]>
Commit | Line | Data |
---|---|---|
ac27a0ec | 1 | /* |
617ba13b | 2 | * linux/fs/ext4/inode.c |
ac27a0ec DK |
3 | * |
4 | * Copyright (C) 1992, 1993, 1994, 1995 | |
5 | * Remy Card (card@masi.ibp.fr) | |
6 | * Laboratoire MASI - Institut Blaise Pascal | |
7 | * Universite Pierre et Marie Curie (Paris VI) | |
8 | * | |
9 | * from | |
10 | * | |
11 | * linux/fs/minix/inode.c | |
12 | * | |
13 | * Copyright (C) 1991, 1992 Linus Torvalds | |
14 | * | |
ac27a0ec DK |
15 | * 64-bit file support on 64-bit platforms by Jakub Jelinek |
16 | * (jj@sunsite.ms.mff.cuni.cz) | |
17 | * | |
617ba13b | 18 | * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000 |
ac27a0ec DK |
19 | */ |
20 | ||
ac27a0ec DK |
21 | #include <linux/fs.h> |
22 | #include <linux/time.h> | |
ac27a0ec DK |
23 | #include <linux/highuid.h> |
24 | #include <linux/pagemap.h> | |
25 | #include <linux/quotaops.h> | |
26 | #include <linux/string.h> | |
27 | #include <linux/buffer_head.h> | |
28 | #include <linux/writeback.h> | |
64769240 | 29 | #include <linux/pagevec.h> |
ac27a0ec | 30 | #include <linux/mpage.h> |
e83c1397 | 31 | #include <linux/namei.h> |
ac27a0ec DK |
32 | #include <linux/uio.h> |
33 | #include <linux/bio.h> | |
4c0425ff | 34 | #include <linux/workqueue.h> |
744692dc | 35 | #include <linux/kernel.h> |
6db26ffc | 36 | #include <linux/printk.h> |
5a0e3ad6 | 37 | #include <linux/slab.h> |
00a1a053 | 38 | #include <linux/bitops.h> |
9bffad1e | 39 | |
3dcf5451 | 40 | #include "ext4_jbd2.h" |
ac27a0ec DK |
41 | #include "xattr.h" |
42 | #include "acl.h" | |
9f125d64 | 43 | #include "truncate.h" |
ac27a0ec | 44 | |
9bffad1e TT |
45 | #include <trace/events/ext4.h> |
46 | ||
a1d6cc56 AK |
47 | #define MPAGE_DA_EXTENT_TAIL 0x01 |
48 | ||
814525f4 DW |
49 | static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw, |
50 | struct ext4_inode_info *ei) | |
51 | { | |
52 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
53 | __u16 csum_lo; | |
54 | __u16 csum_hi = 0; | |
55 | __u32 csum; | |
56 | ||
171a7f21 | 57 | csum_lo = le16_to_cpu(raw->i_checksum_lo); |
814525f4 DW |
58 | raw->i_checksum_lo = 0; |
59 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && | |
60 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) { | |
171a7f21 | 61 | csum_hi = le16_to_cpu(raw->i_checksum_hi); |
814525f4 DW |
62 | raw->i_checksum_hi = 0; |
63 | } | |
64 | ||
65 | csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw, | |
66 | EXT4_INODE_SIZE(inode->i_sb)); | |
67 | ||
171a7f21 | 68 | raw->i_checksum_lo = cpu_to_le16(csum_lo); |
814525f4 DW |
69 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && |
70 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) | |
171a7f21 | 71 | raw->i_checksum_hi = cpu_to_le16(csum_hi); |
814525f4 DW |
72 | |
73 | return csum; | |
74 | } | |
75 | ||
76 | static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw, | |
77 | struct ext4_inode_info *ei) | |
78 | { | |
79 | __u32 provided, calculated; | |
80 | ||
81 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != | |
82 | cpu_to_le32(EXT4_OS_LINUX) || | |
9aa5d32b | 83 | !ext4_has_metadata_csum(inode->i_sb)) |
814525f4 DW |
84 | return 1; |
85 | ||
86 | provided = le16_to_cpu(raw->i_checksum_lo); | |
87 | calculated = ext4_inode_csum(inode, raw, ei); | |
88 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && | |
89 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) | |
90 | provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16; | |
91 | else | |
92 | calculated &= 0xFFFF; | |
93 | ||
94 | return provided == calculated; | |
95 | } | |
96 | ||
97 | static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw, | |
98 | struct ext4_inode_info *ei) | |
99 | { | |
100 | __u32 csum; | |
101 | ||
102 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != | |
103 | cpu_to_le32(EXT4_OS_LINUX) || | |
9aa5d32b | 104 | !ext4_has_metadata_csum(inode->i_sb)) |
814525f4 DW |
105 | return; |
106 | ||
107 | csum = ext4_inode_csum(inode, raw, ei); | |
108 | raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF); | |
109 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && | |
110 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) | |
111 | raw->i_checksum_hi = cpu_to_le16(csum >> 16); | |
112 | } | |
113 | ||
678aaf48 JK |
114 | static inline int ext4_begin_ordered_truncate(struct inode *inode, |
115 | loff_t new_size) | |
116 | { | |
7ff9c073 | 117 | trace_ext4_begin_ordered_truncate(inode, new_size); |
8aefcd55 TT |
118 | /* |
119 | * If jinode is zero, then we never opened the file for | |
120 | * writing, so there's no need to call | |
121 | * jbd2_journal_begin_ordered_truncate() since there's no | |
122 | * outstanding writes we need to flush. | |
123 | */ | |
124 | if (!EXT4_I(inode)->jinode) | |
125 | return 0; | |
126 | return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode), | |
127 | EXT4_I(inode)->jinode, | |
128 | new_size); | |
678aaf48 JK |
129 | } |
130 | ||
d47992f8 LC |
131 | static void ext4_invalidatepage(struct page *page, unsigned int offset, |
132 | unsigned int length); | |
cb20d518 TT |
133 | static int __ext4_journalled_writepage(struct page *page, unsigned int len); |
134 | static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh); | |
fffb2739 JK |
135 | static int ext4_meta_trans_blocks(struct inode *inode, int lblocks, |
136 | int pextents); | |
64769240 | 137 | |
ac27a0ec DK |
138 | /* |
139 | * Test whether an inode is a fast symlink. | |
140 | */ | |
f348c252 | 141 | int ext4_inode_is_fast_symlink(struct inode *inode) |
ac27a0ec | 142 | { |
65eddb56 YY |
143 | int ea_blocks = EXT4_I(inode)->i_file_acl ? |
144 | EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0; | |
ac27a0ec | 145 | |
bd9db175 ZL |
146 | if (ext4_has_inline_data(inode)) |
147 | return 0; | |
148 | ||
ac27a0ec DK |
149 | return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0); |
150 | } | |
151 | ||
ac27a0ec DK |
152 | /* |
153 | * Restart the transaction associated with *handle. This does a commit, | |
154 | * so before we call here everything must be consistently dirtied against | |
155 | * this transaction. | |
156 | */ | |
fa5d1113 | 157 | int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode, |
487caeef | 158 | int nblocks) |
ac27a0ec | 159 | { |
487caeef JK |
160 | int ret; |
161 | ||
162 | /* | |
e35fd660 | 163 | * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this |
487caeef JK |
164 | * moment, get_block can be called only for blocks inside i_size since |
165 | * page cache has been already dropped and writes are blocked by | |
166 | * i_mutex. So we can safely drop the i_data_sem here. | |
167 | */ | |
0390131b | 168 | BUG_ON(EXT4_JOURNAL(inode) == NULL); |
ac27a0ec | 169 | jbd_debug(2, "restarting handle %p\n", handle); |
487caeef | 170 | up_write(&EXT4_I(inode)->i_data_sem); |
8e8eaabe | 171 | ret = ext4_journal_restart(handle, nblocks); |
487caeef | 172 | down_write(&EXT4_I(inode)->i_data_sem); |
fa5d1113 | 173 | ext4_discard_preallocations(inode); |
487caeef JK |
174 | |
175 | return ret; | |
ac27a0ec DK |
176 | } |
177 | ||
178 | /* | |
179 | * Called at the last iput() if i_nlink is zero. | |
180 | */ | |
0930fcc1 | 181 | void ext4_evict_inode(struct inode *inode) |
ac27a0ec DK |
182 | { |
183 | handle_t *handle; | |
bc965ab3 | 184 | int err; |
ac27a0ec | 185 | |
7ff9c073 | 186 | trace_ext4_evict_inode(inode); |
2581fdc8 | 187 | |
0930fcc1 | 188 | if (inode->i_nlink) { |
2d859db3 JK |
189 | /* |
190 | * When journalling data dirty buffers are tracked only in the | |
191 | * journal. So although mm thinks everything is clean and | |
192 | * ready for reaping the inode might still have some pages to | |
193 | * write in the running transaction or waiting to be | |
194 | * checkpointed. Thus calling jbd2_journal_invalidatepage() | |
195 | * (via truncate_inode_pages()) to discard these buffers can | |
196 | * cause data loss. Also even if we did not discard these | |
197 | * buffers, we would have no way to find them after the inode | |
198 | * is reaped and thus user could see stale data if he tries to | |
199 | * read them before the transaction is checkpointed. So be | |
200 | * careful and force everything to disk here... We use | |
201 | * ei->i_datasync_tid to store the newest transaction | |
202 | * containing inode's data. | |
203 | * | |
204 | * Note that directories do not have this problem because they | |
205 | * don't use page cache. | |
206 | */ | |
207 | if (ext4_should_journal_data(inode) && | |
2b405bfa TT |
208 | (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) && |
209 | inode->i_ino != EXT4_JOURNAL_INO) { | |
2d859db3 JK |
210 | journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; |
211 | tid_t commit_tid = EXT4_I(inode)->i_datasync_tid; | |
212 | ||
d76a3a77 | 213 | jbd2_complete_transaction(journal, commit_tid); |
2d859db3 JK |
214 | filemap_write_and_wait(&inode->i_data); |
215 | } | |
91b0abe3 | 216 | truncate_inode_pages_final(&inode->i_data); |
5dc23bdd JK |
217 | |
218 | WARN_ON(atomic_read(&EXT4_I(inode)->i_ioend_count)); | |
0930fcc1 AV |
219 | goto no_delete; |
220 | } | |
221 | ||
e2bfb088 TT |
222 | if (is_bad_inode(inode)) |
223 | goto no_delete; | |
224 | dquot_initialize(inode); | |
907f4554 | 225 | |
678aaf48 JK |
226 | if (ext4_should_order_data(inode)) |
227 | ext4_begin_ordered_truncate(inode, 0); | |
91b0abe3 | 228 | truncate_inode_pages_final(&inode->i_data); |
ac27a0ec | 229 | |
5dc23bdd | 230 | WARN_ON(atomic_read(&EXT4_I(inode)->i_ioend_count)); |
ac27a0ec | 231 | |
8e8ad8a5 JK |
232 | /* |
233 | * Protect us against freezing - iput() caller didn't have to have any | |
234 | * protection against it | |
235 | */ | |
236 | sb_start_intwrite(inode->i_sb); | |
9924a92a TT |
237 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, |
238 | ext4_blocks_for_truncate(inode)+3); | |
ac27a0ec | 239 | if (IS_ERR(handle)) { |
bc965ab3 | 240 | ext4_std_error(inode->i_sb, PTR_ERR(handle)); |
ac27a0ec DK |
241 | /* |
242 | * If we're going to skip the normal cleanup, we still need to | |
243 | * make sure that the in-core orphan linked list is properly | |
244 | * cleaned up. | |
245 | */ | |
617ba13b | 246 | ext4_orphan_del(NULL, inode); |
8e8ad8a5 | 247 | sb_end_intwrite(inode->i_sb); |
ac27a0ec DK |
248 | goto no_delete; |
249 | } | |
250 | ||
251 | if (IS_SYNC(inode)) | |
0390131b | 252 | ext4_handle_sync(handle); |
ac27a0ec | 253 | inode->i_size = 0; |
bc965ab3 TT |
254 | err = ext4_mark_inode_dirty(handle, inode); |
255 | if (err) { | |
12062ddd | 256 | ext4_warning(inode->i_sb, |
bc965ab3 TT |
257 | "couldn't mark inode dirty (err %d)", err); |
258 | goto stop_handle; | |
259 | } | |
ac27a0ec | 260 | if (inode->i_blocks) |
617ba13b | 261 | ext4_truncate(inode); |
bc965ab3 TT |
262 | |
263 | /* | |
264 | * ext4_ext_truncate() doesn't reserve any slop when it | |
265 | * restarts journal transactions; therefore there may not be | |
266 | * enough credits left in the handle to remove the inode from | |
267 | * the orphan list and set the dtime field. | |
268 | */ | |
0390131b | 269 | if (!ext4_handle_has_enough_credits(handle, 3)) { |
bc965ab3 TT |
270 | err = ext4_journal_extend(handle, 3); |
271 | if (err > 0) | |
272 | err = ext4_journal_restart(handle, 3); | |
273 | if (err != 0) { | |
12062ddd | 274 | ext4_warning(inode->i_sb, |
bc965ab3 TT |
275 | "couldn't extend journal (err %d)", err); |
276 | stop_handle: | |
277 | ext4_journal_stop(handle); | |
45388219 | 278 | ext4_orphan_del(NULL, inode); |
8e8ad8a5 | 279 | sb_end_intwrite(inode->i_sb); |
bc965ab3 TT |
280 | goto no_delete; |
281 | } | |
282 | } | |
283 | ||
ac27a0ec | 284 | /* |
617ba13b | 285 | * Kill off the orphan record which ext4_truncate created. |
ac27a0ec | 286 | * AKPM: I think this can be inside the above `if'. |
617ba13b | 287 | * Note that ext4_orphan_del() has to be able to cope with the |
ac27a0ec | 288 | * deletion of a non-existent orphan - this is because we don't |
617ba13b | 289 | * know if ext4_truncate() actually created an orphan record. |
ac27a0ec DK |
290 | * (Well, we could do this if we need to, but heck - it works) |
291 | */ | |
617ba13b MC |
292 | ext4_orphan_del(handle, inode); |
293 | EXT4_I(inode)->i_dtime = get_seconds(); | |
ac27a0ec DK |
294 | |
295 | /* | |
296 | * One subtle ordering requirement: if anything has gone wrong | |
297 | * (transaction abort, IO errors, whatever), then we can still | |
298 | * do these next steps (the fs will already have been marked as | |
299 | * having errors), but we can't free the inode if the mark_dirty | |
300 | * fails. | |
301 | */ | |
617ba13b | 302 | if (ext4_mark_inode_dirty(handle, inode)) |
ac27a0ec | 303 | /* If that failed, just do the required in-core inode clear. */ |
0930fcc1 | 304 | ext4_clear_inode(inode); |
ac27a0ec | 305 | else |
617ba13b MC |
306 | ext4_free_inode(handle, inode); |
307 | ext4_journal_stop(handle); | |
8e8ad8a5 | 308 | sb_end_intwrite(inode->i_sb); |
ac27a0ec DK |
309 | return; |
310 | no_delete: | |
0930fcc1 | 311 | ext4_clear_inode(inode); /* We must guarantee clearing of inode... */ |
ac27a0ec DK |
312 | } |
313 | ||
a9e7f447 DM |
314 | #ifdef CONFIG_QUOTA |
315 | qsize_t *ext4_get_reserved_space(struct inode *inode) | |
60e58e0f | 316 | { |
a9e7f447 | 317 | return &EXT4_I(inode)->i_reserved_quota; |
60e58e0f | 318 | } |
a9e7f447 | 319 | #endif |
9d0be502 | 320 | |
0637c6f4 TT |
321 | /* |
322 | * Called with i_data_sem down, which is important since we can call | |
323 | * ext4_discard_preallocations() from here. | |
324 | */ | |
5f634d06 AK |
325 | void ext4_da_update_reserve_space(struct inode *inode, |
326 | int used, int quota_claim) | |
12219aea AK |
327 | { |
328 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0637c6f4 | 329 | struct ext4_inode_info *ei = EXT4_I(inode); |
0637c6f4 TT |
330 | |
331 | spin_lock(&ei->i_block_reservation_lock); | |
d8990240 | 332 | trace_ext4_da_update_reserve_space(inode, used, quota_claim); |
0637c6f4 | 333 | if (unlikely(used > ei->i_reserved_data_blocks)) { |
8de5c325 | 334 | ext4_warning(inode->i_sb, "%s: ino %lu, used %d " |
1084f252 | 335 | "with only %d reserved data blocks", |
0637c6f4 TT |
336 | __func__, inode->i_ino, used, |
337 | ei->i_reserved_data_blocks); | |
338 | WARN_ON(1); | |
339 | used = ei->i_reserved_data_blocks; | |
340 | } | |
12219aea | 341 | |
0637c6f4 TT |
342 | /* Update per-inode reservations */ |
343 | ei->i_reserved_data_blocks -= used; | |
71d4f7d0 | 344 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, used); |
6bc6e63f | 345 | |
12219aea | 346 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
60e58e0f | 347 | |
72b8ab9d ES |
348 | /* Update quota subsystem for data blocks */ |
349 | if (quota_claim) | |
7b415bf6 | 350 | dquot_claim_block(inode, EXT4_C2B(sbi, used)); |
72b8ab9d | 351 | else { |
5f634d06 AK |
352 | /* |
353 | * We did fallocate with an offset that is already delayed | |
354 | * allocated. So on delayed allocated writeback we should | |
72b8ab9d | 355 | * not re-claim the quota for fallocated blocks. |
5f634d06 | 356 | */ |
7b415bf6 | 357 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, used)); |
5f634d06 | 358 | } |
d6014301 AK |
359 | |
360 | /* | |
361 | * If we have done all the pending block allocations and if | |
362 | * there aren't any writers on the inode, we can discard the | |
363 | * inode's preallocations. | |
364 | */ | |
0637c6f4 TT |
365 | if ((ei->i_reserved_data_blocks == 0) && |
366 | (atomic_read(&inode->i_writecount) == 0)) | |
d6014301 | 367 | ext4_discard_preallocations(inode); |
12219aea AK |
368 | } |
369 | ||
e29136f8 | 370 | static int __check_block_validity(struct inode *inode, const char *func, |
c398eda0 TT |
371 | unsigned int line, |
372 | struct ext4_map_blocks *map) | |
6fd058f7 | 373 | { |
24676da4 TT |
374 | if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk, |
375 | map->m_len)) { | |
c398eda0 TT |
376 | ext4_error_inode(inode, func, line, map->m_pblk, |
377 | "lblock %lu mapped to illegal pblock " | |
378 | "(length %d)", (unsigned long) map->m_lblk, | |
379 | map->m_len); | |
6fd058f7 TT |
380 | return -EIO; |
381 | } | |
382 | return 0; | |
383 | } | |
384 | ||
e29136f8 | 385 | #define check_block_validity(inode, map) \ |
c398eda0 | 386 | __check_block_validity((inode), __func__, __LINE__, (map)) |
e29136f8 | 387 | |
921f266b DM |
388 | #ifdef ES_AGGRESSIVE_TEST |
389 | static void ext4_map_blocks_es_recheck(handle_t *handle, | |
390 | struct inode *inode, | |
391 | struct ext4_map_blocks *es_map, | |
392 | struct ext4_map_blocks *map, | |
393 | int flags) | |
394 | { | |
395 | int retval; | |
396 | ||
397 | map->m_flags = 0; | |
398 | /* | |
399 | * There is a race window that the result is not the same. | |
400 | * e.g. xfstests #223 when dioread_nolock enables. The reason | |
401 | * is that we lookup a block mapping in extent status tree with | |
402 | * out taking i_data_sem. So at the time the unwritten extent | |
403 | * could be converted. | |
404 | */ | |
405 | if (!(flags & EXT4_GET_BLOCKS_NO_LOCK)) | |
c8b459f4 | 406 | down_read(&EXT4_I(inode)->i_data_sem); |
921f266b DM |
407 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
408 | retval = ext4_ext_map_blocks(handle, inode, map, flags & | |
409 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
410 | } else { | |
411 | retval = ext4_ind_map_blocks(handle, inode, map, flags & | |
412 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
413 | } | |
414 | if (!(flags & EXT4_GET_BLOCKS_NO_LOCK)) | |
415 | up_read((&EXT4_I(inode)->i_data_sem)); | |
921f266b DM |
416 | |
417 | /* | |
418 | * We don't check m_len because extent will be collpased in status | |
419 | * tree. So the m_len might not equal. | |
420 | */ | |
421 | if (es_map->m_lblk != map->m_lblk || | |
422 | es_map->m_flags != map->m_flags || | |
423 | es_map->m_pblk != map->m_pblk) { | |
bdafe42a | 424 | printk("ES cache assertion failed for inode: %lu " |
921f266b DM |
425 | "es_cached ex [%d/%d/%llu/%x] != " |
426 | "found ex [%d/%d/%llu/%x] retval %d flags %x\n", | |
427 | inode->i_ino, es_map->m_lblk, es_map->m_len, | |
428 | es_map->m_pblk, es_map->m_flags, map->m_lblk, | |
429 | map->m_len, map->m_pblk, map->m_flags, | |
430 | retval, flags); | |
431 | } | |
432 | } | |
433 | #endif /* ES_AGGRESSIVE_TEST */ | |
434 | ||
f5ab0d1f | 435 | /* |
e35fd660 | 436 | * The ext4_map_blocks() function tries to look up the requested blocks, |
2b2d6d01 | 437 | * and returns if the blocks are already mapped. |
f5ab0d1f | 438 | * |
f5ab0d1f MC |
439 | * Otherwise it takes the write lock of the i_data_sem and allocate blocks |
440 | * and store the allocated blocks in the result buffer head and mark it | |
441 | * mapped. | |
442 | * | |
e35fd660 TT |
443 | * If file type is extents based, it will call ext4_ext_map_blocks(), |
444 | * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping | |
f5ab0d1f MC |
445 | * based files |
446 | * | |
556615dc LC |
447 | * On success, it returns the number of blocks being mapped or allocated. |
448 | * if create==0 and the blocks are pre-allocated and unwritten block, | |
f5ab0d1f MC |
449 | * the result buffer head is unmapped. If the create ==1, it will make sure |
450 | * the buffer head is mapped. | |
451 | * | |
452 | * It returns 0 if plain look up failed (blocks have not been allocated), in | |
df3ab170 | 453 | * that case, buffer head is unmapped |
f5ab0d1f MC |
454 | * |
455 | * It returns the error in case of allocation failure. | |
456 | */ | |
e35fd660 TT |
457 | int ext4_map_blocks(handle_t *handle, struct inode *inode, |
458 | struct ext4_map_blocks *map, int flags) | |
0e855ac8 | 459 | { |
d100eef2 | 460 | struct extent_status es; |
0e855ac8 | 461 | int retval; |
b8a86845 | 462 | int ret = 0; |
921f266b DM |
463 | #ifdef ES_AGGRESSIVE_TEST |
464 | struct ext4_map_blocks orig_map; | |
465 | ||
466 | memcpy(&orig_map, map, sizeof(*map)); | |
467 | #endif | |
f5ab0d1f | 468 | |
e35fd660 TT |
469 | map->m_flags = 0; |
470 | ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u," | |
471 | "logical block %lu\n", inode->i_ino, flags, map->m_len, | |
472 | (unsigned long) map->m_lblk); | |
d100eef2 | 473 | |
e861b5e9 TT |
474 | /* |
475 | * ext4_map_blocks returns an int, and m_len is an unsigned int | |
476 | */ | |
477 | if (unlikely(map->m_len > INT_MAX)) | |
478 | map->m_len = INT_MAX; | |
479 | ||
4adb6ab3 KM |
480 | /* We can handle the block number less than EXT_MAX_BLOCKS */ |
481 | if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS)) | |
482 | return -EIO; | |
483 | ||
d100eef2 ZL |
484 | /* Lookup extent status tree firstly */ |
485 | if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) { | |
486 | if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) { | |
487 | map->m_pblk = ext4_es_pblock(&es) + | |
488 | map->m_lblk - es.es_lblk; | |
489 | map->m_flags |= ext4_es_is_written(&es) ? | |
490 | EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN; | |
491 | retval = es.es_len - (map->m_lblk - es.es_lblk); | |
492 | if (retval > map->m_len) | |
493 | retval = map->m_len; | |
494 | map->m_len = retval; | |
495 | } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) { | |
496 | retval = 0; | |
497 | } else { | |
498 | BUG_ON(1); | |
499 | } | |
921f266b DM |
500 | #ifdef ES_AGGRESSIVE_TEST |
501 | ext4_map_blocks_es_recheck(handle, inode, map, | |
502 | &orig_map, flags); | |
503 | #endif | |
d100eef2 ZL |
504 | goto found; |
505 | } | |
506 | ||
4df3d265 | 507 | /* |
b920c755 TT |
508 | * Try to see if we can get the block without requesting a new |
509 | * file system block. | |
4df3d265 | 510 | */ |
729f52c6 | 511 | if (!(flags & EXT4_GET_BLOCKS_NO_LOCK)) |
c8b459f4 | 512 | down_read(&EXT4_I(inode)->i_data_sem); |
12e9b892 | 513 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
a4e5d88b DM |
514 | retval = ext4_ext_map_blocks(handle, inode, map, flags & |
515 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
0e855ac8 | 516 | } else { |
a4e5d88b DM |
517 | retval = ext4_ind_map_blocks(handle, inode, map, flags & |
518 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
0e855ac8 | 519 | } |
f7fec032 | 520 | if (retval > 0) { |
3be78c73 | 521 | unsigned int status; |
f7fec032 | 522 | |
44fb851d ZL |
523 | if (unlikely(retval != map->m_len)) { |
524 | ext4_warning(inode->i_sb, | |
525 | "ES len assertion failed for inode " | |
526 | "%lu: retval %d != map->m_len %d", | |
527 | inode->i_ino, retval, map->m_len); | |
528 | WARN_ON(1); | |
921f266b | 529 | } |
921f266b | 530 | |
f7fec032 ZL |
531 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
532 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; | |
533 | if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) && | |
d2dc317d | 534 | !(status & EXTENT_STATUS_WRITTEN) && |
f7fec032 ZL |
535 | ext4_find_delalloc_range(inode, map->m_lblk, |
536 | map->m_lblk + map->m_len - 1)) | |
537 | status |= EXTENT_STATUS_DELAYED; | |
538 | ret = ext4_es_insert_extent(inode, map->m_lblk, | |
539 | map->m_len, map->m_pblk, status); | |
540 | if (ret < 0) | |
541 | retval = ret; | |
542 | } | |
729f52c6 ZL |
543 | if (!(flags & EXT4_GET_BLOCKS_NO_LOCK)) |
544 | up_read((&EXT4_I(inode)->i_data_sem)); | |
f5ab0d1f | 545 | |
d100eef2 | 546 | found: |
e35fd660 | 547 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
b8a86845 | 548 | ret = check_block_validity(inode, map); |
6fd058f7 TT |
549 | if (ret != 0) |
550 | return ret; | |
551 | } | |
552 | ||
f5ab0d1f | 553 | /* If it is only a block(s) look up */ |
c2177057 | 554 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) |
f5ab0d1f MC |
555 | return retval; |
556 | ||
557 | /* | |
558 | * Returns if the blocks have already allocated | |
559 | * | |
560 | * Note that if blocks have been preallocated | |
df3ab170 | 561 | * ext4_ext_get_block() returns the create = 0 |
f5ab0d1f MC |
562 | * with buffer head unmapped. |
563 | */ | |
e35fd660 | 564 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) |
b8a86845 LC |
565 | /* |
566 | * If we need to convert extent to unwritten | |
567 | * we continue and do the actual work in | |
568 | * ext4_ext_map_blocks() | |
569 | */ | |
570 | if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) | |
571 | return retval; | |
4df3d265 | 572 | |
2a8964d6 | 573 | /* |
a25a4e1a ZL |
574 | * Here we clear m_flags because after allocating an new extent, |
575 | * it will be set again. | |
2a8964d6 | 576 | */ |
a25a4e1a | 577 | map->m_flags &= ~EXT4_MAP_FLAGS; |
2a8964d6 | 578 | |
4df3d265 | 579 | /* |
556615dc | 580 | * New blocks allocate and/or writing to unwritten extent |
f5ab0d1f | 581 | * will possibly result in updating i_data, so we take |
d91bd2c1 | 582 | * the write lock of i_data_sem, and call get_block() |
f5ab0d1f | 583 | * with create == 1 flag. |
4df3d265 | 584 | */ |
c8b459f4 | 585 | down_write(&EXT4_I(inode)->i_data_sem); |
d2a17637 | 586 | |
4df3d265 AK |
587 | /* |
588 | * We need to check for EXT4 here because migrate | |
589 | * could have changed the inode type in between | |
590 | */ | |
12e9b892 | 591 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
e35fd660 | 592 | retval = ext4_ext_map_blocks(handle, inode, map, flags); |
0e855ac8 | 593 | } else { |
e35fd660 | 594 | retval = ext4_ind_map_blocks(handle, inode, map, flags); |
267e4db9 | 595 | |
e35fd660 | 596 | if (retval > 0 && map->m_flags & EXT4_MAP_NEW) { |
267e4db9 AK |
597 | /* |
598 | * We allocated new blocks which will result in | |
599 | * i_data's format changing. Force the migrate | |
600 | * to fail by clearing migrate flags | |
601 | */ | |
19f5fb7a | 602 | ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE); |
267e4db9 | 603 | } |
d2a17637 | 604 | |
5f634d06 AK |
605 | /* |
606 | * Update reserved blocks/metadata blocks after successful | |
607 | * block allocation which had been deferred till now. We don't | |
608 | * support fallocate for non extent files. So we can update | |
609 | * reserve space here. | |
610 | */ | |
611 | if ((retval > 0) && | |
1296cc85 | 612 | (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)) |
5f634d06 AK |
613 | ext4_da_update_reserve_space(inode, retval, 1); |
614 | } | |
2ac3b6e0 | 615 | |
f7fec032 | 616 | if (retval > 0) { |
3be78c73 | 617 | unsigned int status; |
f7fec032 | 618 | |
44fb851d ZL |
619 | if (unlikely(retval != map->m_len)) { |
620 | ext4_warning(inode->i_sb, | |
621 | "ES len assertion failed for inode " | |
622 | "%lu: retval %d != map->m_len %d", | |
623 | inode->i_ino, retval, map->m_len); | |
624 | WARN_ON(1); | |
921f266b | 625 | } |
921f266b | 626 | |
adb23551 ZL |
627 | /* |
628 | * If the extent has been zeroed out, we don't need to update | |
629 | * extent status tree. | |
630 | */ | |
631 | if ((flags & EXT4_GET_BLOCKS_PRE_IO) && | |
632 | ext4_es_lookup_extent(inode, map->m_lblk, &es)) { | |
633 | if (ext4_es_is_written(&es)) | |
634 | goto has_zeroout; | |
635 | } | |
f7fec032 ZL |
636 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
637 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; | |
638 | if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) && | |
d2dc317d | 639 | !(status & EXTENT_STATUS_WRITTEN) && |
f7fec032 ZL |
640 | ext4_find_delalloc_range(inode, map->m_lblk, |
641 | map->m_lblk + map->m_len - 1)) | |
642 | status |= EXTENT_STATUS_DELAYED; | |
643 | ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, | |
644 | map->m_pblk, status); | |
645 | if (ret < 0) | |
646 | retval = ret; | |
5356f261 AK |
647 | } |
648 | ||
adb23551 | 649 | has_zeroout: |
4df3d265 | 650 | up_write((&EXT4_I(inode)->i_data_sem)); |
e35fd660 | 651 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
b8a86845 | 652 | ret = check_block_validity(inode, map); |
6fd058f7 TT |
653 | if (ret != 0) |
654 | return ret; | |
655 | } | |
0e855ac8 AK |
656 | return retval; |
657 | } | |
658 | ||
923ae0ff RZ |
659 | static void ext4_end_io_unwritten(struct buffer_head *bh, int uptodate) |
660 | { | |
661 | struct inode *inode = bh->b_assoc_map->host; | |
