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> |
8659ac25 | 34 | #include <linux/pipe_fs_i.h> |
7f1a37e3 | 35 | #include <linux/mount.h> |
ccd979bd MF |
36 | |
37 | #define MLOG_MASK_PREFIX ML_INODE | |
38 | #include <cluster/masklog.h> | |
39 | ||
40 | #include "ocfs2.h" | |
41 | ||
42 | #include "alloc.h" | |
43 | #include "aops.h" | |
44 | #include "dir.h" | |
45 | #include "dlmglue.h" | |
46 | #include "extent_map.h" | |
47 | #include "file.h" | |
48 | #include "sysfile.h" | |
49 | #include "inode.h" | |
ca4d147e | 50 | #include "ioctl.h" |
ccd979bd MF |
51 | #include "journal.h" |
52 | #include "mmap.h" | |
53 | #include "suballoc.h" | |
54 | #include "super.h" | |
55 | ||
56 | #include "buffer_head_io.h" | |
57 | ||
58 | static int ocfs2_sync_inode(struct inode *inode) | |
59 | { | |
60 | filemap_fdatawrite(inode->i_mapping); | |
61 | return sync_mapping_buffers(inode->i_mapping); | |
62 | } | |
63 | ||
64 | static int ocfs2_file_open(struct inode *inode, struct file *file) | |
65 | { | |
66 | int status; | |
67 | int mode = file->f_flags; | |
68 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
69 | ||
70 | mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file, | |
71 | file->f_dentry->d_name.len, file->f_dentry->d_name.name); | |
72 | ||
73 | spin_lock(&oi->ip_lock); | |
74 | ||
75 | /* Check that the inode hasn't been wiped from disk by another | |
76 | * node. If it hasn't then we're safe as long as we hold the | |
77 | * spin lock until our increment of open count. */ | |
78 | if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) { | |
79 | spin_unlock(&oi->ip_lock); | |
80 | ||
81 | status = -ENOENT; | |
82 | goto leave; | |
83 | } | |
84 | ||
85 | if (mode & O_DIRECT) | |
86 | oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT; | |
87 | ||
88 | oi->ip_open_count++; | |
89 | spin_unlock(&oi->ip_lock); | |
90 | status = 0; | |
91 | leave: | |
92 | mlog_exit(status); | |
93 | return status; | |
94 | } | |
95 | ||
96 | static int ocfs2_file_release(struct inode *inode, struct file *file) | |
97 | { | |
98 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
99 | ||
100 | mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file, | |
101 | file->f_dentry->d_name.len, | |
102 | file->f_dentry->d_name.name); | |
103 | ||
104 | spin_lock(&oi->ip_lock); | |
105 | if (!--oi->ip_open_count) | |
106 | oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT; | |
107 | spin_unlock(&oi->ip_lock); | |
108 | ||
109 | mlog_exit(0); | |
110 | ||
111 | return 0; | |
112 | } | |
113 | ||
114 | static int ocfs2_sync_file(struct file *file, | |
115 | struct dentry *dentry, | |
116 | int datasync) | |
117 | { | |
118 | int err = 0; | |
119 | journal_t *journal; | |
120 | struct inode *inode = dentry->d_inode; | |
121 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
122 | ||
123 | mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", file, dentry, datasync, | |
124 | dentry->d_name.len, dentry->d_name.name); | |
125 | ||
126 | err = ocfs2_sync_inode(dentry->d_inode); | |
127 | if (err) | |
128 | goto bail; | |
129 | ||
130 | journal = osb->journal->j_journal; | |
131 | err = journal_force_commit(journal); | |
132 | ||
133 | bail: | |
134 | mlog_exit(err); | |
135 | ||
136 | return (err < 0) ? -EIO : 0; | |
137 | } | |
138 | ||
7f1a37e3 TY |
139 | int ocfs2_should_update_atime(struct inode *inode, |
140 | struct vfsmount *vfsmnt) | |
141 | { | |
142 | struct timespec now; | |
143 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
144 | ||
145 | if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb)) | |
146 | return 0; | |
147 | ||
148 | if ((inode->i_flags & S_NOATIME) || | |
149 | ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))) | |
150 | return 0; | |
151 | ||
152 | if ((vfsmnt->mnt_flags & MNT_NOATIME) || | |
153 | ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))) | |
154 | return 0; | |
155 | ||
156 | now = CURRENT_TIME; | |
157 | if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum)) | |
158 | return 0; | |
159 | else | |
160 | return 1; | |
161 | } | |
162 | ||
163 | int ocfs2_update_inode_atime(struct inode *inode, | |
164 | struct buffer_head *bh) | |
165 | { | |
166 | int ret; | |
167 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
168 | handle_t *handle; | |
169 | ||
170 | mlog_entry_void(); | |
171 | ||
172 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); | |
173 | if (handle == NULL) { | |
174 | ret = -ENOMEM; | |
175 | mlog_errno(ret); | |
176 | goto out; | |
177 | } | |
178 | ||
179 | inode->i_atime = CURRENT_TIME; | |
180 | ret = ocfs2_mark_inode_dirty(handle, inode, bh); | |
181 | if (ret < 0) | |
182 | mlog_errno(ret); | |
183 | ||
184 | ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle); | |
185 | out: | |
186 | mlog_exit(ret); | |
187 | return ret; | |
188 | } | |
189 | ||
1fabe148 | 190 | int ocfs2_set_inode_size(handle_t *handle, |
ccd979bd MF |
191 | struct inode *inode, |
192 | struct buffer_head *fe_bh, | |
193 | u64 new_i_size) | |
194 | { | |
195 | int status; | |
196 | ||
197 | mlog_entry_void(); | |
198 | i_size_write(inode, new_i_size); | |
199 | inode->i_blocks = ocfs2_align_bytes_to_sectors(new_i_size); | |
200 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; | |
201 | ||
202 | status = ocfs2_mark_inode_dirty(handle, inode, fe_bh); | |
203 | if (status < 0) { | |
204 | mlog_errno(status); | |
205 | goto bail; | |
206 | } | |
207 | ||
208 | bail: | |
209 | mlog_exit(status); | |
210 | return status; | |
211 | } | |
212 | ||
213 | static int ocfs2_simple_size_update(struct inode *inode, | |
214 | struct buffer_head *di_bh, | |
