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> |
385820a3 | 37 | #include <linux/falloc.h> |
ccd979bd MF |
38 | |
39 | #define MLOG_MASK_PREFIX ML_INODE | |
40 | #include <cluster/masklog.h> | |
41 | ||
42 | #include "ocfs2.h" | |
43 | ||
44 | #include "alloc.h" | |
45 | #include "aops.h" | |
46 | #include "dir.h" | |
47 | #include "dlmglue.h" | |
48 | #include "extent_map.h" | |
49 | #include "file.h" | |
50 | #include "sysfile.h" | |
51 | #include "inode.h" | |
ca4d147e | 52 | #include "ioctl.h" |
ccd979bd | 53 | #include "journal.h" |
53fc622b | 54 | #include "locks.h" |
ccd979bd MF |
55 | #include "mmap.h" |
56 | #include "suballoc.h" | |
57 | #include "super.h" | |
58 | ||
59 | #include "buffer_head_io.h" | |
60 | ||
61 | static int ocfs2_sync_inode(struct inode *inode) | |
62 | { | |
63 | filemap_fdatawrite(inode->i_mapping); | |
64 | return sync_mapping_buffers(inode->i_mapping); | |
65 | } | |
66 | ||
53fc622b MF |
67 | static int ocfs2_init_file_private(struct inode *inode, struct file *file) |
68 | { | |
69 | struct ocfs2_file_private *fp; | |
70 | ||
71 | fp = kzalloc(sizeof(struct ocfs2_file_private), GFP_KERNEL); | |
72 | if (!fp) | |
73 | return -ENOMEM; | |
74 | ||
75 | fp->fp_file = file; | |
76 | mutex_init(&fp->fp_mutex); | |
77 | ocfs2_file_lock_res_init(&fp->fp_flock, fp); | |
78 | file->private_data = fp; | |
79 | ||
80 | return 0; | |
81 | } | |
82 | ||
83 | static void ocfs2_free_file_private(struct inode *inode, struct file *file) | |
84 | { | |
85 | struct ocfs2_file_private *fp = file->private_data; | |
86 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
87 | ||
88 | if (fp) { | |
89 | ocfs2_simple_drop_lockres(osb, &fp->fp_flock); | |
90 | ocfs2_lock_res_free(&fp->fp_flock); | |
91 | kfree(fp); | |
92 | file->private_data = NULL; | |
93 | } | |
94 | } | |
95 | ||
ccd979bd MF |
96 | static int ocfs2_file_open(struct inode *inode, struct file *file) |
97 | { | |
98 | int status; | |
99 | int mode = file->f_flags; | |
100 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
101 | ||
102 | mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file, | |
d28c9174 | 103 | file->f_path.dentry->d_name.len, file->f_path.dentry->d_name.name); |
ccd979bd MF |
104 | |
105 | spin_lock(&oi->ip_lock); | |
106 | ||
107 | /* Check that the inode hasn't been wiped from disk by another | |
108 | * node. If it hasn't then we're safe as long as we hold the | |
109 | * spin lock until our increment of open count. */ | |
110 | if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) { | |
111 | spin_unlock(&oi->ip_lock); | |
112 | ||
113 | status = -ENOENT; | |
114 | goto leave; | |
115 | } | |
116 | ||
117 | if (mode & O_DIRECT) | |
118 | oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT; | |
119 | ||
120 | oi->ip_open_count++; | |
121 | spin_unlock(&oi->ip_lock); | |
53fc622b MF |
122 | |
123 | status = ocfs2_init_file_private(inode, file); | |
124 | if (status) { | |
125 | /* | |
126 | * We want to set open count back if we're failing the | |
127 | * open. | |
128 | */ | |
129 | spin_lock(&oi->ip_lock); | |
130 | oi->ip_open_count--; | |
131 | spin_unlock(&oi->ip_lock); | |
132 | } | |
133 | ||
ccd979bd MF |
134 | leave: |
135 | mlog_exit(status); | |
136 | return status; | |
137 | } | |
138 | ||
139 | static int ocfs2_file_release(struct inode *inode, struct file *file) | |
140 | { | |
141 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
142 | ||
143 | mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file, | |
d28c9174 JS |
144 | file->f_path.dentry->d_name.len, |
145 | file->f_path.dentry->d_name.name); | |
ccd979bd MF |
146 | |
147 | spin_lock(&oi->ip_lock); | |
148 | if (!--oi->ip_open_count) | |
149 | oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT; | |
150 | spin_unlock(&oi->ip_lock); | |
151 | ||
53fc622b MF |
152 | ocfs2_free_file_private(inode, file); |
153 | ||
ccd979bd MF |
154 | mlog_exit(0); |
155 | ||
156 | return 0; | |
157 | } | |
158 | ||
53fc622b MF |
159 | static int ocfs2_dir_open(struct inode *inode, struct file *file) |
160 | { | |
161 | return ocfs2_init_file_private(inode, file); | |
162 | } | |
163 | ||
164 | static int ocfs2_dir_release(struct inode *inode, struct file *file) | |
165 | { | |
166 | ocfs2_free_file_private(inode, file); | |
167 | return 0; | |
168 | } | |
169 | ||
ccd979bd MF |
170 | static int ocfs2_sync_file(struct file *file, |
171 | struct dentry *dentry, | |
172 | int datasync) | |
173 | { | |
174 | int err = 0; | |
175 | journal_t *journal; | |
176 | struct inode *inode = dentry->d_inode; | |
177 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
178 | ||
179 | mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", file, dentry, datasync, | |
180 | dentry->d_name.len, dentry->d_name.name); | |
181 | ||
182 | err = ocfs2_sync_inode(dentry->d_inode); | |
183 | if (err) | |
184 | goto bail; | |
185 | ||
186 | journal = osb->journal->j_journal; | |
187 | err = journal_force_commit(journal); | |
188 | ||
189 | bail: | |
190 | mlog_exit(err); | |
191 | ||
192 | return (err < 0) ? -EIO : 0; | |
193 | } | |
194 | ||
7f1a37e3 TY |
195 | int ocfs2_should_update_atime(struct inode *inode, |
196 | struct vfsmount *vfsmnt) | |
197 | { | |
198 | struct timespec now; | |
199 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
200 | ||
201 | if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb)) | |
202 | return 0; | |
203 | ||
204 | if ((inode->i_flags & S_NOATIME) || | |
205 | ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))) | |
206 | return 0; | |
207 | ||
6c2aad05 MF |
208 | /* |
209 | * We can be called with no vfsmnt structure - NFSD will | |
210 | * sometimes do this. | |
211 | * | |
212 | * Note that our action here is different than touch_atime() - | |
213 | * if we can't tell whether this is a noatime mount, then we | |
214 | * don't know whether to trust the value of s_atime_quantum. | |
215 | */ | |
216 | if (vfsmnt == NULL) | |
217 | return 0; | |
218 | ||
7f1a37e3 TY |
219 | if ((vfsmnt->mnt_flags & MNT_NOATIME) || |
220 | ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))) | |
221 | return 0; | |
222 | ||
7e913c53 MF |
223 | if (vfsmnt->mnt_flags & MNT_RELATIME) { |
224 | if ((timespec_compare(&inode->i_atime, &inode->i_mtime) <= 0) || | |
225 | (timespec_compare(&inode->i_atime, &inode->i_ctime) <= 0)) | |
226 | return 1; | |
227 | ||
228 | return 0; | |
229 | } | |
230 | ||
7f1a37e3 TY |
231 | now = CURRENT_TIME; |
232 | if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum)) | |
233 | return 0; | |
234 | else | |
235 | return 1; | |
236 | } | |
237 | ||
238 | int ocfs2_update_inode_atime(struct inode *inode, | |
239 | struct buffer_head *bh) | |
240 | { | |
241 | int ret; | |
242 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
243 | handle_t *handle; | |
c11e9faf | 244 | struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data; |
7f1a37e3 TY |
245 | |
246 | mlog_entry_void(); | |
247 | ||
248 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); | |
249 | if (handle == NULL) { | |
250 | ret = -ENOMEM; | |
251 | mlog_errno(ret); | |
252 | goto out; | |
253 | } | |
254 | ||
c11e9faf MF |
255 | ret = ocfs2_journal_access(handle, inode, bh, |
256 | OCFS2_JOURNAL_ACCESS_WRITE); | |
257 | if (ret) { | |
258 | mlog_errno(ret); | |
259 | goto out_commit; | |
260 | } | |
261 | ||
262 | /* | |
263 | * Don't use ocfs2_mark_inode_dirty() here as we don't always | |
264 | * have i_mutex to guard against concurrent changes to other | |
265 | * inode fields. | |
266 | */ | |
7f1a37e3 | 267 | inode->i_atime = CURRENT_TIME; |
c11e9faf MF |
268 | di->i_atime = cpu_to_le64(inode->i_atime.tv_sec); |
269 | di->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec); | |
270 | ||
271 | ret = ocfs2_journal_dirty(handle, bh); | |
7f1a37e3 TY |
272 | if (ret < 0) |
273 | mlog_errno(ret); | |
274 | ||
c11e9faf | 275 | out_commit: |
7f1a37e3 TY |
276 | ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle); |
277 | out: | |
278 | mlog_exit(ret); | |
279 | return ret; | |
280 | } | |
281 | ||
6cb129f5 AB |
282 | static int ocfs2_set_inode_size(handle_t *handle, |
283 | struct inode *inode, | |
284 | struct buffer_head *fe_bh, | |
285 | u64 new_i_size) | |
ccd979bd MF |
286 | { |
287 | int status; | |
288 | ||
289 | mlog_entry_void(); | |
290 | i_size_write(inode, new_i_size); | |
8110b073 | 291 | inode->i_blocks = ocfs2_inode_sector_count(inode); |
ccd979bd MF |
292 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; |
293 | ||
294 | status = ocfs2_mark_inode_dirty(handle, inode, fe_bh); | |
295 | if (status < 0) { | |
296 | mlog_errno(status); | |
297 | goto bail; | |
298 | } | |
299 | ||
300 | bail: | |
301 | mlog_exit(status); | |
302 | return status; | |
303 | } | |
304 | ||
305 | static int ocfs2_simple_size_update(struct inode *inode, | |
306 | struct buffer_head *di_bh, | |
307 | u64 new_i_size) | |
308 | { | |
309 | int ret; | |
310 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1fabe148 | 311 | handle_t *handle = NULL; |
ccd979bd | 312 | |
65eff9cc | 313 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
ccd979bd MF |
314 | if (handle == NULL) { |
315 | ret = -ENOMEM; | |
316 | mlog_errno(ret); | |
317 | goto out; | |
318 | } | |
319 | ||
320 | ret = ocfs2_set_inode_size(handle, inode, di_bh, | |
321 | new_i_size); | |
322 | if (ret < 0) | |
323 | mlog_errno(ret); | |
324 | ||
02dc1af4 | 325 | ocfs2_commit_trans(osb, handle); |
ccd979bd MF |
326 | out: |
327 | return ret; | |
328 | } | |
329 | ||
