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> |
a27bb332 | 40 | #include <linux/aio.h> |
9bffad1e | 41 | |
3dcf5451 | 42 | #include "ext4_jbd2.h" |
ac27a0ec DK |
43 | #include "xattr.h" |
44 | #include "acl.h" | |
9f125d64 | 45 | #include "truncate.h" |
ac27a0ec | 46 | |
9bffad1e TT |
47 | #include <trace/events/ext4.h> |
48 | ||
a1d6cc56 AK |
49 | #define MPAGE_DA_EXTENT_TAIL 0x01 |
50 | ||
814525f4 DW |
51 | static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw, |
52 | struct ext4_inode_info *ei) | |
53 | { | |
54 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
55 | __u16 csum_lo; | |
56 | __u16 csum_hi = 0; | |
57 | __u32 csum; | |
58 | ||
171a7f21 | 59 | csum_lo = le16_to_cpu(raw->i_checksum_lo); |
814525f4 DW |
60 | raw->i_checksum_lo = 0; |
61 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && | |
62 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) { | |
171a7f21 | 63 | csum_hi = le16_to_cpu(raw->i_checksum_hi); |
814525f4 DW |
64 | raw->i_checksum_hi = 0; |
65 | } | |
66 | ||
67 | csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw, | |
68 | EXT4_INODE_SIZE(inode->i_sb)); | |
69 | ||
171a7f21 | 70 | raw->i_checksum_lo = cpu_to_le16(csum_lo); |
814525f4 DW |
71 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && |
72 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) | |
171a7f21 | 73 | raw->i_checksum_hi = cpu_to_le16(csum_hi); |
814525f4 DW |
74 | |
75 | return csum; | |
76 | } | |
77 | ||
78 | static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw, | |
79 | struct ext4_inode_info *ei) | |
80 | { | |
81 | __u32 provided, calculated; | |
82 | ||
83 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != | |
84 | cpu_to_le32(EXT4_OS_LINUX) || | |
85 | !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, | |
86 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) | |
87 | return 1; | |
88 | ||
89 | provided = le16_to_cpu(raw->i_checksum_lo); | |
90 | calculated = ext4_inode_csum(inode, raw, ei); | |
91 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && | |
92 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) | |
93 | provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16; | |
94 | else | |
95 | calculated &= 0xFFFF; | |
96 | ||
97 | return provided == calculated; | |
98 | } | |
99 | ||
100 | static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw, | |
101 | struct ext4_inode_info *ei) | |
102 | { | |
103 | __u32 csum; | |
104 | ||
105 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != | |
106 | cpu_to_le32(EXT4_OS_LINUX) || | |
107 | !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, | |
108 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) | |
109 | return; | |
110 | ||
111 | csum = ext4_inode_csum(inode, raw, ei); | |
112 | raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF); | |
113 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && | |
114 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) | |
115 | raw->i_checksum_hi = cpu_to_le16(csum >> 16); | |
116 | } | |
117 | ||
678aaf48 JK |
118 | static inline int ext4_begin_ordered_truncate(struct inode *inode, |
119 | loff_t new_size) | |
120 | { | |
7ff9c073 | 121 | trace_ext4_begin_ordered_truncate(inode, new_size); |
8aefcd55 TT |
122 | /* |
123 | * If jinode is zero, then we never opened the file for | |
124 | * writing, so there's no need to call | |
125 | * jbd2_journal_begin_ordered_truncate() since there's no | |
126 | * outstanding writes we need to flush. | |
127 | */ | |
128 | if (!EXT4_I(inode)->jinode) | |
129 | return 0; | |
130 | return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode), | |
131 | EXT4_I(inode)->jinode, | |
132 | new_size); | |
678aaf48 JK |
133 | } |
134 | ||
d47992f8 LC |
135 | static void ext4_invalidatepage(struct page *page, unsigned int offset, |
136 | unsigned int length); | |
cb20d518 TT |
137 | static int __ext4_journalled_writepage(struct page *page, unsigned int len); |
138 | static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh); | |
64769240 | 139 | |
ac27a0ec DK |
140 | /* |
141 | * Test whether an inode is a fast symlink. | |
142 | */ | |
617ba13b | 143 | static int ext4_inode_is_fast_symlink(struct inode *inode) |
ac27a0ec | 144 | { |
617ba13b | 145 | int ea_blocks = EXT4_I(inode)->i_file_acl ? |
ac27a0ec DK |
146 | (inode->i_sb->s_blocksize >> 9) : 0; |
147 | ||
148 | return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0); | |
149 | } | |
150 | ||
ac27a0ec DK |
151 | /* |
152 | * Restart the transaction associated with *handle. This does a commit, | |
153 | * so before we call here everything must be consistently dirtied against | |
154 | * this transaction. | |
155 | */ | |
fa5d1113 | 156 | int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode, |
487caeef | 157 | int nblocks) |
ac27a0ec | 158 | { |
487caeef JK |
159 | int ret; |
160 | ||
161 | /* | |
e35fd660 | 162 | * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this |
487caeef JK |
163 | * moment, get_block can be called only for blocks inside i_size since |
164 | * page cache has been already dropped and writes are blocked by | |
165 | * i_mutex. So we can safely drop the i_data_sem here. | |
166 | */ | |
0390131b | 167 | BUG_ON(EXT4_JOURNAL(inode) == NULL); |
ac27a0ec | 168 | jbd_debug(2, "restarting handle %p\n", handle); |
487caeef | 169 | up_write(&EXT4_I(inode)->i_data_sem); |
8e8eaabe | 170 | ret = ext4_journal_restart(handle, nblocks); |
487caeef | 171 | down_write(&EXT4_I(inode)->i_data_sem); |
fa5d1113 | 172 | ext4_discard_preallocations(inode); |
487caeef JK |
173 | |
174 | return ret; | |
ac27a0ec DK |
175 | } |
176 | ||
177 | /* | |
178 | * Called at the last iput() if i_nlink is zero. | |
179 | */ | |
0930fcc1 | 180 | void ext4_evict_inode(struct inode *inode) |
ac27a0ec DK |
181 | { |
182 | handle_t *handle; | |
bc965ab3 | 183 | int err; |
ac27a0ec | 184 | |
7ff9c073 | 185 | trace_ext4_evict_inode(inode); |
2581fdc8 | 186 | |
0930fcc1 | 187 | if (inode->i_nlink) { |
2d859db3 JK |
188 | /* |
189 | * When journalling data dirty buffers are tracked only in the | |
190 | * journal. So although mm thinks everything is clean and | |
191 | * ready for reaping the inode might still have some pages to | |
192 | * write in the running transaction or waiting to be | |
193 | * checkpointed. Thus calling jbd2_journal_invalidatepage() | |
194 | * (via truncate_inode_pages()) to discard these buffers can | |
195 | * cause data loss. Also even if we did not discard these | |
196 | * buffers, we would have no way to find them after the inode | |
197 | * is reaped and thus user could see stale data if he tries to | |
198 | * read them before the transaction is checkpointed. So be | |
199 | * careful and force everything to disk here... We use | |
200 | * ei->i_datasync_tid to store the newest transaction | |
201 | * containing inode's data. | |
202 | * | |
203 | * Note that directories do not have this problem because they | |
204 | * don't use page cache. | |
205 | */ | |
206 | if (ext4_should_journal_data(inode) && | |
2b405bfa TT |
207 | (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) && |
208 | inode->i_ino != EXT4_JOURNAL_INO) { | |
2d859db3 JK |
209 | journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; |
210 | tid_t commit_tid = EXT4_I(inode)->i_datasync_tid; | |
211 | ||
d76a3a77 | 212 | jbd2_complete_transaction(journal, commit_tid); |
2d859db3 JK |
213 | filemap_write_and_wait(&inode->i_data); |
214 | } | |
0930fcc1 | 215 | truncate_inode_pages(&inode->i_data, 0); |
1ada47d9 | 216 | ext4_ioend_shutdown(inode); |
0930fcc1 AV |
217 | goto no_delete; |
218 | } | |
219 | ||
907f4554 | 220 | if (!is_bad_inode(inode)) |
871a2931 | 221 | dquot_initialize(inode); |
907f4554 | 222 | |
678aaf48 JK |
223 | if (ext4_should_order_data(inode)) |
224 | ext4_begin_ordered_truncate(inode, 0); | |
ac27a0ec | 225 | truncate_inode_pages(&inode->i_data, 0); |
1ada47d9 | 226 | ext4_ioend_shutdown(inode); |
ac27a0ec DK |
227 | |
228 | if (is_bad_inode(inode)) | |
229 | goto no_delete; | |
230 | ||
8e8ad8a5 JK |
231 | /* |
232 | * Protect us against freezing - iput() caller didn't have to have any | |
233 | * protection against it | |
234 | */ | |
235 | sb_start_intwrite(inode->i_sb); | |
9924a92a TT |
236 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, |
237 | ext4_blocks_for_truncate(inode)+3); | |
ac27a0ec | 238 | if (IS_ERR(handle)) { |
bc965ab3 | 239 | ext4_std_error(inode->i_sb, PTR_ERR(handle)); |
ac27a0ec DK |
240 | /* |
241 | * If we're going to skip the normal cleanup, we still need to | |
242 | * make sure that the in-core orphan linked list is properly | |
243 | * cleaned up. | |
244 | */ | |
617ba13b | 245 | ext4_orphan_del(NULL, inode); |
8e8ad8a5 | 246 | sb_end_intwrite(inode->i_sb); |
ac27a0ec DK |
247 | goto no_delete; |
248 | } | |
249 | ||
250 | if (IS_SYNC(inode)) | |
0390131b | 251 | ext4_handle_sync(handle); |
ac27a0ec | 252 | inode->i_size = 0; |
bc965ab3 TT |
253 | err = ext4_mark_inode_dirty(handle, inode); |
254 | if (err) { | |
12062ddd | 255 | ext4_warning(inode->i_sb, |
bc965ab3 TT |
256 | "couldn't mark inode dirty (err %d)", err); |
257 | goto stop_handle; | |
258 | } | |
ac27a0ec | 259 | if (inode->i_blocks) |
617ba13b | 260 | ext4_truncate(inode); |
bc965ab3 TT |
261 | |
262 | /* | |
263 | * ext4_ext_truncate() doesn't reserve any slop when it | |
264 | * restarts journal transactions; therefore there may not be | |
265 | * enough credits left in the handle to remove the inode from | |
266 | * the orphan list and set the dtime field. | |
267 | */ | |
0390131b | 268 | if (!ext4_handle_has_enough_credits(handle, 3)) { |
bc965ab3 TT |
269 | err = ext4_journal_extend(handle, 3); |
270 | if (err > 0) | |
271 | err = ext4_journal_restart(handle, 3); | |
272 | if (err != 0) { | |
12062ddd | 273 | ext4_warning(inode->i_sb, |
bc965ab3 TT |
274 | "couldn't extend journal (err %d)", err); |
275 | stop_handle: | |
276 | ext4_journal_stop(handle); | |
45388219 | 277 | ext4_orphan_del(NULL, inode); |
8e8ad8a5 | 278 | sb_end_intwrite(inode->i_sb); |
bc965ab3 TT |
279 | goto no_delete; |
280 | } | |
281 | } | |
282 | ||
ac27a0ec | 283 | /* |
617ba13b | 284 | * Kill off the orphan record which ext4_truncate created. |
ac27a0ec | 285 | * AKPM: I think this can be inside the above `if'. |
617ba13b | 286 | * Note that ext4_orphan_del() has to be able to cope with the |
ac27a0ec | 287 | * deletion of a non-existent orphan - this is because we don't |
617ba13b | 288 | * know if ext4_truncate() actually created an orphan record. |
ac27a0ec DK |
289 | * (Well, we could do this if we need to, but heck - it works) |
290 | */ | |
617ba13b MC |
291 | ext4_orphan_del(handle, inode); |
292 | EXT4_I(inode)->i_dtime = get_seconds(); | |
ac27a0ec DK |
293 | |
294 | /* | |
295 | * One subtle ordering requirement: if anything has gone wrong | |
296 | * (transaction abort, IO errors, whatever), then we can still | |
297 | * do these next steps (the fs will already have been marked as | |
298 | * having errors), but we can't free the inode if the mark_dirty | |
299 | * fails. | |
300 | */ | |
617ba13b | 301 | if (ext4_mark_inode_dirty(handle, inode)) |
ac27a0ec | 302 | /* If that failed, just do the required in-core inode clear. */ |
0930fcc1 | 303 | ext4_clear_inode(inode); |
ac27a0ec | 304 | else |
617ba13b MC |
305 | ext4_free_inode(handle, inode); |
306 | ext4_journal_stop(handle); | |
8e8ad8a5 | 307 | sb_end_intwrite(inode->i_sb); |
ac27a0ec DK |
308 | return; |
309 | no_delete: | |
0930fcc1 | 310 | ext4_clear_inode(inode); /* We must guarantee clearing of inode... */ |
ac27a0ec DK |
311 | } |
312 | ||
a9e7f447 DM |
313 | #ifdef CONFIG_QUOTA |
314 | qsize_t *ext4_get_reserved_space(struct inode *inode) | |
60e58e0f | 315 | { |
a9e7f447 | 316 | return &EXT4_I(inode)->i_reserved_quota; |
60e58e0f | 317 | } |
a9e7f447 | 318 | #endif |
9d0be502 | 319 | |
12219aea AK |
320 | /* |
321 | * Calculate the number of metadata blocks need to reserve | |
9d0be502 | 322 | * to allocate a block located at @lblock |
12219aea | 323 | */ |
01f49d0b | 324 | static int ext4_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock) |
12219aea | 325 | { |
12e9b892 | 326 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
9d0be502 | 327 | return ext4_ext_calc_metadata_amount(inode, lblock); |
12219aea | 328 | |
8bb2b247 | 329 | return ext4_ind_calc_metadata_amount(inode, lblock); |
12219aea AK |
330 | } |
331 | ||
0637c6f4 TT |
332 | /* |
333 | * Called with i_data_sem down, which is important since we can call | |
334 | * ext4_discard_preallocations() from here. | |
335 | */ | |
5f634d06 AK |
336 | void ext4_da_update_reserve_space(struct inode *inode, |
337 | int used, int quota_claim) | |
12219aea AK |
338 | { |
339 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0637c6f4 | 340 | struct ext4_inode_info *ei = EXT4_I(inode); |
0637c6f4 TT |
341 | |
342 | spin_lock(&ei->i_block_reservation_lock); | |
d8990240 | 343 | trace_ext4_da_update_reserve_space(inode, used, quota_claim); |
0637c6f4 | 344 | if (unlikely(used > ei->i_reserved_data_blocks)) { |
8de5c325 | 345 | ext4_warning(inode->i_sb, "%s: ino %lu, used %d " |
1084f252 | 346 | "with only %d reserved data blocks", |
0637c6f4 TT |
347 | __func__, inode->i_ino, used, |
348 | ei->i_reserved_data_blocks); | |
349 | WARN_ON(1); | |
350 | used = ei->i_reserved_data_blocks; | |
351 | } | |
12219aea | 352 | |
97795d2a | 353 | if (unlikely(ei->i_allocated_meta_blocks > ei->i_reserved_meta_blocks)) { |
01a523eb TT |
354 | ext4_warning(inode->i_sb, "ino %lu, allocated %d " |
355 | "with only %d reserved metadata blocks " | |
356 | "(releasing %d blocks with reserved %d data blocks)", | |
357 | inode->i_ino, ei->i_allocated_meta_blocks, | |
358 | ei->i_reserved_meta_blocks, used, | |
359 | ei->i_reserved_data_blocks); | |
97795d2a BF |
360 | WARN_ON(1); |
361 | ei->i_allocated_meta_blocks = ei->i_reserved_meta_blocks; | |
362 | } | |
363 | ||
0637c6f4 TT |
364 | /* Update per-inode reservations */ |
365 | ei->i_reserved_data_blocks -= used; | |
0637c6f4 | 366 | ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks; |
57042651 | 367 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, |
72b8ab9d | 368 | used + ei->i_allocated_meta_blocks); |
0637c6f4 | 369 | ei->i_allocated_meta_blocks = 0; |
6bc6e63f | 370 | |
0637c6f4 TT |
371 | if (ei->i_reserved_data_blocks == 0) { |
372 | /* | |
373 | * We can release all of the reserved metadata blocks | |
374 | * only when we have written all of the delayed | |
375 | * allocation blocks. | |
376 | */ | |
57042651 | 377 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, |
72b8ab9d | 378 | ei->i_reserved_meta_blocks); |
ee5f4d9c | 379 | ei->i_reserved_meta_blocks = 0; |
9d0be502 | 380 | ei->i_da_metadata_calc_len = 0; |
6bc6e63f | 381 | } |
12219aea | 382 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
60e58e0f | 383 | |
72b8ab9d ES |
384 | /* Update quota subsystem for data blocks */ |
385 | if (quota_claim) | |
7b415bf6 | 386 | dquot_claim_block(inode, EXT4_C2B(sbi, used)); |
72b8ab9d | 387 | else { |
5f634d06 AK |
388 | /* |
389 | * We did fallocate with an offset that is already delayed | |
390 | * allocated. So on delayed allocated writeback we should | |
72b8ab9d | 391 | * not re-claim the quota for fallocated blocks. |
5f634d06 | 392 | */ |
7b415bf6 | 393 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, used)); |
5f634d06 | 394 | } |
d6014301 AK |
395 | |
396 | /* | |
397 | * If we have done all the pending block allocations and if | |
398 | * there aren't any writers on the inode, we can discard the | |
399 | * inode's preallocations. | |
400 | */ | |
0637c6f4 TT |
401 | if ((ei->i_reserved_data_blocks == 0) && |
402 | (atomic_read(&inode->i_writecount) == 0)) | |
d6014301 | 403 | ext4_discard_preallocations(inode); |
12219aea AK |
404 | } |
405 | ||
e29136f8 | 406 | static int __check_block_validity(struct inode *inode, const char *func, |
c398eda0 TT |
407 | unsigned int line, |
408 | struct ext4_map_blocks *map) | |
6fd058f7 | 409 | { |
24676da4 TT |
410 | if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk, |
411 | map->m_len)) { | |
c398eda0 TT |
412 | ext4_error_inode(inode, func, line, map->m_pblk, |
413 | "lblock %lu mapped to illegal pblock " | |
414 | "(length %d)", (unsigned long) map->m_lblk, | |
415 | map->m_len); | |
6fd058f7 TT |
416 | return -EIO; |
417 | } | |
418 | return 0; | |
419 | } | |
420 | ||
e29136f8 | 421 | #define check_block_validity(inode, map) \ |
c398eda0 | 422 | __check_block_validity((inode), __func__, __LINE__, (map)) |
e29136f8 | 423 | |
55138e0b | 424 | /* |
1f94533d TT |
425 | * Return the number of contiguous dirty pages in a given inode |
426 | * starting at page frame idx. | |
55138e0b TT |
427 | */ |
428 | static pgoff_t ext4_num_dirty_pages(struct inode *inode, pgoff_t idx, | |
429 | unsigned int max_pages) | |
430 | { | |
431 | struct address_space *mapping = inode->i_mapping; | |
432 | pgoff_t index; | |
433 | struct pagevec pvec; | |
434 | pgoff_t num = 0; | |
435 | int i, nr_pages, done = 0; | |
436 | ||
437 | if (max_pages == 0) | |
438 | return 0; | |
439 | pagevec_init(&pvec, 0); | |
440 | while (!done) { | |
441 | index = idx; | |
442 | nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, | |
443 | PAGECACHE_TAG_DIRTY, | |
444 | (pgoff_t)PAGEVEC_SIZE); | |
445 | if (nr_pages == 0) | |
446 | break; | |
447 | for (i = 0; i < nr_pages; i++) { | |
448 | struct page *page = pvec.pages[i]; | |
449 | struct buffer_head *bh, *head; | |
450 | ||
451 | lock_page(page); | |
452 | if (unlikely(page->mapping != mapping) || | |
453 | !PageDirty(page) || | |
454 | PageWriteback(page) || | |
455 | page->index != idx) { | |
456 | done = 1; | |
457 | unlock_page(page); | |
458 | break; | |
459 | } | |
1f94533d TT |
460 | if (page_has_buffers(page)) { |
461 | bh = head = page_buffers(page); | |
462 | do { | |
463 | if (!buffer_delay(bh) && | |
464 | !buffer_unwritten(bh)) | |
465 | done = 1; | |
466 | bh = bh->b_this_page; | |
467 | } while (!done && (bh != head)); | |
468 | } | |
55138e0b TT |
469 | unlock_page(page); |
470 | if (done) | |
471 | break; | |
472 | idx++; | |
473 | num++; | |
659c6009 ES |
474 | if (num >= max_pages) { |
475 | done = 1; | |
55138e0b | 476 | break; |
659c6009 | 477 | } |
55138e0b TT |
478 | } |
479 | pagevec_release(&pvec); | |
480 | } | |
481 | return num; | |
482 | } | |
483 | ||
921f266b DM |
484 | #ifdef ES_AGGRESSIVE_TEST |
485 | static void ext4_map_blocks_es_recheck(handle_t *handle, | |
486 | struct inode *inode, | |
487 | struct ext4_map_blocks *es_map, | |
488 | struct ext4_map_blocks *map, | |
489 | int flags) | |
490 | { | |
491 | int retval; | |
492 | ||
493 | map->m_flags = 0; | |
494 | /* | |
495 | * There is a race window that the result is not the same. | |
496 | * e.g. xfstests #223 when dioread_nolock enables. The reason | |
497 | * is that we lookup a block mapping in extent status tree with | |
498 | * out taking i_data_sem. So at the time the unwritten extent | |
499 | * could be converted. | |
500 | */ | |
501 | if (!(flags & EXT4_GET_BLOCKS_NO_LOCK)) | |
502 | down_read((&EXT4_I(inode)->i_data_sem)); | |
503 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { | |
504 | retval = ext4_ext_map_blocks(handle, inode, map, flags & | |
505 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
506 | } else { | |
507 | retval = ext4_ind_map_blocks(handle, inode, map, flags & | |
508 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
509 | } | |
510 | if (!(flags & EXT4_GET_BLOCKS_NO_LOCK)) | |
511 | up_read((&EXT4_I(inode)->i_data_sem)); | |
512 | /* | |
513 | * Clear EXT4_MAP_FROM_CLUSTER and EXT4_MAP_BOUNDARY flag | |
514 | * because it shouldn't be marked in es_map->m_flags. | |
515 | */ | |
516 | map->m_flags &= ~(EXT4_MAP_FROM_CLUSTER | EXT4_MAP_BOUNDARY); | |
517 | ||
518 | /* | |
519 | * We don't check m_len because extent will be collpased in status | |
520 | * tree. So the m_len might not equal. | |
521 | */ | |
522 | if (es_map->m_lblk != map->m_lblk || | |
523 | es_map->m_flags != map->m_flags || | |
524 | es_map->m_pblk != map->m_pblk) { | |
525 | printk("ES cache assertation failed for inode: %lu " | |
526 | "es_cached ex [%d/%d/%llu/%x] != " | |
527 | "found ex [%d/%d/%llu/%x] retval %d flags %x\n", | |
528 | inode->i_ino, es_map->m_lblk, es_map->m_len, | |
529 | es_map->m_pblk, es_map->m_flags, map->m_lblk, | |
530 | map->m_len, map->m_pblk, map->m_flags, | |
531 | retval, flags); | |
532 | } | |
533 | } | |
534 | #endif /* ES_AGGRESSIVE_TEST */ | |
535 | ||
f5ab0d1f | 536 | /* |
e35fd660 | 537 | * The ext4_map_blocks() function tries to look up the requested blocks, |
2b2d6d01 | 538 | * and returns if the blocks are already mapped. |
f5ab0d1f | 539 | * |
f5ab0d1f MC |
540 | * Otherwise it takes the write lock of the i_data_sem and allocate blocks |
541 | * and store the allocated blocks in the result buffer head and mark it | |
542 | * mapped. | |
543 | * | |
e35fd660 TT |
544 | * If file type is extents based, it will call ext4_ext_map_blocks(), |
545 | * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping | |
f5ab0d1f MC |
546 | * based files |
547 | * | |
548 | * On success, it returns the number of blocks being mapped or allocate. | |
549 | * if create==0 and the blocks are pre-allocated and uninitialized block, | |
550 | * the result buffer head is unmapped. If the create ==1, it will make sure | |
551 | * the buffer head is mapped. | |
552 | * | |
553 | * It returns 0 if plain look up failed (blocks have not been allocated), in | |
df3ab170 | 554 | * that case, buffer head is unmapped |
f5ab0d1f MC |
555 | * |
556 | * It returns the error in case of allocation failure. | |
557 | */ | |
e35fd660 TT |
558 | int ext4_map_blocks(handle_t *handle, struct inode *inode, |
559 | struct ext4_map_blocks *map, int flags) | |
0e855ac8 | 560 | { |
d100eef2 | 561 | struct extent_status es; |
0e855ac8 | 562 | int retval; |
921f266b DM |
563 | #ifdef ES_AGGRESSIVE_TEST |
564 | struct ext4_map_blocks orig_map; | |
565 | ||
566 | memcpy(&orig_map, map, sizeof(*map)); | |
567 | #endif | |
f5ab0d1f | 568 | |
e35fd660 TT |
569 | map->m_flags = 0; |
570 | ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u," | |
571 | "logical block %lu\n", inode->i_ino, flags, map->m_len, | |
572 | (unsigned long) map->m_lblk); | |
d100eef2 ZL |
573 | |
574 | /* Lookup extent status tree firstly */ | |
575 | if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) { | |
576 | if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) { | |
577 | map->m_pblk = ext4_es_pblock(&es) + | |
578 | map->m_lblk - es.es_lblk; | |
579 | map->m_flags |= ext4_es_is_written(&es) ? | |
580 | EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN; | |
581 | retval = es.es_len - (map->m_lblk - es.es_lblk); | |
582 | if (retval > map->m_len) | |
583 | retval = map->m_len; | |
584 | map->m_len = retval; | |
585 | } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) { | |
586 | retval = 0; | |
587 | } else { | |
588 | BUG_ON(1); | |
589 | } | |
921f266b DM |
590 | #ifdef ES_AGGRESSIVE_TEST |
591 | ext4_map_blocks_es_recheck(handle, inode, map, | |
592 | &orig_map, flags); | |
593 | #endif | |
d100eef2 ZL |
594 | goto found; |
595 | } | |
596 | ||
4df3d265 | 597 | /* |
b920c755 TT |
598 | * Try to see if we can get the block without requesting a new |
599 | * file system block. | |
4df3d265 | 600 | */ |
729f52c6 ZL |
601 | if (!(flags & EXT4_GET_BLOCKS_NO_LOCK)) |
602 | down_read((&EXT4_I(inode)->i_data_sem)); | |
12e9b892 | 603 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
a4e5d88b DM |
604 | retval = ext4_ext_map_blocks(handle, inode, map, flags & |
605 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
0e855ac8 | 606 | } else { |
a4e5d88b DM |
607 | retval = ext4_ind_map_blocks(handle, inode, map, flags & |
608 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
0e855ac8 | 609 | } |
f7fec032 ZL |
610 | if (retval > 0) { |
611 | int ret; | |
612 | unsigned long long status; | |
613 | ||
921f266b DM |
614 | #ifdef ES_AGGRESSIVE_TEST |
615 | if (retval != map->m_len) { | |
616 | printk("ES len assertation failed for inode: %lu " | |
617 | "retval %d != map->m_len %d " | |
618 | "in %s (lookup)\n", inode->i_ino, retval, | |
619 | map->m_len, __func__); | |
620 | } | |
621 | #endif | |
622 | ||
f7fec032 ZL |
623 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
624 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; | |
625 | if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) && | |
626 | ext4_find_delalloc_range(inode, map->m_lblk, | |
627 | map->m_lblk + map->m_len - 1)) | |
628 | status |= EXTENT_STATUS_DELAYED; | |
629 | ret = ext4_es_insert_extent(inode, map->m_lblk, | |
630 | map->m_len, map->m_pblk, status); | |
631 | if (ret < 0) | |
632 | retval = ret; | |
633 | } | |
729f52c6 ZL |
634 | if (!(flags & EXT4_GET_BLOCKS_NO_LOCK)) |
635 | up_read((&EXT4_I(inode)->i_data_sem)); | |
f5ab0d1f | 636 | |
d100eef2 | 637 | found: |
e35fd660 | 638 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
f7fec032 | 639 | int ret = check_block_validity(inode, map); |
6fd058f7 TT |
640 | if (ret != 0) |
641 | return ret; | |
642 | } | |
643 | ||
f5ab0d1f | 644 | /* If it is only a block(s) look up */ |
c2177057 | 645 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) |
f5ab0d1f MC |
646 | return retval; |
647 | ||
648 | /* | |
649 | * Returns if the blocks have already allocated | |
650 | * | |
651 | * Note that if blocks have been preallocated | |
df3ab170 | 652 | * ext4_ext_get_block() returns the create = 0 |
f5ab0d1f MC |
653 | * with buffer head unmapped. |
654 | */ | |
e35fd660 | 655 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) |
4df3d265 AK |
656 | return retval; |
657 | ||
2a8964d6 | 658 | /* |
a25a4e1a ZL |
659 | * Here we clear m_flags because after allocating an new extent, |
660 | * it will be set again. | |
2a8964d6 | 661 | */ |
a25a4e1a | 662 | map->m_flags &= ~EXT4_MAP_FLAGS; |
2a8964d6 | 663 | |
4df3d265 | 664 | /* |
f5ab0d1f MC |
665 | * New blocks allocate and/or writing to uninitialized extent |
666 | * will possibly result in updating i_data, so we take | |
667 | * the write lock of i_data_sem, and call get_blocks() | |
668 | * with create == 1 flag. | |
4df3d265 AK |
669 | */ |
670 | down_write((&EXT4_I(inode)->i_data_sem)); | |
d2a17637 MC |
671 | |
672 | /* | |
673 | * if the caller is from delayed allocation writeout path | |
674 | * we have already reserved fs blocks for allocation | |
675 | * let the underlying get_block() function know to | |
676 | * avoid double accounting | |
677 | */ | |
c2177057 | 678 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) |
f2321097 | 679 | ext4_set_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED); |
4df3d265 AK |
680 | /* |
681 | * We need to check for EXT4 here because migrate | |
682 | * could have changed the inode type in between | |
683 | */ | |
12e9b892 | 684 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
