blob: 4636aa6b01177fd75895cd89547235daa1695ed3 [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 Fasheh385820a2007-07-19 00:14:38 -070037#include <linux/falloc.h>
Mark Fashehccd979b2005-12-15 14:31:24 -080038
39#define MLOG_MASK_PREFIX ML_INODE
40#include <cluster/masklog.h>
41
42#include "ocfs2.h"
43
44#include "alloc.h"
45#include "aops.h"
46#include "dir.h"
47#include "dlmglue.h"
48#include "extent_map.h"
49#include "file.h"
50#include "sysfile.h"
51#include "inode.h"
Herbert Poetzlca4d1472006-07-03 17:27:12 -070052#include "ioctl.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080053#include "journal.h"
Mark Fasheh53fc6222007-12-20 16:49:04 -080054#include "locks.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080055#include "mmap.h"
56#include "suballoc.h"
57#include "super.h"
Tiger Yangcf1d6c72008-08-18 17:11:00 +080058#include "xattr.h"
Tiger Yang23fc2702008-11-14 11:17:18 +080059#include "acl.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080060
61#include "buffer_head_io.h"
62
63static int ocfs2_sync_inode(struct inode *inode)
64{
65 filemap_fdatawrite(inode->i_mapping);
66 return sync_mapping_buffers(inode->i_mapping);
67}
68
Mark Fasheh53fc6222007-12-20 16:49:04 -080069static int ocfs2_init_file_private(struct inode *inode, struct file *file)
70{
71 struct ocfs2_file_private *fp;
72
73 fp = kzalloc(sizeof(struct ocfs2_file_private), GFP_KERNEL);
74 if (!fp)
75 return -ENOMEM;
76
77 fp->fp_file = file;
78 mutex_init(&fp->fp_mutex);
79 ocfs2_file_lock_res_init(&fp->fp_flock, fp);
80 file->private_data = fp;
81
82 return 0;
83}
84
85static void ocfs2_free_file_private(struct inode *inode, struct file *file)
86{
87 struct ocfs2_file_private *fp = file->private_data;
88 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
89
90 if (fp) {
91 ocfs2_simple_drop_lockres(osb, &fp->fp_flock);
92 ocfs2_lock_res_free(&fp->fp_flock);
93 kfree(fp);
94 file->private_data = NULL;
95 }
96}
97
Mark Fashehccd979b2005-12-15 14:31:24 -080098static int ocfs2_file_open(struct inode *inode, struct file *file)
99{
100 int status;
101 int mode = file->f_flags;
102 struct ocfs2_inode_info *oi = OCFS2_I(inode);
103
104 mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file,
Josef Sipekd28c9172006-12-08 02:37:25 -0800105 file->f_path.dentry->d_name.len, file->f_path.dentry->d_name.name);
Mark Fashehccd979b2005-12-15 14:31:24 -0800106
107 spin_lock(&oi->ip_lock);
108
109 /* Check that the inode hasn't been wiped from disk by another
110 * node. If it hasn't then we're safe as long as we hold the
111 * spin lock until our increment of open count. */
112 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
113 spin_unlock(&oi->ip_lock);
114
115 status = -ENOENT;
116 goto leave;
117 }
118
119 if (mode & O_DIRECT)
120 oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT;
121
122 oi->ip_open_count++;
123 spin_unlock(&oi->ip_lock);
Mark Fasheh53fc6222007-12-20 16:49:04 -0800124
125 status = ocfs2_init_file_private(inode, file);
126 if (status) {
127 /*
128 * We want to set open count back if we're failing the
129 * open.
130 */
131 spin_lock(&oi->ip_lock);
132 oi->ip_open_count--;
133 spin_unlock(&oi->ip_lock);
134 }
135
Mark Fashehccd979b2005-12-15 14:31:24 -0800136leave:
137 mlog_exit(status);
138 return status;
139}
140
141static int ocfs2_file_release(struct inode *inode, struct file *file)
142{
143 struct ocfs2_inode_info *oi = OCFS2_I(inode);
144
145 mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file,
Josef Sipekd28c9172006-12-08 02:37:25 -0800146 file->f_path.dentry->d_name.len,
147 file->f_path.dentry->d_name.name);
Mark Fashehccd979b2005-12-15 14:31:24 -0800148
149 spin_lock(&oi->ip_lock);
150 if (!--oi->ip_open_count)
151 oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT;
152 spin_unlock(&oi->ip_lock);
153
Mark Fasheh53fc6222007-12-20 16:49:04 -0800154 ocfs2_free_file_private(inode, file);
155
Mark Fashehccd979b2005-12-15 14:31:24 -0800156 mlog_exit(0);
157
158 return 0;
159}
160
Mark Fasheh53fc6222007-12-20 16:49:04 -0800161static int ocfs2_dir_open(struct inode *inode, struct file *file)
162{
163 return ocfs2_init_file_private(inode, file);
164}
165
166static int ocfs2_dir_release(struct inode *inode, struct file *file)
167{
168 ocfs2_free_file_private(inode, file);
169 return 0;
170}
171
Mark Fashehccd979b2005-12-15 14:31:24 -0800172static int ocfs2_sync_file(struct file *file,
173 struct dentry *dentry,
174 int datasync)
175{
176 int err = 0;
177 journal_t *journal;
178 struct inode *inode = dentry->d_inode;
179 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
180
181 mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", file, dentry, datasync,
182 dentry->d_name.len, dentry->d_name.name);
183
184 err = ocfs2_sync_inode(dentry->d_inode);
185 if (err)
186 goto bail;
187
188 journal = osb->journal->j_journal;
Joel Becker2b4e30f2008-09-03 20:03:41 -0700189 err = jbd2_journal_force_commit(journal);
Mark Fashehccd979b2005-12-15 14:31:24 -0800190
191bail:
192 mlog_exit(err);
193
194 return (err < 0) ? -EIO : 0;
195}
196
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800197int ocfs2_should_update_atime(struct inode *inode,
198 struct vfsmount *vfsmnt)
199{
200 struct timespec now;
201 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
202
203 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
204 return 0;
205
206 if ((inode->i_flags & S_NOATIME) ||
207 ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)))
208 return 0;
209
Mark Fasheh6c2aad02006-12-19 15:25:52 -0800210 /*
211 * We can be called with no vfsmnt structure - NFSD will
212 * sometimes do this.
213 *
214 * Note that our action here is different than touch_atime() -
215 * if we can't tell whether this is a noatime mount, then we
216 * don't know whether to trust the value of s_atime_quantum.
217 */
218 if (vfsmnt == NULL)
219 return 0;
220
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800221 if ((vfsmnt->mnt_flags & MNT_NOATIME) ||
222 ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
223 return 0;
224
Mark Fasheh7e913c52006-12-13 00:34:35 -0800225 if (vfsmnt->mnt_flags & MNT_RELATIME) {
226 if ((timespec_compare(&inode->i_atime, &inode->i_mtime) <= 0) ||
227 (timespec_compare(&inode->i_atime, &inode->i_ctime) <= 0))
228 return 1;
229
230 return 0;
231 }
232
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800233 now = CURRENT_TIME;
234 if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum))
235 return 0;
236 else
237 return 1;
238}
239
240int ocfs2_update_inode_atime(struct inode *inode,
241 struct buffer_head *bh)
242{
243 int ret;
244 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
245 handle_t *handle;
Mark Fashehc11e9fa2007-07-20 11:24:53 -0700246 struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data;
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800247
248 mlog_entry_void();
249
250 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
Jan Karafa38e922008-10-20 19:23:51 +0200251 if (IS_ERR(handle)) {
252 ret = PTR_ERR(handle);
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800253 mlog_errno(ret);
254 goto out;
255 }
256
Mark Fashehc11e9fa2007-07-20 11:24:53 -0700257 ret = ocfs2_journal_access(handle, inode, bh,
258 OCFS2_JOURNAL_ACCESS_WRITE);
259 if (ret) {
260 mlog_errno(ret);
261 goto out_commit;
262 }
263
264 /*
265 * Don't use ocfs2_mark_inode_dirty() here as we don't always
266 * have i_mutex to guard against concurrent changes to other
267 * inode fields.
268 */
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800269 inode->i_atime = CURRENT_TIME;
Mark Fashehc11e9fa2007-07-20 11:24:53 -0700270 di->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
271 di->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
272
273 ret = ocfs2_journal_dirty(handle, bh);
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800274 if (ret < 0)
275 mlog_errno(ret);
276
Mark Fashehc11e9fa2007-07-20 11:24:53 -0700277out_commit:
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800278 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
279out:
280 mlog_exit(ret);
281 return ret;
282}
283
Adrian Bunk6cb129f2007-04-26 00:29:35 -0700284static int ocfs2_set_inode_size(handle_t *handle,
285 struct inode *inode,
286 struct buffer_head *fe_bh,
287 u64 new_i_size)
Mark Fashehccd979b2005-12-15 14:31:24 -0800288{
289 int status;
290
291 mlog_entry_void();
292 i_size_write(inode, new_i_size);
Mark Fasheh8110b072007-03-22 16:53:23 -0700293 inode->i_blocks = ocfs2_inode_sector_count(inode);
Mark Fashehccd979b2005-12-15 14:31:24 -0800294 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
295
296 status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
297 if (status < 0) {
298 mlog_errno(status);
299 goto bail;
300 }
301
302bail:
303 mlog_exit(status);
304 return status;
305}
306
307static int ocfs2_simple_size_update(struct inode *inode,
308 struct buffer_head *di_bh,
309 u64 new_i_size)
310{
311 int ret;
312 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
Mark Fasheh1fabe142006-10-09 18:11:45 -0700313 handle_t *handle = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800314
Mark Fasheh65eff9c2006-10-09 17:26:22 -0700315 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
Jan Karafa38e922008-10-20 19:23:51 +0200316 if (IS_ERR(handle)) {
317 ret = PTR_ERR(handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800318 mlog_errno(ret);
319 goto out;
320 }
321
322 ret = ocfs2_set_inode_size(handle, inode, di_bh,
323 new_i_size);
324 if (ret < 0)
325 mlog_errno(ret);
326
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700327 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800328out:
329 return ret;
330}
331
332static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
333 struct inode *inode,
334 struct buffer_head *fe_bh,
335 u64 new_i_size)
336{
337 int status;
Mark Fasheh1fabe142006-10-09 18:11:45 -0700338 handle_t *handle;
Mark Fasheh60b11392007-02-16 11:46:50 -0800339 struct ocfs2_dinode *di;
Mark Fasheh35edec12007-07-06 14:41:18 -0700340 u64 cluster_bytes;
Mark Fashehccd979b2005-12-15 14:31:24 -0800341
342 mlog_entry_void();
343
344 /* TODO: This needs to actually orphan the inode in this
345 * transaction. */
346
Mark Fasheh65eff9c2006-10-09 17:26:22 -0700347 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
Mark Fashehccd979b2005-12-15 14:31:24 -0800348 if (IS_ERR(handle)) {
349 status = PTR_ERR(handle);
350 mlog_errno(status);
351 goto out;
352 }
353
Mark Fasheh60b11392007-02-16 11:46:50 -0800354 status = ocfs2_journal_access(handle, inode, fe_bh,
355 OCFS2_JOURNAL_ACCESS_WRITE);
356 if (status < 0) {
357 mlog_errno(status);
358 goto out_commit;
359 }
360
361 /*
362 * Do this before setting i_size.
