]>
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> |
a90714c1 | 38 | #include <linux/quotaops.h> |
ccd979bd MF |
39 | |
40 | #define MLOG_MASK_PREFIX ML_INODE | |
41 | #include <cluster/masklog.h> | |
42 | ||
43 | #include "ocfs2.h" | |
44 | ||
45 | #include "alloc.h" | |
46 | #include "aops.h" | |
47 | #include "dir.h" | |
48 | #include "dlmglue.h" | |
49 | #include "extent_map.h" | |
50 | #include "file.h" | |
51 | #include "sysfile.h" | |
52 | #include "inode.h" | |
ca4d147e | 53 | #include "ioctl.h" |
ccd979bd | 54 | #include "journal.h" |
53fc622b | 55 | #include "locks.h" |
ccd979bd MF |
56 | #include "mmap.h" |
57 | #include "suballoc.h" | |
58 | #include "super.h" | |
cf1d6c76 | 59 | #include "xattr.h" |
23fc2702 | 60 | #include "acl.h" |
a90714c1 | 61 | #include "quota.h" |
293b2f70 | 62 | #include "refcounttree.h" |
ccd979bd MF |
63 | |
64 | #include "buffer_head_io.h" | |
65 | ||
66 | static int ocfs2_sync_inode(struct inode *inode) | |
67 | { | |
68 | filemap_fdatawrite(inode->i_mapping); | |
69 | return sync_mapping_buffers(inode->i_mapping); | |
70 | } | |
71 | ||
53fc622b MF |
72 | static int ocfs2_init_file_private(struct inode *inode, struct file *file) |
73 | { | |
74 | struct ocfs2_file_private *fp; | |
75 | ||
76 | fp = kzalloc(sizeof(struct ocfs2_file_private), GFP_KERNEL); | |
77 | if (!fp) | |
78 | return -ENOMEM; | |
79 | ||
80 | fp->fp_file = file; | |
81 | mutex_init(&fp->fp_mutex); | |
82 | ocfs2_file_lock_res_init(&fp->fp_flock, fp); | |
83 | file->private_data = fp; | |
84 | ||
85 | return 0; | |
86 | } | |
87 | ||
88 | static void ocfs2_free_file_private(struct inode *inode, struct file *file) | |
89 | { | |
90 | struct ocfs2_file_private *fp = file->private_data; | |
91 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
92 | ||
93 | if (fp) { | |
94 | ocfs2_simple_drop_lockres(osb, &fp->fp_flock); | |
95 | ocfs2_lock_res_free(&fp->fp_flock); | |
96 | kfree(fp); | |
97 | file->private_data = NULL; | |
98 | } | |
99 | } | |
100 | ||
ccd979bd MF |
101 | static int ocfs2_file_open(struct inode *inode, struct file *file) |
102 | { | |
103 | int status; | |
104 | int mode = file->f_flags; | |
105 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
106 | ||
107 | mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file, | |
d28c9174 | 108 | file->f_path.dentry->d_name.len, file->f_path.dentry->d_name.name); |
ccd979bd | 109 | |
907f4554 | 110 | if (file->f_mode & FMODE_WRITE) |
871a2931 | 111 | dquot_initialize(inode); |
907f4554 | 112 | |
ccd979bd MF |
113 | spin_lock(&oi->ip_lock); |
114 | ||
115 | /* Check that the inode hasn't been wiped from disk by another | |
116 | * node. If it hasn't then we're safe as long as we hold the | |
117 | * spin lock until our increment of open count. */ | |
118 | if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) { | |
119 | spin_unlock(&oi->ip_lock); | |
120 | ||
121 | status = -ENOENT; | |
122 | goto leave; | |
123 | } | |
124 | ||
125 | if (mode & O_DIRECT) | |
126 | oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT; | |
127 | ||
128 | oi->ip_open_count++; | |
129 | spin_unlock(&oi->ip_lock); | |
53fc622b MF |
130 | |
131 | status = ocfs2_init_file_private(inode, file); | |
132 | if (status) { | |
133 | /* | |
134 | * We want to set open count back if we're failing the | |
135 | * open. | |
136 | */ | |
137 | spin_lock(&oi->ip_lock); | |
138 | oi->ip_open_count--; | |
139 | spin_unlock(&oi->ip_lock); | |
140 | } | |
141 | ||
ccd979bd MF |
142 | leave: |
143 | mlog_exit(status); | |
144 | return status; | |
145 | } | |
146 | ||
147 | static int ocfs2_file_release(struct inode *inode, struct file *file) | |
148 | { | |
149 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
150 | ||
151 | mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file, | |
d28c9174 JS |
152 | file->f_path.dentry->d_name.len, |
153 | file->f_path.dentry->d_name.name); | |
ccd979bd MF |
154 | |
155 | spin_lock(&oi->ip_lock); | |
156 | if (!--oi->ip_open_count) | |
157 | oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT; | |
158 | spin_unlock(&oi->ip_lock); | |
159 | ||
53fc622b MF |
160 | ocfs2_free_file_private(inode, file); |
161 | ||
ccd979bd MF |
162 | mlog_exit(0); |
163 | ||
164 | return 0; | |
165 | } | |
166 | ||
53fc622b MF |
167 | static int ocfs2_dir_open(struct inode *inode, struct file *file) |
168 | { | |
169 | return ocfs2_init_file_private(inode, file); | |
170 | } | |
171 | ||
172 | static int ocfs2_dir_release(struct inode *inode, struct file *file) | |
173 | { | |
174 | ocfs2_free_file_private(inode, file); | |
175 | return 0; | |
176 | } | |
177 | ||
7ea80859 | 178 | static int ocfs2_sync_file(struct file *file, int datasync) |
ccd979bd MF |
179 | { |
180 | int err = 0; | |
181 | journal_t *journal; | |
7ea80859 CH |
182 | struct dentry *dentry = file->f_path.dentry; |
183 | struct inode *inode = file->f_mapping->host; | |
ccd979bd MF |
184 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
185 | ||
186 | mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", file, dentry, datasync, | |
187 | dentry->d_name.len, dentry->d_name.name); | |
188 | ||
189 | err = ocfs2_sync_inode(dentry->d_inode); | |
190 | if (err) | |
191 | goto bail; | |
192 | ||
e04cc15f HH |
193 | if (datasync && !(inode->i_state & I_DIRTY_DATASYNC)) |
194 | goto bail; | |
195 | ||
ccd979bd | 196 | journal = osb->journal->j_journal; |
2b4e30fb | 197 | err = jbd2_journal_force_commit(journal); |
ccd979bd MF |
198 | |
199 | bail: | |
200 | mlog_exit(err); | |
201 | ||
202 | return (err < 0) ? -EIO : 0; | |
203 | } | |
204 | ||
7f1a37e3 TY |
205 | int ocfs2_should_update_atime(struct inode *inode, |
206 | struct vfsmount *vfsmnt) | |
207 | { | |
208 | struct timespec now; | |
209 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
210 | ||
211 | if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb)) | |
212 | return 0; | |
213 | ||
214 | if ((inode->i_flags & S_NOATIME) || | |
215 | ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))) | |
216 | return 0; | |
217 | ||
6c2aad05 MF |
218 | /* |
219 | * We can be called with no vfsmnt structure - NFSD will | |
220 | * sometimes do this. | |
221 | * | |
222 | * Note that our action here is different than touch_atime() - | |
223 | * if we can't tell whether this is a noatime mount, then we | |
224 | * don't know whether to trust the value of s_atime_quantum. | |
225 | */ | |
226 | if (vfsmnt == NULL) | |
227 | return 0; | |
228 | ||
7f1a37e3 TY |
229 | if ((vfsmnt->mnt_flags & MNT_NOATIME) || |
230 | ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))) | |
231 | return 0; | |
232 | ||
7e913c53 MF |
233 | if (vfsmnt->mnt_flags & MNT_RELATIME) { |
234 | if ((timespec_compare(&inode->i_atime, &inode->i_mtime) <= 0) || | |
235 | (timespec_compare(&inode->i_atime, &inode->i_ctime) <= 0)) | |
236 | return 1; | |
237 | ||
238 | return 0; | |
239 | } | |
240 | ||
7f1a37e3 TY |
241 | now = CURRENT_TIME; |
242 | if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum)) | |
243 | return 0; | |
244 | else | |
245 | return 1; | |
246 | } | |
247 | ||
248 | int ocfs2_update_inode_atime(struct inode *inode, | |
249 | struct buffer_head *bh) | |
250 | { | |
251 | int ret; | |
252 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
253 | handle_t *handle; | |
c11e9faf | 254 | struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data; |
7f1a37e3 TY |
255 | |
256 | mlog_entry_void(); | |
257 | ||
258 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); | |
fa38e92c JK |
259 | if (IS_ERR(handle)) { |
260 | ret = PTR_ERR(handle); | |
7f1a37e3 TY |
261 | mlog_errno(ret); |
262 | goto out; | |
263 | } | |
264 | ||
0cf2f763 | 265 | ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh, |
13723d00 | 266 | OCFS2_JOURNAL_ACCESS_WRITE); |
c11e9faf MF |
267 | if (ret) { |
268 | mlog_errno(ret); | |
269 | goto out_commit; | |
270 | } | |
271 | ||
272 | /* | |
273 | * Don't use ocfs2_mark_inode_dirty() here as we don't always | |
274 | * have i_mutex to guard against concurrent changes to other | |
275 | * inode fields. | |
276 | */ | |
7f1a37e3 | 277 | inode->i_atime = CURRENT_TIME; |
c11e9faf MF |
278 | di->i_atime = cpu_to_le64(inode->i_atime.tv_sec); |
279 | di->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec); | |
ec20cec7 | 280 | ocfs2_journal_dirty(handle, bh); |
7f1a37e3 | 281 | |
c11e9faf | 282 | out_commit: |
7f1a37e3 TY |
283 | ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle); |
284 | out: | |
285 | mlog_exit(ret); | |
286 | return ret; | |
287 | } | |
288 | ||
6cb129f5 AB |
289 | static int ocfs2_set_inode_size(handle_t *handle, |
290 | struct inode *inode, | |
291 | struct buffer_head *fe_bh, | |
292 | u64 new_i_size) | |
ccd979bd MF |
293 | { |
294 | int status; | |
295 | ||
296 | mlog_entry_void(); | |
297 | i_size_write(inode, new_i_size); | |
8110b073 | 298 | inode->i_blocks = ocfs2_inode_sector_count(inode); |
ccd979bd MF |
299 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; |
300 | ||
301 | status = ocfs2_mark_inode_dirty(handle, inode, fe_bh); | |
302 | if (status < 0) { | |
303 | mlog_errno(status); | |
304 | goto bail; | |
305 | } | |
306 | ||
307 | bail: | |
308 | mlog_exit(status); | |
309 | return status; | |
310 | } | |
311 | ||
9e33d69f JK |
312 | int ocfs2_simple_size_update(struct inode *inode, |
313 | struct buffer_head *di_bh, | |
314 | u64 new_i_size) | |
ccd979bd MF |
315 | { |
316 | int ret; | |
317 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1fabe148 | 318 | handle_t *handle = NULL; |
ccd979bd | 319 | |
65eff9cc | 320 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
fa38e92c JK |
321 | if (IS_ERR(handle)) { |
322 | ret = PTR_ERR(handle); | |
ccd979bd MF |
323 | mlog_errno(ret); |
324 | goto out; | |
325 | } | |
326 | ||
327 | ret = ocfs2_set_inode_size(handle, inode, di_bh, | |
328 | new_i_size); | |
329 | if (ret < 0) | |
330 | mlog_errno(ret); | |
331 | ||
02dc1af4 | 332 | ocfs2_commit_trans(osb, handle); |
ccd979bd MF |
333 | out: |
334 | return ret; | |
335 | } | |
336 | ||
37f8a2bf TM |
337 | static int ocfs2_cow_file_pos(struct inode *inode, |
338 | struct buffer_head *fe_bh, | |
339 | u64 offset) | |
340 | { | |
341 | int status; | |
342 | u32 phys, cpos = offset >> OCFS2_SB(inode->i_sb)->s_clustersize_bits; | |
343 | unsigned int num_clusters = 0; | |
344 | unsigned int ext_flags = 0; | |
345 | ||
346 | /* | |
347 | * If the new offset is aligned to the range of the cluster, there is | |
348 | * no space for ocfs2_zero_range_for_truncate to fill, so no need to | |
349 | * CoW either. | |
350 | */ | |
351 | if ((offset & (OCFS2_SB(inode->i_sb)->s_clustersize - 1)) == 0) | |
352 | return 0; | |
353 | ||
354 | status = ocfs2_get_clusters(inode, cpos, &phys, | |
355 | &num_clusters, &ext_flags); | |
356 | if (status) { | |
357 | mlog_errno(status); | |
358 | goto out; | |
359 | } | |
360 | ||
361 | if (!(ext_flags & OCFS2_EXT_REFCOUNTED)) | |
362 | goto out; | |
363 | ||
364 | return ocfs2_refcount_cow(inode, fe_bh, cpos, 1, cpos+1); | |
365 | ||
366 | out: | |
367 | return status; | |
368 | } | |
369 | ||
ccd979bd MF |
370 | static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb, |
