; we need to make sure the inode |
1458 | * is removed from the orphan list in that case. | |
1459 | */ | |
1460 | if (inode->i_nlink) | |
1461 | ext4_orphan_del(NULL, inode); | |
1462 | } | |
1463 | ||
bfc1af65 | 1464 | return ret ? ret : copied; |
ac27a0ec DK |
1465 | } |
1466 | ||
b90197b6 TT |
1467 | /* |
1468 | * This is a private version of page_zero_new_buffers() which doesn't | |
1469 | * set the buffer to be dirty, since in data=journalled mode we need | |
1470 | * to call ext4_handle_dirty_metadata() instead. | |
1471 | */ | |
3b136499 JK |
1472 | static void ext4_journalled_zero_new_buffers(handle_t *handle, |
1473 | struct page *page, | |
1474 | unsigned from, unsigned to) | |
b90197b6 TT |
1475 | { |
1476 | unsigned int block_start = 0, block_end; | |
1477 | struct buffer_head *head, *bh; | |
1478 | ||
1479 | bh = head = page_buffers(page); | |
1480 | do { | |
1481 | block_end = block_start + bh->b_size; | |
1482 | if (buffer_new(bh)) { | |
1483 | if (block_end > from && block_start < to) { | |
1484 | if (!PageUptodate(page)) { | |
1485 | unsigned start, size; | |
1486 | ||
1487 | start = max(from, block_start); | |
1488 | size = min(to, block_end) - start; | |
1489 | ||
1490 | zero_user(page, start, size); | |
3b136499 | 1491 | write_end_fn(handle, bh); |
b90197b6 TT |
1492 | } |
1493 | clear_buffer_new(bh); | |
1494 | } | |
1495 | } | |
1496 | block_start = block_end; | |
1497 | bh = bh->b_this_page; | |
1498 | } while (bh != head); | |
1499 | } | |
1500 | ||
bfc1af65 | 1501 | static int ext4_journalled_write_end(struct file *file, |
de9a55b8 TT |
1502 | struct address_space *mapping, |
1503 | loff_t pos, unsigned len, unsigned copied, | |
1504 | struct page *page, void *fsdata) | |
ac27a0ec | 1505 | { |
617ba13b | 1506 | handle_t *handle = ext4_journal_current_handle(); |
bfc1af65 | 1507 | struct inode *inode = mapping->host; |
0572639f | 1508 | loff_t old_size = inode->i_size; |
ac27a0ec DK |
1509 | int ret = 0, ret2; |
1510 | int partial = 0; | |
bfc1af65 | 1511 | unsigned from, to; |
4631dbf6 | 1512 | int size_changed = 0; |
362eca70 | 1513 | int inline_data = ext4_has_inline_data(inode); |
ac27a0ec | 1514 | |
9bffad1e | 1515 | trace_ext4_journalled_write_end(inode, pos, len, copied); |
09cbfeaf | 1516 | from = pos & (PAGE_SIZE - 1); |
bfc1af65 NP |
1517 | to = from + len; |
1518 | ||
441c8508 CW |
1519 | BUG_ON(!ext4_handle_valid(handle)); |
1520 | ||
362eca70 | 1521 | if (inline_data) { |
eb5efbcb TT |
1522 | ret = ext4_write_inline_data_end(inode, pos, len, |
1523 | copied, page); | |
1524 | if (ret < 0) { | |
1525 | unlock_page(page); | |
1526 | put_page(page); | |
1527 | goto errout; | |
1528 | } | |
1529 | copied = ret; | |
1530 | } else if (unlikely(copied < len) && !PageUptodate(page)) { | |
3b136499 JK |
1531 | copied = 0; |
1532 | ext4_journalled_zero_new_buffers(handle, page, from, to); | |
1533 | } else { | |
1534 | if (unlikely(copied < len)) | |
1535 | ext4_journalled_zero_new_buffers(handle, page, | |
1536 | from + copied, to); | |
3fdcfb66 | 1537 | ret = ext4_walk_page_buffers(handle, page_buffers(page), from, |
3b136499 JK |
1538 | from + copied, &partial, |
1539 | write_end_fn); | |
3fdcfb66 TM |
1540 | if (!partial) |
1541 | SetPageUptodate(page); | |
1542 | } | |
4631dbf6 | 1543 | size_changed = ext4_update_inode_size(inode, pos + copied); |
19f5fb7a | 1544 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
2d859db3 | 1545 | EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; |
4631dbf6 | 1546 | unlock_page(page); |
09cbfeaf | 1547 | put_page(page); |
4631dbf6 | 1548 | |
0572639f XW |
1549 | if (old_size < pos) |
1550 | pagecache_isize_extended(inode, old_size, pos); | |
1551 | ||
362eca70 | 1552 | if (size_changed || inline_data) { |
617ba13b | 1553 | ret2 = ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
1554 | if (!ret) |
1555 | ret = ret2; | |
1556 | } | |
bfc1af65 | 1557 | |
ffacfa7a | 1558 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
f8514083 AK |
1559 | /* if we have allocated more blocks and copied |
1560 | * less. We will have blocks allocated outside | |
1561 | * inode->i_size. So truncate them | |
1562 | */ | |
1563 | ext4_orphan_add(handle, inode); | |
1564 | ||
eb5efbcb | 1565 | errout: |
617ba13b | 1566 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1567 | if (!ret) |
1568 | ret = ret2; | |
f8514083 | 1569 | if (pos + len > inode->i_size) { |
b9a4207d | 1570 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1571 | /* |
ffacfa7a | 1572 | * If truncate failed early the inode might still be |
f8514083 AK |
1573 | * on the orphan list; we need to make sure the inode |
1574 | * is removed from the orphan list in that case. | |
1575 | */ | |
1576 | if (inode->i_nlink) | |
1577 | ext4_orphan_del(NULL, inode); | |
1578 | } | |
bfc1af65 NP |
1579 | |
1580 | return ret ? ret : copied; | |
ac27a0ec | 1581 | } |
d2a17637 | 1582 | |
9d0be502 | 1583 | /* |
c27e43a1 | 1584 | * Reserve space for a single cluster |
9d0be502 | 1585 | */ |
c27e43a1 | 1586 | static int ext4_da_reserve_space(struct inode *inode) |
d2a17637 | 1587 | { |
60e58e0f | 1588 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
0637c6f4 | 1589 | struct ext4_inode_info *ei = EXT4_I(inode); |
5dd4056d | 1590 | int ret; |
03179fe9 TT |
1591 | |
1592 | /* | |
1593 | * We will charge metadata quota at writeout time; this saves | |
1594 | * us from metadata over-estimation, though we may go over by | |
1595 | * a small amount in the end. Here we just reserve for data. | |
1596 | */ | |
1597 | ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1)); | |
1598 | if (ret) | |
1599 | return ret; | |
d2a17637 | 1600 | |
0637c6f4 | 1601 | spin_lock(&ei->i_block_reservation_lock); |
71d4f7d0 | 1602 | if (ext4_claim_free_clusters(sbi, 1, 0)) { |
03179fe9 | 1603 | spin_unlock(&ei->i_block_reservation_lock); |
03179fe9 | 1604 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1)); |
d2a17637 MC |
1605 | return -ENOSPC; |
1606 | } | |
9d0be502 | 1607 | ei->i_reserved_data_blocks++; |
c27e43a1 | 1608 | trace_ext4_da_reserve_space(inode); |
0637c6f4 | 1609 | spin_unlock(&ei->i_block_reservation_lock); |
39bc680a | 1610 | |
d2a17637 MC |
1611 | return 0; /* success */ |
1612 | } | |
1613 | ||
f456767d | 1614 | void ext4_da_release_space(struct inode *inode, int to_free) |
d2a17637 MC |
1615 | { |
1616 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0637c6f4 | 1617 | struct ext4_inode_info *ei = EXT4_I(inode); |
d2a17637 | 1618 | |
cd213226 MC |
1619 | if (!to_free) |
1620 | return; /* Nothing to release, exit */ | |
1621 | ||
d2a17637 | 1622 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); |
cd213226 | 1623 | |
5a58ec87 | 1624 | trace_ext4_da_release_space(inode, to_free); |
0637c6f4 | 1625 | if (unlikely(to_free > ei->i_reserved_data_blocks)) { |
cd213226 | 1626 | /* |
0637c6f4 TT |
1627 | * if there aren't enough reserved blocks, then the |
1628 | * counter is messed up somewhere. Since this | |
1629 | * function is called from invalidate page, it's | |
1630 | * harmless to return without any action. | |
cd213226 | 1631 | */ |
8de5c325 | 1632 | ext4_warning(inode->i_sb, "ext4_da_release_space: " |
0637c6f4 | 1633 | "ino %lu, to_free %d with only %d reserved " |
1084f252 | 1634 | "data blocks", inode->i_ino, to_free, |
0637c6f4 TT |
1635 | ei->i_reserved_data_blocks); |
1636 | WARN_ON(1); | |
1637 | to_free = ei->i_reserved_data_blocks; | |
cd213226 | 1638 | } |
0637c6f4 | 1639 | ei->i_reserved_data_blocks -= to_free; |
cd213226 | 1640 | |
72b8ab9d | 1641 | /* update fs dirty data blocks counter */ |
57042651 | 1642 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free); |
d2a17637 | 1643 | |
d2a17637 | 1644 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
60e58e0f | 1645 | |
7b415bf6 | 1646 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free)); |
d2a17637 MC |
1647 | } |
1648 | ||
1649 | static void ext4_da_page_release_reservation(struct page *page, | |
ca99fdd2 LC |
1650 | unsigned int offset, |
1651 | unsigned int length) | |
d2a17637 | 1652 | { |
f456767d | 1653 | int contiguous_blks = 0; |
d2a17637 MC |
1654 | struct buffer_head *head, *bh; |
1655 | unsigned int curr_off = 0; | |
7b415bf6 | 1656 | struct inode *inode = page->mapping->host; |
ca99fdd2 | 1657 | unsigned int stop = offset + length; |
51865fda | 1658 | ext4_fsblk_t lblk; |
d2a17637 | 1659 | |
09cbfeaf | 1660 | BUG_ON(stop > PAGE_SIZE || stop < length); |
ca99fdd2 | 1661 | |
d2a17637 MC |
1662 | head = page_buffers(page); |
1663 | bh = head; | |
1664 | do { | |
1665 | unsigned int next_off = curr_off + bh->b_size; | |
1666 | ||
ca99fdd2 LC |
1667 | if (next_off > stop) |
1668 | break; | |
1669 | ||
d2a17637 | 1670 | if ((offset <= curr_off) && (buffer_delay(bh))) { |
9705acd6 | 1671 | contiguous_blks++; |
d2a17637 | 1672 | clear_buffer_delay(bh); |
9705acd6 LC |
1673 | } else if (contiguous_blks) { |
1674 | lblk = page->index << | |
09cbfeaf | 1675 | (PAGE_SHIFT - inode->i_blkbits); |
9705acd6 LC |
1676 | lblk += (curr_off >> inode->i_blkbits) - |
1677 | contiguous_blks; | |
f456767d | 1678 | ext4_es_remove_blks(inode, lblk, contiguous_blks); |
9705acd6 | 1679 | contiguous_blks = 0; |
d2a17637 MC |
1680 | } |
1681 | curr_off = next_off; | |
1682 | } while ((bh = bh->b_this_page) != head); | |
7b415bf6 | 1683 | |
9705acd6 | 1684 | if (contiguous_blks) { |
09cbfeaf | 1685 | lblk = page->index << (PAGE_SHIFT - inode->i_blkbits); |
9705acd6 | 1686 | lblk += (curr_off >> inode->i_blkbits) - contiguous_blks; |
f456767d | 1687 | ext4_es_remove_blks(inode, lblk, contiguous_blks); |
51865fda ZL |
1688 | } |
1689 | ||
d2a17637 | 1690 | } |
ac27a0ec | 1691 | |
64769240 AT |
1692 | /* |
1693 | * Delayed allocation stuff | |
1694 | */ | |
1695 | ||
4e7ea81d JK |
1696 | struct mpage_da_data { |
1697 | struct inode *inode; | |
1698 | struct writeback_control *wbc; | |
6b523df4 | 1699 | |
4e7ea81d JK |
1700 | pgoff_t first_page; /* The first page to write */ |
1701 | pgoff_t next_page; /* Current page to examine */ | |
1702 | pgoff_t last_page; /* Last page to examine */ | |
791b7f08 | 1703 | /* |
4e7ea81d JK |
1704 | * Extent to map - this can be after first_page because that can be |
1705 | * fully mapped. We somewhat abuse m_flags to store whether the extent | |
1706 | * is delalloc or unwritten. | |
791b7f08 | 1707 | */ |
4e7ea81d JK |
1708 | struct ext4_map_blocks map; |
1709 | struct ext4_io_submit io_submit; /* IO submission data */ | |
dddbd6ac | 1710 | unsigned int do_map:1; |
4e7ea81d | 1711 | }; |
64769240 | 1712 | |
4e7ea81d JK |
1713 | static void mpage_release_unused_pages(struct mpage_da_data *mpd, |
1714 | bool invalidate) | |
c4a0c46e AK |
1715 | { |
1716 | int nr_pages, i; | |
1717 | pgoff_t index, end; | |
1718 | struct pagevec pvec; | |
1719 | struct inode *inode = mpd->inode; | |
1720 | struct address_space *mapping = inode->i_mapping; | |
4e7ea81d JK |
1721 | |
1722 | /* This is necessary when next_page == 0. */ | |
1723 | if (mpd->first_page >= mpd->next_page) | |
1724 | return; | |
c4a0c46e | 1725 | |
c7f5938a CW |
1726 | index = mpd->first_page; |
1727 | end = mpd->next_page - 1; | |
4e7ea81d JK |
1728 | if (invalidate) { |
1729 | ext4_lblk_t start, last; | |
09cbfeaf KS |
1730 | start = index << (PAGE_SHIFT - inode->i_blkbits); |
1731 | last = end << (PAGE_SHIFT - inode->i_blkbits); | |
4e7ea81d JK |
1732 | ext4_es_remove_extent(inode, start, last - start + 1); |
1733 | } | |
51865fda | 1734 | |
86679820 | 1735 | pagevec_init(&pvec); |
c4a0c46e | 1736 | while (index <= end) { |
397162ff | 1737 | nr_pages = pagevec_lookup_range(&pvec, mapping, &index, end); |
c4a0c46e AK |
1738 | if (nr_pages == 0) |
1739 | break; | |
1740 | for (i = 0; i < nr_pages; i++) { | |
1741 | struct page *page = pvec.pages[i]; | |
2b85a617 | 1742 | |
c4a0c46e AK |
1743 | BUG_ON(!PageLocked(page)); |
1744 | BUG_ON(PageWriteback(page)); | |
4e7ea81d | 1745 | if (invalidate) { |
4e800c03 | 1746 | if (page_mapped(page)) |
1747 | clear_page_dirty_for_io(page); | |
09cbfeaf | 1748 | block_invalidatepage(page, 0, PAGE_SIZE); |
4e7ea81d JK |
1749 | ClearPageUptodate(page); |
1750 | } | |
c4a0c46e AK |
1751 | unlock_page(page); |
1752 | } | |
9b1d0998 | 1753 | pagevec_release(&pvec); |
c4a0c46e | 1754 | } |
c4a0c46e AK |
1755 | } |
1756 | ||
df22291f AK |
1757 | static void ext4_print_free_blocks(struct inode *inode) |
1758 | { | |
1759 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
92b97816 | 1760 | struct super_block *sb = inode->i_sb; |
f78ee70d | 1761 | struct ext4_inode_info *ei = EXT4_I(inode); |
92b97816 TT |
1762 | |
1763 | ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld", | |
5dee5437 | 1764 | EXT4_C2B(EXT4_SB(inode->i_sb), |
f78ee70d | 1765 | ext4_count_free_clusters(sb))); |
92b97816 TT |
1766 | ext4_msg(sb, KERN_CRIT, "Free/Dirty block details"); |
1767 | ext4_msg(sb, KERN_CRIT, "free_blocks=%lld", | |
f78ee70d | 1768 | (long long) EXT4_C2B(EXT4_SB(sb), |
57042651 | 1769 | percpu_counter_sum(&sbi->s_freeclusters_counter))); |
92b97816 | 1770 | ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld", |
f78ee70d | 1771 | (long long) EXT4_C2B(EXT4_SB(sb), |
7b415bf6 | 1772 | percpu_counter_sum(&sbi->s_dirtyclusters_counter))); |
92b97816 TT |
1773 | ext4_msg(sb, KERN_CRIT, "Block reservation details"); |
1774 | ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u", | |
f78ee70d | 1775 | ei->i_reserved_data_blocks); |
df22291f AK |
1776 | return; |
1777 | } | |
1778 | ||
c364b22c | 1779 | static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh) |
29fa89d0 | 1780 | { |
c364b22c | 1781 | return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh); |
29fa89d0 AK |
1782 | } |
1783 | ||
0b02f4c0 EW |
1784 | /* |
1785 | * ext4_insert_delayed_block - adds a delayed block to the extents status | |
1786 | * tree, incrementing the reserved cluster/block | |
1787 | * count or making a pending reservation | |
1788 | * where needed | |
1789 | * | |
1790 | * @inode - file containing the newly added block | |
1791 | * @lblk - logical block to be added | |
1792 | * | |
1793 | * Returns 0 on success, negative error code on failure. | |
1794 | */ | |
1795 | static int ext4_insert_delayed_block(struct inode *inode, ext4_lblk_t lblk) | |
1796 | { | |
1797 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
1798 | int ret; | |
1799 | bool allocated = false; | |
1800 | ||
1801 | /* | |
1802 | * If the cluster containing lblk is shared with a delayed, | |
1803 | * written, or unwritten extent in a bigalloc file system, it's | |
1804 | * already been accounted for and does not need to be reserved. | |
1805 | * A pending reservation must be made for the cluster if it's | |
1806 | * shared with a written or unwritten extent and doesn't already | |
1807 | * have one. Written and unwritten extents can be purged from the | |
1808 | * extents status tree if the system is under memory pressure, so | |
1809 | * it's necessary to examine the extent tree if a search of the | |
1810 | * extents status tree doesn't get a match. | |
1811 | */ | |
1812 | if (sbi->s_cluster_ratio == 1) { | |
1813 | ret = ext4_da_reserve_space(inode); | |
1814 | if (ret != 0) /* ENOSPC */ | |
1815 | goto errout; | |
1816 | } else { /* bigalloc */ | |
1817 | if (!ext4_es_scan_clu(inode, &ext4_es_is_delonly, lblk)) { | |
1818 | if (!ext4_es_scan_clu(inode, | |
1819 | &ext4_es_is_mapped, lblk)) { | |
1820 | ret = ext4_clu_mapped(inode, | |
1821 | EXT4_B2C(sbi, lblk)); | |
1822 | if (ret < 0) | |
1823 | goto errout; | |
1824 | if (ret == 0) { | |
1825 | ret = ext4_da_reserve_space(inode); | |
1826 | if (ret != 0) /* ENOSPC */ | |
1827 | goto errout; | |
1828 | } else { | |
1829 | allocated = true; | |
1830 | } | |
1831 | } else { | |
1832 | allocated = true; | |
1833 | } | |
1834 | } | |
1835 | } | |
1836 | ||
1837 | ret = ext4_es_insert_delayed_block(inode, lblk, allocated); | |
1838 | ||
1839 | errout: | |
1840 | return ret; | |
1841 | } | |
1842 | ||
5356f261 AK |
1843 | /* |
1844 | * This function is grabs code from the very beginning of | |
1845 | * ext4_map_blocks, but assumes that the caller is from delayed write | |
1846 | * time. This function looks up the requested blocks and sets the | |
1847 | * buffer delay bit under the protection of i_data_sem. | |
1848 | */ | |
1849 | static int ext4_da_map_blocks(struct inode *inode, sector_t iblock, | |
1850 | struct ext4_map_blocks *map, | |
1851 | struct buffer_head *bh) | |
1852 | { | |
d100eef2 | 1853 | struct extent_status es; |
5356f261 AK |
1854 | int retval; |
1855 | sector_t invalid_block = ~((sector_t) 0xffff); | |
921f266b DM |
1856 | #ifdef ES_AGGRESSIVE_TEST |
1857 | struct ext4_map_blocks orig_map; | |
1858 | ||
1859 | memcpy(&orig_map, map, sizeof(*map)); | |
1860 | #endif | |
5356f261 AK |
1861 | |
1862 | if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es)) | |
1863 | invalid_block = ~0; | |
1864 | ||
1865 | map->m_flags = 0; | |
1866 | ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u," | |
1867 | "logical block %lu\n", inode->i_ino, map->m_len, | |
1868 | (unsigned long) map->m_lblk); | |
d100eef2 ZL |
1869 | |
1870 | /* Lookup extent status tree firstly */ | |
1871 | if (ext4_es_lookup_extent(inode, iblock, &es)) { | |
d100eef2 ZL |
1872 | if (ext4_es_is_hole(&es)) { |
1873 | retval = 0; | |
c8b459f4 | 1874 | down_read(&EXT4_I(inode)->i_data_sem); |
d100eef2 ZL |
1875 | goto add_delayed; |
1876 | } | |
1877 | ||
1878 | /* | |
1879 | * Delayed extent could be allocated by fallocate. | |
1880 | * So we need to check it. | |
1881 | */ | |
1882 | if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) { | |
1883 | map_bh(bh, inode->i_sb, invalid_block); | |
1884 | set_buffer_new(bh); | |
1885 | set_buffer_delay(bh); | |
1886 | return 0; | |
1887 | } | |
1888 | ||
1889 | map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk; | |
1890 | retval = es.es_len - (iblock - es.es_lblk); | |
1891 | if (retval > map->m_len) | |
1892 | retval = map->m_len; | |
1893 | map->m_len = retval; | |
1894 | if (ext4_es_is_written(&es)) | |
1895 | map->m_flags |= EXT4_MAP_MAPPED; | |
1896 | else if (ext4_es_is_unwritten(&es)) | |
1897 | map->m_flags |= EXT4_MAP_UNWRITTEN; | |
1898 | else | |
1e83bc81 | 1899 | BUG(); |
d100eef2 | 1900 | |
921f266b DM |
1901 | #ifdef ES_AGGRESSIVE_TEST |
1902 | ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0); | |
1903 | #endif | |
d100eef2 ZL |
1904 | return retval; |
1905 | } | |
1906 | ||
5356f261 AK |
1907 | /* |
1908 | * Try to see if we can get the block without requesting a new | |
1909 | * file system block. | |
1910 | */ | |
c8b459f4 | 1911 | down_read(&EXT4_I(inode)->i_data_sem); |
cbd7584e | 1912 | if (ext4_has_inline_data(inode)) |
9c3569b5 | 1913 | retval = 0; |
cbd7584e | 1914 | else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
2f8e0a7c | 1915 | retval = ext4_ext_map_blocks(NULL, inode, map, 0); |
5356f261 | 1916 | else |
2f8e0a7c | 1917 | retval = ext4_ind_map_blocks(NULL, inode, map, 0); |
5356f261 | 1918 | |
d100eef2 | 1919 | add_delayed: |
5356f261 | 1920 | if (retval == 0) { |
f7fec032 | 1921 | int ret; |
ad431025 | 1922 | |
5356f261 AK |
1923 | /* |
1924 | * XXX: __block_prepare_write() unmaps passed block, | |
1925 | * is it OK? | |
1926 | */ | |
5356f261 | 1927 | |
0b02f4c0 EW |
1928 | ret = ext4_insert_delayed_block(inode, map->m_lblk); |
1929 | if (ret != 0) { | |
f7fec032 | 1930 | retval = ret; |
51865fda | 1931 | goto out_unlock; |
f7fec032 | 1932 | } |
51865fda | 1933 | |
5356f261 AK |
1934 | map_bh(bh, inode->i_sb, invalid_block); |
1935 | set_buffer_new(bh); | |
1936 | set_buffer_delay(bh); | |
f7fec032 ZL |
1937 | } else if (retval > 0) { |
1938 | int ret; | |
3be78c73 | 1939 | unsigned int status; |
f7fec032 | 1940 | |
44fb851d ZL |
1941 | if (unlikely(retval != map->m_len)) { |
1942 | ext4_warning(inode->i_sb, | |
1943 | "ES len assertion failed for inode " | |
1944 | "%lu: retval %d != map->m_len %d", | |
1945 | inode->i_ino, retval, map->m_len); | |
1946 | WARN_ON(1); | |
921f266b | 1947 | } |
921f266b | 1948 | |
f7fec032 ZL |
1949 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
1950 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; | |
1951 | ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, | |
1952 | map->m_pblk, status); | |
1953 | if (ret != 0) | |
1954 | retval = ret; | |
5356f261 AK |
1955 | } |
1956 | ||
1957 | out_unlock: | |
1958 | up_read((&EXT4_I(inode)->i_data_sem)); | |
1959 | ||
1960 | return retval; | |
1961 | } | |
1962 | ||
64769240 | 1963 | /* |
d91bd2c1 | 1964 | * This is a special get_block_t callback which is used by |
b920c755 TT |
1965 | * ext4_da_write_begin(). It will either return mapped block or |
1966 | * reserve space for a single block. | |
29fa89d0 AK |
1967 | * |
1968 | * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set. | |
1969 | * We also have b_blocknr = -1 and b_bdev initialized properly | |
1970 | * | |
1971 | * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set. | |
1972 | * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev | |
1973 | * initialized properly. | |
64769240 | 1974 | */ |
9c3569b5 TM |
1975 | int ext4_da_get_block_prep(struct inode *inode, sector_t iblock, |
1976 | struct buffer_head *bh, int create) | |
64769240 | 1977 | { |
2ed88685 | 1978 | struct ext4_map_blocks map; |
64769240 AT |
1979 | int ret = 0; |
1980 | ||
1981 | BUG_ON(create == 0); | |
2ed88685 TT |
1982 | BUG_ON(bh->b_size != inode->i_sb->s_blocksize); |
1983 | ||
1984 | map.m_lblk = iblock; | |
1985 | map.m_len = 1; | |
64769240 AT |
1986 | |
1987 | /* | |
1988 | * first, we need to know whether the block is allocated already | |
1989 | * preallocated blocks are unmapped but should treated | |
1990 | * the same as allocated blocks. | |
1991 | */ | |
5356f261 AK |
1992 | ret = ext4_da_map_blocks(inode, iblock, &map, bh); |
1993 | if (ret <= 0) | |
2ed88685 | 1994 | return ret; |
64769240 | 1995 | |
2ed88685 | 1996 | map_bh(bh, inode->i_sb, map.m_pblk); |
ed8ad838 | 1997 | ext4_update_bh_state(bh, map.m_flags); |
2ed88685 TT |
1998 | |
1999 | if (buffer_unwritten(bh)) { | |
2000 | /* A delayed write to unwritten bh should be marked | |
2001 | * new and mapped. Mapped ensures that we don't do | |
2002 | * get_block multiple times when we write to the same | |
2003 | * offset and new ensures that we do proper zero out | |
2004 | * for partial write. | |
2005 | */ | |
2006 | set_buffer_new(bh); | |
c8205636 | 2007 | set_buffer_mapped(bh); |
2ed88685 TT |
2008 | } |
2009 | return 0; | |
64769240 | 2010 | } |
61628a3f | 2011 | |
62e086be AK |
2012 | static int bget_one(handle_t *handle, struct buffer_head *bh) |
2013 | { | |
2014 | get_bh(bh); | |
2015 | return 0; | |
2016 | } | |
2017 | ||
2018 | static int bput_one(handle_t *handle, struct buffer_head *bh) | |
2019 | { | |
2020 | put_bh(bh); | |
2021 | return 0; | |
2022 | } | |
2023 | ||
2024 | static int __ext4_journalled_writepage(struct page *page, | |
62e086be AK |
2025 | unsigned int len) |
2026 | { | |
2027 | struct address_space *mapping = page->mapping; | |
2028 | struct inode *inode = mapping->host; | |
3fdcfb66 | 2029 | struct buffer_head *page_bufs = NULL; |
62e086be | 2030 | handle_t *handle = NULL; |
3fdcfb66 TM |
2031 | int ret = 0, err = 0; |
2032 | int inline_data = ext4_has_inline_data(inode); | |
2033 | struct buffer_head *inode_bh = NULL; | |
62e086be | 2034 | |
cb20d518 | 2035 | ClearPageChecked(page); |
3fdcfb66 TM |
2036 | |
2037 | if (inline_data) { | |
2038 | BUG_ON(page->index != 0); | |
2039 | BUG_ON(len > ext4_get_max_inline_size(inode)); | |
2040 | inode_bh = ext4_journalled_write_inline_data(inode, len, page); | |
2041 | if (inode_bh == NULL) | |
2042 | goto out; | |
2043 | } else { | |
2044 | page_bufs = page_buffers(page); | |
2045 | if (!page_bufs) { | |
2046 | BUG(); | |
2047 | goto out; | |
2048 | } | |
2049 | ext4_walk_page_buffers(handle, page_bufs, 0, len, | |
2050 | NULL, bget_one); | |
2051 | } | |
bdf96838 TT |
2052 | /* |
2053 | * We need to release the page lock before we start the | |
2054 | * journal, so grab a reference so the page won't disappear | |
2055 | * out from under us. | |
2056 | */ | |
2057 | get_page(page); | |
62e086be AK |
2058 | unlock_page(page); |
2059 | ||
9924a92a TT |
2060 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
2061 | ext4_writepage_trans_blocks(inode)); | |
62e086be AK |
2062 | if (IS_ERR(handle)) { |
2063 | ret = PTR_ERR(handle); | |
bdf96838 TT |
2064 | put_page(page); |
2065 | goto out_no_pagelock; | |
62e086be | 2066 | } |
441c8508 CW |
2067 | BUG_ON(!ext4_handle_valid(handle)); |
2068 | ||
bdf96838 TT |
2069 | lock_page(page); |
2070 | put_page(page); | |
2071 | if (page->mapping != mapping) { | |
2072 | /* The page got truncated from under us */ | |
2073 | ext4_journal_stop(handle); | |
2074 | ret = 0; | |
2075 | goto out; | |
2076 | } | |
2077 | ||
3fdcfb66 | 2078 | if (inline_data) { |
362eca70 | 2079 | ret = ext4_mark_inode_dirty(handle, inode); |
3fdcfb66 TM |
2080 | } else { |
2081 | ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL, | |
2082 | do_journal_get_write_access); | |
2083 | ||
2084 | err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL, | |
2085 | write_end_fn); | |
2086 | } | |
62e086be AK |
2087 | if (ret == 0) |
2088 | ret = err; | |
2d859db3 | 2089 | EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; |
62e086be AK |
2090 | err = ext4_journal_stop(handle); |
2091 | if (!ret) | |
2092 | ret = err; | |
2093 | ||
3fdcfb66 | 2094 | if (!ext4_has_inline_data(inode)) |
8c9367fd | 2095 | ext4_walk_page_buffers(NULL, page_bufs, 0, len, |
3fdcfb66 | 2096 | NULL, bput_one); |
19f5fb7a | 2097 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
62e086be | 2098 | out: |
bdf96838 TT |
2099 | unlock_page(page); |
2100 | out_no_pagelock: | |
