]>
Commit | Line | Data |
---|---|---|
ccd979bd MF |
1 | /* -*- mode: c; c-basic-offset: 8; -*- |
2 | * vim: noexpandtab sw=8 ts=8 sts=0: | |
3 | * | |
4 | * file.c | |
5 | * | |
6 | * File open, close, extend, truncate | |
7 | * | |
8 | * Copyright (C) 2002, 2004 Oracle. All rights reserved. | |
9 | * | |
10 | * This program is free software; you can redistribute it and/or | |
11 | * modify it under the terms of the GNU General Public | |
12 | * License as published by the Free Software Foundation; either | |
13 | * version 2 of the License, or (at your option) any later version. | |
14 | * | |
15 | * This program is distributed in the hope that it will be useful, | |
16 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
17 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
18 | * General Public License for more details. | |
19 | * | |
20 | * You should have received a copy of the GNU General Public | |
21 | * License along with this program; if not, write to the | |
22 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
23 | * Boston, MA 021110-1307, USA. | |
24 | */ | |
25 | ||
16f7e0fe | 26 | #include <linux/capability.h> |
ccd979bd MF |
27 | #include <linux/fs.h> |
28 | #include <linux/types.h> | |
29 | #include <linux/slab.h> | |
30 | #include <linux/highmem.h> | |
31 | #include <linux/pagemap.h> | |
32 | #include <linux/uio.h> | |
e2057c5a | 33 | #include <linux/sched.h> |
d6b29d7c | 34 | #include <linux/splice.h> |
7f1a37e3 | 35 | #include <linux/mount.h> |
9517bac6 | 36 | #include <linux/writeback.h> |
ccd979bd MF |
37 | |
38 | #define MLOG_MASK_PREFIX ML_INODE | |
39 | #include <cluster/masklog.h> | |
40 | ||
41 | #include "ocfs2.h" | |
42 | ||
43 | #include "alloc.h" | |
44 | #include "aops.h" | |
45 | #include "dir.h" | |
46 | #include "dlmglue.h" | |
47 | #include "extent_map.h" | |
48 | #include "file.h" | |
49 | #include "sysfile.h" | |
50 | #include "inode.h" | |
ca4d147e | 51 | #include "ioctl.h" |
ccd979bd MF |
52 | #include "journal.h" |
53 | #include "mmap.h" | |
54 | #include "suballoc.h" | |
55 | #include "super.h" | |
56 | ||
57 | #include "buffer_head_io.h" | |
58 | ||
59 | static int ocfs2_sync_inode(struct inode *inode) | |
60 | { | |
61 | filemap_fdatawrite(inode->i_mapping); | |
62 | return sync_mapping_buffers(inode->i_mapping); | |
63 | } | |
64 | ||
65 | static int ocfs2_file_open(struct inode *inode, struct file *file) | |
66 | { | |
67 | int status; | |
68 | int mode = file->f_flags; | |
69 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
70 | ||
71 | mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file, | |
d28c9174 | 72 | file->f_path.dentry->d_name.len, file->f_path.dentry->d_name.name); |
ccd979bd MF |
73 | |
74 | spin_lock(&oi->ip_lock); | |
75 | ||
76 | /* Check that the inode hasn't been wiped from disk by another | |
77 | * node. If it hasn't then we're safe as long as we hold the | |
78 | * spin lock until our increment of open count. */ | |
79 | if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) { | |
80 | spin_unlock(&oi->ip_lock); | |
81 | ||
82 | status = -ENOENT; | |
83 | goto leave; | |
84 | } | |
85 | ||
86 | if (mode & O_DIRECT) | |
87 | oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT; | |
88 | ||
89 | oi->ip_open_count++; | |
90 | spin_unlock(&oi->ip_lock); | |
91 | status = 0; | |
92 | leave: | |
93 | mlog_exit(status); | |
94 | return status; | |
95 | } | |
96 | ||
97 | static int ocfs2_file_release(struct inode *inode, struct file *file) | |
98 | { | |
99 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
100 | ||
101 | mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file, | |
d28c9174 JS |
102 | file->f_path.dentry->d_name.len, |
103 | file->f_path.dentry->d_name.name); | |
ccd979bd MF |
104 | |
105 | spin_lock(&oi->ip_lock); | |
106 | if (!--oi->ip_open_count) | |
107 | oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT; | |
108 | spin_unlock(&oi->ip_lock); | |
109 | ||
110 | mlog_exit(0); | |
111 | ||
112 | return 0; | |
113 | } | |
114 | ||
115 | static int ocfs2_sync_file(struct file *file, | |
116 | struct dentry *dentry, | |
117 | int datasync) | |
118 | { | |
119 | int err = 0; | |
120 | journal_t *journal; | |
121 | struct inode *inode = dentry->d_inode; | |
122 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
123 | ||
124 | mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", file, dentry, datasync, | |
125 | dentry->d_name.len, dentry->d_name.name); | |
126 | ||
127 | err = ocfs2_sync_inode(dentry->d_inode); | |
128 | if (err) | |
129 | goto bail; | |
130 | ||
131 | journal = osb->journal->j_journal; | |
132 | err = journal_force_commit(journal); | |
133 | ||
134 | bail: | |
135 | mlog_exit(err); | |
136 | ||
137 | return (err < 0) ? -EIO : 0; | |
138 | } | |
139 | ||
7f1a37e3 TY |
140 | int ocfs2_should_update_atime(struct inode *inode, |
141 | struct vfsmount *vfsmnt) | |
142 | { | |
143 | struct timespec now; | |
144 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
145 | ||
146 | if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb)) | |
147 | return 0; | |
148 | ||
149 | if ((inode->i_flags & S_NOATIME) || | |
150 | ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))) | |
151 | return 0; | |
152 | ||
6c2aad05 MF |
153 | /* |
154 | * We can be called with no vfsmnt structure - NFSD will | |
155 | * sometimes do this. | |
156 | * | |
157 | * Note that our action here is different than touch_atime() - | |
158 | * if we can't tell whether this is a noatime mount, then we | |
159 | * don't know whether to trust the value of s_atime_quantum. | |
160 | */ | |
161 | if (vfsmnt == NULL) | |
162 | return 0; | |
163 | ||
7f1a37e3 TY |
164 | if ((vfsmnt->mnt_flags & MNT_NOATIME) || |
165 | ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))) | |
166 | return 0; | |
167 | ||
7e913c53 MF |
168 | if (vfsmnt->mnt_flags & MNT_RELATIME) { |
169 | if ((timespec_compare(&inode->i_atime, &inode->i_mtime) <= 0) || | |
170 | (timespec_compare(&inode->i_atime, &inode->i_ctime) <= 0)) | |
171 | return 1; | |
172 | ||
173 | return 0; | |
174 | } | |
175 | ||
7f1a37e3 TY |
176 | now = CURRENT_TIME; |
177 | if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum)) | |
178 | return 0; | |
179 | else | |
180 | return 1; | |
181 | } | |
182 | ||
183 | int ocfs2_update_inode_atime(struct inode *inode, | |
184 | struct buffer_head *bh) | |
185 | { | |
186 | int ret; | |
187 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
188 | handle_t *handle; | |
189 | ||
190 | mlog_entry_void(); | |
191 | ||
192 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); | |
193 | if (handle == NULL) { | |
194 | ret = -ENOMEM; | |
195 | mlog_errno(ret); | |
196 | goto out; | |
197 | } | |
198 | ||
199 | inode->i_atime = CURRENT_TIME; | |
200 | ret = ocfs2_mark_inode_dirty(handle, inode, bh); | |
201 | if (ret < 0) | |
202 | mlog_errno(ret); | |
203 | ||
204 | ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle); | |
205 | out: | |
206 | mlog_exit(ret); | |
207 | return ret; | |
208 | } | |
209 | ||
6cb129f5 AB |
210 | static int ocfs2_set_inode_size(handle_t *handle, |
211 | struct inode *inode, | |
212 | struct buffer_head *fe_bh, | |
213 | u64 new_i_size) | |
ccd979bd MF |
214 | { |
215 | int status; | |
216 | ||
217 | mlog_entry_void(); | |
218 | i_size_write(inode, new_i_size); | |
8110b073 | 219 | inode->i_blocks = ocfs2_inode_sector_count(inode); |
ccd979bd MF |
220 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; |
221 | ||
222 | status = ocfs2_mark_inode_dirty(handle, inode, fe_bh); | |
223 | if (status < 0) { | |
224 | mlog_errno(status); | |
225 | goto bail; | |
226 | } | |
227 | ||
228 | bail: | |
229 | mlog_exit(status); | |
230 | return status; | |
231 | } | |
232 | ||
233 | static int ocfs2_simple_size_update(struct inode *inode, | |
234 | struct buffer_head *di_bh, | |
235 | u64 new_i_size) | |
236 | { | |
237 | int ret; | |
238 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1fabe148 | 239 | handle_t *handle = NULL; |
ccd979bd | 240 | |
65eff9cc | 241 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
ccd979bd MF |
242 | if (handle == NULL) { |
243 | ret = -ENOMEM; | |
244 | mlog_errno(ret); | |
245 | goto out; | |
246 | } | |
247 | ||
248 | ret = ocfs2_set_inode_size(handle, inode, di_bh, | |
249 | new_i_size); | |
250 | if (ret < 0) | |
251 | mlog_errno(ret); | |
252 | ||
02dc1af4 | 253 | ocfs2_commit_trans(osb, handle); |
ccd979bd MF |
254 | out: |
255 | return ret; | |
256 | } | |
257 | ||
258 | static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb, | |
259 | struct inode *inode, | |
260 | struct buffer_head *fe_bh, | |
261 | u64 new_i_size) | |
262 | { | |
263 | int status; | |
1fabe148 | 264 | handle_t *handle; |
60b11392 | 265 | struct ocfs2_dinode *di; |
35edec1d | 266 | u64 cluster_bytes; |
ccd979bd MF |
267 | |
268 | mlog_entry_void(); | |
269 | ||
270 | /* TODO: This needs to actually orphan the inode in this | |
271 | * transaction. */ | |
272 | ||
65eff9cc | 273 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
ccd979bd MF |
274 | if (IS_ERR(handle)) { |
275 | status = PTR_ERR(handle); | |
276 | mlog_errno(status); | |
277 | goto out; | |
278 | } | |
279 | ||
60b11392 MF |
280 | status = ocfs2_journal_access(handle, inode, fe_bh, |
281 | OCFS2_JOURNAL_ACCESS_WRITE); | |
282 | if (status < 0) { | |
283 | mlog_errno(status); | |
284 | goto out_commit; | |
285 | } | |
286 | ||
287 | /* | |
288 | * Do this before setting i_size. | |
289 | */ | |