371 | struct inode *inode, | |
372 | struct buffer_head *fe_bh, | |
373 | u64 new_i_size) | |
374 | { | |
375 | int status; | |
1fabe148 | 376 | handle_t *handle; |
60b11392 | 377 | struct ocfs2_dinode *di; |
35edec1d | 378 | u64 cluster_bytes; |
ccd979bd MF |
379 | |
380 | mlog_entry_void(); | |
381 | ||
37f8a2bf TM |
382 | /* |
383 | * We need to CoW the cluster contains the offset if it is reflinked | |
384 | * since we will call ocfs2_zero_range_for_truncate later which will | |
385 | * write "0" from offset to the end of the cluster. | |
386 | */ | |
387 | status = ocfs2_cow_file_pos(inode, fe_bh, new_i_size); | |
388 | if (status) { | |
389 | mlog_errno(status); | |
390 | return status; | |
391 | } | |
392 | ||
ccd979bd MF |
393 | /* TODO: This needs to actually orphan the inode in this |
394 | * transaction. */ | |
395 | ||
65eff9cc | 396 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
ccd979bd MF |
397 | if (IS_ERR(handle)) { |
398 | status = PTR_ERR(handle); | |
399 | mlog_errno(status); | |
400 | goto out; | |
401 | } | |
402 | ||
0cf2f763 | 403 | status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh, |
13723d00 | 404 | OCFS2_JOURNAL_ACCESS_WRITE); |
60b11392 MF |
405 | if (status < 0) { |
406 | mlog_errno(status); | |
407 | goto out_commit; | |
408 | } | |
409 | ||
410 | /* | |
411 | * Do this before setting i_size. | |
412 | */ | |
35edec1d MF |
413 | cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size); |
414 | status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size, | |
415 | cluster_bytes); | |
60b11392 MF |
416 | if (status) { |
417 | mlog_errno(status); | |
418 | goto out_commit; | |
419 | } | |
420 | ||
421 | i_size_write(inode, new_i_size); | |
60b11392 MF |
422 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; |
423 | ||
424 | di = (struct ocfs2_dinode *) fe_bh->b_data; | |
425 | di->i_size = cpu_to_le64(new_i_size); | |
426 | di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec); | |
427 | di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec); | |
428 | ||
ec20cec7 | 429 | ocfs2_journal_dirty(handle, fe_bh); |
ccd979bd | 430 | |
60b11392 | 431 | out_commit: |
02dc1af4 | 432 | ocfs2_commit_trans(osb, handle); |
ccd979bd | 433 | out: |
60b11392 | 434 | |
ccd979bd MF |
435 | mlog_exit(status); |
436 | return status; | |
437 | } | |
438 | ||
439 | static int ocfs2_truncate_file(struct inode *inode, | |
440 | struct buffer_head *di_bh, | |
441 | u64 new_i_size) | |
442 | { | |
443 | int status = 0; | |
444 | struct ocfs2_dinode *fe = NULL; | |
445 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
ccd979bd | 446 | |
b0697053 MF |
447 | mlog_entry("(inode = %llu, new_i_size = %llu\n", |
448 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
449 | (unsigned long long)new_i_size); | |
ccd979bd | 450 | |
b657c95c JB |
451 | /* We trust di_bh because it comes from ocfs2_inode_lock(), which |
452 | * already validated it */ | |
ccd979bd | 453 | fe = (struct ocfs2_dinode *) di_bh->b_data; |
ccd979bd MF |
454 | |
455 | mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode), | |
b0697053 MF |
456 | "Inode %llu, inode i_size = %lld != di " |
457 | "i_size = %llu, i_flags = 0x%x\n", | |
458 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
ccd979bd | 459 | i_size_read(inode), |
b0697053 MF |
460 | (unsigned long long)le64_to_cpu(fe->i_size), |
461 | le32_to_cpu(fe->i_flags)); | |
ccd979bd MF |
462 | |
463 | if (new_i_size > le64_to_cpu(fe->i_size)) { | |
b0697053 MF |
464 | mlog(0, "asked to truncate file with size (%llu) to size (%llu)!\n", |
465 | (unsigned long long)le64_to_cpu(fe->i_size), | |
466 | (unsigned long long)new_i_size); | |
ccd979bd MF |
467 | status = -EINVAL; |
468 | mlog_errno(status); | |
469 | goto bail; | |
470 | } | |
471 | ||
b0697053 MF |
472 | mlog(0, "inode %llu, i_size = %llu, new_i_size = %llu\n", |
473 | (unsigned long long)le64_to_cpu(fe->i_blkno), | |
474 | (unsigned long long)le64_to_cpu(fe->i_size), | |
475 | (unsigned long long)new_i_size); | |
ccd979bd MF |
476 | |
477 | /* lets handle the simple truncate cases before doing any more | |
478 | * cluster locking. */ | |
479 | if (new_i_size == le64_to_cpu(fe->i_size)) | |
480 | goto bail; | |
481 | ||
2e89b2e4 MF |
482 | down_write(&OCFS2_I(inode)->ip_alloc_sem); |
483 | ||
4fe370af MF |
484 | ocfs2_resv_discard(&osb->osb_la_resmap, |
485 | &OCFS2_I(inode)->ip_la_data_resv); | |
486 | ||
c934a92d MF |
487 | /* |
488 | * The inode lock forced other nodes to sync and drop their | |
489 | * pages, which (correctly) happens even if we have a truncate | |
490 | * without allocation change - ocfs2 cluster sizes can be much | |
491 | * greater than page size, so we have to truncate them | |
492 | * anyway. | |
493 | */ | |
2e89b2e4 MF |
494 | unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1); |
495 | truncate_inode_pages(inode->i_mapping, new_i_size); | |
496 | ||
1afc32b9 MF |
497 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) { |
498 | status = ocfs2_truncate_inline(inode, di_bh, new_i_size, | |
b1967d0e | 499 | i_size_read(inode), 1); |
1afc32b9 MF |
500 | if (status) |
501 | mlog_errno(status); | |
502 | ||
c934a92d | 503 | goto bail_unlock_sem; |
1afc32b9 MF |
504 | } |
505 | ||
ccd979bd MF |
506 | /* alright, we're going to need to do a full blown alloc size |
507 | * change. Orphan the inode so that recovery can complete the | |
508 | * truncate if necessary. This does the task of marking | |
509 | * i_size. */ | |
510 | status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size); | |
511 | if (status < 0) { | |
512 | mlog_errno(status); | |
c934a92d | 513 | goto bail_unlock_sem; |
ccd979bd MF |
514 | } |
515 | ||
78f94673 | 516 | status = ocfs2_commit_truncate(osb, inode, di_bh); |
ccd979bd MF |
517 | if (status < 0) { |
518 | mlog_errno(status); | |
c934a92d | 519 | goto bail_unlock_sem; |
ccd979bd MF |
520 | } |
521 | ||
522 | /* TODO: orphan dir cleanup here. */ | |
c934a92d | 523 | bail_unlock_sem: |
2e89b2e4 MF |
524 | up_write(&OCFS2_I(inode)->ip_alloc_sem); |
525 | ||
ccd979bd | 526 | bail: |
8b2c0dba TM |
527 | if (!status && OCFS2_I(inode)->ip_clusters == 0) |
528 | status = ocfs2_try_remove_refcount_tree(inode, di_bh); | |
ccd979bd MF |
529 | |
530 | mlog_exit(status); | |
531 | return status; | |
532 | } | |
533 | ||
534 | /* | |
0eb8d47e | 535 | * extend file allocation only here. |
ccd979bd MF |
536 | * we'll update all the disk stuff, and oip->alloc_size |
537 | * | |
538 | * expect stuff to be locked, a transaction started and enough data / | |
539 | * metadata reservations in the contexts. | |
540 | * | |
541 | * Will return -EAGAIN, and a reason if a restart is needed. | |
542 | * If passed in, *reason will always be set, even in error. | |
543 | */ | |
0eb8d47e TM |
544 | int ocfs2_add_inode_data(struct ocfs2_super *osb, |
545 | struct inode *inode, | |
546 | u32 *logical_offset, | |
547 | u32 clusters_to_add, | |
548 | int mark_unwritten, | |
549 | struct buffer_head *fe_bh, | |
550 | handle_t *handle, | |
551 | struct ocfs2_alloc_context *data_ac, | |
552 | struct ocfs2_alloc_context *meta_ac, | |
553 | enum ocfs2_alloc_restarted *reason_ret) | |
ccd979bd | 554 | { |
f99b9b7c JB |
555 | int ret; |
556 | struct ocfs2_extent_tree et; | |
ccd979bd | 557 | |
5e404e9e | 558 | ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), fe_bh); |
cbee7e1a JB |
559 | ret = ocfs2_add_clusters_in_btree(handle, &et, logical_offset, |
560 | clusters_to_add, mark_unwritten, | |
561 | data_ac, meta_ac, reason_ret); | |
f99b9b7c JB |
562 | |
563 | return ret; | |
ccd979bd MF |
564 | } |
565 | ||
2ae99a60 MF |
566 | static int __ocfs2_extend_allocation(struct inode *inode, u32 logical_start, |
567 | u32 clusters_to_add, int mark_unwritten) | |
ccd979bd MF |
568 | { |
569 | int status = 0; | |
570 | int restart_func = 0; | |
abf8b156 | 571 | int credits; |
2ae99a60 | 572 | u32 prev_clusters; |
ccd979bd MF |
573 | struct buffer_head *bh = NULL; |
574 | struct ocfs2_dinode *fe = NULL; | |
1fabe148 | 575 | handle_t *handle = NULL; |
ccd979bd MF |
576 | struct ocfs2_alloc_context *data_ac = NULL; |
577 | struct ocfs2_alloc_context *meta_ac = NULL; | |
578 | enum ocfs2_alloc_restarted why; | |
579 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
f99b9b7c | 580 | struct ocfs2_extent_tree et; |
a90714c1 | 581 | int did_quota = 0; |
ccd979bd MF |
582 | |
583 | mlog_entry("(clusters_to_add = %u)\n", clusters_to_add); | |
584 | ||
dcd0538f MF |
585 | /* |
586 | * This function only exists for file systems which don't | |
587 | * support holes. | |
588 | */ | |
2ae99a60 | 589 | BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb)); |
dcd0538f | 590 | |
b657c95c | 591 | status = ocfs2_read_inode_block(inode, &bh); |
ccd979bd MF |
592 | if (status < 0) { |
593 | mlog_errno(status); | |
594 | goto leave; | |
595 | } | |
ccd979bd | 596 | fe = (struct ocfs2_dinode *) bh->b_data; |
ccd979bd MF |
597 | |
598 | restart_all: | |
599 | BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters); | |
600 | ||
e7d4cb6b TM |
601 | mlog(0, "extend inode %llu, i_size = %lld, di->i_clusters = %u, " |
602 | "clusters_to_add = %u\n", | |
603 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
604 | (long long)i_size_read(inode), le32_to_cpu(fe->i_clusters), | |
605 | clusters_to_add); | |
5e404e9e | 606 | ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), bh); |
f99b9b7c JB |
607 | status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0, |
608 | &data_ac, &meta_ac); | |
9517bac6 MF |
609 | if (status) { |
610 | mlog_errno(status); | |
611 | goto leave; | |
612 | } | |
613 | ||
811f933d TM |
614 | credits = ocfs2_calc_extend_credits(osb->sb, &fe->id2.i_list, |
615 | clusters_to_add); | |
65eff9cc | 616 | handle = ocfs2_start_trans(osb, credits); |
ccd979bd MF |
617 | if (IS_ERR(handle)) { |
618 | status = PTR_ERR(handle); | |
619 | handle = NULL; | |
620 | mlog_errno(status); | |
621 | goto leave; | |
622 | } | |
623 | ||
624 | restarted_transaction: | |
5dd4056d CH |
625 | status = dquot_alloc_space_nodirty(inode, |
626 | ocfs2_clusters_to_bytes(osb->sb, clusters_to_add)); | |
627 | if (status) | |
a90714c1 | 628 | goto leave; |
a90714c1 JK |
629 | did_quota = 1; |
630 | ||
ccd979bd MF |
631 | /* reserve a write to the file entry early on - that we if we |
632 | * run out of credits in the allocation path, we can still | |
633 | * update i_size. */ | |
0cf2f763 | 634 | status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh, |
13723d00 | 635 | OCFS2_JOURNAL_ACCESS_WRITE); |
ccd979bd MF |
636 | if (status < 0) { |
637 | mlog_errno(status); | |
638 | goto leave; | |
639 | } | |
640 | ||
641 | prev_clusters = OCFS2_I(inode)->ip_clusters; | |
642 | ||
0eb8d47e TM |
643 | status = ocfs2_add_inode_data(osb, |
644 | inode, | |
645 | &logical_start, | |
646 | clusters_to_add, | |
647 | mark_unwritten, | |
648 | bh, | |
649 | handle, | |
650 | data_ac, | |
651 | meta_ac, | |
652 | &why); | |
ccd979bd MF |
653 | if ((status < 0) && (status != -EAGAIN)) { |
654 | if (status != -ENOSPC) | |
655 | mlog_errno(status); | |
656 | goto leave; | |
657 | } | |
