blob: 8c97fa1c45f609107692daf65de89ed7772bbaf0 [file] [log] [blame]
Mark Fashehccd979b2005-12-15 14:31:24 -08001/* -*- 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
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080026#include <linux/capability.h>
Mark Fashehccd979b2005-12-15 14:31:24 -080027#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>
Mark Fashehe2057c52006-10-03 17:53:05 -070033#include <linux/sched.h>
Tiger Yang8659ac22006-10-17 18:29:52 -070034#include <linux/pipe_fs_i.h>
Tiger Yang7f1a37e2006-11-15 15:48:42 +080035#include <linux/mount.h>
Mark Fashehccd979b2005-12-15 14:31:24 -080036
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"
Herbert Poetzlca4d1472006-07-03 17:27:12 -070050#include "ioctl.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080051#include "journal.h"
52#include "mmap.h"
53#include "suballoc.h"
54#include "super.h"
55
56#include "buffer_head_io.h"
57
58static int ocfs2_sync_inode(struct inode *inode)
59{
60 filemap_fdatawrite(inode->i_mapping);
61 return sync_mapping_buffers(inode->i_mapping);
62}
63
64static 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,
Josef Sipekd28c9172006-12-08 02:37:25 -080071 file->f_path.dentry->d_name.len, file->f_path.dentry->d_name.name);
Mark Fashehccd979b2005-12-15 14:31:24 -080072
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;
91leave:
92 mlog_exit(status);
93 return status;
94}
95
96static 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,
Josef Sipekd28c9172006-12-08 02:37:25 -0800101 file->f_path.dentry->d_name.len,
102 file->f_path.dentry->d_name.name);
Mark Fashehccd979b2005-12-15 14:31:24 -0800103
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
114static 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
133bail:
134 mlog_exit(err);
135
136 return (err < 0) ? -EIO : 0;
137}
138
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800139int 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
Mark Fasheh6c2aad02006-12-19 15:25:52 -0800152 /*
153 * We can be called with no vfsmnt structure - NFSD will
154 * sometimes do this.
155 *
156 * Note that our action here is different than touch_atime() -
157 * if we can't tell whether this is a noatime mount, then we
158 * don't know whether to trust the value of s_atime_quantum.
159 */
160 if (vfsmnt == NULL)
161 return 0;
162
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800163 if ((vfsmnt->mnt_flags & MNT_NOATIME) ||
164 ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
165 return 0;
166
Mark Fasheh7e913c52006-12-13 00:34:35 -0800167 if (vfsmnt->mnt_flags & MNT_RELATIME) {
168 if ((timespec_compare(&inode->i_atime, &inode->i_mtime) <= 0) ||
169 (timespec_compare(&inode->i_atime, &inode->i_ctime) <= 0))
170 return 1;
171
172 return 0;
173 }
174
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800175 now = CURRENT_TIME;
176 if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum))
177 return 0;
178 else
179 return 1;
180}
181
182int ocfs2_update_inode_atime(struct inode *inode,
183 struct buffer_head *bh)
184{
185 int ret;
186 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
187 handle_t *handle;
188
189 mlog_entry_void();
190
191 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
192 if (handle == NULL) {
193 ret = -ENOMEM;
194 mlog_errno(ret);
195 goto out;
196 }
197
198 inode->i_atime = CURRENT_TIME;
199 ret = ocfs2_mark_inode_dirty(handle, inode, bh);
200 if (ret < 0)
201 mlog_errno(ret);
202
203 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
204out:
205 mlog_exit(ret);
206 return ret;
207}
208
Mark Fasheh1fabe142006-10-09 18:11:45 -0700209int ocfs2_set_inode_size(handle_t *handle,
Mark Fashehccd979b2005-12-15 14:31:24 -0800210 struct inode *inode,
211 struct buffer_head *fe_bh,
212 u64 new_i_size)
213{
214 int status;
215
216 mlog_entry_void();
217 i_size_write(inode, new_i_size);
218 inode->i_blocks = ocfs2_align_bytes_to_sectors(new_i_size);
219 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
220
221 status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
222 if (status < 0) {
223 mlog_errno(status);
224 goto bail;
225 }
226
227bail:
228 mlog_exit(status);
229 return status;
230}
231
232static int ocfs2_simple_size_update(struct inode *inode,
233 struct buffer_head *di_bh,
234 u64 new_i_size)
235{
236 int ret;
237 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
Mark Fasheh1fabe142006-10-09 18:11:45 -0700238 handle_t *handle = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800239
Mark Fasheh65eff9c2006-10-09 17:26:22 -0700240 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
Mark Fashehccd979b2005-12-15 14:31:24 -0800241 if (handle == NULL) {
242 ret = -ENOMEM;
243 mlog_errno(ret);
244 goto out;
245 }
246
247 ret = ocfs2_set_inode_size(handle, inode, di_bh,
248 new_i_size);
249 if (ret < 0)
250 mlog_errno(ret);
251
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700252 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800253out:
254 return ret;
255}
256
257static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
258 struct inode *inode,
259 struct buffer_head *fe_bh,
260 u64 new_i_size)
261{
262 int status;
Mark Fasheh1fabe142006-10-09 18:11:45 -0700263 handle_t *handle;
Mark Fashehccd979b2005-12-15 14:31:24 -0800264
265 mlog_entry_void();
266
267 /* TODO: This needs to actually orphan the inode in this
268 * transaction. */
269
Mark Fasheh65eff9c2006-10-09 17:26:22 -0700270 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
Mark Fashehccd979b2005-12-15 14:31:24 -0800271 if (IS_ERR(handle)) {
272 status = PTR_ERR(handle);
273 mlog_errno(status);
274 goto out;
275 }
276
277 status = ocfs2_set_inode_size(handle, inode, fe_bh, new_i_size);
278 if (status < 0)
279 mlog_errno(status);
280
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700281 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800282out:
283 mlog_exit(status);
284 return status;
285}
286
287static int ocfs2_truncate_file(struct inode *inode,
288 struct buffer_head *di_bh,
289 u64 new_i_size)
290{
291 int status = 0;
292 struct ocfs2_dinode *fe = NULL;
293 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
294 struct ocfs2_truncate_context *tc = NULL;
295
Mark Fashehb06970532006-03-03 10:24:33 -0800296 mlog_entry("(inode = %llu, new_i_size = %llu\n",
297 (unsigned long long)OCFS2_I(inode)->ip_blkno,
298 (unsigned long long)new_i_size);
Mark Fashehccd979b2005-12-15 14:31:24 -0800299
300 truncate_inode_pages(inode->i_mapping, new_i_size);
301
302 fe = (struct ocfs2_dinode *) di_bh->b_data;
303 if (!OCFS2_IS_VALID_DINODE(fe)) {
304 OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe);
305 status = -EIO;
306 goto bail;
307 }
308
309 mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode),
Mark Fashehb06970532006-03-03 10:24:33 -0800310 "Inode %llu, inode i_size = %lld != di "
311 "i_size = %llu, i_flags = 0x%x\n",
312 (unsigned long long)OCFS2_I(inode)->ip_blkno,
Mark Fashehccd979b2005-12-15 14:31:24 -0800313 i_size_read(inode),
