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> |
0a0ae7b3 | 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 | 48 | #include <trace/events/ext4.h> |
6fa3eb70 | 49 | #include <linux/blkdev.h> |
9bffad1e | 50 | |
a1d6cc56 AK |
51 | #define MPAGE_DA_EXTENT_TAIL 0x01 |
52 | ||
814525f4 DW |
53 | static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw, |
54 | struct ext4_inode_info *ei) | |
55 | { | |
56 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
57 | __u16 csum_lo; | |
58 | __u16 csum_hi = 0; | |
59 | __u32 csum; | |
60 | ||
171a7f21 | 61 | csum_lo = le16_to_cpu(raw->i_checksum_lo); |
814525f4 DW |
62 | raw->i_checksum_lo = 0; |
63 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && | |
64 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) { | |
171a7f21 | 65 | csum_hi = le16_to_cpu(raw->i_checksum_hi); |
814525f4 DW |
66 | raw->i_checksum_hi = 0; |
67 | } | |
68 | ||
69 | csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw, | |
70 | EXT4_INODE_SIZE(inode->i_sb)); | |
71 | ||
171a7f21 | 72 | raw->i_checksum_lo = cpu_to_le16(csum_lo); |
814525f4 DW |
73 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && |
74 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) | |
171a7f21 | 75 | raw->i_checksum_hi = cpu_to_le16(csum_hi); |
814525f4 DW |
76 | |
77 | return csum; | |
78 | } | |
79 | ||
80 | static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw, | |
81 | struct ext4_inode_info *ei) | |
82 | { | |
83 | __u32 provided, calculated; | |
84 | ||
85 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != | |
86 | cpu_to_le32(EXT4_OS_LINUX) || | |
87 | !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, | |
88 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) | |
89 | return 1; | |
90 | ||
91 | provided = le16_to_cpu(raw->i_checksum_lo); | |
92 | calculated = ext4_inode_csum(inode, raw, ei); | |
93 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && | |
94 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) | |
95 | provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16; | |
96 | else | |
97 | calculated &= 0xFFFF; | |
98 | ||
99 | return provided == calculated; | |
100 | } | |
101 | ||
102 | static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw, | |
103 | struct ext4_inode_info *ei) | |
104 | { | |
105 | __u32 csum; | |
106 | ||
107 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != | |
108 | cpu_to_le32(EXT4_OS_LINUX) || | |
109 | !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, | |
110 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) | |
111 | return; | |
112 | ||
113 | csum = ext4_inode_csum(inode, raw, ei); | |
114 | raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF); | |
115 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && | |
116 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) | |
117 | raw->i_checksum_hi = cpu_to_le16(csum >> 16); | |
118 | } | |
119 | ||
678aaf48 JK |
120 | static inline int ext4_begin_ordered_truncate(struct inode *inode, |
121 | loff_t new_size) | |
122 | { | |
7ff9c073 | 123 | trace_ext4_begin_ordered_truncate(inode, new_size); |
8aefcd55 TT |
124 | /* |
125 | * If jinode is zero, then we never opened the file for | |
126 | * writing, so there's no need to call | |
127 | * jbd2_journal_begin_ordered_truncate() since there's no | |
128 | * outstanding writes we need to flush. | |
129 | */ | |
130 | if (!EXT4_I(inode)->jinode) | |
131 | return 0; | |
132 | return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode), | |
133 | EXT4_I(inode)->jinode, | |
134 | new_size); | |
678aaf48 JK |
135 | } |
136 | ||
64769240 | 137 | static void ext4_invalidatepage(struct page *page, unsigned long offset); |
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); | |
5f163cc7 ES |
140 | static int ext4_discard_partial_page_buffers_no_lock(handle_t *handle, |
141 | struct inode *inode, struct page *page, loff_t from, | |
142 | loff_t length, int flags); | |
64769240 | 143 | |
ac27a0ec DK |
144 | /* |
145 | * Test whether an inode is a fast symlink. | |
146 | */ | |
617ba13b | 147 | static int ext4_inode_is_fast_symlink(struct inode *inode) |
ac27a0ec | 148 | { |
617ba13b | 149 | int ea_blocks = EXT4_I(inode)->i_file_acl ? |
ac27a0ec DK |
150 | (inode->i_sb->s_blocksize >> 9) : 0; |
151 | ||
152 | return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0); | |
153 | } | |
154 | ||
ac27a0ec DK |
155 | /* |
156 | * Restart the transaction associated with *handle. This does a commit, | |
157 | * so before we call here everything must be consistently dirtied against | |
158 | * this transaction. | |
159 | */ | |
fa5d1113 | 160 | int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode, |
487caeef | 161 | int nblocks) |
ac27a0ec | 162 | { |
487caeef JK |
163 | int ret; |
164 | ||
165 | /* | |
e35fd660 | 166 | * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this |
487caeef JK |
167 | * moment, get_block can be called only for blocks inside i_size since |
168 | * page cache has been already dropped and writes are blocked by | |
169 | * i_mutex. So we can safely drop the i_data_sem here. | |
170 | */ | |
0390131b | 171 | BUG_ON(EXT4_JOURNAL(inode) == NULL); |
ac27a0ec | 172 | jbd_debug(2, "restarting handle %p\n", handle); |
487caeef | 173 | up_write(&EXT4_I(inode)->i_data_sem); |
8e8eaabe | 174 | ret = ext4_journal_restart(handle, nblocks); |
487caeef | 175 | down_write(&EXT4_I(inode)->i_data_sem); |
fa5d1113 | 176 | ext4_discard_preallocations(inode); |
487caeef JK |
177 | |
178 | return ret; | |
ac27a0ec DK |
179 | } |
180 | ||
181 | /* | |
182 | * Called at the last iput() if i_nlink is zero. | |
183 | */ | |
0930fcc1 | 184 | void ext4_evict_inode(struct inode *inode) |
ac27a0ec DK |
185 | { |
186 | handle_t *handle; | |
bc965ab3 | 187 | int err; |
ac27a0ec | 188 | |
7ff9c073 | 189 | trace_ext4_evict_inode(inode); |
2581fdc8 | 190 | |
0930fcc1 | 191 | if (inode->i_nlink) { |
2d859db3 JK |
192 | /* |
193 | * When journalling data dirty buffers are tracked only in the | |
194 | * journal. So although mm thinks everything is clean and | |
195 | * ready for reaping the inode might still have some pages to | |
196 | * write in the running transaction or waiting to be | |
197 | * checkpointed. Thus calling jbd2_journal_invalidatepage() | |
198 | * (via truncate_inode_pages()) to discard these buffers can | |
199 | * cause data loss. Also even if we did not discard these | |
200 | * buffers, we would have no way to find them after the inode | |
201 | * is reaped and thus user could see stale data if he tries to | |
202 | * read them before the transaction is checkpointed. So be | |
203 | * careful and force everything to disk here... We use | |
204 | * ei->i_datasync_tid to store the newest transaction | |
205 | * containing inode's data. | |
206 | * | |
207 | * Note that directories do not have this problem because they | |
208 | * don't use page cache. | |
209 | */ | |
210 | if (ext4_should_journal_data(inode) && | |
2b405bfa TT |
211 | (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) && |
212 | inode->i_ino != EXT4_JOURNAL_INO) { | |
2d859db3 JK |
213 | journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; |
214 | tid_t commit_tid = EXT4_I(inode)->i_datasync_tid; | |
215 | ||
d76a3a77 | 216 | jbd2_complete_transaction(journal, commit_tid); |
2d859db3 JK |
217 | filemap_write_and_wait(&inode->i_data); |
218 | } | |
0930fcc1 | 219 | truncate_inode_pages(&inode->i_data, 0); |
1ada47d9 | 220 | ext4_ioend_shutdown(inode); |
0930fcc1 AV |
221 | goto no_delete; |
222 | } | |
223 | ||
907f4554 | 224 | if (!is_bad_inode(inode)) |
871a2931 | 225 | dquot_initialize(inode); |
907f4554 | 226 | |
678aaf48 JK |
227 | if (ext4_should_order_data(inode)) |
228 | ext4_begin_ordered_truncate(inode, 0); | |
ac27a0ec | 229 | truncate_inode_pages(&inode->i_data, 0); |
1ada47d9 | 230 | ext4_ioend_shutdown(inode); |
ac27a0ec DK |
231 | |
232 | if (is_bad_inode(inode)) | |
233 | goto no_delete; | |
234 | ||
8e8ad8a5 JK |
235 | /* |
236 | * Protect us against freezing - iput() caller didn't have to have any | |
237 | * protection against it | |
238 | */ | |
239 | sb_start_intwrite(inode->i_sb); | |
9924a92a TT |
240 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, |
241 | ext4_blocks_for_truncate(inode)+3); | |
ac27a0ec | 242 | if (IS_ERR(handle)) { |
bc965ab3 | 243 | ext4_std_error(inode->i_sb, PTR_ERR(handle)); |
ac27a0ec DK |
244 | /* |
245 | * If we're going to skip the normal cleanup, we still need to | |
246 | * make sure that the in-core orphan linked list is properly | |
247 | * cleaned up. | |
248 | */ | |
617ba13b | 249 | ext4_orphan_del(NULL, inode); |
8e8ad8a5 | 250 | sb_end_intwrite(inode->i_sb); |
ac27a0ec DK |
251 | goto no_delete; |
252 | } | |
253 | ||
254 | if (IS_SYNC(inode)) | |
0390131b | 255 | ext4_handle_sync(handle); |
ac27a0ec | 256 | inode->i_size = 0; |
bc965ab3 TT |
257 | err = ext4_mark_inode_dirty(handle, inode); |
258 | if (err) { | |
12062ddd | 259 | ext4_warning(inode->i_sb, |
bc965ab3 TT |
260 | "couldn't mark inode dirty (err %d)", err); |
261 | goto stop_handle; | |
262 | } | |
ac27a0ec | 263 | if (inode->i_blocks) |
617ba13b | 264 | ext4_truncate(inode); |
bc965ab3 TT |
265 | |
266 | /* | |
267 | * ext4_ext_truncate() doesn't reserve any slop when it | |
268 | * restarts journal transactions; therefore there may not be | |
269 | * enough credits left in the handle to remove the inode from | |
270 | * the orphan list and set the dtime field. | |
271 | */ | |
0390131b | 272 | if (!ext4_handle_has_enough_credits(handle, 3)) { |
bc965ab3 TT |
273 | err = ext4_journal_extend(handle, 3); |
274 | if (err > 0) | |
275 | err = ext4_journal_restart(handle, 3); | |
276 | if (err != 0) { | |
12062ddd | 277 | ext4_warning(inode->i_sb, |
bc965ab3 TT |
278 | "couldn't extend journal (err %d)", err); |
279 | stop_handle: | |
280 | ext4_journal_stop(handle); | |
45388219 | 281 | ext4_orphan_del(NULL, inode); |
8e8ad8a5 | 282 | sb_end_intwrite(inode->i_sb); |
bc965ab3 TT |
283 | goto no_delete; |
284 | } | |
285 | } | |
286 | ||
ac27a0ec | 287 | /* |
617ba13b | 288 | * Kill off the orphan record which ext4_truncate created. |
ac27a0ec | 289 | * AKPM: I think this can be inside the above `if'. |
617ba13b | 290 | * Note that ext4_orphan_del() has to be able to cope with the |
ac27a0ec | 291 | * deletion of a non-existent orphan - this is because we don't |
617ba13b | 292 | * know if ext4_truncate() actually created an orphan record. |
ac27a0ec DK |
293 | * (Well, we could do this if we need to, but heck - it works) |
294 | */ | |
617ba13b MC |
295 | ext4_orphan_del(handle, inode); |
296 | EXT4_I(inode)->i_dtime = get_seconds(); | |
ac27a0ec DK |
297 | |
298 | /* | |
299 | * One subtle ordering requirement: if anything has gone wrong | |
300 | * (transaction abort, IO errors, whatever), then we can still | |
301 | * do these next steps (the fs will already have been marked as | |
302 | * having errors), but we can't free the inode if the mark_dirty | |
303 | * fails. | |
304 | */ | |
617ba13b | 305 | if (ext4_mark_inode_dirty(handle, inode)) |
ac27a0ec | 306 | /* If that failed, just do the required in-core inode clear. */ |
0930fcc1 | 307 | ext4_clear_inode(inode); |
ac27a0ec | 308 | else |
617ba13b MC |
309 | ext4_free_inode(handle, inode); |
310 | ext4_journal_stop(handle); | |
8e8ad8a5 | 311 | sb_end_intwrite(inode->i_sb); |
ac27a0ec DK |
312 | return; |
313 | no_delete: | |
0930fcc1 | 314 | ext4_clear_inode(inode); /* We must guarantee clearing of inode... */ |
ac27a0ec DK |
315 | } |
316 | ||
a9e7f447 DM |
317 | #ifdef CONFIG_QUOTA |
318 | qsize_t *ext4_get_reserved_space(struct inode *inode) | |
60e58e0f | 319 | { |
a9e7f447 | 320 | return &EXT4_I(inode)->i_reserved_quota; |
60e58e0f | 321 | } |
a9e7f447 | 322 | #endif |
9d0be502 | 323 | |
12219aea AK |
324 | /* |
325 | * Calculate the number of metadata blocks need to reserve | |
9d0be502 | 326 | * to allocate a block located at @lblock |
12219aea | 327 | */ |
01f49d0b | 328 | static int ext4_calc_metadata_amount(struct inode *inode, ext4_lblk_t lblock) |
12219aea | 329 | { |
12e9b892 | 330 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
9d0be502 | 331 | return ext4_ext_calc_metadata_amount(inode, lblock); |
12219aea | 332 | |
8bb2b247 | 333 | return ext4_ind_calc_metadata_amount(inode, lblock); |
12219aea AK |
334 | } |
335 | ||
0637c6f4 TT |
336 | /* |
337 | * Called with i_data_sem down, which is important since we can call | |
338 | * ext4_discard_preallocations() from here. | |
339 | */ | |
5f634d06 AK |
340 | void ext4_da_update_reserve_space(struct inode *inode, |
341 | int used, int quota_claim) | |
12219aea AK |
342 | { |
343 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0637c6f4 | 344 | struct ext4_inode_info *ei = EXT4_I(inode); |
0637c6f4 TT |
345 | |
346 | spin_lock(&ei->i_block_reservation_lock); | |
d8990240 | 347 | trace_ext4_da_update_reserve_space(inode, used, quota_claim); |
0637c6f4 | 348 | if (unlikely(used > ei->i_reserved_data_blocks)) { |
8de5c325 | 349 | ext4_warning(inode->i_sb, "%s: ino %lu, used %d " |
1084f252 | 350 | "with only %d reserved data blocks", |
0637c6f4 TT |
351 | __func__, inode->i_ino, used, |
352 | ei->i_reserved_data_blocks); | |
353 | WARN_ON(1); | |
354 | used = ei->i_reserved_data_blocks; | |
355 | } | |
12219aea | 356 | |
97795d2a | 357 | if (unlikely(ei->i_allocated_meta_blocks > ei->i_reserved_meta_blocks)) { |
01a523eb TT |
358 | ext4_warning(inode->i_sb, "ino %lu, allocated %d " |
359 | "with only %d reserved metadata blocks " | |
360 | "(releasing %d blocks with reserved %d data blocks)", | |
361 | inode->i_ino, ei->i_allocated_meta_blocks, | |
362 | ei->i_reserved_meta_blocks, used, | |
363 | ei->i_reserved_data_blocks); | |
97795d2a BF |
364 | WARN_ON(1); |
365 | ei->i_allocated_meta_blocks = ei->i_reserved_meta_blocks; | |
366 | } | |
367 | ||
0637c6f4 TT |
368 | /* Update per-inode reservations */ |
369 | ei->i_reserved_data_blocks -= used; | |
0637c6f4 | 370 | ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks; |
57042651 | 371 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, |
72b8ab9d | 372 | used + ei->i_allocated_meta_blocks); |
0637c6f4 | 373 | ei->i_allocated_meta_blocks = 0; |
6bc6e63f | 374 | |
0637c6f4 TT |
375 | if (ei->i_reserved_data_blocks == 0) { |
376 | /* | |
377 | * We can release all of the reserved metadata blocks | |
378 | * only when we have written all of the delayed | |
379 | * allocation blocks. | |
380 | */ | |
57042651 | 381 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, |
72b8ab9d | 382 | ei->i_reserved_meta_blocks); |
ee5f4d9c | 383 | ei->i_reserved_meta_blocks = 0; |
9d0be502 | 384 | ei->i_da_metadata_calc_len = 0; |
6bc6e63f | 385 | } |
12219aea | 386 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
60e58e0f | 387 | |
72b8ab9d ES |
388 | /* Update quota subsystem for data blocks */ |
389 | if (quota_claim) | |
7b415bf6 | 390 | dquot_claim_block(inode, EXT4_C2B(sbi, used)); |
72b8ab9d | 391 | else { |
5f634d06 AK |
392 | /* |
393 | * We did fallocate with an offset that is already delayed | |
394 | * allocated. So on delayed allocated writeback we should | |
72b8ab9d | 395 | * not re-claim the quota for fallocated blocks. |
5f634d06 | 396 | */ |
7b415bf6 | 397 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, used)); |
5f634d06 | 398 | } |
d6014301 AK |
399 | |
400 | /* | |
401 | * If we have done all the pending block allocations and if | |
402 | * there aren't any writers on the inode, we can discard the | |
403 | * inode's preallocations. | |
404 | */ | |
0637c6f4 TT |
405 | if ((ei->i_reserved_data_blocks == 0) && |
406 | (atomic_read(&inode->i_writecount) == 0)) | |
d6014301 | 407 | ext4_discard_preallocations(inode); |
12219aea AK |
408 | } |
409 | ||
e29136f8 | 410 | static int __check_block_validity(struct inode *inode, const char *func, |
c398eda0 TT |
411 | unsigned int line, |
412 | struct ext4_map_blocks *map) | |
6fd058f7 | 413 | { |
24676da4 TT |
414 | if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk, |
415 | map->m_len)) { | |
c398eda0 TT |
416 | ext4_error_inode(inode, func, line, map->m_pblk, |
417 | "lblock %lu mapped to illegal pblock " | |
418 | "(length %d)", (unsigned long) map->m_lblk, | |
419 | map->m_len); | |
6fd058f7 TT |
420 | return -EIO; |
421 | } | |
422 | return 0; | |
423 | } | |
424 | ||
e29136f8 | 425 | #define check_block_validity(inode, map) \ |
c398eda0 | 426 | __check_block_validity((inode), __func__, __LINE__, (map)) |
e29136f8 | 427 | |
55138e0b | 428 | /* |
1f94533d TT |
429 | * Return the number of contiguous dirty pages in a given inode |
430 | * starting at page frame idx. | |
55138e0b TT |
431 | */ |
432 | static pgoff_t ext4_num_dirty_pages(struct inode *inode, pgoff_t idx, | |
433 | unsigned int max_pages) | |
434 | { | |
435 | struct address_space *mapping = inode->i_mapping; | |
436 | pgoff_t index; | |
437 | struct pagevec pvec; | |
438 | pgoff_t num = 0; | |
439 | int i, nr_pages, done = 0; | |
440 | ||
441 | if (max_pages == 0) | |
442 | return 0; | |
443 | pagevec_init(&pvec, 0); | |
444 | while (!done) { | |
445 | index = idx; | |
446 | nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, | |
447 | PAGECACHE_TAG_DIRTY, | |
448 | (pgoff_t)PAGEVEC_SIZE); | |
449 | if (nr_pages == 0) | |
450 | break; | |
451 | for (i = 0; i < nr_pages; i++) { | |
452 | struct page *page = pvec.pages[i]; | |
453 | struct buffer_head *bh, *head; | |
454 | ||
455 | lock_page(page); | |
456 | if (unlikely(page->mapping != mapping) || | |
457 | !PageDirty(page) || | |
458 | PageWriteback(page) || | |
459 | page->index != idx) { | |
460 | done = 1; | |
461 | unlock_page(page); | |
462 | break; | |
463 | } | |
1f94533d TT |
464 | if (page_has_buffers(page)) { |
465 | bh = head = page_buffers(page); | |
466 | do { | |
467 | if (!buffer_delay(bh) && | |
468 | !buffer_unwritten(bh)) | |
469 | done = 1; | |
470 | bh = bh->b_this_page; | |
471 | } while (!done && (bh != head)); | |
472 | } | |
55138e0b TT |
473 | unlock_page(page); |
474 | if (done) | |
475 | break; | |
476 | idx++; | |
477 | num++; | |
659c6009 ES |
478 | if (num >= max_pages) { |
479 | done = 1; | |
55138e0b | 480 | break; |
659c6009 | 481 | } |
55138e0b TT |
482 | } |
483 | pagevec_release(&pvec); | |
484 | } | |
485 | return num; | |
486 | } | |
487 | ||
921f266b DM |
488 | #ifdef ES_AGGRESSIVE_TEST |
489 | static void ext4_map_blocks_es_recheck(handle_t *handle, | |
490 | struct inode *inode, | |
491 | struct ext4_map_blocks *es_map, | |
492 | struct ext4_map_blocks *map, | |
493 | int flags) | |
494 | { | |
495 | int retval; | |
496 | ||
497 | map->m_flags = 0; | |
498 | /* | |
499 | * There is a race window that the result is not the same. | |
500 | * e.g. xfstests #223 when dioread_nolock enables. The reason | |
501 | * is that we lookup a block mapping in extent status tree with | |
502 | * out taking i_data_sem. So at the time the unwritten extent | |
503 | * could be converted. | |
504 | */ | |
505 | if (!(flags & EXT4_GET_BLOCKS_NO_LOCK)) | |
506 | down_read((&EXT4_I(inode)->i_data_sem)); | |
507 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { | |
508 | retval = ext4_ext_map_blocks(handle, inode, map, flags & | |
509 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
510 | } else { | |
511 | retval = ext4_ind_map_blocks(handle, inode, map, flags & | |
512 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
513 | } | |
514 | if (!(flags & EXT4_GET_BLOCKS_NO_LOCK)) | |
515 | up_read((&EXT4_I(inode)->i_data_sem)); | |
516 | /* | |
517 | * Clear EXT4_MAP_FROM_CLUSTER and EXT4_MAP_BOUNDARY flag | |
518 | * because it shouldn't be marked in es_map->m_flags. | |
519 | */ | |
520 | map->m_flags &= ~(EXT4_MAP_FROM_CLUSTER | EXT4_MAP_BOUNDARY); | |
521 | ||
522 | /* | |
523 | * We don't check m_len because extent will be collpased in status | |
524 | * tree. So the m_len might not equal. | |
525 | */ | |
526 | if (es_map->m_lblk != map->m_lblk || | |
527 | es_map->m_flags != map->m_flags || | |
528 | es_map->m_pblk != map->m_pblk) { | |
529 | printk("ES cache assertation failed for inode: %lu " | |
530 | "es_cached ex [%d/%d/%llu/%x] != " | |
531 | "found ex [%d/%d/%llu/%x] retval %d flags %x\n", | |
532 | inode->i_ino, es_map->m_lblk, es_map->m_len, | |
533 | es_map->m_pblk, es_map->m_flags, map->m_lblk, | |
534 | map->m_len, map->m_pblk, map->m_flags, | |
535 | retval, flags); | |
536 | } | |
537 | } | |
538 | #endif /* ES_AGGRESSIVE_TEST */ | |
539 | ||
f5ab0d1f | 540 | /* |
e35fd660 | 541 | * The ext4_map_blocks() function tries to look up the requested blocks, |
2b2d6d01 | 542 | * and returns if the blocks are already mapped. |
f5ab0d1f | 543 | * |
f5ab0d1f MC |
544 | * Otherwise it takes the write lock of the i_data_sem and allocate blocks |
545 | * and store the allocated blocks in the result buffer head and mark it | |
546 | * mapped. | |
547 | * | |
e35fd660 TT |
548 | * If file type is extents based, it will call ext4_ext_map_blocks(), |
549 | * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping | |
f5ab0d1f MC |
550 | * based files |
551 | * | |
552 | * On success, it returns the number of blocks being mapped or allocate. | |
553 | * if create==0 and the blocks are pre-allocated and uninitialized block, | |
554 | * the result buffer head is unmapped. If the create ==1, it will make sure | |
555 | * the buffer head is mapped. | |
556 | * | |
557 | * It returns 0 if plain look up failed (blocks have not been allocated), in | |
df3ab170 | 558 | * that case, buffer head is unmapped |
f5ab0d1f MC |
559 | * |
560 | * It returns the error in case of allocation failure. | |
561 | */ | |
e35fd660 TT |
562 | int ext4_map_blocks(handle_t *handle, struct inode *inode, |
563 | struct ext4_map_blocks *map, int flags) | |
0e855ac8 | 564 | { |
d100eef2 | 565 | struct extent_status es; |
0e855ac8 | 566 | int retval; |
921f266b DM |
567 | #ifdef ES_AGGRESSIVE_TEST |
568 | struct ext4_map_blocks orig_map; | |
569 | ||
570 | memcpy(&orig_map, map, sizeof(*map)); | |
571 | #endif | |
f5ab0d1f | 572 | |
e35fd660 TT |
573 | map->m_flags = 0; |
574 | ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u," | |
575 | "logical block %lu\n", inode->i_ino, flags, map->m_len, | |
576 | (unsigned long) map->m_lblk); | |
d100eef2 ZL |
577 | |
578 | /* Lookup extent status tree firstly */ | |
579 | if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) { | |
580 | if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) { | |
581 | map->m_pblk = ext4_es_pblock(&es) + | |
582 | map->m_lblk - es.es_lblk; | |
583 | map->m_flags |= ext4_es_is_written(&es) ? | |
584 | EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN; | |
585 | retval = es.es_len - (map->m_lblk - es.es_lblk); | |
586 | if (retval > map->m_len) | |
587 | retval = map->m_len; | |
588 | map->m_len = retval; | |
589 | } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) { | |
590 | retval = 0; | |
591 | } else { | |
592 | BUG_ON(1); | |
593 | } | |
921f266b DM |
594 | #ifdef ES_AGGRESSIVE_TEST |
595 | ext4_map_blocks_es_recheck(handle, inode, map, | |
596 | &orig_map, flags); | |
597 | #endif | |
d100eef2 ZL |
598 | goto found; |
599 | } | |
600 | ||
4df3d265 | 601 | /* |
b920c755 TT |
602 | * Try to see if we can get the block without requesting a new |
603 | * file system block. | |
4df3d265 | 604 | */ |
729f52c6 ZL |
605 | if (!(flags & EXT4_GET_BLOCKS_NO_LOCK)) |
606 | down_read((&EXT4_I(inode)->i_data_sem)); | |
12e9b892 | 607 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
a4e5d88b DM |
608 | retval = ext4_ext_map_blocks(handle, inode, map, flags & |
609 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
0e855ac8 | 610 | } else { |
a4e5d88b DM |
611 | retval = ext4_ind_map_blocks(handle, inode, map, flags & |
612 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
0e855ac8 | 613 | } |
f7fec032 ZL |
614 | if (retval > 0) { |
615 | int ret; | |
616 | unsigned long long status; | |
617 | ||
921f266b DM |
618 | #ifdef ES_AGGRESSIVE_TEST |
619 | if (retval != map->m_len) { | |
620 | printk("ES len assertation failed for inode: %lu " | |
621 | "retval %d != map->m_len %d " | |
622 | "in %s (lookup)\n", inode->i_ino, retval, | |
623 | map->m_len, __func__); | |
624 | } | |
625 | #endif | |
626 | ||
f7fec032 ZL |
627 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
628 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; | |
629 | if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) && | |
630 | ext4_find_delalloc_range(inode, map->m_lblk, | |
631 | map->m_lblk + map->m_len - 1)) | |
632 | status |= EXTENT_STATUS_DELAYED; | |
633 | ret = ext4_es_insert_extent(inode, map->m_lblk, | |
634 | map->m_len, map->m_pblk, status); | |
635 | if (ret < 0) | |
636 | retval = ret; | |
637 | } | |
729f52c6 ZL |
638 | if (!(flags & EXT4_GET_BLOCKS_NO_LOCK)) |
639 | up_read((&EXT4_I(inode)->i_data_sem)); | |
f5ab0d1f | 640 | |
d100eef2 | 641 | found: |
e35fd660 | 642 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
f7fec032 | 643 | int ret = check_block_validity(inode, map); |
6fd058f7 TT |
644 | if (ret != 0) |
645 | return ret; | |
646 | } | |
647 | ||
f5ab0d1f | 648 | /* If it is only a block(s) look up */ |
c2177057 | 649 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) |
f5ab0d1f MC |
650 | return retval; |
651 | ||
652 | /* | |
653 | * Returns if the blocks have already allocated | |
654 | * | |
655 | * Note that if blocks have been preallocated | |
df3ab170 | 656 | * ext4_ext_get_block() returns the create = 0 |
f5ab0d1f MC |
657 | * with buffer head unmapped. |
658 | */ | |
e35fd660 | 659 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) |
4df3d265 AK |
660 | return retval; |
661 | ||
2a8964d6 | 662 | /* |
a25a4e1a ZL |
663 | * Here we clear m_flags because after allocating an new extent, |
664 | * it will be set again. | |
2a8964d6 | 665 | */ |
a25a4e1a | 666 | map->m_flags &= ~EXT4_MAP_FLAGS; |
2a8964d6 | 667 | |
4df3d265 | 668 | /* |
f5ab0d1f MC |
669 | * New blocks allocate and/or writing to uninitialized extent |
670 | * will possibly result in updating i_data, so we take | |
671 | * the write lock of i_data_sem, and call get_blocks() | |
672 | * with create == 1 flag. | |
4df3d265 AK |
673 | */ |
674 | down_write((&EXT4_I(inode)->i_data_sem)); | |
d2a17637 MC |
675 | |
676 | /* | |
677 | * if the caller is from delayed allocation writeout path | |
678 | * we have already reserved fs blocks for allocation | |
679 | * let the underlying get_block() function know to | |
680 | * avoid double accounting | |
681 | */ | |
c2177057 | 682 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) |
f2321097 | 683 | ext4_set_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED); |
4df3d265 AK |
684 | /* |
685 | * We need to check for EXT4 here because migrate | |
686 | * could have changed the inode type in between | |
687 | */ | |
12e9b892 | 688 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
e35fd660 | 689 | retval = ext4_ext_map_blocks(handle, inode, map, flags); |
0e855ac8 | 690 | } else { |
e35fd660 | 691 | retval = ext4_ind_map_blocks(handle, inode, map, flags); |
267e4db9 | 692 | |
e35fd660 | 693 | if (retval > 0 && map->m_flags & EXT4_MAP_NEW) { |
267e4db9 AK |
694 | /* |
695 | * We allocated new blocks which will result in | |
696 | * i_data's format changing. Force the migrate | |
697 | * to fail by clearing migrate flags | |
698 | */ | |
19f5fb7a | 699 | ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE); |
267e4db9 | 700 | } |
d2a17637 | 701 | |
5f634d06 AK |
702 | /* |
