]>
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> | |
ac27a0ec DK |
23 | #include <linux/highuid.h> |
24 | #include <linux/pagemap.h> | |
c94c2acf | 25 | #include <linux/dax.h> |
ac27a0ec DK |
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> |
00a1a053 | 39 | #include <linux/bitops.h> |
364443cb | 40 | #include <linux/iomap.h> |
9bffad1e | 41 | |
3dcf5451 | 42 | #include "ext4_jbd2.h" |
ac27a0ec DK |
43 | #include "xattr.h" |
44 | #include "acl.h" | |
9f125d64 | 45 | #include "truncate.h" |
ac27a0ec | 46 | |
9bffad1e TT |
47 | #include <trace/events/ext4.h> |
48 | ||
a1d6cc56 AK |
49 | #define MPAGE_DA_EXTENT_TAIL 0x01 |
50 | ||
814525f4 DW |
51 | static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw, |
52 | struct ext4_inode_info *ei) | |
53 | { | |
54 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
814525f4 | 55 | __u32 csum; |
b47820ed DJ |
56 | __u16 dummy_csum = 0; |
57 | int offset = offsetof(struct ext4_inode, i_checksum_lo); | |
58 | unsigned int csum_size = sizeof(dummy_csum); | |
814525f4 | 59 | |
b47820ed DJ |
60 | csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw, offset); |
61 | csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, csum_size); | |
62 | offset += csum_size; | |
63 | csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset, | |
64 | EXT4_GOOD_OLD_INODE_SIZE - offset); | |
814525f4 | 65 | |
b47820ed DJ |
66 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
67 | offset = offsetof(struct ext4_inode, i_checksum_hi); | |
68 | csum = ext4_chksum(sbi, csum, (__u8 *)raw + | |
69 | EXT4_GOOD_OLD_INODE_SIZE, | |
70 | offset - EXT4_GOOD_OLD_INODE_SIZE); | |
71 | if (EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) { | |
72 | csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, | |
73 | csum_size); | |
74 | offset += csum_size; | |
b47820ed | 75 | } |
05ac5aa1 DJ |
76 | csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset, |
77 | EXT4_INODE_SIZE(inode->i_sb) - offset); | |
814525f4 DW |
78 | } |
79 | ||
814525f4 DW |
80 | return csum; |
81 | } | |
82 | ||
83 | static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw, | |
84 | struct ext4_inode_info *ei) | |
85 | { | |
86 | __u32 provided, calculated; | |
87 | ||
88 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != | |
89 | cpu_to_le32(EXT4_OS_LINUX) || | |
9aa5d32b | 90 | !ext4_has_metadata_csum(inode->i_sb)) |
814525f4 DW |
91 | return 1; |
92 | ||
93 | provided = le16_to_cpu(raw->i_checksum_lo); | |
94 | calculated = ext4_inode_csum(inode, raw, ei); | |
95 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && | |
96 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) | |
97 | provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16; | |
98 | else | |
99 | calculated &= 0xFFFF; | |
100 | ||
101 | return provided == calculated; | |
102 | } | |
103 | ||
104 | static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw, | |
105 | struct ext4_inode_info *ei) | |
106 | { | |
107 | __u32 csum; | |
108 | ||
109 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != | |
110 | cpu_to_le32(EXT4_OS_LINUX) || | |
9aa5d32b | 111 | !ext4_has_metadata_csum(inode->i_sb)) |
814525f4 DW |
112 | return; |
113 | ||
114 | csum = ext4_inode_csum(inode, raw, ei); | |
115 | raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF); | |
116 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && | |
117 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) | |
118 | raw->i_checksum_hi = cpu_to_le16(csum >> 16); | |
119 | } | |
120 | ||
678aaf48 JK |
121 | static inline int ext4_begin_ordered_truncate(struct inode *inode, |
122 | loff_t new_size) | |
123 | { | |
7ff9c073 | 124 | trace_ext4_begin_ordered_truncate(inode, new_size); |
8aefcd55 TT |
125 | /* |
126 | * If jinode is zero, then we never opened the file for | |
127 | * writing, so there's no need to call | |
128 | * jbd2_journal_begin_ordered_truncate() since there's no | |
129 | * outstanding writes we need to flush. | |
130 | */ | |
131 | if (!EXT4_I(inode)->jinode) | |
132 | return 0; | |
133 | return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode), | |
134 | EXT4_I(inode)->jinode, | |
135 | new_size); | |
678aaf48 JK |
136 | } |
137 | ||
d47992f8 LC |
138 | static void ext4_invalidatepage(struct page *page, unsigned int offset, |
139 | unsigned int length); | |
cb20d518 TT |
140 | static int __ext4_journalled_writepage(struct page *page, unsigned int len); |
141 | static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh); | |
dec214d0 TE |
142 | static int ext4_meta_trans_blocks(struct inode *inode, int lblocks, |
143 | int pextents); | |
64769240 | 144 | |
ac27a0ec DK |
145 | /* |
146 | * Test whether an inode is a fast symlink. | |
407cd7fb | 147 | * A fast symlink has its symlink data stored in ext4_inode_info->i_data. |
ac27a0ec | 148 | */ |
f348c252 | 149 | int ext4_inode_is_fast_symlink(struct inode *inode) |
ac27a0ec | 150 | { |
407cd7fb TE |
151 | return S_ISLNK(inode->i_mode) && inode->i_size && |
152 | (inode->i_size < EXT4_N_BLOCKS * 4); | |
ac27a0ec DK |
153 | } |
154 | ||
ac27a0ec DK |
155 | /* |
156 | * Restart the transaction associated with *handle. This does a commit, | |
157 | * so before we call here everything must be consistently dirtied against | |
158 | * this transaction. | |
159 | */ | |
fa5d1113 | 160 | int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode, |
487caeef | 161 | int nblocks) |
ac27a0ec | 162 | { |
487caeef JK |
163 | int ret; |
164 | ||
165 | /* | |
e35fd660 | 166 | * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this |
487caeef JK |
167 | * moment, get_block can be called only for blocks inside i_size since |
168 | * page cache has been already dropped and writes are blocked by | |
169 | * i_mutex. So we can safely drop the i_data_sem here. | |
170 | */ | |
0390131b | 171 | BUG_ON(EXT4_JOURNAL(inode) == NULL); |
ac27a0ec | 172 | jbd_debug(2, "restarting handle %p\n", handle); |
487caeef | 173 | up_write(&EXT4_I(inode)->i_data_sem); |
8e8eaabe | 174 | ret = ext4_journal_restart(handle, nblocks); |
487caeef | 175 | down_write(&EXT4_I(inode)->i_data_sem); |
fa5d1113 | 176 | ext4_discard_preallocations(inode); |
487caeef JK |
177 | |
178 | return ret; | |
ac27a0ec DK |
179 | } |
180 | ||
181 | /* | |
182 | * Called at the last iput() if i_nlink is zero. | |
183 | */ | |
0930fcc1 | 184 | void ext4_evict_inode(struct inode *inode) |
ac27a0ec DK |
185 | { |
186 | handle_t *handle; | |
bc965ab3 | 187 | int err; |
e50e5129 | 188 | int extra_credits = 3; |
0421a189 | 189 | struct ext4_xattr_inode_array *ea_inode_array = NULL; |
ac27a0ec | 190 | |
7ff9c073 | 191 | trace_ext4_evict_inode(inode); |
2581fdc8 | 192 | |
0930fcc1 | 193 | if (inode->i_nlink) { |
2d859db3 JK |
194 | /* |
195 | * When journalling data dirty buffers are tracked only in the | |
196 | * journal. So although mm thinks everything is clean and | |
197 | * ready for reaping the inode might still have some pages to | |
198 | * write in the running transaction or waiting to be | |
199 | * checkpointed. Thus calling jbd2_journal_invalidatepage() | |
200 | * (via truncate_inode_pages()) to discard these buffers can | |
201 | * cause data loss. Also even if we did not discard these | |
202 | * buffers, we would have no way to find them after the inode | |
203 | * is reaped and thus user could see stale data if he tries to | |
204 | * read them before the transaction is checkpointed. So be | |
205 | * careful and force everything to disk here... We use | |
206 | * ei->i_datasync_tid to store the newest transaction | |
207 | * containing inode's data. | |
208 | * | |
209 | * Note that directories do not have this problem because they | |
210 | * don't use page cache. | |
211 | */ | |
6a7fd522 VN |
212 | if (inode->i_ino != EXT4_JOURNAL_INO && |
213 | ext4_should_journal_data(inode) && | |
3abb1a0f JK |
214 | (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) && |
215 | inode->i_data.nrpages) { | |
2d859db3 JK |
216 | journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; |
217 | tid_t commit_tid = EXT4_I(inode)->i_datasync_tid; | |
218 | ||
d76a3a77 | 219 | jbd2_complete_transaction(journal, commit_tid); |
2d859db3 JK |
220 | filemap_write_and_wait(&inode->i_data); |
221 | } | |
91b0abe3 | 222 | truncate_inode_pages_final(&inode->i_data); |
5dc23bdd | 223 | |
0930fcc1 AV |
224 | goto no_delete; |
225 | } | |
226 | ||
e2bfb088 TT |
227 | if (is_bad_inode(inode)) |
228 | goto no_delete; | |
229 | dquot_initialize(inode); | |
907f4554 | 230 | |
678aaf48 JK |
231 | if (ext4_should_order_data(inode)) |
232 | ext4_begin_ordered_truncate(inode, 0); | |
91b0abe3 | 233 | truncate_inode_pages_final(&inode->i_data); |
ac27a0ec | 234 | |
8e8ad8a5 JK |
235 | /* |
236 | * Protect us against freezing - iput() caller didn't have to have any | |
237 | * protection against it | |
238 | */ | |
239 | sb_start_intwrite(inode->i_sb); | |
e50e5129 | 240 | |
30a7eb97 TE |
241 | if (!IS_NOQUOTA(inode)) |
242 | extra_credits += EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb); | |
243 | ||
244 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, | |
245 | ext4_blocks_for_truncate(inode)+extra_credits); | |
ac27a0ec | 246 | if (IS_ERR(handle)) { |
bc965ab3 | 247 | ext4_std_error(inode->i_sb, PTR_ERR(handle)); |
ac27a0ec DK |
248 | /* |
249 | * If we're going to skip the normal cleanup, we still need to | |
250 | * make sure that the in-core orphan linked list is properly | |
251 | * cleaned up. | |
252 | */ | |
617ba13b | 253 | ext4_orphan_del(NULL, inode); |
8e8ad8a5 | 254 | sb_end_intwrite(inode->i_sb); |
ac27a0ec DK |
255 | goto no_delete; |
256 | } | |
30a7eb97 | 257 | |
ac27a0ec | 258 | if (IS_SYNC(inode)) |
0390131b | 259 | ext4_handle_sync(handle); |
407cd7fb TE |
260 | |
261 | /* | |
262 | * Set inode->i_size to 0 before calling ext4_truncate(). We need | |
263 | * special handling of symlinks here because i_size is used to | |
264 | * determine whether ext4_inode_info->i_data contains symlink data or | |
265 | * block mappings. Setting i_size to 0 will remove its fast symlink | |
266 | * status. Erase i_data so that it becomes a valid empty block map. | |
267 | */ | |
268 | if (ext4_inode_is_fast_symlink(inode)) | |
269 | memset(EXT4_I(inode)->i_data, 0, sizeof(EXT4_I(inode)->i_data)); | |
ac27a0ec | 270 | inode->i_size = 0; |
bc965ab3 TT |
271 | err = ext4_mark_inode_dirty(handle, inode); |
272 | if (err) { | |
12062ddd | 273 | ext4_warning(inode->i_sb, |
bc965ab3 TT |
274 | "couldn't mark inode dirty (err %d)", err); |
275 | goto stop_handle; | |
276 | } | |
2c98eb5e TT |
277 | if (inode->i_blocks) { |
278 | err = ext4_truncate(inode); | |
279 | if (err) { | |
280 | ext4_error(inode->i_sb, | |
281 | "couldn't truncate inode %lu (err %d)", | |
282 | inode->i_ino, err); | |
283 | goto stop_handle; | |
284 | } | |
285 | } | |
bc965ab3 | 286 | |
30a7eb97 TE |
287 | /* Remove xattr references. */ |
288 | err = ext4_xattr_delete_inode(handle, inode, &ea_inode_array, | |
289 | extra_credits); | |
290 | if (err) { | |
291 | ext4_warning(inode->i_sb, "xattr delete (err %d)", err); | |
292 | stop_handle: | |
293 | ext4_journal_stop(handle); | |
294 | ext4_orphan_del(NULL, inode); | |
295 | sb_end_intwrite(inode->i_sb); | |
296 | ext4_xattr_inode_array_free(ea_inode_array); | |
297 | goto no_delete; | |
bc965ab3 TT |
298 | } |
299 | ||
ac27a0ec | 300 | /* |
617ba13b | 301 | * Kill off the orphan record which ext4_truncate created. |
ac27a0ec | 302 | * AKPM: I think this can be inside the above `if'. |
617ba13b | 303 | * Note that ext4_orphan_del() has to be able to cope with the |
ac27a0ec | 304 | * deletion of a non-existent orphan - this is because we don't |
617ba13b | 305 | * know if ext4_truncate() actually created an orphan record. |
ac27a0ec DK |
306 | * (Well, we could do this if we need to, but heck - it works) |
307 | */ | |
617ba13b MC |
308 | ext4_orphan_del(handle, inode); |
309 | EXT4_I(inode)->i_dtime = get_seconds(); | |
ac27a0ec DK |
310 | |
311 | /* | |
312 | * One subtle ordering requirement: if anything has gone wrong | |
313 | * (transaction abort, IO errors, whatever), then we can still | |
314 | * do these next steps (the fs will already have been marked as | |
315 | * having errors), but we can't free the inode if the mark_dirty | |
316 | * fails. | |
317 | */ | |
617ba13b | 318 | if (ext4_mark_inode_dirty(handle, inode)) |
ac27a0ec | 319 | /* If that failed, just do the required in-core inode clear. */ |
0930fcc1 | 320 | ext4_clear_inode(inode); |
ac27a0ec | 321 | else |
617ba13b MC |
322 | ext4_free_inode(handle, inode); |
323 | ext4_journal_stop(handle); | |
8e8ad8a5 | 324 | sb_end_intwrite(inode->i_sb); |
0421a189 | 325 | ext4_xattr_inode_array_free(ea_inode_array); |
ac27a0ec DK |
326 | return; |
327 | no_delete: | |
0930fcc1 | 328 | ext4_clear_inode(inode); /* We must guarantee clearing of inode... */ |
ac27a0ec DK |
329 | } |
330 | ||
a9e7f447 DM |
331 | #ifdef CONFIG_QUOTA |
332 | qsize_t *ext4_get_reserved_space(struct inode *inode) | |
60e58e0f | 333 | { |
a9e7f447 | 334 | return &EXT4_I(inode)->i_reserved_quota; |
60e58e0f | 335 | } |
a9e7f447 | 336 | #endif |
9d0be502 | 337 | |
0637c6f4 TT |
338 | /* |
339 | * Called with i_data_sem down, which is important since we can call | |
340 | * ext4_discard_preallocations() from here. | |
341 | */ | |
5f634d06 AK |
342 | void ext4_da_update_reserve_space(struct inode *inode, |
343 | int used, int quota_claim) | |
12219aea AK |
344 | { |
345 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0637c6f4 | 346 | struct ext4_inode_info *ei = EXT4_I(inode); |
0637c6f4 TT |
347 | |
348 | spin_lock(&ei->i_block_reservation_lock); | |
d8990240 | 349 | trace_ext4_da_update_reserve_space(inode, used, quota_claim); |
0637c6f4 | 350 | if (unlikely(used > ei->i_reserved_data_blocks)) { |
8de5c325 | 351 | ext4_warning(inode->i_sb, "%s: ino %lu, used %d " |
1084f252 | 352 | "with only %d reserved data blocks", |
0637c6f4 TT |
353 | __func__, inode->i_ino, used, |
354 | ei->i_reserved_data_blocks); | |
355 | WARN_ON(1); | |
356 | used = ei->i_reserved_data_blocks; | |
357 | } | |
12219aea | 358 | |
0637c6f4 TT |
359 | /* Update per-inode reservations */ |
360 | ei->i_reserved_data_blocks -= used; | |
71d4f7d0 | 361 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, used); |
6bc6e63f | 362 | |
12219aea | 363 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
60e58e0f | 364 | |
72b8ab9d ES |
365 | /* Update quota subsystem for data blocks */ |
366 | if (quota_claim) | |
7b415bf6 | 367 | dquot_claim_block(inode, EXT4_C2B(sbi, used)); |
72b8ab9d | 368 | else { |
5f634d06 AK |
369 | /* |
370 | * We did fallocate with an offset that is already delayed | |
371 | * allocated. So on delayed allocated writeback we should | |
72b8ab9d | 372 | * not re-claim the quota for fallocated blocks. |
5f634d06 | 373 | */ |
7b415bf6 | 374 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, used)); |
5f634d06 | 375 | } |
d6014301 AK |
376 | |
377 | /* | |
378 | * If we have done all the pending block allocations and if | |
379 | * there aren't any writers on the inode, we can discard the | |
380 | * inode's preallocations. | |
381 | */ | |
0637c6f4 TT |
382 | if ((ei->i_reserved_data_blocks == 0) && |
383 | (atomic_read(&inode->i_writecount) == 0)) | |
d6014301 | 384 | ext4_discard_preallocations(inode); |
12219aea AK |
385 | } |
386 | ||
e29136f8 | 387 | static int __check_block_validity(struct inode *inode, const char *func, |
c398eda0 TT |
388 | unsigned int line, |
389 | struct ext4_map_blocks *map) | |
6fd058f7 | 390 | { |
24676da4 TT |
391 | if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk, |
392 | map->m_len)) { | |
c398eda0 TT |
393 | ext4_error_inode(inode, func, line, map->m_pblk, |
394 | "lblock %lu mapped to illegal pblock " | |
395 | "(length %d)", (unsigned long) map->m_lblk, | |
396 | map->m_len); | |
6a797d27 | 397 | return -EFSCORRUPTED; |
6fd058f7 TT |
398 | } |
399 | return 0; | |
400 | } | |
401 | ||
53085fac JK |
402 | int ext4_issue_zeroout(struct inode *inode, ext4_lblk_t lblk, ext4_fsblk_t pblk, |
403 | ext4_lblk_t len) | |
404 | { | |
405 | int ret; | |
406 | ||
407 | if (ext4_encrypted_inode(inode)) | |
a7550b30 | 408 | return fscrypt_zeroout_range(inode, lblk, pblk, len); |
53085fac JK |
409 | |
410 | ret = sb_issue_zeroout(inode->i_sb, pblk, len, GFP_NOFS); | |
411 | if (ret > 0) | |
412 | ret = 0; | |
413 | ||
414 | return ret; | |
415 | } | |
416 | ||
e29136f8 | 417 | #define check_block_validity(inode, map) \ |
c398eda0 | 418 | __check_block_validity((inode), __func__, __LINE__, (map)) |
e29136f8 | 419 | |
921f266b DM |
420 | #ifdef ES_AGGRESSIVE_TEST |
421 | static void ext4_map_blocks_es_recheck(handle_t *handle, | |
422 | struct inode *inode, | |
423 | struct ext4_map_blocks *es_map, | |
424 | struct ext4_map_blocks *map, | |
425 | int flags) | |
426 | { | |
427 | int retval; | |
428 | ||
429 | map->m_flags = 0; | |
430 | /* | |
431 | * There is a race window that the result is not the same. | |
432 | * e.g. xfstests #223 when dioread_nolock enables. The reason | |
433 | * is that we lookup a block mapping in extent status tree with | |
434 | * out taking i_data_sem. So at the time the unwritten extent | |
435 | * could be converted. | |
436 | */ | |
2dcba478 | 437 | down_read(&EXT4_I(inode)->i_data_sem); |
921f266b DM |
438 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
439 | retval = ext4_ext_map_blocks(handle, inode, map, flags & | |
440 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
441 | } else { | |
442 | retval = ext4_ind_map_blocks(handle, inode, map, flags & | |
443 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
444 | } | |
2dcba478 | 445 | up_read((&EXT4_I(inode)->i_data_sem)); |
921f266b DM |
446 | |
447 | /* | |
448 | * We don't check m_len because extent will be collpased in status | |
449 | * tree. So the m_len might not equal. | |
450 | */ | |
451 | if (es_map->m_lblk != map->m_lblk || | |
452 | es_map->m_flags != map->m_flags || | |
453 | es_map->m_pblk != map->m_pblk) { | |
bdafe42a | 454 | printk("ES cache assertion failed for inode: %lu " |
921f266b DM |
455 | "es_cached ex [%d/%d/%llu/%x] != " |
456 | "found ex [%d/%d/%llu/%x] retval %d flags %x\n", | |
457 | inode->i_ino, es_map->m_lblk, es_map->m_len, | |
458 | es_map->m_pblk, es_map->m_flags, map->m_lblk, | |
459 | map->m_len, map->m_pblk, map->m_flags, | |
460 | retval, flags); | |
461 | } | |
462 | } | |
463 | #endif /* ES_AGGRESSIVE_TEST */ | |
464 | ||
f5ab0d1f | 465 | /* |
e35fd660 | 466 | * The ext4_map_blocks() function tries to look up the requested blocks, |
2b2d6d01 | 467 | * and returns if the blocks are already mapped. |
f5ab0d1f | 468 | * |
f5ab0d1f MC |
469 | * Otherwise it takes the write lock of the i_data_sem and allocate blocks |
470 | * and store the allocated blocks in the result buffer head and mark it | |
471 | * mapped. | |
472 | * | |
e35fd660 TT |
473 | * If file type is extents based, it will call ext4_ext_map_blocks(), |
474 | * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping | |
f5ab0d1f MC |
475 | * based files |
476 | * | |
facab4d9 JK |
477 | * On success, it returns the number of blocks being mapped or allocated. if |
478 | * create==0 and the blocks are pre-allocated and unwritten, the resulting @map | |
479 | * is marked as unwritten. If the create == 1, it will mark @map as mapped. | |
f5ab0d1f MC |
480 | * |
481 | * It returns 0 if plain look up failed (blocks have not been allocated), in | |
facab4d9 JK |
482 | * that case, @map is returned as unmapped but we still do fill map->m_len to |
483 | * indicate the length of a hole starting at map->m_lblk. | |
f5ab0d1f MC |
484 | * |
485 | * It returns the error in case of allocation failure. | |
486 | */ | |
e35fd660 TT |
487 | int ext4_map_blocks(handle_t *handle, struct inode *inode, |
488 | struct ext4_map_blocks *map, int flags) | |
0e855ac8 | 489 | { |
d100eef2 | 490 | struct extent_status es; |
0e855ac8 | 491 | int retval; |
b8a86845 | 492 | int ret = 0; |
921f266b DM |
493 | #ifdef ES_AGGRESSIVE_TEST |
494 | struct ext4_map_blocks orig_map; | |
495 | ||
496 | memcpy(&orig_map, map, sizeof(*map)); | |
497 | #endif | |
f5ab0d1f | 498 | |
e35fd660 TT |
499 | map->m_flags = 0; |
500 | ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u," | |
501 | "logical block %lu\n", inode->i_ino, flags, map->m_len, | |
502 | (unsigned long) map->m_lblk); | |
d100eef2 | 503 | |
e861b5e9 TT |
504 | /* |
505 | * ext4_map_blocks returns an int, and m_len is an unsigned int | |
506 | */ | |
507 | if (unlikely(map->m_len > INT_MAX)) | |
508 | map->m_len = INT_MAX; | |
509 | ||
4adb6ab3 KM |
510 | /* We can handle the block number less than EXT_MAX_BLOCKS */ |
511 | if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS)) | |
6a797d27 | 512 | return -EFSCORRUPTED; |
4adb6ab3 | 513 | |
d100eef2 ZL |
514 | /* Lookup extent status tree firstly */ |
515 | if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) { | |
516 | if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) { | |
517 | map->m_pblk = ext4_es_pblock(&es) + | |
518 | map->m_lblk - es.es_lblk; | |
519 | map->m_flags |= ext4_es_is_written(&es) ? | |
520 | EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN; | |
521 | retval = es.es_len - (map->m_lblk - es.es_lblk); | |
522 | if (retval > map->m_len) | |
523 | retval = map->m_len; | |
524 | map->m_len = retval; | |
525 | } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) { | |
facab4d9 JK |
526 | map->m_pblk = 0; |
527 | retval = es.es_len - (map->m_lblk - es.es_lblk); | |
528 | if (retval > map->m_len) | |
529 | retval = map->m_len; | |
530 | map->m_len = retval; | |
d100eef2 ZL |
531 | retval = 0; |
532 | } else { | |
533 | BUG_ON(1); | |
534 | } | |
921f266b DM |
535 | #ifdef ES_AGGRESSIVE_TEST |
536 | ext4_map_blocks_es_recheck(handle, inode, map, | |
537 | &orig_map, flags); | |
538 | #endif | |
d100eef2 ZL |
539 | goto found; |
540 | } | |
541 | ||
4df3d265 | 542 | /* |
b920c755 TT |
543 | * Try to see if we can get the block without requesting a new |
544 | * file system block. | |
4df3d265 | 545 | */ |
2dcba478 | 546 | down_read(&EXT4_I(inode)->i_data_sem); |
12e9b892 | 547 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
a4e5d88b DM |
548 | retval = ext4_ext_map_blocks(handle, inode, map, flags & |
549 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
0e855ac8 | 550 | } else { |
a4e5d88b DM |
551 | retval = ext4_ind_map_blocks(handle, inode, map, flags & |
552 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
0e855ac8 | 553 | } |
f7fec032 | 554 | if (retval > 0) { |
3be78c73 | 555 | unsigned int status; |
f7fec032 | 556 | |
44fb851d ZL |
557 | if (unlikely(retval != map->m_len)) { |
558 | ext4_warning(inode->i_sb, | |
559 | "ES len assertion failed for inode " | |
560 | "%lu: retval %d != map->m_len %d", | |
561 | inode->i_ino, retval, map->m_len); | |
562 | WARN_ON(1); | |
921f266b | 563 | } |
921f266b | 564 | |
f7fec032 ZL |
565 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
566 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; | |
567 | if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) && | |
d2dc317d | 568 | !(status & EXTENT_STATUS_WRITTEN) && |
f7fec032 ZL |
569 | ext4_find_delalloc_range(inode, map->m_lblk, |
570 | map->m_lblk + map->m_len - 1)) | |
571 | status |= EXTENT_STATUS_DELAYED; | |
572 | ret = ext4_es_insert_extent(inode, map->m_lblk, | |
573 | map->m_len, map->m_pblk, status); | |
574 | if (ret < 0) | |
575 | retval = ret; | |
576 | } | |
2dcba478 | 577 | up_read((&EXT4_I(inode)->i_data_sem)); |
f5ab0d1f | 578 | |
d100eef2 | 579 | found: |
e35fd660 | 580 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
b8a86845 | 581 | ret = check_block_validity(inode, map); |
6fd058f7 TT |
582 | if (ret != 0) |
583 | return ret; | |
584 | } | |
585 | ||
f5ab0d1f | 586 | /* If it is only a block(s) look up */ |
c2177057 | 587 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) |
f5ab0d1f MC |
588 | return retval; |
589 | ||
590 | /* | |
591 | * Returns if the blocks have already allocated | |
592 | * | |
593 | * Note that if blocks have been preallocated | |
df3ab170 | 594 | * ext4_ext_get_block() returns the create = 0 |
f5ab0d1f MC |
595 | * with buffer head unmapped. |
596 | */ | |
e35fd660 | 597 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) |
b8a86845 LC |
598 | /* |
599 | * If we need to convert extent to unwritten | |
600 | * we continue and do the actual work in | |
601 | * ext4_ext_map_blocks() | |
602 | */ | |
603 | if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) | |
604 | return retval; | |
4df3d265 | 605 | |
2a8964d6 | 606 | /* |
a25a4e1a ZL |
607 | * Here we clear m_flags because after allocating an new extent, |
608 | * it will be set again. | |
2a8964d6 | 609 | */ |
a25a4e1a | 610 | map->m_flags &= ~EXT4_MAP_FLAGS; |
2a8964d6 | 611 | |
4df3d265 | 612 | /* |
556615dc | 613 | * New blocks allocate and/or writing to unwritten extent |
f5ab0d1f | 614 | * will possibly result in updating i_data, so we take |
d91bd2c1 | 615 | * the write lock of i_data_sem, and call get_block() |
f5ab0d1f | 616 | * with create == 1 flag. |
4df3d265 | 617 | */ |
c8b459f4 | 618 | down_write(&EXT4_I(inode)->i_data_sem); |
d2a17637 | 619 | |
4df3d265 AK |
620 | /* |
621 | * We need to check for EXT4 here because migrate | |
622 | * could have changed the inode type in between | |
623 | */ | |
12e9b892 | 624 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
e35fd660 | 625 | retval = ext4_ext_map_blocks(handle, inode, map, flags); |
0e855ac8 | 626 | } else { |
e35fd660 | 627 | retval = ext4_ind_map_blocks(handle, inode, map, flags); |
267e4db9 | 628 | |
e35fd660 | 629 | if (retval > 0 && map->m_flags & EXT4_MAP_NEW) { |
267e4db9 AK |
630 | /* |
631 | * We allocated new blocks which will result in | |
632 | * i_data's format changing. Force the migrate | |
633 | * to fail by clearing migrate flags | |
634 | */ | |
19f5fb7a | 635 | ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE); |
267e4db9 | 636 | } |
d2a17637 | 637 | |
5f634d06 AK |
638 | /* |
639 | * Update reserved blocks/metadata blocks after successful | |
640 | * block allocation which had been deferred till now. We don't | |
641 | * support fallocate for non extent files. So we can update | |
642 | * reserve space here. | |
643 | */ | |
644 | if ((retval > 0) && | |
1296cc85 | 645 | (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)) |
5f634d06 AK |
646 | ext4_da_update_reserve_space(inode, retval, 1); |
647 | } | |
2ac3b6e0 | 648 | |
f7fec032 | 649 | if (retval > 0) { |
3be78c73 | 650 | unsigned int status; |
f7fec032 | 651 | |
44fb851d ZL |
652 | if (unlikely(retval != map->m_len)) { |
653 | ext4_warning(inode->i_sb, | |
654 | "ES len assertion failed for inode " | |
655 | "%lu: retval %d != map->m_len %d", | |
656 | inode->i_ino, retval, map->m_len); | |
657 | WARN_ON(1); | |
921f266b | 658 | } |
921f266b | 659 | |
c86d8db3 JK |
660 | /* |
661 | * We have to zeroout blocks before inserting them into extent | |
662 | * status tree. Otherwise someone could look them up there and | |
9b623df6 JK |
663 | * use them before they are really zeroed. We also have to |
664 | * unmap metadata before zeroing as otherwise writeback can | |
665 | * overwrite zeros with stale data from block device. | |
c86d8db3 JK |
666 | */ |
667 | if (flags & EXT4_GET_BLOCKS_ZERO && | |
668 | map->m_flags & EXT4_MAP_MAPPED && | |
669 | map->m_flags & EXT4_MAP_NEW) { | |
64e1c57f JK |
670 | clean_bdev_aliases(inode->i_sb->s_bdev, map->m_pblk, |
671 | map->m_len); | |
c86d8db3 JK |
672 | ret = ext4_issue_zeroout(inode, map->m_lblk, |
673 | map->m_pblk, map->m_len); | |
674 | if (ret) { | |
675 | retval = ret; | |
676 | goto out_sem; | |
677 | } | |
678 | } | |
679 | ||
adb23551 ZL |
680 | /* |
681 | * If the extent has been zeroed out, we don't need to update | |
682 | * extent status tree. | |
683 | */ | |
684 | if ((flags & EXT4_GET_BLOCKS_PRE_IO) && | |
685 | ext4_es_lookup_extent(inode, map->m_lblk, &es)) { | |
686 | if (ext4_es_is_written(&es)) | |
c86d8db3 | 687 | goto out_sem; |
adb23551 | 688 | } |
f7fec032 ZL |
689 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
690 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; | |
691 | if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) && | |
d2dc317d | 692 | !(status & EXTENT_STATUS_WRITTEN) && |
f7fec032 ZL |
693 | ext4_find_delalloc_range(inode, map->m_lblk, |
694 | map->m_lblk + map->m_len - 1)) | |
695 | status |= EXTENT_STATUS_DELAYED; | |
696 | ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, | |
697 | map->m_pblk, status); | |
c86d8db3 | 698 | if (ret < 0) { |
f7fec032 | 699 | retval = ret; |
c86d8db3 JK |
700 | goto out_sem; |
701 | } | |
5356f261 AK |
702 | } |
703 | ||
c86d8db3 | 704 | out_sem: |
4df3d265 | 705 | up_write((&EXT4_I(inode)->i_data_sem)); |
e35fd660 | 706 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
b8a86845 | 707 | ret = check_block_validity(inode, map); |
6fd058f7 TT |
708 | if (ret != 0) |
709 | return ret; | |
06bd3c36 JK |
710 | |
711 | /* | |
712 | * Inodes with freshly allocated blocks where contents will be | |
713 | * visible after transaction commit must be on transaction's | |
714 | * ordered data list. | |
715 | */ | |
716 | if (map->m_flags & EXT4_MAP_NEW && | |
717 | !(map->m_flags & EXT4_MAP_UNWRITTEN) && | |
718 | !(flags & EXT4_GET_BLOCKS_ZERO) && | |
02749a4c | 719 | !ext4_is_quota_file(inode) && |
06bd3c36 | 720 | ext4_should_order_data(inode)) { |
ee0876bc JK |
721 | if (flags & EXT4_GET_BLOCKS_IO_SUBMIT) |
722 | ret = ext4_jbd2_inode_add_wait(handle, inode); | |
723 | else | |
724 | ret = ext4_jbd2_inode_add_write(handle, inode); | |
06bd3c36 JK |
725 | if (ret) |
726 | return ret; | |
727 | } | |
6fd058f7 | 728 | } |
0e855ac8 AK |
729 | return retval; |
730 | } | |
731 | ||
ed8ad838 JK |
732 | /* |
733 | * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages | |
734 | * we have to be careful as someone else may be manipulating b_state as well. | |
735 | */ | |
736 | static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags) | |
737 | { | |
738 | unsigned long old_state; | |
739 | unsigned long new_state; | |
740 | ||
741 | flags &= EXT4_MAP_FLAGS; | |
742 | ||
743 | /* Dummy buffer_head? Set non-atomically. */ | |
744 | if (!bh->b_page) { | |
745 | bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags; | |
746 | return; | |
747 | } | |
748 | /* | |
749 | * Someone else may be modifying b_state. Be careful! This is ugly but | |
750 | * once we get rid of using bh as a container for mapping information | |
751 | * to pass to / from get_block functions, this can go away. | |
752 | */ | |
753 | do { | |
754 | old_state = READ_ONCE(bh->b_state); | |
755 | new_state = (old_state & ~EXT4_MAP_FLAGS) | flags; | |
756 | } while (unlikely( | |
757 | cmpxchg(&bh->b_state, old_state, new_state) != old_state)); | |
758 | } | |
759 | ||
2ed88685 TT |
760 | static int _ext4_get_block(struct inode *inode, sector_t iblock, |
761 | struct buffer_head *bh, int flags) | |
ac27a0ec | 762 | { |
2ed88685 | 763 | struct ext4_map_blocks map; |
efe70c29 | 764 | int ret = 0; |
ac27a0ec | 765 | |
46c7f254 TM |
766 | if (ext4_has_inline_data(inode)) |
767 | return -ERANGE; | |
768 | ||
2ed88685 TT |
769 | map.m_lblk = iblock; |
770 | map.m_len = bh->b_size >> inode->i_blkbits; | |
771 | ||
efe70c29 JK |
772 | ret = ext4_map_blocks(ext4_journal_current_handle(), inode, &map, |
773 | flags); | |
7fb5409d | 774 | if (ret > 0) { |
2ed88685 | 775 | map_bh(bh, inode->i_sb, map.m_pblk); |
ed8ad838 | 776 | ext4_update_bh_state(bh, map.m_flags); |
2ed88685 | 777 | bh->b_size = inode->i_sb->s_blocksize * map.m_len; |
7fb5409d | 778 | ret = 0; |
547edce3 RZ |
779 | } else if (ret == 0) { |
780 | /* hole case, need to fill in bh->b_size */ | |
781 | bh->b_size = inode->i_sb->s_blocksize * map.m_len; | |
ac27a0ec DK |
782 | } |
783 | return ret; | |
784 | } | |
785 | ||
2ed88685 TT |
786 | int ext4_get_block(struct inode *inode, sector_t iblock, |
787 | struct buffer_head *bh, int create) | |
788 | { | |
789 | return _ext4_get_block(inode, iblock, bh, | |
790 | create ? EXT4_GET_BLOCKS_CREATE : 0); | |
791 | } | |
792 | ||
705965bd JK |
793 | /* |
794 | * Get block function used when preparing for buffered write if we require | |
795 | * creating an unwritten extent if blocks haven't been allocated. The extent | |
796 | * will be converted to written after the IO is complete. | |
797 | */ | |
