blob: 62442e413a001cfbe57607c43e06d408418cf13e [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>
Jan Karaa90714c2008-10-09 19:38:40 +020038#include <linux/quotaops.h>
Mark Fashehccd979b2005-12-15 14:31:24 -080039
40#define MLOG_MASK_PREFIX ML_INODE
41#include <cluster/masklog.h>
42
43#include "ocfs2.h"
44
45#include "alloc.h"
46#include "aops.h"
47#include "dir.h"
48#include "dlmglue.h"
49#include "extent_map.h"
50#include "file.h"
51#include "sysfile.h"
52#include "inode.h"
Herbert Poetzlca4d1472006-07-03 17:27:12 -070053#include "ioctl.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080054#include "journal.h"
Mark Fasheh53fc6222007-12-20 16:49:04 -080055#include "locks.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080056#include "mmap.h"
57#include "suballoc.h"
58#include "super.h"
Tiger Yangcf1d6c72008-08-18 17:11:00 +080059#include "xattr.h"
Tiger Yang23fc2702008-11-14 11:17:18 +080060#include "acl.h"
Jan Karaa90714c2008-10-09 19:38:40 +020061#include "quota.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080062
63#include "buffer_head_io.h"
64
65static int ocfs2_sync_inode(struct inode *inode)
66{
67 filemap_fdatawrite(inode->i_mapping);
68 return sync_mapping_buffers(inode->i_mapping);
69}
70
Mark Fasheh53fc6222007-12-20 16:49:04 -080071static int ocfs2_init_file_private(struct inode *inode, struct file *file)
72{
73 struct ocfs2_file_private *fp;
74
75 fp = kzalloc(sizeof(struct ocfs2_file_private), GFP_KERNEL);
76 if (!fp)
77 return -ENOMEM;
78
79 fp->fp_file = file;
80 mutex_init(&fp->fp_mutex);
81 ocfs2_file_lock_res_init(&fp->fp_flock, fp);
82 file->private_data = fp;
83
84 return 0;
85}
86
87static void ocfs2_free_file_private(struct inode *inode, struct file *file)
88{
89 struct ocfs2_file_private *fp = file->private_data;
90 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
91
92 if (fp) {
93 ocfs2_simple_drop_lockres(osb, &fp->fp_flock);
94 ocfs2_lock_res_free(&fp->fp_flock);
95 kfree(fp);
96 file->private_data = NULL;
97 }
98}
99
Mark Fashehccd979b2005-12-15 14:31:24 -0800100static int ocfs2_file_open(struct inode *inode, struct file *file)
101{
102 int status;
103 int mode = file->f_flags;
104 struct ocfs2_inode_info *oi = OCFS2_I(inode);
105
106 mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file,
Josef Sipekd28c9172006-12-08 02:37:25 -0800107 file->f_path.dentry->d_name.len, file->f_path.dentry->d_name.name);
Mark Fashehccd979b2005-12-15 14:31:24 -0800108
109 spin_lock(&oi->ip_lock);
110
111 /* Check that the inode hasn't been wiped from disk by another
112 * node. If it hasn't then we're safe as long as we hold the
113 * spin lock until our increment of open count. */
114 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
115 spin_unlock(&oi->ip_lock);
116
117 status = -ENOENT;
118 goto leave;
119 }
120
121 if (mode & O_DIRECT)
122 oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT;
123
124 oi->ip_open_count++;
125 spin_unlock(&oi->ip_lock);
Mark Fasheh53fc6222007-12-20 16:49:04 -0800126
127 status = ocfs2_init_file_private(inode, file);
128 if (status) {
129 /*
130 * We want to set open count back if we're failing the
131 * open.