330 | static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb, | |
331 | struct inode *inode, | |
332 | struct buffer_head *fe_bh, | |
333 | u64 new_i_size) | |
334 | { | |
335 | int status; | |
1fabe148 | 336 | handle_t *handle; |
60b11392 | 337 | struct ocfs2_dinode *di; |
35edec1d | 338 | u64 cluster_bytes; |
ccd979bd MF |
339 | |
340 | mlog_entry_void(); | |
341 | ||
342 | /* TODO: This needs to actually orphan the inode in this | |
343 | * transaction. */ | |
344 | ||
65eff9cc | 345 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
ccd979bd MF |
346 | if (IS_ERR(handle)) { |
347 | status = PTR_ERR(handle); | |
348 | mlog_errno(status); | |
349 | goto out; | |
350 | } | |
351 | ||
60b11392 MF |
352 | status = ocfs2_journal_access(handle, inode, fe_bh, |
353 | OCFS2_JOURNAL_ACCESS_WRITE); | |
354 | if (status < 0) { | |
355 | mlog_errno(status); | |
356 | goto out_commit; | |
357 | } | |
358 | ||
359 | /* | |
360 | * Do this before setting i_size. | |
361 | */ | |
35edec1d MF |
362 | cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size); |
363 | status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size, | |
364 | cluster_bytes); | |
60b11392 MF |
365 | if (status) { |
366 | mlog_errno(status); | |
367 | goto out_commit; | |
368 | } | |
369 | ||
370 | i_size_write(inode, new_i_size); | |
60b11392 MF |
371 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; |
372 | ||
373 | di = (struct ocfs2_dinode *) fe_bh->b_data; | |
374 | di->i_size = cpu_to_le64(new_i_size); | |
375 | di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec); | |
376 | di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec); | |
377 | ||
378 | status = ocfs2_journal_dirty(handle, fe_bh); | |
ccd979bd MF |
379 | if (status < 0) |
380 | mlog_errno(status); | |
381 | ||
60b11392 | 382 | out_commit: |
02dc1af4 | 383 | ocfs2_commit_trans(osb, handle); |
ccd979bd | 384 | out: |
60b11392 | 385 | |
ccd979bd MF |
386 | mlog_exit(status); |
387 | return status; | |
388 | } | |
389 | ||
390 | static int ocfs2_truncate_file(struct inode *inode, | |
391 | struct buffer_head *di_bh, | |
392 | u64 new_i_size) | |
393 | { | |
394 | int status = 0; | |
395 | struct ocfs2_dinode *fe = NULL; | |
396 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
397 | struct ocfs2_truncate_context *tc = NULL; | |
398 | ||
b0697053 MF |
399 | mlog_entry("(inode = %llu, new_i_size = %llu\n", |
400 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
401 | (unsigned long long)new_i_size); | |
ccd979bd | 402 | |
ccd979bd MF |
403 | fe = (struct ocfs2_dinode *) di_bh->b_data; |
404 | if (!OCFS2_IS_VALID_DINODE(fe)) { | |
405 | OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe); | |
406 | status = -EIO; | |
407 | goto bail; | |
408 | } | |
409 | ||
410 | mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode), | |
b0697053 MF |
411 | "Inode %llu, inode i_size = %lld != di " |
412 | "i_size = %llu, i_flags = 0x%x\n", | |
413 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
ccd979bd | 414 | i_size_read(inode), |
b0697053 MF |
415 | (unsigned long long)le64_to_cpu(fe->i_size), |
416 | le32_to_cpu(fe->i_flags)); | |
ccd979bd MF |
417 | |
418 | if (new_i_size > le64_to_cpu(fe->i_size)) { | |
b0697053 MF |
419 | mlog(0, "asked to truncate file with size (%llu) to size (%llu)!\n", |
420 | (unsigned long long)le64_to_cpu(fe->i_size), | |
421 | (unsigned long long)new_i_size); | |
ccd979bd MF |
422 | status = -EINVAL; |
423 | mlog_errno(status); | |
424 | goto bail; | |
425 | } | |
426 | ||
b0697053 MF |
427 | mlog(0, "inode %llu, i_size = %llu, new_i_size = %llu\n", |
428 | (unsigned long long)le64_to_cpu(fe->i_blkno), | |
429 | (unsigned long long)le64_to_cpu(fe->i_size), | |
430 | (unsigned long long)new_i_size); | |
ccd979bd MF |
431 | |
432 | /* lets handle the simple truncate cases before doing any more | |
433 | * cluster locking. */ | |
434 | if (new_i_size == le64_to_cpu(fe->i_size)) | |
435 | goto bail; | |
436 | ||
2e89b2e4 MF |
437 | down_write(&OCFS2_I(inode)->ip_alloc_sem); |
438 | ||
c934a92d MF |
439 | /* |
440 | * The inode lock forced other nodes to sync and drop their | |
441 | * pages, which (correctly) happens even if we have a truncate | |
442 | * without allocation change - ocfs2 cluster sizes can be much | |
443 | * greater than page size, so we have to truncate them | |
444 | * anyway. | |
445 | */ | |
2e89b2e4 MF |
446 | unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1); |
447 | truncate_inode_pages(inode->i_mapping, new_i_size); | |
448 | ||
1afc32b9 MF |
449 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) { |
450 | status = ocfs2_truncate_inline(inode, di_bh, new_i_size, | |
b1967d0e | 451 | i_size_read(inode), 1); |
1afc32b9 MF |
452 | if (status) |
453 | mlog_errno(status); | |
454 | ||
c934a92d | 455 | goto bail_unlock_sem; |
1afc32b9 MF |
456 | } |
457 | ||
ccd979bd MF |
458 | /* alright, we're going to need to do a full blown alloc size |
459 | * change. Orphan the inode so that recovery can complete the | |
460 | * truncate if necessary. This does the task of marking | |
461 | * i_size. */ | |
462 | status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size); | |
463 | if (status < 0) { | |
464 | mlog_errno(status); | |
c934a92d | 465 | goto bail_unlock_sem; |
ccd979bd MF |
466 | } |
467 | ||
468 | status = ocfs2_prepare_truncate(osb, inode, di_bh, &tc); | |
469 | if (status < 0) { | |
470 | mlog_errno(status); | |
c934a92d | 471 | goto bail_unlock_sem; |
ccd979bd MF |
472 | } |
473 | ||
474 | status = ocfs2_commit_truncate(osb, inode, di_bh, tc); | |
475 | if (status < 0) { | |
476 | mlog_errno(status); | |
c934a92d | 477 | goto bail_unlock_sem; |
ccd979bd MF |
478 | } |
479 | ||
480 | /* TODO: orphan dir cleanup here. */ | |
c934a92d | 481 | bail_unlock_sem: |
2e89b2e4 MF |
482 | up_write(&OCFS2_I(inode)->ip_alloc_sem); |
483 | ||
ccd979bd MF |
484 | bail: |
485 | ||
486 | mlog_exit(status); | |
487 | return status; | |
488 | } | |
489 | ||
490 | /* | |
491 | * extend allocation only here. | |
492 | * we'll update all the disk stuff, and oip->alloc_size | |
493 | * | |
494 | * expect stuff to be locked, a transaction started and enough data / | |
495 | * metadata reservations in the contexts. | |
496 | * | |
497 | * Will return -EAGAIN, and a reason if a restart is needed. | |
498 | * If passed in, *reason will always be set, even in error. | |
499 | */ | |
500 | int ocfs2_do_extend_allocation(struct ocfs2_super *osb, | |
501 | struct inode *inode, | |
dcd0538f | 502 | u32 *logical_offset, |
ccd979bd | 503 | u32 clusters_to_add, |
2ae99a60 | 504 | int mark_unwritten, |
ccd979bd | 505 | struct buffer_head *fe_bh, |
1fabe148 | 506 | handle_t *handle, |
ccd979bd MF |
507 | struct ocfs2_alloc_context *data_ac, |
508 | struct ocfs2_alloc_context *meta_ac, | |
509 | enum ocfs2_alloc_restarted *reason_ret) | |
510 | { | |
511 | int status = 0; | |
512 | int free_extents; | |
513 | struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data; | |
514 | enum ocfs2_alloc_restarted reason = RESTART_NONE; | |
515 | u32 bit_off, num_bits; | |
516 | u64 block; | |
2ae99a60 | 517 | u8 flags = 0; |
ccd979bd MF |
518 | |
519 | BUG_ON(!clusters_to_add); | |
520 | ||
2ae99a60 MF |
521 | if (mark_unwritten) |
522 | flags = OCFS2_EXT_UNWRITTEN; | |
523 | ||
e7d4cb6b TM |
524 | free_extents = ocfs2_num_free_extents(osb, inode, fe_bh, |
525 | OCFS2_DINODE_EXTENT); | |
ccd979bd MF |
526 | if (free_extents < 0) { |
527 | status = free_extents; | |
528 | mlog_errno(status); | |
529 | goto leave; | |
530 | } | |
531 | ||
532 | /* there are two cases which could cause us to EAGAIN in the | |
533 | * we-need-more-metadata case: | |
534 | * 1) we haven't reserved *any* | |
535 | * 2) we are so fragmented, we've needed to add metadata too | |
536 | * many times. */ | |
537 | if (!free_extents && !meta_ac) { | |
538 | mlog(0, "we haven't reserved any metadata!\n"); | |
539 | status = -EAGAIN; | |
540 | reason = RESTART_META; | |
541 | goto leave; | |
542 | } else if ((!free_extents) | |
543 | && (ocfs2_alloc_context_bits_left(meta_ac) | |
811f933d | 544 | < ocfs2_extend_meta_needed(&fe->id2.i_list))) { |
ccd979bd MF |
545 | mlog(0, "filesystem is really fragmented...\n"); |
546 | status = -EAGAIN; | |
547 | reason = RESTART_META; | |
548 | goto leave; | |
549 | } | |
550 | ||
415cb800 MF |
551 | status = __ocfs2_claim_clusters(osb, handle, data_ac, 1, |
552 | clusters_to_add, &bit_off, &num_bits); | |
ccd979bd MF |
553 | if (status < 0) { |
554 | if (status != -ENOSPC) | |
555 | mlog_errno(status); | |
556 | goto leave; | |
557 | } | |
558 | ||
559 | BUG_ON(num_bits > clusters_to_add); | |
560 | ||
561 | /* reserve our write early -- insert_extent may update the inode */ | |
562 | status = ocfs2_journal_access(handle, inode, fe_bh, | |
563 | OCFS2_JOURNAL_ACCESS_WRITE); | |
564 | if (status < 0) { | |
565 | mlog_errno(status); | |
566 | goto leave; | |
567 | } | |
568 | ||
569 | block = ocfs2_clusters_to_blocks(osb->sb, bit_off); | |
b0697053 MF |
570 | mlog(0, "Allocating %u clusters at block %u for inode %llu\n", |
571 | num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
dcd0538f MF |
572 | status = ocfs2_insert_extent(osb, handle, inode, fe_bh, |
573 | *logical_offset, block, num_bits, | |
e7d4cb6b | 574 | flags, meta_ac, OCFS2_DINODE_EXTENT); |
ccd979bd MF |
575 | if (status < 0) { |
576 | mlog_errno(status); | |
577 | goto leave; | |
578 | } | |
579 | ||
ccd979bd MF |
580 | status = ocfs2_journal_dirty(handle, fe_bh); |
581 | if (status < 0) { | |
