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> | |
dab291af | 23 | #include <linux/jbd2.h> |
ac27a0ec DK |
24 | #include <linux/highuid.h> |
25 | #include <linux/pagemap.h> | |
26 | #include <linux/quotaops.h> | |
27 | #include <linux/string.h> | |
28 | #include <linux/buffer_head.h> | |
29 | #include <linux/writeback.h> | |
64769240 | 30 | #include <linux/pagevec.h> |
ac27a0ec | 31 | #include <linux/mpage.h> |
e83c1397 | 32 | #include <linux/namei.h> |
ac27a0ec DK |
33 | #include <linux/uio.h> |
34 | #include <linux/bio.h> | |
4c0425ff | 35 | #include <linux/workqueue.h> |
744692dc | 36 | #include <linux/kernel.h> |
6db26ffc | 37 | #include <linux/printk.h> |
5a0e3ad6 | 38 | #include <linux/slab.h> |
a8901d34 | 39 | #include <linux/ratelimit.h> |
9bffad1e | 40 | |
3dcf5451 | 41 | #include "ext4_jbd2.h" |
ac27a0ec DK |
42 | #include "xattr.h" |
43 | #include "acl.h" | |
9f125d64 | 44 | #include "truncate.h" |
ac27a0ec | 45 | |
9bffad1e TT |
46 | #include <trace/events/ext4.h> |
47 | ||
a1d6cc56 AK |
48 | #define MPAGE_DA_EXTENT_TAIL 0x01 |
49 | ||
814525f4 DW |
50 | static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw, |
51 | struct ext4_inode_info *ei) | |
52 | { | |
53 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
54 | __u16 csum_lo; | |
55 | __u16 csum_hi = 0; | |
56 | __u32 csum; | |
57 | ||
58 | csum_lo = raw->i_checksum_lo; | |
59 | raw->i_checksum_lo = 0; | |
60 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && | |
61 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) { | |
62 | csum_hi = raw->i_checksum_hi; | |
63 | raw->i_checksum_hi = 0; | |
64 | } | |
65 | ||
66 | csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw, | |
67 | EXT4_INODE_SIZE(inode->i_sb)); | |
68 | ||
69 | raw->i_checksum_lo = csum_lo; | |
70 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && | |
71 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) | |
72 | raw->i_checksum_hi = csum_hi; | |
73 | ||
74 | return csum; | |
75 | } | |
76 | ||
77 | static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw, | |
78 | struct ext4_inode_info *ei) | |
79 | { | |
80 | __u32 provided, calculated; | |
81 | ||
82 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != | |
83 | cpu_to_le32(EXT4_OS_LINUX) || | |
84 | !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, | |
85 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) | |
86 | return 1; | |
87 | ||
88 | provided = le16_to_cpu(raw->i_checksum_lo); | |
89 | calculated = ext4_inode_csum(inode, raw, ei); | |
90 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && | |
91 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) | |
92 | provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16; | |
93 | else | |
94 | calculated &= 0xFFFF; | |
95 | ||
96 | return provided == calculated; | |
97 | } | |
98 | ||
99 | static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw, | |
100 | struct ext4_inode_info *ei) | |
101 | { | |
102 | __u32 csum; | |
103 | ||
104 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != | |
105 | cpu_to_le32(EXT4_OS_LINUX) || | |
106 | !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, | |
107 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) | |
108 | return; | |
109 | ||
110 | csum = ext4_inode_csum(inode, raw, ei); | |
111 | raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF); | |
112 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && | |
113 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) | |
114 | raw->i_checksum_hi = cpu_to_le16(csum >> 16); | |
115 | } | |
116 | ||
678aaf48 JK |
117 | static inline int ext4_begin_ordered_truncate(struct inode *inode, |
118 | loff_t new_size) | |
119 | { | |
7ff9c073 | 120 | trace_ext4_begin_ordered_truncate(inode, new_size); |
8aefcd55 TT |
121 | /* |
122 | * If jinode is zero, then we never opened the file for | |
123 | * writing, so there's no need to call | |
124 | * jbd2_journal_begin_ordered_truncate() since there's no | |
125 | * outstanding writes we need to flush. | |
126 | */ | |
127 | if (!EXT4_I(inode)->jinode) | |
128 | return 0; | |
129 | return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode), | |
130 | EXT4_I(inode)->jinode, | |
131 | new_size); | |
678aaf48 JK |
132 | } |
133 | ||
64769240 | 134 | static void ext4_invalidatepage(struct page *page, unsigned long offset); |
cb20d518 TT |
135 | static int noalloc_get_block_write(struct inode *inode, sector_t iblock, |
136 | struct buffer_head *bh_result, int create); | |
137 | static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode); | |
138 | static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate); | |
139 | static int __ext4_journalled_writepage(struct page *page, unsigned int len); | |
140 | static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh); | |
5f163cc7 ES |
141 | static int ext4_discard_partial_page_buffers_no_lock(handle_t *handle, |
142 | struct inode *inode, struct page *page, loff_t from, | |
143 | loff_t length, int flags); | |
64769240 | 144 | |
ac27a0ec DK |
145 | /* |
146 | * Test whether an inode is a fast symlink. | |
147 | */ | |
617ba13b | 148 | static int ext4_inode_is_fast_symlink(struct inode *inode) |
ac27a0ec | 149 | { |
617ba13b | 150 | int ea_blocks = EXT4_I(inode)->i_file_acl ? |
ac27a0ec DK |
151 | (inode->i_sb->s_blocksize >> 9) : 0; |
152 | ||
153 | return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0); | |
154 | } | |
155 | ||
ac27a0ec DK |
156 | /* |
157 | * Restart the transaction associated with *handle. This does a commit, | |
158 | * so before we call here everything must be consistently dirtied against | |
159 | * this transaction. | |
160 | */ | |
fa5d1113 | 161 | int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode, |
487caeef | 162 | int nblocks) |
ac27a0ec | 163 | { |
487caeef JK |
164 | int ret; |
165 | ||
166 | /* | |
e35fd660 | 167 | * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this |
487caeef JK |
168 | * moment, get_block can be called only for blocks inside i_size since |
169 | * page cache has been already dropped and writes are blocked by | |
170 | * i_mutex. So we can safely drop the i_data_sem here. | |
171 | */ | |
0390131b | 172 | BUG_ON(EXT4_JOURNAL(inode) == NULL); |
ac27a0ec | 173 | jbd_debug(2, "restarting handle %p\n", handle); |
487caeef | 174 | up_write(&EXT4_I(inode)->i_data_sem); |
8e8eaabe | 175 | ret = ext4_journal_restart(handle, nblocks); |
487caeef | 176 | down_write(&EXT4_I(inode)->i_data_sem); |
fa5d1113 | 177 | ext4_discard_preallocations(inode); |
487caeef JK |
178 | |
179 | return ret; | |
ac27a0ec DK |
180 | } |
181 | ||
182 | /* | |
183 | * Called at the last iput() if i_nlink is zero. | |
184 | */ | |
0930fcc1 | 185 | void ext4_evict_inode(struct inode *inode) |
ac27a0ec DK |
186 | { |
187 | handle_t *handle; | |
bc965ab3 | 188 | int err; |
ac27a0ec | 189 | |
7ff9c073 | 190 | trace_ext4_evict_inode(inode); |
2581fdc8 | 191 | |
2581fdc8 JZ |
192 | ext4_ioend_wait(inode); |
193 | ||
0930fcc1 | 194 | if (inode->i_nlink) { |
2d859db3 JK |
195 | /* |
196 | * When journalling data dirty buffers are tracked only in the | |
197 | * journal. So although mm thinks everything is clean and | |
198 | * ready for reaping the inode might still have some pages to | |
199 | * write in the running transaction or waiting to be | |
200 | * checkpointed. Thus calling jbd2_journal_invalidatepage() | |
201 | * (via truncate_inode_pages()) to discard these buffers can | |
202 | * cause data loss. Also even if we did not discard these | |
203 | * buffers, we would have no way to find them after the inode | |
204 | * is reaped and thus user could see stale data if he tries to | |
205 | * read them before the transaction is checkpointed. So be | |
206 | * careful and force everything to disk here... We use | |
207 | * ei->i_datasync_tid to store the newest transaction | |
208 | * containing inode's data. | |
209 | * | |
210 | * Note that directories do not have this problem because they | |
211 | * don't use page cache. | |
212 | */ | |
213 | if (ext4_should_journal_data(inode) && | |
214 | (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode))) { | |
215 | journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; | |
216 | tid_t commit_tid = EXT4_I(inode)->i_datasync_tid; | |
217 | ||
218 | jbd2_log_start_commit(journal, commit_tid); | |
219 | jbd2_log_wait_commit(journal, commit_tid); | |
220 | filemap_write_and_wait(&inode->i_data); | |
221 | } | |
0930fcc1 AV |
222 | truncate_inode_pages(&inode->i_data, 0); |
223 | goto no_delete; | |
224 | } | |
225 | ||
907f4554 | 226 | if (!is_bad_inode(inode)) |
871a2931 | 227 | dquot_initialize(inode); |
907f4554 | 228 | |
678aaf48 JK |
229 | if (ext4_should_order_data(inode)) |
230 | ext4_begin_ordered_truncate(inode, 0); | |
ac27a0ec DK |
231 | truncate_inode_pages(&inode->i_data, 0); |
232 | ||
233 | if (is_bad_inode(inode)) | |
234 | goto no_delete; | |
235 | ||
9f125d64 | 236 | handle = ext4_journal_start(inode, ext4_blocks_for_truncate(inode)+3); |
ac27a0ec | 237 | if (IS_ERR(handle)) { |
bc965ab3 | 238 | ext4_std_error(inode->i_sb, PTR_ERR(handle)); |
ac27a0ec DK |
239 | /* |
240 | * If we're going to skip the normal cleanup, we still need to | |
241 | * make sure that the in-core orphan linked list is properly | |
242 | * cleaned up. | |
243 | */ | |
617ba13b | 244 | ext4_orphan_del(NULL, inode); |
ac27a0ec DK |
245 | goto no_delete; |
246 | } | |
247 | ||
248 | if (IS_SYNC(inode)) | |
0390131b | 249 | ext4_handle_sync(handle); |
ac27a0ec | 250 | inode->i_size = 0; |
bc965ab3 TT |
251 | err = ext4_mark_inode_dirty(handle, inode); |
252 | if (err) { | |
12062ddd | 253 | ext4_warning(inode->i_sb, |
bc965ab3 TT |
254 | "couldn't mark inode dirty (err %d)", err); |
255 | goto stop_handle; | |
256 | } | |
ac27a0ec | 257 | if (inode->i_blocks) |
617ba13b | 258 | ext4_truncate(inode); |
bc965ab3 TT |
259 | |
260 | /* | |
261 | * ext4_ext_truncate() doesn't reserve any slop when it | |
262 | * restarts journal transactions; therefore there may not be | |
263 | * enough credits left in the handle to remove the inode from | |
264 | * the orphan list and set the dtime field. | |
265 | */ | |
0390131b | 266 | if (!ext4_handle_has_enough_credits(handle, 3)) { |
bc965ab3 TT |
267 | err = ext4_journal_extend(handle, 3); |
268 | if (err > 0) | |
269 | err = ext4_journal_restart(handle, 3); | |
270 | if (err != 0) { | |
12062ddd | 271 | ext4_warning(inode->i_sb, |
bc965ab3 TT |
272 | "couldn't extend journal (err %d)", err); |
273 | stop_handle: | |
274 | ext4_journal_stop(handle); | |
45388219 | 275 | ext4_orphan_del(NULL, inode); |
bc965ab3 TT |
276 | goto no_delete; |
277 | } | |
278 | } | |
279 | ||
ac27a0ec | 280 | /* |
617ba13b | 281 | * Kill off the orphan record which ext4_truncate created. |
ac27a0ec | 282 | * AKPM: I think this can be inside the above `if'. |
617ba13b | 283 | * Note that ext4_orphan_del() has to be able to cope with the |
ac27a0ec | 284 | * deletion of a non-existent orphan - this is because we don't |
617ba13b | 285 | * know if ext4_truncate() actually created an orphan record. |
ac27a0ec DK |
286 | * (Well, we could do this if we need to, but heck - it works) |
287 | */ | |
617ba13b MC |
288 | ext4_orphan_del(handle, inode); |
289 | EXT4_I(inode)->i_dtime = get_seconds(); | |
ac27a0ec DK |
290 | |
291 | /* | |
292 | * One subtle ordering requirement: if anything has gone wrong | |
293 | * (transaction abort, IO errors, whatever), then we can still | |
294 | * do these next steps (the fs will already have been marked as | |
295 | * having errors), but we can't free the inode if the mark_dirty | |
296 | * fails. | |
297 | */ | |
617ba13b | 298 | if (ext4_mark_inode_dirty(handle, inode)) |
ac27a0ec | 299 | /* If that failed, just do the required in-core inode clear. */ |
0930fcc1 | 300 | ext4_clear_inode(inode); |
ac27a0ec | 301 | else |
617ba13b MC |
302 | ext4_free_inode(handle, inode); |
303 | ext4_journal_stop(handle); | |
ac27a0ec DK |
304 | return; |
305 | no_delete: | |
0930fcc1 | 306 | ext4_clear_inode(inode); /* We must guarantee clearing of inode... */ |
ac27a0ec DK |
307 | } |
308 | ||
a9e7f447 DM |
309 | #ifdef CONFIG_QUOTA |
310 | qsize_t *ext4_get_reserved_space(struct inode *inode) | |
60e58e0f | 311 | { |
a9e7f447 | 312 | return &EXT4_I(inode)->i_reserved_quota; |
60e58e0f | 313 | } |
a9e7f447 | 314 | #endif |
9d0be502 | 315 | |
12219aea AK |
316 | /* |
317 | * Calculate the number of metadata blocks need to reserve | |
9d0be502 | 318 | * to allocate a block located at @lblock |
12219aea | 319 | */ |
01f49d0b | 320 | static int ext4_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock) |
12219aea | 321 | { |
12e9b892 | 322 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
9d0be502 | 323 | return ext4_ext_calc_metadata_amount(inode, lblock); |
12219aea | 324 | |
8bb2b247 | 325 | return ext4_ind_calc_metadata_amount(inode, lblock); |
12219aea AK |
326 | } |
327 | ||
0637c6f4 TT |
328 | /* |
329 | * Called with i_data_sem down, which is important since we can call | |
330 | * ext4_discard_preallocations() from here. | |
331 | */ | |
5f634d06 AK |
332 | void ext4_da_update_reserve_space(struct inode *inode, |
333 | int used, int quota_claim) | |
12219aea AK |
334 | { |
335 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0637c6f4 | 336 | struct ext4_inode_info *ei = EXT4_I(inode); |
0637c6f4 TT |
337 | |
338 | spin_lock(&ei->i_block_reservation_lock); | |
d8990240 | 339 | trace_ext4_da_update_reserve_space(inode, used, quota_claim); |
0637c6f4 TT |
340 | if (unlikely(used > ei->i_reserved_data_blocks)) { |
341 | ext4_msg(inode->i_sb, KERN_NOTICE, "%s: ino %lu, used %d " | |
1084f252 | 342 | "with only %d reserved data blocks", |
0637c6f4 TT |
343 | __func__, inode->i_ino, used, |
344 | ei->i_reserved_data_blocks); | |
345 | WARN_ON(1); | |
346 | used = ei->i_reserved_data_blocks; | |
347 | } | |
12219aea | 348 | |
0637c6f4 TT |
349 | /* Update per-inode reservations */ |
350 | ei->i_reserved_data_blocks -= used; | |
0637c6f4 | 351 | ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks; |
57042651 | 352 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, |
72b8ab9d | 353 | used + ei->i_allocated_meta_blocks); |
0637c6f4 | 354 | ei->i_allocated_meta_blocks = 0; |
6bc6e63f | 355 | |
0637c6f4 TT |
356 | if (ei->i_reserved_data_blocks == 0) { |
357 | /* | |
358 | * We can release all of the reserved metadata blocks | |
359 | * only when we have written all of the delayed | |
360 | * allocation blocks. | |
361 | */ | |
57042651 | 362 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, |
72b8ab9d | 363 | ei->i_reserved_meta_blocks); |
ee5f4d9c | 364 | ei->i_reserved_meta_blocks = 0; |
9d0be502 | 365 | ei->i_da_metadata_calc_len = 0; |
6bc6e63f | 366 | } |
12219aea | 367 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
60e58e0f | 368 | |
72b8ab9d ES |
369 | /* Update quota subsystem for data blocks */ |
370 | if (quota_claim) | |
7b415bf6 | 371 | dquot_claim_block(inode, EXT4_C2B(sbi, used)); |
72b8ab9d | 372 | else { |
5f634d06 AK |
373 | /* |
374 | * We did fallocate with an offset that is already delayed | |
375 | * allocated. So on delayed allocated writeback we should | |
72b8ab9d | 376 | * not re-claim the quota for fallocated blocks. |
5f634d06 | 377 | */ |
7b415bf6 | 378 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, used)); |
5f634d06 | 379 | } |
d6014301 AK |
380 | |
381 | /* | |
382 | * If we have done all the pending block allocations and if | |
383 | * there aren't any writers on the inode, we can discard the | |
384 | * inode's preallocations. | |
385 | */ | |
0637c6f4 TT |
386 | if ((ei->i_reserved_data_blocks == 0) && |
387 | (atomic_read(&inode->i_writecount) == 0)) | |
d6014301 | 388 | ext4_discard_preallocations(inode); |
12219aea AK |
389 | } |
390 | ||
e29136f8 | 391 | static int __check_block_validity(struct inode *inode, const char *func, |
c398eda0 TT |
392 | unsigned int line, |
393 | struct ext4_map_blocks *map) | |
6fd058f7 | 394 | { |
24676da4 TT |
395 | if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk, |
396 | map->m_len)) { | |
c398eda0 TT |
397 | ext4_error_inode(inode, func, line, map->m_pblk, |
398 | "lblock %lu mapped to illegal pblock " | |
399 | "(length %d)", (unsigned long) map->m_lblk, | |
400 | map->m_len); | |
6fd058f7 TT |
401 | return -EIO; |
402 | } | |
403 | return 0; | |
404 | } | |
405 | ||
e29136f8 | 406 | #define check_block_validity(inode, map) \ |
c398eda0 | 407 | __check_block_validity((inode), __func__, __LINE__, (map)) |
e29136f8 | 408 | |
55138e0b | 409 | /* |
1f94533d TT |
410 | * Return the number of contiguous dirty pages in a given inode |
411 | * starting at page frame idx. | |
55138e0b TT |
412 | */ |
413 | static pgoff_t ext4_num_dirty_pages(struct inode *inode, pgoff_t idx, | |
414 | unsigned int max_pages) | |
415 | { | |
416 | struct address_space *mapping = inode->i_mapping; | |
417 | pgoff_t index; | |
418 | struct pagevec pvec; | |
419 | pgoff_t num = 0; | |
420 | int i, nr_pages, done = 0; | |
421 | ||
422 | if (max_pages == 0) | |
423 | return 0; | |
424 | pagevec_init(&pvec, 0); | |
425 | while (!done) { | |
426 | index = idx; | |
427 | nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, | |
428 | PAGECACHE_TAG_DIRTY, | |
429 | (pgoff_t)PAGEVEC_SIZE); | |
430 | if (nr_pages == 0) | |
431 | break; | |
432 | for (i = 0; i < nr_pages; i++) { | |
433 | struct page *page = pvec.pages[i]; | |
434 | struct buffer_head *bh, *head; | |
435 | ||
436 | lock_page(page); | |
437 | if (unlikely(page->mapping != mapping) || | |
438 | !PageDirty(page) || | |
439 | PageWriteback(page) || | |
440 | page->index != idx) { | |
441 | done = 1; | |
442 | unlock_page(page); | |
443 | break; | |
444 | } | |
1f94533d TT |
445 | if (page_has_buffers(page)) { |
446 | bh = head = page_buffers(page); | |
447 | do { | |
448 | if (!buffer_delay(bh) && | |
449 | !buffer_unwritten(bh)) | |
450 | done = 1; | |
451 | bh = bh->b_this_page; | |
452 | } while (!done && (bh != head)); | |
453 | } | |
55138e0b TT |
454 | unlock_page(page); |
455 | if (done) | |
456 | break; | |
457 | idx++; | |
458 | num++; | |
659c6009 ES |
459 | if (num >= max_pages) { |
460 | done = 1; | |
55138e0b | 461 | break; |
659c6009 | 462 | } |
55138e0b TT |
463 | } |
464 | pagevec_release(&pvec); | |
465 | } | |
466 | return num; | |
467 | } | |
468 | ||
5356f261 AK |
469 | /* |
470 | * Sets the BH_Da_Mapped bit on the buffer heads corresponding to the given map. | |
471 | */ | |
472 | static void set_buffers_da_mapped(struct inode *inode, | |
473 | struct ext4_map_blocks *map) | |
474 | { | |
475 | struct address_space *mapping = inode->i_mapping; | |
476 | struct pagevec pvec; | |
477 | int i, nr_pages; | |
478 | pgoff_t index, end; | |
479 | ||
480 | index = map->m_lblk >> (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
481 | end = (map->m_lblk + map->m_len - 1) >> | |
482 | (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
483 | ||
484 | pagevec_init(&pvec, 0); | |
485 | while (index <= end) { | |
486 | nr_pages = pagevec_lookup(&pvec, mapping, index, | |
487 | min(end - index + 1, | |
488 | (pgoff_t)PAGEVEC_SIZE)); | |
489 | if (nr_pages == 0) | |
490 | break; | |
491 | for (i = 0; i < nr_pages; i++) { | |
492 | struct page *page = pvec.pages[i]; | |
493 | struct buffer_head *bh, *head; | |
494 | ||
495 | if (unlikely(page->mapping != mapping) || | |
496 | !PageDirty(page)) | |
497 | break; | |
498 | ||
499 | if (page_has_buffers(page)) { | |
500 | bh = head = page_buffers(page); | |
501 | do { | |
502 | set_buffer_da_mapped(bh); | |
503 | bh = bh->b_this_page; | |
504 | } while (bh != head); | |
505 | } | |
506 | index++; | |
507 | } | |
508 | pagevec_release(&pvec); | |
509 | } | |
510 | } | |
511 | ||
f5ab0d1f | 512 | /* |
e35fd660 | 513 | * The ext4_map_blocks() function tries to look up the requested blocks, |
2b2d6d01 | 514 | * and returns if the blocks are already mapped. |
f5ab0d1f | 515 | * |
f5ab0d1f MC |
516 | * Otherwise it takes the write lock of the i_data_sem and allocate blocks |
517 | * and store the allocated blocks in the result buffer head and mark it | |
518 | * mapped. | |
519 | * | |
e35fd660 TT |
520 | * If file type is extents based, it will call ext4_ext_map_blocks(), |
521 | * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping | |
f5ab0d1f MC |
522 | * based files |
523 | * | |
524 | * On success, it returns the number of blocks being mapped or allocate. | |
525 | * if create==0 and the blocks are pre-allocated and uninitialized block, | |
526 | * the result buffer head is unmapped. If the create ==1, it will make sure | |
527 | * the buffer head is mapped. | |
528 | * | |
529 | * It returns 0 if plain look up failed (blocks have not been allocated), in | |
df3ab170 | 530 | * that case, buffer head is unmapped |
f5ab0d1f MC |
531 | * |
532 | * It returns the error in case of allocation failure. | |
533 | */ | |
e35fd660 TT |
534 | int ext4_map_blocks(handle_t *handle, struct inode *inode, |
535 | struct ext4_map_blocks *map, int flags) | |
0e855ac8 AK |
536 | { |
537 | int retval; | |
f5ab0d1f | 538 | |
e35fd660 TT |
539 | map->m_flags = 0; |
540 | ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u," | |
541 | "logical block %lu\n", inode->i_ino, flags, map->m_len, | |
542 | (unsigned long) map->m_lblk); | |
4df3d265 | 543 | /* |
b920c755 TT |
544 | * Try to see if we can get the block without requesting a new |
545 | * file system block. | |
4df3d265 AK |
546 | */ |
547 | down_read((&EXT4_I(inode)->i_data_sem)); | |
12e9b892 | 548 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
a4e5d88b DM |
549 | retval = ext4_ext_map_blocks(handle, inode, map, flags & |
550 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
0e855ac8 | 551 | } else { |
a4e5d88b DM |
552 | retval = ext4_ind_map_blocks(handle, inode, map, flags & |
553 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
0e855ac8 | 554 | } |
4df3d265 | 555 | up_read((&EXT4_I(inode)->i_data_sem)); |
f5ab0d1f | 556 | |
e35fd660 | 557 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
e29136f8 | 558 | int ret = check_block_validity(inode, map); |
6fd058f7 TT |
559 | if (ret != 0) |
560 | return ret; | |
561 | } | |
562 | ||
f5ab0d1f | 563 | /* If it is only a block(s) look up */ |
c2177057 | 564 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) |
f5ab0d1f MC |
565 | return retval; |
566 | ||
567 | /* | |
568 | * Returns if the blocks have already allocated | |
569 | * | |
570 | * Note that if blocks have been preallocated | |
df3ab170 | 571 | * ext4_ext_get_block() returns the create = 0 |
f5ab0d1f MC |
572 | * with buffer head unmapped. |
573 | */ | |
e35fd660 | 574 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) |
4df3d265 AK |
575 | return retval; |
576 | ||
2a8964d6 AK |
577 | /* |
578 | * When we call get_blocks without the create flag, the | |
579 | * BH_Unwritten flag could have gotten set if the blocks | |
580 | * requested were part of a uninitialized extent. We need to | |
581 | * clear this flag now that we are committed to convert all or | |
582 | * part of the uninitialized extent to be an initialized | |
583 | * extent. This is because we need to avoid the combination | |
584 | * of BH_Unwritten and BH_Mapped flags being simultaneously | |
585 | * set on the buffer_head. | |
586 | */ | |
e35fd660 | 587 | map->m_flags &= ~EXT4_MAP_UNWRITTEN; |
2a8964d6 | 588 | |
4df3d265 | 589 | /* |
f5ab0d1f MC |
590 | * New blocks allocate and/or writing to uninitialized extent |
591 | * will possibly result in updating i_data, so we take | |
592 | * the write lock of i_data_sem, and call get_blocks() | |
593 | * with create == 1 flag. | |
4df3d265 AK |
594 | */ |
595 | down_write((&EXT4_I(inode)->i_data_sem)); | |
d2a17637 MC |
596 | |
597 | /* | |
598 | * if the caller is from delayed allocation writeout path | |
599 | * we have already reserved fs blocks for allocation | |
600 | * let the underlying get_block() function know to | |
601 | * avoid double accounting | |
602 | */ | |
c2177057 | 603 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) |
f2321097 | 604 | ext4_set_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED); |
4df3d265 AK |
605 | /* |
606 | * We need to check for EXT4 here because migrate | |
607 | * could have changed the inode type in between | |
608 | */ | |
12e9b892 | 609 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
e35fd660 | 610 | retval = ext4_ext_map_blocks(handle, inode, map, flags); |
0e855ac8 | 611 | } else { |
e35fd660 | 612 | retval = ext4_ind_map_blocks(handle, inode, map, flags); |
267e4db9 | 613 | |
e35fd660 | 614 | if (retval > 0 && map->m_flags & EXT4_MAP_NEW) { |
267e4db9 AK |
615 | /* |
616 | * We allocated new blocks which will result in | |
617 | * i_data's format changing. Force the migrate | |
618 | * to fail by clearing migrate flags | |
619 | */ | |
19f5fb7a | 620 | ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE); |
267e4db9 | 621 | } |
d2a17637 | 622 | |
5f634d06 AK |
623 | /* |
624 | * Update reserved blocks/metadata blocks after successful | |