Mark Fashehb06970532006-03-03 10:24:33 -0800314 (unsigned long long)le64_to_cpu(fe->i_size),
315 le32_to_cpu(fe->i_flags));
Mark Fashehccd979b2005-12-15 14:31:24 -0800316
317 if (new_i_size > le64_to_cpu(fe->i_size)) {
Mark Fashehb06970532006-03-03 10:24:33 -0800318 mlog(0, "asked to truncate file with size (%llu) to size (%llu)!\n",
319 (unsigned long long)le64_to_cpu(fe->i_size),
320 (unsigned long long)new_i_size);
Mark Fashehccd979b2005-12-15 14:31:24 -0800321 status = -EINVAL;
322 mlog_errno(status);
323 goto bail;
324 }
325
Mark Fashehb06970532006-03-03 10:24:33 -0800326 mlog(0, "inode %llu, i_size = %llu, new_i_size = %llu\n",
327 (unsigned long long)le64_to_cpu(fe->i_blkno),
328 (unsigned long long)le64_to_cpu(fe->i_size),
329 (unsigned long long)new_i_size);
Mark Fashehccd979b2005-12-15 14:31:24 -0800330
331 /* lets handle the simple truncate cases before doing any more
332 * cluster locking. */
333 if (new_i_size == le64_to_cpu(fe->i_size))
334 goto bail;
335
Mark Fashehab0920c2006-03-16 15:06:37 -0800336 /* This forces other nodes to sync and drop their pages. Do
337 * this even if we have a truncate without allocation change -
338 * ocfs2 cluster sizes can be much greater than page size, so
339 * we have to truncate them anyway. */
340 status = ocfs2_data_lock(inode, 1);
341 if (status < 0) {
342 mlog_errno(status);
343 goto bail;
344 }
345 ocfs2_data_unlock(inode, 1);
346
Mark Fashehccd979b2005-12-15 14:31:24 -0800347 if (le32_to_cpu(fe->i_clusters) ==
348 ocfs2_clusters_for_bytes(osb->sb, new_i_size)) {
349 mlog(0, "fe->i_clusters = %u, so we do a simple truncate\n",
350 fe->i_clusters);
351 /* No allocation change is required, so lets fast path
352 * this truncate. */
353 status = ocfs2_simple_size_update(inode, di_bh, new_i_size);
354 if (status < 0)
355 mlog_errno(status);
356 goto bail;
357 }
358
Mark Fashehccd979b2005-12-15 14:31:24 -0800359 /* alright, we're going to need to do a full blown alloc size
360 * change. Orphan the inode so that recovery can complete the
361 * truncate if necessary. This does the task of marking
362 * i_size. */
363 status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
364 if (status < 0) {
365 mlog_errno(status);
366 goto bail;
367 }
368
369 status = ocfs2_prepare_truncate(osb, inode, di_bh, &tc);
370 if (status < 0) {
371 mlog_errno(status);
372 goto bail;
373 }
374
375 status = ocfs2_commit_truncate(osb, inode, di_bh, tc);
376 if (status < 0) {
377 mlog_errno(status);
378 goto bail;
379 }
380
381 /* TODO: orphan dir cleanup here. */
382bail:
383
384 mlog_exit(status);
385 return status;
386}
387
388/*
389 * extend allocation only here.
390 * we'll update all the disk stuff, and oip->alloc_size
391 *
392 * expect stuff to be locked, a transaction started and enough data /
393 * metadata reservations in the contexts.
394 *
395 * Will return -EAGAIN, and a reason if a restart is needed.
396 * If passed in, *reason will always be set, even in error.
397 */
398int ocfs2_do_extend_allocation(struct ocfs2_super *osb,
399 struct inode *inode,
Mark Fashehdcd05382007-01-16 11:32:23 -0800400 u32 *logical_offset,
Mark Fashehccd979b2005-12-15 14:31:24 -0800401 u32 clusters_to_add,
402 struct buffer_head *fe_bh,
Mark Fasheh1fabe142006-10-09 18:11:45 -0700403 handle_t *handle,
Mark Fashehccd979b2005-12-15 14:31:24 -0800404 struct ocfs2_alloc_context *data_ac,
405 struct ocfs2_alloc_context *meta_ac,
406 enum ocfs2_alloc_restarted *reason_ret)
407{
408 int status = 0;
409 int free_extents;
410 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
411 enum ocfs2_alloc_restarted reason = RESTART_NONE;
412 u32 bit_off, num_bits;
413 u64 block;
414
415 BUG_ON(!clusters_to_add);
416
417 free_extents = ocfs2_num_free_extents(osb, inode, fe);
418 if (free_extents < 0) {
419 status = free_extents;
420 mlog_errno(status);
421 goto leave;
422 }
423
424 /* there are two cases which could cause us to EAGAIN in the
425 * we-need-more-metadata case:
426 * 1) we haven't reserved *any*
427 * 2) we are so fragmented, we've needed to add metadata too
428 * many times. */
429 if (!free_extents && !meta_ac) {
430 mlog(0, "we haven't reserved any metadata!\n");
431 status = -EAGAIN;
432 reason = RESTART_META;
433 goto leave;
434 } else if ((!free_extents)
435 && (ocfs2_alloc_context_bits_left(meta_ac)
436 < ocfs2_extend_meta_needed(fe))) {
437 mlog(0, "filesystem is really fragmented...\n");
438 status = -EAGAIN;
439 reason = RESTART_META;
440 goto leave;
441 }
442
443 status = ocfs2_claim_clusters(osb, handle, data_ac, 1,
444 &bit_off, &num_bits);
445 if (status < 0) {
446 if (status != -ENOSPC)
447 mlog_errno(status);
448 goto leave;
449 }
450
451 BUG_ON(num_bits > clusters_to_add);
452
453 /* reserve our write early -- insert_extent may update the inode */
454 status = ocfs2_journal_access(handle, inode, fe_bh,
455 OCFS2_JOURNAL_ACCESS_WRITE);
456 if (status < 0) {
457 mlog_errno(status);
458 goto leave;
459 }
460
461 block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
Mark Fashehb06970532006-03-03 10:24:33 -0800462 mlog(0, "Allocating %u clusters at block %u for inode %llu\n",
463 num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
Mark Fashehdcd05382007-01-16 11:32:23 -0800464 status = ocfs2_insert_extent(osb, handle, inode, fe_bh,
465 *logical_offset, block, num_bits,
466 meta_ac);
Mark Fashehccd979b2005-12-15 14:31:24 -0800467 if (status < 0) {
468 mlog_errno(status);
469 goto leave;
470 }
471
Mark Fashehccd979b2005-12-15 14:31:24 -0800472 status = ocfs2_journal_dirty(handle, fe_bh);
473 if (status < 0) {
474 mlog_errno(status);
475 goto leave;
476 }
477
478 clusters_to_add -= num_bits;
Mark Fashehdcd05382007-01-16 11:32:23 -0800479 *logical_offset += num_bits;
Mark Fashehccd979b2005-12-15 14:31:24 -0800480
481 if (clusters_to_add) {
482 mlog(0, "need to alloc once more, clusters = %u, wanted = "
483 "%u\n", fe->i_clusters, clusters_to_add);
484 status = -EAGAIN;
485 reason = RESTART_TRANS;
486 }
487
488leave:
489 mlog_exit(status);
490 if (reason_ret)
491 *reason_ret = reason;
492 return status;
493}
494
495static int ocfs2_extend_allocation(struct inode *inode,
496 u32 clusters_to_add)
497{
498 int status = 0;
499 int restart_func = 0;
500 int drop_alloc_sem = 0;
501 int credits, num_free_extents;
Mark Fashehdcd05382007-01-16 11:32:23 -0800502 u32 prev_clusters, logical_start;
Mark Fashehccd979b2005-12-15 14:31:24 -0800503 struct buffer_head *bh = NULL;
504 struct ocfs2_dinode *fe = NULL;
Mark Fasheh1fabe142006-10-09 18:11:45 -0700505 handle_t *handle = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800506 struct ocfs2_alloc_context *data_ac = NULL;
507 struct ocfs2_alloc_context *meta_ac = NULL;
508 enum ocfs2_alloc_restarted why;
509 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
510
511 mlog_entry("(clusters_to_add = %u)\n", clusters_to_add);
512
Mark Fashehdcd05382007-01-16 11:32:23 -0800513 /*
514 * This function only exists for file systems which don't
515 * support holes.
