]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
ac27a0ec | 2 | /* |
617ba13b | 3 | * linux/fs/ext4/inode.c |
ac27a0ec DK |
4 | * |
5 | * Copyright (C) 1992, 1993, 1994, 1995 | |
6 | * Remy Card (card@masi.ibp.fr) | |
7 | * Laboratoire MASI - Institut Blaise Pascal | |
8 | * Universite Pierre et Marie Curie (Paris VI) | |
9 | * | |
10 | * from | |
11 | * | |
12 | * linux/fs/minix/inode.c | |
13 | * | |
14 | * Copyright (C) 1991, 1992 Linus Torvalds | |
15 | * | |
ac27a0ec DK |
16 | * 64-bit file support on 64-bit platforms by Jakub Jelinek |
17 | * (jj@sunsite.ms.mff.cuni.cz) | |
18 | * | |
617ba13b | 19 | * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000 |
ac27a0ec DK |
20 | */ |
21 | ||
ac27a0ec DK |
22 | #include <linux/fs.h> |
23 | #include <linux/time.h> | |
ac27a0ec DK |
24 | #include <linux/highuid.h> |
25 | #include <linux/pagemap.h> | |
c94c2acf | 26 | #include <linux/dax.h> |
ac27a0ec DK |
27 | #include <linux/quotaops.h> |
28 | #include <linux/string.h> | |
29 | #include <linux/buffer_head.h> | |
30 | #include <linux/writeback.h> | |
64769240 | 31 | #include <linux/pagevec.h> |
ac27a0ec | 32 | #include <linux/mpage.h> |
e83c1397 | 33 | #include <linux/namei.h> |
ac27a0ec DK |
34 | #include <linux/uio.h> |
35 | #include <linux/bio.h> | |
4c0425ff | 36 | #include <linux/workqueue.h> |
744692dc | 37 | #include <linux/kernel.h> |
6db26ffc | 38 | #include <linux/printk.h> |
5a0e3ad6 | 39 | #include <linux/slab.h> |
00a1a053 | 40 | #include <linux/bitops.h> |
364443cb | 41 | #include <linux/iomap.h> |
ae5e165d | 42 | #include <linux/iversion.h> |
9bffad1e | 43 | |
3dcf5451 | 44 | #include "ext4_jbd2.h" |
ac27a0ec DK |
45 | #include "xattr.h" |
46 | #include "acl.h" | |
9f125d64 | 47 | #include "truncate.h" |
ac27a0ec | 48 | |
9bffad1e TT |
49 | #include <trace/events/ext4.h> |
50 | ||
a1d6cc56 AK |
51 | #define MPAGE_DA_EXTENT_TAIL 0x01 |
52 | ||
814525f4 DW |
53 | static __u32 ext4_inode_csum(struct inode *inode, struct ext4_inode *raw, |
54 | struct ext4_inode_info *ei) | |
55 | { | |
56 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
814525f4 | 57 | __u32 csum; |
b47820ed DJ |
58 | __u16 dummy_csum = 0; |
59 | int offset = offsetof(struct ext4_inode, i_checksum_lo); | |
60 | unsigned int csum_size = sizeof(dummy_csum); | |
814525f4 | 61 | |
b47820ed DJ |
62 | csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)raw, offset); |
63 | csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, csum_size); | |
64 | offset += csum_size; | |
65 | csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset, | |
66 | EXT4_GOOD_OLD_INODE_SIZE - offset); | |
814525f4 | 67 | |
b47820ed DJ |
68 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
69 | offset = offsetof(struct ext4_inode, i_checksum_hi); | |
70 | csum = ext4_chksum(sbi, csum, (__u8 *)raw + | |
71 | EXT4_GOOD_OLD_INODE_SIZE, | |
72 | offset - EXT4_GOOD_OLD_INODE_SIZE); | |
73 | if (EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) { | |
74 | csum = ext4_chksum(sbi, csum, (__u8 *)&dummy_csum, | |
75 | csum_size); | |
76 | offset += csum_size; | |
b47820ed | 77 | } |
05ac5aa1 DJ |
78 | csum = ext4_chksum(sbi, csum, (__u8 *)raw + offset, |
79 | EXT4_INODE_SIZE(inode->i_sb) - offset); | |
814525f4 DW |
80 | } |
81 | ||
814525f4 DW |
82 | return csum; |
83 | } | |
84 | ||
85 | static int ext4_inode_csum_verify(struct inode *inode, struct ext4_inode *raw, | |
86 | struct ext4_inode_info *ei) | |
87 | { | |
88 | __u32 provided, calculated; | |
89 | ||
90 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != | |
91 | cpu_to_le32(EXT4_OS_LINUX) || | |
9aa5d32b | 92 | !ext4_has_metadata_csum(inode->i_sb)) |
814525f4 DW |
93 | return 1; |
94 | ||
95 | provided = le16_to_cpu(raw->i_checksum_lo); | |
96 | calculated = ext4_inode_csum(inode, raw, ei); | |
97 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && | |
98 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) | |
99 | provided |= ((__u32)le16_to_cpu(raw->i_checksum_hi)) << 16; | |
100 | else | |
101 | calculated &= 0xFFFF; | |
102 | ||
103 | return provided == calculated; | |
104 | } | |
105 | ||
106 | static void ext4_inode_csum_set(struct inode *inode, struct ext4_inode *raw, | |
107 | struct ext4_inode_info *ei) | |
108 | { | |
109 | __u32 csum; | |
110 | ||
111 | if (EXT4_SB(inode->i_sb)->s_es->s_creator_os != | |
112 | cpu_to_le32(EXT4_OS_LINUX) || | |
9aa5d32b | 113 | !ext4_has_metadata_csum(inode->i_sb)) |
814525f4 DW |
114 | return; |
115 | ||
116 | csum = ext4_inode_csum(inode, raw, ei); | |
117 | raw->i_checksum_lo = cpu_to_le16(csum & 0xFFFF); | |
118 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && | |
119 | EXT4_FITS_IN_INODE(raw, ei, i_checksum_hi)) | |
120 | raw->i_checksum_hi = cpu_to_le16(csum >> 16); | |
121 | } | |
122 | ||
678aaf48 JK |
123 | static inline int ext4_begin_ordered_truncate(struct inode *inode, |
124 | loff_t new_size) | |
125 | { | |
7ff9c073 | 126 | trace_ext4_begin_ordered_truncate(inode, new_size); |
8aefcd55 TT |
127 | /* |
128 | * If jinode is zero, then we never opened the file for | |
129 | * writing, so there's no need to call | |
130 | * jbd2_journal_begin_ordered_truncate() since there's no | |
131 | * outstanding writes we need to flush. | |
132 | */ | |
133 | if (!EXT4_I(inode)->jinode) | |
134 | return 0; | |
135 | return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode), | |
136 | EXT4_I(inode)->jinode, | |
137 | new_size); | |
678aaf48 JK |
138 | } |
139 | ||
d47992f8 LC |
140 | static void ext4_invalidatepage(struct page *page, unsigned int offset, |
141 | unsigned int length); | |
cb20d518 TT |
142 | static int __ext4_journalled_writepage(struct page *page, unsigned int len); |
143 | static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh); | |
dec214d0 TE |
144 | static int ext4_meta_trans_blocks(struct inode *inode, int lblocks, |
145 | int pextents); | |
64769240 | 146 | |
ac27a0ec DK |
147 | /* |
148 | * Test whether an inode is a fast symlink. | |
407cd7fb | 149 | * A fast symlink has its symlink data stored in ext4_inode_info->i_data. |
ac27a0ec | 150 | */ |
f348c252 | 151 | int ext4_inode_is_fast_symlink(struct inode *inode) |
ac27a0ec | 152 | { |
fc82228a AK |
153 | if (!(EXT4_I(inode)->i_flags & EXT4_EA_INODE_FL)) { |
154 | int ea_blocks = EXT4_I(inode)->i_file_acl ? | |
155 | EXT4_CLUSTER_SIZE(inode->i_sb) >> 9 : 0; | |
156 | ||
157 | if (ext4_has_inline_data(inode)) | |
158 | return 0; | |
159 | ||
160 | return (S_ISLNK(inode->i_mode) && inode->i_blocks - ea_blocks == 0); | |
161 | } | |
407cd7fb TE |
162 | return S_ISLNK(inode->i_mode) && inode->i_size && |
163 | (inode->i_size < EXT4_N_BLOCKS * 4); | |
ac27a0ec DK |
164 | } |
165 | ||
ac27a0ec DK |
166 | /* |
167 | * Restart the transaction associated with *handle. This does a commit, | |
168 | * so before we call here everything must be consistently dirtied against | |
169 | * this transaction. | |
170 | */ | |
fa5d1113 | 171 | int ext4_truncate_restart_trans(handle_t *handle, struct inode *inode, |
487caeef | 172 | int nblocks) |
ac27a0ec | 173 | { |
487caeef JK |
174 | int ret; |
175 | ||
176 | /* | |
e35fd660 | 177 | * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this |
487caeef JK |
178 | * moment, get_block can be called only for blocks inside i_size since |
179 | * page cache has been already dropped and writes are blocked by | |
180 | * i_mutex. So we can safely drop the i_data_sem here. | |
181 | */ | |
0390131b | 182 | BUG_ON(EXT4_JOURNAL(inode) == NULL); |
ac27a0ec | 183 | jbd_debug(2, "restarting handle %p\n", handle); |
487caeef | 184 | up_write(&EXT4_I(inode)->i_data_sem); |
8e8eaabe | 185 | ret = ext4_journal_restart(handle, nblocks); |
487caeef | 186 | down_write(&EXT4_I(inode)->i_data_sem); |
fa5d1113 | 187 | ext4_discard_preallocations(inode); |
487caeef JK |
188 | |
189 | return ret; | |
ac27a0ec DK |
190 | } |
191 | ||
192 | /* | |
193 | * Called at the last iput() if i_nlink is zero. | |
194 | */ | |
0930fcc1 | 195 | void ext4_evict_inode(struct inode *inode) |
ac27a0ec DK |
196 | { |
197 | handle_t *handle; | |
bc965ab3 | 198 | int err; |
e50e5129 | 199 | int extra_credits = 3; |
0421a189 | 200 | struct ext4_xattr_inode_array *ea_inode_array = NULL; |
ac27a0ec | 201 | |
7ff9c073 | 202 | trace_ext4_evict_inode(inode); |
2581fdc8 | 203 | |
0930fcc1 | 204 | if (inode->i_nlink) { |
2d859db3 JK |
205 | /* |
206 | * When journalling data dirty buffers are tracked only in the | |
207 | * journal. So although mm thinks everything is clean and | |
208 | * ready for reaping the inode might still have some pages to | |
209 | * write in the running transaction or waiting to be | |
210 | * checkpointed. Thus calling jbd2_journal_invalidatepage() | |
211 | * (via truncate_inode_pages()) to discard these buffers can | |
212 | * cause data loss. Also even if we did not discard these | |
213 | * buffers, we would have no way to find them after the inode | |
214 | * is reaped and thus user could see stale data if he tries to | |
215 | * read them before the transaction is checkpointed. So be | |
216 | * careful and force everything to disk here... We use | |
217 | * ei->i_datasync_tid to store the newest transaction | |
218 | * containing inode's data. | |
219 | * | |
220 | * Note that directories do not have this problem because they | |
221 | * don't use page cache. | |
222 | */ | |
6a7fd522 VN |
223 | if (inode->i_ino != EXT4_JOURNAL_INO && |
224 | ext4_should_journal_data(inode) && | |
3abb1a0f JK |
225 | (S_ISLNK(inode->i_mode) || S_ISREG(inode->i_mode)) && |
226 | inode->i_data.nrpages) { | |
2d859db3 JK |
227 | journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; |
228 | tid_t commit_tid = EXT4_I(inode)->i_datasync_tid; | |
229 | ||
d76a3a77 | 230 | jbd2_complete_transaction(journal, commit_tid); |
2d859db3 JK |
231 | filemap_write_and_wait(&inode->i_data); |
232 | } | |
91b0abe3 | 233 | truncate_inode_pages_final(&inode->i_data); |
5dc23bdd | 234 | |
0930fcc1 AV |
235 | goto no_delete; |
236 | } | |
237 | ||
e2bfb088 TT |
238 | if (is_bad_inode(inode)) |
239 | goto no_delete; | |
240 | dquot_initialize(inode); | |
907f4554 | 241 | |
678aaf48 JK |
242 | if (ext4_should_order_data(inode)) |
243 | ext4_begin_ordered_truncate(inode, 0); | |
91b0abe3 | 244 | truncate_inode_pages_final(&inode->i_data); |
ac27a0ec | 245 | |
8e8ad8a5 JK |
246 | /* |
247 | * Protect us against freezing - iput() caller didn't have to have any | |
248 | * protection against it | |
249 | */ | |
250 | sb_start_intwrite(inode->i_sb); | |
e50e5129 | 251 | |
30a7eb97 TE |
252 | if (!IS_NOQUOTA(inode)) |
253 | extra_credits += EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb); | |
254 | ||
255 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, | |
256 | ext4_blocks_for_truncate(inode)+extra_credits); | |
ac27a0ec | 257 | if (IS_ERR(handle)) { |
bc965ab3 | 258 | ext4_std_error(inode->i_sb, PTR_ERR(handle)); |
ac27a0ec DK |
259 | /* |
260 | * If we're going to skip the normal cleanup, we still need to | |
261 | * make sure that the in-core orphan linked list is properly | |
262 | * cleaned up. | |
263 | */ | |
617ba13b | 264 | ext4_orphan_del(NULL, inode); |
8e8ad8a5 | 265 | sb_end_intwrite(inode->i_sb); |
ac27a0ec DK |
266 | goto no_delete; |
267 | } | |
30a7eb97 | 268 | |
ac27a0ec | 269 | if (IS_SYNC(inode)) |
0390131b | 270 | ext4_handle_sync(handle); |
407cd7fb TE |
271 | |
272 | /* | |
273 | * Set inode->i_size to 0 before calling ext4_truncate(). We need | |
274 | * special handling of symlinks here because i_size is used to | |
275 | * determine whether ext4_inode_info->i_data contains symlink data or | |
276 | * block mappings. Setting i_size to 0 will remove its fast symlink | |
277 | * status. Erase i_data so that it becomes a valid empty block map. | |
278 | */ | |
279 | if (ext4_inode_is_fast_symlink(inode)) | |
280 | memset(EXT4_I(inode)->i_data, 0, sizeof(EXT4_I(inode)->i_data)); | |
ac27a0ec | 281 | inode->i_size = 0; |
bc965ab3 TT |
282 | err = ext4_mark_inode_dirty(handle, inode); |
283 | if (err) { | |
12062ddd | 284 | ext4_warning(inode->i_sb, |
bc965ab3 TT |
285 | "couldn't mark inode dirty (err %d)", err); |
286 | goto stop_handle; | |
287 | } | |
2c98eb5e TT |
288 | if (inode->i_blocks) { |
289 | err = ext4_truncate(inode); | |
290 | if (err) { | |
291 | ext4_error(inode->i_sb, | |
292 | "couldn't truncate inode %lu (err %d)", | |
293 | inode->i_ino, err); | |
294 | goto stop_handle; | |
295 | } | |
296 | } | |
bc965ab3 | 297 | |
30a7eb97 TE |
298 | /* Remove xattr references. */ |
299 | err = ext4_xattr_delete_inode(handle, inode, &ea_inode_array, | |
300 | extra_credits); | |
301 | if (err) { | |
302 | ext4_warning(inode->i_sb, "xattr delete (err %d)", err); | |
303 | stop_handle: | |
304 | ext4_journal_stop(handle); | |
305 | ext4_orphan_del(NULL, inode); | |
306 | sb_end_intwrite(inode->i_sb); | |
307 | ext4_xattr_inode_array_free(ea_inode_array); | |
308 | goto no_delete; | |
bc965ab3 TT |
309 | } |
310 | ||
ac27a0ec | 311 | /* |
617ba13b | 312 | * Kill off the orphan record which ext4_truncate created. |
ac27a0ec | 313 | * AKPM: I think this can be inside the above `if'. |
617ba13b | 314 | * Note that ext4_orphan_del() has to be able to cope with the |
ac27a0ec | 315 | * deletion of a non-existent orphan - this is because we don't |
617ba13b | 316 | * know if ext4_truncate() actually created an orphan record. |
ac27a0ec DK |
317 | * (Well, we could do this if we need to, but heck - it works) |
318 | */ | |
617ba13b MC |
319 | ext4_orphan_del(handle, inode); |
320 | EXT4_I(inode)->i_dtime = get_seconds(); | |
ac27a0ec DK |
321 | |
322 | /* | |
323 | * One subtle ordering requirement: if anything has gone wrong | |
324 | * (transaction abort, IO errors, whatever), then we can still | |
325 | * do these next steps (the fs will already have been marked as | |
326 | * having errors), but we can't free the inode if the mark_dirty | |
327 | * fails. | |
328 | */ | |
617ba13b | 329 | if (ext4_mark_inode_dirty(handle, inode)) |
ac27a0ec | 330 | /* If that failed, just do the required in-core inode clear. */ |
0930fcc1 | 331 | ext4_clear_inode(inode); |
ac27a0ec | 332 | else |
617ba13b MC |
333 | ext4_free_inode(handle, inode); |
334 | ext4_journal_stop(handle); | |
8e8ad8a5 | 335 | sb_end_intwrite(inode->i_sb); |
0421a189 | 336 | ext4_xattr_inode_array_free(ea_inode_array); |
ac27a0ec DK |
337 | return; |
338 | no_delete: | |
0930fcc1 | 339 | ext4_clear_inode(inode); /* We must guarantee clearing of inode... */ |
ac27a0ec DK |
340 | } |
341 | ||
a9e7f447 DM |
342 | #ifdef CONFIG_QUOTA |
343 | qsize_t *ext4_get_reserved_space(struct inode *inode) | |
60e58e0f | 344 | { |
a9e7f447 | 345 | return &EXT4_I(inode)->i_reserved_quota; |
60e58e0f | 346 | } |
a9e7f447 | 347 | #endif |
9d0be502 | 348 | |
0637c6f4 TT |
349 | /* |
350 | * Called with i_data_sem down, which is important since we can call | |
351 | * ext4_discard_preallocations() from here. | |
352 | */ | |
5f634d06 AK |
353 | void ext4_da_update_reserve_space(struct inode *inode, |
354 | int used, int quota_claim) | |
12219aea AK |
355 | { |
356 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0637c6f4 | 357 | struct ext4_inode_info *ei = EXT4_I(inode); |
0637c6f4 TT |
358 | |
359 | spin_lock(&ei->i_block_reservation_lock); | |
d8990240 | 360 | trace_ext4_da_update_reserve_space(inode, used, quota_claim); |
0637c6f4 | 361 | if (unlikely(used > ei->i_reserved_data_blocks)) { |
8de5c325 | 362 | ext4_warning(inode->i_sb, "%s: ino %lu, used %d " |
1084f252 | 363 | "with only %d reserved data blocks", |
0637c6f4 TT |
364 | __func__, inode->i_ino, used, |
365 | ei->i_reserved_data_blocks); | |
366 | WARN_ON(1); | |
367 | used = ei->i_reserved_data_blocks; | |
368 | } | |
12219aea | 369 | |
0637c6f4 TT |
370 | /* Update per-inode reservations */ |
371 | ei->i_reserved_data_blocks -= used; | |
71d4f7d0 | 372 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, used); |
6bc6e63f | 373 | |
12219aea | 374 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
60e58e0f | 375 | |
72b8ab9d ES |
376 | /* Update quota subsystem for data blocks */ |
377 | if (quota_claim) | |
7b415bf6 | 378 | dquot_claim_block(inode, EXT4_C2B(sbi, used)); |
72b8ab9d | 379 | else { |
5f634d06 AK |
380 | /* |
381 | * We did fallocate with an offset that is already delayed | |
382 | * allocated. So on delayed allocated writeback we should | |
72b8ab9d | 383 | * not re-claim the quota for fallocated blocks. |
5f634d06 | 384 | */ |
7b415bf6 | 385 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, used)); |
5f634d06 | 386 | } |
d6014301 AK |
387 | |
388 | /* | |
389 | * If we have done all the pending block allocations and if | |
390 | * there aren't any writers on the inode, we can discard the | |
391 | * inode's preallocations. | |
392 | */ | |
0637c6f4 TT |
393 | if ((ei->i_reserved_data_blocks == 0) && |
394 | (atomic_read(&inode->i_writecount) == 0)) | |
d6014301 | 395 | ext4_discard_preallocations(inode); |
12219aea AK |
396 | } |
397 | ||
e29136f8 | 398 | static int __check_block_validity(struct inode *inode, const char *func, |
c398eda0 TT |
399 | unsigned int line, |
400 | struct ext4_map_blocks *map) | |
6fd058f7 | 401 | { |
24676da4 TT |
402 | if (!ext4_data_block_valid(EXT4_SB(inode->i_sb), map->m_pblk, |
403 | map->m_len)) { | |
c398eda0 TT |
404 | ext4_error_inode(inode, func, line, map->m_pblk, |
405 | "lblock %lu mapped to illegal pblock " | |
406 | "(length %d)", (unsigned long) map->m_lblk, | |
407 | map->m_len); | |
6a797d27 | 408 | return -EFSCORRUPTED; |
6fd058f7 TT |
409 | } |
410 | return 0; | |
411 | } | |
412 | ||
53085fac JK |
413 | int ext4_issue_zeroout(struct inode *inode, ext4_lblk_t lblk, ext4_fsblk_t pblk, |
414 | ext4_lblk_t len) | |
415 | { | |
416 | int ret; | |
417 | ||
418 | if (ext4_encrypted_inode(inode)) | |
a7550b30 | 419 | return fscrypt_zeroout_range(inode, lblk, pblk, len); |
53085fac JK |
420 | |
421 | ret = sb_issue_zeroout(inode->i_sb, pblk, len, GFP_NOFS); | |
422 | if (ret > 0) | |
423 | ret = 0; | |
424 | ||
425 | return ret; | |
426 | } | |
427 | ||
e29136f8 | 428 | #define check_block_validity(inode, map) \ |
c398eda0 | 429 | __check_block_validity((inode), __func__, __LINE__, (map)) |
e29136f8 | 430 | |
921f266b DM |
431 | #ifdef ES_AGGRESSIVE_TEST |
432 | static void ext4_map_blocks_es_recheck(handle_t *handle, | |
433 | struct inode *inode, | |
434 | struct ext4_map_blocks *es_map, | |
435 | struct ext4_map_blocks *map, | |
436 | int flags) | |
437 | { | |
438 | int retval; | |
439 | ||
440 | map->m_flags = 0; | |
441 | /* | |
442 | * There is a race window that the result is not the same. | |
443 | * e.g. xfstests #223 when dioread_nolock enables. The reason | |
444 | * is that we lookup a block mapping in extent status tree with | |
445 | * out taking i_data_sem. So at the time the unwritten extent | |
446 | * could be converted. | |
447 | */ | |
2dcba478 | 448 | down_read(&EXT4_I(inode)->i_data_sem); |
921f266b DM |
449 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
450 | retval = ext4_ext_map_blocks(handle, inode, map, flags & | |
451 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
452 | } else { | |
453 | retval = ext4_ind_map_blocks(handle, inode, map, flags & | |
454 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
455 | } | |
2dcba478 | 456 | up_read((&EXT4_I(inode)->i_data_sem)); |
921f266b DM |
457 | |
458 | /* | |
459 | * We don't check m_len because extent will be collpased in status | |
460 | * tree. So the m_len might not equal. | |
461 | */ | |
462 | if (es_map->m_lblk != map->m_lblk || | |
463 | es_map->m_flags != map->m_flags || | |
464 | es_map->m_pblk != map->m_pblk) { | |
bdafe42a | 465 | printk("ES cache assertion failed for inode: %lu " |
921f266b DM |
466 | "es_cached ex [%d/%d/%llu/%x] != " |
467 | "found ex [%d/%d/%llu/%x] retval %d flags %x\n", | |
468 | inode->i_ino, es_map->m_lblk, es_map->m_len, | |
469 | es_map->m_pblk, es_map->m_flags, map->m_lblk, | |
470 | map->m_len, map->m_pblk, map->m_flags, | |
471 | retval, flags); | |
472 | } | |
473 | } | |
474 | #endif /* ES_AGGRESSIVE_TEST */ | |
475 | ||
f5ab0d1f | 476 | /* |
e35fd660 | 477 | * The ext4_map_blocks() function tries to look up the requested blocks, |
2b2d6d01 | 478 | * and returns if the blocks are already mapped. |
f5ab0d1f | 479 | * |
f5ab0d1f MC |
480 | * Otherwise it takes the write lock of the i_data_sem and allocate blocks |
481 | * and store the allocated blocks in the result buffer head and mark it | |
482 | * mapped. | |
483 | * | |
e35fd660 TT |
484 | * If file type is extents based, it will call ext4_ext_map_blocks(), |
485 | * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping | |
f5ab0d1f MC |
486 | * based files |
487 | * | |
facab4d9 JK |
488 | * On success, it returns the number of blocks being mapped or allocated. if |
489 | * create==0 and the blocks are pre-allocated and unwritten, the resulting @map | |
490 | * is marked as unwritten. If the create == 1, it will mark @map as mapped. | |
f5ab0d1f MC |
491 | * |
492 | * It returns 0 if plain look up failed (blocks have not been allocated), in | |
facab4d9 JK |
493 | * that case, @map is returned as unmapped but we still do fill map->m_len to |
494 | * indicate the length of a hole starting at map->m_lblk. | |
f5ab0d1f MC |
495 | * |
496 | * It returns the error in case of allocation failure. | |
497 | */ | |
e35fd660 TT |
498 | int ext4_map_blocks(handle_t *handle, struct inode *inode, |
499 | struct ext4_map_blocks *map, int flags) | |
0e855ac8 | 500 | { |
d100eef2 | 501 | struct extent_status es; |
0e855ac8 | 502 | int retval; |
b8a86845 | 503 | int ret = 0; |
921f266b DM |
504 | #ifdef ES_AGGRESSIVE_TEST |
505 | struct ext4_map_blocks orig_map; | |
506 | ||
507 | memcpy(&orig_map, map, sizeof(*map)); | |
508 | #endif | |
f5ab0d1f | 509 | |
e35fd660 TT |
510 | map->m_flags = 0; |
511 | ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u," | |
512 | "logical block %lu\n", inode->i_ino, flags, map->m_len, | |
513 | (unsigned long) map->m_lblk); | |
d100eef2 | 514 | |
e861b5e9 TT |
515 | /* |
516 | * ext4_map_blocks returns an int, and m_len is an unsigned int | |
517 | */ | |
518 | if (unlikely(map->m_len > INT_MAX)) | |
519 | map->m_len = INT_MAX; | |
520 | ||
4adb6ab3 KM |
521 | /* We can handle the block number less than EXT_MAX_BLOCKS */ |
522 | if (unlikely(map->m_lblk >= EXT_MAX_BLOCKS)) | |
6a797d27 | 523 | return -EFSCORRUPTED; |
4adb6ab3 | 524 | |
d100eef2 ZL |
525 | /* Lookup extent status tree firstly */ |
526 | if (ext4_es_lookup_extent(inode, map->m_lblk, &es)) { | |
527 | if (ext4_es_is_written(&es) || ext4_es_is_unwritten(&es)) { | |
528 | map->m_pblk = ext4_es_pblock(&es) + | |
529 | map->m_lblk - es.es_lblk; | |
530 | map->m_flags |= ext4_es_is_written(&es) ? | |
531 | EXT4_MAP_MAPPED : EXT4_MAP_UNWRITTEN; | |
532 | retval = es.es_len - (map->m_lblk - es.es_lblk); | |
533 | if (retval > map->m_len) | |
534 | retval = map->m_len; | |
535 | map->m_len = retval; | |
536 | } else if (ext4_es_is_delayed(&es) || ext4_es_is_hole(&es)) { | |
facab4d9 JK |
537 | map->m_pblk = 0; |
538 | retval = es.es_len - (map->m_lblk - es.es_lblk); | |
539 | if (retval > map->m_len) | |
540 | retval = map->m_len; | |
541 | map->m_len = retval; | |
d100eef2 ZL |
542 | retval = 0; |
543 | } else { | |
544 | BUG_ON(1); | |
545 | } | |
921f266b DM |
546 | #ifdef ES_AGGRESSIVE_TEST |
547 | ext4_map_blocks_es_recheck(handle, inode, map, | |
548 | &orig_map, flags); | |
549 | #endif | |
d100eef2 ZL |
550 | goto found; |
551 | } | |
552 | ||
4df3d265 | 553 | /* |
b920c755 TT |
554 | * Try to see if we can get the block without requesting a new |
555 | * file system block. | |
4df3d265 | 556 | */ |
2dcba478 | 557 | down_read(&EXT4_I(inode)->i_data_sem); |
12e9b892 | 558 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
a4e5d88b DM |
559 | retval = ext4_ext_map_blocks(handle, inode, map, flags & |
560 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
0e855ac8 | 561 | } else { |
a4e5d88b DM |
562 | retval = ext4_ind_map_blocks(handle, inode, map, flags & |
563 | EXT4_GET_BLOCKS_KEEP_SIZE); | |
0e855ac8 | 564 | } |
f7fec032 | 565 | if (retval > 0) { |
3be78c73 | 566 | unsigned int status; |
f7fec032 | 567 | |
44fb851d ZL |
568 | if (unlikely(retval != map->m_len)) { |
569 | ext4_warning(inode->i_sb, | |
570 | "ES len assertion failed for inode " | |
571 | "%lu: retval %d != map->m_len %d", | |
572 | inode->i_ino, retval, map->m_len); | |
573 | WARN_ON(1); | |
921f266b | 574 | } |
921f266b | 575 | |
f7fec032 ZL |
576 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
577 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; | |
578 | if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) && | |
d2dc317d | 579 | !(status & EXTENT_STATUS_WRITTEN) && |
f7fec032 ZL |
580 | ext4_find_delalloc_range(inode, map->m_lblk, |
581 | map->m_lblk + map->m_len - 1)) | |
582 | status |= EXTENT_STATUS_DELAYED; | |
583 | ret = ext4_es_insert_extent(inode, map->m_lblk, | |
584 | map->m_len, map->m_pblk, status); | |
585 | if (ret < 0) | |
586 | retval = ret; | |
587 | } | |
2dcba478 | 588 | up_read((&EXT4_I(inode)->i_data_sem)); |
f5ab0d1f | 589 | |
d100eef2 | 590 | found: |
e35fd660 | 591 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
b8a86845 | 592 | ret = check_block_validity(inode, map); |
6fd058f7 TT |
593 | if (ret != 0) |
594 | return ret; | |
595 | } | |
596 | ||
f5ab0d1f | 597 | /* If it is only a block(s) look up */ |
c2177057 | 598 | if ((flags & EXT4_GET_BLOCKS_CREATE) == 0) |
f5ab0d1f MC |
599 | return retval; |
600 | ||
601 | /* | |
602 | * Returns if the blocks have already allocated | |
603 | * | |
604 | * Note that if blocks have been preallocated | |
df3ab170 | 605 | * ext4_ext_get_block() returns the create = 0 |
f5ab0d1f MC |
606 | * with buffer head unmapped. |
607 | */ | |
e35fd660 | 608 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) |
b8a86845 LC |
609 | /* |
610 | * If we need to convert extent to unwritten | |
611 | * we continue and do the actual work in | |
612 | * ext4_ext_map_blocks() | |
613 | */ | |
614 | if (!(flags & EXT4_GET_BLOCKS_CONVERT_UNWRITTEN)) | |
615 | return retval; | |
4df3d265 | 616 | |
2a8964d6 | 617 | /* |
a25a4e1a ZL |
618 | * Here we clear m_flags because after allocating an new extent, |
619 | * it will be set again. | |
2a8964d6 | 620 | */ |
a25a4e1a | 621 | map->m_flags &= ~EXT4_MAP_FLAGS; |
2a8964d6 | 622 | |
4df3d265 | 623 | /* |
556615dc | 624 | * New blocks allocate and/or writing to unwritten extent |
f5ab0d1f | 625 | * will possibly result in updating i_data, so we take |
d91bd2c1 | 626 | * the write lock of i_data_sem, and call get_block() |
f5ab0d1f | 627 | * with create == 1 flag. |
4df3d265 | 628 | */ |
c8b459f4 | 629 | down_write(&EXT4_I(inode)->i_data_sem); |
d2a17637 | 630 | |
4df3d265 AK |
631 | /* |
632 | * We need to check for EXT4 here because migrate | |
633 | * could have changed the inode type in between | |
634 | */ | |
12e9b892 | 635 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { |
e35fd660 | 636 | retval = ext4_ext_map_blocks(handle, inode, map, flags); |
0e855ac8 | 637 | } else { |
e35fd660 | 638 | retval = ext4_ind_map_blocks(handle, inode, map, flags); |
267e4db9 | 639 | |
e35fd660 | 640 | if (retval > 0 && map->m_flags & EXT4_MAP_NEW) { |
267e4db9 AK |
641 | /* |
642 | * We allocated new blocks which will result in | |
643 | * i_data's format changing. Force the migrate | |
644 | * to fail by clearing migrate flags | |
645 | */ | |
19f5fb7a | 646 | ext4_clear_inode_state(inode, EXT4_STATE_EXT_MIGRATE); |
267e4db9 | 647 | } |
d2a17637 | 648 | |
5f634d06 AK |
649 | /* |
650 | * Update reserved blocks/metadata blocks after successful | |
651 | * block allocation which had been deferred till now. We don't | |
652 | * support fallocate for non extent files. So we can update | |
653 | * reserve space here. | |
654 | */ | |
655 | if ((retval > 0) && | |
1296cc85 | 656 | (flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE)) |
5f634d06 AK |
657 | ext4_da_update_reserve_space(inode, retval, 1); |
658 | } | |
2ac3b6e0 | 659 | |
f7fec032 | 660 | if (retval > 0) { |
3be78c73 | 661 | unsigned int status; |
f7fec032 | 662 | |
44fb851d ZL |
663 | if (unlikely(retval != map->m_len)) { |
664 | ext4_warning(inode->i_sb, | |
665 | "ES len assertion failed for inode " | |
666 | "%lu: retval %d != map->m_len %d", | |
667 | inode->i_ino, retval, map->m_len); | |
668 | WARN_ON(1); | |
921f266b | 669 | } |
921f266b | 670 | |
c86d8db3 JK |
671 | /* |
672 | * We have to zeroout blocks before inserting them into extent | |
673 | * status tree. Otherwise someone could look them up there and | |
9b623df6 JK |
674 | * use them before they are really zeroed. We also have to |
675 | * unmap metadata before zeroing as otherwise writeback can | |
676 | * overwrite zeros with stale data from block device. | |
c86d8db3 JK |
677 | */ |
678 | if (flags & EXT4_GET_BLOCKS_ZERO && | |
679 | map->m_flags & EXT4_MAP_MAPPED && | |
680 | map->m_flags & EXT4_MAP_NEW) { | |
64e1c57f JK |
681 | clean_bdev_aliases(inode->i_sb->s_bdev, map->m_pblk, |
682 | map->m_len); | |
c86d8db3 JK |
683 | ret = ext4_issue_zeroout(inode, map->m_lblk, |
684 | map->m_pblk, map->m_len); | |
685 | if (ret) { | |
686 | retval = ret; | |
687 | goto out_sem; | |
688 | } | |
689 | } | |
690 | ||
adb23551 ZL |
691 | /* |
692 | * If the extent has been zeroed out, we don't need to update | |
693 | * extent status tree. | |
694 | */ | |
695 | if ((flags & EXT4_GET_BLOCKS_PRE_IO) && | |
696 | ext4_es_lookup_extent(inode, map->m_lblk, &es)) { | |
697 | if (ext4_es_is_written(&es)) | |
c86d8db3 | 698 | goto out_sem; |
adb23551 | 699 | } |
f7fec032 ZL |
700 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
701 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; | |
702 | if (!(flags & EXT4_GET_BLOCKS_DELALLOC_RESERVE) && | |
d2dc317d | 703 | !(status & EXTENT_STATUS_WRITTEN) && |
f7fec032 ZL |
704 | ext4_find_delalloc_range(inode, map->m_lblk, |
705 | map->m_lblk + map->m_len - 1)) | |
706 | status |= EXTENT_STATUS_DELAYED; | |
707 | ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, | |
708 | map->m_pblk, status); | |
c86d8db3 | 709 | if (ret < 0) { |
f7fec032 | 710 | retval = ret; |
c86d8db3 JK |
711 | goto out_sem; |
712 | } | |
5356f261 AK |
713 | } |
714 | ||
c86d8db3 | 715 | out_sem: |
4df3d265 | 716 | up_write((&EXT4_I(inode)->i_data_sem)); |
e35fd660 | 717 | if (retval > 0 && map->m_flags & EXT4_MAP_MAPPED) { |
b8a86845 | 718 | ret = check_block_validity(inode, map); |
6fd058f7 TT |
719 | if (ret != 0) |
720 | return ret; | |
06bd3c36 JK |
721 | |
722 | /* | |
723 | * Inodes with freshly allocated blocks where contents will be | |
724 | * visible after transaction commit must be on transaction's | |
725 | * ordered data list. | |
726 | */ | |
727 | if (map->m_flags & EXT4_MAP_NEW && | |
728 | !(map->m_flags & EXT4_MAP_UNWRITTEN) && | |
729 | !(flags & EXT4_GET_BLOCKS_ZERO) && | |
02749a4c | 730 | !ext4_is_quota_file(inode) && |
06bd3c36 | 731 | ext4_should_order_data(inode)) { |
ee0876bc JK |
732 | if (flags & EXT4_GET_BLOCKS_IO_SUBMIT) |
733 | ret = ext4_jbd2_inode_add_wait(handle, inode); | |
734 | else | |
735 | ret = ext4_jbd2_inode_add_write(handle, inode); | |
06bd3c36 JK |
736 | if (ret) |
737 | return ret; | |
738 | } | |
6fd058f7 | 739 | } |
0e855ac8 AK |
740 | return retval; |
741 | } | |
742 | ||
ed8ad838 JK |
743 | /* |
744 | * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages | |
745 | * we have to be careful as someone else may be manipulating b_state as well. | |
746 | */ | |
747 | static void ext4_update_bh_state(struct buffer_head *bh, unsigned long flags) | |
748 | { | |
749 | unsigned long old_state; | |
750 | unsigned long new_state; | |
751 | ||
752 | flags &= EXT4_MAP_FLAGS; | |
753 | ||
754 | /* Dummy buffer_head? Set non-atomically. */ | |
755 | if (!bh->b_page) { | |
756 | bh->b_state = (bh->b_state & ~EXT4_MAP_FLAGS) | flags; | |
757 | return; | |
758 | } | |
759 | /* | |
760 | * Someone else may be modifying b_state. Be careful! This is ugly but | |
761 | * once we get rid of using bh as a container for mapping information | |
762 | * to pass to / from get_block functions, this can go away. | |
763 | */ | |
764 | do { | |
765 | old_state = READ_ONCE(bh->b_state); | |
766 | new_state = (old_state & ~EXT4_MAP_FLAGS) | flags; | |
767 | } while (unlikely( | |
768 | cmpxchg(&bh->b_state, old_state, new_state) != old_state)); | |
769 | } | |
770 | ||
2ed88685 TT |
771 | static int _ext4_get_block(struct inode *inode, sector_t iblock, |
772 | struct buffer_head *bh, int flags) | |
ac27a0ec | 773 | { |
2ed88685 | 774 | struct ext4_map_blocks map; |