3fdcfb66 | 2101 | brelse(inode_bh); |
62e086be AK |
2102 | return ret; |
2103 | } | |
2104 | ||
61628a3f | 2105 | /* |
43ce1d23 AK |
2106 | * Note that we don't need to start a transaction unless we're journaling data |
2107 | * because we should have holes filled from ext4_page_mkwrite(). We even don't | |
2108 | * need to file the inode to the transaction's list in ordered mode because if | |
2109 | * we are writing back data added by write(), the inode is already there and if | |
25985edc | 2110 | * we are writing back data modified via mmap(), no one guarantees in which |
43ce1d23 AK |
2111 | * transaction the data will hit the disk. In case we are journaling data, we |
2112 | * cannot start transaction directly because transaction start ranks above page | |
2113 | * lock so we have to do some magic. | |
2114 | * | |
b920c755 | 2115 | * This function can get called via... |
20970ba6 | 2116 | * - ext4_writepages after taking page lock (have journal handle) |
b920c755 | 2117 | * - journal_submit_inode_data_buffers (no journal handle) |
f6463b0d | 2118 | * - shrink_page_list via the kswapd/direct reclaim (no journal handle) |
b920c755 | 2119 | * - grab_page_cache when doing write_begin (have journal handle) |
43ce1d23 AK |
2120 | * |
2121 | * We don't do any block allocation in this function. If we have page with | |
2122 | * multiple blocks we need to write those buffer_heads that are mapped. This | |
2123 | * is important for mmaped based write. So if we do with blocksize 1K | |
2124 | * truncate(f, 1024); | |
2125 | * a = mmap(f, 0, 4096); | |
2126 | * a[0] = 'a'; | |
2127 | * truncate(f, 4096); | |
2128 | * we have in the page first buffer_head mapped via page_mkwrite call back | |
90802ed9 | 2129 | * but other buffer_heads would be unmapped but dirty (dirty done via the |
43ce1d23 AK |
2130 | * do_wp_page). So writepage should write the first block. If we modify |
2131 | * the mmap area beyond 1024 we will again get a page_fault and the | |
2132 | * page_mkwrite callback will do the block allocation and mark the | |
2133 | * buffer_heads mapped. | |
2134 | * | |
2135 | * We redirty the page if we have any buffer_heads that is either delay or | |
2136 | * unwritten in the page. | |
2137 | * | |
2138 | * We can get recursively called as show below. | |
2139 | * | |
2140 | * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() -> | |
2141 | * ext4_writepage() | |
2142 | * | |
2143 | * But since we don't do any block allocation we should not deadlock. | |
2144 | * Page also have the dirty flag cleared so we don't get recurive page_lock. | |
61628a3f | 2145 | */ |
43ce1d23 | 2146 | static int ext4_writepage(struct page *page, |
62e086be | 2147 | struct writeback_control *wbc) |
64769240 | 2148 | { |
f8bec370 | 2149 | int ret = 0; |
61628a3f | 2150 | loff_t size; |
498e5f24 | 2151 | unsigned int len; |
744692dc | 2152 | struct buffer_head *page_bufs = NULL; |
61628a3f | 2153 | struct inode *inode = page->mapping->host; |
36ade451 | 2154 | struct ext4_io_submit io_submit; |
1c8349a1 | 2155 | bool keep_towrite = false; |
61628a3f | 2156 | |
0db1ff22 TT |
2157 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) { |
2158 | ext4_invalidatepage(page, 0, PAGE_SIZE); | |
2159 | unlock_page(page); | |
2160 | return -EIO; | |
2161 | } | |
2162 | ||
a9c667f8 | 2163 | trace_ext4_writepage(page); |
f0e6c985 | 2164 | size = i_size_read(inode); |
09cbfeaf KS |
2165 | if (page->index == size >> PAGE_SHIFT) |
2166 | len = size & ~PAGE_MASK; | |
f0e6c985 | 2167 | else |
09cbfeaf | 2168 | len = PAGE_SIZE; |
64769240 | 2169 | |
a42afc5f | 2170 | page_bufs = page_buffers(page); |
a42afc5f | 2171 | /* |
fe386132 JK |
2172 | * We cannot do block allocation or other extent handling in this |
2173 | * function. If there are buffers needing that, we have to redirty | |
2174 | * the page. But we may reach here when we do a journal commit via | |
2175 | * journal_submit_inode_data_buffers() and in that case we must write | |
2176 | * allocated buffers to achieve data=ordered mode guarantees. | |
cccd147a TT |
2177 | * |
2178 | * Also, if there is only one buffer per page (the fs block | |
2179 | * size == the page size), if one buffer needs block | |
2180 | * allocation or needs to modify the extent tree to clear the | |
2181 | * unwritten flag, we know that the page can't be written at | |
2182 | * all, so we might as well refuse the write immediately. | |
2183 | * Unfortunately if the block size != page size, we can't as | |
2184 | * easily detect this case using ext4_walk_page_buffers(), but | |
2185 | * for the extremely common case, this is an optimization that | |
2186 | * skips a useless round trip through ext4_bio_write_page(). | |
a42afc5f | 2187 | */ |
f19d5870 TM |
2188 | if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL, |
2189 | ext4_bh_delay_or_unwritten)) { | |
f8bec370 | 2190 | redirty_page_for_writepage(wbc, page); |
cccd147a | 2191 | if ((current->flags & PF_MEMALLOC) || |
09cbfeaf | 2192 | (inode->i_sb->s_blocksize == PAGE_SIZE)) { |
fe386132 JK |
2193 | /* |
2194 | * For memory cleaning there's no point in writing only | |
2195 | * some buffers. So just bail out. Warn if we came here | |
2196 | * from direct reclaim. | |
2197 | */ | |
2198 | WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) | |
2199 | == PF_MEMALLOC); | |
f0e6c985 AK |
2200 | unlock_page(page); |
2201 | return 0; | |
2202 | } | |
1c8349a1 | 2203 | keep_towrite = true; |
a42afc5f | 2204 | } |
64769240 | 2205 | |
cb20d518 | 2206 | if (PageChecked(page) && ext4_should_journal_data(inode)) |
43ce1d23 AK |
2207 | /* |
2208 | * It's mmapped pagecache. Add buffers and journal it. There | |
2209 | * doesn't seem much point in redirtying the page here. | |
2210 | */ | |
3f0ca309 | 2211 | return __ext4_journalled_writepage(page, len); |
43ce1d23 | 2212 | |
97a851ed JK |
2213 | ext4_io_submit_init(&io_submit, wbc); |
2214 | io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS); | |
2215 | if (!io_submit.io_end) { | |
2216 | redirty_page_for_writepage(wbc, page); | |
2217 | unlock_page(page); | |
2218 | return -ENOMEM; | |
2219 | } | |
1c8349a1 | 2220 | ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite); |
36ade451 | 2221 | ext4_io_submit(&io_submit); |
97a851ed JK |
2222 | /* Drop io_end reference we got from init */ |
2223 | ext4_put_io_end_defer(io_submit.io_end); | |
64769240 AT |
2224 | return ret; |
2225 | } | |
2226 | ||
5f1132b2 JK |
2227 | static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page) |
2228 | { | |
2229 | int len; | |
a056bdaa | 2230 | loff_t size; |
5f1132b2 JK |
2231 | int err; |
2232 | ||
2233 | BUG_ON(page->index != mpd->first_page); | |
a056bdaa JK |
2234 | clear_page_dirty_for_io(page); |
2235 | /* | |
2236 | * We have to be very careful here! Nothing protects writeback path | |
2237 | * against i_size changes and the page can be writeably mapped into | |
2238 | * page tables. So an application can be growing i_size and writing | |
2239 | * data through mmap while writeback runs. clear_page_dirty_for_io() | |
2240 | * write-protects our page in page tables and the page cannot get | |
2241 | * written to again until we release page lock. So only after | |
2242 | * clear_page_dirty_for_io() we are safe to sample i_size for | |
2243 | * ext4_bio_write_page() to zero-out tail of the written page. We rely | |
2244 | * on the barrier provided by TestClearPageDirty in | |
2245 | * clear_page_dirty_for_io() to make sure i_size is really sampled only | |
2246 | * after page tables are updated. | |
2247 | */ | |
2248 | size = i_size_read(mpd->inode); | |
09cbfeaf KS |
2249 | if (page->index == size >> PAGE_SHIFT) |
2250 | len = size & ~PAGE_MASK; | |
5f1132b2 | 2251 | else |
09cbfeaf | 2252 | len = PAGE_SIZE; |
1c8349a1 | 2253 | err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false); |
5f1132b2 JK |
2254 | if (!err) |
2255 | mpd->wbc->nr_to_write--; | |
2256 | mpd->first_page++; | |
2257 | ||
2258 | return err; | |
2259 | } | |
2260 | ||
4e7ea81d JK |
2261 | #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay)) |
2262 | ||
61628a3f | 2263 | /* |
fffb2739 JK |
2264 | * mballoc gives us at most this number of blocks... |
2265 | * XXX: That seems to be only a limitation of ext4_mb_normalize_request(). | |
70261f56 | 2266 | * The rest of mballoc seems to handle chunks up to full group size. |
61628a3f | 2267 | */ |
fffb2739 | 2268 | #define MAX_WRITEPAGES_EXTENT_LEN 2048 |
525f4ed8 | 2269 | |
4e7ea81d JK |
2270 | /* |
2271 | * mpage_add_bh_to_extent - try to add bh to extent of blocks to map | |
2272 | * | |
2273 | * @mpd - extent of blocks | |
2274 | * @lblk - logical number of the block in the file | |
09930042 | 2275 | * @bh - buffer head we want to add to the extent |
4e7ea81d | 2276 | * |
09930042 JK |
2277 | * The function is used to collect contig. blocks in the same state. If the |
2278 | * buffer doesn't require mapping for writeback and we haven't started the | |
2279 | * extent of buffers to map yet, the function returns 'true' immediately - the | |
2280 | * caller can write the buffer right away. Otherwise the function returns true | |
2281 | * if the block has been added to the extent, false if the block couldn't be | |
2282 | * added. | |
4e7ea81d | 2283 | */ |
09930042 JK |
2284 | static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk, |
2285 | struct buffer_head *bh) | |
4e7ea81d JK |
2286 | { |
2287 | struct ext4_map_blocks *map = &mpd->map; | |
2288 | ||
09930042 JK |
2289 | /* Buffer that doesn't need mapping for writeback? */ |
2290 | if (!buffer_dirty(bh) || !buffer_mapped(bh) || | |
2291 | (!buffer_delay(bh) && !buffer_unwritten(bh))) { | |
2292 | /* So far no extent to map => we write the buffer right away */ | |
2293 | if (map->m_len == 0) | |
2294 | return true; | |
2295 | return false; | |
2296 | } | |
4e7ea81d JK |
2297 | |
2298 | /* First block in the extent? */ | |
2299 | if (map->m_len == 0) { | |
dddbd6ac JK |
2300 | /* We cannot map unless handle is started... */ |
2301 | if (!mpd->do_map) | |
2302 | return false; | |
4e7ea81d JK |
2303 | map->m_lblk = lblk; |
2304 | map->m_len = 1; | |
09930042 JK |
2305 | map->m_flags = bh->b_state & BH_FLAGS; |
2306 | return true; | |
4e7ea81d JK |
2307 | } |
2308 | ||
09930042 JK |
2309 | /* Don't go larger than mballoc is willing to allocate */ |
2310 | if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN) | |
2311 | return false; | |
2312 | ||
4e7ea81d JK |
2313 | /* Can we merge the block to our big extent? */ |
2314 | if (lblk == map->m_lblk + map->m_len && | |
09930042 | 2315 | (bh->b_state & BH_FLAGS) == map->m_flags) { |
4e7ea81d | 2316 | map->m_len++; |
09930042 | 2317 | return true; |
4e7ea81d | 2318 | } |
09930042 | 2319 | return false; |
4e7ea81d JK |
2320 | } |
2321 | ||
5f1132b2 JK |
2322 | /* |
2323 | * mpage_process_page_bufs - submit page buffers for IO or add them to extent | |
2324 | * | |
2325 | * @mpd - extent of blocks for mapping | |
2326 | * @head - the first buffer in the page | |
2327 | * @bh - buffer we should start processing from | |
2328 | * @lblk - logical number of the block in the file corresponding to @bh | |
2329 | * | |
2330 | * Walk through page buffers from @bh upto @head (exclusive) and either submit | |
2331 | * the page for IO if all buffers in this page were mapped and there's no | |
2332 | * accumulated extent of buffers to map or add buffers in the page to the | |
2333 | * extent of buffers to map. The function returns 1 if the caller can continue | |
2334 | * by processing the next page, 0 if it should stop adding buffers to the | |
2335 | * extent to map because we cannot extend it anymore. It can also return value | |
2336 | * < 0 in case of error during IO submission. | |
2337 | */ | |
2338 | static int mpage_process_page_bufs(struct mpage_da_data *mpd, | |
2339 | struct buffer_head *head, | |
2340 | struct buffer_head *bh, | |
2341 | ext4_lblk_t lblk) | |
4e7ea81d JK |
2342 | { |
2343 | struct inode *inode = mpd->inode; | |
5f1132b2 | 2344 | int err; |
93407472 | 2345 | ext4_lblk_t blocks = (i_size_read(inode) + i_blocksize(inode) - 1) |
4e7ea81d JK |
2346 | >> inode->i_blkbits; |
2347 | ||
2348 | do { | |
2349 | BUG_ON(buffer_locked(bh)); | |
2350 | ||
09930042 | 2351 | if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) { |
4e7ea81d JK |
2352 | /* Found extent to map? */ |
2353 | if (mpd->map.m_len) | |
5f1132b2 | 2354 | return 0; |
dddbd6ac JK |
2355 | /* Buffer needs mapping and handle is not started? */ |
2356 | if (!mpd->do_map) | |
2357 | return 0; | |
09930042 | 2358 | /* Everything mapped so far and we hit EOF */ |
5f1132b2 | 2359 | break; |
4e7ea81d | 2360 | } |
4e7ea81d | 2361 | } while (lblk++, (bh = bh->b_this_page) != head); |
5f1132b2 JK |
2362 | /* So far everything mapped? Submit the page for IO. */ |
2363 | if (mpd->map.m_len == 0) { | |
2364 | err = mpage_submit_page(mpd, head->b_page); | |
2365 | if (err < 0) | |
2366 | return err; | |
2367 | } | |
2368 | return lblk < blocks; | |
4e7ea81d JK |
2369 | } |
2370 | ||
2371 | /* | |
2372 | * mpage_map_buffers - update buffers corresponding to changed extent and | |
2373 | * submit fully mapped pages for IO | |
2374 | * | |
2375 | * @mpd - description of extent to map, on return next extent to map | |
2376 | * | |
2377 | * Scan buffers corresponding to changed extent (we expect corresponding pages | |
2378 | * to be already locked) and update buffer state according to new extent state. | |
2379 | * We map delalloc buffers to their physical location, clear unwritten bits, | |
556615dc | 2380 | * and mark buffers as uninit when we perform writes to unwritten extents |
4e7ea81d JK |
2381 | * and do extent conversion after IO is finished. If the last page is not fully |
2382 | * mapped, we update @map to the next extent in the last page that needs | |
2383 | * mapping. Otherwise we submit the page for IO. | |
2384 | */ | |
2385 | static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd) | |
2386 | { | |
2387 | struct pagevec pvec; | |
2388 | int nr_pages, i; | |
2389 | struct inode *inode = mpd->inode; | |
2390 | struct buffer_head *head, *bh; | |
09cbfeaf | 2391 | int bpp_bits = PAGE_SHIFT - inode->i_blkbits; |
4e7ea81d JK |
2392 | pgoff_t start, end; |
2393 | ext4_lblk_t lblk; | |
2394 | sector_t pblock; | |
2395 | int err; | |
2396 | ||
2397 | start = mpd->map.m_lblk >> bpp_bits; | |
2398 | end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits; | |
2399 | lblk = start << bpp_bits; | |
2400 | pblock = mpd->map.m_pblk; | |
2401 | ||
86679820 | 2402 | pagevec_init(&pvec); |
4e7ea81d | 2403 | while (start <= end) { |
2b85a617 | 2404 | nr_pages = pagevec_lookup_range(&pvec, inode->i_mapping, |
397162ff | 2405 | &start, end); |
4e7ea81d JK |
2406 | if (nr_pages == 0) |
2407 | break; | |
2408 | for (i = 0; i < nr_pages; i++) { | |
2409 | struct page *page = pvec.pages[i]; | |
2410 | ||
4e7ea81d JK |
2411 | bh = head = page_buffers(page); |
2412 | do { | |
2413 | if (lblk < mpd->map.m_lblk) | |
2414 | continue; | |
2415 | if (lblk >= mpd->map.m_lblk + mpd->map.m_len) { | |
2416 | /* | |
2417 | * Buffer after end of mapped extent. | |
2418 | * Find next buffer in the page to map. | |
2419 | */ | |
2420 | mpd->map.m_len = 0; | |
2421 | mpd->map.m_flags = 0; | |
5f1132b2 JK |
2422 | /* |
2423 | * FIXME: If dioread_nolock supports | |
2424 | * blocksize < pagesize, we need to make | |
2425 | * sure we add size mapped so far to | |
2426 | * io_end->size as the following call | |
2427 | * can submit the page for IO. | |
2428 | */ | |
2429 | err = mpage_process_page_bufs(mpd, head, | |
2430 | bh, lblk); | |
4e7ea81d | 2431 | pagevec_release(&pvec); |
5f1132b2 JK |
2432 | if (err > 0) |
2433 | err = 0; | |
2434 | return err; | |
4e7ea81d JK |
2435 | } |
2436 | if (buffer_delay(bh)) { | |
2437 | clear_buffer_delay(bh); | |
2438 | bh->b_blocknr = pblock++; | |
2439 | } | |
4e7ea81d | 2440 | clear_buffer_unwritten(bh); |
5f1132b2 | 2441 | } while (lblk++, (bh = bh->b_this_page) != head); |
4e7ea81d JK |
2442 | |
2443 | /* | |
2444 | * FIXME: This is going to break if dioread_nolock | |
2445 | * supports blocksize < pagesize as we will try to | |
2446 | * convert potentially unmapped parts of inode. | |
2447 | */ | |
09cbfeaf | 2448 | mpd->io_submit.io_end->size += PAGE_SIZE; |
4e7ea81d JK |
2449 | /* Page fully mapped - let IO run! */ |
2450 | err = mpage_submit_page(mpd, page); | |
2451 | if (err < 0) { | |
2452 | pagevec_release(&pvec); | |
2453 | return err; | |
2454 | } | |
4e7ea81d JK |
2455 | } |
2456 | pagevec_release(&pvec); | |
2457 | } | |
2458 | /* Extent fully mapped and matches with page boundary. We are done. */ | |
2459 | mpd->map.m_len = 0; | |
2460 | mpd->map.m_flags = 0; | |
2461 | return 0; | |
2462 | } | |
2463 | ||
2464 | static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd) | |
2465 | { | |
2466 | struct inode *inode = mpd->inode; | |
2467 | struct ext4_map_blocks *map = &mpd->map; | |
2468 | int get_blocks_flags; | |
090f32ee | 2469 | int err, dioread_nolock; |
4e7ea81d JK |
2470 | |
2471 | trace_ext4_da_write_pages_extent(inode, map); | |
2472 | /* | |
2473 | * Call ext4_map_blocks() to allocate any delayed allocation blocks, or | |
556615dc | 2474 | * to convert an unwritten extent to be initialized (in the case |
4e7ea81d JK |
2475 | * where we have written into one or more preallocated blocks). It is |
2476 | * possible that we're going to need more metadata blocks than | |
2477 | * previously reserved. However we must not fail because we're in | |
2478 | * writeback and there is nothing we can do about it so it might result | |
2479 | * in data loss. So use reserved blocks to allocate metadata if | |
2480 | * possible. | |
2481 | * | |
754cfed6 TT |
2482 | * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if |
2483 | * the blocks in question are delalloc blocks. This indicates | |
2484 | * that the blocks and quotas has already been checked when | |
2485 | * the data was copied into the page cache. | |
4e7ea81d JK |
2486 | */ |
2487 | get_blocks_flags = EXT4_GET_BLOCKS_CREATE | | |
ee0876bc JK |
2488 | EXT4_GET_BLOCKS_METADATA_NOFAIL | |
2489 | EXT4_GET_BLOCKS_IO_SUBMIT; | |
090f32ee LC |
2490 | dioread_nolock = ext4_should_dioread_nolock(inode); |
2491 | if (dioread_nolock) | |
4e7ea81d JK |
2492 | get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT; |
2493 | if (map->m_flags & (1 << BH_Delay)) | |
2494 | get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE; | |
2495 | ||
2496 | err = ext4_map_blocks(handle, inode, map, get_blocks_flags); | |
2497 | if (err < 0) | |
2498 | return err; | |
090f32ee | 2499 | if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) { |
6b523df4 JK |
2500 | if (!mpd->io_submit.io_end->handle && |
2501 | ext4_handle_valid(handle)) { | |
2502 | mpd->io_submit.io_end->handle = handle->h_rsv_handle; | |
2503 | handle->h_rsv_handle = NULL; | |
2504 | } | |
3613d228 | 2505 | ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end); |
6b523df4 | 2506 | } |
4e7ea81d JK |
2507 | |
2508 | BUG_ON(map->m_len == 0); | |
4e7ea81d JK |
2509 | return 0; |
2510 | } | |
2511 | ||
2512 | /* | |
2513 | * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length | |
2514 | * mpd->len and submit pages underlying it for IO | |
2515 | * | |
2516 | * @handle - handle for journal operations | |
2517 | * @mpd - extent to map | |
7534e854 JK |
2518 | * @give_up_on_write - we set this to true iff there is a fatal error and there |
2519 | * is no hope of writing the data. The caller should discard | |
2520 | * dirty pages to avoid infinite loops. | |
4e7ea81d JK |
2521 | * |
2522 | * The function maps extent starting at mpd->lblk of length mpd->len. If it is | |
2523 | * delayed, blocks are allocated, if it is unwritten, we may need to convert | |
2524 | * them to initialized or split the described range from larger unwritten | |
2525 | * extent. Note that we need not map all the described range since allocation | |
2526 | * can return less blocks or the range is covered by more unwritten extents. We | |
2527 | * cannot map more because we are limited by reserved transaction credits. On | |
2528 | * the other hand we always make sure that the last touched page is fully | |
2529 | * mapped so that it can be written out (and thus forward progress is | |
2530 | * guaranteed). After mapping we submit all mapped pages for IO. | |
2531 | */ | |
2532 | static int mpage_map_and_submit_extent(handle_t *handle, | |
cb530541 TT |
2533 | struct mpage_da_data *mpd, |
2534 | bool *give_up_on_write) | |
4e7ea81d JK |
2535 | { |
2536 | struct inode *inode = mpd->inode; | |
2537 | struct ext4_map_blocks *map = &mpd->map; | |
2538 | int err; | |
2539 | loff_t disksize; | |
6603120e | 2540 | int progress = 0; |
4e7ea81d JK |
2541 | |
2542 | mpd->io_submit.io_end->offset = | |
2543 | ((loff_t)map->m_lblk) << inode->i_blkbits; | |
27d7c4ed | 2544 | do { |
4e7ea81d JK |
2545 | err = mpage_map_one_extent(handle, mpd); |
2546 | if (err < 0) { | |
2547 | struct super_block *sb = inode->i_sb; | |
2548 | ||
0db1ff22 TT |
2549 | if (ext4_forced_shutdown(EXT4_SB(sb)) || |
2550 | EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED) | |
cb530541 | 2551 | goto invalidate_dirty_pages; |
4e7ea81d | 2552 | /* |
cb530541 TT |
2553 | * Let the uper layers retry transient errors. |
2554 | * In the case of ENOSPC, if ext4_count_free_blocks() | |
2555 | * is non-zero, a commit should free up blocks. | |
4e7ea81d | 2556 | */ |
cb530541 | 2557 | if ((err == -ENOMEM) || |
6603120e DM |
2558 | (err == -ENOSPC && ext4_count_free_clusters(sb))) { |
2559 | if (progress) | |
2560 | goto update_disksize; | |
cb530541 | 2561 | return err; |
6603120e | 2562 | } |
cb530541 TT |
2563 | ext4_msg(sb, KERN_CRIT, |
2564 | "Delayed block allocation failed for " | |
2565 | "inode %lu at logical offset %llu with" | |
2566 | " max blocks %u with error %d", | |
2567 | inode->i_ino, | |
2568 | (unsigned long long)map->m_lblk, | |
2569 | (unsigned)map->m_len, -err); | |
2570 | ext4_msg(sb, KERN_CRIT, | |
2571 | "This should not happen!! Data will " | |
2572 | "be lost\n"); | |
2573 | if (err == -ENOSPC) | |
2574 | ext4_print_free_blocks(inode); | |
2575 | invalidate_dirty_pages: | |
2576 | *give_up_on_write = true; | |
4e7ea81d JK |
2577 | return err; |
2578 | } | |
6603120e | 2579 | progress = 1; |
4e7ea81d JK |
2580 | /* |
2581 | * Update buffer state, submit mapped pages, and get us new | |
2582 | * extent to map | |
2583 | */ | |
2584 | err = mpage_map_and_submit_buffers(mpd); | |
2585 | if (err < 0) | |
6603120e | 2586 | goto update_disksize; |
27d7c4ed | 2587 | } while (map->m_len); |
4e7ea81d | 2588 | |
6603120e | 2589 | update_disksize: |
622cad13 TT |
2590 | /* |
2591 | * Update on-disk size after IO is submitted. Races with | |
2592 | * truncate are avoided by checking i_size under i_data_sem. | |
2593 | */ | |
09cbfeaf | 2594 | disksize = ((loff_t)mpd->first_page) << PAGE_SHIFT; |
4e7ea81d JK |
2595 | if (disksize > EXT4_I(inode)->i_disksize) { |
2596 | int err2; | |
622cad13 TT |
2597 | loff_t i_size; |
2598 | ||
2599 | down_write(&EXT4_I(inode)->i_data_sem); | |
2600 | i_size = i_size_read(inode); | |
2601 | if (disksize > i_size) | |
2602 | disksize = i_size; | |
2603 | if (disksize > EXT4_I(inode)->i_disksize) | |
2604 | EXT4_I(inode)->i_disksize = disksize; | |
622cad13 | 2605 | up_write(&EXT4_I(inode)->i_data_sem); |
b907f2d5 | 2606 | err2 = ext4_mark_inode_dirty(handle, inode); |
4e7ea81d JK |
2607 | if (err2) |
2608 | ext4_error(inode->i_sb, | |
2609 | "Failed to mark inode %lu dirty", | |
2610 | inode->i_ino); | |
2611 | if (!err) | |
2612 | err = err2; | |
2613 | } | |
2614 | return err; | |
2615 | } | |
2616 | ||
fffb2739 JK |
2617 | /* |
2618 | * Calculate the total number of credits to reserve for one writepages | |
20970ba6 | 2619 | * iteration. This is called from ext4_writepages(). We map an extent of |
70261f56 | 2620 | * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping |
fffb2739 JK |
2621 | * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN + |
2622 | * bpp - 1 blocks in bpp different extents. | |
2623 | */ | |
525f4ed8 MC |
2624 | static int ext4_da_writepages_trans_blocks(struct inode *inode) |
2625 | { | |
fffb2739 | 2626 | int bpp = ext4_journal_blocks_per_page(inode); |
525f4ed8 | 2627 | |
fffb2739 JK |
2628 | return ext4_meta_trans_blocks(inode, |
2629 | MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp); | |
525f4ed8 | 2630 | } |
61628a3f | 2631 | |
8e48dcfb | 2632 | /* |
4e7ea81d JK |
2633 | * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages |
2634 | * and underlying extent to map | |
2635 | * | |
2636 | * @mpd - where to look for pages | |
2637 | * | |
2638 | * Walk dirty pages in the mapping. If they are fully mapped, submit them for | |
2639 | * IO immediately. When we find a page which isn't mapped we start accumulating | |
2640 | * extent of buffers underlying these pages that needs mapping (formed by | |
2641 | * either delayed or unwritten buffers). We also lock the pages containing | |
2642 | * these buffers. The extent found is returned in @mpd structure (starting at | |
2643 | * mpd->lblk with length mpd->len blocks). | |
2644 | * | |
2645 | * Note that this function can attach bios to one io_end structure which are | |
2646 | * neither logically nor physically contiguous. Although it may seem as an | |
2647 | * unnecessary complication, it is actually inevitable in blocksize < pagesize | |
2648 | * case as we need to track IO to all buffers underlying a page in one io_end. | |
8e48dcfb | 2649 | */ |
4e7ea81d | 2650 | static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd) |
8e48dcfb | 2651 | { |
4e7ea81d JK |
2652 | struct address_space *mapping = mpd->inode->i_mapping; |