516 */
517 BUG_ON(ocfs2_sparse_alloc(osb));
518
Mark Fashehccd979b2005-12-15 14:31:24 -0800519 status = ocfs2_read_block(osb, OCFS2_I(inode)->ip_blkno, &bh,
520 OCFS2_BH_CACHED, inode);
521 if (status < 0) {
522 mlog_errno(status);
523 goto leave;
524 }
525
526 fe = (struct ocfs2_dinode *) bh->b_data;
527 if (!OCFS2_IS_VALID_DINODE(fe)) {
528 OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe);
529 status = -EIO;
530 goto leave;
531 }
532
Mark Fashehdcd05382007-01-16 11:32:23 -0800533 logical_start = OCFS2_I(inode)->ip_clusters;
534
Mark Fashehccd979b2005-12-15 14:31:24 -0800535restart_all:
536 BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
537
Mark Fashehb06970532006-03-03 10:24:33 -0800538 mlog(0, "extend inode %llu, i_size = %lld, fe->i_clusters = %u, "
Mark Fashehccd979b2005-12-15 14:31:24 -0800539 "clusters_to_add = %u\n",
Mark Fashehb06970532006-03-03 10:24:33 -0800540 (unsigned long long)OCFS2_I(inode)->ip_blkno, i_size_read(inode),
Mark Fashehccd979b2005-12-15 14:31:24 -0800541 fe->i_clusters, clusters_to_add);
542
Mark Fashehccd979b2005-12-15 14:31:24 -0800543 num_free_extents = ocfs2_num_free_extents(osb,
544 inode,
545 fe);
546 if (num_free_extents < 0) {
547 status = num_free_extents;
548 mlog_errno(status);
549 goto leave;
550 }
551
552 if (!num_free_extents) {
Mark Fashehda5cbf22006-10-06 18:34:35 -0700553 status = ocfs2_reserve_new_metadata(osb, fe, &meta_ac);
Mark Fashehccd979b2005-12-15 14:31:24 -0800554 if (status < 0) {
555 if (status != -ENOSPC)
556 mlog_errno(status);
557 goto leave;
558 }
559 }
560
Mark Fashehda5cbf22006-10-06 18:34:35 -0700561 status = ocfs2_reserve_clusters(osb, clusters_to_add, &data_ac);
Mark Fashehccd979b2005-12-15 14:31:24 -0800562 if (status < 0) {
563 if (status != -ENOSPC)
564 mlog_errno(status);
565 goto leave;
566 }
567
568 /* blocks peope in read/write from reading our allocation
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800569 * until we're done changing it. We depend on i_mutex to block
Mark Fashehccd979b2005-12-15 14:31:24 -0800570 * other extend/truncate calls while we're here. Ordering wrt
571 * start_trans is important here -- always do it before! */
572 down_write(&OCFS2_I(inode)->ip_alloc_sem);
573 drop_alloc_sem = 1;
574
575 credits = ocfs2_calc_extend_credits(osb->sb, fe, clusters_to_add);
Mark Fasheh65eff9c2006-10-09 17:26:22 -0700576 handle = ocfs2_start_trans(osb, credits);
Mark Fashehccd979b2005-12-15 14:31:24 -0800577 if (IS_ERR(handle)) {
578 status = PTR_ERR(handle);
579 handle = NULL;
580 mlog_errno(status);
581 goto leave;
582 }
583
584restarted_transaction:
585 /* reserve a write to the file entry early on - that we if we
586 * run out of credits in the allocation path, we can still
587 * update i_size. */
588 status = ocfs2_journal_access(handle, inode, bh,
589 OCFS2_JOURNAL_ACCESS_WRITE);
590 if (status < 0) {
591 mlog_errno(status);
592 goto leave;
593 }
594
595 prev_clusters = OCFS2_I(inode)->ip_clusters;
596
597 status = ocfs2_do_extend_allocation(osb,
598 inode,
Mark Fashehdcd05382007-01-16 11:32:23 -0800599 &logical_start,
Mark Fashehccd979b2005-12-15 14:31:24 -0800600 clusters_to_add,
601 bh,
602 handle,
603 data_ac,
604 meta_ac,
605 &why);
606 if ((status < 0) && (status != -EAGAIN)) {
607 if (status != -ENOSPC)
608 mlog_errno(status);
609 goto leave;
610 }
611
612 status = ocfs2_journal_dirty(handle, bh);
613 if (status < 0) {
614 mlog_errno(status);
615 goto leave;
616 }
617
618 spin_lock(&OCFS2_I(inode)->ip_lock);
619 clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
620 spin_unlock(&OCFS2_I(inode)->ip_lock);
621
622 if (why != RESTART_NONE && clusters_to_add) {
623 if (why == RESTART_META) {
624 mlog(0, "restarting function.\n");
625 restart_func = 1;
626 } else {
627 BUG_ON(why != RESTART_TRANS);
628
629 mlog(0, "restarting transaction.\n");
630 /* TODO: This can be more intelligent. */
631 credits = ocfs2_calc_extend_credits(osb->sb,
632 fe,
633 clusters_to_add);
Mark Fasheh1fabe142006-10-09 18:11:45 -0700634 status = ocfs2_extend_trans(handle, credits);
Mark Fashehccd979b2005-12-15 14:31:24 -0800635 if (status < 0) {
636 /* handle still has to be committed at
637 * this point. */
638 status = -ENOMEM;
639 mlog_errno(status);
640 goto leave;
641 }
642 goto restarted_transaction;
643 }
644 }
645
Mark Fashehb06970532006-03-03 10:24:33 -0800646 mlog(0, "fe: i_clusters = %u, i_size=%llu\n",
647 fe->i_clusters, (unsigned long long)fe->i_size);
Mark Fashehccd979b2005-12-15 14:31:24 -0800648 mlog(0, "inode: ip_clusters=%u, i_size=%lld\n",
649 OCFS2_I(inode)->ip_clusters, i_size_read(inode));
650
651leave:
652 if (drop_alloc_sem) {
653 up_write(&OCFS2_I(inode)->ip_alloc_sem);
654 drop_alloc_sem = 0;
655 }