215 | u64 new_i_size) | |
216 | { | |
217 | int ret; | |
218 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
1fabe148 | 219 | handle_t *handle = NULL; |
ccd979bd | 220 | |
65eff9cc | 221 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
ccd979bd MF |
222 | if (handle == NULL) { |
223 | ret = -ENOMEM; | |
224 | mlog_errno(ret); | |
225 | goto out; | |
226 | } | |
227 | ||
228 | ret = ocfs2_set_inode_size(handle, inode, di_bh, | |
229 | new_i_size); | |
230 | if (ret < 0) | |
231 | mlog_errno(ret); | |
232 | ||
02dc1af4 | 233 | ocfs2_commit_trans(osb, handle); |
ccd979bd MF |
234 | out: |
235 | return ret; | |
236 | } | |
237 | ||
238 | static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb, | |
239 | struct inode *inode, | |
240 | struct buffer_head *fe_bh, | |
241 | u64 new_i_size) | |
242 | { | |
243 | int status; | |
1fabe148 | 244 | handle_t *handle; |
ccd979bd MF |
245 | |
246 | mlog_entry_void(); | |
247 | ||
248 | /* TODO: This needs to actually orphan the inode in this | |
249 | * transaction. */ | |
250 | ||
65eff9cc | 251 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
ccd979bd MF |
252 | if (IS_ERR(handle)) { |
253 | status = PTR_ERR(handle); | |
254 | mlog_errno(status); | |
255 | goto out; | |
256 | } | |
257 | ||
258 | status = ocfs2_set_inode_size(handle, inode, fe_bh, new_i_size); | |
259 | if (status < 0) | |
260 | mlog_errno(status); | |
261 | ||
02dc1af4 | 262 | ocfs2_commit_trans(osb, handle); |
ccd979bd MF |
263 | out: |
264 | mlog_exit(status); | |
265 | return status; | |
266 | } | |
267 | ||
268 | static int ocfs2_truncate_file(struct inode *inode, | |
269 | struct buffer_head *di_bh, | |
270 | u64 new_i_size) | |
271 | { | |
272 | int status = 0; | |
273 | struct ocfs2_dinode *fe = NULL; | |
274 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
275 | struct ocfs2_truncate_context *tc = NULL; | |
276 | ||
b0697053 MF |
277 | mlog_entry("(inode = %llu, new_i_size = %llu\n", |
278 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
279 | (unsigned long long)new_i_size); | |
ccd979bd MF |
280 | |
281 | truncate_inode_pages(inode->i_mapping, new_i_size); | |
282 | ||
283 | fe = (struct ocfs2_dinode *) di_bh->b_data; | |
284 | if (!OCFS2_IS_VALID_DINODE(fe)) { | |
285 | OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe); | |
286 | status = -EIO; | |
287 | goto bail; | |
288 | } | |
289 | ||
290 | mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode), | |
b0697053 MF |
291 | "Inode %llu, inode i_size = %lld != di " |
292 | "i_size = %llu, i_flags = 0x%x\n", | |
293 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
ccd979bd | 294 | i_size_read(inode), |
b0697053 MF |
295 | (unsigned long long)le64_to_cpu(fe->i_size), |
296 | le32_to_cpu(fe->i_flags)); | |
ccd979bd MF |
297 | |
298 | if (new_i_size > le64_to_cpu(fe->i_size)) { | |
b0697053 MF |
299 | mlog(0, "asked to truncate file with size (%llu) to size (%llu)!\n", |
300 | (unsigned long long)le64_to_cpu(fe->i_size), | |
301 | (unsigned long long)new_i_size); | |
ccd979bd MF |
302 | status = -EINVAL; |
303 | mlog_errno(status); | |
304 | goto bail; | |
305 | } | |
306 | ||
b0697053 MF |
307 | mlog(0, "inode %llu, i_size = %llu, new_i_size = %llu\n", |
308 | (unsigned long long)le64_to_cpu(fe->i_blkno), | |
309 | (unsigned long long)le64_to_cpu(fe->i_size), | |
310 | (unsigned long long)new_i_size); | |
ccd979bd MF |
311 | |
312 | /* lets handle the simple truncate cases before doing any more | |
313 | * cluster locking. */ | |
314 | if (new_i_size == le64_to_cpu(fe->i_size)) | |
315 | goto bail; | |
316 | ||
ab0920ce MF |
317 | /* This forces other nodes to sync and drop their pages. Do |
318 | * this even if we have a truncate without allocation change - | |
319 | * ocfs2 cluster sizes can be much greater than page size, so | |
320 | * we have to truncate them anyway. */ | |
321 | status = ocfs2_data_lock(inode, 1); | |
322 | if (status < 0) { | |
323 | mlog_errno(status); | |
324 | goto bail; | |
325 | } | |
326 | ocfs2_data_unlock(inode, 1); | |
327 | ||
ccd979bd MF |
328 | if (le32_to_cpu(fe->i_clusters) == |
329 | ocfs2_clusters_for_bytes(osb->sb, new_i_size)) { | |
330 | mlog(0, "fe->i_clusters = %u, so we do a simple truncate\n", | |
331 | fe->i_clusters); | |
332 | /* No allocation change is required, so lets fast path | |
333 | * this truncate. */ | |
334 | status = ocfs2_simple_size_update(inode, di_bh, new_i_size); | |
335 | if (status < 0) | |
336 | mlog_errno(status); | |
337 | goto bail; | |
338 | } | |
339 | ||
ccd979bd MF |
340 | /* alright, we're going to need to do a full blown alloc size |
341 | * change. Orphan the inode so that recovery can complete the | |
342 | * truncate if necessary. This does the task of marking | |
343 | * i_size. */ | |
344 | status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size); | |
345 | if (status < 0) { | |
346 | mlog_errno(status); | |
347 | goto bail; | |
348 | } | |
349 | ||
350 | status = ocfs2_prepare_truncate(osb, inode, di_bh, &tc); | |
351 | if (status < 0) { | |
352 | mlog_errno(status); | |
353 | goto bail; | |
354 | } | |
355 | ||
356 | status = ocfs2_commit_truncate(osb, inode, di_bh, tc); | |
357 | if (status < 0) { | |
358 | mlog_errno(status); | |
359 | goto bail; | |
360 | } | |
361 | ||
362 | /* TODO: orphan dir cleanup here. */ | |
363 | bail: | |
364 | ||
365 | mlog_exit(status); | |
366 | return status; | |
367 | } | |
368 | ||
369 | /* | |
370 | * extend allocation only here. | |
371 | * we'll update all the disk stuff, and oip->alloc_size | |
372 | * | |
373 | * expect stuff to be locked, a transaction started and enough data / | |