efe70c29 | 775 | int ret = 0; |
ac27a0ec | 776 | |
46c7f254 TM |
777 | if (ext4_has_inline_data(inode)) |
778 | return -ERANGE; | |
779 | ||
2ed88685 TT |
780 | map.m_lblk = iblock; |
781 | map.m_len = bh->b_size >> inode->i_blkbits; | |
782 | ||
efe70c29 JK |
783 | ret = ext4_map_blocks(ext4_journal_current_handle(), inode, &map, |
784 | flags); | |
7fb5409d | 785 | if (ret > 0) { |
2ed88685 | 786 | map_bh(bh, inode->i_sb, map.m_pblk); |
ed8ad838 | 787 | ext4_update_bh_state(bh, map.m_flags); |
2ed88685 | 788 | bh->b_size = inode->i_sb->s_blocksize * map.m_len; |
7fb5409d | 789 | ret = 0; |
547edce3 RZ |
790 | } else if (ret == 0) { |
791 | /* hole case, need to fill in bh->b_size */ | |
792 | bh->b_size = inode->i_sb->s_blocksize * map.m_len; | |
ac27a0ec DK |
793 | } |
794 | return ret; | |
795 | } | |
796 | ||
2ed88685 TT |
797 | int ext4_get_block(struct inode *inode, sector_t iblock, |
798 | struct buffer_head *bh, int create) | |
799 | { | |
800 | return _ext4_get_block(inode, iblock, bh, | |
801 | create ? EXT4_GET_BLOCKS_CREATE : 0); | |
802 | } | |
803 | ||
705965bd JK |
804 | /* |
805 | * Get block function used when preparing for buffered write if we require | |
806 | * creating an unwritten extent if blocks haven't been allocated. The extent | |
807 | * will be converted to written after the IO is complete. | |
808 | */ | |
809 | int ext4_get_block_unwritten(struct inode *inode, sector_t iblock, | |
810 | struct buffer_head *bh_result, int create) | |
811 | { | |
812 | ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n", | |
813 | inode->i_ino, create); | |
814 | return _ext4_get_block(inode, iblock, bh_result, | |
815 | EXT4_GET_BLOCKS_IO_CREATE_EXT); | |
816 | } | |
817 | ||
efe70c29 JK |
818 | /* Maximum number of blocks we map for direct IO at once. */ |
819 | #define DIO_MAX_BLOCKS 4096 | |
820 | ||
e84dfbe2 JK |
821 | /* |
822 | * Get blocks function for the cases that need to start a transaction - | |
823 | * generally difference cases of direct IO and DAX IO. It also handles retries | |
824 | * in case of ENOSPC. | |
825 | */ | |
826 | static int ext4_get_block_trans(struct inode *inode, sector_t iblock, | |
827 | struct buffer_head *bh_result, int flags) | |
efe70c29 JK |
828 | { |
829 | int dio_credits; | |
e84dfbe2 JK |
830 | handle_t *handle; |
831 | int retries = 0; | |
832 | int ret; | |
efe70c29 JK |
833 | |
834 | /* Trim mapping request to maximum we can map at once for DIO */ | |
835 | if (bh_result->b_size >> inode->i_blkbits > DIO_MAX_BLOCKS) | |
836 | bh_result->b_size = DIO_MAX_BLOCKS << inode->i_blkbits; | |
837 | dio_credits = ext4_chunk_trans_blocks(inode, | |
838 | bh_result->b_size >> inode->i_blkbits); | |
e84dfbe2 JK |
839 | retry: |
840 | handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, dio_credits); | |
841 | if (IS_ERR(handle)) | |
842 | return PTR_ERR(handle); | |
843 | ||
844 | ret = _ext4_get_block(inode, iblock, bh_result, flags); | |
845 | ext4_journal_stop(handle); | |
846 | ||
847 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) | |
848 | goto retry; | |
849 | return ret; | |
efe70c29 JK |
850 | } |
851 | ||
705965bd JK |
852 | /* Get block function for DIO reads and writes to inodes without extents */ |
853 | int ext4_dio_get_block(struct inode *inode, sector_t iblock, | |
854 | struct buffer_head *bh, int create) | |
855 | { | |
efe70c29 JK |
856 | /* We don't expect handle for direct IO */ |
857 | WARN_ON_ONCE(ext4_journal_current_handle()); | |
858 | ||
e84dfbe2 JK |
859 | if (!create) |
860 | return _ext4_get_block(inode, iblock, bh, 0); | |
861 | return ext4_get_block_trans(inode, iblock, bh, EXT4_GET_BLOCKS_CREATE); | |
705965bd JK |
862 | } |
863 | ||
864 | /* | |
109811c2 | 865 | * Get block function for AIO DIO writes when we create unwritten extent if |
705965bd JK |
866 | * blocks are not allocated yet. The extent will be converted to written |
867 | * after IO is complete. | |
868 | */ | |
109811c2 JK |
869 | static int ext4_dio_get_block_unwritten_async(struct inode *inode, |
870 | sector_t iblock, struct buffer_head *bh_result, int create) | |
705965bd | 871 | { |
efe70c29 JK |
872 | int ret; |
873 | ||
efe70c29 JK |
874 | /* We don't expect handle for direct IO */ |
875 | WARN_ON_ONCE(ext4_journal_current_handle()); | |
876 | ||
e84dfbe2 JK |
877 | ret = ext4_get_block_trans(inode, iblock, bh_result, |
878 | EXT4_GET_BLOCKS_IO_CREATE_EXT); | |
efe70c29 | 879 | |
109811c2 JK |
880 | /* |
881 | * When doing DIO using unwritten extents, we need io_end to convert | |
882 | * unwritten extents to written on IO completion. We allocate io_end | |
883 | * once we spot unwritten extent and store it in b_private. Generic | |
884 | * DIO code keeps b_private set and furthermore passes the value to | |
885 | * our completion callback in 'private' argument. | |
886 | */ | |
887 | if (!ret && buffer_unwritten(bh_result)) { | |
888 | if (!bh_result->b_private) { | |
889 | ext4_io_end_t *io_end; | |
890 | ||
891 | io_end = ext4_init_io_end(inode, GFP_KERNEL); | |
892 | if (!io_end) | |
893 | return -ENOMEM; | |
894 | bh_result->b_private = io_end; | |
895 | ext4_set_io_unwritten_flag(inode, io_end); | |
896 | } | |
efe70c29 | 897 | set_buffer_defer_completion(bh_result); |
efe70c29 JK |
898 | } |
899 | ||
900 | return ret; | |
705965bd JK |
901 | } |
902 | ||
109811c2 JK |
903 | /* |
904 | * Get block function for non-AIO DIO writes when we create unwritten extent if | |
905 | * blocks are not allocated yet. The extent will be converted to written | |
1e21196c | 906 | * after IO is complete by ext4_direct_IO_write(). |
109811c2 JK |
907 | */ |
908 | static int ext4_dio_get_block_unwritten_sync(struct inode *inode, | |
909 | sector_t iblock, struct buffer_head *bh_result, int create) | |
910 | { | |
109811c2 JK |
911 | int ret; |
912 | ||
913 | /* We don't expect handle for direct IO */ | |
914 | WARN_ON_ONCE(ext4_journal_current_handle()); | |
915 | ||
e84dfbe2 JK |
916 | ret = ext4_get_block_trans(inode, iblock, bh_result, |
917 | EXT4_GET_BLOCKS_IO_CREATE_EXT); | |
109811c2 JK |
918 | |
919 | /* | |
920 | * Mark inode as having pending DIO writes to unwritten extents. | |
1e21196c | 921 | * ext4_direct_IO_write() checks this flag and converts extents to |
109811c2 JK |
922 | * written. |
923 | */ | |
924 | if (!ret && buffer_unwritten(bh_result)) | |
925 | ext4_set_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN); | |
926 | ||
927 | return ret; | |
928 | } | |
929 | ||
705965bd JK |
930 | static int ext4_dio_get_block_overwrite(struct inode *inode, sector_t iblock, |
931 | struct buffer_head *bh_result, int create) | |
932 | { | |
933 | int ret; | |
934 | ||
935 | ext4_debug("ext4_dio_get_block_overwrite: inode %lu, create flag %d\n", | |
936 | inode->i_ino, create); | |
efe70c29 JK |
937 | /* We don't expect handle for direct IO */ |
938 | WARN_ON_ONCE(ext4_journal_current_handle()); | |
939 | ||
705965bd JK |
940 | ret = _ext4_get_block(inode, iblock, bh_result, 0); |
941 | /* | |
942 | * Blocks should have been preallocated! ext4_file_write_iter() checks | |
943 | * that. | |
944 | */ | |
efe70c29 | 945 | WARN_ON_ONCE(!buffer_mapped(bh_result) || buffer_unwritten(bh_result)); |
705965bd JK |
946 | |
947 | return ret; | |
948 | } | |
949 | ||
950 | ||
ac27a0ec DK |
951 | /* |
952 | * `handle' can be NULL if create is zero | |
953 | */ | |
617ba13b | 954 | struct buffer_head *ext4_getblk(handle_t *handle, struct inode *inode, |
c5e298ae | 955 | ext4_lblk_t block, int map_flags) |
ac27a0ec | 956 | { |
2ed88685 TT |
957 | struct ext4_map_blocks map; |
958 | struct buffer_head *bh; | |
c5e298ae | 959 | int create = map_flags & EXT4_GET_BLOCKS_CREATE; |
10560082 | 960 | int err; |
ac27a0ec DK |
961 | |
962 | J_ASSERT(handle != NULL || create == 0); | |
963 | ||
2ed88685 TT |
964 | map.m_lblk = block; |
965 | map.m_len = 1; | |
c5e298ae | 966 | err = ext4_map_blocks(handle, inode, &map, map_flags); |
ac27a0ec | 967 | |
10560082 TT |
968 | if (err == 0) |
969 | return create ? ERR_PTR(-ENOSPC) : NULL; | |
2ed88685 | 970 | if (err < 0) |
10560082 | 971 | return ERR_PTR(err); |
2ed88685 TT |
972 | |
973 | bh = sb_getblk(inode->i_sb, map.m_pblk); | |
10560082 TT |
974 | if (unlikely(!bh)) |
975 | return ERR_PTR(-ENOMEM); | |
2ed88685 TT |
976 | if (map.m_flags & EXT4_MAP_NEW) { |
977 | J_ASSERT(create != 0); | |
978 | J_ASSERT(handle != NULL); | |
ac27a0ec | 979 | |
2ed88685 TT |
980 | /* |
981 | * Now that we do not always journal data, we should | |
982 | * keep in mind whether this should always journal the | |
983 | * new buffer as metadata. For now, regular file | |
984 | * writes use ext4_get_block instead, so it's not a | |
985 | * problem. | |
986 | */ | |
987 | lock_buffer(bh); | |
988 | BUFFER_TRACE(bh, "call get_create_access"); | |
10560082 TT |
989 | err = ext4_journal_get_create_access(handle, bh); |
990 | if (unlikely(err)) { | |
991 | unlock_buffer(bh); | |
992 | goto errout; | |
993 | } | |
994 | if (!buffer_uptodate(bh)) { | |
2ed88685 TT |
995 | memset(bh->b_data, 0, inode->i_sb->s_blocksize); |
996 | set_buffer_uptodate(bh); | |
ac27a0ec | 997 | } |
2ed88685 TT |
998 | unlock_buffer(bh); |
999 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); | |
1000 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
10560082 TT |
1001 | if (unlikely(err)) |
1002 | goto errout; | |
1003 | } else | |
2ed88685 | 1004 | BUFFER_TRACE(bh, "not a new buffer"); |
2ed88685 | 1005 | return bh; |
10560082 TT |
1006 | errout: |
1007 | brelse(bh); | |
1008 | return ERR_PTR(err); | |
ac27a0ec DK |
1009 | } |
1010 | ||
617ba13b | 1011 | struct buffer_head *ext4_bread(handle_t *handle, struct inode *inode, |
c5e298ae | 1012 | ext4_lblk_t block, int map_flags) |
ac27a0ec | 1013 | { |
af5bc92d | 1014 | struct buffer_head *bh; |
ac27a0ec | 1015 | |
c5e298ae | 1016 | bh = ext4_getblk(handle, inode, block, map_flags); |
1c215028 | 1017 | if (IS_ERR(bh)) |
ac27a0ec | 1018 | return bh; |
1c215028 | 1019 | if (!bh || buffer_uptodate(bh)) |
ac27a0ec | 1020 | return bh; |
dfec8a14 | 1021 | ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, &bh); |
ac27a0ec DK |
1022 | wait_on_buffer(bh); |
1023 | if (buffer_uptodate(bh)) | |
1024 | return bh; | |
1025 | put_bh(bh); | |
1c215028 | 1026 | return ERR_PTR(-EIO); |
ac27a0ec DK |
1027 | } |
1028 | ||
9699d4f9 TE |
1029 | /* Read a contiguous batch of blocks. */ |
1030 | int ext4_bread_batch(struct inode *inode, ext4_lblk_t block, int bh_count, | |
1031 | bool wait, struct buffer_head **bhs) | |
1032 | { | |
1033 | int i, err; | |
1034 | ||
1035 | for (i = 0; i < bh_count; i++) { | |
1036 | bhs[i] = ext4_getblk(NULL, inode, block + i, 0 /* map_flags */); | |
1037 | if (IS_ERR(bhs[i])) { | |
1038 | err = PTR_ERR(bhs[i]); | |
1039 | bh_count = i; | |
1040 | goto out_brelse; | |
1041 | } | |
1042 | } | |
1043 | ||
1044 | for (i = 0; i < bh_count; i++) | |
1045 | /* Note that NULL bhs[i] is valid because of holes. */ | |
1046 | if (bhs[i] && !buffer_uptodate(bhs[i])) | |
1047 | ll_rw_block(REQ_OP_READ, REQ_META | REQ_PRIO, 1, | |
1048 | &bhs[i]); | |
1049 | ||
1050 | if (!wait) | |
1051 | return 0; | |
1052 | ||
1053 | for (i = 0; i < bh_count; i++) | |
1054 | if (bhs[i]) | |
1055 | wait_on_buffer(bhs[i]); | |
1056 | ||
1057 | for (i = 0; i < bh_count; i++) { | |
1058 | if (bhs[i] && !buffer_uptodate(bhs[i])) { | |
1059 | err = -EIO; | |
1060 | goto out_brelse; | |
1061 | } | |
1062 | } | |
1063 | return 0; | |
1064 | ||
1065 | out_brelse: | |
1066 | for (i = 0; i < bh_count; i++) { | |
1067 | brelse(bhs[i]); | |
1068 | bhs[i] = NULL; | |
1069 | } | |
1070 | return err; | |
1071 | } | |
1072 | ||
f19d5870 TM |
1073 | int ext4_walk_page_buffers(handle_t *handle, |
1074 | struct buffer_head *head, | |
1075 | unsigned from, | |
1076 | unsigned to, | |
1077 | int *partial, | |
1078 | int (*fn)(handle_t *handle, | |
1079 | struct buffer_head *bh)) | |
ac27a0ec DK |
1080 | { |
1081 | struct buffer_head *bh; | |
1082 | unsigned block_start, block_end; | |
1083 | unsigned blocksize = head->b_size; | |
1084 | int err, ret = 0; | |
1085 | struct buffer_head *next; | |
1086 | ||
af5bc92d TT |
1087 | for (bh = head, block_start = 0; |
1088 | ret == 0 && (bh != head || !block_start); | |
de9a55b8 | 1089 | block_start = block_end, bh = next) { |
ac27a0ec DK |
1090 | next = bh->b_this_page; |
1091 | block_end = block_start + blocksize; | |
1092 | if (block_end <= from || block_start >= to) { | |
1093 | if (partial && !buffer_uptodate(bh)) | |
1094 | *partial = 1; | |
1095 | continue; | |
1096 | } | |
1097 | err = (*fn)(handle, bh); | |
1098 | if (!ret) | |
1099 | ret = err; | |
1100 | } | |
1101 | return ret; | |
1102 | } | |
1103 | ||
1104 | /* | |
1105 | * To preserve ordering, it is essential that the hole instantiation and | |
1106 | * the data write be encapsulated in a single transaction. We cannot | |
617ba13b | 1107 | * close off a transaction and start a new one between the ext4_get_block() |
dab291af | 1108 | * and the commit_write(). So doing the jbd2_journal_start at the start of |
ac27a0ec DK |
1109 | * prepare_write() is the right place. |
1110 | * | |
36ade451 JK |
1111 | * Also, this function can nest inside ext4_writepage(). In that case, we |
1112 | * *know* that ext4_writepage() has generated enough buffer credits to do the | |
1113 | * whole page. So we won't block on the journal in that case, which is good, | |
1114 | * because the caller may be PF_MEMALLOC. | |
ac27a0ec | 1115 | * |
617ba13b | 1116 | * By accident, ext4 can be reentered when a transaction is open via |
ac27a0ec DK |
1117 | * quota file writes. If we were to commit the transaction while thus |
1118 | * reentered, there can be a deadlock - we would be holding a quota | |
1119 | * lock, and the commit would never complete if another thread had a | |
1120 | * transaction open and was blocking on the quota lock - a ranking | |
1121 | * violation. | |
1122 | * | |
dab291af | 1123 | * So what we do is to rely on the fact that jbd2_journal_stop/journal_start |
ac27a0ec DK |
1124 | * will _not_ run commit under these circumstances because handle->h_ref |
1125 | * is elevated. We'll still have enough credits for the tiny quotafile | |
1126 | * write. | |
1127 | */ | |
f19d5870 TM |
1128 | int do_journal_get_write_access(handle_t *handle, |
1129 | struct buffer_head *bh) | |
ac27a0ec | 1130 | { |
56d35a4c JK |
1131 | int dirty = buffer_dirty(bh); |
1132 | int ret; | |
1133 | ||
ac27a0ec DK |
1134 | if (!buffer_mapped(bh) || buffer_freed(bh)) |
1135 | return 0; | |
56d35a4c | 1136 | /* |
ebdec241 | 1137 | * __block_write_begin() could have dirtied some buffers. Clean |
56d35a4c JK |
1138 | * the dirty bit as jbd2_journal_get_write_access() could complain |
1139 | * otherwise about fs integrity issues. Setting of the dirty bit | |
ebdec241 | 1140 | * by __block_write_begin() isn't a real problem here as we clear |
56d35a4c JK |
1141 | * the bit before releasing a page lock and thus writeback cannot |
1142 | * ever write the buffer. | |
1143 | */ | |
1144 | if (dirty) | |
1145 | clear_buffer_dirty(bh); | |
5d601255 | 1146 | BUFFER_TRACE(bh, "get write access"); |
56d35a4c JK |
1147 | ret = ext4_journal_get_write_access(handle, bh); |
1148 | if (!ret && dirty) | |
1149 | ret = ext4_handle_dirty_metadata(handle, NULL, bh); | |
1150 | return ret; | |
ac27a0ec DK |
1151 | } |
1152 | ||
2058f83a MH |
1153 | #ifdef CONFIG_EXT4_FS_ENCRYPTION |
1154 | static int ext4_block_write_begin(struct page *page, loff_t pos, unsigned len, | |
1155 | get_block_t *get_block) | |
1156 | { | |
09cbfeaf | 1157 | unsigned from = pos & (PAGE_SIZE - 1); |
2058f83a MH |
1158 | unsigned to = from + len; |
1159 | struct inode *inode = page->mapping->host; | |
1160 | unsigned block_start, block_end; | |
1161 | sector_t block; | |
1162 | int err = 0; | |
1163 | unsigned blocksize = inode->i_sb->s_blocksize; | |
1164 | unsigned bbits; | |
1165 | struct buffer_head *bh, *head, *wait[2], **wait_bh = wait; | |
1166 | bool decrypt = false; | |
1167 | ||
1168 | BUG_ON(!PageLocked(page)); | |
09cbfeaf KS |
1169 | BUG_ON(from > PAGE_SIZE); |
1170 | BUG_ON(to > PAGE_SIZE); | |
2058f83a MH |
1171 | BUG_ON(from > to); |
1172 | ||
1173 | if (!page_has_buffers(page)) | |
1174 | create_empty_buffers(page, blocksize, 0); | |
1175 | head = page_buffers(page); | |
1176 | bbits = ilog2(blocksize); | |
09cbfeaf | 1177 | block = (sector_t)page->index << (PAGE_SHIFT - bbits); |
2058f83a MH |
1178 | |
1179 | for (bh = head, block_start = 0; bh != head || !block_start; | |
1180 | block++, block_start = block_end, bh = bh->b_this_page) { | |
1181 | block_end = block_start + blocksize; | |
1182 | if (block_end <= from || block_start >= to) { | |
1183 | if (PageUptodate(page)) { | |
1184 | if (!buffer_uptodate(bh)) | |
1185 | set_buffer_uptodate(bh); | |
1186 | } | |
1187 | continue; | |
1188 | } | |
1189 | if (buffer_new(bh)) | |
1190 | clear_buffer_new(bh); | |
1191 | if (!buffer_mapped(bh)) { | |
1192 | WARN_ON(bh->b_size != blocksize); | |
1193 | err = get_block(inode, block, bh, 1); | |
1194 | if (err) | |
1195 | break; | |
1196 | if (buffer_new(bh)) { | |
e64855c6 | 1197 | clean_bdev_bh_alias(bh); |
2058f83a MH |
1198 | if (PageUptodate(page)) { |
1199 | clear_buffer_new(bh); | |
1200 | set_buffer_uptodate(bh); | |
1201 | mark_buffer_dirty(bh); | |
1202 | continue; | |
1203 | } | |
1204 | if (block_end > to || block_start < from) | |
1205 | zero_user_segments(page, to, block_end, | |
1206 | block_start, from); | |
1207 | continue; | |
1208 | } | |
1209 | } | |
1210 | if (PageUptodate(page)) { | |
1211 | if (!buffer_uptodate(bh)) | |
1212 | set_buffer_uptodate(bh); | |
1213 | continue; | |
1214 | } | |
1215 | if (!buffer_uptodate(bh) && !buffer_delay(bh) && | |
1216 | !buffer_unwritten(bh) && | |
1217 | (block_start < from || block_end > to)) { | |
dfec8a14 | 1218 | ll_rw_block(REQ_OP_READ, 0, 1, &bh); |
2058f83a MH |
1219 | *wait_bh++ = bh; |
1220 | decrypt = ext4_encrypted_inode(inode) && | |
1221 | S_ISREG(inode->i_mode); | |
1222 | } | |
1223 | } | |
1224 | /* | |
1225 | * If we issued read requests, let them complete. | |
1226 | */ | |
1227 | while (wait_bh > wait) { | |
1228 | wait_on_buffer(*--wait_bh); | |
1229 | if (!buffer_uptodate(*wait_bh)) | |
1230 | err = -EIO; | |
1231 | } | |
1232 | if (unlikely(err)) | |
1233 | page_zero_new_buffers(page, from, to); | |
1234 | else if (decrypt) | |
7821d4dd | 1235 | err = fscrypt_decrypt_page(page->mapping->host, page, |
9c4bb8a3 | 1236 | PAGE_SIZE, 0, page->index); |
2058f83a MH |
1237 | return err; |
1238 | } | |
1239 | #endif | |
1240 | ||
bfc1af65 | 1241 | static int ext4_write_begin(struct file *file, struct address_space *mapping, |
de9a55b8 TT |
1242 | loff_t pos, unsigned len, unsigned flags, |
1243 | struct page **pagep, void **fsdata) | |
ac27a0ec | 1244 | { |
af5bc92d | 1245 | struct inode *inode = mapping->host; |
1938a150 | 1246 | int ret, needed_blocks; |
ac27a0ec DK |
1247 | handle_t *handle; |
1248 | int retries = 0; | |
af5bc92d | 1249 | struct page *page; |
de9a55b8 | 1250 | pgoff_t index; |
af5bc92d | 1251 | unsigned from, to; |
bfc1af65 | 1252 | |
0db1ff22 TT |
1253 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
1254 | return -EIO; | |
1255 | ||
9bffad1e | 1256 | trace_ext4_write_begin(inode, pos, len, flags); |
1938a150 AK |
1257 | /* |
1258 | * Reserve one block more for addition to orphan list in case | |
1259 | * we allocate blocks but write fails for some reason | |
1260 | */ | |
1261 | needed_blocks = ext4_writepage_trans_blocks(inode) + 1; | |
09cbfeaf KS |
1262 | index = pos >> PAGE_SHIFT; |
1263 | from = pos & (PAGE_SIZE - 1); | |
af5bc92d | 1264 | to = from + len; |
ac27a0ec | 1265 | |
f19d5870 TM |
1266 | if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { |
1267 | ret = ext4_try_to_write_inline_data(mapping, inode, pos, len, | |
1268 | flags, pagep); | |
1269 | if (ret < 0) | |
47564bfb TT |
1270 | return ret; |
1271 | if (ret == 1) | |
1272 | return 0; | |
f19d5870 TM |
1273 | } |
1274 | ||
47564bfb TT |
1275 | /* |
1276 | * grab_cache_page_write_begin() can take a long time if the | |
1277 | * system is thrashing due to memory pressure, or if the page | |
1278 | * is being written back. So grab it first before we start | |
1279 | * the transaction handle. This also allows us to allocate | |
1280 | * the page (if needed) without using GFP_NOFS. | |
1281 | */ | |
1282 | retry_grab: | |
1283 | page = grab_cache_page_write_begin(mapping, index, flags); | |
1284 | if (!page) | |
1285 | return -ENOMEM; | |
1286 | unlock_page(page); | |
1287 | ||
1288 | retry_journal: | |
9924a92a | 1289 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, needed_blocks); |
af5bc92d | 1290 | if (IS_ERR(handle)) { |
09cbfeaf | 1291 | put_page(page); |
47564bfb | 1292 | return PTR_ERR(handle); |
7479d2b9 | 1293 | } |
ac27a0ec | 1294 | |
47564bfb TT |
1295 | lock_page(page); |
1296 | if (page->mapping != mapping) { | |
1297 | /* The page got truncated from under us */ | |
1298 | unlock_page(page); | |
09cbfeaf | 1299 | put_page(page); |
cf108bca | 1300 | ext4_journal_stop(handle); |
47564bfb | 1301 | goto retry_grab; |
cf108bca | 1302 | } |
7afe5aa5 DM |
1303 | /* In case writeback began while the page was unlocked */ |
1304 | wait_for_stable_page(page); | |
cf108bca | 1305 | |
2058f83a MH |
1306 | #ifdef CONFIG_EXT4_FS_ENCRYPTION |
1307 | if (ext4_should_dioread_nolock(inode)) | |
1308 | ret = ext4_block_write_begin(page, pos, len, | |
705965bd | 1309 | ext4_get_block_unwritten); |
2058f83a MH |
1310 | else |
1311 | ret = ext4_block_write_begin(page, pos, len, | |
1312 | ext4_get_block); | |
1313 | #else | |
744692dc | 1314 | if (ext4_should_dioread_nolock(inode)) |
705965bd JK |
1315 | ret = __block_write_begin(page, pos, len, |
1316 | ext4_get_block_unwritten); | |
744692dc | 1317 | else |
6e1db88d | 1318 | ret = __block_write_begin(page, pos, len, ext4_get_block); |
2058f83a | 1319 | #endif |
bfc1af65 | 1320 | if (!ret && ext4_should_journal_data(inode)) { |
f19d5870 TM |
1321 | ret = ext4_walk_page_buffers(handle, page_buffers(page), |
1322 | from, to, NULL, | |
1323 | do_journal_get_write_access); | |
ac27a0ec | 1324 | } |
bfc1af65 NP |
1325 | |
1326 | if (ret) { | |
af5bc92d | 1327 | unlock_page(page); |
ae4d5372 | 1328 | /* |
6e1db88d | 1329 | * __block_write_begin may have instantiated a few blocks |
ae4d5372 AK |
1330 | * outside i_size. Trim these off again. Don't need |
1331 | * i_size_read because we hold i_mutex. | |
1938a150 AK |
1332 | * |
1333 | * Add inode to orphan list in case we crash before | |
1334 | * truncate finishes | |
ae4d5372 | 1335 | */ |
ffacfa7a | 1336 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
1938a150 AK |
1337 | ext4_orphan_add(handle, inode); |
1338 | ||
1339 | ext4_journal_stop(handle); | |
1340 | if (pos + len > inode->i_size) { | |
b9a4207d | 1341 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1342 | /* |
ffacfa7a | 1343 | * If truncate failed early the inode might |
1938a150 AK |
1344 | * still be on the orphan list; we need to |
1345 | * make sure the inode is removed from the | |
1346 | * orphan list in that case. | |
1347 | */ | |
1348 | if (inode->i_nlink) | |
1349 | ext4_orphan_del(NULL, inode); | |
1350 | } | |
bfc1af65 | 1351 | |
47564bfb TT |
1352 | if (ret == -ENOSPC && |
1353 | ext4_should_retry_alloc(inode->i_sb, &retries)) | |
1354 | goto retry_journal; | |
09cbfeaf | 1355 | put_page(page); |
47564bfb TT |
1356 | return ret; |
1357 | } | |
1358 | *pagep = page; | |
ac27a0ec DK |
1359 | return ret; |
1360 | } | |
1361 | ||
bfc1af65 NP |
1362 | /* For write_end() in data=journal mode */ |
1363 | static int write_end_fn(handle_t *handle, struct buffer_head *bh) | |
ac27a0ec | 1364 | { |
13fca323 | 1365 | int ret; |
ac27a0ec DK |
1366 | if (!buffer_mapped(bh) || buffer_freed(bh)) |
1367 | return 0; | |
1368 | set_buffer_uptodate(bh); | |
13fca323 TT |
1369 | ret = ext4_handle_dirty_metadata(handle, NULL, bh); |
1370 | clear_buffer_meta(bh); | |
1371 | clear_buffer_prio(bh); | |
1372 | return ret; | |
ac27a0ec DK |
1373 | } |
1374 | ||
eed4333f ZL |
1375 | /* |
1376 | * We need to pick up the new inode size which generic_commit_write gave us | |
1377 | * `file' can be NULL - eg, when called from page_symlink(). | |
1378 | * | |
1379 | * ext4 never places buffers on inode->i_mapping->private_list. metadata | |
1380 | * buffers are managed internally. | |
1381 | */ | |
1382 | static int ext4_write_end(struct file *file, | |
1383 | struct address_space *mapping, | |
1384 | loff_t pos, unsigned len, unsigned copied, | |
1385 | struct page *page, void *fsdata) | |
f8514083 | 1386 | { |
f8514083 | 1387 | handle_t *handle = ext4_journal_current_handle(); |
eed4333f | 1388 | struct inode *inode = mapping->host; |
0572639f | 1389 | loff_t old_size = inode->i_size; |
eed4333f ZL |
1390 | int ret = 0, ret2; |
1391 | int i_size_changed = 0; | |
1392 | ||
1393 | trace_ext4_write_end(inode, pos, len, copied); | |
42c832de TT |
1394 | if (ext4_has_inline_data(inode)) { |
1395 | ret = ext4_write_inline_data_end(inode, pos, len, | |
1396 | copied, page); | |
eb5efbcb TT |
1397 | if (ret < 0) { |
1398 | unlock_page(page); | |
1399 | put_page(page); | |
42c832de | 1400 | goto errout; |
eb5efbcb | 1401 | } |
42c832de TT |
1402 | copied = ret; |
1403 | } else | |
f19d5870 TM |
1404 | copied = block_write_end(file, mapping, pos, |
1405 | len, copied, page, fsdata); | |
f8514083 | 1406 | /* |
4631dbf6 | 1407 | * it's important to update i_size while still holding page lock: |
f8514083 AK |
1408 | * page writeout could otherwise come in and zero beyond i_size. |
1409 | */ | |
4631dbf6 | 1410 | i_size_changed = ext4_update_inode_size(inode, pos + copied); |
f8514083 | 1411 | unlock_page(page); |
09cbfeaf | 1412 | put_page(page); |
f8514083 | 1413 | |
0572639f XW |
1414 | if (old_size < pos) |
1415 | pagecache_isize_extended(inode, old_size, pos); | |
f8514083 AK |
1416 | /* |
1417 | * Don't mark the inode dirty under page lock. First, it unnecessarily | |
1418 | * makes the holding time of page lock longer. Second, it forces lock | |
1419 | * ordering of page lock and transaction start for journaling | |
1420 | * filesystems. | |
1421 | */ | |
1422 | if (i_size_changed) | |
1423 | ext4_mark_inode_dirty(handle, inode); | |
1424 | ||
ffacfa7a | 1425 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
f8514083 AK |
1426 | /* if we have allocated more blocks and copied |
1427 | * less. We will have blocks allocated outside | |
1428 | * inode->i_size. So truncate them | |
1429 | */ | |
1430 | ext4_orphan_add(handle, inode); | |
74d553aa | 1431 | errout: |
617ba13b | 1432 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1433 | if (!ret) |
1434 | ret = ret2; | |
bfc1af65 | 1435 | |
f8514083 | 1436 | if (pos + len > inode->i_size) { |
b9a4207d | 1437 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1438 | /* |
ffacfa7a | 1439 | * If truncate failed early the inode might still be |
f8514083 AK |
1440 | * on the orphan list; we need to make sure the inode |
1441 | * is removed from the orphan list in that case. | |
1442 | */ | |
1443 | if (inode->i_nlink) | |
1444 | ext4_orphan_del(NULL, inode); | |
1445 | } | |
1446 | ||
bfc1af65 | 1447 | return ret ? ret : copied; |
ac27a0ec DK |
1448 | } |
1449 | ||
b90197b6 TT |
1450 | /* |
1451 | * This is a private version of page_zero_new_buffers() which doesn't | |
1452 | * set the buffer to be dirty, since in data=journalled mode we need | |
1453 | * to call ext4_handle_dirty_metadata() instead. | |
1454 | */ | |
3b136499 JK |
1455 | static void ext4_journalled_zero_new_buffers(handle_t *handle, |
1456 | struct page *page, | |
1457 | unsigned from, unsigned to) | |
b90197b6 TT |
1458 | { |
1459 | unsigned int block_start = 0, block_end; | |
1460 | struct buffer_head *head, *bh; | |
1461 | ||
1462 | bh = head = page_buffers(page); | |
1463 | do { | |
1464 | block_end = block_start + bh->b_size; | |
1465 | if (buffer_new(bh)) { | |
1466 | if (block_end > from && block_start < to) { | |
1467 | if (!PageUptodate(page)) { | |
1468 | unsigned start, size; | |
1469 | ||
1470 | start = max(from, block_start); | |
1471 | size = min(to, block_end) - start; | |
1472 | ||
1473 | zero_user(page, start, size); | |
3b136499 | 1474 | write_end_fn(handle, bh); |
b90197b6 TT |
1475 | } |
1476 | clear_buffer_new(bh); | |
1477 | } | |
1478 | } | |
1479 | block_start = block_end; | |
1480 | bh = bh->b_this_page; | |
1481 | } while (bh != head); | |
1482 | } | |
1483 | ||
bfc1af65 | 1484 | static int ext4_journalled_write_end(struct file *file, |
de9a55b8 TT |
1485 | struct address_space *mapping, |
1486 | loff_t pos, unsigned len, unsigned copied, | |
1487 | struct page *page, void *fsdata) | |
ac27a0ec | 1488 | { |
617ba13b | 1489 | handle_t *handle = ext4_journal_current_handle(); |
bfc1af65 | 1490 | struct inode *inode = mapping->host; |
0572639f | 1491 | loff_t old_size = inode->i_size; |
ac27a0ec DK |
1492 | int ret = 0, ret2; |
1493 | int partial = 0; | |
bfc1af65 | 1494 | unsigned from, to; |
4631dbf6 | 1495 | int size_changed = 0; |
ac27a0ec | 1496 | |
9bffad1e | 1497 | trace_ext4_journalled_write_end(inode, pos, len, copied); |
09cbfeaf | 1498 | from = pos & (PAGE_SIZE - 1); |
bfc1af65 NP |
1499 | to = from + len; |
1500 | ||
441c8508 CW |
1501 | BUG_ON(!ext4_handle_valid(handle)); |
1502 | ||
eb5efbcb TT |
1503 | if (ext4_has_inline_data(inode)) { |
1504 | ret = ext4_write_inline_data_end(inode, pos, len, | |
1505 | copied, page); | |
1506 | if (ret < 0) { | |
1507 | unlock_page(page); | |
1508 | put_page(page); | |
1509 | goto errout; | |
1510 | } | |
1511 | copied = ret; | |
1512 | } else if (unlikely(copied < len) && !PageUptodate(page)) { | |
3b136499 JK |
1513 | copied = 0; |
1514 | ext4_journalled_zero_new_buffers(handle, page, from, to); | |
1515 | } else { | |
1516 | if (unlikely(copied < len)) | |
1517 | ext4_journalled_zero_new_buffers(handle, page, | |
1518 | from + copied, to); | |
3fdcfb66 | 1519 | ret = ext4_walk_page_buffers(handle, page_buffers(page), from, |
3b136499 JK |
1520 | from + copied, &partial, |
1521 | write_end_fn); | |
3fdcfb66 TM |
1522 | if (!partial) |
1523 | SetPageUptodate(page); | |
1524 | } | |
4631dbf6 | 1525 | size_changed = ext4_update_inode_size(inode, pos + copied); |
19f5fb7a | 1526 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
2d859db3 | 1527 | EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; |
4631dbf6 | 1528 | unlock_page(page); |
09cbfeaf | 1529 | put_page(page); |
4631dbf6 | 1530 | |
0572639f XW |
1531 | if (old_size < pos) |
1532 | pagecache_isize_extended(inode, old_size, pos); | |
1533 | ||
4631dbf6 | 1534 | if (size_changed) { |
617ba13b | 1535 | ret2 = ext4_mark_inode_dirty(handle, inode); |
ac27a0ec DK |
1536 | if (!ret) |
1537 | ret = ret2; | |
1538 | } | |
bfc1af65 | 1539 | |
ffacfa7a | 1540 | if (pos + len > inode->i_size && ext4_can_truncate(inode)) |
f8514083 AK |
1541 | /* if we have allocated more blocks and copied |
1542 | * less. We will have blocks allocated outside | |
1543 | * inode->i_size. So truncate them | |
1544 | */ | |
1545 | ext4_orphan_add(handle, inode); | |
1546 | ||
eb5efbcb | 1547 | errout: |
617ba13b | 1548 | ret2 = ext4_journal_stop(handle); |
ac27a0ec DK |
1549 | if (!ret) |
1550 | ret = ret2; | |
f8514083 | 1551 | if (pos + len > inode->i_size) { |
b9a4207d | 1552 | ext4_truncate_failed_write(inode); |
de9a55b8 | 1553 | /* |
ffacfa7a | 1554 | * If truncate failed early the inode might still be |
f8514083 AK |
1555 | * on the orphan list; we need to make sure the inode |
1556 | * is removed from the orphan list in that case. | |
1557 | */ | |
1558 | if (inode->i_nlink) | |
1559 | ext4_orphan_del(NULL, inode); | |
1560 | } | |
bfc1af65 NP |
1561 | |
1562 | return ret ? ret : copied; | |
ac27a0ec | 1563 | } |
d2a17637 | 1564 | |
9d0be502 | 1565 | /* |
c27e43a1 | 1566 | * Reserve space for a single cluster |
9d0be502 | 1567 | */ |
c27e43a1 | 1568 | static int ext4_da_reserve_space(struct inode *inode) |
d2a17637 | 1569 | { |
60e58e0f | 1570 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
0637c6f4 | 1571 | struct ext4_inode_info *ei = EXT4_I(inode); |
5dd4056d | 1572 | int ret; |
03179fe9 TT |
1573 | |
1574 | /* | |
1575 | * We will charge metadata quota at writeout time; this saves | |
1576 | * us from metadata over-estimation, though we may go over by | |
1577 | * a small amount in the end. Here we just reserve for data. | |
1578 | */ | |
1579 | ret = dquot_reserve_block(inode, EXT4_C2B(sbi, 1)); | |
1580 | if (ret) | |
1581 | return ret; | |
d2a17637 | 1582 | |
0637c6f4 | 1583 | spin_lock(&ei->i_block_reservation_lock); |
71d4f7d0 | 1584 | if (ext4_claim_free_clusters(sbi, 1, 0)) { |