363 */
Mark Fasheh35edec12007-07-06 14:41:18 -0700364 cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size);
365 status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size,
366 cluster_bytes);
Mark Fasheh60b11392007-02-16 11:46:50 -0800367 if (status) {
368 mlog_errno(status);
369 goto out_commit;
370 }
371
372 i_size_write(inode, new_i_size);
Mark Fasheh60b11392007-02-16 11:46:50 -0800373 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
374
375 di = (struct ocfs2_dinode *) fe_bh->b_data;
376 di->i_size = cpu_to_le64(new_i_size);
377 di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
378 di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
379
380 status = ocfs2_journal_dirty(handle, fe_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800381 if (status < 0)
382 mlog_errno(status);
383
Mark Fasheh60b11392007-02-16 11:46:50 -0800384out_commit:
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700385 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800386out:
Mark Fasheh60b11392007-02-16 11:46:50 -0800387
Mark Fashehccd979b2005-12-15 14:31:24 -0800388 mlog_exit(status);
389 return status;
390}
391
392static int ocfs2_truncate_file(struct inode *inode,
393 struct buffer_head *di_bh,
394 u64 new_i_size)
395{
396 int status = 0;
397 struct ocfs2_dinode *fe = NULL;
398 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
399 struct ocfs2_truncate_context *tc = NULL;
400
Mark Fashehb0697052006-03-03 10:24:33 -0800401 mlog_entry("(inode = %llu, new_i_size = %llu\n",
402 (unsigned long long)OCFS2_I(inode)->ip_blkno,
403 (unsigned long long)new_i_size);
Mark Fashehccd979b2005-12-15 14:31:24 -0800404
Mark Fashehccd979b2005-12-15 14:31:24 -0800405 fe = (struct ocfs2_dinode *) di_bh->b_data;
406 if (!OCFS2_IS_VALID_DINODE(fe)) {
407 OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe);
408 status = -EIO;
409 goto bail;
410 }
411
412 mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode),
Mark Fashehb0697052006-03-03 10:24:33 -0800413 "Inode %llu, inode i_size = %lld != di "
414 "i_size = %llu, i_flags = 0x%x\n",
415 (unsigned long long)OCFS2_I(inode)->ip_blkno,
Mark Fashehccd979b2005-12-15 14:31:24 -0800416 i_size_read(inode),
Mark Fashehb0697052006-03-03 10:24:33 -0800417 (unsigned long long)le64_to_cpu(fe->i_size),
418 le32_to_cpu(fe->i_flags));
Mark Fashehccd979b2005-12-15 14:31:24 -0800419
420 if (new_i_size > le64_to_cpu(fe->i_size)) {
Mark Fashehb0697052006-03-03 10:24:33 -0800421 mlog(0, "asked to truncate file with size (%llu) to size (%llu)!\n",
422 (unsigned long long)le64_to_cpu(fe->i_size),
423 (unsigned long long)new_i_size);
Mark Fashehccd979b2005-12-15 14:31:24 -0800424 status = -EINVAL;
425 mlog_errno(status);
426 goto bail;
427 }
428
Mark Fashehb0697052006-03-03 10:24:33 -0800429 mlog(0, "inode %llu, i_size = %llu, new_i_size = %llu\n",
430 (unsigned long long)le64_to_cpu(fe->i_blkno),
431 (unsigned long long)le64_to_cpu(fe->i_size),
432 (unsigned long long)new_i_size);
Mark Fashehccd979b2005-12-15 14:31:24 -0800433
434 /* lets handle the simple truncate cases before doing any more
435 * cluster locking. */
436 if (new_i_size == le64_to_cpu(fe->i_size))
437 goto bail;
438
Mark Fasheh2e89b2e2007-05-09 13:40:18 -0700439 down_write(&OCFS2_I(inode)->ip_alloc_sem);
440
Mark Fashehc934a922007-10-18 15:23:46 -0700441 /*
442 * The inode lock forced other nodes to sync and drop their
443 * pages, which (correctly) happens even if we have a truncate
444 * without allocation change - ocfs2 cluster sizes can be much
445 * greater than page size, so we have to truncate them
446 * anyway.
447 */
Mark Fasheh2e89b2e2007-05-09 13:40:18 -0700448 unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1);
449 truncate_inode_pages(inode->i_mapping, new_i_size);
450
Mark Fasheh1afc32b2007-09-07 14:46:51 -0700451 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
452 status = ocfs2_truncate_inline(inode, di_bh, new_i_size,
Mark Fashehb1967d02007-11-20 11:56:39 -0800453 i_size_read(inode), 1);
Mark Fasheh1afc32b2007-09-07 14:46:51 -0700454 if (status)
455 mlog_errno(status);
456
Mark Fashehc934a922007-10-18 15:23:46 -0700457 goto bail_unlock_sem;
Mark Fasheh1afc32b2007-09-07 14:46:51 -0700458 }
459
Mark Fashehccd979b2005-12-15 14:31:24 -0800460 /* alright, we're going to need to do a full blown alloc size
461 * change. Orphan the inode so that recovery can complete the
462 * truncate if necessary. This does the task of marking
463 * i_size. */
464 status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
465 if (status < 0) {
466 mlog_errno(status);
Mark Fashehc934a922007-10-18 15:23:46 -0700467 goto bail_unlock_sem;
Mark Fashehccd979b2005-12-15 14:31:24 -0800468 }
469
470 status = ocfs2_prepare_truncate(osb, inode, di_bh, &tc);
471 if (status < 0) {
472 mlog_errno(status);
Mark Fashehc934a922007-10-18 15:23:46 -0700473 goto bail_unlock_sem;
Mark Fashehccd979b2005-12-15 14:31:24 -0800474 }
475
476 status = ocfs2_commit_truncate(osb, inode, di_bh, tc);
477 if (status < 0) {
478 mlog_errno(status);
Mark Fashehc934a922007-10-18 15:23:46 -0700479 goto bail_unlock_sem;
Mark Fashehccd979b2005-12-15 14:31:24 -0800480 }
481
482 /* TODO: orphan dir cleanup here. */
Mark Fashehc934a922007-10-18 15:23:46 -0700483bail_unlock_sem:
Mark Fasheh2e89b2e2007-05-09 13:40:18 -0700484 up_write(&OCFS2_I(inode)->ip_alloc_sem);
485
Mark Fashehccd979b2005-12-15 14:31:24 -0800486bail:
487
488 mlog_exit(status);
489 return status;
490}
491
492/*
Tao Ma0eb8d472008-08-18 17:38:45 +0800493 * extend file allocation only here.
Mark Fashehccd979b2005-12-15 14:31:24 -0800494 * we'll update all the disk stuff, and oip->alloc_size
495 *
496 * expect stuff to be locked, a transaction started and enough data /
497 * metadata reservations in the contexts.
498 *
499 * Will return -EAGAIN, and a reason if a restart is needed.
500 * If passed in, *reason will always be set, even in error.
501 */
Tao Ma0eb8d472008-08-18 17:38:45 +0800502int ocfs2_add_inode_data(struct ocfs2_super *osb,
503 struct inode *inode,
504 u32 *logical_offset,
505 u32 clusters_to_add,
506 int mark_unwritten,
507 struct buffer_head *fe_bh,
508 handle_t *handle,
509 struct ocfs2_alloc_context *data_ac,
510 struct ocfs2_alloc_context *meta_ac,
511 enum ocfs2_alloc_restarted *reason_ret)
Mark Fashehccd979b2005-12-15 14:31:24 -0800512{
Joel Beckerf99b9b72008-08-20 19:36:33 -0700513 int ret;
514 struct ocfs2_extent_tree et;
Mark Fashehccd979b2005-12-15 14:31:24 -0800515
Joel Becker8d6220d2008-08-22 12:46:09 -0700516 ocfs2_init_dinode_extent_tree(&et, inode, fe_bh);
Joel Beckerf99b9b72008-08-20 19:36:33 -0700517 ret = ocfs2_add_clusters_in_btree(osb, inode, logical_offset,
Tao Ma0eb8d472008-08-18 17:38:45 +0800518 clusters_to_add, mark_unwritten,
Joel Beckerf99b9b72008-08-20 19:36:33 -0700519 &et, handle,
520 data_ac, meta_ac, reason_ret);
Joel Beckerf99b9b72008-08-20 19:36:33 -0700521
522 return ret;
Mark Fashehccd979b2005-12-15 14:31:24 -0800523}
524
Mark Fasheh2ae99a62007-03-09 16:43:28 -0800525static int __ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
526 u32 clusters_to_add, int mark_unwritten)
Mark Fashehccd979b2005-12-15 14:31:24 -0800527{
528 int status = 0;
529 int restart_func = 0;
Mark Fashehabf8b152007-01-17 13:07:24 -0800530 int credits;
Mark Fasheh2ae99a62007-03-09 16:43:28 -0800531 u32 prev_clusters;
Mark Fashehccd979b2005-12-15 14:31:24 -0800532 struct buffer_head *bh = NULL;
533 struct ocfs2_dinode *fe = NULL;
Mark Fasheh1fabe142006-10-09 18:11:45 -0700534 handle_t *handle = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800535 struct ocfs2_alloc_context *data_ac = NULL;
536 struct ocfs2_alloc_context *meta_ac = NULL;
537 enum ocfs2_alloc_restarted why;
538 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
Joel Beckerf99b9b72008-08-20 19:36:33 -0700539 struct ocfs2_extent_tree et;
Mark Fashehccd979b2005-12-15 14:31:24 -0800540
541 mlog_entry("(clusters_to_add = %u)\n", clusters_to_add);
542
Mark Fashehdcd05382007-01-16 11:32:23 -0800543 /*
544 * This function only exists for file systems which don't
545 * support holes.