798 | int ext4_get_block_unwritten(struct inode *inode, sector_t iblock, | |
799 | struct buffer_head *bh_result, int create) | |
800 | { | |
801 | ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n", | |
802 | inode->i_ino, create); | |
803 | return _ext4_get_block(inode, iblock, bh_result, | |
804 | EXT4_GET_BLOCKS_IO_CREATE_EXT); | |
805 | } | |
806 | ||
efe70c29 JK |
807 | /* Maximum number of blocks we map for direct IO at once. */ |
808 | #define DIO_MAX_BLOCKS 4096 | |
809 | ||
e84dfbe2 JK |
810 | /* |
811 | * Get blocks function for the cases that need to start a transaction - | |
812 | * generally difference cases of direct IO and DAX IO. It also handles retries | |
813 | * in case of ENOSPC. | |
814 | */ | |
815 | static int ext4_get_block_trans(struct inode *inode, sector_t iblock, | |
816 | struct buffer_head *bh_result, int flags) | |
efe70c29 JK |
817 | { |
818 | int dio_credits; | |
e84dfbe2 JK |
819 | handle_t *handle; |
820 | int retries = 0; | |
821 | int ret; | |
efe70c29 JK |
822 | |
823 | /* Trim mapping request to maximum we can map at once for DIO */ | |
824 | if (bh_result->b_size >> inode->i_blkbits > DIO_MAX_BLOCKS) | |
825 | bh_result->b_size = DIO_MAX_BLOCKS << inode->i_blkbits; | |
826 | dio_credits = ext4_chunk_trans_blocks(inode, | |
827 | bh_result->b_size >> inode->i_blkbits); | |
e84dfbe2 JK |
828 | retry: |
829 | handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits); | |
830 | if (IS_ERR(handle)) | |
831 | return PTR_ERR(handle); | |
832 | ||
833 | ret = _ext4_get_block(inode, iblock, bh_result, flags); | |
834 | ext4_journal_stop(handle); | |
835 | ||
836 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) | |
837 | goto retry; | |
838 | return ret; | |
efe70c29 JK |
839 | } |
840 | ||
705965bd JK |
841 | /* Get block function for DIO reads and writes to inodes without extents */ |
842 | int ext4_dio_get_block(struct inode *inode, sector_t iblock, | |
843 | struct buffer_head *bh, int create) | |
844 | { | |
efe70c29 JK |
845 | /* We don't expect handle for direct IO */ |
846 | WARN_ON_ONCE(ext4_journal_current_handle()); | |
847 | ||
e84dfbe2 JK |
848 | if (!create) |
849 | return _ext4_get_block(inode, iblock, bh, 0); | |
850 | return ext4_get_block_trans(inode, iblock, bh, EXT4_GET_BLOCKS_CREATE); | |
705965bd JK |
851 | } |
852 | ||
853 | /* | |
109811c2 | 854 | * Get block function for AIO DIO writes when we create unwritten extent if |
705965bd JK |
855 | * blocks are not allocated yet. The extent will be converted to written |
856 | * after IO is complete. | |
857 | */ | |
109811c2 JK |
858 | static int ext4_dio_get_block_unwritten_async(struct inode *inode, |
859 | sector_t iblock, struct buffer_head *bh_result, int create) | |
705965bd | 860 | { |
efe70c29 JK |
861 | int ret; |
862 | ||
efe70c29 JK |
863 | /* We don't expect handle for direct IO */ |
864 | WARN_ON_ONCE(ext4_journal_current_handle()); | |
865 | ||
e84dfbe2 JK |
866 | ret = ext4_get_block_trans(inode, iblock, bh_result, |
867 | EXT4_GET_BLOCKS_IO_CREATE_EXT); | |
efe70c29 | 868 | |
109811c2 JK |
869 | /* |
870 | * When doing DIO using unwritten extents, we need io_end to convert | |
871 | * unwritten extents to written on IO completion. We allocate io_end | |
872 | * once we spot unwritten extent and store it in b_private. Generic | |
873 | * DIO code keeps b_private set and furthermore passes the value to | |
874 | * our completion callback in 'private' argument. | |
875 | */ | |
876 | if (!ret && buffer_unwritten(bh_result)) { | |
877 | if (!bh_result->b_private) { | |
878 | ext4_io_end_t *io_end; | |
879 | ||
880 | io_end = ext4_init_io_end(inode, GFP_KERNEL); | |
881 | if (!io_end) | |
882 | return -ENOMEM; | |
883 | bh_result->b_private = io_end; | |
884 | ext4_set_io_unwritten_flag(inode, io_end); | |
885 | } | |
efe70c29 | 886 | set_buffer_defer_completion(bh_result); |
efe70c29 JK |
887 | } |
888 | ||
889 | return ret; | |
705965bd JK |
890 | } |
891 | ||
109811c2 JK |
892 | /* |
893 | * Get block function for non-AIO DIO writes when we create unwritten extent if | |
894 | * blocks are not allocated yet. The extent will be converted to written | |
1e21196c | 895 | * after IO is complete by ext4_direct_IO_write(). |
109811c2 JK |
896 | */ |
897 | static int ext4_dio_get_block_unwritten_sync(struct inode *inode, | |
898 | sector_t iblock, struct buffer_head *bh_result, int create) | |
899 | { | |
109811c2 JK |
900 | int ret; |
901 | ||
902 | /* We don't expect handle for direct IO */ | |
903 | WARN_ON_ONCE(ext4_journal_current_handle()); | |
904 | ||
e84dfbe2 JK |
905 | ret = ext4_get_block_trans(inode, iblock, bh_result, |
906 | EXT4_GET_BLOCKS_IO_CREATE_EXT); | |
109811c2 JK |
907 | |
908 | /* | |
909 | * Mark inode as having pending DIO writes to unwritten extents. | |
1e21196c | 910 | * ext4_direct_IO_write() checks this flag and converts extents to |
109811c2 JK |
911 | * written. |
912 | */ | |
913 | if (!ret && buffer_unwritten(bh_result)) | |
914 | ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN); | |
915 | ||
916 | return ret; | |
917 | } | |
918 | ||
705965bd JK |
919 | static int ext4_dio_get_block_overwrite(struct inode *inode, sector_t iblock, |
920 | struct buffer_head *bh_result, int create) | |
921 | { | |
922 | int ret; | |
923 | ||
924 | ext4_debug("ext4_dio_get_block_overwrite: inode %lu, create flag %d\n", | |
925 | inode->i_ino, create); | |
efe70c29 JK |
926 | /* We don't expect handle for direct IO */ |
927 | WARN_ON_ONCE(ext4_journal_current_handle()); | |
928 | ||
705965bd JK |
929 | ret = _ext4_get_block(inode, iblock, bh_result, 0); |
930 | /* | |
931 | * Blocks should have been preallocated! ext4_file_write_iter() checks | |
932 | * that. | |
933 | */ | |
efe70c29 | 934 | WARN_ON_ONCE(!buffer_mapped(bh_result) || buffer_unwritten(bh_result)); |
705965bd JK |
935 | |
936 | return ret; | |
937 | } | |
938 | ||
939 | ||
ac27a0ec DK |
940 | /* |
941 | * `handle' can be NULL if create is zero | |
942 | */ | |
617ba13b | 943 | struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode, |
c5e298ae | 944 | ext4_lblk_t block, int map_flags) |
ac27a0ec | 945 | { |
2ed88685 TT |
946 | struct ext4_map_blocks map; |
947 | struct buffer_head *bh; | |
c5e298ae | 948 | int create = map_flags & EXT4_GET_BLOCKS_CREATE; |
10560082 | 949 | int err; |
ac27a0ec DK |
950 | |
951 | J_ASSERT(handle != NULL || create == 0); | |
952 | ||
2ed88685 TT |
953 | map.m_lblk = block; |
954 | map.m_len = 1; | |
c5e298ae | 955 | err = ext4_map_blocks(handle, inode, &map, map_flags); |
ac27a0ec | 956 | |
10560082 TT |
957 | if (err == 0) |
958 | return create ? ERR_PTR(-ENOSPC) : NULL; | |
2ed88685 | 959 | if (err < 0) |
10560082 | 960 | return ERR_PTR(err); |
2ed88685 TT |
961 | |
962 | bh = sb_getblk(inode->i_sb, map.m_pblk); | |
10560082 TT |
963 | if (unlikely(!bh)) |
964 | return ERR_PTR(-ENOMEM); | |
2ed88685 TT |
965 | if (map.m_flags & EXT4_MAP_NEW) { |
966 | J_ASSERT(create != 0); | |
967 | J_ASSERT(handle != NULL); | |
ac27a0ec | 968 | |
2ed88685 TT |
969 | /* |
970 | * Now that we do not always journal data, we should | |
971 | * keep in mind whether this should always journal the | |
972 | * new buffer as metadata. For now, regular file | |
973 | * writes use ext4_get_block instead, so it's not a | |
974 | * problem. | |
975 | */ | |
976 | lock_buffer(bh); | |
977 | BUFFER_TRACE(bh, "call get_create_access"); | |
10560082 TT |
978 | err = ext4_journal_get_create_access(handle, bh); |
979 | if (unlikely(err)) { | |
980 | unlock_buffer(bh); | |
981 | goto errout; | |
982 | } | |
983 | if (!buffer_uptodate(bh)) { | |
2ed88685 TT |
984 | memset(bh->b_data, 0, inode->i_sb->s_blocksize); |
985 | set_buffer_uptodate(bh); | |
ac27a0ec | 986 | } |
2ed88685 TT |
987 | unlock_buffer(bh); |
988 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); | |
989 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
10560082 TT |
990 | if (unlikely(err)) |
991 | goto errout; | |
992 | } else | |
2ed88685 | 993 | BUFFER_TRACE(bh, "not a new buffer"); |
2ed88685 | 994 | return bh; |
10560082 TT |
995 | errout: |
996 | brelse(bh); | |
997 | return ERR_PTR(err); | |
ac27a0ec DK |
998 | } |
999 | ||
617ba13b | 1000 | struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode, |
c5e298ae | 1001 | ext4_lblk_t block, int map_flags) |
ac27a0ec | 1002 | { |
af5bc92d | 1003 | struct buffer_head *bh; |
ac27a0ec | 1004 | |
c5e298ae | 1005 | bh = ext4_getblk(handle, inode, block, map_flags); |
1c215028 | 1006 | if (IS_ERR(bh)) |
ac27a0ec | 1007 | return bh; |
1c215028 | 1008 | if (!bh || buffer_uptodate(bh)) |
ac27a0ec | 1009 | return bh; |
dfec8a14 | 1010 | ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &bh); |
ac27a0ec DK |
1011 | wait_on_buffer(bh); |
1012 | if (buffer_uptodate(bh)) | |
1013 | return bh; | |
1014 | put_bh(bh); | |
1c215028 | 1015 | return ERR_PTR(-EIO); |
ac27a0ec DK |
1016 | } |
1017 | ||
9699d4f9 TE |
1018 | /* Read a contiguous batch of blocks. */ |
1019 | int ext4_bread_batch(struct inode *inode, ext4_lblk_t block, int bh_count, | |
1020 | bool wait, struct buffer_head **bhs) | |
1021 | { | |
1022 | int i, err; | |
1023 | ||
1024 | for (i = 0; i < bh_count; i++) { | |
1025 | bhs[i] = ext4_getblk(NULL, inode, block + i, 0 /* map_flags */); | |
1026 | if (IS_ERR(bhs[i])) { | |
1027 | err = PTR_ERR(bhs[i]); | |
1028 | bh_count = i; | |
1029 | goto out_brelse; | |
1030 | } | |
1031 | } | |
1032 | ||
1033 | for (i = 0; i < bh_count; i++) | |
1034 | /* Note that NULL bhs[i] is valid because of holes. */ | |
1035 | if (bhs[i] && !buffer_uptodate(bhs[i])) | |
1036 | ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, | |
1037 | &bhs[i]); | |
1038 | ||
1039 | if (!wait) | |
1040 | return 0; | |
1041 | ||
1042 | for (i = 0; i < bh_count; i++) | |
1043 | if (bhs[i]) | |
1044 | wait_on_buffer(bhs[i]); | |
1045 | ||
1046 | for (i = 0; i < bh_count; i++) { | |
1047 | if (bhs[i] && !buffer_uptodate(bhs[i])) { | |
1048 | err = -EIO; | |
1049 | goto out_brelse; | |
1050 | } | |
1051 | } | |
1052 | return 0; | |
1053 | ||
1054 | out_brelse: | |
1055 | for (i = 0; i < bh_count; i++) { | |
1056 | brelse(bhs[i]); | |
1057 | bhs[i] = NULL; | |
1058 | } | |
1059 | return err; | |
1060 | } | |
1061 | ||
f19d5870 TM |
1062 | int ext4_walk_page_buffers(handle_t *handle, |
1063 | struct buffer_head *head, | |
1064 | unsigned from, | |
1065 | unsigned to, | |
1066 | int *partial, | |
1067 | int (*fn)(handle_t *handle, | |
1068 | struct buffer_head *bh)) | |
ac27a0ec DK |
1069 | { |
1070 | struct buffer_head *bh; | |
1071 | unsigned block_start, block_end; | |
1072 | unsigned blocksize = head->b_size; | |
1073 | int err, ret = 0; | |
1074 | struct buffer_head *next; | |
1075 | ||
af5bc92d TT |
1076 | for (bh = head, block_start = 0; |
1077 | ret == 0 && (bh != head || !block_start); | |
de9a55b8 | 1078 | block_start = block_end, bh = next) { |
ac27a0ec DK |
1079 | next = bh->b_this_page; |
1080 | block_end = block_start + blocksize; | |
1081 | if (block_end <= from || block_start >= to) { | |
1082 | if (partial && !buffer_uptodate(bh)) | |
1083 | *partial = 1; | |
1084 | continue; | |
1085 | } | |
1086 | err = (*fn)(handle, bh); | |
1087 | if (!ret) | |
1088 | ret = err; | |
1089 | } | |
1090 | return ret; | |
1091 | } | |
1092 | ||
1093 | /* | |
1094 | * To preserve ordering, it is essential that the hole instantiation and | |
1095 | * the data write be encapsulated in a single transaction. We cannot | |
617ba13b | 1096 | * close off a transaction and start a new one between the ext4_get_block() |
dab291af | 1097 | * and the commit_write(). So doing the jbd2_journal_start at the start of |
ac27a0ec DK |
1098 | * prepare_write() is the right place. |
1099 | * | |
36ade451 JK |
1100 | * Also, this function can nest inside ext4_writepage(). In that case, we |
1101 | * *know* that ext4_writepage() has generated enough buffer credits to do the | |
1102 | * whole page. So we won't block on the journal in that case, which is good, | |
1103 | * because the caller may be PF_MEMALLOC. | |
ac27a0ec | 1104 | * |
617ba13b | 1105 | * By accident, ext4 can be reentered when a transaction is open via |
ac27a0ec DK |
1106 | * quota file writes. If we were to commit the transaction while thus |
1107 | * reentered, there can be a deadlock - we would be holding a quota | |
1108 | * lock, and the commit would never complete if another thread had a | |
1109 | * transaction open and was blocking on the quota lock - a ranking | |
1110 | * violation. | |
1111 | * | |
dab291af | 1112 | * So what we do is to rely on the fact that jbd2_journal_stop/journal_start |
ac27a0ec DK |
1113 | * will _not_ run commit under these circumstances because handle->h_ref |
1114 | * is elevated. We'll still have enough credits for the tiny quotafile | |
1115 | * write. | |
1116 | */ | |
f19d5870 TM |
1117 | int do_journal_get_write_access(handle_t *handle, |
1118 | struct buffer_head *bh) | |
ac27a0ec | 1119 | { |
56d35a4c JK |
1120 | int dirty = buffer_dirty(bh); |
1121 | int ret; | |
1122 | ||
ac27a0ec DK |
1123 | if (!buffer_mapped(bh) || buffer_freed(bh)) |
1124 | return 0; | |
56d35a4c | 1125 | /* |
ebdec241 | 1126 | * __block_write_begin() could have dirtied some buffers. Clean |
56d35a4c JK |
1127 | * the dirty bit as jbd2_journal_get_write_access() could complain |
1128 | * otherwise about fs integrity issues. Setting of the dirty bit | |
ebdec241 | 1129 | * by __block_write_begin() isn't a real problem here as we clear |
56d35a4c JK |
1130 | * the bit before releasing a page lock and thus writeback cannot |
1131 | * ever write the buffer. | |
1132 | */ | |
1133 | if (dirty) | |
1134 | clear_buffer_dirty(bh); | |
5d601255 | 1135 | BUFFER_TRACE(bh, "get write access"); |
56d35a4c JK |
1136 | ret = ext4_journal_get_write_access(handle, bh); |
1137 | if (!ret && dirty) | |
1138 | ret = ext4_handle_dirty_metadata(handle, NULL, bh); | |
1139 | return ret; | |
ac27a0ec DK |
1140 | } |
1141 | ||
2058f83a MH |
1142 | #ifdef CONFIG_EXT4_FS_ENCRYPTION |
1143 | static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len, | |
1144 | get_block_t *get_block) | |
1145 | { | |
09cbfeaf | 1146 | unsigned from = pos & (PAGE_SIZE - 1); |
2058f83a MH |
1147 | unsigned to = from + len; |
1148 | struct inode *inode = page->mapping->host; | |
1149 | unsigned block_start, block_end; | |
1150 | sector_t block; | |
1151 | int err = 0; | |
1152 | unsigned blocksize = inode->i_sb->s_blocksize; | |
1153 | unsigned bbits; | |
1154 | struct buffer_head *bh, *head, *wait[2], **wait_bh = wait; | |
1155 | bool decrypt = false; | |
1156 | ||
1157 | BUG_ON(!PageLocked(page)); | |
09cbfeaf KS |
1158 | BUG_ON(from > PAGE_SIZE); |
1159 | BUG_ON(to > PAGE_SIZE); | |
2058f83a MH |
1160 | BUG_ON(from > to); |
1161 | ||
1162 | if (!page_has_buffers(page)) | |
1163 | create_empty_buffers(page, blocksize, 0); | |
1164 | head = page_buffers(page); | |
1165 | bbits = ilog2(blocksize); | |
09cbfeaf | 1166 | block = (sector_t)page->index << (PAGE_SHIFT - bbits); |
2058f83a MH |
1167 | |
1168 | for (bh = head, block_start = 0; bh != head || !block_start; | |
1169 | block++, block_start = block_end, bh = bh->b_this_page) { | |
1170 | block_end = block_start + blocksize; | |
1171 | if (block_end <= from || block_start >= to) { | |
1172 | if (PageUptodate(page)) { | |
1173 | if (!buffer_uptodate(bh)) | |
1174 | set_buffer_uptodate(bh); | |
1175 | } | |
1176 | continue; | |
1177 | } | |
1178 | if (buffer_new(bh)) | |
1179 | clear_buffer_new(bh); | |
1180 | if (!buffer_mapped(bh)) { | |
1181 | WARN_ON(bh->b_size != blocksize); | |
1182 | err = get_block(inode, block, bh, 1); | |
1183 | if (err) | |
1184 | break; | |
1185 | if (buffer_new(bh)) { | |
e64855c6 | 1186 | clean_bdev_bh_alias(bh); |
2058f83a MH |
1187 | if (PageUptodate(page)) { |
1188 | clear_buffer_new(bh); | |
1189 | set_buffer_uptodate(bh); | |
1190 | mark_buffer_dirty(bh); | |
1191 | continue; | |
1192 | } | |
1193 | if (block_end > to || block_start < from) | |
1194 | zero_user_segments(page, to, block_end, | |
1195 | block_start, from); | |
1196 | continue; | |
1197 | } | |
1198 | } | |
1199 | if (PageUptodate(page)) { | |
1200 | if (!buffer_uptodate(bh)) | |
1201 | set_buffer_uptodate(bh); | |
1202 | continue; | |
1203 | } | |
1204 | if (!buffer_uptodate(bh) && !buffer_delay(bh) && | |
1205 | !buffer_unwritten(bh) && | |
1206 | (block_start < from || block_end > to)) { | |
dfec8a14 | 1207 | ll_rw_block(REQ_OP_READ, 0, 1, &bh); |
2058f83a MH |
1208 | *wait_bh++ = bh; |
1209 | decrypt = ext4_encrypted_inode(inode) && | |
1210 | S_ISREG(inode->i_mode); | |
1211 | } | |
1212 | } | |
1213 | /* | |
1214 | * If we issued read requests, let them complete. | |
1215 | */ | |
1216 | while (wait_bh > wait) { | |
1217 | wait_on_buffer(*--wait_bh); | |
1218 | if (!buffer_uptodate(*wait_bh)) | |
1219 | err = -EIO; | |
1220 | } | |
1221 | if (unlikely(err)) | |
1222 | page_zero_new_buffers(page, from, to); | |
1223 | else if (decrypt) | |
7821d4dd | 1224 | err = fscrypt_decrypt_page(page->mapping->host, page, |
9c4bb8a3 | 1225 | PAGE_SIZE, 0, page->index); |
2058f83a MH |
1226 | return err; |
1227 | } | |
1228 | #endif | |
1229 | ||
bfc1af65 | 1230 | static int ext4_write_begin(struct file *file, struct address_space *mapping, |
de9a55b8 TT |
1231 | loff_t pos, unsigned len, unsigned flags, |
1232 | struct page **pagep, void **fsdata) | |
ac27a0ec | 1233 | { |
af5bc92d | 1234 | struct inode *inode = mapping->host; |
1938a150 | 1235 | int ret, needed_blocks; |
ac27a0ec DK |
1236 | handle_t *handle; |
1237 | int retries = 0; | |
af5bc92d | 1238 | struct page *page; |
de9a55b8 | 1239 | pgoff_t index; |
af5bc92d | 1240 | unsigned from, to; |
bfc1af65 | 1241 | |
0db1ff22 TT |
1242 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
1243 | return -EIO; | |
1244 | ||
9bffad1e | 1245 | trace_ext4_write_begin(inode, pos, len, flags); |
1938a150 AK |
1246 | /* |
1247 | * Reserve one block more for addition to orphan list in case | |
1248 | * we allocate blocks but write fails for some reason | |
1249 | */ | |
1250 | needed_blocks = ext4_writepage_trans_blocks(inode) + 1; | |
09cbfeaf KS |
1251 | index = pos >> PAGE_SHIFT; |
1252 | from = pos & (PAGE_SIZE - 1); | |
af5bc92d | 1253 | to = from + len; |
ac27a0ec | 1254 | |
f19d5870 TM |
1255 | if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { |
1256 | ret = ext4_try_to_write_inline_data(mapping, inode, pos, len, | |
1257 | flags, pagep); | |
1258 | if (ret < 0) | |
47564bfb TT |
1259 | return ret; |
1260 | if (ret == 1) | |
1261 | return 0; | |
f19d5870 TM |
1262 | } |
1263 | ||
47564bfb TT |
1264 | /* |
1265 | * grab_cache_page_write_begin() can take a long time if the | |
1266 | * system is thrashing due to memory pressure, or if the page | |
1267 | * is being written back. So grab it first before we start | |
1268 | * the transaction handle. This also allows us to allocate | |
1269 | * the page (if needed) without using GFP_NOFS. | |
1270 | */ | |
1271 | retry_grab: | |
1272 | page = grab_cache_page_write_begin(mapping, index, flags); | |
1273 | if (!page) | |
1274 | return -ENOMEM; | |
1275 | unlock_page(page); | |
1276 | ||
1277 | retry_journal: | |
9924a92a | 1278 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks); |
af5bc92d | 1279 | if (IS_ERR(handle)) { |
09cbfeaf | 1280 | put_page(page); |
47564bfb | 1281 | return PTR_ERR(handle); |
7479d2b9 | 1282 | } |
ac27a0ec | 1283 | |
47564bfb TT |
1284 | lock_page(page); |
1285 | if (page->mapping != mapping) { | |
1286 | /* The page got truncated from under us */ | |
1287 | unlock_page(page); | |
09cbfeaf | 1288 | put_page(page); |
cf108bca | 1289 | ext4_journal_stop(handle); |
47564bfb | 1290 | goto retry_grab; |
cf108bca | 1291 | } |
7afe5aa5 DM |
1292 | /* In case writeback began while the page was unlocked */ |
1293 | wait_for_stable_page(page); | |
cf108bca | 1294 | |
2058f83a MH |
1295 | #ifdef CONFIG_EXT4_FS_ENCRYPTION |
1296 | if (ext4_should_dioread_nolock(inode)) | |
1297 | ret = ext4_block_write_begin(page, pos, len, | |
705965bd | 1298 | ext4_get_block_unwritten); |
2058f83a MH |
1299 | else |
1300 | ret = ext4_block_write_begin(page, pos, len, | |
1301 | ext4_get_block); | |
1302 | #else | |
744692dc | 1303 | if (ext4_should_dioread_nolock(inode)) |
705965bd JK |
1304 | ret = __block_write_begin(page, pos, len, |
1305 | ext4_get_block_unwritten); | |
744692dc | 1306 | else |
6e1db88d | 1307 | ret = __block_write_begin(page, pos, len, ext4_get_block); |
2058f83a | 1308 | #endif |
bfc1af65 | 1309 | if (!ret && ext4_should_journal_data(inode)) { |
f19d5870 TM |
1310 | ret = ext4_walk_page_buffers(handle, page_buffers(page), |
1311 | from, to, NULL, | |
1312 | do_journal_get_write_access); | |
ac27a0ec | 1313 | } |
bfc1af65 NP |
1314 | |
1315 | if (ret) { | |
af5bc92d | 1316 | unlock_page(page); |
ae4d5372 | 1317 | /* |
6e1db88d | 1318 | * __block_write_begin may have instantiated a few blocks |
ae4d5372 AK |
1319 | * outside i_size. Trim these off again. Don't need |
1320 | * i_size_read because we hold i_mutex. | |
1938a150 AK |
1321 | * |
1322 | * Add inode to orphan list in case we crash before | |
1323 | * truncate finishes | |
ae4d5372 | 1324 | */ |
ffacfa7a | 1325 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
1938a150 AK |
1326 | ext4_orphan_add(handle, inode); |
1327 | ||
1328 | ext4_journal_stop(handle); | |
1329 | if (pos + len > inode->i_size) { | |
b9a4207d | 1330 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1331 | /* |
ffacfa7a | 1332 | * If truncate failed early the inode might |
1938a150 AK |
1333 | * still be on the orphan list; we need to |
1334 | * make sure the inode is removed from the | |
1335 | * orphan list in that case. | |
1336 | */ | |
1337 | if (inode->i_nlink) | |
1338 | ext4_orphan_del(NULL, inode); | |
1339 | } | |
bfc1af65 | 1340 | |
47564bfb TT |
1341 | if (ret == -ENOSPC && |
1342 | ext4_should_retry_alloc(inode->i_sb, &retries)) | |
1343 | goto retry_journal; | |
09cbfeaf | 1344 | put_page(page); |
47564bfb TT |
1345 | return ret; |
1346 | } | |
1347 | *pagep = page; | |
ac27a0ec DK |
1348 | return ret; |
1349 | } | |
1350 | ||
bfc1af65 NP |
1351 | /* For write_end() in data=journal mode */ |
1352 | static int write_end_fn(handle_t *handle, struct buffer_head *bh) | |
ac27a0ec | 1353 | { |
13fca323 | 1354 | int ret; |
ac27a0ec DK |
1355 | if (!buffer_mapped(bh) || buffer_freed(bh)) |
1356 | return 0; | |
1357 | set_buffer_uptodate(bh); | |
13fca323 TT |
1358 | ret = ext4_handle_dirty_metadata(handle, NULL, bh); |
1359 | clear_buffer_meta(bh); | |
1360 | clear_buffer_prio(bh); | |
1361 | return ret; | |
ac27a0ec DK |
1362 | } |
1363 | ||
eed4333f ZL |
1364 | /* |
1365 | * We need to pick up the new inode size which generic_commit_write gave us | |
1366 | * `file' can be NULL - eg, when called from page_symlink(). | |
1367 | * | |
1368 | * ext4 never places buffers on inode->i_mapping->private_list. metadata | |
1369 | * buffers are managed internally. | |
1370 | */ | |
1371 | static int ext4_write_end(struct file *file, | |
1372 | struct address_space *mapping, | |
1373 | loff_t pos, unsigned len, unsigned copied, | |
1374 | struct page *page, void *fsdata) | |
f8514083 | 1375 | { |
f8514083 | 1376 | handle_t *handle = ext4_journal_current_handle(); |
eed4333f | 1377 | struct inode *inode = mapping->host; |
0572639f | 1378 | loff_t old_size = inode->i_size; |
eed4333f ZL |
1379 | int ret = 0, ret2; |
1380 | int i_size_changed = 0; | |
1381 | ||
1382 | trace_ext4_write_end(inode, pos, len, copied); | |
42c832de TT |
1383 | if (ext4_has_inline_data(inode)) { |
1384 | ret = ext4_write_inline_data_end(inode, pos, len, | |
1385 | copied, page); | |
eb5efbcb TT |
1386 | if (ret < 0) { |
1387 | unlock_page(page); | |
1388 | put_page(page); | |
42c832de | 1389 | goto errout; |
eb5efbcb | 1390 | } |
42c832de TT |
1391 | copied = ret; |
1392 | } else | |
f19d5870 TM |
1393 | copied = block_write_end(file, mapping, pos, |
1394 | len, copied, page, fsdata); | |
f8514083 | 1395 | /* |
4631dbf6 | 1396 | * it's important to update i_size while still holding page lock: |
f8514083 AK |
1397 | * page writeout could otherwise come in and zero beyond i_size. |
1398 | */ | |
4631dbf6 | 1399 | i_size_changed = ext4_update_inode_size(inode, pos + copied); |
f8514083 | 1400 | unlock_page(page); |
09cbfeaf | 1401 | put_page(page); |
f8514083 | 1402 | |
0572639f XW |
1403 | if (old_size < pos) |
1404 | pagecache_isize_extended(inode, old_size, pos); | |
f8514083 AK |
1405 | /* |
1406 | * Don't mark the inode dirty under page lock. First, it unnecessarily | |
1407 | * makes the holding time of page lock longer. Second, it forces lock | |
1408 | * ordering of page lock and transaction start for journaling | |
1409 | * filesystems. | |
1410 | */ | |
1411 | if (i_size_changed) | |
1412 | ext4_mark_inode_dirty(handle, inode); | |
1413 | ||
ffacfa7a | 1414 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
f8514083 AK |
1415 | /* if we have allocated more blocks and copied |
1416 | * less. We will have blocks allocated outside | |
1417 | * inode->i_size. So truncate them | |
1418 | */ | |
1419 | ext4_orphan_add(handle, inode); | |
74d553aa | 1420 | errout: |
617ba13b | 1421 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1422 | if (!ret) |
1423 | ret = ret2; | |
bfc1af65 | 1424 | |
f8514083 | 1425 | if (pos + len > inode->i_size) { |
b9a4207d | 1426 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1427 | /* |
ffacfa7a | 1428 | * If truncate failed early the inode might still be |
f8514083 AK |
1429 | * on the orphan list; we need to make sure the inode |
1430 | * is removed from the orphan list in that case. | |
1431 | */ | |
1432 | if (inode->i_nlink) | |
1433 | ext4_orphan_del(NULL, inode); | |
1434 | } | |
1435 | ||
bfc1af65 | 1436 | return ret ? ret : copied; |
ac27a0ec DK |
1437 | } |
1438 | ||
b90197b6 TT |
1439 | /* |
1440 | * This is a private version of page_zero_new_buffers() which doesn't | |
1441 | * set the buffer to be dirty, since in data=journalled mode we need | |
1442 | * to call ext4_handle_dirty_metadata() instead. | |
1443 | */ | |
3b136499 JK |
1444 | static void ext4_journalled_zero_new_buffers(handle_t *handle, |
1445 | struct page *page, | |
1446 | unsigned from, unsigned to) | |
b90197b6 TT |
1447 | { |
1448 | unsigned int block_start = 0, block_end; | |
1449 | struct buffer_head *head, *bh; | |
1450 | ||
1451 | bh = head = page_buffers(page); | |
1452 | do { | |
1453 | block_end = block_start + bh->b_size; | |
1454 | if (buffer_new(bh)) { | |
1455 | if (block_end > from && block_start < to) { | |
1456 | if (!PageUptodate(page)) { | |
1457 | unsigned start, size; | |
1458 | ||
1459 | start = max(from, block_start); | |
1460 | size = min(to, block_end) - start; | |
1461 | ||
1462 | zero_user(page, start, size); | |
3b136499 | 1463 | write_end_fn(handle, bh); |
b90197b6 TT |
1464 | } |
1465 | clear_buffer_new(bh); | |
1466 | } | |
1467 | } | |
1468 | block_start = block_end; | |
1469 | bh = bh->b_this_page; | |
1470 | } while (bh != head); | |
1471 | } | |
1472 | ||
bfc1af65 | 1473 | static int ext4_journalled_write_end(struct file *file, |
de9a55b8 TT |
1474 | struct address_space *mapping, |
1475 | loff_t pos, unsigned len, unsigned copied, | |
1476 | struct page *page, void *fsdata) | |
ac27a0ec | 1477 | { |
617ba13b | 1478 | handle_t *handle = ext4_journal_current_handle(); |
bfc1af65 | 1479 | struct inode *inode = mapping->host; |
0572639f | 1480 | loff_t old_size = inode->i_size; |
ac27a0ec DK |
1481 | int ret = 0, ret2; |
1482 | int partial = 0; | |
bfc1af65 | 1483 | unsigned from, to; |
4631dbf6 | 1484 | int size_changed = 0; |
ac27a0ec | 1485 | |
9bffad1e | 1486 | trace_ext4_journalled_write_end(inode, pos, len, copied); |
09cbfeaf | 1487 | from = pos & (PAGE_SIZE - 1); |
bfc1af65 NP |
1488 | to = from + len; |
1489 | ||
441c8508 CW |
1490 | BUG_ON(!ext4_handle_valid(handle)); |
1491 | ||
eb5efbcb TT |
1492 | if (ext4_has_inline_data(inode)) { |
1493 | ret = ext4_write_inline_data_end(inode, pos, len, | |
1494 | copied, page); | |
1495 | if (ret < 0) { | |
1496 | unlock_page(page); | |
1497 | put_page(page); | |
1498 | goto errout; | |
1499 | } | |
1500 | copied = ret; | |
1501 | } else if (unlikely(copied < len) && !PageUptodate(page)) { | |
3b136499 JK |
1502 | copied = 0; |
1503 | ext4_journalled_zero_new_buffers(handle, page, from, to); | |
1504 | } else { | |
1505 | if (unlikely(copied < len)) | |
1506 | ext4_journalled_zero_new_buffers(handle, page, | |
1507 | from + copied, to); | |
3fdcfb66 | 1508 | ret = ext4_walk_page_buffers(handle, page_buffers(page), from, |
3b136499 JK |
1509 | from + copied, &partial, |
1510 | write_end_fn); | |
3fdcfb66 TM |
1511 | if (!partial) |
1512 | SetPageUptodate(page); | |
1513 | } | |
4631dbf6 | 1514 | size_changed = ext4_update_inode_size(inode, pos + copied); |
19f5fb7a | 1515 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
2d859db3 | 1516 | EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; |
4631dbf6 | 1517 | unlock_page(page); |
09cbfeaf | 1518 | put_page(page); |
4631dbf6 | 1519 | |
0572639f XW |
1520 | if (old_size < pos) |
1521 | pagecache_isize_extended(inode, old_size, pos); | |
1522 | ||
4631dbf6 | 1523 | if (size_changed) { |
617ba13b | 1524 | ret2 = ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
1525 | if (!ret) |
1526 | ret = ret2; | |
1527 | } | |
bfc1af65 | 1528 | |
ffacfa7a | 1529 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
f8514083 AK |
1530 | /* if we have allocated more blocks and copied |
1531 | * less. We will have blocks allocated outside | |
1532 | * inode->i_size. So truncate them | |
1533 | */ | |
1534 | ext4_orphan_add(handle, inode); | |
1535 | ||
eb5efbcb | 1536 | errout: |
617ba13b | 1537 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1538 | if (!ret) |
1539 | ret = ret2; | |
f8514083 | 1540 | if (pos + len > inode->i_size) { |
b9a4207d | 1541 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1542 | /* |
ffacfa7a | 1543 | * If truncate failed early the inode might still be |
f8514083 AK |
1544 | * on the orphan list; we need to make sure the inode |
1545 | * is removed from the orphan list in that case. | |
1546 | */ | |
1547 | if (inode->i_nlink) | |
1548 | ext4_orphan_del(NULL, inode); | |
1549 | } | |
bfc1af65 NP |
1550 | |
1551 | return ret ? ret : copied; | |
ac27a0ec | 1552 | } |
d2a17637 | 1553 | |
9d0be502 | 1554 | /* |
c27e43a1 | 1555 | * Reserve space for a single cluster |
9d0be502 | 1556 | */ |
c27e43a1 | 1557 | static int ext4_da_reserve_space(struct inode *inode) |
d2a17637 | 1558 | { |
60e58e0f | 1559 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
0637c6f4 | 1560 | struct ext4_inode_info *ei = EXT4_I(inode); |
5dd4056d | 1561 | int ret; |
03179fe9 TT |
1562 | |
1563 | /* | |
1564 | * We will charge metadata quota at writeout time; this saves | |
1565 | * us from metadata over-estimation, though we may go over by | |
1566 | * a small amount in the end. Here we just reserve for data. | |
1567 | */ | |
1568 | ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1)); | |
1569 | if (ret) | |
1570 | return ret; | |
d2a17637 | 1571 | |
0637c6f4 | 1572 | spin_lock(&ei->i_block_reservation_lock); |
71d4f7d0 | 1573 | if (ext4_claim_free_clusters(sbi, 1, 0)) { |
03179fe9 | 1574 | spin_unlock(&ei->i_block_reservation_lock); |
03179fe9 | 1575 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1)); |
d2a17637 MC |
1576 | return -ENOSPC; |
1577 | } | |
9d0be502 | 1578 | ei->i_reserved_data_blocks++; |
c27e43a1 | 1579 | trace_ext4_da_reserve_space(inode); |
0637c6f4 | 1580 | spin_unlock(&ei->i_block_reservation_lock); |
39bc680a | 1581 | |
d2a17637 MC |
1582 | return 0; /* success */ |
1583 | } | |
1584 | ||
12219aea | 1585 | static void ext4_da_release_space(struct inode *inode, int to_free) |
d2a17637 MC |
1586 | { |
1587 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0637c6f4 | 1588 | struct ext4_inode_info *ei = EXT4_I(inode); |
d2a17637 | 1589 | |
cd213226 MC |
1590 | if (!to_free) |
1591 | return; /* Nothing to release, exit */ | |
1592 | ||
d2a17637 | 1593 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); |
cd213226 | 1594 | |
5a58ec87 | 1595 | trace_ext4_da_release_space(inode, to_free); |
0637c6f4 | 1596 | if (unlikely(to_free > ei->i_reserved_data_blocks)) { |
cd213226 | 1597 | /* |
0637c6f4 TT |
1598 | * if there aren't enough reserved blocks, then the |
1599 | * counter is messed up somewhere. Since this | |
1600 | * function is called from invalidate page, it's | |
1601 | * harmless to return without any action. | |
cd213226 | 1602 | */ |