582 | mlog_errno(status); | |
583 | goto leave; | |
584 | } | |
585 | ||
586 | clusters_to_add -= num_bits; | |
dcd0538f | 587 | *logical_offset += num_bits; |
ccd979bd MF |
588 | |
589 | if (clusters_to_add) { | |
590 | mlog(0, "need to alloc once more, clusters = %u, wanted = " | |
591 | "%u\n", fe->i_clusters, clusters_to_add); | |
592 | status = -EAGAIN; | |
593 | reason = RESTART_TRANS; | |
594 | } | |
595 | ||
596 | leave: | |
597 | mlog_exit(status); | |
598 | if (reason_ret) | |
599 | *reason_ret = reason; | |
600 | return status; | |
601 | } | |
602 | ||
2ae99a60 MF |
603 | static int __ocfs2_extend_allocation(struct inode *inode, u32 logical_start, |
604 | u32 clusters_to_add, int mark_unwritten) | |
ccd979bd MF |
605 | { |
606 | int status = 0; | |
607 | int restart_func = 0; | |
abf8b156 | 608 | int credits; |
2ae99a60 | 609 | u32 prev_clusters; |
ccd979bd MF |
610 | struct buffer_head *bh = NULL; |
611 | struct ocfs2_dinode *fe = NULL; | |
1fabe148 | 612 | handle_t *handle = NULL; |
ccd979bd MF |
613 | struct ocfs2_alloc_context *data_ac = NULL; |
614 | struct ocfs2_alloc_context *meta_ac = NULL; | |
615 | enum ocfs2_alloc_restarted why; | |
616 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
617 | ||
618 | mlog_entry("(clusters_to_add = %u)\n", clusters_to_add); | |
619 | ||
dcd0538f MF |
620 | /* |
621 | * This function only exists for file systems which don't | |
622 | * support holes. | |
623 | */ | |
2ae99a60 | 624 | BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb)); |
dcd0538f | 625 | |
ccd979bd MF |
626 | status = ocfs2_read_block(osb, OCFS2_I(inode)->ip_blkno, &bh, |
627 | OCFS2_BH_CACHED, inode); | |
628 | if (status < 0) { | |
629 | mlog_errno(status); | |
630 | goto leave; | |
631 | } | |
632 | ||
633 | fe = (struct ocfs2_dinode *) bh->b_data; | |
634 | if (!OCFS2_IS_VALID_DINODE(fe)) { | |
635 | OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe); | |
636 | status = -EIO; | |
637 | goto leave; | |
638 | } | |
639 | ||
640 | restart_all: | |
641 | BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters); | |
642 | ||
e7d4cb6b TM |
643 | mlog(0, "extend inode %llu, i_size = %lld, di->i_clusters = %u, " |
644 | "clusters_to_add = %u\n", | |
645 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
646 | (long long)i_size_read(inode), le32_to_cpu(fe->i_clusters), | |
647 | clusters_to_add); | |
648 | status = ocfs2_lock_allocators(inode, bh, &fe->id2.i_list, | |
649 | clusters_to_add, 0, &data_ac, | |
9517bac6 MF |
650 | &meta_ac); |
651 | if (status) { | |
652 | mlog_errno(status); | |
653 | goto leave; | |
654 | } | |
655 | ||
811f933d TM |
656 | credits = ocfs2_calc_extend_credits(osb->sb, &fe->id2.i_list, |
657 | clusters_to_add); | |
65eff9cc | 658 | handle = ocfs2_start_trans(osb, credits); |
ccd979bd MF |
659 | if (IS_ERR(handle)) { |
660 | status = PTR_ERR(handle); | |
661 | handle = NULL; | |
662 | mlog_errno(status); | |
663 | goto leave; | |
664 | } | |
665 | ||
666 | restarted_transaction: | |
667 | /* reserve a write to the file entry early on - that we if we | |
668 | * run out of credits in the allocation path, we can still | |
669 | * update i_size. */ | |
670 | status = ocfs2_journal_access(handle, inode, bh, | |
671 | OCFS2_JOURNAL_ACCESS_WRITE); | |
672 | if (status < 0) { | |
673 | mlog_errno(status); | |
674 | goto leave; | |
675 | } | |
676 | ||
677 | prev_clusters = OCFS2_I(inode)->ip_clusters; | |
678 | ||
679 | status = ocfs2_do_extend_allocation(osb, | |
680 | inode, | |
dcd0538f | 681 | &logical_start, |
ccd979bd | 682 | clusters_to_add, |
2ae99a60 | 683 | mark_unwritten, |
ccd979bd MF |
684 | bh, |
685 | handle, | |
686 | data_ac, | |
687 | meta_ac, | |
688 | &why); | |
689 | if ((status < 0) && (status != -EAGAIN)) { | |
690 | if (status != -ENOSPC) | |
691 | mlog_errno(status); | |
692 | goto leave; | |
693 | } | |
694 | ||
695 | status = ocfs2_journal_dirty(handle, bh); | |
696 | if (status < 0) { | |
697 | mlog_errno(status); | |
698 | goto leave; | |
699 | } | |
700 | ||
701 | spin_lock(&OCFS2_I(inode)->ip_lock); | |
702 | clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters); | |
703 | spin_unlock(&OCFS2_I(inode)->ip_lock); | |
704 | ||
705 | if (why != RESTART_NONE && clusters_to_add) { | |
706 | if (why == RESTART_META) { | |
707 | mlog(0, "restarting function.\n"); | |
708 | restart_func = 1; | |
709 | } else { | |
710 | BUG_ON(why != RESTART_TRANS); | |
711 | ||
712 | mlog(0, "restarting transaction.\n"); | |
713 | /* TODO: This can be more intelligent. */ | |
714 | credits = ocfs2_calc_extend_credits(osb->sb, | |
811f933d | 715 | &fe->id2.i_list, |
ccd979bd | 716 | clusters_to_add); |
1fabe148 | 717 | status = ocfs2_extend_trans(handle, credits); |
ccd979bd MF |
718 | if (status < 0) { |
719 | /* handle still has to be committed at | |
720 | * this point. */ | |
721 | status = -ENOMEM; | |
722 | mlog_errno(status); | |
723 | goto leave; | |
724 | } | |
725 | goto restarted_transaction; | |
726 | } | |
727 | } | |
728 | ||
b0697053 | 729 | mlog(0, "fe: i_clusters = %u, i_size=%llu\n", |
1ca1a111 MF |
730 | le32_to_cpu(fe->i_clusters), |
731 | (unsigned long long)le64_to_cpu(fe->i_size)); | |
ccd979bd | 732 | mlog(0, "inode: ip_clusters=%u, i_size=%lld\n", |
634bf74d | 733 | OCFS2_I(inode)->ip_clusters, (long long)i_size_read(inode)); |
ccd979bd MF |
734 | |
735 | leave: | |
ccd979bd | 736 | if (handle) { |
02dc1af4 | 737 | ocfs2_commit_trans(osb, handle); |
ccd979bd MF |
738 | handle = NULL; |
739 | } | |
740 | if (data_ac) { | |
741 | ocfs2_free_alloc_context(data_ac); | |
742 | data_ac = NULL; | |
743 | } | |
744 | if (meta_ac) { | |
745 | ocfs2_free_alloc_context(meta_ac); | |
746 | meta_ac = NULL; | |
747 | } | |
748 | if ((!status) && restart_func) { | |
749 | restart_func = 0; | |
750 | goto restart_all; | |
751 | } | |
752 | if (bh) { | |
753 | brelse(bh); | |
754 | bh = NULL; | |
755 | } | |
756 | ||
757 | mlog_exit(status); | |
758 | return status; | |
759 | } | |
760 | ||
761 | /* Some parts of this taken from generic_cont_expand, which turned out | |
762 | * to be too fragile to do exactly what we need without us having to | |
53013cba MF |
763 | * worry about recursive locking in ->prepare_write() and |
764 | * ->commit_write(). */ | |
ccd979bd MF |
765 | static int ocfs2_write_zero_page(struct inode *inode, |
766 | u64 size) | |
767 | { | |
768 | struct address_space *mapping = inode->i_mapping; | |
769 | struct page *page; | |
770 | unsigned long index; | |
771 | unsigned int offset; | |
1fabe148 | 772 | handle_t *handle = NULL; |
ccd979bd MF |
773 | int ret; |
774 | ||
775 | offset = (size & (PAGE_CACHE_SIZE-1)); /* Within page */ | |
776 | /* ugh. in prepare/commit_write, if from==to==start of block, we | |
777 | ** skip the prepare. make sure we never send an offset for the start | |
778 | ** of a block | |
779 | */ | |
780 | if ((offset & (inode->i_sb->s_blocksize - 1)) == 0) { | |
781 | offset++; | |
782 | } | |
783 | index = size >> PAGE_CACHE_SHIFT; | |
784 | ||
785 | page = grab_cache_page(mapping, index); | |
786 | if (!page) { | |
787 | ret = -ENOMEM; | |
788 | mlog_errno(ret); | |
789 | goto out; | |
790 | } | |
791 | ||
53013cba | 792 | ret = ocfs2_prepare_write_nolock(inode, page, offset, offset); |
ccd979bd MF |
793 | if (ret < 0) { |
794 | mlog_errno(ret); | |
795 | goto out_unlock; | |
796 | } | |
797 | ||
798 | if (ocfs2_should_order_data(inode)) { | |
799 | handle = ocfs2_start_walk_page_trans(inode, page, offset, | |
800 | offset); | |
801 | if (IS_ERR(handle)) { | |
802 | ret = PTR_ERR(handle); | |
803 | handle = NULL; | |
804 | goto out_unlock; | |
805 | } | |
806 | } | |
807 | ||
808 | /* must not update i_size! */ | |
809 | ret = block_commit_write(page, offset, offset); | |
810 | if (ret < 0) | |
811 | mlog_errno(ret); | |
812 | else | |
813 | ret = 0; | |
814 | ||
815 | if (handle) | |
02dc1af4 | 816 | ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle); |
ccd979bd MF |
817 | out_unlock: |
818 | unlock_page(page); | |
819 | page_cache_release(page); | |
820 | out: | |
821 | return ret; | |
822 | } | |
823 | ||
824 | static int ocfs2_zero_extend(struct inode *inode, | |
825 | u64 zero_to_size) | |
826 | { | |
827 | int ret = 0; | |
828 | u64 start_off; | |
829 | struct super_block *sb = inode->i_sb; | |
830 | ||
831 | start_off = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode)); | |
832 | while (start_off < zero_to_size) { | |
833 | ret = ocfs2_write_zero_page(inode, start_off); | |
834 | if (ret < 0) { | |
835 | mlog_errno(ret); | |
836 | goto out; | |
837 | } | |
838 | ||
839 | start_off += sb->s_blocksize; | |
e2057c5a MF |
840 | |
841 | /* | |
842 | * Very large extends have the potential to lock up | |
843 | * the cpu for extended periods of time. | |
844 | */ | |
845 | cond_resched(); | |
ccd979bd MF |
846 | } |
847 | ||
848 | out: | |
849 | return ret; | |
850 | } | |
851 | ||
65ed39d6 MF |
852 | int ocfs2_extend_no_holes(struct inode *inode, u64 new_i_size, u64 zero_to) |
853 | { | |
854 | int ret; | |
855 | u32 clusters_to_add; | |
856 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
857 | ||
858 | clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size); | |
859 | if (clusters_to_add < oi->ip_clusters) | |
860 | clusters_to_add = 0; | |
861 | else | |
862 | clusters_to_add -= oi->ip_clusters; | |
863 | ||
864 | if (clusters_to_add) { | |
865 | ret = __ocfs2_extend_allocation(inode, oi->ip_clusters, | |
866 | clusters_to_add, 0); | |
867 | if (ret) { | |
868 | mlog_errno(ret); | |