132 */
133 spin_lock(&oi->ip_lock);
134 oi->ip_open_count--;
135 spin_unlock(&oi->ip_lock);
136 }
137
Mark Fashehccd979b2005-12-15 14:31:24 -0800138leave:
139 mlog_exit(status);
140 return status;
141}
142
143static int ocfs2_file_release(struct inode *inode, struct file *file)
144{
145 struct ocfs2_inode_info *oi = OCFS2_I(inode);
146
147 mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file,
Josef Sipekd28c9172006-12-08 02:37:25 -0800148 file->f_path.dentry->d_name.len,
149 file->f_path.dentry->d_name.name);
Mark Fashehccd979b2005-12-15 14:31:24 -0800150
151 spin_lock(&oi->ip_lock);
152 if (!--oi->ip_open_count)
153 oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT;
154 spin_unlock(&oi->ip_lock);
155
Mark Fasheh53fc6222007-12-20 16:49:04 -0800156 ocfs2_free_file_private(inode, file);
157
Mark Fashehccd979b2005-12-15 14:31:24 -0800158 mlog_exit(0);
159
160 return 0;
161}
162
Mark Fasheh53fc6222007-12-20 16:49:04 -0800163static int ocfs2_dir_open(struct inode *inode, struct file *file)
164{
165 return ocfs2_init_file_private(inode, file);
166}
167
168static int ocfs2_dir_release(struct inode *inode, struct file *file)
169{
170 ocfs2_free_file_private(inode, file);
171 return 0;
172}
173
Mark Fashehccd979b2005-12-15 14:31:24 -0800174static int ocfs2_sync_file(struct file *file,
175 struct dentry *dentry,
176 int datasync)
177{
178 int err = 0;
179 journal_t *journal;
180 struct inode *inode = dentry->d_inode;
181 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
182
183 mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", file, dentry, datasync,
184 dentry->d_name.len, dentry->d_name.name);
185
186 err = ocfs2_sync_inode(dentry->d_inode);
187 if (err)
188 goto bail;
189
Hisashi Hifumie04cc152009-06-09 16:47:45 +0900190 if (datasync && !(inode->i_state & I_DIRTY_DATASYNC))
191 goto bail;
192
Mark Fashehccd979b2005-12-15 14:31:24 -0800193 journal = osb->journal->j_journal;
Joel Becker2b4e30f2008-09-03 20:03:41 -0700194 err = jbd2_journal_force_commit(journal);
Mark Fashehccd979b2005-12-15 14:31:24 -0800195
196bail:
197 mlog_exit(err);
198
199 return (err < 0) ? -EIO : 0;
200}
201
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800202int ocfs2_should_update_atime(struct inode *inode,
203 struct vfsmount *vfsmnt)
204{
205 struct timespec now;
206 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
207
208 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
209 return 0;
210
211 if ((inode->i_flags & S_NOATIME) ||
212 ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)))
213 return 0;
214
Mark Fasheh6c2aad02006-12-19 15:25:52 -0800215 /*
216 * We can be called with no vfsmnt structure - NFSD will
217 * sometimes do this.
218 *
219 * Note that our action here is different than touch_atime() -
220 * if we can't tell whether this is a noatime mount, then we
221 * don't know whether to trust the value of s_atime_quantum.
222 */
223 if (vfsmnt == NULL)
224 return 0;
225
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800226 if ((vfsmnt->mnt_flags & MNT_NOATIME) ||
227 ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
228 return 0;
229
Mark Fasheh7e913c52006-12-13 00:34:35 -0800230 if (vfsmnt->mnt_flags & MNT_RELATIME) {
231 if ((timespec_compare(&inode->i_atime, &inode->i_mtime) <= 0) ||
232 (timespec_compare(&inode->i_atime, &inode->i_ctime) <= 0))
233 return 1;
234
235 return 0;
236 }
237
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800238 now = CURRENT_TIME;
239 if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum))
240 return 0;
241 else
242 return 1;
243}
244
245int ocfs2_update_inode_atime(struct inode *inode,
246 struct buffer_head *bh)
247{
248 int ret;
249 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
250 handle_t *handle;
Mark Fashehc11e9fa2007-07-20 11:24:53 -0700251 struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data;
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800252
253 mlog_entry_void();
254
255 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
Jan Karafa38e922008-10-20 19:23:51 +0200256 if (IS_ERR(handle)) {
257 ret = PTR_ERR(handle);
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800258 mlog_errno(ret);
259 goto out;
260 }
261
Joel Becker13723d02008-10-17 19:25:01 -0700262 ret = ocfs2_journal_access_di(handle, inode, bh,
263 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehc11e9fa2007-07-20 11:24:53 -0700264 if (ret) {
265 mlog_errno(ret);
266 goto out_commit;
267 }
268
269 /*
270 * Don't use ocfs2_mark_inode_dirty() here as we don't always
271 * have i_mutex to guard against concurrent changes to other
272 * inode fields.
