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> |
ccd979bd MF |
34 | |
35 | #define MLOG_MASK_PREFIX ML_INODE | |
36 | #include <cluster/masklog.h> | |
37 | ||
38 | #include "ocfs2.h" | |
39 | ||
40 | #include "alloc.h" | |
41 | #include "aops.h" | |
42 | #include "dir.h" | |
43 | #include "dlmglue.h" | |
44 | #include "extent_map.h" | |
45 | #include "file.h" | |
46 | #include "sysfile.h" | |
47 | #include "inode.h" | |
ca4d147e | 48 | #include "ioctl.h" |
ccd979bd MF |
49 | #include "journal.h" |
50 | #include "mmap.h" | |
51 | #include "suballoc.h" | |
52 | #include "super.h" | |
53 | ||
54 | #include "buffer_head_io.h" | |
55 | ||
56 | static int ocfs2_sync_inode(struct inode *inode) | |
57 | { | |
58 | filemap_fdatawrite(inode->i_mapping); | |
59 | return sync_mapping_buffers(inode->i_mapping); | |
60 | } | |
61 | ||
62 | static int ocfs2_file_open(struct inode *inode, struct file *file) | |
63 | { | |
64 | int status; | |
65 | int mode = file->f_flags; | |
66 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
67 | ||
68 | mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file, | |
69 | file->f_dentry->d_name.len, file->f_dentry->d_name.name); | |
70 | ||
71 | spin_lock(&oi->ip_lock); | |
72 | ||
73 | /* Check that the inode hasn't been wiped from disk by another | |
74 | * node. If it hasn't then we're safe as long as we hold the | |
75 | * spin lock until our increment of open count. */ | |
76 | if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) { | |
77 | spin_unlock(&oi->ip_lock); | |
78 | ||
79 | status = -ENOENT; | |
80 | goto leave; | |
81 | } | |
82 | ||
83 | if (mode & O_DIRECT) | |
84 | oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT; | |
85 | ||
86 | oi->ip_open_count++; | |
87 | spin_unlock(&oi->ip_lock); | |
88 | status = 0; | |
89 | leave: | |
90 | mlog_exit(status); | |
91 | return status; | |
92 | } | |
93 | ||
94 | static int ocfs2_file_release(struct inode *inode, struct file *file) | |
95 | { | |
96 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
97 | ||
98 | mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file, | |
99 | file->f_dentry->d_name.len, | |
100 | file->f_dentry->d_name.name); | |
101 | ||
102 | spin_lock(&oi->ip_lock); | |
103 | if (!--oi->ip_open_count) | |
104 | oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT; | |
105 | spin_unlock(&oi->ip_lock); | |
106 | ||
107 | mlog_exit(0); | |
108 | ||
109 | return 0; | |
110 | } | |
111 | ||
112 | static int ocfs2_sync_file(struct file *file, | |
113 | struct dentry *dentry, | |
114 | int datasync) | |
115 | { | |
116 | int err = 0; | |
117 | journal_t *journal; | |
118 | struct inode *inode = dentry->d_inode; | |
119 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
120 | ||
121 | mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", file, dentry, datasync, | |
122 | dentry->d_name.len, dentry->d_name.name); | |
123 | ||
124 | err = ocfs2_sync_inode(dentry->d_inode); | |
125 | if (err) | |
126 | goto bail; | |
127 | ||
128 | journal = osb->journal->j_journal; | |
129 | err = journal_force_commit(journal); | |
130 | ||
131 | bail: | |
132 | mlog_exit(err); | |
133 | ||
134 | return (err < 0) ? -EIO : 0; | |
135 | } | |
136 | ||
137 | int ocfs2_set_inode_size(struct ocfs2_journal_handle *handle, | |
138 | struct inode *inode, | |
139 | struct buffer_head *fe_bh, | |
140 | u64 new_i_size) | |
141 | { | |
142 | int status; | |
143 | ||
144 | mlog_entry_void(); | |
145 | i_size_write(inode, new_i_size); | |
146 | inode->i_blocks = ocfs2_align_bytes_to_sectors(new_i_size); | |
147 | inode->i_ctime = inode->i_mtime = CURRENT_TIME; | |
148 | ||
149 | status = ocfs2_mark_inode_dirty(handle, inode, fe_bh); | |
150 | if (status < 0) { | |
151 | mlog_errno(status); | |
152 | goto bail; | |
153 | } | |
154 | ||
155 | bail: | |
156 | mlog_exit(status); | |
157 | return status; | |
158 | } | |
159 | ||
160 | static int ocfs2_simple_size_update(struct inode *inode, | |
161 | struct buffer_head *di_bh, | |
162 | u64 new_i_size) | |
163 | { | |
164 | int ret; | |
165 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
166 | struct ocfs2_journal_handle *handle = NULL; | |
167 | ||
168 | handle = ocfs2_start_trans(osb, NULL, | |
169 | OCFS2_INODE_UPDATE_CREDITS); | |
170 | if (handle == NULL) { | |
171 | ret = -ENOMEM; | |
172 | mlog_errno(ret); | |
173 | goto out; | |
174 | } | |
175 | ||
176 | ret = ocfs2_set_inode_size(handle, inode, di_bh, | |
177 | new_i_size); | |
178 | if (ret < 0) | |
179 | mlog_errno(ret); | |
180 | ||
181 | ocfs2_commit_trans(handle); | |
182 | out: | |
183 | return ret; | |
184 | } | |
185 | ||
186 | static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb, | |
187 | struct inode *inode, | |
188 | struct buffer_head *fe_bh, | |
189 | u64 new_i_size) | |
190 | { | |
191 | int status; | |
192 | struct ocfs2_journal_handle *handle; | |
193 | ||
194 | mlog_entry_void(); | |
195 | ||
196 | /* TODO: This needs to actually orphan the inode in this | |
197 | * transaction. */ | |
198 | ||
199 | handle = ocfs2_start_trans(osb, NULL, OCFS2_INODE_UPDATE_CREDITS); | |
200 | if (IS_ERR(handle)) { | |
