blob: 3e21ad9a6dde43bf3aad4970712b307345e2702f [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>
Jens Axboed6b29d72007-06-04 09:59:47 +020034#include <linux/splice.h>
Tiger Yang7f1a37e2006-11-15 15:48:42 +080035#include <linux/mount.h>
Mark Fasheh9517bac2007-02-09 20:24:12 -080036#include <linux/writeback.h>
Mark Fashehccd979b2005-12-15 14:31:24 -080037
38#define MLOG_MASK_PREFIX ML_INODE
39#include <cluster/masklog.h>
40
41#include "ocfs2.h"
42
43#include "alloc.h"
44#include "aops.h"
45#include "dir.h"
46#include "dlmglue.h"
47#include "extent_map.h"
48#include "file.h"
49#include "sysfile.h"
50#include "inode.h"
Herbert Poetzlca4d1472006-07-03 17:27:12 -070051#include "ioctl.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080052#include "journal.h"
53#include "mmap.h"
54#include "suballoc.h"
55#include "super.h"
56
57#include "buffer_head_io.h"
58
59static int ocfs2_sync_inode(struct inode *inode)
60{
61 filemap_fdatawrite(inode->i_mapping);
62 return sync_mapping_buffers(inode->i_mapping);
63}
64
65static int ocfs2_file_open(struct inode *inode, struct file *file)
66{
67 int status;
68 int mode = file->f_flags;
69 struct ocfs2_inode_info *oi = OCFS2_I(inode);
70
71 mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file,
Josef Sipekd28c9172006-12-08 02:37:25 -080072 file->f_path.dentry->d_name.len, file->f_path.dentry->d_name.name);
Mark Fashehccd979b2005-12-15 14:31:24 -080073
74 spin_lock(&oi->ip_lock);
75
76 /* Check that the inode hasn't been wiped from disk by another
77 * node. If it hasn't then we're safe as long as we hold the
78 * spin lock until our increment of open count. */
79 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
80 spin_unlock(&oi->ip_lock);
81
82 status = -ENOENT;
83 goto leave;
84 }
85
86 if (mode & O_DIRECT)
87 oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT;
88
89 oi->ip_open_count++;
90 spin_unlock(&oi->ip_lock);
91 status = 0;
92leave:
93 mlog_exit(status);
94 return status;
95}
96
97static int ocfs2_file_release(struct inode *inode, struct file *file)
98{
99 struct ocfs2_inode_info *oi = OCFS2_I(inode);
100
101 mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file,
Josef Sipekd28c9172006-12-08 02:37:25 -0800102 file->f_path.dentry->d_name.len,
103 file->f_path.dentry->d_name.name);
Mark Fashehccd979b2005-12-15 14:31:24 -0800104
105 spin_lock(&oi->ip_lock);
106 if (!--oi->ip_open_count)
107 oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT;
108 spin_unlock(&oi->ip_lock);
109
110 mlog_exit(0);
111
112 return 0;
113}
114
115static int ocfs2_sync_file(struct file *file,
116 struct dentry *dentry,
117 int datasync)
118{
119 int err = 0;
120 journal_t *journal;
121 struct inode *inode = dentry->d_inode;
122 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
123
124 mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", file, dentry, datasync,
125 dentry->d_name.len, dentry->d_name.name);
126
127 err = ocfs2_sync_inode(dentry->d_inode);
128 if (err)
129 goto bail;
130
131 journal = osb->journal->j_journal;
132 err = journal_force_commit(journal);
133
134bail:
135 mlog_exit(err);
136
137 return (err < 0) ? -EIO : 0;
138}
139
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800140int ocfs2_should_update_atime(struct inode *inode,
141 struct vfsmount *vfsmnt)
142{
143 struct timespec now;
144 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
145
146 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
147 return 0;
148
149 if ((inode->i_flags & S_NOATIME) ||
150 ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)))
151 return 0;
152
Mark Fasheh6c2aad02006-12-19 15:25:52 -0800153 /*
154 * We can be called with no vfsmnt structure - NFSD will
155 * sometimes do this.
156 *
157 * Note that our action here is different than touch_atime() -
158 * if we can't tell whether this is a noatime mount, then we
159 * don't know whether to trust the value of s_atime_quantum.
160 */
161 if (vfsmnt == NULL)
162 return 0;
163
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800164 if ((vfsmnt->mnt_flags & MNT_NOATIME) ||
165 ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
166 return 0;
167
Mark Fasheh7e913c52006-12-13 00:34:35 -0800168 if (vfsmnt->mnt_flags & MNT_RELATIME) {
169 if ((timespec_compare(&inode->i_atime, &inode->i_mtime) <= 0) ||
170 (timespec_compare(&inode->i_atime, &inode->i_ctime) <= 0))
171 return 1;
172
173 return 0;
174 }
175
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800176 now = CURRENT_TIME;
177 if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum))
178 return 0;
179 else
180 return 1;
181}
182
183int ocfs2_update_inode_atime(struct inode *inode,
184 struct buffer_head *bh)
185{
186 int ret;
187 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
188 handle_t *handle;
189
190 mlog_entry_void();
191
192 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
193 if (handle == NULL) {
194 ret = -ENOMEM;
195 mlog_errno(ret);
196 goto out;
197 }
198
199 inode->i_atime = CURRENT_TIME;
200 ret = ocfs2_mark_inode_dirty(handle, inode, bh);
201 if (ret < 0)
202 mlog_errno(ret);
203
204 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
205out:
206 mlog_exit(ret);
207 return ret;
208}
209
Adrian Bunk6cb129f2007-04-26 00:29:35 -0700210static int ocfs2_set_inode_size(handle_t *handle,
211 struct inode *inode,
212 struct buffer_head *fe_bh,
213 u64 new_i_size)
Mark Fashehccd979b2005-12-15 14:31:24 -0800214{
215 int status;
216
217 mlog_entry_void();
218 i_size_write(inode, new_i_size);
Mark Fasheh8110b072007-03-22 16:53:23 -0700219 inode->i_blocks = ocfs2_inode_sector_count(inode);
Mark Fashehccd979b2005-12-15 14:31:24 -0800220 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
221
222 status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
223 if (status < 0) {
224 mlog_errno(status);
225 goto bail;
226 }
227
228bail:
229 mlog_exit(status);
230 return status;
231}
232
233static int ocfs2_simple_size_update(struct inode *inode,
234 struct buffer_head *di_bh,
235 u64 new_i_size)
236{
237 int ret;
238 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
Mark Fasheh1fabe142006-10-09 18:11:45 -0700239 handle_t *handle = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800240
Mark Fasheh65eff9c2006-10-09 17:26:22 -0700241 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
Mark Fashehccd979b2005-12-15 14:31:24 -0800242 if (handle == NULL) {
243 ret = -ENOMEM;
244 mlog_errno(ret);
245 goto out;
246 }
247
248 ret = ocfs2_set_inode_size(handle, inode, di_bh,
249 new_i_size);
250 if (ret < 0)
251 mlog_errno(ret);
252
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700253 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800254out:
255 return ret;
256}
257
258static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
259 struct inode *inode,
260 struct buffer_head *fe_bh,
261 u64 new_i_size)
262{
263 int status;
Mark Fasheh1fabe142006-10-09 18:11:45 -0700264 handle_t *handle;
Mark Fasheh60b11392007-02-16 11:46:50 -0800265 struct ocfs2_dinode *di;
Mark Fashehccd979b2005-12-15 14:31:24 -0800266
267 mlog_entry_void();
268
269 /* TODO: This needs to actually orphan the inode in this
270 * transaction. */
271
Mark Fasheh65eff9c2006-10-09 17:26:22 -0700272 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
Mark Fashehccd979b2005-12-15 14:31:24 -0800273 if (IS_ERR(handle)) {
274 status = PTR_ERR(handle);
275 mlog_errno(status);
276 goto out;
277 }
278
Mark Fasheh60b11392007-02-16 11:46:50 -0800279 status = ocfs2_journal_access(handle, inode, fe_bh,
280 OCFS2_JOURNAL_ACCESS_WRITE);
281 if (status < 0) {
282 mlog_errno(status);
283 goto out_commit;
284 }
285
286 /*
287 * Do this before setting i_size.
