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> |
ccd979bd | 40 | |
ccd979bd MF |
41 | #include <cluster/masklog.h> |
42 | ||
43 | #include "ocfs2.h" | |
44 | ||
45 | #include "alloc.h" | |
46 | #include "aops.h" | |
47 | #include "dir.h" | |
48 | #include "dlmglue.h" | |
49 | #include "extent_map.h" | |
50 | #include "file.h" | |
51 | #include "sysfile.h" | |
52 | #include "inode.h" | |
ca4d147e | 53 | #include "ioctl.h" |
ccd979bd | 54 | #include "journal.h" |
53fc622b | 55 | #include "locks.h" |
ccd979bd MF |
56 | #include "mmap.h" |
57 | #include "suballoc.h" | |
58 | #include "super.h" | |
cf1d6c76 | 59 | #include "xattr.h" |
23fc2702 | 60 | #include "acl.h" |
a90714c1 | 61 | #include "quota.h" |
293b2f70 | 62 | #include "refcounttree.h" |
468eedde | 63 | #include "ocfs2_trace.h" |
ccd979bd MF |
64 | |
65 | #include "buffer_head_io.h" | |
66 | ||
53fc622b MF |
67 | static int ocfs2_init_file_private(struct inode *inode, struct file *file) |
68 | { | |
69 | struct ocfs2_file_private *fp; | |
70 | ||
71 | fp = kzalloc(sizeof(struct ocfs2_file_private), GFP_KERNEL); | |
72 | if (!fp) | |
73 | return -ENOMEM; | |
74 | ||
75 | fp->fp_file = file; | |
76 | mutex_init(&fp->fp_mutex); | |
77 | ocfs2_file_lock_res_init(&fp->fp_flock, fp); | |
78 | file->private_data = fp; | |
79 | ||
80 | return 0; | |
81 | } | |
82 | ||
83 | static void ocfs2_free_file_private(struct inode *inode, struct file *file) | |
84 | { | |
85 | struct ocfs2_file_private *fp = file->private_data; | |
86 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
87 | ||
88 | if (fp) { | |
89 | ocfs2_simple_drop_lockres(osb, &fp->fp_flock); | |
90 | ocfs2_lock_res_free(&fp->fp_flock); | |
91 | kfree(fp); | |
92 | file->private_data = NULL; | |
93 | } | |
94 | } | |
95 | ||
ccd979bd MF |
96 | static int ocfs2_file_open(struct inode *inode, struct file *file) |
97 | { | |
98 | int status; | |
99 | int mode = file->f_flags; | |
100 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
101 | ||
468eedde TM |
102 | trace_ocfs2_file_open(inode, file, file->f_path.dentry, |
103 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
104 | file->f_path.dentry->d_name.len, | |
105 | file->f_path.dentry->d_name.name, mode); | |
ccd979bd | 106 | |
907f4554 | 107 | if (file->f_mode & FMODE_WRITE) |
871a2931 | 108 | dquot_initialize(inode); |
907f4554 | 109 | |
ccd979bd MF |
110 | spin_lock(&oi->ip_lock); |
111 | ||
112 | /* Check that the inode hasn't been wiped from disk by another | |
113 | * node. If it hasn't then we're safe as long as we hold the | |
114 | * spin lock until our increment of open count. */ | |
115 | if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) { | |
116 | spin_unlock(&oi->ip_lock); | |
117 | ||
118 | status = -ENOENT; | |
119 | goto leave; | |
120 | } | |
121 | ||
122 | if (mode & O_DIRECT) | |
123 | oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT; | |
124 | ||
125 | oi->ip_open_count++; | |
126 | spin_unlock(&oi->ip_lock); | |
53fc622b MF |
127 | |
128 | status = ocfs2_init_file_private(inode, file); | |
129 | if (status) { | |
130 | /* | |
131 | * We want to set open count back if we're failing the | |
132 | * open. | |
133 | */ | |
134 | spin_lock(&oi->ip_lock); | |
135 | oi->ip_open_count--; | |
136 | spin_unlock(&oi->ip_lock); | |
137 | } | |
138 | ||
ccd979bd | 139 | leave: |
ccd979bd MF |
140 | return status; |
141 | } | |
142 | ||
143 | static int ocfs2_file_release(struct inode *inode, struct file *file) | |
144 | { | |
145 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
146 | ||
ccd979bd MF |
147 | spin_lock(&oi->ip_lock); |
148 | if (!--oi->ip_open_count) | |
149 | oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT; | |
468eedde TM |
150 | |
151 | trace_ocfs2_file_release(inode, file, file->f_path.dentry, | |
152 | oi->ip_blkno, | |
153 | file->f_path.dentry->d_name.len, | |
154 | file->f_path.dentry->d_name.name, | |
155 | oi->ip_open_count); | |
ccd979bd MF |
156 | spin_unlock(&oi->ip_lock); |
157 | ||
53fc622b MF |
158 | ocfs2_free_file_private(inode, file); |
159 | ||
ccd979bd MF |
160 | return 0; |
161 | } | |
162 | ||
53fc622b MF |
163 | static int ocfs2_dir_open(struct inode *inode, struct file *file) |
164 | { | |
165 | return ocfs2_init_file_private(inode, file); | |
166 | } | |
167 | ||
168 | static int ocfs2_dir_release(struct inode *inode, struct file *file) | |
169 | { | |
170 | ocfs2_free_file_private(inode, file); | |
171 | return 0; | |
172 | } | |
173 | ||
02c24a82 JB |
174 | static int ocfs2_sync_file(struct file *file, loff_t start, loff_t end, |
175 | int datasync) | |
ccd979bd MF |
176 | { |
177 | int err = 0; | |
7ea80859 | 178 | struct inode *inode = file->f_mapping->host; |
ccd979bd | 179 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
2931cdcb DW |
180 | struct ocfs2_inode_info *oi = OCFS2_I(inode); |
181 | journal_t *journal = osb->journal->j_journal; | |
182 | int ret; | |
183 | tid_t commit_tid; | |
184 | bool needs_barrier = false; | |
ccd979bd | 185 | |
468eedde TM |
186 | trace_ocfs2_sync_file(inode, file, file->f_path.dentry, |
187 | OCFS2_I(inode)->ip_blkno, | |
188 | file->f_path.dentry->d_name.len, | |
189 | file->f_path.dentry->d_name.name, | |
190 | (unsigned long long)datasync); | |
ccd979bd | 191 | |
a987c7ca YL |
192 | if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb)) |
193 | return -EROFS; | |
194 | ||
02c24a82 JB |
195 | err = filemap_write_and_wait_range(inode->i_mapping, start, end); |
196 | if (err) | |
197 | return err; | |
198 | ||
2931cdcb DW |
199 | commit_tid = datasync ? oi->i_datasync_tid : oi->i_sync_tid; |
200 | if (journal->j_flags & JBD2_BARRIER && | |
201 | !jbd2_trans_will_send_data_barrier(journal, commit_tid)) | |
202 | needs_barrier = true; | |
203 | err = jbd2_complete_transaction(journal, commit_tid); | |
204 | if (needs_barrier) { | |
205 | ret = blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, NULL); | |
206 | if (!err) | |
207 | err = ret; | |
04eda1a1 | 208 | } |
e04cc15f | 209 | |
c1e8d35e TM |
210 | if (err) |
211 | mlog_errno(err); | |
ccd979bd MF |
212 | |
213 | return (err < 0) ? -EIO : 0; | |
214 | } | |
215 | ||
7f1a37e3 TY |
216 | int ocfs2_should_update_atime(struct inode *inode, |
217 | struct vfsmount *vfsmnt) | |
218 | { | |
219 | struct timespec now; | |
220 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
221 | ||
222 | if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb)) | |
223 | return 0; | |
224 | ||
225 | if ((inode->i_flags & S_NOATIME) || | |
226 | ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))) | |
227 | return 0; | |
228 | ||
6c2aad05 MF |
229 | /* |
230 | * We can be called with no vfsmnt structure - NFSD will | |
231 | * sometimes do this. | |
232 | * | |
233 | * Note that our action here is different than touch_atime() - | |
234 | * if we can't tell whether this is a noatime mount, then we | |
235 | * don't know whether to trust the value of s_atime_quantum. | |
236 | */ | |
237 | if (vfsmnt == NULL) | |
238 | return 0; | |
239 | ||
7f1a37e3 TY |
240 | if ((vfsmnt->mnt_flags & MNT_NOATIME) || |
241 | ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))) | |
242 | return 0; | |
243 | ||
7e913c53 MF |
244 | if (vfsmnt->mnt_flags & MNT_RELATIME) { |
245 | if ((timespec_compare(&inode->i_atime, &inode->i_mtime) <= 0) || | |
246 | (timespec_compare(&inode->i_atime, &inode->i_ctime) <= 0)) | |
247 | return 1; | |
248 | ||
249 | return 0; | |
250 | } | |
251 | ||
7f1a37e3 TY |
252 | now = CURRENT_TIME; |
253 | if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum)) | |
254 | return 0; | |
255 | else | |
256 | return 1; | |
257 | } | |
258 | ||
259 | int ocfs2_update_inode_atime(struct inode *inode, | |
260 | struct buffer_head *bh) | |
261 | { | |
262 | int ret; | |
263 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
264 | handle_t *handle; | |
c11e9faf | 265 | struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data; |
7f1a37e3 | 266 | |
7f1a37e3 | 267 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
fa38e92c JK |
268 | if (IS_ERR(handle)) { |
269 | ret = PTR_ERR(handle); | |
7f1a37e3 TY |
270 | mlog_errno(ret); |
271 | goto out; | |
272 | } | |
273 | ||
0cf2f763 | 274 | ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh, |
13723d00 | 275 | OCFS2_JOURNAL_ACCESS_WRITE); |
c11e9faf MF |
276 | if (ret) { |
277 | mlog_errno(ret); | |
278 | goto out_commit; | |
279 | } | |
280 | ||
281 | /* | |
282 | * Don't use ocfs2_mark_inode_dirty() here as we don't always | |
283 | * have i_mutex to guard against concurrent changes to other | |
284 | * inode fields. | |
285 | */ | |
7f1a37e3 | 286 | inode->i_atime = CURRENT_TIME; |
c11e9faf MF |
287 | di->i_atime = cpu_to_le64(inode->i_atime.tv_sec); |
288 | di->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec); | |
ec20cec7 | 289 | ocfs2_journal_dirty(handle, bh); |
7f1a37e3 | 290 | |
c11e9faf | 291 | out_commit: |
7f1a37e3 TY |
292 | ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle); |
293 | out: | |
7f1a37e3 TY |
294 | return ret; |
295 | } | |
296 | ||
6cb129f5 AB |
297 | static int ocfs2_set_inode_size(handle_t *handle, |
298 | struct inode *inode, | |
299 | struct buffer_head *fe_bh, | |
300 | u64 new_i_size) | |
ccd979bd MF |
301 | { |
302 | int status; | |
303 | ||
ccd979bd | 304 | i_size_write(inode, new_i_size); |
8110b073 | 305 | inode->i_blocks = ocfs2_inode_sector_count(inode); |
ccd979bd MF |
306 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; |
307 | ||
308 | status = ocfs2_mark_inode_dirty(handle, inode, fe_bh); | |
309 | if (status < 0) { | |
310 | mlog_errno(status); | |
311 | goto bail; | |
312 | } | |
313 | ||
314 | bail: | |
ccd979bd MF |
315 | return status; |
316 | } | |
317 | ||
9e33d69f JK |
318 | int ocfs2_simple_size_update(struct inode *inode, |
319 | struct buffer_head *di_bh, | |
320 | u64 new_i_size) | |
ccd979bd MF |
321 | { |
322 | int ret; | |
323 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1fabe148 | 324 | handle_t *handle = NULL; |
ccd979bd | 325 | |
65eff9cc | 326 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
fa38e92c JK |
327 | if (IS_ERR(handle)) { |
328 | ret = PTR_ERR(handle); | |
ccd979bd MF |
329 | mlog_errno(ret); |
330 | goto out; | |
331 | } | |
332 | ||
333 | ret = ocfs2_set_inode_size(handle, inode, di_bh, | |
334 | new_i_size); | |
335 | if (ret < 0) | |
336 | mlog_errno(ret); | |
337 | ||
02dc1af4 | 338 | ocfs2_commit_trans(osb, handle); |
ccd979bd MF |
339 | out: |
340 | return ret; | |
341 | } | |
342 | ||
37f8a2bf TM |
343 | static int ocfs2_cow_file_pos(struct inode *inode, |
344 | struct buffer_head *fe_bh, | |
345 | u64 offset) | |
346 | { | |
347 | int status; | |
348 | u32 phys, cpos = offset >> OCFS2_SB(inode->i_sb)->s_clustersize_bits; | |
349 | unsigned int num_clusters = 0; | |
350 | unsigned int ext_flags = 0; | |
351 | ||
352 | /* | |
353 | * If the new offset is aligned to the range of the cluster, there is | |
354 | * no space for ocfs2_zero_range_for_truncate to fill, so no need to | |
355 | * CoW either. | |
356 | */ | |
357 | if ((offset & (OCFS2_SB(inode->i_sb)->s_clustersize - 1)) == 0) | |
358 | return 0; | |
359 | ||
360 | status = ocfs2_get_clusters(inode, cpos, &phys, | |
361 | &num_clusters, &ext_flags); | |
362 | if (status) { | |
363 | mlog_errno(status); | |
364 | goto out; | |
365 | } | |
366 | ||
367 | if (!(ext_flags & OCFS2_EXT_REFCOUNTED)) | |
368 | goto out; | |
369 | ||
c7dd3392 | 370 | return ocfs2_refcount_cow(inode, fe_bh, cpos, 1, cpos+1); |
37f8a2bf TM |
371 | |
372 | out: | |
373 | return status; | |
374 | } | |
375 | ||
ccd979bd MF |
376 | static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb, |
377 | struct inode *inode, | |
378 | struct buffer_head *fe_bh, | |
379 | u64 new_i_size) | |
380 | { | |
381 | int status; | |
1fabe148 | 382 | handle_t *handle; |
60b11392 | 383 | struct ocfs2_dinode *di; |
35edec1d | 384 | u64 cluster_bytes; |
ccd979bd | 385 | |
37f8a2bf TM |
386 | /* |
387 | * We need to CoW the cluster contains the offset if it is reflinked | |
388 | * since we will call ocfs2_zero_range_for_truncate later which will | |
389 | * write "0" from offset to the end of the cluster. | |
390 | */ | |
391 | status = ocfs2_cow_file_pos(inode, fe_bh, new_i_size); | |
392 | if (status) { | |
393 | mlog_errno(status); | |
394 | return status; | |