201 | status = PTR_ERR(handle); | |
202 | mlog_errno(status); | |
203 | goto out; | |
204 | } | |
205 | ||
206 | status = ocfs2_set_inode_size(handle, inode, fe_bh, new_i_size); | |
207 | if (status < 0) | |
208 | mlog_errno(status); | |
209 | ||
210 | ocfs2_commit_trans(handle); | |
211 | out: | |
212 | mlog_exit(status); | |
213 | return status; | |
214 | } | |
215 | ||
216 | static int ocfs2_truncate_file(struct inode *inode, | |
217 | struct buffer_head *di_bh, | |
218 | u64 new_i_size) | |
219 | { | |
220 | int status = 0; | |
221 | struct ocfs2_dinode *fe = NULL; | |
222 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
223 | struct ocfs2_truncate_context *tc = NULL; | |
224 | ||
b0697053 MF |
225 | mlog_entry("(inode = %llu, new_i_size = %llu\n", |
226 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
227 | (unsigned long long)new_i_size); | |
ccd979bd MF |
228 | |
229 | truncate_inode_pages(inode->i_mapping, new_i_size); | |
230 | ||
231 | fe = (struct ocfs2_dinode *) di_bh->b_data; | |
232 | if (!OCFS2_IS_VALID_DINODE(fe)) { | |
233 | OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe); | |
234 | status = -EIO; | |
235 | goto bail; | |
236 | } | |
237 | ||
238 | mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode), | |
b0697053 MF |
239 | "Inode %llu, inode i_size = %lld != di " |
240 | "i_size = %llu, i_flags = 0x%x\n", | |
241 | (unsigned long long)OCFS2_I(inode)->ip_blkno, | |
ccd979bd | 242 | i_size_read(inode), |
b0697053 MF |
243 | (unsigned long long)le64_to_cpu(fe->i_size), |
244 | le32_to_cpu(fe->i_flags)); | |
ccd979bd MF |
245 | |
246 | if (new_i_size > le64_to_cpu(fe->i_size)) { | |
b0697053 MF |
247 | mlog(0, "asked to truncate file with size (%llu) to size (%llu)!\n", |
248 | (unsigned long long)le64_to_cpu(fe->i_size), | |
249 | (unsigned long long)new_i_size); | |
ccd979bd MF |
250 | status = -EINVAL; |
251 | mlog_errno(status); | |
252 | goto bail; | |
253 | } | |
254 | ||
b0697053 MF |
255 | mlog(0, "inode %llu, i_size = %llu, new_i_size = %llu\n", |
256 | (unsigned long long)le64_to_cpu(fe->i_blkno), | |
257 | (unsigned long long)le64_to_cpu(fe->i_size), | |
258 | (unsigned long long)new_i_size); | |
ccd979bd MF |
259 | |
260 | /* lets handle the simple truncate cases before doing any more | |
261 | * cluster locking. */ | |
262 | if (new_i_size == le64_to_cpu(fe->i_size)) | |
263 | goto bail; | |
264 | ||
ab0920ce MF |
265 | /* This forces other nodes to sync and drop their pages. Do |
266 | * this even if we have a truncate without allocation change - | |
267 | * ocfs2 cluster sizes can be much greater than page size, so | |
268 | * we have to truncate them anyway. */ | |
269 | status = ocfs2_data_lock(inode, 1); | |
270 | if (status < 0) { | |
271 | mlog_errno(status); | |
272 | goto bail; | |
273 | } | |
274 | ocfs2_data_unlock(inode, 1); | |
275 | ||
ccd979bd MF |
276 | if (le32_to_cpu(fe->i_clusters) == |
277 | ocfs2_clusters_for_bytes(osb->sb, new_i_size)) { | |
278 | mlog(0, "fe->i_clusters = %u, so we do a simple truncate\n", | |
279 | fe->i_clusters); | |
280 | /* No allocation change is required, so lets fast path | |
281 | * this truncate. */ | |
282 | status = ocfs2_simple_size_update(inode, di_bh, new_i_size); | |
283 | if (status < 0) | |
284 | mlog_errno(status); | |
285 | goto bail; | |
286 | } | |
287 | ||
ccd979bd MF |
288 | /* alright, we're going to need to do a full blown alloc size |
289 | * change. Orphan the inode so that recovery can complete the | |
290 | * truncate if necessary. This does the task of marking | |
291 | * i_size. */ | |
292 | status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size); | |
293 | if (status < 0) { | |
294 | mlog_errno(status); | |
295 | goto bail; | |
296 | } | |
297 | ||
298 | status = ocfs2_prepare_truncate(osb, inode, di_bh, &tc); | |
299 | if (status < 0) { | |
300 | mlog_errno(status); | |
301 | goto bail; | |
302 | } | |
303 | ||
304 | status = ocfs2_commit_truncate(osb, inode, di_bh, tc); | |
305 | if (status < 0) { | |
306 | mlog_errno(status); | |
307 | goto bail; | |
308 | } | |
309 | ||
310 | /* TODO: orphan dir cleanup here. */ | |
311 | bail: | |
312 | ||
313 | mlog_exit(status); | |
314 | return status; | |
315 | } | |
316 | ||
317 | /* | |
318 | * extend allocation only here. | |
319 | * we'll update all the disk stuff, and oip->alloc_size | |
320 | * | |
321 | * expect stuff to be locked, a transaction started and enough data / | |
322 | * metadata reservations in the contexts. | |
323 | * | |
324 | * Will return -EAGAIN, and a reason if a restart is needed. | |
325 | * If passed in, *reason will always be set, even in error. | |
326 | */ | |
327 | int ocfs2_do_extend_allocation(struct ocfs2_super *osb, | |
328 | struct inode *inode, | |
329 | u32 clusters_to_add, | |
330 | struct buffer_head *fe_bh, | |
331 | struct ocfs2_journal_handle *handle, | |
332 | struct ocfs2_alloc_context *data_ac, | |
333 | struct ocfs2_alloc_context *meta_ac, | |
334 | enum ocfs2_alloc_restarted *reason_ret) | |
335 | { | |
336 | int status = 0; | |
337 | int free_extents; | |
