]>
Commit | Line | Data |
---|---|---|
ac27a0ec | 1 | /* |
617ba13b | 2 | * linux/fs/ext4/inode.c |
ac27a0ec DK |
3 | * |
4 | * Copyright (C) 1992, 1993, 1994, 1995 | |
5 | * Remy Card (card@masi.ibp.fr) | |
6 | * Laboratoire MASI - Institut Blaise Pascal | |
7 | * Universite Pierre et Marie Curie (Paris VI) | |
8 | * | |
9 | * from | |
10 | * | |
11 | * linux/fs/minix/inode.c | |
12 | * | |
13 | * Copyright (C) 1991, 1992 Linus Torvalds | |
14 | * | |
ac27a0ec DK |
15 | * 64-bit file support on 64-bit platforms by Jakub Jelinek |
16 | * (jj@sunsite.ms.mff.cuni.cz) | |
17 | * | |
617ba13b | 18 | * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000 |
ac27a0ec DK |
19 | */ |
20 | ||
ac27a0ec DK |
21 | #include <linux/fs.h> |
22 | #include <linux/time.h> | |
dab291af | 23 | #include <linux/jbd2.h> |
ac27a0ec DK |
24 | #include <linux/highuid.h> |
25 | #include <linux/pagemap.h> | |
26 | #include <linux/quotaops.h> | |
27 | #include <linux/string.h> | |
28 | #include <linux/buffer_head.h> | |
29 | #include <linux/writeback.h> | |
64769240 | 30 | #include <linux/pagevec.h> |
ac27a0ec | 31 | #include <linux/mpage.h> |
e83c1397 | 32 | #include <linux/namei.h> |
ac27a0ec DK |
33 | #include <linux/uio.h> |
34 | #include <linux/bio.h> | |
4c0425ff | 35 | #include <linux/workqueue.h> |
744692dc | 36 | #include <linux/kernel.h> |
6db26ffc | 37 | #include <linux/printk.h> |
5a0e3ad6 | 38 | #include <linux/slab.h> |
a8901d34 | 39 | #include <linux/ratelimit.h> |
9bffad1e | 40 | |
3dcf5451 | 41 | #include "ext4_jbd2.h" |
ac27a0ec DK |
42 | #include "xattr.h" |
43 | #include "acl.h" | |
9f125d64 | 44 | #include "truncate.h" |
ac27a0ec | 45 | |
9bffad1e TT |
46 | #include <trace/events/ext4.h> |
47 | ||
a1d6cc56 AK |
48 | #define MPAGE_DA_EXTENT_TAIL 0x01 |
49 | ||
814525f4 DW |
50 | static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw, |
51 | struct ext4_inode_info *ei) | |
52 | { | |
53 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
54 | __u16 csum_lo; | |
55 | __u16 csum_hi = 0; | |
56 | __u32 csum; | |
57 | ||
58 | csum_lo = raw->i_checksum_lo; | |
59 | raw->i_checksum_lo = 0; | |
60 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && | |
61 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) { | |
62 | csum_hi = raw->i_checksum_hi; | |
63 | raw->i_checksum_hi = 0; | |
64 | } | |
65 | ||
66 | csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw, | |
67 | EXT4_INODE_SIZE(inode->i_sb)); | |
68 | ||
69 | raw->i_checksum_lo = csum_lo; | |
70 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && | |
71 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) | |
72 | raw->i_checksum_hi = csum_hi; | |
73 | ||
74 | return csum; | |
75 | } | |
76 | ||
77 | static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw, | |
78 | struct ext4_inode_info *ei) | |
79 | { | |
80 | __u32 provided, calculated; | |
81 | ||
82 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != | |
83 | cpu_to_le32(EXT4_OS_LINUX) || | |
84 | !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, | |
85 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) | |
86 | return 1; | |
87 | ||
88 | provided = le16_to_cpu(raw->i_checksum_lo); | |
89 | calculated = ext4_inode_csum(inode, raw, ei); | |
90 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && | |
91 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) | |
92 | provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16; | |
93 | else | |
94 | calculated &= 0xFFFF; | |
95 | ||
96 | return provided == calculated; | |
97 | } | |
98 | ||
99 | static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw, | |
100 | struct ext4_inode_info *ei) | |
101 | { | |
102 | __u32 csum; | |
103 | ||
104 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != | |
105 | cpu_to_le32(EXT4_OS_LINUX) || | |
106 | !EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb, | |
107 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) | |
108 | return; | |
109 | ||
110 | csum = ext4_inode_csum(inode, raw, ei); | |
111 | raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF); | |
112 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && | |
113 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) | |
114 | raw->i_checksum_hi = cpu_to_le16(csum >> 16); | |
115 | } | |
116 | ||
678aaf48 JK |
117 | static inline int ext4_begin_ordered_truncate(struct inode *inode, |
118 | loff_t new_size) | |
119 | { | |
7ff9c073 | 120 | trace_ext4_begin_ordered_truncate(inode, new_size); |
8aefcd55 TT |
121 | /* |
122 | * If jinode is zero, then we never opened the file for | |
123 | * writing, so there's no need to call | |
124 | * jbd2_journal_begin_ordered_truncate() since there's no | |
125 | * outstanding writes we need to flush. | |
126 | */ | |
127 | if (!EXT4_I(inode)->jinode) | |
128 | return 0; | |
129 | return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode), | |
130 | EXT4_I(inode)->jinode, | |
131 | new_size); | |
678aaf48 JK |
132 | } |
133 | ||
64769240 | 134 | static void ext4_invalidatepage(struct page *page, unsigned long offset); |
cb20d518 TT |
135 | static int noalloc_get_block_write(struct inode *inode, sector_t iblock, |
136 | struct buffer_head *bh_result, int create); | |
137 | static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode); | |
138 | static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate); | |
139 | static int __ext4_journalled_writepage(struct page *page, unsigned int len); | |
140 | static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh); | |
5f163cc7 ES |
141 | static int ext4_discard_partial_page_buffers_no_lock(handle_t *handle, |
142 | struct inode *inode, struct page *page, loff_t from, | |
143 | loff_t length, int flags); | |
64769240 | 144 | |
ac27a0ec DK |
145 | /* |
146 | * Test whether an inode is a fast symlink. | |
147 | */ | |
617ba13b | 148 | static int ext4_inode_is_fast_symlink(struct inode *inode) |
ac27a0ec | 149 | { |
617ba13b | 150 | int ea_blocks = EXT4_I(inode)->i_file_acl ? |
ac27a0ec DK |
151 | (inode->i_sb->s_blocksize >> 9) : 0; |
152 | ||
153 | return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0); | |
154 | } | |
155 | ||
ac27a0ec DK |
156 | /* |
157 | * Restart the transaction associated with *handle. This does a commit, | |
158 | * so before we call here everything must be consistently dirtied against | |
159 | * this transaction. | |
160 | */ | |
fa5d1113 | 161 | int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode, |
487caeef | 162 | int nblocks) |
ac27a0ec | 163 | { |
487caeef JK |
164 | int ret; |
165 | ||
166 | /* | |
e35fd660 | 167 | * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this |
487caeef JK |
168 | * moment, get_block can be called only for blocks inside i_size since |
169 | * page cache has been already dropped and writes are blocked by | |
170 | * i_mutex. So we can safely drop the i_data_sem here. | |
171 | */ | |
0390131b | 172 | BUG_ON(EXT4_JOURNAL(inode) == NULL); |
ac27a0ec | 173 | jbd_debug(2, "restarting handle %p\n", handle); |
487caeef | 174 | up_write(&EXT4_I(inode)->i_data_sem); |
8e8eaabe | 175 | ret = ext4_journal_restart(handle, nblocks); |
487caeef | 176 | down_write(&EXT4_I(inode)->i_data_sem); |
fa5d1113 | 177 | ext4_discard_preallocations(inode); |
487caeef JK |
178 | |
179 | return ret; | |
ac27a0ec DK |
180 | } |
181 | ||
182 | /* | |
183 | * Called at the last iput() if i_nlink is zero. | |
184 | */ | |
0930fcc1 | 185 | void ext4_evict_inode(struct inode *inode) |
ac27a0ec DK |
186 | { |
187 | handle_t *handle; | |
bc965ab3 | 188 | int err; |
ac27a0ec | 189 | |
7ff9c073 | 190 | trace_ext4_evict_inode(inode); |
2581fdc8 | 191 | |
2581fdc8 JZ |
192 | ext4_ioend_wait(inode); |
193 | ||
0930fcc1 | 194 | if (inode->i_nlink) { |
2d859db3 JK |
195 | /* |
196 | * When journalling data dirty buffers are tracked only in the | |
197 | * journal. So although mm thinks everything is clean and | |
198 | * ready for reaping the inode might still have some pages to | |
199 | * write in the running transaction or waiting to be | |
200 | * checkpointed. Thus calling jbd2_journal_invalidatepage() | |
201 | * (via truncate_inode_pages()) to discard these buffers can | |
202 | * cause data loss. Also even if we did not discard these | |
203 | * buffers, we would have no way to find them after the inode | |
204 | * is reaped and thus user could see stale data if he tries to | |
205 | * read them before the transaction is checkpointed. So be | |
206 | * careful and force everything to disk here... We use | |
207 | * ei->i_datasync_tid to store the newest transaction | |
208 | * containing inode's data. | |
209 | * | |
210 | * Note that directories do not have this problem because they | |
211 | * don't use page cache. | |
212 | */ | |
213 | if (ext4_should_journal_data(inode) && | |
214 | (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode))) { | |
215 | journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; | |
216 | tid_t commit_tid = EXT4_I(inode)->i_datasync_tid; | |
217 | ||
218 | jbd2_log_start_commit(journal, commit_tid); | |
219 | jbd2_log_wait_commit(journal, commit_tid); | |
220 | filemap_write_and_wait(&inode->i_data); | |
221 | } | |
0930fcc1 AV |
222 | truncate_inode_pages(&inode->i_data, 0); |
223 | goto no_delete; | |
224 | } | |
225 | ||
907f4554 | 226 | if (!is_bad_inode(inode)) |
871a2931 | 227 | dquot_initialize(inode); |
907f4554 | 228 | |
678aaf48 JK |
229 | if (ext4_should_order_data(inode)) |
230 | ext4_begin_ordered_truncate(inode, 0); | |
ac27a0ec DK |
231 | truncate_inode_pages(&inode->i_data, 0); |
232 | ||
233 | if (is_bad_inode(inode)) | |
234 | goto no_delete; | |
235 | ||
8e8ad8a5 JK |
236 | /* |
237 | * Protect us against freezing - iput() caller didn't have to have any | |
238 | * protection against it | |
239 | */ | |
240 | sb_start_intwrite(inode->i_sb); | |
9f125d64 | 241 | handle = ext4_journal_start(inode, 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 TT |
348 | if (unlikely(used > ei->i_reserved_data_blocks)) { |
349 | ext4_msg(inode->i_sb, KERN_NOTICE, "%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 BF |
357 | if (unlikely(ei->i_allocated_meta_blocks > ei->i_reserved_meta_blocks)) { |
358 | ext4_msg(inode->i_sb, KERN_NOTICE, "%s: ino %lu, allocated %d " | |
359 | "with only %d reserved metadata blocks\n", __func__, | |
360 | inode->i_ino, ei->i_allocated_meta_blocks, | |
361 | ei->i_reserved_meta_blocks); | |
362 | WARN_ON(1); | |
363 | ei->i_allocated_meta_blocks = ei->i_reserved_meta_blocks; | |
364 | } | |
365 | ||
0637c6f4 TT |
366 | /* Update per-inode reservations */ |
367 | ei->i_reserved_data_blocks -= used; | |
0637c6f4 | 368 | ei->i_reserved_meta_blocks -= ei->i_allocated_meta_blocks; |
57042651 | 369 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, |
72b8ab9d | 370 | used + ei->i_allocated_meta_blocks); |
0637c6f4 | 371 | ei->i_allocated_meta_blocks = 0; |
6bc6e63f | 372 | |
0637c6f4 TT |
373 | if (ei->i_reserved_data_blocks == 0) { |
374 | /* | |
375 | * We can release all of the reserved metadata blocks | |
376 | * only when we have written all of the delayed | |
377 | * allocation blocks. | |
378 | */ | |
57042651 | 379 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, |
72b8ab9d | 380 | ei->i_reserved_meta_blocks); |
ee5f4d9c | 381 | ei->i_reserved_meta_blocks = 0; |
9d0be502 | 382 | ei->i_da_metadata_calc_len = 0; |
6bc6e63f | 383 | } |
12219aea | 384 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
60e58e0f | 385 | |
72b8ab9d ES |
386 | /* Update quota subsystem for data blocks */ |
387 | if (quota_claim) | |
7b415bf6 | 388 | dquot_claim_block(inode, EXT4_C2B(sbi, used)); |
72b8ab9d | 389 | else { |
5f634d06 AK |
390 | /* |
391 | * We did fallocate with an offset that is already delayed | |
392 | * allocated. So on delayed allocated writeback we should | |
72b8ab9d | 393 | * not re-claim the quota for fallocated blocks. |
5f634d06 | 394 | */ |
7b415bf6 | 395 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, used)); |
5f634d06 | 396 | } |
d6014301 AK |
397 | |
398 | /* | |
399 | * If we have done all the pending block allocations and if | |
400 | * there aren't any writers on the inode, we can discard the | |
401 | * inode's preallocations. | |
402 | */ | |
0637c6f4 TT |
403 | if ((ei->i_reserved_data_blocks == 0) && |
404 | (atomic_read(&inode->i_writecount) == 0)) | |
d6014301 | 405 | ext4_discard_preallocations(inode); |
12219aea AK |
406 | } |
407 | ||
e29136f8 | 408 | static int __check_block_validity(struct inode *inode, const char *func, |
c398eda0 TT |
409 | unsigned int line, |
410 | struct ext4_map_blocks *map) | |
6fd058f7 | 411 | { |
24676da4 TT |
412 | if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk, |
413 | map->m_len)) { | |
c398eda0 TT |
414 | ext4_error_inode(inode, func, line, map->m_pblk, |
415 | "lblock %lu mapped to illegal pblock " | |
416 | "(length %d)", (unsigned long) map->m_lblk, | |
417 | map->m_len); | |
6fd058f7 TT |
418 | return -EIO; |
419 | } | |
420 | return 0; | |
421 | } | |
422 | ||
e29136f8 | 423 | #define check_block_validity(inode, map) \ |
c398eda0 | 424 | __check_block_validity((inode), __func__, __LINE__, (map)) |
e29136f8 | 425 | |
55138e0b | 426 | /* |
1f94533d TT |
427 | * Return the number of contiguous dirty pages in a given inode |
428 | * starting at page frame idx. | |
55138e0b TT |
429 | */ |
430 | static pgoff_t ext4_num_dirty_pages(struct inode *inode, pgoff_t idx, | |
431 | unsigned int max_pages) | |
432 | { | |
433 | struct address_space *mapping = inode->i_mapping; | |
434 | pgoff_t index; | |
435 | struct pagevec pvec; | |
436 | pgoff_t num = 0; | |
437 | int i, nr_pages, done = 0; | |
438 | ||
439 | if (max_pages == 0) | |
440 | return 0; | |
441 | pagevec_init(&pvec, 0); | |
442 | while (!done) { | |
443 | index = idx; | |
444 | nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, | |
445 | PAGECACHE_TAG_DIRTY, | |
446 | (pgoff_t)PAGEVEC_SIZE); | |
447 | if (nr_pages == 0) | |
448 | break; | |
449 | for (i = 0; i < nr_pages; i++) { | |
450 | struct page *page = pvec.pages[i]; | |
451 | struct buffer_head *bh, *head; | |
452 | ||
453 | lock_page(page); | |
454 | if (unlikely(page->mapping != mapping) || | |
455 | !PageDirty(page) || | |
456 | PageWriteback(page) || | |
457 | page->index != idx) { | |
458 | done = 1; | |
459 | unlock_page(page); | |
460 | break; | |
461 | } | |
1f94533d TT |
462 | if (page_has_buffers(page)) { |
463 | bh = head = page_buffers(page); | |
464 | do { | |
465 | if (!buffer_delay(bh) && | |
466 | !buffer_unwritten(bh)) | |
467 | done = 1; | |
468 | bh = bh->b_this_page; | |
469 | } while (!done && (bh != head)); | |
470 | } | |
55138e0b TT |
471 | unlock_page(page); |
472 | if (done) | |
473 | break; | |
474 | idx++; | |
475 | num++; | |
659c6009 ES |
476 | if (num >= max_pages) { |
477 | done = 1; | |
55138e0b | 478 | break; |
659c6009 | 479 | } |
55138e0b TT |
480 | } |
481 | pagevec_release(&pvec); | |
482 | } | |
483 | return num; | |
484 | } | |
485 | ||
5356f261 AK |
486 | /* |
487 | * Sets the BH_Da_Mapped bit on the buffer heads corresponding to the given map. | |
488 | */ | |
489 | static void set_buffers_da_mapped(struct inode *inode, | |
490 | struct ext4_map_blocks *map) | |
491 | { | |
492 | struct address_space *mapping = inode->i_mapping; | |
493 | struct pagevec pvec; | |
494 | int i, nr_pages; | |
495 | pgoff_t index, end; | |
496 | ||
497 | index = map->m_lblk >> (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
498 | end = (map->m_lblk + map->m_len - 1) >> | |
499 | (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
500 | ||
501 | pagevec_init(&pvec, 0); | |
502 | while (index <= end) { | |
503 | nr_pages = pagevec_lookup(&pvec, mapping, index, | |
504 | min(end - index + 1, | |
505 | (pgoff_t)PAGEVEC_SIZE)); | |
506 | if (nr_pages == 0) | |
507 | break; | |
508 | for (i = 0; i < nr_pages; i++) { | |
509 | struct page *page = pvec.pages[i]; | |
510 | struct buffer_head *bh, *head; | |
511 | ||
512 | if (unlikely(page->mapping != mapping) || | |
513 | !PageDirty(page)) | |
514 | break; | |
515 | ||
516 | if (page_has_buffers(page)) { | |
517 | bh = head = page_buffers(page); | |
518 | do { | |
519 | set_buffer_da_mapped(bh); | |
520 | bh = bh->b_this_page; | |
521 | } while (bh != head); | |
522 | } | |
523 | index++; | |
524 | } | |
525 | pagevec_release(&pvec); | |
526 | } | |
527 | } | |
528 | ||
f5ab0d1f | 529 | /* |
e35fd660 | 530 | * The ext4_map_blocks() function tries to look up the requested blocks, |
2b2d6d01 | 531 | * and returns if the blocks are already mapped. |
f5ab0d1f | 532 | * |
f5ab0d1f MC |
533 | * Otherwise it takes the write lock of the i_data_sem and allocate blocks |
534 | * and store the allocated blocks in the result buffer head and mark it | |
535 | * mapped. | |
536 | * | |
e35fd660 TT |
537 | * If file type is extents based, it will call ext4_ext_map_blocks(), |
538 | * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping | |
f5ab0d1f MC |
539 | * based files |
540 | * | |
541 | * On success, it returns the number of blocks being mapped or allocate. | |
542 | * if create==0 and the blocks are pre-allocated and uninitialized block, | |
543 | * the result buffer head is unmapped. If the create ==1, it will make sure | |
544 | * the buffer head is mapped. | |
545 | * | |
546 | * It returns 0 if plain look up failed (blocks have not been allocated), in | |
df3ab170 | 547 | * that case, buffer head is unmapped |
f5ab0d1f MC |
548 | * |
549 | * It returns the error in case of allocation failure. | |
550 | */ | |
e35fd660 TT |
551 | int ext4_map_blocks(handle_t *handle, struct inode *inode, |
552 | struct ext4_map_blocks *map, int flags) | |
0e855ac8 AK |
553 | { |
554 | int retval; | |
f5ab0d1f | 555 | |
e35fd660 TT |
556 | map->m_flags = 0; |
557 | ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u," | |
558 | "logical block %lu\n", inode->i_ino, flags, map->m_len, | |
559 | (unsigned long) map->m_lblk); | |
4df3d265 | 560 | /* |
b920c755 TT |
561 | * Try to see if we can get the block without requesting a new |
562 | * file system block. | |
4df3d265 | 563 | */ |
729f52c6 ZL |
564 | if (!(flags & EXT4_GET_BLOCKS_NO_LOCK)) |
565 | down_read((&EXT4_I(inode)->i_data_sem)); | |
12e9b892 | 566 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
a4e5d88b DM |
567 | retval = ext4_ext_map_blocks(handle, inode, map, flags & |
568 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
0e855ac8 | 569 | } else { |
a4e5d88b DM |
570 | retval = ext4_ind_map_blocks(handle, inode, map, flags & |
571 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
0e855ac8 | 572 | } |
729f52c6 ZL |
573 | if (!(flags & EXT4_GET_BLOCKS_NO_LOCK)) |
574 | up_read((&EXT4_I(inode)->i_data_sem)); | |
f5ab0d1f | 575 | |
e35fd660 | 576 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
51865fda ZL |
577 | int ret; |
578 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) { | |
579 | /* delayed alloc may be allocated by fallocate and | |
580 | * coverted to initialized by directIO. | |
581 | * we need to handle delayed extent here. | |
582 | */ | |
583 | down_write((&EXT4_I(inode)->i_data_sem)); | |
584 | goto delayed_mapped; | |
585 | } | |
586 | ret = check_block_validity(inode, map); | |
6fd058f7 TT |
587 | if (ret != 0) |
588 | return ret; | |
589 | } | |
590 | ||
f5ab0d1f | 591 | /* If it is only a block(s) look up */ |
c2177057 | 592 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) |
f5ab0d1f MC |
593 | return retval; |
594 | ||
595 | /* | |
596 | * Returns if the blocks have already allocated | |
597 | * | |
598 | * Note that if blocks have been preallocated | |
df3ab170 | 599 | * ext4_ext_get_block() returns the create = 0 |
f5ab0d1f MC |
600 | * with buffer head unmapped. |
601 | */ | |
e35fd660 | 602 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) |
4df3d265 AK |
603 | return retval; |
604 | ||
2a8964d6 AK |
605 | /* |
606 | * When we call get_blocks without the create flag, the | |
607 | * BH_Unwritten flag could have gotten set if the blocks | |
608 | * requested were part of a uninitialized extent. We need to | |
609 | * clear this flag now that we are committed to convert all or | |
610 | * part of the uninitialized extent to be an initialized | |
611 | * extent. This is because we need to avoid the combination | |
612 | * of BH_Unwritten and BH_Mapped flags being simultaneously | |
613 | * set on the buffer_head. | |
614 | */ | |
e35fd660 | 615 | map->m_flags &= ~EXT4_MAP_UNWRITTEN; |
2a8964d6 | 616 | |
4df3d265 | 617 | /* |
f5ab0d1f MC |
618 | * New blocks allocate and/or writing to uninitialized extent |
619 | * will possibly result in updating i_data, so we take | |
620 | * the write lock of i_data_sem, and call get_blocks() | |
621 | * with create == 1 flag. | |
4df3d265 AK |
622 | */ |
623 | down_write((&EXT4_I(inode)->i_data_sem)); | |
d2a17637 MC |
624 | |
625 | /* | |
626 | * if the caller is from delayed allocation writeout path | |
627 | * we have already reserved fs blocks for allocation | |
628 | * let the underlying get_block() function know to | |
629 | * avoid double accounting | |
630 | */ | |
c2177057 | 631 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) |
f2321097 | 632 | ext4_set_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED); |
4df3d265 AK |
633 | /* |
634 | * We need to check for EXT4 here because migrate | |
635 | * could have changed the inode type in between | |
636 | */ | |
12e9b892 | 637 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
e35fd660 | 638 | retval = ext4_ext_map_blocks(handle, inode, map, flags); |
0e855ac8 | 639 | } else { |
e35fd660 | 640 | retval = ext4_ind_map_blocks(handle, inode, map, flags); |
267e4db9 | 641 | |
e35fd660 | 642 | if (retval > 0 && map->m_flags & EXT4_MAP_NEW) { |
267e4db9 AK |
643 | /* |
644 | * We allocated new blocks which will result in | |
645 | * i_data's format changing. Force the migrate | |
646 | * to fail by clearing migrate flags | |
647 | */ | |
19f5fb7a | 648 | ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE); |
267e4db9 | 649 | } |
d2a17637 | 650 | |
5f634d06 AK |
651 | /* |
652 | * Update reserved blocks/metadata blocks after successful | |
653 | * block allocation which had been deferred till now. We don't | |
654 | * support fallocate for non extent files. So we can update | |
655 | * reserve space here. | |
656 | */ | |
657 | if ((retval > 0) && | |
1296cc85 | 658 | (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)) |
5f634d06 AK |
659 | ext4_da_update_reserve_space(inode, retval, 1); |
660 | } | |
5356f261 | 661 | if (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) { |
f2321097 | 662 | ext4_clear_inode_state(inode, EXT4_STATE_DELALLOC_RESERVED); |
2ac3b6e0 | 663 | |
5356f261 AK |
664 | /* If we have successfully mapped the delayed allocated blocks, |
665 | * set the BH_Da_Mapped bit on them. Its important to do this | |
666 | * under the protection of i_data_sem. | |
667 | */ | |
51865fda ZL |
668 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
669 | int ret; | |
5356f261 | 670 | set_buffers_da_mapped(inode, map); |
51865fda ZL |
671 | delayed_mapped: |
672 | /* delayed allocation blocks has been allocated */ | |
673 | ret = ext4_es_remove_extent(inode, map->m_lblk, | |
674 | map->m_len); | |
675 | if (ret < 0) | |
676 | retval = ret; | |
677 | } | |
5356f261 AK |
678 | } |
679 | ||
4df3d265 | 680 | up_write((&EXT4_I(inode)->i_data_sem)); |
e35fd660 | 681 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
e29136f8 | 682 | int ret = check_block_validity(inode, map); |
6fd058f7 TT |
683 | if (ret != 0) |
684 | return ret; | |
685 | } | |
0e855ac8 AK |
686 | return retval; |
687 | } | |
688 | ||
f3bd1f3f MC |
689 | /* Maximum number of blocks we map for direct IO at once. */ |
690 | #define DIO_MAX_BLOCKS 4096 | |
691 | ||
2ed88685 TT |
692 | static int _ext4_get_block(struct inode *inode, sector_t iblock, |
693 | struct buffer_head *bh, int flags) | |
ac27a0ec | 694 | { |
3e4fdaf8 | 695 | handle_t *handle = ext4_journal_current_handle(); |
2ed88685 | 696 | struct ext4_map_blocks map; |
7fb5409d | 697 | int ret = 0, started = 0; |
f3bd1f3f | 698 | int dio_credits; |
ac27a0ec | 699 | |
2ed88685 TT |
700 | map.m_lblk = iblock; |
701 | map.m_len = bh->b_size >> inode->i_blkbits; | |
702 | ||
8b0f165f | 703 | if (flags && !(flags & EXT4_GET_BLOCKS_NO_LOCK) && !handle) { |
7fb5409d | 704 | /* Direct IO write... */ |
2ed88685 TT |
705 | if (map.m_len > DIO_MAX_BLOCKS) |
706 | map.m_len = DIO_MAX_BLOCKS; | |
707 | dio_credits = ext4_chunk_trans_blocks(inode, map.m_len); | |
f3bd1f3f | 708 | handle = ext4_journal_start(inode, dio_credits); |
7fb5409d | 709 | if (IS_ERR(handle)) { |
ac27a0ec | 710 | ret = PTR_ERR(handle); |
2ed88685 | 711 | return ret; |
ac27a0ec | 712 | } |
7fb5409d | 713 | started = 1; |
ac27a0ec DK |
714 | } |
715 | ||
2ed88685 | 716 | ret = ext4_map_blocks(handle, inode, &map, flags); |
7fb5409d | 717 | if (ret > 0) { |
2ed88685 TT |
718 | map_bh(bh, inode->i_sb, map.m_pblk); |
719 | bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags; | |