e35fd660 | 685 | retval = ext4_ext_map_blocks(handle, inode, map, flags); |
0e855ac8 | 686 | } else { |
e35fd660 | 687 | retval = ext4_ind_map_blocks(handle, inode, map, flags); |
267e4db9 | 688 | |
e35fd660 | 689 | if (retval > 0 && map->m_flags & EXT4_MAP_NEW) { |
267e4db9 AK |
690 | /* |
691 | * We allocated new blocks which will result in | |
692 | * i_data's format changing. Force the migrate | |
693 | * to fail by clearing migrate flags | |
694 | */ | |
19f5fb7a | 695 | ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE); |
267e4db9 | 696 | } |
d2a17637 | 697 | |
5f634d06 AK |
698 | /* |
699 | * Update reserved blocks/metadata blocks after successful | |
700 | * block allocation which had been deferred till now. We don't | |
701 | * support fallocate for non extent files. So we can update | |
702 | * reserve space here. | |
703 | */ | |
704 | if ((retval > 0) && | |
1296cc85 | 705 | (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)) |
5f634d06 AK |
706 | ext4_da_update_reserve_space(inode, retval, 1); |
707 | } | |
f7fec032 | 708 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) |
f2321097 | 709 | ext4_clear_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED); |
2ac3b6e0 | 710 | |
f7fec032 ZL |
711 | if (retval > 0) { |
712 | int ret; | |
713 | unsigned long long status; | |
714 | ||
921f266b DM |
715 | #ifdef ES_AGGRESSIVE_TEST |
716 | if (retval != map->m_len) { | |
717 | printk("ES len assertation failed for inode: %lu " | |
718 | "retval %d != map->m_len %d " | |
719 | "in %s (allocation)\n", inode->i_ino, retval, | |
720 | map->m_len, __func__); | |
721 | } | |
722 | #endif | |
723 | ||
adb23551 ZL |
724 | /* |
725 | * If the extent has been zeroed out, we don't need to update | |
726 | * extent status tree. | |
727 | */ | |
728 | if ((flags & EXT4_GET_BLOCKS_PRE_IO) && | |
729 | ext4_es_lookup_extent(inode, map->m_lblk, &es)) { | |
730 | if (ext4_es_is_written(&es)) | |
731 | goto has_zeroout; | |
732 | } | |
f7fec032 ZL |
733 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
734 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; | |
735 | if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) && | |
736 | ext4_find_delalloc_range(inode, map->m_lblk, | |
737 | map->m_lblk + map->m_len - 1)) | |
738 | status |= EXTENT_STATUS_DELAYED; | |
739 | ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, | |
740 | map->m_pblk, status); | |
741 | if (ret < 0) | |
742 | retval = ret; | |
5356f261 AK |
743 | } |
744 | ||
adb23551 | 745 | has_zeroout: |
4df3d265 | 746 | up_write((&EXT4_I(inode)->i_data_sem)); |
e35fd660 | 747 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
e29136f8 | 748 | int ret = check_block_validity(inode, map); |
6fd058f7 TT |
749 | if (ret != 0) |
750 | return ret; | |
751 | } | |
0e855ac8 AK |
752 | return retval; |
753 | } | |
754 | ||
f3bd1f3f MC |
755 | /* Maximum number of blocks we map for direct IO at once. */ |
756 | #define DIO_MAX_BLOCKS 4096 | |
757 | ||
2ed88685 TT |
758 | static int _ext4_get_block(struct inode *inode, sector_t iblock, |
759 | struct buffer_head *bh, int flags) | |
ac27a0ec | 760 | { |
3e4fdaf8 | 761 | handle_t *handle = ext4_journal_current_handle(); |
2ed88685 | 762 | struct ext4_map_blocks map; |
7fb5409d | 763 | int ret = 0, started = 0; |
f3bd1f3f | 764 | int dio_credits; |
ac27a0ec | 765 | |
46c7f254 TM |
766 | if (ext4_has_inline_data(inode)) |
767 | return -ERANGE; | |
768 | ||
2ed88685 TT |
769 | map.m_lblk = iblock; |
770 | map.m_len = bh->b_size >> inode->i_blkbits; | |
771 | ||
8b0f165f | 772 | if (flags && !(flags & EXT4_GET_BLOCKS_NO_LOCK) && !handle) { |
7fb5409d | 773 | /* Direct IO write... */ |
2ed88685 TT |
774 | if (map.m_len > DIO_MAX_BLOCKS) |
775 | map.m_len = DIO_MAX_BLOCKS; | |
776 | dio_credits = ext4_chunk_trans_blocks(inode, map.m_len); | |
9924a92a TT |
777 | handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, |
778 | dio_credits); | |
7fb5409d | 779 | if (IS_ERR(handle)) { |
ac27a0ec | 780 | ret = PTR_ERR(handle); |
2ed88685 | 781 | return ret; |
ac27a0ec | 782 | } |
7fb5409d | 783 | started = 1; |
ac27a0ec DK |
784 | } |
785 | ||
2ed88685 | 786 | ret = ext4_map_blocks(handle, inode, &map, flags); |
7fb5409d | 787 | if (ret > 0) { |
2ed88685 TT |
788 | map_bh(bh, inode->i_sb, map.m_pblk); |
789 | bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags; | |
790 | bh->b_size = inode->i_sb->s_blocksize * map.m_len; | |
7fb5409d | 791 | ret = 0; |
ac27a0ec | 792 | } |
7fb5409d JK |
793 | if (started) |
794 | ext4_journal_stop(handle); | |
ac27a0ec DK |
795 | return ret; |
796 | } | |
797 | ||
2ed88685 TT |
798 | int ext4_get_block(struct inode *inode, sector_t iblock, |
799 | struct buffer_head *bh, int create) | |
800 | { | |
801 | return _ext4_get_block(inode, iblock, bh, | |
802 | create ? EXT4_GET_BLOCKS_CREATE : 0); | |
803 | } | |
804 | ||
ac27a0ec DK |
805 | /* |
806 | * `handle' can be NULL if create is zero | |
807 | */ | |
617ba13b | 808 | struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode, |
725d26d3 | 809 | ext4_lblk_t block, int create, int *errp) |
ac27a0ec | 810 | { |
2ed88685 TT |
811 | struct ext4_map_blocks map; |
812 | struct buffer_head *bh; | |
ac27a0ec DK |
813 | int fatal = 0, err; |
814 | ||
815 | J_ASSERT(handle != NULL || create == 0); | |
816 | ||
2ed88685 TT |
817 | map.m_lblk = block; |
818 | map.m_len = 1; | |
819 | err = ext4_map_blocks(handle, inode, &map, | |
820 | create ? EXT4_GET_BLOCKS_CREATE : 0); | |
ac27a0ec | 821 | |
90b0a973 CM |
822 | /* ensure we send some value back into *errp */ |
823 | *errp = 0; | |
824 | ||
0f70b406 TT |
825 | if (create && err == 0) |
826 | err = -ENOSPC; /* should never happen */ | |
2ed88685 TT |
827 | if (err < 0) |
828 | *errp = err; | |
829 | if (err <= 0) | |
830 | return NULL; | |
2ed88685 TT |
831 | |
832 | bh = sb_getblk(inode->i_sb, map.m_pblk); | |
aebf0243 | 833 | if (unlikely(!bh)) { |
860d21e2 | 834 | *errp = -ENOMEM; |
2ed88685 | 835 | return NULL; |
ac27a0ec | 836 | } |
2ed88685 TT |
837 | if (map.m_flags & EXT4_MAP_NEW) { |
838 | J_ASSERT(create != 0); | |
839 | J_ASSERT(handle != NULL); | |
ac27a0ec | 840 | |
2ed88685 TT |
841 | /* |
842 | * Now that we do not always journal data, we should | |
843 | * keep in mind whether this should always journal the | |
844 | * new buffer as metadata. For now, regular file | |
845 | * writes use ext4_get_block instead, so it's not a | |
846 | * problem. | |
847 | */ | |
848 | lock_buffer(bh); | |
849 | BUFFER_TRACE(bh, "call get_create_access"); | |
850 | fatal = ext4_journal_get_create_access(handle, bh); | |
851 | if (!fatal && !buffer_uptodate(bh)) { | |
852 | memset(bh->b_data, 0, inode->i_sb->s_blocksize); | |
853 | set_buffer_uptodate(bh); | |
ac27a0ec | 854 | } |
2ed88685 TT |
855 | unlock_buffer(bh); |
856 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); | |
857 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
858 | if (!fatal) | |
859 | fatal = err; | |
860 | } else { | |
861 | BUFFER_TRACE(bh, "not a new buffer"); | |
ac27a0ec | 862 | } |
2ed88685 TT |
863 | if (fatal) { |
864 | *errp = fatal; | |
865 | brelse(bh); | |
866 | bh = NULL; | |
867 | } | |
868 | return bh; | |
ac27a0ec DK |
869 | } |
870 | ||
617ba13b | 871 | struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode, |
725d26d3 | 872 | ext4_lblk_t block, int create, int *err) |
ac27a0ec | 873 | { |
af5bc92d | 874 | struct buffer_head *bh; |
ac27a0ec | 875 | |
617ba13b | 876 | bh = ext4_getblk(handle, inode, block, create, err); |
ac27a0ec DK |
877 | if (!bh) |
878 | return bh; | |
879 | if (buffer_uptodate(bh)) | |
880 | return bh; | |
65299a3b | 881 | ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh); |
ac27a0ec DK |
882 | wait_on_buffer(bh); |
883 | if (buffer_uptodate(bh)) | |
884 | return bh; | |
885 | put_bh(bh); | |
886 | *err = -EIO; | |
887 | return NULL; | |
888 | } | |
889 | ||
f19d5870 TM |
890 | int ext4_walk_page_buffers(handle_t *handle, |
891 | struct buffer_head *head, | |
892 | unsigned from, | |
893 | unsigned to, | |
894 | int *partial, | |
895 | int (*fn)(handle_t *handle, | |
896 | struct buffer_head *bh)) | |
ac27a0ec DK |
897 | { |
898 | struct buffer_head *bh; | |
899 | unsigned block_start, block_end; | |
900 | unsigned blocksize = head->b_size; | |
901 | int err, ret = 0; | |
902 | struct buffer_head *next; | |
903 | ||
af5bc92d TT |
904 | for (bh = head, block_start = 0; |
905 | ret == 0 && (bh != head || !block_start); | |
de9a55b8 | 906 | block_start = block_end, bh = next) { |
ac27a0ec DK |
907 | next = bh->b_this_page; |
908 | block_end = block_start + blocksize; | |
909 | if (block_end <= from || block_start >= to) { | |
910 | if (partial && !buffer_uptodate(bh)) | |
911 | *partial = 1; | |
912 | continue; | |
913 | } | |
914 | err = (*fn)(handle, bh); | |
915 | if (!ret) | |
916 | ret = err; | |
917 | } | |
918 | return ret; | |
919 | } | |
920 | ||
921 | /* | |
922 | * To preserve ordering, it is essential that the hole instantiation and | |
923 | * the data write be encapsulated in a single transaction. We cannot | |
617ba13b | 924 | * close off a transaction and start a new one between the ext4_get_block() |
dab291af | 925 | * and the commit_write(). So doing the jbd2_journal_start at the start of |
ac27a0ec DK |
926 | * prepare_write() is the right place. |
927 | * | |
36ade451 JK |
928 | * Also, this function can nest inside ext4_writepage(). In that case, we |
929 | * *know* that ext4_writepage() has generated enough buffer credits to do the | |
930 | * whole page. So we won't block on the journal in that case, which is good, | |
931 | * because the caller may be PF_MEMALLOC. | |
ac27a0ec | 932 | * |
617ba13b | 933 | * By accident, ext4 can be reentered when a transaction is open via |
ac27a0ec DK |
934 | * quota file writes. If we were to commit the transaction while thus |
935 | * reentered, there can be a deadlock - we would be holding a quota | |
936 | * lock, and the commit would never complete if another thread had a | |
937 | * transaction open and was blocking on the quota lock - a ranking | |
938 | * violation. | |
939 | * | |
dab291af | 940 | * So what we do is to rely on the fact that jbd2_journal_stop/journal_start |
ac27a0ec DK |
941 | * will _not_ run commit under these circumstances because handle->h_ref |
942 | * is elevated. We'll still have enough credits for the tiny quotafile | |
943 | * write. | |
944 | */ | |
f19d5870 TM |
945 | int do_journal_get_write_access(handle_t *handle, |
946 | struct buffer_head *bh) | |
ac27a0ec | 947 | { |
56d35a4c JK |
948 | int dirty = buffer_dirty(bh); |
949 | int ret; | |
950 | ||
ac27a0ec DK |
951 | if (!buffer_mapped(bh) || buffer_freed(bh)) |
952 | return 0; | |
56d35a4c | 953 | /* |
ebdec241 | 954 | * __block_write_begin() could have dirtied some buffers. Clean |
56d35a4c JK |
955 | * the dirty bit as jbd2_journal_get_write_access() could complain |
956 | * otherwise about fs integrity issues. Setting of the dirty bit | |
ebdec241 | 957 | * by __block_write_begin() isn't a real problem here as we clear |
56d35a4c JK |
958 | * the bit before releasing a page lock and thus writeback cannot |
959 | * ever write the buffer. | |
960 | */ | |
961 | if (dirty) | |
962 | clear_buffer_dirty(bh); | |
963 | ret = ext4_journal_get_write_access(handle, bh); | |
964 | if (!ret && dirty) | |
965 | ret = ext4_handle_dirty_metadata(handle, NULL, bh); | |
966 | return ret; | |
ac27a0ec DK |
967 | } |
968 | ||
8b0f165f AP |
969 | static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock, |
970 | struct buffer_head *bh_result, int create); | |
bfc1af65 | 971 | static int ext4_write_begin(struct file *file, struct address_space *mapping, |
de9a55b8 TT |
972 | loff_t pos, unsigned len, unsigned flags, |
973 | struct page **pagep, void **fsdata) | |
ac27a0ec | 974 | { |
af5bc92d | 975 | struct inode *inode = mapping->host; |
1938a150 | 976 | int ret, needed_blocks; |
ac27a0ec DK |
977 | handle_t *handle; |
978 | int retries = 0; | |
af5bc92d | 979 | struct page *page; |
de9a55b8 | 980 | pgoff_t index; |
af5bc92d | 981 | unsigned from, to; |
bfc1af65 | 982 | |
9bffad1e | 983 | trace_ext4_write_begin(inode, pos, len, flags); |
1938a150 AK |
984 | /* |
985 | * Reserve one block more for addition to orphan list in case | |
986 | * we allocate blocks but write fails for some reason | |
987 | */ | |
988 | needed_blocks = ext4_writepage_trans_blocks(inode) + 1; | |
de9a55b8 | 989 | index = pos >> PAGE_CACHE_SHIFT; |
af5bc92d TT |
990 | from = pos & (PAGE_CACHE_SIZE - 1); |
991 | to = from + len; | |
ac27a0ec | 992 | |
f19d5870 TM |
993 | if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { |
994 | ret = ext4_try_to_write_inline_data(mapping, inode, pos, len, | |
995 | flags, pagep); | |
996 | if (ret < 0) | |
47564bfb TT |
997 | return ret; |
998 | if (ret == 1) | |
999 | return 0; | |
f19d5870 TM |
1000 | } |
1001 | ||
47564bfb TT |
1002 | /* |
1003 | * grab_cache_page_write_begin() can take a long time if the | |
1004 | * system is thrashing due to memory pressure, or if the page | |
1005 | * is being written back. So grab it first before we start | |
1006 | * the transaction handle. This also allows us to allocate | |
1007 | * the page (if needed) without using GFP_NOFS. | |
1008 | */ | |
1009 | retry_grab: | |
1010 | page = grab_cache_page_write_begin(mapping, index, flags); | |
1011 | if (!page) | |
1012 | return -ENOMEM; | |
1013 | unlock_page(page); | |
1014 | ||
1015 | retry_journal: | |
9924a92a | 1016 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks); |
af5bc92d | 1017 | if (IS_ERR(handle)) { |
47564bfb TT |
1018 | page_cache_release(page); |
1019 | return PTR_ERR(handle); | |
7479d2b9 | 1020 | } |
ac27a0ec | 1021 | |
47564bfb TT |
1022 | lock_page(page); |
1023 | if (page->mapping != mapping) { | |
1024 | /* The page got truncated from under us */ | |
1025 | unlock_page(page); | |
1026 | page_cache_release(page); | |
cf108bca | 1027 | ext4_journal_stop(handle); |
47564bfb | 1028 | goto retry_grab; |
cf108bca | 1029 | } |
47564bfb | 1030 | wait_on_page_writeback(page); |
cf108bca | 1031 | |
744692dc | 1032 | if (ext4_should_dioread_nolock(inode)) |
6e1db88d | 1033 | ret = __block_write_begin(page, pos, len, ext4_get_block_write); |
744692dc | 1034 | else |
6e1db88d | 1035 | ret = __block_write_begin(page, pos, len, ext4_get_block); |
bfc1af65 NP |
1036 | |
1037 | if (!ret && ext4_should_journal_data(inode)) { | |
f19d5870 TM |
1038 | ret = ext4_walk_page_buffers(handle, page_buffers(page), |
1039 | from, to, NULL, | |
1040 | do_journal_get_write_access); | |
ac27a0ec | 1041 | } |
bfc1af65 NP |
1042 | |
1043 | if (ret) { | |
af5bc92d | 1044 | unlock_page(page); |
ae4d5372 | 1045 | /* |
6e1db88d | 1046 | * __block_write_begin may have instantiated a few blocks |
ae4d5372 AK |
1047 | * outside i_size. Trim these off again. Don't need |
1048 | * i_size_read because we hold i_mutex. | |
1938a150 AK |
1049 | * |
1050 | * Add inode to orphan list in case we crash before | |
1051 | * truncate finishes | |
ae4d5372 | 1052 | */ |
ffacfa7a | 1053 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
1938a150 AK |
1054 | ext4_orphan_add(handle, inode); |
1055 | ||
1056 | ext4_journal_stop(handle); | |
1057 | if (pos + len > inode->i_size) { | |
b9a4207d | 1058 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1059 | /* |
ffacfa7a | 1060 | * If truncate failed early the inode might |
1938a150 AK |
1061 | * still be on the orphan list; we need to |
1062 | * make sure the inode is removed from the | |
1063 | * orphan list in that case. | |
1064 | */ | |
1065 | if (inode->i_nlink) | |
1066 | ext4_orphan_del(NULL, inode); | |
1067 | } | |
bfc1af65 | 1068 | |
47564bfb TT |
1069 | if (ret == -ENOSPC && |
1070 | ext4_should_retry_alloc(inode->i_sb, &retries)) | |
1071 | goto retry_journal; | |
1072 | page_cache_release(page); | |
1073 | return ret; | |
1074 | } | |
1075 | *pagep = page; | |
ac27a0ec DK |
1076 | return ret; |
1077 | } | |
1078 | ||
bfc1af65 NP |
1079 | /* For write_end() in data=journal mode */ |
1080 | static int write_end_fn(handle_t *handle, struct buffer_head *bh) | |
ac27a0ec | 1081 | { |
13fca323 | 1082 | int ret; |
ac27a0ec DK |
1083 | if (!buffer_mapped(bh) || buffer_freed(bh)) |
1084 | return 0; | |
1085 | set_buffer_uptodate(bh); | |
13fca323 TT |
1086 | ret = ext4_handle_dirty_metadata(handle, NULL, bh); |
1087 | clear_buffer_meta(bh); | |
1088 | clear_buffer_prio(bh); | |
1089 | return ret; | |
ac27a0ec DK |
1090 | } |
1091 | ||
eed4333f ZL |
1092 | /* |
1093 | * We need to pick up the new inode size which generic_commit_write gave us | |
1094 | * `file' can be NULL - eg, when called from page_symlink(). | |
1095 | * | |
1096 | * ext4 never places buffers on inode->i_mapping->private_list. metadata | |
1097 | * buffers are managed internally. | |
1098 | */ | |
1099 | static int ext4_write_end(struct file *file, | |
1100 | struct address_space *mapping, | |
1101 | loff_t pos, unsigned len, unsigned copied, | |
1102 | struct page *page, void *fsdata) | |
f8514083 | 1103 | { |
f8514083 | 1104 | handle_t *handle = ext4_journal_current_handle(); |
eed4333f ZL |
1105 | struct inode *inode = mapping->host; |
1106 | int ret = 0, ret2; | |
1107 | int i_size_changed = 0; | |
1108 | ||
1109 | trace_ext4_write_end(inode, pos, len, copied); | |
1110 | if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE)) { | |
1111 | ret = ext4_jbd2_file_inode(handle, inode); | |
1112 | if (ret) { | |
1113 | unlock_page(page); | |
1114 | page_cache_release(page); | |
1115 | goto errout; | |
1116 | } | |
1117 | } | |
f8514083 | 1118 | |
f19d5870 TM |
1119 | if (ext4_has_inline_data(inode)) |
1120 | copied = ext4_write_inline_data_end(inode, pos, len, | |
1121 | copied, page); | |
1122 | else | |
1123 | copied = block_write_end(file, mapping, pos, | |
1124 | len, copied, page, fsdata); | |
f8514083 AK |
1125 | |
1126 | /* | |
1127 | * No need to use i_size_read() here, the i_size | |
eed4333f | 1128 | * cannot change under us because we hole i_mutex. |
f8514083 AK |
1129 | * |
1130 | * But it's important to update i_size while still holding page lock: | |
1131 | * page writeout could otherwise come in and zero beyond i_size. | |
1132 | */ | |
1133 | if (pos + copied > inode->i_size) { | |
1134 | i_size_write(inode, pos + copied); | |
1135 | i_size_changed = 1; | |
1136 | } | |
1137 | ||
eed4333f | 1138 | if (pos + copied > EXT4_I(inode)->i_disksize) { |
f8514083 AK |
1139 | /* We need to mark inode dirty even if |
1140 | * new_i_size is less that inode->i_size | |
eed4333f | 1141 | * but greater than i_disksize. (hint delalloc) |
f8514083 AK |
1142 | */ |
1143 | ext4_update_i_disksize(inode, (pos + copied)); | |
1144 | i_size_changed = 1; | |
1145 | } | |
1146 | unlock_page(page); | |
1147 | page_cache_release(page); | |
1148 | ||
1149 | /* | |
1150 | * Don't mark the inode dirty under page lock. First, it unnecessarily | |
1151 | * makes the holding time of page lock longer. Second, it forces lock | |
1152 | * ordering of page lock and transaction start for journaling | |
1153 | * filesystems. | |
1154 | */ | |
1155 | if (i_size_changed) | |
1156 | ext4_mark_inode_dirty(handle, inode); | |
1157 | ||
74d553aa TT |
1158 | if (copied < 0) |
1159 | ret = copied; | |
ffacfa7a | 1160 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
f8514083 AK |
1161 | /* if we have allocated more blocks and copied |
1162 | * less. We will have blocks allocated outside | |
1163 | * inode->i_size. So truncate them | |
1164 | */ | |
1165 | ext4_orphan_add(handle, inode); | |
74d553aa | 1166 | errout: |
617ba13b | 1167 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1168 | if (!ret) |
1169 | ret = ret2; | |
bfc1af65 | 1170 | |
f8514083 | 1171 | if (pos + len > inode->i_size) { |
b9a4207d | 1172 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1173 | /* |
ffacfa7a | 1174 | * If truncate failed early the inode might still be |
f8514083 AK |
1175 | * on the orphan list; we need to make sure the inode |
1176 | * is removed from the orphan list in that case. | |
1177 | */ | |
1178 | if (inode->i_nlink) | |
1179 | ext4_orphan_del(NULL, inode); | |
1180 | } | |
1181 | ||
bfc1af65 | 1182 | return ret ? ret : copied; |
ac27a0ec DK |
1183 | } |
1184 | ||
bfc1af65 | 1185 | static int ext4_journalled_write_end(struct file *file, |
de9a55b8 TT |
1186 | struct address_space *mapping, |
1187 | loff_t pos, unsigned len, unsigned copied, | |
1188 | struct page *page, void *fsdata) | |
ac27a0ec | 1189 | { |
617ba13b | 1190 | handle_t *handle = ext4_journal_current_handle(); |
bfc1af65 | 1191 | struct inode *inode = mapping->host; |
ac27a0ec DK |
1192 | int ret = 0, ret2; |
1193 | int partial = 0; | |
bfc1af65 | 1194 | unsigned from, to; |
cf17fea6 | 1195 | loff_t new_i_size; |
ac27a0ec | 1196 | |
9bffad1e | 1197 | trace_ext4_journalled_write_end(inode, pos, len, copied); |
bfc1af65 NP |
1198 | from = pos & (PAGE_CACHE_SIZE - 1); |
1199 | to = from + len; | |
1200 | ||
441c8508 CW |
1201 | BUG_ON(!ext4_handle_valid(handle)); |
1202 | ||
3fdcfb66 TM |
1203 | if (ext4_has_inline_data(inode)) |
1204 | copied = ext4_write_inline_data_end(inode, pos, len, | |
1205 | copied, page); | |
1206 | else { | |
1207 | if (copied < len) { | |
1208 | if (!PageUptodate(page)) | |
1209 | copied = 0; | |
1210 | page_zero_new_buffers(page, from+copied, to); | |
1211 | } | |
ac27a0ec | 1212 | |
3fdcfb66 TM |
1213 | ret = ext4_walk_page_buffers(handle, page_buffers(page), from, |
1214 | to, &partial, write_end_fn); | |
1215 | if (!partial) | |
1216 | SetPageUptodate(page); | |
1217 | } | |
cf17fea6 AK |
1218 | new_i_size = pos + copied; |
1219 | if (new_i_size > inode->i_size) | |
bfc1af65 | 1220 | i_size_write(inode, pos+copied); |
19f5fb7a | 1221 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
2d859db3 | 1222 | EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; |
cf17fea6 AK |
1223 | if (new_i_size > EXT4_I(inode)->i_disksize) { |
1224 | ext4_update_i_disksize(inode, new_i_size); | |
617ba13b | 1225 | ret2 = ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
1226 | if (!ret) |
1227 | ret = ret2; | |
1228 | } | |
bfc1af65 | 1229 | |
cf108bca | 1230 | unlock_page(page); |
f8514083 | 1231 | page_cache_release(page); |
ffacfa7a | 1232 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
f8514083 AK |
1233 | /* if we have allocated more blocks and copied |
1234 | * less. We will have blocks allocated outside | |
1235 | * inode->i_size. So truncate them | |
1236 | */ | |
1237 | ext4_orphan_add(handle, inode); | |
1238 | ||
617ba13b | 1239 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1240 | if (!ret) |
1241 | ret = ret2; | |
f8514083 | 1242 | if (pos + len > inode->i_size) { |
b9a4207d | 1243 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1244 | /* |
ffacfa7a | 1245 | * If truncate failed early the inode might still be |
f8514083 AK |
1246 | * on the orphan list; we need to make sure the inode |
1247 | * is removed from the orphan list in that case. | |
1248 | */ | |
1249 | if (inode->i_nlink) | |
1250 | ext4_orphan_del(NULL, inode); | |
1251 | } | |
bfc1af65 NP |
1252 | |
1253 | return ret ? ret : copied; | |
ac27a0ec | 1254 | } |
d2a17637 | 1255 | |
386ad67c LC |
1256 | /* |
1257 | * Reserve a metadata for a single block located at lblock | |
1258 | */ | |
1259 | static int ext4_da_reserve_metadata(struct inode *inode, ext4_lblk_t lblock) | |
1260 | { | |
1261 | int retries = 0; | |
1262 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
1263 | struct ext4_inode_info *ei = EXT4_I(inode); | |
1264 | unsigned int md_needed; | |
1265 | ext4_lblk_t save_last_lblock; | |
1266 | int save_len; | |
1267 | ||
1268 | /* | |
1269 | * recalculate the amount of metadata blocks to reserve | |
1270 | * in order to allocate nrblocks | |
1271 | * worse case is one extent per block | |
1272 | */ | |
1273 | repeat: | |
1274 | spin_lock(&ei->i_block_reservation_lock); | |
1275 | /* | |
1276 | * ext4_calc_metadata_amount() has side effects, which we have | |
1277 | * to be prepared undo if we fail to claim space. | |
1278 | */ | |
1279 | save_len = ei->i_da_metadata_calc_len; | |
1280 | save_last_lblock = ei->i_da_metadata_calc_last_lblock; | |
1281 | md_needed = EXT4_NUM_B2C(sbi, | |
1282 | ext4_calc_metadata_amount(inode, lblock)); | |
1283 | trace_ext4_da_reserve_space(inode, md_needed); | |
1284 | ||
1285 | /* | |
1286 | * We do still charge estimated metadata to the sb though; | |
1287 | * we cannot afford to run out of free blocks. | |
1288 | */ | |
1289 | if (ext4_claim_free_clusters(sbi, md_needed, 0)) { | |
1290 | ei->i_da_metadata_calc_len = save_len; | |
1291 | ei->i_da_metadata_calc_last_lblock = save_last_lblock; | |
1292 | spin_unlock(&ei->i_block_reservation_lock); | |
1293 | if (ext4_should_retry_alloc(inode->i_sb, &retries)) { | |
1294 | cond_resched(); | |
1295 | goto repeat; | |
1296 | } | |
1297 | return -ENOSPC; | |
1298 | } | |
1299 | ei->i_reserved_meta_blocks += md_needed; | |
1300 | spin_unlock(&ei->i_block_reservation_lock); | |
1301 | ||
1302 | return 0; /* success */ | |
1303 | } | |
1304 | ||
9d0be502 | 1305 | /* |
7b415bf6 | 1306 | * Reserve a single cluster located at lblock |
9d0be502 | 1307 | */ |
01f49d0b | 1308 | static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock) |
d2a17637 | 1309 | { |
030ba6bc | 1310 | int retries = 0; |
60e58e0f | 1311 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
0637c6f4 | 1312 | struct ext4_inode_info *ei = EXT4_I(inode); |
7b415bf6 | 1313 | unsigned int md_needed; |
5dd4056d | 1314 | int ret; |
03179fe9 TT |
1315 | ext4_lblk_t save_last_lblock; |
1316 | int save_len; | |
1317 | ||
1318 | /* | |
1319 | * We will charge metadata quota at writeout time; this saves | |
1320 | * us from metadata over-estimation, though we may go over by | |
1321 | * a small amount in the end. Here we just reserve for data. | |
1322 | */ | |
1323 | ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1)); | |
1324 | if (ret) | |
1325 | return ret; | |
d2a17637 MC |
1326 | |
1327 | /* | |
1328 | * recalculate the amount of metadata blocks to reserve | |
1329 | * in order to allocate nrblocks | |
1330 | * worse case is one extent per block | |
1331 | */ | |
030ba6bc | 1332 | repeat: |
0637c6f4 | 1333 | spin_lock(&ei->i_block_reservation_lock); |
03179fe9 TT |
1334 | /* |
1335 | * ext4_calc_metadata_amount() has side effects, which we have | |
1336 | * to be prepared undo if we fail to claim space. | |
1337 | */ | |
1338 | save_len = ei->i_da_metadata_calc_len; | |
1339 | save_last_lblock = ei->i_da_metadata_calc_last_lblock; | |
7b415bf6 AK |
1340 | md_needed = EXT4_NUM_B2C(sbi, |
1341 | ext4_calc_metadata_amount(inode, lblock)); | |
f8ec9d68 | 1342 | trace_ext4_da_reserve_space(inode, md_needed); |
d2a17637 | 1343 | |
72b8ab9d ES |
1344 | /* |
1345 | * We do still charge estimated metadata to the sb though; | |
1346 | * we cannot afford to run out of free blocks. | |
1347 | */ | |
e7d5f315 | 1348 | if (ext4_claim_free_clusters(sbi, md_needed + 1, 0)) { |