2653 | struct pagevec pvec; | |
2654 | unsigned int nr_pages; | |
aeac589a | 2655 | long left = mpd->wbc->nr_to_write; |
4e7ea81d JK |
2656 | pgoff_t index = mpd->first_page; |
2657 | pgoff_t end = mpd->last_page; | |
10bbd235 | 2658 | xa_mark_t tag; |
4e7ea81d JK |
2659 | int i, err = 0; |
2660 | int blkbits = mpd->inode->i_blkbits; | |
2661 | ext4_lblk_t lblk; | |
2662 | struct buffer_head *head; | |
8e48dcfb | 2663 | |
4e7ea81d | 2664 | if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages) |
5b41d924 ES |
2665 | tag = PAGECACHE_TAG_TOWRITE; |
2666 | else | |
2667 | tag = PAGECACHE_TAG_DIRTY; | |
2668 | ||
86679820 | 2669 | pagevec_init(&pvec); |
4e7ea81d JK |
2670 | mpd->map.m_len = 0; |
2671 | mpd->next_page = index; | |
4f01b02c | 2672 | while (index <= end) { |
dc7f3e86 | 2673 | nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end, |
67fd707f | 2674 | tag); |
8e48dcfb | 2675 | if (nr_pages == 0) |
4e7ea81d | 2676 | goto out; |
8e48dcfb TT |
2677 | |
2678 | for (i = 0; i < nr_pages; i++) { | |
2679 | struct page *page = pvec.pages[i]; | |
2680 | ||
aeac589a ML |
2681 | /* |
2682 | * Accumulated enough dirty pages? This doesn't apply | |
2683 | * to WB_SYNC_ALL mode. For integrity sync we have to | |
2684 | * keep going because someone may be concurrently | |
2685 | * dirtying pages, and we might have synced a lot of | |
2686 | * newly appeared dirty pages, but have not synced all | |
2687 | * of the old dirty pages. | |
2688 | */ | |
2689 | if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0) | |
2690 | goto out; | |
2691 | ||
4e7ea81d JK |
2692 | /* If we can't merge this page, we are done. */ |
2693 | if (mpd->map.m_len > 0 && mpd->next_page != page->index) | |
2694 | goto out; | |
78aaced3 | 2695 | |
8e48dcfb | 2696 | lock_page(page); |
8e48dcfb | 2697 | /* |
4e7ea81d JK |
2698 | * If the page is no longer dirty, or its mapping no |
2699 | * longer corresponds to inode we are writing (which | |
2700 | * means it has been truncated or invalidated), or the | |
2701 | * page is already under writeback and we are not doing | |
2702 | * a data integrity writeback, skip the page | |
8e48dcfb | 2703 | */ |
4f01b02c TT |
2704 | if (!PageDirty(page) || |
2705 | (PageWriteback(page) && | |
4e7ea81d | 2706 | (mpd->wbc->sync_mode == WB_SYNC_NONE)) || |
4f01b02c | 2707 | unlikely(page->mapping != mapping)) { |
8e48dcfb TT |
2708 | unlock_page(page); |
2709 | continue; | |
2710 | } | |
2711 | ||
7cb1a535 | 2712 | wait_on_page_writeback(page); |
8e48dcfb | 2713 | BUG_ON(PageWriteback(page)); |
8e48dcfb | 2714 | |
4e7ea81d | 2715 | if (mpd->map.m_len == 0) |
8eb9e5ce | 2716 | mpd->first_page = page->index; |
8eb9e5ce | 2717 | mpd->next_page = page->index + 1; |
f8bec370 | 2718 | /* Add all dirty buffers to mpd */ |
4e7ea81d | 2719 | lblk = ((ext4_lblk_t)page->index) << |
09cbfeaf | 2720 | (PAGE_SHIFT - blkbits); |
f8bec370 | 2721 | head = page_buffers(page); |
5f1132b2 JK |
2722 | err = mpage_process_page_bufs(mpd, head, head, lblk); |
2723 | if (err <= 0) | |
4e7ea81d | 2724 | goto out; |
5f1132b2 | 2725 | err = 0; |
aeac589a | 2726 | left--; |
8e48dcfb TT |
2727 | } |
2728 | pagevec_release(&pvec); | |
2729 | cond_resched(); | |
2730 | } | |
4f01b02c | 2731 | return 0; |
8eb9e5ce TT |
2732 | out: |
2733 | pagevec_release(&pvec); | |
4e7ea81d | 2734 | return err; |
8e48dcfb TT |
2735 | } |
2736 | ||
20970ba6 TT |
2737 | static int ext4_writepages(struct address_space *mapping, |
2738 | struct writeback_control *wbc) | |
64769240 | 2739 | { |
4e7ea81d JK |
2740 | pgoff_t writeback_index = 0; |
2741 | long nr_to_write = wbc->nr_to_write; | |
22208ded | 2742 | int range_whole = 0; |
4e7ea81d | 2743 | int cycled = 1; |
61628a3f | 2744 | handle_t *handle = NULL; |
df22291f | 2745 | struct mpage_da_data mpd; |
5e745b04 | 2746 | struct inode *inode = mapping->host; |
6b523df4 | 2747 | int needed_blocks, rsv_blocks = 0, ret = 0; |
5e745b04 | 2748 | struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb); |
4e7ea81d | 2749 | bool done; |
1bce63d1 | 2750 | struct blk_plug plug; |
cb530541 | 2751 | bool give_up_on_write = false; |
61628a3f | 2752 | |
0db1ff22 TT |
2753 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
2754 | return -EIO; | |
2755 | ||
c8585c6f | 2756 | percpu_down_read(&sbi->s_journal_flag_rwsem); |
20970ba6 | 2757 | trace_ext4_writepages(inode, wbc); |
ba80b101 | 2758 | |
61628a3f MC |
2759 | /* |
2760 | * No pages to write? This is mainly a kludge to avoid starting | |
2761 | * a transaction for special inodes like journal inode on last iput() | |
2762 | * because that could violate lock ordering on umount | |
2763 | */ | |
a1d6cc56 | 2764 | if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) |
bbf023c7 | 2765 | goto out_writepages; |
2a21e37e | 2766 | |
20970ba6 | 2767 | if (ext4_should_journal_data(inode)) { |
043d20d1 | 2768 | ret = generic_writepages(mapping, wbc); |
bbf023c7 | 2769 | goto out_writepages; |
20970ba6 TT |
2770 | } |
2771 | ||
2a21e37e TT |
2772 | /* |
2773 | * If the filesystem has aborted, it is read-only, so return | |
2774 | * right away instead of dumping stack traces later on that | |
2775 | * will obscure the real source of the problem. We test | |
1751e8a6 | 2776 | * EXT4_MF_FS_ABORTED instead of sb->s_flag's SB_RDONLY because |
2a21e37e | 2777 | * the latter could be true if the filesystem is mounted |
20970ba6 | 2778 | * read-only, and in that case, ext4_writepages should |
2a21e37e TT |
2779 | * *never* be called, so if that ever happens, we would want |
2780 | * the stack trace. | |
2781 | */ | |
0db1ff22 TT |
2782 | if (unlikely(ext4_forced_shutdown(EXT4_SB(mapping->host->i_sb)) || |
2783 | sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) { | |
bbf023c7 ML |
2784 | ret = -EROFS; |
2785 | goto out_writepages; | |
2786 | } | |
2a21e37e | 2787 | |
6b523df4 JK |
2788 | if (ext4_should_dioread_nolock(inode)) { |
2789 | /* | |
70261f56 | 2790 | * We may need to convert up to one extent per block in |
6b523df4 JK |
2791 | * the page and we may dirty the inode. |
2792 | */ | |
812c0cab TT |
2793 | rsv_blocks = 1 + ext4_chunk_trans_blocks(inode, |
2794 | PAGE_SIZE >> inode->i_blkbits); | |
6b523df4 JK |
2795 | } |
2796 | ||
4e7ea81d JK |
2797 | /* |
2798 | * If we have inline data and arrive here, it means that | |
2799 | * we will soon create the block for the 1st page, so | |
2800 | * we'd better clear the inline data here. | |
2801 | */ | |
2802 | if (ext4_has_inline_data(inode)) { | |
2803 | /* Just inode will be modified... */ | |
2804 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 1); | |
2805 | if (IS_ERR(handle)) { | |
2806 | ret = PTR_ERR(handle); | |
2807 | goto out_writepages; | |
2808 | } | |
2809 | BUG_ON(ext4_test_inode_state(inode, | |
2810 | EXT4_STATE_MAY_INLINE_DATA)); | |
2811 | ext4_destroy_inline_data(handle, inode); | |
2812 | ext4_journal_stop(handle); | |
2813 | } | |
2814 | ||
22208ded AK |
2815 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
2816 | range_whole = 1; | |
61628a3f | 2817 | |
2acf2c26 | 2818 | if (wbc->range_cyclic) { |
4e7ea81d JK |
2819 | writeback_index = mapping->writeback_index; |
2820 | if (writeback_index) | |
2acf2c26 | 2821 | cycled = 0; |
4e7ea81d JK |
2822 | mpd.first_page = writeback_index; |
2823 | mpd.last_page = -1; | |
5b41d924 | 2824 | } else { |
09cbfeaf KS |
2825 | mpd.first_page = wbc->range_start >> PAGE_SHIFT; |
2826 | mpd.last_page = wbc->range_end >> PAGE_SHIFT; | |
5b41d924 | 2827 | } |
a1d6cc56 | 2828 | |
4e7ea81d JK |
2829 | mpd.inode = inode; |
2830 | mpd.wbc = wbc; | |
2831 | ext4_io_submit_init(&mpd.io_submit, wbc); | |
2acf2c26 | 2832 | retry: |
6e6938b6 | 2833 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
4e7ea81d JK |
2834 | tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page); |
2835 | done = false; | |
1bce63d1 | 2836 | blk_start_plug(&plug); |
dddbd6ac JK |
2837 | |
2838 | /* | |
2839 | * First writeback pages that don't need mapping - we can avoid | |
2840 | * starting a transaction unnecessarily and also avoid being blocked | |
2841 | * in the block layer on device congestion while having transaction | |
2842 | * started. | |
2843 | */ | |
2844 | mpd.do_map = 0; | |
2845 | mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL); | |
2846 | if (!mpd.io_submit.io_end) { | |
2847 | ret = -ENOMEM; | |
2848 | goto unplug; | |
2849 | } | |
2850 | ret = mpage_prepare_extent_to_map(&mpd); | |
a297b2fc XW |
2851 | /* Unlock pages we didn't use */ |
2852 | mpage_release_unused_pages(&mpd, false); | |
dddbd6ac JK |
2853 | /* Submit prepared bio */ |
2854 | ext4_io_submit(&mpd.io_submit); | |
2855 | ext4_put_io_end_defer(mpd.io_submit.io_end); | |
2856 | mpd.io_submit.io_end = NULL; | |
dddbd6ac JK |
2857 | if (ret < 0) |
2858 | goto unplug; | |
2859 | ||
4e7ea81d JK |
2860 | while (!done && mpd.first_page <= mpd.last_page) { |
2861 | /* For each extent of pages we use new io_end */ | |
2862 | mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL); | |
2863 | if (!mpd.io_submit.io_end) { | |
2864 | ret = -ENOMEM; | |
2865 | break; | |
2866 | } | |
a1d6cc56 AK |
2867 | |
2868 | /* | |
4e7ea81d JK |
2869 | * We have two constraints: We find one extent to map and we |
2870 | * must always write out whole page (makes a difference when | |
2871 | * blocksize < pagesize) so that we don't block on IO when we | |
2872 | * try to write out the rest of the page. Journalled mode is | |
2873 | * not supported by delalloc. | |
a1d6cc56 AK |
2874 | */ |
2875 | BUG_ON(ext4_should_journal_data(inode)); | |
525f4ed8 | 2876 | needed_blocks = ext4_da_writepages_trans_blocks(inode); |
a1d6cc56 | 2877 | |
4e7ea81d | 2878 | /* start a new transaction */ |
6b523df4 JK |
2879 | handle = ext4_journal_start_with_reserve(inode, |
2880 | EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks); | |
61628a3f MC |
2881 | if (IS_ERR(handle)) { |
2882 | ret = PTR_ERR(handle); | |
1693918e | 2883 | ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: " |
fbe845dd | 2884 | "%ld pages, ino %lu; err %d", __func__, |
a1d6cc56 | 2885 | wbc->nr_to_write, inode->i_ino, ret); |
4e7ea81d JK |
2886 | /* Release allocated io_end */ |
2887 | ext4_put_io_end(mpd.io_submit.io_end); | |
dddbd6ac | 2888 | mpd.io_submit.io_end = NULL; |
4e7ea81d | 2889 | break; |
61628a3f | 2890 | } |
dddbd6ac | 2891 | mpd.do_map = 1; |
f63e6005 | 2892 | |
4e7ea81d JK |
2893 | trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc); |
2894 | ret = mpage_prepare_extent_to_map(&mpd); | |
2895 | if (!ret) { | |
2896 | if (mpd.map.m_len) | |
cb530541 TT |
2897 | ret = mpage_map_and_submit_extent(handle, &mpd, |
2898 | &give_up_on_write); | |
4e7ea81d JK |
2899 | else { |
2900 | /* | |
2901 | * We scanned the whole range (or exhausted | |
2902 | * nr_to_write), submitted what was mapped and | |
2903 | * didn't find anything needing mapping. We are | |
2904 | * done. | |
2905 | */ | |
2906 | done = true; | |
2907 | } | |
f63e6005 | 2908 | } |
646caa9c JK |
2909 | /* |
2910 | * Caution: If the handle is synchronous, | |
2911 | * ext4_journal_stop() can wait for transaction commit | |
2912 | * to finish which may depend on writeback of pages to | |
2913 | * complete or on page lock to be released. In that | |
2914 | * case, we have to wait until after after we have | |
2915 | * submitted all the IO, released page locks we hold, | |
2916 | * and dropped io_end reference (for extent conversion | |
2917 | * to be able to complete) before stopping the handle. | |
2918 | */ | |
2919 | if (!ext4_handle_valid(handle) || handle->h_sync == 0) { | |
2920 | ext4_journal_stop(handle); | |
2921 | handle = NULL; | |
dddbd6ac | 2922 | mpd.do_map = 0; |
646caa9c | 2923 | } |
4e7ea81d | 2924 | /* Unlock pages we didn't use */ |
cb530541 | 2925 | mpage_release_unused_pages(&mpd, give_up_on_write); |
a297b2fc XW |
2926 | /* Submit prepared bio */ |
2927 | ext4_io_submit(&mpd.io_submit); | |
2928 | ||
646caa9c JK |
2929 | /* |
2930 | * Drop our io_end reference we got from init. We have | |
2931 | * to be careful and use deferred io_end finishing if | |
2932 | * we are still holding the transaction as we can | |
2933 | * release the last reference to io_end which may end | |
2934 | * up doing unwritten extent conversion. | |
2935 | */ | |
2936 | if (handle) { | |
2937 | ext4_put_io_end_defer(mpd.io_submit.io_end); | |
2938 | ext4_journal_stop(handle); | |
2939 | } else | |
2940 | ext4_put_io_end(mpd.io_submit.io_end); | |
dddbd6ac | 2941 | mpd.io_submit.io_end = NULL; |
4e7ea81d JK |
2942 | |
2943 | if (ret == -ENOSPC && sbi->s_journal) { | |
2944 | /* | |
2945 | * Commit the transaction which would | |
22208ded AK |
2946 | * free blocks released in the transaction |
2947 | * and try again | |
2948 | */ | |
df22291f | 2949 | jbd2_journal_force_commit_nested(sbi->s_journal); |
22208ded | 2950 | ret = 0; |
4e7ea81d JK |
2951 | continue; |
2952 | } | |
2953 | /* Fatal error - ENOMEM, EIO... */ | |
2954 | if (ret) | |
61628a3f | 2955 | break; |
a1d6cc56 | 2956 | } |
dddbd6ac | 2957 | unplug: |
1bce63d1 | 2958 | blk_finish_plug(&plug); |
9c12a831 | 2959 | if (!ret && !cycled && wbc->nr_to_write > 0) { |
2acf2c26 | 2960 | cycled = 1; |
4e7ea81d JK |
2961 | mpd.last_page = writeback_index - 1; |
2962 | mpd.first_page = 0; | |
2acf2c26 AK |
2963 | goto retry; |
2964 | } | |
22208ded AK |
2965 | |
2966 | /* Update index */ | |
22208ded AK |
2967 | if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) |
2968 | /* | |
4e7ea81d | 2969 | * Set the writeback_index so that range_cyclic |
22208ded AK |
2970 | * mode will write it back later |
2971 | */ | |
4e7ea81d | 2972 | mapping->writeback_index = mpd.first_page; |
a1d6cc56 | 2973 | |
61628a3f | 2974 | out_writepages: |
20970ba6 TT |
2975 | trace_ext4_writepages_result(inode, wbc, ret, |
2976 | nr_to_write - wbc->nr_to_write); | |
c8585c6f | 2977 | percpu_up_read(&sbi->s_journal_flag_rwsem); |
61628a3f | 2978 | return ret; |
64769240 AT |
2979 | } |
2980 | ||
5f0663bb DW |
2981 | static int ext4_dax_writepages(struct address_space *mapping, |
2982 | struct writeback_control *wbc) | |
2983 | { | |
2984 | int ret; | |
2985 | long nr_to_write = wbc->nr_to_write; | |
2986 | struct inode *inode = mapping->host; | |
2987 | struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb); | |
2988 | ||
2989 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) | |
2990 | return -EIO; | |
2991 | ||
2992 | percpu_down_read(&sbi->s_journal_flag_rwsem); | |
2993 | trace_ext4_writepages(inode, wbc); | |
2994 | ||
2995 | ret = dax_writeback_mapping_range(mapping, inode->i_sb->s_bdev, wbc); | |
2996 | trace_ext4_writepages_result(inode, wbc, ret, | |
2997 | nr_to_write - wbc->nr_to_write); | |
2998 | percpu_up_read(&sbi->s_journal_flag_rwsem); | |
2999 | return ret; | |
3000 | } | |
3001 | ||
79f0be8d AK |
3002 | static int ext4_nonda_switch(struct super_block *sb) |
3003 | { | |
5c1ff336 | 3004 | s64 free_clusters, dirty_clusters; |
79f0be8d AK |
3005 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
3006 | ||
3007 | /* | |
3008 | * switch to non delalloc mode if we are running low | |
3009 | * on free block. The free block accounting via percpu | |
179f7ebf | 3010 | * counters can get slightly wrong with percpu_counter_batch getting |
79f0be8d AK |
3011 | * accumulated on each CPU without updating global counters |
3012 | * Delalloc need an accurate free block accounting. So switch | |
3013 | * to non delalloc when we are near to error range. | |
3014 | */ | |
5c1ff336 EW |
3015 | free_clusters = |
3016 | percpu_counter_read_positive(&sbi->s_freeclusters_counter); | |
3017 | dirty_clusters = | |
3018 | percpu_counter_read_positive(&sbi->s_dirtyclusters_counter); | |
00d4e736 TT |
3019 | /* |
3020 | * Start pushing delalloc when 1/2 of free blocks are dirty. | |
3021 | */ | |
5c1ff336 | 3022 | if (dirty_clusters && (free_clusters < 2 * dirty_clusters)) |
10ee27a0 | 3023 | try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE); |
00d4e736 | 3024 | |
5c1ff336 EW |
3025 | if (2 * free_clusters < 3 * dirty_clusters || |
3026 | free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) { | |
79f0be8d | 3027 | /* |
c8afb446 ES |
3028 | * free block count is less than 150% of dirty blocks |
3029 | * or free blocks is less than watermark | |
79f0be8d AK |
3030 | */ |
3031 | return 1; | |
3032 | } | |
3033 | return 0; | |
3034 | } | |
3035 | ||
0ff8947f ES |
3036 | /* We always reserve for an inode update; the superblock could be there too */ |
3037 | static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len) | |
3038 | { | |
e2b911c5 | 3039 | if (likely(ext4_has_feature_large_file(inode->i_sb))) |
0ff8947f ES |
3040 | return 1; |
3041 | ||
3042 | if (pos + len <= 0x7fffffffULL) | |
3043 | return 1; | |
3044 | ||
3045 | /* We might need to update the superblock to set LARGE_FILE */ | |
3046 | return 2; | |
3047 | } | |
3048 | ||
64769240 | 3049 | static int ext4_da_write_begin(struct file *file, struct address_space *mapping, |
de9a55b8 TT |
3050 | loff_t pos, unsigned len, unsigned flags, |
3051 | struct page **pagep, void **fsdata) | |
64769240 | 3052 | { |
72b8ab9d | 3053 | int ret, retries = 0; |
64769240 AT |
3054 | struct page *page; |
3055 | pgoff_t index; | |
64769240 AT |
3056 | struct inode *inode = mapping->host; |
3057 | handle_t *handle; | |
3058 | ||
0db1ff22 TT |
3059 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
3060 | return -EIO; | |
3061 | ||
09cbfeaf | 3062 | index = pos >> PAGE_SHIFT; |
79f0be8d | 3063 | |
4db0d88e TT |
3064 | if (ext4_nonda_switch(inode->i_sb) || |
3065 | S_ISLNK(inode->i_mode)) { | |
79f0be8d AK |
3066 | *fsdata = (void *)FALL_BACK_TO_NONDELALLOC; |
3067 | return ext4_write_begin(file, mapping, pos, | |
3068 | len, flags, pagep, fsdata); | |
3069 | } | |
3070 | *fsdata = (void *)0; | |
9bffad1e | 3071 | trace_ext4_da_write_begin(inode, pos, len, flags); |
9c3569b5 TM |
3072 | |
3073 | if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { | |
3074 | ret = ext4_da_write_inline_data_begin(mapping, inode, | |
3075 | pos, len, flags, | |
3076 | pagep, fsdata); | |
3077 | if (ret < 0) | |
47564bfb TT |
3078 | return ret; |
3079 | if (ret == 1) | |
3080 | return 0; | |
9c3569b5 TM |
3081 | } |
3082 | ||
47564bfb TT |
3083 | /* |
3084 | * grab_cache_page_write_begin() can take a long time if the | |
3085 | * system is thrashing due to memory pressure, or if the page | |
3086 | * is being written back. So grab it first before we start | |
3087 | * the transaction handle. This also allows us to allocate | |
3088 | * the page (if needed) without using GFP_NOFS. | |
3089 | */ | |
3090 | retry_grab: | |
3091 | page = grab_cache_page_write_begin(mapping, index, flags); | |
3092 | if (!page) | |
3093 | return -ENOMEM; | |
3094 | unlock_page(page); | |
3095 | ||
64769240 AT |
3096 | /* |
3097 | * With delayed allocation, we don't log the i_disksize update | |
3098 | * if there is delayed block allocation. But we still need | |
3099 | * to journalling the i_disksize update if writes to the end | |
3100 | * of file which has an already mapped buffer. | |
3101 | */ | |
47564bfb | 3102 | retry_journal: |
0ff8947f ES |
3103 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
3104 | ext4_da_write_credits(inode, pos, len)); | |
64769240 | 3105 | if (IS_ERR(handle)) { |
09cbfeaf | 3106 | put_page(page); |
47564bfb | 3107 | return PTR_ERR(handle); |
64769240 AT |
3108 | } |
3109 | ||
47564bfb TT |
3110 | lock_page(page); |
3111 | if (page->mapping != mapping) { | |
3112 | /* The page got truncated from under us */ | |
3113 | unlock_page(page); | |
09cbfeaf | 3114 | put_page(page); |
d5a0d4f7 | 3115 | ext4_journal_stop(handle); |
47564bfb | 3116 | goto retry_grab; |
d5a0d4f7 | 3117 | } |
47564bfb | 3118 | /* In case writeback began while the page was unlocked */ |
7afe5aa5 | 3119 | wait_for_stable_page(page); |
64769240 | 3120 | |
643fa961 | 3121 | #ifdef CONFIG_FS_ENCRYPTION |
2058f83a MH |
3122 | ret = ext4_block_write_begin(page, pos, len, |
3123 | ext4_da_get_block_prep); | |
3124 | #else | |
6e1db88d | 3125 | ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep); |
2058f83a | 3126 | #endif |
64769240 AT |
3127 | if (ret < 0) { |
3128 | unlock_page(page); | |
3129 | ext4_journal_stop(handle); | |
ae4d5372 AK |
3130 | /* |
3131 | * block_write_begin may have instantiated a few blocks | |
3132 | * outside i_size. Trim these off again. Don't need | |
3133 | * i_size_read because we hold i_mutex. | |
3134 | */ | |
3135 | if (pos + len > inode->i_size) | |
b9a4207d | 3136 | ext4_truncate_failed_write(inode); |
47564bfb TT |
3137 | |
3138 | if (ret == -ENOSPC && | |
3139 | ext4_should_retry_alloc(inode->i_sb, &retries)) | |
3140 | goto retry_journal; | |
3141 | ||
09cbfeaf | 3142 | put_page(page); |
47564bfb | 3143 | return ret; |
64769240 AT |
3144 | } |
3145 | ||
47564bfb | 3146 | *pagep = page; |
64769240 AT |
3147 | return ret; |
3148 | } | |
3149 | ||
632eaeab MC |
3150 | /* |
3151 | * Check if we should update i_disksize | |
3152 | * when write to the end of file but not require block allocation | |
3153 | */ | |
3154 | static int ext4_da_should_update_i_disksize(struct page *page, | |
de9a55b8 | 3155 | unsigned long offset) |
632eaeab MC |
3156 | { |
3157 | struct buffer_head *bh; | |
3158 | struct inode *inode = page->mapping->host; | |
3159 | unsigned int idx; | |
3160 | int i; | |
3161 | ||
3162 | bh = page_buffers(page); | |
3163 | idx = offset >> inode->i_blkbits; | |
3164 | ||
af5bc92d | 3165 | for (i = 0; i < idx; i++) |
632eaeab MC |
3166 | bh = bh->b_this_page; |
3167 | ||
29fa89d0 | 3168 | if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh)) |
632eaeab MC |
3169 | return 0; |
3170 | return 1; | |
3171 | } | |
3172 | ||
64769240 | 3173 | static int ext4_da_write_end(struct file *file, |
de9a55b8 TT |
3174 | struct address_space *mapping, |
3175 | loff_t pos, unsigned len, unsigned copied, | |
3176 | struct page *page, void *fsdata) | |
64769240 AT |
3177 | { |
3178 | struct inode *inode = mapping->host; | |
3179 | int ret = 0, ret2; | |
3180 | handle_t *handle = ext4_journal_current_handle(); | |
3181 | loff_t new_i_size; | |
632eaeab | 3182 | unsigned long start, end; |
79f0be8d AK |
3183 | int write_mode = (int)(unsigned long)fsdata; |
3184 | ||
74d553aa TT |
3185 | if (write_mode == FALL_BACK_TO_NONDELALLOC) |
3186 | return ext4_write_end(file, mapping, pos, | |
3187 | len, copied, page, fsdata); | |
632eaeab | 3188 | |
9bffad1e | 3189 | trace_ext4_da_write_end(inode, pos, len, copied); |
09cbfeaf | 3190 | start = pos & (PAGE_SIZE - 1); |
af5bc92d | 3191 | end = start + copied - 1; |
64769240 AT |
3192 | |
3193 | /* | |
3194 | * generic_write_end() will run mark_inode_dirty() if i_size | |
3195 | * changes. So let's piggyback the i_disksize mark_inode_dirty | |
3196 | * into that. | |
3197 | */ | |
64769240 | 3198 | new_i_size = pos + copied; |
ea51d132 | 3199 | if (copied && new_i_size > EXT4_I(inode)->i_disksize) { |
9c3569b5 TM |
3200 | if (ext4_has_inline_data(inode) || |
3201 | ext4_da_should_update_i_disksize(page, end)) { | |
ee124d27 | 3202 | ext4_update_i_disksize(inode, new_i_size); |
cf17fea6 AK |
3203 | /* We need to mark inode dirty even if |
3204 | * new_i_size is less that inode->i_size | |
3205 | * bu greater than i_disksize.(hint delalloc) | |
3206 | */ | |
3207 | ext4_mark_inode_dirty(handle, inode); | |
64769240 | 3208 | } |
632eaeab | 3209 | } |
9c3569b5 TM |
3210 | |
3211 | if (write_mode != CONVERT_INLINE_DATA && | |
3212 | ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) && | |
3213 | ext4_has_inline_data(inode)) | |
3214 | ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied, | |
3215 | page); | |
3216 | else | |
3217 | ret2 = generic_write_end(file, mapping, pos, len, copied, | |
64769240 | 3218 | page, fsdata); |
9c3569b5 | 3219 | |
64769240 AT |
3220 | copied = ret2; |
3221 | if (ret2 < 0) | |
3222 | ret = ret2; | |
3223 | ret2 = ext4_journal_stop(handle); | |
3224 | if (!ret) | |
3225 | ret = ret2; | |
3226 | ||
3227 | return ret ? ret : copied; | |
3228 | } | |
3229 | ||
d47992f8 LC |
3230 | static void ext4_da_invalidatepage(struct page *page, unsigned int offset, |
3231 | unsigned int length) | |
64769240 | 3232 | { |
64769240 AT |
3233 | /* |
3234 | * Drop reserved blocks | |
3235 | */ | |
3236 | BUG_ON(!PageLocked(page)); | |
3237 | if (!page_has_buffers(page)) | |
3238 | goto out; | |
3239 | ||
ca99fdd2 | 3240 | ext4_da_page_release_reservation(page, offset, length); |
64769240 AT |
3241 | |
3242 | out: | |
d47992f8 | 3243 | ext4_invalidatepage(page, offset, length); |
64769240 AT |
3244 | |
3245 | return; | |
3246 | } | |
3247 | ||
ccd2506b TT |
3248 | /* |
3249 | * Force all delayed allocation blocks to be allocated for a given inode. | |
3250 | */ | |
3251 | int ext4_alloc_da_blocks(struct inode *inode) | |
3252 | { | |
fb40ba0d TT |
3253 | trace_ext4_alloc_da_blocks(inode); |
3254 | ||
71d4f7d0 | 3255 | if (!EXT4_I(inode)->i_reserved_data_blocks) |
ccd2506b TT |
3256 | return 0; |
3257 | ||
3258 | /* | |
3259 | * We do something simple for now. The filemap_flush() will | |
3260 | * also start triggering a write of the data blocks, which is | |
3261 | * not strictly speaking necessary (and for users of | |
3262 | * laptop_mode, not even desirable). However, to do otherwise | |
3263 | * would require replicating code paths in: | |
de9a55b8 | 3264 | * |
20970ba6 | 3265 | * ext4_writepages() -> |
ccd2506b TT |
3266 | * write_cache_pages() ---> (via passed in callback function) |
3267 | * __mpage_da_writepage() --> | |
3268 | * mpage_add_bh_to_extent() | |
3269 | * mpage_da_map_blocks() | |
3270 | * | |
3271 | * The problem is that write_cache_pages(), located in | |
3272 | * mm/page-writeback.c, marks pages clean in preparation for | |