703 | * Update reserved blocks/metadata blocks after successful | |
704 | * block allocation which had been deferred till now. We don't | |
705 | * support fallocate for non extent files. So we can update | |
706 | * reserve space here. | |
707 | */ | |
708 | if ((retval > 0) && | |
1296cc85 | 709 | (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)) |
5f634d06 AK |
710 | ext4_da_update_reserve_space(inode, retval, 1); |
711 | } | |
f7fec032 | 712 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) |
f2321097 | 713 | ext4_clear_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED); |
2ac3b6e0 | 714 | |
f7fec032 ZL |
715 | if (retval > 0) { |
716 | int ret; | |
717 | unsigned long long status; | |
718 | ||
921f266b DM |
719 | #ifdef ES_AGGRESSIVE_TEST |
720 | if (retval != map->m_len) { | |
721 | printk("ES len assertation failed for inode: %lu " | |
722 | "retval %d != map->m_len %d " | |
723 | "in %s (allocation)\n", inode->i_ino, retval, | |
724 | map->m_len, __func__); | |
725 | } | |
726 | #endif | |
727 | ||
adb23551 ZL |
728 | /* |
729 | * If the extent has been zeroed out, we don't need to update | |
730 | * extent status tree. | |
731 | */ | |
732 | if ((flags & EXT4_GET_BLOCKS_PRE_IO) && | |
733 | ext4_es_lookup_extent(inode, map->m_lblk, &es)) { | |
734 | if (ext4_es_is_written(&es)) | |
735 | goto has_zeroout; | |
736 | } | |
f7fec032 ZL |
737 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
738 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; | |
739 | if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) && | |
740 | ext4_find_delalloc_range(inode, map->m_lblk, | |
741 | map->m_lblk + map->m_len - 1)) | |
742 | status |= EXTENT_STATUS_DELAYED; | |
743 | ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, | |
744 | map->m_pblk, status); | |
745 | if (ret < 0) | |
746 | retval = ret; | |
5356f261 AK |
747 | } |
748 | ||
adb23551 | 749 | has_zeroout: |
4df3d265 | 750 | up_write((&EXT4_I(inode)->i_data_sem)); |
e35fd660 | 751 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
e29136f8 | 752 | int ret = check_block_validity(inode, map); |
6fd058f7 TT |
753 | if (ret != 0) |
754 | return ret; | |
755 | } | |
0e855ac8 AK |
756 | return retval; |
757 | } | |
758 | ||
f3bd1f3f MC |
759 | /* Maximum number of blocks we map for direct IO at once. */ |
760 | #define DIO_MAX_BLOCKS 4096 | |
761 | ||
2ed88685 TT |
762 | static int _ext4_get_block(struct inode *inode, sector_t iblock, |
763 | struct buffer_head *bh, int flags) | |
ac27a0ec | 764 | { |
3e4fdaf8 | 765 | handle_t *handle = ext4_journal_current_handle(); |
2ed88685 | 766 | struct ext4_map_blocks map; |
7fb5409d | 767 | int ret = 0, started = 0; |
f3bd1f3f | 768 | int dio_credits; |
ac27a0ec | 769 | |
46c7f254 TM |
770 | if (ext4_has_inline_data(inode)) |
771 | return -ERANGE; | |
772 | ||
2ed88685 TT |
773 | map.m_lblk = iblock; |
774 | map.m_len = bh->b_size >> inode->i_blkbits; | |
775 | ||
8b0f165f | 776 | if (flags && !(flags & EXT4_GET_BLOCKS_NO_LOCK) && !handle) { |
7fb5409d | 777 | /* Direct IO write... */ |
2ed88685 TT |
778 | if (map.m_len > DIO_MAX_BLOCKS) |
779 | map.m_len = DIO_MAX_BLOCKS; | |
780 | dio_credits = ext4_chunk_trans_blocks(inode, map.m_len); | |
9924a92a TT |
781 | handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, |
782 | dio_credits); | |
7fb5409d | 783 | if (IS_ERR(handle)) { |
ac27a0ec | 784 | ret = PTR_ERR(handle); |
2ed88685 | 785 | return ret; |
ac27a0ec | 786 | } |
7fb5409d | 787 | started = 1; |
ac27a0ec DK |
788 | } |
789 | ||
2ed88685 | 790 | ret = ext4_map_blocks(handle, inode, &map, flags); |
7fb5409d | 791 | if (ret > 0) { |
2ed88685 TT |
792 | map_bh(bh, inode->i_sb, map.m_pblk); |
793 | bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags; | |
794 | bh->b_size = inode->i_sb->s_blocksize * map.m_len; | |
7fb5409d | 795 | ret = 0; |
ac27a0ec | 796 | } |
7fb5409d JK |
797 | if (started) |
798 | ext4_journal_stop(handle); | |
ac27a0ec DK |
799 | return ret; |
800 | } | |
801 | ||
2ed88685 TT |
802 | int ext4_get_block(struct inode *inode, sector_t iblock, |
803 | struct buffer_head *bh, int create) | |
804 | { | |
805 | return _ext4_get_block(inode, iblock, bh, | |
806 | create ? EXT4_GET_BLOCKS_CREATE : 0); | |
807 | } | |
808 | ||
ac27a0ec DK |
809 | /* |
810 | * `handle' can be NULL if create is zero | |
811 | */ | |
617ba13b | 812 | struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode, |
725d26d3 | 813 | ext4_lblk_t block, int create, int *errp) |
ac27a0ec | 814 | { |
2ed88685 TT |
815 | struct ext4_map_blocks map; |
816 | struct buffer_head *bh; | |
ac27a0ec DK |
817 | int fatal = 0, err; |
818 | ||
819 | J_ASSERT(handle != NULL || create == 0); | |
820 | ||
2ed88685 TT |
821 | map.m_lblk = block; |
822 | map.m_len = 1; | |
823 | err = ext4_map_blocks(handle, inode, &map, | |
824 | create ? EXT4_GET_BLOCKS_CREATE : 0); | |
ac27a0ec | 825 | |
90b0a973 CM |
826 | /* ensure we send some value back into *errp */ |
827 | *errp = 0; | |
828 | ||
0f70b406 TT |
829 | if (create && err == 0) |
830 | err = -ENOSPC; /* should never happen */ | |
2ed88685 TT |
831 | if (err < 0) |
832 | *errp = err; | |
833 | if (err <= 0) | |
834 | return NULL; | |
2ed88685 TT |
835 | |
836 | bh = sb_getblk(inode->i_sb, map.m_pblk); | |
aebf0243 | 837 | if (unlikely(!bh)) { |
860d21e2 | 838 | *errp = -ENOMEM; |
2ed88685 | 839 | return NULL; |
ac27a0ec | 840 | } |
2ed88685 TT |
841 | if (map.m_flags & EXT4_MAP_NEW) { |
842 | J_ASSERT(create != 0); | |
843 | J_ASSERT(handle != NULL); | |
ac27a0ec | 844 | |
2ed88685 TT |
845 | /* |
846 | * Now that we do not always journal data, we should | |
847 | * keep in mind whether this should always journal the | |
848 | * new buffer as metadata. For now, regular file | |
849 | * writes use ext4_get_block instead, so it's not a | |
850 | * problem. | |
851 | */ | |
852 | lock_buffer(bh); | |
853 | BUFFER_TRACE(bh, "call get_create_access"); | |
854 | fatal = ext4_journal_get_create_access(handle, bh); | |
855 | if (!fatal && !buffer_uptodate(bh)) { | |
856 | memset(bh->b_data, 0, inode->i_sb->s_blocksize); | |
857 | set_buffer_uptodate(bh); | |
ac27a0ec | 858 | } |
2ed88685 TT |
859 | unlock_buffer(bh); |
860 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); | |
861 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
862 | if (!fatal) | |
863 | fatal = err; | |
864 | } else { | |
865 | BUFFER_TRACE(bh, "not a new buffer"); | |
ac27a0ec | 866 | } |
2ed88685 TT |
867 | if (fatal) { |
868 | *errp = fatal; | |
869 | brelse(bh); | |
870 | bh = NULL; | |
871 | } | |
872 | return bh; | |
ac27a0ec DK |
873 | } |
874 | ||
617ba13b | 875 | struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode, |
725d26d3 | 876 | ext4_lblk_t block, int create, int *err) |
ac27a0ec | 877 | { |
af5bc92d | 878 | struct buffer_head *bh; |
ac27a0ec | 879 | |
617ba13b | 880 | bh = ext4_getblk(handle, inode, block, create, err); |
ac27a0ec DK |
881 | if (!bh) |
882 | return bh; | |
883 | if (buffer_uptodate(bh)) | |
884 | return bh; | |
65299a3b | 885 | ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh); |
ac27a0ec DK |
886 | wait_on_buffer(bh); |
887 | if (buffer_uptodate(bh)) | |
888 | return bh; | |
889 | put_bh(bh); | |
890 | *err = -EIO; | |
891 | return NULL; | |
892 | } | |
893 | ||
f19d5870 TM |
894 | int ext4_walk_page_buffers(handle_t *handle, |
895 | struct buffer_head *head, | |
896 | unsigned from, | |
897 | unsigned to, | |
898 | int *partial, | |
899 | int (*fn)(handle_t *handle, | |
900 | struct buffer_head *bh)) | |
ac27a0ec DK |
901 | { |
902 | struct buffer_head *bh; | |
903 | unsigned block_start, block_end; | |
904 | unsigned blocksize = head->b_size; | |
905 | int err, ret = 0; | |
906 | struct buffer_head *next; | |
907 | ||
af5bc92d TT |
908 | for (bh = head, block_start = 0; |
909 | ret == 0 && (bh != head || !block_start); | |
de9a55b8 | 910 | block_start = block_end, bh = next) { |
ac27a0ec DK |
911 | next = bh->b_this_page; |
912 | block_end = block_start + blocksize; | |
913 | if (block_end <= from || block_start >= to) { | |
914 | if (partial && !buffer_uptodate(bh)) | |
915 | *partial = 1; | |
916 | continue; | |
917 | } | |
918 | err = (*fn)(handle, bh); | |
919 | if (!ret) | |
920 | ret = err; | |
921 | } | |
922 | return ret; | |
923 | } | |
924 | ||
925 | /* | |
926 | * To preserve ordering, it is essential that the hole instantiation and | |
927 | * the data write be encapsulated in a single transaction. We cannot | |
617ba13b | 928 | * close off a transaction and start a new one between the ext4_get_block() |
dab291af | 929 | * and the commit_write(). So doing the jbd2_journal_start at the start of |
ac27a0ec DK |
930 | * prepare_write() is the right place. |
931 | * | |
36ade451 JK |
932 | * Also, this function can nest inside ext4_writepage(). In that case, we |
933 | * *know* that ext4_writepage() has generated enough buffer credits to do the | |
934 | * whole page. So we won't block on the journal in that case, which is good, | |
935 | * because the caller may be PF_MEMALLOC. | |
ac27a0ec | 936 | * |
617ba13b | 937 | * By accident, ext4 can be reentered when a transaction is open via |
ac27a0ec DK |
938 | * quota file writes. If we were to commit the transaction while thus |
939 | * reentered, there can be a deadlock - we would be holding a quota | |
940 | * lock, and the commit would never complete if another thread had a | |
941 | * transaction open and was blocking on the quota lock - a ranking | |
942 | * violation. | |
943 | * | |
dab291af | 944 | * So what we do is to rely on the fact that jbd2_journal_stop/journal_start |
ac27a0ec DK |
945 | * will _not_ run commit under these circumstances because handle->h_ref |
946 | * is elevated. We'll still have enough credits for the tiny quotafile | |
947 | * write. | |
948 | */ | |
f19d5870 TM |
949 | int do_journal_get_write_access(handle_t *handle, |
950 | struct buffer_head *bh) | |
ac27a0ec | 951 | { |
56d35a4c JK |
952 | int dirty = buffer_dirty(bh); |
953 | int ret; | |
954 | ||
ac27a0ec DK |
955 | if (!buffer_mapped(bh) || buffer_freed(bh)) |
956 | return 0; | |
56d35a4c | 957 | /* |
ebdec241 | 958 | * __block_write_begin() could have dirtied some buffers. Clean |
56d35a4c JK |
959 | * the dirty bit as jbd2_journal_get_write_access() could complain |
960 | * otherwise about fs integrity issues. Setting of the dirty bit | |
ebdec241 | 961 | * by __block_write_begin() isn't a real problem here as we clear |
56d35a4c JK |
962 | * the bit before releasing a page lock and thus writeback cannot |
963 | * ever write the buffer. | |
964 | */ | |
965 | if (dirty) | |
966 | clear_buffer_dirty(bh); | |
967 | ret = ext4_journal_get_write_access(handle, bh); | |
968 | if (!ret && dirty) | |
969 | ret = ext4_handle_dirty_metadata(handle, NULL, bh); | |
970 | return ret; | |
ac27a0ec DK |
971 | } |
972 | ||
8b0f165f AP |
973 | static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock, |
974 | struct buffer_head *bh_result, int create); | |
bfc1af65 | 975 | static int ext4_write_begin(struct file *file, struct address_space *mapping, |
de9a55b8 TT |
976 | loff_t pos, unsigned len, unsigned flags, |
977 | struct page **pagep, void **fsdata) | |
ac27a0ec | 978 | { |
af5bc92d | 979 | struct inode *inode = mapping->host; |
1938a150 | 980 | int ret, needed_blocks; |
ac27a0ec DK |
981 | handle_t *handle; |
982 | int retries = 0; | |
af5bc92d | 983 | struct page *page; |
de9a55b8 | 984 | pgoff_t index; |
af5bc92d | 985 | unsigned from, to; |
6fa3eb70 S |
986 | #if defined(FEATURE_STORAGE_PID_LOGGER) |
987 | extern unsigned char *page_logger; | |
988 | struct page_pid_logger *tmp_logger; | |
989 | unsigned long page_index; | |
990 | extern spinlock_t g_locker; | |
991 | unsigned long g_flags; | |
992 | #endif | |
bfc1af65 | 993 | |
9bffad1e | 994 | trace_ext4_write_begin(inode, pos, len, flags); |
1938a150 AK |
995 | /* |
996 | * Reserve one block more for addition to orphan list in case | |
997 | * we allocate blocks but write fails for some reason | |
998 | */ | |
999 | needed_blocks = ext4_writepage_trans_blocks(inode) + 1; | |
de9a55b8 | 1000 | index = pos >> PAGE_CACHE_SHIFT; |
af5bc92d TT |
1001 | from = pos & (PAGE_CACHE_SIZE - 1); |
1002 | to = from + len; | |
ac27a0ec | 1003 | |
f19d5870 TM |
1004 | if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { |
1005 | ret = ext4_try_to_write_inline_data(mapping, inode, pos, len, | |
1006 | flags, pagep); | |
1007 | if (ret < 0) | |
47564bfb TT |
1008 | return ret; |
1009 | if (ret == 1) | |
1010 | return 0; | |
f19d5870 TM |
1011 | } |
1012 | ||
47564bfb TT |
1013 | /* |
1014 | * grab_cache_page_write_begin() can take a long time if the | |
1015 | * system is thrashing due to memory pressure, or if the page | |
1016 | * is being written back. So grab it first before we start | |
1017 | * the transaction handle. This also allows us to allocate | |
1018 | * the page (if needed) without using GFP_NOFS. | |
1019 | */ | |
1020 | retry_grab: | |
1021 | page = grab_cache_page_write_begin(mapping, index, flags); | |
1022 | if (!page) | |
1023 | return -ENOMEM; | |
1024 | unlock_page(page); | |
1025 | ||
1026 | retry_journal: | |
9924a92a | 1027 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks); |
af5bc92d | 1028 | if (IS_ERR(handle)) { |
47564bfb TT |
1029 | page_cache_release(page); |
1030 | return PTR_ERR(handle); | |
7479d2b9 | 1031 | } |
ac27a0ec | 1032 | |
47564bfb TT |
1033 | lock_page(page); |
1034 | if (page->mapping != mapping) { | |
1035 | /* The page got truncated from under us */ | |
1036 | unlock_page(page); | |
1037 | page_cache_release(page); | |
cf108bca | 1038 | ext4_journal_stop(handle); |
47564bfb | 1039 | goto retry_grab; |
cf108bca | 1040 | } |
47564bfb | 1041 | wait_on_page_writeback(page); |
cf108bca | 1042 | |
744692dc | 1043 | if (ext4_should_dioread_nolock(inode)) |
6e1db88d | 1044 | ret = __block_write_begin(page, pos, len, ext4_get_block_write); |
744692dc | 1045 | else |
6e1db88d | 1046 | ret = __block_write_begin(page, pos, len, ext4_get_block); |
bfc1af65 NP |
1047 | |
1048 | if (!ret && ext4_should_journal_data(inode)) { | |
f19d5870 TM |
1049 | ret = ext4_walk_page_buffers(handle, page_buffers(page), |
1050 | from, to, NULL, | |
1051 | do_journal_get_write_access); | |
ac27a0ec | 1052 | } |
bfc1af65 NP |
1053 | |
1054 | if (ret) { | |
af5bc92d | 1055 | unlock_page(page); |
ae4d5372 | 1056 | /* |
6e1db88d | 1057 | * __block_write_begin may have instantiated a few blocks |
ae4d5372 AK |
1058 | * outside i_size. Trim these off again. Don't need |
1059 | * i_size_read because we hold i_mutex. | |
1938a150 AK |
1060 | * |
1061 | * Add inode to orphan list in case we crash before | |
1062 | * truncate finishes | |
ae4d5372 | 1063 | */ |
ffacfa7a | 1064 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
1938a150 AK |
1065 | ext4_orphan_add(handle, inode); |
1066 | ||
1067 | ext4_journal_stop(handle); | |
1068 | if (pos + len > inode->i_size) { | |
b9a4207d | 1069 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1070 | /* |
ffacfa7a | 1071 | * If truncate failed early the inode might |
1938a150 AK |
1072 | * still be on the orphan list; we need to |
1073 | * make sure the inode is removed from the | |
1074 | * orphan list in that case. | |
1075 | */ | |
1076 | if (inode->i_nlink) | |
1077 | ext4_orphan_del(NULL, inode); | |
1078 | } | |
bfc1af65 | 1079 | |
47564bfb TT |
1080 | if (ret == -ENOSPC && |
1081 | ext4_should_retry_alloc(inode->i_sb, &retries)) | |
1082 | goto retry_journal; | |
1083 | page_cache_release(page); | |
1084 | return ret; | |
1085 | } | |
1086 | *pagep = page; | |
6fa3eb70 S |
1087 | #if defined(FEATURE_STORAGE_PID_LOGGER) |
1088 | if( page_logger && (*pagep)) { | |
1089 | //#if defined(CONFIG_FLATMEM) | |
1090 | //page_index = (unsigned long)((*pagep) - mem_map) ; | |
1091 | //#else | |
1092 | page_index = (unsigned long)(__page_to_pfn(*pagep))- PHYS_PFN_OFFSET; | |
1093 | //#endif | |
1094 | tmp_logger =((struct page_pid_logger *)page_logger) + page_index; | |
1095 | spin_lock_irqsave(&g_locker, g_flags); | |
1096 | if( page_index < num_physpages) { | |
1097 | if( tmp_logger->pid1 == 0XFFFF) | |
1098 | tmp_logger->pid1 = current->pid; | |
1099 | else if( tmp_logger->pid1 != current->pid) | |
1100 | tmp_logger->pid2 = current->pid; | |
1101 | } | |
1102 | spin_unlock_irqrestore(&g_locker, g_flags); | |
1103 | } | |
1104 | #endif | |
ac27a0ec DK |
1105 | return ret; |
1106 | } | |
1107 | ||
bfc1af65 NP |
1108 | /* For write_end() in data=journal mode */ |
1109 | static int write_end_fn(handle_t *handle, struct buffer_head *bh) | |
ac27a0ec | 1110 | { |
13fca323 | 1111 | int ret; |
ac27a0ec DK |
1112 | if (!buffer_mapped(bh) || buffer_freed(bh)) |
1113 | return 0; | |
1114 | set_buffer_uptodate(bh); | |
13fca323 TT |
1115 | ret = ext4_handle_dirty_metadata(handle, NULL, bh); |
1116 | clear_buffer_meta(bh); | |
1117 | clear_buffer_prio(bh); | |
1118 | return ret; | |
ac27a0ec DK |
1119 | } |
1120 | ||
eed4333f ZL |
1121 | /* |
1122 | * We need to pick up the new inode size which generic_commit_write gave us | |
1123 | * `file' can be NULL - eg, when called from page_symlink(). | |
1124 | * | |
1125 | * ext4 never places buffers on inode->i_mapping->private_list. metadata | |
1126 | * buffers are managed internally. | |
1127 | */ | |
1128 | static int ext4_write_end(struct file *file, | |
1129 | struct address_space *mapping, | |
1130 | loff_t pos, unsigned len, unsigned copied, | |
1131 | struct page *page, void *fsdata) | |
f8514083 | 1132 | { |
f8514083 | 1133 | handle_t *handle = ext4_journal_current_handle(); |
eed4333f ZL |
1134 | struct inode *inode = mapping->host; |
1135 | int ret = 0, ret2; | |
1136 | int i_size_changed = 0; | |
1137 | ||
1138 | trace_ext4_write_end(inode, pos, len, copied); | |
1139 | if (ext4_test_inode_state(inode, EXT4_STATE_ORDERED_MODE)) { | |
1140 | ret = ext4_jbd2_file_inode(handle, inode); | |
1141 | if (ret) { | |
1142 | unlock_page(page); | |
1143 | page_cache_release(page); | |
1144 | goto errout; | |
1145 | } | |
1146 | } | |
f8514083 | 1147 | |
0529b225 TT |
1148 | if (ext4_has_inline_data(inode)) { |
1149 | ret = ext4_write_inline_data_end(inode, pos, len, | |
1150 | copied, page); | |
1151 | if (ret < 0) | |
1152 | goto errout; | |
1153 | copied = ret; | |
1154 | } else | |
f19d5870 TM |
1155 | copied = block_write_end(file, mapping, pos, |
1156 | len, copied, page, fsdata); | |
f8514083 AK |
1157 | |
1158 | /* | |
1159 | * No need to use i_size_read() here, the i_size | |
eed4333f | 1160 | * cannot change under us because we hole i_mutex. |
f8514083 AK |
1161 | * |
1162 | * But it's important to update i_size while still holding page lock: | |
1163 | * page writeout could otherwise come in and zero beyond i_size. | |
1164 | */ | |
1165 | if (pos + copied > inode->i_size) { | |
1166 | i_size_write(inode, pos + copied); | |
1167 | i_size_changed = 1; | |
1168 | } | |
1169 | ||
eed4333f | 1170 | if (pos + copied > EXT4_I(inode)->i_disksize) { |
f8514083 AK |
1171 | /* We need to mark inode dirty even if |
1172 | * new_i_size is less that inode->i_size | |
eed4333f | 1173 | * but greater than i_disksize. (hint delalloc) |
f8514083 AK |
1174 | */ |
1175 | ext4_update_i_disksize(inode, (pos + copied)); | |
1176 | i_size_changed = 1; | |
1177 | } | |
1178 | unlock_page(page); | |
1179 | page_cache_release(page); | |
1180 | ||
1181 | /* | |
1182 | * Don't mark the inode dirty under page lock. First, it unnecessarily | |
1183 | * makes the holding time of page lock longer. Second, it forces lock | |
1184 | * ordering of page lock and transaction start for journaling | |
1185 | * filesystems. | |
1186 | */ | |
1187 | if (i_size_changed) | |
1188 | ext4_mark_inode_dirty(handle, inode); | |
1189 | ||
74d553aa TT |
1190 | if (copied < 0) |
1191 | ret = copied; | |
ffacfa7a | 1192 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
f8514083 AK |
1193 | /* if we have allocated more blocks and copied |
1194 | * less. We will have blocks allocated outside | |
1195 | * inode->i_size. So truncate them | |
1196 | */ | |
1197 | ext4_orphan_add(handle, inode); | |
74d553aa | 1198 | errout: |
617ba13b | 1199 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1200 | if (!ret) |
1201 | ret = ret2; | |
bfc1af65 | 1202 | |
f8514083 | 1203 | if (pos + len > inode->i_size) { |
b9a4207d | 1204 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1205 | /* |
ffacfa7a | 1206 | * If truncate failed early the inode might still be |
f8514083 AK |
1207 | * on the orphan list; we need to make sure the inode |
1208 | * is removed from the orphan list in that case. | |
1209 | */ | |
1210 | if (inode->i_nlink) | |
1211 | ext4_orphan_del(NULL, inode); | |
1212 | } | |
1213 | ||
bfc1af65 | 1214 | return ret ? ret : copied; |
ac27a0ec DK |
1215 | } |
1216 | ||
bfc1af65 | 1217 | static int ext4_journalled_write_end(struct file *file, |
de9a55b8 TT |
1218 | struct address_space *mapping, |
1219 | loff_t pos, unsigned len, unsigned copied, | |
1220 | struct page *page, void *fsdata) | |
ac27a0ec | 1221 | { |
617ba13b | 1222 | handle_t *handle = ext4_journal_current_handle(); |
bfc1af65 | 1223 | struct inode *inode = mapping->host; |
ac27a0ec DK |
1224 | int ret = 0, ret2; |
1225 | int partial = 0; | |
bfc1af65 | 1226 | unsigned from, to; |
cf17fea6 | 1227 | loff_t new_i_size; |
ac27a0ec | 1228 | |
9bffad1e | 1229 | trace_ext4_journalled_write_end(inode, pos, len, copied); |
bfc1af65 NP |
1230 | from = pos & (PAGE_CACHE_SIZE - 1); |
1231 | to = from + len; | |
1232 | ||
441c8508 CW |
1233 | BUG_ON(!ext4_handle_valid(handle)); |
1234 | ||
3fdcfb66 TM |
1235 | if (ext4_has_inline_data(inode)) |
1236 | copied = ext4_write_inline_data_end(inode, pos, len, | |
1237 | copied, page); | |
1238 | else { | |
1239 | if (copied < len) { | |
1240 | if (!PageUptodate(page)) | |
1241 | copied = 0; | |
1242 | page_zero_new_buffers(page, from+copied, to); | |
1243 | } | |
ac27a0ec | 1244 | |
3fdcfb66 TM |
1245 | ret = ext4_walk_page_buffers(handle, page_buffers(page), from, |
1246 | to, &partial, write_end_fn); | |
1247 | if (!partial) | |
1248 | SetPageUptodate(page); | |
1249 | } | |
cf17fea6 AK |
1250 | new_i_size = pos + copied; |
1251 | if (new_i_size > inode->i_size) | |
bfc1af65 | 1252 | i_size_write(inode, pos+copied); |
19f5fb7a | 1253 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
2d859db3 | 1254 | EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; |
cf17fea6 AK |
1255 | if (new_i_size > EXT4_I(inode)->i_disksize) { |
1256 | ext4_update_i_disksize(inode, new_i_size); | |
617ba13b | 1257 | ret2 = ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
1258 | if (!ret) |
1259 | ret = ret2; | |
1260 | } | |
bfc1af65 | 1261 | |
cf108bca | 1262 | unlock_page(page); |
f8514083 | 1263 | page_cache_release(page); |
ffacfa7a | 1264 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
f8514083 AK |
1265 | /* if we have allocated more blocks and copied |
1266 | * less. We will have blocks allocated outside | |
1267 | * inode->i_size. So truncate them | |
1268 | */ | |
1269 | ext4_orphan_add(handle, inode); | |
1270 | ||
617ba13b | 1271 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1272 | if (!ret) |
1273 | ret = ret2; | |
f8514083 | 1274 | if (pos + len > inode->i_size) { |
b9a4207d | 1275 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1276 | /* |
ffacfa7a | 1277 | * If truncate failed early the inode might still be |
f8514083 AK |
1278 | * on the orphan list; we need to make sure the inode |
1279 | * is removed from the orphan list in that case. | |
1280 | */ | |
1281 | if (inode->i_nlink) | |
1282 | ext4_orphan_del(NULL, inode); | |
1283 | } | |
bfc1af65 NP |
1284 | |
1285 | return ret ? ret : copied; | |
ac27a0ec | 1286 | } |
d2a17637 | 1287 | |
386ad67c LC |
1288 | /* |
1289 | * Reserve a metadata for a single block located at lblock | |
1290 | */ | |
1291 | static int ext4_da_reserve_metadata(struct inode *inode, ext4_lblk_t lblock) | |
1292 | { | |
386ad67c LC |
1293 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
1294 | struct ext4_inode_info *ei = EXT4_I(inode); | |
1295 | unsigned int md_needed; | |
1296 | ext4_lblk_t save_last_lblock; | |
1297 | int save_len; | |
1298 | ||
1299 | /* | |
1300 | * recalculate the amount of metadata blocks to reserve | |
1301 | * in order to allocate nrblocks | |
1302 | * worse case is one extent per block | |
1303 | */ | |
386ad67c LC |
1304 | spin_lock(&ei->i_block_reservation_lock); |
1305 | /* | |
1306 | * ext4_calc_metadata_amount() has side effects, which we have | |
1307 | * to be prepared undo if we fail to claim space. | |
1308 | */ | |
1309 | save_len = ei->i_da_metadata_calc_len; | |
1310 | save_last_lblock = ei->i_da_metadata_calc_last_lblock; | |
1311 | md_needed = EXT4_NUM_B2C(sbi, | |
1312 | ext4_calc_metadata_amount(inode, lblock)); | |
1313 | trace_ext4_da_reserve_space(inode, md_needed); | |
1314 | ||
1315 | /* | |
1316 | * We do still charge estimated metadata to the sb though; | |
1317 | * we cannot afford to run out of free blocks. | |
1318 | */ | |
1319 | if (ext4_claim_free_clusters(sbi, md_needed, 0)) { | |
1320 | ei->i_da_metadata_calc_len = save_len; | |
1321 | ei->i_da_metadata_calc_last_lblock = save_last_lblock; | |
1322 | spin_unlock(&ei->i_block_reservation_lock); | |
386ad67c LC |
1323 | return -ENOSPC; |
1324 | } | |
1325 | ei->i_reserved_meta_blocks += md_needed; | |
1326 | spin_unlock(&ei->i_block_reservation_lock); | |
1327 | ||
1328 | return 0; /* success */ | |
1329 | } | |
1330 | ||
9d0be502 | 1331 | /* |
7b415bf6 | 1332 | * Reserve a single cluster located at lblock |
9d0be502 | 1333 | */ |
01f49d0b | 1334 | static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock) |
d2a17637 | 1335 | { |
60e58e0f | 1336 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
0637c6f4 | 1337 | struct ext4_inode_info *ei = EXT4_I(inode); |
7b415bf6 | 1338 | unsigned int md_needed; |
5dd4056d | 1339 | int ret; |
03179fe9 TT |
1340 | ext4_lblk_t save_last_lblock; |
1341 | int save_len; | |
1342 | ||
1343 | /* | |
1344 | * We will charge metadata quota at writeout time; this saves | |
1345 | * us from metadata over-estimation, though we may go over by | |
1346 | * a small amount in the end. Here we just reserve for data. | |
1347 | */ | |
1348 | ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1)); | |
1349 | if (ret) | |
1350 | return ret; | |
d2a17637 MC |
1351 | |
1352 | /* | |
1353 | * recalculate the amount of metadata blocks to reserve | |
1354 | * in order to allocate nrblocks | |
1355 | * worse case is one extent per block | |
1356 | */ | |
0637c6f4 | 1357 | spin_lock(&ei->i_block_reservation_lock); |
03179fe9 TT |
1358 | /* |
1359 | * ext4_calc_metadata_amount() has side effects, which we have | |
1360 | * to be prepared undo if we fail to claim space. | |
1361 | */ | |
1362 | save_len = ei->i_da_metadata_calc_len; | |
1363 | save_last_lblock = ei->i_da_metadata_calc_last_lblock; | |
7b415bf6 AK |
1364 | md_needed = EXT4_NUM_B2C(sbi, |
1365 | ext4_calc_metadata_amount(inode, lblock)); | |
f8ec9d68 | 1366 | trace_ext4_da_reserve_space(inode, md_needed); |
d2a17637 | 1367 | |
72b8ab9d ES |
1368 | /* |
1369 | * We do still charge estimated metadata to the sb though; | |
1370 | * we cannot afford to run out of free blocks. | |
1371 | */ | |
e7d5f315 | 1372 | if (ext4_claim_free_clusters(sbi, md_needed + 1, 0)) { |
03179fe9 TT |
1373 | ei->i_da_metadata_calc_len = save_len; |
1374 | ei->i_da_metadata_calc_last_lblock = save_last_lblock; | |
1375 | spin_unlock(&ei->i_block_reservation_lock); | |
03179fe9 | 1376 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1)); |