625 | * block allocation which had been deferred till now. We don't | |
626 | * support fallocate for non extent files. So we can update | |
627 | * reserve space here. | |
628 | */ | |
629 | if ((retval > 0) && | |
1296cc85 | 630 | (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)) |
5f634d06 AK |
631 | ext4_da_update_reserve_space(inode, retval, 1); |
632 | } | |
5356f261 | 633 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) { |
f2321097 | 634 | ext4_clear_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED); |
2ac3b6e0 | 635 | |
5356f261 AK |
636 | /* If we have successfully mapped the delayed allocated blocks, |
637 | * set the BH_Da_Mapped bit on them. Its important to do this | |
638 | * under the protection of i_data_sem. | |
639 | */ | |
640 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) | |
641 | set_buffers_da_mapped(inode, map); | |
642 | } | |
643 | ||
4df3d265 | 644 | up_write((&EXT4_I(inode)->i_data_sem)); |
e35fd660 | 645 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
e29136f8 | 646 | int ret = check_block_validity(inode, map); |
6fd058f7 TT |
647 | if (ret != 0) |
648 | return ret; | |
649 | } | |
0e855ac8 AK |
650 | return retval; |
651 | } | |
652 | ||
f3bd1f3f MC |
653 | /* Maximum number of blocks we map for direct IO at once. */ |
654 | #define DIO_MAX_BLOCKS 4096 | |
655 | ||
2ed88685 TT |
656 | static int _ext4_get_block(struct inode *inode, sector_t iblock, |
657 | struct buffer_head *bh, int flags) | |
ac27a0ec | 658 | { |
3e4fdaf8 | 659 | handle_t *handle = ext4_journal_current_handle(); |
2ed88685 | 660 | struct ext4_map_blocks map; |
7fb5409d | 661 | int ret = 0, started = 0; |
f3bd1f3f | 662 | int dio_credits; |
ac27a0ec | 663 | |
2ed88685 TT |
664 | map.m_lblk = iblock; |
665 | map.m_len = bh->b_size >> inode->i_blkbits; | |
666 | ||
667 | if (flags && !handle) { | |
7fb5409d | 668 | /* Direct IO write... */ |
2ed88685 TT |
669 | if (map.m_len > DIO_MAX_BLOCKS) |
670 | map.m_len = DIO_MAX_BLOCKS; | |
671 | dio_credits = ext4_chunk_trans_blocks(inode, map.m_len); | |
f3bd1f3f | 672 | handle = ext4_journal_start(inode, dio_credits); |
7fb5409d | 673 | if (IS_ERR(handle)) { |
ac27a0ec | 674 | ret = PTR_ERR(handle); |
2ed88685 | 675 | return ret; |
ac27a0ec | 676 | } |
7fb5409d | 677 | started = 1; |
ac27a0ec DK |
678 | } |
679 | ||
2ed88685 | 680 | ret = ext4_map_blocks(handle, inode, &map, flags); |
7fb5409d | 681 | if (ret > 0) { |
2ed88685 TT |
682 | map_bh(bh, inode->i_sb, map.m_pblk); |
683 | bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags; | |
684 | bh->b_size = inode->i_sb->s_blocksize * map.m_len; | |
7fb5409d | 685 | ret = 0; |
ac27a0ec | 686 | } |
7fb5409d JK |
687 | if (started) |
688 | ext4_journal_stop(handle); | |
ac27a0ec DK |
689 | return ret; |
690 | } | |
691 | ||
2ed88685 TT |
692 | int ext4_get_block(struct inode *inode, sector_t iblock, |
693 | struct buffer_head *bh, int create) | |
694 | { | |
695 | return _ext4_get_block(inode, iblock, bh, | |
696 | create ? EXT4_GET_BLOCKS_CREATE : 0); | |
697 | } | |
698 | ||
ac27a0ec DK |
699 | /* |
700 | * `handle' can be NULL if create is zero | |
701 | */ | |
617ba13b | 702 | struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode, |
725d26d3 | 703 | ext4_lblk_t block, int create, int *errp) |
ac27a0ec | 704 | { |
2ed88685 TT |
705 | struct ext4_map_blocks map; |
706 | struct buffer_head *bh; | |
ac27a0ec DK |
707 | int fatal = 0, err; |
708 | ||
709 | J_ASSERT(handle != NULL || create == 0); | |
710 | ||
2ed88685 TT |
711 | map.m_lblk = block; |
712 | map.m_len = 1; | |
713 | err = ext4_map_blocks(handle, inode, &map, | |
714 | create ? EXT4_GET_BLOCKS_CREATE : 0); | |
ac27a0ec | 715 | |
2ed88685 TT |
716 | if (err < 0) |
717 | *errp = err; | |
718 | if (err <= 0) | |
719 | return NULL; | |
720 | *errp = 0; | |
721 | ||
722 | bh = sb_getblk(inode->i_sb, map.m_pblk); | |
723 | if (!bh) { | |
724 | *errp = -EIO; | |
725 | return NULL; | |
ac27a0ec | 726 | } |
2ed88685 TT |
727 | if (map.m_flags & EXT4_MAP_NEW) { |
728 | J_ASSERT(create != 0); | |
729 | J_ASSERT(handle != NULL); | |
ac27a0ec | 730 | |
2ed88685 TT |
731 | /* |
732 | * Now that we do not always journal data, we should | |
733 | * keep in mind whether this should always journal the | |
734 | * new buffer as metadata. For now, regular file | |
735 | * writes use ext4_get_block instead, so it's not a | |
736 | * problem. | |
737 | */ | |
738 | lock_buffer(bh); | |
739 | BUFFER_TRACE(bh, "call get_create_access"); | |
740 | fatal = ext4_journal_get_create_access(handle, bh); | |
741 | if (!fatal && !buffer_uptodate(bh)) { | |
742 | memset(bh->b_data, 0, inode->i_sb->s_blocksize); | |
743 | set_buffer_uptodate(bh); | |
ac27a0ec | 744 | } |
2ed88685 TT |
745 | unlock_buffer(bh); |
746 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); | |
747 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
748 | if (!fatal) | |
749 | fatal = err; | |
750 | } else { | |
751 | BUFFER_TRACE(bh, "not a new buffer"); | |
ac27a0ec | 752 | } |
2ed88685 TT |
753 | if (fatal) { |
754 | *errp = fatal; | |
755 | brelse(bh); | |
756 | bh = NULL; | |
757 | } | |
758 | return bh; | |
ac27a0ec DK |
759 | } |
760 | ||
617ba13b | 761 | struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode, |
725d26d3 | 762 | ext4_lblk_t block, int create, int *err) |
ac27a0ec | 763 | { |
af5bc92d | 764 | struct buffer_head *bh; |
ac27a0ec | 765 | |
617ba13b | 766 | bh = ext4_getblk(handle, inode, block, create, err); |
ac27a0ec DK |
767 | if (!bh) |
768 | return bh; | |
769 | if (buffer_uptodate(bh)) | |
770 | return bh; | |
65299a3b | 771 | ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh); |
ac27a0ec DK |
772 | wait_on_buffer(bh); |
773 | if (buffer_uptodate(bh)) | |
774 | return bh; | |
775 | put_bh(bh); | |
776 | *err = -EIO; | |
777 | return NULL; | |
778 | } | |
779 | ||
af5bc92d TT |
780 | static int walk_page_buffers(handle_t *handle, |
781 | struct buffer_head *head, | |
782 | unsigned from, | |
783 | unsigned to, | |
784 | int *partial, | |
785 | int (*fn)(handle_t *handle, | |
786 | struct buffer_head *bh)) | |
ac27a0ec DK |
787 | { |
788 | struct buffer_head *bh; | |
789 | unsigned block_start, block_end; | |
790 | unsigned blocksize = head->b_size; | |
791 | int err, ret = 0; | |
792 | struct buffer_head *next; | |
793 | ||
af5bc92d TT |
794 | for (bh = head, block_start = 0; |
795 | ret == 0 && (bh != head || !block_start); | |
de9a55b8 | 796 | block_start = block_end, bh = next) { |
ac27a0ec DK |
797 | next = bh->b_this_page; |
798 | block_end = block_start + blocksize; | |
799 | if (block_end <= from || block_start >= to) { | |
800 | if (partial && !buffer_uptodate(bh)) | |
801 | *partial = 1; | |
802 | continue; | |
803 | } | |
804 | err = (*fn)(handle, bh); | |
805 | if (!ret) | |
806 | ret = err; | |
807 | } | |
808 | return ret; | |
809 | } | |
810 | ||
811 | /* | |
812 | * To preserve ordering, it is essential that the hole instantiation and | |
813 | * the data write be encapsulated in a single transaction. We cannot | |
617ba13b | 814 | * close off a transaction and start a new one between the ext4_get_block() |
dab291af | 815 | * and the commit_write(). So doing the jbd2_journal_start at the start of |
ac27a0ec DK |
816 | * prepare_write() is the right place. |
817 | * | |
617ba13b MC |
818 | * Also, this function can nest inside ext4_writepage() -> |
819 | * block_write_full_page(). In that case, we *know* that ext4_writepage() | |
ac27a0ec DK |
820 | * has generated enough buffer credits to do the whole page. So we won't |
821 | * block on the journal in that case, which is good, because the caller may | |
822 | * be PF_MEMALLOC. | |
823 | * | |
617ba13b | 824 | * By accident, ext4 can be reentered when a transaction is open via |
ac27a0ec DK |
825 | * quota file writes. If we were to commit the transaction while thus |
826 | * reentered, there can be a deadlock - we would be holding a quota | |
827 | * lock, and the commit would never complete if another thread had a | |
828 | * transaction open and was blocking on the quota lock - a ranking | |
829 | * violation. | |
830 | * | |
dab291af | 831 | * So what we do is to rely on the fact that jbd2_journal_stop/journal_start |
ac27a0ec DK |
832 | * will _not_ run commit under these circumstances because handle->h_ref |
833 | * is elevated. We'll still have enough credits for the tiny quotafile | |
834 | * write. | |
835 | */ | |
836 | static int do_journal_get_write_access(handle_t *handle, | |
de9a55b8 | 837 | struct buffer_head *bh) |
ac27a0ec | 838 | { |
56d35a4c JK |
839 | int dirty = buffer_dirty(bh); |
840 | int ret; | |
841 | ||
ac27a0ec DK |
842 | if (!buffer_mapped(bh) || buffer_freed(bh)) |
843 | return 0; | |
56d35a4c | 844 | /* |
ebdec241 | 845 | * __block_write_begin() could have dirtied some buffers. Clean |
56d35a4c JK |
846 | * the dirty bit as jbd2_journal_get_write_access() could complain |
847 | * otherwise about fs integrity issues. Setting of the dirty bit | |
ebdec241 | 848 | * by __block_write_begin() isn't a real problem here as we clear |
56d35a4c JK |
849 | * the bit before releasing a page lock and thus writeback cannot |
850 | * ever write the buffer. | |
851 | */ | |
852 | if (dirty) | |
853 | clear_buffer_dirty(bh); | |
854 | ret = ext4_journal_get_write_access(handle, bh); | |
855 | if (!ret && dirty) | |
856 | ret = ext4_handle_dirty_metadata(handle, NULL, bh); | |
857 | return ret; | |
ac27a0ec DK |
858 | } |
859 | ||
744692dc JZ |
860 | static int ext4_get_block_write(struct inode *inode, sector_t iblock, |
861 | struct buffer_head *bh_result, int create); | |
bfc1af65 | 862 | static int ext4_write_begin(struct file *file, struct address_space *mapping, |
de9a55b8 TT |
863 | loff_t pos, unsigned len, unsigned flags, |
864 | struct page **pagep, void **fsdata) | |
ac27a0ec | 865 | { |
af5bc92d | 866 | struct inode *inode = mapping->host; |
1938a150 | 867 | int ret, needed_blocks; |
ac27a0ec DK |
868 | handle_t *handle; |
869 | int retries = 0; | |
af5bc92d | 870 | struct page *page; |
de9a55b8 | 871 | pgoff_t index; |
af5bc92d | 872 | unsigned from, to; |
bfc1af65 | 873 | |
9bffad1e | 874 | trace_ext4_write_begin(inode, pos, len, flags); |
1938a150 AK |
875 | /* |
876 | * Reserve one block more for addition to orphan list in case | |
877 | * we allocate blocks but write fails for some reason | |
878 | */ | |
879 | needed_blocks = ext4_writepage_trans_blocks(inode) + 1; | |
de9a55b8 | 880 | index = pos >> PAGE_CACHE_SHIFT; |
af5bc92d TT |
881 | from = pos & (PAGE_CACHE_SIZE - 1); |
882 | to = from + len; | |
ac27a0ec DK |
883 | |
884 | retry: | |
af5bc92d TT |
885 | handle = ext4_journal_start(inode, needed_blocks); |
886 | if (IS_ERR(handle)) { | |
887 | ret = PTR_ERR(handle); | |
888 | goto out; | |
7479d2b9 | 889 | } |
ac27a0ec | 890 | |
ebd3610b JK |
891 | /* We cannot recurse into the filesystem as the transaction is already |
892 | * started */ | |
893 | flags |= AOP_FLAG_NOFS; | |
894 | ||
54566b2c | 895 | page = grab_cache_page_write_begin(mapping, index, flags); |
cf108bca JK |
896 | if (!page) { |
897 | ext4_journal_stop(handle); | |
898 | ret = -ENOMEM; | |
899 | goto out; | |
900 | } | |
901 | *pagep = page; | |
902 | ||
744692dc | 903 | if (ext4_should_dioread_nolock(inode)) |
6e1db88d | 904 | ret = __block_write_begin(page, pos, len, ext4_get_block_write); |
744692dc | 905 | else |
6e1db88d | 906 | ret = __block_write_begin(page, pos, len, ext4_get_block); |
bfc1af65 NP |
907 | |
908 | if (!ret && ext4_should_journal_data(inode)) { | |
ac27a0ec DK |
909 | ret = walk_page_buffers(handle, page_buffers(page), |
910 | from, to, NULL, do_journal_get_write_access); | |
911 | } | |
bfc1af65 NP |
912 | |
913 | if (ret) { | |
af5bc92d | 914 | unlock_page(page); |
af5bc92d | 915 | page_cache_release(page); |
ae4d5372 | 916 | /* |
6e1db88d | 917 | * __block_write_begin may have instantiated a few blocks |
ae4d5372 AK |
918 | * outside i_size. Trim these off again. Don't need |
919 | * i_size_read because we hold i_mutex. | |
1938a150 AK |
920 | * |
921 | * Add inode to orphan list in case we crash before | |
922 | * truncate finishes | |
ae4d5372 | 923 | */ |
ffacfa7a | 924 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
1938a150 AK |
925 | ext4_orphan_add(handle, inode); |
926 | ||
927 | ext4_journal_stop(handle); | |
928 | if (pos + len > inode->i_size) { | |
b9a4207d | 929 | ext4_truncate_failed_write(inode); |
de9a55b8 | 930 | /* |
ffacfa7a | 931 | * If truncate failed early the inode might |
1938a150 AK |
932 | * still be on the orphan list; we need to |
933 | * make sure the inode is removed from the | |
934 | * orphan list in that case. | |
935 | */ | |
936 | if (inode->i_nlink) | |
937 | ext4_orphan_del(NULL, inode); | |
938 | } | |
bfc1af65 NP |
939 | } |
940 | ||
617ba13b | 941 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) |
ac27a0ec | 942 | goto retry; |
7479d2b9 | 943 | out: |
ac27a0ec DK |
944 | return ret; |
945 | } | |
946 | ||
bfc1af65 NP |
947 | /* For write_end() in data=journal mode */ |
948 | static int write_end_fn(handle_t *handle, struct buffer_head *bh) | |
ac27a0ec DK |
949 | { |
950 | if (!buffer_mapped(bh) || buffer_freed(bh)) | |
951 | return 0; | |
952 | set_buffer_uptodate(bh); | |
0390131b | 953 | return ext4_handle_dirty_metadata(handle, NULL, bh); |
ac27a0ec DK |
954 | } |
955 | ||
f8514083 | 956 | static int ext4_generic_write_end(struct file *file, |
de9a55b8 TT |
957 | struct address_space *mapping, |
958 | loff_t pos, unsigned len, unsigned copied, | |
959 | struct page *page, void *fsdata) | |
f8514083 AK |
960 | { |
961 | int i_size_changed = 0; | |
962 | struct inode *inode = mapping->host; | |
963 | handle_t *handle = ext4_journal_current_handle(); | |
964 | ||
965 | copied = block_write_end(file, mapping, pos, len, copied, page, fsdata); | |
966 | ||
967 | /* | |
968 | * No need to use i_size_read() here, the i_size | |
969 | * cannot change under us because we hold i_mutex. | |
970 | * | |
971 | * But it's important to update i_size while still holding page lock: | |
972 | * page writeout could otherwise come in and zero beyond i_size. | |
973 | */ | |
974 | if (pos + copied > inode->i_size) { | |
975 | i_size_write(inode, pos + copied); | |
976 | i_size_changed = 1; | |
977 | } | |
978 | ||
979 | if (pos + copied > EXT4_I(inode)->i_disksize) { | |
980 | /* We need to mark inode dirty even if | |
981 | * new_i_size is less that inode->i_size | |
982 | * bu greater than i_disksize.(hint delalloc) | |
983 | */ | |
984 | ext4_update_i_disksize(inode, (pos + copied)); | |
985 | i_size_changed = 1; | |
986 | } | |
987 | unlock_page(page); | |
988 | page_cache_release(page); | |
989 | ||
990 | /* | |
991 | * Don't mark the inode dirty under page lock. First, it unnecessarily | |
992 | * makes the holding time of page lock longer. Second, it forces lock | |
993 | * ordering of page lock and transaction start for journaling | |
994 | * filesystems. | |
995 | */ | |
996 | if (i_size_changed) | |
997 | ext4_mark_inode_dirty(handle, inode); | |
998 | ||
999 | return copied; | |
1000 | } | |
1001 | ||
ac27a0ec DK |
1002 | /* |
1003 | * We need to pick up the new inode size which generic_commit_write gave us | |
1004 | * `file' can be NULL - eg, when called from page_symlink(). | |
1005 | * | |
617ba13b | 1006 | * ext4 never places buffers on inode->i_mapping->private_list. metadata |
ac27a0ec DK |
1007 | * buffers are managed internally. |
1008 | */ | |
bfc1af65 | 1009 | static int ext4_ordered_write_end(struct file *file, |
de9a55b8 TT |
1010 | struct address_space *mapping, |
1011 | loff_t pos, unsigned len, unsigned copied, | |
1012 | struct page *page, void *fsdata) | |
ac27a0ec | 1013 | { |
617ba13b | 1014 | handle_t *handle = ext4_journal_current_handle(); |
cf108bca | 1015 | struct inode *inode = mapping->host; |
ac27a0ec DK |
1016 | int ret = 0, ret2; |
1017 | ||
9bffad1e | 1018 | trace_ext4_ordered_write_end(inode, pos, len, copied); |
678aaf48 | 1019 | ret = ext4_jbd2_file_inode(handle, inode); |
ac27a0ec DK |
1020 | |
1021 | if (ret == 0) { | |
f8514083 | 1022 | ret2 = ext4_generic_write_end(file, mapping, pos, len, copied, |
bfc1af65 | 1023 | page, fsdata); |
f8a87d89 | 1024 | copied = ret2; |
ffacfa7a | 1025 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
f8514083 AK |
1026 | /* if we have allocated more blocks and copied |
1027 | * less. We will have blocks allocated outside | |
1028 | * inode->i_size. So truncate them | |
1029 | */ | |
1030 | ext4_orphan_add(handle, inode); | |
f8a87d89 RK |
1031 | if (ret2 < 0) |
1032 | ret = ret2; | |
09e0834f AF |
1033 | } else { |
1034 | unlock_page(page); | |
1035 | page_cache_release(page); | |
ac27a0ec | 1036 | } |
09e0834f | 1037 | |
617ba13b | 1038 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1039 | if (!ret) |
1040 | ret = ret2; | |
bfc1af65 | 1041 | |
f8514083 | 1042 | if (pos + len > inode->i_size) { |
b9a4207d | 1043 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1044 | /* |
ffacfa7a | 1045 | * If truncate failed early the inode might still be |
f8514083 AK |
1046 | * on the orphan list; we need to make sure the inode |
1047 | * is removed from the orphan list in that case. | |
1048 | */ | |
1049 | if (inode->i_nlink) | |
1050 | ext4_orphan_del(NULL, inode); | |
1051 | } | |
1052 | ||
1053 | ||
bfc1af65 | 1054 | return ret ? ret : copied; |
ac27a0ec DK |
1055 | } |
1056 | ||
bfc1af65 | 1057 | static int ext4_writeback_write_end(struct file *file, |
de9a55b8 TT |
1058 | struct address_space *mapping, |
1059 | loff_t pos, unsigned len, unsigned copied, | |
1060 | struct page *page, void *fsdata) | |
ac27a0ec | 1061 | { |
617ba13b | 1062 | handle_t *handle = ext4_journal_current_handle(); |
cf108bca | 1063 | struct inode *inode = mapping->host; |
ac27a0ec | 1064 | int ret = 0, ret2; |
ac27a0ec | 1065 | |
9bffad1e | 1066 | trace_ext4_writeback_write_end(inode, pos, len, copied); |
f8514083 | 1067 | ret2 = ext4_generic_write_end(file, mapping, pos, len, copied, |
bfc1af65 | 1068 | page, fsdata); |
f8a87d89 | 1069 | copied = ret2; |
ffacfa7a | 1070 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
f8514083 AK |
1071 | /* if we have allocated more blocks and copied |
1072 | * less. We will have blocks allocated outside | |
1073 | * inode->i_size. So truncate them | |
1074 | */ | |
1075 | ext4_orphan_add(handle, inode); | |
1076 | ||
f8a87d89 RK |
1077 | if (ret2 < 0) |
1078 | ret = ret2; | |
ac27a0ec | 1079 | |
617ba13b | 1080 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1081 | if (!ret) |
1082 | ret = ret2; | |
bfc1af65 | 1083 | |
f8514083 | 1084 | if (pos + len > inode->i_size) { |
b9a4207d | 1085 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1086 | /* |
ffacfa7a | 1087 | * If truncate failed early the inode might still be |
f8514083 AK |
1088 | * on the orphan list; we need to make sure the inode |
1089 | * is removed from the orphan list in that case. | |
1090 | */ | |
1091 | if (inode->i_nlink) | |
1092 | ext4_orphan_del(NULL, inode); | |
1093 | } | |
1094 | ||
bfc1af65 | 1095 | return ret ? ret : copied; |
ac27a0ec DK |
1096 | } |
1097 | ||
bfc1af65 | 1098 | static int ext4_journalled_write_end(struct file *file, |
de9a55b8 TT |
1099 | struct address_space *mapping, |
1100 | loff_t pos, unsigned len, unsigned copied, | |
1101 | struct page *page, void *fsdata) | |
ac27a0ec | 1102 | { |
617ba13b | 1103 | handle_t *handle = ext4_journal_current_handle(); |
bfc1af65 | 1104 | struct inode *inode = mapping->host; |
ac27a0ec DK |
1105 | int ret = 0, ret2; |
1106 | int partial = 0; | |
bfc1af65 | 1107 | unsigned from, to; |
cf17fea6 | 1108 | loff_t new_i_size; |
ac27a0ec | 1109 | |
9bffad1e | 1110 | trace_ext4_journalled_write_end(inode, pos, len, copied); |
bfc1af65 NP |
1111 | from = pos & (PAGE_CACHE_SIZE - 1); |
1112 | to = from + len; | |
1113 | ||
441c8508 CW |
1114 | BUG_ON(!ext4_handle_valid(handle)); |
1115 | ||
bfc1af65 NP |
1116 | if (copied < len) { |
1117 | if (!PageUptodate(page)) | |
1118 | copied = 0; | |
1119 | page_zero_new_buffers(page, from+copied, to); | |
1120 | } | |
ac27a0ec DK |
1121 | |
1122 | ret = walk_page_buffers(handle, page_buffers(page), from, | |
bfc1af65 | 1123 | to, &partial, write_end_fn); |
ac27a0ec DK |
1124 | if (!partial) |
1125 | SetPageUptodate(page); | |
cf17fea6 AK |
1126 | new_i_size = pos + copied; |
1127 | if (new_i_size > inode->i_size) | |
bfc1af65 | 1128 | i_size_write(inode, pos+copied); |
19f5fb7a | 1129 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
2d859db3 | 1130 | EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; |
cf17fea6 AK |
1131 | if (new_i_size > EXT4_I(inode)->i_disksize) { |
1132 | ext4_update_i_disksize(inode, new_i_size); | |
617ba13b | 1133 | ret2 = ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
1134 | if (!ret) |
1135 | ret = ret2; | |
1136 | } | |
bfc1af65 | 1137 | |
cf108bca | 1138 | unlock_page(page); |
f8514083 | 1139 | page_cache_release(page); |
ffacfa7a | 1140 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
f8514083 AK |
1141 | /* if we have allocated more blocks and copied |
1142 | * less. We will have blocks allocated outside | |
1143 | * inode->i_size. So truncate them | |
1144 | */ | |
1145 | ext4_orphan_add(handle, inode); | |
1146 | ||
617ba13b | 1147 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1148 | if (!ret) |
1149 | ret = ret2; | |
f8514083 | 1150 | if (pos + len > inode->i_size) { |
b9a4207d | 1151 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1152 | /* |
ffacfa7a | 1153 | * If truncate failed early the inode might still be |
f8514083 AK |
1154 | * on the orphan list; we need to make sure the inode |
1155 | * is removed from the orphan list in that case. | |
1156 | */ | |
1157 | if (inode->i_nlink) | |
1158 | ext4_orphan_del(NULL, inode); | |
1159 | } | |
bfc1af65 NP |
1160 | |
1161 | return ret ? ret : copied; | |
ac27a0ec | 1162 | } |
d2a17637 | 1163 | |
9d0be502 | 1164 | /* |
7b415bf6 | 1165 | * Reserve a single cluster located at lblock |
9d0be502 | 1166 | */ |
01f49d0b | 1167 | static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock) |
d2a17637 | 1168 | { |
030ba6bc | 1169 | int retries = 0; |
60e58e0f | 1170 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
0637c6f4 | 1171 | struct ext4_inode_info *ei = EXT4_I(inode); |
7b415bf6 | 1172 | unsigned int md_needed; |
5dd4056d | 1173 | int ret; |
d2a17637 MC |
1174 | |
1175 | /* | |
1176 | * recalculate the amount of metadata blocks to reserve | |
1177 | * in order to allocate nrblocks | |
1178 | * worse case is one extent per block | |
1179 | */ | |
030ba6bc | 1180 | repeat: |
0637c6f4 | 1181 | spin_lock(&ei->i_block_reservation_lock); |
7b415bf6 AK |
1182 | md_needed = EXT4_NUM_B2C(sbi, |
1183 | ext4_calc_metadata_amount(inode, lblock)); | |
f8ec9d68 | 1184 | trace_ext4_da_reserve_space(inode, md_needed); |
0637c6f4 | 1185 | spin_unlock(&ei->i_block_reservation_lock); |
d2a17637 | 1186 | |
60e58e0f | 1187 | /* |
72b8ab9d ES |
1188 | * We will charge metadata quota at writeout time; this saves |
1189 | * us from metadata over-estimation, though we may go over by | |
1190 | * a small amount in the end. Here we just reserve for data. | |
60e58e0f | 1191 | */ |
7b415bf6 | 1192 | ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1)); |
5dd4056d CH |
1193 | if (ret) |
1194 | return ret; | |
72b8ab9d ES |
1195 | /* |
1196 | * We do still charge estimated metadata to the sb though; | |
1197 | * we cannot afford to run out of free blocks. | |
1198 | */ | |
e7d5f315 | 1199 | if (ext4_claim_free_clusters(sbi, md_needed + 1, 0)) { |
7b415bf6 | 1200 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1)); |
030ba6bc AK |
1201 | if (ext4_should_retry_alloc(inode->i_sb, &retries)) { |
1202 | yield(); | |
1203 | goto repeat; | |
1204 | } | |
d2a17637 MC |
1205 | return -ENOSPC; |
1206 | } | |
0637c6f4 | 1207 | spin_lock(&ei->i_block_reservation_lock); |
9d0be502 | 1208 | ei->i_reserved_data_blocks++; |
0637c6f4 TT |
1209 | ei->i_reserved_meta_blocks += md_needed; |
1210 | spin_unlock(&ei->i_block_reservation_lock); | |
39bc680a | 1211 | |
d2a17637 MC |
1212 | return 0; /* success */ |
1213 | } | |
1214 | ||
12219aea | 1215 | static void ext4_da_release_space(struct inode *inode, int to_free) |
d2a17637 MC |
1216 | { |
1217 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0637c6f4 | 1218 | struct ext4_inode_info *ei = EXT4_I(inode); |
d2a17637 | 1219 | |
cd213226 MC |
1220 | if (!to_free) |
1221 | return; /* Nothing to release, exit */ | |
1222 | ||
d2a17637 | 1223 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); |
cd213226 | 1224 | |
5a58ec87 | 1225 | trace_ext4_da_release_space(inode, to_free); |
0637c6f4 | 1226 | if (unlikely(to_free > ei->i_reserved_data_blocks)) { |
cd213226 | 1227 | /* |
0637c6f4 TT |
1228 | * if there aren't enough reserved blocks, then the |
1229 | * counter is messed up somewhere. Since this | |
1230 | * function is called from invalidate page, it's | |
1231 | * harmless to return without any action. | |
cd213226 | 1232 | */ |
0637c6f4 TT |