658 | ||
ec20cec7 | 659 | ocfs2_journal_dirty(handle, bh); |
ccd979bd MF |
660 | |
661 | spin_lock(&OCFS2_I(inode)->ip_lock); | |
662 | clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters); | |
663 | spin_unlock(&OCFS2_I(inode)->ip_lock); | |
a90714c1 | 664 | /* Release unused quota reservation */ |
5dd4056d | 665 | dquot_free_space(inode, |
a90714c1 JK |
666 | ocfs2_clusters_to_bytes(osb->sb, clusters_to_add)); |
667 | did_quota = 0; | |
ccd979bd MF |
668 | |
669 | if (why != RESTART_NONE && clusters_to_add) { | |
670 | if (why == RESTART_META) { | |
671 | mlog(0, "restarting function.\n"); | |
672 | restart_func = 1; | |
79681842 | 673 | status = 0; |
ccd979bd MF |
674 | } else { |
675 | BUG_ON(why != RESTART_TRANS); | |
676 | ||
677 | mlog(0, "restarting transaction.\n"); | |
678 | /* TODO: This can be more intelligent. */ | |
679 | credits = ocfs2_calc_extend_credits(osb->sb, | |
811f933d | 680 | &fe->id2.i_list, |
ccd979bd | 681 | clusters_to_add); |
1fabe148 | 682 | status = ocfs2_extend_trans(handle, credits); |
ccd979bd MF |
683 | if (status < 0) { |
684 | /* handle still has to be committed at | |
685 | * this point. */ | |
686 | status = -ENOMEM; | |
687 | mlog_errno(status); | |
688 | goto leave; | |
689 | } | |
690 | goto restarted_transaction; | |
691 | } | |
692 | } | |
693 | ||
b0697053 | 694 | mlog(0, "fe: i_clusters = %u, i_size=%llu\n", |
1ca1a111 MF |
695 | le32_to_cpu(fe->i_clusters), |
696 | (unsigned long long)le64_to_cpu(fe->i_size)); | |
ccd979bd | 697 | mlog(0, "inode: ip_clusters=%u, i_size=%lld\n", |
634bf74d | 698 | OCFS2_I(inode)->ip_clusters, (long long)i_size_read(inode)); |
ccd979bd MF |
699 | |
700 | leave: | |
a90714c1 | 701 | if (status < 0 && did_quota) |
5dd4056d | 702 | dquot_free_space(inode, |
a90714c1 | 703 | ocfs2_clusters_to_bytes(osb->sb, clusters_to_add)); |
ccd979bd | 704 | if (handle) { |
02dc1af4 | 705 | ocfs2_commit_trans(osb, handle); |
ccd979bd MF |
706 | handle = NULL; |
707 | } | |
708 | if (data_ac) { | |
709 | ocfs2_free_alloc_context(data_ac); | |
710 | data_ac = NULL; | |
711 | } | |
712 | if (meta_ac) { | |
713 | ocfs2_free_alloc_context(meta_ac); | |
714 | meta_ac = NULL; | |
715 | } | |
716 | if ((!status) && restart_func) { | |
717 | restart_func = 0; | |
718 | goto restart_all; | |
719 | } | |
a81cb88b MF |
720 | brelse(bh); |
721 | bh = NULL; | |
ccd979bd MF |
722 | |
723 | mlog_exit(status); | |
724 | return status; | |
725 | } | |
726 | ||
a4bfb4cf JB |
727 | /* |
728 | * While a write will already be ordering the data, a truncate will not. | |
729 | * Thus, we need to explicitly order the zeroed pages. | |
730 | */ | |
731 | static handle_t *ocfs2_zero_start_ordered_transaction(struct inode *inode) | |
732 | { | |
733 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
734 | handle_t *handle = NULL; | |
735 | int ret = 0; | |
736 | ||
737 | if (!ocfs2_should_order_data(inode)) | |
738 | goto out; | |
739 | ||
740 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); | |
741 | if (IS_ERR(handle)) { | |
742 | ret = -ENOMEM; | |
743 | mlog_errno(ret); | |
744 | goto out; | |
745 | } | |
746 | ||
747 | ret = ocfs2_jbd2_file_inode(handle, inode); | |
748 | if (ret < 0) | |
749 | mlog_errno(ret); | |
750 | ||
751 | out: | |
752 | if (ret) { | |
753 | if (!IS_ERR(handle)) | |
754 | ocfs2_commit_trans(osb, handle); | |
755 | handle = ERR_PTR(ret); | |
756 | } | |
757 | return handle; | |
758 | } | |
759 | ||
ccd979bd MF |
760 | /* Some parts of this taken from generic_cont_expand, which turned out |
761 | * to be too fragile to do exactly what we need without us having to | |
4e02ed4b | 762 | * worry about recursive locking in ->write_begin() and ->write_end(). */ |
a4bfb4cf JB |
763 | static int ocfs2_write_zero_page(struct inode *inode, u64 abs_from, |
764 | u64 abs_to) | |
ccd979bd MF |
765 | { |
766 | struct address_space *mapping = inode->i_mapping; | |
767 | struct page *page; | |
a4bfb4cf | 768 | unsigned long index = abs_from >> PAGE_CACHE_SHIFT; |
1fabe148 | 769 | handle_t *handle = NULL; |
ccd979bd | 770 | int ret; |
a4bfb4cf | 771 | unsigned zero_from, zero_to, block_start, block_end; |
ccd979bd | 772 | |
a4bfb4cf JB |
773 | BUG_ON(abs_from >= abs_to); |
774 | BUG_ON(abs_to > (((u64)index + 1) << PAGE_CACHE_SHIFT)); | |
775 | BUG_ON(abs_from & (inode->i_blkbits - 1)); | |
ccd979bd MF |
776 | |
777 | page = grab_cache_page(mapping, index); | |
778 | if (!page) { | |
779 | ret = -ENOMEM; | |
780 | mlog_errno(ret); | |
781 | goto out; | |
782 | } | |
783 | ||
a4bfb4cf JB |
784 | /* Get the offsets within the page that we want to zero */ |
785 | zero_from = abs_from & (PAGE_CACHE_SIZE - 1); | |
786 | zero_to = abs_to & (PAGE_CACHE_SIZE - 1); | |
787 | if (!zero_to) | |
788 | zero_to = PAGE_CACHE_SIZE; | |
ccd979bd | 789 | |
a4bfb4cf JB |
790 | /* We know that zero_from is block aligned */ |
791 | for (block_start = zero_from; block_start < zero_to; | |
792 | block_start = block_end) { | |
793 | block_end = block_start + (1 << inode->i_blkbits); | |
794 | ||
795 | /* | |
796 | * block_start is block-aligned. Bump it by one to | |
797 | * force ocfs2_{prepare,commit}_write() to zero the | |
798 | * whole block. | |
799 | */ | |
800 | ret = ocfs2_prepare_write_nolock(inode, page, | |
801 | block_start + 1, | |
802 | block_start + 1); | |
803 | if (ret < 0) { | |
804 | mlog_errno(ret); | |
ccd979bd MF |
805 | goto out_unlock; |
806 | } | |
ccd979bd | 807 | |
a4bfb4cf JB |
808 | if (!handle) { |
809 | handle = ocfs2_zero_start_ordered_transaction(inode); | |
810 | if (IS_ERR(handle)) { | |
811 | ret = PTR_ERR(handle); | |
812 | handle = NULL; | |
813 | break; | |
814 | } | |
815 | } | |
816 | ||
817 | /* must not update i_size! */ | |
818 | ret = block_commit_write(page, block_start + 1, | |
819 | block_start + 1); | |
820 | if (ret < 0) | |
821 | mlog_errno(ret); | |
822 | else | |
823 | ret = 0; | |
824 | } | |
ccd979bd MF |
825 | |
826 | if (handle) | |
02dc1af4 | 827 | ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle); |
a4bfb4cf | 828 | |
ccd979bd MF |
829 | out_unlock: |
830 | unlock_page(page); | |
831 | page_cache_release(page); | |
832 | out: | |
833 | return ret; | |
834 | } | |
835 | ||
836 | static int ocfs2_zero_extend(struct inode *inode, | |
837 | u64 zero_to_size) | |
838 | { | |
839 | int ret = 0; | |
a4bfb4cf | 840 | u64 start_off, next_off; |
ccd979bd MF |
841 | struct super_block *sb = inode->i_sb; |
842 | ||
843 | start_off = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode)); | |
844 | while (start_off < zero_to_size) { | |
a4bfb4cf JB |
845 | next_off = (start_off & PAGE_CACHE_MASK) + PAGE_CACHE_SIZE; |
846 | if (next_off > zero_to_size) | |
847 | next_off = zero_to_size; | |
848 | ret = ocfs2_write_zero_page(inode, start_off, next_off); | |
ccd979bd MF |
849 | if (ret < 0) { |
850 | mlog_errno(ret); | |
851 | goto out; | |
852 | } | |
853 | ||
a4bfb4cf | 854 | start_off = next_off; |
e2057c5a MF |
855 | |
856 | /* | |
857 | * Very large extends have the potential to lock up | |
858 | * the cpu for extended periods of time. | |
859 | */ | |
860 | cond_resched(); | |
ccd979bd MF |
861 | } |
862 | ||
863 | out: | |
864 | return ret; | |
865 | } | |
866 | ||
65ed39d6 MF |
867 | int ocfs2_extend_no_holes(struct inode *inode, u64 new_i_size, u64 zero_to) |
868 | { | |
869 | int ret; | |
870 | u32 clusters_to_add; | |
871 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
872 | ||
873 | clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size); | |
874 | if (clusters_to_add < oi->ip_clusters) | |
875 | clusters_to_add = 0; | |
876 | else | |
877 | clusters_to_add -= oi->ip_clusters; | |
878 | ||
879 | if (clusters_to_add) { | |
880 | ret = __ocfs2_extend_allocation(inode, oi->ip_clusters, | |
881 | clusters_to_add, 0); | |
882 | if (ret) { | |
883 | mlog_errno(ret); | |
884 | goto out; | |
885 | } | |
886 | } | |
887 | ||
888 | /* | |
889 | * Call this even if we don't add any clusters to the tree. We | |
890 | * still need to zero the area between the old i_size and the | |
891 | * new i_size. | |
892 | */ | |
893 | ret = ocfs2_zero_extend(inode, zero_to); | |
894 | if (ret < 0) | |
895 | mlog_errno(ret); | |
896 | ||
897 | out: | |
898 | return ret; | |
899 | } | |
900 | ||
ccd979bd MF |
901 | static int ocfs2_extend_file(struct inode *inode, |
902 | struct buffer_head *di_bh, | |
65ed39d6 | 903 | u64 new_i_size) |
ccd979bd | 904 | { |
c934a92d | 905 | int ret = 0; |
1afc32b9 | 906 | struct ocfs2_inode_info *oi = OCFS2_I(inode); |
ccd979bd | 907 | |
65ed39d6 | 908 | BUG_ON(!di_bh); |
53013cba | 909 | |
ccd979bd MF |
910 | /* setattr sometimes calls us like this. */ |
911 | if (new_i_size == 0) | |
912 | goto out; | |
913 | ||
914 | if (i_size_read(inode) == new_i_size) | |
915 | goto out; | |
916 | BUG_ON(new_i_size < i_size_read(inode)); | |
917 | ||
1afc32b9 MF |
918 | /* |
919 | * Fall through for converting inline data, even if the fs | |
920 | * supports sparse files. | |
921 | * | |
922 | * The check for inline data here is legal - nobody can add | |
923 | * the feature since we have i_mutex. We must check it again | |
924 | * after acquiring ip_alloc_sem though, as paths like mmap | |
925 | * might have raced us to converting the inode to extents. | |
926 | */ | |
927 | if (!(oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) | |
928 | && ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))) | |
3a0782d0 | 929 | goto out_update_size; |
ccd979bd | 930 | |
0effef77 | 931 | /* |
65ed39d6 MF |
932 | * The alloc sem blocks people in read/write from reading our |
933 | * allocation until we're done changing it. We depend on | |
934 | * i_mutex to block other extend/truncate calls while we're | |
935 | * here. | |
0effef77 | 936 | */ |
1afc32b9 MF |
937 | down_write(&oi->ip_alloc_sem); |
938 | ||
939 | if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) { | |
940 | /* | |
941 | * We can optimize small extends by keeping the inodes | |
942 | * inline data. | |
943 | */ | |
944 | if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) { | |
945 | up_write(&oi->ip_alloc_sem); | |
946 | goto out_update_size; | |
947 | } | |
948 | ||
949 | ret = ocfs2_convert_inline_data_to_extents(inode, di_bh); | |
950 | if (ret) { | |
951 | up_write(&oi->ip_alloc_sem); | |
952 | ||
953 | mlog_errno(ret); | |
c934a92d | 954 | goto out; |
1afc32b9 MF |
955 | } |
956 | } | |
957 | ||
958 | if (!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))) | |
959 | ret = ocfs2_extend_no_holes(inode, new_i_size, new_i_size); | |
960 | ||
961 | up_write(&oi->ip_alloc_sem); | |
65ed39d6 | 962 | |
0effef77 MF |
963 | if (ret < 0) { |
964 | mlog_errno(ret); | |
c934a92d | 965 | goto out; |
53013cba MF |
966 | } |
967 | ||
3a0782d0 | 968 | out_update_size: |
65ed39d6 MF |
969 | ret = ocfs2_simple_size_update(inode, di_bh, new_i_size); |
970 | if (ret < 0) | |
971 | mlog_errno(ret); | |
ccd979bd MF |
972 | |
973 | out: | |
974 | return ret; | |
975 | } | |
976 | ||
977 | int ocfs2_setattr(struct dentry *dentry, struct iattr *attr) | |
978 | { | |
979 | int status = 0, size_change; | |
980 | struct inode *inode = dentry->d_inode; | |