273 */
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800274 inode->i_atime = CURRENT_TIME;
Mark Fashehc11e9fa2007-07-20 11:24:53 -0700275 di->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
276 di->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
277
278 ret = ocfs2_journal_dirty(handle, bh);
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800279 if (ret < 0)
280 mlog_errno(ret);
281
Mark Fashehc11e9fa2007-07-20 11:24:53 -0700282out_commit:
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800283 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
284out:
285 mlog_exit(ret);
286 return ret;
287}
288
Adrian Bunk6cb129f2007-04-26 00:29:35 -0700289static int ocfs2_set_inode_size(handle_t *handle,
290 struct inode *inode,
291 struct buffer_head *fe_bh,
292 u64 new_i_size)
Mark Fashehccd979b2005-12-15 14:31:24 -0800293{
294 int status;
295
296 mlog_entry_void();
297 i_size_write(inode, new_i_size);
Mark Fasheh8110b072007-03-22 16:53:23 -0700298 inode->i_blocks = ocfs2_inode_sector_count(inode);
Mark Fashehccd979b2005-12-15 14:31:24 -0800299 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
300
301 status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
302 if (status < 0) {
303 mlog_errno(status);
304 goto bail;
305 }
306
307bail:
308 mlog_exit(status);
309 return status;
310}
311
Jan Kara9e33d692008-08-25 19:56:50 +0200312int ocfs2_simple_size_update(struct inode *inode,
313 struct buffer_head *di_bh,
314 u64 new_i_size)
Mark Fashehccd979b2005-12-15 14:31:24 -0800315{
316 int ret;
317 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
Mark Fasheh1fabe142006-10-09 18:11:45 -0700318 handle_t *handle = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800319
Mark Fasheh65eff9c2006-10-09 17:26:22 -0700320 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
Jan Karafa38e922008-10-20 19:23:51 +0200321 if (IS_ERR(handle)) {
322 ret = PTR_ERR(handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800323 mlog_errno(ret);
324 goto out;
325 }
326
327 ret = ocfs2_set_inode_size(handle, inode, di_bh,
328 new_i_size);
329 if (ret < 0)
330 mlog_errno(ret);
331
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700332 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800333out:
334 return ret;
335}
336
337static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
338 struct inode *inode,
339 struct buffer_head *fe_bh,
340 u64 new_i_size)
341{
342 int status;
Mark Fasheh1fabe142006-10-09 18:11:45 -0700343 handle_t *handle;
Mark Fasheh60b11392007-02-16 11:46:50 -0800344 struct ocfs2_dinode *di;
Mark Fasheh35edec12007-07-06 14:41:18 -0700345 u64 cluster_bytes;
Mark Fashehccd979b2005-12-15 14:31:24 -0800346
347 mlog_entry_void();
348
349 /* TODO: This needs to actually orphan the inode in this
350 * transaction. */
351
Mark Fasheh65eff9c2006-10-09 17:26:22 -0700352 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
Mark Fashehccd979b2005-12-15 14:31:24 -0800353 if (IS_ERR(handle)) {
354 status = PTR_ERR(handle);
355 mlog_errno(status);
356 goto out;
357 }
358
Joel Becker13723d02008-10-17 19:25:01 -0700359 status = ocfs2_journal_access_di(handle, inode, fe_bh,
360 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fasheh60b11392007-02-16 11:46:50 -0800361 if (status < 0) {
362 mlog_errno(status);
363 goto out_commit;
364 }
365
366 /*
367 * Do this before setting i_size.