1233 | ext4_msg(inode->i_sb, KERN_NOTICE, "ext4_da_release_space: " |
1234 | "ino %lu, to_free %d with only %d reserved " | |
1084f252 | 1235 | "data blocks", inode->i_ino, to_free, |
0637c6f4 TT |
1236 | ei->i_reserved_data_blocks); |
1237 | WARN_ON(1); | |
1238 | to_free = ei->i_reserved_data_blocks; | |
cd213226 | 1239 | } |
0637c6f4 | 1240 | ei->i_reserved_data_blocks -= to_free; |
cd213226 | 1241 | |
0637c6f4 TT |
1242 | if (ei->i_reserved_data_blocks == 0) { |
1243 | /* | |
1244 | * We can release all of the reserved metadata blocks | |
1245 | * only when we have written all of the delayed | |
1246 | * allocation blocks. | |
7b415bf6 AK |
1247 | * Note that in case of bigalloc, i_reserved_meta_blocks, |
1248 | * i_reserved_data_blocks, etc. refer to number of clusters. | |
0637c6f4 | 1249 | */ |
57042651 | 1250 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, |
72b8ab9d | 1251 | ei->i_reserved_meta_blocks); |
ee5f4d9c | 1252 | ei->i_reserved_meta_blocks = 0; |
9d0be502 | 1253 | ei->i_da_metadata_calc_len = 0; |
0637c6f4 | 1254 | } |
d2a17637 | 1255 | |
72b8ab9d | 1256 | /* update fs dirty data blocks counter */ |
57042651 | 1257 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free); |
d2a17637 | 1258 | |
d2a17637 | 1259 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
60e58e0f | 1260 | |
7b415bf6 | 1261 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free)); |
d2a17637 MC |
1262 | } |
1263 | ||
1264 | static void ext4_da_page_release_reservation(struct page *page, | |
de9a55b8 | 1265 | unsigned long offset) |
d2a17637 MC |
1266 | { |
1267 | int to_release = 0; | |
1268 | struct buffer_head *head, *bh; | |
1269 | unsigned int curr_off = 0; | |
7b415bf6 AK |
1270 | struct inode *inode = page->mapping->host; |
1271 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
1272 | int num_clusters; | |
d2a17637 MC |
1273 | |
1274 | head = page_buffers(page); | |
1275 | bh = head; | |
1276 | do { | |
1277 | unsigned int next_off = curr_off + bh->b_size; | |
1278 | ||
1279 | if ((offset <= curr_off) && (buffer_delay(bh))) { | |
1280 | to_release++; | |
1281 | clear_buffer_delay(bh); | |
5356f261 | 1282 | clear_buffer_da_mapped(bh); |
d2a17637 MC |
1283 | } |
1284 | curr_off = next_off; | |
1285 | } while ((bh = bh->b_this_page) != head); | |
7b415bf6 AK |
1286 | |
1287 | /* If we have released all the blocks belonging to a cluster, then we | |
1288 | * need to release the reserved space for that cluster. */ | |
1289 | num_clusters = EXT4_NUM_B2C(sbi, to_release); | |
1290 | while (num_clusters > 0) { | |
1291 | ext4_fsblk_t lblk; | |
1292 | lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) + | |
1293 | ((num_clusters - 1) << sbi->s_cluster_bits); | |
1294 | if (sbi->s_cluster_ratio == 1 || | |
1295 | !ext4_find_delalloc_cluster(inode, lblk, 1)) | |
1296 | ext4_da_release_space(inode, 1); | |
1297 | ||
1298 | num_clusters--; | |
1299 | } | |
d2a17637 | 1300 | } |
ac27a0ec | 1301 | |
64769240 AT |
1302 | /* |
1303 | * Delayed allocation stuff | |
1304 | */ | |
1305 | ||
64769240 AT |
1306 | /* |
1307 | * mpage_da_submit_io - walks through extent of pages and try to write | |
a1d6cc56 | 1308 | * them with writepage() call back |
64769240 AT |
1309 | * |
1310 | * @mpd->inode: inode | |
1311 | * @mpd->first_page: first page of the extent | |
1312 | * @mpd->next_page: page after the last page of the extent | |
64769240 AT |
1313 | * |
1314 | * By the time mpage_da_submit_io() is called we expect all blocks | |
1315 | * to be allocated. this may be wrong if allocation failed. | |
1316 | * | |
1317 | * As pages are already locked by write_cache_pages(), we can't use it | |
1318 | */ | |
1de3e3df TT |
1319 | static int mpage_da_submit_io(struct mpage_da_data *mpd, |
1320 | struct ext4_map_blocks *map) | |
64769240 | 1321 | { |
791b7f08 AK |
1322 | struct pagevec pvec; |
1323 | unsigned long index, end; | |
1324 | int ret = 0, err, nr_pages, i; | |
1325 | struct inode *inode = mpd->inode; | |
1326 | struct address_space *mapping = inode->i_mapping; | |
cb20d518 | 1327 | loff_t size = i_size_read(inode); |
3ecdb3a1 TT |
1328 | unsigned int len, block_start; |
1329 | struct buffer_head *bh, *page_bufs = NULL; | |
cb20d518 | 1330 | int journal_data = ext4_should_journal_data(inode); |
1de3e3df | 1331 | sector_t pblock = 0, cur_logical = 0; |
bd2d0210 | 1332 | struct ext4_io_submit io_submit; |
64769240 AT |
1333 | |
1334 | BUG_ON(mpd->next_page <= mpd->first_page); | |
bd2d0210 | 1335 | memset(&io_submit, 0, sizeof(io_submit)); |
791b7f08 AK |
1336 | /* |
1337 | * We need to start from the first_page to the next_page - 1 | |
1338 | * to make sure we also write the mapped dirty buffer_heads. | |
8dc207c0 | 1339 | * If we look at mpd->b_blocknr we would only be looking |
791b7f08 AK |
1340 | * at the currently mapped buffer_heads. |
1341 | */ | |
64769240 AT |
1342 | index = mpd->first_page; |
1343 | end = mpd->next_page - 1; | |
1344 | ||
791b7f08 | 1345 | pagevec_init(&pvec, 0); |
64769240 | 1346 | while (index <= end) { |
791b7f08 | 1347 | nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE); |
64769240 AT |
1348 | if (nr_pages == 0) |
1349 | break; | |
1350 | for (i = 0; i < nr_pages; i++) { | |
97498956 | 1351 | int commit_write = 0, skip_page = 0; |
64769240 AT |
1352 | struct page *page = pvec.pages[i]; |
1353 | ||
791b7f08 AK |
1354 | index = page->index; |
1355 | if (index > end) | |
1356 | break; | |
cb20d518 TT |
1357 | |
1358 | if (index == size >> PAGE_CACHE_SHIFT) | |
1359 | len = size & ~PAGE_CACHE_MASK; | |
1360 | else | |
1361 | len = PAGE_CACHE_SIZE; | |
1de3e3df TT |
1362 | if (map) { |
1363 | cur_logical = index << (PAGE_CACHE_SHIFT - | |
1364 | inode->i_blkbits); | |
1365 | pblock = map->m_pblk + (cur_logical - | |
1366 | map->m_lblk); | |
1367 | } | |
791b7f08 AK |
1368 | index++; |
1369 | ||
1370 | BUG_ON(!PageLocked(page)); | |
1371 | BUG_ON(PageWriteback(page)); | |
1372 | ||
64769240 | 1373 | /* |
cb20d518 TT |
1374 | * If the page does not have buffers (for |
1375 | * whatever reason), try to create them using | |
a107e5a3 | 1376 | * __block_write_begin. If this fails, |
97498956 | 1377 | * skip the page and move on. |
64769240 | 1378 | */ |
cb20d518 | 1379 | if (!page_has_buffers(page)) { |
a107e5a3 | 1380 | if (__block_write_begin(page, 0, len, |
cb20d518 | 1381 | noalloc_get_block_write)) { |
97498956 | 1382 | skip_page: |
cb20d518 TT |
1383 | unlock_page(page); |
1384 | continue; | |
1385 | } | |
1386 | commit_write = 1; | |
1387 | } | |
64769240 | 1388 | |
3ecdb3a1 TT |
1389 | bh = page_bufs = page_buffers(page); |
1390 | block_start = 0; | |
64769240 | 1391 | do { |
1de3e3df | 1392 | if (!bh) |
97498956 | 1393 | goto skip_page; |
1de3e3df TT |
1394 | if (map && (cur_logical >= map->m_lblk) && |
1395 | (cur_logical <= (map->m_lblk + | |
1396 | (map->m_len - 1)))) { | |
29fa89d0 AK |
1397 | if (buffer_delay(bh)) { |
1398 | clear_buffer_delay(bh); | |
1399 | bh->b_blocknr = pblock; | |
29fa89d0 | 1400 | } |
5356f261 AK |
1401 | if (buffer_da_mapped(bh)) |
1402 | clear_buffer_da_mapped(bh); | |
1de3e3df TT |
1403 | if (buffer_unwritten(bh) || |
1404 | buffer_mapped(bh)) | |
1405 | BUG_ON(bh->b_blocknr != pblock); | |
1406 | if (map->m_flags & EXT4_MAP_UNINIT) | |
1407 | set_buffer_uninit(bh); | |
1408 | clear_buffer_unwritten(bh); | |
1409 | } | |
29fa89d0 | 1410 | |
13a79a47 YY |
1411 | /* |
1412 | * skip page if block allocation undone and | |
1413 | * block is dirty | |
1414 | */ | |
1415 | if (ext4_bh_delay_or_unwritten(NULL, bh)) | |
97498956 | 1416 | skip_page = 1; |
3ecdb3a1 TT |
1417 | bh = bh->b_this_page; |
1418 | block_start += bh->b_size; | |
64769240 AT |
1419 | cur_logical++; |
1420 | pblock++; | |
1de3e3df TT |
1421 | } while (bh != page_bufs); |
1422 | ||
97498956 TT |
1423 | if (skip_page) |
1424 | goto skip_page; | |
cb20d518 TT |
1425 | |
1426 | if (commit_write) | |
1427 | /* mark the buffer_heads as dirty & uptodate */ | |
1428 | block_commit_write(page, 0, len); | |
1429 | ||
97498956 | 1430 | clear_page_dirty_for_io(page); |
bd2d0210 TT |
1431 | /* |
1432 | * Delalloc doesn't support data journalling, | |
1433 | * but eventually maybe we'll lift this | |
1434 | * restriction. | |
1435 | */ | |
1436 | if (unlikely(journal_data && PageChecked(page))) | |
cb20d518 | 1437 | err = __ext4_journalled_writepage(page, len); |
1449032b | 1438 | else if (test_opt(inode->i_sb, MBLK_IO_SUBMIT)) |
bd2d0210 TT |
1439 | err = ext4_bio_write_page(&io_submit, page, |
1440 | len, mpd->wbc); | |
9dd75f1f TT |
1441 | else if (buffer_uninit(page_bufs)) { |
1442 | ext4_set_bh_endio(page_bufs, inode); | |
1443 | err = block_write_full_page_endio(page, | |
1444 | noalloc_get_block_write, | |
1445 | mpd->wbc, ext4_end_io_buffer_write); | |
1446 | } else | |
1449032b TT |
1447 | err = block_write_full_page(page, |
1448 | noalloc_get_block_write, mpd->wbc); | |
cb20d518 TT |
1449 | |
1450 | if (!err) | |
a1d6cc56 | 1451 | mpd->pages_written++; |
64769240 AT |
1452 | /* |
1453 | * In error case, we have to continue because | |
1454 | * remaining pages are still locked | |
64769240 AT |
1455 | */ |
1456 | if (ret == 0) | |
1457 | ret = err; | |
64769240 AT |
1458 | } |
1459 | pagevec_release(&pvec); | |
1460 | } | |
bd2d0210 | 1461 | ext4_io_submit(&io_submit); |
64769240 | 1462 | return ret; |
64769240 AT |
1463 | } |
1464 | ||
c7f5938a | 1465 | static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd) |
c4a0c46e AK |
1466 | { |
1467 | int nr_pages, i; | |
1468 | pgoff_t index, end; | |
1469 | struct pagevec pvec; | |
1470 | struct inode *inode = mpd->inode; | |
1471 | struct address_space *mapping = inode->i_mapping; | |
1472 | ||
c7f5938a CW |
1473 | index = mpd->first_page; |
1474 | end = mpd->next_page - 1; | |
c4a0c46e AK |
1475 | while (index <= end) { |
1476 | nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE); | |
1477 | if (nr_pages == 0) | |
1478 | break; | |
1479 | for (i = 0; i < nr_pages; i++) { | |
1480 | struct page *page = pvec.pages[i]; | |
9b1d0998 | 1481 | if (page->index > end) |
c4a0c46e | 1482 | break; |
c4a0c46e AK |
1483 | BUG_ON(!PageLocked(page)); |
1484 | BUG_ON(PageWriteback(page)); | |
1485 | block_invalidatepage(page, 0); | |
1486 | ClearPageUptodate(page); | |
1487 | unlock_page(page); | |
1488 | } | |
9b1d0998 JK |
1489 | index = pvec.pages[nr_pages - 1]->index + 1; |
1490 | pagevec_release(&pvec); | |
c4a0c46e AK |
1491 | } |
1492 | return; | |
1493 | } | |
1494 | ||
df22291f AK |
1495 | static void ext4_print_free_blocks(struct inode *inode) |
1496 | { | |
1497 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
92b97816 TT |
1498 | struct super_block *sb = inode->i_sb; |
1499 | ||
1500 | ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld", | |
5dee5437 TT |
1501 | EXT4_C2B(EXT4_SB(inode->i_sb), |
1502 | ext4_count_free_clusters(inode->i_sb))); | |
92b97816 TT |
1503 | ext4_msg(sb, KERN_CRIT, "Free/Dirty block details"); |
1504 | ext4_msg(sb, KERN_CRIT, "free_blocks=%lld", | |
57042651 TT |
1505 | (long long) EXT4_C2B(EXT4_SB(inode->i_sb), |
1506 | percpu_counter_sum(&sbi->s_freeclusters_counter))); | |
92b97816 | 1507 | ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld", |
7b415bf6 AK |
1508 | (long long) EXT4_C2B(EXT4_SB(inode->i_sb), |
1509 | percpu_counter_sum(&sbi->s_dirtyclusters_counter))); | |
92b97816 TT |
1510 | ext4_msg(sb, KERN_CRIT, "Block reservation details"); |
1511 | ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u", | |
1512 | EXT4_I(inode)->i_reserved_data_blocks); | |
1513 | ext4_msg(sb, KERN_CRIT, "i_reserved_meta_blocks=%u", | |
1693918e | 1514 | EXT4_I(inode)->i_reserved_meta_blocks); |
df22291f AK |
1515 | return; |
1516 | } | |
1517 | ||
64769240 | 1518 | /* |
5a87b7a5 TT |
1519 | * mpage_da_map_and_submit - go through given space, map them |
1520 | * if necessary, and then submit them for I/O | |
64769240 | 1521 | * |
8dc207c0 | 1522 | * @mpd - bh describing space |
64769240 AT |
1523 | * |
1524 | * The function skips space we know is already mapped to disk blocks. | |
1525 | * | |
64769240 | 1526 | */ |
5a87b7a5 | 1527 | static void mpage_da_map_and_submit(struct mpage_da_data *mpd) |
64769240 | 1528 | { |
2ac3b6e0 | 1529 | int err, blks, get_blocks_flags; |
1de3e3df | 1530 | struct ext4_map_blocks map, *mapp = NULL; |
2fa3cdfb TT |
1531 | sector_t next = mpd->b_blocknr; |
1532 | unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits; | |
1533 | loff_t disksize = EXT4_I(mpd->inode)->i_disksize; | |
1534 | handle_t *handle = NULL; | |
64769240 AT |
1535 | |
1536 | /* | |
5a87b7a5 TT |
1537 | * If the blocks are mapped already, or we couldn't accumulate |
1538 | * any blocks, then proceed immediately to the submission stage. | |
2fa3cdfb | 1539 | */ |
5a87b7a5 TT |
1540 | if ((mpd->b_size == 0) || |
1541 | ((mpd->b_state & (1 << BH_Mapped)) && | |
1542 | !(mpd->b_state & (1 << BH_Delay)) && | |
1543 | !(mpd->b_state & (1 << BH_Unwritten)))) | |
1544 | goto submit_io; | |
2fa3cdfb TT |
1545 | |
1546 | handle = ext4_journal_current_handle(); | |
1547 | BUG_ON(!handle); | |
1548 | ||
79ffab34 | 1549 | /* |
79e83036 | 1550 | * Call ext4_map_blocks() to allocate any delayed allocation |
2ac3b6e0 TT |
1551 | * blocks, or to convert an uninitialized extent to be |
1552 | * initialized (in the case where we have written into | |
1553 | * one or more preallocated blocks). | |
1554 | * | |
1555 | * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to | |
1556 | * indicate that we are on the delayed allocation path. This | |
1557 | * affects functions in many different parts of the allocation | |
1558 | * call path. This flag exists primarily because we don't | |
79e83036 | 1559 | * want to change *many* call functions, so ext4_map_blocks() |
f2321097 | 1560 | * will set the EXT4_STATE_DELALLOC_RESERVED flag once the |
2ac3b6e0 TT |
1561 | * inode's allocation semaphore is taken. |
1562 | * | |
1563 | * If the blocks in questions were delalloc blocks, set | |
1564 | * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting | |
1565 | * variables are updated after the blocks have been allocated. | |
79ffab34 | 1566 | */ |
2ed88685 TT |
1567 | map.m_lblk = next; |
1568 | map.m_len = max_blocks; | |
1296cc85 | 1569 | get_blocks_flags = EXT4_GET_BLOCKS_CREATE; |
744692dc JZ |
1570 | if (ext4_should_dioread_nolock(mpd->inode)) |
1571 | get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT; | |
2ac3b6e0 | 1572 | if (mpd->b_state & (1 << BH_Delay)) |
1296cc85 AK |
1573 | get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE; |
1574 | ||
2ed88685 | 1575 | blks = ext4_map_blocks(handle, mpd->inode, &map, get_blocks_flags); |
2fa3cdfb | 1576 | if (blks < 0) { |
e3570639 ES |
1577 | struct super_block *sb = mpd->inode->i_sb; |
1578 | ||
2fa3cdfb | 1579 | err = blks; |
ed5bde0b | 1580 | /* |
5a87b7a5 | 1581 | * If get block returns EAGAIN or ENOSPC and there |
97498956 TT |
1582 | * appears to be free blocks we will just let |
1583 | * mpage_da_submit_io() unlock all of the pages. | |
c4a0c46e AK |
1584 | */ |
1585 | if (err == -EAGAIN) | |
5a87b7a5 | 1586 | goto submit_io; |
df22291f | 1587 | |
5dee5437 | 1588 | if (err == -ENOSPC && ext4_count_free_clusters(sb)) { |
df22291f | 1589 | mpd->retval = err; |
5a87b7a5 | 1590 | goto submit_io; |
df22291f AK |
1591 | } |
1592 | ||
c4a0c46e | 1593 | /* |
ed5bde0b TT |
1594 | * get block failure will cause us to loop in |
1595 | * writepages, because a_ops->writepage won't be able | |
1596 | * to make progress. The page will be redirtied by | |
1597 | * writepage and writepages will again try to write | |
1598 | * the same. | |
c4a0c46e | 1599 | */ |
e3570639 ES |
1600 | if (!(EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)) { |
1601 | ext4_msg(sb, KERN_CRIT, | |
1602 | "delayed block allocation failed for inode %lu " | |
1603 | "at logical offset %llu with max blocks %zd " | |
1604 | "with error %d", mpd->inode->i_ino, | |
1605 | (unsigned long long) next, | |
1606 | mpd->b_size >> mpd->inode->i_blkbits, err); | |
1607 | ext4_msg(sb, KERN_CRIT, | |
1608 | "This should not happen!! Data will be lost\n"); | |
1609 | if (err == -ENOSPC) | |
1610 | ext4_print_free_blocks(mpd->inode); | |
030ba6bc | 1611 | } |
2fa3cdfb | 1612 | /* invalidate all the pages */ |
c7f5938a | 1613 | ext4_da_block_invalidatepages(mpd); |
e0fd9b90 CW |
1614 | |
1615 | /* Mark this page range as having been completed */ | |
1616 | mpd->io_done = 1; | |
5a87b7a5 | 1617 | return; |
c4a0c46e | 1618 | } |
2fa3cdfb TT |
1619 | BUG_ON(blks == 0); |
1620 | ||
1de3e3df | 1621 | mapp = ↦ |
2ed88685 TT |
1622 | if (map.m_flags & EXT4_MAP_NEW) { |
1623 | struct block_device *bdev = mpd->inode->i_sb->s_bdev; | |
1624 | int i; | |
64769240 | 1625 | |
2ed88685 TT |
1626 | for (i = 0; i < map.m_len; i++) |
1627 | unmap_underlying_metadata(bdev, map.m_pblk + i); | |
64769240 | 1628 | |
decbd919 TT |
1629 | if (ext4_should_order_data(mpd->inode)) { |
1630 | err = ext4_jbd2_file_inode(handle, mpd->inode); | |
8de49e67 | 1631 | if (err) { |
decbd919 | 1632 | /* Only if the journal is aborted */ |
8de49e67 KM |
1633 | mpd->retval = err; |
1634 | goto submit_io; | |
1635 | } | |
decbd919 | 1636 | } |
2fa3cdfb TT |
1637 | } |
1638 | ||
1639 | /* | |
03f5d8bc | 1640 | * Update on-disk size along with block allocation. |
2fa3cdfb TT |
1641 | */ |
1642 | disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits; | |
1643 | if (disksize > i_size_read(mpd->inode)) | |
1644 | disksize = i_size_read(mpd->inode); | |
1645 | if (disksize > EXT4_I(mpd->inode)->i_disksize) { | |
1646 | ext4_update_i_disksize(mpd->inode, disksize); | |
5a87b7a5 TT |
1647 | err = ext4_mark_inode_dirty(handle, mpd->inode); |
1648 | if (err) | |
1649 | ext4_error(mpd->inode->i_sb, | |
1650 | "Failed to mark inode %lu dirty", | |
1651 | mpd->inode->i_ino); | |
2fa3cdfb TT |
1652 | } |
1653 | ||
5a87b7a5 | 1654 | submit_io: |
1de3e3df | 1655 | mpage_da_submit_io(mpd, mapp); |
5a87b7a5 | 1656 | mpd->io_done = 1; |
64769240 AT |
1657 | } |
1658 | ||
bf068ee2 AK |
1659 | #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \ |
1660 | (1 << BH_Delay) | (1 << BH_Unwritten)) | |
64769240 AT |
1661 | |
1662 | /* | |
1663 | * mpage_add_bh_to_extent - try to add one more block to extent of blocks | |
1664 | * | |
1665 | * @mpd->lbh - extent of blocks | |
1666 | * @logical - logical number of the block in the file | |
1667 | * @bh - bh of the block (used to access block's state) | |
1668 | * | |
1669 | * the function is used to collect contig. blocks in same state | |
1670 | */ | |
1671 | static void mpage_add_bh_to_extent(struct mpage_da_data *mpd, | |
8dc207c0 TT |
1672 | sector_t logical, size_t b_size, |
1673 | unsigned long b_state) | |
64769240 | 1674 | { |
64769240 | 1675 | sector_t next; |
8dc207c0 | 1676 | int nrblocks = mpd->b_size >> mpd->inode->i_blkbits; |
64769240 | 1677 | |
c445e3e0 ES |
1678 | /* |
1679 | * XXX Don't go larger than mballoc is willing to allocate | |
1680 | * This is a stopgap solution. We eventually need to fold | |
1681 | * mpage_da_submit_io() into this function and then call | |
79e83036 | 1682 | * ext4_map_blocks() multiple times in a loop |
c445e3e0 ES |
1683 | */ |
1684 | if (nrblocks >= 8*1024*1024/mpd->inode->i_sb->s_blocksize) | |
1685 | goto flush_it; | |
1686 | ||
525f4ed8 | 1687 | /* check if thereserved journal credits might overflow */ |
12e9b892 | 1688 | if (!(ext4_test_inode_flag(mpd->inode, EXT4_INODE_EXTENTS))) { |
525f4ed8 MC |
1689 | if (nrblocks >= EXT4_MAX_TRANS_DATA) { |
1690 | /* | |
1691 | * With non-extent format we are limited by the journal | |
1692 | * credit available. Total credit needed to insert | |
1693 | * nrblocks contiguous blocks is dependent on the | |
1694 | * nrblocks. So limit nrblocks. | |
1695 | */ | |
1696 | goto flush_it; | |
1697 | } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) > | |
1698 | EXT4_MAX_TRANS_DATA) { | |
1699 | /* | |
1700 | * Adding the new buffer_head would make it cross the | |
1701 | * allowed limit for which we have journal credit | |
1702 | * reserved. So limit the new bh->b_size | |
1703 | */ | |
1704 | b_size = (EXT4_MAX_TRANS_DATA - nrblocks) << | |
1705 | mpd->inode->i_blkbits; | |
1706 | /* we will do mpage_da_submit_io in the next loop */ | |
1707 | } | |
1708 | } | |
64769240 AT |
1709 | /* |
1710 | * First block in the extent | |
1711 | */ | |
8dc207c0 TT |
1712 | if (mpd->b_size == 0) { |
1713 | mpd->b_blocknr = logical; | |
1714 | mpd->b_size = b_size; | |
1715 | mpd->b_state = b_state & BH_FLAGS; | |
64769240 AT |
1716 | return; |
1717 | } | |
1718 | ||
8dc207c0 | 1719 | next = mpd->b_blocknr + nrblocks; |
64769240 AT |
1720 | /* |
1721 | * Can we merge the block to our big extent? | |
1722 | */ | |
8dc207c0 TT |
1723 | if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) { |
1724 | mpd->b_size += b_size; | |
64769240 AT |
1725 | return; |
1726 | } | |
1727 | ||
525f4ed8 | 1728 | flush_it: |
64769240 AT |
1729 | /* |
1730 | * We couldn't merge the block to our extent, so we | |
1731 | * need to flush current extent and start new one | |
1732 | */ | |
5a87b7a5 | 1733 | mpage_da_map_and_submit(mpd); |
a1d6cc56 | 1734 | return; |
64769240 AT |
1735 | } |
1736 | ||
c364b22c | 1737 | static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh) |
29fa89d0 | 1738 | { |
c364b22c | 1739 | return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh); |
29fa89d0 AK |
1740 | } |
1741 | ||
5356f261 AK |
1742 | /* |
1743 | * This function is grabs code from the very beginning of | |
1744 | * ext4_map_blocks, but assumes that the caller is from delayed write | |
1745 | * time. This function looks up the requested blocks and sets the | |
1746 | * buffer delay bit under the protection of i_data_sem. | |
1747 | */ | |
1748 | static int ext4_da_map_blocks(struct inode *inode, sector_t iblock, | |
1749 | struct ext4_map_blocks *map, | |
1750 | struct buffer_head *bh) | |
1751 | { | |
1752 | int retval; | |
1753 | sector_t invalid_block = ~((sector_t) 0xffff); | |
1754 | ||
1755 | if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es)) | |
1756 | invalid_block = ~0; | |
1757 | ||
1758 | map->m_flags = 0; | |
1759 | ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u," | |
1760 | "logical block %lu\n", inode->i_ino, map->m_len, | |
1761 | (unsigned long) map->m_lblk); | |
1762 | /* | |
1763 | * Try to see if we can get the block without requesting a new | |
1764 | * file system block. | |
1765 | */ | |
1766 | down_read((&EXT4_I(inode)->i_data_sem)); | |
1767 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
1768 | retval = ext4_ext_map_blocks(NULL, inode, map, 0); | |
1769 | else | |
1770 | retval = ext4_ind_map_blocks(NULL, inode, map, 0); | |
1771 | ||
1772 | if (retval == 0) { | |
1773 | /* | |
1774 | * XXX: __block_prepare_write() unmaps passed block, | |
1775 | * is it OK? | |
1776 | */ | |
1777 | /* If the block was allocated from previously allocated cluster, | |
1778 | * then we dont need to reserve it again. */ | |
1779 | if (!(map->m_flags & EXT4_MAP_FROM_CLUSTER)) { | |
1780 | retval = ext4_da_reserve_space(inode, iblock); | |
1781 | if (retval) | |
1782 | /* not enough space to reserve */ | |
1783 | goto out_unlock; | |
1784 | } | |
1785 | ||
1786 | /* Clear EXT4_MAP_FROM_CLUSTER flag since its purpose is served | |
1787 | * and it should not appear on the bh->b_state. | |
1788 | */ | |
1789 | map->m_flags &= ~EXT4_MAP_FROM_CLUSTER; | |
1790 | ||
1791 | map_bh(bh, inode->i_sb, invalid_block); | |
1792 | set_buffer_new(bh); | |
1793 | set_buffer_delay(bh); | |
1794 | } | |
1795 | ||
1796 | out_unlock: | |
1797 | up_read((&EXT4_I(inode)->i_data_sem)); | |
1798 | ||
1799 | return retval; | |
1800 | } | |
1801 | ||
64769240 | 1802 | /* |
b920c755 TT |
1803 | * This is a special get_blocks_t callback which is used by |
1804 | * ext4_da_write_begin(). It will either return mapped block or | |
1805 | * reserve space for a single block. | |
29fa89d0 AK |
1806 | * |
1807 | * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set. | |
1808 | * We also have b_blocknr = -1 and b_bdev initialized properly | |
1809 | * | |
1810 | * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set. | |
1811 | * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev | |
1812 | * initialized properly. | |
64769240 AT |
1813 | */ |
1814 | static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock, | |
2ed88685 | 1815 | struct buffer_head *bh, int create) |
64769240 | 1816 | { |
2ed88685 | 1817 | struct ext4_map_blocks map; |
64769240 AT |
1818 | int ret = 0; |
1819 | ||
1820 | BUG_ON(create == 0); | |
2ed88685 TT |
1821 | BUG_ON(bh->b_size != inode->i_sb->s_blocksize); |
1822 | ||
1823 | map.m_lblk = iblock; | |
1824 | map.m_len = 1; | |
64769240 AT |
1825 | |
1826 | /* | |
1827 | * first, we need to know whether the block is allocated already | |
1828 | * preallocated blocks are unmapped but should treated | |
1829 | * the same as allocated blocks. | |
1830 | */ | |
5356f261 AK |
1831 | ret = ext4_da_map_blocks(inode, iblock, &map, bh); |
1832 | if (ret <= 0) | |
2ed88685 | 1833 | return ret; |
64769240 | 1834 | |
2ed88685 TT |
1835 | map_bh(bh, inode->i_sb, map.m_pblk); |
