]>
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> |
04eda1a1 | 39 | #include <linux/blkdev.h> |
66114cad | 40 | #include <linux/backing-dev.h> |
ccd979bd | 41 | |
ccd979bd MF |
42 | #include <cluster/masklog.h> |
43 | ||
44 | #include "ocfs2.h" | |
45 | ||
46 | #include "alloc.h" | |
47 | #include "aops.h" | |
48 | #include "dir.h" | |
49 | #include "dlmglue.h" | |
50 | #include "extent_map.h" | |
51 | #include "file.h" | |
52 | #include "sysfile.h" | |
53 | #include "inode.h" | |
ca4d147e | 54 | #include "ioctl.h" |
ccd979bd | 55 | #include "journal.h" |
53fc622b | 56 | #include "locks.h" |
ccd979bd MF |
57 | #include "mmap.h" |
58 | #include "suballoc.h" | |
59 | #include "super.h" | |
cf1d6c76 | 60 | #include "xattr.h" |
23fc2702 | 61 | #include "acl.h" |
a90714c1 | 62 | #include "quota.h" |
293b2f70 | 63 | #include "refcounttree.h" |
468eedde | 64 | #include "ocfs2_trace.h" |
ccd979bd MF |
65 | |
66 | #include "buffer_head_io.h" | |
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 | ||
468eedde | 103 | trace_ocfs2_file_open(inode, file, file->f_path.dentry, |
d324cd4c | 104 | (unsigned long long)oi->ip_blkno, |
468eedde TM |
105 | file->f_path.dentry->d_name.len, |
106 | file->f_path.dentry->d_name.name, mode); | |
ccd979bd | 107 | |
9c89fe0a JK |
108 | if (file->f_mode & FMODE_WRITE) { |
109 | status = dquot_initialize(inode); | |
110 | if (status) | |
111 | goto leave; | |
112 | } | |
907f4554 | 113 | |
ccd979bd MF |
114 | spin_lock(&oi->ip_lock); |
115 | ||
116 | /* Check that the inode hasn't been wiped from disk by another | |
117 | * node. If it hasn't then we're safe as long as we hold the | |
118 | * spin lock until our increment of open count. */ | |
d324cd4c | 119 | if (oi->ip_flags & OCFS2_INODE_DELETED) { |
ccd979bd MF |
120 | spin_unlock(&oi->ip_lock); |
121 | ||
122 | status = -ENOENT; | |
123 | goto leave; | |
124 | } | |
125 | ||
126 | if (mode & O_DIRECT) | |
127 | oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT; | |
128 | ||
129 | oi->ip_open_count++; | |
130 | spin_unlock(&oi->ip_lock); | |
53fc622b MF |
131 | |
132 | status = ocfs2_init_file_private(inode, file); | |
133 | if (status) { | |
134 | /* | |
135 | * We want to set open count back if we're failing the | |
136 | * open. | |
137 | */ | |
138 | spin_lock(&oi->ip_lock); | |
139 | oi->ip_open_count--; | |
140 | spin_unlock(&oi->ip_lock); | |
141 | } | |
142 | ||
c4c2416a GH |
143 | file->f_mode |= FMODE_NOWAIT; |
144 | ||
ccd979bd | 145 | leave: |
ccd979bd MF |
146 | return status; |
147 | } | |
148 | ||
149 | static int ocfs2_file_release(struct inode *inode, struct file *file) | |
150 | { | |
151 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
152 | ||
ccd979bd MF |
153 | spin_lock(&oi->ip_lock); |
154 | if (!--oi->ip_open_count) | |
155 | oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT; | |
468eedde TM |
156 | |
157 | trace_ocfs2_file_release(inode, file, file->f_path.dentry, | |
158 | oi->ip_blkno, | |
159 | file->f_path.dentry->d_name.len, | |
160 | file->f_path.dentry->d_name.name, | |
161 | oi->ip_open_count); | |
ccd979bd MF |
162 | spin_unlock(&oi->ip_lock); |
163 | ||
53fc622b MF |
164 | ocfs2_free_file_private(inode, file); |
165 | ||
ccd979bd MF |
166 | return 0; |
167 | } | |
168 | ||
53fc622b MF |
169 | static int ocfs2_dir_open(struct inode *inode, struct file *file) |
170 | { | |
171 | return ocfs2_init_file_private(inode, file); | |
172 | } | |
173 | ||
174 | static int ocfs2_dir_release(struct inode *inode, struct file *file) | |
175 | { | |
176 | ocfs2_free_file_private(inode, file); | |
177 | return 0; | |
178 | } | |
179 | ||
02c24a82 JB |
180 | static int ocfs2_sync_file(struct file *file, loff_t start, loff_t end, |
181 | int datasync) | |
ccd979bd MF |
182 | { |
183 | int err = 0; | |
7ea80859 | 184 | struct inode *inode = file->f_mapping->host; |
ccd979bd | 185 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
2931cdcb DW |
186 | struct ocfs2_inode_info *oi = OCFS2_I(inode); |
187 | journal_t *journal = osb->journal->j_journal; | |
188 | int ret; | |
189 | tid_t commit_tid; | |
190 | bool needs_barrier = false; | |
ccd979bd | 191 | |
468eedde | 192 | trace_ocfs2_sync_file(inode, file, file->f_path.dentry, |
d324cd4c | 193 | oi->ip_blkno, |
468eedde TM |
194 | file->f_path.dentry->d_name.len, |
195 | file->f_path.dentry->d_name.name, | |
196 | (unsigned long long)datasync); | |
ccd979bd | 197 | |
a987c7ca YL |
198 | if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb)) |
199 | return -EROFS; | |
200 | ||
3b49c9a1 | 201 | err = file_write_and_wait_range(file, start, end); |
02c24a82 JB |
202 | if (err) |
203 | return err; | |
204 | ||
2931cdcb DW |
205 | commit_tid = datasync ? oi->i_datasync_tid : oi->i_sync_tid; |
206 | if (journal->j_flags & JBD2_BARRIER && | |
207 | !jbd2_trans_will_send_data_barrier(journal, commit_tid)) | |
208 | needs_barrier = true; | |
209 | err = jbd2_complete_transaction(journal, commit_tid); | |
210 | if (needs_barrier) { | |
211 | ret = blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, NULL); | |
212 | if (!err) | |
213 | err = ret; | |
04eda1a1 | 214 | } |
e04cc15f | 215 | |
c1e8d35e TM |
216 | if (err) |
217 | mlog_errno(err); | |
ccd979bd MF |
218 | |
219 | return (err < 0) ? -EIO : 0; | |
220 | } | |
221 | ||
7f1a37e3 TY |
222 | int ocfs2_should_update_atime(struct inode *inode, |
223 | struct vfsmount *vfsmnt) | |
224 | { | |
95582b00 | 225 | struct timespec64 now; |
7f1a37e3 TY |
226 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
227 | ||
228 | if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb)) | |
229 | return 0; | |
230 | ||
231 | if ((inode->i_flags & S_NOATIME) || | |
1751e8a6 | 232 | ((inode->i_sb->s_flags & SB_NODIRATIME) && S_ISDIR(inode->i_mode))) |
7f1a37e3 TY |
233 | return 0; |
234 | ||
6c2aad05 MF |
235 | /* |
236 | * We can be called with no vfsmnt structure - NFSD will | |
237 | * sometimes do this. | |
238 | * | |
239 | * Note that our action here is different than touch_atime() - | |
240 | * if we can't tell whether this is a noatime mount, then we | |
241 | * don't know whether to trust the value of s_atime_quantum. | |
242 | */ | |
243 | if (vfsmnt == NULL) | |
244 | return 0; | |
245 | ||
7f1a37e3 TY |
246 | if ((vfsmnt->mnt_flags & MNT_NOATIME) || |
247 | ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))) | |
248 | return 0; | |
249 | ||
7e913c53 | 250 | if (vfsmnt->mnt_flags & MNT_RELATIME) { |
95582b00 DD |
251 | if ((timespec64_compare(&inode->i_atime, &inode->i_mtime) <= 0) || |
252 | (timespec64_compare(&inode->i_atime, &inode->i_ctime) <= 0)) | |
7e913c53 MF |
253 | return 1; |
254 | ||
255 | return 0; | |
256 | } | |
257 | ||
078cd827 | 258 | now = current_time(inode); |
7f1a37e3 TY |
259 | if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum)) |
260 | return 0; | |
261 | else | |
262 | return 1; | |
263 | } | |
264 | ||
265 | int ocfs2_update_inode_atime(struct inode *inode, | |
266 | struct buffer_head *bh) | |
267 | { | |
268 | int ret; | |
269 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
270 | handle_t *handle; | |
c11e9faf | 271 | struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data; |
7f1a37e3 | 272 | |
7f1a37e3 | 273 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
fa38e92c JK |
274 | if (IS_ERR(handle)) { |
275 | ret = PTR_ERR(handle); | |
7f1a37e3 TY |
276 | mlog_errno(ret); |
277 | goto out; | |
278 | } | |
279 | ||
0cf2f763 | 280 | ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh, |
13723d00 | 281 | OCFS2_JOURNAL_ACCESS_WRITE); |
c11e9faf MF |
282 | if (ret) { |
283 | mlog_errno(ret); | |
284 | goto out_commit; | |
285 | } | |
286 | ||
287 | /* | |
288 | * Don't use ocfs2_mark_inode_dirty() here as we don't always | |
289 | * have i_mutex to guard against concurrent changes to other | |
290 | * inode fields. | |
291 | */ | |
078cd827 | 292 | inode->i_atime = current_time(inode); |
c11e9faf MF |
293 | di->i_atime = cpu_to_le64(inode->i_atime.tv_sec); |
294 | di->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec); | |
6fdb702d | 295 | ocfs2_update_inode_fsync_trans(handle, inode, 0); |
ec20cec7 | 296 | ocfs2_journal_dirty(handle, bh); |
7f1a37e3 | 297 | |
c11e9faf | 298 | out_commit: |
1119d3c0 | 299 | ocfs2_commit_trans(osb, handle); |
7f1a37e3 | 300 | out: |
7f1a37e3 TY |
301 | return ret; |
302 | } | |
303 | ||
026749a8 | 304 | int ocfs2_set_inode_size(handle_t *handle, |
6cb129f5 AB |
305 | struct inode *inode, |
306 | struct buffer_head *fe_bh, | |
307 | u64 new_i_size) | |
ccd979bd MF |
308 | { |
309 | int status; | |
310 | ||
ccd979bd | 311 | i_size_write(inode, new_i_size); |
8110b073 | 312 | inode->i_blocks = ocfs2_inode_sector_count(inode); |
078cd827 | 313 | inode->i_ctime = inode->i_mtime = current_time(inode); |
ccd979bd MF |
314 | |
315 | status = ocfs2_mark_inode_dirty(handle, inode, fe_bh); | |
316 | if (status < 0) { | |
317 | mlog_errno(status); | |
318 | goto bail; | |
319 | } | |
320 | ||
321 | bail: | |
ccd979bd MF |
322 | return status; |
323 | } | |
324 | ||
9e33d69f JK |
325 | int ocfs2_simple_size_update(struct inode *inode, |
326 | struct buffer_head *di_bh, | |
327 | u64 new_i_size) | |
ccd979bd MF |
328 | { |
329 | int ret; | |
330 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1fabe148 | 331 | handle_t *handle = NULL; |
ccd979bd | 332 | |
65eff9cc | 333 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
fa38e92c JK |
334 | if (IS_ERR(handle)) { |
335 | ret = PTR_ERR(handle); | |
ccd979bd MF |
336 | mlog_errno(ret); |
337 | goto out; | |
338 | } | |
339 | ||
340 | ret = ocfs2_set_inode_size(handle, inode, di_bh, | |
341 | new_i_size); | |
342 | if (ret < 0) | |
343 | mlog_errno(ret); | |
344 | ||
6fdb702d | 345 | ocfs2_update_inode_fsync_trans(handle, inode, 0); |
02dc1af4 | 346 | ocfs2_commit_trans(osb, handle); |
ccd979bd MF |
347 | out: |
348 | return ret; | |
349 | } | |
350 | ||
37f8a2bf TM |
351 | static int ocfs2_cow_file_pos(struct inode *inode, |
352 | struct buffer_head *fe_bh, | |
353 | u64 offset) | |
354 | { | |
355 | int status; | |
356 | u32 phys, cpos = offset >> OCFS2_SB(inode->i_sb)->s_clustersize_bits; | |
357 | unsigned int num_clusters = 0; | |
358 | unsigned int ext_flags = 0; | |
359 | ||
360 | /* | |
361 | * If the new offset is aligned to the range of the cluster, there is | |
362 | * no space for ocfs2_zero_range_for_truncate to fill, so no need to | |
363 | * CoW either. | |
364 | */ | |
365 | if ((offset & (OCFS2_SB(inode->i_sb)->s_clustersize - 1)) == 0) | |
366 | return 0; | |
367 | ||
368 | status = ocfs2_get_clusters(inode, cpos, &phys, | |
369 | &num_clusters, &ext_flags); | |
370 | if (status) { | |
371 | mlog_errno(status); | |
372 | goto out; | |
373 | } | |
374 | ||
375 | if (!(ext_flags & OCFS2_EXT_REFCOUNTED)) | |
376 | goto out; | |
377 | ||
c7dd3392 | 378 | return ocfs2_refcount_cow(inode, fe_bh, cpos, 1, cpos+1); |
37f8a2bf TM |
379 | |
380 | out: | |
381 | return status; | |
382 | } | |
383 | ||
ccd979bd MF |
384 | static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb, |
385 | struct inode *inode, | |
386 | struct buffer_head *fe_bh, | |
387 | u64 new_i_size) | |
388 | { | |
389 | int status; | |
1fabe148 | 390 | handle_t *handle; |
60b11392 | 391 | struct ocfs2_dinode *di; |
35edec1d | 392 | u64 cluster_bytes; |
ccd979bd | 393 | |
37f8a2bf TM |
394 | /* |
395 | * We need to CoW the cluster contains the offset if it is reflinked | |
396 | * since we will call ocfs2_zero_range_for_truncate later which will | |
397 | * write "0" from offset to the end of the cluster. | |
398 | */ | |
399 | status = ocfs2_cow_file_pos(inode, fe_bh, new_i_size); | |
400 | if (status) { | |
401 | mlog_errno(status); | |
402 | return status; | |
403 | } | |
404 | ||
ccd979bd MF |
405 | /* TODO: This needs to actually orphan the inode in this |
406 | * transaction. */ | |
407 | ||
65eff9cc | 408 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