720 | bh->b_size = inode->i_sb->s_blocksize * map.m_len; | |
7fb5409d | 721 | ret = 0; |
ac27a0ec | 722 | } |
7fb5409d JK |
723 | if (started) |
724 | ext4_journal_stop(handle); | |
ac27a0ec DK |
725 | return ret; |
726 | } | |
727 | ||
2ed88685 TT |
728 | int ext4_get_block(struct inode *inode, sector_t iblock, |
729 | struct buffer_head *bh, int create) | |
730 | { | |
731 | return _ext4_get_block(inode, iblock, bh, | |
732 | create ? EXT4_GET_BLOCKS_CREATE : 0); | |
733 | } | |
734 | ||
ac27a0ec DK |
735 | /* |
736 | * `handle' can be NULL if create is zero | |
737 | */ | |
617ba13b | 738 | struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode, |
725d26d3 | 739 | ext4_lblk_t block, int create, int *errp) |
ac27a0ec | 740 | { |
2ed88685 TT |
741 | struct ext4_map_blocks map; |
742 | struct buffer_head *bh; | |
ac27a0ec DK |
743 | int fatal = 0, err; |
744 | ||
745 | J_ASSERT(handle != NULL || create == 0); | |
746 | ||
2ed88685 TT |
747 | map.m_lblk = block; |
748 | map.m_len = 1; | |
749 | err = ext4_map_blocks(handle, inode, &map, | |
750 | create ? EXT4_GET_BLOCKS_CREATE : 0); | |
ac27a0ec | 751 | |
90b0a973 CM |
752 | /* ensure we send some value back into *errp */ |
753 | *errp = 0; | |
754 | ||
2ed88685 TT |
755 | if (err < 0) |
756 | *errp = err; | |
757 | if (err <= 0) | |
758 | return NULL; | |
2ed88685 TT |
759 | |
760 | bh = sb_getblk(inode->i_sb, map.m_pblk); | |
761 | if (!bh) { | |
762 | *errp = -EIO; | |
763 | return NULL; | |
ac27a0ec | 764 | } |
2ed88685 TT |
765 | if (map.m_flags & EXT4_MAP_NEW) { |
766 | J_ASSERT(create != 0); | |
767 | J_ASSERT(handle != NULL); | |
ac27a0ec | 768 | |
2ed88685 TT |
769 | /* |
770 | * Now that we do not always journal data, we should | |
771 | * keep in mind whether this should always journal the | |
772 | * new buffer as metadata. For now, regular file | |
773 | * writes use ext4_get_block instead, so it's not a | |
774 | * problem. | |
775 | */ | |
776 | lock_buffer(bh); | |
777 | BUFFER_TRACE(bh, "call get_create_access"); | |
778 | fatal = ext4_journal_get_create_access(handle, bh); | |
779 | if (!fatal && !buffer_uptodate(bh)) { | |
780 | memset(bh->b_data, 0, inode->i_sb->s_blocksize); | |
781 | set_buffer_uptodate(bh); | |
ac27a0ec | 782 | } |
2ed88685 TT |
783 | unlock_buffer(bh); |
784 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); | |
785 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
786 | if (!fatal) | |
787 | fatal = err; | |
788 | } else { | |
789 | BUFFER_TRACE(bh, "not a new buffer"); | |
ac27a0ec | 790 | } |
2ed88685 TT |
791 | if (fatal) { |
792 | *errp = fatal; | |
793 | brelse(bh); | |
794 | bh = NULL; | |
795 | } | |
796 | return bh; | |
ac27a0ec DK |
797 | } |
798 | ||
617ba13b | 799 | struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode, |
725d26d3 | 800 | ext4_lblk_t block, int create, int *err) |
ac27a0ec | 801 | { |
af5bc92d | 802 | struct buffer_head *bh; |
ac27a0ec | 803 | |
617ba13b | 804 | bh = ext4_getblk(handle, inode, block, create, err); |
ac27a0ec DK |
805 | if (!bh) |
806 | return bh; | |
807 | if (buffer_uptodate(bh)) | |
808 | return bh; | |
65299a3b | 809 | ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &bh); |
ac27a0ec DK |
810 | wait_on_buffer(bh); |
811 | if (buffer_uptodate(bh)) | |
812 | return bh; | |
813 | put_bh(bh); | |
814 | *err = -EIO; | |
815 | return NULL; | |
816 | } | |
817 | ||
af5bc92d TT |
818 | static int walk_page_buffers(handle_t *handle, |
819 | struct buffer_head *head, | |
820 | unsigned from, | |
821 | unsigned to, | |
822 | int *partial, | |
823 | int (*fn)(handle_t *handle, | |
824 | struct buffer_head *bh)) | |
ac27a0ec DK |
825 | { |
826 | struct buffer_head *bh; | |
827 | unsigned block_start, block_end; | |
828 | unsigned blocksize = head->b_size; | |
829 | int err, ret = 0; | |
830 | struct buffer_head *next; | |
831 | ||
af5bc92d TT |
832 | for (bh = head, block_start = 0; |
833 | ret == 0 && (bh != head || !block_start); | |
de9a55b8 | 834 | block_start = block_end, bh = next) { |
ac27a0ec DK |
835 | next = bh->b_this_page; |
836 | block_end = block_start + blocksize; | |
837 | if (block_end <= from || block_start >= to) { | |
838 | if (partial && !buffer_uptodate(bh)) | |
839 | *partial = 1; | |
840 | continue; | |
841 | } | |
842 | err = (*fn)(handle, bh); | |
843 | if (!ret) | |
844 | ret = err; | |
845 | } | |
846 | return ret; | |
847 | } | |
848 | ||
849 | /* | |
850 | * To preserve ordering, it is essential that the hole instantiation and | |
851 | * the data write be encapsulated in a single transaction. We cannot | |
617ba13b | 852 | * close off a transaction and start a new one between the ext4_get_block() |
dab291af | 853 | * and the commit_write(). So doing the jbd2_journal_start at the start of |
ac27a0ec DK |
854 | * prepare_write() is the right place. |
855 | * | |
617ba13b MC |
856 | * Also, this function can nest inside ext4_writepage() -> |
857 | * block_write_full_page(). In that case, we *know* that ext4_writepage() | |
ac27a0ec DK |
858 | * has generated enough buffer credits to do the whole page. So we won't |
859 | * block on the journal in that case, which is good, because the caller may | |
860 | * be PF_MEMALLOC. | |
861 | * | |
617ba13b | 862 | * By accident, ext4 can be reentered when a transaction is open via |
ac27a0ec DK |
863 | * quota file writes. If we were to commit the transaction while thus |
864 | * reentered, there can be a deadlock - we would be holding a quota | |
865 | * lock, and the commit would never complete if another thread had a | |
866 | * transaction open and was blocking on the quota lock - a ranking | |
867 | * violation. | |
868 | * | |
dab291af | 869 | * So what we do is to rely on the fact that jbd2_journal_stop/journal_start |
ac27a0ec DK |
870 | * will _not_ run commit under these circumstances because handle->h_ref |
871 | * is elevated. We'll still have enough credits for the tiny quotafile | |
872 | * write. | |
873 | */ | |
874 | static int do_journal_get_write_access(handle_t *handle, | |
de9a55b8 | 875 | struct buffer_head *bh) |
ac27a0ec | 876 | { |
56d35a4c JK |
877 | int dirty = buffer_dirty(bh); |
878 | int ret; | |
879 | ||
ac27a0ec DK |
880 | if (!buffer_mapped(bh) || buffer_freed(bh)) |
881 | return 0; | |
56d35a4c | 882 | /* |
ebdec241 | 883 | * __block_write_begin() could have dirtied some buffers. Clean |
56d35a4c JK |
884 | * the dirty bit as jbd2_journal_get_write_access() could complain |
885 | * otherwise about fs integrity issues. Setting of the dirty bit | |
ebdec241 | 886 | * by __block_write_begin() isn't a real problem here as we clear |
56d35a4c JK |
887 | * the bit before releasing a page lock and thus writeback cannot |
888 | * ever write the buffer. | |
889 | */ | |
890 | if (dirty) | |
891 | clear_buffer_dirty(bh); | |
892 | ret = ext4_journal_get_write_access(handle, bh); | |
893 | if (!ret && dirty) | |
894 | ret = ext4_handle_dirty_metadata(handle, NULL, bh); | |
895 | return ret; | |
ac27a0ec DK |
896 | } |
897 | ||
744692dc JZ |
898 | static int ext4_get_block_write(struct inode *inode, sector_t iblock, |
899 | struct buffer_head *bh_result, int create); | |
8b0f165f AP |
900 | static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock, |
901 | struct buffer_head *bh_result, int create); | |
bfc1af65 | 902 | static int ext4_write_begin(struct file *file, struct address_space *mapping, |
de9a55b8 TT |
903 | loff_t pos, unsigned len, unsigned flags, |
904 | struct page **pagep, void **fsdata) | |
ac27a0ec | 905 | { |
af5bc92d | 906 | struct inode *inode = mapping->host; |
1938a150 | 907 | int ret, needed_blocks; |
ac27a0ec DK |
908 | handle_t *handle; |
909 | int retries = 0; | |
af5bc92d | 910 | struct page *page; |
de9a55b8 | 911 | pgoff_t index; |
af5bc92d | 912 | unsigned from, to; |
bfc1af65 | 913 | |
9bffad1e | 914 | trace_ext4_write_begin(inode, pos, len, flags); |
1938a150 AK |
915 | /* |
916 | * Reserve one block more for addition to orphan list in case | |
917 | * we allocate blocks but write fails for some reason | |
918 | */ | |
919 | needed_blocks = ext4_writepage_trans_blocks(inode) + 1; | |
de9a55b8 | 920 | index = pos >> PAGE_CACHE_SHIFT; |
af5bc92d TT |
921 | from = pos & (PAGE_CACHE_SIZE - 1); |
922 | to = from + len; | |
ac27a0ec DK |
923 | |
924 | retry: | |
af5bc92d TT |
925 | handle = ext4_journal_start(inode, needed_blocks); |
926 | if (IS_ERR(handle)) { | |
927 | ret = PTR_ERR(handle); | |
928 | goto out; | |
7479d2b9 | 929 | } |
ac27a0ec | 930 | |
ebd3610b JK |
931 | /* We cannot recurse into the filesystem as the transaction is already |
932 | * started */ | |
933 | flags |= AOP_FLAG_NOFS; | |
934 | ||
54566b2c | 935 | page = grab_cache_page_write_begin(mapping, index, flags); |
cf108bca JK |
936 | if (!page) { |
937 | ext4_journal_stop(handle); | |
938 | ret = -ENOMEM; | |
939 | goto out; | |
940 | } | |
941 | *pagep = page; | |
942 | ||
744692dc | 943 | if (ext4_should_dioread_nolock(inode)) |
6e1db88d | 944 | ret = __block_write_begin(page, pos, len, ext4_get_block_write); |
744692dc | 945 | else |
6e1db88d | 946 | ret = __block_write_begin(page, pos, len, ext4_get_block); |
bfc1af65 NP |
947 | |
948 | if (!ret && ext4_should_journal_data(inode)) { | |
ac27a0ec DK |
949 | ret = walk_page_buffers(handle, page_buffers(page), |
950 | from, to, NULL, do_journal_get_write_access); | |
951 | } | |
bfc1af65 NP |
952 | |
953 | if (ret) { | |
af5bc92d | 954 | unlock_page(page); |
af5bc92d | 955 | page_cache_release(page); |
ae4d5372 | 956 | /* |
6e1db88d | 957 | * __block_write_begin may have instantiated a few blocks |
ae4d5372 AK |
958 | * outside i_size. Trim these off again. Don't need |
959 | * i_size_read because we hold i_mutex. | |
1938a150 AK |
960 | * |
961 | * Add inode to orphan list in case we crash before | |
962 | * truncate finishes | |
ae4d5372 | 963 | */ |
ffacfa7a | 964 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
1938a150 AK |
965 | ext4_orphan_add(handle, inode); |
966 | ||
967 | ext4_journal_stop(handle); | |
968 | if (pos + len > inode->i_size) { | |
b9a4207d | 969 | ext4_truncate_failed_write(inode); |
de9a55b8 | 970 | /* |
ffacfa7a | 971 | * If truncate failed early the inode might |
1938a150 AK |
972 | * still be on the orphan list; we need to |
973 | * make sure the inode is removed from the | |
974 | * orphan list in that case. | |
975 | */ | |
976 | if (inode->i_nlink) | |
977 | ext4_orphan_del(NULL, inode); | |
978 | } | |
bfc1af65 NP |
979 | } |
980 | ||
617ba13b | 981 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) |
ac27a0ec | 982 | goto retry; |
7479d2b9 | 983 | out: |
ac27a0ec DK |
984 | return ret; |
985 | } | |
986 | ||
bfc1af65 NP |
987 | /* For write_end() in data=journal mode */ |
988 | static int write_end_fn(handle_t *handle, struct buffer_head *bh) | |
ac27a0ec DK |
989 | { |
990 | if (!buffer_mapped(bh) || buffer_freed(bh)) | |
991 | return 0; | |
992 | set_buffer_uptodate(bh); | |
0390131b | 993 | return ext4_handle_dirty_metadata(handle, NULL, bh); |
ac27a0ec DK |
994 | } |
995 | ||
f8514083 | 996 | static int ext4_generic_write_end(struct file *file, |
de9a55b8 TT |
997 | struct address_space *mapping, |
998 | loff_t pos, unsigned len, unsigned copied, | |
999 | struct page *page, void *fsdata) | |
f8514083 AK |
1000 | { |
1001 | int i_size_changed = 0; | |
1002 | struct inode *inode = mapping->host; | |
1003 | handle_t *handle = ext4_journal_current_handle(); | |
1004 | ||
1005 | copied = block_write_end(file, mapping, pos, len, copied, page, fsdata); | |
1006 | ||
1007 | /* | |
1008 | * No need to use i_size_read() here, the i_size | |
1009 | * cannot change under us because we hold i_mutex. | |
1010 | * | |
1011 | * But it's important to update i_size while still holding page lock: | |
1012 | * page writeout could otherwise come in and zero beyond i_size. | |
1013 | */ | |
1014 | if (pos + copied > inode->i_size) { | |
1015 | i_size_write(inode, pos + copied); | |
1016 | i_size_changed = 1; | |
1017 | } | |
1018 | ||
1019 | if (pos + copied > EXT4_I(inode)->i_disksize) { | |
1020 | /* We need to mark inode dirty even if | |
1021 | * new_i_size is less that inode->i_size | |
1022 | * bu greater than i_disksize.(hint delalloc) | |
1023 | */ | |
1024 | ext4_update_i_disksize(inode, (pos + copied)); | |
1025 | i_size_changed = 1; | |
1026 | } | |
1027 | unlock_page(page); | |
1028 | page_cache_release(page); | |
1029 | ||
1030 | /* | |
1031 | * Don't mark the inode dirty under page lock. First, it unnecessarily | |
1032 | * makes the holding time of page lock longer. Second, it forces lock | |
1033 | * ordering of page lock and transaction start for journaling | |
1034 | * filesystems. | |
1035 | */ | |
1036 | if (i_size_changed) | |
1037 | ext4_mark_inode_dirty(handle, inode); | |
1038 | ||
1039 | return copied; | |
1040 | } | |
1041 | ||
ac27a0ec DK |
1042 | /* |
1043 | * We need to pick up the new inode size which generic_commit_write gave us | |
1044 | * `file' can be NULL - eg, when called from page_symlink(). | |
1045 | * | |
617ba13b | 1046 | * ext4 never places buffers on inode->i_mapping->private_list. metadata |
ac27a0ec DK |
1047 | * buffers are managed internally. |
1048 | */ | |
bfc1af65 | 1049 | static int ext4_ordered_write_end(struct file *file, |
de9a55b8 TT |
1050 | struct address_space *mapping, |
1051 | loff_t pos, unsigned len, unsigned copied, | |
1052 | struct page *page, void *fsdata) | |
ac27a0ec | 1053 | { |
617ba13b | 1054 | handle_t *handle = ext4_journal_current_handle(); |
cf108bca | 1055 | struct inode *inode = mapping->host; |
ac27a0ec DK |
1056 | int ret = 0, ret2; |
1057 | ||
9bffad1e | 1058 | trace_ext4_ordered_write_end(inode, pos, len, copied); |
678aaf48 | 1059 | ret = ext4_jbd2_file_inode(handle, inode); |
ac27a0ec DK |
1060 | |
1061 | if (ret == 0) { | |
f8514083 | 1062 | ret2 = ext4_generic_write_end(file, mapping, pos, len, copied, |
bfc1af65 | 1063 | page, fsdata); |
f8a87d89 | 1064 | copied = ret2; |
ffacfa7a | 1065 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
f8514083 AK |
1066 | /* if we have allocated more blocks and copied |
1067 | * less. We will have blocks allocated outside | |
1068 | * inode->i_size. So truncate them | |
1069 | */ | |
1070 | ext4_orphan_add(handle, inode); | |
f8a87d89 RK |
1071 | if (ret2 < 0) |
1072 | ret = ret2; | |
09e0834f AF |
1073 | } else { |
1074 | unlock_page(page); | |
1075 | page_cache_release(page); | |
ac27a0ec | 1076 | } |
09e0834f | 1077 | |
617ba13b | 1078 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1079 | if (!ret) |
1080 | ret = ret2; | |
bfc1af65 | 1081 | |
f8514083 | 1082 | if (pos + len > inode->i_size) { |
b9a4207d | 1083 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1084 | /* |
ffacfa7a | 1085 | * If truncate failed early the inode might still be |
f8514083 AK |
1086 | * on the orphan list; we need to make sure the inode |
1087 | * is removed from the orphan list in that case. | |
1088 | */ | |
1089 | if (inode->i_nlink) | |
1090 | ext4_orphan_del(NULL, inode); | |
1091 | } | |
1092 | ||
1093 | ||
bfc1af65 | 1094 | return ret ? ret : copied; |
ac27a0ec DK |
1095 | } |
1096 | ||
bfc1af65 | 1097 | static int ext4_writeback_write_end(struct file *file, |
de9a55b8 TT |
1098 | struct address_space *mapping, |
1099 | loff_t pos, unsigned len, unsigned copied, | |
1100 | struct page *page, void *fsdata) | |
ac27a0ec | 1101 | { |
617ba13b | 1102 | handle_t *handle = ext4_journal_current_handle(); |
cf108bca | 1103 | struct inode *inode = mapping->host; |
ac27a0ec | 1104 | int ret = 0, ret2; |
ac27a0ec | 1105 | |
9bffad1e | 1106 | trace_ext4_writeback_write_end(inode, pos, len, copied); |
f8514083 | 1107 | ret2 = ext4_generic_write_end(file, mapping, pos, len, copied, |
bfc1af65 | 1108 | page, fsdata); |
f8a87d89 | 1109 | copied = ret2; |
ffacfa7a | 1110 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
f8514083 AK |
1111 | /* if we have allocated more blocks and copied |
1112 | * less. We will have blocks allocated outside | |
1113 | * inode->i_size. So truncate them | |
1114 | */ | |
1115 | ext4_orphan_add(handle, inode); | |
1116 | ||
f8a87d89 RK |
1117 | if (ret2 < 0) |
1118 | ret = ret2; | |
ac27a0ec | 1119 | |
617ba13b | 1120 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1121 | if (!ret) |
1122 | ret = ret2; | |
bfc1af65 | 1123 | |
f8514083 | 1124 | if (pos + len > inode->i_size) { |
b9a4207d | 1125 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1126 | /* |
ffacfa7a | 1127 | * If truncate failed early the inode might still be |
f8514083 AK |
1128 | * on the orphan list; we need to make sure the inode |
1129 | * is removed from the orphan list in that case. | |
1130 | */ | |
1131 | if (inode->i_nlink) | |
1132 | ext4_orphan_del(NULL, inode); | |
1133 | } | |
1134 | ||
bfc1af65 | 1135 | return ret ? ret : copied; |
ac27a0ec DK |
1136 | } |
1137 | ||
bfc1af65 | 1138 | static int ext4_journalled_write_end(struct file *file, |
de9a55b8 TT |
1139 | struct address_space *mapping, |
1140 | loff_t pos, unsigned len, unsigned copied, | |
1141 | struct page *page, void *fsdata) | |
ac27a0ec | 1142 | { |
617ba13b | 1143 | handle_t *handle = ext4_journal_current_handle(); |
bfc1af65 | 1144 | struct inode *inode = mapping->host; |
ac27a0ec DK |
1145 | int ret = 0, ret2; |
1146 | int partial = 0; | |
bfc1af65 | 1147 | unsigned from, to; |
cf17fea6 | 1148 | loff_t new_i_size; |
ac27a0ec | 1149 | |
9bffad1e | 1150 | trace_ext4_journalled_write_end(inode, pos, len, copied); |
bfc1af65 NP |
1151 | from = pos & (PAGE_CACHE_SIZE - 1); |
1152 | to = from + len; | |
1153 | ||
441c8508 CW |
1154 | BUG_ON(!ext4_handle_valid(handle)); |
1155 | ||
bfc1af65 NP |
1156 | if (copied < len) { |
1157 | if (!PageUptodate(page)) | |
1158 | copied = 0; | |
1159 | page_zero_new_buffers(page, from+copied, to); | |
1160 | } | |
ac27a0ec DK |
1161 | |
1162 | ret = walk_page_buffers(handle, page_buffers(page), from, | |
bfc1af65 | 1163 | to, &partial, write_end_fn); |
ac27a0ec DK |
1164 | if (!partial) |
1165 | SetPageUptodate(page); | |
cf17fea6 AK |
1166 | new_i_size = pos + copied; |
1167 | if (new_i_size > inode->i_size) | |
bfc1af65 | 1168 | i_size_write(inode, pos+copied); |
19f5fb7a | 1169 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
2d859db3 | 1170 | EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; |
cf17fea6 AK |
1171 | if (new_i_size > EXT4_I(inode)->i_disksize) { |
1172 | ext4_update_i_disksize(inode, new_i_size); | |
617ba13b | 1173 | ret2 = ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
1174 | if (!ret) |
1175 | ret = ret2; | |
1176 | } | |
bfc1af65 | 1177 | |
cf108bca | 1178 | unlock_page(page); |
f8514083 | 1179 | page_cache_release(page); |
ffacfa7a | 1180 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
f8514083 AK |
1181 | /* if we have allocated more blocks and copied |
1182 | * less. We will have blocks allocated outside | |
1183 | * inode->i_size. So truncate them | |
1184 | */ | |
1185 | ext4_orphan_add(handle, inode); | |
1186 | ||
617ba13b | 1187 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1188 | if (!ret) |
1189 | ret = ret2; | |
f8514083 | 1190 | if (pos + len > inode->i_size) { |
b9a4207d | 1191 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1192 | /* |
ffacfa7a | 1193 | * If truncate failed early the inode might still be |
f8514083 AK |
1194 | * on the orphan list; we need to make sure the inode |
1195 | * is removed from the orphan list in that case. | |
1196 | */ | |
1197 | if (inode->i_nlink) | |
1198 | ext4_orphan_del(NULL, inode); | |
1199 | } | |
bfc1af65 NP |
1200 | |
1201 | return ret ? ret : copied; | |
ac27a0ec | 1202 | } |
d2a17637 | 1203 | |
9d0be502 | 1204 | /* |
7b415bf6 | 1205 | * Reserve a single cluster located at lblock |
9d0be502 | 1206 | */ |
01f49d0b | 1207 | static int ext4_da_reserve_space(struct inode *inode, ext4_lblk_t lblock) |
d2a17637 | 1208 | { |
030ba6bc | 1209 | int retries = 0; |
60e58e0f | 1210 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
0637c6f4 | 1211 | struct ext4_inode_info *ei = EXT4_I(inode); |
7b415bf6 | 1212 | unsigned int md_needed; |
5dd4056d | 1213 | int ret; |
03179fe9 TT |
1214 | ext4_lblk_t save_last_lblock; |
1215 | int save_len; | |
1216 | ||
1217 | /* | |
1218 | * We will charge metadata quota at writeout time; this saves | |
1219 | * us from metadata over-estimation, though we may go over by | |
1220 | * a small amount in the end. Here we just reserve for data. | |
1221 | */ | |
1222 | ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1)); | |
1223 | if (ret) | |
1224 | return ret; | |
d2a17637 MC |
1225 | |
1226 | /* | |
1227 | * recalculate the amount of metadata blocks to reserve | |
1228 | * in order to allocate nrblocks | |
1229 | * worse case is one extent per block | |
1230 | */ | |
030ba6bc | 1231 | repeat: |
0637c6f4 | 1232 | spin_lock(&ei->i_block_reservation_lock); |
03179fe9 TT |
1233 | /* |
1234 | * ext4_calc_metadata_amount() has side effects, which we have | |
1235 | * to be prepared undo if we fail to claim space. | |
1236 | */ | |
1237 | save_len = ei->i_da_metadata_calc_len; | |
1238 | save_last_lblock = ei->i_da_metadata_calc_last_lblock; | |
7b415bf6 AK |
1239 | md_needed = EXT4_NUM_B2C(sbi, |
1240 | ext4_calc_metadata_amount(inode, lblock)); | |
f8ec9d68 | 1241 | trace_ext4_da_reserve_space(inode, md_needed); |
d2a17637 | 1242 | |
72b8ab9d ES |
1243 | /* |
1244 | * We do still charge estimated metadata to the sb though; | |
1245 | * we cannot afford to run out of free blocks. | |
1246 | */ | |
e7d5f315 | 1247 | if (ext4_claim_free_clusters(sbi, md_needed + 1, 0)) { |
03179fe9 TT |
1248 | ei->i_da_metadata_calc_len = save_len; |
1249 | ei->i_da_metadata_calc_last_lblock = save_last_lblock; | |
1250 | spin_unlock(&ei->i_block_reservation_lock); | |
030ba6bc AK |
1251 | if (ext4_should_retry_alloc(inode->i_sb, &retries)) { |
1252 | yield(); | |
1253 | goto repeat; | |
1254 | } | |
03179fe9 | 1255 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1)); |
d2a17637 MC |
1256 | return -ENOSPC; |
1257 | } | |
9d0be502 | 1258 | ei->i_reserved_data_blocks++; |
0637c6f4 TT |
1259 | ei->i_reserved_meta_blocks += md_needed; |
1260 | spin_unlock(&ei->i_block_reservation_lock); | |
39bc680a | 1261 | |
d2a17637 MC |
1262 | return 0; /* success */ |
1263 | } | |
1264 | ||
12219aea | 1265 | static void ext4_da_release_space(struct inode *inode, int to_free) |
d2a17637 MC |
1266 | { |
1267 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0637c6f4 | 1268 | struct ext4_inode_info *ei = EXT4_I(inode); |
d2a17637 | 1269 | |
cd213226 MC |
1270 | if (!to_free) |
1271 | return; /* Nothing to release, exit */ | |
1272 | ||
d2a17637 | 1273 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); |
cd213226 | 1274 | |
5a58ec87 | 1275 | trace_ext4_da_release_space(inode, to_free); |
0637c6f4 | 1276 | if (unlikely(to_free > ei->i_reserved_data_blocks)) { |
cd213226 | 1277 | /* |
0637c6f4 TT |
1278 | * if there aren't enough reserved blocks, then the |
1279 | * counter is messed up somewhere. Since this | |
1280 | * function is called from invalidate page, it's | |
1281 | * harmless to return without any action. | |
cd213226 | 1282 | */ |
0637c6f4 TT |
1283 | ext4_msg(inode->i_sb, KERN_NOTICE, "ext4_da_release_space: " |
1284 | "ino %lu, to_free %d with only %d reserved " | |
1084f252 | 1285 | "data blocks", inode->i_ino, to_free, |
0637c6f4 TT |
1286 | ei->i_reserved_data_blocks); |
1287 | WARN_ON(1); | |
1288 | to_free = ei->i_reserved_data_blocks; | |
cd213226 | 1289 | } |
0637c6f4 | 1290 | ei->i_reserved_data_blocks -= to_free; |
cd213226 | 1291 | |
0637c6f4 TT |
1292 | if (ei->i_reserved_data_blocks == 0) { |
1293 | /* | |
1294 | * We can release all of the reserved metadata blocks | |
1295 | * only when we have written all of the delayed | |
1296 | * allocation blocks. | |
7b415bf6 AK |
1297 | * Note that in case of bigalloc, i_reserved_meta_blocks, |
1298 | * i_reserved_data_blocks, etc. refer to number of clusters. | |
0637c6f4 | 1299 | */ |
57042651 | 1300 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, |
72b8ab9d | 1301 | ei->i_reserved_meta_blocks); |
ee5f4d9c | 1302 | ei->i_reserved_meta_blocks = 0; |
9d0be502 | 1303 | ei->i_da_metadata_calc_len = 0; |
0637c6f4 | 1304 | } |
d2a17637 | 1305 | |
72b8ab9d | 1306 | /* update fs dirty data blocks counter */ |
57042651 | 1307 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free); |
d2a17637 | 1308 | |
d2a17637 | 1309 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
60e58e0f | 1310 | |
7b415bf6 | 1311 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free)); |
d2a17637 MC |
1312 | } |
1313 | ||
1314 | static void ext4_da_page_release_reservation(struct page *page, | |
de9a55b8 | 1315 | unsigned long offset) |
d2a17637 MC |
1316 | { |
1317 | int to_release = 0; | |
1318 | struct buffer_head *head, *bh; | |
1319 | unsigned int curr_off = 0; | |
7b415bf6 AK |
1320 | struct inode *inode = page->mapping->host; |
1321 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
1322 | int num_clusters; | |
51865fda | 1323 | ext4_fsblk_t lblk; |
d2a17637 MC |
1324 | |