374 | * metadata reservations in the contexts. | |
375 | * | |
376 | * Will return -EAGAIN, and a reason if a restart is needed. | |
377 | * If passed in, *reason will always be set, even in error. | |
378 | */ | |
379 | int ocfs2_do_extend_allocation(struct ocfs2_super *osb, | |
380 | struct inode *inode, | |
381 | u32 clusters_to_add, | |
382 | struct buffer_head *fe_bh, | |
1fabe148 | 383 | handle_t *handle, |
ccd979bd MF |
384 | struct ocfs2_alloc_context *data_ac, |
385 | struct ocfs2_alloc_context *meta_ac, | |
386 | enum ocfs2_alloc_restarted *reason_ret) | |
387 | { | |
388 | int status = 0; | |
389 | int free_extents; | |
390 | struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data; | |
391 | enum ocfs2_alloc_restarted reason = RESTART_NONE; | |
392 | u32 bit_off, num_bits; | |
393 | u64 block; | |
394 | ||
395 | BUG_ON(!clusters_to_add); | |
396 | ||
397 | free_extents = ocfs2_num_free_extents(osb, inode, fe); | |
398 | if (free_extents < 0) { | |
399 | status = free_extents; | |
400 | mlog_errno(status); | |
401 | goto leave; | |
402 | } | |
403 | ||
404 | /* there are two cases which could cause us to EAGAIN in the | |
405 | * we-need-more-metadata case: | |
406 | * 1) we haven't reserved *any* | |
407 | * 2) we are so fragmented, we've needed to add metadata too | |
408 | * many times. */ | |
409 | if (!free_extents && !meta_ac) { | |
410 | mlog(0, "we haven't reserved any metadata!\n"); | |
411 | status = -EAGAIN; | |
412 | reason = RESTART_META; | |
413 | goto leave; | |
414 | } else if ((!free_extents) | |
415 | && (ocfs2_alloc_context_bits_left(meta_ac) | |
416 | < ocfs2_extend_meta_needed(fe))) { | |
417 | mlog(0, "filesystem is really fragmented...\n"); | |
418 | status = -EAGAIN; | |
419 | reason = RESTART_META; | |
420 | goto leave; | |
421 | } | |
422 | ||
423 | status = ocfs2_claim_clusters(osb, handle, data_ac, 1, | |
424 | &bit_off, &num_bits); | |
425 | if (status < 0) { | |
426 | if (status != -ENOSPC) | |
427 | mlog_errno(status); | |
428 | goto leave; | |
429 | } | |
430 | ||
431 | BUG_ON(num_bits > clusters_to_add); | |
432 | ||
433 | /* reserve our write early -- insert_extent may update the inode */ | |
434 | status = ocfs2_journal_access(handle, inode, fe_bh, | |
435 | OCFS2_JOURNAL_ACCESS_WRITE); | |
436 | if (status < 0) { | |
437 | mlog_errno(status); | |
438 | goto leave; | |
439 | } | |
440 | ||
441 | block = ocfs2_clusters_to_blocks(osb->sb, bit_off); | |
b0697053 MF |
442 | mlog(0, "Allocating %u clusters at block %u for inode %llu\n", |
443 | num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
ccd979bd MF |
444 | status = ocfs2_insert_extent(osb, handle, inode, fe_bh, block, |
445 | num_bits, meta_ac); | |
446 | if (status < 0) { | |
447 | mlog_errno(status); | |
448 | goto leave; | |
449 | } | |
450 | ||
451 | le32_add_cpu(&fe->i_clusters, num_bits); | |
452 | spin_lock(&OCFS2_I(inode)->ip_lock); | |
453 | OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters); | |
454 | spin_unlock(&OCFS2_I(inode)->ip_lock); | |
455 | ||
456 | status = ocfs2_journal_dirty(handle, fe_bh); | |
457 | if (status < 0) { | |
458 | mlog_errno(status); | |
459 | goto leave; | |
460 | } | |
461 | ||
462 | clusters_to_add -= num_bits; | |
463 | ||
464 | if (clusters_to_add) { | |
465 | mlog(0, "need to alloc once more, clusters = %u, wanted = " | |
466 | "%u\n", fe->i_clusters, clusters_to_add); | |
467 | status = -EAGAIN; | |
468 | reason = RESTART_TRANS; | |
469 | } | |
470 | ||
471 | leave: | |
472 | mlog_exit(status); | |
473 | if (reason_ret) | |
474 | *reason_ret = reason; | |
475 | return status; | |
476 | } | |
477 | ||
478 | static int ocfs2_extend_allocation(struct inode *inode, | |
479 | u32 clusters_to_add) | |
480 | { | |
481 | int status = 0; | |
482 | int restart_func = 0; | |
483 | int drop_alloc_sem = 0; | |
484 | int credits, num_free_extents; | |
485 | u32 prev_clusters; | |
486 | struct buffer_head *bh = NULL; | |
487 | struct ocfs2_dinode *fe = NULL; | |
1fabe148 | 488 | handle_t *handle = NULL; |
ccd979bd MF |
489 | struct ocfs2_alloc_context *data_ac = NULL; |
490 | struct ocfs2_alloc_context *meta_ac = NULL; | |
491 | enum ocfs2_alloc_restarted why; | |
492 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
493 | ||
494 | mlog_entry("(clusters_to_add = %u)\n", clusters_to_add); | |
495 | ||
496 | status = ocfs2_read_block(osb, OCFS2_I(inode)->ip_blkno, &bh, | |
497 | OCFS2_BH_CACHED, inode); | |
498 | if (status < 0) { | |
499 | mlog_errno(status); | |
500 | goto leave; | |
501 | } | |
502 | ||
503 | fe = (struct ocfs2_dinode *) bh->b_data; | |
504 | if (!OCFS2_IS_VALID_DINODE(fe)) { | |
505 | OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe); | |
506 | status = -EIO; | |
507 | goto leave; | |
508 | } | |
509 | ||
510 | restart_all: | |
511 | BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters); | |
512 | ||
b0697053 | 513 | mlog(0, "extend inode %llu, i_size = %lld, fe->i_clusters = %u, " |
ccd979bd | 514 | "clusters_to_add = %u\n", |
b0697053 | 515 | (unsigned long long)OCFS2_I(inode)->ip_blkno, i_size_read(inode), |
ccd979bd MF |
516 | fe->i_clusters, clusters_to_add); |
517 | ||
ccd979bd MF |
518 | num_free_extents = ocfs2_num_free_extents(osb, |
519 | inode, | |
520 | fe); | |
521 | if (num_free_extents < 0) { | |
522 | status = num_free_extents; | |
523 | mlog_errno(status); | |
524 | goto leave; | |
525 | } | |
526 | ||
527 | if (!num_free_extents) { | |
da5cbf2f | 528 | status = ocfs2_reserve_new_metadata(osb, fe, &meta_ac); |
ccd979bd MF |
529 | if (status < 0) { |
530 | if (status != -ENOSPC) | |
531 | mlog_errno(status); | |
532 | goto leave; | |
533 | } | |