35edec1d MF |
290 | cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size); |
291 | status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size, | |
292 | cluster_bytes); | |
60b11392 MF |
293 | if (status) { |
294 | mlog_errno(status); | |
295 | goto out_commit; | |
296 | } | |
297 | ||
298 | i_size_write(inode, new_i_size); | |
299 | inode->i_blocks = ocfs2_align_bytes_to_sectors(new_i_size); | |
300 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; | |
301 | ||
302 | di = (struct ocfs2_dinode *) fe_bh->b_data; | |
303 | di->i_size = cpu_to_le64(new_i_size); | |
304 | di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec); | |
305 | di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec); | |
306 | ||
307 | status = ocfs2_journal_dirty(handle, fe_bh); | |
ccd979bd MF |
308 | if (status < 0) |
309 | mlog_errno(status); | |
310 | ||
60b11392 | 311 | out_commit: |
02dc1af4 | 312 | ocfs2_commit_trans(osb, handle); |
ccd979bd | 313 | out: |
60b11392 | 314 | |
ccd979bd MF |
315 | mlog_exit(status); |
316 | return status; | |
317 | } | |
318 | ||
319 | static int ocfs2_truncate_file(struct inode *inode, | |
320 | struct buffer_head *di_bh, | |
321 | u64 new_i_size) | |
322 | { | |
323 | int status = 0; | |
324 | struct ocfs2_dinode *fe = NULL; | |
325 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
326 | struct ocfs2_truncate_context *tc = NULL; | |
327 | ||
b0697053 MF |
328 | mlog_entry("(inode = %llu, new_i_size = %llu\n", |
329 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
330 | (unsigned long long)new_i_size); | |
ccd979bd | 331 | |
ccd979bd MF |
332 | fe = (struct ocfs2_dinode *) di_bh->b_data; |
333 | if (!OCFS2_IS_VALID_DINODE(fe)) { | |
334 | OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe); | |
335 | status = -EIO; | |
336 | goto bail; | |
337 | } | |
338 | ||
339 | mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode), | |
b0697053 MF |
340 | "Inode %llu, inode i_size = %lld != di " |
341 | "i_size = %llu, i_flags = 0x%x\n", | |
342 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
ccd979bd | 343 | i_size_read(inode), |
b0697053 MF |
344 | (unsigned long long)le64_to_cpu(fe->i_size), |
345 | le32_to_cpu(fe->i_flags)); | |
ccd979bd MF |
346 | |
347 | if (new_i_size > le64_to_cpu(fe->i_size)) { | |
b0697053 MF |
348 | mlog(0, "asked to truncate file with size (%llu) to size (%llu)!\n", |
349 | (unsigned long long)le64_to_cpu(fe->i_size), | |
350 | (unsigned long long)new_i_size); | |
ccd979bd MF |
351 | status = -EINVAL; |
352 | mlog_errno(status); | |
353 | goto bail; | |
354 | } | |
355 | ||
b0697053 MF |
356 | mlog(0, "inode %llu, i_size = %llu, new_i_size = %llu\n", |
357 | (unsigned long long)le64_to_cpu(fe->i_blkno), | |
358 | (unsigned long long)le64_to_cpu(fe->i_size), | |
359 | (unsigned long long)new_i_size); | |
ccd979bd MF |
360 | |
361 | /* lets handle the simple truncate cases before doing any more | |
362 | * cluster locking. */ | |
363 | if (new_i_size == le64_to_cpu(fe->i_size)) | |
364 | goto bail; | |
365 | ||
2e89b2e4 MF |
366 | down_write(&OCFS2_I(inode)->ip_alloc_sem); |
367 | ||
ab0920ce MF |
368 | /* This forces other nodes to sync and drop their pages. Do |
369 | * this even if we have a truncate without allocation change - | |
370 | * ocfs2 cluster sizes can be much greater than page size, so | |
371 | * we have to truncate them anyway. */ | |
372 | status = ocfs2_data_lock(inode, 1); | |
373 | if (status < 0) { | |
2e89b2e4 MF |
374 | up_write(&OCFS2_I(inode)->ip_alloc_sem); |
375 | ||
ab0920ce MF |
376 | mlog_errno(status); |
377 | goto bail; | |
378 | } | |
ab0920ce | 379 | |
2e89b2e4 MF |
380 | unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1); |
381 | truncate_inode_pages(inode->i_mapping, new_i_size); | |
382 | ||
ccd979bd MF |
383 | /* alright, we're going to need to do a full blown alloc size |
384 | * change. Orphan the inode so that recovery can complete the | |
385 | * truncate if necessary. This does the task of marking | |
386 | * i_size. */ | |
387 | status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size); | |
388 | if (status < 0) { | |
389 | mlog_errno(status); | |
60b11392 | 390 | goto bail_unlock_data; |
ccd979bd MF |
391 | } |
392 | ||
393 | status = ocfs2_prepare_truncate(osb, inode, di_bh, &tc); | |
394 | if (status < 0) { | |
395 | mlog_errno(status); | |
60b11392 | 396 | goto bail_unlock_data; |
ccd979bd MF |
397 | } |
398 | ||
399 | status = ocfs2_commit_truncate(osb, inode, di_bh, tc); | |
400 | if (status < 0) { | |
401 | mlog_errno(status); | |
60b11392 | 402 | goto bail_unlock_data; |
ccd979bd MF |
403 | } |
404 | ||
405 | /* TODO: orphan dir cleanup here. */ | |
60b11392 MF |
406 | bail_unlock_data: |
407 | ocfs2_data_unlock(inode, 1); | |
408 | ||
2e89b2e4 MF |
409 | up_write(&OCFS2_I(inode)->ip_alloc_sem); |
410 | ||
ccd979bd MF |
411 | bail: |
412 | ||
413 | mlog_exit(status); | |
414 | return status; | |
415 | } | |
416 | ||
417 | /* | |
418 | * extend allocation only here. | |
419 | * we'll update all the disk stuff, and oip->alloc_size | |
420 | * | |
421 | * expect stuff to be locked, a transaction started and enough data / | |
422 | * metadata reservations in the contexts. | |
423 | * | |
424 | * Will return -EAGAIN, and a reason if a restart is needed. | |
425 | * If passed in, *reason will always be set, even in error. | |
426 | */ | |
427 | int ocfs2_do_extend_allocation(struct ocfs2_super *osb, | |
428 | struct inode *inode, | |
dcd0538f | 429 | u32 *logical_offset, |
ccd979bd | 430 | u32 clusters_to_add, |
2ae99a60 | 431 | int mark_unwritten, |
ccd979bd | 432 | struct buffer_head *fe_bh, |
1fabe148 | 433 | handle_t *handle, |
ccd979bd MF |
434 | struct ocfs2_alloc_context *data_ac, |
435 | struct ocfs2_alloc_context *meta_ac, | |
436 | enum ocfs2_alloc_restarted *reason_ret) | |
437 | { | |
438 | int status = 0; | |
439 | int free_extents; | |
440 | struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data; | |
441 | enum ocfs2_alloc_restarted reason = RESTART_NONE; | |
442 | u32 bit_off, num_bits; | |
443 | u64 block; | |
2ae99a60 | 444 | u8 flags = 0; |
ccd979bd MF |
445 | |
446 | BUG_ON(!clusters_to_add); | |
447 | ||
2ae99a60 MF |
448 | if (mark_unwritten) |
449 | flags = OCFS2_EXT_UNWRITTEN; | |
450 | ||
ccd979bd MF |
451 | free_extents = ocfs2_num_free_extents(osb, inode, fe); |
452 | if (free_extents < 0) { | |
453 | status = free_extents; | |
454 | mlog_errno(status); | |
455 | goto leave; | |
456 | } | |
457 | ||
458 | /* there are two cases which could cause us to EAGAIN in the | |
459 | * we-need-more-metadata case: | |
460 | * 1) we haven't reserved *any* | |
461 | * 2) we are so fragmented, we've needed to add metadata too | |
462 | * many times. */ | |
463 | if (!free_extents && !meta_ac) { | |
464 | mlog(0, "we haven't reserved any metadata!\n"); | |
465 | status = -EAGAIN; | |
466 | reason = RESTART_META; | |
467 | goto leave; | |
468 | } else if ((!free_extents) | |
469 | && (ocfs2_alloc_context_bits_left(meta_ac) | |
470 | < ocfs2_extend_meta_needed(fe))) { | |
471 | mlog(0, "filesystem is really fragmented...\n"); | |
472 | status = -EAGAIN; | |
473 | reason = RESTART_META; | |
474 | goto leave; | |
475 | } | |
476 | ||
477 | status = ocfs2_claim_clusters(osb, handle, data_ac, 1, | |
478 | &bit_off, &num_bits); | |
479 | if (status < 0) { | |
480 | if (status != -ENOSPC) | |
481 | mlog_errno(status); | |
482 | goto leave; | |
483 | } | |
484 | ||
485 | BUG_ON(num_bits > clusters_to_add); | |
486 | ||
487 | /* reserve our write early -- insert_extent may update the inode */ | |
488 | status = ocfs2_journal_access(handle, inode, fe_bh, | |
489 | OCFS2_JOURNAL_ACCESS_WRITE); | |
490 | if (status < 0) { | |
491 | mlog_errno(status); | |
492 | goto leave; | |
493 | } | |
494 | ||
495 | block = ocfs2_clusters_to_blocks(osb->sb, bit_off); | |
b0697053 MF |
496 | mlog(0, "Allocating %u clusters at block %u for inode %llu\n", |
497 | num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
dcd0538f MF |
498 | status = ocfs2_insert_extent(osb, handle, inode, fe_bh, |
499 | *logical_offset, block, num_bits, | |
2ae99a60 | 500 | flags, meta_ac); |
ccd979bd MF |
501 | if (status < 0) { |
502 | mlog_errno(status); | |
503 | goto leave; | |
504 | } | |
505 | ||
ccd979bd MF |
506 | status = ocfs2_journal_dirty(handle, fe_bh); |
507 | if (status < 0) { | |
508 | mlog_errno(status); | |
509 | goto leave; | |
510 | } | |
511 | ||
512 | clusters_to_add -= num_bits; | |
dcd0538f | 513 | *logical_offset += num_bits; |
ccd979bd MF |
514 | |
515 | if (clusters_to_add) { | |
516 | mlog(0, "need to alloc once more, clusters = %u, wanted = " | |
517 | "%u\n", fe->i_clusters, clusters_to_add); | |
518 | status = -EAGAIN; | |
519 | reason = RESTART_TRANS; | |
520 | } | |
521 | ||
522 | leave: | |
523 | mlog_exit(status); | |
524 | if (reason_ret) | |
525 | *reason_ret = reason; | |
526 | return status; | |
527 | } | |
528 | ||
abf8b156 MF |
529 | /* |
530 | * For a given allocation, determine which allocators will need to be | |
531 | * accessed, and lock them, reserving the appropriate number of bits. | |
532 | * | |
2ae99a60 MF |
533 | * Sparse file systems call this from ocfs2_write_begin_nolock() |
534 | * and ocfs2_allocate_unwritten_extents(). | |
535 | * | |
536 | * File systems which don't support holes call this from | |
537 | * ocfs2_extend_allocation(). | |
abf8b156 | 538 | */ |
9517bac6 | 539 | int ocfs2_lock_allocators(struct inode *inode, struct ocfs2_dinode *di, |