338 | struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data; | |
339 | enum ocfs2_alloc_restarted reason = RESTART_NONE; | |
340 | u32 bit_off, num_bits; | |
341 | u64 block; | |
342 | ||
343 | BUG_ON(!clusters_to_add); | |
344 | ||
345 | free_extents = ocfs2_num_free_extents(osb, inode, fe); | |
346 | if (free_extents < 0) { | |
347 | status = free_extents; | |
348 | mlog_errno(status); | |
349 | goto leave; | |
350 | } | |
351 | ||
352 | /* there are two cases which could cause us to EAGAIN in the | |
353 | * we-need-more-metadata case: | |
354 | * 1) we haven't reserved *any* | |
355 | * 2) we are so fragmented, we've needed to add metadata too | |
356 | * many times. */ | |
357 | if (!free_extents && !meta_ac) { | |
358 | mlog(0, "we haven't reserved any metadata!\n"); | |
359 | status = -EAGAIN; | |
360 | reason = RESTART_META; | |
361 | goto leave; | |
362 | } else if ((!free_extents) | |
363 | && (ocfs2_alloc_context_bits_left(meta_ac) | |
364 | < ocfs2_extend_meta_needed(fe))) { | |
365 | mlog(0, "filesystem is really fragmented...\n"); | |
366 | status = -EAGAIN; | |
367 | reason = RESTART_META; | |
368 | goto leave; | |
369 | } | |
370 | ||
371 | status = ocfs2_claim_clusters(osb, handle, data_ac, 1, | |
372 | &bit_off, &num_bits); | |
373 | if (status < 0) { | |
374 | if (status != -ENOSPC) | |
375 | mlog_errno(status); | |
376 | goto leave; | |
377 | } | |
378 | ||
379 | BUG_ON(num_bits > clusters_to_add); | |
380 | ||
381 | /* reserve our write early -- insert_extent may update the inode */ | |
382 | status = ocfs2_journal_access(handle, inode, fe_bh, | |
383 | OCFS2_JOURNAL_ACCESS_WRITE); | |
384 | if (status < 0) { | |
385 | mlog_errno(status); | |
386 | goto leave; | |
387 | } | |
388 | ||
389 | block = ocfs2_clusters_to_blocks(osb->sb, bit_off); | |
b0697053 MF |
390 | mlog(0, "Allocating %u clusters at block %u for inode %llu\n", |
391 | num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno); | |
ccd979bd MF |
392 | status = ocfs2_insert_extent(osb, handle, inode, fe_bh, block, |
393 | num_bits, meta_ac); | |
394 | if (status < 0) { | |
395 | mlog_errno(status); | |
396 | goto leave; | |
397 | } | |
398 | ||
399 | le32_add_cpu(&fe->i_clusters, num_bits); | |
400 | spin_lock(&OCFS2_I(inode)->ip_lock); | |
401 | OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters); | |
402 | spin_unlock(&OCFS2_I(inode)->ip_lock); | |
403 | ||
404 | status = ocfs2_journal_dirty(handle, fe_bh); | |
405 | if (status < 0) { | |
406 | mlog_errno(status); | |
407 | goto leave; | |
408 | } | |
409 | ||
410 | clusters_to_add -= num_bits; | |
411 | ||
412 | if (clusters_to_add) { | |
413 | mlog(0, "need to alloc once more, clusters = %u, wanted = " | |
414 | "%u\n", fe->i_clusters, clusters_to_add); | |
415 | status = -EAGAIN; | |
416 | reason = RESTART_TRANS; | |
417 | } | |
418 | ||
419 | leave: | |
420 | mlog_exit(status); | |
421 | if (reason_ret) | |
422 | *reason_ret = reason; | |
423 | return status; | |
424 | } | |
425 | ||
426 | static int ocfs2_extend_allocation(struct inode *inode, | |
427 | u32 clusters_to_add) | |
428 | { | |
429 | int status = 0; | |
430 | int restart_func = 0; | |
431 | int drop_alloc_sem = 0; | |
432 | int credits, num_free_extents; | |
433 | u32 prev_clusters; | |
434 | struct buffer_head *bh = NULL; | |
435 | struct ocfs2_dinode *fe = NULL; | |
436 | struct ocfs2_journal_handle *handle = NULL; | |
437 | struct ocfs2_alloc_context *data_ac = NULL; | |
438 | struct ocfs2_alloc_context *meta_ac = NULL; | |
439 | enum ocfs2_alloc_restarted why; | |
440 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
441 | ||
442 | mlog_entry("(clusters_to_add = %u)\n", clusters_to_add); | |
443 | ||
444 | status = ocfs2_read_block(osb, OCFS2_I(inode)->ip_blkno, &bh, | |
445 | OCFS2_BH_CACHED, inode); | |
446 | if (status < 0) { | |
447 | mlog_errno(status); | |
448 | goto leave; | |
449 | } | |
450 | ||
451 | fe = (struct ocfs2_dinode *) bh->b_data; | |
452 | if (!OCFS2_IS_VALID_DINODE(fe)) { | |
453 | OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe); | |
454 | status = -EIO; | |
455 | goto leave; | |
456 | } | |
457 | ||
458 | restart_all: | |
459 | BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters); | |
460 | ||
b0697053 | 461 | mlog(0, "extend inode %llu, i_size = %lld, fe->i_clusters = %u, " |
ccd979bd | 462 | "clusters_to_add = %u\n", |
b0697053 | 463 | (unsigned long long)OCFS2_I(inode)->ip_blkno, i_size_read(inode), |
ccd979bd MF |
464 | fe->i_clusters, clusters_to_add); |
465 | ||
ccd979bd MF |
466 | num_free_extents = ocfs2_num_free_extents(osb, |
467 | inode, | |
468 | fe); | |
469 | if (num_free_extents < 0) { | |
470 | status = num_free_extents; | |
471 | mlog_errno(status); | |
472 | goto leave; | |
473 | } | |
474 | ||
475 | if (!num_free_extents) { | |
da5cbf2f | 476 | status = ocfs2_reserve_new_metadata(osb, fe, &meta_ac); |
ccd979bd MF |
477 | if (status < 0) { |
478 | if (status != -ENOSPC) | |
479 | mlog_errno(status); | |
480 | goto leave; | |
481 | } | |
482 | } | |
483 | ||
da5cbf2f | 484 | status = ocfs2_reserve_clusters(osb, clusters_to_add, &data_ac); |
ccd979bd MF |