662 | /* XXX: breaks on 32-bit > 16GB. Is that even supported? */ | |
663 | loff_t offset = (loff_t)(uintptr_t)bh->b_private << inode->i_blkbits; | |
664 | int err; | |
665 | if (!uptodate) | |
666 | return; | |
667 | WARN_ON(!buffer_unwritten(bh)); | |
668 | err = ext4_convert_unwritten_extents(NULL, inode, offset, bh->b_size); | |
669 | } | |
670 | ||
f3bd1f3f MC |
671 | /* Maximum number of blocks we map for direct IO at once. */ |
672 | #define DIO_MAX_BLOCKS 4096 | |
673 | ||
2ed88685 TT |
674 | static int _ext4_get_block(struct inode *inode, sector_t iblock, |
675 | struct buffer_head *bh, int flags) | |
ac27a0ec | 676 | { |
3e4fdaf8 | 677 | handle_t *handle = ext4_journal_current_handle(); |
2ed88685 | 678 | struct ext4_map_blocks map; |
7fb5409d | 679 | int ret = 0, started = 0; |
f3bd1f3f | 680 | int dio_credits; |
ac27a0ec | 681 | |
46c7f254 TM |
682 | if (ext4_has_inline_data(inode)) |
683 | return -ERANGE; | |
684 | ||
2ed88685 TT |
685 | map.m_lblk = iblock; |
686 | map.m_len = bh->b_size >> inode->i_blkbits; | |
687 | ||
8b0f165f | 688 | if (flags && !(flags & EXT4_GET_BLOCKS_NO_LOCK) && !handle) { |
7fb5409d | 689 | /* Direct IO write... */ |
2ed88685 TT |
690 | if (map.m_len > DIO_MAX_BLOCKS) |
691 | map.m_len = DIO_MAX_BLOCKS; | |
692 | dio_credits = ext4_chunk_trans_blocks(inode, map.m_len); | |
9924a92a TT |
693 | handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, |
694 | dio_credits); | |
7fb5409d | 695 | if (IS_ERR(handle)) { |
ac27a0ec | 696 | ret = PTR_ERR(handle); |
2ed88685 | 697 | return ret; |
ac27a0ec | 698 | } |
7fb5409d | 699 | started = 1; |
ac27a0ec DK |
700 | } |
701 | ||
2ed88685 | 702 | ret = ext4_map_blocks(handle, inode, &map, flags); |
7fb5409d | 703 | if (ret > 0) { |
7b7a8665 CH |
704 | ext4_io_end_t *io_end = ext4_inode_aio(inode); |
705 | ||
2ed88685 TT |
706 | map_bh(bh, inode->i_sb, map.m_pblk); |
707 | bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags; | |
923ae0ff RZ |
708 | if (IS_DAX(inode) && buffer_unwritten(bh) && !io_end) { |
709 | bh->b_assoc_map = inode->i_mapping; | |
710 | bh->b_private = (void *)(unsigned long)iblock; | |
711 | bh->b_end_io = ext4_end_io_unwritten; | |
712 | } | |
7b7a8665 CH |
713 | if (io_end && io_end->flag & EXT4_IO_END_UNWRITTEN) |
714 | set_buffer_defer_completion(bh); | |
2ed88685 | 715 | bh->b_size = inode->i_sb->s_blocksize * map.m_len; |
7fb5409d | 716 | ret = 0; |
ac27a0ec | 717 | } |
7fb5409d JK |
718 | if (started) |
719 | ext4_journal_stop(handle); | |
ac27a0ec DK |
720 | return ret; |
721 | } | |
722 | ||
2ed88685 TT |
723 | int ext4_get_block(struct inode *inode, sector_t iblock, |
724 | struct buffer_head *bh, int create) | |
725 | { | |
726 | return _ext4_get_block(inode, iblock, bh, | |
727 | create ? EXT4_GET_BLOCKS_CREATE : 0); | |
728 | } | |
729 | ||
ac27a0ec DK |
730 | /* |
731 | * `handle' can be NULL if create is zero | |
732 | */ | |
617ba13b | 733 | struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode, |
c5e298ae | 734 | ext4_lblk_t block, int map_flags) |
ac27a0ec | 735 | { |
2ed88685 TT |
736 | struct ext4_map_blocks map; |
737 | struct buffer_head *bh; | |
c5e298ae | 738 | int create = map_flags & EXT4_GET_BLOCKS_CREATE; |
10560082 | 739 | int err; |
ac27a0ec DK |
740 | |
741 | J_ASSERT(handle != NULL || create == 0); | |
742 | ||
2ed88685 TT |
743 | map.m_lblk = block; |
744 | map.m_len = 1; | |
c5e298ae | 745 | err = ext4_map_blocks(handle, inode, &map, map_flags); |
ac27a0ec | 746 | |
10560082 TT |
747 | if (err == 0) |
748 | return create ? ERR_PTR(-ENOSPC) : NULL; | |
2ed88685 | 749 | if (err < 0) |
10560082 | 750 | return ERR_PTR(err); |
2ed88685 TT |
751 | |
752 | bh = sb_getblk(inode->i_sb, map.m_pblk); | |
10560082 TT |
753 | if (unlikely(!bh)) |
754 | return ERR_PTR(-ENOMEM); | |
2ed88685 TT |
755 | if (map.m_flags & EXT4_MAP_NEW) { |
756 | J_ASSERT(create != 0); | |
757 | J_ASSERT(handle != NULL); | |
ac27a0ec | 758 | |
2ed88685 TT |
759 | /* |
760 | * Now that we do not always journal data, we should | |
761 | * keep in mind whether this should always journal the | |
762 | * new buffer as metadata. For now, regular file | |
763 | * writes use ext4_get_block instead, so it's not a | |
764 | * problem. | |
765 | */ | |
766 | lock_buffer(bh); | |
767 | BUFFER_TRACE(bh, "call get_create_access"); | |
10560082 TT |
768 | err = ext4_journal_get_create_access(handle, bh); |
769 | if (unlikely(err)) { | |
770 | unlock_buffer(bh); | |
771 | goto errout; | |
772 | } | |
773 | if (!buffer_uptodate(bh)) { | |
2ed88685 TT |
774 | memset(bh->b_data, 0, inode->i_sb->s_blocksize); |
775 | set_buffer_uptodate(bh); | |
ac27a0ec | 776 | } |
2ed88685 TT |
777 | unlock_buffer(bh); |
778 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); | |
779 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
10560082 TT |
780 | if (unlikely(err)) |
781 | goto errout; | |
782 | } else | |
2ed88685 | 783 | BUFFER_TRACE(bh, "not a new buffer"); |
2ed88685 | 784 | return bh; |
10560082 TT |
785 | errout: |
786 | brelse(bh); | |
787 | return ERR_PTR(err); | |
ac27a0ec DK |
788 | } |
789 | ||
617ba13b | 790 | struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode, |
c5e298ae | 791 | ext4_lblk_t block, int map_flags) |
ac27a0ec | 792 | { |
af5bc92d | 793 | struct buffer_head *bh; |
ac27a0ec | 794 | |
c5e298ae | 795 | bh = ext4_getblk(handle, inode, block, map_flags); |
1c215028 | 796 | if (IS_ERR(bh)) |
ac27a0ec | 797 | return bh; |
1c215028 | 798 | if (!bh || buffer_uptodate(bh)) |
ac27a0ec | 799 | return bh; |
65299a3b | 800 | ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh); |
ac27a0ec DK |
801 | wait_on_buffer(bh); |
802 | if (buffer_uptodate(bh)) | |
803 | return bh; | |
804 | put_bh(bh); | |
1c215028 | 805 | return ERR_PTR(-EIO); |
ac27a0ec DK |
806 | } |
807 | ||
f19d5870 TM |
808 | int ext4_walk_page_buffers(handle_t *handle, |
809 | struct buffer_head *head, | |
810 | unsigned from, | |
811 | unsigned to, | |
812 | int *partial, | |
813 | int (*fn)(handle_t *handle, | |
814 | struct buffer_head *bh)) | |
ac27a0ec DK |
815 | { |
816 | struct buffer_head *bh; | |
817 | unsigned block_start, block_end; | |
818 | unsigned blocksize = head->b_size; | |
819 | int err, ret = 0; | |
820 | struct buffer_head *next; | |
821 | ||
af5bc92d TT |
822 | for (bh = head, block_start = 0; |
823 | ret == 0 && (bh != head || !block_start); | |
de9a55b8 | 824 | block_start = block_end, bh = next) { |
ac27a0ec DK |
825 | next = bh->b_this_page; |
826 | block_end = block_start + blocksize; | |
827 | if (block_end <= from || block_start >= to) { | |
828 | if (partial && !buffer_uptodate(bh)) | |
829 | *partial = 1; | |
830 | continue; | |
831 | } | |
832 | err = (*fn)(handle, bh); | |
833 | if (!ret) | |
834 | ret = err; | |
835 | } | |
836 | return ret; | |
837 | } | |
838 | ||
839 | /* | |
840 | * To preserve ordering, it is essential that the hole instantiation and | |
841 | * the data write be encapsulated in a single transaction. We cannot | |
617ba13b | 842 | * close off a transaction and start a new one between the ext4_get_block() |
dab291af | 843 | * and the commit_write(). So doing the jbd2_journal_start at the start of |
ac27a0ec DK |
844 | * prepare_write() is the right place. |
845 | * | |
36ade451 JK |
846 | * Also, this function can nest inside ext4_writepage(). In that case, we |
847 | * *know* that ext4_writepage() has generated enough buffer credits to do the | |
848 | * whole page. So we won't block on the journal in that case, which is good, | |
849 | * because the caller may be PF_MEMALLOC. | |
ac27a0ec | 850 | * |
617ba13b | 851 | * By accident, ext4 can be reentered when a transaction is open via |
ac27a0ec DK |
852 | * quota file writes. If we were to commit the transaction while thus |
853 | * reentered, there can be a deadlock - we would be holding a quota | |
854 | * lock, and the commit would never complete if another thread had a | |
855 | * transaction open and was blocking on the quota lock - a ranking | |
856 | * violation. | |
857 | * | |
dab291af | 858 | * So what we do is to rely on the fact that jbd2_journal_stop/journal_start |
ac27a0ec DK |
859 | * will _not_ run commit under these circumstances because handle->h_ref |
860 | * is elevated. We'll still have enough credits for the tiny quotafile | |
861 | * write. | |
862 | */ | |
f19d5870 TM |
863 | int do_journal_get_write_access(handle_t *handle, |
864 | struct buffer_head *bh) | |
ac27a0ec | 865 | { |
56d35a4c JK |
866 | int dirty = buffer_dirty(bh); |
867 | int ret; | |
868 | ||
ac27a0ec DK |
869 | if (!buffer_mapped(bh) || buffer_freed(bh)) |
870 | return 0; | |
56d35a4c | 871 | /* |
ebdec241 | 872 | * __block_write_begin() could have dirtied some buffers. Clean |
56d35a4c JK |
873 | * the dirty bit as jbd2_journal_get_write_access() could complain |
874 | * otherwise about fs integrity issues. Setting of the dirty bit | |
ebdec241 | 875 | * by __block_write_begin() isn't a real problem here as we clear |
56d35a4c JK |
876 | * the bit before releasing a page lock and thus writeback cannot |
877 | * ever write the buffer. | |
878 | */ | |
879 | if (dirty) | |
880 | clear_buffer_dirty(bh); | |
5d601255 | 881 | BUFFER_TRACE(bh, "get write access"); |
56d35a4c JK |
882 | ret = ext4_journal_get_write_access(handle, bh); |
883 | if (!ret && dirty) | |
884 | ret = ext4_handle_dirty_metadata(handle, NULL, bh); | |
885 | return ret; | |
ac27a0ec DK |
886 | } |
887 | ||
8b0f165f AP |
888 | static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock, |
889 | struct buffer_head *bh_result, int create); | |
2058f83a MH |
890 | |
891 | #ifdef CONFIG_EXT4_FS_ENCRYPTION | |
892 | static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len, | |
893 | get_block_t *get_block) | |
894 | { | |
895 | unsigned from = pos & (PAGE_CACHE_SIZE - 1); | |
896 | unsigned to = from + len; | |
897 | struct inode *inode = page->mapping->host; | |
898 | unsigned block_start, block_end; | |
899 | sector_t block; | |
900 | int err = 0; | |
901 | unsigned blocksize = inode->i_sb->s_blocksize; | |
902 | unsigned bbits; | |
903 | struct buffer_head *bh, *head, *wait[2], **wait_bh = wait; | |
904 | bool decrypt = false; | |
905 | ||
906 | BUG_ON(!PageLocked(page)); | |
907 | BUG_ON(from > PAGE_CACHE_SIZE); | |
908 | BUG_ON(to > PAGE_CACHE_SIZE); | |
909 | BUG_ON(from > to); | |
910 | ||
911 | if (!page_has_buffers(page)) | |
912 | create_empty_buffers(page, blocksize, 0); | |
913 | head = page_buffers(page); | |
914 | bbits = ilog2(blocksize); | |
915 | block = (sector_t)page->index << (PAGE_CACHE_SHIFT - bbits); | |
916 | ||
917 | for (bh = head, block_start = 0; bh != head || !block_start; | |
918 | block++, block_start = block_end, bh = bh->b_this_page) { | |
919 | block_end = block_start + blocksize; | |
920 | if (block_end <= from || block_start >= to) { | |
921 | if (PageUptodate(page)) { | |
922 | if (!buffer_uptodate(bh)) | |
923 | set_buffer_uptodate(bh); | |
924 | } | |
925 | continue; | |
926 | } | |
927 | if (buffer_new(bh)) | |
928 | clear_buffer_new(bh); | |
929 | if (!buffer_mapped(bh)) { | |
930 | WARN_ON(bh->b_size != blocksize); | |
931 | err = get_block(inode, block, bh, 1); | |
932 | if (err) | |
933 | break; | |
934 | if (buffer_new(bh)) { | |
935 | unmap_underlying_metadata(bh->b_bdev, | |
936 | bh->b_blocknr); | |
937 | if (PageUptodate(page)) { | |
938 | clear_buffer_new(bh); | |
939 | set_buffer_uptodate(bh); | |
940 | mark_buffer_dirty(bh); | |
941 | continue; | |
942 | } | |
943 | if (block_end > to || block_start < from) | |
944 | zero_user_segments(page, to, block_end, | |
945 | block_start, from); | |
946 | continue; | |
947 | } | |
948 | } | |
949 | if (PageUptodate(page)) { | |
950 | if (!buffer_uptodate(bh)) | |
951 | set_buffer_uptodate(bh); | |
952 | continue; | |
953 | } | |
954 | if (!buffer_uptodate(bh) && !buffer_delay(bh) && | |
955 | !buffer_unwritten(bh) && | |
956 | (block_start < from || block_end > to)) { | |
957 | ll_rw_block(READ, 1, &bh); | |
958 | *wait_bh++ = bh; | |
959 | decrypt = ext4_encrypted_inode(inode) && | |
960 | S_ISREG(inode->i_mode); | |
961 | } | |
962 | } | |
963 | /* | |
964 | * If we issued read requests, let them complete. | |
965 | */ | |
966 | while (wait_bh > wait) { | |
967 | wait_on_buffer(*--wait_bh); | |
968 | if (!buffer_uptodate(*wait_bh)) | |
969 | err = -EIO; | |
970 | } | |
971 | if (unlikely(err)) | |
972 | page_zero_new_buffers(page, from, to); | |
973 | else if (decrypt) | |
974 | err = ext4_decrypt_one(inode, page); | |
975 | return err; | |
976 | } | |
977 | #endif | |
978 | ||
bfc1af65 | 979 | static int ext4_write_begin(struct file *file, struct address_space *mapping, |
de9a55b8 TT |
980 | loff_t pos, unsigned len, unsigned flags, |
981 | struct page **pagep, void **fsdata) | |
ac27a0ec | 982 | { |
af5bc92d | 983 | struct inode *inode = mapping->host; |
1938a150 | 984 | int ret, needed_blocks; |
ac27a0ec DK |
985 | handle_t *handle; |
986 | int retries = 0; | |
af5bc92d | 987 | struct page *page; |
de9a55b8 | 988 | pgoff_t index; |
af5bc92d | 989 | unsigned from, to; |
bfc1af65 | 990 | |
9bffad1e | 991 | trace_ext4_write_begin(inode, pos, len, flags); |
1938a150 AK |
992 | /* |
993 | * Reserve one block more for addition to orphan list in case | |
994 | * we allocate blocks but write fails for some reason | |
995 | */ | |
996 | needed_blocks = ext4_writepage_trans_blocks(inode) + 1; | |
de9a55b8 | 997 | index = pos >> PAGE_CACHE_SHIFT; |
af5bc92d TT |
998 | from = pos & (PAGE_CACHE_SIZE - 1); |
999 | to = from + len; | |
ac27a0ec | 1000 | |
f19d5870 TM |
1001 | if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { |
1002 | ret = ext4_try_to_write_inline_data(mapping, inode, pos, len, | |
1003 | flags, pagep); | |
1004 | if (ret < 0) | |
47564bfb TT |
1005 | return ret; |
1006 | if (ret == 1) | |
1007 | return 0; | |
f19d5870 TM |
1008 | } |
1009 | ||
47564bfb TT |
1010 | /* |
1011 | * grab_cache_page_write_begin() can take a long time if the | |
1012 | * system is thrashing due to memory pressure, or if the page | |
1013 | * is being written back. So grab it first before we start | |
1014 | * the transaction handle. This also allows us to allocate | |
1015 | * the page (if needed) without using GFP_NOFS. | |
1016 | */ | |
1017 | retry_grab: | |
1018 | page = grab_cache_page_write_begin(mapping, index, flags); | |
1019 | if (!page) | |
1020 | return -ENOMEM; | |
1021 | unlock_page(page); | |
1022 | ||
1023 | retry_journal: | |
9924a92a | 1024 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks); |
af5bc92d | 1025 | if (IS_ERR(handle)) { |
47564bfb TT |
1026 | page_cache_release(page); |
1027 | return PTR_ERR(handle); | |
7479d2b9 | 1028 | } |
ac27a0ec | 1029 | |
47564bfb TT |
1030 | lock_page(page); |
1031 | if (page->mapping != mapping) { | |
1032 | /* The page got truncated from under us */ | |
1033 | unlock_page(page); | |
1034 | page_cache_release(page); | |
cf108bca | 1035 | ext4_journal_stop(handle); |
47564bfb | 1036 | goto retry_grab; |
cf108bca | 1037 | } |
7afe5aa5 DM |
1038 | /* In case writeback began while the page was unlocked */ |
1039 | wait_for_stable_page(page); | |
cf108bca | 1040 | |
2058f83a MH |
1041 | #ifdef CONFIG_EXT4_FS_ENCRYPTION |
1042 | if (ext4_should_dioread_nolock(inode)) | |
1043 | ret = ext4_block_write_begin(page, pos, len, | |
1044 | ext4_get_block_write); | |
1045 | else | |
1046 | ret = ext4_block_write_begin(page, pos, len, | |
1047 | ext4_get_block); | |
1048 | #else | |
744692dc | 1049 | if (ext4_should_dioread_nolock(inode)) |
6e1db88d | 1050 | ret = __block_write_begin(page, pos, len, ext4_get_block_write); |
744692dc | 1051 | else |
6e1db88d | 1052 | ret = __block_write_begin(page, pos, len, ext4_get_block); |
2058f83a | 1053 | #endif |
bfc1af65 | 1054 | if (!ret && ext4_should_journal_data(inode)) { |
f19d5870 TM |
1055 | ret = ext4_walk_page_buffers(handle, page_buffers(page), |
1056 | from, to, NULL, | |
1057 | do_journal_get_write_access); | |
ac27a0ec | 1058 | } |
bfc1af65 NP |
1059 | |
1060 | if (ret) { | |
af5bc92d | 1061 | unlock_page(page); |
ae4d5372 | 1062 | /* |
6e1db88d | 1063 | * __block_write_begin may have instantiated a few blocks |
ae4d5372 AK |
1064 | * outside i_size. Trim these off again. Don't need |
1065 | * i_size_read because we hold i_mutex. | |
1938a150 AK |
1066 | * |
1067 | * Add inode to orphan list in case we crash before | |
1068 | * truncate finishes | |
ae4d5372 | 1069 | */ |
ffacfa7a | 1070 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
1938a150 AK |
1071 | ext4_orphan_add(handle, inode); |
1072 | ||
1073 | ext4_journal_stop(handle); | |
1074 | if (pos + len > inode->i_size) { | |
b9a4207d | 1075 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1076 | /* |
ffacfa7a | 1077 | * If truncate failed early the inode might |
1938a150 AK |
1078 | * still be on the orphan list; we need to |
1079 | * make sure the inode is removed from the | |
1080 | * orphan list in that case. | |
1081 | */ | |
1082 | if (inode->i_nlink) | |
1083 | ext4_orphan_del(NULL, inode); | |
1084 | } | |
bfc1af65 | 1085 | |
47564bfb TT |
1086 | if (ret == -ENOSPC && |
1087 | ext4_should_retry_alloc(inode->i_sb, &retries)) | |
1088 | goto retry_journal; | |
1089 | page_cache_release(page); | |
1090 | return ret; | |
1091 | } | |
1092 | *pagep = page; | |
ac27a0ec DK |
1093 | return ret; |
1094 | } | |
1095 | ||
bfc1af65 NP |
1096 | /* For write_end() in data=journal mode */ |
1097 | static int write_end_fn(handle_t *handle, struct buffer_head *bh) | |
ac27a0ec | 1098 | { |
13fca323 | 1099 | int ret; |
ac27a0ec DK |
1100 | if (!buffer_mapped(bh) || buffer_freed(bh)) |
1101 | return 0; | |
1102 | set_buffer_uptodate(bh); | |
13fca323 TT |
1103 | ret = ext4_handle_dirty_metadata(handle, NULL, bh); |
1104 | clear_buffer_meta(bh); | |
1105 | clear_buffer_prio(bh); | |
1106 | return ret; | |
ac27a0ec DK |
1107 | } |
1108 | ||
eed4333f ZL |
1109 | /* |
1110 | * We need to pick up the new inode size which generic_commit_write gave us | |
1111 | * `file' can be NULL - eg, when called from page_symlink(). | |
1112 | * | |
1113 | * ext4 never places buffers on inode->i_mapping->private_list. metadata | |
1114 | * buffers are managed internally. | |
1115 | */ | |
1116 | static int ext4_write_end(struct file *file, | |
1117 | struct address_space *mapping, | |
1118 | loff_t pos, unsigned len, unsigned copied, | |
1119 | struct page *page, void *fsdata) | |
f8514083 | 1120 | { |
f8514083 | 1121 | handle_t *handle = ext4_journal_current_handle(); |
eed4333f | 1122 | struct inode *inode = mapping->host; |
0572639f | 1123 | loff_t old_size = inode->i_size; |
eed4333f ZL |
1124 | int ret = 0, ret2; |
1125 | int i_size_changed = 0; | |
1126 | ||
1127 | trace_ext4_write_end(inode, pos, len, copied); | |
1128 | if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE)) { | |
1129 | ret = ext4_jbd2_file_inode(handle, inode); | |
1130 | if (ret) { | |
1131 | unlock_page(page); | |
1132 | page_cache_release(page); | |
1133 | goto errout; | |
1134 | } | |
1135 | } | |
f8514083 | 1136 | |
42c832de TT |
1137 | if (ext4_has_inline_data(inode)) { |
1138 | ret = ext4_write_inline_data_end(inode, pos, len, | |
1139 | copied, page); | |
1140 | if (ret < 0) | |
1141 | goto errout; | |
1142 | copied = ret; | |
1143 | } else | |
f19d5870 TM |
1144 | copied = block_write_end(file, mapping, pos, |
1145 | len, copied, page, fsdata); | |
f8514083 | 1146 | /* |
4631dbf6 | 1147 | * it's important to update i_size while still holding page lock: |
f8514083 AK |
1148 | * page writeout could otherwise come in and zero beyond i_size. |
1149 | */ | |
4631dbf6 | 1150 | i_size_changed = ext4_update_inode_size(inode, pos + copied); |
f8514083 AK |
1151 | unlock_page(page); |
1152 | page_cache_release(page); | |
1153 | ||
0572639f XW |
1154 | if (old_size < pos) |
1155 | pagecache_isize_extended(inode, old_size, pos); | |
f8514083 AK |
1156 | /* |
1157 | * Don't mark the inode dirty under page lock. First, it unnecessarily | |
1158 | * makes the holding time of page lock longer. Second, it forces lock | |
1159 | * ordering of page lock and transaction start for journaling | |
1160 | * filesystems. | |
1161 | */ | |
1162 | if (i_size_changed) | |
1163 | ext4_mark_inode_dirty(handle, inode); | |
1164 | ||
ffacfa7a | 1165 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
f8514083 AK |
1166 | /* if we have allocated more blocks and copied |
1167 | * less. We will have blocks allocated outside | |
1168 | * inode->i_size. So truncate them | |
1169 | */ | |
1170 | ext4_orphan_add(handle, inode); | |
74d553aa | 1171 | errout: |
617ba13b | 1172 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1173 | if (!ret) |
1174 | ret = ret2; | |
bfc1af65 | 1175 | |
f8514083 | 1176 | if (pos + len > inode->i_size) { |
b9a4207d | 1177 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1178 | /* |
ffacfa7a | 1179 | * If truncate failed early the inode might still be |
f8514083 AK |
1180 | * on the orphan list; we need to make sure the inode |
1181 | * is removed from the orphan list in that case. | |
1182 | */ | |
1183 | if (inode->i_nlink) | |
1184 | ext4_orphan_del(NULL, inode); | |
1185 | } | |
1186 | ||
bfc1af65 | 1187 | return ret ? ret : copied; |
ac27a0ec DK |
1188 | } |
1189 | ||
bfc1af65 | 1190 | static int ext4_journalled_write_end(struct file *file, |
de9a55b8 TT |
1191 | struct address_space *mapping, |
1192 | loff_t pos, unsigned len, unsigned copied, | |
1193 | struct page *page, void *fsdata) | |
ac27a0ec | 1194 | { |
617ba13b | 1195 | handle_t *handle = ext4_journal_current_handle(); |
bfc1af65 | 1196 | struct inode *inode = mapping->host; |
0572639f | 1197 | loff_t old_size = inode->i_size; |
ac27a0ec DK |
1198 | int ret = 0, ret2; |
1199 | int partial = 0; | |
bfc1af65 | 1200 | unsigned from, to; |
4631dbf6 | 1201 | int size_changed = 0; |
ac27a0ec | 1202 | |
9bffad1e | 1203 | trace_ext4_journalled_write_end(inode, pos, len, copied); |
bfc1af65 NP |
1204 | from = pos & (PAGE_CACHE_SIZE - 1); |
1205 | to = from + len; | |
1206 | ||
441c8508 CW |
1207 | BUG_ON(!ext4_handle_valid(handle)); |
1208 | ||
3fdcfb66 TM |
1209 | if (ext4_has_inline_data(inode)) |
1210 | copied = ext4_write_inline_data_end(inode, pos, len, | |
1211 | copied, page); | |
1212 | else { | |
1213 | if (copied < len) { | |
1214 | if (!PageUptodate(page)) | |
1215 | copied = 0; | |
1216 | page_zero_new_buffers(page, from+copied, to); | |
1217 | } | |
ac27a0ec | 1218 | |
3fdcfb66 TM |
1219 | ret = ext4_walk_page_buffers(handle, page_buffers(page), from, |
1220 | to, &partial, write_end_fn); | |
1221 | if (!partial) | |
1222 | SetPageUptodate(page); | |
1223 | } | |
4631dbf6 | 1224 | size_changed = ext4_update_inode_size(inode, pos + copied); |
19f5fb7a | 1225 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
2d859db3 | 1226 | EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; |
4631dbf6 DM |
1227 | unlock_page(page); |
1228 | page_cache_release(page); | |
1229 | ||
0572639f XW |
1230 | if (old_size < pos) |
1231 | pagecache_isize_extended(inode, old_size, pos); | |
1232 | ||
4631dbf6 | 1233 | if (size_changed) { |
617ba13b | 1234 | ret2 = ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
1235 | if (!ret) |
1236 | ret = ret2; | |
1237 | } | |
bfc1af65 | 1238 | |
ffacfa7a | 1239 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
f8514083 AK |
1240 | /* if we have allocated more blocks and copied |
1241 | * less. We will have blocks allocated outside | |
1242 | * inode->i_size. So truncate them | |
1243 | */ | |
1244 | ext4_orphan_add(handle, inode); | |
1245 | ||
617ba13b | 1246 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1247 | if (!ret) |
1248 | ret = ret2; | |
f8514083 | 1249 | if (pos + len > inode->i_size) { |
b9a4207d | 1250 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1251 | /* |
ffacfa7a | 1252 | * If truncate failed early the inode might still be |
f8514083 AK |
1253 | * on the orphan list; we need to make sure the inode |
1254 | * is removed from the orphan list in that case. | |
1255 | */ | |
1256 | if (inode->i_nlink) | |
1257 | ext4_orphan_del(NULL, inode); | |
1258 | } | |
bfc1af65 NP |
1259 | |
1260 | return ret ? ret : copied; | |
ac27a0ec | 1261 | } |
d2a17637 | 1262 | |
9d0be502 | 1263 | /* |
c27e43a1 | 1264 | * Reserve space for a single cluster |
9d0be502 | 1265 | */ |
c27e43a1 | 1266 | static int ext4_da_reserve_space(struct inode *inode) |
d2a17637 | 1267 | { |
60e58e0f | 1268 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
0637c6f4 | 1269 | struct ext4_inode_info *ei = EXT4_I(inode); |
5dd4056d | 1270 | int ret; |
03179fe9 TT |
1271 | |
1272 | /* | |
1273 | * We will charge metadata quota at writeout time; this saves | |
1274 | * us from metadata over-estimation, though we may go over by | |
1275 | * a small amount in the end. Here we just reserve for data. | |
1276 | */ | |
1277 | ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1)); | |
1278 | if (ret) | |
1279 | return ret; | |
d2a17637 | 1280 | |
0637c6f4 | 1281 | spin_lock(&ei->i_block_reservation_lock); |
71d4f7d0 | 1282 | if (ext4_claim_free_clusters(sbi, 1, 0)) { |
03179fe9 | 1283 | spin_unlock(&ei->i_block_reservation_lock); |
03179fe9 | 1284 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1)); |
d2a17637 MC |
1285 | return -ENOSPC; |
1286 | } | |
9d0be502 | 1287 | ei->i_reserved_data_blocks++; |
c27e43a1 | 1288 | trace_ext4_da_reserve_space(inode); |
0637c6f4 | 1289 | spin_unlock(&ei->i_block_reservation_lock); |
39bc680a | 1290 | |
d2a17637 MC |
1291 | return 0; /* success */ |
1292 | } | |
1293 | ||
12219aea | 1294 | static void ext4_da_release_space(struct inode *inode, int to_free) |
d2a17637 MC |
1295 | { |
1296 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0637c6f4 | 1297 | struct ext4_inode_info *ei = EXT4_I(inode); |
d2a17637 | 1298 | |
cd213226 MC |
1299 | if (!to_free) |
1300 | return; /* Nothing to release, exit */ | |
1301 | ||
d2a17637 | 1302 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); |
cd213226 | 1303 | |
5a58ec87 | 1304 | trace_ext4_da_release_space(inode, to_free); |
0637c6f4 | 1305 | if (unlikely(to_free > ei->i_reserved_data_blocks)) { |
cd213226 | 1306 | /* |
0637c6f4 TT |
1307 | * if there aren't enough reserved blocks, then the |
1308 | * counter is messed up somewhere. Since this | |
1309 | * function is called from invalidate page, it's | |
1310 | * harmless to return without any action. | |
cd213226 | 1311 | */ |
8de5c325 | 1312 | ext4_warning(inode->i_sb, "ext4_da_release_space: " |
0637c6f4 | 1313 | "ino %lu, to_free %d with only %d reserved " |
1084f252 | 1314 | "data blocks", inode->i_ino, to_free, |
0637c6f4 TT |
1315 | ei->i_reserved_data_blocks); |
1316 | WARN_ON(1); | |
1317 | to_free = ei->i_reserved_data_blocks; | |
cd213226 | 1318 | } |
0637c6f4 | 1319 | ei->i_reserved_data_blocks -= to_free; |
cd213226 | 1320 | |
72b8ab9d | 1321 | /* update fs dirty data blocks counter */ |
57042651 | 1322 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free); |
d2a17637 | 1323 | |
d2a17637 | 1324 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
60e58e0f | 1325 | |
7b415bf6 | 1326 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free)); |
d2a17637 MC |
1327 | } |
1328 | ||
1329 | static void ext4_da_page_release_reservation(struct page *page, | |
ca99fdd2 LC |
1330 | unsigned int offset, |
1331 | unsigned int length) | |
d2a17637 MC |
1332 | { |
1333 | int to_release = 0; | |
1334 | struct buffer_head *head, *bh; | |
1335 | unsigned int curr_off = 0; | |
7b415bf6 AK |
1336 | struct inode *inode = page->mapping->host; |
1337 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
ca99fdd2 | 1338 | unsigned int stop = offset + length; |
7b415bf6 | 1339 | int num_clusters; |
51865fda | 1340 | ext4_fsblk_t lblk; |
d2a17637 | 1341 | |
ca99fdd2 LC |
1342 | BUG_ON(stop > PAGE_CACHE_SIZE || stop < length); |
1343 | ||
d2a17637 MC |
1344 | head = page_buffers(page); |
1345 | bh = head; | |
1346 | do { | |
1347 | unsigned int next_off = curr_off + bh->b_size; | |
1348 | ||
ca99fdd2 LC |