8de5c325 | 1603 | ext4_warning(inode->i_sb, "ext4_da_release_space: " |
0637c6f4 | 1604 | "ino %lu, to_free %d with only %d reserved " |
1084f252 | 1605 | "data blocks", inode->i_ino, to_free, |
0637c6f4 TT |
1606 | ei->i_reserved_data_blocks); |
1607 | WARN_ON(1); | |
1608 | to_free = ei->i_reserved_data_blocks; | |
cd213226 | 1609 | } |
0637c6f4 | 1610 | ei->i_reserved_data_blocks -= to_free; |
cd213226 | 1611 | |
72b8ab9d | 1612 | /* update fs dirty data blocks counter */ |
57042651 | 1613 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free); |
d2a17637 | 1614 | |
d2a17637 | 1615 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
60e58e0f | 1616 | |
7b415bf6 | 1617 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free)); |
d2a17637 MC |
1618 | } |
1619 | ||
1620 | static void ext4_da_page_release_reservation(struct page *page, | |
ca99fdd2 LC |
1621 | unsigned int offset, |
1622 | unsigned int length) | |
d2a17637 | 1623 | { |
9705acd6 | 1624 | int to_release = 0, contiguous_blks = 0; |
d2a17637 MC |
1625 | struct buffer_head *head, *bh; |
1626 | unsigned int curr_off = 0; | |
7b415bf6 AK |
1627 | struct inode *inode = page->mapping->host; |
1628 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
ca99fdd2 | 1629 | unsigned int stop = offset + length; |
7b415bf6 | 1630 | int num_clusters; |
51865fda | 1631 | ext4_fsblk_t lblk; |
d2a17637 | 1632 | |
09cbfeaf | 1633 | BUG_ON(stop > PAGE_SIZE || stop < length); |
ca99fdd2 | 1634 | |
d2a17637 MC |
1635 | head = page_buffers(page); |
1636 | bh = head; | |
1637 | do { | |
1638 | unsigned int next_off = curr_off + bh->b_size; | |
1639 | ||
ca99fdd2 LC |
1640 | if (next_off > stop) |
1641 | break; | |
1642 | ||
d2a17637 MC |
1643 | if ((offset <= curr_off) && (buffer_delay(bh))) { |
1644 | to_release++; | |
9705acd6 | 1645 | contiguous_blks++; |
d2a17637 | 1646 | clear_buffer_delay(bh); |
9705acd6 LC |
1647 | } else if (contiguous_blks) { |
1648 | lblk = page->index << | |
09cbfeaf | 1649 | (PAGE_SHIFT - inode->i_blkbits); |
9705acd6 LC |
1650 | lblk += (curr_off >> inode->i_blkbits) - |
1651 | contiguous_blks; | |
1652 | ext4_es_remove_extent(inode, lblk, contiguous_blks); | |
1653 | contiguous_blks = 0; | |
d2a17637 MC |
1654 | } |
1655 | curr_off = next_off; | |
1656 | } while ((bh = bh->b_this_page) != head); | |
7b415bf6 | 1657 | |
9705acd6 | 1658 | if (contiguous_blks) { |
09cbfeaf | 1659 | lblk = page->index << (PAGE_SHIFT - inode->i_blkbits); |
9705acd6 LC |
1660 | lblk += (curr_off >> inode->i_blkbits) - contiguous_blks; |
1661 | ext4_es_remove_extent(inode, lblk, contiguous_blks); | |
51865fda ZL |
1662 | } |
1663 | ||
7b415bf6 AK |
1664 | /* If we have released all the blocks belonging to a cluster, then we |
1665 | * need to release the reserved space for that cluster. */ | |
1666 | num_clusters = EXT4_NUM_B2C(sbi, to_release); | |
1667 | while (num_clusters > 0) { | |
09cbfeaf | 1668 | lblk = (page->index << (PAGE_SHIFT - inode->i_blkbits)) + |
7b415bf6 AK |
1669 | ((num_clusters - 1) << sbi->s_cluster_bits); |
1670 | if (sbi->s_cluster_ratio == 1 || | |
7d1b1fbc | 1671 | !ext4_find_delalloc_cluster(inode, lblk)) |
7b415bf6 AK |
1672 | ext4_da_release_space(inode, 1); |
1673 | ||
1674 | num_clusters--; | |
1675 | } | |
d2a17637 | 1676 | } |
ac27a0ec | 1677 | |
64769240 AT |
1678 | /* |
1679 | * Delayed allocation stuff | |
1680 | */ | |
1681 | ||
4e7ea81d JK |
1682 | struct mpage_da_data { |
1683 | struct inode *inode; | |
1684 | struct writeback_control *wbc; | |
6b523df4 | 1685 | |
4e7ea81d JK |
1686 | pgoff_t first_page; /* The first page to write */ |
1687 | pgoff_t next_page; /* Current page to examine */ | |
1688 | pgoff_t last_page; /* Last page to examine */ | |
791b7f08 | 1689 | /* |
4e7ea81d JK |
1690 | * Extent to map - this can be after first_page because that can be |
1691 | * fully mapped. We somewhat abuse m_flags to store whether the extent | |
1692 | * is delalloc or unwritten. | |
791b7f08 | 1693 | */ |
4e7ea81d JK |
1694 | struct ext4_map_blocks map; |
1695 | struct ext4_io_submit io_submit; /* IO submission data */ | |
dddbd6ac | 1696 | unsigned int do_map:1; |
4e7ea81d | 1697 | }; |
64769240 | 1698 | |
4e7ea81d JK |
1699 | static void mpage_release_unused_pages(struct mpage_da_data *mpd, |
1700 | bool invalidate) | |
c4a0c46e AK |
1701 | { |
1702 | int nr_pages, i; | |
1703 | pgoff_t index, end; | |
1704 | struct pagevec pvec; | |
1705 | struct inode *inode = mpd->inode; | |
1706 | struct address_space *mapping = inode->i_mapping; | |
4e7ea81d JK |
1707 | |
1708 | /* This is necessary when next_page == 0. */ | |
1709 | if (mpd->first_page >= mpd->next_page) | |
1710 | return; | |
c4a0c46e | 1711 | |
c7f5938a CW |
1712 | index = mpd->first_page; |
1713 | end = mpd->next_page - 1; | |
4e7ea81d JK |
1714 | if (invalidate) { |
1715 | ext4_lblk_t start, last; | |
09cbfeaf KS |
1716 | start = index << (PAGE_SHIFT - inode->i_blkbits); |
1717 | last = end << (PAGE_SHIFT - inode->i_blkbits); | |
4e7ea81d JK |
1718 | ext4_es_remove_extent(inode, start, last - start + 1); |
1719 | } | |
51865fda | 1720 | |
66bea92c | 1721 | pagevec_init(&pvec, 0); |
c4a0c46e AK |
1722 | while (index <= end) { |
1723 | nr_pages = pagevec_lookup(&pvec, mapping, index, PAGEVEC_SIZE); | |
1724 | if (nr_pages == 0) | |
1725 | break; | |
1726 | for (i = 0; i < nr_pages; i++) { | |
1727 | struct page *page = pvec.pages[i]; | |
9b1d0998 | 1728 | if (page->index > end) |
c4a0c46e | 1729 | break; |
c4a0c46e AK |
1730 | BUG_ON(!PageLocked(page)); |
1731 | BUG_ON(PageWriteback(page)); | |
4e7ea81d | 1732 | if (invalidate) { |
4e800c03 | 1733 | if (page_mapped(page)) |
1734 | clear_page_dirty_for_io(page); | |
09cbfeaf | 1735 | block_invalidatepage(page, 0, PAGE_SIZE); |
4e7ea81d JK |
1736 | ClearPageUptodate(page); |
1737 | } | |
c4a0c46e AK |
1738 | unlock_page(page); |
1739 | } | |
9b1d0998 JK |
1740 | index = pvec.pages[nr_pages - 1]->index + 1; |
1741 | pagevec_release(&pvec); | |
c4a0c46e | 1742 | } |
c4a0c46e AK |
1743 | } |
1744 | ||
df22291f AK |
1745 | static void ext4_print_free_blocks(struct inode *inode) |
1746 | { | |
1747 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
92b97816 | 1748 | struct super_block *sb = inode->i_sb; |
f78ee70d | 1749 | struct ext4_inode_info *ei = EXT4_I(inode); |
92b97816 TT |
1750 | |
1751 | ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld", | |
5dee5437 | 1752 | EXT4_C2B(EXT4_SB(inode->i_sb), |
f78ee70d | 1753 | ext4_count_free_clusters(sb))); |
92b97816 TT |
1754 | ext4_msg(sb, KERN_CRIT, "Free/Dirty block details"); |
1755 | ext4_msg(sb, KERN_CRIT, "free_blocks=%lld", | |
f78ee70d | 1756 | (long long) EXT4_C2B(EXT4_SB(sb), |
57042651 | 1757 | percpu_counter_sum(&sbi->s_freeclusters_counter))); |
92b97816 | 1758 | ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld", |
f78ee70d | 1759 | (long long) EXT4_C2B(EXT4_SB(sb), |
7b415bf6 | 1760 | percpu_counter_sum(&sbi->s_dirtyclusters_counter))); |
92b97816 TT |
1761 | ext4_msg(sb, KERN_CRIT, "Block reservation details"); |
1762 | ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u", | |
f78ee70d | 1763 | ei->i_reserved_data_blocks); |
df22291f AK |
1764 | return; |
1765 | } | |
1766 | ||
c364b22c | 1767 | static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh) |
29fa89d0 | 1768 | { |
c364b22c | 1769 | return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh); |
29fa89d0 AK |
1770 | } |
1771 | ||
5356f261 AK |
1772 | /* |
1773 | * This function is grabs code from the very beginning of | |
1774 | * ext4_map_blocks, but assumes that the caller is from delayed write | |
1775 | * time. This function looks up the requested blocks and sets the | |
1776 | * buffer delay bit under the protection of i_data_sem. | |
1777 | */ | |
1778 | static int ext4_da_map_blocks(struct inode *inode, sector_t iblock, | |
1779 | struct ext4_map_blocks *map, | |
1780 | struct buffer_head *bh) | |
1781 | { | |
d100eef2 | 1782 | struct extent_status es; |
5356f261 AK |
1783 | int retval; |
1784 | sector_t invalid_block = ~((sector_t) 0xffff); | |
921f266b DM |
1785 | #ifdef ES_AGGRESSIVE_TEST |
1786 | struct ext4_map_blocks orig_map; | |
1787 | ||
1788 | memcpy(&orig_map, map, sizeof(*map)); | |
1789 | #endif | |
5356f261 AK |
1790 | |
1791 | if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es)) | |
1792 | invalid_block = ~0; | |
1793 | ||
1794 | map->m_flags = 0; | |
1795 | ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u," | |
1796 | "logical block %lu\n", inode->i_ino, map->m_len, | |
1797 | (unsigned long) map->m_lblk); | |
d100eef2 ZL |
1798 | |
1799 | /* Lookup extent status tree firstly */ | |
1800 | if (ext4_es_lookup_extent(inode, iblock, &es)) { | |
d100eef2 ZL |
1801 | if (ext4_es_is_hole(&es)) { |
1802 | retval = 0; | |
c8b459f4 | 1803 | down_read(&EXT4_I(inode)->i_data_sem); |
d100eef2 ZL |
1804 | goto add_delayed; |
1805 | } | |
1806 | ||
1807 | /* | |
1808 | * Delayed extent could be allocated by fallocate. | |
1809 | * So we need to check it. | |
1810 | */ | |
1811 | if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) { | |
1812 | map_bh(bh, inode->i_sb, invalid_block); | |
1813 | set_buffer_new(bh); | |
1814 | set_buffer_delay(bh); | |
1815 | return 0; | |
1816 | } | |
1817 | ||
1818 | map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk; | |
1819 | retval = es.es_len - (iblock - es.es_lblk); | |
1820 | if (retval > map->m_len) | |
1821 | retval = map->m_len; | |
1822 | map->m_len = retval; | |
1823 | if (ext4_es_is_written(&es)) | |
1824 | map->m_flags |= EXT4_MAP_MAPPED; | |
1825 | else if (ext4_es_is_unwritten(&es)) | |
1826 | map->m_flags |= EXT4_MAP_UNWRITTEN; | |
1827 | else | |
1828 | BUG_ON(1); | |
1829 | ||
921f266b DM |
1830 | #ifdef ES_AGGRESSIVE_TEST |
1831 | ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0); | |
1832 | #endif | |
d100eef2 ZL |
1833 | return retval; |
1834 | } | |
1835 | ||
5356f261 AK |
1836 | /* |
1837 | * Try to see if we can get the block without requesting a new | |
1838 | * file system block. | |
1839 | */ | |
c8b459f4 | 1840 | down_read(&EXT4_I(inode)->i_data_sem); |
cbd7584e | 1841 | if (ext4_has_inline_data(inode)) |
9c3569b5 | 1842 | retval = 0; |
cbd7584e | 1843 | else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
2f8e0a7c | 1844 | retval = ext4_ext_map_blocks(NULL, inode, map, 0); |
5356f261 | 1845 | else |
2f8e0a7c | 1846 | retval = ext4_ind_map_blocks(NULL, inode, map, 0); |
5356f261 | 1847 | |
d100eef2 | 1848 | add_delayed: |
5356f261 | 1849 | if (retval == 0) { |
f7fec032 | 1850 | int ret; |
5356f261 AK |
1851 | /* |
1852 | * XXX: __block_prepare_write() unmaps passed block, | |
1853 | * is it OK? | |
1854 | */ | |
386ad67c LC |
1855 | /* |
1856 | * If the block was allocated from previously allocated cluster, | |
1857 | * then we don't need to reserve it again. However we still need | |
1858 | * to reserve metadata for every block we're going to write. | |
1859 | */ | |
c27e43a1 | 1860 | if (EXT4_SB(inode->i_sb)->s_cluster_ratio == 1 || |
cbd7584e | 1861 | !ext4_find_delalloc_cluster(inode, map->m_lblk)) { |
c27e43a1 | 1862 | ret = ext4_da_reserve_space(inode); |
f7fec032 | 1863 | if (ret) { |
5356f261 | 1864 | /* not enough space to reserve */ |
f7fec032 | 1865 | retval = ret; |
5356f261 | 1866 | goto out_unlock; |
f7fec032 | 1867 | } |
5356f261 AK |
1868 | } |
1869 | ||
f7fec032 ZL |
1870 | ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, |
1871 | ~0, EXTENT_STATUS_DELAYED); | |
1872 | if (ret) { | |
1873 | retval = ret; | |
51865fda | 1874 | goto out_unlock; |
f7fec032 | 1875 | } |
51865fda | 1876 | |
5356f261 AK |
1877 | map_bh(bh, inode->i_sb, invalid_block); |
1878 | set_buffer_new(bh); | |
1879 | set_buffer_delay(bh); | |
f7fec032 ZL |
1880 | } else if (retval > 0) { |
1881 | int ret; | |
3be78c73 | 1882 | unsigned int status; |
f7fec032 | 1883 | |
44fb851d ZL |
1884 | if (unlikely(retval != map->m_len)) { |
1885 | ext4_warning(inode->i_sb, | |
1886 | "ES len assertion failed for inode " | |
1887 | "%lu: retval %d != map->m_len %d", | |
1888 | inode->i_ino, retval, map->m_len); | |
1889 | WARN_ON(1); | |
921f266b | 1890 | } |
921f266b | 1891 | |
f7fec032 ZL |
1892 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
1893 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; | |
1894 | ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, | |
1895 | map->m_pblk, status); | |
1896 | if (ret != 0) | |
1897 | retval = ret; | |
5356f261 AK |
1898 | } |
1899 | ||
1900 | out_unlock: | |
1901 | up_read((&EXT4_I(inode)->i_data_sem)); | |
1902 | ||
1903 | return retval; | |
1904 | } | |
1905 | ||
64769240 | 1906 | /* |
d91bd2c1 | 1907 | * This is a special get_block_t callback which is used by |
b920c755 TT |
1908 | * ext4_da_write_begin(). It will either return mapped block or |
1909 | * reserve space for a single block. | |
29fa89d0 AK |
1910 | * |
1911 | * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set. | |
1912 | * We also have b_blocknr = -1 and b_bdev initialized properly | |
1913 | * | |
1914 | * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set. | |
1915 | * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev | |
1916 | * initialized properly. | |
64769240 | 1917 | */ |
9c3569b5 TM |
1918 | int ext4_da_get_block_prep(struct inode *inode, sector_t iblock, |
1919 | struct buffer_head *bh, int create) | |
64769240 | 1920 | { |
2ed88685 | 1921 | struct ext4_map_blocks map; |
64769240 AT |
1922 | int ret = 0; |
1923 | ||
1924 | BUG_ON(create == 0); | |
2ed88685 TT |
1925 | BUG_ON(bh->b_size != inode->i_sb->s_blocksize); |
1926 | ||
1927 | map.m_lblk = iblock; | |
1928 | map.m_len = 1; | |
64769240 AT |
1929 | |
1930 | /* | |
1931 | * first, we need to know whether the block is allocated already | |
1932 | * preallocated blocks are unmapped but should treated | |
1933 | * the same as allocated blocks. | |
1934 | */ | |
5356f261 AK |
1935 | ret = ext4_da_map_blocks(inode, iblock, &map, bh); |
1936 | if (ret <= 0) | |
2ed88685 | 1937 | return ret; |
64769240 | 1938 | |
2ed88685 | 1939 | map_bh(bh, inode->i_sb, map.m_pblk); |
ed8ad838 | 1940 | ext4_update_bh_state(bh, map.m_flags); |
2ed88685 TT |
1941 | |
1942 | if (buffer_unwritten(bh)) { | |
1943 | /* A delayed write to unwritten bh should be marked | |
1944 | * new and mapped. Mapped ensures that we don't do | |
1945 | * get_block multiple times when we write to the same | |
1946 | * offset and new ensures that we do proper zero out | |
1947 | * for partial write. | |
1948 | */ | |
1949 | set_buffer_new(bh); | |
c8205636 | 1950 | set_buffer_mapped(bh); |
2ed88685 TT |
1951 | } |
1952 | return 0; | |
64769240 | 1953 | } |
61628a3f | 1954 | |
62e086be AK |
1955 | static int bget_one(handle_t *handle, struct buffer_head *bh) |
1956 | { | |
1957 | get_bh(bh); | |
1958 | return 0; | |
1959 | } | |
1960 | ||
1961 | static int bput_one(handle_t *handle, struct buffer_head *bh) | |
1962 | { | |
1963 | put_bh(bh); | |
1964 | return 0; | |
1965 | } | |
1966 | ||
1967 | static int __ext4_journalled_writepage(struct page *page, | |
62e086be AK |
1968 | unsigned int len) |
1969 | { | |
1970 | struct address_space *mapping = page->mapping; | |
1971 | struct inode *inode = mapping->host; | |
3fdcfb66 | 1972 | struct buffer_head *page_bufs = NULL; |
62e086be | 1973 | handle_t *handle = NULL; |
3fdcfb66 TM |
1974 | int ret = 0, err = 0; |
1975 | int inline_data = ext4_has_inline_data(inode); | |
1976 | struct buffer_head *inode_bh = NULL; | |
62e086be | 1977 | |
cb20d518 | 1978 | ClearPageChecked(page); |
3fdcfb66 TM |
1979 | |
1980 | if (inline_data) { | |
1981 | BUG_ON(page->index != 0); | |
1982 | BUG_ON(len > ext4_get_max_inline_size(inode)); | |
1983 | inode_bh = ext4_journalled_write_inline_data(inode, len, page); | |
1984 | if (inode_bh == NULL) | |
1985 | goto out; | |
1986 | } else { | |
1987 | page_bufs = page_buffers(page); | |
1988 | if (!page_bufs) { | |
1989 | BUG(); | |
1990 | goto out; | |
1991 | } | |
1992 | ext4_walk_page_buffers(handle, page_bufs, 0, len, | |
1993 | NULL, bget_one); | |
1994 | } | |
bdf96838 TT |
1995 | /* |
1996 | * We need to release the page lock before we start the | |
1997 | * journal, so grab a reference so the page won't disappear | |
1998 | * out from under us. | |
1999 | */ | |
2000 | get_page(page); | |
62e086be AK |
2001 | unlock_page(page); |
2002 | ||
9924a92a TT |
2003 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
2004 | ext4_writepage_trans_blocks(inode)); | |
62e086be AK |
2005 | if (IS_ERR(handle)) { |
2006 | ret = PTR_ERR(handle); | |
bdf96838 TT |
2007 | put_page(page); |
2008 | goto out_no_pagelock; | |
62e086be | 2009 | } |
441c8508 CW |
2010 | BUG_ON(!ext4_handle_valid(handle)); |
2011 | ||
bdf96838 TT |
2012 | lock_page(page); |
2013 | put_page(page); | |
2014 | if (page->mapping != mapping) { | |
2015 | /* The page got truncated from under us */ | |
2016 | ext4_journal_stop(handle); | |
2017 | ret = 0; | |
2018 | goto out; | |
2019 | } | |
2020 | ||
3fdcfb66 | 2021 | if (inline_data) { |
5d601255 | 2022 | BUFFER_TRACE(inode_bh, "get write access"); |
3fdcfb66 | 2023 | ret = ext4_journal_get_write_access(handle, inode_bh); |
62e086be | 2024 | |
3fdcfb66 TM |
2025 | err = ext4_handle_dirty_metadata(handle, inode, inode_bh); |
2026 | ||
2027 | } else { | |
2028 | ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL, | |
2029 | do_journal_get_write_access); | |
2030 | ||
2031 | err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL, | |
2032 | write_end_fn); | |
2033 | } | |
62e086be AK |
2034 | if (ret == 0) |
2035 | ret = err; | |
2d859db3 | 2036 | EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; |
62e086be AK |
2037 | err = ext4_journal_stop(handle); |
2038 | if (!ret) | |
2039 | ret = err; | |
2040 | ||
3fdcfb66 | 2041 | if (!ext4_has_inline_data(inode)) |
8c9367fd | 2042 | ext4_walk_page_buffers(NULL, page_bufs, 0, len, |
3fdcfb66 | 2043 | NULL, bput_one); |
19f5fb7a | 2044 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
62e086be | 2045 | out: |
bdf96838 TT |
2046 | unlock_page(page); |
2047 | out_no_pagelock: | |
3fdcfb66 | 2048 | brelse(inode_bh); |
62e086be AK |
2049 | return ret; |
2050 | } | |
2051 | ||
61628a3f | 2052 | /* |
43ce1d23 AK |
2053 | * Note that we don't need to start a transaction unless we're journaling data |
2054 | * because we should have holes filled from ext4_page_mkwrite(). We even don't | |
2055 | * need to file the inode to the transaction's list in ordered mode because if | |
2056 | * we are writing back data added by write(), the inode is already there and if | |
25985edc | 2057 | * we are writing back data modified via mmap(), no one guarantees in which |
43ce1d23 AK |
2058 | * transaction the data will hit the disk. In case we are journaling data, we |
2059 | * cannot start transaction directly because transaction start ranks above page | |
2060 | * lock so we have to do some magic. | |
2061 | * | |
b920c755 | 2062 | * This function can get called via... |
20970ba6 | 2063 | * - ext4_writepages after taking page lock (have journal handle) |
b920c755 | 2064 | * - journal_submit_inode_data_buffers (no journal handle) |
f6463b0d | 2065 | * - shrink_page_list via the kswapd/direct reclaim (no journal handle) |
b920c755 | 2066 | * - grab_page_cache when doing write_begin (have journal handle) |
43ce1d23 AK |
2067 | * |
2068 | * We don't do any block allocation in this function. If we have page with | |
2069 | * multiple blocks we need to write those buffer_heads that are mapped. This | |
2070 | * is important for mmaped based write. So if we do with blocksize 1K | |
2071 | * truncate(f, 1024); | |
2072 | * a = mmap(f, 0, 4096); | |
2073 | * a[0] = 'a'; | |
2074 | * truncate(f, 4096); | |
2075 | * we have in the page first buffer_head mapped via page_mkwrite call back | |
90802ed9 | 2076 | * but other buffer_heads would be unmapped but dirty (dirty done via the |
43ce1d23 AK |
2077 | * do_wp_page). So writepage should write the first block. If we modify |
2078 | * the mmap area beyond 1024 we will again get a page_fault and the | |
2079 | * page_mkwrite callback will do the block allocation and mark the | |
2080 | * buffer_heads mapped. | |
2081 | * | |
2082 | * We redirty the page if we have any buffer_heads that is either delay or | |
2083 | * unwritten in the page. | |
2084 | * | |
2085 | * We can get recursively called as show below. | |
2086 | * | |
2087 | * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() -> | |
2088 | * ext4_writepage() | |
2089 | * | |
2090 | * But since we don't do any block allocation we should not deadlock. | |
2091 | * Page also have the dirty flag cleared so we don't get recurive page_lock. | |
61628a3f | 2092 | */ |
43ce1d23 | 2093 | static int ext4_writepage(struct page *page, |
62e086be | 2094 | struct writeback_control *wbc) |
64769240 | 2095 | { |
f8bec370 | 2096 | int ret = 0; |
61628a3f | 2097 | loff_t size; |
498e5f24 | 2098 | unsigned int len; |
744692dc | 2099 | struct buffer_head *page_bufs = NULL; |
61628a3f | 2100 | struct inode *inode = page->mapping->host; |
36ade451 | 2101 | struct ext4_io_submit io_submit; |
1c8349a1 | 2102 | bool keep_towrite = false; |
61628a3f | 2103 | |
0db1ff22 TT |
2104 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) { |
2105 | ext4_invalidatepage(page, 0, PAGE_SIZE); | |
2106 | unlock_page(page); | |
2107 | return -EIO; | |
2108 | } | |
2109 | ||
a9c667f8 | 2110 | trace_ext4_writepage(page); |
f0e6c985 | 2111 | size = i_size_read(inode); |
09cbfeaf KS |
2112 | if (page->index == size >> PAGE_SHIFT) |
2113 | len = size & ~PAGE_MASK; | |
f0e6c985 | 2114 | else |
09cbfeaf | 2115 | len = PAGE_SIZE; |
64769240 | 2116 | |
a42afc5f | 2117 | page_bufs = page_buffers(page); |
a42afc5f | 2118 | /* |
fe386132 JK |
2119 | * We cannot do block allocation or other extent handling in this |
2120 | * function. If there are buffers needing that, we have to redirty | |
2121 | * the page. But we may reach here when we do a journal commit via | |
2122 | * journal_submit_inode_data_buffers() and in that case we must write | |
2123 | * allocated buffers to achieve data=ordered mode guarantees. | |
cccd147a TT |
2124 | * |
2125 | * Also, if there is only one buffer per page (the fs block | |
2126 | * size == the page size), if one buffer needs block | |
2127 | * allocation or needs to modify the extent tree to clear the | |
2128 | * unwritten flag, we know that the page can't be written at | |
2129 | * all, so we might as well refuse the write immediately. | |
2130 | * Unfortunately if the block size != page size, we can't as | |
2131 | * easily detect this case using ext4_walk_page_buffers(), but | |
2132 | * for the extremely common case, this is an optimization that | |
2133 | * skips a useless round trip through ext4_bio_write_page(). | |
a42afc5f | 2134 | */ |
f19d5870 TM |
2135 | if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL, |
2136 | ext4_bh_delay_or_unwritten)) { | |
f8bec370 | 2137 | redirty_page_for_writepage(wbc, page); |
cccd147a | 2138 | if ((current->flags & PF_MEMALLOC) || |
09cbfeaf | 2139 | (inode->i_sb->s_blocksize == PAGE_SIZE)) { |
fe386132 JK |
2140 | /* |
2141 | * For memory cleaning there's no point in writing only | |
2142 | * some buffers. So just bail out. Warn if we came here | |
2143 | * from direct reclaim. | |
2144 | */ | |
2145 | WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) | |
2146 | == PF_MEMALLOC); | |
f0e6c985 AK |
2147 | unlock_page(page); |
2148 | return 0; | |
2149 | } | |
1c8349a1 | 2150 | keep_towrite = true; |
a42afc5f | 2151 | } |
64769240 | 2152 | |
cb20d518 | 2153 | if (PageChecked(page) && ext4_should_journal_data(inode)) |
43ce1d23 AK |
2154 | /* |
2155 | * It's mmapped pagecache. Add buffers and journal it. There | |
2156 | * doesn't seem much point in redirtying the page here. | |
2157 | */ | |
3f0ca309 | 2158 | return __ext4_journalled_writepage(page, len); |
43ce1d23 | 2159 | |
97a851ed JK |
2160 | ext4_io_submit_init(&io_submit, wbc); |
2161 | io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS); | |
2162 | if (!io_submit.io_end) { | |
2163 | redirty_page_for_writepage(wbc, page); | |
2164 | unlock_page(page); | |
2165 | return -ENOMEM; | |
2166 | } | |
1c8349a1 | 2167 | ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite); |
36ade451 | 2168 | ext4_io_submit(&io_submit); |
97a851ed JK |
2169 | /* Drop io_end reference we got from init */ |
2170 | ext4_put_io_end_defer(io_submit.io_end); | |
64769240 AT |
2171 | return ret; |
2172 | } | |
2173 | ||
5f1132b2 JK |
2174 | static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page) |
2175 | { | |
2176 | int len; | |
a056bdaa | 2177 | loff_t size; |
5f1132b2 JK |
2178 | int err; |
2179 | ||
2180 | BUG_ON(page->index != mpd->first_page); | |
a056bdaa JK |
2181 | clear_page_dirty_for_io(page); |
2182 | /* | |
2183 | * We have to be very careful here! Nothing protects writeback path | |
2184 | * against i_size changes and the page can be writeably mapped into | |
2185 | * page tables. So an application can be growing i_size and writing | |
2186 | * data through mmap while writeback runs. clear_page_dirty_for_io() | |
2187 | * write-protects our page in page tables and the page cannot get | |
2188 | * written to again until we release page lock. So only after | |
2189 | * clear_page_dirty_for_io() we are safe to sample i_size for | |
2190 | * ext4_bio_write_page() to zero-out tail of the written page. We rely | |
2191 | * on the barrier provided by TestClearPageDirty in | |
2192 | * clear_page_dirty_for_io() to make sure i_size is really sampled only | |
2193 | * after page tables are updated. | |
2194 | */ | |
2195 | size = i_size_read(mpd->inode); | |
09cbfeaf KS |
2196 | if (page->index == size >> PAGE_SHIFT) |
2197 | len = size & ~PAGE_MASK; | |
5f1132b2 | 2198 | else |
09cbfeaf | 2199 | len = PAGE_SIZE; |
1c8349a1 | 2200 | err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false); |
5f1132b2 JK |
2201 | if (!err) |
2202 | mpd->wbc->nr_to_write--; | |
2203 | mpd->first_page++; | |
2204 | ||
2205 | return err; | |
2206 | } | |
2207 | ||
4e7ea81d JK |
2208 | #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay)) |
2209 | ||
61628a3f | 2210 | /* |
fffb2739 JK |
2211 | * mballoc gives us at most this number of blocks... |
2212 | * XXX: That seems to be only a limitation of ext4_mb_normalize_request(). | |
70261f56 | 2213 | * The rest of mballoc seems to handle chunks up to full group size. |
61628a3f | 2214 | */ |
fffb2739 | 2215 | #define MAX_WRITEPAGES_EXTENT_LEN 2048 |
525f4ed8 | 2216 | |
4e7ea81d JK |
2217 | /* |
2218 | * mpage_add_bh_to_extent - try to add bh to extent of blocks to map | |
2219 | * | |
2220 | * @mpd - extent of blocks | |
2221 | * @lblk - logical number of the block in the file | |
09930042 | 2222 | * @bh - buffer head we want to add to the extent |
4e7ea81d | 2223 | * |
09930042 JK |
2224 | * The function is used to collect contig. blocks in the same state. If the |
2225 | * buffer doesn't require mapping for writeback and we haven't started the | |
2226 | * extent of buffers to map yet, the function returns 'true' immediately - the | |
2227 | * caller can write the buffer right away. Otherwise the function returns true | |
2228 | * if the block has been added to the extent, false if the block couldn't be | |
2229 | * added. | |
4e7ea81d | 2230 | */ |
09930042 JK |
2231 | static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk, |
2232 | struct buffer_head *bh) | |
4e7ea81d JK |
2233 | { |
2234 | struct ext4_map_blocks *map = &mpd->map; | |
2235 | ||
09930042 JK |
2236 | /* Buffer that doesn't need mapping for writeback? */ |
2237 | if (!buffer_dirty(bh) || !buffer_mapped(bh) || | |
2238 | (!buffer_delay(bh) && !buffer_unwritten(bh))) { | |
2239 | /* So far no extent to map => we write the buffer right away */ | |
2240 | if (map->m_len == 0) | |
2241 | return true; | |
2242 | return false; | |
2243 | } | |
4e7ea81d JK |
2244 | |
2245 | /* First block in the extent? */ | |
2246 | if (map->m_len == 0) { | |
dddbd6ac JK |
2247 | /* We cannot map unless handle is started... */ |
2248 | if (!mpd->do_map) | |
2249 | return false; | |
4e7ea81d JK |
2250 | map->m_lblk = lblk; |
2251 | map->m_len = 1; | |
09930042 JK |
2252 | map->m_flags = bh->b_state & BH_FLAGS; |
2253 | return true; | |
4e7ea81d JK |
2254 | } |
2255 | ||
09930042 JK |
2256 | /* Don't go larger than mballoc is willing to allocate */ |
2257 | if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN) | |
2258 | return false; | |
2259 | ||
4e7ea81d JK |
2260 | /* Can we merge the block to our big extent? */ |
2261 | if (lblk == map->m_lblk + map->m_len && | |
09930042 | 2262 | (bh->b_state & BH_FLAGS) == map->m_flags) { |
4e7ea81d | 2263 | map->m_len++; |
09930042 | 2264 | return true; |
4e7ea81d | 2265 | } |
09930042 | 2266 | return false; |
4e7ea81d JK |
2267 | } |
2268 | ||
5f1132b2 JK |
2269 | /* |
2270 | * mpage_process_page_bufs - submit page buffers for IO or add them to extent | |
2271 | * | |
2272 | * @mpd - extent of blocks for mapping | |
2273 | * @head - the first buffer in the page | |
2274 | * @bh - buffer we should start processing from | |
2275 | * @lblk - logical number of the block in the file corresponding to @bh | |
2276 | * | |
2277 | * Walk through page buffers from @bh upto @head (exclusive) and either submit | |
2278 | * the page for IO if all buffers in this page were mapped and there's no | |
2279 | * accumulated extent of buffers to map or add buffers in the page to the | |
2280 | * extent of buffers to map. The function returns 1 if the caller can continue | |
2281 | * by processing the next page, 0 if it should stop adding buffers to the | |
2282 | * extent to map because we cannot extend it anymore. It can also return value | |
2283 | * < 0 in case of error during IO submission. | |
2284 | */ | |
2285 | static int mpage_process_page_bufs(struct mpage_da_data *mpd, | |
2286 | struct buffer_head *head, | |
2287 | struct buffer_head *bh, | |
2288 | ext4_lblk_t lblk) | |
4e7ea81d JK |
2289 | { |
2290 | struct inode *inode = mpd->inode; | |
5f1132b2 | 2291 | int err; |
93407472 | 2292 | ext4_lblk_t blocks = (i_size_read(inode) + i_blocksize(inode) - 1) |
4e7ea81d JK |
2293 | >> inode->i_blkbits; |
2294 | ||
2295 | do { | |
2296 | BUG_ON(buffer_locked(bh)); | |
2297 | ||
09930042 | 2298 | if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) { |
4e7ea81d JK |
2299 | /* Found extent to map? */ |
2300 | if (mpd->map.m_len) | |
5f1132b2 | 2301 | return 0; |
dddbd6ac JK |
2302 | /* Buffer needs mapping and handle is not started? */ |
2303 | if (!mpd->do_map) | |
2304 | return 0; | |
09930042 | 2305 | /* Everything mapped so far and we hit EOF */ |
5f1132b2 | 2306 | break; |
4e7ea81d | 2307 | } |
4e7ea81d | 2308 | } while (lblk++, (bh = bh->b_this_page) != head); |
5f1132b2 JK |
2309 | /* So far everything mapped? Submit the page for IO. */ |
2310 | if (mpd->map.m_len == 0) { | |
2311 | err = mpage_submit_page(mpd, head->b_page); | |
2312 | if (err < 0) | |
2313 | return err; | |
2314 | } | |
2315 | return lblk < blocks; | |
4e7ea81d JK |
2316 | } |
2317 | ||
2318 | /* | |
2319 | * mpage_map_buffers - update buffers corresponding to changed extent and | |
2320 | * submit fully mapped pages for IO | |
2321 | * | |
2322 | * @mpd - description of extent to map, on return next extent to map | |
2323 | * | |
2324 | * Scan buffers corresponding to changed extent (we expect corresponding pages | |
2325 | * to be already locked) and update buffer state according to new extent state. | |
2326 | * We map delalloc buffers to their physical location, clear unwritten bits, | |
556615dc | 2327 | * and mark buffers as uninit when we perform writes to unwritten extents |
4e7ea81d JK |
2328 | * and do extent conversion after IO is finished. If the last page is not fully |
2329 | * mapped, we update @map to the next extent in the last page that needs | |
2330 | * mapping. Otherwise we submit the page for IO. | |
2331 | */ | |
2332 | static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd) | |
2333 | { | |
2334 | struct pagevec pvec; | |
2335 | int nr_pages, i; | |
2336 | struct inode *inode = mpd->inode; | |
2337 | struct buffer_head *head, *bh; | |
09cbfeaf | 2338 | int bpp_bits = PAGE_SHIFT - inode->i_blkbits; |
4e7ea81d JK |
2339 | pgoff_t start, end; |
2340 | ext4_lblk_t lblk; | |
2341 | sector_t pblock; | |
2342 | int err; | |
2343 | ||
2344 | start = mpd->map.m_lblk >> bpp_bits; | |
2345 | end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits; | |
2346 | lblk = start << bpp_bits; | |
