]>
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" | |
cf1d6c76 | 58 | #include "xattr.h" |
ccd979bd MF |
59 | |
60 | #include "buffer_head_io.h" | |
61 | ||
62 | static int ocfs2_sync_inode(struct inode *inode) | |
63 | { | |
64 | filemap_fdatawrite(inode->i_mapping); | |
65 | return sync_mapping_buffers(inode->i_mapping); | |
66 | } | |
67 | ||
53fc622b MF |
68 | static int ocfs2_init_file_private(struct inode *inode, struct file *file) |
69 | { | |
70 | struct ocfs2_file_private *fp; | |
71 | ||
72 | fp = kzalloc(sizeof(struct ocfs2_file_private), GFP_KERNEL); | |
73 | if (!fp) | |
74 | return -ENOMEM; | |
75 | ||
76 | fp->fp_file = file; | |
77 | mutex_init(&fp->fp_mutex); | |
78 | ocfs2_file_lock_res_init(&fp->fp_flock, fp); | |
79 | file->private_data = fp; | |
80 | ||
81 | return 0; | |
82 | } | |
83 | ||
84 | static void ocfs2_free_file_private(struct inode *inode, struct file *file) | |
85 | { | |
86 | struct ocfs2_file_private *fp = file->private_data; | |
87 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
88 | ||
89 | if (fp) { | |
90 | ocfs2_simple_drop_lockres(osb, &fp->fp_flock); | |
91 | ocfs2_lock_res_free(&fp->fp_flock); | |
92 | kfree(fp); | |
93 | file->private_data = NULL; | |
94 | } | |
95 | } | |
96 | ||
ccd979bd MF |
97 | static int ocfs2_file_open(struct inode *inode, struct file *file) |
98 | { | |
99 | int status; | |
100 | int mode = file->f_flags; | |
101 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
102 | ||
103 | mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file, | |
d28c9174 | 104 | file->f_path.dentry->d_name.len, file->f_path.dentry->d_name.name); |
ccd979bd MF |
105 | |
106 | spin_lock(&oi->ip_lock); | |
107 | ||
108 | /* Check that the inode hasn't been wiped from disk by another | |
109 | * node. If it hasn't then we're safe as long as we hold the | |
110 | * spin lock until our increment of open count. */ | |
111 | if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) { | |
112 | spin_unlock(&oi->ip_lock); | |
113 | ||
114 | status = -ENOENT; | |
115 | goto leave; | |
116 | } | |
117 | ||
118 | if (mode & O_DIRECT) | |
119 | oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT; | |
120 | ||
121 | oi->ip_open_count++; | |
122 | spin_unlock(&oi->ip_lock); | |
53fc622b MF |
123 | |
124 | status = ocfs2_init_file_private(inode, file); | |
125 | if (status) { | |
126 | /* | |
127 | * We want to set open count back if we're failing the | |
128 | * open. | |
129 | */ | |
130 | spin_lock(&oi->ip_lock); | |
131 | oi->ip_open_count--; | |
132 | spin_unlock(&oi->ip_lock); | |
133 | } | |
134 | ||
ccd979bd MF |
135 | leave: |
136 | mlog_exit(status); | |
137 | return status; | |
138 | } | |
139 | ||
140 | static int ocfs2_file_release(struct inode *inode, struct file *file) | |
141 | { | |
142 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
143 | ||
144 | mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file, | |
d28c9174 JS |
145 | file->f_path.dentry->d_name.len, |
146 | file->f_path.dentry->d_name.name); | |
ccd979bd MF |
147 | |
148 | spin_lock(&oi->ip_lock); | |
149 | if (!--oi->ip_open_count) | |
150 | oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT; | |
151 | spin_unlock(&oi->ip_lock); | |
152 | ||
53fc622b MF |
153 | ocfs2_free_file_private(inode, file); |
154 | ||
ccd979bd MF |
155 | mlog_exit(0); |
156 | ||
157 | return 0; | |
158 | } | |
159 | ||
53fc622b MF |
160 | static int ocfs2_dir_open(struct inode *inode, struct file *file) |
161 | { | |
162 | return ocfs2_init_file_private(inode, file); | |
163 | } | |
164 | ||
165 | static int ocfs2_dir_release(struct inode *inode, struct file *file) | |
166 | { | |
167 | ocfs2_free_file_private(inode, file); | |
168 | return 0; | |
169 | } | |
170 | ||
ccd979bd MF |
171 | static int ocfs2_sync_file(struct file *file, |
172 | struct dentry *dentry, | |
173 | int datasync) | |
174 | { | |
175 | int err = 0; | |
176 | journal_t *journal; | |
177 | struct inode *inode = dentry->d_inode; | |
178 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
179 | ||
180 | mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", file, dentry, datasync, | |
181 | dentry->d_name.len, dentry->d_name.name); | |
182 | ||
183 | err = ocfs2_sync_inode(dentry->d_inode); | |
184 | if (err) | |
185 | goto bail; | |
186 | ||
187 | journal = osb->journal->j_journal; | |
188 | err = journal_force_commit(journal); | |
189 | ||
190 | bail: | |
191 | mlog_exit(err); | |
192 | ||
193 | return (err < 0) ? -EIO : 0; | |
194 | } | |
195 | ||
7f1a37e3 TY |
196 | int ocfs2_should_update_atime(struct inode *inode, |
197 | struct vfsmount *vfsmnt) | |
198 | { | |
199 | struct timespec now; | |
200 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
201 | ||
202 | if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb)) | |
203 | return 0; | |
204 | ||
205 | if ((inode->i_flags & S_NOATIME) || | |
206 | ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))) | |
207 | return 0; | |
208 | ||
6c2aad05 MF |
209 | /* |
210 | * We can be called with no vfsmnt structure - NFSD will | |
211 | * sometimes do this. | |
212 | * | |
213 | * Note that our action here is different than touch_atime() - | |
214 | * if we can't tell whether this is a noatime mount, then we | |
215 | * don't know whether to trust the value of s_atime_quantum. | |
216 | */ | |
217 | if (vfsmnt == NULL) | |
218 | return 0; | |
219 | ||
7f1a37e3 TY |
220 | if ((vfsmnt->mnt_flags & MNT_NOATIME) || |
221 | ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))) | |
222 | return 0; | |
223 | ||
7e913c53 MF |
224 | if (vfsmnt->mnt_flags & MNT_RELATIME) { |
225 | if ((timespec_compare(&inode->i_atime, &inode->i_mtime) <= 0) || | |
226 | (timespec_compare(&inode->i_atime, &inode->i_ctime) <= 0)) | |
227 | return 1; | |
228 | ||
229 | return 0; | |
230 | } | |
231 | ||
7f1a37e3 TY |
232 | now = CURRENT_TIME; |
233 | if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum)) | |
234 | return 0; | |
235 | else | |
236 | return 1; | |
237 | } | |
238 | ||
239 | int ocfs2_update_inode_atime(struct inode *inode, | |
240 | struct buffer_head *bh) | |
241 | { | |
242 | int ret; | |
243 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
244 | handle_t *handle; | |
c11e9faf | 245 | struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data; |
7f1a37e3 TY |
246 | |
247 | mlog_entry_void(); | |
248 | ||
249 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); | |
250 | if (handle == NULL) { | |
251 | ret = -ENOMEM; | |
252 | mlog_errno(ret); | |
253 | goto out; | |
254 | } | |
255 | ||
c11e9faf MF |
256 | ret = ocfs2_journal_access(handle, inode, bh, |
257 | OCFS2_JOURNAL_ACCESS_WRITE); | |
258 | if (ret) { | |
259 | mlog_errno(ret); | |
260 | goto out_commit; | |
261 | } | |
262 | ||
263 | /* | |
264 | * Don't use ocfs2_mark_inode_dirty() here as we don't always | |
265 | * have i_mutex to guard against concurrent changes to other | |
266 | * inode fields. | |
267 | */ | |
7f1a37e3 | 268 | inode->i_atime = CURRENT_TIME; |
c11e9faf MF |
269 | di->i_atime = cpu_to_le64(inode->i_atime.tv_sec); |
270 | di->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec); | |
271 | ||
272 | ret = ocfs2_journal_dirty(handle, bh); | |
7f1a37e3 TY |
273 | if (ret < 0) |
274 | mlog_errno(ret); | |
275 | ||
c11e9faf | 276 | out_commit: |
7f1a37e3 TY |
277 | ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle); |
278 | out: | |
279 | mlog_exit(ret); | |
280 | return ret; | |
281 | } | |
282 | ||
6cb129f5 AB |
283 | static int ocfs2_set_inode_size(handle_t *handle, |
284 | struct inode *inode, | |
285 | struct buffer_head *fe_bh, | |
286 | u64 new_i_size) | |
ccd979bd MF |
287 | { |
288 | int status; | |
289 | ||
290 | mlog_entry_void(); | |
291 | i_size_write(inode, new_i_size); | |
8110b073 | 292 | inode->i_blocks = ocfs2_inode_sector_count(inode); |
ccd979bd MF |
293 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; |
294 | ||
295 | status = ocfs2_mark_inode_dirty(handle, inode, fe_bh); | |
296 | if (status < 0) { | |
297 | mlog_errno(status); | |
298 | goto bail; | |
299 | } | |
300 | ||
301 | bail: | |
302 | mlog_exit(status); | |
303 | return status; | |
304 | } | |
305 | ||
306 | static int ocfs2_simple_size_update(struct inode *inode, | |
307 | struct buffer_head *di_bh, | |
308 | u64 new_i_size) | |
309 | { | |
310 | int ret; | |
311 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1fabe148 | 312 | handle_t *handle = NULL; |
ccd979bd | 313 | |
65eff9cc | 314 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
ccd979bd MF |
315 | if (handle == NULL) { |
316 | ret = -ENOMEM; | |
317 | mlog_errno(ret); | |
318 | goto out; | |
319 | } | |
320 | ||
321 | ret = ocfs2_set_inode_size(handle, inode, di_bh, | |
322 | new_i_size); | |
323 | if (ret < 0) | |
324 | mlog_errno(ret); | |
325 | ||
02dc1af4 | 326 | ocfs2_commit_trans(osb, handle); |
ccd979bd MF |
327 | out: |
328 | return ret; | |
329 | } | |
330 | ||
331 | static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb, | |
332 | struct inode *inode, | |
333 | struct buffer_head *fe_bh, | |
334 | u64 new_i_size) | |
335 | { | |
336 | int status; | |
1fabe148 | 337 | handle_t *handle; |
60b11392 | 338 | struct ocfs2_dinode *di; |
35edec1d | 339 | u64 cluster_bytes; |