368 */
Mark Fasheh35edec12007-07-06 14:41:18 -0700369 cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size);
370 status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size,
371 cluster_bytes);
Mark Fasheh60b11392007-02-16 11:46:50 -0800372 if (status) {
373 mlog_errno(status);
374 goto out_commit;
375 }
376
377 i_size_write(inode, new_i_size);
Mark Fasheh60b11392007-02-16 11:46:50 -0800378 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
379
380 di = (struct ocfs2_dinode *) fe_bh->b_data;
381 di->i_size = cpu_to_le64(new_i_size);
382 di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
383 di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
384
385 status = ocfs2_journal_dirty(handle, fe_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800386 if (status < 0)
387 mlog_errno(status);
388
Mark Fasheh60b11392007-02-16 11:46:50 -0800389out_commit:
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700390 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800391out:
Mark Fasheh60b11392007-02-16 11:46:50 -0800392
Mark Fashehccd979b2005-12-15 14:31:24 -0800393 mlog_exit(status);
394 return status;
395}
396
397static int ocfs2_truncate_file(struct inode *inode,
398 struct buffer_head *di_bh,
399 u64 new_i_size)
400{
401 int status = 0;
402 struct ocfs2_dinode *fe = NULL;
403 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
404 struct ocfs2_truncate_context *tc = NULL;
405
Mark Fashehb0697052006-03-03 10:24:33 -0800406 mlog_entry("(inode = %llu, new_i_size = %llu\n",
407 (unsigned long long)OCFS2_I(inode)->ip_blkno,
408 (unsigned long long)new_i_size);
Mark Fashehccd979b2005-12-15 14:31:24 -0800409
Joel Beckerb657c952008-11-13 14:49:11 -0800410 /* We trust di_bh because it comes from ocfs2_inode_lock(), which
411 * already validated it */
Mark Fashehccd979b2005-12-15 14:31:24 -0800412 fe = (struct ocfs2_dinode *) di_bh->b_data;
Mark Fashehccd979b2005-12-15 14:31:24 -0800413
414 mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode),
Mark Fashehb0697052006-03-03 10:24:33 -0800415 "Inode %llu, inode i_size = %lld != di "
416 "i_size = %llu, i_flags = 0x%x\n",
417 (unsigned long long)OCFS2_I(inode)->ip_blkno,
Mark Fashehccd979b2005-12-15 14:31:24 -0800418 i_size_read(inode),
Mark Fashehb0697052006-03-03 10:24:33 -0800419 (unsigned long long)le64_to_cpu(fe->i_size),
420 le32_to_cpu(fe->i_flags));
Mark Fashehccd979b2005-12-15 14:31:24 -0800421
422 if (new_i_size > le64_to_cpu(fe->i_size)) {
Mark Fashehb0697052006-03-03 10:24:33 -0800423 mlog(0, "asked to truncate file with size (%llu) to size (%llu)!\n",
424 (unsigned long long)le64_to_cpu(fe->i_size),
425 (unsigned long long)new_i_size);
Mark Fashehccd979b2005-12-15 14:31:24 -0800426 status = -EINVAL;
427 mlog_errno(status);
428 goto bail;
429 }
430
Mark Fashehb0697052006-03-03 10:24:33 -0800431 mlog(0, "inode %llu, i_size = %llu, new_i_size = %llu\n",
432 (unsigned long long)le64_to_cpu(fe->i_blkno),
433 (unsigned long long)le64_to_cpu(fe->i_size),
434 (unsigned long long)new_i_size);
Mark Fashehccd979b2005-12-15 14:31:24 -0800435
436 /* lets handle the simple truncate cases before doing any more
437 * cluster locking. */
438 if (new_i_size == le64_to_cpu(fe->i_size))
439 goto bail;
440
Mark Fasheh2e89b2e2007-05-09 13:40:18 -0700441 down_write(&OCFS2_I(inode)->ip_alloc_sem);
442
Mark Fashehc934a922007-10-18 15:23:46 -0700443 /*
444 * The inode lock forced other nodes to sync and drop their
445 * pages, which (correctly) happens even if we have a truncate
446 * without allocation change - ocfs2 cluster sizes can be much
447 * greater than page size, so we have to truncate them
448 * anyway.