395 | } | |
396 | ||
ccd979bd MF |
397 | /* TODO: This needs to actually orphan the inode in this |
398 | * transaction. */ | |
399 | ||
65eff9cc | 400 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
ccd979bd MF |
401 | if (IS_ERR(handle)) { |
402 | status = PTR_ERR(handle); | |
403 | mlog_errno(status); | |
404 | goto out; | |
405 | } | |
406 | ||
0cf2f763 | 407 | status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh, |
13723d00 | 408 | OCFS2_JOURNAL_ACCESS_WRITE); |
60b11392 MF |
409 | if (status < 0) { |
410 | mlog_errno(status); | |
411 | goto out_commit; | |
412 | } | |
413 | ||
414 | /* | |
415 | * Do this before setting i_size. | |
416 | */ | |
35edec1d MF |
417 | cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size); |
418 | status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size, | |
419 | cluster_bytes); | |
60b11392 MF |
420 | if (status) { |
421 | mlog_errno(status); | |
422 | goto out_commit; | |
423 | } | |
424 | ||
425 | i_size_write(inode, new_i_size); | |
60b11392 MF |
426 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; |
427 | ||
428 | di = (struct ocfs2_dinode *) fe_bh->b_data; | |
429 | di->i_size = cpu_to_le64(new_i_size); | |
430 | di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec); | |
431 | di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec); | |
432 | ||
ec20cec7 | 433 | ocfs2_journal_dirty(handle, fe_bh); |
ccd979bd | 434 | |
60b11392 | 435 | out_commit: |
02dc1af4 | 436 | ocfs2_commit_trans(osb, handle); |
ccd979bd | 437 | out: |
ccd979bd MF |
438 | return status; |
439 | } | |
440 | ||
441 | static int ocfs2_truncate_file(struct inode *inode, | |
442 | struct buffer_head *di_bh, | |
443 | u64 new_i_size) | |
444 | { | |
445 | int status = 0; | |
446 | struct ocfs2_dinode *fe = NULL; | |
447 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
ccd979bd | 448 | |
b657c95c JB |
449 | /* We trust di_bh because it comes from ocfs2_inode_lock(), which |
450 | * already validated it */ | |
ccd979bd | 451 | fe = (struct ocfs2_dinode *) di_bh->b_data; |
ccd979bd | 452 | |
468eedde TM |
453 | trace_ocfs2_truncate_file((unsigned long long)OCFS2_I(inode)->ip_blkno, |
454 | (unsigned long long)le64_to_cpu(fe->i_size), | |
455 | (unsigned long long)new_i_size); | |
456 | ||
ccd979bd | 457 | mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode), |
b0697053 MF |
458 | "Inode %llu, inode i_size = %lld != di " |
459 | "i_size = %llu, i_flags = 0x%x\n", | |
460 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
ccd979bd | 461 | i_size_read(inode), |
b0697053 MF |
462 | (unsigned long long)le64_to_cpu(fe->i_size), |
463 | le32_to_cpu(fe->i_flags)); | |
ccd979bd MF |
464 | |
465 | if (new_i_size > le64_to_cpu(fe->i_size)) { | |
468eedde TM |
466 | trace_ocfs2_truncate_file_error( |
467 | (unsigned long long)le64_to_cpu(fe->i_size), | |
468 | (unsigned long long)new_i_size); | |
ccd979bd MF |
469 | status = -EINVAL; |
470 | mlog_errno(status); | |
471 | goto bail; | |
472 | } | |
473 | ||
2e89b2e4 MF |
474 | down_write(&OCFS2_I(inode)->ip_alloc_sem); |
475 | ||
4fe370af MF |
476 | ocfs2_resv_discard(&osb->osb_la_resmap, |
477 | &OCFS2_I(inode)->ip_la_data_resv); | |
478 | ||
c934a92d MF |
479 | /* |
480 | * The inode lock forced other nodes to sync and drop their | |
481 | * pages, which (correctly) happens even if we have a truncate | |
482 | * without allocation change - ocfs2 cluster sizes can be much | |
483 | * greater than page size, so we have to truncate them | |
484 | * anyway. | |
485 | */ | |
2e89b2e4 MF |
486 | unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1); |
487 | truncate_inode_pages(inode->i_mapping, new_i_size); | |
488 | ||
1afc32b9 MF |
489 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) { |
490 | status = ocfs2_truncate_inline(inode, di_bh, new_i_size, | |
b1967d0e | 491 | i_size_read(inode), 1); |
1afc32b9 MF |
492 | if (status) |
493 | mlog_errno(status); | |
494 | ||
c934a92d | 495 | goto bail_unlock_sem; |
1afc32b9 MF |
496 | } |
497 | ||
ccd979bd MF |
498 | /* alright, we're going to need to do a full blown alloc size |
499 | * change. Orphan the inode so that recovery can complete the | |
500 | * truncate if necessary. This does the task of marking | |
501 | * i_size. */ | |
502 | status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size); | |
503 | if (status < 0) { | |
504 | mlog_errno(status); | |
c934a92d | 505 | goto bail_unlock_sem; |
ccd979bd MF |
506 | } |
507 | ||
78f94673 | 508 | status = ocfs2_commit_truncate(osb, inode, di_bh); |
ccd979bd MF |
509 | if (status < 0) { |
510 | mlog_errno(status); | |
c934a92d | 511 | goto bail_unlock_sem; |
ccd979bd MF |
512 | } |
513 | ||
514 | /* TODO: orphan dir cleanup here. */ | |
c934a92d | 515 | bail_unlock_sem: |
2e89b2e4 MF |
516 | up_write(&OCFS2_I(inode)->ip_alloc_sem); |
517 | ||
ccd979bd | 518 | bail: |
8b2c0dba TM |
519 | if (!status && OCFS2_I(inode)->ip_clusters == 0) |
520 | status = ocfs2_try_remove_refcount_tree(inode, di_bh); | |
ccd979bd | 521 | |
ccd979bd MF |
522 | return status; |
523 | } | |
524 | ||
525 | /* | |
0eb8d47e | 526 | * extend file allocation only here. |
ccd979bd MF |
527 | * we'll update all the disk stuff, and oip->alloc_size |
528 | * | |
529 | * expect stuff to be locked, a transaction started and enough data / | |
530 | * metadata reservations in the contexts. | |
531 | * | |
532 | * Will return -EAGAIN, and a reason if a restart is needed. | |
533 | * If passed in, *reason will always be set, even in error. | |
534 | */ | |
0eb8d47e TM |
535 | int ocfs2_add_inode_data(struct ocfs2_super *osb, |
536 | struct inode *inode, | |
537 | u32 *logical_offset, | |
538 | u32 clusters_to_add, | |
539 | int mark_unwritten, | |
540 | struct buffer_head *fe_bh, | |
541 | handle_t *handle, | |
542 | struct ocfs2_alloc_context *data_ac, | |
543 | struct ocfs2_alloc_context *meta_ac, | |
544 | enum ocfs2_alloc_restarted *reason_ret) | |
ccd979bd | 545 | { |
f99b9b7c JB |
546 | int ret; |
547 | struct ocfs2_extent_tree et; | |
ccd979bd | 548 | |
5e404e9e | 549 | ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), fe_bh); |
cbee7e1a JB |
550 | ret = ocfs2_add_clusters_in_btree(handle, &et, logical_offset, |
551 | clusters_to_add, mark_unwritten, | |
552 | data_ac, meta_ac, reason_ret); | |
f99b9b7c JB |
553 | |
554 | return ret; | |
ccd979bd MF |
555 | } |
556 | ||
2ae99a60 MF |
557 | static int __ocfs2_extend_allocation(struct inode *inode, u32 logical_start, |
558 | u32 clusters_to_add, int mark_unwritten) | |
ccd979bd MF |
559 | { |
560 | int status = 0; | |
561 | int restart_func = 0; | |
abf8b156 | 562 | int credits; |
2ae99a60 | 563 | u32 prev_clusters; |
ccd979bd MF |
564 | struct buffer_head *bh = NULL; |
565 | struct ocfs2_dinode *fe = NULL; | |
1fabe148 | 566 | handle_t *handle = NULL; |
ccd979bd MF |
567 | struct ocfs2_alloc_context *data_ac = NULL; |
568 | struct ocfs2_alloc_context *meta_ac = NULL; | |
569 | enum ocfs2_alloc_restarted why; | |
570 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
f99b9b7c | 571 | struct ocfs2_extent_tree et; |
a90714c1 | 572 | int did_quota = 0; |
ccd979bd | 573 | |
dcd0538f | 574 | /* |
f0cb0f0b | 575 | * Unwritten extent only exists for file systems which |
dcd0538f MF |
576 | * support holes. |
577 | */ | |
2ae99a60 | 578 | BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb)); |
dcd0538f | 579 | |
b657c95c | 580 | status = ocfs2_read_inode_block(inode, &bh); |
ccd979bd MF |
581 | if (status < 0) { |
582 | mlog_errno(status); | |
583 | goto leave; | |
584 | } | |
ccd979bd | 585 | fe = (struct ocfs2_dinode *) bh->b_data; |
ccd979bd MF |
586 | |
587 | restart_all: | |
588 | BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters); | |
589 | ||
5e404e9e | 590 | ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), bh); |
f99b9b7c JB |
591 | status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0, |
592 | &data_ac, &meta_ac); | |
9517bac6 MF |
593 | if (status) { |
594 | mlog_errno(status); | |
595 | goto leave; | |
596 | } | |
597 | ||
06f9da6e | 598 | credits = ocfs2_calc_extend_credits(osb->sb, &fe->id2.i_list); |
65eff9cc | 599 | handle = ocfs2_start_trans(osb, credits); |
ccd979bd MF |
600 | if (IS_ERR(handle)) { |
601 | status = PTR_ERR(handle); | |
602 | handle = NULL; | |
603 | mlog_errno(status); | |
604 | goto leave; | |
605 | } | |
606 | ||
607 | restarted_transaction: | |
468eedde TM |
608 | trace_ocfs2_extend_allocation( |
609 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
610 | (unsigned long long)i_size_read(inode), | |
611 | le32_to_cpu(fe->i_clusters), clusters_to_add, | |
612 | why, restart_func); | |
613 | ||
5dd4056d CH |
614 | status = dquot_alloc_space_nodirty(inode, |
615 | ocfs2_clusters_to_bytes(osb->sb, clusters_to_add)); | |
616 | if (status) | |
a90714c1 | 617 | goto leave; |
a90714c1 JK |
618 | did_quota = 1; |
619 | ||
ccd979bd MF |
620 | /* reserve a write to the file entry early on - that we if we |
621 | * run out of credits in the allocation path, we can still | |
622 | * update i_size. */ | |
0cf2f763 | 623 | status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh, |
13723d00 | 624 | OCFS2_JOURNAL_ACCESS_WRITE); |
ccd979bd MF |
625 | if (status < 0) { |
626 | mlog_errno(status); | |
627 | goto leave; | |
628 | } | |
629 | ||
630 | prev_clusters = OCFS2_I(inode)->ip_clusters; | |
631 | ||
0eb8d47e TM |
632 | status = ocfs2_add_inode_data(osb, |
633 | inode, | |
634 | &logical_start, | |
635 | clusters_to_add, | |
636 | mark_unwritten, | |
637 | bh, | |
638 | handle, | |
639 | data_ac, | |
640 | meta_ac, | |
641 | &why); | |
ccd979bd MF |
642 | if ((status < 0) && (status != -EAGAIN)) { |
643 | if (status != -ENOSPC) | |
644 | mlog_errno(status); | |
645 | goto leave; | |
646 | } | |
2931cdcb | 647 | ocfs2_update_inode_fsync_trans(handle, inode, 1); |
ec20cec7 | 648 | ocfs2_journal_dirty(handle, bh); |
ccd979bd MF |
649 | |
650 | spin_lock(&OCFS2_I(inode)->ip_lock); | |
651 | clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters); | |
652 | spin_unlock(&OCFS2_I(inode)->ip_lock); | |
a90714c1 | 653 | /* Release unused quota reservation */ |
5dd4056d | 654 | dquot_free_space(inode, |
a90714c1 JK |
655 | ocfs2_clusters_to_bytes(osb->sb, clusters_to_add)); |
656 | did_quota = 0; | |
ccd979bd MF |
657 | |
658 | if (why != RESTART_NONE && clusters_to_add) { | |
659 | if (why == RESTART_META) { | |
ccd979bd | 660 | restart_func = 1; |
79681842 | 661 | status = 0; |
ccd979bd MF |
662 | } else { |
663 | BUG_ON(why != RESTART_TRANS); | |
664 | ||
2b1e55c3 | 665 | status = ocfs2_allocate_extend_trans(handle, 1); |
ccd979bd MF |
666 | if (status < 0) { |
667 | /* handle still has to be committed at | |
668 | * this point. */ | |
669 | status = -ENOMEM; | |
670 | mlog_errno(status); | |
671 | goto leave; | |
672 | } | |
673 | goto restarted_transaction; | |
674 | } | |
675 | } | |
676 | ||
468eedde | 677 | trace_ocfs2_extend_allocation_end(OCFS2_I(inode)->ip_blkno, |
1ca1a111 | 678 | le32_to_cpu(fe->i_clusters), |
468eedde TM |
679 | (unsigned long long)le64_to_cpu(fe->i_size), |
680 | OCFS2_I(inode)->ip_clusters, | |
681 | (unsigned long long)i_size_read(inode)); | |
ccd979bd MF |
682 | |
683 | leave: | |
a90714c1 | 684 | if (status < 0 && did_quota) |
5dd4056d | 685 | dquot_free_space(inode, |
a90714c1 | 686 | ocfs2_clusters_to_bytes(osb->sb, clusters_to_add)); |
ccd979bd | 687 | if (handle) { |
02dc1af4 | 688 | ocfs2_commit_trans(osb, handle); |
ccd979bd MF |
689 | handle = NULL; |
690 | } | |
691 | if (data_ac) { | |
692 | ocfs2_free_alloc_context(data_ac); | |
693 | data_ac = NULL; | |
694 | } | |
695 | if (meta_ac) { | |
696 | ocfs2_free_alloc_context(meta_ac); | |
697 | meta_ac = NULL; | |
698 | } | |
699 | if ((!status) && restart_func) { | |
700 | restart_func = 0; | |
701 | goto restart_all; | |
702 | } | |
a81cb88b MF |
703 | brelse(bh); |
704 | bh = NULL; | |
ccd979bd | 705 | |
ccd979bd MF |
706 | return status; |
707 | } | |
708 | ||
a4bfb4cf JB |
709 | /* |
710 | * While a write will already be ordering the data, a truncate will not. | |
711 | * Thus, we need to explicitly order the zeroed pages. | |
712 | */ | |
c7d2cbc3 JB |
713 | static handle_t *ocfs2_zero_start_ordered_transaction(struct inode *inode, |
714 | struct buffer_head *di_bh) | |
a4bfb4cf JB |
715 | { |