03179fe9 | 1585 | spin_unlock(&ei->i_block_reservation_lock); |
03179fe9 | 1586 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, 1)); |
d2a17637 MC |
1587 | return -ENOSPC; |
1588 | } | |
9d0be502 | 1589 | ei->i_reserved_data_blocks++; |
c27e43a1 | 1590 | trace_ext4_da_reserve_space(inode); |
0637c6f4 | 1591 | spin_unlock(&ei->i_block_reservation_lock); |
39bc680a | 1592 | |
d2a17637 MC |
1593 | return 0; /* success */ |
1594 | } | |
1595 | ||
12219aea | 1596 | static void ext4_da_release_space(struct inode *inode, int to_free) |
d2a17637 MC |
1597 | { |
1598 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
0637c6f4 | 1599 | struct ext4_inode_info *ei = EXT4_I(inode); |
d2a17637 | 1600 | |
cd213226 MC |
1601 | if (!to_free) |
1602 | return; /* Nothing to release, exit */ | |
1603 | ||
d2a17637 | 1604 | spin_lock(&EXT4_I(inode)->i_block_reservation_lock); |
cd213226 | 1605 | |
5a58ec87 | 1606 | trace_ext4_da_release_space(inode, to_free); |
0637c6f4 | 1607 | if (unlikely(to_free > ei->i_reserved_data_blocks)) { |
cd213226 | 1608 | /* |
0637c6f4 TT |
1609 | * if there aren't enough reserved blocks, then the |
1610 | * counter is messed up somewhere. Since this | |
1611 | * function is called from invalidate page, it's | |
1612 | * harmless to return without any action. | |
cd213226 | 1613 | */ |
8de5c325 | 1614 | ext4_warning(inode->i_sb, "ext4_da_release_space: " |
0637c6f4 | 1615 | "ino %lu, to_free %d with only %d reserved " |
1084f252 | 1616 | "data blocks", inode->i_ino, to_free, |
0637c6f4 TT |
1617 | ei->i_reserved_data_blocks); |
1618 | WARN_ON(1); | |
1619 | to_free = ei->i_reserved_data_blocks; | |
cd213226 | 1620 | } |
0637c6f4 | 1621 | ei->i_reserved_data_blocks -= to_free; |
cd213226 | 1622 | |
72b8ab9d | 1623 | /* update fs dirty data blocks counter */ |
57042651 | 1624 | percpu_counter_sub(&sbi->s_dirtyclusters_counter, to_free); |
d2a17637 | 1625 | |
d2a17637 | 1626 | spin_unlock(&EXT4_I(inode)->i_block_reservation_lock); |
60e58e0f | 1627 | |
7b415bf6 | 1628 | dquot_release_reservation_block(inode, EXT4_C2B(sbi, to_free)); |
d2a17637 MC |
1629 | } |
1630 | ||
1631 | static void ext4_da_page_release_reservation(struct page *page, | |
ca99fdd2 LC |
1632 | unsigned int offset, |
1633 | unsigned int length) | |
d2a17637 | 1634 | { |
9705acd6 | 1635 | int to_release = 0, contiguous_blks = 0; |
d2a17637 MC |
1636 | struct buffer_head *head, *bh; |
1637 | unsigned int curr_off = 0; | |
7b415bf6 AK |
1638 | struct inode *inode = page->mapping->host; |
1639 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
ca99fdd2 | 1640 | unsigned int stop = offset + length; |
7b415bf6 | 1641 | int num_clusters; |
51865fda | 1642 | ext4_fsblk_t lblk; |
d2a17637 | 1643 | |
09cbfeaf | 1644 | BUG_ON(stop > PAGE_SIZE || stop < length); |
ca99fdd2 | 1645 | |
d2a17637 MC |
1646 | head = page_buffers(page); |
1647 | bh = head; | |
1648 | do { | |
1649 | unsigned int next_off = curr_off + bh->b_size; | |
1650 | ||
ca99fdd2 LC |
1651 | if (next_off > stop) |
1652 | break; | |
1653 | ||
d2a17637 MC |
1654 | if ((offset <= curr_off) && (buffer_delay(bh))) { |
1655 | to_release++; | |
9705acd6 | 1656 | contiguous_blks++; |
d2a17637 | 1657 | clear_buffer_delay(bh); |
9705acd6 LC |
1658 | } else if (contiguous_blks) { |
1659 | lblk = page->index << | |
09cbfeaf | 1660 | (PAGE_SHIFT - inode->i_blkbits); |
9705acd6 LC |
1661 | lblk += (curr_off >> inode->i_blkbits) - |
1662 | contiguous_blks; | |
1663 | ext4_es_remove_extent(inode, lblk, contiguous_blks); | |
1664 | contiguous_blks = 0; | |
d2a17637 MC |
1665 | } |
1666 | curr_off = next_off; | |
1667 | } while ((bh = bh->b_this_page) != head); | |
7b415bf6 | 1668 | |
9705acd6 | 1669 | if (contiguous_blks) { |
09cbfeaf | 1670 | lblk = page->index << (PAGE_SHIFT - inode->i_blkbits); |
9705acd6 LC |
1671 | lblk += (curr_off >> inode->i_blkbits) - contiguous_blks; |
1672 | ext4_es_remove_extent(inode, lblk, contiguous_blks); | |
51865fda ZL |
1673 | } |
1674 | ||
7b415bf6 AK |
1675 | /* If we have released all the blocks belonging to a cluster, then we |
1676 | * need to release the reserved space for that cluster. */ | |
1677 | num_clusters = EXT4_NUM_B2C(sbi, to_release); | |
1678 | while (num_clusters > 0) { | |
09cbfeaf | 1679 | lblk = (page->index << (PAGE_SHIFT - inode->i_blkbits)) + |
7b415bf6 AK |
1680 | ((num_clusters - 1) << sbi->s_cluster_bits); |
1681 | if (sbi->s_cluster_ratio == 1 || | |
7d1b1fbc | 1682 | !ext4_find_delalloc_cluster(inode, lblk)) |
7b415bf6 AK |
1683 | ext4_da_release_space(inode, 1); |
1684 | ||
1685 | num_clusters--; | |
1686 | } | |
d2a17637 | 1687 | } |
ac27a0ec | 1688 | |
64769240 AT |
1689 | /* |
1690 | * Delayed allocation stuff | |
1691 | */ | |
1692 | ||
4e7ea81d JK |
1693 | struct mpage_da_data { |
1694 | struct inode *inode; | |
1695 | struct writeback_control *wbc; | |
6b523df4 | 1696 | |
4e7ea81d JK |
1697 | pgoff_t first_page; /* The first page to write */ |
1698 | pgoff_t next_page; /* Current page to examine */ | |
1699 | pgoff_t last_page; /* Last page to examine */ | |
791b7f08 | 1700 | /* |
4e7ea81d JK |
1701 | * Extent to map - this can be after first_page because that can be |
1702 | * fully mapped. We somewhat abuse m_flags to store whether the extent | |
1703 | * is delalloc or unwritten. | |
791b7f08 | 1704 | */ |
4e7ea81d JK |
1705 | struct ext4_map_blocks map; |
1706 | struct ext4_io_submit io_submit; /* IO submission data */ | |
dddbd6ac | 1707 | unsigned int do_map:1; |
4e7ea81d | 1708 | }; |
64769240 | 1709 | |
4e7ea81d JK |
1710 | static void mpage_release_unused_pages(struct mpage_da_data *mpd, |
1711 | bool invalidate) | |
c4a0c46e AK |
1712 | { |
1713 | int nr_pages, i; | |
1714 | pgoff_t index, end; | |
1715 | struct pagevec pvec; | |
1716 | struct inode *inode = mpd->inode; | |
1717 | struct address_space *mapping = inode->i_mapping; | |
4e7ea81d JK |
1718 | |
1719 | /* This is necessary when next_page == 0. */ | |
1720 | if (mpd->first_page >= mpd->next_page) | |
1721 | return; | |
c4a0c46e | 1722 | |
c7f5938a CW |
1723 | index = mpd->first_page; |
1724 | end = mpd->next_page - 1; | |
4e7ea81d JK |
1725 | if (invalidate) { |
1726 | ext4_lblk_t start, last; | |
09cbfeaf KS |
1727 | start = index << (PAGE_SHIFT - inode->i_blkbits); |
1728 | last = end << (PAGE_SHIFT - inode->i_blkbits); | |
4e7ea81d JK |
1729 | ext4_es_remove_extent(inode, start, last - start + 1); |
1730 | } | |
51865fda | 1731 | |
86679820 | 1732 | pagevec_init(&pvec); |
c4a0c46e | 1733 | while (index <= end) { |
397162ff | 1734 | nr_pages = pagevec_lookup_range(&pvec, mapping, &index, end); |
c4a0c46e AK |
1735 | if (nr_pages == 0) |
1736 | break; | |
1737 | for (i = 0; i < nr_pages; i++) { | |
1738 | struct page *page = pvec.pages[i]; | |
2b85a617 | 1739 | |
c4a0c46e AK |
1740 | BUG_ON(!PageLocked(page)); |
1741 | BUG_ON(PageWriteback(page)); | |
4e7ea81d | 1742 | if (invalidate) { |
4e800c03 | 1743 | if (page_mapped(page)) |
1744 | clear_page_dirty_for_io(page); | |
09cbfeaf | 1745 | block_invalidatepage(page, 0, PAGE_SIZE); |
4e7ea81d JK |
1746 | ClearPageUptodate(page); |
1747 | } | |
c4a0c46e AK |
1748 | unlock_page(page); |
1749 | } | |
9b1d0998 | 1750 | pagevec_release(&pvec); |
c4a0c46e | 1751 | } |
c4a0c46e AK |
1752 | } |
1753 | ||
df22291f AK |
1754 | static void ext4_print_free_blocks(struct inode *inode) |
1755 | { | |
1756 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); | |
92b97816 | 1757 | struct super_block *sb = inode->i_sb; |
f78ee70d | 1758 | struct ext4_inode_info *ei = EXT4_I(inode); |
92b97816 TT |
1759 | |
1760 | ext4_msg(sb, KERN_CRIT, "Total free blocks count %lld", | |
5dee5437 | 1761 | EXT4_C2B(EXT4_SB(inode->i_sb), |
f78ee70d | 1762 | ext4_count_free_clusters(sb))); |
92b97816 TT |
1763 | ext4_msg(sb, KERN_CRIT, "Free/Dirty block details"); |
1764 | ext4_msg(sb, KERN_CRIT, "free_blocks=%lld", | |
f78ee70d | 1765 | (long long) EXT4_C2B(EXT4_SB(sb), |
57042651 | 1766 | percpu_counter_sum(&sbi->s_freeclusters_counter))); |
92b97816 | 1767 | ext4_msg(sb, KERN_CRIT, "dirty_blocks=%lld", |
f78ee70d | 1768 | (long long) EXT4_C2B(EXT4_SB(sb), |
7b415bf6 | 1769 | percpu_counter_sum(&sbi->s_dirtyclusters_counter))); |
92b97816 TT |
1770 | ext4_msg(sb, KERN_CRIT, "Block reservation details"); |
1771 | ext4_msg(sb, KERN_CRIT, "i_reserved_data_blocks=%u", | |
f78ee70d | 1772 | ei->i_reserved_data_blocks); |
df22291f AK |
1773 | return; |
1774 | } | |
1775 | ||
c364b22c | 1776 | static int ext4_bh_delay_or_unwritten(handle_t *handle, struct buffer_head *bh) |
29fa89d0 | 1777 | { |
c364b22c | 1778 | return (buffer_delay(bh) || buffer_unwritten(bh)) && buffer_dirty(bh); |
29fa89d0 AK |
1779 | } |
1780 | ||
5356f261 AK |
1781 | /* |
1782 | * This function is grabs code from the very beginning of | |
1783 | * ext4_map_blocks, but assumes that the caller is from delayed write | |
1784 | * time. This function looks up the requested blocks and sets the | |
1785 | * buffer delay bit under the protection of i_data_sem. | |
1786 | */ | |
1787 | static int ext4_da_map_blocks(struct inode *inode, sector_t iblock, | |
1788 | struct ext4_map_blocks *map, | |
1789 | struct buffer_head *bh) | |
1790 | { | |
d100eef2 | 1791 | struct extent_status es; |
5356f261 AK |
1792 | int retval; |
1793 | sector_t invalid_block = ~((sector_t) 0xffff); | |
921f266b DM |
1794 | #ifdef ES_AGGRESSIVE_TEST |
1795 | struct ext4_map_blocks orig_map; | |
1796 | ||
1797 | memcpy(&orig_map, map, sizeof(*map)); | |
1798 | #endif | |
5356f261 AK |
1799 | |
1800 | if (invalid_block < ext4_blocks_count(EXT4_SB(inode->i_sb)->s_es)) | |
1801 | invalid_block = ~0; | |
1802 | ||
1803 | map->m_flags = 0; | |
1804 | ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u," | |
1805 | "logical block %lu\n", inode->i_ino, map->m_len, | |
1806 | (unsigned long) map->m_lblk); | |
d100eef2 ZL |
1807 | |
1808 | /* Lookup extent status tree firstly */ | |
1809 | if (ext4_es_lookup_extent(inode, iblock, &es)) { | |
d100eef2 ZL |
1810 | if (ext4_es_is_hole(&es)) { |
1811 | retval = 0; | |
c8b459f4 | 1812 | down_read(&EXT4_I(inode)->i_data_sem); |
d100eef2 ZL |
1813 | goto add_delayed; |
1814 | } | |
1815 | ||
1816 | /* | |
1817 | * Delayed extent could be allocated by fallocate. | |
1818 | * So we need to check it. | |
1819 | */ | |
1820 | if (ext4_es_is_delayed(&es) && !ext4_es_is_unwritten(&es)) { | |
1821 | map_bh(bh, inode->i_sb, invalid_block); | |
1822 | set_buffer_new(bh); | |
1823 | set_buffer_delay(bh); | |
1824 | return 0; | |
1825 | } | |
1826 | ||
1827 | map->m_pblk = ext4_es_pblock(&es) + iblock - es.es_lblk; | |
1828 | retval = es.es_len - (iblock - es.es_lblk); | |
1829 | if (retval > map->m_len) | |
1830 | retval = map->m_len; | |
1831 | map->m_len = retval; | |
1832 | if (ext4_es_is_written(&es)) | |
1833 | map->m_flags |= EXT4_MAP_MAPPED; | |
1834 | else if (ext4_es_is_unwritten(&es)) | |
1835 | map->m_flags |= EXT4_MAP_UNWRITTEN; | |
1836 | else | |
1837 | BUG_ON(1); | |
1838 | ||
921f266b DM |
1839 | #ifdef ES_AGGRESSIVE_TEST |
1840 | ext4_map_blocks_es_recheck(NULL, inode, map, &orig_map, 0); | |
1841 | #endif | |
d100eef2 ZL |
1842 | return retval; |
1843 | } | |
1844 | ||
5356f261 AK |
1845 | /* |
1846 | * Try to see if we can get the block without requesting a new | |
1847 | * file system block. | |
1848 | */ | |
c8b459f4 | 1849 | down_read(&EXT4_I(inode)->i_data_sem); |
cbd7584e | 1850 | if (ext4_has_inline_data(inode)) |
9c3569b5 | 1851 | retval = 0; |
cbd7584e | 1852 | else if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
2f8e0a7c | 1853 | retval = ext4_ext_map_blocks(NULL, inode, map, 0); |
5356f261 | 1854 | else |
2f8e0a7c | 1855 | retval = ext4_ind_map_blocks(NULL, inode, map, 0); |
5356f261 | 1856 | |
d100eef2 | 1857 | add_delayed: |
5356f261 | 1858 | if (retval == 0) { |
f7fec032 | 1859 | int ret; |
5356f261 AK |
1860 | /* |
1861 | * XXX: __block_prepare_write() unmaps passed block, | |
1862 | * is it OK? | |
1863 | */ | |
386ad67c LC |
1864 | /* |
1865 | * If the block was allocated from previously allocated cluster, | |
1866 | * then we don't need to reserve it again. However we still need | |
1867 | * to reserve metadata for every block we're going to write. | |
1868 | */ | |
c27e43a1 | 1869 | if (EXT4_SB(inode->i_sb)->s_cluster_ratio == 1 || |
cbd7584e | 1870 | !ext4_find_delalloc_cluster(inode, map->m_lblk)) { |
c27e43a1 | 1871 | ret = ext4_da_reserve_space(inode); |
f7fec032 | 1872 | if (ret) { |
5356f261 | 1873 | /* not enough space to reserve */ |
f7fec032 | 1874 | retval = ret; |
5356f261 | 1875 | goto out_unlock; |
f7fec032 | 1876 | } |
5356f261 AK |
1877 | } |
1878 | ||
f7fec032 ZL |
1879 | ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, |
1880 | ~0, EXTENT_STATUS_DELAYED); | |
1881 | if (ret) { | |
1882 | retval = ret; | |
51865fda | 1883 | goto out_unlock; |
f7fec032 | 1884 | } |
51865fda | 1885 | |
5356f261 AK |
1886 | map_bh(bh, inode->i_sb, invalid_block); |
1887 | set_buffer_new(bh); | |
1888 | set_buffer_delay(bh); | |
f7fec032 ZL |
1889 | } else if (retval > 0) { |
1890 | int ret; | |
3be78c73 | 1891 | unsigned int status; |
f7fec032 | 1892 | |
44fb851d ZL |
1893 | if (unlikely(retval != map->m_len)) { |
1894 | ext4_warning(inode->i_sb, | |
1895 | "ES len assertion failed for inode " | |
1896 | "%lu: retval %d != map->m_len %d", | |
1897 | inode->i_ino, retval, map->m_len); | |
1898 | WARN_ON(1); | |
921f266b | 1899 | } |
921f266b | 1900 | |
f7fec032 ZL |
1901 | status = map->m_flags & EXT4_MAP_UNWRITTEN ? |
1902 | EXTENT_STATUS_UNWRITTEN : EXTENT_STATUS_WRITTEN; | |
1903 | ret = ext4_es_insert_extent(inode, map->m_lblk, map->m_len, | |
1904 | map->m_pblk, status); | |
1905 | if (ret != 0) | |
1906 | retval = ret; | |
5356f261 AK |
1907 | } |
1908 | ||
1909 | out_unlock: | |
1910 | up_read((&EXT4_I(inode)->i_data_sem)); | |
1911 | ||
1912 | return retval; | |
1913 | } | |
1914 | ||
64769240 | 1915 | /* |
d91bd2c1 | 1916 | * This is a special get_block_t callback which is used by |
b920c755 TT |
1917 | * ext4_da_write_begin(). It will either return mapped block or |
1918 | * reserve space for a single block. | |
29fa89d0 AK |
1919 | * |
1920 | * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set. | |
1921 | * We also have b_blocknr = -1 and b_bdev initialized properly | |
1922 | * | |
1923 | * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set. | |
1924 | * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev | |
1925 | * initialized properly. | |
64769240 | 1926 | */ |
9c3569b5 TM |
1927 | int ext4_da_get_block_prep(struct inode *inode, sector_t iblock, |
1928 | struct buffer_head *bh, int create) | |
64769240 | 1929 | { |
2ed88685 | 1930 | struct ext4_map_blocks map; |
64769240 AT |
1931 | int ret = 0; |
1932 | ||
1933 | BUG_ON(create == 0); | |
2ed88685 TT |
1934 | BUG_ON(bh->b_size != inode->i_sb->s_blocksize); |
1935 | ||
1936 | map.m_lblk = iblock; | |
1937 | map.m_len = 1; | |
64769240 AT |
1938 | |
1939 | /* | |
1940 | * first, we need to know whether the block is allocated already | |
1941 | * preallocated blocks are unmapped but should treated | |
1942 | * the same as allocated blocks. | |
1943 | */ | |
5356f261 AK |
1944 | ret = ext4_da_map_blocks(inode, iblock, &map, bh); |
1945 | if (ret <= 0) | |
2ed88685 | 1946 | return ret; |
64769240 | 1947 | |
2ed88685 | 1948 | map_bh(bh, inode->i_sb, map.m_pblk); |
ed8ad838 | 1949 | ext4_update_bh_state(bh, map.m_flags); |
2ed88685 TT |
1950 | |
1951 | if (buffer_unwritten(bh)) { | |
1952 | /* A delayed write to unwritten bh should be marked | |
1953 | * new and mapped. Mapped ensures that we don't do | |
1954 | * get_block multiple times when we write to the same | |
1955 | * offset and new ensures that we do proper zero out | |
1956 | * for partial write. | |
1957 | */ | |
1958 | set_buffer_new(bh); | |
c8205636 | 1959 | set_buffer_mapped(bh); |
2ed88685 TT |
1960 | } |
1961 | return 0; | |
64769240 | 1962 | } |
61628a3f | 1963 | |
62e086be AK |
1964 | static int bget_one(handle_t *handle, struct buffer_head *bh) |
1965 | { | |
1966 | get_bh(bh); | |
1967 | return 0; | |
1968 | } | |
1969 | ||
1970 | static int bput_one(handle_t *handle, struct buffer_head *bh) | |
1971 | { | |
1972 | put_bh(bh); | |
1973 | return 0; | |
1974 | } | |
1975 | ||
1976 | static int __ext4_journalled_writepage(struct page *page, | |
62e086be AK |
1977 | unsigned int len) |
1978 | { | |
1979 | struct address_space *mapping = page->mapping; | |
1980 | struct inode *inode = mapping->host; | |
3fdcfb66 | 1981 | struct buffer_head *page_bufs = NULL; |
62e086be | 1982 | handle_t *handle = NULL; |
3fdcfb66 TM |
1983 | int ret = 0, err = 0; |
1984 | int inline_data = ext4_has_inline_data(inode); | |
1985 | struct buffer_head *inode_bh = NULL; | |
62e086be | 1986 | |
cb20d518 | 1987 | ClearPageChecked(page); |
3fdcfb66 TM |
1988 | |
1989 | if (inline_data) { | |
1990 | BUG_ON(page->index != 0); | |
1991 | BUG_ON(len > ext4_get_max_inline_size(inode)); | |
1992 | inode_bh = ext4_journalled_write_inline_data(inode, len, page); | |
1993 | if (inode_bh == NULL) | |
1994 | goto out; | |
1995 | } else { | |
1996 | page_bufs = page_buffers(page); | |
1997 | if (!page_bufs) { | |
1998 | BUG(); | |
1999 | goto out; | |
2000 | } | |
2001 | ext4_walk_page_buffers(handle, page_bufs, 0, len, | |
2002 | NULL, bget_one); | |
2003 | } | |
bdf96838 TT |
2004 | /* |
2005 | * We need to release the page lock before we start the | |
2006 | * journal, so grab a reference so the page won't disappear | |
2007 | * out from under us. | |
2008 | */ | |
2009 | get_page(page); | |
62e086be AK |
2010 | unlock_page(page); |
2011 | ||
9924a92a TT |
2012 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
2013 | ext4_writepage_trans_blocks(inode)); | |
62e086be AK |
2014 | if (IS_ERR(handle)) { |
2015 | ret = PTR_ERR(handle); | |
bdf96838 TT |
2016 | put_page(page); |
2017 | goto out_no_pagelock; | |
62e086be | 2018 | } |
441c8508 CW |
2019 | BUG_ON(!ext4_handle_valid(handle)); |
2020 | ||
bdf96838 TT |
2021 | lock_page(page); |
2022 | put_page(page); | |
2023 | if (page->mapping != mapping) { | |
2024 | /* The page got truncated from under us */ | |
2025 | ext4_journal_stop(handle); | |
2026 | ret = 0; | |
2027 | goto out; | |
2028 | } | |
2029 | ||
3fdcfb66 | 2030 | if (inline_data) { |
5d601255 | 2031 | BUFFER_TRACE(inode_bh, "get write access"); |
3fdcfb66 | 2032 | ret = ext4_journal_get_write_access(handle, inode_bh); |
62e086be | 2033 | |
3fdcfb66 TM |
2034 | err = ext4_handle_dirty_metadata(handle, inode, inode_bh); |
2035 | ||
2036 | } else { | |
2037 | ret = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL, | |
2038 | do_journal_get_write_access); | |
2039 | ||
2040 | err = ext4_walk_page_buffers(handle, page_bufs, 0, len, NULL, | |
2041 | write_end_fn); | |
2042 | } | |
62e086be AK |
2043 | if (ret == 0) |
2044 | ret = err; | |
2d859db3 | 2045 | EXT4_I(inode)->i_datasync_tid = handle->h_transaction->t_tid; |
62e086be AK |
2046 | err = ext4_journal_stop(handle); |
2047 | if (!ret) | |
2048 | ret = err; | |
2049 | ||
3fdcfb66 | 2050 | if (!ext4_has_inline_data(inode)) |
8c9367fd | 2051 | ext4_walk_page_buffers(NULL, page_bufs, 0, len, |
3fdcfb66 | 2052 | NULL, bput_one); |
19f5fb7a | 2053 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); |
62e086be | 2054 | out: |
bdf96838 TT |
2055 | unlock_page(page); |
2056 | out_no_pagelock: | |
3fdcfb66 | 2057 | brelse(inode_bh); |
62e086be AK |
2058 | return ret; |
2059 | } | |
2060 | ||
61628a3f | 2061 | /* |
43ce1d23 AK |
2062 | * Note that we don't need to start a transaction unless we're journaling data |
2063 | * because we should have holes filled from ext4_page_mkwrite(). We even don't | |
2064 | * need to file the inode to the transaction's list in ordered mode because if | |
2065 | * we are writing back data added by write(), the inode is already there and if | |
25985edc | 2066 | * we are writing back data modified via mmap(), no one guarantees in which |
43ce1d23 AK |
2067 | * transaction the data will hit the disk. In case we are journaling data, we |
2068 | * cannot start transaction directly because transaction start ranks above page | |
2069 | * lock so we have to do some magic. | |
2070 | * | |
b920c755 | 2071 | * This function can get called via... |
20970ba6 | 2072 | * - ext4_writepages after taking page lock (have journal handle) |
b920c755 | 2073 | * - journal_submit_inode_data_buffers (no journal handle) |
f6463b0d | 2074 | * - shrink_page_list via the kswapd/direct reclaim (no journal handle) |
b920c755 | 2075 | * - grab_page_cache when doing write_begin (have journal handle) |
43ce1d23 AK |
2076 | * |
2077 | * We don't do any block allocation in this function. If we have page with | |
2078 | * multiple blocks we need to write those buffer_heads that are mapped. This | |
2079 | * is important for mmaped based write. So if we do with blocksize 1K | |
2080 | * truncate(f, 1024); | |
2081 | * a = mmap(f, 0, 4096); | |
2082 | * a[0] = 'a'; | |
2083 | * truncate(f, 4096); | |
2084 | * we have in the page first buffer_head mapped via page_mkwrite call back | |
90802ed9 | 2085 | * but other buffer_heads would be unmapped but dirty (dirty done via the |
43ce1d23 AK |
2086 | * do_wp_page). So writepage should write the first block. If we modify |
2087 | * the mmap area beyond 1024 we will again get a page_fault and the | |
2088 | * page_mkwrite callback will do the block allocation and mark the | |
2089 | * buffer_heads mapped. | |
2090 | * | |
2091 | * We redirty the page if we have any buffer_heads that is either delay or | |
2092 | * unwritten in the page. | |
2093 | * | |
2094 | * We can get recursively called as show below. | |
2095 | * | |
2096 | * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() -> | |
2097 | * ext4_writepage() | |
2098 | * | |
2099 | * But since we don't do any block allocation we should not deadlock. | |
2100 | * Page also have the dirty flag cleared so we don't get recurive page_lock. | |
61628a3f | 2101 | */ |
43ce1d23 | 2102 | static int ext4_writepage(struct page *page, |
62e086be | 2103 | struct writeback_control *wbc) |
64769240 | 2104 | { |
f8bec370 | 2105 | int ret = 0; |
61628a3f | 2106 | loff_t size; |
498e5f24 | 2107 | unsigned int len; |
744692dc | 2108 | struct buffer_head *page_bufs = NULL; |
61628a3f | 2109 | struct inode *inode = page->mapping->host; |
36ade451 | 2110 | struct ext4_io_submit io_submit; |
1c8349a1 | 2111 | bool keep_towrite = false; |
61628a3f | 2112 | |
0db1ff22 TT |
2113 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) { |
2114 | ext4_invalidatepage(page, 0, PAGE_SIZE); | |
2115 | unlock_page(page); | |
2116 | return -EIO; | |
2117 | } | |
2118 | ||
a9c667f8 | 2119 | trace_ext4_writepage(page); |
f0e6c985 | 2120 | size = i_size_read(inode); |
09cbfeaf KS |
2121 | if (page->index == size >> PAGE_SHIFT) |
2122 | len = size & ~PAGE_MASK; | |
f0e6c985 | 2123 | else |
09cbfeaf | 2124 | len = PAGE_SIZE; |
64769240 | 2125 | |
a42afc5f | 2126 | page_bufs = page_buffers(page); |
a42afc5f | 2127 | /* |
fe386132 JK |
2128 | * We cannot do block allocation or other extent handling in this |
2129 | * function. If there are buffers needing that, we have to redirty | |
2130 | * the page. But we may reach here when we do a journal commit via | |
2131 | * journal_submit_inode_data_buffers() and in that case we must write | |
2132 | * allocated buffers to achieve data=ordered mode guarantees. | |
cccd147a TT |
2133 | * |
2134 | * Also, if there is only one buffer per page (the fs block | |
2135 | * size == the page size), if one buffer needs block | |
2136 | * allocation or needs to modify the extent tree to clear the | |
2137 | * unwritten flag, we know that the page can't be written at | |
2138 | * all, so we might as well refuse the write immediately. | |
2139 | * Unfortunately if the block size != page size, we can't as | |
2140 | * easily detect this case using ext4_walk_page_buffers(), but | |
2141 | * for the extremely common case, this is an optimization that | |
2142 | * skips a useless round trip through ext4_bio_write_page(). | |
a42afc5f | 2143 | */ |
f19d5870 TM |
2144 | if (ext4_walk_page_buffers(NULL, page_bufs, 0, len, NULL, |
2145 | ext4_bh_delay_or_unwritten)) { | |
f8bec370 | 2146 | redirty_page_for_writepage(wbc, page); |
cccd147a | 2147 | if ((current->flags & PF_MEMALLOC) || |
09cbfeaf | 2148 | (inode->i_sb->s_blocksize == PAGE_SIZE)) { |
fe386132 JK |
2149 | /* |
2150 | * For memory cleaning there's no point in writing only | |
2151 | * some buffers. So just bail out. Warn if we came here | |
2152 | * from direct reclaim. | |
2153 | */ | |
2154 | WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) | |
2155 | == PF_MEMALLOC); | |
f0e6c985 AK |
2156 | unlock_page(page); |
2157 | return 0; | |
2158 | } | |
1c8349a1 | 2159 | keep_towrite = true; |
a42afc5f | 2160 | } |
64769240 | 2161 | |
cb20d518 | 2162 | if (PageChecked(page) && ext4_should_journal_data(inode)) |
43ce1d23 AK |
2163 | /* |
2164 | * It's mmapped pagecache. Add buffers and journal it. There | |
2165 | * doesn't seem much point in redirtying the page here. | |
2166 | */ | |
3f0ca309 | 2167 | return __ext4_journalled_writepage(page, len); |
43ce1d23 | 2168 | |
97a851ed JK |
2169 | ext4_io_submit_init(&io_submit, wbc); |
2170 | io_submit.io_end = ext4_init_io_end(inode, GFP_NOFS); | |
2171 | if (!io_submit.io_end) { | |
2172 | redirty_page_for_writepage(wbc, page); | |
2173 | unlock_page(page); | |
2174 | return -ENOMEM; | |
2175 | } | |
1c8349a1 | 2176 | ret = ext4_bio_write_page(&io_submit, page, len, wbc, keep_towrite); |
36ade451 | 2177 | ext4_io_submit(&io_submit); |
97a851ed JK |
2178 | /* Drop io_end reference we got from init */ |
2179 | ext4_put_io_end_defer(io_submit.io_end); | |
64769240 AT |
2180 | return ret; |
2181 | } | |
2182 | ||
5f1132b2 JK |
2183 | static int mpage_submit_page(struct mpage_da_data *mpd, struct page *page) |
2184 | { | |
2185 | int len; | |
a056bdaa | 2186 | loff_t size; |
5f1132b2 JK |
2187 | int err; |
2188 | ||
2189 | BUG_ON(page->index != mpd->first_page); | |
a056bdaa JK |
2190 | clear_page_dirty_for_io(page); |
2191 | /* | |
2192 | * We have to be very careful here! Nothing protects writeback path | |
2193 | * against i_size changes and the page can be writeably mapped into | |
2194 | * page tables. So an application can be growing i_size and writing | |
2195 | * data through mmap while writeback runs. clear_page_dirty_for_io() | |
2196 | * write-protects our page in page tables and the page cannot get | |
2197 | * written to again until we release page lock. So only after | |
2198 | * clear_page_dirty_for_io() we are safe to sample i_size for | |
2199 | * ext4_bio_write_page() to zero-out tail of the written page. We rely | |
2200 | * on the barrier provided by TestClearPageDirty in | |
2201 | * clear_page_dirty_for_io() to make sure i_size is really sampled only | |
2202 | * after page tables are updated. | |
2203 | */ | |
2204 | size = i_size_read(mpd->inode); | |
09cbfeaf KS |
2205 | if (page->index == size >> PAGE_SHIFT) |
2206 | len = size & ~PAGE_MASK; | |
5f1132b2 | 2207 | else |
09cbfeaf | 2208 | len = PAGE_SIZE; |
1c8349a1 | 2209 | err = ext4_bio_write_page(&mpd->io_submit, page, len, mpd->wbc, false); |
5f1132b2 JK |
2210 | if (!err) |
2211 | mpd->wbc->nr_to_write--; | |
2212 | mpd->first_page++; | |
2213 | ||
2214 | return err; | |
2215 | } | |
2216 | ||
4e7ea81d JK |
2217 | #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay)) |
2218 | ||
61628a3f | 2219 | /* |
fffb2739 JK |
2220 | * mballoc gives us at most this number of blocks... |
2221 | * XXX: That seems to be only a limitation of ext4_mb_normalize_request(). | |
70261f56 | 2222 | * The rest of mballoc seems to handle chunks up to full group size. |
61628a3f | 2223 | */ |
fffb2739 | 2224 | #define MAX_WRITEPAGES_EXTENT_LEN 2048 |
525f4ed8 | 2225 | |
4e7ea81d JK |
2226 | /* |
2227 | * mpage_add_bh_to_extent - try to add bh to extent of blocks to map | |
2228 | * | |
2229 | * @mpd - extent of blocks | |
2230 | * @lblk - logical number of the block in the file | |
09930042 | 2231 | * @bh - buffer head we want to add to the extent |
4e7ea81d | 2232 | * |
09930042 JK |
2233 | * The function is used to collect contig. blocks in the same state. If the |
2234 | * buffer doesn't require mapping for writeback and we haven't started the | |
2235 | * extent of buffers to map yet, the function returns 'true' immediately - the | |
2236 | * caller can write the buffer right away. Otherwise the function returns true | |
2237 | * if the block has been added to the extent, false if the block couldn't be | |
2238 | * added. | |
4e7ea81d | 2239 | */ |
09930042 JK |
2240 | static bool mpage_add_bh_to_extent(struct mpage_da_data *mpd, ext4_lblk_t lblk, |
2241 | struct buffer_head *bh) | |
4e7ea81d JK |
2242 | { |
2243 | struct ext4_map_blocks *map = &mpd->map; | |
2244 | ||
09930042 JK |
2245 | /* Buffer that doesn't need mapping for writeback? */ |
2246 | if (!buffer_dirty(bh) || !buffer_mapped(bh) || | |
2247 | (!buffer_delay(bh) && !buffer_unwritten(bh))) { | |
2248 | /* So far no extent to map => we write the buffer right away */ | |
2249 | if (map->m_len == 0) | |
2250 | return true; | |
2251 | return false; | |
2252 | } | |
4e7ea81d JK |
2253 | |
2254 | /* First block in the extent? */ | |
2255 | if (map->m_len == 0) { | |
dddbd6ac JK |
2256 | /* We cannot map unless handle is started... */ |
2257 | if (!mpd->do_map) | |
2258 | return false; | |
4e7ea81d JK |
2259 | map->m_lblk = lblk; |
2260 | map->m_len = 1; | |
09930042 JK |
2261 | map->m_flags = bh->b_state & BH_FLAGS; |
2262 | return true; | |
4e7ea81d JK |
2263 | } |
2264 | ||
09930042 JK |
2265 | /* Don't go larger than mballoc is willing to allocate */ |
2266 | if (map->m_len >= MAX_WRITEPAGES_EXTENT_LEN) | |
2267 | return false; | |
2268 | ||
4e7ea81d JK |
2269 | /* Can we merge the block to our big extent? */ |
2270 | if (lblk == map->m_lblk + map->m_len && | |
09930042 | 2271 | (bh->b_state & BH_FLAGS) == map->m_flags) { |
4e7ea81d | 2272 | map->m_len++; |
09930042 | 2273 | return true; |
4e7ea81d | 2274 | } |
09930042 | 2275 | return false; |
4e7ea81d JK |
2276 | } |
2277 | ||
5f1132b2 JK |
2278 | /* |
2279 | * mpage_process_page_bufs - submit page buffers for IO or add them to extent | |
2280 | * | |
2281 | * @mpd - extent of blocks for mapping | |
2282 | * @head - the first buffer in the page | |
2283 | * @bh - buffer we should start processing from | |
2284 | * @lblk - logical number of the block in the file corresponding to @bh | |
2285 | * | |
2286 | * Walk through page buffers from @bh upto @head (exclusive) and either submit | |
2287 | * the page for IO if all buffers in this page were mapped and there's no | |
2288 | * accumulated extent of buffers to map or add buffers in the page to the | |
2289 | * extent of buffers to map. The function returns 1 if the caller can continue | |
2290 | * by processing the next page, 0 if it should stop adding buffers to the | |
2291 | * extent to map because we cannot extend it anymore. It can also return value | |
2292 | * < 0 in case of error during IO submission. | |
2293 | */ | |
2294 | static int mpage_process_page_bufs(struct mpage_da_data *mpd, | |
2295 | struct buffer_head *head, | |
2296 | struct buffer_head *bh, | |
2297 | ext4_lblk_t lblk) | |
4e7ea81d JK |
2298 | { |
2299 | struct inode *inode = mpd->inode; | |
5f1132b2 | 2300 | int err; |
93407472 | 2301 | ext4_lblk_t blocks = (i_size_read(inode) + i_blocksize(inode) - 1) |
4e7ea81d JK |
2302 | >> inode->i_blkbits; |
2303 | ||
2304 | do { | |
2305 | BUG_ON(buffer_locked(bh)); | |
2306 | ||
09930042 | 2307 | if (lblk >= blocks || !mpage_add_bh_to_extent(mpd, lblk, bh)) { |
4e7ea81d JK |
2308 | /* Found extent to map? */ |
2309 | if (mpd->map.m_len) | |
5f1132b2 | 2310 | return 0; |
dddbd6ac JK |
2311 | /* Buffer needs mapping and handle is not started? */ |
2312 | if (!mpd->do_map) | |
2313 | return 0; | |
09930042 | 2314 | /* Everything mapped so far and we hit EOF */ |
5f1132b2 | 2315 | break; |
4e7ea81d | 2316 | } |
4e7ea81d | 2317 | } while (lblk++, (bh = bh->b_this_page) != head); |
5f1132b2 JK |
2318 | /* So far everything mapped? Submit the page for IO. */ |
2319 | if (mpd->map.m_len == 0) { | |
2320 | err = mpage_submit_page(mpd, head->b_page); | |
2321 | if (err < 0) | |
2322 | return err; | |
2323 | } | |
2324 | return lblk < blocks; | |
4e7ea81d JK |
2325 | } |
2326 | ||
2327 | /* | |
2328 | * mpage_map_buffers - update buffers corresponding to changed extent and | |
2329 | * submit fully mapped pages for IO | |
2330 | * | |
2331 | * @mpd - description of extent to map, on return next extent to map | |
2332 | * | |