ccd979bd MF |
340 | |
341 | mlog_entry_void(); | |
342 | ||
343 | /* TODO: This needs to actually orphan the inode in this | |
344 | * transaction. */ | |
345 | ||
65eff9cc | 346 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
ccd979bd MF |
347 | if (IS_ERR(handle)) { |
348 | status = PTR_ERR(handle); | |
349 | mlog_errno(status); | |
350 | goto out; | |
351 | } | |
352 | ||
60b11392 MF |
353 | status = ocfs2_journal_access(handle, inode, fe_bh, |
354 | OCFS2_JOURNAL_ACCESS_WRITE); | |
355 | if (status < 0) { | |
356 | mlog_errno(status); | |
357 | goto out_commit; | |
358 | } | |
359 | ||
360 | /* | |
361 | * Do this before setting i_size. | |
362 | */ | |
35edec1d MF |
363 | cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size); |
364 | status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size, | |
365 | cluster_bytes); | |
60b11392 MF |
366 | if (status) { |
367 | mlog_errno(status); | |
368 | goto out_commit; | |
369 | } | |
370 | ||
371 | i_size_write(inode, new_i_size); | |
60b11392 MF |
372 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; |
373 | ||
374 | di = (struct ocfs2_dinode *) fe_bh->b_data; | |
375 | di->i_size = cpu_to_le64(new_i_size); | |
376 | di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec); | |
377 | di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec); | |
378 | ||
379 | status = ocfs2_journal_dirty(handle, fe_bh); | |
ccd979bd MF |
380 | if (status < 0) |
381 | mlog_errno(status); | |
382 | ||
60b11392 | 383 | out_commit: |
02dc1af4 | 384 | ocfs2_commit_trans(osb, handle); |
ccd979bd | 385 | out: |
60b11392 | 386 | |
ccd979bd MF |
387 | mlog_exit(status); |
388 | return status; | |
389 | } | |
390 | ||
391 | static int ocfs2_truncate_file(struct inode *inode, | |
392 | struct buffer_head *di_bh, | |
393 | u64 new_i_size) | |
394 | { | |
395 | int status = 0; | |
396 | struct ocfs2_dinode *fe = NULL; | |
397 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
398 | struct ocfs2_truncate_context *tc = NULL; | |
399 | ||
b0697053 MF |
400 | mlog_entry("(inode = %llu, new_i_size = %llu\n", |
401 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
402 | (unsigned long long)new_i_size); | |
ccd979bd | 403 | |
ccd979bd MF |
404 | fe = (struct ocfs2_dinode *) di_bh->b_data; |
405 | if (!OCFS2_IS_VALID_DINODE(fe)) { | |
406 | OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe); | |
407 | status = -EIO; | |
408 | goto bail; | |
409 | } | |
410 | ||
411 | mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode), | |
b0697053 MF |
412 | "Inode %llu, inode i_size = %lld != di " |
413 | "i_size = %llu, i_flags = 0x%x\n", | |
414 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
ccd979bd | 415 | i_size_read(inode), |
b0697053 MF |
416 | (unsigned long long)le64_to_cpu(fe->i_size), |
417 | le32_to_cpu(fe->i_flags)); | |
ccd979bd MF |
418 | |
419 | if (new_i_size > le64_to_cpu(fe->i_size)) { | |
b0697053 MF |
420 | mlog(0, "asked to truncate file with size (%llu) to size (%llu)!\n", |
421 | (unsigned long long)le64_to_cpu(fe->i_size), | |
422 | (unsigned long long)new_i_size); | |
ccd979bd MF |
423 | status = -EINVAL; |
424 | mlog_errno(status); | |
425 | goto bail; | |
426 | } | |
427 | ||
b0697053 MF |
428 | mlog(0, "inode %llu, i_size = %llu, new_i_size = %llu\n", |
429 | (unsigned long long)le64_to_cpu(fe->i_blkno), | |
430 | (unsigned long long)le64_to_cpu(fe->i_size), | |
431 | (unsigned long long)new_i_size); | |
ccd979bd MF |
432 | |
433 | /* lets handle the simple truncate cases before doing any more | |
434 | * cluster locking. */ | |
435 | if (new_i_size == le64_to_cpu(fe->i_size)) | |
436 | goto bail; | |
437 | ||
2e89b2e4 MF |
438 | down_write(&OCFS2_I(inode)->ip_alloc_sem); |
439 | ||
c934a92d MF |
440 | /* |
441 | * The inode lock forced other nodes to sync and drop their | |
442 | * pages, which (correctly) happens even if we have a truncate | |
443 | * without allocation change - ocfs2 cluster sizes can be much | |
444 | * greater than page size, so we have to truncate them | |
445 | * anyway. | |
446 | */ | |
2e89b2e4 MF |
447 | unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1); |
448 | truncate_inode_pages(inode->i_mapping, new_i_size); | |
449 | ||
1afc32b9 MF |
450 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) { |
451 | status = ocfs2_truncate_inline(inode, di_bh, new_i_size, | |
b1967d0e | 452 | i_size_read(inode), 1); |
1afc32b9 MF |
453 | if (status) |
454 | mlog_errno(status); | |
455 | ||
c934a92d | 456 | goto bail_unlock_sem; |
1afc32b9 MF |
457 | } |
458 | ||
ccd979bd MF |
459 | /* alright, we're going to need to do a full blown alloc size |
460 | * change. Orphan the inode so that recovery can complete the | |
461 | * truncate if necessary. This does the task of marking | |
462 | * i_size. */ | |
463 | status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size); | |
464 | if (status < 0) { | |
465 | mlog_errno(status); | |
c934a92d | 466 | goto bail_unlock_sem; |
ccd979bd MF |
467 | } |
468 | ||
469 | status = ocfs2_prepare_truncate(osb, inode, di_bh, &tc); | |
470 | if (status < 0) { | |
471 | mlog_errno(status); | |
c934a92d | 472 | goto bail_unlock_sem; |
ccd979bd MF |
473 | } |
474 | ||
475 | status = ocfs2_commit_truncate(osb, inode, di_bh, tc); | |
476 | if (status < 0) { | |
477 | mlog_errno(status); | |
c934a92d | 478 | goto bail_unlock_sem; |
ccd979bd MF |
479 | } |
480 | ||
481 | /* TODO: orphan dir cleanup here. */ | |
c934a92d | 482 | bail_unlock_sem: |
2e89b2e4 MF |
483 | up_write(&OCFS2_I(inode)->ip_alloc_sem); |
484 | ||
ccd979bd MF |
485 | bail: |
486 | ||
487 | mlog_exit(status); | |
488 | return status; | |
489 | } | |
490 | ||
491 | /* | |
0eb8d47e | 492 | * extend file allocation only here. |
ccd979bd MF |
493 | * we'll update all the disk stuff, and oip->alloc_size |
494 | * | |
495 | * expect stuff to be locked, a transaction started and enough data / | |
496 | * metadata reservations in the contexts. | |
497 | * | |
498 | * Will return -EAGAIN, and a reason if a restart is needed. | |
499 | * If passed in, *reason will always be set, even in error. | |
500 | */ | |
0eb8d47e TM |
501 | int ocfs2_add_inode_data(struct ocfs2_super *osb, |
502 | struct inode *inode, | |
503 | u32 *logical_offset, | |
504 | u32 clusters_to_add, | |
505 | int mark_unwritten, | |
506 | struct buffer_head *fe_bh, | |
507 | handle_t *handle, | |
508 | struct ocfs2_alloc_context *data_ac, | |
509 | struct ocfs2_alloc_context *meta_ac, | |
510 | enum ocfs2_alloc_restarted *reason_ret) | |
ccd979bd | 511 | { |
f99b9b7c JB |
512 | int ret; |
513 | struct ocfs2_extent_tree et; | |
ccd979bd | 514 | |
f99b9b7c JB |
515 | ocfs2_get_dinode_extent_tree(&et, inode, fe_bh); |
516 | ret = ocfs2_add_clusters_in_btree(osb, inode, logical_offset, | |
0eb8d47e | 517 | clusters_to_add, mark_unwritten, |
f99b9b7c JB |
518 | &et, handle, |
519 | data_ac, meta_ac, reason_ret); | |
520 | ocfs2_put_extent_tree(&et); | |
521 | ||
522 | return ret; | |
ccd979bd MF |
523 | } |
524 | ||
2ae99a60 MF |
525 | static int __ocfs2_extend_allocation(struct inode *inode, u32 logical_start, |
526 | u32 clusters_to_add, int mark_unwritten) | |
ccd979bd MF |
527 | { |
528 | int status = 0; | |
529 | int restart_func = 0; | |
abf8b156 | 530 | int credits; |
2ae99a60 | 531 | u32 prev_clusters; |
ccd979bd MF |
532 | struct buffer_head *bh = NULL; |
533 | struct ocfs2_dinode *fe = NULL; | |
1fabe148 | 534 | handle_t *handle = NULL; |
ccd979bd MF |
535 | struct ocfs2_alloc_context *data_ac = NULL; |
536 | struct ocfs2_alloc_context *meta_ac = NULL; | |
537 | enum ocfs2_alloc_restarted why; | |
538 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
f99b9b7c | 539 | struct ocfs2_extent_tree et; |
ccd979bd MF |
540 | |
541 | mlog_entry("(clusters_to_add = %u)\n", clusters_to_add); | |
542 | ||
dcd0538f MF |
543 | /* |
544 | * This function only exists for file systems which don't | |
545 | * support holes. | |
546 | */ | |
2ae99a60 | 547 | BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb)); |
dcd0538f | 548 | |
ccd979bd MF |
549 | status = ocfs2_read_block(osb, OCFS2_I(inode)->ip_blkno, &bh, |
550 | OCFS2_BH_CACHED, inode); | |
551 | if (status < 0) { | |
552 | mlog_errno(status); | |
553 | goto leave; | |
554 | } | |
555 | ||
556 | fe = (struct ocfs2_dinode *) bh->b_data; | |
557 | if (!OCFS2_IS_VALID_DINODE(fe)) { | |
558 | OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe); | |
559 | status = -EIO; | |
560 | goto leave; | |
561 | } | |
562 | ||
563 | restart_all: | |
564 | BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters); | |
565 | ||
e7d4cb6b TM |
566 | mlog(0, "extend inode %llu, i_size = %lld, di->i_clusters = %u, " |
567 | "clusters_to_add = %u\n", | |
568 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
569 | (long long)i_size_read(inode), le32_to_cpu(fe->i_clusters), | |
570 | clusters_to_add); | |
f99b9b7c JB |
571 | ocfs2_get_dinode_extent_tree(&et, inode, bh); |
572 | status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0, | |
573 | &data_ac, &meta_ac); | |
574 | ocfs2_put_extent_tree(&et); | |
9517bac6 MF |
575 | if (status) { |
576 | mlog_errno(status); | |
577 | goto leave; | |
578 | } | |
579 | ||
811f933d TM |