ccd979bd MF |
409 | if (IS_ERR(handle)) { |
410 | status = PTR_ERR(handle); | |
411 | mlog_errno(status); | |
412 | goto out; | |
413 | } | |
414 | ||
0cf2f763 | 415 | status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh, |
13723d00 | 416 | OCFS2_JOURNAL_ACCESS_WRITE); |
60b11392 MF |
417 | if (status < 0) { |
418 | mlog_errno(status); | |
419 | goto out_commit; | |
420 | } | |
421 | ||
422 | /* | |
423 | * Do this before setting i_size. | |
424 | */ | |
35edec1d MF |
425 | cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size); |
426 | status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size, | |
427 | cluster_bytes); | |
60b11392 MF |
428 | if (status) { |
429 | mlog_errno(status); | |
430 | goto out_commit; | |
431 | } | |
432 | ||
433 | i_size_write(inode, new_i_size); | |
078cd827 | 434 | inode->i_ctime = inode->i_mtime = current_time(inode); |
60b11392 MF |
435 | |
436 | di = (struct ocfs2_dinode *) fe_bh->b_data; | |
437 | di->i_size = cpu_to_le64(new_i_size); | |
438 | di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec); | |
439 | di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec); | |
6fdb702d | 440 | ocfs2_update_inode_fsync_trans(handle, inode, 0); |
60b11392 | 441 | |
ec20cec7 | 442 | ocfs2_journal_dirty(handle, fe_bh); |
ccd979bd | 443 | |
60b11392 | 444 | out_commit: |
02dc1af4 | 445 | ocfs2_commit_trans(osb, handle); |
ccd979bd | 446 | out: |
ccd979bd MF |
447 | return status; |
448 | } | |
449 | ||
026749a8 | 450 | int ocfs2_truncate_file(struct inode *inode, |
ccd979bd MF |
451 | struct buffer_head *di_bh, |
452 | u64 new_i_size) | |
453 | { | |
454 | int status = 0; | |
455 | struct ocfs2_dinode *fe = NULL; | |
456 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
ccd979bd | 457 | |
b657c95c JB |
458 | /* We trust di_bh because it comes from ocfs2_inode_lock(), which |
459 | * already validated it */ | |
ccd979bd | 460 | fe = (struct ocfs2_dinode *) di_bh->b_data; |
ccd979bd | 461 | |
468eedde TM |
462 | trace_ocfs2_truncate_file((unsigned long long)OCFS2_I(inode)->ip_blkno, |
463 | (unsigned long long)le64_to_cpu(fe->i_size), | |
464 | (unsigned long long)new_i_size); | |
465 | ||
ccd979bd | 466 | mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode), |
b0697053 MF |
467 | "Inode %llu, inode i_size = %lld != di " |
468 | "i_size = %llu, i_flags = 0x%x\n", | |
469 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
ccd979bd | 470 | i_size_read(inode), |
b0697053 MF |
471 | (unsigned long long)le64_to_cpu(fe->i_size), |
472 | le32_to_cpu(fe->i_flags)); | |
ccd979bd MF |
473 | |
474 | if (new_i_size > le64_to_cpu(fe->i_size)) { | |
468eedde TM |
475 | trace_ocfs2_truncate_file_error( |
476 | (unsigned long long)le64_to_cpu(fe->i_size), | |
477 | (unsigned long long)new_i_size); | |
ccd979bd MF |
478 | status = -EINVAL; |
479 | mlog_errno(status); | |
480 | goto bail; | |
481 | } | |
482 | ||
2e89b2e4 MF |
483 | down_write(&OCFS2_I(inode)->ip_alloc_sem); |
484 | ||
4fe370af MF |
485 | ocfs2_resv_discard(&osb->osb_la_resmap, |
486 | &OCFS2_I(inode)->ip_la_data_resv); | |
487 | ||
c934a92d MF |
488 | /* |
489 | * The inode lock forced other nodes to sync and drop their | |
490 | * pages, which (correctly) happens even if we have a truncate | |
491 | * without allocation change - ocfs2 cluster sizes can be much | |
492 | * greater than page size, so we have to truncate them | |
493 | * anyway. | |
494 | */ | |
2e89b2e4 MF |
495 | unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1); |
496 | truncate_inode_pages(inode->i_mapping, new_i_size); | |
497 | ||
1afc32b9 MF |
498 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) { |
499 | status = ocfs2_truncate_inline(inode, di_bh, new_i_size, | |
b1967d0e | 500 | i_size_read(inode), 1); |
1afc32b9 MF |
501 | if (status) |
502 | mlog_errno(status); | |
503 | ||
c934a92d | 504 | goto bail_unlock_sem; |
1afc32b9 MF |
505 | } |
506 | ||
ccd979bd MF |
507 | /* alright, we're going to need to do a full blown alloc size |
508 | * change. Orphan the inode so that recovery can complete the | |
509 | * truncate if necessary. This does the task of marking | |
510 | * i_size. */ | |
511 | status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size); | |
512 | if (status < 0) { | |
513 | mlog_errno(status); | |
c934a92d | 514 | goto bail_unlock_sem; |
ccd979bd MF |
515 | } |
516 | ||
78f94673 | 517 | status = ocfs2_commit_truncate(osb, inode, di_bh); |
ccd979bd MF |
518 | if (status < 0) { |
519 | mlog_errno(status); | |
c934a92d | 520 | goto bail_unlock_sem; |
ccd979bd MF |
521 | } |
522 | ||
523 | /* TODO: orphan dir cleanup here. */ | |
c934a92d | 524 | bail_unlock_sem: |
2e89b2e4 MF |
525 | up_write(&OCFS2_I(inode)->ip_alloc_sem); |
526 | ||
ccd979bd | 527 | bail: |
8b2c0dba TM |
528 | if (!status && OCFS2_I(inode)->ip_clusters == 0) |
529 | status = ocfs2_try_remove_refcount_tree(inode, di_bh); | |
ccd979bd | 530 | |
ccd979bd MF |
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 | ||
5bc55d65 JG |
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; | |
696cdf73 | 578 | enum ocfs2_alloc_restarted why = RESTART_NONE; |
ccd979bd | 579 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
f99b9b7c | 580 | struct ocfs2_extent_tree et; |
a90714c1 | 581 | int did_quota = 0; |
ccd979bd | 582 | |
dcd0538f | 583 | /* |
f0cb0f0b | 584 | * Unwritten extent only exists for file systems which |
dcd0538f MF |
585 | * support holes. |
586 | */ | |
2ae99a60 | 587 | BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb)); |
dcd0538f | 588 | |
b657c95c | 589 | status = ocfs2_read_inode_block(inode, &bh); |
ccd979bd MF |
590 | if (status < 0) { |
591 | mlog_errno(status); | |
592 | goto leave; | |
593 | } | |
ccd979bd | 594 | fe = (struct ocfs2_dinode *) bh->b_data; |
ccd979bd MF |
595 | |
596 | restart_all: | |
597 | BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters); | |
598 | ||
5e404e9e | 599 | ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), bh); |
f99b9b7c JB |
600 | status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0, |
601 | &data_ac, &meta_ac); | |
9517bac6 MF |
602 | if (status) { |
603 | mlog_errno(status); | |
604 | goto leave; | |
605 | } | |
606 | ||
06f9da6e | 607 | credits = ocfs2_calc_extend_credits(osb->sb, &fe->id2.i_list); |
65eff9cc | 608 | handle = ocfs2_start_trans(osb, credits); |
ccd979bd MF |
609 | if (IS_ERR(handle)) { |
610 | status = PTR_ERR(handle); | |
611 | handle = NULL; | |
612 | mlog_errno(status); | |
613 | goto leave; | |
614 | } | |
615 | ||
616 | restarted_transaction: | |
468eedde TM |
617 | trace_ocfs2_extend_allocation( |
618 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
619 | (unsigned long long)i_size_read(inode), | |
620 | le32_to_cpu(fe->i_clusters), clusters_to_add, | |
621 | why, restart_func); | |
622 | ||
5dd4056d CH |
623 | status = dquot_alloc_space_nodirty(inode, |
624 | ocfs2_clusters_to_bytes(osb->sb, clusters_to_add)); | |
625 | if (status) | |
a90714c1 | 626 | goto leave; |
a90714c1 JK |
627 | did_quota = 1; |
628 | ||
ccd979bd MF |
629 | /* reserve a write to the file entry early on - that we if we |
630 | * run out of credits in the allocation path, we can still | |
631 | * update i_size. */ | |
0cf2f763 | 632 | status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh, |
13723d00 | 633 | OCFS2_JOURNAL_ACCESS_WRITE); |
ccd979bd MF |
634 | if (status < 0) { |
635 | mlog_errno(status); | |
636 | goto leave; | |
637 | } | |
638 | ||
639 | prev_clusters = OCFS2_I(inode)->ip_clusters; | |
640 | ||
0eb8d47e TM |
641 | status = ocfs2_add_inode_data(osb, |
642 | inode, | |
643 | &logical_start, | |
644 | clusters_to_add, | |
645 | mark_unwritten, | |
646 | bh, | |
647 | handle, | |
648 | data_ac, | |
649 | meta_ac, | |
650 | &why); | |
ccd979bd MF |
651 | if ((status < 0) && (status != -EAGAIN)) { |
652 | if (status != -ENOSPC) | |
653 | mlog_errno(status); | |
654 | goto leave; | |
655 | } | |
2931cdcb | 656 | ocfs2_update_inode_fsync_trans(handle, inode, 1); |
ec20cec7 | 657 | ocfs2_journal_dirty(handle, bh); |
ccd979bd MF |
658 | |
659 | spin_lock(&OCFS2_I(inode)->ip_lock); | |
660 | clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters); | |
661 | spin_unlock(&OCFS2_I(inode)->ip_lock); | |
a90714c1 | 662 | /* Release unused quota reservation */ |
5dd4056d | 663 | dquot_free_space(inode, |
a90714c1 JK |
664 | ocfs2_clusters_to_bytes(osb->sb, clusters_to_add)); |
665 | did_quota = 0; | |
ccd979bd MF |
666 | |
667 | if (why != RESTART_NONE && clusters_to_add) { | |
668 | if (why == RESTART_META) { | |
ccd979bd | 669 | restart_func = 1; |
79681842 | 670 | status = 0; |
ccd979bd MF |
671 | } else { |
672 | BUG_ON(why != RESTART_TRANS); | |
673 | ||
2b1e55c3 | 674 | status = ocfs2_allocate_extend_trans(handle, 1); |
ccd979bd MF |
675 | if (status < 0) { |
676 | /* handle still has to be committed at | |
677 | * this point. */ | |
678 | status = -ENOMEM; | |
679 | mlog_errno(status); | |
680 | goto leave; | |
681 | } | |
682 | goto restarted_transaction; | |
683 | } | |
684 | } | |
685 | ||
468eedde | 686 | trace_ocfs2_extend_allocation_end(OCFS2_I(inode)->ip_blkno, |
1ca1a111 | 687 | le32_to_cpu(fe->i_clusters), |
468eedde TM |
688 | (unsigned long long)le64_to_cpu(fe->i_size), |
689 | OCFS2_I(inode)->ip_clusters, | |
690 | (unsigned long long)i_size_read(inode)); | |
ccd979bd MF |
691 | |
692 | leave: | |
a90714c1 | 693 | if (status < 0 && did_quota) |
5dd4056d | 694 | dquot_free_space(inode, |
a90714c1 | 695 | ocfs2_clusters_to_bytes(osb->sb, clusters_to_add)); |
ccd979bd | 696 | if (handle) { |
02dc1af4 | 697 | ocfs2_commit_trans(osb, handle); |
ccd979bd MF |
698 | handle = NULL; |
699 | } | |
700 | if (data_ac) { | |
701 | ocfs2_free_alloc_context(data_ac); | |
702 | data_ac = NULL; | |
703 | } | |
704 | if (meta_ac) { | |
705 | ocfs2_free_alloc_context(meta_ac); | |
706 | meta_ac = NULL; | |
707 | } | |
708 | if ((!status) && restart_func) { | |
709 | restart_func = 0; | |
710 | goto restart_all; | |
711 | } | |
a81cb88b MF |
712 | brelse(bh); |
713 | bh = NULL; | |
ccd979bd | 714 | |
ccd979bd MF |
715 | return status; |
716 | } | |
717 | ||
a4bfb4cf JB |
718 | /* |
719 | * While a write will already be ordering the data, a truncate will not. | |
720 | * Thus, we need to explicitly order the zeroed pages. | |
721 | */ | |
c7d2cbc3 JB |
722 | static handle_t *ocfs2_zero_start_ordered_transaction(struct inode *inode, |
723 | struct buffer_head *di_bh) | |
a4bfb4cf JB |
724 | { |
725 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
726 | handle_t *handle = NULL; | |
727 | int ret = 0; | |
728 | ||
729 | if (!ocfs2_should_order_data(inode)) | |
730 | goto out; | |
731 | ||
732 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); | |
733 | if (IS_ERR(handle)) { | |
734 | ret = -ENOMEM; | |
735 | mlog_errno(ret); | |
736 | goto out; | |
737 | } | |
738 | ||
739 | ret = ocfs2_jbd2_file_inode(handle, inode); | |
c7d2cbc3 JB |
740 | if (ret < 0) { |
741 | mlog_errno(ret); | |
742 | goto out; | |
743 | } | |
744 | ||
745 | ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh, | |
746 | OCFS2_JOURNAL_ACCESS_WRITE); | |
747 | if (ret) | |
a4bfb4cf | 748 | mlog_errno(ret); |
6fdb702d | 749 | ocfs2_update_inode_fsync_trans(handle, inode, 1); |
a4bfb4cf JB |
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 | 763 | static int ocfs2_write_zero_page(struct inode *inode, u64 abs_from, |
c7d2cbc3 | 764 | u64 abs_to, struct buffer_head *di_bh) |
ccd979bd MF |
765 | { |
766 | struct address_space *mapping = inode->i_mapping; | |
767 | struct page *page; | |
09cbfeaf | 768 | unsigned long index = abs_from >> PAGE_SHIFT; |
f775da2f | 769 | handle_t *handle; |
5453258d | 770 | int ret = 0; |
a4bfb4cf | 771 | unsigned zero_from, zero_to, block_start, block_end; |
c7d2cbc3 | 772 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; |
ccd979bd | 773 | |
a4bfb4cf | 774 | BUG_ON(abs_from >= abs_to); |