2333 | * Scan buffers corresponding to changed extent (we expect corresponding pages | |
2334 | * to be already locked) and update buffer state according to new extent state. | |
2335 | * We map delalloc buffers to their physical location, clear unwritten bits, | |
556615dc | 2336 | * and mark buffers as uninit when we perform writes to unwritten extents |
4e7ea81d JK |
2337 | * and do extent conversion after IO is finished. If the last page is not fully |
2338 | * mapped, we update @map to the next extent in the last page that needs | |
2339 | * mapping. Otherwise we submit the page for IO. | |
2340 | */ | |
2341 | static int mpage_map_and_submit_buffers(struct mpage_da_data *mpd) | |
2342 | { | |
2343 | struct pagevec pvec; | |
2344 | int nr_pages, i; | |
2345 | struct inode *inode = mpd->inode; | |
2346 | struct buffer_head *head, *bh; | |
09cbfeaf | 2347 | int bpp_bits = PAGE_SHIFT - inode->i_blkbits; |
4e7ea81d JK |
2348 | pgoff_t start, end; |
2349 | ext4_lblk_t lblk; | |
2350 | sector_t pblock; | |
2351 | int err; | |
2352 | ||
2353 | start = mpd->map.m_lblk >> bpp_bits; | |
2354 | end = (mpd->map.m_lblk + mpd->map.m_len - 1) >> bpp_bits; | |
2355 | lblk = start << bpp_bits; | |
2356 | pblock = mpd->map.m_pblk; | |
2357 | ||
86679820 | 2358 | pagevec_init(&pvec); |
4e7ea81d | 2359 | while (start <= end) { |
2b85a617 | 2360 | nr_pages = pagevec_lookup_range(&pvec, inode->i_mapping, |
397162ff | 2361 | &start, end); |
4e7ea81d JK |
2362 | if (nr_pages == 0) |
2363 | break; | |
2364 | for (i = 0; i < nr_pages; i++) { | |
2365 | struct page *page = pvec.pages[i]; | |
2366 | ||
4e7ea81d JK |
2367 | bh = head = page_buffers(page); |
2368 | do { | |
2369 | if (lblk < mpd->map.m_lblk) | |
2370 | continue; | |
2371 | if (lblk >= mpd->map.m_lblk + mpd->map.m_len) { | |
2372 | /* | |
2373 | * Buffer after end of mapped extent. | |
2374 | * Find next buffer in the page to map. | |
2375 | */ | |
2376 | mpd->map.m_len = 0; | |
2377 | mpd->map.m_flags = 0; | |
5f1132b2 JK |
2378 | /* |
2379 | * FIXME: If dioread_nolock supports | |
2380 | * blocksize < pagesize, we need to make | |
2381 | * sure we add size mapped so far to | |
2382 | * io_end->size as the following call | |
2383 | * can submit the page for IO. | |
2384 | */ | |
2385 | err = mpage_process_page_bufs(mpd, head, | |
2386 | bh, lblk); | |
4e7ea81d | 2387 | pagevec_release(&pvec); |
5f1132b2 JK |
2388 | if (err > 0) |
2389 | err = 0; | |
2390 | return err; | |
4e7ea81d JK |
2391 | } |
2392 | if (buffer_delay(bh)) { | |
2393 | clear_buffer_delay(bh); | |
2394 | bh->b_blocknr = pblock++; | |
2395 | } | |
4e7ea81d | 2396 | clear_buffer_unwritten(bh); |
5f1132b2 | 2397 | } while (lblk++, (bh = bh->b_this_page) != head); |
4e7ea81d JK |
2398 | |
2399 | /* | |
2400 | * FIXME: This is going to break if dioread_nolock | |
2401 | * supports blocksize < pagesize as we will try to | |
2402 | * convert potentially unmapped parts of inode. | |
2403 | */ | |
09cbfeaf | 2404 | mpd->io_submit.io_end->size += PAGE_SIZE; |
4e7ea81d JK |
2405 | /* Page fully mapped - let IO run! */ |
2406 | err = mpage_submit_page(mpd, page); | |
2407 | if (err < 0) { | |
2408 | pagevec_release(&pvec); | |
2409 | return err; | |
2410 | } | |
4e7ea81d JK |
2411 | } |
2412 | pagevec_release(&pvec); | |
2413 | } | |
2414 | /* Extent fully mapped and matches with page boundary. We are done. */ | |
2415 | mpd->map.m_len = 0; | |
2416 | mpd->map.m_flags = 0; | |
2417 | return 0; | |
2418 | } | |
2419 | ||
2420 | static int mpage_map_one_extent(handle_t *handle, struct mpage_da_data *mpd) | |
2421 | { | |
2422 | struct inode *inode = mpd->inode; | |
2423 | struct ext4_map_blocks *map = &mpd->map; | |
2424 | int get_blocks_flags; | |
090f32ee | 2425 | int err, dioread_nolock; |
4e7ea81d JK |
2426 | |
2427 | trace_ext4_da_write_pages_extent(inode, map); | |
2428 | /* | |
2429 | * Call ext4_map_blocks() to allocate any delayed allocation blocks, or | |
556615dc | 2430 | * to convert an unwritten extent to be initialized (in the case |
4e7ea81d JK |
2431 | * where we have written into one or more preallocated blocks). It is |
2432 | * possible that we're going to need more metadata blocks than | |
2433 | * previously reserved. However we must not fail because we're in | |
2434 | * writeback and there is nothing we can do about it so it might result | |
2435 | * in data loss. So use reserved blocks to allocate metadata if | |
2436 | * possible. | |
2437 | * | |
754cfed6 TT |
2438 | * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if |
2439 | * the blocks in question are delalloc blocks. This indicates | |
2440 | * that the blocks and quotas has already been checked when | |
2441 | * the data was copied into the page cache. | |
4e7ea81d JK |
2442 | */ |
2443 | get_blocks_flags = EXT4_GET_BLOCKS_CREATE | | |
ee0876bc JK |
2444 | EXT4_GET_BLOCKS_METADATA_NOFAIL | |
2445 | EXT4_GET_BLOCKS_IO_SUBMIT; | |
090f32ee LC |
2446 | dioread_nolock = ext4_should_dioread_nolock(inode); |
2447 | if (dioread_nolock) | |
4e7ea81d JK |
2448 | get_blocks_flags |= EXT4_GET_BLOCKS_IO_CREATE_EXT; |
2449 | if (map->m_flags & (1 << BH_Delay)) | |
2450 | get_blocks_flags |= EXT4_GET_BLOCKS_DELALLOC_RESERVE; | |
2451 | ||
2452 | err = ext4_map_blocks(handle, inode, map, get_blocks_flags); | |
2453 | if (err < 0) | |
2454 | return err; | |
090f32ee | 2455 | if (dioread_nolock && (map->m_flags & EXT4_MAP_UNWRITTEN)) { |
6b523df4 JK |
2456 | if (!mpd->io_submit.io_end->handle && |
2457 | ext4_handle_valid(handle)) { | |
2458 | mpd->io_submit.io_end->handle = handle->h_rsv_handle; | |
2459 | handle->h_rsv_handle = NULL; | |
2460 | } | |
3613d228 | 2461 | ext4_set_io_unwritten_flag(inode, mpd->io_submit.io_end); |
6b523df4 | 2462 | } |
4e7ea81d JK |
2463 | |
2464 | BUG_ON(map->m_len == 0); | |
2465 | if (map->m_flags & EXT4_MAP_NEW) { | |
64e1c57f JK |
2466 | clean_bdev_aliases(inode->i_sb->s_bdev, map->m_pblk, |
2467 | map->m_len); | |
4e7ea81d JK |
2468 | } |
2469 | return 0; | |
2470 | } | |
2471 | ||
2472 | /* | |
2473 | * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length | |
2474 | * mpd->len and submit pages underlying it for IO | |
2475 | * | |
2476 | * @handle - handle for journal operations | |
2477 | * @mpd - extent to map | |
7534e854 JK |
2478 | * @give_up_on_write - we set this to true iff there is a fatal error and there |
2479 | * is no hope of writing the data. The caller should discard | |
2480 | * dirty pages to avoid infinite loops. | |
4e7ea81d JK |
2481 | * |
2482 | * The function maps extent starting at mpd->lblk of length mpd->len. If it is | |
2483 | * delayed, blocks are allocated, if it is unwritten, we may need to convert | |
2484 | * them to initialized or split the described range from larger unwritten | |
2485 | * extent. Note that we need not map all the described range since allocation | |
2486 | * can return less blocks or the range is covered by more unwritten extents. We | |
2487 | * cannot map more because we are limited by reserved transaction credits. On | |
2488 | * the other hand we always make sure that the last touched page is fully | |
2489 | * mapped so that it can be written out (and thus forward progress is | |
2490 | * guaranteed). After mapping we submit all mapped pages for IO. | |
2491 | */ | |
2492 | static int mpage_map_and_submit_extent(handle_t *handle, | |
cb530541 TT |
2493 | struct mpage_da_data *mpd, |
2494 | bool *give_up_on_write) | |
4e7ea81d JK |
2495 | { |
2496 | struct inode *inode = mpd->inode; | |
2497 | struct ext4_map_blocks *map = &mpd->map; | |
2498 | int err; | |
2499 | loff_t disksize; | |
6603120e | 2500 | int progress = 0; |
4e7ea81d JK |
2501 | |
2502 | mpd->io_submit.io_end->offset = | |
2503 | ((loff_t)map->m_lblk) << inode->i_blkbits; | |
27d7c4ed | 2504 | do { |
4e7ea81d JK |
2505 | err = mpage_map_one_extent(handle, mpd); |
2506 | if (err < 0) { | |
2507 | struct super_block *sb = inode->i_sb; | |
2508 | ||
0db1ff22 TT |
2509 | if (ext4_forced_shutdown(EXT4_SB(sb)) || |
2510 | EXT4_SB(sb)->s_mount_flags & EXT4_MF_FS_ABORTED) | |
cb530541 | 2511 | goto invalidate_dirty_pages; |
4e7ea81d | 2512 | /* |
cb530541 TT |
2513 | * Let the uper layers retry transient errors. |
2514 | * In the case of ENOSPC, if ext4_count_free_blocks() | |
2515 | * is non-zero, a commit should free up blocks. | |
4e7ea81d | 2516 | */ |
cb530541 | 2517 | if ((err == -ENOMEM) || |
6603120e DM |
2518 | (err == -ENOSPC && ext4_count_free_clusters(sb))) { |
2519 | if (progress) | |
2520 | goto update_disksize; | |
cb530541 | 2521 | return err; |
6603120e | 2522 | } |
cb530541 TT |
2523 | ext4_msg(sb, KERN_CRIT, |
2524 | "Delayed block allocation failed for " | |
2525 | "inode %lu at logical offset %llu with" | |
2526 | " max blocks %u with error %d", | |
2527 | inode->i_ino, | |
2528 | (unsigned long long)map->m_lblk, | |
2529 | (unsigned)map->m_len, -err); | |
2530 | ext4_msg(sb, KERN_CRIT, | |
2531 | "This should not happen!! Data will " | |
2532 | "be lost\n"); | |
2533 | if (err == -ENOSPC) | |
2534 | ext4_print_free_blocks(inode); | |
2535 | invalidate_dirty_pages: | |
2536 | *give_up_on_write = true; | |
4e7ea81d JK |
2537 | return err; |
2538 | } | |
6603120e | 2539 | progress = 1; |
4e7ea81d JK |
2540 | /* |
2541 | * Update buffer state, submit mapped pages, and get us new | |
2542 | * extent to map | |
2543 | */ | |
2544 | err = mpage_map_and_submit_buffers(mpd); | |
2545 | if (err < 0) | |
6603120e | 2546 | goto update_disksize; |
27d7c4ed | 2547 | } while (map->m_len); |
4e7ea81d | 2548 | |
6603120e | 2549 | update_disksize: |
622cad13 TT |
2550 | /* |
2551 | * Update on-disk size after IO is submitted. Races with | |
2552 | * truncate are avoided by checking i_size under i_data_sem. | |
2553 | */ | |
09cbfeaf | 2554 | disksize = ((loff_t)mpd->first_page) << PAGE_SHIFT; |
4e7ea81d JK |
2555 | if (disksize > EXT4_I(inode)->i_disksize) { |
2556 | int err2; | |
622cad13 TT |
2557 | loff_t i_size; |
2558 | ||
2559 | down_write(&EXT4_I(inode)->i_data_sem); | |
2560 | i_size = i_size_read(inode); | |
2561 | if (disksize > i_size) | |
2562 | disksize = i_size; | |
2563 | if (disksize > EXT4_I(inode)->i_disksize) | |
2564 | EXT4_I(inode)->i_disksize = disksize; | |
622cad13 | 2565 | up_write(&EXT4_I(inode)->i_data_sem); |
b907f2d5 | 2566 | err2 = ext4_mark_inode_dirty(handle, inode); |
4e7ea81d JK |
2567 | if (err2) |
2568 | ext4_error(inode->i_sb, | |
2569 | "Failed to mark inode %lu dirty", | |
2570 | inode->i_ino); | |
2571 | if (!err) | |
2572 | err = err2; | |
2573 | } | |
2574 | return err; | |
2575 | } | |
2576 | ||
fffb2739 JK |
2577 | /* |
2578 | * Calculate the total number of credits to reserve for one writepages | |
20970ba6 | 2579 | * iteration. This is called from ext4_writepages(). We map an extent of |
70261f56 | 2580 | * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping |
fffb2739 JK |
2581 | * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN + |
2582 | * bpp - 1 blocks in bpp different extents. | |
2583 | */ | |
525f4ed8 MC |
2584 | static int ext4_da_writepages_trans_blocks(struct inode *inode) |
2585 | { | |
fffb2739 | 2586 | int bpp = ext4_journal_blocks_per_page(inode); |
525f4ed8 | 2587 | |
fffb2739 JK |
2588 | return ext4_meta_trans_blocks(inode, |
2589 | MAX_WRITEPAGES_EXTENT_LEN + bpp - 1, bpp); | |
525f4ed8 | 2590 | } |
61628a3f | 2591 | |
8e48dcfb | 2592 | /* |
4e7ea81d JK |
2593 | * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages |
2594 | * and underlying extent to map | |
2595 | * | |
2596 | * @mpd - where to look for pages | |
2597 | * | |
2598 | * Walk dirty pages in the mapping. If they are fully mapped, submit them for | |
2599 | * IO immediately. When we find a page which isn't mapped we start accumulating | |
2600 | * extent of buffers underlying these pages that needs mapping (formed by | |
2601 | * either delayed or unwritten buffers). We also lock the pages containing | |
2602 | * these buffers. The extent found is returned in @mpd structure (starting at | |
2603 | * mpd->lblk with length mpd->len blocks). | |
2604 | * | |
2605 | * Note that this function can attach bios to one io_end structure which are | |
2606 | * neither logically nor physically contiguous. Although it may seem as an | |
2607 | * unnecessary complication, it is actually inevitable in blocksize < pagesize | |
2608 | * case as we need to track IO to all buffers underlying a page in one io_end. | |
8e48dcfb | 2609 | */ |
4e7ea81d | 2610 | static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd) |
8e48dcfb | 2611 | { |
4e7ea81d JK |
2612 | struct address_space *mapping = mpd->inode->i_mapping; |
2613 | struct pagevec pvec; | |
2614 | unsigned int nr_pages; | |
aeac589a | 2615 | long left = mpd->wbc->nr_to_write; |
4e7ea81d JK |
2616 | pgoff_t index = mpd->first_page; |
2617 | pgoff_t end = mpd->last_page; | |
2618 | int tag; | |
2619 | int i, err = 0; | |
2620 | int blkbits = mpd->inode->i_blkbits; | |
2621 | ext4_lblk_t lblk; | |
2622 | struct buffer_head *head; | |
8e48dcfb | 2623 | |
4e7ea81d | 2624 | if (mpd->wbc->sync_mode == WB_SYNC_ALL || mpd->wbc->tagged_writepages) |
5b41d924 ES |
2625 | tag = PAGECACHE_TAG_TOWRITE; |
2626 | else | |
2627 | tag = PAGECACHE_TAG_DIRTY; | |
2628 | ||
86679820 | 2629 | pagevec_init(&pvec); |
4e7ea81d JK |
2630 | mpd->map.m_len = 0; |
2631 | mpd->next_page = index; | |
4f01b02c | 2632 | while (index <= end) { |
dc7f3e86 | 2633 | nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end, |
67fd707f | 2634 | tag); |
8e48dcfb | 2635 | if (nr_pages == 0) |
4e7ea81d | 2636 | goto out; |
8e48dcfb TT |
2637 | |
2638 | for (i = 0; i < nr_pages; i++) { | |
2639 | struct page *page = pvec.pages[i]; | |
2640 | ||
aeac589a ML |
2641 | /* |
2642 | * Accumulated enough dirty pages? This doesn't apply | |
2643 | * to WB_SYNC_ALL mode. For integrity sync we have to | |
2644 | * keep going because someone may be concurrently | |
2645 | * dirtying pages, and we might have synced a lot of | |
2646 | * newly appeared dirty pages, but have not synced all | |
2647 | * of the old dirty pages. | |
2648 | */ | |
2649 | if (mpd->wbc->sync_mode == WB_SYNC_NONE && left <= 0) | |
2650 | goto out; | |
2651 | ||
4e7ea81d JK |
2652 | /* If we can't merge this page, we are done. */ |
2653 | if (mpd->map.m_len > 0 && mpd->next_page != page->index) | |
2654 | goto out; | |
78aaced3 | 2655 | |
8e48dcfb | 2656 | lock_page(page); |
8e48dcfb | 2657 | /* |
4e7ea81d JK |
2658 | * If the page is no longer dirty, or its mapping no |
2659 | * longer corresponds to inode we are writing (which | |
2660 | * means it has been truncated or invalidated), or the | |
2661 | * page is already under writeback and we are not doing | |
2662 | * a data integrity writeback, skip the page | |
8e48dcfb | 2663 | */ |
4f01b02c TT |
2664 | if (!PageDirty(page) || |
2665 | (PageWriteback(page) && | |
4e7ea81d | 2666 | (mpd->wbc->sync_mode == WB_SYNC_NONE)) || |
4f01b02c | 2667 | unlikely(page->mapping != mapping)) { |
8e48dcfb TT |
2668 | unlock_page(page); |
2669 | continue; | |
2670 | } | |
2671 | ||
7cb1a535 | 2672 | wait_on_page_writeback(page); |
8e48dcfb | 2673 | BUG_ON(PageWriteback(page)); |
8e48dcfb | 2674 | |
4e7ea81d | 2675 | if (mpd->map.m_len == 0) |
8eb9e5ce | 2676 | mpd->first_page = page->index; |
8eb9e5ce | 2677 | mpd->next_page = page->index + 1; |
f8bec370 | 2678 | /* Add all dirty buffers to mpd */ |
4e7ea81d | 2679 | lblk = ((ext4_lblk_t)page->index) << |
09cbfeaf | 2680 | (PAGE_SHIFT - blkbits); |
f8bec370 | 2681 | head = page_buffers(page); |
5f1132b2 JK |
2682 | err = mpage_process_page_bufs(mpd, head, head, lblk); |
2683 | if (err <= 0) | |
4e7ea81d | 2684 | goto out; |
5f1132b2 | 2685 | err = 0; |
aeac589a | 2686 | left--; |
8e48dcfb TT |
2687 | } |
2688 | pagevec_release(&pvec); | |
2689 | cond_resched(); | |
2690 | } | |
4f01b02c | 2691 | return 0; |
8eb9e5ce TT |
2692 | out: |
2693 | pagevec_release(&pvec); | |
4e7ea81d | 2694 | return err; |
8e48dcfb TT |
2695 | } |
2696 | ||
20970ba6 TT |
2697 | static int ext4_writepages(struct address_space *mapping, |
2698 | struct writeback_control *wbc) | |
64769240 | 2699 | { |
4e7ea81d JK |
2700 | pgoff_t writeback_index = 0; |
2701 | long nr_to_write = wbc->nr_to_write; | |
22208ded | 2702 | int range_whole = 0; |
4e7ea81d | 2703 | int cycled = 1; |
61628a3f | 2704 | handle_t *handle = NULL; |
df22291f | 2705 | struct mpage_da_data mpd; |
5e745b04 | 2706 | struct inode *inode = mapping->host; |
6b523df4 | 2707 | int needed_blocks, rsv_blocks = 0, ret = 0; |
5e745b04 | 2708 | struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb); |
4e7ea81d | 2709 | bool done; |
1bce63d1 | 2710 | struct blk_plug plug; |
cb530541 | 2711 | bool give_up_on_write = false; |
61628a3f | 2712 | |
0db1ff22 TT |
2713 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
2714 | return -EIO; | |
2715 | ||
c8585c6f | 2716 | percpu_down_read(&sbi->s_journal_flag_rwsem); |
20970ba6 | 2717 | trace_ext4_writepages(inode, wbc); |
ba80b101 | 2718 | |
61628a3f MC |
2719 | /* |
2720 | * No pages to write? This is mainly a kludge to avoid starting | |
2721 | * a transaction for special inodes like journal inode on last iput() | |
2722 | * because that could violate lock ordering on umount | |
2723 | */ | |
a1d6cc56 | 2724 | if (!mapping->nrpages || !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) |
bbf023c7 | 2725 | goto out_writepages; |
2a21e37e | 2726 | |
20970ba6 | 2727 | if (ext4_should_journal_data(inode)) { |
043d20d1 | 2728 | ret = generic_writepages(mapping, wbc); |
bbf023c7 | 2729 | goto out_writepages; |
20970ba6 TT |
2730 | } |
2731 | ||
2a21e37e TT |
2732 | /* |
2733 | * If the filesystem has aborted, it is read-only, so return | |
2734 | * right away instead of dumping stack traces later on that | |
2735 | * will obscure the real source of the problem. We test | |
1751e8a6 | 2736 | * EXT4_MF_FS_ABORTED instead of sb->s_flag's SB_RDONLY because |
2a21e37e | 2737 | * the latter could be true if the filesystem is mounted |
20970ba6 | 2738 | * read-only, and in that case, ext4_writepages should |
2a21e37e TT |
2739 | * *never* be called, so if that ever happens, we would want |
2740 | * the stack trace. | |
2741 | */ | |
0db1ff22 TT |
2742 | if (unlikely(ext4_forced_shutdown(EXT4_SB(mapping->host->i_sb)) || |
2743 | sbi->s_mount_flags & EXT4_MF_FS_ABORTED)) { | |
bbf023c7 ML |
2744 | ret = -EROFS; |
2745 | goto out_writepages; | |
2746 | } | |
2a21e37e | 2747 | |
6b523df4 JK |
2748 | if (ext4_should_dioread_nolock(inode)) { |
2749 | /* | |
70261f56 | 2750 | * We may need to convert up to one extent per block in |
6b523df4 JK |
2751 | * the page and we may dirty the inode. |
2752 | */ | |
09cbfeaf | 2753 | rsv_blocks = 1 + (PAGE_SIZE >> inode->i_blkbits); |
6b523df4 JK |
2754 | } |
2755 | ||
4e7ea81d JK |
2756 | /* |
2757 | * If we have inline data and arrive here, it means that | |
2758 | * we will soon create the block for the 1st page, so | |
2759 | * we'd better clear the inline data here. | |
2760 | */ | |
2761 | if (ext4_has_inline_data(inode)) { | |
2762 | /* Just inode will be modified... */ | |
2763 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 1); | |
2764 | if (IS_ERR(handle)) { | |
2765 | ret = PTR_ERR(handle); | |
2766 | goto out_writepages; | |
2767 | } | |
2768 | BUG_ON(ext4_test_inode_state(inode, | |
2769 | EXT4_STATE_MAY_INLINE_DATA)); | |
2770 | ext4_destroy_inline_data(handle, inode); | |
2771 | ext4_journal_stop(handle); | |
2772 | } | |
2773 | ||
22208ded AK |
2774 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
2775 | range_whole = 1; | |
61628a3f | 2776 | |
2acf2c26 | 2777 | if (wbc->range_cyclic) { |
4e7ea81d JK |
2778 | writeback_index = mapping->writeback_index; |
2779 | if (writeback_index) | |
2acf2c26 | 2780 | cycled = 0; |
4e7ea81d JK |
2781 | mpd.first_page = writeback_index; |
2782 | mpd.last_page = -1; | |
5b41d924 | 2783 | } else { |
09cbfeaf KS |
2784 | mpd.first_page = wbc->range_start >> PAGE_SHIFT; |
2785 | mpd.last_page = wbc->range_end >> PAGE_SHIFT; | |
5b41d924 | 2786 | } |
a1d6cc56 | 2787 | |
4e7ea81d JK |
2788 | mpd.inode = inode; |
2789 | mpd.wbc = wbc; | |
2790 | ext4_io_submit_init(&mpd.io_submit, wbc); | |
2acf2c26 | 2791 | retry: |
6e6938b6 | 2792 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
4e7ea81d JK |
2793 | tag_pages_for_writeback(mapping, mpd.first_page, mpd.last_page); |
2794 | done = false; | |
1bce63d1 | 2795 | blk_start_plug(&plug); |
dddbd6ac JK |
2796 | |
2797 | /* | |
2798 | * First writeback pages that don't need mapping - we can avoid | |
2799 | * starting a transaction unnecessarily and also avoid being blocked | |
2800 | * in the block layer on device congestion while having transaction | |
2801 | * started. | |
2802 | */ | |
2803 | mpd.do_map = 0; | |
2804 | mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL); | |
2805 | if (!mpd.io_submit.io_end) { | |
2806 | ret = -ENOMEM; | |
2807 | goto unplug; | |
2808 | } | |
2809 | ret = mpage_prepare_extent_to_map(&mpd); | |
2810 | /* Submit prepared bio */ | |
2811 | ext4_io_submit(&mpd.io_submit); | |
2812 | ext4_put_io_end_defer(mpd.io_submit.io_end); | |
2813 | mpd.io_submit.io_end = NULL; | |
2814 | /* Unlock pages we didn't use */ | |
2815 | mpage_release_unused_pages(&mpd, false); | |
2816 | if (ret < 0) | |
2817 | goto unplug; | |
2818 | ||
4e7ea81d JK |
2819 | while (!done && mpd.first_page <= mpd.last_page) { |
2820 | /* For each extent of pages we use new io_end */ | |
2821 | mpd.io_submit.io_end = ext4_init_io_end(inode, GFP_KERNEL); | |
2822 | if (!mpd.io_submit.io_end) { | |
2823 | ret = -ENOMEM; | |
2824 | break; | |
2825 | } | |
a1d6cc56 AK |
2826 | |
2827 | /* | |
4e7ea81d JK |
2828 | * We have two constraints: We find one extent to map and we |
2829 | * must always write out whole page (makes a difference when | |
2830 | * blocksize < pagesize) so that we don't block on IO when we | |
2831 | * try to write out the rest of the page. Journalled mode is | |
2832 | * not supported by delalloc. | |
a1d6cc56 AK |
2833 | */ |
2834 | BUG_ON(ext4_should_journal_data(inode)); | |
525f4ed8 | 2835 | needed_blocks = ext4_da_writepages_trans_blocks(inode); |
a1d6cc56 | 2836 | |
4e7ea81d | 2837 | /* start a new transaction */ |
6b523df4 JK |
2838 | handle = ext4_journal_start_with_reserve(inode, |
2839 | EXT4_HT_WRITE_PAGE, needed_blocks, rsv_blocks); | |
61628a3f MC |
2840 | if (IS_ERR(handle)) { |
2841 | ret = PTR_ERR(handle); | |
1693918e | 2842 | ext4_msg(inode->i_sb, KERN_CRIT, "%s: jbd2_start: " |
fbe845dd | 2843 | "%ld pages, ino %lu; err %d", __func__, |
a1d6cc56 | 2844 | wbc->nr_to_write, inode->i_ino, ret); |
4e7ea81d JK |
2845 | /* Release allocated io_end */ |
2846 | ext4_put_io_end(mpd.io_submit.io_end); | |
dddbd6ac | 2847 | mpd.io_submit.io_end = NULL; |
4e7ea81d | 2848 | break; |
61628a3f | 2849 | } |
dddbd6ac | 2850 | mpd.do_map = 1; |
f63e6005 | 2851 | |
4e7ea81d JK |
2852 | trace_ext4_da_write_pages(inode, mpd.first_page, mpd.wbc); |
2853 | ret = mpage_prepare_extent_to_map(&mpd); | |
2854 | if (!ret) { | |
2855 | if (mpd.map.m_len) | |
cb530541 TT |
2856 | ret = mpage_map_and_submit_extent(handle, &mpd, |
2857 | &give_up_on_write); | |
4e7ea81d JK |
2858 | else { |
2859 | /* | |
2860 | * We scanned the whole range (or exhausted | |
2861 | * nr_to_write), submitted what was mapped and | |
2862 | * didn't find anything needing mapping. We are | |
2863 | * done. | |
2864 | */ | |
2865 | done = true; | |
2866 | } | |
f63e6005 | 2867 | } |
646caa9c JK |
2868 | /* |
2869 | * Caution: If the handle is synchronous, | |
2870 | * ext4_journal_stop() can wait for transaction commit | |
2871 | * to finish which may depend on writeback of pages to | |
2872 | * complete or on page lock to be released. In that | |
2873 | * case, we have to wait until after after we have | |
2874 | * submitted all the IO, released page locks we hold, | |
2875 | * and dropped io_end reference (for extent conversion | |
2876 | * to be able to complete) before stopping the handle. | |
2877 | */ | |
2878 | if (!ext4_handle_valid(handle) || handle->h_sync == 0) { | |
2879 | ext4_journal_stop(handle); | |
2880 | handle = NULL; | |
dddbd6ac | 2881 | mpd.do_map = 0; |
646caa9c | 2882 | } |
4e7ea81d JK |
2883 | /* Submit prepared bio */ |
2884 | ext4_io_submit(&mpd.io_submit); | |
2885 | /* Unlock pages we didn't use */ | |
cb530541 | 2886 | mpage_release_unused_pages(&mpd, give_up_on_write); |
646caa9c JK |
2887 | /* |
2888 | * Drop our io_end reference we got from init. We have | |
2889 | * to be careful and use deferred io_end finishing if | |
2890 | * we are still holding the transaction as we can | |
2891 | * release the last reference to io_end which may end | |
2892 | * up doing unwritten extent conversion. | |
2893 | */ | |
2894 | if (handle) { | |
2895 | ext4_put_io_end_defer(mpd.io_submit.io_end); | |
2896 | ext4_journal_stop(handle); | |
2897 | } else | |
2898 | ext4_put_io_end(mpd.io_submit.io_end); | |
dddbd6ac | 2899 | mpd.io_submit.io_end = NULL; |
4e7ea81d JK |
2900 | |
2901 | if (ret == -ENOSPC && sbi->s_journal) { | |
2902 | /* | |
2903 | * Commit the transaction which would | |
22208ded AK |
2904 | * free blocks released in the transaction |
2905 | * and try again | |
2906 | */ | |
df22291f | 2907 | jbd2_journal_force_commit_nested(sbi->s_journal); |
22208ded | 2908 | ret = 0; |
4e7ea81d JK |
2909 | continue; |
2910 | } | |
2911 | /* Fatal error - ENOMEM, EIO... */ | |
2912 | if (ret) | |
61628a3f | 2913 | break; |
a1d6cc56 | 2914 | } |
dddbd6ac | 2915 | unplug: |
1bce63d1 | 2916 | blk_finish_plug(&plug); |
9c12a831 | 2917 | if (!ret && !cycled && wbc->nr_to_write > 0) { |
2acf2c26 | 2918 | cycled = 1; |
4e7ea81d JK |
2919 | mpd.last_page = writeback_index - 1; |
2920 | mpd.first_page = 0; | |
2acf2c26 AK |
2921 | goto retry; |
2922 | } | |
22208ded AK |
2923 | |
2924 | /* Update index */ | |
22208ded AK |
2925 | if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) |
2926 | /* | |
4e7ea81d | 2927 | * Set the writeback_index so that range_cyclic |
22208ded AK |
2928 | * mode will write it back later |
2929 | */ | |
4e7ea81d | 2930 | mapping->writeback_index = mpd.first_page; |
a1d6cc56 | 2931 | |
61628a3f | 2932 | out_writepages: |
20970ba6 TT |
2933 | trace_ext4_writepages_result(inode, wbc, ret, |
2934 | nr_to_write - wbc->nr_to_write); | |
c8585c6f | 2935 | percpu_up_read(&sbi->s_journal_flag_rwsem); |
61628a3f | 2936 | return ret; |
64769240 AT |
2937 | } |
2938 | ||
5f0663bb DW |
2939 | static int ext4_dax_writepages(struct address_space *mapping, |
2940 | struct writeback_control *wbc) | |
2941 | { | |
2942 | int ret; | |
2943 | long nr_to_write = wbc->nr_to_write; | |
2944 | struct inode *inode = mapping->host; | |
2945 | struct ext4_sb_info *sbi = EXT4_SB(mapping->host->i_sb); | |
2946 | ||
2947 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) | |
2948 | return -EIO; | |
2949 | ||
2950 | percpu_down_read(&sbi->s_journal_flag_rwsem); | |
2951 | trace_ext4_writepages(inode, wbc); | |
2952 | ||
2953 | ret = dax_writeback_mapping_range(mapping, inode->i_sb->s_bdev, wbc); | |
2954 | trace_ext4_writepages_result(inode, wbc, ret, | |
2955 | nr_to_write - wbc->nr_to_write); | |
2956 | percpu_up_read(&sbi->s_journal_flag_rwsem); | |
2957 | return ret; | |
2958 | } | |
2959 | ||
79f0be8d AK |
2960 | static int ext4_nonda_switch(struct super_block *sb) |
2961 | { | |
5c1ff336 | 2962 | s64 free_clusters, dirty_clusters; |
79f0be8d AK |
2963 | struct ext4_sb_info *sbi = EXT4_SB(sb); |
2964 | ||
2965 | /* | |
2966 | * switch to non delalloc mode if we are running low | |
2967 | * on free block. The free block accounting via percpu | |
179f7ebf | 2968 | * counters can get slightly wrong with percpu_counter_batch getting |
79f0be8d AK |
2969 | * accumulated on each CPU without updating global counters |
2970 | * Delalloc need an accurate free block accounting. So switch | |
2971 | * to non delalloc when we are near to error range. | |
2972 | */ | |
5c1ff336 EW |
2973 | free_clusters = |
2974 | percpu_counter_read_positive(&sbi->s_freeclusters_counter); | |
2975 | dirty_clusters = | |
2976 | percpu_counter_read_positive(&sbi->s_dirtyclusters_counter); | |
00d4e736 TT |
2977 | /* |
2978 | * Start pushing delalloc when 1/2 of free blocks are dirty. | |
2979 | */ | |
5c1ff336 | 2980 | if (dirty_clusters && (free_clusters < 2 * dirty_clusters)) |
10ee27a0 | 2981 | try_to_writeback_inodes_sb(sb, WB_REASON_FS_FREE_SPACE); |
00d4e736 | 2982 | |
5c1ff336 EW |
2983 | if (2 * free_clusters < 3 * dirty_clusters || |
2984 | free_clusters < (dirty_clusters + EXT4_FREECLUSTERS_WATERMARK)) { | |
79f0be8d | 2985 | /* |
c8afb446 ES |
2986 | * free block count is less than 150% of dirty blocks |
2987 | * or free blocks is less than watermark | |
79f0be8d AK |
2988 | */ |
2989 | return 1; | |
2990 | } | |
2991 | return 0; | |
2992 | } | |
2993 | ||
0ff8947f ES |
2994 | /* We always reserve for an inode update; the superblock could be there too */ |
2995 | static int ext4_da_write_credits(struct inode *inode, loff_t pos, unsigned len) | |
2996 | { | |
e2b911c5 | 2997 | if (likely(ext4_has_feature_large_file(inode->i_sb))) |
0ff8947f ES |
2998 | return 1; |
2999 | ||
3000 | if (pos + len <= 0x7fffffffULL) | |
3001 | return 1; | |
3002 | ||
3003 | /* We might need to update the superblock to set LARGE_FILE */ | |
3004 | return 2; | |
3005 | } | |
3006 | ||
64769240 | 3007 | static int ext4_da_write_begin(struct file *file, struct address_space *mapping, |
de9a55b8 TT |
3008 | loff_t pos, unsigned len, unsigned flags, |
3009 | struct page **pagep, void **fsdata) | |
64769240 | 3010 | { |
72b8ab9d | 3011 | int ret, retries = 0; |
64769240 AT |
3012 | struct page *page; |
3013 | pgoff_t index; | |
64769240 AT |
3014 | struct inode *inode = mapping->host; |
3015 | handle_t *handle; | |
3016 | ||
0db1ff22 TT |
3017 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
3018 | return -EIO; | |
3019 | ||
09cbfeaf | 3020 | index = pos >> PAGE_SHIFT; |
79f0be8d | 3021 | |
4db0d88e TT |
3022 | if (ext4_nonda_switch(inode->i_sb) || |
3023 | S_ISLNK(inode->i_mode)) { | |
79f0be8d AK |
3024 | *fsdata = (void *)FALL_BACK_TO_NONDELALLOC; |
3025 | return ext4_write_begin(file, mapping, pos, | |
3026 | len, flags, pagep, fsdata); | |
3027 | } | |
3028 | *fsdata = (void *)0; | |
9bffad1e | 3029 | trace_ext4_da_write_begin(inode, pos, len, flags); |
9c3569b5 TM |
3030 | |
3031 | if (ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA)) { | |
3032 | ret = ext4_da_write_inline_data_begin(mapping, inode, | |
3033 | pos, len, flags, | |
3034 | pagep, fsdata); | |
3035 | if (ret < 0) | |
47564bfb TT |
3036 | return ret; |
3037 | if (ret == 1) | |
3038 | return 0; | |
9c3569b5 TM |
3039 | } |
3040 | ||
47564bfb TT |
3041 | /* |
3042 | * grab_cache_page_write_begin() can take a long time if the | |
3043 | * system is thrashing due to memory pressure, or if the page | |
3044 | * is being written back. So grab it first before we start | |
3045 | * the transaction handle. This also allows us to allocate | |
3046 | * the page (if needed) without using GFP_NOFS. | |
3047 | */ | |
3048 | retry_grab: | |
3049 | page = grab_cache_page_write_begin(mapping, index, flags); | |
3050 | if (!page) | |
3051 | return -ENOMEM; | |
3052 | unlock_page(page); | |
3053 | ||
64769240 AT |
3054 | /* |
3055 | * With delayed allocation, we don't log the i_disksize update | |
3056 | * if there is delayed block allocation. But we still need | |
3057 | * to journalling the i_disksize update if writes to the end | |
3058 | * of file which has an already mapped buffer. | |
3059 | */ | |
47564bfb | 3060 | retry_journal: |
0ff8947f ES |
3061 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
3062 | ext4_da_write_credits(inode, pos, len)); | |
64769240 | 3063 | if (IS_ERR(handle)) { |
09cbfeaf | 3064 | put_page(page); |
47564bfb | 3065 | return PTR_ERR(handle); |
64769240 AT |
3066 | } |
3067 | ||
47564bfb TT |
3068 | lock_page(page); |
3069 | if (page->mapping != mapping) { | |
3070 | /* The page got truncated from under us */ | |
3071 | unlock_page(page); | |
09cbfeaf | 3072 | put_page(page); |
d5a0d4f7 | 3073 | ext4_journal_stop(handle); |
47564bfb | 3074 | goto retry_grab; |
d5a0d4f7 | 3075 | } |
47564bfb | 3076 | /* In case writeback began while the page was unlocked */ |
7afe5aa5 | 3077 | wait_for_stable_page(page); |
64769240 | 3078 | |
2058f83a MH |
3079 | #ifdef CONFIG_EXT4_FS_ENCRYPTION |
3080 | ret = ext4_block_write_begin(page, pos, len, | |
3081 | ext4_da_get_block_prep); | |
3082 | #else | |
6e1db88d | 3083 | ret = __block_write_begin(page, pos, len, ext4_da_get_block_prep); |
2058f83a | 3084 | #endif |
64769240 AT |
3085 | if (ret < 0) { |
3086 | unlock_page(page); | |
3087 | ext4_journal_stop(handle); | |