1349 | if (next_off > stop) |
1350 | break; | |
1351 | ||
d2a17637 MC |
1352 | if ((offset <= curr_off) && (buffer_delay(bh))) { |
1353 | to_release++; | |
1354 | clear_buffer_delay(bh); | |
1355 | } | |
1356 | curr_off = next_off; | |
1357 | } while ((bh = bh->b_this_page) != head); | |
7b415bf6 | 1358 | |
51865fda ZL |
1359 | if (to_release) { |
1360 | lblk = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
1361 | ext4_es_remove_extent(inode, lblk, to_release); | |
1362 | } | |
1363 | ||
7b415bf6 AK |
1364 | /* If we have released all the blocks belonging to a cluster, then we |
1365 | * need to release the reserved space for that cluster. */ | |
1366 | num_clusters = EXT4_NUM_B2C(sbi, to_release); | |
1367 | while (num_clusters > 0) { | |
7b415bf6 AK |
1368 | lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) + |
1369 | ((num_clusters - 1) << sbi->s_cluster_bits); | |
1370 | if (sbi->s_cluster_ratio == 1 || | |
7d1b1fbc | 1371 | !ext4_find_delalloc_cluster(inode, lblk)) |
7b415bf6 AK |
1372 | ext4_da_release_space(inode, 1); |
1373 | ||
1374 | num_clusters--; | |
1375 | } | |
d2a17637 | 1376 | } |
ac27a0ec | 1377 | |
64769240 AT |
1378 | /* |
1379 | * Delayed allocation stuff | |
1380 | */ | |
1381 | ||
4e7ea81d JK |
1382 | struct mpage_da_data { |
1383 | struct inode *inode; | |
1384 | struct writeback_control *wbc; | |
6b523df4 | 1385 | |
4e7ea81d JK |
1386 | pgoff_t first_page; /* The first page to write */ |
1387 | pgoff_t next_page; /* Current page to examine */ | |
1388 | pgoff_t last_page; /* Last page to examine */ | |
791b7f08 | 1389 | /* |
4e7ea81d JK |
1390 | * Extent to map - this can be after first_page because that can be |
1391 | * fully mapped. We somewhat abuse m_flags to store whether the extent | |
1392 | * is delalloc or unwritten. | |
791b7f08 | 1393 | */ |
4e7ea81d JK |
1394 | struct ext4_map_blocks map; |
1395 | struct ext4_io_submit io_submit; /* IO submission data */ | |
1396 | }; | |
64769240 | 1397 | |
4e7ea81d JK |
1398 | static void mpage_release_unused_pages(struct mpage_da_data *mpd, |
1399 | bool invalidate) | |
c4a0c46e AK |
1400 | { |
1401 | int nr_pages, i; | |
1402 | pgoff_t index, end; | |
1403 | struct pagevec pvec; | |
1404 | struct inode *inode = mpd->inode; | |
1405 | struct address_space *mapping = inode->i_mapping; | |
4e7ea81d JK |
1406 | |
1407 | /* This is necessary when next_page == 0. */ | |
1408 | if (mpd->first_page >= mpd->next_page) | |
1409 | return; | |
c4a0c46e | 1410 | |
c7f5938a CW |
1411 | index = mpd->first_page; |
1412 | end = mpd->next_page - 1; | |
4e7ea81d JK |
1413 | if (invalidate) { |
1414 | ext4_lblk_t start, last; | |
1415 | start = index << (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
1416 | last = end << (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
1417 | ext4_es_remove_extent(inode, start, last - start + 1); | |
1418 | } | |
51865fda | 1419 | |
66bea92c | 1420 | pagevec_init(&pvec, 0); |
c4a0c46e AK |
1421 | while (index <= end) { |
1422 | nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE); | |
1423 | if (nr_pages == 0) | |
1424 | break; | |
1425 | for (i = 0; i < nr_pages; i++) { | |
1426 | struct page *page = pvec.pages[i]; | |
9b1d0998 | 1427 | if (page->index > end) |
c4a0c46e | 1428 | break; |
c4a0c46e AK |
1429 | BUG_ON(!PageLocked(page)); |
1430 | BUG_ON(PageWriteback(page)); | |
4e7ea81d JK |
1431 | if (invalidate) { |
1432 | block_invalidatepage(page, 0, PAGE_CACHE_SIZE); | |
1433 | ClearPageUptodate(page); | |
1434 | } | |
c4a0c46e AK |
1435 | unlock_page(page); |
1436 | } | |
9b1d0998 JK |
1437 | index = pvec.pages[nr_pages - 1]->index + 1; |
1438 | pagevec_release(&pvec); | |
c4a0c46e | 1439 | } |
c4a0c46e AK |
1440 | } |
1441 | ||
df22291f AK |
1442 | static void ext4_print_free_blocks(struct inode *inode) |
1443 | { | |
1444 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
92b97816 | 1445 | struct super_block *sb = inode->i_sb; |
f78ee70d | 1446 | struct ext4_inode_info *ei = EXT4_I(inode); |
92b97816 TT |
1447 | |
1448 | ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld", | |
5dee5437 | 1449 | EXT4_C2B(EXT4_SB(inode->i_sb), |
f78ee70d | 1450 | ext4_count_free_clusters(sb))); |
92b97816 TT |
1451 | ext4_msg(sb, KERN_CRIT, "Free/Dirty block details"); |
1452 | ext4_msg(sb, KERN_CRIT, "free_blocks=%lld", | |
f78ee70d | 1453 | (long long) EXT4_C2B(EXT4_SB(sb), |
57042651 | 1454 | percpu_counter_sum(&sbi->s_freeclusters_counter))); |
92b97816 | 1455 | ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld", |
f78ee70d | 1456 | (long long) EXT4_C2B(EXT4_SB(sb), |
7b415bf6 | 1457 | percpu_counter_sum(&sbi->s_dirtyclusters_counter))); |
92b97816 TT |
1458 | ext4_msg(sb, KERN_CRIT, "Block reservation details"); |
1459 | ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u", | |
f78ee70d | 1460 | ei->i_reserved_data_blocks); |
df22291f AK |
1461 | return; |
1462 | } | |
1463 | ||
c364b22c | 1464 | static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh) |
29fa89d0 | 1465 | { |
c364b22c | 1466 | return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh); |
29fa89d0 AK |
1467 | } |
1468 | ||
5356f261 AK |
1469 | /* |
1470 | * This function is grabs code from the very beginning of | |
1471 | * ext4_map_blocks, but assumes that the caller is from delayed write | |
1472 | * time. This function looks up the requested blocks and sets the | |
1473 | * buffer delay bit under the protection of i_data_sem. | |
1474 | */ | |
1475 | static int ext4_da_map_blocks(struct inode *inode, sector_t iblock, | |
1476 | struct ext4_map_blocks *map, | |
1477 | struct buffer_head *bh) | |
1478 | { | |
d100eef2 | 1479 | struct extent_status es; |
5356f261 AK |
1480 | int retval; |
1481 | sector_t invalid_block = ~((sector_t) 0xffff); | |
921f266b DM |
1482 | #ifdef ES_AGGRESSIVE_TEST |
1483 | struct ext4_map_blocks orig_map; | |
1484 | ||
1485 | memcpy(&orig_map, map, sizeof(*map)); | |
1486 | #endif | |
5356f261 AK |
1487 | |
1488 | if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es)) | |
1489 | invalid_block = ~0; | |
1490 | ||
1491 | map->m_flags = 0; | |
1492 | ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u," | |
1493 | "logical block %lu\n", inode->i_ino, map->m_len, | |
1494 | (unsigned long) map->m_lblk); | |
d100eef2 ZL |
1495 | |
1496 | /* Lookup extent status tree firstly */ | |
1497 | if (ext4_es_lookup_extent(inode, iblock, &es)) { | |
d100eef2 ZL |
1498 | if (ext4_es_is_hole(&es)) { |
1499 | retval = 0; | |
c8b459f4 | 1500 | down_read(&EXT4_I(inode)->i_data_sem); |
d100eef2 ZL |
1501 | goto add_delayed; |
1502 | } | |
1503 | ||
1504 | /* | |
1505 | * Delayed extent could be allocated by fallocate. | |
1506 | * So we need to check it. | |
1507 | */ | |
1508 | if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) { | |
1509 | map_bh(bh, inode->i_sb, invalid_block); | |
1510 | set_buffer_new(bh); | |
1511 | set_buffer_delay(bh); | |
1512 | return 0; | |
1513 | } | |
1514 | ||
1515 | map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk; | |
1516 | retval = es.es_len - (iblock - es.es_lblk); | |
1517 | if (retval > map->m_len) | |
1518 | retval = map->m_len; | |
1519 | map->m_len = retval; | |
1520 | if (ext4_es_is_written(&es)) | |
1521 | map->m_flags |= EXT4_MAP_MAPPED; | |
1522 | else if (ext4_es_is_unwritten(&es)) | |
1523 | map->m_flags |= EXT4_MAP_UNWRITTEN; | |
1524 | else | |
1525 | BUG_ON(1); | |
1526 | ||
921f266b DM |
1527 | #ifdef ES_AGGRESSIVE_TEST |
1528 | ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0); | |
1529 | #endif | |
d100eef2 ZL |
1530 | return retval; |
1531 | } | |
1532 | ||
5356f261 AK |
1533 | /* |
1534 | * Try to see if we can get the block without requesting a new | |
1535 | * file system block. | |
1536 | */ | |
c8b459f4 | 1537 | down_read(&EXT4_I(inode)->i_data_sem); |
cbd7584e | 1538 | if (ext4_has_inline_data(inode)) |
9c3569b5 | 1539 | retval = 0; |
cbd7584e | 1540 | else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
2f8e0a7c | 1541 | retval = ext4_ext_map_blocks(NULL, inode, map, 0); |
5356f261 | 1542 | else |
2f8e0a7c | 1543 | retval = ext4_ind_map_blocks(NULL, inode, map, 0); |
5356f261 | 1544 | |
d100eef2 | 1545 | add_delayed: |
5356f261 | 1546 | if (retval == 0) { |
f7fec032 | 1547 | int ret; |
5356f261 AK |
1548 | /* |
1549 | * XXX: __block_prepare_write() unmaps passed block, | |
1550 | * is it OK? | |
1551 | */ | |
386ad67c LC |
1552 | /* |
1553 | * If the block was allocated from previously allocated cluster, | |
1554 | * then we don't need to reserve it again. However we still need | |
1555 | * to reserve metadata for every block we're going to write. | |
1556 | */ | |
c27e43a1 | 1557 | if (EXT4_SB(inode->i_sb)->s_cluster_ratio == 1 || |
cbd7584e | 1558 | !ext4_find_delalloc_cluster(inode, map->m_lblk)) { |
c27e43a1 | 1559 | ret = ext4_da_reserve_space(inode); |
f7fec032 | 1560 | if (ret) { |
5356f261 | 1561 | /* not enough space to reserve */ |
f7fec032 | 1562 | retval = ret; |
5356f261 | 1563 | goto out_unlock; |
f7fec032 | 1564 | } |
5356f261 AK |
1565 | } |
1566 | ||
f7fec032 ZL |
1567 | ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, |
1568 | ~0, EXTENT_STATUS_DELAYED); | |
1569 | if (ret) { | |
1570 | retval = ret; | |
51865fda | 1571 | goto out_unlock; |
f7fec032 | 1572 | } |
51865fda | 1573 | |
5356f261 AK |
1574 | map_bh(bh, inode->i_sb, invalid_block); |
1575 | set_buffer_new(bh); | |
1576 | set_buffer_delay(bh); | |
f7fec032 ZL |
1577 | } else if (retval > 0) { |
1578 | int ret; | |
3be78c73 | 1579 | unsigned int status; |
f7fec032 | 1580 | |
44fb851d ZL |
1581 | if (unlikely(retval != map->m_len)) { |
1582 | ext4_warning(inode->i_sb, | |
1583 | "ES len assertion failed for inode " | |
1584 | "%lu: retval %d != map->m_len %d", | |
1585 | inode->i_ino, retval, map->m_len); | |
1586 | WARN_ON(1); | |
921f266b | 1587 | } |
921f266b | 1588 | |
f7fec032 ZL |
1589 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
1590 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; | |
1591 | ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, | |
1592 | map->m_pblk, status); | |
1593 | if (ret != 0) | |
1594 | retval = ret; | |
5356f261 AK |
1595 | } |
1596 | ||
1597 | out_unlock: | |
1598 | up_read((&EXT4_I(inode)->i_data_sem)); | |
1599 | ||
1600 | return retval; | |
1601 | } | |
1602 | ||
64769240 | 1603 | /* |
d91bd2c1 | 1604 | * This is a special get_block_t callback which is used by |
b920c755 TT |
1605 | * ext4_da_write_begin(). It will either return mapped block or |
1606 | * reserve space for a single block. | |
29fa89d0 AK |
1607 | * |
1608 | * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set. | |
1609 | * We also have b_blocknr = -1 and b_bdev initialized properly | |
1610 | * | |
1611 | * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set. | |
1612 | * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev | |
1613 | * initialized properly. | |
64769240 | 1614 | */ |
9c3569b5 TM |
1615 | int ext4_da_get_block_prep(struct inode *inode, sector_t iblock, |
1616 | struct buffer_head *bh, int create) | |
64769240 | 1617 | { |
2ed88685 | 1618 | struct ext4_map_blocks map; |
64769240 AT |
1619 | int ret = 0; |
1620 | ||
1621 | BUG_ON(create == 0); | |
2ed88685 TT |
1622 | BUG_ON(bh->b_size != inode->i_sb->s_blocksize); |
1623 | ||
1624 | map.m_lblk = iblock; | |
1625 | map.m_len = 1; | |
64769240 AT |
1626 | |
1627 | /* | |
1628 | * first, we need to know whether the block is allocated already | |
1629 | * preallocated blocks are unmapped but should treated | |
1630 | * the same as allocated blocks. | |
1631 | */ | |
5356f261 AK |
1632 | ret = ext4_da_map_blocks(inode, iblock, &map, bh); |
1633 | if (ret <= 0) | |
2ed88685 | 1634 | return ret; |
64769240 | 1635 | |
2ed88685 TT |
1636 | map_bh(bh, inode->i_sb, map.m_pblk); |
1637 | bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags; | |
1638 | ||
1639 | if (buffer_unwritten(bh)) { | |
1640 | /* A delayed write to unwritten bh should be marked | |
1641 | * new and mapped. Mapped ensures that we don't do | |
1642 | * get_block multiple times when we write to the same | |
1643 | * offset and new ensures that we do proper zero out | |
1644 | * for partial write. | |
1645 | */ | |
1646 | set_buffer_new(bh); | |
c8205636 | 1647 | set_buffer_mapped(bh); |
2ed88685 TT |
1648 | } |
1649 | return 0; | |
64769240 | 1650 | } |
61628a3f | 1651 | |
62e086be AK |
1652 | static int bget_one(handle_t *handle, struct buffer_head *bh) |
1653 | { | |
1654 | get_bh(bh); | |
1655 | return 0; | |
1656 | } | |
1657 | ||
1658 | static int bput_one(handle_t *handle, struct buffer_head *bh) | |
1659 | { | |
1660 | put_bh(bh); | |
1661 | return 0; | |
1662 | } | |
1663 | ||
1664 | static int __ext4_journalled_writepage(struct page *page, | |
62e086be AK |
1665 | unsigned int len) |
1666 | { | |
1667 | struct address_space *mapping = page->mapping; | |
1668 | struct inode *inode = mapping->host; | |
3fdcfb66 | 1669 | struct buffer_head *page_bufs = NULL; |
62e086be | 1670 | handle_t *handle = NULL; |
3fdcfb66 TM |
1671 | int ret = 0, err = 0; |
1672 | int inline_data = ext4_has_inline_data(inode); | |
1673 | struct buffer_head *inode_bh = NULL; | |
62e086be | 1674 | |
cb20d518 | 1675 | ClearPageChecked(page); |
3fdcfb66 TM |
1676 | |
1677 | if (inline_data) { | |
1678 | BUG_ON(page->index != 0); | |
1679 | BUG_ON(len > ext4_get_max_inline_size(inode)); | |
1680 | inode_bh = ext4_journalled_write_inline_data(inode, len, page); | |
1681 | if (inode_bh == NULL) | |
1682 | goto out; | |
1683 | } else { | |
1684 | page_bufs = page_buffers(page); | |
1685 | if (!page_bufs) { | |
1686 | BUG(); | |
1687 | goto out; | |
1688 | } | |
1689 | ext4_walk_page_buffers(handle, page_bufs, 0, len, | |
1690 | NULL, bget_one); | |
1691 | } | |
bdf96838 TT |
1692 | /* |
1693 | * We need to release the page lock before we start the | |
1694 | * journal, so grab a reference so the page won't disappear | |
1695 | * out from under us. | |
1696 | */ | |
1697 | get_page(page); | |
62e086be AK |
1698 | unlock_page(page); |
1699 | ||
9924a92a TT |
1700 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
1701 | ext4_writepage_trans_blocks(inode)); | |
62e086be AK |
1702 | if (IS_ERR(handle)) { |
1703 | ret = PTR_ERR(handle); | |
bdf96838 TT |
1704 | put_page(page); |
1705 | goto out_no_pagelock; | |
62e086be | 1706 | } |
441c8508 CW |
1707 | BUG_ON(!ext4_handle_valid(handle)); |
1708 | ||
bdf96838 TT |
1709 | lock_page(page); |
1710 | put_page(page); | |
1711 | if (page->mapping != mapping) { | |
1712 | /* The page got truncated from under us */ | |
1713 | ext4_journal_stop(handle); | |
1714 | ret = 0; | |
1715 | goto out; | |
1716 | } | |
1717 | ||
3fdcfb66 | 1718 | if (inline_data) { |
5d601255 | 1719 | BUFFER_TRACE(inode_bh, "get write access"); |
3fdcfb66 | 1720 | ret = ext4_journal_get_write_access(handle, inode_bh); |
62e086be | 1721 | |
3fdcfb66 TM |
1722 | err = ext4_handle_dirty_metadata(handle, inode, inode_bh); |
1723 | ||
1724 | } else { | |
1725 | ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL, | |
1726 | do_journal_get_write_access); | |
1727 | ||
1728 | err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL, | |
1729 | write_end_fn); | |
1730 | } | |
62e086be AK |
1731 | if (ret == 0) |
1732 | ret = err; | |
2d859db3 | 1733 | EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; |
62e086be AK |
1734 | err = ext4_journal_stop(handle); |
1735 | if (!ret) | |
1736 | ret = err; | |
1737 | ||
3fdcfb66 | 1738 | if (!ext4_has_inline_data(inode)) |
8c9367fd | 1739 | ext4_walk_page_buffers(NULL, page_bufs, 0, len, |
3fdcfb66 | 1740 | NULL, bput_one); |
19f5fb7a | 1741 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
62e086be | 1742 | out: |
bdf96838 TT |
1743 | unlock_page(page); |
1744 | out_no_pagelock: | |
3fdcfb66 | 1745 | brelse(inode_bh); |
62e086be AK |
1746 | return ret; |
1747 | } | |
1748 | ||
61628a3f | 1749 | /* |
43ce1d23 AK |
1750 | * Note that we don't need to start a transaction unless we're journaling data |
1751 | * because we should have holes filled from ext4_page_mkwrite(). We even don't | |
1752 | * need to file the inode to the transaction's list in ordered mode because if | |
1753 | * we are writing back data added by write(), the inode is already there and if | |
25985edc | 1754 | * we are writing back data modified via mmap(), no one guarantees in which |
43ce1d23 AK |
1755 | * transaction the data will hit the disk. In case we are journaling data, we |
1756 | * cannot start transaction directly because transaction start ranks above page | |
1757 | * lock so we have to do some magic. | |
1758 | * | |
b920c755 | 1759 | * This function can get called via... |
20970ba6 | 1760 | * - ext4_writepages after taking page lock (have journal handle) |
b920c755 | 1761 | * - journal_submit_inode_data_buffers (no journal handle) |
f6463b0d | 1762 | * - shrink_page_list via the kswapd/direct reclaim (no journal handle) |
b920c755 | 1763 | * - grab_page_cache when doing write_begin (have journal handle) |
43ce1d23 AK |
1764 | * |
1765 | * We don't do any block allocation in this function. If we have page with | |
1766 | * multiple blocks we need to write those buffer_heads that are mapped. This | |
1767 | * is important for mmaped based write. So if we do with blocksize 1K | |
1768 | * truncate(f, 1024); | |
1769 | * a = mmap(f, 0, 4096); | |
1770 | * a[0] = 'a'; | |
1771 | * truncate(f, 4096); | |
1772 | * we have in the page first buffer_head mapped via page_mkwrite call back | |
90802ed9 | 1773 | * but other buffer_heads would be unmapped but dirty (dirty done via the |
43ce1d23 AK |
1774 | * do_wp_page). So writepage should write the first block. If we modify |
1775 | * the mmap area beyond 1024 we will again get a page_fault and the | |
1776 | * page_mkwrite callback will do the block allocation and mark the | |
1777 | * buffer_heads mapped. | |
1778 | * | |
1779 | * We redirty the page if we have any buffer_heads that is either delay or | |
1780 | * unwritten in the page. | |
1781 | * | |
1782 | * We can get recursively called as show below. | |
1783 | * | |
1784 | * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() -> | |
1785 | * ext4_writepage() | |
1786 | * | |
1787 | * But since we don't do any block allocation we should not deadlock. | |
1788 | * Page also have the dirty flag cleared so we don't get recurive page_lock. | |
61628a3f | 1789 | */ |
43ce1d23 | 1790 | static int ext4_writepage(struct page *page, |
62e086be | 1791 | struct writeback_control *wbc) |
64769240 | 1792 | { |
f8bec370 | 1793 | int ret = 0; |
61628a3f | 1794 | loff_t size; |
498e5f24 | 1795 | unsigned int len; |
744692dc | 1796 | struct buffer_head *page_bufs = NULL; |
61628a3f | 1797 | struct inode *inode = page->mapping->host; |
36ade451 | 1798 | struct ext4_io_submit io_submit; |
1c8349a1 | 1799 | bool keep_towrite = false; |
61628a3f | 1800 | |
a9c667f8 | 1801 | trace_ext4_writepage(page); |
f0e6c985 AK |
1802 | size = i_size_read(inode); |
1803 | if (page->index == size >> PAGE_CACHE_SHIFT) | |
1804 | len = size & ~PAGE_CACHE_MASK; | |
1805 | else | |
1806 | len = PAGE_CACHE_SIZE; | |
64769240 | 1807 | |
a42afc5f | 1808 | page_bufs = page_buffers(page); |
a42afc5f | 1809 | /* |
fe386132 JK |
1810 | * We cannot do block allocation or other extent handling in this |
1811 | * function. If there are buffers needing that, we have to redirty | |
1812 | * the page. But we may reach here when we do a journal commit via | |
1813 | * journal_submit_inode_data_buffers() and in that case we must write | |
1814 | * allocated buffers to achieve data=ordered mode guarantees. | |
a42afc5f | 1815 | */ |
f19d5870 TM |
1816 | if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL, |
1817 | ext4_bh_delay_or_unwritten)) { | |
f8bec370 | 1818 | redirty_page_for_writepage(wbc, page); |
fe386132 JK |
1819 | if (current->flags & PF_MEMALLOC) { |
1820 | /* | |
1821 | * For memory cleaning there's no point in writing only | |
1822 | * some buffers. So just bail out. Warn if we came here | |
1823 | * from direct reclaim. | |
1824 | */ | |
1825 | WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) | |
1826 | == PF_MEMALLOC); | |
f0e6c985 AK |
1827 | unlock_page(page); |
1828 | return 0; | |
1829 | } | |
1c8349a1 | 1830 | keep_towrite = true; |
a42afc5f | 1831 | } |
64769240 | 1832 | |
cb20d518 | 1833 | if (PageChecked(page) && ext4_should_journal_data(inode)) |
43ce1d23 AK |
1834 | /* |
1835 | * It's mmapped pagecache. Add buffers and journal it. There | |
1836 | * doesn't seem much point in redirtying the page here. | |
1837 | */ | |
3f0ca309 | 1838 | return __ext4_journalled_writepage(page, len); |
43ce1d23 | 1839 | |
97a851ed JK |
1840 | ext4_io_submit_init(&io_submit, wbc); |
1841 | io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS); | |
1842 | if (!io_submit.io_end) { | |
1843 | redirty_page_for_writepage(wbc, page); | |
1844 | unlock_page(page); | |
1845 | return -ENOMEM; | |
1846 | } | |
1c8349a1 | 1847 | ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite); |
36ade451 | 1848 | ext4_io_submit(&io_submit); |
97a851ed JK |
1849 | /* Drop io_end reference we got from init */ |
1850 | ext4_put_io_end_defer(io_submit.io_end); | |
64769240 AT |
1851 | return ret; |
1852 | } | |
1853 | ||
5f1132b2 JK |
1854 | static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page) |
1855 | { | |
1856 | int len; | |
1857 | loff_t size = i_size_read(mpd->inode); | |
1858 | int err; | |
1859 | ||
1860 | BUG_ON(page->index != mpd->first_page); | |
1861 | if (page->index == size >> PAGE_CACHE_SHIFT) | |
1862 | len = size & ~PAGE_CACHE_MASK; | |
1863 | else | |
1864 | len = PAGE_CACHE_SIZE; | |
1865 | clear_page_dirty_for_io(page); | |
1c8349a1 | 1866 | err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false); |
5f1132b2 JK |
1867 | if (!err) |
1868 | mpd->wbc->nr_to_write--; | |
1869 | mpd->first_page++; | |
1870 | ||
1871 | return err; | |
1872 | } | |
1873 | ||
4e7ea81d JK |
1874 | #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay)) |
1875 | ||
61628a3f | 1876 | /* |
fffb2739 JK |
1877 | * mballoc gives us at most this number of blocks... |
1878 | * XXX: That seems to be only a limitation of ext4_mb_normalize_request(). | |
70261f56 | 1879 | * The rest of mballoc seems to handle chunks up to full group size. |
61628a3f | 1880 | */ |
fffb2739 | 1881 | #define MAX_WRITEPAGES_EXTENT_LEN 2048 |
525f4ed8 | 1882 | |
4e7ea81d JK |
1883 | /* |
1884 | * mpage_add_bh_to_extent - try to add bh to extent of blocks to map | |
1885 | * | |
1886 | * @mpd - extent of blocks | |
1887 | * @lblk - logical number of the block in the file | |
09930042 | 1888 | * @bh - buffer head we want to add to the extent |
4e7ea81d | 1889 | * |
09930042 JK |
1890 | * The function is used to collect contig. blocks in the same state. If the |
1891 | * buffer doesn't require mapping for writeback and we haven't started the | |
1892 | * extent of buffers to map yet, the function returns 'true' immediately - the | |
1893 | * caller can write the buffer right away. Otherwise the function returns true | |
1894 | * if the block has been added to the extent, false if the block couldn't be | |
1895 | * added. | |
4e7ea81d | 1896 | */ |
09930042 JK |
1897 | static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk, |
1898 | struct buffer_head *bh) | |
4e7ea81d JK |
1899 | { |
1900 | struct ext4_map_blocks *map = &mpd->map; | |
1901 | ||
09930042 JK |
1902 | /* Buffer that doesn't need mapping for writeback? */ |
1903 | if (!buffer_dirty(bh) || !buffer_mapped(bh) || | |
1904 | (!buffer_delay(bh) && !buffer_unwritten(bh))) { | |
1905 | /* So far no extent to map => we write the buffer right away */ | |
1906 | if (map->m_len == 0) | |
1907 | return true; | |
1908 | return false; | |
1909 | } | |
4e7ea81d JK |
1910 | |
1911 | /* First block in the extent? */ | |
1912 | if (map->m_len == 0) { | |
1913 | map->m_lblk = lblk; | |
1914 | map->m_len = 1; | |
09930042 JK |
1915 | map->m_flags = bh->b_state & BH_FLAGS; |
1916 | return true; | |
4e7ea81d JK |
1917 | } |
1918 | ||
09930042 JK |
1919 | /* Don't go larger than mballoc is willing to allocate */ |
1920 | if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN) | |
1921 | return false; | |
1922 | ||
4e7ea81d JK |
1923 | /* Can we merge the block to our big extent? */ |
1924 | if (lblk == map->m_lblk + map->m_len && | |
09930042 | 1925 | (bh->b_state & BH_FLAGS) == map->m_flags) { |
4e7ea81d | 1926 | map->m_len++; |
09930042 | 1927 | return true; |
4e7ea81d | 1928 | } |
09930042 | 1929 | return false; |
4e7ea81d JK |
1930 | } |
1931 | ||
5f1132b2 JK |
1932 | /* |
1933 | * mpage_process_page_bufs - submit page buffers for IO or add them to extent | |
1934 | * | |
1935 | * @mpd - extent of blocks for mapping | |
1936 | * @head - the first buffer in the page | |
1937 | * @bh - buffer we should start processing from | |
1938 | * @lblk - logical number of the block in the file corresponding to @bh | |
1939 | * | |
1940 | * Walk through page buffers from @bh upto @head (exclusive) and either submit | |
1941 | * the page for IO if all buffers in this page were mapped and there's no | |
1942 | * accumulated extent of buffers to map or add buffers in the page to the | |
1943 | * extent of buffers to map. The function returns 1 if the caller can continue | |
1944 | * by processing the next page, 0 if it should stop adding buffers to the | |
1945 | * extent to map because we cannot extend it anymore. It can also return value | |
1946 | * < 0 in case of error during IO submission. | |
1947 | */ | |
1948 | static int mpage_process_page_bufs(struct mpage_da_data *mpd, | |
1949 | struct buffer_head *head, | |
1950 | struct buffer_head *bh, | |
1951 | ext4_lblk_t lblk) | |
4e7ea81d JK |
1952 | { |
1953 | struct inode *inode = mpd->inode; | |
5f1132b2 | 1954 | int err; |
4e7ea81d JK |
1955 | ext4_lblk_t blocks = (i_size_read(inode) + (1 << inode->i_blkbits) - 1) |
1956 | >> inode->i_blkbits; | |
1957 | ||
1958 | do { | |
1959 | BUG_ON(buffer_locked(bh)); | |
1960 | ||
09930042 | 1961 | if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) { |
4e7ea81d JK |
1962 | /* Found extent to map? */ |
1963 | if (mpd->map.m_len) | |
5f1132b2 | 1964 | return 0; |
09930042 | 1965 | /* Everything mapped so far and we hit EOF */ |
5f1132b2 | 1966 | break; |
4e7ea81d | 1967 | } |
4e7ea81d | 1968 | } while (lblk++, (bh = bh->b_this_page) != head); |
5f1132b2 JK |
1969 | /* So far everything mapped? Submit the page for IO. */ |
1970 | if (mpd->map.m_len == 0) { | |
1971 | err = mpage_submit_page(mpd, head->b_page); | |
1972 | if (err < 0) | |
1973 | return err; | |
1974 | } | |
1975 | return lblk < blocks; | |
4e7ea81d JK |
1976 | } |
1977 | ||
1978 | /* | |
1979 | * mpage_map_buffers - update buffers corresponding to changed extent and | |
1980 | * submit fully mapped pages for IO | |
1981 | * | |
1982 | * @mpd - description of extent to map, on return next extent to map | |
1983 | * | |
1984 | * Scan buffers corresponding to changed extent (we expect corresponding pages | |
1985 | * to be already locked) and update buffer state according to new extent state. | |
1986 | * We map delalloc buffers to their physical location, clear unwritten bits, | |
556615dc | 1987 | * and mark buffers as uninit when we perform writes to unwritten extents |
4e7ea81d JK |
1988 | * and do extent conversion after IO is finished. If the last page is not fully |
1989 | * mapped, we update @map to the next extent in the last page that needs | |
1990 | * mapping. Otherwise we submit the page for IO. | |
1991 | */ | |
1992 | static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd) | |
1993 | { | |
1994 | struct pagevec pvec; | |
1995 | int nr_pages, i; | |
1996 | struct inode *inode = mpd->inode; | |
1997 | struct buffer_head *head, *bh; | |
1998 | int bpp_bits = PAGE_CACHE_SHIFT - inode->i_blkbits; | |
4e7ea81d JK |
1999 | pgoff_t start, end; |
2000 | ext4_lblk_t lblk; | |
2001 | sector_t pblock; | |
2002 | int err; | |
2003 | ||