534 | } | |
535 | ||
da5cbf2f | 536 | status = ocfs2_reserve_clusters(osb, clusters_to_add, &data_ac); |
ccd979bd MF |
537 | if (status < 0) { |
538 | if (status != -ENOSPC) | |
539 | mlog_errno(status); | |
540 | goto leave; | |
541 | } | |
542 | ||
543 | /* blocks peope in read/write from reading our allocation | |
1b1dcc1b | 544 | * until we're done changing it. We depend on i_mutex to block |
ccd979bd MF |
545 | * other extend/truncate calls while we're here. Ordering wrt |
546 | * start_trans is important here -- always do it before! */ | |
547 | down_write(&OCFS2_I(inode)->ip_alloc_sem); | |
548 | drop_alloc_sem = 1; | |
549 | ||
550 | credits = ocfs2_calc_extend_credits(osb->sb, fe, clusters_to_add); | |
65eff9cc | 551 | handle = ocfs2_start_trans(osb, credits); |
ccd979bd MF |
552 | if (IS_ERR(handle)) { |
553 | status = PTR_ERR(handle); | |
554 | handle = NULL; | |
555 | mlog_errno(status); | |
556 | goto leave; | |
557 | } | |
558 | ||
559 | restarted_transaction: | |
560 | /* reserve a write to the file entry early on - that we if we | |
561 | * run out of credits in the allocation path, we can still | |
562 | * update i_size. */ | |
563 | status = ocfs2_journal_access(handle, inode, bh, | |
564 | OCFS2_JOURNAL_ACCESS_WRITE); | |
565 | if (status < 0) { | |
566 | mlog_errno(status); | |
567 | goto leave; | |
568 | } | |
569 | ||
570 | prev_clusters = OCFS2_I(inode)->ip_clusters; | |
571 | ||
572 | status = ocfs2_do_extend_allocation(osb, | |
573 | inode, | |
574 | clusters_to_add, | |
575 | bh, | |
576 | handle, | |
577 | data_ac, | |
578 | meta_ac, | |
579 | &why); | |
580 | if ((status < 0) && (status != -EAGAIN)) { | |
581 | if (status != -ENOSPC) | |
582 | mlog_errno(status); | |
583 | goto leave; | |
584 | } | |
585 | ||
586 | status = ocfs2_journal_dirty(handle, bh); | |
587 | if (status < 0) { | |
588 | mlog_errno(status); | |
589 | goto leave; | |
590 | } | |
591 | ||
592 | spin_lock(&OCFS2_I(inode)->ip_lock); | |
593 | clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters); | |
594 | spin_unlock(&OCFS2_I(inode)->ip_lock); | |
595 | ||
596 | if (why != RESTART_NONE && clusters_to_add) { | |
597 | if (why == RESTART_META) { | |
598 | mlog(0, "restarting function.\n"); | |
599 | restart_func = 1; | |
600 | } else { | |
601 | BUG_ON(why != RESTART_TRANS); | |
602 | ||
603 | mlog(0, "restarting transaction.\n"); | |
604 | /* TODO: This can be more intelligent. */ | |
605 | credits = ocfs2_calc_extend_credits(osb->sb, | |
606 | fe, | |
607 | clusters_to_add); | |
1fabe148 | 608 | status = ocfs2_extend_trans(handle, credits); |
ccd979bd MF |
609 | if (status < 0) { |
610 | /* handle still has to be committed at | |
611 | * this point. */ | |
612 | status = -ENOMEM; | |
613 | mlog_errno(status); | |
614 | goto leave; | |
615 | } | |
616 | goto restarted_transaction; | |
617 | } | |
618 | } | |
619 | ||
b0697053 MF |
620 | mlog(0, "fe: i_clusters = %u, i_size=%llu\n", |
621 | fe->i_clusters, (unsigned long long)fe->i_size); | |
ccd979bd MF |
622 | mlog(0, "inode: ip_clusters=%u, i_size=%lld\n", |
623 | OCFS2_I(inode)->ip_clusters, i_size_read(inode)); | |
624 | ||
625 | leave: | |
626 | if (drop_alloc_sem) { | |
627 | up_write(&OCFS2_I(inode)->ip_alloc_sem); | |
628 | drop_alloc_sem = 0; | |
629 | } | |
630 | if (handle) { | |
02dc1af4 | 631 | ocfs2_commit_trans(osb, handle); |
ccd979bd MF |
632 | handle = NULL; |
633 | } | |
634 | if (data_ac) { | |
635 | ocfs2_free_alloc_context(data_ac); | |
636 | data_ac = NULL; | |
637 | } | |
638 | if (meta_ac) { | |
639 | ocfs2_free_alloc_context(meta_ac); | |
640 | meta_ac = NULL; | |
641 | } | |
642 | if ((!status) && restart_func) { | |
643 | restart_func = 0; | |
644 | goto restart_all; | |
645 | } | |
646 | if (bh) { | |
647 | brelse(bh); | |
648 | bh = NULL; | |
649 | } | |
650 | ||
651 | mlog_exit(status); | |
652 | return status; | |
653 | } | |
654 | ||
655 | /* Some parts of this taken from generic_cont_expand, which turned out | |
656 | * to be too fragile to do exactly what we need without us having to | |
53013cba MF |
657 | * worry about recursive locking in ->prepare_write() and |
658 | * ->commit_write(). */ | |
ccd979bd MF |
659 | static int ocfs2_write_zero_page(struct inode *inode, |
660 | u64 size) | |
661 | { | |
662 | struct address_space *mapping = inode->i_mapping; | |
663 | struct page *page; | |
664 | unsigned long index; | |
665 | unsigned int offset; | |
1fabe148 | 666 | handle_t *handle = NULL; |
ccd979bd MF |
667 | int ret; |
668 | ||
669 | offset = (size & (PAGE_CACHE_SIZE-1)); /* Within page */ | |
670 | /* ugh. in prepare/commit_write, if from==to==start of block, we | |
671 | ** skip the prepare. make sure we never send an offset for the start | |
672 | ** of a block | |
673 | */ | |
674 | if ((offset & (inode->i_sb->s_blocksize - 1)) == 0) { | |
675 | offset++; | |
676 | } | |
677 | index = size >> PAGE_CACHE_SHIFT; | |
678 | ||
679 | page = grab_cache_page(mapping, index); | |
680 | if (!page) { | |
681 | ret = -ENOMEM; | |
682 | mlog_errno(ret); | |
683 | goto out; | |
684 | } | |
685 | ||
53013cba | 686 | ret = ocfs2_prepare_write_nolock(inode, page, offset, offset); |
ccd979bd MF |
687 | if (ret < 0) { |
688 | mlog_errno(ret); | |
689 | goto out_unlock; | |
690 | } | |
691 | ||
692 | if (ocfs2_should_order_data(inode)) { | |
693 | handle = ocfs2_start_walk_page_trans(inode, page, offset, | |
694 | offset); | |
695 | if (IS_ERR(handle)) { | |
696 | ret = PTR_ERR(handle); | |
697 | handle = NULL; | |
698 | goto out_unlock; | |
699 | } | |
700 | } | |
701 | ||
702 | /* must not update i_size! */ | |
703 | ret = block_commit_write(page, offset, offset); | |