09cbfeaf | 775 | BUG_ON(abs_to > (((u64)index + 1) << PAGE_SHIFT)); |
a4bfb4cf | 776 | BUG_ON(abs_from & (inode->i_blkbits - 1)); |
ccd979bd | 777 | |
f775da2f JB |
778 | handle = ocfs2_zero_start_ordered_transaction(inode, di_bh); |
779 | if (IS_ERR(handle)) { | |
780 | ret = PTR_ERR(handle); | |
781 | goto out; | |
782 | } | |
783 | ||
9b4c0ff3 | 784 | page = find_or_create_page(mapping, index, GFP_NOFS); |
ccd979bd MF |
785 | if (!page) { |
786 | ret = -ENOMEM; | |
787 | mlog_errno(ret); | |
f775da2f | 788 | goto out_commit_trans; |
ccd979bd MF |
789 | } |
790 | ||
a4bfb4cf | 791 | /* Get the offsets within the page that we want to zero */ |
09cbfeaf KS |
792 | zero_from = abs_from & (PAGE_SIZE - 1); |
793 | zero_to = abs_to & (PAGE_SIZE - 1); | |
a4bfb4cf | 794 | if (!zero_to) |
09cbfeaf | 795 | zero_to = PAGE_SIZE; |
ccd979bd | 796 | |
468eedde TM |
797 | trace_ocfs2_write_zero_page( |
798 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
799 | (unsigned long long)abs_from, | |
800 | (unsigned long long)abs_to, | |
801 | index, zero_from, zero_to); | |
5693486b | 802 | |
a4bfb4cf JB |
803 | /* We know that zero_from is block aligned */ |
804 | for (block_start = zero_from; block_start < zero_to; | |
805 | block_start = block_end) { | |
93407472 | 806 | block_end = block_start + i_blocksize(inode); |
a4bfb4cf JB |
807 | |
808 | /* | |
ebdec241 CH |
809 | * block_start is block-aligned. Bump it by one to force |
810 | * __block_write_begin and block_commit_write to zero the | |
a4bfb4cf JB |
811 | * whole block. |
812 | */ | |
ebdec241 CH |
813 | ret = __block_write_begin(page, block_start + 1, 0, |
814 | ocfs2_get_block); | |
a4bfb4cf JB |
815 | if (ret < 0) { |
816 | mlog_errno(ret); | |
ccd979bd MF |
817 | goto out_unlock; |
818 | } | |
ccd979bd | 819 | |
a4bfb4cf JB |
820 | |
821 | /* must not update i_size! */ | |
822 | ret = block_commit_write(page, block_start + 1, | |
823 | block_start + 1); | |
824 | if (ret < 0) | |
825 | mlog_errno(ret); | |
826 | else | |
827 | ret = 0; | |
828 | } | |
ccd979bd | 829 | |
f775da2f JB |
830 | /* |
831 | * fs-writeback will release the dirty pages without page lock | |
832 | * whose offset are over inode size, the release happens at | |
833 | * block_write_full_page(). | |
834 | */ | |
835 | i_size_write(inode, abs_to); | |
836 | inode->i_blocks = ocfs2_inode_sector_count(inode); | |
837 | di->i_size = cpu_to_le64((u64)i_size_read(inode)); | |
078cd827 | 838 | inode->i_mtime = inode->i_ctime = current_time(inode); |
f775da2f JB |
839 | di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec); |
840 | di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec); | |
841 | di->i_mtime_nsec = di->i_ctime_nsec; | |
c7d2cbc3 | 842 | if (handle) { |
c7d2cbc3 | 843 | ocfs2_journal_dirty(handle, di_bh); |
6fdb702d | 844 | ocfs2_update_inode_fsync_trans(handle, inode, 1); |
c7d2cbc3 | 845 | } |
a4bfb4cf | 846 | |
ccd979bd MF |
847 | out_unlock: |
848 | unlock_page(page); | |
09cbfeaf | 849 | put_page(page); |
f775da2f JB |
850 | out_commit_trans: |
851 | if (handle) | |
852 | ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle); | |
ccd979bd MF |
853 | out: |
854 | return ret; | |
855 | } | |
856 | ||
5693486b JB |
857 | /* |
858 | * Find the next range to zero. We do this in terms of bytes because | |
859 | * that's what ocfs2_zero_extend() wants, and it is dealing with the | |
860 | * pagecache. We may return multiple extents. | |
861 | * | |
862 | * zero_start and zero_end are ocfs2_zero_extend()s current idea of what | |
863 | * needs to be zeroed. range_start and range_end return the next zeroing | |
864 | * range. A subsequent call should pass the previous range_end as its | |
865 | * zero_start. If range_end is 0, there's nothing to do. | |
866 | * | |
867 | * Unwritten extents are skipped over. Refcounted extents are CoWd. | |
868 | */ | |
869 | static int ocfs2_zero_extend_get_range(struct inode *inode, | |
870 | struct buffer_head *di_bh, | |
871 | u64 zero_start, u64 zero_end, | |
872 | u64 *range_start, u64 *range_end) | |
ccd979bd | 873 | { |
5693486b JB |
874 | int rc = 0, needs_cow = 0; |
875 | u32 p_cpos, zero_clusters = 0; | |
876 | u32 zero_cpos = | |
877 | zero_start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits; | |
878 | u32 last_cpos = ocfs2_clusters_for_bytes(inode->i_sb, zero_end); | |
879 | unsigned int num_clusters = 0; | |
880 | unsigned int ext_flags = 0; | |
ccd979bd | 881 | |
5693486b JB |
882 | while (zero_cpos < last_cpos) { |
883 | rc = ocfs2_get_clusters(inode, zero_cpos, &p_cpos, | |
884 | &num_clusters, &ext_flags); | |
885 | if (rc) { | |
886 | mlog_errno(rc); | |
ccd979bd MF |
887 | goto out; |
888 | } | |
889 | ||
5693486b JB |
890 | if (p_cpos && !(ext_flags & OCFS2_EXT_UNWRITTEN)) { |
891 | zero_clusters = num_clusters; | |
892 | if (ext_flags & OCFS2_EXT_REFCOUNTED) | |
893 | needs_cow = 1; | |
894 | break; | |
895 | } | |
896 | ||
897 | zero_cpos += num_clusters; | |
898 | } | |
899 | if (!zero_clusters) { | |
900 | *range_end = 0; | |
901 | goto out; | |
902 | } | |
903 | ||
904 | while ((zero_cpos + zero_clusters) < last_cpos) { | |
905 | rc = ocfs2_get_clusters(inode, zero_cpos + zero_clusters, | |
906 | &p_cpos, &num_clusters, | |
907 | &ext_flags); | |
908 | if (rc) { | |
909 | mlog_errno(rc); | |
910 | goto out; | |
911 | } | |
912 | ||
913 | if (!p_cpos || (ext_flags & OCFS2_EXT_UNWRITTEN)) | |
914 | break; | |
915 | if (ext_flags & OCFS2_EXT_REFCOUNTED) | |
916 | needs_cow = 1; | |
917 | zero_clusters += num_clusters; | |
918 | } | |
919 | if ((zero_cpos + zero_clusters) > last_cpos) | |
920 | zero_clusters = last_cpos - zero_cpos; | |
921 | ||
922 | if (needs_cow) { | |
c7dd3392 | 923 | rc = ocfs2_refcount_cow(inode, di_bh, zero_cpos, |
15502712 | 924 | zero_clusters, UINT_MAX); |
5693486b JB |
925 | if (rc) { |
926 | mlog_errno(rc); | |
927 | goto out; | |
928 | } | |
929 | } | |
930 | ||
931 | *range_start = ocfs2_clusters_to_bytes(inode->i_sb, zero_cpos); | |
932 | *range_end = ocfs2_clusters_to_bytes(inode->i_sb, | |
933 | zero_cpos + zero_clusters); | |
934 | ||
935 | out: | |
936 | return rc; | |
937 | } | |
938 | ||
939 | /* | |
940 | * Zero one range returned from ocfs2_zero_extend_get_range(). The caller | |
941 | * has made sure that the entire range needs zeroing. | |
942 | */ | |
943 | static int ocfs2_zero_extend_range(struct inode *inode, u64 range_start, | |
c7d2cbc3 | 944 | u64 range_end, struct buffer_head *di_bh) |
5693486b JB |
945 | { |
946 | int rc = 0; | |
947 | u64 next_pos; | |
948 | u64 zero_pos = range_start; | |
949 | ||
468eedde TM |
950 | trace_ocfs2_zero_extend_range( |
951 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
952 | (unsigned long long)range_start, | |
953 | (unsigned long long)range_end); | |
5693486b JB |
954 | BUG_ON(range_start >= range_end); |
955 | ||
956 | while (zero_pos < range_end) { | |
09cbfeaf | 957 | next_pos = (zero_pos & PAGE_MASK) + PAGE_SIZE; |
5693486b JB |
958 | if (next_pos > range_end) |
959 | next_pos = range_end; | |
c7d2cbc3 | 960 | rc = ocfs2_write_zero_page(inode, zero_pos, next_pos, di_bh); |
5693486b JB |
961 | if (rc < 0) { |
962 | mlog_errno(rc); | |
963 | break; | |
964 | } | |
965 | zero_pos = next_pos; | |
e2057c5a MF |
966 | |
967 | /* | |
968 | * Very large extends have the potential to lock up | |
969 | * the cpu for extended periods of time. | |
970 | */ | |
971 | cond_resched(); | |
ccd979bd MF |
972 | } |
973 | ||
5693486b JB |
974 | return rc; |
975 | } | |
976 | ||
977 | int ocfs2_zero_extend(struct inode *inode, struct buffer_head *di_bh, | |
978 | loff_t zero_to_size) | |
979 | { | |
980 | int ret = 0; | |
981 | u64 zero_start, range_start = 0, range_end = 0; | |
982 | struct super_block *sb = inode->i_sb; | |
983 | ||
984 | zero_start = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode)); | |
468eedde TM |
985 | trace_ocfs2_zero_extend((unsigned long long)OCFS2_I(inode)->ip_blkno, |
986 | (unsigned long long)zero_start, | |
987 | (unsigned long long)i_size_read(inode)); | |
5693486b JB |
988 | while (zero_start < zero_to_size) { |
989 | ret = ocfs2_zero_extend_get_range(inode, di_bh, zero_start, | |
990 | zero_to_size, | |
991 | &range_start, | |
992 | &range_end); | |
993 | if (ret) { | |
994 | mlog_errno(ret); | |
995 | break; | |
996 | } | |
997 | if (!range_end) | |
998 | break; | |
999 | /* Trim the ends */ | |
1000 | if (range_start < zero_start) | |
1001 | range_start = zero_start; | |
1002 | if (range_end > zero_to_size) | |
1003 | range_end = zero_to_size; | |
1004 | ||
1005 | ret = ocfs2_zero_extend_range(inode, range_start, | |
c7d2cbc3 | 1006 | range_end, di_bh); |
5693486b JB |
1007 | if (ret) { |
1008 | mlog_errno(ret); | |
1009 | break; | |
1010 | } | |
1011 | zero_start = range_end; | |
1012 | } | |
1013 | ||
ccd979bd MF |
1014 | return ret; |
1015 | } | |
1016 | ||
5693486b JB |
1017 | int ocfs2_extend_no_holes(struct inode *inode, struct buffer_head *di_bh, |
1018 | u64 new_i_size, u64 zero_to) | |
65ed39d6 MF |
1019 | { |
1020 | int ret; | |
1021 | u32 clusters_to_add; | |
1022 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
1023 | ||
5693486b JB |
1024 | /* |
1025 | * Only quota files call this without a bh, and they can't be | |
1026 | * refcounted. | |
1027 | */ | |
84e40080 | 1028 | BUG_ON(!di_bh && ocfs2_is_refcount_inode(inode)); |
5693486b JB |
1029 | BUG_ON(!di_bh && !(oi->ip_flags & OCFS2_INODE_SYSTEM_FILE)); |
1030 | ||
65ed39d6 MF |
1031 | clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size); |
1032 | if (clusters_to_add < oi->ip_clusters) | |
1033 | clusters_to_add = 0; | |
1034 | else | |
1035 | clusters_to_add -= oi->ip_clusters; | |
1036 | ||
1037 | if (clusters_to_add) { | |
5bc55d65 JG |
1038 | ret = ocfs2_extend_allocation(inode, oi->ip_clusters, |
1039 | clusters_to_add, 0); | |
65ed39d6 MF |
1040 | if (ret) { |
1041 | mlog_errno(ret); | |
1042 | goto out; | |
1043 | } | |
1044 | } | |
1045 | ||
1046 | /* | |
1047 | * Call this even if we don't add any clusters to the tree. We | |
1048 | * still need to zero the area between the old i_size and the | |
1049 | * new i_size. | |
1050 | */ | |
5693486b | 1051 | ret = ocfs2_zero_extend(inode, di_bh, zero_to); |
65ed39d6 MF |
1052 | if (ret < 0) |
1053 | mlog_errno(ret); | |
1054 | ||
1055 | out: | |
1056 | return ret; | |
1057 | } | |
1058 | ||
ccd979bd MF |
1059 | static int ocfs2_extend_file(struct inode *inode, |
1060 | struct buffer_head *di_bh, | |
65ed39d6 | 1061 | u64 new_i_size) |
ccd979bd | 1062 | { |
c934a92d | 1063 | int ret = 0; |
1afc32b9 | 1064 | struct ocfs2_inode_info *oi = OCFS2_I(inode); |
ccd979bd | 1065 | |
65ed39d6 | 1066 | BUG_ON(!di_bh); |
53013cba | 1067 | |
ccd979bd MF |
1068 | /* setattr sometimes calls us like this. */ |
1069 | if (new_i_size == 0) | |
1070 | goto out; | |
1071 | ||
1072 | if (i_size_read(inode) == new_i_size) | |
5693486b | 1073 | goto out; |
ccd979bd MF |
1074 | BUG_ON(new_i_size < i_size_read(inode)); |
1075 | ||
0effef77 | 1076 | /* |
65ed39d6 MF |
1077 | * The alloc sem blocks people in read/write from reading our |
1078 | * allocation until we're done changing it. We depend on | |
1079 | * i_mutex to block other extend/truncate calls while we're | |
5693486b JB |
1080 | * here. We even have to hold it for sparse files because there |
1081 | * might be some tail zeroing. | |
0effef77 | 1082 | */ |
1afc32b9 MF |
1083 | down_write(&oi->ip_alloc_sem); |
1084 | ||
1085 | if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) { | |
1086 | /* | |
1087 | * We can optimize small extends by keeping the inodes | |
1088 | * inline data. | |
1089 | */ | |
1090 | if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) { | |
1091 | up_write(&oi->ip_alloc_sem); | |
1092 | goto out_update_size; | |
1093 | } | |
1094 | ||
1095 | ret = ocfs2_convert_inline_data_to_extents(inode, di_bh); | |
1096 | if (ret) { | |
1097 | up_write(&oi->ip_alloc_sem); | |