d2a17637 MC |
1377 | return -ENOSPC; |
1378 | } | |
9d0be502 | 1379 | ei->i_reserved_data_blocks++; |
0637c6f4 TT |
1380 | ei->i_reserved_meta_blocks += md_needed; |
1381 | spin_unlock(&ei->i_block_reservation_lock); | |
39bc680a | 1382 | |
d2a17637 MC |
1383 | return 0; /* success */ |
1384 | } | |
1385 | ||
12219aea | 1386 | static void ext4_da_release_space(struct inode *inode, int to_free) |
d2a17637 MC |
1387 | { |
1388 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0637c6f4 | 1389 | struct ext4_inode_info *ei = EXT4_I(inode); |
d2a17637 | 1390 | |
cd213226 MC |
1391 | if (!to_free) |
1392 | return; /* Nothing to release, exit */ | |
1393 | ||
d2a17637 | 1394 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); |
cd213226 | 1395 | |
5a58ec87 | 1396 | trace_ext4_da_release_space(inode, to_free); |
0637c6f4 | 1397 | if (unlikely(to_free > ei->i_reserved_data_blocks)) { |
cd213226 | 1398 | /* |
0637c6f4 TT |
1399 | * if there aren't enough reserved blocks, then the |
1400 | * counter is messed up somewhere. Since this | |
1401 | * function is called from invalidate page, it's | |
1402 | * harmless to return without any action. | |
cd213226 | 1403 | */ |
8de5c325 | 1404 | ext4_warning(inode->i_sb, "ext4_da_release_space: " |
0637c6f4 | 1405 | "ino %lu, to_free %d with only %d reserved " |
1084f252 | 1406 | "data blocks", inode->i_ino, to_free, |
0637c6f4 TT |
1407 | ei->i_reserved_data_blocks); |
1408 | WARN_ON(1); | |
1409 | to_free = ei->i_reserved_data_blocks; | |
cd213226 | 1410 | } |
0637c6f4 | 1411 | ei->i_reserved_data_blocks -= to_free; |
cd213226 | 1412 | |
0637c6f4 TT |
1413 | if (ei->i_reserved_data_blocks == 0) { |
1414 | /* | |
1415 | * We can release all of the reserved metadata blocks | |
1416 | * only when we have written all of the delayed | |
1417 | * allocation blocks. | |
7b415bf6 AK |
1418 | * Note that in case of bigalloc, i_reserved_meta_blocks, |
1419 | * i_reserved_data_blocks, etc. refer to number of clusters. | |
0637c6f4 | 1420 | */ |
57042651 | 1421 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, |
72b8ab9d | 1422 | ei->i_reserved_meta_blocks); |
ee5f4d9c | 1423 | ei->i_reserved_meta_blocks = 0; |
9d0be502 | 1424 | ei->i_da_metadata_calc_len = 0; |
0637c6f4 | 1425 | } |
d2a17637 | 1426 | |
72b8ab9d | 1427 | /* update fs dirty data blocks counter */ |
57042651 | 1428 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free); |
d2a17637 | 1429 | |
d2a17637 | 1430 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
60e58e0f | 1431 | |
7b415bf6 | 1432 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free)); |
d2a17637 MC |
1433 | } |
1434 | ||
1435 | static void ext4_da_page_release_reservation(struct page *page, | |
de9a55b8 | 1436 | unsigned long offset) |
d2a17637 MC |
1437 | { |
1438 | int to_release = 0; | |
1439 | struct buffer_head *head, *bh; | |
1440 | unsigned int curr_off = 0; | |
7b415bf6 AK |
1441 | struct inode *inode = page->mapping->host; |
1442 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
1443 | int num_clusters; | |
51865fda | 1444 | ext4_fsblk_t lblk; |
d2a17637 MC |
1445 | |
1446 | head = page_buffers(page); | |
1447 | bh = head; | |
1448 | do { | |
1449 | unsigned int next_off = curr_off + bh->b_size; | |
1450 | ||
1451 | if ((offset <= curr_off) && (buffer_delay(bh))) { | |
1452 | to_release++; | |
1453 | clear_buffer_delay(bh); | |
1454 | } | |
1455 | curr_off = next_off; | |
1456 | } while ((bh = bh->b_this_page) != head); | |
7b415bf6 | 1457 | |
51865fda ZL |
1458 | if (to_release) { |
1459 | lblk = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
1460 | ext4_es_remove_extent(inode, lblk, to_release); | |
1461 | } | |
1462 | ||
7b415bf6 AK |
1463 | /* If we have released all the blocks belonging to a cluster, then we |
1464 | * need to release the reserved space for that cluster. */ | |
1465 | num_clusters = EXT4_NUM_B2C(sbi, to_release); | |
1466 | while (num_clusters > 0) { | |
7b415bf6 AK |
1467 | lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) + |
1468 | ((num_clusters - 1) << sbi->s_cluster_bits); | |
1469 | if (sbi->s_cluster_ratio == 1 || | |
7d1b1fbc | 1470 | !ext4_find_delalloc_cluster(inode, lblk)) |
7b415bf6 AK |
1471 | ext4_da_release_space(inode, 1); |
1472 | ||
1473 | num_clusters--; | |
1474 | } | |
d2a17637 | 1475 | } |
ac27a0ec | 1476 | |
64769240 AT |
1477 | /* |
1478 | * Delayed allocation stuff | |
1479 | */ | |
1480 | ||
64769240 AT |
1481 | /* |
1482 | * mpage_da_submit_io - walks through extent of pages and try to write | |
a1d6cc56 | 1483 | * them with writepage() call back |
64769240 AT |
1484 | * |
1485 | * @mpd->inode: inode | |
1486 | * @mpd->first_page: first page of the extent | |
1487 | * @mpd->next_page: page after the last page of the extent | |
64769240 AT |
1488 | * |
1489 | * By the time mpage_da_submit_io() is called we expect all blocks | |
1490 | * to be allocated. this may be wrong if allocation failed. | |
1491 | * | |
1492 | * As pages are already locked by write_cache_pages(), we can't use it | |
1493 | */ | |
1de3e3df TT |
1494 | static int mpage_da_submit_io(struct mpage_da_data *mpd, |
1495 | struct ext4_map_blocks *map) | |
64769240 | 1496 | { |
791b7f08 AK |
1497 | struct pagevec pvec; |
1498 | unsigned long index, end; | |
1499 | int ret = 0, err, nr_pages, i; | |
1500 | struct inode *inode = mpd->inode; | |
1501 | struct address_space *mapping = inode->i_mapping; | |
cb20d518 | 1502 | loff_t size = i_size_read(inode); |
3ecdb3a1 TT |
1503 | unsigned int len, block_start; |
1504 | struct buffer_head *bh, *page_bufs = NULL; | |
1de3e3df | 1505 | sector_t pblock = 0, cur_logical = 0; |
bd2d0210 | 1506 | struct ext4_io_submit io_submit; |
64769240 AT |
1507 | |
1508 | BUG_ON(mpd->next_page <= mpd->first_page); | |
a549984b | 1509 | memset(&io_submit, 0, sizeof(io_submit)); |
791b7f08 AK |
1510 | /* |
1511 | * We need to start from the first_page to the next_page - 1 | |
1512 | * to make sure we also write the mapped dirty buffer_heads. | |
8dc207c0 | 1513 | * If we look at mpd->b_blocknr we would only be looking |
791b7f08 AK |
1514 | * at the currently mapped buffer_heads. |
1515 | */ | |
64769240 AT |
1516 | index = mpd->first_page; |
1517 | end = mpd->next_page - 1; | |
1518 | ||
791b7f08 | 1519 | pagevec_init(&pvec, 0); |
64769240 | 1520 | while (index <= end) { |
791b7f08 | 1521 | nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE); |
64769240 AT |
1522 | if (nr_pages == 0) |
1523 | break; | |
1524 | for (i = 0; i < nr_pages; i++) { | |
f8bec370 | 1525 | int skip_page = 0; |
64769240 AT |
1526 | struct page *page = pvec.pages[i]; |
1527 | ||
791b7f08 AK |
1528 | index = page->index; |
1529 | if (index > end) | |
1530 | break; | |
cb20d518 TT |
1531 | |
1532 | if (index == size >> PAGE_CACHE_SHIFT) | |
1533 | len = size & ~PAGE_CACHE_MASK; | |
1534 | else | |
1535 | len = PAGE_CACHE_SIZE; | |
1de3e3df TT |
1536 | if (map) { |
1537 | cur_logical = index << (PAGE_CACHE_SHIFT - | |
1538 | inode->i_blkbits); | |
1539 | pblock = map->m_pblk + (cur_logical - | |
1540 | map->m_lblk); | |
1541 | } | |
791b7f08 AK |
1542 | index++; |
1543 | ||
1544 | BUG_ON(!PageLocked(page)); | |
1545 | BUG_ON(PageWriteback(page)); | |
1546 | ||
3ecdb3a1 TT |
1547 | bh = page_bufs = page_buffers(page); |
1548 | block_start = 0; | |
64769240 | 1549 | do { |
1de3e3df TT |
1550 | if (map && (cur_logical >= map->m_lblk) && |
1551 | (cur_logical <= (map->m_lblk + | |
1552 | (map->m_len - 1)))) { | |
29fa89d0 AK |
1553 | if (buffer_delay(bh)) { |
1554 | clear_buffer_delay(bh); | |
1555 | bh->b_blocknr = pblock; | |
29fa89d0 | 1556 | } |
1de3e3df TT |
1557 | if (buffer_unwritten(bh) || |
1558 | buffer_mapped(bh)) | |
1559 | BUG_ON(bh->b_blocknr != pblock); | |
1560 | if (map->m_flags & EXT4_MAP_UNINIT) | |
1561 | set_buffer_uninit(bh); | |
1562 | clear_buffer_unwritten(bh); | |
1563 | } | |
29fa89d0 | 1564 | |
13a79a47 YY |
1565 | /* |
1566 | * skip page if block allocation undone and | |
1567 | * block is dirty | |
1568 | */ | |
1569 | if (ext4_bh_delay_or_unwritten(NULL, bh)) | |
97498956 | 1570 | skip_page = 1; |
3ecdb3a1 TT |
1571 | bh = bh->b_this_page; |
1572 | block_start += bh->b_size; | |
64769240 AT |
1573 | cur_logical++; |
1574 | pblock++; | |
1de3e3df TT |
1575 | } while (bh != page_bufs); |
1576 | ||
f8bec370 JK |
1577 | if (skip_page) { |
1578 | unlock_page(page); | |
1579 | continue; | |
1580 | } | |
cb20d518 | 1581 | |
97498956 | 1582 | clear_page_dirty_for_io(page); |
fe089c77 JK |
1583 | err = ext4_bio_write_page(&io_submit, page, len, |
1584 | mpd->wbc); | |
cb20d518 | 1585 | if (!err) |
a1d6cc56 | 1586 | mpd->pages_written++; |
64769240 AT |
1587 | /* |
1588 | * In error case, we have to continue because | |
1589 | * remaining pages are still locked | |
64769240 AT |
1590 | */ |
1591 | if (ret == 0) | |
1592 | ret = err; | |
64769240 AT |
1593 | } |
1594 | pagevec_release(&pvec); | |
1595 | } | |
bd2d0210 | 1596 | ext4_io_submit(&io_submit); |
64769240 | 1597 | return ret; |
64769240 AT |
1598 | } |
1599 | ||
c7f5938a | 1600 | static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd) |
c4a0c46e AK |
1601 | { |
1602 | int nr_pages, i; | |
1603 | pgoff_t index, end; | |
1604 | struct pagevec pvec; | |
1605 | struct inode *inode = mpd->inode; | |
1606 | struct address_space *mapping = inode->i_mapping; | |
51865fda | 1607 | ext4_lblk_t start, last; |
c4a0c46e | 1608 | |
c7f5938a CW |
1609 | index = mpd->first_page; |
1610 | end = mpd->next_page - 1; | |
51865fda ZL |
1611 | |
1612 | start = index << (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
1613 | last = end << (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
1614 | ext4_es_remove_extent(inode, start, last - start + 1); | |
1615 | ||
66bea92c | 1616 | pagevec_init(&pvec, 0); |
c4a0c46e AK |
1617 | while (index <= end) { |
1618 | nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE); | |
1619 | if (nr_pages == 0) | |
1620 | break; | |
1621 | for (i = 0; i < nr_pages; i++) { | |
1622 | struct page *page = pvec.pages[i]; | |
9b1d0998 | 1623 | if (page->index > end) |
c4a0c46e | 1624 | break; |
c4a0c46e AK |
1625 | BUG_ON(!PageLocked(page)); |
1626 | BUG_ON(PageWriteback(page)); | |
1627 | block_invalidatepage(page, 0); | |
1628 | ClearPageUptodate(page); | |
1629 | unlock_page(page); | |
1630 | } | |
9b1d0998 JK |
1631 | index = pvec.pages[nr_pages - 1]->index + 1; |
1632 | pagevec_release(&pvec); | |
c4a0c46e AK |
1633 | } |
1634 | return; | |
1635 | } | |
1636 | ||
df22291f AK |
1637 | static void ext4_print_free_blocks(struct inode *inode) |
1638 | { | |
1639 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
92b97816 | 1640 | struct super_block *sb = inode->i_sb; |
f78ee70d | 1641 | struct ext4_inode_info *ei = EXT4_I(inode); |
92b97816 TT |
1642 | |
1643 | ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld", | |
5dee5437 | 1644 | EXT4_C2B(EXT4_SB(inode->i_sb), |
f78ee70d | 1645 | ext4_count_free_clusters(sb))); |
92b97816 TT |
1646 | ext4_msg(sb, KERN_CRIT, "Free/Dirty block details"); |
1647 | ext4_msg(sb, KERN_CRIT, "free_blocks=%lld", | |
f78ee70d | 1648 | (long long) EXT4_C2B(EXT4_SB(sb), |
57042651 | 1649 | percpu_counter_sum(&sbi->s_freeclusters_counter))); |
92b97816 | 1650 | ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld", |
f78ee70d | 1651 | (long long) EXT4_C2B(EXT4_SB(sb), |
7b415bf6 | 1652 | percpu_counter_sum(&sbi->s_dirtyclusters_counter))); |
92b97816 TT |
1653 | ext4_msg(sb, KERN_CRIT, "Block reservation details"); |
1654 | ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u", | |
f78ee70d | 1655 | ei->i_reserved_data_blocks); |
92b97816 | 1656 | ext4_msg(sb, KERN_CRIT, "i_reserved_meta_blocks=%u", |
f78ee70d LC |
1657 | ei->i_reserved_meta_blocks); |
1658 | ext4_msg(sb, KERN_CRIT, "i_allocated_meta_blocks=%u", | |
1659 | ei->i_allocated_meta_blocks); | |
df22291f AK |
1660 | return; |
1661 | } | |
1662 | ||
64769240 | 1663 | /* |
5a87b7a5 TT |
1664 | * mpage_da_map_and_submit - go through given space, map them |
1665 | * if necessary, and then submit them for I/O | |
64769240 | 1666 | * |
8dc207c0 | 1667 | * @mpd - bh describing space |
64769240 AT |
1668 | * |
1669 | * The function skips space we know is already mapped to disk blocks. | |
1670 | * | |
64769240 | 1671 | */ |
5a87b7a5 | 1672 | static void mpage_da_map_and_submit(struct mpage_da_data *mpd) |
64769240 | 1673 | { |
2ac3b6e0 | 1674 | int err, blks, get_blocks_flags; |
1de3e3df | 1675 | struct ext4_map_blocks map, *mapp = NULL; |
2fa3cdfb TT |
1676 | sector_t next = mpd->b_blocknr; |
1677 | unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits; | |
1678 | loff_t disksize = EXT4_I(mpd->inode)->i_disksize; | |
1679 | handle_t *handle = NULL; | |
64769240 AT |
1680 | |
1681 | /* | |
5a87b7a5 TT |
1682 | * If the blocks are mapped already, or we couldn't accumulate |
1683 | * any blocks, then proceed immediately to the submission stage. | |
2fa3cdfb | 1684 | */ |
5a87b7a5 TT |
1685 | if ((mpd->b_size == 0) || |
1686 | ((mpd->b_state & (1 << BH_Mapped)) && | |
1687 | !(mpd->b_state & (1 << BH_Delay)) && | |
1688 | !(mpd->b_state & (1 << BH_Unwritten)))) | |
1689 | goto submit_io; | |
2fa3cdfb TT |
1690 | |
1691 | handle = ext4_journal_current_handle(); | |
1692 | BUG_ON(!handle); | |
1693 | ||
79ffab34 | 1694 | /* |
79e83036 | 1695 | * Call ext4_map_blocks() to allocate any delayed allocation |
2ac3b6e0 TT |
1696 | * blocks, or to convert an uninitialized extent to be |
1697 | * initialized (in the case where we have written into | |
1698 | * one or more preallocated blocks). | |
1699 | * | |
1700 | * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to | |
1701 | * indicate that we are on the delayed allocation path. This | |
1702 | * affects functions in many different parts of the allocation | |
1703 | * call path. This flag exists primarily because we don't | |
79e83036 | 1704 | * want to change *many* call functions, so ext4_map_blocks() |
f2321097 | 1705 | * will set the EXT4_STATE_DELALLOC_RESERVED flag once the |
2ac3b6e0 TT |
1706 | * inode's allocation semaphore is taken. |
1707 | * | |
1708 | * If the blocks in questions were delalloc blocks, set | |
1709 | * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting | |
1710 | * variables are updated after the blocks have been allocated. | |
79ffab34 | 1711 | */ |
2ed88685 TT |
1712 | map.m_lblk = next; |
1713 | map.m_len = max_blocks; | |
27dd4385 LC |
1714 | /* |
1715 | * We're in delalloc path and it is possible that we're going to | |
1716 | * need more metadata blocks than previously reserved. However | |
1717 | * we must not fail because we're in writeback and there is | |
1718 | * nothing we can do about it so it might result in data loss. | |
1719 | * So use reserved blocks to allocate metadata if possible. | |
1720 | */ | |
1721 | get_blocks_flags = EXT4_GET_BLOCKS_CREATE | | |
1722 | EXT4_GET_BLOCKS_METADATA_NOFAIL; | |
744692dc JZ |
1723 | if (ext4_should_dioread_nolock(mpd->inode)) |
1724 | get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT; | |
2ac3b6e0 | 1725 | if (mpd->b_state & (1 << BH_Delay)) |
1296cc85 AK |
1726 | get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE; |
1727 | ||
27dd4385 | 1728 | |
2ed88685 | 1729 | blks = ext4_map_blocks(handle, mpd->inode, &map, get_blocks_flags); |
2fa3cdfb | 1730 | if (blks < 0) { |
e3570639 ES |
1731 | struct super_block *sb = mpd->inode->i_sb; |
1732 | ||
2fa3cdfb | 1733 | err = blks; |
ed5bde0b | 1734 | /* |
5a87b7a5 | 1735 | * If get block returns EAGAIN or ENOSPC and there |
97498956 TT |
1736 | * appears to be free blocks we will just let |
1737 | * mpage_da_submit_io() unlock all of the pages. | |
c4a0c46e AK |
1738 | */ |
1739 | if (err == -EAGAIN) | |
5a87b7a5 | 1740 | goto submit_io; |
df22291f | 1741 | |
5dee5437 | 1742 | if (err == -ENOSPC && ext4_count_free_clusters(sb)) { |
df22291f | 1743 | mpd->retval = err; |
5a87b7a5 | 1744 | goto submit_io; |
df22291f AK |
1745 | } |
1746 | ||
c4a0c46e | 1747 | /* |
ed5bde0b TT |
1748 | * get block failure will cause us to loop in |
1749 | * writepages, because a_ops->writepage won't be able | |
1750 | * to make progress. The page will be redirtied by | |
1751 | * writepage and writepages will again try to write | |
1752 | * the same. | |
c4a0c46e | 1753 | */ |
e3570639 ES |
1754 | if (!(EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)) { |
1755 | ext4_msg(sb, KERN_CRIT, | |
1756 | "delayed block allocation failed for inode %lu " | |
1757 | "at logical offset %llu with max blocks %zd " | |
1758 | "with error %d", mpd->inode->i_ino, | |
1759 | (unsigned long long) next, | |
1760 | mpd->b_size >> mpd->inode->i_blkbits, err); | |
1761 | ext4_msg(sb, KERN_CRIT, | |
01a523eb | 1762 | "This should not happen!! Data will be lost"); |
e3570639 ES |
1763 | if (err == -ENOSPC) |
1764 | ext4_print_free_blocks(mpd->inode); | |
030ba6bc | 1765 | } |
2fa3cdfb | 1766 | /* invalidate all the pages */ |
c7f5938a | 1767 | ext4_da_block_invalidatepages(mpd); |
e0fd9b90 CW |
1768 | |
1769 | /* Mark this page range as having been completed */ | |
1770 | mpd->io_done = 1; | |
5a87b7a5 | 1771 | return; |
c4a0c46e | 1772 | } |
2fa3cdfb TT |
1773 | BUG_ON(blks == 0); |
1774 | ||
1de3e3df | 1775 | mapp = ↦ |
2ed88685 TT |
1776 | if (map.m_flags & EXT4_MAP_NEW) { |
1777 | struct block_device *bdev = mpd->inode->i_sb->s_bdev; | |
1778 | int i; | |
64769240 | 1779 | |
2ed88685 TT |
1780 | for (i = 0; i < map.m_len; i++) |
1781 | unmap_underlying_metadata(bdev, map.m_pblk + i); | |
2fa3cdfb TT |
1782 | } |
1783 | ||
1784 | /* | |
03f5d8bc | 1785 | * Update on-disk size along with block allocation. |
2fa3cdfb TT |
1786 | */ |
1787 | disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits; | |
1788 | if (disksize > i_size_read(mpd->inode)) | |
1789 | disksize = i_size_read(mpd->inode); | |
1790 | if (disksize > EXT4_I(mpd->inode)->i_disksize) { | |
1791 | ext4_update_i_disksize(mpd->inode, disksize); | |
5a87b7a5 TT |
1792 | err = ext4_mark_inode_dirty(handle, mpd->inode); |
1793 | if (err) | |
1794 | ext4_error(mpd->inode->i_sb, | |
1795 | "Failed to mark inode %lu dirty", | |
1796 | mpd->inode->i_ino); | |
2fa3cdfb TT |
1797 | } |
1798 | ||
5a87b7a5 | 1799 | submit_io: |
1de3e3df | 1800 | mpage_da_submit_io(mpd, mapp); |
5a87b7a5 | 1801 | mpd->io_done = 1; |
64769240 AT |
1802 | } |
1803 | ||
bf068ee2 AK |
1804 | #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \ |
1805 | (1 << BH_Delay) | (1 << BH_Unwritten)) | |
64769240 AT |
1806 | |
1807 | /* | |
1808 | * mpage_add_bh_to_extent - try to add one more block to extent of blocks | |
1809 | * | |
1810 | * @mpd->lbh - extent of blocks | |
1811 | * @logical - logical number of the block in the file | |
b6a8e62f | 1812 | * @b_state - b_state of the buffer head added |
64769240 AT |
1813 | * |
1814 | * the function is used to collect contig. blocks in same state | |
1815 | */ | |
b6a8e62f | 1816 | static void mpage_add_bh_to_extent(struct mpage_da_data *mpd, sector_t logical, |
8dc207c0 | 1817 | unsigned long b_state) |
64769240 | 1818 | { |
64769240 | 1819 | sector_t next; |
b6a8e62f JK |
1820 | int blkbits = mpd->inode->i_blkbits; |
1821 | int nrblocks = mpd->b_size >> blkbits; | |
64769240 | 1822 | |
c445e3e0 ES |
1823 | /* |
1824 | * XXX Don't go larger than mballoc is willing to allocate | |
1825 | * This is a stopgap solution. We eventually need to fold | |
1826 | * mpage_da_submit_io() into this function and then call | |
79e83036 | 1827 | * ext4_map_blocks() multiple times in a loop |
c445e3e0 | 1828 | */ |
b6a8e62f | 1829 | if (nrblocks >= (8*1024*1024 >> blkbits)) |
c445e3e0 ES |
1830 | goto flush_it; |
1831 | ||
b6a8e62f JK |
1832 | /* check if the reserved journal credits might overflow */ |
1833 | if (!ext4_test_inode_flag(mpd->inode, EXT4_INODE_EXTENTS)) { | |
525f4ed8 MC |
1834 | if (nrblocks >= EXT4_MAX_TRANS_DATA) { |
1835 | /* | |
1836 | * With non-extent format we are limited by the journal | |
1837 | * credit available. Total credit needed to insert | |
1838 | * nrblocks contiguous blocks is dependent on the | |
1839 | * nrblocks. So limit nrblocks. | |
1840 | */ | |
1841 | goto flush_it; | |
525f4ed8 MC |
1842 | } |
1843 | } | |
64769240 AT |
1844 | /* |
1845 | * First block in the extent | |
1846 | */ | |
8dc207c0 TT |
1847 | if (mpd->b_size == 0) { |
1848 | mpd->b_blocknr = logical; | |
b6a8e62f | 1849 | mpd->b_size = 1 << blkbits; |
8dc207c0 | 1850 | mpd->b_state = b_state & BH_FLAGS; |
64769240 AT |
1851 | return; |
1852 | } | |
1853 | ||
8dc207c0 | 1854 | next = mpd->b_blocknr + nrblocks; |
64769240 AT |
1855 | /* |
1856 | * Can we merge the block to our big extent? | |
1857 | */ | |
8dc207c0 | 1858 | if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) { |
b6a8e62f | 1859 | mpd->b_size += 1 << blkbits; |
64769240 AT |
1860 | return; |
1861 | } | |
1862 | ||
525f4ed8 | 1863 | flush_it: |
64769240 AT |
1864 | /* |
1865 | * We couldn't merge the block to our extent, so we | |
1866 | * need to flush current extent and start new one | |
1867 | */ | |
5a87b7a5 | 1868 | mpage_da_map_and_submit(mpd); |
a1d6cc56 | 1869 | return; |
64769240 AT |
1870 | } |
1871 | ||
c364b22c | 1872 | static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh) |
29fa89d0 | 1873 | { |
c364b22c | 1874 | return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh); |
29fa89d0 AK |
1875 | } |
1876 | ||
5356f261 AK |
1877 | /* |
1878 | * This function is grabs code from the very beginning of | |
1879 | * ext4_map_blocks, but assumes that the caller is from delayed write | |
1880 | * time. This function looks up the requested blocks and sets the | |
1881 | * buffer delay bit under the protection of i_data_sem. | |
1882 | */ | |
1883 | static int ext4_da_map_blocks(struct inode *inode, sector_t iblock, | |
1884 | struct ext4_map_blocks *map, | |
1885 | struct buffer_head *bh) | |
1886 | { | |
d100eef2 | 1887 | struct extent_status es; |
5356f261 AK |
1888 | int retval; |
1889 | sector_t invalid_block = ~((sector_t) 0xffff); | |
921f266b DM |
1890 | #ifdef ES_AGGRESSIVE_TEST |
1891 | struct ext4_map_blocks orig_map; | |
1892 | ||
1893 | memcpy(&orig_map, map, sizeof(*map)); | |
1894 | #endif | |
5356f261 AK |
1895 | |
1896 | if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es)) | |
1897 | invalid_block = ~0; | |
1898 | ||
1899 | map->m_flags = 0; | |
1900 | ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u," | |
1901 | "logical block %lu\n", inode->i_ino, map->m_len, | |
1902 | (unsigned long) map->m_lblk); | |
d100eef2 ZL |
1903 | |
1904 | /* Lookup extent status tree firstly */ | |
1905 | if (ext4_es_lookup_extent(inode, iblock, &es)) { | |
1906 | ||
1907 | if (ext4_es_is_hole(&es)) { | |
1908 | retval = 0; | |
1909 | down_read((&EXT4_I(inode)->i_data_sem)); | |
1910 | goto add_delayed; | |
1911 | } | |
1912 | ||
1913 | /* | |
1914 | * Delayed extent could be allocated by fallocate. | |
1915 | * So we need to check it. | |
1916 | */ | |
1917 | if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) { | |
1918 | map_bh(bh, inode->i_sb, invalid_block); | |
1919 | set_buffer_new(bh); | |
1920 | set_buffer_delay(bh); | |
1921 | return 0; | |
1922 | } | |
1923 | ||
1924 | map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk; | |
1925 | retval = es.es_len - (iblock - es.es_lblk); | |
1926 | if (retval > map->m_len) | |
1927 | retval = map->m_len; | |
1928 | map->m_len = retval; | |
1929 | if (ext4_es_is_written(&es)) | |
1930 | map->m_flags |= EXT4_MAP_MAPPED; | |
1931 | else if (ext4_es_is_unwritten(&es)) | |
1932 | map->m_flags |= EXT4_MAP_UNWRITTEN; | |
1933 | else | |
1934 | BUG_ON(1); | |
1935 | ||
921f266b DM |
1936 | #ifdef ES_AGGRESSIVE_TEST |
1937 | ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0); | |
1938 | #endif | |
d100eef2 ZL |
1939 | return retval; |
1940 | } | |
1941 | ||
5356f261 AK |
1942 | /* |
1943 | * Try to see if we can get the block without requesting a new | |
1944 | * file system block. | |
1945 | */ | |
1946 | down_read((&EXT4_I(inode)->i_data_sem)); | |
9c3569b5 TM |
1947 | if (ext4_has_inline_data(inode)) { |
1948 | /* | |
1949 | * We will soon create blocks for this page, and let | |
1950 | * us pretend as if the blocks aren't allocated yet. | |
1951 | * In case of clusters, we have to handle the work | |
1952 | * of mapping from cluster so that the reserved space | |
1953 | * is calculated properly. | |
1954 | */ | |
1955 | if ((EXT4_SB(inode->i_sb)->s_cluster_ratio > 1) && | |
1956 | ext4_find_delalloc_cluster(inode, map->m_lblk)) | |
1957 | map->m_flags |= EXT4_MAP_FROM_CLUSTER; | |
1958 | retval = 0; | |
1959 | } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
d100eef2 ZL |
1960 | retval = ext4_ext_map_blocks(NULL, inode, map, |
1961 | EXT4_GET_BLOCKS_NO_PUT_HOLE); | |
5356f261 | 1962 | else |
d100eef2 ZL |
1963 | retval = ext4_ind_map_blocks(NULL, inode, map, |
1964 | EXT4_GET_BLOCKS_NO_PUT_HOLE); | |
5356f261 | 1965 | |
d100eef2 | 1966 | add_delayed: |
5356f261 | 1967 | if (retval == 0) { |
f7fec032 | 1968 | int ret; |
5356f261 AK |
1969 | /* |
1970 | * XXX: __block_prepare_write() unmaps passed block, | |
1971 | * is it OK? | |
1972 | */ | |
386ad67c LC |
1973 | /* |
1974 | * If the block was allocated from previously allocated cluster, | |
1975 | * then we don't need to reserve it again. However we still need | |
1976 | * to reserve metadata for every block we're going to write. | |
1977 | */ | |
5356f261 | 1978 | if (!(map->m_flags & EXT4_MAP_FROM_CLUSTER)) { |
f7fec032 ZL |
1979 | ret = ext4_da_reserve_space(inode, iblock); |
1980 | if (ret) { | |
5356f261 | 1981 | /* not enough space to reserve */ |
f7fec032 | 1982 | retval = ret; |
5356f261 | 1983 | goto out_unlock; |
f7fec032 | 1984 | } |
386ad67c LC |
1985 | } else { |
1986 | ret = ext4_da_reserve_metadata(inode, iblock); | |
1987 | if (ret) { | |
1988 | /* not enough space to reserve */ | |
1989 | retval = ret; | |
1990 | goto out_unlock; | |
1991 | } | |
5356f261 AK |
1992 | } |
1993 | ||
f7fec032 ZL |
1994 | ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, |
1995 | ~0, EXTENT_STATUS_DELAYED); | |
1996 | if (ret) { | |
1997 | retval = ret; | |
51865fda | 1998 | goto out_unlock; |
f7fec032 | 1999 | } |
51865fda | 2000 | |
5356f261 AK |
2001 | /* Clear EXT4_MAP_FROM_CLUSTER flag since its purpose is served |
2002 | * and it should not appear on the bh->b_state. | |
2003 | */ | |
2004 | map->m_flags &= ~EXT4_MAP_FROM_CLUSTER; | |
2005 | ||
2006 | map_bh(bh, inode->i_sb, invalid_block); | |
2007 | set_buffer_new(bh); | |
2008 | set_buffer_delay(bh); | |
f7fec032 ZL |
2009 | } else if (retval > 0) { |
2010 | int ret; | |
2011 | unsigned long long status; | |
2012 | ||
921f266b DM |
2013 | #ifdef ES_AGGRESSIVE_TEST |
2014 | if (retval != map->m_len) { | |
2015 | printk("ES len assertation failed for inode: %lu " | |
2016 | "retval %d != map->m_len %d " | |
2017 | "in %s (lookup)\n", inode->i_ino, retval, | |
2018 | map->m_len, __func__); | |
2019 | } | |
2020 | #endif | |
2021 | ||
f7fec032 ZL |
2022 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
2023 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; | |
2024 | ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, | |
2025 | map->m_pblk, status); | |
2026 | if (ret != 0) | |
2027 | retval = ret; | |
5356f261 AK |
2028 | } |
2029 | ||
2030 | out_unlock: | |
2031 | up_read((&EXT4_I(inode)->i_data_sem)); | |
2032 | ||
2033 | return retval; | |
2034 | } | |
2035 | ||
64769240 | 2036 | /* |
b920c755 TT |
2037 | * This is a special get_blocks_t callback which is used by |
2038 | * ext4_da_write_begin(). It will either return mapped block or | |