704 | if (ret < 0) | |
705 | mlog_errno(ret); | |
706 | else | |
707 | ret = 0; | |
708 | ||
709 | if (handle) | |
02dc1af4 | 710 | ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle); |
ccd979bd MF |
711 | out_unlock: |
712 | unlock_page(page); | |
713 | page_cache_release(page); | |
714 | out: | |
715 | return ret; | |
716 | } | |
717 | ||
718 | static int ocfs2_zero_extend(struct inode *inode, | |
719 | u64 zero_to_size) | |
720 | { | |
721 | int ret = 0; | |
722 | u64 start_off; | |
723 | struct super_block *sb = inode->i_sb; | |
724 | ||
725 | start_off = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode)); | |
726 | while (start_off < zero_to_size) { | |
727 | ret = ocfs2_write_zero_page(inode, start_off); | |
728 | if (ret < 0) { | |
729 | mlog_errno(ret); | |
730 | goto out; | |
731 | } | |
732 | ||
733 | start_off += sb->s_blocksize; | |
e2057c5a MF |
734 | |
735 | /* | |
736 | * Very large extends have the potential to lock up | |
737 | * the cpu for extended periods of time. | |
738 | */ | |
739 | cond_resched(); | |
ccd979bd MF |
740 | } |
741 | ||
742 | out: | |
743 | return ret; | |
744 | } | |
745 | ||
53013cba MF |
746 | /* |
747 | * A tail_to_skip value > 0 indicates that we're being called from | |
748 | * ocfs2_file_aio_write(). This has the following implications: | |
749 | * | |
750 | * - we don't want to update i_size | |
751 | * - di_bh will be NULL, which is fine because it's only used in the | |
752 | * case where we want to update i_size. | |
753 | * - ocfs2_zero_extend() will then only be filling the hole created | |
754 | * between i_size and the start of the write. | |
755 | */ | |
ccd979bd MF |
756 | static int ocfs2_extend_file(struct inode *inode, |
757 | struct buffer_head *di_bh, | |
53013cba MF |
758 | u64 new_i_size, |
759 | size_t tail_to_skip) | |
ccd979bd MF |
760 | { |
761 | int ret = 0; | |
762 | u32 clusters_to_add; | |
763 | ||
53013cba MF |
764 | BUG_ON(!tail_to_skip && !di_bh); |
765 | ||
ccd979bd MF |
766 | /* setattr sometimes calls us like this. */ |
767 | if (new_i_size == 0) | |
768 | goto out; | |
769 | ||
770 | if (i_size_read(inode) == new_i_size) | |
771 | goto out; | |
772 | BUG_ON(new_i_size < i_size_read(inode)); | |
773 | ||
774 | clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size) - | |
775 | OCFS2_I(inode)->ip_clusters; | |
776 | ||
0effef77 MF |
777 | /* |
778 | * protect the pages that ocfs2_zero_extend is going to be | |
779 | * pulling into the page cache.. we do this before the | |
780 | * metadata extend so that we don't get into the situation | |
781 | * where we've extended the metadata but can't get the data | |
782 | * lock to zero. | |
783 | */ | |
784 | ret = ocfs2_data_lock(inode, 1); | |
785 | if (ret < 0) { | |
786 | mlog_errno(ret); | |
787 | goto out; | |
788 | } | |
ccd979bd | 789 | |
0effef77 | 790 | if (clusters_to_add) { |
53013cba | 791 | ret = ocfs2_extend_allocation(inode, clusters_to_add); |
ccd979bd MF |
792 | if (ret < 0) { |
793 | mlog_errno(ret); | |
53013cba | 794 | goto out_unlock; |
ccd979bd | 795 | } |
0effef77 | 796 | } |
ccd979bd | 797 | |
0effef77 MF |
798 | /* |
799 | * Call this even if we don't add any clusters to the tree. We | |
800 | * still need to zero the area between the old i_size and the | |
801 | * new i_size. | |
802 | */ | |
803 | ret = ocfs2_zero_extend(inode, (u64)new_i_size - tail_to_skip); | |
804 | if (ret < 0) { | |
805 | mlog_errno(ret); | |
806 | goto out_unlock; | |
53013cba MF |
807 | } |
808 | ||
809 | if (!tail_to_skip) { | |
810 | /* We're being called from ocfs2_setattr() which wants | |
811 | * us to update i_size */ | |
812 | ret = ocfs2_simple_size_update(inode, di_bh, new_i_size); | |
813 | if (ret < 0) | |
814 | mlog_errno(ret); | |
ccd979bd MF |
815 | } |
816 | ||
53013cba | 817 | out_unlock: |
0effef77 | 818 | ocfs2_data_unlock(inode, 1); |
53013cba | 819 | |
ccd979bd MF |
820 | out: |
821 | return ret; | |
822 | } | |
823 | ||
824 | int ocfs2_setattr(struct dentry *dentry, struct iattr *attr) | |
825 | { | |
826 | int status = 0, size_change; | |
827 | struct inode *inode = dentry->d_inode; | |
828 | struct super_block *sb = inode->i_sb; | |
829 | struct ocfs2_super *osb = OCFS2_SB(sb); | |
830 | struct buffer_head *bh = NULL; | |
1fabe148 | 831 | handle_t *handle = NULL; |
ccd979bd MF |
832 | |
833 | mlog_entry("(0x%p, '%.*s')\n", dentry, | |
834 | dentry->d_name.len, dentry->d_name.name); | |
835 | ||
836 | if (attr->ia_valid & ATTR_MODE) | |
837 | mlog(0, "mode change: %d\n", attr->ia_mode); | |
838 | if (attr->ia_valid & ATTR_UID) | |
839 | mlog(0, "uid change: %d\n", attr->ia_uid); | |
840 | if (attr->ia_valid & ATTR_GID) | |
841 | mlog(0, "gid change: %d\n", attr->ia_gid); | |
842 | if (attr->ia_valid & ATTR_SIZE) | |
843 | mlog(0, "size change...\n"); | |
844 | if (attr->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME)) | |
845 | mlog(0, "time change...\n"); | |
846 | ||
847 | #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \ | |
848 | | ATTR_GID | ATTR_UID | ATTR_MODE) | |
849 | if (!(attr->ia_valid & OCFS2_VALID_ATTRS)) { | |
850 | mlog(0, "can't handle attrs: 0x%x\n", attr->ia_valid); | |
851 | return 0; | |
852 | } | |
853 | ||
854 | status = inode_change_ok(inode, attr); | |
855 | if (status) | |
856 | return status; | |
857 | ||
858 | size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE; | |
859 | if (size_change) { | |
860 | status = ocfs2_rw_lock(inode, 1); | |
861 | if (status < 0) { | |
862 | mlog_errno(status); | |
863 | goto bail; | |
864 | } | |
865 | } | |
866 | ||
4bcec184 | 867 | status = ocfs2_meta_lock(inode, &bh, 1); |