1afc32b9 | 1098 | mlog_errno(ret); |
c934a92d | 1099 | goto out; |
1afc32b9 MF |
1100 | } |
1101 | } | |
1102 | ||
5693486b JB |
1103 | if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))) |
1104 | ret = ocfs2_zero_extend(inode, di_bh, new_i_size); | |
1105 | else | |
1106 | ret = ocfs2_extend_no_holes(inode, di_bh, new_i_size, | |
1107 | new_i_size); | |
1afc32b9 MF |
1108 | |
1109 | up_write(&oi->ip_alloc_sem); | |
65ed39d6 | 1110 | |
0effef77 MF |
1111 | if (ret < 0) { |
1112 | mlog_errno(ret); | |
c934a92d | 1113 | goto out; |
53013cba MF |
1114 | } |
1115 | ||
3a0782d0 | 1116 | out_update_size: |
65ed39d6 MF |
1117 | ret = ocfs2_simple_size_update(inode, di_bh, new_i_size); |
1118 | if (ret < 0) | |
1119 | mlog_errno(ret); | |
ccd979bd MF |
1120 | |
1121 | out: | |
1122 | return ret; | |
1123 | } | |
1124 | ||
1125 | int ocfs2_setattr(struct dentry *dentry, struct iattr *attr) | |
1126 | { | |
1127 | int status = 0, size_change; | |
3d46a44a | 1128 | int inode_locked = 0; |
2b0143b5 | 1129 | struct inode *inode = d_inode(dentry); |
ccd979bd MF |
1130 | struct super_block *sb = inode->i_sb; |
1131 | struct ocfs2_super *osb = OCFS2_SB(sb); | |
1132 | struct buffer_head *bh = NULL; | |
1fabe148 | 1133 | handle_t *handle = NULL; |
65bac575 | 1134 | struct dquot *transfer_to[MAXQUOTAS] = { }; |
52a9ee28 | 1135 | int qtype; |
b891fa50 ER |
1136 | int had_lock; |
1137 | struct ocfs2_lock_holder oh; | |
ccd979bd | 1138 | |
468eedde TM |
1139 | trace_ocfs2_setattr(inode, dentry, |
1140 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
1141 | dentry->d_name.len, dentry->d_name.name, | |
1142 | attr->ia_valid, attr->ia_mode, | |
ba613560 EB |
1143 | from_kuid(&init_user_ns, attr->ia_uid), |
1144 | from_kgid(&init_user_ns, attr->ia_gid)); | |
ccd979bd | 1145 | |
bc535809 SM |
1146 | /* ensuring we don't even attempt to truncate a symlink */ |
1147 | if (S_ISLNK(inode->i_mode)) | |
1148 | attr->ia_valid &= ~ATTR_SIZE; | |
1149 | ||
ccd979bd MF |
1150 | #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \ |
1151 | | ATTR_GID | ATTR_UID | ATTR_MODE) | |
468eedde | 1152 | if (!(attr->ia_valid & OCFS2_VALID_ATTRS)) |
ccd979bd | 1153 | return 0; |
ccd979bd | 1154 | |
31051c85 | 1155 | status = setattr_prepare(dentry, attr); |
ccd979bd MF |
1156 | if (status) |
1157 | return status; | |
1158 | ||
9c89fe0a JK |
1159 | if (is_quota_modification(inode, attr)) { |
1160 | status = dquot_initialize(inode); | |
1161 | if (status) | |
1162 | return status; | |
1163 | } | |
ccd979bd MF |
1164 | size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE; |
1165 | if (size_change) { | |
28f5a8a7 | 1166 | /* |
1167 | * Here we should wait dio to finish before inode lock | |
1168 | * to avoid a deadlock between ocfs2_setattr() and | |
1169 | * ocfs2_dio_end_io_write() | |
1170 | */ | |
1171 | inode_dio_wait(inode); | |
1172 | ||
ccd979bd MF |
1173 | status = ocfs2_rw_lock(inode, 1); |
1174 | if (status < 0) { | |
1175 | mlog_errno(status); | |
1176 | goto bail; | |
1177 | } | |
1178 | } | |
1179 | ||
b891fa50 ER |
1180 | had_lock = ocfs2_inode_lock_tracker(inode, &bh, 1, &oh); |
1181 | if (had_lock < 0) { | |
1182 | status = had_lock; | |
ccd979bd | 1183 | goto bail_unlock_rw; |
b891fa50 ER |
1184 | } else if (had_lock) { |
1185 | /* | |
1186 | * As far as we know, ocfs2_setattr() could only be the first | |
1187 | * VFS entry point in the call chain of recursive cluster | |
1188 | * locking issue. | |
1189 | * | |
1190 | * For instance: | |
1191 | * chmod_common() | |
1192 | * notify_change() | |
1193 | * ocfs2_setattr() | |
1194 | * posix_acl_chmod() | |
1195 | * ocfs2_iop_get_acl() | |
1196 | * | |
1197 | * But, we're not 100% sure if it's always true, because the | |
1198 | * ordering of the VFS entry points in the call chain is out | |
1199 | * of our control. So, we'd better dump the stack here to | |
1200 | * catch the other cases of recursive locking. | |
1201 | */ | |
1202 | mlog(ML_ERROR, "Another case of recursive locking:\n"); | |
1203 | dump_stack(); | |
ccd979bd | 1204 | } |
3d46a44a | 1205 | inode_locked = 1; |
ccd979bd | 1206 | |
d62e74be | 1207 | if (size_change) { |
5051f768 WW |
1208 | status = inode_newsize_ok(inode, attr->ia_size); |
1209 | if (status) | |
ce76fd30 | 1210 | goto bail_unlock; |
ce76fd30 | 1211 | |
d62e74be | 1212 | if (i_size_read(inode) >= attr->ia_size) { |
2b4e30fb JB |
1213 | if (ocfs2_should_order_data(inode)) { |
1214 | status = ocfs2_begin_ordered_truncate(inode, | |
1215 | attr->ia_size); | |
1216 | if (status) | |
1217 | goto bail_unlock; | |
1218 | } | |
ccd979bd | 1219 | status = ocfs2_truncate_file(inode, bh, attr->ia_size); |
2b4e30fb | 1220 | } else |
65ed39d6 | 1221 | status = ocfs2_extend_file(inode, bh, attr->ia_size); |
ccd979bd MF |
1222 | if (status < 0) { |
1223 | if (status != -ENOSPC) | |
1224 | mlog_errno(status); | |
1225 | status = -ENOSPC; | |
1226 | goto bail_unlock; | |
1227 | } | |
1228 | } | |
1229 | ||
488c8ef0 EB |
1230 | if ((attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) || |
1231 | (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) { | |
65bac575 JK |
1232 | /* |
1233 | * Gather pointers to quota structures so that allocation / | |
1234 | * freeing of quota structures happens here and not inside | |
b43fa828 | 1235 | * dquot_transfer() where we have problems with lock ordering |
65bac575 | 1236 | */ |
488c8ef0 | 1237 | if (attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid) |
a90714c1 JK |
1238 | && OCFS2_HAS_RO_COMPAT_FEATURE(sb, |
1239 | OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) { | |
aca645a6 | 1240 | transfer_to[USRQUOTA] = dqget(sb, make_kqid_uid(attr->ia_uid)); |
6184fc0b JK |
1241 | if (IS_ERR(transfer_to[USRQUOTA])) { |
1242 | status = PTR_ERR(transfer_to[USRQUOTA]); | |
a90714c1 | 1243 | goto bail_unlock; |
65bac575 | 1244 | } |
a90714c1 | 1245 | } |
488c8ef0 | 1246 | if (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid) |
a90714c1 JK |
1247 | && OCFS2_HAS_RO_COMPAT_FEATURE(sb, |
1248 | OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) { | |
aca645a6 | 1249 | transfer_to[GRPQUOTA] = dqget(sb, make_kqid_gid(attr->ia_gid)); |
6184fc0b JK |
1250 | if (IS_ERR(transfer_to[GRPQUOTA])) { |
1251 | status = PTR_ERR(transfer_to[GRPQUOTA]); | |
a90714c1 | 1252 | goto bail_unlock; |
65bac575 | 1253 | } |
a90714c1 | 1254 | } |
65bac575 JK |
1255 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS + |
1256 | 2 * ocfs2_quota_trans_credits(sb)); | |
a90714c1 JK |
1257 | if (IS_ERR(handle)) { |
1258 | status = PTR_ERR(handle); | |
1259 | mlog_errno(status); | |
1260 | goto bail_unlock; | |
1261 | } | |
52a9ee28 | 1262 | status = __dquot_transfer(inode, transfer_to); |
a90714c1 JK |
1263 | if (status < 0) |
1264 | goto bail_commit; | |
1265 | } else { | |
1266 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); | |
1267 | if (IS_ERR(handle)) { | |
1268 | status = PTR_ERR(handle); | |
1269 | mlog_errno(status); | |
1270 | goto bail_unlock; | |
1271 | } | |
ccd979bd MF |
1272 | } |
1273 | ||
1025774c CH |
1274 | setattr_copy(inode, attr); |
1275 | mark_inode_dirty(inode); | |
1276 | ||
ccd979bd MF |
1277 | status = ocfs2_mark_inode_dirty(handle, inode, bh); |
1278 | if (status < 0) | |
1279 | mlog_errno(status); | |
1280 | ||
1281 | bail_commit: | |
02dc1af4 | 1282 | ocfs2_commit_trans(osb, handle); |
ccd979bd | 1283 | bail_unlock: |
b891fa50 ER |
1284 | if (status && inode_locked) { |
1285 | ocfs2_inode_unlock_tracker(inode, 1, &oh, had_lock); | |
3d46a44a TS |
1286 | inode_locked = 0; |
1287 | } | |
ccd979bd MF |
1288 | bail_unlock_rw: |
1289 | if (size_change) | |
1290 | ocfs2_rw_unlock(inode, 1); | |
1291 | bail: | |
ccd979bd | 1292 | |
65bac575 | 1293 | /* Release quota pointers in case we acquired them */ |
52362810 | 1294 | for (qtype = 0; qtype < OCFS2_MAXQUOTAS; qtype++) |
65bac575 | 1295 | dqput(transfer_to[qtype]); |
65bac575 | 1296 | |
060bc66d | 1297 | if (!status && attr->ia_valid & ATTR_MODE) { |
5ee0fbd5 | 1298 | status = ocfs2_acl_chmod(inode, bh); |
060bc66d TY |
1299 | if (status < 0) |
1300 | mlog_errno(status); | |
1301 | } | |
3d46a44a | 1302 | if (inode_locked) |
b891fa50 | 1303 | ocfs2_inode_unlock_tracker(inode, 1, &oh, had_lock); |
060bc66d | 1304 | |
5ee0fbd5 | 1305 | brelse(bh); |
ccd979bd MF |
1306 | return status; |
1307 | } | |
1308 | ||
a528d35e DH |
1309 | int ocfs2_getattr(const struct path *path, struct kstat *stat, |
1310 | u32 request_mask, unsigned int flags) | |
ccd979bd | 1311 | { |
a528d35e DH |
1312 | struct inode *inode = d_inode(path->dentry); |
1313 | struct super_block *sb = path->dentry->d_sb; | |
ccd979bd MF |
1314 | struct ocfs2_super *osb = sb->s_fs_info; |
1315 | int err; | |
1316 | ||
a528d35e | 1317 | err = ocfs2_inode_revalidate(path->dentry); |
ccd979bd MF |
1318 | if (err) { |
1319 | if (err != -ENOENT) | |
1320 | mlog_errno(err); | |
1321 | goto bail; | |
1322 | } | |
1323 | ||
1324 | generic_fillattr(inode, stat); | |
d6364627 JH |
1325 | /* |
1326 | * If there is inline data in the inode, the inode will normally not | |
1327 | * have data blocks allocated (it may have an external xattr block). | |
1328 | * Report at least one sector for such files, so tools like tar, rsync, | |
1329 | * others don't incorrectly think the file is completely sparse. | |
1330 | */ | |
1331 | if (unlikely(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)) | |
1332 | stat->blocks += (stat->size + 511)>>9; | |
ccd979bd MF |
1333 | |
1334 | /* We set the blksize from the cluster size for performance */ | |
1335 | stat->blksize = osb->s_clustersize; | |
1336 | ||
1337 | bail: | |
ccd979bd MF |
1338 | return err; |
1339 | } | |
1340 | ||
10556cb2 | 1341 | int ocfs2_permission(struct inode *inode, int mask) |
d38eb8db | 1342 | { |
b891fa50 ER |
1343 | int ret, had_lock; |
1344 | struct ocfs2_lock_holder oh; | |
d38eb8db | 1345 | |
10556cb2 | 1346 | if (mask & MAY_NOT_BLOCK) |
b74c79e9 NP |
1347 | return -ECHILD; |
1348 | ||
b891fa50 ER |
1349 | had_lock = ocfs2_inode_lock_tracker(inode, NULL, 0, &oh); |
1350 | if (had_lock < 0) { | |
1351 | ret = had_lock; | |
d38eb8db | 1352 | goto out; |
b891fa50 ER |
1353 | } else if (had_lock) { |
1354 | /* See comments in ocfs2_setattr() for details. | |
1355 | * The call chain of this case could be: | |
1356 | * do_sys_open() | |
1357 | * may_open() | |
1358 | * inode_permission() | |
1359 | * ocfs2_permission() | |
1360 | * ocfs2_iop_get_acl() | |
1361 | */ | |
1362 | mlog(ML_ERROR, "Another case of recursive locking:\n"); | |
1363 | dump_stack(); | |
d38eb8db TY |
1364 | } |
1365 | ||
2830ba7f | 1366 | ret = generic_permission(inode, mask); |
d38eb8db | 1367 | |
b891fa50 | 1368 | ocfs2_inode_unlock_tracker(inode, 0, &oh, had_lock); |
d38eb8db | 1369 | out: |
d38eb8db TY |
1370 | return ret; |
1371 | } | |
1372 | ||
b2580103 MF |
1373 | static int __ocfs2_write_remove_suid(struct inode *inode, |
1374 | struct buffer_head *bh) | |
ccd979bd MF |
1375 | { |
1376 | int ret; | |
1fabe148 | 1377 | handle_t *handle; |
ccd979bd MF |
1378 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
1379 | struct ocfs2_dinode *di; | |
1380 | ||
468eedde TM |
1381 | trace_ocfs2_write_remove_suid( |
1382 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
1383 | inode->i_mode); | |
ccd979bd | 1384 | |
65eff9cc | 1385 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
fa38e92c JK |
1386 | if (IS_ERR(handle)) { |
1387 | ret = PTR_ERR(handle); | |
ccd979bd MF |
1388 | mlog_errno(ret); |
1389 | goto out; | |
1390 | } | |
1391 | ||
0cf2f763 | 1392 | ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh, |
13723d00 | 1393 | OCFS2_JOURNAL_ACCESS_WRITE); |
ccd979bd MF |
1394 | if (ret < 0) { |
1395 | mlog_errno(ret); | |
b2580103 | 1396 | goto out_trans; |
ccd979bd MF |
1397 | } |
1398 | ||
1399 | inode->i_mode &= ~S_ISUID; | |
1400 | if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP)) | |
1401 | inode->i_mode &= ~S_ISGID; | |