1836 | bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags; | |
1837 | ||
1838 | if (buffer_unwritten(bh)) { | |
1839 | /* A delayed write to unwritten bh should be marked | |
1840 | * new and mapped. Mapped ensures that we don't do | |
1841 | * get_block multiple times when we write to the same | |
1842 | * offset and new ensures that we do proper zero out | |
1843 | * for partial write. | |
1844 | */ | |
1845 | set_buffer_new(bh); | |
c8205636 | 1846 | set_buffer_mapped(bh); |
2ed88685 TT |
1847 | } |
1848 | return 0; | |
64769240 | 1849 | } |
61628a3f | 1850 | |
b920c755 TT |
1851 | /* |
1852 | * This function is used as a standard get_block_t calback function | |
1853 | * when there is no desire to allocate any blocks. It is used as a | |
ebdec241 | 1854 | * callback function for block_write_begin() and block_write_full_page(). |
206f7ab4 | 1855 | * These functions should only try to map a single block at a time. |
b920c755 TT |
1856 | * |
1857 | * Since this function doesn't do block allocations even if the caller | |
1858 | * requests it by passing in create=1, it is critically important that | |
1859 | * any caller checks to make sure that any buffer heads are returned | |
1860 | * by this function are either all already mapped or marked for | |
206f7ab4 CH |
1861 | * delayed allocation before calling block_write_full_page(). Otherwise, |
1862 | * b_blocknr could be left unitialized, and the page write functions will | |
1863 | * be taken by surprise. | |
b920c755 TT |
1864 | */ |
1865 | static int noalloc_get_block_write(struct inode *inode, sector_t iblock, | |
f0e6c985 AK |
1866 | struct buffer_head *bh_result, int create) |
1867 | { | |
a2dc52b5 | 1868 | BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize); |
2ed88685 | 1869 | return _ext4_get_block(inode, iblock, bh_result, 0); |
61628a3f MC |
1870 | } |
1871 | ||
62e086be AK |
1872 | static int bget_one(handle_t *handle, struct buffer_head *bh) |
1873 | { | |
1874 | get_bh(bh); | |
1875 | return 0; | |
1876 | } | |
1877 | ||
1878 | static int bput_one(handle_t *handle, struct buffer_head *bh) | |
1879 | { | |
1880 | put_bh(bh); | |
1881 | return 0; | |
1882 | } | |
1883 | ||
1884 | static int __ext4_journalled_writepage(struct page *page, | |
62e086be AK |
1885 | unsigned int len) |
1886 | { | |
1887 | struct address_space *mapping = page->mapping; | |
1888 | struct inode *inode = mapping->host; | |
1889 | struct buffer_head *page_bufs; | |
1890 | handle_t *handle = NULL; | |
1891 | int ret = 0; | |
1892 | int err; | |
1893 | ||
cb20d518 | 1894 | ClearPageChecked(page); |
62e086be AK |
1895 | page_bufs = page_buffers(page); |
1896 | BUG_ON(!page_bufs); | |
1897 | walk_page_buffers(handle, page_bufs, 0, len, NULL, bget_one); | |
1898 | /* As soon as we unlock the page, it can go away, but we have | |
1899 | * references to buffers so we are safe */ | |
1900 | unlock_page(page); | |
1901 | ||
1902 | handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode)); | |
1903 | if (IS_ERR(handle)) { | |
1904 | ret = PTR_ERR(handle); | |
1905 | goto out; | |
1906 | } | |
1907 | ||
441c8508 CW |
1908 | BUG_ON(!ext4_handle_valid(handle)); |
1909 | ||
62e086be AK |
1910 | ret = walk_page_buffers(handle, page_bufs, 0, len, NULL, |
1911 | do_journal_get_write_access); | |
1912 | ||
1913 | err = walk_page_buffers(handle, page_bufs, 0, len, NULL, | |
1914 | write_end_fn); | |
1915 | if (ret == 0) | |
1916 | ret = err; | |
2d859db3 | 1917 | EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; |
62e086be AK |
1918 | err = ext4_journal_stop(handle); |
1919 | if (!ret) | |
1920 | ret = err; | |
1921 | ||
1922 | walk_page_buffers(handle, page_bufs, 0, len, NULL, bput_one); | |
19f5fb7a | 1923 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
62e086be AK |
1924 | out: |
1925 | return ret; | |
1926 | } | |
1927 | ||
744692dc JZ |
1928 | static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode); |
1929 | static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate); | |
1930 | ||
61628a3f | 1931 | /* |
43ce1d23 AK |
1932 | * Note that we don't need to start a transaction unless we're journaling data |
1933 | * because we should have holes filled from ext4_page_mkwrite(). We even don't | |
1934 | * need to file the inode to the transaction's list in ordered mode because if | |
1935 | * we are writing back data added by write(), the inode is already there and if | |
25985edc | 1936 | * we are writing back data modified via mmap(), no one guarantees in which |
43ce1d23 AK |
1937 | * transaction the data will hit the disk. In case we are journaling data, we |
1938 | * cannot start transaction directly because transaction start ranks above page | |
1939 | * lock so we have to do some magic. | |
1940 | * | |
b920c755 TT |
1941 | * This function can get called via... |
1942 | * - ext4_da_writepages after taking page lock (have journal handle) | |
1943 | * - journal_submit_inode_data_buffers (no journal handle) | |
1944 | * - shrink_page_list via pdflush (no journal handle) | |
1945 | * - grab_page_cache when doing write_begin (have journal handle) | |
43ce1d23 AK |
1946 | * |
1947 | * We don't do any block allocation in this function. If we have page with | |
1948 | * multiple blocks we need to write those buffer_heads that are mapped. This | |
1949 | * is important for mmaped based write. So if we do with blocksize 1K | |
1950 | * truncate(f, 1024); | |
1951 | * a = mmap(f, 0, 4096); | |
1952 | * a[0] = 'a'; | |
1953 | * truncate(f, 4096); | |
1954 | * we have in the page first buffer_head mapped via page_mkwrite call back | |
90802ed9 | 1955 | * but other buffer_heads would be unmapped but dirty (dirty done via the |
43ce1d23 AK |
1956 | * do_wp_page). So writepage should write the first block. If we modify |
1957 | * the mmap area beyond 1024 we will again get a page_fault and the | |
1958 | * page_mkwrite callback will do the block allocation and mark the | |
1959 | * buffer_heads mapped. | |
1960 | * | |
1961 | * We redirty the page if we have any buffer_heads that is either delay or | |
1962 | * unwritten in the page. | |
1963 | * | |
1964 | * We can get recursively called as show below. | |
1965 | * | |
1966 | * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() -> | |
1967 | * ext4_writepage() | |
1968 | * | |
1969 | * But since we don't do any block allocation we should not deadlock. | |
1970 | * Page also have the dirty flag cleared so we don't get recurive page_lock. | |
61628a3f | 1971 | */ |
43ce1d23 | 1972 | static int ext4_writepage(struct page *page, |
62e086be | 1973 | struct writeback_control *wbc) |
64769240 | 1974 | { |
a42afc5f | 1975 | int ret = 0, commit_write = 0; |
61628a3f | 1976 | loff_t size; |
498e5f24 | 1977 | unsigned int len; |
744692dc | 1978 | struct buffer_head *page_bufs = NULL; |
61628a3f MC |
1979 | struct inode *inode = page->mapping->host; |
1980 | ||
a9c667f8 | 1981 | trace_ext4_writepage(page); |
f0e6c985 AK |
1982 | size = i_size_read(inode); |
1983 | if (page->index == size >> PAGE_CACHE_SHIFT) | |
1984 | len = size & ~PAGE_CACHE_MASK; | |
1985 | else | |
1986 | len = PAGE_CACHE_SIZE; | |
64769240 | 1987 | |
a42afc5f TT |
1988 | /* |
1989 | * If the page does not have buffers (for whatever reason), | |
a107e5a3 | 1990 | * try to create them using __block_write_begin. If this |
a42afc5f TT |
1991 | * fails, redirty the page and move on. |
1992 | */ | |
b1142e8f | 1993 | if (!page_has_buffers(page)) { |
a107e5a3 | 1994 | if (__block_write_begin(page, 0, len, |
a42afc5f TT |
1995 | noalloc_get_block_write)) { |
1996 | redirty_page: | |
f0e6c985 AK |
1997 | redirty_page_for_writepage(wbc, page); |
1998 | unlock_page(page); | |
1999 | return 0; | |
2000 | } | |
a42afc5f TT |
2001 | commit_write = 1; |
2002 | } | |
2003 | page_bufs = page_buffers(page); | |
2004 | if (walk_page_buffers(NULL, page_bufs, 0, len, NULL, | |
2005 | ext4_bh_delay_or_unwritten)) { | |
f0e6c985 | 2006 | /* |
b1142e8f TT |
2007 | * We don't want to do block allocation, so redirty |
2008 | * the page and return. We may reach here when we do | |
2009 | * a journal commit via journal_submit_inode_data_buffers. | |
966dbde2 MG |
2010 | * We can also reach here via shrink_page_list but it |
2011 | * should never be for direct reclaim so warn if that | |
2012 | * happens | |
f0e6c985 | 2013 | */ |
966dbde2 MG |
2014 | WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) == |
2015 | PF_MEMALLOC); | |
a42afc5f TT |
2016 | goto redirty_page; |
2017 | } | |
2018 | if (commit_write) | |
ed9b3e33 | 2019 | /* now mark the buffer_heads as dirty and uptodate */ |
b767e78a | 2020 | block_commit_write(page, 0, len); |
64769240 | 2021 | |
cb20d518 | 2022 | if (PageChecked(page) && ext4_should_journal_data(inode)) |
43ce1d23 AK |
2023 | /* |
2024 | * It's mmapped pagecache. Add buffers and journal it. There | |
2025 | * doesn't seem much point in redirtying the page here. | |
2026 | */ | |
3f0ca309 | 2027 | return __ext4_journalled_writepage(page, len); |
43ce1d23 | 2028 | |
a42afc5f | 2029 | if (buffer_uninit(page_bufs)) { |
744692dc JZ |
2030 | ext4_set_bh_endio(page_bufs, inode); |
2031 | ret = block_write_full_page_endio(page, noalloc_get_block_write, | |
2032 | wbc, ext4_end_io_buffer_write); | |
2033 | } else | |
b920c755 TT |
2034 | ret = block_write_full_page(page, noalloc_get_block_write, |
2035 | wbc); | |
64769240 | 2036 | |
64769240 AT |
2037 | return ret; |
2038 | } | |
2039 | ||
61628a3f | 2040 | /* |
525f4ed8 | 2041 | * This is called via ext4_da_writepages() to |
25985edc | 2042 | * calculate the total number of credits to reserve to fit |
525f4ed8 MC |
2043 | * a single extent allocation into a single transaction, |
2044 | * ext4_da_writpeages() will loop calling this before | |
2045 | * the block allocation. | |
61628a3f | 2046 | */ |
525f4ed8 MC |
2047 | |
2048 | static int ext4_da_writepages_trans_blocks(struct inode *inode) | |
2049 | { | |
2050 | int max_blocks = EXT4_I(inode)->i_reserved_data_blocks; | |
2051 | ||
2052 | /* | |
2053 | * With non-extent format the journal credit needed to | |
2054 | * insert nrblocks contiguous block is dependent on | |
2055 | * number of contiguous block. So we will limit | |
2056 | * number of contiguous block to a sane value | |
2057 | */ | |
12e9b892 | 2058 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) && |
525f4ed8 MC |
2059 | (max_blocks > EXT4_MAX_TRANS_DATA)) |
2060 | max_blocks = EXT4_MAX_TRANS_DATA; | |
2061 | ||
2062 | return ext4_chunk_trans_blocks(inode, max_blocks); | |
2063 | } | |
61628a3f | 2064 | |
8e48dcfb TT |
2065 | /* |
2066 | * write_cache_pages_da - walk the list of dirty pages of the given | |
8eb9e5ce | 2067 | * address space and accumulate pages that need writing, and call |
168fc022 TT |
2068 | * mpage_da_map_and_submit to map a single contiguous memory region |
2069 | * and then write them. | |
8e48dcfb TT |
2070 | */ |
2071 | static int write_cache_pages_da(struct address_space *mapping, | |
2072 | struct writeback_control *wbc, | |
72f84e65 ES |
2073 | struct mpage_da_data *mpd, |
2074 | pgoff_t *done_index) | |
8e48dcfb | 2075 | { |
4f01b02c | 2076 | struct buffer_head *bh, *head; |
168fc022 | 2077 | struct inode *inode = mapping->host; |
4f01b02c TT |
2078 | struct pagevec pvec; |
2079 | unsigned int nr_pages; | |
2080 | sector_t logical; | |
2081 | pgoff_t index, end; | |
2082 | long nr_to_write = wbc->nr_to_write; | |
2083 | int i, tag, ret = 0; | |
8e48dcfb | 2084 | |
168fc022 TT |
2085 | memset(mpd, 0, sizeof(struct mpage_da_data)); |
2086 | mpd->wbc = wbc; | |
2087 | mpd->inode = inode; | |
8e48dcfb TT |
2088 | pagevec_init(&pvec, 0); |
2089 | index = wbc->range_start >> PAGE_CACHE_SHIFT; | |
2090 | end = wbc->range_end >> PAGE_CACHE_SHIFT; | |
2091 | ||
6e6938b6 | 2092 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
5b41d924 ES |
2093 | tag = PAGECACHE_TAG_TOWRITE; |
2094 | else | |
2095 | tag = PAGECACHE_TAG_DIRTY; | |
2096 | ||
72f84e65 | 2097 | *done_index = index; |
4f01b02c | 2098 | while (index <= end) { |
5b41d924 | 2099 | nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag, |
8e48dcfb TT |
2100 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1); |
2101 | if (nr_pages == 0) | |
4f01b02c | 2102 | return 0; |
8e48dcfb TT |
2103 | |
2104 | for (i = 0; i < nr_pages; i++) { | |
2105 | struct page *page = pvec.pages[i]; | |
2106 | ||
2107 | /* | |
2108 | * At this point, the page may be truncated or | |
2109 | * invalidated (changing page->mapping to NULL), or | |
2110 | * even swizzled back from swapper_space to tmpfs file | |
2111 | * mapping. However, page->index will not change | |
2112 | * because we have a reference on the page. | |
2113 | */ | |
4f01b02c TT |
2114 | if (page->index > end) |
2115 | goto out; | |
8e48dcfb | 2116 | |
72f84e65 ES |
2117 | *done_index = page->index + 1; |
2118 | ||
78aaced3 TT |
2119 | /* |
2120 | * If we can't merge this page, and we have | |
2121 | * accumulated an contiguous region, write it | |
2122 | */ | |
2123 | if ((mpd->next_page != page->index) && | |
2124 | (mpd->next_page != mpd->first_page)) { | |
2125 | mpage_da_map_and_submit(mpd); | |
2126 | goto ret_extent_tail; | |
2127 | } | |
2128 | ||
8e48dcfb TT |
2129 | lock_page(page); |
2130 | ||
2131 | /* | |
4f01b02c TT |
2132 | * If the page is no longer dirty, or its |
2133 | * mapping no longer corresponds to inode we | |
2134 | * are writing (which means it has been | |
2135 | * truncated or invalidated), or the page is | |
2136 | * already under writeback and we are not | |
2137 | * doing a data integrity writeback, skip the page | |
8e48dcfb | 2138 | */ |
4f01b02c TT |
2139 | if (!PageDirty(page) || |
2140 | (PageWriteback(page) && | |
2141 | (wbc->sync_mode == WB_SYNC_NONE)) || | |
2142 | unlikely(page->mapping != mapping)) { | |
8e48dcfb TT |
2143 | unlock_page(page); |
2144 | continue; | |
2145 | } | |
2146 | ||
7cb1a535 | 2147 | wait_on_page_writeback(page); |
8e48dcfb | 2148 | BUG_ON(PageWriteback(page)); |
8e48dcfb | 2149 | |
168fc022 | 2150 | if (mpd->next_page != page->index) |
8eb9e5ce | 2151 | mpd->first_page = page->index; |
8eb9e5ce TT |
2152 | mpd->next_page = page->index + 1; |
2153 | logical = (sector_t) page->index << | |
2154 | (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
2155 | ||
2156 | if (!page_has_buffers(page)) { | |
4f01b02c TT |
2157 | mpage_add_bh_to_extent(mpd, logical, |
2158 | PAGE_CACHE_SIZE, | |
8eb9e5ce | 2159 | (1 << BH_Dirty) | (1 << BH_Uptodate)); |
4f01b02c TT |
2160 | if (mpd->io_done) |
2161 | goto ret_extent_tail; | |
8eb9e5ce TT |
2162 | } else { |
2163 | /* | |
4f01b02c TT |
2164 | * Page with regular buffer heads, |
2165 | * just add all dirty ones | |
8eb9e5ce TT |
2166 | */ |
2167 | head = page_buffers(page); | |
2168 | bh = head; | |
2169 | do { | |
2170 | BUG_ON(buffer_locked(bh)); | |
2171 | /* | |
2172 | * We need to try to allocate | |
2173 | * unmapped blocks in the same page. | |
2174 | * Otherwise we won't make progress | |
2175 | * with the page in ext4_writepage | |
2176 | */ | |
2177 | if (ext4_bh_delay_or_unwritten(NULL, bh)) { | |
2178 | mpage_add_bh_to_extent(mpd, logical, | |
2179 | bh->b_size, | |
2180 | bh->b_state); | |
4f01b02c TT |
2181 | if (mpd->io_done) |
2182 | goto ret_extent_tail; | |
8eb9e5ce TT |
2183 | } else if (buffer_dirty(bh) && (buffer_mapped(bh))) { |
2184 | /* | |
4f01b02c TT |
2185 | * mapped dirty buffer. We need |
2186 | * to update the b_state | |
2187 | * because we look at b_state | |
2188 | * in mpage_da_map_blocks. We | |
2189 | * don't update b_size because | |
2190 | * if we find an unmapped | |
2191 | * buffer_head later we need to | |
2192 | * use the b_state flag of that | |
2193 | * buffer_head. | |
8eb9e5ce TT |
2194 | */ |
2195 | if (mpd->b_size == 0) | |
2196 | mpd->b_state = bh->b_state & BH_FLAGS; | |
2197 | } | |
2198 | logical++; | |
2199 | } while ((bh = bh->b_this_page) != head); | |
8e48dcfb TT |
2200 | } |
2201 | ||
2202 | if (nr_to_write > 0) { | |
2203 | nr_to_write--; | |
2204 | if (nr_to_write == 0 && | |
4f01b02c | 2205 | wbc->sync_mode == WB_SYNC_NONE) |
8e48dcfb TT |
2206 | /* |
2207 | * We stop writing back only if we are | |
2208 | * not doing integrity sync. In case of | |
2209 | * integrity sync we have to keep going | |
2210 | * because someone may be concurrently | |
2211 | * dirtying pages, and we might have | |
2212 | * synced a lot of newly appeared dirty | |
2213 | * pages, but have not synced all of the | |
2214 | * old dirty pages. | |
2215 | */ | |
4f01b02c | 2216 | goto out; |
8e48dcfb TT |
2217 | } |
2218 | } | |
2219 | pagevec_release(&pvec); | |
2220 | cond_resched(); | |
2221 | } | |
4f01b02c TT |
2222 | return 0; |
2223 | ret_extent_tail: | |
2224 | ret = MPAGE_DA_EXTENT_TAIL; | |
8eb9e5ce TT |
2225 | out: |
2226 | pagevec_release(&pvec); | |
2227 | cond_resched(); | |
8e48dcfb TT |
2228 | return ret; |
2229 | } | |
2230 | ||
2231 | ||
64769240 | 2232 | static int ext4_da_writepages(struct address_space *mapping, |
a1d6cc56 | 2233 | struct writeback_control *wbc) |
64769240 | 2234 | { |
22208ded AK |
2235 | pgoff_t index; |
2236 | int range_whole = 0; | |
61628a3f | 2237 | handle_t *handle = NULL; |
df22291f | 2238 | struct mpage_da_data mpd; |
5e745b04 | 2239 | struct inode *inode = mapping->host; |
498e5f24 | 2240 | int pages_written = 0; |
55138e0b | 2241 | unsigned int max_pages; |
2acf2c26 | 2242 | int range_cyclic, cycled = 1, io_done = 0; |
55138e0b TT |
2243 | int needed_blocks, ret = 0; |
2244 | long desired_nr_to_write, nr_to_writebump = 0; | |
de89de6e | 2245 | loff_t range_start = wbc->range_start; |
5e745b04 | 2246 | struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb); |
72f84e65 | 2247 | pgoff_t done_index = 0; |
5b41d924 | 2248 | pgoff_t end; |
1bce63d1 | 2249 | struct blk_plug plug; |
61628a3f | 2250 | |
9bffad1e | 2251 | trace_ext4_da_writepages(inode, wbc); |
ba80b101 | 2252 | |
61628a3f MC |
2253 | /* |
2254 | * No pages to write? This is mainly a kludge to avoid starting | |
2255 | * a transaction for special inodes like journal inode on last iput() | |
2256 | * because that could violate lock ordering on umount | |
2257 | */ | |
a1d6cc56 | 2258 | if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) |
61628a3f | 2259 | return 0; |
2a21e37e TT |
2260 | |
2261 | /* | |
2262 | * If the filesystem has aborted, it is read-only, so return | |
2263 | * right away instead of dumping stack traces later on that | |
2264 | * will obscure the real source of the problem. We test | |
4ab2f15b | 2265 | * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because |
2a21e37e TT |
2266 | * the latter could be true if the filesystem is mounted |
2267 | * read-only, and in that case, ext4_da_writepages should | |
2268 | * *never* be called, so if that ever happens, we would want | |
2269 | * the stack trace. | |
2270 | */ | |
4ab2f15b | 2271 | if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) |
2a21e37e TT |
2272 | return -EROFS; |
2273 | ||
22208ded AK |
2274 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
2275 | range_whole = 1; | |
61628a3f | 2276 | |
2acf2c26 AK |
2277 | range_cyclic = wbc->range_cyclic; |
2278 | if (wbc->range_cyclic) { | |
22208ded | 2279 | index = mapping->writeback_index; |
2acf2c26 AK |
2280 | if (index) |
2281 | cycled = 0; | |
2282 | wbc->range_start = index << PAGE_CACHE_SHIFT; | |
2283 | wbc->range_end = LLONG_MAX; | |
2284 | wbc->range_cyclic = 0; | |
5b41d924 ES |
2285 | end = -1; |
2286 | } else { | |
22208ded | 2287 | index = wbc->range_start >> PAGE_CACHE_SHIFT; |
5b41d924 ES |
2288 | end = wbc->range_end >> PAGE_CACHE_SHIFT; |
2289 | } | |
a1d6cc56 | 2290 | |
55138e0b TT |
2291 | /* |
2292 | * This works around two forms of stupidity. The first is in | |
2293 | * the writeback code, which caps the maximum number of pages | |
2294 | * written to be 1024 pages. This is wrong on multiple | |
2295 | * levels; different architectues have a different page size, | |
2296 | * which changes the maximum amount of data which gets | |
2297 | * written. Secondly, 4 megabytes is way too small. XFS | |
2298 | * forces this value to be 16 megabytes by multiplying | |
2299 | * nr_to_write parameter by four, and then relies on its | |
2300 | * allocator to allocate larger extents to make them | |
2301 | * contiguous. Unfortunately this brings us to the second | |
2302 | * stupidity, which is that ext4's mballoc code only allocates | |
2303 | * at most 2048 blocks. So we force contiguous writes up to | |
2304 | * the number of dirty blocks in the inode, or | |
2305 | * sbi->max_writeback_mb_bump whichever is smaller. | |
2306 | */ | |
2307 | max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT); | |
b443e733 ES |
2308 | if (!range_cyclic && range_whole) { |
2309 | if (wbc->nr_to_write == LONG_MAX) | |
2310 | desired_nr_to_write = wbc->nr_to_write; | |
2311 | else | |
2312 | desired_nr_to_write = wbc->nr_to_write * 8; | |
2313 | } else | |
55138e0b TT |
2314 | desired_nr_to_write = ext4_num_dirty_pages(inode, index, |
2315 | max_pages); | |
2316 | if (desired_nr_to_write > max_pages) | |
2317 | desired_nr_to_write = max_pages; | |
2318 | ||
2319 | if (wbc->nr_to_write < desired_nr_to_write) { | |
2320 | nr_to_writebump = desired_nr_to_write - wbc->nr_to_write; | |
2321 | wbc->nr_to_write = desired_nr_to_write; | |
2322 | } | |
2323 | ||
2acf2c26 | 2324 | retry: |
6e6938b6 | 2325 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
5b41d924 ES |
2326 | tag_pages_for_writeback(mapping, index, end); |
2327 | ||
1bce63d1 | 2328 | blk_start_plug(&plug); |
22208ded | 2329 | while (!ret && wbc->nr_to_write > 0) { |
a1d6cc56 AK |
2330 | |
2331 | /* | |
2332 | * we insert one extent at a time. So we need | |
2333 | * credit needed for single extent allocation. | |
2334 | * journalled mode is currently not supported | |
2335 | * by delalloc | |
2336 | */ | |
2337 | BUG_ON(ext4_should_journal_data(inode)); | |
525f4ed8 | 2338 | needed_blocks = ext4_da_writepages_trans_blocks(inode); |
a1d6cc56 | 2339 | |
61628a3f MC |
2340 | /* start a new transaction*/ |
2341 | handle = ext4_journal_start(inode, needed_blocks); | |
2342 | if (IS_ERR(handle)) { | |
2343 | ret = PTR_ERR(handle); | |
1693918e | 2344 | ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: " |
fbe845dd | 2345 | "%ld pages, ino %lu; err %d", __func__, |
a1d6cc56 | 2346 | wbc->nr_to_write, inode->i_ino, ret); |
3c1fcb2c | 2347 | blk_finish_plug(&plug); |
61628a3f MC |
2348 | goto out_writepages; |
2349 | } | |
f63e6005 TT |
2350 | |
2351 | /* | |
8eb9e5ce | 2352 | * Now call write_cache_pages_da() to find the next |
f63e6005 | 2353 | * contiguous region of logical blocks that need |
8eb9e5ce | 2354 | * blocks to be allocated by ext4 and submit them. |
f63e6005 | 2355 | */ |
72f84e65 | 2356 | ret = write_cache_pages_da(mapping, wbc, &mpd, &done_index); |
f63e6005 | 2357 | /* |
af901ca1 | 2358 | * If we have a contiguous extent of pages and we |
f63e6005 TT |
2359 | * haven't done the I/O yet, map the blocks and submit |
2360 | * them for I/O. | |
2361 | */ | |
2362 | if (!mpd.io_done && mpd.next_page != mpd.first_page) { | |
5a87b7a5 | 2363 | mpage_da_map_and_submit(&mpd); |
f63e6005 TT |
2364 | ret = MPAGE_DA_EXTENT_TAIL; |
2365 | } | |
b3a3ca8c | 2366 | trace_ext4_da_write_pages(inode, &mpd); |
f63e6005 | 2367 | wbc->nr_to_write -= mpd.pages_written; |
df22291f | 2368 | |
61628a3f | 2369 | ext4_journal_stop(handle); |
df22291f | 2370 | |
8f64b32e | 2371 | if ((mpd.retval == -ENOSPC) && sbi->s_journal) { |
22208ded AK |
2372 | /* commit the transaction which would |
2373 | * free blocks released in the transaction | |
2374 | * and try again | |
2375 | */ | |
df22291f | 2376 | jbd2_journal_force_commit_nested(sbi->s_journal); |
22208ded AK |
2377 | ret = 0; |
2378 | } else if (ret == MPAGE_DA_EXTENT_TAIL) { | |
a1d6cc56 | 2379 | /* |
8de49e67 KM |
2380 | * Got one extent now try with rest of the pages. |
2381 | * If mpd.retval is set -EIO, journal is aborted. | |
2382 | * So we don't need to write any more. | |
a1d6cc56 | 2383 | */ |
22208ded | 2384 | pages_written += mpd.pages_written; |
8de49e67 | 2385 | ret = mpd.retval; |
2acf2c26 | 2386 | io_done = 1; |
22208ded | 2387 | } else if (wbc->nr_to_write) |
61628a3f MC |
2388 | /* |
2389 | * There is no more writeout needed | |
2390 | * or we requested for a noblocking writeout | |
2391 | * and we found the device congested | |
2392 | */ | |
61628a3f | 2393 | break; |
a1d6cc56 | 2394 | } |
1bce63d1 | 2395 | blk_finish_plug(&plug); |
2acf2c26 AK |
2396 | if (!io_done && !cycled) { |
2397 | cycled = 1; | |
2398 | index = 0; | |
2399 | wbc->range_start = index << PAGE_CACHE_SHIFT; | |
2400 | wbc->range_end = mapping->writeback_index - 1; | |
2401 | goto retry; | |
2402 | } | |
22208ded AK |
2403 | |
2404 | /* Update index */ | |
2acf2c26 | 2405 | wbc->range_cyclic = range_cyclic; |
22208ded AK |
2406 | if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) |
2407 | /* | |
2408 | * set the writeback_index so that range_cyclic | |
2409 | * mode will write it back later | |
2410 | */ | |