2347 | pblock = mpd->map.m_pblk; | |
2348 | ||
2349 | pagevec_init(&pvec, 0); | |
2350 | while (start <= end) { | |
2351 | nr_pages = pagevec_lookup(&pvec, inode->i_mapping, start, | |
2352 | PAGEVEC_SIZE); | |
2353 | if (nr_pages == 0) | |
2354 | break; | |
2355 | for (i = 0; i < nr_pages; i++) { | |
2356 | struct page *page = pvec.pages[i]; | |
2357 | ||
2358 | if (page->index > end) | |
2359 | break; | |
70261f56 | 2360 | /* Up to 'end' pages must be contiguous */ |
4e7ea81d JK |
2361 | BUG_ON(page->index != start); |
2362 | bh = head = page_buffers(page); | |
2363 | do { | |
2364 | if (lblk < mpd->map.m_lblk) | |
2365 | continue; | |
2366 | if (lblk >= mpd->map.m_lblk + mpd->map.m_len) { | |
2367 | /* | |
2368 | * Buffer after end of mapped extent. | |
2369 | * Find next buffer in the page to map. | |
2370 | */ | |
2371 | mpd->map.m_len = 0; | |
2372 | mpd->map.m_flags = 0; | |
5f1132b2 JK |
2373 | /* |
2374 | * FIXME: If dioread_nolock supports | |
2375 | * blocksize < pagesize, we need to make | |
2376 | * sure we add size mapped so far to | |
2377 | * io_end->size as the following call | |
2378 | * can submit the page for IO. | |
2379 | */ | |
2380 | err = mpage_process_page_bufs(mpd, head, | |
2381 | bh, lblk); | |
4e7ea81d | 2382 | pagevec_release(&pvec); |
5f1132b2 JK |
2383 | if (err > 0) |
2384 | err = 0; | |
2385 | return err; | |
4e7ea81d JK |
2386 | } |
2387 | if (buffer_delay(bh)) { | |
2388 | clear_buffer_delay(bh); | |
2389 | bh->b_blocknr = pblock++; | |
2390 | } | |
4e7ea81d | 2391 | clear_buffer_unwritten(bh); |
5f1132b2 | 2392 | } while (lblk++, (bh = bh->b_this_page) != head); |
4e7ea81d JK |
2393 | |
2394 | /* | |
2395 | * FIXME: This is going to break if dioread_nolock | |
2396 | * supports blocksize < pagesize as we will try to | |
2397 | * convert potentially unmapped parts of inode. | |
2398 | */ | |
09cbfeaf | 2399 | mpd->io_submit.io_end->size += PAGE_SIZE; |
4e7ea81d JK |
2400 | /* Page fully mapped - let IO run! */ |
2401 | err = mpage_submit_page(mpd, page); | |
2402 | if (err < 0) { | |
2403 | pagevec_release(&pvec); | |
2404 | return err; | |
2405 | } | |
2406 | start++; | |
2407 | } | |
2408 | pagevec_release(&pvec); | |
2409 | } | |
2410 | /* Extent fully mapped and matches with page boundary. We are done. */ | |
2411 | mpd->map.m_len = 0; | |
2412 | mpd->map.m_flags = 0; | |
2413 | return 0; | |
2414 | } | |
2415 | ||
2416 | static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd) | |
2417 | { | |
2418 | struct inode *inode = mpd->inode; | |
2419 | struct ext4_map_blocks *map = &mpd->map; | |
2420 | int get_blocks_flags; | |
090f32ee | 2421 | int err, dioread_nolock; |
4e7ea81d JK |
2422 | |
2423 | trace_ext4_da_write_pages_extent(inode, map); | |
2424 | /* | |
2425 | * Call ext4_map_blocks() to allocate any delayed allocation blocks, or | |
556615dc | 2426 | * to convert an unwritten extent to be initialized (in the case |
4e7ea81d JK |
2427 | * where we have written into one or more preallocated blocks). It is |
2428 | * possible that we're going to need more metadata blocks than | |
2429 | * previously reserved. However we must not fail because we're in | |
2430 | * writeback and there is nothing we can do about it so it might result | |
2431 | * in data loss. So use reserved blocks to allocate metadata if | |
2432 | * possible. | |
2433 | * | |
754cfed6 TT |
2434 | * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if |
2435 | * the blocks in question are delalloc blocks. This indicates | |
2436 | * that the blocks and quotas has already been checked when | |
2437 | * the data was copied into the page cache. | |
4e7ea81d JK |
2438 | */ |
2439 | get_blocks_flags = EXT4_GET_BLOCKS_CREATE | | |
ee0876bc JK |
2440 | EXT4_GET_BLOCKS_METADATA_NOFAIL | |
2441 | EXT4_GET_BLOCKS_IO_SUBMIT; | |
090f32ee LC |
2442 | dioread_nolock = ext4_should_dioread_nolock(inode); |
2443 | if (dioread_nolock) | |
4e7ea81d JK |
2444 | get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT; |
2445 | if (map->m_flags & (1 << BH_Delay)) | |
2446 | get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE; | |
2447 | ||
2448 | err = ext4_map_blocks(handle, inode, map, get_blocks_flags); | |
2449 | if (err < 0) | |
2450 | return err; | |
090f32ee | 2451 | if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) { |
6b523df4 JK |
2452 | if (!mpd->io_submit.io_end->handle && |
2453 | ext4_handle_valid(handle)) { | |
2454 | mpd->io_submit.io_end->handle = handle->h_rsv_handle; | |
2455 | handle->h_rsv_handle = NULL; | |
2456 | } | |
3613d228 | 2457 | ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end); |
6b523df4 | 2458 | } |
4e7ea81d JK |
2459 | |
2460 | BUG_ON(map->m_len == 0); | |
2461 | if (map->m_flags & EXT4_MAP_NEW) { | |
64e1c57f JK |
2462 | clean_bdev_aliases(inode->i_sb->s_bdev, map->m_pblk, |
2463 | map->m_len); | |
4e7ea81d JK |
2464 | } |
2465 | return 0; | |
2466 | } | |
2467 | ||
2468 | /* | |
2469 | * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length | |
2470 | * mpd->len and submit pages underlying it for IO | |
2471 | * | |
2472 | * @handle - handle for journal operations | |
2473 | * @mpd - extent to map | |
7534e854 JK |
2474 | * @give_up_on_write - we set this to true iff there is a fatal error and there |
2475 | * is no hope of writing the data. The caller should discard | |
2476 | * dirty pages to avoid infinite loops. | |
4e7ea81d JK |
2477 | * |
2478 | * The function maps extent starting at mpd->lblk of length mpd->len. If it is | |
2479 | * delayed, blocks are allocated, if it is unwritten, we may need to convert | |
2480 | * them to initialized or split the described range from larger unwritten | |
2481 | * extent. Note that we need not map all the described range since allocation | |
2482 | * can return less blocks or the range is covered by more unwritten extents. We | |
2483 | * cannot map more because we are limited by reserved transaction credits. On | |
2484 | * the other hand we always make sure that the last touched page is fully | |
2485 | * mapped so that it can be written out (and thus forward progress is | |
2486 | * guaranteed). After mapping we submit all mapped pages for IO. | |
2487 | */ | |
2488 | static int mpage_map_and_submit_extent(handle_t *handle, | |
cb530541 TT |
2489 | struct mpage_da_data *mpd, |
2490 | bool *give_up_on_write) | |
4e7ea81d JK |
2491 | { |
2492 | struct inode *inode = mpd->inode; | |
2493 | struct ext4_map_blocks *map = &mpd->map; | |
2494 | int err; | |
2495 | loff_t disksize; | |
6603120e | 2496 | int progress = 0; |
4e7ea81d JK |
2497 | |
2498 | mpd->io_submit.io_end->offset = | |
2499 | ((loff_t)map->m_lblk) << inode->i_blkbits; | |
27d7c4ed | 2500 | do { |
4e7ea81d JK |
2501 | err = mpage_map_one_extent(handle, mpd); |
2502 | if (err < 0) { | |
2503 | struct super_block *sb = inode->i_sb; | |
2504 | ||
0db1ff22 TT |
2505 | if (ext4_forced_shutdown(EXT4_SB(sb)) || |
2506 | EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED) | |
cb530541 | 2507 | goto invalidate_dirty_pages; |
4e7ea81d | 2508 | /* |
cb530541 TT |
2509 | * Let the uper layers retry transient errors. |
2510 | * In the case of ENOSPC, if ext4_count_free_blocks() | |
2511 | * is non-zero, a commit should free up blocks. | |
4e7ea81d | 2512 | */ |
cb530541 | 2513 | if ((err == -ENOMEM) || |
6603120e DM |
2514 | (err == -ENOSPC && ext4_count_free_clusters(sb))) { |
2515 | if (progress) | |
2516 | goto update_disksize; | |
cb530541 | 2517 | return err; |
6603120e | 2518 | } |
cb530541 TT |
2519 | ext4_msg(sb, KERN_CRIT, |
2520 | "Delayed block allocation failed for " | |
2521 | "inode %lu at logical offset %llu with" | |
2522 | " max blocks %u with error %d", | |
2523 | inode->i_ino, | |
2524 | (unsigned long long)map->m_lblk, | |
2525 | (unsigned)map->m_len, -err); | |
2526 | ext4_msg(sb, KERN_CRIT, | |
2527 | "This should not happen!! Data will " | |
2528 | "be lost\n"); | |
2529 | if (err == -ENOSPC) | |
2530 | ext4_print_free_blocks(inode); | |
2531 | invalidate_dirty_pages: | |
2532 | *give_up_on_write = true; | |
4e7ea81d JK |
2533 | return err; |
2534 | } | |
6603120e | 2535 | progress = 1; |
4e7ea81d JK |
2536 | /* |
2537 | * Update buffer state, submit mapped pages, and get us new | |
2538 | * extent to map | |
2539 | */ | |
2540 | err = mpage_map_and_submit_buffers(mpd); | |
2541 | if (err < 0) | |
6603120e | 2542 | goto update_disksize; |
27d7c4ed | 2543 | } while (map->m_len); |
4e7ea81d | 2544 | |
6603120e | 2545 | update_disksize: |
622cad13 TT |
2546 | /* |
2547 | * Update on-disk size after IO is submitted. Races with | |
2548 | * truncate are avoided by checking i_size under i_data_sem. | |
2549 | */ | |
09cbfeaf | 2550 | disksize = ((loff_t)mpd->first_page) << PAGE_SHIFT; |
4e7ea81d JK |
2551 | if (disksize > EXT4_I(inode)->i_disksize) { |
2552 | int err2; | |
622cad13 TT |
2553 | loff_t i_size; |
2554 | ||
2555 | down_write(&EXT4_I(inode)->i_data_sem); | |
2556 | i_size = i_size_read(inode); | |
2557 | if (disksize > i_size) | |
2558 | disksize = i_size; | |
2559 | if (disksize > EXT4_I(inode)->i_disksize) | |
2560 | EXT4_I(inode)->i_disksize = disksize; | |
622cad13 | 2561 | up_write(&EXT4_I(inode)->i_data_sem); |
b907f2d5 | 2562 | err2 = ext4_mark_inode_dirty(handle, inode); |
4e7ea81d JK |
2563 | if (err2) |
2564 | ext4_error(inode->i_sb, | |
2565 | "Failed to mark inode %lu dirty", | |
2566 | inode->i_ino); | |
2567 | if (!err) | |
2568 | err = err2; | |
2569 | } | |
2570 | return err; | |
2571 | } | |
2572 | ||
fffb2739 JK |
2573 | /* |
2574 | * Calculate the total number of credits to reserve for one writepages | |
20970ba6 | 2575 | * iteration. This is called from ext4_writepages(). We map an extent of |
70261f56 | 2576 | * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping |
fffb2739 JK |
2577 | * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN + |
2578 | * bpp - 1 blocks in bpp different extents. | |
2579 | */ | |
525f4ed8 MC |
2580 | static int ext4_da_writepages_trans_blocks(struct inode *inode) |
2581 | { | |
fffb2739 | 2582 | int bpp = ext4_journal_blocks_per_page(inode); |
525f4ed8 | 2583 | |
fffb2739 JK |
2584 | return ext4_meta_trans_blocks(inode, |
2585 | MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp); | |
525f4ed8 | 2586 | } |
61628a3f | 2587 | |
8e48dcfb | 2588 | /* |
4e7ea81d JK |
2589 | * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages |
2590 | * and underlying extent to map | |
2591 | * | |
2592 | * @mpd - where to look for pages | |
2593 | * | |
2594 | * Walk dirty pages in the mapping. If they are fully mapped, submit them for | |
2595 | * IO immediately. When we find a page which isn't mapped we start accumulating | |
2596 | * extent of buffers underlying these pages that needs mapping (formed by | |
2597 | * either delayed or unwritten buffers). We also lock the pages containing | |
2598 | * these buffers. The extent found is returned in @mpd structure (starting at | |
2599 | * mpd->lblk with length mpd->len blocks). | |
2600 | * | |
2601 | * Note that this function can attach bios to one io_end structure which are | |
2602 | * neither logically nor physically contiguous. Although it may seem as an | |
2603 | * unnecessary complication, it is actually inevitable in blocksize < pagesize | |
2604 | * case as we need to track IO to all buffers underlying a page in one io_end. | |
8e48dcfb | 2605 | */ |
4e7ea81d | 2606 | static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd) |
8e48dcfb | 2607 | { |
4e7ea81d JK |
2608 | struct address_space *mapping = mpd->inode->i_mapping; |
2609 | struct pagevec pvec; | |
2610 | unsigned int nr_pages; | |
aeac589a | 2611 | long left = mpd->wbc->nr_to_write; |
4e7ea81d JK |
2612 | pgoff_t index = mpd->first_page; |
2613 | pgoff_t end = mpd->last_page; | |
2614 | int tag; | |
2615 | int i, err = 0; | |
2616 | int blkbits = mpd->inode->i_blkbits; | |
2617 | ext4_lblk_t lblk; | |
2618 | struct buffer_head *head; | |
8e48dcfb | 2619 | |
4e7ea81d | 2620 | if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages) |
5b41d924 ES |
2621 | tag = PAGECACHE_TAG_TOWRITE; |
2622 | else | |
2623 | tag = PAGECACHE_TAG_DIRTY; | |
2624 | ||
4e7ea81d JK |
2625 | pagevec_init(&pvec, 0); |
2626 | mpd->map.m_len = 0; | |
2627 | mpd->next_page = index; | |
4f01b02c | 2628 | while (index <= end) { |
5b41d924 | 2629 | nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag, |
8e48dcfb TT |
2630 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1); |
2631 | if (nr_pages == 0) | |
4e7ea81d | 2632 | goto out; |
8e48dcfb TT |
2633 | |
2634 | for (i = 0; i < nr_pages; i++) { | |
2635 | struct page *page = pvec.pages[i]; | |
2636 | ||
2637 | /* | |
2638 | * At this point, the page may be truncated or | |
2639 | * invalidated (changing page->mapping to NULL), or | |
2640 | * even swizzled back from swapper_space to tmpfs file | |
2641 | * mapping. However, page->index will not change | |
2642 | * because we have a reference on the page. | |
2643 | */ | |
4f01b02c TT |
2644 | if (page->index > end) |
2645 | goto out; | |
8e48dcfb | 2646 | |
aeac589a ML |
2647 | /* |
2648 | * Accumulated enough dirty pages? This doesn't apply | |
2649 | * to WB_SYNC_ALL mode. For integrity sync we have to | |
2650 | * keep going because someone may be concurrently | |
2651 | * dirtying pages, and we might have synced a lot of | |
2652 | * newly appeared dirty pages, but have not synced all | |
2653 | * of the old dirty pages. | |
2654 | */ | |
2655 | if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0) | |
2656 | goto out; | |
2657 | ||
4e7ea81d JK |
2658 | /* If we can't merge this page, we are done. */ |
2659 | if (mpd->map.m_len > 0 && mpd->next_page != page->index) | |
2660 | goto out; | |
78aaced3 | 2661 | |
8e48dcfb | 2662 | lock_page(page); |
8e48dcfb | 2663 | /* |
4e7ea81d JK |
2664 | * If the page is no longer dirty, or its mapping no |
2665 | * longer corresponds to inode we are writing (which | |
2666 | * means it has been truncated or invalidated), or the | |
2667 | * page is already under writeback and we are not doing | |
2668 | * a data integrity writeback, skip the page | |
8e48dcfb | 2669 | */ |
4f01b02c TT |
2670 | if (!PageDirty(page) || |
2671 | (PageWriteback(page) && | |
4e7ea81d | 2672 | (mpd->wbc->sync_mode == WB_SYNC_NONE)) || |
4f01b02c | 2673 | unlikely(page->mapping != mapping)) { |
8e48dcfb TT |
2674 | unlock_page(page); |
2675 | continue; | |
2676 | } | |
2677 | ||
7cb1a535 | 2678 | wait_on_page_writeback(page); |
8e48dcfb | 2679 | BUG_ON(PageWriteback(page)); |
8e48dcfb | 2680 | |
4e7ea81d | 2681 | if (mpd->map.m_len == 0) |
8eb9e5ce | 2682 | mpd->first_page = page->index; |
8eb9e5ce | 2683 | mpd->next_page = page->index + 1; |
f8bec370 | 2684 | /* Add all dirty buffers to mpd */ |
4e7ea81d | 2685 | lblk = ((ext4_lblk_t)page->index) << |
09cbfeaf | 2686 | (PAGE_SHIFT - blkbits); |
f8bec370 | 2687 | head = page_buffers(page); |
5f1132b2 JK |
2688 | err = mpage_process_page_bufs(mpd, head, head, lblk); |
2689 | if (err <= 0) | |
4e7ea81d | 2690 | goto out; |
5f1132b2 | 2691 | err = 0; |
aeac589a | 2692 | left--; |
8e48dcfb TT |
2693 | } |
2694 | pagevec_release(&pvec); | |
2695 | cond_resched(); | |
2696 | } | |
4f01b02c | 2697 | return 0; |
8eb9e5ce TT |
2698 | out: |
2699 | pagevec_release(&pvec); | |
4e7ea81d | 2700 | return err; |
8e48dcfb TT |
2701 | } |
2702 | ||
20970ba6 TT |
2703 | static int __writepage(struct page *page, struct writeback_control *wbc, |
2704 | void *data) | |
2705 | { | |
2706 | struct address_space *mapping = data; | |
2707 | int ret = ext4_writepage(page, wbc); | |
2708 | mapping_set_error(mapping, ret); | |
2709 | return ret; | |
2710 | } | |
2711 | ||
2712 | static int ext4_writepages(struct address_space *mapping, | |
2713 | struct writeback_control *wbc) | |
64769240 | 2714 | { |
4e7ea81d JK |
2715 | pgoff_t writeback_index = 0; |
2716 | long nr_to_write = wbc->nr_to_write; | |
22208ded | 2717 | int range_whole = 0; |
4e7ea81d | 2718 | int cycled = 1; |
61628a3f | 2719 | handle_t *handle = NULL; |
df22291f | 2720 | struct mpage_da_data mpd; |
5e745b04 | 2721 | struct inode *inode = mapping->host; |
6b523df4 | 2722 | int needed_blocks, rsv_blocks = 0, ret = 0; |
5e745b04 | 2723 | struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb); |
4e7ea81d | 2724 | bool done; |
1bce63d1 | 2725 | struct blk_plug plug; |
cb530541 | 2726 | bool give_up_on_write = false; |
61628a3f | 2727 | |
0db1ff22 TT |
2728 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
2729 | return -EIO; | |
2730 | ||
c8585c6f | 2731 | percpu_down_read(&sbi->s_journal_flag_rwsem); |
20970ba6 | 2732 | trace_ext4_writepages(inode, wbc); |
ba80b101 | 2733 | |
c8585c6f DJ |
2734 | if (dax_mapping(mapping)) { |
2735 | ret = dax_writeback_mapping_range(mapping, inode->i_sb->s_bdev, | |
2736 | wbc); | |
2737 | goto out_writepages; | |
2738 | } | |
7f6d5b52 | 2739 | |
61628a3f MC |
2740 | /* |
2741 | * No pages to write? This is mainly a kludge to avoid starting | |
2742 | * a transaction for special inodes like journal inode on last iput() | |
2743 | * because that could violate lock ordering on umount | |
2744 | */ | |
a1d6cc56 | 2745 | if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) |
bbf023c7 | 2746 | goto out_writepages; |
2a21e37e | 2747 | |
20970ba6 TT |
2748 | if (ext4_should_journal_data(inode)) { |
2749 | struct blk_plug plug; | |
20970ba6 TT |
2750 | |
2751 | blk_start_plug(&plug); | |
2752 | ret = write_cache_pages(mapping, wbc, __writepage, mapping); | |
2753 | blk_finish_plug(&plug); | |
bbf023c7 | 2754 | goto out_writepages; |
20970ba6 TT |
2755 | } |
2756 | ||
2a21e37e TT |
2757 | /* |
2758 | * If the filesystem has aborted, it is read-only, so return | |
2759 | * right away instead of dumping stack traces later on that | |
2760 | * will obscure the real source of the problem. We test | |
4ab2f15b | 2761 | * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because |
2a21e37e | 2762 | * the latter could be true if the filesystem is mounted |
20970ba6 | 2763 | * read-only, and in that case, ext4_writepages should |
2a21e37e TT |
2764 | * *never* be called, so if that ever happens, we would want |
2765 | * the stack trace. | |
2766 | */ | |
0db1ff22 TT |
2767 | if (unlikely(ext4_forced_shutdown(EXT4_SB(mapping->host->i_sb)) || |
2768 | sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) { | |
bbf023c7 ML |
2769 | ret = -EROFS; |
2770 | goto out_writepages; | |
2771 | } | |
2a21e37e | 2772 | |
6b523df4 JK |
2773 | if (ext4_should_dioread_nolock(inode)) { |
2774 | /* | |
70261f56 | 2775 | * We may need to convert up to one extent per block in |
6b523df4 JK |
2776 | * the page and we may dirty the inode. |
2777 | */ | |
09cbfeaf | 2778 | rsv_blocks = 1 + (PAGE_SIZE >> inode->i_blkbits); |
6b523df4 JK |
2779 | } |
2780 | ||
4e7ea81d JK |
2781 | /* |
2782 | * If we have inline data and arrive here, it means that | |
2783 | * we will soon create the block for the 1st page, so | |
2784 | * we'd better clear the inline data here. | |
2785 | */ | |
2786 | if (ext4_has_inline_data(inode)) { | |
2787 | /* Just inode will be modified... */ | |
2788 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 1); | |
2789 | if (IS_ERR(handle)) { | |
2790 | ret = PTR_ERR(handle); | |
2791 | goto out_writepages; | |
2792 | } | |
2793 | BUG_ON(ext4_test_inode_state(inode, | |
2794 | EXT4_STATE_MAY_INLINE_DATA)); | |
2795 | ext4_destroy_inline_data(handle, inode); | |
2796 | ext4_journal_stop(handle); | |
2797 | } | |
2798 | ||
22208ded AK |
2799 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
2800 | range_whole = 1; | |
61628a3f | 2801 | |
2acf2c26 | 2802 | if (wbc->range_cyclic) { |
4e7ea81d JK |
2803 | writeback_index = mapping->writeback_index; |
2804 | if (writeback_index) | |
2acf2c26 | 2805 | cycled = 0; |
4e7ea81d JK |
2806 | mpd.first_page = writeback_index; |
2807 | mpd.last_page = -1; | |
5b41d924 | 2808 | } else { |
09cbfeaf KS |
2809 | mpd.first_page = wbc->range_start >> PAGE_SHIFT; |
2810 | mpd.last_page = wbc->range_end >> PAGE_SHIFT; | |
5b41d924 | 2811 | } |
a1d6cc56 | 2812 | |
4e7ea81d JK |
2813 | mpd.inode = inode; |
2814 | mpd.wbc = wbc; | |
2815 | ext4_io_submit_init(&mpd.io_submit, wbc); | |
2acf2c26 | 2816 | retry: |
6e6938b6 | 2817 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
4e7ea81d JK |
2818 | tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page); |
2819 | done = false; | |
1bce63d1 | 2820 | blk_start_plug(&plug); |
dddbd6ac JK |
2821 | |
2822 | /* | |
2823 | * First writeback pages that don't need mapping - we can avoid | |
2824 | * starting a transaction unnecessarily and also avoid being blocked | |
2825 | * in the block layer on device congestion while having transaction | |
2826 | * started. | |
2827 | */ | |
2828 | mpd.do_map = 0; | |
2829 | mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL); | |
2830 | if (!mpd.io_submit.io_end) { | |
2831 | ret = -ENOMEM; | |
2832 | goto unplug; | |
2833 | } | |
2834 | ret = mpage_prepare_extent_to_map(&mpd); | |
2835 | /* Submit prepared bio */ | |
2836 | ext4_io_submit(&mpd.io_submit); | |
2837 | ext4_put_io_end_defer(mpd.io_submit.io_end); | |
2838 | mpd.io_submit.io_end = NULL; | |
2839 | /* Unlock pages we didn't use */ | |
2840 | mpage_release_unused_pages(&mpd, false); | |
2841 | if (ret < 0) | |
2842 | goto unplug; | |
2843 | ||
4e7ea81d JK |
2844 | while (!done && mpd.first_page <= mpd.last_page) { |
2845 | /* For each extent of pages we use new io_end */ | |
2846 | mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL); | |
2847 | if (!mpd.io_submit.io_end) { | |
2848 | ret = -ENOMEM; | |
2849 | break; | |
2850 | } | |
a1d6cc56 AK |
2851 | |
2852 | /* | |
4e7ea81d JK |
2853 | * We have two constraints: We find one extent to map and we |
2854 | * must always write out whole page (makes a difference when | |
2855 | * blocksize < pagesize) so that we don't block on IO when we | |
2856 | * try to write out the rest of the page. Journalled mode is | |
2857 | * not supported by delalloc. | |
a1d6cc56 AK |
2858 | */ |
2859 | BUG_ON(ext4_should_journal_data(inode)); | |
525f4ed8 | 2860 | needed_blocks = ext4_da_writepages_trans_blocks(inode); |
a1d6cc56 | 2861 | |
4e7ea81d | 2862 | /* start a new transaction */ |
6b523df4 JK |
2863 | handle = ext4_journal_start_with_reserve(inode, |
2864 | EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks); | |
61628a3f MC |
2865 | if (IS_ERR(handle)) { |
2866 | ret = PTR_ERR(handle); | |
1693918e | 2867 | ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: " |
fbe845dd | 2868 | "%ld pages, ino %lu; err %d", __func__, |
a1d6cc56 | 2869 | wbc->nr_to_write, inode->i_ino, ret); |
4e7ea81d JK |
2870 | /* Release allocated io_end */ |
2871 | ext4_put_io_end(mpd.io_submit.io_end); | |
dddbd6ac | 2872 | mpd.io_submit.io_end = NULL; |
4e7ea81d | 2873 | break; |
61628a3f | 2874 | } |
dddbd6ac | 2875 | mpd.do_map = 1; |
f63e6005 | 2876 | |
4e7ea81d JK |
2877 | trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc); |
2878 | ret = mpage_prepare_extent_to_map(&mpd); | |
2879 | if (!ret) { | |
2880 | if (mpd.map.m_len) | |
cb530541 TT |
2881 | ret = mpage_map_and_submit_extent(handle, &mpd, |
2882 | &give_up_on_write); | |
4e7ea81d JK |
2883 | else { |
2884 | /* | |
2885 | * We scanned the whole range (or exhausted | |
2886 | * nr_to_write), submitted what was mapped and | |
2887 | * didn't find anything needing mapping. We are | |
2888 | * done. | |
2889 | */ | |
2890 | done = true; | |
2891 | } | |
f63e6005 | 2892 | } |
646caa9c JK |
2893 | /* |
2894 | * Caution: If the handle is synchronous, | |
2895 | * ext4_journal_stop() can wait for transaction commit | |
2896 | * to finish which may depend on writeback of pages to | |
2897 | * complete or on page lock to be released. In that | |
2898 | * case, we have to wait until after after we have | |
2899 | * submitted all the IO, released page locks we hold, | |
2900 | * and dropped io_end reference (for extent conversion | |
2901 | * to be able to complete) before stopping the handle. | |
2902 | */ | |
2903 | if (!ext4_handle_valid(handle) || handle->h_sync == 0) { | |
2904 | ext4_journal_stop(handle); | |
2905 | handle = NULL; | |
dddbd6ac | 2906 | mpd.do_map = 0; |
646caa9c | 2907 | } |
4e7ea81d JK |
2908 | /* Submit prepared bio */ |
2909 | ext4_io_submit(&mpd.io_submit); | |
2910 | /* Unlock pages we didn't use */ | |
cb530541 | 2911 | mpage_release_unused_pages(&mpd, give_up_on_write); |
646caa9c JK |
2912 | /* |
2913 | * Drop our io_end reference we got from init. We have | |
2914 | * to be careful and use deferred io_end finishing if | |
2915 | * we are still holding the transaction as we can | |
2916 | * release the last reference to io_end which may end | |
2917 | * up doing unwritten extent conversion. | |
2918 | */ | |
2919 | if (handle) { | |
2920 | ext4_put_io_end_defer(mpd.io_submit.io_end); | |
2921 | ext4_journal_stop(handle); | |
2922 | } else | |
2923 | ext4_put_io_end(mpd.io_submit.io_end); | |
dddbd6ac | 2924 | mpd.io_submit.io_end = NULL; |
4e7ea81d JK |
2925 | |
2926 | if (ret == -ENOSPC && sbi->s_journal) { | |
2927 | /* | |
2928 | * Commit the transaction which would | |
22208ded AK |
2929 | * free blocks released in the transaction |
2930 | * and try again | |
2931 | */ | |
df22291f | 2932 | jbd2_journal_force_commit_nested(sbi->s_journal); |
22208ded | 2933 | ret = 0; |
4e7ea81d JK |
2934 | continue; |
2935 | } | |
2936 | /* Fatal error - ENOMEM, EIO... */ | |
2937 | if (ret) | |
61628a3f | 2938 | break; |
a1d6cc56 | 2939 | } |
dddbd6ac | 2940 | unplug: |
1bce63d1 | 2941 | blk_finish_plug(&plug); |
9c12a831 | 2942 | if (!ret && !cycled && wbc->nr_to_write > 0) { |
2acf2c26 | 2943 | cycled = 1; |
4e7ea81d JK |
2944 | mpd.last_page = writeback_index - 1; |
2945 | mpd.first_page = 0; | |
2acf2c26 AK |
2946 | goto retry; |
2947 | } | |
22208ded AK |
2948 | |
2949 | /* Update index */ | |
22208ded AK |
2950 | if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) |
2951 | /* | |
4e7ea81d | 2952 | * Set the writeback_index so that range_cyclic |
22208ded AK |
2953 | * mode will write it back later |
2954 | */ | |
4e7ea81d | 2955 | mapping->writeback_index = mpd.first_page; |
a1d6cc56 | 2956 | |
61628a3f | 2957 | out_writepages: |
20970ba6 TT |
2958 | trace_ext4_writepages_result(inode, wbc, ret, |
2959 | nr_to_write - wbc->nr_to_write); | |
c8585c6f | 2960 | percpu_up_read(&sbi->s_journal_flag_rwsem); |
61628a3f | 2961 | return ret; |
64769240 AT |
2962 | } |
2963 | ||
79f0be8d AK |
2964 | static int ext4_nonda_switch(struct super_block *sb) |
2965 | { | |
5c1ff336 | 2966 | s64 free_clusters, dirty_clusters; |
79f0be8d AK |
2967 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
2968 | ||
2969 | /* | |
2970 | * switch to non delalloc mode if we are running low | |
2971 | * on free block. The free block accounting via percpu | |
179f7ebf | 2972 | * counters can get slightly wrong with percpu_counter_batch getting |
79f0be8d AK |
2973 | * accumulated on each CPU without updating global counters |
2974 | * Delalloc need an accurate free block accounting. So switch | |
2975 | * to non delalloc when we are near to error range. | |
2976 | */ | |
5c1ff336 EW |
2977 | free_clusters = |
2978 | percpu_counter_read_positive(&sbi->s_freeclusters_counter); | |
2979 | dirty_clusters = | |
2980 | percpu_counter_read_positive(&sbi->s_dirtyclusters_counter); | |
00d4e736 TT |
2981 | /* |
2982 | * Start pushing delalloc when 1/2 of free blocks are dirty. | |
2983 | */ | |
5c1ff336 | 2984 | if (dirty_clusters && (free_clusters < 2 * dirty_clusters)) |
10ee27a0 | 2985 | try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE); |
00d4e736 | 2986 | |
5c1ff336 EW |
2987 | if (2 * free_clusters < 3 * dirty_clusters || |
2988 | free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) { | |
79f0be8d | 2989 | /* |
c8afb446 ES |
2990 | * free block count is less than 150% of dirty blocks |
2991 | * or free blocks is less than watermark | |
79f0be8d AK |
2992 | */ |
2993 | return 1; | |
2994 | } | |
2995 | return 0; | |
2996 | } | |
2997 | ||
0ff8947f ES |
2998 | /* We always reserve for an inode update; the superblock could be there too */ |
2999 | static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len) | |
3000 | { | |
e2b911c5 | 3001 | if (likely(ext4_has_feature_large_file(inode->i_sb))) |
0ff8947f ES |
3002 | return 1; |
3003 | ||
3004 | if (pos + len <= 0x7fffffffULL) | |
3005 | return 1; | |
3006 | ||
3007 | /* We might need to update the superblock to set LARGE_FILE */ | |
3008 | return 2; | |
3009 | } | |
3010 | ||
64769240 | 3011 | static int ext4_da_write_begin(struct file *file, struct address_space *mapping, |
de9a55b8 TT |
3012 | loff_t pos, unsigned len, unsigned flags, |
3013 | struct page **pagep, void **fsdata) | |
64769240 | 3014 | { |
72b8ab9d | 3015 | int ret, retries = 0; |
64769240 AT |
3016 | struct page *page; |
3017 | pgoff_t index; | |
64769240 AT |
3018 | struct inode *inode = mapping->host; |
3019 | handle_t *handle; | |
3020 | ||
0db1ff22 TT |
3021 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
3022 | return -EIO; | |
3023 | ||
09cbfeaf | 3024 | index = pos >> PAGE_SHIFT; |
79f0be8d | 3025 | |
4db0d88e TT |
3026 | if (ext4_nonda_switch(inode->i_sb) || |
3027 | S_ISLNK(inode->i_mode)) { | |
79f0be8d AK |
3028 | *fsdata = (void *)FALL_BACK_TO_NONDELALLOC; |
3029 | return ext4_write_begin(file, mapping, pos, | |
3030 | len, flags, pagep, fsdata); | |
3031 | } | |
3032 | *fsdata = (void *)0; | |
9bffad1e | 3033 | trace_ext4_da_write_begin(inode, pos, len, flags); |
9c3569b5 TM |
3034 | |
3035 | if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { | |
3036 | ret = ext4_da_write_inline_data_begin(mapping, inode, | |
3037 | pos, len, flags, | |
3038 | pagep, fsdata); | |
3039 | if (ret < 0) | |
47564bfb TT |
3040 | return ret; |
3041 | if (ret == 1) | |
3042 | return 0; | |
9c3569b5 TM |
3043 | } |
3044 | ||
47564bfb TT |
3045 | /* |
3046 | * grab_cache_page_write_begin() can take a long time if the | |
3047 | * system is thrashing due to memory pressure, or if the page | |
3048 | * is being written back. So grab it first before we start | |
3049 | * the transaction handle. This also allows us to allocate | |
3050 | * the page (if needed) without using GFP_NOFS. | |
3051 | */ | |
3052 | retry_grab: | |
3053 | page = grab_cache_page_write_begin(mapping, index, flags); | |
3054 | if (!page) | |
3055 | return -ENOMEM; | |
3056 | unlock_page(page); | |
3057 | ||
64769240 AT |
3058 | /* |
3059 | * With delayed allocation, we don't log the i_disksize update | |
3060 | * if there is delayed block allocation. But we still need | |
3061 | * to journalling the i_disksize update if writes to the end | |
3062 | * of file which has an already mapped buffer. | |
3063 | */ | |
47564bfb | 3064 | retry_journal: |
0ff8947f ES |
3065 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
3066 | ext4_da_write_credits(inode, pos, len)); | |
64769240 | 3067 | if (IS_ERR(handle)) { |
09cbfeaf | 3068 | put_page(page); |
47564bfb | 3069 | return PTR_ERR(handle); |
64769240 AT |
3070 | } |
3071 | ||
47564bfb TT |
3072 | lock_page(page); |
3073 | if (page->mapping != mapping) { | |
3074 | /* The page got truncated from under us */ | |
3075 | unlock_page(page); | |
09cbfeaf | 3076 | put_page(page); |
d5a0d4f7 | 3077 | ext4_journal_stop(handle); |
47564bfb | 3078 | goto retry_grab; |
d5a0d4f7 | 3079 | } |
47564bfb | 3080 | /* In case writeback began while the page was unlocked */ |
7afe5aa5 | 3081 | wait_for_stable_page(page); |
64769240 | 3082 | |
2058f83a MH |
3083 | #ifdef CONFIG_EXT4_FS_ENCRYPTION |
3084 | ret = ext4_block_write_begin(page, pos, len, | |
3085 | ext4_da_get_block_prep); | |
3086 | #else | |
6e1db88d | 3087 | ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep); |
2058f83a | 3088 | #endif |
64769240 AT |
3089 | if (ret < 0) { |
3090 | unlock_page(page); | |
3091 | ext4_journal_stop(handle); | |
ae4d5372 AK |
3092 | /* |
3093 | * block_write_begin may have instantiated a few blocks | |
3094 | * outside i_size. Trim these off again. Don't need | |
3095 | * i_size_read because we hold i_mutex. | |
3096 | */ | |
3097 | if (pos + len > inode->i_size) | |
b9a4207d | 3098 | ext4_truncate_failed_write(inode); |
47564bfb TT |
3099 | |
3100 | if (ret == -ENOSPC && | |
3101 | ext4_should_retry_alloc(inode->i_sb, &retries)) | |
3102 | goto retry_journal; | |
3103 | ||
09cbfeaf | 3104 | put_page(page); |
47564bfb | 3105 | return ret; |
64769240 AT |
3106 | } |
3107 | ||
47564bfb | 3108 | *pagep = page; |
64769240 AT |
3109 | return ret; |