ae4d5372 AK |
3088 | /* |
3089 | * block_write_begin may have instantiated a few blocks | |
3090 | * outside i_size. Trim these off again. Don't need | |
3091 | * i_size_read because we hold i_mutex. | |
3092 | */ | |
3093 | if (pos + len > inode->i_size) | |
b9a4207d | 3094 | ext4_truncate_failed_write(inode); |
47564bfb TT |
3095 | |
3096 | if (ret == -ENOSPC && | |
3097 | ext4_should_retry_alloc(inode->i_sb, &retries)) | |
3098 | goto retry_journal; | |
3099 | ||
09cbfeaf | 3100 | put_page(page); |
47564bfb | 3101 | return ret; |
64769240 AT |
3102 | } |
3103 | ||
47564bfb | 3104 | *pagep = page; |
64769240 AT |
3105 | return ret; |
3106 | } | |
3107 | ||
632eaeab MC |
3108 | /* |
3109 | * Check if we should update i_disksize | |
3110 | * when write to the end of file but not require block allocation | |
3111 | */ | |
3112 | static int ext4_da_should_update_i_disksize(struct page *page, | |
de9a55b8 | 3113 | unsigned long offset) |
632eaeab MC |
3114 | { |
3115 | struct buffer_head *bh; | |
3116 | struct inode *inode = page->mapping->host; | |
3117 | unsigned int idx; | |
3118 | int i; | |
3119 | ||
3120 | bh = page_buffers(page); | |
3121 | idx = offset >> inode->i_blkbits; | |
3122 | ||
af5bc92d | 3123 | for (i = 0; i < idx; i++) |
632eaeab MC |
3124 | bh = bh->b_this_page; |
3125 | ||
29fa89d0 | 3126 | if (!buffer_mapped(bh) || (buffer_delay(bh)) || buffer_unwritten(bh)) |
632eaeab MC |
3127 | return 0; |
3128 | return 1; | |
3129 | } | |
3130 | ||
64769240 | 3131 | static int ext4_da_write_end(struct file *file, |
de9a55b8 TT |
3132 | struct address_space *mapping, |
3133 | loff_t pos, unsigned len, unsigned copied, | |
3134 | struct page *page, void *fsdata) | |
64769240 AT |
3135 | { |
3136 | struct inode *inode = mapping->host; | |
3137 | int ret = 0, ret2; | |
3138 | handle_t *handle = ext4_journal_current_handle(); | |
3139 | loff_t new_i_size; | |
632eaeab | 3140 | unsigned long start, end; |
79f0be8d AK |
3141 | int write_mode = (int)(unsigned long)fsdata; |
3142 | ||
74d553aa TT |
3143 | if (write_mode == FALL_BACK_TO_NONDELALLOC) |
3144 | return ext4_write_end(file, mapping, pos, | |
3145 | len, copied, page, fsdata); | |
632eaeab | 3146 | |
9bffad1e | 3147 | trace_ext4_da_write_end(inode, pos, len, copied); |
09cbfeaf | 3148 | start = pos & (PAGE_SIZE - 1); |
af5bc92d | 3149 | end = start + copied - 1; |
64769240 AT |
3150 | |
3151 | /* | |
3152 | * generic_write_end() will run mark_inode_dirty() if i_size | |
3153 | * changes. So let's piggyback the i_disksize mark_inode_dirty | |
3154 | * into that. | |
3155 | */ | |
64769240 | 3156 | new_i_size = pos + copied; |
ea51d132 | 3157 | if (copied && new_i_size > EXT4_I(inode)->i_disksize) { |
9c3569b5 TM |
3158 | if (ext4_has_inline_data(inode) || |
3159 | ext4_da_should_update_i_disksize(page, end)) { | |
ee124d27 | 3160 | ext4_update_i_disksize(inode, new_i_size); |
cf17fea6 AK |
3161 | /* We need to mark inode dirty even if |
3162 | * new_i_size is less that inode->i_size | |
3163 | * bu greater than i_disksize.(hint delalloc) | |
3164 | */ | |
3165 | ext4_mark_inode_dirty(handle, inode); | |
64769240 | 3166 | } |
632eaeab | 3167 | } |
9c3569b5 TM |
3168 | |
3169 | if (write_mode != CONVERT_INLINE_DATA && | |
3170 | ext4_test_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA) && | |
3171 | ext4_has_inline_data(inode)) | |
3172 | ret2 = ext4_da_write_inline_data_end(inode, pos, len, copied, | |
3173 | page); | |
3174 | else | |
3175 | ret2 = generic_write_end(file, mapping, pos, len, copied, | |
64769240 | 3176 | page, fsdata); |
9c3569b5 | 3177 | |
64769240 AT |
3178 | copied = ret2; |
3179 | if (ret2 < 0) | |
3180 | ret = ret2; | |
3181 | ret2 = ext4_journal_stop(handle); | |
3182 | if (!ret) | |
3183 | ret = ret2; | |
3184 | ||
3185 | return ret ? ret : copied; | |
3186 | } | |
3187 | ||
d47992f8 LC |
3188 | static void ext4_da_invalidatepage(struct page *page, unsigned int offset, |
3189 | unsigned int length) | |
64769240 | 3190 | { |
64769240 AT |
3191 | /* |
3192 | * Drop reserved blocks | |
3193 | */ | |
3194 | BUG_ON(!PageLocked(page)); | |
3195 | if (!page_has_buffers(page)) | |
3196 | goto out; | |
3197 | ||
ca99fdd2 | 3198 | ext4_da_page_release_reservation(page, offset, length); |
64769240 AT |
3199 | |
3200 | out: | |
d47992f8 | 3201 | ext4_invalidatepage(page, offset, length); |
64769240 AT |
3202 | |
3203 | return; | |
3204 | } | |
3205 | ||
ccd2506b TT |
3206 | /* |
3207 | * Force all delayed allocation blocks to be allocated for a given inode. | |
3208 | */ | |
3209 | int ext4_alloc_da_blocks(struct inode *inode) | |
3210 | { | |
fb40ba0d TT |
3211 | trace_ext4_alloc_da_blocks(inode); |
3212 | ||
71d4f7d0 | 3213 | if (!EXT4_I(inode)->i_reserved_data_blocks) |
ccd2506b TT |
3214 | return 0; |
3215 | ||
3216 | /* | |
3217 | * We do something simple for now. The filemap_flush() will | |
3218 | * also start triggering a write of the data blocks, which is | |
3219 | * not strictly speaking necessary (and for users of | |
3220 | * laptop_mode, not even desirable). However, to do otherwise | |
3221 | * would require replicating code paths in: | |
de9a55b8 | 3222 | * |
20970ba6 | 3223 | * ext4_writepages() -> |
ccd2506b TT |
3224 | * write_cache_pages() ---> (via passed in callback function) |
3225 | * __mpage_da_writepage() --> | |
3226 | * mpage_add_bh_to_extent() | |
3227 | * mpage_da_map_blocks() | |
3228 | * | |
3229 | * The problem is that write_cache_pages(), located in | |
3230 | * mm/page-writeback.c, marks pages clean in preparation for | |
3231 | * doing I/O, which is not desirable if we're not planning on | |
3232 | * doing I/O at all. | |
3233 | * | |
3234 | * We could call write_cache_pages(), and then redirty all of | |
380cf090 | 3235 | * the pages by calling redirty_page_for_writepage() but that |
ccd2506b TT |
3236 | * would be ugly in the extreme. So instead we would need to |
3237 | * replicate parts of the code in the above functions, | |
25985edc | 3238 | * simplifying them because we wouldn't actually intend to |
ccd2506b TT |
3239 | * write out the pages, but rather only collect contiguous |
3240 | * logical block extents, call the multi-block allocator, and | |
3241 | * then update the buffer heads with the block allocations. | |
de9a55b8 | 3242 | * |
ccd2506b TT |
3243 | * For now, though, we'll cheat by calling filemap_flush(), |
3244 | * which will map the blocks, and start the I/O, but not | |
3245 | * actually wait for the I/O to complete. | |
3246 | */ | |
3247 | return filemap_flush(inode->i_mapping); | |
3248 | } | |
64769240 | 3249 | |
ac27a0ec DK |
3250 | /* |
3251 | * bmap() is special. It gets used by applications such as lilo and by | |
3252 | * the swapper to find the on-disk block of a specific piece of data. | |
3253 | * | |
3254 | * Naturally, this is dangerous if the block concerned is still in the | |
617ba13b | 3255 | * journal. If somebody makes a swapfile on an ext4 data-journaling |
ac27a0ec DK |
3256 | * filesystem and enables swap, then they may get a nasty shock when the |
3257 | * data getting swapped to that swapfile suddenly gets overwritten by | |
3258 | * the original zero's written out previously to the journal and | |
3259 | * awaiting writeback in the kernel's buffer cache. | |
3260 | * | |
3261 | * So, if we see any bmap calls here on a modified, data-journaled file, | |
3262 | * take extra steps to flush any blocks which might be in the cache. | |
3263 | */ | |
617ba13b | 3264 | static sector_t ext4_bmap(struct address_space *mapping, sector_t block) |
ac27a0ec DK |
3265 | { |
3266 | struct inode *inode = mapping->host; | |
3267 | journal_t *journal; | |
3268 | int err; | |
3269 | ||
46c7f254 TM |
3270 | /* |
3271 | * We can get here for an inline file via the FIBMAP ioctl | |
3272 | */ | |
3273 | if (ext4_has_inline_data(inode)) | |
3274 | return 0; | |
3275 | ||
64769240 AT |
3276 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) && |
3277 | test_opt(inode->i_sb, DELALLOC)) { | |
3278 | /* | |
3279 | * With delalloc we want to sync the file | |
3280 | * so that we can make sure we allocate | |
3281 | * blocks for file | |
3282 | */ | |
3283 | filemap_write_and_wait(mapping); | |
3284 | } | |
3285 | ||
19f5fb7a TT |
3286 | if (EXT4_JOURNAL(inode) && |
3287 | ext4_test_inode_state(inode, EXT4_STATE_JDATA)) { | |
ac27a0ec DK |
3288 | /* |
3289 | * This is a REALLY heavyweight approach, but the use of | |
3290 | * bmap on dirty files is expected to be extremely rare: | |
3291 | * only if we run lilo or swapon on a freshly made file | |
3292 | * do we expect this to happen. | |
3293 | * | |
3294 | * (bmap requires CAP_SYS_RAWIO so this does not | |
3295 | * represent an unprivileged user DOS attack --- we'd be | |
3296 | * in trouble if mortal users could trigger this path at | |
3297 | * will.) | |
3298 | * | |
617ba13b | 3299 | * NB. EXT4_STATE_JDATA is not set on files other than |
ac27a0ec DK |
3300 | * regular files. If somebody wants to bmap a directory |
3301 | * or symlink and gets confused because the buffer | |
3302 | * hasn't yet been flushed to disk, they deserve | |
3303 | * everything they get. | |
3304 | */ | |
3305 | ||
19f5fb7a | 3306 | ext4_clear_inode_state(inode, EXT4_STATE_JDATA); |
617ba13b | 3307 | journal = EXT4_JOURNAL(inode); |
dab291af MC |
3308 | jbd2_journal_lock_updates(journal); |
3309 | err = jbd2_journal_flush(journal); | |
3310 | jbd2_journal_unlock_updates(journal); | |
ac27a0ec DK |
3311 | |
3312 | if (err) | |
3313 | return 0; | |
3314 | } | |
3315 | ||
af5bc92d | 3316 | return generic_block_bmap(mapping, block, ext4_get_block); |
ac27a0ec DK |
3317 | } |
3318 | ||
617ba13b | 3319 | static int ext4_readpage(struct file *file, struct page *page) |
ac27a0ec | 3320 | { |
46c7f254 TM |
3321 | int ret = -EAGAIN; |
3322 | struct inode *inode = page->mapping->host; | |
3323 | ||
0562e0ba | 3324 | trace_ext4_readpage(page); |
46c7f254 TM |
3325 | |
3326 | if (ext4_has_inline_data(inode)) | |
3327 | ret = ext4_readpage_inline(inode, page); | |
3328 | ||
3329 | if (ret == -EAGAIN) | |
f64e02fe | 3330 | return ext4_mpage_readpages(page->mapping, NULL, page, 1); |
46c7f254 TM |
3331 | |
3332 | return ret; | |
ac27a0ec DK |
3333 | } |
3334 | ||
3335 | static int | |
617ba13b | 3336 | ext4_readpages(struct file *file, struct address_space *mapping, |
ac27a0ec DK |
3337 | struct list_head *pages, unsigned nr_pages) |
3338 | { | |
46c7f254 TM |
3339 | struct inode *inode = mapping->host; |
3340 | ||
3341 | /* If the file has inline data, no need to do readpages. */ | |
3342 | if (ext4_has_inline_data(inode)) | |
3343 | return 0; | |
3344 | ||
f64e02fe | 3345 | return ext4_mpage_readpages(mapping, pages, NULL, nr_pages); |
ac27a0ec DK |
3346 | } |
3347 | ||
d47992f8 LC |
3348 | static void ext4_invalidatepage(struct page *page, unsigned int offset, |
3349 | unsigned int length) | |
ac27a0ec | 3350 | { |
ca99fdd2 | 3351 | trace_ext4_invalidatepage(page, offset, length); |
0562e0ba | 3352 | |
4520fb3c JK |
3353 | /* No journalling happens on data buffers when this function is used */ |
3354 | WARN_ON(page_has_buffers(page) && buffer_jbd(page_buffers(page))); | |
3355 | ||
ca99fdd2 | 3356 | block_invalidatepage(page, offset, length); |
4520fb3c JK |
3357 | } |
3358 | ||
53e87268 | 3359 | static int __ext4_journalled_invalidatepage(struct page *page, |
ca99fdd2 LC |
3360 | unsigned int offset, |
3361 | unsigned int length) | |
4520fb3c JK |
3362 | { |
3363 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); | |
3364 | ||
ca99fdd2 | 3365 | trace_ext4_journalled_invalidatepage(page, offset, length); |
4520fb3c | 3366 | |
ac27a0ec DK |
3367 | /* |
3368 | * If it's a full truncate we just forget about the pending dirtying | |
3369 | */ | |
09cbfeaf | 3370 | if (offset == 0 && length == PAGE_SIZE) |
ac27a0ec DK |
3371 | ClearPageChecked(page); |
3372 | ||
ca99fdd2 | 3373 | return jbd2_journal_invalidatepage(journal, page, offset, length); |
53e87268 JK |
3374 | } |
3375 | ||
3376 | /* Wrapper for aops... */ | |
3377 | static void ext4_journalled_invalidatepage(struct page *page, | |
d47992f8 LC |
3378 | unsigned int offset, |
3379 | unsigned int length) | |
53e87268 | 3380 | { |
ca99fdd2 | 3381 | WARN_ON(__ext4_journalled_invalidatepage(page, offset, length) < 0); |
ac27a0ec DK |
3382 | } |
3383 | ||
617ba13b | 3384 | static int ext4_releasepage(struct page *page, gfp_t wait) |
ac27a0ec | 3385 | { |
617ba13b | 3386 | journal_t *journal = EXT4_JOURNAL(page->mapping->host); |
ac27a0ec | 3387 | |
0562e0ba JZ |
3388 | trace_ext4_releasepage(page); |
3389 | ||
e1c36595 JK |
3390 | /* Page has dirty journalled data -> cannot release */ |
3391 | if (PageChecked(page)) | |
ac27a0ec | 3392 | return 0; |
0390131b FM |
3393 | if (journal) |
3394 | return jbd2_journal_try_to_free_buffers(journal, page, wait); | |
3395 | else | |
3396 | return try_to_free_buffers(page); | |
ac27a0ec DK |
3397 | } |
3398 | ||
b8a6176c JK |
3399 | static bool ext4_inode_datasync_dirty(struct inode *inode) |
3400 | { | |
3401 | journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; | |
3402 | ||
3403 | if (journal) | |
3404 | return !jbd2_transaction_committed(journal, | |
3405 | EXT4_I(inode)->i_datasync_tid); | |
3406 | /* Any metadata buffers to write? */ | |
3407 | if (!list_empty(&inode->i_mapping->private_list)) | |
3408 | return true; | |
3409 | return inode->i_state & I_DIRTY_DATASYNC; | |
3410 | } | |
3411 | ||
364443cb JK |
3412 | static int ext4_iomap_begin(struct inode *inode, loff_t offset, loff_t length, |
3413 | unsigned flags, struct iomap *iomap) | |
3414 | { | |
5e405595 | 3415 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
364443cb JK |
3416 | unsigned int blkbits = inode->i_blkbits; |
3417 | unsigned long first_block = offset >> blkbits; | |
3418 | unsigned long last_block = (offset + length - 1) >> blkbits; | |
3419 | struct ext4_map_blocks map; | |
545052e9 | 3420 | bool delalloc = false; |
364443cb JK |
3421 | int ret; |
3422 | ||
7046ae35 AG |
3423 | |
3424 | if (flags & IOMAP_REPORT) { | |
3425 | if (ext4_has_inline_data(inode)) { | |
3426 | ret = ext4_inline_data_iomap(inode, iomap); | |
3427 | if (ret != -EAGAIN) { | |
3428 | if (ret == 0 && offset >= iomap->length) | |
3429 | ret = -ENOENT; | |
3430 | return ret; | |
3431 | } | |
3432 | } | |
3433 | } else { | |
3434 | if (WARN_ON_ONCE(ext4_has_inline_data(inode))) | |
3435 | return -ERANGE; | |
3436 | } | |
364443cb JK |
3437 | |
3438 | map.m_lblk = first_block; | |
3439 | map.m_len = last_block - first_block + 1; | |
3440 | ||
545052e9 | 3441 | if (flags & IOMAP_REPORT) { |
776722e8 | 3442 | ret = ext4_map_blocks(NULL, inode, &map, 0); |
545052e9 CH |
3443 | if (ret < 0) |
3444 | return ret; | |
3445 | ||
3446 | if (ret == 0) { | |
3447 | ext4_lblk_t end = map.m_lblk + map.m_len - 1; | |
3448 | struct extent_status es; | |
3449 | ||
3450 | ext4_es_find_delayed_extent_range(inode, map.m_lblk, end, &es); | |
3451 | ||
3452 | if (!es.es_len || es.es_lblk > end) { | |
3453 | /* entire range is a hole */ | |
3454 | } else if (es.es_lblk > map.m_lblk) { | |
3455 | /* range starts with a hole */ | |
3456 | map.m_len = es.es_lblk - map.m_lblk; | |
3457 | } else { | |
3458 | ext4_lblk_t offs = 0; | |
3459 | ||
3460 | if (es.es_lblk < map.m_lblk) | |
3461 | offs = map.m_lblk - es.es_lblk; | |
3462 | map.m_lblk = es.es_lblk + offs; | |
3463 | map.m_len = es.es_len - offs; | |
3464 | delalloc = true; | |
3465 | } | |
3466 | } | |
3467 | } else if (flags & IOMAP_WRITE) { | |
776722e8 JK |
3468 | int dio_credits; |
3469 | handle_t *handle; | |
3470 | int retries = 0; | |
3471 | ||
3472 | /* Trim mapping request to maximum we can map at once for DIO */ | |
3473 | if (map.m_len > DIO_MAX_BLOCKS) | |
3474 | map.m_len = DIO_MAX_BLOCKS; | |
3475 | dio_credits = ext4_chunk_trans_blocks(inode, map.m_len); | |
3476 | retry: | |
3477 | /* | |
3478 | * Either we allocate blocks and then we don't get unwritten | |
3479 | * extent so we have reserved enough credits, or the blocks | |
3480 | * are already allocated and unwritten and in that case | |
3481 | * extent conversion fits in the credits as well. | |
3482 | */ | |
3483 | handle = ext4_journal_start(inode, EXT4_HT_MAP_BLOCKS, | |
3484 | dio_credits); | |
3485 | if (IS_ERR(handle)) | |
3486 | return PTR_ERR(handle); | |
3487 | ||
3488 | ret = ext4_map_blocks(handle, inode, &map, | |
776722e8 JK |
3489 | EXT4_GET_BLOCKS_CREATE_ZERO); |
3490 | if (ret < 0) { | |
3491 | ext4_journal_stop(handle); | |
3492 | if (ret == -ENOSPC && | |
3493 | ext4_should_retry_alloc(inode->i_sb, &retries)) | |
3494 | goto retry; | |
3495 | return ret; | |
3496 | } | |
776722e8 JK |
3497 | |
3498 | /* | |
e2ae766c | 3499 | * If we added blocks beyond i_size, we need to make sure they |
776722e8 | 3500 | * will get truncated if we crash before updating i_size in |
e2ae766c JK |
3501 | * ext4_iomap_end(). For faults we don't need to do that (and |
3502 | * even cannot because for orphan list operations inode_lock is | |
3503 | * required) - if we happen to instantiate block beyond i_size, | |
3504 | * it is because we race with truncate which has already added | |
3505 | * the inode to the orphan list. | |
776722e8 | 3506 | */ |
e2ae766c JK |
3507 | if (!(flags & IOMAP_FAULT) && first_block + map.m_len > |
3508 | (i_size_read(inode) + (1 << blkbits) - 1) >> blkbits) { | |
776722e8 JK |
3509 | int err; |
3510 | ||
3511 | err = ext4_orphan_add(handle, inode); | |
3512 | if (err < 0) { | |
3513 | ext4_journal_stop(handle); | |
3514 | return err; | |
3515 | } | |
3516 | } | |
3517 | ext4_journal_stop(handle); | |
545052e9 CH |
3518 | } else { |
3519 | ret = ext4_map_blocks(NULL, inode, &map, 0); | |
3520 | if (ret < 0) | |
3521 | return ret; | |
776722e8 | 3522 | } |
364443cb JK |
3523 | |
3524 | iomap->flags = 0; | |
aaa422c4 | 3525 | if (ext4_inode_datasync_dirty(inode)) |
b8a6176c | 3526 | iomap->flags |= IOMAP_F_DIRTY; |
5e405595 DW |
3527 | iomap->bdev = inode->i_sb->s_bdev; |
3528 | iomap->dax_dev = sbi->s_daxdev; | |
fe23cb65 | 3529 | iomap->offset = (u64)first_block << blkbits; |
545052e9 | 3530 | iomap->length = (u64)map.m_len << blkbits; |
364443cb JK |
3531 | |
3532 | if (ret == 0) { | |
545052e9 | 3533 | iomap->type = delalloc ? IOMAP_DELALLOC : IOMAP_HOLE; |
19fe5f64 | 3534 | iomap->addr = IOMAP_NULL_ADDR; |
364443cb JK |
3535 | } else { |
3536 | if (map.m_flags & EXT4_MAP_MAPPED) { | |
3537 | iomap->type = IOMAP_MAPPED; | |
3538 | } else if (map.m_flags & EXT4_MAP_UNWRITTEN) { | |
3539 | iomap->type = IOMAP_UNWRITTEN; | |
3540 | } else { | |
3541 | WARN_ON_ONCE(1); | |
3542 | return -EIO; | |
3543 | } | |
19fe5f64 | 3544 | iomap->addr = (u64)map.m_pblk << blkbits; |
364443cb JK |
3545 | } |
3546 | ||
3547 | if (map.m_flags & EXT4_MAP_NEW) | |
3548 | iomap->flags |= IOMAP_F_NEW; | |
545052e9 | 3549 | |
364443cb JK |
3550 | return 0; |
3551 | } | |
3552 | ||
776722e8 JK |
3553 | static int ext4_iomap_end(struct inode *inode, loff_t offset, loff_t length, |
3554 | ssize_t written, unsigned flags, struct iomap *iomap) | |
3555 | { | |
3556 | int ret = 0; | |
3557 | handle_t *handle; | |
3558 | int blkbits = inode->i_blkbits; | |
3559 | bool truncate = false; | |
3560 | ||
e2ae766c | 3561 | if (!(flags & IOMAP_WRITE) || (flags & IOMAP_FAULT)) |
776722e8 JK |
3562 | return 0; |
3563 | ||
3564 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 2); | |
3565 | if (IS_ERR(handle)) { | |
3566 | ret = PTR_ERR(handle); | |
3567 | goto orphan_del; | |
3568 | } | |
3569 | if (ext4_update_inode_size(inode, offset + written)) | |
3570 | ext4_mark_inode_dirty(handle, inode); | |
3571 | /* | |
3572 | * We may need to truncate allocated but not written blocks beyond EOF. | |
3573 | */ | |
3574 | if (iomap->offset + iomap->length > | |
3575 | ALIGN(inode->i_size, 1 << blkbits)) { | |
3576 | ext4_lblk_t written_blk, end_blk; | |
3577 | ||
3578 | written_blk = (offset + written) >> blkbits; | |
3579 | end_blk = (offset + length) >> blkbits; | |
3580 | if (written_blk < end_blk && ext4_can_truncate(inode)) | |
3581 | truncate = true; | |
3582 | } | |
3583 | /* | |
3584 | * Remove inode from orphan list if we were extending a inode and | |
3585 | * everything went fine. | |
3586 | */ | |
3587 | if (!truncate && inode->i_nlink && | |
3588 | !list_empty(&EXT4_I(inode)->i_orphan)) | |
3589 | ext4_orphan_del(handle, inode); | |
3590 | ext4_journal_stop(handle); | |
3591 | if (truncate) { | |
3592 | ext4_truncate_failed_write(inode); | |
3593 | orphan_del: | |
3594 | /* | |
3595 | * If truncate failed early the inode might still be on the | |
3596 | * orphan list; we need to make sure the inode is removed from | |
3597 | * the orphan list in that case. | |
3598 | */ | |
3599 | if (inode->i_nlink) | |
3600 | ext4_orphan_del(NULL, inode); | |
3601 | } | |
3602 | return ret; | |
3603 | } | |
3604 | ||
8ff6daa1 | 3605 | const struct iomap_ops ext4_iomap_ops = { |
364443cb | 3606 | .iomap_begin = ext4_iomap_begin, |
776722e8 | 3607 | .iomap_end = ext4_iomap_end, |
364443cb JK |
3608 | }; |
3609 | ||
187372a3 | 3610 | static int ext4_end_io_dio(struct kiocb *iocb, loff_t offset, |
7b7a8665 | 3611 | ssize_t size, void *private) |
4c0425ff | 3612 | { |
109811c2 | 3613 | ext4_io_end_t *io_end = private; |
4c0425ff | 3614 | |
97a851ed | 3615 | /* if not async direct IO just return */ |
7b7a8665 | 3616 | if (!io_end) |
187372a3 | 3617 | return 0; |
4b70df18 | 3618 | |
88635ca2 | 3619 | ext_debug("ext4_end_io_dio(): io_end 0x%p " |
ace36ad4 | 3620 | "for inode %lu, iocb 0x%p, offset %llu, size %zd\n", |
109811c2 | 3621 | io_end, io_end->inode->i_ino, iocb, offset, size); |
8d5d02e6 | 3622 | |
74c66bcb JK |
3623 | /* |
3624 | * Error during AIO DIO. We cannot convert unwritten extents as the | |
3625 | * data was not written. Just clear the unwritten flag and drop io_end. | |
3626 | */ | |
3627 | if (size <= 0) { | |
3628 | ext4_clear_io_unwritten_flag(io_end); | |
3629 | size = 0; | |
3630 | } | |
4c0425ff MC |
3631 | io_end->offset = offset; |
3632 | io_end->size = size; | |
7b7a8665 | 3633 | ext4_put_io_end(io_end); |
187372a3 CH |
3634 | |
3635 | return 0; | |
4c0425ff | 3636 | } |
c7064ef1 | 3637 | |
4c0425ff | 3638 | /* |
914f82a3 JK |
3639 | * Handling of direct IO writes. |
3640 | * | |
3641 | * For ext4 extent files, ext4 will do direct-io write even to holes, | |
4c0425ff MC |
3642 | * preallocated extents, and those write extend the file, no need to |
3643 | * fall back to buffered IO. | |
3644 | * | |
556615dc | 3645 | * For holes, we fallocate those blocks, mark them as unwritten |
69c499d1 | 3646 | * If those blocks were preallocated, we mark sure they are split, but |
556615dc | 3647 | * still keep the range to write as unwritten. |
4c0425ff | 3648 | * |
69c499d1 | 3649 | * The unwritten extents will be converted to written when DIO is completed. |
8d5d02e6 | 3650 | * For async direct IO, since the IO may still pending when return, we |
25985edc | 3651 | * set up an end_io call back function, which will do the conversion |
8d5d02e6 | 3652 | * when async direct IO completed. |
4c0425ff MC |
3653 | * |
3654 | * If the O_DIRECT write will extend the file then add this inode to the | |
3655 | * orphan list. So recovery will truncate it back to the original size | |
3656 | * if the machine crashes during the write. | |
3657 | * | |
3658 | */ | |
0e01df10 | 3659 | static ssize_t ext4_direct_IO_write(struct kiocb *iocb, struct iov_iter *iter) |
4c0425ff MC |
3660 | { |
3661 | struct file *file = iocb->ki_filp; | |
3662 | struct inode *inode = file->f_mapping->host; | |
45d8ec4d | 3663 | struct ext4_inode_info *ei = EXT4_I(inode); |
4c0425ff | 3664 | ssize_t ret; |
c8b8e32d | 3665 | loff_t offset = iocb->ki_pos; |
a6cbcd4a | 3666 | size_t count = iov_iter_count(iter); |
69c499d1 TT |
3667 | int overwrite = 0; |
3668 | get_block_t *get_block_func = NULL; | |
3669 | int dio_flags = 0; | |
4c0425ff | 3670 | loff_t final_size = offset + count; |
914f82a3 JK |
3671 | int orphan = 0; |
3672 | handle_t *handle; | |
729f52c6 | 3673 | |
45d8ec4d | 3674 | if (final_size > inode->i_size || final_size > ei->i_disksize) { |
914f82a3 JK |
3675 | /* Credits for sb + inode write */ |
3676 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 2); | |
3677 | if (IS_ERR(handle)) { | |
3678 | ret = PTR_ERR(handle); | |
3679 | goto out; | |
3680 | } | |
3681 | ret = ext4_orphan_add(handle, inode); | |
3682 | if (ret) { | |
3683 | ext4_journal_stop(handle); | |
3684 | goto out; | |
3685 | } | |
3686 | orphan = 1; | |
73fdad00 | 3687 | ext4_update_i_disksize(inode, inode->i_size); |
914f82a3 JK |
3688 | ext4_journal_stop(handle); |
3689 | } | |
4bd809db | 3690 | |
69c499d1 | 3691 | BUG_ON(iocb->private == NULL); |
4bd809db | 3692 | |
e8340395 JK |
3693 | /* |
3694 | * Make all waiters for direct IO properly wait also for extent | |
3695 | * conversion. This also disallows race between truncate() and | |
3696 | * overwrite DIO as i_dio_count needs to be incremented under i_mutex. | |
3697 | */ | |
914f82a3 | 3698 | inode_dio_begin(inode); |
e8340395 | 3699 | |
69c499d1 TT |
3700 | /* If we do a overwrite dio, i_mutex locking can be released */ |
3701 | overwrite = *((int *)iocb->private); | |
4bd809db | 3702 | |
2dcba478 | 3703 | if (overwrite) |
5955102c | 3704 | inode_unlock(inode); |
8d5d02e6 | 3705 | |
69c499d1 | 3706 | /* |
914f82a3 | 3707 | * For extent mapped files we could direct write to holes and fallocate. |
69c499d1 | 3708 | * |
109811c2 JK |
3709 | * Allocated blocks to fill the hole are marked as unwritten to prevent |
3710 | * parallel buffered read to expose the stale data before DIO complete | |
3711 | * the data IO. | |
69c499d1 | 3712 | * |
109811c2 JK |
3713 | * As to previously fallocated extents, ext4 get_block will just simply |
3714 | * mark the buffer mapped but still keep the extents unwritten. | |
69c499d1 | 3715 | * |
109811c2 JK |
3716 | * For non AIO case, we will convert those unwritten extents to written |
3717 | * after return back from blockdev_direct_IO. That way we save us from | |
3718 | * allocating io_end structure and also the overhead of offloading | |
3719 | * the extent convertion to a workqueue. | |
69c499d1 TT |
3720 | * |
3721 | * For async DIO, the conversion needs to be deferred when the | |
3722 | * IO is completed. The ext4 end_io callback function will be | |
3723 | * called to take care of the conversion work. Here for async | |
3724 | * case, we allocate an io_end structure to hook to the iocb. | |
3725 | */ | |
3726 | iocb->private = NULL; | |
109811c2 | 3727 | if (overwrite) |
705965bd | 3728 | get_block_func = ext4_dio_get_block_overwrite; |
0bd2d5ec | 3729 | else if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS) || |
93407472 | 3730 | round_down(offset, i_blocksize(inode)) >= inode->i_size) { |
914f82a3 JK |
3731 | get_block_func = ext4_dio_get_block; |
3732 | dio_flags = DIO_LOCKING | DIO_SKIP_HOLES; | |
3733 | } else if (is_sync_kiocb(iocb)) { | |
109811c2 JK |
3734 | get_block_func = ext4_dio_get_block_unwritten_sync; |
3735 | dio_flags = DIO_LOCKING; | |
69c499d1 | 3736 | } else { |
109811c2 | 3737 | get_block_func = ext4_dio_get_block_unwritten_async; |
69c499d1 TT |
3738 | dio_flags = DIO_LOCKING; |
3739 | } | |
0bd2d5ec JK |
3740 | ret = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev, iter, |
3741 | get_block_func, ext4_end_io_dio, NULL, | |
3742 | dio_flags); | |
69c499d1 | 3743 | |
97a851ed | 3744 | if (ret > 0 && !overwrite && ext4_test_inode_state(inode, |
69c499d1 TT |
3745 | EXT4_STATE_DIO_UNWRITTEN)) { |
3746 | int err; | |
3747 | /* | |
3748 | * for non AIO case, since the IO is already | |
3749 | * completed, we could do the conversion right here | |
3750 | */ | |
6b523df4 | 3751 | err = ext4_convert_unwritten_extents(NULL, inode, |
69c499d1 TT |
3752 | offset, ret); |
3753 | if (err < 0) | |
3754 | ret = err; | |
3755 | ext4_clear_inode_state(inode, EXT4_STATE_DIO_UNWRITTEN); | |
3756 | } | |
4bd809db | 3757 | |
914f82a3 | 3758 | inode_dio_end(inode); |
69c499d1 | 3759 | /* take i_mutex locking again if we do a ovewrite dio */ |
2dcba478 | 3760 | if (overwrite) |
5955102c | 3761 | inode_lock(inode); |
8d5d02e6 | 3762 | |
914f82a3 JK |
3763 | if (ret < 0 && final_size > inode->i_size) |
3764 | ext4_truncate_failed_write(inode); | |
3765 | ||
3766 | /* Handle extending of i_size after direct IO write */ | |
3767 | if (orphan) { | |
3768 | int err; | |
3769 | ||
3770 | /* Credits for sb + inode write */ | |
3771 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 2); | |
3772 | if (IS_ERR(handle)) { | |
abbc3f93 HS |
3773 | /* |
3774 | * We wrote the data but cannot extend | |
3775 | * i_size. Bail out. In async io case, we do | |
3776 | * not return error here because we have | |
3777 | * already submmitted the corresponding | |
3778 | * bio. Returning error here makes the caller | |
3779 | * think that this IO is done and failed | |
3780 | * resulting in race with bio's completion | |
3781 | * handler. | |
3782 | */ | |
3783 | if (!ret) | |
3784 | ret = PTR_ERR(handle); | |
914f82a3 JK |
3785 | if (inode->i_nlink) |
3786 | ext4_orphan_del(NULL, inode); | |
3787 | ||
3788 | goto out; | |
3789 | } | |
3790 | if (inode->i_nlink) | |
3791 | ext4_orphan_del(handle, inode); | |
3792 | if (ret > 0) { | |
3793 | loff_t end = offset + ret; | |
45d8ec4d | 3794 | if (end > inode->i_size || end > ei->i_disksize) { |
73fdad00 | 3795 | ext4_update_i_disksize(inode, end); |
45d8ec4d EG |
3796 | if (end > inode->i_size) |
3797 | i_size_write(inode, end); | |
914f82a3 JK |
3798 | /* |
3799 | * We're going to return a positive `ret' | |
3800 | * here due to non-zero-length I/O, so there's | |
3801 | * no way of reporting error returns from | |
3802 | * ext4_mark_inode_dirty() to userspace. So | |
3803 | * ignore it. | |
3804 | */ | |
3805 | ext4_mark_inode_dirty(handle, inode); | |
3806 | } | |
3807 | } | |
3808 | err = ext4_journal_stop(handle); | |
3809 | if (ret == 0) | |
3810 | ret = err; | |
3811 | } | |
3812 | out: | |
3813 | return ret; | |
3814 | } | |
3815 | ||
0e01df10 | 3816 | static ssize_t ext4_direct_IO_read(struct kiocb *iocb, struct iov_iter *iter) |
914f82a3 | 3817 | { |
16c54688 JK |
3818 | struct address_space *mapping = iocb->ki_filp->f_mapping; |
3819 | struct inode *inode = mapping->host; | |
0bd2d5ec | 3820 | size_t count = iov_iter_count(iter); |
914f82a3 JK |
3821 | ssize_t ret; |
3822 | ||
16c54688 JK |
3823 | /* |
3824 | * Shared inode_lock is enough for us - it protects against concurrent | |
3825 | * writes & truncates and since we take care of writing back page cache, | |
3826 | * we are protected against page writeback as well. | |
3827 | */ | |
3828 | inode_lock_shared(inode); | |
0bd2d5ec | 3829 | ret = filemap_write_and_wait_range(mapping, iocb->ki_pos, |
e5465795 | 3830 | iocb->ki_pos + count - 1); |
0bd2d5ec JK |
3831 | if (ret) |
3832 | goto out_unlock; | |
3833 | ret = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev, | |
3834 | iter, ext4_dio_get_block, NULL, NULL, 0); | |
16c54688 JK |
3835 | out_unlock: |
3836 | inode_unlock_shared(inode); | |
69c499d1 | 3837 | return ret; |
4c0425ff MC |
3838 | } |
3839 | ||
c8b8e32d | 3840 | static ssize_t ext4_direct_IO(struct kiocb *iocb, struct iov_iter *iter) |
4c0425ff MC |
3841 | { |
3842 | struct file *file = iocb->ki_filp; | |
3843 | struct inode *inode = file->f_mapping->host; | |
a6cbcd4a | 3844 | size_t count = iov_iter_count(iter); |
c8b8e32d | 3845 | loff_t offset = iocb->ki_pos; |
0562e0ba | 3846 | ssize_t ret; |
4c0425ff | 3847 | |
2058f83a MH |
3848 | #ifdef CONFIG_EXT4_FS_ENCRYPTION |
3849 | if (ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode)) | |
3850 | return 0; | |
3851 | #endif | |
3852 | ||
84ebd795 TT |
3853 | /* |
3854 | * If we are doing data journalling we don't support O_DIRECT | |
3855 | */ | |
3856 | if (ext4_should_journal_data(inode)) | |
3857 | return 0; | |
3858 | ||
46c7f254 TM |
3859 | /* Let buffer I/O handle the inline data case. */ |
3860 | if (ext4_has_inline_data(inode)) | |
3861 | return 0; | |
3862 | ||
6f673763 | 3863 | trace_ext4_direct_IO_enter(inode, offset, count, iov_iter_rw(iter)); |
914f82a3 | 3864 | if (iov_iter_rw(iter) == READ) |
0e01df10 | 3865 | ret = ext4_direct_IO_read(iocb, iter); |
0562e0ba | 3866 | else |
0e01df10 | 3867 | ret = ext4_direct_IO_write(iocb, iter); |
6f673763 | 3868 | trace_ext4_direct_IO_exit(inode, offset, count, iov_iter_rw(iter), ret); |
0562e0ba | 3869 | return ret; |
4c0425ff MC |
3870 | } |
3871 | ||
ac27a0ec | 3872 | /* |
617ba13b | 3873 | * Pages can be marked dirty completely asynchronously from ext4's journalling |
ac27a0ec DK |
3874 | * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do |