ccd979bd MF |
868 | if (status < 0) { |
869 | if (status != -ENOENT) | |
870 | mlog_errno(status); | |
871 | goto bail_unlock_rw; | |
872 | } | |
873 | ||
874 | if (size_change && attr->ia_size != i_size_read(inode)) { | |
875 | if (i_size_read(inode) > attr->ia_size) | |
876 | status = ocfs2_truncate_file(inode, bh, attr->ia_size); | |
877 | else | |
53013cba | 878 | status = ocfs2_extend_file(inode, bh, attr->ia_size, 0); |
ccd979bd MF |
879 | if (status < 0) { |
880 | if (status != -ENOSPC) | |
881 | mlog_errno(status); | |
882 | status = -ENOSPC; | |
883 | goto bail_unlock; | |
884 | } | |
885 | } | |
886 | ||
65eff9cc | 887 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
ccd979bd MF |
888 | if (IS_ERR(handle)) { |
889 | status = PTR_ERR(handle); | |
890 | mlog_errno(status); | |
891 | goto bail_unlock; | |
892 | } | |
893 | ||
894 | status = inode_setattr(inode, attr); | |
895 | if (status < 0) { | |
896 | mlog_errno(status); | |
897 | goto bail_commit; | |
898 | } | |
899 | ||
900 | status = ocfs2_mark_inode_dirty(handle, inode, bh); | |
901 | if (status < 0) | |
902 | mlog_errno(status); | |
903 | ||
904 | bail_commit: | |
02dc1af4 | 905 | ocfs2_commit_trans(osb, handle); |
ccd979bd MF |
906 | bail_unlock: |
907 | ocfs2_meta_unlock(inode, 1); | |
908 | bail_unlock_rw: | |
909 | if (size_change) | |
910 | ocfs2_rw_unlock(inode, 1); | |
911 | bail: | |
912 | if (bh) | |
913 | brelse(bh); | |
914 | ||
915 | mlog_exit(status); | |
916 | return status; | |
917 | } | |
918 | ||
919 | int ocfs2_getattr(struct vfsmount *mnt, | |
920 | struct dentry *dentry, | |
921 | struct kstat *stat) | |
922 | { | |
923 | struct inode *inode = dentry->d_inode; | |
924 | struct super_block *sb = dentry->d_inode->i_sb; | |
925 | struct ocfs2_super *osb = sb->s_fs_info; | |
926 | int err; | |
927 | ||
928 | mlog_entry_void(); | |
929 | ||
930 | err = ocfs2_inode_revalidate(dentry); | |
931 | if (err) { | |
932 | if (err != -ENOENT) | |
933 | mlog_errno(err); | |
934 | goto bail; | |
935 | } | |
936 | ||
937 | generic_fillattr(inode, stat); | |
938 | ||
939 | /* We set the blksize from the cluster size for performance */ | |
940 | stat->blksize = osb->s_clustersize; | |
941 | ||
942 | bail: | |
943 | mlog_exit(err); | |
944 | ||
945 | return err; | |
946 | } | |
947 | ||
948 | static int ocfs2_write_remove_suid(struct inode *inode) | |
949 | { | |
950 | int ret; | |
951 | struct buffer_head *bh = NULL; | |
952 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
1fabe148 | 953 | handle_t *handle; |
ccd979bd MF |
954 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); |
955 | struct ocfs2_dinode *di; | |
956 | ||
b0697053 MF |
957 | mlog_entry("(Inode %llu, mode 0%o)\n", |
958 | (unsigned long long)oi->ip_blkno, inode->i_mode); | |
ccd979bd | 959 | |
65eff9cc | 960 | handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS); |
ccd979bd MF |
961 | if (handle == NULL) { |
962 | ret = -ENOMEM; | |
963 | mlog_errno(ret); | |
964 | goto out; | |
965 | } | |
966 | ||
967 | ret = ocfs2_read_block(osb, oi->ip_blkno, &bh, OCFS2_BH_CACHED, inode); | |
968 | if (ret < 0) { | |
969 | mlog_errno(ret); | |
970 | goto out_trans; | |
971 | } | |
972 | ||
973 | ret = ocfs2_journal_access(handle, inode, bh, | |
974 | OCFS2_JOURNAL_ACCESS_WRITE); | |
975 | if (ret < 0) { | |
976 | mlog_errno(ret); | |
977 | goto out_bh; | |
978 | } | |
979 | ||
980 | inode->i_mode &= ~S_ISUID; | |
981 | if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP)) | |
982 | inode->i_mode &= ~S_ISGID; | |
983 | ||
984 | di = (struct ocfs2_dinode *) bh->b_data; | |
985 | di->i_mode = cpu_to_le16(inode->i_mode); | |
986 | ||
987 | ret = ocfs2_journal_dirty(handle, bh); | |
988 | if (ret < 0) | |
989 | mlog_errno(ret); | |
990 | out_bh: | |
991 | brelse(bh); | |
992 | out_trans: | |
02dc1af4 | 993 | ocfs2_commit_trans(osb, handle); |
ccd979bd MF |
994 | out: |
995 | mlog_exit(ret); | |
996 | return ret; | |
997 | } | |
998 | ||
8659ac25 TY |
999 | static int ocfs2_prepare_inode_for_write(struct dentry *dentry, |
1000 | loff_t *ppos, | |
1001 | size_t count, | |
1002 | int appending) | |
ccd979bd | 1003 | { |
8659ac25 TY |
1004 | int ret = 0, meta_level = appending; |
1005 | struct inode *inode = dentry->d_inode; | |
ccd979bd | 1006 | u32 clusters; |
ccd979bd | 1007 | loff_t newsize, saved_pos; |
ccd979bd | 1008 | |
ccd979bd MF |
1009 | /* |
1010 | * We sample i_size under a read level meta lock to see if our write | |
1011 | * is extending the file, if it is we back off and get a write level | |
1012 | * meta lock. | |
1013 | */ | |
ccd979bd | 1014 | for(;;) { |
4bcec184 | 1015 | ret = ocfs2_meta_lock(inode, NULL, meta_level); |
ccd979bd MF |
1016 | if (ret < 0) { |
1017 | meta_level = -1; | |
1018 | mlog_errno(ret); | |
1019 | goto out; | |
1020 | } | |
1021 | ||
1022 | /* Clear suid / sgid if necessary. We do this here | |
1023 | * instead of later in the write path because | |
1024 | * remove_suid() calls ->setattr without any hint that | |
1025 | * we may have already done our cluster locking. Since | |
1026 | * ocfs2_setattr() *must* take cluster locks to | |
1027 | * proceeed, this will lead us to recursively lock the | |
1028 | * inode. There's also the dinode i_size state which | |
1029 | * can be lost via setattr during extending writes (we | |
1030 | * set inode->i_size at the end of a write. */ | |
8659ac25 | 1031 | if (should_remove_suid(dentry)) { |
ccd979bd MF |
1032 | if (meta_level == 0) { |
1033 | ocfs2_meta_unlock(inode, meta_level); | |
1034 | meta_level = 1; | |
1035 | continue; | |
1036 | } | |
1037 | ||
1038 | ret = ocfs2_write_remove_suid(inode); | |