656 if (handle) {
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700657 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800658 handle = NULL;
659 }
660 if (data_ac) {
661 ocfs2_free_alloc_context(data_ac);
662 data_ac = NULL;
663 }
664 if (meta_ac) {
665 ocfs2_free_alloc_context(meta_ac);
666 meta_ac = NULL;
667 }
668 if ((!status) && restart_func) {
669 restart_func = 0;
670 goto restart_all;
671 }
672 if (bh) {
673 brelse(bh);
674 bh = NULL;
675 }
676
677 mlog_exit(status);
678 return status;
679}
680
681/* Some parts of this taken from generic_cont_expand, which turned out
682 * to be too fragile to do exactly what we need without us having to
Mark Fasheh53013cb2006-05-05 19:04:03 -0700683 * worry about recursive locking in ->prepare_write() and
684 * ->commit_write(). */
Mark Fashehccd979b2005-12-15 14:31:24 -0800685static int ocfs2_write_zero_page(struct inode *inode,
686 u64 size)
687{
688 struct address_space *mapping = inode->i_mapping;
689 struct page *page;
690 unsigned long index;
691 unsigned int offset;
Mark Fasheh1fabe142006-10-09 18:11:45 -0700692 handle_t *handle = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800693 int ret;
694
695 offset = (size & (PAGE_CACHE_SIZE-1)); /* Within page */
696 /* ugh. in prepare/commit_write, if from==to==start of block, we
697 ** skip the prepare. make sure we never send an offset for the start
698 ** of a block
699 */
700 if ((offset & (inode->i_sb->s_blocksize - 1)) == 0) {
701 offset++;
702 }
703 index = size >> PAGE_CACHE_SHIFT;
704
705 page = grab_cache_page(mapping, index);
706 if (!page) {
707 ret = -ENOMEM;
708 mlog_errno(ret);
709 goto out;
710 }
711
Mark Fasheh53013cb2006-05-05 19:04:03 -0700712 ret = ocfs2_prepare_write_nolock(inode, page, offset, offset);
Mark Fashehccd979b2005-12-15 14:31:24 -0800713 if (ret < 0) {
714 mlog_errno(ret);
715 goto out_unlock;
716 }
717
718 if (ocfs2_should_order_data(inode)) {
719 handle = ocfs2_start_walk_page_trans(inode, page, offset,
720 offset);
721 if (IS_ERR(handle)) {
722 ret = PTR_ERR(handle);
723 handle = NULL;
724 goto out_unlock;
725 }
726 }
727
728 /* must not update i_size! */
729 ret = block_commit_write(page, offset, offset);
730 if (ret < 0)
731 mlog_errno(ret);
732 else
733 ret = 0;
734
735 if (handle)
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700736 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800737out_unlock:
738 unlock_page(page);
739 page_cache_release(page);
740out:
741 return ret;
742}
743
744static int ocfs2_zero_extend(struct inode *inode,
745 u64 zero_to_size)
746{
747 int ret = 0;
748 u64 start_off;
749 struct super_block *sb = inode->i_sb;
750
751 start_off = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
752 while (start_off < zero_to_size) {
753 ret = ocfs2_write_zero_page(inode, start_off);
754 if (ret < 0) {
755 mlog_errno(ret);
756 goto out;
757 }
758
759 start_off += sb->s_blocksize;
Mark Fashehe2057c52006-10-03 17:53:05 -0700760
761 /*
762 * Very large extends have the potential to lock up
763 * the cpu for extended periods of time.
764 */
765 cond_resched();
Mark Fashehccd979b2005-12-15 14:31:24 -0800766 }
767
768out:
769 return ret;
770}
771
Mark Fasheh53013cb2006-05-05 19:04:03 -0700772/*
773 * A tail_to_skip value > 0 indicates that we're being called from
774 * ocfs2_file_aio_write(). This has the following implications:
775 *
776 * - we don't want to update i_size
777 * - di_bh will be NULL, which is fine because it's only used in the
778 * case where we want to update i_size.
779 * - ocfs2_zero_extend() will then only be filling the hole created
780 * between i_size and the start of the write.
781 */
Mark Fashehccd979b2005-12-15 14:31:24 -0800782static int ocfs2_extend_file(struct inode *inode,
783 struct buffer_head *di_bh,
Mark Fasheh53013cb2006-05-05 19:04:03 -0700784 u64 new_i_size,
785 size_t tail_to_skip)
Mark Fashehccd979b2005-12-15 14:31:24 -0800786{
787 int ret = 0;
788 u32 clusters_to_add;
789
Mark Fasheh53013cb2006-05-05 19:04:03 -0700790 BUG_ON(!tail_to_skip && !di_bh);
791
Mark Fashehccd979b2005-12-15 14:31:24 -0800792 /* setattr sometimes calls us like this. */
793 if (new_i_size == 0)
794 goto out;
795
796 if (i_size_read(inode) == new_i_size)
797 goto out;
798 BUG_ON(new_i_size < i_size_read(inode));
799
800 clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size) -
801 OCFS2_I(inode)->ip_clusters;
802
Mark Fasheh0effef72006-10-03 17:44:42 -0700803 /*
804 * protect the pages that ocfs2_zero_extend is going to be
805 * pulling into the page cache.. we do this before the
806 * metadata extend so that we don't get into the situation
807 * where we've extended the metadata but can't get the data
808 * lock to zero.