03179fe9 TT |
1349 | ei->i_da_metadata_calc_len = save_len; |
1350 | ei->i_da_metadata_calc_last_lblock = save_last_lblock; | |
1351 | spin_unlock(&ei->i_block_reservation_lock); | |
030ba6bc | 1352 | if (ext4_should_retry_alloc(inode->i_sb, &retries)) { |
bb8b20ed | 1353 | cond_resched(); |
030ba6bc AK |
1354 | goto repeat; |
1355 | } | |
03179fe9 | 1356 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1)); |
d2a17637 MC |
1357 | return -ENOSPC; |
1358 | } | |
9d0be502 | 1359 | ei->i_reserved_data_blocks++; |
0637c6f4 TT |
1360 | ei->i_reserved_meta_blocks += md_needed; |
1361 | spin_unlock(&ei->i_block_reservation_lock); | |
39bc680a | 1362 | |
d2a17637 MC |
1363 | return 0; /* success */ |
1364 | } | |
1365 | ||
12219aea | 1366 | static void ext4_da_release_space(struct inode *inode, int to_free) |
d2a17637 MC |
1367 | { |
1368 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0637c6f4 | 1369 | struct ext4_inode_info *ei = EXT4_I(inode); |
d2a17637 | 1370 | |
cd213226 MC |
1371 | if (!to_free) |
1372 | return; /* Nothing to release, exit */ | |
1373 | ||
d2a17637 | 1374 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); |
cd213226 | 1375 | |
5a58ec87 | 1376 | trace_ext4_da_release_space(inode, to_free); |
0637c6f4 | 1377 | if (unlikely(to_free > ei->i_reserved_data_blocks)) { |
cd213226 | 1378 | /* |
0637c6f4 TT |
1379 | * if there aren't enough reserved blocks, then the |
1380 | * counter is messed up somewhere. Since this | |
1381 | * function is called from invalidate page, it's | |
1382 | * harmless to return without any action. | |
cd213226 | 1383 | */ |
8de5c325 | 1384 | ext4_warning(inode->i_sb, "ext4_da_release_space: " |
0637c6f4 | 1385 | "ino %lu, to_free %d with only %d reserved " |
1084f252 | 1386 | "data blocks", inode->i_ino, to_free, |
0637c6f4 TT |
1387 | ei->i_reserved_data_blocks); |
1388 | WARN_ON(1); | |
1389 | to_free = ei->i_reserved_data_blocks; | |
cd213226 | 1390 | } |
0637c6f4 | 1391 | ei->i_reserved_data_blocks -= to_free; |
cd213226 | 1392 | |
0637c6f4 TT |
1393 | if (ei->i_reserved_data_blocks == 0) { |
1394 | /* | |
1395 | * We can release all of the reserved metadata blocks | |
1396 | * only when we have written all of the delayed | |
1397 | * allocation blocks. | |
7b415bf6 AK |
1398 | * Note that in case of bigalloc, i_reserved_meta_blocks, |
1399 | * i_reserved_data_blocks, etc. refer to number of clusters. | |
0637c6f4 | 1400 | */ |
57042651 | 1401 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, |
72b8ab9d | 1402 | ei->i_reserved_meta_blocks); |
ee5f4d9c | 1403 | ei->i_reserved_meta_blocks = 0; |
9d0be502 | 1404 | ei->i_da_metadata_calc_len = 0; |
0637c6f4 | 1405 | } |
d2a17637 | 1406 | |
72b8ab9d | 1407 | /* update fs dirty data blocks counter */ |
57042651 | 1408 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free); |
d2a17637 | 1409 | |
d2a17637 | 1410 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
60e58e0f | 1411 | |
7b415bf6 | 1412 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free)); |
d2a17637 MC |
1413 | } |
1414 | ||
1415 | static void ext4_da_page_release_reservation(struct page *page, | |
ca99fdd2 LC |
1416 | unsigned int offset, |
1417 | unsigned int length) | |
d2a17637 MC |
1418 | { |
1419 | int to_release = 0; | |
1420 | struct buffer_head *head, *bh; | |
1421 | unsigned int curr_off = 0; | |
7b415bf6 AK |
1422 | struct inode *inode = page->mapping->host; |
1423 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
ca99fdd2 | 1424 | unsigned int stop = offset + length; |
7b415bf6 | 1425 | int num_clusters; |
51865fda | 1426 | ext4_fsblk_t lblk; |
d2a17637 | 1427 | |
ca99fdd2 LC |
1428 | BUG_ON(stop > PAGE_CACHE_SIZE || stop < length); |
1429 | ||
d2a17637 MC |
1430 | head = page_buffers(page); |
1431 | bh = head; | |
1432 | do { | |
1433 | unsigned int next_off = curr_off + bh->b_size; | |
1434 | ||
ca99fdd2 LC |
1435 | if (next_off > stop) |
1436 | break; | |
1437 | ||
d2a17637 MC |
1438 | if ((offset <= curr_off) && (buffer_delay(bh))) { |
1439 | to_release++; | |
1440 | clear_buffer_delay(bh); | |
1441 | } | |
1442 | curr_off = next_off; | |
1443 | } while ((bh = bh->b_this_page) != head); | |
7b415bf6 | 1444 | |
51865fda ZL |
1445 | if (to_release) { |
1446 | lblk = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
1447 | ext4_es_remove_extent(inode, lblk, to_release); | |
1448 | } | |
1449 | ||
7b415bf6 AK |
1450 | /* If we have released all the blocks belonging to a cluster, then we |
1451 | * need to release the reserved space for that cluster. */ | |
1452 | num_clusters = EXT4_NUM_B2C(sbi, to_release); | |
1453 | while (num_clusters > 0) { | |
7b415bf6 AK |
1454 | lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) + |
1455 | ((num_clusters - 1) << sbi->s_cluster_bits); | |
1456 | if (sbi->s_cluster_ratio == 1 || | |
7d1b1fbc | 1457 | !ext4_find_delalloc_cluster(inode, lblk)) |
7b415bf6 AK |
1458 | ext4_da_release_space(inode, 1); |
1459 | ||
1460 | num_clusters--; | |
1461 | } | |
d2a17637 | 1462 | } |
ac27a0ec | 1463 | |
64769240 AT |
1464 | /* |
1465 | * Delayed allocation stuff | |
1466 | */ | |
1467 | ||
64769240 AT |
1468 | /* |
1469 | * mpage_da_submit_io - walks through extent of pages and try to write | |
a1d6cc56 | 1470 | * them with writepage() call back |
64769240 AT |
1471 | * |
1472 | * @mpd->inode: inode | |
1473 | * @mpd->first_page: first page of the extent | |
1474 | * @mpd->next_page: page after the last page of the extent | |
64769240 AT |
1475 | * |
1476 | * By the time mpage_da_submit_io() is called we expect all blocks | |
1477 | * to be allocated. this may be wrong if allocation failed. | |
1478 | * | |
1479 | * As pages are already locked by write_cache_pages(), we can't use it | |
1480 | */ | |
1de3e3df TT |
1481 | static int mpage_da_submit_io(struct mpage_da_data *mpd, |
1482 | struct ext4_map_blocks *map) | |
64769240 | 1483 | { |
791b7f08 AK |
1484 | struct pagevec pvec; |
1485 | unsigned long index, end; | |
1486 | int ret = 0, err, nr_pages, i; | |
1487 | struct inode *inode = mpd->inode; | |
1488 | struct address_space *mapping = inode->i_mapping; | |
cb20d518 | 1489 | loff_t size = i_size_read(inode); |
3ecdb3a1 TT |
1490 | unsigned int len, block_start; |
1491 | struct buffer_head *bh, *page_bufs = NULL; | |
1de3e3df | 1492 | sector_t pblock = 0, cur_logical = 0; |
bd2d0210 | 1493 | struct ext4_io_submit io_submit; |
64769240 AT |
1494 | |
1495 | BUG_ON(mpd->next_page <= mpd->first_page); | |
a549984b | 1496 | memset(&io_submit, 0, sizeof(io_submit)); |
791b7f08 AK |
1497 | /* |
1498 | * We need to start from the first_page to the next_page - 1 | |
1499 | * to make sure we also write the mapped dirty buffer_heads. | |
8dc207c0 | 1500 | * If we look at mpd->b_blocknr we would only be looking |
791b7f08 AK |
1501 | * at the currently mapped buffer_heads. |
1502 | */ | |
64769240 AT |
1503 | index = mpd->first_page; |
1504 | end = mpd->next_page - 1; | |
1505 | ||
791b7f08 | 1506 | pagevec_init(&pvec, 0); |
64769240 | 1507 | while (index <= end) { |
791b7f08 | 1508 | nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE); |
64769240 AT |
1509 | if (nr_pages == 0) |
1510 | break; | |
1511 | for (i = 0; i < nr_pages; i++) { | |
f8bec370 | 1512 | int skip_page = 0; |
64769240 AT |
1513 | struct page *page = pvec.pages[i]; |
1514 | ||
791b7f08 AK |
1515 | index = page->index; |
1516 | if (index > end) | |
1517 | break; | |
cb20d518 TT |
1518 | |
1519 | if (index == size >> PAGE_CACHE_SHIFT) | |
1520 | len = size & ~PAGE_CACHE_MASK; | |
1521 | else | |
1522 | len = PAGE_CACHE_SIZE; | |
1de3e3df TT |
1523 | if (map) { |
1524 | cur_logical = index << (PAGE_CACHE_SHIFT - | |
1525 | inode->i_blkbits); | |
1526 | pblock = map->m_pblk + (cur_logical - | |
1527 | map->m_lblk); | |
1528 | } | |
791b7f08 AK |
1529 | index++; |
1530 | ||
1531 | BUG_ON(!PageLocked(page)); | |
1532 | BUG_ON(PageWriteback(page)); | |
1533 | ||
3ecdb3a1 TT |
1534 | bh = page_bufs = page_buffers(page); |
1535 | block_start = 0; | |
64769240 | 1536 | do { |
1de3e3df TT |
1537 | if (map && (cur_logical >= map->m_lblk) && |
1538 | (cur_logical <= (map->m_lblk + | |
1539 | (map->m_len - 1)))) { | |
29fa89d0 AK |
1540 | if (buffer_delay(bh)) { |
1541 | clear_buffer_delay(bh); | |
1542 | bh->b_blocknr = pblock; | |
29fa89d0 | 1543 | } |
1de3e3df TT |
1544 | if (buffer_unwritten(bh) || |
1545 | buffer_mapped(bh)) | |
1546 | BUG_ON(bh->b_blocknr != pblock); | |
1547 | if (map->m_flags & EXT4_MAP_UNINIT) | |
1548 | set_buffer_uninit(bh); | |
1549 | clear_buffer_unwritten(bh); | |
1550 | } | |
29fa89d0 | 1551 | |
13a79a47 YY |
1552 | /* |
1553 | * skip page if block allocation undone and | |
1554 | * block is dirty | |
1555 | */ | |
1556 | if (ext4_bh_delay_or_unwritten(NULL, bh)) | |
97498956 | 1557 | skip_page = 1; |
3ecdb3a1 TT |
1558 | bh = bh->b_this_page; |
1559 | block_start += bh->b_size; | |
64769240 AT |
1560 | cur_logical++; |
1561 | pblock++; | |
1de3e3df TT |
1562 | } while (bh != page_bufs); |
1563 | ||
f8bec370 JK |
1564 | if (skip_page) { |
1565 | unlock_page(page); | |
1566 | continue; | |
1567 | } | |
cb20d518 | 1568 | |
97498956 | 1569 | clear_page_dirty_for_io(page); |
fe089c77 JK |
1570 | err = ext4_bio_write_page(&io_submit, page, len, |
1571 | mpd->wbc); | |
cb20d518 | 1572 | if (!err) |
a1d6cc56 | 1573 | mpd->pages_written++; |
64769240 AT |
1574 | /* |
1575 | * In error case, we have to continue because | |
1576 | * remaining pages are still locked | |
64769240 AT |
1577 | */ |
1578 | if (ret == 0) | |
1579 | ret = err; | |
64769240 AT |
1580 | } |
1581 | pagevec_release(&pvec); | |
1582 | } | |
bd2d0210 | 1583 | ext4_io_submit(&io_submit); |
64769240 | 1584 | return ret; |
64769240 AT |
1585 | } |
1586 | ||
c7f5938a | 1587 | static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd) |
c4a0c46e AK |
1588 | { |
1589 | int nr_pages, i; | |
1590 | pgoff_t index, end; | |
1591 | struct pagevec pvec; | |
1592 | struct inode *inode = mpd->inode; | |
1593 | struct address_space *mapping = inode->i_mapping; | |
51865fda | 1594 | ext4_lblk_t start, last; |
c4a0c46e | 1595 | |
c7f5938a CW |
1596 | index = mpd->first_page; |
1597 | end = mpd->next_page - 1; | |
51865fda ZL |
1598 | |
1599 | start = index << (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
1600 | last = end << (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
1601 | ext4_es_remove_extent(inode, start, last - start + 1); | |
1602 | ||
66bea92c | 1603 | pagevec_init(&pvec, 0); |
c4a0c46e AK |
1604 | while (index <= end) { |
1605 | nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE); | |
1606 | if (nr_pages == 0) | |
1607 | break; | |
1608 | for (i = 0; i < nr_pages; i++) { | |
1609 | struct page *page = pvec.pages[i]; | |
9b1d0998 | 1610 | if (page->index > end) |
c4a0c46e | 1611 | break; |
c4a0c46e AK |
1612 | BUG_ON(!PageLocked(page)); |
1613 | BUG_ON(PageWriteback(page)); | |
d47992f8 | 1614 | block_invalidatepage(page, 0, PAGE_CACHE_SIZE); |
c4a0c46e AK |
1615 | ClearPageUptodate(page); |
1616 | unlock_page(page); | |
1617 | } | |
9b1d0998 JK |
1618 | index = pvec.pages[nr_pages - 1]->index + 1; |
1619 | pagevec_release(&pvec); | |
c4a0c46e AK |
1620 | } |
1621 | return; | |
1622 | } | |
1623 | ||
df22291f AK |
1624 | static void ext4_print_free_blocks(struct inode *inode) |
1625 | { | |
1626 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
92b97816 | 1627 | struct super_block *sb = inode->i_sb; |
f78ee70d | 1628 | struct ext4_inode_info *ei = EXT4_I(inode); |
92b97816 TT |
1629 | |
1630 | ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld", | |
5dee5437 | 1631 | EXT4_C2B(EXT4_SB(inode->i_sb), |
f78ee70d | 1632 | ext4_count_free_clusters(sb))); |
92b97816 TT |
1633 | ext4_msg(sb, KERN_CRIT, "Free/Dirty block details"); |
1634 | ext4_msg(sb, KERN_CRIT, "free_blocks=%lld", | |
f78ee70d | 1635 | (long long) EXT4_C2B(EXT4_SB(sb), |
57042651 | 1636 | percpu_counter_sum(&sbi->s_freeclusters_counter))); |
92b97816 | 1637 | ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld", |
f78ee70d | 1638 | (long long) EXT4_C2B(EXT4_SB(sb), |
7b415bf6 | 1639 | percpu_counter_sum(&sbi->s_dirtyclusters_counter))); |
92b97816 TT |
1640 | ext4_msg(sb, KERN_CRIT, "Block reservation details"); |
1641 | ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u", | |
f78ee70d | 1642 | ei->i_reserved_data_blocks); |
92b97816 | 1643 | ext4_msg(sb, KERN_CRIT, "i_reserved_meta_blocks=%u", |
f78ee70d LC |
1644 | ei->i_reserved_meta_blocks); |
1645 | ext4_msg(sb, KERN_CRIT, "i_allocated_meta_blocks=%u", | |
1646 | ei->i_allocated_meta_blocks); | |
df22291f AK |
1647 | return; |
1648 | } | |
1649 | ||
64769240 | 1650 | /* |
5a87b7a5 TT |
1651 | * mpage_da_map_and_submit - go through given space, map them |
1652 | * if necessary, and then submit them for I/O | |
64769240 | 1653 | * |
8dc207c0 | 1654 | * @mpd - bh describing space |
64769240 AT |
1655 | * |
1656 | * The function skips space we know is already mapped to disk blocks. | |
1657 | * | |
64769240 | 1658 | */ |
5a87b7a5 | 1659 | static void mpage_da_map_and_submit(struct mpage_da_data *mpd) |
64769240 | 1660 | { |
2ac3b6e0 | 1661 | int err, blks, get_blocks_flags; |
1de3e3df | 1662 | struct ext4_map_blocks map, *mapp = NULL; |
2fa3cdfb TT |
1663 | sector_t next = mpd->b_blocknr; |
1664 | unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits; | |
1665 | loff_t disksize = EXT4_I(mpd->inode)->i_disksize; | |
1666 | handle_t *handle = NULL; | |
64769240 AT |
1667 | |
1668 | /* | |
5a87b7a5 TT |
1669 | * If the blocks are mapped already, or we couldn't accumulate |
1670 | * any blocks, then proceed immediately to the submission stage. | |
2fa3cdfb | 1671 | */ |
5a87b7a5 TT |
1672 | if ((mpd->b_size == 0) || |
1673 | ((mpd->b_state & (1 << BH_Mapped)) && | |
1674 | !(mpd->b_state & (1 << BH_Delay)) && | |
1675 | !(mpd->b_state & (1 << BH_Unwritten)))) | |
1676 | goto submit_io; | |
2fa3cdfb TT |
1677 | |
1678 | handle = ext4_journal_current_handle(); | |
1679 | BUG_ON(!handle); | |
1680 | ||
79ffab34 | 1681 | /* |
79e83036 | 1682 | * Call ext4_map_blocks() to allocate any delayed allocation |
2ac3b6e0 TT |
1683 | * blocks, or to convert an uninitialized extent to be |
1684 | * initialized (in the case where we have written into | |
1685 | * one or more preallocated blocks). | |
1686 | * | |
1687 | * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to | |
1688 | * indicate that we are on the delayed allocation path. This | |
1689 | * affects functions in many different parts of the allocation | |
1690 | * call path. This flag exists primarily because we don't | |
79e83036 | 1691 | * want to change *many* call functions, so ext4_map_blocks() |
f2321097 | 1692 | * will set the EXT4_STATE_DELALLOC_RESERVED flag once the |
2ac3b6e0 TT |
1693 | * inode's allocation semaphore is taken. |
1694 | * | |
1695 | * If the blocks in questions were delalloc blocks, set | |
1696 | * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting | |
1697 | * variables are updated after the blocks have been allocated. | |
79ffab34 | 1698 | */ |
2ed88685 TT |
1699 | map.m_lblk = next; |
1700 | map.m_len = max_blocks; | |
27dd4385 LC |
1701 | /* |
1702 | * We're in delalloc path and it is possible that we're going to | |
1703 | * need more metadata blocks than previously reserved. However | |
1704 | * we must not fail because we're in writeback and there is | |
1705 | * nothing we can do about it so it might result in data loss. | |
1706 | * So use reserved blocks to allocate metadata if possible. | |
1707 | */ | |
1708 | get_blocks_flags = EXT4_GET_BLOCKS_CREATE | | |
1709 | EXT4_GET_BLOCKS_METADATA_NOFAIL; | |
744692dc JZ |
1710 | if (ext4_should_dioread_nolock(mpd->inode)) |
1711 | get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT; | |
2ac3b6e0 | 1712 | if (mpd->b_state & (1 << BH_Delay)) |
1296cc85 AK |
1713 | get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE; |
1714 | ||
27dd4385 | 1715 | |
2ed88685 | 1716 | blks = ext4_map_blocks(handle, mpd->inode, &map, get_blocks_flags); |
2fa3cdfb | 1717 | if (blks < 0) { |
e3570639 ES |
1718 | struct super_block *sb = mpd->inode->i_sb; |
1719 | ||
2fa3cdfb | 1720 | err = blks; |
ed5bde0b | 1721 | /* |
5a87b7a5 | 1722 | * If get block returns EAGAIN or ENOSPC and there |
97498956 TT |
1723 | * appears to be free blocks we will just let |
1724 | * mpage_da_submit_io() unlock all of the pages. | |
c4a0c46e AK |
1725 | */ |
1726 | if (err == -EAGAIN) | |
5a87b7a5 | 1727 | goto submit_io; |
df22291f | 1728 | |
5dee5437 | 1729 | if (err == -ENOSPC && ext4_count_free_clusters(sb)) { |
df22291f | 1730 | mpd->retval = err; |
5a87b7a5 | 1731 | goto submit_io; |
df22291f AK |
1732 | } |
1733 | ||
c4a0c46e | 1734 | /* |
ed5bde0b TT |
1735 | * get block failure will cause us to loop in |
1736 | * writepages, because a_ops->writepage won't be able | |
1737 | * to make progress. The page will be redirtied by | |
1738 | * writepage and writepages will again try to write | |
1739 | * the same. | |
c4a0c46e | 1740 | */ |
e3570639 ES |
1741 | if (!(EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)) { |
1742 | ext4_msg(sb, KERN_CRIT, | |
1743 | "delayed block allocation failed for inode %lu " | |
1744 | "at logical offset %llu with max blocks %zd " | |
1745 | "with error %d", mpd->inode->i_ino, | |
1746 | (unsigned long long) next, | |
1747 | mpd->b_size >> mpd->inode->i_blkbits, err); | |
1748 | ext4_msg(sb, KERN_CRIT, | |
01a523eb | 1749 | "This should not happen!! Data will be lost"); |
e3570639 ES |
1750 | if (err == -ENOSPC) |
1751 | ext4_print_free_blocks(mpd->inode); | |
030ba6bc | 1752 | } |
2fa3cdfb | 1753 | /* invalidate all the pages */ |
c7f5938a | 1754 | ext4_da_block_invalidatepages(mpd); |
e0fd9b90 CW |
1755 | |
1756 | /* Mark this page range as having been completed */ | |
1757 | mpd->io_done = 1; | |
5a87b7a5 | 1758 | return; |
c4a0c46e | 1759 | } |
2fa3cdfb TT |
1760 | BUG_ON(blks == 0); |
1761 | ||
1de3e3df | 1762 | mapp = ↦ |
2ed88685 TT |
1763 | if (map.m_flags & EXT4_MAP_NEW) { |
1764 | struct block_device *bdev = mpd->inode->i_sb->s_bdev; | |
1765 | int i; | |
64769240 | 1766 | |
2ed88685 TT |
1767 | for (i = 0; i < map.m_len; i++) |
1768 | unmap_underlying_metadata(bdev, map.m_pblk + i); | |
2fa3cdfb TT |
1769 | } |
1770 | ||
1771 | /* | |
03f5d8bc | 1772 | * Update on-disk size along with block allocation. |
2fa3cdfb TT |
1773 | */ |
1774 | disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits; | |
1775 | if (disksize > i_size_read(mpd->inode)) | |
1776 | disksize = i_size_read(mpd->inode); | |
1777 | if (disksize > EXT4_I(mpd->inode)->i_disksize) { | |
1778 | ext4_update_i_disksize(mpd->inode, disksize); | |
5a87b7a5 TT |
1779 | err = ext4_mark_inode_dirty(handle, mpd->inode); |
1780 | if (err) | |
1781 | ext4_error(mpd->inode->i_sb, | |
1782 | "Failed to mark inode %lu dirty", | |
1783 | mpd->inode->i_ino); | |
2fa3cdfb TT |
1784 | } |
1785 | ||
5a87b7a5 | 1786 | submit_io: |
1de3e3df | 1787 | mpage_da_submit_io(mpd, mapp); |
5a87b7a5 | 1788 | mpd->io_done = 1; |
64769240 AT |
1789 | } |
1790 | ||
bf068ee2 AK |
1791 | #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \ |
1792 | (1 << BH_Delay) | (1 << BH_Unwritten)) | |
64769240 AT |
1793 | |
1794 | /* | |
1795 | * mpage_add_bh_to_extent - try to add one more block to extent of blocks | |
1796 | * | |
1797 | * @mpd->lbh - extent of blocks | |
1798 | * @logical - logical number of the block in the file | |
b6a8e62f | 1799 | * @b_state - b_state of the buffer head added |
64769240 AT |
1800 | * |
1801 | * the function is used to collect contig. blocks in same state | |
1802 | */ | |
b6a8e62f | 1803 | static void mpage_add_bh_to_extent(struct mpage_da_data *mpd, sector_t logical, |
8dc207c0 | 1804 | unsigned long b_state) |
64769240 | 1805 | { |
64769240 | 1806 | sector_t next; |
b6a8e62f JK |
1807 | int blkbits = mpd->inode->i_blkbits; |
1808 | int nrblocks = mpd->b_size >> blkbits; | |
64769240 | 1809 | |
c445e3e0 ES |
1810 | /* |
1811 | * XXX Don't go larger than mballoc is willing to allocate | |
1812 | * This is a stopgap solution. We eventually need to fold | |
1813 | * mpage_da_submit_io() into this function and then call | |
79e83036 | 1814 | * ext4_map_blocks() multiple times in a loop |
c445e3e0 | 1815 | */ |
b6a8e62f | 1816 | if (nrblocks >= (8*1024*1024 >> blkbits)) |
c445e3e0 ES |
1817 | goto flush_it; |
1818 | ||
b6a8e62f JK |
1819 | /* check if the reserved journal credits might overflow */ |
1820 | if (!ext4_test_inode_flag(mpd->inode, EXT4_INODE_EXTENTS)) { | |
525f4ed8 MC |
1821 | if (nrblocks >= EXT4_MAX_TRANS_DATA) { |
1822 | /* | |
1823 | * With non-extent format we are limited by the journal | |
1824 | * credit available. Total credit needed to insert | |
1825 | * nrblocks contiguous blocks is dependent on the | |
1826 | * nrblocks. So limit nrblocks. | |
1827 | */ | |
1828 | goto flush_it; | |
525f4ed8 MC |
1829 | } |
1830 | } | |
64769240 AT |
1831 | /* |
1832 | * First block in the extent | |
1833 | */ | |
8dc207c0 TT |
1834 | if (mpd->b_size == 0) { |
1835 | mpd->b_blocknr = logical; | |
b6a8e62f | 1836 | mpd->b_size = 1 << blkbits; |
8dc207c0 | 1837 | mpd->b_state = b_state & BH_FLAGS; |
64769240 AT |
1838 | return; |
1839 | } | |
1840 | ||
8dc207c0 | 1841 | next = mpd->b_blocknr + nrblocks; |
64769240 AT |
1842 | /* |
1843 | * Can we merge the block to our big extent? | |
1844 | */ | |
8dc207c0 | 1845 | if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) { |
b6a8e62f | 1846 | mpd->b_size += 1 << blkbits; |
64769240 AT |
1847 | return; |
1848 | } | |
1849 | ||
525f4ed8 | 1850 | flush_it: |
64769240 AT |
1851 | /* |
1852 | * We couldn't merge the block to our extent, so we | |
1853 | * need to flush current extent and start new one | |
1854 | */ | |
5a87b7a5 | 1855 | mpage_da_map_and_submit(mpd); |
a1d6cc56 | 1856 | return; |
64769240 AT |
1857 | } |
1858 | ||
c364b22c | 1859 | static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh) |
29fa89d0 | 1860 | { |
c364b22c | 1861 | return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh); |
29fa89d0 AK |
1862 | } |
1863 | ||
5356f261 AK |
1864 | /* |
1865 | * This function is grabs code from the very beginning of | |
1866 | * ext4_map_blocks, but assumes that the caller is from delayed write | |
1867 | * time. This function looks up the requested blocks and sets the | |
1868 | * buffer delay bit under the protection of i_data_sem. | |
1869 | */ | |
1870 | static int ext4_da_map_blocks(struct inode *inode, sector_t iblock, | |
1871 | struct ext4_map_blocks *map, | |
1872 | struct buffer_head *bh) | |
1873 | { | |
d100eef2 | 1874 | struct extent_status es; |
5356f261 AK |
1875 | int retval; |
1876 | sector_t invalid_block = ~((sector_t) 0xffff); | |
921f266b DM |
1877 | #ifdef ES_AGGRESSIVE_TEST |
1878 | struct ext4_map_blocks orig_map; | |
1879 | ||
1880 | memcpy(&orig_map, map, sizeof(*map)); | |
1881 | #endif | |
5356f261 AK |
1882 | |
1883 | if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es)) | |
1884 | invalid_block = ~0; | |
1885 | ||
1886 | map->m_flags = 0; | |
1887 | ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u," | |
1888 | "logical block %lu\n", inode->i_ino, map->m_len, | |
1889 | (unsigned long) map->m_lblk); | |
d100eef2 ZL |
1890 | |
1891 | /* Lookup extent status tree firstly */ | |
1892 | if (ext4_es_lookup_extent(inode, iblock, &es)) { | |
1893 | ||
1894 | if (ext4_es_is_hole(&es)) { | |
1895 | retval = 0; | |
1896 | down_read((&EXT4_I(inode)->i_data_sem)); | |
1897 | goto add_delayed; | |
1898 | } | |
1899 | ||
1900 | /* | |
1901 | * Delayed extent could be allocated by fallocate. | |
1902 | * So we need to check it. | |
1903 | */ | |
1904 | if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) { | |
1905 | map_bh(bh, inode->i_sb, invalid_block); | |
1906 | set_buffer_new(bh); | |
1907 | set_buffer_delay(bh); | |
1908 | return 0; | |
1909 | } | |
1910 | ||
1911 | map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk; | |
1912 | retval = es.es_len - (iblock - es.es_lblk); | |
1913 | if (retval > map->m_len) | |
1914 | retval = map->m_len; | |
1915 | map->m_len = retval; | |
1916 | if (ext4_es_is_written(&es)) | |
1917 | map->m_flags |= EXT4_MAP_MAPPED; | |
1918 | else if (ext4_es_is_unwritten(&es)) | |
1919 | map->m_flags |= EXT4_MAP_UNWRITTEN; | |
1920 | else | |
1921 | BUG_ON(1); | |
1922 | ||
921f266b DM |
1923 | #ifdef ES_AGGRESSIVE_TEST |
1924 | ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0); | |
1925 | #endif | |
d100eef2 ZL |
1926 | return retval; |
1927 | } | |
1928 | ||
5356f261 AK |
1929 | /* |
1930 | * Try to see if we can get the block without requesting a new | |
1931 | * file system block. | |
1932 | */ | |
1933 | down_read((&EXT4_I(inode)->i_data_sem)); | |
9c3569b5 TM |
1934 | if (ext4_has_inline_data(inode)) { |
1935 | /* | |
1936 | * We will soon create blocks for this page, and let | |
1937 | * us pretend as if the blocks aren't allocated yet. | |
1938 | * In case of clusters, we have to handle the work | |
1939 | * of mapping from cluster so that the reserved space | |
1940 | * is calculated properly. | |
1941 | */ | |
1942 | if ((EXT4_SB(inode->i_sb)->s_cluster_ratio > 1) && | |
1943 | ext4_find_delalloc_cluster(inode, map->m_lblk)) | |
1944 | map->m_flags |= EXT4_MAP_FROM_CLUSTER; | |
1945 | retval = 0; | |
1946 | } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
d100eef2 ZL |
1947 | retval = ext4_ext_map_blocks(NULL, inode, map, |
1948 | EXT4_GET_BLOCKS_NO_PUT_HOLE); | |
5356f261 | 1949 | else |
d100eef2 ZL |
1950 | retval = ext4_ind_map_blocks(NULL, inode, map, |
1951 | EXT4_GET_BLOCKS_NO_PUT_HOLE); | |
5356f261 | 1952 | |
d100eef2 | 1953 | add_delayed: |
5356f261 | 1954 | if (retval == 0) { |
f7fec032 | 1955 | int ret; |
5356f261 AK |
1956 | /* |
1957 | * XXX: __block_prepare_write() unmaps passed block, | |
1958 | * is it OK? | |
1959 | */ | |
386ad67c LC |
1960 | /* |
1961 | * If the block was allocated from previously allocated cluster, | |
1962 | * then we don't need to reserve it again. However we still need | |
1963 | * to reserve metadata for every block we're going to write. | |
1964 | */ | |
5356f261 | 1965 | if (!(map->m_flags & EXT4_MAP_FROM_CLUSTER)) { |
f7fec032 ZL |
1966 | ret = ext4_da_reserve_space(inode, iblock); |
1967 | if (ret) { | |
5356f261 | 1968 | /* not enough space to reserve */ |
f7fec032 | 1969 | retval = ret; |
5356f261 | 1970 | goto out_unlock; |
f7fec032 | 1971 | } |
386ad67c LC |
1972 | } else { |
1973 | ret = ext4_da_reserve_metadata(inode, iblock); | |
1974 | if (ret) { | |
1975 | /* not enough space to reserve */ | |
1976 | retval = ret; | |
1977 | goto out_unlock; | |
1978 | } | |
5356f261 AK |
1979 | } |
1980 | ||
f7fec032 ZL |
1981 | ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, |
1982 | ~0, EXTENT_STATUS_DELAYED); | |
1983 | if (ret) { | |
1984 | retval = ret; | |
51865fda | 1985 | goto out_unlock; |
f7fec032 | 1986 | } |
51865fda | 1987 | |
5356f261 AK |
1988 | /* Clear EXT4_MAP_FROM_CLUSTER flag since its purpose is served |