1325 | head = page_buffers(page); | |
1326 | bh = head; | |
1327 | do { | |
1328 | unsigned int next_off = curr_off + bh->b_size; | |
1329 | ||
1330 | if ((offset <= curr_off) && (buffer_delay(bh))) { | |
1331 | to_release++; | |
1332 | clear_buffer_delay(bh); | |
5356f261 | 1333 | clear_buffer_da_mapped(bh); |
d2a17637 MC |
1334 | } |
1335 | curr_off = next_off; | |
1336 | } while ((bh = bh->b_this_page) != head); | |
7b415bf6 | 1337 | |
51865fda ZL |
1338 | if (to_release) { |
1339 | lblk = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
1340 | ext4_es_remove_extent(inode, lblk, to_release); | |
1341 | } | |
1342 | ||
7b415bf6 AK |
1343 | /* If we have released all the blocks belonging to a cluster, then we |
1344 | * need to release the reserved space for that cluster. */ | |
1345 | num_clusters = EXT4_NUM_B2C(sbi, to_release); | |
1346 | while (num_clusters > 0) { | |
7b415bf6 AK |
1347 | lblk = (page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits)) + |
1348 | ((num_clusters - 1) << sbi->s_cluster_bits); | |
1349 | if (sbi->s_cluster_ratio == 1 || | |
1350 | !ext4_find_delalloc_cluster(inode, lblk, 1)) | |
1351 | ext4_da_release_space(inode, 1); | |
1352 | ||
1353 | num_clusters--; | |
1354 | } | |
d2a17637 | 1355 | } |
ac27a0ec | 1356 | |
64769240 AT |
1357 | /* |
1358 | * Delayed allocation stuff | |
1359 | */ | |
1360 | ||
64769240 AT |
1361 | /* |
1362 | * mpage_da_submit_io - walks through extent of pages and try to write | |
a1d6cc56 | 1363 | * them with writepage() call back |
64769240 AT |
1364 | * |
1365 | * @mpd->inode: inode | |
1366 | * @mpd->first_page: first page of the extent | |
1367 | * @mpd->next_page: page after the last page of the extent | |
64769240 AT |
1368 | * |
1369 | * By the time mpage_da_submit_io() is called we expect all blocks | |
1370 | * to be allocated. this may be wrong if allocation failed. | |
1371 | * | |
1372 | * As pages are already locked by write_cache_pages(), we can't use it | |
1373 | */ | |
1de3e3df TT |
1374 | static int mpage_da_submit_io(struct mpage_da_data *mpd, |
1375 | struct ext4_map_blocks *map) | |
64769240 | 1376 | { |
791b7f08 AK |
1377 | struct pagevec pvec; |
1378 | unsigned long index, end; | |
1379 | int ret = 0, err, nr_pages, i; | |
1380 | struct inode *inode = mpd->inode; | |
1381 | struct address_space *mapping = inode->i_mapping; | |
cb20d518 | 1382 | loff_t size = i_size_read(inode); |
3ecdb3a1 TT |
1383 | unsigned int len, block_start; |
1384 | struct buffer_head *bh, *page_bufs = NULL; | |
cb20d518 | 1385 | int journal_data = ext4_should_journal_data(inode); |
1de3e3df | 1386 | sector_t pblock = 0, cur_logical = 0; |
bd2d0210 | 1387 | struct ext4_io_submit io_submit; |
64769240 AT |
1388 | |
1389 | BUG_ON(mpd->next_page <= mpd->first_page); | |
bd2d0210 | 1390 | memset(&io_submit, 0, sizeof(io_submit)); |
791b7f08 AK |
1391 | /* |
1392 | * We need to start from the first_page to the next_page - 1 | |
1393 | * to make sure we also write the mapped dirty buffer_heads. | |
8dc207c0 | 1394 | * If we look at mpd->b_blocknr we would only be looking |
791b7f08 AK |
1395 | * at the currently mapped buffer_heads. |
1396 | */ | |
64769240 AT |
1397 | index = mpd->first_page; |
1398 | end = mpd->next_page - 1; | |
1399 | ||
791b7f08 | 1400 | pagevec_init(&pvec, 0); |
64769240 | 1401 | while (index <= end) { |
791b7f08 | 1402 | nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE); |
64769240 AT |
1403 | if (nr_pages == 0) |
1404 | break; | |
1405 | for (i = 0; i < nr_pages; i++) { | |
97498956 | 1406 | int commit_write = 0, skip_page = 0; |
64769240 AT |
1407 | struct page *page = pvec.pages[i]; |
1408 | ||
791b7f08 AK |
1409 | index = page->index; |
1410 | if (index > end) | |
1411 | break; | |
cb20d518 TT |
1412 | |
1413 | if (index == size >> PAGE_CACHE_SHIFT) | |
1414 | len = size & ~PAGE_CACHE_MASK; | |
1415 | else | |
1416 | len = PAGE_CACHE_SIZE; | |
1de3e3df TT |
1417 | if (map) { |
1418 | cur_logical = index << (PAGE_CACHE_SHIFT - | |
1419 | inode->i_blkbits); | |
1420 | pblock = map->m_pblk + (cur_logical - | |
1421 | map->m_lblk); | |
1422 | } | |
791b7f08 AK |
1423 | index++; |
1424 | ||
1425 | BUG_ON(!PageLocked(page)); | |
1426 | BUG_ON(PageWriteback(page)); | |
1427 | ||
64769240 | 1428 | /* |
cb20d518 TT |
1429 | * If the page does not have buffers (for |
1430 | * whatever reason), try to create them using | |
a107e5a3 | 1431 | * __block_write_begin. If this fails, |
97498956 | 1432 | * skip the page and move on. |
64769240 | 1433 | */ |
cb20d518 | 1434 | if (!page_has_buffers(page)) { |
a107e5a3 | 1435 | if (__block_write_begin(page, 0, len, |
cb20d518 | 1436 | noalloc_get_block_write)) { |
97498956 | 1437 | skip_page: |
cb20d518 TT |
1438 | unlock_page(page); |
1439 | continue; | |
1440 | } | |
1441 | commit_write = 1; | |
1442 | } | |
64769240 | 1443 | |
3ecdb3a1 TT |
1444 | bh = page_bufs = page_buffers(page); |
1445 | block_start = 0; | |
64769240 | 1446 | do { |
1de3e3df | 1447 | if (!bh) |
97498956 | 1448 | goto skip_page; |
1de3e3df TT |
1449 | if (map && (cur_logical >= map->m_lblk) && |
1450 | (cur_logical <= (map->m_lblk + | |
1451 | (map->m_len - 1)))) { | |
29fa89d0 AK |
1452 | if (buffer_delay(bh)) { |
1453 | clear_buffer_delay(bh); | |
1454 | bh->b_blocknr = pblock; | |
29fa89d0 | 1455 | } |
5356f261 AK |
1456 | if (buffer_da_mapped(bh)) |
1457 | clear_buffer_da_mapped(bh); | |
1de3e3df TT |
1458 | if (buffer_unwritten(bh) || |
1459 | buffer_mapped(bh)) | |
1460 | BUG_ON(bh->b_blocknr != pblock); | |
1461 | if (map->m_flags & EXT4_MAP_UNINIT) | |
1462 | set_buffer_uninit(bh); | |
1463 | clear_buffer_unwritten(bh); | |
1464 | } | |
29fa89d0 | 1465 | |
13a79a47 YY |
1466 | /* |
1467 | * skip page if block allocation undone and | |
1468 | * block is dirty | |
1469 | */ | |
1470 | if (ext4_bh_delay_or_unwritten(NULL, bh)) | |
97498956 | 1471 | skip_page = 1; |
3ecdb3a1 TT |
1472 | bh = bh->b_this_page; |
1473 | block_start += bh->b_size; | |
64769240 AT |
1474 | cur_logical++; |
1475 | pblock++; | |
1de3e3df TT |
1476 | } while (bh != page_bufs); |
1477 | ||
97498956 TT |
1478 | if (skip_page) |
1479 | goto skip_page; | |
cb20d518 TT |
1480 | |
1481 | if (commit_write) | |
1482 | /* mark the buffer_heads as dirty & uptodate */ | |
1483 | block_commit_write(page, 0, len); | |
1484 | ||
97498956 | 1485 | clear_page_dirty_for_io(page); |
bd2d0210 TT |
1486 | /* |
1487 | * Delalloc doesn't support data journalling, | |
1488 | * but eventually maybe we'll lift this | |
1489 | * restriction. | |
1490 | */ | |
1491 | if (unlikely(journal_data && PageChecked(page))) | |
cb20d518 | 1492 | err = __ext4_journalled_writepage(page, len); |
1449032b | 1493 | else if (test_opt(inode->i_sb, MBLK_IO_SUBMIT)) |
bd2d0210 TT |
1494 | err = ext4_bio_write_page(&io_submit, page, |
1495 | len, mpd->wbc); | |
9dd75f1f TT |
1496 | else if (buffer_uninit(page_bufs)) { |
1497 | ext4_set_bh_endio(page_bufs, inode); | |
1498 | err = block_write_full_page_endio(page, | |
1499 | noalloc_get_block_write, | |
1500 | mpd->wbc, ext4_end_io_buffer_write); | |
1501 | } else | |
1449032b TT |
1502 | err = block_write_full_page(page, |
1503 | noalloc_get_block_write, mpd->wbc); | |
cb20d518 TT |
1504 | |
1505 | if (!err) | |
a1d6cc56 | 1506 | mpd->pages_written++; |
64769240 AT |
1507 | /* |
1508 | * In error case, we have to continue because | |
1509 | * remaining pages are still locked | |
64769240 AT |
1510 | */ |
1511 | if (ret == 0) | |
1512 | ret = err; | |
64769240 AT |
1513 | } |
1514 | pagevec_release(&pvec); | |
1515 | } | |
bd2d0210 | 1516 | ext4_io_submit(&io_submit); |
64769240 | 1517 | return ret; |
64769240 AT |
1518 | } |
1519 | ||
c7f5938a | 1520 | static void ext4_da_block_invalidatepages(struct mpage_da_data *mpd) |
c4a0c46e AK |
1521 | { |
1522 | int nr_pages, i; | |
1523 | pgoff_t index, end; | |
1524 | struct pagevec pvec; | |
1525 | struct inode *inode = mpd->inode; | |
1526 | struct address_space *mapping = inode->i_mapping; | |
51865fda | 1527 | ext4_lblk_t start, last; |
c4a0c46e | 1528 | |
c7f5938a CW |
1529 | index = mpd->first_page; |
1530 | end = mpd->next_page - 1; | |
51865fda ZL |
1531 | |
1532 | start = index << (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
1533 | last = end << (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
1534 | ext4_es_remove_extent(inode, start, last - start + 1); | |
1535 | ||
c4a0c46e AK |
1536 | while (index <= end) { |
1537 | nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE); | |
1538 | if (nr_pages == 0) | |
1539 | break; | |
1540 | for (i = 0; i < nr_pages; i++) { | |
1541 | struct page *page = pvec.pages[i]; | |
9b1d0998 | 1542 | if (page->index > end) |
c4a0c46e | 1543 | break; |
c4a0c46e AK |
1544 | BUG_ON(!PageLocked(page)); |
1545 | BUG_ON(PageWriteback(page)); | |
1546 | block_invalidatepage(page, 0); | |
1547 | ClearPageUptodate(page); | |
1548 | unlock_page(page); | |
1549 | } | |
9b1d0998 JK |
1550 | index = pvec.pages[nr_pages - 1]->index + 1; |
1551 | pagevec_release(&pvec); | |
c4a0c46e AK |
1552 | } |
1553 | return; | |
1554 | } | |
1555 | ||
df22291f AK |
1556 | static void ext4_print_free_blocks(struct inode *inode) |
1557 | { | |
1558 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
92b97816 TT |
1559 | struct super_block *sb = inode->i_sb; |
1560 | ||
1561 | ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld", | |
5dee5437 TT |
1562 | EXT4_C2B(EXT4_SB(inode->i_sb), |
1563 | ext4_count_free_clusters(inode->i_sb))); | |
92b97816 TT |
1564 | ext4_msg(sb, KERN_CRIT, "Free/Dirty block details"); |
1565 | ext4_msg(sb, KERN_CRIT, "free_blocks=%lld", | |
57042651 TT |
1566 | (long long) EXT4_C2B(EXT4_SB(inode->i_sb), |
1567 | percpu_counter_sum(&sbi->s_freeclusters_counter))); | |
92b97816 | 1568 | ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld", |
7b415bf6 AK |
1569 | (long long) EXT4_C2B(EXT4_SB(inode->i_sb), |
1570 | percpu_counter_sum(&sbi->s_dirtyclusters_counter))); | |
92b97816 TT |
1571 | ext4_msg(sb, KERN_CRIT, "Block reservation details"); |
1572 | ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u", | |
1573 | EXT4_I(inode)->i_reserved_data_blocks); | |
1574 | ext4_msg(sb, KERN_CRIT, "i_reserved_meta_blocks=%u", | |
1693918e | 1575 | EXT4_I(inode)->i_reserved_meta_blocks); |
df22291f AK |
1576 | return; |
1577 | } | |
1578 | ||
64769240 | 1579 | /* |
5a87b7a5 TT |
1580 | * mpage_da_map_and_submit - go through given space, map them |
1581 | * if necessary, and then submit them for I/O | |
64769240 | 1582 | * |
8dc207c0 | 1583 | * @mpd - bh describing space |
64769240 AT |
1584 | * |
1585 | * The function skips space we know is already mapped to disk blocks. | |
1586 | * | |
64769240 | 1587 | */ |
5a87b7a5 | 1588 | static void mpage_da_map_and_submit(struct mpage_da_data *mpd) |
64769240 | 1589 | { |
2ac3b6e0 | 1590 | int err, blks, get_blocks_flags; |
1de3e3df | 1591 | struct ext4_map_blocks map, *mapp = NULL; |
2fa3cdfb TT |
1592 | sector_t next = mpd->b_blocknr; |
1593 | unsigned max_blocks = mpd->b_size >> mpd->inode->i_blkbits; | |
1594 | loff_t disksize = EXT4_I(mpd->inode)->i_disksize; | |
1595 | handle_t *handle = NULL; | |
64769240 AT |
1596 | |
1597 | /* | |
5a87b7a5 TT |
1598 | * If the blocks are mapped already, or we couldn't accumulate |
1599 | * any blocks, then proceed immediately to the submission stage. | |
2fa3cdfb | 1600 | */ |
5a87b7a5 TT |
1601 | if ((mpd->b_size == 0) || |
1602 | ((mpd->b_state & (1 << BH_Mapped)) && | |
1603 | !(mpd->b_state & (1 << BH_Delay)) && | |
1604 | !(mpd->b_state & (1 << BH_Unwritten)))) | |
1605 | goto submit_io; | |
2fa3cdfb TT |
1606 | |
1607 | handle = ext4_journal_current_handle(); | |
1608 | BUG_ON(!handle); | |
1609 | ||
79ffab34 | 1610 | /* |
79e83036 | 1611 | * Call ext4_map_blocks() to allocate any delayed allocation |
2ac3b6e0 TT |
1612 | * blocks, or to convert an uninitialized extent to be |
1613 | * initialized (in the case where we have written into | |
1614 | * one or more preallocated blocks). | |
1615 | * | |
1616 | * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to | |
1617 | * indicate that we are on the delayed allocation path. This | |
1618 | * affects functions in many different parts of the allocation | |
1619 | * call path. This flag exists primarily because we don't | |
79e83036 | 1620 | * want to change *many* call functions, so ext4_map_blocks() |
f2321097 | 1621 | * will set the EXT4_STATE_DELALLOC_RESERVED flag once the |
2ac3b6e0 TT |
1622 | * inode's allocation semaphore is taken. |
1623 | * | |
1624 | * If the blocks in questions were delalloc blocks, set | |
1625 | * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting | |
1626 | * variables are updated after the blocks have been allocated. | |
79ffab34 | 1627 | */ |
2ed88685 TT |
1628 | map.m_lblk = next; |
1629 | map.m_len = max_blocks; | |
1296cc85 | 1630 | get_blocks_flags = EXT4_GET_BLOCKS_CREATE; |
744692dc JZ |
1631 | if (ext4_should_dioread_nolock(mpd->inode)) |
1632 | get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT; | |
2ac3b6e0 | 1633 | if (mpd->b_state & (1 << BH_Delay)) |
1296cc85 AK |
1634 | get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE; |
1635 | ||
2ed88685 | 1636 | blks = ext4_map_blocks(handle, mpd->inode, &map, get_blocks_flags); |
2fa3cdfb | 1637 | if (blks < 0) { |
e3570639 ES |
1638 | struct super_block *sb = mpd->inode->i_sb; |
1639 | ||
2fa3cdfb | 1640 | err = blks; |
ed5bde0b | 1641 | /* |
5a87b7a5 | 1642 | * If get block returns EAGAIN or ENOSPC and there |
97498956 TT |
1643 | * appears to be free blocks we will just let |
1644 | * mpage_da_submit_io() unlock all of the pages. | |
c4a0c46e AK |
1645 | */ |
1646 | if (err == -EAGAIN) | |
5a87b7a5 | 1647 | goto submit_io; |
df22291f | 1648 | |
5dee5437 | 1649 | if (err == -ENOSPC && ext4_count_free_clusters(sb)) { |
df22291f | 1650 | mpd->retval = err; |
5a87b7a5 | 1651 | goto submit_io; |
df22291f AK |
1652 | } |
1653 | ||
c4a0c46e | 1654 | /* |
ed5bde0b TT |
1655 | * get block failure will cause us to loop in |
1656 | * writepages, because a_ops->writepage won't be able | |
1657 | * to make progress. The page will be redirtied by | |
1658 | * writepage and writepages will again try to write | |
1659 | * the same. | |
c4a0c46e | 1660 | */ |
e3570639 ES |
1661 | if (!(EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED)) { |
1662 | ext4_msg(sb, KERN_CRIT, | |
1663 | "delayed block allocation failed for inode %lu " | |
1664 | "at logical offset %llu with max blocks %zd " | |
1665 | "with error %d", mpd->inode->i_ino, | |
1666 | (unsigned long long) next, | |
1667 | mpd->b_size >> mpd->inode->i_blkbits, err); | |
1668 | ext4_msg(sb, KERN_CRIT, | |
1669 | "This should not happen!! Data will be lost\n"); | |
1670 | if (err == -ENOSPC) | |
1671 | ext4_print_free_blocks(mpd->inode); | |
030ba6bc | 1672 | } |
2fa3cdfb | 1673 | /* invalidate all the pages */ |
c7f5938a | 1674 | ext4_da_block_invalidatepages(mpd); |
e0fd9b90 CW |
1675 | |
1676 | /* Mark this page range as having been completed */ | |
1677 | mpd->io_done = 1; | |
5a87b7a5 | 1678 | return; |
c4a0c46e | 1679 | } |
2fa3cdfb TT |
1680 | BUG_ON(blks == 0); |
1681 | ||
1de3e3df | 1682 | mapp = ↦ |
2ed88685 TT |
1683 | if (map.m_flags & EXT4_MAP_NEW) { |
1684 | struct block_device *bdev = mpd->inode->i_sb->s_bdev; | |
1685 | int i; | |
64769240 | 1686 | |
2ed88685 TT |
1687 | for (i = 0; i < map.m_len; i++) |
1688 | unmap_underlying_metadata(bdev, map.m_pblk + i); | |
64769240 | 1689 | |
decbd919 TT |
1690 | if (ext4_should_order_data(mpd->inode)) { |
1691 | err = ext4_jbd2_file_inode(handle, mpd->inode); | |
8de49e67 | 1692 | if (err) { |
decbd919 | 1693 | /* Only if the journal is aborted */ |
8de49e67 KM |
1694 | mpd->retval = err; |
1695 | goto submit_io; | |
1696 | } | |
decbd919 | 1697 | } |
2fa3cdfb TT |
1698 | } |
1699 | ||
1700 | /* | |
03f5d8bc | 1701 | * Update on-disk size along with block allocation. |
2fa3cdfb TT |
1702 | */ |
1703 | disksize = ((loff_t) next + blks) << mpd->inode->i_blkbits; | |
1704 | if (disksize > i_size_read(mpd->inode)) | |
1705 | disksize = i_size_read(mpd->inode); | |
1706 | if (disksize > EXT4_I(mpd->inode)->i_disksize) { | |
1707 | ext4_update_i_disksize(mpd->inode, disksize); | |
5a87b7a5 TT |
1708 | err = ext4_mark_inode_dirty(handle, mpd->inode); |
1709 | if (err) | |
1710 | ext4_error(mpd->inode->i_sb, | |
1711 | "Failed to mark inode %lu dirty", | |
1712 | mpd->inode->i_ino); | |
2fa3cdfb TT |
1713 | } |
1714 | ||
5a87b7a5 | 1715 | submit_io: |
1de3e3df | 1716 | mpage_da_submit_io(mpd, mapp); |
5a87b7a5 | 1717 | mpd->io_done = 1; |
64769240 AT |
1718 | } |
1719 | ||
bf068ee2 AK |
1720 | #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \ |
1721 | (1 << BH_Delay) | (1 << BH_Unwritten)) | |
64769240 AT |
1722 | |
1723 | /* | |
1724 | * mpage_add_bh_to_extent - try to add one more block to extent of blocks | |
1725 | * | |
1726 | * @mpd->lbh - extent of blocks | |
1727 | * @logical - logical number of the block in the file | |
1728 | * @bh - bh of the block (used to access block's state) | |
1729 | * | |
1730 | * the function is used to collect contig. blocks in same state | |
1731 | */ | |
1732 | static void mpage_add_bh_to_extent(struct mpage_da_data *mpd, | |
8dc207c0 TT |
1733 | sector_t logical, size_t b_size, |
1734 | unsigned long b_state) | |
64769240 | 1735 | { |
64769240 | 1736 | sector_t next; |
8dc207c0 | 1737 | int nrblocks = mpd->b_size >> mpd->inode->i_blkbits; |
64769240 | 1738 | |
c445e3e0 ES |
1739 | /* |
1740 | * XXX Don't go larger than mballoc is willing to allocate | |
1741 | * This is a stopgap solution. We eventually need to fold | |
1742 | * mpage_da_submit_io() into this function and then call | |
79e83036 | 1743 | * ext4_map_blocks() multiple times in a loop |
c445e3e0 ES |
1744 | */ |
1745 | if (nrblocks >= 8*1024*1024/mpd->inode->i_sb->s_blocksize) | |
1746 | goto flush_it; | |
1747 | ||
525f4ed8 | 1748 | /* check if thereserved journal credits might overflow */ |
12e9b892 | 1749 | if (!(ext4_test_inode_flag(mpd->inode, EXT4_INODE_EXTENTS))) { |
525f4ed8 MC |
1750 | if (nrblocks >= EXT4_MAX_TRANS_DATA) { |
1751 | /* | |
1752 | * With non-extent format we are limited by the journal | |
1753 | * credit available. Total credit needed to insert | |
1754 | * nrblocks contiguous blocks is dependent on the | |
1755 | * nrblocks. So limit nrblocks. | |
1756 | */ | |
1757 | goto flush_it; | |
1758 | } else if ((nrblocks + (b_size >> mpd->inode->i_blkbits)) > | |
1759 | EXT4_MAX_TRANS_DATA) { | |
1760 | /* | |
1761 | * Adding the new buffer_head would make it cross the | |
1762 | * allowed limit for which we have journal credit | |
1763 | * reserved. So limit the new bh->b_size | |
1764 | */ | |
1765 | b_size = (EXT4_MAX_TRANS_DATA - nrblocks) << | |
1766 | mpd->inode->i_blkbits; | |
1767 | /* we will do mpage_da_submit_io in the next loop */ | |
1768 | } | |
1769 | } | |
64769240 AT |
1770 | /* |
1771 | * First block in the extent | |
1772 | */ | |
8dc207c0 TT |
1773 | if (mpd->b_size == 0) { |
1774 | mpd->b_blocknr = logical; | |
1775 | mpd->b_size = b_size; | |
1776 | mpd->b_state = b_state & BH_FLAGS; | |
64769240 AT |
1777 | return; |
1778 | } | |
1779 | ||
8dc207c0 | 1780 | next = mpd->b_blocknr + nrblocks; |
64769240 AT |
1781 | /* |
1782 | * Can we merge the block to our big extent? | |
1783 | */ | |
8dc207c0 TT |
1784 | if (logical == next && (b_state & BH_FLAGS) == mpd->b_state) { |
1785 | mpd->b_size += b_size; | |
64769240 AT |
1786 | return; |
1787 | } | |
1788 | ||
525f4ed8 | 1789 | flush_it: |
64769240 AT |
1790 | /* |
1791 | * We couldn't merge the block to our extent, so we | |
1792 | * need to flush current extent and start new one | |
1793 | */ | |
5a87b7a5 | 1794 | mpage_da_map_and_submit(mpd); |
a1d6cc56 | 1795 | return; |
64769240 AT |
1796 | } |
1797 | ||
c364b22c | 1798 | static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh) |
29fa89d0 | 1799 | { |
c364b22c | 1800 | return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh); |
29fa89d0 AK |
1801 | } |
1802 | ||
5356f261 AK |
1803 | /* |
1804 | * This function is grabs code from the very beginning of | |
1805 | * ext4_map_blocks, but assumes that the caller is from delayed write | |
1806 | * time. This function looks up the requested blocks and sets the | |
1807 | * buffer delay bit under the protection of i_data_sem. | |
1808 | */ | |
1809 | static int ext4_da_map_blocks(struct inode *inode, sector_t iblock, | |
1810 | struct ext4_map_blocks *map, | |
1811 | struct buffer_head *bh) | |
1812 | { | |
1813 | int retval; | |
1814 | sector_t invalid_block = ~((sector_t) 0xffff); | |
1815 | ||
1816 | if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es)) | |
1817 | invalid_block = ~0; | |
1818 | ||
1819 | map->m_flags = 0; | |
1820 | ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u," | |
1821 | "logical block %lu\n", inode->i_ino, map->m_len, | |
1822 | (unsigned long) map->m_lblk); | |
1823 | /* | |
1824 | * Try to see if we can get the block without requesting a new | |
1825 | * file system block. | |
1826 | */ | |
1827 | down_read((&EXT4_I(inode)->i_data_sem)); | |
1828 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
1829 | retval = ext4_ext_map_blocks(NULL, inode, map, 0); | |
1830 | else | |
1831 | retval = ext4_ind_map_blocks(NULL, inode, map, 0); | |
1832 | ||
1833 | if (retval == 0) { | |
1834 | /* | |
1835 | * XXX: __block_prepare_write() unmaps passed block, | |
1836 | * is it OK? | |
1837 | */ | |
1838 | /* If the block was allocated from previously allocated cluster, | |
1839 | * then we dont need to reserve it again. */ | |
1840 | if (!(map->m_flags & EXT4_MAP_FROM_CLUSTER)) { | |
1841 | retval = ext4_da_reserve_space(inode, iblock); | |
1842 | if (retval) | |
1843 | /* not enough space to reserve */ | |
1844 | goto out_unlock; | |
1845 | } | |
1846 | ||
51865fda ZL |
1847 | retval = ext4_es_insert_extent(inode, map->m_lblk, map->m_len); |
1848 | if (retval) | |
1849 | goto out_unlock; | |
1850 | ||
5356f261 AK |
1851 | /* Clear EXT4_MAP_FROM_CLUSTER flag since its purpose is served |
1852 | * and it should not appear on the bh->b_state. | |
1853 | */ | |
1854 | map->m_flags &= ~EXT4_MAP_FROM_CLUSTER; | |
1855 | ||
1856 | map_bh(bh, inode->i_sb, invalid_block); | |
1857 | set_buffer_new(bh); | |
1858 | set_buffer_delay(bh); | |
1859 | } | |
1860 | ||
1861 | out_unlock: | |
1862 | up_read((&EXT4_I(inode)->i_data_sem)); | |
1863 | ||
1864 | return retval; | |
1865 | } | |
1866 | ||
64769240 | 1867 | /* |
b920c755 TT |
1868 | * This is a special get_blocks_t callback which is used by |
1869 | * ext4_da_write_begin(). It will either return mapped block or | |
1870 | * reserve space for a single block. | |
29fa89d0 AK |
1871 | * |
1872 | * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set. | |
1873 | * We also have b_blocknr = -1 and b_bdev initialized properly | |
1874 | * | |
1875 | * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set. | |
1876 | * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev | |
1877 | * initialized properly. | |
64769240 AT |
1878 | */ |
1879 | static int ext4_da_get_block_prep(struct inode *inode, sector_t iblock, | |
2ed88685 | 1880 | struct buffer_head *bh, int create) |
64769240 | 1881 | { |
2ed88685 | 1882 | struct ext4_map_blocks map; |
64769240 AT |
1883 | int ret = 0; |
1884 | ||
1885 | BUG_ON(create == 0); | |
2ed88685 TT |
1886 | BUG_ON(bh->b_size != inode->i_sb->s_blocksize); |
1887 | ||
1888 | map.m_lblk = iblock; | |
1889 | map.m_len = 1; | |
64769240 AT |
1890 | |
1891 | /* | |
1892 | * first, we need to know whether the block is allocated already | |
1893 | * preallocated blocks are unmapped but should treated | |
1894 | * the same as allocated blocks. | |
1895 | */ | |
5356f261 AK |
1896 | ret = ext4_da_map_blocks(inode, iblock, &map, bh); |
1897 | if (ret <= 0) | |
2ed88685 | 1898 | return ret; |
64769240 | 1899 | |
2ed88685 TT |
1900 | map_bh(bh, inode->i_sb, map.m_pblk); |
1901 | bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | map.m_flags; | |
1902 | ||
1903 | if (buffer_unwritten(bh)) { | |
1904 | /* A delayed write to unwritten bh should be marked | |
1905 | * new and mapped. Mapped ensures that we don't do | |
1906 | * get_block multiple times when we write to the same | |
1907 | * offset and new ensures that we do proper zero out | |
1908 | * for partial write. | |
1909 | */ | |
1910 | set_buffer_new(bh); | |
c8205636 | 1911 | set_buffer_mapped(bh); |
2ed88685 TT |
1912 | } |
1913 | return 0; | |
64769240 | 1914 | } |
61628a3f | 1915 | |
b920c755 TT |
1916 | /* |
1917 | * This function is used as a standard get_block_t calback function | |
1918 | * when there is no desire to allocate any blocks. It is used as a | |
ebdec241 | 1919 | * callback function for block_write_begin() and block_write_full_page(). |
206f7ab4 | 1920 | * These functions should only try to map a single block at a time. |
b920c755 TT |
1921 | * |
1922 | * Since this function doesn't do block allocations even if the caller | |
1923 | * requests it by passing in create=1, it is critically important that | |