809 */
810 ret = ocfs2_data_lock(inode, 1);
811 if (ret < 0) {
812 mlog_errno(ret);
813 goto out;
814 }
Mark Fasheh53013cb2006-05-05 19:04:03 -0700815
Mark Fasheh0effef72006-10-03 17:44:42 -0700816 if (clusters_to_add) {
Mark Fashehccd979b2005-12-15 14:31:24 -0800817 ret = ocfs2_extend_allocation(inode, clusters_to_add);
818 if (ret < 0) {
819 mlog_errno(ret);
Mark Fasheh53013cb2006-05-05 19:04:03 -0700820 goto out_unlock;
Mark Fashehccd979b2005-12-15 14:31:24 -0800821 }
Mark Fasheh0effef72006-10-03 17:44:42 -0700822 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800823
Mark Fasheh0effef72006-10-03 17:44:42 -0700824 /*
825 * Call this even if we don't add any clusters to the tree. We
826 * still need to zero the area between the old i_size and the
827 * new i_size.
828 */
829 ret = ocfs2_zero_extend(inode, (u64)new_i_size - tail_to_skip);
830 if (ret < 0) {
831 mlog_errno(ret);
832 goto out_unlock;
Mark Fashehccd979b2005-12-15 14:31:24 -0800833 }
834
Mark Fasheh53013cb2006-05-05 19:04:03 -0700835 if (!tail_to_skip) {
836 /* We're being called from ocfs2_setattr() which wants
837 * us to update i_size */
838 ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
839 if (ret < 0)
840 mlog_errno(ret);
841 }
842
843out_unlock:
Mark Fasheh0effef72006-10-03 17:44:42 -0700844 ocfs2_data_unlock(inode, 1);
Mark Fasheh53013cb2006-05-05 19:04:03 -0700845
Mark Fashehccd979b2005-12-15 14:31:24 -0800846out:
847 return ret;
848}
849
850int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
851{
852 int status = 0, size_change;
853 struct inode *inode = dentry->d_inode;
854 struct super_block *sb = inode->i_sb;
855 struct ocfs2_super *osb = OCFS2_SB(sb);
856 struct buffer_head *bh = NULL;
Mark Fasheh1fabe142006-10-09 18:11:45 -0700857 handle_t *handle = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800858
859 mlog_entry("(0x%p, '%.*s')\n", dentry,
860 dentry->d_name.len, dentry->d_name.name);
861
862 if (attr->ia_valid & ATTR_MODE)
863 mlog(0, "mode change: %d\n", attr->ia_mode);
864 if (attr->ia_valid & ATTR_UID)
865 mlog(0, "uid change: %d\n", attr->ia_uid);
866 if (attr->ia_valid & ATTR_GID)
867 mlog(0, "gid change: %d\n", attr->ia_gid);
868 if (attr->ia_valid & ATTR_SIZE)
869 mlog(0, "size change...\n");
870 if (attr->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME))
871 mlog(0, "time change...\n");
872
873#define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
874 | ATTR_GID | ATTR_UID | ATTR_MODE)
875 if (!(attr->ia_valid & OCFS2_VALID_ATTRS)) {
876 mlog(0, "can't handle attrs: 0x%x\n", attr->ia_valid);
877 return 0;
878 }
879
880 status = inode_change_ok(inode, attr);
881 if (status)
882 return status;
883
884 size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
885 if (size_change) {
886 status = ocfs2_rw_lock(inode, 1);
887 if (status < 0) {
888 mlog_errno(status);
889 goto bail;
890 }
891 }
892
Mark Fasheh4bcec182006-10-09 16:02:40 -0700893 status = ocfs2_meta_lock(inode, &bh, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -0800894 if (status < 0) {
895 if (status != -ENOENT)
896 mlog_errno(status);
897 goto bail_unlock_rw;
898 }
899
900 if (size_change && attr->ia_size != i_size_read(inode)) {
901 if (i_size_read(inode) > attr->ia_size)
902 status = ocfs2_truncate_file(inode, bh, attr->ia_size);
903 else
Mark Fasheh53013cb2006-05-05 19:04:03 -0700904 status = ocfs2_extend_file(inode, bh, attr->ia_size, 0);
Mark Fashehccd979b2005-12-15 14:31:24 -0800905 if (status < 0) {
906 if (status != -ENOSPC)
907 mlog_errno(status);
908 status = -ENOSPC;
909 goto bail_unlock;
910 }
911 }
912
Mark Fasheh65eff9c2006-10-09 17:26:22 -0700913 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
Mark Fashehccd979b2005-12-15 14:31:24 -0800914 if (IS_ERR(handle)) {
915 status = PTR_ERR(handle);
916 mlog_errno(status);
917 goto bail_unlock;
918 }
919
920 status = inode_setattr(inode, attr);
921 if (status < 0) {
922 mlog_errno(status);
923 goto bail_commit;
924 }
925
926 status = ocfs2_mark_inode_dirty(handle, inode, bh);
927 if (status < 0)
928 mlog_errno(status);
929
930bail_commit:
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700931 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800932bail_unlock:
933 ocfs2_meta_unlock(inode, 1);
934bail_unlock_rw:
935 if (size_change)
936 ocfs2_rw_unlock(inode, 1);
937bail:
938 if (bh)
939 brelse(bh);
940
941 mlog_exit(status);
942 return status;
943}
944
945int ocfs2_getattr(struct vfsmount *mnt,
946 struct dentry *dentry,
947 struct kstat *stat)
948{
949 struct inode *inode = dentry->d_inode;
950 struct super_block *sb = dentry->d_inode->i_sb;
951 struct ocfs2_super *osb = sb->s_fs_info;
952 int err;
953
954 mlog_entry_void();
955
956 err = ocfs2_inode_revalidate(dentry);
957 if (err) {
958 if (err != -ENOENT)
959 mlog_errno(err);
960 goto bail;
961 }
962
963 generic_fillattr(inode, stat);
964
965 /* We set the blksize from the cluster size for performance */
966 stat->blksize = osb->s_clustersize;
967
968bail:
969 mlog_exit(err);
970
971 return err;
972}
973
Tiger Yangd38eb8d2006-11-27 09:59:21 +0800974int ocfs2_permission(struct inode *inode, int mask, struct nameidata *nd)
975{
976 int ret;
977
978 mlog_entry_void();
979
980 ret = ocfs2_meta_lock(inode, NULL, 0);
981 if (ret) {
Mark Fasheha9f5f702007-04-26 11:43:43 -0700982 if (ret != -ENOENT)
983 mlog_errno(ret);
Tiger Yangd38eb8d2006-11-27 09:59:21 +0800984 goto out;
985 }
986