2039 | * reserve space for a single block. | |
29fa89d0 AK |
2040 | * |
2041 | * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set. | |
2042 | * We also have b_blocknr = -1 and b_bdev initialized properly | |
2043 | * | |
2044 | * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set. | |
2045 | * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev | |
2046 | * initialized properly. | |
64769240 | 2047 | */ |
9c3569b5 TM |
2048 | int ext4_da_get_block_prep(struct inode *inode, sector_t iblock, |
2049 | struct buffer_head *bh, int create) | |
64769240 | 2050 | { |
2ed88685 | 2051 | struct ext4_map_blocks map; |
64769240 AT |
2052 | int ret = 0; |
2053 | ||
2054 | BUG_ON(create == 0); | |
2ed88685 TT |
2055 | BUG_ON(bh->b_size != inode->i_sb->s_blocksize); |
2056 | ||
2057 | map.m_lblk = iblock; | |
2058 | map.m_len = 1; | |
64769240 AT |
2059 | |
2060 | /* | |
2061 | * first, we need to know whether the block is allocated already | |
2062 | * preallocated blocks are unmapped but should treated | |
2063 | * the same as allocated blocks. | |
2064 | */ | |
5356f261 AK |
2065 | ret = ext4_da_map_blocks(inode, iblock, &map, bh); |
2066 | if (ret <= 0) | |
2ed88685 | 2067 | return ret; |
64769240 | 2068 | |
2ed88685 TT |
2069 | map_bh(bh, inode->i_sb, map.m_pblk); |
2070 | bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags; | |
2071 | ||
2072 | if (buffer_unwritten(bh)) { | |
2073 | /* A delayed write to unwritten bh should be marked | |
2074 | * new and mapped. Mapped ensures that we don't do | |
2075 | * get_block multiple times when we write to the same | |
2076 | * offset and new ensures that we do proper zero out | |
2077 | * for partial write. | |
2078 | */ | |
2079 | set_buffer_new(bh); | |
c8205636 | 2080 | set_buffer_mapped(bh); |
2ed88685 TT |
2081 | } |
2082 | return 0; | |
64769240 | 2083 | } |
61628a3f | 2084 | |
62e086be AK |
2085 | static int bget_one(handle_t *handle, struct buffer_head *bh) |
2086 | { | |
2087 | get_bh(bh); | |
2088 | return 0; | |
2089 | } | |
2090 | ||
2091 | static int bput_one(handle_t *handle, struct buffer_head *bh) | |
2092 | { | |
2093 | put_bh(bh); | |
2094 | return 0; | |
2095 | } | |
2096 | ||
2097 | static int __ext4_journalled_writepage(struct page *page, | |
62e086be AK |
2098 | unsigned int len) |
2099 | { | |
2100 | struct address_space *mapping = page->mapping; | |
2101 | struct inode *inode = mapping->host; | |
3fdcfb66 | 2102 | struct buffer_head *page_bufs = NULL; |
62e086be | 2103 | handle_t *handle = NULL; |
3fdcfb66 TM |
2104 | int ret = 0, err = 0; |
2105 | int inline_data = ext4_has_inline_data(inode); | |
2106 | struct buffer_head *inode_bh = NULL; | |
62e086be | 2107 | |
cb20d518 | 2108 | ClearPageChecked(page); |
3fdcfb66 TM |
2109 | |
2110 | if (inline_data) { | |
2111 | BUG_ON(page->index != 0); | |
2112 | BUG_ON(len > ext4_get_max_inline_size(inode)); | |
2113 | inode_bh = ext4_journalled_write_inline_data(inode, len, page); | |
2114 | if (inode_bh == NULL) | |
2115 | goto out; | |
2116 | } else { | |
2117 | page_bufs = page_buffers(page); | |
2118 | if (!page_bufs) { | |
2119 | BUG(); | |
2120 | goto out; | |
2121 | } | |
2122 | ext4_walk_page_buffers(handle, page_bufs, 0, len, | |
2123 | NULL, bget_one); | |
2124 | } | |
62e086be AK |
2125 | /* As soon as we unlock the page, it can go away, but we have |
2126 | * references to buffers so we are safe */ | |
2127 | unlock_page(page); | |
2128 | ||
9924a92a TT |
2129 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
2130 | ext4_writepage_trans_blocks(inode)); | |
62e086be AK |
2131 | if (IS_ERR(handle)) { |
2132 | ret = PTR_ERR(handle); | |
2133 | goto out; | |
2134 | } | |
2135 | ||
441c8508 CW |
2136 | BUG_ON(!ext4_handle_valid(handle)); |
2137 | ||
3fdcfb66 TM |
2138 | if (inline_data) { |
2139 | ret = ext4_journal_get_write_access(handle, inode_bh); | |
62e086be | 2140 | |
3fdcfb66 TM |
2141 | err = ext4_handle_dirty_metadata(handle, inode, inode_bh); |
2142 | ||
2143 | } else { | |
2144 | ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL, | |
2145 | do_journal_get_write_access); | |
2146 | ||
2147 | err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL, | |
2148 | write_end_fn); | |
2149 | } | |
62e086be AK |
2150 | if (ret == 0) |
2151 | ret = err; | |
2d859db3 | 2152 | EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; |
62e086be AK |
2153 | err = ext4_journal_stop(handle); |
2154 | if (!ret) | |
2155 | ret = err; | |
2156 | ||
3fdcfb66 TM |
2157 | if (!ext4_has_inline_data(inode)) |
2158 | ext4_walk_page_buffers(handle, page_bufs, 0, len, | |
2159 | NULL, bput_one); | |
19f5fb7a | 2160 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
62e086be | 2161 | out: |
3fdcfb66 | 2162 | brelse(inode_bh); |
62e086be AK |
2163 | return ret; |
2164 | } | |
2165 | ||
61628a3f | 2166 | /* |
43ce1d23 AK |
2167 | * Note that we don't need to start a transaction unless we're journaling data |
2168 | * because we should have holes filled from ext4_page_mkwrite(). We even don't | |
2169 | * need to file the inode to the transaction's list in ordered mode because if | |
2170 | * we are writing back data added by write(), the inode is already there and if | |
25985edc | 2171 | * we are writing back data modified via mmap(), no one guarantees in which |
43ce1d23 AK |
2172 | * transaction the data will hit the disk. In case we are journaling data, we |
2173 | * cannot start transaction directly because transaction start ranks above page | |
2174 | * lock so we have to do some magic. | |
2175 | * | |
b920c755 TT |
2176 | * This function can get called via... |
2177 | * - ext4_da_writepages after taking page lock (have journal handle) | |
2178 | * - journal_submit_inode_data_buffers (no journal handle) | |
f6463b0d | 2179 | * - shrink_page_list via the kswapd/direct reclaim (no journal handle) |
b920c755 | 2180 | * - grab_page_cache when doing write_begin (have journal handle) |
43ce1d23 AK |
2181 | * |
2182 | * We don't do any block allocation in this function. If we have page with | |
2183 | * multiple blocks we need to write those buffer_heads that are mapped. This | |
2184 | * is important for mmaped based write. So if we do with blocksize 1K | |
2185 | * truncate(f, 1024); | |
2186 | * a = mmap(f, 0, 4096); | |
2187 | * a[0] = 'a'; | |
2188 | * truncate(f, 4096); | |
2189 | * we have in the page first buffer_head mapped via page_mkwrite call back | |
90802ed9 | 2190 | * but other buffer_heads would be unmapped but dirty (dirty done via the |
43ce1d23 AK |
2191 | * do_wp_page). So writepage should write the first block. If we modify |
2192 | * the mmap area beyond 1024 we will again get a page_fault and the | |
2193 | * page_mkwrite callback will do the block allocation and mark the | |
2194 | * buffer_heads mapped. | |
2195 | * | |
2196 | * We redirty the page if we have any buffer_heads that is either delay or | |
2197 | * unwritten in the page. | |
2198 | * | |
2199 | * We can get recursively called as show below. | |
2200 | * | |
2201 | * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() -> | |
2202 | * ext4_writepage() | |
2203 | * | |
2204 | * But since we don't do any block allocation we should not deadlock. | |
2205 | * Page also have the dirty flag cleared so we don't get recurive page_lock. | |
61628a3f | 2206 | */ |
43ce1d23 | 2207 | static int ext4_writepage(struct page *page, |
62e086be | 2208 | struct writeback_control *wbc) |
64769240 | 2209 | { |
f8bec370 | 2210 | int ret = 0; |
61628a3f | 2211 | loff_t size; |
498e5f24 | 2212 | unsigned int len; |
744692dc | 2213 | struct buffer_head *page_bufs = NULL; |
61628a3f | 2214 | struct inode *inode = page->mapping->host; |
36ade451 | 2215 | struct ext4_io_submit io_submit; |
61628a3f | 2216 | |
a9c667f8 | 2217 | trace_ext4_writepage(page); |
f0e6c985 AK |
2218 | size = i_size_read(inode); |
2219 | if (page->index == size >> PAGE_CACHE_SHIFT) | |
2220 | len = size & ~PAGE_CACHE_MASK; | |
2221 | else | |
2222 | len = PAGE_CACHE_SIZE; | |
64769240 | 2223 | |
a42afc5f | 2224 | page_bufs = page_buffers(page); |
a42afc5f | 2225 | /* |
fe386132 JK |
2226 | * We cannot do block allocation or other extent handling in this |
2227 | * function. If there are buffers needing that, we have to redirty | |
2228 | * the page. But we may reach here when we do a journal commit via | |
2229 | * journal_submit_inode_data_buffers() and in that case we must write | |
2230 | * allocated buffers to achieve data=ordered mode guarantees. | |
a42afc5f | 2231 | */ |
f19d5870 TM |
2232 | if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL, |
2233 | ext4_bh_delay_or_unwritten)) { | |
f8bec370 | 2234 | redirty_page_for_writepage(wbc, page); |
fe386132 JK |
2235 | if (current->flags & PF_MEMALLOC) { |
2236 | /* | |
2237 | * For memory cleaning there's no point in writing only | |
2238 | * some buffers. So just bail out. Warn if we came here | |
2239 | * from direct reclaim. | |
2240 | */ | |
2241 | WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) | |
2242 | == PF_MEMALLOC); | |
f0e6c985 AK |
2243 | unlock_page(page); |
2244 | return 0; | |
2245 | } | |
a42afc5f | 2246 | } |
64769240 | 2247 | |
cb20d518 | 2248 | if (PageChecked(page) && ext4_should_journal_data(inode)) |
43ce1d23 AK |
2249 | /* |
2250 | * It's mmapped pagecache. Add buffers and journal it. There | |
2251 | * doesn't seem much point in redirtying the page here. | |
2252 | */ | |
3f0ca309 | 2253 | return __ext4_journalled_writepage(page, len); |
43ce1d23 | 2254 | |
a549984b | 2255 | memset(&io_submit, 0, sizeof(io_submit)); |
36ade451 JK |
2256 | ret = ext4_bio_write_page(&io_submit, page, len, wbc); |
2257 | ext4_io_submit(&io_submit); | |
64769240 AT |
2258 | return ret; |
2259 | } | |
2260 | ||
61628a3f | 2261 | /* |
525f4ed8 | 2262 | * This is called via ext4_da_writepages() to |
25985edc | 2263 | * calculate the total number of credits to reserve to fit |
525f4ed8 MC |
2264 | * a single extent allocation into a single transaction, |
2265 | * ext4_da_writpeages() will loop calling this before | |
2266 | * the block allocation. | |
61628a3f | 2267 | */ |
525f4ed8 MC |
2268 | |
2269 | static int ext4_da_writepages_trans_blocks(struct inode *inode) | |
2270 | { | |
2271 | int max_blocks = EXT4_I(inode)->i_reserved_data_blocks; | |
2272 | ||
2273 | /* | |
2274 | * With non-extent format the journal credit needed to | |
2275 | * insert nrblocks contiguous block is dependent on | |
2276 | * number of contiguous block. So we will limit | |
2277 | * number of contiguous block to a sane value | |
2278 | */ | |
12e9b892 | 2279 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) && |
525f4ed8 MC |
2280 | (max_blocks > EXT4_MAX_TRANS_DATA)) |
2281 | max_blocks = EXT4_MAX_TRANS_DATA; | |
2282 | ||
2283 | return ext4_chunk_trans_blocks(inode, max_blocks); | |
2284 | } | |
61628a3f | 2285 | |
8e48dcfb TT |
2286 | /* |
2287 | * write_cache_pages_da - walk the list of dirty pages of the given | |
8eb9e5ce | 2288 | * address space and accumulate pages that need writing, and call |
168fc022 TT |
2289 | * mpage_da_map_and_submit to map a single contiguous memory region |
2290 | * and then write them. | |
8e48dcfb | 2291 | */ |
9c3569b5 TM |
2292 | static int write_cache_pages_da(handle_t *handle, |
2293 | struct address_space *mapping, | |
8e48dcfb | 2294 | struct writeback_control *wbc, |
72f84e65 ES |
2295 | struct mpage_da_data *mpd, |
2296 | pgoff_t *done_index) | |
8e48dcfb | 2297 | { |
4f01b02c | 2298 | struct buffer_head *bh, *head; |
168fc022 | 2299 | struct inode *inode = mapping->host; |
4f01b02c TT |
2300 | struct pagevec pvec; |
2301 | unsigned int nr_pages; | |
2302 | sector_t logical; | |
2303 | pgoff_t index, end; | |
2304 | long nr_to_write = wbc->nr_to_write; | |
2305 | int i, tag, ret = 0; | |
8e48dcfb | 2306 | |
168fc022 TT |
2307 | memset(mpd, 0, sizeof(struct mpage_da_data)); |
2308 | mpd->wbc = wbc; | |
2309 | mpd->inode = inode; | |
8e48dcfb TT |
2310 | pagevec_init(&pvec, 0); |
2311 | index = wbc->range_start >> PAGE_CACHE_SHIFT; | |
2312 | end = wbc->range_end >> PAGE_CACHE_SHIFT; | |
2313 | ||
6e6938b6 | 2314 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
5b41d924 ES |
2315 | tag = PAGECACHE_TAG_TOWRITE; |
2316 | else | |
2317 | tag = PAGECACHE_TAG_DIRTY; | |
2318 | ||
72f84e65 | 2319 | *done_index = index; |
4f01b02c | 2320 | while (index <= end) { |
5b41d924 | 2321 | nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag, |
8e48dcfb TT |
2322 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1); |
2323 | if (nr_pages == 0) | |
4f01b02c | 2324 | return 0; |
8e48dcfb TT |
2325 | |
2326 | for (i = 0; i < nr_pages; i++) { | |
2327 | struct page *page = pvec.pages[i]; | |
2328 | ||
2329 | /* | |
2330 | * At this point, the page may be truncated or | |
2331 | * invalidated (changing page->mapping to NULL), or | |
2332 | * even swizzled back from swapper_space to tmpfs file | |
2333 | * mapping. However, page->index will not change | |
2334 | * because we have a reference on the page. | |
2335 | */ | |
4f01b02c TT |
2336 | if (page->index > end) |
2337 | goto out; | |
8e48dcfb | 2338 | |
72f84e65 ES |
2339 | *done_index = page->index + 1; |
2340 | ||
78aaced3 TT |
2341 | /* |
2342 | * If we can't merge this page, and we have | |
2343 | * accumulated an contiguous region, write it | |
2344 | */ | |
2345 | if ((mpd->next_page != page->index) && | |
2346 | (mpd->next_page != mpd->first_page)) { | |
2347 | mpage_da_map_and_submit(mpd); | |
2348 | goto ret_extent_tail; | |
2349 | } | |
2350 | ||
8e48dcfb TT |
2351 | lock_page(page); |
2352 | ||
2353 | /* | |
4f01b02c TT |
2354 | * If the page is no longer dirty, or its |
2355 | * mapping no longer corresponds to inode we | |
2356 | * are writing (which means it has been | |
2357 | * truncated or invalidated), or the page is | |
2358 | * already under writeback and we are not | |
2359 | * doing a data integrity writeback, skip the page | |
8e48dcfb | 2360 | */ |
4f01b02c TT |
2361 | if (!PageDirty(page) || |
2362 | (PageWriteback(page) && | |
2363 | (wbc->sync_mode == WB_SYNC_NONE)) || | |
2364 | unlikely(page->mapping != mapping)) { | |
8e48dcfb TT |
2365 | unlock_page(page); |
2366 | continue; | |
2367 | } | |
2368 | ||
7cb1a535 | 2369 | wait_on_page_writeback(page); |
8e48dcfb | 2370 | BUG_ON(PageWriteback(page)); |
8e48dcfb | 2371 | |
9c3569b5 TM |
2372 | /* |
2373 | * If we have inline data and arrive here, it means that | |
2374 | * we will soon create the block for the 1st page, so | |
2375 | * we'd better clear the inline data here. | |
2376 | */ | |
2377 | if (ext4_has_inline_data(inode)) { | |
2378 | BUG_ON(ext4_test_inode_state(inode, | |
2379 | EXT4_STATE_MAY_INLINE_DATA)); | |
2380 | ext4_destroy_inline_data(handle, inode); | |
2381 | } | |
2382 | ||
168fc022 | 2383 | if (mpd->next_page != page->index) |
8eb9e5ce | 2384 | mpd->first_page = page->index; |
8eb9e5ce TT |
2385 | mpd->next_page = page->index + 1; |
2386 | logical = (sector_t) page->index << | |
2387 | (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
2388 | ||
f8bec370 JK |
2389 | /* Add all dirty buffers to mpd */ |
2390 | head = page_buffers(page); | |
2391 | bh = head; | |
2392 | do { | |
2393 | BUG_ON(buffer_locked(bh)); | |
8eb9e5ce | 2394 | /* |
f8bec370 JK |
2395 | * We need to try to allocate unmapped blocks |
2396 | * in the same page. Otherwise we won't make | |
2397 | * progress with the page in ext4_writepage | |
8eb9e5ce | 2398 | */ |
f8bec370 JK |
2399 | if (ext4_bh_delay_or_unwritten(NULL, bh)) { |
2400 | mpage_add_bh_to_extent(mpd, logical, | |
f8bec370 JK |
2401 | bh->b_state); |
2402 | if (mpd->io_done) | |
2403 | goto ret_extent_tail; | |
2404 | } else if (buffer_dirty(bh) && | |
2405 | buffer_mapped(bh)) { | |
8eb9e5ce | 2406 | /* |
f8bec370 JK |
2407 | * mapped dirty buffer. We need to |
2408 | * update the b_state because we look | |
2409 | * at b_state in mpage_da_map_blocks. | |
2410 | * We don't update b_size because if we | |
2411 | * find an unmapped buffer_head later | |
2412 | * we need to use the b_state flag of | |
2413 | * that buffer_head. | |
8eb9e5ce | 2414 | */ |
f8bec370 JK |
2415 | if (mpd->b_size == 0) |
2416 | mpd->b_state = | |
2417 | bh->b_state & BH_FLAGS; | |
2418 | } | |
2419 | logical++; | |
2420 | } while ((bh = bh->b_this_page) != head); | |
8e48dcfb TT |
2421 | |
2422 | if (nr_to_write > 0) { | |
2423 | nr_to_write--; | |
2424 | if (nr_to_write == 0 && | |
4f01b02c | 2425 | wbc->sync_mode == WB_SYNC_NONE) |
8e48dcfb TT |
2426 | /* |
2427 | * We stop writing back only if we are | |
2428 | * not doing integrity sync. In case of | |
2429 | * integrity sync we have to keep going | |
2430 | * because someone may be concurrently | |
2431 | * dirtying pages, and we might have | |
2432 | * synced a lot of newly appeared dirty | |
2433 | * pages, but have not synced all of the | |
2434 | * old dirty pages. | |
2435 | */ | |
4f01b02c | 2436 | goto out; |
8e48dcfb TT |
2437 | } |
2438 | } | |
2439 | pagevec_release(&pvec); | |
2440 | cond_resched(); | |
2441 | } | |
4f01b02c TT |
2442 | return 0; |
2443 | ret_extent_tail: | |
2444 | ret = MPAGE_DA_EXTENT_TAIL; | |
8eb9e5ce TT |
2445 | out: |
2446 | pagevec_release(&pvec); | |
2447 | cond_resched(); | |
8e48dcfb TT |
2448 | return ret; |
2449 | } | |
2450 | ||
2451 | ||
64769240 | 2452 | static int ext4_da_writepages(struct address_space *mapping, |
a1d6cc56 | 2453 | struct writeback_control *wbc) |
64769240 | 2454 | { |
22208ded AK |
2455 | pgoff_t index; |
2456 | int range_whole = 0; | |
61628a3f | 2457 | handle_t *handle = NULL; |
df22291f | 2458 | struct mpage_da_data mpd; |
5e745b04 | 2459 | struct inode *inode = mapping->host; |
498e5f24 | 2460 | int pages_written = 0; |
55138e0b | 2461 | unsigned int max_pages; |
2acf2c26 | 2462 | int range_cyclic, cycled = 1, io_done = 0; |
55138e0b TT |
2463 | int needed_blocks, ret = 0; |
2464 | long desired_nr_to_write, nr_to_writebump = 0; | |
de89de6e | 2465 | loff_t range_start = wbc->range_start; |
5e745b04 | 2466 | struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb); |
72f84e65 | 2467 | pgoff_t done_index = 0; |
5b41d924 | 2468 | pgoff_t end; |
1bce63d1 | 2469 | struct blk_plug plug; |
61628a3f | 2470 | |
9bffad1e | 2471 | trace_ext4_da_writepages(inode, wbc); |
ba80b101 | 2472 | |
61628a3f MC |
2473 | /* |
2474 | * No pages to write? This is mainly a kludge to avoid starting | |
2475 | * a transaction for special inodes like journal inode on last iput() | |
2476 | * because that could violate lock ordering on umount | |
2477 | */ | |
a1d6cc56 | 2478 | if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) |
61628a3f | 2479 | return 0; |
2a21e37e TT |
2480 | |
2481 | /* | |
2482 | * If the filesystem has aborted, it is read-only, so return | |
2483 | * right away instead of dumping stack traces later on that | |
2484 | * will obscure the real source of the problem. We test | |
4ab2f15b | 2485 | * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because |
2a21e37e TT |
2486 | * the latter could be true if the filesystem is mounted |
2487 | * read-only, and in that case, ext4_da_writepages should | |
2488 | * *never* be called, so if that ever happens, we would want | |
2489 | * the stack trace. | |
2490 | */ | |
4ab2f15b | 2491 | if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) |
2a21e37e TT |
2492 | return -EROFS; |
2493 | ||
22208ded AK |
2494 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
2495 | range_whole = 1; | |
61628a3f | 2496 | |
2acf2c26 AK |
2497 | range_cyclic = wbc->range_cyclic; |
2498 | if (wbc->range_cyclic) { | |
22208ded | 2499 | index = mapping->writeback_index; |
2acf2c26 AK |
2500 | if (index) |
2501 | cycled = 0; | |
2502 | wbc->range_start = index << PAGE_CACHE_SHIFT; | |
2503 | wbc->range_end = LLONG_MAX; | |
2504 | wbc->range_cyclic = 0; | |
5b41d924 ES |
2505 | end = -1; |
2506 | } else { | |
22208ded | 2507 | index = wbc->range_start >> PAGE_CACHE_SHIFT; |
5b41d924 ES |
2508 | end = wbc->range_end >> PAGE_CACHE_SHIFT; |
2509 | } | |
a1d6cc56 | 2510 | |
55138e0b TT |
2511 | /* |
2512 | * This works around two forms of stupidity. The first is in | |
2513 | * the writeback code, which caps the maximum number of pages | |
2514 | * written to be 1024 pages. This is wrong on multiple | |
2515 | * levels; different architectues have a different page size, | |
2516 | * which changes the maximum amount of data which gets | |
2517 | * written. Secondly, 4 megabytes is way too small. XFS | |
2518 | * forces this value to be 16 megabytes by multiplying | |
2519 | * nr_to_write parameter by four, and then relies on its | |
2520 | * allocator to allocate larger extents to make them | |
2521 | * contiguous. Unfortunately this brings us to the second | |
2522 | * stupidity, which is that ext4's mballoc code only allocates | |
2523 | * at most 2048 blocks. So we force contiguous writes up to | |
2524 | * the number of dirty blocks in the inode, or | |
2525 | * sbi->max_writeback_mb_bump whichever is smaller. | |
2526 | */ | |
2527 | max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT); | |
b443e733 ES |
2528 | if (!range_cyclic && range_whole) { |
2529 | if (wbc->nr_to_write == LONG_MAX) | |
2530 | desired_nr_to_write = wbc->nr_to_write; | |
2531 | else | |
2532 | desired_nr_to_write = wbc->nr_to_write * 8; | |
2533 | } else | |
55138e0b TT |
2534 | desired_nr_to_write = ext4_num_dirty_pages(inode, index, |
2535 | max_pages); | |
2536 | if (desired_nr_to_write > max_pages) | |
2537 | desired_nr_to_write = max_pages; | |
2538 | ||
2539 | if (wbc->nr_to_write < desired_nr_to_write) { | |
2540 | nr_to_writebump = desired_nr_to_write - wbc->nr_to_write; | |
2541 | wbc->nr_to_write = desired_nr_to_write; | |
2542 | } | |
2543 | ||
2acf2c26 | 2544 | retry: |
6e6938b6 | 2545 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
5b41d924 ES |
2546 | tag_pages_for_writeback(mapping, index, end); |
2547 | ||
1bce63d1 | 2548 | blk_start_plug(&plug); |
22208ded | 2549 | while (!ret && wbc->nr_to_write > 0) { |
a1d6cc56 AK |
2550 | |
2551 | /* | |
2552 | * we insert one extent at a time. So we need | |
2553 | * credit needed for single extent allocation. | |
2554 | * journalled mode is currently not supported | |
2555 | * by delalloc | |
2556 | */ | |
2557 | BUG_ON(ext4_should_journal_data(inode)); | |
525f4ed8 | 2558 | needed_blocks = ext4_da_writepages_trans_blocks(inode); |
a1d6cc56 | 2559 | |
61628a3f | 2560 | /* start a new transaction*/ |
9924a92a TT |
2561 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
2562 | needed_blocks); | |
61628a3f MC |
2563 | if (IS_ERR(handle)) { |
2564 | ret = PTR_ERR(handle); | |
1693918e | 2565 | ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: " |
fbe845dd | 2566 | "%ld pages, ino %lu; err %d", __func__, |
a1d6cc56 | 2567 | wbc->nr_to_write, inode->i_ino, ret); |
3c1fcb2c | 2568 | blk_finish_plug(&plug); |
61628a3f MC |
2569 | goto out_writepages; |
2570 | } | |
f63e6005 TT |
2571 | |
2572 | /* | |
8eb9e5ce | 2573 | * Now call write_cache_pages_da() to find the next |
f63e6005 | 2574 | * contiguous region of logical blocks that need |
8eb9e5ce | 2575 | * blocks to be allocated by ext4 and submit them. |
f63e6005 | 2576 | */ |
9c3569b5 TM |
2577 | ret = write_cache_pages_da(handle, mapping, |
2578 | wbc, &mpd, &done_index); | |
f63e6005 | 2579 | /* |
af901ca1 | 2580 | * If we have a contiguous extent of pages and we |
f63e6005 TT |
2581 | * haven't done the I/O yet, map the blocks and submit |
2582 | * them for I/O. | |
2583 | */ | |
2584 | if (!mpd.io_done && mpd.next_page != mpd.first_page) { | |
5a87b7a5 | 2585 | mpage_da_map_and_submit(&mpd); |
f63e6005 TT |
2586 | ret = MPAGE_DA_EXTENT_TAIL; |
2587 | } | |
b3a3ca8c | 2588 | trace_ext4_da_write_pages(inode, &mpd); |
f63e6005 | 2589 | wbc->nr_to_write -= mpd.pages_written; |
df22291f | 2590 | |
61628a3f | 2591 | ext4_journal_stop(handle); |
df22291f | 2592 | |
8f64b32e | 2593 | if ((mpd.retval == -ENOSPC) && sbi->s_journal) { |
22208ded AK |
2594 | /* commit the transaction which would |
2595 | * free blocks released in the transaction | |
2596 | * and try again | |
2597 | */ | |
df22291f | 2598 | jbd2_journal_force_commit_nested(sbi->s_journal); |
22208ded AK |
2599 | ret = 0; |
2600 | } else if (ret == MPAGE_DA_EXTENT_TAIL) { | |
a1d6cc56 | 2601 | /* |
8de49e67 KM |
2602 | * Got one extent now try with rest of the pages. |
2603 | * If mpd.retval is set -EIO, journal is aborted. | |
2604 | * So we don't need to write any more. | |
a1d6cc56 | 2605 | */ |
22208ded | 2606 | pages_written += mpd.pages_written; |
8de49e67 | 2607 | ret = mpd.retval; |
2acf2c26 | 2608 | io_done = 1; |
22208ded | 2609 | } else if (wbc->nr_to_write) |
61628a3f MC |
2610 | /* |
2611 | * There is no more writeout needed | |
2612 | * or we requested for a noblocking writeout | |
2613 | * and we found the device congested | |
2614 | */ | |
61628a3f | 2615 | break; |
a1d6cc56 | 2616 | } |
1bce63d1 | 2617 | blk_finish_plug(&plug); |
2acf2c26 AK |
2618 | if (!io_done && !cycled) { |
2619 | cycled = 1; | |
2620 | index = 0; | |
2621 | wbc->range_start = index << PAGE_CACHE_SHIFT; | |
2622 | wbc->range_end = mapping->writeback_index - 1; | |
2623 | goto retry; | |
2624 | } | |
22208ded AK |
2625 | |
2626 | /* Update index */ | |
2acf2c26 | 2627 | wbc->range_cyclic = range_cyclic; |
22208ded AK |
2628 | if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) |
2629 | /* | |
2630 | * set the writeback_index so that range_cyclic | |
2631 | * mode will write it back later | |
2632 | */ | |
72f84e65 | 2633 | mapping->writeback_index = done_index; |
a1d6cc56 | 2634 | |
61628a3f | 2635 | out_writepages: |
2faf2e19 | 2636 | wbc->nr_to_write -= nr_to_writebump; |
de89de6e | 2637 | wbc->range_start = range_start; |
9bffad1e | 2638 | trace_ext4_da_writepages_result(inode, wbc, ret, pages_written); |
61628a3f | 2639 | return ret; |
64769240 AT |
2640 | } |
2641 | ||
79f0be8d AK |
2642 | static int ext4_nonda_switch(struct super_block *sb) |
2643 | { | |
5c1ff336 | 2644 | s64 free_clusters, dirty_clusters; |
79f0be8d AK |
2645 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
2646 | ||
2647 | /* | |
2648 | * switch to non delalloc mode if we are running low | |
2649 | * on free block. The free block accounting via percpu | |
179f7ebf | 2650 | * counters can get slightly wrong with percpu_counter_batch getting |
79f0be8d AK |
2651 | * accumulated on each CPU without updating global counters |
2652 | * Delalloc need an accurate free block accounting. So switch | |
2653 | * to non delalloc when we are near to error range. | |
2654 | */ | |
5c1ff336 EW |
2655 | free_clusters = |
2656 | percpu_counter_read_positive(&sbi->s_freeclusters_counter); | |
2657 | dirty_clusters = | |
2658 | percpu_counter_read_positive(&sbi->s_dirtyclusters_counter); | |
00d4e736 TT |
2659 | /* |
2660 | * Start pushing delalloc when 1/2 of free blocks are dirty. | |
2661 | */ | |
5c1ff336 | 2662 | if (dirty_clusters && (free_clusters < 2 * dirty_clusters)) |
10ee27a0 | 2663 | try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE); |
00d4e736 | 2664 | |
5c1ff336 EW |
2665 | if (2 * free_clusters < 3 * dirty_clusters || |
2666 | free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) { | |
79f0be8d | 2667 | /* |
c8afb446 ES |
2668 | * free block count is less than 150% of dirty blocks |
2669 | * or free blocks is less than watermark | |
79f0be8d AK |
2670 | */ |
2671 | return 1; | |
2672 | } | |
2673 | return 0; | |
2674 | } | |
2675 | ||
64769240 | 2676 | static int ext4_da_write_begin(struct file *file, struct address_space *mapping, |