1924 | * any caller checks to make sure that any buffer heads are returned | |
1925 | * by this function are either all already mapped or marked for | |
206f7ab4 CH |
1926 | * delayed allocation before calling block_write_full_page(). Otherwise, |
1927 | * b_blocknr could be left unitialized, and the page write functions will | |
1928 | * be taken by surprise. | |
b920c755 TT |
1929 | */ |
1930 | static int noalloc_get_block_write(struct inode *inode, sector_t iblock, | |
f0e6c985 AK |
1931 | struct buffer_head *bh_result, int create) |
1932 | { | |
a2dc52b5 | 1933 | BUG_ON(bh_result->b_size != inode->i_sb->s_blocksize); |
2ed88685 | 1934 | return _ext4_get_block(inode, iblock, bh_result, 0); |
61628a3f MC |
1935 | } |
1936 | ||
62e086be AK |
1937 | static int bget_one(handle_t *handle, struct buffer_head *bh) |
1938 | { | |
1939 | get_bh(bh); | |
1940 | return 0; | |
1941 | } | |
1942 | ||
1943 | static int bput_one(handle_t *handle, struct buffer_head *bh) | |
1944 | { | |
1945 | put_bh(bh); | |
1946 | return 0; | |
1947 | } | |
1948 | ||
1949 | static int __ext4_journalled_writepage(struct page *page, | |
62e086be AK |
1950 | unsigned int len) |
1951 | { | |
1952 | struct address_space *mapping = page->mapping; | |
1953 | struct inode *inode = mapping->host; | |
1954 | struct buffer_head *page_bufs; | |
1955 | handle_t *handle = NULL; | |
1956 | int ret = 0; | |
1957 | int err; | |
1958 | ||
cb20d518 | 1959 | ClearPageChecked(page); |
62e086be AK |
1960 | page_bufs = page_buffers(page); |
1961 | BUG_ON(!page_bufs); | |
1962 | walk_page_buffers(handle, page_bufs, 0, len, NULL, bget_one); | |
1963 | /* As soon as we unlock the page, it can go away, but we have | |
1964 | * references to buffers so we are safe */ | |
1965 | unlock_page(page); | |
1966 | ||
1967 | handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode)); | |
1968 | if (IS_ERR(handle)) { | |
1969 | ret = PTR_ERR(handle); | |
1970 | goto out; | |
1971 | } | |
1972 | ||
441c8508 CW |
1973 | BUG_ON(!ext4_handle_valid(handle)); |
1974 | ||
62e086be AK |
1975 | ret = walk_page_buffers(handle, page_bufs, 0, len, NULL, |
1976 | do_journal_get_write_access); | |
1977 | ||
1978 | err = walk_page_buffers(handle, page_bufs, 0, len, NULL, | |
1979 | write_end_fn); | |
1980 | if (ret == 0) | |
1981 | ret = err; | |
2d859db3 | 1982 | EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; |
62e086be AK |
1983 | err = ext4_journal_stop(handle); |
1984 | if (!ret) | |
1985 | ret = err; | |
1986 | ||
1987 | walk_page_buffers(handle, page_bufs, 0, len, NULL, bput_one); | |
19f5fb7a | 1988 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
62e086be AK |
1989 | out: |
1990 | return ret; | |
1991 | } | |
1992 | ||
61628a3f | 1993 | /* |
43ce1d23 AK |
1994 | * Note that we don't need to start a transaction unless we're journaling data |
1995 | * because we should have holes filled from ext4_page_mkwrite(). We even don't | |
1996 | * need to file the inode to the transaction's list in ordered mode because if | |
1997 | * we are writing back data added by write(), the inode is already there and if | |
25985edc | 1998 | * we are writing back data modified via mmap(), no one guarantees in which |
43ce1d23 AK |
1999 | * transaction the data will hit the disk. In case we are journaling data, we |
2000 | * cannot start transaction directly because transaction start ranks above page | |
2001 | * lock so we have to do some magic. | |
2002 | * | |
b920c755 TT |
2003 | * This function can get called via... |
2004 | * - ext4_da_writepages after taking page lock (have journal handle) | |
2005 | * - journal_submit_inode_data_buffers (no journal handle) | |
f6463b0d | 2006 | * - shrink_page_list via the kswapd/direct reclaim (no journal handle) |
b920c755 | 2007 | * - grab_page_cache when doing write_begin (have journal handle) |
43ce1d23 AK |
2008 | * |
2009 | * We don't do any block allocation in this function. If we have page with | |
2010 | * multiple blocks we need to write those buffer_heads that are mapped. This | |
2011 | * is important for mmaped based write. So if we do with blocksize 1K | |
2012 | * truncate(f, 1024); | |
2013 | * a = mmap(f, 0, 4096); | |
2014 | * a[0] = 'a'; | |
2015 | * truncate(f, 4096); | |
2016 | * we have in the page first buffer_head mapped via page_mkwrite call back | |
90802ed9 | 2017 | * but other buffer_heads would be unmapped but dirty (dirty done via the |
43ce1d23 AK |
2018 | * do_wp_page). So writepage should write the first block. If we modify |
2019 | * the mmap area beyond 1024 we will again get a page_fault and the | |
2020 | * page_mkwrite callback will do the block allocation and mark the | |
2021 | * buffer_heads mapped. | |
2022 | * | |
2023 | * We redirty the page if we have any buffer_heads that is either delay or | |
2024 | * unwritten in the page. | |
2025 | * | |
2026 | * We can get recursively called as show below. | |
2027 | * | |
2028 | * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() -> | |
2029 | * ext4_writepage() | |
2030 | * | |
2031 | * But since we don't do any block allocation we should not deadlock. | |
2032 | * Page also have the dirty flag cleared so we don't get recurive page_lock. | |
61628a3f | 2033 | */ |
43ce1d23 | 2034 | static int ext4_writepage(struct page *page, |
62e086be | 2035 | struct writeback_control *wbc) |
64769240 | 2036 | { |
a42afc5f | 2037 | int ret = 0, commit_write = 0; |
61628a3f | 2038 | loff_t size; |
498e5f24 | 2039 | unsigned int len; |
744692dc | 2040 | struct buffer_head *page_bufs = NULL; |
61628a3f MC |
2041 | struct inode *inode = page->mapping->host; |
2042 | ||
a9c667f8 | 2043 | trace_ext4_writepage(page); |
f0e6c985 AK |
2044 | size = i_size_read(inode); |
2045 | if (page->index == size >> PAGE_CACHE_SHIFT) | |
2046 | len = size & ~PAGE_CACHE_MASK; | |
2047 | else | |
2048 | len = PAGE_CACHE_SIZE; | |
64769240 | 2049 | |
a42afc5f TT |
2050 | /* |
2051 | * If the page does not have buffers (for whatever reason), | |
a107e5a3 | 2052 | * try to create them using __block_write_begin. If this |
a42afc5f TT |
2053 | * fails, redirty the page and move on. |
2054 | */ | |
b1142e8f | 2055 | if (!page_has_buffers(page)) { |
a107e5a3 | 2056 | if (__block_write_begin(page, 0, len, |
a42afc5f TT |
2057 | noalloc_get_block_write)) { |
2058 | redirty_page: | |
f0e6c985 AK |
2059 | redirty_page_for_writepage(wbc, page); |
2060 | unlock_page(page); | |
2061 | return 0; | |
2062 | } | |
a42afc5f TT |
2063 | commit_write = 1; |
2064 | } | |
2065 | page_bufs = page_buffers(page); | |
2066 | if (walk_page_buffers(NULL, page_bufs, 0, len, NULL, | |
2067 | ext4_bh_delay_or_unwritten)) { | |
f0e6c985 | 2068 | /* |
b1142e8f TT |
2069 | * We don't want to do block allocation, so redirty |
2070 | * the page and return. We may reach here when we do | |
2071 | * a journal commit via journal_submit_inode_data_buffers. | |
966dbde2 MG |
2072 | * We can also reach here via shrink_page_list but it |
2073 | * should never be for direct reclaim so warn if that | |
2074 | * happens | |
f0e6c985 | 2075 | */ |
966dbde2 MG |
2076 | WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) == |
2077 | PF_MEMALLOC); | |
a42afc5f TT |
2078 | goto redirty_page; |
2079 | } | |
2080 | if (commit_write) | |
ed9b3e33 | 2081 | /* now mark the buffer_heads as dirty and uptodate */ |
b767e78a | 2082 | block_commit_write(page, 0, len); |
64769240 | 2083 | |
cb20d518 | 2084 | if (PageChecked(page) && ext4_should_journal_data(inode)) |
43ce1d23 AK |
2085 | /* |
2086 | * It's mmapped pagecache. Add buffers and journal it. There | |
2087 | * doesn't seem much point in redirtying the page here. | |
2088 | */ | |
3f0ca309 | 2089 | return __ext4_journalled_writepage(page, len); |
43ce1d23 | 2090 | |
a42afc5f | 2091 | if (buffer_uninit(page_bufs)) { |
744692dc JZ |
2092 | ext4_set_bh_endio(page_bufs, inode); |
2093 | ret = block_write_full_page_endio(page, noalloc_get_block_write, | |
2094 | wbc, ext4_end_io_buffer_write); | |
2095 | } else | |
b920c755 TT |
2096 | ret = block_write_full_page(page, noalloc_get_block_write, |
2097 | wbc); | |
64769240 | 2098 | |
64769240 AT |
2099 | return ret; |
2100 | } | |
2101 | ||
61628a3f | 2102 | /* |
525f4ed8 | 2103 | * This is called via ext4_da_writepages() to |
25985edc | 2104 | * calculate the total number of credits to reserve to fit |
525f4ed8 MC |
2105 | * a single extent allocation into a single transaction, |
2106 | * ext4_da_writpeages() will loop calling this before | |
2107 | * the block allocation. | |
61628a3f | 2108 | */ |
525f4ed8 MC |
2109 | |
2110 | static int ext4_da_writepages_trans_blocks(struct inode *inode) | |
2111 | { | |
2112 | int max_blocks = EXT4_I(inode)->i_reserved_data_blocks; | |
2113 | ||
2114 | /* | |
2115 | * With non-extent format the journal credit needed to | |
2116 | * insert nrblocks contiguous block is dependent on | |
2117 | * number of contiguous block. So we will limit | |
2118 | * number of contiguous block to a sane value | |
2119 | */ | |
12e9b892 | 2120 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) && |
525f4ed8 MC |
2121 | (max_blocks > EXT4_MAX_TRANS_DATA)) |
2122 | max_blocks = EXT4_MAX_TRANS_DATA; | |
2123 | ||
2124 | return ext4_chunk_trans_blocks(inode, max_blocks); | |
2125 | } | |
61628a3f | 2126 | |
8e48dcfb TT |
2127 | /* |
2128 | * write_cache_pages_da - walk the list of dirty pages of the given | |
8eb9e5ce | 2129 | * address space and accumulate pages that need writing, and call |
168fc022 TT |
2130 | * mpage_da_map_and_submit to map a single contiguous memory region |
2131 | * and then write them. | |
8e48dcfb TT |
2132 | */ |
2133 | static int write_cache_pages_da(struct address_space *mapping, | |
2134 | struct writeback_control *wbc, | |
72f84e65 ES |
2135 | struct mpage_da_data *mpd, |
2136 | pgoff_t *done_index) | |
8e48dcfb | 2137 | { |
4f01b02c | 2138 | struct buffer_head *bh, *head; |
168fc022 | 2139 | struct inode *inode = mapping->host; |
4f01b02c TT |
2140 | struct pagevec pvec; |
2141 | unsigned int nr_pages; | |
2142 | sector_t logical; | |
2143 | pgoff_t index, end; | |
2144 | long nr_to_write = wbc->nr_to_write; | |
2145 | int i, tag, ret = 0; | |
8e48dcfb | 2146 | |
168fc022 TT |
2147 | memset(mpd, 0, sizeof(struct mpage_da_data)); |
2148 | mpd->wbc = wbc; | |
2149 | mpd->inode = inode; | |
8e48dcfb TT |
2150 | pagevec_init(&pvec, 0); |
2151 | index = wbc->range_start >> PAGE_CACHE_SHIFT; | |
2152 | end = wbc->range_end >> PAGE_CACHE_SHIFT; | |
2153 | ||
6e6938b6 | 2154 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
5b41d924 ES |
2155 | tag = PAGECACHE_TAG_TOWRITE; |
2156 | else | |
2157 | tag = PAGECACHE_TAG_DIRTY; | |
2158 | ||
72f84e65 | 2159 | *done_index = index; |
4f01b02c | 2160 | while (index <= end) { |
5b41d924 | 2161 | nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag, |
8e48dcfb TT |
2162 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1); |
2163 | if (nr_pages == 0) | |
4f01b02c | 2164 | return 0; |
8e48dcfb TT |
2165 | |
2166 | for (i = 0; i < nr_pages; i++) { | |
2167 | struct page *page = pvec.pages[i]; | |
2168 | ||
2169 | /* | |
2170 | * At this point, the page may be truncated or | |
2171 | * invalidated (changing page->mapping to NULL), or | |
2172 | * even swizzled back from swapper_space to tmpfs file | |
2173 | * mapping. However, page->index will not change | |
2174 | * because we have a reference on the page. | |
2175 | */ | |
4f01b02c TT |
2176 | if (page->index > end) |
2177 | goto out; | |
8e48dcfb | 2178 | |
72f84e65 ES |
2179 | *done_index = page->index + 1; |
2180 | ||
78aaced3 TT |
2181 | /* |
2182 | * If we can't merge this page, and we have | |
2183 | * accumulated an contiguous region, write it | |
2184 | */ | |
2185 | if ((mpd->next_page != page->index) && | |
2186 | (mpd->next_page != mpd->first_page)) { | |
2187 | mpage_da_map_and_submit(mpd); | |
2188 | goto ret_extent_tail; | |
2189 | } | |
2190 | ||
8e48dcfb TT |
2191 | lock_page(page); |
2192 | ||
2193 | /* | |
4f01b02c TT |
2194 | * If the page is no longer dirty, or its |
2195 | * mapping no longer corresponds to inode we | |
2196 | * are writing (which means it has been | |
2197 | * truncated or invalidated), or the page is | |
2198 | * already under writeback and we are not | |
2199 | * doing a data integrity writeback, skip the page | |
8e48dcfb | 2200 | */ |
4f01b02c TT |
2201 | if (!PageDirty(page) || |
2202 | (PageWriteback(page) && | |
2203 | (wbc->sync_mode == WB_SYNC_NONE)) || | |
2204 | unlikely(page->mapping != mapping)) { | |
8e48dcfb TT |
2205 | unlock_page(page); |
2206 | continue; | |
2207 | } | |
2208 | ||
7cb1a535 | 2209 | wait_on_page_writeback(page); |
8e48dcfb | 2210 | BUG_ON(PageWriteback(page)); |
8e48dcfb | 2211 | |
168fc022 | 2212 | if (mpd->next_page != page->index) |
8eb9e5ce | 2213 | mpd->first_page = page->index; |
8eb9e5ce TT |
2214 | mpd->next_page = page->index + 1; |
2215 | logical = (sector_t) page->index << | |
2216 | (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
2217 | ||
2218 | if (!page_has_buffers(page)) { | |
4f01b02c TT |
2219 | mpage_add_bh_to_extent(mpd, logical, |
2220 | PAGE_CACHE_SIZE, | |
8eb9e5ce | 2221 | (1 << BH_Dirty) | (1 << BH_Uptodate)); |
4f01b02c TT |
2222 | if (mpd->io_done) |
2223 | goto ret_extent_tail; | |
8eb9e5ce TT |
2224 | } else { |
2225 | /* | |
4f01b02c TT |
2226 | * Page with regular buffer heads, |
2227 | * just add all dirty ones | |
8eb9e5ce TT |
2228 | */ |
2229 | head = page_buffers(page); | |
2230 | bh = head; | |
2231 | do { | |
2232 | BUG_ON(buffer_locked(bh)); | |
2233 | /* | |
2234 | * We need to try to allocate | |
2235 | * unmapped blocks in the same page. | |
2236 | * Otherwise we won't make progress | |
2237 | * with the page in ext4_writepage | |
2238 | */ | |
2239 | if (ext4_bh_delay_or_unwritten(NULL, bh)) { | |
2240 | mpage_add_bh_to_extent(mpd, logical, | |
2241 | bh->b_size, | |
2242 | bh->b_state); | |
4f01b02c TT |
2243 | if (mpd->io_done) |
2244 | goto ret_extent_tail; | |
8eb9e5ce TT |
2245 | } else if (buffer_dirty(bh) && (buffer_mapped(bh))) { |
2246 | /* | |
4f01b02c TT |
2247 | * mapped dirty buffer. We need |
2248 | * to update the b_state | |
2249 | * because we look at b_state | |
2250 | * in mpage_da_map_blocks. We | |
2251 | * don't update b_size because | |
2252 | * if we find an unmapped | |
2253 | * buffer_head later we need to | |
2254 | * use the b_state flag of that | |
2255 | * buffer_head. | |
8eb9e5ce TT |
2256 | */ |
2257 | if (mpd->b_size == 0) | |
2258 | mpd->b_state = bh->b_state & BH_FLAGS; | |
2259 | } | |
2260 | logical++; | |
2261 | } while ((bh = bh->b_this_page) != head); | |
8e48dcfb TT |
2262 | } |
2263 | ||
2264 | if (nr_to_write > 0) { | |
2265 | nr_to_write--; | |
2266 | if (nr_to_write == 0 && | |
4f01b02c | 2267 | wbc->sync_mode == WB_SYNC_NONE) |
8e48dcfb TT |
2268 | /* |
2269 | * We stop writing back only if we are | |
2270 | * not doing integrity sync. In case of | |
2271 | * integrity sync we have to keep going | |
2272 | * because someone may be concurrently | |
2273 | * dirtying pages, and we might have | |
2274 | * synced a lot of newly appeared dirty | |
2275 | * pages, but have not synced all of the | |
2276 | * old dirty pages. | |
2277 | */ | |
4f01b02c | 2278 | goto out; |
8e48dcfb TT |
2279 | } |
2280 | } | |
2281 | pagevec_release(&pvec); | |
2282 | cond_resched(); | |
2283 | } | |
4f01b02c TT |
2284 | return 0; |
2285 | ret_extent_tail: | |
2286 | ret = MPAGE_DA_EXTENT_TAIL; | |
8eb9e5ce TT |
2287 | out: |
2288 | pagevec_release(&pvec); | |
2289 | cond_resched(); | |
8e48dcfb TT |
2290 | return ret; |
2291 | } | |
2292 | ||
2293 | ||
64769240 | 2294 | static int ext4_da_writepages(struct address_space *mapping, |
a1d6cc56 | 2295 | struct writeback_control *wbc) |
64769240 | 2296 | { |
22208ded AK |
2297 | pgoff_t index; |
2298 | int range_whole = 0; | |
61628a3f | 2299 | handle_t *handle = NULL; |
df22291f | 2300 | struct mpage_da_data mpd; |
5e745b04 | 2301 | struct inode *inode = mapping->host; |
498e5f24 | 2302 | int pages_written = 0; |
55138e0b | 2303 | unsigned int max_pages; |
2acf2c26 | 2304 | int range_cyclic, cycled = 1, io_done = 0; |
55138e0b TT |
2305 | int needed_blocks, ret = 0; |
2306 | long desired_nr_to_write, nr_to_writebump = 0; | |
de89de6e | 2307 | loff_t range_start = wbc->range_start; |
5e745b04 | 2308 | struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb); |
72f84e65 | 2309 | pgoff_t done_index = 0; |
5b41d924 | 2310 | pgoff_t end; |
1bce63d1 | 2311 | struct blk_plug plug; |
61628a3f | 2312 | |
9bffad1e | 2313 | trace_ext4_da_writepages(inode, wbc); |
ba80b101 | 2314 | |
61628a3f MC |
2315 | /* |
2316 | * No pages to write? This is mainly a kludge to avoid starting | |
2317 | * a transaction for special inodes like journal inode on last iput() | |
2318 | * because that could violate lock ordering on umount | |
2319 | */ | |
a1d6cc56 | 2320 | if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) |
61628a3f | 2321 | return 0; |
2a21e37e TT |
2322 | |
2323 | /* | |
2324 | * If the filesystem has aborted, it is read-only, so return | |
2325 | * right away instead of dumping stack traces later on that | |
2326 | * will obscure the real source of the problem. We test | |
4ab2f15b | 2327 | * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because |
2a21e37e TT |
2328 | * the latter could be true if the filesystem is mounted |
2329 | * read-only, and in that case, ext4_da_writepages should | |
2330 | * *never* be called, so if that ever happens, we would want | |
2331 | * the stack trace. | |
2332 | */ | |
4ab2f15b | 2333 | if (unlikely(sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) |
2a21e37e TT |
2334 | return -EROFS; |
2335 | ||
22208ded AK |
2336 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
2337 | range_whole = 1; | |
61628a3f | 2338 | |
2acf2c26 AK |
2339 | range_cyclic = wbc->range_cyclic; |
2340 | if (wbc->range_cyclic) { | |
22208ded | 2341 | index = mapping->writeback_index; |
2acf2c26 AK |
2342 | if (index) |
2343 | cycled = 0; | |
2344 | wbc->range_start = index << PAGE_CACHE_SHIFT; | |
2345 | wbc->range_end = LLONG_MAX; | |
2346 | wbc->range_cyclic = 0; | |
5b41d924 ES |
2347 | end = -1; |
2348 | } else { | |
22208ded | 2349 | index = wbc->range_start >> PAGE_CACHE_SHIFT; |
5b41d924 ES |
2350 | end = wbc->range_end >> PAGE_CACHE_SHIFT; |
2351 | } | |
a1d6cc56 | 2352 | |
55138e0b TT |
2353 | /* |
2354 | * This works around two forms of stupidity. The first is in | |
2355 | * the writeback code, which caps the maximum number of pages | |
2356 | * written to be 1024 pages. This is wrong on multiple | |
2357 | * levels; different architectues have a different page size, | |
2358 | * which changes the maximum amount of data which gets | |
2359 | * written. Secondly, 4 megabytes is way too small. XFS | |
2360 | * forces this value to be 16 megabytes by multiplying | |
2361 | * nr_to_write parameter by four, and then relies on its | |
2362 | * allocator to allocate larger extents to make them | |
2363 | * contiguous. Unfortunately this brings us to the second | |
2364 | * stupidity, which is that ext4's mballoc code only allocates | |
2365 | * at most 2048 blocks. So we force contiguous writes up to | |
2366 | * the number of dirty blocks in the inode, or | |
2367 | * sbi->max_writeback_mb_bump whichever is smaller. | |
2368 | */ | |
2369 | max_pages = sbi->s_max_writeback_mb_bump << (20 - PAGE_CACHE_SHIFT); | |
b443e733 ES |
2370 | if (!range_cyclic && range_whole) { |
2371 | if (wbc->nr_to_write == LONG_MAX) | |
2372 | desired_nr_to_write = wbc->nr_to_write; | |
2373 | else | |
2374 | desired_nr_to_write = wbc->nr_to_write * 8; | |
2375 | } else | |
55138e0b TT |
2376 | desired_nr_to_write = ext4_num_dirty_pages(inode, index, |
2377 | max_pages); | |
2378 | if (desired_nr_to_write > max_pages) | |
2379 | desired_nr_to_write = max_pages; | |
2380 | ||
2381 | if (wbc->nr_to_write < desired_nr_to_write) { | |
2382 | nr_to_writebump = desired_nr_to_write - wbc->nr_to_write; | |
2383 | wbc->nr_to_write = desired_nr_to_write; | |
2384 | } | |
2385 | ||
2acf2c26 | 2386 | retry: |
6e6938b6 | 2387 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
5b41d924 ES |
2388 | tag_pages_for_writeback(mapping, index, end); |
2389 | ||
1bce63d1 | 2390 | blk_start_plug(&plug); |
22208ded | 2391 | while (!ret && wbc->nr_to_write > 0) { |
a1d6cc56 AK |
2392 | |
2393 | /* | |
2394 | * we insert one extent at a time. So we need | |
2395 | * credit needed for single extent allocation. | |
2396 | * journalled mode is currently not supported | |
2397 | * by delalloc | |
2398 | */ | |
2399 | BUG_ON(ext4_should_journal_data(inode)); | |
525f4ed8 | 2400 | needed_blocks = ext4_da_writepages_trans_blocks(inode); |
a1d6cc56 | 2401 | |
61628a3f MC |
2402 | /* start a new transaction*/ |
2403 | handle = ext4_journal_start(inode, needed_blocks); | |
2404 | if (IS_ERR(handle)) { | |
2405 | ret = PTR_ERR(handle); | |
1693918e | 2406 | ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: " |
fbe845dd | 2407 | "%ld pages, ino %lu; err %d", __func__, |
a1d6cc56 | 2408 | wbc->nr_to_write, inode->i_ino, ret); |
3c1fcb2c | 2409 | blk_finish_plug(&plug); |
61628a3f MC |
2410 | goto out_writepages; |
2411 | } | |
f63e6005 TT |
2412 | |
2413 | /* | |
8eb9e5ce | 2414 | * Now call write_cache_pages_da() to find the next |
f63e6005 | 2415 | * contiguous region of logical blocks that need |
8eb9e5ce | 2416 | * blocks to be allocated by ext4 and submit them. |
f63e6005 | 2417 | */ |
72f84e65 | 2418 | ret = write_cache_pages_da(mapping, wbc, &mpd, &done_index); |
f63e6005 | 2419 | /* |
af901ca1 | 2420 | * If we have a contiguous extent of pages and we |
f63e6005 TT |
2421 | * haven't done the I/O yet, map the blocks and submit |
2422 | * them for I/O. | |
2423 | */ | |
2424 | if (!mpd.io_done && mpd.next_page != mpd.first_page) { | |
5a87b7a5 | 2425 | mpage_da_map_and_submit(&mpd); |
f63e6005 TT |
2426 | ret = MPAGE_DA_EXTENT_TAIL; |
2427 | } | |
b3a3ca8c | 2428 | trace_ext4_da_write_pages(inode, &mpd); |
f63e6005 | 2429 | wbc->nr_to_write -= mpd.pages_written; |
df22291f | 2430 | |
61628a3f | 2431 | ext4_journal_stop(handle); |
df22291f | 2432 | |
8f64b32e | 2433 | if ((mpd.retval == -ENOSPC) && sbi->s_journal) { |
22208ded AK |
2434 | /* commit the transaction which would |
2435 | * free blocks released in the transaction | |
2436 | * and try again | |
2437 | */ | |
df22291f | 2438 | jbd2_journal_force_commit_nested(sbi->s_journal); |
22208ded AK |
2439 | ret = 0; |
2440 | } else if (ret == MPAGE_DA_EXTENT_TAIL) { | |
a1d6cc56 | 2441 | /* |
8de49e67 KM |
2442 | * Got one extent now try with rest of the pages. |
2443 | * If mpd.retval is set -EIO, journal is aborted. | |
2444 | * So we don't need to write any more. | |
a1d6cc56 | 2445 | */ |
22208ded | 2446 | pages_written += mpd.pages_written; |
8de49e67 | 2447 | ret = mpd.retval; |
2acf2c26 | 2448 | io_done = 1; |
22208ded | 2449 | } else if (wbc->nr_to_write) |
61628a3f MC |
2450 | /* |
2451 | * There is no more writeout needed | |
2452 | * or we requested for a noblocking writeout | |
2453 | * and we found the device congested | |
2454 | */ | |
61628a3f | 2455 | break; |
a1d6cc56 | 2456 | } |
1bce63d1 | 2457 | blk_finish_plug(&plug); |
2acf2c26 AK |
2458 | if (!io_done && !cycled) { |
2459 | cycled = 1; | |
2460 | index = 0; | |
2461 | wbc->range_start = index << PAGE_CACHE_SHIFT; | |
2462 | wbc->range_end = mapping->writeback_index - 1; | |
2463 | goto retry; | |
2464 | } | |
22208ded AK |
2465 | |
2466 | /* Update index */ | |
2acf2c26 | 2467 | wbc->range_cyclic = range_cyclic; |
22208ded AK |
2468 | if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) |
2469 | /* | |
2470 | * set the writeback_index so that range_cyclic | |
2471 | * mode will write it back later | |
2472 | */ | |
72f84e65 | 2473 | mapping->writeback_index = done_index; |
a1d6cc56 | 2474 | |
61628a3f | 2475 | out_writepages: |
2faf2e19 | 2476 | wbc->nr_to_write -= nr_to_writebump; |
de89de6e | 2477 | wbc->range_start = range_start; |
9bffad1e | 2478 | trace_ext4_da_writepages_result(inode, wbc, ret, pages_written); |
61628a3f | 2479 | return ret; |
64769240 AT |
2480 | } |
2481 | ||
79f0be8d AK |
2482 | #define FALL_BACK_TO_NONDELALLOC 1 |
2483 | static int ext4_nonda_switch(struct super_block *sb) | |
2484 | { | |
2485 | s64 free_blocks, dirty_blocks; | |
2486 | struct ext4_sb_info *sbi = EXT4_SB(sb); | |
2487 | ||
2488 | /* | |
2489 | * switch to non delalloc mode if we are running low | |
2490 | * on free block. The free block accounting via percpu | |
179f7ebf | 2491 | * counters can get slightly wrong with percpu_counter_batch getting |
79f0be8d AK |
2492 | * accumulated on each CPU without updating global counters |
2493 | * Delalloc need an accurate free block accounting. So switch | |
2494 | * to non delalloc when we are near to error range. | |
2495 | */ | |
57042651 TT |
2496 | free_blocks = EXT4_C2B(sbi, |
2497 | percpu_counter_read_positive(&sbi->s_freeclusters_counter)); | |
2498 | dirty_blocks = percpu_counter_read_positive(&sbi->s_dirtyclusters_counter); | |