3273 | * doing I/O, which is not desirable if we're not planning on | |
3274 | * doing I/O at all. | |
3275 | * | |
3276 | * We could call write_cache_pages(), and then redirty all of | |
380cf090 | 3277 | * the pages by calling redirty_page_for_writepage() but that |
ccd2506b TT |
3278 | * would be ugly in the extreme. So instead we would need to |
3279 | * replicate parts of the code in the above functions, | |
25985edc | 3280 | * simplifying them because we wouldn't actually intend to |
ccd2506b TT |
3281 | * write out the pages, but rather only collect contiguous |
3282 | * logical block extents, call the multi-block allocator, and | |
3283 | * then update the buffer heads with the block allocations. | |
de9a55b8 | 3284 | * |
ccd2506b TT |
3285 | * For now, though, we'll cheat by calling filemap_flush(), |
3286 | * which will map the blocks, and start the I/O, but not | |
3287 | * actually wait for the I/O to complete. | |
3288 | */ | |
3289 | return filemap_flush(inode->i_mapping); | |
3290 | } | |
64769240 | 3291 | |
ac27a0ec DK |
3292 | /* |
3293 | * bmap() is special. It gets used by applications such as lilo and by | |
3294 | * the swapper to find the on-disk block of a specific piece of data. | |
3295 | * | |
3296 | * Naturally, this is dangerous if the block concerned is still in the | |
617ba13b | 3297 | * journal. If somebody makes a swapfile on an ext4 data-journaling |
ac27a0ec DK |
3298 | * filesystem and enables swap, then they may get a nasty shock when the |
3299 | * data getting swapped to that swapfile suddenly gets overwritten by | |
3300 | * the original zero's written out previously to the journal and | |
3301 | * awaiting writeback in the kernel's buffer cache. | |
3302 | * | |
3303 | * So, if we see any bmap calls here on a modified, data-journaled file, | |
3304 | * take extra steps to flush any blocks which might be in the cache. | |
3305 | */ | |
617ba13b | 3306 | static sector_t ext4_bmap(struct address_space *mapping, sector_t block) |
ac27a0ec DK |
3307 | { |
3308 | struct inode *inode = mapping->host; | |
3309 | journal_t *journal; | |
3310 | int err; | |
3311 | ||
46c7f254 TM |
3312 | /* |
3313 | * We can get here for an inline file via the FIBMAP ioctl | |
3314 | */ | |
3315 | if (ext4_has_inline_data(inode)) | |
3316 | return 0; | |
3317 | ||
64769240 AT |
3318 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) && |
3319 | test_opt(inode->i_sb, DELALLOC)) { | |
3320 | /* | |
3321 | * With delalloc we want to sync the file | |
3322 | * so that we can make sure we allocate | |
3323 | * blocks for file | |
3324 | */ | |
3325 | filemap_write_and_wait(mapping); | |
3326 | } | |
3327 | ||
19f5fb7a TT |
3328 | if (EXT4_JOURNAL(inode) && |
3329 | ext4_test_inode_state(inode, EXT4_STATE_JDATA)) { | |
ac27a0ec DK |
3330 | /* |
3331 | * This is a REALLY heavyweight approach, but the use of | |
3332 | * bmap on dirty files is expected to be extremely rare: | |
3333 | * only if we run lilo or swapon on a freshly made file | |
3334 | * do we expect this to happen. | |
3335 | * | |
3336 | * (bmap requires CAP_SYS_RAWIO so this does not | |
3337 | * represent an unprivileged user DOS attack --- we'd be | |
3338 | * in trouble if mortal users could trigger this path at | |
3339 | * will.) | |
3340 | * | |
617ba13b | 3341 | * NB. EXT4_STATE_JDATA is not set on files other than |
ac27a0ec DK |
3342 | * regular files. If somebody wants to bmap a directory |
3343 | * or symlink and gets confused because the buffer | |
3344 | * hasn't yet been flushed to disk, they deserve | |
3345 | * everything they get. | |
3346 | */ | |
3347 | ||
19f5fb7a | 3348 | ext4_clear_inode_state(inode, EXT4_STATE_JDATA); |
617ba13b | 3349 | journal = EXT4_JOURNAL(inode); |
dab291af MC |
3350 | jbd2_journal_lock_updates(journal); |
3351 | err = jbd2_journal_flush(journal); | |
3352 | jbd2_journal_unlock_updates(journal); | |
ac27a0ec DK |
3353 | |
3354 | if (err) | |
3355 | return 0; | |
3356 | } | |
3357 | ||
af5bc92d | 3358 | return generic_block_bmap(mapping, block, ext4_get_block); |
ac27a0ec DK |
3359 | } |
3360 | ||
617ba13b | 3361 | static int ext4_readpage(struct file *file, struct page *page) |
ac27a0ec | 3362 | { |
46c7f254 TM |
3363 | int ret = -EAGAIN; |
3364 | struct inode *inode = page->mapping->host; | |
3365 | ||
0562e0ba | 3366 | trace_ext4_readpage(page); |
46c7f254 TM |
3367 | |
3368 | if (ext4_has_inline_data(inode)) | |
3369 | ret = ext4_readpage_inline(inode, page); | |
3370 | ||
3371 | if (ret == -EAGAIN) | |
ac22b46a JA |
3372 | return ext4_mpage_readpages(page->mapping, NULL, page, 1, |
3373 | false); | |
46c7f254 TM |
3374 | |
3375 | return ret; | |
ac27a0ec DK |
3376 | } |
3377 | ||
3378 | static int | |
617ba13b | 3379 | ext4_readpages(struct file *file, struct address_space *mapping, |
ac27a0ec DK |
3380 | struct list_head *pages, unsigned nr_pages) |
3381 | { | |
46c7f254 TM |
3382 | struct inode *inode = mapping->host; |
3383 | ||
3384 | /* If the file has inline data, no need to do readpages. */ | |
3385 | if (ext4_has_inline_data(inode)) | |
3386 | return 0; | |
3387 | ||
ac22b46a | 3388 | return ext4_mpage_readpages(mapping, pages, NULL, nr_pages, true); |
ac27a0ec DK |
3389 | } |
3390 | ||
d47992f8 LC |
3391 | static void ext4_invalidatepage(struct page *page, unsigned int offset, |
3392 | unsigned int length) | |
ac27a0ec | 3393 | { |
ca99fdd2 | 3394 | trace_ext4_invalidatepage(page, offset, length); |
0562e0ba | 3395 | |
4520fb3c JK |
3396 | /* No journalling happens on data buffers when this function is used */ |
3397 | WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page))); | |
3398 | ||
ca99fdd2 | 3399 | block_invalidatepage(page, offset, length); |
4520fb3c JK |
3400 | } |
3401 | ||
53e87268 | 3402 | static int __ext4_journalled_invalidatepage(struct page *page, |
ca99fdd2 LC |
3403 | unsigned int offset, |
3404 | unsigned int length) | |
4520fb3c JK |
3405 | { |
3406 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); | |
3407 | ||
ca99fdd2 | 3408 | trace_ext4_journalled_invalidatepage(page, offset, length); |
4520fb3c | 3409 | |
ac27a0ec DK |
3410 | /* |
3411 | * If it's a full truncate we just forget about the pending dirtying | |
3412 | */ | |
09cbfeaf | 3413 | if (offset == 0 && length == PAGE_SIZE) |
ac27a0ec DK |
3414 | ClearPageChecked(page); |
3415 | ||
ca99fdd2 | 3416 | return jbd2_journal_invalidatepage(journal, page, offset, length); |
53e87268 JK |
3417 | } |
3418 | ||
3419 | /* Wrapper for aops... */ | |
3420 | static void ext4_journalled_invalidatepage(struct page *page, | |
d47992f8 LC |
3421 | unsigned int offset, |
3422 | unsigned int length) | |
53e87268 | 3423 | { |
ca99fdd2 | 3424 | WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0); |
ac27a0ec DK |
3425 | } |
3426 | ||
617ba13b | 3427 | static int ext4_releasepage(struct page *page, gfp_t wait) |
ac27a0ec | 3428 | { |
617ba13b | 3429 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); |
ac27a0ec | 3430 | |
0562e0ba JZ |
3431 | trace_ext4_releasepage(page); |
3432 | ||
e1c36595 JK |
3433 | /* Page has dirty journalled data -> cannot release */ |
3434 | if (PageChecked(page)) | |
ac27a0ec | 3435 | return 0; |
0390131b FM |
3436 | if (journal) |
3437 | return jbd2_journal_try_to_free_buffers(journal, page, wait); | |
3438 | else | |
3439 | return try_to_free_buffers(page); | |
ac27a0ec DK |
3440 | } |
3441 | ||
b8a6176c JK |
3442 | static bool ext4_inode_datasync_dirty(struct inode *inode) |
3443 | { | |
3444 | journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; | |
3445 | ||
3446 | if (journal) | |
3447 | return !jbd2_transaction_committed(journal, | |
3448 | EXT4_I(inode)->i_datasync_tid); | |
3449 | /* Any metadata buffers to write? */ | |
3450 | if (!list_empty(&inode->i_mapping->private_list)) | |
3451 | return true; | |
3452 | return inode->i_state & I_DIRTY_DATASYNC; | |
3453 | } | |
3454 | ||
364443cb JK |
3455 | static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length, |
3456 | unsigned flags, struct iomap *iomap) | |
3457 | { | |
5e405595 | 3458 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
364443cb | 3459 | unsigned int blkbits = inode->i_blkbits; |
bcd8e91f | 3460 | unsigned long first_block, last_block; |
364443cb | 3461 | struct ext4_map_blocks map; |
545052e9 | 3462 | bool delalloc = false; |
364443cb JK |
3463 | int ret; |
3464 | ||
bcd8e91f TT |
3465 | if ((offset >> blkbits) > EXT4_MAX_LOGICAL_BLOCK) |
3466 | return -EINVAL; | |
3467 | first_block = offset >> blkbits; | |
3468 | last_block = min_t(loff_t, (offset + length - 1) >> blkbits, | |
3469 | EXT4_MAX_LOGICAL_BLOCK); | |
7046ae35 AG |
3470 | |
3471 | if (flags & IOMAP_REPORT) { | |
3472 | if (ext4_has_inline_data(inode)) { | |
3473 | ret = ext4_inline_data_iomap(inode, iomap); | |
3474 | if (ret != -EAGAIN) { | |
3475 | if (ret == 0 && offset >= iomap->length) | |
3476 | ret = -ENOENT; | |
3477 | return ret; | |
3478 | } | |
3479 | } | |
3480 | } else { | |
3481 | if (WARN_ON_ONCE(ext4_has_inline_data(inode))) | |
3482 | return -ERANGE; | |
3483 | } | |
364443cb JK |
3484 | |
3485 | map.m_lblk = first_block; | |
3486 | map.m_len = last_block - first_block + 1; | |
3487 | ||
545052e9 | 3488 | if (flags & IOMAP_REPORT) { |
776722e8 | 3489 | ret = ext4_map_blocks(NULL, inode, &map, 0); |
545052e9 CH |
3490 | if (ret < 0) |
3491 | return ret; | |
3492 | ||
3493 | if (ret == 0) { | |
3494 | ext4_lblk_t end = map.m_lblk + map.m_len - 1; | |
3495 | struct extent_status es; | |
3496 | ||
ad431025 EW |
3497 | ext4_es_find_extent_range(inode, &ext4_es_is_delayed, |
3498 | map.m_lblk, end, &es); | |
545052e9 CH |
3499 | |
3500 | if (!es.es_len || es.es_lblk > end) { | |
3501 | /* entire range is a hole */ | |
3502 | } else if (es.es_lblk > map.m_lblk) { | |
3503 | /* range starts with a hole */ | |
3504 | map.m_len = es.es_lblk - map.m_lblk; | |
3505 | } else { | |
3506 | ext4_lblk_t offs = 0; | |
3507 | ||
3508 | if (es.es_lblk < map.m_lblk) | |
3509 | offs = map.m_lblk - es.es_lblk; | |
3510 | map.m_lblk = es.es_lblk + offs; | |
3511 | map.m_len = es.es_len - offs; | |
3512 | delalloc = true; | |
3513 | } | |
3514 | } | |
3515 | } else if (flags & IOMAP_WRITE) { | |
776722e8 JK |
3516 | int dio_credits; |
3517 | handle_t *handle; | |
3518 | int retries = 0; | |
3519 | ||
3520 | /* Trim mapping request to maximum we can map at once for DIO */ | |
3521 | if (map.m_len > DIO_MAX_BLOCKS) | |
3522 | map.m_len = DIO_MAX_BLOCKS; | |
3523 | dio_credits = ext4_chunk_trans_blocks(inode, map.m_len); | |
3524 | retry: | |
3525 | /* | |
3526 | * Either we allocate blocks and then we don't get unwritten | |
3527 | * extent so we have reserved enough credits, or the blocks | |
3528 | * are already allocated and unwritten and in that case | |
3529 | * extent conversion fits in the credits as well. | |
3530 | */ | |
3531 | handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, | |
3532 | dio_credits); | |
3533 | if (IS_ERR(handle)) | |
3534 | return PTR_ERR(handle); | |
3535 | ||
3536 | ret = ext4_map_blocks(handle, inode, &map, | |
776722e8 JK |
3537 | EXT4_GET_BLOCKS_CREATE_ZERO); |
3538 | if (ret < 0) { | |
3539 | ext4_journal_stop(handle); | |
3540 | if (ret == -ENOSPC && | |
3541 | ext4_should_retry_alloc(inode->i_sb, &retries)) | |
3542 | goto retry; | |
3543 | return ret; | |
3544 | } | |
776722e8 JK |
3545 | |
3546 | /* | |
e2ae766c | 3547 | * If we added blocks beyond i_size, we need to make sure they |
776722e8 | 3548 | * will get truncated if we crash before updating i_size in |
e2ae766c JK |
3549 | * ext4_iomap_end(). For faults we don't need to do that (and |
3550 | * even cannot because for orphan list operations inode_lock is | |
3551 | * required) - if we happen to instantiate block beyond i_size, | |
3552 | * it is because we race with truncate which has already added | |
3553 | * the inode to the orphan list. | |
776722e8 | 3554 | */ |
e2ae766c JK |
3555 | if (!(flags & IOMAP_FAULT) && first_block + map.m_len > |
3556 | (i_size_read(inode) + (1 << blkbits) - 1) >> blkbits) { | |
776722e8 JK |
3557 | int err; |
3558 | ||
3559 | err = ext4_orphan_add(handle, inode); | |
3560 | if (err < 0) { | |
3561 | ext4_journal_stop(handle); | |
3562 | return err; | |
3563 | } | |
3564 | } | |
3565 | ext4_journal_stop(handle); | |
545052e9 CH |
3566 | } else { |
3567 | ret = ext4_map_blocks(NULL, inode, &map, 0); | |
3568 | if (ret < 0) | |
3569 | return ret; | |
776722e8 | 3570 | } |
364443cb JK |
3571 | |
3572 | iomap->flags = 0; | |
aaa422c4 | 3573 | if (ext4_inode_datasync_dirty(inode)) |
b8a6176c | 3574 | iomap->flags |= IOMAP_F_DIRTY; |
5e405595 DW |
3575 | iomap->bdev = inode->i_sb->s_bdev; |
3576 | iomap->dax_dev = sbi->s_daxdev; | |
fe23cb65 | 3577 | iomap->offset = (u64)first_block << blkbits; |
545052e9 | 3578 | iomap->length = (u64)map.m_len << blkbits; |
364443cb JK |
3579 | |
3580 | if (ret == 0) { | |
545052e9 | 3581 | iomap->type = delalloc ? IOMAP_DELALLOC : IOMAP_HOLE; |
19fe5f64 | 3582 | iomap->addr = IOMAP_NULL_ADDR; |
364443cb JK |
3583 | } else { |
3584 | if (map.m_flags & EXT4_MAP_MAPPED) { | |
3585 | iomap->type = IOMAP_MAPPED; | |
3586 | } else if (map.m_flags & EXT4_MAP_UNWRITTEN) { | |
3587 | iomap->type = IOMAP_UNWRITTEN; | |
3588 | } else { | |
3589 | WARN_ON_ONCE(1); | |
3590 | return -EIO; | |
3591 | } | |
19fe5f64 | 3592 | iomap->addr = (u64)map.m_pblk << blkbits; |
364443cb JK |
3593 | } |
3594 | ||
3595 | if (map.m_flags & EXT4_MAP_NEW) | |
3596 | iomap->flags |= IOMAP_F_NEW; | |
545052e9 | 3597 | |
364443cb JK |
3598 | return 0; |
3599 | } | |
3600 | ||
776722e8 JK |
3601 | static int ext4_iomap_end(struct inode *inode, loff_t offset, loff_t length, |
3602 | ssize_t written, unsigned flags, struct iomap *iomap) | |
3603 | { | |
3604 | int ret = 0; | |
3605 | handle_t *handle; | |
3606 | int blkbits = inode->i_blkbits; | |
3607 | bool truncate = false; | |
3608 | ||
e2ae766c | 3609 | if (!(flags & IOMAP_WRITE) || (flags & IOMAP_FAULT)) |
776722e8 JK |
3610 | return 0; |
3611 | ||
3612 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 2); | |
3613 | if (IS_ERR(handle)) { | |
3614 | ret = PTR_ERR(handle); | |
3615 | goto orphan_del; | |
3616 | } | |
3617 | if (ext4_update_inode_size(inode, offset + written)) | |
3618 | ext4_mark_inode_dirty(handle, inode); | |
3619 | /* | |
3620 | * We may need to truncate allocated but not written blocks beyond EOF. | |
3621 | */ | |
3622 | if (iomap->offset + iomap->length > | |
3623 | ALIGN(inode->i_size, 1 << blkbits)) { | |
3624 | ext4_lblk_t written_blk, end_blk; | |
3625 | ||
3626 | written_blk = (offset + written) >> blkbits; | |
3627 | end_blk = (offset + length) >> blkbits; | |
3628 | if (written_blk < end_blk && ext4_can_truncate(inode)) | |
3629 | truncate = true; | |
3630 | } | |
3631 | /* | |
3632 | * Remove inode from orphan list if we were extending a inode and | |
3633 | * everything went fine. | |
3634 | */ | |
3635 | if (!truncate && inode->i_nlink && | |
3636 | !list_empty(&EXT4_I(inode)->i_orphan)) | |
3637 | ext4_orphan_del(handle, inode); | |
3638 | ext4_journal_stop(handle); | |
3639 | if (truncate) { | |
3640 | ext4_truncate_failed_write(inode); | |
3641 | orphan_del: | |
3642 | /* | |
3643 | * If truncate failed early the inode might still be on the | |
3644 | * orphan list; we need to make sure the inode is removed from | |
3645 | * the orphan list in that case. | |
3646 | */ | |
3647 | if (inode->i_nlink) | |
3648 | ext4_orphan_del(NULL, inode); | |
3649 | } | |
3650 | return ret; | |
3651 | } | |
3652 | ||
8ff6daa1 | 3653 | const struct iomap_ops ext4_iomap_ops = { |
364443cb | 3654 | .iomap_begin = ext4_iomap_begin, |
776722e8 | 3655 | .iomap_end = ext4_iomap_end, |
364443cb JK |
3656 | }; |
3657 | ||
187372a3 | 3658 | static int ext4_end_io_dio(struct kiocb *iocb, loff_t offset, |
7b7a8665 | 3659 | ssize_t size, void *private) |
4c0425ff | 3660 | { |
109811c2 | 3661 | ext4_io_end_t *io_end = private; |
4c0425ff | 3662 | |
97a851ed | 3663 | /* if not async direct IO just return */ |
7b7a8665 | 3664 | if (!io_end) |
187372a3 | 3665 | return 0; |
4b70df18 | 3666 | |
88635ca2 | 3667 | ext_debug("ext4_end_io_dio(): io_end 0x%p " |
ace36ad4 | 3668 | "for inode %lu, iocb 0x%p, offset %llu, size %zd\n", |
109811c2 | 3669 | io_end, io_end->inode->i_ino, iocb, offset, size); |
8d5d02e6 | 3670 | |
74c66bcb JK |
3671 | /* |
3672 | * Error during AIO DIO. We cannot convert unwritten extents as the | |
3673 | * data was not written. Just clear the unwritten flag and drop io_end. | |
3674 | */ | |
3675 | if (size <= 0) { | |
3676 | ext4_clear_io_unwritten_flag(io_end); | |
3677 | size = 0; | |
3678 | } | |
4c0425ff MC |
3679 | io_end->offset = offset; |
3680 | io_end->size = size; | |
7b7a8665 | 3681 | ext4_put_io_end(io_end); |
187372a3 CH |
3682 | |
3683 | return 0; | |
4c0425ff | 3684 | } |
c7064ef1 | 3685 | |
4c0425ff | 3686 | /* |
914f82a3 JK |
3687 | * Handling of direct IO writes. |
3688 | * | |
3689 | * For ext4 extent files, ext4 will do direct-io write even to holes, | |
4c0425ff MC |
3690 | * preallocated extents, and those write extend the file, no need to |
3691 | * fall back to buffered IO. | |
3692 | * | |
556615dc | 3693 | * For holes, we fallocate those blocks, mark them as unwritten |
69c499d1 | 3694 | * If those blocks were preallocated, we mark sure they are split, but |
556615dc | 3695 | * still keep the range to write as unwritten. |
4c0425ff | 3696 | * |
69c499d1 | 3697 | * The unwritten extents will be converted to written when DIO is completed. |
8d5d02e6 | 3698 | * For async direct IO, since the IO may still pending when return, we |
25985edc | 3699 | * set up an end_io call back function, which will do the conversion |
8d5d02e6 | 3700 | * when async direct IO completed. |
4c0425ff MC |
3701 | * |
3702 | * If the O_DIRECT write will extend the file then add this inode to the | |
3703 | * orphan list. So recovery will truncate it back to the original size | |
3704 | * if the machine crashes during the write. | |
3705 | * | |
3706 | */ | |
0e01df10 | 3707 | static ssize_t ext4_direct_IO_write(struct kiocb *iocb, struct iov_iter *iter) |
4c0425ff MC |
3708 | { |
3709 | struct file *file = iocb->ki_filp; | |
3710 | struct inode *inode = file->f_mapping->host; | |
45d8ec4d | 3711 | struct ext4_inode_info *ei = EXT4_I(inode); |
4c0425ff | 3712 | ssize_t ret; |
c8b8e32d | 3713 | loff_t offset = iocb->ki_pos; |
a6cbcd4a | 3714 | size_t count = iov_iter_count(iter); |
69c499d1 TT |
3715 | int overwrite = 0; |
3716 | get_block_t *get_block_func = NULL; | |
3717 | int dio_flags = 0; | |
4c0425ff | 3718 | loff_t final_size = offset + count; |
914f82a3 JK |
3719 | int orphan = 0; |
3720 | handle_t *handle; | |
729f52c6 | 3721 | |
45d8ec4d | 3722 | if (final_size > inode->i_size || final_size > ei->i_disksize) { |
914f82a3 JK |
3723 | /* Credits for sb + inode write */ |
3724 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 2); | |
3725 | if (IS_ERR(handle)) { | |
3726 | ret = PTR_ERR(handle); | |
3727 | goto out; | |
3728 | } | |
3729 | ret = ext4_orphan_add(handle, inode); | |
3730 | if (ret) { | |
3731 | ext4_journal_stop(handle); | |
3732 | goto out; | |
3733 | } | |
3734 | orphan = 1; | |
73fdad00 | 3735 | ext4_update_i_disksize(inode, inode->i_size); |
914f82a3 JK |
3736 | ext4_journal_stop(handle); |
3737 | } | |
4bd809db | 3738 | |
69c499d1 | 3739 | BUG_ON(iocb->private == NULL); |
4bd809db | 3740 | |
e8340395 JK |
3741 | /* |
3742 | * Make all waiters for direct IO properly wait also for extent | |
3743 | * conversion. This also disallows race between truncate() and | |
3744 | * overwrite DIO as i_dio_count needs to be incremented under i_mutex. | |
3745 | */ | |
914f82a3 | 3746 | inode_dio_begin(inode); |
e8340395 | 3747 | |
69c499d1 TT |
3748 | /* If we do a overwrite dio, i_mutex locking can be released */ |
3749 | overwrite = *((int *)iocb->private); | |
4bd809db | 3750 | |
2dcba478 | 3751 | if (overwrite) |
5955102c | 3752 | inode_unlock(inode); |
8d5d02e6 | 3753 | |
69c499d1 | 3754 | /* |
914f82a3 | 3755 | * For extent mapped files we could direct write to holes and fallocate. |
69c499d1 | 3756 | * |
109811c2 JK |
3757 | * Allocated blocks to fill the hole are marked as unwritten to prevent |
3758 | * parallel buffered read to expose the stale data before DIO complete | |
3759 | * the data IO. | |
69c499d1 | 3760 | * |
109811c2 JK |
3761 | * As to previously fallocated extents, ext4 get_block will just simply |
3762 | * mark the buffer mapped but still keep the extents unwritten. | |
69c499d1 | 3763 | * |
109811c2 JK |
3764 | * For non AIO case, we will convert those unwritten extents to written |
3765 | * after return back from blockdev_direct_IO. That way we save us from | |
3766 | * allocating io_end structure and also the overhead of offloading | |
3767 | * the extent convertion to a workqueue. | |
69c499d1 TT |
3768 | * |
3769 | * For async DIO, the conversion needs to be deferred when the | |
3770 | * IO is completed. The ext4 end_io callback function will be | |
3771 | * called to take care of the conversion work. Here for async | |
3772 | * case, we allocate an io_end structure to hook to the iocb. | |
3773 | */ | |
3774 | iocb->private = NULL; | |
109811c2 | 3775 | if (overwrite) |
705965bd | 3776 | get_block_func = ext4_dio_get_block_overwrite; |
0bd2d5ec | 3777 | else if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) || |
93407472 | 3778 | round_down(offset, i_blocksize(inode)) >= inode->i_size) { |
914f82a3 JK |
3779 | get_block_func = ext4_dio_get_block; |
3780 | dio_flags = DIO_LOCKING | DIO_SKIP_HOLES; | |
3781 | } else if (is_sync_kiocb(iocb)) { | |
109811c2 JK |
3782 | get_block_func = ext4_dio_get_block_unwritten_sync; |
3783 | dio_flags = DIO_LOCKING; | |
69c499d1 | 3784 | } else { |
109811c2 | 3785 | get_block_func = ext4_dio_get_block_unwritten_async; |
69c499d1 TT |
3786 | dio_flags = DIO_LOCKING; |
3787 | } | |
0bd2d5ec JK |
3788 | ret = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev, iter, |
3789 | get_block_func, ext4_end_io_dio, NULL, | |
3790 | dio_flags); | |
69c499d1 | 3791 | |
97a851ed | 3792 | if (ret > 0 && !overwrite && ext4_test_inode_state(inode, |
69c499d1 TT |
3793 | EXT4_STATE_DIO_UNWRITTEN)) { |
3794 | int err; | |
3795 | /* | |
3796 | * for non AIO case, since the IO is already | |
3797 | * completed, we could do the conversion right here | |
3798 | */ | |
6b523df4 | 3799 | err = ext4_convert_unwritten_extents(NULL, inode, |
69c499d1 TT |
3800 | offset, ret); |
3801 | if (err < 0) | |
3802 | ret = err; | |
3803 | ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN); | |
3804 | } | |
4bd809db | 3805 | |
914f82a3 | 3806 | inode_dio_end(inode); |
69c499d1 | 3807 | /* take i_mutex locking again if we do a ovewrite dio */ |
2dcba478 | 3808 | if (overwrite) |
5955102c | 3809 | inode_lock(inode); |
8d5d02e6 | 3810 | |
914f82a3 JK |
3811 | if (ret < 0 && final_size > inode->i_size) |
3812 | ext4_truncate_failed_write(inode); | |
3813 | ||
3814 | /* Handle extending of i_size after direct IO write */ | |
3815 | if (orphan) { | |
3816 | int err; | |
3817 | ||
3818 | /* Credits for sb + inode write */ | |
3819 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 2); | |
3820 | if (IS_ERR(handle)) { | |
abbc3f93 HS |
3821 | /* |
3822 | * We wrote the data but cannot extend | |
3823 | * i_size. Bail out. In async io case, we do | |
3824 | * not return error here because we have | |
3825 | * already submmitted the corresponding | |
3826 | * bio. Returning error here makes the caller | |
3827 | * think that this IO is done and failed | |
3828 | * resulting in race with bio's completion | |
3829 | * handler. | |
3830 | */ | |
3831 | if (!ret) | |
3832 | ret = PTR_ERR(handle); | |
914f82a3 JK |
3833 | if (inode->i_nlink) |
3834 | ext4_orphan_del(NULL, inode); | |
3835 | ||
3836 | goto out; | |
3837 | } | |
3838 | if (inode->i_nlink) | |
3839 | ext4_orphan_del(handle, inode); | |
3840 | if (ret > 0) { | |
3841 | loff_t end = offset + ret; | |
45d8ec4d | 3842 | if (end > inode->i_size || end > ei->i_disksize) { |
73fdad00 | 3843 | ext4_update_i_disksize(inode, end); |
45d8ec4d EG |
3844 | if (end > inode->i_size) |
3845 | i_size_write(inode, end); | |
914f82a3 JK |
3846 | /* |
3847 | * We're going to return a positive `ret' | |
3848 | * here due to non-zero-length I/O, so there's | |
3849 | * no way of reporting error returns from | |
3850 | * ext4_mark_inode_dirty() to userspace. So | |
3851 | * ignore it. | |
3852 | */ | |
3853 | ext4_mark_inode_dirty(handle, inode); | |
3854 | } | |
3855 | } | |
3856 | err = ext4_journal_stop(handle); | |
3857 | if (ret == 0) | |
3858 | ret = err; | |
3859 | } | |
3860 | out: | |
3861 | return ret; | |
3862 | } | |
3863 | ||
0e01df10 | 3864 | static ssize_t ext4_direct_IO_read(struct kiocb *iocb, struct iov_iter *iter) |
914f82a3 | 3865 | { |
16c54688 JK |
3866 | struct address_space *mapping = iocb->ki_filp->f_mapping; |
3867 | struct inode *inode = mapping->host; | |
0bd2d5ec | 3868 | size_t count = iov_iter_count(iter); |
914f82a3 JK |
3869 | ssize_t ret; |
3870 | ||
16c54688 JK |
3871 | /* |
3872 | * Shared inode_lock is enough for us - it protects against concurrent | |
3873 | * writes & truncates and since we take care of writing back page cache, | |
3874 | * we are protected against page writeback as well. | |
3875 | */ | |
3876 | inode_lock_shared(inode); | |
0bd2d5ec | 3877 | ret = filemap_write_and_wait_range(mapping, iocb->ki_pos, |
e5465795 | 3878 | iocb->ki_pos + count - 1); |
0bd2d5ec JK |
3879 | if (ret) |
3880 | goto out_unlock; | |
3881 | ret = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev, | |
3882 | iter, ext4_dio_get_block, NULL, NULL, 0); | |
16c54688 JK |
3883 | out_unlock: |
3884 | inode_unlock_shared(inode); | |