3110 | } | |
3111 | ||
632eaeab MC |
3112 | /* |
3113 | * Check if we should update i_disksize | |
3114 | * when write to the end of file but not require block allocation | |
3115 | */ | |
3116 | static int ext4_da_should_update_i_disksize(struct page *page, | |
de9a55b8 | 3117 | unsigned long offset) |
632eaeab MC |
3118 | { |
3119 | struct buffer_head *bh; | |
3120 | struct inode *inode = page->mapping->host; | |
3121 | unsigned int idx; | |
3122 | int i; | |
3123 | ||
3124 | bh = page_buffers(page); | |
3125 | idx = offset >> inode->i_blkbits; | |
3126 | ||
af5bc92d | 3127 | for (i = 0; i < idx; i++) |
632eaeab MC |
3128 | bh = bh->b_this_page; |
3129 | ||
29fa89d0 | 3130 | if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh)) |
632eaeab MC |
3131 | return 0; |
3132 | return 1; | |
3133 | } | |
3134 | ||
64769240 | 3135 | static int ext4_da_write_end(struct file *file, |
de9a55b8 TT |
3136 | struct address_space *mapping, |
3137 | loff_t pos, unsigned len, unsigned copied, | |
3138 | struct page *page, void *fsdata) | |
64769240 AT |
3139 | { |
3140 | struct inode *inode = mapping->host; | |
3141 | int ret = 0, ret2; | |
3142 | handle_t *handle = ext4_journal_current_handle(); | |
3143 | loff_t new_i_size; | |
632eaeab | 3144 | unsigned long start, end; |
79f0be8d AK |
3145 | int write_mode = (int)(unsigned long)fsdata; |
3146 | ||
74d553aa TT |
3147 | if (write_mode == FALL_BACK_TO_NONDELALLOC) |
3148 | return ext4_write_end(file, mapping, pos, | |
3149 | len, copied, page, fsdata); | |
632eaeab | 3150 | |
9bffad1e | 3151 | trace_ext4_da_write_end(inode, pos, len, copied); |
09cbfeaf | 3152 | start = pos & (PAGE_SIZE - 1); |
af5bc92d | 3153 | end = start + copied - 1; |
64769240 AT |
3154 | |
3155 | /* | |
3156 | * generic_write_end() will run mark_inode_dirty() if i_size | |
3157 | * changes. So let's piggyback the i_disksize mark_inode_dirty | |
3158 | * into that. | |
3159 | */ | |
64769240 | 3160 | new_i_size = pos + copied; |
ea51d132 | 3161 | if (copied && new_i_size > EXT4_I(inode)->i_disksize) { |
9c3569b5 TM |
3162 | if (ext4_has_inline_data(inode) || |
3163 | ext4_da_should_update_i_disksize(page, end)) { | |
ee124d27 | 3164 | ext4_update_i_disksize(inode, new_i_size); |
cf17fea6 AK |
3165 | /* We need to mark inode dirty even if |
3166 | * new_i_size is less that inode->i_size | |
3167 | * bu greater than i_disksize.(hint delalloc) | |
3168 | */ | |
3169 | ext4_mark_inode_dirty(handle, inode); | |
64769240 | 3170 | } |
632eaeab | 3171 | } |
9c3569b5 TM |
3172 | |
3173 | if (write_mode != CONVERT_INLINE_DATA && | |
3174 | ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) && | |
3175 | ext4_has_inline_data(inode)) | |
3176 | ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied, | |
3177 | page); | |
3178 | else | |
3179 | ret2 = generic_write_end(file, mapping, pos, len, copied, | |
64769240 | 3180 | page, fsdata); |
9c3569b5 | 3181 | |
64769240 AT |
3182 | copied = ret2; |
3183 | if (ret2 < 0) | |
3184 | ret = ret2; | |
3185 | ret2 = ext4_journal_stop(handle); | |
3186 | if (!ret) | |
3187 | ret = ret2; | |
3188 | ||
3189 | return ret ? ret : copied; | |
3190 | } | |
3191 | ||
d47992f8 LC |
3192 | static void ext4_da_invalidatepage(struct page *page, unsigned int offset, |
3193 | unsigned int length) | |
64769240 | 3194 | { |
64769240 AT |
3195 | /* |
3196 | * Drop reserved blocks | |
3197 | */ | |
3198 | BUG_ON(!PageLocked(page)); | |
3199 | if (!page_has_buffers(page)) | |
3200 | goto out; | |
3201 | ||
ca99fdd2 | 3202 | ext4_da_page_release_reservation(page, offset, length); |
64769240 AT |
3203 | |
3204 | out: | |
d47992f8 | 3205 | ext4_invalidatepage(page, offset, length); |
64769240 AT |
3206 | |
3207 | return; | |
3208 | } | |
3209 | ||
ccd2506b TT |
3210 | /* |
3211 | * Force all delayed allocation blocks to be allocated for a given inode. | |
3212 | */ | |
3213 | int ext4_alloc_da_blocks(struct inode *inode) | |
3214 | { | |
fb40ba0d TT |
3215 | trace_ext4_alloc_da_blocks(inode); |
3216 | ||
71d4f7d0 | 3217 | if (!EXT4_I(inode)->i_reserved_data_blocks) |
ccd2506b TT |
3218 | return 0; |
3219 | ||
3220 | /* | |
3221 | * We do something simple for now. The filemap_flush() will | |
3222 | * also start triggering a write of the data blocks, which is | |
3223 | * not strictly speaking necessary (and for users of | |
3224 | * laptop_mode, not even desirable). However, to do otherwise | |
3225 | * would require replicating code paths in: | |
de9a55b8 | 3226 | * |
20970ba6 | 3227 | * ext4_writepages() -> |
ccd2506b TT |
3228 | * write_cache_pages() ---> (via passed in callback function) |
3229 | * __mpage_da_writepage() --> | |
3230 | * mpage_add_bh_to_extent() | |
3231 | * mpage_da_map_blocks() | |
3232 | * | |
3233 | * The problem is that write_cache_pages(), located in | |
3234 | * mm/page-writeback.c, marks pages clean in preparation for | |
3235 | * doing I/O, which is not desirable if we're not planning on | |
3236 | * doing I/O at all. | |
3237 | * | |
3238 | * We could call write_cache_pages(), and then redirty all of | |
380cf090 | 3239 | * the pages by calling redirty_page_for_writepage() but that |
ccd2506b TT |
3240 | * would be ugly in the extreme. So instead we would need to |
3241 | * replicate parts of the code in the above functions, | |
25985edc | 3242 | * simplifying them because we wouldn't actually intend to |
ccd2506b TT |
3243 | * write out the pages, but rather only collect contiguous |
3244 | * logical block extents, call the multi-block allocator, and | |
3245 | * then update the buffer heads with the block allocations. | |
de9a55b8 | 3246 | * |
ccd2506b TT |
3247 | * For now, though, we'll cheat by calling filemap_flush(), |
3248 | * which will map the blocks, and start the I/O, but not | |
3249 | * actually wait for the I/O to complete. | |
3250 | */ | |
3251 | return filemap_flush(inode->i_mapping); | |
3252 | } | |
64769240 | 3253 | |
ac27a0ec DK |
3254 | /* |
3255 | * bmap() is special. It gets used by applications such as lilo and by | |
3256 | * the swapper to find the on-disk block of a specific piece of data. | |
3257 | * | |
3258 | * Naturally, this is dangerous if the block concerned is still in the | |
617ba13b | 3259 | * journal. If somebody makes a swapfile on an ext4 data-journaling |
ac27a0ec DK |
3260 | * filesystem and enables swap, then they may get a nasty shock when the |
3261 | * data getting swapped to that swapfile suddenly gets overwritten by | |
3262 | * the original zero's written out previously to the journal and | |
3263 | * awaiting writeback in the kernel's buffer cache. | |
3264 | * | |
3265 | * So, if we see any bmap calls here on a modified, data-journaled file, | |
3266 | * take extra steps to flush any blocks which might be in the cache. | |
3267 | */ | |
617ba13b | 3268 | static sector_t ext4_bmap(struct address_space *mapping, sector_t block) |
ac27a0ec DK |
3269 | { |
3270 | struct inode *inode = mapping->host; | |
3271 | journal_t *journal; | |
3272 | int err; | |
3273 | ||
46c7f254 TM |
3274 | /* |
3275 | * We can get here for an inline file via the FIBMAP ioctl | |
3276 | */ | |
3277 | if (ext4_has_inline_data(inode)) | |
3278 | return 0; | |
3279 | ||
64769240 AT |
3280 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) && |
3281 | test_opt(inode->i_sb, DELALLOC)) { | |
3282 | /* | |
3283 | * With delalloc we want to sync the file | |
3284 | * so that we can make sure we allocate | |
3285 | * blocks for file | |
3286 | */ | |
3287 | filemap_write_and_wait(mapping); | |
3288 | } | |
3289 | ||
19f5fb7a TT |
3290 | if (EXT4_JOURNAL(inode) && |
3291 | ext4_test_inode_state(inode, EXT4_STATE_JDATA)) { | |
ac27a0ec DK |
3292 | /* |
3293 | * This is a REALLY heavyweight approach, but the use of | |
3294 | * bmap on dirty files is expected to be extremely rare: | |
3295 | * only if we run lilo or swapon on a freshly made file | |
3296 | * do we expect this to happen. | |
3297 | * | |
3298 | * (bmap requires CAP_SYS_RAWIO so this does not | |
3299 | * represent an unprivileged user DOS attack --- we'd be | |
3300 | * in trouble if mortal users could trigger this path at | |
3301 | * will.) | |
3302 | * | |
617ba13b | 3303 | * NB. EXT4_STATE_JDATA is not set on files other than |
ac27a0ec DK |
3304 | * regular files. If somebody wants to bmap a directory |
3305 | * or symlink and gets confused because the buffer | |
3306 | * hasn't yet been flushed to disk, they deserve | |
3307 | * everything they get. | |
3308 | */ | |
3309 | ||
19f5fb7a | 3310 | ext4_clear_inode_state(inode, EXT4_STATE_JDATA); |
617ba13b | 3311 | journal = EXT4_JOURNAL(inode); |
dab291af MC |
3312 | jbd2_journal_lock_updates(journal); |
3313 | err = jbd2_journal_flush(journal); | |
3314 | jbd2_journal_unlock_updates(journal); | |
ac27a0ec DK |
3315 | |
3316 | if (err) | |
3317 | return 0; | |
3318 | } | |
3319 | ||
af5bc92d | 3320 | return generic_block_bmap(mapping, block, ext4_get_block); |
ac27a0ec DK |
3321 | } |
3322 | ||
617ba13b | 3323 | static int ext4_readpage(struct file *file, struct page *page) |
ac27a0ec | 3324 | { |
46c7f254 TM |
3325 | int ret = -EAGAIN; |
3326 | struct inode *inode = page->mapping->host; | |
3327 | ||
0562e0ba | 3328 | trace_ext4_readpage(page); |
46c7f254 TM |
3329 | |
3330 | if (ext4_has_inline_data(inode)) | |
3331 | ret = ext4_readpage_inline(inode, page); | |
3332 | ||
3333 | if (ret == -EAGAIN) | |
f64e02fe | 3334 | return ext4_mpage_readpages(page->mapping, NULL, page, 1); |
46c7f254 TM |
3335 | |
3336 | return ret; | |
ac27a0ec DK |
3337 | } |
3338 | ||
3339 | static int | |
617ba13b | 3340 | ext4_readpages(struct file *file, struct address_space *mapping, |
ac27a0ec DK |
3341 | struct list_head *pages, unsigned nr_pages) |
3342 | { | |
46c7f254 TM |
3343 | struct inode *inode = mapping->host; |
3344 | ||
3345 | /* If the file has inline data, no need to do readpages. */ | |
3346 | if (ext4_has_inline_data(inode)) | |
3347 | return 0; | |
3348 | ||
f64e02fe | 3349 | return ext4_mpage_readpages(mapping, pages, NULL, nr_pages); |
ac27a0ec DK |
3350 | } |
3351 | ||
d47992f8 LC |
3352 | static void ext4_invalidatepage(struct page *page, unsigned int offset, |
3353 | unsigned int length) | |
ac27a0ec | 3354 | { |
ca99fdd2 | 3355 | trace_ext4_invalidatepage(page, offset, length); |
0562e0ba | 3356 | |
4520fb3c JK |
3357 | /* No journalling happens on data buffers when this function is used */ |
3358 | WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page))); | |
3359 | ||
ca99fdd2 | 3360 | block_invalidatepage(page, offset, length); |
4520fb3c JK |
3361 | } |
3362 | ||
53e87268 | 3363 | static int __ext4_journalled_invalidatepage(struct page *page, |
ca99fdd2 LC |
3364 | unsigned int offset, |
3365 | unsigned int length) | |
4520fb3c JK |
3366 | { |
3367 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); | |
3368 | ||
ca99fdd2 | 3369 | trace_ext4_journalled_invalidatepage(page, offset, length); |
4520fb3c | 3370 | |
ac27a0ec DK |
3371 | /* |
3372 | * If it's a full truncate we just forget about the pending dirtying | |
3373 | */ | |
09cbfeaf | 3374 | if (offset == 0 && length == PAGE_SIZE) |
ac27a0ec DK |
3375 | ClearPageChecked(page); |
3376 | ||
ca99fdd2 | 3377 | return jbd2_journal_invalidatepage(journal, page, offset, length); |
53e87268 JK |
3378 | } |
3379 | ||
3380 | /* Wrapper for aops... */ | |
3381 | static void ext4_journalled_invalidatepage(struct page *page, | |
d47992f8 LC |
3382 | unsigned int offset, |
3383 | unsigned int length) | |
53e87268 | 3384 | { |
ca99fdd2 | 3385 | WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0); |
ac27a0ec DK |
3386 | } |
3387 | ||
617ba13b | 3388 | static int ext4_releasepage(struct page *page, gfp_t wait) |
ac27a0ec | 3389 | { |
617ba13b | 3390 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); |
ac27a0ec | 3391 | |
0562e0ba JZ |
3392 | trace_ext4_releasepage(page); |
3393 | ||
e1c36595 JK |
3394 | /* Page has dirty journalled data -> cannot release */ |
3395 | if (PageChecked(page)) | |
ac27a0ec | 3396 | return 0; |
0390131b FM |
3397 | if (journal) |
3398 | return jbd2_journal_try_to_free_buffers(journal, page, wait); | |
3399 | else | |
3400 | return try_to_free_buffers(page); | |
ac27a0ec DK |
3401 | } |
3402 | ||
ba5843f5 | 3403 | #ifdef CONFIG_FS_DAX |
364443cb JK |
3404 | static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length, |
3405 | unsigned flags, struct iomap *iomap) | |
3406 | { | |
fa5d932c | 3407 | struct block_device *bdev; |
364443cb JK |
3408 | unsigned int blkbits = inode->i_blkbits; |
3409 | unsigned long first_block = offset >> blkbits; | |
3410 | unsigned long last_block = (offset + length - 1) >> blkbits; | |
3411 | struct ext4_map_blocks map; | |
3412 | int ret; | |
3413 | ||
364443cb JK |
3414 | if (WARN_ON_ONCE(ext4_has_inline_data(inode))) |
3415 | return -ERANGE; | |
3416 | ||
3417 | map.m_lblk = first_block; | |
3418 | map.m_len = last_block - first_block + 1; | |
3419 | ||
776722e8 JK |
3420 | if (!(flags & IOMAP_WRITE)) { |
3421 | ret = ext4_map_blocks(NULL, inode, &map, 0); | |
3422 | } else { | |
3423 | int dio_credits; | |
3424 | handle_t *handle; | |
3425 | int retries = 0; | |
3426 | ||
3427 | /* Trim mapping request to maximum we can map at once for DIO */ | |
3428 | if (map.m_len > DIO_MAX_BLOCKS) | |
3429 | map.m_len = DIO_MAX_BLOCKS; | |
3430 | dio_credits = ext4_chunk_trans_blocks(inode, map.m_len); | |
3431 | retry: | |
3432 | /* | |
3433 | * Either we allocate blocks and then we don't get unwritten | |
3434 | * extent so we have reserved enough credits, or the blocks | |
3435 | * are already allocated and unwritten and in that case | |
3436 | * extent conversion fits in the credits as well. | |
3437 | */ | |
3438 | handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, | |
3439 | dio_credits); | |
3440 | if (IS_ERR(handle)) | |
3441 | return PTR_ERR(handle); | |
3442 | ||
3443 | ret = ext4_map_blocks(handle, inode, &map, | |
776722e8 JK |
3444 | EXT4_GET_BLOCKS_CREATE_ZERO); |
3445 | if (ret < 0) { | |
3446 | ext4_journal_stop(handle); | |
3447 | if (ret == -ENOSPC && | |
3448 | ext4_should_retry_alloc(inode->i_sb, &retries)) | |
3449 | goto retry; | |
3450 | return ret; | |
3451 | } | |
776722e8 JK |
3452 | |
3453 | /* | |
e2ae766c | 3454 | * If we added blocks beyond i_size, we need to make sure they |
776722e8 | 3455 | * will get truncated if we crash before updating i_size in |
e2ae766c JK |
3456 | * ext4_iomap_end(). For faults we don't need to do that (and |
3457 | * even cannot because for orphan list operations inode_lock is | |
3458 | * required) - if we happen to instantiate block beyond i_size, | |
3459 | * it is because we race with truncate which has already added | |
3460 | * the inode to the orphan list. | |
776722e8 | 3461 | */ |
e2ae766c JK |
3462 | if (!(flags & IOMAP_FAULT) && first_block + map.m_len > |
3463 | (i_size_read(inode) + (1 << blkbits) - 1) >> blkbits) { | |
776722e8 JK |
3464 | int err; |
3465 | ||
3466 | err = ext4_orphan_add(handle, inode); | |
3467 | if (err < 0) { | |
3468 | ext4_journal_stop(handle); | |
3469 | return err; | |
3470 | } | |
3471 | } | |
3472 | ext4_journal_stop(handle); | |
3473 | } | |
364443cb JK |
3474 | |
3475 | iomap->flags = 0; | |
fa5d932c DW |
3476 | bdev = inode->i_sb->s_bdev; |
3477 | iomap->bdev = bdev; | |
3478 | if (blk_queue_dax(bdev->bd_queue)) | |
f5705aa8 | 3479 | iomap->dax_dev = fs_dax_get_by_host(bdev->bd_disk->disk_name); |
fa5d932c DW |
3480 | else |
3481 | iomap->dax_dev = NULL; | |
364443cb JK |
3482 | iomap->offset = first_block << blkbits; |
3483 | ||
3484 | if (ret == 0) { | |
3485 | iomap->type = IOMAP_HOLE; | |
3486 | iomap->blkno = IOMAP_NULL_BLOCK; | |
3487 | iomap->length = (u64)map.m_len << blkbits; | |
3488 | } else { | |
3489 | if (map.m_flags & EXT4_MAP_MAPPED) { | |
3490 | iomap->type = IOMAP_MAPPED; | |
3491 | } else if (map.m_flags & EXT4_MAP_UNWRITTEN) { | |
3492 | iomap->type = IOMAP_UNWRITTEN; | |
3493 | } else { | |
3494 | WARN_ON_ONCE(1); | |
3495 | return -EIO; | |
3496 | } | |
3497 | iomap->blkno = (sector_t)map.m_pblk << (blkbits - 9); | |
3498 | iomap->length = (u64)map.m_len << blkbits; | |
3499 | } | |
3500 | ||
3501 | if (map.m_flags & EXT4_MAP_NEW) | |
3502 | iomap->flags |= IOMAP_F_NEW; | |
3503 | return 0; | |
3504 | } | |
3505 | ||
776722e8 JK |
3506 | static int ext4_iomap_end(struct inode *inode, loff_t offset, loff_t length, |
3507 | ssize_t written, unsigned flags, struct iomap *iomap) | |
3508 | { | |
3509 | int ret = 0; | |
3510 | handle_t *handle; | |
3511 | int blkbits = inode->i_blkbits; | |
3512 | bool truncate = false; | |
3513 | ||
f5705aa8 | 3514 | fs_put_dax(iomap->dax_dev); |
e2ae766c | 3515 | if (!(flags & IOMAP_WRITE) || (flags & IOMAP_FAULT)) |
776722e8 JK |
3516 | return 0; |
3517 | ||
3518 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 2); | |
3519 | if (IS_ERR(handle)) { | |
3520 | ret = PTR_ERR(handle); | |
3521 | goto orphan_del; | |
3522 | } | |
3523 | if (ext4_update_inode_size(inode, offset + written)) | |
3524 | ext4_mark_inode_dirty(handle, inode); | |
3525 | /* | |
3526 | * We may need to truncate allocated but not written blocks beyond EOF. | |
3527 | */ | |
3528 | if (iomap->offset + iomap->length > | |
3529 | ALIGN(inode->i_size, 1 << blkbits)) { | |
3530 | ext4_lblk_t written_blk, end_blk; | |
3531 | ||
3532 | written_blk = (offset + written) >> blkbits; | |
3533 | end_blk = (offset + length) >> blkbits; | |
3534 | if (written_blk < end_blk && ext4_can_truncate(inode)) | |
3535 | truncate = true; | |
3536 | } | |
3537 | /* | |
3538 | * Remove inode from orphan list if we were extending a inode and | |
3539 | * everything went fine. | |
3540 | */ | |
3541 | if (!truncate && inode->i_nlink && | |
3542 | !list_empty(&EXT4_I(inode)->i_orphan)) | |
3543 | ext4_orphan_del(handle, inode); | |
3544 | ext4_journal_stop(handle); | |
3545 | if (truncate) { | |
3546 | ext4_truncate_failed_write(inode); | |
3547 | orphan_del: | |
3548 | /* | |
3549 | * If truncate failed early the inode might still be on the | |
3550 | * orphan list; we need to make sure the inode is removed from | |
3551 | * the orphan list in that case. | |
3552 | */ | |
3553 | if (inode->i_nlink) | |
3554 | ext4_orphan_del(NULL, inode); | |
3555 | } | |
3556 | return ret; | |
3557 | } | |
3558 | ||
8ff6daa1 | 3559 | const struct iomap_ops ext4_iomap_ops = { |
364443cb | 3560 | .iomap_begin = ext4_iomap_begin, |
776722e8 | 3561 | .iomap_end = ext4_iomap_end, |
364443cb JK |
3562 | }; |
3563 | ||
ba5843f5 | 3564 | #endif |
ed923b57 | 3565 | |
187372a3 | 3566 | static int ext4_end_io_dio(struct kiocb *iocb, loff_t offset, |
7b7a8665 | 3567 | ssize_t size, void *private) |
4c0425ff | 3568 | { |
109811c2 | 3569 | ext4_io_end_t *io_end = private; |
4c0425ff | 3570 | |
97a851ed | 3571 | /* if not async direct IO just return */ |
7b7a8665 | 3572 | if (!io_end) |
187372a3 | 3573 | return 0; |
4b70df18 | 3574 | |
88635ca2 | 3575 | ext_debug("ext4_end_io_dio(): io_end 0x%p " |
ace36ad4 | 3576 | "for inode %lu, iocb 0x%p, offset %llu, size %zd\n", |
109811c2 | 3577 | io_end, io_end->inode->i_ino, iocb, offset, size); |
8d5d02e6 | 3578 | |
74c66bcb JK |
3579 | /* |
3580 | * Error during AIO DIO. We cannot convert unwritten extents as the | |
3581 | * data was not written. Just clear the unwritten flag and drop io_end. | |
3582 | */ | |
3583 | if (size <= 0) { | |
3584 | ext4_clear_io_unwritten_flag(io_end); | |
3585 | size = 0; | |
3586 | } | |
4c0425ff MC |
3587 | io_end->offset = offset; |
3588 | io_end->size = size; | |
7b7a8665 | 3589 | ext4_put_io_end(io_end); |
187372a3 CH |
3590 | |
3591 | return 0; | |
4c0425ff | 3592 | } |
c7064ef1 | 3593 | |
4c0425ff | 3594 | /* |
914f82a3 JK |
3595 | * Handling of direct IO writes. |
3596 | * | |
3597 | * For ext4 extent files, ext4 will do direct-io write even to holes, | |
4c0425ff MC |
3598 | * preallocated extents, and those write extend the file, no need to |
3599 | * fall back to buffered IO. | |
3600 | * | |
556615dc | 3601 | * For holes, we fallocate those blocks, mark them as unwritten |
69c499d1 | 3602 | * If those blocks were preallocated, we mark sure they are split, but |
556615dc | 3603 | * still keep the range to write as unwritten. |
4c0425ff | 3604 | * |
69c499d1 | 3605 | * The unwritten extents will be converted to written when DIO is completed. |
8d5d02e6 | 3606 | * For async direct IO, since the IO may still pending when return, we |
25985edc | 3607 | * set up an end_io call back function, which will do the conversion |
8d5d02e6 | 3608 | * when async direct IO completed. |
4c0425ff MC |
3609 | * |
3610 | * If the O_DIRECT write will extend the file then add this inode to the | |
3611 | * orphan list. So recovery will truncate it back to the original size | |
3612 | * if the machine crashes during the write. | |
3613 | * | |
3614 | */ | |
0e01df10 | 3615 | static ssize_t ext4_direct_IO_write(struct kiocb *iocb, struct iov_iter *iter) |
4c0425ff MC |
3616 | { |
3617 | struct file *file = iocb->ki_filp; | |
3618 | struct inode *inode = file->f_mapping->host; | |
914f82a3 | 3619 | struct ext4_inode_info *ei = EXT4_I(inode); |
4c0425ff | 3620 | ssize_t ret; |
c8b8e32d | 3621 | loff_t offset = iocb->ki_pos; |
a6cbcd4a | 3622 | size_t count = iov_iter_count(iter); |
69c499d1 TT |
3623 | int overwrite = 0; |
3624 | get_block_t *get_block_func = NULL; | |
3625 | int dio_flags = 0; | |
4c0425ff | 3626 | loff_t final_size = offset + count; |
914f82a3 JK |
3627 | int orphan = 0; |
3628 | handle_t *handle; | |
729f52c6 | 3629 | |
914f82a3 JK |
3630 | if (final_size > inode->i_size) { |
3631 | /* Credits for sb + inode write */ | |
3632 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 2); | |
3633 | if (IS_ERR(handle)) { | |
3634 | ret = PTR_ERR(handle); | |
3635 | goto out; | |
3636 | } | |
3637 | ret = ext4_orphan_add(handle, inode); | |
3638 | if (ret) { | |
3639 | ext4_journal_stop(handle); | |
3640 | goto out; | |
3641 | } | |
3642 | orphan = 1; | |
3643 | ei->i_disksize = inode->i_size; | |
3644 | ext4_journal_stop(handle); | |
3645 | } | |
4bd809db | 3646 | |
69c499d1 | 3647 | BUG_ON(iocb->private == NULL); |
4bd809db | 3648 | |
e8340395 JK |
3649 | /* |
3650 | * Make all waiters for direct IO properly wait also for extent | |
3651 | * conversion. This also disallows race between truncate() and | |
3652 | * overwrite DIO as i_dio_count needs to be incremented under i_mutex. | |
3653 | */ | |
914f82a3 | 3654 | inode_dio_begin(inode); |
e8340395 | 3655 | |
69c499d1 TT |
3656 | /* If we do a overwrite dio, i_mutex locking can be released */ |
3657 | overwrite = *((int *)iocb->private); | |
4bd809db | 3658 | |
2dcba478 | 3659 | if (overwrite) |
5955102c | 3660 | inode_unlock(inode); |
8d5d02e6 | 3661 | |
69c499d1 | 3662 | /* |
914f82a3 | 3663 | * For extent mapped files we could direct write to holes and fallocate. |
69c499d1 | 3664 | * |
109811c2 JK |
3665 | * Allocated blocks to fill the hole are marked as unwritten to prevent |
3666 | * parallel buffered read to expose the stale data before DIO complete | |
3667 | * the data IO. | |
69c499d1 | 3668 | * |
109811c2 JK |
3669 | * As to previously fallocated extents, ext4 get_block will just simply |
3670 | * mark the buffer mapped but still keep the extents unwritten. | |
69c499d1 | 3671 | * |
109811c2 JK |
3672 | * For non AIO case, we will convert those unwritten extents to written |
3673 | * after return back from blockdev_direct_IO. That way we save us from | |
3674 | * allocating io_end structure and also the overhead of offloading | |
3675 | * the extent convertion to a workqueue. | |
69c499d1 TT |
3676 | * |
3677 | * For async DIO, the conversion needs to be deferred when the | |
3678 | * IO is completed. The ext4 end_io callback function will be | |
3679 | * called to take care of the conversion work. Here for async | |
3680 | * case, we allocate an io_end structure to hook to the iocb. | |
3681 | */ | |
3682 | iocb->private = NULL; | |
109811c2 | 3683 | if (overwrite) |
705965bd | 3684 | get_block_func = ext4_dio_get_block_overwrite; |
0bd2d5ec | 3685 | else if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) || |
93407472 | 3686 | round_down(offset, i_blocksize(inode)) >= inode->i_size) { |
914f82a3 JK |
3687 | get_block_func = ext4_dio_get_block; |
3688 | dio_flags = DIO_LOCKING | DIO_SKIP_HOLES; | |
3689 | } else if (is_sync_kiocb(iocb)) { | |
109811c2 JK |
3690 | get_block_func = ext4_dio_get_block_unwritten_sync; |
3691 | dio_flags = DIO_LOCKING; | |
69c499d1 | 3692 | } else { |
109811c2 | 3693 | get_block_func = ext4_dio_get_block_unwritten_async; |
69c499d1 TT |
3694 | dio_flags = DIO_LOCKING; |
3695 | } | |
0bd2d5ec JK |
3696 | ret = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev, iter, |
3697 | get_block_func, ext4_end_io_dio, NULL, | |
3698 | dio_flags); | |
69c499d1 | 3699 | |
97a851ed | 3700 | if (ret > 0 && !overwrite && ext4_test_inode_state(inode, |
69c499d1 TT |
3701 | EXT4_STATE_DIO_UNWRITTEN)) { |
3702 | int err; | |
3703 | /* | |
3704 | * for non AIO case, since the IO is already | |
3705 | * completed, we could do the conversion right here | |
3706 | */ | |
6b523df4 | 3707 | err = ext4_convert_unwritten_extents(NULL, inode, |
69c499d1 TT |
3708 | offset, ret); |
3709 | if (err < 0) | |
3710 | ret = err; | |
3711 | ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN); | |
3712 | } | |
4bd809db | 3713 | |
914f82a3 | 3714 | inode_dio_end(inode); |
69c499d1 | 3715 | /* take i_mutex locking again if we do a ovewrite dio */ |
2dcba478 | 3716 | if (overwrite) |
5955102c | 3717 | inode_lock(inode); |
8d5d02e6 | 3718 | |
914f82a3 JK |
3719 | if (ret < 0 && final_size > inode->i_size) |
3720 | ext4_truncate_failed_write(inode); | |
3721 | ||
3722 | /* Handle extending of i_size after direct IO write */ | |
3723 | if (orphan) { | |
3724 | int err; | |
3725 | ||
3726 | /* Credits for sb + inode write */ | |
3727 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 2); | |
3728 | if (IS_ERR(handle)) { | |
3729 | /* This is really bad luck. We've written the data | |
3730 | * but cannot extend i_size. Bail out and pretend | |
3731 | * the write failed... */ | |
3732 | ret = PTR_ERR(handle); | |
3733 | if (inode->i_nlink) | |
3734 | ext4_orphan_del(NULL, inode); | |
3735 | ||
3736 | goto out; | |
3737 | } | |
3738 | if (inode->i_nlink) | |
3739 | ext4_orphan_del(handle, inode); | |
3740 | if (ret > 0) { | |
3741 | loff_t end = offset + ret; | |
3742 | if (end > inode->i_size) { | |
3743 | ei->i_disksize = end; | |
3744 | i_size_write(inode, end); | |
3745 | /* | |
3746 | * We're going to return a positive `ret' | |
3747 | * here due to non-zero-length I/O, so there's | |
3748 | * no way of reporting error returns from | |
3749 | * ext4_mark_inode_dirty() to userspace. So | |
3750 | * ignore it. | |
3751 | */ | |
3752 | ext4_mark_inode_dirty(handle, inode); | |
3753 | } | |
3754 | } | |
3755 | err = ext4_journal_stop(handle); | |
3756 | if (ret == 0) | |
3757 | ret = err; | |
3758 | } | |
3759 | out: | |
3760 | return ret; | |
3761 | } | |
3762 | ||
0e01df10 | 3763 | static ssize_t ext4_direct_IO_read(struct kiocb *iocb, struct iov_iter *iter) |
914f82a3 | 3764 | { |
16c54688 JK |
3765 | struct address_space *mapping = iocb->ki_filp->f_mapping; |
3766 | struct inode *inode = mapping->host; | |
0bd2d5ec | 3767 | size_t count = iov_iter_count(iter); |
914f82a3 JK |
3768 | ssize_t ret; |
3769 | ||
16c54688 JK |
3770 | /* |
3771 | * Shared inode_lock is enough for us - it protects against concurrent | |
3772 | * writes & truncates and since we take care of writing back page cache, | |
3773 | * we are protected against page writeback as well. | |
3774 | */ | |
3775 | inode_lock_shared(inode); | |
0bd2d5ec | 3776 | ret = filemap_write_and_wait_range(mapping, iocb->ki_pos, |
e5465795 | 3777 | iocb->ki_pos + count - 1); |
0bd2d5ec JK |
3778 | if (ret) |
3779 | goto out_unlock; | |
3780 | ret = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev, | |
3781 | iter, ext4_dio_get_block, NULL, NULL, 0); | |
16c54688 JK |
3782 | out_unlock: |
3783 | inode_unlock_shared(inode); | |
69c499d1 | 3784 | return ret; |
4c0425ff MC |
3785 | } |
3786 | ||
c8b8e32d | 3787 | static ssize_t ext4_direct_IO(struct kiocb *iocb, struct iov_iter *iter) |
4c0425ff MC |
3788 | { |
3789 | struct file *file = iocb->ki_filp; | |
3790 | struct inode *inode = file->f_mapping->host; | |
a6cbcd4a | 3791 | size_t count = iov_iter_count(iter); |
c8b8e32d | 3792 | loff_t offset = iocb->ki_pos; |
0562e0ba | 3793 | ssize_t ret; |
4c0425ff | 3794 | |
2058f83a MH |
3795 | #ifdef CONFIG_EXT4_FS_ENCRYPTION |
3796 | if (ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode)) | |
3797 | return 0; | |
3798 | #endif | |
3799 | ||
84ebd795 TT |
3800 | /* |
3801 | * If we are doing data journalling we don't support O_DIRECT | |
3802 | */ | |
3803 | if (ext4_should_journal_data(inode)) | |
3804 | return 0; | |
3805 | ||
46c7f254 TM |
3806 | /* Let buffer I/O handle the inline data case. */ |
3807 | if (ext4_has_inline_data(inode)) | |
3808 | return 0; | |
3809 | ||
0bd2d5ec JK |
3810 | /* DAX uses iomap path now */ |
3811 | if (WARN_ON_ONCE(IS_DAX(inode))) | |
3812 | return 0; | |
3813 | ||
6f673763 | 3814 | trace_ext4_direct_IO_enter(inode, offset, count, iov_iter_rw(iter)); |
914f82a3 | 3815 | if (iov_iter_rw(iter) == READ) |
0e01df10 | 3816 | ret = ext4_direct_IO_read(iocb, iter); |
0562e0ba | 3817 | else |
0e01df10 | 3818 | ret = ext4_direct_IO_write(iocb, iter); |
6f673763 | 3819 | trace_ext4_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), ret); |
0562e0ba | 3820 | return ret; |
4c0425ff MC |
3821 | } |
3822 | ||
ac27a0ec | 3823 | /* |
617ba13b | 3824 | * Pages can be marked dirty completely asynchronously from ext4's journalling |
ac27a0ec DK |
3825 | * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do |
3826 | * much here because ->set_page_dirty is called under VFS locks. The page is | |
3827 | * not necessarily locked. | |
3828 | * | |
3829 | * We cannot just dirty the page and leave attached buffers clean, because the | |
3830 | * buffers' dirty state is "definitive". We cannot just set the buffers dirty | |
3831 | * or jbddirty because all the journalling code will explode. | |
3832 | * | |
3833 | * So what we do is to mark the page "pending dirty" and next time writepage | |
3834 | * is called, propagate that into the buffers appropriately. | |
3835 | */ | |
617ba13b | 3836 | static int ext4_journalled_set_page_dirty(struct page *page) |
ac27a0ec DK |
3837 | { |
3838 | SetPageChecked(page); | |
3839 | return __set_page_dirty_nobuffers(page); | |
3840 | } | |
3841 | ||
6dcc693b JK |
3842 | static int ext4_set_page_dirty(struct page *page) |
3843 | { | |
3844 | WARN_ON_ONCE(!PageLocked(page) && !PageDirty(page)); | |
3845 | WARN_ON_ONCE(!page_has_buffers(page)); | |
3846 | return __set_page_dirty_buffers(page); | |
3847 | } | |
3848 | ||
74d553aa | 3849 | static const struct address_space_operations ext4_aops = { |
8ab22b9a HH |
3850 | .readpage = ext4_readpage, |
3851 | .readpages = ext4_readpages, | |
43ce1d23 | 3852 | .writepage = ext4_writepage, |
20970ba6 | 3853 | .writepages = ext4_writepages, |
8ab22b9a | 3854 | .write_begin = ext4_write_begin, |
74d553aa | 3855 | .write_end = ext4_write_end, |
6dcc693b | 3856 | .set_page_dirty = ext4_set_page_dirty, |
8ab22b9a HH |
3857 | .bmap = ext4_bmap, |
3858 | .invalidatepage = ext4_invalidatepage, | |
3859 | .releasepage = ext4_releasepage, | |
3860 | .direct_IO = ext4_direct_IO, | |
3861 | .migratepage = buffer_migrate_page, | |
3862 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 3863 | .error_remove_page = generic_error_remove_page, |
ac27a0ec DK |
3864 | }; |
3865 | ||
617ba13b | 3866 | static const struct address_space_operations ext4_journalled_aops = { |
8ab22b9a HH |
3867 | .readpage = ext4_readpage, |
3868 | .readpages = ext4_readpages, | |
43ce1d23 | 3869 | .writepage = ext4_writepage, |
20970ba6 | 3870 | .writepages = ext4_writepages, |
8ab22b9a HH |
3871 | .write_begin = ext4_write_begin, |
3872 | .write_end = ext4_journalled_write_end, | |
3873 | .set_page_dirty = ext4_journalled_set_page_dirty, | |
3874 | .bmap = ext4_bmap, | |
4520fb3c | 3875 | .invalidatepage = ext4_journalled_invalidatepage, |
8ab22b9a | 3876 | .releasepage = ext4_releasepage, |
84ebd795 | 3877 | .direct_IO = ext4_direct_IO, |
8ab22b9a | 3878 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 3879 | .error_remove_page = generic_error_remove_page, |
ac27a0ec DK |
3880 | }; |
3881 | ||
64769240 | 3882 | static const struct address_space_operations ext4_da_aops = { |
