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> |
00a1a053 | 41 | #include <linux/bitops.h> |
9bffad1e | 42 | |
3dcf5451 | 43 | #include "ext4_jbd2.h" |
ac27a0ec DK |
44 | #include "xattr.h" |
45 | #include "acl.h" | |
9f125d64 | 46 | #include "truncate.h" |
ac27a0ec | 47 | |
9bffad1e TT |
48 | #include <trace/events/ext4.h> |
49 | ||
a1d6cc56 AK |
50 | #define MPAGE_DA_EXTENT_TAIL 0x01 |
51 | ||
814525f4 DW |
52 | static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw, |
53 | struct ext4_inode_info *ei) | |
54 | { | |
55 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
56 | __u16 csum_lo; | |
57 | __u16 csum_hi = 0; | |
58 | __u32 csum; | |
59 | ||
171a7f21 | 60 | csum_lo = le16_to_cpu(raw->i_checksum_lo); |
814525f4 DW |
61 | raw->i_checksum_lo = 0; |
62 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && | |
63 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) { | |
171a7f21 | 64 | csum_hi = le16_to_cpu(raw->i_checksum_hi); |
814525f4 DW |
65 | raw->i_checksum_hi = 0; |
66 | } | |
67 | ||
68 | csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw, | |
69 | EXT4_INODE_SIZE(inode->i_sb)); | |
70 | ||
171a7f21 | 71 | raw->i_checksum_lo = cpu_to_le16(csum_lo); |
814525f4 DW |
72 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && |
73 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) | |
171a7f21 | 74 | raw->i_checksum_hi = cpu_to_le16(csum_hi); |
814525f4 DW |
75 | |
76 | return csum; | |
77 | } | |
78 | ||
79 | static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw, | |
80 | struct ext4_inode_info *ei) | |
81 | { | |
82 | __u32 provided, calculated; | |
83 | ||
84 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != | |
85 | cpu_to_le32(EXT4_OS_LINUX) || | |
86 | !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, | |
87 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) | |
88 | return 1; | |
89 | ||
90 | provided = le16_to_cpu(raw->i_checksum_lo); | |
91 | calculated = ext4_inode_csum(inode, raw, ei); | |
92 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && | |
93 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) | |
94 | provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16; | |
95 | else | |
96 | calculated &= 0xFFFF; | |
97 | ||
98 | return provided == calculated; | |
99 | } | |
100 | ||
101 | static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw, | |
102 | struct ext4_inode_info *ei) | |
103 | { | |
104 | __u32 csum; | |
105 | ||
106 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != | |
107 | cpu_to_le32(EXT4_OS_LINUX) || | |
108 | !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, | |
109 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) | |
110 | return; | |
111 | ||
112 | csum = ext4_inode_csum(inode, raw, ei); | |
113 | raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF); | |
114 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && | |
115 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) | |
116 | raw->i_checksum_hi = cpu_to_le16(csum >> 16); | |
117 | } | |
118 | ||
678aaf48 JK |
119 | static inline int ext4_begin_ordered_truncate(struct inode *inode, |
120 | loff_t new_size) | |
121 | { | |
7ff9c073 | 122 | trace_ext4_begin_ordered_truncate(inode, new_size); |
8aefcd55 TT |
123 | /* |
124 | * If jinode is zero, then we never opened the file for | |
125 | * writing, so there's no need to call | |
126 | * jbd2_journal_begin_ordered_truncate() since there's no | |
127 | * outstanding writes we need to flush. | |
128 | */ | |
129 | if (!EXT4_I(inode)->jinode) | |
130 | return 0; | |
131 | return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode), | |
132 | EXT4_I(inode)->jinode, | |
133 | new_size); | |
678aaf48 JK |
134 | } |
135 | ||
d47992f8 LC |
136 | static void ext4_invalidatepage(struct page *page, unsigned int offset, |
137 | unsigned int length); | |
cb20d518 TT |
138 | static int __ext4_journalled_writepage(struct page *page, unsigned int len); |
139 | static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh); | |
fffb2739 JK |
140 | static int ext4_meta_trans_blocks(struct inode *inode, int lblocks, |
141 | int pextents); | |
64769240 | 142 | |
ac27a0ec DK |
143 | /* |
144 | * Test whether an inode is a fast symlink. | |
145 | */ | |
617ba13b | 146 | static int ext4_inode_is_fast_symlink(struct inode *inode) |
ac27a0ec | 147 | { |
65eddb56 YY |
148 | int ea_blocks = EXT4_I(inode)->i_file_acl ? |
149 | EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0; | |
ac27a0ec | 150 | |
bd9db175 ZL |
151 | if (ext4_has_inline_data(inode)) |
152 | return 0; | |
153 | ||
ac27a0ec DK |
154 | return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0); |
155 | } | |
156 | ||
ac27a0ec DK |
157 | /* |
158 | * Restart the transaction associated with *handle. This does a commit, | |
159 | * so before we call here everything must be consistently dirtied against | |
160 | * this transaction. | |
161 | */ | |
fa5d1113 | 162 | int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode, |
487caeef | 163 | int nblocks) |
ac27a0ec | 164 | { |
487caeef JK |
165 | int ret; |
166 | ||
167 | /* | |
e35fd660 | 168 | * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this |
487caeef JK |
169 | * moment, get_block can be called only for blocks inside i_size since |
170 | * page cache has been already dropped and writes are blocked by | |
171 | * i_mutex. So we can safely drop the i_data_sem here. | |
172 | */ | |
0390131b | 173 | BUG_ON(EXT4_JOURNAL(inode) == NULL); |
ac27a0ec | 174 | jbd_debug(2, "restarting handle %p\n", handle); |
487caeef | 175 | up_write(&EXT4_I(inode)->i_data_sem); |
8e8eaabe | 176 | ret = ext4_journal_restart(handle, nblocks); |
487caeef | 177 | down_write(&EXT4_I(inode)->i_data_sem); |
fa5d1113 | 178 | ext4_discard_preallocations(inode); |
487caeef JK |
179 | |
180 | return ret; | |
ac27a0ec DK |
181 | } |
182 | ||
183 | /* | |
184 | * Called at the last iput() if i_nlink is zero. | |
185 | */ | |
0930fcc1 | 186 | void ext4_evict_inode(struct inode *inode) |
ac27a0ec DK |
187 | { |
188 | handle_t *handle; | |
bc965ab3 | 189 | int err; |
ac27a0ec | 190 | |
7ff9c073 | 191 | trace_ext4_evict_inode(inode); |
2581fdc8 | 192 | |
0930fcc1 | 193 | if (inode->i_nlink) { |
2d859db3 JK |
194 | /* |
195 | * When journalling data dirty buffers are tracked only in the | |
196 | * journal. So although mm thinks everything is clean and | |
197 | * ready for reaping the inode might still have some pages to | |
198 | * write in the running transaction or waiting to be | |
199 | * checkpointed. Thus calling jbd2_journal_invalidatepage() | |
200 | * (via truncate_inode_pages()) to discard these buffers can | |
201 | * cause data loss. Also even if we did not discard these | |
202 | * buffers, we would have no way to find them after the inode | |
203 | * is reaped and thus user could see stale data if he tries to | |
204 | * read them before the transaction is checkpointed. So be | |
205 | * careful and force everything to disk here... We use | |
206 | * ei->i_datasync_tid to store the newest transaction | |
207 | * containing inode's data. | |
208 | * | |
209 | * Note that directories do not have this problem because they | |
210 | * don't use page cache. | |
211 | */ | |
212 | if (ext4_should_journal_data(inode) && | |
2b405bfa TT |
213 | (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) && |
214 | inode->i_ino != EXT4_JOURNAL_INO) { | |
2d859db3 JK |
215 | journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; |
216 | tid_t commit_tid = EXT4_I(inode)->i_datasync_tid; | |
217 | ||
d76a3a77 | 218 | jbd2_complete_transaction(journal, commit_tid); |
2d859db3 JK |
219 | filemap_write_and_wait(&inode->i_data); |
220 | } | |
91b0abe3 | 221 | truncate_inode_pages_final(&inode->i_data); |
5dc23bdd JK |
222 | |
223 | WARN_ON(atomic_read(&EXT4_I(inode)->i_ioend_count)); | |
0930fcc1 AV |
224 | goto no_delete; |
225 | } | |
226 | ||
907f4554 | 227 | if (!is_bad_inode(inode)) |
871a2931 | 228 | dquot_initialize(inode); |
907f4554 | 229 | |
678aaf48 JK |
230 | if (ext4_should_order_data(inode)) |
231 | ext4_begin_ordered_truncate(inode, 0); | |
91b0abe3 | 232 | truncate_inode_pages_final(&inode->i_data); |
ac27a0ec | 233 | |
5dc23bdd | 234 | WARN_ON(atomic_read(&EXT4_I(inode)->i_ioend_count)); |
ac27a0ec DK |
235 | if (is_bad_inode(inode)) |
236 | goto no_delete; | |
237 | ||
8e8ad8a5 JK |
238 | /* |
239 | * Protect us against freezing - iput() caller didn't have to have any | |
240 | * protection against it | |
241 | */ | |
242 | sb_start_intwrite(inode->i_sb); | |
9924a92a TT |
243 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, |
244 | ext4_blocks_for_truncate(inode)+3); | |
ac27a0ec | 245 | if (IS_ERR(handle)) { |
bc965ab3 | 246 | ext4_std_error(inode->i_sb, PTR_ERR(handle)); |
ac27a0ec DK |
247 | /* |
248 | * If we're going to skip the normal cleanup, we still need to | |
249 | * make sure that the in-core orphan linked list is properly | |
250 | * cleaned up. | |
251 | */ | |
617ba13b | 252 | ext4_orphan_del(NULL, inode); |
8e8ad8a5 | 253 | sb_end_intwrite(inode->i_sb); |
ac27a0ec DK |
254 | goto no_delete; |
255 | } | |
256 | ||
257 | if (IS_SYNC(inode)) | |
0390131b | 258 | ext4_handle_sync(handle); |
ac27a0ec | 259 | inode->i_size = 0; |
bc965ab3 TT |
260 | err = ext4_mark_inode_dirty(handle, inode); |
261 | if (err) { | |
12062ddd | 262 | ext4_warning(inode->i_sb, |
bc965ab3 TT |
263 | "couldn't mark inode dirty (err %d)", err); |
264 | goto stop_handle; | |
265 | } | |
ac27a0ec | 266 | if (inode->i_blocks) |
617ba13b | 267 | ext4_truncate(inode); |
bc965ab3 TT |
268 | |
269 | /* | |
270 | * ext4_ext_truncate() doesn't reserve any slop when it | |
271 | * restarts journal transactions; therefore there may not be | |
272 | * enough credits left in the handle to remove the inode from | |
273 | * the orphan list and set the dtime field. | |
274 | */ | |
0390131b | 275 | if (!ext4_handle_has_enough_credits(handle, 3)) { |
bc965ab3 TT |
276 | err = ext4_journal_extend(handle, 3); |
277 | if (err > 0) | |
278 | err = ext4_journal_restart(handle, 3); | |
279 | if (err != 0) { | |
12062ddd | 280 | ext4_warning(inode->i_sb, |
bc965ab3 TT |
281 | "couldn't extend journal (err %d)", err); |
282 | stop_handle: | |
283 | ext4_journal_stop(handle); | |
45388219 | 284 | ext4_orphan_del(NULL, inode); |
8e8ad8a5 | 285 | sb_end_intwrite(inode->i_sb); |
bc965ab3 TT |
286 | goto no_delete; |
287 | } | |
288 | } | |
289 | ||
ac27a0ec | 290 | /* |
617ba13b | 291 | * Kill off the orphan record which ext4_truncate created. |
ac27a0ec | 292 | * AKPM: I think this can be inside the above `if'. |
617ba13b | 293 | * Note that ext4_orphan_del() has to be able to cope with the |
ac27a0ec | 294 | * deletion of a non-existent orphan - this is because we don't |
617ba13b | 295 | * know if ext4_truncate() actually created an orphan record. |
ac27a0ec DK |
296 | * (Well, we could do this if we need to, but heck - it works) |
297 | */ | |
617ba13b MC |
298 | ext4_orphan_del(handle, inode); |
299 | EXT4_I(inode)->i_dtime = get_seconds(); | |
ac27a0ec DK |
300 | |
301 | /* | |
302 | * One subtle ordering requirement: if anything has gone wrong | |
303 | * (transaction abort, IO errors, whatever), then we can still | |
304 | * do these next steps (the fs will already have been marked as | |
305 | * having errors), but we can't free the inode if the mark_dirty | |
306 | * fails. | |
307 | */ | |
617ba13b | 308 | if (ext4_mark_inode_dirty(handle, inode)) |
ac27a0ec | 309 | /* If that failed, just do the required in-core inode clear. */ |
0930fcc1 | 310 | ext4_clear_inode(inode); |
ac27a0ec | 311 | else |
617ba13b MC |
312 | ext4_free_inode(handle, inode); |
313 | ext4_journal_stop(handle); | |
8e8ad8a5 | 314 | sb_end_intwrite(inode->i_sb); |
ac27a0ec DK |
315 | return; |
316 | no_delete: | |
0930fcc1 | 317 | ext4_clear_inode(inode); /* We must guarantee clearing of inode... */ |
ac27a0ec DK |
318 | } |
319 | ||
a9e7f447 DM |
320 | #ifdef CONFIG_QUOTA |
321 | qsize_t *ext4_get_reserved_space(struct inode *inode) | |
60e58e0f | 322 | { |
a9e7f447 | 323 | return &EXT4_I(inode)->i_reserved_quota; |
60e58e0f | 324 | } |
a9e7f447 | 325 | #endif |
9d0be502 | 326 | |
12219aea AK |
327 | /* |
328 | * Calculate the number of metadata blocks need to reserve | |
9d0be502 | 329 | * to allocate a block located at @lblock |
12219aea | 330 | */ |
01f49d0b | 331 | static int ext4_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock) |
12219aea | 332 | { |
12e9b892 | 333 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
9d0be502 | 334 | return ext4_ext_calc_metadata_amount(inode, lblock); |
12219aea | 335 | |
8bb2b247 | 336 | return ext4_ind_calc_metadata_amount(inode, lblock); |
12219aea AK |
337 | } |
338 | ||
0637c6f4 TT |
339 | /* |
340 | * Called with i_data_sem down, which is important since we can call | |
341 | * ext4_discard_preallocations() from here. | |
342 | */ | |
5f634d06 AK |
343 | void ext4_da_update_reserve_space(struct inode *inode, |
344 | int used, int quota_claim) | |
12219aea AK |
345 | { |
346 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0637c6f4 | 347 | struct ext4_inode_info *ei = EXT4_I(inode); |
0637c6f4 TT |
348 | |
349 | spin_lock(&ei->i_block_reservation_lock); | |
d8990240 | 350 | trace_ext4_da_update_reserve_space(inode, used, quota_claim); |
0637c6f4 | 351 | if (unlikely(used > ei->i_reserved_data_blocks)) { |
8de5c325 | 352 | ext4_warning(inode->i_sb, "%s: ino %lu, used %d " |
1084f252 | 353 | "with only %d reserved data blocks", |
0637c6f4 TT |
354 | __func__, inode->i_ino, used, |
355 | ei->i_reserved_data_blocks); | |
356 | WARN_ON(1); | |
357 | used = ei->i_reserved_data_blocks; | |
358 | } | |
12219aea | 359 | |
97795d2a | 360 | if (unlikely(ei->i_allocated_meta_blocks > ei->i_reserved_meta_blocks)) { |
01a523eb TT |
361 | ext4_warning(inode->i_sb, "ino %lu, allocated %d " |
362 | "with only %d reserved metadata blocks " | |
363 | "(releasing %d blocks with reserved %d data blocks)", | |
364 | inode->i_ino, ei->i_allocated_meta_blocks, | |
365 | ei->i_reserved_meta_blocks, used, | |
366 | ei->i_reserved_data_blocks); | |
97795d2a BF |
367 | WARN_ON(1); |
368 | ei->i_allocated_meta_blocks = ei->i_reserved_meta_blocks; | |
369 | } | |
370 | ||
0637c6f4 TT |
371 | /* Update per-inode reservations */ |
372 | ei->i_reserved_data_blocks -= used; | |
0637c6f4 | 373 | ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks; |
57042651 | 374 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, |
72b8ab9d | 375 | used + ei->i_allocated_meta_blocks); |
0637c6f4 | 376 | ei->i_allocated_meta_blocks = 0; |
6bc6e63f | 377 | |
0637c6f4 TT |
378 | if (ei->i_reserved_data_blocks == 0) { |
379 | /* | |
380 | * We can release all of the reserved metadata blocks | |
381 | * only when we have written all of the delayed | |
382 | * allocation blocks. | |
383 | */ | |
57042651 | 384 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, |
72b8ab9d | 385 | ei->i_reserved_meta_blocks); |
ee5f4d9c | 386 | ei->i_reserved_meta_blocks = 0; |
9d0be502 | 387 | ei->i_da_metadata_calc_len = 0; |
6bc6e63f | 388 | } |
12219aea | 389 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
60e58e0f | 390 | |
72b8ab9d ES |
391 | /* Update quota subsystem for data blocks */ |
392 | if (quota_claim) | |
7b415bf6 | 393 | dquot_claim_block(inode, EXT4_C2B(sbi, used)); |
72b8ab9d | 394 | else { |
5f634d06 AK |
395 | /* |
396 | * We did fallocate with an offset that is already delayed | |
397 | * allocated. So on delayed allocated writeback we should | |
72b8ab9d | 398 | * not re-claim the quota for fallocated blocks. |
5f634d06 | 399 | */ |
7b415bf6 | 400 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, used)); |
5f634d06 | 401 | } |
d6014301 AK |
402 | |
403 | /* | |
404 | * If we have done all the pending block allocations and if | |
405 | * there aren't any writers on the inode, we can discard the | |
406 | * inode's preallocations. | |
407 | */ | |
0637c6f4 TT |
408 | if ((ei->i_reserved_data_blocks == 0) && |
409 | (atomic_read(&inode->i_writecount) == 0)) | |
d6014301 | 410 | ext4_discard_preallocations(inode); |
12219aea AK |
411 | } |
412 | ||
e29136f8 | 413 | static int __check_block_validity(struct inode *inode, const char *func, |
c398eda0 TT |
414 | unsigned int line, |
415 | struct ext4_map_blocks *map) | |
6fd058f7 | 416 | { |
24676da4 TT |
417 | if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk, |
418 | map->m_len)) { | |
c398eda0 TT |
419 | ext4_error_inode(inode, func, line, map->m_pblk, |
420 | "lblock %lu mapped to illegal pblock " | |
421 | "(length %d)", (unsigned long) map->m_lblk, | |
422 | map->m_len); | |
6fd058f7 TT |
423 | return -EIO; |
424 | } | |
425 | return 0; | |
426 | } | |
427 | ||
e29136f8 | 428 | #define check_block_validity(inode, map) \ |
c398eda0 | 429 | __check_block_validity((inode), __func__, __LINE__, (map)) |
e29136f8 | 430 | |
921f266b DM |
431 | #ifdef ES_AGGRESSIVE_TEST |
432 | static void ext4_map_blocks_es_recheck(handle_t *handle, | |
433 | struct inode *inode, | |
434 | struct ext4_map_blocks *es_map, | |
435 | struct ext4_map_blocks *map, | |
436 | int flags) | |
437 | { | |
438 | int retval; | |
439 | ||
440 | map->m_flags = 0; | |
441 | /* | |
442 | * There is a race window that the result is not the same. | |
443 | * e.g. xfstests #223 when dioread_nolock enables. The reason | |
444 | * is that we lookup a block mapping in extent status tree with | |
445 | * out taking i_data_sem. So at the time the unwritten extent | |
446 | * could be converted. | |
447 | */ | |
448 | if (!(flags & EXT4_GET_BLOCKS_NO_LOCK)) | |
c8b459f4 | 449 | down_read(&EXT4_I(inode)->i_data_sem); |
921f266b DM |
450 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
451 | retval = ext4_ext_map_blocks(handle, inode, map, flags & | |
452 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
453 | } else { | |
454 | retval = ext4_ind_map_blocks(handle, inode, map, flags & | |
455 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
456 | } | |
457 | if (!(flags & EXT4_GET_BLOCKS_NO_LOCK)) | |
458 | up_read((&EXT4_I(inode)->i_data_sem)); | |
459 | /* | |
460 | * Clear EXT4_MAP_FROM_CLUSTER and EXT4_MAP_BOUNDARY flag | |
461 | * because it shouldn't be marked in es_map->m_flags. | |
462 | */ | |
463 | map->m_flags &= ~(EXT4_MAP_FROM_CLUSTER | EXT4_MAP_BOUNDARY); | |
464 | ||
465 | /* | |
466 | * We don't check m_len because extent will be collpased in status | |
467 | * tree. So the m_len might not equal. | |
468 | */ | |
469 | if (es_map->m_lblk != map->m_lblk || | |
470 | es_map->m_flags != map->m_flags || | |
471 | es_map->m_pblk != map->m_pblk) { | |
bdafe42a | 472 | printk("ES cache assertion failed for inode: %lu " |
921f266b DM |
473 | "es_cached ex [%d/%d/%llu/%x] != " |
474 | "found ex [%d/%d/%llu/%x] retval %d flags %x\n", | |
475 | inode->i_ino, es_map->m_lblk, es_map->m_len, | |
476 | es_map->m_pblk, es_map->m_flags, map->m_lblk, | |
477 | map->m_len, map->m_pblk, map->m_flags, | |
478 | retval, flags); | |
479 | } | |
480 | } | |
481 | #endif /* ES_AGGRESSIVE_TEST */ | |
482 | ||
f5ab0d1f | 483 | /* |
e35fd660 | 484 | * The ext4_map_blocks() function tries to look up the requested blocks, |
2b2d6d01 | 485 | * and returns if the blocks are already mapped. |
f5ab0d1f | 486 | * |
f5ab0d1f MC |
487 | * Otherwise it takes the write lock of the i_data_sem and allocate blocks |
488 | * and store the allocated blocks in the result buffer head and mark it | |
489 | * mapped. | |
490 | * | |
e35fd660 TT |
491 | * If file type is extents based, it will call ext4_ext_map_blocks(), |
492 | * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping | |
f5ab0d1f MC |
493 | * based files |
494 | * | |
556615dc LC |
495 | * On success, it returns the number of blocks being mapped or allocated. |
496 | * if create==0 and the blocks are pre-allocated and unwritten block, | |
f5ab0d1f MC |
497 | * the result buffer head is unmapped. If the create ==1, it will make sure |
498 | * the buffer head is mapped. | |
499 | * | |
500 | * It returns 0 if plain look up failed (blocks have not been allocated), in | |
df3ab170 | 501 | * that case, buffer head is unmapped |
f5ab0d1f MC |
502 | * |
503 | * It returns the error in case of allocation failure. | |
504 | */ | |
e35fd660 TT |
505 | int ext4_map_blocks(handle_t *handle, struct inode *inode, |
506 | struct ext4_map_blocks *map, int flags) | |
0e855ac8 | 507 | { |
d100eef2 | 508 | struct extent_status es; |
0e855ac8 | 509 | int retval; |
b8a86845 | 510 | int ret = 0; |
921f266b DM |
511 | #ifdef ES_AGGRESSIVE_TEST |
512 | struct ext4_map_blocks orig_map; | |
513 | ||
514 | memcpy(&orig_map, map, sizeof(*map)); | |
515 | #endif | |
f5ab0d1f | 516 | |
e35fd660 TT |
517 | map->m_flags = 0; |
518 | ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u," | |
519 | "logical block %lu\n", inode->i_ino, flags, map->m_len, | |
520 | (unsigned long) map->m_lblk); | |
d100eef2 | 521 | |
e861b5e9 TT |
522 | /* |
523 | * ext4_map_blocks returns an int, and m_len is an unsigned int | |
524 | */ | |
525 | if (unlikely(map->m_len > INT_MAX)) | |
526 | map->m_len = INT_MAX; | |
527 | ||
4adb6ab3 KM |
528 | /* We can handle the block number less than EXT_MAX_BLOCKS */ |
529 | if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS)) | |
530 | return -EIO; | |
531 | ||
d100eef2 ZL |
532 | /* Lookup extent status tree firstly */ |
533 | if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) { | |
63b99968 | 534 | ext4_es_lru_add(inode); |
d100eef2 ZL |
535 | if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) { |
536 | map->m_pblk = ext4_es_pblock(&es) + | |
537 | map->m_lblk - es.es_lblk; | |
538 | map->m_flags |= ext4_es_is_written(&es) ? | |
539 | EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN; | |
540 | retval = es.es_len - (map->m_lblk - es.es_lblk); | |
541 | if (retval > map->m_len) | |
542 | retval = map->m_len; | |
543 | map->m_len = retval; | |
544 | } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) { | |
545 | retval = 0; | |
546 | } else { | |
547 | BUG_ON(1); | |
548 | } | |
921f266b DM |
549 | #ifdef ES_AGGRESSIVE_TEST |
550 | ext4_map_blocks_es_recheck(handle, inode, map, | |
551 | &orig_map, flags); | |
552 | #endif | |
d100eef2 ZL |
553 | goto found; |
554 | } | |
555 | ||
4df3d265 | 556 | /* |
b920c755 TT |
557 | * Try to see if we can get the block without requesting a new |
558 | * file system block. | |
4df3d265 | 559 | */ |
729f52c6 | 560 | if (!(flags & EXT4_GET_BLOCKS_NO_LOCK)) |
c8b459f4 | 561 | down_read(&EXT4_I(inode)->i_data_sem); |
12e9b892 | 562 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
a4e5d88b DM |
563 | retval = ext4_ext_map_blocks(handle, inode, map, flags & |
564 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
0e855ac8 | 565 | } else { |
a4e5d88b DM |
566 | retval = ext4_ind_map_blocks(handle, inode, map, flags & |
567 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
0e855ac8 | 568 | } |
f7fec032 | 569 | if (retval > 0) { |
3be78c73 | 570 | unsigned int status; |
f7fec032 | 571 | |
44fb851d ZL |
572 | if (unlikely(retval != map->m_len)) { |
573 | ext4_warning(inode->i_sb, | |
574 | "ES len assertion failed for inode " | |
575 | "%lu: retval %d != map->m_len %d", | |
576 | inode->i_ino, retval, map->m_len); | |
577 | WARN_ON(1); | |
921f266b | 578 | } |
921f266b | 579 | |
f7fec032 ZL |
580 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
581 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; | |
582 | if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) && | |
583 | ext4_find_delalloc_range(inode, map->m_lblk, | |
584 | map->m_lblk + map->m_len - 1)) | |
585 | status |= EXTENT_STATUS_DELAYED; | |
586 | ret = ext4_es_insert_extent(inode, map->m_lblk, | |
587 | map->m_len, map->m_pblk, status); | |
588 | if (ret < 0) | |
589 | retval = ret; | |
590 | } | |
729f52c6 ZL |
591 | if (!(flags & EXT4_GET_BLOCKS_NO_LOCK)) |
592 | up_read((&EXT4_I(inode)->i_data_sem)); | |
f5ab0d1f | 593 | |
d100eef2 | 594 | found: |
e35fd660 | 595 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
b8a86845 | 596 | ret = check_block_validity(inode, map); |
6fd058f7 TT |
597 | if (ret != 0) |
598 | return ret; | |
599 | } | |
600 | ||
f5ab0d1f | 601 | /* If it is only a block(s) look up */ |
c2177057 | 602 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) |
f5ab0d1f MC |
603 | return retval; |
604 | ||
605 | /* | |
606 | * Returns if the blocks have already allocated | |
607 | * | |
608 | * Note that if blocks have been preallocated | |
df3ab170 | 609 | * ext4_ext_get_block() returns the create = 0 |
f5ab0d1f MC |
610 | * with buffer head unmapped. |
611 | */ | |
e35fd660 | 612 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) |
b8a86845 LC |
613 | /* |
614 | * If we need to convert extent to unwritten | |
615 | * we continue and do the actual work in | |
616 | * ext4_ext_map_blocks() | |
617 | */ | |
618 | if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) | |
619 | return retval; | |
4df3d265 | 620 | |
2a8964d6 | 621 | /* |
a25a4e1a ZL |
622 | * Here we clear m_flags because after allocating an new extent, |
623 | * it will be set again. | |
2a8964d6 | 624 | */ |
a25a4e1a | 625 | map->m_flags &= ~EXT4_MAP_FLAGS; |
2a8964d6 | 626 | |
4df3d265 | 627 | /* |
556615dc | 628 | * New blocks allocate and/or writing to unwritten extent |
f5ab0d1f MC |
629 | * will possibly result in updating i_data, so we take |
630 | * the write lock of i_data_sem, and call get_blocks() | |
631 | * with create == 1 flag. | |
4df3d265 | 632 | */ |
c8b459f4 | 633 | down_write(&EXT4_I(inode)->i_data_sem); |
d2a17637 MC |
634 | |
635 | /* | |
636 | * if the caller is from delayed allocation writeout path | |
637 | * we have already reserved fs blocks for allocation | |
638 | * let the underlying get_block() function know to | |
639 | * avoid double accounting | |
640 | */ | |
c2177057 | 641 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) |
f2321097 | 642 | ext4_set_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED); |
4df3d265 AK |