72f84e65 | 2411 | mapping->writeback_index = done_index; |
a1d6cc56 | 2412 | |
61628a3f | 2413 | out_writepages: |
2faf2e19 | 2414 | wbc->nr_to_write -= nr_to_writebump; |
de89de6e | 2415 | wbc->range_start = range_start; |
9bffad1e | 2416 | trace_ext4_da_writepages_result(inode, wbc, ret, pages_written); |
61628a3f | 2417 | return ret; |
64769240 AT |
2418 | } |
2419 | ||
79f0be8d AK |
2420 | #define FALL_BACK_TO_NONDELALLOC 1 |
2421 | static int ext4_nonda_switch(struct super_block *sb) | |
2422 | { | |
2423 | s64 free_blocks, dirty_blocks; | |
2424 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
2425 | ||
2426 | /* | |
2427 | * switch to non delalloc mode if we are running low | |
2428 | * on free block. The free block accounting via percpu | |
179f7ebf | 2429 | * counters can get slightly wrong with percpu_counter_batch getting |
79f0be8d AK |
2430 | * accumulated on each CPU without updating global counters |
2431 | * Delalloc need an accurate free block accounting. So switch | |
2432 | * to non delalloc when we are near to error range. | |
2433 | */ | |
57042651 TT |
2434 | free_blocks = EXT4_C2B(sbi, |
2435 | percpu_counter_read_positive(&sbi->s_freeclusters_counter)); | |
2436 | dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyclusters_counter); | |
79f0be8d | 2437 | if (2 * free_blocks < 3 * dirty_blocks || |
df55c99d | 2438 | free_blocks < (dirty_blocks + EXT4_FREECLUSTERS_WATERMARK)) { |
79f0be8d | 2439 | /* |
c8afb446 ES |
2440 | * free block count is less than 150% of dirty blocks |
2441 | * or free blocks is less than watermark | |
79f0be8d AK |
2442 | */ |
2443 | return 1; | |
2444 | } | |
c8afb446 ES |
2445 | /* |
2446 | * Even if we don't switch but are nearing capacity, | |
2447 | * start pushing delalloc when 1/2 of free blocks are dirty. | |
2448 | */ | |
2449 | if (free_blocks < 2 * dirty_blocks) | |
0e175a18 | 2450 | writeback_inodes_sb_if_idle(sb, WB_REASON_FS_FREE_SPACE); |
c8afb446 | 2451 | |
79f0be8d AK |
2452 | return 0; |
2453 | } | |
2454 | ||
64769240 | 2455 | static int ext4_da_write_begin(struct file *file, struct address_space *mapping, |
de9a55b8 TT |
2456 | loff_t pos, unsigned len, unsigned flags, |
2457 | struct page **pagep, void **fsdata) | |
64769240 | 2458 | { |
72b8ab9d | 2459 | int ret, retries = 0; |
64769240 AT |
2460 | struct page *page; |
2461 | pgoff_t index; | |
64769240 AT |
2462 | struct inode *inode = mapping->host; |
2463 | handle_t *handle; | |
2464 | ||
2465 | index = pos >> PAGE_CACHE_SHIFT; | |
79f0be8d AK |
2466 | |
2467 | if (ext4_nonda_switch(inode->i_sb)) { | |
2468 | *fsdata = (void *)FALL_BACK_TO_NONDELALLOC; | |
2469 | return ext4_write_begin(file, mapping, pos, | |
2470 | len, flags, pagep, fsdata); | |
2471 | } | |
2472 | *fsdata = (void *)0; | |
9bffad1e | 2473 | trace_ext4_da_write_begin(inode, pos, len, flags); |
d2a17637 | 2474 | retry: |
64769240 AT |
2475 | /* |
2476 | * With delayed allocation, we don't log the i_disksize update | |
2477 | * if there is delayed block allocation. But we still need | |
2478 | * to journalling the i_disksize update if writes to the end | |
2479 | * of file which has an already mapped buffer. | |
2480 | */ | |
2481 | handle = ext4_journal_start(inode, 1); | |
2482 | if (IS_ERR(handle)) { | |
2483 | ret = PTR_ERR(handle); | |
2484 | goto out; | |
2485 | } | |
ebd3610b JK |
2486 | /* We cannot recurse into the filesystem as the transaction is already |
2487 | * started */ | |
2488 | flags |= AOP_FLAG_NOFS; | |
64769240 | 2489 | |
54566b2c | 2490 | page = grab_cache_page_write_begin(mapping, index, flags); |
d5a0d4f7 ES |
2491 | if (!page) { |
2492 | ext4_journal_stop(handle); | |
2493 | ret = -ENOMEM; | |
2494 | goto out; | |
2495 | } | |
64769240 AT |
2496 | *pagep = page; |
2497 | ||
6e1db88d | 2498 | ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep); |
64769240 AT |
2499 | if (ret < 0) { |
2500 | unlock_page(page); | |
2501 | ext4_journal_stop(handle); | |
2502 | page_cache_release(page); | |
ae4d5372 AK |
2503 | /* |
2504 | * block_write_begin may have instantiated a few blocks | |
2505 | * outside i_size. Trim these off again. Don't need | |
2506 | * i_size_read because we hold i_mutex. | |
2507 | */ | |
2508 | if (pos + len > inode->i_size) | |
b9a4207d | 2509 | ext4_truncate_failed_write(inode); |
64769240 AT |
2510 | } |
2511 | ||
d2a17637 MC |
2512 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) |
2513 | goto retry; | |
64769240 AT |
2514 | out: |
2515 | return ret; | |
2516 | } | |
2517 | ||
632eaeab MC |
2518 | /* |
2519 | * Check if we should update i_disksize | |
2520 | * when write to the end of file but not require block allocation | |
2521 | */ | |
2522 | static int ext4_da_should_update_i_disksize(struct page *page, | |
de9a55b8 | 2523 | unsigned long offset) |
632eaeab MC |
2524 | { |
2525 | struct buffer_head *bh; | |
2526 | struct inode *inode = page->mapping->host; | |
2527 | unsigned int idx; | |
2528 | int i; | |
2529 | ||
2530 | bh = page_buffers(page); | |
2531 | idx = offset >> inode->i_blkbits; | |
2532 | ||
af5bc92d | 2533 | for (i = 0; i < idx; i++) |
632eaeab MC |
2534 | bh = bh->b_this_page; |
2535 | ||
29fa89d0 | 2536 | if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh)) |
632eaeab MC |
2537 | return 0; |
2538 | return 1; | |
2539 | } | |
2540 | ||
64769240 | 2541 | static int ext4_da_write_end(struct file *file, |
de9a55b8 TT |
2542 | struct address_space *mapping, |
2543 | loff_t pos, unsigned len, unsigned copied, | |
2544 | struct page *page, void *fsdata) | |
64769240 AT |
2545 | { |
2546 | struct inode *inode = mapping->host; | |
2547 | int ret = 0, ret2; | |
2548 | handle_t *handle = ext4_journal_current_handle(); | |
2549 | loff_t new_i_size; | |
632eaeab | 2550 | unsigned long start, end; |
79f0be8d AK |
2551 | int write_mode = (int)(unsigned long)fsdata; |
2552 | ||
2553 | if (write_mode == FALL_BACK_TO_NONDELALLOC) { | |
3d2b1582 LC |
2554 | switch (ext4_inode_journal_mode(inode)) { |
2555 | case EXT4_INODE_ORDERED_DATA_MODE: | |
79f0be8d AK |
2556 | return ext4_ordered_write_end(file, mapping, pos, |
2557 | len, copied, page, fsdata); | |
3d2b1582 | 2558 | case EXT4_INODE_WRITEBACK_DATA_MODE: |
79f0be8d AK |
2559 | return ext4_writeback_write_end(file, mapping, pos, |
2560 | len, copied, page, fsdata); | |
3d2b1582 | 2561 | default: |
79f0be8d AK |
2562 | BUG(); |
2563 | } | |
2564 | } | |
632eaeab | 2565 | |
9bffad1e | 2566 | trace_ext4_da_write_end(inode, pos, len, copied); |
632eaeab | 2567 | start = pos & (PAGE_CACHE_SIZE - 1); |
af5bc92d | 2568 | end = start + copied - 1; |
64769240 AT |
2569 | |
2570 | /* | |
2571 | * generic_write_end() will run mark_inode_dirty() if i_size | |
2572 | * changes. So let's piggyback the i_disksize mark_inode_dirty | |
2573 | * into that. | |
2574 | */ | |
2575 | ||
2576 | new_i_size = pos + copied; | |
ea51d132 | 2577 | if (copied && new_i_size > EXT4_I(inode)->i_disksize) { |
632eaeab MC |
2578 | if (ext4_da_should_update_i_disksize(page, end)) { |
2579 | down_write(&EXT4_I(inode)->i_data_sem); | |
2580 | if (new_i_size > EXT4_I(inode)->i_disksize) { | |
2581 | /* | |
2582 | * Updating i_disksize when extending file | |
2583 | * without needing block allocation | |
2584 | */ | |
2585 | if (ext4_should_order_data(inode)) | |
2586 | ret = ext4_jbd2_file_inode(handle, | |
2587 | inode); | |
64769240 | 2588 | |
632eaeab MC |
2589 | EXT4_I(inode)->i_disksize = new_i_size; |
2590 | } | |
2591 | up_write(&EXT4_I(inode)->i_data_sem); | |
cf17fea6 AK |
2592 | /* We need to mark inode dirty even if |
2593 | * new_i_size is less that inode->i_size | |
2594 | * bu greater than i_disksize.(hint delalloc) | |
2595 | */ | |
2596 | ext4_mark_inode_dirty(handle, inode); | |
64769240 | 2597 | } |
632eaeab | 2598 | } |
64769240 AT |
2599 | ret2 = generic_write_end(file, mapping, pos, len, copied, |
2600 | page, fsdata); | |
2601 | copied = ret2; | |
2602 | if (ret2 < 0) | |
2603 | ret = ret2; | |
2604 | ret2 = ext4_journal_stop(handle); | |
2605 | if (!ret) | |
2606 | ret = ret2; | |
2607 | ||
2608 | return ret ? ret : copied; | |
2609 | } | |
2610 | ||
2611 | static void ext4_da_invalidatepage(struct page *page, unsigned long offset) | |
2612 | { | |
64769240 AT |
2613 | /* |
2614 | * Drop reserved blocks | |
2615 | */ | |
2616 | BUG_ON(!PageLocked(page)); | |
2617 | if (!page_has_buffers(page)) | |
2618 | goto out; | |
2619 | ||
d2a17637 | 2620 | ext4_da_page_release_reservation(page, offset); |
64769240 AT |
2621 | |
2622 | out: | |
2623 | ext4_invalidatepage(page, offset); | |
2624 | ||
2625 | return; | |
2626 | } | |
2627 | ||
ccd2506b TT |
2628 | /* |
2629 | * Force all delayed allocation blocks to be allocated for a given inode. | |
2630 | */ | |
2631 | int ext4_alloc_da_blocks(struct inode *inode) | |
2632 | { | |
fb40ba0d TT |
2633 | trace_ext4_alloc_da_blocks(inode); |
2634 | ||
ccd2506b TT |
2635 | if (!EXT4_I(inode)->i_reserved_data_blocks && |
2636 | !EXT4_I(inode)->i_reserved_meta_blocks) | |
2637 | return 0; | |
2638 | ||
2639 | /* | |
2640 | * We do something simple for now. The filemap_flush() will | |
2641 | * also start triggering a write of the data blocks, which is | |
2642 | * not strictly speaking necessary (and for users of | |
2643 | * laptop_mode, not even desirable). However, to do otherwise | |
2644 | * would require replicating code paths in: | |
de9a55b8 | 2645 | * |
ccd2506b TT |
2646 | * ext4_da_writepages() -> |
2647 | * write_cache_pages() ---> (via passed in callback function) | |
2648 | * __mpage_da_writepage() --> | |
2649 | * mpage_add_bh_to_extent() | |
2650 | * mpage_da_map_blocks() | |
2651 | * | |
2652 | * The problem is that write_cache_pages(), located in | |
2653 | * mm/page-writeback.c, marks pages clean in preparation for | |
2654 | * doing I/O, which is not desirable if we're not planning on | |
2655 | * doing I/O at all. | |
2656 | * | |
2657 | * We could call write_cache_pages(), and then redirty all of | |
380cf090 | 2658 | * the pages by calling redirty_page_for_writepage() but that |
ccd2506b TT |
2659 | * would be ugly in the extreme. So instead we would need to |
2660 | * replicate parts of the code in the above functions, | |
25985edc | 2661 | * simplifying them because we wouldn't actually intend to |
ccd2506b TT |
2662 | * write out the pages, but rather only collect contiguous |
2663 | * logical block extents, call the multi-block allocator, and | |
2664 | * then update the buffer heads with the block allocations. | |
de9a55b8 | 2665 | * |
ccd2506b TT |
2666 | * For now, though, we'll cheat by calling filemap_flush(), |
2667 | * which will map the blocks, and start the I/O, but not | |
2668 | * actually wait for the I/O to complete. | |
2669 | */ | |
2670 | return filemap_flush(inode->i_mapping); | |
2671 | } | |
64769240 | 2672 | |
ac27a0ec DK |
2673 | /* |
2674 | * bmap() is special. It gets used by applications such as lilo and by | |
2675 | * the swapper to find the on-disk block of a specific piece of data. | |
2676 | * | |
2677 | * Naturally, this is dangerous if the block concerned is still in the | |
617ba13b | 2678 | * journal. If somebody makes a swapfile on an ext4 data-journaling |
ac27a0ec DK |
2679 | * filesystem and enables swap, then they may get a nasty shock when the |
2680 | * data getting swapped to that swapfile suddenly gets overwritten by | |
2681 | * the original zero's written out previously to the journal and | |
2682 | * awaiting writeback in the kernel's buffer cache. | |
2683 | * | |
2684 | * So, if we see any bmap calls here on a modified, data-journaled file, | |
2685 | * take extra steps to flush any blocks which might be in the cache. | |
2686 | */ | |
617ba13b | 2687 | static sector_t ext4_bmap(struct address_space *mapping, sector_t block) |
ac27a0ec DK |
2688 | { |
2689 | struct inode *inode = mapping->host; | |
2690 | journal_t *journal; | |
2691 | int err; | |
2692 | ||
64769240 AT |
2693 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) && |
2694 | test_opt(inode->i_sb, DELALLOC)) { | |
2695 | /* | |
2696 | * With delalloc we want to sync the file | |
2697 | * so that we can make sure we allocate | |
2698 | * blocks for file | |
2699 | */ | |
2700 | filemap_write_and_wait(mapping); | |
2701 | } | |
2702 | ||
19f5fb7a TT |
2703 | if (EXT4_JOURNAL(inode) && |
2704 | ext4_test_inode_state(inode, EXT4_STATE_JDATA)) { | |
ac27a0ec DK |
2705 | /* |
2706 | * This is a REALLY heavyweight approach, but the use of | |
2707 | * bmap on dirty files is expected to be extremely rare: | |
2708 | * only if we run lilo or swapon on a freshly made file | |
2709 | * do we expect this to happen. | |
2710 | * | |
2711 | * (bmap requires CAP_SYS_RAWIO so this does not | |
2712 | * represent an unprivileged user DOS attack --- we'd be | |
2713 | * in trouble if mortal users could trigger this path at | |
2714 | * will.) | |
2715 | * | |
617ba13b | 2716 | * NB. EXT4_STATE_JDATA is not set on files other than |
ac27a0ec DK |
2717 | * regular files. If somebody wants to bmap a directory |
2718 | * or symlink and gets confused because the buffer | |
2719 | * hasn't yet been flushed to disk, they deserve | |
2720 | * everything they get. | |
2721 | */ | |
2722 | ||
19f5fb7a | 2723 | ext4_clear_inode_state(inode, EXT4_STATE_JDATA); |
617ba13b | 2724 | journal = EXT4_JOURNAL(inode); |
dab291af MC |
2725 | jbd2_journal_lock_updates(journal); |
2726 | err = jbd2_journal_flush(journal); | |
2727 | jbd2_journal_unlock_updates(journal); | |
ac27a0ec DK |
2728 | |
2729 | if (err) | |
2730 | return 0; | |
2731 | } | |
2732 | ||
af5bc92d | 2733 | return generic_block_bmap(mapping, block, ext4_get_block); |
ac27a0ec DK |
2734 | } |
2735 | ||
617ba13b | 2736 | static int ext4_readpage(struct file *file, struct page *page) |
ac27a0ec | 2737 | { |
0562e0ba | 2738 | trace_ext4_readpage(page); |
617ba13b | 2739 | return mpage_readpage(page, ext4_get_block); |
ac27a0ec DK |
2740 | } |
2741 | ||
2742 | static int | |
617ba13b | 2743 | ext4_readpages(struct file *file, struct address_space *mapping, |
ac27a0ec DK |
2744 | struct list_head *pages, unsigned nr_pages) |
2745 | { | |
617ba13b | 2746 | return mpage_readpages(mapping, pages, nr_pages, ext4_get_block); |
ac27a0ec DK |
2747 | } |
2748 | ||
744692dc JZ |
2749 | static void ext4_invalidatepage_free_endio(struct page *page, unsigned long offset) |
2750 | { | |
2751 | struct buffer_head *head, *bh; | |
2752 | unsigned int curr_off = 0; | |
2753 | ||
2754 | if (!page_has_buffers(page)) | |
2755 | return; | |
2756 | head = bh = page_buffers(page); | |
2757 | do { | |
2758 | if (offset <= curr_off && test_clear_buffer_uninit(bh) | |
2759 | && bh->b_private) { | |
2760 | ext4_free_io_end(bh->b_private); | |
2761 | bh->b_private = NULL; | |
2762 | bh->b_end_io = NULL; | |
2763 | } | |
2764 | curr_off = curr_off + bh->b_size; | |
2765 | bh = bh->b_this_page; | |
2766 | } while (bh != head); | |
2767 | } | |
2768 | ||
617ba13b | 2769 | static void ext4_invalidatepage(struct page *page, unsigned long offset) |
ac27a0ec | 2770 | { |
617ba13b | 2771 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); |
ac27a0ec | 2772 | |
0562e0ba JZ |
2773 | trace_ext4_invalidatepage(page, offset); |
2774 | ||
744692dc JZ |
2775 | /* |
2776 | * free any io_end structure allocated for buffers to be discarded | |
2777 | */ | |
2778 | if (ext4_should_dioread_nolock(page->mapping->host)) | |
2779 | ext4_invalidatepage_free_endio(page, offset); | |
ac27a0ec DK |
2780 | /* |
2781 | * If it's a full truncate we just forget about the pending dirtying | |
2782 | */ | |
2783 | if (offset == 0) | |
2784 | ClearPageChecked(page); | |
2785 | ||
0390131b FM |
2786 | if (journal) |
2787 | jbd2_journal_invalidatepage(journal, page, offset); | |
2788 | else | |
2789 | block_invalidatepage(page, offset); | |
ac27a0ec DK |
2790 | } |
2791 | ||
617ba13b | 2792 | static int ext4_releasepage(struct page *page, gfp_t wait) |
ac27a0ec | 2793 | { |
617ba13b | 2794 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); |
ac27a0ec | 2795 | |
0562e0ba JZ |
2796 | trace_ext4_releasepage(page); |
2797 | ||
ac27a0ec DK |
2798 | WARN_ON(PageChecked(page)); |
2799 | if (!page_has_buffers(page)) | |
2800 | return 0; | |
0390131b FM |
2801 | if (journal) |
2802 | return jbd2_journal_try_to_free_buffers(journal, page, wait); | |
2803 | else | |
2804 | return try_to_free_buffers(page); | |
ac27a0ec DK |
2805 | } |
2806 | ||
2ed88685 TT |
2807 | /* |
2808 | * ext4_get_block used when preparing for a DIO write or buffer write. | |
2809 | * We allocate an uinitialized extent if blocks haven't been allocated. | |
2810 | * The extent will be converted to initialized after the IO is complete. | |
2811 | */ | |
c7064ef1 | 2812 | static int ext4_get_block_write(struct inode *inode, sector_t iblock, |
4c0425ff MC |
2813 | struct buffer_head *bh_result, int create) |
2814 | { | |
c7064ef1 | 2815 | ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n", |
8d5d02e6 | 2816 | inode->i_ino, create); |
2ed88685 TT |
2817 | return _ext4_get_block(inode, iblock, bh_result, |
2818 | EXT4_GET_BLOCKS_IO_CREATE_EXT); | |
4c0425ff MC |
2819 | } |
2820 | ||
4c0425ff | 2821 | static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset, |
552ef802 CH |
2822 | ssize_t size, void *private, int ret, |
2823 | bool is_async) | |
4c0425ff | 2824 | { |
72c5052d | 2825 | struct inode *inode = iocb->ki_filp->f_path.dentry->d_inode; |
4c0425ff MC |
2826 | ext4_io_end_t *io_end = iocb->private; |
2827 | struct workqueue_struct *wq; | |
744692dc JZ |
2828 | unsigned long flags; |
2829 | struct ext4_inode_info *ei; | |
4c0425ff | 2830 | |
4b70df18 M |
2831 | /* if not async direct IO or dio with 0 bytes write, just return */ |
2832 | if (!io_end || !size) | |
552ef802 | 2833 | goto out; |
4b70df18 | 2834 | |
88635ca2 | 2835 | ext_debug("ext4_end_io_dio(): io_end 0x%p " |
ace36ad4 | 2836 | "for inode %lu, iocb 0x%p, offset %llu, size %zd\n", |
8d5d02e6 MC |
2837 | iocb->private, io_end->inode->i_ino, iocb, offset, |
2838 | size); | |
8d5d02e6 | 2839 | |
b5a7e970 TT |
2840 | iocb->private = NULL; |
2841 | ||
8d5d02e6 | 2842 | /* if not aio dio with unwritten extents, just free io and return */ |
bd2d0210 | 2843 | if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) { |
8d5d02e6 | 2844 | ext4_free_io_end(io_end); |
5b3ff237 JZ |
2845 | out: |
2846 | if (is_async) | |
2847 | aio_complete(iocb, ret, 0); | |
72c5052d | 2848 | inode_dio_done(inode); |
5b3ff237 | 2849 | return; |
8d5d02e6 MC |
2850 | } |
2851 | ||
4c0425ff MC |
2852 | io_end->offset = offset; |
2853 | io_end->size = size; | |
5b3ff237 JZ |
2854 | if (is_async) { |
2855 | io_end->iocb = iocb; | |
2856 | io_end->result = ret; | |
2857 | } | |
4c0425ff MC |
2858 | wq = EXT4_SB(io_end->inode->i_sb)->dio_unwritten_wq; |
2859 | ||
8d5d02e6 | 2860 | /* Add the io_end to per-inode completed aio dio list*/ |
744692dc JZ |
2861 | ei = EXT4_I(io_end->inode); |
2862 | spin_lock_irqsave(&ei->i_completed_io_lock, flags); | |
2863 | list_add_tail(&io_end->list, &ei->i_completed_io_list); | |
2864 | spin_unlock_irqrestore(&ei->i_completed_io_lock, flags); | |
c999af2b ES |
2865 | |
2866 | /* queue the work to convert unwritten extents to written */ | |
4c81f045 | 2867 | queue_work(wq, &io_end->work); |
4c0425ff | 2868 | } |
c7064ef1 | 2869 | |
744692dc JZ |
2870 | static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate) |
2871 | { | |
2872 | ext4_io_end_t *io_end = bh->b_private; | |
2873 | struct workqueue_struct *wq; | |
2874 | struct inode *inode; | |
2875 | unsigned long flags; | |
2876 | ||
2877 | if (!test_clear_buffer_uninit(bh) || !io_end) | |
2878 | goto out; | |
2879 | ||
2880 | if (!(io_end->inode->i_sb->s_flags & MS_ACTIVE)) { | |
92b97816 TT |
2881 | ext4_msg(io_end->inode->i_sb, KERN_INFO, |
2882 | "sb umounted, discard end_io request for inode %lu", | |
2883 | io_end->inode->i_ino); | |
744692dc JZ |
2884 | ext4_free_io_end(io_end); |
2885 | goto out; | |
2886 | } | |
2887 | ||
32c80b32 TM |
2888 | /* |
2889 | * It may be over-defensive here to check EXT4_IO_END_UNWRITTEN now, | |
2890 | * but being more careful is always safe for the future change. | |
2891 | */ | |
744692dc | 2892 | inode = io_end->inode; |
0edeb71d | 2893 | ext4_set_io_unwritten_flag(inode, io_end); |
744692dc JZ |
2894 | |
2895 | /* Add the io_end to per-inode completed io list*/ | |
2896 | spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags); | |
2897 | list_add_tail(&io_end->list, &EXT4_I(inode)->i_completed_io_list); | |
2898 | spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags); | |
2899 | ||
2900 | wq = EXT4_SB(inode->i_sb)->dio_unwritten_wq; | |
2901 | /* queue the work to convert unwritten extents to written */ | |
2902 | queue_work(wq, &io_end->work); | |
2903 | out: | |
2904 | bh->b_private = NULL; | |
2905 | bh->b_end_io = NULL; | |
2906 | clear_buffer_uninit(bh); | |
2907 | end_buffer_async_write(bh, uptodate); | |
2908 | } | |
2909 | ||
2910 | static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode) | |
2911 | { | |
2912 | ext4_io_end_t *io_end; | |
2913 | struct page *page = bh->b_page; | |
2914 | loff_t offset = (sector_t)page->index << PAGE_CACHE_SHIFT; | |
2915 | size_t size = bh->b_size; | |
2916 | ||
2917 | retry: | |
2918 | io_end = ext4_init_io_end(inode, GFP_ATOMIC); | |
2919 | if (!io_end) { | |
6db26ffc | 2920 | pr_warn_ratelimited("%s: allocation fail\n", __func__); |
744692dc JZ |
2921 | schedule(); |
2922 | goto retry; | |
2923 | } | |
2924 | io_end->offset = offset; | |
2925 | io_end->size = size; | |
2926 | /* | |
2927 | * We need to hold a reference to the page to make sure it | |
2928 | * doesn't get evicted before ext4_end_io_work() has a chance | |
2929 | * to convert the extent from written to unwritten. | |
2930 | */ | |
2931 | io_end->page = page; | |
2932 | get_page(io_end->page); | |
2933 | ||
2934 | bh->b_private = io_end; | |
2935 | bh->b_end_io = ext4_end_io_buffer_write; | |
2936 | return 0; | |
2937 | } | |
2938 | ||
4c0425ff MC |
2939 | /* |
2940 | * For ext4 extent files, ext4 will do direct-io write to holes, | |
2941 | * preallocated extents, and those write extend the file, no need to | |
2942 | * fall back to buffered IO. | |
2943 | * | |
b595076a | 2944 | * For holes, we fallocate those blocks, mark them as uninitialized |
4c0425ff | 2945 | * If those blocks were preallocated, we mark sure they are splited, but |
b595076a | 2946 | * still keep the range to write as uninitialized. |
4c0425ff | 2947 | * |
8d5d02e6 MC |
2948 | * The unwrritten extents will be converted to written when DIO is completed. |
2949 | * For async direct IO, since the IO may still pending when return, we | |
25985edc | 2950 | * set up an end_io call back function, which will do the conversion |
8d5d02e6 | 2951 | * when async direct IO completed. |
4c0425ff MC |
2952 | * |
2953 | * If the O_DIRECT write will extend the file then add this inode to the | |
2954 | * orphan list. So recovery will truncate it back to the original size | |
2955 | * if the machine crashes during the write. | |
2956 | * | |
2957 | */ | |
2958 | static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb, | |
2959 | const struct iovec *iov, loff_t offset, | |
2960 | unsigned long nr_segs) | |
2961 | { | |
2962 | struct file *file = iocb->ki_filp; | |
2963 | struct inode *inode = file->f_mapping->host; | |
2964 | ssize_t ret; | |
2965 | size_t count = iov_length(iov, nr_segs); | |
2966 | ||
2967 | loff_t final_size = offset + count; | |
2968 | if (rw == WRITE && final_size <= inode->i_size) { | |
2969 | /* | |
8d5d02e6 MC |
2970 | * We could direct write to holes and fallocate. |
2971 | * | |
2972 | * Allocated blocks to fill the hole are marked as uninitialized | |
25985edc | 2973 | * to prevent parallel buffered read to expose the stale data |
4c0425ff | 2974 | * before DIO complete the data IO. |
8d5d02e6 MC |
2975 | * |
2976 | * As to previously fallocated extents, ext4 get_block | |
4c0425ff MC |
2977 | * will just simply mark the buffer mapped but still |
2978 | * keep the extents uninitialized. | |
2979 | * | |
8d5d02e6 MC |
2980 | * for non AIO case, we will convert those unwritten extents |
2981 | * to written after return back from blockdev_direct_IO. | |
2982 | * | |
2983 | * for async DIO, the conversion needs to be defered when | |
2984 | * the IO is completed. The ext4 end_io callback function | |
2985 | * will be called to take care of the conversion work. | |
2986 | * Here for async case, we allocate an io_end structure to | |
2987 | * hook to the iocb. | |
4c0425ff | 2988 | */ |
8d5d02e6 MC |
2989 | iocb->private = NULL; |
2990 | EXT4_I(inode)->cur_aio_dio = NULL; | |
2991 | if (!is_sync_kiocb(iocb)) { | |
266991b1 JM |
2992 | ext4_io_end_t *io_end = |
2993 | ext4_init_io_end(inode, GFP_NOFS); | |
2994 | if (!io_end) | |
8d5d02e6 | 2995 | return -ENOMEM; |
266991b1 JM |
2996 | io_end->flag |= EXT4_IO_END_DIRECT; |
2997 | iocb->private = io_end; | |
8d5d02e6 MC |
2998 | /* |
2999 | * we save the io structure for current async | |
79e83036 | 3000 | * direct IO, so that later ext4_map_blocks() |
8d5d02e6 MC |
3001 | * could flag the io structure whether there |