1989 | * and it should not appear on the bh->b_state. | |
1990 | */ | |
1991 | map->m_flags &= ~EXT4_MAP_FROM_CLUSTER; | |
1992 | ||
1993 | map_bh(bh, inode->i_sb, invalid_block); | |
1994 | set_buffer_new(bh); | |
1995 | set_buffer_delay(bh); | |
f7fec032 ZL |
1996 | } else if (retval > 0) { |
1997 | int ret; | |
1998 | unsigned long long status; | |
1999 | ||
921f266b DM |
2000 | #ifdef ES_AGGRESSIVE_TEST |
2001 | if (retval != map->m_len) { | |
2002 | printk("ES len assertation failed for inode: %lu " | |
2003 | "retval %d != map->m_len %d " | |
2004 | "in %s (lookup)\n", inode->i_ino, retval, | |
2005 | map->m_len, __func__); | |
2006 | } | |
2007 | #endif | |
2008 | ||
f7fec032 ZL |
2009 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
2010 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; | |
2011 | ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, | |
2012 | map->m_pblk, status); | |
2013 | if (ret != 0) | |
2014 | retval = ret; | |
5356f261 AK |
2015 | } |
2016 | ||
2017 | out_unlock: | |
2018 | up_read((&EXT4_I(inode)->i_data_sem)); | |
2019 | ||
2020 | return retval; | |
2021 | } | |
2022 | ||
64769240 | 2023 | /* |
b920c755 TT |
2024 | * This is a special get_blocks_t callback which is used by |
2025 | * ext4_da_write_begin(). It will either return mapped block or | |
2026 | * reserve space for a single block. | |
29fa89d0 AK |
2027 | * |
2028 | * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set. | |
2029 | * We also have b_blocknr = -1 and b_bdev initialized properly | |
2030 | * | |
2031 | * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set. | |
2032 | * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev | |
2033 | * initialized properly. | |
64769240 | 2034 | */ |
9c3569b5 TM |
2035 | int ext4_da_get_block_prep(struct inode *inode, sector_t iblock, |
2036 | struct buffer_head *bh, int create) | |
64769240 | 2037 | { |
2ed88685 | 2038 | struct ext4_map_blocks map; |
64769240 AT |
2039 | int ret = 0; |
2040 | ||
2041 | BUG_ON(create == 0); | |
2ed88685 TT |
2042 | BUG_ON(bh->b_size != inode->i_sb->s_blocksize); |
2043 | ||
2044 | map.m_lblk = iblock; | |
2045 | map.m_len = 1; | |
64769240 AT |
2046 | |
2047 | /* | |
2048 | * first, we need to know whether the block is allocated already | |
2049 | * preallocated blocks are unmapped but should treated | |
2050 | * the same as allocated blocks. | |
2051 | */ | |
5356f261 AK |
2052 | ret = ext4_da_map_blocks(inode, iblock, &map, bh); |
2053 | if (ret <= 0) | |
2ed88685 | 2054 | return ret; |
64769240 | 2055 | |
2ed88685 TT |
2056 | map_bh(bh, inode->i_sb, map.m_pblk); |
2057 | bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags; | |
2058 | ||
2059 | if (buffer_unwritten(bh)) { | |
2060 | /* A delayed write to unwritten bh should be marked | |
2061 | * new and mapped. Mapped ensures that we don't do | |
2062 | * get_block multiple times when we write to the same | |
2063 | * offset and new ensures that we do proper zero out | |
2064 | * for partial write. | |
2065 | */ | |
2066 | set_buffer_new(bh); | |
c8205636 | 2067 | set_buffer_mapped(bh); |
2ed88685 TT |
2068 | } |
2069 | return 0; | |
64769240 | 2070 | } |
61628a3f | 2071 | |
62e086be AK |
2072 | static int bget_one(handle_t *handle, struct buffer_head *bh) |
2073 | { | |
2074 | get_bh(bh); | |
2075 | return 0; | |
2076 | } | |
2077 | ||
2078 | static int bput_one(handle_t *handle, struct buffer_head *bh) | |
2079 | { | |
2080 | put_bh(bh); | |
2081 | return 0; | |
2082 | } | |
2083 | ||
2084 | static int __ext4_journalled_writepage(struct page *page, | |
62e086be AK |
2085 | unsigned int len) |
2086 | { | |
2087 | struct address_space *mapping = page->mapping; | |
2088 | struct inode *inode = mapping->host; | |
3fdcfb66 | 2089 | struct buffer_head *page_bufs = NULL; |
62e086be | 2090 | handle_t *handle = NULL; |
3fdcfb66 TM |
2091 | int ret = 0, err = 0; |
2092 | int inline_data = ext4_has_inline_data(inode); | |
2093 | struct buffer_head *inode_bh = NULL; | |
62e086be | 2094 | |
cb20d518 | 2095 | ClearPageChecked(page); |
3fdcfb66 TM |
2096 | |
2097 | if (inline_data) { | |
2098 | BUG_ON(page->index != 0); | |
2099 | BUG_ON(len > ext4_get_max_inline_size(inode)); | |
2100 | inode_bh = ext4_journalled_write_inline_data(inode, len, page); | |
2101 | if (inode_bh == NULL) | |
2102 | goto out; | |
2103 | } else { | |
2104 | page_bufs = page_buffers(page); | |
2105 | if (!page_bufs) { | |
2106 | BUG(); | |
2107 | goto out; | |
2108 | } | |
2109 | ext4_walk_page_buffers(handle, page_bufs, 0, len, | |
2110 | NULL, bget_one); | |
2111 | } | |
62e086be AK |
2112 | /* As soon as we unlock the page, it can go away, but we have |
2113 | * references to buffers so we are safe */ | |
2114 | unlock_page(page); | |
2115 | ||
9924a92a TT |
2116 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
2117 | ext4_writepage_trans_blocks(inode)); | |
62e086be AK |
2118 | if (IS_ERR(handle)) { |
2119 | ret = PTR_ERR(handle); | |
2120 | goto out; | |
2121 | } | |
2122 | ||
441c8508 CW |
2123 | BUG_ON(!ext4_handle_valid(handle)); |
2124 | ||
3fdcfb66 TM |
2125 | if (inline_data) { |
2126 | ret = ext4_journal_get_write_access(handle, inode_bh); | |
62e086be | 2127 | |
3fdcfb66 TM |
2128 | err = ext4_handle_dirty_metadata(handle, inode, inode_bh); |
2129 | ||
2130 | } else { | |
2131 | ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL, | |
2132 | do_journal_get_write_access); | |
2133 | ||
2134 | err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL, | |
2135 | write_end_fn); | |
2136 | } | |
62e086be AK |
2137 | if (ret == 0) |
2138 | ret = err; | |
2d859db3 | 2139 | EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; |
62e086be AK |
2140 | err = ext4_journal_stop(handle); |
2141 | if (!ret) | |
2142 | ret = err; | |
2143 | ||
3fdcfb66 TM |
2144 | if (!ext4_has_inline_data(inode)) |
2145 | ext4_walk_page_buffers(handle, page_bufs, 0, len, | |
2146 | NULL, bput_one); | |
19f5fb7a | 2147 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
62e086be | 2148 | out: |
3fdcfb66 | 2149 | brelse(inode_bh); |
62e086be AK |
2150 | return ret; |
2151 | } | |
2152 | ||
61628a3f | 2153 | /* |
43ce1d23 AK |
2154 | * Note that we don't need to start a transaction unless we're journaling data |
2155 | * because we should have holes filled from ext4_page_mkwrite(). We even don't | |
2156 | * need to file the inode to the transaction's list in ordered mode because if | |
2157 | * we are writing back data added by write(), the inode is already there and if | |
25985edc | 2158 | * we are writing back data modified via mmap(), no one guarantees in which |
43ce1d23 AK |
2159 | * transaction the data will hit the disk. In case we are journaling data, we |
2160 | * cannot start transaction directly because transaction start ranks above page | |
2161 | * lock so we have to do some magic. | |
2162 | * | |
b920c755 TT |
2163 | * This function can get called via... |
2164 | * - ext4_da_writepages after taking page lock (have journal handle) | |
2165 | * - journal_submit_inode_data_buffers (no journal handle) | |
f6463b0d | 2166 | * - shrink_page_list via the kswapd/direct reclaim (no journal handle) |
b920c755 | 2167 | * - grab_page_cache when doing write_begin (have journal handle) |
43ce1d23 AK |
2168 | * |
2169 | * We don't do any block allocation in this function. If we have page with | |
2170 | * multiple blocks we need to write those buffer_heads that are mapped. This | |
2171 | * is important for mmaped based write. So if we do with blocksize 1K | |
2172 | * truncate(f, 1024); | |
2173 | * a = mmap(f, 0, 4096); | |
2174 | * a[0] = 'a'; | |
2175 | * truncate(f, 4096); | |
2176 | * we have in the page first buffer_head mapped via page_mkwrite call back | |
90802ed9 | 2177 | * but other buffer_heads would be unmapped but dirty (dirty done via the |
43ce1d23 AK |
2178 | * do_wp_page). So writepage should write the first block. If we modify |
2179 | * the mmap area beyond 1024 we will again get a page_fault and the | |
2180 | * page_mkwrite callback will do the block allocation and mark the | |
2181 | * buffer_heads mapped. | |
2182 | * | |
2183 | * We redirty the page if we have any buffer_heads that is either delay or | |
2184 | * unwritten in the page. | |
2185 | * | |
2186 | * We can get recursively called as show below. | |
2187 | * | |
2188 | * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() -> | |
2189 | * ext4_writepage() | |
2190 | * | |
2191 | * But since we don't do any block allocation we should not deadlock. | |
2192 | * Page also have the dirty flag cleared so we don't get recurive page_lock. | |
61628a3f | 2193 | */ |
43ce1d23 | 2194 | static int ext4_writepage(struct page *page, |
62e086be | 2195 | struct writeback_control *wbc) |
64769240 | 2196 | { |
f8bec370 | 2197 | int ret = 0; |
61628a3f | 2198 | loff_t size; |
498e5f24 | 2199 | unsigned int len; |
744692dc | 2200 | struct buffer_head *page_bufs = NULL; |
61628a3f | 2201 | struct inode *inode = page->mapping->host; |
36ade451 | 2202 | struct ext4_io_submit io_submit; |
61628a3f | 2203 | |
a9c667f8 | 2204 | trace_ext4_writepage(page); |
f0e6c985 AK |
2205 | size = i_size_read(inode); |
2206 | if (page->index == size >> PAGE_CACHE_SHIFT) | |
2207 | len = size & ~PAGE_CACHE_MASK; | |
2208 | else | |
2209 | len = PAGE_CACHE_SIZE; | |
64769240 | 2210 | |
a42afc5f | 2211 | page_bufs = page_buffers(page); |
a42afc5f | 2212 | /* |
fe386132 JK |
2213 | * We cannot do block allocation or other extent handling in this |
2214 | * function. If there are buffers needing that, we have to redirty | |
2215 | * the page. But we may reach here when we do a journal commit via | |
2216 | * journal_submit_inode_data_buffers() and in that case we must write | |
2217 | * allocated buffers to achieve data=ordered mode guarantees. | |
a42afc5f | 2218 | */ |
f19d5870 TM |
2219 | if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL, |
2220 | ext4_bh_delay_or_unwritten)) { | |
f8bec370 | 2221 | redirty_page_for_writepage(wbc, page); |
fe386132 JK |
2222 | if (current->flags & PF_MEMALLOC) { |
2223 | /* | |
2224 | * For memory cleaning there's no point in writing only | |
2225 | * some buffers. So just bail out. Warn if we came here | |
2226 | * from direct reclaim. | |
2227 | */ | |
2228 | WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) | |
2229 | == PF_MEMALLOC); | |
f0e6c985 AK |
2230 | unlock_page(page); |
2231 | return 0; | |
2232 | } | |
a42afc5f | 2233 | } |
64769240 | 2234 | |
cb20d518 | 2235 | if (PageChecked(page) && ext4_should_journal_data(inode)) |
43ce1d23 AK |
2236 | /* |
2237 | * It's mmapped pagecache. Add buffers and journal it. There | |
2238 | * doesn't seem much point in redirtying the page here. | |
2239 | */ | |
3f0ca309 | 2240 | return __ext4_journalled_writepage(page, len); |
43ce1d23 | 2241 | |
a549984b | 2242 | memset(&io_submit, 0, sizeof(io_submit)); |
36ade451 JK |
2243 | ret = ext4_bio_write_page(&io_submit, page, len, wbc); |
2244 | ext4_io_submit(&io_submit); | |
64769240 AT |
2245 | return ret; |
2246 | } | |
2247 | ||
61628a3f | 2248 | /* |
525f4ed8 | 2249 | * This is called via ext4_da_writepages() to |
25985edc | 2250 | * calculate the total number of credits to reserve to fit |
525f4ed8 MC |
2251 | * a single extent allocation into a single transaction, |
2252 | * ext4_da_writpeages() will loop calling this before | |
2253 | * the block allocation. | |
61628a3f | 2254 | */ |
525f4ed8 MC |
2255 | |
2256 | static int ext4_da_writepages_trans_blocks(struct inode *inode) | |
2257 | { | |
2258 | int max_blocks = EXT4_I(inode)->i_reserved_data_blocks; | |
2259 | ||
2260 | /* | |
2261 | * With non-extent format the journal credit needed to | |
2262 | * insert nrblocks contiguous block is dependent on | |
2263 | * number of contiguous block. So we will limit | |
2264 | * number of contiguous block to a sane value | |
2265 | */ | |
12e9b892 | 2266 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) && |
525f4ed8 MC |
2267 | (max_blocks > EXT4_MAX_TRANS_DATA)) |
2268 | max_blocks = EXT4_MAX_TRANS_DATA; | |
2269 | ||
2270 | return ext4_chunk_trans_blocks(inode, max_blocks); | |
2271 | } | |
61628a3f | 2272 | |
8e48dcfb TT |
2273 | /* |
2274 | * write_cache_pages_da - walk the list of dirty pages of the given | |
8eb9e5ce | 2275 | * address space and accumulate pages that need writing, and call |
168fc022 TT |
2276 | * mpage_da_map_and_submit to map a single contiguous memory region |
2277 | * and then write them. | |
8e48dcfb | 2278 | */ |
9c3569b5 TM |
2279 | static int write_cache_pages_da(handle_t *handle, |
2280 | struct address_space *mapping, | |
8e48dcfb | 2281 | struct writeback_control *wbc, |
72f84e65 ES |
2282 | struct mpage_da_data *mpd, |
2283 | pgoff_t *done_index) | |
8e48dcfb | 2284 | { |
4f01b02c | 2285 | struct buffer_head *bh, *head; |
168fc022 | 2286 | struct inode *inode = mapping->host; |
4f01b02c TT |
2287 | struct pagevec pvec; |
2288 | unsigned int nr_pages; | |
2289 | sector_t logical; | |
2290 | pgoff_t index, end; | |
2291 | long nr_to_write = wbc->nr_to_write; | |
2292 | int i, tag, ret = 0; | |
8e48dcfb | 2293 | |
168fc022 TT |
2294 | memset(mpd, 0, sizeof(struct mpage_da_data)); |
2295 | mpd->wbc = wbc; | |
2296 | mpd->inode = inode; | |
8e48dcfb TT |
2297 | pagevec_init(&pvec, 0); |
2298 | index = wbc->range_start >> PAGE_CACHE_SHIFT; | |
2299 | end = wbc->range_end >> PAGE_CACHE_SHIFT; | |
2300 | ||
6e6938b6 | 2301 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
5b41d924 ES |
2302 | tag = PAGECACHE_TAG_TOWRITE; |
2303 | else | |
2304 | tag = PAGECACHE_TAG_DIRTY; | |
2305 | ||
72f84e65 | 2306 | *done_index = index; |
4f01b02c | 2307 | while (index <= end) { |
5b41d924 | 2308 | nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag, |
8e48dcfb TT |
2309 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1); |
2310 | if (nr_pages == 0) | |
4f01b02c | 2311 | return 0; |
8e48dcfb TT |
2312 | |
2313 | for (i = 0; i < nr_pages; i++) { | |
2314 | struct page *page = pvec.pages[i]; | |
2315 | ||
2316 | /* | |
2317 | * At this point, the page may be truncated or | |
2318 | * invalidated (changing page->mapping to NULL), or | |
2319 | * even swizzled back from swapper_space to tmpfs file | |
2320 | * mapping. However, page->index will not change | |
2321 | * because we have a reference on the page. | |
2322 | */ | |
4f01b02c TT |
2323 | if (page->index > end) |
2324 | goto out; | |
8e48dcfb | 2325 | |
72f84e65 ES |
2326 | *done_index = page->index + 1; |
2327 | ||
78aaced3 TT |
2328 | /* |
2329 | * If we can't merge this page, and we have | |
2330 | * accumulated an contiguous region, write it | |
2331 | */ | |
2332 | if ((mpd->next_page != page->index) && | |
2333 | (mpd->next_page != mpd->first_page)) { | |
2334 | mpage_da_map_and_submit(mpd); | |
2335 | goto ret_extent_tail; | |
2336 | } | |
2337 | ||
8e48dcfb TT |
2338 | lock_page(page); |
2339 | ||
2340 | /* | |
4f01b02c TT |
2341 | * If the page is no longer dirty, or its |
2342 | * mapping no longer corresponds to inode we | |
2343 | * are writing (which means it has been | |
2344 | * truncated or invalidated), or the page is | |
2345 | * already under writeback and we are not | |
2346 | * doing a data integrity writeback, skip the page | |
8e48dcfb | 2347 | */ |
4f01b02c TT |
2348 | if (!PageDirty(page) || |
2349 | (PageWriteback(page) && | |
2350 | (wbc->sync_mode == WB_SYNC_NONE)) || | |
2351 | unlikely(page->mapping != mapping)) { | |
8e48dcfb TT |
2352 | unlock_page(page); |
2353 | continue; | |
2354 | } | |
2355 | ||
7cb1a535 | 2356 | wait_on_page_writeback(page); |
8e48dcfb | 2357 | BUG_ON(PageWriteback(page)); |
8e48dcfb | 2358 | |
9c3569b5 TM |
2359 | /* |
2360 | * If we have inline data and arrive here, it means that | |
2361 | * we will soon create the block for the 1st page, so | |
2362 | * we'd better clear the inline data here. | |
2363 | */ | |
2364 | if (ext4_has_inline_data(inode)) { | |
2365 | BUG_ON(ext4_test_inode_state(inode, | |
2366 | EXT4_STATE_MAY_INLINE_DATA)); | |
2367 | ext4_destroy_inline_data(handle, inode); | |
2368 | } | |
2369 | ||
168fc022 | 2370 | if (mpd->next_page != page->index) |
8eb9e5ce | 2371 | mpd->first_page = page->index; |
8eb9e5ce TT |
2372 | mpd->next_page = page->index + 1; |
2373 | logical = (sector_t) page->index << | |
2374 | (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
2375 | ||
f8bec370 JK |
2376 | /* Add all dirty buffers to mpd */ |
2377 | head = page_buffers(page); | |
2378 | bh = head; | |
2379 | do { | |
2380 | BUG_ON(buffer_locked(bh)); | |
8eb9e5ce | 2381 | /* |
f8bec370 JK |
2382 | * We need to try to allocate unmapped blocks |
2383 | * in the same page. Otherwise we won't make | |
2384 | * progress with the page in ext4_writepage | |
8eb9e5ce | 2385 | */ |
f8bec370 JK |
2386 | if (ext4_bh_delay_or_unwritten(NULL, bh)) { |
2387 | mpage_add_bh_to_extent(mpd, logical, | |
f8bec370 JK |
2388 | bh->b_state); |
2389 | if (mpd->io_done) | |
2390 | goto ret_extent_tail; | |
2391 | } else if (buffer_dirty(bh) && | |
2392 | buffer_mapped(bh)) { | |
8eb9e5ce | 2393 | /* |
f8bec370 JK |
2394 | * mapped dirty buffer. We need to |
2395 | * update the b_state because we look | |
2396 | * at b_state in mpage_da_map_blocks. | |
2397 | * We don't update b_size because if we | |
2398 | * find an unmapped buffer_head later | |
2399 | * we need to use the b_state flag of | |
2400 | * that buffer_head. | |
8eb9e5ce | 2401 | */ |
f8bec370 JK |
2402 | if (mpd->b_size == 0) |
2403 | mpd->b_state = | |
2404 | bh->b_state & BH_FLAGS; | |
2405 | } | |
2406 | logical++; | |
2407 | } while ((bh = bh->b_this_page) != head); | |
8e48dcfb TT |
2408 | |
2409 | if (nr_to_write > 0) { | |
2410 | nr_to_write--; | |
2411 | if (nr_to_write == 0 && | |
4f01b02c | 2412 | wbc->sync_mode == WB_SYNC_NONE) |
8e48dcfb TT |
2413 | /* |
2414 | * We stop writing back only if we are | |
2415 | * not doing integrity sync. In case of | |
2416 | * integrity sync we have to keep going | |
2417 | * because someone may be concurrently | |
2418 | * dirtying pages, and we might have | |
2419 | * synced a lot of newly appeared dirty | |
2420 | * pages, but have not synced all of the | |
2421 | * old dirty pages. | |
2422 | */ | |
4f01b02c | 2423 | goto out; |
8e48dcfb TT |
2424 | } |
2425 | } | |
2426 | pagevec_release(&pvec); | |
2427 | cond_resched(); | |
2428 | } | |
4f01b02c TT |
2429 | return 0; |
2430 | ret_extent_tail: | |
2431 | ret = MPAGE_DA_EXTENT_TAIL; | |
8eb9e5ce TT |
2432 | out: |
2433 | pagevec_release(&pvec); | |
2434 | cond_resched(); | |
8e48dcfb TT |
2435 | return ret; |
2436 | } | |
2437 | ||
2438 | ||
64769240 | 2439 | static int ext4_da_writepages(struct address_space *mapping, |
a1d6cc56 | 2440 | struct writeback_control *wbc) |
64769240 | 2441 | { |
22208ded AK |
2442 | pgoff_t index; |
2443 | int range_whole = 0; | |
61628a3f | 2444 | handle_t *handle = NULL; |
df22291f | 2445 | struct mpage_da_data mpd; |
5e745b04 | 2446 | struct inode *inode = mapping->host; |
498e5f24 | 2447 | int pages_written = 0; |
55138e0b | 2448 | unsigned int max_pages; |
2acf2c26 | 2449 | int range_cyclic, cycled = 1, io_done = 0; |
55138e0b TT |
2450 | int needed_blocks, ret = 0; |
2451 | long desired_nr_to_write, nr_to_writebump = 0; | |
de89de6e | 2452 | loff_t range_start = wbc->range_start; |
5e745b04 | 2453 | struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb); |
72f84e65 | 2454 | pgoff_t done_index = 0; |
5b41d924 | 2455 | pgoff_t end; |
1bce63d1 | 2456 | struct blk_plug plug; |
61628a3f | 2457 | |
9bffad1e | 2458 | trace_ext4_da_writepages(inode, wbc); |
ba80b101 | 2459 | |
61628a3f MC |
2460 | /* |
2461 | * No pages to write? This is mainly a kludge to avoid starting | |
2462 | * a transaction for special inodes like journal inode on last iput() | |
2463 | * because that could violate lock ordering on umount | |
2464 | */ | |
a1d6cc56 | 2465 | if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) |
61628a3f | 2466 | return 0; |
2a21e37e TT |
2467 | |
2468 | /* | |
2469 | * If the filesystem has aborted, it is read-only, so return | |
2470 | * right away instead of dumping stack traces later on that | |
2471 | * will obscure the real source of the problem. We test | |
4ab2f15b | 2472 | * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because |
2a21e37e TT |
2473 | * the latter could be true if the filesystem is mounted |
2474 | * read-only, and in that case, ext4_da_writepages should | |
2475 | * *never* be called, so if that ever happens, we would want | |
2476 | * the stack trace. | |
2477 | */ | |
4ab2f15b | 2478 | if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) |
2a21e37e TT |
2479 | return -EROFS; |
2480 | ||
22208ded AK |
2481 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
2482 | range_whole = 1; | |
61628a3f | 2483 | |
2acf2c26 AK |
2484 | range_cyclic = wbc->range_cyclic; |
2485 | if (wbc->range_cyclic) { | |
22208ded | 2486 | index = mapping->writeback_index; |
2acf2c26 AK |
2487 | if (index) |
2488 | cycled = 0; | |
2489 | wbc->range_start = index << PAGE_CACHE_SHIFT; | |
2490 | wbc->range_end = LLONG_MAX; | |
2491 | wbc->range_cyclic = 0; | |
5b41d924 ES |
2492 | end = -1; |
2493 | } else { | |
22208ded | 2494 | index = wbc->range_start >> PAGE_CACHE_SHIFT; |
5b41d924 ES |
2495 | end = wbc->range_end >> PAGE_CACHE_SHIFT; |
2496 | } | |
a1d6cc56 | 2497 | |
55138e0b TT |
2498 | /* |
2499 | * This works around two forms of stupidity. The first is in | |
2500 | * the writeback code, which caps the maximum number of pages | |
2501 | * written to be 1024 pages. This is wrong on multiple | |
2502 | * levels; different architectues have a different page size, | |
2503 | * which changes the maximum amount of data which gets | |
2504 | * written. Secondly, 4 megabytes is way too small. XFS | |
2505 | * forces this value to be 16 megabytes by multiplying | |
2506 | * nr_to_write parameter by four, and then relies on its | |
2507 | * allocator to allocate larger extents to make them | |
2508 | * contiguous. Unfortunately this brings us to the second | |
2509 | * stupidity, which is that ext4's mballoc code only allocates | |
2510 | * at most 2048 blocks. So we force contiguous writes up to | |
2511 | * the number of dirty blocks in the inode, or | |
2512 | * sbi->max_writeback_mb_bump whichever is smaller. | |
2513 | */ | |
2514 | max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT); | |
b443e733 ES |
2515 | if (!range_cyclic && range_whole) { |
2516 | if (wbc->nr_to_write == LONG_MAX) | |
2517 | desired_nr_to_write = wbc->nr_to_write; | |
2518 | else | |
2519 | desired_nr_to_write = wbc->nr_to_write * 8; | |
2520 | } else | |
55138e0b TT |
2521 | desired_nr_to_write = ext4_num_dirty_pages(inode, index, |
2522 | max_pages); | |
2523 | if (desired_nr_to_write > max_pages) | |
2524 | desired_nr_to_write = max_pages; | |
2525 | ||
2526 | if (wbc->nr_to_write < desired_nr_to_write) { | |
2527 | nr_to_writebump = desired_nr_to_write - wbc->nr_to_write; | |
2528 | wbc->nr_to_write = desired_nr_to_write; | |
2529 | } | |
2530 | ||
2acf2c26 | 2531 | retry: |
6e6938b6 | 2532 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
5b41d924 ES |
2533 | tag_pages_for_writeback(mapping, index, end); |
2534 | ||
1bce63d1 | 2535 | blk_start_plug(&plug); |
22208ded | 2536 | while (!ret && wbc->nr_to_write > 0) { |
a1d6cc56 AK |
2537 | |
2538 | /* | |
2539 | * we insert one extent at a time. So we need | |
2540 | * credit needed for single extent allocation. | |
2541 | * journalled mode is currently not supported | |
2542 | * by delalloc | |
2543 | */ | |
2544 | BUG_ON(ext4_should_journal_data(inode)); | |
525f4ed8 | 2545 | needed_blocks = ext4_da_writepages_trans_blocks(inode); |
a1d6cc56 | 2546 | |
61628a3f | 2547 | /* start a new transaction*/ |
9924a92a TT |
2548 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
2549 | needed_blocks); | |
61628a3f MC |
2550 | if (IS_ERR(handle)) { |
2551 | ret = PTR_ERR(handle); | |
1693918e | 2552 | ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: " |
fbe845dd | 2553 | "%ld pages, ino %lu; err %d", __func__, |
a1d6cc56 | 2554 | wbc->nr_to_write, inode->i_ino, ret); |
3c1fcb2c | 2555 | blk_finish_plug(&plug); |
61628a3f MC |
2556 | goto out_writepages; |
2557 | } | |
f63e6005 TT |
2558 | |
2559 | /* | |
8eb9e5ce | 2560 | * Now call write_cache_pages_da() to find the next |
f63e6005 | 2561 | * contiguous region of logical blocks that need |
8eb9e5ce | 2562 | * blocks to be allocated by ext4 and submit them. |
f63e6005 | 2563 | */ |
9c3569b5 TM |
2564 | ret = write_cache_pages_da(handle, mapping, |
2565 | wbc, &mpd, &done_index); | |
f63e6005 | 2566 | /* |
af901ca1 | 2567 | * If we have a contiguous extent of pages and we |
f63e6005 TT |
2568 | * haven't done the I/O yet, map the blocks and submit |
2569 | * them for I/O. | |
2570 | */ | |
2571 | if (!mpd.io_done && mpd.next_page != mpd.first_page) { | |
5a87b7a5 | 2572 | mpage_da_map_and_submit(&mpd); |
f63e6005 TT |
2573 | ret = MPAGE_DA_EXTENT_TAIL; |
2574 | } | |
b3a3ca8c | 2575 | trace_ext4_da_write_pages(inode, &mpd); |
f63e6005 | 2576 | wbc->nr_to_write -= mpd.pages_written; |
df22291f | 2577 | |
61628a3f | 2578 | ext4_journal_stop(handle); |
df22291f | 2579 | |
8f64b32e | 2580 | if ((mpd.retval == -ENOSPC) && sbi->s_journal) { |
22208ded AK |
2581 | /* commit the transaction which would |
2582 | * free blocks released in the transaction | |
2583 | * and try again | |
2584 | */ | |
df22291f | 2585 | jbd2_journal_force_commit_nested(sbi->s_journal); |
22208ded AK |
2586 | ret = 0; |
2587 | } else if (ret == MPAGE_DA_EXTENT_TAIL) { | |
a1d6cc56 | 2588 | /* |
8de49e67 KM |
2589 | * Got one extent now try with rest of the pages. |
2590 | * If mpd.retval is set -EIO, journal is aborted. | |
2591 | * So we don't need to write any more. | |
a1d6cc56 | 2592 | */ |
22208ded | 2593 | pages_written += mpd.pages_written; |
8de49e67 | 2594 | ret = mpd.retval; |
2acf2c26 | 2595 | io_done = 1; |
22208ded | 2596 | } else if (wbc->nr_to_write) |
61628a3f MC |
2597 | /* |
2598 | * There is no more writeout needed | |
2599 | * or we requested for a noblocking writeout | |
2600 | * and we found the device congested | |
2601 | */ | |
61628a3f | 2602 | break; |
a1d6cc56 | 2603 | } |
1bce63d1 | 2604 | blk_finish_plug(&plug); |
2acf2c26 AK |
2605 | if (!io_done && !cycled) { |
2606 | cycled = 1; | |
2607 | index = 0; | |
2608 | wbc->range_start = index << PAGE_CACHE_SHIFT; | |
2609 | wbc->range_end = mapping->writeback_index - 1; | |
2610 | goto retry; | |
2611 | } | |
22208ded AK |
2612 | |
2613 | /* Update index */ | |
2acf2c26 | 2614 | wbc->range_cyclic = range_cyclic; |
22208ded AK |
2615 | if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) |
2616 | /* | |
2617 | * set the writeback_index so that range_cyclic | |
2618 | * mode will write it back later | |
2619 | */ | |
72f84e65 | 2620 | mapping->writeback_index = done_index; |
a1d6cc56 | 2621 | |
61628a3f | 2622 | out_writepages: |
2faf2e19 | 2623 | wbc->nr_to_write -= nr_to_writebump; |
de89de6e | 2624 | wbc->range_start = range_start; |