449 */
Mark Fasheh2e89b2e2007-05-09 13:40:18 -0700450 unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1);
451 truncate_inode_pages(inode->i_mapping, new_i_size);
452
Mark Fasheh1afc32b2007-09-07 14:46:51 -0700453 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
454 status = ocfs2_truncate_inline(inode, di_bh, new_i_size,
Mark Fashehb1967d02007-11-20 11:56:39 -0800455 i_size_read(inode), 1);
Mark Fasheh1afc32b2007-09-07 14:46:51 -0700456 if (status)
457 mlog_errno(status);
458
Mark Fashehc934a922007-10-18 15:23:46 -0700459 goto bail_unlock_sem;
Mark Fasheh1afc32b2007-09-07 14:46:51 -0700460 }
461
Mark Fashehccd979b2005-12-15 14:31:24 -0800462 /* alright, we're going to need to do a full blown alloc size
463 * change. Orphan the inode so that recovery can complete the
464 * truncate if necessary. This does the task of marking
465 * i_size. */
466 status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
467 if (status < 0) {
468 mlog_errno(status);
Mark Fashehc934a922007-10-18 15:23:46 -0700469 goto bail_unlock_sem;
Mark Fashehccd979b2005-12-15 14:31:24 -0800470 }
471
472 status = ocfs2_prepare_truncate(osb, inode, di_bh, &tc);
473 if (status < 0) {
474 mlog_errno(status);
Mark Fashehc934a922007-10-18 15:23:46 -0700475 goto bail_unlock_sem;
Mark Fashehccd979b2005-12-15 14:31:24 -0800476 }
477
478 status = ocfs2_commit_truncate(osb, inode, di_bh, tc);
479 if (status < 0) {
480 mlog_errno(status);
Mark Fashehc934a922007-10-18 15:23:46 -0700481 goto bail_unlock_sem;
Mark Fashehccd979b2005-12-15 14:31:24 -0800482 }
483
484 /* TODO: orphan dir cleanup here. */
Mark Fashehc934a922007-10-18 15:23:46 -0700485bail_unlock_sem:
Mark Fasheh2e89b2e2007-05-09 13:40:18 -0700486 up_write(&OCFS2_I(inode)->ip_alloc_sem);
487
Mark Fashehccd979b2005-12-15 14:31:24 -0800488bail:
489
490 mlog_exit(status);
491 return status;
492}
493
494/*
Tao Ma0eb8d472008-08-18 17:38:45 +0800495 * extend file allocation only here.
Mark Fashehccd979b2005-12-15 14:31:24 -0800496 * we'll update all the disk stuff, and oip->alloc_size
497 *
498 * expect stuff to be locked, a transaction started and enough data /
499 * metadata reservations in the contexts.
500 *
501 * Will return -EAGAIN, and a reason if a restart is needed.
502 * If passed in, *reason will always be set, even in error.
503 */
Tao Ma0eb8d472008-08-18 17:38:45 +0800504int ocfs2_add_inode_data(struct ocfs2_super *osb,
505 struct inode *inode,
506 u32 *logical_offset,
507 u32 clusters_to_add,
508 int mark_unwritten,
509 struct buffer_head *fe_bh,
510 handle_t *handle,
511 struct ocfs2_alloc_context *data_ac,
512 struct ocfs2_alloc_context *meta_ac,
513 enum ocfs2_alloc_restarted *reason_ret)
Mark Fashehccd979b2005-12-15 14:31:24 -0800514{
Joel Beckerf99b9b72008-08-20 19:36:33 -0700515 int ret;
516 struct ocfs2_extent_tree et;
Mark Fashehccd979b2005-12-15 14:31:24 -0800517
Joel Becker8d6220d2008-08-22 12:46:09 -0700518 ocfs2_init_dinode_extent_tree(&et, inode, fe_bh);
Joel Beckerf99b9b72008-08-20 19:36:33 -0700519 ret = ocfs2_add_clusters_in_btree(osb, inode, logical_offset,
Tao Ma0eb8d472008-08-18 17:38:45 +0800520 clusters_to_add, mark_unwritten,
Joel Beckerf99b9b72008-08-20 19:36:33 -0700521 &et, handle,
522 data_ac, meta_ac, reason_ret);
Joel Beckerf99b9b72008-08-20 19:36:33 -0700523
524 return ret;
Mark Fashehccd979b2005-12-15 14:31:24 -0800525}
526
Mark Fasheh2ae99a62007-03-09 16:43:28 -0800527static int __ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
528 u32 clusters_to_add, int mark_unwritten)
Mark Fashehccd979b2005-12-15 14:31:24 -0800529{
530 int status = 0;
531 int restart_func = 0;
Mark Fashehabf8b152007-01-17 13:07:24 -0800532 int credits;
Mark Fasheh2ae99a62007-03-09 16:43:28 -0800533 u32 prev_clusters;
Mark Fashehccd979b2005-12-15 14:31:24 -0800534 struct buffer_head *bh = NULL;
535 struct ocfs2_dinode *fe = NULL;
Mark Fasheh1fabe142006-10-09 18:11:45 -0700536 handle_t *handle = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800537 struct ocfs2_alloc_context *data_ac = NULL;
538 struct ocfs2_alloc_context *meta_ac = NULL;
539 enum ocfs2_alloc_restarted why;
540 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
Joel Beckerf99b9b72008-08-20 19:36:33 -0700541 struct ocfs2_extent_tree et;
Jan Karaa90714c2008-10-09 19:38:40 +0200542 int did_quota = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -0800543
544 mlog_entry("(clusters_to_add = %u)\n", clusters_to_add);
545
Mark Fashehdcd05382007-01-16 11:32:23 -0800546 /*
547 * This function only exists for file systems which don't
548 * support holes.