b27b7cbc | 540 | u32 clusters_to_add, u32 extents_to_split, |
9517bac6 MF |
541 | struct ocfs2_alloc_context **data_ac, |
542 | struct ocfs2_alloc_context **meta_ac) | |
abf8b156 | 543 | { |
063c4561 | 544 | int ret = 0, num_free_extents; |
b27b7cbc | 545 | unsigned int max_recs_needed = clusters_to_add + 2 * extents_to_split; |
abf8b156 MF |
546 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
547 | ||
548 | *meta_ac = NULL; | |
063c4561 MF |
549 | if (data_ac) |
550 | *data_ac = NULL; | |
551 | ||
552 | BUG_ON(clusters_to_add != 0 && data_ac == NULL); | |
abf8b156 MF |
553 | |
554 | mlog(0, "extend inode %llu, i_size = %lld, di->i_clusters = %u, " | |
b27b7cbc | 555 | "clusters_to_add = %u, extents_to_split = %u\n", |
abf8b156 | 556 | (unsigned long long)OCFS2_I(inode)->ip_blkno, i_size_read(inode), |
b27b7cbc | 557 | le32_to_cpu(di->i_clusters), clusters_to_add, extents_to_split); |
abf8b156 MF |
558 | |
559 | num_free_extents = ocfs2_num_free_extents(osb, inode, di); | |
560 | if (num_free_extents < 0) { | |
561 | ret = num_free_extents; | |
562 | mlog_errno(ret); | |
563 | goto out; | |
564 | } | |
565 | ||
566 | /* | |
567 | * Sparse allocation file systems need to be more conservative | |
568 | * with reserving room for expansion - the actual allocation | |
569 | * happens while we've got a journal handle open so re-taking | |
570 | * a cluster lock (because we ran out of room for another | |
571 | * extent) will violate ordering rules. | |
572 | * | |
9517bac6 | 573 | * Most of the time we'll only be seeing this 1 cluster at a time |
abf8b156 | 574 | * anyway. |
b27b7cbc MF |
575 | * |
576 | * Always lock for any unwritten extents - we might want to | |
577 | * add blocks during a split. | |
abf8b156 MF |
578 | */ |
579 | if (!num_free_extents || | |
b27b7cbc | 580 | (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed)) { |
abf8b156 MF |
581 | ret = ocfs2_reserve_new_metadata(osb, di, meta_ac); |
582 | if (ret < 0) { | |
583 | if (ret != -ENOSPC) | |
584 | mlog_errno(ret); | |
585 | goto out; | |
586 | } | |
587 | } | |
588 | ||
063c4561 MF |
589 | if (clusters_to_add == 0) |
590 | goto out; | |
591 | ||
abf8b156 MF |
592 | ret = ocfs2_reserve_clusters(osb, clusters_to_add, data_ac); |
593 | if (ret < 0) { | |
594 | if (ret != -ENOSPC) | |
595 | mlog_errno(ret); | |
596 | goto out; | |
597 | } | |
598 | ||
599 | out: | |
600 | if (ret) { | |
601 | if (*meta_ac) { | |
602 | ocfs2_free_alloc_context(*meta_ac); | |
603 | *meta_ac = NULL; | |
604 | } | |
605 | ||
606 | /* | |
607 | * We cannot have an error and a non null *data_ac. | |
608 | */ | |
609 | } | |
610 | ||
611 | return ret; | |
612 | } | |
613 | ||
2ae99a60 MF |
614 | static int __ocfs2_extend_allocation(struct inode *inode, u32 logical_start, |
615 | u32 clusters_to_add, int mark_unwritten) | |
ccd979bd MF |
616 | { |
617 | int status = 0; | |
618 | int restart_func = 0; | |
abf8b156 | 619 | int credits; |
2ae99a60 | 620 | u32 prev_clusters; |
ccd979bd MF |
621 | struct buffer_head *bh = NULL; |
622 | struct ocfs2_dinode *fe = NULL; | |
1fabe148 | 623 | handle_t *handle = NULL; |
ccd979bd MF |
624 | struct ocfs2_alloc_context *data_ac = NULL; |
625 | struct ocfs2_alloc_context *meta_ac = NULL; | |
626 | enum ocfs2_alloc_restarted why; | |
627 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
628 | ||
629 | mlog_entry("(clusters_to_add = %u)\n", clusters_to_add); | |
630 | ||
dcd0538f MF |
631 | /* |
632 | * This function only exists for file systems which don't | |
633 | * support holes. | |
634 | */ | |
2ae99a60 | 635 | BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb)); |
dcd0538f | 636 | |
ccd979bd MF |
637 | status = ocfs2_read_block(osb, OCFS2_I(inode)->ip_blkno, &bh, |
638 | OCFS2_BH_CACHED, inode); | |
639 | if (status < 0) { | |
640 | mlog_errno(status); | |
641 | goto leave; | |
642 | } | |
643 | ||
644 | fe = (struct ocfs2_dinode *) bh->b_data; | |
645 | if (!OCFS2_IS_VALID_DINODE(fe)) { | |
646 | OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe); | |
647 | status = -EIO; | |
648 | goto leave; | |
649 | } | |
650 | ||
651 | restart_all: | |
652 | BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters); | |
653 | ||
b27b7cbc | 654 | status = ocfs2_lock_allocators(inode, fe, clusters_to_add, 0, &data_ac, |
9517bac6 MF |
655 | &meta_ac); |
656 | if (status) { | |
657 | mlog_errno(status); | |
658 | goto leave; | |
659 | } | |
660 | ||
ccd979bd | 661 | credits = ocfs2_calc_extend_credits(osb->sb, fe, clusters_to_add); |
65eff9cc | 662 | handle = ocfs2_start_trans(osb, credits); |
ccd979bd MF |
663 | if (IS_ERR(handle)) { |
664 | status = PTR_ERR(handle); | |
665 | handle = NULL; | |
666 | mlog_errno(status); | |
667 | goto leave; | |
668 | } | |
669 | ||
670 | restarted_transaction: | |
671 | /* reserve a write to the file entry early on - that we if we | |
672 | * run out of credits in the allocation path, we can still | |
673 | * update i_size. */ | |
674 | status = ocfs2_journal_access(handle, inode, bh, | |
675 | OCFS2_JOURNAL_ACCESS_WRITE); | |
676 | if (status < 0) { | |
677 | mlog_errno(status); | |
678 | goto leave; | |
679 | } | |
680 | ||
681 | prev_clusters = OCFS2_I(inode)->ip_clusters; | |
682 | ||
683 | status = ocfs2_do_extend_allocation(osb, | |
684 | inode, | |
dcd0538f | 685 | &logical_start, |
ccd979bd | 686 | clusters_to_add, |
2ae99a60 | 687 | mark_unwritten, |
ccd979bd MF |
688 | bh, |
689 | handle, | |
690 | data_ac, | |
691 | meta_ac, | |
692 | &why); | |
693 | if ((status < 0) && (status != -EAGAIN)) { | |
694 | if (status != -ENOSPC) | |
695 | mlog_errno(status); | |
696 | goto leave; | |
697 | } | |
698 | ||
699 | status = ocfs2_journal_dirty(handle, bh); | |
700 | if (status < 0) { | |
701 | mlog_errno(status); | |
702 | goto leave; | |
703 | } | |
704 | ||
705 | spin_lock(&OCFS2_I(inode)->ip_lock); | |
706 | clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters); | |
707 | spin_unlock(&OCFS2_I(inode)->ip_lock); | |
708 | ||
709 | if (why != RESTART_NONE && clusters_to_add) { | |
710 | if (why == RESTART_META) { | |
711 | mlog(0, "restarting function.\n"); | |
712 | restart_func = 1; | |
713 | } else { | |
714 | BUG_ON(why != RESTART_TRANS); | |
715 | ||
716 | mlog(0, "restarting transaction.\n"); | |
717 | /* TODO: This can be more intelligent. */ | |
718 | credits = ocfs2_calc_extend_credits(osb->sb, | |
719 | fe, | |
720 | clusters_to_add); | |
1fabe148 | 721 | status = ocfs2_extend_trans(handle, credits); |
ccd979bd MF |
722 | if (status < 0) { |
723 | /* handle still has to be committed at | |
724 | * this point. */ | |
725 | status = -ENOMEM; | |
726 | mlog_errno(status); | |
727 | goto leave; | |
728 | } | |
729 | goto restarted_transaction; | |
730 | } | |
731 | } | |
732 | ||
b0697053 | 733 | mlog(0, "fe: i_clusters = %u, i_size=%llu\n", |
1ca1a111 MF |
734 | le32_to_cpu(fe->i_clusters), |
735 | (unsigned long long)le64_to_cpu(fe->i_size)); | |
ccd979bd MF |
736 | mlog(0, "inode: ip_clusters=%u, i_size=%lld\n", |
737 | OCFS2_I(inode)->ip_clusters, i_size_read(inode)); | |
738 | ||
739 | leave: | |
ccd979bd | 740 | if (handle) { |
02dc1af4 | 741 | ocfs2_commit_trans(osb, handle); |
ccd979bd MF |
742 | handle = NULL; |
743 | } | |
744 | if (data_ac) { | |
745 | ocfs2_free_alloc_context(data_ac); | |
746 | data_ac = NULL; | |
747 | } | |
748 | if (meta_ac) { | |
749 | ocfs2_free_alloc_context(meta_ac); | |
750 | meta_ac = NULL; | |
751 | } | |
752 | if ((!status) && restart_func) { | |
753 | restart_func = 0; | |
754 | goto restart_all; | |
755 | } | |
756 | if (bh) { | |
757 | brelse(bh); | |
758 | bh = NULL; | |
759 | } | |
760 | ||
761 | mlog_exit(status); | |
762 | return status; | |
763 | } | |
764 | ||
2ae99a60 MF |
765 | static int ocfs2_extend_allocation(struct inode *inode, u32 logical_start, |
766 | u32 clusters_to_add, int mark_unwritten) | |
767 | { | |
768 | int ret; | |
769 | ||
770 | /* | |
771 | * The alloc sem blocks peope in read/write from reading our | |
772 | * allocation until we're done changing it. We depend on | |
773 | * i_mutex to block other extend/truncate calls while we're | |
774 | * here. | |
775 | */ | |
776 | down_write(&OCFS2_I(inode)->ip_alloc_sem); | |
777 | ret = __ocfs2_extend_allocation(inode, logical_start, clusters_to_add, | |
778 | mark_unwritten); | |
779 | up_write(&OCFS2_I(inode)->ip_alloc_sem); | |
780 | ||
781 | return ret; | |
782 | } | |
783 | ||
ccd979bd MF |
784 | /* Some parts of this taken from generic_cont_expand, which turned out |
785 | * to be too fragile to do exactly what we need without us having to | |
53013cba MF |
786 | * worry about recursive locking in ->prepare_write() and |
787 | * ->commit_write(). */ | |
ccd979bd MF |
788 | static int ocfs2_write_zero_page(struct inode *inode, |
789 | u64 size) | |
790 | { | |
791 | struct address_space *mapping = inode->i_mapping; | |
792 | struct page *page; | |
793 | unsigned long index; | |
794 | unsigned int offset; | |
1fabe148 | 795 | handle_t *handle = NULL; |
ccd979bd MF |
796 | int ret; |
797 | ||
798 | offset = (size & (PAGE_CACHE_SIZE-1)); /* Within page */ | |
799 | /* ugh. in prepare/commit_write, if from==to==start of block, we | |
800 | ** skip the prepare. make sure we never send an offset for the start | |
801 | ** of a block | |
802 | */ | |
803 | if ((offset & (inode->i_sb->s_blocksize - 1)) == 0) { | |
804 | offset++; | |
805 | } | |
806 | index = size >> PAGE_CACHE_SHIFT; | |
807 | ||
808 | page = grab_cache_page(mapping, index); | |