580 | credits = ocfs2_calc_extend_credits(osb->sb, &fe->id2.i_list, |
581 | clusters_to_add); | |
65eff9cc | 582 | handle = ocfs2_start_trans(osb, credits); |
ccd979bd MF |
583 | if (IS_ERR(handle)) { |
584 | status = PTR_ERR(handle); | |
585 | handle = NULL; | |
586 | mlog_errno(status); | |
587 | goto leave; | |
588 | } | |
589 | ||
590 | restarted_transaction: | |
591 | /* reserve a write to the file entry early on - that we if we | |
592 | * run out of credits in the allocation path, we can still | |
593 | * update i_size. */ | |
594 | status = ocfs2_journal_access(handle, inode, bh, | |
595 | OCFS2_JOURNAL_ACCESS_WRITE); | |
596 | if (status < 0) { | |
597 | mlog_errno(status); | |
598 | goto leave; | |
599 | } | |
600 | ||
601 | prev_clusters = OCFS2_I(inode)->ip_clusters; | |
602 | ||
0eb8d47e TM |
603 | status = ocfs2_add_inode_data(osb, |
604 | inode, | |
605 | &logical_start, | |
606 | clusters_to_add, | |
607 | mark_unwritten, | |
608 | bh, | |
609 | handle, | |
610 | data_ac, | |
611 | meta_ac, | |
612 | &why); | |
ccd979bd MF |
613 | if ((status < 0) && (status != -EAGAIN)) { |
614 | if (status != -ENOSPC) | |
615 | mlog_errno(status); | |
616 | goto leave; | |
617 | } | |
618 | ||
619 | status = ocfs2_journal_dirty(handle, bh); | |
620 | if (status < 0) { | |
621 | mlog_errno(status); | |
622 | goto leave; | |
623 | } | |
624 | ||
625 | spin_lock(&OCFS2_I(inode)->ip_lock); | |
626 | clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters); | |
627 | spin_unlock(&OCFS2_I(inode)->ip_lock); | |
628 | ||
629 | if (why != RESTART_NONE && clusters_to_add) { | |
630 | if (why == RESTART_META) { | |
631 | mlog(0, "restarting function.\n"); | |
632 | restart_func = 1; | |
633 | } else { | |
634 | BUG_ON(why != RESTART_TRANS); | |
635 | ||
636 | mlog(0, "restarting transaction.\n"); | |
637 | /* TODO: This can be more intelligent. */ | |
638 | credits = ocfs2_calc_extend_credits(osb->sb, | |
811f933d | 639 | &fe->id2.i_list, |
ccd979bd | 640 | clusters_to_add); |
1fabe148 | 641 | status = ocfs2_extend_trans(handle, credits); |
ccd979bd MF |
642 | if (status < 0) { |
643 | /* handle still has to be committed at | |
644 | * this point. */ | |
645 | status = -ENOMEM; | |
646 | mlog_errno(status); | |
647 | goto leave; | |
648 | } | |
649 | goto restarted_transaction; | |
650 | } | |
651 | } | |
652 | ||
b0697053 | 653 | mlog(0, "fe: i_clusters = %u, i_size=%llu\n", |
1ca1a111 MF |
654 | le32_to_cpu(fe->i_clusters), |
655 | (unsigned long long)le64_to_cpu(fe->i_size)); | |
ccd979bd | 656 | mlog(0, "inode: ip_clusters=%u, i_size=%lld\n", |
634bf74d | 657 | OCFS2_I(inode)->ip_clusters, (long long)i_size_read(inode)); |
ccd979bd MF |
658 | |
659 | leave: | |
ccd979bd | 660 | if (handle) { |
02dc1af4 | 661 | ocfs2_commit_trans(osb, handle); |
ccd979bd MF |
662 | handle = NULL; |
663 | } | |
664 | if (data_ac) { | |
665 | ocfs2_free_alloc_context(data_ac); | |
666 | data_ac = NULL; | |
667 | } | |
668 | if (meta_ac) { | |
669 | ocfs2_free_alloc_context(meta_ac); | |
670 | meta_ac = NULL; | |
671 | } | |
672 | if ((!status) && restart_func) { | |
673 | restart_func = 0; | |
674 | goto restart_all; | |
675 | } | |
676 | if (bh) { | |
677 | brelse(bh); | |
678 | bh = NULL; | |
679 | } | |
680 | ||
681 | mlog_exit(status); | |
682 | return status; | |
683 | } | |
684 | ||
685 | /* Some parts of this taken from generic_cont_expand, which turned out | |
686 | * to be too fragile to do exactly what we need without us having to | |
53013cba MF |
687 | * worry about recursive locking in ->prepare_write() and |
688 | * ->commit_write(). */ | |
ccd979bd MF |
689 | static int ocfs2_write_zero_page(struct inode *inode, |
690 | u64 size) | |
691 | { | |
692 | struct address_space *mapping = inode->i_mapping; | |
693 | struct page *page; | |
694 | unsigned long index; | |
695 | unsigned int offset; | |
1fabe148 | 696 | handle_t *handle = NULL; |
ccd979bd MF |
697 | int ret; |
698 | ||
699 | offset = (size & (PAGE_CACHE_SIZE-1)); /* Within page */ | |
700 | /* ugh. in prepare/commit_write, if from==to==start of block, we | |
701 | ** skip the prepare. make sure we never send an offset for the start | |
702 | ** of a block | |
703 | */ | |
704 | if ((offset & (inode->i_sb->s_blocksize - 1)) == 0) { | |
705 | offset++; | |
706 | } | |
707 | index = size >> PAGE_CACHE_SHIFT; | |
708 | ||
709 | page = grab_cache_page(mapping, index); | |
710 | if (!page) { | |
711 | ret = -ENOMEM; | |
712 | mlog_errno(ret); | |
713 | goto out; | |
714 | } | |
715 | ||
53013cba | 716 | ret = ocfs2_prepare_write_nolock(inode, page, offset, offset); |
ccd979bd MF |
717 | if (ret < 0) { |
718 | mlog_errno(ret); | |
719 | goto out_unlock; | |
720 | } | |
721 | ||
722 | if (ocfs2_should_order_data(inode)) { | |
723 | handle = ocfs2_start_walk_page_trans(inode, page, offset, | |
724 | offset); | |
725 | if (IS_ERR(handle)) { | |
726 | ret = PTR_ERR(handle); | |
727 | handle = NULL; | |
728 | goto out_unlock; | |
729 | } | |
730 | } | |
731 | ||
732 | /* must not update i_size! */ | |
733 | ret = block_commit_write(page, offset, offset); | |
734 | if (ret < 0) | |
735 | mlog_errno(ret); | |
736 | else | |
737 | ret = 0; | |
738 | ||
739 | if (handle) | |
02dc1af4 | 740 | ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle); |
ccd979bd MF |
741 | out_unlock: |
742 | unlock_page(page); | |
743 | page_cache_release(page); | |
744 | out: | |
745 | return ret; | |
746 | } | |
747 | ||
748 | static int ocfs2_zero_extend(struct inode *inode, | |
749 | u64 zero_to_size) | |
750 | { | |
751 | int ret = 0; | |
752 | u64 start_off; | |
753 | struct super_block *sb = inode->i_sb; | |
754 | ||
755 | start_off = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode)); | |
756 | while (start_off < zero_to_size) { | |
757 | ret = ocfs2_write_zero_page(inode, start_off); | |
758 | if (ret < 0) { | |
759 | mlog_errno(ret); | |
760 | goto out; | |
761 | } | |
762 | ||
763 | start_off += sb->s_blocksize; | |
e2057c5a MF |
764 | |
765 | /* | |
766 | * Very large extends have the potential to lock up | |
767 | * the cpu for extended periods of time. | |
768 | */ | |
769 | cond_resched(); | |
ccd979bd MF |
770 | } |
771 | ||
772 | out: | |
773 | return ret; | |
774 | } | |
775 | ||
65ed39d6 MF |
776 | int ocfs2_extend_no_holes(struct inode *inode, u64 new_i_size, u64 zero_to) |
777 | { | |
778 | int ret; | |
779 | u32 clusters_to_add; | |
780 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
781 | ||
782 | clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size); | |
783 | if (clusters_to_add < oi->ip_clusters) | |
784 | clusters_to_add = 0; | |
785 | else | |
786 | clusters_to_add -= oi->ip_clusters; | |
787 | ||
788 | if (clusters_to_add) { | |
789 | ret = __ocfs2_extend_allocation(inode, oi->ip_clusters, | |
790 | clusters_to_add, 0); | |
791 | if (ret) { | |
792 | mlog_errno(ret); | |
793 | goto out; | |
794 | } | |
795 | } | |
796 | ||
797 | /* | |
798 | * Call this even if we don't add any clusters to the tree. We | |
799 | * still need to zero the area between the old i_size and the | |
800 | * new i_size. | |
801 | */ | |
802 | ret = ocfs2_zero_extend(inode, zero_to); | |
803 | if (ret < 0) | |
804 | mlog_errno(ret); | |
805 | ||
806 | out: | |
807 | return ret; | |
808 | } | |
809 | ||
ccd979bd MF |
810 | static int ocfs2_extend_file(struct inode *inode, |
811 | struct buffer_head *di_bh, | |
65ed39d6 | 812 | u64 new_i_size) |
ccd979bd | 813 | { |
c934a92d | 814 | int ret = 0; |
1afc32b9 | 815 | struct ocfs2_inode_info *oi = OCFS2_I(inode); |
ccd979bd | 816 | |
65ed39d6 | 817 | BUG_ON(!di_bh); |
53013cba | 818 | |
ccd979bd MF |
819 | /* setattr sometimes calls us like this. */ |
820 | if (new_i_size == 0) | |
821 | goto out; | |
822 | ||
823 | if (i_size_read(inode) == new_i_size) | |
824 | goto out; | |
825 | BUG_ON(new_i_size < i_size_read(inode)); | |
826 | ||
1afc32b9 MF |
827 | /* |
828 | * Fall through for converting inline data, even if the fs | |
829 | * supports sparse files. | |
830 | * | |
831 | * The check for inline data here is legal - nobody can add | |
832 | * the feature since we have i_mutex. We must check it again | |
833 | * after acquiring ip_alloc_sem though, as paths like mmap | |
834 | * might have raced us to converting the inode to extents. | |
835 | */ | |
836 | if (!(oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) | |
837 | && ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))) | |
3a0782d0 | 838 | goto out_update_size; |
ccd979bd | 839 | |
0effef77 | 840 | /* |
65ed39d6 MF |
841 | * The alloc sem blocks people in read/write from reading our |
842 | * allocation until we're done changing it. We depend on | |
843 | * i_mutex to block other extend/truncate calls while we're | |
844 | * here. | |
0effef77 | 845 | */ |
1afc32b9 MF |
846 | down_write(&oi->ip_alloc_sem); |
847 | ||
848 | if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) { | |
849 | /* | |
850 | * We can optimize small extends by keeping the inodes | |
851 | * inline data. | |
852 | */ | |
853 | if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) { | |
854 | up_write(&oi->ip_alloc_sem); | |
855 | goto out_update_size; | |
856 | } | |
857 | ||
858 | ret = ocfs2_convert_inline_data_to_extents(inode, di_bh); | |