549 */
Mark Fasheh2ae99a62007-03-09 16:43:28 -0800550 BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb));
Mark Fashehdcd05382007-01-16 11:32:23 -0800551
Joel Beckerb657c952008-11-13 14:49:11 -0800552 status = ocfs2_read_inode_block(inode, &bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800553 if (status < 0) {
554 mlog_errno(status);
555 goto leave;
556 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800557 fe = (struct ocfs2_dinode *) bh->b_data;
Mark Fashehccd979b2005-12-15 14:31:24 -0800558
559restart_all:
560 BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
561
Tao Mae7d4cb62008-08-18 17:38:44 +0800562 mlog(0, "extend inode %llu, i_size = %lld, di->i_clusters = %u, "
563 "clusters_to_add = %u\n",
564 (unsigned long long)OCFS2_I(inode)->ip_blkno,
565 (long long)i_size_read(inode), le32_to_cpu(fe->i_clusters),
566 clusters_to_add);
Joel Becker8d6220d2008-08-22 12:46:09 -0700567 ocfs2_init_dinode_extent_tree(&et, inode, bh);
Joel Beckerf99b9b72008-08-20 19:36:33 -0700568 status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
569 &data_ac, &meta_ac);
Mark Fasheh9517bac2007-02-09 20:24:12 -0800570 if (status) {
571 mlog_errno(status);
572 goto leave;
573 }
574
Tao Ma811f9332008-08-18 17:38:43 +0800575 credits = ocfs2_calc_extend_credits(osb->sb, &fe->id2.i_list,
576 clusters_to_add);
Mark Fasheh65eff9c2006-10-09 17:26:22 -0700577 handle = ocfs2_start_trans(osb, credits);
Mark Fashehccd979b2005-12-15 14:31:24 -0800578 if (IS_ERR(handle)) {
579 status = PTR_ERR(handle);
580 handle = NULL;
581 mlog_errno(status);
582 goto leave;
583 }
584
585restarted_transaction:
Jan Karaa90714c2008-10-09 19:38:40 +0200586 if (vfs_dq_alloc_space_nodirty(inode, ocfs2_clusters_to_bytes(osb->sb,
587 clusters_to_add))) {
588 status = -EDQUOT;
589 goto leave;
590 }
591 did_quota = 1;
592
Mark Fashehccd979b2005-12-15 14:31:24 -0800593 /* reserve a write to the file entry early on - that we if we
594 * run out of credits in the allocation path, we can still
595 * update i_size. */
Joel Becker13723d02008-10-17 19:25:01 -0700596 status = ocfs2_journal_access_di(handle, inode, bh,
597 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehccd979b2005-12-15 14:31:24 -0800598 if (status < 0) {
599 mlog_errno(status);
600 goto leave;
601 }
602
603 prev_clusters = OCFS2_I(inode)->ip_clusters;
604
Tao Ma0eb8d472008-08-18 17:38:45 +0800605 status = ocfs2_add_inode_data(osb,
606 inode,
607 &logical_start,
608 clusters_to_add,
609 mark_unwritten,
610 bh,
611 handle,
612 data_ac,
613 meta_ac,
614 &why);
Mark Fashehccd979b2005-12-15 14:31:24 -0800615 if ((status < 0) && (status != -EAGAIN)) {
616 if (status != -ENOSPC)
617 mlog_errno(status);
618 goto leave;
619 }
620
621 status = ocfs2_journal_dirty(handle, bh);
622 if (status < 0) {
623 mlog_errno(status);
624 goto leave;
625 }
626
627 spin_lock(&OCFS2_I(inode)->ip_lock);
628 clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
629 spin_unlock(&OCFS2_I(inode)->ip_lock);
Jan Karaa90714c2008-10-09 19:38:40 +0200630 /* Release unused quota reservation */
631 vfs_dq_free_space(inode,
632 ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
633 did_quota = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -0800634