69c499d1 | 3885 | return ret; |
4c0425ff MC |
3886 | } |
3887 | ||
c8b8e32d | 3888 | static ssize_t ext4_direct_IO(struct kiocb *iocb, struct iov_iter *iter) |
4c0425ff MC |
3889 | { |
3890 | struct file *file = iocb->ki_filp; | |
3891 | struct inode *inode = file->f_mapping->host; | |
a6cbcd4a | 3892 | size_t count = iov_iter_count(iter); |
c8b8e32d | 3893 | loff_t offset = iocb->ki_pos; |
0562e0ba | 3894 | ssize_t ret; |
4c0425ff | 3895 | |
643fa961 | 3896 | #ifdef CONFIG_FS_ENCRYPTION |
592ddec7 | 3897 | if (IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode)) |
2058f83a MH |
3898 | return 0; |
3899 | #endif | |
3900 | ||
84ebd795 TT |
3901 | /* |
3902 | * If we are doing data journalling we don't support O_DIRECT | |
3903 | */ | |
3904 | if (ext4_should_journal_data(inode)) | |
3905 | return 0; | |
3906 | ||
46c7f254 TM |
3907 | /* Let buffer I/O handle the inline data case. */ |
3908 | if (ext4_has_inline_data(inode)) | |
3909 | return 0; | |
3910 | ||
6f673763 | 3911 | trace_ext4_direct_IO_enter(inode, offset, count, iov_iter_rw(iter)); |
914f82a3 | 3912 | if (iov_iter_rw(iter) == READ) |
0e01df10 | 3913 | ret = ext4_direct_IO_read(iocb, iter); |
0562e0ba | 3914 | else |
0e01df10 | 3915 | ret = ext4_direct_IO_write(iocb, iter); |
6f673763 | 3916 | trace_ext4_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), ret); |
0562e0ba | 3917 | return ret; |
4c0425ff MC |
3918 | } |
3919 | ||
ac27a0ec | 3920 | /* |
617ba13b | 3921 | * Pages can be marked dirty completely asynchronously from ext4's journalling |
ac27a0ec DK |
3922 | * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do |
3923 | * much here because ->set_page_dirty is called under VFS locks. The page is | |
3924 | * not necessarily locked. | |
3925 | * | |
3926 | * We cannot just dirty the page and leave attached buffers clean, because the | |
3927 | * buffers' dirty state is "definitive". We cannot just set the buffers dirty | |
3928 | * or jbddirty because all the journalling code will explode. | |
3929 | * | |
3930 | * So what we do is to mark the page "pending dirty" and next time writepage | |
3931 | * is called, propagate that into the buffers appropriately. | |
3932 | */ | |
617ba13b | 3933 | static int ext4_journalled_set_page_dirty(struct page *page) |
ac27a0ec DK |
3934 | { |
3935 | SetPageChecked(page); | |
3936 | return __set_page_dirty_nobuffers(page); | |
3937 | } | |
3938 | ||
6dcc693b JK |
3939 | static int ext4_set_page_dirty(struct page *page) |
3940 | { | |
3941 | WARN_ON_ONCE(!PageLocked(page) && !PageDirty(page)); | |
3942 | WARN_ON_ONCE(!page_has_buffers(page)); | |
3943 | return __set_page_dirty_buffers(page); | |
3944 | } | |
3945 | ||
74d553aa | 3946 | static const struct address_space_operations ext4_aops = { |
8ab22b9a HH |
3947 | .readpage = ext4_readpage, |
3948 | .readpages = ext4_readpages, | |
43ce1d23 | 3949 | .writepage = ext4_writepage, |
20970ba6 | 3950 | .writepages = ext4_writepages, |
8ab22b9a | 3951 | .write_begin = ext4_write_begin, |
74d553aa | 3952 | .write_end = ext4_write_end, |
6dcc693b | 3953 | .set_page_dirty = ext4_set_page_dirty, |
8ab22b9a HH |
3954 | .bmap = ext4_bmap, |
3955 | .invalidatepage = ext4_invalidatepage, | |
3956 | .releasepage = ext4_releasepage, | |
3957 | .direct_IO = ext4_direct_IO, | |
3958 | .migratepage = buffer_migrate_page, | |
3959 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 3960 | .error_remove_page = generic_error_remove_page, |
ac27a0ec DK |
3961 | }; |
3962 | ||
617ba13b | 3963 | static const struct address_space_operations ext4_journalled_aops = { |
8ab22b9a HH |
3964 | .readpage = ext4_readpage, |
3965 | .readpages = ext4_readpages, | |
43ce1d23 | 3966 | .writepage = ext4_writepage, |
20970ba6 | 3967 | .writepages = ext4_writepages, |
8ab22b9a HH |
3968 | .write_begin = ext4_write_begin, |
3969 | .write_end = ext4_journalled_write_end, | |
3970 | .set_page_dirty = ext4_journalled_set_page_dirty, | |
3971 | .bmap = ext4_bmap, | |
4520fb3c | 3972 | .invalidatepage = ext4_journalled_invalidatepage, |
8ab22b9a | 3973 | .releasepage = ext4_releasepage, |
84ebd795 | 3974 | .direct_IO = ext4_direct_IO, |
8ab22b9a | 3975 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 3976 | .error_remove_page = generic_error_remove_page, |
ac27a0ec DK |
3977 | }; |
3978 | ||
64769240 | 3979 | static const struct address_space_operations ext4_da_aops = { |
8ab22b9a HH |
3980 | .readpage = ext4_readpage, |
3981 | .readpages = ext4_readpages, | |
43ce1d23 | 3982 | .writepage = ext4_writepage, |
20970ba6 | 3983 | .writepages = ext4_writepages, |
8ab22b9a HH |
3984 | .write_begin = ext4_da_write_begin, |
3985 | .write_end = ext4_da_write_end, | |
6dcc693b | 3986 | .set_page_dirty = ext4_set_page_dirty, |
8ab22b9a HH |
3987 | .bmap = ext4_bmap, |
3988 | .invalidatepage = ext4_da_invalidatepage, | |
3989 | .releasepage = ext4_releasepage, | |
3990 | .direct_IO = ext4_direct_IO, | |
3991 | .migratepage = buffer_migrate_page, | |
3992 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 3993 | .error_remove_page = generic_error_remove_page, |
64769240 AT |
3994 | }; |
3995 | ||
5f0663bb DW |
3996 | static const struct address_space_operations ext4_dax_aops = { |
3997 | .writepages = ext4_dax_writepages, | |
3998 | .direct_IO = noop_direct_IO, | |
3999 | .set_page_dirty = noop_set_page_dirty, | |
94dbb631 | 4000 | .bmap = ext4_bmap, |
5f0663bb DW |
4001 | .invalidatepage = noop_invalidatepage, |
4002 | }; | |
4003 | ||
617ba13b | 4004 | void ext4_set_aops(struct inode *inode) |
ac27a0ec | 4005 | { |
3d2b1582 LC |
4006 | switch (ext4_inode_journal_mode(inode)) { |
4007 | case EXT4_INODE_ORDERED_DATA_MODE: | |
3d2b1582 | 4008 | case EXT4_INODE_WRITEBACK_DATA_MODE: |
3d2b1582 LC |
4009 | break; |
4010 | case EXT4_INODE_JOURNAL_DATA_MODE: | |
617ba13b | 4011 | inode->i_mapping->a_ops = &ext4_journalled_aops; |
74d553aa | 4012 | return; |
3d2b1582 LC |
4013 | default: |
4014 | BUG(); | |
4015 | } | |
5f0663bb DW |
4016 | if (IS_DAX(inode)) |
4017 | inode->i_mapping->a_ops = &ext4_dax_aops; | |
4018 | else if (test_opt(inode->i_sb, DELALLOC)) | |
74d553aa TT |
4019 | inode->i_mapping->a_ops = &ext4_da_aops; |
4020 | else | |
4021 | inode->i_mapping->a_ops = &ext4_aops; | |
ac27a0ec DK |
4022 | } |
4023 | ||
923ae0ff | 4024 | static int __ext4_block_zero_page_range(handle_t *handle, |
d863dc36 LC |
4025 | struct address_space *mapping, loff_t from, loff_t length) |
4026 | { | |
09cbfeaf KS |
4027 | ext4_fsblk_t index = from >> PAGE_SHIFT; |
4028 | unsigned offset = from & (PAGE_SIZE-1); | |
923ae0ff | 4029 | unsigned blocksize, pos; |
d863dc36 LC |
4030 | ext4_lblk_t iblock; |
4031 | struct inode *inode = mapping->host; | |
4032 | struct buffer_head *bh; | |
4033 | struct page *page; | |
4034 | int err = 0; | |
4035 | ||
09cbfeaf | 4036 | page = find_or_create_page(mapping, from >> PAGE_SHIFT, |
c62d2555 | 4037 | mapping_gfp_constraint(mapping, ~__GFP_FS)); |
d863dc36 LC |
4038 | if (!page) |
4039 | return -ENOMEM; | |
4040 | ||
4041 | blocksize = inode->i_sb->s_blocksize; | |
d863dc36 | 4042 | |
09cbfeaf | 4043 | iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits); |
d863dc36 LC |
4044 | |
4045 | if (!page_has_buffers(page)) | |
4046 | create_empty_buffers(page, blocksize, 0); | |
4047 | ||
4048 | /* Find the buffer that contains "offset" */ | |
4049 | bh = page_buffers(page); | |
4050 | pos = blocksize; | |
4051 | while (offset >= pos) { | |
4052 | bh = bh->b_this_page; | |
4053 | iblock++; | |
4054 | pos += blocksize; | |
4055 | } | |
d863dc36 LC |
4056 | if (buffer_freed(bh)) { |
4057 | BUFFER_TRACE(bh, "freed: skip"); | |
4058 | goto unlock; | |
4059 | } | |
d863dc36 LC |
4060 | if (!buffer_mapped(bh)) { |
4061 | BUFFER_TRACE(bh, "unmapped"); | |
4062 | ext4_get_block(inode, iblock, bh, 0); | |
4063 | /* unmapped? It's a hole - nothing to do */ | |
4064 | if (!buffer_mapped(bh)) { | |
4065 | BUFFER_TRACE(bh, "still unmapped"); | |
4066 | goto unlock; | |
4067 | } | |
4068 | } | |
4069 | ||
4070 | /* Ok, it's mapped. Make sure it's up-to-date */ | |
4071 | if (PageUptodate(page)) | |
4072 | set_buffer_uptodate(bh); | |
4073 | ||
4074 | if (!buffer_uptodate(bh)) { | |
4075 | err = -EIO; | |
dfec8a14 | 4076 | ll_rw_block(REQ_OP_READ, 0, 1, &bh); |
d863dc36 LC |
4077 | wait_on_buffer(bh); |
4078 | /* Uhhuh. Read error. Complain and punt. */ | |
4079 | if (!buffer_uptodate(bh)) | |
4080 | goto unlock; | |
592ddec7 | 4081 | if (S_ISREG(inode->i_mode) && IS_ENCRYPTED(inode)) { |
c9c7429c | 4082 | /* We expect the key to be set. */ |
a7550b30 | 4083 | BUG_ON(!fscrypt_has_encryption_key(inode)); |
aa8bc1ac | 4084 | WARN_ON_ONCE(fscrypt_decrypt_pagecache_blocks( |
ec39a368 | 4085 | page, blocksize, bh_offset(bh))); |
c9c7429c | 4086 | } |
d863dc36 | 4087 | } |
d863dc36 LC |
4088 | if (ext4_should_journal_data(inode)) { |
4089 | BUFFER_TRACE(bh, "get write access"); | |
4090 | err = ext4_journal_get_write_access(handle, bh); | |
4091 | if (err) | |
4092 | goto unlock; | |
4093 | } | |
d863dc36 | 4094 | zero_user(page, offset, length); |
d863dc36 LC |
4095 | BUFFER_TRACE(bh, "zeroed end of block"); |
4096 | ||
d863dc36 LC |
4097 | if (ext4_should_journal_data(inode)) { |
4098 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
0713ed0c | 4099 | } else { |
353eefd3 | 4100 | err = 0; |
d863dc36 | 4101 | mark_buffer_dirty(bh); |
3957ef53 | 4102 | if (ext4_should_order_data(inode)) |
73131fbb RZ |
4103 | err = ext4_jbd2_inode_add_write(handle, inode, from, |
4104 | length); | |
0713ed0c | 4105 | } |
d863dc36 LC |
4106 | |
4107 | unlock: | |
4108 | unlock_page(page); | |
09cbfeaf | 4109 | put_page(page); |
d863dc36 LC |
4110 | return err; |
4111 | } | |
4112 | ||
923ae0ff RZ |
4113 | /* |
4114 | * ext4_block_zero_page_range() zeros out a mapping of length 'length' | |
4115 | * starting from file offset 'from'. The range to be zero'd must | |
4116 | * be contained with in one block. If the specified range exceeds | |
4117 | * the end of the block it will be shortened to end of the block | |
4118 | * that cooresponds to 'from' | |
4119 | */ | |
4120 | static int ext4_block_zero_page_range(handle_t *handle, | |
4121 | struct address_space *mapping, loff_t from, loff_t length) | |
4122 | { | |
4123 | struct inode *inode = mapping->host; | |
09cbfeaf | 4124 | unsigned offset = from & (PAGE_SIZE-1); |
923ae0ff RZ |
4125 | unsigned blocksize = inode->i_sb->s_blocksize; |
4126 | unsigned max = blocksize - (offset & (blocksize - 1)); | |
4127 | ||
4128 | /* | |
4129 | * correct length if it does not fall between | |
4130 | * 'from' and the end of the block | |
4131 | */ | |
4132 | if (length > max || length < 0) | |
4133 | length = max; | |
4134 | ||
47e69351 JK |
4135 | if (IS_DAX(inode)) { |
4136 | return iomap_zero_range(inode, from, length, NULL, | |
4137 | &ext4_iomap_ops); | |
4138 | } | |
923ae0ff RZ |
4139 | return __ext4_block_zero_page_range(handle, mapping, from, length); |
4140 | } | |
4141 | ||
94350ab5 MW |
4142 | /* |
4143 | * ext4_block_truncate_page() zeroes out a mapping from file offset `from' | |
4144 | * up to the end of the block which corresponds to `from'. | |
4145 | * This required during truncate. We need to physically zero the tail end | |
4146 | * of that block so it doesn't yield old data if the file is later grown. | |
4147 | */ | |
c197855e | 4148 | static int ext4_block_truncate_page(handle_t *handle, |
94350ab5 MW |
4149 | struct address_space *mapping, loff_t from) |
4150 | { | |
09cbfeaf | 4151 | unsigned offset = from & (PAGE_SIZE-1); |
94350ab5 MW |
4152 | unsigned length; |
4153 | unsigned blocksize; | |
4154 | struct inode *inode = mapping->host; | |
4155 | ||
0d06863f | 4156 | /* If we are processing an encrypted inode during orphan list handling */ |
592ddec7 | 4157 | if (IS_ENCRYPTED(inode) && !fscrypt_has_encryption_key(inode)) |
0d06863f TT |
4158 | return 0; |
4159 | ||
94350ab5 MW |
4160 | blocksize = inode->i_sb->s_blocksize; |
4161 | length = blocksize - (offset & (blocksize - 1)); | |
4162 | ||
4163 | return ext4_block_zero_page_range(handle, mapping, from, length); | |
4164 | } | |
4165 | ||
a87dd18c LC |
4166 | int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode, |
4167 | loff_t lstart, loff_t length) | |
4168 | { | |
4169 | struct super_block *sb = inode->i_sb; | |
4170 | struct address_space *mapping = inode->i_mapping; | |
e1be3a92 | 4171 | unsigned partial_start, partial_end; |
a87dd18c LC |
4172 | ext4_fsblk_t start, end; |
4173 | loff_t byte_end = (lstart + length - 1); | |
4174 | int err = 0; | |
4175 | ||
e1be3a92 LC |
4176 | partial_start = lstart & (sb->s_blocksize - 1); |
4177 | partial_end = byte_end & (sb->s_blocksize - 1); | |
4178 | ||
a87dd18c LC |
4179 | start = lstart >> sb->s_blocksize_bits; |
4180 | end = byte_end >> sb->s_blocksize_bits; | |
4181 | ||
4182 | /* Handle partial zero within the single block */ | |
e1be3a92 LC |
4183 | if (start == end && |
4184 | (partial_start || (partial_end != sb->s_blocksize - 1))) { | |
a87dd18c LC |
4185 | err = ext4_block_zero_page_range(handle, mapping, |
4186 | lstart, length); | |
4187 | return err; | |
4188 | } | |
4189 | /* Handle partial zero out on the start of the range */ | |
e1be3a92 | 4190 | if (partial_start) { |
a87dd18c LC |
4191 | err = ext4_block_zero_page_range(handle, mapping, |
4192 | lstart, sb->s_blocksize); | |
4193 | if (err) | |
4194 | return err; | |
4195 | } | |
4196 | /* Handle partial zero out on the end of the range */ | |
e1be3a92 | 4197 | if (partial_end != sb->s_blocksize - 1) |
a87dd18c | 4198 | err = ext4_block_zero_page_range(handle, mapping, |
e1be3a92 LC |
4199 | byte_end - partial_end, |
4200 | partial_end + 1); | |
a87dd18c LC |
4201 | return err; |
4202 | } | |
4203 | ||
91ef4caf DG |
4204 | int ext4_can_truncate(struct inode *inode) |
4205 | { | |
91ef4caf DG |
4206 | if (S_ISREG(inode->i_mode)) |
4207 | return 1; | |
4208 | if (S_ISDIR(inode->i_mode)) | |
4209 | return 1; | |
4210 | if (S_ISLNK(inode->i_mode)) | |
4211 | return !ext4_inode_is_fast_symlink(inode); | |
4212 | return 0; | |
4213 | } | |
4214 | ||
01127848 JK |
4215 | /* |
4216 | * We have to make sure i_disksize gets properly updated before we truncate | |
4217 | * page cache due to hole punching or zero range. Otherwise i_disksize update | |
4218 | * can get lost as it may have been postponed to submission of writeback but | |
4219 | * that will never happen after we truncate page cache. | |
4220 | */ | |
4221 | int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset, | |
4222 | loff_t len) | |
4223 | { | |
4224 | handle_t *handle; | |
4225 | loff_t size = i_size_read(inode); | |
4226 | ||
5955102c | 4227 | WARN_ON(!inode_is_locked(inode)); |
01127848 JK |
4228 | if (offset > size || offset + len < size) |
4229 | return 0; | |
4230 | ||
4231 | if (EXT4_I(inode)->i_disksize >= size) | |
4232 | return 0; | |
4233 | ||
4234 | handle = ext4_journal_start(inode, EXT4_HT_MISC, 1); | |
4235 | if (IS_ERR(handle)) | |
4236 | return PTR_ERR(handle); | |
4237 | ext4_update_i_disksize(inode, size); | |
4238 | ext4_mark_inode_dirty(handle, inode); | |
4239 | ext4_journal_stop(handle); | |
4240 | ||
4241 | return 0; | |
4242 | } | |
4243 | ||
b1f38217 | 4244 | static void ext4_wait_dax_page(struct ext4_inode_info *ei) |
430657b6 | 4245 | { |
430657b6 RZ |
4246 | up_write(&ei->i_mmap_sem); |
4247 | schedule(); | |
4248 | down_write(&ei->i_mmap_sem); | |
4249 | } | |
4250 | ||
4251 | int ext4_break_layouts(struct inode *inode) | |
4252 | { | |
4253 | struct ext4_inode_info *ei = EXT4_I(inode); | |
4254 | struct page *page; | |
430657b6 RZ |
4255 | int error; |
4256 | ||
4257 | if (WARN_ON_ONCE(!rwsem_is_locked(&ei->i_mmap_sem))) | |
4258 | return -EINVAL; | |
4259 | ||
4260 | do { | |
430657b6 RZ |
4261 | page = dax_layout_busy_page(inode->i_mapping); |
4262 | if (!page) | |
4263 | return 0; | |
4264 | ||
4265 | error = ___wait_var_event(&page->_refcount, | |
4266 | atomic_read(&page->_refcount) == 1, | |
4267 | TASK_INTERRUPTIBLE, 0, 0, | |
b1f38217 RZ |
4268 | ext4_wait_dax_page(ei)); |
4269 | } while (error == 0); | |
430657b6 RZ |
4270 | |
4271 | return error; | |
4272 | } | |
4273 | ||
a4bb6b64 | 4274 | /* |
cca32b7e | 4275 | * ext4_punch_hole: punches a hole in a file by releasing the blocks |
a4bb6b64 AH |
4276 | * associated with the given offset and length |
4277 | * | |
4278 | * @inode: File inode | |
4279 | * @offset: The offset where the hole will begin | |
4280 | * @len: The length of the hole | |
4281 | * | |
4907cb7b | 4282 | * Returns: 0 on success or negative on failure |
a4bb6b64 AH |
4283 | */ |
4284 | ||
aeb2817a | 4285 | int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length) |
a4bb6b64 | 4286 | { |
26a4c0c6 TT |
4287 | struct super_block *sb = inode->i_sb; |
4288 | ext4_lblk_t first_block, stop_block; | |
4289 | struct address_space *mapping = inode->i_mapping; | |
a87dd18c | 4290 | loff_t first_block_offset, last_block_offset; |
26a4c0c6 TT |
4291 | handle_t *handle; |
4292 | unsigned int credits; | |
4293 | int ret = 0; | |
4294 | ||
a4bb6b64 | 4295 | if (!S_ISREG(inode->i_mode)) |
73355192 | 4296 | return -EOPNOTSUPP; |
a4bb6b64 | 4297 | |
b8a86845 | 4298 | trace_ext4_punch_hole(inode, offset, length, 0); |
aaddea81 | 4299 | |
26a4c0c6 TT |
4300 | /* |
4301 | * Write out all dirty pages to avoid race conditions | |
4302 | * Then release them. | |
4303 | */ | |
cca32b7e | 4304 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { |
26a4c0c6 TT |
4305 | ret = filemap_write_and_wait_range(mapping, offset, |
4306 | offset + length - 1); | |
4307 | if (ret) | |
4308 | return ret; | |
4309 | } | |
4310 | ||
5955102c | 4311 | inode_lock(inode); |
9ef06cec | 4312 | |
26a4c0c6 TT |
4313 | /* No need to punch hole beyond i_size */ |
4314 | if (offset >= inode->i_size) | |
4315 | goto out_mutex; | |
4316 | ||
4317 | /* | |
4318 | * If the hole extends beyond i_size, set the hole | |
4319 | * to end after the page that contains i_size | |
4320 | */ | |
4321 | if (offset + length > inode->i_size) { | |
4322 | length = inode->i_size + | |
09cbfeaf | 4323 | PAGE_SIZE - (inode->i_size & (PAGE_SIZE - 1)) - |
26a4c0c6 TT |
4324 | offset; |
4325 | } | |
4326 | ||
a361293f JK |
4327 | if (offset & (sb->s_blocksize - 1) || |
4328 | (offset + length) & (sb->s_blocksize - 1)) { | |
4329 | /* | |
4330 | * Attach jinode to inode for jbd2 if we do any zeroing of | |
4331 | * partial block | |
4332 | */ | |
4333 | ret = ext4_inode_attach_jinode(inode); | |
4334 | if (ret < 0) | |
4335 | goto out_mutex; | |
4336 | ||
4337 | } | |
4338 | ||
ea3d7209 | 4339 | /* Wait all existing dio workers, newcomers will block on i_mutex */ |
ea3d7209 JK |
4340 | inode_dio_wait(inode); |
4341 | ||
4342 | /* | |
4343 | * Prevent page faults from reinstantiating pages we have released from | |
4344 | * page cache. | |
4345 | */ | |
4346 | down_write(&EXT4_I(inode)->i_mmap_sem); | |
430657b6 RZ |
4347 | |
4348 | ret = ext4_break_layouts(inode); | |
4349 | if (ret) | |
4350 | goto out_dio; | |
4351 | ||
a87dd18c LC |
4352 | first_block_offset = round_up(offset, sb->s_blocksize); |
4353 | last_block_offset = round_down((offset + length), sb->s_blocksize) - 1; | |
26a4c0c6 | 4354 | |
a87dd18c | 4355 | /* Now release the pages and zero block aligned part of pages*/ |
01127848 JK |
4356 | if (last_block_offset > first_block_offset) { |
4357 | ret = ext4_update_disksize_before_punch(inode, offset, length); | |
4358 | if (ret) | |
4359 | goto out_dio; | |
a87dd18c LC |
4360 | truncate_pagecache_range(inode, first_block_offset, |
4361 | last_block_offset); | |
01127848 | 4362 | } |
26a4c0c6 TT |
4363 | |
4364 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
4365 | credits = ext4_writepage_trans_blocks(inode); | |
4366 | else | |
4367 | credits = ext4_blocks_for_truncate(inode); | |
4368 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); | |
4369 | if (IS_ERR(handle)) { | |
4370 | ret = PTR_ERR(handle); | |
4371 | ext4_std_error(sb, ret); | |
4372 | goto out_dio; | |
4373 | } | |
4374 | ||
a87dd18c LC |
4375 | ret = ext4_zero_partial_blocks(handle, inode, offset, |
4376 | length); | |
4377 | if (ret) | |
4378 | goto out_stop; | |
26a4c0c6 TT |
4379 | |
4380 | first_block = (offset + sb->s_blocksize - 1) >> | |
4381 | EXT4_BLOCK_SIZE_BITS(sb); | |
4382 | stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb); | |
4383 | ||
eee597ac LC |
4384 | /* If there are blocks to remove, do it */ |
4385 | if (stop_block > first_block) { | |
26a4c0c6 | 4386 | |
eee597ac LC |
4387 | down_write(&EXT4_I(inode)->i_data_sem); |
4388 | ext4_discard_preallocations(inode); | |
26a4c0c6 | 4389 | |
eee597ac LC |
4390 | ret = ext4_es_remove_extent(inode, first_block, |
4391 | stop_block - first_block); | |
4392 | if (ret) { | |
4393 | up_write(&EXT4_I(inode)->i_data_sem); | |
4394 | goto out_stop; | |
4395 | } | |
26a4c0c6 | 4396 | |
eee597ac LC |
4397 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
4398 | ret = ext4_ext_remove_space(inode, first_block, | |
4399 | stop_block - 1); | |
4400 | else | |
4401 | ret = ext4_ind_remove_space(handle, inode, first_block, | |
4402 | stop_block); | |
26a4c0c6 | 4403 | |
eee597ac LC |
4404 | up_write(&EXT4_I(inode)->i_data_sem); |
4405 | } | |
26a4c0c6 TT |
4406 | if (IS_SYNC(inode)) |
4407 | ext4_handle_sync(handle); | |
e251f9bc | 4408 | |
eeca7ea1 | 4409 | inode->i_mtime = inode->i_ctime = current_time(inode); |
26a4c0c6 | 4410 | ext4_mark_inode_dirty(handle, inode); |
67a7d5f5 JK |
4411 | if (ret >= 0) |
4412 | ext4_update_inode_fsync_trans(handle, inode, 1); | |
26a4c0c6 TT |
4413 | out_stop: |
4414 | ext4_journal_stop(handle); | |
4415 | out_dio: | |
ea3d7209 | 4416 | up_write(&EXT4_I(inode)->i_mmap_sem); |
26a4c0c6 | 4417 | out_mutex: |
5955102c | 4418 | inode_unlock(inode); |
26a4c0c6 | 4419 | return ret; |
a4bb6b64 AH |
4420 | } |
4421 | ||
a361293f JK |
4422 | int ext4_inode_attach_jinode(struct inode *inode) |
4423 | { | |
4424 | struct ext4_inode_info *ei = EXT4_I(inode); | |
4425 | struct jbd2_inode *jinode; | |
4426 | ||
4427 | if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal) | |
4428 | return 0; | |
4429 | ||
4430 | jinode = jbd2_alloc_inode(GFP_KERNEL); | |
4431 | spin_lock(&inode->i_lock); | |
4432 | if (!ei->jinode) { | |
4433 | if (!jinode) { | |
4434 | spin_unlock(&inode->i_lock); | |
4435 | return -ENOMEM; | |
4436 | } | |
4437 | ei->jinode = jinode; | |
4438 | jbd2_journal_init_jbd_inode(ei->jinode, inode); | |
4439 | jinode = NULL; | |
4440 | } | |
4441 | spin_unlock(&inode->i_lock); | |
4442 | if (unlikely(jinode != NULL)) | |
4443 | jbd2_free_inode(jinode); | |
4444 | return 0; | |
4445 | } | |
4446 | ||
ac27a0ec | 4447 | /* |
617ba13b | 4448 | * ext4_truncate() |
ac27a0ec | 4449 | * |
617ba13b MC |
4450 | * We block out ext4_get_block() block instantiations across the entire |
4451 | * transaction, and VFS/VM ensures that ext4_truncate() cannot run | |
ac27a0ec DK |
4452 | * simultaneously on behalf of the same inode. |
4453 | * | |
42b2aa86 | 4454 | * As we work through the truncate and commit bits of it to the journal there |
ac27a0ec DK |
4455 | * is one core, guiding principle: the file's tree must always be consistent on |
4456 | * disk. We must be able to restart the truncate after a crash. | |
4457 | * | |
4458 | * The file's tree may be transiently inconsistent in memory (although it | |
4459 | * probably isn't), but whenever we close off and commit a journal transaction, | |
4460 | * the contents of (the filesystem + the journal) must be consistent and | |
4461 | * restartable. It's pretty simple, really: bottom up, right to left (although | |
4462 | * left-to-right works OK too). | |
4463 | * | |
4464 | * Note that at recovery time, journal replay occurs *before* the restart of | |
4465 | * truncate against the orphan inode list. | |
4466 | * | |
4467 | * The committed inode has the new, desired i_size (which is the same as | |
617ba13b | 4468 | * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see |
ac27a0ec | 4469 | * that this inode's truncate did not complete and it will again call |
617ba13b MC |
4470 | * ext4_truncate() to have another go. So there will be instantiated blocks |
4471 | * to the right of the truncation point in a crashed ext4 filesystem. But | |
ac27a0ec | 4472 | * that's fine - as long as they are linked from the inode, the post-crash |
617ba13b | 4473 | * ext4_truncate() run will find them and release them. |
ac27a0ec | 4474 | */ |
2c98eb5e | 4475 | int ext4_truncate(struct inode *inode) |
ac27a0ec | 4476 | { |
819c4920 TT |
4477 | struct ext4_inode_info *ei = EXT4_I(inode); |
4478 | unsigned int credits; | |
2c98eb5e | 4479 | int err = 0; |
819c4920 TT |
4480 | handle_t *handle; |
4481 | struct address_space *mapping = inode->i_mapping; | |
819c4920 | 4482 | |
19b5ef61 TT |
4483 | /* |
4484 | * There is a possibility that we're either freeing the inode | |
e04027e8 | 4485 | * or it's a completely new inode. In those cases we might not |
19b5ef61 TT |
4486 | * have i_mutex locked because it's not necessary. |
4487 | */ | |
4488 | if (!(inode->i_state & (I_NEW|I_FREEING))) | |
5955102c | 4489 | WARN_ON(!inode_is_locked(inode)); |
0562e0ba JZ |
4490 | trace_ext4_truncate_enter(inode); |
4491 | ||
91ef4caf | 4492 | if (!ext4_can_truncate(inode)) |
2c98eb5e | 4493 | return 0; |
ac27a0ec | 4494 | |
12e9b892 | 4495 | ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS); |
c8d46e41 | 4496 | |
5534fb5b | 4497 | if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC)) |
19f5fb7a | 4498 | ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE); |
7d8f9f7d | 4499 | |
aef1c851 TM |
4500 | if (ext4_has_inline_data(inode)) { |
4501 | int has_inline = 1; | |
4502 | ||
01daf945 TT |
4503 | err = ext4_inline_data_truncate(inode, &has_inline); |