809 | if (!page) { | |
810 | ret = -ENOMEM; | |
811 | mlog_errno(ret); | |
812 | goto out; | |
813 | } | |
814 | ||
53013cba | 815 | ret = ocfs2_prepare_write_nolock(inode, page, offset, offset); |
ccd979bd MF |
816 | if (ret < 0) { |
817 | mlog_errno(ret); | |
818 | goto out_unlock; | |
819 | } | |
820 | ||
821 | if (ocfs2_should_order_data(inode)) { | |
822 | handle = ocfs2_start_walk_page_trans(inode, page, offset, | |
823 | offset); | |
824 | if (IS_ERR(handle)) { | |
825 | ret = PTR_ERR(handle); | |
826 | handle = NULL; | |
827 | goto out_unlock; | |
828 | } | |
829 | } | |
830 | ||
831 | /* must not update i_size! */ | |
832 | ret = block_commit_write(page, offset, offset); | |
833 | if (ret < 0) | |
834 | mlog_errno(ret); | |
835 | else | |
836 | ret = 0; | |
837 | ||
838 | if (handle) | |
02dc1af4 | 839 | ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle); |
ccd979bd MF |
840 | out_unlock: |
841 | unlock_page(page); | |
842 | page_cache_release(page); | |
843 | out: | |
844 | return ret; | |
845 | } | |
846 | ||
847 | static int ocfs2_zero_extend(struct inode *inode, | |
848 | u64 zero_to_size) | |
849 | { | |
850 | int ret = 0; | |
851 | u64 start_off; | |
852 | struct super_block *sb = inode->i_sb; | |
853 | ||
854 | start_off = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode)); | |
855 | while (start_off < zero_to_size) { | |
856 | ret = ocfs2_write_zero_page(inode, start_off); | |
857 | if (ret < 0) { | |
858 | mlog_errno(ret); | |
859 | goto out; | |
860 | } | |
861 | ||
862 | start_off += sb->s_blocksize; | |
e2057c5a MF |
863 | |
864 | /* | |
865 | * Very large extends have the potential to lock up | |
866 | * the cpu for extended periods of time. | |
867 | */ | |
868 | cond_resched(); | |
ccd979bd MF |
869 | } |
870 | ||
871 | out: | |
872 | return ret; | |
873 | } | |
874 | ||
53013cba MF |
875 | /* |
876 | * A tail_to_skip value > 0 indicates that we're being called from | |
877 | * ocfs2_file_aio_write(). This has the following implications: | |
878 | * | |
879 | * - we don't want to update i_size | |
880 | * - di_bh will be NULL, which is fine because it's only used in the | |
881 | * case where we want to update i_size. | |
882 | * - ocfs2_zero_extend() will then only be filling the hole created | |
883 | * between i_size and the start of the write. | |
884 | */ | |
ccd979bd MF |
885 | static int ocfs2_extend_file(struct inode *inode, |
886 | struct buffer_head *di_bh, | |
53013cba MF |
887 | u64 new_i_size, |
888 | size_t tail_to_skip) | |
ccd979bd MF |
889 | { |
890 | int ret = 0; | |
3a0782d0 | 891 | u32 clusters_to_add = 0; |
ccd979bd | 892 | |
53013cba MF |
893 | BUG_ON(!tail_to_skip && !di_bh); |
894 | ||
ccd979bd MF |
895 | /* setattr sometimes calls us like this. */ |
896 | if (new_i_size == 0) | |
897 | goto out; | |
898 | ||
899 | if (i_size_read(inode) == new_i_size) | |
900 | goto out; | |
901 | BUG_ON(new_i_size < i_size_read(inode)); | |
902 | ||
3a0782d0 MF |
903 | if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))) { |
904 | BUG_ON(tail_to_skip != 0); | |
905 | goto out_update_size; | |
906 | } | |
907 | ||
ccd979bd MF |
908 | clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size) - |
909 | OCFS2_I(inode)->ip_clusters; | |
910 | ||
0effef77 MF |
911 | /* |
912 | * protect the pages that ocfs2_zero_extend is going to be | |
913 | * pulling into the page cache.. we do this before the | |
914 | * metadata extend so that we don't get into the situation | |
915 | * where we've extended the metadata but can't get the data | |
916 | * lock to zero. | |
917 | */ | |
918 | ret = ocfs2_data_lock(inode, 1); | |
919 | if (ret < 0) { | |
920 | mlog_errno(ret); | |
921 | goto out; | |
922 | } | |
ccd979bd | 923 | |
0effef77 | 924 | if (clusters_to_add) { |
2ae99a60 MF |
925 | ret = ocfs2_extend_allocation(inode, |
926 | OCFS2_I(inode)->ip_clusters, | |
927 | clusters_to_add, 0); | |
ccd979bd MF |
928 | if (ret < 0) { |
929 | mlog_errno(ret); | |
53013cba | 930 | goto out_unlock; |
ccd979bd | 931 | } |
0effef77 | 932 | } |
ccd979bd | 933 | |
0effef77 MF |
934 | /* |
935 | * Call this even if we don't add any clusters to the tree. We | |
936 | * still need to zero the area between the old i_size and the | |
937 | * new i_size. | |
938 | */ | |
939 | ret = ocfs2_zero_extend(inode, (u64)new_i_size - tail_to_skip); | |
940 | if (ret < 0) { | |
941 | mlog_errno(ret); | |
942 | goto out_unlock; | |
53013cba MF |
943 | } |
944 | ||
3a0782d0 | 945 | out_update_size: |
53013cba MF |
946 | if (!tail_to_skip) { |
947 | /* We're being called from ocfs2_setattr() which wants | |
948 | * us to update i_size */ | |
949 | ret = ocfs2_simple_size_update(inode, di_bh, new_i_size); | |
950 | if (ret < 0) | |
951 | mlog_errno(ret); | |
ccd979bd MF |
952 | } |
953 | ||
53013cba | 954 | out_unlock: |
3a0782d0 MF |
955 | if (!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))) |
956 | ocfs2_data_unlock(inode, 1); | |
53013cba | 957 | |
ccd979bd MF |
958 | out: |
959 | return ret; | |
960 | } | |
961 | ||
962 | int ocfs2_setattr(struct dentry *dentry, struct iattr *attr) | |
963 | { | |
964 | int status = 0, size_change; | |
965 | struct inode *inode = dentry->d_inode; | |
966 | struct super_block *sb = inode->i_sb; | |
967 | struct ocfs2_super *osb = OCFS2_SB(sb); | |
968 | struct buffer_head *bh = NULL; | |
1fabe148 | 969 | handle_t *handle = NULL; |
ccd979bd MF |
970 | |
971 | mlog_entry("(0x%p, '%.*s')\n", dentry, | |
972 | dentry->d_name.len, dentry->d_name.name); | |
973 | ||
974 | if (attr->ia_valid & ATTR_MODE) | |
975 | mlog(0, "mode change: %d\n", attr->ia_mode); | |
976 | if (attr->ia_valid & ATTR_UID) | |
977 | mlog(0, "uid change: %d\n", attr->ia_uid); | |
978 | if (attr->ia_valid & ATTR_GID) | |
979 | mlog(0, "gid change: %d\n", attr->ia_gid); | |
980 | if (attr->ia_valid & ATTR_SIZE) | |
981 | mlog(0, "size change...\n"); | |
982 | if (attr->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME)) | |
983 | mlog(0, "time change...\n"); | |
984 | ||
985 | #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \ | |
986 | | ATTR_GID | ATTR_UID | ATTR_MODE) | |
987 | if (!(attr->ia_valid & OCFS2_VALID_ATTRS)) { | |
988 | mlog(0, "can't handle attrs: 0x%x\n", attr->ia_valid); | |
989 | return 0; | |
990 | } | |
991 | ||
992 | status = inode_change_ok(inode, attr); | |
993 | if (status) | |
994 | return status; | |
995 | ||
996 | size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE; | |
997 | if (size_change) { | |
998 | status = ocfs2_rw_lock(inode, 1); | |
999 | if (status < 0) { | |
1000 | mlog_errno(status); | |
1001 | goto bail; | |
1002 | } | |
1003 | } | |
1004 | ||
4bcec184 | 1005 | status = ocfs2_meta_lock(inode, &bh, 1); |
ccd979bd MF |
1006 | if (status < 0) { |
1007 | if (status != -ENOENT) | |
1008 | mlog_errno(status); | |
1009 | goto bail_unlock_rw; | |
1010 | } | |
1011 | ||
1012 | if (size_change && attr->ia_size != i_size_read(inode)) { | |
1013 | if (i_size_read(inode) > attr->ia_size) | |
1014 | status = ocfs2_truncate_file(inode, bh, attr->ia_size); | |
1015 | else | |
53013cba | 1016 | status = ocfs2_extend_file(inode, bh, attr->ia_size, 0); |
ccd979bd MF |
1017 | if (status < 0) { |
1018 | if (status != -ENOSPC) | |
1019 | mlog_errno(status); | |
1020 | status = -ENOSPC; | |
1021 | goto bail_unlock; | |
1022 | } | |
1023 | } | |
1024 | ||
65eff9cc | 1025 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
ccd979bd MF |
1026 | if (IS_ERR(handle)) { |
1027 | status = PTR_ERR(handle); | |
1028 | mlog_errno(status); | |
1029 | goto bail_unlock; | |
1030 | } | |
1031 | ||
7307de80 MF |
1032 | /* |
1033 | * This will intentionally not wind up calling vmtruncate(), | |
1034 | * since all the work for a size change has been done above. | |
1035 | * Otherwise, we could get into problems with truncate as | |
1036 | * ip_alloc_sem is used there to protect against i_size | |
1037 | * changes. | |
1038 | */ | |
ccd979bd MF |
1039 | status = inode_setattr(inode, attr); |
1040 | if (status < 0) { | |
1041 | mlog_errno(status); | |
1042 | goto bail_commit; | |
1043 | } | |
1044 | ||
1045 | status = ocfs2_mark_inode_dirty(handle, inode, bh); | |
1046 | if (status < 0) | |
1047 | mlog_errno(status); | |
1048 | ||
1049 | bail_commit: | |
02dc1af4 | 1050 | ocfs2_commit_trans(osb, handle); |
ccd979bd MF |
1051 | bail_unlock: |
1052 | ocfs2_meta_unlock(inode, 1); | |
1053 | bail_unlock_rw: | |
1054 | if (size_change) | |
1055 | ocfs2_rw_unlock(inode, 1); | |
1056 | bail: | |
1057 | if (bh) | |
1058 | brelse(bh); | |
1059 | ||
1060 | mlog_exit(status); | |
1061 | return status; | |
1062 | } | |
1063 | ||
1064 | int ocfs2_getattr(struct vfsmount *mnt, | |
1065 | struct dentry *dentry, | |
1066 | struct kstat *stat) | |
1067 | { | |
1068 | struct inode *inode = dentry->d_inode; | |
1069 | struct super_block *sb = dentry->d_inode->i_sb; | |
1070 | struct ocfs2_super *osb = sb->s_fs_info; | |
1071 | int err; | |
1072 | ||
1073 | mlog_entry_void(); | |
1074 | ||
1075 | err = ocfs2_inode_revalidate(dentry); | |
1076 | if (err) { | |
1077 | if (err != -ENOENT) | |
1078 | mlog_errno(err); | |
1079 | goto bail; | |
1080 | } | |
1081 | ||
1082 | generic_fillattr(inode, stat); | |
1083 | ||
1084 | /* We set the blksize from the cluster size for performance */ | |
1085 | stat->blksize = osb->s_clustersize; | |
1086 | ||
1087 | bail: | |
1088 | mlog_exit(err); | |
1089 | ||
1090 | return err; | |
1091 | } | |
1092 | ||
d38eb8db TY |