546 */
Mark Fasheh2ae99a62007-03-09 16:43:28 -0800547 BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb));
Mark Fashehdcd05382007-01-16 11:32:23 -0800548
Joel Becker0fcaa562008-10-09 17:20:31 -0700549 status = ocfs2_read_block(inode, OCFS2_I(inode)->ip_blkno, &bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800550 if (status < 0) {
551 mlog_errno(status);
552 goto leave;
553 }
554
555 fe = (struct ocfs2_dinode *) bh->b_data;
556 if (!OCFS2_IS_VALID_DINODE(fe)) {
557 OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe);
558 status = -EIO;
559 goto leave;
560 }
561
562restart_all:
563 BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
564
Tao Mae7d4cb62008-08-18 17:38:44 +0800565 mlog(0, "extend inode %llu, i_size = %lld, di->i_clusters = %u, "
566 "clusters_to_add = %u\n",
567 (unsigned long long)OCFS2_I(inode)->ip_blkno,
568 (long long)i_size_read(inode), le32_to_cpu(fe->i_clusters),
569 clusters_to_add);
Joel Becker8d6220d2008-08-22 12:46:09 -0700570 ocfs2_init_dinode_extent_tree(&et, inode, bh);
Joel Beckerf99b9b72008-08-20 19:36:33 -0700571 status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
572 &data_ac, &meta_ac);
Mark Fasheh9517bac2007-02-09 20:24:12 -0800573 if (status) {
574 mlog_errno(status);
575 goto leave;
576 }
577
Tao Ma811f9332008-08-18 17:38:43 +0800578 credits = ocfs2_calc_extend_credits(osb->sb, &fe->id2.i_list,
579 clusters_to_add);
Mark Fasheh65eff9c2006-10-09 17:26:22 -0700580 handle = ocfs2_start_trans(osb, credits);
Mark Fashehccd979b2005-12-15 14:31:24 -0800581 if (IS_ERR(handle)) {
582 status = PTR_ERR(handle);
583 handle = NULL;
584 mlog_errno(status);
585 goto leave;
586 }
587
588restarted_transaction:
589 /* reserve a write to the file entry early on - that we if we
590 * run out of credits in the allocation path, we can still
591 * update i_size. */
592 status = ocfs2_journal_access(handle, inode, bh,
593 OCFS2_JOURNAL_ACCESS_WRITE);
594 if (status < 0) {
595 mlog_errno(status);
596 goto leave;
597 }
598
599 prev_clusters = OCFS2_I(inode)->ip_clusters;
600
Tao Ma0eb8d472008-08-18 17:38:45 +0800601 status = ocfs2_add_inode_data(osb,
602 inode,
603 &logical_start,
604 clusters_to_add,
605 mark_unwritten,
606 bh,
607 handle,
608 data_ac,
609 meta_ac,
610 &why);
Mark Fashehccd979b2005-12-15 14:31:24 -0800611 if ((status < 0) && (status != -EAGAIN)) {
612 if (status != -ENOSPC)
613 mlog_errno(status);
614 goto leave;
615 }
616
617 status = ocfs2_journal_dirty(handle, bh);
618 if (status < 0) {
619 mlog_errno(status);
620 goto leave;
621 }
622
623 spin_lock(&OCFS2_I(inode)->ip_lock);
624 clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
625 spin_unlock(&OCFS2_I(inode)->ip_lock);
626
627 if (why != RESTART_NONE && clusters_to_add) {
628 if (why == RESTART_META) {
629 mlog(0, "restarting function.\n");
630 restart_func = 1;
631 } else {
632 BUG_ON(why != RESTART_TRANS);
633
634 mlog(0, "restarting transaction.\n");
635 /* TODO: This can be more intelligent. */
636 credits = ocfs2_calc_extend_credits(osb->sb,
Tao Ma811f9332008-08-18 17:38:43 +0800637 &fe->id2.i_list,
Mark Fashehccd979b2005-12-15 14:31:24 -0800638 clusters_to_add);
Mark Fasheh1fabe142006-10-09 18:11:45 -0700639 status = ocfs2_extend_trans(handle, credits);
Mark Fashehccd979b2005-12-15 14:31:24 -0800640 if (status < 0) {
641 /* handle still has to be committed at
642 * this point. */
643 status = -ENOMEM;
644 mlog_errno(status);
645 goto leave;
646 }
647 goto restarted_transaction;
648 }
649 }
650
Mark Fashehb0697052006-03-03 10:24:33 -0800651 mlog(0, "fe: i_clusters = %u, i_size=%llu\n",
Mark Fasheh1ca1a112007-04-27 16:01:25 -0700652 le32_to_cpu(fe->i_clusters),
653 (unsigned long long)le64_to_cpu(fe->i_size));
Mark Fashehccd979b2005-12-15 14:31:24 -0800654 mlog(0, "inode: ip_clusters=%u, i_size=%lld\n",
Jan Kara634bf742007-12-19 15:25:42 +0100655 OCFS2_I(inode)->ip_clusters, (long long)i_size_read(inode));
Mark Fashehccd979b2005-12-15 14:31:24 -0800656
657leave:
Mark Fashehccd979b2005-12-15 14:31:24 -0800658 if (handle) {
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700659 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800660 handle = NULL;
661 }
662 if (data_ac) {
663 ocfs2_free_alloc_context(data_ac);
664 data_ac = NULL;
665 }
666 if (meta_ac) {
667 ocfs2_free_alloc_context(meta_ac);
668 meta_ac = NULL;
669 }
670 if ((!status) && restart_func) {
671 restart_func = 0;
672 goto restart_all;
673 }
Mark Fasheha81cb882008-10-07 14:25:16 -0700674 brelse(bh);
675 bh = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800676
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
Nick Piggin4e02ed42008-10-29 14:00:55 -0700683 * worry about recursive locking in ->write_begin() and ->write_end(). */
Mark Fashehccd979b2005-12-15 14:31:24 -0800684static int ocfs2_write_zero_page(struct inode *inode,
685 u64 size)
686{
687 struct address_space *mapping = inode->i_mapping;
688 struct page *page;
689 unsigned long index;
690 unsigned int offset;
Mark Fasheh1fabe142006-10-09 18:11:45 -0700691 handle_t *handle = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800692 int ret;
693
694 offset = (size & (PAGE_CACHE_SIZE-1)); /* Within page */
695 /* ugh. in prepare/commit_write, if from==to==start of block, we
696 ** skip the prepare. make sure we never send an offset for the start
697 ** of a block
698 */
699 if ((offset & (inode->i_sb->s_blocksize - 1)) == 0) {
700 offset++;
701 }
702 index = size >> PAGE_CACHE_SHIFT;
703
704 page = grab_cache_page(mapping, index);
705 if (!page) {
706 ret = -ENOMEM;
707 mlog_errno(ret);
708 goto out;
709 }
710
Mark Fasheh53013cb2006-05-05 19:04:03 -0700711 ret = ocfs2_prepare_write_nolock(inode, page, offset, offset);
Mark Fashehccd979b2005-12-15 14:31:24 -0800712 if (ret < 0) {
713 mlog_errno(ret);
714 goto out_unlock;
715 }
716
717 if (ocfs2_should_order_data(inode)) {
718 handle = ocfs2_start_walk_page_trans(inode, page, offset,
719 offset);
720 if (IS_ERR(handle)) {
721 ret = PTR_ERR(handle);
722 handle = NULL;
723 goto out_unlock;
724 }
725 }
726
727 /* must not update i_size! */
728 ret = block_commit_write(page, offset, offset);
729 if (ret < 0)
730 mlog_errno(ret);
731 else
732 ret = 0;
733
734 if (handle)
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700735 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800736out_unlock:
737 unlock_page(page);
738 page_cache_release(page);
739out:
740 return ret;
741}
742
743static int ocfs2_zero_extend(struct inode *inode,
744 u64 zero_to_size)
745{
746 int ret = 0;
747 u64 start_off;
748 struct super_block *sb = inode->i_sb;
749
750 start_off = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
751 while (start_off < zero_to_size) {
752 ret = ocfs2_write_zero_page(inode, start_off);
753 if (ret < 0) {
754 mlog_errno(ret);
755 goto out;
756 }
757
758 start_off += sb->s_blocksize;
Mark Fashehe2057c52006-10-03 17:53:05 -0700759
760 /*
761 * Very large extends have the potential to lock up
762 * the cpu for extended periods of time.
763 */
764 cond_resched();
Mark Fashehccd979b2005-12-15 14:31:24 -0800765 }
766
767out:
768 return ret;
769}
770
Mark Fasheh65ed39d2007-08-28 17:13:23 -0700771int ocfs2_extend_no_holes(struct inode *inode, u64 new_i_size, u64 zero_to)
772{
773 int ret;
774 u32 clusters_to_add;
775 struct ocfs2_inode_info *oi = OCFS2_I(inode);
776
777 clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size);
778 if (clusters_to_add < oi->ip_clusters)
779 clusters_to_add = 0;
780 else
781 clusters_to_add -= oi->ip_clusters;
782
783 if (clusters_to_add) {
784 ret = __ocfs2_extend_allocation(inode, oi->ip_clusters,
785 clusters_to_add, 0);
786 if (ret) {
787 mlog_errno(ret);
788 goto out;
789 }
790 }
791
792 /*
793 * Call this even if we don't add any clusters to the tree. We
794 * still need to zero the area between the old i_size and the
795 * new i_size.
796 */
797 ret = ocfs2_zero_extend(inode, zero_to);
798 if (ret < 0)
799 mlog_errno(ret);
800
801out:
802 return ret;
803}
804
Mark Fashehccd979b2005-12-15 14:31:24 -0800805static int ocfs2_extend_file(struct inode *inode,
806 struct buffer_head *di_bh,
Mark Fasheh65ed39d2007-08-28 17:13:23 -0700807 u64 new_i_size)
Mark Fashehccd979b2005-12-15 14:31:24 -0800808{
Mark Fashehc934a922007-10-18 15:23:46 -0700809 int ret = 0;
Mark Fasheh1afc32b2007-09-07 14:46:51 -0700810 struct ocfs2_inode_info *oi = OCFS2_I(inode);
Mark Fashehccd979b2005-12-15 14:31:24 -0800811
Mark Fasheh65ed39d2007-08-28 17:13:23 -0700812 BUG_ON(!di_bh);
Mark Fasheh53013cb2006-05-05 19:04:03 -0700813
Mark Fashehccd979b2005-12-15 14:31:24 -0800814 /* setattr sometimes calls us like this. */
815 if (new_i_size == 0)
816 goto out;
817
818 if (i_size_read(inode) == new_i_size)
819 goto out;
820 BUG_ON(new_i_size < i_size_read(inode));
821
Mark Fasheh1afc32b2007-09-07 14:46:51 -0700822 /*
823 * Fall through for converting inline data, even if the fs
824 * supports sparse files.
825 *
826 * The check for inline data here is legal - nobody can add
827 * the feature since we have i_mutex. We must check it again
828 * after acquiring ip_alloc_sem though, as paths like mmap
829 * might have raced us to converting the inode to extents.
830 */
831 if (!(oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
832 && ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
Mark Fasheh3a0782d2007-01-17 12:53:31 -0800833 goto out_update_size;
Mark Fashehccd979b2005-12-15 14:31:24 -0800834
Mark Fasheh0effef72006-10-03 17:44:42 -0700835 /*
Mark Fasheh65ed39d2007-08-28 17:13:23 -0700836 * The alloc sem blocks people in read/write from reading our
837 * allocation until we're done changing it. We depend on
838 * i_mutex to block other extend/truncate calls while we're
839 * here.
Mark Fasheh0effef72006-10-03 17:44:42 -0700840 */
Mark Fasheh1afc32b2007-09-07 14:46:51 -0700841 down_write(&oi->ip_alloc_sem);
842
843 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
844 /*
845 * We can optimize small extends by keeping the inodes
846 * inline data.