635 if (why != RESTART_NONE && clusters_to_add) {
636 if (why == RESTART_META) {
637 mlog(0, "restarting function.\n");
638 restart_func = 1;
639 } else {
640 BUG_ON(why != RESTART_TRANS);
641
642 mlog(0, "restarting transaction.\n");
643 /* TODO: This can be more intelligent. */
644 credits = ocfs2_calc_extend_credits(osb->sb,
Tao Ma811f9332008-08-18 17:38:43 +0800645 &fe->id2.i_list,
Mark Fashehccd979b2005-12-15 14:31:24 -0800646 clusters_to_add);
Mark Fasheh1fabe142006-10-09 18:11:45 -0700647 status = ocfs2_extend_trans(handle, credits);
Mark Fashehccd979b2005-12-15 14:31:24 -0800648 if (status < 0) {
649 /* handle still has to be committed at
650 * this point. */
651 status = -ENOMEM;
652 mlog_errno(status);
653 goto leave;
654 }
655 goto restarted_transaction;
656 }
657 }
658
Mark Fashehb0697052006-03-03 10:24:33 -0800659 mlog(0, "fe: i_clusters = %u, i_size=%llu\n",
Mark Fasheh1ca1a112007-04-27 16:01:25 -0700660 le32_to_cpu(fe->i_clusters),
661 (unsigned long long)le64_to_cpu(fe->i_size));
Mark Fashehccd979b2005-12-15 14:31:24 -0800662 mlog(0, "inode: ip_clusters=%u, i_size=%lld\n",
Jan Kara634bf742007-12-19 15:25:42 +0100663 OCFS2_I(inode)->ip_clusters, (long long)i_size_read(inode));
Mark Fashehccd979b2005-12-15 14:31:24 -0800664
665leave:
Jan Karaa90714c2008-10-09 19:38:40 +0200666 if (status < 0 && did_quota)
667 vfs_dq_free_space(inode,
668 ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
Mark Fashehccd979b2005-12-15 14:31:24 -0800669 if (handle) {
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700670 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800671 handle = NULL;
672 }
673 if (data_ac) {
674 ocfs2_free_alloc_context(data_ac);
675 data_ac = NULL;
676 }
677 if (meta_ac) {
678 ocfs2_free_alloc_context(meta_ac);
679 meta_ac = NULL;
680 }
681 if ((!status) && restart_func) {
682 restart_func = 0;
683 goto restart_all;
684 }
Mark Fasheha81cb882008-10-07 14:25:16 -0700685 brelse(bh);
686 bh = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800687
688 mlog_exit(status);
689 return status;
690}
691
692/* Some parts of this taken from generic_cont_expand, which turned out
693 * to be too fragile to do exactly what we need without us having to
Nick Piggin4e02ed42008-10-29 14:00:55 -0700694 * worry about recursive locking in ->write_begin() and ->write_end(). */
Mark Fashehccd979b2005-12-15 14:31:24 -0800695static int ocfs2_write_zero_page(struct inode *inode,
696 u64 size)
697{
698 struct address_space *mapping = inode->i_mapping;
699 struct page *page;
700 unsigned long index;
701 unsigned int offset;
Mark Fasheh1fabe142006-10-09 18:11:45 -0700702 handle_t *handle = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800703 int ret;
704
705 offset = (size & (PAGE_CACHE_SIZE-1)); /* Within page */
706 /* ugh. in prepare/commit_write, if from==to==start of block, we
707 ** skip the prepare. make sure we never send an offset for the start
708 ** of a block
709 */
710 if ((offset & (inode->i_sb->s_blocksize - 1)) == 0) {
711 offset++;
712 }
713 index = size >> PAGE_CACHE_SHIFT;
714
715 page = grab_cache_page(mapping, index);
716 if (!page) {
717 ret = -ENOMEM;
718 mlog_errno(ret);
719 goto out;
720 }
721