1402 | ||
1403 | di = (struct ocfs2_dinode *) bh->b_data; | |
1404 | di->i_mode = cpu_to_le16(inode->i_mode); | |
6fdb702d | 1405 | ocfs2_update_inode_fsync_trans(handle, inode, 0); |
ccd979bd | 1406 | |
ec20cec7 | 1407 | ocfs2_journal_dirty(handle, bh); |
b2580103 | 1408 | |
ccd979bd | 1409 | out_trans: |
02dc1af4 | 1410 | ocfs2_commit_trans(osb, handle); |
ccd979bd | 1411 | out: |
ccd979bd MF |
1412 | return ret; |
1413 | } | |
1414 | ||
b2580103 MF |
1415 | static int ocfs2_write_remove_suid(struct inode *inode) |
1416 | { | |
1417 | int ret; | |
1418 | struct buffer_head *bh = NULL; | |
b2580103 | 1419 | |
b657c95c | 1420 | ret = ocfs2_read_inode_block(inode, &bh); |
b2580103 MF |
1421 | if (ret < 0) { |
1422 | mlog_errno(ret); | |
1423 | goto out; | |
1424 | } | |
1425 | ||
1426 | ret = __ocfs2_write_remove_suid(inode, bh); | |
1427 | out: | |
1428 | brelse(bh); | |
1429 | return ret; | |
1430 | } | |
1431 | ||
2ae99a60 MF |
1432 | /* |
1433 | * Allocate enough extents to cover the region starting at byte offset | |
1434 | * start for len bytes. Existing extents are skipped, any extents | |
1435 | * added are marked as "unwritten". | |
1436 | */ | |
1437 | static int ocfs2_allocate_unwritten_extents(struct inode *inode, | |
1438 | u64 start, u64 len) | |
1439 | { | |
1440 | int ret; | |
1441 | u32 cpos, phys_cpos, clusters, alloc_size; | |
1afc32b9 MF |
1442 | u64 end = start + len; |
1443 | struct buffer_head *di_bh = NULL; | |
1444 | ||
1445 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) { | |
b657c95c | 1446 | ret = ocfs2_read_inode_block(inode, &di_bh); |
1afc32b9 MF |
1447 | if (ret) { |
1448 | mlog_errno(ret); | |
1449 | goto out; | |
1450 | } | |
1451 | ||
1452 | /* | |
1453 | * Nothing to do if the requested reservation range | |
1454 | * fits within the inode. | |
1455 | */ | |
1456 | if (ocfs2_size_fits_inline_data(di_bh, end)) | |
1457 | goto out; | |
1458 | ||
1459 | ret = ocfs2_convert_inline_data_to_extents(inode, di_bh); | |
1460 | if (ret) { | |
1461 | mlog_errno(ret); | |
1462 | goto out; | |
1463 | } | |
1464 | } | |
2ae99a60 MF |
1465 | |
1466 | /* | |
1467 | * We consider both start and len to be inclusive. | |
1468 | */ | |
1469 | cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits; | |
1470 | clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len); | |
1471 | clusters -= cpos; | |
1472 | ||
1473 | while (clusters) { | |
1474 | ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, | |
1475 | &alloc_size, NULL); | |
1476 | if (ret) { | |
1477 | mlog_errno(ret); | |
1478 | goto out; | |
1479 | } | |
1480 | ||
1481 | /* | |
1482 | * Hole or existing extent len can be arbitrary, so | |
1483 | * cap it to our own allocation request. | |
1484 | */ | |
1485 | if (alloc_size > clusters) | |
1486 | alloc_size = clusters; | |
1487 | ||
1488 | if (phys_cpos) { | |
1489 | /* | |
1490 | * We already have an allocation at this | |
1491 | * region so we can safely skip it. | |
1492 | */ | |
1493 | goto next; | |
1494 | } | |
1495 | ||
5bc55d65 | 1496 | ret = ocfs2_extend_allocation(inode, cpos, alloc_size, 1); |
2ae99a60 MF |
1497 | if (ret) { |
1498 | if (ret != -ENOSPC) | |
1499 | mlog_errno(ret); | |
1500 | goto out; | |
1501 | } | |
1502 | ||
1503 | next: | |
1504 | cpos += alloc_size; | |
1505 | clusters -= alloc_size; | |
1506 | } | |
1507 | ||
1508 | ret = 0; | |
1509 | out: | |
1afc32b9 MF |
1510 | |
1511 | brelse(di_bh); | |
2ae99a60 MF |
1512 | return ret; |
1513 | } | |
1514 | ||
063c4561 MF |
1515 | /* |
1516 | * Truncate a byte range, avoiding pages within partial clusters. This | |
1517 | * preserves those pages for the zeroing code to write to. | |
1518 | */ | |
1519 | static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start, | |
1520 | u64 byte_len) | |
1521 | { | |
1522 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1523 | loff_t start, end; | |
1524 | struct address_space *mapping = inode->i_mapping; | |
1525 | ||
1526 | start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start); | |
1527 | end = byte_start + byte_len; | |
1528 | end = end & ~(osb->s_clustersize - 1); | |
1529 | ||
1530 | if (start < end) { | |
1531 | unmap_mapping_range(mapping, start, end - start, 0); | |
1532 | truncate_inode_pages_range(mapping, start, end - 1); | |
1533 | } | |
1534 | } | |
1535 | ||
1536 | static int ocfs2_zero_partial_clusters(struct inode *inode, | |
1537 | u64 start, u64 len) | |
1538 | { | |
1539 | int ret = 0; | |
d21c353d AS |
1540 | u64 tmpend = 0; |
1541 | u64 end = start + len; | |
063c4561 MF |
1542 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
1543 | unsigned int csize = osb->s_clustersize; | |
1544 | handle_t *handle; | |
1545 | ||
1546 | /* | |
1547 | * The "start" and "end" values are NOT necessarily part of | |
1548 | * the range whose allocation is being deleted. Rather, this | |
1549 | * is what the user passed in with the request. We must zero | |
1550 | * partial clusters here. There's no need to worry about | |
1551 | * physical allocation - the zeroing code knows to skip holes. | |
1552 | */ | |
468eedde TM |
1553 | trace_ocfs2_zero_partial_clusters( |
1554 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
1555 | (unsigned long long)start, (unsigned long long)end); | |
063c4561 MF |
1556 | |
1557 | /* | |
1558 | * If both edges are on a cluster boundary then there's no | |
1559 | * zeroing required as the region is part of the allocation to | |
1560 | * be truncated. | |
1561 | */ | |
1562 | if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0) | |
1563 | goto out; | |
1564 | ||
1565 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); | |
fa38e92c JK |
1566 | if (IS_ERR(handle)) { |
1567 | ret = PTR_ERR(handle); | |
063c4561 MF |
1568 | mlog_errno(ret); |
1569 | goto out; | |
1570 | } | |
1571 | ||
1572 | /* | |
d21c353d AS |
1573 | * If start is on a cluster boundary and end is somewhere in another |
1574 | * cluster, we have not COWed the cluster starting at start, unless | |
1575 | * end is also within the same cluster. So, in this case, we skip this | |
1576 | * first call to ocfs2_zero_range_for_truncate() truncate and move on | |
1577 | * to the next one. | |
063c4561 | 1578 | */ |
d21c353d AS |
1579 | if ((start & (csize - 1)) != 0) { |
1580 | /* | |
1581 | * We want to get the byte offset of the end of the 1st | |
1582 | * cluster. | |
1583 | */ | |
1584 | tmpend = (u64)osb->s_clustersize + | |
1585 | (start & ~(osb->s_clustersize - 1)); | |
1586 | if (tmpend > end) | |
1587 | tmpend = end; | |
063c4561 | 1588 | |
d21c353d AS |
1589 | trace_ocfs2_zero_partial_clusters_range1( |
1590 | (unsigned long long)start, | |
1591 | (unsigned long long)tmpend); | |
063c4561 | 1592 | |
d21c353d AS |
1593 | ret = ocfs2_zero_range_for_truncate(inode, handle, start, |
1594 | tmpend); | |
1595 | if (ret) | |
1596 | mlog_errno(ret); | |
1597 | } | |
063c4561 MF |
1598 | |
1599 | if (tmpend < end) { | |
1600 | /* | |
1601 | * This may make start and end equal, but the zeroing | |
1602 | * code will skip any work in that case so there's no | |
1603 | * need to catch it up here. | |
1604 | */ | |
1605 | start = end & ~(osb->s_clustersize - 1); | |
1606 | ||
468eedde TM |
1607 | trace_ocfs2_zero_partial_clusters_range2( |
1608 | (unsigned long long)start, (unsigned long long)end); | |
063c4561 MF |
1609 | |
1610 | ret = ocfs2_zero_range_for_truncate(inode, handle, start, end); | |
1611 | if (ret) | |
1612 | mlog_errno(ret); | |
1613 | } | |
6fdb702d | 1614 | ocfs2_update_inode_fsync_trans(handle, inode, 1); |
063c4561 MF |
1615 | |
1616 | ocfs2_commit_trans(osb, handle); | |
1617 | out: | |
1618 | return ret; | |
1619 | } | |
1620 | ||
c1631d4a TY |
1621 | static int ocfs2_find_rec(struct ocfs2_extent_list *el, u32 pos) |
1622 | { | |
1623 | int i; | |
1624 | struct ocfs2_extent_rec *rec = NULL; | |
1625 | ||
1626 | for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) { | |
1627 | ||
1628 | rec = &el->l_recs[i]; | |
1629 | ||
1630 | if (le32_to_cpu(rec->e_cpos) < pos) | |
1631 | break; | |
1632 | } | |
1633 | ||
1634 | return i; | |
1635 | } | |
1636 | ||
1637 | /* | |
1638 | * Helper to calculate the punching pos and length in one run, we handle the | |
1639 | * following three cases in order: | |
1640 | * | |
1641 | * - remove the entire record | |
1642 | * - remove a partial record | |
1643 | * - no record needs to be removed (hole-punching completed) | |
1644 | */ | |
1645 | static void ocfs2_calc_trunc_pos(struct inode *inode, | |
1646 | struct ocfs2_extent_list *el, | |
1647 | struct ocfs2_extent_rec *rec, | |
1648 | u32 trunc_start, u32 *trunc_cpos, | |
1649 | u32 *trunc_len, u32 *trunc_end, | |
1650 | u64 *blkno, int *done) | |
1651 | { | |
1652 | int ret = 0; | |
1653 | u32 coff, range; | |
1654 | ||
1655 | range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec); | |
1656 | ||
1657 | if (le32_to_cpu(rec->e_cpos) >= trunc_start) { | |
9a790ba1 TY |
1658 | /* |
1659 | * remove an entire extent record. | |
1660 | */ | |
c1631d4a TY |
1661 | *trunc_cpos = le32_to_cpu(rec->e_cpos); |
1662 | /* | |
1663 | * Skip holes if any. | |
1664 | */ | |
1665 | if (range < *trunc_end) | |
1666 | *trunc_end = range; | |
1667 | *trunc_len = *trunc_end - le32_to_cpu(rec->e_cpos); | |
1668 | *blkno = le64_to_cpu(rec->e_blkno); | |
1669 | *trunc_end = le32_to_cpu(rec->e_cpos); | |
1670 | } else if (range > trunc_start) { | |
9a790ba1 TY |
1671 | /* |
1672 | * remove a partial extent record, which means we're | |
1673 | * removing the last extent record. | |
1674 | */ | |
c1631d4a | 1675 | *trunc_cpos = trunc_start; |
9a790ba1 TY |
1676 | /* |
1677 | * skip hole if any. | |
1678 | */ | |
1679 | if (range < *trunc_end) | |
1680 | *trunc_end = range; | |
c1631d4a TY |
1681 | *trunc_len = *trunc_end - trunc_start; |
1682 | coff = trunc_start - le32_to_cpu(rec->e_cpos); | |
1683 | *blkno = le64_to_cpu(rec->e_blkno) + | |
1684 | ocfs2_clusters_to_blocks(inode->i_sb, coff); | |
1685 | *trunc_end = trunc_start; | |
1686 | } else { | |
1687 | /* | |
1688 | * It may have two following possibilities: | |
1689 | * | |
1690 | * - last record has been removed | |
1691 | * - trunc_start was within a hole | |
1692 | * | |
1693 | * both two cases mean the completion of hole punching. | |
1694 | */ | |
1695 | ret = 1; | |
1696 | } | |
1697 | ||
1698 | *done = ret; | |
1699 | } | |
1700 | ||
29ac8e85 DW |
1701 | int ocfs2_remove_inode_range(struct inode *inode, |
1702 | struct buffer_head *di_bh, u64 byte_start, | |
1703 | u64 byte_len) | |
063c4561 | 1704 | { |
c1631d4a TY |
1705 | int ret = 0, flags = 0, done = 0, i; |
1706 | u32 trunc_start, trunc_len, trunc_end, trunc_cpos, phys_cpos; | |
1707 | u32 cluster_in_el; | |
063c4561 MF |
1708 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
1709 | struct ocfs2_cached_dealloc_ctxt dealloc; | |
b1967d0e | 1710 | struct address_space *mapping = inode->i_mapping; |
fecc0112 | 1711 | struct ocfs2_extent_tree et; |
c1631d4a TY |
1712 | struct ocfs2_path *path = NULL; |
1713 | struct ocfs2_extent_list *el = NULL; | |
1714 | struct ocfs2_extent_rec *rec = NULL; | |
e8aec068 | 1715 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; |
c1631d4a | 1716 | u64 blkno, refcount_loc = le64_to_cpu(di->i_refcount_loc); |
063c4561 | 1717 | |
5e404e9e | 1718 | ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh); |
063c4561 MF |
1719 | ocfs2_init_dealloc_ctxt(&dealloc); |
1720 | ||
468eedde TM |
1721 | trace_ocfs2_remove_inode_range( |
1722 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
1723 | (unsigned long long)byte_start, | |
1724 | (unsigned long long)byte_len); | |
1725 | ||
063c4561 MF |
1726 | if (byte_len == 0) |
1727 | return 0; | |
1728 | ||
1afc32b9 MF |
1729 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) { |
1730 | ret = ocfs2_truncate_inline(inode, di_bh, byte_start, | |
b1967d0e MF |
1731 | byte_start + byte_len, 0); |
1732 | if (ret) { | |