288 */
289 status = ocfs2_zero_tail_for_truncate(inode, handle, new_i_size);
290 if (status) {
291 mlog_errno(status);
292 goto out_commit;
293 }
294
295 i_size_write(inode, new_i_size);
296 inode->i_blocks = ocfs2_align_bytes_to_sectors(new_i_size);
297 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
298
299 di = (struct ocfs2_dinode *) fe_bh->b_data;
300 di->i_size = cpu_to_le64(new_i_size);
301 di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
302 di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
303
304 status = ocfs2_journal_dirty(handle, fe_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800305 if (status < 0)
306 mlog_errno(status);
307
Mark Fasheh60b11392007-02-16 11:46:50 -0800308out_commit:
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700309 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800310out:
Mark Fasheh60b11392007-02-16 11:46:50 -0800311
Mark Fashehccd979b2005-12-15 14:31:24 -0800312 mlog_exit(status);
313 return status;
314}
315
316static int ocfs2_truncate_file(struct inode *inode,
317 struct buffer_head *di_bh,
318 u64 new_i_size)
319{
320 int status = 0;
321 struct ocfs2_dinode *fe = NULL;
322 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
323 struct ocfs2_truncate_context *tc = NULL;
324
Mark Fashehb06970532006-03-03 10:24:33 -0800325 mlog_entry("(inode = %llu, new_i_size = %llu\n",
326 (unsigned long long)OCFS2_I(inode)->ip_blkno,
327 (unsigned long long)new_i_size);
Mark Fashehccd979b2005-12-15 14:31:24 -0800328
Mark Fashehccd979b2005-12-15 14:31:24 -0800329 fe = (struct ocfs2_dinode *) di_bh->b_data;
330 if (!OCFS2_IS_VALID_DINODE(fe)) {
331 OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe);
332 status = -EIO;
333 goto bail;
334 }
335
336 mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode),
Mark Fashehb06970532006-03-03 10:24:33 -0800337 "Inode %llu, inode i_size = %lld != di "
338 "i_size = %llu, i_flags = 0x%x\n",
339 (unsigned long long)OCFS2_I(inode)->ip_blkno,
Mark Fashehccd979b2005-12-15 14:31:24 -0800340 i_size_read(inode),
Mark Fashehb06970532006-03-03 10:24:33 -0800341 (unsigned long long)le64_to_cpu(fe->i_size),
342 le32_to_cpu(fe->i_flags));
Mark Fashehccd979b2005-12-15 14:31:24 -0800343
344 if (new_i_size > le64_to_cpu(fe->i_size)) {
Mark Fashehb06970532006-03-03 10:24:33 -0800345 mlog(0, "asked to truncate file with size (%llu) to size (%llu)!\n",
346 (unsigned long long)le64_to_cpu(fe->i_size),
347 (unsigned long long)new_i_size);
Mark Fashehccd979b2005-12-15 14:31:24 -0800348 status = -EINVAL;
349 mlog_errno(status);
350 goto bail;
351 }
352
Mark Fashehb06970532006-03-03 10:24:33 -0800353 mlog(0, "inode %llu, i_size = %llu, new_i_size = %llu\n",
354 (unsigned long long)le64_to_cpu(fe->i_blkno),
355 (unsigned long long)le64_to_cpu(fe->i_size),
356 (unsigned long long)new_i_size);
Mark Fashehccd979b2005-12-15 14:31:24 -0800357
358 /* lets handle the simple truncate cases before doing any more
359 * cluster locking. */
360 if (new_i_size == le64_to_cpu(fe->i_size))
361 goto bail;
362
Mark Fasheh2e89b2e2007-05-09 13:40:18 -0700363 down_write(&OCFS2_I(inode)->ip_alloc_sem);
364
Mark Fashehab0920c2006-03-16 15:06:37 -0800365 /* This forces other nodes to sync and drop their pages. Do
366 * this even if we have a truncate without allocation change -
367 * ocfs2 cluster sizes can be much greater than page size, so
368 * we have to truncate them anyway. */
369 status = ocfs2_data_lock(inode, 1);
370 if (status < 0) {
Mark Fasheh2e89b2e2007-05-09 13:40:18 -0700371 up_write(&OCFS2_I(inode)->ip_alloc_sem);
372
Mark Fashehab0920c2006-03-16 15:06:37 -0800373 mlog_errno(status);
374 goto bail;
375 }
Mark Fashehab0920c2006-03-16 15:06:37 -0800376
Mark Fasheh2e89b2e2007-05-09 13:40:18 -0700377 unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1);
378 truncate_inode_pages(inode->i_mapping, new_i_size);
379
Mark Fashehccd979b2005-12-15 14:31:24 -0800380 /* alright, we're going to need to do a full blown alloc size
381 * change. Orphan the inode so that recovery can complete the
382 * truncate if necessary. This does the task of marking
383 * i_size. */
384 status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
385 if (status < 0) {
386 mlog_errno(status);
Mark Fasheh60b11392007-02-16 11:46:50 -0800387 goto bail_unlock_data;
Mark Fashehccd979b2005-12-15 14:31:24 -0800388 }
389
390 status = ocfs2_prepare_truncate(osb, inode, di_bh, &tc);
391 if (status < 0) {
392 mlog_errno(status);
Mark Fasheh60b11392007-02-16 11:46:50 -0800393 goto bail_unlock_data;
Mark Fashehccd979b2005-12-15 14:31:24 -0800394 }
395
396 status = ocfs2_commit_truncate(osb, inode, di_bh, tc);
397 if (status < 0) {
398 mlog_errno(status);
Mark Fasheh60b11392007-02-16 11:46:50 -0800399 goto bail_unlock_data;
Mark Fashehccd979b2005-12-15 14:31:24 -0800400 }
401
402 /* TODO: orphan dir cleanup here. */
Mark Fasheh60b11392007-02-16 11:46:50 -0800403bail_unlock_data:
404 ocfs2_data_unlock(inode, 1);
405
Mark Fasheh2e89b2e2007-05-09 13:40:18 -0700406 up_write(&OCFS2_I(inode)->ip_alloc_sem);
407
Mark Fashehccd979b2005-12-15 14:31:24 -0800408bail:
409
410 mlog_exit(status);
411 return status;
412}
413
414/*
415 * extend allocation only here.
416 * we'll update all the disk stuff, and oip->alloc_size
417 *
418 * expect stuff to be locked, a transaction started and enough data /
419 * metadata reservations in the contexts.
420 *
421 * Will return -EAGAIN, and a reason if a restart is needed.
422 * If passed in, *reason will always be set, even in error.
423 */
424int ocfs2_do_extend_allocation(struct ocfs2_super *osb,
425 struct inode *inode,
Mark Fashehdcd05382007-01-16 11:32:23 -0800426 u32 *logical_offset,
Mark Fashehccd979b2005-12-15 14:31:24 -0800427 u32 clusters_to_add,
Mark Fasheh2ae99a62007-03-09 16:43:28 -0800428 int mark_unwritten,
Mark Fashehccd979b2005-12-15 14:31:24 -0800429 struct buffer_head *fe_bh,
Mark Fasheh1fabe142006-10-09 18:11:45 -0700430 handle_t *handle,
Mark Fashehccd979b2005-12-15 14:31:24 -0800431 struct ocfs2_alloc_context *data_ac,
432 struct ocfs2_alloc_context *meta_ac,
433 enum ocfs2_alloc_restarted *reason_ret)
434{
435 int status = 0;
436 int free_extents;
437 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
438 enum ocfs2_alloc_restarted reason = RESTART_NONE;
439 u32 bit_off, num_bits;
440 u64 block;
Mark Fasheh2ae99a62007-03-09 16:43:28 -0800441 u8 flags = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -0800442
443 BUG_ON(!clusters_to_add);
444
Mark Fasheh2ae99a62007-03-09 16:43:28 -0800445 if (mark_unwritten)
446 flags = OCFS2_EXT_UNWRITTEN;
447
Mark Fashehccd979b2005-12-15 14:31:24 -0800448 free_extents = ocfs2_num_free_extents(osb, inode, fe);
449 if (free_extents < 0) {
450 status = free_extents;
451 mlog_errno(status);
452 goto leave;
453 }
454
455 /* there are two cases which could cause us to EAGAIN in the
456 * we-need-more-metadata case:
457 * 1) we haven't reserved *any*
458 * 2) we are so fragmented, we've needed to add metadata too
459 * many times. */
460 if (!free_extents && !meta_ac) {
461 mlog(0, "we haven't reserved any metadata!\n");
462 status = -EAGAIN;
463 reason = RESTART_META;
464 goto leave;
465 } else if ((!free_extents)
466 && (ocfs2_alloc_context_bits_left(meta_ac)
467 < ocfs2_extend_meta_needed(fe))) {
468 mlog(0, "filesystem is really fragmented...\n");
469 status = -EAGAIN;
470 reason = RESTART_META;
471 goto leave;
472 }
473
474 status = ocfs2_claim_clusters(osb, handle, data_ac, 1,
475 &bit_off, &num_bits);
476 if (status < 0) {
477 if (status != -ENOSPC)
478 mlog_errno(status);
479 goto leave;
480 }
481
482 BUG_ON(num_bits > clusters_to_add);
483
484 /* reserve our write early -- insert_extent may update the inode */
485 status = ocfs2_journal_access(handle, inode, fe_bh,
486 OCFS2_JOURNAL_ACCESS_WRITE);
487 if (status < 0) {
488 mlog_errno(status);
489 goto leave;
490 }
491
492 block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
Mark Fashehb06970532006-03-03 10:24:33 -0800493 mlog(0, "Allocating %u clusters at block %u for inode %llu\n",
494 num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
Mark Fashehdcd05382007-01-16 11:32:23 -0800495 status = ocfs2_insert_extent(osb, handle, inode, fe_bh,
496 *logical_offset, block, num_bits,
Mark Fasheh2ae99a62007-03-09 16:43:28 -0800497 flags, meta_ac);
Mark Fashehccd979b2005-12-15 14:31:24 -0800498 if (status < 0) {
499 mlog_errno(status);
500 goto leave;
501 }
502
Mark Fashehccd979b2005-12-15 14:31:24 -0800503 status = ocfs2_journal_dirty(handle, fe_bh);
504 if (status < 0) {
505 mlog_errno(status);
506 goto leave;
507 }
508
509 clusters_to_add -= num_bits;
Mark Fashehdcd05382007-01-16 11:32:23 -0800510 *logical_offset += num_bits;
Mark Fashehccd979b2005-12-15 14:31:24 -0800511
512 if (clusters_to_add) {
513 mlog(0, "need to alloc once more, clusters = %u, wanted = "
514 "%u\n", fe->i_clusters, clusters_to_add);
515 status = -EAGAIN;
516 reason = RESTART_TRANS;
517 }
518
519leave:
520 mlog_exit(status);
521 if (reason_ret)
522 *reason_ret = reason;
523 return status;
524}
525
Mark Fashehabf8b152007-01-17 13:07:24 -0800526/*
527 * For a given allocation, determine which allocators will need to be
528 * accessed, and lock them, reserving the appropriate number of bits.
529 *
Mark Fasheh2ae99a62007-03-09 16:43:28 -0800530 * Sparse file systems call this from ocfs2_write_begin_nolock()
531 * and ocfs2_allocate_unwritten_extents().
532 *
533 * File systems which don't support holes call this from
534 * ocfs2_extend_allocation().