847 */
848 if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) {
849 up_write(&oi->ip_alloc_sem);
850 goto out_update_size;
851 }
852
853 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
854 if (ret) {
855 up_write(&oi->ip_alloc_sem);
856
857 mlog_errno(ret);
Mark Fashehc934a922007-10-18 15:23:46 -0700858 goto out;
Mark Fasheh1afc32b2007-09-07 14:46:51 -0700859 }
860 }
861
862 if (!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
863 ret = ocfs2_extend_no_holes(inode, new_i_size, new_i_size);
864
865 up_write(&oi->ip_alloc_sem);
Mark Fasheh65ed39d2007-08-28 17:13:23 -0700866
Mark Fasheh0effef72006-10-03 17:44:42 -0700867 if (ret < 0) {
868 mlog_errno(ret);
Mark Fashehc934a922007-10-18 15:23:46 -0700869 goto out;
Mark Fashehccd979b2005-12-15 14:31:24 -0800870 }
871
Mark Fasheh3a0782d2007-01-17 12:53:31 -0800872out_update_size:
Mark Fasheh65ed39d2007-08-28 17:13:23 -0700873 ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
874 if (ret < 0)
875 mlog_errno(ret);
Mark Fasheh53013cb2006-05-05 19:04:03 -0700876
Mark Fashehccd979b2005-12-15 14:31:24 -0800877out:
878 return ret;
879}
880
881int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
882{
883 int status = 0, size_change;
884 struct inode *inode = dentry->d_inode;
885 struct super_block *sb = inode->i_sb;
886 struct ocfs2_super *osb = OCFS2_SB(sb);
887 struct buffer_head *bh = NULL;
Mark Fasheh1fabe142006-10-09 18:11:45 -0700888 handle_t *handle = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800889
890 mlog_entry("(0x%p, '%.*s')\n", dentry,
891 dentry->d_name.len, dentry->d_name.name);
892
Sunil Mushranbc535802008-04-18 10:23:53 -0700893 /* ensuring we don't even attempt to truncate a symlink */
894 if (S_ISLNK(inode->i_mode))
895 attr->ia_valid &= ~ATTR_SIZE;
896
Mark Fashehccd979b2005-12-15 14:31:24 -0800897 if (attr->ia_valid & ATTR_MODE)
898 mlog(0, "mode change: %d\n", attr->ia_mode);
899 if (attr->ia_valid & ATTR_UID)
900 mlog(0, "uid change: %d\n", attr->ia_uid);
901 if (attr->ia_valid & ATTR_GID)
902 mlog(0, "gid change: %d\n", attr->ia_gid);
903 if (attr->ia_valid & ATTR_SIZE)
904 mlog(0, "size change...\n");
905 if (attr->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME))
906 mlog(0, "time change...\n");
907
908#define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
909 | ATTR_GID | ATTR_UID | ATTR_MODE)
910 if (!(attr->ia_valid & OCFS2_VALID_ATTRS)) {
911 mlog(0, "can't handle attrs: 0x%x\n", attr->ia_valid);
912 return 0;
913 }
914
915 status = inode_change_ok(inode, attr);
916 if (status)
917 return status;
918
919 size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
920 if (size_change) {
921 status = ocfs2_rw_lock(inode, 1);
922 if (status < 0) {
923 mlog_errno(status);
924 goto bail;
925 }
926 }
927
Mark Fashehe63aecb62007-10-18 15:30:42 -0700928 status = ocfs2_inode_lock(inode, &bh, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -0800929 if (status < 0) {
930 if (status != -ENOENT)
931 mlog_errno(status);
932 goto bail_unlock_rw;
933 }
934
935 if (size_change && attr->ia_size != i_size_read(inode)) {
Mark Fashehce76fd32007-07-20 12:02:14 -0700936 if (attr->ia_size > sb->s_maxbytes) {
937 status = -EFBIG;
938 goto bail_unlock;
939 }
940
Joel Becker2b4e30f2008-09-03 20:03:41 -0700941 if (i_size_read(inode) > attr->ia_size) {
942 if (ocfs2_should_order_data(inode)) {
943 status = ocfs2_begin_ordered_truncate(inode,
944 attr->ia_size);
945 if (status)
946 goto bail_unlock;
947 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800948 status = ocfs2_truncate_file(inode, bh, attr->ia_size);
Joel Becker2b4e30f2008-09-03 20:03:41 -0700949 } else
Mark Fasheh65ed39d2007-08-28 17:13:23 -0700950 status = ocfs2_extend_file(inode, bh, attr->ia_size);
Mark Fashehccd979b2005-12-15 14:31:24 -0800951 if (status < 0) {
952 if (status != -ENOSPC)
953 mlog_errno(status);
954 status = -ENOSPC;
955 goto bail_unlock;
956 }
957 }
958
Mark Fasheh65eff9c2006-10-09 17:26:22 -0700959 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
Mark Fashehccd979b2005-12-15 14:31:24 -0800960 if (IS_ERR(handle)) {
961 status = PTR_ERR(handle);
962 mlog_errno(status);
963 goto bail_unlock;
964 }
965
Mark Fasheh7307de82007-05-09 15:16:19 -0700966 /*
967 * This will intentionally not wind up calling vmtruncate(),
968 * since all the work for a size change has been done above.
969 * Otherwise, we could get into problems with truncate as
970 * ip_alloc_sem is used there to protect against i_size
971 * changes.
972 */
Mark Fashehccd979b2005-12-15 14:31:24 -0800973 status = inode_setattr(inode, attr);
974 if (status < 0) {
975 mlog_errno(status);
976 goto bail_commit;
977 }
978
979 status = ocfs2_mark_inode_dirty(handle, inode, bh);
980 if (status < 0)
981 mlog_errno(status);
982
983bail_commit:
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700984 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800985bail_unlock:
Mark Fashehe63aecb62007-10-18 15:30:42 -0700986 ocfs2_inode_unlock(inode, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -0800987bail_unlock_rw:
988 if (size_change)
989 ocfs2_rw_unlock(inode, 1);
990bail:
Mark Fasheha81cb882008-10-07 14:25:16 -0700991 brelse(bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800992
Tiger Yang060bc662008-11-14 11:17:29 +0800993 if (!status && attr->ia_valid & ATTR_MODE) {
994 status = ocfs2_acl_chmod(inode);
995 if (status < 0)
996 mlog_errno(status);
997 }
998
Mark Fashehccd979b2005-12-15 14:31:24 -0800999 mlog_exit(status);
1000 return status;
1001}
1002
1003int ocfs2_getattr(struct vfsmount *mnt,
1004 struct dentry *dentry,
1005 struct kstat *stat)
1006{
1007 struct inode *inode = dentry->d_inode;
1008 struct super_block *sb = dentry->d_inode->i_sb;
1009 struct ocfs2_super *osb = sb->s_fs_info;
1010 int err;
1011
1012 mlog_entry_void();
1013
1014 err = ocfs2_inode_revalidate(dentry);
1015 if (err) {
1016 if (err != -ENOENT)
1017 mlog_errno(err);
1018 goto bail;
1019 }
1020
1021 generic_fillattr(inode, stat);
1022
1023 /* We set the blksize from the cluster size for performance */
1024 stat->blksize = osb->s_clustersize;
1025
1026bail:
1027 mlog_exit(err);
1028
1029 return err;
1030}
1031
Al Viroe6305c42008-07-15 21:03:57 -04001032int ocfs2_permission(struct inode *inode, int mask)
Tiger Yangd38eb8d2006-11-27 09:59:21 +08001033{
1034 int ret;
1035
1036 mlog_entry_void();
1037
Mark Fashehe63aecb62007-10-18 15:30:42 -07001038 ret = ocfs2_inode_lock(inode, NULL, 0);
Tiger Yangd38eb8d2006-11-27 09:59:21 +08001039 if (ret) {
Mark Fasheha9f5f702007-04-26 11:43:43 -07001040 if (ret != -ENOENT)
1041 mlog_errno(ret);
Tiger Yangd38eb8d2006-11-27 09:59:21 +08001042 goto out;
1043 }
1044
Tiger Yang23fc2702008-11-14 11:17:18 +08001045 ret = generic_permission(inode, mask, ocfs2_check_acl);
Tiger Yangd38eb8d2006-11-27 09:59:21 +08001046
Mark Fashehe63aecb62007-10-18 15:30:42 -07001047 ocfs2_inode_unlock(inode, 0);
Tiger Yangd38eb8d2006-11-27 09:59:21 +08001048out:
1049 mlog_exit(ret);
1050 return ret;
1051}
1052
Mark Fashehb2580102007-03-09 16:53:21 -08001053static int __ocfs2_write_remove_suid(struct inode *inode,
1054 struct buffer_head *bh)
Mark Fashehccd979b2005-12-15 14:31:24 -08001055{
1056 int ret;
Mark Fasheh1fabe142006-10-09 18:11:45 -07001057 handle_t *handle;
Mark Fashehccd979b2005-12-15 14:31:24 -08001058 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1059 struct ocfs2_dinode *di;
1060
Mark Fashehb0697052006-03-03 10:24:33 -08001061 mlog_entry("(Inode %llu, mode 0%o)\n",
Mark Fashehb2580102007-03-09 16:53:21 -08001062 (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_mode);
Mark Fashehccd979b2005-12-15 14:31:24 -08001063
Mark Fasheh65eff9c2006-10-09 17:26:22 -07001064 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
Jan Karafa38e922008-10-20 19:23:51 +02001065 if (IS_ERR(handle)) {
1066 ret = PTR_ERR(handle);
Mark Fashehccd979b2005-12-15 14:31:24 -08001067 mlog_errno(ret);
1068 goto out;
1069 }
1070
Mark Fashehccd979b2005-12-15 14:31:24 -08001071 ret = ocfs2_journal_access(handle, inode, bh,
1072 OCFS2_JOURNAL_ACCESS_WRITE);
1073 if (ret < 0) {
1074 mlog_errno(ret);
Mark Fashehb2580102007-03-09 16:53:21 -08001075 goto out_trans;
Mark Fashehccd979b2005-12-15 14:31:24 -08001076 }
1077
1078 inode->i_mode &= ~S_ISUID;
1079 if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
1080 inode->i_mode &= ~S_ISGID;
1081
1082 di = (struct ocfs2_dinode *) bh->b_data;
1083 di->i_mode = cpu_to_le16(inode->i_mode);
1084
1085 ret = ocfs2_journal_dirty(handle, bh);
1086 if (ret < 0)
1087 mlog_errno(ret);
Mark Fashehb2580102007-03-09 16:53:21 -08001088
Mark Fashehccd979b2005-12-15 14:31:24 -08001089out_trans:
Mark Fasheh02dc1af2006-10-09 16:48:10 -07001090 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -08001091out:
1092 mlog_exit(ret);
1093 return ret;
1094}
1095
Mark Fasheh9517bac2007-02-09 20:24:12 -08001096/*
1097 * Will look for holes and unwritten extents in the range starting at
1098 * pos for count bytes (inclusive).
1099 */
1100static int ocfs2_check_range_for_holes(struct inode *inode, loff_t pos,
1101 size_t count)
1102{
1103 int ret = 0;
Mark Fasheh49cb8d22007-03-09 16:21:46 -08001104 unsigned int extent_flags;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001105 u32 cpos, clusters, extent_len, phys_cpos;
1106 struct super_block *sb = inode->i_sb;
1107
1108 cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
1109 clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
1110
1111 while (clusters) {
Mark Fasheh49cb8d22007-03-09 16:21:46 -08001112 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
1113 &extent_flags);
Mark Fasheh9517bac2007-02-09 20:24:12 -08001114 if (ret < 0) {
1115 mlog_errno(ret);
1116 goto out;
1117 }
1118
Mark Fasheh49cb8d22007-03-09 16:21:46 -08001119 if (phys_cpos == 0 || (extent_flags & OCFS2_EXT_UNWRITTEN)) {
Mark Fasheh9517bac2007-02-09 20:24:12 -08001120 ret = 1;
1121 break;
1122 }
1123
1124 if (extent_len > clusters)
1125 extent_len = clusters;
1126
1127 clusters -= extent_len;
1128 cpos += extent_len;
1129 }
1130out:
1131 return ret;
1132}
1133
Mark Fashehb2580102007-03-09 16:53:21 -08001134static int ocfs2_write_remove_suid(struct inode *inode)
1135{
1136 int ret;
1137 struct buffer_head *bh = NULL;
1138 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1139
Joel Becker0fcaa562008-10-09 17:20:31 -07001140 ret = ocfs2_read_block(inode, oi->ip_blkno, &bh);
Mark Fashehb2580102007-03-09 16:53:21 -08001141 if (ret < 0) {
1142 mlog_errno(ret);
1143 goto out;
1144 }
1145
1146 ret = __ocfs2_write_remove_suid(inode, bh);
1147out:
1148 brelse(bh);
1149 return ret;
1150}
1151
Mark Fasheh2ae99a62007-03-09 16:43:28 -08001152/*
1153 * Allocate enough extents to cover the region starting at byte offset
1154 * start for len bytes. Existing extents are skipped, any extents
1155 * added are marked as "unwritten".