859 | if (ret) { | |
860 | up_write(&oi->ip_alloc_sem); | |
861 | ||
862 | mlog_errno(ret); | |
c934a92d | 863 | goto out; |
1afc32b9 MF |
864 | } |
865 | } | |
866 | ||
867 | if (!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))) | |
868 | ret = ocfs2_extend_no_holes(inode, new_i_size, new_i_size); | |
869 | ||
870 | up_write(&oi->ip_alloc_sem); | |
65ed39d6 | 871 | |
0effef77 MF |
872 | if (ret < 0) { |
873 | mlog_errno(ret); | |
c934a92d | 874 | goto out; |
53013cba MF |
875 | } |
876 | ||
3a0782d0 | 877 | out_update_size: |
65ed39d6 MF |
878 | ret = ocfs2_simple_size_update(inode, di_bh, new_i_size); |
879 | if (ret < 0) | |
880 | mlog_errno(ret); | |
ccd979bd MF |
881 | |
882 | out: | |
883 | return ret; | |
884 | } | |
885 | ||
886 | int ocfs2_setattr(struct dentry *dentry, struct iattr *attr) | |
887 | { | |
888 | int status = 0, size_change; | |
889 | struct inode *inode = dentry->d_inode; | |
890 | struct super_block *sb = inode->i_sb; | |
891 | struct ocfs2_super *osb = OCFS2_SB(sb); | |
892 | struct buffer_head *bh = NULL; | |
1fabe148 | 893 | handle_t *handle = NULL; |
ccd979bd MF |
894 | |
895 | mlog_entry("(0x%p, '%.*s')\n", dentry, | |
896 | dentry->d_name.len, dentry->d_name.name); | |
897 | ||
bc535809 SM |
898 | /* ensuring we don't even attempt to truncate a symlink */ |
899 | if (S_ISLNK(inode->i_mode)) | |
900 | attr->ia_valid &= ~ATTR_SIZE; | |
901 | ||
ccd979bd MF |
902 | if (attr->ia_valid & ATTR_MODE) |
903 | mlog(0, "mode change: %d\n", attr->ia_mode); | |
904 | if (attr->ia_valid & ATTR_UID) | |
905 | mlog(0, "uid change: %d\n", attr->ia_uid); | |
906 | if (attr->ia_valid & ATTR_GID) | |
907 | mlog(0, "gid change: %d\n", attr->ia_gid); | |
908 | if (attr->ia_valid & ATTR_SIZE) | |
909 | mlog(0, "size change...\n"); | |
910 | if (attr->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME)) | |
911 | mlog(0, "time change...\n"); | |
912 | ||
913 | #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \ | |
914 | | ATTR_GID | ATTR_UID | ATTR_MODE) | |
915 | if (!(attr->ia_valid & OCFS2_VALID_ATTRS)) { | |
916 | mlog(0, "can't handle attrs: 0x%x\n", attr->ia_valid); | |
917 | return 0; | |
918 | } | |
919 | ||
920 | status = inode_change_ok(inode, attr); | |
921 | if (status) | |
922 | return status; | |
923 | ||
924 | size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE; | |
925 | if (size_change) { | |
926 | status = ocfs2_rw_lock(inode, 1); | |
927 | if (status < 0) { | |
928 | mlog_errno(status); | |
929 | goto bail; | |
930 | } | |
931 | } | |
932 | ||
e63aecb6 | 933 | status = ocfs2_inode_lock(inode, &bh, 1); |
ccd979bd MF |
934 | if (status < 0) { |
935 | if (status != -ENOENT) | |
936 | mlog_errno(status); | |
937 | goto bail_unlock_rw; | |
938 | } | |
939 | ||
940 | if (size_change && attr->ia_size != i_size_read(inode)) { | |
ce76fd30 MF |
941 | if (attr->ia_size > sb->s_maxbytes) { |
942 | status = -EFBIG; | |
943 | goto bail_unlock; | |
944 | } | |
945 | ||
ccd979bd MF |
946 | if (i_size_read(inode) > attr->ia_size) |
947 | status = ocfs2_truncate_file(inode, bh, attr->ia_size); | |
948 | else | |
65ed39d6 | 949 | status = ocfs2_extend_file(inode, bh, attr->ia_size); |
ccd979bd MF |
950 | if (status < 0) { |
951 | if (status != -ENOSPC) | |
952 | mlog_errno(status); | |
953 | status = -ENOSPC; | |
954 | goto bail_unlock; | |
955 | } | |
956 | } | |
957 | ||
65eff9cc | 958 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
ccd979bd MF |
959 | if (IS_ERR(handle)) { |
960 | status = PTR_ERR(handle); | |
961 | mlog_errno(status); | |
962 | goto bail_unlock; | |
963 | } | |
964 | ||
7307de80 MF |
965 | /* |
966 | * This will intentionally not wind up calling vmtruncate(), | |
967 | * since all the work for a size change has been done above. | |
968 | * Otherwise, we could get into problems with truncate as | |
969 | * ip_alloc_sem is used there to protect against i_size | |
970 | * changes. | |
971 | */ | |
ccd979bd MF |
972 | status = inode_setattr(inode, attr); |
973 | if (status < 0) { | |
974 | mlog_errno(status); | |
975 | goto bail_commit; | |
976 | } | |
977 | ||
978 | status = ocfs2_mark_inode_dirty(handle, inode, bh); | |
979 | if (status < 0) | |
980 | mlog_errno(status); | |
981 | ||
982 | bail_commit: | |
02dc1af4 | 983 | ocfs2_commit_trans(osb, handle); |
ccd979bd | 984 | bail_unlock: |
e63aecb6 | 985 | ocfs2_inode_unlock(inode, 1); |
ccd979bd MF |
986 | bail_unlock_rw: |
987 | if (size_change) | |
988 | ocfs2_rw_unlock(inode, 1); | |
989 | bail: | |
990 | if (bh) | |
991 | brelse(bh); | |
992 | ||
993 | mlog_exit(status); | |
994 | return status; | |
995 | } | |
996 | ||
997 | int ocfs2_getattr(struct vfsmount *mnt, | |
998 | struct dentry *dentry, | |
999 | struct kstat *stat) | |
1000 | { | |
1001 | struct inode *inode = dentry->d_inode; | |
1002 | struct super_block *sb = dentry->d_inode->i_sb; | |
1003 | struct ocfs2_super *osb = sb->s_fs_info; | |
1004 | int err; | |
1005 | ||
1006 | mlog_entry_void(); | |
1007 | ||
1008 | err = ocfs2_inode_revalidate(dentry); | |
1009 | if (err) { | |
1010 | if (err != -ENOENT) | |
1011 | mlog_errno(err); | |
1012 | goto bail; | |
1013 | } | |
1014 | ||
1015 | generic_fillattr(inode, stat); | |
1016 | ||
1017 | /* We set the blksize from the cluster size for performance */ | |
1018 | stat->blksize = osb->s_clustersize; | |
1019 | ||
1020 | bail: | |
1021 | mlog_exit(err); | |
1022 | ||
1023 | return err; | |
1024 | } | |
1025 | ||
e6305c43 | 1026 | int ocfs2_permission(struct inode *inode, int mask) |
d38eb8db TY |
1027 | { |
1028 | int ret; | |
1029 | ||
1030 | mlog_entry_void(); | |
1031 | ||
e63aecb6 | 1032 | ret = ocfs2_inode_lock(inode, NULL, 0); |
d38eb8db | 1033 | if (ret) { |
a9f5f707 MF |
1034 | if (ret != -ENOENT) |
1035 | mlog_errno(ret); | |
d38eb8db TY |
1036 | goto out; |
1037 | } | |
1038 | ||
1039 | ret = generic_permission(inode, mask, NULL); | |
d38eb8db | 1040 | |
e63aecb6 | 1041 | ocfs2_inode_unlock(inode, 0); |
d38eb8db TY |
1042 | out: |
1043 | mlog_exit(ret); | |
1044 | return ret; | |
1045 | } | |
1046 | ||
b2580103 MF |
1047 | static int __ocfs2_write_remove_suid(struct inode *inode, |
1048 | struct buffer_head *bh) | |
ccd979bd MF |
1049 | { |
1050 | int ret; | |
1fabe148 | 1051 | handle_t *handle; |
ccd979bd MF |
1052 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
1053 | struct ocfs2_dinode *di; | |
1054 | ||
b0697053 | 1055 | mlog_entry("(Inode %llu, mode 0%o)\n", |
b2580103 | 1056 | (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_mode); |
ccd979bd | 1057 | |
65eff9cc | 1058 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
ccd979bd MF |
1059 | if (handle == NULL) { |
1060 | ret = -ENOMEM; | |
1061 | mlog_errno(ret); | |
1062 | goto out; | |
1063 | } | |
1064 | ||
ccd979bd MF |
1065 | ret = ocfs2_journal_access(handle, inode, bh, |
1066 | OCFS2_JOURNAL_ACCESS_WRITE); | |
1067 | if (ret < 0) { | |
1068 | mlog_errno(ret); | |
b2580103 | 1069 | goto out_trans; |
ccd979bd MF |
1070 | } |
1071 | ||
1072 | inode->i_mode &= ~S_ISUID; | |
1073 | if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP)) | |
1074 | inode->i_mode &= ~S_ISGID; | |
1075 | ||
1076 | di = (struct ocfs2_dinode *) bh->b_data; | |
1077 | di->i_mode = cpu_to_le16(inode->i_mode); | |
1078 | ||
1079 | ret = ocfs2_journal_dirty(handle, bh); | |
1080 | if (ret < 0) | |
1081 | mlog_errno(ret); | |
b2580103 | 1082 | |
ccd979bd | 1083 | out_trans: |
02dc1af4 | 1084 | ocfs2_commit_trans(osb, handle); |
ccd979bd MF |
1085 | out: |
1086 | mlog_exit(ret); | |
1087 | return ret; | |
1088 | } | |
1089 | ||
9517bac6 MF |
1090 | /* |
1091 | * Will look for holes and unwritten extents in the range starting at | |
1092 | * pos for count bytes (inclusive). | |
1093 | */ | |
1094 | static int ocfs2_check_range_for_holes(struct inode *inode, loff_t pos, | |
1095 | size_t count) | |
1096 | { | |
1097 | int ret = 0; | |
49cb8d2d | 1098 | unsigned int extent_flags; |
9517bac6 MF |
1099 | u32 cpos, clusters, extent_len, phys_cpos; |
1100 | struct super_block *sb = inode->i_sb; | |
1101 | ||
1102 | cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits; | |
1103 | clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos; | |
1104 | ||
1105 | while (clusters) { | |
49cb8d2d MF |
1106 | ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len, |
1107 | &extent_flags); | |
9517bac6 MF |
1108 | if (ret < 0) { |
1109 | mlog_errno(ret); | |
1110 | goto out; | |
1111 | } | |
1112 | ||
49cb8d2d | 1113 | if (phys_cpos == 0 || (extent_flags & OCFS2_EXT_UNWRITTEN)) { |
9517bac6 MF |
1114 | ret = 1; |
1115 | break; | |
1116 | } | |
1117 | ||
1118 | if (extent_len > clusters) | |
1119 | extent_len = clusters; | |
1120 | ||
1121 | clusters -= extent_len; | |
1122 | cpos += extent_len; | |
1123 | } | |
1124 | out: | |
1125 | return ret; | |
1126 | } | |
1127 | ||
b2580103 MF |
1128 | static int ocfs2_write_remove_suid(struct inode *inode) |
1129 | { | |
1130 | int ret; | |
1131 | struct buffer_head *bh = NULL; | |
1132 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
1133 | ||
1134 | ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), | |
1135 | oi->ip_blkno, &bh, OCFS2_BH_CACHED, inode); | |
1136 | if (ret < 0) { | |
1137 | mlog_errno(ret); | |
1138 | goto out; | |
1139 | } | |