1039 | if (ret < 0) { | |
1040 | mlog_errno(ret); | |
8659ac25 | 1041 | goto out_unlock; |
ccd979bd MF |
1042 | } |
1043 | } | |
1044 | ||
1045 | /* work on a copy of ppos until we're sure that we won't have | |
1046 | * to recalculate it due to relocking. */ | |
8659ac25 | 1047 | if (appending) { |
ccd979bd MF |
1048 | saved_pos = i_size_read(inode); |
1049 | mlog(0, "O_APPEND: inode->i_size=%llu\n", saved_pos); | |
1050 | } else { | |
8659ac25 | 1051 | saved_pos = *ppos; |
ccd979bd | 1052 | } |
8659ac25 | 1053 | newsize = count + saved_pos; |
ccd979bd | 1054 | |
215c7f9f MF |
1055 | mlog(0, "pos=%lld newsize=%lld cursize=%lld\n", |
1056 | (long long) saved_pos, (long long) newsize, | |
1057 | (long long) i_size_read(inode)); | |
ccd979bd MF |
1058 | |
1059 | /* No need for a higher level metadata lock if we're | |
1060 | * never going past i_size. */ | |
1061 | if (newsize <= i_size_read(inode)) | |
1062 | break; | |
1063 | ||
1064 | if (meta_level == 0) { | |
1065 | ocfs2_meta_unlock(inode, meta_level); | |
1066 | meta_level = 1; | |
1067 | continue; | |
1068 | } | |
1069 | ||
1070 | spin_lock(&OCFS2_I(inode)->ip_lock); | |
1071 | clusters = ocfs2_clusters_for_bytes(inode->i_sb, newsize) - | |
1072 | OCFS2_I(inode)->ip_clusters; | |
1073 | spin_unlock(&OCFS2_I(inode)->ip_lock); | |
1074 | ||
1075 | mlog(0, "Writing at EOF, may need more allocation: " | |
215c7f9f MF |
1076 | "i_size = %lld, newsize = %lld, need %u clusters\n", |
1077 | (long long) i_size_read(inode), (long long) newsize, | |
1078 | clusters); | |
ccd979bd MF |
1079 | |
1080 | /* We only want to continue the rest of this loop if | |
1081 | * our extend will actually require more | |
1082 | * allocation. */ | |
1083 | if (!clusters) | |
1084 | break; | |
1085 | ||
8659ac25 | 1086 | ret = ocfs2_extend_file(inode, NULL, newsize, count); |
ccd979bd MF |
1087 | if (ret < 0) { |
1088 | if (ret != -ENOSPC) | |
1089 | mlog_errno(ret); | |
8659ac25 | 1090 | goto out_unlock; |
ccd979bd | 1091 | } |
ccd979bd MF |
1092 | break; |
1093 | } | |
1094 | ||
8659ac25 TY |
1095 | if (appending) |
1096 | *ppos = saved_pos; | |
1097 | ||
1098 | out_unlock: | |
ccd979bd | 1099 | ocfs2_meta_unlock(inode, meta_level); |
8659ac25 TY |
1100 | |
1101 | out: | |
1102 | return ret; | |
1103 | } | |
1104 | ||
1105 | static ssize_t ocfs2_file_aio_write(struct kiocb *iocb, | |
1106 | const struct iovec *iov, | |
1107 | unsigned long nr_segs, | |
1108 | loff_t pos) | |
1109 | { | |
1110 | int ret, rw_level, have_alloc_sem = 0; | |
1111 | struct file *filp = iocb->ki_filp; | |
1112 | struct inode *inode = filp->f_dentry->d_inode; | |
1113 | int appending = filp->f_flags & O_APPEND ? 1 : 0; | |
1114 | ||
1115 | mlog_entry("(0x%p, %u, '%.*s')\n", filp, | |
1116 | (unsigned int)nr_segs, | |
1117 | filp->f_dentry->d_name.len, | |
1118 | filp->f_dentry->d_name.name); | |
1119 | ||
1120 | /* happy write of zero bytes */ | |
1121 | if (iocb->ki_left == 0) | |
1122 | return 0; | |
1123 | ||
1124 | mutex_lock(&inode->i_mutex); | |
1125 | /* to match setattr's i_mutex -> i_alloc_sem -> rw_lock ordering */ | |
1126 | if (filp->f_flags & O_DIRECT) { | |
1127 | have_alloc_sem = 1; | |
1128 | down_read(&inode->i_alloc_sem); | |
1129 | } | |
1130 | ||
1131 | /* concurrent O_DIRECT writes are allowed */ | |
1132 | rw_level = (filp->f_flags & O_DIRECT) ? 0 : 1; | |
1133 | ret = ocfs2_rw_lock(inode, rw_level); | |
1134 | if (ret < 0) { | |
1135 | rw_level = -1; | |
1136 | mlog_errno(ret); | |
1137 | goto out; | |
1138 | } | |
1139 | ||
1140 | ret = ocfs2_prepare_inode_for_write(filp->f_dentry, &iocb->ki_pos, | |
1141 | iocb->ki_left, appending); | |
1142 | if (ret < 0) { | |
1143 | mlog_errno(ret); | |
1144 | goto out; | |
1145 | } | |
ccd979bd MF |
1146 | |
1147 | /* communicate with ocfs2_dio_end_io */ | |
1148 | ocfs2_iocb_set_rw_locked(iocb); | |
1149 | ||
027445c3 | 1150 | ret = generic_file_aio_write_nolock(iocb, iov, nr_segs, iocb->ki_pos); |
ccd979bd MF |
1151 | |
1152 | /* buffered aio wouldn't have proper lock coverage today */ | |
1153 | BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT)); | |
1154 | ||
1155 | /* | |
1156 | * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io | |
1157 | * function pointer which is called when o_direct io completes so that | |
1158 | * it can unlock our rw lock. (it's the clustered equivalent of | |
1159 | * i_alloc_sem; protects truncate from racing with pending ios). | |
1160 | * Unfortunately there are error cases which call end_io and others | |
1161 | * that don't. so we don't have to unlock the rw_lock if either an | |
1162 | * async dio is going to do it in the future or an end_io after an | |
1163 | * error has already done it. | |
1164 | */ | |
1165 | if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) { | |
1166 | rw_level = -1; | |
1167 | have_alloc_sem = 0; | |
1168 | } | |
1169 | ||
1170 | out: | |
ccd979bd MF |
1171 | if (have_alloc_sem) |
1172 | up_read(&inode->i_alloc_sem); | |
1173 | if (rw_level != -1) | |
1174 | ocfs2_rw_unlock(inode, rw_level); | |
1b1dcc1b | 1175 | mutex_unlock(&inode->i_mutex); |
ccd979bd MF |
1176 | |
1177 | mlog_exit(ret); | |
1178 | return ret; | |
1179 | } | |
1180 | ||
8659ac25 TY |
1181 | static ssize_t ocfs2_file_splice_write(struct pipe_inode_info *pipe, |
1182 | struct file *out, | |
1183 | loff_t *ppos, | |
1184 | size_t len, | |
1185 | unsigned int flags) | |
1186 | { | |
1187 | int ret; | |
1188 | struct inode *inode = out->f_dentry->d_inode; | |
1189 | ||
1190 | mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", out, pipe, | |
1191 | (unsigned int)len, | |
1192 | out->f_dentry->d_name.len, | |