981 | struct super_block *sb = inode->i_sb; | |
982 | struct ocfs2_super *osb = OCFS2_SB(sb); | |
983 | struct buffer_head *bh = NULL; | |
1fabe148 | 984 | handle_t *handle = NULL; |
65bac575 | 985 | struct dquot *transfer_to[MAXQUOTAS] = { }; |
52a9ee28 | 986 | int qtype; |
ccd979bd MF |
987 | |
988 | mlog_entry("(0x%p, '%.*s')\n", dentry, | |
989 | dentry->d_name.len, dentry->d_name.name); | |
990 | ||
bc535809 SM |
991 | /* ensuring we don't even attempt to truncate a symlink */ |
992 | if (S_ISLNK(inode->i_mode)) | |
993 | attr->ia_valid &= ~ATTR_SIZE; | |
994 | ||
ccd979bd MF |
995 | if (attr->ia_valid & ATTR_MODE) |
996 | mlog(0, "mode change: %d\n", attr->ia_mode); | |
997 | if (attr->ia_valid & ATTR_UID) | |
998 | mlog(0, "uid change: %d\n", attr->ia_uid); | |
999 | if (attr->ia_valid & ATTR_GID) | |
1000 | mlog(0, "gid change: %d\n", attr->ia_gid); | |
1001 | if (attr->ia_valid & ATTR_SIZE) | |
1002 | mlog(0, "size change...\n"); | |
1003 | if (attr->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME)) | |
1004 | mlog(0, "time change...\n"); | |
1005 | ||
1006 | #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \ | |
1007 | | ATTR_GID | ATTR_UID | ATTR_MODE) | |
1008 | if (!(attr->ia_valid & OCFS2_VALID_ATTRS)) { | |
1009 | mlog(0, "can't handle attrs: 0x%x\n", attr->ia_valid); | |
1010 | return 0; | |
1011 | } | |
1012 | ||
1013 | status = inode_change_ok(inode, attr); | |
1014 | if (status) | |
1015 | return status; | |
1016 | ||
12755627 DM |
1017 | if (is_quota_modification(inode, attr)) |
1018 | dquot_initialize(inode); | |
ccd979bd MF |
1019 | size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE; |
1020 | if (size_change) { | |
1021 | status = ocfs2_rw_lock(inode, 1); | |
1022 | if (status < 0) { | |
1023 | mlog_errno(status); | |
1024 | goto bail; | |
1025 | } | |
1026 | } | |
1027 | ||
e63aecb6 | 1028 | status = ocfs2_inode_lock(inode, &bh, 1); |
ccd979bd MF |
1029 | if (status < 0) { |
1030 | if (status != -ENOENT) | |
1031 | mlog_errno(status); | |
1032 | goto bail_unlock_rw; | |
1033 | } | |
1034 | ||
1035 | if (size_change && attr->ia_size != i_size_read(inode)) { | |
5051f768 WW |
1036 | status = inode_newsize_ok(inode, attr->ia_size); |
1037 | if (status) | |
ce76fd30 | 1038 | goto bail_unlock; |
ce76fd30 | 1039 | |
2b4e30fb JB |
1040 | if (i_size_read(inode) > attr->ia_size) { |
1041 | if (ocfs2_should_order_data(inode)) { | |
1042 | status = ocfs2_begin_ordered_truncate(inode, | |
1043 | attr->ia_size); | |
1044 | if (status) | |
1045 | goto bail_unlock; | |
1046 | } | |
ccd979bd | 1047 | status = ocfs2_truncate_file(inode, bh, attr->ia_size); |
2b4e30fb | 1048 | } else |
65ed39d6 | 1049 | status = ocfs2_extend_file(inode, bh, attr->ia_size); |
ccd979bd MF |
1050 | if (status < 0) { |
1051 | if (status != -ENOSPC) | |
1052 | mlog_errno(status); | |
1053 | status = -ENOSPC; | |
1054 | goto bail_unlock; | |
1055 | } | |
1056 | } | |
1057 | ||
a90714c1 JK |
1058 | if ((attr->ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) || |
1059 | (attr->ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) { | |
65bac575 JK |
1060 | /* |
1061 | * Gather pointers to quota structures so that allocation / | |
1062 | * freeing of quota structures happens here and not inside | |
b43fa828 | 1063 | * dquot_transfer() where we have problems with lock ordering |
65bac575 | 1064 | */ |
a90714c1 JK |
1065 | if (attr->ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid |
1066 | && OCFS2_HAS_RO_COMPAT_FEATURE(sb, | |
1067 | OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) { | |
65bac575 JK |
1068 | transfer_to[USRQUOTA] = dqget(sb, attr->ia_uid, |
1069 | USRQUOTA); | |
52a9ee28 | 1070 | if (!transfer_to[USRQUOTA]) { |
65bac575 | 1071 | status = -ESRCH; |
a90714c1 | 1072 | goto bail_unlock; |
65bac575 | 1073 | } |
a90714c1 JK |
1074 | } |
1075 | if (attr->ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid | |
1076 | && OCFS2_HAS_RO_COMPAT_FEATURE(sb, | |
1077 | OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) { | |
65bac575 JK |
1078 | transfer_to[GRPQUOTA] = dqget(sb, attr->ia_gid, |
1079 | GRPQUOTA); | |
52a9ee28 | 1080 | if (!transfer_to[GRPQUOTA]) { |
65bac575 | 1081 | status = -ESRCH; |
a90714c1 | 1082 | goto bail_unlock; |
65bac575 | 1083 | } |
a90714c1 | 1084 | } |
65bac575 JK |
1085 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS + |
1086 | 2 * ocfs2_quota_trans_credits(sb)); | |
a90714c1 JK |
1087 | if (IS_ERR(handle)) { |
1088 | status = PTR_ERR(handle); | |
1089 | mlog_errno(status); | |
1090 | goto bail_unlock; | |
1091 | } | |
52a9ee28 | 1092 | status = __dquot_transfer(inode, transfer_to); |
a90714c1 JK |
1093 | if (status < 0) |
1094 | goto bail_commit; | |
1095 | } else { | |
1096 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); | |
1097 | if (IS_ERR(handle)) { | |
1098 | status = PTR_ERR(handle); | |
1099 | mlog_errno(status); | |
1100 | goto bail_unlock; | |
1101 | } | |
ccd979bd MF |
1102 | } |
1103 | ||
7307de80 | 1104 | /* |
15c6fd97 | 1105 | * This will intentionally not wind up calling simple_setsize(), |
7307de80 MF |
1106 | * since all the work for a size change has been done above. |
1107 | * Otherwise, we could get into problems with truncate as | |
1108 | * ip_alloc_sem is used there to protect against i_size | |
1109 | * changes. | |
1110 | */ | |
ccd979bd MF |
1111 | status = inode_setattr(inode, attr); |
1112 | if (status < 0) { | |
1113 | mlog_errno(status); | |
1114 | goto bail_commit; | |
1115 | } | |
1116 | ||
1117 | status = ocfs2_mark_inode_dirty(handle, inode, bh); | |
1118 | if (status < 0) | |
1119 | mlog_errno(status); | |
1120 | ||
1121 | bail_commit: | |
02dc1af4 | 1122 | ocfs2_commit_trans(osb, handle); |
ccd979bd | 1123 | bail_unlock: |
e63aecb6 | 1124 | ocfs2_inode_unlock(inode, 1); |
ccd979bd MF |
1125 | bail_unlock_rw: |
1126 | if (size_change) | |
1127 | ocfs2_rw_unlock(inode, 1); | |
1128 | bail: | |
a81cb88b | 1129 | brelse(bh); |
ccd979bd | 1130 | |
65bac575 | 1131 | /* Release quota pointers in case we acquired them */ |
52a9ee28 | 1132 | for (qtype = 0; qtype < MAXQUOTAS; qtype++) |
65bac575 | 1133 | dqput(transfer_to[qtype]); |
65bac575 | 1134 | |
060bc66d TY |
1135 | if (!status && attr->ia_valid & ATTR_MODE) { |
1136 | status = ocfs2_acl_chmod(inode); | |
1137 | if (status < 0) | |
1138 | mlog_errno(status); | |
1139 | } | |
1140 | ||
ccd979bd MF |
1141 | mlog_exit(status); |
1142 | return status; | |
1143 | } | |
1144 | ||
1145 | int ocfs2_getattr(struct vfsmount *mnt, | |
1146 | struct dentry *dentry, | |
1147 | struct kstat *stat) | |
1148 | { | |
1149 | struct inode *inode = dentry->d_inode; | |
1150 | struct super_block *sb = dentry->d_inode->i_sb; | |
1151 | struct ocfs2_super *osb = sb->s_fs_info; | |
1152 | int err; | |
1153 | ||
1154 | mlog_entry_void(); | |
1155 | ||
1156 | err = ocfs2_inode_revalidate(dentry); | |
1157 | if (err) { | |
1158 | if (err != -ENOENT) | |
1159 | mlog_errno(err); | |
1160 | goto bail; | |
1161 | } | |
1162 | ||
1163 | generic_fillattr(inode, stat); | |
1164 | ||
1165 | /* We set the blksize from the cluster size for performance */ | |
1166 | stat->blksize = osb->s_clustersize; | |
1167 | ||
1168 | bail: | |
1169 | mlog_exit(err); | |
1170 | ||
1171 | return err; | |
1172 | } | |
1173 | ||
e6305c43 | 1174 | int ocfs2_permission(struct inode *inode, int mask) |
d38eb8db TY |
1175 | { |
1176 | int ret; | |
1177 | ||
1178 | mlog_entry_void(); | |
1179 | ||
e63aecb6 | 1180 | ret = ocfs2_inode_lock(inode, NULL, 0); |
d38eb8db | 1181 | if (ret) { |
a9f5f707 MF |
1182 | if (ret != -ENOENT) |
1183 | mlog_errno(ret); | |
d38eb8db TY |
1184 | goto out; |
1185 | } | |
1186 | ||
23fc2702 | 1187 | ret = generic_permission(inode, mask, ocfs2_check_acl); |
d38eb8db | 1188 | |
e63aecb6 | 1189 | ocfs2_inode_unlock(inode, 0); |
d38eb8db TY |
1190 | out: |
1191 | mlog_exit(ret); | |
1192 | return ret; | |
1193 | } | |
1194 | ||
b2580103 MF |
1195 | static int __ocfs2_write_remove_suid(struct inode *inode, |
1196 | struct buffer_head *bh) | |
ccd979bd MF |
1197 | { |
1198 | int ret; | |
1fabe148 | 1199 | handle_t *handle; |
ccd979bd MF |
1200 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
1201 | struct ocfs2_dinode *di; | |
1202 | ||
b0697053 | 1203 | mlog_entry("(Inode %llu, mode 0%o)\n", |
b2580103 | 1204 | (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_mode); |
ccd979bd | 1205 | |
65eff9cc | 1206 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
fa38e92c JK |
1207 | if (IS_ERR(handle)) { |
1208 | ret = PTR_ERR(handle); | |
ccd979bd MF |
1209 | mlog_errno(ret); |
1210 | goto out; | |
1211 | } | |
1212 | ||
0cf2f763 | 1213 | ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh, |
13723d00 | 1214 | OCFS2_JOURNAL_ACCESS_WRITE); |
ccd979bd MF |
1215 | if (ret < 0) { |
1216 | mlog_errno(ret); | |
b2580103 | 1217 | goto out_trans; |
ccd979bd MF |
1218 | } |
1219 | ||
1220 | inode->i_mode &= ~S_ISUID; | |
1221 | if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP)) | |
1222 | inode->i_mode &= ~S_ISGID; | |
1223 | ||
1224 | di = (struct ocfs2_dinode *) bh->b_data; | |
1225 | di->i_mode = cpu_to_le16(inode->i_mode); | |
1226 | ||
ec20cec7 | 1227 | ocfs2_journal_dirty(handle, bh); |
b2580103 | 1228 | |
ccd979bd | 1229 | out_trans: |
02dc1af4 | 1230 | ocfs2_commit_trans(osb, handle); |
ccd979bd MF |
1231 | out: |
1232 | mlog_exit(ret); | |
1233 | return ret; | |
1234 | } | |
1235 | ||
9517bac6 MF |
1236 | /* |
1237 | * Will look for holes and unwritten extents in the range starting at | |
1238 | * pos for count bytes (inclusive). | |
1239 | */ | |
1240 | static int ocfs2_check_range_for_holes(struct inode *inode, loff_t pos, | |
1241 | size_t count) | |
1242 | { | |
1243 | int ret = 0; | |
49cb8d2d | 1244 | unsigned int extent_flags; |
9517bac6 MF |
1245 | u32 cpos, clusters, extent_len, phys_cpos; |
1246 | struct super_block *sb = inode->i_sb; | |
1247 | ||
1248 | cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits; | |
1249 | clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos; | |
1250 | ||
1251 | while (clusters) { | |
49cb8d2d MF |
1252 | ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len, |
1253 | &extent_flags); | |
9517bac6 MF |
1254 | if (ret < 0) { |
1255 | mlog_errno(ret); | |
1256 | goto out; | |
1257 | } | |
1258 | ||
49cb8d2d | 1259 | if (phys_cpos == 0 || (extent_flags & OCFS2_EXT_UNWRITTEN)) { |
9517bac6 MF |
1260 | ret = 1; |
1261 | break; | |
1262 | } | |
1263 | ||
1264 | if (extent_len > clusters) | |
1265 | extent_len = clusters; | |
1266 | ||
1267 | clusters -= extent_len; | |
1268 | cpos += extent_len; | |
1269 | } | |
1270 | out: | |
1271 | return ret; | |
1272 | } | |
1273 | ||
b2580103 MF |
1274 | static int ocfs2_write_remove_suid(struct inode *inode) |
1275 | { | |
1276 | int ret; | |
1277 | struct buffer_head *bh = NULL; | |
b2580103 | 1278 | |
b657c95c | 1279 | ret = ocfs2_read_inode_block(inode, &bh); |
b2580103 MF |
1280 | if (ret < 0) { |
1281 | mlog_errno(ret); | |
1282 | goto out; | |
1283 | } | |
1284 | ||
1285 | ret = __ocfs2_write_remove_suid(inode, bh); | |
1286 | out: | |
1287 | brelse(bh); | |
1288 | return ret; | |
1289 | } | |
1290 | ||