8ab22b9a HH |
3883 | .readpage = ext4_readpage, |
3884 | .readpages = ext4_readpages, | |
43ce1d23 | 3885 | .writepage = ext4_writepage, |
20970ba6 | 3886 | .writepages = ext4_writepages, |
8ab22b9a HH |
3887 | .write_begin = ext4_da_write_begin, |
3888 | .write_end = ext4_da_write_end, | |
6dcc693b | 3889 | .set_page_dirty = ext4_set_page_dirty, |
8ab22b9a HH |
3890 | .bmap = ext4_bmap, |
3891 | .invalidatepage = ext4_da_invalidatepage, | |
3892 | .releasepage = ext4_releasepage, | |
3893 | .direct_IO = ext4_direct_IO, | |
3894 | .migratepage = buffer_migrate_page, | |
3895 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 3896 | .error_remove_page = generic_error_remove_page, |
64769240 AT |
3897 | }; |
3898 | ||
617ba13b | 3899 | void ext4_set_aops(struct inode *inode) |
ac27a0ec | 3900 | { |
3d2b1582 LC |
3901 | switch (ext4_inode_journal_mode(inode)) { |
3902 | case EXT4_INODE_ORDERED_DATA_MODE: | |
3d2b1582 | 3903 | case EXT4_INODE_WRITEBACK_DATA_MODE: |
3d2b1582 LC |
3904 | break; |
3905 | case EXT4_INODE_JOURNAL_DATA_MODE: | |
617ba13b | 3906 | inode->i_mapping->a_ops = &ext4_journalled_aops; |
74d553aa | 3907 | return; |
3d2b1582 LC |
3908 | default: |
3909 | BUG(); | |
3910 | } | |
74d553aa TT |
3911 | if (test_opt(inode->i_sb, DELALLOC)) |
3912 | inode->i_mapping->a_ops = &ext4_da_aops; | |
3913 | else | |
3914 | inode->i_mapping->a_ops = &ext4_aops; | |
ac27a0ec DK |
3915 | } |
3916 | ||
923ae0ff | 3917 | static int __ext4_block_zero_page_range(handle_t *handle, |
d863dc36 LC |
3918 | struct address_space *mapping, loff_t from, loff_t length) |
3919 | { | |
09cbfeaf KS |
3920 | ext4_fsblk_t index = from >> PAGE_SHIFT; |
3921 | unsigned offset = from & (PAGE_SIZE-1); | |
923ae0ff | 3922 | unsigned blocksize, pos; |
d863dc36 LC |
3923 | ext4_lblk_t iblock; |
3924 | struct inode *inode = mapping->host; | |
3925 | struct buffer_head *bh; | |
3926 | struct page *page; | |
3927 | int err = 0; | |
3928 | ||
09cbfeaf | 3929 | page = find_or_create_page(mapping, from >> PAGE_SHIFT, |
c62d2555 | 3930 | mapping_gfp_constraint(mapping, ~__GFP_FS)); |
d863dc36 LC |
3931 | if (!page) |
3932 | return -ENOMEM; | |
3933 | ||
3934 | blocksize = inode->i_sb->s_blocksize; | |
d863dc36 | 3935 | |
09cbfeaf | 3936 | iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits); |
d863dc36 LC |
3937 | |
3938 | if (!page_has_buffers(page)) | |
3939 | create_empty_buffers(page, blocksize, 0); | |
3940 | ||
3941 | /* Find the buffer that contains "offset" */ | |
3942 | bh = page_buffers(page); | |
3943 | pos = blocksize; | |
3944 | while (offset >= pos) { | |
3945 | bh = bh->b_this_page; | |
3946 | iblock++; | |
3947 | pos += blocksize; | |
3948 | } | |
d863dc36 LC |
3949 | if (buffer_freed(bh)) { |
3950 | BUFFER_TRACE(bh, "freed: skip"); | |
3951 | goto unlock; | |
3952 | } | |
d863dc36 LC |
3953 | if (!buffer_mapped(bh)) { |
3954 | BUFFER_TRACE(bh, "unmapped"); | |
3955 | ext4_get_block(inode, iblock, bh, 0); | |
3956 | /* unmapped? It's a hole - nothing to do */ | |
3957 | if (!buffer_mapped(bh)) { | |
3958 | BUFFER_TRACE(bh, "still unmapped"); | |
3959 | goto unlock; | |
3960 | } | |
3961 | } | |
3962 | ||
3963 | /* Ok, it's mapped. Make sure it's up-to-date */ | |
3964 | if (PageUptodate(page)) | |
3965 | set_buffer_uptodate(bh); | |
3966 | ||
3967 | if (!buffer_uptodate(bh)) { | |
3968 | err = -EIO; | |
dfec8a14 | 3969 | ll_rw_block(REQ_OP_READ, 0, 1, &bh); |
d863dc36 LC |
3970 | wait_on_buffer(bh); |
3971 | /* Uhhuh. Read error. Complain and punt. */ | |
3972 | if (!buffer_uptodate(bh)) | |
3973 | goto unlock; | |
c9c7429c MH |
3974 | if (S_ISREG(inode->i_mode) && |
3975 | ext4_encrypted_inode(inode)) { | |
3976 | /* We expect the key to be set. */ | |
a7550b30 | 3977 | BUG_ON(!fscrypt_has_encryption_key(inode)); |
09cbfeaf | 3978 | BUG_ON(blocksize != PAGE_SIZE); |
b50f7b26 | 3979 | WARN_ON_ONCE(fscrypt_decrypt_page(page->mapping->host, |
9c4bb8a3 | 3980 | page, PAGE_SIZE, 0, page->index)); |
c9c7429c | 3981 | } |
d863dc36 | 3982 | } |
d863dc36 LC |
3983 | if (ext4_should_journal_data(inode)) { |
3984 | BUFFER_TRACE(bh, "get write access"); | |
3985 | err = ext4_journal_get_write_access(handle, bh); | |
3986 | if (err) | |
3987 | goto unlock; | |
3988 | } | |
d863dc36 | 3989 | zero_user(page, offset, length); |
d863dc36 LC |
3990 | BUFFER_TRACE(bh, "zeroed end of block"); |
3991 | ||
d863dc36 LC |
3992 | if (ext4_should_journal_data(inode)) { |
3993 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
0713ed0c | 3994 | } else { |
353eefd3 | 3995 | err = 0; |
d863dc36 | 3996 | mark_buffer_dirty(bh); |
3957ef53 | 3997 | if (ext4_should_order_data(inode)) |
ee0876bc | 3998 | err = ext4_jbd2_inode_add_write(handle, inode); |
0713ed0c | 3999 | } |
d863dc36 LC |
4000 | |
4001 | unlock: | |
4002 | unlock_page(page); | |
09cbfeaf | 4003 | put_page(page); |
d863dc36 LC |
4004 | return err; |
4005 | } | |
4006 | ||
923ae0ff RZ |
4007 | /* |
4008 | * ext4_block_zero_page_range() zeros out a mapping of length 'length' | |
4009 | * starting from file offset 'from'. The range to be zero'd must | |
4010 | * be contained with in one block. If the specified range exceeds | |
4011 | * the end of the block it will be shortened to end of the block | |
4012 | * that cooresponds to 'from' | |
4013 | */ | |
4014 | static int ext4_block_zero_page_range(handle_t *handle, | |
4015 | struct address_space *mapping, loff_t from, loff_t length) | |
4016 | { | |
4017 | struct inode *inode = mapping->host; | |
09cbfeaf | 4018 | unsigned offset = from & (PAGE_SIZE-1); |
923ae0ff RZ |
4019 | unsigned blocksize = inode->i_sb->s_blocksize; |
4020 | unsigned max = blocksize - (offset & (blocksize - 1)); | |
4021 | ||
4022 | /* | |
4023 | * correct length if it does not fall between | |
4024 | * 'from' and the end of the block | |
4025 | */ | |
4026 | if (length > max || length < 0) | |
4027 | length = max; | |
4028 | ||
47e69351 JK |
4029 | if (IS_DAX(inode)) { |
4030 | return iomap_zero_range(inode, from, length, NULL, | |
4031 | &ext4_iomap_ops); | |
4032 | } | |
923ae0ff RZ |
4033 | return __ext4_block_zero_page_range(handle, mapping, from, length); |
4034 | } | |
4035 | ||
94350ab5 MW |
4036 | /* |
4037 | * ext4_block_truncate_page() zeroes out a mapping from file offset `from' | |
4038 | * up to the end of the block which corresponds to `from'. | |
4039 | * This required during truncate. We need to physically zero the tail end | |
4040 | * of that block so it doesn't yield old data if the file is later grown. | |
4041 | */ | |
c197855e | 4042 | static int ext4_block_truncate_page(handle_t *handle, |
94350ab5 MW |
4043 | struct address_space *mapping, loff_t from) |
4044 | { | |
09cbfeaf | 4045 | unsigned offset = from & (PAGE_SIZE-1); |
94350ab5 MW |
4046 | unsigned length; |
4047 | unsigned blocksize; | |
4048 | struct inode *inode = mapping->host; | |
4049 | ||
0d06863f TT |
4050 | /* If we are processing an encrypted inode during orphan list handling */ |
4051 | if (ext4_encrypted_inode(inode) && !fscrypt_has_encryption_key(inode)) | |
4052 | return 0; | |
4053 | ||
94350ab5 MW |
4054 | blocksize = inode->i_sb->s_blocksize; |
4055 | length = blocksize - (offset & (blocksize - 1)); | |
4056 | ||
4057 | return ext4_block_zero_page_range(handle, mapping, from, length); | |
4058 | } | |
4059 | ||
a87dd18c LC |
4060 | int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode, |
4061 | loff_t lstart, loff_t length) | |
4062 | { | |
4063 | struct super_block *sb = inode->i_sb; | |
4064 | struct address_space *mapping = inode->i_mapping; | |
e1be3a92 | 4065 | unsigned partial_start, partial_end; |
a87dd18c LC |
4066 | ext4_fsblk_t start, end; |
4067 | loff_t byte_end = (lstart + length - 1); | |
4068 | int err = 0; | |
4069 | ||
e1be3a92 LC |
4070 | partial_start = lstart & (sb->s_blocksize - 1); |
4071 | partial_end = byte_end & (sb->s_blocksize - 1); | |
4072 | ||
a87dd18c LC |
4073 | start = lstart >> sb->s_blocksize_bits; |
4074 | end = byte_end >> sb->s_blocksize_bits; | |
4075 | ||
4076 | /* Handle partial zero within the single block */ | |
e1be3a92 LC |
4077 | if (start == end && |
4078 | (partial_start || (partial_end != sb->s_blocksize - 1))) { | |
a87dd18c LC |
4079 | err = ext4_block_zero_page_range(handle, mapping, |
4080 | lstart, length); | |
4081 | return err; | |
4082 | } | |
4083 | /* Handle partial zero out on the start of the range */ | |
e1be3a92 | 4084 | if (partial_start) { |
a87dd18c LC |
4085 | err = ext4_block_zero_page_range(handle, mapping, |
4086 | lstart, sb->s_blocksize); | |
4087 | if (err) | |
4088 | return err; | |
4089 | } | |
4090 | /* Handle partial zero out on the end of the range */ | |
e1be3a92 | 4091 | if (partial_end != sb->s_blocksize - 1) |
a87dd18c | 4092 | err = ext4_block_zero_page_range(handle, mapping, |
e1be3a92 LC |
4093 | byte_end - partial_end, |
4094 | partial_end + 1); | |
a87dd18c LC |
4095 | return err; |
4096 | } | |
4097 | ||
91ef4caf DG |
4098 | int ext4_can_truncate(struct inode *inode) |
4099 | { | |
91ef4caf DG |
4100 | if (S_ISREG(inode->i_mode)) |
4101 | return 1; | |
4102 | if (S_ISDIR(inode->i_mode)) | |
4103 | return 1; | |
4104 | if (S_ISLNK(inode->i_mode)) | |
4105 | return !ext4_inode_is_fast_symlink(inode); | |
4106 | return 0; | |
4107 | } | |
4108 | ||
01127848 JK |
4109 | /* |
4110 | * We have to make sure i_disksize gets properly updated before we truncate | |
4111 | * page cache due to hole punching or zero range. Otherwise i_disksize update | |
4112 | * can get lost as it may have been postponed to submission of writeback but | |
4113 | * that will never happen after we truncate page cache. | |
4114 | */ | |
4115 | int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset, | |
4116 | loff_t len) | |
4117 | { | |
4118 | handle_t *handle; | |
4119 | loff_t size = i_size_read(inode); | |
4120 | ||
5955102c | 4121 | WARN_ON(!inode_is_locked(inode)); |
01127848 JK |
4122 | if (offset > size || offset + len < size) |
4123 | return 0; | |
4124 | ||
4125 | if (EXT4_I(inode)->i_disksize >= size) | |
4126 | return 0; | |
4127 | ||
4128 | handle = ext4_journal_start(inode, EXT4_HT_MISC, 1); | |
4129 | if (IS_ERR(handle)) | |
4130 | return PTR_ERR(handle); | |
4131 | ext4_update_i_disksize(inode, size); | |
4132 | ext4_mark_inode_dirty(handle, inode); | |
4133 | ext4_journal_stop(handle); | |
4134 | ||
4135 | return 0; | |
4136 | } | |
4137 | ||
a4bb6b64 | 4138 | /* |
cca32b7e | 4139 | * ext4_punch_hole: punches a hole in a file by releasing the blocks |
a4bb6b64 AH |
4140 | * associated with the given offset and length |
4141 | * | |
4142 | * @inode: File inode | |
4143 | * @offset: The offset where the hole will begin | |
4144 | * @len: The length of the hole | |
4145 | * | |
4907cb7b | 4146 | * Returns: 0 on success or negative on failure |
a4bb6b64 AH |
4147 | */ |
4148 | ||
aeb2817a | 4149 | int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length) |
a4bb6b64 | 4150 | { |
26a4c0c6 TT |
4151 | struct super_block *sb = inode->i_sb; |
4152 | ext4_lblk_t first_block, stop_block; | |
4153 | struct address_space *mapping = inode->i_mapping; | |
a87dd18c | 4154 | loff_t first_block_offset, last_block_offset; |
26a4c0c6 TT |
4155 | handle_t *handle; |
4156 | unsigned int credits; | |
4157 | int ret = 0; | |
4158 | ||
a4bb6b64 | 4159 | if (!S_ISREG(inode->i_mode)) |
73355192 | 4160 | return -EOPNOTSUPP; |
a4bb6b64 | 4161 | |
b8a86845 | 4162 | trace_ext4_punch_hole(inode, offset, length, 0); |
aaddea81 | 4163 | |
26a4c0c6 TT |
4164 | /* |
4165 | * Write out all dirty pages to avoid race conditions | |
4166 | * Then release them. | |
4167 | */ | |
cca32b7e | 4168 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { |
26a4c0c6 TT |
4169 | ret = filemap_write_and_wait_range(mapping, offset, |
4170 | offset + length - 1); | |
4171 | if (ret) | |
4172 | return ret; | |
4173 | } | |
4174 | ||
5955102c | 4175 | inode_lock(inode); |
9ef06cec | 4176 | |
26a4c0c6 TT |
4177 | /* No need to punch hole beyond i_size */ |
4178 | if (offset >= inode->i_size) | |
4179 | goto out_mutex; | |
4180 | ||
4181 | /* | |
4182 | * If the hole extends beyond i_size, set the hole | |
4183 | * to end after the page that contains i_size | |
4184 | */ | |
4185 | if (offset + length > inode->i_size) { | |
4186 | length = inode->i_size + | |
09cbfeaf | 4187 | PAGE_SIZE - (inode->i_size & (PAGE_SIZE - 1)) - |
26a4c0c6 TT |
4188 | offset; |
4189 | } | |
4190 | ||
a361293f JK |
4191 | if (offset & (sb->s_blocksize - 1) || |
4192 | (offset + length) & (sb->s_blocksize - 1)) { | |
4193 | /* | |
4194 | * Attach jinode to inode for jbd2 if we do any zeroing of | |
4195 | * partial block | |
4196 | */ | |
4197 | ret = ext4_inode_attach_jinode(inode); | |
4198 | if (ret < 0) | |
4199 | goto out_mutex; | |
4200 | ||
4201 | } | |
4202 | ||
ea3d7209 JK |
4203 | /* Wait all existing dio workers, newcomers will block on i_mutex */ |
4204 | ext4_inode_block_unlocked_dio(inode); | |
4205 | inode_dio_wait(inode); | |
4206 | ||
4207 | /* | |
4208 | * Prevent page faults from reinstantiating pages we have released from | |
4209 | * page cache. | |
4210 | */ | |
4211 | down_write(&EXT4_I(inode)->i_mmap_sem); | |
a87dd18c LC |
4212 | first_block_offset = round_up(offset, sb->s_blocksize); |
4213 | last_block_offset = round_down((offset + length), sb->s_blocksize) - 1; | |
26a4c0c6 | 4214 | |
a87dd18c | 4215 | /* Now release the pages and zero block aligned part of pages*/ |
01127848 JK |
4216 | if (last_block_offset > first_block_offset) { |
4217 | ret = ext4_update_disksize_before_punch(inode, offset, length); | |
4218 | if (ret) | |
4219 | goto out_dio; | |
a87dd18c LC |
4220 | truncate_pagecache_range(inode, first_block_offset, |
4221 | last_block_offset); | |
01127848 | 4222 | } |
26a4c0c6 TT |
4223 | |
4224 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
4225 | credits = ext4_writepage_trans_blocks(inode); | |
4226 | else | |
4227 | credits = ext4_blocks_for_truncate(inode); | |
4228 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); | |
4229 | if (IS_ERR(handle)) { | |
4230 | ret = PTR_ERR(handle); | |
4231 | ext4_std_error(sb, ret); | |
4232 | goto out_dio; | |
4233 | } | |
4234 | ||
a87dd18c LC |
4235 | ret = ext4_zero_partial_blocks(handle, inode, offset, |
4236 | length); | |
4237 | if (ret) | |
4238 | goto out_stop; | |
26a4c0c6 TT |
4239 | |
4240 | first_block = (offset + sb->s_blocksize - 1) >> | |
4241 | EXT4_BLOCK_SIZE_BITS(sb); | |
4242 | stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb); | |
4243 | ||
4244 | /* If there are no blocks to remove, return now */ | |
4245 | if (first_block >= stop_block) | |
4246 | goto out_stop; | |
4247 | ||
4248 | down_write(&EXT4_I(inode)->i_data_sem); | |
4249 | ext4_discard_preallocations(inode); | |
4250 | ||
4251 | ret = ext4_es_remove_extent(inode, first_block, | |
4252 | stop_block - first_block); | |
4253 | if (ret) { | |
4254 | up_write(&EXT4_I(inode)->i_data_sem); | |
4255 | goto out_stop; | |
4256 | } | |
4257 | ||
4258 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
4259 | ret = ext4_ext_remove_space(inode, first_block, | |
4260 | stop_block - 1); | |
4261 | else | |
4f579ae7 | 4262 | ret = ext4_ind_remove_space(handle, inode, first_block, |
26a4c0c6 TT |
4263 | stop_block); |
4264 | ||
819c4920 | 4265 | up_write(&EXT4_I(inode)->i_data_sem); |
26a4c0c6 TT |
4266 | if (IS_SYNC(inode)) |
4267 | ext4_handle_sync(handle); | |
e251f9bc | 4268 | |
eeca7ea1 | 4269 | inode->i_mtime = inode->i_ctime = current_time(inode); |
26a4c0c6 | 4270 | ext4_mark_inode_dirty(handle, inode); |
67a7d5f5 JK |
4271 | if (ret >= 0) |
4272 | ext4_update_inode_fsync_trans(handle, inode, 1); | |
26a4c0c6 TT |
4273 | out_stop: |
4274 | ext4_journal_stop(handle); | |
4275 | out_dio: | |
ea3d7209 | 4276 | up_write(&EXT4_I(inode)->i_mmap_sem); |
26a4c0c6 TT |
4277 | ext4_inode_resume_unlocked_dio(inode); |
4278 | out_mutex: | |
5955102c | 4279 | inode_unlock(inode); |
26a4c0c6 | 4280 | return ret; |
a4bb6b64 AH |
4281 | } |
4282 | ||
a361293f JK |
4283 | int ext4_inode_attach_jinode(struct inode *inode) |
4284 | { | |
4285 | struct ext4_inode_info *ei = EXT4_I(inode); | |
4286 | struct jbd2_inode *jinode; | |
4287 | ||
4288 | if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal) | |
4289 | return 0; | |
4290 | ||
4291 | jinode = jbd2_alloc_inode(GFP_KERNEL); | |
4292 | spin_lock(&inode->i_lock); | |
4293 | if (!ei->jinode) { | |
4294 | if (!jinode) { | |
4295 | spin_unlock(&inode->i_lock); | |
4296 | return -ENOMEM; | |
4297 | } | |
4298 | ei->jinode = jinode; | |
4299 | jbd2_journal_init_jbd_inode(ei->jinode, inode); | |
4300 | jinode = NULL; | |
4301 | } | |
4302 | spin_unlock(&inode->i_lock); | |
4303 | if (unlikely(jinode != NULL)) | |
4304 | jbd2_free_inode(jinode); | |
4305 | return 0; | |
4306 | } | |
4307 | ||
ac27a0ec | 4308 | /* |
617ba13b | 4309 | * ext4_truncate() |
ac27a0ec | 4310 | * |
617ba13b MC |
4311 | * We block out ext4_get_block() block instantiations across the entire |
4312 | * transaction, and VFS/VM ensures that ext4_truncate() cannot run | |
ac27a0ec DK |
4313 | * simultaneously on behalf of the same inode. |
4314 | * | |
42b2aa86 | 4315 | * As we work through the truncate and commit bits of it to the journal there |
ac27a0ec DK |
4316 | * is one core, guiding principle: the file's tree must always be consistent on |
4317 | * disk. We must be able to restart the truncate after a crash. | |
4318 | * | |
4319 | * The file's tree may be transiently inconsistent in memory (although it | |
4320 | * probably isn't), but whenever we close off and commit a journal transaction, | |
4321 | * the contents of (the filesystem + the journal) must be consistent and | |
4322 | * restartable. It's pretty simple, really: bottom up, right to left (although | |
4323 | * left-to-right works OK too). | |
4324 | * | |
4325 | * Note that at recovery time, journal replay occurs *before* the restart of | |
4326 | * truncate against the orphan inode list. | |
4327 | * | |
4328 | * The committed inode has the new, desired i_size (which is the same as | |
617ba13b | 4329 | * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see |
ac27a0ec | 4330 | * that this inode's truncate did not complete and it will again call |
617ba13b MC |
4331 | * ext4_truncate() to have another go. So there will be instantiated blocks |
4332 | * to the right of the truncation point in a crashed ext4 filesystem. But | |
ac27a0ec | 4333 | * that's fine - as long as they are linked from the inode, the post-crash |
617ba13b | 4334 | * ext4_truncate() run will find them and release them. |
ac27a0ec | 4335 | */ |
2c98eb5e | 4336 | int ext4_truncate(struct inode *inode) |
ac27a0ec | 4337 | { |
819c4920 TT |
4338 | struct ext4_inode_info *ei = EXT4_I(inode); |
4339 | unsigned int credits; | |
2c98eb5e | 4340 | int err = 0; |
819c4920 TT |
4341 | handle_t *handle; |
4342 | struct address_space *mapping = inode->i_mapping; | |
819c4920 | 4343 | |
19b5ef61 TT |
4344 | /* |
4345 | * There is a possibility that we're either freeing the inode | |
e04027e8 | 4346 | * or it's a completely new inode. In those cases we might not |
19b5ef61 TT |
4347 | * have i_mutex locked because it's not necessary. |
4348 | */ | |
4349 | if (!(inode->i_state & (I_NEW|I_FREEING))) | |
5955102c | 4350 | WARN_ON(!inode_is_locked(inode)); |
0562e0ba JZ |
4351 | trace_ext4_truncate_enter(inode); |
4352 | ||
91ef4caf | 4353 | if (!ext4_can_truncate(inode)) |
2c98eb5e | 4354 | return 0; |
ac27a0ec | 4355 | |
12e9b892 | 4356 | ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS); |
c8d46e41 | 4357 | |
5534fb5b | 4358 | if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC)) |
19f5fb7a | 4359 | ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE); |
7d8f9f7d | 4360 | |
aef1c851 TM |
4361 | if (ext4_has_inline_data(inode)) { |
4362 | int has_inline = 1; | |
4363 | ||
01daf945 TT |
4364 | err = ext4_inline_data_truncate(inode, &has_inline); |
4365 | if (err) | |
4366 | return err; | |
aef1c851 | 4367 | if (has_inline) |
2c98eb5e | 4368 | return 0; |
aef1c851 TM |
4369 | } |
4370 | ||
a361293f JK |
4371 | /* If we zero-out tail of the page, we have to create jinode for jbd2 */ |
4372 | if (inode->i_size & (inode->i_sb->s_blocksize - 1)) { | |
4373 | if (ext4_inode_attach_jinode(inode) < 0) | |
2c98eb5e | 4374 | return 0; |
a361293f JK |
4375 | } |
4376 | ||
819c4920 TT |
4377 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
4378 | credits = ext4_writepage_trans_blocks(inode); | |
4379 | else | |
4380 | credits = ext4_blocks_for_truncate(inode); | |
4381 | ||
4382 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); | |
2c98eb5e TT |
4383 | if (IS_ERR(handle)) |
4384 | return PTR_ERR(handle); | |
819c4920 | 4385 | |
eb3544c6 LC |
4386 | if (inode->i_size & (inode->i_sb->s_blocksize - 1)) |
4387 | ext4_block_truncate_page(handle, mapping, inode->i_size); | |
819c4920 TT |
4388 | |
4389 | /* | |
4390 | * We add the inode to the orphan list, so that if this | |
4391 | * truncate spans multiple transactions, and we crash, we will | |
4392 | * resume the truncate when the filesystem recovers. It also | |
4393 | * marks the inode dirty, to catch the new size. | |
4394 | * | |
4395 | * Implication: the file must always be in a sane, consistent | |
4396 | * truncatable state while each transaction commits. | |
4397 | */ | |
2c98eb5e TT |
4398 | err = ext4_orphan_add(handle, inode); |
4399 | if (err) | |
819c4920 TT |
4400 | goto out_stop; |
4401 | ||
4402 | down_write(&EXT4_I(inode)->i_data_sem); | |
4403 | ||
4404 | ext4_discard_preallocations(inode); | |
4405 | ||
ff9893dc | 4406 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
d0abb36d | 4407 | err = ext4_ext_truncate(handle, inode); |
ff9893dc | 4408 | else |
819c4920 TT |
4409 | ext4_ind_truncate(handle, inode); |
4410 | ||
4411 | up_write(&ei->i_data_sem); | |
d0abb36d TT |
4412 | if (err) |
4413 | goto out_stop; | |
819c4920 TT |
4414 | |
4415 | if (IS_SYNC(inode)) | |
4416 | ext4_handle_sync(handle); | |
4417 | ||
4418 | out_stop: | |
4419 | /* | |
4420 | * If this was a simple ftruncate() and the file will remain alive, | |
4421 | * then we need to clear up the orphan record which we created above. | |
4422 | * However, if this was a real unlink then we were called by | |
58d86a50 | 4423 | * ext4_evict_inode(), and we allow that function to clean up the |
819c4920 TT |
4424 | * orphan info for us. |
4425 | */ | |
4426 | if (inode->i_nlink) | |
4427 | ext4_orphan_del(handle, inode); | |
4428 | ||
eeca7ea1 | 4429 | inode->i_mtime = inode->i_ctime = current_time(inode); |
819c4920 TT |
4430 | ext4_mark_inode_dirty(handle, inode); |
4431 | ext4_journal_stop(handle); | |
ac27a0ec | 4432 | |
0562e0ba | 4433 | trace_ext4_truncate_exit(inode); |
2c98eb5e | 4434 | return err; |
ac27a0ec DK |
4435 | } |
4436 | ||
ac27a0ec | 4437 | /* |
617ba13b | 4438 | * ext4_get_inode_loc returns with an extra refcount against the inode's |
ac27a0ec DK |
4439 | * underlying buffer_head on success. If 'in_mem' is true, we have all |
4440 | * data in memory that is needed to recreate the on-disk version of this | |
4441 | * inode. | |
4442 | */ | |
617ba13b MC |
4443 | static int __ext4_get_inode_loc(struct inode *inode, |
4444 | struct ext4_iloc *iloc, int in_mem) | |
ac27a0ec | 4445 | { |
240799cd TT |
4446 | struct ext4_group_desc *gdp; |
4447 | struct buffer_head *bh; | |
4448 | struct super_block *sb = inode->i_sb; | |
4449 | ext4_fsblk_t block; | |
4450 | int inodes_per_block, inode_offset; | |
4451 | ||
3a06d778 | 4452 | iloc->bh = NULL; |
240799cd | 4453 | if (!ext4_valid_inum(sb, inode->i_ino)) |
6a797d27 | 4454 | return -EFSCORRUPTED; |
ac27a0ec | 4455 | |
240799cd TT |
4456 | iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb); |
4457 | gdp = ext4_get_group_desc(sb, iloc->block_group, NULL); | |
4458 | if (!gdp) | |
ac27a0ec DK |
4459 | return -EIO; |
4460 | ||
240799cd TT |
4461 | /* |
4462 | * Figure out the offset within the block group inode table | |
4463 | */ | |
00d09882 | 4464 | inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; |
240799cd TT |
4465 | inode_offset = ((inode->i_ino - 1) % |
4466 | EXT4_INODES_PER_GROUP(sb)); | |
4467 | block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block); | |
4468 | iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb); | |
4469 | ||
4470 | bh = sb_getblk(sb, block); | |
aebf0243 | 4471 | if (unlikely(!bh)) |
860d21e2 | 4472 | return -ENOMEM; |
ac27a0ec DK |
4473 | if (!buffer_uptodate(bh)) { |
4474 | lock_buffer(bh); | |
9c83a923 HK |
4475 | |
4476 | /* | |
4477 | * If the buffer has the write error flag, we have failed | |
4478 | * to write out another inode in the same block. In this | |
4479 | * case, we don't have to read the block because we may | |
4480 | * read the old inode data successfully. | |
4481 | */ | |
4482 | if (buffer_write_io_error(bh) && !buffer_uptodate(bh)) | |
4483 | set_buffer_uptodate(bh); | |
4484 | ||
ac27a0ec DK |
4485 | if (buffer_uptodate(bh)) { |
4486 | /* someone brought it uptodate while we waited */ | |
4487 | unlock_buffer(bh); | |
4488 | goto has_buffer; | |
4489 | } | |
4490 | ||
4491 | /* | |
4492 | * If we have all information of the inode in memory and this | |
4493 | * is the only valid inode in the block, we need not read the | |
4494 | * block. | |
4495 | */ | |
4496 | if (in_mem) { | |
4497 | struct buffer_head *bitmap_bh; | |
240799cd | 4498 | int i, start; |
ac27a0ec | 4499 | |
240799cd | 4500 | start = inode_offset & ~(inodes_per_block - 1); |
ac27a0ec | 4501 | |
240799cd TT |
4502 | /* Is the inode bitmap in cache? */ |
4503 | bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp)); | |
aebf0243 | 4504 | if (unlikely(!bitmap_bh)) |
ac27a0ec DK |
4505 | goto make_io; |
4506 | ||
4507 | /* | |
4508 | * If the inode bitmap isn't in cache then the | |
4509 | * optimisation may end up performing two reads instead | |
4510 | * of one, so skip it. | |
4511 | */ | |
4512 | if (!buffer_uptodate(bitmap_bh)) { | |
4513 | brelse(bitmap_bh); | |
4514 | goto make_io; | |
4515 | } | |
240799cd | 4516 | for (i = start; i < start + inodes_per_block; i++) { |
ac27a0ec DK |
4517 | if (i == inode_offset) |
4518 | continue; | |
617ba13b | 4519 | if (ext4_test_bit(i, bitmap_bh->b_data)) |
ac27a0ec DK |
4520 | break; |
4521 | } | |
4522 | brelse(bitmap_bh); | |
240799cd | 4523 | if (i == start + inodes_per_block) { |
ac27a0ec DK |
4524 | /* all other inodes are free, so skip I/O */ |
4525 | memset(bh->b_data, 0, bh->b_size); | |
4526 | set_buffer_uptodate(bh); | |
4527 | unlock_buffer(bh); | |
4528 | goto has_buffer; | |
4529 | } | |
4530 | } | |
4531 | ||
4532 | make_io: | |
240799cd TT |
4533 | /* |
4534 | * If we need to do any I/O, try to pre-readahead extra | |
4535 | * blocks from the inode table. | |
4536 | */ | |
4537 | if (EXT4_SB(sb)->s_inode_readahead_blks) { | |
4538 | ext4_fsblk_t b, end, table; | |
4539 | unsigned num; | |
0d606e2c | 4540 | __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks; |
240799cd TT |
4541 | |
4542 | table = ext4_inode_table(sb, gdp); | |
b713a5ec | 4543 | /* s_inode_readahead_blks is always a power of 2 */ |
0d606e2c | 4544 | b = block & ~((ext4_fsblk_t) ra_blks - 1); |
240799cd TT |
4545 | if (table > b) |
4546 | b = table; | |
0d606e2c | 4547 | end = b + ra_blks; |
240799cd | 4548 | num = EXT4_INODES_PER_GROUP(sb); |
feb0ab32 | 4549 | if (ext4_has_group_desc_csum(sb)) |
560671a0 | 4550 | num -= ext4_itable_unused_count(sb, gdp); |
240799cd TT |
4551 | table += num / inodes_per_block; |
4552 | if (end > table) | |
4553 | end = table; | |
4554 | while (b <= end) | |
4555 | sb_breadahead(sb, b++); | |
4556 | } | |
4557 | ||
ac27a0ec DK |
4558 | /* |
4559 | * There are other valid inodes in the buffer, this inode | |
4560 | * has in-inode xattrs, or we don't have this inode in memory. | |
4561 | * Read the block from disk. | |
4562 | */ | |
0562e0ba | 4563 | trace_ext4_load_inode(inode); |
ac27a0ec DK |
4564 | get_bh(bh); |
4565 | bh->b_end_io = end_buffer_read_sync; | |
2a222ca9 | 4566 | submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh); |
ac27a0ec DK |
4567 | wait_on_buffer(bh); |
4568 | if (!buffer_uptodate(bh)) { | |
c398eda0 TT |
4569 | EXT4_ERROR_INODE_BLOCK(inode, block, |
4570 | "unable to read itable block"); | |
ac27a0ec DK |
4571 | brelse(bh); |
4572 | return -EIO; | |
4573 | } | |
4574 | } | |
4575 | has_buffer: | |
4576 | iloc->bh = bh; | |
4577 | return 0; | |
4578 | } | |
4579 | ||
617ba13b | 4580 | int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec DK |
4581 | { |
4582 | /* We have all inode data except xattrs in memory here. */ | |
617ba13b | 4583 | return __ext4_get_inode_loc(inode, iloc, |
19f5fb7a | 4584 | !ext4_test_inode_state(inode, EXT4_STATE_XATTR)); |
ac27a0ec DK |
4585 | } |
4586 | ||
617ba13b | 4587 | void ext4_set_inode_flags(struct inode *inode) |
ac27a0ec | 4588 | { |
617ba13b | 4589 | unsigned int flags = EXT4_I(inode)->i_flags; |
00a1a053 | 4590 | unsigned int new_fl = 0; |
ac27a0ec | 4591 | |
617ba13b | 4592 | if (flags & EXT4_SYNC_FL) |
00a1a053 | 4593 | new_fl |= S_SYNC; |
617ba13b | 4594 | if (flags & EXT4_APPEND_FL) |
00a1a053 | 4595 | new_fl |= S_APPEND; |
617ba13b | 4596 | if (flags & EXT4_IMMUTABLE_FL) |
00a1a053 | 4597 | new_fl |= S_IMMUTABLE; |
617ba13b | 4598 | if (flags & EXT4_NOATIME_FL) |
00a1a053 | 4599 | new_fl |= S_NOATIME; |
617ba13b | 4600 | if (flags & EXT4_DIRSYNC_FL) |
00a1a053 | 4601 | new_fl |= S_DIRSYNC; |
a3caa24b JK |
4602 | if (test_opt(inode->i_sb, DAX) && S_ISREG(inode->i_mode) && |
4603 | !ext4_should_journal_data(inode) && !ext4_has_inline_data(inode) && | |
4604 | !ext4_encrypted_inode(inode)) | |
923ae0ff | 4605 | new_fl |= S_DAX; |
5f16f322 | 4606 | inode_set_flags(inode, new_fl, |
923ae0ff | 4607 | S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX); |
ac27a0ec DK |
4608 | } |
4609 | ||
0fc1b451 | 4610 | static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode, |
de9a55b8 | 4611 | struct ext4_inode_info *ei) |
0fc1b451 AK |
4612 | { |
4613 | blkcnt_t i_blocks ; | |
8180a562 AK |
4614 | struct inode *inode = &(ei->vfs_inode); |
4615 | struct super_block *sb = inode->i_sb; | |
0fc1b451 | 4616 | |
e2b911c5 | 4617 | if (ext4_has_feature_huge_file(sb)) { |
0fc1b451 AK |
4618 | /* we are using combined 48 bit field */ |
4619 | i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 | | |
4620 | le32_to_cpu(raw_inode->i_blocks_lo); | |
07a03824 | 4621 | if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) { |
8180a562 AK |
4622 | /* i_blocks represent file system block size */ |
4623 | return i_blocks << (inode->i_blkbits - 9); | |
4624 | } else { | |
4625 | return i_blocks; | |
4626 | } | |
0fc1b451 AK |
4627 | } else { |
4628 | return le32_to_cpu(raw_inode->i_blocks_lo); | |
4629 | } | |
4630 | } | |
ff9ddf7e | 4631 | |
152a7b0a TM |
4632 | static inline void ext4_iget_extra_inode(struct inode *inode, |
4633 | struct ext4_inode *raw_inode, | |
4634 | struct ext4_inode_info *ei) | |
4635 | { | |
4636 | __le32 *magic = (void *)raw_inode + | |
4637 | EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize; | |
290ab230 EB |
4638 | if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize + sizeof(__le32) <= |
4639 | EXT4_INODE_SIZE(inode->i_sb) && | |
4640 | *magic == cpu_to_le32(EXT4_XATTR_MAGIC)) { | |
152a7b0a | 4641 | ext4_set_inode_state(inode, EXT4_STATE_XATTR); |
67cf5b09 | 4642 | ext4_find_inline_data_nolock(inode); |
f19d5870 TM |
4643 | } else |
4644 | EXT4_I(inode)->i_inline_off = 0; | |
152a7b0a TM |
4645 | } |
4646 | ||
040cb378 LX |
4647 | int ext4_get_projid(struct inode *inode, kprojid_t *projid) |
4648 | { | |
0b7b7779 | 4649 | if (!ext4_has_feature_project(inode->i_sb)) |
040cb378 LX |
4650 | return -EOPNOTSUPP; |
4651 | *projid = EXT4_I(inode)->i_projid; | |
4652 | return 0; | |
4653 | } | |
4654 | ||
1d1fe1ee | 4655 | struct inode *ext4_iget(struct super_block *sb, unsigned long ino) |
ac27a0ec | 4656 | { |
617ba13b MC |
4657 | struct ext4_iloc iloc; |
4658 | struct ext4_inode *raw_inode; | |
1d1fe1ee | 4659 | struct ext4_inode_info *ei; |
1d1fe1ee | 4660 | struct inode *inode; |
b436b9be | 4661 | journal_t *journal = EXT4_SB(sb)->s_journal; |
1d1fe1ee | 4662 | long ret; |
7e6e1ef4 | 4663 | loff_t size; |
ac27a0ec | 4664 | int block; |
08cefc7a EB |
4665 | uid_t i_uid; |
4666 | gid_t i_gid; | |
040cb378 | 4667 | projid_t i_projid; |
ac27a0ec | 4668 | |
1d1fe1ee DH |
4669 | inode = iget_locked(sb, ino); |
4670 | if (!inode) | |
4671 | return ERR_PTR(-ENOMEM); | |
4672 | if (!(inode->i_state & I_NEW)) | |