485 | if (status < 0) { |
486 | if (status != -ENOSPC) | |
487 | mlog_errno(status); | |
488 | goto leave; | |
489 | } | |
490 | ||
491 | /* blocks peope in read/write from reading our allocation | |
1b1dcc1b | 492 | * until we're done changing it. We depend on i_mutex to block |
ccd979bd MF |
493 | * other extend/truncate calls while we're here. Ordering wrt |
494 | * start_trans is important here -- always do it before! */ | |
495 | down_write(&OCFS2_I(inode)->ip_alloc_sem); | |
496 | drop_alloc_sem = 1; | |
497 | ||
498 | credits = ocfs2_calc_extend_credits(osb->sb, fe, clusters_to_add); | |
da5cbf2f | 499 | handle = ocfs2_start_trans(osb, NULL, credits); |
ccd979bd MF |
500 | if (IS_ERR(handle)) { |
501 | status = PTR_ERR(handle); | |
502 | handle = NULL; | |
503 | mlog_errno(status); | |
504 | goto leave; | |
505 | } | |
506 | ||
507 | restarted_transaction: | |
508 | /* reserve a write to the file entry early on - that we if we | |
509 | * run out of credits in the allocation path, we can still | |
510 | * update i_size. */ | |
511 | status = ocfs2_journal_access(handle, inode, bh, | |
512 | OCFS2_JOURNAL_ACCESS_WRITE); | |
513 | if (status < 0) { | |
514 | mlog_errno(status); | |
515 | goto leave; | |
516 | } | |
517 | ||
518 | prev_clusters = OCFS2_I(inode)->ip_clusters; | |
519 | ||
520 | status = ocfs2_do_extend_allocation(osb, | |
521 | inode, | |
522 | clusters_to_add, | |
523 | bh, | |
524 | handle, | |
525 | data_ac, | |
526 | meta_ac, | |
527 | &why); | |
528 | if ((status < 0) && (status != -EAGAIN)) { | |
529 | if (status != -ENOSPC) | |
530 | mlog_errno(status); | |
531 | goto leave; | |
532 | } | |
533 | ||
534 | status = ocfs2_journal_dirty(handle, bh); | |
535 | if (status < 0) { | |
536 | mlog_errno(status); | |
537 | goto leave; | |
538 | } | |
539 | ||
540 | spin_lock(&OCFS2_I(inode)->ip_lock); | |
541 | clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters); | |
542 | spin_unlock(&OCFS2_I(inode)->ip_lock); | |
543 | ||
544 | if (why != RESTART_NONE && clusters_to_add) { | |
545 | if (why == RESTART_META) { | |
546 | mlog(0, "restarting function.\n"); | |
547 | restart_func = 1; | |
548 | } else { | |
549 | BUG_ON(why != RESTART_TRANS); | |
550 | ||
551 | mlog(0, "restarting transaction.\n"); | |
552 | /* TODO: This can be more intelligent. */ | |
553 | credits = ocfs2_calc_extend_credits(osb->sb, | |
554 | fe, | |
555 | clusters_to_add); | |
1fc58146 | 556 | status = ocfs2_extend_trans(handle->k_handle, credits); |
ccd979bd MF |
557 | if (status < 0) { |
558 | /* handle still has to be committed at | |
559 | * this point. */ | |
560 | status = -ENOMEM; | |
561 | mlog_errno(status); | |
562 | goto leave; | |
563 | } | |
564 | goto restarted_transaction; | |
565 | } | |
566 | } | |
567 | ||
b0697053 MF |
568 | mlog(0, "fe: i_clusters = %u, i_size=%llu\n", |
569 | fe->i_clusters, (unsigned long long)fe->i_size); | |
ccd979bd MF |
570 | mlog(0, "inode: ip_clusters=%u, i_size=%lld\n", |
571 | OCFS2_I(inode)->ip_clusters, i_size_read(inode)); | |
572 | ||
573 | leave: | |
574 | if (drop_alloc_sem) { | |
575 | up_write(&OCFS2_I(inode)->ip_alloc_sem); | |
576 | drop_alloc_sem = 0; | |
577 | } | |
578 | if (handle) { | |
579 | ocfs2_commit_trans(handle); | |
580 | handle = NULL; | |
581 | } | |
582 | if (data_ac) { | |
583 | ocfs2_free_alloc_context(data_ac); | |
584 | data_ac = NULL; | |
585 | } | |
586 | if (meta_ac) { | |
587 | ocfs2_free_alloc_context(meta_ac); | |
588 | meta_ac = NULL; | |
589 | } | |
590 | if ((!status) && restart_func) { | |
591 | restart_func = 0; | |
592 | goto restart_all; | |
593 | } | |
594 | if (bh) { | |
595 | brelse(bh); | |
596 | bh = NULL; | |
597 | } | |
598 | ||
599 | mlog_exit(status); | |
600 | return status; | |
601 | } | |
602 | ||
603 | /* Some parts of this taken from generic_cont_expand, which turned out | |
604 | * to be too fragile to do exactly what we need without us having to | |
53013cba MF |
605 | * worry about recursive locking in ->prepare_write() and |
606 | * ->commit_write(). */ | |
ccd979bd MF |
607 | static int ocfs2_write_zero_page(struct inode *inode, |
608 | u64 size) | |
609 | { | |
610 | struct address_space *mapping = inode->i_mapping; | |
611 | struct page *page; | |
612 | unsigned long index; | |
613 | unsigned int offset; | |
614 | struct ocfs2_journal_handle *handle = NULL; | |
615 | int ret; | |
616 | ||
617 | offset = (size & (PAGE_CACHE_SIZE-1)); /* Within page */ | |
618 | /* ugh. in prepare/commit_write, if from==to==start of block, we | |
619 | ** skip the prepare. make sure we never send an offset for the start | |
620 | ** of a block | |
621 | */ | |
622 | if ((offset & (inode->i_sb->s_blocksize - 1)) == 0) { | |
623 | offset++; | |
624 | } | |
625 | index = size >> PAGE_CACHE_SHIFT; | |
626 | ||
627 | page = grab_cache_page(mapping, index); | |
628 | if (!page) { | |
629 | ret = -ENOMEM; | |
630 | mlog_errno(ret); | |
631 | goto out; | |
632 | } | |
633 | ||
53013cba | 634 | ret = ocfs2_prepare_write_nolock(inode, page, offset, offset); |
ccd979bd MF |
635 | if (ret < 0) { |
636 | mlog_errno(ret); | |
637 | goto out_unlock; | |