643 | /* |
644 | * We need to check for EXT4 here because migrate | |
645 | * could have changed the inode type in between | |
646 | */ | |
12e9b892 | 647 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
e35fd660 | 648 | retval = ext4_ext_map_blocks(handle, inode, map, flags); |
0e855ac8 | 649 | } else { |
e35fd660 | 650 | retval = ext4_ind_map_blocks(handle, inode, map, flags); |
267e4db9 | 651 | |
e35fd660 | 652 | if (retval > 0 && map->m_flags & EXT4_MAP_NEW) { |
267e4db9 AK |
653 | /* |
654 | * We allocated new blocks which will result in | |
655 | * i_data's format changing. Force the migrate | |
656 | * to fail by clearing migrate flags | |
657 | */ | |
19f5fb7a | 658 | ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE); |
267e4db9 | 659 | } |
d2a17637 | 660 | |
5f634d06 AK |
661 | /* |
662 | * Update reserved blocks/metadata blocks after successful | |
663 | * block allocation which had been deferred till now. We don't | |
664 | * support fallocate for non extent files. So we can update | |
665 | * reserve space here. | |
666 | */ | |
667 | if ((retval > 0) && | |
1296cc85 | 668 | (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)) |
5f634d06 AK |
669 | ext4_da_update_reserve_space(inode, retval, 1); |
670 | } | |
f7fec032 | 671 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) |
f2321097 | 672 | ext4_clear_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED); |
2ac3b6e0 | 673 | |
f7fec032 | 674 | if (retval > 0) { |
3be78c73 | 675 | unsigned int status; |
f7fec032 | 676 | |
44fb851d ZL |
677 | if (unlikely(retval != map->m_len)) { |
678 | ext4_warning(inode->i_sb, | |
679 | "ES len assertion failed for inode " | |
680 | "%lu: retval %d != map->m_len %d", | |
681 | inode->i_ino, retval, map->m_len); | |
682 | WARN_ON(1); | |
921f266b | 683 | } |
921f266b | 684 | |
adb23551 ZL |
685 | /* |
686 | * If the extent has been zeroed out, we don't need to update | |
687 | * extent status tree. | |
688 | */ | |
689 | if ((flags & EXT4_GET_BLOCKS_PRE_IO) && | |
690 | ext4_es_lookup_extent(inode, map->m_lblk, &es)) { | |
691 | if (ext4_es_is_written(&es)) | |
692 | goto has_zeroout; | |
693 | } | |
f7fec032 ZL |
694 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
695 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; | |
696 | if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) && | |
697 | ext4_find_delalloc_range(inode, map->m_lblk, | |
698 | map->m_lblk + map->m_len - 1)) | |
699 | status |= EXTENT_STATUS_DELAYED; | |
700 | ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, | |
701 | map->m_pblk, status); | |
702 | if (ret < 0) | |
703 | retval = ret; | |
5356f261 AK |
704 | } |
705 | ||
adb23551 | 706 | has_zeroout: |
4df3d265 | 707 | up_write((&EXT4_I(inode)->i_data_sem)); |
e35fd660 | 708 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
b8a86845 | 709 | ret = check_block_validity(inode, map); |
6fd058f7 TT |
710 | if (ret != 0) |
711 | return ret; | |
712 | } | |
0e855ac8 AK |
713 | return retval; |
714 | } | |
715 | ||
f3bd1f3f MC |
716 | /* Maximum number of blocks we map for direct IO at once. */ |
717 | #define DIO_MAX_BLOCKS 4096 | |
718 | ||
2ed88685 TT |
719 | static int _ext4_get_block(struct inode *inode, sector_t iblock, |
720 | struct buffer_head *bh, int flags) | |
ac27a0ec | 721 | { |
3e4fdaf8 | 722 | handle_t *handle = ext4_journal_current_handle(); |
2ed88685 | 723 | struct ext4_map_blocks map; |
7fb5409d | 724 | int ret = 0, started = 0; |
f3bd1f3f | 725 | int dio_credits; |
ac27a0ec | 726 | |
46c7f254 TM |
727 | if (ext4_has_inline_data(inode)) |
728 | return -ERANGE; | |
729 | ||
2ed88685 TT |
730 | map.m_lblk = iblock; |
731 | map.m_len = bh->b_size >> inode->i_blkbits; | |
732 | ||
8b0f165f | 733 | if (flags && !(flags & EXT4_GET_BLOCKS_NO_LOCK) && !handle) { |
7fb5409d | 734 | /* Direct IO write... */ |
2ed88685 TT |
735 | if (map.m_len > DIO_MAX_BLOCKS) |
736 | map.m_len = DIO_MAX_BLOCKS; | |
737 | dio_credits = ext4_chunk_trans_blocks(inode, map.m_len); | |
9924a92a TT |
738 | handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, |
739 | dio_credits); | |
7fb5409d | 740 | if (IS_ERR(handle)) { |
ac27a0ec | 741 | ret = PTR_ERR(handle); |
2ed88685 | 742 | return ret; |
ac27a0ec | 743 | } |
7fb5409d | 744 | started = 1; |
ac27a0ec DK |
745 | } |
746 | ||
2ed88685 | 747 | ret = ext4_map_blocks(handle, inode, &map, flags); |
7fb5409d | 748 | if (ret > 0) { |
7b7a8665 CH |
749 | ext4_io_end_t *io_end = ext4_inode_aio(inode); |
750 | ||
2ed88685 TT |
751 | map_bh(bh, inode->i_sb, map.m_pblk); |
752 | bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags; | |
7b7a8665 CH |
753 | if (io_end && io_end->flag & EXT4_IO_END_UNWRITTEN) |
754 | set_buffer_defer_completion(bh); | |
2ed88685 | 755 | bh->b_size = inode->i_sb->s_blocksize * map.m_len; |
7fb5409d | 756 | ret = 0; |
ac27a0ec | 757 | } |
7fb5409d JK |
758 | if (started) |
759 | ext4_journal_stop(handle); | |
ac27a0ec DK |
760 | return ret; |
761 | } | |
762 | ||
2ed88685 TT |
763 | int ext4_get_block(struct inode *inode, sector_t iblock, |
764 | struct buffer_head *bh, int create) | |
765 | { | |
766 | return _ext4_get_block(inode, iblock, bh, | |
767 | create ? EXT4_GET_BLOCKS_CREATE : 0); | |
768 | } | |
769 | ||
ac27a0ec DK |
770 | /* |
771 | * `handle' can be NULL if create is zero | |
772 | */ | |
617ba13b | 773 | struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode, |
725d26d3 | 774 | ext4_lblk_t block, int create, int *errp) |
ac27a0ec | 775 | { |
2ed88685 TT |
776 | struct ext4_map_blocks map; |
777 | struct buffer_head *bh; | |
ac27a0ec DK |
778 | int fatal = 0, err; |
779 | ||
780 | J_ASSERT(handle != NULL || create == 0); | |
781 | ||
2ed88685 TT |
782 | map.m_lblk = block; |
783 | map.m_len = 1; | |
784 | err = ext4_map_blocks(handle, inode, &map, | |
785 | create ? EXT4_GET_BLOCKS_CREATE : 0); | |
ac27a0ec | 786 | |
90b0a973 CM |
787 | /* ensure we send some value back into *errp */ |
788 | *errp = 0; | |
789 | ||
0f70b406 TT |
790 | if (create && err == 0) |
791 | err = -ENOSPC; /* should never happen */ | |
2ed88685 TT |
792 | if (err < 0) |
793 | *errp = err; | |
794 | if (err <= 0) | |
795 | return NULL; | |
2ed88685 TT |
796 | |
797 | bh = sb_getblk(inode->i_sb, map.m_pblk); | |
aebf0243 | 798 | if (unlikely(!bh)) { |
860d21e2 | 799 | *errp = -ENOMEM; |
2ed88685 | 800 | return NULL; |
ac27a0ec | 801 | } |
2ed88685 TT |
802 | if (map.m_flags & EXT4_MAP_NEW) { |
803 | J_ASSERT(create != 0); | |
804 | J_ASSERT(handle != NULL); | |
ac27a0ec | 805 | |
2ed88685 TT |
806 | /* |
807 | * Now that we do not always journal data, we should | |
808 | * keep in mind whether this should always journal the | |
809 | * new buffer as metadata. For now, regular file | |
810 | * writes use ext4_get_block instead, so it's not a | |
811 | * problem. | |
812 | */ | |
813 | lock_buffer(bh); | |
814 | BUFFER_TRACE(bh, "call get_create_access"); | |
815 | fatal = ext4_journal_get_create_access(handle, bh); | |
816 | if (!fatal && !buffer_uptodate(bh)) { | |
817 | memset(bh->b_data, 0, inode->i_sb->s_blocksize); | |
818 | set_buffer_uptodate(bh); | |
ac27a0ec | 819 | } |
2ed88685 TT |
820 | unlock_buffer(bh); |
821 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); | |
822 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
823 | if (!fatal) | |
824 | fatal = err; | |
825 | } else { | |
826 | BUFFER_TRACE(bh, "not a new buffer"); | |
ac27a0ec | 827 | } |
2ed88685 TT |
828 | if (fatal) { |
829 | *errp = fatal; | |
830 | brelse(bh); | |
831 | bh = NULL; | |
832 | } | |
833 | return bh; | |
ac27a0ec DK |
834 | } |
835 | ||
617ba13b | 836 | struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode, |
725d26d3 | 837 | ext4_lblk_t block, int create, int *err) |
ac27a0ec | 838 | { |
af5bc92d | 839 | struct buffer_head *bh; |
ac27a0ec | 840 | |
617ba13b | 841 | bh = ext4_getblk(handle, inode, block, create, err); |
ac27a0ec DK |
842 | if (!bh) |
843 | return bh; | |
844 | if (buffer_uptodate(bh)) | |
845 | return bh; | |
65299a3b | 846 | ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh); |
ac27a0ec DK |
847 | wait_on_buffer(bh); |
848 | if (buffer_uptodate(bh)) | |
849 | return bh; | |
850 | put_bh(bh); | |
851 | *err = -EIO; | |
852 | return NULL; | |
853 | } | |
854 | ||
f19d5870 TM |
855 | int ext4_walk_page_buffers(handle_t *handle, |
856 | struct buffer_head *head, | |
857 | unsigned from, | |
858 | unsigned to, | |
859 | int *partial, | |
860 | int (*fn)(handle_t *handle, | |
861 | struct buffer_head *bh)) | |
ac27a0ec DK |
862 | { |
863 | struct buffer_head *bh; | |
864 | unsigned block_start, block_end; | |
865 | unsigned blocksize = head->b_size; | |
866 | int err, ret = 0; | |
867 | struct buffer_head *next; | |
868 | ||
af5bc92d TT |
869 | for (bh = head, block_start = 0; |
870 | ret == 0 && (bh != head || !block_start); | |
de9a55b8 | 871 | block_start = block_end, bh = next) { |
ac27a0ec DK |
872 | next = bh->b_this_page; |
873 | block_end = block_start + blocksize; | |
874 | if (block_end <= from || block_start >= to) { | |
875 | if (partial && !buffer_uptodate(bh)) | |
876 | *partial = 1; | |
877 | continue; | |
878 | } | |
879 | err = (*fn)(handle, bh); | |
880 | if (!ret) | |
881 | ret = err; | |
882 | } | |
883 | return ret; | |
884 | } | |
885 | ||
886 | /* | |
887 | * To preserve ordering, it is essential that the hole instantiation and | |
888 | * the data write be encapsulated in a single transaction. We cannot | |
617ba13b | 889 | * close off a transaction and start a new one between the ext4_get_block() |
dab291af | 890 | * and the commit_write(). So doing the jbd2_journal_start at the start of |
ac27a0ec DK |
891 | * prepare_write() is the right place. |
892 | * | |
36ade451 JK |
893 | * Also, this function can nest inside ext4_writepage(). In that case, we |
894 | * *know* that ext4_writepage() has generated enough buffer credits to do the | |
895 | * whole page. So we won't block on the journal in that case, which is good, | |
896 | * because the caller may be PF_MEMALLOC. | |
ac27a0ec | 897 | * |
617ba13b | 898 | * By accident, ext4 can be reentered when a transaction is open via |
ac27a0ec DK |
899 | * quota file writes. If we were to commit the transaction while thus |
900 | * reentered, there can be a deadlock - we would be holding a quota | |
901 | * lock, and the commit would never complete if another thread had a | |
902 | * transaction open and was blocking on the quota lock - a ranking | |
903 | * violation. | |
904 | * | |
dab291af | 905 | * So what we do is to rely on the fact that jbd2_journal_stop/journal_start |
ac27a0ec DK |
906 | * will _not_ run commit under these circumstances because handle->h_ref |
907 | * is elevated. We'll still have enough credits for the tiny quotafile | |
908 | * write. | |
909 | */ | |
f19d5870 TM |
910 | int do_journal_get_write_access(handle_t *handle, |
911 | struct buffer_head *bh) | |
ac27a0ec | 912 | { |
56d35a4c JK |
913 | int dirty = buffer_dirty(bh); |
914 | int ret; | |
915 | ||
ac27a0ec DK |
916 | if (!buffer_mapped(bh) || buffer_freed(bh)) |
917 | return 0; | |
56d35a4c | 918 | /* |
ebdec241 | 919 | * __block_write_begin() could have dirtied some buffers. Clean |
56d35a4c JK |
920 | * the dirty bit as jbd2_journal_get_write_access() could complain |
921 | * otherwise about fs integrity issues. Setting of the dirty bit | |
ebdec241 | 922 | * by __block_write_begin() isn't a real problem here as we clear |
56d35a4c JK |
923 | * the bit before releasing a page lock and thus writeback cannot |
924 | * ever write the buffer. | |
925 | */ | |
926 | if (dirty) | |
927 | clear_buffer_dirty(bh); | |
5d601255 | 928 | BUFFER_TRACE(bh, "get write access"); |
56d35a4c JK |
929 | ret = ext4_journal_get_write_access(handle, bh); |
930 | if (!ret && dirty) | |
931 | ret = ext4_handle_dirty_metadata(handle, NULL, bh); | |
932 | return ret; | |
ac27a0ec DK |
933 | } |
934 | ||
8b0f165f AP |
935 | static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock, |
936 | struct buffer_head *bh_result, int create); | |
bfc1af65 | 937 | static int ext4_write_begin(struct file *file, struct address_space *mapping, |
de9a55b8 TT |
938 | loff_t pos, unsigned len, unsigned flags, |
939 | struct page **pagep, void **fsdata) | |
ac27a0ec | 940 | { |
af5bc92d | 941 | struct inode *inode = mapping->host; |
1938a150 | 942 | int ret, needed_blocks; |
ac27a0ec DK |
943 | handle_t *handle; |
944 | int retries = 0; | |
af5bc92d | 945 | struct page *page; |
de9a55b8 | 946 | pgoff_t index; |
af5bc92d | 947 | unsigned from, to; |
bfc1af65 | 948 | |
9bffad1e | 949 | trace_ext4_write_begin(inode, pos, len, flags); |
1938a150 AK |
950 | /* |
951 | * Reserve one block more for addition to orphan list in case | |
952 | * we allocate blocks but write fails for some reason | |
953 | */ | |
954 | needed_blocks = ext4_writepage_trans_blocks(inode) + 1; | |
de9a55b8 | 955 | index = pos >> PAGE_CACHE_SHIFT; |
af5bc92d TT |
956 | from = pos & (PAGE_CACHE_SIZE - 1); |
957 | to = from + len; | |
ac27a0ec | 958 | |
f19d5870 TM |
959 | if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { |
960 | ret = ext4_try_to_write_inline_data(mapping, inode, pos, len, | |
961 | flags, pagep); | |
962 | if (ret < 0) | |
47564bfb TT |
963 | return ret; |
964 | if (ret == 1) | |
965 | return 0; | |
f19d5870 TM |
966 | } |
967 | ||
47564bfb TT |
968 | /* |
969 | * grab_cache_page_write_begin() can take a long time if the | |
970 | * system is thrashing due to memory pressure, or if the page | |
971 | * is being written back. So grab it first before we start | |
972 | * the transaction handle. This also allows us to allocate | |
973 | * the page (if needed) without using GFP_NOFS. | |
974 | */ | |
975 | retry_grab: | |
976 | page = grab_cache_page_write_begin(mapping, index, flags); | |
977 | if (!page) | |
978 | return -ENOMEM; | |
979 | unlock_page(page); | |
980 | ||
981 | retry_journal: | |
9924a92a | 982 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks); |
af5bc92d | 983 | if (IS_ERR(handle)) { |
47564bfb TT |
984 | page_cache_release(page); |
985 | return PTR_ERR(handle); | |
7479d2b9 | 986 | } |
ac27a0ec | 987 | |
47564bfb TT |
988 | lock_page(page); |
989 | if (page->mapping != mapping) { | |
990 | /* The page got truncated from under us */ | |
991 | unlock_page(page); | |
992 | page_cache_release(page); | |
cf108bca | 993 | ext4_journal_stop(handle); |
47564bfb | 994 | goto retry_grab; |
cf108bca | 995 | } |
7afe5aa5 DM |
996 | /* In case writeback began while the page was unlocked */ |
997 | wait_for_stable_page(page); | |
cf108bca | 998 | |
744692dc | 999 | if (ext4_should_dioread_nolock(inode)) |
6e1db88d | 1000 | ret = __block_write_begin(page, pos, len, ext4_get_block_write); |
744692dc | 1001 | else |
6e1db88d | 1002 | ret = __block_write_begin(page, pos, len, ext4_get_block); |
bfc1af65 NP |
1003 | |
1004 | if (!ret && ext4_should_journal_data(inode)) { | |
f19d5870 TM |
1005 | ret = ext4_walk_page_buffers(handle, page_buffers(page), |
1006 | from, to, NULL, | |
1007 | do_journal_get_write_access); | |
ac27a0ec | 1008 | } |
bfc1af65 NP |
1009 | |
1010 | if (ret) { | |
af5bc92d | 1011 | unlock_page(page); |
ae4d5372 | 1012 | /* |
6e1db88d | 1013 | * __block_write_begin may have instantiated a few blocks |
ae4d5372 AK |
1014 | * outside i_size. Trim these off again. Don't need |
1015 | * i_size_read because we hold i_mutex. | |
1938a150 AK |
1016 | * |
1017 | * Add inode to orphan list in case we crash before | |
1018 | * truncate finishes | |
ae4d5372 | 1019 | */ |
ffacfa7a | 1020 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
1938a150 AK |
1021 | ext4_orphan_add(handle, inode); |
1022 | ||
1023 | ext4_journal_stop(handle); | |
1024 | if (pos + len > inode->i_size) { | |
b9a4207d | 1025 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1026 | /* |
ffacfa7a | 1027 | * If truncate failed early the inode might |
1938a150 AK |
1028 | * still be on the orphan list; we need to |
1029 | * make sure the inode is removed from the | |
1030 | * orphan list in that case. | |
1031 | */ | |
1032 | if (inode->i_nlink) | |
1033 | ext4_orphan_del(NULL, inode); | |
1034 | } | |
bfc1af65 | 1035 | |
47564bfb TT |
1036 | if (ret == -ENOSPC && |
1037 | ext4_should_retry_alloc(inode->i_sb, &retries)) | |
1038 | goto retry_journal; | |
1039 | page_cache_release(page); | |
1040 | return ret; | |
1041 | } | |
1042 | *pagep = page; | |
ac27a0ec DK |
1043 | return ret; |
1044 | } | |
1045 | ||
bfc1af65 NP |
1046 | /* For write_end() in data=journal mode */ |
1047 | static int write_end_fn(handle_t *handle, struct buffer_head *bh) | |
ac27a0ec | 1048 | { |
13fca323 | 1049 | int ret; |
ac27a0ec DK |
1050 | if (!buffer_mapped(bh) || buffer_freed(bh)) |
1051 | return 0; | |
1052 | set_buffer_uptodate(bh); | |
13fca323 TT |
1053 | ret = ext4_handle_dirty_metadata(handle, NULL, bh); |
1054 | clear_buffer_meta(bh); | |
1055 | clear_buffer_prio(bh); | |
1056 | return ret; | |
ac27a0ec DK |
1057 | } |
1058 | ||
eed4333f ZL |
1059 | /* |
1060 | * We need to pick up the new inode size which generic_commit_write gave us | |
1061 | * `file' can be NULL - eg, when called from page_symlink(). | |
1062 | * | |
1063 | * ext4 never places buffers on inode->i_mapping->private_list. metadata | |
1064 | * buffers are managed internally. | |
1065 | */ | |
1066 | static int ext4_write_end(struct file *file, | |
1067 | struct address_space *mapping, | |
1068 | loff_t pos, unsigned len, unsigned copied, | |
1069 | struct page *page, void *fsdata) | |
f8514083 | 1070 | { |
f8514083 | 1071 | handle_t *handle = ext4_journal_current_handle(); |
eed4333f ZL |
1072 | struct inode *inode = mapping->host; |
1073 | int ret = 0, ret2; | |
1074 | int i_size_changed = 0; | |
1075 | ||
1076 | trace_ext4_write_end(inode, pos, len, copied); | |
1077 | if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE)) { | |
1078 | ret = ext4_jbd2_file_inode(handle, inode); | |
1079 | if (ret) { | |
1080 | unlock_page(page); | |
1081 | page_cache_release(page); | |
1082 | goto errout; | |
1083 | } | |
1084 | } | |
f8514083 | 1085 | |
42c832de TT |
1086 | if (ext4_has_inline_data(inode)) { |
1087 | ret = ext4_write_inline_data_end(inode, pos, len, | |
1088 | copied, page); | |
1089 | if (ret < 0) | |
1090 | goto errout; | |
1091 | copied = ret; | |
1092 | } else | |
f19d5870 TM |
1093 | copied = block_write_end(file, mapping, pos, |
1094 | len, copied, page, fsdata); | |
f8514083 AK |
1095 | |
1096 | /* | |
1097 | * No need to use i_size_read() here, the i_size | |
eed4333f | 1098 | * cannot change under us because we hole i_mutex. |
f8514083 AK |
1099 | * |
1100 | * But it's important to update i_size while still holding page lock: | |
1101 | * page writeout could otherwise come in and zero beyond i_size. | |
1102 | */ | |
1103 | if (pos + copied > inode->i_size) { | |
1104 | i_size_write(inode, pos + copied); | |
1105 | i_size_changed = 1; | |
1106 | } | |
1107 | ||
eed4333f | 1108 | if (pos + copied > EXT4_I(inode)->i_disksize) { |
f8514083 AK |
1109 | /* We need to mark inode dirty even if |
1110 | * new_i_size is less that inode->i_size | |
eed4333f | 1111 | * but greater than i_disksize. (hint delalloc) |
f8514083 AK |
1112 | */ |
1113 | ext4_update_i_disksize(inode, (pos + copied)); | |
1114 | i_size_changed = 1; | |
1115 | } | |
1116 | unlock_page(page); | |
1117 | page_cache_release(page); | |
1118 | ||
1119 | /* | |
1120 | * Don't mark the inode dirty under page lock. First, it unnecessarily | |
1121 | * makes the holding time of page lock longer. Second, it forces lock | |
1122 | * ordering of page lock and transaction start for journaling | |
1123 | * filesystems. | |
1124 | */ | |
1125 | if (i_size_changed) | |
1126 | ext4_mark_inode_dirty(handle, inode); | |
1127 | ||
ffacfa7a | 1128 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
f8514083 AK |
1129 | /* if we have allocated more blocks and copied |
1130 | * less. We will have blocks allocated outside | |
1131 | * inode->i_size. So truncate them | |
1132 | */ | |
1133 | ext4_orphan_add(handle, inode); | |
74d553aa | 1134 | errout: |
617ba13b | 1135 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1136 | if (!ret) |
1137 | ret = ret2; | |
bfc1af65 | 1138 | |
f8514083 | 1139 | if (pos + len > inode->i_size) { |
b9a4207d | 1140 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1141 | /* |
ffacfa7a | 1142 | * If truncate failed early the inode might still be |
f8514083 AK |
1143 | * on the orphan list; we need to make sure the inode |
1144 | * is removed from the orphan list in that case. | |
1145 | */ | |
1146 | if (inode->i_nlink) | |
1147 | ext4_orphan_del(NULL, inode); | |
1148 | } | |
1149 | ||
bfc1af65 | 1150 | return ret ? ret : copied; |
ac27a0ec DK |
1151 | } |
1152 | ||
bfc1af65 | 1153 | static int ext4_journalled_write_end(struct file *file, |
de9a55b8 TT |
1154 | struct address_space *mapping, |
1155 | loff_t pos, unsigned len, unsigned copied, | |
1156 | struct page *page, void *fsdata) | |
ac27a0ec | 1157 | { |
617ba13b | 1158 | handle_t *handle = ext4_journal_current_handle(); |
bfc1af65 | 1159 | struct inode *inode = mapping->host; |
ac27a0ec DK |
1160 | int ret = 0, ret2; |
1161 | int partial = 0; | |
bfc1af65 | 1162 | unsigned from, to; |
cf17fea6 | 1163 | loff_t new_i_size; |
ac27a0ec | 1164 | |
9bffad1e | 1165 | trace_ext4_journalled_write_end(inode, pos, len, copied); |
bfc1af65 NP |
1166 | from = pos & (PAGE_CACHE_SIZE - 1); |
1167 | to = from + len; | |
1168 | ||
441c8508 CW |
1169 | BUG_ON(!ext4_handle_valid(handle)); |
1170 | ||
3fdcfb66 TM |
1171 | if (ext4_has_inline_data(inode)) |
1172 | copied = ext4_write_inline_data_end(inode, pos, len, | |
1173 | copied, page); | |
1174 | else { | |
1175 | if (copied < len) { | |
1176 | if (!PageUptodate(page)) | |
1177 | copied = 0; | |
1178 | page_zero_new_buffers(page, from+copied, to); | |
1179 | } | |
ac27a0ec | 1180 | |
3fdcfb66 TM |
1181 | ret = ext4_walk_page_buffers(handle, page_buffers(page), from, |
1182 | to, &partial, write_end_fn); | |
1183 | if (!partial) | |
1184 | SetPageUptodate(page); | |
1185 | } | |
cf17fea6 AK |
1186 | new_i_size = pos + copied; |
1187 | if (new_i_size > inode->i_size) | |
bfc1af65 | 1188 | i_size_write(inode, pos+copied); |
19f5fb7a | 1189 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
2d859db3 | 1190 | EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; |
cf17fea6 AK |
1191 | if (new_i_size > EXT4_I(inode)->i_disksize) { |
1192 | ext4_update_i_disksize(inode, new_i_size); | |
617ba13b | 1193 | ret2 = ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
1194 | if (!ret) |
1195 | ret = ret2; | |
1196 | } | |
bfc1af65 | 1197 | |
cf108bca | 1198 | unlock_page(page); |
f8514083 | 1199 | page_cache_release(page); |
ffacfa7a | 1200 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
f8514083 AK |
1201 | /* if we have allocated more blocks and copied |
1202 | * less. We will have blocks allocated outside | |
1203 | * inode->i_size. So truncate them | |
1204 | */ | |
1205 | ext4_orphan_add(handle, inode); | |
1206 | ||
617ba13b | 1207 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1208 | if (!ret) |
1209 | ret = ret2; | |
f8514083 | 1210 | if (pos + len > inode->i_size) { |
b9a4207d | 1211 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1212 | /* |
ffacfa7a | 1213 | * If truncate failed early the inode might still be |
f8514083 AK |
1214 | * on the orphan list; we need to make sure the inode |
1215 | * is removed from the orphan list in that case. | |
1216 | */ | |
1217 | if (inode->i_nlink) | |
1218 | ext4_orphan_del(NULL, inode); | |
1219 | } | |
bfc1af65 NP |
1220 | |
1221 | return ret ? ret : copied; | |
ac27a0ec | 1222 | } |
d2a17637 | 1223 | |
386ad67c LC |
1224 | /* |
1225 | * Reserve a metadata for a single block located at lblock | |
1226 | */ | |
1227 | static int ext4_da_reserve_metadata(struct inode *inode, ext4_lblk_t lblock) | |
1228 | { | |
386ad67c LC |
1229 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
1230 | struct ext4_inode_info *ei = EXT4_I(inode); | |
1231 | unsigned int md_needed; | |
1232 | ext4_lblk_t save_last_lblock; | |
1233 | int save_len; | |
1234 | ||
1235 | /* | |
1236 | * recalculate the amount of metadata blocks to reserve | |
1237 | * in order to allocate nrblocks | |
1238 | * worse case is one extent per block | |
1239 | */ | |
386ad67c LC |
1240 | spin_lock(&ei->i_block_reservation_lock); |
1241 | /* | |
1242 | * ext4_calc_metadata_amount() has side effects, which we have | |
1243 | * to be prepared undo if we fail to claim space. | |
1244 | */ | |
1245 | save_len = ei->i_da_metadata_calc_len; | |
1246 | save_last_lblock = ei->i_da_metadata_calc_last_lblock; | |
1247 | md_needed = EXT4_NUM_B2C(sbi, | |
1248 | ext4_calc_metadata_amount(inode, lblock)); | |
1249 | trace_ext4_da_reserve_space(inode, md_needed); | |
1250 | ||
1251 | /* | |
1252 | * We do still charge estimated metadata to the sb though; | |
1253 | * we cannot afford to run out of free blocks. | |
1254 | */ | |
1255 | if (ext4_claim_free_clusters(sbi, md_needed, 0)) { | |
1256 | ei->i_da_metadata_calc_len = save_len; | |
1257 | ei->i_da_metadata_calc_last_lblock = save_last_lblock; | |
1258 | spin_unlock(&ei->i_block_reservation_lock); | |
386ad67c LC |
1259 | return -ENOSPC; |
1260 | } | |
1261 | ei->i_reserved_meta_blocks += md_needed; | |
1262 | spin_unlock(&ei->i_block_reservation_lock); | |
1263 | ||
1264 | return 0; /* success */ | |
1265 | } | |
1266 | ||
9d0be502 | 1267 | /* |
7b415bf6 | 1268 | * Reserve a single cluster located at lblock |
9d0be502 | 1269 | */ |
01f49d0b | 1270 | static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock) |
d2a17637 | 1271 | { |
60e58e0f | 1272 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
0637c6f4 | 1273 | struct ext4_inode_info *ei = EXT4_I(inode); |
7b415bf6 | 1274 | unsigned int md_needed; |
5dd4056d | 1275 | int ret; |
03179fe9 TT |
1276 | ext4_lblk_t save_last_lblock; |
1277 | int save_len; | |
1278 | ||
1279 | /* | |
1280 | * We will charge metadata quota at writeout time; this saves | |