3002 | * is a unwritten extents needs to be converted | |
3003 | * when IO is completed. | |
3004 | */ | |
3005 | EXT4_I(inode)->cur_aio_dio = iocb->private; | |
3006 | } | |
3007 | ||
aacfc19c | 3008 | ret = __blockdev_direct_IO(rw, iocb, inode, |
4c0425ff MC |
3009 | inode->i_sb->s_bdev, iov, |
3010 | offset, nr_segs, | |
c7064ef1 | 3011 | ext4_get_block_write, |
aacfc19c CH |
3012 | ext4_end_io_dio, |
3013 | NULL, | |
93ef8541 | 3014 | DIO_LOCKING); |
8d5d02e6 MC |
3015 | if (iocb->private) |
3016 | EXT4_I(inode)->cur_aio_dio = NULL; | |
3017 | /* | |
3018 | * The io_end structure takes a reference to the inode, | |
3019 | * that structure needs to be destroyed and the | |
3020 | * reference to the inode need to be dropped, when IO is | |
3021 | * complete, even with 0 byte write, or failed. | |
3022 | * | |
3023 | * In the successful AIO DIO case, the io_end structure will be | |
3024 | * desctroyed and the reference to the inode will be dropped | |
3025 | * after the end_io call back function is called. | |
3026 | * | |
3027 | * In the case there is 0 byte write, or error case, since | |
3028 | * VFS direct IO won't invoke the end_io call back function, | |
3029 | * we need to free the end_io structure here. | |
3030 | */ | |
3031 | if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) { | |
3032 | ext4_free_io_end(iocb->private); | |
3033 | iocb->private = NULL; | |
19f5fb7a TT |
3034 | } else if (ret > 0 && ext4_test_inode_state(inode, |
3035 | EXT4_STATE_DIO_UNWRITTEN)) { | |
109f5565 | 3036 | int err; |
8d5d02e6 MC |
3037 | /* |
3038 | * for non AIO case, since the IO is already | |
25985edc | 3039 | * completed, we could do the conversion right here |
8d5d02e6 | 3040 | */ |
109f5565 M |
3041 | err = ext4_convert_unwritten_extents(inode, |
3042 | offset, ret); | |
3043 | if (err < 0) | |
3044 | ret = err; | |
19f5fb7a | 3045 | ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN); |
109f5565 | 3046 | } |
4c0425ff MC |
3047 | return ret; |
3048 | } | |
8d5d02e6 MC |
3049 | |
3050 | /* for write the the end of file case, we fall back to old way */ | |
4c0425ff MC |
3051 | return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs); |
3052 | } | |
3053 | ||
3054 | static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb, | |
3055 | const struct iovec *iov, loff_t offset, | |
3056 | unsigned long nr_segs) | |
3057 | { | |
3058 | struct file *file = iocb->ki_filp; | |
3059 | struct inode *inode = file->f_mapping->host; | |
0562e0ba | 3060 | ssize_t ret; |
4c0425ff | 3061 | |
84ebd795 TT |
3062 | /* |
3063 | * If we are doing data journalling we don't support O_DIRECT | |
3064 | */ | |
3065 | if (ext4_should_journal_data(inode)) | |
3066 | return 0; | |
3067 | ||
0562e0ba | 3068 | trace_ext4_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw); |
12e9b892 | 3069 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
0562e0ba JZ |
3070 | ret = ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs); |
3071 | else | |
3072 | ret = ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs); | |
3073 | trace_ext4_direct_IO_exit(inode, offset, | |
3074 | iov_length(iov, nr_segs), rw, ret); | |
3075 | return ret; | |
4c0425ff MC |
3076 | } |
3077 | ||
ac27a0ec | 3078 | /* |
617ba13b | 3079 | * Pages can be marked dirty completely asynchronously from ext4's journalling |
ac27a0ec DK |
3080 | * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do |
3081 | * much here because ->set_page_dirty is called under VFS locks. The page is | |
3082 | * not necessarily locked. | |
3083 | * | |
3084 | * We cannot just dirty the page and leave attached buffers clean, because the | |
3085 | * buffers' dirty state is "definitive". We cannot just set the buffers dirty | |
3086 | * or jbddirty because all the journalling code will explode. | |
3087 | * | |
3088 | * So what we do is to mark the page "pending dirty" and next time writepage | |
3089 | * is called, propagate that into the buffers appropriately. | |
3090 | */ | |
617ba13b | 3091 | static int ext4_journalled_set_page_dirty(struct page *page) |
ac27a0ec DK |
3092 | { |
3093 | SetPageChecked(page); | |
3094 | return __set_page_dirty_nobuffers(page); | |
3095 | } | |
3096 | ||
617ba13b | 3097 | static const struct address_space_operations ext4_ordered_aops = { |
8ab22b9a HH |
3098 | .readpage = ext4_readpage, |
3099 | .readpages = ext4_readpages, | |
43ce1d23 | 3100 | .writepage = ext4_writepage, |
8ab22b9a HH |
3101 | .write_begin = ext4_write_begin, |
3102 | .write_end = ext4_ordered_write_end, | |
3103 | .bmap = ext4_bmap, | |
3104 | .invalidatepage = ext4_invalidatepage, | |
3105 | .releasepage = ext4_releasepage, | |
3106 | .direct_IO = ext4_direct_IO, | |
3107 | .migratepage = buffer_migrate_page, | |
3108 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 3109 | .error_remove_page = generic_error_remove_page, |
ac27a0ec DK |
3110 | }; |
3111 | ||
617ba13b | 3112 | static const struct address_space_operations ext4_writeback_aops = { |
8ab22b9a HH |
3113 | .readpage = ext4_readpage, |
3114 | .readpages = ext4_readpages, | |
43ce1d23 | 3115 | .writepage = ext4_writepage, |
8ab22b9a HH |
3116 | .write_begin = ext4_write_begin, |
3117 | .write_end = ext4_writeback_write_end, | |
3118 | .bmap = ext4_bmap, | |
3119 | .invalidatepage = ext4_invalidatepage, | |
3120 | .releasepage = ext4_releasepage, | |
3121 | .direct_IO = ext4_direct_IO, | |
3122 | .migratepage = buffer_migrate_page, | |
3123 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 3124 | .error_remove_page = generic_error_remove_page, |
ac27a0ec DK |
3125 | }; |
3126 | ||
617ba13b | 3127 | static const struct address_space_operations ext4_journalled_aops = { |
8ab22b9a HH |
3128 | .readpage = ext4_readpage, |
3129 | .readpages = ext4_readpages, | |
43ce1d23 | 3130 | .writepage = ext4_writepage, |
8ab22b9a HH |
3131 | .write_begin = ext4_write_begin, |
3132 | .write_end = ext4_journalled_write_end, | |
3133 | .set_page_dirty = ext4_journalled_set_page_dirty, | |
3134 | .bmap = ext4_bmap, | |
3135 | .invalidatepage = ext4_invalidatepage, | |
3136 | .releasepage = ext4_releasepage, | |
84ebd795 | 3137 | .direct_IO = ext4_direct_IO, |
8ab22b9a | 3138 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 3139 | .error_remove_page = generic_error_remove_page, |
ac27a0ec DK |
3140 | }; |
3141 | ||
64769240 | 3142 | static const struct address_space_operations ext4_da_aops = { |
8ab22b9a HH |
3143 | .readpage = ext4_readpage, |
3144 | .readpages = ext4_readpages, | |
43ce1d23 | 3145 | .writepage = ext4_writepage, |
8ab22b9a | 3146 | .writepages = ext4_da_writepages, |
8ab22b9a HH |
3147 | .write_begin = ext4_da_write_begin, |
3148 | .write_end = ext4_da_write_end, | |
3149 | .bmap = ext4_bmap, | |
3150 | .invalidatepage = ext4_da_invalidatepage, | |
3151 | .releasepage = ext4_releasepage, | |
3152 | .direct_IO = ext4_direct_IO, | |
3153 | .migratepage = buffer_migrate_page, | |
3154 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 3155 | .error_remove_page = generic_error_remove_page, |
64769240 AT |
3156 | }; |
3157 | ||
617ba13b | 3158 | void ext4_set_aops(struct inode *inode) |
ac27a0ec | 3159 | { |
3d2b1582 LC |
3160 | switch (ext4_inode_journal_mode(inode)) { |
3161 | case EXT4_INODE_ORDERED_DATA_MODE: | |
3162 | if (test_opt(inode->i_sb, DELALLOC)) | |
3163 | inode->i_mapping->a_ops = &ext4_da_aops; | |
3164 | else | |
3165 | inode->i_mapping->a_ops = &ext4_ordered_aops; | |
3166 | break; | |
3167 | case EXT4_INODE_WRITEBACK_DATA_MODE: | |
3168 | if (test_opt(inode->i_sb, DELALLOC)) | |
3169 | inode->i_mapping->a_ops = &ext4_da_aops; | |
3170 | else | |
3171 | inode->i_mapping->a_ops = &ext4_writeback_aops; | |
3172 | break; | |
3173 | case EXT4_INODE_JOURNAL_DATA_MODE: | |
617ba13b | 3174 | inode->i_mapping->a_ops = &ext4_journalled_aops; |
3d2b1582 LC |
3175 | break; |
3176 | default: | |
3177 | BUG(); | |
3178 | } | |
ac27a0ec DK |
3179 | } |
3180 | ||
4e96b2db AH |
3181 | |
3182 | /* | |
3183 | * ext4_discard_partial_page_buffers() | |
3184 | * Wrapper function for ext4_discard_partial_page_buffers_no_lock. | |
3185 | * This function finds and locks the page containing the offset | |
3186 | * "from" and passes it to ext4_discard_partial_page_buffers_no_lock. | |
3187 | * Calling functions that already have the page locked should call | |
3188 | * ext4_discard_partial_page_buffers_no_lock directly. | |
3189 | */ | |
3190 | int ext4_discard_partial_page_buffers(handle_t *handle, | |
3191 | struct address_space *mapping, loff_t from, | |
3192 | loff_t length, int flags) | |
3193 | { | |
3194 | struct inode *inode = mapping->host; | |
3195 | struct page *page; | |
3196 | int err = 0; | |
3197 | ||
3198 | page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT, | |
3199 | mapping_gfp_mask(mapping) & ~__GFP_FS); | |
3200 | if (!page) | |
5129d05f | 3201 | return -ENOMEM; |
4e96b2db AH |
3202 | |
3203 | err = ext4_discard_partial_page_buffers_no_lock(handle, inode, page, | |
3204 | from, length, flags); | |
3205 | ||
3206 | unlock_page(page); | |
3207 | page_cache_release(page); | |
3208 | return err; | |
3209 | } | |
3210 | ||
3211 | /* | |
3212 | * ext4_discard_partial_page_buffers_no_lock() | |
3213 | * Zeros a page range of length 'length' starting from offset 'from'. | |
3214 | * Buffer heads that correspond to the block aligned regions of the | |
3215 | * zeroed range will be unmapped. Unblock aligned regions | |
3216 | * will have the corresponding buffer head mapped if needed so that | |
3217 | * that region of the page can be updated with the partial zero out. | |
3218 | * | |
3219 | * This function assumes that the page has already been locked. The | |
3220 | * The range to be discarded must be contained with in the given page. | |
3221 | * If the specified range exceeds the end of the page it will be shortened | |
3222 | * to the end of the page that corresponds to 'from'. This function is | |
3223 | * appropriate for updating a page and it buffer heads to be unmapped and | |
3224 | * zeroed for blocks that have been either released, or are going to be | |
3225 | * released. | |
3226 | * | |
3227 | * handle: The journal handle | |
3228 | * inode: The files inode | |
3229 | * page: A locked page that contains the offset "from" | |
3230 | * from: The starting byte offset (from the begining of the file) | |
3231 | * to begin discarding | |
3232 | * len: The length of bytes to discard | |
3233 | * flags: Optional flags that may be used: | |
3234 | * | |
3235 | * EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED | |
3236 | * Only zero the regions of the page whose buffer heads | |
3237 | * have already been unmapped. This flag is appropriate | |
3238 | * for updateing the contents of a page whose blocks may | |
3239 | * have already been released, and we only want to zero | |
3240 | * out the regions that correspond to those released blocks. | |
3241 | * | |
3242 | * Returns zero on sucess or negative on failure. | |
3243 | */ | |
5f163cc7 | 3244 | static int ext4_discard_partial_page_buffers_no_lock(handle_t *handle, |
4e96b2db AH |
3245 | struct inode *inode, struct page *page, loff_t from, |
3246 | loff_t length, int flags) | |
3247 | { | |
3248 | ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT; | |
3249 | unsigned int offset = from & (PAGE_CACHE_SIZE-1); | |
3250 | unsigned int blocksize, max, pos; | |
4e96b2db AH |
3251 | ext4_lblk_t iblock; |
3252 | struct buffer_head *bh; | |
3253 | int err = 0; | |
3254 | ||
3255 | blocksize = inode->i_sb->s_blocksize; | |
3256 | max = PAGE_CACHE_SIZE - offset; | |
3257 | ||
3258 | if (index != page->index) | |
3259 | return -EINVAL; | |
3260 | ||
3261 | /* | |
3262 | * correct length if it does not fall between | |
3263 | * 'from' and the end of the page | |
3264 | */ | |
3265 | if (length > max || length < 0) | |
3266 | length = max; | |
3267 | ||
3268 | iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits); | |
3269 | ||
093e6e36 YY |
3270 | if (!page_has_buffers(page)) |
3271 | create_empty_buffers(page, blocksize, 0); | |
4e96b2db AH |
3272 | |
3273 | /* Find the buffer that contains "offset" */ | |
3274 | bh = page_buffers(page); | |
3275 | pos = blocksize; | |
3276 | while (offset >= pos) { | |
3277 | bh = bh->b_this_page; | |
3278 | iblock++; | |
3279 | pos += blocksize; | |
3280 | } | |
3281 | ||
3282 | pos = offset; | |
3283 | while (pos < offset + length) { | |
e260daf2 YY |
3284 | unsigned int end_of_block, range_to_discard; |
3285 | ||
4e96b2db AH |
3286 | err = 0; |
3287 | ||
3288 | /* The length of space left to zero and unmap */ | |
3289 | range_to_discard = offset + length - pos; | |
3290 | ||
3291 | /* The length of space until the end of the block */ | |
3292 | end_of_block = blocksize - (pos & (blocksize-1)); | |
3293 | ||
3294 | /* | |
3295 | * Do not unmap or zero past end of block | |
3296 | * for this buffer head | |
3297 | */ | |
3298 | if (range_to_discard > end_of_block) | |
3299 | range_to_discard = end_of_block; | |
3300 | ||
3301 | ||
3302 | /* | |
3303 | * Skip this buffer head if we are only zeroing unampped | |
3304 | * regions of the page | |
3305 | */ | |
3306 | if (flags & EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED && | |
3307 | buffer_mapped(bh)) | |
3308 | goto next; | |
3309 | ||
3310 | /* If the range is block aligned, unmap */ | |
3311 | if (range_to_discard == blocksize) { | |
3312 | clear_buffer_dirty(bh); | |
3313 | bh->b_bdev = NULL; | |
3314 | clear_buffer_mapped(bh); | |
3315 | clear_buffer_req(bh); | |
3316 | clear_buffer_new(bh); | |
3317 | clear_buffer_delay(bh); | |
3318 | clear_buffer_unwritten(bh); | |
3319 | clear_buffer_uptodate(bh); | |
3320 | zero_user(page, pos, range_to_discard); | |
3321 | BUFFER_TRACE(bh, "Buffer discarded"); | |
3322 | goto next; | |
3323 | } | |
3324 | ||
3325 | /* | |
3326 | * If this block is not completely contained in the range | |
3327 | * to be discarded, then it is not going to be released. Because | |
3328 | * we need to keep this block, we need to make sure this part | |
3329 | * of the page is uptodate before we modify it by writeing | |
3330 | * partial zeros on it. | |
3331 | */ | |
3332 | if (!buffer_mapped(bh)) { | |
3333 | /* | |
3334 | * Buffer head must be mapped before we can read | |
3335 | * from the block | |
3336 | */ | |
3337 | BUFFER_TRACE(bh, "unmapped"); | |
3338 | ext4_get_block(inode, iblock, bh, 0); | |
3339 | /* unmapped? It's a hole - nothing to do */ | |
3340 | if (!buffer_mapped(bh)) { | |
3341 | BUFFER_TRACE(bh, "still unmapped"); | |
3342 | goto next; | |
3343 | } | |
3344 | } | |
3345 | ||
3346 | /* Ok, it's mapped. Make sure it's up-to-date */ | |
3347 | if (PageUptodate(page)) | |
3348 | set_buffer_uptodate(bh); | |
3349 | ||
3350 | if (!buffer_uptodate(bh)) { | |
3351 | err = -EIO; | |
3352 | ll_rw_block(READ, 1, &bh); | |
3353 | wait_on_buffer(bh); | |
3354 | /* Uhhuh. Read error. Complain and punt.*/ | |
3355 | if (!buffer_uptodate(bh)) | |
3356 | goto next; | |
3357 | } | |
3358 | ||
3359 | if (ext4_should_journal_data(inode)) { | |
3360 | BUFFER_TRACE(bh, "get write access"); | |
3361 | err = ext4_journal_get_write_access(handle, bh); | |
3362 | if (err) | |
3363 | goto next; | |
3364 | } | |
3365 | ||
3366 | zero_user(page, pos, range_to_discard); | |
3367 | ||
3368 | err = 0; | |
3369 | if (ext4_should_journal_data(inode)) { | |
3370 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
decbd919 | 3371 | } else |
4e96b2db | 3372 | mark_buffer_dirty(bh); |
4e96b2db AH |
3373 | |
3374 | BUFFER_TRACE(bh, "Partial buffer zeroed"); | |
3375 | next: | |
3376 | bh = bh->b_this_page; | |
3377 | iblock++; | |
3378 | pos += range_to_discard; | |
3379 | } | |
3380 | ||
3381 | return err; | |
3382 | } | |
3383 | ||
91ef4caf DG |
3384 | int ext4_can_truncate(struct inode *inode) |
3385 | { | |
91ef4caf DG |
3386 | if (S_ISREG(inode->i_mode)) |
3387 | return 1; | |
3388 | if (S_ISDIR(inode->i_mode)) | |
3389 | return 1; | |
3390 | if (S_ISLNK(inode->i_mode)) | |
3391 | return !ext4_inode_is_fast_symlink(inode); | |
3392 | return 0; | |
3393 | } | |
3394 | ||
a4bb6b64 AH |
3395 | /* |
3396 | * ext4_punch_hole: punches a hole in a file by releaseing the blocks | |
3397 | * associated with the given offset and length | |
3398 | * | |
3399 | * @inode: File inode | |
3400 | * @offset: The offset where the hole will begin | |
3401 | * @len: The length of the hole | |
3402 | * | |
3403 | * Returns: 0 on sucess or negative on failure | |
3404 | */ | |
3405 | ||
3406 | int ext4_punch_hole(struct file *file, loff_t offset, loff_t length) | |
3407 | { | |
3408 | struct inode *inode = file->f_path.dentry->d_inode; | |
3409 | if (!S_ISREG(inode->i_mode)) | |
73355192 | 3410 | return -EOPNOTSUPP; |
a4bb6b64 AH |
3411 | |
3412 | if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { | |
3413 | /* TODO: Add support for non extent hole punching */ | |
73355192 | 3414 | return -EOPNOTSUPP; |
a4bb6b64 AH |
3415 | } |
3416 | ||
bab08ab9 TT |
3417 | if (EXT4_SB(inode->i_sb)->s_cluster_ratio > 1) { |
3418 | /* TODO: Add support for bigalloc file systems */ | |
73355192 | 3419 | return -EOPNOTSUPP; |
bab08ab9 TT |
3420 | } |
3421 | ||
a4bb6b64 AH |
3422 | return ext4_ext_punch_hole(file, offset, length); |
3423 | } | |
3424 | ||
ac27a0ec | 3425 | /* |
617ba13b | 3426 | * ext4_truncate() |
ac27a0ec | 3427 | * |
617ba13b MC |
3428 | * We block out ext4_get_block() block instantiations across the entire |
3429 | * transaction, and VFS/VM ensures that ext4_truncate() cannot run | |
ac27a0ec DK |
3430 | * simultaneously on behalf of the same inode. |
3431 | * | |
42b2aa86 | 3432 | * As we work through the truncate and commit bits of it to the journal there |
ac27a0ec DK |
3433 | * is one core, guiding principle: the file's tree must always be consistent on |
3434 | * disk. We must be able to restart the truncate after a crash. | |
3435 | * | |
3436 | * The file's tree may be transiently inconsistent in memory (although it | |
3437 | * probably isn't), but whenever we close off and commit a journal transaction, | |
3438 | * the contents of (the filesystem + the journal) must be consistent and | |
3439 | * restartable. It's pretty simple, really: bottom up, right to left (although | |
3440 | * left-to-right works OK too). | |
3441 | * | |
3442 | * Note that at recovery time, journal replay occurs *before* the restart of | |
3443 | * truncate against the orphan inode list. | |
3444 | * | |
3445 | * The committed inode has the new, desired i_size (which is the same as | |
617ba13b | 3446 | * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see |
ac27a0ec | 3447 | * that this inode's truncate did not complete and it will again call |
617ba13b MC |
3448 | * ext4_truncate() to have another go. So there will be instantiated blocks |
3449 | * to the right of the truncation point in a crashed ext4 filesystem. But | |
ac27a0ec | 3450 | * that's fine - as long as they are linked from the inode, the post-crash |
617ba13b | 3451 | * ext4_truncate() run will find them and release them. |
ac27a0ec | 3452 | */ |
617ba13b | 3453 | void ext4_truncate(struct inode *inode) |
ac27a0ec | 3454 | { |
0562e0ba JZ |
3455 | trace_ext4_truncate_enter(inode); |
3456 | ||
91ef4caf | 3457 | if (!ext4_can_truncate(inode)) |
ac27a0ec DK |
3458 | return; |
3459 | ||
12e9b892 | 3460 | ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS); |
c8d46e41 | 3461 | |
5534fb5b | 3462 | if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC)) |
19f5fb7a | 3463 | ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE); |
7d8f9f7d | 3464 | |
ff9893dc | 3465 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
cf108bca | 3466 | ext4_ext_truncate(inode); |
ff9893dc AG |
3467 | else |
3468 | ext4_ind_truncate(inode); | |
ac27a0ec | 3469 | |
0562e0ba | 3470 | trace_ext4_truncate_exit(inode); |
ac27a0ec DK |
3471 | } |
3472 | ||
ac27a0ec | 3473 | /* |
617ba13b | 3474 | * ext4_get_inode_loc returns with an extra refcount against the inode's |
ac27a0ec DK |
3475 | * underlying buffer_head on success. If 'in_mem' is true, we have all |
3476 | * data in memory that is needed to recreate the on-disk version of this | |
3477 | * inode. | |
3478 | */ | |
617ba13b MC |
3479 | static int __ext4_get_inode_loc(struct inode *inode, |
3480 | struct ext4_iloc *iloc, int in_mem) | |
ac27a0ec | 3481 | { |
240799cd TT |
3482 | struct ext4_group_desc *gdp; |
3483 | struct buffer_head *bh; | |
3484 | struct super_block *sb = inode->i_sb; | |
3485 | ext4_fsblk_t block; | |
3486 | int inodes_per_block, inode_offset; | |
3487 | ||
3a06d778 | 3488 | iloc->bh = NULL; |
240799cd TT |
3489 | if (!ext4_valid_inum(sb, inode->i_ino)) |
3490 | return -EIO; | |
ac27a0ec | 3491 | |
240799cd TT |
3492 | iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb); |
3493 | gdp = ext4_get_group_desc(sb, iloc->block_group, NULL); | |
3494 | if (!gdp) | |
ac27a0ec DK |
3495 | return -EIO; |
3496 | ||
240799cd TT |
3497 | /* |
3498 | * Figure out the offset within the block group inode table | |
3499 | */ | |
00d09882 | 3500 | inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; |
240799cd TT |
3501 | inode_offset = ((inode->i_ino - 1) % |
3502 | EXT4_INODES_PER_GROUP(sb)); | |
3503 | block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block); | |
3504 | iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb); | |
3505 | ||
3506 | bh = sb_getblk(sb, block); | |
ac27a0ec | 3507 | if (!bh) { |
c398eda0 TT |
3508 | EXT4_ERROR_INODE_BLOCK(inode, block, |
3509 | "unable to read itable block"); | |
ac27a0ec DK |
3510 | return -EIO; |
3511 | } | |
3512 | if (!buffer_uptodate(bh)) { | |
3513 | lock_buffer(bh); | |
9c83a923 HK |
3514 | |
3515 | /* | |
3516 | * If the buffer has the write error flag, we have failed | |
3517 | * to write out another inode in the same block. In this | |
3518 | * case, we don't have to read the block because we may | |
3519 | * read the old inode data successfully. | |
3520 | */ | |
3521 | if (buffer_write_io_error(bh) && !buffer_uptodate(bh)) | |
3522 | set_buffer_uptodate(bh); | |
3523 | ||
ac27a0ec DK |
3524 | if (buffer_uptodate(bh)) { |
3525 | /* someone brought it uptodate while we waited */ | |
3526 | unlock_buffer(bh); | |
3527 | goto has_buffer; | |
3528 | } | |
3529 | ||
3530 | /* | |
3531 | * If we have all information of the inode in memory and this | |
3532 | * is the only valid inode in the block, we need not read the | |
3533 | * block. | |
3534 | */ | |
3535 | if (in_mem) { | |
3536 | struct buffer_head *bitmap_bh; | |
240799cd | 3537 | int i, start; |
ac27a0ec | 3538 | |
240799cd | 3539 | start = inode_offset & ~(inodes_per_block - 1); |
ac27a0ec | 3540 | |
240799cd TT |
3541 | /* Is the inode bitmap in cache? */ |
3542 | bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp)); | |
ac27a0ec DK |
3543 | if (!bitmap_bh) |
3544 | goto make_io; | |
3545 | ||
3546 | /* | |
3547 | * If the inode bitmap isn't in cache then the | |
3548 | * optimisation may end up performing two reads instead | |
3549 | * of one, so skip it. | |
3550 | */ | |
3551 | if (!buffer_uptodate(bitmap_bh)) { | |
3552 | brelse(bitmap_bh); | |
3553 | goto make_io; | |
3554 | } | |
240799cd | 3555 | for (i = start; i < start + inodes_per_block; i++) { |
ac27a0ec DK |
3556 | if (i == inode_offset) |
3557 | continue; | |
617ba13b | 3558 | if (ext4_test_bit(i, bitmap_bh->b_data)) |
ac27a0ec DK |
3559 | break; |
3560 | } | |
3561 | brelse(bitmap_bh); | |
240799cd | 3562 | if (i == start + inodes_per_block) { |
ac27a0ec DK |
3563 | /* all other inodes are free, so skip I/O */ |
3564 | memset(bh->b_data, 0, bh->b_size); | |
3565 | set_buffer_uptodate(bh); | |
3566 | unlock_buffer(bh); | |
3567 | goto has_buffer; | |
3568 | } | |
3569 | } | |
3570 | ||
3571 | make_io: | |
240799cd TT |
3572 | /* |
3573 | * If we need to do any I/O, try to pre-readahead extra | |
3574 | * blocks from the inode table. | |
3575 | */ | |
3576 | if (EXT4_SB(sb)->s_inode_readahead_blks) { | |
3577 | ext4_fsblk_t b, end, table; | |
3578 | unsigned num; | |
3579 | ||
3580 | table = ext4_inode_table(sb, gdp); | |
b713a5ec | 3581 | /* s_inode_readahead_blks is always a power of 2 */ |
240799cd TT |
3582 | b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1); |
3583 | if (table > b) | |
3584 | b = table; | |
3585 | end = b + EXT4_SB(sb)->s_inode_readahead_blks; | |
3586 | num = EXT4_INODES_PER_GROUP(sb); | |
feb0ab32 | 3587 | if (ext4_has_group_desc_csum(sb)) |
560671a0 | 3588 | num -= ext4_itable_unused_count(sb, gdp); |
240799cd TT |
3589 | table += num / inodes_per_block; |
3590 | if (end > table) | |
3591 | end = table; | |
3592 | while (b <= end) | |
3593 | sb_breadahead(sb, b++); | |
3594 | } | |
3595 | ||
ac27a0ec DK |
3596 | /* |
3597 | * There are other valid inodes in the buffer, this inode | |
3598 | * has in-inode xattrs, or we don't have this inode in memory. | |
3599 | * Read the block from disk. | |
3600 | */ | |
0562e0ba | 3601 | trace_ext4_load_inode(inode); |
ac27a0ec DK |
3602 | get_bh(bh); |
3603 | bh->b_end_io = end_buffer_read_sync; | |
65299a3b | 3604 | submit_bh(READ | REQ_META | REQ_PRIO, bh); |