1afc32b9 | 1733 | mlog_errno(ret); |
b1967d0e MF |
1734 | goto out; |
1735 | } | |
1736 | /* | |
1737 | * There's no need to get fancy with the page cache | |
1738 | * truncate of an inline-data inode. We're talking | |
1739 | * about less than a page here, which will be cached | |
1740 | * in the dinode buffer anyway. | |
1741 | */ | |
1742 | unmap_mapping_range(mapping, 0, 0, 0); | |
1743 | truncate_inode_pages(mapping, 0); | |
1744 | goto out; | |
1afc32b9 MF |
1745 | } |
1746 | ||
e8aec068 TY |
1747 | /* |
1748 | * For reflinks, we may need to CoW 2 clusters which might be | |
1749 | * partially zero'd later, if hole's start and end offset were | |
1750 | * within one cluster(means is not exactly aligned to clustersize). | |
1751 | */ | |
1752 | ||
84e40080 | 1753 | if (ocfs2_is_refcount_inode(inode)) { |
e8aec068 TY |
1754 | ret = ocfs2_cow_file_pos(inode, di_bh, byte_start); |
1755 | if (ret) { | |
1756 | mlog_errno(ret); | |
1757 | goto out; | |
1758 | } | |
1759 | ||
1760 | ret = ocfs2_cow_file_pos(inode, di_bh, byte_start + byte_len); | |
1761 | if (ret) { | |
1762 | mlog_errno(ret); | |
1763 | goto out; | |
1764 | } | |
1765 | } | |
1766 | ||
063c4561 | 1767 | trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start); |
c1631d4a TY |
1768 | trunc_end = (byte_start + byte_len) >> osb->s_clustersize_bits; |
1769 | cluster_in_el = trunc_end; | |
063c4561 | 1770 | |
063c4561 MF |
1771 | ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len); |
1772 | if (ret) { | |
1773 | mlog_errno(ret); | |
1774 | goto out; | |
1775 | } | |
1776 | ||
c1631d4a TY |
1777 | path = ocfs2_new_path_from_et(&et); |
1778 | if (!path) { | |
1779 | ret = -ENOMEM; | |
1780 | mlog_errno(ret); | |
1781 | goto out; | |
1782 | } | |
1783 | ||
1784 | while (trunc_end > trunc_start) { | |
1785 | ||
1786 | ret = ocfs2_find_path(INODE_CACHE(inode), path, | |
1787 | cluster_in_el); | |
063c4561 MF |
1788 | if (ret) { |
1789 | mlog_errno(ret); | |
1790 | goto out; | |
1791 | } | |
1792 | ||
c1631d4a | 1793 | el = path_leaf_el(path); |
063c4561 | 1794 | |
c1631d4a TY |
1795 | i = ocfs2_find_rec(el, trunc_end); |
1796 | /* | |
1797 | * Need to go to previous extent block. | |
1798 | */ | |
1799 | if (i < 0) { | |
1800 | if (path->p_tree_depth == 0) | |
1801 | break; | |
063c4561 | 1802 | |
c1631d4a TY |
1803 | ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb, |
1804 | path, | |
1805 | &cluster_in_el); | |
063c4561 MF |
1806 | if (ret) { |
1807 | mlog_errno(ret); | |
1808 | goto out; | |
1809 | } | |
c1631d4a TY |
1810 | |
1811 | /* | |
1812 | * We've reached the leftmost extent block, | |
1813 | * it's safe to leave. | |
1814 | */ | |
1815 | if (cluster_in_el == 0) | |
1816 | break; | |
1817 | ||
1818 | /* | |
1819 | * The 'pos' searched for previous extent block is | |
1820 | * always one cluster less than actual trunc_end. | |
1821 | */ | |
1822 | trunc_end = cluster_in_el + 1; | |
1823 | ||
1824 | ocfs2_reinit_path(path, 1); | |
1825 | ||
1826 | continue; | |
1827 | ||
1828 | } else | |
1829 | rec = &el->l_recs[i]; | |
1830 | ||
1831 | ocfs2_calc_trunc_pos(inode, el, rec, trunc_start, &trunc_cpos, | |
1832 | &trunc_len, &trunc_end, &blkno, &done); | |
1833 | if (done) | |
1834 | break; | |
1835 | ||
1836 | flags = rec->e_flags; | |
1837 | phys_cpos = ocfs2_blocks_to_clusters(inode->i_sb, blkno); | |
1838 | ||
1839 | ret = ocfs2_remove_btree_range(inode, &et, trunc_cpos, | |
1840 | phys_cpos, trunc_len, flags, | |
f62f12b3 | 1841 | &dealloc, refcount_loc, false); |
c1631d4a TY |
1842 | if (ret < 0) { |
1843 | mlog_errno(ret); | |
1844 | goto out; | |
063c4561 MF |
1845 | } |
1846 | ||
c1631d4a TY |
1847 | cluster_in_el = trunc_end; |
1848 | ||
1849 | ocfs2_reinit_path(path, 1); | |
063c4561 MF |
1850 | } |
1851 | ||
1852 | ocfs2_truncate_cluster_pages(inode, byte_start, byte_len); | |
1853 | ||
1854 | out: | |
7aebff18 | 1855 | ocfs2_free_path(path); |
063c4561 MF |
1856 | ocfs2_schedule_truncate_log_flush(osb, 1); |
1857 | ocfs2_run_deallocs(osb, &dealloc); | |
1858 | ||
1859 | return ret; | |
1860 | } | |
1861 | ||
b2580103 MF |
1862 | /* |
1863 | * Parts of this function taken from xfs_change_file_space() | |
1864 | */ | |
385820a3 MF |
1865 | static int __ocfs2_change_file_space(struct file *file, struct inode *inode, |
1866 | loff_t f_pos, unsigned int cmd, | |
1867 | struct ocfs2_space_resv *sr, | |
1868 | int change_size) | |
b2580103 MF |
1869 | { |
1870 | int ret; | |
1871 | s64 llen; | |
385820a3 | 1872 | loff_t size; |
b2580103 MF |
1873 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
1874 | struct buffer_head *di_bh = NULL; | |
1875 | handle_t *handle; | |
a00cce35 | 1876 | unsigned long long max_off = inode->i_sb->s_maxbytes; |
b2580103 | 1877 | |
b2580103 MF |
1878 | if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb)) |
1879 | return -EROFS; | |
1880 | ||
5955102c | 1881 | inode_lock(inode); |
b2580103 MF |
1882 | |
1883 | /* | |
1884 | * This prevents concurrent writes on other nodes | |
1885 | */ | |
1886 | ret = ocfs2_rw_lock(inode, 1); | |
1887 | if (ret) { | |
1888 | mlog_errno(ret); | |
1889 | goto out; | |
1890 | } | |
1891 | ||
e63aecb6 | 1892 | ret = ocfs2_inode_lock(inode, &di_bh, 1); |
b2580103 MF |
1893 | if (ret) { |
1894 | mlog_errno(ret); | |
1895 | goto out_rw_unlock; | |
1896 | } | |
1897 | ||
1898 | if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) { | |
1899 | ret = -EPERM; | |
e63aecb6 | 1900 | goto out_inode_unlock; |
b2580103 MF |
1901 | } |
1902 | ||
1903 | switch (sr->l_whence) { | |
1904 | case 0: /*SEEK_SET*/ | |
1905 | break; | |
1906 | case 1: /*SEEK_CUR*/ | |
385820a3 | 1907 | sr->l_start += f_pos; |
b2580103 MF |
1908 | break; |
1909 | case 2: /*SEEK_END*/ | |
1910 | sr->l_start += i_size_read(inode); | |
1911 | break; | |
1912 | default: | |
1913 | ret = -EINVAL; | |
e63aecb6 | 1914 | goto out_inode_unlock; |
b2580103 MF |
1915 | } |
1916 | sr->l_whence = 0; | |
1917 | ||
1918 | llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len; | |
1919 | ||
1920 | if (sr->l_start < 0 | |
1921 | || sr->l_start > max_off | |
1922 | || (sr->l_start + llen) < 0 | |
1923 | || (sr->l_start + llen) > max_off) { | |
1924 | ret = -EINVAL; | |
e63aecb6 | 1925 | goto out_inode_unlock; |
b2580103 | 1926 | } |
385820a3 | 1927 | size = sr->l_start + sr->l_len; |
b2580103 | 1928 | |
a2a3b398 TS |
1929 | if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64 || |
1930 | cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) { | |
b2580103 MF |
1931 | if (sr->l_len <= 0) { |
1932 | ret = -EINVAL; | |
e63aecb6 | 1933 | goto out_inode_unlock; |
b2580103 MF |
1934 | } |
1935 | } | |
1936 | ||
385820a3 | 1937 | if (file && should_remove_suid(file->f_path.dentry)) { |
b2580103 MF |
1938 | ret = __ocfs2_write_remove_suid(inode, di_bh); |
1939 | if (ret) { | |
1940 | mlog_errno(ret); | |
e63aecb6 | 1941 | goto out_inode_unlock; |
b2580103 MF |
1942 | } |
1943 | } | |
1944 | ||
1945 | down_write(&OCFS2_I(inode)->ip_alloc_sem); | |
1946 | switch (cmd) { | |
1947 | case OCFS2_IOC_RESVSP: | |
1948 | case OCFS2_IOC_RESVSP64: | |
1949 | /* | |
1950 | * This takes unsigned offsets, but the signed ones we | |
1951 | * pass have been checked against overflow above. | |
1952 | */ | |
1953 | ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start, | |
1954 | sr->l_len); | |
1955 | break; | |
1956 | case OCFS2_IOC_UNRESVSP: | |
1957 | case OCFS2_IOC_UNRESVSP64: | |
1958 | ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start, | |
1959 | sr->l_len); | |
1960 | break; | |
1961 | default: | |
1962 | ret = -EINVAL; | |
1963 | } | |
1964 | up_write(&OCFS2_I(inode)->ip_alloc_sem); | |
1965 | if (ret) { | |
1966 | mlog_errno(ret); | |
e63aecb6 | 1967 | goto out_inode_unlock; |
b2580103 MF |
1968 | } |
1969 | ||
1970 | /* | |
1971 | * We update c/mtime for these changes | |
1972 | */ | |
1973 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); | |
1974 | if (IS_ERR(handle)) { | |
1975 | ret = PTR_ERR(handle); | |
1976 | mlog_errno(ret); | |
e63aecb6 | 1977 | goto out_inode_unlock; |
b2580103 MF |
1978 | } |
1979 | ||
385820a3 MF |
1980 | if (change_size && i_size_read(inode) < size) |
1981 | i_size_write(inode, size); | |
1982 | ||
078cd827 | 1983 | inode->i_ctime = inode->i_mtime = current_time(inode); |
b2580103 MF |
1984 | ret = ocfs2_mark_inode_dirty(handle, inode, di_bh); |
1985 | if (ret < 0) | |
1986 | mlog_errno(ret); | |
1987 | ||
a4e08d00 | 1988 | if (file && (file->f_flags & O_SYNC)) |
df295d4a MF |
1989 | handle->h_sync = 1; |
1990 | ||
b2580103 MF |
1991 | ocfs2_commit_trans(osb, handle); |
1992 | ||
e63aecb6 | 1993 | out_inode_unlock: |
b2580103 | 1994 | brelse(di_bh); |
e63aecb6 | 1995 | ocfs2_inode_unlock(inode, 1); |
b2580103 MF |
1996 | out_rw_unlock: |
1997 | ocfs2_rw_unlock(inode, 1); | |
1998 | ||
b2580103 | 1999 | out: |
5955102c | 2000 | inode_unlock(inode); |
b2580103 MF |
2001 | return ret; |
2002 | } | |
2003 | ||
385820a3 MF |
2004 | int ocfs2_change_file_space(struct file *file, unsigned int cmd, |
2005 | struct ocfs2_space_resv *sr) | |
2006 | { | |
496ad9aa | 2007 | struct inode *inode = file_inode(file); |
c19a28e1 | 2008 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
fef6925c | 2009 | int ret; |
385820a3 MF |
2010 | |
2011 | if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) && | |
2012 | !ocfs2_writes_unwritten_extents(osb)) | |
2013 | return -ENOTTY; | |
2014 | else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) && | |
2015 | !ocfs2_sparse_alloc(osb)) | |
2016 | return -ENOTTY; | |
2017 | ||
2018 | if (!S_ISREG(inode->i_mode)) | |
2019 | return -EINVAL; | |
2020 | ||
2021 | if (!(file->f_mode & FMODE_WRITE)) | |
2022 | return -EBADF; | |
2023 | ||
fef6925c JK |
2024 | ret = mnt_want_write_file(file); |
2025 | if (ret) | |
2026 | return ret; | |
2027 | ret = __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0); | |
2028 | mnt_drop_write_file(file); | |
2029 | return ret; | |
385820a3 MF |
2030 | } |
2031 | ||
2fe17c10 | 2032 | static long ocfs2_fallocate(struct file *file, int mode, loff_t offset, |
385820a3 MF |
2033 | loff_t len) |
2034 | { | |
496ad9aa | 2035 | struct inode *inode = file_inode(file); |
385820a3 MF |
2036 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
2037 | struct ocfs2_space_resv sr; | |
2038 | int change_size = 1; | |
db47fef2 | 2039 | int cmd = OCFS2_IOC_RESVSP64; |
385820a3 | 2040 | |
64c23e86 CH |
2041 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) |
2042 | return -EOPNOTSUPP; | |
385820a3 MF |
2043 | if (!ocfs2_writes_unwritten_extents(osb)) |
2044 | return -EOPNOTSUPP; | |
2045 | ||
385820a3 MF |
2046 | if (mode & FALLOC_FL_KEEP_SIZE) |
2047 | change_size = 0; | |
2048 | ||
db47fef2 JB |
2049 | if (mode & FALLOC_FL_PUNCH_HOLE) |
2050 | cmd = OCFS2_IOC_UNRESVSP64; | |
2051 | ||
385820a3 MF |
2052 | sr.l_whence = 0; |
2053 | sr.l_start = (s64)offset; | |
2054 | sr.l_len = (s64)len; | |
2055 | ||
db47fef2 JB |
2056 | return __ocfs2_change_file_space(NULL, inode, offset, cmd, &sr, |
2057 | change_size); | |
385820a3 MF |
2058 | } |
2059 | ||
293b2f70 TM |
2060 | int ocfs2_check_range_for_refcount(struct inode *inode, loff_t pos, |
2061 | size_t count) | |
2062 | { | |
2063 | int ret = 0; | |
2064 | unsigned int extent_flags; | |
2065 | u32 cpos, clusters, extent_len, phys_cpos; | |
2066 | struct super_block *sb = inode->i_sb; | |
2067 | ||
2068 | if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)) || | |
84e40080 | 2069 | !ocfs2_is_refcount_inode(inode) || |
2f48d593 | 2070 | OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) |
293b2f70 TM |
2071 | return 0; |
2072 | ||
2073 | cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits; | |
2074 | clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos; | |
2075 | ||
2076 | while (clusters) { | |
2077 | ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len, | |
2078 | &extent_flags); | |
2079 | if (ret < 0) { | |