de9a55b8 TT |
2677 | loff_t pos, unsigned len, unsigned flags, |
2678 | struct page **pagep, void **fsdata) | |
64769240 | 2679 | { |
72b8ab9d | 2680 | int ret, retries = 0; |
64769240 AT |
2681 | struct page *page; |
2682 | pgoff_t index; | |
64769240 AT |
2683 | struct inode *inode = mapping->host; |
2684 | handle_t *handle; | |
2685 | ||
2686 | index = pos >> PAGE_CACHE_SHIFT; | |
79f0be8d AK |
2687 | |
2688 | if (ext4_nonda_switch(inode->i_sb)) { | |
2689 | *fsdata = (void *)FALL_BACK_TO_NONDELALLOC; | |
2690 | return ext4_write_begin(file, mapping, pos, | |
2691 | len, flags, pagep, fsdata); | |
2692 | } | |
2693 | *fsdata = (void *)0; | |
9bffad1e | 2694 | trace_ext4_da_write_begin(inode, pos, len, flags); |
9c3569b5 TM |
2695 | |
2696 | if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { | |
2697 | ret = ext4_da_write_inline_data_begin(mapping, inode, | |
2698 | pos, len, flags, | |
2699 | pagep, fsdata); | |
2700 | if (ret < 0) | |
47564bfb TT |
2701 | return ret; |
2702 | if (ret == 1) | |
2703 | return 0; | |
9c3569b5 TM |
2704 | } |
2705 | ||
47564bfb TT |
2706 | /* |
2707 | * grab_cache_page_write_begin() can take a long time if the | |
2708 | * system is thrashing due to memory pressure, or if the page | |
2709 | * is being written back. So grab it first before we start | |
2710 | * the transaction handle. This also allows us to allocate | |
2711 | * the page (if needed) without using GFP_NOFS. | |
2712 | */ | |
2713 | retry_grab: | |
2714 | page = grab_cache_page_write_begin(mapping, index, flags); | |
2715 | if (!page) | |
2716 | return -ENOMEM; | |
2717 | unlock_page(page); | |
2718 | ||
64769240 AT |
2719 | /* |
2720 | * With delayed allocation, we don't log the i_disksize update | |
2721 | * if there is delayed block allocation. But we still need | |
2722 | * to journalling the i_disksize update if writes to the end | |
2723 | * of file which has an already mapped buffer. | |
2724 | */ | |
47564bfb | 2725 | retry_journal: |
9924a92a | 2726 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, 1); |
64769240 | 2727 | if (IS_ERR(handle)) { |
47564bfb TT |
2728 | page_cache_release(page); |
2729 | return PTR_ERR(handle); | |
64769240 AT |
2730 | } |
2731 | ||
47564bfb TT |
2732 | lock_page(page); |
2733 | if (page->mapping != mapping) { | |
2734 | /* The page got truncated from under us */ | |
2735 | unlock_page(page); | |
2736 | page_cache_release(page); | |
d5a0d4f7 | 2737 | ext4_journal_stop(handle); |
47564bfb | 2738 | goto retry_grab; |
d5a0d4f7 | 2739 | } |
47564bfb TT |
2740 | /* In case writeback began while the page was unlocked */ |
2741 | wait_on_page_writeback(page); | |
64769240 | 2742 | |
6e1db88d | 2743 | ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep); |
64769240 AT |
2744 | if (ret < 0) { |
2745 | unlock_page(page); | |
2746 | ext4_journal_stop(handle); | |
ae4d5372 AK |
2747 | /* |
2748 | * block_write_begin may have instantiated a few blocks | |
2749 | * outside i_size. Trim these off again. Don't need | |
2750 | * i_size_read because we hold i_mutex. | |
2751 | */ | |
2752 | if (pos + len > inode->i_size) | |
b9a4207d | 2753 | ext4_truncate_failed_write(inode); |
47564bfb TT |
2754 | |
2755 | if (ret == -ENOSPC && | |
2756 | ext4_should_retry_alloc(inode->i_sb, &retries)) | |
2757 | goto retry_journal; | |
2758 | ||
2759 | page_cache_release(page); | |
2760 | return ret; | |
64769240 AT |
2761 | } |
2762 | ||
47564bfb | 2763 | *pagep = page; |
64769240 AT |
2764 | return ret; |
2765 | } | |
2766 | ||
632eaeab MC |
2767 | /* |
2768 | * Check if we should update i_disksize | |
2769 | * when write to the end of file but not require block allocation | |
2770 | */ | |
2771 | static int ext4_da_should_update_i_disksize(struct page *page, | |
de9a55b8 | 2772 | unsigned long offset) |
632eaeab MC |
2773 | { |
2774 | struct buffer_head *bh; | |
2775 | struct inode *inode = page->mapping->host; | |
2776 | unsigned int idx; | |
2777 | int i; | |
2778 | ||
2779 | bh = page_buffers(page); | |
2780 | idx = offset >> inode->i_blkbits; | |
2781 | ||
af5bc92d | 2782 | for (i = 0; i < idx; i++) |
632eaeab MC |
2783 | bh = bh->b_this_page; |
2784 | ||
29fa89d0 | 2785 | if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh)) |
632eaeab MC |
2786 | return 0; |
2787 | return 1; | |
2788 | } | |
2789 | ||
64769240 | 2790 | static int ext4_da_write_end(struct file *file, |
de9a55b8 TT |
2791 | struct address_space *mapping, |
2792 | loff_t pos, unsigned len, unsigned copied, | |
2793 | struct page *page, void *fsdata) | |
64769240 AT |
2794 | { |
2795 | struct inode *inode = mapping->host; | |
2796 | int ret = 0, ret2; | |
2797 | handle_t *handle = ext4_journal_current_handle(); | |
2798 | loff_t new_i_size; | |
632eaeab | 2799 | unsigned long start, end; |
79f0be8d AK |
2800 | int write_mode = (int)(unsigned long)fsdata; |
2801 | ||
74d553aa TT |
2802 | if (write_mode == FALL_BACK_TO_NONDELALLOC) |
2803 | return ext4_write_end(file, mapping, pos, | |
2804 | len, copied, page, fsdata); | |
632eaeab | 2805 | |
9bffad1e | 2806 | trace_ext4_da_write_end(inode, pos, len, copied); |
632eaeab | 2807 | start = pos & (PAGE_CACHE_SIZE - 1); |
af5bc92d | 2808 | end = start + copied - 1; |
64769240 AT |
2809 | |
2810 | /* | |
2811 | * generic_write_end() will run mark_inode_dirty() if i_size | |
2812 | * changes. So let's piggyback the i_disksize mark_inode_dirty | |
2813 | * into that. | |
2814 | */ | |
64769240 | 2815 | new_i_size = pos + copied; |
ea51d132 | 2816 | if (copied && new_i_size > EXT4_I(inode)->i_disksize) { |
9c3569b5 TM |
2817 | if (ext4_has_inline_data(inode) || |
2818 | ext4_da_should_update_i_disksize(page, end)) { | |
632eaeab | 2819 | down_write(&EXT4_I(inode)->i_data_sem); |
f3b59291 | 2820 | if (new_i_size > EXT4_I(inode)->i_disksize) |
632eaeab | 2821 | EXT4_I(inode)->i_disksize = new_i_size; |
632eaeab | 2822 | up_write(&EXT4_I(inode)->i_data_sem); |
cf17fea6 AK |
2823 | /* We need to mark inode dirty even if |
2824 | * new_i_size is less that inode->i_size | |
2825 | * bu greater than i_disksize.(hint delalloc) | |
2826 | */ | |
2827 | ext4_mark_inode_dirty(handle, inode); | |
64769240 | 2828 | } |
632eaeab | 2829 | } |
9c3569b5 TM |
2830 | |
2831 | if (write_mode != CONVERT_INLINE_DATA && | |
2832 | ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) && | |
2833 | ext4_has_inline_data(inode)) | |
2834 | ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied, | |
2835 | page); | |
2836 | else | |
2837 | ret2 = generic_write_end(file, mapping, pos, len, copied, | |
64769240 | 2838 | page, fsdata); |
9c3569b5 | 2839 | |
64769240 AT |
2840 | copied = ret2; |
2841 | if (ret2 < 0) | |
2842 | ret = ret2; | |
2843 | ret2 = ext4_journal_stop(handle); | |
2844 | if (!ret) | |
2845 | ret = ret2; | |
2846 | ||
2847 | return ret ? ret : copied; | |
2848 | } | |
2849 | ||
2850 | static void ext4_da_invalidatepage(struct page *page, unsigned long offset) | |
2851 | { | |
64769240 AT |
2852 | /* |
2853 | * Drop reserved blocks | |
2854 | */ | |
2855 | BUG_ON(!PageLocked(page)); | |
2856 | if (!page_has_buffers(page)) | |
2857 | goto out; | |
2858 | ||
d2a17637 | 2859 | ext4_da_page_release_reservation(page, offset); |
64769240 AT |
2860 | |
2861 | out: | |
2862 | ext4_invalidatepage(page, offset); | |
2863 | ||
2864 | return; | |
2865 | } | |
2866 | ||
ccd2506b TT |
2867 | /* |
2868 | * Force all delayed allocation blocks to be allocated for a given inode. | |
2869 | */ | |
2870 | int ext4_alloc_da_blocks(struct inode *inode) | |
2871 | { | |
fb40ba0d TT |
2872 | trace_ext4_alloc_da_blocks(inode); |
2873 | ||
ccd2506b TT |
2874 | if (!EXT4_I(inode)->i_reserved_data_blocks && |
2875 | !EXT4_I(inode)->i_reserved_meta_blocks) | |
2876 | return 0; | |
2877 | ||
2878 | /* | |
2879 | * We do something simple for now. The filemap_flush() will | |
2880 | * also start triggering a write of the data blocks, which is | |
2881 | * not strictly speaking necessary (and for users of | |
2882 | * laptop_mode, not even desirable). However, to do otherwise | |
2883 | * would require replicating code paths in: | |
de9a55b8 | 2884 | * |
ccd2506b TT |
2885 | * ext4_da_writepages() -> |
2886 | * write_cache_pages() ---> (via passed in callback function) | |
2887 | * __mpage_da_writepage() --> | |
2888 | * mpage_add_bh_to_extent() | |
2889 | * mpage_da_map_blocks() | |
2890 | * | |
2891 | * The problem is that write_cache_pages(), located in | |
2892 | * mm/page-writeback.c, marks pages clean in preparation for | |
2893 | * doing I/O, which is not desirable if we're not planning on | |
2894 | * doing I/O at all. | |
2895 | * | |
2896 | * We could call write_cache_pages(), and then redirty all of | |
380cf090 | 2897 | * the pages by calling redirty_page_for_writepage() but that |
ccd2506b TT |
2898 | * would be ugly in the extreme. So instead we would need to |
2899 | * replicate parts of the code in the above functions, | |
25985edc | 2900 | * simplifying them because we wouldn't actually intend to |
ccd2506b TT |
2901 | * write out the pages, but rather only collect contiguous |
2902 | * logical block extents, call the multi-block allocator, and | |
2903 | * then update the buffer heads with the block allocations. | |
de9a55b8 | 2904 | * |
ccd2506b TT |
2905 | * For now, though, we'll cheat by calling filemap_flush(), |
2906 | * which will map the blocks, and start the I/O, but not | |
2907 | * actually wait for the I/O to complete. | |
2908 | */ | |
2909 | return filemap_flush(inode->i_mapping); | |
2910 | } | |
64769240 | 2911 | |
ac27a0ec DK |
2912 | /* |
2913 | * bmap() is special. It gets used by applications such as lilo and by | |
2914 | * the swapper to find the on-disk block of a specific piece of data. | |
2915 | * | |
2916 | * Naturally, this is dangerous if the block concerned is still in the | |
617ba13b | 2917 | * journal. If somebody makes a swapfile on an ext4 data-journaling |
ac27a0ec DK |
2918 | * filesystem and enables swap, then they may get a nasty shock when the |
2919 | * data getting swapped to that swapfile suddenly gets overwritten by | |
2920 | * the original zero's written out previously to the journal and | |
2921 | * awaiting writeback in the kernel's buffer cache. | |
2922 | * | |
2923 | * So, if we see any bmap calls here on a modified, data-journaled file, | |
2924 | * take extra steps to flush any blocks which might be in the cache. | |
2925 | */ | |
617ba13b | 2926 | static sector_t ext4_bmap(struct address_space *mapping, sector_t block) |
ac27a0ec DK |
2927 | { |
2928 | struct inode *inode = mapping->host; | |
2929 | journal_t *journal; | |
2930 | int err; | |
2931 | ||
46c7f254 TM |
2932 | /* |
2933 | * We can get here for an inline file via the FIBMAP ioctl | |
2934 | */ | |
2935 | if (ext4_has_inline_data(inode)) | |
2936 | return 0; | |
2937 | ||
64769240 AT |
2938 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) && |
2939 | test_opt(inode->i_sb, DELALLOC)) { | |
2940 | /* | |
2941 | * With delalloc we want to sync the file | |
2942 | * so that we can make sure we allocate | |
2943 | * blocks for file | |
2944 | */ | |
2945 | filemap_write_and_wait(mapping); | |
2946 | } | |
2947 | ||
19f5fb7a TT |
2948 | if (EXT4_JOURNAL(inode) && |
2949 | ext4_test_inode_state(inode, EXT4_STATE_JDATA)) { | |
ac27a0ec DK |
2950 | /* |
2951 | * This is a REALLY heavyweight approach, but the use of | |
2952 | * bmap on dirty files is expected to be extremely rare: | |
2953 | * only if we run lilo or swapon on a freshly made file | |
2954 | * do we expect this to happen. | |
2955 | * | |
2956 | * (bmap requires CAP_SYS_RAWIO so this does not | |
2957 | * represent an unprivileged user DOS attack --- we'd be | |
2958 | * in trouble if mortal users could trigger this path at | |
2959 | * will.) | |
2960 | * | |
617ba13b | 2961 | * NB. EXT4_STATE_JDATA is not set on files other than |
ac27a0ec DK |
2962 | * regular files. If somebody wants to bmap a directory |
2963 | * or symlink and gets confused because the buffer | |
2964 | * hasn't yet been flushed to disk, they deserve | |
2965 | * everything they get. | |
2966 | */ | |
2967 | ||
19f5fb7a | 2968 | ext4_clear_inode_state(inode, EXT4_STATE_JDATA); |
617ba13b | 2969 | journal = EXT4_JOURNAL(inode); |
dab291af MC |
2970 | jbd2_journal_lock_updates(journal); |
2971 | err = jbd2_journal_flush(journal); | |
2972 | jbd2_journal_unlock_updates(journal); | |
ac27a0ec DK |
2973 | |
2974 | if (err) | |
2975 | return 0; | |
2976 | } | |
2977 | ||
af5bc92d | 2978 | return generic_block_bmap(mapping, block, ext4_get_block); |
ac27a0ec DK |
2979 | } |
2980 | ||
617ba13b | 2981 | static int ext4_readpage(struct file *file, struct page *page) |
ac27a0ec | 2982 | { |
46c7f254 TM |
2983 | int ret = -EAGAIN; |
2984 | struct inode *inode = page->mapping->host; | |
2985 | ||
0562e0ba | 2986 | trace_ext4_readpage(page); |
46c7f254 TM |
2987 | |
2988 | if (ext4_has_inline_data(inode)) | |
2989 | ret = ext4_readpage_inline(inode, page); | |
2990 | ||
2991 | if (ret == -EAGAIN) | |
2992 | return mpage_readpage(page, ext4_get_block); | |
2993 | ||
2994 | return ret; | |
ac27a0ec DK |
2995 | } |
2996 | ||
2997 | static int | |
617ba13b | 2998 | ext4_readpages(struct file *file, struct address_space *mapping, |
ac27a0ec DK |
2999 | struct list_head *pages, unsigned nr_pages) |
3000 | { | |
46c7f254 TM |
3001 | struct inode *inode = mapping->host; |
3002 | ||
3003 | /* If the file has inline data, no need to do readpages. */ | |
3004 | if (ext4_has_inline_data(inode)) | |
3005 | return 0; | |
3006 | ||
617ba13b | 3007 | return mpage_readpages(mapping, pages, nr_pages, ext4_get_block); |
ac27a0ec DK |
3008 | } |
3009 | ||
617ba13b | 3010 | static void ext4_invalidatepage(struct page *page, unsigned long offset) |
ac27a0ec | 3011 | { |
0562e0ba JZ |
3012 | trace_ext4_invalidatepage(page, offset); |
3013 | ||
4520fb3c JK |
3014 | /* No journalling happens on data buffers when this function is used */ |
3015 | WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page))); | |
3016 | ||
3017 | block_invalidatepage(page, offset); | |
3018 | } | |
3019 | ||
53e87268 JK |
3020 | static int __ext4_journalled_invalidatepage(struct page *page, |
3021 | unsigned long offset) | |
4520fb3c JK |
3022 | { |
3023 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); | |
3024 | ||
3025 | trace_ext4_journalled_invalidatepage(page, offset); | |
3026 | ||
ac27a0ec DK |
3027 | /* |
3028 | * If it's a full truncate we just forget about the pending dirtying | |
3029 | */ | |
3030 | if (offset == 0) | |
3031 | ClearPageChecked(page); | |
3032 | ||
53e87268 JK |
3033 | return jbd2_journal_invalidatepage(journal, page, offset); |
3034 | } | |
3035 | ||
3036 | /* Wrapper for aops... */ | |
3037 | static void ext4_journalled_invalidatepage(struct page *page, | |
3038 | unsigned long offset) | |
3039 | { | |
3040 | WARN_ON(__ext4_journalled_invalidatepage(page, offset) < 0); | |
ac27a0ec DK |
3041 | } |
3042 | ||
617ba13b | 3043 | static int ext4_releasepage(struct page *page, gfp_t wait) |
ac27a0ec | 3044 | { |
617ba13b | 3045 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); |
ac27a0ec | 3046 | |
0562e0ba JZ |
3047 | trace_ext4_releasepage(page); |
3048 | ||
e1c36595 JK |
3049 | /* Page has dirty journalled data -> cannot release */ |
3050 | if (PageChecked(page)) | |
ac27a0ec | 3051 | return 0; |
0390131b FM |
3052 | if (journal) |
3053 | return jbd2_journal_try_to_free_buffers(journal, page, wait); | |
3054 | else | |
3055 | return try_to_free_buffers(page); | |
ac27a0ec DK |
3056 | } |
3057 | ||
2ed88685 TT |
3058 | /* |
3059 | * ext4_get_block used when preparing for a DIO write or buffer write. | |
3060 | * We allocate an uinitialized extent if blocks haven't been allocated. | |
3061 | * The extent will be converted to initialized after the IO is complete. | |
3062 | */ | |
f19d5870 | 3063 | int ext4_get_block_write(struct inode *inode, sector_t iblock, |
4c0425ff MC |
3064 | struct buffer_head *bh_result, int create) |
3065 | { | |
c7064ef1 | 3066 | ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n", |
8d5d02e6 | 3067 | inode->i_ino, create); |
2ed88685 TT |
3068 | return _ext4_get_block(inode, iblock, bh_result, |
3069 | EXT4_GET_BLOCKS_IO_CREATE_EXT); | |
4c0425ff MC |
3070 | } |
3071 | ||
729f52c6 | 3072 | static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock, |
8b0f165f | 3073 | struct buffer_head *bh_result, int create) |
729f52c6 | 3074 | { |
8b0f165f AP |
3075 | ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n", |
3076 | inode->i_ino, create); | |
3077 | return _ext4_get_block(inode, iblock, bh_result, | |
3078 | EXT4_GET_BLOCKS_NO_LOCK); | |
729f52c6 ZL |
3079 | } |
3080 | ||
4c0425ff | 3081 | static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset, |
552ef802 CH |
3082 | ssize_t size, void *private, int ret, |
3083 | bool is_async) | |
4c0425ff | 3084 | { |
496ad9aa | 3085 | struct inode *inode = file_inode(iocb->ki_filp); |
4c0425ff | 3086 | ext4_io_end_t *io_end = iocb->private; |
4c0425ff | 3087 | |
a549984b TT |
3088 | /* if not async direct IO or dio with 0 bytes write, just return */ |
3089 | if (!io_end || !size) | |
3090 | goto out; | |
4b70df18 | 3091 | |
88635ca2 | 3092 | ext_debug("ext4_end_io_dio(): io_end 0x%p " |
ace36ad4 | 3093 | "for inode %lu, iocb 0x%p, offset %llu, size %zd\n", |
8d5d02e6 MC |
3094 | iocb->private, io_end->inode->i_ino, iocb, offset, |
3095 | size); | |
8d5d02e6 | 3096 | |
b5a7e970 | 3097 | iocb->private = NULL; |
a549984b TT |
3098 | |
3099 | /* if not aio dio with unwritten extents, just free io and return */ | |
3100 | if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) { | |
3101 | ext4_free_io_end(io_end); | |
3102 | out: | |
3103 | inode_dio_done(inode); | |
3104 | if (is_async) | |
3105 | aio_complete(iocb, ret, 0); | |
3106 | return; | |
3107 | } | |
3108 | ||
4c0425ff MC |
3109 | io_end->offset = offset; |
3110 | io_end->size = size; | |
5b3ff237 JZ |
3111 | if (is_async) { |
3112 | io_end->iocb = iocb; | |
3113 | io_end->result = ret; | |
3114 | } | |
a549984b TT |
3115 | |
3116 | ext4_add_complete_io(io_end); | |
4c0425ff | 3117 | } |
c7064ef1 | 3118 | |
4c0425ff MC |
3119 | /* |
3120 | * For ext4 extent files, ext4 will do direct-io write to holes, | |
3121 | * preallocated extents, and those write extend the file, no need to | |
3122 | * fall back to buffered IO. | |
3123 | * | |
b595076a | 3124 | * For holes, we fallocate those blocks, mark them as uninitialized |
69c499d1 | 3125 | * If those blocks were preallocated, we mark sure they are split, but |
b595076a | 3126 | * still keep the range to write as uninitialized. |
4c0425ff | 3127 | * |
69c499d1 | 3128 | * The unwritten extents will be converted to written when DIO is completed. |
8d5d02e6 | 3129 | * For async direct IO, since the IO may still pending when return, we |
25985edc | 3130 | * set up an end_io call back function, which will do the conversion |
8d5d02e6 | 3131 | * when async direct IO completed. |
4c0425ff MC |
3132 | * |
3133 | * If the O_DIRECT write will extend the file then add this inode to the | |
3134 | * orphan list. So recovery will truncate it back to the original size | |
3135 | * if the machine crashes during the write. | |
3136 | * | |
3137 | */ | |
3138 | static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb, | |
3139 | const struct iovec *iov, loff_t offset, | |
3140 | unsigned long nr_segs) | |
3141 | { | |
3142 | struct file *file = iocb->ki_filp; | |
3143 | struct inode *inode = file->f_mapping->host; | |
3144 | ssize_t ret; | |
3145 | size_t count = iov_length(iov, nr_segs); | |
69c499d1 TT |
3146 | int overwrite = 0; |
3147 | get_block_t *get_block_func = NULL; | |
3148 | int dio_flags = 0; | |
4c0425ff | 3149 | loff_t final_size = offset + count; |
729f52c6 | 3150 | |
69c499d1 TT |
3151 | /* Use the old path for reads and writes beyond i_size. */ |
3152 | if (rw != WRITE || final_size > inode->i_size) | |
3153 | return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs); | |
4bd809db | 3154 | |
69c499d1 | 3155 | BUG_ON(iocb->private == NULL); |
4bd809db | 3156 | |
69c499d1 TT |
3157 | /* If we do a overwrite dio, i_mutex locking can be released */ |
3158 | overwrite = *((int *)iocb->private); | |
4bd809db | 3159 | |
69c499d1 TT |
3160 | if (overwrite) { |
3161 | atomic_inc(&inode->i_dio_count); | |
3162 | down_read(&EXT4_I(inode)->i_data_sem); | |
3163 | mutex_unlock(&inode->i_mutex); | |
3164 | } | |
8d5d02e6 | 3165 | |
69c499d1 TT |
3166 | /* |
3167 | * We could direct write to holes and fallocate. | |
3168 | * | |
3169 | * Allocated blocks to fill the hole are marked as | |
3170 | * uninitialized to prevent parallel buffered read to expose | |
3171 | * the stale data before DIO complete the data IO. | |
3172 | * | |
3173 | * As to previously fallocated extents, ext4 get_block will | |
3174 | * just simply mark the buffer mapped but still keep the | |
3175 | * extents uninitialized. | |
3176 | * | |
3177 | * For non AIO case, we will convert those unwritten extents | |
3178 | * to written after return back from blockdev_direct_IO. | |
3179 | * | |
3180 | * For async DIO, the conversion needs to be deferred when the | |
3181 | * IO is completed. The ext4 end_io callback function will be | |
3182 | * called to take care of the conversion work. Here for async | |
3183 | * case, we allocate an io_end structure to hook to the iocb. | |
3184 | */ | |
3185 | iocb->private = NULL; | |
3186 | ext4_inode_aio_set(inode, NULL); | |
3187 | if (!is_sync_kiocb(iocb)) { | |
a549984b | 3188 | ext4_io_end_t *io_end = ext4_init_io_end(inode, GFP_NOFS); |
69c499d1 TT |
3189 | if (!io_end) { |
3190 | ret = -ENOMEM; | |
3191 | goto retake_lock; | |
8b0f165f | 3192 | } |
69c499d1 | 3193 | io_end->flag |= EXT4_IO_END_DIRECT; |
a549984b | 3194 | iocb->private = io_end; |
8d5d02e6 | 3195 | /* |
69c499d1 TT |
3196 | * we save the io structure for current async direct |
3197 | * IO, so that later ext4_map_blocks() could flag the | |
3198 | * io structure whether there is a unwritten extents | |
3199 | * needs to be converted when IO is completed. | |
8d5d02e6 | 3200 | */ |
69c499d1 TT |
3201 | ext4_inode_aio_set(inode, io_end); |
3202 | } | |
4bd809db | 3203 | |
69c499d1 TT |
3204 | if (overwrite) { |
3205 | get_block_func = ext4_get_block_write_nolock; | |
3206 | } else { | |
3207 | get_block_func = ext4_get_block_write; | |
3208 | dio_flags = DIO_LOCKING; | |
3209 | } | |
3210 | ret = __blockdev_direct_IO(rw, iocb, inode, | |
3211 | inode->i_sb->s_bdev, iov, | |
3212 | offset, nr_segs, | |
3213 | get_block_func, | |
3214 | ext4_end_io_dio, | |
3215 | NULL, | |
3216 | dio_flags); | |
3217 | ||
a549984b TT |
3218 | if (iocb->private) |
3219 | ext4_inode_aio_set(inode, NULL); | |
69c499d1 | 3220 | /* |
a549984b TT |
3221 | * The io_end structure takes a reference to the inode, that |
3222 | * structure needs to be destroyed and the reference to the | |
3223 | * inode need to be dropped, when IO is complete, even with 0 | |
3224 | * byte write, or failed. | |
3225 | * | |
3226 | * In the successful AIO DIO case, the io_end structure will | |
3227 | * be destroyed and the reference to the inode will be dropped | |
3228 | * after the end_io call back function is called. | |
3229 | * | |
3230 | * In the case there is 0 byte write, or error case, since VFS | |
3231 | * direct IO won't invoke the end_io call back function, we | |
3232 | * need to free the end_io structure here. | |
69c499d1 | 3233 | */ |
a549984b TT |
3234 | if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) { |
3235 | ext4_free_io_end(iocb->private); | |
3236 | iocb->private = NULL; | |
3237 | } else if (ret > 0 && !overwrite && ext4_test_inode_state(inode, | |
69c499d1 TT |
3238 | EXT4_STATE_DIO_UNWRITTEN)) { |
3239 | int err; | |
3240 | /* | |
3241 | * for non AIO case, since the IO is already | |
3242 | * completed, we could do the conversion right here | |
3243 | */ | |
3244 | err = ext4_convert_unwritten_extents(inode, | |
3245 | offset, ret); | |
3246 | if (err < 0) | |
3247 | ret = err; | |
3248 | ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN); | |
3249 | } | |
4bd809db | 3250 | |
69c499d1 TT |
3251 | retake_lock: |
3252 | /* take i_mutex locking again if we do a ovewrite dio */ | |
3253 | if (overwrite) { | |
3254 | inode_dio_done(inode); | |
3255 | up_read(&EXT4_I(inode)->i_data_sem); | |
3256 | mutex_lock(&inode->i_mutex); | |
4c0425ff | 3257 | } |
8d5d02e6 | 3258 | |
69c499d1 | 3259 | return ret; |
4c0425ff MC |
3260 | } |
3261 | ||
3262 | static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb, | |
3263 | const struct iovec *iov, loff_t offset, | |
3264 | unsigned long nr_segs) | |
3265 | { | |
3266 | struct file *file = iocb->ki_filp; | |
3267 | struct inode *inode = file->f_mapping->host; | |
0562e0ba | 3268 | ssize_t ret; |
4c0425ff | 3269 | |
84ebd795 TT |
3270 | /* |
3271 | * If we are doing data journalling we don't support O_DIRECT | |
3272 | */ | |
3273 | if (ext4_should_journal_data(inode)) | |
3274 | return 0; | |
3275 | ||
46c7f254 TM |
3276 | /* Let buffer I/O handle the inline data case. */ |
3277 | if (ext4_has_inline_data(inode)) | |
3278 | return 0; | |
3279 | ||
0562e0ba | 3280 | trace_ext4_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw); |
12e9b892 | 3281 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
0562e0ba JZ |
3282 | ret = ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs); |
3283 | else | |
3284 | ret = ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs); | |
3285 | trace_ext4_direct_IO_exit(inode, offset, | |
3286 | iov_length(iov, nr_segs), rw, ret); | |
3287 | return ret; | |
4c0425ff MC |
3288 | } |
3289 | ||
ac27a0ec | 3290 | /* |
617ba13b | 3291 | * Pages can be marked dirty completely asynchronously from ext4's journalling |
ac27a0ec DK |
3292 | * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do |
3293 | * much here because ->set_page_dirty is called under VFS locks. The page is | |
3294 | * not necessarily locked. | |
3295 | * | |
3296 | * We cannot just dirty the page and leave attached buffers clean, because the | |
3297 | * buffers' dirty state is "definitive". We cannot just set the buffers dirty | |
3298 | * or jbddirty because all the journalling code will explode. | |
3299 | * | |
3300 | * So what we do is to mark the page "pending dirty" and next time writepage | |
3301 | * is called, propagate that into the buffers appropriately. | |
3302 | */ | |
617ba13b | 3303 | static int ext4_journalled_set_page_dirty(struct page *page) |
ac27a0ec DK |
3304 | { |
3305 | SetPageChecked(page); | |
3306 | return __set_page_dirty_nobuffers(page); | |
3307 | } | |
3308 | ||
74d553aa | 3309 | static const struct address_space_operations ext4_aops = { |
8ab22b9a HH |
3310 | .readpage = ext4_readpage, |
3311 | .readpages = ext4_readpages, | |
43ce1d23 | 3312 | .writepage = ext4_writepage, |
8ab22b9a | 3313 | .write_begin = ext4_write_begin, |
74d553aa | 3314 | .write_end = ext4_write_end, |
8ab22b9a HH |
3315 | .bmap = ext4_bmap, |
3316 | .invalidatepage = ext4_invalidatepage, | |
3317 | .releasepage = ext4_releasepage, | |
3318 | .direct_IO = ext4_direct_IO, | |
3319 | .migratepage = buffer_migrate_page, | |
3320 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 3321 | .error_remove_page = generic_error_remove_page, |
ac27a0ec DK |
3322 | }; |
3323 | ||
617ba13b | 3324 | static const struct address_space_operations ext4_journalled_aops = { |
8ab22b9a HH |
3325 | .readpage = ext4_readpage, |
3326 | .readpages = ext4_readpages, | |
43ce1d23 | 3327 | .writepage = ext4_writepage, |