1281 | * us from metadata over-estimation, though we may go over by | |
1282 | * a small amount in the end. Here we just reserve for data. | |
1283 | */ | |
1284 | ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1)); | |
1285 | if (ret) | |
1286 | return ret; | |
d2a17637 MC |
1287 | |
1288 | /* | |
1289 | * recalculate the amount of metadata blocks to reserve | |
1290 | * in order to allocate nrblocks | |
1291 | * worse case is one extent per block | |
1292 | */ | |
0637c6f4 | 1293 | spin_lock(&ei->i_block_reservation_lock); |
03179fe9 TT |
1294 | /* |
1295 | * ext4_calc_metadata_amount() has side effects, which we have | |
1296 | * to be prepared undo if we fail to claim space. | |
1297 | */ | |
1298 | save_len = ei->i_da_metadata_calc_len; | |
1299 | save_last_lblock = ei->i_da_metadata_calc_last_lblock; | |
7b415bf6 AK |
1300 | md_needed = EXT4_NUM_B2C(sbi, |
1301 | ext4_calc_metadata_amount(inode, lblock)); | |
f8ec9d68 | 1302 | trace_ext4_da_reserve_space(inode, md_needed); |
d2a17637 | 1303 | |
72b8ab9d ES |
1304 | /* |
1305 | * We do still charge estimated metadata to the sb though; | |
1306 | * we cannot afford to run out of free blocks. | |
1307 | */ | |
e7d5f315 | 1308 | if (ext4_claim_free_clusters(sbi, md_needed + 1, 0)) { |
03179fe9 TT |
1309 | ei->i_da_metadata_calc_len = save_len; |
1310 | ei->i_da_metadata_calc_last_lblock = save_last_lblock; | |
1311 | spin_unlock(&ei->i_block_reservation_lock); | |
03179fe9 | 1312 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1)); |
d2a17637 MC |
1313 | return -ENOSPC; |
1314 | } | |
9d0be502 | 1315 | ei->i_reserved_data_blocks++; |
0637c6f4 TT |
1316 | ei->i_reserved_meta_blocks += md_needed; |
1317 | spin_unlock(&ei->i_block_reservation_lock); | |
39bc680a | 1318 | |
d2a17637 MC |
1319 | return 0; /* success */ |
1320 | } | |
1321 | ||
12219aea | 1322 | static void ext4_da_release_space(struct inode *inode, int to_free) |
d2a17637 MC |
1323 | { |
1324 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0637c6f4 | 1325 | struct ext4_inode_info *ei = EXT4_I(inode); |
d2a17637 | 1326 | |
cd213226 MC |
1327 | if (!to_free) |
1328 | return; /* Nothing to release, exit */ | |
1329 | ||
d2a17637 | 1330 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); |
cd213226 | 1331 | |
5a58ec87 | 1332 | trace_ext4_da_release_space(inode, to_free); |
0637c6f4 | 1333 | if (unlikely(to_free > ei->i_reserved_data_blocks)) { |
cd213226 | 1334 | /* |
0637c6f4 TT |
1335 | * if there aren't enough reserved blocks, then the |
1336 | * counter is messed up somewhere. Since this | |
1337 | * function is called from invalidate page, it's | |
1338 | * harmless to return without any action. | |
cd213226 | 1339 | */ |
8de5c325 | 1340 | ext4_warning(inode->i_sb, "ext4_da_release_space: " |
0637c6f4 | 1341 | "ino %lu, to_free %d with only %d reserved " |
1084f252 | 1342 | "data blocks", inode->i_ino, to_free, |
0637c6f4 TT |
1343 | ei->i_reserved_data_blocks); |
1344 | WARN_ON(1); | |
1345 | to_free = ei->i_reserved_data_blocks; | |
cd213226 | 1346 | } |
0637c6f4 | 1347 | ei->i_reserved_data_blocks -= to_free; |
cd213226 | 1348 | |
0637c6f4 TT |
1349 | if (ei->i_reserved_data_blocks == 0) { |
1350 | /* | |
1351 | * We can release all of the reserved metadata blocks | |
1352 | * only when we have written all of the delayed | |
1353 | * allocation blocks. | |
7b415bf6 AK |
1354 | * Note that in case of bigalloc, i_reserved_meta_blocks, |
1355 | * i_reserved_data_blocks, etc. refer to number of clusters. | |
0637c6f4 | 1356 | */ |
57042651 | 1357 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, |
72b8ab9d | 1358 | ei->i_reserved_meta_blocks); |
ee5f4d9c | 1359 | ei->i_reserved_meta_blocks = 0; |
9d0be502 | 1360 | ei->i_da_metadata_calc_len = 0; |
0637c6f4 | 1361 | } |
d2a17637 | 1362 | |
72b8ab9d | 1363 | /* update fs dirty data blocks counter */ |
57042651 | 1364 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free); |
d2a17637 | 1365 | |
d2a17637 | 1366 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
60e58e0f | 1367 | |
7b415bf6 | 1368 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free)); |
d2a17637 MC |
1369 | } |
1370 | ||
1371 | static void ext4_da_page_release_reservation(struct page *page, | |
ca99fdd2 LC |
1372 | unsigned int offset, |
1373 | unsigned int length) | |
d2a17637 MC |
1374 | { |
1375 | int to_release = 0; | |
1376 | struct buffer_head *head, *bh; | |
1377 | unsigned int curr_off = 0; | |
7b415bf6 AK |
1378 | struct inode *inode = page->mapping->host; |
1379 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
ca99fdd2 | 1380 | unsigned int stop = offset + length; |
7b415bf6 | 1381 | int num_clusters; |
51865fda | 1382 | ext4_fsblk_t lblk; |
d2a17637 | 1383 | |
ca99fdd2 LC |
1384 | BUG_ON(stop > PAGE_CACHE_SIZE || stop < length); |
1385 | ||
d2a17637 MC |
1386 | head = page_buffers(page); |
1387 | bh = head; | |
1388 | do { | |
1389 | unsigned int next_off = curr_off + bh->b_size; | |
1390 | ||
ca99fdd2 LC |
1391 | if (next_off > stop) |
1392 | break; | |
1393 | ||
d2a17637 MC |
1394 | if ((offset <= curr_off) && (buffer_delay(bh))) { |
1395 | to_release++; | |
1396 | clear_buffer_delay(bh); | |
1397 | } | |
1398 | curr_off = next_off; | |
1399 | } while ((bh = bh->b_this_page) != head); | |
7b415bf6 | 1400 | |
51865fda ZL |
1401 | if (to_release) { |
1402 | lblk = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
1403 | ext4_es_remove_extent(inode, lblk, to_release); | |
1404 | } | |
1405 | ||
7b415bf6 AK |
1406 | /* If we have released all the blocks belonging to a cluster, then we |
1407 | * need to release the reserved space for that cluster. */ | |
1408 | num_clusters = EXT4_NUM_B2C(sbi, to_release); | |
1409 | while (num_clusters > 0) { | |
7b415bf6 AK |
1410 | lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) + |
1411 | ((num_clusters - 1) << sbi->s_cluster_bits); | |
1412 | if (sbi->s_cluster_ratio == 1 || | |
7d1b1fbc | 1413 | !ext4_find_delalloc_cluster(inode, lblk)) |
7b415bf6 AK |
1414 | ext4_da_release_space(inode, 1); |
1415 | ||
1416 | num_clusters--; | |
1417 | } | |
d2a17637 | 1418 | } |
ac27a0ec | 1419 | |
64769240 AT |
1420 | /* |
1421 | * Delayed allocation stuff | |
1422 | */ | |
1423 | ||
4e7ea81d JK |
1424 | struct mpage_da_data { |
1425 | struct inode *inode; | |
1426 | struct writeback_control *wbc; | |
6b523df4 | 1427 | |
4e7ea81d JK |
1428 | pgoff_t first_page; /* The first page to write */ |
1429 | pgoff_t next_page; /* Current page to examine */ | |
1430 | pgoff_t last_page; /* Last page to examine */ | |
791b7f08 | 1431 | /* |
4e7ea81d JK |
1432 | * Extent to map - this can be after first_page because that can be |
1433 | * fully mapped. We somewhat abuse m_flags to store whether the extent | |
1434 | * is delalloc or unwritten. | |
791b7f08 | 1435 | */ |
4e7ea81d JK |
1436 | struct ext4_map_blocks map; |
1437 | struct ext4_io_submit io_submit; /* IO submission data */ | |
1438 | }; | |
64769240 | 1439 | |
4e7ea81d JK |
1440 | static void mpage_release_unused_pages(struct mpage_da_data *mpd, |
1441 | bool invalidate) | |
c4a0c46e AK |
1442 | { |
1443 | int nr_pages, i; | |
1444 | pgoff_t index, end; | |
1445 | struct pagevec pvec; | |
1446 | struct inode *inode = mpd->inode; | |
1447 | struct address_space *mapping = inode->i_mapping; | |
4e7ea81d JK |
1448 | |
1449 | /* This is necessary when next_page == 0. */ | |
1450 | if (mpd->first_page >= mpd->next_page) | |
1451 | return; | |
c4a0c46e | 1452 | |
c7f5938a CW |
1453 | index = mpd->first_page; |
1454 | end = mpd->next_page - 1; | |
4e7ea81d JK |
1455 | if (invalidate) { |
1456 | ext4_lblk_t start, last; | |
1457 | start = index << (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
1458 | last = end << (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
1459 | ext4_es_remove_extent(inode, start, last - start + 1); | |
1460 | } | |
51865fda | 1461 | |
66bea92c | 1462 | pagevec_init(&pvec, 0); |
c4a0c46e AK |
1463 | while (index <= end) { |
1464 | nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE); | |
1465 | if (nr_pages == 0) | |
1466 | break; | |
1467 | for (i = 0; i < nr_pages; i++) { | |
1468 | struct page *page = pvec.pages[i]; | |
9b1d0998 | 1469 | if (page->index > end) |
c4a0c46e | 1470 | break; |
c4a0c46e AK |
1471 | BUG_ON(!PageLocked(page)); |
1472 | BUG_ON(PageWriteback(page)); | |
4e7ea81d JK |
1473 | if (invalidate) { |
1474 | block_invalidatepage(page, 0, PAGE_CACHE_SIZE); | |
1475 | ClearPageUptodate(page); | |
1476 | } | |
c4a0c46e AK |
1477 | unlock_page(page); |
1478 | } | |
9b1d0998 JK |
1479 | index = pvec.pages[nr_pages - 1]->index + 1; |
1480 | pagevec_release(&pvec); | |
c4a0c46e | 1481 | } |
c4a0c46e AK |
1482 | } |
1483 | ||
df22291f AK |
1484 | static void ext4_print_free_blocks(struct inode *inode) |
1485 | { | |
1486 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
92b97816 | 1487 | struct super_block *sb = inode->i_sb; |
f78ee70d | 1488 | struct ext4_inode_info *ei = EXT4_I(inode); |
92b97816 TT |
1489 | |
1490 | ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld", | |
5dee5437 | 1491 | EXT4_C2B(EXT4_SB(inode->i_sb), |
f78ee70d | 1492 | ext4_count_free_clusters(sb))); |
92b97816 TT |
1493 | ext4_msg(sb, KERN_CRIT, "Free/Dirty block details"); |
1494 | ext4_msg(sb, KERN_CRIT, "free_blocks=%lld", | |
f78ee70d | 1495 | (long long) EXT4_C2B(EXT4_SB(sb), |
57042651 | 1496 | percpu_counter_sum(&sbi->s_freeclusters_counter))); |
92b97816 | 1497 | ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld", |
f78ee70d | 1498 | (long long) EXT4_C2B(EXT4_SB(sb), |
7b415bf6 | 1499 | percpu_counter_sum(&sbi->s_dirtyclusters_counter))); |
92b97816 TT |
1500 | ext4_msg(sb, KERN_CRIT, "Block reservation details"); |
1501 | ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u", | |
f78ee70d | 1502 | ei->i_reserved_data_blocks); |
92b97816 | 1503 | ext4_msg(sb, KERN_CRIT, "i_reserved_meta_blocks=%u", |
f78ee70d LC |
1504 | ei->i_reserved_meta_blocks); |
1505 | ext4_msg(sb, KERN_CRIT, "i_allocated_meta_blocks=%u", | |
1506 | ei->i_allocated_meta_blocks); | |
df22291f AK |
1507 | return; |
1508 | } | |
1509 | ||
c364b22c | 1510 | static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh) |
29fa89d0 | 1511 | { |
c364b22c | 1512 | return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh); |
29fa89d0 AK |
1513 | } |
1514 | ||
5356f261 AK |
1515 | /* |
1516 | * This function is grabs code from the very beginning of | |
1517 | * ext4_map_blocks, but assumes that the caller is from delayed write | |
1518 | * time. This function looks up the requested blocks and sets the | |
1519 | * buffer delay bit under the protection of i_data_sem. | |
1520 | */ | |
1521 | static int ext4_da_map_blocks(struct inode *inode, sector_t iblock, | |
1522 | struct ext4_map_blocks *map, | |
1523 | struct buffer_head *bh) | |
1524 | { | |
d100eef2 | 1525 | struct extent_status es; |
5356f261 AK |
1526 | int retval; |
1527 | sector_t invalid_block = ~((sector_t) 0xffff); | |
921f266b DM |
1528 | #ifdef ES_AGGRESSIVE_TEST |
1529 | struct ext4_map_blocks orig_map; | |
1530 | ||
1531 | memcpy(&orig_map, map, sizeof(*map)); | |
1532 | #endif | |
5356f261 AK |
1533 | |
1534 | if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es)) | |
1535 | invalid_block = ~0; | |
1536 | ||
1537 | map->m_flags = 0; | |
1538 | ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u," | |
1539 | "logical block %lu\n", inode->i_ino, map->m_len, | |
1540 | (unsigned long) map->m_lblk); | |
d100eef2 ZL |
1541 | |
1542 | /* Lookup extent status tree firstly */ | |
1543 | if (ext4_es_lookup_extent(inode, iblock, &es)) { | |
63b99968 | 1544 | ext4_es_lru_add(inode); |
d100eef2 ZL |
1545 | if (ext4_es_is_hole(&es)) { |
1546 | retval = 0; | |
c8b459f4 | 1547 | down_read(&EXT4_I(inode)->i_data_sem); |
d100eef2 ZL |
1548 | goto add_delayed; |
1549 | } | |
1550 | ||
1551 | /* | |
1552 | * Delayed extent could be allocated by fallocate. | |
1553 | * So we need to check it. | |
1554 | */ | |
1555 | if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) { | |
1556 | map_bh(bh, inode->i_sb, invalid_block); | |
1557 | set_buffer_new(bh); | |
1558 | set_buffer_delay(bh); | |
1559 | return 0; | |
1560 | } | |
1561 | ||
1562 | map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk; | |
1563 | retval = es.es_len - (iblock - es.es_lblk); | |
1564 | if (retval > map->m_len) | |
1565 | retval = map->m_len; | |
1566 | map->m_len = retval; | |
1567 | if (ext4_es_is_written(&es)) | |
1568 | map->m_flags |= EXT4_MAP_MAPPED; | |
1569 | else if (ext4_es_is_unwritten(&es)) | |
1570 | map->m_flags |= EXT4_MAP_UNWRITTEN; | |
1571 | else | |
1572 | BUG_ON(1); | |
1573 | ||
921f266b DM |
1574 | #ifdef ES_AGGRESSIVE_TEST |
1575 | ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0); | |
1576 | #endif | |
d100eef2 ZL |
1577 | return retval; |
1578 | } | |
1579 | ||
5356f261 AK |
1580 | /* |
1581 | * Try to see if we can get the block without requesting a new | |
1582 | * file system block. | |
1583 | */ | |
c8b459f4 | 1584 | down_read(&EXT4_I(inode)->i_data_sem); |
9c3569b5 TM |
1585 | if (ext4_has_inline_data(inode)) { |
1586 | /* | |
1587 | * We will soon create blocks for this page, and let | |
1588 | * us pretend as if the blocks aren't allocated yet. | |
1589 | * In case of clusters, we have to handle the work | |
1590 | * of mapping from cluster so that the reserved space | |
1591 | * is calculated properly. | |
1592 | */ | |
1593 | if ((EXT4_SB(inode->i_sb)->s_cluster_ratio > 1) && | |
1594 | ext4_find_delalloc_cluster(inode, map->m_lblk)) | |
1595 | map->m_flags |= EXT4_MAP_FROM_CLUSTER; | |
1596 | retval = 0; | |
1597 | } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
d100eef2 ZL |
1598 | retval = ext4_ext_map_blocks(NULL, inode, map, |
1599 | EXT4_GET_BLOCKS_NO_PUT_HOLE); | |
5356f261 | 1600 | else |
d100eef2 ZL |
1601 | retval = ext4_ind_map_blocks(NULL, inode, map, |
1602 | EXT4_GET_BLOCKS_NO_PUT_HOLE); | |
5356f261 | 1603 | |
d100eef2 | 1604 | add_delayed: |
5356f261 | 1605 | if (retval == 0) { |
f7fec032 | 1606 | int ret; |
5356f261 AK |
1607 | /* |
1608 | * XXX: __block_prepare_write() unmaps passed block, | |
1609 | * is it OK? | |
1610 | */ | |
386ad67c LC |
1611 | /* |
1612 | * If the block was allocated from previously allocated cluster, | |
1613 | * then we don't need to reserve it again. However we still need | |
1614 | * to reserve metadata for every block we're going to write. | |
1615 | */ | |
5356f261 | 1616 | if (!(map->m_flags & EXT4_MAP_FROM_CLUSTER)) { |
f7fec032 ZL |
1617 | ret = ext4_da_reserve_space(inode, iblock); |
1618 | if (ret) { | |
5356f261 | 1619 | /* not enough space to reserve */ |
f7fec032 | 1620 | retval = ret; |
5356f261 | 1621 | goto out_unlock; |
f7fec032 | 1622 | } |
386ad67c LC |
1623 | } else { |
1624 | ret = ext4_da_reserve_metadata(inode, iblock); | |
1625 | if (ret) { | |
1626 | /* not enough space to reserve */ | |
1627 | retval = ret; | |
1628 | goto out_unlock; | |
1629 | } | |
5356f261 AK |
1630 | } |
1631 | ||
f7fec032 ZL |
1632 | ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, |
1633 | ~0, EXTENT_STATUS_DELAYED); | |
1634 | if (ret) { | |
1635 | retval = ret; | |
51865fda | 1636 | goto out_unlock; |
f7fec032 | 1637 | } |
51865fda | 1638 | |
5356f261 AK |
1639 | /* Clear EXT4_MAP_FROM_CLUSTER flag since its purpose is served |
1640 | * and it should not appear on the bh->b_state. | |
1641 | */ | |
1642 | map->m_flags &= ~EXT4_MAP_FROM_CLUSTER; | |
1643 | ||
1644 | map_bh(bh, inode->i_sb, invalid_block); | |
1645 | set_buffer_new(bh); | |
1646 | set_buffer_delay(bh); | |
f7fec032 ZL |
1647 | } else if (retval > 0) { |
1648 | int ret; | |
3be78c73 | 1649 | unsigned int status; |
f7fec032 | 1650 | |
44fb851d ZL |
1651 | if (unlikely(retval != map->m_len)) { |
1652 | ext4_warning(inode->i_sb, | |
1653 | "ES len assertion failed for inode " | |
1654 | "%lu: retval %d != map->m_len %d", | |
1655 | inode->i_ino, retval, map->m_len); | |
1656 | WARN_ON(1); | |
921f266b | 1657 | } |
921f266b | 1658 | |
f7fec032 ZL |
1659 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
1660 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; | |
1661 | ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, | |
1662 | map->m_pblk, status); | |
1663 | if (ret != 0) | |
1664 | retval = ret; | |
5356f261 AK |
1665 | } |
1666 | ||
1667 | out_unlock: | |
1668 | up_read((&EXT4_I(inode)->i_data_sem)); | |
1669 | ||
1670 | return retval; | |
1671 | } | |
1672 | ||
64769240 | 1673 | /* |
b920c755 TT |
1674 | * This is a special get_blocks_t callback which is used by |
1675 | * ext4_da_write_begin(). It will either return mapped block or | |
1676 | * reserve space for a single block. | |
29fa89d0 AK |
1677 | * |
1678 | * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set. | |
1679 | * We also have b_blocknr = -1 and b_bdev initialized properly | |
1680 | * | |
1681 | * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set. | |
1682 | * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev | |
1683 | * initialized properly. | |
64769240 | 1684 | */ |
9c3569b5 TM |
1685 | int ext4_da_get_block_prep(struct inode *inode, sector_t iblock, |
1686 | struct buffer_head *bh, int create) | |
64769240 | 1687 | { |
2ed88685 | 1688 | struct ext4_map_blocks map; |
64769240 AT |
1689 | int ret = 0; |
1690 | ||
1691 | BUG_ON(create == 0); | |
2ed88685 TT |
1692 | BUG_ON(bh->b_size != inode->i_sb->s_blocksize); |
1693 | ||
1694 | map.m_lblk = iblock; | |
1695 | map.m_len = 1; | |
64769240 AT |
1696 | |
1697 | /* | |
1698 | * first, we need to know whether the block is allocated already | |
1699 | * preallocated blocks are unmapped but should treated | |
1700 | * the same as allocated blocks. | |
1701 | */ | |
5356f261 AK |
1702 | ret = ext4_da_map_blocks(inode, iblock, &map, bh); |
1703 | if (ret <= 0) | |
2ed88685 | 1704 | return ret; |
64769240 | 1705 | |
2ed88685 TT |
1706 | map_bh(bh, inode->i_sb, map.m_pblk); |
1707 | bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags; | |
1708 | ||
1709 | if (buffer_unwritten(bh)) { | |
1710 | /* A delayed write to unwritten bh should be marked | |
1711 | * new and mapped. Mapped ensures that we don't do | |
1712 | * get_block multiple times when we write to the same | |
1713 | * offset and new ensures that we do proper zero out | |
1714 | * for partial write. | |
1715 | */ | |
1716 | set_buffer_new(bh); | |
c8205636 | 1717 | set_buffer_mapped(bh); |
2ed88685 TT |
1718 | } |
1719 | return 0; | |
64769240 | 1720 | } |
61628a3f | 1721 | |
62e086be AK |
1722 | static int bget_one(handle_t *handle, struct buffer_head *bh) |
1723 | { | |
1724 | get_bh(bh); | |
1725 | return 0; | |
1726 | } | |
1727 | ||
1728 | static int bput_one(handle_t *handle, struct buffer_head *bh) | |
1729 | { | |
1730 | put_bh(bh); | |
1731 | return 0; | |
1732 | } | |
1733 | ||
1734 | static int __ext4_journalled_writepage(struct page *page, | |
62e086be AK |
1735 | unsigned int len) |
1736 | { | |
1737 | struct address_space *mapping = page->mapping; | |
1738 | struct inode *inode = mapping->host; | |
3fdcfb66 | 1739 | struct buffer_head *page_bufs = NULL; |
62e086be | 1740 | handle_t *handle = NULL; |
3fdcfb66 TM |
1741 | int ret = 0, err = 0; |
1742 | int inline_data = ext4_has_inline_data(inode); | |
1743 | struct buffer_head *inode_bh = NULL; | |
62e086be | 1744 | |
cb20d518 | 1745 | ClearPageChecked(page); |
3fdcfb66 TM |
1746 | |
1747 | if (inline_data) { | |
1748 | BUG_ON(page->index != 0); | |
1749 | BUG_ON(len > ext4_get_max_inline_size(inode)); | |
1750 | inode_bh = ext4_journalled_write_inline_data(inode, len, page); | |
1751 | if (inode_bh == NULL) | |
1752 | goto out; | |
1753 | } else { | |
1754 | page_bufs = page_buffers(page); | |
1755 | if (!page_bufs) { | |
1756 | BUG(); | |
1757 | goto out; | |
1758 | } | |
1759 | ext4_walk_page_buffers(handle, page_bufs, 0, len, | |
1760 | NULL, bget_one); | |
1761 | } | |
62e086be AK |
1762 | /* As soon as we unlock the page, it can go away, but we have |
1763 | * references to buffers so we are safe */ | |
1764 | unlock_page(page); | |
1765 | ||
9924a92a TT |
1766 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
1767 | ext4_writepage_trans_blocks(inode)); | |
62e086be AK |
1768 | if (IS_ERR(handle)) { |
1769 | ret = PTR_ERR(handle); | |
1770 | goto out; | |
1771 | } | |
1772 | ||
441c8508 CW |
1773 | BUG_ON(!ext4_handle_valid(handle)); |
1774 | ||
3fdcfb66 | 1775 | if (inline_data) { |
5d601255 | 1776 | BUFFER_TRACE(inode_bh, "get write access"); |
3fdcfb66 | 1777 | ret = ext4_journal_get_write_access(handle, inode_bh); |
62e086be | 1778 | |
3fdcfb66 TM |
1779 | err = ext4_handle_dirty_metadata(handle, inode, inode_bh); |
1780 | ||
1781 | } else { | |
1782 | ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL, | |
1783 | do_journal_get_write_access); | |
1784 | ||
1785 | err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL, | |
1786 | write_end_fn); | |
1787 | } | |
62e086be AK |
1788 | if (ret == 0) |
1789 | ret = err; | |
2d859db3 | 1790 | EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; |
62e086be AK |
1791 | err = ext4_journal_stop(handle); |
1792 | if (!ret) | |
1793 | ret = err; | |
1794 | ||
3fdcfb66 | 1795 | if (!ext4_has_inline_data(inode)) |
8c9367fd | 1796 | ext4_walk_page_buffers(NULL, page_bufs, 0, len, |
3fdcfb66 | 1797 | NULL, bput_one); |
19f5fb7a | 1798 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
62e086be | 1799 | out: |
3fdcfb66 | 1800 | brelse(inode_bh); |
62e086be AK |
1801 | return ret; |
1802 | } | |
1803 | ||
61628a3f | 1804 | /* |
43ce1d23 AK |
1805 | * Note that we don't need to start a transaction unless we're journaling data |
1806 | * because we should have holes filled from ext4_page_mkwrite(). We even don't | |
1807 | * need to file the inode to the transaction's list in ordered mode because if | |
1808 | * we are writing back data added by write(), the inode is already there and if | |
25985edc | 1809 | * we are writing back data modified via mmap(), no one guarantees in which |
43ce1d23 AK |
1810 | * transaction the data will hit the disk. In case we are journaling data, we |
1811 | * cannot start transaction directly because transaction start ranks above page | |
1812 | * lock so we have to do some magic. | |
1813 | * | |
b920c755 | 1814 | * This function can get called via... |
20970ba6 | 1815 | * - ext4_writepages after taking page lock (have journal handle) |
b920c755 | 1816 | * - journal_submit_inode_data_buffers (no journal handle) |
f6463b0d | 1817 | * - shrink_page_list via the kswapd/direct reclaim (no journal handle) |
b920c755 | 1818 | * - grab_page_cache when doing write_begin (have journal handle) |
43ce1d23 AK |
1819 | * |
1820 | * We don't do any block allocation in this function. If we have page with | |
1821 | * multiple blocks we need to write those buffer_heads that are mapped. This | |
1822 | * is important for mmaped based write. So if we do with blocksize 1K | |
1823 | * truncate(f, 1024); | |
1824 | * a = mmap(f, 0, 4096); | |
1825 | * a[0] = 'a'; | |
1826 | * truncate(f, 4096); | |
1827 | * we have in the page first buffer_head mapped via page_mkwrite call back | |
90802ed9 | 1828 | * but other buffer_heads would be unmapped but dirty (dirty done via the |
43ce1d23 AK |
1829 | * do_wp_page). So writepage should write the first block. If we modify |
1830 | * the mmap area beyond 1024 we will again get a page_fault and the | |
1831 | * page_mkwrite callback will do the block allocation and mark the | |
1832 | * buffer_heads mapped. | |
1833 | * | |
1834 | * We redirty the page if we have any buffer_heads that is either delay or | |
1835 | * unwritten in the page. | |
1836 | * | |
1837 | * We can get recursively called as show below. | |
1838 | * | |
1839 | * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() -> | |
1840 | * ext4_writepage() | |
1841 | * | |
1842 | * But since we don't do any block allocation we should not deadlock. | |
1843 | * Page also have the dirty flag cleared so we don't get recurive page_lock. | |
61628a3f | 1844 | */ |
43ce1d23 | 1845 | static int ext4_writepage(struct page *page, |
62e086be | 1846 | struct writeback_control *wbc) |
64769240 | 1847 | { |
f8bec370 | 1848 | int ret = 0; |
61628a3f | 1849 | loff_t size; |
498e5f24 | 1850 | unsigned int len; |
744692dc | 1851 | struct buffer_head *page_bufs = NULL; |
61628a3f | 1852 | struct inode *inode = page->mapping->host; |
36ade451 | 1853 | struct ext4_io_submit io_submit; |
1c8349a1 | 1854 | bool keep_towrite = false; |
61628a3f | 1855 | |
a9c667f8 | 1856 | trace_ext4_writepage(page); |
f0e6c985 AK |
1857 | size = i_size_read(inode); |
1858 | if (page->index == size >> PAGE_CACHE_SHIFT) | |
1859 | len = size & ~PAGE_CACHE_MASK; | |
1860 | else | |
1861 | len = PAGE_CACHE_SIZE; | |
64769240 | 1862 | |
a42afc5f | 1863 | page_bufs = page_buffers(page); |
a42afc5f | 1864 | /* |
fe386132 JK |
1865 | * We cannot do block allocation or other extent handling in this |
1866 | * function. If there are buffers needing that, we have to redirty | |
1867 | * the page. But we may reach here when we do a journal commit via | |
1868 | * journal_submit_inode_data_buffers() and in that case we must write | |
1869 | * allocated buffers to achieve data=ordered mode guarantees. | |
a42afc5f | 1870 | */ |
f19d5870 TM |
1871 | if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL, |
1872 | ext4_bh_delay_or_unwritten)) { | |
f8bec370 | 1873 | redirty_page_for_writepage(wbc, page); |
fe386132 JK |
1874 | if (current->flags & PF_MEMALLOC) { |
1875 | /* | |
1876 | * For memory cleaning there's no point in writing only | |
1877 | * some buffers. So just bail out. Warn if we came here | |
1878 | * from direct reclaim. | |
1879 | */ | |
1880 | WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) | |
1881 | == PF_MEMALLOC); | |
f0e6c985 AK |
1882 | unlock_page(page); |
1883 | return 0; | |
1884 | } | |
1c8349a1 | 1885 | keep_towrite = true; |
a42afc5f | 1886 | } |
64769240 | 1887 | |
cb20d518 | 1888 | if (PageChecked(page) && ext4_should_journal_data(inode)) |
43ce1d23 AK |
1889 | /* |