4673 | return inode; | |
4674 | ||
4675 | ei = EXT4_I(inode); | |
7dc57615 | 4676 | iloc.bh = NULL; |
ac27a0ec | 4677 | |
1d1fe1ee DH |
4678 | ret = __ext4_get_inode_loc(inode, &iloc, 0); |
4679 | if (ret < 0) | |
ac27a0ec | 4680 | goto bad_inode; |
617ba13b | 4681 | raw_inode = ext4_raw_inode(&iloc); |
814525f4 DW |
4682 | |
4683 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { | |
4684 | ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize); | |
4685 | if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > | |
2dc8d9e1 EB |
4686 | EXT4_INODE_SIZE(inode->i_sb) || |
4687 | (ei->i_extra_isize & 3)) { | |
4688 | EXT4_ERROR_INODE(inode, | |
4689 | "bad extra_isize %u (inode size %u)", | |
4690 | ei->i_extra_isize, | |
4691 | EXT4_INODE_SIZE(inode->i_sb)); | |
6a797d27 | 4692 | ret = -EFSCORRUPTED; |
814525f4 DW |
4693 | goto bad_inode; |
4694 | } | |
4695 | } else | |
4696 | ei->i_extra_isize = 0; | |
4697 | ||
4698 | /* Precompute checksum seed for inode metadata */ | |
9aa5d32b | 4699 | if (ext4_has_metadata_csum(sb)) { |
814525f4 DW |
4700 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
4701 | __u32 csum; | |
4702 | __le32 inum = cpu_to_le32(inode->i_ino); | |
4703 | __le32 gen = raw_inode->i_generation; | |
4704 | csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum, | |
4705 | sizeof(inum)); | |
4706 | ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen, | |
4707 | sizeof(gen)); | |
4708 | } | |
4709 | ||
4710 | if (!ext4_inode_csum_verify(inode, raw_inode, ei)) { | |
4711 | EXT4_ERROR_INODE(inode, "checksum invalid"); | |
6a797d27 | 4712 | ret = -EFSBADCRC; |
814525f4 DW |
4713 | goto bad_inode; |
4714 | } | |
4715 | ||
ac27a0ec | 4716 | inode->i_mode = le16_to_cpu(raw_inode->i_mode); |
08cefc7a EB |
4717 | i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low); |
4718 | i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low); | |
0b7b7779 | 4719 | if (ext4_has_feature_project(sb) && |
040cb378 LX |
4720 | EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE && |
4721 | EXT4_FITS_IN_INODE(raw_inode, ei, i_projid)) | |
4722 | i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid); | |
4723 | else | |
4724 | i_projid = EXT4_DEF_PROJID; | |
4725 | ||
af5bc92d | 4726 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
08cefc7a EB |
4727 | i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16; |
4728 | i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16; | |
ac27a0ec | 4729 | } |
08cefc7a EB |
4730 | i_uid_write(inode, i_uid); |
4731 | i_gid_write(inode, i_gid); | |
afdbd4a6 | 4732 | ei->i_projid = make_kprojid(sb->s_user_ns, i_projid); |
bfe86848 | 4733 | set_nlink(inode, le16_to_cpu(raw_inode->i_links_count)); |
ac27a0ec | 4734 | |
353eb83c | 4735 | ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */ |
67cf5b09 | 4736 | ei->i_inline_off = 0; |
ac27a0ec DK |
4737 | ei->i_dir_start_lookup = 0; |
4738 | ei->i_dtime = le32_to_cpu(raw_inode->i_dtime); | |
4739 | /* We now have enough fields to check if the inode was active or not. | |
4740 | * This is needed because nfsd might try to access dead inodes | |
4741 | * the test is that same one that e2fsck uses | |
4742 | * NeilBrown 1999oct15 | |
4743 | */ | |
4744 | if (inode->i_nlink == 0) { | |
393d1d1d DTB |
4745 | if ((inode->i_mode == 0 || |
4746 | !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) && | |
4747 | ino != EXT4_BOOT_LOADER_INO) { | |
ac27a0ec | 4748 | /* this inode is deleted */ |
1d1fe1ee | 4749 | ret = -ESTALE; |
ac27a0ec DK |
4750 | goto bad_inode; |
4751 | } | |
4752 | /* The only unlinked inodes we let through here have | |
4753 | * valid i_mode and are being read by the orphan | |
4754 | * recovery code: that's fine, we're about to complete | |
393d1d1d DTB |
4755 | * the process of deleting those. |
4756 | * OR it is the EXT4_BOOT_LOADER_INO which is | |
4757 | * not initialized on a new filesystem. */ | |
ac27a0ec | 4758 | } |
ac27a0ec | 4759 | ei->i_flags = le32_to_cpu(raw_inode->i_flags); |
0fc1b451 | 4760 | inode->i_blocks = ext4_inode_blocks(raw_inode, ei); |
7973c0c1 | 4761 | ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo); |
e2b911c5 | 4762 | if (ext4_has_feature_64bit(sb)) |
a1ddeb7e BP |
4763 | ei->i_file_acl |= |
4764 | ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32; | |
e08ac99f | 4765 | inode->i_size = ext4_isize(sb, raw_inode); |
7e6e1ef4 DW |
4766 | if ((size = i_size_read(inode)) < 0) { |
4767 | EXT4_ERROR_INODE(inode, "bad i_size value: %lld", size); | |
4768 | ret = -EFSCORRUPTED; | |
4769 | goto bad_inode; | |
4770 | } | |
ac27a0ec | 4771 | ei->i_disksize = inode->i_size; |
a9e7f447 DM |
4772 | #ifdef CONFIG_QUOTA |
4773 | ei->i_reserved_quota = 0; | |
4774 | #endif | |
ac27a0ec DK |
4775 | inode->i_generation = le32_to_cpu(raw_inode->i_generation); |
4776 | ei->i_block_group = iloc.block_group; | |
a4912123 | 4777 | ei->i_last_alloc_group = ~0; |
ac27a0ec DK |
4778 | /* |
4779 | * NOTE! The in-memory inode i_data array is in little-endian order | |
4780 | * even on big-endian machines: we do NOT byteswap the block numbers! | |
4781 | */ | |
617ba13b | 4782 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
ac27a0ec DK |
4783 | ei->i_data[block] = raw_inode->i_block[block]; |
4784 | INIT_LIST_HEAD(&ei->i_orphan); | |
4785 | ||
b436b9be JK |
4786 | /* |
4787 | * Set transaction id's of transactions that have to be committed | |
4788 | * to finish f[data]sync. We set them to currently running transaction | |
4789 | * as we cannot be sure that the inode or some of its metadata isn't | |
4790 | * part of the transaction - the inode could have been reclaimed and | |
4791 | * now it is reread from disk. | |
4792 | */ | |
4793 | if (journal) { | |
4794 | transaction_t *transaction; | |
4795 | tid_t tid; | |
4796 | ||
a931da6a | 4797 | read_lock(&journal->j_state_lock); |
b436b9be JK |
4798 | if (journal->j_running_transaction) |
4799 | transaction = journal->j_running_transaction; | |
4800 | else | |
4801 | transaction = journal->j_committing_transaction; | |
4802 | if (transaction) | |
4803 | tid = transaction->t_tid; | |
4804 | else | |
4805 | tid = journal->j_commit_sequence; | |
a931da6a | 4806 | read_unlock(&journal->j_state_lock); |
b436b9be JK |
4807 | ei->i_sync_tid = tid; |
4808 | ei->i_datasync_tid = tid; | |
4809 | } | |
4810 | ||
0040d987 | 4811 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
ac27a0ec DK |
4812 | if (ei->i_extra_isize == 0) { |
4813 | /* The extra space is currently unused. Use it. */ | |
2dc8d9e1 | 4814 | BUILD_BUG_ON(sizeof(struct ext4_inode) & 3); |
617ba13b MC |
4815 | ei->i_extra_isize = sizeof(struct ext4_inode) - |
4816 | EXT4_GOOD_OLD_INODE_SIZE; | |
ac27a0ec | 4817 | } else { |
152a7b0a | 4818 | ext4_iget_extra_inode(inode, raw_inode, ei); |
ac27a0ec | 4819 | } |
814525f4 | 4820 | } |
ac27a0ec | 4821 | |
ef7f3835 KS |
4822 | EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode); |
4823 | EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode); | |
4824 | EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode); | |
4825 | EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode); | |
4826 | ||
ed3654eb | 4827 | if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) { |
c4f65706 TT |
4828 | inode->i_version = le32_to_cpu(raw_inode->i_disk_version); |
4829 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { | |
4830 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
4831 | inode->i_version |= | |
4832 | (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32; | |
4833 | } | |
25ec56b5 JNC |
4834 | } |
4835 | ||
c4b5a614 | 4836 | ret = 0; |
485c26ec | 4837 | if (ei->i_file_acl && |
1032988c | 4838 | !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) { |
24676da4 TT |
4839 | EXT4_ERROR_INODE(inode, "bad extended attribute block %llu", |
4840 | ei->i_file_acl); | |
6a797d27 | 4841 | ret = -EFSCORRUPTED; |
485c26ec | 4842 | goto bad_inode; |
f19d5870 TM |
4843 | } else if (!ext4_has_inline_data(inode)) { |
4844 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { | |
4845 | if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || | |
4846 | (S_ISLNK(inode->i_mode) && | |
4847 | !ext4_inode_is_fast_symlink(inode)))) | |
4848 | /* Validate extent which is part of inode */ | |
4849 | ret = ext4_ext_check_inode(inode); | |
4850 | } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || | |
4851 | (S_ISLNK(inode->i_mode) && | |
4852 | !ext4_inode_is_fast_symlink(inode))) { | |
4853 | /* Validate block references which are part of inode */ | |
4854 | ret = ext4_ind_check_inode(inode); | |
4855 | } | |
fe2c8191 | 4856 | } |
567f3e9a | 4857 | if (ret) |
de9a55b8 | 4858 | goto bad_inode; |
7a262f7c | 4859 | |
ac27a0ec | 4860 | if (S_ISREG(inode->i_mode)) { |
617ba13b | 4861 | inode->i_op = &ext4_file_inode_operations; |
be64f884 | 4862 | inode->i_fop = &ext4_file_operations; |
617ba13b | 4863 | ext4_set_aops(inode); |
ac27a0ec | 4864 | } else if (S_ISDIR(inode->i_mode)) { |
617ba13b MC |
4865 | inode->i_op = &ext4_dir_inode_operations; |
4866 | inode->i_fop = &ext4_dir_operations; | |
ac27a0ec | 4867 | } else if (S_ISLNK(inode->i_mode)) { |
a7a67e8a AV |
4868 | if (ext4_encrypted_inode(inode)) { |
4869 | inode->i_op = &ext4_encrypted_symlink_inode_operations; | |
4870 | ext4_set_aops(inode); | |
4871 | } else if (ext4_inode_is_fast_symlink(inode)) { | |
75e7566b | 4872 | inode->i_link = (char *)ei->i_data; |
617ba13b | 4873 | inode->i_op = &ext4_fast_symlink_inode_operations; |
e83c1397 DG |
4874 | nd_terminate_link(ei->i_data, inode->i_size, |
4875 | sizeof(ei->i_data) - 1); | |
4876 | } else { | |
617ba13b MC |
4877 | inode->i_op = &ext4_symlink_inode_operations; |
4878 | ext4_set_aops(inode); | |
ac27a0ec | 4879 | } |
21fc61c7 | 4880 | inode_nohighmem(inode); |
563bdd61 TT |
4881 | } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) || |
4882 | S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) { | |
617ba13b | 4883 | inode->i_op = &ext4_special_inode_operations; |
ac27a0ec DK |
4884 | if (raw_inode->i_block[0]) |
4885 | init_special_inode(inode, inode->i_mode, | |
4886 | old_decode_dev(le32_to_cpu(raw_inode->i_block[0]))); | |
4887 | else | |
4888 | init_special_inode(inode, inode->i_mode, | |
4889 | new_decode_dev(le32_to_cpu(raw_inode->i_block[1]))); | |
393d1d1d DTB |
4890 | } else if (ino == EXT4_BOOT_LOADER_INO) { |
4891 | make_bad_inode(inode); | |
563bdd61 | 4892 | } else { |
6a797d27 | 4893 | ret = -EFSCORRUPTED; |
24676da4 | 4894 | EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode); |
563bdd61 | 4895 | goto bad_inode; |
ac27a0ec | 4896 | } |
af5bc92d | 4897 | brelse(iloc.bh); |
617ba13b | 4898 | ext4_set_inode_flags(inode); |
dec214d0 TE |
4899 | |
4900 | if (ei->i_flags & EXT4_EA_INODE_FL) { | |
33d201e0 | 4901 | ext4_xattr_inode_set_class(inode); |
dec214d0 TE |
4902 | |
4903 | inode_lock(inode); | |
4904 | inode->i_flags |= S_NOQUOTA; | |
4905 | inode_unlock(inode); | |
4906 | } | |
4907 | ||
1d1fe1ee DH |
4908 | unlock_new_inode(inode); |
4909 | return inode; | |
ac27a0ec DK |
4910 | |
4911 | bad_inode: | |
567f3e9a | 4912 | brelse(iloc.bh); |
1d1fe1ee DH |
4913 | iget_failed(inode); |
4914 | return ERR_PTR(ret); | |
ac27a0ec DK |
4915 | } |
4916 | ||
f4bb2981 TT |
4917 | struct inode *ext4_iget_normal(struct super_block *sb, unsigned long ino) |
4918 | { | |
4919 | if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO) | |
6a797d27 | 4920 | return ERR_PTR(-EFSCORRUPTED); |
f4bb2981 TT |
4921 | return ext4_iget(sb, ino); |
4922 | } | |
4923 | ||
0fc1b451 AK |
4924 | static int ext4_inode_blocks_set(handle_t *handle, |
4925 | struct ext4_inode *raw_inode, | |
4926 | struct ext4_inode_info *ei) | |
4927 | { | |
4928 | struct inode *inode = &(ei->vfs_inode); | |
4929 | u64 i_blocks = inode->i_blocks; | |
4930 | struct super_block *sb = inode->i_sb; | |
0fc1b451 AK |
4931 | |
4932 | if (i_blocks <= ~0U) { | |
4933 | /* | |
4907cb7b | 4934 | * i_blocks can be represented in a 32 bit variable |
0fc1b451 AK |
4935 | * as multiple of 512 bytes |
4936 | */ | |
8180a562 | 4937 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 4938 | raw_inode->i_blocks_high = 0; |
84a8dce2 | 4939 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
f287a1a5 TT |
4940 | return 0; |
4941 | } | |
e2b911c5 | 4942 | if (!ext4_has_feature_huge_file(sb)) |
f287a1a5 TT |
4943 | return -EFBIG; |
4944 | ||
4945 | if (i_blocks <= 0xffffffffffffULL) { | |
0fc1b451 AK |
4946 | /* |
4947 | * i_blocks can be represented in a 48 bit variable | |
4948 | * as multiple of 512 bytes | |
4949 | */ | |
8180a562 | 4950 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 4951 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); |
84a8dce2 | 4952 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
0fc1b451 | 4953 | } else { |
84a8dce2 | 4954 | ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
8180a562 AK |
4955 | /* i_block is stored in file system block size */ |
4956 | i_blocks = i_blocks >> (inode->i_blkbits - 9); | |
4957 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); | |
4958 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); | |
0fc1b451 | 4959 | } |
f287a1a5 | 4960 | return 0; |
0fc1b451 AK |
4961 | } |
4962 | ||
a26f4992 TT |
4963 | struct other_inode { |
4964 | unsigned long orig_ino; | |
4965 | struct ext4_inode *raw_inode; | |
4966 | }; | |
4967 | ||
4968 | static int other_inode_match(struct inode * inode, unsigned long ino, | |
4969 | void *data) | |
4970 | { | |
4971 | struct other_inode *oi = (struct other_inode *) data; | |
4972 | ||
4973 | if ((inode->i_ino != ino) || | |
4974 | (inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW | | |
4975 | I_DIRTY_SYNC | I_DIRTY_DATASYNC)) || | |
4976 | ((inode->i_state & I_DIRTY_TIME) == 0)) | |
4977 | return 0; | |
4978 | spin_lock(&inode->i_lock); | |
4979 | if (((inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW | | |
4980 | I_DIRTY_SYNC | I_DIRTY_DATASYNC)) == 0) && | |
4981 | (inode->i_state & I_DIRTY_TIME)) { | |
4982 | struct ext4_inode_info *ei = EXT4_I(inode); | |
4983 | ||
4984 | inode->i_state &= ~(I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED); | |
4985 | spin_unlock(&inode->i_lock); | |
4986 | ||
4987 | spin_lock(&ei->i_raw_lock); | |
4988 | EXT4_INODE_SET_XTIME(i_ctime, inode, oi->raw_inode); | |
4989 | EXT4_INODE_SET_XTIME(i_mtime, inode, oi->raw_inode); | |
4990 | EXT4_INODE_SET_XTIME(i_atime, inode, oi->raw_inode); | |
4991 | ext4_inode_csum_set(inode, oi->raw_inode, ei); | |
4992 | spin_unlock(&ei->i_raw_lock); | |
4993 | trace_ext4_other_inode_update_time(inode, oi->orig_ino); | |
4994 | return -1; | |
4995 | } | |
4996 | spin_unlock(&inode->i_lock); | |
4997 | return -1; | |
4998 | } | |
4999 | ||
5000 | /* | |
5001 | * Opportunistically update the other time fields for other inodes in | |
5002 | * the same inode table block. | |
5003 | */ | |
5004 | static void ext4_update_other_inodes_time(struct super_block *sb, | |
5005 | unsigned long orig_ino, char *buf) | |
5006 | { | |
5007 | struct other_inode oi; | |
5008 | unsigned long ino; | |
5009 | int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; | |
5010 | int inode_size = EXT4_INODE_SIZE(sb); | |
5011 | ||
5012 | oi.orig_ino = orig_ino; | |
0f0ff9a9 TT |
5013 | /* |
5014 | * Calculate the first inode in the inode table block. Inode | |
5015 | * numbers are one-based. That is, the first inode in a block | |
5016 | * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1). | |
5017 | */ | |
5018 | ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1; | |
a26f4992 TT |
5019 | for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) { |
5020 | if (ino == orig_ino) | |
5021 | continue; | |
5022 | oi.raw_inode = (struct ext4_inode *) buf; | |
5023 | (void) find_inode_nowait(sb, ino, other_inode_match, &oi); | |
5024 | } | |
5025 | } | |
5026 | ||
ac27a0ec DK |
5027 | /* |
5028 | * Post the struct inode info into an on-disk inode location in the | |
5029 | * buffer-cache. This gobbles the caller's reference to the | |
5030 | * buffer_head in the inode location struct. | |
5031 | * | |
5032 | * The caller must have write access to iloc->bh. | |
5033 | */ | |
617ba13b | 5034 | static int ext4_do_update_inode(handle_t *handle, |
ac27a0ec | 5035 | struct inode *inode, |
830156c7 | 5036 | struct ext4_iloc *iloc) |
ac27a0ec | 5037 | { |
617ba13b MC |
5038 | struct ext4_inode *raw_inode = ext4_raw_inode(iloc); |
5039 | struct ext4_inode_info *ei = EXT4_I(inode); | |
ac27a0ec | 5040 | struct buffer_head *bh = iloc->bh; |
202ee5df | 5041 | struct super_block *sb = inode->i_sb; |
ac27a0ec | 5042 | int err = 0, rc, block; |
202ee5df | 5043 | int need_datasync = 0, set_large_file = 0; |
08cefc7a EB |
5044 | uid_t i_uid; |
5045 | gid_t i_gid; | |
040cb378 | 5046 | projid_t i_projid; |
ac27a0ec | 5047 | |
202ee5df TT |
5048 | spin_lock(&ei->i_raw_lock); |
5049 | ||
5050 | /* For fields not tracked in the in-memory inode, | |
ac27a0ec | 5051 | * initialise them to zero for new inodes. */ |
19f5fb7a | 5052 | if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) |
617ba13b | 5053 | memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size); |
ac27a0ec DK |
5054 | |
5055 | raw_inode->i_mode = cpu_to_le16(inode->i_mode); | |
08cefc7a EB |
5056 | i_uid = i_uid_read(inode); |
5057 | i_gid = i_gid_read(inode); | |
afdbd4a6 | 5058 | i_projid = from_kprojid(sb->s_user_ns, ei->i_projid); |
af5bc92d | 5059 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
08cefc7a EB |
5060 | raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid)); |
5061 | raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid)); | |
ac27a0ec DK |
5062 | /* |
5063 | * Fix up interoperability with old kernels. Otherwise, old inodes get | |
5064 | * re-used with the upper 16 bits of the uid/gid intact | |
5065 | */ | |
93e3b4e6 DJ |
5066 | if (ei->i_dtime && list_empty(&ei->i_orphan)) { |
5067 | raw_inode->i_uid_high = 0; | |
5068 | raw_inode->i_gid_high = 0; | |
5069 | } else { | |
ac27a0ec | 5070 | raw_inode->i_uid_high = |
08cefc7a | 5071 | cpu_to_le16(high_16_bits(i_uid)); |
ac27a0ec | 5072 | raw_inode->i_gid_high = |
08cefc7a | 5073 | cpu_to_le16(high_16_bits(i_gid)); |
ac27a0ec DK |
5074 | } |
5075 | } else { | |
08cefc7a EB |
5076 | raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid)); |
5077 | raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid)); | |
ac27a0ec DK |
5078 | raw_inode->i_uid_high = 0; |
5079 | raw_inode->i_gid_high = 0; | |
5080 | } | |
5081 | raw_inode->i_links_count = cpu_to_le16(inode->i_nlink); | |
ef7f3835 KS |
5082 | |
5083 | EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode); | |
5084 | EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode); | |
5085 | EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode); | |
5086 | EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode); | |
5087 | ||
bce92d56 LX |
5088 | err = ext4_inode_blocks_set(handle, raw_inode, ei); |
5089 | if (err) { | |
202ee5df | 5090 | spin_unlock(&ei->i_raw_lock); |
0fc1b451 | 5091 | goto out_brelse; |
202ee5df | 5092 | } |
ac27a0ec | 5093 | raw_inode->i_dtime = cpu_to_le32(ei->i_dtime); |
353eb83c | 5094 | raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF); |
ed3654eb | 5095 | if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) |
a1ddeb7e BP |
5096 | raw_inode->i_file_acl_high = |
5097 | cpu_to_le16(ei->i_file_acl >> 32); | |
7973c0c1 | 5098 | raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl); |
e08ac99f | 5099 | if (ei->i_disksize != ext4_isize(inode->i_sb, raw_inode)) { |
b71fc079 JK |
5100 | ext4_isize_set(raw_inode, ei->i_disksize); |
5101 | need_datasync = 1; | |
5102 | } | |
a48380f7 | 5103 | if (ei->i_disksize > 0x7fffffffULL) { |
e2b911c5 | 5104 | if (!ext4_has_feature_large_file(sb) || |
a48380f7 | 5105 | EXT4_SB(sb)->s_es->s_rev_level == |
202ee5df TT |
5106 | cpu_to_le32(EXT4_GOOD_OLD_REV)) |
5107 | set_large_file = 1; | |
ac27a0ec DK |
5108 | } |
5109 | raw_inode->i_generation = cpu_to_le32(inode->i_generation); | |
5110 | if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) { | |
5111 | if (old_valid_dev(inode->i_rdev)) { | |
5112 | raw_inode->i_block[0] = | |
5113 | cpu_to_le32(old_encode_dev(inode->i_rdev)); | |
5114 | raw_inode->i_block[1] = 0; | |
5115 | } else { | |
5116 | raw_inode->i_block[0] = 0; | |
5117 | raw_inode->i_block[1] = | |
5118 | cpu_to_le32(new_encode_dev(inode->i_rdev)); | |
5119 | raw_inode->i_block[2] = 0; | |
5120 | } | |
f19d5870 | 5121 | } else if (!ext4_has_inline_data(inode)) { |
de9a55b8 TT |
5122 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
5123 | raw_inode->i_block[block] = ei->i_data[block]; | |
f19d5870 | 5124 | } |
ac27a0ec | 5125 | |
ed3654eb | 5126 | if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) { |
c4f65706 TT |
5127 | raw_inode->i_disk_version = cpu_to_le32(inode->i_version); |
5128 | if (ei->i_extra_isize) { | |
5129 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
5130 | raw_inode->i_version_hi = | |
5131 | cpu_to_le32(inode->i_version >> 32); | |
5132 | raw_inode->i_extra_isize = | |
5133 | cpu_to_le16(ei->i_extra_isize); | |
5134 | } | |
25ec56b5 | 5135 | } |
040cb378 | 5136 | |
afdbd4a6 SF |
5137 | if (i_projid != (projid_t)-1) { |
5138 | BUG_ON(!ext4_has_feature_project(inode->i_sb) && | |
5139 | i_projid != EXT4_DEF_PROJID); | |
040cb378 | 5140 | |
afdbd4a6 SF |
5141 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && |
5142 | EXT4_FITS_IN_INODE(raw_inode, ei, i_projid)) | |
5143 | raw_inode->i_projid = cpu_to_le32(i_projid); | |
5144 | } | |
040cb378 | 5145 | |
814525f4 | 5146 | ext4_inode_csum_set(inode, raw_inode, ei); |
202ee5df | 5147 | spin_unlock(&ei->i_raw_lock); |
a26f4992 TT |
5148 | if (inode->i_sb->s_flags & MS_LAZYTIME) |
5149 | ext4_update_other_inodes_time(inode->i_sb, inode->i_ino, | |
5150 | bh->b_data); | |
202ee5df | 5151 | |
830156c7 | 5152 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
73b50c1c | 5153 | rc = ext4_handle_dirty_metadata(handle, NULL, bh); |
830156c7 FM |
5154 | if (!err) |
5155 | err = rc; | |
19f5fb7a | 5156 | ext4_clear_inode_state(inode, EXT4_STATE_NEW); |
202ee5df | 5157 | if (set_large_file) { |
5d601255 | 5158 | BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access"); |
202ee5df TT |
5159 | err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh); |
5160 | if (err) | |
5161 | goto out_brelse; | |
5162 | ext4_update_dynamic_rev(sb); | |
e2b911c5 | 5163 | ext4_set_feature_large_file(sb); |
202ee5df TT |
5164 | ext4_handle_sync(handle); |
5165 | err = ext4_handle_dirty_super(handle, sb); | |
5166 | } | |
b71fc079 | 5167 | ext4_update_inode_fsync_trans(handle, inode, need_datasync); |
ac27a0ec | 5168 | out_brelse: |
af5bc92d | 5169 | brelse(bh); |
617ba13b | 5170 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
5171 | return err; |
5172 | } | |
5173 | ||
5174 | /* | |
617ba13b | 5175 | * ext4_write_inode() |
ac27a0ec DK |
5176 | * |
5177 | * We are called from a few places: | |
5178 | * | |
87f7e416 | 5179 | * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files. |
ac27a0ec | 5180 | * Here, there will be no transaction running. We wait for any running |
4907cb7b | 5181 | * transaction to commit. |
ac27a0ec | 5182 | * |
87f7e416 TT |
5183 | * - Within flush work (sys_sync(), kupdate and such). |
5184 | * We wait on commit, if told to. | |
ac27a0ec | 5185 | * |
87f7e416 TT |
5186 | * - Within iput_final() -> write_inode_now() |
5187 | * We wait on commit, if told to. | |
ac27a0ec DK |
5188 | * |
5189 | * In all cases it is actually safe for us to return without doing anything, | |
5190 | * because the inode has been copied into a raw inode buffer in | |
87f7e416 TT |
5191 | * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL |
5192 | * writeback. | |
ac27a0ec DK |
5193 | * |
5194 | * Note that we are absolutely dependent upon all inode dirtiers doing the | |
5195 | * right thing: they *must* call mark_inode_dirty() after dirtying info in | |
5196 | * which we are interested. | |
5197 | * | |
5198 | * It would be a bug for them to not do this. The code: | |
5199 | * | |
5200 | * mark_inode_dirty(inode) | |
5201 | * stuff(); | |
5202 | * inode->i_size = expr; | |
5203 | * | |
87f7e416 TT |
5204 | * is in error because write_inode() could occur while `stuff()' is running, |
5205 | * and the new i_size will be lost. Plus the inode will no longer be on the | |
5206 | * superblock's dirty inode list. | |
ac27a0ec | 5207 | */ |
a9185b41 | 5208 | int ext4_write_inode(struct inode *inode, struct writeback_control *wbc) |
ac27a0ec | 5209 | { |
91ac6f43 FM |
5210 | int err; |
5211 | ||
87f7e416 | 5212 | if (WARN_ON_ONCE(current->flags & PF_MEMALLOC)) |
ac27a0ec DK |
5213 | return 0; |
5214 | ||
91ac6f43 FM |
5215 | if (EXT4_SB(inode->i_sb)->s_journal) { |
5216 | if (ext4_journal_current_handle()) { | |
5217 | jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n"); | |
5218 | dump_stack(); | |
5219 | return -EIO; | |
5220 | } | |
ac27a0ec | 5221 | |
10542c22 JK |
5222 | /* |
5223 | * No need to force transaction in WB_SYNC_NONE mode. Also | |
5224 | * ext4_sync_fs() will force the commit after everything is | |
5225 | * written. | |
5226 | */ | |
5227 | if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync) | |
91ac6f43 FM |
5228 | return 0; |
5229 | ||
5230 | err = ext4_force_commit(inode->i_sb); | |
5231 | } else { | |
5232 | struct ext4_iloc iloc; | |
ac27a0ec | 5233 | |
8b472d73 | 5234 | err = __ext4_get_inode_loc(inode, &iloc, 0); |
91ac6f43 FM |
5235 | if (err) |
5236 | return err; | |
10542c22 JK |
5237 | /* |
5238 | * sync(2) will flush the whole buffer cache. No need to do | |
5239 | * it here separately for each inode. | |
5240 | */ | |
5241 | if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) | |
830156c7 FM |
5242 | sync_dirty_buffer(iloc.bh); |
5243 | if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) { | |
c398eda0 TT |
5244 | EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr, |
5245 | "IO error syncing inode"); | |
830156c7 FM |
5246 | err = -EIO; |
5247 | } | |
fd2dd9fb | 5248 | brelse(iloc.bh); |
91ac6f43 FM |
5249 | } |
5250 | return err; | |
ac27a0ec DK |
5251 | } |
5252 | ||
53e87268 JK |
5253 | /* |
5254 | * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate | |
5255 | * buffers that are attached to a page stradding i_size and are undergoing | |
5256 | * commit. In that case we have to wait for commit to finish and try again. | |
5257 | */ | |
5258 | static void ext4_wait_for_tail_page_commit(struct inode *inode) | |
5259 | { | |
5260 | struct page *page; | |
5261 | unsigned offset; | |
5262 | journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; | |
5263 | tid_t commit_tid = 0; | |
5264 | int ret; | |
5265 | ||
09cbfeaf | 5266 | offset = inode->i_size & (PAGE_SIZE - 1); |
53e87268 JK |
5267 | /* |
5268 | * All buffers in the last page remain valid? Then there's nothing to | |
ea1754a0 | 5269 | * do. We do the check mainly to optimize the common PAGE_SIZE == |
53e87268 JK |
5270 | * blocksize case |
5271 | */ | |
93407472 | 5272 | if (offset > PAGE_SIZE - i_blocksize(inode)) |
53e87268 JK |
5273 | return; |
5274 | while (1) { | |
5275 | page = find_lock_page(inode->i_mapping, | |
09cbfeaf | 5276 | inode->i_size >> PAGE_SHIFT); |
53e87268 JK |
5277 | if (!page) |
5278 | return; | |
ca99fdd2 | 5279 | ret = __ext4_journalled_invalidatepage(page, offset, |
09cbfeaf | 5280 | PAGE_SIZE - offset); |
53e87268 | 5281 | unlock_page(page); |
09cbfeaf | 5282 | put_page(page); |
53e87268 JK |
5283 | if (ret != -EBUSY) |
5284 | return; | |
5285 | commit_tid = 0; | |
5286 | read_lock(&journal->j_state_lock); | |
5287 | if (journal->j_committing_transaction) | |
5288 | commit_tid = journal->j_committing_transaction->t_tid; | |
5289 | read_unlock(&journal->j_state_lock); | |
5290 | if (commit_tid) | |
5291 | jbd2_log_wait_commit(journal, commit_tid); | |
5292 | } | |
5293 | } | |
5294 | ||
ac27a0ec | 5295 | /* |
617ba13b | 5296 | * ext4_setattr() |
ac27a0ec DK |
5297 | * |
5298 | * Called from notify_change. | |
5299 | * | |
5300 | * We want to trap VFS attempts to truncate the file as soon as | |
5301 | * possible. In particular, we want to make sure that when the VFS | |
5302 | * shrinks i_size, we put the inode on the orphan list and modify | |
5303 | * i_disksize immediately, so that during the subsequent flushing of | |
5304 | * dirty pages and freeing of disk blocks, we can guarantee that any | |
5305 | * commit will leave the blocks being flushed in an unused state on | |
5306 | * disk. (On recovery, the inode will get truncated and the blocks will | |
5307 | * be freed, so we have a strong guarantee that no future commit will | |
5308 | * leave these blocks visible to the user.) | |
5309 | * | |
678aaf48 JK |
5310 | * Another thing we have to assure is that if we are in ordered mode |
5311 | * and inode is still attached to the committing transaction, we must | |
5312 | * we start writeout of all the dirty pages which are being truncated. | |
5313 | * This way we are sure that all the data written in the previous | |
5314 | * transaction are already on disk (truncate waits for pages under | |
5315 | * writeback). | |
5316 | * | |
5317 | * Called with inode->i_mutex down. | |
ac27a0ec | 5318 | */ |
617ba13b | 5319 | int ext4_setattr(struct dentry *dentry, struct iattr *attr) |
ac27a0ec | 5320 | { |
2b0143b5 | 5321 | struct inode *inode = d_inode(dentry); |
ac27a0ec | 5322 | int error, rc = 0; |
3d287de3 | 5323 | int orphan = 0; |
ac27a0ec DK |
5324 | const unsigned int ia_valid = attr->ia_valid; |
5325 | ||
0db1ff22 TT |
5326 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
5327 | return -EIO; | |
5328 | ||
31051c85 | 5329 | error = setattr_prepare(dentry, attr); |
ac27a0ec DK |
5330 | if (error) |
5331 | return error; | |
5332 | ||
a7cdadee JK |
5333 | if (is_quota_modification(inode, attr)) { |
5334 | error = dquot_initialize(inode); | |
5335 | if (error) | |
5336 | return error; | |
5337 | } | |
08cefc7a EB |
5338 | if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) || |
5339 | (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) { | |
ac27a0ec DK |
5340 | handle_t *handle; |
5341 | ||
5342 | /* (user+group)*(old+new) structure, inode write (sb, | |
5343 | * inode block, ? - but truncate inode update has it) */ | |
9924a92a TT |
5344 | handle = ext4_journal_start(inode, EXT4_HT_QUOTA, |
5345 | (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) + | |
5346 | EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3); | |
ac27a0ec DK |
5347 | if (IS_ERR(handle)) { |
5348 | error = PTR_ERR(handle); | |
5349 | goto err_out; | |
5350 | } | |
7a9ca53a TE |
5351 | |
5352 | /* dquot_transfer() calls back ext4_get_inode_usage() which | |
5353 | * counts xattr inode references. | |
5354 | */ | |
5355 | down_read(&EXT4_I(inode)->xattr_sem); | |
b43fa828 | 5356 | error = dquot_transfer(inode, attr); |
7a9ca53a TE |
5357 | up_read(&EXT4_I(inode)->xattr_sem); |
5358 | ||
ac27a0ec | 5359 | if (error) { |
617ba13b | 5360 | ext4_journal_stop(handle); |
ac27a0ec DK |
5361 | return error; |
5362 | } | |
5363 | /* Update corresponding info in inode so that everything is in | |
5364 | * one transaction */ | |
5365 | if (attr->ia_valid & ATTR_UID) | |
5366 | inode->i_uid = attr->ia_uid; | |
5367 | if (attr->ia_valid & ATTR_GID) | |
5368 | inode->i_gid = attr->ia_gid; | |
617ba13b MC |
5369 | error = ext4_mark_inode_dirty(handle, inode); |
5370 | ext4_journal_stop(handle); | |
ac27a0ec DK |
5371 | } |
5372 | ||
3da40c7b | 5373 | if (attr->ia_valid & ATTR_SIZE) { |
5208386c | 5374 | handle_t *handle; |
3da40c7b JB |
5375 | loff_t oldsize = inode->i_size; |
5376 | int shrink = (attr->ia_size <= inode->i_size); | |
562c72aa | 5377 | |
63136858 EB |
5378 | if (ext4_encrypted_inode(inode)) { |
5379 | error = fscrypt_get_encryption_info(inode); | |
5380 | if (error) | |
5381 | return error; | |
5382 | if (!fscrypt_has_encryption_key(inode)) | |
5383 | return -ENOKEY; | |
5384 | } | |
5385 | ||
12e9b892 | 5386 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) { |
e2b46574 ES |
5387 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
5388 | ||
0c095c7f TT |
5389 | if (attr->ia_size > sbi->s_bitmap_maxbytes) |
5390 | return -EFBIG; | |
e2b46574 | 5391 | } |
3da40c7b JB |
5392 | if (!S_ISREG(inode->i_mode)) |
5393 | return -EINVAL; | |
dff6efc3 CH |
5394 | |
5395 | if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size) | |
5396 | inode_inc_iversion(inode); | |
5397 | ||
3da40c7b | 5398 | if (ext4_should_order_data(inode) && |