2080 | mlog_errno(ret); | |
2081 | goto out; | |
2082 | } | |
2083 | ||
2084 | if (phys_cpos && (extent_flags & OCFS2_EXT_REFCOUNTED)) { | |
2085 | ret = 1; | |
2086 | break; | |
2087 | } | |
2088 | ||
2089 | if (extent_len > clusters) | |
2090 | extent_len = clusters; | |
2091 | ||
2092 | clusters -= extent_len; | |
2093 | cpos += extent_len; | |
2094 | } | |
2095 | out: | |
2096 | return ret; | |
2097 | } | |
2098 | ||
a11f7e63 MF |
2099 | static int ocfs2_is_io_unaligned(struct inode *inode, size_t count, loff_t pos) |
2100 | { | |
2101 | int blockmask = inode->i_sb->s_blocksize - 1; | |
2102 | loff_t final_size = pos + count; | |
2103 | ||
2104 | if ((pos & blockmask) || (final_size & blockmask)) | |
2105 | return 1; | |
2106 | return 0; | |
2107 | } | |
2108 | ||
293b2f70 | 2109 | static int ocfs2_prepare_inode_for_refcount(struct inode *inode, |
b8908236 | 2110 | struct file *file, |
293b2f70 TM |
2111 | loff_t pos, size_t count, |
2112 | int *meta_level) | |
2113 | { | |
2114 | int ret; | |
2115 | struct buffer_head *di_bh = NULL; | |
2116 | u32 cpos = pos >> OCFS2_SB(inode->i_sb)->s_clustersize_bits; | |
2117 | u32 clusters = | |
2118 | ocfs2_clusters_for_bytes(inode->i_sb, pos + count) - cpos; | |
2119 | ||
2120 | ret = ocfs2_inode_lock(inode, &di_bh, 1); | |
2121 | if (ret) { | |
2122 | mlog_errno(ret); | |
2123 | goto out; | |
2124 | } | |
2125 | ||
2126 | *meta_level = 1; | |
2127 | ||
c7dd3392 | 2128 | ret = ocfs2_refcount_cow(inode, di_bh, cpos, clusters, UINT_MAX); |
293b2f70 TM |
2129 | if (ret) |
2130 | mlog_errno(ret); | |
2131 | out: | |
2132 | brelse(di_bh); | |
2133 | return ret; | |
2134 | } | |
2135 | ||
b8908236 | 2136 | static int ocfs2_prepare_inode_for_write(struct file *file, |
c4c2416a | 2137 | loff_t pos, size_t count, int wait) |
ccd979bd | 2138 | { |
c4c2416a | 2139 | int ret = 0, meta_level = 0, overwrite_io = 0; |
b8908236 | 2140 | struct dentry *dentry = file->f_path.dentry; |
2b0143b5 | 2141 | struct inode *inode = d_inode(dentry); |
c4c2416a | 2142 | struct buffer_head *di_bh = NULL; |
90320251 | 2143 | loff_t end; |
ccd979bd | 2144 | |
2bd63216 | 2145 | /* |
65ed39d6 MF |
2146 | * We start with a read level meta lock and only jump to an ex |
2147 | * if we need to make modifications here. | |
ccd979bd | 2148 | */ |
ccd979bd | 2149 | for(;;) { |
c4c2416a GH |
2150 | if (wait) |
2151 | ret = ocfs2_inode_lock(inode, NULL, meta_level); | |
2152 | else | |
2153 | ret = ocfs2_try_inode_lock(inode, | |
2154 | overwrite_io ? NULL : &di_bh, meta_level); | |
ccd979bd MF |
2155 | if (ret < 0) { |
2156 | meta_level = -1; | |
c4c2416a GH |
2157 | if (ret != -EAGAIN) |
2158 | mlog_errno(ret); | |
ccd979bd MF |
2159 | goto out; |
2160 | } | |
2161 | ||
c4c2416a GH |
2162 | /* |
2163 | * Check if IO will overwrite allocated blocks in case | |
2164 | * IOCB_NOWAIT flag is set. | |
2165 | */ | |
2166 | if (!wait && !overwrite_io) { | |
2167 | overwrite_io = 1; | |
2168 | if (!down_read_trylock(&OCFS2_I(inode)->ip_alloc_sem)) { | |
2169 | ret = -EAGAIN; | |
2170 | goto out_unlock; | |
2171 | } | |
2172 | ||
2173 | ret = ocfs2_overwrite_io(inode, di_bh, pos, count); | |
2174 | brelse(di_bh); | |
2175 | di_bh = NULL; | |
2176 | up_read(&OCFS2_I(inode)->ip_alloc_sem); | |
2177 | if (ret < 0) { | |
2178 | if (ret != -EAGAIN) | |
2179 | mlog_errno(ret); | |
2180 | goto out_unlock; | |
2181 | } | |
2182 | } | |
2183 | ||
ccd979bd MF |
2184 | /* Clear suid / sgid if necessary. We do this here |
2185 | * instead of later in the write path because | |
2186 | * remove_suid() calls ->setattr without any hint that | |
2187 | * we may have already done our cluster locking. Since | |
2188 | * ocfs2_setattr() *must* take cluster locks to | |
42b2aa86 | 2189 | * proceed, this will lead us to recursively lock the |
ccd979bd MF |
2190 | * inode. There's also the dinode i_size state which |
2191 | * can be lost via setattr during extending writes (we | |
2192 | * set inode->i_size at the end of a write. */ | |
8659ac25 | 2193 | if (should_remove_suid(dentry)) { |
ccd979bd | 2194 | if (meta_level == 0) { |
e63aecb6 | 2195 | ocfs2_inode_unlock(inode, meta_level); |
ccd979bd MF |
2196 | meta_level = 1; |
2197 | continue; | |
2198 | } | |
2199 | ||
2200 | ret = ocfs2_write_remove_suid(inode); | |
2201 | if (ret < 0) { | |
2202 | mlog_errno(ret); | |
8659ac25 | 2203 | goto out_unlock; |
ccd979bd MF |
2204 | } |
2205 | } | |
2206 | ||
90320251 | 2207 | end = pos + count; |
9517bac6 | 2208 | |
90320251 | 2209 | ret = ocfs2_check_range_for_refcount(inode, pos, count); |
293b2f70 TM |
2210 | if (ret == 1) { |
2211 | ocfs2_inode_unlock(inode, meta_level); | |
2212 | meta_level = -1; | |
2213 | ||
2214 | ret = ocfs2_prepare_inode_for_refcount(inode, | |
b8908236 | 2215 | file, |
90320251 | 2216 | pos, |
293b2f70 TM |
2217 | count, |
2218 | &meta_level); | |
2219 | } | |
2220 | ||
2221 | if (ret < 0) { | |
2222 | mlog_errno(ret); | |
2223 | goto out_unlock; | |
2224 | } | |
2225 | ||
ccd979bd MF |
2226 | break; |
2227 | } | |
2228 | ||
8659ac25 | 2229 | out_unlock: |
468eedde | 2230 | trace_ocfs2_prepare_inode_for_write(OCFS2_I(inode)->ip_blkno, |
c4c2416a GH |
2231 | pos, count, wait); |
2232 | ||
2233 | brelse(di_bh); | |
468eedde | 2234 | |
293b2f70 TM |
2235 | if (meta_level >= 0) |
2236 | ocfs2_inode_unlock(inode, meta_level); | |
8659ac25 TY |
2237 | |
2238 | out: | |
2239 | return ret; | |
2240 | } | |
2241 | ||
3ef045c3 AV |
2242 | static ssize_t ocfs2_file_write_iter(struct kiocb *iocb, |
2243 | struct iov_iter *from) | |
8659ac25 | 2244 | { |
c4c2416a | 2245 | int rw_level; |
9517bac6 | 2246 | ssize_t written = 0; |
3309dd04 | 2247 | ssize_t ret; |
f1f973ff | 2248 | size_t count = iov_iter_count(from); |
9517bac6 | 2249 | struct file *file = iocb->ki_filp; |
496ad9aa | 2250 | struct inode *inode = file_inode(file); |
9ea2d32f | 2251 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
7bdb0d18 TY |
2252 | int full_coherency = !(osb->s_mount_opt & |
2253 | OCFS2_MOUNT_COHERENCY_BUFFERED); | |
e63890f3 | 2254 | void *saved_ki_complete = NULL; |
faaebf18 JQ |
2255 | int append_write = ((iocb->ki_pos + count) >= |
2256 | i_size_read(inode) ? 1 : 0); | |
c4c2416a GH |
2257 | int direct_io = iocb->ki_flags & IOCB_DIRECT ? 1 : 0; |
2258 | int nowait = iocb->ki_flags & IOCB_NOWAIT ? 1 : 0; | |
9517bac6 | 2259 | |
1202d4ba | 2260 | trace_ocfs2_file_write_iter(inode, file, file->f_path.dentry, |
468eedde TM |
2261 | (unsigned long long)OCFS2_I(inode)->ip_blkno, |
2262 | file->f_path.dentry->d_name.len, | |
2263 | file->f_path.dentry->d_name.name, | |
3ef045c3 | 2264 | (unsigned int)from->nr_segs); /* GRRRRR */ |
8659ac25 | 2265 | |
c4c2416a GH |
2266 | if (!direct_io && nowait) |
2267 | return -EOPNOTSUPP; | |
2268 | ||
66ee59af | 2269 | if (count == 0) |
8659ac25 TY |
2270 | return 0; |
2271 | ||
c4c2416a GH |
2272 | if (nowait) { |
2273 | if (!inode_trylock(inode)) | |
2274 | return -EAGAIN; | |
2275 | } else | |
2276 | inode_lock(inode); | |
9517bac6 | 2277 | |
7bdb0d18 TY |
2278 | /* |
2279 | * Concurrent O_DIRECT writes are allowed with | |
2280 | * mount_option "coherency=buffered". | |
faaebf18 | 2281 | * For append write, we must take rw EX. |
7bdb0d18 | 2282 | */ |
faaebf18 | 2283 | rw_level = (!direct_io || full_coherency || append_write); |
7bdb0d18 | 2284 | |
c4c2416a GH |
2285 | if (nowait) |
2286 | ret = ocfs2_try_rw_lock(inode, rw_level); | |
2287 | else | |
2288 | ret = ocfs2_rw_lock(inode, rw_level); | |
8659ac25 | 2289 | if (ret < 0) { |
c4c2416a GH |
2290 | if (ret != -EAGAIN) |
2291 | mlog_errno(ret); | |
fa5a0eb3 | 2292 | goto out_mutex; |
8659ac25 TY |
2293 | } |
2294 | ||
7bdb0d18 TY |
2295 | /* |
2296 | * O_DIRECT writes with "coherency=full" need to take EX cluster | |
2297 | * inode_lock to guarantee coherency. | |
2298 | */ | |
2299 | if (direct_io && full_coherency) { | |
2300 | /* | |
2301 | * We need to take and drop the inode lock to force | |
2302 | * other nodes to drop their caches. Buffered I/O | |
2303 | * already does this in write_begin(). | |
2304 | */ | |
c4c2416a GH |
2305 | if (nowait) |
2306 | ret = ocfs2_try_inode_lock(inode, NULL, 1); | |
2307 | else | |
2308 | ret = ocfs2_inode_lock(inode, NULL, 1); | |
7bdb0d18 | 2309 | if (ret < 0) { |
c4c2416a GH |
2310 | if (ret != -EAGAIN) |
2311 | mlog_errno(ret); | |
afe1bb73 | 2312 | goto out; |
7bdb0d18 TY |
2313 | } |
2314 | ||
2315 | ocfs2_inode_unlock(inode, 1); | |
2316 | } | |
2317 | ||
3309dd04 AV |
2318 | ret = generic_write_checks(iocb, from); |
2319 | if (ret <= 0) { | |
2320 | if (ret) | |
2321 | mlog_errno(ret); | |
90320251 AV |
2322 | goto out; |
2323 | } | |
3309dd04 | 2324 | count = ret; |
90320251 | 2325 | |
c4c2416a | 2326 | ret = ocfs2_prepare_inode_for_write(file, iocb->ki_pos, count, !nowait); |
8659ac25 | 2327 | if (ret < 0) { |
c4c2416a GH |
2328 | if (ret != -EAGAIN) |
2329 | mlog_errno(ret); | |
8659ac25 TY |
2330 | goto out; |
2331 | } | |
ccd979bd | 2332 | |
e63890f3 RD |
2333 | if (direct_io && !is_sync_kiocb(iocb) && |
2334 | ocfs2_is_io_unaligned(inode, count, iocb->ki_pos)) { | |
a11f7e63 | 2335 | /* |
e63890f3 | 2336 | * Make it a sync io if it's an unaligned aio. |
a11f7e63 | 2337 | */ |
e63890f3 | 2338 | saved_ki_complete = xchg(&iocb->ki_complete, NULL); |
a11f7e63 MF |
2339 | } |
2340 | ||
ccd979bd | 2341 | /* communicate with ocfs2_dio_end_io */ |
7cdfc3a1 | 2342 | ocfs2_iocb_set_rw_locked(iocb, rw_level); |
ccd979bd | 2343 | |
7da839c4 | 2344 | written = __generic_file_write_iter(iocb, from); |
ccd979bd | 2345 | /* buffered aio wouldn't have proper lock coverage today */ |
9e985787 | 2346 | BUG_ON(written == -EIOCBQUEUED && !direct_io); |
ccd979bd | 2347 | |
aa1057b3 RD |
2348 | /* |
2349 | * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io | |
2350 | * function pointer which is called when o_direct io completes so that | |
2351 | * it can unlock our rw lock. | |
2352 | * Unfortunately there are error cases which call end_io and others | |
2353 | * that don't. so we don't have to unlock the rw_lock if either an | |
2354 | * async dio is going to do it in the future or an end_io after an | |
2355 | * error has already done it. | |
2356 | */ | |
2357 | if ((written == -EIOCBQUEUED) || (!ocfs2_iocb_is_rw_locked(iocb))) { | |
2358 | rw_level = -1; | |
aa1057b3 RD |
2359 | } |
2360 | ||
64b4e252 | 2361 | if (unlikely(written <= 0)) |
e63890f3 | 2362 | goto out; |
64b4e252 | 2363 | |
7da839c4 | 2364 | if (((file->f_flags & O_DSYNC) && !direct_io) || |
f1f973ff | 2365 | IS_SYNC(inode)) { |
64b4e252 AV |
2366 | ret = filemap_fdatawrite_range(file->f_mapping, |
2367 | iocb->ki_pos - written, | |
2368 | iocb->ki_pos - 1); | |
918941a3 JK |
2369 | if (ret < 0) |
2370 | written = ret; | |
2371 | ||
86b9c6f3 | 2372 | if (!ret) { |
2b4e30fb | 2373 | ret = jbd2_journal_force_commit(osb->journal->j_journal); |
9ea2d32f MF |
2374 | if (ret < 0) |
2375 | written = ret; | |
2376 | } | |
918941a3 JK |
2377 | |
2378 | if (!ret) | |
64b4e252 AV |
2379 | ret = filemap_fdatawait_range(file->f_mapping, |
2380 | iocb->ki_pos - written, | |
2381 | iocb->ki_pos - 1); | |
9ea2d32f MF |
2382 | } |
2383 | ||
ccd979bd | 2384 | out: |
e63890f3 RD |
2385 | if (saved_ki_complete) |
2386 | xchg(&iocb->ki_complete, saved_ki_complete); | |
2387 | ||
9517bac6 MF |
2388 | if (rw_level != -1) |
2389 | ocfs2_rw_unlock(inode, rw_level); | |
2390 | ||
fa5a0eb3 | 2391 | out_mutex: |
5955102c | 2392 | inode_unlock(inode); |
ccd979bd | 2393 | |
812e7a6a WW |
2394 | if (written) |
2395 | ret = written; | |
812e7a6a | 2396 | return ret; |
ccd979bd MF |
2397 | } |
2398 | ||
3cd9ad5a AV |
2399 | static ssize_t ocfs2_file_read_iter(struct kiocb *iocb, |
2400 | struct iov_iter *to) | |
ccd979bd | 2401 | { |
fa5a0eb3 | 2402 | int ret = 0, rw_level = -1, lock_level = 0; |
ccd979bd | 2403 | struct file *filp = iocb->ki_filp; |
496ad9aa | 2404 | struct inode *inode = file_inode(filp); |
c4c2416a GH |