716 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
717 | handle_t *handle = NULL; | |
718 | int ret = 0; | |
719 | ||
720 | if (!ocfs2_should_order_data(inode)) | |
721 | goto out; | |
722 | ||
723 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); | |
724 | if (IS_ERR(handle)) { | |
725 | ret = -ENOMEM; | |
726 | mlog_errno(ret); | |
727 | goto out; | |
728 | } | |
729 | ||
730 | ret = ocfs2_jbd2_file_inode(handle, inode); | |
c7d2cbc3 JB |
731 | if (ret < 0) { |
732 | mlog_errno(ret); | |
733 | goto out; | |
734 | } | |
735 | ||
736 | ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh, | |
737 | OCFS2_JOURNAL_ACCESS_WRITE); | |
738 | if (ret) | |
a4bfb4cf JB |
739 | mlog_errno(ret); |
740 | ||
741 | out: | |
742 | if (ret) { | |
743 | if (!IS_ERR(handle)) | |
744 | ocfs2_commit_trans(osb, handle); | |
745 | handle = ERR_PTR(ret); | |
746 | } | |
747 | return handle; | |
748 | } | |
749 | ||
ccd979bd MF |
750 | /* Some parts of this taken from generic_cont_expand, which turned out |
751 | * to be too fragile to do exactly what we need without us having to | |
4e02ed4b | 752 | * worry about recursive locking in ->write_begin() and ->write_end(). */ |
a4bfb4cf | 753 | static int ocfs2_write_zero_page(struct inode *inode, u64 abs_from, |
c7d2cbc3 | 754 | u64 abs_to, struct buffer_head *di_bh) |
ccd979bd MF |
755 | { |
756 | struct address_space *mapping = inode->i_mapping; | |
757 | struct page *page; | |
a4bfb4cf | 758 | unsigned long index = abs_from >> PAGE_CACHE_SHIFT; |
1fabe148 | 759 | handle_t *handle = NULL; |
5453258d | 760 | int ret = 0; |
a4bfb4cf | 761 | unsigned zero_from, zero_to, block_start, block_end; |
c7d2cbc3 | 762 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; |
ccd979bd | 763 | |
a4bfb4cf JB |
764 | BUG_ON(abs_from >= abs_to); |
765 | BUG_ON(abs_to > (((u64)index + 1) << PAGE_CACHE_SHIFT)); | |
766 | BUG_ON(abs_from & (inode->i_blkbits - 1)); | |
ccd979bd | 767 | |
9b4c0ff3 | 768 | page = find_or_create_page(mapping, index, GFP_NOFS); |
ccd979bd MF |
769 | if (!page) { |
770 | ret = -ENOMEM; | |
771 | mlog_errno(ret); | |
772 | goto out; | |
773 | } | |
774 | ||
a4bfb4cf JB |
775 | /* Get the offsets within the page that we want to zero */ |
776 | zero_from = abs_from & (PAGE_CACHE_SIZE - 1); | |
777 | zero_to = abs_to & (PAGE_CACHE_SIZE - 1); | |
778 | if (!zero_to) | |
779 | zero_to = PAGE_CACHE_SIZE; | |
ccd979bd | 780 | |
468eedde TM |
781 | trace_ocfs2_write_zero_page( |
782 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
783 | (unsigned long long)abs_from, | |
784 | (unsigned long long)abs_to, | |
785 | index, zero_from, zero_to); | |
5693486b | 786 | |
a4bfb4cf JB |
787 | /* We know that zero_from is block aligned */ |
788 | for (block_start = zero_from; block_start < zero_to; | |
789 | block_start = block_end) { | |
790 | block_end = block_start + (1 << inode->i_blkbits); | |
791 | ||
792 | /* | |
ebdec241 CH |
793 | * block_start is block-aligned. Bump it by one to force |
794 | * __block_write_begin and block_commit_write to zero the | |
a4bfb4cf JB |
795 | * whole block. |
796 | */ | |
ebdec241 CH |
797 | ret = __block_write_begin(page, block_start + 1, 0, |
798 | ocfs2_get_block); | |
a4bfb4cf JB |
799 | if (ret < 0) { |
800 | mlog_errno(ret); | |
ccd979bd MF |
801 | goto out_unlock; |
802 | } | |
ccd979bd | 803 | |
a4bfb4cf | 804 | if (!handle) { |
c7d2cbc3 JB |
805 | handle = ocfs2_zero_start_ordered_transaction(inode, |
806 | di_bh); | |
a4bfb4cf JB |
807 | if (IS_ERR(handle)) { |
808 | ret = PTR_ERR(handle); | |
809 | handle = NULL; | |
810 | break; | |
811 | } | |
812 | } | |
813 | ||
814 | /* must not update i_size! */ | |
815 | ret = block_commit_write(page, block_start + 1, | |
816 | block_start + 1); | |
817 | if (ret < 0) | |
818 | mlog_errno(ret); | |
819 | else | |
820 | ret = 0; | |
821 | } | |
ccd979bd | 822 | |
c7d2cbc3 JB |
823 | if (handle) { |
824 | /* | |
825 | * fs-writeback will release the dirty pages without page lock | |
826 | * whose offset are over inode size, the release happens at | |
827 | * block_write_full_page_endio(). | |
828 | */ | |
829 | i_size_write(inode, abs_to); | |
830 | inode->i_blocks = ocfs2_inode_sector_count(inode); | |
831 | di->i_size = cpu_to_le64((u64)i_size_read(inode)); | |
832 | inode->i_mtime = inode->i_ctime = CURRENT_TIME; | |
833 | di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec); | |
834 | di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec); | |
835 | di->i_mtime_nsec = di->i_ctime_nsec; | |
836 | ocfs2_journal_dirty(handle, di_bh); | |
02dc1af4 | 837 | ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle); |
c7d2cbc3 | 838 | } |
a4bfb4cf | 839 | |
ccd979bd MF |
840 | out_unlock: |
841 | unlock_page(page); | |
842 | page_cache_release(page); | |
843 | out: | |
844 | return ret; | |
845 | } | |
846 | ||
5693486b JB |
847 | /* |
848 | * Find the next range to zero. We do this in terms of bytes because | |
849 | * that's what ocfs2_zero_extend() wants, and it is dealing with the | |
850 | * pagecache. We may return multiple extents. | |
851 | * | |
852 | * zero_start and zero_end are ocfs2_zero_extend()s current idea of what | |
853 | * needs to be zeroed. range_start and range_end return the next zeroing | |
854 | * range. A subsequent call should pass the previous range_end as its | |
855 | * zero_start. If range_end is 0, there's nothing to do. | |
856 | * | |
857 | * Unwritten extents are skipped over. Refcounted extents are CoWd. | |
858 | */ | |
859 | static int ocfs2_zero_extend_get_range(struct inode *inode, | |
860 | struct buffer_head *di_bh, | |
861 | u64 zero_start, u64 zero_end, | |
862 | u64 *range_start, u64 *range_end) | |
ccd979bd | 863 | { |
5693486b JB |
864 | int rc = 0, needs_cow = 0; |
865 | u32 p_cpos, zero_clusters = 0; | |
866 | u32 zero_cpos = | |
867 | zero_start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits; | |
868 | u32 last_cpos = ocfs2_clusters_for_bytes(inode->i_sb, zero_end); | |
869 | unsigned int num_clusters = 0; | |
870 | unsigned int ext_flags = 0; | |
ccd979bd | 871 | |
5693486b JB |
872 | while (zero_cpos < last_cpos) { |
873 | rc = ocfs2_get_clusters(inode, zero_cpos, &p_cpos, | |
874 | &num_clusters, &ext_flags); | |
875 | if (rc) { | |
876 | mlog_errno(rc); | |
ccd979bd MF |
877 | goto out; |
878 | } | |
879 | ||
5693486b JB |
880 | if (p_cpos && !(ext_flags & OCFS2_EXT_UNWRITTEN)) { |
881 | zero_clusters = num_clusters; | |
882 | if (ext_flags & OCFS2_EXT_REFCOUNTED) | |
883 | needs_cow = 1; | |
884 | break; | |
885 | } | |
886 | ||
887 | zero_cpos += num_clusters; | |
888 | } | |
889 | if (!zero_clusters) { | |
890 | *range_end = 0; | |
891 | goto out; | |
892 | } | |
893 | ||
894 | while ((zero_cpos + zero_clusters) < last_cpos) { | |
895 | rc = ocfs2_get_clusters(inode, zero_cpos + zero_clusters, | |
896 | &p_cpos, &num_clusters, | |
897 | &ext_flags); | |
898 | if (rc) { | |
899 | mlog_errno(rc); | |
900 | goto out; | |
901 | } | |
902 | ||
903 | if (!p_cpos || (ext_flags & OCFS2_EXT_UNWRITTEN)) | |
904 | break; | |
905 | if (ext_flags & OCFS2_EXT_REFCOUNTED) | |
906 | needs_cow = 1; | |
907 | zero_clusters += num_clusters; | |
908 | } | |
909 | if ((zero_cpos + zero_clusters) > last_cpos) | |
910 | zero_clusters = last_cpos - zero_cpos; | |
911 | ||
912 | if (needs_cow) { | |
c7dd3392 | 913 | rc = ocfs2_refcount_cow(inode, di_bh, zero_cpos, |
15502712 | 914 | zero_clusters, UINT_MAX); |
5693486b JB |
915 | if (rc) { |
916 | mlog_errno(rc); | |
917 | goto out; | |
918 | } | |
919 | } | |
920 | ||
921 | *range_start = ocfs2_clusters_to_bytes(inode->i_sb, zero_cpos); | |
922 | *range_end = ocfs2_clusters_to_bytes(inode->i_sb, | |
923 | zero_cpos + zero_clusters); | |
924 | ||
925 | out: | |
926 | return rc; | |
927 | } | |
928 | ||
929 | /* | |
930 | * Zero one range returned from ocfs2_zero_extend_get_range(). The caller | |
931 | * has made sure that the entire range needs zeroing. | |
932 | */ | |
933 | static int ocfs2_zero_extend_range(struct inode *inode, u64 range_start, | |
c7d2cbc3 | 934 | u64 range_end, struct buffer_head *di_bh) |
5693486b JB |
935 | { |
936 | int rc = 0; | |
937 | u64 next_pos; | |
938 | u64 zero_pos = range_start; | |
939 | ||
468eedde TM |
940 | trace_ocfs2_zero_extend_range( |
941 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
942 | (unsigned long long)range_start, | |
943 | (unsigned long long)range_end); | |
5693486b JB |
944 | BUG_ON(range_start >= range_end); |
945 | ||
946 | while (zero_pos < range_end) { | |
947 | next_pos = (zero_pos & PAGE_CACHE_MASK) + PAGE_CACHE_SIZE; | |
948 | if (next_pos > range_end) | |
949 | next_pos = range_end; | |
c7d2cbc3 | 950 | rc = ocfs2_write_zero_page(inode, zero_pos, next_pos, di_bh); |
5693486b JB |
951 | if (rc < 0) { |
952 | mlog_errno(rc); | |
953 | break; | |
954 | } | |
955 | zero_pos = next_pos; | |
e2057c5a MF |
956 | |
957 | /* | |
958 | * Very large extends have the potential to lock up | |
959 | * the cpu for extended periods of time. | |
960 | */ | |
961 | cond_resched(); | |
ccd979bd MF |
962 | } |
963 | ||
5693486b JB |
964 | return rc; |
965 | } | |
966 | ||
967 | int ocfs2_zero_extend(struct inode *inode, struct buffer_head *di_bh, | |
968 | loff_t zero_to_size) | |
969 | { | |
970 | int ret = 0; | |
971 | u64 zero_start, range_start = 0, range_end = 0; | |
972 | struct super_block *sb = inode->i_sb; | |
973 | ||
974 | zero_start = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode)); | |
468eedde TM |
975 | trace_ocfs2_zero_extend((unsigned long long)OCFS2_I(inode)->ip_blkno, |
976 | (unsigned long long)zero_start, | |
977 | (unsigned long long)i_size_read(inode)); | |
5693486b JB |
978 | while (zero_start < zero_to_size) { |
979 | ret = ocfs2_zero_extend_get_range(inode, di_bh, zero_start, | |
980 | zero_to_size, | |
981 | &range_start, | |
982 | &range_end); | |
983 | if (ret) { | |
984 | mlog_errno(ret); | |
985 | break; | |
986 | } | |
987 | if (!range_end) | |
988 | break; | |
989 | /* Trim the ends */ | |
990 | if (range_start < zero_start) | |
991 | range_start = zero_start; | |
992 | if (range_end > zero_to_size) | |
993 | range_end = zero_to_size; | |
994 | ||
995 | ret = ocfs2_zero_extend_range(inode, range_start, | |
c7d2cbc3 | 996 | range_end, di_bh); |
5693486b JB |
997 | if (ret) { |
998 | mlog_errno(ret); | |
999 | break; | |
1000 | } | |
1001 | zero_start = range_end; | |
1002 | } | |
1003 | ||
ccd979bd MF |
1004 | return ret; |
1005 | } | |
1006 | ||
5693486b JB |
1007 | int ocfs2_extend_no_holes(struct inode *inode, struct buffer_head *di_bh, |
1008 | u64 new_i_size, u64 zero_to) | |
65ed39d6 MF |
1009 | { |
1010 | int ret; | |
1011 | u32 clusters_to_add; | |
1012 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
1013 | ||
5693486b JB |
1014 | /* |
1015 | * Only quota files call this without a bh, and they can't be | |
1016 | * refcounted. | |
1017 | */ | |
1018 | BUG_ON(!di_bh && (oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL)); | |
1019 | BUG_ON(!di_bh && !(oi->ip_flags & OCFS2_INODE_SYSTEM_FILE)); | |
1020 | ||
65ed39d6 MF |
1021 | clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size); |
1022 | if (clusters_to_add < oi->ip_clusters) | |
1023 | clusters_to_add = 0; | |
1024 | else | |
1025 | clusters_to_add -= oi->ip_clusters; | |
1026 | ||
1027 | if (clusters_to_add) { | |
1028 | ret = __ocfs2_extend_allocation(inode, oi->ip_clusters, | |
1029 | clusters_to_add, 0); | |
1030 | if (ret) { | |
1031 | mlog_errno(ret); | |
1032 | goto out; | |
1033 | } | |
1034 | } | |
1035 | ||
1036 | /* | |
1037 | * Call this even if we don't add any clusters to the tree. We | |
1038 | * still need to zero the area between the old i_size and the | |
1039 | * new i_size. | |
1040 | */ | |
5693486b | 1041 | ret = ocfs2_zero_extend(inode, di_bh, zero_to); |
65ed39d6 MF |
1042 | if (ret < 0) |
1043 | mlog_errno(ret); | |
1044 | ||
1045 | out: | |
1046 | return ret; | |
1047 | } | |
1048 | ||
ccd979bd MF |
1049 | static int ocfs2_extend_file(struct inode *inode, |
1050 | struct buffer_head *di_bh, | |
65ed39d6 | 1051 | u64 new_i_size) |
ccd979bd | 1052 | { |
c934a92d | 1053 | int ret = 0; |
1afc32b9 | 1054 | struct ocfs2_inode_info *oi = OCFS2_I(inode); |
ccd979bd | 1055 | |
65ed39d6 | 1056 | BUG_ON(!di_bh); |
53013cba | 1057 | |
ccd979bd MF |
1058 | /* setattr sometimes calls us like this. */ |