Mark Fashehabf8b152007-01-17 13:07:24 -0800535 */
Mark Fasheh9517bac2007-02-09 20:24:12 -0800536int ocfs2_lock_allocators(struct inode *inode, struct ocfs2_dinode *di,
Mark Fashehb27b7cb2007-06-18 11:22:56 -0700537 u32 clusters_to_add, u32 extents_to_split,
Mark Fasheh9517bac2007-02-09 20:24:12 -0800538 struct ocfs2_alloc_context **data_ac,
539 struct ocfs2_alloc_context **meta_ac)
Mark Fashehabf8b152007-01-17 13:07:24 -0800540{
541 int ret, num_free_extents;
Mark Fashehb27b7cb2007-06-18 11:22:56 -0700542 unsigned int max_recs_needed = clusters_to_add + 2 * extents_to_split;
Mark Fashehabf8b152007-01-17 13:07:24 -0800543 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
544
545 *meta_ac = NULL;
546 *data_ac = NULL;
547
548 mlog(0, "extend inode %llu, i_size = %lld, di->i_clusters = %u, "
Mark Fashehb27b7cb2007-06-18 11:22:56 -0700549 "clusters_to_add = %u, extents_to_split = %u\n",
Mark Fashehabf8b152007-01-17 13:07:24 -0800550 (unsigned long long)OCFS2_I(inode)->ip_blkno, i_size_read(inode),
Mark Fashehb27b7cb2007-06-18 11:22:56 -0700551 le32_to_cpu(di->i_clusters), clusters_to_add, extents_to_split);
Mark Fashehabf8b152007-01-17 13:07:24 -0800552
553 num_free_extents = ocfs2_num_free_extents(osb, inode, di);
554 if (num_free_extents < 0) {
555 ret = num_free_extents;
556 mlog_errno(ret);
557 goto out;
558 }
559
560 /*
561 * Sparse allocation file systems need to be more conservative
562 * with reserving room for expansion - the actual allocation
563 * happens while we've got a journal handle open so re-taking
564 * a cluster lock (because we ran out of room for another
565 * extent) will violate ordering rules.
566 *
Mark Fasheh9517bac2007-02-09 20:24:12 -0800567 * Most of the time we'll only be seeing this 1 cluster at a time
Mark Fashehabf8b152007-01-17 13:07:24 -0800568 * anyway.
Mark Fashehb27b7cb2007-06-18 11:22:56 -0700569 *
570 * Always lock for any unwritten extents - we might want to
571 * add blocks during a split.
Mark Fashehabf8b152007-01-17 13:07:24 -0800572 */
573 if (!num_free_extents ||
Mark Fashehb27b7cb2007-06-18 11:22:56 -0700574 (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed)) {
Mark Fashehabf8b152007-01-17 13:07:24 -0800575 ret = ocfs2_reserve_new_metadata(osb, di, meta_ac);
576 if (ret < 0) {
577 if (ret != -ENOSPC)
578 mlog_errno(ret);
579 goto out;
580 }
581 }
582
583 ret = ocfs2_reserve_clusters(osb, clusters_to_add, data_ac);
584 if (ret < 0) {
585 if (ret != -ENOSPC)
586 mlog_errno(ret);
587 goto out;
588 }
589
590out:
591 if (ret) {
592 if (*meta_ac) {
593 ocfs2_free_alloc_context(*meta_ac);
594 *meta_ac = NULL;
595 }
596
597 /*
598 * We cannot have an error and a non null *data_ac.
599 */
600 }
601
602 return ret;
603}
604
Mark Fasheh2ae99a62007-03-09 16:43:28 -0800605static int __ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
606 u32 clusters_to_add, int mark_unwritten)
Mark Fashehccd979b2005-12-15 14:31:24 -0800607{
608 int status = 0;
609 int restart_func = 0;
Mark Fashehabf8b152007-01-17 13:07:24 -0800610 int credits;
Mark Fasheh2ae99a62007-03-09 16:43:28 -0800611 u32 prev_clusters;
Mark Fashehccd979b2005-12-15 14:31:24 -0800612 struct buffer_head *bh = NULL;
613 struct ocfs2_dinode *fe = NULL;
Mark Fasheh1fabe142006-10-09 18:11:45 -0700614 handle_t *handle = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800615 struct ocfs2_alloc_context *data_ac = NULL;
616 struct ocfs2_alloc_context *meta_ac = NULL;
617 enum ocfs2_alloc_restarted why;
618 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
619
620 mlog_entry("(clusters_to_add = %u)\n", clusters_to_add);
621
Mark Fashehdcd05382007-01-16 11:32:23 -0800622 /*
623 * This function only exists for file systems which don't
624 * support holes.
625 */
Mark Fasheh2ae99a62007-03-09 16:43:28 -0800626 BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb));
Mark Fashehdcd05382007-01-16 11:32:23 -0800627
Mark Fashehccd979b2005-12-15 14:31:24 -0800628 status = ocfs2_read_block(osb, OCFS2_I(inode)->ip_blkno, &bh,
629 OCFS2_BH_CACHED, inode);
630 if (status < 0) {
631 mlog_errno(status);
632 goto leave;
633 }
634
635 fe = (struct ocfs2_dinode *) bh->b_data;
636 if (!OCFS2_IS_VALID_DINODE(fe)) {
637 OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe);
638 status = -EIO;
639 goto leave;
640 }
641
642restart_all:
643 BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
644
Mark Fashehb27b7cb2007-06-18 11:22:56 -0700645 status = ocfs2_lock_allocators(inode, fe, clusters_to_add, 0, &data_ac,
Mark Fasheh9517bac2007-02-09 20:24:12 -0800646 &meta_ac);
647 if (status) {
648 mlog_errno(status);
649 goto leave;
650 }
651
Mark Fashehccd979b2005-12-15 14:31:24 -0800652 credits = ocfs2_calc_extend_credits(osb->sb, fe, clusters_to_add);
Mark Fasheh65eff9c2006-10-09 17:26:22 -0700653 handle = ocfs2_start_trans(osb, credits);
Mark Fashehccd979b2005-12-15 14:31:24 -0800654 if (IS_ERR(handle)) {
655 status = PTR_ERR(handle);
656 handle = NULL;
657 mlog_errno(status);
658 goto leave;
659 }
660
661restarted_transaction:
662 /* reserve a write to the file entry early on - that we if we
663 * run out of credits in the allocation path, we can still
664 * update i_size. */
665 status = ocfs2_journal_access(handle, inode, bh,
666 OCFS2_JOURNAL_ACCESS_WRITE);
667 if (status < 0) {
668 mlog_errno(status);
669 goto leave;
670 }
671
672 prev_clusters = OCFS2_I(inode)->ip_clusters;
673
674 status = ocfs2_do_extend_allocation(osb,
675 inode,
Mark Fashehdcd05382007-01-16 11:32:23 -0800676 &logical_start,
Mark Fashehccd979b2005-12-15 14:31:24 -0800677 clusters_to_add,
Mark Fasheh2ae99a62007-03-09 16:43:28 -0800678 mark_unwritten,
Mark Fashehccd979b2005-12-15 14:31:24 -0800679 bh,
680 handle,
681 data_ac,
682 meta_ac,
683 &why);
684 if ((status < 0) && (status != -EAGAIN)) {
685 if (status != -ENOSPC)
686 mlog_errno(status);
687 goto leave;
688 }
689
690 status = ocfs2_journal_dirty(handle, bh);
691 if (status < 0) {
692 mlog_errno(status);
693 goto leave;
694 }
695
696 spin_lock(&OCFS2_I(inode)->ip_lock);
697 clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
698 spin_unlock(&OCFS2_I(inode)->ip_lock);
699
700 if (why != RESTART_NONE && clusters_to_add) {
701 if (why == RESTART_META) {
702 mlog(0, "restarting function.\n");
703 restart_func = 1;
704 } else {
705 BUG_ON(why != RESTART_TRANS);
706
707 mlog(0, "restarting transaction.\n");
708 /* TODO: This can be more intelligent. */
709 credits = ocfs2_calc_extend_credits(osb->sb,
710 fe,
711 clusters_to_add);
Mark Fasheh1fabe142006-10-09 18:11:45 -0700712 status = ocfs2_extend_trans(handle, credits);
Mark Fashehccd979b2005-12-15 14:31:24 -0800713 if (status < 0) {
714 /* handle still has to be committed at
715 * this point. */
716 status = -ENOMEM;
717 mlog_errno(status);
718 goto leave;
719 }
720 goto restarted_transaction;
721 }
722 }
723
Mark Fashehb06970532006-03-03 10:24:33 -0800724 mlog(0, "fe: i_clusters = %u, i_size=%llu\n",
Mark Fasheh1ca1a112007-04-27 16:01:25 -0700725 le32_to_cpu(fe->i_clusters),
726 (unsigned long long)le64_to_cpu(fe->i_size));
Mark Fashehccd979b2005-12-15 14:31:24 -0800727 mlog(0, "inode: ip_clusters=%u, i_size=%lld\n",
728 OCFS2_I(inode)->ip_clusters, i_size_read(inode));
729
730leave:
Mark Fashehccd979b2005-12-15 14:31:24 -0800731 if (handle) {
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700732 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800733 handle = NULL;
734 }
735 if (data_ac) {
736 ocfs2_free_alloc_context(data_ac);
737 data_ac = NULL;
738 }
739 if (meta_ac) {
740 ocfs2_free_alloc_context(meta_ac);
741 meta_ac = NULL;
742 }
743 if ((!status) && restart_func) {
744 restart_func = 0;
745 goto restart_all;
746 }
747 if (bh) {
748 brelse(bh);
749 bh = NULL;
750 }
751
752 mlog_exit(status);
753 return status;
754}
755
Mark Fasheh2ae99a62007-03-09 16:43:28 -0800756static int ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
757 u32 clusters_to_add, int mark_unwritten)
758{
759 int ret;
760
761 /*
762 * The alloc sem blocks peope in read/write from reading our
763 * allocation until we're done changing it. We depend on
764 * i_mutex to block other extend/truncate calls while we're
765 * here.