1093 | int ocfs2_permission(struct inode *inode, int mask, struct nameidata *nd) |
1094 | { | |
1095 | int ret; | |
1096 | ||
1097 | mlog_entry_void(); | |
1098 | ||
1099 | ret = ocfs2_meta_lock(inode, NULL, 0); | |
1100 | if (ret) { | |
a9f5f707 MF |
1101 | if (ret != -ENOENT) |
1102 | mlog_errno(ret); | |
d38eb8db TY |
1103 | goto out; |
1104 | } | |
1105 | ||
1106 | ret = generic_permission(inode, mask, NULL); | |
d38eb8db TY |
1107 | |
1108 | ocfs2_meta_unlock(inode, 0); | |
1109 | out: | |
1110 | mlog_exit(ret); | |
1111 | return ret; | |
1112 | } | |
1113 | ||
ccd979bd MF |
1114 | static int ocfs2_write_remove_suid(struct inode *inode) |
1115 | { | |
1116 | int ret; | |
1117 | struct buffer_head *bh = NULL; | |
1118 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
1fabe148 | 1119 | handle_t *handle; |
ccd979bd MF |
1120 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
1121 | struct ocfs2_dinode *di; | |
1122 | ||
b0697053 MF |
1123 | mlog_entry("(Inode %llu, mode 0%o)\n", |
1124 | (unsigned long long)oi->ip_blkno, inode->i_mode); | |
ccd979bd | 1125 | |
65eff9cc | 1126 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
ccd979bd MF |
1127 | if (handle == NULL) { |
1128 | ret = -ENOMEM; | |
1129 | mlog_errno(ret); | |
1130 | goto out; | |
1131 | } | |
1132 | ||
1133 | ret = ocfs2_read_block(osb, oi->ip_blkno, &bh, OCFS2_BH_CACHED, inode); | |
1134 | if (ret < 0) { | |
1135 | mlog_errno(ret); | |
1136 | goto out_trans; | |
1137 | } | |
1138 | ||
1139 | ret = ocfs2_journal_access(handle, inode, bh, | |
1140 | OCFS2_JOURNAL_ACCESS_WRITE); | |
1141 | if (ret < 0) { | |
1142 | mlog_errno(ret); | |
1143 | goto out_bh; | |
1144 | } | |
1145 | ||
1146 | inode->i_mode &= ~S_ISUID; | |
1147 | if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP)) | |
1148 | inode->i_mode &= ~S_ISGID; | |
1149 | ||
1150 | di = (struct ocfs2_dinode *) bh->b_data; | |
1151 | di->i_mode = cpu_to_le16(inode->i_mode); | |
1152 | ||
1153 | ret = ocfs2_journal_dirty(handle, bh); | |
1154 | if (ret < 0) | |
1155 | mlog_errno(ret); | |
1156 | out_bh: | |
1157 | brelse(bh); | |
1158 | out_trans: | |
02dc1af4 | 1159 | ocfs2_commit_trans(osb, handle); |
ccd979bd MF |
1160 | out: |
1161 | mlog_exit(ret); | |
1162 | return ret; | |
1163 | } | |
1164 | ||
9517bac6 MF |
1165 | /* |
1166 | * Will look for holes and unwritten extents in the range starting at | |
1167 | * pos for count bytes (inclusive). | |
1168 | */ | |
1169 | static int ocfs2_check_range_for_holes(struct inode *inode, loff_t pos, | |
1170 | size_t count) | |
1171 | { | |
1172 | int ret = 0; | |
49cb8d2d | 1173 | unsigned int extent_flags; |
9517bac6 MF |
1174 | u32 cpos, clusters, extent_len, phys_cpos; |
1175 | struct super_block *sb = inode->i_sb; | |
1176 | ||
1177 | cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits; | |
1178 | clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos; | |
1179 | ||
1180 | while (clusters) { | |
49cb8d2d MF |
1181 | ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len, |
1182 | &extent_flags); | |
9517bac6 MF |
1183 | if (ret < 0) { |
1184 | mlog_errno(ret); | |
1185 | goto out; | |
1186 | } | |
1187 | ||
49cb8d2d | 1188 | if (phys_cpos == 0 || (extent_flags & OCFS2_EXT_UNWRITTEN)) { |
9517bac6 MF |
1189 | ret = 1; |
1190 | break; | |
1191 | } | |
1192 | ||
1193 | if (extent_len > clusters) | |
1194 | extent_len = clusters; | |
1195 | ||
1196 | clusters -= extent_len; | |
1197 | cpos += extent_len; | |
1198 | } | |
1199 | out: | |
1200 | return ret; | |
1201 | } | |
1202 | ||
2ae99a60 MF |
1203 | /* |
1204 | * Allocate enough extents to cover the region starting at byte offset | |
1205 | * start for len bytes. Existing extents are skipped, any extents | |
1206 | * added are marked as "unwritten". | |
1207 | */ | |
1208 | static int ocfs2_allocate_unwritten_extents(struct inode *inode, | |
1209 | u64 start, u64 len) | |
1210 | { | |
1211 | int ret; | |
1212 | u32 cpos, phys_cpos, clusters, alloc_size; | |
1213 | ||
1214 | /* | |
1215 | * We consider both start and len to be inclusive. | |
1216 | */ | |
1217 | cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits; | |
1218 | clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len); | |
1219 | clusters -= cpos; | |
1220 | ||
1221 | while (clusters) { | |
1222 | ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, | |
1223 | &alloc_size, NULL); | |
1224 | if (ret) { | |
1225 | mlog_errno(ret); | |
1226 | goto out; | |
1227 | } | |
1228 | ||
1229 | /* | |
1230 | * Hole or existing extent len can be arbitrary, so | |
1231 | * cap it to our own allocation request. | |
1232 | */ | |
1233 | if (alloc_size > clusters) | |
1234 | alloc_size = clusters; | |
1235 | ||
1236 | if (phys_cpos) { | |
1237 | /* | |
1238 | * We already have an allocation at this | |
1239 | * region so we can safely skip it. | |
1240 | */ | |
1241 | goto next; | |
1242 | } | |
1243 | ||
1244 | ret = __ocfs2_extend_allocation(inode, cpos, alloc_size, 1); | |
1245 | if (ret) { | |
1246 | if (ret != -ENOSPC) | |
1247 | mlog_errno(ret); | |
1248 | goto out; | |
1249 | } | |
1250 | ||
1251 | next: | |
1252 | cpos += alloc_size; | |
1253 | clusters -= alloc_size; | |
1254 | } | |
1255 | ||
1256 | ret = 0; | |
1257 | out: | |
1258 | return ret; | |
1259 | } | |
1260 | ||
063c4561 MF |
1261 | static int __ocfs2_remove_inode_range(struct inode *inode, |
1262 | struct buffer_head *di_bh, | |
1263 | u32 cpos, u32 phys_cpos, u32 len, | |
1264 | struct ocfs2_cached_dealloc_ctxt *dealloc) | |
1265 | { | |
1266 | int ret; | |
1267 | u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos); | |
1268 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1269 | struct inode *tl_inode = osb->osb_tl_inode; | |
1270 | handle_t *handle; | |
1271 | struct ocfs2_alloc_context *meta_ac = NULL; | |
1272 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; | |
1273 | ||
1274 | ret = ocfs2_lock_allocators(inode, di, 0, 1, NULL, &meta_ac); | |
1275 | if (ret) { | |
1276 | mlog_errno(ret); | |
1277 | return ret; | |
1278 | } | |
1279 | ||
1280 | mutex_lock(&tl_inode->i_mutex); | |
1281 | ||
1282 | if (ocfs2_truncate_log_needs_flush(osb)) { | |
1283 | ret = __ocfs2_flush_truncate_log(osb); | |
1284 | if (ret < 0) { | |
1285 | mlog_errno(ret); | |
1286 | goto out; | |
1287 | } | |
1288 | } | |
1289 | ||
1290 | handle = ocfs2_start_trans(osb, OCFS2_REMOVE_EXTENT_CREDITS); | |
1291 | if (handle == NULL) { | |
1292 | ret = -ENOMEM; | |
1293 | mlog_errno(ret); | |
1294 | goto out; | |
1295 | } | |
1296 | ||
1297 | ret = ocfs2_journal_access(handle, inode, di_bh, | |
1298 | OCFS2_JOURNAL_ACCESS_WRITE); | |
1299 | if (ret) { | |
1300 | mlog_errno(ret); | |
1301 | goto out; | |
1302 | } | |
1303 | ||
1304 | ret = ocfs2_remove_extent(inode, di_bh, cpos, len, handle, meta_ac, | |
1305 | dealloc); | |
1306 | if (ret) { | |
1307 | mlog_errno(ret); | |
1308 | goto out_commit; | |
1309 | } | |
1310 | ||
1311 | OCFS2_I(inode)->ip_clusters -= len; | |
1312 | di->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters); | |
1313 | ||
1314 | ret = ocfs2_journal_dirty(handle, di_bh); | |
1315 | if (ret) { | |
1316 | mlog_errno(ret); | |
1317 | goto out_commit; | |
1318 | } | |
1319 | ||
1320 | ret = ocfs2_truncate_log_append(osb, handle, phys_blkno, len); | |
1321 | if (ret) | |
1322 | mlog_errno(ret); | |
1323 | ||
1324 | out_commit: | |
1325 | ocfs2_commit_trans(osb, handle); | |
1326 | out: | |
1327 | mutex_unlock(&tl_inode->i_mutex); | |
1328 | ||
1329 | if (meta_ac) | |
1330 | ocfs2_free_alloc_context(meta_ac); | |
1331 | ||
1332 | return ret; | |
1333 | } | |
1334 | ||
1335 | /* | |
1336 | * Truncate a byte range, avoiding pages within partial clusters. This | |
1337 | * preserves those pages for the zeroing code to write to. | |
1338 | */ | |
1339 | static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start, | |
1340 | u64 byte_len) | |
1341 | { | |
1342 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1343 | loff_t start, end; | |
1344 | struct address_space *mapping = inode->i_mapping; | |
1345 | ||
1346 | start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start); | |
1347 | end = byte_start + byte_len; | |
1348 | end = end & ~(osb->s_clustersize - 1); | |
1349 | ||
1350 | if (start < end) { | |
1351 | unmap_mapping_range(mapping, start, end - start, 0); | |
1352 | truncate_inode_pages_range(mapping, start, end - 1); | |
1353 | } | |
1354 | } | |
1355 | ||
1356 | static int ocfs2_zero_partial_clusters(struct inode *inode, | |
1357 | u64 start, u64 len) | |
1358 | { | |
1359 | int ret = 0; | |
1360 | u64 tmpend, end = start + len; | |
1361 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1362 | unsigned int csize = osb->s_clustersize; | |
1363 | handle_t *handle; | |
1364 | ||
1365 | /* | |
1366 | * The "start" and "end" values are NOT necessarily part of | |
1367 | * the range whose allocation is being deleted. Rather, this | |
1368 | * is what the user passed in with the request. We must zero | |
1369 | * partial clusters here. There's no need to worry about | |
1370 | * physical allocation - the zeroing code knows to skip holes. | |
1371 | */ | |
1372 | mlog(0, "byte start: %llu, end: %llu\n", | |
1373 | (unsigned long long)start, (unsigned long long)end); | |
1374 | ||
1375 | /* | |
1376 | * If both edges are on a cluster boundary then there's no | |
1377 | * zeroing required as the region is part of the allocation to | |
1378 | * be truncated. | |
1379 | */ | |
1380 | if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0) | |