8ab22b9a HH |
3328 | .write_begin = ext4_write_begin, |
3329 | .write_end = ext4_journalled_write_end, | |
3330 | .set_page_dirty = ext4_journalled_set_page_dirty, | |
3331 | .bmap = ext4_bmap, | |
4520fb3c | 3332 | .invalidatepage = ext4_journalled_invalidatepage, |
8ab22b9a | 3333 | .releasepage = ext4_releasepage, |
84ebd795 | 3334 | .direct_IO = ext4_direct_IO, |
8ab22b9a | 3335 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 3336 | .error_remove_page = generic_error_remove_page, |
ac27a0ec DK |
3337 | }; |
3338 | ||
64769240 | 3339 | static const struct address_space_operations ext4_da_aops = { |
8ab22b9a HH |
3340 | .readpage = ext4_readpage, |
3341 | .readpages = ext4_readpages, | |
43ce1d23 | 3342 | .writepage = ext4_writepage, |
8ab22b9a | 3343 | .writepages = ext4_da_writepages, |
8ab22b9a HH |
3344 | .write_begin = ext4_da_write_begin, |
3345 | .write_end = ext4_da_write_end, | |
3346 | .bmap = ext4_bmap, | |
3347 | .invalidatepage = ext4_da_invalidatepage, | |
3348 | .releasepage = ext4_releasepage, | |
3349 | .direct_IO = ext4_direct_IO, | |
3350 | .migratepage = buffer_migrate_page, | |
3351 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 3352 | .error_remove_page = generic_error_remove_page, |
64769240 AT |
3353 | }; |
3354 | ||
617ba13b | 3355 | void ext4_set_aops(struct inode *inode) |
ac27a0ec | 3356 | { |
3d2b1582 LC |
3357 | switch (ext4_inode_journal_mode(inode)) { |
3358 | case EXT4_INODE_ORDERED_DATA_MODE: | |
74d553aa | 3359 | ext4_set_inode_state(inode, EXT4_STATE_ORDERED_MODE); |
3d2b1582 LC |
3360 | break; |
3361 | case EXT4_INODE_WRITEBACK_DATA_MODE: | |
74d553aa | 3362 | ext4_clear_inode_state(inode, EXT4_STATE_ORDERED_MODE); |
3d2b1582 LC |
3363 | break; |
3364 | case EXT4_INODE_JOURNAL_DATA_MODE: | |
617ba13b | 3365 | inode->i_mapping->a_ops = &ext4_journalled_aops; |
74d553aa | 3366 | return; |
3d2b1582 LC |
3367 | default: |
3368 | BUG(); | |
3369 | } | |
74d553aa TT |
3370 | if (test_opt(inode->i_sb, DELALLOC)) |
3371 | inode->i_mapping->a_ops = &ext4_da_aops; | |
3372 | else | |
3373 | inode->i_mapping->a_ops = &ext4_aops; | |
ac27a0ec DK |
3374 | } |
3375 | ||
4e96b2db AH |
3376 | |
3377 | /* | |
3378 | * ext4_discard_partial_page_buffers() | |
3379 | * Wrapper function for ext4_discard_partial_page_buffers_no_lock. | |
3380 | * This function finds and locks the page containing the offset | |
3381 | * "from" and passes it to ext4_discard_partial_page_buffers_no_lock. | |
3382 | * Calling functions that already have the page locked should call | |
3383 | * ext4_discard_partial_page_buffers_no_lock directly. | |
3384 | */ | |
3385 | int ext4_discard_partial_page_buffers(handle_t *handle, | |
3386 | struct address_space *mapping, loff_t from, | |
3387 | loff_t length, int flags) | |
3388 | { | |
3389 | struct inode *inode = mapping->host; | |
3390 | struct page *page; | |
3391 | int err = 0; | |
3392 | ||
3393 | page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT, | |
3394 | mapping_gfp_mask(mapping) & ~__GFP_FS); | |
3395 | if (!page) | |
5129d05f | 3396 | return -ENOMEM; |
4e96b2db AH |
3397 | |
3398 | err = ext4_discard_partial_page_buffers_no_lock(handle, inode, page, | |
3399 | from, length, flags); | |
3400 | ||
3401 | unlock_page(page); | |
3402 | page_cache_release(page); | |
3403 | return err; | |
3404 | } | |
3405 | ||
3406 | /* | |
3407 | * ext4_discard_partial_page_buffers_no_lock() | |
3408 | * Zeros a page range of length 'length' starting from offset 'from'. | |
3409 | * Buffer heads that correspond to the block aligned regions of the | |
3410 | * zeroed range will be unmapped. Unblock aligned regions | |
3411 | * will have the corresponding buffer head mapped if needed so that | |
3412 | * that region of the page can be updated with the partial zero out. | |
3413 | * | |
3414 | * This function assumes that the page has already been locked. The | |
3415 | * The range to be discarded must be contained with in the given page. | |
3416 | * If the specified range exceeds the end of the page it will be shortened | |
3417 | * to the end of the page that corresponds to 'from'. This function is | |
3418 | * appropriate for updating a page and it buffer heads to be unmapped and | |
3419 | * zeroed for blocks that have been either released, or are going to be | |
3420 | * released. | |
3421 | * | |
3422 | * handle: The journal handle | |
3423 | * inode: The files inode | |
3424 | * page: A locked page that contains the offset "from" | |
4907cb7b | 3425 | * from: The starting byte offset (from the beginning of the file) |
4e96b2db AH |
3426 | * to begin discarding |
3427 | * len: The length of bytes to discard | |
3428 | * flags: Optional flags that may be used: | |
3429 | * | |
3430 | * EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED | |
3431 | * Only zero the regions of the page whose buffer heads | |
3432 | * have already been unmapped. This flag is appropriate | |
4907cb7b | 3433 | * for updating the contents of a page whose blocks may |
4e96b2db AH |
3434 | * have already been released, and we only want to zero |
3435 | * out the regions that correspond to those released blocks. | |
3436 | * | |
4907cb7b | 3437 | * Returns zero on success or negative on failure. |
4e96b2db | 3438 | */ |
5f163cc7 | 3439 | static int ext4_discard_partial_page_buffers_no_lock(handle_t *handle, |
4e96b2db AH |
3440 | struct inode *inode, struct page *page, loff_t from, |
3441 | loff_t length, int flags) | |
3442 | { | |
3443 | ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT; | |
3444 | unsigned int offset = from & (PAGE_CACHE_SIZE-1); | |
3445 | unsigned int blocksize, max, pos; | |
4e96b2db AH |
3446 | ext4_lblk_t iblock; |
3447 | struct buffer_head *bh; | |
3448 | int err = 0; | |
3449 | ||
3450 | blocksize = inode->i_sb->s_blocksize; | |
3451 | max = PAGE_CACHE_SIZE - offset; | |
3452 | ||
3453 | if (index != page->index) | |
3454 | return -EINVAL; | |
3455 | ||
3456 | /* | |
3457 | * correct length if it does not fall between | |
3458 | * 'from' and the end of the page | |
3459 | */ | |
3460 | if (length > max || length < 0) | |
3461 | length = max; | |
3462 | ||
3463 | iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits); | |
3464 | ||
093e6e36 YY |
3465 | if (!page_has_buffers(page)) |
3466 | create_empty_buffers(page, blocksize, 0); | |
4e96b2db AH |
3467 | |
3468 | /* Find the buffer that contains "offset" */ | |
3469 | bh = page_buffers(page); | |
3470 | pos = blocksize; | |
3471 | while (offset >= pos) { | |
3472 | bh = bh->b_this_page; | |
3473 | iblock++; | |
3474 | pos += blocksize; | |
3475 | } | |
3476 | ||
3477 | pos = offset; | |
3478 | while (pos < offset + length) { | |
e260daf2 YY |
3479 | unsigned int end_of_block, range_to_discard; |
3480 | ||
4e96b2db AH |
3481 | err = 0; |
3482 | ||
3483 | /* The length of space left to zero and unmap */ | |
3484 | range_to_discard = offset + length - pos; | |
3485 | ||
3486 | /* The length of space until the end of the block */ | |
3487 | end_of_block = blocksize - (pos & (blocksize-1)); | |
3488 | ||
3489 | /* | |
3490 | * Do not unmap or zero past end of block | |
3491 | * for this buffer head | |
3492 | */ | |
3493 | if (range_to_discard > end_of_block) | |
3494 | range_to_discard = end_of_block; | |
3495 | ||
3496 | ||
3497 | /* | |
3498 | * Skip this buffer head if we are only zeroing unampped | |
3499 | * regions of the page | |
3500 | */ | |
3501 | if (flags & EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED && | |
3502 | buffer_mapped(bh)) | |
3503 | goto next; | |
3504 | ||
3505 | /* If the range is block aligned, unmap */ | |
3506 | if (range_to_discard == blocksize) { | |
3507 | clear_buffer_dirty(bh); | |
3508 | bh->b_bdev = NULL; | |
3509 | clear_buffer_mapped(bh); | |
3510 | clear_buffer_req(bh); | |
3511 | clear_buffer_new(bh); | |
3512 | clear_buffer_delay(bh); | |
3513 | clear_buffer_unwritten(bh); | |
3514 | clear_buffer_uptodate(bh); | |
3515 | zero_user(page, pos, range_to_discard); | |
3516 | BUFFER_TRACE(bh, "Buffer discarded"); | |
3517 | goto next; | |
3518 | } | |
3519 | ||
3520 | /* | |
3521 | * If this block is not completely contained in the range | |
3522 | * to be discarded, then it is not going to be released. Because | |
3523 | * we need to keep this block, we need to make sure this part | |
3524 | * of the page is uptodate before we modify it by writeing | |
3525 | * partial zeros on it. | |
3526 | */ | |
3527 | if (!buffer_mapped(bh)) { | |
3528 | /* | |
3529 | * Buffer head must be mapped before we can read | |
3530 | * from the block | |
3531 | */ | |
3532 | BUFFER_TRACE(bh, "unmapped"); | |
3533 | ext4_get_block(inode, iblock, bh, 0); | |
3534 | /* unmapped? It's a hole - nothing to do */ | |
3535 | if (!buffer_mapped(bh)) { | |
3536 | BUFFER_TRACE(bh, "still unmapped"); | |
3537 | goto next; | |
3538 | } | |
3539 | } | |
3540 | ||
3541 | /* Ok, it's mapped. Make sure it's up-to-date */ | |
3542 | if (PageUptodate(page)) | |
3543 | set_buffer_uptodate(bh); | |
3544 | ||
3545 | if (!buffer_uptodate(bh)) { | |
3546 | err = -EIO; | |
3547 | ll_rw_block(READ, 1, &bh); | |
3548 | wait_on_buffer(bh); | |
3549 | /* Uhhuh. Read error. Complain and punt.*/ | |
3550 | if (!buffer_uptodate(bh)) | |
3551 | goto next; | |
3552 | } | |
3553 | ||
3554 | if (ext4_should_journal_data(inode)) { | |
3555 | BUFFER_TRACE(bh, "get write access"); | |
3556 | err = ext4_journal_get_write_access(handle, bh); | |
3557 | if (err) | |
3558 | goto next; | |
3559 | } | |
3560 | ||
3561 | zero_user(page, pos, range_to_discard); | |
3562 | ||
3563 | err = 0; | |
3564 | if (ext4_should_journal_data(inode)) { | |
3565 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
decbd919 | 3566 | } else |
4e96b2db | 3567 | mark_buffer_dirty(bh); |
4e96b2db AH |
3568 | |
3569 | BUFFER_TRACE(bh, "Partial buffer zeroed"); | |
3570 | next: | |
3571 | bh = bh->b_this_page; | |
3572 | iblock++; | |
3573 | pos += range_to_discard; | |
3574 | } | |
3575 | ||
3576 | return err; | |
3577 | } | |
3578 | ||
91ef4caf DG |
3579 | int ext4_can_truncate(struct inode *inode) |
3580 | { | |
91ef4caf DG |
3581 | if (S_ISREG(inode->i_mode)) |
3582 | return 1; | |
3583 | if (S_ISDIR(inode->i_mode)) | |
3584 | return 1; | |
3585 | if (S_ISLNK(inode->i_mode)) | |
3586 | return !ext4_inode_is_fast_symlink(inode); | |
3587 | return 0; | |
3588 | } | |
3589 | ||
a4bb6b64 AH |
3590 | /* |
3591 | * ext4_punch_hole: punches a hole in a file by releaseing the blocks | |
3592 | * associated with the given offset and length | |
3593 | * | |
3594 | * @inode: File inode | |
3595 | * @offset: The offset where the hole will begin | |
3596 | * @len: The length of the hole | |
3597 | * | |
4907cb7b | 3598 | * Returns: 0 on success or negative on failure |
a4bb6b64 AH |
3599 | */ |
3600 | ||
3601 | int ext4_punch_hole(struct file *file, loff_t offset, loff_t length) | |
3602 | { | |
4b9e9796 | 3603 | #if 0 |
496ad9aa | 3604 | struct inode *inode = file_inode(file); |
26a4c0c6 TT |
3605 | struct super_block *sb = inode->i_sb; |
3606 | ext4_lblk_t first_block, stop_block; | |
3607 | struct address_space *mapping = inode->i_mapping; | |
3608 | loff_t first_page, last_page, page_len; | |
3609 | loff_t first_page_offset, last_page_offset; | |
3610 | handle_t *handle; | |
3611 | unsigned int credits; | |
3612 | int ret = 0; | |
3613 | ||
a4bb6b64 | 3614 | if (!S_ISREG(inode->i_mode)) |
73355192 | 3615 | return -EOPNOTSUPP; |
a4bb6b64 | 3616 | |
26a4c0c6 | 3617 | if (EXT4_SB(sb)->s_cluster_ratio > 1) { |
bab08ab9 | 3618 | /* TODO: Add support for bigalloc file systems */ |
73355192 | 3619 | return -EOPNOTSUPP; |
bab08ab9 TT |
3620 | } |
3621 | ||
aaddea81 ZL |
3622 | trace_ext4_punch_hole(inode, offset, length); |
3623 | ||
26a4c0c6 TT |
3624 | /* |
3625 | * Write out all dirty pages to avoid race conditions | |
3626 | * Then release them. | |
3627 | */ | |
3628 | if (mapping->nrpages && mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { | |
3629 | ret = filemap_write_and_wait_range(mapping, offset, | |
3630 | offset + length - 1); | |
3631 | if (ret) | |
3632 | return ret; | |
3633 | } | |
3634 | ||
3635 | mutex_lock(&inode->i_mutex); | |
3636 | /* It's not possible punch hole on append only file */ | |
3637 | if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) { | |
3638 | ret = -EPERM; | |
3639 | goto out_mutex; | |
3640 | } | |
3641 | if (IS_SWAPFILE(inode)) { | |
3642 | ret = -ETXTBSY; | |
3643 | goto out_mutex; | |
3644 | } | |
3645 | ||
3646 | /* No need to punch hole beyond i_size */ | |
3647 | if (offset >= inode->i_size) | |
3648 | goto out_mutex; | |
3649 | ||
3650 | /* | |
3651 | * If the hole extends beyond i_size, set the hole | |
3652 | * to end after the page that contains i_size | |
3653 | */ | |
3654 | if (offset + length > inode->i_size) { | |
3655 | length = inode->i_size + | |
3656 | PAGE_CACHE_SIZE - (inode->i_size & (PAGE_CACHE_SIZE - 1)) - | |
3657 | offset; | |
3658 | } | |
3659 | ||
3660 | first_page = (offset + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; | |
3661 | last_page = (offset + length) >> PAGE_CACHE_SHIFT; | |
3662 | ||
3663 | first_page_offset = first_page << PAGE_CACHE_SHIFT; | |
3664 | last_page_offset = last_page << PAGE_CACHE_SHIFT; | |
3665 | ||
3666 | /* Now release the pages */ | |
3667 | if (last_page_offset > first_page_offset) { | |
3668 | truncate_pagecache_range(inode, first_page_offset, | |
3669 | last_page_offset - 1); | |
3670 | } | |
3671 | ||
3672 | /* Wait all existing dio workers, newcomers will block on i_mutex */ | |
3673 | ext4_inode_block_unlocked_dio(inode); | |
3674 | ret = ext4_flush_unwritten_io(inode); | |
3675 | if (ret) | |
3676 | goto out_dio; | |
3677 | inode_dio_wait(inode); | |
3678 | ||
3679 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
3680 | credits = ext4_writepage_trans_blocks(inode); | |
3681 | else | |
3682 | credits = ext4_blocks_for_truncate(inode); | |
3683 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); | |
3684 | if (IS_ERR(handle)) { | |
3685 | ret = PTR_ERR(handle); | |
3686 | ext4_std_error(sb, ret); | |
3687 | goto out_dio; | |
3688 | } | |
3689 | ||
3690 | /* | |
3691 | * Now we need to zero out the non-page-aligned data in the | |
3692 | * pages at the start and tail of the hole, and unmap the | |
3693 | * buffer heads for the block aligned regions of the page that | |
3694 | * were completely zeroed. | |
3695 | */ | |
3696 | if (first_page > last_page) { | |
3697 | /* | |
3698 | * If the file space being truncated is contained | |
3699 | * within a page just zero out and unmap the middle of | |
3700 | * that page | |
3701 | */ | |
3702 | ret = ext4_discard_partial_page_buffers(handle, | |
3703 | mapping, offset, length, 0); | |
3704 | ||
3705 | if (ret) | |
3706 | goto out_stop; | |
3707 | } else { | |
3708 | /* | |
3709 | * zero out and unmap the partial page that contains | |
3710 | * the start of the hole | |
3711 | */ | |
3712 | page_len = first_page_offset - offset; | |
3713 | if (page_len > 0) { | |
3714 | ret = ext4_discard_partial_page_buffers(handle, mapping, | |
3715 | offset, page_len, 0); | |
3716 | if (ret) | |
3717 | goto out_stop; | |
3718 | } | |
3719 | ||
3720 | /* | |
3721 | * zero out and unmap the partial page that contains | |
3722 | * the end of the hole | |
3723 | */ | |
3724 | page_len = offset + length - last_page_offset; | |
3725 | if (page_len > 0) { | |
3726 | ret = ext4_discard_partial_page_buffers(handle, mapping, | |
3727 | last_page_offset, page_len, 0); | |
3728 | if (ret) | |
3729 | goto out_stop; | |
3730 | } | |
3731 | } | |
3732 | ||
3733 | /* | |
3734 | * If i_size is contained in the last page, we need to | |
3735 | * unmap and zero the partial page after i_size | |
3736 | */ | |
3737 | if (inode->i_size >> PAGE_CACHE_SHIFT == last_page && | |
3738 | inode->i_size % PAGE_CACHE_SIZE != 0) { | |
3739 | page_len = PAGE_CACHE_SIZE - | |
3740 | (inode->i_size & (PAGE_CACHE_SIZE - 1)); | |
3741 | ||
3742 | if (page_len > 0) { | |
3743 | ret = ext4_discard_partial_page_buffers(handle, | |
3744 | mapping, inode->i_size, page_len, 0); | |
3745 | ||
3746 | if (ret) | |
3747 | goto out_stop; | |
3748 | } | |
3749 | } | |
3750 | ||
3751 | first_block = (offset + sb->s_blocksize - 1) >> | |
3752 | EXT4_BLOCK_SIZE_BITS(sb); | |
3753 | stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb); | |
3754 | ||
3755 | /* If there are no blocks to remove, return now */ | |
3756 | if (first_block >= stop_block) | |
3757 | goto out_stop; | |
3758 | ||
3759 | down_write(&EXT4_I(inode)->i_data_sem); | |
3760 | ext4_discard_preallocations(inode); | |
3761 | ||
3762 | ret = ext4_es_remove_extent(inode, first_block, | |
3763 | stop_block - first_block); | |
3764 | if (ret) { | |
3765 | up_write(&EXT4_I(inode)->i_data_sem); | |
3766 | goto out_stop; | |
3767 | } | |
3768 | ||
3769 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
3770 | ret = ext4_ext_remove_space(inode, first_block, | |
3771 | stop_block - 1); | |
3772 | else | |
3773 | ret = ext4_free_hole_blocks(handle, inode, first_block, | |
3774 | stop_block); | |
3775 | ||
3776 | ext4_discard_preallocations(inode); | |
819c4920 | 3777 | up_write(&EXT4_I(inode)->i_data_sem); |
26a4c0c6 TT |
3778 | if (IS_SYNC(inode)) |
3779 | ext4_handle_sync(handle); | |
26a4c0c6 TT |
3780 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); |
3781 | ext4_mark_inode_dirty(handle, inode); | |
3782 | out_stop: | |
3783 | ext4_journal_stop(handle); | |
3784 | out_dio: | |
3785 | ext4_inode_resume_unlocked_dio(inode); | |
3786 | out_mutex: | |
3787 | mutex_unlock(&inode->i_mutex); | |
3788 | return ret; | |
4b9e9796 S |
3789 | #else |
3790 | /* | |
3791 | * Disabled as per b/28760453 | |
3792 | */ | |
3793 | return -EOPNOTSUPP; | |
3794 | #endif | |
a4bb6b64 AH |
3795 | } |
3796 | ||
ac27a0ec | 3797 | /* |
617ba13b | 3798 | * ext4_truncate() |
ac27a0ec | 3799 | * |
617ba13b MC |
3800 | * We block out ext4_get_block() block instantiations across the entire |
3801 | * transaction, and VFS/VM ensures that ext4_truncate() cannot run | |
ac27a0ec DK |
3802 | * simultaneously on behalf of the same inode. |
3803 | * | |
42b2aa86 | 3804 | * As we work through the truncate and commit bits of it to the journal there |
ac27a0ec DK |
3805 | * is one core, guiding principle: the file's tree must always be consistent on |
3806 | * disk. We must be able to restart the truncate after a crash. | |
3807 | * | |
3808 | * The file's tree may be transiently inconsistent in memory (although it | |
3809 | * probably isn't), but whenever we close off and commit a journal transaction, | |
3810 | * the contents of (the filesystem + the journal) must be consistent and | |
3811 | * restartable. It's pretty simple, really: bottom up, right to left (although | |
3812 | * left-to-right works OK too). | |
3813 | * | |
3814 | * Note that at recovery time, journal replay occurs *before* the restart of | |
3815 | * truncate against the orphan inode list. | |
3816 | * | |
3817 | * The committed inode has the new, desired i_size (which is the same as | |
617ba13b | 3818 | * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see |
ac27a0ec | 3819 | * that this inode's truncate did not complete and it will again call |
617ba13b MC |
3820 | * ext4_truncate() to have another go. So there will be instantiated blocks |
3821 | * to the right of the truncation point in a crashed ext4 filesystem. But | |
ac27a0ec | 3822 | * that's fine - as long as they are linked from the inode, the post-crash |
617ba13b | 3823 | * ext4_truncate() run will find them and release them. |
ac27a0ec | 3824 | */ |
617ba13b | 3825 | void ext4_truncate(struct inode *inode) |
ac27a0ec | 3826 | { |
819c4920 TT |
3827 | struct ext4_inode_info *ei = EXT4_I(inode); |
3828 | unsigned int credits; | |
3829 | handle_t *handle; | |
3830 | struct address_space *mapping = inode->i_mapping; | |
3831 | loff_t page_len; | |
3832 | ||
19b5ef61 TT |
3833 | /* |
3834 | * There is a possibility that we're either freeing the inode | |
3835 | * or it completely new indode. In those cases we might not | |
3836 | * have i_mutex locked because it's not necessary. | |
3837 | */ | |
3838 | if (!(inode->i_state & (I_NEW|I_FREEING))) | |
3839 | WARN_ON(!mutex_is_locked(&inode->i_mutex)); | |
0562e0ba JZ |
3840 | trace_ext4_truncate_enter(inode); |
3841 | ||
91ef4caf | 3842 | if (!ext4_can_truncate(inode)) |
ac27a0ec DK |
3843 | return; |
3844 | ||
12e9b892 | 3845 | ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS); |
c8d46e41 | 3846 | |
5534fb5b | 3847 | if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC)) |
19f5fb7a | 3848 | ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE); |
7d8f9f7d | 3849 | |
aef1c851 TM |
3850 | if (ext4_has_inline_data(inode)) { |
3851 | int has_inline = 1; | |
3852 | ||
3853 | ext4_inline_data_truncate(inode, &has_inline); | |
3854 | if (has_inline) | |
3855 | return; | |
3856 | } | |
3857 | ||
819c4920 TT |
3858 | /* |
3859 | * finish any pending end_io work so we won't run the risk of | |
3860 | * converting any truncated blocks to initialized later | |
3861 | */ | |
3862 | ext4_flush_unwritten_io(inode); | |
3863 | ||
3864 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
3865 | credits = ext4_writepage_trans_blocks(inode); | |
3866 | else | |
3867 | credits = ext4_blocks_for_truncate(inode); | |
3868 | ||
3869 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); | |
3870 | if (IS_ERR(handle)) { | |
3871 | ext4_std_error(inode->i_sb, PTR_ERR(handle)); | |
3872 | return; | |
3873 | } | |
3874 | ||
3875 | if (inode->i_size % PAGE_CACHE_SIZE != 0) { | |
3876 | page_len = PAGE_CACHE_SIZE - | |
3877 | (inode->i_size & (PAGE_CACHE_SIZE - 1)); | |
3878 | ||
3879 | if (ext4_discard_partial_page_buffers(handle, | |
3880 | mapping, inode->i_size, page_len, 0)) | |
3881 | goto out_stop; | |
3882 | } | |
3883 | ||
3884 | /* | |
3885 | * We add the inode to the orphan list, so that if this | |
3886 | * truncate spans multiple transactions, and we crash, we will | |
3887 | * resume the truncate when the filesystem recovers. It also | |
3888 | * marks the inode dirty, to catch the new size. | |
3889 | * | |
3890 | * Implication: the file must always be in a sane, consistent | |
3891 | * truncatable state while each transaction commits. | |
3892 | */ | |
3893 | if (ext4_orphan_add(handle, inode)) | |
3894 | goto out_stop; | |
3895 | ||
3896 | down_write(&EXT4_I(inode)->i_data_sem); | |
3897 | ||
3898 | ext4_discard_preallocations(inode); | |
3899 | ||
ff9893dc | 3900 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
819c4920 | 3901 | ext4_ext_truncate(handle, inode); |
ff9893dc | 3902 | else |
819c4920 TT |
3903 | ext4_ind_truncate(handle, inode); |
3904 | ||
3905 | up_write(&ei->i_data_sem); | |
3906 | ||
3907 | if (IS_SYNC(inode)) | |
3908 | ext4_handle_sync(handle); | |
3909 | ||
3910 | out_stop: | |
3911 | /* | |
3912 | * If this was a simple ftruncate() and the file will remain alive, | |
3913 | * then we need to clear up the orphan record which we created above. | |
3914 | * However, if this was a real unlink then we were called by | |
3915 | * ext4_delete_inode(), and we allow that function to clean up the | |
3916 | * orphan info for us. | |
3917 | */ | |
3918 | if (inode->i_nlink) | |
3919 | ext4_orphan_del(handle, inode); | |
3920 | ||
3921 | inode->i_mtime = inode->i_ctime = ext4_current_time(inode); | |
3922 | ext4_mark_inode_dirty(handle, inode); | |
3923 | ext4_journal_stop(handle); | |
ac27a0ec | 3924 | |
0562e0ba | 3925 | trace_ext4_truncate_exit(inode); |
ac27a0ec DK |
3926 | } |
3927 | ||
ac27a0ec | 3928 | /* |
617ba13b | 3929 | * ext4_get_inode_loc returns with an extra refcount against the inode's |
ac27a0ec DK |
3930 | * underlying buffer_head on success. If 'in_mem' is true, we have all |
3931 | * data in memory that is needed to recreate the on-disk version of this | |
3932 | * inode. | |
3933 | */ | |
617ba13b MC |
3934 | static int __ext4_get_inode_loc(struct inode *inode, |
3935 | struct ext4_iloc *iloc, int in_mem) | |
ac27a0ec | 3936 | { |
240799cd TT |
3937 | struct ext4_group_desc *gdp; |
3938 | struct buffer_head *bh; | |
3939 | struct super_block *sb = inode->i_sb; | |
3940 | ext4_fsblk_t block; | |
3941 | int inodes_per_block, inode_offset; | |
3942 | ||
3a06d778 | 3943 | iloc->bh = NULL; |
240799cd TT |
3944 | if (!ext4_valid_inum(sb, inode->i_ino)) |
3945 | return -EIO; | |
ac27a0ec | 3946 | |
240799cd TT |
3947 | iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb); |
3948 | gdp = ext4_get_group_desc(sb, iloc->block_group, NULL); | |
3949 | if (!gdp) | |
ac27a0ec DK |
3950 | return -EIO; |
3951 | ||
240799cd TT |
3952 | /* |
3953 | * Figure out the offset within the block group inode table | |
3954 | */ | |
00d09882 | 3955 | inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; |
240799cd TT |
3956 | inode_offset = ((inode->i_ino - 1) % |
3957 | EXT4_INODES_PER_GROUP(sb)); | |
3958 | block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block); | |
3959 | iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb); | |
3960 | ||
3961 | bh = sb_getblk(sb, block); | |
aebf0243 | 3962 | if (unlikely(!bh)) |
860d21e2 | 3963 | return -ENOMEM; |
ac27a0ec DK |
3964 | if (!buffer_uptodate(bh)) { |
3965 | lock_buffer(bh); | |
9c83a923 HK |
3966 | |
3967 | /* | |
3968 | * If the buffer has the write error flag, we have failed | |
3969 | * to write out another inode in the same block. In this | |
3970 | * case, we don't have to read the block because we may | |
3971 | * read the old inode data successfully. | |
3972 | */ | |
3973 | if (buffer_write_io_error(bh) && !buffer_uptodate(bh)) | |
3974 | set_buffer_uptodate(bh); | |
3975 | ||
ac27a0ec DK |
3976 | if (buffer_uptodate(bh)) { |
3977 | /* someone brought it uptodate while we waited */ | |
3978 | unlock_buffer(bh); | |
3979 | goto has_buffer; | |
3980 | } | |
3981 | ||
3982 | /* | |
3983 | * If we have all information of the inode in memory and this | |
3984 | * is the only valid inode in the block, we need not read the | |
3985 | * block. | |
3986 | */ | |
3987 | if (in_mem) { | |
3988 | struct buffer_head *bitmap_bh; | |
240799cd | 3989 | int i, start; |
ac27a0ec | 3990 | |
240799cd | 3991 | start = inode_offset & ~(inodes_per_block - 1); |
ac27a0ec | 3992 | |
240799cd TT |
3993 | /* Is the inode bitmap in cache? */ |
3994 | bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp)); | |
aebf0243 | 3995 | if (unlikely(!bitmap_bh)) |
ac27a0ec DK |
3996 | goto make_io; |
3997 | ||
3998 | /* | |
3999 | * If the inode bitmap isn't in cache then the | |
4000 | * optimisation may end up performing two reads instead | |
4001 | * of one, so skip it. | |
4002 | */ | |
4003 | if (!buffer_uptodate(bitmap_bh)) { | |
4004 | brelse(bitmap_bh); | |
4005 | goto make_io; | |
4006 | } | |
240799cd | 4007 | for (i = start; i < start + inodes_per_block; i++) { |
ac27a0ec DK |
4008 | if (i == inode_offset) |
4009 | continue; | |
617ba13b | 4010 | if (ext4_test_bit(i, bitmap_bh->b_data)) |
ac27a0ec DK |
4011 | break; |
4012 | } | |
4013 | brelse(bitmap_bh); | |
240799cd | 4014 | if (i == start + inodes_per_block) { |
ac27a0ec DK |
4015 | /* all other inodes are free, so skip I/O */ |
4016 | memset(bh->b_data, 0, bh->b_size); | |
4017 | set_buffer_uptodate(bh); | |
4018 | unlock_buffer(bh); | |
4019 | goto has_buffer; | |
4020 | } | |
4021 | } | |
4022 | ||
4023 | make_io: | |
240799cd TT |
4024 | /* |
4025 | * If we need to do any I/O, try to pre-readahead extra | |
4026 | * blocks from the inode table. | |
4027 | */ | |
4028 | if (EXT4_SB(sb)->s_inode_readahead_blks) { | |