3875 | * much here because ->set_page_dirty is called under VFS locks. The page is | |
3876 | * not necessarily locked. | |
3877 | * | |
3878 | * We cannot just dirty the page and leave attached buffers clean, because the | |
3879 | * buffers' dirty state is "definitive". We cannot just set the buffers dirty | |
3880 | * or jbddirty because all the journalling code will explode. | |
3881 | * | |
3882 | * So what we do is to mark the page "pending dirty" and next time writepage | |
3883 | * is called, propagate that into the buffers appropriately. | |
3884 | */ | |
617ba13b | 3885 | static int ext4_journalled_set_page_dirty(struct page *page) |
ac27a0ec DK |
3886 | { |
3887 | SetPageChecked(page); | |
3888 | return __set_page_dirty_nobuffers(page); | |
3889 | } | |
3890 | ||
6dcc693b JK |
3891 | static int ext4_set_page_dirty(struct page *page) |
3892 | { | |
3893 | WARN_ON_ONCE(!PageLocked(page) && !PageDirty(page)); | |
3894 | WARN_ON_ONCE(!page_has_buffers(page)); | |
3895 | return __set_page_dirty_buffers(page); | |
3896 | } | |
3897 | ||
74d553aa | 3898 | static const struct address_space_operations ext4_aops = { |
8ab22b9a HH |
3899 | .readpage = ext4_readpage, |
3900 | .readpages = ext4_readpages, | |
43ce1d23 | 3901 | .writepage = ext4_writepage, |
20970ba6 | 3902 | .writepages = ext4_writepages, |
8ab22b9a | 3903 | .write_begin = ext4_write_begin, |
74d553aa | 3904 | .write_end = ext4_write_end, |
6dcc693b | 3905 | .set_page_dirty = ext4_set_page_dirty, |
8ab22b9a HH |
3906 | .bmap = ext4_bmap, |
3907 | .invalidatepage = ext4_invalidatepage, | |
3908 | .releasepage = ext4_releasepage, | |
3909 | .direct_IO = ext4_direct_IO, | |
3910 | .migratepage = buffer_migrate_page, | |
3911 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 3912 | .error_remove_page = generic_error_remove_page, |
ac27a0ec DK |
3913 | }; |
3914 | ||
617ba13b | 3915 | static const struct address_space_operations ext4_journalled_aops = { |
8ab22b9a HH |
3916 | .readpage = ext4_readpage, |
3917 | .readpages = ext4_readpages, | |
43ce1d23 | 3918 | .writepage = ext4_writepage, |
20970ba6 | 3919 | .writepages = ext4_writepages, |
8ab22b9a HH |
3920 | .write_begin = ext4_write_begin, |
3921 | .write_end = ext4_journalled_write_end, | |
3922 | .set_page_dirty = ext4_journalled_set_page_dirty, | |
3923 | .bmap = ext4_bmap, | |
4520fb3c | 3924 | .invalidatepage = ext4_journalled_invalidatepage, |
8ab22b9a | 3925 | .releasepage = ext4_releasepage, |
84ebd795 | 3926 | .direct_IO = ext4_direct_IO, |
8ab22b9a | 3927 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 3928 | .error_remove_page = generic_error_remove_page, |
ac27a0ec DK |
3929 | }; |
3930 | ||
64769240 | 3931 | static const struct address_space_operations ext4_da_aops = { |
8ab22b9a HH |
3932 | .readpage = ext4_readpage, |
3933 | .readpages = ext4_readpages, | |
43ce1d23 | 3934 | .writepage = ext4_writepage, |
20970ba6 | 3935 | .writepages = ext4_writepages, |
8ab22b9a HH |
3936 | .write_begin = ext4_da_write_begin, |
3937 | .write_end = ext4_da_write_end, | |
6dcc693b | 3938 | .set_page_dirty = ext4_set_page_dirty, |
8ab22b9a HH |
3939 | .bmap = ext4_bmap, |
3940 | .invalidatepage = ext4_da_invalidatepage, | |
3941 | .releasepage = ext4_releasepage, | |
3942 | .direct_IO = ext4_direct_IO, | |
3943 | .migratepage = buffer_migrate_page, | |
3944 | .is_partially_uptodate = block_is_partially_uptodate, | |
aa261f54 | 3945 | .error_remove_page = generic_error_remove_page, |
64769240 AT |
3946 | }; |
3947 | ||
5f0663bb DW |
3948 | static const struct address_space_operations ext4_dax_aops = { |
3949 | .writepages = ext4_dax_writepages, | |
3950 | .direct_IO = noop_direct_IO, | |
3951 | .set_page_dirty = noop_set_page_dirty, | |
3952 | .invalidatepage = noop_invalidatepage, | |
3953 | }; | |
3954 | ||
617ba13b | 3955 | void ext4_set_aops(struct inode *inode) |
ac27a0ec | 3956 | { |
3d2b1582 LC |
3957 | switch (ext4_inode_journal_mode(inode)) { |
3958 | case EXT4_INODE_ORDERED_DATA_MODE: | |
3d2b1582 | 3959 | case EXT4_INODE_WRITEBACK_DATA_MODE: |
3d2b1582 LC |
3960 | break; |
3961 | case EXT4_INODE_JOURNAL_DATA_MODE: | |
617ba13b | 3962 | inode->i_mapping->a_ops = &ext4_journalled_aops; |
74d553aa | 3963 | return; |
3d2b1582 LC |
3964 | default: |
3965 | BUG(); | |
3966 | } | |
5f0663bb DW |
3967 | if (IS_DAX(inode)) |
3968 | inode->i_mapping->a_ops = &ext4_dax_aops; | |
3969 | else if (test_opt(inode->i_sb, DELALLOC)) | |
74d553aa TT |
3970 | inode->i_mapping->a_ops = &ext4_da_aops; |
3971 | else | |
3972 | inode->i_mapping->a_ops = &ext4_aops; | |
ac27a0ec DK |
3973 | } |
3974 | ||
923ae0ff | 3975 | static int __ext4_block_zero_page_range(handle_t *handle, |
d863dc36 LC |
3976 | struct address_space *mapping, loff_t from, loff_t length) |
3977 | { | |
09cbfeaf KS |
3978 | ext4_fsblk_t index = from >> PAGE_SHIFT; |
3979 | unsigned offset = from & (PAGE_SIZE-1); | |
923ae0ff | 3980 | unsigned blocksize, pos; |
d863dc36 LC |
3981 | ext4_lblk_t iblock; |
3982 | struct inode *inode = mapping->host; | |
3983 | struct buffer_head *bh; | |
3984 | struct page *page; | |
3985 | int err = 0; | |
3986 | ||
09cbfeaf | 3987 | page = find_or_create_page(mapping, from >> PAGE_SHIFT, |
c62d2555 | 3988 | mapping_gfp_constraint(mapping, ~__GFP_FS)); |
d863dc36 LC |
3989 | if (!page) |
3990 | return -ENOMEM; | |
3991 | ||
3992 | blocksize = inode->i_sb->s_blocksize; | |
d863dc36 | 3993 | |
09cbfeaf | 3994 | iblock = index << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits); |
d863dc36 LC |
3995 | |
3996 | if (!page_has_buffers(page)) | |
3997 | create_empty_buffers(page, blocksize, 0); | |
3998 | ||
3999 | /* Find the buffer that contains "offset" */ | |
4000 | bh = page_buffers(page); | |
4001 | pos = blocksize; | |
4002 | while (offset >= pos) { | |
4003 | bh = bh->b_this_page; | |
4004 | iblock++; | |
4005 | pos += blocksize; | |
4006 | } | |
d863dc36 LC |
4007 | if (buffer_freed(bh)) { |
4008 | BUFFER_TRACE(bh, "freed: skip"); | |
4009 | goto unlock; | |
4010 | } | |
d863dc36 LC |
4011 | if (!buffer_mapped(bh)) { |
4012 | BUFFER_TRACE(bh, "unmapped"); | |
4013 | ext4_get_block(inode, iblock, bh, 0); | |
4014 | /* unmapped? It's a hole - nothing to do */ | |
4015 | if (!buffer_mapped(bh)) { | |
4016 | BUFFER_TRACE(bh, "still unmapped"); | |
4017 | goto unlock; | |
4018 | } | |
4019 | } | |
4020 | ||
4021 | /* Ok, it's mapped. Make sure it's up-to-date */ | |
4022 | if (PageUptodate(page)) | |
4023 | set_buffer_uptodate(bh); | |
4024 | ||
4025 | if (!buffer_uptodate(bh)) { | |
4026 | err = -EIO; | |
dfec8a14 | 4027 | ll_rw_block(REQ_OP_READ, 0, 1, &bh); |
d863dc36 LC |
4028 | wait_on_buffer(bh); |
4029 | /* Uhhuh. Read error. Complain and punt. */ | |
4030 | if (!buffer_uptodate(bh)) | |
4031 | goto unlock; | |
c9c7429c MH |
4032 | if (S_ISREG(inode->i_mode) && |
4033 | ext4_encrypted_inode(inode)) { | |
4034 | /* We expect the key to be set. */ | |
a7550b30 | 4035 | BUG_ON(!fscrypt_has_encryption_key(inode)); |
09cbfeaf | 4036 | BUG_ON(blocksize != PAGE_SIZE); |
b50f7b26 | 4037 | WARN_ON_ONCE(fscrypt_decrypt_page(page->mapping->host, |
9c4bb8a3 | 4038 | page, PAGE_SIZE, 0, page->index)); |
c9c7429c | 4039 | } |
d863dc36 | 4040 | } |
d863dc36 LC |
4041 | if (ext4_should_journal_data(inode)) { |
4042 | BUFFER_TRACE(bh, "get write access"); | |
4043 | err = ext4_journal_get_write_access(handle, bh); | |
4044 | if (err) | |
4045 | goto unlock; | |
4046 | } | |
d863dc36 | 4047 | zero_user(page, offset, length); |
d863dc36 LC |
4048 | BUFFER_TRACE(bh, "zeroed end of block"); |
4049 | ||
d863dc36 LC |
4050 | if (ext4_should_journal_data(inode)) { |
4051 | err = ext4_handle_dirty_metadata(handle, inode, bh); | |
0713ed0c | 4052 | } else { |
353eefd3 | 4053 | err = 0; |
d863dc36 | 4054 | mark_buffer_dirty(bh); |
3957ef53 | 4055 | if (ext4_should_order_data(inode)) |
ee0876bc | 4056 | err = ext4_jbd2_inode_add_write(handle, inode); |
0713ed0c | 4057 | } |
d863dc36 LC |
4058 | |
4059 | unlock: | |
4060 | unlock_page(page); | |
09cbfeaf | 4061 | put_page(page); |
d863dc36 LC |
4062 | return err; |
4063 | } | |
4064 | ||
923ae0ff RZ |
4065 | /* |
4066 | * ext4_block_zero_page_range() zeros out a mapping of length 'length' | |
4067 | * starting from file offset 'from'. The range to be zero'd must | |
4068 | * be contained with in one block. If the specified range exceeds | |
4069 | * the end of the block it will be shortened to end of the block | |
4070 | * that cooresponds to 'from' | |
4071 | */ | |
4072 | static int ext4_block_zero_page_range(handle_t *handle, | |
4073 | struct address_space *mapping, loff_t from, loff_t length) | |
4074 | { | |
4075 | struct inode *inode = mapping->host; | |
09cbfeaf | 4076 | unsigned offset = from & (PAGE_SIZE-1); |
923ae0ff RZ |
4077 | unsigned blocksize = inode->i_sb->s_blocksize; |
4078 | unsigned max = blocksize - (offset & (blocksize - 1)); | |
4079 | ||
4080 | /* | |
4081 | * correct length if it does not fall between | |
4082 | * 'from' and the end of the block | |
4083 | */ | |
4084 | if (length > max || length < 0) | |
4085 | length = max; | |
4086 | ||
47e69351 JK |
4087 | if (IS_DAX(inode)) { |
4088 | return iomap_zero_range(inode, from, length, NULL, | |
4089 | &ext4_iomap_ops); | |
4090 | } | |
923ae0ff RZ |
4091 | return __ext4_block_zero_page_range(handle, mapping, from, length); |
4092 | } | |
4093 | ||
94350ab5 MW |
4094 | /* |
4095 | * ext4_block_truncate_page() zeroes out a mapping from file offset `from' | |
4096 | * up to the end of the block which corresponds to `from'. | |
4097 | * This required during truncate. We need to physically zero the tail end | |
4098 | * of that block so it doesn't yield old data if the file is later grown. | |
4099 | */ | |
c197855e | 4100 | static int ext4_block_truncate_page(handle_t *handle, |
94350ab5 MW |
4101 | struct address_space *mapping, loff_t from) |
4102 | { | |
09cbfeaf | 4103 | unsigned offset = from & (PAGE_SIZE-1); |
94350ab5 MW |
4104 | unsigned length; |
4105 | unsigned blocksize; | |
4106 | struct inode *inode = mapping->host; | |
4107 | ||
0d06863f TT |
4108 | /* If we are processing an encrypted inode during orphan list handling */ |
4109 | if (ext4_encrypted_inode(inode) && !fscrypt_has_encryption_key(inode)) | |
4110 | return 0; | |
4111 | ||
94350ab5 MW |
4112 | blocksize = inode->i_sb->s_blocksize; |
4113 | length = blocksize - (offset & (blocksize - 1)); | |
4114 | ||
4115 | return ext4_block_zero_page_range(handle, mapping, from, length); | |
4116 | } | |
4117 | ||
a87dd18c LC |
4118 | int ext4_zero_partial_blocks(handle_t *handle, struct inode *inode, |
4119 | loff_t lstart, loff_t length) | |
4120 | { | |
4121 | struct super_block *sb = inode->i_sb; | |
4122 | struct address_space *mapping = inode->i_mapping; | |
e1be3a92 | 4123 | unsigned partial_start, partial_end; |
a87dd18c LC |
4124 | ext4_fsblk_t start, end; |
4125 | loff_t byte_end = (lstart + length - 1); | |
4126 | int err = 0; | |
4127 | ||
e1be3a92 LC |
4128 | partial_start = lstart & (sb->s_blocksize - 1); |
4129 | partial_end = byte_end & (sb->s_blocksize - 1); | |
4130 | ||
a87dd18c LC |
4131 | start = lstart >> sb->s_blocksize_bits; |
4132 | end = byte_end >> sb->s_blocksize_bits; | |
4133 | ||
4134 | /* Handle partial zero within the single block */ | |
e1be3a92 LC |
4135 | if (start == end && |
4136 | (partial_start || (partial_end != sb->s_blocksize - 1))) { | |
a87dd18c LC |
4137 | err = ext4_block_zero_page_range(handle, mapping, |
4138 | lstart, length); | |
4139 | return err; | |
4140 | } | |
4141 | /* Handle partial zero out on the start of the range */ | |
e1be3a92 | 4142 | if (partial_start) { |
a87dd18c LC |
4143 | err = ext4_block_zero_page_range(handle, mapping, |
4144 | lstart, sb->s_blocksize); | |
4145 | if (err) | |
4146 | return err; | |
4147 | } | |
4148 | /* Handle partial zero out on the end of the range */ | |
e1be3a92 | 4149 | if (partial_end != sb->s_blocksize - 1) |
a87dd18c | 4150 | err = ext4_block_zero_page_range(handle, mapping, |
e1be3a92 LC |
4151 | byte_end - partial_end, |
4152 | partial_end + 1); | |
a87dd18c LC |
4153 | return err; |
4154 | } | |
4155 | ||
91ef4caf DG |
4156 | int ext4_can_truncate(struct inode *inode) |
4157 | { | |
91ef4caf DG |
4158 | if (S_ISREG(inode->i_mode)) |
4159 | return 1; | |
4160 | if (S_ISDIR(inode->i_mode)) | |
4161 | return 1; | |
4162 | if (S_ISLNK(inode->i_mode)) | |
4163 | return !ext4_inode_is_fast_symlink(inode); | |
4164 | return 0; | |
4165 | } | |
4166 | ||
01127848 JK |
4167 | /* |
4168 | * We have to make sure i_disksize gets properly updated before we truncate | |
4169 | * page cache due to hole punching or zero range. Otherwise i_disksize update | |
4170 | * can get lost as it may have been postponed to submission of writeback but | |
4171 | * that will never happen after we truncate page cache. | |
4172 | */ | |
4173 | int ext4_update_disksize_before_punch(struct inode *inode, loff_t offset, | |
4174 | loff_t len) | |
4175 | { | |
4176 | handle_t *handle; | |
4177 | loff_t size = i_size_read(inode); | |
4178 | ||
5955102c | 4179 | WARN_ON(!inode_is_locked(inode)); |
01127848 JK |
4180 | if (offset > size || offset + len < size) |
4181 | return 0; | |
4182 | ||
4183 | if (EXT4_I(inode)->i_disksize >= size) | |
4184 | return 0; | |
4185 | ||
4186 | handle = ext4_journal_start(inode, EXT4_HT_MISC, 1); | |
4187 | if (IS_ERR(handle)) | |
4188 | return PTR_ERR(handle); | |
4189 | ext4_update_i_disksize(inode, size); | |
4190 | ext4_mark_inode_dirty(handle, inode); | |
4191 | ext4_journal_stop(handle); | |
4192 | ||
4193 | return 0; | |
4194 | } | |
4195 | ||
a4bb6b64 | 4196 | /* |
cca32b7e | 4197 | * ext4_punch_hole: punches a hole in a file by releasing the blocks |
a4bb6b64 AH |
4198 | * associated with the given offset and length |
4199 | * | |
4200 | * @inode: File inode | |
4201 | * @offset: The offset where the hole will begin | |
4202 | * @len: The length of the hole | |
4203 | * | |
4907cb7b | 4204 | * Returns: 0 on success or negative on failure |
a4bb6b64 AH |
4205 | */ |
4206 | ||
aeb2817a | 4207 | int ext4_punch_hole(struct inode *inode, loff_t offset, loff_t length) |
a4bb6b64 | 4208 | { |
26a4c0c6 TT |
4209 | struct super_block *sb = inode->i_sb; |
4210 | ext4_lblk_t first_block, stop_block; | |
4211 | struct address_space *mapping = inode->i_mapping; | |
a87dd18c | 4212 | loff_t first_block_offset, last_block_offset; |
26a4c0c6 TT |
4213 | handle_t *handle; |
4214 | unsigned int credits; | |
4215 | int ret = 0; | |
4216 | ||
a4bb6b64 | 4217 | if (!S_ISREG(inode->i_mode)) |
73355192 | 4218 | return -EOPNOTSUPP; |
a4bb6b64 | 4219 | |
b8a86845 | 4220 | trace_ext4_punch_hole(inode, offset, length, 0); |
aaddea81 | 4221 | |
26a4c0c6 TT |
4222 | /* |
4223 | * Write out all dirty pages to avoid race conditions | |
4224 | * Then release them. | |
4225 | */ | |
cca32b7e | 4226 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { |
26a4c0c6 TT |
4227 | ret = filemap_write_and_wait_range(mapping, offset, |
4228 | offset + length - 1); | |
4229 | if (ret) | |
4230 | return ret; | |
4231 | } | |
4232 | ||
5955102c | 4233 | inode_lock(inode); |
9ef06cec | 4234 | |
26a4c0c6 TT |
4235 | /* No need to punch hole beyond i_size */ |
4236 | if (offset >= inode->i_size) | |
4237 | goto out_mutex; | |
4238 | ||
4239 | /* | |
4240 | * If the hole extends beyond i_size, set the hole | |
4241 | * to end after the page that contains i_size | |
4242 | */ | |
4243 | if (offset + length > inode->i_size) { | |
4244 | length = inode->i_size + | |
09cbfeaf | 4245 | PAGE_SIZE - (inode->i_size & (PAGE_SIZE - 1)) - |
26a4c0c6 TT |
4246 | offset; |
4247 | } | |
4248 | ||
a361293f JK |
4249 | if (offset & (sb->s_blocksize - 1) || |
4250 | (offset + length) & (sb->s_blocksize - 1)) { | |
4251 | /* | |
4252 | * Attach jinode to inode for jbd2 if we do any zeroing of | |
4253 | * partial block | |
4254 | */ | |
4255 | ret = ext4_inode_attach_jinode(inode); | |
4256 | if (ret < 0) | |
4257 | goto out_mutex; | |
4258 | ||
4259 | } | |
4260 | ||
ea3d7209 | 4261 | /* Wait all existing dio workers, newcomers will block on i_mutex */ |
ea3d7209 JK |
4262 | inode_dio_wait(inode); |
4263 | ||
4264 | /* | |
4265 | * Prevent page faults from reinstantiating pages we have released from | |
4266 | * page cache. | |
4267 | */ | |
4268 | down_write(&EXT4_I(inode)->i_mmap_sem); | |
a87dd18c LC |
4269 | first_block_offset = round_up(offset, sb->s_blocksize); |
4270 | last_block_offset = round_down((offset + length), sb->s_blocksize) - 1; | |
26a4c0c6 | 4271 | |
a87dd18c | 4272 | /* Now release the pages and zero block aligned part of pages*/ |
01127848 JK |
4273 | if (last_block_offset > first_block_offset) { |
4274 | ret = ext4_update_disksize_before_punch(inode, offset, length); | |
4275 | if (ret) | |
4276 | goto out_dio; | |
a87dd18c LC |
4277 | truncate_pagecache_range(inode, first_block_offset, |
4278 | last_block_offset); | |
01127848 | 4279 | } |
26a4c0c6 TT |
4280 | |
4281 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
4282 | credits = ext4_writepage_trans_blocks(inode); | |
4283 | else | |
4284 | credits = ext4_blocks_for_truncate(inode); | |
4285 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); | |
4286 | if (IS_ERR(handle)) { | |
4287 | ret = PTR_ERR(handle); | |
4288 | ext4_std_error(sb, ret); | |
4289 | goto out_dio; | |
4290 | } | |
4291 | ||
a87dd18c LC |
4292 | ret = ext4_zero_partial_blocks(handle, inode, offset, |
4293 | length); | |
4294 | if (ret) | |
4295 | goto out_stop; | |
26a4c0c6 TT |
4296 | |
4297 | first_block = (offset + sb->s_blocksize - 1) >> | |
4298 | EXT4_BLOCK_SIZE_BITS(sb); | |
4299 | stop_block = (offset + length) >> EXT4_BLOCK_SIZE_BITS(sb); | |
4300 | ||
4301 | /* If there are no blocks to remove, return now */ | |
4302 | if (first_block >= stop_block) | |
4303 | goto out_stop; | |
4304 | ||
4305 | down_write(&EXT4_I(inode)->i_data_sem); | |
4306 | ext4_discard_preallocations(inode); | |
4307 | ||
4308 | ret = ext4_es_remove_extent(inode, first_block, | |
4309 | stop_block - first_block); | |
4310 | if (ret) { | |
4311 | up_write(&EXT4_I(inode)->i_data_sem); | |
4312 | goto out_stop; | |
4313 | } | |
4314 | ||
4315 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) | |
4316 | ret = ext4_ext_remove_space(inode, first_block, | |
4317 | stop_block - 1); | |
4318 | else | |
4f579ae7 | 4319 | ret = ext4_ind_remove_space(handle, inode, first_block, |
26a4c0c6 TT |
4320 | stop_block); |
4321 | ||
819c4920 | 4322 | up_write(&EXT4_I(inode)->i_data_sem); |
26a4c0c6 TT |
4323 | if (IS_SYNC(inode)) |
4324 | ext4_handle_sync(handle); | |
e251f9bc | 4325 | |
eeca7ea1 | 4326 | inode->i_mtime = inode->i_ctime = current_time(inode); |
26a4c0c6 | 4327 | ext4_mark_inode_dirty(handle, inode); |
67a7d5f5 JK |
4328 | if (ret >= 0) |
4329 | ext4_update_inode_fsync_trans(handle, inode, 1); | |
26a4c0c6 TT |
4330 | out_stop: |
4331 | ext4_journal_stop(handle); | |
4332 | out_dio: | |
ea3d7209 | 4333 | up_write(&EXT4_I(inode)->i_mmap_sem); |
26a4c0c6 | 4334 | out_mutex: |
5955102c | 4335 | inode_unlock(inode); |
26a4c0c6 | 4336 | return ret; |
a4bb6b64 AH |
4337 | } |
4338 | ||
a361293f JK |
4339 | int ext4_inode_attach_jinode(struct inode *inode) |
4340 | { | |
4341 | struct ext4_inode_info *ei = EXT4_I(inode); | |
4342 | struct jbd2_inode *jinode; | |
4343 | ||
4344 | if (ei->jinode || !EXT4_SB(inode->i_sb)->s_journal) | |
4345 | return 0; | |
4346 | ||
4347 | jinode = jbd2_alloc_inode(GFP_KERNEL); | |
4348 | spin_lock(&inode->i_lock); | |
4349 | if (!ei->jinode) { | |
4350 | if (!jinode) { | |
4351 | spin_unlock(&inode->i_lock); | |
4352 | return -ENOMEM; | |
4353 | } | |
4354 | ei->jinode = jinode; | |
4355 | jbd2_journal_init_jbd_inode(ei->jinode, inode); | |
4356 | jinode = NULL; | |
4357 | } | |
4358 | spin_unlock(&inode->i_lock); | |
4359 | if (unlikely(jinode != NULL)) | |
4360 | jbd2_free_inode(jinode); | |
4361 | return 0; | |
4362 | } | |
4363 | ||
ac27a0ec | 4364 | /* |
617ba13b | 4365 | * ext4_truncate() |
ac27a0ec | 4366 | * |
617ba13b MC |
4367 | * We block out ext4_get_block() block instantiations across the entire |
4368 | * transaction, and VFS/VM ensures that ext4_truncate() cannot run | |
ac27a0ec DK |
4369 | * simultaneously on behalf of the same inode. |
4370 | * | |
42b2aa86 | 4371 | * As we work through the truncate and commit bits of it to the journal there |
ac27a0ec DK |
4372 | * is one core, guiding principle: the file's tree must always be consistent on |
4373 | * disk. We must be able to restart the truncate after a crash. | |
4374 | * | |
4375 | * The file's tree may be transiently inconsistent in memory (although it | |
4376 | * probably isn't), but whenever we close off and commit a journal transaction, | |
4377 | * the contents of (the filesystem + the journal) must be consistent and | |
4378 | * restartable. It's pretty simple, really: bottom up, right to left (although | |
4379 | * left-to-right works OK too). | |
4380 | * | |
4381 | * Note that at recovery time, journal replay occurs *before* the restart of | |
4382 | * truncate against the orphan inode list. | |
4383 | * | |
4384 | * The committed inode has the new, desired i_size (which is the same as | |
617ba13b | 4385 | * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see |
ac27a0ec | 4386 | * that this inode's truncate did not complete and it will again call |
617ba13b MC |
4387 | * ext4_truncate() to have another go. So there will be instantiated blocks |
4388 | * to the right of the truncation point in a crashed ext4 filesystem. But | |
ac27a0ec | 4389 | * that's fine - as long as they are linked from the inode, the post-crash |
617ba13b | 4390 | * ext4_truncate() run will find them and release them. |
ac27a0ec | 4391 | */ |
2c98eb5e | 4392 | int ext4_truncate(struct inode *inode) |
ac27a0ec | 4393 | { |
819c4920 TT |
4394 | struct ext4_inode_info *ei = EXT4_I(inode); |
4395 | unsigned int credits; | |
2c98eb5e | 4396 | int err = 0; |
819c4920 TT |
4397 | handle_t *handle; |
4398 | struct address_space *mapping = inode->i_mapping; | |
819c4920 | 4399 | |
19b5ef61 TT |
4400 | /* |
4401 | * There is a possibility that we're either freeing the inode | |
e04027e8 | 4402 | * or it's a completely new inode. In those cases we might not |
19b5ef61 TT |
4403 | * have i_mutex locked because it's not necessary. |
4404 | */ | |
4405 | if (!(inode->i_state & (I_NEW|I_FREEING))) | |
5955102c | 4406 | WARN_ON(!inode_is_locked(inode)); |
0562e0ba JZ |
4407 | trace_ext4_truncate_enter(inode); |
4408 | ||
91ef4caf | 4409 | if (!ext4_can_truncate(inode)) |
2c98eb5e | 4410 | return 0; |
ac27a0ec | 4411 | |
12e9b892 | 4412 | ext4_clear_inode_flag(inode, EXT4_INODE_EOFBLOCKS); |
c8d46e41 | 4413 | |
5534fb5b | 4414 | if (inode->i_size == 0 && !test_opt(inode->i_sb, NO_AUTO_DA_ALLOC)) |
19f5fb7a | 4415 | ext4_set_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE); |
7d8f9f7d | 4416 | |
aef1c851 TM |
4417 | if (ext4_has_inline_data(inode)) { |
4418 | int has_inline = 1; | |
4419 | ||
01daf945 TT |
4420 | err = ext4_inline_data_truncate(inode, &has_inline); |
4421 | if (err) | |
4422 | return err; | |
aef1c851 | 4423 | if (has_inline) |
2c98eb5e | 4424 | return 0; |
aef1c851 TM |
4425 | } |
4426 | ||
a361293f JK |
4427 | /* If we zero-out tail of the page, we have to create jinode for jbd2 */ |
4428 | if (inode->i_size & (inode->i_sb->s_blocksize - 1)) { | |
4429 | if (ext4_inode_attach_jinode(inode) < 0) | |
2c98eb5e | 4430 | return 0; |
a361293f JK |
4431 | } |
4432 | ||
819c4920 TT |
4433 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
4434 | credits = ext4_writepage_trans_blocks(inode); | |
4435 | else | |
4436 | credits = ext4_blocks_for_truncate(inode); | |
4437 | ||
4438 | handle = ext4_journal_start(inode, EXT4_HT_TRUNCATE, credits); | |
2c98eb5e TT |
4439 | if (IS_ERR(handle)) |
4440 | return PTR_ERR(handle); | |
819c4920 | 4441 | |
eb3544c6 LC |
4442 | if (inode->i_size & (inode->i_sb->s_blocksize - 1)) |
4443 | ext4_block_truncate_page(handle, mapping, inode->i_size); | |
819c4920 TT |
4444 | |
4445 | /* | |
4446 | * We add the inode to the orphan list, so that if this | |
4447 | * truncate spans multiple transactions, and we crash, we will | |
4448 | * resume the truncate when the filesystem recovers. It also | |
4449 | * marks the inode dirty, to catch the new size. | |
4450 | * | |
4451 | * Implication: the file must always be in a sane, consistent | |
4452 | * truncatable state while each transaction commits. | |
4453 | */ | |
2c98eb5e TT |
4454 | err = ext4_orphan_add(handle, inode); |
4455 | if (err) | |
819c4920 TT |
4456 | goto out_stop; |
4457 | ||
4458 | down_write(&EXT4_I(inode)->i_data_sem); | |
4459 | ||
4460 | ext4_discard_preallocations(inode); | |
4461 | ||
ff9893dc | 4462 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) |
d0abb36d | 4463 | err = ext4_ext_truncate(handle, inode); |
ff9893dc | 4464 | else |
819c4920 TT |
4465 | ext4_ind_truncate(handle, inode); |
4466 | ||
4467 | up_write(&ei->i_data_sem); | |
d0abb36d TT |
4468 | if (err) |
4469 | goto out_stop; | |
819c4920 TT |
4470 | |
4471 | if (IS_SYNC(inode)) | |
4472 | ext4_handle_sync(handle); | |
4473 | ||
4474 | out_stop: | |
4475 | /* | |
4476 | * If this was a simple ftruncate() and the file will remain alive, | |
4477 | * then we need to clear up the orphan record which we created above. | |
4478 | * However, if this was a real unlink then we were called by | |
58d86a50 | 4479 | * ext4_evict_inode(), and we allow that function to clean up the |
819c4920 TT |
4480 | * orphan info for us. |
4481 | */ | |
4482 | if (inode->i_nlink) | |
4483 | ext4_orphan_del(handle, inode); | |
4484 | ||
eeca7ea1 | 4485 | inode->i_mtime = inode->i_ctime = current_time(inode); |
819c4920 TT |
4486 | ext4_mark_inode_dirty(handle, inode); |
4487 | ext4_journal_stop(handle); | |
ac27a0ec | 4488 | |
0562e0ba | 4489 | trace_ext4_truncate_exit(inode); |
2c98eb5e | 4490 | return err; |
ac27a0ec DK |
4491 | } |
4492 | ||
ac27a0ec | 4493 | /* |
617ba13b | 4494 | * ext4_get_inode_loc returns with an extra refcount against the inode's |
ac27a0ec DK |
4495 | * underlying buffer_head on success. If 'in_mem' is true, we have all |
4496 | * data in memory that is needed to recreate the on-disk version of this | |
4497 | * inode. | |
4498 | */ | |
617ba13b MC |
4499 | static int __ext4_get_inode_loc(struct inode *inode, |
4500 | struct ext4_iloc *iloc, int in_mem) | |
ac27a0ec | 4501 | { |
240799cd TT |
4502 | struct ext4_group_desc *gdp; |
4503 | struct buffer_head *bh; | |
4504 | struct super_block *sb = inode->i_sb; | |
4505 | ext4_fsblk_t block; | |
4506 | int inodes_per_block, inode_offset; | |
4507 | ||
3a06d778 | 4508 | iloc->bh = NULL; |
240799cd | 4509 | if (!ext4_valid_inum(sb, inode->i_ino)) |
6a797d27 | 4510 | return -EFSCORRUPTED; |
ac27a0ec | 4511 | |
240799cd TT |
4512 | iloc->block_group = (inode->i_ino - 1) / EXT4_INODES_PER_GROUP(sb); |
4513 | gdp = ext4_get_group_desc(sb, iloc->block_group, NULL); | |
4514 | if (!gdp) | |
ac27a0ec DK |
4515 | return -EIO; |
4516 | ||
240799cd TT |
4517 | /* |
4518 | * Figure out the offset within the block group inode table | |
4519 | */ | |
00d09882 | 4520 | inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; |
240799cd TT |
4521 | inode_offset = ((inode->i_ino - 1) % |
4522 | EXT4_INODES_PER_GROUP(sb)); | |
4523 | block = ext4_inode_table(sb, gdp) + (inode_offset / inodes_per_block); | |
4524 | iloc->offset = (inode_offset % inodes_per_block) * EXT4_INODE_SIZE(sb); | |
4525 | ||
4526 | bh = sb_getblk(sb, block); | |
aebf0243 | 4527 | if (unlikely(!bh)) |
860d21e2 | 4528 | return -ENOMEM; |
ac27a0ec DK |
4529 | if (!buffer_uptodate(bh)) { |
4530 | lock_buffer(bh); | |
9c83a923 HK |
4531 | |
4532 | /* | |
4533 | * If the buffer has the write error flag, we have failed | |
4534 | * to write out another inode in the same block. In this | |
4535 | * case, we don't have to read the block because we may | |
4536 | * read the old inode data successfully. | |
4537 | */ | |
4538 | if (buffer_write_io_error(bh) && !buffer_uptodate(bh)) | |
4539 | set_buffer_uptodate(bh); | |
4540 | ||
ac27a0ec DK |
4541 | if (buffer_uptodate(bh)) { |
4542 | /* someone brought it uptodate while we waited */ | |
4543 | unlock_buffer(bh); | |
4544 | goto has_buffer; | |
4545 | } | |
4546 | ||
4547 | /* | |
4548 | * If we have all information of the inode in memory and this | |
4549 | * is the only valid inode in the block, we need not read the | |
4550 | * block. | |
4551 | */ | |
4552 | if (in_mem) { | |
4553 | struct buffer_head *bitmap_bh; | |
240799cd | 4554 | int i, start; |
ac27a0ec | 4555 | |
240799cd | 4556 | start = inode_offset & ~(inodes_per_block - 1); |
ac27a0ec | 4557 | |
240799cd TT |
4558 | /* Is the inode bitmap in cache? */ |
4559 | bitmap_bh = sb_getblk(sb, ext4_inode_bitmap(sb, gdp)); | |
aebf0243 | 4560 | if (unlikely(!bitmap_bh)) |
ac27a0ec DK |
4561 | goto make_io; |
4562 | ||
4563 | /* | |
4564 | * If the inode bitmap isn't in cache then the | |
4565 | * optimisation may end up performing two reads instead | |
4566 | * of one, so skip it. | |
4567 | */ | |
4568 | if (!buffer_uptodate(bitmap_bh)) { | |
4569 | brelse(bitmap_bh); | |
4570 | goto make_io; | |
4571 | } | |
240799cd | 4572 | for (i = start; i < start + inodes_per_block; i++) { |
ac27a0ec DK |
4573 | if (i == inode_offset) |
4574 | continue; | |
617ba13b | 4575 | if (ext4_test_bit(i, bitmap_bh->b_data)) |
ac27a0ec DK |
4576 | break; |
4577 | } | |
4578 | brelse(bitmap_bh); | |
240799cd | 4579 | if (i == start + inodes_per_block) { |
ac27a0ec DK |
4580 | /* all other inodes are free, so skip I/O */ |
4581 | memset(bh->b_data, 0, bh->b_size); | |
4582 | set_buffer_uptodate(bh); | |
4583 | unlock_buffer(bh); | |
4584 | goto has_buffer; | |
4585 | } | |
4586 | } | |
4587 | ||
4588 | make_io: | |
240799cd TT |
4589 | /* |
4590 | * If we need to do any I/O, try to pre-readahead extra | |
4591 | * blocks from the inode table. | |
4592 | */ | |
4593 | if (EXT4_SB(sb)->s_inode_readahead_blks) { | |
4594 | ext4_fsblk_t b, end, table; | |
4595 | unsigned num; | |
0d606e2c | 4596 | __u32 ra_blks = EXT4_SB(sb)->s_inode_readahead_blks; |
240799cd TT |
4597 | |
4598 | table = ext4_inode_table(sb, gdp); | |
b713a5ec | 4599 | /* s_inode_readahead_blks is always a power of 2 */ |
0d606e2c | 4600 | b = block & ~((ext4_fsblk_t) ra_blks - 1); |
240799cd TT |
4601 | if (table > b) |
4602 | b = table; | |
0d606e2c | 4603 | end = b + ra_blks; |
240799cd | 4604 | num = EXT4_INODES_PER_GROUP(sb); |
feb0ab32 | 4605 | if (ext4_has_group_desc_csum(sb)) |
560671a0 | 4606 | num -= ext4_itable_unused_count(sb, gdp); |
240799cd TT |
4607 | table += num / inodes_per_block; |
4608 | if (end > table) | |
4609 | end = table; | |
4610 | while (b <= end) | |
4611 | sb_breadahead(sb, b++); | |
4612 | } | |
4613 | ||
ac27a0ec DK |
4614 | /* |
4615 | * There are other valid inodes in the buffer, this inode | |
4616 | * has in-inode xattrs, or we don't have this inode in memory. | |
4617 | * Read the block from disk. | |
4618 | */ | |
0562e0ba | 4619 | trace_ext4_load_inode(inode); |
ac27a0ec DK |
4620 | get_bh(bh); |
4621 | bh->b_end_io = end_buffer_read_sync; | |
2a222ca9 | 4622 | submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh); |
ac27a0ec DK |
4623 | wait_on_buffer(bh); |
4624 | if (!buffer_uptodate(bh)) { | |
c398eda0 TT |
4625 | EXT4_ERROR_INODE_BLOCK(inode, block, |
4626 | "unable to read itable block"); | |
ac27a0ec DK |
4627 | brelse(bh); |
4628 | return -EIO; | |
4629 | } | |
4630 | } | |
4631 | has_buffer: | |
4632 | iloc->bh = bh; | |
4633 | return 0; | |
4634 | } | |
4635 | ||
617ba13b | 4636 | int ext4_get_inode_loc(struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec DK |
4637 | { |
4638 | /* We have all inode data except xattrs in memory here. */ | |
617ba13b | 4639 | return __ext4_get_inode_loc(inode, iloc, |
19f5fb7a | 4640 | !ext4_test_inode_state(inode, EXT4_STATE_XATTR)); |
ac27a0ec DK |
4641 | } |
4642 | ||
6642586b RZ |
4643 | static bool ext4_should_use_dax(struct inode *inode) |
4644 | { | |
4645 | if (!test_opt(inode->i_sb, DAX)) | |
4646 | return false; | |
4647 | if (!S_ISREG(inode->i_mode)) | |
4648 | return false; | |
4649 | if (ext4_should_journal_data(inode)) | |
4650 | return false; | |
4651 | if (ext4_has_inline_data(inode)) | |
4652 | return false; | |
4653 | if (ext4_encrypted_inode(inode)) | |
4654 | return false; | |
4655 | return true; | |
4656 | } | |
4657 | ||
617ba13b | 4658 | void ext4_set_inode_flags(struct inode *inode) |
ac27a0ec | 4659 | { |
617ba13b | 4660 | unsigned int flags = EXT4_I(inode)->i_flags; |
00a1a053 | 4661 | unsigned int new_fl = 0; |
ac27a0ec | 4662 | |
617ba13b | 4663 | if (flags & EXT4_SYNC_FL) |
00a1a053 | 4664 | new_fl |= S_SYNC; |