1381 | goto out; | |
1382 | ||
1383 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); | |
1384 | if (handle == NULL) { | |
1385 | ret = -ENOMEM; | |
1386 | mlog_errno(ret); | |
1387 | goto out; | |
1388 | } | |
1389 | ||
1390 | /* | |
1391 | * We want to get the byte offset of the end of the 1st cluster. | |
1392 | */ | |
1393 | tmpend = (u64)osb->s_clustersize + (start & ~(osb->s_clustersize - 1)); | |
1394 | if (tmpend > end) | |
1395 | tmpend = end; | |
1396 | ||
1397 | mlog(0, "1st range: start: %llu, tmpend: %llu\n", | |
1398 | (unsigned long long)start, (unsigned long long)tmpend); | |
1399 | ||
1400 | ret = ocfs2_zero_range_for_truncate(inode, handle, start, tmpend); | |
1401 | if (ret) | |
1402 | mlog_errno(ret); | |
1403 | ||
1404 | if (tmpend < end) { | |
1405 | /* | |
1406 | * This may make start and end equal, but the zeroing | |
1407 | * code will skip any work in that case so there's no | |
1408 | * need to catch it up here. | |
1409 | */ | |
1410 | start = end & ~(osb->s_clustersize - 1); | |
1411 | ||
1412 | mlog(0, "2nd range: start: %llu, end: %llu\n", | |
1413 | (unsigned long long)start, (unsigned long long)end); | |
1414 | ||
1415 | ret = ocfs2_zero_range_for_truncate(inode, handle, start, end); | |
1416 | if (ret) | |
1417 | mlog_errno(ret); | |
1418 | } | |
1419 | ||
1420 | ocfs2_commit_trans(osb, handle); | |
1421 | out: | |
1422 | return ret; | |
1423 | } | |
1424 | ||
1425 | static int ocfs2_remove_inode_range(struct inode *inode, | |
1426 | struct buffer_head *di_bh, u64 byte_start, | |
1427 | u64 byte_len) | |
1428 | { | |
1429 | int ret = 0; | |
1430 | u32 trunc_start, trunc_len, cpos, phys_cpos, alloc_size; | |
1431 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1432 | struct ocfs2_cached_dealloc_ctxt dealloc; | |
1433 | ||
1434 | ocfs2_init_dealloc_ctxt(&dealloc); | |
1435 | ||
1436 | if (byte_len == 0) | |
1437 | return 0; | |
1438 | ||
1439 | trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start); | |
1440 | trunc_len = (byte_start + byte_len) >> osb->s_clustersize_bits; | |
1441 | if (trunc_len >= trunc_start) | |
1442 | trunc_len -= trunc_start; | |
1443 | else | |
1444 | trunc_len = 0; | |
1445 | ||
1446 | mlog(0, "Inode: %llu, start: %llu, len: %llu, cstart: %u, clen: %u\n", | |
1447 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
1448 | (unsigned long long)byte_start, | |
1449 | (unsigned long long)byte_len, trunc_start, trunc_len); | |
1450 | ||
1451 | ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len); | |
1452 | if (ret) { | |
1453 | mlog_errno(ret); | |
1454 | goto out; | |
1455 | } | |
1456 | ||
1457 | cpos = trunc_start; | |
1458 | while (trunc_len) { | |
1459 | ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, | |
1460 | &alloc_size, NULL); | |
1461 | if (ret) { | |
1462 | mlog_errno(ret); | |
1463 | goto out; | |
1464 | } | |
1465 | ||
1466 | if (alloc_size > trunc_len) | |
1467 | alloc_size = trunc_len; | |
1468 | ||
1469 | /* Only do work for non-holes */ | |
1470 | if (phys_cpos != 0) { | |
1471 | ret = __ocfs2_remove_inode_range(inode, di_bh, cpos, | |
1472 | phys_cpos, alloc_size, | |
1473 | &dealloc); | |
1474 | if (ret) { | |
1475 | mlog_errno(ret); | |
1476 | goto out; | |
1477 | } | |
1478 | } | |
1479 | ||
1480 | cpos += alloc_size; | |
1481 | trunc_len -= alloc_size; | |
1482 | } | |
1483 | ||
1484 | ocfs2_truncate_cluster_pages(inode, byte_start, byte_len); | |
1485 | ||
1486 | out: | |
1487 | ocfs2_schedule_truncate_log_flush(osb, 1); | |
1488 | ocfs2_run_deallocs(osb, &dealloc); | |
1489 | ||
1490 | return ret; | |
1491 | } | |
1492 | ||
8659ac25 TY |
1493 | static int ocfs2_prepare_inode_for_write(struct dentry *dentry, |
1494 | loff_t *ppos, | |
1495 | size_t count, | |
9517bac6 MF |
1496 | int appending, |
1497 | int *direct_io) | |
ccd979bd | 1498 | { |
8659ac25 TY |
1499 | int ret = 0, meta_level = appending; |
1500 | struct inode *inode = dentry->d_inode; | |
ccd979bd | 1501 | u32 clusters; |
ccd979bd | 1502 | loff_t newsize, saved_pos; |
ccd979bd | 1503 | |
ccd979bd MF |
1504 | /* |
1505 | * We sample i_size under a read level meta lock to see if our write | |
1506 | * is extending the file, if it is we back off and get a write level | |
1507 | * meta lock. | |
1508 | */ | |
ccd979bd | 1509 | for(;;) { |
4bcec184 | 1510 | ret = ocfs2_meta_lock(inode, NULL, meta_level); |
ccd979bd MF |
1511 | if (ret < 0) { |
1512 | meta_level = -1; | |
1513 | mlog_errno(ret); | |
1514 | goto out; | |
1515 | } | |
1516 | ||
1517 | /* Clear suid / sgid if necessary. We do this here | |
1518 | * instead of later in the write path because | |
1519 | * remove_suid() calls ->setattr without any hint that | |
1520 | * we may have already done our cluster locking. Since | |
1521 | * ocfs2_setattr() *must* take cluster locks to | |
1522 | * proceeed, this will lead us to recursively lock the | |
1523 | * inode. There's also the dinode i_size state which | |
1524 | * can be lost via setattr during extending writes (we | |
1525 | * set inode->i_size at the end of a write. */ | |
8659ac25 | 1526 | if (should_remove_suid(dentry)) { |
ccd979bd MF |
1527 | if (meta_level == 0) { |
1528 | ocfs2_meta_unlock(inode, meta_level); | |
1529 | meta_level = 1; | |
1530 | continue; | |
1531 | } | |
1532 | ||
1533 | ret = ocfs2_write_remove_suid(inode); | |
1534 | if (ret < 0) { | |
1535 | mlog_errno(ret); | |
8659ac25 | 1536 | goto out_unlock; |
ccd979bd MF |
1537 | } |
1538 | } | |
1539 | ||
1540 | /* work on a copy of ppos until we're sure that we won't have | |
1541 | * to recalculate it due to relocking. */ | |
8659ac25 | 1542 | if (appending) { |
ccd979bd MF |
1543 | saved_pos = i_size_read(inode); |
1544 | mlog(0, "O_APPEND: inode->i_size=%llu\n", saved_pos); | |
1545 | } else { | |
8659ac25 | 1546 | saved_pos = *ppos; |
ccd979bd | 1547 | } |
3a0782d0 | 1548 | |
9517bac6 MF |
1549 | if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))) { |
1550 | loff_t end = saved_pos + count; | |
1551 | ||
1552 | /* | |
1553 | * Skip the O_DIRECT checks if we don't need | |
1554 | * them. | |
1555 | */ | |
1556 | if (!direct_io || !(*direct_io)) | |
1557 | break; | |
1558 | ||
1559 | /* | |
1560 | * Allowing concurrent direct writes means | |
1561 | * i_size changes wouldn't be synchronized, so | |
1562 | * one node could wind up truncating another | |
1563 | * nodes writes. | |
1564 | */ | |
1565 | if (end > i_size_read(inode)) { | |
1566 | *direct_io = 0; | |
1567 | break; | |
1568 | } | |
1569 | ||
1570 | /* | |
1571 | * We don't fill holes during direct io, so | |
1572 | * check for them here. If any are found, the | |
1573 | * caller will have to retake some cluster | |
1574 | * locks and initiate the io as buffered. | |
1575 | */ | |
1576 | ret = ocfs2_check_range_for_holes(inode, saved_pos, | |
1577 | count); | |
1578 | if (ret == 1) { | |
1579 | *direct_io = 0; | |
1580 | ret = 0; | |
1581 | } else if (ret < 0) | |
1582 | mlog_errno(ret); | |
1583 | break; | |
1584 | } | |
1585 | ||
3a0782d0 MF |
1586 | /* |
1587 | * The rest of this loop is concerned with legacy file | |
1588 | * systems which don't support sparse files. | |
1589 | */ | |
3a0782d0 | 1590 | |
8659ac25 | 1591 | newsize = count + saved_pos; |
ccd979bd | 1592 | |
215c7f9f MF |
1593 | mlog(0, "pos=%lld newsize=%lld cursize=%lld\n", |
1594 | (long long) saved_pos, (long long) newsize, | |
1595 | (long long) i_size_read(inode)); | |
ccd979bd MF |
1596 | |
1597 | /* No need for a higher level metadata lock if we're | |
1598 | * never going past i_size. */ | |
1599 | if (newsize <= i_size_read(inode)) | |
1600 | break; | |
1601 | ||
1602 | if (meta_level == 0) { | |
1603 | ocfs2_meta_unlock(inode, meta_level); | |
1604 | meta_level = 1; | |
1605 | continue; | |
1606 | } | |
1607 | ||
1608 | spin_lock(&OCFS2_I(inode)->ip_lock); | |
1609 | clusters = ocfs2_clusters_for_bytes(inode->i_sb, newsize) - | |
1610 | OCFS2_I(inode)->ip_clusters; | |
1611 | spin_unlock(&OCFS2_I(inode)->ip_lock); | |
1612 | ||
1613 | mlog(0, "Writing at EOF, may need more allocation: " | |
215c7f9f MF |
1614 | "i_size = %lld, newsize = %lld, need %u clusters\n", |
1615 | (long long) i_size_read(inode), (long long) newsize, | |
1616 | clusters); | |
ccd979bd MF |
1617 | |
1618 | /* We only want to continue the rest of this loop if | |
1619 | * our extend will actually require more | |
1620 | * allocation. */ | |
1621 | if (!clusters) | |
1622 | break; | |
1623 | ||
8659ac25 | 1624 | ret = ocfs2_extend_file(inode, NULL, newsize, count); |
ccd979bd MF |
1625 | if (ret < 0) { |
1626 | if (ret != -ENOSPC) | |
1627 | mlog_errno(ret); | |
8659ac25 | 1628 | goto out_unlock; |
ccd979bd | 1629 | } |
ccd979bd MF |
1630 | break; |
1631 | } | |
1632 | ||
8659ac25 TY |
1633 | if (appending) |
1634 | *ppos = saved_pos; | |
1635 | ||
1636 | out_unlock: | |
ccd979bd | 1637 | ocfs2_meta_unlock(inode, meta_level); |
8659ac25 TY |
1638 | |
1639 | out: | |
1640 | return ret; | |
1641 | } | |
1642 | ||
9517bac6 MF |
1643 | static inline void |
1644 | ocfs2_set_next_iovec(const struct iovec **iovp, size_t *basep, size_t bytes) | |
1645 | { | |
1646 | const struct iovec *iov = *iovp; | |
1647 | size_t base = *basep; | |
1648 | ||
1649 | do { | |
1650 | int copy = min(bytes, iov->iov_len - base); | |
1651 | ||
1652 | bytes -= copy; | |
1653 | base += copy; | |
1654 | if (iov->iov_len == base) { | |
1655 | iov++; | |
1656 | base = 0; | |
1657 | } | |
1658 | } while (bytes); | |
1659 | *iovp = iov; | |