4029 | ext4_fsblk_t b, end, table; | |
4030 | unsigned num; | |
0d606e2c | 4031 | __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks; |
240799cd TT |
4032 | |
4033 | table = ext4_inode_table(sb, gdp); | |
b713a5ec | 4034 | /* s_inode_readahead_blks is always a power of 2 */ |
0d606e2c | 4035 | b = block & ~((ext4_fsblk_t) ra_blks - 1); |
240799cd TT |
4036 | if (table > b) |
4037 | b = table; | |
0d606e2c | 4038 | end = b + ra_blks; |
240799cd | 4039 | num = EXT4_INODES_PER_GROUP(sb); |
feb0ab32 | 4040 | if (ext4_has_group_desc_csum(sb)) |
560671a0 | 4041 | num -= ext4_itable_unused_count(sb, gdp); |
240799cd TT |
4042 | table += num / inodes_per_block; |
4043 | if (end > table) | |
4044 | end = table; | |
4045 | while (b <= end) | |
4046 | sb_breadahead(sb, b++); | |
4047 | } | |
4048 | ||
ac27a0ec DK |
4049 | /* |
4050 | * There are other valid inodes in the buffer, this inode | |
4051 | * has in-inode xattrs, or we don't have this inode in memory. | |
4052 | * Read the block from disk. | |
4053 | */ | |
0562e0ba | 4054 | trace_ext4_load_inode(inode); |
ac27a0ec DK |
4055 | get_bh(bh); |
4056 | bh->b_end_io = end_buffer_read_sync; | |
6fa3eb70 S |
4057 | #ifdef FEATURE_STORAGE_META_LOG |
4058 | if( bh && bh->b_bdev && bh->b_bdev->bd_disk) | |
4059 | set_metadata_rw_status(bh->b_bdev->bd_disk->first_minor, WAIT_READ_CNT); | |
4060 | #endif | |
65299a3b | 4061 | submit_bh(READ | REQ_META | REQ_PRIO, bh); |
ac27a0ec DK |
4062 | wait_on_buffer(bh); |
4063 | if (!buffer_uptodate(bh)) { | |
c398eda0 TT |
4064 | EXT4_ERROR_INODE_BLOCK(inode, block, |
4065 | "unable to read itable block"); | |
ac27a0ec DK |
4066 | brelse(bh); |
4067 | return -EIO; | |
4068 | } | |
4069 | } | |
4070 | has_buffer: | |
4071 | iloc->bh = bh; | |
4072 | return 0; | |
4073 | } | |
4074 | ||
617ba13b | 4075 | int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec DK |
4076 | { |
4077 | /* We have all inode data except xattrs in memory here. */ | |
617ba13b | 4078 | return __ext4_get_inode_loc(inode, iloc, |
19f5fb7a | 4079 | !ext4_test_inode_state(inode, EXT4_STATE_XATTR)); |
ac27a0ec DK |
4080 | } |
4081 | ||
617ba13b | 4082 | void ext4_set_inode_flags(struct inode *inode) |
ac27a0ec | 4083 | { |
617ba13b | 4084 | unsigned int flags = EXT4_I(inode)->i_flags; |
0a0ae7b3 | 4085 | unsigned int new_fl = 0; |
ac27a0ec | 4086 | |
617ba13b | 4087 | if (flags & EXT4_SYNC_FL) |
0a0ae7b3 | 4088 | new_fl |= S_SYNC; |
617ba13b | 4089 | if (flags & EXT4_APPEND_FL) |
0a0ae7b3 | 4090 | new_fl |= S_APPEND; |
617ba13b | 4091 | if (flags & EXT4_IMMUTABLE_FL) |
0a0ae7b3 | 4092 | new_fl |= S_IMMUTABLE; |
617ba13b | 4093 | if (flags & EXT4_NOATIME_FL) |
0a0ae7b3 | 4094 | new_fl |= S_NOATIME; |
617ba13b | 4095 | if (flags & EXT4_DIRSYNC_FL) |
0a0ae7b3 TT |
4096 | new_fl |= S_DIRSYNC; |
4097 | set_mask_bits(&inode->i_flags, | |
4098 | S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC, new_fl); | |
ac27a0ec DK |
4099 | } |
4100 | ||
ff9ddf7e JK |
4101 | /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */ |
4102 | void ext4_get_inode_flags(struct ext4_inode_info *ei) | |
4103 | { | |
84a8dce2 DM |
4104 | unsigned int vfs_fl; |
4105 | unsigned long old_fl, new_fl; | |
4106 | ||
4107 | do { | |
4108 | vfs_fl = ei->vfs_inode.i_flags; | |
4109 | old_fl = ei->i_flags; | |
4110 | new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL| | |
4111 | EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL| | |
4112 | EXT4_DIRSYNC_FL); | |
4113 | if (vfs_fl & S_SYNC) | |
4114 | new_fl |= EXT4_SYNC_FL; | |
4115 | if (vfs_fl & S_APPEND) | |
4116 | new_fl |= EXT4_APPEND_FL; | |
4117 | if (vfs_fl & S_IMMUTABLE) | |
4118 | new_fl |= EXT4_IMMUTABLE_FL; | |
4119 | if (vfs_fl & S_NOATIME) | |
4120 | new_fl |= EXT4_NOATIME_FL; | |
4121 | if (vfs_fl & S_DIRSYNC) | |
4122 | new_fl |= EXT4_DIRSYNC_FL; | |
4123 | } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl); | |
ff9ddf7e | 4124 | } |
de9a55b8 | 4125 | |
0fc1b451 | 4126 | static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode, |
de9a55b8 | 4127 | struct ext4_inode_info *ei) |
0fc1b451 AK |
4128 | { |
4129 | blkcnt_t i_blocks ; | |
8180a562 AK |
4130 | struct inode *inode = &(ei->vfs_inode); |
4131 | struct super_block *sb = inode->i_sb; | |
0fc1b451 AK |
4132 | |
4133 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, | |
4134 | EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) { | |
4135 | /* we are using combined 48 bit field */ | |
4136 | i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 | | |
4137 | le32_to_cpu(raw_inode->i_blocks_lo); | |
07a03824 | 4138 | if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) { |
8180a562 AK |
4139 | /* i_blocks represent file system block size */ |
4140 | return i_blocks << (inode->i_blkbits - 9); | |
4141 | } else { | |
4142 | return i_blocks; | |
4143 | } | |
0fc1b451 AK |
4144 | } else { |
4145 | return le32_to_cpu(raw_inode->i_blocks_lo); | |
4146 | } | |
4147 | } | |
ff9ddf7e | 4148 | |
152a7b0a TM |
4149 | static inline void ext4_iget_extra_inode(struct inode *inode, |
4150 | struct ext4_inode *raw_inode, | |
4151 | struct ext4_inode_info *ei) | |
4152 | { | |
4153 | __le32 *magic = (void *)raw_inode + | |
4154 | EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize; | |
67cf5b09 | 4155 | if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) { |
152a7b0a | 4156 | ext4_set_inode_state(inode, EXT4_STATE_XATTR); |
67cf5b09 | 4157 | ext4_find_inline_data_nolock(inode); |
f19d5870 TM |
4158 | } else |
4159 | EXT4_I(inode)->i_inline_off = 0; | |
152a7b0a TM |
4160 | } |
4161 | ||
1d1fe1ee | 4162 | struct inode *ext4_iget(struct super_block *sb, unsigned long ino) |
ac27a0ec | 4163 | { |
617ba13b MC |
4164 | struct ext4_iloc iloc; |
4165 | struct ext4_inode *raw_inode; | |
1d1fe1ee | 4166 | struct ext4_inode_info *ei; |
1d1fe1ee | 4167 | struct inode *inode; |
b436b9be | 4168 | journal_t *journal = EXT4_SB(sb)->s_journal; |
1d1fe1ee | 4169 | long ret; |
ac27a0ec | 4170 | int block; |
08cefc7a EB |
4171 | uid_t i_uid; |
4172 | gid_t i_gid; | |
ac27a0ec | 4173 | |
1d1fe1ee DH |
4174 | inode = iget_locked(sb, ino); |
4175 | if (!inode) | |
4176 | return ERR_PTR(-ENOMEM); | |
4177 | if (!(inode->i_state & I_NEW)) | |
4178 | return inode; | |
4179 | ||
4180 | ei = EXT4_I(inode); | |
7dc57615 | 4181 | iloc.bh = NULL; |
ac27a0ec | 4182 | |
1d1fe1ee DH |
4183 | ret = __ext4_get_inode_loc(inode, &iloc, 0); |
4184 | if (ret < 0) | |
ac27a0ec | 4185 | goto bad_inode; |
617ba13b | 4186 | raw_inode = ext4_raw_inode(&iloc); |
814525f4 DW |
4187 | |
4188 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { | |
4189 | ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize); | |
4190 | if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > | |
4191 | EXT4_INODE_SIZE(inode->i_sb)) { | |
4192 | EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)", | |
4193 | EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize, | |
4194 | EXT4_INODE_SIZE(inode->i_sb)); | |
4195 | ret = -EIO; | |
4196 | goto bad_inode; | |
4197 | } | |
4198 | } else | |
4199 | ei->i_extra_isize = 0; | |
4200 | ||
4201 | /* Precompute checksum seed for inode metadata */ | |
4202 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, | |
4203 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) { | |
4204 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
4205 | __u32 csum; | |
4206 | __le32 inum = cpu_to_le32(inode->i_ino); | |
4207 | __le32 gen = raw_inode->i_generation; | |
4208 | csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum, | |
4209 | sizeof(inum)); | |
4210 | ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen, | |
4211 | sizeof(gen)); | |
4212 | } | |
4213 | ||
4214 | if (!ext4_inode_csum_verify(inode, raw_inode, ei)) { | |
4215 | EXT4_ERROR_INODE(inode, "checksum invalid"); | |
4216 | ret = -EIO; | |
4217 | goto bad_inode; | |
4218 | } | |
4219 | ||
ac27a0ec | 4220 | inode->i_mode = le16_to_cpu(raw_inode->i_mode); |
08cefc7a EB |
4221 | i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low); |
4222 | i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low); | |
af5bc92d | 4223 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
08cefc7a EB |
4224 | i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16; |
4225 | i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16; | |
ac27a0ec | 4226 | } |
08cefc7a EB |
4227 | i_uid_write(inode, i_uid); |
4228 | i_gid_write(inode, i_gid); | |
bfe86848 | 4229 | set_nlink(inode, le16_to_cpu(raw_inode->i_links_count)); |
ac27a0ec | 4230 | |
353eb83c | 4231 | ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */ |
67cf5b09 | 4232 | ei->i_inline_off = 0; |
ac27a0ec DK |
4233 | ei->i_dir_start_lookup = 0; |
4234 | ei->i_dtime = le32_to_cpu(raw_inode->i_dtime); | |
4235 | /* We now have enough fields to check if the inode was active or not. | |
4236 | * This is needed because nfsd might try to access dead inodes | |
4237 | * the test is that same one that e2fsck uses | |
4238 | * NeilBrown 1999oct15 | |
4239 | */ | |
4240 | if (inode->i_nlink == 0) { | |
393d1d1d DTB |
4241 | if ((inode->i_mode == 0 || |
4242 | !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) && | |
4243 | ino != EXT4_BOOT_LOADER_INO) { | |
ac27a0ec | 4244 | /* this inode is deleted */ |
1d1fe1ee | 4245 | ret = -ESTALE; |
ac27a0ec DK |
4246 | goto bad_inode; |
4247 | } | |
4248 | /* The only unlinked inodes we let through here have | |
4249 | * valid i_mode and are being read by the orphan | |
4250 | * recovery code: that's fine, we're about to complete | |
393d1d1d DTB |
4251 | * the process of deleting those. |
4252 | * OR it is the EXT4_BOOT_LOADER_INO which is | |
4253 | * not initialized on a new filesystem. */ | |
ac27a0ec | 4254 | } |
ac27a0ec | 4255 | ei->i_flags = le32_to_cpu(raw_inode->i_flags); |
0fc1b451 | 4256 | inode->i_blocks = ext4_inode_blocks(raw_inode, ei); |
7973c0c1 | 4257 | ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo); |
a9e81742 | 4258 | if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) |
a1ddeb7e BP |
4259 | ei->i_file_acl |= |
4260 | ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32; | |
a48380f7 | 4261 | inode->i_size = ext4_isize(raw_inode); |
ac27a0ec | 4262 | ei->i_disksize = inode->i_size; |
a9e7f447 DM |
4263 | #ifdef CONFIG_QUOTA |
4264 | ei->i_reserved_quota = 0; | |
4265 | #endif | |
ac27a0ec DK |
4266 | inode->i_generation = le32_to_cpu(raw_inode->i_generation); |
4267 | ei->i_block_group = iloc.block_group; | |
a4912123 | 4268 | ei->i_last_alloc_group = ~0; |
ac27a0ec DK |
4269 | /* |
4270 | * NOTE! The in-memory inode i_data array is in little-endian order | |
4271 | * even on big-endian machines: we do NOT byteswap the block numbers! | |
4272 | */ | |
617ba13b | 4273 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
ac27a0ec DK |
4274 | ei->i_data[block] = raw_inode->i_block[block]; |
4275 | INIT_LIST_HEAD(&ei->i_orphan); | |
4276 | ||
b436b9be JK |
4277 | /* |
4278 | * Set transaction id's of transactions that have to be committed | |
4279 | * to finish f[data]sync. We set them to currently running transaction | |
4280 | * as we cannot be sure that the inode or some of its metadata isn't | |
4281 | * part of the transaction - the inode could have been reclaimed and | |
4282 | * now it is reread from disk. | |
4283 | */ | |
4284 | if (journal) { | |
4285 | transaction_t *transaction; | |
4286 | tid_t tid; | |
4287 | ||
a931da6a | 4288 | read_lock(&journal->j_state_lock); |
b436b9be JK |
4289 | if (journal->j_running_transaction) |
4290 | transaction = journal->j_running_transaction; | |
4291 | else | |
4292 | transaction = journal->j_committing_transaction; | |
4293 | if (transaction) | |
4294 | tid = transaction->t_tid; | |
4295 | else | |
4296 | tid = journal->j_commit_sequence; | |
a931da6a | 4297 | read_unlock(&journal->j_state_lock); |
b436b9be JK |
4298 | ei->i_sync_tid = tid; |
4299 | ei->i_datasync_tid = tid; | |
4300 | } | |
4301 | ||
0040d987 | 4302 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
ac27a0ec DK |
4303 | if (ei->i_extra_isize == 0) { |
4304 | /* The extra space is currently unused. Use it. */ | |
617ba13b MC |
4305 | ei->i_extra_isize = sizeof(struct ext4_inode) - |
4306 | EXT4_GOOD_OLD_INODE_SIZE; | |
ac27a0ec | 4307 | } else { |
152a7b0a | 4308 | ext4_iget_extra_inode(inode, raw_inode, ei); |
ac27a0ec | 4309 | } |
814525f4 | 4310 | } |
ac27a0ec | 4311 | |
ef7f3835 KS |
4312 | EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode); |
4313 | EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode); | |
4314 | EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode); | |
4315 | EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode); | |
4316 | ||
25ec56b5 JNC |
4317 | inode->i_version = le32_to_cpu(raw_inode->i_disk_version); |
4318 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { | |
4319 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
4320 | inode->i_version |= | |
4321 | (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32; | |
4322 | } | |
4323 | ||
c4b5a614 | 4324 | ret = 0; |
485c26ec | 4325 | if (ei->i_file_acl && |
1032988c | 4326 | !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) { |
24676da4 TT |
4327 | EXT4_ERROR_INODE(inode, "bad extended attribute block %llu", |
4328 | ei->i_file_acl); | |
485c26ec TT |
4329 | ret = -EIO; |
4330 | goto bad_inode; | |
f19d5870 TM |
4331 | } else if (!ext4_has_inline_data(inode)) { |
4332 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { | |
4333 | if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || | |
4334 | (S_ISLNK(inode->i_mode) && | |
4335 | !ext4_inode_is_fast_symlink(inode)))) | |
4336 | /* Validate extent which is part of inode */ | |
4337 | ret = ext4_ext_check_inode(inode); | |
4338 | } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || | |
4339 | (S_ISLNK(inode->i_mode) && | |
4340 | !ext4_inode_is_fast_symlink(inode))) { | |
4341 | /* Validate block references which are part of inode */ | |
4342 | ret = ext4_ind_check_inode(inode); | |
4343 | } | |
fe2c8191 | 4344 | } |
567f3e9a | 4345 | if (ret) |
de9a55b8 | 4346 | goto bad_inode; |
7a262f7c | 4347 | |
ac27a0ec | 4348 | if (S_ISREG(inode->i_mode)) { |
617ba13b MC |
4349 | inode->i_op = &ext4_file_inode_operations; |
4350 | inode->i_fop = &ext4_file_operations; | |
4351 | ext4_set_aops(inode); | |
ac27a0ec | 4352 | } else if (S_ISDIR(inode->i_mode)) { |
617ba13b MC |
4353 | inode->i_op = &ext4_dir_inode_operations; |
4354 | inode->i_fop = &ext4_dir_operations; | |
ac27a0ec | 4355 | } else if (S_ISLNK(inode->i_mode)) { |
e83c1397 | 4356 | if (ext4_inode_is_fast_symlink(inode)) { |
617ba13b | 4357 | inode->i_op = &ext4_fast_symlink_inode_operations; |
e83c1397 DG |
4358 | nd_terminate_link(ei->i_data, inode->i_size, |
4359 | sizeof(ei->i_data) - 1); | |
4360 | } else { | |
617ba13b MC |
4361 | inode->i_op = &ext4_symlink_inode_operations; |
4362 | ext4_set_aops(inode); | |
ac27a0ec | 4363 | } |
563bdd61 TT |
4364 | } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) || |
4365 | S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) { | |
617ba13b | 4366 | inode->i_op = &ext4_special_inode_operations; |
ac27a0ec DK |
4367 | if (raw_inode->i_block[0]) |
4368 | init_special_inode(inode, inode->i_mode, | |
4369 | old_decode_dev(le32_to_cpu(raw_inode->i_block[0]))); | |
4370 | else | |
4371 | init_special_inode(inode, inode->i_mode, | |
4372 | new_decode_dev(le32_to_cpu(raw_inode->i_block[1]))); | |
393d1d1d DTB |
4373 | } else if (ino == EXT4_BOOT_LOADER_INO) { |
4374 | make_bad_inode(inode); | |
563bdd61 | 4375 | } else { |
563bdd61 | 4376 | ret = -EIO; |
24676da4 | 4377 | EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode); |
563bdd61 | 4378 | goto bad_inode; |
ac27a0ec | 4379 | } |
af5bc92d | 4380 | brelse(iloc.bh); |
617ba13b | 4381 | ext4_set_inode_flags(inode); |
1d1fe1ee DH |
4382 | unlock_new_inode(inode); |
4383 | return inode; | |
ac27a0ec DK |
4384 | |
4385 | bad_inode: | |
567f3e9a | 4386 | brelse(iloc.bh); |
1d1fe1ee DH |
4387 | iget_failed(inode); |
4388 | return ERR_PTR(ret); | |
ac27a0ec DK |
4389 | } |
4390 | ||
0fc1b451 AK |
4391 | static int ext4_inode_blocks_set(handle_t *handle, |
4392 | struct ext4_inode *raw_inode, | |
4393 | struct ext4_inode_info *ei) | |
4394 | { | |
4395 | struct inode *inode = &(ei->vfs_inode); | |
4396 | u64 i_blocks = inode->i_blocks; | |
4397 | struct super_block *sb = inode->i_sb; | |
0fc1b451 AK |
4398 | |
4399 | if (i_blocks <= ~0U) { | |
4400 | /* | |
4907cb7b | 4401 | * i_blocks can be represented in a 32 bit variable |
0fc1b451 AK |
4402 | * as multiple of 512 bytes |
4403 | */ | |
8180a562 | 4404 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 4405 | raw_inode->i_blocks_high = 0; |
84a8dce2 | 4406 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
f287a1a5 TT |
4407 | return 0; |
4408 | } | |
4409 | if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) | |
4410 | return -EFBIG; | |
4411 | ||
4412 | if (i_blocks <= 0xffffffffffffULL) { | |
0fc1b451 AK |
4413 | /* |
4414 | * i_blocks can be represented in a 48 bit variable | |
4415 | * as multiple of 512 bytes | |
4416 | */ | |
8180a562 | 4417 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 4418 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); |
84a8dce2 | 4419 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
0fc1b451 | 4420 | } else { |
84a8dce2 | 4421 | ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
8180a562 AK |
4422 | /* i_block is stored in file system block size */ |
4423 | i_blocks = i_blocks >> (inode->i_blkbits - 9); | |
4424 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); | |
4425 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); | |
0fc1b451 | 4426 | } |
f287a1a5 | 4427 | return 0; |
0fc1b451 AK |
4428 | } |
4429 | ||
ac27a0ec DK |
4430 | /* |
4431 | * Post the struct inode info into an on-disk inode location in the | |
4432 | * buffer-cache. This gobbles the caller's reference to the | |
4433 | * buffer_head in the inode location struct. | |
4434 | * | |
4435 | * The caller must have write access to iloc->bh. | |
4436 | */ | |
617ba13b | 4437 | static int ext4_do_update_inode(handle_t *handle, |
ac27a0ec | 4438 | struct inode *inode, |
830156c7 | 4439 | struct ext4_iloc *iloc) |
ac27a0ec | 4440 | { |
617ba13b MC |
4441 | struct ext4_inode *raw_inode = ext4_raw_inode(iloc); |
4442 | struct ext4_inode_info *ei = EXT4_I(inode); | |
ac27a0ec DK |
4443 | struct buffer_head *bh = iloc->bh; |
4444 | int err = 0, rc, block; | |
b71fc079 | 4445 | int need_datasync = 0; |
08cefc7a EB |
4446 | uid_t i_uid; |
4447 | gid_t i_gid; | |
ac27a0ec DK |
4448 | |
4449 | /* For fields not not tracking in the in-memory inode, | |
4450 | * initialise them to zero for new inodes. */ | |
19f5fb7a | 4451 | if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) |
617ba13b | 4452 | memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size); |
ac27a0ec | 4453 | |
ff9ddf7e | 4454 | ext4_get_inode_flags(ei); |
ac27a0ec | 4455 | raw_inode->i_mode = cpu_to_le16(inode->i_mode); |
08cefc7a EB |
4456 | i_uid = i_uid_read(inode); |
4457 | i_gid = i_gid_read(inode); | |
af5bc92d | 4458 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
08cefc7a EB |
4459 | raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid)); |
4460 | raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid)); | |
ac27a0ec DK |
4461 | /* |
4462 | * Fix up interoperability with old kernels. Otherwise, old inodes get | |
4463 | * re-used with the upper 16 bits of the uid/gid intact | |
4464 | */ | |
af5bc92d | 4465 | if (!ei->i_dtime) { |
ac27a0ec | 4466 | raw_inode->i_uid_high = |
08cefc7a | 4467 | cpu_to_le16(high_16_bits(i_uid)); |
ac27a0ec | 4468 | raw_inode->i_gid_high = |
08cefc7a | 4469 | cpu_to_le16(high_16_bits(i_gid)); |
ac27a0ec DK |
4470 | } else { |
4471 | raw_inode->i_uid_high = 0; | |
4472 | raw_inode->i_gid_high = 0; | |
4473 | } | |
4474 | } else { | |
08cefc7a EB |
4475 | raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid)); |
4476 | raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid)); | |
ac27a0ec DK |
4477 | raw_inode->i_uid_high = 0; |
4478 | raw_inode->i_gid_high = 0; | |
4479 | } | |
4480 | raw_inode->i_links_count = cpu_to_le16(inode->i_nlink); | |
ef7f3835 KS |
4481 | |
4482 | EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode); | |
4483 | EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode); | |
4484 | EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode); | |
4485 | EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode); | |
4486 | ||
0fc1b451 AK |
4487 | if (ext4_inode_blocks_set(handle, raw_inode, ei)) |
4488 | goto out_brelse; | |
ac27a0ec | 4489 | raw_inode->i_dtime = cpu_to_le32(ei->i_dtime); |
353eb83c | 4490 | raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF); |
9b8f1f01 MC |
4491 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != |
4492 | cpu_to_le32(EXT4_OS_HURD)) | |
a1ddeb7e BP |
4493 | raw_inode->i_file_acl_high = |
4494 | cpu_to_le16(ei->i_file_acl >> 32); | |
7973c0c1 | 4495 | raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl); |
b71fc079 JK |
4496 | if (ei->i_disksize != ext4_isize(raw_inode)) { |
4497 | ext4_isize_set(raw_inode, ei->i_disksize); | |
4498 | need_datasync = 1; | |
4499 | } | |
a48380f7 AK |
4500 | if (ei->i_disksize > 0x7fffffffULL) { |
4501 | struct super_block *sb = inode->i_sb; | |
4502 | if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, | |
4503 | EXT4_FEATURE_RO_COMPAT_LARGE_FILE) || | |
4504 | EXT4_SB(sb)->s_es->s_rev_level == | |
4505 | cpu_to_le32(EXT4_GOOD_OLD_REV)) { | |
4506 | /* If this is the first large file | |
4507 | * created, add a flag to the superblock. | |
4508 | */ | |
4509 | err = ext4_journal_get_write_access(handle, | |
4510 | EXT4_SB(sb)->s_sbh); | |
4511 | if (err) | |
4512 | goto out_brelse; | |
4513 | ext4_update_dynamic_rev(sb); | |
4514 | EXT4_SET_RO_COMPAT_FEATURE(sb, | |
617ba13b | 4515 | EXT4_FEATURE_RO_COMPAT_LARGE_FILE); |
0390131b | 4516 | ext4_handle_sync(handle); |
b50924c2 | 4517 | err = ext4_handle_dirty_super(handle, sb); |
ac27a0ec DK |
4518 | } |
4519 | } | |
4520 | raw_inode->i_generation = cpu_to_le32(inode->i_generation); | |
4521 | if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) { | |
4522 | if (old_valid_dev(inode->i_rdev)) { | |
4523 | raw_inode->i_block[0] = | |
4524 | cpu_to_le32(old_encode_dev(inode->i_rdev)); | |
4525 | raw_inode->i_block[1] = 0; | |
4526 | } else { | |
4527 | raw_inode->i_block[0] = 0; | |
4528 | raw_inode->i_block[1] = | |
4529 | cpu_to_le32(new_encode_dev(inode->i_rdev)); | |
4530 | raw_inode->i_block[2] = 0; | |
4531 | } | |
f19d5870 | 4532 | } else if (!ext4_has_inline_data(inode)) { |
de9a55b8 TT |
4533 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
4534 | raw_inode->i_block[block] = ei->i_data[block]; | |
f19d5870 | 4535 | } |
ac27a0ec | 4536 | |
25ec56b5 JNC |
4537 | raw_inode->i_disk_version = cpu_to_le32(inode->i_version); |
4538 | if (ei->i_extra_isize) { | |
4539 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
4540 | raw_inode->i_version_hi = | |
4541 | cpu_to_le32(inode->i_version >> 32); | |
ac27a0ec | 4542 | raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize); |
25ec56b5 JNC |
4543 | } |
4544 | ||
814525f4 DW |
4545 | ext4_inode_csum_set(inode, raw_inode, ei); |
4546 | ||
830156c7 | 4547 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
73b50c1c | 4548 | rc = ext4_handle_dirty_metadata(handle, NULL, bh); |
830156c7 FM |
4549 | if (!err) |
4550 | err = rc; | |
19f5fb7a | 4551 | ext4_clear_inode_state(inode, EXT4_STATE_NEW); |
ac27a0ec | 4552 | |
b71fc079 | 4553 | ext4_update_inode_fsync_trans(handle, inode, need_datasync); |
ac27a0ec | 4554 | out_brelse: |
af5bc92d | 4555 | brelse(bh); |
617ba13b | 4556 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
4557 | return err; |
4558 | } | |
4559 | ||
4560 | /* | |
617ba13b | 4561 | * ext4_write_inode() |
ac27a0ec DK |
4562 | * |
4563 | * We are called from a few places: | |
4564 | * | |
4565 | * - Within generic_file_write() for O_SYNC files. | |
4566 | * Here, there will be no transaction running. We wait for any running | |
4907cb7b | 4567 | * transaction to commit. |
ac27a0ec DK |
4568 | * |
4569 | * - Within sys_sync(), kupdate and such. | |
4570 | * We wait on commit, if tol to. | |
4571 | * | |
4572 | * - Within prune_icache() (PF_MEMALLOC == true) | |
4573 | * Here we simply return. We can't afford to block kswapd on the | |
4574 | * journal commit. | |
4575 | * | |
4576 | * In all cases it is actually safe for us to return without doing anything, | |
4577 | * because the inode has been copied into a raw inode buffer in | |
617ba13b | 4578 | * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for |
ac27a0ec DK |
4579 | * knfsd. |
4580 | * | |
4581 | * Note that we are absolutely dependent upon all inode dirtiers doing the | |
4582 | * right thing: they *must* call mark_inode_dirty() after dirtying info in | |
4583 | * which we are interested. | |
4584 | * | |
4585 | * It would be a bug for them to not do this. The code: | |
4586 | * | |
4587 | * mark_inode_dirty(inode) | |
4588 | * stuff(); | |
4589 | * inode->i_size = expr; | |
4590 | * | |
4591 | * is in error because a kswapd-driven write_inode() could occur while | |
4592 | * `stuff()' is running, and the new i_size will be lost. Plus the inode | |
4593 | * will no longer be on the superblock's dirty inode list. | |
4594 | */ | |
a9185b41 | 4595 | int ext4_write_inode(struct inode *inode, struct writeback_control *wbc) |
ac27a0ec | 4596 | { |
91ac6f43 FM |
4597 | int err; |
4598 | ||
ac27a0ec DK |
4599 | if (current->flags & PF_MEMALLOC) |
4600 | return 0; | |
4601 | ||
91ac6f43 FM |
4602 | if (EXT4_SB(inode->i_sb)->s_journal) { |
4603 | if (ext4_journal_current_handle()) { | |
4604 | jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n"); | |
4605 | dump_stack(); | |
4606 | return -EIO; | |
4607 | } | |
ac27a0ec | 4608 | |
a9185b41 | 4609 | if (wbc->sync_mode != WB_SYNC_ALL) |
91ac6f43 FM |
4610 | return 0; |
4611 | ||
4612 | err = ext4_force_commit(inode->i_sb); | |
4613 | } else { | |
4614 | struct ext4_iloc iloc; | |
ac27a0ec | 4615 | |
8b472d73 | 4616 | err = __ext4_get_inode_loc(inode, &iloc, 0); |
91ac6f43 FM |
4617 | if (err) |
4618 | return err; | |
a9185b41 | 4619 | if (wbc->sync_mode == WB_SYNC_ALL) |
830156c7 FM |
4620 | sync_dirty_buffer(iloc.bh); |
4621 | if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) { | |
c398eda0 TT |
4622 | EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr, |
4623 | "IO error syncing inode"); | |
830156c7 FM |
4624 | err = -EIO; |
4625 | } | |
fd2dd9fb | 4626 | brelse(iloc.bh); |
91ac6f43 FM |
4627 | } |
4628 | return err; | |
ac27a0ec DK |
4629 | } |
4630 | ||
53e87268 JK |
4631 | /* |
4632 | * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate | |
4633 | * buffers that are attached to a page stradding i_size and are undergoing | |
4634 | * commit. In that case we have to wait for commit to finish and try again. | |
4635 | */ | |
4636 | static void ext4_wait_for_tail_page_commit(struct inode *inode) | |
4637 | { | |
4638 | struct page *page; | |
4639 | unsigned offset; | |
4640 | journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; | |
4641 | tid_t commit_tid = 0; | |
4642 | int ret; | |
4643 | ||
4644 | offset = inode->i_size & (PAGE_CACHE_SIZE - 1); | |
4645 | /* | |
4646 | * All buffers in the last page remain valid? Then there's nothing to | |
4647 | * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE == | |
4648 | * blocksize case | |
4649 | */ | |