766 */
767 down_write(&OCFS2_I(inode)->ip_alloc_sem);
768 ret = __ocfs2_extend_allocation(inode, logical_start, clusters_to_add,
769 mark_unwritten);
770 up_write(&OCFS2_I(inode)->ip_alloc_sem);
771
772 return ret;
773}
774
Mark Fashehccd979b2005-12-15 14:31:24 -0800775/* Some parts of this taken from generic_cont_expand, which turned out
776 * to be too fragile to do exactly what we need without us having to
Mark Fasheh53013cb2006-05-05 19:04:03 -0700777 * worry about recursive locking in ->prepare_write() and
778 * ->commit_write(). */
Mark Fashehccd979b2005-12-15 14:31:24 -0800779static int ocfs2_write_zero_page(struct inode *inode,
780 u64 size)
781{
782 struct address_space *mapping = inode->i_mapping;
783 struct page *page;
784 unsigned long index;
785 unsigned int offset;
Mark Fasheh1fabe142006-10-09 18:11:45 -0700786 handle_t *handle = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800787 int ret;
788
789 offset = (size & (PAGE_CACHE_SIZE-1)); /* Within page */
790 /* ugh. in prepare/commit_write, if from==to==start of block, we
791 ** skip the prepare. make sure we never send an offset for the start
792 ** of a block
793 */
794 if ((offset & (inode->i_sb->s_blocksize - 1)) == 0) {
795 offset++;
796 }
797 index = size >> PAGE_CACHE_SHIFT;
798
799 page = grab_cache_page(mapping, index);
800 if (!page) {
801 ret = -ENOMEM;
802 mlog_errno(ret);
803 goto out;
804 }
805
Mark Fasheh53013cb2006-05-05 19:04:03 -0700806 ret = ocfs2_prepare_write_nolock(inode, page, offset, offset);
Mark Fashehccd979b2005-12-15 14:31:24 -0800807 if (ret < 0) {
808 mlog_errno(ret);
809 goto out_unlock;
810 }
811
812 if (ocfs2_should_order_data(inode)) {
813 handle = ocfs2_start_walk_page_trans(inode, page, offset,
814 offset);
815 if (IS_ERR(handle)) {
816 ret = PTR_ERR(handle);
817 handle = NULL;
818 goto out_unlock;
819 }
820 }
821
822 /* must not update i_size! */
823 ret = block_commit_write(page, offset, offset);
824 if (ret < 0)
825 mlog_errno(ret);
826 else
827 ret = 0;
828
829 if (handle)
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700830 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800831out_unlock:
832 unlock_page(page);
833 page_cache_release(page);
834out:
835 return ret;
836}
837
838static int ocfs2_zero_extend(struct inode *inode,
839 u64 zero_to_size)
840{
841 int ret = 0;
842 u64 start_off;
843 struct super_block *sb = inode->i_sb;
844
845 start_off = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
846 while (start_off < zero_to_size) {
847 ret = ocfs2_write_zero_page(inode, start_off);
848 if (ret < 0) {
849 mlog_errno(ret);
850 goto out;
851 }
852
853 start_off += sb->s_blocksize;
Mark Fashehe2057c52006-10-03 17:53:05 -0700854
855 /*
856 * Very large extends have the potential to lock up
857 * the cpu for extended periods of time.
858 */
859 cond_resched();
Mark Fashehccd979b2005-12-15 14:31:24 -0800860 }
861
862out:
863 return ret;
864}
865
Mark Fasheh53013cb2006-05-05 19:04:03 -0700866/*
867 * A tail_to_skip value > 0 indicates that we're being called from
868 * ocfs2_file_aio_write(). This has the following implications:
869 *
870 * - we don't want to update i_size
871 * - di_bh will be NULL, which is fine because it's only used in the
872 * case where we want to update i_size.
873 * - ocfs2_zero_extend() will then only be filling the hole created
874 * between i_size and the start of the write.
875 */
Mark Fashehccd979b2005-12-15 14:31:24 -0800876static int ocfs2_extend_file(struct inode *inode,
877 struct buffer_head *di_bh,
Mark Fasheh53013cb2006-05-05 19:04:03 -0700878 u64 new_i_size,
879 size_t tail_to_skip)
Mark Fashehccd979b2005-12-15 14:31:24 -0800880{
881 int ret = 0;
Mark Fasheh3a0782d2007-01-17 12:53:31 -0800882 u32 clusters_to_add = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -0800883
Mark Fasheh53013cb2006-05-05 19:04:03 -0700884 BUG_ON(!tail_to_skip && !di_bh);
885
Mark Fashehccd979b2005-12-15 14:31:24 -0800886 /* setattr sometimes calls us like this. */
887 if (new_i_size == 0)
888 goto out;
889
890 if (i_size_read(inode) == new_i_size)
891 goto out;
892 BUG_ON(new_i_size < i_size_read(inode));
893
Mark Fasheh3a0782d2007-01-17 12:53:31 -0800894 if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))) {
895 BUG_ON(tail_to_skip != 0);
896 goto out_update_size;
897 }
898
Mark Fashehccd979b2005-12-15 14:31:24 -0800899 clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size) -
900 OCFS2_I(inode)->ip_clusters;
901
Mark Fasheh0effef72006-10-03 17:44:42 -0700902 /*
903 * protect the pages that ocfs2_zero_extend is going to be
904 * pulling into the page cache.. we do this before the
905 * metadata extend so that we don't get into the situation
906 * where we've extended the metadata but can't get the data
907 * lock to zero.
908 */
909 ret = ocfs2_data_lock(inode, 1);
910 if (ret < 0) {
911 mlog_errno(ret);
912 goto out;
913 }
Mark Fasheh53013cb2006-05-05 19:04:03 -0700914
Mark Fasheh0effef72006-10-03 17:44:42 -0700915 if (clusters_to_add) {
Mark Fasheh2ae99a62007-03-09 16:43:28 -0800916 ret = ocfs2_extend_allocation(inode,
917 OCFS2_I(inode)->ip_clusters,
918 clusters_to_add, 0);
Mark Fashehccd979b2005-12-15 14:31:24 -0800919 if (ret < 0) {
920 mlog_errno(ret);
Mark Fasheh53013cb2006-05-05 19:04:03 -0700921 goto out_unlock;
Mark Fashehccd979b2005-12-15 14:31:24 -0800922 }
Mark Fasheh0effef72006-10-03 17:44:42 -0700923 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800924
Mark Fasheh0effef72006-10-03 17:44:42 -0700925 /*
926 * Call this even if we don't add any clusters to the tree. We
927 * still need to zero the area between the old i_size and the
928 * new i_size.
929 */
930 ret = ocfs2_zero_extend(inode, (u64)new_i_size - tail_to_skip);
931 if (ret < 0) {
932 mlog_errno(ret);
933 goto out_unlock;
Mark Fashehccd979b2005-12-15 14:31:24 -0800934 }
935
Mark Fasheh3a0782d2007-01-17 12:53:31 -0800936out_update_size:
Mark Fasheh53013cb2006-05-05 19:04:03 -0700937 if (!tail_to_skip) {
938 /* We're being called from ocfs2_setattr() which wants
939 * us to update i_size */
940 ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
941 if (ret < 0)
942 mlog_errno(ret);
943 }
944
945out_unlock:
Mark Fasheh3a0782d2007-01-17 12:53:31 -0800946 if (!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
947 ocfs2_data_unlock(inode, 1);
Mark Fasheh53013cb2006-05-05 19:04:03 -0700948
Mark Fashehccd979b2005-12-15 14:31:24 -0800949out:
950 return ret;
951}
952
953int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
954{
955 int status = 0, size_change;
956 struct inode *inode = dentry->d_inode;
957 struct super_block *sb = inode->i_sb;
958 struct ocfs2_super *osb = OCFS2_SB(sb);
959 struct buffer_head *bh = NULL;
Mark Fasheh1fabe142006-10-09 18:11:45 -0700960 handle_t *handle = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800961
962 mlog_entry("(0x%p, '%.*s')\n", dentry,
963 dentry->d_name.len, dentry->d_name.name);
964
965 if (attr->ia_valid & ATTR_MODE)
966 mlog(0, "mode change: %d\n", attr->ia_mode);
967 if (attr->ia_valid & ATTR_UID)
968 mlog(0, "uid change: %d\n", attr->ia_uid);
969 if (attr->ia_valid & ATTR_GID)
970 mlog(0, "gid change: %d\n", attr->ia_gid);
971 if (attr->ia_valid & ATTR_SIZE)
972 mlog(0, "size change...\n");
973 if (attr->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME))
974 mlog(0, "time change...\n");
975
976#define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
977 | ATTR_GID | ATTR_UID | ATTR_MODE)
978 if (!(attr->ia_valid & OCFS2_VALID_ATTRS)) {
979 mlog(0, "can't handle attrs: 0x%x\n", attr->ia_valid);
980 return 0;
981 }
982
983 status = inode_change_ok(inode, attr);
984 if (status)
985 return status;
986
987 size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
988 if (size_change) {
989 status = ocfs2_rw_lock(inode, 1);
990 if (status < 0) {
991 mlog_errno(status);
992 goto bail;
993 }
994 }
995
Mark Fasheh4bcec182006-10-09 16:02:40 -0700996 status = ocfs2_meta_lock(inode, &bh, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -0800997 if (status < 0) {
998 if (status != -ENOENT)
999 mlog_errno(status);
1000 goto bail_unlock_rw;
1001 }
1002
1003 if (size_change && attr->ia_size != i_size_read(inode)) {
1004 if (i_size_read(inode) > attr->ia_size)
1005 status = ocfs2_truncate_file(inode, bh, attr->ia_size);
1006 else
Mark Fasheh53013cb2006-05-05 19:04:03 -07001007 status = ocfs2_extend_file(inode, bh, attr->ia_size, 0);
Mark Fashehccd979b2005-12-15 14:31:24 -08001008 if (status < 0) {
1009 if (status != -ENOSPC)
1010 mlog_errno(status);
1011 status = -ENOSPC;
1012 goto bail_unlock;
1013 }
1014 }
1015
Mark Fasheh65eff9c2006-10-09 17:26:22 -07001016 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
Mark Fashehccd979b2005-12-15 14:31:24 -08001017 if (IS_ERR(handle)) {
1018 status = PTR_ERR(handle);
1019 mlog_errno(status);
1020 goto bail_unlock;
1021 }
1022
Mark Fasheh7307de82007-05-09 15:16:19 -07001023 /*
1024 * This will intentionally not wind up calling vmtruncate(),
1025 * since all the work for a size change has been done above.
1026 * Otherwise, we could get into problems with truncate as
1027 * ip_alloc_sem is used there to protect against i_size
1028 * changes.