9bffad1e | 2625 | trace_ext4_da_writepages_result(inode, wbc, ret, pages_written); |
61628a3f | 2626 | return ret; |
64769240 AT |
2627 | } |
2628 | ||
79f0be8d AK |
2629 | static int ext4_nonda_switch(struct super_block *sb) |
2630 | { | |
5c1ff336 | 2631 | s64 free_clusters, dirty_clusters; |
79f0be8d AK |
2632 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
2633 | ||
2634 | /* | |
2635 | * switch to non delalloc mode if we are running low | |
2636 | * on free block. The free block accounting via percpu | |
179f7ebf | 2637 | * counters can get slightly wrong with percpu_counter_batch getting |
79f0be8d AK |
2638 | * accumulated on each CPU without updating global counters |
2639 | * Delalloc need an accurate free block accounting. So switch | |
2640 | * to non delalloc when we are near to error range. | |
2641 | */ | |
5c1ff336 EW |
2642 | free_clusters = |
2643 | percpu_counter_read_positive(&sbi->s_freeclusters_counter); | |
2644 | dirty_clusters = | |
2645 | percpu_counter_read_positive(&sbi->s_dirtyclusters_counter); | |
00d4e736 TT |
2646 | /* |
2647 | * Start pushing delalloc when 1/2 of free blocks are dirty. | |
2648 | */ | |
5c1ff336 | 2649 | if (dirty_clusters && (free_clusters < 2 * dirty_clusters)) |
10ee27a0 | 2650 | try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE); |
00d4e736 | 2651 | |
5c1ff336 EW |
2652 | if (2 * free_clusters < 3 * dirty_clusters || |
2653 | free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) { | |
79f0be8d | 2654 | /* |
c8afb446 ES |
2655 | * free block count is less than 150% of dirty blocks |
2656 | * or free blocks is less than watermark | |
79f0be8d AK |
2657 | */ |
2658 | return 1; | |
2659 | } | |
2660 | return 0; | |
2661 | } | |
2662 | ||
64769240 | 2663 | static int ext4_da_write_begin(struct file *file, struct address_space *mapping, |
de9a55b8 TT |
2664 | loff_t pos, unsigned len, unsigned flags, |
2665 | struct page **pagep, void **fsdata) | |
64769240 | 2666 | { |
72b8ab9d | 2667 | int ret, retries = 0; |
64769240 AT |
2668 | struct page *page; |
2669 | pgoff_t index; | |
64769240 AT |
2670 | struct inode *inode = mapping->host; |
2671 | handle_t *handle; | |
2672 | ||
2673 | index = pos >> PAGE_CACHE_SHIFT; | |
79f0be8d AK |
2674 | |
2675 | if (ext4_nonda_switch(inode->i_sb)) { | |
2676 | *fsdata = (void *)FALL_BACK_TO_NONDELALLOC; | |
2677 | return ext4_write_begin(file, mapping, pos, | |
2678 | len, flags, pagep, fsdata); | |
2679 | } | |
2680 | *fsdata = (void *)0; | |
9bffad1e | 2681 | trace_ext4_da_write_begin(inode, pos, len, flags); |
9c3569b5 TM |
2682 | |
2683 | if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { | |
2684 | ret = ext4_da_write_inline_data_begin(mapping, inode, | |
2685 | pos, len, flags, | |
2686 | pagep, fsdata); | |
2687 | if (ret < 0) | |
47564bfb TT |
2688 | return ret; |
2689 | if (ret == 1) | |
2690 | return 0; | |
9c3569b5 TM |
2691 | } |
2692 | ||
47564bfb TT |
2693 | /* |
2694 | * grab_cache_page_write_begin() can take a long time if the | |
2695 | * system is thrashing due to memory pressure, or if the page | |
2696 | * is being written back. So grab it first before we start | |
2697 | * the transaction handle. This also allows us to allocate | |
2698 | * the page (if needed) without using GFP_NOFS. | |
2699 | */ | |
2700 | retry_grab: | |
2701 | page = grab_cache_page_write_begin(mapping, index, flags); | |
2702 | if (!page) | |
2703 | return -ENOMEM; | |
2704 | unlock_page(page); | |
2705 | ||
64769240 AT |
2706 | /* |
2707 | * With delayed allocation, we don't log the i_disksize update | |
2708 | * if there is delayed block allocation. But we still need | |
2709 | * to journalling the i_disksize update if writes to the end | |
2710 | * of file which has an already mapped buffer. | |
2711 | */ | |
47564bfb | 2712 | retry_journal: |
9924a92a | 2713 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, 1); |
64769240 | 2714 | if (IS_ERR(handle)) { |
47564bfb TT |
2715 | page_cache_release(page); |
2716 | return PTR_ERR(handle); | |
64769240 AT |
2717 | } |
2718 | ||
47564bfb TT |
2719 | lock_page(page); |
2720 | if (page->mapping != mapping) { | |
2721 | /* The page got truncated from under us */ | |
2722 | unlock_page(page); | |
2723 | page_cache_release(page); | |
d5a0d4f7 | 2724 | ext4_journal_stop(handle); |
47564bfb | 2725 | goto retry_grab; |
d5a0d4f7 | 2726 | } |
47564bfb TT |
2727 | /* In case writeback began while the page was unlocked */ |
2728 | wait_on_page_writeback(page); | |
64769240 | 2729 | |
6e1db88d | 2730 | ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep); |
64769240 AT |
2731 | if (ret < 0) { |
2732 | unlock_page(page); | |
2733 | ext4_journal_stop(handle); | |
ae4d5372 AK |
2734 | /* |
2735 | * block_write_begin may have instantiated a few blocks | |
2736 | * outside i_size. Trim these off again. Don't need | |
2737 | * i_size_read because we hold i_mutex. | |
2738 | */ | |
2739 | if (pos + len > inode->i_size) | |
b9a4207d | 2740 | ext4_truncate_failed_write(inode); |
47564bfb TT |
2741 | |
2742 | if (ret == -ENOSPC && | |
2743 | ext4_should_retry_alloc(inode->i_sb, &retries)) | |
2744 | goto retry_journal; | |
2745 | ||
2746 | page_cache_release(page); | |
2747 | return ret; | |
64769240 AT |
2748 | } |
2749 | ||
47564bfb | 2750 | *pagep = page; |
64769240 AT |
2751 | return ret; |
2752 | } | |
2753 | ||
632eaeab MC |
2754 | /* |
2755 | * Check if we should update i_disksize | |
2756 | * when write to the end of file but not require block allocation | |
2757 | */ | |
2758 | static int ext4_da_should_update_i_disksize(struct page *page, | |
de9a55b8 | 2759 | unsigned long offset) |
632eaeab MC |
2760 | { |
2761 | struct buffer_head *bh; | |
2762 | struct inode *inode = page->mapping->host; | |
2763 | unsigned int idx; | |
2764 | int i; | |
2765 | ||
2766 | bh = page_buffers(page); | |
2767 | idx = offset >> inode->i_blkbits; | |
2768 | ||
af5bc92d | 2769 | for (i = 0; i < idx; i++) |
632eaeab MC |
2770 | bh = bh->b_this_page; |
2771 | ||
29fa89d0 | 2772 | if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh)) |
632eaeab MC |
2773 | return 0; |
2774 | return 1; | |
2775 | } | |
2776 | ||
64769240 | 2777 | static int ext4_da_write_end(struct file *file, |
de9a55b8 TT |
2778 | struct address_space *mapping, |
2779 | loff_t pos, unsigned len, unsigned copied, | |
2780 | struct page *page, void *fsdata) | |
64769240 AT |
2781 | { |
2782 | struct inode *inode = mapping->host; | |
2783 | int ret = 0, ret2; | |
2784 | handle_t *handle = ext4_journal_current_handle(); | |
2785 | loff_t new_i_size; | |
632eaeab | 2786 | unsigned long start, end; |
79f0be8d AK |
2787 | int write_mode = (int)(unsigned long)fsdata; |
2788 | ||
74d553aa TT |
2789 | if (write_mode == FALL_BACK_TO_NONDELALLOC) |
2790 | return ext4_write_end(file, mapping, pos, | |
2791 | len, copied, page, fsdata); | |
632eaeab | 2792 | |
9bffad1e | 2793 | trace_ext4_da_write_end(inode, pos, len, copied); |
632eaeab | 2794 | start = pos & (PAGE_CACHE_SIZE - 1); |
af5bc92d | 2795 | end = start + copied - 1; |
64769240 AT |
2796 | |
2797 | /* | |
2798 | * generic_write_end() will run mark_inode_dirty() if i_size | |
2799 | * changes. So let's piggyback the i_disksize mark_inode_dirty | |
2800 | * into that. | |
2801 | */ | |
64769240 | 2802 | new_i_size = pos + copied; |
ea51d132 | 2803 | if (copied && new_i_size > EXT4_I(inode)->i_disksize) { |
9c3569b5 TM |
2804 | if (ext4_has_inline_data(inode) || |
2805 | ext4_da_should_update_i_disksize(page, end)) { | |
632eaeab | 2806 | down_write(&EXT4_I(inode)->i_data_sem); |
f3b59291 | 2807 | if (new_i_size > EXT4_I(inode)->i_disksize) |
632eaeab | 2808 | EXT4_I(inode)->i_disksize = new_i_size; |
632eaeab | 2809 | up_write(&EXT4_I(inode)->i_data_sem); |
cf17fea6 AK |
2810 | /* We need to mark inode dirty even if |
2811 | * new_i_size is less that inode->i_size | |
2812 | * bu greater than i_disksize.(hint delalloc) | |
2813 | */ | |
2814 | ext4_mark_inode_dirty(handle, inode); | |
64769240 | 2815 | } |
632eaeab | 2816 | } |
9c3569b5 TM |
2817 | |
2818 | if (write_mode != CONVERT_INLINE_DATA && | |
2819 | ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) && | |
2820 | ext4_has_inline_data(inode)) | |
2821 | ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied, | |
2822 | page); | |
2823 | else | |
2824 | ret2 = generic_write_end(file, mapping, pos, len, copied, | |
64769240 | 2825 | page, fsdata); |
9c3569b5 | 2826 | |
64769240 AT |
2827 | copied = ret2; |
2828 | if (ret2 < 0) | |
2829 | ret = ret2; | |
2830 | ret2 = ext4_journal_stop(handle); | |
2831 | if (!ret) | |
2832 | ret = ret2; | |
2833 | ||
2834 | return ret ? ret : copied; | |
2835 | } | |
2836 | ||
d47992f8 LC |
2837 | static void ext4_da_invalidatepage(struct page *page, unsigned int offset, |
2838 | unsigned int length) | |
64769240 | 2839 | { |
64769240 AT |
2840 | /* |
2841 | * Drop reserved blocks | |
2842 | */ | |
2843 | BUG_ON(!PageLocked(page)); | |
2844 | if (!page_has_buffers(page)) | |
2845 | goto out; | |
2846 | ||
ca99fdd2 | 2847 | ext4_da_page_release_reservation(page, offset, length); |
64769240 AT |
2848 | |
2849 | out: | |
d47992f8 | 2850 | ext4_invalidatepage(page, offset, length); |
64769240 AT |
2851 | |
2852 | return; | |
2853 | } | |
2854 | ||
ccd2506b TT |
2855 | /* |
2856 | * Force all delayed allocation blocks to be allocated for a given inode. | |
2857 | */ | |
2858 | int ext4_alloc_da_blocks(struct inode *inode) | |
2859 | { | |
fb40ba0d TT |
2860 | trace_ext4_alloc_da_blocks(inode); |
2861 | ||
ccd2506b TT |
2862 | if (!EXT4_I(inode)->i_reserved_data_blocks && |
2863 | !EXT4_I(inode)->i_reserved_meta_blocks) | |
2864 | return 0; | |
2865 | ||
2866 | /* | |
2867 | * We do something simple for now. The filemap_flush() will | |
2868 | * also start triggering a write of the data blocks, which is | |
2869 | * not strictly speaking necessary (and for users of | |
2870 | * laptop_mode, not even desirable). However, to do otherwise | |
2871 | * would require replicating code paths in: | |
de9a55b8 | 2872 | * |
ccd2506b TT |
2873 | * ext4_da_writepages() -> |
2874 | * write_cache_pages() ---> (via passed in callback function) | |
2875 | * __mpage_da_writepage() --> | |
2876 | * mpage_add_bh_to_extent() | |
2877 | * mpage_da_map_blocks() | |
2878 | * | |
2879 | * The problem is that write_cache_pages(), located in | |
2880 | * mm/page-writeback.c, marks pages clean in preparation for | |
2881 | * doing I/O, which is not desirable if we're not planning on | |
2882 | * doing I/O at all. | |
2883 | * | |
2884 | * We could call write_cache_pages(), and then redirty all of | |
380cf090 | 2885 | * the pages by calling redirty_page_for_writepage() but that |
ccd2506b TT |
2886 | * would be ugly in the extreme. So instead we would need to |
2887 | * replicate parts of the code in the above functions, | |
25985edc | 2888 | * simplifying them because we wouldn't actually intend to |
ccd2506b TT |
2889 | * write out the pages, but rather only collect contiguous |
2890 | * logical block extents, call the multi-block allocator, and | |
2891 | * then update the buffer heads with the block allocations. | |
de9a55b8 | 2892 | * |
ccd2506b TT |
2893 | * For now, though, we'll cheat by calling filemap_flush(), |
2894 | * which will map the blocks, and start the I/O, but not | |
2895 | * actually wait for the I/O to complete. | |
2896 | */ | |
2897 | return filemap_flush(inode->i_mapping); | |
2898 | } | |
64769240 | 2899 | |
ac27a0ec DK |
2900 | /* |
2901 | * bmap() is special. It gets used by applications such as lilo and by | |
2902 | * the swapper to find the on-disk block of a specific piece of data. | |
2903 | * | |
2904 | * Naturally, this is dangerous if the block concerned is still in the | |
617ba13b | 2905 | * journal. If somebody makes a swapfile on an ext4 data-journaling |
ac27a0ec DK |
2906 | * filesystem and enables swap, then they may get a nasty shock when the |
2907 | * data getting swapped to that swapfile suddenly gets overwritten by | |
2908 | * the original zero's written out previously to the journal and | |
2909 | * awaiting writeback in the kernel's buffer cache. | |
2910 | * | |
2911 | * So, if we see any bmap calls here on a modified, data-journaled file, | |
2912 | * take extra steps to flush any blocks which might be in the cache. | |
2913 | */ | |
617ba13b | 2914 | static sector_t ext4_bmap(struct address_space *mapping, sector_t block) |
ac27a0ec DK |
2915 | { |
2916 | struct inode *inode = mapping->host; | |
2917 | journal_t *journal; | |
2918 | int err; | |
2919 | ||
46c7f254 TM |
2920 | /* |
2921 | * We can get here for an inline file via the FIBMAP ioctl | |
2922 | */ | |
2923 | if (ext4_has_inline_data(inode)) | |
2924 | return 0; | |
2925 | ||
64769240 AT |
2926 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) && |
2927 | test_opt(inode->i_sb, DELALLOC)) { | |
2928 | /* | |
2929 | * With delalloc we want to sync the file | |
2930 | * so that we can make sure we allocate | |
2931 | * blocks for file | |
2932 | */ | |
2933 | filemap_write_and_wait(mapping); | |
2934 | } | |
2935 | ||
19f5fb7a TT |
2936 | if (EXT4_JOURNAL(inode) && |
2937 | ext4_test_inode_state(inode, EXT4_STATE_JDATA)) { | |
ac27a0ec DK |
2938 | /* |
2939 | * This is a REALLY heavyweight approach, but the use of | |
2940 | * bmap on dirty files is expected to be extremely rare: | |
2941 | * only if we run lilo or swapon on a freshly made file | |
2942 | * do we expect this to happen. | |
2943 | * | |
2944 | * (bmap requires CAP_SYS_RAWIO so this does not | |
2945 | * represent an unprivileged user DOS attack --- we'd be | |
2946 | * in trouble if mortal users could trigger this path at | |
2947 | * will.) | |
2948 | * | |
617ba13b | 2949 | * NB. EXT4_STATE_JDATA is not set on files other than |
ac27a0ec DK |
2950 | * regular files. If somebody wants to bmap a directory |
2951 | * or symlink and gets confused because the buffer | |
2952 | * hasn't yet been flushed to disk, they deserve | |
2953 | * everything they get. | |
2954 | */ | |
2955 | ||
19f5fb7a | 2956 | ext4_clear_inode_state(inode, EXT4_STATE_JDATA); |
617ba13b | 2957 | journal = EXT4_JOURNAL(inode); |
dab291af MC |
2958 | jbd2_journal_lock_updates(journal); |
2959 | err = jbd2_journal_flush(journal); | |
2960 | jbd2_journal_unlock_updates(journal); | |
ac27a0ec DK |
2961 | |
2962 | if (err) | |
2963 | return 0; | |
2964 | } | |
2965 | ||
af5bc92d | 2966 | return generic_block_bmap(mapping, block, ext4_get_block); |
ac27a0ec DK |
2967 | } |
2968 | ||
617ba13b | 2969 | static int ext4_readpage(struct file *file, struct page *page) |
ac27a0ec | 2970 | { |
46c7f254 TM |
2971 | int ret = -EAGAIN; |
2972 | struct inode *inode = page->mapping->host; | |
2973 | ||
0562e0ba | 2974 | trace_ext4_readpage(page); |
46c7f254 TM |
2975 | |
2976 | if (ext4_has_inline_data(inode)) | |
2977 | ret = ext4_readpage_inline(inode, page); | |
2978 | ||
2979 | if (ret == -EAGAIN) | |
2980 | return mpage_readpage(page, ext4_get_block); | |
2981 | ||
2982 | return ret; | |
ac27a0ec DK |
2983 | } |
2984 | ||
2985 | static int | |
617ba13b | 2986 | ext4_readpages(struct file *file, struct address_space *mapping, |
ac27a0ec DK |
2987 | struct list_head *pages, unsigned nr_pages) |
2988 | { | |
46c7f254 TM |
2989 | struct inode *inode = mapping->host; |
2990 | ||
2991 | /* If the file has inline data, no need to do readpages. */ | |
2992 | if (ext4_has_inline_data(inode)) | |
2993 | return 0; | |
2994 | ||
617ba13b | 2995 | return mpage_readpages(mapping, pages, nr_pages, ext4_get_block); |
ac27a0ec DK |
2996 | } |
2997 | ||
d47992f8 LC |
2998 | static void ext4_invalidatepage(struct page *page, unsigned int offset, |
2999 | unsigned int length) | |
ac27a0ec | 3000 | { |
ca99fdd2 | 3001 | trace_ext4_invalidatepage(page, offset, length); |
0562e0ba | 3002 | |
4520fb3c JK |
3003 | /* No journalling happens on data buffers when this function is used */ |
3004 | WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page))); | |
3005 | ||
ca99fdd2 | 3006 | block_invalidatepage(page, offset, length); |
4520fb3c JK |
3007 | } |
3008 | ||
53e87268 | 3009 | static int __ext4_journalled_invalidatepage(struct page *page, |
ca99fdd2 LC |
3010 | unsigned int offset, |
3011 | unsigned int length) | |
4520fb3c JK |
3012 | { |
3013 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); | |
3014 | ||
ca99fdd2 | 3015 | trace_ext4_journalled_invalidatepage(page, offset, length); |
4520fb3c | 3016 | |
ac27a0ec DK |
3017 | /* |
3018 | * If it's a full truncate we just forget about the pending dirtying | |
3019 | */ | |
ca99fdd2 | 3020 | if (offset == 0 && length == PAGE_CACHE_SIZE) |
ac27a0ec DK |
3021 | ClearPageChecked(page); |
3022 | ||
ca99fdd2 | 3023 | return jbd2_journal_invalidatepage(journal, page, offset, length); |
53e87268 JK |
3024 | } |
3025 | ||
3026 | /* Wrapper for aops... */ | |
3027 | static void ext4_journalled_invalidatepage(struct page *page, | |
d47992f8 LC |
3028 | unsigned int offset, |
3029 | unsigned int length) | |
53e87268 | 3030 | { |
ca99fdd2 | 3031 | WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0); |
ac27a0ec DK |
3032 | } |
3033 | ||
617ba13b | 3034 | static int ext4_releasepage(struct page *page, gfp_t wait) |
ac27a0ec | 3035 | { |
617ba13b | 3036 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); |
ac27a0ec | 3037 | |
0562e0ba JZ |
3038 | trace_ext4_releasepage(page); |
3039 | ||
e1c36595 JK |
3040 | /* Page has dirty journalled data -> cannot release */ |
3041 | if (PageChecked(page)) | |
ac27a0ec | 3042 | return 0; |
0390131b FM |
3043 | if (journal) |
3044 | return jbd2_journal_try_to_free_buffers(journal, page, wait); | |
3045 | else | |
3046 | return try_to_free_buffers(page); | |
ac27a0ec DK |
3047 | } |
3048 | ||
2ed88685 TT |
3049 | /* |
3050 | * ext4_get_block used when preparing for a DIO write or buffer write. | |
3051 | * We allocate an uinitialized extent if blocks haven't been allocated. | |
3052 | * The extent will be converted to initialized after the IO is complete. | |
3053 | */ | |
f19d5870 | 3054 | int ext4_get_block_write(struct inode *inode, sector_t iblock, |
4c0425ff MC |
3055 | struct buffer_head *bh_result, int create) |
3056 | { | |
c7064ef1 | 3057 | ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n", |
8d5d02e6 | 3058 | inode->i_ino, create); |
2ed88685 TT |
3059 | return _ext4_get_block(inode, iblock, bh_result, |
3060 | EXT4_GET_BLOCKS_IO_CREATE_EXT); | |
4c0425ff MC |
3061 | } |
3062 | ||
729f52c6 | 3063 | static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock, |
8b0f165f | 3064 | struct buffer_head *bh_result, int create) |
729f52c6 | 3065 | { |
8b0f165f AP |
3066 | ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n", |
3067 | inode->i_ino, create); | |
3068 | return _ext4_get_block(inode, iblock, bh_result, | |
3069 | EXT4_GET_BLOCKS_NO_LOCK); | |
729f52c6 ZL |
3070 | } |
3071 | ||
4c0425ff | 3072 | static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset, |
552ef802 CH |
3073 | ssize_t size, void *private, int ret, |
3074 | bool is_async) | |
4c0425ff | 3075 | { |
496ad9aa | 3076 | struct inode *inode = file_inode(iocb->ki_filp); |
4c0425ff | 3077 | ext4_io_end_t *io_end = iocb->private; |
4c0425ff | 3078 | |
a549984b TT |
3079 | /* if not async direct IO or dio with 0 bytes write, just return */ |
3080 | if (!io_end || !size) | |
3081 | goto out; | |
4b70df18 | 3082 | |
88635ca2 | 3083 | ext_debug("ext4_end_io_dio(): io_end 0x%p " |
ace36ad4 | 3084 | "for inode %lu, iocb 0x%p, offset %llu, size %zd\n", |
8d5d02e6 MC |
3085 | iocb->private, io_end->inode->i_ino, iocb, offset, |
3086 | size); | |
8d5d02e6 | 3087 | |
b5a7e970 | 3088 | iocb->private = NULL; |
a549984b TT |
3089 | |
3090 | /* if not aio dio with unwritten extents, just free io and return */ | |
3091 | if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) { | |
3092 | ext4_free_io_end(io_end); | |
3093 | out: | |
3094 | inode_dio_done(inode); | |
3095 | if (is_async) | |
3096 | aio_complete(iocb, ret, 0); | |
3097 | return; | |
3098 | } | |
3099 | ||
4c0425ff MC |
3100 | io_end->offset = offset; |
3101 | io_end->size = size; | |
5b3ff237 JZ |
3102 | if (is_async) { |
3103 | io_end->iocb = iocb; | |
3104 | io_end->result = ret; | |
3105 | } | |
a549984b TT |
3106 | |
3107 | ext4_add_complete_io(io_end); | |
4c0425ff | 3108 | } |
c7064ef1 | 3109 | |
4c0425ff MC |
3110 | /* |
3111 | * For ext4 extent files, ext4 will do direct-io write to holes, | |
3112 | * preallocated extents, and those write extend the file, no need to | |
3113 | * fall back to buffered IO. | |
3114 | * | |
b595076a | 3115 | * For holes, we fallocate those blocks, mark them as uninitialized |
69c499d1 | 3116 | * If those blocks were preallocated, we mark sure they are split, but |
b595076a | 3117 | * still keep the range to write as uninitialized. |
4c0425ff | 3118 | * |
69c499d1 | 3119 | * The unwritten extents will be converted to written when DIO is completed. |
8d5d02e6 | 3120 | * For async direct IO, since the IO may still pending when return, we |
25985edc | 3121 | * set up an end_io call back function, which will do the conversion |
8d5d02e6 | 3122 | * when async direct IO completed. |
4c0425ff MC |
3123 | * |
3124 | * If the O_DIRECT write will extend the file then add this inode to the | |
3125 | * orphan list. So recovery will truncate it back to the original size | |
3126 | * if the machine crashes during the write. | |
3127 | * | |
3128 | */ | |
3129 | static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb, | |
3130 | const struct iovec *iov, loff_t offset, | |
3131 | unsigned long nr_segs) | |
3132 | { | |
3133 | struct file *file = iocb->ki_filp; | |
3134 | struct inode *inode = file->f_mapping->host; | |
3135 | ssize_t ret; | |
3136 | size_t count = iov_length(iov, nr_segs); | |
69c499d1 TT |
3137 | int overwrite = 0; |
3138 | get_block_t *get_block_func = NULL; | |
3139 | int dio_flags = 0; | |
4c0425ff | 3140 | loff_t final_size = offset + count; |
729f52c6 | 3141 | |
69c499d1 TT |
3142 | /* Use the old path for reads and writes beyond i_size. */ |
3143 | if (rw != WRITE || final_size > inode->i_size) | |
3144 | return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs); | |
4bd809db | 3145 | |
69c499d1 | 3146 | BUG_ON(iocb->private == NULL); |
4bd809db | 3147 | |
69c499d1 TT |
3148 | /* If we do a overwrite dio, i_mutex locking can be released */ |
3149 | overwrite = *((int *)iocb->private); | |
4bd809db | 3150 | |
69c499d1 TT |
3151 | if (overwrite) { |
3152 | atomic_inc(&inode->i_dio_count); | |
3153 | down_read(&EXT4_I(inode)->i_data_sem); | |
3154 | mutex_unlock(&inode->i_mutex); | |
3155 | } | |
8d5d02e6 | 3156 | |
69c499d1 TT |
3157 | /* |
3158 | * We could direct write to holes and fallocate. | |
3159 | * | |
3160 | * Allocated blocks to fill the hole are marked as | |
3161 | * uninitialized to prevent parallel buffered read to expose | |
3162 | * the stale data before DIO complete the data IO. | |
3163 | * | |
3164 | * As to previously fallocated extents, ext4 get_block will | |
3165 | * just simply mark the buffer mapped but still keep the | |
3166 | * extents uninitialized. | |
3167 | * | |
3168 | * For non AIO case, we will convert those unwritten extents | |
3169 | * to written after return back from blockdev_direct_IO. | |
3170 | * | |
3171 | * For async DIO, the conversion needs to be deferred when the | |
3172 | * IO is completed. The ext4 end_io callback function will be | |
3173 | * called to take care of the conversion work. Here for async | |
3174 | * case, we allocate an io_end structure to hook to the iocb. | |
3175 | */ | |
3176 | iocb->private = NULL; | |
3177 | ext4_inode_aio_set(inode, NULL); | |
3178 | if (!is_sync_kiocb(iocb)) { | |
a549984b | 3179 | ext4_io_end_t *io_end = ext4_init_io_end(inode, GFP_NOFS); |
69c499d1 TT |
3180 | if (!io_end) { |
3181 | ret = -ENOMEM; | |
3182 | goto retake_lock; | |
8b0f165f | 3183 | } |
69c499d1 | 3184 | io_end->flag |= EXT4_IO_END_DIRECT; |
a549984b | 3185 | iocb->private = io_end; |
8d5d02e6 | 3186 | /* |
69c499d1 TT |
3187 | * we save the io structure for current async direct |
3188 | * IO, so that later ext4_map_blocks() could flag the | |
3189 | * io structure whether there is a unwritten extents | |
3190 | * needs to be converted when IO is completed. | |
8d5d02e6 | 3191 | */ |
69c499d1 TT |
3192 | ext4_inode_aio_set(inode, io_end); |
3193 | } | |
4bd809db | 3194 | |
69c499d1 TT |
3195 | if (overwrite) { |
3196 | get_block_func = ext4_get_block_write_nolock; | |
3197 | } else { | |
3198 | get_block_func = ext4_get_block_write; | |
3199 | dio_flags = DIO_LOCKING; | |
3200 | } | |
3201 | ret = __blockdev_direct_IO(rw, iocb, inode, | |
3202 | inode->i_sb->s_bdev, iov, | |
3203 | offset, nr_segs, | |
3204 | get_block_func, | |
3205 | ext4_end_io_dio, | |
3206 | NULL, | |
3207 | dio_flags); | |
3208 | ||
a549984b TT |
3209 | if (iocb->private) |
3210 | ext4_inode_aio_set(inode, NULL); | |
69c499d1 | 3211 | /* |
a549984b TT |
3212 | * The io_end structure takes a reference to the inode, that |
3213 | * structure needs to be destroyed and the reference to the | |
3214 | * inode need to be dropped, when IO is complete, even with 0 | |
3215 | * byte write, or failed. | |
3216 | * | |
3217 | * In the successful AIO DIO case, the io_end structure will | |
3218 | * be destroyed and the reference to the inode will be dropped | |
3219 | * after the end_io call back function is called. | |
3220 | * | |
3221 | * In the case there is 0 byte write, or error case, since VFS | |
3222 | * direct IO won't invoke the end_io call back function, we | |
3223 | * need to free the end_io structure here. | |
69c499d1 | 3224 | */ |
a549984b TT |
3225 | if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) { |
3226 | ext4_free_io_end(iocb->private); | |
3227 | iocb->private = NULL; | |
3228 | } else if (ret > 0 && !overwrite && ext4_test_inode_state(inode, | |
69c499d1 TT |
3229 | EXT4_STATE_DIO_UNWRITTEN)) { |
3230 | int err; | |
3231 | /* | |
3232 | * for non AIO case, since the IO is already | |
3233 | * completed, we could do the conversion right here | |
3234 | */ | |
3235 | err = ext4_convert_unwritten_extents(inode, | |
3236 | offset, ret); | |
3237 | if (err < 0) | |
3238 | ret = err; | |
3239 | ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN); | |
3240 | } | |
4bd809db | 3241 | |
69c499d1 TT |
3242 | retake_lock: |
3243 | /* take i_mutex locking again if we do a ovewrite dio */ | |
3244 | if (overwrite) { | |
3245 | inode_dio_done(inode); | |
3246 | up_read(&EXT4_I(inode)->i_data_sem); | |
3247 | mutex_lock(&inode->i_mutex); | |
4c0425ff | 3248 | } |
8d5d02e6 | 3249 | |
69c499d1 | 3250 | return ret; |
4c0425ff MC |
3251 | } |
3252 | ||
3253 | static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb, | |
3254 | const struct iovec *iov, loff_t offset, | |
3255 | unsigned long nr_segs) | |
3256 | { | |
3257 | struct file *file = iocb->ki_filp; | |
3258 | struct inode *inode = file->f_mapping->host; | |
0562e0ba | 3259 | ssize_t ret; |