638 | } | |
639 | ||
640 | if (ocfs2_should_order_data(inode)) { | |
641 | handle = ocfs2_start_walk_page_trans(inode, page, offset, | |
642 | offset); | |
643 | if (IS_ERR(handle)) { | |
644 | ret = PTR_ERR(handle); | |
645 | handle = NULL; | |
646 | goto out_unlock; | |
647 | } | |
648 | } | |
649 | ||
650 | /* must not update i_size! */ | |
651 | ret = block_commit_write(page, offset, offset); | |
652 | if (ret < 0) | |
653 | mlog_errno(ret); | |
654 | else | |
655 | ret = 0; | |
656 | ||
657 | if (handle) | |
658 | ocfs2_commit_trans(handle); | |
659 | out_unlock: | |
660 | unlock_page(page); | |
661 | page_cache_release(page); | |
662 | out: | |
663 | return ret; | |
664 | } | |
665 | ||
666 | static int ocfs2_zero_extend(struct inode *inode, | |
667 | u64 zero_to_size) | |
668 | { | |
669 | int ret = 0; | |
670 | u64 start_off; | |
671 | struct super_block *sb = inode->i_sb; | |
672 | ||
673 | start_off = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode)); | |
674 | while (start_off < zero_to_size) { | |
675 | ret = ocfs2_write_zero_page(inode, start_off); | |
676 | if (ret < 0) { | |
677 | mlog_errno(ret); | |
678 | goto out; | |
679 | } | |
680 | ||
681 | start_off += sb->s_blocksize; | |
e2057c5a MF |
682 | |
683 | /* | |
684 | * Very large extends have the potential to lock up | |
685 | * the cpu for extended periods of time. | |
686 | */ | |
687 | cond_resched(); | |
ccd979bd MF |
688 | } |
689 | ||
690 | out: | |
691 | return ret; | |
692 | } | |
693 | ||
53013cba MF |
694 | /* |
695 | * A tail_to_skip value > 0 indicates that we're being called from | |
696 | * ocfs2_file_aio_write(). This has the following implications: | |
697 | * | |
698 | * - we don't want to update i_size | |
699 | * - di_bh will be NULL, which is fine because it's only used in the | |
700 | * case where we want to update i_size. | |
701 | * - ocfs2_zero_extend() will then only be filling the hole created | |
702 | * between i_size and the start of the write. | |
703 | */ | |
ccd979bd MF |
704 | static int ocfs2_extend_file(struct inode *inode, |
705 | struct buffer_head *di_bh, | |
53013cba MF |
706 | u64 new_i_size, |
707 | size_t tail_to_skip) | |
ccd979bd MF |
708 | { |
709 | int ret = 0; | |
710 | u32 clusters_to_add; | |
711 | ||
53013cba MF |
712 | BUG_ON(!tail_to_skip && !di_bh); |
713 | ||
ccd979bd MF |
714 | /* setattr sometimes calls us like this. */ |
715 | if (new_i_size == 0) | |
716 | goto out; | |
717 | ||
718 | if (i_size_read(inode) == new_i_size) | |
719 | goto out; | |
720 | BUG_ON(new_i_size < i_size_read(inode)); | |
721 | ||
722 | clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size) - | |
723 | OCFS2_I(inode)->ip_clusters; | |
724 | ||
0effef77 MF |
725 | /* |
726 | * protect the pages that ocfs2_zero_extend is going to be | |
727 | * pulling into the page cache.. we do this before the | |
728 | * metadata extend so that we don't get into the situation | |
729 | * where we've extended the metadata but can't get the data | |
730 | * lock to zero. | |
731 | */ | |
732 | ret = ocfs2_data_lock(inode, 1); | |
733 | if (ret < 0) { | |
734 | mlog_errno(ret); | |
735 | goto out; | |
736 | } | |
ccd979bd | 737 | |
0effef77 | 738 | if (clusters_to_add) { |
53013cba | 739 | ret = ocfs2_extend_allocation(inode, clusters_to_add); |
ccd979bd MF |
740 | if (ret < 0) { |
741 | mlog_errno(ret); | |
53013cba | 742 | goto out_unlock; |
ccd979bd | 743 | } |
0effef77 | 744 | } |
ccd979bd | 745 | |
0effef77 MF |
746 | /* |
747 | * Call this even if we don't add any clusters to the tree. We | |
748 | * still need to zero the area between the old i_size and the | |
749 | * new i_size. | |
750 | */ | |
751 | ret = ocfs2_zero_extend(inode, (u64)new_i_size - tail_to_skip); | |
752 | if (ret < 0) { | |
753 | mlog_errno(ret); | |
754 | goto out_unlock; | |
53013cba MF |
755 | } |
756 | ||
757 | if (!tail_to_skip) { | |
758 | /* We're being called from ocfs2_setattr() which wants | |
759 | * us to update i_size */ | |
760 | ret = ocfs2_simple_size_update(inode, di_bh, new_i_size); | |
761 | if (ret < 0) | |
762 | mlog_errno(ret); | |
ccd979bd MF |
763 | } |
764 | ||
53013cba | 765 | out_unlock: |
0effef77 | 766 | ocfs2_data_unlock(inode, 1); |
53013cba | 767 | |
ccd979bd MF |
768 | out: |
769 | return ret; | |
770 | } | |
771 | ||
772 | int ocfs2_setattr(struct dentry *dentry, struct iattr *attr) | |
773 | { | |
774 | int status = 0, size_change; | |
775 | struct inode *inode = dentry->d_inode; | |
776 | struct super_block *sb = inode->i_sb; | |
777 | struct ocfs2_super *osb = OCFS2_SB(sb); | |
778 | struct buffer_head *bh = NULL; | |
779 | struct ocfs2_journal_handle *handle = NULL; | |
780 | ||
781 | mlog_entry("(0x%p, '%.*s')\n", dentry, | |
782 | dentry->d_name.len, dentry->d_name.name); | |
783 | ||
784 | if (attr->ia_valid & ATTR_MODE) | |
785 | mlog(0, "mode change: %d\n", attr->ia_mode); | |
786 | if (attr->ia_valid & ATTR_UID) | |
787 | mlog(0, "uid change: %d\n", attr->ia_uid); | |
788 | if (attr->ia_valid & ATTR_GID) | |
789 | mlog(0, "gid change: %d\n", attr->ia_gid); | |
790 | if (attr->ia_valid & ATTR_SIZE) | |