869 | goto out; | |
870 | } | |
871 | } | |
872 | ||
873 | /* | |
874 | * Call this even if we don't add any clusters to the tree. We | |
875 | * still need to zero the area between the old i_size and the | |
876 | * new i_size. | |
877 | */ | |
878 | ret = ocfs2_zero_extend(inode, zero_to); | |
879 | if (ret < 0) | |
880 | mlog_errno(ret); | |
881 | ||
882 | out: | |
883 | return ret; | |
884 | } | |
885 | ||
ccd979bd MF |
886 | static int ocfs2_extend_file(struct inode *inode, |
887 | struct buffer_head *di_bh, | |
65ed39d6 | 888 | u64 new_i_size) |
ccd979bd | 889 | { |
c934a92d | 890 | int ret = 0; |
1afc32b9 | 891 | struct ocfs2_inode_info *oi = OCFS2_I(inode); |
ccd979bd | 892 | |
65ed39d6 | 893 | BUG_ON(!di_bh); |
53013cba | 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 | ||
1afc32b9 MF |
903 | /* |
904 | * Fall through for converting inline data, even if the fs | |
905 | * supports sparse files. | |
906 | * | |
907 | * The check for inline data here is legal - nobody can add | |
908 | * the feature since we have i_mutex. We must check it again | |
909 | * after acquiring ip_alloc_sem though, as paths like mmap | |
910 | * might have raced us to converting the inode to extents. | |
911 | */ | |
912 | if (!(oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) | |
913 | && ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))) | |
3a0782d0 | 914 | goto out_update_size; |
ccd979bd | 915 | |
0effef77 | 916 | /* |
65ed39d6 MF |
917 | * The alloc sem blocks people in read/write from reading our |
918 | * allocation until we're done changing it. We depend on | |
919 | * i_mutex to block other extend/truncate calls while we're | |
920 | * here. | |
0effef77 | 921 | */ |
1afc32b9 MF |
922 | down_write(&oi->ip_alloc_sem); |
923 | ||
924 | if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) { | |
925 | /* | |
926 | * We can optimize small extends by keeping the inodes | |
927 | * inline data. | |
928 | */ | |
929 | if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) { | |
930 | up_write(&oi->ip_alloc_sem); | |
931 | goto out_update_size; | |
932 | } | |
933 | ||
934 | ret = ocfs2_convert_inline_data_to_extents(inode, di_bh); | |
935 | if (ret) { | |
936 | up_write(&oi->ip_alloc_sem); | |
937 | ||
938 | mlog_errno(ret); | |
c934a92d | 939 | goto out; |
1afc32b9 MF |
940 | } |
941 | } | |
942 | ||
943 | if (!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))) | |
944 | ret = ocfs2_extend_no_holes(inode, new_i_size, new_i_size); | |
945 | ||
946 | up_write(&oi->ip_alloc_sem); | |
65ed39d6 | 947 | |
0effef77 MF |
948 | if (ret < 0) { |
949 | mlog_errno(ret); | |
c934a92d | 950 | goto out; |
53013cba MF |
951 | } |
952 | ||
3a0782d0 | 953 | out_update_size: |
65ed39d6 MF |
954 | ret = ocfs2_simple_size_update(inode, di_bh, new_i_size); |
955 | if (ret < 0) | |
956 | mlog_errno(ret); | |
ccd979bd MF |
957 | |
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 | ||
bc535809 SM |
974 | /* ensuring we don't even attempt to truncate a symlink */ |
975 | if (S_ISLNK(inode->i_mode)) | |
976 | attr->ia_valid &= ~ATTR_SIZE; | |
977 | ||
ccd979bd MF |
978 | if (attr->ia_valid & ATTR_MODE) |
979 | mlog(0, "mode change: %d\n", attr->ia_mode); | |
980 | if (attr->ia_valid & ATTR_UID) | |
981 | mlog(0, "uid change: %d\n", attr->ia_uid); | |
982 | if (attr->ia_valid & ATTR_GID) | |
983 | mlog(0, "gid change: %d\n", attr->ia_gid); | |
984 | if (attr->ia_valid & ATTR_SIZE) | |
985 | mlog(0, "size change...\n"); | |
986 | if (attr->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME)) | |
987 | mlog(0, "time change...\n"); | |
988 | ||
989 | #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \ | |
990 | | ATTR_GID | ATTR_UID | ATTR_MODE) | |
991 | if (!(attr->ia_valid & OCFS2_VALID_ATTRS)) { | |
992 | mlog(0, "can't handle attrs: 0x%x\n", attr->ia_valid); | |
993 | return 0; | |
994 | } | |
995 | ||
996 | status = inode_change_ok(inode, attr); | |
997 | if (status) | |
998 | return status; | |
999 | ||
1000 | size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE; | |
1001 | if (size_change) { | |
1002 | status = ocfs2_rw_lock(inode, 1); | |
1003 | if (status < 0) { | |
1004 | mlog_errno(status); | |
1005 | goto bail; | |
1006 | } | |
1007 | } | |
1008 | ||
e63aecb6 | 1009 | status = ocfs2_inode_lock(inode, &bh, 1); |
ccd979bd MF |
1010 | if (status < 0) { |
1011 | if (status != -ENOENT) | |
1012 | mlog_errno(status); | |
1013 | goto bail_unlock_rw; | |
1014 | } | |
1015 | ||
1016 | if (size_change && attr->ia_size != i_size_read(inode)) { | |
ce76fd30 MF |
1017 | if (attr->ia_size > sb->s_maxbytes) { |
1018 | status = -EFBIG; | |
1019 | goto bail_unlock; | |
1020 | } | |
1021 | ||
ccd979bd MF |
1022 | if (i_size_read(inode) > attr->ia_size) |
1023 | status = ocfs2_truncate_file(inode, bh, attr->ia_size); | |
1024 | else | |
65ed39d6 | 1025 | status = ocfs2_extend_file(inode, bh, attr->ia_size); |
ccd979bd MF |
1026 | if (status < 0) { |
1027 | if (status != -ENOSPC) | |
1028 | mlog_errno(status); | |
1029 | status = -ENOSPC; | |
1030 | goto bail_unlock; | |
1031 | } | |
1032 | } | |
1033 | ||
65eff9cc | 1034 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
ccd979bd MF |
1035 | if (IS_ERR(handle)) { |
1036 | status = PTR_ERR(handle); | |
1037 | mlog_errno(status); | |
1038 | goto bail_unlock; | |
1039 | } | |
1040 | ||
7307de80 MF |
1041 | /* |
1042 | * This will intentionally not wind up calling vmtruncate(), | |
1043 | * since all the work for a size change has been done above. | |
1044 | * Otherwise, we could get into problems with truncate as | |
1045 | * ip_alloc_sem is used there to protect against i_size | |
1046 | * changes. | |
1047 | */ | |
ccd979bd MF |
1048 | status = inode_setattr(inode, attr); |
1049 | if (status < 0) { | |
1050 | mlog_errno(status); | |
1051 | goto bail_commit; | |
1052 | } | |
1053 | ||
1054 | status = ocfs2_mark_inode_dirty(handle, inode, bh); | |
1055 | if (status < 0) | |
1056 | mlog_errno(status); | |
1057 | ||
1058 | bail_commit: | |
02dc1af4 | 1059 | ocfs2_commit_trans(osb, handle); |
ccd979bd | 1060 | bail_unlock: |
e63aecb6 | 1061 | ocfs2_inode_unlock(inode, 1); |
ccd979bd MF |
1062 | bail_unlock_rw: |
1063 | if (size_change) | |
1064 | ocfs2_rw_unlock(inode, 1); | |
1065 | bail: | |
1066 | if (bh) | |
1067 | brelse(bh); | |
1068 | ||
1069 | mlog_exit(status); | |
1070 | return status; | |
1071 | } | |
1072 | ||
1073 | int ocfs2_getattr(struct vfsmount *mnt, | |
1074 | struct dentry *dentry, | |
1075 | struct kstat *stat) | |
1076 | { | |
1077 | struct inode *inode = dentry->d_inode; | |
1078 | struct super_block *sb = dentry->d_inode->i_sb; | |
1079 | struct ocfs2_super *osb = sb->s_fs_info; | |
1080 | int err; | |
1081 | ||
1082 | mlog_entry_void(); | |
1083 | ||
1084 | err = ocfs2_inode_revalidate(dentry); | |
1085 | if (err) { | |
1086 | if (err != -ENOENT) | |
1087 | mlog_errno(err); | |
1088 | goto bail; | |
1089 | } | |
1090 | ||
1091 | generic_fillattr(inode, stat); | |
1092 | ||
1093 | /* We set the blksize from the cluster size for performance */ | |
1094 | stat->blksize = osb->s_clustersize; | |
1095 | ||
1096 | bail: | |
1097 | mlog_exit(err); | |
1098 | ||
1099 | return err; | |
1100 | } | |
1101 | ||
e6305c43 | 1102 | int ocfs2_permission(struct inode *inode, int mask) |
d38eb8db TY |
1103 | { |
1104 | int ret; | |
1105 | ||
1106 | mlog_entry_void(); | |
1107 | ||
e63aecb6 | 1108 | ret = ocfs2_inode_lock(inode, NULL, 0); |
d38eb8db | 1109 | if (ret) { |
a9f5f707 MF |
1110 | if (ret != -ENOENT) |
1111 | mlog_errno(ret); | |
d38eb8db TY |
1112 | goto out; |
1113 | } | |
1114 | ||
1115 | ret = generic_permission(inode, mask, NULL); | |
d38eb8db | 1116 | |
e63aecb6 | 1117 | ocfs2_inode_unlock(inode, 0); |
d38eb8db TY |
1118 | out: |
1119 | mlog_exit(ret); | |
1120 | return ret; | |
1121 | } | |
1122 | ||
b2580103 MF |
1123 | static int __ocfs2_write_remove_suid(struct inode *inode, |
1124 | struct buffer_head *bh) | |
ccd979bd MF |
1125 | { |
1126 | int ret; | |
1fabe148 | 1127 | handle_t *handle; |
ccd979bd MF |
1128 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
1129 | struct ocfs2_dinode *di; | |
1130 | ||
b0697053 | 1131 | mlog_entry("(Inode %llu, mode 0%o)\n", |
b2580103 | 1132 | (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_mode); |
ccd979bd | 1133 | |
65eff9cc | 1134 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
ccd979bd MF |
1135 | if (handle == NULL) { |
1136 | ret = -ENOMEM; | |
1137 | mlog_errno(ret); | |
1138 | goto out; | |
1139 | } | |
1140 | ||
ccd979bd MF |
1141 | ret = ocfs2_journal_access(handle, inode, bh, |
1142 | OCFS2_JOURNAL_ACCESS_WRITE); | |
1143 | if (ret < 0) { | |
1144 | mlog_errno(ret); | |
b2580103 | 1145 | goto out_trans; |
ccd979bd MF |
1146 | } |
1147 | ||
1148 | inode->i_mode &= ~S_ISUID; | |
1149 | if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP)) | |
1150 | inode->i_mode &= ~S_ISGID; | |
1151 | ||
1152 | di = (struct ocfs2_dinode *) bh->b_data; | |
1153 | di->i_mode = cpu_to_le16(inode->i_mode); | |
1154 | ||
1155 | ret = ocfs2_journal_dirty(handle, bh); | |
1156 | if (ret < 0) | |
1157 | mlog_errno(ret); | |
b2580103 | 1158 | |
ccd979bd | 1159 | out_trans: |
02dc1af4 | 1160 | ocfs2_commit_trans(osb, handle); |
ccd979bd MF |
1161 | out: |
1162 | mlog_exit(ret); | |
1163 | return ret; | |
1164 | } | |
1165 | ||
9517bac6 MF |
1166 | /* |
1167 | * Will look for holes and unwritten extents in the range starting at | |
1168 | * pos for count bytes (inclusive). | |
1169 | */ | |