5208386c | 5399 | (attr->ia_size < inode->i_size)) { |
3da40c7b | 5400 | error = ext4_begin_ordered_truncate(inode, |
678aaf48 | 5401 | attr->ia_size); |
3da40c7b JB |
5402 | if (error) |
5403 | goto err_out; | |
5404 | } | |
5405 | if (attr->ia_size != inode->i_size) { | |
5208386c JK |
5406 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 3); |
5407 | if (IS_ERR(handle)) { | |
5408 | error = PTR_ERR(handle); | |
5409 | goto err_out; | |
5410 | } | |
3da40c7b | 5411 | if (ext4_handle_valid(handle) && shrink) { |
5208386c JK |
5412 | error = ext4_orphan_add(handle, inode); |
5413 | orphan = 1; | |
5414 | } | |
911af577 EG |
5415 | /* |
5416 | * Update c/mtime on truncate up, ext4_truncate() will | |
5417 | * update c/mtime in shrink case below | |
5418 | */ | |
5419 | if (!shrink) { | |
eeca7ea1 | 5420 | inode->i_mtime = current_time(inode); |
911af577 EG |
5421 | inode->i_ctime = inode->i_mtime; |
5422 | } | |
90e775b7 | 5423 | down_write(&EXT4_I(inode)->i_data_sem); |
5208386c JK |
5424 | EXT4_I(inode)->i_disksize = attr->ia_size; |
5425 | rc = ext4_mark_inode_dirty(handle, inode); | |
5426 | if (!error) | |
5427 | error = rc; | |
90e775b7 JK |
5428 | /* |
5429 | * We have to update i_size under i_data_sem together | |
5430 | * with i_disksize to avoid races with writeback code | |
5431 | * running ext4_wb_update_i_disksize(). | |
5432 | */ | |
5433 | if (!error) | |
5434 | i_size_write(inode, attr->ia_size); | |
5435 | up_write(&EXT4_I(inode)->i_data_sem); | |
5208386c JK |
5436 | ext4_journal_stop(handle); |
5437 | if (error) { | |
3da40c7b JB |
5438 | if (orphan) |
5439 | ext4_orphan_del(NULL, inode); | |
678aaf48 JK |
5440 | goto err_out; |
5441 | } | |
d6320cbf | 5442 | } |
3da40c7b JB |
5443 | if (!shrink) |
5444 | pagecache_isize_extended(inode, oldsize, inode->i_size); | |
53e87268 | 5445 | |
5208386c JK |
5446 | /* |
5447 | * Blocks are going to be removed from the inode. Wait | |
5448 | * for dio in flight. Temporarily disable | |
5449 | * dioread_nolock to prevent livelock. | |
5450 | */ | |
5451 | if (orphan) { | |
5452 | if (!ext4_should_journal_data(inode)) { | |
5453 | ext4_inode_block_unlocked_dio(inode); | |
5454 | inode_dio_wait(inode); | |
5455 | ext4_inode_resume_unlocked_dio(inode); | |
5456 | } else | |
5457 | ext4_wait_for_tail_page_commit(inode); | |
1c9114f9 | 5458 | } |
ea3d7209 | 5459 | down_write(&EXT4_I(inode)->i_mmap_sem); |
5208386c JK |
5460 | /* |
5461 | * Truncate pagecache after we've waited for commit | |
5462 | * in data=journal mode to make pages freeable. | |
5463 | */ | |
923ae0ff | 5464 | truncate_pagecache(inode, inode->i_size); |
2c98eb5e TT |
5465 | if (shrink) { |
5466 | rc = ext4_truncate(inode); | |
5467 | if (rc) | |
5468 | error = rc; | |
5469 | } | |
ea3d7209 | 5470 | up_write(&EXT4_I(inode)->i_mmap_sem); |
072bd7ea | 5471 | } |
ac27a0ec | 5472 | |
2c98eb5e | 5473 | if (!error) { |
1025774c CH |
5474 | setattr_copy(inode, attr); |
5475 | mark_inode_dirty(inode); | |
5476 | } | |
5477 | ||
5478 | /* | |
5479 | * If the call to ext4_truncate failed to get a transaction handle at | |
5480 | * all, we need to clean up the in-core orphan list manually. | |
5481 | */ | |
3d287de3 | 5482 | if (orphan && inode->i_nlink) |
617ba13b | 5483 | ext4_orphan_del(NULL, inode); |
ac27a0ec | 5484 | |
2c98eb5e | 5485 | if (!error && (ia_valid & ATTR_MODE)) |
64e178a7 | 5486 | rc = posix_acl_chmod(inode, inode->i_mode); |
ac27a0ec DK |
5487 | |
5488 | err_out: | |
617ba13b | 5489 | ext4_std_error(inode->i_sb, error); |
ac27a0ec DK |
5490 | if (!error) |
5491 | error = rc; | |
5492 | return error; | |
5493 | } | |
5494 | ||
a528d35e DH |
5495 | int ext4_getattr(const struct path *path, struct kstat *stat, |
5496 | u32 request_mask, unsigned int query_flags) | |
3e3398a0 | 5497 | { |
99652ea5 DH |
5498 | struct inode *inode = d_inode(path->dentry); |
5499 | struct ext4_inode *raw_inode; | |
5500 | struct ext4_inode_info *ei = EXT4_I(inode); | |
5501 | unsigned int flags; | |
5502 | ||
5503 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_crtime)) { | |
5504 | stat->result_mask |= STATX_BTIME; | |
5505 | stat->btime.tv_sec = ei->i_crtime.tv_sec; | |
5506 | stat->btime.tv_nsec = ei->i_crtime.tv_nsec; | |
5507 | } | |
5508 | ||
5509 | flags = ei->i_flags & EXT4_FL_USER_VISIBLE; | |
5510 | if (flags & EXT4_APPEND_FL) | |
5511 | stat->attributes |= STATX_ATTR_APPEND; | |
5512 | if (flags & EXT4_COMPR_FL) | |
5513 | stat->attributes |= STATX_ATTR_COMPRESSED; | |
5514 | if (flags & EXT4_ENCRYPT_FL) | |
5515 | stat->attributes |= STATX_ATTR_ENCRYPTED; | |
5516 | if (flags & EXT4_IMMUTABLE_FL) | |
5517 | stat->attributes |= STATX_ATTR_IMMUTABLE; | |
5518 | if (flags & EXT4_NODUMP_FL) | |
5519 | stat->attributes |= STATX_ATTR_NODUMP; | |
3e3398a0 | 5520 | |
3209f68b DH |
5521 | stat->attributes_mask |= (STATX_ATTR_APPEND | |
5522 | STATX_ATTR_COMPRESSED | | |
5523 | STATX_ATTR_ENCRYPTED | | |
5524 | STATX_ATTR_IMMUTABLE | | |
5525 | STATX_ATTR_NODUMP); | |
5526 | ||
3e3398a0 | 5527 | generic_fillattr(inode, stat); |
99652ea5 DH |
5528 | return 0; |
5529 | } | |
5530 | ||
5531 | int ext4_file_getattr(const struct path *path, struct kstat *stat, | |
5532 | u32 request_mask, unsigned int query_flags) | |
5533 | { | |
5534 | struct inode *inode = d_inode(path->dentry); | |
5535 | u64 delalloc_blocks; | |
5536 | ||
5537 | ext4_getattr(path, stat, request_mask, query_flags); | |
3e3398a0 | 5538 | |
9206c561 AD |
5539 | /* |
5540 | * If there is inline data in the inode, the inode will normally not | |
5541 | * have data blocks allocated (it may have an external xattr block). | |
5542 | * Report at least one sector for such files, so tools like tar, rsync, | |
d67d64f4 | 5543 | * others don't incorrectly think the file is completely sparse. |
9206c561 AD |
5544 | */ |
5545 | if (unlikely(ext4_has_inline_data(inode))) | |
5546 | stat->blocks += (stat->size + 511) >> 9; | |
5547 | ||
3e3398a0 MC |
5548 | /* |
5549 | * We can't update i_blocks if the block allocation is delayed | |
5550 | * otherwise in the case of system crash before the real block | |
5551 | * allocation is done, we will have i_blocks inconsistent with | |
5552 | * on-disk file blocks. | |
5553 | * We always keep i_blocks updated together with real | |
5554 | * allocation. But to not confuse with user, stat | |
5555 | * will return the blocks that include the delayed allocation | |
5556 | * blocks for this file. | |
5557 | */ | |
96607551 | 5558 | delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb), |
9206c561 AD |
5559 | EXT4_I(inode)->i_reserved_data_blocks); |
5560 | stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9); | |
3e3398a0 MC |
5561 | return 0; |
5562 | } | |
ac27a0ec | 5563 | |
fffb2739 JK |
5564 | static int ext4_index_trans_blocks(struct inode *inode, int lblocks, |
5565 | int pextents) | |
a02908f1 | 5566 | { |
12e9b892 | 5567 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) |
fffb2739 JK |
5568 | return ext4_ind_trans_blocks(inode, lblocks); |
5569 | return ext4_ext_index_trans_blocks(inode, pextents); | |
a02908f1 | 5570 | } |
ac51d837 | 5571 | |
ac27a0ec | 5572 | /* |
a02908f1 MC |
5573 | * Account for index blocks, block groups bitmaps and block group |
5574 | * descriptor blocks if modify datablocks and index blocks | |
5575 | * worse case, the indexs blocks spread over different block groups | |
ac27a0ec | 5576 | * |
a02908f1 | 5577 | * If datablocks are discontiguous, they are possible to spread over |
4907cb7b | 5578 | * different block groups too. If they are contiguous, with flexbg, |
a02908f1 | 5579 | * they could still across block group boundary. |
ac27a0ec | 5580 | * |
a02908f1 MC |
5581 | * Also account for superblock, inode, quota and xattr blocks |
5582 | */ | |
dec214d0 | 5583 | static int ext4_meta_trans_blocks(struct inode *inode, int lblocks, |
fffb2739 | 5584 | int pextents) |
a02908f1 | 5585 | { |
8df9675f TT |
5586 | ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb); |
5587 | int gdpblocks; | |
a02908f1 MC |
5588 | int idxblocks; |
5589 | int ret = 0; | |
5590 | ||
5591 | /* | |
fffb2739 JK |
5592 | * How many index blocks need to touch to map @lblocks logical blocks |
5593 | * to @pextents physical extents? | |
a02908f1 | 5594 | */ |
fffb2739 | 5595 | idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents); |
a02908f1 MC |
5596 | |
5597 | ret = idxblocks; | |
5598 | ||
5599 | /* | |
5600 | * Now let's see how many group bitmaps and group descriptors need | |
5601 | * to account | |
5602 | */ | |
fffb2739 | 5603 | groups = idxblocks + pextents; |
a02908f1 | 5604 | gdpblocks = groups; |
8df9675f TT |
5605 | if (groups > ngroups) |
5606 | groups = ngroups; | |
a02908f1 MC |
5607 | if (groups > EXT4_SB(inode->i_sb)->s_gdb_count) |
5608 | gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count; | |
5609 | ||
5610 | /* bitmaps and block group descriptor blocks */ | |
5611 | ret += groups + gdpblocks; | |
5612 | ||
5613 | /* Blocks for super block, inode, quota and xattr blocks */ | |
5614 | ret += EXT4_META_TRANS_BLOCKS(inode->i_sb); | |
5615 | ||
5616 | return ret; | |
5617 | } | |
5618 | ||
5619 | /* | |
25985edc | 5620 | * Calculate the total number of credits to reserve to fit |
f3bd1f3f MC |
5621 | * the modification of a single pages into a single transaction, |
5622 | * which may include multiple chunks of block allocations. | |
ac27a0ec | 5623 | * |
525f4ed8 | 5624 | * This could be called via ext4_write_begin() |
ac27a0ec | 5625 | * |
525f4ed8 | 5626 | * We need to consider the worse case, when |
a02908f1 | 5627 | * one new block per extent. |
ac27a0ec | 5628 | */ |
a86c6181 | 5629 | int ext4_writepage_trans_blocks(struct inode *inode) |
ac27a0ec | 5630 | { |
617ba13b | 5631 | int bpp = ext4_journal_blocks_per_page(inode); |
ac27a0ec DK |
5632 | int ret; |
5633 | ||
fffb2739 | 5634 | ret = ext4_meta_trans_blocks(inode, bpp, bpp); |
a86c6181 | 5635 | |
a02908f1 | 5636 | /* Account for data blocks for journalled mode */ |
617ba13b | 5637 | if (ext4_should_journal_data(inode)) |
a02908f1 | 5638 | ret += bpp; |
ac27a0ec DK |
5639 | return ret; |
5640 | } | |
f3bd1f3f MC |
5641 | |
5642 | /* | |
5643 | * Calculate the journal credits for a chunk of data modification. | |
5644 | * | |
5645 | * This is called from DIO, fallocate or whoever calling | |
79e83036 | 5646 | * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks. |
f3bd1f3f MC |
5647 | * |
5648 | * journal buffers for data blocks are not included here, as DIO | |
5649 | * and fallocate do no need to journal data buffers. | |
5650 | */ | |
5651 | int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks) | |
5652 | { | |
5653 | return ext4_meta_trans_blocks(inode, nrblocks, 1); | |
5654 | } | |
5655 | ||
ac27a0ec | 5656 | /* |
617ba13b | 5657 | * The caller must have previously called ext4_reserve_inode_write(). |
ac27a0ec DK |
5658 | * Give this, we know that the caller already has write access to iloc->bh. |
5659 | */ | |
617ba13b | 5660 | int ext4_mark_iloc_dirty(handle_t *handle, |
de9a55b8 | 5661 | struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec DK |
5662 | { |
5663 | int err = 0; | |
5664 | ||
0db1ff22 TT |
5665 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
5666 | return -EIO; | |
5667 | ||
c64db50e | 5668 | if (IS_I_VERSION(inode)) |
25ec56b5 JNC |
5669 | inode_inc_iversion(inode); |
5670 | ||
ac27a0ec DK |
5671 | /* the do_update_inode consumes one bh->b_count */ |
5672 | get_bh(iloc->bh); | |
5673 | ||
dab291af | 5674 | /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */ |
830156c7 | 5675 | err = ext4_do_update_inode(handle, inode, iloc); |
ac27a0ec DK |
5676 | put_bh(iloc->bh); |
5677 | return err; | |
5678 | } | |
5679 | ||
5680 | /* | |
5681 | * On success, We end up with an outstanding reference count against | |
5682 | * iloc->bh. This _must_ be cleaned up later. | |
5683 | */ | |
5684 | ||
5685 | int | |
617ba13b MC |
5686 | ext4_reserve_inode_write(handle_t *handle, struct inode *inode, |
5687 | struct ext4_iloc *iloc) | |
ac27a0ec | 5688 | { |
0390131b FM |
5689 | int err; |
5690 | ||
0db1ff22 TT |
5691 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
5692 | return -EIO; | |
5693 | ||
0390131b FM |
5694 | err = ext4_get_inode_loc(inode, iloc); |
5695 | if (!err) { | |
5696 | BUFFER_TRACE(iloc->bh, "get_write_access"); | |
5697 | err = ext4_journal_get_write_access(handle, iloc->bh); | |
5698 | if (err) { | |
5699 | brelse(iloc->bh); | |
5700 | iloc->bh = NULL; | |
ac27a0ec DK |
5701 | } |
5702 | } | |
617ba13b | 5703 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
5704 | return err; |
5705 | } | |
5706 | ||
c03b45b8 MX |
5707 | static int __ext4_expand_extra_isize(struct inode *inode, |
5708 | unsigned int new_extra_isize, | |
5709 | struct ext4_iloc *iloc, | |
5710 | handle_t *handle, int *no_expand) | |
5711 | { | |
5712 | struct ext4_inode *raw_inode; | |
5713 | struct ext4_xattr_ibody_header *header; | |
5714 | int error; | |
5715 | ||
5716 | raw_inode = ext4_raw_inode(iloc); | |
5717 | ||
5718 | header = IHDR(inode, raw_inode); | |
5719 | ||
5720 | /* No extended attributes present */ | |
5721 | if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) || | |
5722 | header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) { | |
5723 | memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE + | |
5724 | EXT4_I(inode)->i_extra_isize, 0, | |
5725 | new_extra_isize - EXT4_I(inode)->i_extra_isize); | |
5726 | EXT4_I(inode)->i_extra_isize = new_extra_isize; | |
5727 | return 0; | |
5728 | } | |
5729 | ||
5730 | /* try to expand with EAs present */ | |
5731 | error = ext4_expand_extra_isize_ea(inode, new_extra_isize, | |
5732 | raw_inode, handle); | |
5733 | if (error) { | |
5734 | /* | |
5735 | * Inode size expansion failed; don't try again | |
5736 | */ | |
5737 | *no_expand = 1; | |
5738 | } | |
5739 | ||
5740 | return error; | |
5741 | } | |
5742 | ||
6dd4ee7c KS |
5743 | /* |
5744 | * Expand an inode by new_extra_isize bytes. | |
5745 | * Returns 0 on success or negative error number on failure. | |
5746 | */ | |
cf0a5e81 MX |
5747 | static int ext4_try_to_expand_extra_isize(struct inode *inode, |
5748 | unsigned int new_extra_isize, | |
5749 | struct ext4_iloc iloc, | |
5750 | handle_t *handle) | |
6dd4ee7c | 5751 | { |
3b10fdc6 MX |
5752 | int no_expand; |
5753 | int error; | |
6dd4ee7c | 5754 | |
cf0a5e81 MX |
5755 | if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) |
5756 | return -EOVERFLOW; | |
5757 | ||
5758 | /* | |
5759 | * In nojournal mode, we can immediately attempt to expand | |
5760 | * the inode. When journaled, we first need to obtain extra | |
5761 | * buffer credits since we may write into the EA block | |
5762 | * with this same handle. If journal_extend fails, then it will | |
5763 | * only result in a minor loss of functionality for that inode. | |
5764 | * If this is felt to be critical, then e2fsck should be run to | |
5765 | * force a large enough s_min_extra_isize. | |
5766 | */ | |
5767 | if (ext4_handle_valid(handle) && | |
5768 | jbd2_journal_extend(handle, | |
5769 | EXT4_DATA_TRANS_BLOCKS(inode->i_sb)) != 0) | |
5770 | return -ENOSPC; | |
6dd4ee7c | 5771 | |
3b10fdc6 | 5772 | if (ext4_write_trylock_xattr(inode, &no_expand) == 0) |
cf0a5e81 | 5773 | return -EBUSY; |
3b10fdc6 | 5774 | |
c03b45b8 MX |
5775 | error = __ext4_expand_extra_isize(inode, new_extra_isize, &iloc, |
5776 | handle, &no_expand); | |
5777 | ext4_write_unlock_xattr(inode, &no_expand); | |
6dd4ee7c | 5778 | |
c03b45b8 MX |
5779 | return error; |
5780 | } | |
6dd4ee7c | 5781 | |
c03b45b8 MX |
5782 | int ext4_expand_extra_isize(struct inode *inode, |
5783 | unsigned int new_extra_isize, | |
5784 | struct ext4_iloc *iloc) | |
5785 | { | |
5786 | handle_t *handle; | |
5787 | int no_expand; | |
5788 | int error, rc; | |
5789 | ||
5790 | if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) { | |
5791 | brelse(iloc->bh); | |
5792 | return -EOVERFLOW; | |
6dd4ee7c KS |
5793 | } |
5794 | ||
c03b45b8 MX |
5795 | handle = ext4_journal_start(inode, EXT4_HT_INODE, |
5796 | EXT4_DATA_TRANS_BLOCKS(inode->i_sb)); | |
5797 | if (IS_ERR(handle)) { | |
5798 | error = PTR_ERR(handle); | |
5799 | brelse(iloc->bh); | |
5800 | return error; | |
5801 | } | |
5802 | ||
5803 | ext4_write_lock_xattr(inode, &no_expand); | |
5804 | ||
5805 | BUFFER_TRACE(iloc.bh, "get_write_access"); | |
5806 | error = ext4_journal_get_write_access(handle, iloc->bh); | |
3b10fdc6 | 5807 | if (error) { |
c03b45b8 MX |
5808 | brelse(iloc->bh); |
5809 | goto out_stop; | |
3b10fdc6 | 5810 | } |
cf0a5e81 | 5811 | |
c03b45b8 MX |
5812 | error = __ext4_expand_extra_isize(inode, new_extra_isize, iloc, |
5813 | handle, &no_expand); | |
5814 | ||
5815 | rc = ext4_mark_iloc_dirty(handle, inode, iloc); | |
5816 | if (!error) | |
5817 | error = rc; | |
5818 | ||
5819 | ext4_write_unlock_xattr(inode, &no_expand); | |
5820 | out_stop: | |
5821 | ext4_journal_stop(handle); | |
3b10fdc6 | 5822 | return error; |
6dd4ee7c KS |
5823 | } |
5824 | ||
ac27a0ec DK |
5825 | /* |
5826 | * What we do here is to mark the in-core inode as clean with respect to inode | |
5827 | * dirtiness (it may still be data-dirty). | |
5828 | * This means that the in-core inode may be reaped by prune_icache | |
5829 | * without having to perform any I/O. This is a very good thing, | |
5830 | * because *any* task may call prune_icache - even ones which | |
5831 | * have a transaction open against a different journal. | |
5832 | * | |
5833 | * Is this cheating? Not really. Sure, we haven't written the | |
5834 | * inode out, but prune_icache isn't a user-visible syncing function. | |
5835 | * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync) | |
5836 | * we start and wait on commits. | |
ac27a0ec | 5837 | */ |
617ba13b | 5838 | int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode) |
ac27a0ec | 5839 | { |
617ba13b | 5840 | struct ext4_iloc iloc; |
6dd4ee7c | 5841 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
cf0a5e81 | 5842 | int err; |
ac27a0ec DK |
5843 | |
5844 | might_sleep(); | |
7ff9c073 | 5845 | trace_ext4_mark_inode_dirty(inode, _RET_IP_); |
617ba13b | 5846 | err = ext4_reserve_inode_write(handle, inode, &iloc); |
5e1021f2 EG |
5847 | if (err) |
5848 | return err; | |
cf0a5e81 MX |
5849 | |
5850 | if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize) | |
5851 | ext4_try_to_expand_extra_isize(inode, sbi->s_want_extra_isize, | |
5852 | iloc, handle); | |
5853 | ||
5e1021f2 | 5854 | return ext4_mark_iloc_dirty(handle, inode, &iloc); |
ac27a0ec DK |
5855 | } |
5856 | ||
5857 | /* | |
617ba13b | 5858 | * ext4_dirty_inode() is called from __mark_inode_dirty() |
ac27a0ec DK |
5859 | * |
5860 | * We're really interested in the case where a file is being extended. | |
5861 | * i_size has been changed by generic_commit_write() and we thus need | |
5862 | * to include the updated inode in the current transaction. | |
5863 | * | |
5dd4056d | 5864 | * Also, dquot_alloc_block() will always dirty the inode when blocks |
ac27a0ec DK |
5865 | * are allocated to the file. |
5866 | * | |
5867 | * If the inode is marked synchronous, we don't honour that here - doing | |
5868 | * so would cause a commit on atime updates, which we don't bother doing. | |
5869 | * We handle synchronous inodes at the highest possible level. | |
0ae45f63 TT |
5870 | * |
5871 | * If only the I_DIRTY_TIME flag is set, we can skip everything. If | |
5872 | * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need | |
5873 | * to copy into the on-disk inode structure are the timestamp files. | |
ac27a0ec | 5874 | */ |
aa385729 | 5875 | void ext4_dirty_inode(struct inode *inode, int flags) |
ac27a0ec | 5876 | { |
ac27a0ec DK |
5877 | handle_t *handle; |
5878 | ||
0ae45f63 TT |
5879 | if (flags == I_DIRTY_TIME) |
5880 | return; | |
9924a92a | 5881 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 2); |
ac27a0ec DK |
5882 | if (IS_ERR(handle)) |
5883 | goto out; | |
f3dc272f | 5884 | |
f3dc272f CW |
5885 | ext4_mark_inode_dirty(handle, inode); |
5886 | ||
617ba13b | 5887 | ext4_journal_stop(handle); |
ac27a0ec DK |
5888 | out: |
5889 | return; | |
5890 | } | |
5891 | ||
5892 | #if 0 | |
5893 | /* | |
5894 | * Bind an inode's backing buffer_head into this transaction, to prevent | |
5895 | * it from being flushed to disk early. Unlike | |
617ba13b | 5896 | * ext4_reserve_inode_write, this leaves behind no bh reference and |
ac27a0ec DK |
5897 | * returns no iloc structure, so the caller needs to repeat the iloc |
5898 | * lookup to mark the inode dirty later. | |
5899 | */ | |
617ba13b | 5900 | static int ext4_pin_inode(handle_t *handle, struct inode *inode) |
ac27a0ec | 5901 | { |
617ba13b | 5902 | struct ext4_iloc iloc; |
ac27a0ec DK |
5903 | |
5904 | int err = 0; | |
5905 | if (handle) { | |
617ba13b | 5906 | err = ext4_get_inode_loc(inode, &iloc); |
ac27a0ec DK |
5907 | if (!err) { |
5908 | BUFFER_TRACE(iloc.bh, "get_write_access"); | |
dab291af | 5909 | err = jbd2_journal_get_write_access(handle, iloc.bh); |
ac27a0ec | 5910 | if (!err) |
0390131b | 5911 | err = ext4_handle_dirty_metadata(handle, |
73b50c1c | 5912 | NULL, |
0390131b | 5913 | iloc.bh); |
ac27a0ec DK |
5914 | brelse(iloc.bh); |
5915 | } | |
5916 | } | |
617ba13b | 5917 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
5918 | return err; |
5919 | } | |
5920 | #endif | |
5921 | ||
617ba13b | 5922 | int ext4_change_inode_journal_flag(struct inode *inode, int val) |
ac27a0ec DK |
5923 | { |
5924 | journal_t *journal; | |
5925 | handle_t *handle; | |
5926 | int err; | |
c8585c6f | 5927 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
ac27a0ec DK |
5928 | |
5929 | /* | |
5930 | * We have to be very careful here: changing a data block's | |
5931 | * journaling status dynamically is dangerous. If we write a | |
5932 | * data block to the journal, change the status and then delete | |
5933 | * that block, we risk forgetting to revoke the old log record | |
5934 | * from the journal and so a subsequent replay can corrupt data. | |
5935 | * So, first we make sure that the journal is empty and that | |
5936 | * nobody is changing anything. | |
5937 | */ | |
5938 | ||
617ba13b | 5939 | journal = EXT4_JOURNAL(inode); |
0390131b FM |
5940 | if (!journal) |
5941 | return 0; | |
d699594d | 5942 | if (is_journal_aborted(journal)) |
ac27a0ec DK |
5943 | return -EROFS; |
5944 | ||
17335dcc DM |
5945 | /* Wait for all existing dio workers */ |
5946 | ext4_inode_block_unlocked_dio(inode); | |
5947 | inode_dio_wait(inode); | |
5948 | ||
4c546592 DJ |
5949 | /* |
5950 | * Before flushing the journal and switching inode's aops, we have | |
5951 | * to flush all dirty data the inode has. There can be outstanding | |
5952 | * delayed allocations, there can be unwritten extents created by | |
5953 | * fallocate or buffered writes in dioread_nolock mode covered by | |
5954 | * dirty data which can be converted only after flushing the dirty | |
5955 | * data (and journalled aops don't know how to handle these cases). | |
5956 | */ | |
5957 | if (val) { | |
5958 | down_write(&EXT4_I(inode)->i_mmap_sem); | |
5959 | err = filemap_write_and_wait(inode->i_mapping); | |
5960 | if (err < 0) { | |
5961 | up_write(&EXT4_I(inode)->i_mmap_sem); | |
5962 | ext4_inode_resume_unlocked_dio(inode); | |
5963 | return err; | |
5964 | } | |
5965 | } | |
5966 | ||
c8585c6f | 5967 | percpu_down_write(&sbi->s_journal_flag_rwsem); |
dab291af | 5968 | jbd2_journal_lock_updates(journal); |
ac27a0ec DK |
5969 | |
5970 | /* | |
5971 | * OK, there are no updates running now, and all cached data is | |
5972 | * synced to disk. We are now in a completely consistent state | |
5973 | * which doesn't have anything in the journal, and we know that | |
5974 | * no filesystem updates are running, so it is safe to modify | |
5975 | * the inode's in-core data-journaling state flag now. | |
5976 | */ | |
5977 | ||
5978 | if (val) | |
12e9b892 | 5979 | ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
5872ddaa | 5980 | else { |
4f879ca6 JK |
5981 | err = jbd2_journal_flush(journal); |
5982 | if (err < 0) { | |
5983 | jbd2_journal_unlock_updates(journal); | |
c8585c6f | 5984 | percpu_up_write(&sbi->s_journal_flag_rwsem); |
4f879ca6 JK |
5985 | ext4_inode_resume_unlocked_dio(inode); |
5986 | return err; | |
5987 | } | |
12e9b892 | 5988 | ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
5872ddaa | 5989 | } |
617ba13b | 5990 | ext4_set_aops(inode); |
a3caa24b JK |
5991 | /* |
5992 | * Update inode->i_flags after EXT4_INODE_JOURNAL_DATA was updated. | |
5993 | * E.g. S_DAX may get cleared / set. | |
5994 | */ | |
5995 | ext4_set_inode_flags(inode); | |
ac27a0ec | 5996 | |
dab291af | 5997 | jbd2_journal_unlock_updates(journal); |
c8585c6f DJ |
5998 | percpu_up_write(&sbi->s_journal_flag_rwsem); |
5999 | ||
4c546592 DJ |
6000 | if (val) |
6001 | up_write(&EXT4_I(inode)->i_mmap_sem); | |
17335dcc | 6002 | ext4_inode_resume_unlocked_dio(inode); |
ac27a0ec DK |
6003 | |
6004 | /* Finally we can mark the inode as dirty. */ | |
6005 | ||
9924a92a | 6006 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 1); |
ac27a0ec DK |
6007 | if (IS_ERR(handle)) |
6008 | return PTR_ERR(handle); | |
6009 | ||
617ba13b | 6010 | err = ext4_mark_inode_dirty(handle, inode); |
0390131b | 6011 | ext4_handle_sync(handle); |
617ba13b MC |
6012 | ext4_journal_stop(handle); |
6013 | ext4_std_error(inode->i_sb, err); | |
ac27a0ec DK |
6014 | |
6015 | return err; | |
6016 | } | |
2e9ee850 AK |
6017 | |
6018 | static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh) | |
6019 | { | |
6020 | return !buffer_mapped(bh); | |
6021 | } | |
6022 | ||
11bac800 | 6023 | int ext4_page_mkwrite(struct vm_fault *vmf) |
2e9ee850 | 6024 | { |
11bac800 | 6025 | struct vm_area_struct *vma = vmf->vma; |
c2ec175c | 6026 | struct page *page = vmf->page; |
2e9ee850 AK |
6027 | loff_t size; |
6028 | unsigned long len; | |
9ea7df53 | 6029 | int ret; |
2e9ee850 | 6030 | struct file *file = vma->vm_file; |
496ad9aa | 6031 | struct inode *inode = file_inode(file); |
2e9ee850 | 6032 | struct address_space *mapping = inode->i_mapping; |
9ea7df53 JK |
6033 | handle_t *handle; |
6034 | get_block_t *get_block; | |
6035 | int retries = 0; | |
2e9ee850 | 6036 | |
8e8ad8a5 | 6037 | sb_start_pagefault(inode->i_sb); |
041bbb6d | 6038 | file_update_time(vma->vm_file); |
ea3d7209 JK |
6039 | |
6040 | down_read(&EXT4_I(inode)->i_mmap_sem); | |
7b4cc978 EB |
6041 | |
6042 | ret = ext4_convert_inline_data(inode); | |
6043 | if (ret) | |
6044 | goto out_ret; | |
6045 | ||
9ea7df53 JK |
6046 | /* Delalloc case is easy... */ |
6047 | if (test_opt(inode->i_sb, DELALLOC) && | |
6048 | !ext4_should_journal_data(inode) && | |
6049 | !ext4_nonda_switch(inode->i_sb)) { | |
6050 | do { | |
5c500029 | 6051 | ret = block_page_mkwrite(vma, vmf, |
9ea7df53 JK |
6052 | ext4_da_get_block_prep); |
6053 | } while (ret == -ENOSPC && | |
6054 | ext4_should_retry_alloc(inode->i_sb, &retries)); | |
6055 | goto out_ret; | |
2e9ee850 | 6056 | } |
0e499890 DW |
6057 | |
6058 | lock_page(page); | |
9ea7df53 JK |
6059 | size = i_size_read(inode); |
6060 | /* Page got truncated from under us? */ | |
6061 | if (page->mapping != mapping || page_offset(page) > size) { | |
6062 | unlock_page(page); | |
6063 | ret = VM_FAULT_NOPAGE; | |
6064 | goto out; | |
0e499890 | 6065 | } |
2e9ee850 | 6066 | |
09cbfeaf KS |
6067 | if (page->index == size >> PAGE_SHIFT) |
6068 | len = size & ~PAGE_MASK; | |
2e9ee850 | 6069 | else |
09cbfeaf | 6070 | len = PAGE_SIZE; |
a827eaff | 6071 | /* |
9ea7df53 JK |
6072 | * Return if we have all the buffers mapped. This avoids the need to do |
6073 | * journal_start/journal_stop which can block and take a long time | |
a827eaff | 6074 | */ |
2e9ee850 | 6075 | if (page_has_buffers(page)) { |
f19d5870 TM |
6076 | if (!ext4_walk_page_buffers(NULL, page_buffers(page), |
6077 | 0, len, NULL, | |
6078 | ext4_bh_unmapped)) { | |
9ea7df53 | 6079 | /* Wait so that we don't change page under IO */ |
1d1d1a76 | 6080 | wait_for_stable_page(page); |
9ea7df53 JK |
6081 | ret = VM_FAULT_LOCKED; |
6082 | goto out; | |
a827eaff | 6083 | } |
2e9ee850 | 6084 | } |
a827eaff | 6085 | unlock_page(page); |
9ea7df53 JK |
6086 | /* OK, we need to fill the hole... */ |
6087 | if (ext4_should_dioread_nolock(inode)) | |
705965bd | 6088 | get_block = ext4_get_block_unwritten; |
9ea7df53 JK |
6089 | else |
6090 | get_block = ext4_get_block; | |
6091 | retry_alloc: | |
9924a92a TT |
6092 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
6093 | ext4_writepage_trans_blocks(inode)); | |
9ea7df53 | 6094 | if (IS_ERR(handle)) { |
c2ec175c | 6095 | ret = VM_FAULT_SIGBUS; |
9ea7df53 JK |
6096 | goto out; |
6097 | } | |
5c500029 | 6098 | ret = block_page_mkwrite(vma, vmf, get_block); |
9ea7df53 | 6099 | if (!ret && ext4_should_journal_data(inode)) { |
f19d5870 | 6100 | if (ext4_walk_page_buffers(handle, page_buffers(page), 0, |
09cbfeaf | 6101 | PAGE_SIZE, NULL, do_journal_get_write_access)) { |
9ea7df53 JK |
6102 | unlock_page(page); |
6103 | ret = VM_FAULT_SIGBUS; | |
fcbb5515 | 6104 | ext4_journal_stop(handle); |
9ea7df53 JK |
6105 | goto out; |
6106 | } | |
6107 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); | |
6108 | } | |
6109 | ext4_journal_stop(handle); | |
6110 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) | |
6111 | goto retry_alloc; | |
6112 | out_ret: | |
6113 | ret = block_page_mkwrite_return(ret); | |
6114 | out: | |
ea3d7209 | 6115 | up_read(&EXT4_I(inode)->i_mmap_sem); |
8e8ad8a5 | 6116 | sb_end_pagefault(inode->i_sb); |
2e9ee850 AK |
6117 | return ret; |
6118 | } | |
ea3d7209 | 6119 | |
11bac800 | 6120 | int ext4_filemap_fault(struct vm_fault *vmf) |
ea3d7209 | 6121 | { |
11bac800 | 6122 | struct inode *inode = file_inode(vmf->vma->vm_file); |
ea3d7209 JK |
6123 | int err; |
6124 | ||
6125 | down_read(&EXT4_I(inode)->i_mmap_sem); | |
11bac800 | 6126 | err = filemap_fault(vmf); |
ea3d7209 JK |
6127 | up_read(&EXT4_I(inode)->i_mmap_sem); |
6128 | ||
6129 | return err; | |
6130 | } | |
2d90c160 JK |
6131 | |
6132 | /* | |
6133 | * Find the first extent at or after @lblk in an inode that is not a hole. | |
6134 | * Search for @map_len blocks at most. The extent is returned in @result. | |
6135 | * | |
6136 | * The function returns 1 if we found an extent. The function returns 0 in | |
6137 | * case there is no extent at or after @lblk and in that case also sets | |
6138 | * @result->es_len to 0. In case of error, the error code is returned. | |
6139 | */ | |
6140 | int ext4_get_next_extent(struct inode *inode, ext4_lblk_t lblk, | |
6141 | unsigned int map_len, struct extent_status *result) | |
6142 | { | |
6143 | struct ext4_map_blocks map; | |
6144 | struct extent_status es = {}; | |
6145 | int ret; | |
6146 | ||
6147 | map.m_lblk = lblk; | |
6148 | map.m_len = map_len; | |
6149 | ||
6150 | /* | |
6151 | * For non-extent based files this loop may iterate several times since | |
6152 | * we do not determine full hole size. | |
6153 | */ | |
6154 | while (map.m_len > 0) { | |
6155 | ret = ext4_map_blocks(NULL, inode, &map, 0); | |
6156 | if (ret < 0) | |
6157 | return ret; | |
6158 | /* There's extent covering m_lblk? Just return it. */ | |
6159 | if (ret > 0) { | |
6160 | int status; | |
6161 | ||
6162 | ext4_es_store_pblock(result, map.m_pblk); | |
6163 | result->es_lblk = map.m_lblk; | |
6164 | result->es_len = map.m_len; | |
6165 | if (map.m_flags & EXT4_MAP_UNWRITTEN) | |
6166 | status = EXTENT_STATUS_UNWRITTEN; | |
6167 | else | |
6168 | status = EXTENT_STATUS_WRITTEN; | |
6169 | ext4_es_store_status(result, status); | |
6170 | return 1; | |
6171 | } | |
6172 | ext4_es_find_delayed_extent_range(inode, map.m_lblk, | |
6173 | map.m_lblk + map.m_len - 1, | |
6174 | &es); | |
6175 | /* Is delalloc data before next block in extent tree? */ | |
6176 | if (es.es_len && es.es_lblk < map.m_lblk + map.m_len) { | |
6177 | ext4_lblk_t offset = 0; | |
6178 | ||
6179 | if (es.es_lblk < lblk) | |
6180 | offset = lblk - es.es_lblk; | |
6181 | result->es_lblk = es.es_lblk + offset; | |
6182 | ext4_es_store_pblock(result, | |
6183 | ext4_es_pblock(&es) + offset); | |
6184 | result->es_len = es.es_len - offset; | |
6185 | ext4_es_store_status(result, ext4_es_status(&es)); | |
6186 | ||
6187 | return 1; | |
6188 | } | |
6189 | /* There's a hole at m_lblk, advance us after it */ | |
6190 | map.m_lblk += map.m_len; | |
6191 | map_len -= map.m_len; | |
6192 | map.m_len = map_len; | |
6193 | cond_resched(); | |
6194 | } | |
6195 | result->es_len = 0; | |
6196 | return 0; | |
6197 | } |