2ae99a60 MF |
1291 | /* |
1292 | * Allocate enough extents to cover the region starting at byte offset | |
1293 | * start for len bytes. Existing extents are skipped, any extents | |
1294 | * added are marked as "unwritten". | |
1295 | */ | |
1296 | static int ocfs2_allocate_unwritten_extents(struct inode *inode, | |
1297 | u64 start, u64 len) | |
1298 | { | |
1299 | int ret; | |
1300 | u32 cpos, phys_cpos, clusters, alloc_size; | |
1afc32b9 MF |
1301 | u64 end = start + len; |
1302 | struct buffer_head *di_bh = NULL; | |
1303 | ||
1304 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) { | |
b657c95c | 1305 | ret = ocfs2_read_inode_block(inode, &di_bh); |
1afc32b9 MF |
1306 | if (ret) { |
1307 | mlog_errno(ret); | |
1308 | goto out; | |
1309 | } | |
1310 | ||
1311 | /* | |
1312 | * Nothing to do if the requested reservation range | |
1313 | * fits within the inode. | |
1314 | */ | |
1315 | if (ocfs2_size_fits_inline_data(di_bh, end)) | |
1316 | goto out; | |
1317 | ||
1318 | ret = ocfs2_convert_inline_data_to_extents(inode, di_bh); | |
1319 | if (ret) { | |
1320 | mlog_errno(ret); | |
1321 | goto out; | |
1322 | } | |
1323 | } | |
2ae99a60 MF |
1324 | |
1325 | /* | |
1326 | * We consider both start and len to be inclusive. | |
1327 | */ | |
1328 | cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits; | |
1329 | clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len); | |
1330 | clusters -= cpos; | |
1331 | ||
1332 | while (clusters) { | |
1333 | ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, | |
1334 | &alloc_size, NULL); | |
1335 | if (ret) { | |
1336 | mlog_errno(ret); | |
1337 | goto out; | |
1338 | } | |
1339 | ||
1340 | /* | |
1341 | * Hole or existing extent len can be arbitrary, so | |
1342 | * cap it to our own allocation request. | |
1343 | */ | |
1344 | if (alloc_size > clusters) | |
1345 | alloc_size = clusters; | |
1346 | ||
1347 | if (phys_cpos) { | |
1348 | /* | |
1349 | * We already have an allocation at this | |
1350 | * region so we can safely skip it. | |
1351 | */ | |
1352 | goto next; | |
1353 | } | |
1354 | ||
1355 | ret = __ocfs2_extend_allocation(inode, cpos, alloc_size, 1); | |
1356 | if (ret) { | |
1357 | if (ret != -ENOSPC) | |
1358 | mlog_errno(ret); | |
1359 | goto out; | |
1360 | } | |
1361 | ||
1362 | next: | |
1363 | cpos += alloc_size; | |
1364 | clusters -= alloc_size; | |
1365 | } | |
1366 | ||
1367 | ret = 0; | |
1368 | out: | |
1afc32b9 MF |
1369 | |
1370 | brelse(di_bh); | |
2ae99a60 MF |
1371 | return ret; |
1372 | } | |
1373 | ||
063c4561 MF |
1374 | /* |
1375 | * Truncate a byte range, avoiding pages within partial clusters. This | |
1376 | * preserves those pages for the zeroing code to write to. | |
1377 | */ | |
1378 | static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start, | |
1379 | u64 byte_len) | |
1380 | { | |
1381 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1382 | loff_t start, end; | |
1383 | struct address_space *mapping = inode->i_mapping; | |
1384 | ||
1385 | start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start); | |
1386 | end = byte_start + byte_len; | |
1387 | end = end & ~(osb->s_clustersize - 1); | |
1388 | ||
1389 | if (start < end) { | |
1390 | unmap_mapping_range(mapping, start, end - start, 0); | |
1391 | truncate_inode_pages_range(mapping, start, end - 1); | |
1392 | } | |
1393 | } | |
1394 | ||
1395 | static int ocfs2_zero_partial_clusters(struct inode *inode, | |
1396 | u64 start, u64 len) | |
1397 | { | |
1398 | int ret = 0; | |
1399 | u64 tmpend, end = start + len; | |
1400 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1401 | unsigned int csize = osb->s_clustersize; | |
1402 | handle_t *handle; | |
1403 | ||
1404 | /* | |
1405 | * The "start" and "end" values are NOT necessarily part of | |
1406 | * the range whose allocation is being deleted. Rather, this | |
1407 | * is what the user passed in with the request. We must zero | |
1408 | * partial clusters here. There's no need to worry about | |
1409 | * physical allocation - the zeroing code knows to skip holes. | |
1410 | */ | |
1411 | mlog(0, "byte start: %llu, end: %llu\n", | |
1412 | (unsigned long long)start, (unsigned long long)end); | |
1413 | ||
1414 | /* | |
1415 | * If both edges are on a cluster boundary then there's no | |
1416 | * zeroing required as the region is part of the allocation to | |
1417 | * be truncated. | |
1418 | */ | |
1419 | if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0) | |
1420 | goto out; | |
1421 | ||
1422 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); | |
fa38e92c JK |
1423 | if (IS_ERR(handle)) { |
1424 | ret = PTR_ERR(handle); | |
063c4561 MF |
1425 | mlog_errno(ret); |
1426 | goto out; | |
1427 | } | |
1428 | ||
1429 | /* | |
1430 | * We want to get the byte offset of the end of the 1st cluster. | |
1431 | */ | |
1432 | tmpend = (u64)osb->s_clustersize + (start & ~(osb->s_clustersize - 1)); | |
1433 | if (tmpend > end) | |
1434 | tmpend = end; | |
1435 | ||
1436 | mlog(0, "1st range: start: %llu, tmpend: %llu\n", | |
1437 | (unsigned long long)start, (unsigned long long)tmpend); | |
1438 | ||
1439 | ret = ocfs2_zero_range_for_truncate(inode, handle, start, tmpend); | |
1440 | if (ret) | |
1441 | mlog_errno(ret); | |
1442 | ||
1443 | if (tmpend < end) { | |
1444 | /* | |
1445 | * This may make start and end equal, but the zeroing | |
1446 | * code will skip any work in that case so there's no | |
1447 | * need to catch it up here. | |
1448 | */ | |
1449 | start = end & ~(osb->s_clustersize - 1); | |
1450 | ||
1451 | mlog(0, "2nd range: start: %llu, end: %llu\n", | |
1452 | (unsigned long long)start, (unsigned long long)end); | |
1453 | ||
1454 | ret = ocfs2_zero_range_for_truncate(inode, handle, start, end); | |
1455 | if (ret) | |
1456 | mlog_errno(ret); | |
1457 | } | |
1458 | ||
1459 | ocfs2_commit_trans(osb, handle); | |
1460 | out: | |
1461 | return ret; | |
1462 | } | |
1463 | ||
c1631d4a TY |
1464 | static int ocfs2_find_rec(struct ocfs2_extent_list *el, u32 pos) |
1465 | { | |
1466 | int i; | |
1467 | struct ocfs2_extent_rec *rec = NULL; | |
1468 | ||
1469 | for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) { | |
1470 | ||
1471 | rec = &el->l_recs[i]; | |
1472 | ||
1473 | if (le32_to_cpu(rec->e_cpos) < pos) | |
1474 | break; | |
1475 | } | |
1476 | ||
1477 | return i; | |
1478 | } | |
1479 | ||
1480 | /* | |
1481 | * Helper to calculate the punching pos and length in one run, we handle the | |
1482 | * following three cases in order: | |
1483 | * | |
1484 | * - remove the entire record | |
1485 | * - remove a partial record | |
1486 | * - no record needs to be removed (hole-punching completed) | |
1487 | */ | |
1488 | static void ocfs2_calc_trunc_pos(struct inode *inode, | |
1489 | struct ocfs2_extent_list *el, | |
1490 | struct ocfs2_extent_rec *rec, | |
1491 | u32 trunc_start, u32 *trunc_cpos, | |
1492 | u32 *trunc_len, u32 *trunc_end, | |
1493 | u64 *blkno, int *done) | |
1494 | { | |
1495 | int ret = 0; | |
1496 | u32 coff, range; | |
1497 | ||
1498 | range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec); | |
1499 | ||
1500 | if (le32_to_cpu(rec->e_cpos) >= trunc_start) { | |
1501 | *trunc_cpos = le32_to_cpu(rec->e_cpos); | |
1502 | /* | |
1503 | * Skip holes if any. | |
1504 | */ | |
1505 | if (range < *trunc_end) | |
1506 | *trunc_end = range; | |
1507 | *trunc_len = *trunc_end - le32_to_cpu(rec->e_cpos); | |
1508 | *blkno = le64_to_cpu(rec->e_blkno); | |
1509 | *trunc_end = le32_to_cpu(rec->e_cpos); | |
1510 | } else if (range > trunc_start) { | |
1511 | *trunc_cpos = trunc_start; | |
1512 | *trunc_len = *trunc_end - trunc_start; | |
1513 | coff = trunc_start - le32_to_cpu(rec->e_cpos); | |
1514 | *blkno = le64_to_cpu(rec->e_blkno) + | |
1515 | ocfs2_clusters_to_blocks(inode->i_sb, coff); | |
1516 | *trunc_end = trunc_start; | |
1517 | } else { | |
1518 | /* | |
1519 | * It may have two following possibilities: | |
1520 | * | |
1521 | * - last record has been removed | |
1522 | * - trunc_start was within a hole | |
1523 | * | |
1524 | * both two cases mean the completion of hole punching. | |
1525 | */ | |
1526 | ret = 1; | |
1527 | } | |
1528 | ||
1529 | *done = ret; | |
1530 | } | |
1531 | ||
063c4561 MF |
1532 | static int ocfs2_remove_inode_range(struct inode *inode, |
1533 | struct buffer_head *di_bh, u64 byte_start, | |
1534 | u64 byte_len) | |
1535 | { | |
c1631d4a TY |
1536 | int ret = 0, flags = 0, done = 0, i; |
1537 | u32 trunc_start, trunc_len, trunc_end, trunc_cpos, phys_cpos; | |
1538 | u32 cluster_in_el; | |
063c4561 MF |
1539 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
1540 | struct ocfs2_cached_dealloc_ctxt dealloc; | |
b1967d0e | 1541 | struct address_space *mapping = inode->i_mapping; |
fecc0112 | 1542 | struct ocfs2_extent_tree et; |
c1631d4a TY |
1543 | struct ocfs2_path *path = NULL; |
1544 | struct ocfs2_extent_list *el = NULL; | |
1545 | struct ocfs2_extent_rec *rec = NULL; | |
e8aec068 | 1546 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; |
c1631d4a | 1547 | u64 blkno, refcount_loc = le64_to_cpu(di->i_refcount_loc); |
063c4561 | 1548 | |
5e404e9e | 1549 | ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh); |
063c4561 MF |
1550 | ocfs2_init_dealloc_ctxt(&dealloc); |
1551 | ||
1552 | if (byte_len == 0) | |
1553 | return 0; | |
1554 | ||
1afc32b9 MF |
1555 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) { |
1556 | ret = ocfs2_truncate_inline(inode, di_bh, byte_start, | |
b1967d0e MF |
1557 | byte_start + byte_len, 0); |
1558 | if (ret) { | |
1afc32b9 | 1559 | mlog_errno(ret); |
b1967d0e MF |
1560 | goto out; |
1561 | } | |
1562 | /* | |
1563 | * There's no need to get fancy with the page cache | |
1564 | * truncate of an inline-data inode. We're talking | |
1565 | * about less than a page here, which will be cached | |
1566 | * in the dinode buffer anyway. | |
1567 | */ | |
1568 | unmap_mapping_range(mapping, 0, 0, 0); | |
1569 | truncate_inode_pages(mapping, 0); | |
1570 | goto out; | |
1afc32b9 MF |
1571 | } |
1572 | ||
e8aec068 TY |
1573 | /* |
1574 | * For reflinks, we may need to CoW 2 clusters which might be | |
1575 | * partially zero'd later, if hole's start and end offset were | |
1576 | * within one cluster(means is not exactly aligned to clustersize). | |
1577 | */ | |
1578 | ||
1579 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) { | |
1580 | ||
1581 | ret = ocfs2_cow_file_pos(inode, di_bh, byte_start); | |
1582 | if (ret) { | |
1583 | mlog_errno(ret); | |
1584 | goto out; | |
1585 | } | |
1586 | ||
1587 | ret = ocfs2_cow_file_pos(inode, di_bh, byte_start + byte_len); | |
1588 | if (ret) { | |
1589 | mlog_errno(ret); | |
1590 | goto out; | |
1591 | } | |
1592 | } | |
1593 | ||
063c4561 | 1594 | trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start); |
c1631d4a TY |
1595 | trunc_end = (byte_start + byte_len) >> osb->s_clustersize_bits; |
1596 | cluster_in_el = trunc_end; | |
063c4561 | 1597 | |
c1631d4a | 1598 | mlog(0, "Inode: %llu, start: %llu, len: %llu, cstart: %u, cend: %u\n", |
063c4561 MF |
1599 | (unsigned long long)OCFS2_I(inode)->ip_blkno, |
1600 | (unsigned long long)byte_start, | |