1140 | ||
1141 | ret = __ocfs2_write_remove_suid(inode, bh); | |
1142 | out: | |
1143 | brelse(bh); | |
1144 | return ret; | |
1145 | } | |
1146 | ||
2ae99a60 MF |
1147 | /* |
1148 | * Allocate enough extents to cover the region starting at byte offset | |
1149 | * start for len bytes. Existing extents are skipped, any extents | |
1150 | * added are marked as "unwritten". | |
1151 | */ | |
1152 | static int ocfs2_allocate_unwritten_extents(struct inode *inode, | |
1153 | u64 start, u64 len) | |
1154 | { | |
1155 | int ret; | |
1156 | u32 cpos, phys_cpos, clusters, alloc_size; | |
1afc32b9 MF |
1157 | u64 end = start + len; |
1158 | struct buffer_head *di_bh = NULL; | |
1159 | ||
1160 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) { | |
1161 | ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), | |
1162 | OCFS2_I(inode)->ip_blkno, &di_bh, | |
1163 | OCFS2_BH_CACHED, inode); | |
1164 | if (ret) { | |
1165 | mlog_errno(ret); | |
1166 | goto out; | |
1167 | } | |
1168 | ||
1169 | /* | |
1170 | * Nothing to do if the requested reservation range | |
1171 | * fits within the inode. | |
1172 | */ | |
1173 | if (ocfs2_size_fits_inline_data(di_bh, end)) | |
1174 | goto out; | |
1175 | ||
1176 | ret = ocfs2_convert_inline_data_to_extents(inode, di_bh); | |
1177 | if (ret) { | |
1178 | mlog_errno(ret); | |
1179 | goto out; | |
1180 | } | |
1181 | } | |
2ae99a60 MF |
1182 | |
1183 | /* | |
1184 | * We consider both start and len to be inclusive. | |
1185 | */ | |
1186 | cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits; | |
1187 | clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len); | |
1188 | clusters -= cpos; | |
1189 | ||
1190 | while (clusters) { | |
1191 | ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, | |
1192 | &alloc_size, NULL); | |
1193 | if (ret) { | |
1194 | mlog_errno(ret); | |
1195 | goto out; | |
1196 | } | |
1197 | ||
1198 | /* | |
1199 | * Hole or existing extent len can be arbitrary, so | |
1200 | * cap it to our own allocation request. | |
1201 | */ | |
1202 | if (alloc_size > clusters) | |
1203 | alloc_size = clusters; | |
1204 | ||
1205 | if (phys_cpos) { | |
1206 | /* | |
1207 | * We already have an allocation at this | |
1208 | * region so we can safely skip it. | |
1209 | */ | |
1210 | goto next; | |
1211 | } | |
1212 | ||
1213 | ret = __ocfs2_extend_allocation(inode, cpos, alloc_size, 1); | |
1214 | if (ret) { | |
1215 | if (ret != -ENOSPC) | |
1216 | mlog_errno(ret); | |
1217 | goto out; | |
1218 | } | |
1219 | ||
1220 | next: | |
1221 | cpos += alloc_size; | |
1222 | clusters -= alloc_size; | |
1223 | } | |
1224 | ||
1225 | ret = 0; | |
1226 | out: | |
1afc32b9 MF |
1227 | |
1228 | brelse(di_bh); | |
2ae99a60 MF |
1229 | return ret; |
1230 | } | |
1231 | ||
063c4561 MF |
1232 | static int __ocfs2_remove_inode_range(struct inode *inode, |
1233 | struct buffer_head *di_bh, | |
1234 | u32 cpos, u32 phys_cpos, u32 len, | |
1235 | struct ocfs2_cached_dealloc_ctxt *dealloc) | |
1236 | { | |
1237 | int ret; | |
1238 | u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos); | |
1239 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1240 | struct inode *tl_inode = osb->osb_tl_inode; | |
1241 | handle_t *handle; | |
1242 | struct ocfs2_alloc_context *meta_ac = NULL; | |
1243 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; | |
f99b9b7c JB |
1244 | struct ocfs2_extent_tree et; |
1245 | ||
1246 | ocfs2_get_dinode_extent_tree(&et, inode, di_bh); | |
063c4561 | 1247 | |
f99b9b7c | 1248 | ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac); |
063c4561 | 1249 | if (ret) { |
f99b9b7c | 1250 | ocfs2_put_extent_tree(&et); |
063c4561 MF |
1251 | mlog_errno(ret); |
1252 | return ret; | |
1253 | } | |
1254 | ||
1255 | mutex_lock(&tl_inode->i_mutex); | |
1256 | ||
1257 | if (ocfs2_truncate_log_needs_flush(osb)) { | |
1258 | ret = __ocfs2_flush_truncate_log(osb); | |
1259 | if (ret < 0) { | |
1260 | mlog_errno(ret); | |
1261 | goto out; | |
1262 | } | |
1263 | } | |
1264 | ||
1265 | handle = ocfs2_start_trans(osb, OCFS2_REMOVE_EXTENT_CREDITS); | |
1266 | if (handle == NULL) { | |
1267 | ret = -ENOMEM; | |
1268 | mlog_errno(ret); | |
1269 | goto out; | |
1270 | } | |
1271 | ||
1272 | ret = ocfs2_journal_access(handle, inode, di_bh, | |
1273 | OCFS2_JOURNAL_ACCESS_WRITE); | |
1274 | if (ret) { | |
1275 | mlog_errno(ret); | |
1276 | goto out; | |
1277 | } | |
1278 | ||
f99b9b7c JB |
1279 | ret = ocfs2_remove_extent(inode, &et, cpos, len, handle, meta_ac, |
1280 | dealloc); | |
063c4561 MF |
1281 | if (ret) { |
1282 | mlog_errno(ret); | |
1283 | goto out_commit; | |
1284 | } | |
1285 | ||
1286 | OCFS2_I(inode)->ip_clusters -= len; | |
1287 | di->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters); | |
1288 | ||
1289 | ret = ocfs2_journal_dirty(handle, di_bh); | |
1290 | if (ret) { | |
1291 | mlog_errno(ret); | |
1292 | goto out_commit; | |
1293 | } | |
1294 | ||
1295 | ret = ocfs2_truncate_log_append(osb, handle, phys_blkno, len); | |
1296 | if (ret) | |
1297 | mlog_errno(ret); | |
1298 | ||
1299 | out_commit: | |
1300 | ocfs2_commit_trans(osb, handle); | |
1301 | out: | |
1302 | mutex_unlock(&tl_inode->i_mutex); | |
1303 | ||
1304 | if (meta_ac) | |
1305 | ocfs2_free_alloc_context(meta_ac); | |
1306 | ||
f99b9b7c | 1307 | ocfs2_put_extent_tree(&et); |
063c4561 MF |
1308 | return ret; |
1309 | } | |
1310 | ||
1311 | /* | |
1312 | * Truncate a byte range, avoiding pages within partial clusters. This | |
1313 | * preserves those pages for the zeroing code to write to. | |
1314 | */ | |
1315 | static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start, | |
1316 | u64 byte_len) | |
1317 | { | |
1318 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1319 | loff_t start, end; | |
1320 | struct address_space *mapping = inode->i_mapping; | |
1321 | ||
1322 | start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start); | |
1323 | end = byte_start + byte_len; | |
1324 | end = end & ~(osb->s_clustersize - 1); | |
1325 | ||
1326 | if (start < end) { | |
1327 | unmap_mapping_range(mapping, start, end - start, 0); | |
1328 | truncate_inode_pages_range(mapping, start, end - 1); | |
1329 | } | |
1330 | } | |
1331 | ||
1332 | static int ocfs2_zero_partial_clusters(struct inode *inode, | |
1333 | u64 start, u64 len) | |
1334 | { | |
1335 | int ret = 0; | |
1336 | u64 tmpend, end = start + len; | |
1337 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1338 | unsigned int csize = osb->s_clustersize; | |
1339 | handle_t *handle; | |
1340 | ||
1341 | /* | |
1342 | * The "start" and "end" values are NOT necessarily part of | |
1343 | * the range whose allocation is being deleted. Rather, this | |
1344 | * is what the user passed in with the request. We must zero | |
1345 | * partial clusters here. There's no need to worry about | |
1346 | * physical allocation - the zeroing code knows to skip holes. | |
1347 | */ | |
1348 | mlog(0, "byte start: %llu, end: %llu\n", | |
1349 | (unsigned long long)start, (unsigned long long)end); | |
1350 | ||
1351 | /* | |
1352 | * If both edges are on a cluster boundary then there's no | |
1353 | * zeroing required as the region is part of the allocation to | |
1354 | * be truncated. | |
1355 | */ | |
1356 | if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0) | |
1357 | goto out; | |
1358 | ||
1359 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); | |
1360 | if (handle == NULL) { | |
1361 | ret = -ENOMEM; | |
1362 | mlog_errno(ret); | |
1363 | goto out; | |
1364 | } | |
1365 | ||
1366 | /* | |
1367 | * We want to get the byte offset of the end of the 1st cluster. | |
1368 | */ | |
1369 | tmpend = (u64)osb->s_clustersize + (start & ~(osb->s_clustersize - 1)); | |
1370 | if (tmpend > end) | |
1371 | tmpend = end; | |
1372 | ||
1373 | mlog(0, "1st range: start: %llu, tmpend: %llu\n", | |
1374 | (unsigned long long)start, (unsigned long long)tmpend); | |
1375 | ||
1376 | ret = ocfs2_zero_range_for_truncate(inode, handle, start, tmpend); | |
1377 | if (ret) | |
1378 | mlog_errno(ret); | |
1379 | ||
1380 | if (tmpend < end) { | |
1381 | /* | |
1382 | * This may make start and end equal, but the zeroing | |
1383 | * code will skip any work in that case so there's no | |
1384 | * need to catch it up here. | |
1385 | */ | |
1386 | start = end & ~(osb->s_clustersize - 1); | |
1387 | ||
1388 | mlog(0, "2nd range: start: %llu, end: %llu\n", | |
1389 | (unsigned long long)start, (unsigned long long)end); | |
1390 | ||
1391 | ret = ocfs2_zero_range_for_truncate(inode, handle, start, end); | |
1392 | if (ret) | |
1393 | mlog_errno(ret); | |
1394 | } | |
1395 | ||
1396 | ocfs2_commit_trans(osb, handle); | |
1397 | out: | |
1398 | return ret; | |
1399 | } | |
1400 | ||
1401 | static int ocfs2_remove_inode_range(struct inode *inode, | |
1402 | struct buffer_head *di_bh, u64 byte_start, | |
1403 | u64 byte_len) | |
1404 | { | |
1405 | int ret = 0; | |
1406 | u32 trunc_start, trunc_len, cpos, phys_cpos, alloc_size; | |
1407 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1408 | struct ocfs2_cached_dealloc_ctxt dealloc; | |
b1967d0e | 1409 | struct address_space *mapping = inode->i_mapping; |
063c4561 MF |
1410 | |
1411 | ocfs2_init_dealloc_ctxt(&dealloc); | |
1412 | ||
1413 | if (byte_len == 0) | |