791 | mlog(0, "size change...\n"); | |
792 | if (attr->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME)) | |
793 | mlog(0, "time change...\n"); | |
794 | ||
795 | #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \ | |
796 | | ATTR_GID | ATTR_UID | ATTR_MODE) | |
797 | if (!(attr->ia_valid & OCFS2_VALID_ATTRS)) { | |
798 | mlog(0, "can't handle attrs: 0x%x\n", attr->ia_valid); | |
799 | return 0; | |
800 | } | |
801 | ||
802 | status = inode_change_ok(inode, attr); | |
803 | if (status) | |
804 | return status; | |
805 | ||
806 | size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE; | |
807 | if (size_change) { | |
808 | status = ocfs2_rw_lock(inode, 1); | |
809 | if (status < 0) { | |
810 | mlog_errno(status); | |
811 | goto bail; | |
812 | } | |
813 | } | |
814 | ||
4bcec184 | 815 | status = ocfs2_meta_lock(inode, &bh, 1); |
ccd979bd MF |
816 | if (status < 0) { |
817 | if (status != -ENOENT) | |
818 | mlog_errno(status); | |
819 | goto bail_unlock_rw; | |
820 | } | |
821 | ||
822 | if (size_change && attr->ia_size != i_size_read(inode)) { | |
823 | if (i_size_read(inode) > attr->ia_size) | |
824 | status = ocfs2_truncate_file(inode, bh, attr->ia_size); | |
825 | else | |
53013cba | 826 | status = ocfs2_extend_file(inode, bh, attr->ia_size, 0); |
ccd979bd MF |
827 | if (status < 0) { |
828 | if (status != -ENOSPC) | |
829 | mlog_errno(status); | |
830 | status = -ENOSPC; | |
831 | goto bail_unlock; | |
832 | } | |
833 | } | |
834 | ||
835 | handle = ocfs2_start_trans(osb, NULL, OCFS2_INODE_UPDATE_CREDITS); | |
836 | if (IS_ERR(handle)) { | |
837 | status = PTR_ERR(handle); | |
838 | mlog_errno(status); | |
839 | goto bail_unlock; | |
840 | } | |
841 | ||
842 | status = inode_setattr(inode, attr); | |
843 | if (status < 0) { | |
844 | mlog_errno(status); | |
845 | goto bail_commit; | |
846 | } | |
847 | ||
848 | status = ocfs2_mark_inode_dirty(handle, inode, bh); | |
849 | if (status < 0) | |
850 | mlog_errno(status); | |
851 | ||
852 | bail_commit: | |
853 | ocfs2_commit_trans(handle); | |
854 | bail_unlock: | |
855 | ocfs2_meta_unlock(inode, 1); | |
856 | bail_unlock_rw: | |
857 | if (size_change) | |
858 | ocfs2_rw_unlock(inode, 1); | |
859 | bail: | |
860 | if (bh) | |
861 | brelse(bh); | |
862 | ||
863 | mlog_exit(status); | |
864 | return status; | |
865 | } | |
866 | ||
867 | int ocfs2_getattr(struct vfsmount *mnt, | |
868 | struct dentry *dentry, | |
869 | struct kstat *stat) | |
870 | { | |
871 | struct inode *inode = dentry->d_inode; | |
872 | struct super_block *sb = dentry->d_inode->i_sb; | |
873 | struct ocfs2_super *osb = sb->s_fs_info; | |
874 | int err; | |
875 | ||
876 | mlog_entry_void(); | |
877 | ||
878 | err = ocfs2_inode_revalidate(dentry); | |
879 | if (err) { | |
880 | if (err != -ENOENT) | |
881 | mlog_errno(err); | |
882 | goto bail; | |
883 | } | |
884 | ||
885 | generic_fillattr(inode, stat); | |
886 | ||
887 | /* We set the blksize from the cluster size for performance */ | |
888 | stat->blksize = osb->s_clustersize; | |
889 | ||
890 | bail: | |
891 | mlog_exit(err); | |
892 | ||
893 | return err; | |
894 | } | |
895 | ||
896 | static int ocfs2_write_remove_suid(struct inode *inode) | |
897 | { | |
898 | int ret; | |
899 | struct buffer_head *bh = NULL; | |
900 | struct ocfs2_inode_info *oi = OCFS2_I(inode); | |
901 | struct ocfs2_journal_handle *handle; | |
902 | struct ocfs2_super *osb = OCFS2_SB(inode->i_sb); | |
903 | struct ocfs2_dinode *di; | |
904 | ||
b0697053 MF |
905 | mlog_entry("(Inode %llu, mode 0%o)\n", |
906 | (unsigned long long)oi->ip_blkno, inode->i_mode); | |
ccd979bd MF |
907 | |
908 | handle = ocfs2_start_trans(osb, NULL, OCFS2_INODE_UPDATE_CREDITS); | |
909 | if (handle == NULL) { | |
910 | ret = -ENOMEM; | |
911 | mlog_errno(ret); | |
912 | goto out; | |
913 | } | |
914 | ||
915 | ret = ocfs2_read_block(osb, oi->ip_blkno, &bh, OCFS2_BH_CACHED, inode); | |
916 | if (ret < 0) { | |
917 | mlog_errno(ret); | |
918 | goto out_trans; | |
919 | } | |
920 | ||
921 | ret = ocfs2_journal_access(handle, inode, bh, | |
922 | OCFS2_JOURNAL_ACCESS_WRITE); | |
923 | if (ret < 0) { | |
924 | mlog_errno(ret); | |
925 | goto out_bh; | |
926 | } | |
927 | ||
928 | inode->i_mode &= ~S_ISUID; | |
929 | if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP)) | |
930 | inode->i_mode &= ~S_ISGID; | |
931 | ||
932 | di = (struct ocfs2_dinode *) bh->b_data; | |
933 | di->i_mode = cpu_to_le16(inode->i_mode); | |
934 | ||
935 | ret = ocfs2_journal_dirty(handle, bh); | |
936 | if (ret < 0) | |
937 | mlog_errno(ret); | |
938 | out_bh: | |
939 | brelse(bh); | |
940 | out_trans: | |
941 | ocfs2_commit_trans(handle); | |
942 | out: | |
943 | mlog_exit(ret); | |
944 | return ret; | |
945 | } | |
946 | ||
947 | static inline int ocfs2_write_should_remove_suid(struct inode *inode) | |
948 | { | |
949 | mode_t mode = inode->i_mode; | |
950 | ||
951 | if (!capable(CAP_FSETID)) { | |
952 | if (unlikely(mode & S_ISUID)) | |
953 | return 1; | |
954 | ||
955 | if (unlikely((mode & S_ISGID) && (mode & S_IXGRP))) | |
956 | return 1; | |
957 | } | |
958 | return 0; | |
959 | } | |
960 | ||