2004 | start = mpd->map.m_lblk >> bpp_bits; | |
2005 | end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits; | |
2006 | lblk = start << bpp_bits; | |
2007 | pblock = mpd->map.m_pblk; | |
2008 | ||
2009 | pagevec_init(&pvec, 0); | |
2010 | while (start <= end) { | |
2011 | nr_pages = pagevec_lookup(&pvec, inode->i_mapping, start, | |
2012 | PAGEVEC_SIZE); | |
2013 | if (nr_pages == 0) | |
2014 | break; | |
2015 | for (i = 0; i < nr_pages; i++) { | |
2016 | struct page *page = pvec.pages[i]; | |
2017 | ||
2018 | if (page->index > end) | |
2019 | break; | |
70261f56 | 2020 | /* Up to 'end' pages must be contiguous */ |
4e7ea81d JK |
2021 | BUG_ON(page->index != start); |
2022 | bh = head = page_buffers(page); | |
2023 | do { | |
2024 | if (lblk < mpd->map.m_lblk) | |
2025 | continue; | |
2026 | if (lblk >= mpd->map.m_lblk + mpd->map.m_len) { | |
2027 | /* | |
2028 | * Buffer after end of mapped extent. | |
2029 | * Find next buffer in the page to map. | |
2030 | */ | |
2031 | mpd->map.m_len = 0; | |
2032 | mpd->map.m_flags = 0; | |
5f1132b2 JK |
2033 | /* |
2034 | * FIXME: If dioread_nolock supports | |
2035 | * blocksize < pagesize, we need to make | |
2036 | * sure we add size mapped so far to | |
2037 | * io_end->size as the following call | |
2038 | * can submit the page for IO. | |
2039 | */ | |
2040 | err = mpage_process_page_bufs(mpd, head, | |
2041 | bh, lblk); | |
4e7ea81d | 2042 | pagevec_release(&pvec); |
5f1132b2 JK |
2043 | if (err > 0) |
2044 | err = 0; | |
2045 | return err; | |
4e7ea81d JK |
2046 | } |
2047 | if (buffer_delay(bh)) { | |
2048 | clear_buffer_delay(bh); | |
2049 | bh->b_blocknr = pblock++; | |
2050 | } | |
4e7ea81d | 2051 | clear_buffer_unwritten(bh); |
5f1132b2 | 2052 | } while (lblk++, (bh = bh->b_this_page) != head); |
4e7ea81d JK |
2053 | |
2054 | /* | |
2055 | * FIXME: This is going to break if dioread_nolock | |
2056 | * supports blocksize < pagesize as we will try to | |
2057 | * convert potentially unmapped parts of inode. | |
2058 | */ | |
2059 | mpd->io_submit.io_end->size += PAGE_CACHE_SIZE; | |
2060 | /* Page fully mapped - let IO run! */ | |
2061 | err = mpage_submit_page(mpd, page); | |
2062 | if (err < 0) { | |
2063 | pagevec_release(&pvec); | |
2064 | return err; | |
2065 | } | |
2066 | start++; | |
2067 | } | |
2068 | pagevec_release(&pvec); | |
2069 | } | |
2070 | /* Extent fully mapped and matches with page boundary. We are done. */ | |
2071 | mpd->map.m_len = 0; | |
2072 | mpd->map.m_flags = 0; | |
2073 | return 0; | |
2074 | } | |
2075 | ||
2076 | static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd) | |
2077 | { | |
2078 | struct inode *inode = mpd->inode; | |
2079 | struct ext4_map_blocks *map = &mpd->map; | |
2080 | int get_blocks_flags; | |
090f32ee | 2081 | int err, dioread_nolock; |
4e7ea81d JK |
2082 | |
2083 | trace_ext4_da_write_pages_extent(inode, map); | |
2084 | /* | |
2085 | * Call ext4_map_blocks() to allocate any delayed allocation blocks, or | |
556615dc | 2086 | * to convert an unwritten extent to be initialized (in the case |
4e7ea81d JK |
2087 | * where we have written into one or more preallocated blocks). It is |
2088 | * possible that we're going to need more metadata blocks than | |
2089 | * previously reserved. However we must not fail because we're in | |
2090 | * writeback and there is nothing we can do about it so it might result | |
2091 | * in data loss. So use reserved blocks to allocate metadata if | |
2092 | * possible. | |
2093 | * | |
754cfed6 TT |
2094 | * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if |
2095 | * the blocks in question are delalloc blocks. This indicates | |
2096 | * that the blocks and quotas has already been checked when | |
2097 | * the data was copied into the page cache. | |
4e7ea81d JK |
2098 | */ |
2099 | get_blocks_flags = EXT4_GET_BLOCKS_CREATE | | |
2100 | EXT4_GET_BLOCKS_METADATA_NOFAIL; | |
090f32ee LC |
2101 | dioread_nolock = ext4_should_dioread_nolock(inode); |
2102 | if (dioread_nolock) | |
4e7ea81d JK |
2103 | get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT; |
2104 | if (map->m_flags & (1 << BH_Delay)) | |
2105 | get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE; | |
2106 | ||
2107 | err = ext4_map_blocks(handle, inode, map, get_blocks_flags); | |
2108 | if (err < 0) | |
2109 | return err; | |
090f32ee | 2110 | if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) { |
6b523df4 JK |
2111 | if (!mpd->io_submit.io_end->handle && |
2112 | ext4_handle_valid(handle)) { | |
2113 | mpd->io_submit.io_end->handle = handle->h_rsv_handle; | |
2114 | handle->h_rsv_handle = NULL; | |
2115 | } | |
3613d228 | 2116 | ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end); |
6b523df4 | 2117 | } |
4e7ea81d JK |
2118 | |
2119 | BUG_ON(map->m_len == 0); | |
2120 | if (map->m_flags & EXT4_MAP_NEW) { | |
2121 | struct block_device *bdev = inode->i_sb->s_bdev; | |
2122 | int i; | |
2123 | ||
2124 | for (i = 0; i < map->m_len; i++) | |
2125 | unmap_underlying_metadata(bdev, map->m_pblk + i); | |
2126 | } | |
2127 | return 0; | |
2128 | } | |
2129 | ||
2130 | /* | |
2131 | * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length | |
2132 | * mpd->len and submit pages underlying it for IO | |
2133 | * | |
2134 | * @handle - handle for journal operations | |
2135 | * @mpd - extent to map | |
7534e854 JK |
2136 | * @give_up_on_write - we set this to true iff there is a fatal error and there |
2137 | * is no hope of writing the data. The caller should discard | |
2138 | * dirty pages to avoid infinite loops. | |
4e7ea81d JK |
2139 | * |
2140 | * The function maps extent starting at mpd->lblk of length mpd->len. If it is | |
2141 | * delayed, blocks are allocated, if it is unwritten, we may need to convert | |
2142 | * them to initialized or split the described range from larger unwritten | |
2143 | * extent. Note that we need not map all the described range since allocation | |
2144 | * can return less blocks or the range is covered by more unwritten extents. We | |
2145 | * cannot map more because we are limited by reserved transaction credits. On | |
2146 | * the other hand we always make sure that the last touched page is fully | |
2147 | * mapped so that it can be written out (and thus forward progress is | |
2148 | * guaranteed). After mapping we submit all mapped pages for IO. | |
2149 | */ | |
2150 | static int mpage_map_and_submit_extent(handle_t *handle, | |
cb530541 TT |
2151 | struct mpage_da_data *mpd, |
2152 | bool *give_up_on_write) | |
4e7ea81d JK |
2153 | { |
2154 | struct inode *inode = mpd->inode; | |
2155 | struct ext4_map_blocks *map = &mpd->map; | |
2156 | int err; | |
2157 | loff_t disksize; | |
6603120e | 2158 | int progress = 0; |
4e7ea81d JK |
2159 | |
2160 | mpd->io_submit.io_end->offset = | |
2161 | ((loff_t)map->m_lblk) << inode->i_blkbits; | |
27d7c4ed | 2162 | do { |
4e7ea81d JK |
2163 | err = mpage_map_one_extent(handle, mpd); |
2164 | if (err < 0) { | |
2165 | struct super_block *sb = inode->i_sb; | |
2166 | ||
cb530541 TT |
2167 | if (EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED) |
2168 | goto invalidate_dirty_pages; | |
4e7ea81d | 2169 | /* |
cb530541 TT |
2170 | * Let the uper layers retry transient errors. |
2171 | * In the case of ENOSPC, if ext4_count_free_blocks() | |
2172 | * is non-zero, a commit should free up blocks. | |
4e7ea81d | 2173 | */ |
cb530541 | 2174 | if ((err == -ENOMEM) || |
6603120e DM |
2175 | (err == -ENOSPC && ext4_count_free_clusters(sb))) { |
2176 | if (progress) | |
2177 | goto update_disksize; | |
cb530541 | 2178 | return err; |
6603120e | 2179 | } |
cb530541 TT |
2180 | ext4_msg(sb, KERN_CRIT, |
2181 | "Delayed block allocation failed for " | |
2182 | "inode %lu at logical offset %llu with" | |
2183 | " max blocks %u with error %d", | |
2184 | inode->i_ino, | |
2185 | (unsigned long long)map->m_lblk, | |
2186 | (unsigned)map->m_len, -err); | |
2187 | ext4_msg(sb, KERN_CRIT, | |
2188 | "This should not happen!! Data will " | |
2189 | "be lost\n"); | |
2190 | if (err == -ENOSPC) | |
2191 | ext4_print_free_blocks(inode); | |
2192 | invalidate_dirty_pages: | |
2193 | *give_up_on_write = true; | |
4e7ea81d JK |
2194 | return err; |
2195 | } | |
6603120e | 2196 | progress = 1; |
4e7ea81d JK |
2197 | /* |
2198 | * Update buffer state, submit mapped pages, and get us new | |
2199 | * extent to map | |
2200 | */ | |
2201 | err = mpage_map_and_submit_buffers(mpd); | |
2202 | if (err < 0) | |
6603120e | 2203 | goto update_disksize; |
27d7c4ed | 2204 | } while (map->m_len); |
4e7ea81d | 2205 | |
6603120e | 2206 | update_disksize: |
622cad13 TT |
2207 | /* |
2208 | * Update on-disk size after IO is submitted. Races with | |
2209 | * truncate are avoided by checking i_size under i_data_sem. | |
2210 | */ | |
4e7ea81d | 2211 | disksize = ((loff_t)mpd->first_page) << PAGE_CACHE_SHIFT; |
4e7ea81d JK |
2212 | if (disksize > EXT4_I(inode)->i_disksize) { |
2213 | int err2; | |
622cad13 TT |
2214 | loff_t i_size; |
2215 | ||
2216 | down_write(&EXT4_I(inode)->i_data_sem); | |
2217 | i_size = i_size_read(inode); | |
2218 | if (disksize > i_size) | |
2219 | disksize = i_size; | |
2220 | if (disksize > EXT4_I(inode)->i_disksize) | |
2221 | EXT4_I(inode)->i_disksize = disksize; | |
4e7ea81d | 2222 | err2 = ext4_mark_inode_dirty(handle, inode); |
622cad13 | 2223 | up_write(&EXT4_I(inode)->i_data_sem); |
4e7ea81d JK |
2224 | if (err2) |
2225 | ext4_error(inode->i_sb, | |
2226 | "Failed to mark inode %lu dirty", | |
2227 | inode->i_ino); | |
2228 | if (!err) | |
2229 | err = err2; | |
2230 | } | |
2231 | return err; | |
2232 | } | |
2233 | ||
fffb2739 JK |
2234 | /* |
2235 | * Calculate the total number of credits to reserve for one writepages | |
20970ba6 | 2236 | * iteration. This is called from ext4_writepages(). We map an extent of |
70261f56 | 2237 | * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping |
fffb2739 JK |
2238 | * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN + |
2239 | * bpp - 1 blocks in bpp different extents. | |
2240 | */ | |
525f4ed8 MC |
2241 | static int ext4_da_writepages_trans_blocks(struct inode *inode) |
2242 | { | |
fffb2739 | 2243 | int bpp = ext4_journal_blocks_per_page(inode); |
525f4ed8 | 2244 | |
fffb2739 JK |
2245 | return ext4_meta_trans_blocks(inode, |
2246 | MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp); | |
525f4ed8 | 2247 | } |
61628a3f | 2248 | |
8e48dcfb | 2249 | /* |
4e7ea81d JK |
2250 | * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages |
2251 | * and underlying extent to map | |
2252 | * | |
2253 | * @mpd - where to look for pages | |
2254 | * | |
2255 | * Walk dirty pages in the mapping. If they are fully mapped, submit them for | |
2256 | * IO immediately. When we find a page which isn't mapped we start accumulating | |
2257 | * extent of buffers underlying these pages that needs mapping (formed by | |
2258 | * either delayed or unwritten buffers). We also lock the pages containing | |
2259 | * these buffers. The extent found is returned in @mpd structure (starting at | |
2260 | * mpd->lblk with length mpd->len blocks). | |
2261 | * | |
2262 | * Note that this function can attach bios to one io_end structure which are | |
2263 | * neither logically nor physically contiguous. Although it may seem as an | |
2264 | * unnecessary complication, it is actually inevitable in blocksize < pagesize | |
2265 | * case as we need to track IO to all buffers underlying a page in one io_end. | |
8e48dcfb | 2266 | */ |
4e7ea81d | 2267 | static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd) |
8e48dcfb | 2268 | { |
4e7ea81d JK |
2269 | struct address_space *mapping = mpd->inode->i_mapping; |
2270 | struct pagevec pvec; | |
2271 | unsigned int nr_pages; | |
aeac589a | 2272 | long left = mpd->wbc->nr_to_write; |
4e7ea81d JK |
2273 | pgoff_t index = mpd->first_page; |
2274 | pgoff_t end = mpd->last_page; | |
2275 | int tag; | |
2276 | int i, err = 0; | |
2277 | int blkbits = mpd->inode->i_blkbits; | |
2278 | ext4_lblk_t lblk; | |
2279 | struct buffer_head *head; | |
8e48dcfb | 2280 | |
4e7ea81d | 2281 | if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages) |
5b41d924 ES |
2282 | tag = PAGECACHE_TAG_TOWRITE; |
2283 | else | |
2284 | tag = PAGECACHE_TAG_DIRTY; | |
2285 | ||
4e7ea81d JK |
2286 | pagevec_init(&pvec, 0); |
2287 | mpd->map.m_len = 0; | |
2288 | mpd->next_page = index; | |
4f01b02c | 2289 | while (index <= end) { |
5b41d924 | 2290 | nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag, |
8e48dcfb TT |
2291 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1); |
2292 | if (nr_pages == 0) | |
4e7ea81d | 2293 | goto out; |
8e48dcfb TT |
2294 | |
2295 | for (i = 0; i < nr_pages; i++) { | |
2296 | struct page *page = pvec.pages[i]; | |
2297 | ||
2298 | /* | |
2299 | * At this point, the page may be truncated or | |
2300 | * invalidated (changing page->mapping to NULL), or | |
2301 | * even swizzled back from swapper_space to tmpfs file | |
2302 | * mapping. However, page->index will not change | |
2303 | * because we have a reference on the page. | |
2304 | */ | |
4f01b02c TT |
2305 | if (page->index > end) |
2306 | goto out; | |
8e48dcfb | 2307 | |
aeac589a ML |
2308 | /* |
2309 | * Accumulated enough dirty pages? This doesn't apply | |
2310 | * to WB_SYNC_ALL mode. For integrity sync we have to | |
2311 | * keep going because someone may be concurrently | |
2312 | * dirtying pages, and we might have synced a lot of | |
2313 | * newly appeared dirty pages, but have not synced all | |
2314 | * of the old dirty pages. | |
2315 | */ | |
2316 | if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0) | |
2317 | goto out; | |
2318 | ||
4e7ea81d JK |
2319 | /* If we can't merge this page, we are done. */ |
2320 | if (mpd->map.m_len > 0 && mpd->next_page != page->index) | |
2321 | goto out; | |
78aaced3 | 2322 | |
8e48dcfb | 2323 | lock_page(page); |
8e48dcfb | 2324 | /* |
4e7ea81d JK |
2325 | * If the page is no longer dirty, or its mapping no |
2326 | * longer corresponds to inode we are writing (which | |
2327 | * means it has been truncated or invalidated), or the | |
2328 | * page is already under writeback and we are not doing | |
2329 | * a data integrity writeback, skip the page | |
8e48dcfb | 2330 | */ |
4f01b02c TT |
2331 | if (!PageDirty(page) || |
2332 | (PageWriteback(page) && | |
4e7ea81d | 2333 | (mpd->wbc->sync_mode == WB_SYNC_NONE)) || |
4f01b02c | 2334 | unlikely(page->mapping != mapping)) { |
8e48dcfb TT |
2335 | unlock_page(page); |
2336 | continue; | |
2337 | } | |
2338 | ||
7cb1a535 | 2339 | wait_on_page_writeback(page); |
8e48dcfb | 2340 | BUG_ON(PageWriteback(page)); |
8e48dcfb | 2341 | |
4e7ea81d | 2342 | if (mpd->map.m_len == 0) |
8eb9e5ce | 2343 | mpd->first_page = page->index; |
8eb9e5ce | 2344 | mpd->next_page = page->index + 1; |
f8bec370 | 2345 | /* Add all dirty buffers to mpd */ |
4e7ea81d JK |
2346 | lblk = ((ext4_lblk_t)page->index) << |
2347 | (PAGE_CACHE_SHIFT - blkbits); | |
f8bec370 | 2348 | head = page_buffers(page); |
5f1132b2 JK |
2349 | err = mpage_process_page_bufs(mpd, head, head, lblk); |
2350 | if (err <= 0) | |
4e7ea81d | 2351 | goto out; |
5f1132b2 | 2352 | err = 0; |
aeac589a | 2353 | left--; |
8e48dcfb TT |
2354 | } |
2355 | pagevec_release(&pvec); | |
2356 | cond_resched(); | |
2357 | } | |
4f01b02c | 2358 | return 0; |
8eb9e5ce TT |
2359 | out: |
2360 | pagevec_release(&pvec); | |
4e7ea81d | 2361 | return err; |
8e48dcfb TT |
2362 | } |
2363 | ||
20970ba6 TT |
2364 | static int __writepage(struct page *page, struct writeback_control *wbc, |
2365 | void *data) | |
2366 | { | |
2367 | struct address_space *mapping = data; | |
2368 | int ret = ext4_writepage(page, wbc); | |
2369 | mapping_set_error(mapping, ret); | |
2370 | return ret; | |
2371 | } | |
2372 | ||
2373 | static int ext4_writepages(struct address_space *mapping, | |
2374 | struct writeback_control *wbc) | |
64769240 | 2375 | { |
4e7ea81d JK |
2376 | pgoff_t writeback_index = 0; |
2377 | long nr_to_write = wbc->nr_to_write; | |
22208ded | 2378 | int range_whole = 0; |
4e7ea81d | 2379 | int cycled = 1; |
61628a3f | 2380 | handle_t *handle = NULL; |
df22291f | 2381 | struct mpage_da_data mpd; |
5e745b04 | 2382 | struct inode *inode = mapping->host; |
6b523df4 | 2383 | int needed_blocks, rsv_blocks = 0, ret = 0; |
5e745b04 | 2384 | struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb); |
4e7ea81d | 2385 | bool done; |
1bce63d1 | 2386 | struct blk_plug plug; |
cb530541 | 2387 | bool give_up_on_write = false; |
61628a3f | 2388 | |
20970ba6 | 2389 | trace_ext4_writepages(inode, wbc); |
ba80b101 | 2390 | |
61628a3f MC |
2391 | /* |
2392 | * No pages to write? This is mainly a kludge to avoid starting | |
2393 | * a transaction for special inodes like journal inode on last iput() | |
2394 | * because that could violate lock ordering on umount | |
2395 | */ | |
a1d6cc56 | 2396 | if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) |
bbf023c7 | 2397 | goto out_writepages; |
2a21e37e | 2398 | |
20970ba6 TT |
2399 | if (ext4_should_journal_data(inode)) { |
2400 | struct blk_plug plug; | |
20970ba6 TT |
2401 | |
2402 | blk_start_plug(&plug); | |
2403 | ret = write_cache_pages(mapping, wbc, __writepage, mapping); | |
2404 | blk_finish_plug(&plug); | |
bbf023c7 | 2405 | goto out_writepages; |
20970ba6 TT |
2406 | } |
2407 | ||
2a21e37e TT |
2408 | /* |
2409 | * If the filesystem has aborted, it is read-only, so return | |
2410 | * right away instead of dumping stack traces later on that | |
2411 | * will obscure the real source of the problem. We test | |
4ab2f15b | 2412 | * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because |
2a21e37e | 2413 | * the latter could be true if the filesystem is mounted |
20970ba6 | 2414 | * read-only, and in that case, ext4_writepages should |
2a21e37e TT |
2415 | * *never* be called, so if that ever happens, we would want |
2416 | * the stack trace. | |
2417 | */ | |
bbf023c7 ML |
2418 | if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) { |
2419 | ret = -EROFS; | |
2420 | goto out_writepages; | |
2421 | } | |
2a21e37e | 2422 | |
6b523df4 JK |
2423 | if (ext4_should_dioread_nolock(inode)) { |
2424 | /* | |
70261f56 | 2425 | * We may need to convert up to one extent per block in |
6b523df4 JK |
2426 | * the page and we may dirty the inode. |
2427 | */ | |
2428 | rsv_blocks = 1 + (PAGE_CACHE_SIZE >> inode->i_blkbits); | |
2429 | } | |
2430 | ||
4e7ea81d JK |
2431 | /* |
2432 | * If we have inline data and arrive here, it means that | |
2433 | * we will soon create the block for the 1st page, so | |
2434 | * we'd better clear the inline data here. | |
2435 | */ | |
2436 | if (ext4_has_inline_data(inode)) { | |
2437 | /* Just inode will be modified... */ | |
2438 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 1); | |
2439 | if (IS_ERR(handle)) { | |
2440 | ret = PTR_ERR(handle); | |
2441 | goto out_writepages; | |
2442 | } | |
2443 | BUG_ON(ext4_test_inode_state(inode, | |
2444 | EXT4_STATE_MAY_INLINE_DATA)); | |
2445 | ext4_destroy_inline_data(handle, inode); | |
2446 | ext4_journal_stop(handle); | |
2447 | } | |
2448 | ||
22208ded AK |
2449 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
2450 | range_whole = 1; | |
61628a3f | 2451 | |
2acf2c26 | 2452 | if (wbc->range_cyclic) { |
4e7ea81d JK |
2453 | writeback_index = mapping->writeback_index; |
2454 | if (writeback_index) | |
2acf2c26 | 2455 | cycled = 0; |
4e7ea81d JK |
2456 | mpd.first_page = writeback_index; |
2457 | mpd.last_page = -1; | |
5b41d924 | 2458 | } else { |
4e7ea81d JK |
2459 | mpd.first_page = wbc->range_start >> PAGE_CACHE_SHIFT; |
2460 | mpd.last_page = wbc->range_end >> PAGE_CACHE_SHIFT; | |
5b41d924 | 2461 | } |
a1d6cc56 | 2462 | |
4e7ea81d JK |
2463 | mpd.inode = inode; |
2464 | mpd.wbc = wbc; | |
2465 | ext4_io_submit_init(&mpd.io_submit, wbc); | |
2acf2c26 | 2466 | retry: |
6e6938b6 | 2467 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
4e7ea81d JK |
2468 | tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page); |
2469 | done = false; | |
1bce63d1 | 2470 | blk_start_plug(&plug); |
4e7ea81d JK |
2471 | while (!done && mpd.first_page <= mpd.last_page) { |
2472 | /* For each extent of pages we use new io_end */ | |
2473 | mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL); | |
2474 | if (!mpd.io_submit.io_end) { | |
2475 | ret = -ENOMEM; | |
2476 | break; | |
2477 | } | |
a1d6cc56 AK |
2478 | |
2479 | /* | |
4e7ea81d JK |
2480 | * We have two constraints: We find one extent to map and we |
2481 | * must always write out whole page (makes a difference when | |
2482 | * blocksize < pagesize) so that we don't block on IO when we | |
2483 | * try to write out the rest of the page. Journalled mode is | |
2484 | * not supported by delalloc. | |
a1d6cc56 AK |
2485 | */ |
2486 | BUG_ON(ext4_should_journal_data(inode)); | |
525f4ed8 | 2487 | needed_blocks = ext4_da_writepages_trans_blocks(inode); |
a1d6cc56 | 2488 | |
4e7ea81d | 2489 | /* start a new transaction */ |
6b523df4 JK |
2490 | handle = ext4_journal_start_with_reserve(inode, |
2491 | EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks); | |
61628a3f MC |
2492 | if (IS_ERR(handle)) { |
2493 | ret = PTR_ERR(handle); | |
1693918e | 2494 | ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: " |
fbe845dd | 2495 | "%ld pages, ino %lu; err %d", __func__, |
a1d6cc56 | 2496 | wbc->nr_to_write, inode->i_ino, ret); |
4e7ea81d JK |
2497 | /* Release allocated io_end */ |
2498 | ext4_put_io_end(mpd.io_submit.io_end); | |
2499 | break; | |
61628a3f | 2500 | } |
f63e6005 | 2501 | |
4e7ea81d JK |
2502 | trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc); |
2503 | ret = mpage_prepare_extent_to_map(&mpd); | |
2504 | if (!ret) { | |
2505 | if (mpd.map.m_len) | |
cb530541 TT |
2506 | ret = mpage_map_and_submit_extent(handle, &mpd, |
2507 | &give_up_on_write); | |
4e7ea81d JK |
2508 | else { |
2509 | /* | |
2510 | * We scanned the whole range (or exhausted | |
2511 | * nr_to_write), submitted what was mapped and | |
2512 | * didn't find anything needing mapping. We are | |
2513 | * done. | |
2514 | */ | |
2515 | done = true; | |
2516 | } | |
f63e6005 | 2517 | } |
61628a3f | 2518 | ext4_journal_stop(handle); |
4e7ea81d JK |
2519 | /* Submit prepared bio */ |
2520 | ext4_io_submit(&mpd.io_submit); | |
2521 | /* Unlock pages we didn't use */ | |
cb530541 | 2522 | mpage_release_unused_pages(&mpd, give_up_on_write); |
4e7ea81d JK |
2523 | /* Drop our io_end reference we got from init */ |
2524 | ext4_put_io_end(mpd.io_submit.io_end); | |
2525 | ||
2526 | if (ret == -ENOSPC && sbi->s_journal) { | |
2527 | /* | |
2528 | * Commit the transaction which would | |
22208ded AK |
2529 | * free blocks released in the transaction |
2530 | * and try again | |
2531 | */ | |
df22291f | 2532 | jbd2_journal_force_commit_nested(sbi->s_journal); |
22208ded | 2533 | ret = 0; |
4e7ea81d JK |
2534 | continue; |
2535 | } | |
2536 | /* Fatal error - ENOMEM, EIO... */ | |
2537 | if (ret) | |
61628a3f | 2538 | break; |
a1d6cc56 | 2539 | } |
1bce63d1 | 2540 | blk_finish_plug(&plug); |
9c12a831 | 2541 | if (!ret && !cycled && wbc->nr_to_write > 0) { |
2acf2c26 | 2542 | cycled = 1; |
4e7ea81d JK |
2543 | mpd.last_page = writeback_index - 1; |
2544 | mpd.first_page = 0; | |
2acf2c26 AK |
2545 | goto retry; |
2546 | } | |
22208ded AK |
2547 | |
2548 | /* Update index */ | |
22208ded AK |
2549 | if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) |
2550 | /* | |
4e7ea81d | 2551 | * Set the writeback_index so that range_cyclic |
22208ded AK |
2552 | * mode will write it back later |
2553 | */ | |
4e7ea81d | 2554 | mapping->writeback_index = mpd.first_page; |
a1d6cc56 | 2555 | |
61628a3f | 2556 | out_writepages: |
20970ba6 TT |
2557 | trace_ext4_writepages_result(inode, wbc, ret, |
2558 | nr_to_write - wbc->nr_to_write); | |
61628a3f | 2559 | return ret; |
64769240 AT |
2560 | } |
2561 | ||
79f0be8d AK |
2562 | static int ext4_nonda_switch(struct super_block *sb) |
2563 | { | |
5c1ff336 | 2564 | s64 free_clusters, dirty_clusters; |
79f0be8d AK |
2565 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
2566 | ||
2567 | /* | |
2568 | * switch to non delalloc mode if we are running low | |
2569 | * on free block. The free block accounting via percpu | |
179f7ebf | 2570 | * counters can get slightly wrong with percpu_counter_batch getting |
79f0be8d AK |
2571 | * accumulated on each CPU without updating global counters |
2572 | * Delalloc need an accurate free block accounting. So switch | |
2573 | * to non delalloc when we are near to error range. | |
2574 | */ | |
5c1ff336 EW |
2575 | free_clusters = |
2576 | percpu_counter_read_positive(&sbi->s_freeclusters_counter); | |
2577 | dirty_clusters = | |
2578 | percpu_counter_read_positive(&sbi->s_dirtyclusters_counter); | |
00d4e736 TT |
2579 | /* |
2580 | * Start pushing delalloc when 1/2 of free blocks are dirty. | |
2581 | */ | |
5c1ff336 | 2582 | if (dirty_clusters && (free_clusters < 2 * dirty_clusters)) |
10ee27a0 | 2583 | try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE); |
00d4e736 | 2584 | |
5c1ff336 EW |
2585 | if (2 * free_clusters < 3 * dirty_clusters || |
2586 | free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) { | |
79f0be8d | 2587 | /* |
c8afb446 ES |
2588 | * free block count is less than 150% of dirty blocks |
2589 | * or free blocks is less than watermark | |
79f0be8d AK |
2590 | */ |
2591 | return 1; | |
2592 | } | |
2593 | return 0; | |
2594 | } | |
2595 | ||
0ff8947f ES |
2596 | /* We always reserve for an inode update; the superblock could be there too */ |
2597 | static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len) | |
2598 | { | |
2599 | if (likely(EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, | |
2600 | EXT4_FEATURE_RO_COMPAT_LARGE_FILE))) | |
2601 | return 1; | |
2602 | ||
2603 | if (pos + len <= 0x7fffffffULL) | |
2604 | return 1; | |
2605 | ||
2606 | /* We might need to update the superblock to set LARGE_FILE */ | |
2607 | return 2; | |
2608 | } | |
2609 | ||
64769240 | 2610 | static int ext4_da_write_begin(struct file *file, struct address_space *mapping, |
de9a55b8 TT |
2611 | loff_t pos, unsigned len, unsigned flags, |
2612 | struct page **pagep, void **fsdata) | |
64769240 | 2613 | { |
72b8ab9d | 2614 | int ret, retries = 0; |
64769240 AT |
2615 | struct page *page; |
2616 | pgoff_t index; | |
64769240 AT |
2617 | struct inode *inode = mapping->host; |
2618 | handle_t *handle; | |
2619 | ||
2620 | index = pos >> PAGE_CACHE_SHIFT; | |
79f0be8d AK |
2621 | |
2622 | if (ext4_nonda_switch(inode->i_sb)) { | |
2623 | *fsdata = (void *)FALL_BACK_TO_NONDELALLOC; | |
2624 | return ext4_write_begin(file, mapping, pos, | |
2625 | len, flags, pagep, fsdata); | |
2626 | } | |
2627 | *fsdata = (void *)0; | |
9bffad1e | 2628 | trace_ext4_da_write_begin(inode, pos, len, flags); |
9c3569b5 TM |
2629 | |
2630 | if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { | |
2631 | ret = ext4_da_write_inline_data_begin(mapping, inode, | |
2632 | pos, len, flags, | |
2633 | pagep, fsdata); | |
2634 | if (ret < 0) | |
47564bfb TT |
2635 | return ret; |
2636 | if (ret == 1) | |
2637 | return 0; | |
9c3569b5 TM |
2638 | } |
2639 | ||
47564bfb TT |
2640 | /* |
2641 | * grab_cache_page_write_begin() can take a long time if the | |
2642 | * system is thrashing due to memory pressure, or if the page | |
2643 | * is being written back. So grab it first before we start | |
2644 | * the transaction handle. This also allows us to allocate | |
2645 | * the page (if needed) without using GFP_NOFS. | |
2646 | */ | |
2647 | retry_grab: | |
2648 | page = grab_cache_page_write_begin(mapping, index, flags); | |
2649 | if (!page) | |
2650 | return -ENOMEM; | |
2651 | unlock_page(page); | |
2652 | ||
64769240 AT |
2653 | /* |
2654 | * With delayed allocation, we don't log the i_disksize update | |
2655 | * if there is delayed block allocation. But we still need | |
2656 | * to journalling the i_disksize update if writes to the end | |