1170 | static int ocfs2_check_range_for_holes(struct inode *inode, loff_t pos, | |
1171 | size_t count) | |
1172 | { | |
1173 | int ret = 0; | |
49cb8d2d | 1174 | unsigned int extent_flags; |
9517bac6 MF |
1175 | u32 cpos, clusters, extent_len, phys_cpos; |
1176 | struct super_block *sb = inode->i_sb; | |
1177 | ||
1178 | cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits; | |
1179 | clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos; | |
1180 | ||
1181 | while (clusters) { | |
49cb8d2d MF |
1182 | ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len, |
1183 | &extent_flags); | |
9517bac6 MF |
1184 | if (ret < 0) { |
1185 | mlog_errno(ret); | |
1186 | goto out; | |
1187 | } | |
1188 | ||
49cb8d2d | 1189 | if (phys_cpos == 0 || (extent_flags & OCFS2_EXT_UNWRITTEN)) { |
9517bac6 MF |
1190 | ret = 1; |
1191 | break; | |
1192 | } | |
1193 | ||
1194 | if (extent_len > clusters) | |
1195 | extent_len = clusters; | |
1196 | ||
1197 | clusters -= extent_len; | |
1198 | cpos += extent_len; | |
1199 | } | |
1200 | out: | |
1201 | return ret; | |
1202 | } | |
1203 | ||
b2580103 MF |
1204 | static int ocfs2_write_remove_suid(struct inode *inode) |
1205 | { | |
1206 | int ret; | |
1207 | struct buffer_head *bh = NULL; | |
1208 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
1209 | ||
1210 | ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), | |
1211 | oi->ip_blkno, &bh, OCFS2_BH_CACHED, inode); | |
1212 | if (ret < 0) { | |
1213 | mlog_errno(ret); | |
1214 | goto out; | |
1215 | } | |
1216 | ||
1217 | ret = __ocfs2_write_remove_suid(inode, bh); | |
1218 | out: | |
1219 | brelse(bh); | |
1220 | return ret; | |
1221 | } | |
1222 | ||
2ae99a60 MF |
1223 | /* |
1224 | * Allocate enough extents to cover the region starting at byte offset | |
1225 | * start for len bytes. Existing extents are skipped, any extents | |
1226 | * added are marked as "unwritten". | |
1227 | */ | |
1228 | static int ocfs2_allocate_unwritten_extents(struct inode *inode, | |
1229 | u64 start, u64 len) | |
1230 | { | |
1231 | int ret; | |
1232 | u32 cpos, phys_cpos, clusters, alloc_size; | |
1afc32b9 MF |
1233 | u64 end = start + len; |
1234 | struct buffer_head *di_bh = NULL; | |
1235 | ||
1236 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) { | |
1237 | ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), | |
1238 | OCFS2_I(inode)->ip_blkno, &di_bh, | |
1239 | OCFS2_BH_CACHED, inode); | |
1240 | if (ret) { | |
1241 | mlog_errno(ret); | |
1242 | goto out; | |
1243 | } | |
1244 | ||
1245 | /* | |
1246 | * Nothing to do if the requested reservation range | |
1247 | * fits within the inode. | |
1248 | */ | |
1249 | if (ocfs2_size_fits_inline_data(di_bh, end)) | |
1250 | goto out; | |
1251 | ||
1252 | ret = ocfs2_convert_inline_data_to_extents(inode, di_bh); | |
1253 | if (ret) { | |
1254 | mlog_errno(ret); | |
1255 | goto out; | |
1256 | } | |
1257 | } | |
2ae99a60 MF |
1258 | |
1259 | /* | |
1260 | * We consider both start and len to be inclusive. | |
1261 | */ | |
1262 | cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits; | |
1263 | clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len); | |
1264 | clusters -= cpos; | |
1265 | ||
1266 | while (clusters) { | |
1267 | ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, | |
1268 | &alloc_size, NULL); | |
1269 | if (ret) { | |
1270 | mlog_errno(ret); | |
1271 | goto out; | |
1272 | } | |
1273 | ||
1274 | /* | |
1275 | * Hole or existing extent len can be arbitrary, so | |
1276 | * cap it to our own allocation request. | |
1277 | */ | |
1278 | if (alloc_size > clusters) | |
1279 | alloc_size = clusters; | |
1280 | ||
1281 | if (phys_cpos) { | |
1282 | /* | |
1283 | * We already have an allocation at this | |
1284 | * region so we can safely skip it. | |
1285 | */ | |
1286 | goto next; | |
1287 | } | |
1288 | ||
1289 | ret = __ocfs2_extend_allocation(inode, cpos, alloc_size, 1); | |
1290 | if (ret) { | |
1291 | if (ret != -ENOSPC) | |
1292 | mlog_errno(ret); | |
1293 | goto out; | |
1294 | } | |
1295 | ||
1296 | next: | |
1297 | cpos += alloc_size; | |
1298 | clusters -= alloc_size; | |
1299 | } | |
1300 | ||
1301 | ret = 0; | |
1302 | out: | |
1afc32b9 MF |
1303 | |
1304 | brelse(di_bh); | |
2ae99a60 MF |
1305 | return ret; |
1306 | } | |
1307 | ||
063c4561 MF |
1308 | static int __ocfs2_remove_inode_range(struct inode *inode, |
1309 | struct buffer_head *di_bh, | |
1310 | u32 cpos, u32 phys_cpos, u32 len, | |
1311 | struct ocfs2_cached_dealloc_ctxt *dealloc) | |
1312 | { | |
1313 | int ret; | |
1314 | u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos); | |
1315 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1316 | struct inode *tl_inode = osb->osb_tl_inode; | |
1317 | handle_t *handle; | |
1318 | struct ocfs2_alloc_context *meta_ac = NULL; | |
1319 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; | |
1320 | ||
e7d4cb6b TM |
1321 | ret = ocfs2_lock_allocators(inode, di_bh, &di->id2.i_list, |
1322 | 0, 1, NULL, &meta_ac); | |
063c4561 MF |
1323 | if (ret) { |
1324 | mlog_errno(ret); | |
1325 | return ret; | |
1326 | } | |
1327 | ||
1328 | mutex_lock(&tl_inode->i_mutex); | |
1329 | ||
1330 | if (ocfs2_truncate_log_needs_flush(osb)) { | |
1331 | ret = __ocfs2_flush_truncate_log(osb); | |
1332 | if (ret < 0) { | |
1333 | mlog_errno(ret); | |
1334 | goto out; | |
1335 | } | |
1336 | } | |
1337 | ||
1338 | handle = ocfs2_start_trans(osb, OCFS2_REMOVE_EXTENT_CREDITS); | |
1339 | if (handle == NULL) { | |
1340 | ret = -ENOMEM; | |
1341 | mlog_errno(ret); | |
1342 | goto out; | |
1343 | } | |
1344 | ||
1345 | ret = ocfs2_journal_access(handle, inode, di_bh, | |
1346 | OCFS2_JOURNAL_ACCESS_WRITE); | |
1347 | if (ret) { | |
1348 | mlog_errno(ret); | |
1349 | goto out; | |
1350 | } | |
1351 | ||
1352 | ret = ocfs2_remove_extent(inode, di_bh, cpos, len, handle, meta_ac, | |
e7d4cb6b | 1353 | dealloc, OCFS2_DINODE_EXTENT); |
063c4561 MF |
1354 | if (ret) { |
1355 | mlog_errno(ret); | |
1356 | goto out_commit; | |
1357 | } | |
1358 | ||
1359 | OCFS2_I(inode)->ip_clusters -= len; | |
1360 | di->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters); | |
1361 | ||
1362 | ret = ocfs2_journal_dirty(handle, di_bh); | |
1363 | if (ret) { | |
1364 | mlog_errno(ret); | |
1365 | goto out_commit; | |
1366 | } | |
1367 | ||
1368 | ret = ocfs2_truncate_log_append(osb, handle, phys_blkno, len); | |
1369 | if (ret) | |
1370 | mlog_errno(ret); | |
1371 | ||
1372 | out_commit: | |
1373 | ocfs2_commit_trans(osb, handle); | |
1374 | out: | |
1375 | mutex_unlock(&tl_inode->i_mutex); | |
1376 | ||
1377 | if (meta_ac) | |
1378 | ocfs2_free_alloc_context(meta_ac); | |
1379 | ||
1380 | return ret; | |
1381 | } | |
1382 | ||
1383 | /* | |
1384 | * Truncate a byte range, avoiding pages within partial clusters. This | |
1385 | * preserves those pages for the zeroing code to write to. | |
1386 | */ | |
1387 | static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start, | |
1388 | u64 byte_len) | |
1389 | { | |
1390 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1391 | loff_t start, end; | |
1392 | struct address_space *mapping = inode->i_mapping; | |
1393 | ||
1394 | start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start); | |
1395 | end = byte_start + byte_len; | |
1396 | end = end & ~(osb->s_clustersize - 1); | |
1397 | ||
1398 | if (start < end) { | |
1399 | unmap_mapping_range(mapping, start, end - start, 0); | |
1400 | truncate_inode_pages_range(mapping, start, end - 1); | |
1401 | } | |
1402 | } | |
1403 | ||
1404 | static int ocfs2_zero_partial_clusters(struct inode *inode, | |
1405 | u64 start, u64 len) | |
1406 | { | |
1407 | int ret = 0; | |
1408 | u64 tmpend, end = start + len; | |
1409 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1410 | unsigned int csize = osb->s_clustersize; | |
1411 | handle_t *handle; | |
1412 | ||
1413 | /* | |
1414 | * The "start" and "end" values are NOT necessarily part of | |
1415 | * the range whose allocation is being deleted. Rather, this | |
1416 | * is what the user passed in with the request. We must zero | |
1417 | * partial clusters here. There's no need to worry about | |
1418 | * physical allocation - the zeroing code knows to skip holes. | |
1419 | */ | |
1420 | mlog(0, "byte start: %llu, end: %llu\n", | |
1421 | (unsigned long long)start, (unsigned long long)end); | |
1422 | ||
1423 | /* | |
1424 | * If both edges are on a cluster boundary then there's no | |
1425 | * zeroing required as the region is part of the allocation to | |
1426 | * be truncated. | |
1427 | */ | |
1428 | if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0) | |
1429 | goto out; | |
1430 | ||
1431 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); | |
1432 | if (handle == NULL) { | |
1433 | ret = -ENOMEM; | |
1434 | mlog_errno(ret); | |
1435 | goto out; | |
1436 | } | |
1437 | ||
1438 | /* | |
1439 | * We want to get the byte offset of the end of the 1st cluster. | |
1440 | */ | |
1441 | tmpend = (u64)osb->s_clustersize + (start & ~(osb->s_clustersize - 1)); | |
1442 | if (tmpend > end) | |
1443 | tmpend = end; | |
1444 | ||
1445 | mlog(0, "1st range: start: %llu, tmpend: %llu\n", | |
1446 | (unsigned long long)start, (unsigned long long)tmpend); | |
1447 | ||
1448 | ret = ocfs2_zero_range_for_truncate(inode, handle, start, tmpend); | |
1449 | if (ret) | |
1450 | mlog_errno(ret); | |
1451 | ||
1452 | if (tmpend < end) { | |
1453 | /* | |
1454 | * This may make start and end equal, but the zeroing | |
1455 | * code will skip any work in that case so there's no | |