4c0425ff | 3260 | |
84ebd795 TT |
3261 | /* |
3262 | * If we are doing data journalling we don't support O_DIRECT | |
3263 | */ | |
3264 | if (ext4_should_journal_data(inode)) | |
3265 | return 0; | |
3266 | ||
46c7f254 TM |
3267 | /* Let buffer I/O handle the inline data case. */ |
3268 | if (ext4_has_inline_data(inode)) | |
3269 | return 0; | |
3270 | ||
0562e0ba | 3271 | trace_ext4_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw); |
12e9b892 | 3272 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
0562e0ba JZ |
3273 | ret = ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs); |
3274 | else | |
3275 | ret = ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs); | |
3276 | trace_ext4_direct_IO_exit(inode, offset, | |
3277 | iov_length(iov, nr_segs), rw, ret); | |
3278 | return ret; | |
4c0425ff MC |
3279 | } |
3280 | ||
ac27a0ec | 3281 | /* |
617ba13b | 3282 | * Pages can be marked dirty completely asynchronously from ext4's journalling |
ac27a0ec DK |
3283 | * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do |
3284 | * much here because ->set_page_dirty is called under VFS locks. The page is | |
3285 | * not necessarily locked. | |
3286 | * | |
3287 | * We cannot just dirty the page and leave attached buffers clean, because the | |
3288 | * buffers' dirty state is "definitive". We cannot just set the buffers dirty | |
3289 | * or jbddirty because all the journalling code will explode. | |
3290 | * | |
3291 | * So what we do is to mark the page "pending dirty" and next time writepage | |
3292 | * is called, propagate that into the buffers appropriately. | |
3293 | */ | |
617ba13b | 3294 | static int ext4_journalled_set_page_dirty(struct page *page) |
ac27a0ec DK |
3295 | { |
3296 | SetPageChecked(page); | |
3297 | return __set_page_dirty_nobuffers(page); | |
3298 | } | |
3299 | ||
74d553aa | 3300 | static const struct address_space_operations ext4_aops = { |
8ab22b9a HH |
3301 | .readpage = ext4_readpage, |
3302 | .readpages = ext4_readpages, | |
43ce1d23 | 3303 | .writepage = ext4_writepage, |
8ab22b9a | 3304 | .write_begin = ext4_write_begin, |
74d553aa | 3305 | .write_end = ext4_write_end, |
8ab22b9a HH |
3306 | .bmap = ext4_bmap, |
3307 | .invalidatepage = ext4_invalidatepage, | |
3308 | .releasepage = ext4_releasepage, | |
3309 | .direct_IO = ext4_direct_IO, | |
3310 | .migratepage = buffer_migrate_page, | |
3311 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 3312 | .error_remove_page = generic_error_remove_page, |
ac27a0ec DK |
3313 | }; |
3314 | ||
617ba13b | 3315 | static const struct address_space_operations ext4_journalled_aops = { |
8ab22b9a HH |
3316 | .readpage = ext4_readpage, |
3317 | .readpages = ext4_readpages, | |
43ce1d23 | 3318 | .writepage = ext4_writepage, |
8ab22b9a HH |
3319 | .write_begin = ext4_write_begin, |
3320 | .write_end = ext4_journalled_write_end, | |
3321 | .set_page_dirty = ext4_journalled_set_page_dirty, | |
3322 | .bmap = ext4_bmap, | |
4520fb3c | 3323 | .invalidatepage = ext4_journalled_invalidatepage, |
8ab22b9a | 3324 | .releasepage = ext4_releasepage, |
84ebd795 | 3325 | .direct_IO = ext4_direct_IO, |
8ab22b9a | 3326 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 3327 | .error_remove_page = generic_error_remove_page, |
ac27a0ec DK |
3328 | }; |
3329 | ||
64769240 | 3330 | static const struct address_space_operations ext4_da_aops = { |
8ab22b9a HH |
3331 | .readpage = ext4_readpage, |
3332 | .readpages = ext4_readpages, | |
43ce1d23 | 3333 | .writepage = ext4_writepage, |
8ab22b9a | 3334 | .writepages = ext4_da_writepages, |
8ab22b9a HH |
3335 | .write_begin = ext4_da_write_begin, |
3336 | .write_end = ext4_da_write_end, | |
3337 | .bmap = ext4_bmap, | |
3338 | .invalidatepage = ext4_da_invalidatepage, | |
3339 | .releasepage = ext4_releasepage, | |
3340 | .direct_IO = ext4_direct_IO, | |
3341 | .migratepage = buffer_migrate_page, | |
3342 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 3343 | .error_remove_page = generic_error_remove_page, |
64769240 AT |
3344 | }; |
3345 | ||
617ba13b | 3346 | void ext4_set_aops(struct inode *inode) |
ac27a0ec | 3347 | { |
3d2b1582 LC |
3348 | switch (ext4_inode_journal_mode(inode)) { |
3349 | case EXT4_INODE_ORDERED_DATA_MODE: | |
74d553aa | 3350 | ext4_set_inode_state(inode, EXT4_STATE_ORDERED_MODE); |
3d2b1582 LC |
3351 | break; |
3352 | case EXT4_INODE_WRITEBACK_DATA_MODE: | |
74d553aa | 3353 | ext4_clear_inode_state(inode, EXT4_STATE_ORDERED_MODE); |
3d2b1582 LC |
3354 | break; |
3355 | case EXT4_INODE_JOURNAL_DATA_MODE: | |
617ba13b | 3356 | inode->i_mapping->a_ops = &ext4_journalled_aops; |
74d553aa | 3357 | return; |
3d2b1582 LC |
3358 | default: |
3359 | BUG(); | |
3360 | } | |
74d553aa TT |
3361 | if (test_opt(inode->i_sb, DELALLOC)) |
3362 | inode->i_mapping->a_ops = &ext4_da_aops; | |
3363 | else | |
3364 | inode->i_mapping->a_ops = &ext4_aops; | |
ac27a0ec DK |
3365 | } |
3366 | ||
d863dc36 LC |
3367 | /* |
3368 | * ext4_block_truncate_page() zeroes out a mapping from file offset `from' | |
3369 | * up to the end of the block which corresponds to `from'. | |
3370 | * This required during truncate. We need to physically zero the tail end | |
3371 | * of that block so it doesn't yield old data if the file is later grown. | |
3372 | */ | |
3373 | int ext4_block_truncate_page(handle_t *handle, | |
3374 | struct address_space *mapping, loff_t from) | |
3375 | { | |
3376 | unsigned offset = from & (PAGE_CACHE_SIZE-1); | |
3377 | unsigned length; | |
3378 | unsigned blocksize; | |
3379 | struct inode *inode = mapping->host; | |
3380 | ||
3381 | blocksize = inode->i_sb->s_blocksize; | |
3382 | length = blocksize - (offset & (blocksize - 1)); | |
3383 | ||
3384 | return ext4_block_zero_page_range(handle, mapping, from, length); | |
3385 | } | |
3386 | ||
3387 | /* | |
3388 | * ext4_block_zero_page_range() zeros out a mapping of length 'length' | |
3389 | * starting from file offset 'from'. The range to be zero'd must | |
3390 | * be contained with in one block. If the specified range exceeds | |
3391 | * the end of the block it will be shortened to end of the block | |
3392 | * that cooresponds to 'from' | |
3393 | */ | |
3394 | int ext4_block_zero_page_range(handle_t *handle, | |
3395 | struct address_space *mapping, loff_t from, loff_t length) | |
3396 | { | |
3397 | ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT; | |
3398 | unsigned offset = from & (PAGE_CACHE_SIZE-1); | |
3399 | unsigned blocksize, max, pos; | |
3400 | ext4_lblk_t iblock; | |
3401 | struct inode *inode = mapping->host; | |
3402 | struct buffer_head *bh; | |
3403 | struct page *page; | |
3404 | int err = 0; | |
3405 | ||
3406 | page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT, | |
3407 | mapping_gfp_mask(mapping) & ~__GFP_FS); | |
3408 | if (!page) | |
3409 | return -ENOMEM; | |
3410 | ||
3411 | blocksize = inode->i_sb->s_blocksize; | |
3412 | max = blocksize - (offset & (blocksize - 1)); | |
3413 | ||
3414 | /* | |
3415 | * correct length if it does not fall between | |
3416 | * 'from' and the end of the block | |
3417 | */ | |
3418 | if (length > max || length < 0) | |
3419 | length = max; | |
3420 | ||
3421 | iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits); | |
3422 | ||
3423 | if (!page_has_buffers(page)) | |
3424 | create_empty_buffers(page, blocksize, 0); | |
3425 | ||
3426 | /* Find the buffer that contains "offset" */ | |
3427 | bh = page_buffers(page); | |
3428 | pos = blocksize; | |
3429 | while (offset >= pos) { | |
3430 | bh = bh->b_this_page; | |
3431 | iblock++; | |
3432 | pos += blocksize; | |
3433 | } | |
3434 | ||
3435 | err = 0; | |
3436 | if (buffer_freed(bh)) { | |
3437 | BUFFER_TRACE(bh, "freed: skip"); | |
3438 | goto unlock; | |
3439 | } | |
3440 | ||
3441 | if (!buffer_mapped(bh)) { | |
3442 | BUFFER_TRACE(bh, "unmapped"); | |
3443 | ext4_get_block(inode, iblock, bh, 0); | |
3444 | /* unmapped? It's a hole - nothing to do */ | |
3445 | if (!buffer_mapped(bh)) { | |
3446 | BUFFER_TRACE(bh, "still unmapped"); | |
3447 | goto unlock; | |
3448 | } | |
3449 | } | |
3450 | ||
3451 | /* Ok, it's mapped. Make sure it's up-to-date */ | |
3452 | if (PageUptodate(page)) | |
3453 | set_buffer_uptodate(bh); | |
3454 | ||
3455 | if (!buffer_uptodate(bh)) { | |
3456 | err = -EIO; | |
3457 | ll_rw_block(READ, 1, &bh); | |
3458 | wait_on_buffer(bh); | |
3459 | /* Uhhuh. Read error. Complain and punt. */ | |
3460 | if (!buffer_uptodate(bh)) | |
3461 | goto unlock; | |
3462 | } | |
3463 | ||
3464 | if (ext4_should_journal_data(inode)) { | |
3465 | BUFFER_TRACE(bh, "get write access"); | |
3466 | err = ext4_journal_get_write_access(handle, bh); | |
3467 | if (err) | |
3468 | goto unlock; | |
3469 | } | |
3470 | ||
3471 | zero_user(page, offset, length); | |
3472 | ||
3473 | BUFFER_TRACE(bh, "zeroed end of block"); | |
3474 | ||
3475 | err = 0; | |
3476 | if (ext4_should_journal_data(inode)) { | |
3477 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
0713ed0c | 3478 | } else { |
d863dc36 | 3479 | mark_buffer_dirty(bh); |
0713ed0c LC |
3480 | if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE)) |
3481 | err = ext4_jbd2_file_inode(handle, inode); | |
3482 | } | |
d863dc36 LC |
3483 | |
3484 | unlock: | |
3485 | unlock_page(page); | |
3486 | page_cache_release(page); | |
3487 | return err; | |
3488 | } | |
3489 | ||
a87dd18c LC |
3490 | int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode, |
3491 | loff_t lstart, loff_t length) | |
3492 | { | |
3493 | struct super_block *sb = inode->i_sb; | |
3494 | struct address_space *mapping = inode->i_mapping; | |
3495 | unsigned partial = lstart & (sb->s_blocksize - 1); | |
3496 | ext4_fsblk_t start, end; | |
3497 | loff_t byte_end = (lstart + length - 1); | |
3498 | int err = 0; | |
3499 | ||
3500 | start = lstart >> sb->s_blocksize_bits; | |
3501 | end = byte_end >> sb->s_blocksize_bits; | |
3502 | ||
3503 | /* Handle partial zero within the single block */ | |
3504 | if (start == end) { | |
3505 | err = ext4_block_zero_page_range(handle, mapping, | |
3506 | lstart, length); | |
3507 | return err; | |
3508 | } | |
3509 | /* Handle partial zero out on the start of the range */ | |
3510 | if (partial) { | |
3511 | err = ext4_block_zero_page_range(handle, mapping, | |
3512 | lstart, sb->s_blocksize); | |
3513 | if (err) | |
3514 | return err; | |
3515 | } | |
3516 | /* Handle partial zero out on the end of the range */ | |
3517 | partial = byte_end & (sb->s_blocksize - 1); | |
3518 | if (partial != sb->s_blocksize - 1) | |
3519 | err = ext4_block_zero_page_range(handle, mapping, | |
3520 | byte_end - partial, | |
3521 | partial + 1); | |
3522 | return err; | |
3523 | } | |
3524 | ||
91ef4caf DG |
3525 | int ext4_can_truncate(struct inode *inode) |
3526 | { | |
91ef4caf DG |
3527 | if (S_ISREG(inode->i_mode)) |
3528 | return 1; | |
3529 | if (S_ISDIR(inode->i_mode)) | |
3530 | return 1; | |
3531 | if (S_ISLNK(inode->i_mode)) | |
3532 | return !ext4_inode_is_fast_symlink(inode); | |
3533 | return 0; | |
3534 | } | |
3535 | ||
a4bb6b64 AH |
3536 | /* |
3537 | * ext4_punch_hole: punches a hole in a file by releaseing the blocks | |
3538 | * associated with the given offset and length | |
3539 | * | |
3540 | * @inode: File inode | |
3541 | * @offset: The offset where the hole will begin | |
3542 | * @len: The length of the hole | |
3543 | * | |
4907cb7b | 3544 | * Returns: 0 on success or negative on failure |
a4bb6b64 AH |
3545 | */ |
3546 | ||
3547 | int ext4_punch_hole(struct file *file, loff_t offset, loff_t length) | |
3548 | { | |
496ad9aa | 3549 | struct inode *inode = file_inode(file); |
26a4c0c6 TT |
3550 | struct super_block *sb = inode->i_sb; |
3551 | ext4_lblk_t first_block, stop_block; | |
3552 | struct address_space *mapping = inode->i_mapping; | |
a87dd18c | 3553 | loff_t first_block_offset, last_block_offset; |
26a4c0c6 TT |
3554 | handle_t *handle; |
3555 | unsigned int credits; | |
3556 | int ret = 0; | |
3557 | ||
a4bb6b64 | 3558 | if (!S_ISREG(inode->i_mode)) |
73355192 | 3559 | return -EOPNOTSUPP; |
a4bb6b64 | 3560 | |
26a4c0c6 | 3561 | if (EXT4_SB(sb)->s_cluster_ratio > 1) { |
bab08ab9 | 3562 | /* TODO: Add support for bigalloc file systems */ |
73355192 | 3563 | return -EOPNOTSUPP; |
bab08ab9 TT |
3564 | } |
3565 | ||
aaddea81 ZL |
3566 | trace_ext4_punch_hole(inode, offset, length); |
3567 | ||
26a4c0c6 TT |
3568 | /* |
3569 | * Write out all dirty pages to avoid race conditions | |
3570 | * Then release them. | |
3571 | */ | |
3572 | if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { | |
3573 | ret = filemap_write_and_wait_range(mapping, offset, | |
3574 | offset + length - 1); | |
3575 | if (ret) | |
3576 | return ret; | |
3577 | } | |
3578 | ||
3579 | mutex_lock(&inode->i_mutex); | |
3580 | /* It's not possible punch hole on append only file */ | |
3581 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) { | |
3582 | ret = -EPERM; | |
3583 | goto out_mutex; | |
3584 | } | |
3585 | if (IS_SWAPFILE(inode)) { | |
3586 | ret = -ETXTBSY; | |
3587 | goto out_mutex; | |
3588 | } | |
3589 | ||
3590 | /* No need to punch hole beyond i_size */ | |
3591 | if (offset >= inode->i_size) | |
3592 | goto out_mutex; | |
3593 | ||
3594 | /* | |
3595 | * If the hole extends beyond i_size, set the hole | |
3596 | * to end after the page that contains i_size | |
3597 | */ | |
3598 | if (offset + length > inode->i_size) { | |
3599 | length = inode->i_size + | |
3600 | PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) - | |
3601 | offset; | |
3602 | } | |
3603 | ||
a87dd18c LC |
3604 | first_block_offset = round_up(offset, sb->s_blocksize); |
3605 | last_block_offset = round_down((offset + length), sb->s_blocksize) - 1; | |
26a4c0c6 | 3606 | |
a87dd18c LC |
3607 | /* Now release the pages and zero block aligned part of pages*/ |
3608 | if (last_block_offset > first_block_offset) | |
3609 | truncate_pagecache_range(inode, first_block_offset, | |
3610 | last_block_offset); | |
26a4c0c6 TT |
3611 | |
3612 | /* Wait all existing dio workers, newcomers will block on i_mutex */ | |
3613 | ext4_inode_block_unlocked_dio(inode); | |
3614 | ret = ext4_flush_unwritten_io(inode); | |
3615 | if (ret) | |
3616 | goto out_dio; | |
3617 | inode_dio_wait(inode); | |
3618 | ||
3619 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
3620 | credits = ext4_writepage_trans_blocks(inode); | |
3621 | else | |
3622 | credits = ext4_blocks_for_truncate(inode); | |
3623 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); | |
3624 | if (IS_ERR(handle)) { | |
3625 | ret = PTR_ERR(handle); | |
3626 | ext4_std_error(sb, ret); | |
3627 | goto out_dio; | |
3628 | } | |
3629 | ||
a87dd18c LC |
3630 | ret = ext4_zero_partial_blocks(handle, inode, offset, |
3631 | length); | |
3632 | if (ret) | |
3633 | goto out_stop; | |
26a4c0c6 TT |
3634 | |
3635 | first_block = (offset + sb->s_blocksize - 1) >> | |
3636 | EXT4_BLOCK_SIZE_BITS(sb); | |
3637 | stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb); | |
3638 | ||
3639 | /* If there are no blocks to remove, return now */ | |
3640 | if (first_block >= stop_block) | |
3641 | goto out_stop; | |
3642 | ||
3643 | down_write(&EXT4_I(inode)->i_data_sem); | |
3644 | ext4_discard_preallocations(inode); | |
3645 | ||
3646 | ret = ext4_es_remove_extent(inode, first_block, | |
3647 | stop_block - first_block); | |
3648 | if (ret) { | |
3649 | up_write(&EXT4_I(inode)->i_data_sem); | |
3650 | goto out_stop; | |
3651 | } | |
3652 | ||
3653 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
3654 | ret = ext4_ext_remove_space(inode, first_block, | |
3655 | stop_block - 1); | |
3656 | else | |
3657 | ret = ext4_free_hole_blocks(handle, inode, first_block, | |
3658 | stop_block); | |
3659 | ||
3660 | ext4_discard_preallocations(inode); | |
819c4920 | 3661 | up_write(&EXT4_I(inode)->i_data_sem); |
26a4c0c6 TT |
3662 | if (IS_SYNC(inode)) |
3663 | ext4_handle_sync(handle); | |
26a4c0c6 TT |
3664 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); |
3665 | ext4_mark_inode_dirty(handle, inode); | |
3666 | out_stop: | |
3667 | ext4_journal_stop(handle); | |
3668 | out_dio: | |
3669 | ext4_inode_resume_unlocked_dio(inode); | |
3670 | out_mutex: | |
3671 | mutex_unlock(&inode->i_mutex); | |
3672 | return ret; | |
a4bb6b64 AH |
3673 | } |
3674 | ||
ac27a0ec | 3675 | /* |
617ba13b | 3676 | * ext4_truncate() |
ac27a0ec | 3677 | * |
617ba13b MC |
3678 | * We block out ext4_get_block() block instantiations across the entire |
3679 | * transaction, and VFS/VM ensures that ext4_truncate() cannot run | |
ac27a0ec DK |
3680 | * simultaneously on behalf of the same inode. |
3681 | * | |
42b2aa86 | 3682 | * As we work through the truncate and commit bits of it to the journal there |
ac27a0ec DK |
3683 | * is one core, guiding principle: the file's tree must always be consistent on |
3684 | * disk. We must be able to restart the truncate after a crash. | |
3685 | * | |
3686 | * The file's tree may be transiently inconsistent in memory (although it | |
3687 | * probably isn't), but whenever we close off and commit a journal transaction, | |
3688 | * the contents of (the filesystem + the journal) must be consistent and | |
3689 | * restartable. It's pretty simple, really: bottom up, right to left (although | |
3690 | * left-to-right works OK too). | |
3691 | * | |
3692 | * Note that at recovery time, journal replay occurs *before* the restart of | |
3693 | * truncate against the orphan inode list. | |
3694 | * | |
3695 | * The committed inode has the new, desired i_size (which is the same as | |
617ba13b | 3696 | * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see |
ac27a0ec | 3697 | * that this inode's truncate did not complete and it will again call |
617ba13b MC |
3698 | * ext4_truncate() to have another go. So there will be instantiated blocks |
3699 | * to the right of the truncation point in a crashed ext4 filesystem. But | |
ac27a0ec | 3700 | * that's fine - as long as they are linked from the inode, the post-crash |
617ba13b | 3701 | * ext4_truncate() run will find them and release them. |
ac27a0ec | 3702 | */ |
617ba13b | 3703 | void ext4_truncate(struct inode *inode) |
ac27a0ec | 3704 | { |
819c4920 TT |
3705 | struct ext4_inode_info *ei = EXT4_I(inode); |
3706 | unsigned int credits; | |
3707 | handle_t *handle; | |
3708 | struct address_space *mapping = inode->i_mapping; | |
819c4920 | 3709 | |
19b5ef61 TT |
3710 | /* |
3711 | * There is a possibility that we're either freeing the inode | |
3712 | * or it completely new indode. In those cases we might not | |
3713 | * have i_mutex locked because it's not necessary. | |
3714 | */ | |
3715 | if (!(inode->i_state & (I_NEW|I_FREEING))) | |
3716 | WARN_ON(!mutex_is_locked(&inode->i_mutex)); | |
0562e0ba JZ |
3717 | trace_ext4_truncate_enter(inode); |
3718 | ||
91ef4caf | 3719 | if (!ext4_can_truncate(inode)) |
ac27a0ec DK |
3720 | return; |
3721 | ||
12e9b892 | 3722 | ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS); |
c8d46e41 | 3723 | |
5534fb5b | 3724 | if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC)) |
19f5fb7a | 3725 | ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE); |
7d8f9f7d | 3726 | |
aef1c851 TM |
3727 | if (ext4_has_inline_data(inode)) { |
3728 | int has_inline = 1; | |
3729 | ||
3730 | ext4_inline_data_truncate(inode, &has_inline); | |
3731 | if (has_inline) | |
3732 | return; | |
3733 | } | |
3734 | ||
819c4920 TT |
3735 | /* |
3736 | * finish any pending end_io work so we won't run the risk of | |
3737 | * converting any truncated blocks to initialized later | |
3738 | */ | |
3739 | ext4_flush_unwritten_io(inode); | |
3740 | ||
3741 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
3742 | credits = ext4_writepage_trans_blocks(inode); | |
3743 | else | |
3744 | credits = ext4_blocks_for_truncate(inode); | |
3745 | ||
3746 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); | |
3747 | if (IS_ERR(handle)) { | |
3748 | ext4_std_error(inode->i_sb, PTR_ERR(handle)); | |
3749 | return; | |
3750 | } | |
3751 | ||
eb3544c6 LC |
3752 | if (inode->i_size & (inode->i_sb->s_blocksize - 1)) |
3753 | ext4_block_truncate_page(handle, mapping, inode->i_size); | |
819c4920 TT |
3754 | |
3755 | /* | |
3756 | * We add the inode to the orphan list, so that if this | |
3757 | * truncate spans multiple transactions, and we crash, we will | |
3758 | * resume the truncate when the filesystem recovers. It also | |
3759 | * marks the inode dirty, to catch the new size. | |
3760 | * | |
3761 | * Implication: the file must always be in a sane, consistent | |
3762 | * truncatable state while each transaction commits. | |
3763 | */ | |
3764 | if (ext4_orphan_add(handle, inode)) | |
3765 | goto out_stop; | |
3766 | ||
3767 | down_write(&EXT4_I(inode)->i_data_sem); | |
3768 | ||
3769 | ext4_discard_preallocations(inode); | |
3770 | ||
ff9893dc | 3771 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
819c4920 | 3772 | ext4_ext_truncate(handle, inode); |
ff9893dc | 3773 | else |
819c4920 TT |
3774 | ext4_ind_truncate(handle, inode); |
3775 | ||
3776 | up_write(&ei->i_data_sem); | |
3777 | ||
3778 | if (IS_SYNC(inode)) | |
3779 | ext4_handle_sync(handle); | |
3780 | ||
3781 | out_stop: | |
3782 | /* | |
3783 | * If this was a simple ftruncate() and the file will remain alive, | |
3784 | * then we need to clear up the orphan record which we created above. | |
3785 | * However, if this was a real unlink then we were called by | |
3786 | * ext4_delete_inode(), and we allow that function to clean up the | |
3787 | * orphan info for us. | |
3788 | */ | |
3789 | if (inode->i_nlink) | |
3790 | ext4_orphan_del(handle, inode); | |
3791 | ||
3792 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); | |
3793 | ext4_mark_inode_dirty(handle, inode); | |
3794 | ext4_journal_stop(handle); | |
ac27a0ec | 3795 | |
0562e0ba | 3796 | trace_ext4_truncate_exit(inode); |
ac27a0ec DK |
3797 | } |
3798 | ||
ac27a0ec | 3799 | /* |
617ba13b | 3800 | * ext4_get_inode_loc returns with an extra refcount against the inode's |
ac27a0ec DK |
3801 | * underlying buffer_head on success. If 'in_mem' is true, we have all |
3802 | * data in memory that is needed to recreate the on-disk version of this | |
3803 | * inode. | |
3804 | */ | |
617ba13b MC |
3805 | static int __ext4_get_inode_loc(struct inode *inode, |
3806 | struct ext4_iloc *iloc, int in_mem) | |
ac27a0ec | 3807 | { |
240799cd TT |
3808 | struct ext4_group_desc *gdp; |
3809 | struct buffer_head *bh; | |
3810 | struct super_block *sb = inode->i_sb; | |
3811 | ext4_fsblk_t block; | |
3812 | int inodes_per_block, inode_offset; | |
3813 | ||
3a06d778 | 3814 | iloc->bh = NULL; |
240799cd TT |
3815 | if (!ext4_valid_inum(sb, inode->i_ino)) |
3816 | return -EIO; | |
ac27a0ec | 3817 | |
240799cd TT |
3818 | iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb); |
3819 | gdp = ext4_get_group_desc(sb, iloc->block_group, NULL); | |
3820 | if (!gdp) | |
ac27a0ec DK |
3821 | return -EIO; |
3822 | ||
240799cd TT |
3823 | /* |
3824 | * Figure out the offset within the block group inode table | |
3825 | */ | |
00d09882 | 3826 | inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; |
240799cd TT |
3827 | inode_offset = ((inode->i_ino - 1) % |
3828 | EXT4_INODES_PER_GROUP(sb)); | |
3829 | block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block); | |
3830 | iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb); | |
3831 | ||
3832 | bh = sb_getblk(sb, block); | |
aebf0243 | 3833 | if (unlikely(!bh)) |
860d21e2 | 3834 | return -ENOMEM; |
ac27a0ec DK |
3835 | if (!buffer_uptodate(bh)) { |
3836 | lock_buffer(bh); | |
9c83a923 HK |
3837 | |
3838 | /* | |
3839 | * If the buffer has the write error flag, we have failed | |
3840 | * to write out another inode in the same block. In this | |
3841 | * case, we don't have to read the block because we may | |
3842 | * read the old inode data successfully. | |
3843 | */ | |
3844 | if (buffer_write_io_error(bh) && !buffer_uptodate(bh)) | |
3845 | set_buffer_uptodate(bh); | |
3846 | ||
ac27a0ec DK |
3847 | if (buffer_uptodate(bh)) { |
3848 | /* someone brought it uptodate while we waited */ | |
3849 | unlock_buffer(bh); | |
3850 | goto has_buffer; | |
3851 | } | |
3852 | ||
3853 | /* | |
3854 | * If we have all information of the inode in memory and this | |
3855 | * is the only valid inode in the block, we need not read the | |
3856 | * block. | |
3857 | */ | |
3858 | if (in_mem) { | |
3859 | struct buffer_head *bitmap_bh; | |
240799cd | 3860 | int i, start; |
ac27a0ec | 3861 | |
240799cd | 3862 | start = inode_offset & ~(inodes_per_block - 1); |
ac27a0ec | 3863 | |
240799cd TT |
3864 | /* Is the inode bitmap in cache? */ |
3865 | bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp)); | |
aebf0243 | 3866 | if (unlikely(!bitmap_bh)) |
ac27a0ec DK |
3867 | goto make_io; |
3868 | ||
3869 | /* | |
3870 | * If the inode bitmap isn't in cache then the | |
3871 | * optimisation may end up performing two reads instead | |
3872 | * of one, so skip it. | |
3873 | */ | |
3874 | if (!buffer_uptodate(bitmap_bh)) { | |
3875 | brelse(bitmap_bh); | |
3876 | goto make_io; | |
3877 | } | |
240799cd | 3878 | for (i = start; i < start + inodes_per_block; i++) { |
ac27a0ec DK |
3879 | if (i == inode_offset) |
3880 | continue; | |
617ba13b | 3881 | if (ext4_test_bit(i, bitmap_bh->b_data)) |
ac27a0ec DK |
3882 | break; |
3883 | } | |
3884 | brelse(bitmap_bh); | |
240799cd | 3885 | if (i == start + inodes_per_block) { |
ac27a0ec DK |
3886 | /* all other inodes are free, so skip I/O */ |
3887 | memset(bh->b_data, 0, bh->b_size); | |
3888 | set_buffer_uptodate(bh); | |
3889 | unlock_buffer(bh); | |
3890 | goto has_buffer; | |
3891 | } | |
3892 | } | |
3893 | ||
3894 | make_io: | |
240799cd TT |
3895 | /* |
3896 | * If we need to do any I/O, try to pre-readahead extra | |
3897 | * blocks from the inode table. | |
3898 | */ | |
3899 | if (EXT4_SB(sb)->s_inode_readahead_blks) { | |
3900 | ext4_fsblk_t b, end, table; | |
3901 | unsigned num; | |
0d606e2c | 3902 | __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks; |
240799cd TT |
3903 | |
3904 | table = ext4_inode_table(sb, gdp); | |
b713a5ec | 3905 | /* s_inode_readahead_blks is always a power of 2 */ |
0d606e2c | 3906 | b = block & ~((ext4_fsblk_t) ra_blks - 1); |
240799cd TT |
3907 | if (table > b) |
3908 | b = table; | |
0d606e2c | 3909 | end = b + ra_blks; |
240799cd | 3910 | num = EXT4_INODES_PER_GROUP(sb); |
feb0ab32 | 3911 | if (ext4_has_group_desc_csum(sb)) |
560671a0 | 3912 | num -= ext4_itable_unused_count(sb, gdp); |
240799cd TT |
3913 | table += num / inodes_per_block; |
3914 | if (end > table) | |
3915 | end = table; | |
3916 | while (b <= end) | |
3917 | sb_breadahead(sb, b++); | |
3918 | } | |
3919 | ||
ac27a0ec DK |
3920 | /* |
3921 | * There are other valid inodes in the buffer, this inode | |
3922 | * has in-inode xattrs, or we don't have this inode in memory. | |
3923 | * Read the block from disk. | |
3924 | */ | |
0562e0ba | 3925 | trace_ext4_load_inode(inode); |
ac27a0ec DK |
3926 | get_bh(bh); |