c1631d4a | 1601 | (unsigned long long)byte_len, trunc_start, trunc_end); |
063c4561 MF |
1602 | |
1603 | ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len); | |
1604 | if (ret) { | |
1605 | mlog_errno(ret); | |
1606 | goto out; | |
1607 | } | |
1608 | ||
c1631d4a TY |
1609 | path = ocfs2_new_path_from_et(&et); |
1610 | if (!path) { | |
1611 | ret = -ENOMEM; | |
1612 | mlog_errno(ret); | |
1613 | goto out; | |
1614 | } | |
1615 | ||
1616 | while (trunc_end > trunc_start) { | |
1617 | ||
1618 | ret = ocfs2_find_path(INODE_CACHE(inode), path, | |
1619 | cluster_in_el); | |
063c4561 MF |
1620 | if (ret) { |
1621 | mlog_errno(ret); | |
1622 | goto out; | |
1623 | } | |
1624 | ||
c1631d4a | 1625 | el = path_leaf_el(path); |
063c4561 | 1626 | |
c1631d4a TY |
1627 | i = ocfs2_find_rec(el, trunc_end); |
1628 | /* | |
1629 | * Need to go to previous extent block. | |
1630 | */ | |
1631 | if (i < 0) { | |
1632 | if (path->p_tree_depth == 0) | |
1633 | break; | |
063c4561 | 1634 | |
c1631d4a TY |
1635 | ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb, |
1636 | path, | |
1637 | &cluster_in_el); | |
063c4561 MF |
1638 | if (ret) { |
1639 | mlog_errno(ret); | |
1640 | goto out; | |
1641 | } | |
c1631d4a TY |
1642 | |
1643 | /* | |
1644 | * We've reached the leftmost extent block, | |
1645 | * it's safe to leave. | |
1646 | */ | |
1647 | if (cluster_in_el == 0) | |
1648 | break; | |
1649 | ||
1650 | /* | |
1651 | * The 'pos' searched for previous extent block is | |
1652 | * always one cluster less than actual trunc_end. | |
1653 | */ | |
1654 | trunc_end = cluster_in_el + 1; | |
1655 | ||
1656 | ocfs2_reinit_path(path, 1); | |
1657 | ||
1658 | continue; | |
1659 | ||
1660 | } else | |
1661 | rec = &el->l_recs[i]; | |
1662 | ||
1663 | ocfs2_calc_trunc_pos(inode, el, rec, trunc_start, &trunc_cpos, | |
1664 | &trunc_len, &trunc_end, &blkno, &done); | |
1665 | if (done) | |
1666 | break; | |
1667 | ||
1668 | flags = rec->e_flags; | |
1669 | phys_cpos = ocfs2_blocks_to_clusters(inode->i_sb, blkno); | |
1670 | ||
1671 | ret = ocfs2_remove_btree_range(inode, &et, trunc_cpos, | |
1672 | phys_cpos, trunc_len, flags, | |
1673 | &dealloc, refcount_loc); | |
1674 | if (ret < 0) { | |
1675 | mlog_errno(ret); | |
1676 | goto out; | |
063c4561 MF |
1677 | } |
1678 | ||
c1631d4a TY |
1679 | cluster_in_el = trunc_end; |
1680 | ||
1681 | ocfs2_reinit_path(path, 1); | |
063c4561 MF |
1682 | } |
1683 | ||
1684 | ocfs2_truncate_cluster_pages(inode, byte_start, byte_len); | |
1685 | ||
1686 | out: | |
1687 | ocfs2_schedule_truncate_log_flush(osb, 1); | |
1688 | ocfs2_run_deallocs(osb, &dealloc); | |
1689 | ||
1690 | return ret; | |
1691 | } | |
1692 | ||
b2580103 MF |
1693 | /* |
1694 | * Parts of this function taken from xfs_change_file_space() | |
1695 | */ | |
385820a3 MF |
1696 | static int __ocfs2_change_file_space(struct file *file, struct inode *inode, |
1697 | loff_t f_pos, unsigned int cmd, | |
1698 | struct ocfs2_space_resv *sr, | |
1699 | int change_size) | |
b2580103 MF |
1700 | { |
1701 | int ret; | |
1702 | s64 llen; | |
385820a3 | 1703 | loff_t size; |
b2580103 MF |
1704 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
1705 | struct buffer_head *di_bh = NULL; | |
1706 | handle_t *handle; | |
a00cce35 | 1707 | unsigned long long max_off = inode->i_sb->s_maxbytes; |
b2580103 | 1708 | |
b2580103 MF |
1709 | if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb)) |
1710 | return -EROFS; | |
1711 | ||
1712 | mutex_lock(&inode->i_mutex); | |
1713 | ||
1714 | /* | |
1715 | * This prevents concurrent writes on other nodes | |
1716 | */ | |
1717 | ret = ocfs2_rw_lock(inode, 1); | |
1718 | if (ret) { | |
1719 | mlog_errno(ret); | |
1720 | goto out; | |
1721 | } | |
1722 | ||
e63aecb6 | 1723 | ret = ocfs2_inode_lock(inode, &di_bh, 1); |
b2580103 MF |
1724 | if (ret) { |
1725 | mlog_errno(ret); | |
1726 | goto out_rw_unlock; | |
1727 | } | |
1728 | ||
1729 | if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) { | |
1730 | ret = -EPERM; | |
e63aecb6 | 1731 | goto out_inode_unlock; |
b2580103 MF |
1732 | } |
1733 | ||
1734 | switch (sr->l_whence) { | |
1735 | case 0: /*SEEK_SET*/ | |
1736 | break; | |
1737 | case 1: /*SEEK_CUR*/ | |
385820a3 | 1738 | sr->l_start += f_pos; |
b2580103 MF |
1739 | break; |
1740 | case 2: /*SEEK_END*/ | |
1741 | sr->l_start += i_size_read(inode); | |
1742 | break; | |
1743 | default: | |
1744 | ret = -EINVAL; | |
e63aecb6 | 1745 | goto out_inode_unlock; |
b2580103 MF |
1746 | } |
1747 | sr->l_whence = 0; | |
1748 | ||
1749 | llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len; | |
1750 | ||
1751 | if (sr->l_start < 0 | |
1752 | || sr->l_start > max_off | |
1753 | || (sr->l_start + llen) < 0 | |
1754 | || (sr->l_start + llen) > max_off) { | |
1755 | ret = -EINVAL; | |
e63aecb6 | 1756 | goto out_inode_unlock; |
b2580103 | 1757 | } |
385820a3 | 1758 | size = sr->l_start + sr->l_len; |
b2580103 MF |
1759 | |
1760 | if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) { | |
1761 | if (sr->l_len <= 0) { | |
1762 | ret = -EINVAL; | |
e63aecb6 | 1763 | goto out_inode_unlock; |
b2580103 MF |
1764 | } |
1765 | } | |
1766 | ||
385820a3 | 1767 | if (file && should_remove_suid(file->f_path.dentry)) { |
b2580103 MF |
1768 | ret = __ocfs2_write_remove_suid(inode, di_bh); |
1769 | if (ret) { | |
1770 | mlog_errno(ret); | |
e63aecb6 | 1771 | goto out_inode_unlock; |
b2580103 MF |
1772 | } |
1773 | } | |
1774 | ||
1775 | down_write(&OCFS2_I(inode)->ip_alloc_sem); | |
1776 | switch (cmd) { | |
1777 | case OCFS2_IOC_RESVSP: | |
1778 | case OCFS2_IOC_RESVSP64: | |
1779 | /* | |
1780 | * This takes unsigned offsets, but the signed ones we | |
1781 | * pass have been checked against overflow above. | |
1782 | */ | |
1783 | ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start, | |
1784 | sr->l_len); | |
1785 | break; | |
1786 | case OCFS2_IOC_UNRESVSP: | |
1787 | case OCFS2_IOC_UNRESVSP64: | |
1788 | ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start, | |
1789 | sr->l_len); | |
1790 | break; | |
1791 | default: | |
1792 | ret = -EINVAL; | |
1793 | } | |
1794 | up_write(&OCFS2_I(inode)->ip_alloc_sem); | |
1795 | if (ret) { | |
1796 | mlog_errno(ret); | |
e63aecb6 | 1797 | goto out_inode_unlock; |
b2580103 MF |
1798 | } |
1799 | ||
1800 | /* | |
1801 | * We update c/mtime for these changes | |
1802 | */ | |
1803 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); | |
1804 | if (IS_ERR(handle)) { | |
1805 | ret = PTR_ERR(handle); | |
1806 | mlog_errno(ret); | |
e63aecb6 | 1807 | goto out_inode_unlock; |
b2580103 MF |
1808 | } |
1809 | ||
385820a3 MF |
1810 | if (change_size && i_size_read(inode) < size) |
1811 | i_size_write(inode, size); | |
1812 | ||
b2580103 MF |
1813 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; |
1814 | ret = ocfs2_mark_inode_dirty(handle, inode, di_bh); | |
1815 | if (ret < 0) | |
1816 | mlog_errno(ret); | |
1817 | ||
1818 | ocfs2_commit_trans(osb, handle); | |
1819 | ||
e63aecb6 | 1820 | out_inode_unlock: |
b2580103 | 1821 | brelse(di_bh); |
e63aecb6 | 1822 | ocfs2_inode_unlock(inode, 1); |
b2580103 MF |
1823 | out_rw_unlock: |
1824 | ocfs2_rw_unlock(inode, 1); | |
1825 | ||
b2580103 | 1826 | out: |
c259ae52 | 1827 | mutex_unlock(&inode->i_mutex); |
b2580103 MF |
1828 | return ret; |
1829 | } | |
1830 | ||
385820a3 MF |
1831 | int ocfs2_change_file_space(struct file *file, unsigned int cmd, |
1832 | struct ocfs2_space_resv *sr) | |
1833 | { | |
1834 | struct inode *inode = file->f_path.dentry->d_inode; | |
c19a28e1 | 1835 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
385820a3 MF |
1836 | |
1837 | if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) && | |
1838 | !ocfs2_writes_unwritten_extents(osb)) | |
1839 | return -ENOTTY; | |
1840 | else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) && | |
1841 | !ocfs2_sparse_alloc(osb)) | |
1842 | return -ENOTTY; | |
1843 | ||
1844 | if (!S_ISREG(inode->i_mode)) | |
1845 | return -EINVAL; | |
1846 | ||
1847 | if (!(file->f_mode & FMODE_WRITE)) | |
1848 | return -EBADF; | |
1849 | ||
1850 | return __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0); | |
1851 | } | |
1852 | ||
1853 | static long ocfs2_fallocate(struct inode *inode, int mode, loff_t offset, | |
1854 | loff_t len) | |
1855 | { | |
1856 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1857 | struct ocfs2_space_resv sr; | |
1858 | int change_size = 1; | |
1859 | ||
1860 | if (!ocfs2_writes_unwritten_extents(osb)) | |
1861 | return -EOPNOTSUPP; | |
1862 | ||
1863 | if (S_ISDIR(inode->i_mode)) | |
1864 | return -ENODEV; | |
1865 | ||
1866 | if (mode & FALLOC_FL_KEEP_SIZE) | |
1867 | change_size = 0; | |
1868 | ||
1869 | sr.l_whence = 0; | |
1870 | sr.l_start = (s64)offset; | |
1871 | sr.l_len = (s64)len; | |
1872 | ||
1873 | return __ocfs2_change_file_space(NULL, inode, offset, | |
1874 | OCFS2_IOC_RESVSP64, &sr, change_size); | |
1875 | } | |
1876 | ||
293b2f70 TM |
1877 | int ocfs2_check_range_for_refcount(struct inode *inode, loff_t pos, |
1878 | size_t count) | |
1879 | { | |
1880 | int ret = 0; | |
1881 | unsigned int extent_flags; | |
1882 | u32 cpos, clusters, extent_len, phys_cpos; | |
1883 | struct super_block *sb = inode->i_sb; | |
1884 | ||
1885 | if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)) || | |
2f48d593 TM |
1886 | !(OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) || |
1887 | OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) | |
293b2f70 TM |
1888 | return 0; |
1889 | ||
1890 | cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits; | |
1891 | clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos; | |
1892 | ||
1893 | while (clusters) { | |
1894 | ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len, | |
1895 | &extent_flags); | |
1896 | if (ret < 0) { | |
1897 | mlog_errno(ret); | |
1898 | goto out; | |
1899 | } | |
1900 | ||
1901 | if (phys_cpos && (extent_flags & OCFS2_EXT_REFCOUNTED)) { | |
1902 | ret = 1; | |
1903 | break; | |
1904 | } | |
1905 | ||
1906 | if (extent_len > clusters) | |
1907 | extent_len = clusters; | |
1908 | ||
1909 | clusters -= extent_len; | |
1910 | cpos += extent_len; | |
1911 | } | |
1912 | out: | |
1913 | return ret; | |
1914 | } | |
1915 | ||
1916 | static int ocfs2_prepare_inode_for_refcount(struct inode *inode, | |
1917 | loff_t pos, size_t count, | |
1918 | int *meta_level) | |
1919 | { | |
1920 | int ret; | |
1921 | struct buffer_head *di_bh = NULL; | |
1922 | u32 cpos = pos >> OCFS2_SB(inode->i_sb)->s_clustersize_bits; | |
1923 | u32 clusters = | |
1924 | ocfs2_clusters_for_bytes(inode->i_sb, pos + count) - cpos; | |
1925 | ||
1926 | ret = ocfs2_inode_lock(inode, &di_bh, 1); | |
1927 | if (ret) { | |
1928 | mlog_errno(ret); | |
1929 | goto out; | |
1930 | } | |
1931 | ||
1932 | *meta_level = 1; | |
1933 | ||
37f8a2bf | 1934 | ret = ocfs2_refcount_cow(inode, di_bh, cpos, clusters, UINT_MAX); |
293b2f70 TM |
1935 | if (ret) |
1936 | mlog_errno(ret); | |
1937 | out: | |
1938 | brelse(di_bh); | |
1939 | return ret; | |
1940 | } | |
1941 | ||
8659ac25 TY |
1942 | static int ocfs2_prepare_inode_for_write(struct dentry *dentry, |
1943 | loff_t *ppos, | |
1944 | size_t count, | |