1193 | out->f_dentry->d_name.name); | |
1194 | ||
1195 | inode_double_lock(inode, pipe->inode); | |
1196 | ||
1197 | ret = ocfs2_rw_lock(inode, 1); | |
1198 | if (ret < 0) { | |
1199 | mlog_errno(ret); | |
1200 | goto out; | |
1201 | } | |
1202 | ||
1203 | ret = ocfs2_prepare_inode_for_write(out->f_dentry, ppos, len, 0); | |
1204 | if (ret < 0) { | |
1205 | mlog_errno(ret); | |
1206 | goto out_unlock; | |
1207 | } | |
1208 | ||
1209 | /* ok, we're done with i_size and alloc work */ | |
1210 | ret = generic_file_splice_write_nolock(pipe, out, ppos, len, flags); | |
1211 | ||
1212 | out_unlock: | |
1213 | ocfs2_rw_unlock(inode, 1); | |
1214 | out: | |
1215 | inode_double_unlock(inode, pipe->inode); | |
1216 | ||
1217 | mlog_exit(ret); | |
1218 | return ret; | |
1219 | } | |
1220 | ||
1221 | static ssize_t ocfs2_file_splice_read(struct file *in, | |
1222 | loff_t *ppos, | |
1223 | struct pipe_inode_info *pipe, | |
1224 | size_t len, | |
1225 | unsigned int flags) | |
1226 | { | |
1227 | int ret = 0; | |
1228 | struct inode *inode = in->f_dentry->d_inode; | |
1229 | ||
1230 | mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", in, pipe, | |
1231 | (unsigned int)len, | |
1232 | in->f_dentry->d_name.len, | |
1233 | in->f_dentry->d_name.name); | |
1234 | ||
1235 | /* | |
1236 | * See the comment in ocfs2_file_aio_read() | |
1237 | */ | |
1238 | ret = ocfs2_meta_lock(inode, NULL, 0); | |
1239 | if (ret < 0) { | |
1240 | mlog_errno(ret); | |
1241 | goto bail; | |
1242 | } | |
1243 | ocfs2_meta_unlock(inode, 0); | |
1244 | ||
1245 | ret = generic_file_splice_read(in, ppos, pipe, len, flags); | |
1246 | ||
1247 | bail: | |
1248 | mlog_exit(ret); | |
1249 | return ret; | |
1250 | } | |
1251 | ||
ccd979bd | 1252 | static ssize_t ocfs2_file_aio_read(struct kiocb *iocb, |
027445c3 BP |
1253 | const struct iovec *iov, |
1254 | unsigned long nr_segs, | |
ccd979bd MF |
1255 | loff_t pos) |
1256 | { | |
1257 | int ret = 0, rw_level = -1, have_alloc_sem = 0; | |
1258 | struct file *filp = iocb->ki_filp; | |
1259 | struct inode *inode = filp->f_dentry->d_inode; | |
ccd979bd | 1260 | |
027445c3 BP |
1261 | mlog_entry("(0x%p, %u, '%.*s')\n", filp, |
1262 | (unsigned int)nr_segs, | |
ccd979bd MF |
1263 | filp->f_dentry->d_name.len, |
1264 | filp->f_dentry->d_name.name); | |
1265 | ||
1266 | if (!inode) { | |
1267 | ret = -EINVAL; | |
1268 | mlog_errno(ret); | |
1269 | goto bail; | |
1270 | } | |
1271 | ||
ccd979bd MF |
1272 | /* |
1273 | * buffered reads protect themselves in ->readpage(). O_DIRECT reads | |
1274 | * need locks to protect pending reads from racing with truncate. | |
1275 | */ | |
1276 | if (filp->f_flags & O_DIRECT) { | |
1277 | down_read(&inode->i_alloc_sem); | |
1278 | have_alloc_sem = 1; | |
1279 | ||
1280 | ret = ocfs2_rw_lock(inode, 0); | |
1281 | if (ret < 0) { | |
1282 | mlog_errno(ret); | |
1283 | goto bail; | |
1284 | } | |
1285 | rw_level = 0; | |
1286 | /* communicate with ocfs2_dio_end_io */ | |
1287 | ocfs2_iocb_set_rw_locked(iocb); | |
1288 | } | |
1289 | ||
c4374f8a MF |
1290 | /* |
1291 | * We're fine letting folks race truncates and extending | |
1292 | * writes with read across the cluster, just like they can | |
1293 | * locally. Hence no rw_lock during read. | |
1294 | * | |
1295 | * Take and drop the meta data lock to update inode fields | |
1296 | * like i_size. This allows the checks down below | |
1297 | * generic_file_aio_read() a chance of actually working. | |
1298 | */ | |
4bcec184 | 1299 | ret = ocfs2_meta_lock(inode, NULL, 0); |
c4374f8a MF |
1300 | if (ret < 0) { |
1301 | mlog_errno(ret); | |
1302 | goto bail; | |
1303 | } | |
1304 | ocfs2_meta_unlock(inode, 0); | |
1305 | ||
027445c3 | 1306 | ret = generic_file_aio_read(iocb, iov, nr_segs, iocb->ki_pos); |
ccd979bd MF |
1307 | if (ret == -EINVAL) |
1308 | mlog(ML_ERROR, "generic_file_aio_read returned -EINVAL\n"); | |
1309 | ||
1310 | /* buffered aio wouldn't have proper lock coverage today */ | |
1311 | BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT)); | |
1312 | ||
1313 | /* see ocfs2_file_aio_write */ | |
1314 | if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) { | |
1315 | rw_level = -1; | |
1316 | have_alloc_sem = 0; | |
1317 | } | |
1318 | ||
1319 | bail: | |
1320 | if (have_alloc_sem) | |
1321 | up_read(&inode->i_alloc_sem); | |
1322 | if (rw_level != -1) | |
1323 | ocfs2_rw_unlock(inode, rw_level); | |
1324 | mlog_exit(ret); | |
1325 | ||
1326 | return ret; | |
1327 | } | |
1328 | ||
1329 | struct inode_operations ocfs2_file_iops = { | |
1330 | .setattr = ocfs2_setattr, | |
1331 | .getattr = ocfs2_getattr, | |
1332 | }; | |
1333 | ||
1334 | struct inode_operations ocfs2_special_file_iops = { | |
1335 | .setattr = ocfs2_setattr, | |
1336 | .getattr = ocfs2_getattr, | |
1337 | }; | |
1338 | ||
4b6f5d20 | 1339 | const struct file_operations ocfs2_fops = { |
ccd979bd MF |
1340 | .read = do_sync_read, |
1341 | .write = do_sync_write, | |
1342 | .sendfile = generic_file_sendfile, | |
1343 | .mmap = ocfs2_mmap, | |
1344 | .fsync = ocfs2_sync_file, | |
1345 | .release = ocfs2_file_release, | |
1346 | .open = ocfs2_file_open, | |
1347 | .aio_read = ocfs2_file_aio_read, | |
1348 | .aio_write = ocfs2_file_aio_write, | |
ca4d147e | 1349 | .ioctl = ocfs2_ioctl, |
8659ac25 TY |
1350 | .splice_read = ocfs2_file_splice_read, |
1351 | .splice_write = ocfs2_file_splice_write, | |
ccd979bd MF |
1352 | }; |
1353 | ||
4b6f5d20 | 1354 | const struct file_operations ocfs2_dops = { |
ccd979bd MF |
1355 | .read = generic_read_dir, |
1356 | .readdir = ocfs2_readdir, | |
1357 | .fsync = ocfs2_sync_file, | |
ca4d147e | 1358 | .ioctl = ocfs2_ioctl, |
ccd979bd | 1359 | }; |