617ba13b | 4665 | if (flags & EXT4_APPEND_FL) |
00a1a053 | 4666 | new_fl |= S_APPEND; |
617ba13b | 4667 | if (flags & EXT4_IMMUTABLE_FL) |
00a1a053 | 4668 | new_fl |= S_IMMUTABLE; |
617ba13b | 4669 | if (flags & EXT4_NOATIME_FL) |
00a1a053 | 4670 | new_fl |= S_NOATIME; |
617ba13b | 4671 | if (flags & EXT4_DIRSYNC_FL) |
00a1a053 | 4672 | new_fl |= S_DIRSYNC; |
6642586b | 4673 | if (ext4_should_use_dax(inode)) |
923ae0ff | 4674 | new_fl |= S_DAX; |
2ee6a576 EB |
4675 | if (flags & EXT4_ENCRYPT_FL) |
4676 | new_fl |= S_ENCRYPTED; | |
5f16f322 | 4677 | inode_set_flags(inode, new_fl, |
2ee6a576 EB |
4678 | S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC|S_DAX| |
4679 | S_ENCRYPTED); | |
ac27a0ec DK |
4680 | } |
4681 | ||
0fc1b451 | 4682 | static blkcnt_t ext4_inode_blocks(struct ext4_inode *raw_inode, |
de9a55b8 | 4683 | struct ext4_inode_info *ei) |
0fc1b451 AK |
4684 | { |
4685 | blkcnt_t i_blocks ; | |
8180a562 AK |
4686 | struct inode *inode = &(ei->vfs_inode); |
4687 | struct super_block *sb = inode->i_sb; | |
0fc1b451 | 4688 | |
e2b911c5 | 4689 | if (ext4_has_feature_huge_file(sb)) { |
0fc1b451 AK |
4690 | /* we are using combined 48 bit field */ |
4691 | i_blocks = ((u64)le16_to_cpu(raw_inode->i_blocks_high)) << 32 | | |
4692 | le32_to_cpu(raw_inode->i_blocks_lo); | |
07a03824 | 4693 | if (ext4_test_inode_flag(inode, EXT4_INODE_HUGE_FILE)) { |
8180a562 AK |
4694 | /* i_blocks represent file system block size */ |
4695 | return i_blocks << (inode->i_blkbits - 9); | |
4696 | } else { | |
4697 | return i_blocks; | |
4698 | } | |
0fc1b451 AK |
4699 | } else { |
4700 | return le32_to_cpu(raw_inode->i_blocks_lo); | |
4701 | } | |
4702 | } | |
ff9ddf7e | 4703 | |
152a7b0a TM |
4704 | static inline void ext4_iget_extra_inode(struct inode *inode, |
4705 | struct ext4_inode *raw_inode, | |
4706 | struct ext4_inode_info *ei) | |
4707 | { | |
4708 | __le32 *magic = (void *)raw_inode + | |
4709 | EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize; | |
290ab230 EB |
4710 | if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize + sizeof(__le32) <= |
4711 | EXT4_INODE_SIZE(inode->i_sb) && | |
4712 | *magic == cpu_to_le32(EXT4_XATTR_MAGIC)) { | |
152a7b0a | 4713 | ext4_set_inode_state(inode, EXT4_STATE_XATTR); |
67cf5b09 | 4714 | ext4_find_inline_data_nolock(inode); |
f19d5870 TM |
4715 | } else |
4716 | EXT4_I(inode)->i_inline_off = 0; | |
152a7b0a TM |
4717 | } |
4718 | ||
040cb378 LX |
4719 | int ext4_get_projid(struct inode *inode, kprojid_t *projid) |
4720 | { | |
0b7b7779 | 4721 | if (!ext4_has_feature_project(inode->i_sb)) |
040cb378 LX |
4722 | return -EOPNOTSUPP; |
4723 | *projid = EXT4_I(inode)->i_projid; | |
4724 | return 0; | |
4725 | } | |
4726 | ||
1d1fe1ee | 4727 | struct inode *ext4_iget(struct super_block *sb, unsigned long ino) |
ac27a0ec | 4728 | { |
617ba13b MC |
4729 | struct ext4_iloc iloc; |
4730 | struct ext4_inode *raw_inode; | |
1d1fe1ee | 4731 | struct ext4_inode_info *ei; |
1d1fe1ee | 4732 | struct inode *inode; |
b436b9be | 4733 | journal_t *journal = EXT4_SB(sb)->s_journal; |
1d1fe1ee | 4734 | long ret; |
7e6e1ef4 | 4735 | loff_t size; |
ac27a0ec | 4736 | int block; |
08cefc7a EB |
4737 | uid_t i_uid; |
4738 | gid_t i_gid; | |
040cb378 | 4739 | projid_t i_projid; |
ac27a0ec | 4740 | |
1d1fe1ee DH |
4741 | inode = iget_locked(sb, ino); |
4742 | if (!inode) | |
4743 | return ERR_PTR(-ENOMEM); | |
4744 | if (!(inode->i_state & I_NEW)) | |
4745 | return inode; | |
4746 | ||
4747 | ei = EXT4_I(inode); | |
7dc57615 | 4748 | iloc.bh = NULL; |
ac27a0ec | 4749 | |
1d1fe1ee DH |
4750 | ret = __ext4_get_inode_loc(inode, &iloc, 0); |
4751 | if (ret < 0) | |
ac27a0ec | 4752 | goto bad_inode; |
617ba13b | 4753 | raw_inode = ext4_raw_inode(&iloc); |
814525f4 | 4754 | |
8e4b5eae TT |
4755 | if ((ino == EXT4_ROOT_INO) && (raw_inode->i_links_count == 0)) { |
4756 | EXT4_ERROR_INODE(inode, "root inode unallocated"); | |
4757 | ret = -EFSCORRUPTED; | |
4758 | goto bad_inode; | |
4759 | } | |
4760 | ||
814525f4 DW |
4761 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
4762 | ei->i_extra_isize = le16_to_cpu(raw_inode->i_extra_isize); | |
4763 | if (EXT4_GOOD_OLD_INODE_SIZE + ei->i_extra_isize > | |
2dc8d9e1 EB |
4764 | EXT4_INODE_SIZE(inode->i_sb) || |
4765 | (ei->i_extra_isize & 3)) { | |
4766 | EXT4_ERROR_INODE(inode, | |
4767 | "bad extra_isize %u (inode size %u)", | |
4768 | ei->i_extra_isize, | |
4769 | EXT4_INODE_SIZE(inode->i_sb)); | |
6a797d27 | 4770 | ret = -EFSCORRUPTED; |
814525f4 DW |
4771 | goto bad_inode; |
4772 | } | |
4773 | } else | |
4774 | ei->i_extra_isize = 0; | |
4775 | ||
4776 | /* Precompute checksum seed for inode metadata */ | |
9aa5d32b | 4777 | if (ext4_has_metadata_csum(sb)) { |
814525f4 DW |
4778 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
4779 | __u32 csum; | |
4780 | __le32 inum = cpu_to_le32(inode->i_ino); | |
4781 | __le32 gen = raw_inode->i_generation; | |
4782 | csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum, | |
4783 | sizeof(inum)); | |
4784 | ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen, | |
4785 | sizeof(gen)); | |
4786 | } | |
4787 | ||
4788 | if (!ext4_inode_csum_verify(inode, raw_inode, ei)) { | |
4789 | EXT4_ERROR_INODE(inode, "checksum invalid"); | |
6a797d27 | 4790 | ret = -EFSBADCRC; |
814525f4 DW |
4791 | goto bad_inode; |
4792 | } | |
4793 | ||
ac27a0ec | 4794 | inode->i_mode = le16_to_cpu(raw_inode->i_mode); |
08cefc7a EB |
4795 | i_uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low); |
4796 | i_gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low); | |
0b7b7779 | 4797 | if (ext4_has_feature_project(sb) && |
040cb378 LX |
4798 | EXT4_INODE_SIZE(sb) > EXT4_GOOD_OLD_INODE_SIZE && |
4799 | EXT4_FITS_IN_INODE(raw_inode, ei, i_projid)) | |
4800 | i_projid = (projid_t)le32_to_cpu(raw_inode->i_projid); | |
4801 | else | |
4802 | i_projid = EXT4_DEF_PROJID; | |
4803 | ||
af5bc92d | 4804 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
08cefc7a EB |
4805 | i_uid |= le16_to_cpu(raw_inode->i_uid_high) << 16; |
4806 | i_gid |= le16_to_cpu(raw_inode->i_gid_high) << 16; | |
ac27a0ec | 4807 | } |
08cefc7a EB |
4808 | i_uid_write(inode, i_uid); |
4809 | i_gid_write(inode, i_gid); | |
040cb378 | 4810 | ei->i_projid = make_kprojid(&init_user_ns, i_projid); |
bfe86848 | 4811 | set_nlink(inode, le16_to_cpu(raw_inode->i_links_count)); |
ac27a0ec | 4812 | |
353eb83c | 4813 | ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */ |
67cf5b09 | 4814 | ei->i_inline_off = 0; |
ac27a0ec DK |
4815 | ei->i_dir_start_lookup = 0; |
4816 | ei->i_dtime = le32_to_cpu(raw_inode->i_dtime); | |
4817 | /* We now have enough fields to check if the inode was active or not. | |
4818 | * This is needed because nfsd might try to access dead inodes | |
4819 | * the test is that same one that e2fsck uses | |
4820 | * NeilBrown 1999oct15 | |
4821 | */ | |
4822 | if (inode->i_nlink == 0) { | |
393d1d1d DTB |
4823 | if ((inode->i_mode == 0 || |
4824 | !(EXT4_SB(inode->i_sb)->s_mount_state & EXT4_ORPHAN_FS)) && | |
4825 | ino != EXT4_BOOT_LOADER_INO) { | |
ac27a0ec | 4826 | /* this inode is deleted */ |
1d1fe1ee | 4827 | ret = -ESTALE; |
ac27a0ec DK |
4828 | goto bad_inode; |
4829 | } | |
4830 | /* The only unlinked inodes we let through here have | |
4831 | * valid i_mode and are being read by the orphan | |
4832 | * recovery code: that's fine, we're about to complete | |
393d1d1d DTB |
4833 | * the process of deleting those. |
4834 | * OR it is the EXT4_BOOT_LOADER_INO which is | |
4835 | * not initialized on a new filesystem. */ | |
ac27a0ec | 4836 | } |
ac27a0ec | 4837 | ei->i_flags = le32_to_cpu(raw_inode->i_flags); |
0fc1b451 | 4838 | inode->i_blocks = ext4_inode_blocks(raw_inode, ei); |
7973c0c1 | 4839 | ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl_lo); |
e2b911c5 | 4840 | if (ext4_has_feature_64bit(sb)) |
a1ddeb7e BP |
4841 | ei->i_file_acl |= |
4842 | ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32; | |
e08ac99f | 4843 | inode->i_size = ext4_isize(sb, raw_inode); |
7e6e1ef4 DW |
4844 | if ((size = i_size_read(inode)) < 0) { |
4845 | EXT4_ERROR_INODE(inode, "bad i_size value: %lld", size); | |
4846 | ret = -EFSCORRUPTED; | |
4847 | goto bad_inode; | |
4848 | } | |
ac27a0ec | 4849 | ei->i_disksize = inode->i_size; |
a9e7f447 DM |
4850 | #ifdef CONFIG_QUOTA |
4851 | ei->i_reserved_quota = 0; | |
4852 | #endif | |
ac27a0ec DK |
4853 | inode->i_generation = le32_to_cpu(raw_inode->i_generation); |
4854 | ei->i_block_group = iloc.block_group; | |
a4912123 | 4855 | ei->i_last_alloc_group = ~0; |
ac27a0ec DK |
4856 | /* |
4857 | * NOTE! The in-memory inode i_data array is in little-endian order | |
4858 | * even on big-endian machines: we do NOT byteswap the block numbers! | |
4859 | */ | |
617ba13b | 4860 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
ac27a0ec DK |
4861 | ei->i_data[block] = raw_inode->i_block[block]; |
4862 | INIT_LIST_HEAD(&ei->i_orphan); | |
4863 | ||
b436b9be JK |
4864 | /* |
4865 | * Set transaction id's of transactions that have to be committed | |
4866 | * to finish f[data]sync. We set them to currently running transaction | |
4867 | * as we cannot be sure that the inode or some of its metadata isn't | |
4868 | * part of the transaction - the inode could have been reclaimed and | |
4869 | * now it is reread from disk. | |
4870 | */ | |
4871 | if (journal) { | |
4872 | transaction_t *transaction; | |
4873 | tid_t tid; | |
4874 | ||
a931da6a | 4875 | read_lock(&journal->j_state_lock); |
b436b9be JK |
4876 | if (journal->j_running_transaction) |
4877 | transaction = journal->j_running_transaction; | |
4878 | else | |
4879 | transaction = journal->j_committing_transaction; | |
4880 | if (transaction) | |
4881 | tid = transaction->t_tid; | |
4882 | else | |
4883 | tid = journal->j_commit_sequence; | |
a931da6a | 4884 | read_unlock(&journal->j_state_lock); |
b436b9be JK |
4885 | ei->i_sync_tid = tid; |
4886 | ei->i_datasync_tid = tid; | |
4887 | } | |
4888 | ||
0040d987 | 4889 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
ac27a0ec DK |
4890 | if (ei->i_extra_isize == 0) { |
4891 | /* The extra space is currently unused. Use it. */ | |
2dc8d9e1 | 4892 | BUILD_BUG_ON(sizeof(struct ext4_inode) & 3); |
617ba13b MC |
4893 | ei->i_extra_isize = sizeof(struct ext4_inode) - |
4894 | EXT4_GOOD_OLD_INODE_SIZE; | |
ac27a0ec | 4895 | } else { |
152a7b0a | 4896 | ext4_iget_extra_inode(inode, raw_inode, ei); |
ac27a0ec | 4897 | } |
814525f4 | 4898 | } |
ac27a0ec | 4899 | |
ef7f3835 KS |
4900 | EXT4_INODE_GET_XTIME(i_ctime, inode, raw_inode); |
4901 | EXT4_INODE_GET_XTIME(i_mtime, inode, raw_inode); | |
4902 | EXT4_INODE_GET_XTIME(i_atime, inode, raw_inode); | |
4903 | EXT4_EINODE_GET_XTIME(i_crtime, ei, raw_inode); | |
4904 | ||
ed3654eb | 4905 | if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) { |
ee73f9a5 JL |
4906 | u64 ivers = le32_to_cpu(raw_inode->i_disk_version); |
4907 | ||
c4f65706 TT |
4908 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE) { |
4909 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
ee73f9a5 | 4910 | ivers |= |
c4f65706 TT |
4911 | (__u64)(le32_to_cpu(raw_inode->i_version_hi)) << 32; |
4912 | } | |
ee73f9a5 | 4913 | inode_set_iversion_queried(inode, ivers); |
25ec56b5 JNC |
4914 | } |
4915 | ||
c4b5a614 | 4916 | ret = 0; |
485c26ec | 4917 | if (ei->i_file_acl && |
1032988c | 4918 | !ext4_data_block_valid(EXT4_SB(sb), ei->i_file_acl, 1)) { |
24676da4 TT |
4919 | EXT4_ERROR_INODE(inode, "bad extended attribute block %llu", |
4920 | ei->i_file_acl); | |
6a797d27 | 4921 | ret = -EFSCORRUPTED; |
485c26ec | 4922 | goto bad_inode; |
f19d5870 TM |
4923 | } else if (!ext4_has_inline_data(inode)) { |
4924 | if (ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { | |
4925 | if ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || | |
4926 | (S_ISLNK(inode->i_mode) && | |
4927 | !ext4_inode_is_fast_symlink(inode)))) | |
4928 | /* Validate extent which is part of inode */ | |
4929 | ret = ext4_ext_check_inode(inode); | |
4930 | } else if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || | |
4931 | (S_ISLNK(inode->i_mode) && | |
4932 | !ext4_inode_is_fast_symlink(inode))) { | |
4933 | /* Validate block references which are part of inode */ | |
4934 | ret = ext4_ind_check_inode(inode); | |
4935 | } | |
fe2c8191 | 4936 | } |
567f3e9a | 4937 | if (ret) |
de9a55b8 | 4938 | goto bad_inode; |
7a262f7c | 4939 | |
ac27a0ec | 4940 | if (S_ISREG(inode->i_mode)) { |
617ba13b | 4941 | inode->i_op = &ext4_file_inode_operations; |
be64f884 | 4942 | inode->i_fop = &ext4_file_operations; |
617ba13b | 4943 | ext4_set_aops(inode); |
ac27a0ec | 4944 | } else if (S_ISDIR(inode->i_mode)) { |
617ba13b MC |
4945 | inode->i_op = &ext4_dir_inode_operations; |
4946 | inode->i_fop = &ext4_dir_operations; | |
ac27a0ec | 4947 | } else if (S_ISLNK(inode->i_mode)) { |
a7a67e8a AV |
4948 | if (ext4_encrypted_inode(inode)) { |
4949 | inode->i_op = &ext4_encrypted_symlink_inode_operations; | |
4950 | ext4_set_aops(inode); | |
4951 | } else if (ext4_inode_is_fast_symlink(inode)) { | |
75e7566b | 4952 | inode->i_link = (char *)ei->i_data; |
617ba13b | 4953 | inode->i_op = &ext4_fast_symlink_inode_operations; |
e83c1397 DG |
4954 | nd_terminate_link(ei->i_data, inode->i_size, |
4955 | sizeof(ei->i_data) - 1); | |
4956 | } else { | |
617ba13b MC |
4957 | inode->i_op = &ext4_symlink_inode_operations; |
4958 | ext4_set_aops(inode); | |
ac27a0ec | 4959 | } |
21fc61c7 | 4960 | inode_nohighmem(inode); |
563bdd61 TT |
4961 | } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) || |
4962 | S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) { | |
617ba13b | 4963 | inode->i_op = &ext4_special_inode_operations; |
ac27a0ec DK |
4964 | if (raw_inode->i_block[0]) |
4965 | init_special_inode(inode, inode->i_mode, | |
4966 | old_decode_dev(le32_to_cpu(raw_inode->i_block[0]))); | |
4967 | else | |
4968 | init_special_inode(inode, inode->i_mode, | |
4969 | new_decode_dev(le32_to_cpu(raw_inode->i_block[1]))); | |
393d1d1d DTB |
4970 | } else if (ino == EXT4_BOOT_LOADER_INO) { |
4971 | make_bad_inode(inode); | |
563bdd61 | 4972 | } else { |
6a797d27 | 4973 | ret = -EFSCORRUPTED; |
24676da4 | 4974 | EXT4_ERROR_INODE(inode, "bogus i_mode (%o)", inode->i_mode); |
563bdd61 | 4975 | goto bad_inode; |
ac27a0ec | 4976 | } |
af5bc92d | 4977 | brelse(iloc.bh); |
617ba13b | 4978 | ext4_set_inode_flags(inode); |
dec214d0 | 4979 | |
1d1fe1ee DH |
4980 | unlock_new_inode(inode); |
4981 | return inode; | |
ac27a0ec DK |
4982 | |
4983 | bad_inode: | |
567f3e9a | 4984 | brelse(iloc.bh); |
1d1fe1ee DH |
4985 | iget_failed(inode); |
4986 | return ERR_PTR(ret); | |
ac27a0ec DK |
4987 | } |
4988 | ||
f4bb2981 TT |
4989 | struct inode *ext4_iget_normal(struct super_block *sb, unsigned long ino) |
4990 | { | |
4991 | if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO) | |
6a797d27 | 4992 | return ERR_PTR(-EFSCORRUPTED); |
f4bb2981 TT |
4993 | return ext4_iget(sb, ino); |
4994 | } | |
4995 | ||
0fc1b451 AK |
4996 | static int ext4_inode_blocks_set(handle_t *handle, |
4997 | struct ext4_inode *raw_inode, | |
4998 | struct ext4_inode_info *ei) | |
4999 | { | |
5000 | struct inode *inode = &(ei->vfs_inode); | |
5001 | u64 i_blocks = inode->i_blocks; | |
5002 | struct super_block *sb = inode->i_sb; | |
0fc1b451 AK |
5003 | |
5004 | if (i_blocks <= ~0U) { | |
5005 | /* | |
4907cb7b | 5006 | * i_blocks can be represented in a 32 bit variable |
0fc1b451 AK |
5007 | * as multiple of 512 bytes |
5008 | */ | |
8180a562 | 5009 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 5010 | raw_inode->i_blocks_high = 0; |
84a8dce2 | 5011 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
f287a1a5 TT |
5012 | return 0; |
5013 | } | |
e2b911c5 | 5014 | if (!ext4_has_feature_huge_file(sb)) |
f287a1a5 TT |
5015 | return -EFBIG; |
5016 | ||
5017 | if (i_blocks <= 0xffffffffffffULL) { | |
0fc1b451 AK |
5018 | /* |
5019 | * i_blocks can be represented in a 48 bit variable | |
5020 | * as multiple of 512 bytes | |
5021 | */ | |
8180a562 | 5022 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); |
0fc1b451 | 5023 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); |
84a8dce2 | 5024 | ext4_clear_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
0fc1b451 | 5025 | } else { |
84a8dce2 | 5026 | ext4_set_inode_flag(inode, EXT4_INODE_HUGE_FILE); |
8180a562 AK |
5027 | /* i_block is stored in file system block size */ |
5028 | i_blocks = i_blocks >> (inode->i_blkbits - 9); | |
5029 | raw_inode->i_blocks_lo = cpu_to_le32(i_blocks); | |
5030 | raw_inode->i_blocks_high = cpu_to_le16(i_blocks >> 32); | |
0fc1b451 | 5031 | } |
f287a1a5 | 5032 | return 0; |
0fc1b451 AK |
5033 | } |
5034 | ||
a26f4992 TT |
5035 | struct other_inode { |
5036 | unsigned long orig_ino; | |
5037 | struct ext4_inode *raw_inode; | |
5038 | }; | |
5039 | ||
5040 | static int other_inode_match(struct inode * inode, unsigned long ino, | |
5041 | void *data) | |
5042 | { | |
5043 | struct other_inode *oi = (struct other_inode *) data; | |
5044 | ||
5045 | if ((inode->i_ino != ino) || | |
5046 | (inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW | | |
0e11f644 | 5047 | I_DIRTY_INODE)) || |
a26f4992 TT |
5048 | ((inode->i_state & I_DIRTY_TIME) == 0)) |
5049 | return 0; | |
5050 | spin_lock(&inode->i_lock); | |
5051 | if (((inode->i_state & (I_FREEING | I_WILL_FREE | I_NEW | | |
0e11f644 | 5052 | I_DIRTY_INODE)) == 0) && |
a26f4992 TT |
5053 | (inode->i_state & I_DIRTY_TIME)) { |
5054 | struct ext4_inode_info *ei = EXT4_I(inode); | |
5055 | ||
5056 | inode->i_state &= ~(I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED); | |
5057 | spin_unlock(&inode->i_lock); | |
5058 | ||
5059 | spin_lock(&ei->i_raw_lock); | |
5060 | EXT4_INODE_SET_XTIME(i_ctime, inode, oi->raw_inode); | |
5061 | EXT4_INODE_SET_XTIME(i_mtime, inode, oi->raw_inode); | |
5062 | EXT4_INODE_SET_XTIME(i_atime, inode, oi->raw_inode); | |
5063 | ext4_inode_csum_set(inode, oi->raw_inode, ei); | |
5064 | spin_unlock(&ei->i_raw_lock); | |
5065 | trace_ext4_other_inode_update_time(inode, oi->orig_ino); | |
5066 | return -1; | |
5067 | } | |
5068 | spin_unlock(&inode->i_lock); | |
5069 | return -1; | |
5070 | } | |
5071 | ||
5072 | /* | |
5073 | * Opportunistically update the other time fields for other inodes in | |
5074 | * the same inode table block. | |
5075 | */ | |
5076 | static void ext4_update_other_inodes_time(struct super_block *sb, | |
5077 | unsigned long orig_ino, char *buf) | |
5078 | { | |
5079 | struct other_inode oi; | |
5080 | unsigned long ino; | |
5081 | int i, inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; | |
5082 | int inode_size = EXT4_INODE_SIZE(sb); | |
5083 | ||
5084 | oi.orig_ino = orig_ino; | |
0f0ff9a9 TT |
5085 | /* |
5086 | * Calculate the first inode in the inode table block. Inode | |
5087 | * numbers are one-based. That is, the first inode in a block | |
5088 | * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1). | |
5089 | */ | |
5090 | ino = ((orig_ino - 1) & ~(inodes_per_block - 1)) + 1; | |
a26f4992 TT |
5091 | for (i = 0; i < inodes_per_block; i++, ino++, buf += inode_size) { |
5092 | if (ino == orig_ino) | |
5093 | continue; | |
5094 | oi.raw_inode = (struct ext4_inode *) buf; | |
5095 | (void) find_inode_nowait(sb, ino, other_inode_match, &oi); | |
5096 | } | |
5097 | } | |
5098 | ||
ac27a0ec DK |
5099 | /* |
5100 | * Post the struct inode info into an on-disk inode location in the | |
5101 | * buffer-cache. This gobbles the caller's reference to the | |
5102 | * buffer_head in the inode location struct. | |
5103 | * | |
5104 | * The caller must have write access to iloc->bh. | |
5105 | */ | |
617ba13b | 5106 | static int ext4_do_update_inode(handle_t *handle, |
ac27a0ec | 5107 | struct inode *inode, |
830156c7 | 5108 | struct ext4_iloc *iloc) |
ac27a0ec | 5109 | { |
617ba13b MC |
5110 | struct ext4_inode *raw_inode = ext4_raw_inode(iloc); |
5111 | struct ext4_inode_info *ei = EXT4_I(inode); | |
ac27a0ec | 5112 | struct buffer_head *bh = iloc->bh; |
202ee5df | 5113 | struct super_block *sb = inode->i_sb; |
ac27a0ec | 5114 | int err = 0, rc, block; |
202ee5df | 5115 | int need_datasync = 0, set_large_file = 0; |
08cefc7a EB |
5116 | uid_t i_uid; |
5117 | gid_t i_gid; | |
040cb378 | 5118 | projid_t i_projid; |
ac27a0ec | 5119 | |
202ee5df TT |
5120 | spin_lock(&ei->i_raw_lock); |
5121 | ||
5122 | /* For fields not tracked in the in-memory inode, | |
ac27a0ec | 5123 | * initialise them to zero for new inodes. */ |
19f5fb7a | 5124 | if (ext4_test_inode_state(inode, EXT4_STATE_NEW)) |
617ba13b | 5125 | memset(raw_inode, 0, EXT4_SB(inode->i_sb)->s_inode_size); |
ac27a0ec DK |
5126 | |
5127 | raw_inode->i_mode = cpu_to_le16(inode->i_mode); | |
08cefc7a EB |
5128 | i_uid = i_uid_read(inode); |
5129 | i_gid = i_gid_read(inode); | |
040cb378 | 5130 | i_projid = from_kprojid(&init_user_ns, ei->i_projid); |
af5bc92d | 5131 | if (!(test_opt(inode->i_sb, NO_UID32))) { |
08cefc7a EB |
5132 | raw_inode->i_uid_low = cpu_to_le16(low_16_bits(i_uid)); |
5133 | raw_inode->i_gid_low = cpu_to_le16(low_16_bits(i_gid)); | |
ac27a0ec DK |
5134 | /* |
5135 | * Fix up interoperability with old kernels. Otherwise, old inodes get | |
5136 | * re-used with the upper 16 bits of the uid/gid intact | |
5137 | */ | |
93e3b4e6 DJ |
5138 | if (ei->i_dtime && list_empty(&ei->i_orphan)) { |
5139 | raw_inode->i_uid_high = 0; | |
5140 | raw_inode->i_gid_high = 0; | |
5141 | } else { | |
ac27a0ec | 5142 | raw_inode->i_uid_high = |
08cefc7a | 5143 | cpu_to_le16(high_16_bits(i_uid)); |
ac27a0ec | 5144 | raw_inode->i_gid_high = |
08cefc7a | 5145 | cpu_to_le16(high_16_bits(i_gid)); |
ac27a0ec DK |
5146 | } |
5147 | } else { | |
08cefc7a EB |
5148 | raw_inode->i_uid_low = cpu_to_le16(fs_high2lowuid(i_uid)); |
5149 | raw_inode->i_gid_low = cpu_to_le16(fs_high2lowgid(i_gid)); | |
ac27a0ec DK |
5150 | raw_inode->i_uid_high = 0; |
5151 | raw_inode->i_gid_high = 0; | |
5152 | } | |
5153 | raw_inode->i_links_count = cpu_to_le16(inode->i_nlink); | |
ef7f3835 KS |
5154 | |
5155 | EXT4_INODE_SET_XTIME(i_ctime, inode, raw_inode); | |
5156 | EXT4_INODE_SET_XTIME(i_mtime, inode, raw_inode); | |
5157 | EXT4_INODE_SET_XTIME(i_atime, inode, raw_inode); | |
5158 | EXT4_EINODE_SET_XTIME(i_crtime, ei, raw_inode); | |
5159 | ||
bce92d56 LX |
5160 | err = ext4_inode_blocks_set(handle, raw_inode, ei); |
5161 | if (err) { | |
202ee5df | 5162 | spin_unlock(&ei->i_raw_lock); |
0fc1b451 | 5163 | goto out_brelse; |
202ee5df | 5164 | } |
ac27a0ec | 5165 | raw_inode->i_dtime = cpu_to_le32(ei->i_dtime); |
353eb83c | 5166 | raw_inode->i_flags = cpu_to_le32(ei->i_flags & 0xFFFFFFFF); |
ed3654eb | 5167 | if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) |
a1ddeb7e BP |
5168 | raw_inode->i_file_acl_high = |
5169 | cpu_to_le16(ei->i_file_acl >> 32); | |
7973c0c1 | 5170 | raw_inode->i_file_acl_lo = cpu_to_le32(ei->i_file_acl); |
e08ac99f | 5171 | if (ei->i_disksize != ext4_isize(inode->i_sb, raw_inode)) { |
b71fc079 JK |
5172 | ext4_isize_set(raw_inode, ei->i_disksize); |
5173 | need_datasync = 1; | |
5174 | } | |
a48380f7 | 5175 | if (ei->i_disksize > 0x7fffffffULL) { |
e2b911c5 | 5176 | if (!ext4_has_feature_large_file(sb) || |
a48380f7 | 5177 | EXT4_SB(sb)->s_es->s_rev_level == |
202ee5df TT |
5178 | cpu_to_le32(EXT4_GOOD_OLD_REV)) |
5179 | set_large_file = 1; | |
ac27a0ec DK |
5180 | } |
5181 | raw_inode->i_generation = cpu_to_le32(inode->i_generation); | |
5182 | if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) { | |
5183 | if (old_valid_dev(inode->i_rdev)) { | |
5184 | raw_inode->i_block[0] = | |
5185 | cpu_to_le32(old_encode_dev(inode->i_rdev)); | |
5186 | raw_inode->i_block[1] = 0; | |
5187 | } else { | |
5188 | raw_inode->i_block[0] = 0; | |
5189 | raw_inode->i_block[1] = | |
5190 | cpu_to_le32(new_encode_dev(inode->i_rdev)); | |
5191 | raw_inode->i_block[2] = 0; | |
5192 | } | |
f19d5870 | 5193 | } else if (!ext4_has_inline_data(inode)) { |
de9a55b8 TT |
5194 | for (block = 0; block < EXT4_N_BLOCKS; block++) |
5195 | raw_inode->i_block[block] = ei->i_data[block]; | |
f19d5870 | 5196 | } |
ac27a0ec | 5197 | |
ed3654eb | 5198 | if (likely(!test_opt2(inode->i_sb, HURD_COMPAT))) { |
ee73f9a5 JL |
5199 | u64 ivers = inode_peek_iversion(inode); |
5200 | ||
5201 | raw_inode->i_disk_version = cpu_to_le32(ivers); | |
c4f65706 TT |
5202 | if (ei->i_extra_isize) { |
5203 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_version_hi)) | |
5204 | raw_inode->i_version_hi = | |
ee73f9a5 | 5205 | cpu_to_le32(ivers >> 32); |
c4f65706 TT |
5206 | raw_inode->i_extra_isize = |
5207 | cpu_to_le16(ei->i_extra_isize); | |
5208 | } | |
25ec56b5 | 5209 | } |
040cb378 | 5210 | |
0b7b7779 | 5211 | BUG_ON(!ext4_has_feature_project(inode->i_sb) && |
040cb378 LX |
5212 | i_projid != EXT4_DEF_PROJID); |
5213 | ||
5214 | if (EXT4_INODE_SIZE(inode->i_sb) > EXT4_GOOD_OLD_INODE_SIZE && | |
5215 | EXT4_FITS_IN_INODE(raw_inode, ei, i_projid)) | |
5216 | raw_inode->i_projid = cpu_to_le32(i_projid); | |
5217 | ||
814525f4 | 5218 | ext4_inode_csum_set(inode, raw_inode, ei); |
202ee5df | 5219 | spin_unlock(&ei->i_raw_lock); |
1751e8a6 | 5220 | if (inode->i_sb->s_flags & SB_LAZYTIME) |
a26f4992 TT |
5221 | ext4_update_other_inodes_time(inode->i_sb, inode->i_ino, |
5222 | bh->b_data); | |
202ee5df | 5223 | |
830156c7 | 5224 | BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata"); |
73b50c1c | 5225 | rc = ext4_handle_dirty_metadata(handle, NULL, bh); |
830156c7 FM |
5226 | if (!err) |
5227 | err = rc; | |
19f5fb7a | 5228 | ext4_clear_inode_state(inode, EXT4_STATE_NEW); |
202ee5df | 5229 | if (set_large_file) { |
5d601255 | 5230 | BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get write access"); |
202ee5df TT |
5231 | err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh); |
5232 | if (err) | |
5233 | goto out_brelse; | |
5234 | ext4_update_dynamic_rev(sb); | |
e2b911c5 | 5235 | ext4_set_feature_large_file(sb); |
202ee5df TT |
5236 | ext4_handle_sync(handle); |
5237 | err = ext4_handle_dirty_super(handle, sb); | |
5238 | } | |
b71fc079 | 5239 | ext4_update_inode_fsync_trans(handle, inode, need_datasync); |
ac27a0ec | 5240 | out_brelse: |
af5bc92d | 5241 | brelse(bh); |
617ba13b | 5242 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
5243 | return err; |
5244 | } | |
5245 | ||
5246 | /* | |
617ba13b | 5247 | * ext4_write_inode() |
ac27a0ec DK |
5248 | * |
5249 | * We are called from a few places: | |
5250 | * | |
87f7e416 | 5251 | * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files. |
ac27a0ec | 5252 | * Here, there will be no transaction running. We wait for any running |
4907cb7b | 5253 | * transaction to commit. |
ac27a0ec | 5254 | * |
87f7e416 TT |
5255 | * - Within flush work (sys_sync(), kupdate and such). |
5256 | * We wait on commit, if told to. | |
ac27a0ec | 5257 | * |
87f7e416 TT |
5258 | * - Within iput_final() -> write_inode_now() |
5259 | * We wait on commit, if told to. | |
ac27a0ec DK |
5260 | * |
5261 | * In all cases it is actually safe for us to return without doing anything, | |
5262 | * because the inode has been copied into a raw inode buffer in | |
87f7e416 TT |
5263 | * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL |
5264 | * writeback. | |
ac27a0ec DK |
5265 | * |
5266 | * Note that we are absolutely dependent upon all inode dirtiers doing the | |
5267 | * right thing: they *must* call mark_inode_dirty() after dirtying info in | |
5268 | * which we are interested. | |
5269 | * | |
5270 | * It would be a bug for them to not do this. The code: | |
5271 | * | |
5272 | * mark_inode_dirty(inode) | |
5273 | * stuff(); | |
5274 | * inode->i_size = expr; | |
5275 | * | |
87f7e416 TT |
5276 | * is in error because write_inode() could occur while `stuff()' is running, |
5277 | * and the new i_size will be lost. Plus the inode will no longer be on the | |
5278 | * superblock's dirty inode list. | |
ac27a0ec | 5279 | */ |
a9185b41 | 5280 | int ext4_write_inode(struct inode *inode, struct writeback_control *wbc) |
ac27a0ec | 5281 | { |
91ac6f43 FM |
5282 | int err; |
5283 | ||
87f7e416 | 5284 | if (WARN_ON_ONCE(current->flags & PF_MEMALLOC)) |
ac27a0ec DK |
5285 | return 0; |
5286 | ||
91ac6f43 FM |
5287 | if (EXT4_SB(inode->i_sb)->s_journal) { |
5288 | if (ext4_journal_current_handle()) { | |
5289 | jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n"); | |
5290 | dump_stack(); | |
5291 | return -EIO; | |
5292 | } | |
ac27a0ec | 5293 | |
10542c22 JK |
5294 | /* |
5295 | * No need to force transaction in WB_SYNC_NONE mode. Also | |
5296 | * ext4_sync_fs() will force the commit after everything is | |
5297 | * written. | |
5298 | */ | |
5299 | if (wbc->sync_mode != WB_SYNC_ALL || wbc->for_sync) | |
91ac6f43 FM |
5300 | return 0; |
5301 | ||
5302 | err = ext4_force_commit(inode->i_sb); | |
5303 | } else { | |
5304 | struct ext4_iloc iloc; | |
ac27a0ec | 5305 | |
8b472d73 | 5306 | err = __ext4_get_inode_loc(inode, &iloc, 0); |
91ac6f43 FM |
5307 | if (err) |
5308 | return err; | |
10542c22 JK |
5309 | /* |
5310 | * sync(2) will flush the whole buffer cache. No need to do | |
5311 | * it here separately for each inode. | |
5312 | */ | |
5313 | if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) | |
830156c7 FM |
5314 | sync_dirty_buffer(iloc.bh); |
5315 | if (buffer_req(iloc.bh) && !buffer_uptodate(iloc.bh)) { | |
c398eda0 TT |
5316 | EXT4_ERROR_INODE_BLOCK(inode, iloc.bh->b_blocknr, |
5317 | "IO error syncing inode"); | |
830156c7 FM |
5318 | err = -EIO; |
5319 | } | |
fd2dd9fb | 5320 | brelse(iloc.bh); |
91ac6f43 FM |
5321 | } |
5322 | return err; | |
ac27a0ec DK |
5323 | } |
5324 | ||
53e87268 JK |
5325 | /* |
5326 | * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate | |
5327 | * buffers that are attached to a page stradding i_size and are undergoing | |
5328 | * commit. In that case we have to wait for commit to finish and try again. | |
5329 | */ | |
5330 | static void ext4_wait_for_tail_page_commit(struct inode *inode) | |
5331 | { | |
5332 | struct page *page; | |
5333 | unsigned offset; | |
5334 | journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; | |
5335 | tid_t commit_tid = 0; | |
5336 | int ret; | |
5337 | ||
09cbfeaf | 5338 | offset = inode->i_size & (PAGE_SIZE - 1); |
53e87268 JK |
5339 | /* |
5340 | * All buffers in the last page remain valid? Then there's nothing to | |
ea1754a0 | 5341 | * do. We do the check mainly to optimize the common PAGE_SIZE == |
53e87268 JK |
5342 | * blocksize case |
5343 | */ | |
93407472 | 5344 | if (offset > PAGE_SIZE - i_blocksize(inode)) |
53e87268 JK |
5345 | return; |
5346 | while (1) { | |
5347 | page = find_lock_page(inode->i_mapping, | |
09cbfeaf | 5348 | inode->i_size >> PAGE_SHIFT); |
53e87268 JK |
5349 | if (!page) |
5350 | return; | |
ca99fdd2 | 5351 | ret = __ext4_journalled_invalidatepage(page, offset, |
09cbfeaf | 5352 | PAGE_SIZE - offset); |
53e87268 | 5353 | unlock_page(page); |
09cbfeaf | 5354 | put_page(page); |
53e87268 JK |
5355 | if (ret != -EBUSY) |
5356 | return; | |
5357 | commit_tid = 0; | |
5358 | read_lock(&journal->j_state_lock); | |
5359 | if (journal->j_committing_transaction) | |
5360 | commit_tid = journal->j_committing_transaction->t_tid; | |
5361 | read_unlock(&journal->j_state_lock); | |
5362 | if (commit_tid) | |
5363 | jbd2_log_wait_commit(journal, commit_tid); | |
5364 | } | |
5365 | } | |
5366 | ||
ac27a0ec | 5367 | /* |
617ba13b | 5368 | * ext4_setattr() |
ac27a0ec DK |
5369 | * |
5370 | * Called from notify_change. | |
5371 | * | |
5372 | * We want to trap VFS attempts to truncate the file as soon as | |
5373 | * possible. In particular, we want to make sure that when the VFS | |
5374 | * shrinks i_size, we put the inode on the orphan list and modify | |
5375 | * i_disksize immediately, so that during the subsequent flushing of | |
5376 | * dirty pages and freeing of disk blocks, we can guarantee that any | |
5377 | * commit will leave the blocks being flushed in an unused state on | |
5378 | * disk. (On recovery, the inode will get truncated and the blocks will | |
5379 | * be freed, so we have a strong guarantee that no future commit will | |
5380 | * leave these blocks visible to the user.) | |
5381 | * | |
678aaf48 JK |
5382 | * Another thing we have to assure is that if we are in ordered mode |
5383 | * and inode is still attached to the committing transaction, we must | |
5384 | * we start writeout of all the dirty pages which are being truncated. | |
5385 | * This way we are sure that all the data written in the previous | |
5386 | * transaction are already on disk (truncate waits for pages under | |
5387 | * writeback). | |
5388 | * | |
5389 | * Called with inode->i_mutex down. | |
ac27a0ec | 5390 | */ |