00d4e736 TT |
2499 | /* |
2500 | * Start pushing delalloc when 1/2 of free blocks are dirty. | |
2501 | */ | |
2502 | if (dirty_blocks && (free_blocks < 2 * dirty_blocks) && | |
2503 | !writeback_in_progress(sb->s_bdi) && | |
2504 | down_read_trylock(&sb->s_umount)) { | |
2505 | writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE); | |
2506 | up_read(&sb->s_umount); | |
2507 | } | |
2508 | ||
79f0be8d | 2509 | if (2 * free_blocks < 3 * dirty_blocks || |
df55c99d | 2510 | free_blocks < (dirty_blocks + EXT4_FREECLUSTERS_WATERMARK)) { |
79f0be8d | 2511 | /* |
c8afb446 ES |
2512 | * free block count is less than 150% of dirty blocks |
2513 | * or free blocks is less than watermark | |
79f0be8d AK |
2514 | */ |
2515 | return 1; | |
2516 | } | |
2517 | return 0; | |
2518 | } | |
2519 | ||
64769240 | 2520 | static int ext4_da_write_begin(struct file *file, struct address_space *mapping, |
de9a55b8 TT |
2521 | loff_t pos, unsigned len, unsigned flags, |
2522 | struct page **pagep, void **fsdata) | |
64769240 | 2523 | { |
72b8ab9d | 2524 | int ret, retries = 0; |
64769240 AT |
2525 | struct page *page; |
2526 | pgoff_t index; | |
64769240 AT |
2527 | struct inode *inode = mapping->host; |
2528 | handle_t *handle; | |
2529 | ||
2530 | index = pos >> PAGE_CACHE_SHIFT; | |
79f0be8d AK |
2531 | |
2532 | if (ext4_nonda_switch(inode->i_sb)) { | |
2533 | *fsdata = (void *)FALL_BACK_TO_NONDELALLOC; | |
2534 | return ext4_write_begin(file, mapping, pos, | |
2535 | len, flags, pagep, fsdata); | |
2536 | } | |
2537 | *fsdata = (void *)0; | |
9bffad1e | 2538 | trace_ext4_da_write_begin(inode, pos, len, flags); |
d2a17637 | 2539 | retry: |
64769240 AT |
2540 | /* |
2541 | * With delayed allocation, we don't log the i_disksize update | |
2542 | * if there is delayed block allocation. But we still need | |
2543 | * to journalling the i_disksize update if writes to the end | |
2544 | * of file which has an already mapped buffer. | |
2545 | */ | |
2546 | handle = ext4_journal_start(inode, 1); | |
2547 | if (IS_ERR(handle)) { | |
2548 | ret = PTR_ERR(handle); | |
2549 | goto out; | |
2550 | } | |
ebd3610b JK |
2551 | /* We cannot recurse into the filesystem as the transaction is already |
2552 | * started */ | |
2553 | flags |= AOP_FLAG_NOFS; | |
64769240 | 2554 | |
54566b2c | 2555 | page = grab_cache_page_write_begin(mapping, index, flags); |
d5a0d4f7 ES |
2556 | if (!page) { |
2557 | ext4_journal_stop(handle); | |
2558 | ret = -ENOMEM; | |
2559 | goto out; | |
2560 | } | |
64769240 AT |
2561 | *pagep = page; |
2562 | ||
6e1db88d | 2563 | ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep); |
64769240 AT |
2564 | if (ret < 0) { |
2565 | unlock_page(page); | |
2566 | ext4_journal_stop(handle); | |
2567 | page_cache_release(page); | |
ae4d5372 AK |
2568 | /* |
2569 | * block_write_begin may have instantiated a few blocks | |
2570 | * outside i_size. Trim these off again. Don't need | |
2571 | * i_size_read because we hold i_mutex. | |
2572 | */ | |
2573 | if (pos + len > inode->i_size) | |
b9a4207d | 2574 | ext4_truncate_failed_write(inode); |
64769240 AT |
2575 | } |
2576 | ||
d2a17637 MC |
2577 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) |
2578 | goto retry; | |
64769240 AT |
2579 | out: |
2580 | return ret; | |
2581 | } | |
2582 | ||
632eaeab MC |
2583 | /* |
2584 | * Check if we should update i_disksize | |
2585 | * when write to the end of file but not require block allocation | |
2586 | */ | |
2587 | static int ext4_da_should_update_i_disksize(struct page *page, | |
de9a55b8 | 2588 | unsigned long offset) |
632eaeab MC |
2589 | { |
2590 | struct buffer_head *bh; | |
2591 | struct inode *inode = page->mapping->host; | |
2592 | unsigned int idx; | |
2593 | int i; | |
2594 | ||
2595 | bh = page_buffers(page); | |
2596 | idx = offset >> inode->i_blkbits; | |
2597 | ||
af5bc92d | 2598 | for (i = 0; i < idx; i++) |
632eaeab MC |
2599 | bh = bh->b_this_page; |
2600 | ||
29fa89d0 | 2601 | if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh)) |
632eaeab MC |
2602 | return 0; |
2603 | return 1; | |
2604 | } | |
2605 | ||
64769240 | 2606 | static int ext4_da_write_end(struct file *file, |
de9a55b8 TT |
2607 | struct address_space *mapping, |
2608 | loff_t pos, unsigned len, unsigned copied, | |
2609 | struct page *page, void *fsdata) | |
64769240 AT |
2610 | { |
2611 | struct inode *inode = mapping->host; | |
2612 | int ret = 0, ret2; | |
2613 | handle_t *handle = ext4_journal_current_handle(); | |
2614 | loff_t new_i_size; | |
632eaeab | 2615 | unsigned long start, end; |
79f0be8d AK |
2616 | int write_mode = (int)(unsigned long)fsdata; |
2617 | ||
2618 | if (write_mode == FALL_BACK_TO_NONDELALLOC) { | |
3d2b1582 LC |
2619 | switch (ext4_inode_journal_mode(inode)) { |
2620 | case EXT4_INODE_ORDERED_DATA_MODE: | |
79f0be8d AK |
2621 | return ext4_ordered_write_end(file, mapping, pos, |
2622 | len, copied, page, fsdata); | |
3d2b1582 | 2623 | case EXT4_INODE_WRITEBACK_DATA_MODE: |
79f0be8d AK |
2624 | return ext4_writeback_write_end(file, mapping, pos, |
2625 | len, copied, page, fsdata); | |
3d2b1582 | 2626 | default: |
79f0be8d AK |
2627 | BUG(); |
2628 | } | |
2629 | } | |
632eaeab | 2630 | |
9bffad1e | 2631 | trace_ext4_da_write_end(inode, pos, len, copied); |
632eaeab | 2632 | start = pos & (PAGE_CACHE_SIZE - 1); |
af5bc92d | 2633 | end = start + copied - 1; |
64769240 AT |
2634 | |
2635 | /* | |
2636 | * generic_write_end() will run mark_inode_dirty() if i_size | |
2637 | * changes. So let's piggyback the i_disksize mark_inode_dirty | |
2638 | * into that. | |
2639 | */ | |
2640 | ||
2641 | new_i_size = pos + copied; | |
ea51d132 | 2642 | if (copied && new_i_size > EXT4_I(inode)->i_disksize) { |
632eaeab MC |
2643 | if (ext4_da_should_update_i_disksize(page, end)) { |
2644 | down_write(&EXT4_I(inode)->i_data_sem); | |
2645 | if (new_i_size > EXT4_I(inode)->i_disksize) { | |
2646 | /* | |
2647 | * Updating i_disksize when extending file | |
2648 | * without needing block allocation | |
2649 | */ | |
2650 | if (ext4_should_order_data(inode)) | |
2651 | ret = ext4_jbd2_file_inode(handle, | |
2652 | inode); | |
64769240 | 2653 | |
632eaeab MC |
2654 | EXT4_I(inode)->i_disksize = new_i_size; |
2655 | } | |
2656 | up_write(&EXT4_I(inode)->i_data_sem); | |
cf17fea6 AK |
2657 | /* We need to mark inode dirty even if |
2658 | * new_i_size is less that inode->i_size | |
2659 | * bu greater than i_disksize.(hint delalloc) | |
2660 | */ | |
2661 | ext4_mark_inode_dirty(handle, inode); | |
64769240 | 2662 | } |
632eaeab | 2663 | } |
64769240 AT |
2664 | ret2 = generic_write_end(file, mapping, pos, len, copied, |
2665 | page, fsdata); | |
2666 | copied = ret2; | |
2667 | if (ret2 < 0) | |
2668 | ret = ret2; | |
2669 | ret2 = ext4_journal_stop(handle); | |
2670 | if (!ret) | |
2671 | ret = ret2; | |
2672 | ||
2673 | return ret ? ret : copied; | |
2674 | } | |
2675 | ||
2676 | static void ext4_da_invalidatepage(struct page *page, unsigned long offset) | |
2677 | { | |
64769240 AT |
2678 | /* |
2679 | * Drop reserved blocks | |
2680 | */ | |
2681 | BUG_ON(!PageLocked(page)); | |
2682 | if (!page_has_buffers(page)) | |
2683 | goto out; | |
2684 | ||
d2a17637 | 2685 | ext4_da_page_release_reservation(page, offset); |
64769240 AT |
2686 | |
2687 | out: | |
2688 | ext4_invalidatepage(page, offset); | |
2689 | ||
2690 | return; | |
2691 | } | |
2692 | ||
ccd2506b TT |
2693 | /* |
2694 | * Force all delayed allocation blocks to be allocated for a given inode. | |
2695 | */ | |
2696 | int ext4_alloc_da_blocks(struct inode *inode) | |
2697 | { | |
fb40ba0d TT |
2698 | trace_ext4_alloc_da_blocks(inode); |
2699 | ||
ccd2506b TT |
2700 | if (!EXT4_I(inode)->i_reserved_data_blocks && |
2701 | !EXT4_I(inode)->i_reserved_meta_blocks) | |
2702 | return 0; | |
2703 | ||
2704 | /* | |
2705 | * We do something simple for now. The filemap_flush() will | |
2706 | * also start triggering a write of the data blocks, which is | |
2707 | * not strictly speaking necessary (and for users of | |
2708 | * laptop_mode, not even desirable). However, to do otherwise | |
2709 | * would require replicating code paths in: | |
de9a55b8 | 2710 | * |
ccd2506b TT |
2711 | * ext4_da_writepages() -> |
2712 | * write_cache_pages() ---> (via passed in callback function) | |
2713 | * __mpage_da_writepage() --> | |
2714 | * mpage_add_bh_to_extent() | |
2715 | * mpage_da_map_blocks() | |
2716 | * | |
2717 | * The problem is that write_cache_pages(), located in | |
2718 | * mm/page-writeback.c, marks pages clean in preparation for | |
2719 | * doing I/O, which is not desirable if we're not planning on | |
2720 | * doing I/O at all. | |
2721 | * | |
2722 | * We could call write_cache_pages(), and then redirty all of | |
380cf090 | 2723 | * the pages by calling redirty_page_for_writepage() but that |
ccd2506b TT |
2724 | * would be ugly in the extreme. So instead we would need to |
2725 | * replicate parts of the code in the above functions, | |
25985edc | 2726 | * simplifying them because we wouldn't actually intend to |
ccd2506b TT |
2727 | * write out the pages, but rather only collect contiguous |
2728 | * logical block extents, call the multi-block allocator, and | |
2729 | * then update the buffer heads with the block allocations. | |
de9a55b8 | 2730 | * |
ccd2506b TT |
2731 | * For now, though, we'll cheat by calling filemap_flush(), |
2732 | * which will map the blocks, and start the I/O, but not | |
2733 | * actually wait for the I/O to complete. | |
2734 | */ | |
2735 | return filemap_flush(inode->i_mapping); | |
2736 | } | |
64769240 | 2737 | |
ac27a0ec DK |
2738 | /* |
2739 | * bmap() is special. It gets used by applications such as lilo and by | |
2740 | * the swapper to find the on-disk block of a specific piece of data. | |
2741 | * | |
2742 | * Naturally, this is dangerous if the block concerned is still in the | |
617ba13b | 2743 | * journal. If somebody makes a swapfile on an ext4 data-journaling |
ac27a0ec DK |
2744 | * filesystem and enables swap, then they may get a nasty shock when the |
2745 | * data getting swapped to that swapfile suddenly gets overwritten by | |
2746 | * the original zero's written out previously to the journal and | |
2747 | * awaiting writeback in the kernel's buffer cache. | |
2748 | * | |
2749 | * So, if we see any bmap calls here on a modified, data-journaled file, | |
2750 | * take extra steps to flush any blocks which might be in the cache. | |
2751 | */ | |
617ba13b | 2752 | static sector_t ext4_bmap(struct address_space *mapping, sector_t block) |
ac27a0ec DK |
2753 | { |
2754 | struct inode *inode = mapping->host; | |
2755 | journal_t *journal; | |
2756 | int err; | |
2757 | ||
64769240 AT |
2758 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) && |
2759 | test_opt(inode->i_sb, DELALLOC)) { | |
2760 | /* | |
2761 | * With delalloc we want to sync the file | |
2762 | * so that we can make sure we allocate | |
2763 | * blocks for file | |
2764 | */ | |
2765 | filemap_write_and_wait(mapping); | |
2766 | } | |
2767 | ||
19f5fb7a TT |
2768 | if (EXT4_JOURNAL(inode) && |
2769 | ext4_test_inode_state(inode, EXT4_STATE_JDATA)) { | |
ac27a0ec DK |
2770 | /* |
2771 | * This is a REALLY heavyweight approach, but the use of | |
2772 | * bmap on dirty files is expected to be extremely rare: | |
2773 | * only if we run lilo or swapon on a freshly made file | |
2774 | * do we expect this to happen. | |
2775 | * | |
2776 | * (bmap requires CAP_SYS_RAWIO so this does not | |
2777 | * represent an unprivileged user DOS attack --- we'd be | |
2778 | * in trouble if mortal users could trigger this path at | |
2779 | * will.) | |
2780 | * | |
617ba13b | 2781 | * NB. EXT4_STATE_JDATA is not set on files other than |
ac27a0ec DK |
2782 | * regular files. If somebody wants to bmap a directory |
2783 | * or symlink and gets confused because the buffer | |
2784 | * hasn't yet been flushed to disk, they deserve | |
2785 | * everything they get. | |
2786 | */ | |
2787 | ||
19f5fb7a | 2788 | ext4_clear_inode_state(inode, EXT4_STATE_JDATA); |
617ba13b | 2789 | journal = EXT4_JOURNAL(inode); |
dab291af MC |
2790 | jbd2_journal_lock_updates(journal); |
2791 | err = jbd2_journal_flush(journal); | |
2792 | jbd2_journal_unlock_updates(journal); | |
ac27a0ec DK |
2793 | |
2794 | if (err) | |
2795 | return 0; | |
2796 | } | |
2797 | ||
af5bc92d | 2798 | return generic_block_bmap(mapping, block, ext4_get_block); |
ac27a0ec DK |
2799 | } |
2800 | ||
617ba13b | 2801 | static int ext4_readpage(struct file *file, struct page *page) |
ac27a0ec | 2802 | { |
0562e0ba | 2803 | trace_ext4_readpage(page); |
617ba13b | 2804 | return mpage_readpage(page, ext4_get_block); |
ac27a0ec DK |
2805 | } |
2806 | ||
2807 | static int | |
617ba13b | 2808 | ext4_readpages(struct file *file, struct address_space *mapping, |
ac27a0ec DK |
2809 | struct list_head *pages, unsigned nr_pages) |
2810 | { | |
617ba13b | 2811 | return mpage_readpages(mapping, pages, nr_pages, ext4_get_block); |
ac27a0ec DK |
2812 | } |
2813 | ||
744692dc JZ |
2814 | static void ext4_invalidatepage_free_endio(struct page *page, unsigned long offset) |
2815 | { | |
2816 | struct buffer_head *head, *bh; | |
2817 | unsigned int curr_off = 0; | |
2818 | ||
2819 | if (!page_has_buffers(page)) | |
2820 | return; | |
2821 | head = bh = page_buffers(page); | |
2822 | do { | |
2823 | if (offset <= curr_off && test_clear_buffer_uninit(bh) | |
2824 | && bh->b_private) { | |
2825 | ext4_free_io_end(bh->b_private); | |
2826 | bh->b_private = NULL; | |
2827 | bh->b_end_io = NULL; | |
2828 | } | |
2829 | curr_off = curr_off + bh->b_size; | |
2830 | bh = bh->b_this_page; | |
2831 | } while (bh != head); | |
2832 | } | |
2833 | ||
617ba13b | 2834 | static void ext4_invalidatepage(struct page *page, unsigned long offset) |
ac27a0ec | 2835 | { |
617ba13b | 2836 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); |
ac27a0ec | 2837 | |
0562e0ba JZ |
2838 | trace_ext4_invalidatepage(page, offset); |
2839 | ||
744692dc JZ |
2840 | /* |
2841 | * free any io_end structure allocated for buffers to be discarded | |
2842 | */ | |
2843 | if (ext4_should_dioread_nolock(page->mapping->host)) | |
2844 | ext4_invalidatepage_free_endio(page, offset); | |
ac27a0ec DK |
2845 | /* |
2846 | * If it's a full truncate we just forget about the pending dirtying | |
2847 | */ | |
2848 | if (offset == 0) | |
2849 | ClearPageChecked(page); | |
2850 | ||
0390131b FM |
2851 | if (journal) |
2852 | jbd2_journal_invalidatepage(journal, page, offset); | |
2853 | else | |
2854 | block_invalidatepage(page, offset); | |
ac27a0ec DK |
2855 | } |
2856 | ||
617ba13b | 2857 | static int ext4_releasepage(struct page *page, gfp_t wait) |
ac27a0ec | 2858 | { |
617ba13b | 2859 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); |
ac27a0ec | 2860 | |
0562e0ba JZ |
2861 | trace_ext4_releasepage(page); |
2862 | ||
ac27a0ec DK |
2863 | WARN_ON(PageChecked(page)); |
2864 | if (!page_has_buffers(page)) | |
2865 | return 0; | |
0390131b FM |
2866 | if (journal) |
2867 | return jbd2_journal_try_to_free_buffers(journal, page, wait); | |
2868 | else | |
2869 | return try_to_free_buffers(page); | |
ac27a0ec DK |
2870 | } |
2871 | ||
2ed88685 TT |
2872 | /* |
2873 | * ext4_get_block used when preparing for a DIO write or buffer write. | |
2874 | * We allocate an uinitialized extent if blocks haven't been allocated. | |
2875 | * The extent will be converted to initialized after the IO is complete. | |
2876 | */ | |
c7064ef1 | 2877 | static int ext4_get_block_write(struct inode *inode, sector_t iblock, |
4c0425ff MC |
2878 | struct buffer_head *bh_result, int create) |
2879 | { | |
c7064ef1 | 2880 | ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n", |
8d5d02e6 | 2881 | inode->i_ino, create); |
2ed88685 TT |
2882 | return _ext4_get_block(inode, iblock, bh_result, |
2883 | EXT4_GET_BLOCKS_IO_CREATE_EXT); | |
4c0425ff MC |
2884 | } |
2885 | ||
729f52c6 | 2886 | static int ext4_get_block_write_nolock(struct inode *inode, sector_t iblock, |
8b0f165f | 2887 | struct buffer_head *bh_result, int create) |
729f52c6 | 2888 | { |
8b0f165f AP |
2889 | ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n", |
2890 | inode->i_ino, create); | |
2891 | return _ext4_get_block(inode, iblock, bh_result, | |
2892 | EXT4_GET_BLOCKS_NO_LOCK); | |
729f52c6 ZL |
2893 | } |
2894 | ||
4c0425ff | 2895 | static void ext4_end_io_dio(struct kiocb *iocb, loff_t offset, |
552ef802 CH |
2896 | ssize_t size, void *private, int ret, |
2897 | bool is_async) | |
4c0425ff | 2898 | { |
72c5052d | 2899 | struct inode *inode = iocb->ki_filp->f_path.dentry->d_inode; |
4c0425ff | 2900 | ext4_io_end_t *io_end = iocb->private; |
4c0425ff | 2901 | |
4b70df18 M |
2902 | /* if not async direct IO or dio with 0 bytes write, just return */ |
2903 | if (!io_end || !size) | |
552ef802 | 2904 | goto out; |
4b70df18 | 2905 | |
88635ca2 | 2906 | ext_debug("ext4_end_io_dio(): io_end 0x%p " |
ace36ad4 | 2907 | "for inode %lu, iocb 0x%p, offset %llu, size %zd\n", |
8d5d02e6 MC |
2908 | iocb->private, io_end->inode->i_ino, iocb, offset, |
2909 | size); | |
8d5d02e6 | 2910 | |
b5a7e970 TT |
2911 | iocb->private = NULL; |
2912 | ||
8d5d02e6 | 2913 | /* if not aio dio with unwritten extents, just free io and return */ |
bd2d0210 | 2914 | if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) { |
8d5d02e6 | 2915 | ext4_free_io_end(io_end); |
5b3ff237 JZ |
2916 | out: |
2917 | if (is_async) | |
2918 | aio_complete(iocb, ret, 0); | |
72c5052d | 2919 | inode_dio_done(inode); |
5b3ff237 | 2920 | return; |
8d5d02e6 MC |
2921 | } |
2922 | ||
4c0425ff MC |
2923 | io_end->offset = offset; |
2924 | io_end->size = size; | |
5b3ff237 JZ |
2925 | if (is_async) { |
2926 | io_end->iocb = iocb; | |
2927 | io_end->result = ret; | |
2928 | } | |
4c0425ff | 2929 | |
28a535f9 | 2930 | ext4_add_complete_io(io_end); |
4c0425ff | 2931 | } |
c7064ef1 | 2932 | |
744692dc JZ |
2933 | static void ext4_end_io_buffer_write(struct buffer_head *bh, int uptodate) |
2934 | { | |
2935 | ext4_io_end_t *io_end = bh->b_private; | |
744692dc | 2936 | struct inode *inode; |
744692dc JZ |
2937 | |
2938 | if (!test_clear_buffer_uninit(bh) || !io_end) | |
2939 | goto out; | |
2940 | ||
2941 | if (!(io_end->inode->i_sb->s_flags & MS_ACTIVE)) { | |
92b97816 TT |
2942 | ext4_msg(io_end->inode->i_sb, KERN_INFO, |
2943 | "sb umounted, discard end_io request for inode %lu", | |
2944 | io_end->inode->i_ino); | |
744692dc JZ |
2945 | ext4_free_io_end(io_end); |
2946 | goto out; | |
2947 | } | |
2948 | ||
32c80b32 TM |
2949 | /* |
2950 | * It may be over-defensive here to check EXT4_IO_END_UNWRITTEN now, | |
2951 | * but being more careful is always safe for the future change. | |
2952 | */ | |
744692dc | 2953 | inode = io_end->inode; |
0edeb71d | 2954 | ext4_set_io_unwritten_flag(inode, io_end); |
28a535f9 | 2955 | ext4_add_complete_io(io_end); |
744692dc JZ |
2956 | out: |
2957 | bh->b_private = NULL; | |
2958 | bh->b_end_io = NULL; | |
2959 | clear_buffer_uninit(bh); | |
2960 | end_buffer_async_write(bh, uptodate); | |
2961 | } | |
2962 | ||
2963 | static int ext4_set_bh_endio(struct buffer_head *bh, struct inode *inode) | |
2964 | { | |
2965 | ext4_io_end_t *io_end; | |
2966 | struct page *page = bh->b_page; | |
2967 | loff_t offset = (sector_t)page->index << PAGE_CACHE_SHIFT; | |
2968 | size_t size = bh->b_size; | |
2969 | ||
2970 | retry: | |
2971 | io_end = ext4_init_io_end(inode, GFP_ATOMIC); | |
2972 | if (!io_end) { | |
6db26ffc | 2973 | pr_warn_ratelimited("%s: allocation fail\n", __func__); |
744692dc JZ |
2974 | schedule(); |
2975 | goto retry; | |
2976 | } | |
2977 | io_end->offset = offset; | |
2978 | io_end->size = size; | |
2979 | /* | |
2980 | * We need to hold a reference to the page to make sure it | |
2981 | * doesn't get evicted before ext4_end_io_work() has a chance | |
2982 | * to convert the extent from written to unwritten. | |
2983 | */ | |
2984 | io_end->page = page; | |
2985 | get_page(io_end->page); | |
2986 | ||
2987 | bh->b_private = io_end; | |
2988 | bh->b_end_io = ext4_end_io_buffer_write; | |
2989 | return 0; | |
2990 | } | |
2991 | ||
4c0425ff MC |
2992 | /* |
2993 | * For ext4 extent files, ext4 will do direct-io write to holes, | |
2994 | * preallocated extents, and those write extend the file, no need to | |
2995 | * fall back to buffered IO. | |
2996 | * | |
b595076a | 2997 | * For holes, we fallocate those blocks, mark them as uninitialized |
4c0425ff | 2998 | * If those blocks were preallocated, we mark sure they are splited, but |
b595076a | 2999 | * still keep the range to write as uninitialized. |
4c0425ff | 3000 | * |
8d5d02e6 MC |
3001 | * The unwrritten extents will be converted to written when DIO is completed. |
3002 | * For async direct IO, since the IO may still pending when return, we | |
25985edc | 3003 | * set up an end_io call back function, which will do the conversion |
8d5d02e6 | 3004 | * when async direct IO completed. |
4c0425ff MC |
3005 | * |
3006 | * If the O_DIRECT write will extend the file then add this inode to the | |
3007 | * orphan list. So recovery will truncate it back to the original size | |
3008 | * if the machine crashes during the write. | |
3009 | * | |
3010 | */ | |
3011 | static ssize_t ext4_ext_direct_IO(int rw, struct kiocb *iocb, | |
3012 | const struct iovec *iov, loff_t offset, | |
3013 | unsigned long nr_segs) | |
3014 | { | |
3015 | struct file *file = iocb->ki_filp; | |
3016 | struct inode *inode = file->f_mapping->host; | |
3017 | ssize_t ret; | |
3018 | size_t count = iov_length(iov, nr_segs); | |
3019 | ||
3020 | loff_t final_size = offset + count; | |
3021 | if (rw == WRITE && final_size <= inode->i_size) { | |
729f52c6 | 3022 | int overwrite = 0; |
8b0f165f AP |
3023 | get_block_t *get_block_func = NULL; |
3024 | int dio_flags = 0; | |
729f52c6 | 3025 | |
4bd809db ZL |
3026 | BUG_ON(iocb->private == NULL); |
3027 | ||
3028 | /* If we do a overwrite dio, i_mutex locking can be released */ | |
3029 | overwrite = *((int *)iocb->private); | |
3030 | ||
3031 | if (overwrite) { | |
1f555cfa | 3032 | atomic_inc(&inode->i_dio_count); |
4bd809db ZL |
3033 | down_read(&EXT4_I(inode)->i_data_sem); |
3034 | mutex_unlock(&inode->i_mutex); | |
3035 | } | |
3036 | ||
4c0425ff | 3037 | /* |
8d5d02e6 MC |
3038 | * We could direct write to holes and fallocate. |
3039 | * | |
3040 | * Allocated blocks to fill the hole are marked as uninitialized | |
25985edc | 3041 | * to prevent parallel buffered read to expose the stale data |
4c0425ff | 3042 | * before DIO complete the data IO. |
8d5d02e6 MC |
3043 | * |
3044 | * As to previously fallocated extents, ext4 get_block | |
4c0425ff MC |
3045 | * will just simply mark the buffer mapped but still |
3046 | * keep the extents uninitialized. | |
3047 | * | |
8d5d02e6 MC |
3048 | * for non AIO case, we will convert those unwritten extents |
3049 | * to written after return back from blockdev_direct_IO. | |
3050 | * | |
3051 | * for async DIO, the conversion needs to be defered when | |
3052 | * the IO is completed. The ext4 end_io callback function | |
3053 | * will be called to take care of the conversion work. | |
3054 | * Here for async case, we allocate an io_end structure to | |
3055 | * hook to the iocb. | |
4c0425ff | 3056 | */ |
8d5d02e6 | 3057 | iocb->private = NULL; |
f45ee3a1 | 3058 | ext4_inode_aio_set(inode, NULL); |
8d5d02e6 | 3059 | if (!is_sync_kiocb(iocb)) { |
266991b1 JM |
3060 | ext4_io_end_t *io_end = |
3061 | ext4_init_io_end(inode, GFP_NOFS); | |
4bd809db ZL |
3062 | if (!io_end) { |
3063 | ret = -ENOMEM; | |
3064 | goto retake_lock; | |
3065 | } | |
266991b1 JM |
3066 | io_end->flag |= EXT4_IO_END_DIRECT; |
3067 | iocb->private = io_end; | |
8d5d02e6 MC |
3068 | /* |
3069 | * we save the io structure for current async | |
79e83036 | 3070 | * direct IO, so that later ext4_map_blocks() |
8d5d02e6 MC |
3071 | * could flag the io structure whether there |
3072 | * is a unwritten extents needs to be converted | |
3073 | * when IO is completed. | |
3074 | */ | |
f45ee3a1 | 3075 | ext4_inode_aio_set(inode, io_end); |
8d5d02e6 MC |
3076 | } |
3077 | ||
8b0f165f AP |
3078 | if (overwrite) { |
3079 | get_block_func = ext4_get_block_write_nolock; | |
3080 | } else { | |
3081 | get_block_func = ext4_get_block_write; | |
3082 | dio_flags = DIO_LOCKING; | |
3083 | } | |
3084 | ret = __blockdev_direct_IO(rw, iocb, inode, | |
3085 | inode->i_sb->s_bdev, iov, | |
3086 | offset, nr_segs, | |
3087 | get_block_func, | |
3088 | ext4_end_io_dio, | |
3089 | NULL, | |
3090 | dio_flags); | |
3091 | ||
8d5d02e6 | 3092 | if (iocb->private) |
f45ee3a1 | 3093 | ext4_inode_aio_set(inode, NULL); |
8d5d02e6 MC |
3094 | /* |
3095 | * The io_end structure takes a reference to the inode, | |
3096 | * that structure needs to be destroyed and the | |
3097 | * reference to the inode need to be dropped, when IO is | |
3098 | * complete, even with 0 byte write, or failed. | |
3099 | * | |
3100 | * In the successful AIO DIO case, the io_end structure will be | |
3101 | * desctroyed and the reference to the inode will be dropped | |