1456 | * need to catch it up here. | |
1457 | */ | |
1458 | start = end & ~(osb->s_clustersize - 1); | |
1459 | ||
1460 | mlog(0, "2nd range: start: %llu, end: %llu\n", | |
1461 | (unsigned long long)start, (unsigned long long)end); | |
1462 | ||
1463 | ret = ocfs2_zero_range_for_truncate(inode, handle, start, end); | |
1464 | if (ret) | |
1465 | mlog_errno(ret); | |
1466 | } | |
1467 | ||
1468 | ocfs2_commit_trans(osb, handle); | |
1469 | out: | |
1470 | return ret; | |
1471 | } | |
1472 | ||
1473 | static int ocfs2_remove_inode_range(struct inode *inode, | |
1474 | struct buffer_head *di_bh, u64 byte_start, | |
1475 | u64 byte_len) | |
1476 | { | |
1477 | int ret = 0; | |
1478 | u32 trunc_start, trunc_len, cpos, phys_cpos, alloc_size; | |
1479 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1480 | struct ocfs2_cached_dealloc_ctxt dealloc; | |
b1967d0e | 1481 | struct address_space *mapping = inode->i_mapping; |
063c4561 MF |
1482 | |
1483 | ocfs2_init_dealloc_ctxt(&dealloc); | |
1484 | ||
1485 | if (byte_len == 0) | |
1486 | return 0; | |
1487 | ||
1afc32b9 MF |
1488 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) { |
1489 | ret = ocfs2_truncate_inline(inode, di_bh, byte_start, | |
b1967d0e MF |
1490 | byte_start + byte_len, 0); |
1491 | if (ret) { | |
1afc32b9 | 1492 | mlog_errno(ret); |
b1967d0e MF |
1493 | goto out; |
1494 | } | |
1495 | /* | |
1496 | * There's no need to get fancy with the page cache | |
1497 | * truncate of an inline-data inode. We're talking | |
1498 | * about less than a page here, which will be cached | |
1499 | * in the dinode buffer anyway. | |
1500 | */ | |
1501 | unmap_mapping_range(mapping, 0, 0, 0); | |
1502 | truncate_inode_pages(mapping, 0); | |
1503 | goto out; | |
1afc32b9 MF |
1504 | } |
1505 | ||
063c4561 MF |
1506 | trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start); |
1507 | trunc_len = (byte_start + byte_len) >> osb->s_clustersize_bits; | |
1508 | if (trunc_len >= trunc_start) | |
1509 | trunc_len -= trunc_start; | |
1510 | else | |
1511 | trunc_len = 0; | |
1512 | ||
1513 | mlog(0, "Inode: %llu, start: %llu, len: %llu, cstart: %u, clen: %u\n", | |
1514 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
1515 | (unsigned long long)byte_start, | |
1516 | (unsigned long long)byte_len, trunc_start, trunc_len); | |
1517 | ||
1518 | ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len); | |
1519 | if (ret) { | |
1520 | mlog_errno(ret); | |
1521 | goto out; | |
1522 | } | |
1523 | ||
1524 | cpos = trunc_start; | |
1525 | while (trunc_len) { | |
1526 | ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, | |
1527 | &alloc_size, NULL); | |
1528 | if (ret) { | |
1529 | mlog_errno(ret); | |
1530 | goto out; | |
1531 | } | |
1532 | ||
1533 | if (alloc_size > trunc_len) | |
1534 | alloc_size = trunc_len; | |
1535 | ||
1536 | /* Only do work for non-holes */ | |
1537 | if (phys_cpos != 0) { | |
1538 | ret = __ocfs2_remove_inode_range(inode, di_bh, cpos, | |
1539 | phys_cpos, alloc_size, | |
1540 | &dealloc); | |
1541 | if (ret) { | |
1542 | mlog_errno(ret); | |
1543 | goto out; | |
1544 | } | |
1545 | } | |
1546 | ||
1547 | cpos += alloc_size; | |
1548 | trunc_len -= alloc_size; | |
1549 | } | |
1550 | ||
1551 | ocfs2_truncate_cluster_pages(inode, byte_start, byte_len); | |
1552 | ||
1553 | out: | |
1554 | ocfs2_schedule_truncate_log_flush(osb, 1); | |
1555 | ocfs2_run_deallocs(osb, &dealloc); | |
1556 | ||
1557 | return ret; | |
1558 | } | |
1559 | ||
b2580103 MF |
1560 | /* |
1561 | * Parts of this function taken from xfs_change_file_space() | |
1562 | */ | |
385820a3 MF |
1563 | static int __ocfs2_change_file_space(struct file *file, struct inode *inode, |
1564 | loff_t f_pos, unsigned int cmd, | |
1565 | struct ocfs2_space_resv *sr, | |
1566 | int change_size) | |
b2580103 MF |
1567 | { |
1568 | int ret; | |
1569 | s64 llen; | |
385820a3 | 1570 | loff_t size; |
b2580103 MF |
1571 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
1572 | struct buffer_head *di_bh = NULL; | |
1573 | handle_t *handle; | |
a00cce35 | 1574 | unsigned long long max_off = inode->i_sb->s_maxbytes; |
b2580103 | 1575 | |
b2580103 MF |
1576 | if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb)) |
1577 | return -EROFS; | |
1578 | ||
1579 | mutex_lock(&inode->i_mutex); | |
1580 | ||
1581 | /* | |
1582 | * This prevents concurrent writes on other nodes | |
1583 | */ | |
1584 | ret = ocfs2_rw_lock(inode, 1); | |
1585 | if (ret) { | |
1586 | mlog_errno(ret); | |
1587 | goto out; | |
1588 | } | |
1589 | ||
e63aecb6 | 1590 | ret = ocfs2_inode_lock(inode, &di_bh, 1); |
b2580103 MF |
1591 | if (ret) { |
1592 | mlog_errno(ret); | |
1593 | goto out_rw_unlock; | |
1594 | } | |
1595 | ||
1596 | if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) { | |
1597 | ret = -EPERM; | |
e63aecb6 | 1598 | goto out_inode_unlock; |
b2580103 MF |
1599 | } |
1600 | ||
1601 | switch (sr->l_whence) { | |
1602 | case 0: /*SEEK_SET*/ | |
1603 | break; | |
1604 | case 1: /*SEEK_CUR*/ | |
385820a3 | 1605 | sr->l_start += f_pos; |
b2580103 MF |
1606 | break; |
1607 | case 2: /*SEEK_END*/ | |
1608 | sr->l_start += i_size_read(inode); | |
1609 | break; | |
1610 | default: | |
1611 | ret = -EINVAL; | |
e63aecb6 | 1612 | goto out_inode_unlock; |
b2580103 MF |
1613 | } |
1614 | sr->l_whence = 0; | |
1615 | ||
1616 | llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len; | |
1617 | ||
1618 | if (sr->l_start < 0 | |
1619 | || sr->l_start > max_off | |
1620 | || (sr->l_start + llen) < 0 | |
1621 | || (sr->l_start + llen) > max_off) { | |
1622 | ret = -EINVAL; | |
e63aecb6 | 1623 | goto out_inode_unlock; |
b2580103 | 1624 | } |
385820a3 | 1625 | size = sr->l_start + sr->l_len; |
b2580103 MF |
1626 | |
1627 | if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) { | |
1628 | if (sr->l_len <= 0) { | |
1629 | ret = -EINVAL; | |
e63aecb6 | 1630 | goto out_inode_unlock; |
b2580103 MF |
1631 | } |
1632 | } | |
1633 | ||
385820a3 | 1634 | if (file && should_remove_suid(file->f_path.dentry)) { |
b2580103 MF |
1635 | ret = __ocfs2_write_remove_suid(inode, di_bh); |
1636 | if (ret) { | |
1637 | mlog_errno(ret); | |
e63aecb6 | 1638 | goto out_inode_unlock; |
b2580103 MF |
1639 | } |
1640 | } | |
1641 | ||
1642 | down_write(&OCFS2_I(inode)->ip_alloc_sem); | |
1643 | switch (cmd) { | |
1644 | case OCFS2_IOC_RESVSP: | |
1645 | case OCFS2_IOC_RESVSP64: | |
1646 | /* | |
1647 | * This takes unsigned offsets, but the signed ones we | |
1648 | * pass have been checked against overflow above. | |
1649 | */ | |
1650 | ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start, | |
1651 | sr->l_len); | |
1652 | break; | |
1653 | case OCFS2_IOC_UNRESVSP: | |
1654 | case OCFS2_IOC_UNRESVSP64: | |
1655 | ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start, | |
1656 | sr->l_len); | |
1657 | break; | |
1658 | default: | |
1659 | ret = -EINVAL; | |
1660 | } | |
1661 | up_write(&OCFS2_I(inode)->ip_alloc_sem); | |
1662 | if (ret) { | |
1663 | mlog_errno(ret); | |
e63aecb6 | 1664 | goto out_inode_unlock; |
b2580103 MF |
1665 | } |
1666 | ||
1667 | /* | |
1668 | * We update c/mtime for these changes | |
1669 | */ | |
1670 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); | |
1671 | if (IS_ERR(handle)) { | |
1672 | ret = PTR_ERR(handle); | |
1673 | mlog_errno(ret); | |
e63aecb6 | 1674 | goto out_inode_unlock; |
b2580103 MF |
1675 | } |
1676 | ||
385820a3 MF |
1677 | if (change_size && i_size_read(inode) < size) |
1678 | i_size_write(inode, size); | |
1679 | ||
b2580103 MF |
1680 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; |
1681 | ret = ocfs2_mark_inode_dirty(handle, inode, di_bh); | |
1682 | if (ret < 0) | |
1683 | mlog_errno(ret); | |
1684 | ||
1685 | ocfs2_commit_trans(osb, handle); | |
1686 | ||
e63aecb6 | 1687 | out_inode_unlock: |
b2580103 | 1688 | brelse(di_bh); |
e63aecb6 | 1689 | ocfs2_inode_unlock(inode, 1); |
b2580103 MF |
1690 | out_rw_unlock: |
1691 | ocfs2_rw_unlock(inode, 1); | |
1692 | ||
b2580103 | 1693 | out: |
c259ae52 | 1694 | mutex_unlock(&inode->i_mutex); |
b2580103 MF |
1695 | return ret; |
1696 | } | |
1697 | ||
385820a3 MF |
1698 | int ocfs2_change_file_space(struct file *file, unsigned int cmd, |
1699 | struct ocfs2_space_resv *sr) | |
1700 | { | |
1701 | struct inode *inode = file->f_path.dentry->d_inode; | |
1702 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);; | |
1703 | ||
1704 | if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) && | |
1705 | !ocfs2_writes_unwritten_extents(osb)) | |
1706 | return -ENOTTY; | |
1707 | else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) && | |
1708 | !ocfs2_sparse_alloc(osb)) | |
1709 | return -ENOTTY; | |
1710 | ||
1711 | if (!S_ISREG(inode->i_mode)) | |
1712 | return -EINVAL; | |
1713 | ||
1714 | if (!(file->f_mode & FMODE_WRITE)) | |
1715 | return -EBADF; | |
1716 | ||
1717 | return __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0); | |
1718 | } | |
1719 | ||
1720 | static long ocfs2_fallocate(struct inode *inode, int mode, loff_t offset, | |
1721 | loff_t len) | |
1722 | { | |
1723 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1724 | struct ocfs2_space_resv sr; | |
1725 | int change_size = 1; | |
1726 | ||
1727 | if (!ocfs2_writes_unwritten_extents(osb)) | |
1728 | return -EOPNOTSUPP; | |
1729 | ||
1730 | if (S_ISDIR(inode->i_mode)) | |
1731 | return -ENODEV; | |
1732 | ||
1733 | if (mode & FALLOC_FL_KEEP_SIZE) | |
1734 | change_size = 0; | |
1735 | ||
1736 | sr.l_whence = 0; | |
1737 | sr.l_start = (s64)offset; | |
1738 | sr.l_len = (s64)len; | |