1414 | return 0; | |
1415 | ||
1afc32b9 MF |
1416 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) { |
1417 | ret = ocfs2_truncate_inline(inode, di_bh, byte_start, | |
b1967d0e MF |
1418 | byte_start + byte_len, 0); |
1419 | if (ret) { | |
1afc32b9 | 1420 | mlog_errno(ret); |
b1967d0e MF |
1421 | goto out; |
1422 | } | |
1423 | /* | |
1424 | * There's no need to get fancy with the page cache | |
1425 | * truncate of an inline-data inode. We're talking | |
1426 | * about less than a page here, which will be cached | |
1427 | * in the dinode buffer anyway. | |
1428 | */ | |
1429 | unmap_mapping_range(mapping, 0, 0, 0); | |
1430 | truncate_inode_pages(mapping, 0); | |
1431 | goto out; | |
1afc32b9 MF |
1432 | } |
1433 | ||
063c4561 MF |
1434 | trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start); |
1435 | trunc_len = (byte_start + byte_len) >> osb->s_clustersize_bits; | |
1436 | if (trunc_len >= trunc_start) | |
1437 | trunc_len -= trunc_start; | |
1438 | else | |
1439 | trunc_len = 0; | |
1440 | ||
1441 | mlog(0, "Inode: %llu, start: %llu, len: %llu, cstart: %u, clen: %u\n", | |
1442 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
1443 | (unsigned long long)byte_start, | |
1444 | (unsigned long long)byte_len, trunc_start, trunc_len); | |
1445 | ||
1446 | ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len); | |
1447 | if (ret) { | |
1448 | mlog_errno(ret); | |
1449 | goto out; | |
1450 | } | |
1451 | ||
1452 | cpos = trunc_start; | |
1453 | while (trunc_len) { | |
1454 | ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, | |
1455 | &alloc_size, NULL); | |
1456 | if (ret) { | |
1457 | mlog_errno(ret); | |
1458 | goto out; | |
1459 | } | |
1460 | ||
1461 | if (alloc_size > trunc_len) | |
1462 | alloc_size = trunc_len; | |
1463 | ||
1464 | /* Only do work for non-holes */ | |
1465 | if (phys_cpos != 0) { | |
1466 | ret = __ocfs2_remove_inode_range(inode, di_bh, cpos, | |
1467 | phys_cpos, alloc_size, | |
1468 | &dealloc); | |
1469 | if (ret) { | |
1470 | mlog_errno(ret); | |
1471 | goto out; | |
1472 | } | |
1473 | } | |
1474 | ||
1475 | cpos += alloc_size; | |
1476 | trunc_len -= alloc_size; | |
1477 | } | |
1478 | ||
1479 | ocfs2_truncate_cluster_pages(inode, byte_start, byte_len); | |
1480 | ||
1481 | out: | |
1482 | ocfs2_schedule_truncate_log_flush(osb, 1); | |
1483 | ocfs2_run_deallocs(osb, &dealloc); | |
1484 | ||
1485 | return ret; | |
1486 | } | |
1487 | ||
b2580103 MF |
1488 | /* |
1489 | * Parts of this function taken from xfs_change_file_space() | |
1490 | */ | |
385820a3 MF |
1491 | static int __ocfs2_change_file_space(struct file *file, struct inode *inode, |
1492 | loff_t f_pos, unsigned int cmd, | |
1493 | struct ocfs2_space_resv *sr, | |
1494 | int change_size) | |
b2580103 MF |
1495 | { |
1496 | int ret; | |
1497 | s64 llen; | |
385820a3 | 1498 | loff_t size; |
b2580103 MF |
1499 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
1500 | struct buffer_head *di_bh = NULL; | |
1501 | handle_t *handle; | |
a00cce35 | 1502 | unsigned long long max_off = inode->i_sb->s_maxbytes; |
b2580103 | 1503 | |
b2580103 MF |
1504 | if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb)) |
1505 | return -EROFS; | |
1506 | ||
1507 | mutex_lock(&inode->i_mutex); | |
1508 | ||
1509 | /* | |
1510 | * This prevents concurrent writes on other nodes | |
1511 | */ | |
1512 | ret = ocfs2_rw_lock(inode, 1); | |
1513 | if (ret) { | |
1514 | mlog_errno(ret); | |
1515 | goto out; | |
1516 | } | |
1517 | ||
e63aecb6 | 1518 | ret = ocfs2_inode_lock(inode, &di_bh, 1); |
b2580103 MF |
1519 | if (ret) { |
1520 | mlog_errno(ret); | |
1521 | goto out_rw_unlock; | |
1522 | } | |
1523 | ||
1524 | if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) { | |
1525 | ret = -EPERM; | |
e63aecb6 | 1526 | goto out_inode_unlock; |
b2580103 MF |
1527 | } |
1528 | ||
1529 | switch (sr->l_whence) { | |
1530 | case 0: /*SEEK_SET*/ | |
1531 | break; | |
1532 | case 1: /*SEEK_CUR*/ | |
385820a3 | 1533 | sr->l_start += f_pos; |
b2580103 MF |
1534 | break; |
1535 | case 2: /*SEEK_END*/ | |
1536 | sr->l_start += i_size_read(inode); | |
1537 | break; | |
1538 | default: | |
1539 | ret = -EINVAL; | |
e63aecb6 | 1540 | goto out_inode_unlock; |
b2580103 MF |
1541 | } |
1542 | sr->l_whence = 0; | |
1543 | ||
1544 | llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len; | |
1545 | ||
1546 | if (sr->l_start < 0 | |
1547 | || sr->l_start > max_off | |
1548 | || (sr->l_start + llen) < 0 | |
1549 | || (sr->l_start + llen) > max_off) { | |
1550 | ret = -EINVAL; | |
e63aecb6 | 1551 | goto out_inode_unlock; |
b2580103 | 1552 | } |
385820a3 | 1553 | size = sr->l_start + sr->l_len; |
b2580103 MF |
1554 | |
1555 | if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) { | |
1556 | if (sr->l_len <= 0) { | |
1557 | ret = -EINVAL; | |
e63aecb6 | 1558 | goto out_inode_unlock; |
b2580103 MF |
1559 | } |
1560 | } | |
1561 | ||
385820a3 | 1562 | if (file && should_remove_suid(file->f_path.dentry)) { |
b2580103 MF |
1563 | ret = __ocfs2_write_remove_suid(inode, di_bh); |
1564 | if (ret) { | |
1565 | mlog_errno(ret); | |
e63aecb6 | 1566 | goto out_inode_unlock; |
b2580103 MF |
1567 | } |
1568 | } | |
1569 | ||
1570 | down_write(&OCFS2_I(inode)->ip_alloc_sem); | |
1571 | switch (cmd) { | |
1572 | case OCFS2_IOC_RESVSP: | |
1573 | case OCFS2_IOC_RESVSP64: | |
1574 | /* | |
1575 | * This takes unsigned offsets, but the signed ones we | |
1576 | * pass have been checked against overflow above. | |
1577 | */ | |
1578 | ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start, | |
1579 | sr->l_len); | |
1580 | break; | |
1581 | case OCFS2_IOC_UNRESVSP: | |
1582 | case OCFS2_IOC_UNRESVSP64: | |
1583 | ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start, | |
1584 | sr->l_len); | |
1585 | break; | |
1586 | default: | |
1587 | ret = -EINVAL; | |
1588 | } | |
1589 | up_write(&OCFS2_I(inode)->ip_alloc_sem); | |
1590 | if (ret) { | |
1591 | mlog_errno(ret); | |
e63aecb6 | 1592 | goto out_inode_unlock; |
b2580103 MF |
1593 | } |
1594 | ||
1595 | /* | |
1596 | * We update c/mtime for these changes | |
1597 | */ | |
1598 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); | |
1599 | if (IS_ERR(handle)) { | |
1600 | ret = PTR_ERR(handle); | |
1601 | mlog_errno(ret); | |
e63aecb6 | 1602 | goto out_inode_unlock; |
b2580103 MF |
1603 | } |
1604 | ||
385820a3 MF |
1605 | if (change_size && i_size_read(inode) < size) |
1606 | i_size_write(inode, size); | |
1607 | ||
b2580103 MF |
1608 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; |
1609 | ret = ocfs2_mark_inode_dirty(handle, inode, di_bh); | |
1610 | if (ret < 0) | |
1611 | mlog_errno(ret); | |
1612 | ||
1613 | ocfs2_commit_trans(osb, handle); | |
1614 | ||
e63aecb6 | 1615 | out_inode_unlock: |
b2580103 | 1616 | brelse(di_bh); |
e63aecb6 | 1617 | ocfs2_inode_unlock(inode, 1); |
b2580103 MF |
1618 | out_rw_unlock: |
1619 | ocfs2_rw_unlock(inode, 1); | |
1620 | ||
b2580103 | 1621 | out: |
c259ae52 | 1622 | mutex_unlock(&inode->i_mutex); |
b2580103 MF |
1623 | return ret; |
1624 | } | |
1625 | ||
385820a3 MF |
1626 | int ocfs2_change_file_space(struct file *file, unsigned int cmd, |
1627 | struct ocfs2_space_resv *sr) | |
1628 | { | |
1629 | struct inode *inode = file->f_path.dentry->d_inode; | |
1630 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);; | |
1631 | ||
1632 | if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) && | |
1633 | !ocfs2_writes_unwritten_extents(osb)) | |
1634 | return -ENOTTY; | |
1635 | else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) && | |
1636 | !ocfs2_sparse_alloc(osb)) | |
1637 | return -ENOTTY; | |
1638 | ||
1639 | if (!S_ISREG(inode->i_mode)) | |
1640 | return -EINVAL; | |
1641 | ||
1642 | if (!(file->f_mode & FMODE_WRITE)) | |
1643 | return -EBADF; | |
1644 | ||
1645 | return __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0); | |
1646 | } | |
1647 | ||
1648 | static long ocfs2_fallocate(struct inode *inode, int mode, loff_t offset, | |
1649 | loff_t len) | |
1650 | { | |
1651 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1652 | struct ocfs2_space_resv sr; | |
1653 | int change_size = 1; | |
1654 | ||
1655 | if (!ocfs2_writes_unwritten_extents(osb)) | |
1656 | return -EOPNOTSUPP; | |
1657 | ||
1658 | if (S_ISDIR(inode->i_mode)) | |
1659 | return -ENODEV; | |
1660 | ||
1661 | if (mode & FALLOC_FL_KEEP_SIZE) | |
1662 | change_size = 0; | |
1663 | ||
1664 | sr.l_whence = 0; | |
1665 | sr.l_start = (s64)offset; | |
1666 | sr.l_len = (s64)len; | |
1667 | ||
1668 | return __ocfs2_change_file_space(NULL, inode, offset, | |
1669 | OCFS2_IOC_RESVSP64, &sr, change_size); | |
1670 | } | |
1671 | ||
8659ac25 TY |
1672 | static int ocfs2_prepare_inode_for_write(struct dentry *dentry, |
1673 | loff_t *ppos, | |
1674 | size_t count, | |
9517bac6 MF |
1675 | int appending, |
1676 | int *direct_io) | |
ccd979bd | 1677 | { |
65ed39d6 | 1678 | int ret = 0, meta_level = 0; |
8659ac25 | 1679 | struct inode *inode = dentry->d_inode; |
65ed39d6 | 1680 | loff_t saved_pos, end; |
ccd979bd | 1681 | |
ccd979bd | 1682 | /* |
65ed39d6 MF |
1683 | * We start with a read level meta lock and only jump to an ex |
1684 | * if we need to make modifications here. | |
ccd979bd | 1685 | */ |