3927 | bh->b_end_io = end_buffer_read_sync; | |
65299a3b | 3928 | submit_bh(READ | REQ_META | REQ_PRIO, bh); |
ac27a0ec DK |
3929 | wait_on_buffer(bh); |
3930 | if (!buffer_uptodate(bh)) { | |
c398eda0 TT |
3931 | EXT4_ERROR_INODE_BLOCK(inode, block, |
3932 | "unable to read itable block"); | |
ac27a0ec DK |
3933 | brelse(bh); |
3934 | return -EIO; | |
3935 | } | |
3936 | } | |
3937 | has_buffer: | |
3938 | iloc->bh = bh; | |
3939 | return 0; | |
3940 | } | |
3941 | ||
617ba13b | 3942 | int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec DK |
3943 | { |
3944 | /* We have all inode data except xattrs in memory here. */ | |
617ba13b | 3945 | return __ext4_get_inode_loc(inode, iloc, |
19f5fb7a | 3946 | !ext4_test_inode_state(inode, EXT4_STATE_XATTR)); |
ac27a0ec DK |
3947 | } |
3948 | ||
617ba13b | 3949 | void ext4_set_inode_flags(struct inode *inode) |
ac27a0ec | 3950 | { |
617ba13b | 3951 | unsigned int flags = EXT4_I(inode)->i_flags; |
ac27a0ec DK |
3952 | |
3953 | inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC); | |
617ba13b | 3954 | if (flags & EXT4_SYNC_FL) |
ac27a0ec | 3955 | inode->i_flags |= S_SYNC; |
617ba13b | 3956 | if (flags & EXT4_APPEND_FL) |
ac27a0ec | 3957 | inode->i_flags |= S_APPEND; |
617ba13b | 3958 | if (flags & EXT4_IMMUTABLE_FL) |
ac27a0ec | 3959 | inode->i_flags |= S_IMMUTABLE; |
617ba13b | 3960 | if (flags & EXT4_NOATIME_FL) |
ac27a0ec | 3961 | inode->i_flags |= S_NOATIME; |
617ba13b | 3962 | if (flags & EXT4_DIRSYNC_FL) |
ac27a0ec DK |
3963 | inode->i_flags |= S_DIRSYNC; |
3964 | } | |
3965 | ||
ff9ddf7e JK |
3966 | /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */ |
3967 | void ext4_get_inode_flags(struct ext4_inode_info *ei) | |
3968 | { | |
84a8dce2 DM |
3969 | unsigned int vfs_fl; |
3970 | unsigned long old_fl, new_fl; | |
3971 | ||
3972 | do { | |
3973 | vfs_fl = ei->vfs_inode.i_flags; | |
3974 | old_fl = ei->i_flags; | |
3975 | new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL| | |
3976 | EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL| | |
3977 | EXT4_DIRSYNC_FL); | |
3978 | if (vfs_fl & S_SYNC) | |
3979 | new_fl |= EXT4_SYNC_FL; | |
3980 | if (vfs_fl & S_APPEND) | |
3981 | new_fl |= EXT4_APPEND_FL; | |
3982 | if (vfs_fl & S_IMMUTABLE) | |
3983 | new_fl |= EXT4_IMMUTABLE_FL; | |
3984 | if (vfs_fl & S_NOATIME) | |
3985 | new_fl |= EXT4_NOATIME_FL; | |
3986 | if (vfs_fl & S_DIRSYNC) | |
3987 | new_fl |= EXT4_DIRSYNC_FL; | |
3988 | } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl); | |
ff9ddf7e | 3989 | } |
de9a55b8 | 3990 | |
0fc1b451 | 3991 | static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode, |
de9a55b8 | 3992 | struct ext4_inode_info *ei) |
0fc1b451 AK |
3993 | { |
3994 | blkcnt_t i_blocks ; | |
8180a562 AK |
3995 | struct inode *inode = &(ei->vfs_inode); |
3996 | struct super_block *sb = inode->i_sb; | |
0fc1b451 AK |
3997 | |
3998 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, | |
3999 | EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) { | |
4000 | /* we are using combined 48 bit field */ | |
4001 | i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 | | |
4002 | le32_to_cpu(raw_inode->i_blocks_lo); | |
07a03824 | 4003 | if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) { |
8180a562 AK |
4004 | /* i_blocks represent file system block size */ |
4005 | return i_blocks << (inode->i_blkbits - 9); | |
4006 | } else { | |
4007 | return i_blocks; | |
4008 | } | |
0fc1b451 AK |
4009 | } else { |
4010 | return le32_to_cpu(raw_inode->i_blocks_lo); | |
4011 | } | |
4012 | } | |
ff9ddf7e | 4013 | |
152a7b0a TM |
4014 | static inline void ext4_iget_extra_inode(struct inode *inode, |
4015 | struct ext4_inode *raw_inode, | |
4016 | struct ext4_inode_info *ei) | |
4017 | { | |
4018 | __le32 *magic = (void *)raw_inode + | |
4019 | EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize; | |
67cf5b09 | 4020 | if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) { |
152a7b0a | 4021 | ext4_set_inode_state(inode, EXT4_STATE_XATTR); |
67cf5b09 | 4022 | ext4_find_inline_data_nolock(inode); |
f19d5870 TM |
4023 | } else |
4024 | EXT4_I(inode)->i_inline_off = 0; | |
152a7b0a TM |
4025 | } |
4026 | ||
1d1fe1ee | 4027 | struct inode *ext4_iget(struct super_block *sb, unsigned long ino) |
ac27a0ec | 4028 | { |
617ba13b MC |
4029 | struct ext4_iloc iloc; |
4030 | struct ext4_inode *raw_inode; | |
1d1fe1ee | 4031 | struct ext4_inode_info *ei; |
1d1fe1ee | 4032 | struct inode *inode; |
b436b9be | 4033 | journal_t *journal = EXT4_SB(sb)->s_journal; |
1d1fe1ee | 4034 | long ret; |
ac27a0ec | 4035 | int block; |
08cefc7a EB |
4036 | uid_t i_uid; |
4037 | gid_t i_gid; | |
ac27a0ec | 4038 | |
1d1fe1ee DH |
4039 | inode = iget_locked(sb, ino); |
4040 | if (!inode) | |
4041 | return ERR_PTR(-ENOMEM); | |
4042 | if (!(inode->i_state & I_NEW)) | |
4043 | return inode; | |
4044 | ||
4045 | ei = EXT4_I(inode); | |
7dc57615 | 4046 | iloc.bh = NULL; |
ac27a0ec | 4047 | |
1d1fe1ee DH |
4048 | ret = __ext4_get_inode_loc(inode, &iloc, 0); |
4049 | if (ret < 0) | |
ac27a0ec | 4050 | goto bad_inode; |
617ba13b | 4051 | raw_inode = ext4_raw_inode(&iloc); |
814525f4 DW |
4052 | |
4053 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { | |
4054 | ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize); | |
4055 | if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > | |
4056 | EXT4_INODE_SIZE(inode->i_sb)) { | |
4057 | EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)", | |
4058 | EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize, | |
4059 | EXT4_INODE_SIZE(inode->i_sb)); | |
4060 | ret = -EIO; | |
4061 | goto bad_inode; | |
4062 | } | |
4063 | } else | |
4064 | ei->i_extra_isize = 0; | |
4065 | ||
4066 | /* Precompute checksum seed for inode metadata */ | |
4067 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, | |
4068 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) { | |
4069 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
4070 | __u32 csum; | |
4071 | __le32 inum = cpu_to_le32(inode->i_ino); | |
4072 | __le32 gen = raw_inode->i_generation; | |
4073 | csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum, | |
4074 | sizeof(inum)); | |
4075 | ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen, | |
4076 | sizeof(gen)); | |
4077 | } | |
4078 | ||
4079 | if (!ext4_inode_csum_verify(inode, raw_inode, ei)) { | |
4080 | EXT4_ERROR_INODE(inode, "checksum invalid"); | |
4081 | ret = -EIO; | |
4082 | goto bad_inode; | |
4083 | } | |
4084 | ||
ac27a0ec | 4085 | inode->i_mode = le16_to_cpu(raw_inode->i_mode); |
08cefc7a EB |
4086 | i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low); |
4087 | i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low); | |
af5bc92d | 4088 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
08cefc7a EB |
4089 | i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16; |
4090 | i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16; | |
ac27a0ec | 4091 | } |
08cefc7a EB |
4092 | i_uid_write(inode, i_uid); |
4093 | i_gid_write(inode, i_gid); | |
bfe86848 | 4094 | set_nlink(inode, le16_to_cpu(raw_inode->i_links_count)); |
ac27a0ec | 4095 | |
353eb83c | 4096 | ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */ |
67cf5b09 | 4097 | ei->i_inline_off = 0; |
ac27a0ec DK |
4098 | ei->i_dir_start_lookup = 0; |
4099 | ei->i_dtime = le32_to_cpu(raw_inode->i_dtime); | |
4100 | /* We now have enough fields to check if the inode was active or not. | |
4101 | * This is needed because nfsd might try to access dead inodes | |
4102 | * the test is that same one that e2fsck uses | |
4103 | * NeilBrown 1999oct15 | |
4104 | */ | |
4105 | if (inode->i_nlink == 0) { | |
393d1d1d DTB |
4106 | if ((inode->i_mode == 0 || |
4107 | !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) && | |
4108 | ino != EXT4_BOOT_LOADER_INO) { | |
ac27a0ec | 4109 | /* this inode is deleted */ |
1d1fe1ee | 4110 | ret = -ESTALE; |
ac27a0ec DK |
4111 | goto bad_inode; |
4112 | } | |
4113 | /* The only unlinked inodes we let through here have | |
4114 | * valid i_mode and are being read by the orphan | |
4115 | * recovery code: that's fine, we're about to complete | |
393d1d1d DTB |
4116 | * the process of deleting those. |
4117 | * OR it is the EXT4_BOOT_LOADER_INO which is | |
4118 | * not initialized on a new filesystem. */ | |
ac27a0ec | 4119 | } |
ac27a0ec | 4120 | ei->i_flags = le32_to_cpu(raw_inode->i_flags); |
0fc1b451 | 4121 | inode->i_blocks = ext4_inode_blocks(raw_inode, ei); |
7973c0c1 | 4122 | ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo); |
a9e81742 | 4123 | if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) |
a1ddeb7e BP |
4124 | ei->i_file_acl |= |
4125 | ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32; | |
a48380f7 | 4126 | inode->i_size = ext4_isize(raw_inode); |
ac27a0ec | 4127 | ei->i_disksize = inode->i_size; |
a9e7f447 DM |
4128 | #ifdef CONFIG_QUOTA |
4129 | ei->i_reserved_quota = 0; | |
4130 | #endif | |
ac27a0ec DK |
4131 | inode->i_generation = le32_to_cpu(raw_inode->i_generation); |
4132 | ei->i_block_group = iloc.block_group; | |
a4912123 | 4133 | ei->i_last_alloc_group = ~0; |
ac27a0ec DK |
4134 | /* |
4135 | * NOTE! The in-memory inode i_data array is in little-endian order | |
4136 | * even on big-endian machines: we do NOT byteswap the block numbers! | |
4137 | */ | |
617ba13b | 4138 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
ac27a0ec DK |
4139 | ei->i_data[block] = raw_inode->i_block[block]; |
4140 | INIT_LIST_HEAD(&ei->i_orphan); | |
4141 | ||
b436b9be JK |
4142 | /* |
4143 | * Set transaction id's of transactions that have to be committed | |
4144 | * to finish f[data]sync. We set them to currently running transaction | |
4145 | * as we cannot be sure that the inode or some of its metadata isn't | |
4146 | * part of the transaction - the inode could have been reclaimed and | |
4147 | * now it is reread from disk. | |
4148 | */ | |
4149 | if (journal) { | |
4150 | transaction_t *transaction; | |
4151 | tid_t tid; | |
4152 | ||
a931da6a | 4153 | read_lock(&journal->j_state_lock); |
b436b9be JK |
4154 | if (journal->j_running_transaction) |
4155 | transaction = journal->j_running_transaction; | |
4156 | else | |
4157 | transaction = journal->j_committing_transaction; | |
4158 | if (transaction) | |
4159 | tid = transaction->t_tid; | |
4160 | else | |
4161 | tid = journal->j_commit_sequence; | |
a931da6a | 4162 | read_unlock(&journal->j_state_lock); |
b436b9be JK |
4163 | ei->i_sync_tid = tid; |
4164 | ei->i_datasync_tid = tid; | |
4165 | } | |
4166 | ||
0040d987 | 4167 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
ac27a0ec DK |
4168 | if (ei->i_extra_isize == 0) { |
4169 | /* The extra space is currently unused. Use it. */ | |
617ba13b MC |
4170 | ei->i_extra_isize = sizeof(struct ext4_inode) - |
4171 | EXT4_GOOD_OLD_INODE_SIZE; | |
ac27a0ec | 4172 | } else { |
152a7b0a | 4173 | ext4_iget_extra_inode(inode, raw_inode, ei); |
ac27a0ec | 4174 | } |
814525f4 | 4175 | } |
ac27a0ec | 4176 | |
ef7f3835 KS |
4177 | EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode); |
4178 | EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode); | |
4179 | EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode); | |
4180 | EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode); | |
4181 | ||
25ec56b5 JNC |
4182 | inode->i_version = le32_to_cpu(raw_inode->i_disk_version); |
4183 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { | |
4184 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
4185 | inode->i_version |= | |
4186 | (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32; | |
4187 | } | |
4188 | ||
c4b5a614 | 4189 | ret = 0; |
485c26ec | 4190 | if (ei->i_file_acl && |
1032988c | 4191 | !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) { |
24676da4 TT |
4192 | EXT4_ERROR_INODE(inode, "bad extended attribute block %llu", |
4193 | ei->i_file_acl); | |
485c26ec TT |
4194 | ret = -EIO; |
4195 | goto bad_inode; | |
f19d5870 TM |
4196 | } else if (!ext4_has_inline_data(inode)) { |
4197 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { | |
4198 | if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || | |
4199 | (S_ISLNK(inode->i_mode) && | |
4200 | !ext4_inode_is_fast_symlink(inode)))) | |
4201 | /* Validate extent which is part of inode */ | |
4202 | ret = ext4_ext_check_inode(inode); | |
4203 | } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || | |
4204 | (S_ISLNK(inode->i_mode) && | |
4205 | !ext4_inode_is_fast_symlink(inode))) { | |
4206 | /* Validate block references which are part of inode */ | |
4207 | ret = ext4_ind_check_inode(inode); | |
4208 | } | |
fe2c8191 | 4209 | } |
567f3e9a | 4210 | if (ret) |
de9a55b8 | 4211 | goto bad_inode; |
7a262f7c | 4212 | |
ac27a0ec | 4213 | if (S_ISREG(inode->i_mode)) { |
617ba13b MC |
4214 | inode->i_op = &ext4_file_inode_operations; |
4215 | inode->i_fop = &ext4_file_operations; | |
4216 | ext4_set_aops(inode); | |
ac27a0ec | 4217 | } else if (S_ISDIR(inode->i_mode)) { |
617ba13b MC |
4218 | inode->i_op = &ext4_dir_inode_operations; |
4219 | inode->i_fop = &ext4_dir_operations; | |
ac27a0ec | 4220 | } else if (S_ISLNK(inode->i_mode)) { |
e83c1397 | 4221 | if (ext4_inode_is_fast_symlink(inode)) { |
617ba13b | 4222 | inode->i_op = &ext4_fast_symlink_inode_operations; |
e83c1397 DG |
4223 | nd_terminate_link(ei->i_data, inode->i_size, |
4224 | sizeof(ei->i_data) - 1); | |
4225 | } else { | |
617ba13b MC |
4226 | inode->i_op = &ext4_symlink_inode_operations; |
4227 | ext4_set_aops(inode); | |
ac27a0ec | 4228 | } |
563bdd61 TT |
4229 | } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) || |
4230 | S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) { | |
617ba13b | 4231 | inode->i_op = &ext4_special_inode_operations; |
ac27a0ec DK |
4232 | if (raw_inode->i_block[0]) |
4233 | init_special_inode(inode, inode->i_mode, | |
4234 | old_decode_dev(le32_to_cpu(raw_inode->i_block[0]))); | |
4235 | else | |
4236 | init_special_inode(inode, inode->i_mode, | |
4237 | new_decode_dev(le32_to_cpu(raw_inode->i_block[1]))); | |
393d1d1d DTB |
4238 | } else if (ino == EXT4_BOOT_LOADER_INO) { |
4239 | make_bad_inode(inode); | |
563bdd61 | 4240 | } else { |
563bdd61 | 4241 | ret = -EIO; |
24676da4 | 4242 | EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode); |
563bdd61 | 4243 | goto bad_inode; |
ac27a0ec | 4244 | } |
af5bc92d | 4245 | brelse(iloc.bh); |
617ba13b | 4246 | ext4_set_inode_flags(inode); |
1d1fe1ee DH |
4247 | unlock_new_inode(inode); |
4248 | return inode; | |
ac27a0ec DK |
4249 | |
4250 | bad_inode: | |
567f3e9a | 4251 | brelse(iloc.bh); |
1d1fe1ee DH |
4252 | iget_failed(inode); |
4253 | return ERR_PTR(ret); | |
ac27a0ec DK |
4254 | } |
4255 | ||
0fc1b451 AK |
4256 | static int ext4_inode_blocks_set(handle_t *handle, |
4257 | struct ext4_inode *raw_inode, | |
4258 | struct ext4_inode_info *ei) | |
4259 | { | |
4260 | struct inode *inode = &(ei->vfs_inode); | |
4261 | u64 i_blocks = inode->i_blocks; | |
4262 | struct super_block *sb = inode->i_sb; | |
0fc1b451 AK |
4263 | |
4264 | if (i_blocks <= ~0U) { | |
4265 | /* | |
4907cb7b | 4266 | * i_blocks can be represented in a 32 bit variable |
0fc1b451 AK |
4267 | * as multiple of 512 bytes |
4268 | */ | |
8180a562 | 4269 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 4270 | raw_inode->i_blocks_high = 0; |
84a8dce2 | 4271 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
f287a1a5 TT |
4272 | return 0; |
4273 | } | |
4274 | if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) | |
4275 | return -EFBIG; | |
4276 | ||
4277 | if (i_blocks <= 0xffffffffffffULL) { | |
0fc1b451 AK |
4278 | /* |
4279 | * i_blocks can be represented in a 48 bit variable | |
4280 | * as multiple of 512 bytes | |
4281 | */ | |
8180a562 | 4282 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 4283 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); |
84a8dce2 | 4284 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
0fc1b451 | 4285 | } else { |
84a8dce2 | 4286 | ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
8180a562 AK |
4287 | /* i_block is stored in file system block size */ |
4288 | i_blocks = i_blocks >> (inode->i_blkbits - 9); | |
4289 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); | |
4290 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); | |
0fc1b451 | 4291 | } |
f287a1a5 | 4292 | return 0; |
0fc1b451 AK |
4293 | } |
4294 | ||
ac27a0ec DK |
4295 | /* |
4296 | * Post the struct inode info into an on-disk inode location in the | |
4297 | * buffer-cache. This gobbles the caller's reference to the | |
4298 | * buffer_head in the inode location struct. | |
4299 | * | |
4300 | * The caller must have write access to iloc->bh. | |
4301 | */ | |
617ba13b | 4302 | static int ext4_do_update_inode(handle_t *handle, |
ac27a0ec | 4303 | struct inode *inode, |
830156c7 | 4304 | struct ext4_iloc *iloc) |
ac27a0ec | 4305 | { |
617ba13b MC |
4306 | struct ext4_inode *raw_inode = ext4_raw_inode(iloc); |
4307 | struct ext4_inode_info *ei = EXT4_I(inode); | |
ac27a0ec DK |
4308 | struct buffer_head *bh = iloc->bh; |
4309 | int err = 0, rc, block; | |
b71fc079 | 4310 | int need_datasync = 0; |
08cefc7a EB |
4311 | uid_t i_uid; |
4312 | gid_t i_gid; | |
ac27a0ec DK |
4313 | |
4314 | /* For fields not not tracking in the in-memory inode, | |
4315 | * initialise them to zero for new inodes. */ | |
19f5fb7a | 4316 | if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) |
617ba13b | 4317 | memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size); |
ac27a0ec | 4318 | |
ff9ddf7e | 4319 | ext4_get_inode_flags(ei); |
ac27a0ec | 4320 | raw_inode->i_mode = cpu_to_le16(inode->i_mode); |
08cefc7a EB |
4321 | i_uid = i_uid_read(inode); |
4322 | i_gid = i_gid_read(inode); | |
af5bc92d | 4323 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
08cefc7a EB |
4324 | raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid)); |
4325 | raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid)); | |
ac27a0ec DK |
4326 | /* |
4327 | * Fix up interoperability with old kernels. Otherwise, old inodes get | |
4328 | * re-used with the upper 16 bits of the uid/gid intact | |
4329 | */ | |
af5bc92d | 4330 | if (!ei->i_dtime) { |
ac27a0ec | 4331 | raw_inode->i_uid_high = |
08cefc7a | 4332 | cpu_to_le16(high_16_bits(i_uid)); |
ac27a0ec | 4333 | raw_inode->i_gid_high = |
08cefc7a | 4334 | cpu_to_le16(high_16_bits(i_gid)); |
ac27a0ec DK |
4335 | } else { |
4336 | raw_inode->i_uid_high = 0; | |
4337 | raw_inode->i_gid_high = 0; | |
4338 | } | |
4339 | } else { | |
08cefc7a EB |
4340 | raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid)); |
4341 | raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid)); | |
ac27a0ec DK |
4342 | raw_inode->i_uid_high = 0; |
4343 | raw_inode->i_gid_high = 0; | |
4344 | } | |
4345 | raw_inode->i_links_count = cpu_to_le16(inode->i_nlink); | |
ef7f3835 KS |
4346 | |
4347 | EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode); | |
4348 | EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode); | |
4349 | EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode); | |
4350 | EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode); | |
4351 | ||
0fc1b451 AK |
4352 | if (ext4_inode_blocks_set(handle, raw_inode, ei)) |
4353 | goto out_brelse; | |
ac27a0ec | 4354 | raw_inode->i_dtime = cpu_to_le32(ei->i_dtime); |
353eb83c | 4355 | raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF); |
9b8f1f01 MC |
4356 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != |
4357 | cpu_to_le32(EXT4_OS_HURD)) | |
a1ddeb7e BP |
4358 | raw_inode->i_file_acl_high = |
4359 | cpu_to_le16(ei->i_file_acl >> 32); | |
7973c0c1 | 4360 | raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl); |
b71fc079 JK |
4361 | if (ei->i_disksize != ext4_isize(raw_inode)) { |
4362 | ext4_isize_set(raw_inode, ei->i_disksize); | |
4363 | need_datasync = 1; | |
4364 | } | |
a48380f7 AK |
4365 | if (ei->i_disksize > 0x7fffffffULL) { |
4366 | struct super_block *sb = inode->i_sb; | |
4367 | if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, | |
4368 | EXT4_FEATURE_RO_COMPAT_LARGE_FILE) || | |
4369 | EXT4_SB(sb)->s_es->s_rev_level == | |
4370 | cpu_to_le32(EXT4_GOOD_OLD_REV)) { | |
4371 | /* If this is the first large file | |
4372 | * created, add a flag to the superblock. | |
4373 | */ | |
4374 | err = ext4_journal_get_write_access(handle, | |
4375 | EXT4_SB(sb)->s_sbh); | |
4376 | if (err) | |
4377 | goto out_brelse; | |
4378 | ext4_update_dynamic_rev(sb); | |
4379 | EXT4_SET_RO_COMPAT_FEATURE(sb, | |
617ba13b | 4380 | EXT4_FEATURE_RO_COMPAT_LARGE_FILE); |
0390131b | 4381 | ext4_handle_sync(handle); |
b50924c2 | 4382 | err = ext4_handle_dirty_super(handle, sb); |
ac27a0ec DK |
4383 | } |
4384 | } | |
4385 | raw_inode->i_generation = cpu_to_le32(inode->i_generation); | |
4386 | if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) { | |
4387 | if (old_valid_dev(inode->i_rdev)) { | |
4388 | raw_inode->i_block[0] = | |
4389 | cpu_to_le32(old_encode_dev(inode->i_rdev)); | |
4390 | raw_inode->i_block[1] = 0; | |
4391 | } else { | |
4392 | raw_inode->i_block[0] = 0; | |
4393 | raw_inode->i_block[1] = | |
4394 | cpu_to_le32(new_encode_dev(inode->i_rdev)); | |
4395 | raw_inode->i_block[2] = 0; | |
4396 | } | |
f19d5870 | 4397 | } else if (!ext4_has_inline_data(inode)) { |
de9a55b8 TT |
4398 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
4399 | raw_inode->i_block[block] = ei->i_data[block]; | |
f19d5870 | 4400 | } |
ac27a0ec | 4401 | |
25ec56b5 JNC |
4402 | raw_inode->i_disk_version = cpu_to_le32(inode->i_version); |
4403 | if (ei->i_extra_isize) { | |
4404 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
4405 | raw_inode->i_version_hi = | |
4406 | cpu_to_le32(inode->i_version >> 32); | |
ac27a0ec | 4407 | raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize); |
25ec56b5 JNC |
4408 | } |
4409 | ||
814525f4 DW |
4410 | ext4_inode_csum_set(inode, raw_inode, ei); |
4411 | ||
830156c7 | 4412 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
73b50c1c | 4413 | rc = ext4_handle_dirty_metadata(handle, NULL, bh); |
830156c7 FM |
4414 | if (!err) |
4415 | err = rc; | |
19f5fb7a | 4416 | ext4_clear_inode_state(inode, EXT4_STATE_NEW); |
ac27a0ec | 4417 | |
b71fc079 | 4418 | ext4_update_inode_fsync_trans(handle, inode, need_datasync); |
ac27a0ec | 4419 | out_brelse: |
af5bc92d | 4420 | brelse(bh); |
617ba13b | 4421 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
4422 | return err; |
4423 | } | |
4424 | ||
4425 | /* | |
617ba13b | 4426 | * ext4_write_inode() |
ac27a0ec DK |
4427 | * |
4428 | * We are called from a few places: | |
4429 | * | |
4430 | * - Within generic_file_write() for O_SYNC files. | |
4431 | * Here, there will be no transaction running. We wait for any running | |
4907cb7b | 4432 | * transaction to commit. |
ac27a0ec DK |
4433 | * |
4434 | * - Within sys_sync(), kupdate and such. | |
4435 | * We wait on commit, if tol to. | |
4436 | * | |
4437 | * - Within prune_icache() (PF_MEMALLOC == true) | |
4438 | * Here we simply return. We can't afford to block kswapd on the | |
4439 | * journal commit. | |
4440 | * | |
4441 | * In all cases it is actually safe for us to return without doing anything, | |
4442 | * because the inode has been copied into a raw inode buffer in | |
617ba13b | 4443 | * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for |
ac27a0ec DK |
4444 | * knfsd. |
4445 | * | |
4446 | * Note that we are absolutely dependent upon all inode dirtiers doing the | |
4447 | * right thing: they *must* call mark_inode_dirty() after dirtying info in | |
4448 | * which we are interested. | |
4449 | * | |
4450 | * It would be a bug for them to not do this. The code: | |
4451 | * | |
4452 | * mark_inode_dirty(inode) | |
4453 | * stuff(); | |
4454 | * inode->i_size = expr; | |
4455 | * | |
4456 | * is in error because a kswapd-driven write_inode() could occur while | |
4457 | * `stuff()' is running, and the new i_size will be lost. Plus the inode | |
4458 | * will no longer be on the superblock's dirty inode list. | |
4459 | */ | |
a9185b41 | 4460 | int ext4_write_inode(struct inode *inode, struct writeback_control *wbc) |
ac27a0ec | 4461 | { |
91ac6f43 FM |
4462 | int err; |
4463 | ||
ac27a0ec DK |
4464 | if (current->flags & PF_MEMALLOC) |
4465 | return 0; | |
4466 | ||
91ac6f43 FM |
4467 | if (EXT4_SB(inode->i_sb)->s_journal) { |
4468 | if (ext4_journal_current_handle()) { | |
4469 | jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n"); | |
4470 | dump_stack(); | |
4471 | return -EIO; | |
4472 | } | |
ac27a0ec | 4473 | |
a9185b41 | 4474 | if (wbc->sync_mode != WB_SYNC_ALL) |
91ac6f43 FM |
4475 | return 0; |
4476 | ||
4477 | err = ext4_force_commit(inode->i_sb); | |
4478 | } else { | |
4479 | struct ext4_iloc iloc; | |
ac27a0ec | 4480 | |
8b472d73 | 4481 | err = __ext4_get_inode_loc(inode, &iloc, 0); |
91ac6f43 FM |
4482 | if (err) |
4483 | return err; | |
a9185b41 | 4484 | if (wbc->sync_mode == WB_SYNC_ALL) |
830156c7 FM |
4485 | sync_dirty_buffer(iloc.bh); |
4486 | if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) { | |
c398eda0 TT |
4487 | EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr, |
4488 | "IO error syncing inode"); | |
830156c7 FM |
4489 | err = -EIO; |
4490 | } | |
fd2dd9fb | 4491 | brelse(iloc.bh); |
91ac6f43 FM |
4492 | } |
4493 | return err; | |
ac27a0ec DK |
4494 | } |
4495 | ||
53e87268 JK |
4496 | /* |
4497 | * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate | |
4498 | * buffers that are attached to a page stradding i_size and are undergoing | |
4499 | * commit. In that case we have to wait for commit to finish and try again. | |
4500 | */ | |
4501 | static void ext4_wait_for_tail_page_commit(struct inode *inode) | |
4502 | { | |
4503 | struct page *page; | |
4504 | unsigned offset; | |
4505 | journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; | |
4506 | tid_t commit_tid = 0; | |
4507 | int ret; | |
4508 | ||
4509 | offset = inode->i_size & (PAGE_CACHE_SIZE - 1); | |
4510 | /* | |
4511 | * All buffers in the last page remain valid? Then there's nothing to | |
4512 | * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE == | |
4513 | * blocksize case | |
4514 | */ | |
4515 | if (offset > PAGE_CACHE_SIZE - (1 << inode->i_blkbits)) | |
4516 | return; | |
4517 | while (1) { | |
4518 | page = find_lock_page(inode->i_mapping, | |
4519 | inode->i_size >> PAGE_CACHE_SHIFT); | |
4520 | if (!page) | |
4521 | return; | |
ca99fdd2 LC |
4522 | ret = __ext4_journalled_invalidatepage(page, offset, |
4523 | PAGE_CACHE_SIZE - offset); | |
53e87268 JK |
4524 | unlock_page(page); |
4525 | page_cache_release(page); | |
4526 | if (ret != -EBUSY) | |
4527 | return; | |
4528 | commit_tid = 0; | |
4529 | read_lock(&journal->j_state_lock); | |
4530 | if (journal->j_committing_transaction) | |
4531 | commit_tid = journal->j_committing_transaction->t_tid; | |
4532 | read_unlock(&journal->j_state_lock); | |
4533 | if (commit_tid) | |
4534 | jbd2_log_wait_commit(journal, commit_tid); | |
4535 | } | |
4536 | } | |
4537 | ||