1890 | * It's mmapped pagecache. Add buffers and journal it. There | |
1891 | * doesn't seem much point in redirtying the page here. | |
1892 | */ | |
3f0ca309 | 1893 | return __ext4_journalled_writepage(page, len); |
43ce1d23 | 1894 | |
97a851ed JK |
1895 | ext4_io_submit_init(&io_submit, wbc); |
1896 | io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS); | |
1897 | if (!io_submit.io_end) { | |
1898 | redirty_page_for_writepage(wbc, page); | |
1899 | unlock_page(page); | |
1900 | return -ENOMEM; | |
1901 | } | |
1c8349a1 | 1902 | ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite); |
36ade451 | 1903 | ext4_io_submit(&io_submit); |
97a851ed JK |
1904 | /* Drop io_end reference we got from init */ |
1905 | ext4_put_io_end_defer(io_submit.io_end); | |
64769240 AT |
1906 | return ret; |
1907 | } | |
1908 | ||
5f1132b2 JK |
1909 | static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page) |
1910 | { | |
1911 | int len; | |
1912 | loff_t size = i_size_read(mpd->inode); | |
1913 | int err; | |
1914 | ||
1915 | BUG_ON(page->index != mpd->first_page); | |
1916 | if (page->index == size >> PAGE_CACHE_SHIFT) | |
1917 | len = size & ~PAGE_CACHE_MASK; | |
1918 | else | |
1919 | len = PAGE_CACHE_SIZE; | |
1920 | clear_page_dirty_for_io(page); | |
1c8349a1 | 1921 | err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false); |
5f1132b2 JK |
1922 | if (!err) |
1923 | mpd->wbc->nr_to_write--; | |
1924 | mpd->first_page++; | |
1925 | ||
1926 | return err; | |
1927 | } | |
1928 | ||
4e7ea81d JK |
1929 | #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay)) |
1930 | ||
61628a3f | 1931 | /* |
fffb2739 JK |
1932 | * mballoc gives us at most this number of blocks... |
1933 | * XXX: That seems to be only a limitation of ext4_mb_normalize_request(). | |
70261f56 | 1934 | * The rest of mballoc seems to handle chunks up to full group size. |
61628a3f | 1935 | */ |
fffb2739 | 1936 | #define MAX_WRITEPAGES_EXTENT_LEN 2048 |
525f4ed8 | 1937 | |
4e7ea81d JK |
1938 | /* |
1939 | * mpage_add_bh_to_extent - try to add bh to extent of blocks to map | |
1940 | * | |
1941 | * @mpd - extent of blocks | |
1942 | * @lblk - logical number of the block in the file | |
09930042 | 1943 | * @bh - buffer head we want to add to the extent |
4e7ea81d | 1944 | * |
09930042 JK |
1945 | * The function is used to collect contig. blocks in the same state. If the |
1946 | * buffer doesn't require mapping for writeback and we haven't started the | |
1947 | * extent of buffers to map yet, the function returns 'true' immediately - the | |
1948 | * caller can write the buffer right away. Otherwise the function returns true | |
1949 | * if the block has been added to the extent, false if the block couldn't be | |
1950 | * added. | |
4e7ea81d | 1951 | */ |
09930042 JK |
1952 | static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk, |
1953 | struct buffer_head *bh) | |
4e7ea81d JK |
1954 | { |
1955 | struct ext4_map_blocks *map = &mpd->map; | |
1956 | ||
09930042 JK |
1957 | /* Buffer that doesn't need mapping for writeback? */ |
1958 | if (!buffer_dirty(bh) || !buffer_mapped(bh) || | |
1959 | (!buffer_delay(bh) && !buffer_unwritten(bh))) { | |
1960 | /* So far no extent to map => we write the buffer right away */ | |
1961 | if (map->m_len == 0) | |
1962 | return true; | |
1963 | return false; | |
1964 | } | |
4e7ea81d JK |
1965 | |
1966 | /* First block in the extent? */ | |
1967 | if (map->m_len == 0) { | |
1968 | map->m_lblk = lblk; | |
1969 | map->m_len = 1; | |
09930042 JK |
1970 | map->m_flags = bh->b_state & BH_FLAGS; |
1971 | return true; | |
4e7ea81d JK |
1972 | } |
1973 | ||
09930042 JK |
1974 | /* Don't go larger than mballoc is willing to allocate */ |
1975 | if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN) | |
1976 | return false; | |
1977 | ||
4e7ea81d JK |
1978 | /* Can we merge the block to our big extent? */ |
1979 | if (lblk == map->m_lblk + map->m_len && | |
09930042 | 1980 | (bh->b_state & BH_FLAGS) == map->m_flags) { |
4e7ea81d | 1981 | map->m_len++; |
09930042 | 1982 | return true; |
4e7ea81d | 1983 | } |
09930042 | 1984 | return false; |
4e7ea81d JK |
1985 | } |
1986 | ||
5f1132b2 JK |
1987 | /* |
1988 | * mpage_process_page_bufs - submit page buffers for IO or add them to extent | |
1989 | * | |
1990 | * @mpd - extent of blocks for mapping | |
1991 | * @head - the first buffer in the page | |
1992 | * @bh - buffer we should start processing from | |
1993 | * @lblk - logical number of the block in the file corresponding to @bh | |
1994 | * | |
1995 | * Walk through page buffers from @bh upto @head (exclusive) and either submit | |
1996 | * the page for IO if all buffers in this page were mapped and there's no | |
1997 | * accumulated extent of buffers to map or add buffers in the page to the | |
1998 | * extent of buffers to map. The function returns 1 if the caller can continue | |
1999 | * by processing the next page, 0 if it should stop adding buffers to the | |
2000 | * extent to map because we cannot extend it anymore. It can also return value | |
2001 | * < 0 in case of error during IO submission. | |
2002 | */ | |
2003 | static int mpage_process_page_bufs(struct mpage_da_data *mpd, | |
2004 | struct buffer_head *head, | |
2005 | struct buffer_head *bh, | |
2006 | ext4_lblk_t lblk) | |
4e7ea81d JK |
2007 | { |
2008 | struct inode *inode = mpd->inode; | |
5f1132b2 | 2009 | int err; |
4e7ea81d JK |
2010 | ext4_lblk_t blocks = (i_size_read(inode) + (1 << inode->i_blkbits) - 1) |
2011 | >> inode->i_blkbits; | |
2012 | ||
2013 | do { | |
2014 | BUG_ON(buffer_locked(bh)); | |
2015 | ||
09930042 | 2016 | if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) { |
4e7ea81d JK |
2017 | /* Found extent to map? */ |
2018 | if (mpd->map.m_len) | |
5f1132b2 | 2019 | return 0; |
09930042 | 2020 | /* Everything mapped so far and we hit EOF */ |
5f1132b2 | 2021 | break; |
4e7ea81d | 2022 | } |
4e7ea81d | 2023 | } while (lblk++, (bh = bh->b_this_page) != head); |
5f1132b2 JK |
2024 | /* So far everything mapped? Submit the page for IO. */ |
2025 | if (mpd->map.m_len == 0) { | |
2026 | err = mpage_submit_page(mpd, head->b_page); | |
2027 | if (err < 0) | |
2028 | return err; | |
2029 | } | |
2030 | return lblk < blocks; | |
4e7ea81d JK |
2031 | } |
2032 | ||
2033 | /* | |
2034 | * mpage_map_buffers - update buffers corresponding to changed extent and | |
2035 | * submit fully mapped pages for IO | |
2036 | * | |
2037 | * @mpd - description of extent to map, on return next extent to map | |
2038 | * | |
2039 | * Scan buffers corresponding to changed extent (we expect corresponding pages | |
2040 | * to be already locked) and update buffer state according to new extent state. | |
2041 | * We map delalloc buffers to their physical location, clear unwritten bits, | |
556615dc | 2042 | * and mark buffers as uninit when we perform writes to unwritten extents |
4e7ea81d JK |
2043 | * and do extent conversion after IO is finished. If the last page is not fully |
2044 | * mapped, we update @map to the next extent in the last page that needs | |
2045 | * mapping. Otherwise we submit the page for IO. | |
2046 | */ | |
2047 | static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd) | |
2048 | { | |
2049 | struct pagevec pvec; | |
2050 | int nr_pages, i; | |
2051 | struct inode *inode = mpd->inode; | |
2052 | struct buffer_head *head, *bh; | |
2053 | int bpp_bits = PAGE_CACHE_SHIFT - inode->i_blkbits; | |
4e7ea81d JK |
2054 | pgoff_t start, end; |
2055 | ext4_lblk_t lblk; | |
2056 | sector_t pblock; | |
2057 | int err; | |
2058 | ||
2059 | start = mpd->map.m_lblk >> bpp_bits; | |
2060 | end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits; | |
2061 | lblk = start << bpp_bits; | |
2062 | pblock = mpd->map.m_pblk; | |
2063 | ||
2064 | pagevec_init(&pvec, 0); | |
2065 | while (start <= end) { | |
2066 | nr_pages = pagevec_lookup(&pvec, inode->i_mapping, start, | |
2067 | PAGEVEC_SIZE); | |
2068 | if (nr_pages == 0) | |
2069 | break; | |
2070 | for (i = 0; i < nr_pages; i++) { | |
2071 | struct page *page = pvec.pages[i]; | |
2072 | ||
2073 | if (page->index > end) | |
2074 | break; | |
70261f56 | 2075 | /* Up to 'end' pages must be contiguous */ |
4e7ea81d JK |
2076 | BUG_ON(page->index != start); |
2077 | bh = head = page_buffers(page); | |
2078 | do { | |
2079 | if (lblk < mpd->map.m_lblk) | |
2080 | continue; | |
2081 | if (lblk >= mpd->map.m_lblk + mpd->map.m_len) { | |
2082 | /* | |
2083 | * Buffer after end of mapped extent. | |
2084 | * Find next buffer in the page to map. | |
2085 | */ | |
2086 | mpd->map.m_len = 0; | |
2087 | mpd->map.m_flags = 0; | |
5f1132b2 JK |
2088 | /* |
2089 | * FIXME: If dioread_nolock supports | |
2090 | * blocksize < pagesize, we need to make | |
2091 | * sure we add size mapped so far to | |
2092 | * io_end->size as the following call | |
2093 | * can submit the page for IO. | |
2094 | */ | |
2095 | err = mpage_process_page_bufs(mpd, head, | |
2096 | bh, lblk); | |
4e7ea81d | 2097 | pagevec_release(&pvec); |
5f1132b2 JK |
2098 | if (err > 0) |
2099 | err = 0; | |
2100 | return err; | |
4e7ea81d JK |
2101 | } |
2102 | if (buffer_delay(bh)) { | |
2103 | clear_buffer_delay(bh); | |
2104 | bh->b_blocknr = pblock++; | |
2105 | } | |
4e7ea81d | 2106 | clear_buffer_unwritten(bh); |
5f1132b2 | 2107 | } while (lblk++, (bh = bh->b_this_page) != head); |
4e7ea81d JK |
2108 | |
2109 | /* | |
2110 | * FIXME: This is going to break if dioread_nolock | |
2111 | * supports blocksize < pagesize as we will try to | |
2112 | * convert potentially unmapped parts of inode. | |
2113 | */ | |
2114 | mpd->io_submit.io_end->size += PAGE_CACHE_SIZE; | |
2115 | /* Page fully mapped - let IO run! */ | |
2116 | err = mpage_submit_page(mpd, page); | |
2117 | if (err < 0) { | |
2118 | pagevec_release(&pvec); | |
2119 | return err; | |
2120 | } | |
2121 | start++; | |
2122 | } | |
2123 | pagevec_release(&pvec); | |
2124 | } | |
2125 | /* Extent fully mapped and matches with page boundary. We are done. */ | |
2126 | mpd->map.m_len = 0; | |
2127 | mpd->map.m_flags = 0; | |
2128 | return 0; | |
2129 | } | |
2130 | ||
2131 | static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd) | |
2132 | { | |
2133 | struct inode *inode = mpd->inode; | |
2134 | struct ext4_map_blocks *map = &mpd->map; | |
2135 | int get_blocks_flags; | |
090f32ee | 2136 | int err, dioread_nolock; |
4e7ea81d JK |
2137 | |
2138 | trace_ext4_da_write_pages_extent(inode, map); | |
2139 | /* | |
2140 | * Call ext4_map_blocks() to allocate any delayed allocation blocks, or | |
556615dc | 2141 | * to convert an unwritten extent to be initialized (in the case |
4e7ea81d JK |
2142 | * where we have written into one or more preallocated blocks). It is |
2143 | * possible that we're going to need more metadata blocks than | |
2144 | * previously reserved. However we must not fail because we're in | |
2145 | * writeback and there is nothing we can do about it so it might result | |
2146 | * in data loss. So use reserved blocks to allocate metadata if | |
2147 | * possible. | |
2148 | * | |
2149 | * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if the blocks | |
2150 | * in question are delalloc blocks. This affects functions in many | |
2151 | * different parts of the allocation call path. This flag exists | |
2152 | * primarily because we don't want to change *many* call functions, so | |
2153 | * ext4_map_blocks() will set the EXT4_STATE_DELALLOC_RESERVED flag | |
2154 | * once the inode's allocation semaphore is taken. | |
2155 | */ | |
2156 | get_blocks_flags = EXT4_GET_BLOCKS_CREATE | | |
2157 | EXT4_GET_BLOCKS_METADATA_NOFAIL; | |
090f32ee LC |
2158 | dioread_nolock = ext4_should_dioread_nolock(inode); |
2159 | if (dioread_nolock) | |
4e7ea81d JK |
2160 | get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT; |
2161 | if (map->m_flags & (1 << BH_Delay)) | |
2162 | get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE; | |
2163 | ||
2164 | err = ext4_map_blocks(handle, inode, map, get_blocks_flags); | |
2165 | if (err < 0) | |
2166 | return err; | |
090f32ee | 2167 | if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) { |
6b523df4 JK |
2168 | if (!mpd->io_submit.io_end->handle && |
2169 | ext4_handle_valid(handle)) { | |
2170 | mpd->io_submit.io_end->handle = handle->h_rsv_handle; | |
2171 | handle->h_rsv_handle = NULL; | |
2172 | } | |
3613d228 | 2173 | ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end); |
6b523df4 | 2174 | } |
4e7ea81d JK |
2175 | |
2176 | BUG_ON(map->m_len == 0); | |
2177 | if (map->m_flags & EXT4_MAP_NEW) { | |
2178 | struct block_device *bdev = inode->i_sb->s_bdev; | |
2179 | int i; | |
2180 | ||
2181 | for (i = 0; i < map->m_len; i++) | |
2182 | unmap_underlying_metadata(bdev, map->m_pblk + i); | |
2183 | } | |
2184 | return 0; | |
2185 | } | |
2186 | ||
2187 | /* | |
2188 | * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length | |
2189 | * mpd->len and submit pages underlying it for IO | |
2190 | * | |
2191 | * @handle - handle for journal operations | |
2192 | * @mpd - extent to map | |
7534e854 JK |
2193 | * @give_up_on_write - we set this to true iff there is a fatal error and there |
2194 | * is no hope of writing the data. The caller should discard | |
2195 | * dirty pages to avoid infinite loops. | |
4e7ea81d JK |
2196 | * |
2197 | * The function maps extent starting at mpd->lblk of length mpd->len. If it is | |
2198 | * delayed, blocks are allocated, if it is unwritten, we may need to convert | |
2199 | * them to initialized or split the described range from larger unwritten | |
2200 | * extent. Note that we need not map all the described range since allocation | |
2201 | * can return less blocks or the range is covered by more unwritten extents. We | |
2202 | * cannot map more because we are limited by reserved transaction credits. On | |
2203 | * the other hand we always make sure that the last touched page is fully | |
2204 | * mapped so that it can be written out (and thus forward progress is | |
2205 | * guaranteed). After mapping we submit all mapped pages for IO. | |
2206 | */ | |
2207 | static int mpage_map_and_submit_extent(handle_t *handle, | |
cb530541 TT |
2208 | struct mpage_da_data *mpd, |
2209 | bool *give_up_on_write) | |
4e7ea81d JK |
2210 | { |
2211 | struct inode *inode = mpd->inode; | |
2212 | struct ext4_map_blocks *map = &mpd->map; | |
2213 | int err; | |
2214 | loff_t disksize; | |
2215 | ||
2216 | mpd->io_submit.io_end->offset = | |
2217 | ((loff_t)map->m_lblk) << inode->i_blkbits; | |
27d7c4ed | 2218 | do { |
4e7ea81d JK |
2219 | err = mpage_map_one_extent(handle, mpd); |
2220 | if (err < 0) { | |
2221 | struct super_block *sb = inode->i_sb; | |
2222 | ||
cb530541 TT |
2223 | if (EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED) |
2224 | goto invalidate_dirty_pages; | |
4e7ea81d | 2225 | /* |
cb530541 TT |
2226 | * Let the uper layers retry transient errors. |
2227 | * In the case of ENOSPC, if ext4_count_free_blocks() | |
2228 | * is non-zero, a commit should free up blocks. | |
4e7ea81d | 2229 | */ |
cb530541 TT |
2230 | if ((err == -ENOMEM) || |
2231 | (err == -ENOSPC && ext4_count_free_clusters(sb))) | |
2232 | return err; | |
2233 | ext4_msg(sb, KERN_CRIT, | |
2234 | "Delayed block allocation failed for " | |
2235 | "inode %lu at logical offset %llu with" | |
2236 | " max blocks %u with error %d", | |
2237 | inode->i_ino, | |
2238 | (unsigned long long)map->m_lblk, | |
2239 | (unsigned)map->m_len, -err); | |
2240 | ext4_msg(sb, KERN_CRIT, | |
2241 | "This should not happen!! Data will " | |
2242 | "be lost\n"); | |
2243 | if (err == -ENOSPC) | |
2244 | ext4_print_free_blocks(inode); | |
2245 | invalidate_dirty_pages: | |
2246 | *give_up_on_write = true; | |
4e7ea81d JK |
2247 | return err; |
2248 | } | |
2249 | /* | |
2250 | * Update buffer state, submit mapped pages, and get us new | |
2251 | * extent to map | |
2252 | */ | |
2253 | err = mpage_map_and_submit_buffers(mpd); | |
2254 | if (err < 0) | |
2255 | return err; | |
27d7c4ed | 2256 | } while (map->m_len); |
4e7ea81d | 2257 | |
622cad13 TT |
2258 | /* |
2259 | * Update on-disk size after IO is submitted. Races with | |
2260 | * truncate are avoided by checking i_size under i_data_sem. | |
2261 | */ | |
4e7ea81d | 2262 | disksize = ((loff_t)mpd->first_page) << PAGE_CACHE_SHIFT; |
4e7ea81d JK |
2263 | if (disksize > EXT4_I(inode)->i_disksize) { |
2264 | int err2; | |
622cad13 TT |
2265 | loff_t i_size; |
2266 | ||
2267 | down_write(&EXT4_I(inode)->i_data_sem); | |
2268 | i_size = i_size_read(inode); | |
2269 | if (disksize > i_size) | |
2270 | disksize = i_size; | |
2271 | if (disksize > EXT4_I(inode)->i_disksize) | |
2272 | EXT4_I(inode)->i_disksize = disksize; | |
4e7ea81d | 2273 | err2 = ext4_mark_inode_dirty(handle, inode); |
622cad13 | 2274 | up_write(&EXT4_I(inode)->i_data_sem); |
4e7ea81d JK |
2275 | if (err2) |
2276 | ext4_error(inode->i_sb, | |
2277 | "Failed to mark inode %lu dirty", | |
2278 | inode->i_ino); | |
2279 | if (!err) | |
2280 | err = err2; | |
2281 | } | |
2282 | return err; | |
2283 | } | |
2284 | ||
fffb2739 JK |
2285 | /* |
2286 | * Calculate the total number of credits to reserve for one writepages | |
20970ba6 | 2287 | * iteration. This is called from ext4_writepages(). We map an extent of |
70261f56 | 2288 | * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping |
fffb2739 JK |
2289 | * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN + |
2290 | * bpp - 1 blocks in bpp different extents. | |
2291 | */ | |
525f4ed8 MC |
2292 | static int ext4_da_writepages_trans_blocks(struct inode *inode) |
2293 | { | |
fffb2739 | 2294 | int bpp = ext4_journal_blocks_per_page(inode); |
525f4ed8 | 2295 | |
fffb2739 JK |
2296 | return ext4_meta_trans_blocks(inode, |
2297 | MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp); | |
525f4ed8 | 2298 | } |
61628a3f | 2299 | |
8e48dcfb | 2300 | /* |
4e7ea81d JK |
2301 | * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages |
2302 | * and underlying extent to map | |
2303 | * | |
2304 | * @mpd - where to look for pages | |
2305 | * | |
2306 | * Walk dirty pages in the mapping. If they are fully mapped, submit them for | |
2307 | * IO immediately. When we find a page which isn't mapped we start accumulating | |
2308 | * extent of buffers underlying these pages that needs mapping (formed by | |
2309 | * either delayed or unwritten buffers). We also lock the pages containing | |
2310 | * these buffers. The extent found is returned in @mpd structure (starting at | |
2311 | * mpd->lblk with length mpd->len blocks). | |
2312 | * | |
2313 | * Note that this function can attach bios to one io_end structure which are | |
2314 | * neither logically nor physically contiguous. Although it may seem as an | |
2315 | * unnecessary complication, it is actually inevitable in blocksize < pagesize | |
2316 | * case as we need to track IO to all buffers underlying a page in one io_end. | |
8e48dcfb | 2317 | */ |
4e7ea81d | 2318 | static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd) |
8e48dcfb | 2319 | { |
4e7ea81d JK |
2320 | struct address_space *mapping = mpd->inode->i_mapping; |
2321 | struct pagevec pvec; | |
2322 | unsigned int nr_pages; | |
aeac589a | 2323 | long left = mpd->wbc->nr_to_write; |
4e7ea81d JK |
2324 | pgoff_t index = mpd->first_page; |
2325 | pgoff_t end = mpd->last_page; | |
2326 | int tag; | |
2327 | int i, err = 0; | |
2328 | int blkbits = mpd->inode->i_blkbits; | |
2329 | ext4_lblk_t lblk; | |
2330 | struct buffer_head *head; | |
8e48dcfb | 2331 | |
4e7ea81d | 2332 | if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages) |
5b41d924 ES |
2333 | tag = PAGECACHE_TAG_TOWRITE; |
2334 | else | |
2335 | tag = PAGECACHE_TAG_DIRTY; | |
2336 | ||
4e7ea81d JK |
2337 | pagevec_init(&pvec, 0); |
2338 | mpd->map.m_len = 0; | |
2339 | mpd->next_page = index; | |
4f01b02c | 2340 | while (index <= end) { |
5b41d924 | 2341 | nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag, |
8e48dcfb TT |
2342 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1); |
2343 | if (nr_pages == 0) | |
4e7ea81d | 2344 | goto out; |
8e48dcfb TT |
2345 | |
2346 | for (i = 0; i < nr_pages; i++) { | |
2347 | struct page *page = pvec.pages[i]; | |
2348 | ||
2349 | /* | |
2350 | * At this point, the page may be truncated or | |
2351 | * invalidated (changing page->mapping to NULL), or | |
2352 | * even swizzled back from swapper_space to tmpfs file | |
2353 | * mapping. However, page->index will not change | |
2354 | * because we have a reference on the page. | |
2355 | */ | |
4f01b02c TT |
2356 | if (page->index > end) |
2357 | goto out; | |
8e48dcfb | 2358 | |
aeac589a ML |
2359 | /* |
2360 | * Accumulated enough dirty pages? This doesn't apply | |
2361 | * to WB_SYNC_ALL mode. For integrity sync we have to | |
2362 | * keep going because someone may be concurrently | |
2363 | * dirtying pages, and we might have synced a lot of | |
2364 | * newly appeared dirty pages, but have not synced all | |
2365 | * of the old dirty pages. | |
2366 | */ | |
2367 | if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0) | |
2368 | goto out; | |
2369 | ||
4e7ea81d JK |
2370 | /* If we can't merge this page, we are done. */ |
2371 | if (mpd->map.m_len > 0 && mpd->next_page != page->index) | |
2372 | goto out; | |
78aaced3 | 2373 | |
8e48dcfb | 2374 | lock_page(page); |
8e48dcfb | 2375 | /* |
4e7ea81d JK |
2376 | * If the page is no longer dirty, or its mapping no |
2377 | * longer corresponds to inode we are writing (which | |
2378 | * means it has been truncated or invalidated), or the | |
2379 | * page is already under writeback and we are not doing | |
2380 | * a data integrity writeback, skip the page | |
8e48dcfb | 2381 | */ |
4f01b02c TT |
2382 | if (!PageDirty(page) || |
2383 | (PageWriteback(page) && | |
4e7ea81d | 2384 | (mpd->wbc->sync_mode == WB_SYNC_NONE)) || |
4f01b02c | 2385 | unlikely(page->mapping != mapping)) { |
8e48dcfb TT |
2386 | unlock_page(page); |
2387 | continue; | |
2388 | } | |
2389 | ||
7cb1a535 | 2390 | wait_on_page_writeback(page); |
8e48dcfb | 2391 | BUG_ON(PageWriteback(page)); |
8e48dcfb | 2392 | |
4e7ea81d | 2393 | if (mpd->map.m_len == 0) |
8eb9e5ce | 2394 | mpd->first_page = page->index; |
8eb9e5ce | 2395 | mpd->next_page = page->index + 1; |
f8bec370 | 2396 | /* Add all dirty buffers to mpd */ |
4e7ea81d JK |
2397 | lblk = ((ext4_lblk_t)page->index) << |
2398 | (PAGE_CACHE_SHIFT - blkbits); | |
f8bec370 | 2399 | head = page_buffers(page); |
5f1132b2 JK |
2400 | err = mpage_process_page_bufs(mpd, head, head, lblk); |
2401 | if (err <= 0) | |
4e7ea81d | 2402 | goto out; |
5f1132b2 | 2403 | err = 0; |
aeac589a | 2404 | left--; |
8e48dcfb TT |
2405 | } |
2406 | pagevec_release(&pvec); | |
2407 | cond_resched(); | |
2408 | } | |
4f01b02c | 2409 | return 0; |
8eb9e5ce TT |
2410 | out: |
2411 | pagevec_release(&pvec); | |
4e7ea81d | 2412 | return err; |
8e48dcfb TT |
2413 | } |
2414 | ||
20970ba6 TT |
2415 | static int __writepage(struct page *page, struct writeback_control *wbc, |
2416 | void *data) | |
2417 | { | |
2418 | struct address_space *mapping = data; | |
2419 | int ret = ext4_writepage(page, wbc); | |
2420 | mapping_set_error(mapping, ret); | |
2421 | return ret; | |
2422 | } | |
2423 | ||
2424 | static int ext4_writepages(struct address_space *mapping, | |
2425 | struct writeback_control *wbc) | |
64769240 | 2426 | { |
4e7ea81d JK |
2427 | pgoff_t writeback_index = 0; |
2428 | long nr_to_write = wbc->nr_to_write; | |
22208ded | 2429 | int range_whole = 0; |
4e7ea81d | 2430 | int cycled = 1; |
61628a3f | 2431 | handle_t *handle = NULL; |
df22291f | 2432 | struct mpage_da_data mpd; |
5e745b04 | 2433 | struct inode *inode = mapping->host; |
6b523df4 | 2434 | int needed_blocks, rsv_blocks = 0, ret = 0; |
5e745b04 | 2435 | struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb); |
4e7ea81d | 2436 | bool done; |
1bce63d1 | 2437 | struct blk_plug plug; |
cb530541 | 2438 | bool give_up_on_write = false; |
61628a3f | 2439 | |
20970ba6 | 2440 | trace_ext4_writepages(inode, wbc); |
ba80b101 | 2441 | |
61628a3f MC |
2442 | /* |
2443 | * No pages to write? This is mainly a kludge to avoid starting | |
2444 | * a transaction for special inodes like journal inode on last iput() | |
2445 | * because that could violate lock ordering on umount | |
2446 | */ | |
a1d6cc56 | 2447 | if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) |
bbf023c7 | 2448 | goto out_writepages; |
2a21e37e | 2449 | |
20970ba6 TT |
2450 | if (ext4_should_journal_data(inode)) { |
2451 | struct blk_plug plug; | |
20970ba6 TT |
2452 | |
2453 | blk_start_plug(&plug); | |
2454 | ret = write_cache_pages(mapping, wbc, __writepage, mapping); | |
2455 | blk_finish_plug(&plug); | |
bbf023c7 | 2456 | goto out_writepages; |
20970ba6 TT |
2457 | } |
2458 | ||
2a21e37e TT |
2459 | /* |
2460 | * If the filesystem has aborted, it is read-only, so return | |
2461 | * right away instead of dumping stack traces later on that | |
2462 | * will obscure the real source of the problem. We test | |
4ab2f15b | 2463 | * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because |
2a21e37e | 2464 | * the latter could be true if the filesystem is mounted |
20970ba6 | 2465 | * read-only, and in that case, ext4_writepages should |
2a21e37e TT |
2466 | * *never* be called, so if that ever happens, we would want |
2467 | * the stack trace. | |
2468 | */ | |
bbf023c7 ML |
2469 | if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) { |
2470 | ret = -EROFS; | |
2471 | goto out_writepages; | |
2472 | } | |
2a21e37e | 2473 | |
6b523df4 JK |
2474 | if (ext4_should_dioread_nolock(inode)) { |
2475 | /* | |
70261f56 | 2476 | * We may need to convert up to one extent per block in |
6b523df4 JK |
2477 | * the page and we may dirty the inode. |
2478 | */ | |
2479 | rsv_blocks = 1 + (PAGE_CACHE_SIZE >> inode->i_blkbits); | |
2480 | } | |
2481 | ||
4e7ea81d JK |
2482 | /* |
2483 | * If we have inline data and arrive here, it means that | |
2484 | * we will soon create the block for the 1st page, so | |