1660 | *basep = base; | |
1661 | } | |
1662 | ||
3a307ffc | 1663 | static struct page * ocfs2_get_write_source(char **ret_src_buf, |
9517bac6 MF |
1664 | const struct iovec *cur_iov, |
1665 | size_t iov_offset) | |
1666 | { | |
1667 | int ret; | |
3a307ffc | 1668 | char *buf = cur_iov->iov_base + iov_offset; |
9517bac6 | 1669 | struct page *src_page = NULL; |
3a307ffc | 1670 | unsigned long off; |
9517bac6 | 1671 | |
3a307ffc | 1672 | off = (unsigned long)(buf) & ~PAGE_CACHE_MASK; |
9517bac6 MF |
1673 | |
1674 | if (!segment_eq(get_fs(), KERNEL_DS)) { | |
1675 | /* | |
1676 | * Pull in the user page. We want to do this outside | |
1677 | * of the meta data locks in order to preserve locking | |
1678 | * order in case of page fault. | |
1679 | */ | |
1680 | ret = get_user_pages(current, current->mm, | |
1681 | (unsigned long)buf & PAGE_CACHE_MASK, 1, | |
1682 | 0, 0, &src_page, NULL); | |
1683 | if (ret == 1) | |
3a307ffc | 1684 | *ret_src_buf = kmap(src_page) + off; |
9517bac6 MF |
1685 | else |
1686 | src_page = ERR_PTR(-EFAULT); | |
1687 | } else { | |
3a307ffc | 1688 | *ret_src_buf = buf; |
9517bac6 MF |
1689 | } |
1690 | ||
1691 | return src_page; | |
1692 | } | |
1693 | ||
3a307ffc | 1694 | static void ocfs2_put_write_source(struct page *page) |
9517bac6 MF |
1695 | { |
1696 | if (page) { | |
1697 | kunmap(page); | |
1698 | page_cache_release(page); | |
1699 | } | |
1700 | } | |
1701 | ||
1702 | static ssize_t ocfs2_file_buffered_write(struct file *file, loff_t *ppos, | |
1703 | const struct iovec *iov, | |
1704 | unsigned long nr_segs, | |
1705 | size_t count, | |
1706 | ssize_t o_direct_written) | |
1707 | { | |
1708 | int ret = 0; | |
1709 | ssize_t copied, total = 0; | |
3a307ffc MF |
1710 | size_t iov_offset = 0, bytes; |
1711 | loff_t pos; | |
9517bac6 | 1712 | const struct iovec *cur_iov = iov; |
3a307ffc MF |
1713 | struct page *user_page, *page; |
1714 | char *buf, *dst; | |
1715 | void *fsdata; | |
9517bac6 MF |
1716 | |
1717 | /* | |
1718 | * handle partial DIO write. Adjust cur_iov if needed. | |
1719 | */ | |
1720 | ocfs2_set_next_iovec(&cur_iov, &iov_offset, o_direct_written); | |
1721 | ||
1722 | do { | |
3a307ffc | 1723 | pos = *ppos; |
9517bac6 | 1724 | |
3a307ffc MF |
1725 | user_page = ocfs2_get_write_source(&buf, cur_iov, iov_offset); |
1726 | if (IS_ERR(user_page)) { | |
1727 | ret = PTR_ERR(user_page); | |
9517bac6 MF |
1728 | goto out; |
1729 | } | |
1730 | ||
3a307ffc MF |
1731 | /* Stay within our page boundaries */ |
1732 | bytes = min((PAGE_CACHE_SIZE - ((unsigned long)pos & ~PAGE_CACHE_MASK)), | |
1733 | (PAGE_CACHE_SIZE - ((unsigned long)buf & ~PAGE_CACHE_MASK))); | |
1734 | /* Stay within the vector boundary */ | |
1735 | bytes = min_t(size_t, bytes, cur_iov->iov_len - iov_offset); | |
1736 | /* Stay within count */ | |
1737 | bytes = min(bytes, count); | |
1738 | ||
1739 | page = NULL; | |
1740 | ret = ocfs2_write_begin(file, file->f_mapping, pos, bytes, 0, | |
1741 | &page, &fsdata); | |
1742 | if (ret) { | |
1743 | mlog_errno(ret); | |
1744 | goto out; | |
1745 | } | |
9517bac6 | 1746 | |
3a307ffc MF |
1747 | dst = kmap_atomic(page, KM_USER0); |
1748 | memcpy(dst + (pos & (PAGE_CACHE_SIZE - 1)), buf, bytes); | |
1749 | kunmap_atomic(dst, KM_USER0); | |
1750 | flush_dcache_page(page); | |
1751 | ocfs2_put_write_source(user_page); | |
9517bac6 | 1752 | |
3a307ffc MF |
1753 | copied = ocfs2_write_end(file, file->f_mapping, pos, bytes, |
1754 | bytes, page, fsdata); | |
9517bac6 MF |
1755 | if (copied < 0) { |
1756 | mlog_errno(copied); | |
1757 | ret = copied; | |
1758 | goto out; | |
1759 | } | |
1760 | ||
1761 | total += copied; | |
3a307ffc | 1762 | *ppos = pos + copied; |
9517bac6 MF |
1763 | count -= copied; |
1764 | ||
1765 | ocfs2_set_next_iovec(&cur_iov, &iov_offset, copied); | |
1766 | } while(count); | |
1767 | ||
1768 | out: | |
1769 | return total ? total : ret; | |
1770 | } | |
1771 | ||
8659ac25 TY |
1772 | static ssize_t ocfs2_file_aio_write(struct kiocb *iocb, |
1773 | const struct iovec *iov, | |
1774 | unsigned long nr_segs, | |
1775 | loff_t pos) | |
1776 | { | |
9517bac6 MF |
1777 | int ret, direct_io, appending, rw_level, have_alloc_sem = 0; |
1778 | int can_do_direct, sync = 0; | |
1779 | ssize_t written = 0; | |
1780 | size_t ocount; /* original count */ | |
1781 | size_t count; /* after file limit checks */ | |
1782 | loff_t *ppos = &iocb->ki_pos; | |
1783 | struct file *file = iocb->ki_filp; | |
1784 | struct inode *inode = file->f_path.dentry->d_inode; | |
1785 | ||
1786 | mlog_entry("(0x%p, %u, '%.*s')\n", file, | |
8659ac25 | 1787 | (unsigned int)nr_segs, |
9517bac6 MF |
1788 | file->f_path.dentry->d_name.len, |
1789 | file->f_path.dentry->d_name.name); | |
8659ac25 | 1790 | |
8659ac25 TY |
1791 | if (iocb->ki_left == 0) |
1792 | return 0; | |
1793 | ||
d9b08b9e | 1794 | ret = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ); |
9517bac6 MF |
1795 | if (ret) |
1796 | return ret; | |
1797 | ||
1798 | count = ocount; | |
1799 | ||
1800 | vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE); | |
1801 | ||
1802 | appending = file->f_flags & O_APPEND ? 1 : 0; | |
1803 | direct_io = file->f_flags & O_DIRECT ? 1 : 0; | |
1804 | ||
8659ac25 | 1805 | mutex_lock(&inode->i_mutex); |
9517bac6 MF |
1806 | |
1807 | relock: | |
8659ac25 | 1808 | /* to match setattr's i_mutex -> i_alloc_sem -> rw_lock ordering */ |
9517bac6 | 1809 | if (direct_io) { |
8659ac25 | 1810 | down_read(&inode->i_alloc_sem); |
9517bac6 | 1811 | have_alloc_sem = 1; |
8659ac25 TY |
1812 | } |
1813 | ||
1814 | /* concurrent O_DIRECT writes are allowed */ | |
9517bac6 | 1815 | rw_level = !direct_io; |
8659ac25 TY |
1816 | ret = ocfs2_rw_lock(inode, rw_level); |
1817 | if (ret < 0) { | |
8659ac25 | 1818 | mlog_errno(ret); |
9517bac6 | 1819 | goto out_sems; |
8659ac25 TY |
1820 | } |
1821 | ||
9517bac6 MF |
1822 | can_do_direct = direct_io; |
1823 | ret = ocfs2_prepare_inode_for_write(file->f_path.dentry, ppos, | |
1824 | iocb->ki_left, appending, | |
1825 | &can_do_direct); | |
8659ac25 TY |
1826 | if (ret < 0) { |
1827 | mlog_errno(ret); | |
1828 | goto out; | |
1829 | } | |
ccd979bd | 1830 | |
9517bac6 MF |
1831 | /* |
1832 | * We can't complete the direct I/O as requested, fall back to | |
1833 | * buffered I/O. | |
1834 | */ | |
1835 | if (direct_io && !can_do_direct) { | |
1836 | ocfs2_rw_unlock(inode, rw_level); | |
1837 | up_read(&inode->i_alloc_sem); | |
1838 | ||
1839 | have_alloc_sem = 0; | |
1840 | rw_level = -1; | |
1841 | ||
1842 | direct_io = 0; | |
1843 | sync = 1; | |
1844 | goto relock; | |
1845 | } | |
1846 | ||
1847 | if (!sync && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) | |
1848 | sync = 1; | |
1849 | ||
1850 | /* | |
1851 | * XXX: Is it ok to execute these checks a second time? | |
1852 | */ | |
1853 | ret = generic_write_checks(file, ppos, &count, S_ISBLK(inode->i_mode)); | |
1854 | if (ret) | |
1855 | goto out; | |
1856 | ||
1857 | /* | |
1858 | * Set pos so that sync_page_range_nolock() below understands | |
1859 | * where to start from. We might've moved it around via the | |
1860 | * calls above. The range we want to actually sync starts from | |
1861 | * *ppos here. | |
1862 | * | |
1863 | */ | |
1864 | pos = *ppos; | |
1865 | ||
ccd979bd | 1866 | /* communicate with ocfs2_dio_end_io */ |
7cdfc3a1 | 1867 | ocfs2_iocb_set_rw_locked(iocb, rw_level); |
ccd979bd | 1868 | |
9517bac6 MF |
1869 | if (direct_io) { |
1870 | written = generic_file_direct_write(iocb, iov, &nr_segs, *ppos, | |
1871 | ppos, count, ocount); | |
1872 | if (written < 0) { | |
1873 | ret = written; | |
1874 | goto out_dio; | |
1875 | } | |
1876 | } else { | |
1877 | written = ocfs2_file_buffered_write(file, ppos, iov, nr_segs, | |
1878 | count, written); | |
1879 | if (written < 0) { | |
1880 | ret = written; | |
1881 | if (ret != -EFAULT || ret != -ENOSPC) | |
1882 | mlog_errno(ret); | |
1883 | goto out; | |
1884 | } | |
1885 | } | |
ccd979bd | 1886 | |
9517bac6 | 1887 | out_dio: |
ccd979bd | 1888 | /* buffered aio wouldn't have proper lock coverage today */ |
9517bac6 | 1889 | BUG_ON(ret == -EIOCBQUEUED && !(file->f_flags & O_DIRECT)); |
ccd979bd MF |
1890 | |
1891 | /* | |
1892 | * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io | |
1893 | * function pointer which is called when o_direct io completes so that | |
1894 | * it can unlock our rw lock. (it's the clustered equivalent of | |
1895 | * i_alloc_sem; protects truncate from racing with pending ios). | |
1896 | * Unfortunately there are error cases which call end_io and others | |
1897 | * that don't. so we don't have to unlock the rw_lock if either an | |
1898 | * async dio is going to do it in the future or an end_io after an | |
1899 | * error has already done it. | |
1900 | */ | |
1901 | if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) { | |
1902 | rw_level = -1; | |
1903 | have_alloc_sem = 0; | |
1904 | } | |
1905 | ||
1906 | out: | |
9517bac6 MF |
1907 | if (rw_level != -1) |
1908 | ocfs2_rw_unlock(inode, rw_level); | |
1909 | ||
1910 | out_sems: | |
ccd979bd MF |
1911 | if (have_alloc_sem) |
1912 | up_read(&inode->i_alloc_sem); | |
9517bac6 MF |
1913 | |
1914 | if (written > 0 && sync) { | |
1915 | ssize_t err; | |
1916 | ||
1917 | err = sync_page_range_nolock(inode, file->f_mapping, pos, count); | |
1918 | if (err < 0) | |
1919 | written = err; | |
1920 | } | |
1921 | ||
1b1dcc1b | 1922 | mutex_unlock(&inode->i_mutex); |
ccd979bd MF |
1923 | |
1924 | mlog_exit(ret); | |