ac27a0ec DK |
3605 | wait_on_buffer(bh); |
3606 | if (!buffer_uptodate(bh)) { | |
c398eda0 TT |
3607 | EXT4_ERROR_INODE_BLOCK(inode, block, |
3608 | "unable to read itable block"); | |
ac27a0ec DK |
3609 | brelse(bh); |
3610 | return -EIO; | |
3611 | } | |
3612 | } | |
3613 | has_buffer: | |
3614 | iloc->bh = bh; | |
3615 | return 0; | |
3616 | } | |
3617 | ||
617ba13b | 3618 | int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec DK |
3619 | { |
3620 | /* We have all inode data except xattrs in memory here. */ | |
617ba13b | 3621 | return __ext4_get_inode_loc(inode, iloc, |
19f5fb7a | 3622 | !ext4_test_inode_state(inode, EXT4_STATE_XATTR)); |
ac27a0ec DK |
3623 | } |
3624 | ||
617ba13b | 3625 | void ext4_set_inode_flags(struct inode *inode) |
ac27a0ec | 3626 | { |
617ba13b | 3627 | unsigned int flags = EXT4_I(inode)->i_flags; |
ac27a0ec DK |
3628 | |
3629 | inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC); | |
617ba13b | 3630 | if (flags & EXT4_SYNC_FL) |
ac27a0ec | 3631 | inode->i_flags |= S_SYNC; |
617ba13b | 3632 | if (flags & EXT4_APPEND_FL) |
ac27a0ec | 3633 | inode->i_flags |= S_APPEND; |
617ba13b | 3634 | if (flags & EXT4_IMMUTABLE_FL) |
ac27a0ec | 3635 | inode->i_flags |= S_IMMUTABLE; |
617ba13b | 3636 | if (flags & EXT4_NOATIME_FL) |
ac27a0ec | 3637 | inode->i_flags |= S_NOATIME; |
617ba13b | 3638 | if (flags & EXT4_DIRSYNC_FL) |
ac27a0ec DK |
3639 | inode->i_flags |= S_DIRSYNC; |
3640 | } | |
3641 | ||
ff9ddf7e JK |
3642 | /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */ |
3643 | void ext4_get_inode_flags(struct ext4_inode_info *ei) | |
3644 | { | |
84a8dce2 DM |
3645 | unsigned int vfs_fl; |
3646 | unsigned long old_fl, new_fl; | |
3647 | ||
3648 | do { | |
3649 | vfs_fl = ei->vfs_inode.i_flags; | |
3650 | old_fl = ei->i_flags; | |
3651 | new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL| | |
3652 | EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL| | |
3653 | EXT4_DIRSYNC_FL); | |
3654 | if (vfs_fl & S_SYNC) | |
3655 | new_fl |= EXT4_SYNC_FL; | |
3656 | if (vfs_fl & S_APPEND) | |
3657 | new_fl |= EXT4_APPEND_FL; | |
3658 | if (vfs_fl & S_IMMUTABLE) | |
3659 | new_fl |= EXT4_IMMUTABLE_FL; | |
3660 | if (vfs_fl & S_NOATIME) | |
3661 | new_fl |= EXT4_NOATIME_FL; | |
3662 | if (vfs_fl & S_DIRSYNC) | |
3663 | new_fl |= EXT4_DIRSYNC_FL; | |
3664 | } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl); | |
ff9ddf7e | 3665 | } |
de9a55b8 | 3666 | |
0fc1b451 | 3667 | static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode, |
de9a55b8 | 3668 | struct ext4_inode_info *ei) |
0fc1b451 AK |
3669 | { |
3670 | blkcnt_t i_blocks ; | |
8180a562 AK |
3671 | struct inode *inode = &(ei->vfs_inode); |
3672 | struct super_block *sb = inode->i_sb; | |
0fc1b451 AK |
3673 | |
3674 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, | |
3675 | EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) { | |
3676 | /* we are using combined 48 bit field */ | |
3677 | i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 | | |
3678 | le32_to_cpu(raw_inode->i_blocks_lo); | |
07a03824 | 3679 | if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) { |
8180a562 AK |
3680 | /* i_blocks represent file system block size */ |
3681 | return i_blocks << (inode->i_blkbits - 9); | |
3682 | } else { | |
3683 | return i_blocks; | |
3684 | } | |
0fc1b451 AK |
3685 | } else { |
3686 | return le32_to_cpu(raw_inode->i_blocks_lo); | |
3687 | } | |
3688 | } | |
ff9ddf7e | 3689 | |
1d1fe1ee | 3690 | struct inode *ext4_iget(struct super_block *sb, unsigned long ino) |
ac27a0ec | 3691 | { |
617ba13b MC |
3692 | struct ext4_iloc iloc; |
3693 | struct ext4_inode *raw_inode; | |
1d1fe1ee | 3694 | struct ext4_inode_info *ei; |
1d1fe1ee | 3695 | struct inode *inode; |
b436b9be | 3696 | journal_t *journal = EXT4_SB(sb)->s_journal; |
1d1fe1ee | 3697 | long ret; |
ac27a0ec DK |
3698 | int block; |
3699 | ||
1d1fe1ee DH |
3700 | inode = iget_locked(sb, ino); |
3701 | if (!inode) | |
3702 | return ERR_PTR(-ENOMEM); | |
3703 | if (!(inode->i_state & I_NEW)) | |
3704 | return inode; | |
3705 | ||
3706 | ei = EXT4_I(inode); | |
7dc57615 | 3707 | iloc.bh = NULL; |
ac27a0ec | 3708 | |
1d1fe1ee DH |
3709 | ret = __ext4_get_inode_loc(inode, &iloc, 0); |
3710 | if (ret < 0) | |
ac27a0ec | 3711 | goto bad_inode; |
617ba13b | 3712 | raw_inode = ext4_raw_inode(&iloc); |
814525f4 DW |
3713 | |
3714 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { | |
3715 | ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize); | |
3716 | if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > | |
3717 | EXT4_INODE_SIZE(inode->i_sb)) { | |
3718 | EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)", | |
3719 | EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize, | |
3720 | EXT4_INODE_SIZE(inode->i_sb)); | |
3721 | ret = -EIO; | |
3722 | goto bad_inode; | |
3723 | } | |
3724 | } else | |
3725 | ei->i_extra_isize = 0; | |
3726 | ||
3727 | /* Precompute checksum seed for inode metadata */ | |
3728 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, | |
3729 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) { | |
3730 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
3731 | __u32 csum; | |
3732 | __le32 inum = cpu_to_le32(inode->i_ino); | |
3733 | __le32 gen = raw_inode->i_generation; | |
3734 | csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum, | |
3735 | sizeof(inum)); | |
3736 | ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen, | |
3737 | sizeof(gen)); | |
3738 | } | |
3739 | ||
3740 | if (!ext4_inode_csum_verify(inode, raw_inode, ei)) { | |
3741 | EXT4_ERROR_INODE(inode, "checksum invalid"); | |
3742 | ret = -EIO; | |
3743 | goto bad_inode; | |
3744 | } | |
3745 | ||
ac27a0ec DK |
3746 | inode->i_mode = le16_to_cpu(raw_inode->i_mode); |
3747 | inode->i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low); | |
3748 | inode->i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low); | |
af5bc92d | 3749 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
ac27a0ec DK |
3750 | inode->i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16; |
3751 | inode->i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16; | |
3752 | } | |
bfe86848 | 3753 | set_nlink(inode, le16_to_cpu(raw_inode->i_links_count)); |
ac27a0ec | 3754 | |
353eb83c | 3755 | ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */ |
ac27a0ec DK |
3756 | ei->i_dir_start_lookup = 0; |
3757 | ei->i_dtime = le32_to_cpu(raw_inode->i_dtime); | |
3758 | /* We now have enough fields to check if the inode was active or not. | |
3759 | * This is needed because nfsd might try to access dead inodes | |
3760 | * the test is that same one that e2fsck uses | |
3761 | * NeilBrown 1999oct15 | |
3762 | */ | |
3763 | if (inode->i_nlink == 0) { | |
3764 | if (inode->i_mode == 0 || | |
617ba13b | 3765 | !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) { |
ac27a0ec | 3766 | /* this inode is deleted */ |
1d1fe1ee | 3767 | ret = -ESTALE; |
ac27a0ec DK |
3768 | goto bad_inode; |
3769 | } | |
3770 | /* The only unlinked inodes we let through here have | |
3771 | * valid i_mode and are being read by the orphan | |
3772 | * recovery code: that's fine, we're about to complete | |
3773 | * the process of deleting those. */ | |
3774 | } | |
ac27a0ec | 3775 | ei->i_flags = le32_to_cpu(raw_inode->i_flags); |
0fc1b451 | 3776 | inode->i_blocks = ext4_inode_blocks(raw_inode, ei); |
7973c0c1 | 3777 | ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo); |
a9e81742 | 3778 | if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) |
a1ddeb7e BP |
3779 | ei->i_file_acl |= |
3780 | ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32; | |
a48380f7 | 3781 | inode->i_size = ext4_isize(raw_inode); |
ac27a0ec | 3782 | ei->i_disksize = inode->i_size; |
a9e7f447 DM |
3783 | #ifdef CONFIG_QUOTA |
3784 | ei->i_reserved_quota = 0; | |
3785 | #endif | |
ac27a0ec DK |
3786 | inode->i_generation = le32_to_cpu(raw_inode->i_generation); |
3787 | ei->i_block_group = iloc.block_group; | |
a4912123 | 3788 | ei->i_last_alloc_group = ~0; |
ac27a0ec DK |
3789 | /* |
3790 | * NOTE! The in-memory inode i_data array is in little-endian order | |
3791 | * even on big-endian machines: we do NOT byteswap the block numbers! | |
3792 | */ | |
617ba13b | 3793 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
ac27a0ec DK |
3794 | ei->i_data[block] = raw_inode->i_block[block]; |
3795 | INIT_LIST_HEAD(&ei->i_orphan); | |
3796 | ||
b436b9be JK |
3797 | /* |
3798 | * Set transaction id's of transactions that have to be committed | |
3799 | * to finish f[data]sync. We set them to currently running transaction | |
3800 | * as we cannot be sure that the inode or some of its metadata isn't | |
3801 | * part of the transaction - the inode could have been reclaimed and | |
3802 | * now it is reread from disk. | |
3803 | */ | |
3804 | if (journal) { | |
3805 | transaction_t *transaction; | |
3806 | tid_t tid; | |
3807 | ||
a931da6a | 3808 | read_lock(&journal->j_state_lock); |
b436b9be JK |
3809 | if (journal->j_running_transaction) |
3810 | transaction = journal->j_running_transaction; | |
3811 | else | |
3812 | transaction = journal->j_committing_transaction; | |
3813 | if (transaction) | |
3814 | tid = transaction->t_tid; | |
3815 | else | |
3816 | tid = journal->j_commit_sequence; | |
a931da6a | 3817 | read_unlock(&journal->j_state_lock); |
b436b9be JK |
3818 | ei->i_sync_tid = tid; |
3819 | ei->i_datasync_tid = tid; | |
3820 | } | |
3821 | ||
0040d987 | 3822 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
ac27a0ec DK |
3823 | if (ei->i_extra_isize == 0) { |
3824 | /* The extra space is currently unused. Use it. */ | |
617ba13b MC |
3825 | ei->i_extra_isize = sizeof(struct ext4_inode) - |
3826 | EXT4_GOOD_OLD_INODE_SIZE; | |
ac27a0ec DK |
3827 | } else { |
3828 | __le32 *magic = (void *)raw_inode + | |
617ba13b | 3829 | EXT4_GOOD_OLD_INODE_SIZE + |
ac27a0ec | 3830 | ei->i_extra_isize; |
617ba13b | 3831 | if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) |
19f5fb7a | 3832 | ext4_set_inode_state(inode, EXT4_STATE_XATTR); |
ac27a0ec | 3833 | } |
814525f4 | 3834 | } |
ac27a0ec | 3835 | |
ef7f3835 KS |
3836 | EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode); |
3837 | EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode); | |
3838 | EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode); | |
3839 | EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode); | |
3840 | ||
25ec56b5 JNC |
3841 | inode->i_version = le32_to_cpu(raw_inode->i_disk_version); |
3842 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { | |
3843 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
3844 | inode->i_version |= | |
3845 | (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32; | |
3846 | } | |
3847 | ||
c4b5a614 | 3848 | ret = 0; |
485c26ec | 3849 | if (ei->i_file_acl && |
1032988c | 3850 | !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) { |
24676da4 TT |
3851 | EXT4_ERROR_INODE(inode, "bad extended attribute block %llu", |
3852 | ei->i_file_acl); | |
485c26ec TT |
3853 | ret = -EIO; |
3854 | goto bad_inode; | |
07a03824 | 3855 | } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
c4b5a614 TT |
3856 | if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || |
3857 | (S_ISLNK(inode->i_mode) && | |
3858 | !ext4_inode_is_fast_symlink(inode))) | |
3859 | /* Validate extent which is part of inode */ | |
3860 | ret = ext4_ext_check_inode(inode); | |
de9a55b8 | 3861 | } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || |
fe2c8191 TN |
3862 | (S_ISLNK(inode->i_mode) && |
3863 | !ext4_inode_is_fast_symlink(inode))) { | |
de9a55b8 | 3864 | /* Validate block references which are part of inode */ |
1f7d1e77 | 3865 | ret = ext4_ind_check_inode(inode); |
fe2c8191 | 3866 | } |
567f3e9a | 3867 | if (ret) |
de9a55b8 | 3868 | goto bad_inode; |
7a262f7c | 3869 | |
ac27a0ec | 3870 | if (S_ISREG(inode->i_mode)) { |
617ba13b MC |
3871 | inode->i_op = &ext4_file_inode_operations; |
3872 | inode->i_fop = &ext4_file_operations; | |
3873 | ext4_set_aops(inode); | |
ac27a0ec | 3874 | } else if (S_ISDIR(inode->i_mode)) { |
617ba13b MC |
3875 | inode->i_op = &ext4_dir_inode_operations; |
3876 | inode->i_fop = &ext4_dir_operations; | |
ac27a0ec | 3877 | } else if (S_ISLNK(inode->i_mode)) { |
e83c1397 | 3878 | if (ext4_inode_is_fast_symlink(inode)) { |
617ba13b | 3879 | inode->i_op = &ext4_fast_symlink_inode_operations; |
e83c1397 DG |
3880 | nd_terminate_link(ei->i_data, inode->i_size, |
3881 | sizeof(ei->i_data) - 1); | |
3882 | } else { | |
617ba13b MC |
3883 | inode->i_op = &ext4_symlink_inode_operations; |
3884 | ext4_set_aops(inode); | |
ac27a0ec | 3885 | } |
563bdd61 TT |
3886 | } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) || |
3887 | S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) { | |
617ba13b | 3888 | inode->i_op = &ext4_special_inode_operations; |
ac27a0ec DK |
3889 | if (raw_inode->i_block[0]) |
3890 | init_special_inode(inode, inode->i_mode, | |
3891 | old_decode_dev(le32_to_cpu(raw_inode->i_block[0]))); | |
3892 | else | |
3893 | init_special_inode(inode, inode->i_mode, | |
3894 | new_decode_dev(le32_to_cpu(raw_inode->i_block[1]))); | |
563bdd61 | 3895 | } else { |
563bdd61 | 3896 | ret = -EIO; |
24676da4 | 3897 | EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode); |
563bdd61 | 3898 | goto bad_inode; |
ac27a0ec | 3899 | } |
af5bc92d | 3900 | brelse(iloc.bh); |
617ba13b | 3901 | ext4_set_inode_flags(inode); |
1d1fe1ee DH |
3902 | unlock_new_inode(inode); |
3903 | return inode; | |
ac27a0ec DK |
3904 | |
3905 | bad_inode: | |
567f3e9a | 3906 | brelse(iloc.bh); |
1d1fe1ee DH |
3907 | iget_failed(inode); |
3908 | return ERR_PTR(ret); | |
ac27a0ec DK |
3909 | } |
3910 | ||
0fc1b451 AK |
3911 | static int ext4_inode_blocks_set(handle_t *handle, |
3912 | struct ext4_inode *raw_inode, | |
3913 | struct ext4_inode_info *ei) | |
3914 | { | |
3915 | struct inode *inode = &(ei->vfs_inode); | |
3916 | u64 i_blocks = inode->i_blocks; | |
3917 | struct super_block *sb = inode->i_sb; | |
0fc1b451 AK |
3918 | |
3919 | if (i_blocks <= ~0U) { | |
3920 | /* | |
3921 | * i_blocks can be represnted in a 32 bit variable | |
3922 | * as multiple of 512 bytes | |
3923 | */ | |
8180a562 | 3924 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 3925 | raw_inode->i_blocks_high = 0; |
84a8dce2 | 3926 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
f287a1a5 TT |
3927 | return 0; |
3928 | } | |
3929 | if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) | |
3930 | return -EFBIG; | |
3931 | ||
3932 | if (i_blocks <= 0xffffffffffffULL) { | |
0fc1b451 AK |
3933 | /* |
3934 | * i_blocks can be represented in a 48 bit variable | |
3935 | * as multiple of 512 bytes | |
3936 | */ | |
8180a562 | 3937 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 3938 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); |
84a8dce2 | 3939 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
0fc1b451 | 3940 | } else { |
84a8dce2 | 3941 | ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
8180a562 AK |
3942 | /* i_block is stored in file system block size */ |
3943 | i_blocks = i_blocks >> (inode->i_blkbits - 9); | |
3944 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); | |
3945 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); | |
0fc1b451 | 3946 | } |
f287a1a5 | 3947 | return 0; |
0fc1b451 AK |
3948 | } |
3949 | ||
ac27a0ec DK |
3950 | /* |
3951 | * Post the struct inode info into an on-disk inode location in the | |
3952 | * buffer-cache. This gobbles the caller's reference to the | |
3953 | * buffer_head in the inode location struct. | |
3954 | * | |
3955 | * The caller must have write access to iloc->bh. | |
3956 | */ | |
617ba13b | 3957 | static int ext4_do_update_inode(handle_t *handle, |
ac27a0ec | 3958 | struct inode *inode, |
830156c7 | 3959 | struct ext4_iloc *iloc) |
ac27a0ec | 3960 | { |
617ba13b MC |
3961 | struct ext4_inode *raw_inode = ext4_raw_inode(iloc); |
3962 | struct ext4_inode_info *ei = EXT4_I(inode); | |
ac27a0ec DK |
3963 | struct buffer_head *bh = iloc->bh; |
3964 | int err = 0, rc, block; | |
3965 | ||
3966 | /* For fields not not tracking in the in-memory inode, | |
3967 | * initialise them to zero for new inodes. */ | |
19f5fb7a | 3968 | if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) |
617ba13b | 3969 | memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size); |
ac27a0ec | 3970 | |
ff9ddf7e | 3971 | ext4_get_inode_flags(ei); |
ac27a0ec | 3972 | raw_inode->i_mode = cpu_to_le16(inode->i_mode); |
af5bc92d | 3973 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
ac27a0ec DK |
3974 | raw_inode->i_uid_low = cpu_to_le16(low_16_bits(inode->i_uid)); |
3975 | raw_inode->i_gid_low = cpu_to_le16(low_16_bits(inode->i_gid)); | |
3976 | /* | |
3977 | * Fix up interoperability with old kernels. Otherwise, old inodes get | |
3978 | * re-used with the upper 16 bits of the uid/gid intact | |
3979 | */ | |
af5bc92d | 3980 | if (!ei->i_dtime) { |
ac27a0ec DK |
3981 | raw_inode->i_uid_high = |
3982 | cpu_to_le16(high_16_bits(inode->i_uid)); | |
3983 | raw_inode->i_gid_high = | |
3984 | cpu_to_le16(high_16_bits(inode->i_gid)); | |
3985 | } else { | |
3986 | raw_inode->i_uid_high = 0; | |
3987 | raw_inode->i_gid_high = 0; | |
3988 | } | |
3989 | } else { | |
3990 | raw_inode->i_uid_low = | |
3991 | cpu_to_le16(fs_high2lowuid(inode->i_uid)); | |
3992 | raw_inode->i_gid_low = | |
3993 | cpu_to_le16(fs_high2lowgid(inode->i_gid)); | |
3994 | raw_inode->i_uid_high = 0; | |
3995 | raw_inode->i_gid_high = 0; | |
3996 | } | |
3997 | raw_inode->i_links_count = cpu_to_le16(inode->i_nlink); | |
ef7f3835 KS |
3998 | |
3999 | EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode); | |
4000 | EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode); | |
4001 | EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode); | |
4002 | EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode); | |
4003 | ||
0fc1b451 AK |
4004 | if (ext4_inode_blocks_set(handle, raw_inode, ei)) |
4005 | goto out_brelse; | |
ac27a0ec | 4006 | raw_inode->i_dtime = cpu_to_le32(ei->i_dtime); |
353eb83c | 4007 | raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF); |
9b8f1f01 MC |
4008 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != |
4009 | cpu_to_le32(EXT4_OS_HURD)) | |
a1ddeb7e BP |
4010 | raw_inode->i_file_acl_high = |
4011 | cpu_to_le16(ei->i_file_acl >> 32); | |
7973c0c1 | 4012 | raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl); |
a48380f7 AK |
4013 | ext4_isize_set(raw_inode, ei->i_disksize); |
4014 | if (ei->i_disksize > 0x7fffffffULL) { | |
4015 | struct super_block *sb = inode->i_sb; | |
4016 | if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, | |
4017 | EXT4_FEATURE_RO_COMPAT_LARGE_FILE) || | |
4018 | EXT4_SB(sb)->s_es->s_rev_level == | |
4019 | cpu_to_le32(EXT4_GOOD_OLD_REV)) { | |
4020 | /* If this is the first large file | |
4021 | * created, add a flag to the superblock. | |
4022 | */ | |
4023 | err = ext4_journal_get_write_access(handle, | |
4024 | EXT4_SB(sb)->s_sbh); | |
4025 | if (err) | |
4026 | goto out_brelse; | |
4027 | ext4_update_dynamic_rev(sb); | |
4028 | EXT4_SET_RO_COMPAT_FEATURE(sb, | |
617ba13b | 4029 | EXT4_FEATURE_RO_COMPAT_LARGE_FILE); |
0390131b | 4030 | ext4_handle_sync(handle); |
a9c47317 | 4031 | err = ext4_handle_dirty_super_now(handle, sb); |
ac27a0ec DK |
4032 | } |
4033 | } | |
4034 | raw_inode->i_generation = cpu_to_le32(inode->i_generation); | |
4035 | if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) { | |
4036 | if (old_valid_dev(inode->i_rdev)) { | |
4037 | raw_inode->i_block[0] = | |
4038 | cpu_to_le32(old_encode_dev(inode->i_rdev)); | |
4039 | raw_inode->i_block[1] = 0; | |
4040 | } else { | |
4041 | raw_inode->i_block[0] = 0; | |
4042 | raw_inode->i_block[1] = | |
4043 | cpu_to_le32(new_encode_dev(inode->i_rdev)); | |
4044 | raw_inode->i_block[2] = 0; | |
4045 | } | |
de9a55b8 TT |
4046 | } else |
4047 | for (block = 0; block < EXT4_N_BLOCKS; block++) | |
4048 | raw_inode->i_block[block] = ei->i_data[block]; | |
ac27a0ec | 4049 | |
25ec56b5 JNC |
4050 | raw_inode->i_disk_version = cpu_to_le32(inode->i_version); |
4051 | if (ei->i_extra_isize) { | |
4052 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
4053 | raw_inode->i_version_hi = | |
4054 | cpu_to_le32(inode->i_version >> 32); | |
ac27a0ec | 4055 | raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize); |
25ec56b5 JNC |
4056 | } |
4057 | ||
814525f4 DW |
4058 | ext4_inode_csum_set(inode, raw_inode, ei); |
4059 | ||
830156c7 | 4060 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
73b50c1c | 4061 | rc = ext4_handle_dirty_metadata(handle, NULL, bh); |
830156c7 FM |
4062 | if (!err) |
4063 | err = rc; | |
19f5fb7a | 4064 | ext4_clear_inode_state(inode, EXT4_STATE_NEW); |
ac27a0ec | 4065 | |
b436b9be | 4066 | ext4_update_inode_fsync_trans(handle, inode, 0); |
ac27a0ec | 4067 | out_brelse: |
af5bc92d | 4068 | brelse(bh); |
617ba13b | 4069 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
4070 | return err; |
4071 | } | |
4072 | ||
4073 | /* | |
617ba13b | 4074 | * ext4_write_inode() |
ac27a0ec DK |
4075 | * |
4076 | * We are called from a few places: | |
4077 | * | |
4078 | * - Within generic_file_write() for O_SYNC files. | |
4079 | * Here, there will be no transaction running. We wait for any running | |
4080 | * trasnaction to commit. | |
4081 | * | |
4082 | * - Within sys_sync(), kupdate and such. | |
4083 | * We wait on commit, if tol to. | |
4084 | * | |
4085 | * - Within prune_icache() (PF_MEMALLOC == true) | |
4086 | * Here we simply return. We can't afford to block kswapd on the | |
4087 | * journal commit. | |
4088 | * | |
4089 | * In all cases it is actually safe for us to return without doing anything, | |
4090 | * because the inode has been copied into a raw inode buffer in | |
617ba13b | 4091 | * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for |
ac27a0ec DK |
4092 | * knfsd. |
4093 | * | |
4094 | * Note that we are absolutely dependent upon all inode dirtiers doing the | |
4095 | * right thing: they *must* call mark_inode_dirty() after dirtying info in | |
4096 | * which we are interested. | |
4097 | * | |
4098 | * It would be a bug for them to not do this. The code: | |
4099 | * | |
4100 | * mark_inode_dirty(inode) | |
4101 | * stuff(); | |
4102 | * inode->i_size = expr; | |
4103 | * | |
4104 | * is in error because a kswapd-driven write_inode() could occur while | |
4105 | * `stuff()' is running, and the new i_size will be lost. Plus the inode | |
4106 | * will no longer be on the superblock's dirty inode list. | |
4107 | */ | |
a9185b41 | 4108 | int ext4_write_inode(struct inode *inode, struct writeback_control *wbc) |
ac27a0ec | 4109 | { |
91ac6f43 FM |
4110 | int err; |
4111 | ||
ac27a0ec DK |
4112 | if (current->flags & PF_MEMALLOC) |
4113 | return 0; | |
4114 | ||
91ac6f43 FM |
4115 | if (EXT4_SB(inode->i_sb)->s_journal) { |
4116 | if (ext4_journal_current_handle()) { | |
4117 | jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n"); | |
4118 | dump_stack(); | |
4119 | return -EIO; | |
4120 | } | |
ac27a0ec | 4121 | |
a9185b41 | 4122 | if (wbc->sync_mode != WB_SYNC_ALL) |
91ac6f43 FM |
4123 | return 0; |
4124 | ||
4125 | err = ext4_force_commit(inode->i_sb); | |
4126 | } else { | |
4127 | struct ext4_iloc iloc; | |
ac27a0ec | 4128 | |
8b472d73 | 4129 | err = __ext4_get_inode_loc(inode, &iloc, 0); |
91ac6f43 FM |
4130 | if (err) |
4131 | return err; | |
a9185b41 | 4132 | if (wbc->sync_mode == WB_SYNC_ALL) |
830156c7 FM |
4133 | sync_dirty_buffer(iloc.bh); |
4134 | if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) { | |
c398eda0 TT |
4135 | EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr, |
4136 | "IO error syncing inode"); | |
830156c7 FM |
4137 | err = -EIO; |
4138 | } | |
fd2dd9fb | 4139 | brelse(iloc.bh); |
91ac6f43 FM |
4140 | } |
4141 | return err; | |
ac27a0ec DK |
4142 | } |
4143 | ||
4144 | /* | |
617ba13b | 4145 | * ext4_setattr() |
ac27a0ec DK |
4146 | * |
4147 | * Called from notify_change. | |
4148 | * | |
4149 | * We want to trap VFS attempts to truncate the file as soon as | |
4150 | * possible. In particular, we want to make sure that when the VFS | |
4151 | * shrinks i_size, we put the inode on the orphan list and modify | |
4152 | * i_disksize immediately, so that during the subsequent flushing of | |
4153 | * dirty pages and freeing of disk blocks, we can guarantee that any | |
4154 | * commit will leave the blocks being flushed in an unused state on | |
4155 | * disk. (On recovery, the inode will get truncated and the blocks will | |
4156 | * be freed, so we have a strong guarantee that no future commit will | |
4157 | * leave these blocks visible to the user.) | |
4158 | * | |
678aaf48 JK |
4159 | * Another thing we have to assure is that if we are in ordered mode |
4160 | * and inode is still attached to the committing transaction, we must | |
4161 | * we start writeout of all the dirty pages which are being truncated. | |
4162 | * This way we are sure that all the data written in the previous | |