2485 | * we'd better clear the inline data here. | |
2486 | */ | |
2487 | if (ext4_has_inline_data(inode)) { | |
2488 | /* Just inode will be modified... */ | |
2489 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 1); | |
2490 | if (IS_ERR(handle)) { | |
2491 | ret = PTR_ERR(handle); | |
2492 | goto out_writepages; | |
2493 | } | |
2494 | BUG_ON(ext4_test_inode_state(inode, | |
2495 | EXT4_STATE_MAY_INLINE_DATA)); | |
2496 | ext4_destroy_inline_data(handle, inode); | |
2497 | ext4_journal_stop(handle); | |
2498 | } | |
2499 | ||
22208ded AK |
2500 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
2501 | range_whole = 1; | |
61628a3f | 2502 | |
2acf2c26 | 2503 | if (wbc->range_cyclic) { |
4e7ea81d JK |
2504 | writeback_index = mapping->writeback_index; |
2505 | if (writeback_index) | |
2acf2c26 | 2506 | cycled = 0; |
4e7ea81d JK |
2507 | mpd.first_page = writeback_index; |
2508 | mpd.last_page = -1; | |
5b41d924 | 2509 | } else { |
4e7ea81d JK |
2510 | mpd.first_page = wbc->range_start >> PAGE_CACHE_SHIFT; |
2511 | mpd.last_page = wbc->range_end >> PAGE_CACHE_SHIFT; | |
5b41d924 | 2512 | } |
a1d6cc56 | 2513 | |
4e7ea81d JK |
2514 | mpd.inode = inode; |
2515 | mpd.wbc = wbc; | |
2516 | ext4_io_submit_init(&mpd.io_submit, wbc); | |
2acf2c26 | 2517 | retry: |
6e6938b6 | 2518 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
4e7ea81d JK |
2519 | tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page); |
2520 | done = false; | |
1bce63d1 | 2521 | blk_start_plug(&plug); |
4e7ea81d JK |
2522 | while (!done && mpd.first_page <= mpd.last_page) { |
2523 | /* For each extent of pages we use new io_end */ | |
2524 | mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL); | |
2525 | if (!mpd.io_submit.io_end) { | |
2526 | ret = -ENOMEM; | |
2527 | break; | |
2528 | } | |
a1d6cc56 AK |
2529 | |
2530 | /* | |
4e7ea81d JK |
2531 | * We have two constraints: We find one extent to map and we |
2532 | * must always write out whole page (makes a difference when | |
2533 | * blocksize < pagesize) so that we don't block on IO when we | |
2534 | * try to write out the rest of the page. Journalled mode is | |
2535 | * not supported by delalloc. | |
a1d6cc56 AK |
2536 | */ |
2537 | BUG_ON(ext4_should_journal_data(inode)); | |
525f4ed8 | 2538 | needed_blocks = ext4_da_writepages_trans_blocks(inode); |
a1d6cc56 | 2539 | |
4e7ea81d | 2540 | /* start a new transaction */ |
6b523df4 JK |
2541 | handle = ext4_journal_start_with_reserve(inode, |
2542 | EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks); | |
61628a3f MC |
2543 | if (IS_ERR(handle)) { |
2544 | ret = PTR_ERR(handle); | |
1693918e | 2545 | ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: " |
fbe845dd | 2546 | "%ld pages, ino %lu; err %d", __func__, |
a1d6cc56 | 2547 | wbc->nr_to_write, inode->i_ino, ret); |
4e7ea81d JK |
2548 | /* Release allocated io_end */ |
2549 | ext4_put_io_end(mpd.io_submit.io_end); | |
2550 | break; | |
61628a3f | 2551 | } |
f63e6005 | 2552 | |
4e7ea81d JK |
2553 | trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc); |
2554 | ret = mpage_prepare_extent_to_map(&mpd); | |
2555 | if (!ret) { | |
2556 | if (mpd.map.m_len) | |
cb530541 TT |
2557 | ret = mpage_map_and_submit_extent(handle, &mpd, |
2558 | &give_up_on_write); | |
4e7ea81d JK |
2559 | else { |
2560 | /* | |
2561 | * We scanned the whole range (or exhausted | |
2562 | * nr_to_write), submitted what was mapped and | |
2563 | * didn't find anything needing mapping. We are | |
2564 | * done. | |
2565 | */ | |
2566 | done = true; | |
2567 | } | |
f63e6005 | 2568 | } |
61628a3f | 2569 | ext4_journal_stop(handle); |
4e7ea81d JK |
2570 | /* Submit prepared bio */ |
2571 | ext4_io_submit(&mpd.io_submit); | |
2572 | /* Unlock pages we didn't use */ | |
cb530541 | 2573 | mpage_release_unused_pages(&mpd, give_up_on_write); |
4e7ea81d JK |
2574 | /* Drop our io_end reference we got from init */ |
2575 | ext4_put_io_end(mpd.io_submit.io_end); | |
2576 | ||
2577 | if (ret == -ENOSPC && sbi->s_journal) { | |
2578 | /* | |
2579 | * Commit the transaction which would | |
22208ded AK |
2580 | * free blocks released in the transaction |
2581 | * and try again | |
2582 | */ | |
df22291f | 2583 | jbd2_journal_force_commit_nested(sbi->s_journal); |
22208ded | 2584 | ret = 0; |
4e7ea81d JK |
2585 | continue; |
2586 | } | |
2587 | /* Fatal error - ENOMEM, EIO... */ | |
2588 | if (ret) | |
61628a3f | 2589 | break; |
a1d6cc56 | 2590 | } |
1bce63d1 | 2591 | blk_finish_plug(&plug); |
9c12a831 | 2592 | if (!ret && !cycled && wbc->nr_to_write > 0) { |
2acf2c26 | 2593 | cycled = 1; |
4e7ea81d JK |
2594 | mpd.last_page = writeback_index - 1; |
2595 | mpd.first_page = 0; | |
2acf2c26 AK |
2596 | goto retry; |
2597 | } | |
22208ded AK |
2598 | |
2599 | /* Update index */ | |
22208ded AK |
2600 | if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) |
2601 | /* | |
4e7ea81d | 2602 | * Set the writeback_index so that range_cyclic |
22208ded AK |
2603 | * mode will write it back later |
2604 | */ | |
4e7ea81d | 2605 | mapping->writeback_index = mpd.first_page; |
a1d6cc56 | 2606 | |
61628a3f | 2607 | out_writepages: |
20970ba6 TT |
2608 | trace_ext4_writepages_result(inode, wbc, ret, |
2609 | nr_to_write - wbc->nr_to_write); | |
61628a3f | 2610 | return ret; |
64769240 AT |
2611 | } |
2612 | ||
79f0be8d AK |
2613 | static int ext4_nonda_switch(struct super_block *sb) |
2614 | { | |
5c1ff336 | 2615 | s64 free_clusters, dirty_clusters; |
79f0be8d AK |
2616 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
2617 | ||
2618 | /* | |
2619 | * switch to non delalloc mode if we are running low | |
2620 | * on free block. The free block accounting via percpu | |
179f7ebf | 2621 | * counters can get slightly wrong with percpu_counter_batch getting |
79f0be8d AK |
2622 | * accumulated on each CPU without updating global counters |
2623 | * Delalloc need an accurate free block accounting. So switch | |
2624 | * to non delalloc when we are near to error range. | |
2625 | */ | |
5c1ff336 EW |
2626 | free_clusters = |
2627 | percpu_counter_read_positive(&sbi->s_freeclusters_counter); | |
2628 | dirty_clusters = | |
2629 | percpu_counter_read_positive(&sbi->s_dirtyclusters_counter); | |
00d4e736 TT |
2630 | /* |
2631 | * Start pushing delalloc when 1/2 of free blocks are dirty. | |
2632 | */ | |
5c1ff336 | 2633 | if (dirty_clusters && (free_clusters < 2 * dirty_clusters)) |
10ee27a0 | 2634 | try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE); |
00d4e736 | 2635 | |
5c1ff336 EW |
2636 | if (2 * free_clusters < 3 * dirty_clusters || |
2637 | free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) { | |
79f0be8d | 2638 | /* |
c8afb446 ES |
2639 | * free block count is less than 150% of dirty blocks |
2640 | * or free blocks is less than watermark | |
79f0be8d AK |
2641 | */ |
2642 | return 1; | |
2643 | } | |
2644 | return 0; | |
2645 | } | |
2646 | ||
64769240 | 2647 | static int ext4_da_write_begin(struct file *file, struct address_space *mapping, |
de9a55b8 TT |
2648 | loff_t pos, unsigned len, unsigned flags, |
2649 | struct page **pagep, void **fsdata) | |
64769240 | 2650 | { |
72b8ab9d | 2651 | int ret, retries = 0; |
64769240 AT |
2652 | struct page *page; |
2653 | pgoff_t index; | |
64769240 AT |
2654 | struct inode *inode = mapping->host; |
2655 | handle_t *handle; | |
2656 | ||
2657 | index = pos >> PAGE_CACHE_SHIFT; | |
79f0be8d AK |
2658 | |
2659 | if (ext4_nonda_switch(inode->i_sb)) { | |
2660 | *fsdata = (void *)FALL_BACK_TO_NONDELALLOC; | |
2661 | return ext4_write_begin(file, mapping, pos, | |
2662 | len, flags, pagep, fsdata); | |
2663 | } | |
2664 | *fsdata = (void *)0; | |
9bffad1e | 2665 | trace_ext4_da_write_begin(inode, pos, len, flags); |
9c3569b5 TM |
2666 | |
2667 | if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { | |
2668 | ret = ext4_da_write_inline_data_begin(mapping, inode, | |
2669 | pos, len, flags, | |
2670 | pagep, fsdata); | |
2671 | if (ret < 0) | |
47564bfb TT |
2672 | return ret; |
2673 | if (ret == 1) | |
2674 | return 0; | |
9c3569b5 TM |
2675 | } |
2676 | ||
47564bfb TT |
2677 | /* |
2678 | * grab_cache_page_write_begin() can take a long time if the | |
2679 | * system is thrashing due to memory pressure, or if the page | |
2680 | * is being written back. So grab it first before we start | |
2681 | * the transaction handle. This also allows us to allocate | |
2682 | * the page (if needed) without using GFP_NOFS. | |
2683 | */ | |
2684 | retry_grab: | |
2685 | page = grab_cache_page_write_begin(mapping, index, flags); | |
2686 | if (!page) | |
2687 | return -ENOMEM; | |
2688 | unlock_page(page); | |
2689 | ||
64769240 AT |
2690 | /* |
2691 | * With delayed allocation, we don't log the i_disksize update | |
2692 | * if there is delayed block allocation. But we still need | |
2693 | * to journalling the i_disksize update if writes to the end | |
2694 | * of file which has an already mapped buffer. | |
2695 | */ | |
47564bfb | 2696 | retry_journal: |
9924a92a | 2697 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, 1); |
64769240 | 2698 | if (IS_ERR(handle)) { |
47564bfb TT |
2699 | page_cache_release(page); |
2700 | return PTR_ERR(handle); | |
64769240 AT |
2701 | } |
2702 | ||
47564bfb TT |
2703 | lock_page(page); |
2704 | if (page->mapping != mapping) { | |
2705 | /* The page got truncated from under us */ | |
2706 | unlock_page(page); | |
2707 | page_cache_release(page); | |
d5a0d4f7 | 2708 | ext4_journal_stop(handle); |
47564bfb | 2709 | goto retry_grab; |
d5a0d4f7 | 2710 | } |
47564bfb | 2711 | /* In case writeback began while the page was unlocked */ |
7afe5aa5 | 2712 | wait_for_stable_page(page); |
64769240 | 2713 | |
6e1db88d | 2714 | ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep); |
64769240 AT |
2715 | if (ret < 0) { |
2716 | unlock_page(page); | |
2717 | ext4_journal_stop(handle); | |
ae4d5372 AK |
2718 | /* |
2719 | * block_write_begin may have instantiated a few blocks | |
2720 | * outside i_size. Trim these off again. Don't need | |
2721 | * i_size_read because we hold i_mutex. | |
2722 | */ | |
2723 | if (pos + len > inode->i_size) | |
b9a4207d | 2724 | ext4_truncate_failed_write(inode); |
47564bfb TT |
2725 | |
2726 | if (ret == -ENOSPC && | |
2727 | ext4_should_retry_alloc(inode->i_sb, &retries)) | |
2728 | goto retry_journal; | |
2729 | ||
2730 | page_cache_release(page); | |
2731 | return ret; | |
64769240 AT |
2732 | } |
2733 | ||
47564bfb | 2734 | *pagep = page; |
64769240 AT |
2735 | return ret; |
2736 | } | |
2737 | ||
632eaeab MC |
2738 | /* |
2739 | * Check if we should update i_disksize | |
2740 | * when write to the end of file but not require block allocation | |
2741 | */ | |
2742 | static int ext4_da_should_update_i_disksize(struct page *page, | |
de9a55b8 | 2743 | unsigned long offset) |
632eaeab MC |
2744 | { |
2745 | struct buffer_head *bh; | |
2746 | struct inode *inode = page->mapping->host; | |
2747 | unsigned int idx; | |
2748 | int i; | |
2749 | ||
2750 | bh = page_buffers(page); | |
2751 | idx = offset >> inode->i_blkbits; | |
2752 | ||
af5bc92d | 2753 | for (i = 0; i < idx; i++) |
632eaeab MC |
2754 | bh = bh->b_this_page; |
2755 | ||
29fa89d0 | 2756 | if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh)) |
632eaeab MC |
2757 | return 0; |
2758 | return 1; | |
2759 | } | |
2760 | ||
64769240 | 2761 | static int ext4_da_write_end(struct file *file, |
de9a55b8 TT |
2762 | struct address_space *mapping, |
2763 | loff_t pos, unsigned len, unsigned copied, | |
2764 | struct page *page, void *fsdata) | |
64769240 AT |
2765 | { |
2766 | struct inode *inode = mapping->host; | |
2767 | int ret = 0, ret2; | |
2768 | handle_t *handle = ext4_journal_current_handle(); | |
2769 | loff_t new_i_size; | |
632eaeab | 2770 | unsigned long start, end; |
79f0be8d AK |
2771 | int write_mode = (int)(unsigned long)fsdata; |
2772 | ||
74d553aa TT |
2773 | if (write_mode == FALL_BACK_TO_NONDELALLOC) |
2774 | return ext4_write_end(file, mapping, pos, | |
2775 | len, copied, page, fsdata); | |
632eaeab | 2776 | |
9bffad1e | 2777 | trace_ext4_da_write_end(inode, pos, len, copied); |
632eaeab | 2778 | start = pos & (PAGE_CACHE_SIZE - 1); |
af5bc92d | 2779 | end = start + copied - 1; |
64769240 AT |
2780 | |
2781 | /* | |
2782 | * generic_write_end() will run mark_inode_dirty() if i_size | |
2783 | * changes. So let's piggyback the i_disksize mark_inode_dirty | |
2784 | * into that. | |
2785 | */ | |
64769240 | 2786 | new_i_size = pos + copied; |
ea51d132 | 2787 | if (copied && new_i_size > EXT4_I(inode)->i_disksize) { |
9c3569b5 TM |
2788 | if (ext4_has_inline_data(inode) || |
2789 | ext4_da_should_update_i_disksize(page, end)) { | |
632eaeab | 2790 | down_write(&EXT4_I(inode)->i_data_sem); |
f3b59291 | 2791 | if (new_i_size > EXT4_I(inode)->i_disksize) |
632eaeab | 2792 | EXT4_I(inode)->i_disksize = new_i_size; |
632eaeab | 2793 | up_write(&EXT4_I(inode)->i_data_sem); |
cf17fea6 AK |
2794 | /* We need to mark inode dirty even if |
2795 | * new_i_size is less that inode->i_size | |
2796 | * bu greater than i_disksize.(hint delalloc) | |
2797 | */ | |
2798 | ext4_mark_inode_dirty(handle, inode); | |
64769240 | 2799 | } |
632eaeab | 2800 | } |
9c3569b5 TM |
2801 | |
2802 | if (write_mode != CONVERT_INLINE_DATA && | |
2803 | ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) && | |
2804 | ext4_has_inline_data(inode)) | |
2805 | ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied, | |
2806 | page); | |
2807 | else | |
2808 | ret2 = generic_write_end(file, mapping, pos, len, copied, | |
64769240 | 2809 | page, fsdata); |
9c3569b5 | 2810 | |
64769240 AT |
2811 | copied = ret2; |
2812 | if (ret2 < 0) | |
2813 | ret = ret2; | |
2814 | ret2 = ext4_journal_stop(handle); | |
2815 | if (!ret) | |
2816 | ret = ret2; | |
2817 | ||
2818 | return ret ? ret : copied; | |
2819 | } | |
2820 | ||
d47992f8 LC |
2821 | static void ext4_da_invalidatepage(struct page *page, unsigned int offset, |
2822 | unsigned int length) | |
64769240 | 2823 | { |
64769240 AT |
2824 | /* |
2825 | * Drop reserved blocks | |
2826 | */ | |
2827 | BUG_ON(!PageLocked(page)); | |
2828 | if (!page_has_buffers(page)) | |
2829 | goto out; | |
2830 | ||
ca99fdd2 | 2831 | ext4_da_page_release_reservation(page, offset, length); |
64769240 AT |
2832 | |
2833 | out: | |
d47992f8 | 2834 | ext4_invalidatepage(page, offset, length); |
64769240 AT |
2835 | |
2836 | return; | |
2837 | } | |
2838 | ||
ccd2506b TT |
2839 | /* |
2840 | * Force all delayed allocation blocks to be allocated for a given inode. | |
2841 | */ | |
2842 | int ext4_alloc_da_blocks(struct inode *inode) | |
2843 | { | |
fb40ba0d TT |
2844 | trace_ext4_alloc_da_blocks(inode); |
2845 | ||
ccd2506b TT |
2846 | if (!EXT4_I(inode)->i_reserved_data_blocks && |
2847 | !EXT4_I(inode)->i_reserved_meta_blocks) | |
2848 | return 0; | |
2849 | ||
2850 | /* | |
2851 | * We do something simple for now. The filemap_flush() will | |
2852 | * also start triggering a write of the data blocks, which is | |
2853 | * not strictly speaking necessary (and for users of | |
2854 | * laptop_mode, not even desirable). However, to do otherwise | |
2855 | * would require replicating code paths in: | |
de9a55b8 | 2856 | * |
20970ba6 | 2857 | * ext4_writepages() -> |
ccd2506b TT |
2858 | * write_cache_pages() ---> (via passed in callback function) |
2859 | * __mpage_da_writepage() --> | |
2860 | * mpage_add_bh_to_extent() | |
2861 | * mpage_da_map_blocks() | |
2862 | * | |
2863 | * The problem is that write_cache_pages(), located in | |
2864 | * mm/page-writeback.c, marks pages clean in preparation for | |
2865 | * doing I/O, which is not desirable if we're not planning on | |
2866 | * doing I/O at all. | |
2867 | * | |
2868 | * We could call write_cache_pages(), and then redirty all of | |
380cf090 | 2869 | * the pages by calling redirty_page_for_writepage() but that |
ccd2506b TT |
2870 | * would be ugly in the extreme. So instead we would need to |
2871 | * replicate parts of the code in the above functions, | |
25985edc | 2872 | * simplifying them because we wouldn't actually intend to |
ccd2506b TT |
2873 | * write out the pages, but rather only collect contiguous |
2874 | * logical block extents, call the multi-block allocator, and | |
2875 | * then update the buffer heads with the block allocations. | |
de9a55b8 | 2876 | * |
ccd2506b TT |
2877 | * For now, though, we'll cheat by calling filemap_flush(), |
2878 | * which will map the blocks, and start the I/O, but not | |
2879 | * actually wait for the I/O to complete. | |
2880 | */ | |
2881 | return filemap_flush(inode->i_mapping); | |
2882 | } | |
64769240 | 2883 | |
ac27a0ec DK |
2884 | /* |
2885 | * bmap() is special. It gets used by applications such as lilo and by | |
2886 | * the swapper to find the on-disk block of a specific piece of data. | |
2887 | * | |
2888 | * Naturally, this is dangerous if the block concerned is still in the | |
617ba13b | 2889 | * journal. If somebody makes a swapfile on an ext4 data-journaling |
ac27a0ec DK |
2890 | * filesystem and enables swap, then they may get a nasty shock when the |
2891 | * data getting swapped to that swapfile suddenly gets overwritten by | |
2892 | * the original zero's written out previously to the journal and | |
2893 | * awaiting writeback in the kernel's buffer cache. | |
2894 | * | |
2895 | * So, if we see any bmap calls here on a modified, data-journaled file, | |
2896 | * take extra steps to flush any blocks which might be in the cache. | |
2897 | */ | |
617ba13b | 2898 | static sector_t ext4_bmap(struct address_space *mapping, sector_t block) |
ac27a0ec DK |
2899 | { |
2900 | struct inode *inode = mapping->host; | |
2901 | journal_t *journal; | |
2902 | int err; | |
2903 | ||
46c7f254 TM |
2904 | /* |
2905 | * We can get here for an inline file via the FIBMAP ioctl | |
2906 | */ | |
2907 | if (ext4_has_inline_data(inode)) | |
2908 | return 0; | |
2909 | ||
64769240 AT |
2910 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) && |
2911 | test_opt(inode->i_sb, DELALLOC)) { | |
2912 | /* | |
2913 | * With delalloc we want to sync the file | |
2914 | * so that we can make sure we allocate | |
2915 | * blocks for file | |
2916 | */ | |
2917 | filemap_write_and_wait(mapping); | |
2918 | } | |
2919 | ||
19f5fb7a TT |
2920 | if (EXT4_JOURNAL(inode) && |
2921 | ext4_test_inode_state(inode, EXT4_STATE_JDATA)) { | |
ac27a0ec DK |
2922 | /* |
2923 | * This is a REALLY heavyweight approach, but the use of | |
2924 | * bmap on dirty files is expected to be extremely rare: | |
2925 | * only if we run lilo or swapon on a freshly made file | |
2926 | * do we expect this to happen. | |
2927 | * | |
2928 | * (bmap requires CAP_SYS_RAWIO so this does not | |
2929 | * represent an unprivileged user DOS attack --- we'd be | |
2930 | * in trouble if mortal users could trigger this path at | |
2931 | * will.) | |
2932 | * | |
617ba13b | 2933 | * NB. EXT4_STATE_JDATA is not set on files other than |
ac27a0ec DK |
2934 | * regular files. If somebody wants to bmap a directory |
2935 | * or symlink and gets confused because the buffer | |
2936 | * hasn't yet been flushed to disk, they deserve | |
2937 | * everything they get. | |
2938 | */ | |
2939 | ||
19f5fb7a | 2940 | ext4_clear_inode_state(inode, EXT4_STATE_JDATA); |
617ba13b | 2941 | journal = EXT4_JOURNAL(inode); |
dab291af MC |
2942 | jbd2_journal_lock_updates(journal); |
2943 | err = jbd2_journal_flush(journal); | |
2944 | jbd2_journal_unlock_updates(journal); | |
ac27a0ec DK |
2945 | |
2946 | if (err) | |
2947 | return 0; | |
2948 | } | |
2949 | ||
af5bc92d | 2950 | return generic_block_bmap(mapping, block, ext4_get_block); |
ac27a0ec DK |
2951 | } |
2952 | ||
617ba13b | 2953 | static int ext4_readpage(struct file *file, struct page *page) |
ac27a0ec | 2954 | { |
46c7f254 TM |
2955 | int ret = -EAGAIN; |
2956 | struct inode *inode = page->mapping->host; | |
2957 | ||
0562e0ba | 2958 | trace_ext4_readpage(page); |
46c7f254 TM |
2959 | |
2960 | if (ext4_has_inline_data(inode)) | |
2961 | ret = ext4_readpage_inline(inode, page); | |
2962 | ||
2963 | if (ret == -EAGAIN) | |
2964 | return mpage_readpage(page, ext4_get_block); | |
2965 | ||
2966 | return ret; | |
ac27a0ec DK |
2967 | } |
2968 | ||
2969 | static int | |
617ba13b | 2970 | ext4_readpages(struct file *file, struct address_space *mapping, |
ac27a0ec DK |
2971 | struct list_head *pages, unsigned nr_pages) |
2972 | { | |
46c7f254 TM |
2973 | struct inode *inode = mapping->host; |
2974 | ||
2975 | /* If the file has inline data, no need to do readpages. */ | |
2976 | if (ext4_has_inline_data(inode)) | |
2977 | return 0; | |
2978 | ||
617ba13b | 2979 | return mpage_readpages(mapping, pages, nr_pages, ext4_get_block); |
ac27a0ec DK |
2980 | } |
2981 | ||
d47992f8 LC |
2982 | static void ext4_invalidatepage(struct page *page, unsigned int offset, |
2983 | unsigned int length) | |
ac27a0ec | 2984 | { |
ca99fdd2 | 2985 | trace_ext4_invalidatepage(page, offset, length); |
0562e0ba | 2986 | |
4520fb3c JK |
2987 | /* No journalling happens on data buffers when this function is used */ |
2988 | WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page))); | |
2989 | ||
ca99fdd2 | 2990 | block_invalidatepage(page, offset, length); |
4520fb3c JK |
2991 | } |
2992 | ||
53e87268 | 2993 | static int __ext4_journalled_invalidatepage(struct page *page, |
ca99fdd2 LC |
2994 | unsigned int offset, |
2995 | unsigned int length) | |
4520fb3c JK |
2996 | { |
2997 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); | |
2998 | ||
ca99fdd2 | 2999 | trace_ext4_journalled_invalidatepage(page, offset, length); |
4520fb3c | 3000 | |
ac27a0ec DK |
3001 | /* |
3002 | * If it's a full truncate we just forget about the pending dirtying | |
3003 | */ | |
ca99fdd2 | 3004 | if (offset == 0 && length == PAGE_CACHE_SIZE) |
ac27a0ec DK |
3005 | ClearPageChecked(page); |
3006 | ||
ca99fdd2 | 3007 | return jbd2_journal_invalidatepage(journal, page, offset, length); |
53e87268 JK |
3008 | } |
3009 | ||
3010 | /* Wrapper for aops... */ | |
3011 | static void ext4_journalled_invalidatepage(struct page *page, | |
d47992f8 LC |
3012 | unsigned int offset, |
3013 | unsigned int length) | |
53e87268 | 3014 | { |
ca99fdd2 | 3015 | WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0); |
ac27a0ec DK |
3016 | } |
3017 | ||
617ba13b | 3018 | static int ext4_releasepage(struct page *page, gfp_t wait) |
ac27a0ec | 3019 | { |
617ba13b | 3020 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); |
ac27a0ec | 3021 | |
0562e0ba JZ |
3022 | trace_ext4_releasepage(page); |
3023 | ||
e1c36595 JK |
3024 | /* Page has dirty journalled data -> cannot release */ |
3025 | if (PageChecked(page)) | |
ac27a0ec | 3026 | return 0; |
0390131b FM |
3027 | if (journal) |
3028 | return jbd2_journal_try_to_free_buffers(journal, page, wait); | |
3029 | else | |
3030 | return try_to_free_buffers(page); | |
ac27a0ec DK |
3031 | } |
3032 | ||
2ed88685 TT |
3033 | /* |
3034 | * ext4_get_block used when preparing for a DIO write or buffer write. | |
3035 | * We allocate an uinitialized extent if blocks haven't been allocated. | |
3036 | * The extent will be converted to initialized after the IO is complete. | |
3037 | */ | |
f19d5870 | 3038 | int ext4_get_block_write(struct inode *inode, sector_t iblock, |
4c0425ff MC |
3039 | struct buffer_head *bh_result, int create) |
3040 | { | |
c7064ef1 | 3041 | ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n", |
8d5d02e6 | 3042 | inode->i_ino, create); |
2ed88685 TT |
3043 | return _ext4_get_block(inode, iblock, bh_result, |
3044 | EXT4_GET_BLOCKS_IO_CREATE_EXT); | |
4c0425ff MC |
3045 | } |
3046 | ||
729f52c6 | 3047 | static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock, |
8b0f165f | 3048 | struct buffer_head *bh_result, int create) |
729f52c6 | 3049 | { |
8b0f165f AP |
3050 | ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n", |
3051 | inode->i_ino, create); | |
3052 | return _ext4_get_block(inode, iblock, bh_result, | |
3053 | EXT4_GET_BLOCKS_NO_LOCK); | |
729f52c6 ZL |
3054 | } |
3055 | ||
4c0425ff | 3056 | static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset, |
7b7a8665 | 3057 | ssize_t size, void *private) |
4c0425ff MC |
3058 | { |
3059 | ext4_io_end_t *io_end = iocb->private; | |
4c0425ff | 3060 | |
97a851ed | 3061 | /* if not async direct IO just return */ |
7b7a8665 | 3062 | if (!io_end) |
97a851ed | 3063 | return; |
4b70df18 | 3064 | |
88635ca2 | 3065 | ext_debug("ext4_end_io_dio(): io_end 0x%p " |
ace36ad4 | 3066 | "for inode %lu, iocb 0x%p, offset %llu, size %zd\n", |
8d5d02e6 MC |
3067 | iocb->private, io_end->inode->i_ino, iocb, offset, |
3068 | size); | |
8d5d02e6 | 3069 | |
b5a7e970 | 3070 | iocb->private = NULL; |
4c0425ff MC |
3071 | io_end->offset = offset; |
3072 | io_end->size = size; | |
7b7a8665 | 3073 | ext4_put_io_end(io_end); |
4c0425ff | 3074 | } |
c7064ef1 | 3075 | |
4c0425ff MC |
3076 | /* |
3077 | * For ext4 extent files, ext4 will do direct-io write to holes, | |
3078 | * preallocated extents, and those write extend the file, no need to | |
3079 | * fall back to buffered IO. | |
3080 | * | |
556615dc | 3081 | * For holes, we fallocate those blocks, mark them as unwritten |
69c499d1 | 3082 | * If those blocks were preallocated, we mark sure they are split, but |
556615dc | 3083 | * still keep the range to write as unwritten. |
4c0425ff | 3084 | * |
69c499d1 | 3085 | * The unwritten extents will be converted to written when DIO is completed. |
8d5d02e6 | 3086 | * For async direct IO, since the IO may still pending when return, we |
25985edc | 3087 | * set up an end_io call back function, which will do the conversion |
8d5d02e6 | 3088 | * when async direct IO completed. |
4c0425ff MC |
3089 | * |
3090 | * If the O_DIRECT write will extend the file then add this inode to the | |
3091 | * orphan list. So recovery will truncate it back to the original size | |
3092 | * if the machine crashes during the write. | |
3093 | * | |
3094 | */ | |
3095 | static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb, | |
16b1f05d | 3096 | struct iov_iter *iter, loff_t offset) |