1029 */
Mark Fashehccd979b2005-12-15 14:31:24 -08001030 status = inode_setattr(inode, attr);
1031 if (status < 0) {
1032 mlog_errno(status);
1033 goto bail_commit;
1034 }
1035
1036 status = ocfs2_mark_inode_dirty(handle, inode, bh);
1037 if (status < 0)
1038 mlog_errno(status);
1039
1040bail_commit:
Mark Fasheh02dc1af2006-10-09 16:48:10 -07001041 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -08001042bail_unlock:
1043 ocfs2_meta_unlock(inode, 1);
1044bail_unlock_rw:
1045 if (size_change)
1046 ocfs2_rw_unlock(inode, 1);
1047bail:
1048 if (bh)
1049 brelse(bh);
1050
1051 mlog_exit(status);
1052 return status;
1053}
1054
1055int ocfs2_getattr(struct vfsmount *mnt,
1056 struct dentry *dentry,
1057 struct kstat *stat)
1058{
1059 struct inode *inode = dentry->d_inode;
1060 struct super_block *sb = dentry->d_inode->i_sb;
1061 struct ocfs2_super *osb = sb->s_fs_info;
1062 int err;
1063
1064 mlog_entry_void();
1065
1066 err = ocfs2_inode_revalidate(dentry);
1067 if (err) {
1068 if (err != -ENOENT)
1069 mlog_errno(err);
1070 goto bail;
1071 }
1072
1073 generic_fillattr(inode, stat);
1074
1075 /* We set the blksize from the cluster size for performance */
1076 stat->blksize = osb->s_clustersize;
1077
1078bail:
1079 mlog_exit(err);
1080
1081 return err;
1082}
1083
Tiger Yangd38eb8d2006-11-27 09:59:21 +08001084int ocfs2_permission(struct inode *inode, int mask, struct nameidata *nd)
1085{
1086 int ret;
1087
1088 mlog_entry_void();
1089
1090 ret = ocfs2_meta_lock(inode, NULL, 0);
1091 if (ret) {
Mark Fasheha9f5f702007-04-26 11:43:43 -07001092 if (ret != -ENOENT)
1093 mlog_errno(ret);
Tiger Yangd38eb8d2006-11-27 09:59:21 +08001094 goto out;
1095 }
1096
1097 ret = generic_permission(inode, mask, NULL);
Tiger Yangd38eb8d2006-11-27 09:59:21 +08001098
1099 ocfs2_meta_unlock(inode, 0);
1100out:
1101 mlog_exit(ret);
1102 return ret;
1103}
1104
Mark Fashehccd979b2005-12-15 14:31:24 -08001105static int ocfs2_write_remove_suid(struct inode *inode)
1106{
1107 int ret;
1108 struct buffer_head *bh = NULL;
1109 struct ocfs2_inode_info *oi = OCFS2_I(inode);
Mark Fasheh1fabe142006-10-09 18:11:45 -07001110 handle_t *handle;
Mark Fashehccd979b2005-12-15 14:31:24 -08001111 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1112 struct ocfs2_dinode *di;
1113
Mark Fashehb06970532006-03-03 10:24:33 -08001114 mlog_entry("(Inode %llu, mode 0%o)\n",
1115 (unsigned long long)oi->ip_blkno, inode->i_mode);
Mark Fashehccd979b2005-12-15 14:31:24 -08001116
Mark Fasheh65eff9c2006-10-09 17:26:22 -07001117 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
Mark Fashehccd979b2005-12-15 14:31:24 -08001118 if (handle == NULL) {
1119 ret = -ENOMEM;
1120 mlog_errno(ret);
1121 goto out;
1122 }
1123
1124 ret = ocfs2_read_block(osb, oi->ip_blkno, &bh, OCFS2_BH_CACHED, inode);
1125 if (ret < 0) {
1126 mlog_errno(ret);
1127 goto out_trans;
1128 }
1129
1130 ret = ocfs2_journal_access(handle, inode, bh,
1131 OCFS2_JOURNAL_ACCESS_WRITE);
1132 if (ret < 0) {
1133 mlog_errno(ret);
1134 goto out_bh;
1135 }
1136
1137 inode->i_mode &= ~S_ISUID;
1138 if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
1139 inode->i_mode &= ~S_ISGID;
1140
1141 di = (struct ocfs2_dinode *) bh->b_data;
1142 di->i_mode = cpu_to_le16(inode->i_mode);
1143
1144 ret = ocfs2_journal_dirty(handle, bh);
1145 if (ret < 0)
1146 mlog_errno(ret);
1147out_bh:
1148 brelse(bh);
1149out_trans:
Mark Fasheh02dc1af2006-10-09 16:48:10 -07001150 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -08001151out:
1152 mlog_exit(ret);
1153 return ret;
1154}
1155
Mark Fasheh9517bac2007-02-09 20:24:12 -08001156/*
1157 * Will look for holes and unwritten extents in the range starting at
1158 * pos for count bytes (inclusive).
1159 */
1160static int ocfs2_check_range_for_holes(struct inode *inode, loff_t pos,
1161 size_t count)
1162{
1163 int ret = 0;
Mark Fasheh49cb8d22007-03-09 16:21:46 -08001164 unsigned int extent_flags;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001165 u32 cpos, clusters, extent_len, phys_cpos;
1166 struct super_block *sb = inode->i_sb;
1167
1168 cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
1169 clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
1170
1171 while (clusters) {
Mark Fasheh49cb8d22007-03-09 16:21:46 -08001172 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
1173 &extent_flags);
Mark Fasheh9517bac2007-02-09 20:24:12 -08001174 if (ret < 0) {
1175 mlog_errno(ret);
1176 goto out;
1177 }
1178
Mark Fasheh49cb8d22007-03-09 16:21:46 -08001179 if (phys_cpos == 0 || (extent_flags & OCFS2_EXT_UNWRITTEN)) {
Mark Fasheh9517bac2007-02-09 20:24:12 -08001180 ret = 1;
1181 break;
1182 }
1183
1184 if (extent_len > clusters)
1185 extent_len = clusters;
1186
1187 clusters -= extent_len;
1188 cpos += extent_len;
1189 }
1190out:
1191 return ret;
1192}
1193
Mark Fasheh2ae99a62007-03-09 16:43:28 -08001194/*
1195 * Allocate enough extents to cover the region starting at byte offset
1196 * start for len bytes. Existing extents are skipped, any extents
1197 * added are marked as "unwritten".
1198 */
1199static int ocfs2_allocate_unwritten_extents(struct inode *inode,
1200 u64 start, u64 len)
1201{
1202 int ret;
1203 u32 cpos, phys_cpos, clusters, alloc_size;
1204
1205 /*
1206 * We consider both start and len to be inclusive.
1207 */
1208 cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
1209 clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len);
1210 clusters -= cpos;
1211
1212 while (clusters) {
1213 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1214 &alloc_size, NULL);
1215 if (ret) {
1216 mlog_errno(ret);
1217 goto out;
1218 }
1219
1220 /*
1221 * Hole or existing extent len can be arbitrary, so
1222 * cap it to our own allocation request.
1223 */
1224 if (alloc_size > clusters)
1225 alloc_size = clusters;
1226
1227 if (phys_cpos) {
1228 /*
1229 * We already have an allocation at this
1230 * region so we can safely skip it.
1231 */
1232 goto next;
1233 }
1234
1235 ret = __ocfs2_extend_allocation(inode, cpos, alloc_size, 1);
1236 if (ret) {
1237 if (ret != -ENOSPC)
1238 mlog_errno(ret);
1239 goto out;
1240 }
1241
1242next:
1243 cpos += alloc_size;
1244 clusters -= alloc_size;
1245 }
1246
1247 ret = 0;
1248out:
1249 return ret;
1250}
1251
Tiger Yang8659ac22006-10-17 18:29:52 -07001252static int ocfs2_prepare_inode_for_write(struct dentry *dentry,
1253 loff_t *ppos,
1254 size_t count,
Mark Fasheh9517bac2007-02-09 20:24:12 -08001255 int appending,
1256 int *direct_io)
Mark Fashehccd979b2005-12-15 14:31:24 -08001257{
Tiger Yang8659ac22006-10-17 18:29:52 -07001258 int ret = 0, meta_level = appending;
1259 struct inode *inode = dentry->d_inode;
Mark Fashehccd979b2005-12-15 14:31:24 -08001260 u32 clusters;
Mark Fashehccd979b2005-12-15 14:31:24 -08001261 loff_t newsize, saved_pos;
Mark Fashehccd979b2005-12-15 14:31:24 -08001262
Mark Fashehccd979b2005-12-15 14:31:24 -08001263 /*
1264 * We sample i_size under a read level meta lock to see if our write
1265 * is extending the file, if it is we back off and get a write level
1266 * meta lock.
1267 */
Mark Fashehccd979b2005-12-15 14:31:24 -08001268 for(;;) {
Mark Fasheh4bcec182006-10-09 16:02:40 -07001269 ret = ocfs2_meta_lock(inode, NULL, meta_level);
Mark Fashehccd979b2005-12-15 14:31:24 -08001270 if (ret < 0) {
1271 meta_level = -1;
1272 mlog_errno(ret);
1273 goto out;
1274 }
1275
1276 /* Clear suid / sgid if necessary. We do this here
1277 * instead of later in the write path because
1278 * remove_suid() calls ->setattr without any hint that
1279 * we may have already done our cluster locking. Since
1280 * ocfs2_setattr() *must* take cluster locks to
1281 * proceeed, this will lead us to recursively lock the
1282 * inode. There's also the dinode i_size state which
1283 * can be lost via setattr during extending writes (we
1284 * set inode->i_size at the end of a write. */
Tiger Yang8659ac22006-10-17 18:29:52 -07001285 if (should_remove_suid(dentry)) {
Mark Fashehccd979b2005-12-15 14:31:24 -08001286 if (meta_level == 0) {
1287 ocfs2_meta_unlock(inode, meta_level);
1288 meta_level = 1;
1289 continue;
1290 }
1291
1292 ret = ocfs2_write_remove_suid(inode);
1293 if (ret < 0) {
1294 mlog_errno(ret);
Tiger Yang8659ac22006-10-17 18:29:52 -07001295 goto out_unlock;
Mark Fashehccd979b2005-12-15 14:31:24 -08001296 }
1297 }
1298
1299 /* work on a copy of ppos until we're sure that we won't have
1300 * to recalculate it due to relocking. */
Tiger Yang8659ac22006-10-17 18:29:52 -07001301 if (appending) {
Mark Fashehccd979b2005-12-15 14:31:24 -08001302 saved_pos = i_size_read(inode);
1303 mlog(0, "O_APPEND: inode->i_size=%llu\n", saved_pos);
1304 } else {
Tiger Yang8659ac22006-10-17 18:29:52 -07001305 saved_pos = *ppos;
Mark Fashehccd979b2005-12-15 14:31:24 -08001306 }
Mark Fasheh3a0782d2007-01-17 12:53:31 -08001307
Mark Fasheh9517bac2007-02-09 20:24:12 -08001308 if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb))) {
1309 loff_t end = saved_pos + count;
1310
1311 /*
1312 * Skip the O_DIRECT checks if we don't need
1313 * them.