2657 | * of file which has an already mapped buffer. | |
2658 | */ | |
47564bfb | 2659 | retry_journal: |
0ff8947f ES |
2660 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
2661 | ext4_da_write_credits(inode, pos, len)); | |
64769240 | 2662 | if (IS_ERR(handle)) { |
47564bfb TT |
2663 | page_cache_release(page); |
2664 | return PTR_ERR(handle); | |
64769240 AT |
2665 | } |
2666 | ||
47564bfb TT |
2667 | lock_page(page); |
2668 | if (page->mapping != mapping) { | |
2669 | /* The page got truncated from under us */ | |
2670 | unlock_page(page); | |
2671 | page_cache_release(page); | |
d5a0d4f7 | 2672 | ext4_journal_stop(handle); |
47564bfb | 2673 | goto retry_grab; |
d5a0d4f7 | 2674 | } |
47564bfb | 2675 | /* In case writeback began while the page was unlocked */ |
7afe5aa5 | 2676 | wait_for_stable_page(page); |
64769240 | 2677 | |
2058f83a MH |
2678 | #ifdef CONFIG_EXT4_FS_ENCRYPTION |
2679 | ret = ext4_block_write_begin(page, pos, len, | |
2680 | ext4_da_get_block_prep); | |
2681 | #else | |
6e1db88d | 2682 | ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep); |
2058f83a | 2683 | #endif |
64769240 AT |
2684 | if (ret < 0) { |
2685 | unlock_page(page); | |
2686 | ext4_journal_stop(handle); | |
ae4d5372 AK |
2687 | /* |
2688 | * block_write_begin may have instantiated a few blocks | |
2689 | * outside i_size. Trim these off again. Don't need | |
2690 | * i_size_read because we hold i_mutex. | |
2691 | */ | |
2692 | if (pos + len > inode->i_size) | |
b9a4207d | 2693 | ext4_truncate_failed_write(inode); |
47564bfb TT |
2694 | |
2695 | if (ret == -ENOSPC && | |
2696 | ext4_should_retry_alloc(inode->i_sb, &retries)) | |
2697 | goto retry_journal; | |
2698 | ||
2699 | page_cache_release(page); | |
2700 | return ret; | |
64769240 AT |
2701 | } |
2702 | ||
47564bfb | 2703 | *pagep = page; |
64769240 AT |
2704 | return ret; |
2705 | } | |
2706 | ||
632eaeab MC |
2707 | /* |
2708 | * Check if we should update i_disksize | |
2709 | * when write to the end of file but not require block allocation | |
2710 | */ | |
2711 | static int ext4_da_should_update_i_disksize(struct page *page, | |
de9a55b8 | 2712 | unsigned long offset) |
632eaeab MC |
2713 | { |
2714 | struct buffer_head *bh; | |
2715 | struct inode *inode = page->mapping->host; | |
2716 | unsigned int idx; | |
2717 | int i; | |
2718 | ||
2719 | bh = page_buffers(page); | |
2720 | idx = offset >> inode->i_blkbits; | |
2721 | ||
af5bc92d | 2722 | for (i = 0; i < idx; i++) |
632eaeab MC |
2723 | bh = bh->b_this_page; |
2724 | ||
29fa89d0 | 2725 | if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh)) |
632eaeab MC |
2726 | return 0; |
2727 | return 1; | |
2728 | } | |
2729 | ||
64769240 | 2730 | static int ext4_da_write_end(struct file *file, |
de9a55b8 TT |
2731 | struct address_space *mapping, |
2732 | loff_t pos, unsigned len, unsigned copied, | |
2733 | struct page *page, void *fsdata) | |
64769240 AT |
2734 | { |
2735 | struct inode *inode = mapping->host; | |
2736 | int ret = 0, ret2; | |
2737 | handle_t *handle = ext4_journal_current_handle(); | |
2738 | loff_t new_i_size; | |
632eaeab | 2739 | unsigned long start, end; |
79f0be8d AK |
2740 | int write_mode = (int)(unsigned long)fsdata; |
2741 | ||
74d553aa TT |
2742 | if (write_mode == FALL_BACK_TO_NONDELALLOC) |
2743 | return ext4_write_end(file, mapping, pos, | |
2744 | len, copied, page, fsdata); | |
632eaeab | 2745 | |
9bffad1e | 2746 | trace_ext4_da_write_end(inode, pos, len, copied); |
632eaeab | 2747 | start = pos & (PAGE_CACHE_SIZE - 1); |
af5bc92d | 2748 | end = start + copied - 1; |
64769240 AT |
2749 | |
2750 | /* | |
2751 | * generic_write_end() will run mark_inode_dirty() if i_size | |
2752 | * changes. So let's piggyback the i_disksize mark_inode_dirty | |
2753 | * into that. | |
2754 | */ | |
64769240 | 2755 | new_i_size = pos + copied; |
ea51d132 | 2756 | if (copied && new_i_size > EXT4_I(inode)->i_disksize) { |
9c3569b5 TM |
2757 | if (ext4_has_inline_data(inode) || |
2758 | ext4_da_should_update_i_disksize(page, end)) { | |
ee124d27 | 2759 | ext4_update_i_disksize(inode, new_i_size); |
cf17fea6 AK |
2760 | /* We need to mark inode dirty even if |
2761 | * new_i_size is less that inode->i_size | |
2762 | * bu greater than i_disksize.(hint delalloc) | |
2763 | */ | |
2764 | ext4_mark_inode_dirty(handle, inode); | |
64769240 | 2765 | } |
632eaeab | 2766 | } |
9c3569b5 TM |
2767 | |
2768 | if (write_mode != CONVERT_INLINE_DATA && | |
2769 | ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) && | |
2770 | ext4_has_inline_data(inode)) | |
2771 | ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied, | |
2772 | page); | |
2773 | else | |
2774 | ret2 = generic_write_end(file, mapping, pos, len, copied, | |
64769240 | 2775 | page, fsdata); |
9c3569b5 | 2776 | |
64769240 AT |
2777 | copied = ret2; |
2778 | if (ret2 < 0) | |
2779 | ret = ret2; | |
2780 | ret2 = ext4_journal_stop(handle); | |
2781 | if (!ret) | |
2782 | ret = ret2; | |
2783 | ||
2784 | return ret ? ret : copied; | |
2785 | } | |
2786 | ||
d47992f8 LC |
2787 | static void ext4_da_invalidatepage(struct page *page, unsigned int offset, |
2788 | unsigned int length) | |
64769240 | 2789 | { |
64769240 AT |
2790 | /* |
2791 | * Drop reserved blocks | |
2792 | */ | |
2793 | BUG_ON(!PageLocked(page)); | |
2794 | if (!page_has_buffers(page)) | |
2795 | goto out; | |
2796 | ||
ca99fdd2 | 2797 | ext4_da_page_release_reservation(page, offset, length); |
64769240 AT |
2798 | |
2799 | out: | |
d47992f8 | 2800 | ext4_invalidatepage(page, offset, length); |
64769240 AT |
2801 | |
2802 | return; | |
2803 | } | |
2804 | ||
ccd2506b TT |
2805 | /* |
2806 | * Force all delayed allocation blocks to be allocated for a given inode. | |
2807 | */ | |
2808 | int ext4_alloc_da_blocks(struct inode *inode) | |
2809 | { | |
fb40ba0d TT |
2810 | trace_ext4_alloc_da_blocks(inode); |
2811 | ||
71d4f7d0 | 2812 | if (!EXT4_I(inode)->i_reserved_data_blocks) |
ccd2506b TT |
2813 | return 0; |
2814 | ||
2815 | /* | |
2816 | * We do something simple for now. The filemap_flush() will | |
2817 | * also start triggering a write of the data blocks, which is | |
2818 | * not strictly speaking necessary (and for users of | |
2819 | * laptop_mode, not even desirable). However, to do otherwise | |
2820 | * would require replicating code paths in: | |
de9a55b8 | 2821 | * |
20970ba6 | 2822 | * ext4_writepages() -> |
ccd2506b TT |
2823 | * write_cache_pages() ---> (via passed in callback function) |
2824 | * __mpage_da_writepage() --> | |
2825 | * mpage_add_bh_to_extent() | |
2826 | * mpage_da_map_blocks() | |
2827 | * | |
2828 | * The problem is that write_cache_pages(), located in | |
2829 | * mm/page-writeback.c, marks pages clean in preparation for | |
2830 | * doing I/O, which is not desirable if we're not planning on | |
2831 | * doing I/O at all. | |
2832 | * | |
2833 | * We could call write_cache_pages(), and then redirty all of | |
380cf090 | 2834 | * the pages by calling redirty_page_for_writepage() but that |
ccd2506b TT |
2835 | * would be ugly in the extreme. So instead we would need to |
2836 | * replicate parts of the code in the above functions, | |
25985edc | 2837 | * simplifying them because we wouldn't actually intend to |
ccd2506b TT |
2838 | * write out the pages, but rather only collect contiguous |
2839 | * logical block extents, call the multi-block allocator, and | |
2840 | * then update the buffer heads with the block allocations. | |
de9a55b8 | 2841 | * |
ccd2506b TT |
2842 | * For now, though, we'll cheat by calling filemap_flush(), |
2843 | * which will map the blocks, and start the I/O, but not | |
2844 | * actually wait for the I/O to complete. | |
2845 | */ | |
2846 | return filemap_flush(inode->i_mapping); | |
2847 | } | |
64769240 | 2848 | |
ac27a0ec DK |
2849 | /* |
2850 | * bmap() is special. It gets used by applications such as lilo and by | |
2851 | * the swapper to find the on-disk block of a specific piece of data. | |
2852 | * | |
2853 | * Naturally, this is dangerous if the block concerned is still in the | |
617ba13b | 2854 | * journal. If somebody makes a swapfile on an ext4 data-journaling |
ac27a0ec DK |
2855 | * filesystem and enables swap, then they may get a nasty shock when the |
2856 | * data getting swapped to that swapfile suddenly gets overwritten by | |
2857 | * the original zero's written out previously to the journal and | |
2858 | * awaiting writeback in the kernel's buffer cache. | |
2859 | * | |
2860 | * So, if we see any bmap calls here on a modified, data-journaled file, | |
2861 | * take extra steps to flush any blocks which might be in the cache. | |
2862 | */ | |
617ba13b | 2863 | static sector_t ext4_bmap(struct address_space *mapping, sector_t block) |
ac27a0ec DK |
2864 | { |
2865 | struct inode *inode = mapping->host; | |
2866 | journal_t *journal; | |
2867 | int err; | |
2868 | ||
46c7f254 TM |
2869 | /* |
2870 | * We can get here for an inline file via the FIBMAP ioctl | |
2871 | */ | |
2872 | if (ext4_has_inline_data(inode)) | |
2873 | return 0; | |
2874 | ||
64769240 AT |
2875 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) && |
2876 | test_opt(inode->i_sb, DELALLOC)) { | |
2877 | /* | |
2878 | * With delalloc we want to sync the file | |
2879 | * so that we can make sure we allocate | |
2880 | * blocks for file | |
2881 | */ | |
2882 | filemap_write_and_wait(mapping); | |
2883 | } | |
2884 | ||
19f5fb7a TT |
2885 | if (EXT4_JOURNAL(inode) && |
2886 | ext4_test_inode_state(inode, EXT4_STATE_JDATA)) { | |
ac27a0ec DK |
2887 | /* |
2888 | * This is a REALLY heavyweight approach, but the use of | |
2889 | * bmap on dirty files is expected to be extremely rare: | |
2890 | * only if we run lilo or swapon on a freshly made file | |
2891 | * do we expect this to happen. | |
2892 | * | |
2893 | * (bmap requires CAP_SYS_RAWIO so this does not | |
2894 | * represent an unprivileged user DOS attack --- we'd be | |
2895 | * in trouble if mortal users could trigger this path at | |
2896 | * will.) | |
2897 | * | |
617ba13b | 2898 | * NB. EXT4_STATE_JDATA is not set on files other than |
ac27a0ec DK |
2899 | * regular files. If somebody wants to bmap a directory |
2900 | * or symlink and gets confused because the buffer | |
2901 | * hasn't yet been flushed to disk, they deserve | |
2902 | * everything they get. | |
2903 | */ | |
2904 | ||
19f5fb7a | 2905 | ext4_clear_inode_state(inode, EXT4_STATE_JDATA); |
617ba13b | 2906 | journal = EXT4_JOURNAL(inode); |
dab291af MC |
2907 | jbd2_journal_lock_updates(journal); |
2908 | err = jbd2_journal_flush(journal); | |
2909 | jbd2_journal_unlock_updates(journal); | |
ac27a0ec DK |
2910 | |
2911 | if (err) | |
2912 | return 0; | |
2913 | } | |
2914 | ||
af5bc92d | 2915 | return generic_block_bmap(mapping, block, ext4_get_block); |
ac27a0ec DK |
2916 | } |
2917 | ||
617ba13b | 2918 | static int ext4_readpage(struct file *file, struct page *page) |
ac27a0ec | 2919 | { |
46c7f254 TM |
2920 | int ret = -EAGAIN; |
2921 | struct inode *inode = page->mapping->host; | |
2922 | ||
0562e0ba | 2923 | trace_ext4_readpage(page); |
46c7f254 TM |
2924 | |
2925 | if (ext4_has_inline_data(inode)) | |
2926 | ret = ext4_readpage_inline(inode, page); | |
2927 | ||
2928 | if (ret == -EAGAIN) | |
f64e02fe | 2929 | return ext4_mpage_readpages(page->mapping, NULL, page, 1); |
46c7f254 TM |
2930 | |
2931 | return ret; | |
ac27a0ec DK |
2932 | } |
2933 | ||
2934 | static int | |
617ba13b | 2935 | ext4_readpages(struct file *file, struct address_space *mapping, |
ac27a0ec DK |
2936 | struct list_head *pages, unsigned nr_pages) |
2937 | { | |
46c7f254 TM |
2938 | struct inode *inode = mapping->host; |
2939 | ||
2940 | /* If the file has inline data, no need to do readpages. */ | |
2941 | if (ext4_has_inline_data(inode)) | |
2942 | return 0; | |
2943 | ||
f64e02fe | 2944 | return ext4_mpage_readpages(mapping, pages, NULL, nr_pages); |
ac27a0ec DK |
2945 | } |
2946 | ||
d47992f8 LC |
2947 | static void ext4_invalidatepage(struct page *page, unsigned int offset, |
2948 | unsigned int length) | |
ac27a0ec | 2949 | { |
ca99fdd2 | 2950 | trace_ext4_invalidatepage(page, offset, length); |
0562e0ba | 2951 | |
4520fb3c JK |
2952 | /* No journalling happens on data buffers when this function is used */ |
2953 | WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page))); | |
2954 | ||
ca99fdd2 | 2955 | block_invalidatepage(page, offset, length); |
4520fb3c JK |
2956 | } |
2957 | ||
53e87268 | 2958 | static int __ext4_journalled_invalidatepage(struct page *page, |
ca99fdd2 LC |
2959 | unsigned int offset, |
2960 | unsigned int length) | |
4520fb3c JK |
2961 | { |
2962 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); | |
2963 | ||
ca99fdd2 | 2964 | trace_ext4_journalled_invalidatepage(page, offset, length); |
4520fb3c | 2965 | |
ac27a0ec DK |
2966 | /* |
2967 | * If it's a full truncate we just forget about the pending dirtying | |
2968 | */ | |
ca99fdd2 | 2969 | if (offset == 0 && length == PAGE_CACHE_SIZE) |
ac27a0ec DK |
2970 | ClearPageChecked(page); |
2971 | ||
ca99fdd2 | 2972 | return jbd2_journal_invalidatepage(journal, page, offset, length); |
53e87268 JK |
2973 | } |
2974 | ||
2975 | /* Wrapper for aops... */ | |
2976 | static void ext4_journalled_invalidatepage(struct page *page, | |
d47992f8 LC |
2977 | unsigned int offset, |
2978 | unsigned int length) | |
53e87268 | 2979 | { |
ca99fdd2 | 2980 | WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0); |
ac27a0ec DK |
2981 | } |
2982 | ||
617ba13b | 2983 | static int ext4_releasepage(struct page *page, gfp_t wait) |
ac27a0ec | 2984 | { |
617ba13b | 2985 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); |
ac27a0ec | 2986 | |
0562e0ba JZ |
2987 | trace_ext4_releasepage(page); |
2988 | ||
e1c36595 JK |
2989 | /* Page has dirty journalled data -> cannot release */ |
2990 | if (PageChecked(page)) | |
ac27a0ec | 2991 | return 0; |
0390131b FM |
2992 | if (journal) |
2993 | return jbd2_journal_try_to_free_buffers(journal, page, wait); | |
2994 | else | |
2995 | return try_to_free_buffers(page); | |
ac27a0ec DK |
2996 | } |
2997 | ||
2ed88685 TT |
2998 | /* |
2999 | * ext4_get_block used when preparing for a DIO write or buffer write. | |
3000 | * We allocate an uinitialized extent if blocks haven't been allocated. | |
3001 | * The extent will be converted to initialized after the IO is complete. | |
3002 | */ | |
f19d5870 | 3003 | int ext4_get_block_write(struct inode *inode, sector_t iblock, |
4c0425ff MC |
3004 | struct buffer_head *bh_result, int create) |
3005 | { | |
c7064ef1 | 3006 | ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n", |
8d5d02e6 | 3007 | inode->i_ino, create); |
2ed88685 TT |
3008 | return _ext4_get_block(inode, iblock, bh_result, |
3009 | EXT4_GET_BLOCKS_IO_CREATE_EXT); | |
4c0425ff MC |
3010 | } |
3011 | ||
729f52c6 | 3012 | static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock, |
8b0f165f | 3013 | struct buffer_head *bh_result, int create) |
729f52c6 | 3014 | { |
8b0f165f AP |
3015 | ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n", |
3016 | inode->i_ino, create); | |
3017 | return _ext4_get_block(inode, iblock, bh_result, | |
3018 | EXT4_GET_BLOCKS_NO_LOCK); | |
729f52c6 ZL |
3019 | } |
3020 | ||
4c0425ff | 3021 | static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset, |
7b7a8665 | 3022 | ssize_t size, void *private) |
4c0425ff MC |
3023 | { |
3024 | ext4_io_end_t *io_end = iocb->private; | |
4c0425ff | 3025 | |
97a851ed | 3026 | /* if not async direct IO just return */ |
7b7a8665 | 3027 | if (!io_end) |
97a851ed | 3028 | return; |
4b70df18 | 3029 | |
88635ca2 | 3030 | ext_debug("ext4_end_io_dio(): io_end 0x%p " |
ace36ad4 | 3031 | "for inode %lu, iocb 0x%p, offset %llu, size %zd\n", |
8d5d02e6 MC |
3032 | iocb->private, io_end->inode->i_ino, iocb, offset, |
3033 | size); | |
8d5d02e6 | 3034 | |
b5a7e970 | 3035 | iocb->private = NULL; |
4c0425ff MC |
3036 | io_end->offset = offset; |
3037 | io_end->size = size; | |
7b7a8665 | 3038 | ext4_put_io_end(io_end); |
4c0425ff | 3039 | } |
c7064ef1 | 3040 | |
4c0425ff MC |
3041 | /* |
3042 | * For ext4 extent files, ext4 will do direct-io write to holes, | |
3043 | * preallocated extents, and those write extend the file, no need to | |
3044 | * fall back to buffered IO. | |
3045 | * | |
556615dc | 3046 | * For holes, we fallocate those blocks, mark them as unwritten |
69c499d1 | 3047 | * If those blocks were preallocated, we mark sure they are split, but |
556615dc | 3048 | * still keep the range to write as unwritten. |
4c0425ff | 3049 | * |
69c499d1 | 3050 | * The unwritten extents will be converted to written when DIO is completed. |
8d5d02e6 | 3051 | * For async direct IO, since the IO may still pending when return, we |
25985edc | 3052 | * set up an end_io call back function, which will do the conversion |
8d5d02e6 | 3053 | * when async direct IO completed. |
4c0425ff MC |
3054 | * |
3055 | * If the O_DIRECT write will extend the file then add this inode to the | |
3056 | * orphan list. So recovery will truncate it back to the original size | |
3057 | * if the machine crashes during the write. | |
3058 | * | |
3059 | */ | |
6f673763 OS |
3060 | static ssize_t ext4_ext_direct_IO(struct kiocb *iocb, struct iov_iter *iter, |
3061 | loff_t offset) | |
4c0425ff MC |
3062 | { |
3063 | struct file *file = iocb->ki_filp; | |
3064 | struct inode *inode = file->f_mapping->host; | |
3065 | ssize_t ret; | |
a6cbcd4a | 3066 | size_t count = iov_iter_count(iter); |
69c499d1 TT |
3067 | int overwrite = 0; |
3068 | get_block_t *get_block_func = NULL; | |
3069 | int dio_flags = 0; | |
4c0425ff | 3070 | loff_t final_size = offset + count; |
97a851ed | 3071 | ext4_io_end_t *io_end = NULL; |
729f52c6 | 3072 | |
69c499d1 | 3073 | /* Use the old path for reads and writes beyond i_size. */ |
6f673763 OS |
3074 | if (iov_iter_rw(iter) != WRITE || final_size > inode->i_size) |
3075 | return ext4_ind_direct_IO(iocb, iter, offset); | |
4bd809db | 3076 | |
69c499d1 | 3077 | BUG_ON(iocb->private == NULL); |
4bd809db | 3078 | |
e8340395 JK |
3079 | /* |
3080 | * Make all waiters for direct IO properly wait also for extent | |
3081 | * conversion. This also disallows race between truncate() and | |
3082 | * overwrite DIO as i_dio_count needs to be incremented under i_mutex. | |
3083 | */ | |
6f673763 | 3084 | if (iov_iter_rw(iter) == WRITE) |
fe0f07d0 | 3085 | inode_dio_begin(inode); |
e8340395 | 3086 | |
69c499d1 TT |
3087 | /* If we do a overwrite dio, i_mutex locking can be released */ |
3088 | overwrite = *((int *)iocb->private); | |
4bd809db | 3089 | |
69c499d1 | 3090 | if (overwrite) { |
69c499d1 TT |
3091 | down_read(&EXT4_I(inode)->i_data_sem); |
3092 | mutex_unlock(&inode->i_mutex); | |
3093 | } | |
8d5d02e6 | 3094 | |
69c499d1 TT |
3095 | /* |
3096 | * We could direct write to holes and fallocate. | |
3097 | * | |
3098 | * Allocated blocks to fill the hole are marked as | |
556615dc | 3099 | * unwritten to prevent parallel buffered read to expose |
69c499d1 TT |
3100 | * the stale data before DIO complete the data IO. |
3101 | * | |
3102 | * As to previously fallocated extents, ext4 get_block will | |
3103 | * just simply mark the buffer mapped but still keep the | |
556615dc | 3104 | * extents unwritten. |
69c499d1 TT |
3105 | * |
3106 | * For non AIO case, we will convert those unwritten extents | |
3107 | * to written after return back from blockdev_direct_IO. | |
3108 | * | |
3109 | * For async DIO, the conversion needs to be deferred when the | |
3110 | * IO is completed. The ext4 end_io callback function will be | |
3111 | * called to take care of the conversion work. Here for async | |
3112 | * case, we allocate an io_end structure to hook to the iocb. | |
3113 | */ | |
3114 | iocb->private = NULL; | |
3115 | ext4_inode_aio_set(inode, NULL); | |
3116 | if (!is_sync_kiocb(iocb)) { | |
97a851ed | 3117 | io_end = ext4_init_io_end(inode, GFP_NOFS); |
69c499d1 TT |
3118 | if (!io_end) { |
3119 | ret = -ENOMEM; | |
3120 | goto retake_lock; | |
8b0f165f | 3121 | } |
97a851ed JK |
3122 | /* |
3123 | * Grab reference for DIO. Will be dropped in ext4_end_io_dio() | |
3124 | */ | |
3125 | iocb->private = ext4_get_io_end(io_end); | |
8d5d02e6 | 3126 | /* |
69c499d1 TT |
3127 | * we save the io structure for current async direct |
3128 | * IO, so that later ext4_map_blocks() could flag the | |
3129 | * io structure whether there is a unwritten extents | |
3130 | * needs to be converted when IO is completed. | |
8d5d02e6 | 3131 | */ |
69c499d1 TT |
3132 | ext4_inode_aio_set(inode, io_end); |
3133 | } | |
4bd809db | 3134 | |
69c499d1 TT |
3135 | if (overwrite) { |
3136 | get_block_func = ext4_get_block_write_nolock; | |
3137 | } else { | |
3138 | get_block_func = ext4_get_block_write; | |
3139 | dio_flags = DIO_LOCKING; | |
3140 | } | |
2058f83a MH |
3141 | #ifdef CONFIG_EXT4_FS_ENCRYPTION |
3142 | BUG_ON(ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode)); | |
3143 | #endif | |
923ae0ff | 3144 | if (IS_DAX(inode)) |
a95cd631 | 3145 | ret = dax_do_io(iocb, inode, iter, offset, get_block_func, |
923ae0ff RZ |
3146 | ext4_end_io_dio, dio_flags); |
3147 | else | |
17f8c842 | 3148 | ret = __blockdev_direct_IO(iocb, inode, |
923ae0ff RZ |
3149 | inode->i_sb->s_bdev, iter, offset, |
3150 | get_block_func, | |
3151 | ext4_end_io_dio, NULL, dio_flags); | |
69c499d1 | 3152 | |
69c499d1 | 3153 | /* |
97a851ed JK |
3154 | * Put our reference to io_end. This can free the io_end structure e.g. |
3155 | * in sync IO case or in case of error. It can even perform extent | |
3156 | * conversion if all bios we submitted finished before we got here. | |
3157 | * Note that in that case iocb->private can be already set to NULL | |
3158 | * here. | |
69c499d1 | 3159 | */ |
97a851ed JK |
3160 | if (io_end) { |
3161 | ext4_inode_aio_set(inode, NULL); | |
3162 | ext4_put_io_end(io_end); | |
3163 | /* | |
3164 | * When no IO was submitted ext4_end_io_dio() was not | |
3165 | * called so we have to put iocb's reference. | |
3166 | */ | |
3167 | if (ret <= 0 && ret != -EIOCBQUEUED && iocb->private) { | |
3168 | WARN_ON(iocb->private != io_end); | |
3169 | WARN_ON(io_end->flag & EXT4_IO_END_UNWRITTEN); | |
97a851ed JK |
3170 | ext4_put_io_end(io_end); |
3171 | iocb->private = NULL; | |
3172 | } | |
3173 | } | |
3174 | if (ret > 0 && !overwrite && ext4_test_inode_state(inode, | |
69c499d1 TT |
3175 | EXT4_STATE_DIO_UNWRITTEN)) { |
3176 | int err; | |
3177 | /* | |
3178 | * for non AIO case, since the IO is already | |
3179 | * completed, we could do the conversion right here | |
3180 | */ | |
6b523df4 | 3181 | err = ext4_convert_unwritten_extents(NULL, inode, |
69c499d1 TT |
3182 | offset, ret); |
3183 | if (err < 0) | |
3184 | ret = err; | |
3185 | ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN); | |
3186 | } | |
4bd809db | 3187 | |
69c499d1 | 3188 | retake_lock: |
6f673763 | 3189 | if (iov_iter_rw(iter) == WRITE) |
fe0f07d0 | 3190 | inode_dio_end(inode); |
69c499d1 TT |
3191 | /* take i_mutex locking again if we do a ovewrite dio */ |
3192 | if (overwrite) { | |
69c499d1 TT |
3193 | up_read(&EXT4_I(inode)->i_data_sem); |
3194 | mutex_lock(&inode->i_mutex); | |
4c0425ff | 3195 | } |
8d5d02e6 | 3196 | |
69c499d1 | 3197 | return ret; |
4c0425ff MC |
3198 | } |
3199 | ||
22c6186e OS |
3200 | static ssize_t ext4_direct_IO(struct kiocb *iocb, struct iov_iter *iter, |
3201 | loff_t offset) | |
4c0425ff MC |
3202 | { |
3203 | struct file *file = iocb->ki_filp; | |
3204 | struct inode *inode = file->f_mapping->host; | |
a6cbcd4a | 3205 | size_t count = iov_iter_count(iter); |
0562e0ba | 3206 | ssize_t ret; |
4c0425ff | 3207 | |
2058f83a MH |
3208 | #ifdef CONFIG_EXT4_FS_ENCRYPTION |
3209 | if (ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode)) | |
3210 | return 0; | |
3211 | #endif | |
3212 | ||
84ebd795 TT |
3213 | /* |
3214 | * If we are doing data journalling we don't support O_DIRECT | |
3215 | */ | |
3216 | if (ext4_should_journal_data(inode)) | |
3217 | return 0; | |
3218 | ||
46c7f254 TM |
3219 | /* Let buffer I/O handle the inline data case. */ |
3220 | if (ext4_has_inline_data(inode)) | |
3221 | return 0; | |
3222 | ||
6f673763 | 3223 | trace_ext4_direct_IO_enter(inode, offset, count, iov_iter_rw(iter)); |
12e9b892 | 3224 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
6f673763 | 3225 | ret = ext4_ext_direct_IO(iocb, iter, offset); |
0562e0ba | 3226 | else |
6f673763 OS |
3227 | ret = ext4_ind_direct_IO(iocb, iter, offset); |
3228 | trace_ext4_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), ret); | |
0562e0ba | 3229 | return ret; |
4c0425ff MC |
3230 | } |
3231 | ||
ac27a0ec | 3232 | /* |
617ba13b | 3233 | * Pages can be marked dirty completely asynchronously from ext4's journalling |
ac27a0ec DK |
3234 | * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do |
3235 | * much here because ->set_page_dirty is called under VFS locks. The page is | |
3236 | * not necessarily locked. | |
3237 | * | |
3238 | * We cannot just dirty the page and leave attached buffers clean, because the | |
3239 | * buffers' dirty state is "definitive". We cannot just set the buffers dirty | |
3240 | * or jbddirty because all the journalling code will explode. | |
3241 | * | |
3242 | * So what we do is to mark the page "pending dirty" and next time writepage | |
3243 | * is called, propagate that into the buffers appropriately. | |
3244 | */ | |
617ba13b | 3245 | static int ext4_journalled_set_page_dirty(struct page *page) |
ac27a0ec DK |
3246 | { |
3247 | SetPageChecked(page); | |
3248 | return __set_page_dirty_nobuffers(page); | |
3249 | } | |
3250 | ||
74d553aa | 3251 | static const struct address_space_operations ext4_aops = { |
8ab22b9a HH |
3252 | .readpage = ext4_readpage, |
3253 | .readpages = ext4_readpages, | |
43ce1d23 | 3254 | .writepage = ext4_writepage, |
20970ba6 | 3255 | .writepages = ext4_writepages, |
8ab22b9a | 3256 | .write_begin = ext4_write_begin, |
74d553aa | 3257 | .write_end = ext4_write_end, |
8ab22b9a HH |
3258 | .bmap = ext4_bmap, |
3259 | .invalidatepage = ext4_invalidatepage, | |
3260 | .releasepage = ext4_releasepage, | |
3261 | .direct_IO = ext4_direct_IO, | |
3262 | .migratepage = buffer_migrate_page, | |
3263 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 3264 | .error_remove_page = generic_error_remove_page, |
ac27a0ec DK |
3265 | }; |
3266 | ||
617ba13b | 3267 | static const struct address_space_operations ext4_journalled_aops = { |
8ab22b9a HH |
3268 | .readpage = ext4_readpage, |
3269 | .readpages = ext4_readpages, | |
43ce1d23 | 3270 | .writepage = ext4_writepage, |
20970ba6 | 3271 | .writepages = ext4_writepages, |
8ab22b9a HH |
3272 | .write_begin = ext4_write_begin, |
3273 | .write_end = ext4_journalled_write_end, | |
3274 | .set_page_dirty = ext4_journalled_set_page_dirty, | |
3275 | .bmap = ext4_bmap, | |
4520fb3c | 3276 | .invalidatepage = ext4_journalled_invalidatepage, |
8ab22b9a | 3277 | .releasepage = ext4_releasepage, |
84ebd795 | 3278 | .direct_IO = ext4_direct_IO, |
8ab22b9a | 3279 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 3280 | .error_remove_page = generic_error_remove_page, |
ac27a0ec DK |
3281 | }; |
3282 | ||
64769240 | 3283 | static const struct address_space_operations ext4_da_aops = { |
8ab22b9a HH |
3284 | .readpage = ext4_readpage, |
3285 | .readpages = ext4_readpages, | |
43ce1d23 | 3286 | .writepage = ext4_writepage, |
20970ba6 | 3287 | .writepages = ext4_writepages, |
8ab22b9a HH |
3288 | .write_begin = ext4_da_write_begin, |
3289 | .write_end = ext4_da_write_end, | |
3290 | .bmap = ext4_bmap, | |
3291 | .invalidatepage = ext4_da_invalidatepage, | |
3292 | .releasepage = ext4_releasepage, | |
3293 | .direct_IO = ext4_direct_IO, | |
3294 | .migratepage = buffer_migrate_page, | |
3295 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 3296 | .error_remove_page = generic_error_remove_page, |
64769240 AT |
3297 | }; |
3298 | ||
617ba13b | 3299 | void ext4_set_aops(struct inode *inode) |
ac27a0ec | 3300 | { |
3d2b1582 LC |