987 ret = generic_permission(inode, mask, NULL);
Tiger Yangd38eb8d2006-11-27 09:59:21 +0800988
989 ocfs2_meta_unlock(inode, 0);
990out:
991 mlog_exit(ret);
992 return ret;
993}
994
Mark Fashehccd979b2005-12-15 14:31:24 -0800995static int ocfs2_write_remove_suid(struct inode *inode)
996{
997 int ret;
998 struct buffer_head *bh = NULL;
999 struct ocfs2_inode_info *oi = OCFS2_I(inode);
Mark Fasheh1fabe142006-10-09 18:11:45 -07001000 handle_t *handle;
Mark Fashehccd979b2005-12-15 14:31:24 -08001001 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1002 struct ocfs2_dinode *di;
1003
Mark Fashehb06970532006-03-03 10:24:33 -08001004 mlog_entry("(Inode %llu, mode 0%o)\n",
1005 (unsigned long long)oi->ip_blkno, inode->i_mode);
Mark Fashehccd979b2005-12-15 14:31:24 -08001006
Mark Fasheh65eff9c2006-10-09 17:26:22 -07001007 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
Mark Fashehccd979b2005-12-15 14:31:24 -08001008 if (handle == NULL) {
1009 ret = -ENOMEM;
1010 mlog_errno(ret);
1011 goto out;
1012 }
1013
1014 ret = ocfs2_read_block(osb, oi->ip_blkno, &bh, OCFS2_BH_CACHED, inode);
1015 if (ret < 0) {
1016 mlog_errno(ret);
1017 goto out_trans;
1018 }
1019
1020 ret = ocfs2_journal_access(handle, inode, bh,
1021 OCFS2_JOURNAL_ACCESS_WRITE);
1022 if (ret < 0) {
1023 mlog_errno(ret);
1024 goto out_bh;
1025 }
1026
1027 inode->i_mode &= ~S_ISUID;
1028 if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
1029 inode->i_mode &= ~S_ISGID;
1030
1031 di = (struct ocfs2_dinode *) bh->b_data;
1032 di->i_mode = cpu_to_le16(inode->i_mode);
1033
1034 ret = ocfs2_journal_dirty(handle, bh);
1035 if (ret < 0)
1036 mlog_errno(ret);
1037out_bh:
1038 brelse(bh);
1039out_trans:
Mark Fasheh02dc1af2006-10-09 16:48:10 -07001040 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -08001041out:
1042 mlog_exit(ret);
1043 return ret;
1044}
1045
Tiger Yang8659ac22006-10-17 18:29:52 -07001046static int ocfs2_prepare_inode_for_write(struct dentry *dentry,
1047 loff_t *ppos,
1048 size_t count,
1049 int appending)
Mark Fashehccd979b2005-12-15 14:31:24 -08001050{
Tiger Yang8659ac22006-10-17 18:29:52 -07001051 int ret = 0, meta_level = appending;
1052 struct inode *inode = dentry->d_inode;
Mark Fashehccd979b2005-12-15 14:31:24 -08001053 u32 clusters;
Mark Fashehccd979b2005-12-15 14:31:24 -08001054 loff_t newsize, saved_pos;
Mark Fashehccd979b2005-12-15 14:31:24 -08001055
Mark Fashehccd979b2005-12-15 14:31:24 -08001056 /*
1057 * We sample i_size under a read level meta lock to see if our write
1058 * is extending the file, if it is we back off and get a write level
1059 * meta lock.
1060 */
Mark Fashehccd979b2005-12-15 14:31:24 -08001061 for(;;) {
Mark Fasheh4bcec182006-10-09 16:02:40 -07001062 ret = ocfs2_meta_lock(inode, NULL, meta_level);
Mark Fashehccd979b2005-12-15 14:31:24 -08001063 if (ret < 0) {
1064 meta_level = -1;
1065 mlog_errno(ret);
1066 goto out;
1067 }
1068
1069 /* Clear suid / sgid if necessary. We do this here
1070 * instead of later in the write path because
1071 * remove_suid() calls ->setattr without any hint that
1072 * we may have already done our cluster locking. Since
1073 * ocfs2_setattr() *must* take cluster locks to
1074 * proceeed, this will lead us to recursively lock the
1075 * inode. There's also the dinode i_size state which
1076 * can be lost via setattr during extending writes (we
1077 * set inode->i_size at the end of a write. */
Tiger Yang8659ac22006-10-17 18:29:52 -07001078 if (should_remove_suid(dentry)) {
Mark Fashehccd979b2005-12-15 14:31:24 -08001079 if (meta_level == 0) {
1080 ocfs2_meta_unlock(inode, meta_level);
1081 meta_level = 1;
1082 continue;
1083 }
1084
1085 ret = ocfs2_write_remove_suid(inode);
1086 if (ret < 0) {
1087 mlog_errno(ret);
Tiger Yang8659ac22006-10-17 18:29:52 -07001088 goto out_unlock;
Mark Fashehccd979b2005-12-15 14:31:24 -08001089 }
1090 }
1091
1092 /* work on a copy of ppos until we're sure that we won't have
1093 * to recalculate it due to relocking. */
Tiger Yang8659ac22006-10-17 18:29:52 -07001094 if (appending) {
Mark Fashehccd979b2005-12-15 14:31:24 -08001095 saved_pos = i_size_read(inode);
1096 mlog(0, "O_APPEND: inode->i_size=%llu\n", saved_pos);
1097 } else {
Tiger Yang8659ac22006-10-17 18:29:52 -07001098 saved_pos = *ppos;
Mark Fashehccd979b2005-12-15 14:31:24 -08001099 }
Tiger Yang8659ac22006-10-17 18:29:52 -07001100 newsize = count + saved_pos;
Mark Fashehccd979b2005-12-15 14:31:24 -08001101
Mark Fasheh215c7f92006-02-01 16:42:10 -08001102 mlog(0, "pos=%lld newsize=%lld cursize=%lld\n",
1103 (long long) saved_pos, (long long) newsize,
1104 (long long) i_size_read(inode));
Mark Fashehccd979b2005-12-15 14:31:24 -08001105
1106 /* No need for a higher level metadata lock if we're
1107 * never going past i_size. */
1108 if (newsize <= i_size_read(inode))
1109 break;
1110
1111 if (meta_level == 0) {
1112 ocfs2_meta_unlock(inode, meta_level);
1113 meta_level = 1;
1114 continue;
1115 }
1116
1117 spin_lock(&OCFS2_I(inode)->ip_lock);
1118 clusters = ocfs2_clusters_for_bytes(inode->i_sb, newsize) -
1119 OCFS2_I(inode)->ip_clusters;
1120 spin_unlock(&OCFS2_I(inode)->ip_lock);
1121
1122 mlog(0, "Writing at EOF, may need more allocation: "
Mark Fasheh215c7f92006-02-01 16:42:10 -08001123 "i_size = %lld, newsize = %lld, need %u clusters\n",
1124 (long long) i_size_read(inode), (long long) newsize,
1125 clusters);
Mark Fashehccd979b2005-12-15 14:31:24 -08001126