ac27a0ec | 4538 | /* |
617ba13b | 4539 | * ext4_setattr() |
ac27a0ec DK |
4540 | * |
4541 | * Called from notify_change. | |
4542 | * | |
4543 | * We want to trap VFS attempts to truncate the file as soon as | |
4544 | * possible. In particular, we want to make sure that when the VFS | |
4545 | * shrinks i_size, we put the inode on the orphan list and modify | |
4546 | * i_disksize immediately, so that during the subsequent flushing of | |
4547 | * dirty pages and freeing of disk blocks, we can guarantee that any | |
4548 | * commit will leave the blocks being flushed in an unused state on | |
4549 | * disk. (On recovery, the inode will get truncated and the blocks will | |
4550 | * be freed, so we have a strong guarantee that no future commit will | |
4551 | * leave these blocks visible to the user.) | |
4552 | * | |
678aaf48 JK |
4553 | * Another thing we have to assure is that if we are in ordered mode |
4554 | * and inode is still attached to the committing transaction, we must | |
4555 | * we start writeout of all the dirty pages which are being truncated. | |
4556 | * This way we are sure that all the data written in the previous | |
4557 | * transaction are already on disk (truncate waits for pages under | |
4558 | * writeback). | |
4559 | * | |
4560 | * Called with inode->i_mutex down. | |
ac27a0ec | 4561 | */ |
617ba13b | 4562 | int ext4_setattr(struct dentry *dentry, struct iattr *attr) |
ac27a0ec DK |
4563 | { |
4564 | struct inode *inode = dentry->d_inode; | |
4565 | int error, rc = 0; | |
3d287de3 | 4566 | int orphan = 0; |
ac27a0ec DK |
4567 | const unsigned int ia_valid = attr->ia_valid; |
4568 | ||
4569 | error = inode_change_ok(inode, attr); | |
4570 | if (error) | |
4571 | return error; | |
4572 | ||
12755627 | 4573 | if (is_quota_modification(inode, attr)) |
871a2931 | 4574 | dquot_initialize(inode); |
08cefc7a EB |
4575 | if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) || |
4576 | (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) { | |
ac27a0ec DK |
4577 | handle_t *handle; |
4578 | ||
4579 | /* (user+group)*(old+new) structure, inode write (sb, | |
4580 | * inode block, ? - but truncate inode update has it) */ | |
9924a92a TT |
4581 | handle = ext4_journal_start(inode, EXT4_HT_QUOTA, |
4582 | (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) + | |
4583 | EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3); | |
ac27a0ec DK |
4584 | if (IS_ERR(handle)) { |
4585 | error = PTR_ERR(handle); | |
4586 | goto err_out; | |
4587 | } | |
b43fa828 | 4588 | error = dquot_transfer(inode, attr); |
ac27a0ec | 4589 | if (error) { |
617ba13b | 4590 | ext4_journal_stop(handle); |
ac27a0ec DK |
4591 | return error; |
4592 | } | |
4593 | /* Update corresponding info in inode so that everything is in | |
4594 | * one transaction */ | |
4595 | if (attr->ia_valid & ATTR_UID) | |
4596 | inode->i_uid = attr->ia_uid; | |
4597 | if (attr->ia_valid & ATTR_GID) | |
4598 | inode->i_gid = attr->ia_gid; | |
617ba13b MC |
4599 | error = ext4_mark_inode_dirty(handle, inode); |
4600 | ext4_journal_stop(handle); | |
ac27a0ec DK |
4601 | } |
4602 | ||
e2b46574 | 4603 | if (attr->ia_valid & ATTR_SIZE) { |
562c72aa | 4604 | |
12e9b892 | 4605 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) { |
e2b46574 ES |
4606 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
4607 | ||
0c095c7f TT |
4608 | if (attr->ia_size > sbi->s_bitmap_maxbytes) |
4609 | return -EFBIG; | |
e2b46574 ES |
4610 | } |
4611 | } | |
4612 | ||
ac27a0ec | 4613 | if (S_ISREG(inode->i_mode) && |
c8d46e41 | 4614 | attr->ia_valid & ATTR_SIZE && |
072bd7ea | 4615 | (attr->ia_size < inode->i_size)) { |
ac27a0ec DK |
4616 | handle_t *handle; |
4617 | ||
9924a92a | 4618 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 3); |
ac27a0ec DK |
4619 | if (IS_ERR(handle)) { |
4620 | error = PTR_ERR(handle); | |
4621 | goto err_out; | |
4622 | } | |
3d287de3 DM |
4623 | if (ext4_handle_valid(handle)) { |
4624 | error = ext4_orphan_add(handle, inode); | |
4625 | orphan = 1; | |
4626 | } | |
617ba13b MC |
4627 | EXT4_I(inode)->i_disksize = attr->ia_size; |
4628 | rc = ext4_mark_inode_dirty(handle, inode); | |
ac27a0ec DK |
4629 | if (!error) |
4630 | error = rc; | |
617ba13b | 4631 | ext4_journal_stop(handle); |
678aaf48 JK |
4632 | |
4633 | if (ext4_should_order_data(inode)) { | |
4634 | error = ext4_begin_ordered_truncate(inode, | |
4635 | attr->ia_size); | |
4636 | if (error) { | |
4637 | /* Do as much error cleanup as possible */ | |
9924a92a TT |
4638 | handle = ext4_journal_start(inode, |
4639 | EXT4_HT_INODE, 3); | |
678aaf48 JK |
4640 | if (IS_ERR(handle)) { |
4641 | ext4_orphan_del(NULL, inode); | |
4642 | goto err_out; | |
4643 | } | |
4644 | ext4_orphan_del(handle, inode); | |
3d287de3 | 4645 | orphan = 0; |
678aaf48 JK |
4646 | ext4_journal_stop(handle); |
4647 | goto err_out; | |
4648 | } | |
4649 | } | |
ac27a0ec DK |
4650 | } |
4651 | ||
072bd7ea | 4652 | if (attr->ia_valid & ATTR_SIZE) { |
53e87268 JK |
4653 | if (attr->ia_size != inode->i_size) { |
4654 | loff_t oldsize = inode->i_size; | |
4655 | ||
4656 | i_size_write(inode, attr->ia_size); | |
4657 | /* | |
4658 | * Blocks are going to be removed from the inode. Wait | |
4659 | * for dio in flight. Temporarily disable | |
4660 | * dioread_nolock to prevent livelock. | |
4661 | */ | |
1b65007e | 4662 | if (orphan) { |
53e87268 JK |
4663 | if (!ext4_should_journal_data(inode)) { |
4664 | ext4_inode_block_unlocked_dio(inode); | |
4665 | inode_dio_wait(inode); | |
4666 | ext4_inode_resume_unlocked_dio(inode); | |
4667 | } else | |
4668 | ext4_wait_for_tail_page_commit(inode); | |
1b65007e | 4669 | } |
53e87268 JK |
4670 | /* |
4671 | * Truncate pagecache after we've waited for commit | |
4672 | * in data=journal mode to make pages freeable. | |
4673 | */ | |
4674 | truncate_pagecache(inode, oldsize, inode->i_size); | |
1c9114f9 | 4675 | } |
afcff5d8 | 4676 | ext4_truncate(inode); |
072bd7ea | 4677 | } |
ac27a0ec | 4678 | |
1025774c CH |
4679 | if (!rc) { |
4680 | setattr_copy(inode, attr); | |
4681 | mark_inode_dirty(inode); | |
4682 | } | |
4683 | ||
4684 | /* | |
4685 | * If the call to ext4_truncate failed to get a transaction handle at | |
4686 | * all, we need to clean up the in-core orphan list manually. | |
4687 | */ | |
3d287de3 | 4688 | if (orphan && inode->i_nlink) |
617ba13b | 4689 | ext4_orphan_del(NULL, inode); |
ac27a0ec DK |
4690 | |
4691 | if (!rc && (ia_valid & ATTR_MODE)) | |
617ba13b | 4692 | rc = ext4_acl_chmod(inode); |
ac27a0ec DK |
4693 | |
4694 | err_out: | |
617ba13b | 4695 | ext4_std_error(inode->i_sb, error); |
ac27a0ec DK |
4696 | if (!error) |
4697 | error = rc; | |
4698 | return error; | |
4699 | } | |
4700 | ||
3e3398a0 MC |
4701 | int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry, |
4702 | struct kstat *stat) | |
4703 | { | |
4704 | struct inode *inode; | |
8af8eecc | 4705 | unsigned long long delalloc_blocks; |
3e3398a0 MC |
4706 | |
4707 | inode = dentry->d_inode; | |
4708 | generic_fillattr(inode, stat); | |
4709 | ||
4710 | /* | |
4711 | * We can't update i_blocks if the block allocation is delayed | |
4712 | * otherwise in the case of system crash before the real block | |
4713 | * allocation is done, we will have i_blocks inconsistent with | |
4714 | * on-disk file blocks. | |
4715 | * We always keep i_blocks updated together with real | |
4716 | * allocation. But to not confuse with user, stat | |
4717 | * will return the blocks that include the delayed allocation | |
4718 | * blocks for this file. | |
4719 | */ | |
96607551 TM |
4720 | delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb), |
4721 | EXT4_I(inode)->i_reserved_data_blocks); | |
3e3398a0 | 4722 | |
8af8eecc | 4723 | stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits-9); |
3e3398a0 MC |
4724 | return 0; |
4725 | } | |
ac27a0ec | 4726 | |
a02908f1 MC |
4727 | static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk) |
4728 | { | |
12e9b892 | 4729 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) |
8bb2b247 | 4730 | return ext4_ind_trans_blocks(inode, nrblocks, chunk); |
ac51d837 | 4731 | return ext4_ext_index_trans_blocks(inode, nrblocks, chunk); |
a02908f1 | 4732 | } |
ac51d837 | 4733 | |
ac27a0ec | 4734 | /* |
a02908f1 MC |
4735 | * Account for index blocks, block groups bitmaps and block group |
4736 | * descriptor blocks if modify datablocks and index blocks | |
4737 | * worse case, the indexs blocks spread over different block groups | |
ac27a0ec | 4738 | * |
a02908f1 | 4739 | * If datablocks are discontiguous, they are possible to spread over |
4907cb7b | 4740 | * different block groups too. If they are contiguous, with flexbg, |
a02908f1 | 4741 | * they could still across block group boundary. |
ac27a0ec | 4742 | * |
a02908f1 MC |
4743 | * Also account for superblock, inode, quota and xattr blocks |
4744 | */ | |
1f109d5a | 4745 | static int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk) |
a02908f1 | 4746 | { |
8df9675f TT |
4747 | ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb); |
4748 | int gdpblocks; | |
a02908f1 MC |
4749 | int idxblocks; |
4750 | int ret = 0; | |
4751 | ||
4752 | /* | |
4753 | * How many index blocks need to touch to modify nrblocks? | |
4754 | * The "Chunk" flag indicating whether the nrblocks is | |
4755 | * physically contiguous on disk | |
4756 | * | |
4757 | * For Direct IO and fallocate, they calls get_block to allocate | |
4758 | * one single extent at a time, so they could set the "Chunk" flag | |
4759 | */ | |
4760 | idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk); | |
4761 | ||
4762 | ret = idxblocks; | |
4763 | ||
4764 | /* | |
4765 | * Now let's see how many group bitmaps and group descriptors need | |
4766 | * to account | |
4767 | */ | |
4768 | groups = idxblocks; | |
4769 | if (chunk) | |
4770 | groups += 1; | |
4771 | else | |
4772 | groups += nrblocks; | |
4773 | ||
4774 | gdpblocks = groups; | |
8df9675f TT |
4775 | if (groups > ngroups) |
4776 | groups = ngroups; | |
a02908f1 MC |
4777 | if (groups > EXT4_SB(inode->i_sb)->s_gdb_count) |
4778 | gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count; | |
4779 | ||
4780 | /* bitmaps and block group descriptor blocks */ | |
4781 | ret += groups + gdpblocks; | |
4782 | ||
4783 | /* Blocks for super block, inode, quota and xattr blocks */ | |
4784 | ret += EXT4_META_TRANS_BLOCKS(inode->i_sb); | |
4785 | ||
4786 | return ret; | |
4787 | } | |
4788 | ||
4789 | /* | |
25985edc | 4790 | * Calculate the total number of credits to reserve to fit |
f3bd1f3f MC |
4791 | * the modification of a single pages into a single transaction, |
4792 | * which may include multiple chunks of block allocations. | |
ac27a0ec | 4793 | * |
525f4ed8 | 4794 | * This could be called via ext4_write_begin() |
ac27a0ec | 4795 | * |
525f4ed8 | 4796 | * We need to consider the worse case, when |
a02908f1 | 4797 | * one new block per extent. |
ac27a0ec | 4798 | */ |
a86c6181 | 4799 | int ext4_writepage_trans_blocks(struct inode *inode) |
ac27a0ec | 4800 | { |
617ba13b | 4801 | int bpp = ext4_journal_blocks_per_page(inode); |
ac27a0ec DK |
4802 | int ret; |
4803 | ||
a02908f1 | 4804 | ret = ext4_meta_trans_blocks(inode, bpp, 0); |
a86c6181 | 4805 | |
a02908f1 | 4806 | /* Account for data blocks for journalled mode */ |
617ba13b | 4807 | if (ext4_should_journal_data(inode)) |
a02908f1 | 4808 | ret += bpp; |
ac27a0ec DK |
4809 | return ret; |
4810 | } | |
f3bd1f3f MC |
4811 | |
4812 | /* | |
4813 | * Calculate the journal credits for a chunk of data modification. | |
4814 | * | |
4815 | * This is called from DIO, fallocate or whoever calling | |
79e83036 | 4816 | * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks. |
f3bd1f3f MC |
4817 | * |
4818 | * journal buffers for data blocks are not included here, as DIO | |
4819 | * and fallocate do no need to journal data buffers. | |
4820 | */ | |
4821 | int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks) | |
4822 | { | |
4823 | return ext4_meta_trans_blocks(inode, nrblocks, 1); | |
4824 | } | |
4825 | ||
ac27a0ec | 4826 | /* |
617ba13b | 4827 | * The caller must have previously called ext4_reserve_inode_write(). |
ac27a0ec DK |
4828 | * Give this, we know that the caller already has write access to iloc->bh. |
4829 | */ | |
617ba13b | 4830 | int ext4_mark_iloc_dirty(handle_t *handle, |
de9a55b8 | 4831 | struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec DK |
4832 | { |
4833 | int err = 0; | |
4834 | ||
c64db50e | 4835 | if (IS_I_VERSION(inode)) |
25ec56b5 JNC |
4836 | inode_inc_iversion(inode); |
4837 | ||
ac27a0ec DK |
4838 | /* the do_update_inode consumes one bh->b_count */ |
4839 | get_bh(iloc->bh); | |
4840 | ||
dab291af | 4841 | /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */ |
830156c7 | 4842 | err = ext4_do_update_inode(handle, inode, iloc); |
ac27a0ec DK |
4843 | put_bh(iloc->bh); |
4844 | return err; | |
4845 | } | |
4846 | ||
4847 | /* | |
4848 | * On success, We end up with an outstanding reference count against | |
4849 | * iloc->bh. This _must_ be cleaned up later. | |
4850 | */ | |
4851 | ||
4852 | int | |
617ba13b MC |
4853 | ext4_reserve_inode_write(handle_t *handle, struct inode *inode, |
4854 | struct ext4_iloc *iloc) | |
ac27a0ec | 4855 | { |
0390131b FM |
4856 | int err; |
4857 | ||
4858 | err = ext4_get_inode_loc(inode, iloc); | |
4859 | if (!err) { | |
4860 | BUFFER_TRACE(iloc->bh, "get_write_access"); | |
4861 | err = ext4_journal_get_write_access(handle, iloc->bh); | |
4862 | if (err) { | |
4863 | brelse(iloc->bh); | |
4864 | iloc->bh = NULL; | |
ac27a0ec DK |
4865 | } |
4866 | } | |
617ba13b | 4867 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
4868 | return err; |
4869 | } | |
4870 | ||
6dd4ee7c KS |
4871 | /* |
4872 | * Expand an inode by new_extra_isize bytes. | |
4873 | * Returns 0 on success or negative error number on failure. | |
4874 | */ | |
1d03ec98 AK |
4875 | static int ext4_expand_extra_isize(struct inode *inode, |
4876 | unsigned int new_extra_isize, | |
4877 | struct ext4_iloc iloc, | |
4878 | handle_t *handle) | |
6dd4ee7c KS |
4879 | { |
4880 | struct ext4_inode *raw_inode; | |
4881 | struct ext4_xattr_ibody_header *header; | |
6dd4ee7c KS |
4882 | |
4883 | if (EXT4_I(inode)->i_extra_isize >= new_extra_isize) | |
4884 | return 0; | |
4885 | ||
4886 | raw_inode = ext4_raw_inode(&iloc); | |
4887 | ||
4888 | header = IHDR(inode, raw_inode); | |
6dd4ee7c KS |
4889 | |
4890 | /* No extended attributes present */ | |
19f5fb7a TT |
4891 | if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) || |
4892 | header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) { | |
6dd4ee7c KS |
4893 | memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0, |
4894 | new_extra_isize); | |
4895 | EXT4_I(inode)->i_extra_isize = new_extra_isize; | |
4896 | return 0; | |
4897 | } | |
4898 | ||
4899 | /* try to expand with EAs present */ | |
4900 | return ext4_expand_extra_isize_ea(inode, new_extra_isize, | |
4901 | raw_inode, handle); | |
4902 | } | |
4903 | ||
ac27a0ec DK |
4904 | /* |
4905 | * What we do here is to mark the in-core inode as clean with respect to inode | |
4906 | * dirtiness (it may still be data-dirty). | |
4907 | * This means that the in-core inode may be reaped by prune_icache | |
4908 | * without having to perform any I/O. This is a very good thing, | |
4909 | * because *any* task may call prune_icache - even ones which | |
4910 | * have a transaction open against a different journal. | |
4911 | * | |
4912 | * Is this cheating? Not really. Sure, we haven't written the | |
4913 | * inode out, but prune_icache isn't a user-visible syncing function. | |
4914 | * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync) | |
4915 | * we start and wait on commits. | |
ac27a0ec | 4916 | */ |
617ba13b | 4917 | int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode) |
ac27a0ec | 4918 | { |
617ba13b | 4919 | struct ext4_iloc iloc; |
6dd4ee7c KS |
4920 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
4921 | static unsigned int mnt_count; | |
4922 | int err, ret; | |
ac27a0ec DK |
4923 | |
4924 | might_sleep(); | |
7ff9c073 | 4925 | trace_ext4_mark_inode_dirty(inode, _RET_IP_); |
617ba13b | 4926 | err = ext4_reserve_inode_write(handle, inode, &iloc); |
0390131b FM |
4927 | if (ext4_handle_valid(handle) && |
4928 | EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize && | |
19f5fb7a | 4929 | !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) { |
6dd4ee7c KS |
4930 | /* |
4931 | * We need extra buffer credits since we may write into EA block | |
4932 | * with this same handle. If journal_extend fails, then it will | |
4933 | * only result in a minor loss of functionality for that inode. | |
4934 | * If this is felt to be critical, then e2fsck should be run to | |
4935 | * force a large enough s_min_extra_isize. | |
4936 | */ | |
4937 | if ((jbd2_journal_extend(handle, | |
4938 | EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) { | |
4939 | ret = ext4_expand_extra_isize(inode, | |
4940 | sbi->s_want_extra_isize, | |
4941 | iloc, handle); | |
4942 | if (ret) { | |
19f5fb7a TT |
4943 | ext4_set_inode_state(inode, |
4944 | EXT4_STATE_NO_EXPAND); | |
c1bddad9 AK |
4945 | if (mnt_count != |
4946 | le16_to_cpu(sbi->s_es->s_mnt_count)) { | |
12062ddd | 4947 | ext4_warning(inode->i_sb, |
6dd4ee7c KS |
4948 | "Unable to expand inode %lu. Delete" |
4949 | " some EAs or run e2fsck.", | |
4950 | inode->i_ino); | |
c1bddad9 AK |
4951 | mnt_count = |
4952 | le16_to_cpu(sbi->s_es->s_mnt_count); | |
6dd4ee7c KS |
4953 | } |
4954 | } | |
4955 | } | |
4956 | } | |
ac27a0ec | 4957 | if (!err) |
617ba13b | 4958 | err = ext4_mark_iloc_dirty(handle, inode, &iloc); |
ac27a0ec DK |
4959 | return err; |
4960 | } | |
4961 | ||
4962 | /* | |
617ba13b | 4963 | * ext4_dirty_inode() is called from __mark_inode_dirty() |
ac27a0ec DK |
4964 | * |
4965 | * We're really interested in the case where a file is being extended. | |
4966 | * i_size has been changed by generic_commit_write() and we thus need | |
4967 | * to include the updated inode in the current transaction. | |
4968 | * | |
5dd4056d | 4969 | * Also, dquot_alloc_block() will always dirty the inode when blocks |
ac27a0ec DK |
4970 | * are allocated to the file. |
4971 | * | |
4972 | * If the inode is marked synchronous, we don't honour that here - doing | |
4973 | * so would cause a commit on atime updates, which we don't bother doing. | |
4974 | * We handle synchronous inodes at the highest possible level. | |
4975 | */ | |
aa385729 | 4976 | void ext4_dirty_inode(struct inode *inode, int flags) |
ac27a0ec | 4977 | { |
ac27a0ec DK |
4978 | handle_t *handle; |
4979 | ||
9924a92a | 4980 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 2); |
ac27a0ec DK |
4981 | if (IS_ERR(handle)) |
4982 | goto out; | |
f3dc272f | 4983 | |
f3dc272f CW |
4984 | ext4_mark_inode_dirty(handle, inode); |
4985 | ||
617ba13b | 4986 | ext4_journal_stop(handle); |
ac27a0ec DK |
4987 | out: |
4988 | return; | |
4989 | } | |
4990 | ||
4991 | #if 0 | |
4992 | /* | |
4993 | * Bind an inode's backing buffer_head into this transaction, to prevent | |
4994 | * it from being flushed to disk early. Unlike | |
617ba13b | 4995 | * ext4_reserve_inode_write, this leaves behind no bh reference and |
ac27a0ec DK |
4996 | * returns no iloc structure, so the caller needs to repeat the iloc |
4997 | * lookup to mark the inode dirty later. | |
4998 | */ | |
617ba13b | 4999 | static int ext4_pin_inode(handle_t *handle, struct inode *inode) |
ac27a0ec | 5000 | { |
617ba13b | 5001 | struct ext4_iloc iloc; |
ac27a0ec DK |
5002 | |
5003 | int err = 0; | |
5004 | if (handle) { | |
617ba13b | 5005 | err = ext4_get_inode_loc(inode, &iloc); |
ac27a0ec DK |
5006 | if (!err) { |
5007 | BUFFER_TRACE(iloc.bh, "get_write_access"); | |
dab291af | 5008 | err = jbd2_journal_get_write_access(handle, iloc.bh); |
ac27a0ec | 5009 | if (!err) |
0390131b | 5010 | err = ext4_handle_dirty_metadata(handle, |
73b50c1c | 5011 | NULL, |
0390131b | 5012 | iloc.bh); |
ac27a0ec DK |
5013 | brelse(iloc.bh); |
5014 | } | |
5015 | } | |
617ba13b | 5016 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
5017 | return err; |
5018 | } | |
5019 | #endif | |
5020 | ||
617ba13b | 5021 | int ext4_change_inode_journal_flag(struct inode *inode, int val) |
ac27a0ec DK |
5022 | { |
5023 | journal_t *journal; | |
5024 | handle_t *handle; | |
5025 | int err; | |
5026 | ||
5027 | /* | |
5028 | * We have to be very careful here: changing a data block's | |
5029 | * journaling status dynamically is dangerous. If we write a | |
5030 | * data block to the journal, change the status and then delete | |
5031 | * that block, we risk forgetting to revoke the old log record | |
5032 | * from the journal and so a subsequent replay can corrupt data. | |
5033 | * So, first we make sure that the journal is empty and that | |
5034 | * nobody is changing anything. | |
5035 | */ | |
5036 | ||
617ba13b | 5037 | journal = EXT4_JOURNAL(inode); |
0390131b FM |
5038 | if (!journal) |
5039 | return 0; | |
d699594d | 5040 | if (is_journal_aborted(journal)) |
ac27a0ec | 5041 | return -EROFS; |
2aff57b0 YY |
5042 | /* We have to allocate physical blocks for delalloc blocks |
5043 | * before flushing journal. otherwise delalloc blocks can not | |
5044 | * be allocated any more. even more truncate on delalloc blocks | |
5045 | * could trigger BUG by flushing delalloc blocks in journal. | |
5046 | * There is no delalloc block in non-journal data mode. | |
5047 | */ | |
5048 | if (val && test_opt(inode->i_sb, DELALLOC)) { | |
5049 | err = ext4_alloc_da_blocks(inode); | |
5050 | if (err < 0) | |
5051 | return err; | |
5052 | } | |
ac27a0ec | 5053 | |
17335dcc DM |
5054 | /* Wait for all existing dio workers */ |
5055 | ext4_inode_block_unlocked_dio(inode); | |
5056 | inode_dio_wait(inode); | |
5057 | ||
dab291af | 5058 | jbd2_journal_lock_updates(journal); |
ac27a0ec DK |
5059 | |
5060 | /* | |
5061 | * OK, there are no updates running now, and all cached data is | |
5062 | * synced to disk. We are now in a completely consistent state | |
5063 | * which doesn't have anything in the journal, and we know that | |
5064 | * no filesystem updates are running, so it is safe to modify | |
5065 | * the inode's in-core data-journaling state flag now. | |
5066 | */ | |
5067 | ||
5068 | if (val) | |
12e9b892 | 5069 | ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
5872ddaa YY |
5070 | else { |
5071 | jbd2_journal_flush(journal); | |
12e9b892 | 5072 | ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
5872ddaa | 5073 | } |
617ba13b | 5074 | ext4_set_aops(inode); |
ac27a0ec | 5075 | |
dab291af | 5076 | jbd2_journal_unlock_updates(journal); |
17335dcc | 5077 | ext4_inode_resume_unlocked_dio(inode); |
ac27a0ec DK |
5078 | |
5079 | /* Finally we can mark the inode as dirty. */ | |
5080 | ||
9924a92a | 5081 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 1); |
ac27a0ec DK |
5082 | if (IS_ERR(handle)) |
5083 | return PTR_ERR(handle); | |
5084 | ||
617ba13b | 5085 | err = ext4_mark_inode_dirty(handle, inode); |
0390131b | 5086 | ext4_handle_sync(handle); |
617ba13b MC |
5087 | ext4_journal_stop(handle); |
5088 | ext4_std_error(inode->i_sb, err); | |
ac27a0ec DK |
5089 | |
5090 | return err; | |
5091 | } | |
2e9ee850 AK |
5092 | |
5093 | static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh) | |
5094 | { | |
5095 | return !buffer_mapped(bh); | |
5096 | } | |
5097 | ||
c2ec175c | 5098 | int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
2e9ee850 | 5099 | { |
c2ec175c | 5100 | struct page *page = vmf->page; |
2e9ee850 AK |
5101 | loff_t size; |
5102 | unsigned long len; | |
9ea7df53 | 5103 | int ret; |
2e9ee850 | 5104 | struct file *file = vma->vm_file; |
496ad9aa | 5105 | struct inode *inode = file_inode(file); |
2e9ee850 | 5106 | struct address_space *mapping = inode->i_mapping; |
9ea7df53 JK |
5107 | handle_t *handle; |
5108 | get_block_t *get_block; | |
5109 | int retries = 0; | |
2e9ee850 | 5110 | |
8e8ad8a5 | 5111 | sb_start_pagefault(inode->i_sb); |
041bbb6d | 5112 | file_update_time(vma->vm_file); |
9ea7df53 JK |
5113 | /* Delalloc case is easy... */ |
5114 | if (test_opt(inode->i_sb, DELALLOC) && | |
5115 | !ext4_should_journal_data(inode) && | |
5116 | !ext4_nonda_switch(inode->i_sb)) { | |
5117 | do { | |
5118 | ret = __block_page_mkwrite(vma, vmf, | |
5119 | ext4_da_get_block_prep); | |
5120 | } while (ret == -ENOSPC && | |
5121 | ext4_should_retry_alloc(inode->i_sb, &retries)); | |
5122 | goto out_ret; | |
2e9ee850 | 5123 | } |
0e499890 DW |
5124 | |
5125 | lock_page(page); | |
9ea7df53 JK |
5126 | size = i_size_read(inode); |
5127 | /* Page got truncated from under us? */ | |
5128 | if (page->mapping != mapping || page_offset(page) > size) { | |
5129 | unlock_page(page); | |
5130 | ret = VM_FAULT_NOPAGE; | |
5131 | goto out; | |
0e499890 | 5132 | } |
2e9ee850 AK |
5133 | |
5134 | if (page->index == size >> PAGE_CACHE_SHIFT) | |
5135 | len = size & ~PAGE_CACHE_MASK; | |
5136 | else | |
5137 | len = PAGE_CACHE_SIZE; | |
a827eaff | 5138 | /* |
9ea7df53 JK |
5139 | * Return if we have all the buffers mapped. This avoids the need to do |
5140 | * journal_start/journal_stop which can block and take a long time | |
a827eaff | 5141 | */ |
2e9ee850 | 5142 | if (page_has_buffers(page)) { |
f19d5870 TM |
5143 | if (!ext4_walk_page_buffers(NULL, page_buffers(page), |
5144 | 0, len, NULL, | |
5145 | ext4_bh_unmapped)) { | |
9ea7df53 | 5146 | /* Wait so that we don't change page under IO */ |
1d1d1a76 | 5147 | wait_for_stable_page(page); |
9ea7df53 JK |
5148 | ret = VM_FAULT_LOCKED; |
5149 | goto out; | |
a827eaff | 5150 | } |
2e9ee850 | 5151 | } |
a827eaff | 5152 | unlock_page(page); |
9ea7df53 JK |
5153 | /* OK, we need to fill the hole... */ |
5154 | if (ext4_should_dioread_nolock(inode)) | |
5155 | get_block = ext4_get_block_write; | |
5156 | else | |
5157 | get_block = ext4_get_block; | |
5158 | retry_alloc: | |
9924a92a TT |
5159 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
5160 | ext4_writepage_trans_blocks(inode)); | |
9ea7df53 | 5161 | if (IS_ERR(handle)) { |
c2ec175c | 5162 | ret = VM_FAULT_SIGBUS; |
9ea7df53 JK |
5163 | goto out; |
5164 | } | |
5165 | ret = __block_page_mkwrite(vma, vmf, get_block); | |
5166 | if (!ret && ext4_should_journal_data(inode)) { | |
f19d5870 | 5167 | if (ext4_walk_page_buffers(handle, page_buffers(page), 0, |
9ea7df53 JK |
5168 | PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) { |
5169 | unlock_page(page); | |
5170 | ret = VM_FAULT_SIGBUS; | |
fcbb5515 | 5171 | ext4_journal_stop(handle); |
9ea7df53 JK |
5172 | goto out; |
5173 | } | |
5174 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); | |
5175 | } | |
5176 | ext4_journal_stop(handle); | |
5177 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) | |
5178 | goto retry_alloc; | |
5179 | out_ret: | |
5180 | ret = block_page_mkwrite_return(ret); | |
5181 | out: | |
8e8ad8a5 | 5182 | sb_end_pagefault(inode->i_sb); |
2e9ee850 AK |
5183 | return ret; |
5184 | } |