3102 | * after the end_io call back function is called. | |
3103 | * | |
3104 | * In the case there is 0 byte write, or error case, since | |
3105 | * VFS direct IO won't invoke the end_io call back function, | |
3106 | * we need to free the end_io structure here. | |
3107 | */ | |
3108 | if (ret != -EIOCBQUEUED && ret <= 0 && iocb->private) { | |
3109 | ext4_free_io_end(iocb->private); | |
3110 | iocb->private = NULL; | |
729f52c6 | 3111 | } else if (ret > 0 && !overwrite && ext4_test_inode_state(inode, |
19f5fb7a | 3112 | EXT4_STATE_DIO_UNWRITTEN)) { |
109f5565 | 3113 | int err; |
8d5d02e6 MC |
3114 | /* |
3115 | * for non AIO case, since the IO is already | |
25985edc | 3116 | * completed, we could do the conversion right here |
8d5d02e6 | 3117 | */ |
109f5565 M |
3118 | err = ext4_convert_unwritten_extents(inode, |
3119 | offset, ret); | |
3120 | if (err < 0) | |
3121 | ret = err; | |
19f5fb7a | 3122 | ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN); |
109f5565 | 3123 | } |
4bd809db ZL |
3124 | |
3125 | retake_lock: | |
3126 | /* take i_mutex locking again if we do a ovewrite dio */ | |
3127 | if (overwrite) { | |
1f555cfa | 3128 | inode_dio_done(inode); |
4bd809db ZL |
3129 | up_read(&EXT4_I(inode)->i_data_sem); |
3130 | mutex_lock(&inode->i_mutex); | |
3131 | } | |
3132 | ||
4c0425ff MC |
3133 | return ret; |
3134 | } | |
8d5d02e6 MC |
3135 | |
3136 | /* for write the the end of file case, we fall back to old way */ | |
4c0425ff MC |
3137 | return ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs); |
3138 | } | |
3139 | ||
3140 | static ssize_t ext4_direct_IO(int rw, struct kiocb *iocb, | |
3141 | const struct iovec *iov, loff_t offset, | |
3142 | unsigned long nr_segs) | |
3143 | { | |
3144 | struct file *file = iocb->ki_filp; | |
3145 | struct inode *inode = file->f_mapping->host; | |
0562e0ba | 3146 | ssize_t ret; |
4c0425ff | 3147 | |
84ebd795 TT |
3148 | /* |
3149 | * If we are doing data journalling we don't support O_DIRECT | |
3150 | */ | |
3151 | if (ext4_should_journal_data(inode)) | |
3152 | return 0; | |
3153 | ||
0562e0ba | 3154 | trace_ext4_direct_IO_enter(inode, offset, iov_length(iov, nr_segs), rw); |
12e9b892 | 3155 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
0562e0ba JZ |
3156 | ret = ext4_ext_direct_IO(rw, iocb, iov, offset, nr_segs); |
3157 | else | |
3158 | ret = ext4_ind_direct_IO(rw, iocb, iov, offset, nr_segs); | |
3159 | trace_ext4_direct_IO_exit(inode, offset, | |
3160 | iov_length(iov, nr_segs), rw, ret); | |
3161 | return ret; | |
4c0425ff MC |
3162 | } |
3163 | ||
ac27a0ec | 3164 | /* |
617ba13b | 3165 | * Pages can be marked dirty completely asynchronously from ext4's journalling |
ac27a0ec DK |
3166 | * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do |
3167 | * much here because ->set_page_dirty is called under VFS locks. The page is | |
3168 | * not necessarily locked. | |
3169 | * | |
3170 | * We cannot just dirty the page and leave attached buffers clean, because the | |
3171 | * buffers' dirty state is "definitive". We cannot just set the buffers dirty | |
3172 | * or jbddirty because all the journalling code will explode. | |
3173 | * | |
3174 | * So what we do is to mark the page "pending dirty" and next time writepage | |
3175 | * is called, propagate that into the buffers appropriately. | |
3176 | */ | |
617ba13b | 3177 | static int ext4_journalled_set_page_dirty(struct page *page) |
ac27a0ec DK |
3178 | { |
3179 | SetPageChecked(page); | |
3180 | return __set_page_dirty_nobuffers(page); | |
3181 | } | |
3182 | ||
617ba13b | 3183 | static const struct address_space_operations ext4_ordered_aops = { |
8ab22b9a HH |
3184 | .readpage = ext4_readpage, |
3185 | .readpages = ext4_readpages, | |
43ce1d23 | 3186 | .writepage = ext4_writepage, |
8ab22b9a HH |
3187 | .write_begin = ext4_write_begin, |
3188 | .write_end = ext4_ordered_write_end, | |
3189 | .bmap = ext4_bmap, | |
3190 | .invalidatepage = ext4_invalidatepage, | |
3191 | .releasepage = ext4_releasepage, | |
3192 | .direct_IO = ext4_direct_IO, | |
3193 | .migratepage = buffer_migrate_page, | |
3194 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 3195 | .error_remove_page = generic_error_remove_page, |
ac27a0ec DK |
3196 | }; |
3197 | ||
617ba13b | 3198 | static const struct address_space_operations ext4_writeback_aops = { |
8ab22b9a HH |
3199 | .readpage = ext4_readpage, |
3200 | .readpages = ext4_readpages, | |
43ce1d23 | 3201 | .writepage = ext4_writepage, |
8ab22b9a HH |
3202 | .write_begin = ext4_write_begin, |
3203 | .write_end = ext4_writeback_write_end, | |
3204 | .bmap = ext4_bmap, | |
3205 | .invalidatepage = ext4_invalidatepage, | |
3206 | .releasepage = ext4_releasepage, | |
3207 | .direct_IO = ext4_direct_IO, | |
3208 | .migratepage = buffer_migrate_page, | |
3209 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 3210 | .error_remove_page = generic_error_remove_page, |
ac27a0ec DK |
3211 | }; |
3212 | ||
617ba13b | 3213 | static const struct address_space_operations ext4_journalled_aops = { |
8ab22b9a HH |
3214 | .readpage = ext4_readpage, |
3215 | .readpages = ext4_readpages, | |
43ce1d23 | 3216 | .writepage = ext4_writepage, |
8ab22b9a HH |
3217 | .write_begin = ext4_write_begin, |
3218 | .write_end = ext4_journalled_write_end, | |
3219 | .set_page_dirty = ext4_journalled_set_page_dirty, | |
3220 | .bmap = ext4_bmap, | |
3221 | .invalidatepage = ext4_invalidatepage, | |
3222 | .releasepage = ext4_releasepage, | |
84ebd795 | 3223 | .direct_IO = ext4_direct_IO, |
8ab22b9a | 3224 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 3225 | .error_remove_page = generic_error_remove_page, |
ac27a0ec DK |
3226 | }; |
3227 | ||
64769240 | 3228 | static const struct address_space_operations ext4_da_aops = { |
8ab22b9a HH |
3229 | .readpage = ext4_readpage, |
3230 | .readpages = ext4_readpages, | |
43ce1d23 | 3231 | .writepage = ext4_writepage, |
8ab22b9a | 3232 | .writepages = ext4_da_writepages, |
8ab22b9a HH |
3233 | .write_begin = ext4_da_write_begin, |
3234 | .write_end = ext4_da_write_end, | |
3235 | .bmap = ext4_bmap, | |
3236 | .invalidatepage = ext4_da_invalidatepage, | |
3237 | .releasepage = ext4_releasepage, | |
3238 | .direct_IO = ext4_direct_IO, | |
3239 | .migratepage = buffer_migrate_page, | |
3240 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 3241 | .error_remove_page = generic_error_remove_page, |
64769240 AT |
3242 | }; |
3243 | ||
617ba13b | 3244 | void ext4_set_aops(struct inode *inode) |
ac27a0ec | 3245 | { |
3d2b1582 LC |
3246 | switch (ext4_inode_journal_mode(inode)) { |
3247 | case EXT4_INODE_ORDERED_DATA_MODE: | |
3248 | if (test_opt(inode->i_sb, DELALLOC)) | |
3249 | inode->i_mapping->a_ops = &ext4_da_aops; | |
3250 | else | |
3251 | inode->i_mapping->a_ops = &ext4_ordered_aops; | |
3252 | break; | |
3253 | case EXT4_INODE_WRITEBACK_DATA_MODE: | |
3254 | if (test_opt(inode->i_sb, DELALLOC)) | |
3255 | inode->i_mapping->a_ops = &ext4_da_aops; | |
3256 | else | |
3257 | inode->i_mapping->a_ops = &ext4_writeback_aops; | |
3258 | break; | |
3259 | case EXT4_INODE_JOURNAL_DATA_MODE: | |
617ba13b | 3260 | inode->i_mapping->a_ops = &ext4_journalled_aops; |
3d2b1582 LC |
3261 | break; |
3262 | default: | |
3263 | BUG(); | |
3264 | } | |
ac27a0ec DK |
3265 | } |
3266 | ||
4e96b2db AH |
3267 | |
3268 | /* | |
3269 | * ext4_discard_partial_page_buffers() | |
3270 | * Wrapper function for ext4_discard_partial_page_buffers_no_lock. | |
3271 | * This function finds and locks the page containing the offset | |
3272 | * "from" and passes it to ext4_discard_partial_page_buffers_no_lock. | |
3273 | * Calling functions that already have the page locked should call | |
3274 | * ext4_discard_partial_page_buffers_no_lock directly. | |
3275 | */ | |
3276 | int ext4_discard_partial_page_buffers(handle_t *handle, | |
3277 | struct address_space *mapping, loff_t from, | |
3278 | loff_t length, int flags) | |
3279 | { | |
3280 | struct inode *inode = mapping->host; | |
3281 | struct page *page; | |
3282 | int err = 0; | |
3283 | ||
3284 | page = find_or_create_page(mapping, from >> PAGE_CACHE_SHIFT, | |
3285 | mapping_gfp_mask(mapping) & ~__GFP_FS); | |
3286 | if (!page) | |
5129d05f | 3287 | return -ENOMEM; |
4e96b2db AH |
3288 | |
3289 | err = ext4_discard_partial_page_buffers_no_lock(handle, inode, page, | |
3290 | from, length, flags); | |
3291 | ||
3292 | unlock_page(page); | |
3293 | page_cache_release(page); | |
3294 | return err; | |
3295 | } | |
3296 | ||
3297 | /* | |
3298 | * ext4_discard_partial_page_buffers_no_lock() | |
3299 | * Zeros a page range of length 'length' starting from offset 'from'. | |
3300 | * Buffer heads that correspond to the block aligned regions of the | |
3301 | * zeroed range will be unmapped. Unblock aligned regions | |
3302 | * will have the corresponding buffer head mapped if needed so that | |
3303 | * that region of the page can be updated with the partial zero out. | |
3304 | * | |
3305 | * This function assumes that the page has already been locked. The | |
3306 | * The range to be discarded must be contained with in the given page. | |
3307 | * If the specified range exceeds the end of the page it will be shortened | |
3308 | * to the end of the page that corresponds to 'from'. This function is | |
3309 | * appropriate for updating a page and it buffer heads to be unmapped and | |
3310 | * zeroed for blocks that have been either released, or are going to be | |
3311 | * released. | |
3312 | * | |
3313 | * handle: The journal handle | |
3314 | * inode: The files inode | |
3315 | * page: A locked page that contains the offset "from" | |
4907cb7b | 3316 | * from: The starting byte offset (from the beginning of the file) |
4e96b2db AH |
3317 | * to begin discarding |
3318 | * len: The length of bytes to discard | |
3319 | * flags: Optional flags that may be used: | |
3320 | * | |
3321 | * EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED | |
3322 | * Only zero the regions of the page whose buffer heads | |
3323 | * have already been unmapped. This flag is appropriate | |
4907cb7b | 3324 | * for updating the contents of a page whose blocks may |
4e96b2db AH |
3325 | * have already been released, and we only want to zero |
3326 | * out the regions that correspond to those released blocks. | |
3327 | * | |
4907cb7b | 3328 | * Returns zero on success or negative on failure. |
4e96b2db | 3329 | */ |
5f163cc7 | 3330 | static int ext4_discard_partial_page_buffers_no_lock(handle_t *handle, |
4e96b2db AH |
3331 | struct inode *inode, struct page *page, loff_t from, |
3332 | loff_t length, int flags) | |
3333 | { | |
3334 | ext4_fsblk_t index = from >> PAGE_CACHE_SHIFT; | |
3335 | unsigned int offset = from & (PAGE_CACHE_SIZE-1); | |
3336 | unsigned int blocksize, max, pos; | |
4e96b2db AH |
3337 | ext4_lblk_t iblock; |
3338 | struct buffer_head *bh; | |
3339 | int err = 0; | |
3340 | ||
3341 | blocksize = inode->i_sb->s_blocksize; | |
3342 | max = PAGE_CACHE_SIZE - offset; | |
3343 | ||
3344 | if (index != page->index) | |
3345 | return -EINVAL; | |
3346 | ||
3347 | /* | |
3348 | * correct length if it does not fall between | |
3349 | * 'from' and the end of the page | |
3350 | */ | |
3351 | if (length > max || length < 0) | |
3352 | length = max; | |
3353 | ||
3354 | iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits); | |
3355 | ||
093e6e36 YY |
3356 | if (!page_has_buffers(page)) |
3357 | create_empty_buffers(page, blocksize, 0); | |
4e96b2db AH |
3358 | |
3359 | /* Find the buffer that contains "offset" */ | |
3360 | bh = page_buffers(page); | |
3361 | pos = blocksize; | |
3362 | while (offset >= pos) { | |
3363 | bh = bh->b_this_page; | |
3364 | iblock++; | |
3365 | pos += blocksize; | |
3366 | } | |
3367 | ||
3368 | pos = offset; | |
3369 | while (pos < offset + length) { | |
e260daf2 YY |
3370 | unsigned int end_of_block, range_to_discard; |
3371 | ||
4e96b2db AH |
3372 | err = 0; |
3373 | ||
3374 | /* The length of space left to zero and unmap */ | |
3375 | range_to_discard = offset + length - pos; | |
3376 | ||
3377 | /* The length of space until the end of the block */ | |
3378 | end_of_block = blocksize - (pos & (blocksize-1)); | |
3379 | ||
3380 | /* | |
3381 | * Do not unmap or zero past end of block | |
3382 | * for this buffer head | |
3383 | */ | |
3384 | if (range_to_discard > end_of_block) | |
3385 | range_to_discard = end_of_block; | |
3386 | ||
3387 | ||
3388 | /* | |
3389 | * Skip this buffer head if we are only zeroing unampped | |
3390 | * regions of the page | |
3391 | */ | |
3392 | if (flags & EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED && | |
3393 | buffer_mapped(bh)) | |
3394 | goto next; | |
3395 | ||
3396 | /* If the range is block aligned, unmap */ | |
3397 | if (range_to_discard == blocksize) { | |
3398 | clear_buffer_dirty(bh); | |
3399 | bh->b_bdev = NULL; | |
3400 | clear_buffer_mapped(bh); | |
3401 | clear_buffer_req(bh); | |
3402 | clear_buffer_new(bh); | |
3403 | clear_buffer_delay(bh); | |
3404 | clear_buffer_unwritten(bh); | |
3405 | clear_buffer_uptodate(bh); | |
3406 | zero_user(page, pos, range_to_discard); | |
3407 | BUFFER_TRACE(bh, "Buffer discarded"); | |
3408 | goto next; | |
3409 | } | |
3410 | ||
3411 | /* | |
3412 | * If this block is not completely contained in the range | |
3413 | * to be discarded, then it is not going to be released. Because | |
3414 | * we need to keep this block, we need to make sure this part | |
3415 | * of the page is uptodate before we modify it by writeing | |
3416 | * partial zeros on it. | |
3417 | */ | |
3418 | if (!buffer_mapped(bh)) { | |
3419 | /* | |
3420 | * Buffer head must be mapped before we can read | |
3421 | * from the block | |
3422 | */ | |
3423 | BUFFER_TRACE(bh, "unmapped"); | |
3424 | ext4_get_block(inode, iblock, bh, 0); | |
3425 | /* unmapped? It's a hole - nothing to do */ | |
3426 | if (!buffer_mapped(bh)) { | |
3427 | BUFFER_TRACE(bh, "still unmapped"); | |
3428 | goto next; | |
3429 | } | |
3430 | } | |
3431 | ||
3432 | /* Ok, it's mapped. Make sure it's up-to-date */ | |
3433 | if (PageUptodate(page)) | |
3434 | set_buffer_uptodate(bh); | |
3435 | ||
3436 | if (!buffer_uptodate(bh)) { | |
3437 | err = -EIO; | |
3438 | ll_rw_block(READ, 1, &bh); | |
3439 | wait_on_buffer(bh); | |
3440 | /* Uhhuh. Read error. Complain and punt.*/ | |
3441 | if (!buffer_uptodate(bh)) | |
3442 | goto next; | |
3443 | } | |
3444 | ||
3445 | if (ext4_should_journal_data(inode)) { | |
3446 | BUFFER_TRACE(bh, "get write access"); | |
3447 | err = ext4_journal_get_write_access(handle, bh); | |
3448 | if (err) | |
3449 | goto next; | |
3450 | } | |
3451 | ||
3452 | zero_user(page, pos, range_to_discard); | |
3453 | ||
3454 | err = 0; | |
3455 | if (ext4_should_journal_data(inode)) { | |
3456 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
decbd919 | 3457 | } else |
4e96b2db | 3458 | mark_buffer_dirty(bh); |
4e96b2db AH |
3459 | |
3460 | BUFFER_TRACE(bh, "Partial buffer zeroed"); | |
3461 | next: | |
3462 | bh = bh->b_this_page; | |
3463 | iblock++; | |
3464 | pos += range_to_discard; | |
3465 | } | |
3466 | ||
3467 | return err; | |
3468 | } | |
3469 | ||
91ef4caf DG |
3470 | int ext4_can_truncate(struct inode *inode) |
3471 | { | |
91ef4caf DG |
3472 | if (S_ISREG(inode->i_mode)) |
3473 | return 1; | |
3474 | if (S_ISDIR(inode->i_mode)) | |
3475 | return 1; | |
3476 | if (S_ISLNK(inode->i_mode)) | |
3477 | return !ext4_inode_is_fast_symlink(inode); | |
3478 | return 0; | |
3479 | } | |
3480 | ||
a4bb6b64 AH |
3481 | /* |
3482 | * ext4_punch_hole: punches a hole in a file by releaseing the blocks | |
3483 | * associated with the given offset and length | |
3484 | * | |
3485 | * @inode: File inode | |
3486 | * @offset: The offset where the hole will begin | |
3487 | * @len: The length of the hole | |
3488 | * | |
4907cb7b | 3489 | * Returns: 0 on success or negative on failure |
a4bb6b64 AH |
3490 | */ |
3491 | ||
3492 | int ext4_punch_hole(struct file *file, loff_t offset, loff_t length) | |
3493 | { | |
3494 | struct inode *inode = file->f_path.dentry->d_inode; | |
3495 | if (!S_ISREG(inode->i_mode)) | |
73355192 | 3496 | return -EOPNOTSUPP; |
a4bb6b64 AH |
3497 | |
3498 | if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { | |
3499 | /* TODO: Add support for non extent hole punching */ | |
73355192 | 3500 | return -EOPNOTSUPP; |
a4bb6b64 AH |
3501 | } |
3502 | ||
bab08ab9 TT |
3503 | if (EXT4_SB(inode->i_sb)->s_cluster_ratio > 1) { |
3504 | /* TODO: Add support for bigalloc file systems */ | |
73355192 | 3505 | return -EOPNOTSUPP; |
bab08ab9 TT |
3506 | } |
3507 | ||
a4bb6b64 AH |
3508 | return ext4_ext_punch_hole(file, offset, length); |
3509 | } | |
3510 | ||
ac27a0ec | 3511 | /* |
617ba13b | 3512 | * ext4_truncate() |
ac27a0ec | 3513 | * |
617ba13b MC |
3514 | * We block out ext4_get_block() block instantiations across the entire |
3515 | * transaction, and VFS/VM ensures that ext4_truncate() cannot run | |
ac27a0ec DK |
3516 | * simultaneously on behalf of the same inode. |
3517 | * | |
42b2aa86 | 3518 | * As we work through the truncate and commit bits of it to the journal there |
ac27a0ec DK |
3519 | * is one core, guiding principle: the file's tree must always be consistent on |
3520 | * disk. We must be able to restart the truncate after a crash. | |
3521 | * | |
3522 | * The file's tree may be transiently inconsistent in memory (although it | |
3523 | * probably isn't), but whenever we close off and commit a journal transaction, | |
3524 | * the contents of (the filesystem + the journal) must be consistent and | |
3525 | * restartable. It's pretty simple, really: bottom up, right to left (although | |
3526 | * left-to-right works OK too). | |
3527 | * | |
3528 | * Note that at recovery time, journal replay occurs *before* the restart of | |
3529 | * truncate against the orphan inode list. | |
3530 | * | |
3531 | * The committed inode has the new, desired i_size (which is the same as | |
617ba13b | 3532 | * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see |
ac27a0ec | 3533 | * that this inode's truncate did not complete and it will again call |
617ba13b MC |
3534 | * ext4_truncate() to have another go. So there will be instantiated blocks |
3535 | * to the right of the truncation point in a crashed ext4 filesystem. But | |
ac27a0ec | 3536 | * that's fine - as long as they are linked from the inode, the post-crash |
617ba13b | 3537 | * ext4_truncate() run will find them and release them. |
ac27a0ec | 3538 | */ |
617ba13b | 3539 | void ext4_truncate(struct inode *inode) |
ac27a0ec | 3540 | { |
0562e0ba JZ |
3541 | trace_ext4_truncate_enter(inode); |
3542 | ||
91ef4caf | 3543 | if (!ext4_can_truncate(inode)) |
ac27a0ec DK |
3544 | return; |
3545 | ||
12e9b892 | 3546 | ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS); |
c8d46e41 | 3547 | |
5534fb5b | 3548 | if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC)) |
19f5fb7a | 3549 | ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE); |
7d8f9f7d | 3550 | |
ff9893dc | 3551 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
cf108bca | 3552 | ext4_ext_truncate(inode); |
ff9893dc AG |
3553 | else |
3554 | ext4_ind_truncate(inode); | |
ac27a0ec | 3555 | |
0562e0ba | 3556 | trace_ext4_truncate_exit(inode); |
ac27a0ec DK |
3557 | } |
3558 | ||
ac27a0ec | 3559 | /* |
617ba13b | 3560 | * ext4_get_inode_loc returns with an extra refcount against the inode's |
ac27a0ec DK |
3561 | * underlying buffer_head on success. If 'in_mem' is true, we have all |
3562 | * data in memory that is needed to recreate the on-disk version of this | |
3563 | * inode. | |
3564 | */ | |
617ba13b MC |
3565 | static int __ext4_get_inode_loc(struct inode *inode, |
3566 | struct ext4_iloc *iloc, int in_mem) | |
ac27a0ec | 3567 | { |
240799cd TT |
3568 | struct ext4_group_desc *gdp; |
3569 | struct buffer_head *bh; | |
3570 | struct super_block *sb = inode->i_sb; | |
3571 | ext4_fsblk_t block; | |
3572 | int inodes_per_block, inode_offset; | |
3573 | ||
3a06d778 | 3574 | iloc->bh = NULL; |
240799cd TT |
3575 | if (!ext4_valid_inum(sb, inode->i_ino)) |
3576 | return -EIO; | |
ac27a0ec | 3577 | |
240799cd TT |
3578 | iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb); |
3579 | gdp = ext4_get_group_desc(sb, iloc->block_group, NULL); | |
3580 | if (!gdp) | |
ac27a0ec DK |
3581 | return -EIO; |
3582 | ||
240799cd TT |
3583 | /* |
3584 | * Figure out the offset within the block group inode table | |
3585 | */ | |
00d09882 | 3586 | inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; |
240799cd TT |
3587 | inode_offset = ((inode->i_ino - 1) % |
3588 | EXT4_INODES_PER_GROUP(sb)); | |
3589 | block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block); | |
3590 | iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb); | |
3591 | ||
3592 | bh = sb_getblk(sb, block); | |
ac27a0ec | 3593 | if (!bh) { |
c398eda0 TT |
3594 | EXT4_ERROR_INODE_BLOCK(inode, block, |
3595 | "unable to read itable block"); | |
ac27a0ec DK |
3596 | return -EIO; |
3597 | } | |
3598 | if (!buffer_uptodate(bh)) { | |
3599 | lock_buffer(bh); | |
9c83a923 HK |
3600 | |
3601 | /* | |
3602 | * If the buffer has the write error flag, we have failed | |
3603 | * to write out another inode in the same block. In this | |
3604 | * case, we don't have to read the block because we may | |
3605 | * read the old inode data successfully. | |
3606 | */ | |
3607 | if (buffer_write_io_error(bh) && !buffer_uptodate(bh)) | |
3608 | set_buffer_uptodate(bh); | |
3609 | ||
ac27a0ec DK |
3610 | if (buffer_uptodate(bh)) { |
3611 | /* someone brought it uptodate while we waited */ | |
3612 | unlock_buffer(bh); | |
3613 | goto has_buffer; | |
3614 | } | |
3615 | ||
3616 | /* | |
3617 | * If we have all information of the inode in memory and this | |
3618 | * is the only valid inode in the block, we need not read the | |
3619 | * block. | |
3620 | */ | |
3621 | if (in_mem) { | |
3622 | struct buffer_head *bitmap_bh; | |
240799cd | 3623 | int i, start; |
ac27a0ec | 3624 | |
240799cd | 3625 | start = inode_offset & ~(inodes_per_block - 1); |
ac27a0ec | 3626 | |
240799cd TT |
3627 | /* Is the inode bitmap in cache? */ |
3628 | bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp)); | |
ac27a0ec DK |
3629 | if (!bitmap_bh) |
3630 | goto make_io; | |
3631 | ||
3632 | /* | |
3633 | * If the inode bitmap isn't in cache then the | |
3634 | * optimisation may end up performing two reads instead | |
3635 | * of one, so skip it. | |
3636 | */ | |
3637 | if (!buffer_uptodate(bitmap_bh)) { | |
3638 | brelse(bitmap_bh); | |
3639 | goto make_io; | |
3640 | } | |
240799cd | 3641 | for (i = start; i < start + inodes_per_block; i++) { |
ac27a0ec DK |
3642 | if (i == inode_offset) |
3643 | continue; | |
617ba13b | 3644 | if (ext4_test_bit(i, bitmap_bh->b_data)) |
ac27a0ec DK |
3645 | break; |
3646 | } | |
3647 | brelse(bitmap_bh); | |
240799cd | 3648 | if (i == start + inodes_per_block) { |
ac27a0ec DK |
3649 | /* all other inodes are free, so skip I/O */ |
3650 | memset(bh->b_data, 0, bh->b_size); | |
3651 | set_buffer_uptodate(bh); | |
3652 | unlock_buffer(bh); | |
3653 | goto has_buffer; | |
3654 | } | |
3655 | } | |
3656 | ||
3657 | make_io: | |
240799cd TT |
3658 | /* |
3659 | * If we need to do any I/O, try to pre-readahead extra | |
3660 | * blocks from the inode table. | |
3661 | */ | |
3662 | if (EXT4_SB(sb)->s_inode_readahead_blks) { | |
3663 | ext4_fsblk_t b, end, table; | |
3664 | unsigned num; | |
3665 | ||
3666 | table = ext4_inode_table(sb, gdp); | |
b713a5ec | 3667 | /* s_inode_readahead_blks is always a power of 2 */ |
240799cd TT |
3668 | b = block & ~(EXT4_SB(sb)->s_inode_readahead_blks-1); |
3669 | if (table > b) | |
3670 | b = table; | |
3671 | end = b + EXT4_SB(sb)->s_inode_readahead_blks; | |
3672 | num = EXT4_INODES_PER_GROUP(sb); | |
feb0ab32 | 3673 | if (ext4_has_group_desc_csum(sb)) |
560671a0 | 3674 | num -= ext4_itable_unused_count(sb, gdp); |
240799cd TT |
3675 | table += num / inodes_per_block; |
3676 | if (end > table) | |
3677 | end = table; | |
3678 | while (b <= end) | |
3679 | sb_breadahead(sb, b++); | |
3680 | } | |
3681 | ||
ac27a0ec DK |
3682 | /* |
3683 | * There are other valid inodes in the buffer, this inode | |
3684 | * has in-inode xattrs, or we don't have this inode in memory. | |
3685 | * Read the block from disk. | |
3686 | */ | |
0562e0ba | 3687 | trace_ext4_load_inode(inode); |
ac27a0ec DK |
3688 | get_bh(bh); |
3689 | bh->b_end_io = end_buffer_read_sync; | |
65299a3b | 3690 | submit_bh(READ | REQ_META | REQ_PRIO, bh); |
ac27a0ec DK |
3691 | wait_on_buffer(bh); |
3692 | if (!buffer_uptodate(bh)) { | |
c398eda0 TT |
3693 | EXT4_ERROR_INODE_BLOCK(inode, block, |
3694 | "unable to read itable block"); | |
ac27a0ec DK |
3695 | brelse(bh); |
3696 | return -EIO; | |
3697 | } | |
3698 | } | |
3699 | has_buffer: | |
3700 | iloc->bh = bh; | |
3701 | return 0; | |
3702 | } | |
3703 | ||
617ba13b | 3704 | int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec DK |
3705 | { |
3706 | /* We have all inode data except xattrs in memory here. */ | |
617ba13b | 3707 | return __ext4_get_inode_loc(inode, iloc, |