3301 | switch (ext4_inode_journal_mode(inode)) { |
3302 | case EXT4_INODE_ORDERED_DATA_MODE: | |
74d553aa | 3303 | ext4_set_inode_state(inode, EXT4_STATE_ORDERED_MODE); |
3d2b1582 LC |
3304 | break; |
3305 | case EXT4_INODE_WRITEBACK_DATA_MODE: | |
74d553aa | 3306 | ext4_clear_inode_state(inode, EXT4_STATE_ORDERED_MODE); |
3d2b1582 LC |
3307 | break; |
3308 | case EXT4_INODE_JOURNAL_DATA_MODE: | |
617ba13b | 3309 | inode->i_mapping->a_ops = &ext4_journalled_aops; |
74d553aa | 3310 | return; |
3d2b1582 LC |
3311 | default: |
3312 | BUG(); | |
3313 | } | |
74d553aa TT |
3314 | if (test_opt(inode->i_sb, DELALLOC)) |
3315 | inode->i_mapping->a_ops = &ext4_da_aops; | |
3316 | else | |
3317 | inode->i_mapping->a_ops = &ext4_aops; | |
ac27a0ec DK |
3318 | } |
3319 | ||
923ae0ff | 3320 | static int __ext4_block_zero_page_range(handle_t *handle, |
d863dc36 LC |
3321 | struct address_space *mapping, loff_t from, loff_t length) |
3322 | { | |
3323 | ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT; | |
3324 | unsigned offset = from & (PAGE_CACHE_SIZE-1); | |
923ae0ff | 3325 | unsigned blocksize, pos; |
d863dc36 LC |
3326 | ext4_lblk_t iblock; |
3327 | struct inode *inode = mapping->host; | |
3328 | struct buffer_head *bh; | |
3329 | struct page *page; | |
3330 | int err = 0; | |
3331 | ||
3332 | page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT, | |
3333 | mapping_gfp_mask(mapping) & ~__GFP_FS); | |
3334 | if (!page) | |
3335 | return -ENOMEM; | |
3336 | ||
3337 | blocksize = inode->i_sb->s_blocksize; | |
d863dc36 LC |
3338 | |
3339 | iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits); | |
3340 | ||
3341 | if (!page_has_buffers(page)) | |
3342 | create_empty_buffers(page, blocksize, 0); | |
3343 | ||
3344 | /* Find the buffer that contains "offset" */ | |
3345 | bh = page_buffers(page); | |
3346 | pos = blocksize; | |
3347 | while (offset >= pos) { | |
3348 | bh = bh->b_this_page; | |
3349 | iblock++; | |
3350 | pos += blocksize; | |
3351 | } | |
d863dc36 LC |
3352 | if (buffer_freed(bh)) { |
3353 | BUFFER_TRACE(bh, "freed: skip"); | |
3354 | goto unlock; | |
3355 | } | |
d863dc36 LC |
3356 | if (!buffer_mapped(bh)) { |
3357 | BUFFER_TRACE(bh, "unmapped"); | |
3358 | ext4_get_block(inode, iblock, bh, 0); | |
3359 | /* unmapped? It's a hole - nothing to do */ | |
3360 | if (!buffer_mapped(bh)) { | |
3361 | BUFFER_TRACE(bh, "still unmapped"); | |
3362 | goto unlock; | |
3363 | } | |
3364 | } | |
3365 | ||
3366 | /* Ok, it's mapped. Make sure it's up-to-date */ | |
3367 | if (PageUptodate(page)) | |
3368 | set_buffer_uptodate(bh); | |
3369 | ||
3370 | if (!buffer_uptodate(bh)) { | |
3371 | err = -EIO; | |
3372 | ll_rw_block(READ, 1, &bh); | |
3373 | wait_on_buffer(bh); | |
3374 | /* Uhhuh. Read error. Complain and punt. */ | |
3375 | if (!buffer_uptodate(bh)) | |
3376 | goto unlock; | |
c9c7429c MH |
3377 | if (S_ISREG(inode->i_mode) && |
3378 | ext4_encrypted_inode(inode)) { | |
3379 | /* We expect the key to be set. */ | |
3380 | BUG_ON(!ext4_has_encryption_key(inode)); | |
3381 | BUG_ON(blocksize != PAGE_CACHE_SIZE); | |
3382 | WARN_ON_ONCE(ext4_decrypt_one(inode, page)); | |
3383 | } | |
d863dc36 | 3384 | } |
d863dc36 LC |
3385 | if (ext4_should_journal_data(inode)) { |
3386 | BUFFER_TRACE(bh, "get write access"); | |
3387 | err = ext4_journal_get_write_access(handle, bh); | |
3388 | if (err) | |
3389 | goto unlock; | |
3390 | } | |
d863dc36 | 3391 | zero_user(page, offset, length); |
d863dc36 LC |
3392 | BUFFER_TRACE(bh, "zeroed end of block"); |
3393 | ||
d863dc36 LC |
3394 | if (ext4_should_journal_data(inode)) { |
3395 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
0713ed0c | 3396 | } else { |
353eefd3 | 3397 | err = 0; |
d863dc36 | 3398 | mark_buffer_dirty(bh); |
0713ed0c LC |
3399 | if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE)) |
3400 | err = ext4_jbd2_file_inode(handle, inode); | |
3401 | } | |
d863dc36 LC |
3402 | |
3403 | unlock: | |
3404 | unlock_page(page); | |
3405 | page_cache_release(page); | |
3406 | return err; | |
3407 | } | |
3408 | ||
923ae0ff RZ |
3409 | /* |
3410 | * ext4_block_zero_page_range() zeros out a mapping of length 'length' | |
3411 | * starting from file offset 'from'. The range to be zero'd must | |
3412 | * be contained with in one block. If the specified range exceeds | |
3413 | * the end of the block it will be shortened to end of the block | |
3414 | * that cooresponds to 'from' | |
3415 | */ | |
3416 | static int ext4_block_zero_page_range(handle_t *handle, | |
3417 | struct address_space *mapping, loff_t from, loff_t length) | |
3418 | { | |
3419 | struct inode *inode = mapping->host; | |
3420 | unsigned offset = from & (PAGE_CACHE_SIZE-1); | |
3421 | unsigned blocksize = inode->i_sb->s_blocksize; | |
3422 | unsigned max = blocksize - (offset & (blocksize - 1)); | |
3423 | ||
3424 | /* | |
3425 | * correct length if it does not fall between | |
3426 | * 'from' and the end of the block | |
3427 | */ | |
3428 | if (length > max || length < 0) | |
3429 | length = max; | |
3430 | ||
3431 | if (IS_DAX(inode)) | |
3432 | return dax_zero_page_range(inode, from, length, ext4_get_block); | |
3433 | return __ext4_block_zero_page_range(handle, mapping, from, length); | |
3434 | } | |
3435 | ||
94350ab5 MW |
3436 | /* |
3437 | * ext4_block_truncate_page() zeroes out a mapping from file offset `from' | |
3438 | * up to the end of the block which corresponds to `from'. | |
3439 | * This required during truncate. We need to physically zero the tail end | |
3440 | * of that block so it doesn't yield old data if the file is later grown. | |
3441 | */ | |
c197855e | 3442 | static int ext4_block_truncate_page(handle_t *handle, |
94350ab5 MW |
3443 | struct address_space *mapping, loff_t from) |
3444 | { | |
3445 | unsigned offset = from & (PAGE_CACHE_SIZE-1); | |
3446 | unsigned length; | |
3447 | unsigned blocksize; | |
3448 | struct inode *inode = mapping->host; | |
3449 | ||
3450 | blocksize = inode->i_sb->s_blocksize; | |
3451 | length = blocksize - (offset & (blocksize - 1)); | |
3452 | ||
3453 | return ext4_block_zero_page_range(handle, mapping, from, length); | |
3454 | } | |
3455 | ||
a87dd18c LC |
3456 | int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode, |
3457 | loff_t lstart, loff_t length) | |
3458 | { | |
3459 | struct super_block *sb = inode->i_sb; | |
3460 | struct address_space *mapping = inode->i_mapping; | |
e1be3a92 | 3461 | unsigned partial_start, partial_end; |
a87dd18c LC |
3462 | ext4_fsblk_t start, end; |
3463 | loff_t byte_end = (lstart + length - 1); | |
3464 | int err = 0; | |
3465 | ||
e1be3a92 LC |
3466 | partial_start = lstart & (sb->s_blocksize - 1); |
3467 | partial_end = byte_end & (sb->s_blocksize - 1); | |
3468 | ||
a87dd18c LC |
3469 | start = lstart >> sb->s_blocksize_bits; |
3470 | end = byte_end >> sb->s_blocksize_bits; | |
3471 | ||
3472 | /* Handle partial zero within the single block */ | |
e1be3a92 LC |
3473 | if (start == end && |
3474 | (partial_start || (partial_end != sb->s_blocksize - 1))) { | |
a87dd18c LC |
3475 | err = ext4_block_zero_page_range(handle, mapping, |
3476 | lstart, length); | |
3477 | return err; | |
3478 | } | |
3479 | /* Handle partial zero out on the start of the range */ | |
e1be3a92 | 3480 | if (partial_start) { |
a87dd18c LC |
3481 | err = ext4_block_zero_page_range(handle, mapping, |
3482 | lstart, sb->s_blocksize); | |
3483 | if (err) | |
3484 | return err; | |
3485 | } | |
3486 | /* Handle partial zero out on the end of the range */ | |
e1be3a92 | 3487 | if (partial_end != sb->s_blocksize - 1) |
a87dd18c | 3488 | err = ext4_block_zero_page_range(handle, mapping, |
e1be3a92 LC |
3489 | byte_end - partial_end, |
3490 | partial_end + 1); | |
a87dd18c LC |
3491 | return err; |
3492 | } | |
3493 | ||
91ef4caf DG |
3494 | int ext4_can_truncate(struct inode *inode) |
3495 | { | |
91ef4caf DG |
3496 | if (S_ISREG(inode->i_mode)) |
3497 | return 1; | |
3498 | if (S_ISDIR(inode->i_mode)) | |
3499 | return 1; | |
3500 | if (S_ISLNK(inode->i_mode)) | |
3501 | return !ext4_inode_is_fast_symlink(inode); | |
3502 | return 0; | |
3503 | } | |
3504 | ||
a4bb6b64 AH |
3505 | /* |
3506 | * ext4_punch_hole: punches a hole in a file by releaseing the blocks | |
3507 | * associated with the given offset and length | |
3508 | * | |
3509 | * @inode: File inode | |
3510 | * @offset: The offset where the hole will begin | |
3511 | * @len: The length of the hole | |
3512 | * | |
4907cb7b | 3513 | * Returns: 0 on success or negative on failure |
a4bb6b64 AH |
3514 | */ |
3515 | ||
aeb2817a | 3516 | int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length) |
a4bb6b64 | 3517 | { |
26a4c0c6 TT |
3518 | struct super_block *sb = inode->i_sb; |
3519 | ext4_lblk_t first_block, stop_block; | |
3520 | struct address_space *mapping = inode->i_mapping; | |
a87dd18c | 3521 | loff_t first_block_offset, last_block_offset; |
26a4c0c6 TT |
3522 | handle_t *handle; |
3523 | unsigned int credits; | |
3524 | int ret = 0; | |
3525 | ||
a4bb6b64 | 3526 | if (!S_ISREG(inode->i_mode)) |
73355192 | 3527 | return -EOPNOTSUPP; |
a4bb6b64 | 3528 | |
b8a86845 | 3529 | trace_ext4_punch_hole(inode, offset, length, 0); |
aaddea81 | 3530 | |
26a4c0c6 TT |
3531 | /* |
3532 | * Write out all dirty pages to avoid race conditions | |
3533 | * Then release them. | |
3534 | */ | |
3535 | if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { | |
3536 | ret = filemap_write_and_wait_range(mapping, offset, | |
3537 | offset + length - 1); | |
3538 | if (ret) | |
3539 | return ret; | |
3540 | } | |
3541 | ||
3542 | mutex_lock(&inode->i_mutex); | |
9ef06cec | 3543 | |
26a4c0c6 TT |
3544 | /* No need to punch hole beyond i_size */ |
3545 | if (offset >= inode->i_size) | |
3546 | goto out_mutex; | |
3547 | ||
3548 | /* | |
3549 | * If the hole extends beyond i_size, set the hole | |
3550 | * to end after the page that contains i_size | |
3551 | */ | |
3552 | if (offset + length > inode->i_size) { | |
3553 | length = inode->i_size + | |
3554 | PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) - | |
3555 | offset; | |
3556 | } | |
3557 | ||
a361293f JK |
3558 | if (offset & (sb->s_blocksize - 1) || |
3559 | (offset + length) & (sb->s_blocksize - 1)) { | |
3560 | /* | |
3561 | * Attach jinode to inode for jbd2 if we do any zeroing of | |
3562 | * partial block | |
3563 | */ | |
3564 | ret = ext4_inode_attach_jinode(inode); | |
3565 | if (ret < 0) | |
3566 | goto out_mutex; | |
3567 | ||
3568 | } | |
3569 | ||
a87dd18c LC |
3570 | first_block_offset = round_up(offset, sb->s_blocksize); |
3571 | last_block_offset = round_down((offset + length), sb->s_blocksize) - 1; | |
26a4c0c6 | 3572 | |
a87dd18c LC |
3573 | /* Now release the pages and zero block aligned part of pages*/ |
3574 | if (last_block_offset > first_block_offset) | |
3575 | truncate_pagecache_range(inode, first_block_offset, | |
3576 | last_block_offset); | |
26a4c0c6 TT |
3577 | |
3578 | /* Wait all existing dio workers, newcomers will block on i_mutex */ | |
3579 | ext4_inode_block_unlocked_dio(inode); | |
26a4c0c6 TT |
3580 | inode_dio_wait(inode); |
3581 | ||
3582 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
3583 | credits = ext4_writepage_trans_blocks(inode); | |
3584 | else | |
3585 | credits = ext4_blocks_for_truncate(inode); | |
3586 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); | |
3587 | if (IS_ERR(handle)) { | |
3588 | ret = PTR_ERR(handle); | |
3589 | ext4_std_error(sb, ret); | |
3590 | goto out_dio; | |
3591 | } | |
3592 | ||
a87dd18c LC |
3593 | ret = ext4_zero_partial_blocks(handle, inode, offset, |
3594 | length); | |
3595 | if (ret) | |
3596 | goto out_stop; | |
26a4c0c6 TT |
3597 | |
3598 | first_block = (offset + sb->s_blocksize - 1) >> | |
3599 | EXT4_BLOCK_SIZE_BITS(sb); | |
3600 | stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb); | |
3601 | ||
3602 | /* If there are no blocks to remove, return now */ | |
3603 | if (first_block >= stop_block) | |
3604 | goto out_stop; | |
3605 | ||
3606 | down_write(&EXT4_I(inode)->i_data_sem); | |
3607 | ext4_discard_preallocations(inode); | |
3608 | ||
3609 | ret = ext4_es_remove_extent(inode, first_block, | |
3610 | stop_block - first_block); | |
3611 | if (ret) { | |
3612 | up_write(&EXT4_I(inode)->i_data_sem); | |
3613 | goto out_stop; | |
3614 | } | |
3615 | ||
3616 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
3617 | ret = ext4_ext_remove_space(inode, first_block, | |
3618 | stop_block - 1); | |
3619 | else | |
4f579ae7 | 3620 | ret = ext4_ind_remove_space(handle, inode, first_block, |
26a4c0c6 TT |
3621 | stop_block); |
3622 | ||
819c4920 | 3623 | up_write(&EXT4_I(inode)->i_data_sem); |
26a4c0c6 TT |
3624 | if (IS_SYNC(inode)) |
3625 | ext4_handle_sync(handle); | |
e251f9bc MP |
3626 | |
3627 | /* Now release the pages again to reduce race window */ | |
3628 | if (last_block_offset > first_block_offset) | |
3629 | truncate_pagecache_range(inode, first_block_offset, | |
3630 | last_block_offset); | |
3631 | ||
26a4c0c6 TT |
3632 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); |
3633 | ext4_mark_inode_dirty(handle, inode); | |
3634 | out_stop: | |
3635 | ext4_journal_stop(handle); | |
3636 | out_dio: | |
3637 | ext4_inode_resume_unlocked_dio(inode); | |
3638 | out_mutex: | |
3639 | mutex_unlock(&inode->i_mutex); | |
3640 | return ret; | |
a4bb6b64 AH |
3641 | } |
3642 | ||
a361293f JK |
3643 | int ext4_inode_attach_jinode(struct inode *inode) |
3644 | { | |
3645 | struct ext4_inode_info *ei = EXT4_I(inode); | |
3646 | struct jbd2_inode *jinode; | |
3647 | ||
3648 | if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal) | |
3649 | return 0; | |
3650 | ||
3651 | jinode = jbd2_alloc_inode(GFP_KERNEL); | |
3652 | spin_lock(&inode->i_lock); | |
3653 | if (!ei->jinode) { | |
3654 | if (!jinode) { | |
3655 | spin_unlock(&inode->i_lock); | |
3656 | return -ENOMEM; | |
3657 | } | |
3658 | ei->jinode = jinode; | |
3659 | jbd2_journal_init_jbd_inode(ei->jinode, inode); | |
3660 | jinode = NULL; | |
3661 | } | |
3662 | spin_unlock(&inode->i_lock); | |
3663 | if (unlikely(jinode != NULL)) | |
3664 | jbd2_free_inode(jinode); | |
3665 | return 0; | |
3666 | } | |
3667 | ||
ac27a0ec | 3668 | /* |
617ba13b | 3669 | * ext4_truncate() |
ac27a0ec | 3670 | * |
617ba13b MC |
3671 | * We block out ext4_get_block() block instantiations across the entire |
3672 | * transaction, and VFS/VM ensures that ext4_truncate() cannot run | |
ac27a0ec DK |
3673 | * simultaneously on behalf of the same inode. |
3674 | * | |
42b2aa86 | 3675 | * As we work through the truncate and commit bits of it to the journal there |
ac27a0ec DK |
3676 | * is one core, guiding principle: the file's tree must always be consistent on |
3677 | * disk. We must be able to restart the truncate after a crash. | |
3678 | * | |
3679 | * The file's tree may be transiently inconsistent in memory (although it | |
3680 | * probably isn't), but whenever we close off and commit a journal transaction, | |
3681 | * the contents of (the filesystem + the journal) must be consistent and | |
3682 | * restartable. It's pretty simple, really: bottom up, right to left (although | |
3683 | * left-to-right works OK too). | |
3684 | * | |
3685 | * Note that at recovery time, journal replay occurs *before* the restart of | |
3686 | * truncate against the orphan inode list. | |
3687 | * | |
3688 | * The committed inode has the new, desired i_size (which is the same as | |
617ba13b | 3689 | * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see |
ac27a0ec | 3690 | * that this inode's truncate did not complete and it will again call |
617ba13b MC |
3691 | * ext4_truncate() to have another go. So there will be instantiated blocks |
3692 | * to the right of the truncation point in a crashed ext4 filesystem. But | |
ac27a0ec | 3693 | * that's fine - as long as they are linked from the inode, the post-crash |
617ba13b | 3694 | * ext4_truncate() run will find them and release them. |
ac27a0ec | 3695 | */ |
617ba13b | 3696 | void ext4_truncate(struct inode *inode) |
ac27a0ec | 3697 | { |
819c4920 TT |
3698 | struct ext4_inode_info *ei = EXT4_I(inode); |
3699 | unsigned int credits; | |
3700 | handle_t *handle; | |
3701 | struct address_space *mapping = inode->i_mapping; | |
819c4920 | 3702 | |
19b5ef61 TT |
3703 | /* |
3704 | * There is a possibility that we're either freeing the inode | |
e04027e8 | 3705 | * or it's a completely new inode. In those cases we might not |
19b5ef61 TT |
3706 | * have i_mutex locked because it's not necessary. |
3707 | */ | |
3708 | if (!(inode->i_state & (I_NEW|I_FREEING))) | |
3709 | WARN_ON(!mutex_is_locked(&inode->i_mutex)); | |
0562e0ba JZ |
3710 | trace_ext4_truncate_enter(inode); |
3711 | ||
91ef4caf | 3712 | if (!ext4_can_truncate(inode)) |
ac27a0ec DK |
3713 | return; |
3714 | ||
12e9b892 | 3715 | ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS); |
c8d46e41 | 3716 | |
5534fb5b | 3717 | if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC)) |
19f5fb7a | 3718 | ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE); |
7d8f9f7d | 3719 | |
aef1c851 TM |
3720 | if (ext4_has_inline_data(inode)) { |
3721 | int has_inline = 1; | |
3722 | ||
3723 | ext4_inline_data_truncate(inode, &has_inline); | |
3724 | if (has_inline) | |
3725 | return; | |
3726 | } | |
3727 | ||
a361293f JK |
3728 | /* If we zero-out tail of the page, we have to create jinode for jbd2 */ |
3729 | if (inode->i_size & (inode->i_sb->s_blocksize - 1)) { | |
3730 | if (ext4_inode_attach_jinode(inode) < 0) | |
3731 | return; | |
3732 | } | |
3733 | ||
819c4920 TT |
3734 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
3735 | credits = ext4_writepage_trans_blocks(inode); | |
3736 | else | |
3737 | credits = ext4_blocks_for_truncate(inode); | |
3738 | ||
3739 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); | |
3740 | if (IS_ERR(handle)) { | |
3741 | ext4_std_error(inode->i_sb, PTR_ERR(handle)); | |
3742 | return; | |
3743 | } | |
3744 | ||
eb3544c6 LC |
3745 | if (inode->i_size & (inode->i_sb->s_blocksize - 1)) |
3746 | ext4_block_truncate_page(handle, mapping, inode->i_size); | |
819c4920 TT |
3747 | |
3748 | /* | |
3749 | * We add the inode to the orphan list, so that if this | |
3750 | * truncate spans multiple transactions, and we crash, we will | |
3751 | * resume the truncate when the filesystem recovers. It also | |
3752 | * marks the inode dirty, to catch the new size. | |
3753 | * | |
3754 | * Implication: the file must always be in a sane, consistent | |
3755 | * truncatable state while each transaction commits. | |
3756 | */ | |
3757 | if (ext4_orphan_add(handle, inode)) | |
3758 | goto out_stop; | |
3759 | ||
3760 | down_write(&EXT4_I(inode)->i_data_sem); | |
3761 | ||
3762 | ext4_discard_preallocations(inode); | |
3763 | ||
ff9893dc | 3764 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
819c4920 | 3765 | ext4_ext_truncate(handle, inode); |
ff9893dc | 3766 | else |
819c4920 TT |
3767 | ext4_ind_truncate(handle, inode); |
3768 | ||
3769 | up_write(&ei->i_data_sem); | |
3770 | ||
3771 | if (IS_SYNC(inode)) | |
3772 | ext4_handle_sync(handle); | |
3773 | ||
3774 | out_stop: | |
3775 | /* | |
3776 | * If this was a simple ftruncate() and the file will remain alive, | |
3777 | * then we need to clear up the orphan record which we created above. | |
3778 | * However, if this was a real unlink then we were called by | |
58d86a50 | 3779 | * ext4_evict_inode(), and we allow that function to clean up the |
819c4920 TT |
3780 | * orphan info for us. |
3781 | */ | |
3782 | if (inode->i_nlink) | |
3783 | ext4_orphan_del(handle, inode); | |
3784 | ||
3785 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); | |
3786 | ext4_mark_inode_dirty(handle, inode); | |
3787 | ext4_journal_stop(handle); | |
ac27a0ec | 3788 | |
0562e0ba | 3789 | trace_ext4_truncate_exit(inode); |
ac27a0ec DK |
3790 | } |
3791 | ||
ac27a0ec | 3792 | /* |
617ba13b | 3793 | * ext4_get_inode_loc returns with an extra refcount against the inode's |
ac27a0ec DK |
3794 | * underlying buffer_head on success. If 'in_mem' is true, we have all |
3795 | * data in memory that is needed to recreate the on-disk version of this | |
3796 | * inode. | |
3797 | */ | |
617ba13b MC |
3798 | static int __ext4_get_inode_loc(struct inode *inode, |
3799 | struct ext4_iloc *iloc, int in_mem) | |
ac27a0ec | 3800 | { |
240799cd TT |
3801 | struct ext4_group_desc *gdp; |
3802 | struct buffer_head *bh; | |
3803 | struct super_block *sb = inode->i_sb; | |
3804 | ext4_fsblk_t block; | |
3805 | int inodes_per_block, inode_offset; | |
3806 | ||
3a06d778 | 3807 | iloc->bh = NULL; |
240799cd TT |
3808 | if (!ext4_valid_inum(sb, inode->i_ino)) |
3809 | return -EIO; | |
ac27a0ec | 3810 | |
240799cd TT |
3811 | iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb); |
3812 | gdp = ext4_get_group_desc(sb, iloc->block_group, NULL); | |
3813 | if (!gdp) | |
ac27a0ec DK |
3814 | return -EIO; |
3815 | ||
240799cd TT |
3816 | /* |
3817 | * Figure out the offset within the block group inode table | |
3818 | */ | |
00d09882 | 3819 | inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; |
240799cd TT |
3820 | inode_offset = ((inode->i_ino - 1) % |
3821 | EXT4_INODES_PER_GROUP(sb)); | |
3822 | block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block); | |
3823 | iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb); | |
3824 | ||
3825 | bh = sb_getblk(sb, block); | |
aebf0243 | 3826 | if (unlikely(!bh)) |
860d21e2 | 3827 | return -ENOMEM; |
ac27a0ec DK |
3828 | if (!buffer_uptodate(bh)) { |
3829 | lock_buffer(bh); | |
9c83a923 HK |
3830 | |
3831 | /* | |
3832 | * If the buffer has the write error flag, we have failed | |
3833 | * to write out another inode in the same block. In this | |
3834 | * case, we don't have to read the block because we may | |
3835 | * read the old inode data successfully. | |
3836 | */ | |
3837 | if (buffer_write_io_error(bh) && !buffer_uptodate(bh)) | |
3838 | set_buffer_uptodate(bh); | |
3839 | ||
ac27a0ec DK |
3840 | if (buffer_uptodate(bh)) { |
3841 | /* someone brought it uptodate while we waited */ | |
3842 | unlock_buffer(bh); | |
3843 | goto has_buffer; | |
3844 | } | |
3845 | ||
3846 | /* | |
3847 | * If we have all information of the inode in memory and this | |
3848 | * is the only valid inode in the block, we need not read the | |
3849 | * block. | |
3850 | */ | |
3851 | if (in_mem) { | |
3852 | struct buffer_head *bitmap_bh; | |
240799cd | 3853 | int i, start; |
ac27a0ec | 3854 | |
240799cd | 3855 | start = inode_offset & ~(inodes_per_block - 1); |
ac27a0ec | 3856 | |
240799cd TT |
3857 | /* Is the inode bitmap in cache? */ |
3858 | bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp)); | |
aebf0243 | 3859 | if (unlikely(!bitmap_bh)) |
ac27a0ec DK |
3860 | goto make_io; |
3861 | ||
3862 | /* | |
3863 | * If the inode bitmap isn't in cache then the | |
3864 | * optimisation may end up performing two reads instead | |
3865 | * of one, so skip it. | |
3866 | */ | |
3867 | if (!buffer_uptodate(bitmap_bh)) { | |
3868 | brelse(bitmap_bh); | |
3869 | goto make_io; | |
3870 | } | |
240799cd | 3871 | for (i = start; i < start + inodes_per_block; i++) { |
ac27a0ec DK |
3872 | if (i == inode_offset) |
3873 | continue; | |
617ba13b | 3874 | if (ext4_test_bit(i, bitmap_bh->b_data)) |
ac27a0ec DK |
3875 | break; |
3876 | } | |
3877 | brelse(bitmap_bh); | |
240799cd | 3878 | if (i == start + inodes_per_block) { |
ac27a0ec DK |
3879 | /* all other inodes are free, so skip I/O */ |
3880 | memset(bh->b_data, 0, bh->b_size); | |
3881 | set_buffer_uptodate(bh); | |
3882 | unlock_buffer(bh); | |
3883 | goto has_buffer; | |
3884 | } | |
3885 | } | |
3886 | ||
3887 | make_io: | |
240799cd TT |
3888 | /* |
3889 | * If we need to do any I/O, try to pre-readahead extra | |
3890 | * blocks from the inode table. | |
3891 | */ | |
3892 | if (EXT4_SB(sb)->s_inode_readahead_blks) { | |
3893 | ext4_fsblk_t b, end, table; | |
3894 | unsigned num; | |
0d606e2c | 3895 | __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks; |
240799cd TT |
3896 | |
3897 | table = ext4_inode_table(sb, gdp); | |
b713a5ec | 3898 | /* s_inode_readahead_blks is always a power of 2 */ |
0d606e2c | 3899 | b = block & ~((ext4_fsblk_t) ra_blks - 1); |
240799cd TT |
3900 | if (table > b) |
3901 | b = table; | |
0d606e2c | 3902 | end = b + ra_blks; |
240799cd | 3903 | num = EXT4_INODES_PER_GROUP(sb); |
feb0ab32 | 3904 | if (ext4_has_group_desc_csum(sb)) |
560671a0 | 3905 | num -= ext4_itable_unused_count(sb, gdp); |
240799cd TT |
3906 | table += num / inodes_per_block; |
3907 | if (end > table) | |
3908 | end = table; | |
3909 | while (b <= end) | |
3910 | sb_breadahead(sb, b++); | |
3911 | } | |
3912 | ||
ac27a0ec DK |
3913 | /* |
3914 | * There are other valid inodes in the buffer, this inode | |
3915 | * has in-inode xattrs, or we don't have this inode in memory. | |
3916 | * Read the block from disk. | |
3917 | */ | |
0562e0ba | 3918 | trace_ext4_load_inode(inode); |
ac27a0ec DK |
3919 | get_bh(bh); |
3920 | bh->b_end_io = end_buffer_read_sync; | |
65299a3b | 3921 | submit_bh(READ | REQ_META | REQ_PRIO, bh); |
ac27a0ec DK |
3922 | wait_on_buffer(bh); |
3923 | if (!buffer_uptodate(bh)) { | |
c398eda0 TT |
3924 | EXT4_ERROR_INODE_BLOCK(inode, block, |
3925 | "unable to read itable block"); | |
ac27a0ec DK |
3926 | brelse(bh); |
3927 | return -EIO; | |
3928 | } | |
3929 | } | |
3930 | has_buffer: | |
3931 | iloc->bh = bh; | |
3932 | return 0; | |
3933 | } | |
3934 | ||
617ba13b | 3935 | int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec DK |
3936 | { |
3937 | /* We have all inode data except xattrs in memory here. */ | |
617ba13b | 3938 | return __ext4_get_inode_loc(inode, iloc, |
19f5fb7a | 3939 | !ext4_test_inode_state(inode, EXT4_STATE_XATTR)); |
ac27a0ec DK |
3940 | } |
3941 | ||
617ba13b | 3942 | void ext4_set_inode_flags(struct inode *inode) |
ac27a0ec | 3943 | { |
617ba13b | 3944 | unsigned int flags = EXT4_I(inode)->i_flags; |
00a1a053 | 3945 | unsigned int new_fl = 0; |
ac27a0ec | 3946 | |
617ba13b | 3947 | if (flags & EXT4_SYNC_FL) |
00a1a053 | 3948 | new_fl |= S_SYNC; |
617ba13b | 3949 | if (flags & EXT4_APPEND_FL) |
00a1a053 | 3950 | new_fl |= S_APPEND; |
617ba13b | 3951 | if (flags & EXT4_IMMUTABLE_FL) |
00a1a053 | 3952 | new_fl |= S_IMMUTABLE; |
617ba13b | 3953 | if (flags & EXT4_NOATIME_FL) |
00a1a053 | 3954 | new_fl |= S_NOATIME; |
617ba13b | 3955 | if (flags & EXT4_DIRSYNC_FL) |
00a1a053 | 3956 | new_fl |= S_DIRSYNC; |
923ae0ff RZ |
3957 | if (test_opt(inode->i_sb, DAX)) |
3958 | new_fl |= S_DAX; | |
5f16f322 | 3959 | inode_set_flags(inode, new_fl, |
923ae0ff | 3960 | S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX); |
ac27a0ec DK |
3961 | } |
3962 | ||
ff9ddf7e JK |
3963 | /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */ |
3964 | void ext4_get_inode_flags(struct ext4_inode_info *ei) | |
3965 | { | |
84a8dce2 DM |
3966 | unsigned int vfs_fl; |
3967 | unsigned long old_fl, new_fl; | |
3968 | ||
3969 | do { | |
3970 | vfs_fl = ei->vfs_inode.i_flags; | |
3971 | old_fl = ei->i_flags; | |
3972 | new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL| | |
3973 | EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL| | |
3974 | EXT4_DIRSYNC_FL); | |
3975 | if (vfs_fl & S_SYNC) | |
3976 | new_fl |= EXT4_SYNC_FL; | |
3977 | if (vfs_fl & S_APPEND) | |
3978 | new_fl |= EXT4_APPEND_FL; | |
3979 | if (vfs_fl & S_IMMUTABLE) | |
3980 | new_fl |= EXT4_IMMUTABLE_FL; | |
3981 | if (vfs_fl & S_NOATIME) | |
3982 | new_fl |= EXT4_NOATIME_FL; | |
3983 | if (vfs_fl & S_DIRSYNC) | |
3984 | new_fl |= EXT4_DIRSYNC_FL; | |
3985 | } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl); | |
ff9ddf7e | 3986 | } |
de9a55b8 | 3987 | |
0fc1b451 | 3988 | static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode, |
de9a55b8 | 3989 | struct ext4_inode_info *ei) |
0fc1b451 AK |
3990 | { |
3991 | blkcnt_t i_blocks ; | |
8180a562 AK |
3992 | struct inode *inode = &(ei->vfs_inode); |
3993 | struct super_block *sb = inode->i_sb; | |
0fc1b451 AK |
3994 | |