ccd979bd | 1686 | for(;;) { |
e63aecb6 | 1687 | ret = ocfs2_inode_lock(inode, NULL, meta_level); |
ccd979bd MF |
1688 | if (ret < 0) { |
1689 | meta_level = -1; | |
1690 | mlog_errno(ret); | |
1691 | goto out; | |
1692 | } | |
1693 | ||
1694 | /* Clear suid / sgid if necessary. We do this here | |
1695 | * instead of later in the write path because | |
1696 | * remove_suid() calls ->setattr without any hint that | |
1697 | * we may have already done our cluster locking. Since | |
1698 | * ocfs2_setattr() *must* take cluster locks to | |
1699 | * proceeed, this will lead us to recursively lock the | |
1700 | * inode. There's also the dinode i_size state which | |
1701 | * can be lost via setattr during extending writes (we | |
1702 | * set inode->i_size at the end of a write. */ | |
8659ac25 | 1703 | if (should_remove_suid(dentry)) { |
ccd979bd | 1704 | if (meta_level == 0) { |
e63aecb6 | 1705 | ocfs2_inode_unlock(inode, meta_level); |
ccd979bd MF |
1706 | meta_level = 1; |
1707 | continue; | |
1708 | } | |
1709 | ||
1710 | ret = ocfs2_write_remove_suid(inode); | |
1711 | if (ret < 0) { | |
1712 | mlog_errno(ret); | |
8659ac25 | 1713 | goto out_unlock; |
ccd979bd MF |
1714 | } |
1715 | } | |
1716 | ||
1717 | /* work on a copy of ppos until we're sure that we won't have | |
1718 | * to recalculate it due to relocking. */ | |
8659ac25 | 1719 | if (appending) { |
ccd979bd MF |
1720 | saved_pos = i_size_read(inode); |
1721 | mlog(0, "O_APPEND: inode->i_size=%llu\n", saved_pos); | |
1722 | } else { | |
8659ac25 | 1723 | saved_pos = *ppos; |
ccd979bd | 1724 | } |
3a0782d0 | 1725 | |
65ed39d6 | 1726 | end = saved_pos + count; |
9517bac6 | 1727 | |
65ed39d6 MF |
1728 | /* |
1729 | * Skip the O_DIRECT checks if we don't need | |
1730 | * them. | |
1731 | */ | |
1732 | if (!direct_io || !(*direct_io)) | |
9517bac6 | 1733 | break; |
9517bac6 | 1734 | |
1afc32b9 MF |
1735 | /* |
1736 | * There's no sane way to do direct writes to an inode | |
1737 | * with inline data. | |
1738 | */ | |
1739 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) { | |
1740 | *direct_io = 0; | |
1741 | break; | |
1742 | } | |
1743 | ||
3a0782d0 | 1744 | /* |
65ed39d6 MF |
1745 | * Allowing concurrent direct writes means |
1746 | * i_size changes wouldn't be synchronized, so | |
1747 | * one node could wind up truncating another | |
1748 | * nodes writes. | |
3a0782d0 | 1749 | */ |
65ed39d6 MF |
1750 | if (end > i_size_read(inode)) { |
1751 | *direct_io = 0; | |
ccd979bd | 1752 | break; |
ccd979bd MF |
1753 | } |
1754 | ||
65ed39d6 MF |
1755 | /* |
1756 | * We don't fill holes during direct io, so | |
1757 | * check for them here. If any are found, the | |
1758 | * caller will have to retake some cluster | |
1759 | * locks and initiate the io as buffered. | |
1760 | */ | |
1761 | ret = ocfs2_check_range_for_holes(inode, saved_pos, count); | |
1762 | if (ret == 1) { | |
1763 | *direct_io = 0; | |
1764 | ret = 0; | |
1765 | } else if (ret < 0) | |
1766 | mlog_errno(ret); | |
ccd979bd MF |
1767 | break; |
1768 | } | |
1769 | ||
8659ac25 TY |
1770 | if (appending) |
1771 | *ppos = saved_pos; | |
1772 | ||
1773 | out_unlock: | |
e63aecb6 | 1774 | ocfs2_inode_unlock(inode, meta_level); |
8659ac25 TY |
1775 | |
1776 | out: | |
1777 | return ret; | |
1778 | } | |
1779 | ||
1780 | static ssize_t ocfs2_file_aio_write(struct kiocb *iocb, | |
1781 | const struct iovec *iov, | |
1782 | unsigned long nr_segs, | |
1783 | loff_t pos) | |
1784 | { | |
9517bac6 | 1785 | int ret, direct_io, appending, rw_level, have_alloc_sem = 0; |
b6af1bcd | 1786 | int can_do_direct; |
9517bac6 MF |
1787 | ssize_t written = 0; |
1788 | size_t ocount; /* original count */ | |
1789 | size_t count; /* after file limit checks */ | |
9ea2d32f MF |
1790 | loff_t old_size, *ppos = &iocb->ki_pos; |
1791 | u32 old_clusters; | |
9517bac6 MF |
1792 | struct file *file = iocb->ki_filp; |
1793 | struct inode *inode = file->f_path.dentry->d_inode; | |
9ea2d32f | 1794 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
9517bac6 MF |
1795 | |
1796 | mlog_entry("(0x%p, %u, '%.*s')\n", file, | |
8659ac25 | 1797 | (unsigned int)nr_segs, |
9517bac6 MF |
1798 | file->f_path.dentry->d_name.len, |
1799 | file->f_path.dentry->d_name.name); | |
8659ac25 | 1800 | |
8659ac25 TY |
1801 | if (iocb->ki_left == 0) |
1802 | return 0; | |
1803 | ||
9517bac6 MF |
1804 | vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE); |
1805 | ||
1806 | appending = file->f_flags & O_APPEND ? 1 : 0; | |
1807 | direct_io = file->f_flags & O_DIRECT ? 1 : 0; | |
1808 | ||
8659ac25 | 1809 | mutex_lock(&inode->i_mutex); |
9517bac6 MF |
1810 | |
1811 | relock: | |
8659ac25 | 1812 | /* to match setattr's i_mutex -> i_alloc_sem -> rw_lock ordering */ |
9517bac6 | 1813 | if (direct_io) { |
8659ac25 | 1814 | down_read(&inode->i_alloc_sem); |
9517bac6 | 1815 | have_alloc_sem = 1; |
8659ac25 TY |
1816 | } |
1817 | ||
1818 | /* concurrent O_DIRECT writes are allowed */ | |
9517bac6 | 1819 | rw_level = !direct_io; |
8659ac25 TY |
1820 | ret = ocfs2_rw_lock(inode, rw_level); |
1821 | if (ret < 0) { | |
8659ac25 | 1822 | mlog_errno(ret); |
9517bac6 | 1823 | goto out_sems; |
8659ac25 TY |
1824 | } |
1825 | ||
9517bac6 MF |
1826 | can_do_direct = direct_io; |
1827 | ret = ocfs2_prepare_inode_for_write(file->f_path.dentry, ppos, | |
1828 | iocb->ki_left, appending, | |
1829 | &can_do_direct); | |
8659ac25 TY |
1830 | if (ret < 0) { |
1831 | mlog_errno(ret); | |
1832 | goto out; | |
1833 | } | |
ccd979bd | 1834 | |
9517bac6 MF |
1835 | /* |
1836 | * We can't complete the direct I/O as requested, fall back to | |
1837 | * buffered I/O. | |
1838 | */ | |
1839 | if (direct_io && !can_do_direct) { | |
1840 | ocfs2_rw_unlock(inode, rw_level); | |
1841 | up_read(&inode->i_alloc_sem); | |
1842 | ||
1843 | have_alloc_sem = 0; | |
1844 | rw_level = -1; | |
1845 | ||
1846 | direct_io = 0; | |
9517bac6 MF |
1847 | goto relock; |
1848 | } | |
1849 | ||
9ea2d32f MF |
1850 | /* |
1851 | * To later detect whether a journal commit for sync writes is | |
1852 | * necessary, we sample i_size, and cluster count here. | |
1853 | */ | |
1854 | old_size = i_size_read(inode); | |
1855 | old_clusters = OCFS2_I(inode)->ip_clusters; | |
1856 | ||
ccd979bd | 1857 | /* communicate with ocfs2_dio_end_io */ |
7cdfc3a1 | 1858 | ocfs2_iocb_set_rw_locked(iocb, rw_level); |
ccd979bd | 1859 | |
9517bac6 | 1860 | if (direct_io) { |
b6af1bcd NP |
1861 | ret = generic_segment_checks(iov, &nr_segs, &ocount, |
1862 | VERIFY_READ); | |
1863 | if (ret) | |
1864 | goto out_dio; | |
1865 | ||
1866 | ret = generic_write_checks(file, ppos, &count, | |
1867 | S_ISBLK(inode->i_mode)); | |
1868 | if (ret) | |
1869 | goto out_dio; | |
1870 | ||
9517bac6 MF |
1871 | written = generic_file_direct_write(iocb, iov, &nr_segs, *ppos, |
1872 | ppos, count, ocount); | |
1873 | if (written < 0) { | |
1874 | ret = written; | |
1875 | goto out_dio; | |
1876 | } | |
1877 | } else { | |
b6af1bcd NP |
1878 | written = generic_file_aio_write_nolock(iocb, iov, nr_segs, |
1879 | *ppos); | |
9517bac6 | 1880 | } |
ccd979bd | 1881 | |
9517bac6 | 1882 | out_dio: |
ccd979bd | 1883 | /* buffered aio wouldn't have proper lock coverage today */ |
9517bac6 | 1884 | BUG_ON(ret == -EIOCBQUEUED && !(file->f_flags & O_DIRECT)); |
ccd979bd | 1885 | |
9ea2d32f MF |
1886 | if ((file->f_flags & O_SYNC && !direct_io) || IS_SYNC(inode)) { |
1887 | /* | |
1888 | * The generic write paths have handled getting data | |
1889 | * to disk, but since we don't make use of the dirty | |
1890 | * inode list, a manual journal commit is necessary | |
1891 | * here. | |
1892 | */ | |
1893 | if (old_size != i_size_read(inode) || | |
1894 | old_clusters != OCFS2_I(inode)->ip_clusters) { | |
1895 | ret = journal_force_commit(osb->journal->j_journal); | |
1896 | if (ret < 0) | |
1897 | written = ret; | |
1898 | } | |
1899 | } | |
1900 | ||
ccd979bd MF |
1901 | /* |
1902 | * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io | |
1903 | * function pointer which is called when o_direct io completes so that | |
1904 | * it can unlock our rw lock. (it's the clustered equivalent of | |
1905 | * i_alloc_sem; protects truncate from racing with pending ios). | |
1906 | * Unfortunately there are error cases which call end_io and others | |
1907 | * that don't. so we don't have to unlock the rw_lock if either an | |
1908 | * async dio is going to do it in the future or an end_io after an | |
1909 | * error has already done it. | |
1910 | */ | |
1911 | if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) { | |
1912 | rw_level = -1; | |
1913 | have_alloc_sem = 0; | |
1914 | } | |
1915 | ||
1916 | out: | |
9517bac6 MF |
1917 | if (rw_level != -1) |
1918 | ocfs2_rw_unlock(inode, rw_level); | |
1919 | ||
1920 | out_sems: | |
ccd979bd MF |
1921 | if (have_alloc_sem) |
1922 | up_read(&inode->i_alloc_sem); | |
9517bac6 | 1923 | |
1b1dcc1b | 1924 | mutex_unlock(&inode->i_mutex); |
ccd979bd MF |
1925 | |
1926 | mlog_exit(ret); | |
9517bac6 | 1927 | return written ? written : ret; |
ccd979bd MF |
1928 | } |
1929 | ||
8659ac25 TY |
1930 | static ssize_t ocfs2_file_splice_write(struct pipe_inode_info *pipe, |
1931 | struct file *out, | |
1932 | loff_t *ppos, | |
1933 | size_t len, | |
1934 | unsigned int flags) | |