4504 | if (err) | |
4505 | return err; | |
aef1c851 | 4506 | if (has_inline) |
2c98eb5e | 4507 | return 0; |
aef1c851 TM |
4508 | } |
4509 | ||
a361293f JK |
4510 | /* If we zero-out tail of the page, we have to create jinode for jbd2 */ |
4511 | if (inode->i_size & (inode->i_sb->s_blocksize - 1)) { | |
4512 | if (ext4_inode_attach_jinode(inode) < 0) | |
2c98eb5e | 4513 | return 0; |
a361293f JK |
4514 | } |
4515 | ||
819c4920 TT |
4516 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
4517 | credits = ext4_writepage_trans_blocks(inode); | |
4518 | else | |
4519 | credits = ext4_blocks_for_truncate(inode); | |
4520 | ||
4521 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); | |
2c98eb5e TT |
4522 | if (IS_ERR(handle)) |
4523 | return PTR_ERR(handle); | |
819c4920 | 4524 | |
eb3544c6 LC |
4525 | if (inode->i_size & (inode->i_sb->s_blocksize - 1)) |
4526 | ext4_block_truncate_page(handle, mapping, inode->i_size); | |
819c4920 TT |
4527 | |
4528 | /* | |
4529 | * We add the inode to the orphan list, so that if this | |
4530 | * truncate spans multiple transactions, and we crash, we will | |
4531 | * resume the truncate when the filesystem recovers. It also | |
4532 | * marks the inode dirty, to catch the new size. | |
4533 | * | |
4534 | * Implication: the file must always be in a sane, consistent | |
4535 | * truncatable state while each transaction commits. | |
4536 | */ | |
2c98eb5e TT |
4537 | err = ext4_orphan_add(handle, inode); |
4538 | if (err) | |
819c4920 TT |
4539 | goto out_stop; |
4540 | ||
4541 | down_write(&EXT4_I(inode)->i_data_sem); | |
4542 | ||
4543 | ext4_discard_preallocations(inode); | |
4544 | ||
ff9893dc | 4545 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
d0abb36d | 4546 | err = ext4_ext_truncate(handle, inode); |
ff9893dc | 4547 | else |
819c4920 TT |
4548 | ext4_ind_truncate(handle, inode); |
4549 | ||
4550 | up_write(&ei->i_data_sem); | |
d0abb36d TT |
4551 | if (err) |
4552 | goto out_stop; | |
819c4920 TT |
4553 | |
4554 | if (IS_SYNC(inode)) | |
4555 | ext4_handle_sync(handle); | |
4556 | ||
4557 | out_stop: | |
4558 | /* | |
4559 | * If this was a simple ftruncate() and the file will remain alive, | |
4560 | * then we need to clear up the orphan record which we created above. | |
4561 | * However, if this was a real unlink then we were called by | |
58d86a50 | 4562 | * ext4_evict_inode(), and we allow that function to clean up the |
819c4920 TT |
4563 | * orphan info for us. |
4564 | */ | |
4565 | if (inode->i_nlink) | |
4566 | ext4_orphan_del(handle, inode); | |
4567 | ||
eeca7ea1 | 4568 | inode->i_mtime = inode->i_ctime = current_time(inode); |
819c4920 TT |
4569 | ext4_mark_inode_dirty(handle, inode); |
4570 | ext4_journal_stop(handle); | |
ac27a0ec | 4571 | |
0562e0ba | 4572 | trace_ext4_truncate_exit(inode); |
2c98eb5e | 4573 | return err; |
ac27a0ec DK |
4574 | } |
4575 | ||
ac27a0ec | 4576 | /* |
617ba13b | 4577 | * ext4_get_inode_loc returns with an extra refcount against the inode's |
ac27a0ec DK |
4578 | * underlying buffer_head on success. If 'in_mem' is true, we have all |
4579 | * data in memory that is needed to recreate the on-disk version of this | |
4580 | * inode. | |
4581 | */ | |
617ba13b MC |
4582 | static int __ext4_get_inode_loc(struct inode *inode, |
4583 | struct ext4_iloc *iloc, int in_mem) | |
ac27a0ec | 4584 | { |
240799cd TT |
4585 | struct ext4_group_desc *gdp; |
4586 | struct buffer_head *bh; | |
4587 | struct super_block *sb = inode->i_sb; | |
4588 | ext4_fsblk_t block; | |
b03755ad | 4589 | struct blk_plug plug; |
240799cd TT |
4590 | int inodes_per_block, inode_offset; |
4591 | ||
3a06d778 | 4592 | iloc->bh = NULL; |
c37e9e01 TT |
4593 | if (inode->i_ino < EXT4_ROOT_INO || |
4594 | inode->i_ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count)) | |
6a797d27 | 4595 | return -EFSCORRUPTED; |
ac27a0ec | 4596 | |
240799cd TT |
4597 | iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb); |
4598 | gdp = ext4_get_group_desc(sb, iloc->block_group, NULL); | |
4599 | if (!gdp) | |
ac27a0ec DK |
4600 | return -EIO; |
4601 | ||
240799cd TT |
4602 | /* |
4603 | * Figure out the offset within the block group inode table | |
4604 | */ | |
00d09882 | 4605 | inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; |
240799cd TT |
4606 | inode_offset = ((inode->i_ino - 1) % |
4607 | EXT4_INODES_PER_GROUP(sb)); | |
4608 | block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block); | |
4609 | iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb); | |
4610 | ||
4611 | bh = sb_getblk(sb, block); | |
aebf0243 | 4612 | if (unlikely(!bh)) |
860d21e2 | 4613 | return -ENOMEM; |
ac27a0ec DK |
4614 | if (!buffer_uptodate(bh)) { |
4615 | lock_buffer(bh); | |
9c83a923 HK |
4616 | |
4617 | /* | |
4618 | * If the buffer has the write error flag, we have failed | |
4619 | * to write out another inode in the same block. In this | |
4620 | * case, we don't have to read the block because we may | |
4621 | * read the old inode data successfully. | |
4622 | */ | |
4623 | if (buffer_write_io_error(bh) && !buffer_uptodate(bh)) | |
4624 | set_buffer_uptodate(bh); | |
4625 | ||
ac27a0ec DK |
4626 | if (buffer_uptodate(bh)) { |
4627 | /* someone brought it uptodate while we waited */ | |
4628 | unlock_buffer(bh); | |
4629 | goto has_buffer; | |
4630 | } | |
4631 | ||
4632 | /* | |
4633 | * If we have all information of the inode in memory and this | |
4634 | * is the only valid inode in the block, we need not read the | |
4635 | * block. | |
4636 | */ | |
4637 | if (in_mem) { | |
4638 | struct buffer_head *bitmap_bh; | |
240799cd | 4639 | int i, start; |
ac27a0ec | 4640 | |
240799cd | 4641 | start = inode_offset & ~(inodes_per_block - 1); |
ac27a0ec | 4642 | |
240799cd TT |
4643 | /* Is the inode bitmap in cache? */ |
4644 | bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp)); | |
aebf0243 | 4645 | if (unlikely(!bitmap_bh)) |
ac27a0ec DK |
4646 | goto make_io; |
4647 | ||
4648 | /* | |
4649 | * If the inode bitmap isn't in cache then the | |
4650 | * optimisation may end up performing two reads instead | |
4651 | * of one, so skip it. | |
4652 | */ | |
4653 | if (!buffer_uptodate(bitmap_bh)) { | |
4654 | brelse(bitmap_bh); | |
4655 | goto make_io; | |
4656 | } | |
240799cd | 4657 | for (i = start; i < start + inodes_per_block; i++) { |
ac27a0ec DK |
4658 | if (i == inode_offset) |
4659 | continue; | |
617ba13b | 4660 | if (ext4_test_bit(i, bitmap_bh->b_data)) |
ac27a0ec DK |
4661 | break; |
4662 | } | |
4663 | brelse(bitmap_bh); | |
240799cd | 4664 | if (i == start + inodes_per_block) { |
ac27a0ec DK |
4665 | /* all other inodes are free, so skip I/O */ |
4666 | memset(bh->b_data, 0, bh->b_size); | |
4667 | set_buffer_uptodate(bh); | |
4668 | unlock_buffer(bh); | |
4669 | goto has_buffer; | |
4670 | } | |
4671 | } | |
4672 | ||
4673 | make_io: | |
240799cd TT |
4674 | /* |
4675 | * If we need to do any I/O, try to pre-readahead extra | |
4676 | * blocks from the inode table. | |
4677 | */ | |
b03755ad | 4678 | blk_start_plug(&plug); |
240799cd TT |
4679 | if (EXT4_SB(sb)->s_inode_readahead_blks) { |
4680 | ext4_fsblk_t b, end, table; | |
4681 | unsigned num; | |
0d606e2c | 4682 | __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks; |
240799cd TT |
4683 | |
4684 | table = ext4_inode_table(sb, gdp); | |
b713a5ec | 4685 | /* s_inode_readahead_blks is always a power of 2 */ |
0d606e2c | 4686 | b = block & ~((ext4_fsblk_t) ra_blks - 1); |
240799cd TT |
4687 | if (table > b) |
4688 | b = table; | |
0d606e2c | 4689 | end = b + ra_blks; |
240799cd | 4690 | num = EXT4_INODES_PER_GROUP(sb); |
feb0ab32 | 4691 | if (ext4_has_group_desc_csum(sb)) |
560671a0 | 4692 | num -= ext4_itable_unused_count(sb, gdp); |
240799cd TT |
4693 | table += num / inodes_per_block; |
4694 | if (end > table) | |
4695 | end = table; | |
4696 | while (b <= end) | |
4697 | sb_breadahead(sb, b++); | |
4698 | } | |
4699 | ||
ac27a0ec DK |
4700 | /* |
4701 | * There are other valid inodes in the buffer, this inode | |
4702 | * has in-inode xattrs, or we don't have this inode in memory. | |
4703 | * Read the block from disk. | |
4704 | */ | |
0562e0ba | 4705 | trace_ext4_load_inode(inode); |
ac27a0ec DK |
4706 | get_bh(bh); |
4707 | bh->b_end_io = end_buffer_read_sync; | |
2a222ca9 | 4708 | submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh); |
b03755ad | 4709 | blk_finish_plug(&plug); |
ac27a0ec DK |
4710 | wait_on_buffer(bh); |
4711 | if (!buffer_uptodate(bh)) { | |
c398eda0 TT |
4712 | EXT4_ERROR_INODE_BLOCK(inode, block, |
4713 | "unable to read itable block"); | |
ac27a0ec DK |
4714 | brelse(bh); |
4715 | return -EIO; | |
4716 | } | |
4717 | } | |
4718 | has_buffer: | |
4719 | iloc->bh = bh; | |
4720 | return 0; | |
4721 | } | |
4722 | ||
617ba13b | 4723 | int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec DK |
4724 | { |
4725 | /* We have all inode data except xattrs in memory here. */ | |
617ba13b | 4726 | return __ext4_get_inode_loc(inode, iloc, |
19f5fb7a | 4727 | !ext4_test_inode_state(inode, EXT4_STATE_XATTR)); |
ac27a0ec DK |
4728 | } |
4729 | ||
6642586b RZ |
4730 | static bool ext4_should_use_dax(struct inode *inode) |
4731 | { | |
4732 | if (!test_opt(inode->i_sb, DAX)) | |
4733 | return false; | |
4734 | if (!S_ISREG(inode->i_mode)) | |
4735 | return false; | |
4736 | if (ext4_should_journal_data(inode)) | |
4737 | return false; | |
4738 | if (ext4_has_inline_data(inode)) | |
4739 | return false; | |
592ddec7 | 4740 | if (ext4_test_inode_flag(inode, EXT4_INODE_ENCRYPT)) |
6642586b RZ |
4741 | return false; |
4742 | return true; | |
4743 | } | |
4744 | ||
617ba13b | 4745 | void ext4_set_inode_flags(struct inode *inode) |
ac27a0ec | 4746 | { |
617ba13b | 4747 | unsigned int flags = EXT4_I(inode)->i_flags; |
00a1a053 | 4748 | unsigned int new_fl = 0; |
ac27a0ec | 4749 | |
617ba13b | 4750 | if (flags & EXT4_SYNC_FL) |
00a1a053 | 4751 | new_fl |= S_SYNC; |
617ba13b | 4752 | if (flags & EXT4_APPEND_FL) |
00a1a053 | 4753 | new_fl |= S_APPEND; |
617ba13b | 4754 | if (flags & EXT4_IMMUTABLE_FL) |
00a1a053 | 4755 | new_fl |= S_IMMUTABLE; |
617ba13b | 4756 | if (flags & EXT4_NOATIME_FL) |
00a1a053 | 4757 | new_fl |= S_NOATIME; |
617ba13b | 4758 | if (flags & EXT4_DIRSYNC_FL) |
00a1a053 | 4759 | new_fl |= S_DIRSYNC; |
6642586b | 4760 | if (ext4_should_use_dax(inode)) |
923ae0ff | 4761 | new_fl |= S_DAX; |
2ee6a576 EB |
4762 | if (flags & EXT4_ENCRYPT_FL) |
4763 | new_fl |= S_ENCRYPTED; | |
b886ee3e GKB |
4764 | if (flags & EXT4_CASEFOLD_FL) |
4765 | new_fl |= S_CASEFOLD; | |
5f16f322 | 4766 | inode_set_flags(inode, new_fl, |
2ee6a576 | 4767 | S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX| |
b886ee3e | 4768 | S_ENCRYPTED|S_CASEFOLD); |
ac27a0ec DK |
4769 | } |
4770 | ||
0fc1b451 | 4771 | static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode, |
de9a55b8 | 4772 | struct ext4_inode_info *ei) |
0fc1b451 AK |
4773 | { |
4774 | blkcnt_t i_blocks ; | |
8180a562 AK |
4775 | struct inode *inode = &(ei->vfs_inode); |
4776 | struct super_block *sb = inode->i_sb; | |
0fc1b451 | 4777 | |
e2b911c5 | 4778 | if (ext4_has_feature_huge_file(sb)) { |
0fc1b451 AK |
4779 | /* we are using combined 48 bit field */ |
4780 | i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 | | |
4781 | le32_to_cpu(raw_inode->i_blocks_lo); | |
07a03824 | 4782 | if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) { |
8180a562 AK |
4783 | /* i_blocks represent file system block size */ |
4784 | return i_blocks << (inode->i_blkbits - 9); | |
4785 | } else { | |
4786 | return i_blocks; | |
4787 | } | |
0fc1b451 AK |
4788 | } else { |
4789 | return le32_to_cpu(raw_inode->i_blocks_lo); | |
4790 | } | |
4791 | } | |
ff9ddf7e | 4792 | |
eb9b5f01 | 4793 | static inline int ext4_iget_extra_inode(struct inode *inode, |
152a7b0a TM |
4794 | struct ext4_inode *raw_inode, |
4795 | struct ext4_inode_info *ei) | |
4796 | { | |
4797 | __le32 *magic = (void *)raw_inode + | |
4798 | EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize; | |
eb9b5f01 | 4799 | |
290ab230 EB |
4800 | if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize + sizeof(__le32) <= |
4801 | EXT4_INODE_SIZE(inode->i_sb) && | |
4802 | *magic == cpu_to_le32(EXT4_XATTR_MAGIC)) { | |
152a7b0a | 4803 | ext4_set_inode_state(inode, EXT4_STATE_XATTR); |
eb9b5f01 | 4804 | return ext4_find_inline_data_nolock(inode); |
f19d5870 TM |
4805 | } else |
4806 | EXT4_I(inode)->i_inline_off = 0; | |
eb9b5f01 | 4807 | return 0; |
152a7b0a TM |
4808 | } |
4809 | ||
040cb378 LX |
4810 | int ext4_get_projid(struct inode *inode, kprojid_t *projid) |
4811 | { | |
0b7b7779 | 4812 | if (!ext4_has_feature_project(inode->i_sb)) |
040cb378 LX |
4813 | return -EOPNOTSUPP; |
4814 | *projid = EXT4_I(inode)->i_projid; | |
4815 | return 0; | |
4816 | } | |
4817 | ||
e254d1af EG |
4818 | /* |
4819 | * ext4 has self-managed i_version for ea inodes, it stores the lower 32bit of | |
4820 | * refcount in i_version, so use raw values if inode has EXT4_EA_INODE_FL flag | |
4821 | * set. | |
4822 | */ | |
4823 | static inline void ext4_inode_set_iversion_queried(struct inode *inode, u64 val) | |
4824 | { | |
4825 | if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) | |
4826 | inode_set_iversion_raw(inode, val); | |
4827 | else | |
4828 | inode_set_iversion_queried(inode, val); | |
4829 | } | |
4830 | static inline u64 ext4_inode_peek_iversion(const struct inode *inode) | |
4831 | { | |
4832 | if (unlikely(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) | |
4833 | return inode_peek_iversion_raw(inode); | |
4834 | else | |
4835 | return inode_peek_iversion(inode); | |
4836 | } | |
4837 | ||
8a363970 TT |
4838 | struct inode *__ext4_iget(struct super_block *sb, unsigned long ino, |
4839 | ext4_iget_flags flags, const char *function, | |
4840 | unsigned int line) | |
ac27a0ec | 4841 | { |
617ba13b MC |
4842 | struct ext4_iloc iloc; |
4843 | struct ext4_inode *raw_inode; | |
1d1fe1ee | 4844 | struct ext4_inode_info *ei; |
1d1fe1ee | 4845 | struct inode *inode; |
b436b9be | 4846 | journal_t *journal = EXT4_SB(sb)->s_journal; |
1d1fe1ee | 4847 | long ret; |
7e6e1ef4 | 4848 | loff_t size; |
ac27a0ec | 4849 | int block; |
08cefc7a EB |
4850 | uid_t i_uid; |
4851 | gid_t i_gid; | |
040cb378 | 4852 | projid_t i_projid; |
ac27a0ec | 4853 | |
191ce178 | 4854 | if ((!(flags & EXT4_IGET_SPECIAL) && |
8a363970 TT |
4855 | (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)) || |
4856 | (ino < EXT4_ROOT_INO) || | |
4857 | (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))) { | |
4858 | if (flags & EXT4_IGET_HANDLE) | |
4859 | return ERR_PTR(-ESTALE); | |
4860 | __ext4_error(sb, function, line, | |
4861 | "inode #%lu: comm %s: iget: illegal inode #", | |
4862 | ino, current->comm); | |
4863 | return ERR_PTR(-EFSCORRUPTED); | |
4864 | } | |
4865 | ||
1d1fe1ee DH |
4866 | inode = iget_locked(sb, ino); |
4867 | if (!inode) | |
4868 | return ERR_PTR(-ENOMEM); | |
4869 | if (!(inode->i_state & I_NEW)) | |
4870 | return inode; | |
4871 | ||
4872 | ei = EXT4_I(inode); | |
7dc57615 | 4873 | iloc.bh = NULL; |
ac27a0ec | 4874 | |
1d1fe1ee DH |
4875 | ret = __ext4_get_inode_loc(inode, &iloc, 0); |
4876 | if (ret < 0) | |
ac27a0ec | 4877 | goto bad_inode; |
617ba13b | 4878 | raw_inode = ext4_raw_inode(&iloc); |
814525f4 | 4879 | |
8e4b5eae | 4880 | if ((ino == EXT4_ROOT_INO) && (raw_inode->i_links_count == 0)) { |
8a363970 TT |
4881 | ext4_error_inode(inode, function, line, 0, |
4882 | "iget: root inode unallocated"); | |
8e4b5eae TT |
4883 | ret = -EFSCORRUPTED; |
4884 | goto bad_inode; | |
4885 | } | |
4886 | ||
8a363970 TT |
4887 | if ((flags & EXT4_IGET_HANDLE) && |
4888 | (raw_inode->i_links_count == 0) && (raw_inode->i_mode == 0)) { | |
4889 | ret = -ESTALE; | |
4890 | goto bad_inode; | |
4891 | } | |
4892 | ||
814525f4 DW |
4893 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
4894 | ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize); | |
4895 | if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > | |
2dc8d9e1 EB |
4896 | EXT4_INODE_SIZE(inode->i_sb) || |
4897 | (ei->i_extra_isize & 3)) { | |
8a363970 TT |
4898 | ext4_error_inode(inode, function, line, 0, |
4899 | "iget: bad extra_isize %u " | |
4900 | "(inode size %u)", | |
2dc8d9e1 EB |
4901 | ei->i_extra_isize, |
4902 | EXT4_INODE_SIZE(inode->i_sb)); | |
6a797d27 | 4903 | ret = -EFSCORRUPTED; |
814525f4 DW |
4904 | goto bad_inode; |
4905 | } | |
4906 | } else | |
4907 | ei->i_extra_isize = 0; | |
4908 | ||
4909 | /* Precompute checksum seed for inode metadata */ | |
9aa5d32b | 4910 | if (ext4_has_metadata_csum(sb)) { |
814525f4 DW |
4911 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
4912 | __u32 csum; | |
4913 | __le32 inum = cpu_to_le32(inode->i_ino); | |
4914 | __le32 gen = raw_inode->i_generation; | |
4915 | csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum, | |
4916 | sizeof(inum)); | |
4917 | ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen, | |
4918 | sizeof(gen)); | |
4919 | } | |
4920 | ||
4921 | if (!ext4_inode_csum_verify(inode, raw_inode, ei)) { | |
8a363970 TT |
4922 | ext4_error_inode(inode, function, line, 0, |
4923 | "iget: checksum invalid"); | |
6a797d27 | 4924 | ret = -EFSBADCRC; |
814525f4 DW |
4925 | goto bad_inode; |
4926 | } | |
4927 | ||
ac27a0ec | 4928 | inode->i_mode = le16_to_cpu(raw_inode->i_mode); |
08cefc7a EB |
4929 | i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low); |
4930 | i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low); | |
0b7b7779 | 4931 | if (ext4_has_feature_project(sb) && |
040cb378 LX |
4932 | EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE && |
4933 | EXT4_FITS_IN_INODE(raw_inode, ei, i_projid)) | |
4934 | i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid); | |
4935 | else | |
4936 | i_projid = EXT4_DEF_PROJID; | |
4937 | ||
af5bc92d | 4938 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
08cefc7a EB |
4939 | i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16; |
4940 | i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16; | |
ac27a0ec | 4941 | } |
08cefc7a EB |
4942 | i_uid_write(inode, i_uid); |
4943 | i_gid_write(inode, i_gid); | |
040cb378 | 4944 | ei->i_projid = make_kprojid(&init_user_ns, i_projid); |
bfe86848 | 4945 | set_nlink(inode, le16_to_cpu(raw_inode->i_links_count)); |
ac27a0ec | 4946 | |
353eb83c | 4947 | ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */ |
67cf5b09 | 4948 | ei->i_inline_off = 0; |
ac27a0ec DK |
4949 | ei->i_dir_start_lookup = 0; |
4950 | ei->i_dtime = le32_to_cpu(raw_inode->i_dtime); | |
4951 | /* We now have enough fields to check if the inode was active or not. | |
4952 | * This is needed because nfsd might try to access dead inodes | |
4953 | * the test is that same one that e2fsck uses | |
4954 | * NeilBrown 1999oct15 | |
4955 | */ | |
4956 | if (inode->i_nlink == 0) { | |
393d1d1d DTB |
4957 | if ((inode->i_mode == 0 || |
4958 | !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) && | |
4959 | ino != EXT4_BOOT_LOADER_INO) { | |
ac27a0ec | 4960 | /* this inode is deleted */ |
1d1fe1ee | 4961 | ret = -ESTALE; |
ac27a0ec DK |
4962 | goto bad_inode; |
4963 | } | |
4964 | /* The only unlinked inodes we let through here have | |
4965 | * valid i_mode and are being read by the orphan | |
4966 | * recovery code: that's fine, we're about to complete | |
393d1d1d DTB |
4967 | * the process of deleting those. |
4968 | * OR it is the EXT4_BOOT_LOADER_INO which is | |
4969 | * not initialized on a new filesystem. */ | |
ac27a0ec | 4970 | } |
ac27a0ec | 4971 | ei->i_flags = le32_to_cpu(raw_inode->i_flags); |
cce6c9f7 | 4972 | ext4_set_inode_flags(inode); |
0fc1b451 | 4973 | inode->i_blocks = ext4_inode_blocks(raw_inode, ei); |
7973c0c1 | 4974 | ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo); |
e2b911c5 | 4975 | if (ext4_has_feature_64bit(sb)) |
a1ddeb7e BP |
4976 | ei->i_file_acl |= |
4977 | ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32; | |
e08ac99f | 4978 | inode->i_size = ext4_isize(sb, raw_inode); |
7e6e1ef4 | 4979 | if ((size = i_size_read(inode)) < 0) { |
8a363970 TT |
4980 | ext4_error_inode(inode, function, line, 0, |
4981 | "iget: bad i_size value: %lld", size); | |
7e6e1ef4 DW |
4982 | ret = -EFSCORRUPTED; |
4983 | goto bad_inode; | |
4984 | } | |
ac27a0ec | 4985 | ei->i_disksize = inode->i_size; |
a9e7f447 DM |
4986 | #ifdef CONFIG_QUOTA |
4987 | ei->i_reserved_quota = 0; | |
4988 | #endif | |
ac27a0ec DK |
4989 | inode->i_generation = le32_to_cpu(raw_inode->i_generation); |
4990 | ei->i_block_group = iloc.block_group; | |
a4912123 | 4991 | ei->i_last_alloc_group = ~0; |
ac27a0ec DK |
4992 | /* |
4993 | * NOTE! The in-memory inode i_data array is in little-endian order | |
4994 | * even on big-endian machines: we do NOT byteswap the block numbers! | |
4995 | */ | |
617ba13b | 4996 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
ac27a0ec DK |
4997 | ei->i_data[block] = raw_inode->i_block[block]; |
4998 | INIT_LIST_HEAD(&ei->i_orphan); | |
4999 | ||
b436b9be JK |
5000 | /* |
5001 | * Set transaction id's of transactions that have to be committed | |
5002 | * to finish f[data]sync. We set them to currently running transaction | |
5003 | * as we cannot be sure that the inode or some of its metadata isn't | |
5004 | * part of the transaction - the inode could have been reclaimed and | |
5005 | * now it is reread from disk. | |
5006 | */ | |
5007 | if (journal) { | |
5008 | transaction_t *transaction; | |
5009 | tid_t tid; | |
5010 | ||
a931da6a | 5011 | read_lock(&journal->j_state_lock); |
b436b9be JK |
5012 | if (journal->j_running_transaction) |
5013 | transaction = journal->j_running_transaction; | |
5014 | else | |
5015 | transaction = journal->j_committing_transaction; | |
5016 | if (transaction) | |
5017 | tid = transaction->t_tid; | |
5018 | else | |
5019 | tid = journal->j_commit_sequence; | |
a931da6a | 5020 | read_unlock(&journal->j_state_lock); |
b436b9be JK |
5021 | ei->i_sync_tid = tid; |
5022 | ei->i_datasync_tid = tid; | |
5023 | } | |
5024 | ||
0040d987 | 5025 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
ac27a0ec DK |
5026 | if (ei->i_extra_isize == 0) { |
5027 | /* The extra space is currently unused. Use it. */ | |
2dc8d9e1 | 5028 | BUILD_BUG_ON(sizeof(struct ext4_inode) & 3); |
617ba13b MC |
5029 | ei->i_extra_isize = sizeof(struct ext4_inode) - |
5030 | EXT4_GOOD_OLD_INODE_SIZE; | |
ac27a0ec | 5031 | } else { |
eb9b5f01 TT |
5032 | ret = ext4_iget_extra_inode(inode, raw_inode, ei); |
5033 | if (ret) | |
5034 | goto bad_inode; | |
ac27a0ec | 5035 | } |
814525f4 | 5036 | } |
ac27a0ec | 5037 | |
ef7f3835 KS |
5038 | EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode); |
5039 | EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode); | |
5040 | EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode); | |
5041 | EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode); | |
5042 | ||
ed3654eb | 5043 | if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) { |
ee73f9a5 JL |
5044 | u64 ivers = le32_to_cpu(raw_inode->i_disk_version); |
5045 | ||
c4f65706 TT |
5046 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
5047 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
ee73f9a5 | 5048 | ivers |= |
c4f65706 TT |
5049 | (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32; |
5050 | } | |
e254d1af | 5051 | ext4_inode_set_iversion_queried(inode, ivers); |
25ec56b5 JNC |
5052 | } |
5053 | ||
c4b5a614 | 5054 | ret = 0; |
485c26ec | 5055 | if (ei->i_file_acl && |
1032988c | 5056 | !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) { |
8a363970 TT |
5057 | ext4_error_inode(inode, function, line, 0, |
5058 | "iget: bad extended attribute block %llu", | |
24676da4 | 5059 | ei->i_file_acl); |
6a797d27 | 5060 | ret = -EFSCORRUPTED; |
485c26ec | 5061 | goto bad_inode; |
f19d5870 | 5062 | } else if (!ext4_has_inline_data(inode)) { |
bc716523 LS |
5063 | /* validate the block references in the inode */ |
5064 | if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || | |
5065 | (S_ISLNK(inode->i_mode) && | |
5066 | !ext4_inode_is_fast_symlink(inode))) { | |
5067 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
f19d5870 | 5068 | ret = ext4_ext_check_inode(inode); |
bc716523 LS |
5069 | else |
5070 | ret = ext4_ind_check_inode(inode); | |
f19d5870 | 5071 | } |
fe2c8191 | 5072 | } |
567f3e9a | 5073 | if (ret) |
de9a55b8 | 5074 | goto bad_inode; |
7a262f7c | 5075 | |
ac27a0ec | 5076 | if (S_ISREG(inode->i_mode)) { |
617ba13b | 5077 | inode->i_op = &ext4_file_inode_operations; |
be64f884 | 5078 | inode->i_fop = &ext4_file_operations; |
617ba13b | 5079 | ext4_set_aops(inode); |
ac27a0ec | 5080 | } else if (S_ISDIR(inode->i_mode)) { |
617ba13b MC |
5081 | inode->i_op = &ext4_dir_inode_operations; |
5082 | inode->i_fop = &ext4_dir_operations; | |
ac27a0ec | 5083 | } else if (S_ISLNK(inode->i_mode)) { |
6390d33b LR |
5084 | /* VFS does not allow setting these so must be corruption */ |
5085 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) { | |
8a363970 TT |
5086 | ext4_error_inode(inode, function, line, 0, |
5087 | "iget: immutable or append flags " | |
5088 | "not allowed on symlinks"); | |
6390d33b LR |
5089 | ret = -EFSCORRUPTED; |
5090 | goto bad_inode; | |
5091 | } | |
592ddec7 | 5092 | if (IS_ENCRYPTED(inode)) { |
a7a67e8a AV |
5093 | inode->i_op = &ext4_encrypted_symlink_inode_operations; |
5094 | ext4_set_aops(inode); | |
5095 | } else if (ext4_inode_is_fast_symlink(inode)) { | |
75e7566b | 5096 | inode->i_link = (char *)ei->i_data; |
617ba13b | 5097 | inode->i_op = &ext4_fast_symlink_inode_operations; |
e83c1397 DG |
5098 | nd_terminate_link(ei->i_data, inode->i_size, |
5099 | sizeof(ei->i_data) - 1); | |
5100 | } else { | |
617ba13b MC |
5101 | inode->i_op = &ext4_symlink_inode_operations; |