961 | static ssize_t ocfs2_file_aio_write(struct kiocb *iocb, | |
027445c3 BP |
962 | const struct iovec *iov, |
963 | unsigned long nr_segs, | |
ccd979bd MF |
964 | loff_t pos) |
965 | { | |
ccd979bd MF |
966 | int ret, rw_level = -1, meta_level = -1, have_alloc_sem = 0; |
967 | u32 clusters; | |
968 | struct file *filp = iocb->ki_filp; | |
969 | struct inode *inode = filp->f_dentry->d_inode; | |
970 | loff_t newsize, saved_pos; | |
ccd979bd | 971 | |
027445c3 BP |
972 | mlog_entry("(0x%p, %u, '%.*s')\n", filp, |
973 | (unsigned int)nr_segs, | |
ccd979bd MF |
974 | filp->f_dentry->d_name.len, |
975 | filp->f_dentry->d_name.name); | |
976 | ||
977 | /* happy write of zero bytes */ | |
027445c3 | 978 | if (iocb->ki_left == 0) |
ccd979bd MF |
979 | return 0; |
980 | ||
981 | if (!inode) { | |
982 | mlog(0, "bad inode\n"); | |
983 | return -EIO; | |
984 | } | |
985 | ||
1b1dcc1b JS |
986 | mutex_lock(&inode->i_mutex); |
987 | /* to match setattr's i_mutex -> i_alloc_sem -> rw_lock ordering */ | |
ccd979bd MF |
988 | if (filp->f_flags & O_DIRECT) { |
989 | have_alloc_sem = 1; | |
990 | down_read(&inode->i_alloc_sem); | |
991 | } | |
992 | ||
993 | /* concurrent O_DIRECT writes are allowed */ | |
994 | rw_level = (filp->f_flags & O_DIRECT) ? 0 : 1; | |
995 | ret = ocfs2_rw_lock(inode, rw_level); | |
996 | if (ret < 0) { | |
997 | rw_level = -1; | |
998 | mlog_errno(ret); | |
999 | goto out; | |
1000 | } | |
1001 | ||
1002 | /* | |
1003 | * We sample i_size under a read level meta lock to see if our write | |
1004 | * is extending the file, if it is we back off and get a write level | |
1005 | * meta lock. | |
1006 | */ | |
1007 | meta_level = (filp->f_flags & O_APPEND) ? 1 : 0; | |
1008 | for(;;) { | |
4bcec184 | 1009 | ret = ocfs2_meta_lock(inode, NULL, meta_level); |
ccd979bd MF |
1010 | if (ret < 0) { |
1011 | meta_level = -1; | |
1012 | mlog_errno(ret); | |
1013 | goto out; | |
1014 | } | |
1015 | ||
1016 | /* Clear suid / sgid if necessary. We do this here | |
1017 | * instead of later in the write path because | |
1018 | * remove_suid() calls ->setattr without any hint that | |
1019 | * we may have already done our cluster locking. Since | |
1020 | * ocfs2_setattr() *must* take cluster locks to | |
1021 | * proceeed, this will lead us to recursively lock the | |
1022 | * inode. There's also the dinode i_size state which | |
1023 | * can be lost via setattr during extending writes (we | |
1024 | * set inode->i_size at the end of a write. */ | |
1025 | if (ocfs2_write_should_remove_suid(inode)) { | |
1026 | if (meta_level == 0) { | |
1027 | ocfs2_meta_unlock(inode, meta_level); | |
1028 | meta_level = 1; | |
1029 | continue; | |
1030 | } | |
1031 | ||
1032 | ret = ocfs2_write_remove_suid(inode); | |
1033 | if (ret < 0) { | |
1034 | mlog_errno(ret); | |
1035 | goto out; | |
1036 | } | |
1037 | } | |
1038 | ||
1039 | /* work on a copy of ppos until we're sure that we won't have | |
1040 | * to recalculate it due to relocking. */ | |
1041 | if (filp->f_flags & O_APPEND) { | |
1042 | saved_pos = i_size_read(inode); | |
1043 | mlog(0, "O_APPEND: inode->i_size=%llu\n", saved_pos); | |
1044 | } else { | |
1045 | saved_pos = iocb->ki_pos; | |
1046 | } | |
027445c3 | 1047 | newsize = iocb->ki_left + saved_pos; |
ccd979bd | 1048 | |
215c7f9f MF |
1049 | mlog(0, "pos=%lld newsize=%lld cursize=%lld\n", |
1050 | (long long) saved_pos, (long long) newsize, | |
1051 | (long long) i_size_read(inode)); | |
ccd979bd MF |
1052 | |
1053 | /* No need for a higher level metadata lock if we're | |
1054 | * never going past i_size. */ | |
1055 | if (newsize <= i_size_read(inode)) | |
1056 | break; | |
1057 | ||
1058 | if (meta_level == 0) { | |
1059 | ocfs2_meta_unlock(inode, meta_level); | |
1060 | meta_level = 1; | |
1061 | continue; | |
1062 | } | |
1063 | ||
1064 | spin_lock(&OCFS2_I(inode)->ip_lock); | |
1065 | clusters = ocfs2_clusters_for_bytes(inode->i_sb, newsize) - | |
1066 | OCFS2_I(inode)->ip_clusters; | |
1067 | spin_unlock(&OCFS2_I(inode)->ip_lock); | |
1068 | ||
1069 | mlog(0, "Writing at EOF, may need more allocation: " | |
215c7f9f MF |
1070 | "i_size = %lld, newsize = %lld, need %u clusters\n", |
1071 | (long long) i_size_read(inode), (long long) newsize, | |
1072 | clusters); | |
ccd979bd MF |
1073 | |
1074 | /* We only want to continue the rest of this loop if | |
1075 | * our extend will actually require more | |
1076 | * allocation. */ | |
1077 | if (!clusters) | |
1078 | break; | |
1079 | ||
027445c3 | 1080 | ret = ocfs2_extend_file(inode, NULL, newsize, iocb->ki_left); |
ccd979bd MF |
1081 | if (ret < 0) { |
1082 | if (ret != -ENOSPC) | |
1083 | mlog_errno(ret); | |
1084 | goto out; | |
1085 | } | |
ccd979bd MF |
1086 | break; |
1087 | } | |
1088 | ||
1089 | /* ok, we're done with i_size and alloc work */ | |
1090 | iocb->ki_pos = saved_pos; | |
1091 | ocfs2_meta_unlock(inode, meta_level); | |
1092 | meta_level = -1; | |
1093 | ||
1094 | /* communicate with ocfs2_dio_end_io */ | |
1095 | ocfs2_iocb_set_rw_locked(iocb); | |
1096 | ||
027445c3 | 1097 | ret = generic_file_aio_write_nolock(iocb, iov, nr_segs, iocb->ki_pos); |
ccd979bd MF |