1739 | ||
1740 | return __ocfs2_change_file_space(NULL, inode, offset, | |
1741 | OCFS2_IOC_RESVSP64, &sr, change_size); | |
1742 | } | |
1743 | ||
8659ac25 TY |
1744 | static int ocfs2_prepare_inode_for_write(struct dentry *dentry, |
1745 | loff_t *ppos, | |
1746 | size_t count, | |
9517bac6 MF |
1747 | int appending, |
1748 | int *direct_io) | |
ccd979bd | 1749 | { |
65ed39d6 | 1750 | int ret = 0, meta_level = 0; |
8659ac25 | 1751 | struct inode *inode = dentry->d_inode; |
65ed39d6 | 1752 | loff_t saved_pos, end; |
ccd979bd | 1753 | |
ccd979bd | 1754 | /* |
65ed39d6 MF |
1755 | * We start with a read level meta lock and only jump to an ex |
1756 | * if we need to make modifications here. | |
ccd979bd | 1757 | */ |
ccd979bd | 1758 | for(;;) { |
e63aecb6 | 1759 | ret = ocfs2_inode_lock(inode, NULL, meta_level); |
ccd979bd MF |
1760 | if (ret < 0) { |
1761 | meta_level = -1; | |
1762 | mlog_errno(ret); | |
1763 | goto out; | |
1764 | } | |
1765 | ||
1766 | /* Clear suid / sgid if necessary. We do this here | |
1767 | * instead of later in the write path because | |
1768 | * remove_suid() calls ->setattr without any hint that | |
1769 | * we may have already done our cluster locking. Since | |
1770 | * ocfs2_setattr() *must* take cluster locks to | |
1771 | * proceeed, this will lead us to recursively lock the | |
1772 | * inode. There's also the dinode i_size state which | |
1773 | * can be lost via setattr during extending writes (we | |
1774 | * set inode->i_size at the end of a write. */ | |
8659ac25 | 1775 | if (should_remove_suid(dentry)) { |
ccd979bd | 1776 | if (meta_level == 0) { |
e63aecb6 | 1777 | ocfs2_inode_unlock(inode, meta_level); |
ccd979bd MF |
1778 | meta_level = 1; |
1779 | continue; | |
1780 | } | |
1781 | ||
1782 | ret = ocfs2_write_remove_suid(inode); | |
1783 | if (ret < 0) { | |
1784 | mlog_errno(ret); | |
8659ac25 | 1785 | goto out_unlock; |
ccd979bd MF |
1786 | } |
1787 | } | |
1788 | ||
1789 | /* work on a copy of ppos until we're sure that we won't have | |
1790 | * to recalculate it due to relocking. */ | |
8659ac25 | 1791 | if (appending) { |
ccd979bd MF |
1792 | saved_pos = i_size_read(inode); |
1793 | mlog(0, "O_APPEND: inode->i_size=%llu\n", saved_pos); | |
1794 | } else { | |
8659ac25 | 1795 | saved_pos = *ppos; |
ccd979bd | 1796 | } |
3a0782d0 | 1797 | |
65ed39d6 | 1798 | end = saved_pos + count; |
9517bac6 | 1799 | |
65ed39d6 MF |
1800 | /* |
1801 | * Skip the O_DIRECT checks if we don't need | |
1802 | * them. | |
1803 | */ | |
1804 | if (!direct_io || !(*direct_io)) | |
9517bac6 | 1805 | break; |
9517bac6 | 1806 | |
1afc32b9 MF |
1807 | /* |
1808 | * There's no sane way to do direct writes to an inode | |
1809 | * with inline data. | |
1810 | */ | |
1811 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) { | |
1812 | *direct_io = 0; | |
1813 | break; | |
1814 | } | |
1815 | ||
3a0782d0 | 1816 | /* |
65ed39d6 MF |
1817 | * Allowing concurrent direct writes means |
1818 | * i_size changes wouldn't be synchronized, so | |
1819 | * one node could wind up truncating another | |
1820 | * nodes writes. | |
3a0782d0 | 1821 | */ |
65ed39d6 MF |
1822 | if (end > i_size_read(inode)) { |
1823 | *direct_io = 0; | |
ccd979bd | 1824 | break; |
ccd979bd MF |
1825 | } |
1826 | ||
65ed39d6 MF |
1827 | /* |
1828 | * We don't fill holes during direct io, so | |
1829 | * check for them here. If any are found, the | |
1830 | * caller will have to retake some cluster | |
1831 | * locks and initiate the io as buffered. | |
1832 | */ | |
1833 | ret = ocfs2_check_range_for_holes(inode, saved_pos, count); | |
1834 | if (ret == 1) { | |
1835 | *direct_io = 0; | |
1836 | ret = 0; | |
1837 | } else if (ret < 0) | |
1838 | mlog_errno(ret); | |
ccd979bd MF |
1839 | break; |
1840 | } | |
1841 | ||
8659ac25 TY |
1842 | if (appending) |
1843 | *ppos = saved_pos; | |
1844 | ||
1845 | out_unlock: | |
e63aecb6 | 1846 | ocfs2_inode_unlock(inode, meta_level); |
8659ac25 TY |
1847 | |
1848 | out: | |
1849 | return ret; | |
1850 | } | |
1851 | ||
1852 | static ssize_t ocfs2_file_aio_write(struct kiocb *iocb, | |
1853 | const struct iovec *iov, | |
1854 | unsigned long nr_segs, | |
1855 | loff_t pos) | |
1856 | { | |
9517bac6 | 1857 | int ret, direct_io, appending, rw_level, have_alloc_sem = 0; |
b6af1bcd | 1858 | int can_do_direct; |
9517bac6 MF |
1859 | ssize_t written = 0; |
1860 | size_t ocount; /* original count */ | |
1861 | size_t count; /* after file limit checks */ | |
9ea2d32f MF |
1862 | loff_t old_size, *ppos = &iocb->ki_pos; |
1863 | u32 old_clusters; | |
9517bac6 MF |
1864 | struct file *file = iocb->ki_filp; |
1865 | struct inode *inode = file->f_path.dentry->d_inode; | |
9ea2d32f | 1866 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
9517bac6 MF |
1867 | |
1868 | mlog_entry("(0x%p, %u, '%.*s')\n", file, | |
8659ac25 | 1869 | (unsigned int)nr_segs, |
9517bac6 MF |
1870 | file->f_path.dentry->d_name.len, |
1871 | file->f_path.dentry->d_name.name); | |
8659ac25 | 1872 | |
8659ac25 TY |
1873 | if (iocb->ki_left == 0) |
1874 | return 0; | |
1875 | ||
9517bac6 MF |
1876 | vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE); |
1877 | ||
1878 | appending = file->f_flags & O_APPEND ? 1 : 0; | |
1879 | direct_io = file->f_flags & O_DIRECT ? 1 : 0; | |
1880 | ||
8659ac25 | 1881 | mutex_lock(&inode->i_mutex); |
9517bac6 MF |
1882 | |
1883 | relock: | |
8659ac25 | 1884 | /* to match setattr's i_mutex -> i_alloc_sem -> rw_lock ordering */ |
9517bac6 | 1885 | if (direct_io) { |
8659ac25 | 1886 | down_read(&inode->i_alloc_sem); |
9517bac6 | 1887 | have_alloc_sem = 1; |
8659ac25 TY |
1888 | } |
1889 | ||
1890 | /* concurrent O_DIRECT writes are allowed */ | |
9517bac6 | 1891 | rw_level = !direct_io; |
8659ac25 TY |
1892 | ret = ocfs2_rw_lock(inode, rw_level); |
1893 | if (ret < 0) { | |
8659ac25 | 1894 | mlog_errno(ret); |
9517bac6 | 1895 | goto out_sems; |
8659ac25 TY |
1896 | } |
1897 | ||
9517bac6 MF |
1898 | can_do_direct = direct_io; |
1899 | ret = ocfs2_prepare_inode_for_write(file->f_path.dentry, ppos, | |
1900 | iocb->ki_left, appending, | |
1901 | &can_do_direct); | |
8659ac25 TY |
1902 | if (ret < 0) { |
1903 | mlog_errno(ret); | |
1904 | goto out; | |
1905 | } | |
ccd979bd | 1906 | |
9517bac6 MF |
1907 | /* |
1908 | * We can't complete the direct I/O as requested, fall back to | |
1909 | * buffered I/O. | |
1910 | */ | |
1911 | if (direct_io && !can_do_direct) { | |
1912 | ocfs2_rw_unlock(inode, rw_level); | |
1913 | up_read(&inode->i_alloc_sem); | |
1914 | ||
1915 | have_alloc_sem = 0; | |
1916 | rw_level = -1; | |
1917 | ||
1918 | direct_io = 0; | |
9517bac6 MF |
1919 | goto relock; |
1920 | } | |
1921 | ||
9ea2d32f MF |
1922 | /* |
1923 | * To later detect whether a journal commit for sync writes is | |
1924 | * necessary, we sample i_size, and cluster count here. | |
1925 | */ | |
1926 | old_size = i_size_read(inode); | |
1927 | old_clusters = OCFS2_I(inode)->ip_clusters; | |
1928 | ||
ccd979bd | 1929 | /* communicate with ocfs2_dio_end_io */ |
7cdfc3a1 | 1930 | ocfs2_iocb_set_rw_locked(iocb, rw_level); |
ccd979bd | 1931 | |
9517bac6 | 1932 | if (direct_io) { |
b6af1bcd NP |
1933 | ret = generic_segment_checks(iov, &nr_segs, &ocount, |
1934 | VERIFY_READ); | |
1935 | if (ret) | |
1936 | goto out_dio; | |
1937 | ||
1938 | ret = generic_write_checks(file, ppos, &count, | |
1939 | S_ISBLK(inode->i_mode)); | |
1940 | if (ret) | |
1941 | goto out_dio; | |
1942 | ||
9517bac6 MF |
1943 | written = generic_file_direct_write(iocb, iov, &nr_segs, *ppos, |
1944 | ppos, count, ocount); | |
1945 | if (written < 0) { | |
1946 | ret = written; | |
1947 | goto out_dio; | |
1948 | } | |
1949 | } else { | |
b6af1bcd NP |
1950 | written = generic_file_aio_write_nolock(iocb, iov, nr_segs, |
1951 | *ppos); | |
9517bac6 | 1952 | } |
ccd979bd | 1953 | |
9517bac6 | 1954 | out_dio: |
ccd979bd | 1955 | /* buffered aio wouldn't have proper lock coverage today */ |
9517bac6 | 1956 | BUG_ON(ret == -EIOCBQUEUED && !(file->f_flags & O_DIRECT)); |
ccd979bd | 1957 | |
9ea2d32f MF |
1958 | if ((file->f_flags & O_SYNC && !direct_io) || IS_SYNC(inode)) { |
1959 | /* | |
1960 | * The generic write paths have handled getting data | |
1961 | * to disk, but since we don't make use of the dirty | |
1962 | * inode list, a manual journal commit is necessary | |
1963 | * here. | |
1964 | */ | |
1965 | if (old_size != i_size_read(inode) || | |
1966 | old_clusters != OCFS2_I(inode)->ip_clusters) { | |
1967 | ret = journal_force_commit(osb->journal->j_journal); | |
1968 | if (ret < 0) | |
1969 | written = ret; | |
1970 | } | |
1971 | } | |
1972 | ||
ccd979bd MF |
1973 | /* |
1974 | * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io | |
1975 | * function pointer which is called when o_direct io completes so that | |
1976 | * it can unlock our rw lock. (it's the clustered equivalent of | |
1977 | * i_alloc_sem; protects truncate from racing with pending ios). | |
1978 | * Unfortunately there are error cases which call end_io and others | |
1979 | * that don't. so we don't have to unlock the rw_lock if either an | |
1980 | * async dio is going to do it in the future or an end_io after an | |
1981 | * error has already done it. | |
1982 | */ | |
1983 | if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) { | |
1984 | rw_level = -1; | |
1985 | have_alloc_sem = 0; | |
1986 | } | |
1987 | ||
1988 | out: | |
9517bac6 MF |
1989 | if (rw_level != -1) |
1990 | ocfs2_rw_unlock(inode, rw_level); | |
1991 | ||
1992 | out_sems: | |
ccd979bd MF |