1156 */
1157static int ocfs2_allocate_unwritten_extents(struct inode *inode,
1158 u64 start, u64 len)
1159{
1160 int ret;
1161 u32 cpos, phys_cpos, clusters, alloc_size;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001162 u64 end = start + len;
1163 struct buffer_head *di_bh = NULL;
1164
1165 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
Joel Becker31d33072008-10-09 17:20:30 -07001166 ret = ocfs2_read_block(inode, OCFS2_I(inode)->ip_blkno,
Joel Becker0fcaa562008-10-09 17:20:31 -07001167 &di_bh);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001168 if (ret) {
1169 mlog_errno(ret);
1170 goto out;
1171 }
1172
1173 /*
1174 * Nothing to do if the requested reservation range
1175 * fits within the inode.
1176 */
1177 if (ocfs2_size_fits_inline_data(di_bh, end))
1178 goto out;
1179
1180 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1181 if (ret) {
1182 mlog_errno(ret);
1183 goto out;
1184 }
1185 }
Mark Fasheh2ae99a62007-03-09 16:43:28 -08001186
1187 /*
1188 * We consider both start and len to be inclusive.
1189 */
1190 cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
1191 clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len);
1192 clusters -= cpos;
1193
1194 while (clusters) {
1195 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1196 &alloc_size, NULL);
1197 if (ret) {
1198 mlog_errno(ret);
1199 goto out;
1200 }
1201
1202 /*
1203 * Hole or existing extent len can be arbitrary, so
1204 * cap it to our own allocation request.
1205 */
1206 if (alloc_size > clusters)
1207 alloc_size = clusters;
1208
1209 if (phys_cpos) {
1210 /*
1211 * We already have an allocation at this
1212 * region so we can safely skip it.
1213 */
1214 goto next;
1215 }
1216
1217 ret = __ocfs2_extend_allocation(inode, cpos, alloc_size, 1);
1218 if (ret) {
1219 if (ret != -ENOSPC)
1220 mlog_errno(ret);
1221 goto out;
1222 }
1223
1224next:
1225 cpos += alloc_size;
1226 clusters -= alloc_size;
1227 }
1228
1229 ret = 0;
1230out:
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001231
1232 brelse(di_bh);
Mark Fasheh2ae99a62007-03-09 16:43:28 -08001233 return ret;
1234}
1235
Mark Fasheh063c4562007-07-03 13:34:11 -07001236/*
1237 * Truncate a byte range, avoiding pages within partial clusters. This
1238 * preserves those pages for the zeroing code to write to.
1239 */
1240static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start,
1241 u64 byte_len)
1242{
1243 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1244 loff_t start, end;
1245 struct address_space *mapping = inode->i_mapping;
1246
1247 start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start);
1248 end = byte_start + byte_len;
1249 end = end & ~(osb->s_clustersize - 1);
1250
1251 if (start < end) {
1252 unmap_mapping_range(mapping, start, end - start, 0);
1253 truncate_inode_pages_range(mapping, start, end - 1);
1254 }
1255}
1256
1257static int ocfs2_zero_partial_clusters(struct inode *inode,
1258 u64 start, u64 len)
1259{
1260 int ret = 0;
1261 u64 tmpend, end = start + len;
1262 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1263 unsigned int csize = osb->s_clustersize;
1264 handle_t *handle;
1265
1266 /*
1267 * The "start" and "end" values are NOT necessarily part of
1268 * the range whose allocation is being deleted. Rather, this
1269 * is what the user passed in with the request. We must zero
1270 * partial clusters here. There's no need to worry about
1271 * physical allocation - the zeroing code knows to skip holes.
1272 */
1273 mlog(0, "byte start: %llu, end: %llu\n",
1274 (unsigned long long)start, (unsigned long long)end);
1275
1276 /*
1277 * If both edges are on a cluster boundary then there's no
1278 * zeroing required as the region is part of the allocation to
1279 * be truncated.
1280 */
1281 if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0)
1282 goto out;
1283
1284 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
Jan Karafa38e922008-10-20 19:23:51 +02001285 if (IS_ERR(handle)) {
1286 ret = PTR_ERR(handle);
Mark Fasheh063c4562007-07-03 13:34:11 -07001287 mlog_errno(ret);
1288 goto out;
1289 }
1290
1291 /*
1292 * We want to get the byte offset of the end of the 1st cluster.
1293 */
1294 tmpend = (u64)osb->s_clustersize + (start & ~(osb->s_clustersize - 1));
1295 if (tmpend > end)
1296 tmpend = end;
1297
1298 mlog(0, "1st range: start: %llu, tmpend: %llu\n",
1299 (unsigned long long)start, (unsigned long long)tmpend);
1300
1301 ret = ocfs2_zero_range_for_truncate(inode, handle, start, tmpend);
1302 if (ret)
1303 mlog_errno(ret);
1304
1305 if (tmpend < end) {
1306 /*
1307 * This may make start and end equal, but the zeroing
1308 * code will skip any work in that case so there's no
1309 * need to catch it up here.
1310 */
1311 start = end & ~(osb->s_clustersize - 1);
1312
1313 mlog(0, "2nd range: start: %llu, end: %llu\n",
1314 (unsigned long long)start, (unsigned long long)end);
1315
1316 ret = ocfs2_zero_range_for_truncate(inode, handle, start, end);
1317 if (ret)
1318 mlog_errno(ret);
1319 }
1320
1321 ocfs2_commit_trans(osb, handle);
1322out:
1323 return ret;
1324}
1325
1326static int ocfs2_remove_inode_range(struct inode *inode,
1327 struct buffer_head *di_bh, u64 byte_start,
1328 u64 byte_len)
1329{
1330 int ret = 0;
1331 u32 trunc_start, trunc_len, cpos, phys_cpos, alloc_size;
1332 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1333 struct ocfs2_cached_dealloc_ctxt dealloc;
Mark Fashehb1967d02007-11-20 11:56:39 -08001334 struct address_space *mapping = inode->i_mapping;
Mark Fashehfecc0112008-11-12 15:16:38 -08001335 struct ocfs2_extent_tree et;
Mark Fasheh063c4562007-07-03 13:34:11 -07001336
Mark Fashehfecc0112008-11-12 15:16:38 -08001337 ocfs2_init_dinode_extent_tree(&et, inode, di_bh);
Mark Fasheh063c4562007-07-03 13:34:11 -07001338 ocfs2_init_dealloc_ctxt(&dealloc);
1339
1340 if (byte_len == 0)
1341 return 0;
1342
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001343 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1344 ret = ocfs2_truncate_inline(inode, di_bh, byte_start,
Mark Fashehb1967d02007-11-20 11:56:39 -08001345 byte_start + byte_len, 0);
1346 if (ret) {
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001347 mlog_errno(ret);
Mark Fashehb1967d02007-11-20 11:56:39 -08001348 goto out;
1349 }
1350 /*
1351 * There's no need to get fancy with the page cache
1352 * truncate of an inline-data inode. We're talking
1353 * about less than a page here, which will be cached
1354 * in the dinode buffer anyway.
1355 */
1356 unmap_mapping_range(mapping, 0, 0, 0);
1357 truncate_inode_pages(mapping, 0);
1358 goto out;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001359 }
1360
Mark Fasheh063c4562007-07-03 13:34:11 -07001361 trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start);
1362 trunc_len = (byte_start + byte_len) >> osb->s_clustersize_bits;
1363 if (trunc_len >= trunc_start)
1364 trunc_len -= trunc_start;
1365 else
1366 trunc_len = 0;
1367
1368 mlog(0, "Inode: %llu, start: %llu, len: %llu, cstart: %u, clen: %u\n",
1369 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1370 (unsigned long long)byte_start,
1371 (unsigned long long)byte_len, trunc_start, trunc_len);
1372
1373 ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len);
1374 if (ret) {
1375 mlog_errno(ret);
1376 goto out;
1377 }
1378
1379 cpos = trunc_start;
1380 while (trunc_len) {
1381 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1382 &alloc_size, NULL);
1383 if (ret) {
1384 mlog_errno(ret);
1385 goto out;
1386 }
1387
1388 if (alloc_size > trunc_len)
1389 alloc_size = trunc_len;
1390
1391 /* Only do work for non-holes */
1392 if (phys_cpos != 0) {
Mark Fashehfecc0112008-11-12 15:16:38 -08001393 ret = ocfs2_remove_btree_range(inode, &et, cpos,
1394 phys_cpos, alloc_size,
1395 &dealloc);
Mark Fasheh063c4562007-07-03 13:34:11 -07001396 if (ret) {
1397 mlog_errno(ret);
1398 goto out;
1399 }
1400 }
1401
1402 cpos += alloc_size;
1403 trunc_len -= alloc_size;
1404 }
1405
1406 ocfs2_truncate_cluster_pages(inode, byte_start, byte_len);
1407
1408out:
1409 ocfs2_schedule_truncate_log_flush(osb, 1);
1410 ocfs2_run_deallocs(osb, &dealloc);
1411
1412 return ret;
1413}
1414
Mark Fashehb2580102007-03-09 16:53:21 -08001415/*
1416 * Parts of this function taken from xfs_change_file_space()
1417 */
Mark Fasheh385820a2007-07-19 00:14:38 -07001418static int __ocfs2_change_file_space(struct file *file, struct inode *inode,
1419 loff_t f_pos, unsigned int cmd,
1420 struct ocfs2_space_resv *sr,
1421 int change_size)
Mark Fashehb2580102007-03-09 16:53:21 -08001422{
1423 int ret;
1424 s64 llen;
Mark Fasheh385820a2007-07-19 00:14:38 -07001425 loff_t size;
Mark Fashehb2580102007-03-09 16:53:21 -08001426 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1427 struct buffer_head *di_bh = NULL;
1428 handle_t *handle;
Mark Fasheha00cce32007-07-20 11:28:30 -07001429 unsigned long long max_off = inode->i_sb->s_maxbytes;
Mark Fashehb2580102007-03-09 16:53:21 -08001430
Mark Fashehb2580102007-03-09 16:53:21 -08001431 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
1432 return -EROFS;
1433
1434 mutex_lock(&inode->i_mutex);
1435
1436 /*
1437 * This prevents concurrent writes on other nodes
1438 */
1439 ret = ocfs2_rw_lock(inode, 1);
1440 if (ret) {
1441 mlog_errno(ret);
1442 goto out;
1443 }
1444
Mark Fashehe63aecb62007-10-18 15:30:42 -07001445 ret = ocfs2_inode_lock(inode, &di_bh, 1);
Mark Fashehb2580102007-03-09 16:53:21 -08001446 if (ret) {
1447 mlog_errno(ret);
1448 goto out_rw_unlock;
1449 }
1450
1451 if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
1452 ret = -EPERM;
Mark Fashehe63aecb62007-10-18 15:30:42 -07001453 goto out_inode_unlock;
Mark Fashehb2580102007-03-09 16:53:21 -08001454 }
1455
1456 switch (sr->l_whence) {
1457 case 0: /*SEEK_SET*/
1458 break;
1459 case 1: /*SEEK_CUR*/
Mark Fasheh385820a2007-07-19 00:14:38 -07001460 sr->l_start += f_pos;
Mark Fashehb2580102007-03-09 16:53:21 -08001461 break;
1462 case 2: /*SEEK_END*/
1463 sr->l_start += i_size_read(inode);
1464 break;
1465 default:
1466 ret = -EINVAL;
Mark Fashehe63aecb62007-10-18 15:30:42 -07001467 goto out_inode_unlock;
Mark Fashehb2580102007-03-09 16:53:21 -08001468 }
1469 sr->l_whence = 0;
1470
1471 llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len;
1472
1473 if (sr->l_start < 0
1474 || sr->l_start > max_off
1475 || (sr->l_start + llen) < 0
1476 || (sr->l_start + llen) > max_off) {
1477 ret = -EINVAL;
Mark Fashehe63aecb62007-10-18 15:30:42 -07001478 goto out_inode_unlock;
Mark Fashehb2580102007-03-09 16:53:21 -08001479 }
Mark Fasheh385820a2007-07-19 00:14:38 -07001480 size = sr->l_start + sr->l_len;
Mark Fashehb2580102007-03-09 16:53:21 -08001481
1482 if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) {
1483 if (sr->l_len <= 0) {
1484 ret = -EINVAL;
Mark Fashehe63aecb62007-10-18 15:30:42 -07001485 goto out_inode_unlock;
Mark Fashehb2580102007-03-09 16:53:21 -08001486 }
1487 }
1488
Mark Fasheh385820a2007-07-19 00:14:38 -07001489 if (file && should_remove_suid(file->f_path.dentry)) {
Mark Fashehb2580102007-03-09 16:53:21 -08001490 ret = __ocfs2_write_remove_suid(inode, di_bh);
1491 if (ret) {
1492 mlog_errno(ret);
Mark Fashehe63aecb62007-10-18 15:30:42 -07001493 goto out_inode_unlock;
Mark Fashehb2580102007-03-09 16:53:21 -08001494 }
1495 }
1496
1497 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1498 switch (cmd) {
1499 case OCFS2_IOC_RESVSP:
1500 case OCFS2_IOC_RESVSP64:
1501 /*
1502 * This takes unsigned offsets, but the signed ones we
1503 * pass have been checked against overflow above.