4c0425ff MC |
3097 | { |
3098 | struct file *file = iocb->ki_filp; | |
3099 | struct inode *inode = file->f_mapping->host; | |
3100 | ssize_t ret; | |
a6cbcd4a | 3101 | size_t count = iov_iter_count(iter); |
69c499d1 TT |
3102 | int overwrite = 0; |
3103 | get_block_t *get_block_func = NULL; | |
3104 | int dio_flags = 0; | |
4c0425ff | 3105 | loff_t final_size = offset + count; |
97a851ed | 3106 | ext4_io_end_t *io_end = NULL; |
729f52c6 | 3107 | |
69c499d1 TT |
3108 | /* Use the old path for reads and writes beyond i_size. */ |
3109 | if (rw != WRITE || final_size > inode->i_size) | |
16b1f05d | 3110 | return ext4_ind_direct_IO(rw, iocb, iter, offset); |
4bd809db | 3111 | |
69c499d1 | 3112 | BUG_ON(iocb->private == NULL); |
4bd809db | 3113 | |
e8340395 JK |
3114 | /* |
3115 | * Make all waiters for direct IO properly wait also for extent | |
3116 | * conversion. This also disallows race between truncate() and | |
3117 | * overwrite DIO as i_dio_count needs to be incremented under i_mutex. | |
3118 | */ | |
3119 | if (rw == WRITE) | |
3120 | atomic_inc(&inode->i_dio_count); | |
3121 | ||
69c499d1 TT |
3122 | /* If we do a overwrite dio, i_mutex locking can be released */ |
3123 | overwrite = *((int *)iocb->private); | |
4bd809db | 3124 | |
69c499d1 | 3125 | if (overwrite) { |
69c499d1 TT |
3126 | down_read(&EXT4_I(inode)->i_data_sem); |
3127 | mutex_unlock(&inode->i_mutex); | |
3128 | } | |
8d5d02e6 | 3129 | |
69c499d1 TT |
3130 | /* |
3131 | * We could direct write to holes and fallocate. | |
3132 | * | |
3133 | * Allocated blocks to fill the hole are marked as | |
556615dc | 3134 | * unwritten to prevent parallel buffered read to expose |
69c499d1 TT |
3135 | * the stale data before DIO complete the data IO. |
3136 | * | |
3137 | * As to previously fallocated extents, ext4 get_block will | |
3138 | * just simply mark the buffer mapped but still keep the | |
556615dc | 3139 | * extents unwritten. |
69c499d1 TT |
3140 | * |
3141 | * For non AIO case, we will convert those unwritten extents | |
3142 | * to written after return back from blockdev_direct_IO. | |
3143 | * | |
3144 | * For async DIO, the conversion needs to be deferred when the | |
3145 | * IO is completed. The ext4 end_io callback function will be | |
3146 | * called to take care of the conversion work. Here for async | |
3147 | * case, we allocate an io_end structure to hook to the iocb. | |
3148 | */ | |
3149 | iocb->private = NULL; | |
3150 | ext4_inode_aio_set(inode, NULL); | |
3151 | if (!is_sync_kiocb(iocb)) { | |
97a851ed | 3152 | io_end = ext4_init_io_end(inode, GFP_NOFS); |
69c499d1 TT |
3153 | if (!io_end) { |
3154 | ret = -ENOMEM; | |
3155 | goto retake_lock; | |
8b0f165f | 3156 | } |
97a851ed JK |
3157 | /* |
3158 | * Grab reference for DIO. Will be dropped in ext4_end_io_dio() | |
3159 | */ | |
3160 | iocb->private = ext4_get_io_end(io_end); | |
8d5d02e6 | 3161 | /* |
69c499d1 TT |
3162 | * we save the io structure for current async direct |
3163 | * IO, so that later ext4_map_blocks() could flag the | |
3164 | * io structure whether there is a unwritten extents | |
3165 | * needs to be converted when IO is completed. | |
8d5d02e6 | 3166 | */ |
69c499d1 TT |
3167 | ext4_inode_aio_set(inode, io_end); |
3168 | } | |
4bd809db | 3169 | |
69c499d1 TT |
3170 | if (overwrite) { |
3171 | get_block_func = ext4_get_block_write_nolock; | |
3172 | } else { | |
3173 | get_block_func = ext4_get_block_write; | |
3174 | dio_flags = DIO_LOCKING; | |
3175 | } | |
3176 | ret = __blockdev_direct_IO(rw, iocb, inode, | |
31b14039 AV |
3177 | inode->i_sb->s_bdev, iter, |
3178 | offset, | |
69c499d1 TT |
3179 | get_block_func, |
3180 | ext4_end_io_dio, | |
3181 | NULL, | |
3182 | dio_flags); | |
3183 | ||
69c499d1 | 3184 | /* |
97a851ed JK |
3185 | * Put our reference to io_end. This can free the io_end structure e.g. |
3186 | * in sync IO case or in case of error. It can even perform extent | |
3187 | * conversion if all bios we submitted finished before we got here. | |
3188 | * Note that in that case iocb->private can be already set to NULL | |
3189 | * here. | |
69c499d1 | 3190 | */ |
97a851ed JK |
3191 | if (io_end) { |
3192 | ext4_inode_aio_set(inode, NULL); | |
3193 | ext4_put_io_end(io_end); | |
3194 | /* | |
3195 | * When no IO was submitted ext4_end_io_dio() was not | |
3196 | * called so we have to put iocb's reference. | |
3197 | */ | |
3198 | if (ret <= 0 && ret != -EIOCBQUEUED && iocb->private) { | |
3199 | WARN_ON(iocb->private != io_end); | |
3200 | WARN_ON(io_end->flag & EXT4_IO_END_UNWRITTEN); | |
97a851ed JK |
3201 | ext4_put_io_end(io_end); |
3202 | iocb->private = NULL; | |
3203 | } | |
3204 | } | |
3205 | if (ret > 0 && !overwrite && ext4_test_inode_state(inode, | |
69c499d1 TT |
3206 | EXT4_STATE_DIO_UNWRITTEN)) { |
3207 | int err; | |
3208 | /* | |
3209 | * for non AIO case, since the IO is already | |
3210 | * completed, we could do the conversion right here | |
3211 | */ | |
6b523df4 | 3212 | err = ext4_convert_unwritten_extents(NULL, inode, |
69c499d1 TT |
3213 | offset, ret); |
3214 | if (err < 0) | |
3215 | ret = err; | |
3216 | ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN); | |
3217 | } | |
4bd809db | 3218 | |
69c499d1 | 3219 | retake_lock: |
e8340395 JK |
3220 | if (rw == WRITE) |
3221 | inode_dio_done(inode); | |
69c499d1 TT |
3222 | /* take i_mutex locking again if we do a ovewrite dio */ |
3223 | if (overwrite) { | |
69c499d1 TT |
3224 | up_read(&EXT4_I(inode)->i_data_sem); |
3225 | mutex_lock(&inode->i_mutex); | |
4c0425ff | 3226 | } |
8d5d02e6 | 3227 | |
69c499d1 | 3228 | return ret; |
4c0425ff MC |
3229 | } |
3230 | ||
3231 | static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb, | |
d8d3d94b | 3232 | struct iov_iter *iter, loff_t offset) |
4c0425ff MC |
3233 | { |
3234 | struct file *file = iocb->ki_filp; | |
3235 | struct inode *inode = file->f_mapping->host; | |
a6cbcd4a | 3236 | size_t count = iov_iter_count(iter); |
0562e0ba | 3237 | ssize_t ret; |
4c0425ff | 3238 | |
84ebd795 TT |
3239 | /* |
3240 | * If we are doing data journalling we don't support O_DIRECT | |
3241 | */ | |
3242 | if (ext4_should_journal_data(inode)) | |
3243 | return 0; | |
3244 | ||
46c7f254 TM |
3245 | /* Let buffer I/O handle the inline data case. */ |
3246 | if (ext4_has_inline_data(inode)) | |
3247 | return 0; | |
3248 | ||
a6cbcd4a | 3249 | trace_ext4_direct_IO_enter(inode, offset, count, rw); |
12e9b892 | 3250 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
16b1f05d | 3251 | ret = ext4_ext_direct_IO(rw, iocb, iter, offset); |
0562e0ba | 3252 | else |
16b1f05d | 3253 | ret = ext4_ind_direct_IO(rw, iocb, iter, offset); |
a6cbcd4a | 3254 | trace_ext4_direct_IO_exit(inode, offset, count, rw, ret); |
0562e0ba | 3255 | return ret; |
4c0425ff MC |
3256 | } |
3257 | ||
ac27a0ec | 3258 | /* |
617ba13b | 3259 | * Pages can be marked dirty completely asynchronously from ext4's journalling |
ac27a0ec DK |
3260 | * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do |
3261 | * much here because ->set_page_dirty is called under VFS locks. The page is | |
3262 | * not necessarily locked. | |
3263 | * | |
3264 | * We cannot just dirty the page and leave attached buffers clean, because the | |
3265 | * buffers' dirty state is "definitive". We cannot just set the buffers dirty | |
3266 | * or jbddirty because all the journalling code will explode. | |
3267 | * | |
3268 | * So what we do is to mark the page "pending dirty" and next time writepage | |
3269 | * is called, propagate that into the buffers appropriately. | |
3270 | */ | |
617ba13b | 3271 | static int ext4_journalled_set_page_dirty(struct page *page) |
ac27a0ec DK |
3272 | { |
3273 | SetPageChecked(page); | |
3274 | return __set_page_dirty_nobuffers(page); | |
3275 | } | |
3276 | ||
74d553aa | 3277 | static const struct address_space_operations ext4_aops = { |
8ab22b9a HH |
3278 | .readpage = ext4_readpage, |
3279 | .readpages = ext4_readpages, | |
43ce1d23 | 3280 | .writepage = ext4_writepage, |
20970ba6 | 3281 | .writepages = ext4_writepages, |
8ab22b9a | 3282 | .write_begin = ext4_write_begin, |
74d553aa | 3283 | .write_end = ext4_write_end, |
8ab22b9a HH |
3284 | .bmap = ext4_bmap, |
3285 | .invalidatepage = ext4_invalidatepage, | |
3286 | .releasepage = ext4_releasepage, | |
3287 | .direct_IO = ext4_direct_IO, | |
3288 | .migratepage = buffer_migrate_page, | |
3289 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 3290 | .error_remove_page = generic_error_remove_page, |
ac27a0ec DK |
3291 | }; |
3292 | ||
617ba13b | 3293 | static const struct address_space_operations ext4_journalled_aops = { |
8ab22b9a HH |
3294 | .readpage = ext4_readpage, |
3295 | .readpages = ext4_readpages, | |
43ce1d23 | 3296 | .writepage = ext4_writepage, |
20970ba6 | 3297 | .writepages = ext4_writepages, |
8ab22b9a HH |
3298 | .write_begin = ext4_write_begin, |
3299 | .write_end = ext4_journalled_write_end, | |
3300 | .set_page_dirty = ext4_journalled_set_page_dirty, | |
3301 | .bmap = ext4_bmap, | |
4520fb3c | 3302 | .invalidatepage = ext4_journalled_invalidatepage, |
8ab22b9a | 3303 | .releasepage = ext4_releasepage, |
84ebd795 | 3304 | .direct_IO = ext4_direct_IO, |
8ab22b9a | 3305 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 3306 | .error_remove_page = generic_error_remove_page, |
ac27a0ec DK |
3307 | }; |
3308 | ||
64769240 | 3309 | static const struct address_space_operations ext4_da_aops = { |
8ab22b9a HH |
3310 | .readpage = ext4_readpage, |
3311 | .readpages = ext4_readpages, | |
43ce1d23 | 3312 | .writepage = ext4_writepage, |
20970ba6 | 3313 | .writepages = ext4_writepages, |
8ab22b9a HH |
3314 | .write_begin = ext4_da_write_begin, |
3315 | .write_end = ext4_da_write_end, | |
3316 | .bmap = ext4_bmap, | |
3317 | .invalidatepage = ext4_da_invalidatepage, | |
3318 | .releasepage = ext4_releasepage, | |
3319 | .direct_IO = ext4_direct_IO, | |
3320 | .migratepage = buffer_migrate_page, | |
3321 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 3322 | .error_remove_page = generic_error_remove_page, |
64769240 AT |
3323 | }; |
3324 | ||
617ba13b | 3325 | void ext4_set_aops(struct inode *inode) |
ac27a0ec | 3326 | { |
3d2b1582 LC |
3327 | switch (ext4_inode_journal_mode(inode)) { |
3328 | case EXT4_INODE_ORDERED_DATA_MODE: | |
74d553aa | 3329 | ext4_set_inode_state(inode, EXT4_STATE_ORDERED_MODE); |
3d2b1582 LC |
3330 | break; |
3331 | case EXT4_INODE_WRITEBACK_DATA_MODE: | |
74d553aa | 3332 | ext4_clear_inode_state(inode, EXT4_STATE_ORDERED_MODE); |
3d2b1582 LC |
3333 | break; |
3334 | case EXT4_INODE_JOURNAL_DATA_MODE: | |
617ba13b | 3335 | inode->i_mapping->a_ops = &ext4_journalled_aops; |
74d553aa | 3336 | return; |
3d2b1582 LC |
3337 | default: |
3338 | BUG(); | |
3339 | } | |
74d553aa TT |
3340 | if (test_opt(inode->i_sb, DELALLOC)) |
3341 | inode->i_mapping->a_ops = &ext4_da_aops; | |
3342 | else | |
3343 | inode->i_mapping->a_ops = &ext4_aops; | |
ac27a0ec DK |
3344 | } |
3345 | ||
d863dc36 LC |
3346 | /* |
3347 | * ext4_block_zero_page_range() zeros out a mapping of length 'length' | |
3348 | * starting from file offset 'from'. The range to be zero'd must | |
3349 | * be contained with in one block. If the specified range exceeds | |
3350 | * the end of the block it will be shortened to end of the block | |
3351 | * that cooresponds to 'from' | |
3352 | */ | |
94350ab5 | 3353 | static int ext4_block_zero_page_range(handle_t *handle, |
d863dc36 LC |
3354 | struct address_space *mapping, loff_t from, loff_t length) |
3355 | { | |
3356 | ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT; | |
3357 | unsigned offset = from & (PAGE_CACHE_SIZE-1); | |
3358 | unsigned blocksize, max, pos; | |
3359 | ext4_lblk_t iblock; | |
3360 | struct inode *inode = mapping->host; | |
3361 | struct buffer_head *bh; | |
3362 | struct page *page; | |
3363 | int err = 0; | |
3364 | ||
3365 | page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT, | |
3366 | mapping_gfp_mask(mapping) & ~__GFP_FS); | |
3367 | if (!page) | |
3368 | return -ENOMEM; | |
3369 | ||
3370 | blocksize = inode->i_sb->s_blocksize; | |
3371 | max = blocksize - (offset & (blocksize - 1)); | |
3372 | ||
3373 | /* | |
3374 | * correct length if it does not fall between | |
3375 | * 'from' and the end of the block | |
3376 | */ | |
3377 | if (length > max || length < 0) | |
3378 | length = max; | |
3379 | ||
3380 | iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits); | |
3381 | ||
3382 | if (!page_has_buffers(page)) | |
3383 | create_empty_buffers(page, blocksize, 0); | |
3384 | ||
3385 | /* Find the buffer that contains "offset" */ | |
3386 | bh = page_buffers(page); | |
3387 | pos = blocksize; | |
3388 | while (offset >= pos) { | |
3389 | bh = bh->b_this_page; | |
3390 | iblock++; | |
3391 | pos += blocksize; | |
3392 | } | |
d863dc36 LC |
3393 | if (buffer_freed(bh)) { |
3394 | BUFFER_TRACE(bh, "freed: skip"); | |
3395 | goto unlock; | |
3396 | } | |
d863dc36 LC |
3397 | if (!buffer_mapped(bh)) { |
3398 | BUFFER_TRACE(bh, "unmapped"); | |
3399 | ext4_get_block(inode, iblock, bh, 0); | |
3400 | /* unmapped? It's a hole - nothing to do */ | |
3401 | if (!buffer_mapped(bh)) { | |
3402 | BUFFER_TRACE(bh, "still unmapped"); | |
3403 | goto unlock; | |
3404 | } | |
3405 | } | |
3406 | ||
3407 | /* Ok, it's mapped. Make sure it's up-to-date */ | |
3408 | if (PageUptodate(page)) | |
3409 | set_buffer_uptodate(bh); | |
3410 | ||
3411 | if (!buffer_uptodate(bh)) { | |
3412 | err = -EIO; | |
3413 | ll_rw_block(READ, 1, &bh); | |
3414 | wait_on_buffer(bh); | |
3415 | /* Uhhuh. Read error. Complain and punt. */ | |
3416 | if (!buffer_uptodate(bh)) | |
3417 | goto unlock; | |
3418 | } | |
d863dc36 LC |
3419 | if (ext4_should_journal_data(inode)) { |
3420 | BUFFER_TRACE(bh, "get write access"); | |
3421 | err = ext4_journal_get_write_access(handle, bh); | |
3422 | if (err) | |
3423 | goto unlock; | |
3424 | } | |
d863dc36 | 3425 | zero_user(page, offset, length); |
d863dc36 LC |
3426 | BUFFER_TRACE(bh, "zeroed end of block"); |
3427 | ||
d863dc36 LC |
3428 | if (ext4_should_journal_data(inode)) { |
3429 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
0713ed0c | 3430 | } else { |
353eefd3 | 3431 | err = 0; |
d863dc36 | 3432 | mark_buffer_dirty(bh); |
0713ed0c LC |
3433 | if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE)) |
3434 | err = ext4_jbd2_file_inode(handle, inode); | |
3435 | } | |
d863dc36 LC |
3436 | |
3437 | unlock: | |
3438 | unlock_page(page); | |
3439 | page_cache_release(page); | |
3440 | return err; | |
3441 | } | |
3442 | ||
94350ab5 MW |
3443 | /* |
3444 | * ext4_block_truncate_page() zeroes out a mapping from file offset `from' | |
3445 | * up to the end of the block which corresponds to `from'. | |
3446 | * This required during truncate. We need to physically zero the tail end | |
3447 | * of that block so it doesn't yield old data if the file is later grown. | |
3448 | */ | |
c197855e | 3449 | static int ext4_block_truncate_page(handle_t *handle, |
94350ab5 MW |
3450 | struct address_space *mapping, loff_t from) |
3451 | { | |
3452 | unsigned offset = from & (PAGE_CACHE_SIZE-1); | |
3453 | unsigned length; | |
3454 | unsigned blocksize; | |
3455 | struct inode *inode = mapping->host; | |
3456 | ||
3457 | blocksize = inode->i_sb->s_blocksize; | |
3458 | length = blocksize - (offset & (blocksize - 1)); | |
3459 | ||
3460 | return ext4_block_zero_page_range(handle, mapping, from, length); | |
3461 | } | |
3462 | ||
a87dd18c LC |
3463 | int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode, |
3464 | loff_t lstart, loff_t length) | |
3465 | { | |
3466 | struct super_block *sb = inode->i_sb; | |
3467 | struct address_space *mapping = inode->i_mapping; | |
e1be3a92 | 3468 | unsigned partial_start, partial_end; |
a87dd18c LC |
3469 | ext4_fsblk_t start, end; |
3470 | loff_t byte_end = (lstart + length - 1); | |
3471 | int err = 0; | |
3472 | ||
e1be3a92 LC |
3473 | partial_start = lstart & (sb->s_blocksize - 1); |
3474 | partial_end = byte_end & (sb->s_blocksize - 1); | |
3475 | ||
a87dd18c LC |
3476 | start = lstart >> sb->s_blocksize_bits; |
3477 | end = byte_end >> sb->s_blocksize_bits; | |
3478 | ||
3479 | /* Handle partial zero within the single block */ | |
e1be3a92 LC |
3480 | if (start == end && |
3481 | (partial_start || (partial_end != sb->s_blocksize - 1))) { | |
a87dd18c LC |
3482 | err = ext4_block_zero_page_range(handle, mapping, |
3483 | lstart, length); | |
3484 | return err; | |
3485 | } | |
3486 | /* Handle partial zero out on the start of the range */ | |
e1be3a92 | 3487 | if (partial_start) { |
a87dd18c LC |
3488 | err = ext4_block_zero_page_range(handle, mapping, |
3489 | lstart, sb->s_blocksize); | |
3490 | if (err) | |
3491 | return err; | |
3492 | } | |
3493 | /* Handle partial zero out on the end of the range */ | |
e1be3a92 | 3494 | if (partial_end != sb->s_blocksize - 1) |
a87dd18c | 3495 | err = ext4_block_zero_page_range(handle, mapping, |
e1be3a92 LC |
3496 | byte_end - partial_end, |
3497 | partial_end + 1); | |
a87dd18c LC |
3498 | return err; |
3499 | } | |
3500 | ||
91ef4caf DG |
3501 | int ext4_can_truncate(struct inode *inode) |
3502 | { | |
91ef4caf DG |
3503 | if (S_ISREG(inode->i_mode)) |
3504 | return 1; | |
3505 | if (S_ISDIR(inode->i_mode)) | |
3506 | return 1; | |
3507 | if (S_ISLNK(inode->i_mode)) | |
3508 | return !ext4_inode_is_fast_symlink(inode); | |
3509 | return 0; | |
3510 | } | |
3511 | ||
a4bb6b64 AH |
3512 | /* |
3513 | * ext4_punch_hole: punches a hole in a file by releaseing the blocks | |
3514 | * associated with the given offset and length | |
3515 | * | |
3516 | * @inode: File inode | |
3517 | * @offset: The offset where the hole will begin | |
3518 | * @len: The length of the hole | |
3519 | * | |
4907cb7b | 3520 | * Returns: 0 on success or negative on failure |
a4bb6b64 AH |
3521 | */ |
3522 | ||
aeb2817a | 3523 | int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length) |
a4bb6b64 | 3524 | { |
26a4c0c6 TT |
3525 | struct super_block *sb = inode->i_sb; |
3526 | ext4_lblk_t first_block, stop_block; | |
3527 | struct address_space *mapping = inode->i_mapping; | |
a87dd18c | 3528 | loff_t first_block_offset, last_block_offset; |
26a4c0c6 TT |
3529 | handle_t *handle; |
3530 | unsigned int credits; | |
3531 | int ret = 0; | |
3532 | ||
a4bb6b64 | 3533 | if (!S_ISREG(inode->i_mode)) |
73355192 | 3534 | return -EOPNOTSUPP; |
a4bb6b64 | 3535 | |
b8a86845 | 3536 | trace_ext4_punch_hole(inode, offset, length, 0); |
aaddea81 | 3537 | |
26a4c0c6 TT |
3538 | /* |
3539 | * Write out all dirty pages to avoid race conditions | |
3540 | * Then release them. | |
3541 | */ | |
3542 | if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { | |
3543 | ret = filemap_write_and_wait_range(mapping, offset, | |
3544 | offset + length - 1); | |
3545 | if (ret) | |
3546 | return ret; | |
3547 | } | |
3548 | ||
3549 | mutex_lock(&inode->i_mutex); | |
9ef06cec | 3550 | |
26a4c0c6 TT |
3551 | /* No need to punch hole beyond i_size */ |
3552 | if (offset >= inode->i_size) | |
3553 | goto out_mutex; | |
3554 | ||
3555 | /* | |
3556 | * If the hole extends beyond i_size, set the hole | |
3557 | * to end after the page that contains i_size | |
3558 | */ | |
3559 | if (offset + length > inode->i_size) { | |
3560 | length = inode->i_size + | |
3561 | PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) - | |
3562 | offset; | |
3563 | } | |
3564 | ||
a361293f JK |
3565 | if (offset & (sb->s_blocksize - 1) || |
3566 | (offset + length) & (sb->s_blocksize - 1)) { | |
3567 | /* | |
3568 | * Attach jinode to inode for jbd2 if we do any zeroing of | |
3569 | * partial block | |
3570 | */ | |
3571 | ret = ext4_inode_attach_jinode(inode); | |
3572 | if (ret < 0) | |
3573 | goto out_mutex; | |
3574 | ||
3575 | } | |
3576 | ||
a87dd18c LC |
3577 | first_block_offset = round_up(offset, sb->s_blocksize); |
3578 | last_block_offset = round_down((offset + length), sb->s_blocksize) - 1; | |
26a4c0c6 | 3579 | |
a87dd18c LC |
3580 | /* Now release the pages and zero block aligned part of pages*/ |
3581 | if (last_block_offset > first_block_offset) | |
3582 | truncate_pagecache_range(inode, first_block_offset, | |
3583 | last_block_offset); | |
26a4c0c6 TT |
3584 | |
3585 | /* Wait all existing dio workers, newcomers will block on i_mutex */ | |
3586 | ext4_inode_block_unlocked_dio(inode); | |
26a4c0c6 TT |
3587 | inode_dio_wait(inode); |
3588 | ||
3589 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
3590 | credits = ext4_writepage_trans_blocks(inode); | |
3591 | else | |
3592 | credits = ext4_blocks_for_truncate(inode); | |
3593 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); | |
3594 | if (IS_ERR(handle)) { | |
3595 | ret = PTR_ERR(handle); | |
3596 | ext4_std_error(sb, ret); | |
3597 | goto out_dio; | |
3598 | } | |
3599 | ||
a87dd18c LC |
3600 | ret = ext4_zero_partial_blocks(handle, inode, offset, |
3601 | length); | |
3602 | if (ret) | |
3603 | goto out_stop; | |
26a4c0c6 TT |
3604 | |
3605 | first_block = (offset + sb->s_blocksize - 1) >> | |
3606 | EXT4_BLOCK_SIZE_BITS(sb); | |
3607 | stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb); | |
3608 | ||
3609 | /* If there are no blocks to remove, return now */ | |
3610 | if (first_block >= stop_block) | |
3611 | goto out_stop; | |
3612 | ||
3613 | down_write(&EXT4_I(inode)->i_data_sem); | |
3614 | ext4_discard_preallocations(inode); | |
3615 | ||
3616 | ret = ext4_es_remove_extent(inode, first_block, | |
3617 | stop_block - first_block); | |
3618 | if (ret) { | |
3619 | up_write(&EXT4_I(inode)->i_data_sem); | |
3620 | goto out_stop; | |
3621 | } | |
3622 | ||
3623 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
3624 | ret = ext4_ext_remove_space(inode, first_block, | |
3625 | stop_block - 1); | |
3626 | else | |
3627 | ret = ext4_free_hole_blocks(handle, inode, first_block, | |
3628 | stop_block); | |
3629 | ||
819c4920 | 3630 | up_write(&EXT4_I(inode)->i_data_sem); |
26a4c0c6 TT |
3631 | if (IS_SYNC(inode)) |
3632 | ext4_handle_sync(handle); | |
e251f9bc MP |
3633 | |
3634 | /* Now release the pages again to reduce race window */ | |
3635 | if (last_block_offset > first_block_offset) | |
3636 | truncate_pagecache_range(inode, first_block_offset, | |
3637 | last_block_offset); | |
3638 | ||
26a4c0c6 TT |
3639 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); |
3640 | ext4_mark_inode_dirty(handle, inode); | |
3641 | out_stop: | |
3642 | ext4_journal_stop(handle); | |
3643 | out_dio: | |
3644 | ext4_inode_resume_unlocked_dio(inode); | |
3645 | out_mutex: | |
3646 | mutex_unlock(&inode->i_mutex); | |
3647 | return ret; | |
a4bb6b64 AH |
3648 | } |
3649 | ||
a361293f JK |
3650 | int ext4_inode_attach_jinode(struct inode *inode) |
3651 | { | |
3652 | struct ext4_inode_info *ei = EXT4_I(inode); | |
3653 | struct jbd2_inode *jinode; | |
3654 | ||
3655 | if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal) | |
3656 | return 0; | |
3657 | ||
3658 | jinode = jbd2_alloc_inode(GFP_KERNEL); | |
3659 | spin_lock(&inode->i_lock); | |
3660 | if (!ei->jinode) { | |
3661 | if (!jinode) { | |
3662 | spin_unlock(&inode->i_lock); | |
3663 | return -ENOMEM; | |
3664 | } | |
3665 | ei->jinode = jinode; | |
3666 | jbd2_journal_init_jbd_inode(ei->jinode, inode); | |
3667 | jinode = NULL; | |
3668 | } | |
3669 | spin_unlock(&inode->i_lock); | |
3670 | if (unlikely(jinode != NULL)) | |
3671 | jbd2_free_inode(jinode); | |
3672 | return 0; | |
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 | |
e04027e8 | 3712 | * or it's a completely new inode. In those cases we might not |
19b5ef61 TT |
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 | ||
a361293f JK |
3735 | /* If we zero-out tail of the page, we have to create jinode for jbd2 */ |
3736 | if (inode->i_size & (inode->i_sb->s_blocksize - 1)) { | |
3737 | if (ext4_inode_attach_jinode(inode) < 0) | |
3738 | return; | |
3739 | } | |
3740 | ||
819c4920 TT |
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; |
00a1a053 | 3952 | unsigned int new_fl = 0; |
ac27a0ec | 3953 | |
617ba13b | 3954 | if (flags & EXT4_SYNC_FL) |
00a1a053 | 3955 | new_fl |= S_SYNC; |
617ba13b | 3956 | if (flags & EXT4_APPEND_FL) |
00a1a053 | 3957 | new_fl |= S_APPEND; |
617ba13b | 3958 | if (flags & EXT4_IMMUTABLE_FL) |
00a1a053 | 3959 | new_fl |= S_IMMUTABLE; |
617ba13b | 3960 | if (flags & EXT4_NOATIME_FL) |
00a1a053 | 3961 | new_fl |= S_NOATIME; |
617ba13b | 3962 | if (flags & EXT4_DIRSYNC_FL) |
00a1a053 | 3963 | new_fl |= S_DIRSYNC; |
5f16f322 TT |
3964 | inode_set_flags(inode, new_fl, |
3965 | S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC); | |
ac27a0ec DK |
3966 | } |
3967 | ||
ff9ddf7e JK |
3968 | /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */ |
3969 | void ext4_get_inode_flags(struct ext4_inode_info *ei) | |
3970 | { | |
84a8dce2 DM |
3971 | unsigned int vfs_fl; |
3972 | unsigned long old_fl, new_fl; | |
3973 | ||
3974 | do { | |
3975 | vfs_fl = ei->vfs_inode.i_flags; | |
3976 | old_fl = ei->i_flags; | |
3977 | new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL| | |
3978 | EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL| | |
3979 | EXT4_DIRSYNC_FL); | |
3980 | if (vfs_fl & S_SYNC) | |
3981 | new_fl |= EXT4_SYNC_FL; | |
3982 | if (vfs_fl & S_APPEND) | |
3983 | new_fl |= EXT4_APPEND_FL; | |
3984 | if (vfs_fl & S_IMMUTABLE) | |
3985 | new_fl |= EXT4_IMMUTABLE_FL; | |
3986 | if (vfs_fl & S_NOATIME) | |
3987 | new_fl |= EXT4_NOATIME_FL; | |
3988 | if (vfs_fl & S_DIRSYNC) | |
3989 | new_fl |= EXT4_DIRSYNC_FL; | |
3990 | } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl); | |
ff9ddf7e | 3991 | } |
de9a55b8 | 3992 | |
0fc1b451 | 3993 | static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode, |
de9a55b8 | 3994 | struct ext4_inode_info *ei) |
0fc1b451 AK |
3995 | { |
3996 | blkcnt_t i_blocks ; | |
8180a562 AK |
3997 | struct inode *inode = &(ei->vfs_inode); |
3998 | struct super_block *sb = inode->i_sb; | |
0fc1b451 AK |
3999 | |
4000 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, | |
4001 | EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) { | |