4650 | if (offset > PAGE_CACHE_SIZE - (1 << inode->i_blkbits)) | |
4651 | return; | |
4652 | while (1) { | |
4653 | page = find_lock_page(inode->i_mapping, | |
4654 | inode->i_size >> PAGE_CACHE_SHIFT); | |
4655 | if (!page) | |
4656 | return; | |
4657 | ret = __ext4_journalled_invalidatepage(page, offset); | |
4658 | unlock_page(page); | |
4659 | page_cache_release(page); | |
4660 | if (ret != -EBUSY) | |
4661 | return; | |
4662 | commit_tid = 0; | |
4663 | read_lock(&journal->j_state_lock); | |
4664 | if (journal->j_committing_transaction) | |
4665 | commit_tid = journal->j_committing_transaction->t_tid; | |
4666 | read_unlock(&journal->j_state_lock); | |
4667 | if (commit_tid) | |
4668 | jbd2_log_wait_commit(journal, commit_tid); | |
4669 | } | |
4670 | } | |
4671 | ||
ac27a0ec | 4672 | /* |
617ba13b | 4673 | * ext4_setattr() |
ac27a0ec DK |
4674 | * |
4675 | * Called from notify_change. | |
4676 | * | |
4677 | * We want to trap VFS attempts to truncate the file as soon as | |
4678 | * possible. In particular, we want to make sure that when the VFS | |
4679 | * shrinks i_size, we put the inode on the orphan list and modify | |
4680 | * i_disksize immediately, so that during the subsequent flushing of | |
4681 | * dirty pages and freeing of disk blocks, we can guarantee that any | |
4682 | * commit will leave the blocks being flushed in an unused state on | |
4683 | * disk. (On recovery, the inode will get truncated and the blocks will | |
4684 | * be freed, so we have a strong guarantee that no future commit will | |
4685 | * leave these blocks visible to the user.) | |
4686 | * | |
678aaf48 JK |
4687 | * Another thing we have to assure is that if we are in ordered mode |
4688 | * and inode is still attached to the committing transaction, we must | |
4689 | * we start writeout of all the dirty pages which are being truncated. | |
4690 | * This way we are sure that all the data written in the previous | |
4691 | * transaction are already on disk (truncate waits for pages under | |
4692 | * writeback). | |
4693 | * | |
4694 | * Called with inode->i_mutex down. | |
ac27a0ec | 4695 | */ |
617ba13b | 4696 | int ext4_setattr(struct dentry *dentry, struct iattr *attr) |
ac27a0ec DK |
4697 | { |
4698 | struct inode *inode = dentry->d_inode; | |
4699 | int error, rc = 0; | |
3d287de3 | 4700 | int orphan = 0; |
ac27a0ec DK |
4701 | const unsigned int ia_valid = attr->ia_valid; |
4702 | ||
4703 | error = inode_change_ok(inode, attr); | |
4704 | if (error) | |
4705 | return error; | |
4706 | ||
12755627 | 4707 | if (is_quota_modification(inode, attr)) |
871a2931 | 4708 | dquot_initialize(inode); |
08cefc7a EB |
4709 | if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) || |
4710 | (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) { | |
ac27a0ec DK |
4711 | handle_t *handle; |
4712 | ||
4713 | /* (user+group)*(old+new) structure, inode write (sb, | |
4714 | * inode block, ? - but truncate inode update has it) */ | |
9924a92a TT |
4715 | handle = ext4_journal_start(inode, EXT4_HT_QUOTA, |
4716 | (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) + | |
4717 | EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3); | |
ac27a0ec DK |
4718 | if (IS_ERR(handle)) { |
4719 | error = PTR_ERR(handle); | |
4720 | goto err_out; | |
4721 | } | |
b43fa828 | 4722 | error = dquot_transfer(inode, attr); |
ac27a0ec | 4723 | if (error) { |
617ba13b | 4724 | ext4_journal_stop(handle); |
ac27a0ec DK |
4725 | return error; |
4726 | } | |
4727 | /* Update corresponding info in inode so that everything is in | |
4728 | * one transaction */ | |
4729 | if (attr->ia_valid & ATTR_UID) | |
4730 | inode->i_uid = attr->ia_uid; | |
4731 | if (attr->ia_valid & ATTR_GID) | |
4732 | inode->i_gid = attr->ia_gid; | |
617ba13b MC |
4733 | error = ext4_mark_inode_dirty(handle, inode); |
4734 | ext4_journal_stop(handle); | |
ac27a0ec DK |
4735 | } |
4736 | ||
263c784f JK |
4737 | if (attr->ia_valid & ATTR_SIZE && attr->ia_size != inode->i_size) { |
4738 | handle_t *handle; | |
4739 | loff_t oldsize = inode->i_size; | |
562c72aa | 4740 | |
12e9b892 | 4741 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) { |
e2b46574 ES |
4742 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
4743 | ||
0c095c7f TT |
4744 | if (attr->ia_size > sbi->s_bitmap_maxbytes) |
4745 | return -EFBIG; | |
e2b46574 | 4746 | } |
d5685be1 CH |
4747 | |
4748 | if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size) | |
4749 | inode_inc_iversion(inode); | |
4750 | ||
263c784f JK |
4751 | if (S_ISREG(inode->i_mode) && |
4752 | (attr->ia_size < inode->i_size)) { | |
4753 | if (ext4_should_order_data(inode)) { | |
4754 | error = ext4_begin_ordered_truncate(inode, | |
678aaf48 | 4755 | attr->ia_size); |
263c784f | 4756 | if (error) |
678aaf48 | 4757 | goto err_out; |
263c784f JK |
4758 | } |
4759 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 3); | |
4760 | if (IS_ERR(handle)) { | |
4761 | error = PTR_ERR(handle); | |
4762 | goto err_out; | |
4763 | } | |
4764 | if (ext4_handle_valid(handle)) { | |
4765 | error = ext4_orphan_add(handle, inode); | |
4766 | orphan = 1; | |
4767 | } | |
4768 | EXT4_I(inode)->i_disksize = attr->ia_size; | |
4769 | rc = ext4_mark_inode_dirty(handle, inode); | |
4770 | if (!error) | |
4771 | error = rc; | |
4772 | ext4_journal_stop(handle); | |
4773 | if (error) { | |
4774 | ext4_orphan_del(NULL, inode); | |
678aaf48 JK |
4775 | goto err_out; |
4776 | } | |
4777 | } | |
53e87268 | 4778 | |
263c784f JK |
4779 | i_size_write(inode, attr->ia_size); |
4780 | /* | |
4781 | * Blocks are going to be removed from the inode. Wait | |
4782 | * for dio in flight. Temporarily disable | |
4783 | * dioread_nolock to prevent livelock. | |
4784 | */ | |
4785 | if (orphan) { | |
4786 | if (!ext4_should_journal_data(inode)) { | |
4787 | ext4_inode_block_unlocked_dio(inode); | |
4788 | inode_dio_wait(inode); | |
4789 | ext4_inode_resume_unlocked_dio(inode); | |
4790 | } else | |
4791 | ext4_wait_for_tail_page_commit(inode); | |
1c9114f9 | 4792 | } |
263c784f JK |
4793 | /* |
4794 | * Truncate pagecache after we've waited for commit | |
4795 | * in data=journal mode to make pages freeable. | |
4796 | */ | |
4797 | truncate_pagecache(inode, oldsize, inode->i_size); | |
072bd7ea | 4798 | } |
263c784f JK |
4799 | /* |
4800 | * We want to call ext4_truncate() even if attr->ia_size == | |
4801 | * inode->i_size for cases like truncation of fallocated space | |
4802 | */ | |
4803 | if (attr->ia_valid & ATTR_SIZE) | |
4804 | ext4_truncate(inode); | |
ac27a0ec | 4805 | |
1025774c CH |
4806 | if (!rc) { |
4807 | setattr_copy(inode, attr); | |
4808 | mark_inode_dirty(inode); | |
4809 | } | |
4810 | ||
4811 | /* | |
4812 | * If the call to ext4_truncate failed to get a transaction handle at | |
4813 | * all, we need to clean up the in-core orphan list manually. | |
4814 | */ | |
3d287de3 | 4815 | if (orphan && inode->i_nlink) |
617ba13b | 4816 | ext4_orphan_del(NULL, inode); |
ac27a0ec DK |
4817 | |
4818 | if (!rc && (ia_valid & ATTR_MODE)) | |
617ba13b | 4819 | rc = ext4_acl_chmod(inode); |
ac27a0ec DK |
4820 | |
4821 | err_out: | |
617ba13b | 4822 | ext4_std_error(inode->i_sb, error); |
ac27a0ec DK |
4823 | if (!error) |
4824 | error = rc; | |
4825 | return error; | |
4826 | } | |
4827 | ||
3e3398a0 MC |
4828 | int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry, |
4829 | struct kstat *stat) | |
4830 | { | |
4831 | struct inode *inode; | |
bb39c83c | 4832 | unsigned long long delalloc_blocks; |
3e3398a0 MC |
4833 | |
4834 | inode = dentry->d_inode; | |
4835 | generic_fillattr(inode, stat); | |
4836 | ||
4837 | /* | |
4838 | * We can't update i_blocks if the block allocation is delayed | |
4839 | * otherwise in the case of system crash before the real block | |
4840 | * allocation is done, we will have i_blocks inconsistent with | |
4841 | * on-disk file blocks. | |
4842 | * We always keep i_blocks updated together with real | |
4843 | * allocation. But to not confuse with user, stat | |
4844 | * will return the blocks that include the delayed allocation | |
4845 | * blocks for this file. | |
4846 | */ | |
96607551 TM |
4847 | delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb), |
4848 | EXT4_I(inode)->i_reserved_data_blocks); | |
3e3398a0 | 4849 | |
bb39c83c | 4850 | stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits-9); |
3e3398a0 MC |
4851 | return 0; |
4852 | } | |
ac27a0ec | 4853 | |
a02908f1 MC |
4854 | static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk) |
4855 | { | |
12e9b892 | 4856 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) |
8bb2b247 | 4857 | return ext4_ind_trans_blocks(inode, nrblocks, chunk); |
ac51d837 | 4858 | return ext4_ext_index_trans_blocks(inode, nrblocks, chunk); |
a02908f1 | 4859 | } |
ac51d837 | 4860 | |
ac27a0ec | 4861 | /* |
a02908f1 MC |
4862 | * Account for index blocks, block groups bitmaps and block group |
4863 | * descriptor blocks if modify datablocks and index blocks | |
4864 | * worse case, the indexs blocks spread over different block groups | |
ac27a0ec | 4865 | * |
a02908f1 | 4866 | * If datablocks are discontiguous, they are possible to spread over |
4907cb7b | 4867 | * different block groups too. If they are contiguous, with flexbg, |
a02908f1 | 4868 | * they could still across block group boundary. |
ac27a0ec | 4869 | * |
a02908f1 MC |
4870 | * Also account for superblock, inode, quota and xattr blocks |
4871 | */ | |
1f109d5a | 4872 | static int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk) |
a02908f1 | 4873 | { |
8df9675f TT |
4874 | ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb); |
4875 | int gdpblocks; | |
a02908f1 MC |
4876 | int idxblocks; |
4877 | int ret = 0; | |
4878 | ||
4879 | /* | |
4880 | * How many index blocks need to touch to modify nrblocks? | |
4881 | * The "Chunk" flag indicating whether the nrblocks is | |
4882 | * physically contiguous on disk | |
4883 | * | |
4884 | * For Direct IO and fallocate, they calls get_block to allocate | |
4885 | * one single extent at a time, so they could set the "Chunk" flag | |
4886 | */ | |
4887 | idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk); | |
4888 | ||
4889 | ret = idxblocks; | |
4890 | ||
4891 | /* | |
4892 | * Now let's see how many group bitmaps and group descriptors need | |
4893 | * to account | |
4894 | */ | |
4895 | groups = idxblocks; | |
4896 | if (chunk) | |
4897 | groups += 1; | |
4898 | else | |
4899 | groups += nrblocks; | |
4900 | ||
4901 | gdpblocks = groups; | |
8df9675f TT |
4902 | if (groups > ngroups) |
4903 | groups = ngroups; | |
a02908f1 MC |
4904 | if (groups > EXT4_SB(inode->i_sb)->s_gdb_count) |
4905 | gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count; | |
4906 | ||
4907 | /* bitmaps and block group descriptor blocks */ | |
4908 | ret += groups + gdpblocks; | |
4909 | ||
4910 | /* Blocks for super block, inode, quota and xattr blocks */ | |
4911 | ret += EXT4_META_TRANS_BLOCKS(inode->i_sb); | |
4912 | ||
4913 | return ret; | |
4914 | } | |
4915 | ||
4916 | /* | |
25985edc | 4917 | * Calculate the total number of credits to reserve to fit |
f3bd1f3f MC |
4918 | * the modification of a single pages into a single transaction, |
4919 | * which may include multiple chunks of block allocations. | |
ac27a0ec | 4920 | * |
525f4ed8 | 4921 | * This could be called via ext4_write_begin() |
ac27a0ec | 4922 | * |
525f4ed8 | 4923 | * We need to consider the worse case, when |
a02908f1 | 4924 | * one new block per extent. |
ac27a0ec | 4925 | */ |
a86c6181 | 4926 | int ext4_writepage_trans_blocks(struct inode *inode) |
ac27a0ec | 4927 | { |
617ba13b | 4928 | int bpp = ext4_journal_blocks_per_page(inode); |
ac27a0ec DK |
4929 | int ret; |
4930 | ||
a02908f1 | 4931 | ret = ext4_meta_trans_blocks(inode, bpp, 0); |
a86c6181 | 4932 | |
a02908f1 | 4933 | /* Account for data blocks for journalled mode */ |
617ba13b | 4934 | if (ext4_should_journal_data(inode)) |
a02908f1 | 4935 | ret += bpp; |
ac27a0ec DK |
4936 | return ret; |
4937 | } | |
f3bd1f3f MC |
4938 | |
4939 | /* | |
4940 | * Calculate the journal credits for a chunk of data modification. | |
4941 | * | |
4942 | * This is called from DIO, fallocate or whoever calling | |
79e83036 | 4943 | * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks. |
f3bd1f3f MC |
4944 | * |
4945 | * journal buffers for data blocks are not included here, as DIO | |
4946 | * and fallocate do no need to journal data buffers. | |
4947 | */ | |
4948 | int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks) | |
4949 | { | |
4950 | return ext4_meta_trans_blocks(inode, nrblocks, 1); | |
4951 | } | |
4952 | ||
ac27a0ec | 4953 | /* |
617ba13b | 4954 | * The caller must have previously called ext4_reserve_inode_write(). |
ac27a0ec DK |
4955 | * Give this, we know that the caller already has write access to iloc->bh. |
4956 | */ | |
617ba13b | 4957 | int ext4_mark_iloc_dirty(handle_t *handle, |
de9a55b8 | 4958 | struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec DK |
4959 | { |
4960 | int err = 0; | |
4961 | ||
c64db50e | 4962 | if (IS_I_VERSION(inode)) |
25ec56b5 JNC |
4963 | inode_inc_iversion(inode); |
4964 | ||
ac27a0ec DK |
4965 | /* the do_update_inode consumes one bh->b_count */ |
4966 | get_bh(iloc->bh); | |
4967 | ||
dab291af | 4968 | /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */ |
830156c7 | 4969 | err = ext4_do_update_inode(handle, inode, iloc); |
ac27a0ec DK |
4970 | put_bh(iloc->bh); |
4971 | return err; | |
4972 | } | |
4973 | ||
4974 | /* | |
4975 | * On success, We end up with an outstanding reference count against | |
4976 | * iloc->bh. This _must_ be cleaned up later. | |
4977 | */ | |
4978 | ||
4979 | int | |
617ba13b MC |
4980 | ext4_reserve_inode_write(handle_t *handle, struct inode *inode, |
4981 | struct ext4_iloc *iloc) | |
ac27a0ec | 4982 | { |
0390131b FM |
4983 | int err; |
4984 | ||
4985 | err = ext4_get_inode_loc(inode, iloc); | |
4986 | if (!err) { | |
4987 | BUFFER_TRACE(iloc->bh, "get_write_access"); | |
4988 | err = ext4_journal_get_write_access(handle, iloc->bh); | |
4989 | if (err) { | |
4990 | brelse(iloc->bh); | |
4991 | iloc->bh = NULL; | |
ac27a0ec DK |
4992 | } |
4993 | } | |
617ba13b | 4994 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
4995 | return err; |
4996 | } | |
4997 | ||
6dd4ee7c KS |
4998 | /* |
4999 | * Expand an inode by new_extra_isize bytes. | |
5000 | * Returns 0 on success or negative error number on failure. | |
5001 | */ | |
1d03ec98 AK |
5002 | static int ext4_expand_extra_isize(struct inode *inode, |
5003 | unsigned int new_extra_isize, | |
5004 | struct ext4_iloc iloc, | |
5005 | handle_t *handle) | |
6dd4ee7c KS |
5006 | { |
5007 | struct ext4_inode *raw_inode; | |
5008 | struct ext4_xattr_ibody_header *header; | |
6dd4ee7c KS |
5009 | |
5010 | if (EXT4_I(inode)->i_extra_isize >= new_extra_isize) | |
5011 | return 0; | |
5012 | ||
5013 | raw_inode = ext4_raw_inode(&iloc); | |
5014 | ||
5015 | header = IHDR(inode, raw_inode); | |
6dd4ee7c KS |
5016 | |
5017 | /* No extended attributes present */ | |
19f5fb7a TT |
5018 | if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) || |
5019 | header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) { | |
6dd4ee7c KS |
5020 | memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0, |
5021 | new_extra_isize); | |
5022 | EXT4_I(inode)->i_extra_isize = new_extra_isize; | |
5023 | return 0; | |
5024 | } | |
5025 | ||
5026 | /* try to expand with EAs present */ | |
5027 | return ext4_expand_extra_isize_ea(inode, new_extra_isize, | |
5028 | raw_inode, handle); | |
5029 | } | |
5030 | ||
ac27a0ec DK |
5031 | /* |
5032 | * What we do here is to mark the in-core inode as clean with respect to inode | |
5033 | * dirtiness (it may still be data-dirty). | |
5034 | * This means that the in-core inode may be reaped by prune_icache | |
5035 | * without having to perform any I/O. This is a very good thing, | |
5036 | * because *any* task may call prune_icache - even ones which | |
5037 | * have a transaction open against a different journal. | |
5038 | * | |
5039 | * Is this cheating? Not really. Sure, we haven't written the | |
5040 | * inode out, but prune_icache isn't a user-visible syncing function. | |
5041 | * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync) | |
5042 | * we start and wait on commits. | |
ac27a0ec | 5043 | */ |
617ba13b | 5044 | int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode) |
ac27a0ec | 5045 | { |
617ba13b | 5046 | struct ext4_iloc iloc; |
6dd4ee7c KS |
5047 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
5048 | static unsigned int mnt_count; | |
5049 | int err, ret; | |
ac27a0ec DK |
5050 | |
5051 | might_sleep(); | |
7ff9c073 | 5052 | trace_ext4_mark_inode_dirty(inode, _RET_IP_); |
617ba13b | 5053 | err = ext4_reserve_inode_write(handle, inode, &iloc); |
0390131b FM |
5054 | if (ext4_handle_valid(handle) && |
5055 | EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize && | |
19f5fb7a | 5056 | !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) { |
6dd4ee7c KS |
5057 | /* |
5058 | * We need extra buffer credits since we may write into EA block | |
5059 | * with this same handle. If journal_extend fails, then it will | |
5060 | * only result in a minor loss of functionality for that inode. | |
5061 | * If this is felt to be critical, then e2fsck should be run to | |
5062 | * force a large enough s_min_extra_isize. | |
5063 | */ | |
5064 | if ((jbd2_journal_extend(handle, | |
5065 | EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) { | |
5066 | ret = ext4_expand_extra_isize(inode, | |
5067 | sbi->s_want_extra_isize, | |
5068 | iloc, handle); | |
5069 | if (ret) { | |
19f5fb7a TT |
5070 | ext4_set_inode_state(inode, |
5071 | EXT4_STATE_NO_EXPAND); | |
c1bddad9 AK |
5072 | if (mnt_count != |
5073 | le16_to_cpu(sbi->s_es->s_mnt_count)) { | |
12062ddd | 5074 | ext4_warning(inode->i_sb, |
6dd4ee7c KS |
5075 | "Unable to expand inode %lu. Delete" |
5076 | " some EAs or run e2fsck.", | |
5077 | inode->i_ino); | |
c1bddad9 AK |
5078 | mnt_count = |
5079 | le16_to_cpu(sbi->s_es->s_mnt_count); | |
6dd4ee7c KS |
5080 | } |
5081 | } | |
5082 | } | |
5083 | } | |
ac27a0ec | 5084 | if (!err) |
617ba13b | 5085 | err = ext4_mark_iloc_dirty(handle, inode, &iloc); |
ac27a0ec DK |
5086 | return err; |
5087 | } | |
5088 | ||
5089 | /* | |
617ba13b | 5090 | * ext4_dirty_inode() is called from __mark_inode_dirty() |
ac27a0ec DK |
5091 | * |
5092 | * We're really interested in the case where a file is being extended. | |
5093 | * i_size has been changed by generic_commit_write() and we thus need | |
5094 | * to include the updated inode in the current transaction. | |
5095 | * | |
5dd4056d | 5096 | * Also, dquot_alloc_block() will always dirty the inode when blocks |
ac27a0ec DK |
5097 | * are allocated to the file. |
5098 | * | |
5099 | * If the inode is marked synchronous, we don't honour that here - doing | |
5100 | * so would cause a commit on atime updates, which we don't bother doing. | |
5101 | * We handle synchronous inodes at the highest possible level. | |
5102 | */ | |
aa385729 | 5103 | void ext4_dirty_inode(struct inode *inode, int flags) |
ac27a0ec | 5104 | { |
ac27a0ec DK |
5105 | handle_t *handle; |
5106 | ||
9924a92a | 5107 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 2); |
ac27a0ec DK |
5108 | if (IS_ERR(handle)) |
5109 | goto out; | |
f3dc272f | 5110 | |
f3dc272f CW |
5111 | ext4_mark_inode_dirty(handle, inode); |
5112 | ||
617ba13b | 5113 | ext4_journal_stop(handle); |
ac27a0ec DK |
5114 | out: |
5115 | return; | |
5116 | } | |
5117 | ||
5118 | #if 0 | |
5119 | /* | |
5120 | * Bind an inode's backing buffer_head into this transaction, to prevent | |
5121 | * it from being flushed to disk early. Unlike | |
617ba13b | 5122 | * ext4_reserve_inode_write, this leaves behind no bh reference and |
ac27a0ec DK |
5123 | * returns no iloc structure, so the caller needs to repeat the iloc |
5124 | * lookup to mark the inode dirty later. | |
5125 | */ | |
617ba13b | 5126 | static int ext4_pin_inode(handle_t *handle, struct inode *inode) |
ac27a0ec | 5127 | { |
617ba13b | 5128 | struct ext4_iloc iloc; |
ac27a0ec DK |
5129 | |
5130 | int err = 0; | |
5131 | if (handle) { | |
617ba13b | 5132 | err = ext4_get_inode_loc(inode, &iloc); |
ac27a0ec DK |
5133 | if (!err) { |
5134 | BUFFER_TRACE(iloc.bh, "get_write_access"); | |
dab291af | 5135 | err = jbd2_journal_get_write_access(handle, iloc.bh); |
ac27a0ec | 5136 | if (!err) |
0390131b | 5137 | err = ext4_handle_dirty_metadata(handle, |
73b50c1c | 5138 | NULL, |
0390131b | 5139 | iloc.bh); |
ac27a0ec DK |
5140 | brelse(iloc.bh); |
5141 | } | |
5142 | } | |
617ba13b | 5143 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
5144 | return err; |
5145 | } | |
5146 | #endif | |
5147 | ||
617ba13b | 5148 | int ext4_change_inode_journal_flag(struct inode *inode, int val) |
ac27a0ec DK |
5149 | { |
5150 | journal_t *journal; | |
5151 | handle_t *handle; | |
5152 | int err; | |
5153 | ||
5154 | /* | |
5155 | * We have to be very careful here: changing a data block's | |
5156 | * journaling status dynamically is dangerous. If we write a | |
5157 | * data block to the journal, change the status and then delete | |
5158 | * that block, we risk forgetting to revoke the old log record | |
5159 | * from the journal and so a subsequent replay can corrupt data. | |
5160 | * So, first we make sure that the journal is empty and that | |
5161 | * nobody is changing anything. | |
5162 | */ | |
5163 | ||
617ba13b | 5164 | journal = EXT4_JOURNAL(inode); |
0390131b FM |
5165 | if (!journal) |
5166 | return 0; | |
d699594d | 5167 | if (is_journal_aborted(journal)) |
ac27a0ec | 5168 | return -EROFS; |
2aff57b0 YY |
5169 | /* We have to allocate physical blocks for delalloc blocks |
5170 | * before flushing journal. otherwise delalloc blocks can not | |
5171 | * be allocated any more. even more truncate on delalloc blocks | |
5172 | * could trigger BUG by flushing delalloc blocks in journal. | |
5173 | * There is no delalloc block in non-journal data mode. | |
5174 | */ | |
5175 | if (val && test_opt(inode->i_sb, DELALLOC)) { | |
5176 | err = ext4_alloc_da_blocks(inode); | |
5177 | if (err < 0) | |
5178 | return err; | |
5179 | } | |
ac27a0ec | 5180 | |
17335dcc DM |
5181 | /* Wait for all existing dio workers */ |
5182 | ext4_inode_block_unlocked_dio(inode); | |
5183 | inode_dio_wait(inode); | |
5184 | ||
dab291af | 5185 | jbd2_journal_lock_updates(journal); |
ac27a0ec DK |
5186 | |
5187 | /* | |
5188 | * OK, there are no updates running now, and all cached data is | |
5189 | * synced to disk. We are now in a completely consistent state | |
5190 | * which doesn't have anything in the journal, and we know that | |
5191 | * no filesystem updates are running, so it is safe to modify | |
5192 | * the inode's in-core data-journaling state flag now. | |
5193 | */ | |
5194 | ||
5195 | if (val) | |
12e9b892 | 5196 | ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
5872ddaa YY |
5197 | else { |
5198 | jbd2_journal_flush(journal); | |
12e9b892 | 5199 | ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
5872ddaa | 5200 | } |
617ba13b | 5201 | ext4_set_aops(inode); |
ac27a0ec | 5202 | |
dab291af | 5203 | jbd2_journal_unlock_updates(journal); |
17335dcc | 5204 | ext4_inode_resume_unlocked_dio(inode); |
ac27a0ec DK |
5205 | |
5206 | /* Finally we can mark the inode as dirty. */ | |
5207 | ||
9924a92a | 5208 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 1); |
ac27a0ec DK |
5209 | if (IS_ERR(handle)) |
5210 | return PTR_ERR(handle); | |
5211 | ||
617ba13b | 5212 | err = ext4_mark_inode_dirty(handle, inode); |
0390131b | 5213 | ext4_handle_sync(handle); |
617ba13b MC |
5214 | ext4_journal_stop(handle); |
5215 | ext4_std_error(inode->i_sb, err); | |
ac27a0ec DK |
5216 | |
5217 | return err; | |
5218 | } | |
2e9ee850 AK |
5219 | |
5220 | static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh) | |
5221 | { | |
5222 | return !buffer_mapped(bh); | |
5223 | } | |
5224 | ||
c2ec175c | 5225 | int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
2e9ee850 | 5226 | { |
c2ec175c | 5227 | struct page *page = vmf->page; |
2e9ee850 AK |
5228 | loff_t size; |
5229 | unsigned long len; | |
9ea7df53 | 5230 | int ret; |
2e9ee850 | 5231 | struct file *file = vma->vm_file; |
496ad9aa | 5232 | struct inode *inode = file_inode(file); |
2e9ee850 | 5233 | struct address_space *mapping = inode->i_mapping; |
9ea7df53 JK |
5234 | handle_t *handle; |
5235 | get_block_t *get_block; | |
5236 | int retries = 0; | |
2e9ee850 | 5237 | |
8e8ad8a5 | 5238 | sb_start_pagefault(inode->i_sb); |
041bbb6d | 5239 | file_update_time(vma->vm_file); |
9ea7df53 JK |
5240 | /* Delalloc case is easy... */ |
5241 | if (test_opt(inode->i_sb, DELALLOC) && | |
5242 | !ext4_should_journal_data(inode) && | |
5243 | !ext4_nonda_switch(inode->i_sb)) { | |
5244 | do { | |
5245 | ret = __block_page_mkwrite(vma, vmf, | |
5246 | ext4_da_get_block_prep); | |
5247 | } while (ret == -ENOSPC && | |
5248 | ext4_should_retry_alloc(inode->i_sb, &retries)); | |
5249 | goto out_ret; | |
2e9ee850 | 5250 | } |
0e499890 DW |
5251 | |
5252 | lock_page(page); | |
9ea7df53 JK |
5253 | size = i_size_read(inode); |
5254 | /* Page got truncated from under us? */ | |
5255 | if (page->mapping != mapping || page_offset(page) > size) { | |
5256 | unlock_page(page); | |
5257 | ret = VM_FAULT_NOPAGE; | |
5258 | goto out; | |
0e499890 | 5259 | } |
2e9ee850 AK |
5260 | |
5261 | if (page->index == size >> PAGE_CACHE_SHIFT) | |
5262 | len = size & ~PAGE_CACHE_MASK; | |
5263 | else | |
5264 | len = PAGE_CACHE_SIZE; | |
a827eaff | 5265 | /* |
9ea7df53 JK |
5266 | * Return if we have all the buffers mapped. This avoids the need to do |
5267 | * journal_start/journal_stop which can block and take a long time | |
a827eaff | 5268 | */ |
2e9ee850 | 5269 | if (page_has_buffers(page)) { |
f19d5870 TM |
5270 | if (!ext4_walk_page_buffers(NULL, page_buffers(page), |
5271 | 0, len, NULL, | |
5272 | ext4_bh_unmapped)) { | |
9ea7df53 | 5273 | /* Wait so that we don't change page under IO */ |
1d1d1a76 | 5274 | wait_for_stable_page(page); |
9ea7df53 JK |
5275 | ret = VM_FAULT_LOCKED; |
5276 | goto out; | |
a827eaff | 5277 | } |
2e9ee850 | 5278 | } |
a827eaff | 5279 | unlock_page(page); |
9ea7df53 JK |
5280 | /* OK, we need to fill the hole... */ |
5281 | if (ext4_should_dioread_nolock(inode)) | |
5282 | get_block = ext4_get_block_write; | |
5283 | else | |
5284 | get_block = ext4_get_block; | |
5285 | retry_alloc: | |
9924a92a TT |
5286 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
5287 | ext4_writepage_trans_blocks(inode)); | |
9ea7df53 | 5288 | if (IS_ERR(handle)) { |
c2ec175c | 5289 | ret = VM_FAULT_SIGBUS; |
9ea7df53 JK |
5290 | goto out; |
5291 | } | |
5292 | ret = __block_page_mkwrite(vma, vmf, get_block); | |
5293 | if (!ret && ext4_should_journal_data(inode)) { | |
f19d5870 | 5294 | if (ext4_walk_page_buffers(handle, page_buffers(page), 0, |
9ea7df53 JK |
5295 | PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) { |
5296 | unlock_page(page); | |
5297 | ret = VM_FAULT_SIGBUS; | |
fcbb5515 | 5298 | ext4_journal_stop(handle); |
9ea7df53 JK |
5299 | goto out; |
5300 | } | |
5301 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); | |
5302 | } | |
5303 | ext4_journal_stop(handle); | |
5304 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) | |
5305 | goto retry_alloc; | |
5306 | out_ret: | |
5307 | ret = block_page_mkwrite_return(ret); | |
5308 | out: | |
8e8ad8a5 | 5309 | sb_end_pagefault(inode->i_sb); |
2e9ee850 AK |
5310 | return ret; |
5311 | } |