1993 | if (have_alloc_sem) |
1994 | up_read(&inode->i_alloc_sem); | |
9517bac6 | 1995 | |
1b1dcc1b | 1996 | mutex_unlock(&inode->i_mutex); |
ccd979bd MF |
1997 | |
1998 | mlog_exit(ret); | |
9517bac6 | 1999 | return written ? written : ret; |
ccd979bd MF |
2000 | } |
2001 | ||
8659ac25 TY |
2002 | static ssize_t ocfs2_file_splice_write(struct pipe_inode_info *pipe, |
2003 | struct file *out, | |
2004 | loff_t *ppos, | |
2005 | size_t len, | |
2006 | unsigned int flags) | |
2007 | { | |
2008 | int ret; | |
d28c9174 | 2009 | struct inode *inode = out->f_path.dentry->d_inode; |
8659ac25 TY |
2010 | |
2011 | mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", out, pipe, | |
2012 | (unsigned int)len, | |
d28c9174 JS |
2013 | out->f_path.dentry->d_name.len, |
2014 | out->f_path.dentry->d_name.name); | |
8659ac25 TY |
2015 | |
2016 | inode_double_lock(inode, pipe->inode); | |
2017 | ||
2018 | ret = ocfs2_rw_lock(inode, 1); | |
2019 | if (ret < 0) { | |
2020 | mlog_errno(ret); | |
2021 | goto out; | |
2022 | } | |
2023 | ||
9517bac6 MF |
2024 | ret = ocfs2_prepare_inode_for_write(out->f_path.dentry, ppos, len, 0, |
2025 | NULL); | |
8659ac25 TY |
2026 | if (ret < 0) { |
2027 | mlog_errno(ret); | |
2028 | goto out_unlock; | |
2029 | } | |
2030 | ||
b6af1bcd | 2031 | ret = generic_file_splice_write_nolock(pipe, out, ppos, len, flags); |
8659ac25 TY |
2032 | |
2033 | out_unlock: | |
2034 | ocfs2_rw_unlock(inode, 1); | |
2035 | out: | |
2036 | inode_double_unlock(inode, pipe->inode); | |
2037 | ||
2038 | mlog_exit(ret); | |
2039 | return ret; | |
2040 | } | |
2041 | ||
2042 | static ssize_t ocfs2_file_splice_read(struct file *in, | |
2043 | loff_t *ppos, | |
2044 | struct pipe_inode_info *pipe, | |
2045 | size_t len, | |
2046 | unsigned int flags) | |
2047 | { | |
2048 | int ret = 0; | |
d28c9174 | 2049 | struct inode *inode = in->f_path.dentry->d_inode; |
8659ac25 TY |
2050 | |
2051 | mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", in, pipe, | |
2052 | (unsigned int)len, | |
d28c9174 JS |
2053 | in->f_path.dentry->d_name.len, |
2054 | in->f_path.dentry->d_name.name); | |
8659ac25 TY |
2055 | |
2056 | /* | |
2057 | * See the comment in ocfs2_file_aio_read() | |
2058 | */ | |
e63aecb6 | 2059 | ret = ocfs2_inode_lock(inode, NULL, 0); |
8659ac25 TY |
2060 | if (ret < 0) { |
2061 | mlog_errno(ret); | |
2062 | goto bail; | |
2063 | } | |
e63aecb6 | 2064 | ocfs2_inode_unlock(inode, 0); |
8659ac25 TY |
2065 | |
2066 | ret = generic_file_splice_read(in, ppos, pipe, len, flags); | |
2067 | ||
2068 | bail: | |
2069 | mlog_exit(ret); | |
2070 | return ret; | |
2071 | } | |
2072 | ||
ccd979bd | 2073 | static ssize_t ocfs2_file_aio_read(struct kiocb *iocb, |
027445c3 BP |
2074 | const struct iovec *iov, |
2075 | unsigned long nr_segs, | |
ccd979bd MF |
2076 | loff_t pos) |
2077 | { | |
25899dee | 2078 | int ret = 0, rw_level = -1, have_alloc_sem = 0, lock_level = 0; |
ccd979bd | 2079 | struct file *filp = iocb->ki_filp; |
d28c9174 | 2080 | struct inode *inode = filp->f_path.dentry->d_inode; |
ccd979bd | 2081 | |
027445c3 BP |
2082 | mlog_entry("(0x%p, %u, '%.*s')\n", filp, |
2083 | (unsigned int)nr_segs, | |
d28c9174 JS |
2084 | filp->f_path.dentry->d_name.len, |
2085 | filp->f_path.dentry->d_name.name); | |
ccd979bd MF |
2086 | |
2087 | if (!inode) { | |
2088 | ret = -EINVAL; | |
2089 | mlog_errno(ret); | |
2090 | goto bail; | |
2091 | } | |
2092 | ||
ccd979bd MF |
2093 | /* |
2094 | * buffered reads protect themselves in ->readpage(). O_DIRECT reads | |
2095 | * need locks to protect pending reads from racing with truncate. | |
2096 | */ | |
2097 | if (filp->f_flags & O_DIRECT) { | |
2098 | down_read(&inode->i_alloc_sem); | |
2099 | have_alloc_sem = 1; | |
2100 | ||
2101 | ret = ocfs2_rw_lock(inode, 0); | |
2102 | if (ret < 0) { | |
2103 | mlog_errno(ret); | |
2104 | goto bail; | |
2105 | } | |
2106 | rw_level = 0; | |
2107 | /* communicate with ocfs2_dio_end_io */ | |
7cdfc3a1 | 2108 | ocfs2_iocb_set_rw_locked(iocb, rw_level); |
ccd979bd MF |
2109 | } |
2110 | ||
c4374f8a MF |
2111 | /* |
2112 | * We're fine letting folks race truncates and extending | |
2113 | * writes with read across the cluster, just like they can | |
2114 | * locally. Hence no rw_lock during read. | |
2115 | * | |
2116 | * Take and drop the meta data lock to update inode fields | |
2117 | * like i_size. This allows the checks down below | |
2118 | * generic_file_aio_read() a chance of actually working. | |
2119 | */ | |
e63aecb6 | 2120 | ret = ocfs2_inode_lock_atime(inode, filp->f_vfsmnt, &lock_level); |
c4374f8a MF |
2121 | if (ret < 0) { |
2122 | mlog_errno(ret); | |
2123 | goto bail; | |
2124 | } | |
e63aecb6 | 2125 | ocfs2_inode_unlock(inode, lock_level); |
c4374f8a | 2126 | |
027445c3 | 2127 | ret = generic_file_aio_read(iocb, iov, nr_segs, iocb->ki_pos); |
ccd979bd | 2128 | if (ret == -EINVAL) |
56753bd3 | 2129 | mlog(0, "generic_file_aio_read returned -EINVAL\n"); |
ccd979bd MF |
2130 | |
2131 | /* buffered aio wouldn't have proper lock coverage today */ | |
2132 | BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT)); | |
2133 | ||
2134 | /* see ocfs2_file_aio_write */ | |
2135 | if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) { | |
2136 | rw_level = -1; | |
2137 | have_alloc_sem = 0; | |
2138 | } | |
2139 | ||
2140 | bail: | |
2141 | if (have_alloc_sem) | |
2142 | up_read(&inode->i_alloc_sem); | |
2143 | if (rw_level != -1) | |
2144 | ocfs2_rw_unlock(inode, rw_level); | |
2145 | mlog_exit(ret); | |
2146 | ||
2147 | return ret; | |
2148 | } | |
2149 | ||
92e1d5be | 2150 | const struct inode_operations ocfs2_file_iops = { |
ccd979bd MF |
2151 | .setattr = ocfs2_setattr, |
2152 | .getattr = ocfs2_getattr, | |
d38eb8db | 2153 | .permission = ocfs2_permission, |
385820a3 | 2154 | .fallocate = ocfs2_fallocate, |
00dc417f | 2155 | .fiemap = ocfs2_fiemap, |
ccd979bd MF |
2156 | }; |
2157 | ||
92e1d5be | 2158 | const struct inode_operations ocfs2_special_file_iops = { |
ccd979bd MF |
2159 | .setattr = ocfs2_setattr, |
2160 | .getattr = ocfs2_getattr, | |
d38eb8db | 2161 | .permission = ocfs2_permission, |
ccd979bd MF |
2162 | }; |
2163 | ||
53da4939 MF |
2164 | /* |
2165 | * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with | |
2166 | * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks! | |
2167 | */ | |
4b6f5d20 | 2168 | const struct file_operations ocfs2_fops = { |
32c3c0e2 | 2169 | .llseek = generic_file_llseek, |
ccd979bd MF |
2170 | .read = do_sync_read, |
2171 | .write = do_sync_write, | |
ccd979bd MF |
2172 | .mmap = ocfs2_mmap, |
2173 | .fsync = ocfs2_sync_file, | |
2174 | .release = ocfs2_file_release, | |
2175 | .open = ocfs2_file_open, | |
2176 | .aio_read = ocfs2_file_aio_read, | |
2177 | .aio_write = ocfs2_file_aio_write, | |
c9ec1488 | 2178 | .unlocked_ioctl = ocfs2_ioctl, |
586d232b MF |
2179 | #ifdef CONFIG_COMPAT |
2180 | .compat_ioctl = ocfs2_compat_ioctl, | |
2181 | #endif | |
53da4939 | 2182 | .lock = ocfs2_lock, |
53fc622b | 2183 | .flock = ocfs2_flock, |
8659ac25 TY |
2184 | .splice_read = ocfs2_file_splice_read, |
2185 | .splice_write = ocfs2_file_splice_write, | |
ccd979bd MF |
2186 | }; |
2187 | ||
4b6f5d20 | 2188 | const struct file_operations ocfs2_dops = { |
32c3c0e2 | 2189 | .llseek = generic_file_llseek, |
ccd979bd MF |
2190 | .read = generic_read_dir, |
2191 | .readdir = ocfs2_readdir, | |
2192 | .fsync = ocfs2_sync_file, | |
53fc622b MF |
2193 | .release = ocfs2_dir_release, |
2194 | .open = ocfs2_dir_open, | |
c9ec1488 | 2195 | .unlocked_ioctl = ocfs2_ioctl, |
586d232b MF |
2196 | #ifdef CONFIG_COMPAT |
2197 | .compat_ioctl = ocfs2_compat_ioctl, | |
53da4939 MF |
2198 | #endif |
2199 | .lock = ocfs2_lock, | |
2200 | .flock = ocfs2_flock, | |
2201 | }; | |
2202 | ||
2203 | /* | |
2204 | * POSIX-lockless variants of our file_operations. | |
2205 | * | |
2206 | * These will be used if the underlying cluster stack does not support | |
2207 | * posix file locking, if the user passes the "localflocks" mount | |
2208 | * option, or if we have a local-only fs. | |
2209 | * | |
2210 | * ocfs2_flock is in here because all stacks handle UNIX file locks, | |
2211 | * so we still want it in the case of no stack support for | |
2212 | * plocks. Internally, it will do the right thing when asked to ignore | |
2213 | * the cluster. | |
2214 | */ | |
2215 | const struct file_operations ocfs2_fops_no_plocks = { | |
2216 | .llseek = generic_file_llseek, | |
2217 | .read = do_sync_read, | |
2218 | .write = do_sync_write, | |
2219 | .mmap = ocfs2_mmap, | |
2220 | .fsync = ocfs2_sync_file, | |
2221 | .release = ocfs2_file_release, | |
2222 | .open = ocfs2_file_open, | |
2223 | .aio_read = ocfs2_file_aio_read, | |
2224 | .aio_write = ocfs2_file_aio_write, | |
2225 | .unlocked_ioctl = ocfs2_ioctl, | |
2226 | #ifdef CONFIG_COMPAT | |
2227 | .compat_ioctl = ocfs2_compat_ioctl, | |
2228 | #endif | |
2229 | .flock = ocfs2_flock, | |
2230 | .splice_read = ocfs2_file_splice_read, | |
2231 | .splice_write = ocfs2_file_splice_write, | |
2232 | }; | |
2233 | ||
2234 | const struct file_operations ocfs2_dops_no_plocks = { | |
2235 | .llseek = generic_file_llseek, | |
2236 | .read = generic_read_dir, | |
2237 | .readdir = ocfs2_readdir, | |
2238 | .fsync = ocfs2_sync_file, | |
2239 | .release = ocfs2_dir_release, | |
2240 | .open = ocfs2_dir_open, | |
2241 | .unlocked_ioctl = ocfs2_ioctl, | |
2242 | #ifdef CONFIG_COMPAT | |
2243 | .compat_ioctl = ocfs2_compat_ioctl, | |
586d232b | 2244 | #endif |
53fc622b | 2245 | .flock = ocfs2_flock, |
ccd979bd | 2246 | }; |