5102 | ext4_set_aops(inode); | |
ac27a0ec | 5103 | } |
21fc61c7 | 5104 | inode_nohighmem(inode); |
563bdd61 TT |
5105 | } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) || |
5106 | S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) { | |
617ba13b | 5107 | inode->i_op = &ext4_special_inode_operations; |
ac27a0ec DK |
5108 | if (raw_inode->i_block[0]) |
5109 | init_special_inode(inode, inode->i_mode, | |
5110 | old_decode_dev(le32_to_cpu(raw_inode->i_block[0]))); | |
5111 | else | |
5112 | init_special_inode(inode, inode->i_mode, | |
5113 | new_decode_dev(le32_to_cpu(raw_inode->i_block[1]))); | |
393d1d1d DTB |
5114 | } else if (ino == EXT4_BOOT_LOADER_INO) { |
5115 | make_bad_inode(inode); | |
563bdd61 | 5116 | } else { |
6a797d27 | 5117 | ret = -EFSCORRUPTED; |
8a363970 TT |
5118 | ext4_error_inode(inode, function, line, 0, |
5119 | "iget: bogus i_mode (%o)", inode->i_mode); | |
563bdd61 | 5120 | goto bad_inode; |
ac27a0ec | 5121 | } |
af5bc92d | 5122 | brelse(iloc.bh); |
dec214d0 | 5123 | |
1d1fe1ee DH |
5124 | unlock_new_inode(inode); |
5125 | return inode; | |
ac27a0ec DK |
5126 | |
5127 | bad_inode: | |
567f3e9a | 5128 | brelse(iloc.bh); |
1d1fe1ee DH |
5129 | iget_failed(inode); |
5130 | return ERR_PTR(ret); | |
ac27a0ec DK |
5131 | } |
5132 | ||
0fc1b451 AK |
5133 | static int ext4_inode_blocks_set(handle_t *handle, |
5134 | struct ext4_inode *raw_inode, | |
5135 | struct ext4_inode_info *ei) | |
5136 | { | |
5137 | struct inode *inode = &(ei->vfs_inode); | |
5138 | u64 i_blocks = inode->i_blocks; | |
5139 | struct super_block *sb = inode->i_sb; | |
0fc1b451 AK |
5140 | |
5141 | if (i_blocks <= ~0U) { | |
5142 | /* | |
4907cb7b | 5143 | * i_blocks can be represented in a 32 bit variable |
0fc1b451 AK |
5144 | * as multiple of 512 bytes |
5145 | */ | |
8180a562 | 5146 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 5147 | raw_inode->i_blocks_high = 0; |
84a8dce2 | 5148 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
f287a1a5 TT |
5149 | return 0; |
5150 | } | |
e2b911c5 | 5151 | if (!ext4_has_feature_huge_file(sb)) |
f287a1a5 TT |
5152 | return -EFBIG; |
5153 | ||
5154 | if (i_blocks <= 0xffffffffffffULL) { | |
0fc1b451 AK |
5155 | /* |
5156 | * i_blocks can be represented in a 48 bit variable | |
5157 | * as multiple of 512 bytes | |
5158 | */ | |
8180a562 | 5159 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 5160 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); |
84a8dce2 | 5161 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
0fc1b451 | 5162 | } else { |
84a8dce2 | 5163 | ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
8180a562 AK |
5164 | /* i_block is stored in file system block size */ |
5165 | i_blocks = i_blocks >> (inode->i_blkbits - 9); | |
5166 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); | |
5167 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); | |
0fc1b451 | 5168 | } |
f287a1a5 | 5169 | return 0; |
0fc1b451 AK |
5170 | } |
5171 | ||
a26f4992 TT |
5172 | struct other_inode { |
5173 | unsigned long orig_ino; | |
5174 | struct ext4_inode *raw_inode; | |
5175 | }; | |
5176 | ||
5177 | static int other_inode_match(struct inode * inode, unsigned long ino, | |
5178 | void *data) | |
5179 | { | |
5180 | struct other_inode *oi = (struct other_inode *) data; | |
5181 | ||
5182 | if ((inode->i_ino != ino) || | |
5183 | (inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW | | |
0e11f644 | 5184 | I_DIRTY_INODE)) || |
a26f4992 TT |
5185 | ((inode->i_state & I_DIRTY_TIME) == 0)) |
5186 | return 0; | |
5187 | spin_lock(&inode->i_lock); | |
5188 | if (((inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW | | |
0e11f644 | 5189 | I_DIRTY_INODE)) == 0) && |
a26f4992 TT |
5190 | (inode->i_state & I_DIRTY_TIME)) { |
5191 | struct ext4_inode_info *ei = EXT4_I(inode); | |
5192 | ||
5193 | inode->i_state &= ~(I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED); | |
5194 | spin_unlock(&inode->i_lock); | |
5195 | ||
5196 | spin_lock(&ei->i_raw_lock); | |
5197 | EXT4_INODE_SET_XTIME(i_ctime, inode, oi->raw_inode); | |
5198 | EXT4_INODE_SET_XTIME(i_mtime, inode, oi->raw_inode); | |
5199 | EXT4_INODE_SET_XTIME(i_atime, inode, oi->raw_inode); | |
5200 | ext4_inode_csum_set(inode, oi->raw_inode, ei); | |
5201 | spin_unlock(&ei->i_raw_lock); | |
5202 | trace_ext4_other_inode_update_time(inode, oi->orig_ino); | |
5203 | return -1; | |
5204 | } | |
5205 | spin_unlock(&inode->i_lock); | |
5206 | return -1; | |
5207 | } | |
5208 | ||
5209 | /* | |
5210 | * Opportunistically update the other time fields for other inodes in | |
5211 | * the same inode table block. | |
5212 | */ | |
5213 | static void ext4_update_other_inodes_time(struct super_block *sb, | |
5214 | unsigned long orig_ino, char *buf) | |
5215 | { | |
5216 | struct other_inode oi; | |
5217 | unsigned long ino; | |
5218 | int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; | |
5219 | int inode_size = EXT4_INODE_SIZE(sb); | |
5220 | ||
5221 | oi.orig_ino = orig_ino; | |
0f0ff9a9 TT |
5222 | /* |
5223 | * Calculate the first inode in the inode table block. Inode | |
5224 | * numbers are one-based. That is, the first inode in a block | |
5225 | * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1). | |
5226 | */ | |
5227 | ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1; | |
a26f4992 TT |
5228 | for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) { |
5229 | if (ino == orig_ino) | |
5230 | continue; | |
5231 | oi.raw_inode = (struct ext4_inode *) buf; | |
5232 | (void) find_inode_nowait(sb, ino, other_inode_match, &oi); | |
5233 | } | |
5234 | } | |
5235 | ||
ac27a0ec DK |
5236 | /* |
5237 | * Post the struct inode info into an on-disk inode location in the | |
5238 | * buffer-cache. This gobbles the caller's reference to the | |
5239 | * buffer_head in the inode location struct. | |
5240 | * | |
5241 | * The caller must have write access to iloc->bh. | |
5242 | */ | |
617ba13b | 5243 | static int ext4_do_update_inode(handle_t *handle, |
ac27a0ec | 5244 | struct inode *inode, |
830156c7 | 5245 | struct ext4_iloc *iloc) |
ac27a0ec | 5246 | { |
617ba13b MC |
5247 | struct ext4_inode *raw_inode = ext4_raw_inode(iloc); |
5248 | struct ext4_inode_info *ei = EXT4_I(inode); | |
ac27a0ec | 5249 | struct buffer_head *bh = iloc->bh; |
202ee5df | 5250 | struct super_block *sb = inode->i_sb; |
ac27a0ec | 5251 | int err = 0, rc, block; |
202ee5df | 5252 | int need_datasync = 0, set_large_file = 0; |
08cefc7a EB |
5253 | uid_t i_uid; |
5254 | gid_t i_gid; | |
040cb378 | 5255 | projid_t i_projid; |
ac27a0ec | 5256 | |
202ee5df TT |
5257 | spin_lock(&ei->i_raw_lock); |
5258 | ||
5259 | /* For fields not tracked in the in-memory inode, | |
ac27a0ec | 5260 | * initialise them to zero for new inodes. */ |
19f5fb7a | 5261 | if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) |
617ba13b | 5262 | memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size); |
ac27a0ec DK |
5263 | |
5264 | raw_inode->i_mode = cpu_to_le16(inode->i_mode); | |
08cefc7a EB |
5265 | i_uid = i_uid_read(inode); |
5266 | i_gid = i_gid_read(inode); | |
040cb378 | 5267 | i_projid = from_kprojid(&init_user_ns, ei->i_projid); |
af5bc92d | 5268 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
08cefc7a EB |
5269 | raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid)); |
5270 | raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid)); | |
ac27a0ec DK |
5271 | /* |
5272 | * Fix up interoperability with old kernels. Otherwise, old inodes get | |
5273 | * re-used with the upper 16 bits of the uid/gid intact | |
5274 | */ | |
93e3b4e6 DJ |
5275 | if (ei->i_dtime && list_empty(&ei->i_orphan)) { |
5276 | raw_inode->i_uid_high = 0; | |
5277 | raw_inode->i_gid_high = 0; | |
5278 | } else { | |
ac27a0ec | 5279 | raw_inode->i_uid_high = |
08cefc7a | 5280 | cpu_to_le16(high_16_bits(i_uid)); |
ac27a0ec | 5281 | raw_inode->i_gid_high = |
08cefc7a | 5282 | cpu_to_le16(high_16_bits(i_gid)); |
ac27a0ec DK |
5283 | } |
5284 | } else { | |
08cefc7a EB |
5285 | raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid)); |
5286 | raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid)); | |
ac27a0ec DK |
5287 | raw_inode->i_uid_high = 0; |
5288 | raw_inode->i_gid_high = 0; | |
5289 | } | |
5290 | raw_inode->i_links_count = cpu_to_le16(inode->i_nlink); | |
ef7f3835 KS |
5291 | |
5292 | EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode); | |
5293 | EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode); | |
5294 | EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode); | |
5295 | EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode); | |
5296 | ||
bce92d56 LX |
5297 | err = ext4_inode_blocks_set(handle, raw_inode, ei); |
5298 | if (err) { | |
202ee5df | 5299 | spin_unlock(&ei->i_raw_lock); |
0fc1b451 | 5300 | goto out_brelse; |
202ee5df | 5301 | } |
ac27a0ec | 5302 | raw_inode->i_dtime = cpu_to_le32(ei->i_dtime); |
353eb83c | 5303 | raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF); |
ed3654eb | 5304 | if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) |
a1ddeb7e BP |
5305 | raw_inode->i_file_acl_high = |
5306 | cpu_to_le16(ei->i_file_acl >> 32); | |
7973c0c1 | 5307 | raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl); |
e08ac99f | 5308 | if (ei->i_disksize != ext4_isize(inode->i_sb, raw_inode)) { |
b71fc079 JK |
5309 | ext4_isize_set(raw_inode, ei->i_disksize); |
5310 | need_datasync = 1; | |
5311 | } | |
a48380f7 | 5312 | if (ei->i_disksize > 0x7fffffffULL) { |
e2b911c5 | 5313 | if (!ext4_has_feature_large_file(sb) || |
a48380f7 | 5314 | EXT4_SB(sb)->s_es->s_rev_level == |
202ee5df TT |
5315 | cpu_to_le32(EXT4_GOOD_OLD_REV)) |
5316 | set_large_file = 1; | |
ac27a0ec DK |
5317 | } |
5318 | raw_inode->i_generation = cpu_to_le32(inode->i_generation); | |
5319 | if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) { | |
5320 | if (old_valid_dev(inode->i_rdev)) { | |
5321 | raw_inode->i_block[0] = | |
5322 | cpu_to_le32(old_encode_dev(inode->i_rdev)); | |
5323 | raw_inode->i_block[1] = 0; | |
5324 | } else { | |
5325 | raw_inode->i_block[0] = 0; | |
5326 | raw_inode->i_block[1] = | |
5327 | cpu_to_le32(new_encode_dev(inode->i_rdev)); | |
5328 | raw_inode->i_block[2] = 0; | |
5329 | } | |
f19d5870 | 5330 | } else if (!ext4_has_inline_data(inode)) { |
de9a55b8 TT |
5331 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
5332 | raw_inode->i_block[block] = ei->i_data[block]; | |
f19d5870 | 5333 | } |
ac27a0ec | 5334 | |
ed3654eb | 5335 | if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) { |
e254d1af | 5336 | u64 ivers = ext4_inode_peek_iversion(inode); |
ee73f9a5 JL |
5337 | |
5338 | raw_inode->i_disk_version = cpu_to_le32(ivers); | |
c4f65706 TT |
5339 | if (ei->i_extra_isize) { |
5340 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
5341 | raw_inode->i_version_hi = | |
ee73f9a5 | 5342 | cpu_to_le32(ivers >> 32); |
c4f65706 TT |
5343 | raw_inode->i_extra_isize = |
5344 | cpu_to_le16(ei->i_extra_isize); | |
5345 | } | |
25ec56b5 | 5346 | } |
040cb378 | 5347 | |
0b7b7779 | 5348 | BUG_ON(!ext4_has_feature_project(inode->i_sb) && |
040cb378 LX |
5349 | i_projid != EXT4_DEF_PROJID); |
5350 | ||
5351 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && | |
5352 | EXT4_FITS_IN_INODE(raw_inode, ei, i_projid)) | |
5353 | raw_inode->i_projid = cpu_to_le32(i_projid); | |
5354 | ||
814525f4 | 5355 | ext4_inode_csum_set(inode, raw_inode, ei); |
202ee5df | 5356 | spin_unlock(&ei->i_raw_lock); |
1751e8a6 | 5357 | if (inode->i_sb->s_flags & SB_LAZYTIME) |
a26f4992 TT |
5358 | ext4_update_other_inodes_time(inode->i_sb, inode->i_ino, |
5359 | bh->b_data); | |
202ee5df | 5360 | |
830156c7 | 5361 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
73b50c1c | 5362 | rc = ext4_handle_dirty_metadata(handle, NULL, bh); |
830156c7 FM |
5363 | if (!err) |
5364 | err = rc; | |
19f5fb7a | 5365 | ext4_clear_inode_state(inode, EXT4_STATE_NEW); |
202ee5df | 5366 | if (set_large_file) { |
5d601255 | 5367 | BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access"); |
202ee5df TT |
5368 | err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh); |
5369 | if (err) | |
5370 | goto out_brelse; | |
e2b911c5 | 5371 | ext4_set_feature_large_file(sb); |
202ee5df TT |
5372 | ext4_handle_sync(handle); |
5373 | err = ext4_handle_dirty_super(handle, sb); | |
5374 | } | |
b71fc079 | 5375 | ext4_update_inode_fsync_trans(handle, inode, need_datasync); |
ac27a0ec | 5376 | out_brelse: |
af5bc92d | 5377 | brelse(bh); |
617ba13b | 5378 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
5379 | return err; |
5380 | } | |
5381 | ||
5382 | /* | |
617ba13b | 5383 | * ext4_write_inode() |
ac27a0ec DK |
5384 | * |
5385 | * We are called from a few places: | |
5386 | * | |
87f7e416 | 5387 | * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files. |
ac27a0ec | 5388 | * Here, there will be no transaction running. We wait for any running |
4907cb7b | 5389 | * transaction to commit. |
ac27a0ec | 5390 | * |
87f7e416 TT |
5391 | * - Within flush work (sys_sync(), kupdate and such). |
5392 | * We wait on commit, if told to. | |
ac27a0ec | 5393 | * |
87f7e416 TT |
5394 | * - Within iput_final() -> write_inode_now() |
5395 | * We wait on commit, if told to. | |
ac27a0ec DK |
5396 | * |
5397 | * In all cases it is actually safe for us to return without doing anything, | |
5398 | * because the inode has been copied into a raw inode buffer in | |
87f7e416 TT |
5399 | * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL |
5400 | * writeback. | |
ac27a0ec DK |
5401 | * |
5402 | * Note that we are absolutely dependent upon all inode dirtiers doing the | |
5403 | * right thing: they *must* call mark_inode_dirty() after dirtying info in | |
5404 | * which we are interested. | |
5405 | * | |
5406 | * It would be a bug for them to not do this. The code: | |
5407 | * | |
5408 | * mark_inode_dirty(inode) | |
5409 | * stuff(); | |
5410 | * inode->i_size = expr; | |
5411 | * | |
87f7e416 TT |
5412 | * is in error because write_inode() could occur while `stuff()' is running, |
5413 | * and the new i_size will be lost. Plus the inode will no longer be on the | |
5414 | * superblock's dirty inode list. | |
ac27a0ec | 5415 | */ |
a9185b41 | 5416 | int ext4_write_inode(struct inode *inode, struct writeback_control *wbc) |
ac27a0ec | 5417 | { |
91ac6f43 FM |
5418 | int err; |
5419 | ||
18f2c4fc TT |
5420 | if (WARN_ON_ONCE(current->flags & PF_MEMALLOC) || |
5421 | sb_rdonly(inode->i_sb)) | |
ac27a0ec DK |
5422 | return 0; |
5423 | ||
18f2c4fc TT |
5424 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
5425 | return -EIO; | |
5426 | ||
91ac6f43 FM |
5427 | if (EXT4_SB(inode->i_sb)->s_journal) { |
5428 | if (ext4_journal_current_handle()) { | |
5429 | jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n"); | |
5430 | dump_stack(); | |
5431 | return -EIO; | |
5432 | } | |
ac27a0ec | 5433 | |
10542c22 JK |
5434 | /* |
5435 | * No need to force transaction in WB_SYNC_NONE mode. Also | |
5436 | * ext4_sync_fs() will force the commit after everything is | |
5437 | * written. | |
5438 | */ | |
5439 | if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync) | |
91ac6f43 FM |
5440 | return 0; |
5441 | ||
18f2c4fc TT |
5442 | err = jbd2_complete_transaction(EXT4_SB(inode->i_sb)->s_journal, |
5443 | EXT4_I(inode)->i_sync_tid); | |
91ac6f43 FM |
5444 | } else { |
5445 | struct ext4_iloc iloc; | |
ac27a0ec | 5446 | |
8b472d73 | 5447 | err = __ext4_get_inode_loc(inode, &iloc, 0); |
91ac6f43 FM |
5448 | if (err) |
5449 | return err; | |
10542c22 JK |
5450 | /* |
5451 | * sync(2) will flush the whole buffer cache. No need to do | |
5452 | * it here separately for each inode. | |
5453 | */ | |
5454 | if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) | |
830156c7 FM |
5455 | sync_dirty_buffer(iloc.bh); |
5456 | if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) { | |
c398eda0 TT |
5457 | EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr, |
5458 | "IO error syncing inode"); | |
830156c7 FM |
5459 | err = -EIO; |
5460 | } | |
fd2dd9fb | 5461 | brelse(iloc.bh); |
91ac6f43 FM |
5462 | } |
5463 | return err; | |
ac27a0ec DK |
5464 | } |
5465 | ||
53e87268 JK |
5466 | /* |
5467 | * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate | |
5468 | * buffers that are attached to a page stradding i_size and are undergoing | |
5469 | * commit. In that case we have to wait for commit to finish and try again. | |
5470 | */ | |
5471 | static void ext4_wait_for_tail_page_commit(struct inode *inode) | |
5472 | { | |
5473 | struct page *page; | |
5474 | unsigned offset; | |
5475 | journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; | |
5476 | tid_t commit_tid = 0; | |
5477 | int ret; | |
5478 | ||
09cbfeaf | 5479 | offset = inode->i_size & (PAGE_SIZE - 1); |
53e87268 JK |
5480 | /* |
5481 | * All buffers in the last page remain valid? Then there's nothing to | |
ea1754a0 | 5482 | * do. We do the check mainly to optimize the common PAGE_SIZE == |
53e87268 JK |
5483 | * blocksize case |
5484 | */ | |
93407472 | 5485 | if (offset > PAGE_SIZE - i_blocksize(inode)) |
53e87268 JK |
5486 | return; |
5487 | while (1) { | |
5488 | page = find_lock_page(inode->i_mapping, | |
09cbfeaf | 5489 | inode->i_size >> PAGE_SHIFT); |
53e87268 JK |
5490 | if (!page) |
5491 | return; | |
ca99fdd2 | 5492 | ret = __ext4_journalled_invalidatepage(page, offset, |
09cbfeaf | 5493 | PAGE_SIZE - offset); |
53e87268 | 5494 | unlock_page(page); |
09cbfeaf | 5495 | put_page(page); |
53e87268 JK |
5496 | if (ret != -EBUSY) |
5497 | return; | |
5498 | commit_tid = 0; | |
5499 | read_lock(&journal->j_state_lock); | |
5500 | if (journal->j_committing_transaction) | |
5501 | commit_tid = journal->j_committing_transaction->t_tid; | |
5502 | read_unlock(&journal->j_state_lock); | |
5503 | if (commit_tid) | |
5504 | jbd2_log_wait_commit(journal, commit_tid); | |
5505 | } | |
5506 | } | |
5507 | ||
ac27a0ec | 5508 | /* |
617ba13b | 5509 | * ext4_setattr() |
ac27a0ec DK |
5510 | * |
5511 | * Called from notify_change. | |
5512 | * | |
5513 | * We want to trap VFS attempts to truncate the file as soon as | |
5514 | * possible. In particular, we want to make sure that when the VFS | |
5515 | * shrinks i_size, we put the inode on the orphan list and modify | |
5516 | * i_disksize immediately, so that during the subsequent flushing of | |
5517 | * dirty pages and freeing of disk blocks, we can guarantee that any | |
5518 | * commit will leave the blocks being flushed in an unused state on | |
5519 | * disk. (On recovery, the inode will get truncated and the blocks will | |
5520 | * be freed, so we have a strong guarantee that no future commit will | |
5521 | * leave these blocks visible to the user.) | |
5522 | * | |
678aaf48 JK |
5523 | * Another thing we have to assure is that if we are in ordered mode |
5524 | * and inode is still attached to the committing transaction, we must | |
5525 | * we start writeout of all the dirty pages which are being truncated. | |
5526 | * This way we are sure that all the data written in the previous | |
5527 | * transaction are already on disk (truncate waits for pages under | |
5528 | * writeback). | |
5529 | * | |
5530 | * Called with inode->i_mutex down. | |
ac27a0ec | 5531 | */ |
617ba13b | 5532 | int ext4_setattr(struct dentry *dentry, struct iattr *attr) |
ac27a0ec | 5533 | { |
2b0143b5 | 5534 | struct inode *inode = d_inode(dentry); |
ac27a0ec | 5535 | int error, rc = 0; |
3d287de3 | 5536 | int orphan = 0; |
ac27a0ec DK |
5537 | const unsigned int ia_valid = attr->ia_valid; |
5538 | ||
0db1ff22 TT |
5539 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
5540 | return -EIO; | |
5541 | ||
02b016ca TT |
5542 | if (unlikely(IS_IMMUTABLE(inode))) |
5543 | return -EPERM; | |
5544 | ||
5545 | if (unlikely(IS_APPEND(inode) && | |
5546 | (ia_valid & (ATTR_MODE | ATTR_UID | | |
5547 | ATTR_GID | ATTR_TIMES_SET)))) | |
5548 | return -EPERM; | |
5549 | ||
31051c85 | 5550 | error = setattr_prepare(dentry, attr); |
ac27a0ec DK |
5551 | if (error) |
5552 | return error; | |
5553 | ||
3ce2b8dd EB |
5554 | error = fscrypt_prepare_setattr(dentry, attr); |
5555 | if (error) | |
5556 | return error; | |
5557 | ||
a7cdadee JK |
5558 | if (is_quota_modification(inode, attr)) { |
5559 | error = dquot_initialize(inode); | |
5560 | if (error) | |
5561 | return error; | |
5562 | } | |
08cefc7a EB |
5563 | if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) || |
5564 | (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) { | |
ac27a0ec DK |
5565 | handle_t *handle; |
5566 | ||
5567 | /* (user+group)*(old+new) structure, inode write (sb, | |
5568 | * inode block, ? - but truncate inode update has it) */ | |
9924a92a TT |
5569 | handle = ext4_journal_start(inode, EXT4_HT_QUOTA, |
5570 | (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) + | |
5571 | EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3); | |
ac27a0ec DK |
5572 | if (IS_ERR(handle)) { |
5573 | error = PTR_ERR(handle); | |
5574 | goto err_out; | |
5575 | } | |
7a9ca53a TE |
5576 | |
5577 | /* dquot_transfer() calls back ext4_get_inode_usage() which | |
5578 | * counts xattr inode references. | |
5579 | */ | |
5580 | down_read(&EXT4_I(inode)->xattr_sem); | |
b43fa828 | 5581 | error = dquot_transfer(inode, attr); |
7a9ca53a TE |
5582 | up_read(&EXT4_I(inode)->xattr_sem); |
5583 | ||
ac27a0ec | 5584 | if (error) { |
617ba13b | 5585 | ext4_journal_stop(handle); |
ac27a0ec DK |
5586 | return error; |
5587 | } | |
5588 | /* Update corresponding info in inode so that everything is in | |
5589 | * one transaction */ | |
5590 | if (attr->ia_valid & ATTR_UID) | |
5591 | inode->i_uid = attr->ia_uid; | |
5592 | if (attr->ia_valid & ATTR_GID) | |
5593 | inode->i_gid = attr->ia_gid; | |
617ba13b MC |
5594 | error = ext4_mark_inode_dirty(handle, inode); |
5595 | ext4_journal_stop(handle); | |
ac27a0ec DK |
5596 | } |
5597 | ||
3da40c7b | 5598 | if (attr->ia_valid & ATTR_SIZE) { |
5208386c | 5599 | handle_t *handle; |
3da40c7b | 5600 | loff_t oldsize = inode->i_size; |
b9c1c267 | 5601 | int shrink = (attr->ia_size < inode->i_size); |
562c72aa | 5602 | |
12e9b892 | 5603 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) { |
e2b46574 ES |
5604 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
5605 | ||
0c095c7f TT |
5606 | if (attr->ia_size > sbi->s_bitmap_maxbytes) |
5607 | return -EFBIG; | |
e2b46574 | 5608 | } |
3da40c7b JB |
5609 | if (!S_ISREG(inode->i_mode)) |
5610 | return -EINVAL; | |
dff6efc3 CH |
5611 | |
5612 | if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size) | |
5613 | inode_inc_iversion(inode); | |
5614 | ||
b9c1c267 JK |
5615 | if (shrink) { |
5616 | if (ext4_should_order_data(inode)) { | |
5617 | error = ext4_begin_ordered_truncate(inode, | |
678aaf48 | 5618 | attr->ia_size); |
b9c1c267 JK |
5619 | if (error) |
5620 | goto err_out; | |
5621 | } | |
5622 | /* | |
5623 | * Blocks are going to be removed from the inode. Wait | |
5624 | * for dio in flight. | |
5625 | */ | |
5626 | inode_dio_wait(inode); | |
5627 | } | |
5628 | ||
5629 | down_write(&EXT4_I(inode)->i_mmap_sem); | |
5630 | ||
5631 | rc = ext4_break_layouts(inode); | |
5632 | if (rc) { | |
5633 | up_write(&EXT4_I(inode)->i_mmap_sem); | |
5634 | return rc; | |
3da40c7b | 5635 | } |
b9c1c267 | 5636 | |
3da40c7b | 5637 | if (attr->ia_size != inode->i_size) { |
5208386c JK |
5638 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 3); |
5639 | if (IS_ERR(handle)) { | |
5640 | error = PTR_ERR(handle); | |
b9c1c267 | 5641 | goto out_mmap_sem; |
5208386c | 5642 | } |
3da40c7b | 5643 | if (ext4_handle_valid(handle) && shrink) { |
5208386c JK |
5644 | error = ext4_orphan_add(handle, inode); |
5645 | orphan = 1; | |
5646 | } | |
911af577 EG |
5647 | /* |
5648 | * Update c/mtime on truncate up, ext4_truncate() will | |
5649 | * update c/mtime in shrink case below | |
5650 | */ | |
5651 | if (!shrink) { | |
eeca7ea1 | 5652 | inode->i_mtime = current_time(inode); |
911af577 EG |
5653 | inode->i_ctime = inode->i_mtime; |
5654 | } | |
90e775b7 | 5655 | down_write(&EXT4_I(inode)->i_data_sem); |
5208386c JK |
5656 | EXT4_I(inode)->i_disksize = attr->ia_size; |
5657 | rc = ext4_mark_inode_dirty(handle, inode); | |
5658 | if (!error) | |
5659 | error = rc; | |
90e775b7 JK |
5660 | /* |
5661 | * We have to update i_size under i_data_sem together | |
5662 | * with i_disksize to avoid races with writeback code | |
5663 | * running ext4_wb_update_i_disksize(). | |
5664 | */ | |
5665 | if (!error) | |
5666 | i_size_write(inode, attr->ia_size); | |
5667 | up_write(&EXT4_I(inode)->i_data_sem); | |
5208386c | 5668 | ext4_journal_stop(handle); |
b9c1c267 JK |
5669 | if (error) |
5670 | goto out_mmap_sem; | |
5671 | if (!shrink) { | |
5672 | pagecache_isize_extended(inode, oldsize, | |
5673 | inode->i_size); | |
5674 | } else if (ext4_should_journal_data(inode)) { | |
5675 | ext4_wait_for_tail_page_commit(inode); | |
678aaf48 | 5676 | } |
d6320cbf | 5677 | } |
430657b6 | 5678 | |
5208386c JK |
5679 | /* |
5680 | * Truncate pagecache after we've waited for commit | |
5681 | * in data=journal mode to make pages freeable. | |
5682 | */ | |
923ae0ff | 5683 | truncate_pagecache(inode, inode->i_size); |
b9c1c267 JK |
5684 | /* |
5685 | * Call ext4_truncate() even if i_size didn't change to | |
5686 | * truncate possible preallocated blocks. | |
5687 | */ | |
5688 | if (attr->ia_size <= oldsize) { | |
2c98eb5e TT |
5689 | rc = ext4_truncate(inode); |
5690 | if (rc) | |
5691 | error = rc; | |
5692 | } | |
b9c1c267 | 5693 | out_mmap_sem: |
ea3d7209 | 5694 | up_write(&EXT4_I(inode)->i_mmap_sem); |
072bd7ea | 5695 | } |
ac27a0ec | 5696 | |
2c98eb5e | 5697 | if (!error) { |
1025774c CH |
5698 | setattr_copy(inode, attr); |
5699 | mark_inode_dirty(inode); | |
5700 | } | |
5701 | ||