1314 */
1315 if (!direct_io || !(*direct_io))
1316 break;
1317
1318 /*
1319 * Allowing concurrent direct writes means
1320 * i_size changes wouldn't be synchronized, so
1321 * one node could wind up truncating another
1322 * nodes writes.
1323 */
1324 if (end > i_size_read(inode)) {
1325 *direct_io = 0;
1326 break;
1327 }
1328
1329 /*
1330 * We don't fill holes during direct io, so
1331 * check for them here. If any are found, the
1332 * caller will have to retake some cluster
1333 * locks and initiate the io as buffered.
1334 */
1335 ret = ocfs2_check_range_for_holes(inode, saved_pos,
1336 count);
1337 if (ret == 1) {
1338 *direct_io = 0;
1339 ret = 0;
1340 } else if (ret < 0)
1341 mlog_errno(ret);
1342 break;
1343 }
1344
Mark Fasheh3a0782d2007-01-17 12:53:31 -08001345 /*
1346 * The rest of this loop is concerned with legacy file
1347 * systems which don't support sparse files.
1348 */
Mark Fasheh3a0782d2007-01-17 12:53:31 -08001349
Tiger Yang8659ac22006-10-17 18:29:52 -07001350 newsize = count + saved_pos;
Mark Fashehccd979b2005-12-15 14:31:24 -08001351
Mark Fasheh215c7f92006-02-01 16:42:10 -08001352 mlog(0, "pos=%lld newsize=%lld cursize=%lld\n",
1353 (long long) saved_pos, (long long) newsize,
1354 (long long) i_size_read(inode));
Mark Fashehccd979b2005-12-15 14:31:24 -08001355
1356 /* No need for a higher level metadata lock if we're
1357 * never going past i_size. */
1358 if (newsize <= i_size_read(inode))
1359 break;
1360
1361 if (meta_level == 0) {
1362 ocfs2_meta_unlock(inode, meta_level);
1363 meta_level = 1;
1364 continue;
1365 }
1366
1367 spin_lock(&OCFS2_I(inode)->ip_lock);
1368 clusters = ocfs2_clusters_for_bytes(inode->i_sb, newsize) -
1369 OCFS2_I(inode)->ip_clusters;
1370 spin_unlock(&OCFS2_I(inode)->ip_lock);
1371
1372 mlog(0, "Writing at EOF, may need more allocation: "
Mark Fasheh215c7f92006-02-01 16:42:10 -08001373 "i_size = %lld, newsize = %lld, need %u clusters\n",
1374 (long long) i_size_read(inode), (long long) newsize,
1375 clusters);
Mark Fashehccd979b2005-12-15 14:31:24 -08001376
1377 /* We only want to continue the rest of this loop if
1378 * our extend will actually require more
1379 * allocation. */
1380 if (!clusters)
1381 break;
1382
Tiger Yang8659ac22006-10-17 18:29:52 -07001383 ret = ocfs2_extend_file(inode, NULL, newsize, count);
Mark Fashehccd979b2005-12-15 14:31:24 -08001384 if (ret < 0) {
1385 if (ret != -ENOSPC)
1386 mlog_errno(ret);
Tiger Yang8659ac22006-10-17 18:29:52 -07001387 goto out_unlock;
Mark Fashehccd979b2005-12-15 14:31:24 -08001388 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001389 break;
1390 }
1391
Tiger Yang8659ac22006-10-17 18:29:52 -07001392 if (appending)
1393 *ppos = saved_pos;
1394
1395out_unlock:
Mark Fashehccd979b2005-12-15 14:31:24 -08001396 ocfs2_meta_unlock(inode, meta_level);
Tiger Yang8659ac22006-10-17 18:29:52 -07001397
1398out:
1399 return ret;
1400}
1401
Mark Fasheh9517bac2007-02-09 20:24:12 -08001402static inline void
1403ocfs2_set_next_iovec(const struct iovec **iovp, size_t *basep, size_t bytes)
1404{
1405 const struct iovec *iov = *iovp;
1406 size_t base = *basep;
1407
1408 do {
1409 int copy = min(bytes, iov->iov_len - base);
1410
1411 bytes -= copy;
1412 base += copy;
1413 if (iov->iov_len == base) {
1414 iov++;
1415 base = 0;
1416 }
1417 } while (bytes);
1418 *iovp = iov;
1419 *basep = base;
1420}
1421
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001422static struct page * ocfs2_get_write_source(char **ret_src_buf,
Mark Fasheh9517bac2007-02-09 20:24:12 -08001423 const struct iovec *cur_iov,
1424 size_t iov_offset)
1425{
1426 int ret;
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001427 char *buf = cur_iov->iov_base + iov_offset;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001428 struct page *src_page = NULL;
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001429 unsigned long off;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001430
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001431 off = (unsigned long)(buf) & ~PAGE_CACHE_MASK;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001432
1433 if (!segment_eq(get_fs(), KERNEL_DS)) {
1434 /*
1435 * Pull in the user page. We want to do this outside
1436 * of the meta data locks in order to preserve locking
1437 * order in case of page fault.
1438 */
1439 ret = get_user_pages(current, current->mm,
1440 (unsigned long)buf & PAGE_CACHE_MASK, 1,
1441 0, 0, &src_page, NULL);
1442 if (ret == 1)
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001443 *ret_src_buf = kmap(src_page) + off;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001444 else
1445 src_page = ERR_PTR(-EFAULT);
1446 } else {
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001447 *ret_src_buf = buf;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001448 }
1449
1450 return src_page;
1451}
1452
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001453static void ocfs2_put_write_source(struct page *page)
Mark Fasheh9517bac2007-02-09 20:24:12 -08001454{
1455 if (page) {
1456 kunmap(page);
1457 page_cache_release(page);
1458 }
1459}
1460
1461static ssize_t ocfs2_file_buffered_write(struct file *file, loff_t *ppos,
1462 const struct iovec *iov,
1463 unsigned long nr_segs,
1464 size_t count,
1465 ssize_t o_direct_written)
1466{
1467 int ret = 0;
1468 ssize_t copied, total = 0;
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001469 size_t iov_offset = 0, bytes;
1470 loff_t pos;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001471 const struct iovec *cur_iov = iov;
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001472 struct page *user_page, *page;
1473 char *buf, *dst;
1474 void *fsdata;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001475
1476 /*
1477 * handle partial DIO write. Adjust cur_iov if needed.
1478 */
1479 ocfs2_set_next_iovec(&cur_iov, &iov_offset, o_direct_written);
1480
1481 do {
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001482 pos = *ppos;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001483
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001484 user_page = ocfs2_get_write_source(&buf, cur_iov, iov_offset);
1485 if (IS_ERR(user_page)) {
1486 ret = PTR_ERR(user_page);
Mark Fasheh9517bac2007-02-09 20:24:12 -08001487 goto out;
1488 }
1489
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001490 /* Stay within our page boundaries */
1491 bytes = min((PAGE_CACHE_SIZE - ((unsigned long)pos & ~PAGE_CACHE_MASK)),
1492 (PAGE_CACHE_SIZE - ((unsigned long)buf & ~PAGE_CACHE_MASK)));
1493 /* Stay within the vector boundary */
1494 bytes = min_t(size_t, bytes, cur_iov->iov_len - iov_offset);
1495 /* Stay within count */
1496 bytes = min(bytes, count);
Mark Fasheh9517bac2007-02-09 20:24:12 -08001497
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001498 page = NULL;
1499 ret = ocfs2_write_begin(file, file->f_mapping, pos, bytes, 0,
1500 &page, &fsdata);
1501 if (ret) {
1502 mlog_errno(ret);
1503 goto out;
1504 }
Mark Fasheh9517bac2007-02-09 20:24:12 -08001505
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001506 dst = kmap_atomic(page, KM_USER0);
1507 memcpy(dst + (pos & (PAGE_CACHE_SIZE - 1)), buf, bytes);
1508 kunmap_atomic(dst, KM_USER0);
1509 flush_dcache_page(page);
1510 ocfs2_put_write_source(user_page);
1511
1512 copied = ocfs2_write_end(file, file->f_mapping, pos, bytes,
1513 bytes, page, fsdata);
Mark Fasheh9517bac2007-02-09 20:24:12 -08001514 if (copied < 0) {
1515 mlog_errno(copied);
1516 ret = copied;
1517 goto out;
1518 }
1519
1520 total += copied;
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001521 *ppos = pos + copied;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001522 count -= copied;
1523
1524 ocfs2_set_next_iovec(&cur_iov, &iov_offset, copied);
1525 } while(count);
1526
1527out:
1528 return total ? total : ret;
1529}
1530
Tiger Yang8659ac22006-10-17 18:29:52 -07001531static ssize_t ocfs2_file_aio_write(struct kiocb *iocb,
1532 const struct iovec *iov,
1533 unsigned long nr_segs,
1534 loff_t pos)
1535{
Mark Fasheh9517bac2007-02-09 20:24:12 -08001536 int ret, direct_io, appending, rw_level, have_alloc_sem = 0;
1537 int can_do_direct, sync = 0;
1538 ssize_t written = 0;
1539 size_t ocount; /* original count */
1540 size_t count; /* after file limit checks */
1541 loff_t *ppos = &iocb->ki_pos;
1542 struct file *file = iocb->ki_filp;
1543 struct inode *inode = file->f_path.dentry->d_inode;
Tiger Yang8659ac22006-10-17 18:29:52 -07001544
Mark Fasheh9517bac2007-02-09 20:24:12 -08001545 mlog_entry("(0x%p, %u, '%.*s')\n", file,
Tiger Yang8659ac22006-10-17 18:29:52 -07001546 (unsigned int)nr_segs,
Mark Fasheh9517bac2007-02-09 20:24:12 -08001547 file->f_path.dentry->d_name.len,
1548 file->f_path.dentry->d_name.name);
Tiger Yang8659ac22006-10-17 18:29:52 -07001549
Tiger Yang8659ac22006-10-17 18:29:52 -07001550 if (iocb->ki_left == 0)
1551 return 0;
1552
Christoph Hellwigd9b08b92007-05-18 13:12:40 +02001553 ret = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
Mark Fasheh9517bac2007-02-09 20:24:12 -08001554 if (ret)
1555 return ret;
1556
1557 count = ocount;
1558
1559 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
1560
1561 appending = file->f_flags & O_APPEND ? 1 : 0;
1562 direct_io = file->f_flags & O_DIRECT ? 1 : 0;
1563
Tiger Yang8659ac22006-10-17 18:29:52 -07001564 mutex_lock(&inode->i_mutex);
Mark Fasheh9517bac2007-02-09 20:24:12 -08001565
1566relock:
Tiger Yang8659ac22006-10-17 18:29:52 -07001567 /* to match setattr's i_mutex -> i_alloc_sem -> rw_lock ordering */
Mark Fasheh9517bac2007-02-09 20:24:12 -08001568 if (direct_io) {
Tiger Yang8659ac22006-10-17 18:29:52 -07001569 down_read(&inode->i_alloc_sem);
Mark Fasheh9517bac2007-02-09 20:24:12 -08001570 have_alloc_sem = 1;
Tiger Yang8659ac22006-10-17 18:29:52 -07001571 }
1572
1573 /* concurrent O_DIRECT writes are allowed */
Mark Fasheh9517bac2007-02-09 20:24:12 -08001574 rw_level = !direct_io;
Tiger Yang8659ac22006-10-17 18:29:52 -07001575 ret = ocfs2_rw_lock(inode, rw_level);
1576 if (ret < 0) {
Mark Fasheh9517bac2007-02-09 20:24:12 -08001577 mlog_errno(ret);
1578 goto out_sems;
1579 }
1580
1581 can_do_direct = direct_io;
1582 ret = ocfs2_prepare_inode_for_write(file->f_path.dentry, ppos,
1583 iocb->ki_left, appending,
1584 &can_do_direct);
1585 if (ret < 0) {
Tiger Yang8659ac22006-10-17 18:29:52 -07001586 mlog_errno(ret);
1587 goto out;
1588 }
1589
Mark Fasheh9517bac2007-02-09 20:24:12 -08001590 /*
1591 * We can't complete the direct I/O as requested, fall back to
1592 * buffered I/O.