4163 | * transaction are already on disk (truncate waits for pages under | |
4164 | * writeback). | |
4165 | * | |
4166 | * Called with inode->i_mutex down. | |
ac27a0ec | 4167 | */ |
617ba13b | 4168 | int ext4_setattr(struct dentry *dentry, struct iattr *attr) |
ac27a0ec DK |
4169 | { |
4170 | struct inode *inode = dentry->d_inode; | |
4171 | int error, rc = 0; | |
3d287de3 | 4172 | int orphan = 0; |
ac27a0ec DK |
4173 | const unsigned int ia_valid = attr->ia_valid; |
4174 | ||
4175 | error = inode_change_ok(inode, attr); | |
4176 | if (error) | |
4177 | return error; | |
4178 | ||
12755627 | 4179 | if (is_quota_modification(inode, attr)) |
871a2931 | 4180 | dquot_initialize(inode); |
ac27a0ec DK |
4181 | if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) || |
4182 | (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) { | |
4183 | handle_t *handle; | |
4184 | ||
4185 | /* (user+group)*(old+new) structure, inode write (sb, | |
4186 | * inode block, ? - but truncate inode update has it) */ | |
5aca07eb | 4187 | handle = ext4_journal_start(inode, (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb)+ |
194074ac | 4188 | EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb))+3); |
ac27a0ec DK |
4189 | if (IS_ERR(handle)) { |
4190 | error = PTR_ERR(handle); | |
4191 | goto err_out; | |
4192 | } | |
b43fa828 | 4193 | error = dquot_transfer(inode, attr); |
ac27a0ec | 4194 | if (error) { |
617ba13b | 4195 | ext4_journal_stop(handle); |
ac27a0ec DK |
4196 | return error; |
4197 | } | |
4198 | /* Update corresponding info in inode so that everything is in | |
4199 | * one transaction */ | |
4200 | if (attr->ia_valid & ATTR_UID) | |
4201 | inode->i_uid = attr->ia_uid; | |
4202 | if (attr->ia_valid & ATTR_GID) | |
4203 | inode->i_gid = attr->ia_gid; | |
617ba13b MC |
4204 | error = ext4_mark_inode_dirty(handle, inode); |
4205 | ext4_journal_stop(handle); | |
ac27a0ec DK |
4206 | } |
4207 | ||
e2b46574 | 4208 | if (attr->ia_valid & ATTR_SIZE) { |
562c72aa CH |
4209 | inode_dio_wait(inode); |
4210 | ||
12e9b892 | 4211 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) { |
e2b46574 ES |
4212 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
4213 | ||
0c095c7f TT |
4214 | if (attr->ia_size > sbi->s_bitmap_maxbytes) |
4215 | return -EFBIG; | |
e2b46574 ES |
4216 | } |
4217 | } | |
4218 | ||
ac27a0ec | 4219 | if (S_ISREG(inode->i_mode) && |
c8d46e41 | 4220 | attr->ia_valid & ATTR_SIZE && |
072bd7ea | 4221 | (attr->ia_size < inode->i_size)) { |
ac27a0ec DK |
4222 | handle_t *handle; |
4223 | ||
617ba13b | 4224 | handle = ext4_journal_start(inode, 3); |
ac27a0ec DK |
4225 | if (IS_ERR(handle)) { |
4226 | error = PTR_ERR(handle); | |
4227 | goto err_out; | |
4228 | } | |
3d287de3 DM |
4229 | if (ext4_handle_valid(handle)) { |
4230 | error = ext4_orphan_add(handle, inode); | |
4231 | orphan = 1; | |
4232 | } | |
617ba13b MC |
4233 | EXT4_I(inode)->i_disksize = attr->ia_size; |
4234 | rc = ext4_mark_inode_dirty(handle, inode); | |
ac27a0ec DK |
4235 | if (!error) |
4236 | error = rc; | |
617ba13b | 4237 | ext4_journal_stop(handle); |
678aaf48 JK |
4238 | |
4239 | if (ext4_should_order_data(inode)) { | |
4240 | error = ext4_begin_ordered_truncate(inode, | |
4241 | attr->ia_size); | |
4242 | if (error) { | |
4243 | /* Do as much error cleanup as possible */ | |
4244 | handle = ext4_journal_start(inode, 3); | |
4245 | if (IS_ERR(handle)) { | |
4246 | ext4_orphan_del(NULL, inode); | |
4247 | goto err_out; | |
4248 | } | |
4249 | ext4_orphan_del(handle, inode); | |
3d287de3 | 4250 | orphan = 0; |
678aaf48 JK |
4251 | ext4_journal_stop(handle); |
4252 | goto err_out; | |
4253 | } | |
4254 | } | |
ac27a0ec DK |
4255 | } |
4256 | ||
072bd7ea | 4257 | if (attr->ia_valid & ATTR_SIZE) { |
afcff5d8 | 4258 | if (attr->ia_size != i_size_read(inode)) |
072bd7ea | 4259 | truncate_setsize(inode, attr->ia_size); |
afcff5d8 | 4260 | ext4_truncate(inode); |
072bd7ea | 4261 | } |
ac27a0ec | 4262 | |
1025774c CH |
4263 | if (!rc) { |
4264 | setattr_copy(inode, attr); | |
4265 | mark_inode_dirty(inode); | |
4266 | } | |
4267 | ||
4268 | /* | |
4269 | * If the call to ext4_truncate failed to get a transaction handle at | |
4270 | * all, we need to clean up the in-core orphan list manually. | |
4271 | */ | |
3d287de3 | 4272 | if (orphan && inode->i_nlink) |
617ba13b | 4273 | ext4_orphan_del(NULL, inode); |
ac27a0ec DK |
4274 | |
4275 | if (!rc && (ia_valid & ATTR_MODE)) | |
617ba13b | 4276 | rc = ext4_acl_chmod(inode); |
ac27a0ec DK |
4277 | |
4278 | err_out: | |
617ba13b | 4279 | ext4_std_error(inode->i_sb, error); |
ac27a0ec DK |
4280 | if (!error) |
4281 | error = rc; | |
4282 | return error; | |
4283 | } | |
4284 | ||
3e3398a0 MC |
4285 | int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry, |
4286 | struct kstat *stat) | |
4287 | { | |
4288 | struct inode *inode; | |
4289 | unsigned long delalloc_blocks; | |
4290 | ||
4291 | inode = dentry->d_inode; | |
4292 | generic_fillattr(inode, stat); | |
4293 | ||
4294 | /* | |
4295 | * We can't update i_blocks if the block allocation is delayed | |
4296 | * otherwise in the case of system crash before the real block | |
4297 | * allocation is done, we will have i_blocks inconsistent with | |
4298 | * on-disk file blocks. | |
4299 | * We always keep i_blocks updated together with real | |
4300 | * allocation. But to not confuse with user, stat | |
4301 | * will return the blocks that include the delayed allocation | |
4302 | * blocks for this file. | |
4303 | */ | |
3e3398a0 | 4304 | delalloc_blocks = EXT4_I(inode)->i_reserved_data_blocks; |
3e3398a0 MC |
4305 | |
4306 | stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9; | |
4307 | return 0; | |
4308 | } | |
ac27a0ec | 4309 | |
a02908f1 MC |
4310 | static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk) |
4311 | { | |
12e9b892 | 4312 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) |
8bb2b247 | 4313 | return ext4_ind_trans_blocks(inode, nrblocks, chunk); |
ac51d837 | 4314 | return ext4_ext_index_trans_blocks(inode, nrblocks, chunk); |
a02908f1 | 4315 | } |
ac51d837 | 4316 | |
ac27a0ec | 4317 | /* |
a02908f1 MC |
4318 | * Account for index blocks, block groups bitmaps and block group |
4319 | * descriptor blocks if modify datablocks and index blocks | |
4320 | * worse case, the indexs blocks spread over different block groups | |
ac27a0ec | 4321 | * |
a02908f1 | 4322 | * If datablocks are discontiguous, they are possible to spread over |
af901ca1 | 4323 | * different block groups too. If they are contiuguous, with flexbg, |
a02908f1 | 4324 | * they could still across block group boundary. |
ac27a0ec | 4325 | * |
a02908f1 MC |
4326 | * Also account for superblock, inode, quota and xattr blocks |
4327 | */ | |
1f109d5a | 4328 | static int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk) |
a02908f1 | 4329 | { |
8df9675f TT |
4330 | ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb); |
4331 | int gdpblocks; | |
a02908f1 MC |
4332 | int idxblocks; |
4333 | int ret = 0; | |
4334 | ||
4335 | /* | |
4336 | * How many index blocks need to touch to modify nrblocks? | |
4337 | * The "Chunk" flag indicating whether the nrblocks is | |
4338 | * physically contiguous on disk | |
4339 | * | |
4340 | * For Direct IO and fallocate, they calls get_block to allocate | |
4341 | * one single extent at a time, so they could set the "Chunk" flag | |
4342 | */ | |
4343 | idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk); | |
4344 | ||
4345 | ret = idxblocks; | |
4346 | ||
4347 | /* | |
4348 | * Now let's see how many group bitmaps and group descriptors need | |
4349 | * to account | |
4350 | */ | |
4351 | groups = idxblocks; | |
4352 | if (chunk) | |
4353 | groups += 1; | |
4354 | else | |
4355 | groups += nrblocks; | |
4356 | ||
4357 | gdpblocks = groups; | |
8df9675f TT |
4358 | if (groups > ngroups) |
4359 | groups = ngroups; | |
a02908f1 MC |
4360 | if (groups > EXT4_SB(inode->i_sb)->s_gdb_count) |
4361 | gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count; | |
4362 | ||
4363 | /* bitmaps and block group descriptor blocks */ | |
4364 | ret += groups + gdpblocks; | |
4365 | ||
4366 | /* Blocks for super block, inode, quota and xattr blocks */ | |
4367 | ret += EXT4_META_TRANS_BLOCKS(inode->i_sb); | |
4368 | ||
4369 | return ret; | |
4370 | } | |
4371 | ||
4372 | /* | |
25985edc | 4373 | * Calculate the total number of credits to reserve to fit |
f3bd1f3f MC |
4374 | * the modification of a single pages into a single transaction, |
4375 | * which may include multiple chunks of block allocations. | |
ac27a0ec | 4376 | * |
525f4ed8 | 4377 | * This could be called via ext4_write_begin() |
ac27a0ec | 4378 | * |
525f4ed8 | 4379 | * We need to consider the worse case, when |
a02908f1 | 4380 | * one new block per extent. |
ac27a0ec | 4381 | */ |
a86c6181 | 4382 | int ext4_writepage_trans_blocks(struct inode *inode) |
ac27a0ec | 4383 | { |
617ba13b | 4384 | int bpp = ext4_journal_blocks_per_page(inode); |
ac27a0ec DK |
4385 | int ret; |
4386 | ||
a02908f1 | 4387 | ret = ext4_meta_trans_blocks(inode, bpp, 0); |
a86c6181 | 4388 | |
a02908f1 | 4389 | /* Account for data blocks for journalled mode */ |
617ba13b | 4390 | if (ext4_should_journal_data(inode)) |
a02908f1 | 4391 | ret += bpp; |
ac27a0ec DK |
4392 | return ret; |
4393 | } | |
f3bd1f3f MC |
4394 | |
4395 | /* | |
4396 | * Calculate the journal credits for a chunk of data modification. | |
4397 | * | |
4398 | * This is called from DIO, fallocate or whoever calling | |
79e83036 | 4399 | * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks. |
f3bd1f3f MC |
4400 | * |
4401 | * journal buffers for data blocks are not included here, as DIO | |
4402 | * and fallocate do no need to journal data buffers. | |
4403 | */ | |
4404 | int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks) | |
4405 | { | |
4406 | return ext4_meta_trans_blocks(inode, nrblocks, 1); | |
4407 | } | |
4408 | ||
ac27a0ec | 4409 | /* |
617ba13b | 4410 | * The caller must have previously called ext4_reserve_inode_write(). |
ac27a0ec DK |
4411 | * Give this, we know that the caller already has write access to iloc->bh. |
4412 | */ | |
617ba13b | 4413 | int ext4_mark_iloc_dirty(handle_t *handle, |
de9a55b8 | 4414 | struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec DK |
4415 | { |
4416 | int err = 0; | |
4417 | ||
c64db50e | 4418 | if (IS_I_VERSION(inode)) |
25ec56b5 JNC |
4419 | inode_inc_iversion(inode); |
4420 | ||
ac27a0ec DK |
4421 | /* the do_update_inode consumes one bh->b_count */ |
4422 | get_bh(iloc->bh); | |
4423 | ||
dab291af | 4424 | /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */ |
830156c7 | 4425 | err = ext4_do_update_inode(handle, inode, iloc); |
ac27a0ec DK |
4426 | put_bh(iloc->bh); |
4427 | return err; | |
4428 | } | |
4429 | ||
4430 | /* | |
4431 | * On success, We end up with an outstanding reference count against | |
4432 | * iloc->bh. This _must_ be cleaned up later. | |
4433 | */ | |
4434 | ||
4435 | int | |
617ba13b MC |
4436 | ext4_reserve_inode_write(handle_t *handle, struct inode *inode, |
4437 | struct ext4_iloc *iloc) | |
ac27a0ec | 4438 | { |
0390131b FM |
4439 | int err; |
4440 | ||
4441 | err = ext4_get_inode_loc(inode, iloc); | |
4442 | if (!err) { | |
4443 | BUFFER_TRACE(iloc->bh, "get_write_access"); | |
4444 | err = ext4_journal_get_write_access(handle, iloc->bh); | |
4445 | if (err) { | |
4446 | brelse(iloc->bh); | |
4447 | iloc->bh = NULL; | |
ac27a0ec DK |
4448 | } |
4449 | } | |
617ba13b | 4450 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
4451 | return err; |
4452 | } | |
4453 | ||
6dd4ee7c KS |
4454 | /* |
4455 | * Expand an inode by new_extra_isize bytes. | |
4456 | * Returns 0 on success or negative error number on failure. | |
4457 | */ | |
1d03ec98 AK |
4458 | static int ext4_expand_extra_isize(struct inode *inode, |
4459 | unsigned int new_extra_isize, | |
4460 | struct ext4_iloc iloc, | |
4461 | handle_t *handle) | |
6dd4ee7c KS |
4462 | { |
4463 | struct ext4_inode *raw_inode; | |
4464 | struct ext4_xattr_ibody_header *header; | |
6dd4ee7c KS |
4465 | |
4466 | if (EXT4_I(inode)->i_extra_isize >= new_extra_isize) | |
4467 | return 0; | |
4468 | ||
4469 | raw_inode = ext4_raw_inode(&iloc); | |
4470 | ||
4471 | header = IHDR(inode, raw_inode); | |
6dd4ee7c KS |
4472 | |
4473 | /* No extended attributes present */ | |
19f5fb7a TT |
4474 | if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) || |
4475 | header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) { | |
6dd4ee7c KS |
4476 | memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0, |
4477 | new_extra_isize); | |
4478 | EXT4_I(inode)->i_extra_isize = new_extra_isize; | |
4479 | return 0; | |
4480 | } | |
4481 | ||
4482 | /* try to expand with EAs present */ | |
4483 | return ext4_expand_extra_isize_ea(inode, new_extra_isize, | |
4484 | raw_inode, handle); | |
4485 | } | |
4486 | ||
ac27a0ec DK |
4487 | /* |
4488 | * What we do here is to mark the in-core inode as clean with respect to inode | |
4489 | * dirtiness (it may still be data-dirty). | |
4490 | * This means that the in-core inode may be reaped by prune_icache | |
4491 | * without having to perform any I/O. This is a very good thing, | |
4492 | * because *any* task may call prune_icache - even ones which | |
4493 | * have a transaction open against a different journal. | |
4494 | * | |
4495 | * Is this cheating? Not really. Sure, we haven't written the | |
4496 | * inode out, but prune_icache isn't a user-visible syncing function. | |
4497 | * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync) | |
4498 | * we start and wait on commits. | |
4499 | * | |
4500 | * Is this efficient/effective? Well, we're being nice to the system | |
4501 | * by cleaning up our inodes proactively so they can be reaped | |
4502 | * without I/O. But we are potentially leaving up to five seconds' | |
4503 | * worth of inodes floating about which prune_icache wants us to | |
4504 | * write out. One way to fix that would be to get prune_icache() | |
4505 | * to do a write_super() to free up some memory. It has the desired | |
4506 | * effect. | |
4507 | */ | |
617ba13b | 4508 | int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode) |
ac27a0ec | 4509 | { |
617ba13b | 4510 | struct ext4_iloc iloc; |
6dd4ee7c KS |
4511 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
4512 | static unsigned int mnt_count; | |
4513 | int err, ret; | |
ac27a0ec DK |
4514 | |
4515 | might_sleep(); | |
7ff9c073 | 4516 | trace_ext4_mark_inode_dirty(inode, _RET_IP_); |
617ba13b | 4517 | err = ext4_reserve_inode_write(handle, inode, &iloc); |
0390131b FM |
4518 | if (ext4_handle_valid(handle) && |
4519 | EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize && | |
19f5fb7a | 4520 | !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) { |
6dd4ee7c KS |
4521 | /* |
4522 | * We need extra buffer credits since we may write into EA block | |
4523 | * with this same handle. If journal_extend fails, then it will | |
4524 | * only result in a minor loss of functionality for that inode. | |
4525 | * If this is felt to be critical, then e2fsck should be run to | |
4526 | * force a large enough s_min_extra_isize. | |
4527 | */ | |
4528 | if ((jbd2_journal_extend(handle, | |
4529 | EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) { | |
4530 | ret = ext4_expand_extra_isize(inode, | |
4531 | sbi->s_want_extra_isize, | |
4532 | iloc, handle); | |
4533 | if (ret) { | |
19f5fb7a TT |
4534 | ext4_set_inode_state(inode, |
4535 | EXT4_STATE_NO_EXPAND); | |
c1bddad9 AK |
4536 | if (mnt_count != |
4537 | le16_to_cpu(sbi->s_es->s_mnt_count)) { | |
12062ddd | 4538 | ext4_warning(inode->i_sb, |
6dd4ee7c KS |
4539 | "Unable to expand inode %lu. Delete" |
4540 | " some EAs or run e2fsck.", | |
4541 | inode->i_ino); | |
c1bddad9 AK |
4542 | mnt_count = |
4543 | le16_to_cpu(sbi->s_es->s_mnt_count); | |
6dd4ee7c KS |
4544 | } |
4545 | } | |
4546 | } | |
4547 | } | |
ac27a0ec | 4548 | if (!err) |
617ba13b | 4549 | err = ext4_mark_iloc_dirty(handle, inode, &iloc); |
ac27a0ec DK |
4550 | return err; |
4551 | } | |
4552 | ||
4553 | /* | |
617ba13b | 4554 | * ext4_dirty_inode() is called from __mark_inode_dirty() |
ac27a0ec DK |
4555 | * |
4556 | * We're really interested in the case where a file is being extended. | |
4557 | * i_size has been changed by generic_commit_write() and we thus need | |
4558 | * to include the updated inode in the current transaction. | |
4559 | * | |
5dd4056d | 4560 | * Also, dquot_alloc_block() will always dirty the inode when blocks |
ac27a0ec DK |
4561 | * are allocated to the file. |
4562 | * | |
4563 | * If the inode is marked synchronous, we don't honour that here - doing | |
4564 | * so would cause a commit on atime updates, which we don't bother doing. | |
4565 | * We handle synchronous inodes at the highest possible level. | |
4566 | */ | |
aa385729 | 4567 | void ext4_dirty_inode(struct inode *inode, int flags) |
ac27a0ec | 4568 | { |
ac27a0ec DK |
4569 | handle_t *handle; |
4570 | ||
617ba13b | 4571 | handle = ext4_journal_start(inode, 2); |
ac27a0ec DK |
4572 | if (IS_ERR(handle)) |
4573 | goto out; | |
f3dc272f | 4574 | |
f3dc272f CW |
4575 | ext4_mark_inode_dirty(handle, inode); |
4576 | ||
617ba13b | 4577 | ext4_journal_stop(handle); |
ac27a0ec DK |
4578 | out: |
4579 | return; | |
4580 | } | |
4581 | ||
4582 | #if 0 | |
4583 | /* | |
4584 | * Bind an inode's backing buffer_head into this transaction, to prevent | |
4585 | * it from being flushed to disk early. Unlike | |
617ba13b | 4586 | * ext4_reserve_inode_write, this leaves behind no bh reference and |
ac27a0ec DK |
4587 | * returns no iloc structure, so the caller needs to repeat the iloc |
4588 | * lookup to mark the inode dirty later. | |
4589 | */ | |
617ba13b | 4590 | static int ext4_pin_inode(handle_t *handle, struct inode *inode) |
ac27a0ec | 4591 | { |
617ba13b | 4592 | struct ext4_iloc iloc; |
ac27a0ec DK |
4593 | |
4594 | int err = 0; | |
4595 | if (handle) { | |
617ba13b | 4596 | err = ext4_get_inode_loc(inode, &iloc); |
ac27a0ec DK |
4597 | if (!err) { |
4598 | BUFFER_TRACE(iloc.bh, "get_write_access"); | |
dab291af | 4599 | err = jbd2_journal_get_write_access(handle, iloc.bh); |
ac27a0ec | 4600 | if (!err) |
0390131b | 4601 | err = ext4_handle_dirty_metadata(handle, |
73b50c1c | 4602 | NULL, |
0390131b | 4603 | iloc.bh); |
ac27a0ec DK |
4604 | brelse(iloc.bh); |
4605 | } | |
4606 | } | |
617ba13b | 4607 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
4608 | return err; |
4609 | } | |
4610 | #endif | |
4611 | ||
617ba13b | 4612 | int ext4_change_inode_journal_flag(struct inode *inode, int val) |
ac27a0ec DK |
4613 | { |
4614 | journal_t *journal; | |
4615 | handle_t *handle; | |
4616 | int err; | |
4617 | ||
4618 | /* | |
4619 | * We have to be very careful here: changing a data block's | |
4620 | * journaling status dynamically is dangerous. If we write a | |
4621 | * data block to the journal, change the status and then delete | |
4622 | * that block, we risk forgetting to revoke the old log record | |
4623 | * from the journal and so a subsequent replay can corrupt data. | |
4624 | * So, first we make sure that the journal is empty and that | |
4625 | * nobody is changing anything. | |
4626 | */ | |
4627 | ||
617ba13b | 4628 | journal = EXT4_JOURNAL(inode); |
0390131b FM |
4629 | if (!journal) |
4630 | return 0; | |
d699594d | 4631 | if (is_journal_aborted(journal)) |
ac27a0ec | 4632 | return -EROFS; |
2aff57b0 YY |
4633 | /* We have to allocate physical blocks for delalloc blocks |
4634 | * before flushing journal. otherwise delalloc blocks can not | |
4635 | * be allocated any more. even more truncate on delalloc blocks | |
4636 | * could trigger BUG by flushing delalloc blocks in journal. | |
4637 | * There is no delalloc block in non-journal data mode. | |
4638 | */ | |
4639 | if (val && test_opt(inode->i_sb, DELALLOC)) { | |
4640 | err = ext4_alloc_da_blocks(inode); | |
4641 | if (err < 0) | |
4642 | return err; | |
4643 | } | |
ac27a0ec | 4644 | |
dab291af | 4645 | jbd2_journal_lock_updates(journal); |
ac27a0ec DK |
4646 | |
4647 | /* | |
4648 | * OK, there are no updates running now, and all cached data is | |
4649 | * synced to disk. We are now in a completely consistent state | |
4650 | * which doesn't have anything in the journal, and we know that | |
4651 | * no filesystem updates are running, so it is safe to modify | |
4652 | * the inode's in-core data-journaling state flag now. | |
4653 | */ | |
4654 | ||
4655 | if (val) | |
12e9b892 | 4656 | ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
5872ddaa YY |
4657 | else { |
4658 | jbd2_journal_flush(journal); | |
12e9b892 | 4659 | ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
5872ddaa | 4660 | } |
617ba13b | 4661 | ext4_set_aops(inode); |
ac27a0ec | 4662 | |
dab291af | 4663 | jbd2_journal_unlock_updates(journal); |
ac27a0ec DK |
4664 | |
4665 | /* Finally we can mark the inode as dirty. */ | |
4666 | ||
617ba13b | 4667 | handle = ext4_journal_start(inode, 1); |
ac27a0ec DK |
4668 | if (IS_ERR(handle)) |
4669 | return PTR_ERR(handle); | |
4670 | ||
617ba13b | 4671 | err = ext4_mark_inode_dirty(handle, inode); |
0390131b | 4672 | ext4_handle_sync(handle); |
617ba13b MC |
4673 | ext4_journal_stop(handle); |
4674 | ext4_std_error(inode->i_sb, err); | |
ac27a0ec DK |
4675 | |
4676 | return err; | |
4677 | } | |
2e9ee850 AK |
4678 | |
4679 | static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh) | |
4680 | { | |
4681 | return !buffer_mapped(bh); | |
4682 | } | |
4683 | ||
c2ec175c | 4684 | int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
2e9ee850 | 4685 | { |
c2ec175c | 4686 | struct page *page = vmf->page; |
2e9ee850 AK |
4687 | loff_t size; |
4688 | unsigned long len; | |
9ea7df53 | 4689 | int ret; |
2e9ee850 AK |
4690 | struct file *file = vma->vm_file; |
4691 | struct inode *inode = file->f_path.dentry->d_inode; | |
4692 | struct address_space *mapping = inode->i_mapping; | |
9ea7df53 JK |
4693 | handle_t *handle; |
4694 | get_block_t *get_block; | |
4695 | int retries = 0; | |
2e9ee850 AK |
4696 | |
4697 | /* | |
9ea7df53 JK |
4698 | * This check is racy but catches the common case. We rely on |
4699 | * __block_page_mkwrite() to do a reliable check. | |
2e9ee850 | 4700 | */ |
9ea7df53 JK |
4701 | vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE); |
4702 | /* Delalloc case is easy... */ | |
4703 | if (test_opt(inode->i_sb, DELALLOC) && | |
4704 | !ext4_should_journal_data(inode) && | |
4705 | !ext4_nonda_switch(inode->i_sb)) { | |
4706 | do { | |
4707 | ret = __block_page_mkwrite(vma, vmf, | |
4708 | ext4_da_get_block_prep); | |
4709 | } while (ret == -ENOSPC && | |
4710 | ext4_should_retry_alloc(inode->i_sb, &retries)); | |
4711 | goto out_ret; | |
2e9ee850 | 4712 | } |
0e499890 DW |
4713 | |
4714 | lock_page(page); | |
9ea7df53 JK |
4715 | size = i_size_read(inode); |
4716 | /* Page got truncated from under us? */ | |
4717 | if (page->mapping != mapping || page_offset(page) > size) { | |
4718 | unlock_page(page); | |
4719 | ret = VM_FAULT_NOPAGE; | |
4720 | goto out; | |
0e499890 | 4721 | } |
2e9ee850 AK |
4722 | |
4723 | if (page->index == size >> PAGE_CACHE_SHIFT) | |
4724 | len = size & ~PAGE_CACHE_MASK; | |
4725 | else | |
4726 | len = PAGE_CACHE_SIZE; | |
a827eaff | 4727 | /* |
9ea7df53 JK |
4728 | * Return if we have all the buffers mapped. This avoids the need to do |
4729 | * journal_start/journal_stop which can block and take a long time | |
a827eaff | 4730 | */ |
2e9ee850 | 4731 | if (page_has_buffers(page)) { |
2e9ee850 | 4732 | if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL, |
a827eaff | 4733 | ext4_bh_unmapped)) { |
9ea7df53 JK |
4734 | /* Wait so that we don't change page under IO */ |
4735 | wait_on_page_writeback(page); | |
4736 | ret = VM_FAULT_LOCKED; | |
4737 | goto out; | |
a827eaff | 4738 | } |
2e9ee850 | 4739 | } |
a827eaff | 4740 | unlock_page(page); |
9ea7df53 JK |
4741 | /* OK, we need to fill the hole... */ |
4742 | if (ext4_should_dioread_nolock(inode)) | |
4743 | get_block = ext4_get_block_write; | |
4744 | else | |
4745 | get_block = ext4_get_block; | |
4746 | retry_alloc: | |
4747 | handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode)); | |
4748 | if (IS_ERR(handle)) { | |
c2ec175c | 4749 | ret = VM_FAULT_SIGBUS; |
9ea7df53 JK |
4750 | goto out; |
4751 | } | |
4752 | ret = __block_page_mkwrite(vma, vmf, get_block); | |
4753 | if (!ret && ext4_should_journal_data(inode)) { | |
4754 | if (walk_page_buffers(handle, page_buffers(page), 0, | |
4755 | PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) { | |
4756 | unlock_page(page); | |
4757 | ret = VM_FAULT_SIGBUS; | |
fcbb5515 | 4758 | ext4_journal_stop(handle); |
9ea7df53 JK |
4759 | goto out; |
4760 | } | |
4761 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); | |
4762 | } | |
4763 | ext4_journal_stop(handle); | |
4764 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) | |
4765 | goto retry_alloc; | |
4766 | out_ret: | |
4767 | ret = block_page_mkwrite_return(ret); | |
4768 | out: | |
2e9ee850 AK |
4769 | return ret; |
4770 | } |