5702 | /* | |
5703 | * If the call to ext4_truncate failed to get a transaction handle at | |
5704 | * all, we need to clean up the in-core orphan list manually. | |
5705 | */ | |
3d287de3 | 5706 | if (orphan && inode->i_nlink) |
617ba13b | 5707 | ext4_orphan_del(NULL, inode); |
ac27a0ec | 5708 | |
2c98eb5e | 5709 | if (!error && (ia_valid & ATTR_MODE)) |
64e178a7 | 5710 | rc = posix_acl_chmod(inode, inode->i_mode); |
ac27a0ec DK |
5711 | |
5712 | err_out: | |
617ba13b | 5713 | ext4_std_error(inode->i_sb, error); |
ac27a0ec DK |
5714 | if (!error) |
5715 | error = rc; | |
5716 | return error; | |
5717 | } | |
5718 | ||
a528d35e DH |
5719 | int ext4_getattr(const struct path *path, struct kstat *stat, |
5720 | u32 request_mask, unsigned int query_flags) | |
3e3398a0 | 5721 | { |
99652ea5 DH |
5722 | struct inode *inode = d_inode(path->dentry); |
5723 | struct ext4_inode *raw_inode; | |
5724 | struct ext4_inode_info *ei = EXT4_I(inode); | |
5725 | unsigned int flags; | |
5726 | ||
5727 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_crtime)) { | |
5728 | stat->result_mask |= STATX_BTIME; | |
5729 | stat->btime.tv_sec = ei->i_crtime.tv_sec; | |
5730 | stat->btime.tv_nsec = ei->i_crtime.tv_nsec; | |
5731 | } | |
5732 | ||
5733 | flags = ei->i_flags & EXT4_FL_USER_VISIBLE; | |
5734 | if (flags & EXT4_APPEND_FL) | |
5735 | stat->attributes |= STATX_ATTR_APPEND; | |
5736 | if (flags & EXT4_COMPR_FL) | |
5737 | stat->attributes |= STATX_ATTR_COMPRESSED; | |
5738 | if (flags & EXT4_ENCRYPT_FL) | |
5739 | stat->attributes |= STATX_ATTR_ENCRYPTED; | |
5740 | if (flags & EXT4_IMMUTABLE_FL) | |
5741 | stat->attributes |= STATX_ATTR_IMMUTABLE; | |
5742 | if (flags & EXT4_NODUMP_FL) | |
5743 | stat->attributes |= STATX_ATTR_NODUMP; | |
3e3398a0 | 5744 | |
3209f68b DH |
5745 | stat->attributes_mask |= (STATX_ATTR_APPEND | |
5746 | STATX_ATTR_COMPRESSED | | |
5747 | STATX_ATTR_ENCRYPTED | | |
5748 | STATX_ATTR_IMMUTABLE | | |
5749 | STATX_ATTR_NODUMP); | |
5750 | ||
3e3398a0 | 5751 | generic_fillattr(inode, stat); |
99652ea5 DH |
5752 | return 0; |
5753 | } | |
5754 | ||
5755 | int ext4_file_getattr(const struct path *path, struct kstat *stat, | |
5756 | u32 request_mask, unsigned int query_flags) | |
5757 | { | |
5758 | struct inode *inode = d_inode(path->dentry); | |
5759 | u64 delalloc_blocks; | |
5760 | ||
5761 | ext4_getattr(path, stat, request_mask, query_flags); | |
3e3398a0 | 5762 | |
9206c561 AD |
5763 | /* |
5764 | * If there is inline data in the inode, the inode will normally not | |
5765 | * have data blocks allocated (it may have an external xattr block). | |
5766 | * Report at least one sector for such files, so tools like tar, rsync, | |
d67d64f4 | 5767 | * others don't incorrectly think the file is completely sparse. |
9206c561 AD |
5768 | */ |
5769 | if (unlikely(ext4_has_inline_data(inode))) | |
5770 | stat->blocks += (stat->size + 511) >> 9; | |
5771 | ||
3e3398a0 MC |
5772 | /* |
5773 | * We can't update i_blocks if the block allocation is delayed | |
5774 | * otherwise in the case of system crash before the real block | |
5775 | * allocation is done, we will have i_blocks inconsistent with | |
5776 | * on-disk file blocks. | |
5777 | * We always keep i_blocks updated together with real | |
5778 | * allocation. But to not confuse with user, stat | |
5779 | * will return the blocks that include the delayed allocation | |
5780 | * blocks for this file. | |
5781 | */ | |
96607551 | 5782 | delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb), |
9206c561 AD |
5783 | EXT4_I(inode)->i_reserved_data_blocks); |
5784 | stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9); | |
3e3398a0 MC |
5785 | return 0; |
5786 | } | |
ac27a0ec | 5787 | |
fffb2739 JK |
5788 | static int ext4_index_trans_blocks(struct inode *inode, int lblocks, |
5789 | int pextents) | |
a02908f1 | 5790 | { |
12e9b892 | 5791 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) |
fffb2739 JK |
5792 | return ext4_ind_trans_blocks(inode, lblocks); |
5793 | return ext4_ext_index_trans_blocks(inode, pextents); | |
a02908f1 | 5794 | } |
ac51d837 | 5795 | |
ac27a0ec | 5796 | /* |
a02908f1 MC |
5797 | * Account for index blocks, block groups bitmaps and block group |
5798 | * descriptor blocks if modify datablocks and index blocks | |
5799 | * worse case, the indexs blocks spread over different block groups | |
ac27a0ec | 5800 | * |
a02908f1 | 5801 | * If datablocks are discontiguous, they are possible to spread over |
4907cb7b | 5802 | * different block groups too. If they are contiguous, with flexbg, |
a02908f1 | 5803 | * they could still across block group boundary. |
ac27a0ec | 5804 | * |
a02908f1 MC |
5805 | * Also account for superblock, inode, quota and xattr blocks |
5806 | */ | |
dec214d0 | 5807 | static int ext4_meta_trans_blocks(struct inode *inode, int lblocks, |
fffb2739 | 5808 | int pextents) |
a02908f1 | 5809 | { |
8df9675f TT |
5810 | ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb); |
5811 | int gdpblocks; | |
a02908f1 MC |
5812 | int idxblocks; |
5813 | int ret = 0; | |
5814 | ||
5815 | /* | |
fffb2739 JK |
5816 | * How many index blocks need to touch to map @lblocks logical blocks |
5817 | * to @pextents physical extents? | |
a02908f1 | 5818 | */ |
fffb2739 | 5819 | idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents); |
a02908f1 MC |
5820 | |
5821 | ret = idxblocks; | |
5822 | ||
5823 | /* | |
5824 | * Now let's see how many group bitmaps and group descriptors need | |
5825 | * to account | |
5826 | */ | |
fffb2739 | 5827 | groups = idxblocks + pextents; |
a02908f1 | 5828 | gdpblocks = groups; |
8df9675f TT |
5829 | if (groups > ngroups) |
5830 | groups = ngroups; | |
a02908f1 MC |
5831 | if (groups > EXT4_SB(inode->i_sb)->s_gdb_count) |
5832 | gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count; | |
5833 | ||
5834 | /* bitmaps and block group descriptor blocks */ | |
5835 | ret += groups + gdpblocks; | |
5836 | ||
5837 | /* Blocks for super block, inode, quota and xattr blocks */ | |
5838 | ret += EXT4_META_TRANS_BLOCKS(inode->i_sb); | |
5839 | ||
5840 | return ret; | |
5841 | } | |
5842 | ||
5843 | /* | |
25985edc | 5844 | * Calculate the total number of credits to reserve to fit |
f3bd1f3f MC |
5845 | * the modification of a single pages into a single transaction, |
5846 | * which may include multiple chunks of block allocations. | |
ac27a0ec | 5847 | * |
525f4ed8 | 5848 | * This could be called via ext4_write_begin() |
ac27a0ec | 5849 | * |
525f4ed8 | 5850 | * We need to consider the worse case, when |
a02908f1 | 5851 | * one new block per extent. |
ac27a0ec | 5852 | */ |
a86c6181 | 5853 | int ext4_writepage_trans_blocks(struct inode *inode) |
ac27a0ec | 5854 | { |
617ba13b | 5855 | int bpp = ext4_journal_blocks_per_page(inode); |
ac27a0ec DK |
5856 | int ret; |
5857 | ||
fffb2739 | 5858 | ret = ext4_meta_trans_blocks(inode, bpp, bpp); |
a86c6181 | 5859 | |
a02908f1 | 5860 | /* Account for data blocks for journalled mode */ |
617ba13b | 5861 | if (ext4_should_journal_data(inode)) |
a02908f1 | 5862 | ret += bpp; |
ac27a0ec DK |
5863 | return ret; |
5864 | } | |
f3bd1f3f MC |
5865 | |
5866 | /* | |
5867 | * Calculate the journal credits for a chunk of data modification. | |
5868 | * | |
5869 | * This is called from DIO, fallocate or whoever calling | |
79e83036 | 5870 | * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks. |
f3bd1f3f MC |
5871 | * |
5872 | * journal buffers for data blocks are not included here, as DIO | |
5873 | * and fallocate do no need to journal data buffers. | |
5874 | */ | |
5875 | int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks) | |
5876 | { | |
5877 | return ext4_meta_trans_blocks(inode, nrblocks, 1); | |
5878 | } | |
5879 | ||
ac27a0ec | 5880 | /* |
617ba13b | 5881 | * The caller must have previously called ext4_reserve_inode_write(). |
ac27a0ec DK |
5882 | * Give this, we know that the caller already has write access to iloc->bh. |
5883 | */ | |
617ba13b | 5884 | int ext4_mark_iloc_dirty(handle_t *handle, |
de9a55b8 | 5885 | struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec DK |
5886 | { |
5887 | int err = 0; | |
5888 | ||
a6758309 VA |
5889 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) { |
5890 | put_bh(iloc->bh); | |
0db1ff22 | 5891 | return -EIO; |
a6758309 | 5892 | } |
c64db50e | 5893 | if (IS_I_VERSION(inode)) |
25ec56b5 JNC |
5894 | inode_inc_iversion(inode); |
5895 | ||
ac27a0ec DK |
5896 | /* the do_update_inode consumes one bh->b_count */ |
5897 | get_bh(iloc->bh); | |
5898 | ||
dab291af | 5899 | /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */ |
830156c7 | 5900 | err = ext4_do_update_inode(handle, inode, iloc); |
ac27a0ec DK |
5901 | put_bh(iloc->bh); |
5902 | return err; | |
5903 | } | |
5904 | ||
5905 | /* | |
5906 | * On success, We end up with an outstanding reference count against | |
5907 | * iloc->bh. This _must_ be cleaned up later. | |
5908 | */ | |
5909 | ||
5910 | int | |
617ba13b MC |
5911 | ext4_reserve_inode_write(handle_t *handle, struct inode *inode, |
5912 | struct ext4_iloc *iloc) | |
ac27a0ec | 5913 | { |
0390131b FM |
5914 | int err; |
5915 | ||
0db1ff22 TT |
5916 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
5917 | return -EIO; | |
5918 | ||
0390131b FM |
5919 | err = ext4_get_inode_loc(inode, iloc); |
5920 | if (!err) { | |
5921 | BUFFER_TRACE(iloc->bh, "get_write_access"); | |
5922 | err = ext4_journal_get_write_access(handle, iloc->bh); | |
5923 | if (err) { | |
5924 | brelse(iloc->bh); | |
5925 | iloc->bh = NULL; | |
ac27a0ec DK |
5926 | } |
5927 | } | |
617ba13b | 5928 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
5929 | return err; |
5930 | } | |
5931 | ||
c03b45b8 MX |
5932 | static int __ext4_expand_extra_isize(struct inode *inode, |
5933 | unsigned int new_extra_isize, | |
5934 | struct ext4_iloc *iloc, | |
5935 | handle_t *handle, int *no_expand) | |
5936 | { | |
5937 | struct ext4_inode *raw_inode; | |
5938 | struct ext4_xattr_ibody_header *header; | |
5939 | int error; | |
5940 | ||
5941 | raw_inode = ext4_raw_inode(iloc); | |
5942 | ||
5943 | header = IHDR(inode, raw_inode); | |
5944 | ||
5945 | /* No extended attributes present */ | |
5946 | if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) || | |
5947 | header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) { | |
5948 | memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE + | |
5949 | EXT4_I(inode)->i_extra_isize, 0, | |
5950 | new_extra_isize - EXT4_I(inode)->i_extra_isize); | |
5951 | EXT4_I(inode)->i_extra_isize = new_extra_isize; | |
5952 | return 0; | |
5953 | } | |
5954 | ||
5955 | /* try to expand with EAs present */ | |
5956 | error = ext4_expand_extra_isize_ea(inode, new_extra_isize, | |
5957 | raw_inode, handle); | |
5958 | if (error) { | |
5959 | /* | |
5960 | * Inode size expansion failed; don't try again | |
5961 | */ | |
5962 | *no_expand = 1; | |
5963 | } | |
5964 | ||
5965 | return error; | |
5966 | } | |
5967 | ||
6dd4ee7c KS |
5968 | /* |
5969 | * Expand an inode by new_extra_isize bytes. | |
5970 | * Returns 0 on success or negative error number on failure. | |
5971 | */ | |
cf0a5e81 MX |
5972 | static int ext4_try_to_expand_extra_isize(struct inode *inode, |
5973 | unsigned int new_extra_isize, | |
5974 | struct ext4_iloc iloc, | |
5975 | handle_t *handle) | |
6dd4ee7c | 5976 | { |
3b10fdc6 MX |
5977 | int no_expand; |
5978 | int error; | |
6dd4ee7c | 5979 | |
cf0a5e81 MX |
5980 | if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) |
5981 | return -EOVERFLOW; | |
5982 | ||
5983 | /* | |
5984 | * In nojournal mode, we can immediately attempt to expand | |
5985 | * the inode. When journaled, we first need to obtain extra | |
5986 | * buffer credits since we may write into the EA block | |
5987 | * with this same handle. If journal_extend fails, then it will | |
5988 | * only result in a minor loss of functionality for that inode. | |
5989 | * If this is felt to be critical, then e2fsck should be run to | |
5990 | * force a large enough s_min_extra_isize. | |
5991 | */ | |
5992 | if (ext4_handle_valid(handle) && | |
5993 | jbd2_journal_extend(handle, | |
5994 | EXT4_DATA_TRANS_BLOCKS(inode->i_sb)) != 0) | |
5995 | return -ENOSPC; | |
6dd4ee7c | 5996 | |
3b10fdc6 | 5997 | if (ext4_write_trylock_xattr(inode, &no_expand) == 0) |
cf0a5e81 | 5998 | return -EBUSY; |
3b10fdc6 | 5999 | |
c03b45b8 MX |
6000 | error = __ext4_expand_extra_isize(inode, new_extra_isize, &iloc, |
6001 | handle, &no_expand); | |
6002 | ext4_write_unlock_xattr(inode, &no_expand); | |
6dd4ee7c | 6003 | |
c03b45b8 MX |
6004 | return error; |
6005 | } | |
6dd4ee7c | 6006 | |
c03b45b8 MX |
6007 | int ext4_expand_extra_isize(struct inode *inode, |
6008 | unsigned int new_extra_isize, | |
6009 | struct ext4_iloc *iloc) | |
6010 | { | |
6011 | handle_t *handle; | |
6012 | int no_expand; | |
6013 | int error, rc; | |
6014 | ||
6015 | if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) { | |
6016 | brelse(iloc->bh); | |
6017 | return -EOVERFLOW; | |
6dd4ee7c KS |
6018 | } |
6019 | ||
c03b45b8 MX |
6020 | handle = ext4_journal_start(inode, EXT4_HT_INODE, |
6021 | EXT4_DATA_TRANS_BLOCKS(inode->i_sb)); | |
6022 | if (IS_ERR(handle)) { | |
6023 | error = PTR_ERR(handle); | |
6024 | brelse(iloc->bh); | |
6025 | return error; | |
6026 | } | |
6027 | ||
6028 | ext4_write_lock_xattr(inode, &no_expand); | |
6029 | ||
ddccb6db | 6030 | BUFFER_TRACE(iloc->bh, "get_write_access"); |
c03b45b8 | 6031 | error = ext4_journal_get_write_access(handle, iloc->bh); |
3b10fdc6 | 6032 | if (error) { |
c03b45b8 MX |
6033 | brelse(iloc->bh); |
6034 | goto out_stop; | |
3b10fdc6 | 6035 | } |
cf0a5e81 | 6036 | |
c03b45b8 MX |
6037 | error = __ext4_expand_extra_isize(inode, new_extra_isize, iloc, |
6038 | handle, &no_expand); | |
6039 | ||
6040 | rc = ext4_mark_iloc_dirty(handle, inode, iloc); | |
6041 | if (!error) | |
6042 | error = rc; | |
6043 | ||
6044 | ext4_write_unlock_xattr(inode, &no_expand); | |
6045 | out_stop: | |
6046 | ext4_journal_stop(handle); | |
3b10fdc6 | 6047 | return error; |
6dd4ee7c KS |
6048 | } |
6049 | ||
ac27a0ec DK |
6050 | /* |
6051 | * What we do here is to mark the in-core inode as clean with respect to inode | |
6052 | * dirtiness (it may still be data-dirty). | |
6053 | * This means that the in-core inode may be reaped by prune_icache | |
6054 | * without having to perform any I/O. This is a very good thing, | |
6055 | * because *any* task may call prune_icache - even ones which | |
6056 | * have a transaction open against a different journal. | |
6057 | * | |
6058 | * Is this cheating? Not really. Sure, we haven't written the | |
6059 | * inode out, but prune_icache isn't a user-visible syncing function. | |
6060 | * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync) | |
6061 | * we start and wait on commits. | |
ac27a0ec | 6062 | */ |
617ba13b | 6063 | int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode) |
ac27a0ec | 6064 | { |
617ba13b | 6065 | struct ext4_iloc iloc; |
6dd4ee7c | 6066 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
cf0a5e81 | 6067 | int err; |
ac27a0ec DK |
6068 | |
6069 | might_sleep(); | |
7ff9c073 | 6070 | trace_ext4_mark_inode_dirty(inode, _RET_IP_); |
617ba13b | 6071 | err = ext4_reserve_inode_write(handle, inode, &iloc); |
5e1021f2 EG |
6072 | if (err) |
6073 | return err; | |
cf0a5e81 MX |
6074 | |
6075 | if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize) | |
6076 | ext4_try_to_expand_extra_isize(inode, sbi->s_want_extra_isize, | |
6077 | iloc, handle); | |
6078 | ||
5e1021f2 | 6079 | return ext4_mark_iloc_dirty(handle, inode, &iloc); |
ac27a0ec DK |
6080 | } |
6081 | ||
6082 | /* | |
617ba13b | 6083 | * ext4_dirty_inode() is called from __mark_inode_dirty() |
ac27a0ec DK |
6084 | * |
6085 | * We're really interested in the case where a file is being extended. | |
6086 | * i_size has been changed by generic_commit_write() and we thus need | |
6087 | * to include the updated inode in the current transaction. | |
6088 | * | |
5dd4056d | 6089 | * Also, dquot_alloc_block() will always dirty the inode when blocks |
ac27a0ec DK |
6090 | * are allocated to the file. |
6091 | * | |
6092 | * If the inode is marked synchronous, we don't honour that here - doing | |
6093 | * so would cause a commit on atime updates, which we don't bother doing. | |
6094 | * We handle synchronous inodes at the highest possible level. | |
0ae45f63 TT |
6095 | * |
6096 | * If only the I_DIRTY_TIME flag is set, we can skip everything. If | |
6097 | * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need | |
6098 | * to copy into the on-disk inode structure are the timestamp files. | |
ac27a0ec | 6099 | */ |
aa385729 | 6100 | void ext4_dirty_inode(struct inode *inode, int flags) |
ac27a0ec | 6101 | { |
ac27a0ec DK |
6102 | handle_t *handle; |
6103 | ||
0ae45f63 TT |
6104 | if (flags == I_DIRTY_TIME) |
6105 | return; | |
9924a92a | 6106 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 2); |
ac27a0ec DK |
6107 | if (IS_ERR(handle)) |
6108 | goto out; | |
f3dc272f | 6109 | |
f3dc272f CW |
6110 | ext4_mark_inode_dirty(handle, inode); |
6111 | ||
617ba13b | 6112 | ext4_journal_stop(handle); |
ac27a0ec DK |
6113 | out: |
6114 | return; | |
6115 | } | |
6116 | ||
617ba13b | 6117 | int ext4_change_inode_journal_flag(struct inode *inode, int val) |
ac27a0ec DK |
6118 | { |
6119 | journal_t *journal; | |
6120 | handle_t *handle; | |
6121 | int err; | |
c8585c6f | 6122 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
ac27a0ec DK |
6123 | |
6124 | /* | |
6125 | * We have to be very careful here: changing a data block's | |
6126 | * journaling status dynamically is dangerous. If we write a | |
6127 | * data block to the journal, change the status and then delete | |
6128 | * that block, we risk forgetting to revoke the old log record | |
6129 | * from the journal and so a subsequent replay can corrupt data. | |
6130 | * So, first we make sure that the journal is empty and that | |
6131 | * nobody is changing anything. | |
6132 | */ | |
6133 | ||
617ba13b | 6134 | journal = EXT4_JOURNAL(inode); |
0390131b FM |
6135 | if (!journal) |
6136 | return 0; | |
d699594d | 6137 | if (is_journal_aborted(journal)) |
ac27a0ec DK |
6138 | return -EROFS; |
6139 | ||
17335dcc | 6140 | /* Wait for all existing dio workers */ |
17335dcc DM |
6141 | inode_dio_wait(inode); |
6142 | ||
4c546592 DJ |
6143 | /* |
6144 | * Before flushing the journal and switching inode's aops, we have | |
6145 | * to flush all dirty data the inode has. There can be outstanding | |
6146 | * delayed allocations, there can be unwritten extents created by | |
6147 | * fallocate or buffered writes in dioread_nolock mode covered by | |
6148 | * dirty data which can be converted only after flushing the dirty | |
6149 | * data (and journalled aops don't know how to handle these cases). | |
6150 | */ | |
6151 | if (val) { | |
6152 | down_write(&EXT4_I(inode)->i_mmap_sem); | |
6153 | err = filemap_write_and_wait(inode->i_mapping); | |
6154 | if (err < 0) { | |
6155 | up_write(&EXT4_I(inode)->i_mmap_sem); | |
4c546592 DJ |
6156 | return err; |
6157 | } | |
6158 | } | |
6159 | ||
c8585c6f | 6160 | percpu_down_write(&sbi->s_journal_flag_rwsem); |
dab291af | 6161 | jbd2_journal_lock_updates(journal); |
ac27a0ec DK |
6162 | |
6163 | /* | |
6164 | * OK, there are no updates running now, and all cached data is | |
6165 | * synced to disk. We are now in a completely consistent state | |
6166 | * which doesn't have anything in the journal, and we know that | |
6167 | * no filesystem updates are running, so it is safe to modify | |
6168 | * the inode's in-core data-journaling state flag now. | |
6169 | */ | |
6170 | ||
6171 | if (val) | |
12e9b892 | 6172 | ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
5872ddaa | 6173 | else { |
4f879ca6 JK |
6174 | err = jbd2_journal_flush(journal); |
6175 | if (err < 0) { | |
6176 | jbd2_journal_unlock_updates(journal); | |
c8585c6f | 6177 | percpu_up_write(&sbi->s_journal_flag_rwsem); |
4f879ca6 JK |
6178 | return err; |
6179 | } | |
12e9b892 | 6180 | ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
5872ddaa | 6181 | } |
617ba13b | 6182 | ext4_set_aops(inode); |
ac27a0ec | 6183 | |
dab291af | 6184 | jbd2_journal_unlock_updates(journal); |
c8585c6f DJ |
6185 | percpu_up_write(&sbi->s_journal_flag_rwsem); |
6186 | ||
4c546592 DJ |
6187 | if (val) |
6188 | up_write(&EXT4_I(inode)->i_mmap_sem); | |
ac27a0ec DK |
6189 | |
6190 | /* Finally we can mark the inode as dirty. */ | |
6191 | ||
9924a92a | 6192 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 1); |
ac27a0ec DK |
6193 | if (IS_ERR(handle)) |
6194 | return PTR_ERR(handle); | |
6195 | ||
617ba13b | 6196 | err = ext4_mark_inode_dirty(handle, inode); |
0390131b | 6197 | ext4_handle_sync(handle); |
617ba13b MC |
6198 | ext4_journal_stop(handle); |
6199 | ext4_std_error(inode->i_sb, err); | |
ac27a0ec DK |
6200 | |
6201 | return err; | |
6202 | } | |
2e9ee850 AK |
6203 | |
6204 | static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh) | |
6205 | { | |
6206 | return !buffer_mapped(bh); | |
6207 | } | |
6208 | ||
401b25aa | 6209 | vm_fault_t ext4_page_mkwrite(struct vm_fault *vmf) |
2e9ee850 | 6210 | { |
11bac800 | 6211 | struct vm_area_struct *vma = vmf->vma; |
c2ec175c | 6212 | struct page *page = vmf->page; |
2e9ee850 AK |
6213 | loff_t size; |
6214 | unsigned long len; | |
401b25aa SJ |
6215 | int err; |
6216 | vm_fault_t ret; | |
2e9ee850 | 6217 | struct file *file = vma->vm_file; |
496ad9aa | 6218 | struct inode *inode = file_inode(file); |
2e9ee850 | 6219 | struct address_space *mapping = inode->i_mapping; |
9ea7df53 JK |
6220 | handle_t *handle; |
6221 | get_block_t *get_block; | |
6222 | int retries = 0; | |
2e9ee850 | 6223 | |
02b016ca TT |
6224 | if (unlikely(IS_IMMUTABLE(inode))) |
6225 | return VM_FAULT_SIGBUS; | |
6226 | ||
8e8ad8a5 | 6227 | sb_start_pagefault(inode->i_sb); |
041bbb6d | 6228 | file_update_time(vma->vm_file); |
ea3d7209 JK |
6229 | |
6230 | down_read(&EXT4_I(inode)->i_mmap_sem); | |
7b4cc978 | 6231 | |
401b25aa SJ |
6232 | err = ext4_convert_inline_data(inode); |
6233 | if (err) | |
7b4cc978 EB |
6234 | goto out_ret; |
6235 | ||
9ea7df53 JK |
6236 | /* Delalloc case is easy... */ |
6237 | if (test_opt(inode->i_sb, DELALLOC) && | |
6238 | !ext4_should_journal_data(inode) && | |
6239 | !ext4_nonda_switch(inode->i_sb)) { | |
6240 | do { | |
401b25aa | 6241 | err = block_page_mkwrite(vma, vmf, |
9ea7df53 | 6242 | ext4_da_get_block_prep); |
401b25aa | 6243 | } while (err == -ENOSPC && |
9ea7df53 JK |
6244 | ext4_should_retry_alloc(inode->i_sb, &retries)); |
6245 | goto out_ret; | |
2e9ee850 | 6246 | } |
0e499890 DW |
6247 | |
6248 | lock_page(page); | |
9ea7df53 JK |
6249 | size = i_size_read(inode); |
6250 | /* Page got truncated from under us? */ | |
6251 | if (page->mapping != mapping || page_offset(page) > size) { | |
6252 | unlock_page(page); | |
6253 | ret = VM_FAULT_NOPAGE; | |
6254 | goto out; | |
0e499890 | 6255 | } |
2e9ee850 | 6256 | |
09cbfeaf KS |
6257 | if (page->index == size >> PAGE_SHIFT) |
6258 | len = size & ~PAGE_MASK; | |
2e9ee850 | 6259 | else |
09cbfeaf | 6260 | len = PAGE_SIZE; |
a827eaff | 6261 | /* |
9ea7df53 JK |
6262 | * Return if we have all the buffers mapped. This avoids the need to do |
6263 | * journal_start/journal_stop which can block and take a long time | |
a827eaff | 6264 | */ |
2e9ee850 | 6265 | if (page_has_buffers(page)) { |
f19d5870 TM |
6266 | if (!ext4_walk_page_buffers(NULL, page_buffers(page), |
6267 | 0, len, NULL, | |
6268 | ext4_bh_unmapped)) { | |
9ea7df53 | 6269 | /* Wait so that we don't change page under IO */ |
1d1d1a76 | 6270 | wait_for_stable_page(page); |
9ea7df53 JK |
6271 | ret = VM_FAULT_LOCKED; |
6272 | goto out; | |
a827eaff | 6273 | } |
2e9ee850 | 6274 | } |
a827eaff | 6275 | unlock_page(page); |
9ea7df53 JK |
6276 | /* OK, we need to fill the hole... */ |
6277 | if (ext4_should_dioread_nolock(inode)) | |
705965bd | 6278 | get_block = ext4_get_block_unwritten; |
9ea7df53 JK |
6279 | else |
6280 | get_block = ext4_get_block; | |
6281 | retry_alloc: | |
9924a92a TT |
6282 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
6283 | ext4_writepage_trans_blocks(inode)); | |
9ea7df53 | 6284 | if (IS_ERR(handle)) { |
c2ec175c | 6285 | ret = VM_FAULT_SIGBUS; |
9ea7df53 JK |
6286 | goto out; |
6287 | } | |
401b25aa SJ |
6288 | err = block_page_mkwrite(vma, vmf, get_block); |
6289 | if (!err && ext4_should_journal_data(inode)) { | |
f19d5870 | 6290 | if (ext4_walk_page_buffers(handle, page_buffers(page), 0, |
09cbfeaf | 6291 | PAGE_SIZE, NULL, do_journal_get_write_access)) { |
9ea7df53 JK |
6292 | unlock_page(page); |
6293 | ret = VM_FAULT_SIGBUS; | |
fcbb5515 | 6294 | ext4_journal_stop(handle); |
9ea7df53 JK |
6295 | goto out; |
6296 | } | |
6297 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); | |
6298 | } | |
6299 | ext4_journal_stop(handle); | |
401b25aa | 6300 | if (err == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) |
9ea7df53 JK |
6301 | goto retry_alloc; |
6302 | out_ret: | |
401b25aa | 6303 | ret = block_page_mkwrite_return(err); |
9ea7df53 | 6304 | out: |
ea3d7209 | 6305 | up_read(&EXT4_I(inode)->i_mmap_sem); |
8e8ad8a5 | 6306 | sb_end_pagefault(inode->i_sb); |
2e9ee850 AK |
6307 | return ret; |
6308 | } | |
ea3d7209 | 6309 | |
401b25aa | 6310 | vm_fault_t ext4_filemap_fault(struct vm_fault *vmf) |
ea3d7209 | 6311 | { |
11bac800 | 6312 | struct inode *inode = file_inode(vmf->vma->vm_file); |
401b25aa | 6313 | vm_fault_t ret; |
ea3d7209 JK |
6314 | |
6315 | down_read(&EXT4_I(inode)->i_mmap_sem); | |
401b25aa | 6316 | ret = filemap_fault(vmf); |
ea3d7209 JK |
6317 | up_read(&EXT4_I(inode)->i_mmap_sem); |
6318 | ||
401b25aa | 6319 | return ret; |
ea3d7209 | 6320 | } |