617ba13b | 5391 | int ext4_setattr(struct dentry *dentry, struct iattr *attr) |
ac27a0ec | 5392 | { |
2b0143b5 | 5393 | struct inode *inode = d_inode(dentry); |
ac27a0ec | 5394 | int error, rc = 0; |
3d287de3 | 5395 | int orphan = 0; |
ac27a0ec DK |
5396 | const unsigned int ia_valid = attr->ia_valid; |
5397 | ||
0db1ff22 TT |
5398 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
5399 | return -EIO; | |
5400 | ||
31051c85 | 5401 | error = setattr_prepare(dentry, attr); |
ac27a0ec DK |
5402 | if (error) |
5403 | return error; | |
5404 | ||
3ce2b8dd EB |
5405 | error = fscrypt_prepare_setattr(dentry, attr); |
5406 | if (error) | |
5407 | return error; | |
5408 | ||
a7cdadee JK |
5409 | if (is_quota_modification(inode, attr)) { |
5410 | error = dquot_initialize(inode); | |
5411 | if (error) | |
5412 | return error; | |
5413 | } | |
08cefc7a EB |
5414 | if ((ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) || |
5415 | (ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) { | |
ac27a0ec DK |
5416 | handle_t *handle; |
5417 | ||
5418 | /* (user+group)*(old+new) structure, inode write (sb, | |
5419 | * inode block, ? - but truncate inode update has it) */ | |
9924a92a TT |
5420 | handle = ext4_journal_start(inode, EXT4_HT_QUOTA, |
5421 | (EXT4_MAXQUOTAS_INIT_BLOCKS(inode->i_sb) + | |
5422 | EXT4_MAXQUOTAS_DEL_BLOCKS(inode->i_sb)) + 3); | |
ac27a0ec DK |
5423 | if (IS_ERR(handle)) { |
5424 | error = PTR_ERR(handle); | |
5425 | goto err_out; | |
5426 | } | |
7a9ca53a TE |
5427 | |
5428 | /* dquot_transfer() calls back ext4_get_inode_usage() which | |
5429 | * counts xattr inode references. | |
5430 | */ | |
5431 | down_read(&EXT4_I(inode)->xattr_sem); | |
b43fa828 | 5432 | error = dquot_transfer(inode, attr); |
7a9ca53a TE |
5433 | up_read(&EXT4_I(inode)->xattr_sem); |
5434 | ||
ac27a0ec | 5435 | if (error) { |
617ba13b | 5436 | ext4_journal_stop(handle); |
ac27a0ec DK |
5437 | return error; |
5438 | } | |
5439 | /* Update corresponding info in inode so that everything is in | |
5440 | * one transaction */ | |
5441 | if (attr->ia_valid & ATTR_UID) | |
5442 | inode->i_uid = attr->ia_uid; | |
5443 | if (attr->ia_valid & ATTR_GID) | |
5444 | inode->i_gid = attr->ia_gid; | |
617ba13b MC |
5445 | error = ext4_mark_inode_dirty(handle, inode); |
5446 | ext4_journal_stop(handle); | |
ac27a0ec DK |
5447 | } |
5448 | ||
3da40c7b | 5449 | if (attr->ia_valid & ATTR_SIZE) { |
5208386c | 5450 | handle_t *handle; |
3da40c7b JB |
5451 | loff_t oldsize = inode->i_size; |
5452 | int shrink = (attr->ia_size <= inode->i_size); | |
562c72aa | 5453 | |
12e9b892 | 5454 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) { |
e2b46574 ES |
5455 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
5456 | ||
0c095c7f TT |
5457 | if (attr->ia_size > sbi->s_bitmap_maxbytes) |
5458 | return -EFBIG; | |
e2b46574 | 5459 | } |
3da40c7b JB |
5460 | if (!S_ISREG(inode->i_mode)) |
5461 | return -EINVAL; | |
dff6efc3 CH |
5462 | |
5463 | if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size) | |
5464 | inode_inc_iversion(inode); | |
5465 | ||
3da40c7b | 5466 | if (ext4_should_order_data(inode) && |
5208386c | 5467 | (attr->ia_size < inode->i_size)) { |
3da40c7b | 5468 | error = ext4_begin_ordered_truncate(inode, |
678aaf48 | 5469 | attr->ia_size); |
3da40c7b JB |
5470 | if (error) |
5471 | goto err_out; | |
5472 | } | |
5473 | if (attr->ia_size != inode->i_size) { | |
5208386c JK |
5474 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 3); |
5475 | if (IS_ERR(handle)) { | |
5476 | error = PTR_ERR(handle); | |
5477 | goto err_out; | |
5478 | } | |
3da40c7b | 5479 | if (ext4_handle_valid(handle) && shrink) { |
5208386c JK |
5480 | error = ext4_orphan_add(handle, inode); |
5481 | orphan = 1; | |
5482 | } | |
911af577 EG |
5483 | /* |
5484 | * Update c/mtime on truncate up, ext4_truncate() will | |
5485 | * update c/mtime in shrink case below | |
5486 | */ | |
5487 | if (!shrink) { | |
eeca7ea1 | 5488 | inode->i_mtime = current_time(inode); |
911af577 EG |
5489 | inode->i_ctime = inode->i_mtime; |
5490 | } | |
90e775b7 | 5491 | down_write(&EXT4_I(inode)->i_data_sem); |
5208386c JK |
5492 | EXT4_I(inode)->i_disksize = attr->ia_size; |
5493 | rc = ext4_mark_inode_dirty(handle, inode); | |
5494 | if (!error) | |
5495 | error = rc; | |
90e775b7 JK |
5496 | /* |
5497 | * We have to update i_size under i_data_sem together | |
5498 | * with i_disksize to avoid races with writeback code | |
5499 | * running ext4_wb_update_i_disksize(). | |
5500 | */ | |
5501 | if (!error) | |
5502 | i_size_write(inode, attr->ia_size); | |
5503 | up_write(&EXT4_I(inode)->i_data_sem); | |
5208386c JK |
5504 | ext4_journal_stop(handle); |
5505 | if (error) { | |
3da40c7b JB |
5506 | if (orphan) |
5507 | ext4_orphan_del(NULL, inode); | |
678aaf48 JK |
5508 | goto err_out; |
5509 | } | |
d6320cbf | 5510 | } |
3da40c7b JB |
5511 | if (!shrink) |
5512 | pagecache_isize_extended(inode, oldsize, inode->i_size); | |
53e87268 | 5513 | |
5208386c JK |
5514 | /* |
5515 | * Blocks are going to be removed from the inode. Wait | |
5516 | * for dio in flight. Temporarily disable | |
5517 | * dioread_nolock to prevent livelock. | |
5518 | */ | |
5519 | if (orphan) { | |
5520 | if (!ext4_should_journal_data(inode)) { | |
5208386c | 5521 | inode_dio_wait(inode); |
5208386c JK |
5522 | } else |
5523 | ext4_wait_for_tail_page_commit(inode); | |
1c9114f9 | 5524 | } |
ea3d7209 | 5525 | down_write(&EXT4_I(inode)->i_mmap_sem); |
5208386c JK |
5526 | /* |
5527 | * Truncate pagecache after we've waited for commit | |
5528 | * in data=journal mode to make pages freeable. | |
5529 | */ | |
923ae0ff | 5530 | truncate_pagecache(inode, inode->i_size); |
2c98eb5e TT |
5531 | if (shrink) { |
5532 | rc = ext4_truncate(inode); | |
5533 | if (rc) | |
5534 | error = rc; | |
5535 | } | |
ea3d7209 | 5536 | up_write(&EXT4_I(inode)->i_mmap_sem); |
072bd7ea | 5537 | } |
ac27a0ec | 5538 | |
2c98eb5e | 5539 | if (!error) { |
1025774c CH |
5540 | setattr_copy(inode, attr); |
5541 | mark_inode_dirty(inode); | |
5542 | } | |
5543 | ||
5544 | /* | |
5545 | * If the call to ext4_truncate failed to get a transaction handle at | |
5546 | * all, we need to clean up the in-core orphan list manually. | |
5547 | */ | |
3d287de3 | 5548 | if (orphan && inode->i_nlink) |
617ba13b | 5549 | ext4_orphan_del(NULL, inode); |
ac27a0ec | 5550 | |
2c98eb5e | 5551 | if (!error && (ia_valid & ATTR_MODE)) |
64e178a7 | 5552 | rc = posix_acl_chmod(inode, inode->i_mode); |
ac27a0ec DK |
5553 | |
5554 | err_out: | |
617ba13b | 5555 | ext4_std_error(inode->i_sb, error); |
ac27a0ec DK |
5556 | if (!error) |
5557 | error = rc; | |
5558 | return error; | |
5559 | } | |
5560 | ||
a528d35e DH |
5561 | int ext4_getattr(const struct path *path, struct kstat *stat, |
5562 | u32 request_mask, unsigned int query_flags) | |
3e3398a0 | 5563 | { |
99652ea5 DH |
5564 | struct inode *inode = d_inode(path->dentry); |
5565 | struct ext4_inode *raw_inode; | |
5566 | struct ext4_inode_info *ei = EXT4_I(inode); | |
5567 | unsigned int flags; | |
5568 | ||
5569 | if (EXT4_FITS_IN_INODE(raw_inode, ei, i_crtime)) { | |
5570 | stat->result_mask |= STATX_BTIME; | |
5571 | stat->btime.tv_sec = ei->i_crtime.tv_sec; | |
5572 | stat->btime.tv_nsec = ei->i_crtime.tv_nsec; | |
5573 | } | |
5574 | ||
5575 | flags = ei->i_flags & EXT4_FL_USER_VISIBLE; | |
5576 | if (flags & EXT4_APPEND_FL) | |
5577 | stat->attributes |= STATX_ATTR_APPEND; | |
5578 | if (flags & EXT4_COMPR_FL) | |
5579 | stat->attributes |= STATX_ATTR_COMPRESSED; | |
5580 | if (flags & EXT4_ENCRYPT_FL) | |
5581 | stat->attributes |= STATX_ATTR_ENCRYPTED; | |
5582 | if (flags & EXT4_IMMUTABLE_FL) | |
5583 | stat->attributes |= STATX_ATTR_IMMUTABLE; | |
5584 | if (flags & EXT4_NODUMP_FL) | |
5585 | stat->attributes |= STATX_ATTR_NODUMP; | |
3e3398a0 | 5586 | |
3209f68b DH |
5587 | stat->attributes_mask |= (STATX_ATTR_APPEND | |
5588 | STATX_ATTR_COMPRESSED | | |
5589 | STATX_ATTR_ENCRYPTED | | |
5590 | STATX_ATTR_IMMUTABLE | | |
5591 | STATX_ATTR_NODUMP); | |
5592 | ||
3e3398a0 | 5593 | generic_fillattr(inode, stat); |
99652ea5 DH |
5594 | return 0; |
5595 | } | |
5596 | ||
5597 | int ext4_file_getattr(const struct path *path, struct kstat *stat, | |
5598 | u32 request_mask, unsigned int query_flags) | |
5599 | { | |
5600 | struct inode *inode = d_inode(path->dentry); | |
5601 | u64 delalloc_blocks; | |
5602 | ||
5603 | ext4_getattr(path, stat, request_mask, query_flags); | |
3e3398a0 | 5604 | |
9206c561 AD |
5605 | /* |
5606 | * If there is inline data in the inode, the inode will normally not | |
5607 | * have data blocks allocated (it may have an external xattr block). | |
5608 | * Report at least one sector for such files, so tools like tar, rsync, | |
d67d64f4 | 5609 | * others don't incorrectly think the file is completely sparse. |
9206c561 AD |
5610 | */ |
5611 | if (unlikely(ext4_has_inline_data(inode))) | |
5612 | stat->blocks += (stat->size + 511) >> 9; | |
5613 | ||
3e3398a0 MC |
5614 | /* |
5615 | * We can't update i_blocks if the block allocation is delayed | |
5616 | * otherwise in the case of system crash before the real block | |
5617 | * allocation is done, we will have i_blocks inconsistent with | |
5618 | * on-disk file blocks. | |
5619 | * We always keep i_blocks updated together with real | |
5620 | * allocation. But to not confuse with user, stat | |
5621 | * will return the blocks that include the delayed allocation | |
5622 | * blocks for this file. | |
5623 | */ | |
96607551 | 5624 | delalloc_blocks = EXT4_C2B(EXT4_SB(inode->i_sb), |
9206c561 AD |
5625 | EXT4_I(inode)->i_reserved_data_blocks); |
5626 | stat->blocks += delalloc_blocks << (inode->i_sb->s_blocksize_bits - 9); | |
3e3398a0 MC |
5627 | return 0; |
5628 | } | |
ac27a0ec | 5629 | |
fffb2739 JK |
5630 | static int ext4_index_trans_blocks(struct inode *inode, int lblocks, |
5631 | int pextents) | |
a02908f1 | 5632 | { |
12e9b892 | 5633 | if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) |
fffb2739 JK |
5634 | return ext4_ind_trans_blocks(inode, lblocks); |
5635 | return ext4_ext_index_trans_blocks(inode, pextents); | |
a02908f1 | 5636 | } |
ac51d837 | 5637 | |
ac27a0ec | 5638 | /* |
a02908f1 MC |
5639 | * Account for index blocks, block groups bitmaps and block group |
5640 | * descriptor blocks if modify datablocks and index blocks | |
5641 | * worse case, the indexs blocks spread over different block groups | |
ac27a0ec | 5642 | * |
a02908f1 | 5643 | * If datablocks are discontiguous, they are possible to spread over |
4907cb7b | 5644 | * different block groups too. If they are contiguous, with flexbg, |
a02908f1 | 5645 | * they could still across block group boundary. |
ac27a0ec | 5646 | * |
a02908f1 MC |
5647 | * Also account for superblock, inode, quota and xattr blocks |
5648 | */ | |
dec214d0 | 5649 | static int ext4_meta_trans_blocks(struct inode *inode, int lblocks, |
fffb2739 | 5650 | int pextents) |
a02908f1 | 5651 | { |
8df9675f TT |
5652 | ext4_group_t groups, ngroups = ext4_get_groups_count(inode->i_sb); |
5653 | int gdpblocks; | |
a02908f1 MC |
5654 | int idxblocks; |
5655 | int ret = 0; | |
5656 | ||
5657 | /* | |
fffb2739 JK |
5658 | * How many index blocks need to touch to map @lblocks logical blocks |
5659 | * to @pextents physical extents? | |
a02908f1 | 5660 | */ |
fffb2739 | 5661 | idxblocks = ext4_index_trans_blocks(inode, lblocks, pextents); |
a02908f1 MC |
5662 | |
5663 | ret = idxblocks; | |
5664 | ||
5665 | /* | |
5666 | * Now let's see how many group bitmaps and group descriptors need | |
5667 | * to account | |
5668 | */ | |
fffb2739 | 5669 | groups = idxblocks + pextents; |
a02908f1 | 5670 | gdpblocks = groups; |
8df9675f TT |
5671 | if (groups > ngroups) |
5672 | groups = ngroups; | |
a02908f1 MC |
5673 | if (groups > EXT4_SB(inode->i_sb)->s_gdb_count) |
5674 | gdpblocks = EXT4_SB(inode->i_sb)->s_gdb_count; | |
5675 | ||
5676 | /* bitmaps and block group descriptor blocks */ | |
5677 | ret += groups + gdpblocks; | |
5678 | ||
5679 | /* Blocks for super block, inode, quota and xattr blocks */ | |
5680 | ret += EXT4_META_TRANS_BLOCKS(inode->i_sb); | |
5681 | ||
5682 | return ret; | |
5683 | } | |
5684 | ||
5685 | /* | |
25985edc | 5686 | * Calculate the total number of credits to reserve to fit |
f3bd1f3f MC |
5687 | * the modification of a single pages into a single transaction, |
5688 | * which may include multiple chunks of block allocations. | |
ac27a0ec | 5689 | * |
525f4ed8 | 5690 | * This could be called via ext4_write_begin() |
ac27a0ec | 5691 | * |
525f4ed8 | 5692 | * We need to consider the worse case, when |
a02908f1 | 5693 | * one new block per extent. |
ac27a0ec | 5694 | */ |
a86c6181 | 5695 | int ext4_writepage_trans_blocks(struct inode *inode) |
ac27a0ec | 5696 | { |
617ba13b | 5697 | int bpp = ext4_journal_blocks_per_page(inode); |
ac27a0ec DK |
5698 | int ret; |
5699 | ||
fffb2739 | 5700 | ret = ext4_meta_trans_blocks(inode, bpp, bpp); |
a86c6181 | 5701 | |
a02908f1 | 5702 | /* Account for data blocks for journalled mode */ |
617ba13b | 5703 | if (ext4_should_journal_data(inode)) |
a02908f1 | 5704 | ret += bpp; |
ac27a0ec DK |
5705 | return ret; |
5706 | } | |
f3bd1f3f MC |
5707 | |
5708 | /* | |
5709 | * Calculate the journal credits for a chunk of data modification. | |
5710 | * | |
5711 | * This is called from DIO, fallocate or whoever calling | |
79e83036 | 5712 | * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks. |
f3bd1f3f MC |
5713 | * |
5714 | * journal buffers for data blocks are not included here, as DIO | |
5715 | * and fallocate do no need to journal data buffers. | |
5716 | */ | |
5717 | int ext4_chunk_trans_blocks(struct inode *inode, int nrblocks) | |
5718 | { | |
5719 | return ext4_meta_trans_blocks(inode, nrblocks, 1); | |
5720 | } | |
5721 | ||
ac27a0ec | 5722 | /* |
617ba13b | 5723 | * The caller must have previously called ext4_reserve_inode_write(). |
ac27a0ec DK |
5724 | * Give this, we know that the caller already has write access to iloc->bh. |
5725 | */ | |
617ba13b | 5726 | int ext4_mark_iloc_dirty(handle_t *handle, |
de9a55b8 | 5727 | struct inode *inode, struct ext4_iloc *iloc) |
ac27a0ec DK |
5728 | { |
5729 | int err = 0; | |
5730 | ||
0db1ff22 TT |
5731 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
5732 | return -EIO; | |
5733 | ||
c64db50e | 5734 | if (IS_I_VERSION(inode)) |
25ec56b5 JNC |
5735 | inode_inc_iversion(inode); |
5736 | ||
ac27a0ec DK |
5737 | /* the do_update_inode consumes one bh->b_count */ |
5738 | get_bh(iloc->bh); | |
5739 | ||
dab291af | 5740 | /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */ |
830156c7 | 5741 | err = ext4_do_update_inode(handle, inode, iloc); |
ac27a0ec DK |
5742 | put_bh(iloc->bh); |
5743 | return err; | |
5744 | } | |
5745 | ||
5746 | /* | |
5747 | * On success, We end up with an outstanding reference count against | |
5748 | * iloc->bh. This _must_ be cleaned up later. | |
5749 | */ | |
5750 | ||
5751 | int | |
617ba13b MC |
5752 | ext4_reserve_inode_write(handle_t *handle, struct inode *inode, |
5753 | struct ext4_iloc *iloc) | |
ac27a0ec | 5754 | { |
0390131b FM |
5755 | int err; |
5756 | ||
0db1ff22 TT |
5757 | if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb)))) |
5758 | return -EIO; | |
5759 | ||
0390131b FM |
5760 | err = ext4_get_inode_loc(inode, iloc); |
5761 | if (!err) { | |
5762 | BUFFER_TRACE(iloc->bh, "get_write_access"); | |
5763 | err = ext4_journal_get_write_access(handle, iloc->bh); | |
5764 | if (err) { | |
5765 | brelse(iloc->bh); | |
5766 | iloc->bh = NULL; | |
ac27a0ec DK |
5767 | } |
5768 | } | |
617ba13b | 5769 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
5770 | return err; |
5771 | } | |
5772 | ||
c03b45b8 MX |
5773 | static int __ext4_expand_extra_isize(struct inode *inode, |
5774 | unsigned int new_extra_isize, | |
5775 | struct ext4_iloc *iloc, | |
5776 | handle_t *handle, int *no_expand) | |
5777 | { | |
5778 | struct ext4_inode *raw_inode; | |
5779 | struct ext4_xattr_ibody_header *header; | |
5780 | int error; | |
5781 | ||
5782 | raw_inode = ext4_raw_inode(iloc); | |
5783 | ||
5784 | header = IHDR(inode, raw_inode); | |
5785 | ||
5786 | /* No extended attributes present */ | |
5787 | if (!ext4_test_inode_state(inode, EXT4_STATE_XATTR) || | |
5788 | header->h_magic != cpu_to_le32(EXT4_XATTR_MAGIC)) { | |
5789 | memset((void *)raw_inode + EXT4_GOOD_OLD_INODE_SIZE + | |
5790 | EXT4_I(inode)->i_extra_isize, 0, | |
5791 | new_extra_isize - EXT4_I(inode)->i_extra_isize); | |
5792 | EXT4_I(inode)->i_extra_isize = new_extra_isize; | |
5793 | return 0; | |
5794 | } | |
5795 | ||
5796 | /* try to expand with EAs present */ | |
5797 | error = ext4_expand_extra_isize_ea(inode, new_extra_isize, | |
5798 | raw_inode, handle); | |
5799 | if (error) { | |
5800 | /* | |
5801 | * Inode size expansion failed; don't try again | |
5802 | */ | |
5803 | *no_expand = 1; | |
5804 | } | |
5805 | ||
5806 | return error; | |
5807 | } | |
5808 | ||
6dd4ee7c KS |
5809 | /* |
5810 | * Expand an inode by new_extra_isize bytes. | |
5811 | * Returns 0 on success or negative error number on failure. | |
5812 | */ | |
cf0a5e81 MX |
5813 | static int ext4_try_to_expand_extra_isize(struct inode *inode, |
5814 | unsigned int new_extra_isize, | |
5815 | struct ext4_iloc iloc, | |
5816 | handle_t *handle) | |
6dd4ee7c | 5817 | { |
3b10fdc6 MX |
5818 | int no_expand; |
5819 | int error; | |
6dd4ee7c | 5820 | |
cf0a5e81 MX |
5821 | if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) |
5822 | return -EOVERFLOW; | |
5823 | ||
5824 | /* | |
5825 | * In nojournal mode, we can immediately attempt to expand | |
5826 | * the inode. When journaled, we first need to obtain extra | |
5827 | * buffer credits since we may write into the EA block | |
5828 | * with this same handle. If journal_extend fails, then it will | |
5829 | * only result in a minor loss of functionality for that inode. | |
5830 | * If this is felt to be critical, then e2fsck should be run to | |
5831 | * force a large enough s_min_extra_isize. | |
5832 | */ | |
5833 | if (ext4_handle_valid(handle) && | |
5834 | jbd2_journal_extend(handle, | |
5835 | EXT4_DATA_TRANS_BLOCKS(inode->i_sb)) != 0) | |
5836 | return -ENOSPC; | |
6dd4ee7c | 5837 | |
3b10fdc6 | 5838 | if (ext4_write_trylock_xattr(inode, &no_expand) == 0) |
cf0a5e81 | 5839 | return -EBUSY; |
3b10fdc6 | 5840 | |
c03b45b8 MX |
5841 | error = __ext4_expand_extra_isize(inode, new_extra_isize, &iloc, |
5842 | handle, &no_expand); | |
5843 | ext4_write_unlock_xattr(inode, &no_expand); | |
6dd4ee7c | 5844 | |
c03b45b8 MX |
5845 | return error; |
5846 | } | |
6dd4ee7c | 5847 | |
c03b45b8 MX |
5848 | int ext4_expand_extra_isize(struct inode *inode, |
5849 | unsigned int new_extra_isize, | |
5850 | struct ext4_iloc *iloc) | |
5851 | { | |
5852 | handle_t *handle; | |
5853 | int no_expand; | |
5854 | int error, rc; | |
5855 | ||
5856 | if (ext4_test_inode_state(inode, EXT4_STATE_NO_EXPAND)) { | |
5857 | brelse(iloc->bh); | |
5858 | return -EOVERFLOW; | |
6dd4ee7c KS |
5859 | } |
5860 | ||
c03b45b8 MX |
5861 | handle = ext4_journal_start(inode, EXT4_HT_INODE, |
5862 | EXT4_DATA_TRANS_BLOCKS(inode->i_sb)); | |
5863 | if (IS_ERR(handle)) { | |
5864 | error = PTR_ERR(handle); | |
5865 | brelse(iloc->bh); | |
5866 | return error; | |
5867 | } | |
5868 | ||
5869 | ext4_write_lock_xattr(inode, &no_expand); | |
5870 | ||
5871 | BUFFER_TRACE(iloc.bh, "get_write_access"); | |
5872 | error = ext4_journal_get_write_access(handle, iloc->bh); | |
3b10fdc6 | 5873 | if (error) { |
c03b45b8 MX |
5874 | brelse(iloc->bh); |
5875 | goto out_stop; | |
3b10fdc6 | 5876 | } |
cf0a5e81 | 5877 | |
c03b45b8 MX |
5878 | error = __ext4_expand_extra_isize(inode, new_extra_isize, iloc, |
5879 | handle, &no_expand); | |
5880 | ||
5881 | rc = ext4_mark_iloc_dirty(handle, inode, iloc); | |
5882 | if (!error) | |
5883 | error = rc; | |
5884 | ||
5885 | ext4_write_unlock_xattr(inode, &no_expand); | |
5886 | out_stop: | |
5887 | ext4_journal_stop(handle); | |
3b10fdc6 | 5888 | return error; |
6dd4ee7c KS |
5889 | } |
5890 | ||
ac27a0ec DK |
5891 | /* |
5892 | * What we do here is to mark the in-core inode as clean with respect to inode | |
5893 | * dirtiness (it may still be data-dirty). | |
5894 | * This means that the in-core inode may be reaped by prune_icache | |
5895 | * without having to perform any I/O. This is a very good thing, | |
5896 | * because *any* task may call prune_icache - even ones which | |
5897 | * have a transaction open against a different journal. | |
5898 | * | |
5899 | * Is this cheating? Not really. Sure, we haven't written the | |
5900 | * inode out, but prune_icache isn't a user-visible syncing function. | |
5901 | * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync) | |
5902 | * we start and wait on commits. | |
ac27a0ec | 5903 | */ |
617ba13b | 5904 | int ext4_mark_inode_dirty(handle_t *handle, struct inode *inode) |
ac27a0ec | 5905 | { |
617ba13b | 5906 | struct ext4_iloc iloc; |
6dd4ee7c | 5907 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
cf0a5e81 | 5908 | int err; |
ac27a0ec DK |
5909 | |
5910 | might_sleep(); | |
7ff9c073 | 5911 | trace_ext4_mark_inode_dirty(inode, _RET_IP_); |
617ba13b | 5912 | err = ext4_reserve_inode_write(handle, inode, &iloc); |
5e1021f2 EG |
5913 | if (err) |
5914 | return err; | |
cf0a5e81 MX |
5915 | |
5916 | if (EXT4_I(inode)->i_extra_isize < sbi->s_want_extra_isize) | |
5917 | ext4_try_to_expand_extra_isize(inode, sbi->s_want_extra_isize, | |
5918 | iloc, handle); | |
5919 | ||
5e1021f2 | 5920 | return ext4_mark_iloc_dirty(handle, inode, &iloc); |
ac27a0ec DK |
5921 | } |
5922 | ||
5923 | /* | |
617ba13b | 5924 | * ext4_dirty_inode() is called from __mark_inode_dirty() |
ac27a0ec DK |
5925 | * |
5926 | * We're really interested in the case where a file is being extended. | |
5927 | * i_size has been changed by generic_commit_write() and we thus need | |
5928 | * to include the updated inode in the current transaction. | |
5929 | * | |
5dd4056d | 5930 | * Also, dquot_alloc_block() will always dirty the inode when blocks |
ac27a0ec DK |
5931 | * are allocated to the file. |
5932 | * | |
5933 | * If the inode is marked synchronous, we don't honour that here - doing | |
5934 | * so would cause a commit on atime updates, which we don't bother doing. | |
5935 | * We handle synchronous inodes at the highest possible level. | |
0ae45f63 TT |
5936 | * |
5937 | * If only the I_DIRTY_TIME flag is set, we can skip everything. If | |
5938 | * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need | |
5939 | * to copy into the on-disk inode structure are the timestamp files. | |
ac27a0ec | 5940 | */ |
aa385729 | 5941 | void ext4_dirty_inode(struct inode *inode, int flags) |
ac27a0ec | 5942 | { |
ac27a0ec DK |
5943 | handle_t *handle; |
5944 | ||
0ae45f63 TT |
5945 | if (flags == I_DIRTY_TIME) |
5946 | return; | |
9924a92a | 5947 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 2); |
ac27a0ec DK |
5948 | if (IS_ERR(handle)) |
5949 | goto out; | |
f3dc272f | 5950 | |
f3dc272f CW |
5951 | ext4_mark_inode_dirty(handle, inode); |
5952 | ||
617ba13b | 5953 | ext4_journal_stop(handle); |
ac27a0ec DK |
5954 | out: |
5955 | return; | |
5956 | } | |
5957 | ||
5958 | #if 0 | |
5959 | /* | |
5960 | * Bind an inode's backing buffer_head into this transaction, to prevent | |
5961 | * it from being flushed to disk early. Unlike | |
617ba13b | 5962 | * ext4_reserve_inode_write, this leaves behind no bh reference and |
ac27a0ec DK |
5963 | * returns no iloc structure, so the caller needs to repeat the iloc |
5964 | * lookup to mark the inode dirty later. | |
5965 | */ | |
617ba13b | 5966 | static int ext4_pin_inode(handle_t *handle, struct inode *inode) |
ac27a0ec | 5967 | { |
617ba13b | 5968 | struct ext4_iloc iloc; |
ac27a0ec DK |
5969 | |
5970 | int err = 0; | |
5971 | if (handle) { | |
617ba13b | 5972 | err = ext4_get_inode_loc(inode, &iloc); |
ac27a0ec DK |
5973 | if (!err) { |
5974 | BUFFER_TRACE(iloc.bh, "get_write_access"); | |
dab291af | 5975 | err = jbd2_journal_get_write_access(handle, iloc.bh); |
ac27a0ec | 5976 | if (!err) |
0390131b | 5977 | err = ext4_handle_dirty_metadata(handle, |
73b50c1c | 5978 | NULL, |
0390131b | 5979 | iloc.bh); |
ac27a0ec DK |
5980 | brelse(iloc.bh); |
5981 | } | |
5982 | } | |
617ba13b | 5983 | ext4_std_error(inode->i_sb, err); |
ac27a0ec DK |
5984 | return err; |
5985 | } | |
5986 | #endif | |
5987 | ||
617ba13b | 5988 | int ext4_change_inode_journal_flag(struct inode *inode, int val) |
ac27a0ec DK |
5989 | { |
5990 | journal_t *journal; | |
5991 | handle_t *handle; | |
5992 | int err; | |
c8585c6f | 5993 | struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb); |
ac27a0ec DK |
5994 | |
5995 | /* | |
5996 | * We have to be very careful here: changing a data block's | |
5997 | * journaling status dynamically is dangerous. If we write a | |
5998 | * data block to the journal, change the status and then delete | |
5999 | * that block, we risk forgetting to revoke the old log record | |
6000 | * from the journal and so a subsequent replay can corrupt data. | |
6001 | * So, first we make sure that the journal is empty and that | |
6002 | * nobody is changing anything. | |
6003 | */ | |
6004 | ||
617ba13b | 6005 | journal = EXT4_JOURNAL(inode); |
0390131b FM |
6006 | if (!journal) |
6007 | return 0; | |
d699594d | 6008 | if (is_journal_aborted(journal)) |
ac27a0ec DK |
6009 | return -EROFS; |
6010 | ||
17335dcc | 6011 | /* Wait for all existing dio workers */ |
17335dcc DM |
6012 | inode_dio_wait(inode); |
6013 | ||
4c546592 DJ |
6014 | /* |
6015 | * Before flushing the journal and switching inode's aops, we have | |
6016 | * to flush all dirty data the inode has. There can be outstanding | |
6017 | * delayed allocations, there can be unwritten extents created by | |
6018 | * fallocate or buffered writes in dioread_nolock mode covered by | |
6019 | * dirty data which can be converted only after flushing the dirty | |
6020 | * data (and journalled aops don't know how to handle these cases). | |
6021 | */ | |
6022 | if (val) { | |
6023 | down_write(&EXT4_I(inode)->i_mmap_sem); | |
6024 | err = filemap_write_and_wait(inode->i_mapping); | |
6025 | if (err < 0) { | |
6026 | up_write(&EXT4_I(inode)->i_mmap_sem); | |
4c546592 DJ |
6027 | return err; |
6028 | } | |
6029 | } | |
6030 | ||
c8585c6f | 6031 | percpu_down_write(&sbi->s_journal_flag_rwsem); |
dab291af | 6032 | jbd2_journal_lock_updates(journal); |
ac27a0ec DK |
6033 | |
6034 | /* | |
6035 | * OK, there are no updates running now, and all cached data is | |
6036 | * synced to disk. We are now in a completely consistent state | |
6037 | * which doesn't have anything in the journal, and we know that | |
6038 | * no filesystem updates are running, so it is safe to modify | |
6039 | * the inode's in-core data-journaling state flag now. | |
6040 | */ | |
6041 | ||
6042 | if (val) | |
12e9b892 | 6043 | ext4_set_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
5872ddaa | 6044 | else { |
4f879ca6 JK |
6045 | err = jbd2_journal_flush(journal); |
6046 | if (err < 0) { | |
6047 | jbd2_journal_unlock_updates(journal); | |
c8585c6f | 6048 | percpu_up_write(&sbi->s_journal_flag_rwsem); |
4f879ca6 JK |
6049 | return err; |
6050 | } | |
12e9b892 | 6051 | ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA); |
5872ddaa | 6052 | } |
617ba13b | 6053 | ext4_set_aops(inode); |
ac27a0ec | 6054 | |
dab291af | 6055 | jbd2_journal_unlock_updates(journal); |
c8585c6f DJ |
6056 | percpu_up_write(&sbi->s_journal_flag_rwsem); |
6057 | ||
4c546592 DJ |
6058 | if (val) |
6059 | up_write(&EXT4_I(inode)->i_mmap_sem); | |
ac27a0ec DK |
6060 | |
6061 | /* Finally we can mark the inode as dirty. */ | |
6062 | ||
9924a92a | 6063 | handle = ext4_journal_start(inode, EXT4_HT_INODE, 1); |
ac27a0ec DK |
6064 | if (IS_ERR(handle)) |
6065 | return PTR_ERR(handle); | |
6066 | ||
617ba13b | 6067 | err = ext4_mark_inode_dirty(handle, inode); |
0390131b | 6068 | ext4_handle_sync(handle); |
617ba13b MC |
6069 | ext4_journal_stop(handle); |
6070 | ext4_std_error(inode->i_sb, err); | |
ac27a0ec DK |
6071 | |
6072 | return err; | |
6073 | } | |
2e9ee850 AK |
6074 | |
6075 | static int ext4_bh_unmapped(handle_t *handle, struct buffer_head *bh) | |
6076 | { | |
6077 | return !buffer_mapped(bh); | |
6078 | } | |
6079 | ||
11bac800 | 6080 | int ext4_page_mkwrite(struct vm_fault *vmf) |
2e9ee850 | 6081 | { |
11bac800 | 6082 | struct vm_area_struct *vma = vmf->vma; |
c2ec175c | 6083 | struct page *page = vmf->page; |
2e9ee850 AK |
6084 | loff_t size; |
6085 | unsigned long len; | |
9ea7df53 | 6086 | int ret; |
2e9ee850 | 6087 | struct file *file = vma->vm_file; |
496ad9aa | 6088 | struct inode *inode = file_inode(file); |
2e9ee850 | 6089 | struct address_space *mapping = inode->i_mapping; |
9ea7df53 JK |
6090 | handle_t *handle; |
6091 | get_block_t *get_block; | |
6092 | int retries = 0; | |
2e9ee850 | 6093 | |
8e8ad8a5 | 6094 | sb_start_pagefault(inode->i_sb); |
041bbb6d | 6095 | file_update_time(vma->vm_file); |
ea3d7209 JK |
6096 | |
6097 | down_read(&EXT4_I(inode)->i_mmap_sem); | |
7b4cc978 EB |
6098 | |
6099 | ret = ext4_convert_inline_data(inode); | |
6100 | if (ret) | |
6101 | goto out_ret; | |
6102 | ||
9ea7df53 JK |
6103 | /* Delalloc case is easy... */ |
6104 | if (test_opt(inode->i_sb, DELALLOC) && | |
6105 | !ext4_should_journal_data(inode) && | |
6106 | !ext4_nonda_switch(inode->i_sb)) { | |
6107 | do { | |
5c500029 | 6108 | ret = block_page_mkwrite(vma, vmf, |
9ea7df53 JK |
6109 | ext4_da_get_block_prep); |
6110 | } while (ret == -ENOSPC && | |
6111 | ext4_should_retry_alloc(inode->i_sb, &retries)); | |
6112 | goto out_ret; | |
2e9ee850 | 6113 | } |
0e499890 DW |
6114 | |
6115 | lock_page(page); | |
9ea7df53 JK |
6116 | size = i_size_read(inode); |
6117 | /* Page got truncated from under us? */ | |
6118 | if (page->mapping != mapping || page_offset(page) > size) { | |
6119 | unlock_page(page); | |
6120 | ret = VM_FAULT_NOPAGE; | |
6121 | goto out; | |
0e499890 | 6122 | } |
2e9ee850 | 6123 | |
09cbfeaf KS |
6124 | if (page->index == size >> PAGE_SHIFT) |
6125 | len = size & ~PAGE_MASK; | |
2e9ee850 | 6126 | else |
09cbfeaf | 6127 | len = PAGE_SIZE; |
a827eaff | 6128 | /* |
9ea7df53 JK |
6129 | * Return if we have all the buffers mapped. This avoids the need to do |
6130 | * journal_start/journal_stop which can block and take a long time | |
a827eaff | 6131 | */ |
2e9ee850 | 6132 | if (page_has_buffers(page)) { |
f19d5870 TM |
6133 | if (!ext4_walk_page_buffers(NULL, page_buffers(page), |
6134 | 0, len, NULL, | |
6135 | ext4_bh_unmapped)) { | |
9ea7df53 | 6136 | /* Wait so that we don't change page under IO */ |
1d1d1a76 | 6137 | wait_for_stable_page(page); |
9ea7df53 JK |
6138 | ret = VM_FAULT_LOCKED; |
6139 | goto out; | |
a827eaff | 6140 | } |
2e9ee850 | 6141 | } |
a827eaff | 6142 | unlock_page(page); |
9ea7df53 JK |
6143 | /* OK, we need to fill the hole... */ |
6144 | if (ext4_should_dioread_nolock(inode)) | |
705965bd | 6145 | get_block = ext4_get_block_unwritten; |
9ea7df53 JK |
6146 | else |
6147 | get_block = ext4_get_block; | |
6148 | retry_alloc: | |
9924a92a TT |
6149 | handle = ext4_journal_start(inode, EXT4_HT_WRITE_PAGE, |
6150 | ext4_writepage_trans_blocks(inode)); | |
9ea7df53 | 6151 | if (IS_ERR(handle)) { |
c2ec175c | 6152 | ret = VM_FAULT_SIGBUS; |
9ea7df53 JK |
6153 | goto out; |
6154 | } | |
5c500029 | 6155 | ret = block_page_mkwrite(vma, vmf, get_block); |
9ea7df53 | 6156 | if (!ret && ext4_should_journal_data(inode)) { |
f19d5870 | 6157 | if (ext4_walk_page_buffers(handle, page_buffers(page), 0, |
09cbfeaf | 6158 | PAGE_SIZE, NULL, do_journal_get_write_access)) { |
9ea7df53 JK |
6159 | unlock_page(page); |
6160 | ret = VM_FAULT_SIGBUS; | |
fcbb5515 | 6161 | ext4_journal_stop(handle); |
9ea7df53 JK |
6162 | goto out; |
6163 | } | |
6164 | ext4_set_inode_state(inode, EXT4_STATE_JDATA); | |
6165 | } | |
6166 | ext4_journal_stop(handle); | |
6167 | if (ret == -ENOSPC && ext4_should_retry_alloc(inode->i_sb, &retries)) | |
6168 | goto retry_alloc; | |
6169 | out_ret: | |
6170 | ret = block_page_mkwrite_return(ret); | |
6171 | out: | |
ea3d7209 | 6172 | up_read(&EXT4_I(inode)->i_mmap_sem); |
8e8ad8a5 | 6173 | sb_end_pagefault(inode->i_sb); |
2e9ee850 AK |
6174 | return ret; |
6175 | } | |
ea3d7209 | 6176 | |
11bac800 | 6177 | int ext4_filemap_fault(struct vm_fault *vmf) |
ea3d7209 | 6178 | { |
11bac800 | 6179 | struct inode *inode = file_inode(vmf->vma->vm_file); |
ea3d7209 JK |
6180 | int err; |
6181 | ||
6182 | down_read(&EXT4_I(inode)->i_mmap_sem); | |
11bac800 | 6183 | err = filemap_fault(vmf); |
ea3d7209 JK |
6184 | up_read(&EXT4_I(inode)->i_mmap_sem); |
6185 | ||
6186 | return err; | |
6187 | } |