1059 | if (new_i_size == 0) | |
1060 | goto out; | |
1061 | ||
1062 | if (i_size_read(inode) == new_i_size) | |
5693486b | 1063 | goto out; |
ccd979bd MF |
1064 | BUG_ON(new_i_size < i_size_read(inode)); |
1065 | ||
0effef77 | 1066 | /* |
65ed39d6 MF |
1067 | * The alloc sem blocks people in read/write from reading our |
1068 | * allocation until we're done changing it. We depend on | |
1069 | * i_mutex to block other extend/truncate calls while we're | |
5693486b JB |
1070 | * here. We even have to hold it for sparse files because there |
1071 | * might be some tail zeroing. | |
0effef77 | 1072 | */ |
1afc32b9 MF |
1073 | down_write(&oi->ip_alloc_sem); |
1074 | ||
1075 | if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) { | |
1076 | /* | |
1077 | * We can optimize small extends by keeping the inodes | |
1078 | * inline data. | |
1079 | */ | |
1080 | if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) { | |
1081 | up_write(&oi->ip_alloc_sem); | |
1082 | goto out_update_size; | |
1083 | } | |
1084 | ||
1085 | ret = ocfs2_convert_inline_data_to_extents(inode, di_bh); | |
1086 | if (ret) { | |
1087 | up_write(&oi->ip_alloc_sem); | |
1afc32b9 | 1088 | mlog_errno(ret); |
c934a92d | 1089 | goto out; |
1afc32b9 MF |
1090 | } |
1091 | } | |
1092 | ||
5693486b JB |
1093 | if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))) |
1094 | ret = ocfs2_zero_extend(inode, di_bh, new_i_size); | |
1095 | else | |
1096 | ret = ocfs2_extend_no_holes(inode, di_bh, new_i_size, | |
1097 | new_i_size); | |
1afc32b9 MF |
1098 | |
1099 | up_write(&oi->ip_alloc_sem); | |
65ed39d6 | 1100 | |
0effef77 MF |
1101 | if (ret < 0) { |
1102 | mlog_errno(ret); | |
c934a92d | 1103 | goto out; |
53013cba MF |
1104 | } |
1105 | ||
3a0782d0 | 1106 | out_update_size: |
65ed39d6 MF |
1107 | ret = ocfs2_simple_size_update(inode, di_bh, new_i_size); |
1108 | if (ret < 0) | |
1109 | mlog_errno(ret); | |
ccd979bd MF |
1110 | |
1111 | out: | |
1112 | return ret; | |
1113 | } | |
1114 | ||
1115 | int ocfs2_setattr(struct dentry *dentry, struct iattr *attr) | |
1116 | { | |
1117 | int status = 0, size_change; | |
1118 | struct inode *inode = dentry->d_inode; | |
1119 | struct super_block *sb = inode->i_sb; | |
1120 | struct ocfs2_super *osb = OCFS2_SB(sb); | |
1121 | struct buffer_head *bh = NULL; | |
1fabe148 | 1122 | handle_t *handle = NULL; |
65bac575 | 1123 | struct dquot *transfer_to[MAXQUOTAS] = { }; |
52a9ee28 | 1124 | int qtype; |
ccd979bd | 1125 | |
468eedde TM |
1126 | trace_ocfs2_setattr(inode, dentry, |
1127 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
1128 | dentry->d_name.len, dentry->d_name.name, | |
1129 | attr->ia_valid, attr->ia_mode, | |
ba613560 EB |
1130 | from_kuid(&init_user_ns, attr->ia_uid), |
1131 | from_kgid(&init_user_ns, attr->ia_gid)); | |
ccd979bd | 1132 | |
bc535809 SM |
1133 | /* ensuring we don't even attempt to truncate a symlink */ |
1134 | if (S_ISLNK(inode->i_mode)) | |
1135 | attr->ia_valid &= ~ATTR_SIZE; | |
1136 | ||
ccd979bd MF |
1137 | #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \ |
1138 | | ATTR_GID | ATTR_UID | ATTR_MODE) | |
468eedde | 1139 | if (!(attr->ia_valid & OCFS2_VALID_ATTRS)) |
ccd979bd | 1140 | return 0; |
ccd979bd MF |
1141 | |
1142 | status = inode_change_ok(inode, attr); | |
1143 | if (status) | |
1144 | return status; | |
1145 | ||
12755627 DM |
1146 | if (is_quota_modification(inode, attr)) |
1147 | dquot_initialize(inode); | |
ccd979bd MF |
1148 | size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE; |
1149 | if (size_change) { | |
1150 | status = ocfs2_rw_lock(inode, 1); | |
1151 | if (status < 0) { | |
1152 | mlog_errno(status); | |
1153 | goto bail; | |
1154 | } | |
1155 | } | |
1156 | ||
e63aecb6 | 1157 | status = ocfs2_inode_lock(inode, &bh, 1); |
ccd979bd MF |
1158 | if (status < 0) { |
1159 | if (status != -ENOENT) | |
1160 | mlog_errno(status); | |
1161 | goto bail_unlock_rw; | |
1162 | } | |
1163 | ||
d62e74be | 1164 | if (size_change) { |
5051f768 WW |
1165 | status = inode_newsize_ok(inode, attr->ia_size); |
1166 | if (status) | |
ce76fd30 | 1167 | goto bail_unlock; |
ce76fd30 | 1168 | |
562c72aa CH |
1169 | inode_dio_wait(inode); |
1170 | ||
d62e74be | 1171 | if (i_size_read(inode) >= attr->ia_size) { |
2b4e30fb JB |
1172 | if (ocfs2_should_order_data(inode)) { |
1173 | status = ocfs2_begin_ordered_truncate(inode, | |
1174 | attr->ia_size); | |
1175 | if (status) | |
1176 | goto bail_unlock; | |
1177 | } | |
ccd979bd | 1178 | status = ocfs2_truncate_file(inode, bh, attr->ia_size); |
2b4e30fb | 1179 | } else |
65ed39d6 | 1180 | status = ocfs2_extend_file(inode, bh, attr->ia_size); |
ccd979bd MF |
1181 | if (status < 0) { |
1182 | if (status != -ENOSPC) | |
1183 | mlog_errno(status); | |
1184 | status = -ENOSPC; | |
1185 | goto bail_unlock; | |
1186 | } | |
1187 | } | |
1188 | ||
488c8ef0 EB |
1189 | if ((attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) || |
1190 | (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) { | |
65bac575 JK |
1191 | /* |
1192 | * Gather pointers to quota structures so that allocation / | |
1193 | * freeing of quota structures happens here and not inside | |
b43fa828 | 1194 | * dquot_transfer() where we have problems with lock ordering |
65bac575 | 1195 | */ |
488c8ef0 | 1196 | if (attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid) |
a90714c1 JK |
1197 | && OCFS2_HAS_RO_COMPAT_FEATURE(sb, |
1198 | OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) { | |
aca645a6 | 1199 | transfer_to[USRQUOTA] = dqget(sb, make_kqid_uid(attr->ia_uid)); |
52a9ee28 | 1200 | if (!transfer_to[USRQUOTA]) { |
65bac575 | 1201 | status = -ESRCH; |
a90714c1 | 1202 | goto bail_unlock; |
65bac575 | 1203 | } |
a90714c1 | 1204 | } |
488c8ef0 | 1205 | if (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid) |
a90714c1 JK |
1206 | && OCFS2_HAS_RO_COMPAT_FEATURE(sb, |
1207 | OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) { | |
aca645a6 | 1208 | transfer_to[GRPQUOTA] = dqget(sb, make_kqid_gid(attr->ia_gid)); |
52a9ee28 | 1209 | if (!transfer_to[GRPQUOTA]) { |
65bac575 | 1210 | status = -ESRCH; |
a90714c1 | 1211 | goto bail_unlock; |
65bac575 | 1212 | } |
a90714c1 | 1213 | } |
65bac575 JK |
1214 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS + |
1215 | 2 * ocfs2_quota_trans_credits(sb)); | |
a90714c1 JK |
1216 | if (IS_ERR(handle)) { |
1217 | status = PTR_ERR(handle); | |
1218 | mlog_errno(status); | |
1219 | goto bail_unlock; | |
1220 | } | |
52a9ee28 | 1221 | status = __dquot_transfer(inode, transfer_to); |
a90714c1 JK |
1222 | if (status < 0) |
1223 | goto bail_commit; | |
1224 | } else { | |
1225 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); | |
1226 | if (IS_ERR(handle)) { | |
1227 | status = PTR_ERR(handle); | |
1228 | mlog_errno(status); | |
1229 | goto bail_unlock; | |
1230 | } | |
ccd979bd MF |
1231 | } |
1232 | ||
1025774c CH |
1233 | setattr_copy(inode, attr); |
1234 | mark_inode_dirty(inode); | |
1235 | ||
ccd979bd MF |
1236 | status = ocfs2_mark_inode_dirty(handle, inode, bh); |
1237 | if (status < 0) | |
1238 | mlog_errno(status); | |
1239 | ||
1240 | bail_commit: | |
02dc1af4 | 1241 | ocfs2_commit_trans(osb, handle); |
ccd979bd | 1242 | bail_unlock: |
e63aecb6 | 1243 | ocfs2_inode_unlock(inode, 1); |
ccd979bd MF |
1244 | bail_unlock_rw: |
1245 | if (size_change) | |
1246 | ocfs2_rw_unlock(inode, 1); | |
1247 | bail: | |
a81cb88b | 1248 | brelse(bh); |
ccd979bd | 1249 | |
65bac575 | 1250 | /* Release quota pointers in case we acquired them */ |
52a9ee28 | 1251 | for (qtype = 0; qtype < MAXQUOTAS; qtype++) |
65bac575 | 1252 | dqput(transfer_to[qtype]); |
65bac575 | 1253 | |
060bc66d | 1254 | if (!status && attr->ia_valid & ATTR_MODE) { |
702e5bc6 | 1255 | status = posix_acl_chmod(inode, inode->i_mode); |
060bc66d TY |
1256 | if (status < 0) |
1257 | mlog_errno(status); | |
1258 | } | |
1259 | ||
ccd979bd MF |
1260 | return status; |
1261 | } | |
1262 | ||
1263 | int ocfs2_getattr(struct vfsmount *mnt, | |
1264 | struct dentry *dentry, | |
1265 | struct kstat *stat) | |
1266 | { | |
1267 | struct inode *inode = dentry->d_inode; | |
1268 | struct super_block *sb = dentry->d_inode->i_sb; | |
1269 | struct ocfs2_super *osb = sb->s_fs_info; | |
1270 | int err; | |
1271 | ||
ccd979bd MF |
1272 | err = ocfs2_inode_revalidate(dentry); |
1273 | if (err) { | |
1274 | if (err != -ENOENT) | |
1275 | mlog_errno(err); | |
1276 | goto bail; | |
1277 | } | |
1278 | ||
1279 | generic_fillattr(inode, stat); | |
1280 | ||
1281 | /* We set the blksize from the cluster size for performance */ | |
1282 | stat->blksize = osb->s_clustersize; | |
1283 | ||
1284 | bail: | |
ccd979bd MF |
1285 | return err; |
1286 | } | |
1287 | ||
10556cb2 | 1288 | int ocfs2_permission(struct inode *inode, int mask) |
d38eb8db TY |
1289 | { |
1290 | int ret; | |
1291 | ||
10556cb2 | 1292 | if (mask & MAY_NOT_BLOCK) |
b74c79e9 NP |
1293 | return -ECHILD; |
1294 | ||
e63aecb6 | 1295 | ret = ocfs2_inode_lock(inode, NULL, 0); |
d38eb8db | 1296 | if (ret) { |
a9f5f707 MF |
1297 | if (ret != -ENOENT) |
1298 | mlog_errno(ret); | |
d38eb8db TY |
1299 | goto out; |
1300 | } | |
1301 | ||
2830ba7f | 1302 | ret = generic_permission(inode, mask); |
d38eb8db | 1303 | |
e63aecb6 | 1304 | ocfs2_inode_unlock(inode, 0); |
d38eb8db | 1305 | out: |
d38eb8db TY |
1306 | return ret; |
1307 | } | |
1308 | ||
b2580103 MF |
1309 | static int __ocfs2_write_remove_suid(struct inode *inode, |
1310 | struct buffer_head *bh) | |
ccd979bd MF |
1311 | { |
1312 | int ret; | |
1fabe148 | 1313 | handle_t *handle; |
ccd979bd MF |
1314 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
1315 | struct ocfs2_dinode *di; | |
1316 | ||
468eedde TM |
1317 | trace_ocfs2_write_remove_suid( |
1318 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
1319 | inode->i_mode); | |
ccd979bd | 1320 | |
65eff9cc | 1321 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
fa38e92c JK |
1322 | if (IS_ERR(handle)) { |
1323 | ret = PTR_ERR(handle); | |
ccd979bd MF |
1324 | mlog_errno(ret); |
1325 | goto out; | |
1326 | } | |
1327 | ||
0cf2f763 | 1328 | ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh, |
13723d00 | 1329 | OCFS2_JOURNAL_ACCESS_WRITE); |
ccd979bd MF |
1330 | if (ret < 0) { |
1331 | mlog_errno(ret); | |
b2580103 | 1332 | goto out_trans; |
ccd979bd MF |
1333 | } |
1334 | ||
1335 | inode->i_mode &= ~S_ISUID; | |
1336 | if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP)) | |
1337 | inode->i_mode &= ~S_ISGID; | |
1338 | ||
1339 | di = (struct ocfs2_dinode *) bh->b_data; | |
1340 | di->i_mode = cpu_to_le16(inode->i_mode); | |
1341 | ||
ec20cec7 | 1342 | ocfs2_journal_dirty(handle, bh); |
b2580103 | 1343 | |
ccd979bd | 1344 | out_trans: |
02dc1af4 | 1345 | ocfs2_commit_trans(osb, handle); |
ccd979bd | 1346 | out: |
ccd979bd MF |
1347 | return ret; |
1348 | } | |
1349 | ||
9517bac6 MF |
1350 | /* |
1351 | * Will look for holes and unwritten extents in the range starting at | |
1352 | * pos for count bytes (inclusive). | |
1353 | */ | |
1354 | static int ocfs2_check_range_for_holes(struct inode *inode, loff_t pos, | |
1355 | size_t count) | |
1356 | { | |
1357 | int ret = 0; | |
49cb8d2d | 1358 | unsigned int extent_flags; |
9517bac6 MF |
1359 | u32 cpos, clusters, extent_len, phys_cpos; |
1360 | struct super_block *sb = inode->i_sb; | |
1361 | ||
1362 | cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits; | |
1363 | clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos; | |
1364 | ||
1365 | while (clusters) { | |
49cb8d2d MF |
1366 | ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len, |
1367 | &extent_flags); | |
9517bac6 MF |
1368 | if (ret < 0) { |
1369 | mlog_errno(ret); | |
1370 | goto out; | |
1371 | } | |
1372 | ||
49cb8d2d | 1373 | if (phys_cpos == 0 || (extent_flags & OCFS2_EXT_UNWRITTEN)) { |
9517bac6 MF |
1374 | ret = 1; |
1375 | break; | |
1376 | } | |
1377 | ||
1378 | if (extent_len > clusters) | |
1379 | extent_len = clusters; | |
1380 | ||
1381 | clusters -= extent_len; | |
1382 | cpos += extent_len; | |
1383 | } | |
1384 | out: | |
1385 | return ret; | |
1386 | } | |
1387 | ||
b2580103 MF |
1388 | static int ocfs2_write_remove_suid(struct inode *inode) |
1389 | { | |
1390 | int ret; | |
1391 | struct buffer_head *bh = NULL; | |
b2580103 | 1392 | |
b657c95c | 1393 | ret = ocfs2_read_inode_block(inode, &bh); |
b2580103 MF |