2405 | int direct_io = iocb->ki_flags & IOCB_DIRECT ? 1 : 0; |
2406 | int nowait = iocb->ki_flags & IOCB_NOWAIT ? 1 : 0; | |
ccd979bd | 2407 | |
1202d4ba | 2408 | trace_ocfs2_file_read_iter(inode, filp, filp->f_path.dentry, |
468eedde TM |
2409 | (unsigned long long)OCFS2_I(inode)->ip_blkno, |
2410 | filp->f_path.dentry->d_name.len, | |
3cd9ad5a AV |
2411 | filp->f_path.dentry->d_name.name, |
2412 | to->nr_segs); /* GRRRRR */ | |
468eedde | 2413 | |
ccd979bd MF |
2414 | |
2415 | if (!inode) { | |
2416 | ret = -EINVAL; | |
2417 | mlog_errno(ret); | |
2418 | goto bail; | |
2419 | } | |
2420 | ||
c4c2416a GH |
2421 | if (!direct_io && nowait) |
2422 | return -EOPNOTSUPP; | |
2423 | ||
2bd63216 | 2424 | /* |
ccd979bd MF |
2425 | * buffered reads protect themselves in ->readpage(). O_DIRECT reads |
2426 | * need locks to protect pending reads from racing with truncate. | |
2427 | */ | |
c4c2416a GH |
2428 | if (direct_io) { |
2429 | if (nowait) | |
2430 | ret = ocfs2_try_rw_lock(inode, 0); | |
2431 | else | |
2432 | ret = ocfs2_rw_lock(inode, 0); | |
2433 | ||
ccd979bd | 2434 | if (ret < 0) { |
c4c2416a GH |
2435 | if (ret != -EAGAIN) |
2436 | mlog_errno(ret); | |
ccd979bd MF |
2437 | goto bail; |
2438 | } | |
2439 | rw_level = 0; | |
2440 | /* communicate with ocfs2_dio_end_io */ | |
7cdfc3a1 | 2441 | ocfs2_iocb_set_rw_locked(iocb, rw_level); |
ccd979bd MF |
2442 | } |
2443 | ||
c4374f8a MF |
2444 | /* |
2445 | * We're fine letting folks race truncates and extending | |
2446 | * writes with read across the cluster, just like they can | |
2447 | * locally. Hence no rw_lock during read. | |
2bd63216 | 2448 | * |
c4374f8a MF |
2449 | * Take and drop the meta data lock to update inode fields |
2450 | * like i_size. This allows the checks down below | |
1202d4ba | 2451 | * generic_file_read_iter() a chance of actually working. |
c4374f8a | 2452 | */ |
c4c2416a GH |
2453 | ret = ocfs2_inode_lock_atime(inode, filp->f_path.mnt, &lock_level, |
2454 | !nowait); | |
c4374f8a | 2455 | if (ret < 0) { |
c4c2416a GH |
2456 | if (ret != -EAGAIN) |
2457 | mlog_errno(ret); | |
c4374f8a MF |
2458 | goto bail; |
2459 | } | |
e63aecb6 | 2460 | ocfs2_inode_unlock(inode, lock_level); |
c4374f8a | 2461 | |
3cd9ad5a | 2462 | ret = generic_file_read_iter(iocb, to); |
1202d4ba | 2463 | trace_generic_file_read_iter_ret(ret); |
ccd979bd MF |
2464 | |
2465 | /* buffered aio wouldn't have proper lock coverage today */ | |
9e985787 | 2466 | BUG_ON(ret == -EIOCBQUEUED && !direct_io); |
ccd979bd | 2467 | |
3ef045c3 | 2468 | /* see ocfs2_file_write_iter */ |
ccd979bd MF |
2469 | if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) { |
2470 | rw_level = -1; | |
ccd979bd MF |
2471 | } |
2472 | ||
2473 | bail: | |
2bd63216 | 2474 | if (rw_level != -1) |
ccd979bd | 2475 | ocfs2_rw_unlock(inode, rw_level); |
ccd979bd MF |
2476 | |
2477 | return ret; | |
2478 | } | |
2479 | ||
93862d5e | 2480 | /* Refer generic_file_llseek_unlocked() */ |
965c8e59 | 2481 | static loff_t ocfs2_file_llseek(struct file *file, loff_t offset, int whence) |
93862d5e SM |
2482 | { |
2483 | struct inode *inode = file->f_mapping->host; | |
2484 | int ret = 0; | |
2485 | ||
5955102c | 2486 | inode_lock(inode); |
93862d5e | 2487 | |
965c8e59 | 2488 | switch (whence) { |
93862d5e SM |
2489 | case SEEK_SET: |
2490 | break; | |
2491 | case SEEK_END: | |
c8d888d9 J |
2492 | /* SEEK_END requires the OCFS2 inode lock for the file |
2493 | * because it references the file's size. | |
2494 | */ | |
2495 | ret = ocfs2_inode_lock(inode, NULL, 0); | |
2496 | if (ret < 0) { | |
2497 | mlog_errno(ret); | |
2498 | goto out; | |
2499 | } | |
2500 | offset += i_size_read(inode); | |
2501 | ocfs2_inode_unlock(inode, 0); | |
93862d5e SM |
2502 | break; |
2503 | case SEEK_CUR: | |
2504 | if (offset == 0) { | |
2505 | offset = file->f_pos; | |
2506 | goto out; | |
2507 | } | |
2508 | offset += file->f_pos; | |
2509 | break; | |
2510 | case SEEK_DATA: | |
2511 | case SEEK_HOLE: | |
965c8e59 | 2512 | ret = ocfs2_seek_data_hole_offset(file, &offset, whence); |
93862d5e SM |
2513 | if (ret) |
2514 | goto out; | |
2515 | break; | |
2516 | default: | |
2517 | ret = -EINVAL; | |
2518 | goto out; | |
2519 | } | |
2520 | ||
46a1c2c7 | 2521 | offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes); |
93862d5e SM |
2522 | |
2523 | out: | |
5955102c | 2524 | inode_unlock(inode); |
93862d5e SM |
2525 | if (ret) |
2526 | return ret; | |
2527 | return offset; | |
2528 | } | |
2529 | ||
42ec3d4c DW |
2530 | static loff_t ocfs2_remap_file_range(struct file *file_in, loff_t pos_in, |
2531 | struct file *file_out, loff_t pos_out, | |
2532 | loff_t len, unsigned int remap_flags) | |
29ac8e85 | 2533 | { |
65f098e9 DW |
2534 | struct inode *inode_in = file_inode(file_in); |
2535 | struct inode *inode_out = file_inode(file_out); | |
2536 | struct ocfs2_super *osb = OCFS2_SB(inode_in->i_sb); | |
2537 | struct buffer_head *in_bh = NULL, *out_bh = NULL; | |
2538 | bool same_inode = (inode_in == inode_out); | |
2539 | loff_t remapped = 0; | |
2540 | ssize_t ret; | |
2541 | ||
2e5dfc99 DW |
2542 | if (remap_flags & ~(REMAP_FILE_DEDUP | REMAP_FILE_ADVISORY)) |
2543 | return -EINVAL; | |
65f098e9 DW |
2544 | if (!ocfs2_refcount_tree(osb)) |
2545 | return -EOPNOTSUPP; | |
2546 | if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb)) | |
2547 | return -EROFS; | |
29ac8e85 | 2548 | |
65f098e9 DW |
2549 | /* Lock both files against IO */ |
2550 | ret = ocfs2_reflink_inodes_lock(inode_in, &in_bh, inode_out, &out_bh); | |
2551 | if (ret) | |
2552 | return ret; | |
2553 | ||
2554 | /* Check file eligibility and prepare for block sharing. */ | |
2555 | ret = -EINVAL; | |
2556 | if ((OCFS2_I(inode_in)->ip_flags & OCFS2_INODE_SYSTEM_FILE) || | |
2557 | (OCFS2_I(inode_out)->ip_flags & OCFS2_INODE_SYSTEM_FILE)) | |
2558 | goto out_unlock; | |
2559 | ||
2560 | ret = generic_remap_file_range_prep(file_in, pos_in, file_out, pos_out, | |
2561 | &len, remap_flags); | |
2562 | if (ret < 0 || len == 0) | |
2563 | goto out_unlock; | |
2564 | ||
2565 | /* Lock out changes to the allocation maps and remap. */ | |
2566 | down_write(&OCFS2_I(inode_in)->ip_alloc_sem); | |
2567 | if (!same_inode) | |
2568 | down_write_nested(&OCFS2_I(inode_out)->ip_alloc_sem, | |
2569 | SINGLE_DEPTH_NESTING); | |
2570 | ||
2571 | /* Zap any page cache for the destination file's range. */ | |
2572 | truncate_inode_pages_range(&inode_out->i_data, | |
2573 | round_down(pos_out, PAGE_SIZE), | |
2574 | round_up(pos_out + len, PAGE_SIZE) - 1); | |
2575 | ||
2576 | remapped = ocfs2_reflink_remap_blocks(inode_in, in_bh, pos_in, | |
2577 | inode_out, out_bh, pos_out, len); | |
2578 | up_write(&OCFS2_I(inode_in)->ip_alloc_sem); | |
2579 | if (!same_inode) | |
2580 | up_write(&OCFS2_I(inode_out)->ip_alloc_sem); | |
2581 | if (remapped < 0) { | |
2582 | ret = remapped; | |
2583 | mlog_errno(ret); | |
2584 | goto out_unlock; | |
2585 | } | |
2586 | ||
2587 | /* | |
2588 | * Empty the extent map so that we may get the right extent | |
2589 | * record from the disk. | |
2590 | */ | |
2591 | ocfs2_extent_map_trunc(inode_in, 0); | |
2592 | ocfs2_extent_map_trunc(inode_out, 0); | |
2593 | ||
2594 | ret = ocfs2_reflink_update_dest(inode_out, out_bh, pos_out + len); | |
2595 | if (ret) { | |
2596 | mlog_errno(ret); | |
2597 | goto out_unlock; | |
2598 | } | |
2599 | ||
2600 | out_unlock: | |
2601 | ocfs2_reflink_inodes_unlock(inode_in, in_bh, inode_out, out_bh); | |
2602 | return remapped > 0 ? remapped : ret; | |
29ac8e85 DW |
2603 | } |
2604 | ||
92e1d5be | 2605 | const struct inode_operations ocfs2_file_iops = { |
ccd979bd MF |
2606 | .setattr = ocfs2_setattr, |
2607 | .getattr = ocfs2_getattr, | |
d38eb8db | 2608 | .permission = ocfs2_permission, |
cf1d6c76 | 2609 | .listxattr = ocfs2_listxattr, |
00dc417f | 2610 | .fiemap = ocfs2_fiemap, |
4e34e719 | 2611 | .get_acl = ocfs2_iop_get_acl, |
702e5bc6 | 2612 | .set_acl = ocfs2_iop_set_acl, |
ccd979bd MF |
2613 | }; |
2614 | ||
92e1d5be | 2615 | const struct inode_operations ocfs2_special_file_iops = { |
ccd979bd MF |
2616 | .setattr = ocfs2_setattr, |
2617 | .getattr = ocfs2_getattr, | |
d38eb8db | 2618 | .permission = ocfs2_permission, |
4e34e719 | 2619 | .get_acl = ocfs2_iop_get_acl, |
702e5bc6 | 2620 | .set_acl = ocfs2_iop_set_acl, |
ccd979bd MF |
2621 | }; |
2622 | ||
53da4939 MF |
2623 | /* |
2624 | * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with | |
2625 | * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks! | |
2626 | */ | |
4b6f5d20 | 2627 | const struct file_operations ocfs2_fops = { |
93862d5e | 2628 | .llseek = ocfs2_file_llseek, |
ccd979bd MF |
2629 | .mmap = ocfs2_mmap, |
2630 | .fsync = ocfs2_sync_file, | |
2631 | .release = ocfs2_file_release, | |
2632 | .open = ocfs2_file_open, | |
3cd9ad5a | 2633 | .read_iter = ocfs2_file_read_iter, |
3ef045c3 | 2634 | .write_iter = ocfs2_file_write_iter, |
c9ec1488 | 2635 | .unlocked_ioctl = ocfs2_ioctl, |
586d232b MF |
2636 | #ifdef CONFIG_COMPAT |
2637 | .compat_ioctl = ocfs2_compat_ioctl, | |
2638 | #endif | |
53da4939 | 2639 | .lock = ocfs2_lock, |
53fc622b | 2640 | .flock = ocfs2_flock, |
82c156f8 | 2641 | .splice_read = generic_file_splice_read, |
6dc8bc0f | 2642 | .splice_write = iter_file_splice_write, |
2fe17c10 | 2643 | .fallocate = ocfs2_fallocate, |
2e5dfc99 | 2644 | .remap_file_range = ocfs2_remap_file_range, |
ccd979bd MF |
2645 | }; |
2646 | ||
4b6f5d20 | 2647 | const struct file_operations ocfs2_dops = { |
32c3c0e2 | 2648 | .llseek = generic_file_llseek, |
ccd979bd | 2649 | .read = generic_read_dir, |
3704412b | 2650 | .iterate = ocfs2_readdir, |
ccd979bd | 2651 | .fsync = ocfs2_sync_file, |
53fc622b MF |
2652 | .release = ocfs2_dir_release, |
2653 | .open = ocfs2_dir_open, | |
c9ec1488 | 2654 | .unlocked_ioctl = ocfs2_ioctl, |
586d232b MF |
2655 | #ifdef CONFIG_COMPAT |
2656 | .compat_ioctl = ocfs2_compat_ioctl, | |
53da4939 MF |
2657 | #endif |
2658 | .lock = ocfs2_lock, | |
2659 | .flock = ocfs2_flock, | |
2660 | }; | |
2661 | ||
2662 | /* | |
2663 | * POSIX-lockless variants of our file_operations. | |
2664 | * | |
2665 | * These will be used if the underlying cluster stack does not support | |
2666 | * posix file locking, if the user passes the "localflocks" mount | |
2667 | * option, or if we have a local-only fs. | |
2668 | * | |
2669 | * ocfs2_flock is in here because all stacks handle UNIX file locks, | |
2670 | * so we still want it in the case of no stack support for | |
2671 | * plocks. Internally, it will do the right thing when asked to ignore | |
2672 | * the cluster. | |
2673 | */ | |
2674 | const struct file_operations ocfs2_fops_no_plocks = { | |
93862d5e | 2675 | .llseek = ocfs2_file_llseek, |
53da4939 MF |
2676 | .mmap = ocfs2_mmap, |
2677 | .fsync = ocfs2_sync_file, | |
2678 | .release = ocfs2_file_release, | |
2679 | .open = ocfs2_file_open, | |
3cd9ad5a | 2680 | .read_iter = ocfs2_file_read_iter, |
3ef045c3 | 2681 | .write_iter = ocfs2_file_write_iter, |
53da4939 MF |
2682 | .unlocked_ioctl = ocfs2_ioctl, |
2683 | #ifdef CONFIG_COMPAT | |
2684 | .compat_ioctl = ocfs2_compat_ioctl, | |
2685 | #endif | |
2686 | .flock = ocfs2_flock, | |
82c156f8 | 2687 | .splice_read = generic_file_splice_read, |
6dc8bc0f | 2688 | .splice_write = iter_file_splice_write, |
3d1c1829 | 2689 | .fallocate = ocfs2_fallocate, |
2e5dfc99 | 2690 | .remap_file_range = ocfs2_remap_file_range, |
53da4939 MF |
2691 | }; |
2692 | ||
2693 | const struct file_operations ocfs2_dops_no_plocks = { | |
2694 | .llseek = generic_file_llseek, | |
2695 | .read = generic_read_dir, | |
3704412b | 2696 | .iterate = ocfs2_readdir, |
53da4939 MF |
2697 | .fsync = ocfs2_sync_file, |
2698 | .release = ocfs2_dir_release, | |
2699 | .open = ocfs2_dir_open, | |
2700 | .unlocked_ioctl = ocfs2_ioctl, | |
2701 | #ifdef CONFIG_COMPAT | |
2702 | .compat_ioctl = ocfs2_compat_ioctl, | |
586d232b | 2703 | #endif |
53fc622b | 2704 | .flock = ocfs2_flock, |
ccd979bd | 2705 | }; |