3995 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, | |
3996 | EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) { | |
3997 | /* we are using combined 48 bit field */ | |
3998 | i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 | | |
3999 | le32_to_cpu(raw_inode->i_blocks_lo); | |
07a03824 | 4000 | if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) { |
8180a562 AK |
4001 | /* i_blocks represent file system block size */ |
4002 | return i_blocks << (inode->i_blkbits - 9); | |
4003 | } else { | |
4004 | return i_blocks; | |
4005 | } | |
0fc1b451 AK |
4006 | } else { |
4007 | return le32_to_cpu(raw_inode->i_blocks_lo); | |
4008 | } | |
4009 | } | |
ff9ddf7e | 4010 | |
152a7b0a TM |
4011 | static inline void ext4_iget_extra_inode(struct inode *inode, |
4012 | struct ext4_inode *raw_inode, | |
4013 | struct ext4_inode_info *ei) | |
4014 | { | |
4015 | __le32 *magic = (void *)raw_inode + | |
4016 | EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize; | |
67cf5b09 | 4017 | if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) { |
152a7b0a | 4018 | ext4_set_inode_state(inode, EXT4_STATE_XATTR); |
67cf5b09 | 4019 | ext4_find_inline_data_nolock(inode); |
f19d5870 TM |
4020 | } else |
4021 | EXT4_I(inode)->i_inline_off = 0; | |
152a7b0a TM |
4022 | } |
4023 | ||
1d1fe1ee | 4024 | struct inode *ext4_iget(struct super_block *sb, unsigned long ino) |
ac27a0ec | 4025 | { |
617ba13b MC |
4026 | struct ext4_iloc iloc; |
4027 | struct ext4_inode *raw_inode; | |
1d1fe1ee | 4028 | struct ext4_inode_info *ei; |
1d1fe1ee | 4029 | struct inode *inode; |
b436b9be | 4030 | journal_t *journal = EXT4_SB(sb)->s_journal; |
1d1fe1ee | 4031 | long ret; |
ac27a0ec | 4032 | int block; |
08cefc7a EB |
4033 | uid_t i_uid; |
4034 | gid_t i_gid; | |
ac27a0ec | 4035 | |
1d1fe1ee DH |
4036 | inode = iget_locked(sb, ino); |
4037 | if (!inode) | |
4038 | return ERR_PTR(-ENOMEM); | |
4039 | if (!(inode->i_state & I_NEW)) | |
4040 | return inode; | |
4041 | ||
4042 | ei = EXT4_I(inode); | |
7dc57615 | 4043 | iloc.bh = NULL; |
ac27a0ec | 4044 | |
1d1fe1ee DH |
4045 | ret = __ext4_get_inode_loc(inode, &iloc, 0); |
4046 | if (ret < 0) | |
ac27a0ec | 4047 | goto bad_inode; |
617ba13b | 4048 | raw_inode = ext4_raw_inode(&iloc); |
814525f4 DW |
4049 | |
4050 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { | |
4051 | ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize); | |
4052 | if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > | |
4053 | EXT4_INODE_SIZE(inode->i_sb)) { | |
4054 | EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)", | |
4055 | EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize, | |
4056 | EXT4_INODE_SIZE(inode->i_sb)); | |
4057 | ret = -EIO; | |
4058 | goto bad_inode; | |
4059 | } | |
4060 | } else | |
4061 | ei->i_extra_isize = 0; | |
4062 | ||
4063 | /* Precompute checksum seed for inode metadata */ | |
9aa5d32b | 4064 | if (ext4_has_metadata_csum(sb)) { |
814525f4 DW |
4065 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
4066 | __u32 csum; | |
4067 | __le32 inum = cpu_to_le32(inode->i_ino); | |
4068 | __le32 gen = raw_inode->i_generation; | |
4069 | csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum, | |
4070 | sizeof(inum)); | |
4071 | ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen, | |
4072 | sizeof(gen)); | |
4073 | } | |
4074 | ||
4075 | if (!ext4_inode_csum_verify(inode, raw_inode, ei)) { | |
4076 | EXT4_ERROR_INODE(inode, "checksum invalid"); | |
4077 | ret = -EIO; | |
4078 | goto bad_inode; | |
4079 | } | |
4080 | ||
ac27a0ec | 4081 | inode->i_mode = le16_to_cpu(raw_inode->i_mode); |
08cefc7a EB |
4082 | i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low); |
4083 | i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low); | |
af5bc92d | 4084 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
08cefc7a EB |
4085 | i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16; |
4086 | i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16; | |
ac27a0ec | 4087 | } |
08cefc7a EB |
4088 | i_uid_write(inode, i_uid); |
4089 | i_gid_write(inode, i_gid); | |
bfe86848 | 4090 | set_nlink(inode, le16_to_cpu(raw_inode->i_links_count)); |
ac27a0ec | 4091 | |
353eb83c | 4092 | ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */ |
67cf5b09 | 4093 | ei->i_inline_off = 0; |
ac27a0ec DK |
4094 | ei->i_dir_start_lookup = 0; |
4095 | ei->i_dtime = le32_to_cpu(raw_inode->i_dtime); | |
4096 | /* We now have enough fields to check if the inode was active or not. | |
4097 | * This is needed because nfsd might try to access dead inodes | |
4098 | * the test is that same one that e2fsck uses | |
4099 | * NeilBrown 1999oct15 | |
4100 | */ | |
4101 | if (inode->i_nlink == 0) { | |
393d1d1d DTB |
4102 | if ((inode->i_mode == 0 || |
4103 | !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) && | |
4104 | ino != EXT4_BOOT_LOADER_INO) { | |
ac27a0ec | 4105 | /* this inode is deleted */ |
1d1fe1ee | 4106 | ret = -ESTALE; |
ac27a0ec DK |
4107 | goto bad_inode; |
4108 | } | |
4109 | /* The only unlinked inodes we let through here have | |
4110 | * valid i_mode and are being read by the orphan | |
4111 | * recovery code: that's fine, we're about to complete | |
393d1d1d DTB |
4112 | * the process of deleting those. |
4113 | * OR it is the EXT4_BOOT_LOADER_INO which is | |
4114 | * not initialized on a new filesystem. */ | |
ac27a0ec | 4115 | } |
ac27a0ec | 4116 | ei->i_flags = le32_to_cpu(raw_inode->i_flags); |
0fc1b451 | 4117 | inode->i_blocks = ext4_inode_blocks(raw_inode, ei); |
7973c0c1 | 4118 | ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo); |
a9e81742 | 4119 | if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) |
a1ddeb7e BP |
4120 | ei->i_file_acl |= |
4121 | ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32; | |
a48380f7 | 4122 | inode->i_size = ext4_isize(raw_inode); |
ac27a0ec | 4123 | ei->i_disksize = inode->i_size; |
a9e7f447 DM |
4124 | #ifdef CONFIG_QUOTA |
4125 | ei->i_reserved_quota = 0; | |
4126 | #endif | |
ac27a0ec DK |
4127 | inode->i_generation = le32_to_cpu(raw_inode->i_generation); |
4128 | ei->i_block_group = iloc.block_group; | |
a4912123 | 4129 | ei->i_last_alloc_group = ~0; |
ac27a0ec DK |
4130 | /* |
4131 | * NOTE! The in-memory inode i_data array is in little-endian order | |
4132 | * even on big-endian machines: we do NOT byteswap the block numbers! | |
4133 | */ | |
617ba13b | 4134 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
ac27a0ec DK |
4135 | ei->i_data[block] = raw_inode->i_block[block]; |
4136 | INIT_LIST_HEAD(&ei->i_orphan); | |
4137 | ||
b436b9be JK |
4138 | /* |
4139 | * Set transaction id's of transactions that have to be committed | |
4140 | * to finish f[data]sync. We set them to currently running transaction | |
4141 | * as we cannot be sure that the inode or some of its metadata isn't | |
4142 | * part of the transaction - the inode could have been reclaimed and | |
4143 | * now it is reread from disk. | |
4144 | */ | |
4145 | if (journal) { | |
4146 | transaction_t *transaction; | |
4147 | tid_t tid; | |
4148 | ||
a931da6a | 4149 | read_lock(&journal->j_state_lock); |
b436b9be JK |
4150 | if (journal->j_running_transaction) |
4151 | transaction = journal->j_running_transaction; | |
4152 | else | |
4153 | transaction = journal->j_committing_transaction; | |
4154 | if (transaction) | |
4155 | tid = transaction->t_tid; | |
4156 | else | |
4157 | tid = journal->j_commit_sequence; | |
a931da6a | 4158 | read_unlock(&journal->j_state_lock); |
b436b9be JK |
4159 | ei->i_sync_tid = tid; |
4160 | ei->i_datasync_tid = tid; | |
4161 | } | |
4162 | ||
0040d987 | 4163 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
ac27a0ec DK |
4164 | if (ei->i_extra_isize == 0) { |
4165 | /* The extra space is currently unused. Use it. */ | |
617ba13b MC |
4166 | ei->i_extra_isize = sizeof(struct ext4_inode) - |
4167 | EXT4_GOOD_OLD_INODE_SIZE; | |
ac27a0ec | 4168 | } else { |
152a7b0a | 4169 | ext4_iget_extra_inode(inode, raw_inode, ei); |
ac27a0ec | 4170 | } |
814525f4 | 4171 | } |
ac27a0ec | 4172 | |
ef7f3835 KS |
4173 | EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode); |
4174 | EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode); | |
4175 | EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode); | |
4176 | EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode); | |
4177 | ||
ed3654eb | 4178 | if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) { |
c4f65706 TT |
4179 | inode->i_version = le32_to_cpu(raw_inode->i_disk_version); |
4180 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { | |
4181 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
4182 | inode->i_version |= | |
4183 | (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32; | |
4184 | } | |
25ec56b5 JNC |
4185 | } |
4186 | ||
c4b5a614 | 4187 | ret = 0; |
485c26ec | 4188 | if (ei->i_file_acl && |
1032988c | 4189 | !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) { |
24676da4 TT |
4190 | EXT4_ERROR_INODE(inode, "bad extended attribute block %llu", |
4191 | ei->i_file_acl); | |
485c26ec TT |
4192 | ret = -EIO; |
4193 | goto bad_inode; | |
f19d5870 TM |
4194 | } else if (!ext4_has_inline_data(inode)) { |
4195 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { | |
4196 | if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || | |
4197 | (S_ISLNK(inode->i_mode) && | |
4198 | !ext4_inode_is_fast_symlink(inode)))) | |
4199 | /* Validate extent which is part of inode */ | |
4200 | ret = ext4_ext_check_inode(inode); | |
4201 | } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || | |
4202 | (S_ISLNK(inode->i_mode) && | |
4203 | !ext4_inode_is_fast_symlink(inode))) { | |
4204 | /* Validate block references which are part of inode */ | |
4205 | ret = ext4_ind_check_inode(inode); | |
4206 | } | |
fe2c8191 | 4207 | } |
567f3e9a | 4208 | if (ret) |
de9a55b8 | 4209 | goto bad_inode; |
7a262f7c | 4210 | |
ac27a0ec | 4211 | if (S_ISREG(inode->i_mode)) { |
617ba13b | 4212 | inode->i_op = &ext4_file_inode_operations; |
be64f884 | 4213 | inode->i_fop = &ext4_file_operations; |
617ba13b | 4214 | ext4_set_aops(inode); |
ac27a0ec | 4215 | } else if (S_ISDIR(inode->i_mode)) { |
617ba13b MC |
4216 | inode->i_op = &ext4_dir_inode_operations; |
4217 | inode->i_fop = &ext4_dir_operations; | |
ac27a0ec | 4218 | } else if (S_ISLNK(inode->i_mode)) { |
f348c252 TT |
4219 | if (ext4_inode_is_fast_symlink(inode) && |
4220 | !ext4_encrypted_inode(inode)) { | |
617ba13b | 4221 | inode->i_op = &ext4_fast_symlink_inode_operations; |
e83c1397 DG |
4222 | nd_terminate_link(ei->i_data, inode->i_size, |
4223 | sizeof(ei->i_data) - 1); | |
4224 | } else { | |
617ba13b MC |
4225 | inode->i_op = &ext4_symlink_inode_operations; |
4226 | ext4_set_aops(inode); | |
ac27a0ec | 4227 | } |
563bdd61 TT |
4228 | } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) || |
4229 | S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) { | |
617ba13b | 4230 | inode->i_op = &ext4_special_inode_operations; |
ac27a0ec DK |
4231 | if (raw_inode->i_block[0]) |
4232 | init_special_inode(inode, inode->i_mode, | |
4233 | old_decode_dev(le32_to_cpu(raw_inode->i_block[0]))); | |
4234 | else | |
4235 | init_special_inode(inode, inode->i_mode, | |
4236 | new_decode_dev(le32_to_cpu(raw_inode->i_block[1]))); | |
393d1d1d DTB |
4237 | } else if (ino == EXT4_BOOT_LOADER_INO) { |
4238 | make_bad_inode(inode); | |
563bdd61 | 4239 | } else { |
563bdd61 | 4240 | ret = -EIO; |
24676da4 | 4241 | EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode); |
563bdd61 | 4242 | goto bad_inode; |
ac27a0ec | 4243 | } |
af5bc92d | 4244 | brelse(iloc.bh); |
617ba13b | 4245 | ext4_set_inode_flags(inode); |
1d1fe1ee DH |
4246 | unlock_new_inode(inode); |
4247 | return inode; | |
ac27a0ec DK |
4248 | |
4249 | bad_inode: | |
567f3e9a | 4250 | brelse(iloc.bh); |
1d1fe1ee DH |
4251 | iget_failed(inode); |
4252 | return ERR_PTR(ret); | |
ac27a0ec DK |
4253 | } |
4254 | ||
f4bb2981 TT |
4255 | struct inode *ext4_iget_normal(struct super_block *sb, unsigned long ino) |
4256 | { | |
4257 | if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO) | |
4258 | return ERR_PTR(-EIO); | |
4259 | return ext4_iget(sb, ino); | |
4260 | } | |
4261 | ||
0fc1b451 AK |
4262 | static int ext4_inode_blocks_set(handle_t *handle, |
4263 | struct ext4_inode *raw_inode, | |
4264 | struct ext4_inode_info *ei) | |
4265 | { | |
4266 | struct inode *inode = &(ei->vfs_inode); | |
4267 | u64 i_blocks = inode->i_blocks; | |
4268 | struct super_block *sb = inode->i_sb; | |
0fc1b451 AK |
4269 | |
4270 | if (i_blocks <= ~0U) { | |
4271 | /* | |
4907cb7b | 4272 | * i_blocks can be represented in a 32 bit variable |
0fc1b451 AK |
4273 | * as multiple of 512 bytes |
4274 | */ | |
8180a562 | 4275 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 4276 | raw_inode->i_blocks_high = 0; |
84a8dce2 | 4277 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
f287a1a5 TT |
4278 | return 0; |
4279 | } | |
4280 | if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) | |
4281 | return -EFBIG; | |
4282 | ||
4283 | if (i_blocks <= 0xffffffffffffULL) { | |
0fc1b451 AK |
4284 | /* |
4285 | * i_blocks can be represented in a 48 bit variable | |
4286 | * as multiple of 512 bytes | |
4287 | */ | |
8180a562 | 4288 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 4289 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); |
84a8dce2 | 4290 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
0fc1b451 | 4291 | } else { |
84a8dce2 | 4292 | ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
8180a562 AK |
4293 | /* i_block is stored in file system block size */ |
4294 | i_blocks = i_blocks >> (inode->i_blkbits - 9); | |
4295 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); | |
4296 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); | |
0fc1b451 | 4297 | } |
f287a1a5 | 4298 | return 0; |
0fc1b451 AK |
4299 | } |
4300 | ||
a26f4992 TT |
4301 | struct other_inode { |
4302 | unsigned long orig_ino; | |
4303 | struct ext4_inode *raw_inode; | |
4304 | }; | |
4305 | ||
4306 | static int other_inode_match(struct inode * inode, unsigned long ino, | |
4307 | void *data) | |
4308 | { | |
4309 | struct other_inode *oi = (struct other_inode *) data; | |
4310 | ||
4311 | if ((inode->i_ino != ino) || | |
4312 | (inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW | | |
4313 | I_DIRTY_SYNC | I_DIRTY_DATASYNC)) || | |
4314 | ((inode->i_state & I_DIRTY_TIME) == 0)) | |
4315 | return 0; | |
4316 | spin_lock(&inode->i_lock); | |
4317 | if (((inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW | | |
4318 | I_DIRTY_SYNC | I_DIRTY_DATASYNC)) == 0) && | |
4319 | (inode->i_state & I_DIRTY_TIME)) { | |
4320 | struct ext4_inode_info *ei = EXT4_I(inode); | |
4321 | ||
4322 | inode->i_state &= ~(I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED); | |
4323 | spin_unlock(&inode->i_lock); | |
4324 | ||
4325 | spin_lock(&ei->i_raw_lock); | |
4326 | EXT4_INODE_SET_XTIME(i_ctime, inode, oi->raw_inode); | |
4327 | EXT4_INODE_SET_XTIME(i_mtime, inode, oi->raw_inode); | |
4328 | EXT4_INODE_SET_XTIME(i_atime, inode, oi->raw_inode); | |
4329 | ext4_inode_csum_set(inode, oi->raw_inode, ei); | |
4330 | spin_unlock(&ei->i_raw_lock); | |
4331 | trace_ext4_other_inode_update_time(inode, oi->orig_ino); | |
4332 | return -1; | |
4333 | } | |
4334 | spin_unlock(&inode->i_lock); | |
4335 | return -1; | |
4336 | } | |
4337 | ||
4338 | /* | |
4339 | * Opportunistically update the other time fields for other inodes in | |
4340 | * the same inode table block. | |
4341 | */ | |
4342 | static void ext4_update_other_inodes_time(struct super_block *sb, | |
4343 | unsigned long orig_ino, char *buf) | |
4344 | { | |
4345 | struct other_inode oi; | |
4346 | unsigned long ino; | |
4347 | int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; | |
4348 | int inode_size = EXT4_INODE_SIZE(sb); | |
4349 | ||
4350 | oi.orig_ino = orig_ino; | |
8f4d8558 | 4351 | ino = (orig_ino & ~(inodes_per_block - 1)) + 1; |
a26f4992 TT |
4352 | for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) { |
4353 | if (ino == orig_ino) | |
4354 | continue; | |
4355 | oi.raw_inode = (struct ext4_inode *) buf; | |
4356 | (void) find_inode_nowait(sb, ino, other_inode_match, &oi); | |
4357 | } | |
4358 | } | |
4359 | ||
ac27a0ec DK |
4360 | /* |
4361 | * Post the struct inode info into an on-disk inode location in the | |
4362 | * buffer-cache. This gobbles the caller's reference to the | |
4363 | * buffer_head in the inode location struct. | |
4364 | * | |
4365 | * The caller must have write access to iloc->bh. | |
4366 | */ | |
617ba13b | 4367 | static int ext4_do_update_inode(handle_t *handle, |
ac27a0ec | 4368 | struct inode *inode, |
830156c7 | 4369 | struct ext4_iloc *iloc) |
ac27a0ec | 4370 | { |
617ba13b MC |
4371 | struct ext4_inode *raw_inode = ext4_raw_inode(iloc); |
4372 | struct ext4_inode_info *ei = EXT4_I(inode); | |
ac27a0ec | 4373 | struct buffer_head *bh = iloc->bh; |
202ee5df | 4374 | struct super_block *sb = inode->i_sb; |
ac27a0ec | 4375 | int err = 0, rc, block; |
202ee5df | 4376 | int need_datasync = 0, set_large_file = 0; |
08cefc7a EB |
4377 | uid_t i_uid; |
4378 | gid_t i_gid; | |
ac27a0ec | 4379 | |
202ee5df TT |
4380 | spin_lock(&ei->i_raw_lock); |
4381 | ||
4382 | /* For fields not tracked in the in-memory inode, | |
ac27a0ec | 4383 | * initialise them to zero for new inodes. */ |
19f5fb7a | 4384 | if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) |
617ba13b | 4385 | memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size); |
ac27a0ec | 4386 | |
ff9ddf7e | 4387 | ext4_get_inode_flags(ei); |
ac27a0ec | 4388 | raw_inode->i_mode = cpu_to_le16(inode->i_mode); |
08cefc7a EB |
4389 | i_uid = i_uid_read(inode); |
4390 | i_gid = i_gid_read(inode); | |
af5bc92d | 4391 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
08cefc7a EB |
4392 | raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid)); |
4393 | raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid)); | |
ac27a0ec DK |
4394 | /* |
4395 | * Fix up interoperability with old kernels. Otherwise, old inodes get | |
4396 | * re-used with the upper 16 bits of the uid/gid intact | |
4397 | */ | |
af5bc92d | 4398 | if (!ei->i_dtime) { |
ac27a0ec | 4399 | raw_inode->i_uid_high = |
08cefc7a | 4400 | cpu_to_le16(high_16_bits(i_uid)); |
ac27a0ec | 4401 | raw_inode->i_gid_high = |
08cefc7a | 4402 | cpu_to_le16(high_16_bits(i_gid)); |
ac27a0ec DK |
4403 | } else { |
4404 | raw_inode->i_uid_high = 0; | |
4405 | raw_inode->i_gid_high = 0; | |
4406 | } | |
4407 | } else { | |
08cefc7a EB |
4408 | raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid)); |
4409 | raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid)); | |
ac27a0ec DK |
4410 | raw_inode->i_uid_high = 0; |
4411 | raw_inode->i_gid_high = 0; | |
4412 | } | |
4413 | raw_inode->i_links_count = cpu_to_le16(inode->i_nlink); | |
ef7f3835 KS |
4414 | |
4415 | EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode); | |
4416 | EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode); | |
4417 | EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode); | |
4418 | EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode); | |
4419 | ||
bce92d56 LX |
4420 | err = ext4_inode_blocks_set(handle, raw_inode, ei); |
4421 | if (err) { | |
202ee5df | 4422 | spin_unlock(&ei->i_raw_lock); |
0fc1b451 | 4423 | goto out_brelse; |
202ee5df | 4424 | } |
ac27a0ec | 4425 | raw_inode->i_dtime = cpu_to_le32(ei->i_dtime); |
353eb83c | 4426 | raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF); |
ed3654eb | 4427 | if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) |
a1ddeb7e BP |
4428 | raw_inode->i_file_acl_high = |
4429 | cpu_to_le16(ei->i_file_acl >> 32); | |
7973c0c1 | 4430 | raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl); |
b71fc079 JK |
4431 | if (ei->i_disksize != ext4_isize(raw_inode)) { |
4432 | ext4_isize_set(raw_inode, ei->i_disksize); | |
4433 | need_datasync = 1; | |
4434 | } | |
a48380f7 | 4435 | if (ei->i_disksize > 0x7fffffffULL) { |
a48380f7 AK |
4436 | if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, |
4437 | EXT4_FEATURE_RO_COMPAT_LARGE_FILE) || | |
4438 | EXT4_SB(sb)->s_es->s_rev_level == | |
202ee5df TT |
4439 | cpu_to_le32(EXT4_GOOD_OLD_REV)) |
4440 | set_large_file = 1; | |
ac27a0ec DK |
4441 | } |
4442 | raw_inode->i_generation = cpu_to_le32(inode->i_generation); | |
4443 | if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) { | |
4444 | if (old_valid_dev(inode->i_rdev)) { | |
4445 | raw_inode->i_block[0] = | |
4446 | cpu_to_le32(old_encode_dev(inode->i_rdev)); | |
4447 | raw_inode->i_block[1] = 0; | |
4448 | } else { | |
4449 | raw_inode->i_block[0] = 0; | |
4450 | raw_inode->i_block[1] = | |
4451 | cpu_to_le32(new_encode_dev(inode->i_rdev)); | |
4452 | raw_inode->i_block[2] = 0; | |
4453 | } | |
f19d5870 | 4454 | } else if (!ext4_has_inline_data(inode)) { |
de9a55b8 TT |
4455 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
4456 | raw_inode->i_block[block] = ei->i_data[block]; | |
f19d5870 | 4457 | } |
ac27a0ec | 4458 | |
ed3654eb | 4459 | if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) { |
c4f65706 TT |
4460 | raw_inode->i_disk_version = cpu_to_le32(inode->i_version); |
4461 | if (ei->i_extra_isize) { | |
4462 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
4463 | raw_inode->i_version_hi = | |
4464 | cpu_to_le32(inode->i_version >> 32); | |
4465 | raw_inode->i_extra_isize = | |
4466 | cpu_to_le16(ei->i_extra_isize); | |
4467 | } | |
25ec56b5 | 4468 | } |
814525f4 | 4469 | ext4_inode_csum_set(inode, raw_inode, ei); |
202ee5df | 4470 | spin_unlock(&ei->i_raw_lock); |
a26f4992 TT |
4471 | if (inode->i_sb->s_flags & MS_LAZYTIME) |
4472 | ext4_update_other_inodes_time(inode->i_sb, inode->i_ino, | |
4473 | bh->b_data); | |
202ee5df | 4474 | |
830156c7 | 4475 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
73b50c1c | 4476 | rc = ext4_handle_dirty_metadata(handle, NULL, bh); |
830156c7 FM |
4477 | if (!err) |
4478 | err = rc; | |
19f5fb7a | 4479 | ext4_clear_inode_state(inode, EXT4_STATE_NEW); |
202ee5df | 4480 | if (set_large_file) { |
5d601255 | 4481 | BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access"); |
202ee5df TT |
4482 | err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh); |
4483 | if (err) | |
4484 | goto out_brelse; | |
4485 | ext4_update_dynamic_rev(sb); | |
4486 | EXT4_SET_RO_COMPAT_FEATURE(sb, | |
4487 | EXT4_FEATURE_RO_COMPAT_LARGE_FILE); | |
4488 | ext4_handle_sync(handle); | |
4489 | err = ext4_handle_dirty_super(handle, sb); | |
4490 | } | |
b71fc079 | 4491 | ext4_update_inode_fsync_trans(handle, inode, need_datasync); |
ac27a0ec | 4492 | out_brelse: |
af5bc92d | 4493 | brelse(bh); |
617ba13b | 4494 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
4495 | return err; |
4496 | } | |
4497 | ||
4498 | /* | |
617ba13b | 4499 | * ext4_write_inode() |
ac27a0ec DK |
4500 | * |
4501 | * We are called from a few places: | |
4502 | * | |
87f7e416 | 4503 | * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files. |
ac27a0ec | 4504 | * Here, there will be no transaction running. We wait for any running |
4907cb7b | 4505 | * transaction to commit. |
ac27a0ec | 4506 | * |
87f7e416 TT |
4507 | * - Within flush work (sys_sync(), kupdate and such). |
4508 | * We wait on commit, if told to. | |
ac27a0ec | 4509 | * |
87f7e416 TT |
4510 | * - Within iput_final() -> write_inode_now() |
4511 | * We wait on commit, if told to. | |
ac27a0ec DK |
4512 | * |
4513 | * In all cases it is actually safe for us to return without doing anything, | |
4514 | * because the inode has been copied into a raw inode buffer in | |
87f7e416 TT |
4515 | * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL |
4516 | * writeback. | |
ac27a0ec DK |
4517 | * |
4518 | * Note that we are absolutely dependent upon all inode dirtiers doing the | |
4519 | * right thing: they *must* call mark_inode_dirty() after dirtying info in | |
4520 | * which we are interested. | |
4521 | * | |
4522 | * It would be a bug for them to not do this. The code: | |
4523 | * | |
4524 | * mark_inode_dirty(inode) | |
4525 | * stuff(); | |
4526 | * inode->i_size = expr; | |
4527 | * | |
87f7e416 TT |
4528 | * is in error because write_inode() could occur while `stuff()' is running, |
4529 | * and the new i_size will be lost. Plus the inode will no longer be on the | |
4530 | * superblock's dirty inode list. | |
ac27a0ec | 4531 | */ |
a9185b41 | 4532 | int ext4_write_inode(struct inode *inode, struct writeback_control *wbc) |
ac27a0ec | 4533 | { |
91ac6f43 FM |
4534 | int err; |
4535 | ||
87f7e416 | 4536 | if (WARN_ON_ONCE(current->flags & PF_MEMALLOC)) |
ac27a0ec DK |
4537 | return 0; |
4538 | ||
91ac6f43 FM |
4539 | if (EXT4_SB(inode->i_sb)->s_journal) { |
4540 | if (ext4_journal_current_handle()) { | |
4541 | jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n"); | |
4542 | dump_stack(); | |
4543 | return -EIO; | |
4544 | } | |
ac27a0ec | 4545 | |
10542c22 JK |
4546 | /* |
4547 | * No need to force transaction in WB_SYNC_NONE mode. Also | |
4548 | * ext4_sync_fs() will force the commit after everything is | |
4549 | * written. | |
4550 | */ | |
4551 | if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync) | |
91ac6f43 FM |
4552 | return 0; |
4553 | ||
4554 | err = ext4_force_commit(inode->i_sb); | |
4555 | } else { | |
4556 | struct ext4_iloc iloc; | |
ac27a0ec | 4557 | |
8b472d73 | 4558 | err = __ext4_get_inode_loc(inode, &iloc, 0); |
91ac6f43 FM |
4559 | if (err) |
4560 | return err; | |
10542c22 JK |
4561 | /* |
4562 | * sync(2) will flush the whole buffer cache. No need to do | |
4563 | * it here separately for each inode. | |
4564 | */ | |
4565 | if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) | |
830156c7 FM |
4566 | sync_dirty_buffer(iloc.bh); |
4567 | if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) { | |
c398eda0 TT |
4568 | EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr, |
4569 | "IO error syncing inode"); | |
830156c7 FM |
4570 | err = -EIO; |
4571 | } | |
fd2dd9fb | 4572 | brelse(iloc.bh); |
91ac6f43 FM |
4573 | } |
4574 | return err; | |
ac27a0ec DK |
4575 | } |
4576 | ||
53e87268 JK |
4577 | /* |
4578 | * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate | |
4579 | * buffers that are attached to a page stradding i_size and are undergoing | |
4580 | * commit. In that case we have to wait for commit to finish and try again. | |
4581 | */ | |
4582 | static void ext4_wait_for_tail_page_commit(struct inode *inode) | |
4583 | { | |
4584 | struct page *page; | |
4585 | unsigned offset; | |
4586 | journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; | |
4587 | tid_t commit_tid = 0; | |
4588 | int ret; | |
4589 | ||
4590 | offset = inode->i_size & (PAGE_CACHE_SIZE - 1); | |
4591 | /* | |
4592 | * All buffers in the last page remain valid? Then there's nothing to | |
4593 | * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE == | |
4594 | * blocksize case | |
4595 | */ | |
4596 | if (offset > PAGE_CACHE_SIZE - (1 << inode->i_blkbits)) | |
4597 | return; | |
4598 | while (1) { | |
4599 | page = find_lock_page(inode->i_mapping, | |
4600 | inode->i_size >> PAGE_CACHE_SHIFT); | |
4601 | if (!page) | |
4602 | return; | |
ca99fdd2 LC |
4603 | ret = __ext4_journalled_invalidatepage(page, offset, |
4604 | PAGE_CACHE_SIZE - offset); | |
53e87268 JK |
4605 | unlock_page(page); |
4606 | page_cache_release(page); | |
4607 | if (ret != -EBUSY) | |
4608 | return; | |
4609 | commit_tid = 0; | |
4610 | read_lock(&journal->j_state_lock); | |
4611 | if (journal->j_committing_transaction) | |
4612 | commit_tid = journal->j_committing_transaction->t_tid; | |
4613 | read_unlock(&journal->j_state_lock); | |
4614 | if (commit_tid) | |
4615 | jbd2_log_wait_commit(journal, commit_tid); | |
4616 | } | |
4617 | } | |
4618 | ||
ac27a0ec | 4619 | /* |
617ba13b | 4620 | * ext4_setattr() |
ac27a0ec DK |
4621 | * |
4622 | * Called from notify_change. | |
4623 | * | |
4624 | * We want to trap VFS attempts to truncate the file as soon as | |