1394 | if (ret < 0) { |
1395 | mlog_errno(ret); | |
1396 | goto out; | |
1397 | } | |
1398 | ||
1399 | ret = __ocfs2_write_remove_suid(inode, bh); | |
1400 | out: | |
1401 | brelse(bh); | |
1402 | return ret; | |
1403 | } | |
1404 | ||
2ae99a60 MF |
1405 | /* |
1406 | * Allocate enough extents to cover the region starting at byte offset | |
1407 | * start for len bytes. Existing extents are skipped, any extents | |
1408 | * added are marked as "unwritten". | |
1409 | */ | |
1410 | static int ocfs2_allocate_unwritten_extents(struct inode *inode, | |
1411 | u64 start, u64 len) | |
1412 | { | |
1413 | int ret; | |
1414 | u32 cpos, phys_cpos, clusters, alloc_size; | |
1afc32b9 MF |
1415 | u64 end = start + len; |
1416 | struct buffer_head *di_bh = NULL; | |
1417 | ||
1418 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) { | |
b657c95c | 1419 | ret = ocfs2_read_inode_block(inode, &di_bh); |
1afc32b9 MF |
1420 | if (ret) { |
1421 | mlog_errno(ret); | |
1422 | goto out; | |
1423 | } | |
1424 | ||
1425 | /* | |
1426 | * Nothing to do if the requested reservation range | |
1427 | * fits within the inode. | |
1428 | */ | |
1429 | if (ocfs2_size_fits_inline_data(di_bh, end)) | |
1430 | goto out; | |
1431 | ||
1432 | ret = ocfs2_convert_inline_data_to_extents(inode, di_bh); | |
1433 | if (ret) { | |
1434 | mlog_errno(ret); | |
1435 | goto out; | |
1436 | } | |
1437 | } | |
2ae99a60 MF |
1438 | |
1439 | /* | |
1440 | * We consider both start and len to be inclusive. | |
1441 | */ | |
1442 | cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits; | |
1443 | clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len); | |
1444 | clusters -= cpos; | |
1445 | ||
1446 | while (clusters) { | |
1447 | ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, | |
1448 | &alloc_size, NULL); | |
1449 | if (ret) { | |
1450 | mlog_errno(ret); | |
1451 | goto out; | |
1452 | } | |
1453 | ||
1454 | /* | |
1455 | * Hole or existing extent len can be arbitrary, so | |
1456 | * cap it to our own allocation request. | |
1457 | */ | |
1458 | if (alloc_size > clusters) | |
1459 | alloc_size = clusters; | |
1460 | ||
1461 | if (phys_cpos) { | |
1462 | /* | |
1463 | * We already have an allocation at this | |
1464 | * region so we can safely skip it. | |
1465 | */ | |
1466 | goto next; | |
1467 | } | |
1468 | ||
1469 | ret = __ocfs2_extend_allocation(inode, cpos, alloc_size, 1); | |
1470 | if (ret) { | |
1471 | if (ret != -ENOSPC) | |
1472 | mlog_errno(ret); | |
1473 | goto out; | |
1474 | } | |
1475 | ||
1476 | next: | |
1477 | cpos += alloc_size; | |
1478 | clusters -= alloc_size; | |
1479 | } | |
1480 | ||
1481 | ret = 0; | |
1482 | out: | |
1afc32b9 MF |
1483 | |
1484 | brelse(di_bh); | |
2ae99a60 MF |
1485 | return ret; |
1486 | } | |
1487 | ||
063c4561 MF |
1488 | /* |
1489 | * Truncate a byte range, avoiding pages within partial clusters. This | |
1490 | * preserves those pages for the zeroing code to write to. | |
1491 | */ | |
1492 | static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start, | |
1493 | u64 byte_len) | |
1494 | { | |
1495 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1496 | loff_t start, end; | |
1497 | struct address_space *mapping = inode->i_mapping; | |
1498 | ||
1499 | start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start); | |
1500 | end = byte_start + byte_len; | |
1501 | end = end & ~(osb->s_clustersize - 1); | |
1502 | ||
1503 | if (start < end) { | |
1504 | unmap_mapping_range(mapping, start, end - start, 0); | |
1505 | truncate_inode_pages_range(mapping, start, end - 1); | |
1506 | } | |
1507 | } | |
1508 | ||
1509 | static int ocfs2_zero_partial_clusters(struct inode *inode, | |
1510 | u64 start, u64 len) | |
1511 | { | |
1512 | int ret = 0; | |
1513 | u64 tmpend, end = start + len; | |
1514 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1515 | unsigned int csize = osb->s_clustersize; | |
1516 | handle_t *handle; | |
1517 | ||
1518 | /* | |
1519 | * The "start" and "end" values are NOT necessarily part of | |
1520 | * the range whose allocation is being deleted. Rather, this | |
1521 | * is what the user passed in with the request. We must zero | |
1522 | * partial clusters here. There's no need to worry about | |
1523 | * physical allocation - the zeroing code knows to skip holes. | |
1524 | */ | |
468eedde TM |
1525 | trace_ocfs2_zero_partial_clusters( |
1526 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
1527 | (unsigned long long)start, (unsigned long long)end); | |
063c4561 MF |
1528 | |
1529 | /* | |
1530 | * If both edges are on a cluster boundary then there's no | |
1531 | * zeroing required as the region is part of the allocation to | |
1532 | * be truncated. | |
1533 | */ | |
1534 | if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0) | |
1535 | goto out; | |
1536 | ||
1537 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); | |
fa38e92c JK |
1538 | if (IS_ERR(handle)) { |
1539 | ret = PTR_ERR(handle); | |
063c4561 MF |
1540 | mlog_errno(ret); |
1541 | goto out; | |
1542 | } | |
1543 | ||
1544 | /* | |
1545 | * We want to get the byte offset of the end of the 1st cluster. | |
1546 | */ | |
1547 | tmpend = (u64)osb->s_clustersize + (start & ~(osb->s_clustersize - 1)); | |
1548 | if (tmpend > end) | |
1549 | tmpend = end; | |
1550 | ||
468eedde TM |
1551 | trace_ocfs2_zero_partial_clusters_range1((unsigned long long)start, |
1552 | (unsigned long long)tmpend); | |
063c4561 MF |
1553 | |
1554 | ret = ocfs2_zero_range_for_truncate(inode, handle, start, tmpend); | |
1555 | if (ret) | |
1556 | mlog_errno(ret); | |
1557 | ||
1558 | if (tmpend < end) { | |
1559 | /* | |
1560 | * This may make start and end equal, but the zeroing | |
1561 | * code will skip any work in that case so there's no | |
1562 | * need to catch it up here. | |
1563 | */ | |
1564 | start = end & ~(osb->s_clustersize - 1); | |
1565 | ||
468eedde TM |
1566 | trace_ocfs2_zero_partial_clusters_range2( |
1567 | (unsigned long long)start, (unsigned long long)end); | |
063c4561 MF |
1568 | |
1569 | ret = ocfs2_zero_range_for_truncate(inode, handle, start, end); | |
1570 | if (ret) | |
1571 | mlog_errno(ret); | |
1572 | } | |
1573 | ||
1574 | ocfs2_commit_trans(osb, handle); | |
1575 | out: | |
1576 | return ret; | |
1577 | } | |
1578 | ||
c1631d4a TY |
1579 | static int ocfs2_find_rec(struct ocfs2_extent_list *el, u32 pos) |
1580 | { | |
1581 | int i; | |
1582 | struct ocfs2_extent_rec *rec = NULL; | |
1583 | ||
1584 | for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) { | |
1585 | ||
1586 | rec = &el->l_recs[i]; | |
1587 | ||
1588 | if (le32_to_cpu(rec->e_cpos) < pos) | |
1589 | break; | |
1590 | } | |
1591 | ||
1592 | return i; | |
1593 | } | |
1594 | ||
1595 | /* | |
1596 | * Helper to calculate the punching pos and length in one run, we handle the | |
1597 | * following three cases in order: | |
1598 | * | |
1599 | * - remove the entire record | |
1600 | * - remove a partial record | |
1601 | * - no record needs to be removed (hole-punching completed) | |
1602 | */ | |
1603 | static void ocfs2_calc_trunc_pos(struct inode *inode, | |
1604 | struct ocfs2_extent_list *el, | |
1605 | struct ocfs2_extent_rec *rec, | |
1606 | u32 trunc_start, u32 *trunc_cpos, | |
1607 | u32 *trunc_len, u32 *trunc_end, | |
1608 | u64 *blkno, int *done) | |
1609 | { | |
1610 | int ret = 0; | |
1611 | u32 coff, range; | |
1612 | ||
1613 | range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec); | |
1614 | ||
1615 | if (le32_to_cpu(rec->e_cpos) >= trunc_start) { | |
9a790ba1 TY |
1616 | /* |
1617 | * remove an entire extent record. | |
1618 | */ | |
c1631d4a TY |
1619 | *trunc_cpos = le32_to_cpu(rec->e_cpos); |
1620 | /* | |
1621 | * Skip holes if any. | |
1622 | */ | |
1623 | if (range < *trunc_end) | |
1624 | *trunc_end = range; | |
1625 | *trunc_len = *trunc_end - le32_to_cpu(rec->e_cpos); | |
1626 | *blkno = le64_to_cpu(rec->e_blkno); | |
1627 | *trunc_end = le32_to_cpu(rec->e_cpos); | |
1628 | } else if (range > trunc_start) { | |
9a790ba1 TY |
1629 | /* |
1630 | * remove a partial extent record, which means we're | |
1631 | * removing the last extent record. | |
1632 | */ | |
c1631d4a | 1633 | *trunc_cpos = trunc_start; |
9a790ba1 TY |
1634 | /* |
1635 | * skip hole if any. | |
1636 | */ | |
1637 | if (range < *trunc_end) | |
1638 | *trunc_end = range; | |
c1631d4a TY |
1639 | *trunc_len = *trunc_end - trunc_start; |
1640 | coff = trunc_start - le32_to_cpu(rec->e_cpos); | |
1641 | *blkno = le64_to_cpu(rec->e_blkno) + | |
1642 | ocfs2_clusters_to_blocks(inode->i_sb, coff); | |
1643 | *trunc_end = trunc_start; | |
1644 | } else { | |
1645 | /* | |
1646 | * It may have two following possibilities: | |
1647 | * | |
1648 | * - last record has been removed | |
1649 | * - trunc_start was within a hole | |
1650 | * | |
1651 | * both two cases mean the completion of hole punching. | |
1652 | */ | |
1653 | ret = 1; | |
1654 | } | |
1655 | ||
1656 | *done = ret; | |
1657 | } | |
1658 | ||
063c4561 MF |
1659 | static int ocfs2_remove_inode_range(struct inode *inode, |
1660 | struct buffer_head *di_bh, u64 byte_start, | |
1661 | u64 byte_len) | |
1662 | { | |
c1631d4a TY |
1663 | int ret = 0, flags = 0, done = 0, i; |
1664 | u32 trunc_start, trunc_len, trunc_end, trunc_cpos, phys_cpos; | |
1665 | u32 cluster_in_el; | |
063c4561 MF |
1666 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
1667 | struct ocfs2_cached_dealloc_ctxt dealloc; | |
b1967d0e | 1668 | struct address_space *mapping = inode->i_mapping; |
fecc0112 | 1669 | struct ocfs2_extent_tree et; |
c1631d4a TY |
1670 | struct ocfs2_path *path = NULL; |
1671 | struct ocfs2_extent_list *el = NULL; | |
1672 | struct ocfs2_extent_rec *rec = NULL; | |
e8aec068 | 1673 | struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data; |
c1631d4a | 1674 | u64 blkno, refcount_loc = le64_to_cpu(di->i_refcount_loc); |
063c4561 | 1675 | |
5e404e9e | 1676 | ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh); |
063c4561 MF |
1677 | ocfs2_init_dealloc_ctxt(&dealloc); |
1678 | ||
468eedde TM |
1679 | trace_ocfs2_remove_inode_range( |
1680 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
1681 | (unsigned long long)byte_start, | |
1682 | (unsigned long long)byte_len); | |
1683 | ||
063c4561 MF |
1684 | if (byte_len == 0) |
1685 | return 0; | |
1686 | ||
1afc32b9 MF |
1687 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) { |
1688 | ret = ocfs2_truncate_inline(inode, di_bh, byte_start, | |
b1967d0e MF |
1689 | byte_start + byte_len, 0); |
1690 | if (ret) { | |
1afc32b9 | 1691 | mlog_errno(ret); |
b1967d0e MF |
1692 | goto out; |
1693 | } | |
1694 | /* | |
1695 | * There's no need to get fancy with the page cache | |
1696 | * truncate of an inline-data inode. We're talking | |
1697 | * about less than a page here, which will be cached | |
1698 | * in the dinode buffer anyway. | |
1699 | */ | |
1700 | unmap_mapping_range(mapping, 0, 0, 0); | |
1701 | truncate_inode_pages(mapping, 0); | |
1702 | goto out; | |
1afc32b9 MF |
1703 | } |
1704 | ||
e8aec068 TY |
1705 | /* |
1706 | * For reflinks, we may need to CoW 2 clusters which might be | |
1707 | * partially zero'd later, if hole's start and end offset were | |
1708 | * within one cluster(means is not exactly aligned to clustersize). | |
1709 | */ | |
1710 | ||
1711 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) { | |
1712 | ||
1713 | ret = ocfs2_cow_file_pos(inode, di_bh, byte_start); | |
1714 | if (ret) { | |
1715 | mlog_errno(ret); | |
1716 | goto out; | |
1717 | } | |
1718 | ||
1719 | ret = ocfs2_cow_file_pos(inode, di_bh, byte_start + byte_len); | |
1720 | if (ret) { | |
1721 | mlog_errno(ret); | |
1722 | goto out; | |
1723 | } | |
1724 | } | |
1725 | ||
063c4561 | 1726 | trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start); |
c1631d4a TY |
1727 | trunc_end = (byte_start + byte_len) >> osb->s_clustersize_bits; |
1728 | cluster_in_el = trunc_end; | |
063c4561 | 1729 | |
063c4561 MF |
1730 | ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len); |
1731 | if (ret) { | |
1732 | mlog_errno(ret); | |
1733 | goto out; | |
1734 | } | |
1735 | ||
c1631d4a TY |
1736 | path = ocfs2_new_path_from_et(&et); |
1737 | if (!path) { | |
1738 | ret = -ENOMEM; | |
1739 | mlog_errno(ret); | |
1740 | goto out; | |
1741 | } | |
1742 | ||
1743 | while (trunc_end > trunc_start) { | |