9517bac6 | 1945 | int appending, |
86470e98 TM |
1946 | int *direct_io, |
1947 | int *has_refcount) | |
ccd979bd | 1948 | { |
65ed39d6 | 1949 | int ret = 0, meta_level = 0; |
8659ac25 | 1950 | struct inode *inode = dentry->d_inode; |
65ed39d6 | 1951 | loff_t saved_pos, end; |
ccd979bd | 1952 | |
2bd63216 | 1953 | /* |
65ed39d6 MF |
1954 | * We start with a read level meta lock and only jump to an ex |
1955 | * if we need to make modifications here. | |
ccd979bd | 1956 | */ |
ccd979bd | 1957 | for(;;) { |
e63aecb6 | 1958 | ret = ocfs2_inode_lock(inode, NULL, meta_level); |
ccd979bd MF |
1959 | if (ret < 0) { |
1960 | meta_level = -1; | |
1961 | mlog_errno(ret); | |
1962 | goto out; | |
1963 | } | |
1964 | ||
1965 | /* Clear suid / sgid if necessary. We do this here | |
1966 | * instead of later in the write path because | |
1967 | * remove_suid() calls ->setattr without any hint that | |
1968 | * we may have already done our cluster locking. Since | |
1969 | * ocfs2_setattr() *must* take cluster locks to | |
1970 | * proceeed, this will lead us to recursively lock the | |
1971 | * inode. There's also the dinode i_size state which | |
1972 | * can be lost via setattr during extending writes (we | |
1973 | * set inode->i_size at the end of a write. */ | |
8659ac25 | 1974 | if (should_remove_suid(dentry)) { |
ccd979bd | 1975 | if (meta_level == 0) { |
e63aecb6 | 1976 | ocfs2_inode_unlock(inode, meta_level); |
ccd979bd MF |
1977 | meta_level = 1; |
1978 | continue; | |
1979 | } | |
1980 | ||
1981 | ret = ocfs2_write_remove_suid(inode); | |
1982 | if (ret < 0) { | |
1983 | mlog_errno(ret); | |
8659ac25 | 1984 | goto out_unlock; |
ccd979bd MF |
1985 | } |
1986 | } | |
1987 | ||
1988 | /* work on a copy of ppos until we're sure that we won't have | |
1989 | * to recalculate it due to relocking. */ | |
8659ac25 | 1990 | if (appending) { |
ccd979bd MF |
1991 | saved_pos = i_size_read(inode); |
1992 | mlog(0, "O_APPEND: inode->i_size=%llu\n", saved_pos); | |
1993 | } else { | |
8659ac25 | 1994 | saved_pos = *ppos; |
ccd979bd | 1995 | } |
3a0782d0 | 1996 | |
65ed39d6 | 1997 | end = saved_pos + count; |
9517bac6 | 1998 | |
293b2f70 TM |
1999 | ret = ocfs2_check_range_for_refcount(inode, saved_pos, count); |
2000 | if (ret == 1) { | |
2001 | ocfs2_inode_unlock(inode, meta_level); | |
2002 | meta_level = -1; | |
2003 | ||
2004 | ret = ocfs2_prepare_inode_for_refcount(inode, | |
2005 | saved_pos, | |
2006 | count, | |
2007 | &meta_level); | |
86470e98 TM |
2008 | if (has_refcount) |
2009 | *has_refcount = 1; | |
96a1cc73 WW |
2010 | if (direct_io) |
2011 | *direct_io = 0; | |
293b2f70 TM |
2012 | } |
2013 | ||
2014 | if (ret < 0) { | |
2015 | mlog_errno(ret); | |
2016 | goto out_unlock; | |
2017 | } | |
2018 | ||
65ed39d6 MF |
2019 | /* |
2020 | * Skip the O_DIRECT checks if we don't need | |
2021 | * them. | |
2022 | */ | |
2023 | if (!direct_io || !(*direct_io)) | |
9517bac6 | 2024 | break; |
9517bac6 | 2025 | |
1afc32b9 MF |
2026 | /* |
2027 | * There's no sane way to do direct writes to an inode | |
2028 | * with inline data. | |
2029 | */ | |
2030 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) { | |
2031 | *direct_io = 0; | |
2032 | break; | |
2033 | } | |
2034 | ||
3a0782d0 | 2035 | /* |
65ed39d6 MF |
2036 | * Allowing concurrent direct writes means |
2037 | * i_size changes wouldn't be synchronized, so | |
2038 | * one node could wind up truncating another | |
2039 | * nodes writes. | |
3a0782d0 | 2040 | */ |
65ed39d6 MF |
2041 | if (end > i_size_read(inode)) { |
2042 | *direct_io = 0; | |
ccd979bd | 2043 | break; |
ccd979bd MF |
2044 | } |
2045 | ||
65ed39d6 MF |
2046 | /* |
2047 | * We don't fill holes during direct io, so | |
2048 | * check for them here. If any are found, the | |
2049 | * caller will have to retake some cluster | |
2050 | * locks and initiate the io as buffered. | |
2051 | */ | |
2052 | ret = ocfs2_check_range_for_holes(inode, saved_pos, count); | |
2053 | if (ret == 1) { | |
2054 | *direct_io = 0; | |
2055 | ret = 0; | |
2056 | } else if (ret < 0) | |
2057 | mlog_errno(ret); | |
ccd979bd MF |
2058 | break; |
2059 | } | |
2060 | ||
8659ac25 TY |
2061 | if (appending) |
2062 | *ppos = saved_pos; | |
2063 | ||
2064 | out_unlock: | |
293b2f70 TM |
2065 | if (meta_level >= 0) |
2066 | ocfs2_inode_unlock(inode, meta_level); | |
8659ac25 TY |
2067 | |
2068 | out: | |
2069 | return ret; | |
2070 | } | |
2071 | ||
2072 | static ssize_t ocfs2_file_aio_write(struct kiocb *iocb, | |
2073 | const struct iovec *iov, | |
2074 | unsigned long nr_segs, | |
2075 | loff_t pos) | |
2076 | { | |
9517bac6 | 2077 | int ret, direct_io, appending, rw_level, have_alloc_sem = 0; |
86470e98 | 2078 | int can_do_direct, has_refcount = 0; |
9517bac6 MF |
2079 | ssize_t written = 0; |
2080 | size_t ocount; /* original count */ | |
2081 | size_t count; /* after file limit checks */ | |
9ea2d32f MF |
2082 | loff_t old_size, *ppos = &iocb->ki_pos; |
2083 | u32 old_clusters; | |
9517bac6 MF |
2084 | struct file *file = iocb->ki_filp; |
2085 | struct inode *inode = file->f_path.dentry->d_inode; | |
9ea2d32f | 2086 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
9517bac6 MF |
2087 | |
2088 | mlog_entry("(0x%p, %u, '%.*s')\n", file, | |
8659ac25 | 2089 | (unsigned int)nr_segs, |
9517bac6 MF |
2090 | file->f_path.dentry->d_name.len, |
2091 | file->f_path.dentry->d_name.name); | |
8659ac25 | 2092 | |
8659ac25 TY |
2093 | if (iocb->ki_left == 0) |
2094 | return 0; | |
2095 | ||
9517bac6 MF |
2096 | vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE); |
2097 | ||
2098 | appending = file->f_flags & O_APPEND ? 1 : 0; | |
2099 | direct_io = file->f_flags & O_DIRECT ? 1 : 0; | |
2100 | ||
8659ac25 | 2101 | mutex_lock(&inode->i_mutex); |
9517bac6 MF |
2102 | |
2103 | relock: | |
8659ac25 | 2104 | /* to match setattr's i_mutex -> i_alloc_sem -> rw_lock ordering */ |
9517bac6 | 2105 | if (direct_io) { |
8659ac25 | 2106 | down_read(&inode->i_alloc_sem); |
9517bac6 | 2107 | have_alloc_sem = 1; |
8659ac25 TY |
2108 | } |
2109 | ||
2110 | /* concurrent O_DIRECT writes are allowed */ | |
9517bac6 | 2111 | rw_level = !direct_io; |
8659ac25 TY |
2112 | ret = ocfs2_rw_lock(inode, rw_level); |
2113 | if (ret < 0) { | |
8659ac25 | 2114 | mlog_errno(ret); |
9517bac6 | 2115 | goto out_sems; |
8659ac25 TY |
2116 | } |
2117 | ||
9517bac6 MF |
2118 | can_do_direct = direct_io; |
2119 | ret = ocfs2_prepare_inode_for_write(file->f_path.dentry, ppos, | |
2120 | iocb->ki_left, appending, | |
86470e98 | 2121 | &can_do_direct, &has_refcount); |
8659ac25 TY |
2122 | if (ret < 0) { |
2123 | mlog_errno(ret); | |
2124 | goto out; | |
2125 | } | |
ccd979bd | 2126 | |
9517bac6 MF |
2127 | /* |
2128 | * We can't complete the direct I/O as requested, fall back to | |
2129 | * buffered I/O. | |
2130 | */ | |
2131 | if (direct_io && !can_do_direct) { | |
2132 | ocfs2_rw_unlock(inode, rw_level); | |
2133 | up_read(&inode->i_alloc_sem); | |
2134 | ||
2135 | have_alloc_sem = 0; | |
2136 | rw_level = -1; | |
2137 | ||
2138 | direct_io = 0; | |
9517bac6 MF |
2139 | goto relock; |
2140 | } | |
2141 | ||
9ea2d32f MF |
2142 | /* |
2143 | * To later detect whether a journal commit for sync writes is | |
2144 | * necessary, we sample i_size, and cluster count here. | |
2145 | */ | |
2146 | old_size = i_size_read(inode); | |
2147 | old_clusters = OCFS2_I(inode)->ip_clusters; | |
2148 | ||
ccd979bd | 2149 | /* communicate with ocfs2_dio_end_io */ |
7cdfc3a1 | 2150 | ocfs2_iocb_set_rw_locked(iocb, rw_level); |
ccd979bd | 2151 | |
6b933c8e LD |
2152 | ret = generic_segment_checks(iov, &nr_segs, &ocount, |
2153 | VERIFY_READ); | |
2154 | if (ret) | |
2155 | goto out_dio; | |
b6af1bcd | 2156 | |
6b933c8e LD |
2157 | count = ocount; |
2158 | ret = generic_write_checks(file, ppos, &count, | |
2159 | S_ISBLK(inode->i_mode)); | |
2160 | if (ret) | |
2161 | goto out_dio; | |
b6af1bcd | 2162 | |
6b933c8e | 2163 | if (direct_io) { |
9517bac6 MF |
2164 | written = generic_file_direct_write(iocb, iov, &nr_segs, *ppos, |
2165 | ppos, count, ocount); | |
2166 | if (written < 0) { | |
c4354001 DM |
2167 | /* |
2168 | * direct write may have instantiated a few | |
2169 | * blocks outside i_size. Trim these off again. | |
2170 | * Don't need i_size_read because we hold i_mutex. | |
15c6fd97 | 2171 | * |
2172 | * XXX(hch): this looks buggy because ocfs2 did not | |
2173 | * actually implement ->truncate. Take a look at | |
2174 | * the new truncate sequence and update this accordingly | |
c4354001 DM |
2175 | */ |
2176 | if (*ppos + count > inode->i_size) | |
15c6fd97 | 2177 | simple_setsize(inode, inode->i_size); |
9517bac6 MF |
2178 | ret = written; |
2179 | goto out_dio; | |
2180 | } | |
2181 | } else { | |
6b933c8e LD |
2182 | current->backing_dev_info = file->f_mapping->backing_dev_info; |
2183 | written = generic_file_buffered_write(iocb, iov, nr_segs, *ppos, | |
2184 | ppos, count, 0); | |
2185 | current->backing_dev_info = NULL; | |
9517bac6 | 2186 | } |
ccd979bd | 2187 | |
9517bac6 | 2188 | out_dio: |
ccd979bd | 2189 | /* buffered aio wouldn't have proper lock coverage today */ |
9517bac6 | 2190 | BUG_ON(ret == -EIOCBQUEUED && !(file->f_flags & O_DIRECT)); |
ccd979bd | 2191 | |
60c48674 TM |
2192 | if (((file->f_flags & O_DSYNC) && !direct_io) || IS_SYNC(inode) || |
2193 | ((file->f_flags & O_DIRECT) && has_refcount)) { | |
918941a3 JK |
2194 | ret = filemap_fdatawrite_range(file->f_mapping, pos, |
2195 | pos + count - 1); | |
2196 | if (ret < 0) | |
2197 | written = ret; | |
2198 | ||
a03ab788 CL |
2199 | if (!ret && ((old_size != i_size_read(inode)) || |
2200 | (old_clusters != OCFS2_I(inode)->ip_clusters) || | |
2201 | has_refcount)) { | |
2b4e30fb | 2202 | ret = jbd2_journal_force_commit(osb->journal->j_journal); |
9ea2d32f MF |
2203 | if (ret < 0) |
2204 | written = ret; | |
2205 | } | |
918941a3 JK |
2206 | |
2207 | if (!ret) | |
2208 | ret = filemap_fdatawait_range(file->f_mapping, pos, | |
2209 | pos + count - 1); | |
9ea2d32f MF |
2210 | } |
2211 | ||
2bd63216 | 2212 | /* |
ccd979bd MF |
2213 | * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io |
2214 | * function pointer which is called when o_direct io completes so that | |
2215 | * it can unlock our rw lock. (it's the clustered equivalent of | |
2216 | * i_alloc_sem; protects truncate from racing with pending ios). | |
2217 | * Unfortunately there are error cases which call end_io and others | |
2218 | * that don't. so we don't have to unlock the rw_lock if either an | |
2219 | * async dio is going to do it in the future or an end_io after an | |
2220 | * error has already done it. | |
2221 | */ | |
66b116c9 | 2222 | if ((ret == -EIOCBQUEUED) || (!ocfs2_iocb_is_rw_locked(iocb))) { |
ccd979bd MF |
2223 | rw_level = -1; |
2224 | have_alloc_sem = 0; | |
2225 | } | |
2226 | ||
2227 | out: | |
9517bac6 MF |
2228 | if (rw_level != -1) |
2229 | ocfs2_rw_unlock(inode, rw_level); | |
2230 | ||
2231 | out_sems: | |
ccd979bd MF |
2232 | if (have_alloc_sem) |
2233 | up_read(&inode->i_alloc_sem); | |
9517bac6 | 2234 | |
1b1dcc1b | 2235 | mutex_unlock(&inode->i_mutex); |
ccd979bd | 2236 | |
812e7a6a WW |
2237 | if (written) |
2238 | ret = written; | |