1593 */
1594 if (direct_io && !can_do_direct) {
1595 ocfs2_rw_unlock(inode, rw_level);
1596 up_read(&inode->i_alloc_sem);
1597
1598 have_alloc_sem = 0;
1599 rw_level = -1;
1600
1601 direct_io = 0;
1602 sync = 1;
1603 goto relock;
Tiger Yang8659ac22006-10-17 18:29:52 -07001604 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001605
Mark Fasheh9517bac2007-02-09 20:24:12 -08001606 if (!sync && ((file->f_flags & O_SYNC) || IS_SYNC(inode)))
1607 sync = 1;
1608
1609 /*
1610 * XXX: Is it ok to execute these checks a second time?
1611 */
1612 ret = generic_write_checks(file, ppos, &count, S_ISBLK(inode->i_mode));
1613 if (ret)
1614 goto out;
1615
1616 /*
1617 * Set pos so that sync_page_range_nolock() below understands
1618 * where to start from. We might've moved it around via the
1619 * calls above. The range we want to actually sync starts from
1620 * *ppos here.
1621 *
1622 */
1623 pos = *ppos;
1624
Mark Fashehccd979b2005-12-15 14:31:24 -08001625 /* communicate with ocfs2_dio_end_io */
Mark Fasheh7cdfc3a2007-04-16 17:28:51 -07001626 ocfs2_iocb_set_rw_locked(iocb, rw_level);
Mark Fashehccd979b2005-12-15 14:31:24 -08001627
Mark Fasheh9517bac2007-02-09 20:24:12 -08001628 if (direct_io) {
1629 written = generic_file_direct_write(iocb, iov, &nr_segs, *ppos,
1630 ppos, count, ocount);
1631 if (written < 0) {
1632 ret = written;
1633 goto out_dio;
1634 }
1635 } else {
1636 written = ocfs2_file_buffered_write(file, ppos, iov, nr_segs,
1637 count, written);
1638 if (written < 0) {
1639 ret = written;
1640 if (ret != -EFAULT || ret != -ENOSPC)
1641 mlog_errno(ret);
1642 goto out;
1643 }
1644 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001645
Mark Fasheh9517bac2007-02-09 20:24:12 -08001646out_dio:
Mark Fashehccd979b2005-12-15 14:31:24 -08001647 /* buffered aio wouldn't have proper lock coverage today */
Mark Fasheh9517bac2007-02-09 20:24:12 -08001648 BUG_ON(ret == -EIOCBQUEUED && !(file->f_flags & O_DIRECT));
Mark Fashehccd979b2005-12-15 14:31:24 -08001649
1650 /*
1651 * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
1652 * function pointer which is called when o_direct io completes so that
1653 * it can unlock our rw lock. (it's the clustered equivalent of
1654 * i_alloc_sem; protects truncate from racing with pending ios).
1655 * Unfortunately there are error cases which call end_io and others
1656 * that don't. so we don't have to unlock the rw_lock if either an
1657 * async dio is going to do it in the future or an end_io after an
1658 * error has already done it.
1659 */
1660 if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
1661 rw_level = -1;
1662 have_alloc_sem = 0;
1663 }
1664
1665out:
Mark Fasheh9517bac2007-02-09 20:24:12 -08001666 if (rw_level != -1)
1667 ocfs2_rw_unlock(inode, rw_level);
1668
1669out_sems:
Mark Fashehccd979b2005-12-15 14:31:24 -08001670 if (have_alloc_sem)
1671 up_read(&inode->i_alloc_sem);
Mark Fasheh9517bac2007-02-09 20:24:12 -08001672
1673 if (written > 0 && sync) {
1674 ssize_t err;
1675
1676 err = sync_page_range_nolock(inode, file->f_mapping, pos, count);
1677 if (err < 0)
1678 written = err;
1679 }
1680
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001681 mutex_unlock(&inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -08001682
1683 mlog_exit(ret);
Mark Fasheh9517bac2007-02-09 20:24:12 -08001684 return written ? written : ret;
Mark Fashehccd979b2005-12-15 14:31:24 -08001685}
1686
Mark Fasheh6af67d82007-03-06 17:24:46 -08001687static int ocfs2_splice_write_actor(struct pipe_inode_info *pipe,
1688 struct pipe_buffer *buf,
1689 struct splice_desc *sd)
1690{
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001691 int ret, count;
Mark Fasheh6af67d82007-03-06 17:24:46 -08001692 ssize_t copied = 0;
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001693 struct file *file = sd->u.file;
1694 unsigned int offset;
1695 struct page *page = NULL;
1696 void *fsdata;
1697 char *src, *dst;
Mark Fasheh6af67d82007-03-06 17:24:46 -08001698
Jens Axboecac36bb02007-06-14 13:10:48 +02001699 ret = buf->ops->confirm(pipe, buf);
Mark Fasheh6af67d82007-03-06 17:24:46 -08001700 if (ret)
1701 goto out;
1702
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001703 offset = sd->pos & ~PAGE_CACHE_MASK;
Mark Fasheh6af67d82007-03-06 17:24:46 -08001704 count = sd->len;
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001705 if (count + offset > PAGE_CACHE_SIZE)
1706 count = PAGE_CACHE_SIZE - offset;
Mark Fasheh6af67d82007-03-06 17:24:46 -08001707
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001708 ret = ocfs2_write_begin(file, file->f_mapping, sd->pos, count, 0,
1709 &page, &fsdata);
1710 if (ret) {
1711 mlog_errno(ret);
1712 goto out;
1713 }
Mark Fasheh6af67d82007-03-06 17:24:46 -08001714
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001715 src = buf->ops->map(pipe, buf, 1);
1716 dst = kmap_atomic(page, KM_USER1);
1717 memcpy(dst + offset, src + buf->offset, count);
1718 kunmap_atomic(page, KM_USER1);
1719 buf->ops->unmap(pipe, buf, src);
Mark Fasheh6af67d82007-03-06 17:24:46 -08001720
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001721 copied = ocfs2_write_end(file, file->f_mapping, sd->pos, count, count,
1722 page, fsdata);
1723 if (copied < 0) {
1724 mlog_errno(copied);
1725 ret = copied;
1726 goto out;
1727 }
Mark Fasheh6af67d82007-03-06 17:24:46 -08001728out:
1729
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001730 return copied ? copied : ret;
Mark Fasheh6af67d82007-03-06 17:24:46 -08001731}
1732
1733static ssize_t __ocfs2_file_splice_write(struct pipe_inode_info *pipe,
1734 struct file *out,
1735 loff_t *ppos,
1736 size_t len,
1737 unsigned int flags)
1738{
1739 int ret, err;
1740 struct address_space *mapping = out->f_mapping;
1741 struct inode *inode = mapping->host;
Jens Axboec66ab6f2007-06-12 21:17:17 +02001742 struct splice_desc sd = {
1743 .total_len = len,
1744 .flags = flags,
1745 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +02001746 .u.file = out,
Jens Axboec66ab6f2007-06-12 21:17:17 +02001747 };
Mark Fasheh6af67d82007-03-06 17:24:46 -08001748
Jens Axboec66ab6f2007-06-12 21:17:17 +02001749 ret = __splice_from_pipe(pipe, &sd, ocfs2_splice_write_actor);
Mark Fasheh6af67d82007-03-06 17:24:46 -08001750 if (ret > 0) {
1751 *ppos += ret;
1752
1753 if (unlikely((out->f_flags & O_SYNC) || IS_SYNC(inode))) {
1754 err = generic_osync_inode(inode, mapping,
1755 OSYNC_METADATA|OSYNC_DATA);
1756 if (err)
1757 ret = err;
1758 }
1759 }
1760
1761 return ret;
1762}
1763
Tiger Yang8659ac22006-10-17 18:29:52 -07001764static ssize_t ocfs2_file_splice_write(struct pipe_inode_info *pipe,
1765 struct file *out,
1766 loff_t *ppos,
1767 size_t len,
1768 unsigned int flags)
1769{
1770 int ret;
Josef Sipekd28c9172006-12-08 02:37:25 -08001771 struct inode *inode = out->f_path.dentry->d_inode;
Tiger Yang8659ac22006-10-17 18:29:52 -07001772
1773 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", out, pipe,
1774 (unsigned int)len,