1744 | ||
1745 | ret = ocfs2_find_path(INODE_CACHE(inode), path, | |
1746 | cluster_in_el); | |
063c4561 MF |
1747 | if (ret) { |
1748 | mlog_errno(ret); | |
1749 | goto out; | |
1750 | } | |
1751 | ||
c1631d4a | 1752 | el = path_leaf_el(path); |
063c4561 | 1753 | |
c1631d4a TY |
1754 | i = ocfs2_find_rec(el, trunc_end); |
1755 | /* | |
1756 | * Need to go to previous extent block. | |
1757 | */ | |
1758 | if (i < 0) { | |
1759 | if (path->p_tree_depth == 0) | |
1760 | break; | |
063c4561 | 1761 | |
c1631d4a TY |
1762 | ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb, |
1763 | path, | |
1764 | &cluster_in_el); | |
063c4561 MF |
1765 | if (ret) { |
1766 | mlog_errno(ret); | |
1767 | goto out; | |
1768 | } | |
c1631d4a TY |
1769 | |
1770 | /* | |
1771 | * We've reached the leftmost extent block, | |
1772 | * it's safe to leave. | |
1773 | */ | |
1774 | if (cluster_in_el == 0) | |
1775 | break; | |
1776 | ||
1777 | /* | |
1778 | * The 'pos' searched for previous extent block is | |
1779 | * always one cluster less than actual trunc_end. | |
1780 | */ | |
1781 | trunc_end = cluster_in_el + 1; | |
1782 | ||
1783 | ocfs2_reinit_path(path, 1); | |
1784 | ||
1785 | continue; | |
1786 | ||
1787 | } else | |
1788 | rec = &el->l_recs[i]; | |
1789 | ||
1790 | ocfs2_calc_trunc_pos(inode, el, rec, trunc_start, &trunc_cpos, | |
1791 | &trunc_len, &trunc_end, &blkno, &done); | |
1792 | if (done) | |
1793 | break; | |
1794 | ||
1795 | flags = rec->e_flags; | |
1796 | phys_cpos = ocfs2_blocks_to_clusters(inode->i_sb, blkno); | |
1797 | ||
1798 | ret = ocfs2_remove_btree_range(inode, &et, trunc_cpos, | |
1799 | phys_cpos, trunc_len, flags, | |
1800 | &dealloc, refcount_loc); | |
1801 | if (ret < 0) { | |
1802 | mlog_errno(ret); | |
1803 | goto out; | |
063c4561 MF |
1804 | } |
1805 | ||
c1631d4a TY |
1806 | cluster_in_el = trunc_end; |
1807 | ||
1808 | ocfs2_reinit_path(path, 1); | |
063c4561 MF |
1809 | } |
1810 | ||
1811 | ocfs2_truncate_cluster_pages(inode, byte_start, byte_len); | |
1812 | ||
1813 | out: | |
7aebff18 | 1814 | ocfs2_free_path(path); |
063c4561 MF |
1815 | ocfs2_schedule_truncate_log_flush(osb, 1); |
1816 | ocfs2_run_deallocs(osb, &dealloc); | |
1817 | ||
1818 | return ret; | |
1819 | } | |
1820 | ||
b2580103 MF |
1821 | /* |
1822 | * Parts of this function taken from xfs_change_file_space() | |
1823 | */ | |
385820a3 MF |
1824 | static int __ocfs2_change_file_space(struct file *file, struct inode *inode, |
1825 | loff_t f_pos, unsigned int cmd, | |
1826 | struct ocfs2_space_resv *sr, | |
1827 | int change_size) | |
b2580103 MF |
1828 | { |
1829 | int ret; | |
1830 | s64 llen; | |
385820a3 | 1831 | loff_t size; |
b2580103 MF |
1832 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
1833 | struct buffer_head *di_bh = NULL; | |
1834 | handle_t *handle; | |
a00cce35 | 1835 | unsigned long long max_off = inode->i_sb->s_maxbytes; |
b2580103 | 1836 | |
b2580103 MF |
1837 | if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb)) |
1838 | return -EROFS; | |
1839 | ||
1840 | mutex_lock(&inode->i_mutex); | |
1841 | ||
1842 | /* | |
1843 | * This prevents concurrent writes on other nodes | |
1844 | */ | |
1845 | ret = ocfs2_rw_lock(inode, 1); | |
1846 | if (ret) { | |
1847 | mlog_errno(ret); | |
1848 | goto out; | |
1849 | } | |
1850 | ||
e63aecb6 | 1851 | ret = ocfs2_inode_lock(inode, &di_bh, 1); |
b2580103 MF |
1852 | if (ret) { |
1853 | mlog_errno(ret); | |
1854 | goto out_rw_unlock; | |
1855 | } | |
1856 | ||
1857 | if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) { | |
1858 | ret = -EPERM; | |
e63aecb6 | 1859 | goto out_inode_unlock; |
b2580103 MF |
1860 | } |
1861 | ||
1862 | switch (sr->l_whence) { | |
1863 | case 0: /*SEEK_SET*/ | |
1864 | break; | |
1865 | case 1: /*SEEK_CUR*/ | |
385820a3 | 1866 | sr->l_start += f_pos; |
b2580103 MF |
1867 | break; |
1868 | case 2: /*SEEK_END*/ | |
1869 | sr->l_start += i_size_read(inode); | |
1870 | break; | |
1871 | default: | |
1872 | ret = -EINVAL; | |
e63aecb6 | 1873 | goto out_inode_unlock; |
b2580103 MF |
1874 | } |
1875 | sr->l_whence = 0; | |
1876 | ||
1877 | llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len; | |
1878 | ||
1879 | if (sr->l_start < 0 | |
1880 | || sr->l_start > max_off | |
1881 | || (sr->l_start + llen) < 0 | |
1882 | || (sr->l_start + llen) > max_off) { | |
1883 | ret = -EINVAL; | |
e63aecb6 | 1884 | goto out_inode_unlock; |
b2580103 | 1885 | } |
385820a3 | 1886 | size = sr->l_start + sr->l_len; |
b2580103 | 1887 | |
a2a3b398 TS |
1888 | if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64 || |
1889 | cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) { | |
b2580103 MF |
1890 | if (sr->l_len <= 0) { |
1891 | ret = -EINVAL; | |
e63aecb6 | 1892 | goto out_inode_unlock; |
b2580103 MF |
1893 | } |
1894 | } | |
1895 | ||
385820a3 | 1896 | if (file && should_remove_suid(file->f_path.dentry)) { |
b2580103 MF |
1897 | ret = __ocfs2_write_remove_suid(inode, di_bh); |
1898 | if (ret) { | |
1899 | mlog_errno(ret); | |
e63aecb6 | 1900 | goto out_inode_unlock; |
b2580103 MF |
1901 | } |
1902 | } | |
1903 | ||
1904 | down_write(&OCFS2_I(inode)->ip_alloc_sem); | |
1905 | switch (cmd) { | |
1906 | case OCFS2_IOC_RESVSP: | |
1907 | case OCFS2_IOC_RESVSP64: | |
1908 | /* | |
1909 | * This takes unsigned offsets, but the signed ones we | |
1910 | * pass have been checked against overflow above. | |
1911 | */ | |
1912 | ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start, | |
1913 | sr->l_len); | |
1914 | break; | |
1915 | case OCFS2_IOC_UNRESVSP: | |
1916 | case OCFS2_IOC_UNRESVSP64: | |
1917 | ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start, | |
1918 | sr->l_len); | |
1919 | break; | |
1920 | default: | |
1921 | ret = -EINVAL; | |
1922 | } | |
1923 | up_write(&OCFS2_I(inode)->ip_alloc_sem); | |
1924 | if (ret) { | |
1925 | mlog_errno(ret); | |
e63aecb6 | 1926 | goto out_inode_unlock; |
b2580103 MF |
1927 | } |
1928 | ||
1929 | /* | |
1930 | * We update c/mtime for these changes | |
1931 | */ | |
1932 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); | |
1933 | if (IS_ERR(handle)) { | |
1934 | ret = PTR_ERR(handle); | |
1935 | mlog_errno(ret); | |
e63aecb6 | 1936 | goto out_inode_unlock; |
b2580103 MF |
1937 | } |
1938 | ||
385820a3 MF |
1939 | if (change_size && i_size_read(inode) < size) |
1940 | i_size_write(inode, size); | |
1941 | ||
b2580103 MF |
1942 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; |
1943 | ret = ocfs2_mark_inode_dirty(handle, inode, di_bh); | |
1944 | if (ret < 0) | |
1945 | mlog_errno(ret); | |
1946 | ||
a4e08d00 | 1947 | if (file && (file->f_flags & O_SYNC)) |
df295d4a MF |
1948 | handle->h_sync = 1; |
1949 | ||
b2580103 MF |
1950 | ocfs2_commit_trans(osb, handle); |
1951 | ||
e63aecb6 | 1952 | out_inode_unlock: |
b2580103 | 1953 | brelse(di_bh); |
e63aecb6 | 1954 | ocfs2_inode_unlock(inode, 1); |
b2580103 MF |
1955 | out_rw_unlock: |
1956 | ocfs2_rw_unlock(inode, 1); | |
1957 | ||
b2580103 | 1958 | out: |
c259ae52 | 1959 | mutex_unlock(&inode->i_mutex); |
b2580103 MF |
1960 | return ret; |
1961 | } | |
1962 | ||
385820a3 MF |
1963 | int ocfs2_change_file_space(struct file *file, unsigned int cmd, |
1964 | struct ocfs2_space_resv *sr) | |
1965 | { | |
496ad9aa | 1966 | struct inode *inode = file_inode(file); |
c19a28e1 | 1967 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
fef6925c | 1968 | int ret; |
385820a3 MF |
1969 | |
1970 | if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) && | |
1971 | !ocfs2_writes_unwritten_extents(osb)) | |
1972 | return -ENOTTY; | |
1973 | else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) && | |
1974 | !ocfs2_sparse_alloc(osb)) | |
1975 | return -ENOTTY; | |
1976 | ||
1977 | if (!S_ISREG(inode->i_mode)) | |
1978 | return -EINVAL; | |
1979 | ||
1980 | if (!(file->f_mode & FMODE_WRITE)) | |
1981 | return -EBADF; | |
1982 | ||
fef6925c JK |
1983 | ret = mnt_want_write_file(file); |
1984 | if (ret) | |
1985 | return ret; | |
1986 | ret = __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0); | |
1987 | mnt_drop_write_file(file); | |
1988 | return ret; | |
385820a3 MF |
1989 | } |
1990 | ||
2fe17c10 | 1991 | static long ocfs2_fallocate(struct file *file, int mode, loff_t offset, |
385820a3 MF |
1992 | loff_t len) |
1993 | { | |
496ad9aa | 1994 | struct inode *inode = file_inode(file); |
385820a3 MF |
1995 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
1996 | struct ocfs2_space_resv sr; | |
1997 | int change_size = 1; | |
db47fef2 | 1998 | int cmd = OCFS2_IOC_RESVSP64; |
385820a3 | 1999 | |
64c23e86 CH |
2000 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) |
2001 | return -EOPNOTSUPP; | |
385820a3 MF |
2002 | if (!ocfs2_writes_unwritten_extents(osb)) |
2003 | return -EOPNOTSUPP; | |
2004 | ||
385820a3 MF |
2005 | if (mode & FALLOC_FL_KEEP_SIZE) |
2006 | change_size = 0; | |
2007 | ||
db47fef2 JB |
2008 | if (mode & FALLOC_FL_PUNCH_HOLE) |
2009 | cmd = OCFS2_IOC_UNRESVSP64; | |
2010 | ||
385820a3 MF |
2011 | sr.l_whence = 0; |
2012 | sr.l_start = (s64)offset; | |
2013 | sr.l_len = (s64)len; | |
2014 | ||
db47fef2 JB |
2015 | return __ocfs2_change_file_space(NULL, inode, offset, cmd, &sr, |
2016 | change_size); | |
385820a3 MF |
2017 | } |
2018 | ||
293b2f70 TM |
2019 | int ocfs2_check_range_for_refcount(struct inode *inode, loff_t pos, |
2020 | size_t count) | |
2021 | { | |
2022 | int ret = 0; | |
2023 | unsigned int extent_flags; | |
2024 | u32 cpos, clusters, extent_len, phys_cpos; | |
2025 | struct super_block *sb = inode->i_sb; | |
2026 | ||
2027 | if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)) || | |
2f48d593 TM |
2028 | !(OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) || |
2029 | OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) | |
293b2f70 TM |
2030 | return 0; |
2031 | ||
2032 | cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits; | |
2033 | clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos; | |
2034 | ||
2035 | while (clusters) { | |
2036 | ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len, | |
2037 | &extent_flags); | |
2038 | if (ret < 0) { | |
2039 | mlog_errno(ret); | |
2040 | goto out; | |
2041 | } | |
2042 | ||
2043 | if (phys_cpos && (extent_flags & OCFS2_EXT_REFCOUNTED)) { | |
2044 | ret = 1; | |
2045 | break; | |
2046 | } | |
2047 | ||
2048 | if (extent_len > clusters) | |
2049 | extent_len = clusters; | |
2050 | ||
2051 | clusters -= extent_len; | |
2052 | cpos += extent_len; | |
2053 | } | |
2054 | out: | |
2055 | return ret; | |
2056 | } | |
2057 | ||
a11f7e63 MF |
2058 | static int ocfs2_is_io_unaligned(struct inode *inode, size_t count, loff_t pos) |
2059 | { | |
2060 | int blockmask = inode->i_sb->s_blocksize - 1; | |
2061 | loff_t final_size = pos + count; | |
2062 | ||
2063 | if ((pos & blockmask) || (final_size & blockmask)) | |
2064 | return 1; | |
2065 | return 0; | |
2066 | } | |
2067 | ||
293b2f70 | 2068 | static int ocfs2_prepare_inode_for_refcount(struct inode *inode, |
b8908236 | 2069 | struct file *file, |
293b2f70 TM |
2070 | loff_t pos, size_t count, |
2071 | int *meta_level) | |
2072 | { | |
2073 | int ret; | |
2074 | struct buffer_head *di_bh = NULL; | |
2075 | u32 cpos = pos >> OCFS2_SB(inode->i_sb)->s_clustersize_bits; | |
2076 | u32 clusters = | |
2077 | ocfs2_clusters_for_bytes(inode->i_sb, pos + count) - cpos; | |
2078 | ||
2079 | ret = ocfs2_inode_lock(inode, &di_bh, 1); | |
2080 | if (ret) { | |
2081 | mlog_errno(ret); | |
2082 | goto out; | |
2083 | } | |
2084 | ||
2085 | *meta_level = 1; | |
2086 | ||
c7dd3392 | 2087 | ret = ocfs2_refcount_cow(inode, di_bh, cpos, clusters, UINT_MAX); |
293b2f70 TM |
2088 | if (ret) |
2089 | mlog_errno(ret); | |
2090 | out: | |
2091 | brelse(di_bh); | |
2092 | return ret; | |
2093 | } | |
2094 | ||
b8908236 | 2095 | static int ocfs2_prepare_inode_for_write(struct file *file, |
8659ac25 TY |
2096 | loff_t *ppos, |
2097 | size_t count, | |
9517bac6 | 2098 | int appending, |
86470e98 TM |
2099 | int *direct_io, |
2100 | int *has_refcount) | |
ccd979bd | 2101 | { |
65ed39d6 | 2102 | int ret = 0, meta_level = 0; |
b8908236 | 2103 | struct dentry *dentry = file->f_path.dentry; |
8659ac25 | 2104 | struct inode *inode = dentry->d_inode; |
468eedde | 2105 | loff_t saved_pos = 0, end; |
ccd979bd | 2106 | |
2bd63216 | 2107 | /* |
65ed39d6 MF |