ccd979bd | 2239 | mlog_exit(ret); |
812e7a6a | 2240 | return ret; |
ccd979bd MF |
2241 | } |
2242 | ||
328eaaba MS |
2243 | static int ocfs2_splice_to_file(struct pipe_inode_info *pipe, |
2244 | struct file *out, | |
2245 | struct splice_desc *sd) | |
2246 | { | |
2247 | int ret; | |
2248 | ||
2249 | ret = ocfs2_prepare_inode_for_write(out->f_path.dentry, &sd->pos, | |
86470e98 | 2250 | sd->total_len, 0, NULL, NULL); |
328eaaba MS |
2251 | if (ret < 0) { |
2252 | mlog_errno(ret); | |
2253 | return ret; | |
2254 | } | |
2255 | ||
2256 | return splice_from_pipe_feed(pipe, sd, pipe_to_file); | |
2257 | } | |
2258 | ||
8659ac25 TY |
2259 | static ssize_t ocfs2_file_splice_write(struct pipe_inode_info *pipe, |
2260 | struct file *out, | |
2261 | loff_t *ppos, | |
2262 | size_t len, | |
2263 | unsigned int flags) | |
2264 | { | |
2265 | int ret; | |
328eaaba MS |
2266 | struct address_space *mapping = out->f_mapping; |
2267 | struct inode *inode = mapping->host; | |
2268 | struct splice_desc sd = { | |
2269 | .total_len = len, | |
2270 | .flags = flags, | |
2271 | .pos = *ppos, | |
2272 | .u.file = out, | |
2273 | }; | |
8659ac25 TY |
2274 | |
2275 | mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", out, pipe, | |
2276 | (unsigned int)len, | |
d28c9174 JS |
2277 | out->f_path.dentry->d_name.len, |
2278 | out->f_path.dentry->d_name.name); | |
8659ac25 | 2279 | |
328eaaba MS |
2280 | if (pipe->inode) |
2281 | mutex_lock_nested(&pipe->inode->i_mutex, I_MUTEX_PARENT); | |
8659ac25 | 2282 | |
328eaaba MS |
2283 | splice_from_pipe_begin(&sd); |
2284 | do { | |
2285 | ret = splice_from_pipe_next(pipe, &sd); | |
2286 | if (ret <= 0) | |
2287 | break; | |
8659ac25 | 2288 | |
328eaaba MS |
2289 | mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD); |
2290 | ret = ocfs2_rw_lock(inode, 1); | |
2291 | if (ret < 0) | |
2292 | mlog_errno(ret); | |
2293 | else { | |
2294 | ret = ocfs2_splice_to_file(pipe, out, &sd); | |
2295 | ocfs2_rw_unlock(inode, 1); | |
2296 | } | |
2297 | mutex_unlock(&inode->i_mutex); | |
2298 | } while (ret > 0); | |
2299 | splice_from_pipe_end(pipe, &sd); | |
8659ac25 | 2300 | |
7bfac9ec MS |
2301 | if (pipe->inode) |
2302 | mutex_unlock(&pipe->inode->i_mutex); | |
8659ac25 | 2303 | |
328eaaba MS |
2304 | if (sd.num_spliced) |
2305 | ret = sd.num_spliced; | |
2306 | ||
2307 | if (ret > 0) { | |
2308 | unsigned long nr_pages; | |
d23c937b | 2309 | int err; |
328eaaba | 2310 | |
328eaaba MS |
2311 | nr_pages = (ret + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; |
2312 | ||
d23c937b JK |
2313 | err = generic_write_sync(out, *ppos, ret); |
2314 | if (err) | |
2315 | ret = err; | |
2316 | else | |
2317 | *ppos += ret; | |
328eaaba | 2318 | |
328eaaba MS |
2319 | balance_dirty_pages_ratelimited_nr(mapping, nr_pages); |
2320 | } | |
8659ac25 TY |
2321 | |
2322 | mlog_exit(ret); | |
2323 | return ret; | |
2324 | } | |
2325 | ||
2326 | static ssize_t ocfs2_file_splice_read(struct file *in, | |
2327 | loff_t *ppos, | |
2328 | struct pipe_inode_info *pipe, | |
2329 | size_t len, | |
2330 | unsigned int flags) | |
2331 | { | |
1962f39a | 2332 | int ret = 0, lock_level = 0; |
d28c9174 | 2333 | struct inode *inode = in->f_path.dentry->d_inode; |
8659ac25 TY |
2334 | |
2335 | mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", in, pipe, | |
2336 | (unsigned int)len, | |
d28c9174 JS |
2337 | in->f_path.dentry->d_name.len, |
2338 | in->f_path.dentry->d_name.name); | |
8659ac25 TY |
2339 | |
2340 | /* | |
2341 | * See the comment in ocfs2_file_aio_read() | |
2342 | */ | |
1962f39a | 2343 | ret = ocfs2_inode_lock_atime(inode, in->f_vfsmnt, &lock_level); |
8659ac25 TY |
2344 | if (ret < 0) { |
2345 | mlog_errno(ret); | |
2346 | goto bail; | |
2347 | } | |
1962f39a | 2348 | ocfs2_inode_unlock(inode, lock_level); |
8659ac25 TY |
2349 | |
2350 | ret = generic_file_splice_read(in, ppos, pipe, len, flags); | |
2351 | ||
2352 | bail: | |
2353 | mlog_exit(ret); | |
2354 | return ret; | |
2355 | } | |
2356 | ||
ccd979bd | 2357 | static ssize_t ocfs2_file_aio_read(struct kiocb *iocb, |
027445c3 BP |
2358 | const struct iovec *iov, |
2359 | unsigned long nr_segs, | |
ccd979bd MF |
2360 | loff_t pos) |
2361 | { | |
25899dee | 2362 | int ret = 0, rw_level = -1, have_alloc_sem = 0, lock_level = 0; |
ccd979bd | 2363 | struct file *filp = iocb->ki_filp; |
d28c9174 | 2364 | struct inode *inode = filp->f_path.dentry->d_inode; |
ccd979bd | 2365 | |
027445c3 BP |
2366 | mlog_entry("(0x%p, %u, '%.*s')\n", filp, |
2367 | (unsigned int)nr_segs, | |
d28c9174 JS |
2368 | filp->f_path.dentry->d_name.len, |
2369 | filp->f_path.dentry->d_name.name); | |
ccd979bd MF |
2370 | |
2371 | if (!inode) { | |
2372 | ret = -EINVAL; | |
2373 | mlog_errno(ret); | |
2374 | goto bail; | |
2375 | } | |
2376 | ||
2bd63216 | 2377 | /* |
ccd979bd MF |
2378 | * buffered reads protect themselves in ->readpage(). O_DIRECT reads |
2379 | * need locks to protect pending reads from racing with truncate. | |
2380 | */ | |
2381 | if (filp->f_flags & O_DIRECT) { | |
2382 | down_read(&inode->i_alloc_sem); | |
2383 | have_alloc_sem = 1; | |
2384 | ||
2385 | ret = ocfs2_rw_lock(inode, 0); | |
2386 | if (ret < 0) { | |
2387 | mlog_errno(ret); | |
2388 | goto bail; | |
2389 | } | |
2390 | rw_level = 0; | |
2391 | /* communicate with ocfs2_dio_end_io */ | |
7cdfc3a1 | 2392 | ocfs2_iocb_set_rw_locked(iocb, rw_level); |
ccd979bd MF |
2393 | } |
2394 | ||
c4374f8a MF |
2395 | /* |
2396 | * We're fine letting folks race truncates and extending | |
2397 | * writes with read across the cluster, just like they can | |
2398 | * locally. Hence no rw_lock during read. | |
2bd63216 | 2399 | * |
c4374f8a MF |
2400 | * Take and drop the meta data lock to update inode fields |
2401 | * like i_size. This allows the checks down below | |
2bd63216 | 2402 | * generic_file_aio_read() a chance of actually working. |
c4374f8a | 2403 | */ |
e63aecb6 | 2404 | ret = ocfs2_inode_lock_atime(inode, filp->f_vfsmnt, &lock_level); |
c4374f8a MF |
2405 | if (ret < 0) { |
2406 | mlog_errno(ret); | |
2407 | goto bail; | |
2408 | } | |
e63aecb6 | 2409 | ocfs2_inode_unlock(inode, lock_level); |
c4374f8a | 2410 | |
027445c3 | 2411 | ret = generic_file_aio_read(iocb, iov, nr_segs, iocb->ki_pos); |
ccd979bd | 2412 | if (ret == -EINVAL) |
56753bd3 | 2413 | mlog(0, "generic_file_aio_read returned -EINVAL\n"); |
ccd979bd MF |
2414 | |
2415 | /* buffered aio wouldn't have proper lock coverage today */ | |
2416 | BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT)); | |
2417 | ||
2418 | /* see ocfs2_file_aio_write */ | |
2419 | if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) { | |
2420 | rw_level = -1; | |
2421 | have_alloc_sem = 0; | |
2422 | } | |
2423 | ||
2424 | bail: | |
2425 | if (have_alloc_sem) | |
2426 | up_read(&inode->i_alloc_sem); | |
2bd63216 | 2427 | if (rw_level != -1) |
ccd979bd MF |
2428 | ocfs2_rw_unlock(inode, rw_level); |
2429 | mlog_exit(ret); | |
2430 | ||
2431 | return ret; | |
2432 | } | |
2433 | ||
92e1d5be | 2434 | const struct inode_operations ocfs2_file_iops = { |
ccd979bd MF |
2435 | .setattr = ocfs2_setattr, |
2436 | .getattr = ocfs2_getattr, | |
d38eb8db | 2437 | .permission = ocfs2_permission, |
cf1d6c76 TY |
2438 | .setxattr = generic_setxattr, |
2439 | .getxattr = generic_getxattr, | |
2440 | .listxattr = ocfs2_listxattr, | |
2441 | .removexattr = generic_removexattr, | |
385820a3 | 2442 | .fallocate = ocfs2_fallocate, |
00dc417f | 2443 | .fiemap = ocfs2_fiemap, |
ccd979bd MF |
2444 | }; |
2445 | ||
92e1d5be | 2446 | const struct inode_operations ocfs2_special_file_iops = { |
ccd979bd MF |
2447 | .setattr = ocfs2_setattr, |
2448 | .getattr = ocfs2_getattr, | |
d38eb8db | 2449 | .permission = ocfs2_permission, |
ccd979bd MF |
2450 | }; |
2451 | ||
53da4939 MF |
2452 | /* |
2453 | * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with | |
2454 | * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks! | |
2455 | */ | |
4b6f5d20 | 2456 | const struct file_operations ocfs2_fops = { |
32c3c0e2 | 2457 | .llseek = generic_file_llseek, |
ccd979bd MF |
2458 | .read = do_sync_read, |
2459 | .write = do_sync_write, | |
ccd979bd MF |
2460 | .mmap = ocfs2_mmap, |
2461 | .fsync = ocfs2_sync_file, | |
2462 | .release = ocfs2_file_release, | |
2463 | .open = ocfs2_file_open, | |
2464 | .aio_read = ocfs2_file_aio_read, | |
2465 | .aio_write = ocfs2_file_aio_write, | |
c9ec1488 | 2466 | .unlocked_ioctl = ocfs2_ioctl, |
586d232b MF |
2467 | #ifdef CONFIG_COMPAT |
2468 | .compat_ioctl = ocfs2_compat_ioctl, | |
2469 | #endif | |
53da4939 | 2470 | .lock = ocfs2_lock, |
53fc622b | 2471 | .flock = ocfs2_flock, |
8659ac25 TY |
2472 | .splice_read = ocfs2_file_splice_read, |
2473 | .splice_write = ocfs2_file_splice_write, | |
ccd979bd MF |
2474 | }; |
2475 | ||
4b6f5d20 | 2476 | const struct file_operations ocfs2_dops = { |
32c3c0e2 | 2477 | .llseek = generic_file_llseek, |
ccd979bd MF |
2478 | .read = generic_read_dir, |
2479 | .readdir = ocfs2_readdir, | |
2480 | .fsync = ocfs2_sync_file, | |
53fc622b MF |
2481 | .release = ocfs2_dir_release, |
2482 | .open = ocfs2_dir_open, | |
c9ec1488 | 2483 | .unlocked_ioctl = ocfs2_ioctl, |
586d232b MF |
2484 | #ifdef CONFIG_COMPAT |
2485 | .compat_ioctl = ocfs2_compat_ioctl, | |
53da4939 MF |
2486 | #endif |
2487 | .lock = ocfs2_lock, | |
2488 | .flock = ocfs2_flock, | |
2489 | }; | |
2490 | ||
2491 | /* | |
2492 | * POSIX-lockless variants of our file_operations. | |
2493 | * | |
2494 | * These will be used if the underlying cluster stack does not support | |
2495 | * posix file locking, if the user passes the "localflocks" mount | |
2496 | * option, or if we have a local-only fs. | |
2497 | * | |
2498 | * ocfs2_flock is in here because all stacks handle UNIX file locks, | |
2499 | * so we still want it in the case of no stack support for | |
2500 | * plocks. Internally, it will do the right thing when asked to ignore | |
2501 | * the cluster. | |
2502 | */ | |
2503 | const struct file_operations ocfs2_fops_no_plocks = { | |
2504 | .llseek = generic_file_llseek, | |
2505 | .read = do_sync_read, | |
2506 | .write = do_sync_write, | |
2507 | .mmap = ocfs2_mmap, | |
2508 | .fsync = ocfs2_sync_file, | |
2509 | .release = ocfs2_file_release, | |
2510 | .open = ocfs2_file_open, | |
2511 | .aio_read = ocfs2_file_aio_read, | |
2512 | .aio_write = ocfs2_file_aio_write, | |
2513 | .unlocked_ioctl = ocfs2_ioctl, | |
2514 | #ifdef CONFIG_COMPAT | |
2515 | .compat_ioctl = ocfs2_compat_ioctl, | |
2516 | #endif | |
2517 | .flock = ocfs2_flock, | |
2518 | .splice_read = ocfs2_file_splice_read, | |
2519 | .splice_write = ocfs2_file_splice_write, | |
2520 | }; | |
2521 | ||
2522 | const struct file_operations ocfs2_dops_no_plocks = { | |
2523 | .llseek = generic_file_llseek, | |
2524 | .read = generic_read_dir, | |
2525 | .readdir = ocfs2_readdir, | |
2526 | .fsync = ocfs2_sync_file, | |
2527 | .release = ocfs2_dir_release, | |
2528 | .open = ocfs2_dir_open, | |
2529 | .unlocked_ioctl = ocfs2_ioctl, | |
2530 | #ifdef CONFIG_COMPAT | |
2531 | .compat_ioctl = ocfs2_compat_ioctl, | |
586d232b | 2532 | #endif |
53fc622b | 2533 | .flock = ocfs2_flock, |
ccd979bd | 2534 | }; |