Josef Sipekd28c9172006-12-08 02:37:25 -08001775 out->f_path.dentry->d_name.len,
1776 out->f_path.dentry->d_name.name);
Tiger Yang8659ac22006-10-17 18:29:52 -07001777
1778 inode_double_lock(inode, pipe->inode);
1779
1780 ret = ocfs2_rw_lock(inode, 1);
1781 if (ret < 0) {
1782 mlog_errno(ret);
1783 goto out;
1784 }
1785
Mark Fasheh9517bac2007-02-09 20:24:12 -08001786 ret = ocfs2_prepare_inode_for_write(out->f_path.dentry, ppos, len, 0,
1787 NULL);
Tiger Yang8659ac22006-10-17 18:29:52 -07001788 if (ret < 0) {
1789 mlog_errno(ret);
1790 goto out_unlock;
1791 }
1792
1793 /* ok, we're done with i_size and alloc work */
Mark Fasheh6af67d82007-03-06 17:24:46 -08001794 ret = __ocfs2_file_splice_write(pipe, out, ppos, len, flags);
Tiger Yang8659ac22006-10-17 18:29:52 -07001795
1796out_unlock:
1797 ocfs2_rw_unlock(inode, 1);
1798out:
1799 inode_double_unlock(inode, pipe->inode);
1800
1801 mlog_exit(ret);
1802 return ret;
1803}
1804
1805static ssize_t ocfs2_file_splice_read(struct file *in,
1806 loff_t *ppos,
1807 struct pipe_inode_info *pipe,
1808 size_t len,
1809 unsigned int flags)
1810{
1811 int ret = 0;
Josef Sipekd28c9172006-12-08 02:37:25 -08001812 struct inode *inode = in->f_path.dentry->d_inode;
Tiger Yang8659ac22006-10-17 18:29:52 -07001813
1814 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", in, pipe,
1815 (unsigned int)len,
Josef Sipekd28c9172006-12-08 02:37:25 -08001816 in->f_path.dentry->d_name.len,
1817 in->f_path.dentry->d_name.name);
Tiger Yang8659ac22006-10-17 18:29:52 -07001818
1819 /*
1820 * See the comment in ocfs2_file_aio_read()
1821 */
1822 ret = ocfs2_meta_lock(inode, NULL, 0);
1823 if (ret < 0) {
1824 mlog_errno(ret);
1825 goto bail;
1826 }
1827 ocfs2_meta_unlock(inode, 0);
1828
1829 ret = generic_file_splice_read(in, ppos, pipe, len, flags);
1830
1831bail:
1832 mlog_exit(ret);
1833 return ret;
1834}
1835
Mark Fashehccd979b2005-12-15 14:31:24 -08001836static ssize_t ocfs2_file_aio_read(struct kiocb *iocb,
Badari Pulavarty027445c2006-09-30 23:28:46 -07001837 const struct iovec *iov,
1838 unsigned long nr_segs,
Mark Fashehccd979b2005-12-15 14:31:24 -08001839 loff_t pos)
1840{
Tiger Yang25899de2006-11-15 15:49:02 +08001841 int ret = 0, rw_level = -1, have_alloc_sem = 0, lock_level = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -08001842 struct file *filp = iocb->ki_filp;
Josef Sipekd28c9172006-12-08 02:37:25 -08001843 struct inode *inode = filp->f_path.dentry->d_inode;
Mark Fashehccd979b2005-12-15 14:31:24 -08001844
Badari Pulavarty027445c2006-09-30 23:28:46 -07001845 mlog_entry("(0x%p, %u, '%.*s')\n", filp,
1846 (unsigned int)nr_segs,
Josef Sipekd28c9172006-12-08 02:37:25 -08001847 filp->f_path.dentry->d_name.len,
1848 filp->f_path.dentry->d_name.name);
Mark Fashehccd979b2005-12-15 14:31:24 -08001849
1850 if (!inode) {
1851 ret = -EINVAL;
1852 mlog_errno(ret);
1853 goto bail;
1854 }
1855
Mark Fashehccd979b2005-12-15 14:31:24 -08001856 /*
1857 * buffered reads protect themselves in ->readpage(). O_DIRECT reads
1858 * need locks to protect pending reads from racing with truncate.
1859 */
1860 if (filp->f_flags & O_DIRECT) {
1861 down_read(&inode->i_alloc_sem);
1862 have_alloc_sem = 1;
1863
1864 ret = ocfs2_rw_lock(inode, 0);
1865 if (ret < 0) {
1866 mlog_errno(ret);
1867 goto bail;
1868 }
1869 rw_level = 0;
1870 /* communicate with ocfs2_dio_end_io */
Mark Fasheh7cdfc3a2007-04-16 17:28:51 -07001871 ocfs2_iocb_set_rw_locked(iocb, rw_level);
Mark Fashehccd979b2005-12-15 14:31:24 -08001872 }
1873
Mark Fashehc4374f82006-05-05 19:04:35 -07001874 /*
1875 * We're fine letting folks race truncates and extending
1876 * writes with read across the cluster, just like they can
1877 * locally. Hence no rw_lock during read.
1878 *
1879 * Take and drop the meta data lock to update inode fields
1880 * like i_size. This allows the checks down below
1881 * generic_file_aio_read() a chance of actually working.
1882 */
Tiger Yang25899de2006-11-15 15:49:02 +08001883 ret = ocfs2_meta_lock_atime(inode, filp->f_vfsmnt, &lock_level);
Mark Fashehc4374f82006-05-05 19:04:35 -07001884 if (ret < 0) {
1885 mlog_errno(ret);
1886 goto bail;
1887 }
Tiger Yang25899de2006-11-15 15:49:02 +08001888 ocfs2_meta_unlock(inode, lock_level);
Mark Fashehc4374f82006-05-05 19:04:35 -07001889
Badari Pulavarty027445c2006-09-30 23:28:46 -07001890 ret = generic_file_aio_read(iocb, iov, nr_segs, iocb->ki_pos);
Mark Fashehccd979b2005-12-15 14:31:24 -08001891 if (ret == -EINVAL)
1892 mlog(ML_ERROR, "generic_file_aio_read returned -EINVAL\n");
1893
1894 /* buffered aio wouldn't have proper lock coverage today */
1895 BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT));
1896
1897 /* see ocfs2_file_aio_write */
1898 if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
1899 rw_level = -1;
1900 have_alloc_sem = 0;
1901 }
1902
1903bail:
1904 if (have_alloc_sem)
1905 up_read(&inode->i_alloc_sem);
1906 if (rw_level != -1)
1907 ocfs2_rw_unlock(inode, rw_level);
1908 mlog_exit(ret);
1909
1910 return ret;
1911}
1912
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08001913const struct inode_operations ocfs2_file_iops = {
Mark Fashehccd979b2005-12-15 14:31:24 -08001914 .setattr = ocfs2_setattr,
1915 .getattr = ocfs2_getattr,
Tiger Yangd38eb8d2006-11-27 09:59:21 +08001916 .permission = ocfs2_permission,
Mark Fashehccd979b2005-12-15 14:31:24 -08001917};
1918
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08001919const struct inode_operations ocfs2_special_file_iops = {
Mark Fashehccd979b2005-12-15 14:31:24 -08001920 .setattr = ocfs2_setattr,
1921 .getattr = ocfs2_getattr,
Tiger Yangd38eb8d2006-11-27 09:59:21 +08001922 .permission = ocfs2_permission,
Mark Fashehccd979b2005-12-15 14:31:24 -08001923};
1924
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08001925const struct file_operations ocfs2_fops = {
Mark Fashehccd979b2005-12-15 14:31:24 -08001926 .read = do_sync_read,
1927 .write = do_sync_write,
Mark Fashehccd979b2005-12-15 14:31:24 -08001928 .mmap = ocfs2_mmap,
1929 .fsync = ocfs2_sync_file,
1930 .release = ocfs2_file_release,
1931 .open = ocfs2_file_open,
1932 .aio_read = ocfs2_file_aio_read,
1933 .aio_write = ocfs2_file_aio_write,
Herbert Poetzlca4d1472006-07-03 17:27:12 -07001934 .ioctl = ocfs2_ioctl,
Mark Fasheh586d2322007-03-09 15:56:28 -08001935#ifdef CONFIG_COMPAT
1936 .compat_ioctl = ocfs2_compat_ioctl,
1937#endif
Tiger Yang8659ac22006-10-17 18:29:52 -07001938 .splice_read = ocfs2_file_splice_read,
1939 .splice_write = ocfs2_file_splice_write,
Mark Fashehccd979b2005-12-15 14:31:24 -08001940};
1941
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08001942const struct file_operations ocfs2_dops = {
Mark Fashehccd979b2005-12-15 14:31:24 -08001943 .read = generic_read_dir,
1944 .readdir = ocfs2_readdir,
1945 .fsync = ocfs2_sync_file,
Herbert Poetzlca4d1472006-07-03 17:27:12 -07001946 .ioctl = ocfs2_ioctl,
Mark Fasheh586d2322007-03-09 15:56:28 -08001947#ifdef CONFIG_COMPAT
1948 .compat_ioctl = ocfs2_compat_ioctl,
1949#endif
Mark Fashehccd979b2005-12-15 14:31:24 -08001950};