2108 | * We start with a read level meta lock and only jump to an ex |
2109 | * if we need to make modifications here. | |
ccd979bd | 2110 | */ |
ccd979bd | 2111 | for(;;) { |
e63aecb6 | 2112 | ret = ocfs2_inode_lock(inode, NULL, meta_level); |
ccd979bd MF |
2113 | if (ret < 0) { |
2114 | meta_level = -1; | |
2115 | mlog_errno(ret); | |
2116 | goto out; | |
2117 | } | |
2118 | ||
2119 | /* Clear suid / sgid if necessary. We do this here | |
2120 | * instead of later in the write path because | |
2121 | * remove_suid() calls ->setattr without any hint that | |
2122 | * we may have already done our cluster locking. Since | |
2123 | * ocfs2_setattr() *must* take cluster locks to | |
42b2aa86 | 2124 | * proceed, this will lead us to recursively lock the |
ccd979bd MF |
2125 | * inode. There's also the dinode i_size state which |
2126 | * can be lost via setattr during extending writes (we | |
2127 | * set inode->i_size at the end of a write. */ | |
8659ac25 | 2128 | if (should_remove_suid(dentry)) { |
ccd979bd | 2129 | if (meta_level == 0) { |
e63aecb6 | 2130 | ocfs2_inode_unlock(inode, meta_level); |
ccd979bd MF |
2131 | meta_level = 1; |
2132 | continue; | |
2133 | } | |
2134 | ||
2135 | ret = ocfs2_write_remove_suid(inode); | |
2136 | if (ret < 0) { | |
2137 | mlog_errno(ret); | |
8659ac25 | 2138 | goto out_unlock; |
ccd979bd MF |
2139 | } |
2140 | } | |
2141 | ||
2142 | /* work on a copy of ppos until we're sure that we won't have | |
2143 | * to recalculate it due to relocking. */ | |
468eedde | 2144 | if (appending) |
ccd979bd | 2145 | saved_pos = i_size_read(inode); |
468eedde | 2146 | else |
8659ac25 | 2147 | saved_pos = *ppos; |
3a0782d0 | 2148 | |
65ed39d6 | 2149 | end = saved_pos + count; |
9517bac6 | 2150 | |
293b2f70 TM |
2151 | ret = ocfs2_check_range_for_refcount(inode, saved_pos, count); |
2152 | if (ret == 1) { | |
2153 | ocfs2_inode_unlock(inode, meta_level); | |
2154 | meta_level = -1; | |
2155 | ||
2156 | ret = ocfs2_prepare_inode_for_refcount(inode, | |
b8908236 | 2157 | file, |
293b2f70 TM |
2158 | saved_pos, |
2159 | count, | |
2160 | &meta_level); | |
86470e98 TM |
2161 | if (has_refcount) |
2162 | *has_refcount = 1; | |
96a1cc73 WW |
2163 | if (direct_io) |
2164 | *direct_io = 0; | |
293b2f70 TM |
2165 | } |
2166 | ||
2167 | if (ret < 0) { | |
2168 | mlog_errno(ret); | |
2169 | goto out_unlock; | |
2170 | } | |
2171 | ||
65ed39d6 MF |
2172 | /* |
2173 | * Skip the O_DIRECT checks if we don't need | |
2174 | * them. | |
2175 | */ | |
2176 | if (!direct_io || !(*direct_io)) | |
9517bac6 | 2177 | break; |
9517bac6 | 2178 | |
1afc32b9 MF |
2179 | /* |
2180 | * There's no sane way to do direct writes to an inode | |
2181 | * with inline data. | |
2182 | */ | |
2183 | if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) { | |
2184 | *direct_io = 0; | |
2185 | break; | |
2186 | } | |
2187 | ||
3a0782d0 | 2188 | /* |
65ed39d6 MF |
2189 | * Allowing concurrent direct writes means |
2190 | * i_size changes wouldn't be synchronized, so | |
2191 | * one node could wind up truncating another | |
2192 | * nodes writes. | |
3a0782d0 | 2193 | */ |
65ed39d6 MF |
2194 | if (end > i_size_read(inode)) { |
2195 | *direct_io = 0; | |
ccd979bd | 2196 | break; |
ccd979bd MF |
2197 | } |
2198 | ||
65ed39d6 MF |
2199 | /* |
2200 | * We don't fill holes during direct io, so | |
2201 | * check for them here. If any are found, the | |
2202 | * caller will have to retake some cluster | |
2203 | * locks and initiate the io as buffered. | |
2204 | */ | |
2205 | ret = ocfs2_check_range_for_holes(inode, saved_pos, count); | |
2206 | if (ret == 1) { | |
2207 | *direct_io = 0; | |
2208 | ret = 0; | |
2209 | } else if (ret < 0) | |
2210 | mlog_errno(ret); | |
ccd979bd MF |
2211 | break; |
2212 | } | |
2213 | ||
8659ac25 TY |
2214 | if (appending) |
2215 | *ppos = saved_pos; | |
2216 | ||
2217 | out_unlock: | |
468eedde TM |
2218 | trace_ocfs2_prepare_inode_for_write(OCFS2_I(inode)->ip_blkno, |
2219 | saved_pos, appending, count, | |
2220 | direct_io, has_refcount); | |
2221 | ||
293b2f70 TM |
2222 | if (meta_level >= 0) |
2223 | ocfs2_inode_unlock(inode, meta_level); | |
8659ac25 TY |
2224 | |
2225 | out: | |
2226 | return ret; | |
2227 | } | |
2228 | ||
2229 | static ssize_t ocfs2_file_aio_write(struct kiocb *iocb, | |
2230 | const struct iovec *iov, | |
2231 | unsigned long nr_segs, | |
2232 | loff_t pos) | |
2233 | { | |
9517bac6 | 2234 | int ret, direct_io, appending, rw_level, have_alloc_sem = 0; |
86470e98 | 2235 | int can_do_direct, has_refcount = 0; |
9517bac6 MF |
2236 | ssize_t written = 0; |
2237 | size_t ocount; /* original count */ | |
2238 | size_t count; /* after file limit checks */ | |
9ea2d32f MF |
2239 | loff_t old_size, *ppos = &iocb->ki_pos; |
2240 | u32 old_clusters; | |
9517bac6 | 2241 | struct file *file = iocb->ki_filp; |
496ad9aa | 2242 | struct inode *inode = file_inode(file); |
9ea2d32f | 2243 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
7bdb0d18 TY |
2244 | int full_coherency = !(osb->s_mount_opt & |
2245 | OCFS2_MOUNT_COHERENCY_BUFFERED); | |
a11f7e63 | 2246 | int unaligned_dio = 0; |
9517bac6 | 2247 | |
468eedde TM |
2248 | trace_ocfs2_file_aio_write(inode, file, file->f_path.dentry, |
2249 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
2250 | file->f_path.dentry->d_name.len, | |
2251 | file->f_path.dentry->d_name.name, | |
2252 | (unsigned int)nr_segs); | |
8659ac25 | 2253 | |
73a7075e | 2254 | if (iocb->ki_nbytes == 0) |
8659ac25 TY |
2255 | return 0; |
2256 | ||
9517bac6 MF |
2257 | appending = file->f_flags & O_APPEND ? 1 : 0; |
2258 | direct_io = file->f_flags & O_DIRECT ? 1 : 0; | |
2259 | ||
8659ac25 | 2260 | mutex_lock(&inode->i_mutex); |
9517bac6 | 2261 | |
39c99f12 TY |
2262 | ocfs2_iocb_clear_sem_locked(iocb); |
2263 | ||
9517bac6 | 2264 | relock: |
bd5fe6c5 | 2265 | /* to match setattr's i_mutex -> rw_lock ordering */ |
9517bac6 | 2266 | if (direct_io) { |
9517bac6 | 2267 | have_alloc_sem = 1; |
39c99f12 TY |
2268 | /* communicate with ocfs2_dio_end_io */ |
2269 | ocfs2_iocb_set_sem_locked(iocb); | |
8659ac25 TY |
2270 | } |
2271 | ||
7bdb0d18 TY |
2272 | /* |
2273 | * Concurrent O_DIRECT writes are allowed with | |
2274 | * mount_option "coherency=buffered". | |
2275 | */ | |
2276 | rw_level = (!direct_io || full_coherency); | |
2277 | ||
8659ac25 TY |
2278 | ret = ocfs2_rw_lock(inode, rw_level); |
2279 | if (ret < 0) { | |
8659ac25 | 2280 | mlog_errno(ret); |
9517bac6 | 2281 | goto out_sems; |
8659ac25 TY |
2282 | } |
2283 | ||
7bdb0d18 TY |
2284 | /* |
2285 | * O_DIRECT writes with "coherency=full" need to take EX cluster | |
2286 | * inode_lock to guarantee coherency. | |
2287 | */ | |
2288 | if (direct_io && full_coherency) { | |
2289 | /* | |
2290 | * We need to take and drop the inode lock to force | |
2291 | * other nodes to drop their caches. Buffered I/O | |
2292 | * already does this in write_begin(). | |
2293 | */ | |
2294 | ret = ocfs2_inode_lock(inode, NULL, 1); | |
2295 | if (ret < 0) { | |
2296 | mlog_errno(ret); | |
afe1bb73 | 2297 | goto out; |
7bdb0d18 TY |
2298 | } |
2299 | ||
2300 | ocfs2_inode_unlock(inode, 1); | |
2301 | } | |
2302 | ||
9517bac6 | 2303 | can_do_direct = direct_io; |
b8908236 | 2304 | ret = ocfs2_prepare_inode_for_write(file, ppos, |
73a7075e | 2305 | iocb->ki_nbytes, appending, |
86470e98 | 2306 | &can_do_direct, &has_refcount); |
8659ac25 TY |
2307 | if (ret < 0) { |
2308 | mlog_errno(ret); | |
2309 | goto out; | |
2310 | } | |
ccd979bd | 2311 | |
a11f7e63 | 2312 | if (direct_io && !is_sync_kiocb(iocb)) |
73a7075e | 2313 | unaligned_dio = ocfs2_is_io_unaligned(inode, iocb->ki_nbytes, |
a11f7e63 MF |
2314 | *ppos); |
2315 | ||
9517bac6 MF |
2316 | /* |
2317 | * We can't complete the direct I/O as requested, fall back to | |
2318 | * buffered I/O. | |
2319 | */ | |
2320 | if (direct_io && !can_do_direct) { | |
2321 | ocfs2_rw_unlock(inode, rw_level); | |
9517bac6 MF |
2322 | |
2323 | have_alloc_sem = 0; | |
2324 | rw_level = -1; | |
2325 | ||
2326 | direct_io = 0; | |
9517bac6 MF |
2327 | goto relock; |
2328 | } | |
2329 | ||
a11f7e63 MF |
2330 | if (unaligned_dio) { |
2331 | /* | |
2332 | * Wait on previous unaligned aio to complete before | |
2333 | * proceeding. | |
2334 | */ | |
c18ceab0 WW |
2335 | mutex_lock(&OCFS2_I(inode)->ip_unaligned_aio); |
2336 | /* Mark the iocb as needing an unlock in ocfs2_dio_end_io */ | |
a11f7e63 MF |
2337 | ocfs2_iocb_set_unaligned_aio(iocb); |
2338 | } | |
2339 | ||
9ea2d32f MF |
2340 | /* |
2341 | * To later detect whether a journal commit for sync writes is | |
2342 | * necessary, we sample i_size, and cluster count here. | |
2343 | */ | |
2344 | old_size = i_size_read(inode); | |
2345 | old_clusters = OCFS2_I(inode)->ip_clusters; | |
2346 | ||
ccd979bd | 2347 | /* communicate with ocfs2_dio_end_io */ |
7cdfc3a1 | 2348 | ocfs2_iocb_set_rw_locked(iocb, rw_level); |
ccd979bd | 2349 | |
6b933c8e LD |
2350 | ret = generic_segment_checks(iov, &nr_segs, &ocount, |
2351 | VERIFY_READ); | |
2352 | if (ret) | |
2353 | goto out_dio; | |
b6af1bcd | 2354 | |
6b933c8e LD |
2355 | count = ocount; |
2356 | ret = generic_write_checks(file, ppos, &count, | |
2357 | S_ISBLK(inode->i_mode)); | |
2358 | if (ret) | |
2359 | goto out_dio; | |
b6af1bcd | 2360 | |
6b933c8e | 2361 | if (direct_io) { |
9517bac6 MF |
2362 | written = generic_file_direct_write(iocb, iov, &nr_segs, *ppos, |
2363 | ppos, count, ocount); | |
2364 | if (written < 0) { | |
2365 | ret = written; | |
2366 | goto out_dio; | |
2367 | } | |
2368 | } else { | |
6b933c8e LD |
2369 | current->backing_dev_info = file->f_mapping->backing_dev_info; |
2370 | written = generic_file_buffered_write(iocb, iov, nr_segs, *ppos, | |
2371 | ppos, count, 0); | |
2372 | current->backing_dev_info = NULL; | |
9517bac6 | 2373 | } |
ccd979bd | 2374 | |
9517bac6 | 2375 | out_dio: |
ccd979bd | 2376 | /* buffered aio wouldn't have proper lock coverage today */ |
9517bac6 | 2377 | BUG_ON(ret == -EIOCBQUEUED && !(file->f_flags & O_DIRECT)); |
ccd979bd | 2378 | |
60c48674 | 2379 | if (((file->f_flags & O_DSYNC) && !direct_io) || IS_SYNC(inode) || |
81c8c82b | 2380 | ((file->f_flags & O_DIRECT) && !direct_io)) { |
1b56e989 AV |
2381 | ret = filemap_fdatawrite_range(file->f_mapping, *ppos, |
2382 | *ppos + count - 1); | |
918941a3 JK |
2383 | if (ret < 0) |
2384 | written = ret; | |
2385 | ||
a03ab788 CL |
2386 | if (!ret && ((old_size != i_size_read(inode)) || |
2387 | (old_clusters != OCFS2_I(inode)->ip_clusters) || | |
2388 | has_refcount)) { | |
2b4e30fb | 2389 | ret = jbd2_journal_force_commit(osb->journal->j_journal); |
9ea2d32f MF |
2390 | if (ret < 0) |
2391 | written = ret; | |
2392 | } | |
918941a3 JK |
2393 | |
2394 | if (!ret) | |
1b56e989 AV |
2395 | ret = filemap_fdatawait_range(file->f_mapping, *ppos, |
2396 | *ppos + count - 1); | |
9ea2d32f MF |
2397 | } |
2398 | ||
2bd63216 | 2399 | /* |
ccd979bd MF |
2400 | * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io |
2401 | * function pointer which is called when o_direct io completes so that | |
bd5fe6c5 | 2402 | * it can unlock our rw lock. |
ccd979bd MF |
2403 | * Unfortunately there are error cases which call end_io and others |
2404 | * that don't. so we don't have to unlock the rw_lock if either an | |
2405 | * async dio is going to do it in the future or an end_io after an | |
2406 | * error has already done it. | |
2407 | */ | |
66b116c9 | 2408 | if ((ret == -EIOCBQUEUED) || (!ocfs2_iocb_is_rw_locked(iocb))) { |
ccd979bd MF |
2409 | rw_level = -1; |
2410 | have_alloc_sem = 0; | |
a11f7e63 | 2411 | unaligned_dio = 0; |
ccd979bd MF |
2412 | } |
2413 | ||
3e5d3c35 JB |
2414 | if (unaligned_dio) { |
2415 | ocfs2_iocb_clear_unaligned_aio(iocb); | |
c18ceab0 | 2416 | mutex_unlock(&OCFS2_I(inode)->ip_unaligned_aio); |
3e5d3c35 | 2417 | } |
a11f7e63 | 2418 | |
ccd979bd | 2419 | out: |
9517bac6 MF |
2420 | if (rw_level != -1) |
2421 | ocfs2_rw_unlock(inode, rw_level); | |
2422 | ||
2423 | out_sems: | |
df2d6f26 | 2424 | if (have_alloc_sem) |
39c99f12 | 2425 | ocfs2_iocb_clear_sem_locked(iocb); |
9517bac6 | 2426 | |
1b1dcc1b | 2427 | mutex_unlock(&inode->i_mutex); |
ccd979bd | 2428 | |
812e7a6a WW |
2429 | if (written) |
2430 | ret = written; | |
812e7a6a | 2431 | return ret; |
ccd979bd MF |
2432 | } |
2433 | ||
328eaaba MS |
2434 | static int ocfs2_splice_to_file(struct pipe_inode_info *pipe, |
2435 | struct file *out, | |