4625 | * possible. In particular, we want to make sure that when the VFS | |
4626 | * shrinks i_size, we put the inode on the orphan list and modify | |
4627 | * i_disksize immediately, so that during the subsequent flushing of | |
4628 | * dirty pages and freeing of disk blocks, we can guarantee that any | |
4629 | * commit will leave the blocks being flushed in an unused state on | |
4630 | * disk. (On recovery, the inode will get truncated and the blocks will | |
4631 | * be freed, so we have a strong guarantee that no future commit will | |
4632 | * leave these blocks visible to the user.) | |
4633 | * | |
678aaf48 JK |
4634 | * Another thing we have to assure is that if we are in ordered mode |
4635 | * and inode is still attached to the committing transaction, we must | |
4636 | * we start writeout of all the dirty pages which are being truncated. | |
4637 | * This way we are sure that all the data written in the previous | |
4638 | * transaction are already on disk (truncate waits for pages under | |
4639 | * writeback). | |
4640 | * | |
4641 | * Called with inode->i_mutex down. | |
ac27a0ec | 4642 | */ |
617ba13b | 4643 | int ext4_setattr(struct dentry *dentry, struct iattr *attr) |
ac27a0ec | 4644 | { |
2b0143b5 | 4645 | struct inode *inode = d_inode(dentry); |
ac27a0ec | 4646 | int error, rc = 0; |
3d287de3 | 4647 | int orphan = 0; |
ac27a0ec DK |
4648 | const unsigned int ia_valid = attr->ia_valid; |
4649 | ||
4650 | error = inode_change_ok(inode, attr); | |
4651 | if (error) | |
4652 | return error; | |
4653 | ||
12755627 | 4654 | if (is_quota_modification(inode, attr)) |
871a2931 | 4655 | dquot_initialize(inode); |
08cefc7a EB |
4656 | if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) || |
4657 | (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) { | |
ac27a0ec DK |
4658 | handle_t *handle; |
4659 | ||
4660 | /* (user+group)*(old+new) structure, inode write (sb, | |
4661 | * inode block, ? - but truncate inode update has it) */ | |
9924a92a TT |
4662 | handle = ext4_journal_start(inode, EXT4_HT_QUOTA, |
4663 | (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) + | |
4664 | EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3); | |
ac27a0ec DK |
4665 | if (IS_ERR(handle)) { |
4666 | error = PTR_ERR(handle); | |
4667 | goto err_out; | |
4668 | } | |
b43fa828 | 4669 | error = dquot_transfer(inode, attr); |
ac27a0ec | 4670 | if (error) { |
617ba13b | 4671 | ext4_journal_stop(handle); |
ac27a0ec DK |
4672 | return error; |
4673 | } | |
4674 | /* Update corresponding info in inode so that everything is in | |
4675 | * one transaction */ | |
4676 | if (attr->ia_valid & ATTR_UID) | |
4677 | inode->i_uid = attr->ia_uid; | |
4678 | if (attr->ia_valid & ATTR_GID) | |
4679 | inode->i_gid = attr->ia_gid; | |
617ba13b MC |
4680 | error = ext4_mark_inode_dirty(handle, inode); |
4681 | ext4_journal_stop(handle); | |
ac27a0ec DK |
4682 | } |
4683 | ||
5208386c JK |
4684 | if (attr->ia_valid & ATTR_SIZE && attr->ia_size != inode->i_size) { |
4685 | handle_t *handle; | |
562c72aa | 4686 | |
12e9b892 | 4687 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) { |
e2b46574 ES |
4688 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
4689 | ||
0c095c7f TT |
4690 | if (attr->ia_size > sbi->s_bitmap_maxbytes) |
4691 | return -EFBIG; | |
e2b46574 | 4692 | } |
dff6efc3 CH |
4693 | |
4694 | if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size) | |
4695 | inode_inc_iversion(inode); | |
4696 | ||
5208386c JK |
4697 | if (S_ISREG(inode->i_mode) && |
4698 | (attr->ia_size < inode->i_size)) { | |
4699 | if (ext4_should_order_data(inode)) { | |
4700 | error = ext4_begin_ordered_truncate(inode, | |
678aaf48 | 4701 | attr->ia_size); |
5208386c | 4702 | if (error) |
678aaf48 | 4703 | goto err_out; |
5208386c JK |
4704 | } |
4705 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 3); | |
4706 | if (IS_ERR(handle)) { | |
4707 | error = PTR_ERR(handle); | |
4708 | goto err_out; | |
4709 | } | |
4710 | if (ext4_handle_valid(handle)) { | |
4711 | error = ext4_orphan_add(handle, inode); | |
4712 | orphan = 1; | |
4713 | } | |
90e775b7 | 4714 | down_write(&EXT4_I(inode)->i_data_sem); |
5208386c JK |
4715 | EXT4_I(inode)->i_disksize = attr->ia_size; |
4716 | rc = ext4_mark_inode_dirty(handle, inode); | |
4717 | if (!error) | |
4718 | error = rc; | |
90e775b7 JK |
4719 | /* |
4720 | * We have to update i_size under i_data_sem together | |
4721 | * with i_disksize to avoid races with writeback code | |
4722 | * running ext4_wb_update_i_disksize(). | |
4723 | */ | |
4724 | if (!error) | |
4725 | i_size_write(inode, attr->ia_size); | |
4726 | up_write(&EXT4_I(inode)->i_data_sem); | |
5208386c JK |
4727 | ext4_journal_stop(handle); |
4728 | if (error) { | |
4729 | ext4_orphan_del(NULL, inode); | |
678aaf48 JK |
4730 | goto err_out; |
4731 | } | |
d6320cbf JK |
4732 | } else { |
4733 | loff_t oldsize = inode->i_size; | |
4734 | ||
90e775b7 | 4735 | i_size_write(inode, attr->ia_size); |
d6320cbf JK |
4736 | pagecache_isize_extended(inode, oldsize, inode->i_size); |
4737 | } | |
53e87268 | 4738 | |
5208386c JK |
4739 | /* |
4740 | * Blocks are going to be removed from the inode. Wait | |
4741 | * for dio in flight. Temporarily disable | |
4742 | * dioread_nolock to prevent livelock. | |
4743 | */ | |
4744 | if (orphan) { | |
4745 | if (!ext4_should_journal_data(inode)) { | |
4746 | ext4_inode_block_unlocked_dio(inode); | |
4747 | inode_dio_wait(inode); | |
4748 | ext4_inode_resume_unlocked_dio(inode); | |
4749 | } else | |
4750 | ext4_wait_for_tail_page_commit(inode); | |
1c9114f9 | 4751 | } |
5208386c JK |
4752 | /* |
4753 | * Truncate pagecache after we've waited for commit | |
4754 | * in data=journal mode to make pages freeable. | |
4755 | */ | |
923ae0ff | 4756 | truncate_pagecache(inode, inode->i_size); |
072bd7ea | 4757 | } |
5208386c JK |
4758 | /* |
4759 | * We want to call ext4_truncate() even if attr->ia_size == | |
4760 | * inode->i_size for cases like truncation of fallocated space | |
4761 | */ | |
4762 | if (attr->ia_valid & ATTR_SIZE) | |
4763 | ext4_truncate(inode); | |
ac27a0ec | 4764 | |
1025774c CH |
4765 | if (!rc) { |
4766 | setattr_copy(inode, attr); | |
4767 | mark_inode_dirty(inode); | |
4768 | } | |
4769 | ||
4770 | /* | |
4771 | * If the call to ext4_truncate failed to get a transaction handle at | |
4772 | * all, we need to clean up the in-core orphan list manually. | |
4773 | */ | |
3d287de3 | 4774 | if (orphan && inode->i_nlink) |
617ba13b | 4775 | ext4_orphan_del(NULL, inode); |
ac27a0ec DK |
4776 | |
4777 | if (!rc && (ia_valid & ATTR_MODE)) | |
64e178a7 | 4778 | rc = posix_acl_chmod(inode, inode->i_mode); |
ac27a0ec DK |
4779 | |
4780 | err_out: | |
617ba13b | 4781 | ext4_std_error(inode->i_sb, error); |
ac27a0ec DK |
4782 | if (!error) |
4783 | error = rc; | |
4784 | return error; | |
4785 | } | |
4786 | ||
3e3398a0 MC |
4787 | int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry, |
4788 | struct kstat *stat) | |
4789 | { | |
4790 | struct inode *inode; | |
8af8eecc | 4791 | unsigned long long delalloc_blocks; |
3e3398a0 | 4792 | |
2b0143b5 | 4793 | inode = d_inode(dentry); |
3e3398a0 MC |
4794 | generic_fillattr(inode, stat); |
4795 | ||
9206c561 AD |
4796 | /* |
4797 | * If there is inline data in the inode, the inode will normally not | |
4798 | * have data blocks allocated (it may have an external xattr block). | |
4799 | * Report at least one sector for such files, so tools like tar, rsync, | |
4800 | * others doen't incorrectly think the file is completely sparse. | |
4801 | */ | |
4802 | if (unlikely(ext4_has_inline_data(inode))) | |
4803 | stat->blocks += (stat->size + 511) >> 9; | |
4804 | ||
3e3398a0 MC |
4805 | /* |
4806 | * We can't update i_blocks if the block allocation is delayed | |
4807 | * otherwise in the case of system crash before the real block | |
4808 | * allocation is done, we will have i_blocks inconsistent with | |
4809 | * on-disk file blocks. | |
4810 | * We always keep i_blocks updated together with real | |
4811 | * allocation. But to not confuse with user, stat | |
4812 | * will return the blocks that include the delayed allocation | |
4813 | * blocks for this file. | |
4814 | */ | |
96607551 | 4815 | delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb), |
9206c561 AD |
4816 | EXT4_I(inode)->i_reserved_data_blocks); |
4817 | stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9); | |
3e3398a0 MC |
4818 | return 0; |
4819 | } | |
ac27a0ec | 4820 | |
fffb2739 JK |
4821 | static int ext4_index_trans_blocks(struct inode *inode, int lblocks, |
4822 | int pextents) | |
a02908f1 | 4823 | { |
12e9b892 | 4824 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) |
fffb2739 JK |
4825 | return ext4_ind_trans_blocks(inode, lblocks); |
4826 | return ext4_ext_index_trans_blocks(inode, pextents); | |
a02908f1 | 4827 | } |
ac51d837 | 4828 | |
ac27a0ec | 4829 | /* |
a02908f1 MC |
4830 | * Account for index blocks, block groups bitmaps and block group |
4831 | * descriptor blocks if modify datablocks and index blocks | |
4832 | * worse case, the indexs blocks spread over different block groups | |
ac27a0ec | 4833 | * |
a02908f1 | 4834 | * If datablocks are discontiguous, they are possible to spread over |
4907cb7b | 4835 | * different block groups too. If they are contiguous, with flexbg, |
a02908f1 | 4836 | * they could still across block group boundary. |
ac27a0ec | 4837 | * |
a02908f1 MC |
4838 | * Also account for superblock, inode, quota and xattr blocks |
4839 | */ | |
fffb2739 JK |
4840 | static int ext4_meta_trans_blocks(struct inode *inode, int lblocks, |
4841 | int pextents) | |
a02908f1 | 4842 | { |
8df9675f TT |
4843 | ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb); |
4844 | int gdpblocks; | |
a02908f1 MC |
4845 | int idxblocks; |
4846 | int ret = 0; | |
4847 | ||
4848 | /* | |
fffb2739 JK |
4849 | * How many index blocks need to touch to map @lblocks logical blocks |
4850 | * to @pextents physical extents? | |
a02908f1 | 4851 | */ |
fffb2739 | 4852 | idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents); |
a02908f1 MC |
4853 | |
4854 | ret = idxblocks; | |
4855 | ||
4856 | /* | |
4857 | * Now let's see how many group bitmaps and group descriptors need | |
4858 | * to account | |
4859 | */ | |
fffb2739 | 4860 | groups = idxblocks + pextents; |
a02908f1 | 4861 | gdpblocks = groups; |
8df9675f TT |
4862 | if (groups > ngroups) |
4863 | groups = ngroups; | |
a02908f1 MC |
4864 | if (groups > EXT4_SB(inode->i_sb)->s_gdb_count) |
4865 | gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count; | |
4866 | ||
4867 | /* bitmaps and block group descriptor blocks */ | |
4868 | ret += groups + gdpblocks; | |
4869 | ||
4870 | /* Blocks for super block, inode, quota and xattr blocks */ | |
4871 | ret += EXT4_META_TRANS_BLOCKS(inode->i_sb); | |
4872 | ||
4873 | return ret; | |
4874 | } | |
4875 | ||
4876 | /* | |
25985edc | 4877 | * Calculate the total number of credits to reserve to fit |
f3bd1f3f MC |
4878 | * the modification of a single pages into a single transaction, |
4879 | * which may include multiple chunks of block allocations. | |
ac27a0ec | 4880 | * |
525f4ed8 | 4881 | * This could be called via ext4_write_begin() |
ac27a0ec | 4882 | * |
525f4ed8 | 4883 | * We need to consider the worse case, when |
a02908f1 | 4884 | * one new block per extent. |
ac27a0ec | 4885 | */ |
a86c6181 | 4886 | int ext4_writepage_trans_blocks(struct inode *inode) |
ac27a0ec | 4887 | { |
617ba13b | 4888 | int bpp = ext4_journal_blocks_per_page(inode); |
ac27a0ec DK |
4889 | int ret; |
4890 | ||
fffb2739 | 4891 | ret = ext4_meta_trans_blocks(inode, bpp, bpp); |
a86c6181 | 4892 | |
a02908f1 | 4893 | /* Account for data blocks for journalled mode */ |
617ba13b | 4894 | if (ext4_should_journal_data(inode)) |
a02908f1 | 4895 | ret += bpp; |
ac27a0ec DK |
4896 | return ret; |
4897 | } | |
f3bd1f3f MC |
4898 | |
4899 | /* | |
4900 | * Calculate the journal credits for a chunk of data modification. | |
4901 | * | |
4902 | * This is called from DIO, fallocate or whoever calling | |
79e83036 | 4903 | * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks. |
f3bd1f3f MC |
4904 | * |
4905 | * journal buffers for data blocks are not included here, as DIO | |
4906 | * and fallocate do no need to journal data buffers. | |
4907 | */ | |
4908 | int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks) | |
4909 | { | |
4910 | return ext4_meta_trans_blocks(inode, nrblocks, 1); | |
4911 | } | |
4912 | ||
ac27a0ec | 4913 | /* |
617ba13b | 4914 | * The caller must have previously called ext4_reserve_inode_write(). |
ac27a0ec DK |
4915 | * Give this, we know that the caller already has write access to iloc->bh. |
4916 | */ | |
617ba13b | 4917 | int ext4_mark_iloc_dirty(handle_t *handle, |
de9a55b8 | 4918 | struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec DK |
4919 | { |
4920 | int err = 0; | |
4921 | ||
c64db50e | 4922 | if (IS_I_VERSION(inode)) |
25ec56b5 JNC |
4923 | inode_inc_iversion(inode); |
4924 | ||
ac27a0ec DK |
4925 | /* the do_update_inode consumes one bh->b_count */ |
4926 | get_bh(iloc->bh); | |
4927 | ||
dab291af | 4928 | /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */ |
830156c7 | 4929 | err = ext4_do_update_inode(handle, inode, iloc); |
ac27a0ec DK |
4930 | put_bh(iloc->bh); |
4931 | return err; | |
4932 | } | |
4933 | ||
4934 | /* | |
4935 | * On success, We end up with an outstanding reference count against | |
4936 | * iloc->bh. This _must_ be cleaned up later. | |
4937 | */ | |
4938 | ||
4939 | int | |
617ba13b MC |
4940 | ext4_reserve_inode_write(handle_t *handle, struct inode *inode, |
4941 | struct ext4_iloc *iloc) | |
ac27a0ec | 4942 | { |
0390131b FM |
4943 | int err; |
4944 | ||
4945 | err = ext4_get_inode_loc(inode, iloc); | |
4946 | if (!err) { | |
4947 | BUFFER_TRACE(iloc->bh, "get_write_access"); | |
4948 | err = ext4_journal_get_write_access(handle, iloc->bh); | |
4949 | if (err) { | |
4950 | brelse(iloc->bh); | |
4951 | iloc->bh = NULL; | |
ac27a0ec DK |
4952 | } |
4953 | } | |
617ba13b | 4954 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
4955 | return err; |
4956 | } | |
4957 | ||
6dd4ee7c KS |
4958 | /* |
4959 | * Expand an inode by new_extra_isize bytes. | |
4960 | * Returns 0 on success or negative error number on failure. | |
4961 | */ | |
1d03ec98 AK |
4962 | static int ext4_expand_extra_isize(struct inode *inode, |
4963 | unsigned int new_extra_isize, | |
4964 | struct ext4_iloc iloc, | |
4965 | handle_t *handle) | |
6dd4ee7c KS |
4966 | { |
4967 | struct ext4_inode *raw_inode; | |
4968 | struct ext4_xattr_ibody_header *header; | |
6dd4ee7c KS |
4969 | |
4970 | if (EXT4_I(inode)->i_extra_isize >= new_extra_isize) | |
4971 | return 0; | |
4972 | ||
4973 | raw_inode = ext4_raw_inode(&iloc); | |
4974 | ||
4975 | header = IHDR(inode, raw_inode); | |
6dd4ee7c KS |
4976 | |
4977 | /* No extended attributes present */ | |
19f5fb7a TT |
4978 | if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) || |
4979 | header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) { | |
6dd4ee7c KS |
4980 | memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0, |
4981 | new_extra_isize); | |
4982 | EXT4_I(inode)->i_extra_isize = new_extra_isize; | |
4983 | return 0; | |
4984 | } | |
4985 | ||
4986 | /* try to expand with EAs present */ | |
4987 | return ext4_expand_extra_isize_ea(inode, new_extra_isize, | |
4988 | raw_inode, handle); | |
4989 | } | |
4990 | ||
ac27a0ec DK |
4991 | /* |
4992 | * What we do here is to mark the in-core inode as clean with respect to inode | |
4993 | * dirtiness (it may still be data-dirty). | |
4994 | * This means that the in-core inode may be reaped by prune_icache | |
4995 | * without having to perform any I/O. This is a very good thing, | |
4996 | * because *any* task may call prune_icache - even ones which | |
4997 | * have a transaction open against a different journal. | |
4998 | * | |
4999 | * Is this cheating? Not really. Sure, we haven't written the | |
5000 | * inode out, but prune_icache isn't a user-visible syncing function. | |
5001 | * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync) | |
5002 | * we start and wait on commits. | |
ac27a0ec | 5003 | */ |
617ba13b | 5004 | int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode) |
ac27a0ec | 5005 | { |
617ba13b | 5006 | struct ext4_iloc iloc; |
6dd4ee7c KS |
5007 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
5008 | static unsigned int mnt_count; | |
5009 | int err, ret; | |
ac27a0ec DK |
5010 | |
5011 | might_sleep(); | |
7ff9c073 | 5012 | trace_ext4_mark_inode_dirty(inode, _RET_IP_); |
617ba13b | 5013 | err = ext4_reserve_inode_write(handle, inode, &iloc); |
0390131b FM |
5014 | if (ext4_handle_valid(handle) && |
5015 | EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize && | |
19f5fb7a | 5016 | !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) { |
6dd4ee7c KS |
5017 | /* |
5018 | * We need extra buffer credits since we may write into EA block | |
5019 | * with this same handle. If journal_extend fails, then it will | |
5020 | * only result in a minor loss of functionality for that inode. | |
5021 | * If this is felt to be critical, then e2fsck should be run to | |
5022 | * force a large enough s_min_extra_isize. | |
5023 | */ | |
5024 | if ((jbd2_journal_extend(handle, | |
5025 | EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) { | |
5026 | ret = ext4_expand_extra_isize(inode, | |
5027 | sbi->s_want_extra_isize, | |
5028 | iloc, handle); | |
5029 | if (ret) { | |
19f5fb7a TT |
5030 | ext4_set_inode_state(inode, |
5031 | EXT4_STATE_NO_EXPAND); | |
c1bddad9 AK |
5032 | if (mnt_count != |
5033 | le16_to_cpu(sbi->s_es->s_mnt_count)) { | |
12062ddd | 5034 | ext4_warning(inode->i_sb, |
6dd4ee7c KS |
5035 | "Unable to expand inode %lu. Delete" |
5036 | " some EAs or run e2fsck.", | |
5037 | inode->i_ino); | |
c1bddad9 AK |
5038 | mnt_count = |
5039 | le16_to_cpu(sbi->s_es->s_mnt_count); | |
6dd4ee7c KS |
5040 | } |
5041 | } | |
5042 | } | |
5043 | } | |
ac27a0ec | 5044 | if (!err) |
617ba13b | 5045 | err = ext4_mark_iloc_dirty(handle, inode, &iloc); |
ac27a0ec DK |
5046 | return err; |
5047 | } | |
5048 | ||
5049 | /* | |
617ba13b | 5050 | * ext4_dirty_inode() is called from __mark_inode_dirty() |
ac27a0ec DK |
5051 | * |
5052 | * We're really interested in the case where a file is being extended. | |
5053 | * i_size has been changed by generic_commit_write() and we thus need | |
5054 | * to include the updated inode in the current transaction. | |
5055 | * | |
5dd4056d | 5056 | * Also, dquot_alloc_block() will always dirty the inode when blocks |
ac27a0ec DK |
5057 | * are allocated to the file. |
5058 | * | |
5059 | * If the inode is marked synchronous, we don't honour that here - doing | |
5060 | * so would cause a commit on atime updates, which we don't bother doing. | |
5061 | * We handle synchronous inodes at the highest possible level. | |
0ae45f63 TT |
5062 | * |
5063 | * If only the I_DIRTY_TIME flag is set, we can skip everything. If | |
5064 | * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need | |
5065 | * to copy into the on-disk inode structure are the timestamp files. | |
ac27a0ec | 5066 | */ |
aa385729 | 5067 | void ext4_dirty_inode(struct inode *inode, int flags) |
ac27a0ec | 5068 | { |
ac27a0ec DK |
5069 | handle_t *handle; |
5070 | ||
0ae45f63 TT |
5071 | if (flags == I_DIRTY_TIME) |
5072 | return; | |
9924a92a | 5073 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 2); |
ac27a0ec DK |
5074 | if (IS_ERR(handle)) |
5075 | goto out; | |
f3dc272f | 5076 | |
f3dc272f CW |
5077 | ext4_mark_inode_dirty(handle, inode); |
5078 | ||
617ba13b | 5079 | ext4_journal_stop(handle); |
ac27a0ec DK |
5080 | out: |
5081 | return; | |
5082 | } | |
5083 | ||
5084 | #if 0 | |
5085 | /* | |
5086 | * Bind an inode's backing buffer_head into this transaction, to prevent | |
5087 | * it from being flushed to disk early. Unlike | |
617ba13b | 5088 | * ext4_reserve_inode_write, this leaves behind no bh reference and |
ac27a0ec DK |
5089 | * returns no iloc structure, so the caller needs to repeat the iloc |
5090 | * lookup to mark the inode dirty later. | |
5091 | */ | |
617ba13b | 5092 | static int ext4_pin_inode(handle_t *handle, struct inode *inode) |
ac27a0ec | 5093 | { |
617ba13b | 5094 | struct ext4_iloc iloc; |
ac27a0ec DK |
5095 | |
5096 | int err = 0; | |
5097 | if (handle) { | |
617ba13b | 5098 | err = ext4_get_inode_loc(inode, &iloc); |
ac27a0ec DK |
5099 | if (!err) { |
5100 | BUFFER_TRACE(iloc.bh, "get_write_access"); | |
dab291af | 5101 | err = jbd2_journal_get_write_access(handle, iloc.bh); |
ac27a0ec | 5102 | if (!err) |
0390131b | 5103 | err = ext4_handle_dirty_metadata(handle, |
73b50c1c | 5104 | NULL, |
0390131b | 5105 | iloc.bh); |
ac27a0ec DK |
5106 | brelse(iloc.bh); |
5107 | } | |
5108 | } | |
617ba13b | 5109 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
5110 | return err; |
5111 | } | |
5112 | #endif | |
5113 | ||
617ba13b | 5114 | int ext4_change_inode_journal_flag(struct inode *inode, int val) |
ac27a0ec DK |
5115 | { |
5116 | journal_t *journal; | |
5117 | handle_t *handle; | |
5118 | int err; | |
5119 | ||
5120 | /* | |
5121 | * We have to be very careful here: changing a data block's | |
5122 | * journaling status dynamically is dangerous. If we write a | |
5123 | * data block to the journal, change the status and then delete | |
5124 | * that block, we risk forgetting to revoke the old log record | |
5125 | * from the journal and so a subsequent replay can corrupt data. | |
5126 | * So, first we make sure that the journal is empty and that | |
5127 | * nobody is changing anything. | |
5128 | */ | |
5129 | ||
617ba13b | 5130 | journal = EXT4_JOURNAL(inode); |
0390131b FM |
5131 | if (!journal) |
5132 | return 0; | |
d699594d | 5133 | if (is_journal_aborted(journal)) |
ac27a0ec | 5134 | return -EROFS; |
2aff57b0 YY |
5135 | /* We have to allocate physical blocks for delalloc blocks |
5136 | * before flushing journal. otherwise delalloc blocks can not | |
5137 | * be allocated any more. even more truncate on delalloc blocks | |
5138 | * could trigger BUG by flushing delalloc blocks in journal. | |
5139 | * There is no delalloc block in non-journal data mode. | |
5140 | */ | |
5141 | if (val && test_opt(inode->i_sb, DELALLOC)) { | |
5142 | err = ext4_alloc_da_blocks(inode); | |
5143 | if (err < 0) | |
5144 | return err; | |
5145 | } | |
ac27a0ec | 5146 | |
17335dcc DM |
5147 | /* Wait for all existing dio workers */ |
5148 | ext4_inode_block_unlocked_dio(inode); | |
5149 | inode_dio_wait(inode); | |
5150 | ||
dab291af | 5151 | jbd2_journal_lock_updates(journal); |
ac27a0ec DK |
5152 | |
5153 | /* | |
5154 | * OK, there are no updates running now, and all cached data is | |
5155 | * synced to disk. We are now in a completely consistent state | |
5156 | * which doesn't have anything in the journal, and we know that | |
5157 | * no filesystem updates are running, so it is safe to modify | |
5158 | * the inode's in-core data-journaling state flag now. | |
5159 | */ | |
5160 | ||
5161 | if (val) | |
12e9b892 | 5162 | ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
5872ddaa | 5163 | else { |
4f879ca6 JK |
5164 | err = jbd2_journal_flush(journal); |
5165 | if (err < 0) { | |
5166 | jbd2_journal_unlock_updates(journal); | |
5167 | ext4_inode_resume_unlocked_dio(inode); | |
5168 | return err; | |
5169 | } | |
12e9b892 | 5170 | ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
5872ddaa | 5171 | } |
617ba13b | 5172 | ext4_set_aops(inode); |
ac27a0ec | 5173 | |
dab291af | 5174 | jbd2_journal_unlock_updates(journal); |
17335dcc | 5175 | ext4_inode_resume_unlocked_dio(inode); |
ac27a0ec DK |
5176 | |
5177 | /* Finally we can mark the inode as dirty. */ | |
5178 | ||
9924a92a | 5179 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 1); |
ac27a0ec DK |
5180 | if (IS_ERR(handle)) |
5181 | return PTR_ERR(handle); | |
5182 | ||
617ba13b | 5183 | err = ext4_mark_inode_dirty(handle, inode); |
0390131b | 5184 | ext4_handle_sync(handle); |
617ba13b MC |
5185 | ext4_journal_stop(handle); |
5186 | ext4_std_error(inode->i_sb, err); | |
ac27a0ec DK |
5187 | |
5188 | return err; | |
5189 | } | |
2e9ee850 AK |
5190 | |
5191 | static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh) | |
5192 | { | |
5193 | return !buffer_mapped(bh); | |
5194 | } | |
5195 | ||
c2ec175c | 5196 | int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
2e9ee850 | 5197 | { |
c2ec175c | 5198 | struct page *page = vmf->page; |
2e9ee850 AK |
5199 | loff_t size; |
5200 | unsigned long len; | |
9ea7df53 | 5201 | int ret; |
2e9ee850 | 5202 | struct file *file = vma->vm_file; |
496ad9aa | 5203 | struct inode *inode = file_inode(file); |
2e9ee850 | 5204 | struct address_space *mapping = inode->i_mapping; |
9ea7df53 JK |
5205 | handle_t *handle; |
5206 | get_block_t *get_block; | |
5207 | int retries = 0; | |
2e9ee850 | 5208 | |
8e8ad8a5 | 5209 | sb_start_pagefault(inode->i_sb); |
041bbb6d | 5210 | file_update_time(vma->vm_file); |
9ea7df53 JK |
5211 | /* Delalloc case is easy... */ |
5212 | if (test_opt(inode->i_sb, DELALLOC) && | |
5213 | !ext4_should_journal_data(inode) && | |
5214 | !ext4_nonda_switch(inode->i_sb)) { | |
5215 | do { | |
5216 | ret = __block_page_mkwrite(vma, vmf, | |
5217 | ext4_da_get_block_prep); | |
5218 | } while (ret == -ENOSPC && | |
5219 | ext4_should_retry_alloc(inode->i_sb, &retries)); | |
5220 | goto out_ret; | |
2e9ee850 | 5221 | } |
0e499890 DW |
5222 | |
5223 | lock_page(page); | |
9ea7df53 JK |
5224 | size = i_size_read(inode); |
5225 | /* Page got truncated from under us? */ | |
5226 | if (page->mapping != mapping || page_offset(page) > size) { | |
5227 | unlock_page(page); | |
5228 | ret = VM_FAULT_NOPAGE; | |
5229 | goto out; | |
0e499890 | 5230 | } |
2e9ee850 AK |
5231 | |
5232 | if (page->index == size >> PAGE_CACHE_SHIFT) | |
5233 | len = size & ~PAGE_CACHE_MASK; | |
5234 | else | |
5235 | len = PAGE_CACHE_SIZE; | |
a827eaff | 5236 | /* |
9ea7df53 JK |
5237 | * Return if we have all the buffers mapped. This avoids the need to do |
5238 | * journal_start/journal_stop which can block and take a long time | |
a827eaff | 5239 | */ |
2e9ee850 | 5240 | if (page_has_buffers(page)) { |
f19d5870 TM |
5241 | if (!ext4_walk_page_buffers(NULL, page_buffers(page), |
5242 | 0, len, NULL, | |
5243 | ext4_bh_unmapped)) { | |
9ea7df53 | 5244 | /* Wait so that we don't change page under IO */ |
1d1d1a76 | 5245 | wait_for_stable_page(page); |
9ea7df53 JK |
5246 | ret = VM_FAULT_LOCKED; |
5247 | goto out; | |
a827eaff | 5248 | } |
2e9ee850 | 5249 | } |
a827eaff | 5250 | unlock_page(page); |
9ea7df53 JK |
5251 | /* OK, we need to fill the hole... */ |
5252 | if (ext4_should_dioread_nolock(inode)) | |
5253 | get_block = ext4_get_block_write; | |
5254 | else | |
5255 | get_block = ext4_get_block; | |
5256 | retry_alloc: | |
9924a92a TT |
5257 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
5258 | ext4_writepage_trans_blocks(inode)); | |
9ea7df53 | 5259 | if (IS_ERR(handle)) { |
c2ec175c | 5260 | ret = VM_FAULT_SIGBUS; |
9ea7df53 JK |
5261 | goto out; |
5262 | } | |
5263 | ret = __block_page_mkwrite(vma, vmf, get_block); | |
5264 | if (!ret && ext4_should_journal_data(inode)) { | |
f19d5870 | 5265 | if (ext4_walk_page_buffers(handle, page_buffers(page), 0, |
9ea7df53 JK |
5266 | PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) { |
5267 | unlock_page(page); | |
5268 | ret = VM_FAULT_SIGBUS; | |
fcbb5515 | 5269 | ext4_journal_stop(handle); |
9ea7df53 JK |
5270 | goto out; |
5271 | } | |
5272 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); | |
5273 | } | |
5274 | ext4_journal_stop(handle); | |
5275 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) | |
5276 | goto retry_alloc; | |
5277 | out_ret: | |
5278 | ret = block_page_mkwrite_return(ret); | |
5279 | out: | |
8e8ad8a5 | 5280 | sb_end_pagefault(inode->i_sb); |
2e9ee850 AK |
5281 | return ret; |
5282 | } |