1504 */
1505 ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start,
1506 sr->l_len);
1507 break;
1508 case OCFS2_IOC_UNRESVSP:
1509 case OCFS2_IOC_UNRESVSP64:
1510 ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start,
1511 sr->l_len);
1512 break;
1513 default:
1514 ret = -EINVAL;
1515 }
1516 up_write(&OCFS2_I(inode)->ip_alloc_sem);
1517 if (ret) {
1518 mlog_errno(ret);
Mark Fashehe63aecb62007-10-18 15:30:42 -07001519 goto out_inode_unlock;
Mark Fashehb2580102007-03-09 16:53:21 -08001520 }
1521
1522 /*
1523 * We update c/mtime for these changes
1524 */
1525 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1526 if (IS_ERR(handle)) {
1527 ret = PTR_ERR(handle);
1528 mlog_errno(ret);
Mark Fashehe63aecb62007-10-18 15:30:42 -07001529 goto out_inode_unlock;
Mark Fashehb2580102007-03-09 16:53:21 -08001530 }
1531
Mark Fasheh385820a2007-07-19 00:14:38 -07001532 if (change_size && i_size_read(inode) < size)
1533 i_size_write(inode, size);
1534
Mark Fashehb2580102007-03-09 16:53:21 -08001535 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1536 ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
1537 if (ret < 0)
1538 mlog_errno(ret);
1539
1540 ocfs2_commit_trans(osb, handle);
1541
Mark Fashehe63aecb62007-10-18 15:30:42 -07001542out_inode_unlock:
Mark Fashehb2580102007-03-09 16:53:21 -08001543 brelse(di_bh);
Mark Fashehe63aecb62007-10-18 15:30:42 -07001544 ocfs2_inode_unlock(inode, 1);
Mark Fashehb2580102007-03-09 16:53:21 -08001545out_rw_unlock:
1546 ocfs2_rw_unlock(inode, 1);
1547
Mark Fashehb2580102007-03-09 16:53:21 -08001548out:
Julia Lawallc259ae52008-07-21 09:59:15 +02001549 mutex_unlock(&inode->i_mutex);
Mark Fashehb2580102007-03-09 16:53:21 -08001550 return ret;
1551}
1552
Mark Fasheh385820a2007-07-19 00:14:38 -07001553int ocfs2_change_file_space(struct file *file, unsigned int cmd,
1554 struct ocfs2_space_resv *sr)
1555{
1556 struct inode *inode = file->f_path.dentry->d_inode;
1557 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);;
1558
1559 if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) &&
1560 !ocfs2_writes_unwritten_extents(osb))
1561 return -ENOTTY;
1562 else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) &&
1563 !ocfs2_sparse_alloc(osb))
1564 return -ENOTTY;
1565
1566 if (!S_ISREG(inode->i_mode))
1567 return -EINVAL;
1568
1569 if (!(file->f_mode & FMODE_WRITE))
1570 return -EBADF;
1571
1572 return __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0);
1573}
1574
1575static long ocfs2_fallocate(struct inode *inode, int mode, loff_t offset,
1576 loff_t len)
1577{
1578 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1579 struct ocfs2_space_resv sr;
1580 int change_size = 1;
1581
1582 if (!ocfs2_writes_unwritten_extents(osb))
1583 return -EOPNOTSUPP;
1584
1585 if (S_ISDIR(inode->i_mode))
1586 return -ENODEV;
1587
1588 if (mode & FALLOC_FL_KEEP_SIZE)
1589 change_size = 0;
1590
1591 sr.l_whence = 0;
1592 sr.l_start = (s64)offset;
1593 sr.l_len = (s64)len;
1594
1595 return __ocfs2_change_file_space(NULL, inode, offset,
1596 OCFS2_IOC_RESVSP64, &sr, change_size);
1597}
1598
Tiger Yang8659ac22006-10-17 18:29:52 -07001599static int ocfs2_prepare_inode_for_write(struct dentry *dentry,
1600 loff_t *ppos,
1601 size_t count,
Mark Fasheh9517bac2007-02-09 20:24:12 -08001602 int appending,
1603 int *direct_io)
Mark Fashehccd979b2005-12-15 14:31:24 -08001604{
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001605 int ret = 0, meta_level = 0;
Tiger Yang8659ac22006-10-17 18:29:52 -07001606 struct inode *inode = dentry->d_inode;
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001607 loff_t saved_pos, end;
Mark Fashehccd979b2005-12-15 14:31:24 -08001608
Mark Fashehccd979b2005-12-15 14:31:24 -08001609 /*
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001610 * We start with a read level meta lock and only jump to an ex
1611 * if we need to make modifications here.
Mark Fashehccd979b2005-12-15 14:31:24 -08001612 */
Mark Fashehccd979b2005-12-15 14:31:24 -08001613 for(;;) {
Mark Fashehe63aecb62007-10-18 15:30:42 -07001614 ret = ocfs2_inode_lock(inode, NULL, meta_level);
Mark Fashehccd979b2005-12-15 14:31:24 -08001615 if (ret < 0) {
1616 meta_level = -1;
1617 mlog_errno(ret);
1618 goto out;
1619 }
1620
1621 /* Clear suid / sgid if necessary. We do this here
1622 * instead of later in the write path because
1623 * remove_suid() calls ->setattr without any hint that
1624 * we may have already done our cluster locking. Since
1625 * ocfs2_setattr() *must* take cluster locks to
1626 * proceeed, this will lead us to recursively lock the
1627 * inode. There's also the dinode i_size state which
1628 * can be lost via setattr during extending writes (we
1629 * set inode->i_size at the end of a write. */
Tiger Yang8659ac22006-10-17 18:29:52 -07001630 if (should_remove_suid(dentry)) {
Mark Fashehccd979b2005-12-15 14:31:24 -08001631 if (meta_level == 0) {
Mark Fashehe63aecb62007-10-18 15:30:42 -07001632 ocfs2_inode_unlock(inode, meta_level);
Mark Fashehccd979b2005-12-15 14:31:24 -08001633 meta_level = 1;
1634 continue;
1635 }
1636
1637 ret = ocfs2_write_remove_suid(inode);
1638 if (ret < 0) {
1639 mlog_errno(ret);
Tiger Yang8659ac22006-10-17 18:29:52 -07001640 goto out_unlock;
Mark Fashehccd979b2005-12-15 14:31:24 -08001641 }
1642 }
1643
1644 /* work on a copy of ppos until we're sure that we won't have
1645 * to recalculate it due to relocking. */
Tiger Yang8659ac22006-10-17 18:29:52 -07001646 if (appending) {
Mark Fashehccd979b2005-12-15 14:31:24 -08001647 saved_pos = i_size_read(inode);
1648 mlog(0, "O_APPEND: inode->i_size=%llu\n", saved_pos);
1649 } else {
Tiger Yang8659ac22006-10-17 18:29:52 -07001650 saved_pos = *ppos;
Mark Fashehccd979b2005-12-15 14:31:24 -08001651 }
Mark Fasheh3a0782d2007-01-17 12:53:31 -08001652
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001653 end = saved_pos + count;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001654
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001655 /*
1656 * Skip the O_DIRECT checks if we don't need
1657 * them.
1658 */
1659 if (!direct_io || !(*direct_io))
1660 break;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001661
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001662 /*
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001663 * There's no sane way to do direct writes to an inode
1664 * with inline data.
1665 */
1666 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1667 *direct_io = 0;
1668 break;
1669 }
1670
1671 /*
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001672 * Allowing concurrent direct writes means
1673 * i_size changes wouldn't be synchronized, so
1674 * one node could wind up truncating another
1675 * nodes writes.
1676 */
1677 if (end > i_size_read(inode)) {
1678 *direct_io = 0;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001679 break;
1680 }
1681
Mark Fasheh3a0782d2007-01-17 12:53:31 -08001682 /*
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001683 * We don't fill holes during direct io, so
1684 * check for them here. If any are found, the
1685 * caller will have to retake some cluster
1686 * locks and initiate the io as buffered.