4002 | /* we are using combined 48 bit field */ | |
4003 | i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 | | |
4004 | le32_to_cpu(raw_inode->i_blocks_lo); | |
07a03824 | 4005 | if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) { |
8180a562 AK |
4006 | /* i_blocks represent file system block size */ |
4007 | return i_blocks << (inode->i_blkbits - 9); | |
4008 | } else { | |
4009 | return i_blocks; | |
4010 | } | |
0fc1b451 AK |
4011 | } else { |
4012 | return le32_to_cpu(raw_inode->i_blocks_lo); | |
4013 | } | |
4014 | } | |
ff9ddf7e | 4015 | |
152a7b0a TM |
4016 | static inline void ext4_iget_extra_inode(struct inode *inode, |
4017 | struct ext4_inode *raw_inode, | |
4018 | struct ext4_inode_info *ei) | |
4019 | { | |
4020 | __le32 *magic = (void *)raw_inode + | |
4021 | EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize; | |
67cf5b09 | 4022 | if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) { |
152a7b0a | 4023 | ext4_set_inode_state(inode, EXT4_STATE_XATTR); |
67cf5b09 | 4024 | ext4_find_inline_data_nolock(inode); |
f19d5870 TM |
4025 | } else |
4026 | EXT4_I(inode)->i_inline_off = 0; | |
152a7b0a TM |
4027 | } |
4028 | ||
1d1fe1ee | 4029 | struct inode *ext4_iget(struct super_block *sb, unsigned long ino) |
ac27a0ec | 4030 | { |
617ba13b MC |
4031 | struct ext4_iloc iloc; |
4032 | struct ext4_inode *raw_inode; | |
1d1fe1ee | 4033 | struct ext4_inode_info *ei; |
1d1fe1ee | 4034 | struct inode *inode; |
b436b9be | 4035 | journal_t *journal = EXT4_SB(sb)->s_journal; |
1d1fe1ee | 4036 | long ret; |
ac27a0ec | 4037 | int block; |
08cefc7a EB |
4038 | uid_t i_uid; |
4039 | gid_t i_gid; | |
ac27a0ec | 4040 | |
1d1fe1ee DH |
4041 | inode = iget_locked(sb, ino); |
4042 | if (!inode) | |
4043 | return ERR_PTR(-ENOMEM); | |
4044 | if (!(inode->i_state & I_NEW)) | |
4045 | return inode; | |
4046 | ||
4047 | ei = EXT4_I(inode); | |
7dc57615 | 4048 | iloc.bh = NULL; |
ac27a0ec | 4049 | |
1d1fe1ee DH |
4050 | ret = __ext4_get_inode_loc(inode, &iloc, 0); |
4051 | if (ret < 0) | |
ac27a0ec | 4052 | goto bad_inode; |
617ba13b | 4053 | raw_inode = ext4_raw_inode(&iloc); |
814525f4 DW |
4054 | |
4055 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { | |
4056 | ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize); | |
4057 | if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > | |
4058 | EXT4_INODE_SIZE(inode->i_sb)) { | |
4059 | EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)", | |
4060 | EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize, | |
4061 | EXT4_INODE_SIZE(inode->i_sb)); | |
4062 | ret = -EIO; | |
4063 | goto bad_inode; | |
4064 | } | |
4065 | } else | |
4066 | ei->i_extra_isize = 0; | |
4067 | ||
4068 | /* Precompute checksum seed for inode metadata */ | |
4069 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, | |
4070 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) { | |
4071 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
4072 | __u32 csum; | |
4073 | __le32 inum = cpu_to_le32(inode->i_ino); | |
4074 | __le32 gen = raw_inode->i_generation; | |
4075 | csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum, | |
4076 | sizeof(inum)); | |
4077 | ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen, | |
4078 | sizeof(gen)); | |
4079 | } | |
4080 | ||
4081 | if (!ext4_inode_csum_verify(inode, raw_inode, ei)) { | |
4082 | EXT4_ERROR_INODE(inode, "checksum invalid"); | |
4083 | ret = -EIO; | |
4084 | goto bad_inode; | |
4085 | } | |
4086 | ||
ac27a0ec | 4087 | inode->i_mode = le16_to_cpu(raw_inode->i_mode); |
08cefc7a EB |
4088 | i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low); |
4089 | i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low); | |
af5bc92d | 4090 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
08cefc7a EB |
4091 | i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16; |
4092 | i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16; | |
ac27a0ec | 4093 | } |
08cefc7a EB |
4094 | i_uid_write(inode, i_uid); |
4095 | i_gid_write(inode, i_gid); | |
bfe86848 | 4096 | set_nlink(inode, le16_to_cpu(raw_inode->i_links_count)); |
ac27a0ec | 4097 | |
353eb83c | 4098 | ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */ |
67cf5b09 | 4099 | ei->i_inline_off = 0; |
ac27a0ec DK |
4100 | ei->i_dir_start_lookup = 0; |
4101 | ei->i_dtime = le32_to_cpu(raw_inode->i_dtime); | |
4102 | /* We now have enough fields to check if the inode was active or not. | |
4103 | * This is needed because nfsd might try to access dead inodes | |
4104 | * the test is that same one that e2fsck uses | |
4105 | * NeilBrown 1999oct15 | |
4106 | */ | |
4107 | if (inode->i_nlink == 0) { | |
393d1d1d DTB |
4108 | if ((inode->i_mode == 0 || |
4109 | !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) && | |
4110 | ino != EXT4_BOOT_LOADER_INO) { | |
ac27a0ec | 4111 | /* this inode is deleted */ |
1d1fe1ee | 4112 | ret = -ESTALE; |
ac27a0ec DK |
4113 | goto bad_inode; |
4114 | } | |
4115 | /* The only unlinked inodes we let through here have | |
4116 | * valid i_mode and are being read by the orphan | |
4117 | * recovery code: that's fine, we're about to complete | |
393d1d1d DTB |
4118 | * the process of deleting those. |
4119 | * OR it is the EXT4_BOOT_LOADER_INO which is | |
4120 | * not initialized on a new filesystem. */ | |
ac27a0ec | 4121 | } |
ac27a0ec | 4122 | ei->i_flags = le32_to_cpu(raw_inode->i_flags); |
0fc1b451 | 4123 | inode->i_blocks = ext4_inode_blocks(raw_inode, ei); |
7973c0c1 | 4124 | ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo); |
a9e81742 | 4125 | if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) |
a1ddeb7e BP |
4126 | ei->i_file_acl |= |
4127 | ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32; | |
a48380f7 | 4128 | inode->i_size = ext4_isize(raw_inode); |
ac27a0ec | 4129 | ei->i_disksize = inode->i_size; |
a9e7f447 DM |
4130 | #ifdef CONFIG_QUOTA |
4131 | ei->i_reserved_quota = 0; | |
4132 | #endif | |
ac27a0ec DK |
4133 | inode->i_generation = le32_to_cpu(raw_inode->i_generation); |
4134 | ei->i_block_group = iloc.block_group; | |
a4912123 | 4135 | ei->i_last_alloc_group = ~0; |
ac27a0ec DK |
4136 | /* |
4137 | * NOTE! The in-memory inode i_data array is in little-endian order | |
4138 | * even on big-endian machines: we do NOT byteswap the block numbers! | |
4139 | */ | |
617ba13b | 4140 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
ac27a0ec DK |
4141 | ei->i_data[block] = raw_inode->i_block[block]; |
4142 | INIT_LIST_HEAD(&ei->i_orphan); | |
4143 | ||
b436b9be JK |
4144 | /* |
4145 | * Set transaction id's of transactions that have to be committed | |
4146 | * to finish f[data]sync. We set them to currently running transaction | |
4147 | * as we cannot be sure that the inode or some of its metadata isn't | |
4148 | * part of the transaction - the inode could have been reclaimed and | |
4149 | * now it is reread from disk. | |
4150 | */ | |
4151 | if (journal) { | |
4152 | transaction_t *transaction; | |
4153 | tid_t tid; | |
4154 | ||
a931da6a | 4155 | read_lock(&journal->j_state_lock); |
b436b9be JK |
4156 | if (journal->j_running_transaction) |
4157 | transaction = journal->j_running_transaction; | |
4158 | else | |
4159 | transaction = journal->j_committing_transaction; | |
4160 | if (transaction) | |
4161 | tid = transaction->t_tid; | |
4162 | else | |
4163 | tid = journal->j_commit_sequence; | |
a931da6a | 4164 | read_unlock(&journal->j_state_lock); |
b436b9be JK |
4165 | ei->i_sync_tid = tid; |
4166 | ei->i_datasync_tid = tid; | |
4167 | } | |
4168 | ||
0040d987 | 4169 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
ac27a0ec DK |
4170 | if (ei->i_extra_isize == 0) { |
4171 | /* The extra space is currently unused. Use it. */ | |
617ba13b MC |
4172 | ei->i_extra_isize = sizeof(struct ext4_inode) - |
4173 | EXT4_GOOD_OLD_INODE_SIZE; | |
ac27a0ec | 4174 | } else { |
152a7b0a | 4175 | ext4_iget_extra_inode(inode, raw_inode, ei); |
ac27a0ec | 4176 | } |
814525f4 | 4177 | } |
ac27a0ec | 4178 | |
ef7f3835 KS |
4179 | EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode); |
4180 | EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode); | |
4181 | EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode); | |
4182 | EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode); | |
4183 | ||
ed3654eb | 4184 | if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) { |
c4f65706 TT |
4185 | inode->i_version = le32_to_cpu(raw_inode->i_disk_version); |
4186 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { | |
4187 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
4188 | inode->i_version |= | |
4189 | (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32; | |
4190 | } | |
25ec56b5 JNC |
4191 | } |
4192 | ||
c4b5a614 | 4193 | ret = 0; |
485c26ec | 4194 | if (ei->i_file_acl && |
1032988c | 4195 | !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) { |
24676da4 TT |
4196 | EXT4_ERROR_INODE(inode, "bad extended attribute block %llu", |
4197 | ei->i_file_acl); | |
485c26ec TT |
4198 | ret = -EIO; |
4199 | goto bad_inode; | |
f19d5870 TM |
4200 | } else if (!ext4_has_inline_data(inode)) { |
4201 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { | |
4202 | if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || | |
4203 | (S_ISLNK(inode->i_mode) && | |
4204 | !ext4_inode_is_fast_symlink(inode)))) | |
4205 | /* Validate extent which is part of inode */ | |
4206 | ret = ext4_ext_check_inode(inode); | |
4207 | } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || | |
4208 | (S_ISLNK(inode->i_mode) && | |
4209 | !ext4_inode_is_fast_symlink(inode))) { | |
4210 | /* Validate block references which are part of inode */ | |
4211 | ret = ext4_ind_check_inode(inode); | |
4212 | } | |
fe2c8191 | 4213 | } |
567f3e9a | 4214 | if (ret) |
de9a55b8 | 4215 | goto bad_inode; |
7a262f7c | 4216 | |
ac27a0ec | 4217 | if (S_ISREG(inode->i_mode)) { |
617ba13b MC |
4218 | inode->i_op = &ext4_file_inode_operations; |
4219 | inode->i_fop = &ext4_file_operations; | |
4220 | ext4_set_aops(inode); | |
ac27a0ec | 4221 | } else if (S_ISDIR(inode->i_mode)) { |
617ba13b MC |
4222 | inode->i_op = &ext4_dir_inode_operations; |
4223 | inode->i_fop = &ext4_dir_operations; | |
ac27a0ec | 4224 | } else if (S_ISLNK(inode->i_mode)) { |
e83c1397 | 4225 | if (ext4_inode_is_fast_symlink(inode)) { |
617ba13b | 4226 | inode->i_op = &ext4_fast_symlink_inode_operations; |
e83c1397 DG |
4227 | nd_terminate_link(ei->i_data, inode->i_size, |
4228 | sizeof(ei->i_data) - 1); | |
4229 | } else { | |
617ba13b MC |
4230 | inode->i_op = &ext4_symlink_inode_operations; |
4231 | ext4_set_aops(inode); | |
ac27a0ec | 4232 | } |
563bdd61 TT |
4233 | } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) || |
4234 | S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) { | |
617ba13b | 4235 | inode->i_op = &ext4_special_inode_operations; |
ac27a0ec DK |
4236 | if (raw_inode->i_block[0]) |
4237 | init_special_inode(inode, inode->i_mode, | |
4238 | old_decode_dev(le32_to_cpu(raw_inode->i_block[0]))); | |
4239 | else | |
4240 | init_special_inode(inode, inode->i_mode, | |
4241 | new_decode_dev(le32_to_cpu(raw_inode->i_block[1]))); | |
393d1d1d DTB |
4242 | } else if (ino == EXT4_BOOT_LOADER_INO) { |
4243 | make_bad_inode(inode); | |
563bdd61 | 4244 | } else { |
563bdd61 | 4245 | ret = -EIO; |
24676da4 | 4246 | EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode); |
563bdd61 | 4247 | goto bad_inode; |
ac27a0ec | 4248 | } |
af5bc92d | 4249 | brelse(iloc.bh); |
617ba13b | 4250 | ext4_set_inode_flags(inode); |
1d1fe1ee DH |
4251 | unlock_new_inode(inode); |
4252 | return inode; | |
ac27a0ec DK |
4253 | |
4254 | bad_inode: | |
567f3e9a | 4255 | brelse(iloc.bh); |
1d1fe1ee DH |
4256 | iget_failed(inode); |
4257 | return ERR_PTR(ret); | |
ac27a0ec DK |
4258 | } |
4259 | ||
0fc1b451 AK |
4260 | static int ext4_inode_blocks_set(handle_t *handle, |
4261 | struct ext4_inode *raw_inode, | |
4262 | struct ext4_inode_info *ei) | |
4263 | { | |
4264 | struct inode *inode = &(ei->vfs_inode); | |
4265 | u64 i_blocks = inode->i_blocks; | |
4266 | struct super_block *sb = inode->i_sb; | |
0fc1b451 AK |
4267 | |
4268 | if (i_blocks <= ~0U) { | |
4269 | /* | |
4907cb7b | 4270 | * i_blocks can be represented in a 32 bit variable |
0fc1b451 AK |
4271 | * as multiple of 512 bytes |
4272 | */ | |
8180a562 | 4273 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 4274 | raw_inode->i_blocks_high = 0; |
84a8dce2 | 4275 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
f287a1a5 TT |
4276 | return 0; |
4277 | } | |
4278 | if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) | |
4279 | return -EFBIG; | |
4280 | ||
4281 | if (i_blocks <= 0xffffffffffffULL) { | |
0fc1b451 AK |
4282 | /* |
4283 | * i_blocks can be represented in a 48 bit variable | |
4284 | * as multiple of 512 bytes | |
4285 | */ | |
8180a562 | 4286 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 4287 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); |
84a8dce2 | 4288 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
0fc1b451 | 4289 | } else { |
84a8dce2 | 4290 | ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
8180a562 AK |
4291 | /* i_block is stored in file system block size */ |
4292 | i_blocks = i_blocks >> (inode->i_blkbits - 9); | |
4293 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); | |
4294 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); | |
0fc1b451 | 4295 | } |
f287a1a5 | 4296 | return 0; |
0fc1b451 AK |
4297 | } |
4298 | ||
ac27a0ec DK |
4299 | /* |
4300 | * Post the struct inode info into an on-disk inode location in the | |
4301 | * buffer-cache. This gobbles the caller's reference to the | |
4302 | * buffer_head in the inode location struct. | |
4303 | * | |
4304 | * The caller must have write access to iloc->bh. | |
4305 | */ | |
617ba13b | 4306 | static int ext4_do_update_inode(handle_t *handle, |
ac27a0ec | 4307 | struct inode *inode, |
830156c7 | 4308 | struct ext4_iloc *iloc) |
ac27a0ec | 4309 | { |
617ba13b MC |
4310 | struct ext4_inode *raw_inode = ext4_raw_inode(iloc); |
4311 | struct ext4_inode_info *ei = EXT4_I(inode); | |
ac27a0ec | 4312 | struct buffer_head *bh = iloc->bh; |
202ee5df | 4313 | struct super_block *sb = inode->i_sb; |
ac27a0ec | 4314 | int err = 0, rc, block; |
202ee5df | 4315 | int need_datasync = 0, set_large_file = 0; |
08cefc7a EB |
4316 | uid_t i_uid; |
4317 | gid_t i_gid; | |
ac27a0ec | 4318 | |
202ee5df TT |
4319 | spin_lock(&ei->i_raw_lock); |
4320 | ||
4321 | /* For fields not tracked in the in-memory inode, | |
ac27a0ec | 4322 | * initialise them to zero for new inodes. */ |
19f5fb7a | 4323 | if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) |
617ba13b | 4324 | memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size); |
ac27a0ec | 4325 | |
ff9ddf7e | 4326 | ext4_get_inode_flags(ei); |
ac27a0ec | 4327 | raw_inode->i_mode = cpu_to_le16(inode->i_mode); |
08cefc7a EB |
4328 | i_uid = i_uid_read(inode); |
4329 | i_gid = i_gid_read(inode); | |
af5bc92d | 4330 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
08cefc7a EB |
4331 | raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid)); |
4332 | raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid)); | |
ac27a0ec DK |
4333 | /* |
4334 | * Fix up interoperability with old kernels. Otherwise, old inodes get | |
4335 | * re-used with the upper 16 bits of the uid/gid intact | |
4336 | */ | |
af5bc92d | 4337 | if (!ei->i_dtime) { |
ac27a0ec | 4338 | raw_inode->i_uid_high = |
08cefc7a | 4339 | cpu_to_le16(high_16_bits(i_uid)); |
ac27a0ec | 4340 | raw_inode->i_gid_high = |
08cefc7a | 4341 | cpu_to_le16(high_16_bits(i_gid)); |
ac27a0ec DK |
4342 | } else { |
4343 | raw_inode->i_uid_high = 0; | |
4344 | raw_inode->i_gid_high = 0; | |
4345 | } | |
4346 | } else { | |
08cefc7a EB |
4347 | raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid)); |
4348 | raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid)); | |
ac27a0ec DK |
4349 | raw_inode->i_uid_high = 0; |
4350 | raw_inode->i_gid_high = 0; | |
4351 | } | |
4352 | raw_inode->i_links_count = cpu_to_le16(inode->i_nlink); | |
ef7f3835 KS |
4353 | |
4354 | EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode); | |
4355 | EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode); | |
4356 | EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode); | |
4357 | EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode); | |
4358 | ||
202ee5df TT |
4359 | if (ext4_inode_blocks_set(handle, raw_inode, ei)) { |
4360 | spin_unlock(&ei->i_raw_lock); | |
0fc1b451 | 4361 | goto out_brelse; |
202ee5df | 4362 | } |
ac27a0ec | 4363 | raw_inode->i_dtime = cpu_to_le32(ei->i_dtime); |
353eb83c | 4364 | raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF); |
ed3654eb | 4365 | if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) |
a1ddeb7e BP |
4366 | raw_inode->i_file_acl_high = |
4367 | cpu_to_le16(ei->i_file_acl >> 32); | |
7973c0c1 | 4368 | raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl); |
b71fc079 JK |
4369 | if (ei->i_disksize != ext4_isize(raw_inode)) { |
4370 | ext4_isize_set(raw_inode, ei->i_disksize); | |
4371 | need_datasync = 1; | |
4372 | } | |
a48380f7 | 4373 | if (ei->i_disksize > 0x7fffffffULL) { |
a48380f7 AK |
4374 | if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, |
4375 | EXT4_FEATURE_RO_COMPAT_LARGE_FILE) || | |
4376 | EXT4_SB(sb)->s_es->s_rev_level == | |
202ee5df TT |
4377 | cpu_to_le32(EXT4_GOOD_OLD_REV)) |
4378 | set_large_file = 1; | |
ac27a0ec DK |
4379 | } |
4380 | raw_inode->i_generation = cpu_to_le32(inode->i_generation); | |
4381 | if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) { | |
4382 | if (old_valid_dev(inode->i_rdev)) { | |
4383 | raw_inode->i_block[0] = | |
4384 | cpu_to_le32(old_encode_dev(inode->i_rdev)); | |
4385 | raw_inode->i_block[1] = 0; | |
4386 | } else { | |
4387 | raw_inode->i_block[0] = 0; | |
4388 | raw_inode->i_block[1] = | |
4389 | cpu_to_le32(new_encode_dev(inode->i_rdev)); | |
4390 | raw_inode->i_block[2] = 0; | |
4391 | } | |
f19d5870 | 4392 | } else if (!ext4_has_inline_data(inode)) { |
de9a55b8 TT |
4393 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
4394 | raw_inode->i_block[block] = ei->i_data[block]; | |
f19d5870 | 4395 | } |
ac27a0ec | 4396 | |
ed3654eb | 4397 | if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) { |
c4f65706 TT |
4398 | raw_inode->i_disk_version = cpu_to_le32(inode->i_version); |
4399 | if (ei->i_extra_isize) { | |
4400 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
4401 | raw_inode->i_version_hi = | |
4402 | cpu_to_le32(inode->i_version >> 32); | |
4403 | raw_inode->i_extra_isize = | |
4404 | cpu_to_le16(ei->i_extra_isize); | |
4405 | } | |
25ec56b5 JNC |
4406 | } |
4407 | ||
814525f4 DW |
4408 | ext4_inode_csum_set(inode, raw_inode, ei); |
4409 | ||
202ee5df TT |
4410 | spin_unlock(&ei->i_raw_lock); |
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); |
202ee5df | 4417 | if (set_large_file) { |
5d601255 | 4418 | BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access"); |
202ee5df TT |
4419 | err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh); |
4420 | if (err) | |
4421 | goto out_brelse; | |
4422 | ext4_update_dynamic_rev(sb); | |
4423 | EXT4_SET_RO_COMPAT_FEATURE(sb, | |
4424 | EXT4_FEATURE_RO_COMPAT_LARGE_FILE); | |
4425 | ext4_handle_sync(handle); | |
4426 | err = ext4_handle_dirty_super(handle, sb); | |
4427 | } | |
b71fc079 | 4428 | ext4_update_inode_fsync_trans(handle, inode, need_datasync); |
ac27a0ec | 4429 | out_brelse: |
af5bc92d | 4430 | brelse(bh); |
617ba13b | 4431 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
4432 | return err; |
4433 | } | |
4434 | ||
4435 | /* | |
617ba13b | 4436 | * ext4_write_inode() |
ac27a0ec DK |
4437 | * |
4438 | * We are called from a few places: | |
4439 | * | |
87f7e416 | 4440 | * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files. |
ac27a0ec | 4441 | * Here, there will be no transaction running. We wait for any running |
4907cb7b | 4442 | * transaction to commit. |
ac27a0ec | 4443 | * |
87f7e416 TT |
4444 | * - Within flush work (sys_sync(), kupdate and such). |
4445 | * We wait on commit, if told to. | |
ac27a0ec | 4446 | * |
87f7e416 TT |
4447 | * - Within iput_final() -> write_inode_now() |
4448 | * We wait on commit, if told to. | |
ac27a0ec DK |
4449 | * |
4450 | * In all cases it is actually safe for us to return without doing anything, | |
4451 | * because the inode has been copied into a raw inode buffer in | |
87f7e416 TT |
4452 | * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL |
4453 | * writeback. | |
ac27a0ec DK |
4454 | * |
4455 | * Note that we are absolutely dependent upon all inode dirtiers doing the | |
4456 | * right thing: they *must* call mark_inode_dirty() after dirtying info in | |
4457 | * which we are interested. | |
4458 | * | |
4459 | * It would be a bug for them to not do this. The code: | |
4460 | * | |
4461 | * mark_inode_dirty(inode) | |
4462 | * stuff(); | |
4463 | * inode->i_size = expr; | |
4464 | * | |
87f7e416 TT |
4465 | * is in error because write_inode() could occur while `stuff()' is running, |
4466 | * and the new i_size will be lost. Plus the inode will no longer be on the | |
4467 | * superblock's dirty inode list. | |
ac27a0ec | 4468 | */ |
a9185b41 | 4469 | int ext4_write_inode(struct inode *inode, struct writeback_control *wbc) |
ac27a0ec | 4470 | { |
91ac6f43 FM |
4471 | int err; |
4472 | ||
87f7e416 | 4473 | if (WARN_ON_ONCE(current->flags & PF_MEMALLOC)) |
ac27a0ec DK |
4474 | return 0; |
4475 | ||
91ac6f43 FM |
4476 | if (EXT4_SB(inode->i_sb)->s_journal) { |
4477 | if (ext4_journal_current_handle()) { | |
4478 | jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n"); | |
4479 | dump_stack(); | |
4480 | return -EIO; | |
4481 | } | |
ac27a0ec | 4482 | |
10542c22 JK |
4483 | /* |
4484 | * No need to force transaction in WB_SYNC_NONE mode. Also | |
4485 | * ext4_sync_fs() will force the commit after everything is | |
4486 | * written. | |
4487 | */ | |
4488 | if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync) | |
91ac6f43 FM |
4489 | return 0; |
4490 | ||
4491 | err = ext4_force_commit(inode->i_sb); | |
4492 | } else { | |
4493 | struct ext4_iloc iloc; | |
ac27a0ec | 4494 | |
8b472d73 | 4495 | err = __ext4_get_inode_loc(inode, &iloc, 0); |
91ac6f43 FM |
4496 | if (err) |
4497 | return err; | |
10542c22 JK |
4498 | /* |
4499 | * sync(2) will flush the whole buffer cache. No need to do | |
4500 | * it here separately for each inode. | |
4501 | */ | |
4502 | if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) | |
830156c7 FM |
4503 | sync_dirty_buffer(iloc.bh); |
4504 | if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) { | |
c398eda0 TT |
4505 | EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr, |
4506 | "IO error syncing inode"); | |
830156c7 FM |
4507 | err = -EIO; |
4508 | } | |
fd2dd9fb | 4509 | brelse(iloc.bh); |
91ac6f43 FM |
4510 | } |
4511 | return err; | |
ac27a0ec DK |
4512 | } |
4513 | ||
53e87268 JK |
4514 | /* |
4515 | * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate | |
4516 | * buffers that are attached to a page stradding i_size and are undergoing | |
4517 | * commit. In that case we have to wait for commit to finish and try again. | |
4518 | */ | |
4519 | static void ext4_wait_for_tail_page_commit(struct inode *inode) | |
4520 | { | |
4521 | struct page *page; | |
4522 | unsigned offset; | |
4523 | journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; | |
4524 | tid_t commit_tid = 0; | |
4525 | int ret; | |
4526 | ||
4527 | offset = inode->i_size & (PAGE_CACHE_SIZE - 1); | |
4528 | /* | |
4529 | * All buffers in the last page remain valid? Then there's nothing to | |
4530 | * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE == | |
4531 | * blocksize case | |
4532 | */ | |
4533 | if (offset > PAGE_CACHE_SIZE - (1 << inode->i_blkbits)) | |
4534 | return; | |
4535 | while (1) { | |
4536 | page = find_lock_page(inode->i_mapping, | |
4537 | inode->i_size >> PAGE_CACHE_SHIFT); | |
4538 | if (!page) | |
4539 | return; | |
ca99fdd2 LC |
4540 | ret = __ext4_journalled_invalidatepage(page, offset, |
4541 | PAGE_CACHE_SIZE - offset); | |
53e87268 JK |
4542 | unlock_page(page); |
4543 | page_cache_release(page); | |
4544 | if (ret != -EBUSY) | |
4545 | return; | |
4546 | commit_tid = 0; | |
4547 | read_lock(&journal->j_state_lock); | |
4548 | if (journal->j_committing_transaction) | |
4549 | commit_tid = journal->j_committing_transaction->t_tid; | |
4550 | read_unlock(&journal->j_state_lock); | |
4551 | if (commit_tid) | |
4552 | jbd2_log_wait_commit(journal, commit_tid); | |
4553 | } | |
4554 | } | |
4555 | ||
ac27a0ec | 4556 | /* |
617ba13b | 4557 | * ext4_setattr() |
ac27a0ec DK |
4558 | * |
4559 | * Called from notify_change. | |
4560 | * | |
4561 | * We want to trap VFS attempts to truncate the file as soon as | |
4562 | * possible. In particular, we want to make sure that when the VFS | |
4563 | * shrinks i_size, we put the inode on the orphan list and modify | |
4564 | * i_disksize immediately, so that during the subsequent flushing of | |
4565 | * dirty pages and freeing of disk blocks, we can guarantee that any | |
4566 | * commit will leave the blocks being flushed in an unused state on | |
4567 | * disk. (On recovery, the inode will get truncated and the blocks will | |
4568 | * be freed, so we have a strong guarantee that no future commit will | |
4569 | * leave these blocks visible to the user.) | |
4570 | * | |
678aaf48 JK |
4571 | * Another thing we have to assure is that if we are in ordered mode |
4572 | * and inode is still attached to the committing transaction, we must | |
4573 | * we start writeout of all the dirty pages which are being truncated. | |
4574 | * This way we are sure that all the data written in the previous | |
4575 | * transaction are already on disk (truncate waits for pages under | |
4576 | * writeback). | |
4577 | * | |
4578 | * Called with inode->i_mutex down. | |
ac27a0ec | 4579 | */ |
617ba13b | 4580 | int ext4_setattr(struct dentry *dentry, struct iattr *attr) |