2436 | struct splice_desc *sd) | |
2437 | { | |
2438 | int ret; | |
2439 | ||
b8908236 | 2440 | ret = ocfs2_prepare_inode_for_write(out, &sd->pos, |
86470e98 | 2441 | sd->total_len, 0, NULL, NULL); |
328eaaba MS |
2442 | if (ret < 0) { |
2443 | mlog_errno(ret); | |
2444 | return ret; | |
2445 | } | |
2446 | ||
2447 | return splice_from_pipe_feed(pipe, sd, pipe_to_file); | |
2448 | } | |
2449 | ||
8659ac25 TY |
2450 | static ssize_t ocfs2_file_splice_write(struct pipe_inode_info *pipe, |
2451 | struct file *out, | |
2452 | loff_t *ppos, | |
2453 | size_t len, | |
2454 | unsigned int flags) | |
2455 | { | |
2456 | int ret; | |
328eaaba MS |
2457 | struct address_space *mapping = out->f_mapping; |
2458 | struct inode *inode = mapping->host; | |
2459 | struct splice_desc sd = { | |
2460 | .total_len = len, | |
2461 | .flags = flags, | |
2462 | .pos = *ppos, | |
2463 | .u.file = out, | |
2464 | }; | |
8659ac25 | 2465 | |
468eedde TM |
2466 | |
2467 | trace_ocfs2_file_splice_write(inode, out, out->f_path.dentry, | |
2468 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
2469 | out->f_path.dentry->d_name.len, | |
2470 | out->f_path.dentry->d_name.name, len); | |
8659ac25 | 2471 | |
72b0d9aa | 2472 | pipe_lock(pipe); |
8659ac25 | 2473 | |
328eaaba MS |
2474 | splice_from_pipe_begin(&sd); |
2475 | do { | |
2476 | ret = splice_from_pipe_next(pipe, &sd); | |
2477 | if (ret <= 0) | |
2478 | break; | |
8659ac25 | 2479 | |
328eaaba MS |
2480 | mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD); |
2481 | ret = ocfs2_rw_lock(inode, 1); | |
2482 | if (ret < 0) | |
2483 | mlog_errno(ret); | |
2484 | else { | |
2485 | ret = ocfs2_splice_to_file(pipe, out, &sd); | |
2486 | ocfs2_rw_unlock(inode, 1); | |
2487 | } | |
2488 | mutex_unlock(&inode->i_mutex); | |
2489 | } while (ret > 0); | |
2490 | splice_from_pipe_end(pipe, &sd); | |
8659ac25 | 2491 | |
72b0d9aa | 2492 | pipe_unlock(pipe); |
8659ac25 | 2493 | |
328eaaba MS |
2494 | if (sd.num_spliced) |
2495 | ret = sd.num_spliced; | |
2496 | ||
2497 | if (ret > 0) { | |
d23c937b | 2498 | int err; |
328eaaba | 2499 | |
d23c937b JK |
2500 | err = generic_write_sync(out, *ppos, ret); |
2501 | if (err) | |
2502 | ret = err; | |
2503 | else | |
2504 | *ppos += ret; | |
328eaaba | 2505 | |
d0e1d66b | 2506 | balance_dirty_pages_ratelimited(mapping); |
328eaaba | 2507 | } |
8659ac25 | 2508 | |
8659ac25 TY |
2509 | return ret; |
2510 | } | |
2511 | ||
2512 | static ssize_t ocfs2_file_splice_read(struct file *in, | |
2513 | loff_t *ppos, | |
2514 | struct pipe_inode_info *pipe, | |
2515 | size_t len, | |
2516 | unsigned int flags) | |
2517 | { | |
1962f39a | 2518 | int ret = 0, lock_level = 0; |
496ad9aa | 2519 | struct inode *inode = file_inode(in); |
8659ac25 | 2520 | |
468eedde TM |
2521 | trace_ocfs2_file_splice_read(inode, in, in->f_path.dentry, |
2522 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
2523 | in->f_path.dentry->d_name.len, | |
2524 | in->f_path.dentry->d_name.name, len); | |
8659ac25 TY |
2525 | |
2526 | /* | |
2527 | * See the comment in ocfs2_file_aio_read() | |
2528 | */ | |
182be684 | 2529 | ret = ocfs2_inode_lock_atime(inode, in->f_path.mnt, &lock_level); |
8659ac25 TY |
2530 | if (ret < 0) { |
2531 | mlog_errno(ret); | |
2532 | goto bail; | |
2533 | } | |
1962f39a | 2534 | ocfs2_inode_unlock(inode, lock_level); |
8659ac25 TY |
2535 | |
2536 | ret = generic_file_splice_read(in, ppos, pipe, len, flags); | |
2537 | ||
2538 | bail: | |
8659ac25 TY |
2539 | return ret; |
2540 | } | |
2541 | ||
ccd979bd | 2542 | static ssize_t ocfs2_file_aio_read(struct kiocb *iocb, |
027445c3 BP |
2543 | const struct iovec *iov, |
2544 | unsigned long nr_segs, | |
ccd979bd MF |
2545 | loff_t pos) |
2546 | { | |
25899dee | 2547 | int ret = 0, rw_level = -1, have_alloc_sem = 0, lock_level = 0; |
ccd979bd | 2548 | struct file *filp = iocb->ki_filp; |
496ad9aa | 2549 | struct inode *inode = file_inode(filp); |
ccd979bd | 2550 | |
468eedde TM |
2551 | trace_ocfs2_file_aio_read(inode, filp, filp->f_path.dentry, |
2552 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
2553 | filp->f_path.dentry->d_name.len, | |
2554 | filp->f_path.dentry->d_name.name, nr_segs); | |
2555 | ||
ccd979bd MF |
2556 | |
2557 | if (!inode) { | |
2558 | ret = -EINVAL; | |
2559 | mlog_errno(ret); | |
2560 | goto bail; | |
2561 | } | |
2562 | ||
39c99f12 TY |
2563 | ocfs2_iocb_clear_sem_locked(iocb); |
2564 | ||
2bd63216 | 2565 | /* |
ccd979bd MF |
2566 | * buffered reads protect themselves in ->readpage(). O_DIRECT reads |
2567 | * need locks to protect pending reads from racing with truncate. | |
2568 | */ | |
2569 | if (filp->f_flags & O_DIRECT) { | |
ccd979bd | 2570 | have_alloc_sem = 1; |
39c99f12 | 2571 | ocfs2_iocb_set_sem_locked(iocb); |
ccd979bd MF |
2572 | |
2573 | ret = ocfs2_rw_lock(inode, 0); | |
2574 | if (ret < 0) { | |
2575 | mlog_errno(ret); | |
2576 | goto bail; | |
2577 | } | |
2578 | rw_level = 0; | |
2579 | /* communicate with ocfs2_dio_end_io */ | |
7cdfc3a1 | 2580 | ocfs2_iocb_set_rw_locked(iocb, rw_level); |
ccd979bd MF |
2581 | } |
2582 | ||
c4374f8a MF |
2583 | /* |
2584 | * We're fine letting folks race truncates and extending | |
2585 | * writes with read across the cluster, just like they can | |
2586 | * locally. Hence no rw_lock during read. | |
2bd63216 | 2587 | * |
c4374f8a MF |
2588 | * Take and drop the meta data lock to update inode fields |
2589 | * like i_size. This allows the checks down below | |
2bd63216 | 2590 | * generic_file_aio_read() a chance of actually working. |
c4374f8a | 2591 | */ |
182be684 | 2592 | ret = ocfs2_inode_lock_atime(inode, filp->f_path.mnt, &lock_level); |
c4374f8a MF |
2593 | if (ret < 0) { |
2594 | mlog_errno(ret); | |
2595 | goto bail; | |
2596 | } | |
e63aecb6 | 2597 | ocfs2_inode_unlock(inode, lock_level); |
c4374f8a | 2598 | |
027445c3 | 2599 | ret = generic_file_aio_read(iocb, iov, nr_segs, iocb->ki_pos); |
468eedde | 2600 | trace_generic_file_aio_read_ret(ret); |
ccd979bd MF |
2601 | |
2602 | /* buffered aio wouldn't have proper lock coverage today */ | |
2603 | BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT)); | |
2604 | ||
2605 | /* see ocfs2_file_aio_write */ | |
2606 | if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) { | |
2607 | rw_level = -1; | |
2608 | have_alloc_sem = 0; | |
2609 | } | |
2610 | ||
2611 | bail: | |
df2d6f26 | 2612 | if (have_alloc_sem) |
39c99f12 | 2613 | ocfs2_iocb_clear_sem_locked(iocb); |
df2d6f26 | 2614 | |
2bd63216 | 2615 | if (rw_level != -1) |
ccd979bd | 2616 | ocfs2_rw_unlock(inode, rw_level); |
ccd979bd MF |
2617 | |
2618 | return ret; | |
2619 | } | |
2620 | ||
93862d5e | 2621 | /* Refer generic_file_llseek_unlocked() */ |
965c8e59 | 2622 | static loff_t ocfs2_file_llseek(struct file *file, loff_t offset, int whence) |
93862d5e SM |
2623 | { |
2624 | struct inode *inode = file->f_mapping->host; | |
2625 | int ret = 0; | |
2626 | ||
2627 | mutex_lock(&inode->i_mutex); | |
2628 | ||
965c8e59 | 2629 | switch (whence) { |
93862d5e SM |
2630 | case SEEK_SET: |
2631 | break; | |
2632 | case SEEK_END: | |
2633 | offset += inode->i_size; | |
2634 | break; | |
2635 | case SEEK_CUR: | |
2636 | if (offset == 0) { | |
2637 | offset = file->f_pos; | |
2638 | goto out; | |
2639 | } | |
2640 | offset += file->f_pos; | |
2641 | break; | |
2642 | case SEEK_DATA: | |
2643 | case SEEK_HOLE: | |
965c8e59 | 2644 | ret = ocfs2_seek_data_hole_offset(file, &offset, whence); |
93862d5e SM |
2645 | if (ret) |
2646 | goto out; | |
2647 | break; | |
2648 | default: | |
2649 | ret = -EINVAL; | |
2650 | goto out; | |
2651 | } | |
2652 | ||
46a1c2c7 | 2653 | offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes); |
93862d5e SM |
2654 | |
2655 | out: | |
2656 | mutex_unlock(&inode->i_mutex); | |
2657 | if (ret) | |
2658 | return ret; | |
2659 | return offset; | |
2660 | } | |
2661 | ||
92e1d5be | 2662 | const struct inode_operations ocfs2_file_iops = { |
ccd979bd MF |
2663 | .setattr = ocfs2_setattr, |
2664 | .getattr = ocfs2_getattr, | |
d38eb8db | 2665 | .permission = ocfs2_permission, |
cf1d6c76 TY |
2666 | .setxattr = generic_setxattr, |
2667 | .getxattr = generic_getxattr, | |
2668 | .listxattr = ocfs2_listxattr, | |
2669 | .removexattr = generic_removexattr, | |
00dc417f | 2670 | .fiemap = ocfs2_fiemap, |
4e34e719 | 2671 | .get_acl = ocfs2_iop_get_acl, |
702e5bc6 | 2672 | .set_acl = ocfs2_iop_set_acl, |
ccd979bd MF |
2673 | }; |
2674 | ||
92e1d5be | 2675 | const struct inode_operations ocfs2_special_file_iops = { |
ccd979bd MF |
2676 | .setattr = ocfs2_setattr, |
2677 | .getattr = ocfs2_getattr, | |
d38eb8db | 2678 | .permission = ocfs2_permission, |
4e34e719 | 2679 | .get_acl = ocfs2_iop_get_acl, |
702e5bc6 | 2680 | .set_acl = ocfs2_iop_set_acl, |
ccd979bd MF |
2681 | }; |
2682 | ||
53da4939 MF |
2683 | /* |
2684 | * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with | |
2685 | * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks! | |
2686 | */ | |
4b6f5d20 | 2687 | const struct file_operations ocfs2_fops = { |
93862d5e | 2688 | .llseek = ocfs2_file_llseek, |
ccd979bd MF |
2689 | .read = do_sync_read, |
2690 | .write = do_sync_write, | |
ccd979bd MF |
2691 | .mmap = ocfs2_mmap, |
2692 | .fsync = ocfs2_sync_file, | |
2693 | .release = ocfs2_file_release, | |
2694 | .open = ocfs2_file_open, | |
2695 | .aio_read = ocfs2_file_aio_read, | |
2696 | .aio_write = ocfs2_file_aio_write, | |
c9ec1488 | 2697 | .unlocked_ioctl = ocfs2_ioctl, |
586d232b MF |
2698 | #ifdef CONFIG_COMPAT |
2699 | .compat_ioctl = ocfs2_compat_ioctl, | |
2700 | #endif | |
53da4939 | 2701 | .lock = ocfs2_lock, |
53fc622b | 2702 | .flock = ocfs2_flock, |
8659ac25 TY |
2703 | .splice_read = ocfs2_file_splice_read, |
2704 | .splice_write = ocfs2_file_splice_write, | |
2fe17c10 | 2705 | .fallocate = ocfs2_fallocate, |
ccd979bd MF |
2706 | }; |
2707 | ||
4b6f5d20 | 2708 | const struct file_operations ocfs2_dops = { |
32c3c0e2 | 2709 | .llseek = generic_file_llseek, |
ccd979bd | 2710 | .read = generic_read_dir, |
3704412b | 2711 | .iterate = ocfs2_readdir, |
ccd979bd | 2712 | .fsync = ocfs2_sync_file, |
53fc622b MF |
2713 | .release = ocfs2_dir_release, |
2714 | .open = ocfs2_dir_open, | |
c9ec1488 | 2715 | .unlocked_ioctl = ocfs2_ioctl, |
586d232b MF |
2716 | #ifdef CONFIG_COMPAT |
2717 | .compat_ioctl = ocfs2_compat_ioctl, | |
53da4939 MF |
2718 | #endif |
2719 | .lock = ocfs2_lock, | |
2720 | .flock = ocfs2_flock, | |
2721 | }; | |
2722 | ||
2723 | /* | |
2724 | * POSIX-lockless variants of our file_operations. | |
2725 | * | |
2726 | * These will be used if the underlying cluster stack does not support | |
2727 | * posix file locking, if the user passes the "localflocks" mount | |
2728 | * option, or if we have a local-only fs. | |
2729 | * | |
2730 | * ocfs2_flock is in here because all stacks handle UNIX file locks, | |
2731 | * so we still want it in the case of no stack support for | |
2732 | * plocks. Internally, it will do the right thing when asked to ignore | |
2733 | * the cluster. | |
2734 | */ | |
2735 | const struct file_operations ocfs2_fops_no_plocks = { | |
93862d5e | 2736 | .llseek = ocfs2_file_llseek, |
53da4939 MF |
2737 | .read = do_sync_read, |
2738 | .write = do_sync_write, | |
2739 | .mmap = ocfs2_mmap, | |
2740 | .fsync = ocfs2_sync_file, | |
2741 | .release = ocfs2_file_release, | |
2742 | .open = ocfs2_file_open, | |
2743 | .aio_read = ocfs2_file_aio_read, | |
2744 | .aio_write = ocfs2_file_aio_write, | |
2745 | .unlocked_ioctl = ocfs2_ioctl, | |
2746 | #ifdef CONFIG_COMPAT | |
2747 | .compat_ioctl = ocfs2_compat_ioctl, | |
2748 | #endif | |
2749 | .flock = ocfs2_flock, | |
2750 | .splice_read = ocfs2_file_splice_read, | |
2751 | .splice_write = ocfs2_file_splice_write, | |
3d1c1829 | 2752 | .fallocate = ocfs2_fallocate, |
53da4939 MF |
2753 | }; |
2754 | ||
2755 | const struct file_operations ocfs2_dops_no_plocks = { | |
2756 | .llseek = generic_file_llseek, | |
2757 | .read = generic_read_dir, | |
3704412b | 2758 | .iterate = ocfs2_readdir, |
53da4939 MF |
2759 | .fsync = ocfs2_sync_file, |
2760 | .release = ocfs2_dir_release, | |
2761 | .open = ocfs2_dir_open, | |
2762 | .unlocked_ioctl = ocfs2_ioctl, | |
2763 | #ifdef CONFIG_COMPAT | |
2764 | .compat_ioctl = ocfs2_compat_ioctl, | |
586d232b | 2765 | #endif |
53fc622b | 2766 | .flock = ocfs2_flock, |
ccd979bd | 2767 | }; |