9517bac6 | 1925 | return written ? written : ret; |
ccd979bd MF |
1926 | } |
1927 | ||
6af67d82 MF |
1928 | static int ocfs2_splice_write_actor(struct pipe_inode_info *pipe, |
1929 | struct pipe_buffer *buf, | |
1930 | struct splice_desc *sd) | |
1931 | { | |
3a307ffc | 1932 | int ret, count; |
6af67d82 | 1933 | ssize_t copied = 0; |
3a307ffc MF |
1934 | struct file *file = sd->u.file; |
1935 | unsigned int offset; | |
1936 | struct page *page = NULL; | |
1937 | void *fsdata; | |
1938 | char *src, *dst; | |
6af67d82 | 1939 | |
cac36bb0 | 1940 | ret = buf->ops->confirm(pipe, buf); |
6af67d82 MF |
1941 | if (ret) |
1942 | goto out; | |
1943 | ||
3a307ffc | 1944 | offset = sd->pos & ~PAGE_CACHE_MASK; |
6af67d82 | 1945 | count = sd->len; |
3a307ffc MF |
1946 | if (count + offset > PAGE_CACHE_SIZE) |
1947 | count = PAGE_CACHE_SIZE - offset; | |
6af67d82 | 1948 | |
3a307ffc MF |
1949 | ret = ocfs2_write_begin(file, file->f_mapping, sd->pos, count, 0, |
1950 | &page, &fsdata); | |
1951 | if (ret) { | |
1952 | mlog_errno(ret); | |
1953 | goto out; | |
1954 | } | |
6af67d82 | 1955 | |
3a307ffc MF |
1956 | src = buf->ops->map(pipe, buf, 1); |
1957 | dst = kmap_atomic(page, KM_USER1); | |
1958 | memcpy(dst + offset, src + buf->offset, count); | |
1959 | kunmap_atomic(page, KM_USER1); | |
1960 | buf->ops->unmap(pipe, buf, src); | |
6af67d82 | 1961 | |
3a307ffc MF |
1962 | copied = ocfs2_write_end(file, file->f_mapping, sd->pos, count, count, |
1963 | page, fsdata); | |
1964 | if (copied < 0) { | |
1965 | mlog_errno(copied); | |
1966 | ret = copied; | |
1967 | goto out; | |
1968 | } | |
6af67d82 MF |
1969 | out: |
1970 | ||
3a307ffc | 1971 | return copied ? copied : ret; |
6af67d82 MF |
1972 | } |
1973 | ||
1974 | static ssize_t __ocfs2_file_splice_write(struct pipe_inode_info *pipe, | |
1975 | struct file *out, | |
1976 | loff_t *ppos, | |
1977 | size_t len, | |
1978 | unsigned int flags) | |
1979 | { | |
1980 | int ret, err; | |
1981 | struct address_space *mapping = out->f_mapping; | |
1982 | struct inode *inode = mapping->host; | |
c66ab6fa JA |
1983 | struct splice_desc sd = { |
1984 | .total_len = len, | |
1985 | .flags = flags, | |
1986 | .pos = *ppos, | |
6a14b90b | 1987 | .u.file = out, |
c66ab6fa JA |
1988 | }; |
1989 | ||
1990 | ret = __splice_from_pipe(pipe, &sd, ocfs2_splice_write_actor); | |
6af67d82 MF |
1991 | if (ret > 0) { |
1992 | *ppos += ret; | |
1993 | ||
1994 | if (unlikely((out->f_flags & O_SYNC) || IS_SYNC(inode))) { | |
1995 | err = generic_osync_inode(inode, mapping, | |
1996 | OSYNC_METADATA|OSYNC_DATA); | |
1997 | if (err) | |
1998 | ret = err; | |
1999 | } | |
2000 | } | |
2001 | ||
2002 | return ret; | |
2003 | } | |
2004 | ||
8659ac25 TY |
2005 | static ssize_t ocfs2_file_splice_write(struct pipe_inode_info *pipe, |
2006 | struct file *out, | |
2007 | loff_t *ppos, | |
2008 | size_t len, | |
2009 | unsigned int flags) | |
2010 | { | |
2011 | int ret; | |
d28c9174 | 2012 | struct inode *inode = out->f_path.dentry->d_inode; |
8659ac25 TY |
2013 | |
2014 | mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", out, pipe, | |
2015 | (unsigned int)len, | |
d28c9174 JS |
2016 | out->f_path.dentry->d_name.len, |
2017 | out->f_path.dentry->d_name.name); | |
8659ac25 TY |
2018 | |
2019 | inode_double_lock(inode, pipe->inode); | |
2020 | ||
2021 | ret = ocfs2_rw_lock(inode, 1); | |
2022 | if (ret < 0) { | |
2023 | mlog_errno(ret); | |
2024 | goto out; | |
2025 | } | |
2026 | ||
9517bac6 MF |
2027 | ret = ocfs2_prepare_inode_for_write(out->f_path.dentry, ppos, len, 0, |
2028 | NULL); | |
8659ac25 TY |
2029 | if (ret < 0) { |
2030 | mlog_errno(ret); | |
2031 | goto out_unlock; | |
2032 | } | |
2033 | ||
2034 | /* ok, we're done with i_size and alloc work */ | |
6af67d82 | 2035 | ret = __ocfs2_file_splice_write(pipe, out, ppos, len, flags); |
8659ac25 TY |
2036 | |
2037 | out_unlock: | |
2038 | ocfs2_rw_unlock(inode, 1); | |
2039 | out: | |
2040 | inode_double_unlock(inode, pipe->inode); | |
2041 | ||
2042 | mlog_exit(ret); | |
2043 | return ret; | |
2044 | } | |
2045 | ||
2046 | static ssize_t ocfs2_file_splice_read(struct file *in, | |
2047 | loff_t *ppos, | |
2048 | struct pipe_inode_info *pipe, | |
2049 | size_t len, | |
2050 | unsigned int flags) | |
2051 | { | |
2052 | int ret = 0; | |
d28c9174 | 2053 | struct inode *inode = in->f_path.dentry->d_inode; |
8659ac25 TY |
2054 | |
2055 | mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", in, pipe, | |
2056 | (unsigned int)len, | |
d28c9174 JS |
2057 | in->f_path.dentry->d_name.len, |
2058 | in->f_path.dentry->d_name.name); | |
8659ac25 TY |
2059 | |
2060 | /* | |
2061 | * See the comment in ocfs2_file_aio_read() | |
2062 | */ | |
2063 | ret = ocfs2_meta_lock(inode, NULL, 0); | |
2064 | if (ret < 0) { | |
2065 | mlog_errno(ret); | |
2066 | goto bail; | |
2067 | } | |
2068 | ocfs2_meta_unlock(inode, 0); | |
2069 | ||
2070 | ret = generic_file_splice_read(in, ppos, pipe, len, flags); | |
2071 | ||
2072 | bail: | |
2073 | mlog_exit(ret); | |
2074 | return ret; | |
2075 | } | |
2076 | ||
ccd979bd | 2077 | static ssize_t ocfs2_file_aio_read(struct kiocb *iocb, |
027445c3 BP |
2078 | const struct iovec *iov, |
2079 | unsigned long nr_segs, | |
ccd979bd MF |
2080 | loff_t pos) |
2081 | { | |
25899dee | 2082 | int ret = 0, rw_level = -1, have_alloc_sem = 0, lock_level = 0; |
ccd979bd | 2083 | struct file *filp = iocb->ki_filp; |
d28c9174 | 2084 | struct inode *inode = filp->f_path.dentry->d_inode; |
ccd979bd | 2085 | |
027445c3 BP |
2086 | mlog_entry("(0x%p, %u, '%.*s')\n", filp, |
2087 | (unsigned int)nr_segs, | |
d28c9174 JS |
2088 | filp->f_path.dentry->d_name.len, |
2089 | filp->f_path.dentry->d_name.name); | |
ccd979bd MF |
2090 | |
2091 | if (!inode) { | |
2092 | ret = -EINVAL; | |
2093 | mlog_errno(ret); | |
2094 | goto bail; | |
2095 | } | |
2096 | ||
ccd979bd MF |
2097 | /* |
2098 | * buffered reads protect themselves in ->readpage(). O_DIRECT reads | |
2099 | * need locks to protect pending reads from racing with truncate. | |
2100 | */ | |
2101 | if (filp->f_flags & O_DIRECT) { | |
2102 | down_read(&inode->i_alloc_sem); | |
2103 | have_alloc_sem = 1; | |
2104 | ||
2105 | ret = ocfs2_rw_lock(inode, 0); | |
2106 | if (ret < 0) { | |
2107 | mlog_errno(ret); | |
2108 | goto bail; | |
2109 | } | |
2110 | rw_level = 0; | |
2111 | /* communicate with ocfs2_dio_end_io */ | |
7cdfc3a1 | 2112 | ocfs2_iocb_set_rw_locked(iocb, rw_level); |
ccd979bd MF |
2113 | } |
2114 | ||
c4374f8a MF |
2115 | /* |
2116 | * We're fine letting folks race truncates and extending | |
2117 | * writes with read across the cluster, just like they can | |
2118 | * locally. Hence no rw_lock during read. | |
2119 | * | |
2120 | * Take and drop the meta data lock to update inode fields | |
2121 | * like i_size. This allows the checks down below | |
2122 | * generic_file_aio_read() a chance of actually working. | |
2123 | */ | |
25899dee | 2124 | ret = ocfs2_meta_lock_atime(inode, filp->f_vfsmnt, &lock_level); |
c4374f8a MF |
2125 | if (ret < 0) { |
2126 | mlog_errno(ret); | |
2127 | goto bail; | |
2128 | } | |
25899dee | 2129 | ocfs2_meta_unlock(inode, lock_level); |
c4374f8a | 2130 | |
027445c3 | 2131 | ret = generic_file_aio_read(iocb, iov, nr_segs, iocb->ki_pos); |
ccd979bd MF |
2132 | if (ret == -EINVAL) |
2133 | mlog(ML_ERROR, "generic_file_aio_read returned -EINVAL\n"); | |
2134 | ||
2135 | /* buffered aio wouldn't have proper lock coverage today */ | |
2136 | BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT)); | |
2137 | ||
2138 | /* see ocfs2_file_aio_write */ | |
2139 | if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) { | |
2140 | rw_level = -1; | |
2141 | have_alloc_sem = 0; | |
2142 | } | |
2143 | ||
2144 | bail: | |
2145 | if (have_alloc_sem) | |
2146 | up_read(&inode->i_alloc_sem); | |
2147 | if (rw_level != -1) | |
2148 | ocfs2_rw_unlock(inode, rw_level); | |
2149 | mlog_exit(ret); | |
2150 | ||
2151 | return ret; | |
2152 | } | |
2153 | ||
92e1d5be | 2154 | const struct inode_operations ocfs2_file_iops = { |
ccd979bd MF |
2155 | .setattr = ocfs2_setattr, |
2156 | .getattr = ocfs2_getattr, | |
d38eb8db | 2157 | .permission = ocfs2_permission, |
ccd979bd MF |
2158 | }; |
2159 | ||
92e1d5be | 2160 | const struct inode_operations ocfs2_special_file_iops = { |
ccd979bd MF |
2161 | .setattr = ocfs2_setattr, |
2162 | .getattr = ocfs2_getattr, | |
d38eb8db | 2163 | .permission = ocfs2_permission, |
ccd979bd MF |
2164 | }; |
2165 | ||
4b6f5d20 | 2166 | const struct file_operations ocfs2_fops = { |
ccd979bd MF |
2167 | .read = do_sync_read, |
2168 | .write = do_sync_write, | |
ccd979bd MF |
2169 | .mmap = ocfs2_mmap, |
2170 | .fsync = ocfs2_sync_file, | |
2171 | .release = ocfs2_file_release, | |
2172 | .open = ocfs2_file_open, | |
2173 | .aio_read = ocfs2_file_aio_read, | |
2174 | .aio_write = ocfs2_file_aio_write, | |
ca4d147e | 2175 | .ioctl = ocfs2_ioctl, |
586d232b MF |
2176 | #ifdef CONFIG_COMPAT |
2177 | .compat_ioctl = ocfs2_compat_ioctl, | |
2178 | #endif | |
8659ac25 TY |
2179 | .splice_read = ocfs2_file_splice_read, |
2180 | .splice_write = ocfs2_file_splice_write, | |
ccd979bd MF |
2181 | }; |
2182 | ||
4b6f5d20 | 2183 | const struct file_operations ocfs2_dops = { |
ccd979bd MF |
2184 | .read = generic_read_dir, |
2185 | .readdir = ocfs2_readdir, | |
2186 | .fsync = ocfs2_sync_file, | |
ca4d147e | 2187 | .ioctl = ocfs2_ioctl, |
586d232b MF |
2188 | #ifdef CONFIG_COMPAT |
2189 | .compat_ioctl = ocfs2_compat_ioctl, | |
2190 | #endif | |
ccd979bd | 2191 | }; |