1098 | |
1099 | /* buffered aio wouldn't have proper lock coverage today */ | |
1100 | BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT)); | |
1101 | ||
1102 | /* | |
1103 | * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io | |
1104 | * function pointer which is called when o_direct io completes so that | |
1105 | * it can unlock our rw lock. (it's the clustered equivalent of | |
1106 | * i_alloc_sem; protects truncate from racing with pending ios). | |
1107 | * Unfortunately there are error cases which call end_io and others | |
1108 | * that don't. so we don't have to unlock the rw_lock if either an | |
1109 | * async dio is going to do it in the future or an end_io after an | |
1110 | * error has already done it. | |
1111 | */ | |
1112 | if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) { | |
1113 | rw_level = -1; | |
1114 | have_alloc_sem = 0; | |
1115 | } | |
1116 | ||
1117 | out: | |
1118 | if (meta_level != -1) | |
1119 | ocfs2_meta_unlock(inode, meta_level); | |
1120 | if (have_alloc_sem) | |
1121 | up_read(&inode->i_alloc_sem); | |
1122 | if (rw_level != -1) | |
1123 | ocfs2_rw_unlock(inode, rw_level); | |
1b1dcc1b | 1124 | mutex_unlock(&inode->i_mutex); |
ccd979bd MF |
1125 | |
1126 | mlog_exit(ret); | |
1127 | return ret; | |
1128 | } | |
1129 | ||
1130 | static ssize_t ocfs2_file_aio_read(struct kiocb *iocb, | |
027445c3 BP |
1131 | const struct iovec *iov, |
1132 | unsigned long nr_segs, | |
ccd979bd MF |
1133 | loff_t pos) |
1134 | { | |
1135 | int ret = 0, rw_level = -1, have_alloc_sem = 0; | |
1136 | struct file *filp = iocb->ki_filp; | |
1137 | struct inode *inode = filp->f_dentry->d_inode; | |
ccd979bd | 1138 | |
027445c3 BP |
1139 | mlog_entry("(0x%p, %u, '%.*s')\n", filp, |
1140 | (unsigned int)nr_segs, | |
ccd979bd MF |
1141 | filp->f_dentry->d_name.len, |
1142 | filp->f_dentry->d_name.name); | |
1143 | ||
1144 | if (!inode) { | |
1145 | ret = -EINVAL; | |
1146 | mlog_errno(ret); | |
1147 | goto bail; | |
1148 | } | |
1149 | ||
ccd979bd MF |
1150 | /* |
1151 | * buffered reads protect themselves in ->readpage(). O_DIRECT reads | |
1152 | * need locks to protect pending reads from racing with truncate. | |
1153 | */ | |
1154 | if (filp->f_flags & O_DIRECT) { | |
1155 | down_read(&inode->i_alloc_sem); | |
1156 | have_alloc_sem = 1; | |
1157 | ||
1158 | ret = ocfs2_rw_lock(inode, 0); | |
1159 | if (ret < 0) { | |
1160 | mlog_errno(ret); | |
1161 | goto bail; | |
1162 | } | |
1163 | rw_level = 0; | |
1164 | /* communicate with ocfs2_dio_end_io */ | |
1165 | ocfs2_iocb_set_rw_locked(iocb); | |
1166 | } | |
1167 | ||
c4374f8a MF |
1168 | /* |
1169 | * We're fine letting folks race truncates and extending | |
1170 | * writes with read across the cluster, just like they can | |
1171 | * locally. Hence no rw_lock during read. | |
1172 | * | |
1173 | * Take and drop the meta data lock to update inode fields | |
1174 | * like i_size. This allows the checks down below | |
1175 | * generic_file_aio_read() a chance of actually working. | |
1176 | */ | |
4bcec184 | 1177 | ret = ocfs2_meta_lock(inode, NULL, 0); |
c4374f8a MF |
1178 | if (ret < 0) { |
1179 | mlog_errno(ret); | |
1180 | goto bail; | |
1181 | } | |
1182 | ocfs2_meta_unlock(inode, 0); | |
1183 | ||
027445c3 | 1184 | ret = generic_file_aio_read(iocb, iov, nr_segs, iocb->ki_pos); |
ccd979bd MF |
1185 | if (ret == -EINVAL) |
1186 | mlog(ML_ERROR, "generic_file_aio_read returned -EINVAL\n"); | |
1187 | ||
1188 | /* buffered aio wouldn't have proper lock coverage today */ | |
1189 | BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT)); | |
1190 | ||
1191 | /* see ocfs2_file_aio_write */ | |
1192 | if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) { | |
1193 | rw_level = -1; | |
1194 | have_alloc_sem = 0; | |
1195 | } | |
1196 | ||
1197 | bail: | |
1198 | if (have_alloc_sem) | |
1199 | up_read(&inode->i_alloc_sem); | |
1200 | if (rw_level != -1) | |
1201 | ocfs2_rw_unlock(inode, rw_level); | |
1202 | mlog_exit(ret); | |
1203 | ||
1204 | return ret; | |
1205 | } | |
1206 | ||
1207 | struct inode_operations ocfs2_file_iops = { | |
1208 | .setattr = ocfs2_setattr, | |
1209 | .getattr = ocfs2_getattr, | |
1210 | }; | |
1211 | ||
1212 | struct inode_operations ocfs2_special_file_iops = { | |
1213 | .setattr = ocfs2_setattr, | |
1214 | .getattr = ocfs2_getattr, | |
1215 | }; | |
1216 | ||
4b6f5d20 | 1217 | const struct file_operations ocfs2_fops = { |
ccd979bd MF |
1218 | .read = do_sync_read, |
1219 | .write = do_sync_write, | |
1220 | .sendfile = generic_file_sendfile, | |
1221 | .mmap = ocfs2_mmap, | |
1222 | .fsync = ocfs2_sync_file, | |
1223 | .release = ocfs2_file_release, | |
1224 | .open = ocfs2_file_open, | |
1225 | .aio_read = ocfs2_file_aio_read, | |
1226 | .aio_write = ocfs2_file_aio_write, | |
ca4d147e | 1227 | .ioctl = ocfs2_ioctl, |
ccd979bd MF |
1228 | }; |
1229 | ||
4b6f5d20 | 1230 | const struct file_operations ocfs2_dops = { |
ccd979bd MF |
1231 | .read = generic_read_dir, |
1232 | .readdir = ocfs2_readdir, | |
1233 | .fsync = ocfs2_sync_file, | |
ca4d147e | 1234 | .ioctl = ocfs2_ioctl, |
ccd979bd | 1235 | }; |