ac27a0ec DK |
4581 | { |
4582 | struct inode *inode = dentry->d_inode; | |
4583 | int error, rc = 0; | |
3d287de3 | 4584 | int orphan = 0; |
ac27a0ec DK |
4585 | const unsigned int ia_valid = attr->ia_valid; |
4586 | ||
4587 | error = inode_change_ok(inode, attr); | |
4588 | if (error) | |
4589 | return error; | |
4590 | ||
12755627 | 4591 | if (is_quota_modification(inode, attr)) |
871a2931 | 4592 | dquot_initialize(inode); |
08cefc7a EB |
4593 | if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) || |
4594 | (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) { | |
ac27a0ec DK |
4595 | handle_t *handle; |
4596 | ||
4597 | /* (user+group)*(old+new) structure, inode write (sb, | |
4598 | * inode block, ? - but truncate inode update has it) */ | |
9924a92a TT |
4599 | handle = ext4_journal_start(inode, EXT4_HT_QUOTA, |
4600 | (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) + | |
4601 | EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3); | |
ac27a0ec DK |
4602 | if (IS_ERR(handle)) { |
4603 | error = PTR_ERR(handle); | |
4604 | goto err_out; | |
4605 | } | |
b43fa828 | 4606 | error = dquot_transfer(inode, attr); |
ac27a0ec | 4607 | if (error) { |
617ba13b | 4608 | ext4_journal_stop(handle); |
ac27a0ec DK |
4609 | return error; |
4610 | } | |
4611 | /* Update corresponding info in inode so that everything is in | |
4612 | * one transaction */ | |
4613 | if (attr->ia_valid & ATTR_UID) | |
4614 | inode->i_uid = attr->ia_uid; | |
4615 | if (attr->ia_valid & ATTR_GID) | |
4616 | inode->i_gid = attr->ia_gid; | |
617ba13b MC |
4617 | error = ext4_mark_inode_dirty(handle, inode); |
4618 | ext4_journal_stop(handle); | |
ac27a0ec DK |
4619 | } |
4620 | ||
5208386c JK |
4621 | if (attr->ia_valid & ATTR_SIZE && attr->ia_size != inode->i_size) { |
4622 | handle_t *handle; | |
562c72aa | 4623 | |
12e9b892 | 4624 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) { |
e2b46574 ES |
4625 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
4626 | ||
0c095c7f TT |
4627 | if (attr->ia_size > sbi->s_bitmap_maxbytes) |
4628 | return -EFBIG; | |
e2b46574 | 4629 | } |
dff6efc3 CH |
4630 | |
4631 | if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size) | |
4632 | inode_inc_iversion(inode); | |
4633 | ||
5208386c JK |
4634 | if (S_ISREG(inode->i_mode) && |
4635 | (attr->ia_size < inode->i_size)) { | |
4636 | if (ext4_should_order_data(inode)) { | |
4637 | error = ext4_begin_ordered_truncate(inode, | |
678aaf48 | 4638 | attr->ia_size); |
5208386c | 4639 | if (error) |
678aaf48 | 4640 | goto err_out; |
5208386c JK |
4641 | } |
4642 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 3); | |
4643 | if (IS_ERR(handle)) { | |
4644 | error = PTR_ERR(handle); | |
4645 | goto err_out; | |
4646 | } | |
4647 | if (ext4_handle_valid(handle)) { | |
4648 | error = ext4_orphan_add(handle, inode); | |
4649 | orphan = 1; | |
4650 | } | |
90e775b7 | 4651 | down_write(&EXT4_I(inode)->i_data_sem); |
5208386c JK |
4652 | EXT4_I(inode)->i_disksize = attr->ia_size; |
4653 | rc = ext4_mark_inode_dirty(handle, inode); | |
4654 | if (!error) | |
4655 | error = rc; | |
90e775b7 JK |
4656 | /* |
4657 | * We have to update i_size under i_data_sem together | |
4658 | * with i_disksize to avoid races with writeback code | |
4659 | * running ext4_wb_update_i_disksize(). | |
4660 | */ | |
4661 | if (!error) | |
4662 | i_size_write(inode, attr->ia_size); | |
4663 | up_write(&EXT4_I(inode)->i_data_sem); | |
5208386c JK |
4664 | ext4_journal_stop(handle); |
4665 | if (error) { | |
4666 | ext4_orphan_del(NULL, inode); | |
678aaf48 JK |
4667 | goto err_out; |
4668 | } | |
90e775b7 JK |
4669 | } else |
4670 | i_size_write(inode, attr->ia_size); | |
53e87268 | 4671 | |
5208386c JK |
4672 | /* |
4673 | * Blocks are going to be removed from the inode. Wait | |
4674 | * for dio in flight. Temporarily disable | |
4675 | * dioread_nolock to prevent livelock. | |
4676 | */ | |
4677 | if (orphan) { | |
4678 | if (!ext4_should_journal_data(inode)) { | |
4679 | ext4_inode_block_unlocked_dio(inode); | |
4680 | inode_dio_wait(inode); | |
4681 | ext4_inode_resume_unlocked_dio(inode); | |
4682 | } else | |
4683 | ext4_wait_for_tail_page_commit(inode); | |
1c9114f9 | 4684 | } |
5208386c JK |
4685 | /* |
4686 | * Truncate pagecache after we've waited for commit | |
4687 | * in data=journal mode to make pages freeable. | |
4688 | */ | |
7caef267 | 4689 | truncate_pagecache(inode, inode->i_size); |
072bd7ea | 4690 | } |
5208386c JK |
4691 | /* |
4692 | * We want to call ext4_truncate() even if attr->ia_size == | |
4693 | * inode->i_size for cases like truncation of fallocated space | |
4694 | */ | |
4695 | if (attr->ia_valid & ATTR_SIZE) | |
4696 | ext4_truncate(inode); | |
ac27a0ec | 4697 | |
1025774c CH |
4698 | if (!rc) { |
4699 | setattr_copy(inode, attr); | |
4700 | mark_inode_dirty(inode); | |
4701 | } | |
4702 | ||
4703 | /* | |
4704 | * If the call to ext4_truncate failed to get a transaction handle at | |
4705 | * all, we need to clean up the in-core orphan list manually. | |
4706 | */ | |
3d287de3 | 4707 | if (orphan && inode->i_nlink) |
617ba13b | 4708 | ext4_orphan_del(NULL, inode); |
ac27a0ec DK |
4709 | |
4710 | if (!rc && (ia_valid & ATTR_MODE)) | |
64e178a7 | 4711 | rc = posix_acl_chmod(inode, inode->i_mode); |
ac27a0ec DK |
4712 | |
4713 | err_out: | |
617ba13b | 4714 | ext4_std_error(inode->i_sb, error); |
ac27a0ec DK |
4715 | if (!error) |
4716 | error = rc; | |
4717 | return error; | |
4718 | } | |
4719 | ||
3e3398a0 MC |
4720 | int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry, |
4721 | struct kstat *stat) | |
4722 | { | |
4723 | struct inode *inode; | |
8af8eecc | 4724 | unsigned long long delalloc_blocks; |
3e3398a0 MC |
4725 | |
4726 | inode = dentry->d_inode; | |
4727 | generic_fillattr(inode, stat); | |
4728 | ||
9206c561 AD |
4729 | /* |
4730 | * If there is inline data in the inode, the inode will normally not | |
4731 | * have data blocks allocated (it may have an external xattr block). | |
4732 | * Report at least one sector for such files, so tools like tar, rsync, | |
4733 | * others doen't incorrectly think the file is completely sparse. | |
4734 | */ | |
4735 | if (unlikely(ext4_has_inline_data(inode))) | |
4736 | stat->blocks += (stat->size + 511) >> 9; | |
4737 | ||
3e3398a0 MC |
4738 | /* |
4739 | * We can't update i_blocks if the block allocation is delayed | |
4740 | * otherwise in the case of system crash before the real block | |
4741 | * allocation is done, we will have i_blocks inconsistent with | |
4742 | * on-disk file blocks. | |
4743 | * We always keep i_blocks updated together with real | |
4744 | * allocation. But to not confuse with user, stat | |
4745 | * will return the blocks that include the delayed allocation | |
4746 | * blocks for this file. | |
4747 | */ | |
96607551 | 4748 | delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb), |
9206c561 AD |
4749 | EXT4_I(inode)->i_reserved_data_blocks); |
4750 | stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9); | |
3e3398a0 MC |
4751 | return 0; |
4752 | } | |
ac27a0ec | 4753 | |
fffb2739 JK |
4754 | static int ext4_index_trans_blocks(struct inode *inode, int lblocks, |
4755 | int pextents) | |
a02908f1 | 4756 | { |
12e9b892 | 4757 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) |
fffb2739 JK |
4758 | return ext4_ind_trans_blocks(inode, lblocks); |
4759 | return ext4_ext_index_trans_blocks(inode, pextents); | |
a02908f1 | 4760 | } |
ac51d837 | 4761 | |
ac27a0ec | 4762 | /* |
a02908f1 MC |
4763 | * Account for index blocks, block groups bitmaps and block group |
4764 | * descriptor blocks if modify datablocks and index blocks | |
4765 | * worse case, the indexs blocks spread over different block groups | |
ac27a0ec | 4766 | * |
a02908f1 | 4767 | * If datablocks are discontiguous, they are possible to spread over |
4907cb7b | 4768 | * different block groups too. If they are contiguous, with flexbg, |
a02908f1 | 4769 | * they could still across block group boundary. |
ac27a0ec | 4770 | * |
a02908f1 MC |
4771 | * Also account for superblock, inode, quota and xattr blocks |
4772 | */ | |
fffb2739 JK |
4773 | static int ext4_meta_trans_blocks(struct inode *inode, int lblocks, |
4774 | int pextents) | |
a02908f1 | 4775 | { |
8df9675f TT |
4776 | ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb); |
4777 | int gdpblocks; | |
a02908f1 MC |
4778 | int idxblocks; |
4779 | int ret = 0; | |
4780 | ||
4781 | /* | |
fffb2739 JK |
4782 | * How many index blocks need to touch to map @lblocks logical blocks |
4783 | * to @pextents physical extents? | |
a02908f1 | 4784 | */ |
fffb2739 | 4785 | idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents); |
a02908f1 MC |
4786 | |
4787 | ret = idxblocks; | |
4788 | ||
4789 | /* | |
4790 | * Now let's see how many group bitmaps and group descriptors need | |
4791 | * to account | |
4792 | */ | |
fffb2739 | 4793 | groups = idxblocks + pextents; |
a02908f1 | 4794 | gdpblocks = groups; |
8df9675f TT |
4795 | if (groups > ngroups) |
4796 | groups = ngroups; | |
a02908f1 MC |
4797 | if (groups > EXT4_SB(inode->i_sb)->s_gdb_count) |
4798 | gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count; | |
4799 | ||
4800 | /* bitmaps and block group descriptor blocks */ | |
4801 | ret += groups + gdpblocks; | |
4802 | ||
4803 | /* Blocks for super block, inode, quota and xattr blocks */ | |
4804 | ret += EXT4_META_TRANS_BLOCKS(inode->i_sb); | |
4805 | ||
4806 | return ret; | |
4807 | } | |
4808 | ||
4809 | /* | |
25985edc | 4810 | * Calculate the total number of credits to reserve to fit |
f3bd1f3f MC |
4811 | * the modification of a single pages into a single transaction, |
4812 | * which may include multiple chunks of block allocations. | |
ac27a0ec | 4813 | * |
525f4ed8 | 4814 | * This could be called via ext4_write_begin() |
ac27a0ec | 4815 | * |
525f4ed8 | 4816 | * We need to consider the worse case, when |
a02908f1 | 4817 | * one new block per extent. |
ac27a0ec | 4818 | */ |
a86c6181 | 4819 | int ext4_writepage_trans_blocks(struct inode *inode) |
ac27a0ec | 4820 | { |
617ba13b | 4821 | int bpp = ext4_journal_blocks_per_page(inode); |
ac27a0ec DK |
4822 | int ret; |
4823 | ||
fffb2739 | 4824 | ret = ext4_meta_trans_blocks(inode, bpp, bpp); |
a86c6181 | 4825 | |
a02908f1 | 4826 | /* Account for data blocks for journalled mode */ |
617ba13b | 4827 | if (ext4_should_journal_data(inode)) |
a02908f1 | 4828 | ret += bpp; |
ac27a0ec DK |
4829 | return ret; |
4830 | } | |
f3bd1f3f MC |
4831 | |
4832 | /* | |
4833 | * Calculate the journal credits for a chunk of data modification. | |
4834 | * | |
4835 | * This is called from DIO, fallocate or whoever calling | |
79e83036 | 4836 | * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks. |
f3bd1f3f MC |
4837 | * |
4838 | * journal buffers for data blocks are not included here, as DIO | |
4839 | * and fallocate do no need to journal data buffers. | |
4840 | */ | |
4841 | int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks) | |
4842 | { | |
4843 | return ext4_meta_trans_blocks(inode, nrblocks, 1); | |
4844 | } | |
4845 | ||
ac27a0ec | 4846 | /* |
617ba13b | 4847 | * The caller must have previously called ext4_reserve_inode_write(). |
ac27a0ec DK |
4848 | * Give this, we know that the caller already has write access to iloc->bh. |
4849 | */ | |
617ba13b | 4850 | int ext4_mark_iloc_dirty(handle_t *handle, |
de9a55b8 | 4851 | struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec DK |
4852 | { |
4853 | int err = 0; | |
4854 | ||
c64db50e | 4855 | if (IS_I_VERSION(inode)) |
25ec56b5 JNC |
4856 | inode_inc_iversion(inode); |
4857 | ||
ac27a0ec DK |
4858 | /* the do_update_inode consumes one bh->b_count */ |
4859 | get_bh(iloc->bh); | |
4860 | ||
dab291af | 4861 | /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */ |
830156c7 | 4862 | err = ext4_do_update_inode(handle, inode, iloc); |
ac27a0ec DK |
4863 | put_bh(iloc->bh); |
4864 | return err; | |
4865 | } | |
4866 | ||
4867 | /* | |
4868 | * On success, We end up with an outstanding reference count against | |
4869 | * iloc->bh. This _must_ be cleaned up later. | |
4870 | */ | |
4871 | ||
4872 | int | |
617ba13b MC |
4873 | ext4_reserve_inode_write(handle_t *handle, struct inode *inode, |
4874 | struct ext4_iloc *iloc) | |
ac27a0ec | 4875 | { |
0390131b FM |
4876 | int err; |
4877 | ||
4878 | err = ext4_get_inode_loc(inode, iloc); | |
4879 | if (!err) { | |
4880 | BUFFER_TRACE(iloc->bh, "get_write_access"); | |
4881 | err = ext4_journal_get_write_access(handle, iloc->bh); | |
4882 | if (err) { | |
4883 | brelse(iloc->bh); | |
4884 | iloc->bh = NULL; | |
ac27a0ec DK |
4885 | } |
4886 | } | |
617ba13b | 4887 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
4888 | return err; |
4889 | } | |
4890 | ||
6dd4ee7c KS |
4891 | /* |
4892 | * Expand an inode by new_extra_isize bytes. | |
4893 | * Returns 0 on success or negative error number on failure. | |
4894 | */ | |
1d03ec98 AK |
4895 | static int ext4_expand_extra_isize(struct inode *inode, |
4896 | unsigned int new_extra_isize, | |
4897 | struct ext4_iloc iloc, | |
4898 | handle_t *handle) | |
6dd4ee7c KS |
4899 | { |
4900 | struct ext4_inode *raw_inode; | |
4901 | struct ext4_xattr_ibody_header *header; | |
6dd4ee7c KS |
4902 | |
4903 | if (EXT4_I(inode)->i_extra_isize >= new_extra_isize) | |
4904 | return 0; | |
4905 | ||
4906 | raw_inode = ext4_raw_inode(&iloc); | |
4907 | ||
4908 | header = IHDR(inode, raw_inode); | |
6dd4ee7c KS |
4909 | |
4910 | /* No extended attributes present */ | |
19f5fb7a TT |
4911 | if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) || |
4912 | header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) { | |
6dd4ee7c KS |
4913 | memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0, |
4914 | new_extra_isize); | |
4915 | EXT4_I(inode)->i_extra_isize = new_extra_isize; | |
4916 | return 0; | |
4917 | } | |
4918 | ||
4919 | /* try to expand with EAs present */ | |
4920 | return ext4_expand_extra_isize_ea(inode, new_extra_isize, | |
4921 | raw_inode, handle); | |
4922 | } | |
4923 | ||
ac27a0ec DK |
4924 | /* |
4925 | * What we do here is to mark the in-core inode as clean with respect to inode | |
4926 | * dirtiness (it may still be data-dirty). | |
4927 | * This means that the in-core inode may be reaped by prune_icache | |
4928 | * without having to perform any I/O. This is a very good thing, | |
4929 | * because *any* task may call prune_icache - even ones which | |
4930 | * have a transaction open against a different journal. | |
4931 | * | |
4932 | * Is this cheating? Not really. Sure, we haven't written the | |
4933 | * inode out, but prune_icache isn't a user-visible syncing function. | |
4934 | * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync) | |
4935 | * we start and wait on commits. | |
ac27a0ec | 4936 | */ |
617ba13b | 4937 | int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode) |
ac27a0ec | 4938 | { |
617ba13b | 4939 | struct ext4_iloc iloc; |
6dd4ee7c KS |
4940 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
4941 | static unsigned int mnt_count; | |
4942 | int err, ret; | |
ac27a0ec DK |
4943 | |
4944 | might_sleep(); | |
7ff9c073 | 4945 | trace_ext4_mark_inode_dirty(inode, _RET_IP_); |
617ba13b | 4946 | err = ext4_reserve_inode_write(handle, inode, &iloc); |
0390131b FM |
4947 | if (ext4_handle_valid(handle) && |
4948 | EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize && | |
19f5fb7a | 4949 | !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) { |
6dd4ee7c KS |
4950 | /* |
4951 | * We need extra buffer credits since we may write into EA block | |
4952 | * with this same handle. If journal_extend fails, then it will | |
4953 | * only result in a minor loss of functionality for that inode. | |
4954 | * If this is felt to be critical, then e2fsck should be run to | |
4955 | * force a large enough s_min_extra_isize. | |
4956 | */ | |
4957 | if ((jbd2_journal_extend(handle, | |
4958 | EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) { | |
4959 | ret = ext4_expand_extra_isize(inode, | |
4960 | sbi->s_want_extra_isize, | |
4961 | iloc, handle); | |
4962 | if (ret) { | |
19f5fb7a TT |
4963 | ext4_set_inode_state(inode, |
4964 | EXT4_STATE_NO_EXPAND); | |
c1bddad9 AK |
4965 | if (mnt_count != |
4966 | le16_to_cpu(sbi->s_es->s_mnt_count)) { | |
12062ddd | 4967 | ext4_warning(inode->i_sb, |
6dd4ee7c KS |
4968 | "Unable to expand inode %lu. Delete" |
4969 | " some EAs or run e2fsck.", | |
4970 | inode->i_ino); | |
c1bddad9 AK |
4971 | mnt_count = |
4972 | le16_to_cpu(sbi->s_es->s_mnt_count); | |
6dd4ee7c KS |
4973 | } |
4974 | } | |
4975 | } | |
4976 | } | |
ac27a0ec | 4977 | if (!err) |
617ba13b | 4978 | err = ext4_mark_iloc_dirty(handle, inode, &iloc); |
ac27a0ec DK |
4979 | return err; |
4980 | } | |
4981 | ||
4982 | /* | |
617ba13b | 4983 | * ext4_dirty_inode() is called from __mark_inode_dirty() |
ac27a0ec DK |
4984 | * |
4985 | * We're really interested in the case where a file is being extended. | |
4986 | * i_size has been changed by generic_commit_write() and we thus need | |
4987 | * to include the updated inode in the current transaction. | |
4988 | * | |
5dd4056d | 4989 | * Also, dquot_alloc_block() will always dirty the inode when blocks |
ac27a0ec DK |
4990 | * are allocated to the file. |
4991 | * | |
4992 | * If the inode is marked synchronous, we don't honour that here - doing | |
4993 | * so would cause a commit on atime updates, which we don't bother doing. | |
4994 | * We handle synchronous inodes at the highest possible level. | |
4995 | */ | |
aa385729 | 4996 | void ext4_dirty_inode(struct inode *inode, int flags) |
ac27a0ec | 4997 | { |
ac27a0ec DK |
4998 | handle_t *handle; |
4999 | ||
9924a92a | 5000 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 2); |
ac27a0ec DK |
5001 | if (IS_ERR(handle)) |
5002 | goto out; | |
f3dc272f | 5003 | |
f3dc272f CW |
5004 | ext4_mark_inode_dirty(handle, inode); |
5005 | ||
617ba13b | 5006 | ext4_journal_stop(handle); |
ac27a0ec DK |
5007 | out: |
5008 | return; | |
5009 | } | |
5010 | ||
5011 | #if 0 | |
5012 | /* | |
5013 | * Bind an inode's backing buffer_head into this transaction, to prevent | |
5014 | * it from being flushed to disk early. Unlike | |
617ba13b | 5015 | * ext4_reserve_inode_write, this leaves behind no bh reference and |
ac27a0ec DK |
5016 | * returns no iloc structure, so the caller needs to repeat the iloc |
5017 | * lookup to mark the inode dirty later. | |
5018 | */ | |
617ba13b | 5019 | static int ext4_pin_inode(handle_t *handle, struct inode *inode) |
ac27a0ec | 5020 | { |
617ba13b | 5021 | struct ext4_iloc iloc; |
ac27a0ec DK |
5022 | |
5023 | int err = 0; | |
5024 | if (handle) { | |
617ba13b | 5025 | err = ext4_get_inode_loc(inode, &iloc); |
ac27a0ec DK |
5026 | if (!err) { |
5027 | BUFFER_TRACE(iloc.bh, "get_write_access"); | |
dab291af | 5028 | err = jbd2_journal_get_write_access(handle, iloc.bh); |
ac27a0ec | 5029 | if (!err) |
0390131b | 5030 | err = ext4_handle_dirty_metadata(handle, |
73b50c1c | 5031 | NULL, |
0390131b | 5032 | iloc.bh); |
ac27a0ec DK |
5033 | brelse(iloc.bh); |
5034 | } | |
5035 | } | |
617ba13b | 5036 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
5037 | return err; |
5038 | } | |
5039 | #endif | |
5040 | ||
617ba13b | 5041 | int ext4_change_inode_journal_flag(struct inode *inode, int val) |
ac27a0ec DK |
5042 | { |
5043 | journal_t *journal; | |
5044 | handle_t *handle; | |
5045 | int err; | |
5046 | ||
5047 | /* | |
5048 | * We have to be very careful here: changing a data block's | |
5049 | * journaling status dynamically is dangerous. If we write a | |
5050 | * data block to the journal, change the status and then delete | |
5051 | * that block, we risk forgetting to revoke the old log record | |
5052 | * from the journal and so a subsequent replay can corrupt data. | |
5053 | * So, first we make sure that the journal is empty and that | |
5054 | * nobody is changing anything. | |
5055 | */ | |
5056 | ||
617ba13b | 5057 | journal = EXT4_JOURNAL(inode); |
0390131b FM |
5058 | if (!journal) |
5059 | return 0; | |
d699594d | 5060 | if (is_journal_aborted(journal)) |
ac27a0ec | 5061 | return -EROFS; |
2aff57b0 YY |
5062 | /* We have to allocate physical blocks for delalloc blocks |
5063 | * before flushing journal. otherwise delalloc blocks can not | |
5064 | * be allocated any more. even more truncate on delalloc blocks | |
5065 | * could trigger BUG by flushing delalloc blocks in journal. | |
5066 | * There is no delalloc block in non-journal data mode. | |
5067 | */ | |
5068 | if (val && test_opt(inode->i_sb, DELALLOC)) { | |
5069 | err = ext4_alloc_da_blocks(inode); | |
5070 | if (err < 0) | |
5071 | return err; | |
5072 | } | |
ac27a0ec | 5073 | |
17335dcc DM |
5074 | /* Wait for all existing dio workers */ |
5075 | ext4_inode_block_unlocked_dio(inode); | |
5076 | inode_dio_wait(inode); | |
5077 | ||
dab291af | 5078 | jbd2_journal_lock_updates(journal); |
ac27a0ec DK |
5079 | |
5080 | /* | |
5081 | * OK, there are no updates running now, and all cached data is | |
5082 | * synced to disk. We are now in a completely consistent state | |
5083 | * which doesn't have anything in the journal, and we know that | |
5084 | * no filesystem updates are running, so it is safe to modify | |
5085 | * the inode's in-core data-journaling state flag now. | |
5086 | */ | |
5087 | ||
5088 | if (val) | |
12e9b892 | 5089 | ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
5872ddaa YY |
5090 | else { |
5091 | jbd2_journal_flush(journal); | |
12e9b892 | 5092 | ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
5872ddaa | 5093 | } |
617ba13b | 5094 | ext4_set_aops(inode); |
ac27a0ec | 5095 | |
dab291af | 5096 | jbd2_journal_unlock_updates(journal); |
17335dcc | 5097 | ext4_inode_resume_unlocked_dio(inode); |
ac27a0ec DK |
5098 | |
5099 | /* Finally we can mark the inode as dirty. */ | |
5100 | ||
9924a92a | 5101 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 1); |
ac27a0ec DK |
5102 | if (IS_ERR(handle)) |
5103 | return PTR_ERR(handle); | |
5104 | ||
617ba13b | 5105 | err = ext4_mark_inode_dirty(handle, inode); |
0390131b | 5106 | ext4_handle_sync(handle); |
617ba13b MC |
5107 | ext4_journal_stop(handle); |
5108 | ext4_std_error(inode->i_sb, err); | |
ac27a0ec DK |
5109 | |
5110 | return err; | |
5111 | } | |
2e9ee850 AK |
5112 | |
5113 | static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh) | |
5114 | { | |
5115 | return !buffer_mapped(bh); | |
5116 | } | |
5117 | ||
c2ec175c | 5118 | int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
2e9ee850 | 5119 | { |
c2ec175c | 5120 | struct page *page = vmf->page; |
2e9ee850 AK |
5121 | loff_t size; |
5122 | unsigned long len; | |
9ea7df53 | 5123 | int ret; |
2e9ee850 | 5124 | struct file *file = vma->vm_file; |
496ad9aa | 5125 | struct inode *inode = file_inode(file); |
2e9ee850 | 5126 | struct address_space *mapping = inode->i_mapping; |
9ea7df53 JK |
5127 | handle_t *handle; |
5128 | get_block_t *get_block; | |
5129 | int retries = 0; | |
2e9ee850 | 5130 | |
8e8ad8a5 | 5131 | sb_start_pagefault(inode->i_sb); |
041bbb6d | 5132 | file_update_time(vma->vm_file); |
9ea7df53 JK |
5133 | /* Delalloc case is easy... */ |
5134 | if (test_opt(inode->i_sb, DELALLOC) && | |
5135 | !ext4_should_journal_data(inode) && | |
5136 | !ext4_nonda_switch(inode->i_sb)) { | |
5137 | do { | |
5138 | ret = __block_page_mkwrite(vma, vmf, | |
5139 | ext4_da_get_block_prep); | |
5140 | } while (ret == -ENOSPC && | |
5141 | ext4_should_retry_alloc(inode->i_sb, &retries)); | |
5142 | goto out_ret; | |
2e9ee850 | 5143 | } |
0e499890 DW |
5144 | |
5145 | lock_page(page); | |
9ea7df53 JK |
5146 | size = i_size_read(inode); |
5147 | /* Page got truncated from under us? */ | |
5148 | if (page->mapping != mapping || page_offset(page) > size) { | |
5149 | unlock_page(page); | |
5150 | ret = VM_FAULT_NOPAGE; | |
5151 | goto out; | |
0e499890 | 5152 | } |
2e9ee850 AK |
5153 | |
5154 | if (page->index == size >> PAGE_CACHE_SHIFT) | |
5155 | len = size & ~PAGE_CACHE_MASK; | |
5156 | else | |
5157 | len = PAGE_CACHE_SIZE; | |
a827eaff | 5158 | /* |
9ea7df53 JK |
5159 | * Return if we have all the buffers mapped. This avoids the need to do |
5160 | * journal_start/journal_stop which can block and take a long time | |
a827eaff | 5161 | */ |
2e9ee850 | 5162 | if (page_has_buffers(page)) { |
f19d5870 TM |
5163 | if (!ext4_walk_page_buffers(NULL, page_buffers(page), |
5164 | 0, len, NULL, | |
5165 | ext4_bh_unmapped)) { | |
9ea7df53 | 5166 | /* Wait so that we don't change page under IO */ |
1d1d1a76 | 5167 | wait_for_stable_page(page); |
9ea7df53 JK |
5168 | ret = VM_FAULT_LOCKED; |
5169 | goto out; | |
a827eaff | 5170 | } |
2e9ee850 | 5171 | } |
a827eaff | 5172 | unlock_page(page); |
9ea7df53 JK |
5173 | /* OK, we need to fill the hole... */ |
5174 | if (ext4_should_dioread_nolock(inode)) | |
5175 | get_block = ext4_get_block_write; | |
5176 | else | |
5177 | get_block = ext4_get_block; | |
5178 | retry_alloc: | |
9924a92a TT |
5179 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
5180 | ext4_writepage_trans_blocks(inode)); | |
9ea7df53 | 5181 | if (IS_ERR(handle)) { |
c2ec175c | 5182 | ret = VM_FAULT_SIGBUS; |
9ea7df53 JK |
5183 | goto out; |
5184 | } | |
5185 | ret = __block_page_mkwrite(vma, vmf, get_block); | |
5186 | if (!ret && ext4_should_journal_data(inode)) { | |
f19d5870 | 5187 | if (ext4_walk_page_buffers(handle, page_buffers(page), 0, |
9ea7df53 JK |
5188 | PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) { |
5189 | unlock_page(page); | |
5190 | ret = VM_FAULT_SIGBUS; | |
fcbb5515 | 5191 | ext4_journal_stop(handle); |
9ea7df53 JK |
5192 | goto out; |
5193 | } | |
5194 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); | |
5195 | } | |
5196 | ext4_journal_stop(handle); | |
5197 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) | |
5198 | goto retry_alloc; | |
5199 | out_ret: | |
5200 | ret = block_page_mkwrite_return(ret); | |
5201 | out: | |
8e8ad8a5 | 5202 | sb_end_pagefault(inode->i_sb); |
2e9ee850 AK |
5203 | return ret; |
5204 | } |