Mark Fasheh3a0782d2007-01-17 12:53:31 -08001687 */
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001688 ret = ocfs2_check_range_for_holes(inode, saved_pos, count);
1689 if (ret == 1) {
1690 *direct_io = 0;
1691 ret = 0;
1692 } else if (ret < 0)
1693 mlog_errno(ret);
Mark Fashehccd979b2005-12-15 14:31:24 -08001694 break;
1695 }
1696
Tiger Yang8659ac22006-10-17 18:29:52 -07001697 if (appending)
1698 *ppos = saved_pos;
1699
1700out_unlock:
Mark Fashehe63aecb62007-10-18 15:30:42 -07001701 ocfs2_inode_unlock(inode, meta_level);
Tiger Yang8659ac22006-10-17 18:29:52 -07001702
1703out:
1704 return ret;
1705}
1706
1707static ssize_t ocfs2_file_aio_write(struct kiocb *iocb,
1708 const struct iovec *iov,
1709 unsigned long nr_segs,
1710 loff_t pos)
1711{
Mark Fasheh9517bac2007-02-09 20:24:12 -08001712 int ret, direct_io, appending, rw_level, have_alloc_sem = 0;
Nick Pigginb6af1bc2007-10-16 01:25:24 -07001713 int can_do_direct;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001714 ssize_t written = 0;
1715 size_t ocount; /* original count */
1716 size_t count; /* after file limit checks */
Mark Fasheh9ea2d322007-10-18 14:14:45 -07001717 loff_t old_size, *ppos = &iocb->ki_pos;
1718 u32 old_clusters;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001719 struct file *file = iocb->ki_filp;
1720 struct inode *inode = file->f_path.dentry->d_inode;
Mark Fasheh9ea2d322007-10-18 14:14:45 -07001721 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
Tiger Yang8659ac22006-10-17 18:29:52 -07001722
Mark Fasheh9517bac2007-02-09 20:24:12 -08001723 mlog_entry("(0x%p, %u, '%.*s')\n", file,
Tiger Yang8659ac22006-10-17 18:29:52 -07001724 (unsigned int)nr_segs,
Mark Fasheh9517bac2007-02-09 20:24:12 -08001725 file->f_path.dentry->d_name.len,
1726 file->f_path.dentry->d_name.name);
Tiger Yang8659ac22006-10-17 18:29:52 -07001727
Tiger Yang8659ac22006-10-17 18:29:52 -07001728 if (iocb->ki_left == 0)
1729 return 0;
1730
Mark Fasheh9517bac2007-02-09 20:24:12 -08001731 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
1732
1733 appending = file->f_flags & O_APPEND ? 1 : 0;
1734 direct_io = file->f_flags & O_DIRECT ? 1 : 0;
1735
Tiger Yang8659ac22006-10-17 18:29:52 -07001736 mutex_lock(&inode->i_mutex);
Mark Fasheh9517bac2007-02-09 20:24:12 -08001737
1738relock:
Tiger Yang8659ac22006-10-17 18:29:52 -07001739 /* to match setattr's i_mutex -> i_alloc_sem -> rw_lock ordering */
Mark Fasheh9517bac2007-02-09 20:24:12 -08001740 if (direct_io) {
Tiger Yang8659ac22006-10-17 18:29:52 -07001741 down_read(&inode->i_alloc_sem);
Mark Fasheh9517bac2007-02-09 20:24:12 -08001742 have_alloc_sem = 1;
Tiger Yang8659ac22006-10-17 18:29:52 -07001743 }
1744
1745 /* concurrent O_DIRECT writes are allowed */
Mark Fasheh9517bac2007-02-09 20:24:12 -08001746 rw_level = !direct_io;
Tiger Yang8659ac22006-10-17 18:29:52 -07001747 ret = ocfs2_rw_lock(inode, rw_level);
1748 if (ret < 0) {
Mark Fasheh9517bac2007-02-09 20:24:12 -08001749 mlog_errno(ret);
1750 goto out_sems;
1751 }
1752
1753 can_do_direct = direct_io;
1754 ret = ocfs2_prepare_inode_for_write(file->f_path.dentry, ppos,
1755 iocb->ki_left, appending,
1756 &can_do_direct);
1757 if (ret < 0) {
Tiger Yang8659ac22006-10-17 18:29:52 -07001758 mlog_errno(ret);
1759 goto out;
1760 }
1761
Mark Fasheh9517bac2007-02-09 20:24:12 -08001762 /*
1763 * We can't complete the direct I/O as requested, fall back to
1764 * buffered I/O.
1765 */
1766 if (direct_io && !can_do_direct) {
1767 ocfs2_rw_unlock(inode, rw_level);
1768 up_read(&inode->i_alloc_sem);
1769
1770 have_alloc_sem = 0;
1771 rw_level = -1;
1772
1773 direct_io = 0;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001774 goto relock;
Tiger Yang8659ac22006-10-17 18:29:52 -07001775 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001776
Mark Fasheh9ea2d322007-10-18 14:14:45 -07001777 /*
1778 * To later detect whether a journal commit for sync writes is
1779 * necessary, we sample i_size, and cluster count here.
1780 */
1781 old_size = i_size_read(inode);
1782 old_clusters = OCFS2_I(inode)->ip_clusters;
1783
Mark Fashehccd979b2005-12-15 14:31:24 -08001784 /* communicate with ocfs2_dio_end_io */
Mark Fasheh7cdfc3a2007-04-16 17:28:51 -07001785 ocfs2_iocb_set_rw_locked(iocb, rw_level);
Mark Fashehccd979b2005-12-15 14:31:24 -08001786
Mark Fasheh9517bac2007-02-09 20:24:12 -08001787 if (direct_io) {
Nick Pigginb6af1bc2007-10-16 01:25:24 -07001788 ret = generic_segment_checks(iov, &nr_segs, &ocount,
1789 VERIFY_READ);
1790 if (ret)
1791 goto out_dio;
1792
1793 ret = generic_write_checks(file, ppos, &count,
1794 S_ISBLK(inode->i_mode));
1795 if (ret)
1796 goto out_dio;
1797
Mark Fasheh9517bac2007-02-09 20:24:12 -08001798 written = generic_file_direct_write(iocb, iov, &nr_segs, *ppos,
1799 ppos, count, ocount);
1800 if (written < 0) {
Dmitri Monakhovc4354002008-10-27 13:01:49 -07001801 /*
1802 * direct write may have instantiated a few
1803 * blocks outside i_size. Trim these off again.
1804 * Don't need i_size_read because we hold i_mutex.
1805 */
1806 if (*ppos + count > inode->i_size)
1807 vmtruncate(inode, inode->i_size);
Mark Fasheh9517bac2007-02-09 20:24:12 -08001808 ret = written;
1809 goto out_dio;
1810 }
1811 } else {
Nick Pigginb6af1bc2007-10-16 01:25:24 -07001812 written = generic_file_aio_write_nolock(iocb, iov, nr_segs,
1813 *ppos);
Mark Fasheh9517bac2007-02-09 20:24:12 -08001814 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001815
Mark Fasheh9517bac2007-02-09 20:24:12 -08001816out_dio:
Mark Fashehccd979b2005-12-15 14:31:24 -08001817 /* buffered aio wouldn't have proper lock coverage today */
Mark Fasheh9517bac2007-02-09 20:24:12 -08001818 BUG_ON(ret == -EIOCBQUEUED && !(file->f_flags & O_DIRECT));
Mark Fashehccd979b2005-12-15 14:31:24 -08001819
Mark Fasheh9ea2d322007-10-18 14:14:45 -07001820 if ((file->f_flags & O_SYNC && !direct_io) || IS_SYNC(inode)) {
1821 /*
1822 * The generic write paths have handled getting data
1823 * to disk, but since we don't make use of the dirty
1824 * inode list, a manual journal commit is necessary
1825 * here.
1826 */
1827 if (old_size != i_size_read(inode) ||
1828 old_clusters != OCFS2_I(inode)->ip_clusters) {
Joel Becker2b4e30f2008-09-03 20:03:41 -07001829 ret = jbd2_journal_force_commit(osb->journal->j_journal);
Mark Fasheh9ea2d322007-10-18 14:14:45 -07001830 if (ret < 0)
1831 written = ret;
1832 }
1833 }
1834
Mark Fashehccd979b2005-12-15 14:31:24 -08001835 /*
1836 * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
1837 * function pointer which is called when o_direct io completes so that
1838 * it can unlock our rw lock. (it's the clustered equivalent of
1839 * i_alloc_sem; protects truncate from racing with pending ios).
1840 * Unfortunately there are error cases which call end_io and others
1841 * that don't. so we don't have to unlock the rw_lock if either an
1842 * async dio is going to do it in the future or an end_io after an
1843 * error has already done it.
1844 */
1845 if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
1846 rw_level = -1;
1847 have_alloc_sem = 0;
1848 }
1849
1850out:
Mark Fasheh9517bac2007-02-09 20:24:12 -08001851 if (rw_level != -1)
1852 ocfs2_rw_unlock(inode, rw_level);
1853
1854out_sems:
Mark Fashehccd979b2005-12-15 14:31:24 -08001855 if (have_alloc_sem)
1856 up_read(&inode->i_alloc_sem);
Mark Fasheh9517bac2007-02-09 20:24:12 -08001857
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001858 mutex_unlock(&inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -08001859
1860 mlog_exit(ret);
Mark Fasheh9517bac2007-02-09 20:24:12 -08001861 return written ? written : ret;
Mark Fashehccd979b2005-12-15 14:31:24 -08001862}
1863
Tiger Yang8659ac22006-10-17 18:29:52 -07001864static ssize_t ocfs2_file_splice_write(struct pipe_inode_info *pipe,
1865 struct file *out,
1866 loff_t *ppos,
1867 size_t len,
1868 unsigned int flags)
1869{
1870 int ret;
Josef Sipekd28c9172006-12-08 02:37:25 -08001871 struct inode *inode = out->f_path.dentry->d_inode;
Tiger Yang8659ac22006-10-17 18:29:52 -07001872
1873 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", out, pipe,
1874 (unsigned int)len,
Josef Sipekd28c9172006-12-08 02:37:25 -08001875 out->f_path.dentry->d_name.len,
1876 out->f_path.dentry->d_name.name);
Tiger Yang8659ac22006-10-17 18:29:52 -07001877
1878 inode_double_lock(inode, pipe->inode);
1879
1880 ret = ocfs2_rw_lock(inode, 1);
1881 if (ret < 0) {
1882 mlog_errno(ret);
1883 goto out;
1884 }
1885
Mark Fasheh9517bac2007-02-09 20:24:12 -08001886 ret = ocfs2_prepare_inode_for_write(out->f_path.dentry, ppos, len, 0,
1887 NULL);
Tiger Yang8659ac22006-10-17 18:29:52 -07001888 if (ret < 0) {
1889 mlog_errno(ret);
1890 goto out_unlock;
1891 }
1892
Nick Pigginb6af1bc2007-10-16 01:25:24 -07001893 ret = generic_file_splice_write_nolock(pipe, out, ppos, len, flags);
Tiger Yang8659ac22006-10-17 18:29:52 -07001894
1895out_unlock:
1896 ocfs2_rw_unlock(inode, 1);
1897out:
1898 inode_double_unlock(inode, pipe->inode);
1899
1900 mlog_exit(ret);
1901 return ret;
1902}
1903
1904static ssize_t ocfs2_file_splice_read(struct file *in,
1905 loff_t *ppos,
1906 struct pipe_inode_info *pipe,
1907 size_t len,
1908 unsigned int flags)
1909{
1910 int ret = 0;
Josef Sipekd28c9172006-12-08 02:37:25 -08001911 struct inode *inode = in->f_path.dentry->d_inode;
Tiger Yang8659ac22006-10-17 18:29:52 -07001912
1913 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", in, pipe,
1914 (unsigned int)len,
Josef Sipekd28c9172006-12-08 02:37:25 -08001915 in->f_path.dentry->d_name.len,
1916 in->f_path.dentry->d_name.name);
Tiger Yang8659ac22006-10-17 18:29:52 -07001917
1918 /*
1919 * See the comment in ocfs2_file_aio_read()
1920 */
Mark Fashehe63aecb62007-10-18 15:30:42 -07001921 ret = ocfs2_inode_lock(inode, NULL, 0);
Tiger Yang8659ac22006-10-17 18:29:52 -07001922 if (ret < 0) {
1923 mlog_errno(ret);
1924 goto bail;
1925 }
Mark Fashehe63aecb62007-10-18 15:30:42 -07001926 ocfs2_inode_unlock(inode, 0);
Tiger Yang8659ac22006-10-17 18:29:52 -07001927
1928 ret = generic_file_splice_read(in, ppos, pipe, len, flags);
1929
1930bail:
1931 mlog_exit(ret);
1932 return ret;
1933}
1934
Mark Fashehccd979b2005-12-15 14:31:24 -08001935static ssize_t ocfs2_file_aio_read(struct kiocb *iocb,
Badari Pulavarty027445c2006-09-30 23:28:46 -07001936 const struct iovec *iov,
1937 unsigned long nr_segs,
Mark Fashehccd979b2005-12-15 14:31:24 -08001938 loff_t pos)
1939{
Tiger Yang25899de2006-11-15 15:49:02 +08001940 int ret = 0, rw_level = -1, have_alloc_sem = 0, lock_level = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -08001941 struct file *filp = iocb->ki_filp;
Josef Sipekd28c9172006-12-08 02:37:25 -08001942 struct inode *inode = filp->f_path.dentry->d_inode;
Mark Fashehccd979b2005-12-15 14:31:24 -08001943
Badari Pulavarty027445c2006-09-30 23:28:46 -07001944 mlog_entry("(0x%p, %u, '%.*s')\n", filp,
1945 (unsigned int)nr_segs,
Josef Sipekd28c9172006-12-08 02:37:25 -08001946 filp->f_path.dentry->d_name.len,
1947 filp->f_path.dentry->d_name.name);
Mark Fashehccd979b2005-12-15 14:31:24 -08001948
1949 if (!inode) {
1950 ret = -EINVAL;
1951 mlog_errno(ret);
1952 goto bail;
1953 }
1954
Mark Fashehccd979b2005-12-15 14:31:24 -08001955 /*
1956 * buffered reads protect themselves in ->readpage(). O_DIRECT reads
1957 * need locks to protect pending reads from racing with truncate.