1127 /* We only want to continue the rest of this loop if
1128 * our extend will actually require more
1129 * allocation. */
1130 if (!clusters)
1131 break;
1132
Tiger Yang8659ac22006-10-17 18:29:52 -07001133 ret = ocfs2_extend_file(inode, NULL, newsize, count);
Mark Fashehccd979b2005-12-15 14:31:24 -08001134 if (ret < 0) {
1135 if (ret != -ENOSPC)
1136 mlog_errno(ret);
Tiger Yang8659ac22006-10-17 18:29:52 -07001137 goto out_unlock;
Mark Fashehccd979b2005-12-15 14:31:24 -08001138 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001139 break;
1140 }
1141
Tiger Yang8659ac22006-10-17 18:29:52 -07001142 if (appending)
1143 *ppos = saved_pos;
1144
1145out_unlock:
Mark Fashehccd979b2005-12-15 14:31:24 -08001146 ocfs2_meta_unlock(inode, meta_level);
Tiger Yang8659ac22006-10-17 18:29:52 -07001147
1148out:
1149 return ret;
1150}
1151
1152static ssize_t ocfs2_file_aio_write(struct kiocb *iocb,
1153 const struct iovec *iov,
1154 unsigned long nr_segs,
1155 loff_t pos)
1156{
1157 int ret, rw_level, have_alloc_sem = 0;
1158 struct file *filp = iocb->ki_filp;
Josef Sipekd28c9172006-12-08 02:37:25 -08001159 struct inode *inode = filp->f_path.dentry->d_inode;
Tiger Yang8659ac22006-10-17 18:29:52 -07001160 int appending = filp->f_flags & O_APPEND ? 1 : 0;
1161
1162 mlog_entry("(0x%p, %u, '%.*s')\n", filp,
1163 (unsigned int)nr_segs,
Josef Sipekd28c9172006-12-08 02:37:25 -08001164 filp->f_path.dentry->d_name.len,
1165 filp->f_path.dentry->d_name.name);
Tiger Yang8659ac22006-10-17 18:29:52 -07001166
1167 /* happy write of zero bytes */
1168 if (iocb->ki_left == 0)
1169 return 0;
1170
1171 mutex_lock(&inode->i_mutex);
1172 /* to match setattr's i_mutex -> i_alloc_sem -> rw_lock ordering */
1173 if (filp->f_flags & O_DIRECT) {
1174 have_alloc_sem = 1;
1175 down_read(&inode->i_alloc_sem);
1176 }
1177
1178 /* concurrent O_DIRECT writes are allowed */
1179 rw_level = (filp->f_flags & O_DIRECT) ? 0 : 1;
1180 ret = ocfs2_rw_lock(inode, rw_level);
1181 if (ret < 0) {
1182 rw_level = -1;
1183 mlog_errno(ret);
1184 goto out;
1185 }
1186
Josef Sipekd28c9172006-12-08 02:37:25 -08001187 ret = ocfs2_prepare_inode_for_write(filp->f_path.dentry, &iocb->ki_pos,
Tiger Yang8659ac22006-10-17 18:29:52 -07001188 iocb->ki_left, appending);
1189 if (ret < 0) {
1190 mlog_errno(ret);
1191 goto out;
1192 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001193
1194 /* communicate with ocfs2_dio_end_io */
1195 ocfs2_iocb_set_rw_locked(iocb);
1196
Badari Pulavarty027445c2006-09-30 23:28:46 -07001197 ret = generic_file_aio_write_nolock(iocb, iov, nr_segs, iocb->ki_pos);
Mark Fashehccd979b2005-12-15 14:31:24 -08001198
1199 /* buffered aio wouldn't have proper lock coverage today */
1200 BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT));
1201
1202 /*
1203 * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
1204 * function pointer which is called when o_direct io completes so that
1205 * it can unlock our rw lock. (it's the clustered equivalent of
1206 * i_alloc_sem; protects truncate from racing with pending ios).
1207 * Unfortunately there are error cases which call end_io and others
1208 * that don't. so we don't have to unlock the rw_lock if either an
1209 * async dio is going to do it in the future or an end_io after an
1210 * error has already done it.
1211 */
1212 if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
1213 rw_level = -1;
1214 have_alloc_sem = 0;
1215 }
1216
1217out:
Mark Fashehccd979b2005-12-15 14:31:24 -08001218 if (have_alloc_sem)
1219 up_read(&inode->i_alloc_sem);
1220 if (rw_level != -1)
1221 ocfs2_rw_unlock(inode, rw_level);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001222 mutex_unlock(&inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -08001223
1224 mlog_exit(ret);
1225 return ret;
1226}
1227
Tiger Yang8659ac22006-10-17 18:29:52 -07001228static ssize_t ocfs2_file_splice_write(struct pipe_inode_info *pipe,
1229 struct file *out,
1230 loff_t *ppos,
1231 size_t len,
1232 unsigned int flags)
1233{
1234 int ret;
Josef Sipekd28c9172006-12-08 02:37:25 -08001235 struct inode *inode = out->f_path.dentry->d_inode;
Tiger Yang8659ac22006-10-17 18:29:52 -07001236
1237 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", out, pipe,
1238 (unsigned int)len,
Josef Sipekd28c9172006-12-08 02:37:25 -08001239 out->f_path.dentry->d_name.len,
1240 out->f_path.dentry->d_name.name);
Tiger Yang8659ac22006-10-17 18:29:52 -07001241
1242 inode_double_lock(inode, pipe->inode);
1243
1244 ret = ocfs2_rw_lock(inode, 1);
1245 if (ret < 0) {
1246 mlog_errno(ret);
1247 goto out;
1248 }
1249
Josef Sipekd28c9172006-12-08 02:37:25 -08001250 ret = ocfs2_prepare_inode_for_write(out->f_path.dentry, ppos, len, 0);
Tiger Yang8659ac22006-10-17 18:29:52 -07001251 if (ret < 0) {
1252 mlog_errno(ret);
1253 goto out_unlock;
1254 }
1255
1256 /* ok, we're done with i_size and alloc work */
1257 ret = generic_file_splice_write_nolock(pipe, out, ppos, len, flags);
1258
1259out_unlock:
1260 ocfs2_rw_unlock(inode, 1);
1261out:
1262 inode_double_unlock(inode, pipe->inode);
1263
1264 mlog_exit(ret);
1265 return ret;
1266}
1267
1268static ssize_t ocfs2_file_splice_read(struct file *in,
1269 loff_t *ppos,
1270 struct pipe_inode_info *pipe,
1271 size_t len,
1272 unsigned int flags)
1273{
1274 int ret = 0;
Josef Sipekd28c9172006-12-08 02:37:25 -08001275 struct inode *inode = in->f_path.dentry->d_inode;
Tiger Yang8659ac22006-10-17 18:29:52 -07001276