19f5fb7a | 3708 | !ext4_test_inode_state(inode, EXT4_STATE_XATTR)); |
ac27a0ec DK |
3709 | } |
3710 | ||
617ba13b | 3711 | void ext4_set_inode_flags(struct inode *inode) |
ac27a0ec | 3712 | { |
617ba13b | 3713 | unsigned int flags = EXT4_I(inode)->i_flags; |
ac27a0ec DK |
3714 | |
3715 | inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC); | |
617ba13b | 3716 | if (flags & EXT4_SYNC_FL) |
ac27a0ec | 3717 | inode->i_flags |= S_SYNC; |
617ba13b | 3718 | if (flags & EXT4_APPEND_FL) |
ac27a0ec | 3719 | inode->i_flags |= S_APPEND; |
617ba13b | 3720 | if (flags & EXT4_IMMUTABLE_FL) |
ac27a0ec | 3721 | inode->i_flags |= S_IMMUTABLE; |
617ba13b | 3722 | if (flags & EXT4_NOATIME_FL) |
ac27a0ec | 3723 | inode->i_flags |= S_NOATIME; |
617ba13b | 3724 | if (flags & EXT4_DIRSYNC_FL) |
ac27a0ec DK |
3725 | inode->i_flags |= S_DIRSYNC; |
3726 | } | |
3727 | ||
ff9ddf7e JK |
3728 | /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */ |
3729 | void ext4_get_inode_flags(struct ext4_inode_info *ei) | |
3730 | { | |
84a8dce2 DM |
3731 | unsigned int vfs_fl; |
3732 | unsigned long old_fl, new_fl; | |
3733 | ||
3734 | do { | |
3735 | vfs_fl = ei->vfs_inode.i_flags; | |
3736 | old_fl = ei->i_flags; | |
3737 | new_fl = old_fl & ~(EXT4_SYNC_FL|EXT4_APPEND_FL| | |
3738 | EXT4_IMMUTABLE_FL|EXT4_NOATIME_FL| | |
3739 | EXT4_DIRSYNC_FL); | |
3740 | if (vfs_fl & S_SYNC) | |
3741 | new_fl |= EXT4_SYNC_FL; | |
3742 | if (vfs_fl & S_APPEND) | |
3743 | new_fl |= EXT4_APPEND_FL; | |
3744 | if (vfs_fl & S_IMMUTABLE) | |
3745 | new_fl |= EXT4_IMMUTABLE_FL; | |
3746 | if (vfs_fl & S_NOATIME) | |
3747 | new_fl |= EXT4_NOATIME_FL; | |
3748 | if (vfs_fl & S_DIRSYNC) | |
3749 | new_fl |= EXT4_DIRSYNC_FL; | |
3750 | } while (cmpxchg(&ei->i_flags, old_fl, new_fl) != old_fl); | |
ff9ddf7e | 3751 | } |
de9a55b8 | 3752 | |
0fc1b451 | 3753 | static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode, |
de9a55b8 | 3754 | struct ext4_inode_info *ei) |
0fc1b451 AK |
3755 | { |
3756 | blkcnt_t i_blocks ; | |
8180a562 AK |
3757 | struct inode *inode = &(ei->vfs_inode); |
3758 | struct super_block *sb = inode->i_sb; | |
0fc1b451 AK |
3759 | |
3760 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, | |
3761 | EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) { | |
3762 | /* we are using combined 48 bit field */ | |
3763 | i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 | | |
3764 | le32_to_cpu(raw_inode->i_blocks_lo); | |
07a03824 | 3765 | if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) { |
8180a562 AK |
3766 | /* i_blocks represent file system block size */ |
3767 | return i_blocks << (inode->i_blkbits - 9); | |
3768 | } else { | |
3769 | return i_blocks; | |
3770 | } | |
0fc1b451 AK |
3771 | } else { |
3772 | return le32_to_cpu(raw_inode->i_blocks_lo); | |
3773 | } | |
3774 | } | |
ff9ddf7e | 3775 | |
1d1fe1ee | 3776 | struct inode *ext4_iget(struct super_block *sb, unsigned long ino) |
ac27a0ec | 3777 | { |
617ba13b MC |
3778 | struct ext4_iloc iloc; |
3779 | struct ext4_inode *raw_inode; | |
1d1fe1ee | 3780 | struct ext4_inode_info *ei; |
1d1fe1ee | 3781 | struct inode *inode; |
b436b9be | 3782 | journal_t *journal = EXT4_SB(sb)->s_journal; |
1d1fe1ee | 3783 | long ret; |
ac27a0ec | 3784 | int block; |
08cefc7a EB |
3785 | uid_t i_uid; |
3786 | gid_t i_gid; | |
ac27a0ec | 3787 | |
1d1fe1ee DH |
3788 | inode = iget_locked(sb, ino); |
3789 | if (!inode) | |
3790 | return ERR_PTR(-ENOMEM); | |
3791 | if (!(inode->i_state & I_NEW)) | |
3792 | return inode; | |
3793 | ||
3794 | ei = EXT4_I(inode); | |
7dc57615 | 3795 | iloc.bh = NULL; |
ac27a0ec | 3796 | |
1d1fe1ee DH |
3797 | ret = __ext4_get_inode_loc(inode, &iloc, 0); |
3798 | if (ret < 0) | |
ac27a0ec | 3799 | goto bad_inode; |
617ba13b | 3800 | raw_inode = ext4_raw_inode(&iloc); |
814525f4 DW |
3801 | |
3802 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { | |
3803 | ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize); | |
3804 | if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > | |
3805 | EXT4_INODE_SIZE(inode->i_sb)) { | |
3806 | EXT4_ERROR_INODE(inode, "bad extra_isize (%u != %u)", | |
3807 | EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize, | |
3808 | EXT4_INODE_SIZE(inode->i_sb)); | |
3809 | ret = -EIO; | |
3810 | goto bad_inode; | |
3811 | } | |
3812 | } else | |
3813 | ei->i_extra_isize = 0; | |
3814 | ||
3815 | /* Precompute checksum seed for inode metadata */ | |
3816 | if (EXT4_HAS_RO_COMPAT_FEATURE(sb, | |
3817 | EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) { | |
3818 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
3819 | __u32 csum; | |
3820 | __le32 inum = cpu_to_le32(inode->i_ino); | |
3821 | __le32 gen = raw_inode->i_generation; | |
3822 | csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum, | |
3823 | sizeof(inum)); | |
3824 | ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen, | |
3825 | sizeof(gen)); | |
3826 | } | |
3827 | ||
3828 | if (!ext4_inode_csum_verify(inode, raw_inode, ei)) { | |
3829 | EXT4_ERROR_INODE(inode, "checksum invalid"); | |
3830 | ret = -EIO; | |
3831 | goto bad_inode; | |
3832 | } | |
3833 | ||
ac27a0ec | 3834 | inode->i_mode = le16_to_cpu(raw_inode->i_mode); |
08cefc7a EB |
3835 | i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low); |
3836 | i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low); | |
af5bc92d | 3837 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
08cefc7a EB |
3838 | i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16; |
3839 | i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16; | |
ac27a0ec | 3840 | } |
08cefc7a EB |
3841 | i_uid_write(inode, i_uid); |
3842 | i_gid_write(inode, i_gid); | |
bfe86848 | 3843 | set_nlink(inode, le16_to_cpu(raw_inode->i_links_count)); |
ac27a0ec | 3844 | |
353eb83c | 3845 | ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */ |
ac27a0ec DK |
3846 | ei->i_dir_start_lookup = 0; |
3847 | ei->i_dtime = le32_to_cpu(raw_inode->i_dtime); | |
3848 | /* We now have enough fields to check if the inode was active or not. | |
3849 | * This is needed because nfsd might try to access dead inodes | |
3850 | * the test is that same one that e2fsck uses | |
3851 | * NeilBrown 1999oct15 | |
3852 | */ | |
3853 | if (inode->i_nlink == 0) { | |
3854 | if (inode->i_mode == 0 || | |
617ba13b | 3855 | !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) { |
ac27a0ec | 3856 | /* this inode is deleted */ |
1d1fe1ee | 3857 | ret = -ESTALE; |
ac27a0ec DK |
3858 | goto bad_inode; |
3859 | } | |
3860 | /* The only unlinked inodes we let through here have | |
3861 | * valid i_mode and are being read by the orphan | |
3862 | * recovery code: that's fine, we're about to complete | |
3863 | * the process of deleting those. */ | |
3864 | } | |
ac27a0ec | 3865 | ei->i_flags = le32_to_cpu(raw_inode->i_flags); |
0fc1b451 | 3866 | inode->i_blocks = ext4_inode_blocks(raw_inode, ei); |
7973c0c1 | 3867 | ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo); |
a9e81742 | 3868 | if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) |
a1ddeb7e BP |
3869 | ei->i_file_acl |= |
3870 | ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32; | |
a48380f7 | 3871 | inode->i_size = ext4_isize(raw_inode); |
ac27a0ec | 3872 | ei->i_disksize = inode->i_size; |
a9e7f447 DM |
3873 | #ifdef CONFIG_QUOTA |
3874 | ei->i_reserved_quota = 0; | |
3875 | #endif | |
ac27a0ec DK |
3876 | inode->i_generation = le32_to_cpu(raw_inode->i_generation); |
3877 | ei->i_block_group = iloc.block_group; | |
a4912123 | 3878 | ei->i_last_alloc_group = ~0; |
ac27a0ec DK |
3879 | /* |
3880 | * NOTE! The in-memory inode i_data array is in little-endian order | |
3881 | * even on big-endian machines: we do NOT byteswap the block numbers! | |
3882 | */ | |
617ba13b | 3883 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
ac27a0ec DK |
3884 | ei->i_data[block] = raw_inode->i_block[block]; |
3885 | INIT_LIST_HEAD(&ei->i_orphan); | |
3886 | ||
b436b9be JK |
3887 | /* |
3888 | * Set transaction id's of transactions that have to be committed | |
3889 | * to finish f[data]sync. We set them to currently running transaction | |
3890 | * as we cannot be sure that the inode or some of its metadata isn't | |
3891 | * part of the transaction - the inode could have been reclaimed and | |
3892 | * now it is reread from disk. | |
3893 | */ | |
3894 | if (journal) { | |
3895 | transaction_t *transaction; | |
3896 | tid_t tid; | |
3897 | ||
a931da6a | 3898 | read_lock(&journal->j_state_lock); |
b436b9be JK |
3899 | if (journal->j_running_transaction) |
3900 | transaction = journal->j_running_transaction; | |
3901 | else | |
3902 | transaction = journal->j_committing_transaction; | |
3903 | if (transaction) | |
3904 | tid = transaction->t_tid; | |
3905 | else | |
3906 | tid = journal->j_commit_sequence; | |
a931da6a | 3907 | read_unlock(&journal->j_state_lock); |
b436b9be JK |
3908 | ei->i_sync_tid = tid; |
3909 | ei->i_datasync_tid = tid; | |
3910 | } | |
3911 | ||
0040d987 | 3912 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
ac27a0ec DK |
3913 | if (ei->i_extra_isize == 0) { |
3914 | /* The extra space is currently unused. Use it. */ | |
617ba13b MC |
3915 | ei->i_extra_isize = sizeof(struct ext4_inode) - |
3916 | EXT4_GOOD_OLD_INODE_SIZE; | |
ac27a0ec DK |
3917 | } else { |
3918 | __le32 *magic = (void *)raw_inode + | |
617ba13b | 3919 | EXT4_GOOD_OLD_INODE_SIZE + |
ac27a0ec | 3920 | ei->i_extra_isize; |
617ba13b | 3921 | if (*magic == cpu_to_le32(EXT4_XATTR_MAGIC)) |
19f5fb7a | 3922 | ext4_set_inode_state(inode, EXT4_STATE_XATTR); |
ac27a0ec | 3923 | } |
814525f4 | 3924 | } |
ac27a0ec | 3925 | |
ef7f3835 KS |
3926 | EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode); |
3927 | EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode); | |
3928 | EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode); | |
3929 | EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode); | |
3930 | ||
25ec56b5 JNC |
3931 | inode->i_version = le32_to_cpu(raw_inode->i_disk_version); |
3932 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { | |
3933 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
3934 | inode->i_version |= | |
3935 | (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32; | |
3936 | } | |
3937 | ||
c4b5a614 | 3938 | ret = 0; |
485c26ec | 3939 | if (ei->i_file_acl && |
1032988c | 3940 | !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) { |
24676da4 TT |
3941 | EXT4_ERROR_INODE(inode, "bad extended attribute block %llu", |
3942 | ei->i_file_acl); | |
485c26ec TT |
3943 | ret = -EIO; |
3944 | goto bad_inode; | |
07a03824 | 3945 | } else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
c4b5a614 TT |
3946 | if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || |
3947 | (S_ISLNK(inode->i_mode) && | |
3948 | !ext4_inode_is_fast_symlink(inode))) | |
3949 | /* Validate extent which is part of inode */ | |
3950 | ret = ext4_ext_check_inode(inode); | |
de9a55b8 | 3951 | } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || |
fe2c8191 TN |
3952 | (S_ISLNK(inode->i_mode) && |
3953 | !ext4_inode_is_fast_symlink(inode))) { | |
de9a55b8 | 3954 | /* Validate block references which are part of inode */ |
1f7d1e77 | 3955 | ret = ext4_ind_check_inode(inode); |
fe2c8191 | 3956 | } |
567f3e9a | 3957 | if (ret) |
de9a55b8 | 3958 | goto bad_inode; |
7a262f7c | 3959 | |
ac27a0ec | 3960 | if (S_ISREG(inode->i_mode)) { |
617ba13b MC |
3961 | inode->i_op = &ext4_file_inode_operations; |
3962 | inode->i_fop = &ext4_file_operations; | |
3963 | ext4_set_aops(inode); | |
ac27a0ec | 3964 | } else if (S_ISDIR(inode->i_mode)) { |
617ba13b MC |
3965 | inode->i_op = &ext4_dir_inode_operations; |
3966 | inode->i_fop = &ext4_dir_operations; | |
ac27a0ec | 3967 | } else if (S_ISLNK(inode->i_mode)) { |
e83c1397 | 3968 | if (ext4_inode_is_fast_symlink(inode)) { |
617ba13b | 3969 | inode->i_op = &ext4_fast_symlink_inode_operations; |
e83c1397 DG |
3970 | nd_terminate_link(ei->i_data, inode->i_size, |
3971 | sizeof(ei->i_data) - 1); | |
3972 | } else { | |
617ba13b MC |
3973 | inode->i_op = &ext4_symlink_inode_operations; |
3974 | ext4_set_aops(inode); | |
ac27a0ec | 3975 | } |
563bdd61 TT |
3976 | } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) || |
3977 | S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) { | |
617ba13b | 3978 | inode->i_op = &ext4_special_inode_operations; |
ac27a0ec DK |
3979 | if (raw_inode->i_block[0]) |
3980 | init_special_inode(inode, inode->i_mode, | |
3981 | old_decode_dev(le32_to_cpu(raw_inode->i_block[0]))); | |
3982 | else | |
3983 | init_special_inode(inode, inode->i_mode, | |
3984 | new_decode_dev(le32_to_cpu(raw_inode->i_block[1]))); | |
563bdd61 | 3985 | } else { |
563bdd61 | 3986 | ret = -EIO; |
24676da4 | 3987 | EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode); |
563bdd61 | 3988 | goto bad_inode; |
ac27a0ec | 3989 | } |
af5bc92d | 3990 | brelse(iloc.bh); |
617ba13b | 3991 | ext4_set_inode_flags(inode); |
1d1fe1ee DH |
3992 | unlock_new_inode(inode); |
3993 | return inode; | |
ac27a0ec DK |
3994 | |
3995 | bad_inode: | |
567f3e9a | 3996 | brelse(iloc.bh); |
1d1fe1ee DH |
3997 | iget_failed(inode); |
3998 | return ERR_PTR(ret); | |
ac27a0ec DK |
3999 | } |
4000 | ||
0fc1b451 AK |
4001 | static int ext4_inode_blocks_set(handle_t *handle, |
4002 | struct ext4_inode *raw_inode, | |
4003 | struct ext4_inode_info *ei) | |
4004 | { | |
4005 | struct inode *inode = &(ei->vfs_inode); | |
4006 | u64 i_blocks = inode->i_blocks; | |
4007 | struct super_block *sb = inode->i_sb; | |
0fc1b451 AK |
4008 | |
4009 | if (i_blocks <= ~0U) { | |
4010 | /* | |
4907cb7b | 4011 | * i_blocks can be represented in a 32 bit variable |
0fc1b451 AK |
4012 | * as multiple of 512 bytes |
4013 | */ | |
8180a562 | 4014 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 4015 | raw_inode->i_blocks_high = 0; |
84a8dce2 | 4016 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
f287a1a5 TT |
4017 | return 0; |
4018 | } | |
4019 | if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) | |
4020 | return -EFBIG; | |
4021 | ||
4022 | if (i_blocks <= 0xffffffffffffULL) { | |
0fc1b451 AK |
4023 | /* |
4024 | * i_blocks can be represented in a 48 bit variable | |
4025 | * as multiple of 512 bytes | |
4026 | */ | |
8180a562 | 4027 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 4028 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); |
84a8dce2 | 4029 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
0fc1b451 | 4030 | } else { |
84a8dce2 | 4031 | ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
8180a562 AK |
4032 | /* i_block is stored in file system block size */ |
4033 | i_blocks = i_blocks >> (inode->i_blkbits - 9); | |
4034 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); | |
4035 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); | |
0fc1b451 | 4036 | } |
f287a1a5 | 4037 | return 0; |
0fc1b451 AK |
4038 | } |
4039 | ||
ac27a0ec DK |
4040 | /* |
4041 | * Post the struct inode info into an on-disk inode location in the | |
4042 | * buffer-cache. This gobbles the caller's reference to the | |
4043 | * buffer_head in the inode location struct. | |
4044 | * | |
4045 | * The caller must have write access to iloc->bh. | |
4046 | */ | |
617ba13b | 4047 | static int ext4_do_update_inode(handle_t *handle, |
ac27a0ec | 4048 | struct inode *inode, |
830156c7 | 4049 | struct ext4_iloc *iloc) |
ac27a0ec | 4050 | { |
617ba13b MC |
4051 | struct ext4_inode *raw_inode = ext4_raw_inode(iloc); |
4052 | struct ext4_inode_info *ei = EXT4_I(inode); | |
ac27a0ec DK |
4053 | struct buffer_head *bh = iloc->bh; |
4054 | int err = 0, rc, block; | |
b71fc079 | 4055 | int need_datasync = 0; |
08cefc7a EB |
4056 | uid_t i_uid; |
4057 | gid_t i_gid; | |
ac27a0ec DK |
4058 | |
4059 | /* For fields not not tracking in the in-memory inode, | |
4060 | * initialise them to zero for new inodes. */ | |
19f5fb7a | 4061 | if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) |
617ba13b | 4062 | memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size); |
ac27a0ec | 4063 | |
ff9ddf7e | 4064 | ext4_get_inode_flags(ei); |
ac27a0ec | 4065 | raw_inode->i_mode = cpu_to_le16(inode->i_mode); |
08cefc7a EB |
4066 | i_uid = i_uid_read(inode); |
4067 | i_gid = i_gid_read(inode); | |
af5bc92d | 4068 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
08cefc7a EB |
4069 | raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid)); |
4070 | raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid)); | |
ac27a0ec DK |
4071 | /* |
4072 | * Fix up interoperability with old kernels. Otherwise, old inodes get | |
4073 | * re-used with the upper 16 bits of the uid/gid intact | |
4074 | */ | |
af5bc92d | 4075 | if (!ei->i_dtime) { |
ac27a0ec | 4076 | raw_inode->i_uid_high = |
08cefc7a | 4077 | cpu_to_le16(high_16_bits(i_uid)); |
ac27a0ec | 4078 | raw_inode->i_gid_high = |
08cefc7a | 4079 | cpu_to_le16(high_16_bits(i_gid)); |
ac27a0ec DK |
4080 | } else { |
4081 | raw_inode->i_uid_high = 0; | |
4082 | raw_inode->i_gid_high = 0; | |
4083 | } | |
4084 | } else { | |
08cefc7a EB |
4085 | raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid)); |
4086 | raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid)); | |
ac27a0ec DK |
4087 | raw_inode->i_uid_high = 0; |
4088 | raw_inode->i_gid_high = 0; | |
4089 | } | |
4090 | raw_inode->i_links_count = cpu_to_le16(inode->i_nlink); | |
ef7f3835 KS |
4091 | |
4092 | EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode); | |
4093 | EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode); | |
4094 | EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode); | |
4095 | EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode); | |
4096 | ||
0fc1b451 AK |
4097 | if (ext4_inode_blocks_set(handle, raw_inode, ei)) |
4098 | goto out_brelse; | |
ac27a0ec | 4099 | raw_inode->i_dtime = cpu_to_le32(ei->i_dtime); |
353eb83c | 4100 | raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF); |
9b8f1f01 MC |
4101 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != |
4102 | cpu_to_le32(EXT4_OS_HURD)) | |
a1ddeb7e BP |
4103 | raw_inode->i_file_acl_high = |
4104 | cpu_to_le16(ei->i_file_acl >> 32); | |
7973c0c1 | 4105 | raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl); |
b71fc079 JK |
4106 | if (ei->i_disksize != ext4_isize(raw_inode)) { |
4107 | ext4_isize_set(raw_inode, ei->i_disksize); | |
4108 | need_datasync = 1; | |
4109 | } | |
a48380f7 AK |
4110 | if (ei->i_disksize > 0x7fffffffULL) { |
4111 | struct super_block *sb = inode->i_sb; | |
4112 | if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, | |
4113 | EXT4_FEATURE_RO_COMPAT_LARGE_FILE) || | |
4114 | EXT4_SB(sb)->s_es->s_rev_level == | |
4115 | cpu_to_le32(EXT4_GOOD_OLD_REV)) { | |
4116 | /* If this is the first large file | |
4117 | * created, add a flag to the superblock. | |
4118 | */ | |
4119 | err = ext4_journal_get_write_access(handle, | |
4120 | EXT4_SB(sb)->s_sbh); | |
4121 | if (err) | |
4122 | goto out_brelse; | |
4123 | ext4_update_dynamic_rev(sb); | |
4124 | EXT4_SET_RO_COMPAT_FEATURE(sb, | |
617ba13b | 4125 | EXT4_FEATURE_RO_COMPAT_LARGE_FILE); |
0390131b | 4126 | ext4_handle_sync(handle); |
b50924c2 | 4127 | err = ext4_handle_dirty_super(handle, sb); |
ac27a0ec DK |
4128 | } |
4129 | } | |
4130 | raw_inode->i_generation = cpu_to_le32(inode->i_generation); | |
4131 | if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) { | |
4132 | if (old_valid_dev(inode->i_rdev)) { | |
4133 | raw_inode->i_block[0] = | |
4134 | cpu_to_le32(old_encode_dev(inode->i_rdev)); | |
4135 | raw_inode->i_block[1] = 0; | |
4136 | } else { | |
4137 | raw_inode->i_block[0] = 0; | |
4138 | raw_inode->i_block[1] = | |
4139 | cpu_to_le32(new_encode_dev(inode->i_rdev)); | |
4140 | raw_inode->i_block[2] = 0; | |
4141 | } | |
de9a55b8 TT |
4142 | } else |
4143 | for (block = 0; block < EXT4_N_BLOCKS; block++) | |
4144 | raw_inode->i_block[block] = ei->i_data[block]; | |
ac27a0ec | 4145 | |
25ec56b5 JNC |
4146 | raw_inode->i_disk_version = cpu_to_le32(inode->i_version); |
4147 | if (ei->i_extra_isize) { | |
4148 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
4149 | raw_inode->i_version_hi = | |
4150 | cpu_to_le32(inode->i_version >> 32); | |
ac27a0ec | 4151 | raw_inode->i_extra_isize = cpu_to_le16(ei->i_extra_isize); |
25ec56b5 JNC |
4152 | } |
4153 | ||
814525f4 DW |
4154 | ext4_inode_csum_set(inode, raw_inode, ei); |
4155 | ||
830156c7 | 4156 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
73b50c1c | 4157 | rc = ext4_handle_dirty_metadata(handle, NULL, bh); |
830156c7 FM |
4158 | if (!err) |
4159 | err = rc; | |
19f5fb7a | 4160 | ext4_clear_inode_state(inode, EXT4_STATE_NEW); |
ac27a0ec | 4161 | |
b71fc079 | 4162 | ext4_update_inode_fsync_trans(handle, inode, need_datasync); |
ac27a0ec | 4163 | out_brelse: |
af5bc92d | 4164 | brelse(bh); |
617ba13b | 4165 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
4166 | return err; |
4167 | } | |
4168 | ||
4169 | /* | |
617ba13b | 4170 | * ext4_write_inode() |
ac27a0ec DK |
4171 | * |
4172 | * We are called from a few places: | |
4173 | * | |
4174 | * - Within generic_file_write() for O_SYNC files. | |
4175 | * Here, there will be no transaction running. We wait for any running | |
4907cb7b | 4176 | * transaction to commit. |
ac27a0ec DK |
4177 | * |
4178 | * - Within sys_sync(), kupdate and such. | |
4179 | * We wait on commit, if tol to. | |
4180 | * | |
4181 | * - Within prune_icache() (PF_MEMALLOC == true) | |
4182 | * Here we simply return. We can't afford to block kswapd on the | |
4183 | * journal commit. | |
4184 | * | |
4185 | * In all cases it is actually safe for us to return without doing anything, | |
4186 | * because the inode has been copied into a raw inode buffer in | |
617ba13b | 4187 | * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for |
ac27a0ec DK |
4188 | * knfsd. |
4189 | * | |
4190 | * Note that we are absolutely dependent upon all inode dirtiers doing the | |
4191 | * right thing: they *must* call mark_inode_dirty() after dirtying info in | |
4192 | * which we are interested. | |
4193 | * | |
4194 | * It would be a bug for them to not do this. The code: | |
4195 | * | |
4196 | * mark_inode_dirty(inode) | |
4197 | * stuff(); | |
4198 | * inode->i_size = expr; | |
4199 | * | |
4200 | * is in error because a kswapd-driven write_inode() could occur while | |
4201 | * `stuff()' is running, and the new i_size will be lost. Plus the inode | |
4202 | * will no longer be on the superblock's dirty inode list. | |
4203 | */ | |
a9185b41 | 4204 | int ext4_write_inode(struct inode *inode, struct writeback_control *wbc) |
ac27a0ec | 4205 | { |
91ac6f43 FM |
4206 | int err; |
4207 | ||
ac27a0ec DK |
4208 | if (current->flags & PF_MEMALLOC) |
4209 | return 0; | |
4210 | ||
91ac6f43 FM |
4211 | if (EXT4_SB(inode->i_sb)->s_journal) { |
4212 | if (ext4_journal_current_handle()) { | |
4213 | jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n"); | |
4214 | dump_stack(); | |
4215 | return -EIO; | |
4216 | } | |
ac27a0ec | 4217 | |
a9185b41 | 4218 | if (wbc->sync_mode != WB_SYNC_ALL) |
91ac6f43 FM |
4219 | return 0; |
4220 | ||
4221 | err = ext4_force_commit(inode->i_sb); | |
4222 | } else { | |
4223 | struct ext4_iloc iloc; | |
ac27a0ec | 4224 | |
8b472d73 | 4225 | err = __ext4_get_inode_loc(inode, &iloc, 0); |
91ac6f43 FM |
4226 | if (err) |
4227 | return err; | |
a9185b41 | 4228 | if (wbc->sync_mode == WB_SYNC_ALL) |
830156c7 FM |
4229 | sync_dirty_buffer(iloc.bh); |
4230 | if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) { | |
c398eda0 TT |
4231 | EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr, |
4232 | "IO error syncing inode"); | |
830156c7 FM |
4233 | err = -EIO; |
4234 | } | |
fd2dd9fb | 4235 | brelse(iloc.bh); |
91ac6f43 FM |
4236 | } |
4237 | return err; | |
ac27a0ec DK |
4238 | } |
4239 | ||
4240 | /* | |
617ba13b | 4241 | * ext4_setattr() |
ac27a0ec DK |
4242 | * |
4243 | * Called from notify_change. | |
4244 | * | |
4245 | * We want to trap VFS attempts to truncate the file as soon as | |
4246 | * possible. In particular, we want to make sure that when the VFS | |
4247 | * shrinks i_size, we put the inode on the orphan list and modify | |
4248 | * i_disksize immediately, so that during the subsequent flushing of | |
4249 | * dirty pages and freeing of disk blocks, we can guarantee that any | |
4250 | * commit will leave the blocks being flushed in an unused state on | |
4251 | * disk. (On recovery, the inode will get truncated and the blocks will | |
4252 | * be freed, so we have a strong guarantee that no future commit will | |
4253 | * leave these blocks visible to the user.) | |
4254 | * | |
678aaf48 JK |
4255 | * Another thing we have to assure is that if we are in ordered mode |
4256 | * and inode is still attached to the committing transaction, we must | |
4257 | * we start writeout of all the dirty pages which are being truncated. | |
4258 | * This way we are sure that all the data written in the previous | |
4259 | * transaction are already on disk (truncate waits for pages under | |
4260 | * writeback). | |
4261 | * | |