1958 */
1959 if (filp->f_flags & O_DIRECT) {
1960 down_read(&inode->i_alloc_sem);
1961 have_alloc_sem = 1;
1962
1963 ret = ocfs2_rw_lock(inode, 0);
1964 if (ret < 0) {
1965 mlog_errno(ret);
1966 goto bail;
1967 }
1968 rw_level = 0;
1969 /* communicate with ocfs2_dio_end_io */
Mark Fasheh7cdfc3a2007-04-16 17:28:51 -07001970 ocfs2_iocb_set_rw_locked(iocb, rw_level);
Mark Fashehccd979b2005-12-15 14:31:24 -08001971 }
1972
Mark Fashehc4374f82006-05-05 19:04:35 -07001973 /*
1974 * We're fine letting folks race truncates and extending
1975 * writes with read across the cluster, just like they can
1976 * locally. Hence no rw_lock during read.
1977 *
1978 * Take and drop the meta data lock to update inode fields
1979 * like i_size. This allows the checks down below
1980 * generic_file_aio_read() a chance of actually working.
1981 */
Mark Fashehe63aecb62007-10-18 15:30:42 -07001982 ret = ocfs2_inode_lock_atime(inode, filp->f_vfsmnt, &lock_level);
Mark Fashehc4374f82006-05-05 19:04:35 -07001983 if (ret < 0) {
1984 mlog_errno(ret);
1985 goto bail;
1986 }
Mark Fashehe63aecb62007-10-18 15:30:42 -07001987 ocfs2_inode_unlock(inode, lock_level);
Mark Fashehc4374f82006-05-05 19:04:35 -07001988
Badari Pulavarty027445c2006-09-30 23:28:46 -07001989 ret = generic_file_aio_read(iocb, iov, nr_segs, iocb->ki_pos);
Mark Fashehccd979b2005-12-15 14:31:24 -08001990 if (ret == -EINVAL)
Sunil Mushran56753bd2008-06-09 11:24:41 -07001991 mlog(0, "generic_file_aio_read returned -EINVAL\n");
Mark Fashehccd979b2005-12-15 14:31:24 -08001992
1993 /* buffered aio wouldn't have proper lock coverage today */
1994 BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT));
1995
1996 /* see ocfs2_file_aio_write */
1997 if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
1998 rw_level = -1;
1999 have_alloc_sem = 0;
2000 }
2001
2002bail:
2003 if (have_alloc_sem)
2004 up_read(&inode->i_alloc_sem);
2005 if (rw_level != -1)
2006 ocfs2_rw_unlock(inode, rw_level);
2007 mlog_exit(ret);
2008
2009 return ret;
2010}
2011
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002012const struct inode_operations ocfs2_file_iops = {
Mark Fashehccd979b2005-12-15 14:31:24 -08002013 .setattr = ocfs2_setattr,
2014 .getattr = ocfs2_getattr,
Tiger Yangd38eb8d2006-11-27 09:59:21 +08002015 .permission = ocfs2_permission,
Tiger Yangcf1d6c72008-08-18 17:11:00 +08002016 .setxattr = generic_setxattr,
2017 .getxattr = generic_getxattr,
2018 .listxattr = ocfs2_listxattr,
2019 .removexattr = generic_removexattr,
Mark Fasheh385820a2007-07-19 00:14:38 -07002020 .fallocate = ocfs2_fallocate,
Mark Fasheh00dc4172008-10-03 17:32:11 -04002021 .fiemap = ocfs2_fiemap,
Mark Fashehccd979b2005-12-15 14:31:24 -08002022};
2023
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002024const struct inode_operations ocfs2_special_file_iops = {
Mark Fashehccd979b2005-12-15 14:31:24 -08002025 .setattr = ocfs2_setattr,
2026 .getattr = ocfs2_getattr,
Tiger Yangd38eb8d2006-11-27 09:59:21 +08002027 .permission = ocfs2_permission,
Mark Fashehccd979b2005-12-15 14:31:24 -08002028};
2029
Mark Fasheh53da4932008-07-21 14:29:16 -07002030/*
2031 * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with
2032 * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks!
2033 */
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08002034const struct file_operations ocfs2_fops = {
Jan Kara32c3c0e2007-12-19 15:24:52 +01002035 .llseek = generic_file_llseek,
Mark Fashehccd979b2005-12-15 14:31:24 -08002036 .read = do_sync_read,
2037 .write = do_sync_write,
Mark Fashehccd979b2005-12-15 14:31:24 -08002038 .mmap = ocfs2_mmap,
2039 .fsync = ocfs2_sync_file,
2040 .release = ocfs2_file_release,
2041 .open = ocfs2_file_open,
2042 .aio_read = ocfs2_file_aio_read,
2043 .aio_write = ocfs2_file_aio_write,
Andi Kleenc9ec1482008-01-27 03:17:17 +01002044 .unlocked_ioctl = ocfs2_ioctl,
Mark Fasheh586d2322007-03-09 15:56:28 -08002045#ifdef CONFIG_COMPAT
2046 .compat_ioctl = ocfs2_compat_ioctl,
2047#endif
Mark Fasheh53da4932008-07-21 14:29:16 -07002048 .lock = ocfs2_lock,
2049 .flock = ocfs2_flock,
2050 .splice_read = ocfs2_file_splice_read,
2051 .splice_write = ocfs2_file_splice_write,
2052};
2053
2054const struct file_operations ocfs2_dops = {
2055 .llseek = generic_file_llseek,
2056 .read = generic_read_dir,
2057 .readdir = ocfs2_readdir,
2058 .fsync = ocfs2_sync_file,
2059 .release = ocfs2_dir_release,
2060 .open = ocfs2_dir_open,
2061 .unlocked_ioctl = ocfs2_ioctl,
2062#ifdef CONFIG_COMPAT
2063 .compat_ioctl = ocfs2_compat_ioctl,
2064#endif
2065 .lock = ocfs2_lock,
2066 .flock = ocfs2_flock,
2067};
2068
2069/*
2070 * POSIX-lockless variants of our file_operations.
2071 *
2072 * These will be used if the underlying cluster stack does not support
2073 * posix file locking, if the user passes the "localflocks" mount
2074 * option, or if we have a local-only fs.
2075 *
2076 * ocfs2_flock is in here because all stacks handle UNIX file locks,
2077 * so we still want it in the case of no stack support for
2078 * plocks. Internally, it will do the right thing when asked to ignore
2079 * the cluster.
2080 */
2081const struct file_operations ocfs2_fops_no_plocks = {
2082 .llseek = generic_file_llseek,
2083 .read = do_sync_read,
2084 .write = do_sync_write,
2085 .mmap = ocfs2_mmap,
2086 .fsync = ocfs2_sync_file,
2087 .release = ocfs2_file_release,
2088 .open = ocfs2_file_open,
2089 .aio_read = ocfs2_file_aio_read,
2090 .aio_write = ocfs2_file_aio_write,
2091 .unlocked_ioctl = ocfs2_ioctl,
2092#ifdef CONFIG_COMPAT
2093 .compat_ioctl = ocfs2_compat_ioctl,
2094#endif
Mark Fasheh53fc6222007-12-20 16:49:04 -08002095 .flock = ocfs2_flock,
Tiger Yang8659ac22006-10-17 18:29:52 -07002096 .splice_read = ocfs2_file_splice_read,
2097 .splice_write = ocfs2_file_splice_write,
Mark Fashehccd979b2005-12-15 14:31:24 -08002098};
2099
Mark Fasheh53da4932008-07-21 14:29:16 -07002100const struct file_operations ocfs2_dops_no_plocks = {
Jan Kara32c3c0e2007-12-19 15:24:52 +01002101 .llseek = generic_file_llseek,
Mark Fashehccd979b2005-12-15 14:31:24 -08002102 .read = generic_read_dir,
2103 .readdir = ocfs2_readdir,
2104 .fsync = ocfs2_sync_file,
Mark Fasheh53fc6222007-12-20 16:49:04 -08002105 .release = ocfs2_dir_release,
2106 .open = ocfs2_dir_open,
Andi Kleenc9ec1482008-01-27 03:17:17 +01002107 .unlocked_ioctl = ocfs2_ioctl,
Mark Fasheh586d2322007-03-09 15:56:28 -08002108#ifdef CONFIG_COMPAT
2109 .compat_ioctl = ocfs2_compat_ioctl,
2110#endif
Mark Fasheh53fc6222007-12-20 16:49:04 -08002111 .flock = ocfs2_flock,
Mark Fashehccd979b2005-12-15 14:31:24 -08002112};