1277 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", in, pipe,
1278 (unsigned int)len,
Josef Sipekd28c9172006-12-08 02:37:25 -08001279 in->f_path.dentry->d_name.len,
1280 in->f_path.dentry->d_name.name);
Tiger Yang8659ac22006-10-17 18:29:52 -07001281
1282 /*
1283 * See the comment in ocfs2_file_aio_read()
1284 */
1285 ret = ocfs2_meta_lock(inode, NULL, 0);
1286 if (ret < 0) {
1287 mlog_errno(ret);
1288 goto bail;
1289 }
1290 ocfs2_meta_unlock(inode, 0);
1291
1292 ret = generic_file_splice_read(in, ppos, pipe, len, flags);
1293
1294bail:
1295 mlog_exit(ret);
1296 return ret;
1297}
1298
Mark Fashehccd979b2005-12-15 14:31:24 -08001299static ssize_t ocfs2_file_aio_read(struct kiocb *iocb,
Badari Pulavarty027445c2006-09-30 23:28:46 -07001300 const struct iovec *iov,
1301 unsigned long nr_segs,
Mark Fashehccd979b2005-12-15 14:31:24 -08001302 loff_t pos)
1303{
Tiger Yang25899de2006-11-15 15:49:02 +08001304 int ret = 0, rw_level = -1, have_alloc_sem = 0, lock_level = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -08001305 struct file *filp = iocb->ki_filp;
Josef Sipekd28c9172006-12-08 02:37:25 -08001306 struct inode *inode = filp->f_path.dentry->d_inode;
Mark Fashehccd979b2005-12-15 14:31:24 -08001307
Badari Pulavarty027445c2006-09-30 23:28:46 -07001308 mlog_entry("(0x%p, %u, '%.*s')\n", filp,
1309 (unsigned int)nr_segs,
Josef Sipekd28c9172006-12-08 02:37:25 -08001310 filp->f_path.dentry->d_name.len,
1311 filp->f_path.dentry->d_name.name);
Mark Fashehccd979b2005-12-15 14:31:24 -08001312
1313 if (!inode) {
1314 ret = -EINVAL;
1315 mlog_errno(ret);
1316 goto bail;
1317 }
1318
Mark Fashehccd979b2005-12-15 14:31:24 -08001319 /*
1320 * buffered reads protect themselves in ->readpage(). O_DIRECT reads
1321 * need locks to protect pending reads from racing with truncate.
1322 */
1323 if (filp->f_flags & O_DIRECT) {
1324 down_read(&inode->i_alloc_sem);
1325 have_alloc_sem = 1;
1326
1327 ret = ocfs2_rw_lock(inode, 0);
1328 if (ret < 0) {
1329 mlog_errno(ret);
1330 goto bail;
1331 }
1332 rw_level = 0;
1333 /* communicate with ocfs2_dio_end_io */
1334 ocfs2_iocb_set_rw_locked(iocb);
1335 }
1336
Mark Fashehc4374f82006-05-05 19:04:35 -07001337 /*
1338 * We're fine letting folks race truncates and extending
1339 * writes with read across the cluster, just like they can
1340 * locally. Hence no rw_lock during read.
1341 *
1342 * Take and drop the meta data lock to update inode fields
1343 * like i_size. This allows the checks down below
1344 * generic_file_aio_read() a chance of actually working.
1345 */
Tiger Yang25899de2006-11-15 15:49:02 +08001346 ret = ocfs2_meta_lock_atime(inode, filp->f_vfsmnt, &lock_level);
Mark Fashehc4374f82006-05-05 19:04:35 -07001347 if (ret < 0) {
1348 mlog_errno(ret);
1349 goto bail;
1350 }
Tiger Yang25899de2006-11-15 15:49:02 +08001351 ocfs2_meta_unlock(inode, lock_level);
Mark Fashehc4374f82006-05-05 19:04:35 -07001352
Badari Pulavarty027445c2006-09-30 23:28:46 -07001353 ret = generic_file_aio_read(iocb, iov, nr_segs, iocb->ki_pos);
Mark Fashehccd979b2005-12-15 14:31:24 -08001354 if (ret == -EINVAL)
1355 mlog(ML_ERROR, "generic_file_aio_read returned -EINVAL\n");
1356
1357 /* buffered aio wouldn't have proper lock coverage today */
1358 BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT));
1359
1360 /* see ocfs2_file_aio_write */
1361 if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
1362 rw_level = -1;
1363 have_alloc_sem = 0;
1364 }
1365
1366bail:
1367 if (have_alloc_sem)
1368 up_read(&inode->i_alloc_sem);
1369 if (rw_level != -1)
1370 ocfs2_rw_unlock(inode, rw_level);
1371 mlog_exit(ret);
1372
1373 return ret;
1374}
1375
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08001376const struct inode_operations ocfs2_file_iops = {
Mark Fashehccd979b2005-12-15 14:31:24 -08001377 .setattr = ocfs2_setattr,
1378 .getattr = ocfs2_getattr,
Tiger Yangd38eb8d2006-11-27 09:59:21 +08001379 .permission = ocfs2_permission,
Mark Fashehccd979b2005-12-15 14:31:24 -08001380};
1381
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08001382const struct inode_operations ocfs2_special_file_iops = {
Mark Fashehccd979b2005-12-15 14:31:24 -08001383 .setattr = ocfs2_setattr,
1384 .getattr = ocfs2_getattr,
Tiger Yangd38eb8d2006-11-27 09:59:21 +08001385 .permission = ocfs2_permission,
Mark Fashehccd979b2005-12-15 14:31:24 -08001386};
1387
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08001388const struct file_operations ocfs2_fops = {
Mark Fashehccd979b2005-12-15 14:31:24 -08001389 .read = do_sync_read,
1390 .write = do_sync_write,
1391 .sendfile = generic_file_sendfile,
1392 .mmap = ocfs2_mmap,
1393 .fsync = ocfs2_sync_file,
1394 .release = ocfs2_file_release,
1395 .open = ocfs2_file_open,
1396 .aio_read = ocfs2_file_aio_read,
1397 .aio_write = ocfs2_file_aio_write,
Herbert Poetzlca4d1472006-07-03 17:27:12 -07001398 .ioctl = ocfs2_ioctl,
Tiger Yang8659ac22006-10-17 18:29:52 -07001399 .splice_read = ocfs2_file_splice_read,
1400 .splice_write = ocfs2_file_splice_write,
Mark Fashehccd979b2005-12-15 14:31:24 -08001401};
1402
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08001403const struct file_operations ocfs2_dops = {
Mark Fashehccd979b2005-12-15 14:31:24 -08001404 .read = generic_read_dir,
1405 .readdir = ocfs2_readdir,
1406 .fsync = ocfs2_sync_file,
Herbert Poetzlca4d1472006-07-03 17:27:12 -07001407 .ioctl = ocfs2_ioctl,
Mark Fashehccd979b2005-12-15 14:31:24 -08001408};