1935 | { | |
1936 | int ret; | |
d28c9174 | 1937 | struct inode *inode = out->f_path.dentry->d_inode; |
8659ac25 TY |
1938 | |
1939 | mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", out, pipe, | |
1940 | (unsigned int)len, | |
d28c9174 JS |
1941 | out->f_path.dentry->d_name.len, |
1942 | out->f_path.dentry->d_name.name); | |
8659ac25 TY |
1943 | |
1944 | inode_double_lock(inode, pipe->inode); | |
1945 | ||
1946 | ret = ocfs2_rw_lock(inode, 1); | |
1947 | if (ret < 0) { | |
1948 | mlog_errno(ret); | |
1949 | goto out; | |
1950 | } | |
1951 | ||
9517bac6 MF |
1952 | ret = ocfs2_prepare_inode_for_write(out->f_path.dentry, ppos, len, 0, |
1953 | NULL); | |
8659ac25 TY |
1954 | if (ret < 0) { |
1955 | mlog_errno(ret); | |
1956 | goto out_unlock; | |
1957 | } | |
1958 | ||
b6af1bcd | 1959 | ret = generic_file_splice_write_nolock(pipe, out, ppos, len, flags); |
8659ac25 TY |
1960 | |
1961 | out_unlock: | |
1962 | ocfs2_rw_unlock(inode, 1); | |
1963 | out: | |
1964 | inode_double_unlock(inode, pipe->inode); | |
1965 | ||
1966 | mlog_exit(ret); | |
1967 | return ret; | |
1968 | } | |
1969 | ||
1970 | static ssize_t ocfs2_file_splice_read(struct file *in, | |
1971 | loff_t *ppos, | |
1972 | struct pipe_inode_info *pipe, | |
1973 | size_t len, | |
1974 | unsigned int flags) | |
1975 | { | |
1976 | int ret = 0; | |
d28c9174 | 1977 | struct inode *inode = in->f_path.dentry->d_inode; |
8659ac25 TY |
1978 | |
1979 | mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", in, pipe, | |
1980 | (unsigned int)len, | |
d28c9174 JS |
1981 | in->f_path.dentry->d_name.len, |
1982 | in->f_path.dentry->d_name.name); | |
8659ac25 TY |
1983 | |
1984 | /* | |
1985 | * See the comment in ocfs2_file_aio_read() | |
1986 | */ | |
e63aecb6 | 1987 | ret = ocfs2_inode_lock(inode, NULL, 0); |
8659ac25 TY |
1988 | if (ret < 0) { |
1989 | mlog_errno(ret); | |
1990 | goto bail; | |
1991 | } | |
e63aecb6 | 1992 | ocfs2_inode_unlock(inode, 0); |
8659ac25 TY |
1993 | |
1994 | ret = generic_file_splice_read(in, ppos, pipe, len, flags); | |
1995 | ||
1996 | bail: | |
1997 | mlog_exit(ret); | |
1998 | return ret; | |
1999 | } | |
2000 | ||
ccd979bd | 2001 | static ssize_t ocfs2_file_aio_read(struct kiocb *iocb, |
027445c3 BP |
2002 | const struct iovec *iov, |
2003 | unsigned long nr_segs, | |
ccd979bd MF |
2004 | loff_t pos) |
2005 | { | |
25899dee | 2006 | int ret = 0, rw_level = -1, have_alloc_sem = 0, lock_level = 0; |
ccd979bd | 2007 | struct file *filp = iocb->ki_filp; |
d28c9174 | 2008 | struct inode *inode = filp->f_path.dentry->d_inode; |
ccd979bd | 2009 | |
027445c3 BP |
2010 | mlog_entry("(0x%p, %u, '%.*s')\n", filp, |
2011 | (unsigned int)nr_segs, | |
d28c9174 JS |
2012 | filp->f_path.dentry->d_name.len, |
2013 | filp->f_path.dentry->d_name.name); | |
ccd979bd MF |
2014 | |
2015 | if (!inode) { | |
2016 | ret = -EINVAL; | |
2017 | mlog_errno(ret); | |
2018 | goto bail; | |
2019 | } | |
2020 | ||
ccd979bd MF |
2021 | /* |
2022 | * buffered reads protect themselves in ->readpage(). O_DIRECT reads | |
2023 | * need locks to protect pending reads from racing with truncate. | |
2024 | */ | |
2025 | if (filp->f_flags & O_DIRECT) { | |
2026 | down_read(&inode->i_alloc_sem); | |
2027 | have_alloc_sem = 1; | |
2028 | ||
2029 | ret = ocfs2_rw_lock(inode, 0); | |
2030 | if (ret < 0) { | |
2031 | mlog_errno(ret); | |
2032 | goto bail; | |
2033 | } | |
2034 | rw_level = 0; | |
2035 | /* communicate with ocfs2_dio_end_io */ | |
7cdfc3a1 | 2036 | ocfs2_iocb_set_rw_locked(iocb, rw_level); |
ccd979bd MF |
2037 | } |
2038 | ||
c4374f8a MF |
2039 | /* |
2040 | * We're fine letting folks race truncates and extending | |
2041 | * writes with read across the cluster, just like they can | |
2042 | * locally. Hence no rw_lock during read. | |
2043 | * | |
2044 | * Take and drop the meta data lock to update inode fields | |
2045 | * like i_size. This allows the checks down below | |
2046 | * generic_file_aio_read() a chance of actually working. | |
2047 | */ | |
e63aecb6 | 2048 | ret = ocfs2_inode_lock_atime(inode, filp->f_vfsmnt, &lock_level); |
c4374f8a MF |
2049 | if (ret < 0) { |
2050 | mlog_errno(ret); | |
2051 | goto bail; | |
2052 | } | |
e63aecb6 | 2053 | ocfs2_inode_unlock(inode, lock_level); |
c4374f8a | 2054 | |
027445c3 | 2055 | ret = generic_file_aio_read(iocb, iov, nr_segs, iocb->ki_pos); |
ccd979bd | 2056 | if (ret == -EINVAL) |
56753bd3 | 2057 | mlog(0, "generic_file_aio_read returned -EINVAL\n"); |
ccd979bd MF |
2058 | |
2059 | /* buffered aio wouldn't have proper lock coverage today */ | |
2060 | BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT)); | |
2061 | ||
2062 | /* see ocfs2_file_aio_write */ | |
2063 | if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) { | |
2064 | rw_level = -1; | |
2065 | have_alloc_sem = 0; | |
2066 | } | |
2067 | ||
2068 | bail: | |
2069 | if (have_alloc_sem) | |
2070 | up_read(&inode->i_alloc_sem); | |
2071 | if (rw_level != -1) | |
2072 | ocfs2_rw_unlock(inode, rw_level); | |
2073 | mlog_exit(ret); | |
2074 | ||
2075 | return ret; | |
2076 | } | |
2077 | ||
92e1d5be | 2078 | const struct inode_operations ocfs2_file_iops = { |
ccd979bd MF |
2079 | .setattr = ocfs2_setattr, |
2080 | .getattr = ocfs2_getattr, | |
d38eb8db | 2081 | .permission = ocfs2_permission, |
cf1d6c76 TY |
2082 | .setxattr = generic_setxattr, |
2083 | .getxattr = generic_getxattr, | |
2084 | .listxattr = ocfs2_listxattr, | |
2085 | .removexattr = generic_removexattr, | |
385820a3 | 2086 | .fallocate = ocfs2_fallocate, |
00dc417f | 2087 | .fiemap = ocfs2_fiemap, |
ccd979bd MF |
2088 | }; |
2089 | ||
92e1d5be | 2090 | const struct inode_operations ocfs2_special_file_iops = { |
ccd979bd MF |
2091 | .setattr = ocfs2_setattr, |
2092 | .getattr = ocfs2_getattr, | |
d38eb8db | 2093 | .permission = ocfs2_permission, |
ccd979bd MF |
2094 | }; |
2095 | ||
53da4939 MF |
2096 | /* |
2097 | * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with | |
2098 | * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks! | |
2099 | */ | |
4b6f5d20 | 2100 | const struct file_operations ocfs2_fops = { |
32c3c0e2 | 2101 | .llseek = generic_file_llseek, |
ccd979bd MF |
2102 | .read = do_sync_read, |
2103 | .write = do_sync_write, | |
ccd979bd MF |
2104 | .mmap = ocfs2_mmap, |
2105 | .fsync = ocfs2_sync_file, | |
2106 | .release = ocfs2_file_release, | |
2107 | .open = ocfs2_file_open, | |
2108 | .aio_read = ocfs2_file_aio_read, | |
2109 | .aio_write = ocfs2_file_aio_write, | |
c9ec1488 | 2110 | .unlocked_ioctl = ocfs2_ioctl, |
586d232b MF |
2111 | #ifdef CONFIG_COMPAT |
2112 | .compat_ioctl = ocfs2_compat_ioctl, | |
2113 | #endif | |
53da4939 | 2114 | .lock = ocfs2_lock, |
53fc622b | 2115 | .flock = ocfs2_flock, |
8659ac25 TY |
2116 | .splice_read = ocfs2_file_splice_read, |
2117 | .splice_write = ocfs2_file_splice_write, | |
ccd979bd MF |
2118 | }; |
2119 | ||
4b6f5d20 | 2120 | const struct file_operations ocfs2_dops = { |
32c3c0e2 | 2121 | .llseek = generic_file_llseek, |
ccd979bd MF |
2122 | .read = generic_read_dir, |
2123 | .readdir = ocfs2_readdir, | |
2124 | .fsync = ocfs2_sync_file, | |
53fc622b MF |
2125 | .release = ocfs2_dir_release, |
2126 | .open = ocfs2_dir_open, | |
c9ec1488 | 2127 | .unlocked_ioctl = ocfs2_ioctl, |
586d232b MF |
2128 | #ifdef CONFIG_COMPAT |
2129 | .compat_ioctl = ocfs2_compat_ioctl, | |
53da4939 MF |
2130 | #endif |
2131 | .lock = ocfs2_lock, | |
2132 | .flock = ocfs2_flock, | |
2133 | }; | |
2134 | ||
2135 | /* | |
2136 | * POSIX-lockless variants of our file_operations. | |
2137 | * | |
2138 | * These will be used if the underlying cluster stack does not support | |
2139 | * posix file locking, if the user passes the "localflocks" mount | |
2140 | * option, or if we have a local-only fs. | |
2141 | * | |
2142 | * ocfs2_flock is in here because all stacks handle UNIX file locks, | |
2143 | * so we still want it in the case of no stack support for | |
2144 | * plocks. Internally, it will do the right thing when asked to ignore | |
2145 | * the cluster. | |
2146 | */ | |
2147 | const struct file_operations ocfs2_fops_no_plocks = { | |
2148 | .llseek = generic_file_llseek, | |
2149 | .read = do_sync_read, | |
2150 | .write = do_sync_write, | |
2151 | .mmap = ocfs2_mmap, | |
2152 | .fsync = ocfs2_sync_file, | |
2153 | .release = ocfs2_file_release, | |
2154 | .open = ocfs2_file_open, | |
2155 | .aio_read = ocfs2_file_aio_read, | |
2156 | .aio_write = ocfs2_file_aio_write, | |
2157 | .unlocked_ioctl = ocfs2_ioctl, | |
2158 | #ifdef CONFIG_COMPAT | |
2159 | .compat_ioctl = ocfs2_compat_ioctl, | |
2160 | #endif | |
2161 | .flock = ocfs2_flock, | |
2162 | .splice_read = ocfs2_file_splice_read, | |
2163 | .splice_write = ocfs2_file_splice_write, | |
2164 | }; | |
2165 | ||
2166 | const struct file_operations ocfs2_dops_no_plocks = { | |
2167 | .llseek = generic_file_llseek, | |
2168 | .read = generic_read_dir, | |
2169 | .readdir = ocfs2_readdir, | |
2170 | .fsync = ocfs2_sync_file, | |
2171 | .release = ocfs2_dir_release, | |
2172 | .open = ocfs2_dir_open, | |
2173 | .unlocked_ioctl = ocfs2_ioctl, | |
2174 | #ifdef CONFIG_COMPAT | |
2175 | .compat_ioctl = ocfs2_compat_ioctl, | |
586d232b | 2176 | #endif |
53fc622b | 2177 | .flock = ocfs2_flock, |
ccd979bd | 2178 | }; |