4262 | * Called with inode->i_mutex down. | |
ac27a0ec | 4263 | */ |
617ba13b | 4264 | int ext4_setattr(struct dentry *dentry, struct iattr *attr) |
ac27a0ec DK |
4265 | { |
4266 | struct inode *inode = dentry->d_inode; | |
4267 | int error, rc = 0; | |
3d287de3 | 4268 | int orphan = 0; |
ac27a0ec DK |
4269 | const unsigned int ia_valid = attr->ia_valid; |
4270 | ||
4271 | error = inode_change_ok(inode, attr); | |
4272 | if (error) | |
4273 | return error; | |
4274 | ||
12755627 | 4275 | if (is_quota_modification(inode, attr)) |
871a2931 | 4276 | dquot_initialize(inode); |
08cefc7a EB |
4277 | if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) || |
4278 | (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) { | |
ac27a0ec DK |
4279 | handle_t *handle; |
4280 | ||
4281 | /* (user+group)*(old+new) structure, inode write (sb, | |
4282 | * inode block, ? - but truncate inode update has it) */ | |
5aca07eb | 4283 | handle = ext4_journal_start(inode, (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb)+ |
194074ac | 4284 | EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb))+3); |
ac27a0ec DK |
4285 | if (IS_ERR(handle)) { |
4286 | error = PTR_ERR(handle); | |
4287 | goto err_out; | |
4288 | } | |
b43fa828 | 4289 | error = dquot_transfer(inode, attr); |
ac27a0ec | 4290 | if (error) { |
617ba13b | 4291 | ext4_journal_stop(handle); |
ac27a0ec DK |
4292 | return error; |
4293 | } | |
4294 | /* Update corresponding info in inode so that everything is in | |
4295 | * one transaction */ | |
4296 | if (attr->ia_valid & ATTR_UID) | |
4297 | inode->i_uid = attr->ia_uid; | |
4298 | if (attr->ia_valid & ATTR_GID) | |
4299 | inode->i_gid = attr->ia_gid; | |
617ba13b MC |
4300 | error = ext4_mark_inode_dirty(handle, inode); |
4301 | ext4_journal_stop(handle); | |
ac27a0ec DK |
4302 | } |
4303 | ||
e2b46574 | 4304 | if (attr->ia_valid & ATTR_SIZE) { |
562c72aa | 4305 | |
12e9b892 | 4306 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) { |
e2b46574 ES |
4307 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
4308 | ||
0c095c7f TT |
4309 | if (attr->ia_size > sbi->s_bitmap_maxbytes) |
4310 | return -EFBIG; | |
e2b46574 ES |
4311 | } |
4312 | } | |
4313 | ||
ac27a0ec | 4314 | if (S_ISREG(inode->i_mode) && |
c8d46e41 | 4315 | attr->ia_valid & ATTR_SIZE && |
072bd7ea | 4316 | (attr->ia_size < inode->i_size)) { |
ac27a0ec DK |
4317 | handle_t *handle; |
4318 | ||
617ba13b | 4319 | handle = ext4_journal_start(inode, 3); |
ac27a0ec DK |
4320 | if (IS_ERR(handle)) { |
4321 | error = PTR_ERR(handle); | |
4322 | goto err_out; | |
4323 | } | |
3d287de3 DM |
4324 | if (ext4_handle_valid(handle)) { |
4325 | error = ext4_orphan_add(handle, inode); | |
4326 | orphan = 1; | |
4327 | } | |
617ba13b MC |
4328 | EXT4_I(inode)->i_disksize = attr->ia_size; |
4329 | rc = ext4_mark_inode_dirty(handle, inode); | |
ac27a0ec DK |
4330 | if (!error) |
4331 | error = rc; | |
617ba13b | 4332 | ext4_journal_stop(handle); |
678aaf48 JK |
4333 | |
4334 | if (ext4_should_order_data(inode)) { | |
4335 | error = ext4_begin_ordered_truncate(inode, | |
4336 | attr->ia_size); | |
4337 | if (error) { | |
4338 | /* Do as much error cleanup as possible */ | |
4339 | handle = ext4_journal_start(inode, 3); | |
4340 | if (IS_ERR(handle)) { | |
4341 | ext4_orphan_del(NULL, inode); | |
4342 | goto err_out; | |
4343 | } | |
4344 | ext4_orphan_del(handle, inode); | |
3d287de3 | 4345 | orphan = 0; |
678aaf48 JK |
4346 | ext4_journal_stop(handle); |
4347 | goto err_out; | |
4348 | } | |
4349 | } | |
ac27a0ec DK |
4350 | } |
4351 | ||
072bd7ea | 4352 | if (attr->ia_valid & ATTR_SIZE) { |
1c9114f9 | 4353 | if (attr->ia_size != i_size_read(inode)) { |
072bd7ea | 4354 | truncate_setsize(inode, attr->ia_size); |
1b65007e DM |
4355 | /* Inode size will be reduced, wait for dio in flight. |
4356 | * Temporarily disable dioread_nolock to prevent | |
4357 | * livelock. */ | |
4358 | if (orphan) { | |
4359 | ext4_inode_block_unlocked_dio(inode); | |
1c9114f9 | 4360 | inode_dio_wait(inode); |
1b65007e DM |
4361 | ext4_inode_resume_unlocked_dio(inode); |
4362 | } | |
1c9114f9 | 4363 | } |
afcff5d8 | 4364 | ext4_truncate(inode); |
072bd7ea | 4365 | } |
ac27a0ec | 4366 | |
1025774c CH |
4367 | if (!rc) { |
4368 | setattr_copy(inode, attr); | |
4369 | mark_inode_dirty(inode); | |
4370 | } | |
4371 | ||
4372 | /* | |
4373 | * If the call to ext4_truncate failed to get a transaction handle at | |
4374 | * all, we need to clean up the in-core orphan list manually. | |
4375 | */ | |
3d287de3 | 4376 | if (orphan && inode->i_nlink) |
617ba13b | 4377 | ext4_orphan_del(NULL, inode); |
ac27a0ec DK |
4378 | |
4379 | if (!rc && (ia_valid & ATTR_MODE)) | |
617ba13b | 4380 | rc = ext4_acl_chmod(inode); |
ac27a0ec DK |
4381 | |
4382 | err_out: | |
617ba13b | 4383 | ext4_std_error(inode->i_sb, error); |
ac27a0ec DK |
4384 | if (!error) |
4385 | error = rc; | |
4386 | return error; | |
4387 | } | |
4388 | ||
3e3398a0 MC |
4389 | int ext4_getattr(struct vfsmount *mnt, struct dentry *dentry, |
4390 | struct kstat *stat) | |
4391 | { | |
4392 | struct inode *inode; | |
4393 | unsigned long delalloc_blocks; | |
4394 | ||
4395 | inode = dentry->d_inode; | |
4396 | generic_fillattr(inode, stat); | |
4397 | ||
4398 | /* | |
4399 | * We can't update i_blocks if the block allocation is delayed | |
4400 | * otherwise in the case of system crash before the real block | |
4401 | * allocation is done, we will have i_blocks inconsistent with | |
4402 | * on-disk file blocks. | |
4403 | * We always keep i_blocks updated together with real | |
4404 | * allocation. But to not confuse with user, stat | |
4405 | * will return the blocks that include the delayed allocation | |
4406 | * blocks for this file. | |
4407 | */ | |
96607551 TM |
4408 | delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb), |
4409 | EXT4_I(inode)->i_reserved_data_blocks); | |
3e3398a0 MC |
4410 | |
4411 | stat->blocks += (delalloc_blocks << inode->i_sb->s_blocksize_bits)>>9; | |
4412 | return 0; | |
4413 | } | |
ac27a0ec | 4414 | |
a02908f1 MC |
4415 | static int ext4_index_trans_blocks(struct inode *inode, int nrblocks, int chunk) |
4416 | { | |
12e9b892 | 4417 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) |
8bb2b247 | 4418 | return ext4_ind_trans_blocks(inode, nrblocks, chunk); |
ac51d837 | 4419 | return ext4_ext_index_trans_blocks(inode, nrblocks, chunk); |
a02908f1 | 4420 | } |
ac51d837 | 4421 | |
ac27a0ec | 4422 | /* |
a02908f1 MC |
4423 | * Account for index blocks, block groups bitmaps and block group |
4424 | * descriptor blocks if modify datablocks and index blocks | |
4425 | * worse case, the indexs blocks spread over different block groups | |
ac27a0ec | 4426 | * |
a02908f1 | 4427 | * If datablocks are discontiguous, they are possible to spread over |
4907cb7b | 4428 | * different block groups too. If they are contiguous, with flexbg, |
a02908f1 | 4429 | * they could still across block group boundary. |
ac27a0ec | 4430 | * |
a02908f1 MC |
4431 | * Also account for superblock, inode, quota and xattr blocks |
4432 | */ | |
1f109d5a | 4433 | static int ext4_meta_trans_blocks(struct inode *inode, int nrblocks, int chunk) |
a02908f1 | 4434 | { |
8df9675f TT |
4435 | ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb); |
4436 | int gdpblocks; | |
a02908f1 MC |
4437 | int idxblocks; |
4438 | int ret = 0; | |
4439 | ||
4440 | /* | |
4441 | * How many index blocks need to touch to modify nrblocks? | |
4442 | * The "Chunk" flag indicating whether the nrblocks is | |
4443 | * physically contiguous on disk | |
4444 | * | |
4445 | * For Direct IO and fallocate, they calls get_block to allocate | |
4446 | * one single extent at a time, so they could set the "Chunk" flag | |
4447 | */ | |
4448 | idxblocks = ext4_index_trans_blocks(inode, nrblocks, chunk); | |
4449 | ||
4450 | ret = idxblocks; | |
4451 | ||
4452 | /* | |
4453 | * Now let's see how many group bitmaps and group descriptors need | |
4454 | * to account | |
4455 | */ | |
4456 | groups = idxblocks; | |
4457 | if (chunk) | |
4458 | groups += 1; | |
4459 | else | |
4460 | groups += nrblocks; | |
4461 | ||
4462 | gdpblocks = groups; | |
8df9675f TT |
4463 | if (groups > ngroups) |
4464 | groups = ngroups; | |
a02908f1 MC |
4465 | if (groups > EXT4_SB(inode->i_sb)->s_gdb_count) |
4466 | gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count; | |
4467 | ||
4468 | /* bitmaps and block group descriptor blocks */ | |
4469 | ret += groups + gdpblocks; | |
4470 | ||
4471 | /* Blocks for super block, inode, quota and xattr blocks */ | |
4472 | ret += EXT4_META_TRANS_BLOCKS(inode->i_sb); | |
4473 | ||
4474 | return ret; | |
4475 | } | |
4476 | ||
4477 | /* | |
25985edc | 4478 | * Calculate the total number of credits to reserve to fit |
f3bd1f3f MC |
4479 | * the modification of a single pages into a single transaction, |
4480 | * which may include multiple chunks of block allocations. | |
ac27a0ec | 4481 | * |
525f4ed8 | 4482 | * This could be called via ext4_write_begin() |
ac27a0ec | 4483 | * |
525f4ed8 | 4484 | * We need to consider the worse case, when |
a02908f1 | 4485 | * one new block per extent. |
ac27a0ec | 4486 | */ |
a86c6181 | 4487 | int ext4_writepage_trans_blocks(struct inode *inode) |
ac27a0ec | 4488 | { |
617ba13b | 4489 | int bpp = ext4_journal_blocks_per_page(inode); |
ac27a0ec DK |
4490 | int ret; |
4491 | ||
a02908f1 | 4492 | ret = ext4_meta_trans_blocks(inode, bpp, 0); |
a86c6181 | 4493 | |
a02908f1 | 4494 | /* Account for data blocks for journalled mode */ |
617ba13b | 4495 | if (ext4_should_journal_data(inode)) |
a02908f1 | 4496 | ret += bpp; |
ac27a0ec DK |
4497 | return ret; |
4498 | } | |
f3bd1f3f MC |
4499 | |
4500 | /* | |
4501 | * Calculate the journal credits for a chunk of data modification. | |
4502 | * | |
4503 | * This is called from DIO, fallocate or whoever calling | |
79e83036 | 4504 | * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks. |
f3bd1f3f MC |
4505 | * |
4506 | * journal buffers for data blocks are not included here, as DIO | |
4507 | * and fallocate do no need to journal data buffers. | |
4508 | */ | |
4509 | int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks) | |
4510 | { | |
4511 | return ext4_meta_trans_blocks(inode, nrblocks, 1); | |
4512 | } | |
4513 | ||
ac27a0ec | 4514 | /* |
617ba13b | 4515 | * The caller must have previously called ext4_reserve_inode_write(). |
ac27a0ec DK |
4516 | * Give this, we know that the caller already has write access to iloc->bh. |
4517 | */ | |
617ba13b | 4518 | int ext4_mark_iloc_dirty(handle_t *handle, |
de9a55b8 | 4519 | struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec DK |
4520 | { |
4521 | int err = 0; | |
4522 | ||
c64db50e | 4523 | if (IS_I_VERSION(inode)) |
25ec56b5 JNC |
4524 | inode_inc_iversion(inode); |
4525 | ||
ac27a0ec DK |
4526 | /* the do_update_inode consumes one bh->b_count */ |
4527 | get_bh(iloc->bh); | |
4528 | ||
dab291af | 4529 | /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */ |
830156c7 | 4530 | err = ext4_do_update_inode(handle, inode, iloc); |
ac27a0ec DK |
4531 | put_bh(iloc->bh); |
4532 | return err; | |
4533 | } | |
4534 | ||
4535 | /* | |
4536 | * On success, We end up with an outstanding reference count against | |
4537 | * iloc->bh. This _must_ be cleaned up later. | |
4538 | */ | |
4539 | ||
4540 | int | |
617ba13b MC |
4541 | ext4_reserve_inode_write(handle_t *handle, struct inode *inode, |
4542 | struct ext4_iloc *iloc) | |
ac27a0ec | 4543 | { |
0390131b FM |
4544 | int err; |
4545 | ||
4546 | err = ext4_get_inode_loc(inode, iloc); | |
4547 | if (!err) { | |
4548 | BUFFER_TRACE(iloc->bh, "get_write_access"); | |
4549 | err = ext4_journal_get_write_access(handle, iloc->bh); | |
4550 | if (err) { | |
4551 | brelse(iloc->bh); | |
4552 | iloc->bh = NULL; | |
ac27a0ec DK |
4553 | } |
4554 | } | |
617ba13b | 4555 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
4556 | return err; |
4557 | } | |
4558 | ||
6dd4ee7c KS |
4559 | /* |
4560 | * Expand an inode by new_extra_isize bytes. | |
4561 | * Returns 0 on success or negative error number on failure. | |
4562 | */ | |
1d03ec98 AK |
4563 | static int ext4_expand_extra_isize(struct inode *inode, |
4564 | unsigned int new_extra_isize, | |
4565 | struct ext4_iloc iloc, | |
4566 | handle_t *handle) | |
6dd4ee7c KS |
4567 | { |
4568 | struct ext4_inode *raw_inode; | |
4569 | struct ext4_xattr_ibody_header *header; | |
6dd4ee7c KS |
4570 | |
4571 | if (EXT4_I(inode)->i_extra_isize >= new_extra_isize) | |
4572 | return 0; | |
4573 | ||
4574 | raw_inode = ext4_raw_inode(&iloc); | |
4575 | ||
4576 | header = IHDR(inode, raw_inode); | |
6dd4ee7c KS |
4577 | |
4578 | /* No extended attributes present */ | |
19f5fb7a TT |
4579 | if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) || |
4580 | header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) { | |
6dd4ee7c KS |
4581 | memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE, 0, |
4582 | new_extra_isize); | |
4583 | EXT4_I(inode)->i_extra_isize = new_extra_isize; | |
4584 | return 0; | |
4585 | } | |
4586 | ||
4587 | /* try to expand with EAs present */ | |
4588 | return ext4_expand_extra_isize_ea(inode, new_extra_isize, | |
4589 | raw_inode, handle); | |
4590 | } | |
4591 | ||
ac27a0ec DK |
4592 | /* |
4593 | * What we do here is to mark the in-core inode as clean with respect to inode | |
4594 | * dirtiness (it may still be data-dirty). | |
4595 | * This means that the in-core inode may be reaped by prune_icache | |
4596 | * without having to perform any I/O. This is a very good thing, | |
4597 | * because *any* task may call prune_icache - even ones which | |
4598 | * have a transaction open against a different journal. | |
4599 | * | |
4600 | * Is this cheating? Not really. Sure, we haven't written the | |
4601 | * inode out, but prune_icache isn't a user-visible syncing function. | |
4602 | * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync) | |
4603 | * we start and wait on commits. | |
ac27a0ec | 4604 | */ |
617ba13b | 4605 | int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode) |
ac27a0ec | 4606 | { |
617ba13b | 4607 | struct ext4_iloc iloc; |
6dd4ee7c KS |
4608 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
4609 | static unsigned int mnt_count; | |
4610 | int err, ret; | |
ac27a0ec DK |
4611 | |
4612 | might_sleep(); | |
7ff9c073 | 4613 | trace_ext4_mark_inode_dirty(inode, _RET_IP_); |
617ba13b | 4614 | err = ext4_reserve_inode_write(handle, inode, &iloc); |
0390131b FM |
4615 | if (ext4_handle_valid(handle) && |
4616 | EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize && | |
19f5fb7a | 4617 | !ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) { |
6dd4ee7c KS |
4618 | /* |
4619 | * We need extra buffer credits since we may write into EA block | |
4620 | * with this same handle. If journal_extend fails, then it will | |
4621 | * only result in a minor loss of functionality for that inode. | |
4622 | * If this is felt to be critical, then e2fsck should be run to | |
4623 | * force a large enough s_min_extra_isize. | |
4624 | */ | |
4625 | if ((jbd2_journal_extend(handle, | |
4626 | EXT4_DATA_TRANS_BLOCKS(inode->i_sb))) == 0) { | |
4627 | ret = ext4_expand_extra_isize(inode, | |
4628 | sbi->s_want_extra_isize, | |
4629 | iloc, handle); | |
4630 | if (ret) { | |
19f5fb7a TT |
4631 | ext4_set_inode_state(inode, |
4632 | EXT4_STATE_NO_EXPAND); | |
c1bddad9 AK |
4633 | if (mnt_count != |
4634 | le16_to_cpu(sbi->s_es->s_mnt_count)) { | |
12062ddd | 4635 | ext4_warning(inode->i_sb, |
6dd4ee7c KS |
4636 | "Unable to expand inode %lu. Delete" |
4637 | " some EAs or run e2fsck.", | |
4638 | inode->i_ino); | |
c1bddad9 AK |
4639 | mnt_count = |
4640 | le16_to_cpu(sbi->s_es->s_mnt_count); | |
6dd4ee7c KS |
4641 | } |
4642 | } | |
4643 | } | |
4644 | } | |
ac27a0ec | 4645 | if (!err) |
617ba13b | 4646 | err = ext4_mark_iloc_dirty(handle, inode, &iloc); |
ac27a0ec DK |
4647 | return err; |
4648 | } | |
4649 | ||
4650 | /* | |
617ba13b | 4651 | * ext4_dirty_inode() is called from __mark_inode_dirty() |
ac27a0ec DK |
4652 | * |
4653 | * We're really interested in the case where a file is being extended. | |
4654 | * i_size has been changed by generic_commit_write() and we thus need | |
4655 | * to include the updated inode in the current transaction. | |
4656 | * | |
5dd4056d | 4657 | * Also, dquot_alloc_block() will always dirty the inode when blocks |
ac27a0ec DK |
4658 | * are allocated to the file. |
4659 | * | |
4660 | * If the inode is marked synchronous, we don't honour that here - doing | |
4661 | * so would cause a commit on atime updates, which we don't bother doing. | |
4662 | * We handle synchronous inodes at the highest possible level. | |
4663 | */ | |
aa385729 | 4664 | void ext4_dirty_inode(struct inode *inode, int flags) |
ac27a0ec | 4665 | { |
ac27a0ec DK |
4666 | handle_t *handle; |
4667 | ||
617ba13b | 4668 | handle = ext4_journal_start(inode, 2); |
ac27a0ec DK |
4669 | if (IS_ERR(handle)) |
4670 | goto out; | |
f3dc272f | 4671 | |
f3dc272f CW |
4672 | ext4_mark_inode_dirty(handle, inode); |
4673 | ||
617ba13b | 4674 | ext4_journal_stop(handle); |
ac27a0ec DK |
4675 | out: |
4676 | return; | |
4677 | } | |
4678 | ||
4679 | #if 0 | |
4680 | /* | |
4681 | * Bind an inode's backing buffer_head into this transaction, to prevent | |
4682 | * it from being flushed to disk early. Unlike | |
617ba13b | 4683 | * ext4_reserve_inode_write, this leaves behind no bh reference and |
ac27a0ec DK |
4684 | * returns no iloc structure, so the caller needs to repeat the iloc |
4685 | * lookup to mark the inode dirty later. | |
4686 | */ | |
617ba13b | 4687 | static int ext4_pin_inode(handle_t *handle, struct inode *inode) |
ac27a0ec | 4688 | { |
617ba13b | 4689 | struct ext4_iloc iloc; |
ac27a0ec DK |
4690 | |
4691 | int err = 0; | |
4692 | if (handle) { | |
617ba13b | 4693 | err = ext4_get_inode_loc(inode, &iloc); |
ac27a0ec DK |
4694 | if (!err) { |
4695 | BUFFER_TRACE(iloc.bh, "get_write_access"); | |
dab291af | 4696 | err = jbd2_journal_get_write_access(handle, iloc.bh); |
ac27a0ec | 4697 | if (!err) |
0390131b | 4698 | err = ext4_handle_dirty_metadata(handle, |
73b50c1c | 4699 | NULL, |
0390131b | 4700 | iloc.bh); |
ac27a0ec DK |
4701 | brelse(iloc.bh); |
4702 | } | |
4703 | } | |
617ba13b | 4704 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
4705 | return err; |
4706 | } | |
4707 | #endif | |
4708 | ||
617ba13b | 4709 | int ext4_change_inode_journal_flag(struct inode *inode, int val) |
ac27a0ec DK |
4710 | { |
4711 | journal_t *journal; | |
4712 | handle_t *handle; | |
4713 | int err; | |
4714 | ||
4715 | /* | |
4716 | * We have to be very careful here: changing a data block's | |
4717 | * journaling status dynamically is dangerous. If we write a | |
4718 | * data block to the journal, change the status and then delete | |
4719 | * that block, we risk forgetting to revoke the old log record | |
4720 | * from the journal and so a subsequent replay can corrupt data. | |
4721 | * So, first we make sure that the journal is empty and that | |
4722 | * nobody is changing anything. | |
4723 | */ | |
4724 | ||
617ba13b | 4725 | journal = EXT4_JOURNAL(inode); |
0390131b FM |
4726 | if (!journal) |
4727 | return 0; | |
d699594d | 4728 | if (is_journal_aborted(journal)) |
ac27a0ec | 4729 | return -EROFS; |
2aff57b0 YY |
4730 | /* We have to allocate physical blocks for delalloc blocks |
4731 | * before flushing journal. otherwise delalloc blocks can not | |
4732 | * be allocated any more. even more truncate on delalloc blocks | |
4733 | * could trigger BUG by flushing delalloc blocks in journal. | |
4734 | * There is no delalloc block in non-journal data mode. | |
4735 | */ | |
4736 | if (val && test_opt(inode->i_sb, DELALLOC)) { | |
4737 | err = ext4_alloc_da_blocks(inode); | |
4738 | if (err < 0) | |
4739 | return err; | |
4740 | } | |
ac27a0ec | 4741 | |
17335dcc DM |
4742 | /* Wait for all existing dio workers */ |
4743 | ext4_inode_block_unlocked_dio(inode); | |
4744 | inode_dio_wait(inode); | |
4745 | ||
dab291af | 4746 | jbd2_journal_lock_updates(journal); |
ac27a0ec DK |
4747 | |
4748 | /* | |
4749 | * OK, there are no updates running now, and all cached data is | |
4750 | * synced to disk. We are now in a completely consistent state | |
4751 | * which doesn't have anything in the journal, and we know that | |
4752 | * no filesystem updates are running, so it is safe to modify | |
4753 | * the inode's in-core data-journaling state flag now. | |
4754 | */ | |
4755 | ||
4756 | if (val) | |
12e9b892 | 4757 | ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
5872ddaa YY |
4758 | else { |
4759 | jbd2_journal_flush(journal); | |
12e9b892 | 4760 | ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
5872ddaa | 4761 | } |
617ba13b | 4762 | ext4_set_aops(inode); |
ac27a0ec | 4763 | |
dab291af | 4764 | jbd2_journal_unlock_updates(journal); |
17335dcc | 4765 | ext4_inode_resume_unlocked_dio(inode); |
ac27a0ec DK |
4766 | |
4767 | /* Finally we can mark the inode as dirty. */ | |
4768 | ||
617ba13b | 4769 | handle = ext4_journal_start(inode, 1); |
ac27a0ec DK |
4770 | if (IS_ERR(handle)) |
4771 | return PTR_ERR(handle); | |
4772 | ||
617ba13b | 4773 | err = ext4_mark_inode_dirty(handle, inode); |
0390131b | 4774 | ext4_handle_sync(handle); |
617ba13b MC |
4775 | ext4_journal_stop(handle); |
4776 | ext4_std_error(inode->i_sb, err); | |
ac27a0ec DK |
4777 | |
4778 | return err; | |
4779 | } | |
2e9ee850 AK |
4780 | |
4781 | static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh) | |
4782 | { | |
4783 | return !buffer_mapped(bh); | |
4784 | } | |
4785 | ||
c2ec175c | 4786 | int ext4_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
2e9ee850 | 4787 | { |
c2ec175c | 4788 | struct page *page = vmf->page; |
2e9ee850 AK |
4789 | loff_t size; |
4790 | unsigned long len; | |
9ea7df53 | 4791 | int ret; |
2e9ee850 AK |
4792 | struct file *file = vma->vm_file; |
4793 | struct inode *inode = file->f_path.dentry->d_inode; | |
4794 | struct address_space *mapping = inode->i_mapping; | |
9ea7df53 JK |
4795 | handle_t *handle; |
4796 | get_block_t *get_block; | |
4797 | int retries = 0; | |
2e9ee850 | 4798 | |
8e8ad8a5 | 4799 | sb_start_pagefault(inode->i_sb); |
041bbb6d | 4800 | file_update_time(vma->vm_file); |
9ea7df53 JK |
4801 | /* Delalloc case is easy... */ |
4802 | if (test_opt(inode->i_sb, DELALLOC) && | |
4803 | !ext4_should_journal_data(inode) && | |
4804 | !ext4_nonda_switch(inode->i_sb)) { | |
4805 | do { | |
4806 | ret = __block_page_mkwrite(vma, vmf, | |
4807 | ext4_da_get_block_prep); | |
4808 | } while (ret == -ENOSPC && | |
4809 | ext4_should_retry_alloc(inode->i_sb, &retries)); | |
4810 | goto out_ret; | |
2e9ee850 | 4811 | } |
0e499890 DW |
4812 | |
4813 | lock_page(page); | |
9ea7df53 JK |
4814 | size = i_size_read(inode); |
4815 | /* Page got truncated from under us? */ | |
4816 | if (page->mapping != mapping || page_offset(page) > size) { | |
4817 | unlock_page(page); | |
4818 | ret = VM_FAULT_NOPAGE; | |
4819 | goto out; | |
0e499890 | 4820 | } |
2e9ee850 AK |
4821 | |
4822 | if (page->index == size >> PAGE_CACHE_SHIFT) | |
4823 | len = size & ~PAGE_CACHE_MASK; | |
4824 | else | |
4825 | len = PAGE_CACHE_SIZE; | |
a827eaff | 4826 | /* |
9ea7df53 JK |
4827 | * Return if we have all the buffers mapped. This avoids the need to do |
4828 | * journal_start/journal_stop which can block and take a long time | |
a827eaff | 4829 | */ |
2e9ee850 | 4830 | if (page_has_buffers(page)) { |
2e9ee850 | 4831 | if (!walk_page_buffers(NULL, page_buffers(page), 0, len, NULL, |
a827eaff | 4832 | ext4_bh_unmapped)) { |
9ea7df53 JK |
4833 | /* Wait so that we don't change page under IO */ |
4834 | wait_on_page_writeback(page); | |
4835 | ret = VM_FAULT_LOCKED; | |
4836 | goto out; | |
a827eaff | 4837 | } |
2e9ee850 | 4838 | } |
a827eaff | 4839 | unlock_page(page); |
9ea7df53 JK |
4840 | /* OK, we need to fill the hole... */ |
4841 | if (ext4_should_dioread_nolock(inode)) | |
4842 | get_block = ext4_get_block_write; | |
4843 | else | |
4844 | get_block = ext4_get_block; | |
4845 | retry_alloc: | |
4846 | handle = ext4_journal_start(inode, ext4_writepage_trans_blocks(inode)); | |
4847 | if (IS_ERR(handle)) { | |
c2ec175c | 4848 | ret = VM_FAULT_SIGBUS; |
9ea7df53 JK |
4849 | goto out; |
4850 | } | |
4851 | ret = __block_page_mkwrite(vma, vmf, get_block); | |
4852 | if (!ret && ext4_should_journal_data(inode)) { | |
4853 | if (walk_page_buffers(handle, page_buffers(page), 0, | |
4854 | PAGE_CACHE_SIZE, NULL, do_journal_get_write_access)) { | |
4855 | unlock_page(page); | |
4856 | ret = VM_FAULT_SIGBUS; | |
fcbb5515 | 4857 | ext4_journal_stop(handle); |
9ea7df53 JK |
4858 | goto out; |
4859 | } | |
4860 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); | |
4861 | } | |
4862 | ext4_journal_stop(handle); | |
4863 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) | |
4864 | goto retry_alloc; | |
4865 | out_ret: | |
4866 | ret = block_page_mkwrite_return(ret); | |
4867 | out: | |
8e8ad8a5 | 4868 | sb_end_pagefault(inode->i_sb); |
2e9ee850 AK |
4869 | return ret; |
4870 | } |