Mark Fasheh53013cb2006-05-05 19:04:03 -0700722 ret = ocfs2_prepare_write_nolock(inode, page, offset, offset);
Mark Fashehccd979b2005-12-15 14:31:24 -0800723 if (ret < 0) {
724 mlog_errno(ret);
725 goto out_unlock;
726 }
727
728 if (ocfs2_should_order_data(inode)) {
729 handle = ocfs2_start_walk_page_trans(inode, page, offset,
730 offset);
731 if (IS_ERR(handle)) {
732 ret = PTR_ERR(handle);
733 handle = NULL;
734 goto out_unlock;
735 }
736 }
737
738 /* must not update i_size! */
739 ret = block_commit_write(page, offset, offset);
740 if (ret < 0)
741 mlog_errno(ret);
742 else
743 ret = 0;
744
745 if (handle)
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700746 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800747out_unlock:
748 unlock_page(page);
749 page_cache_release(page);
750out:
751 return ret;
752}
753
754static int ocfs2_zero_extend(struct inode *inode,
755 u64 zero_to_size)
756{
757 int ret = 0;
758 u64 start_off;
759 struct super_block *sb = inode->i_sb;
760
761 start_off = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
762 while (start_off < zero_to_size) {
763 ret = ocfs2_write_zero_page(inode, start_off);
764 if (ret < 0) {
765 mlog_errno(ret);
766 goto out;
767 }
768
769 start_off += sb->s_blocksize;
Mark Fashehe2057c52006-10-03 17:53:05 -0700770
771 /*
772 * Very large extends have the potential to lock up
773 * the cpu for extended periods of time.
774 */
775 cond_resched();
Mark Fashehccd979b2005-12-15 14:31:24 -0800776 }
777
778out:
779 return ret;
780}
781
Mark Fasheh65ed39d2007-08-28 17:13:23 -0700782int ocfs2_extend_no_holes(struct inode *inode, u64 new_i_size, u64 zero_to)
783{
784 int ret;
785 u32 clusters_to_add;
786 struct ocfs2_inode_info *oi = OCFS2_I(inode);
787
788 clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size);
789 if (clusters_to_add < oi->ip_clusters)
790 clusters_to_add = 0;
791 else
792 clusters_to_add -= oi->ip_clusters;
793
794 if (clusters_to_add) {
795 ret = __ocfs2_extend_allocation(inode, oi->ip_clusters,
796 clusters_to_add, 0);
797 if (ret) {
798 mlog_errno(ret);
799 goto out;
800 }
801 }
802
803 /*
804 * Call this even if we don't add any clusters to the tree. We
805 * still need to zero the area between the old i_size and the
806 * new i_size.
807 */
808 ret = ocfs2_zero_extend(inode, zero_to);
809 if (ret < 0)
810 mlog_errno(ret);
811
812out:
813 return ret;
814}
815
Mark Fashehccd979b2005-12-15 14:31:24 -0800816static int ocfs2_extend_file(struct inode *inode,
817 struct buffer_head *di_bh,
Mark Fasheh65ed39d2007-08-28 17:13:23 -0700818 u64 new_i_size)
Mark Fashehccd979b2005-12-15 14:31:24 -0800819{
Mark Fashehc934a922007-10-18 15:23:46 -0700820 int ret = 0;
Mark Fasheh1afc32b2007-09-07 14:46:51 -0700821 struct ocfs2_inode_info *oi = OCFS2_I(inode);
Mark Fashehccd979b2005-12-15 14:31:24 -0800822
Mark Fasheh65ed39d2007-08-28 17:13:23 -0700823 BUG_ON(!di_bh);
Mark Fasheh53013cb2006-05-05 19:04:03 -0700824
Mark Fashehccd979b2005-12-15 14:31:24 -0800825 /* setattr sometimes calls us like this. */
826 if (new_i_size == 0)
827 goto out;
828
829 if (i_size_read(inode) == new_i_size)
830 goto out;
831 BUG_ON(new_i_size < i_size_read(inode));
832
Mark Fasheh1afc32b2007-09-07 14:46:51 -0700833 /*
834 * Fall through for converting inline data, even if the fs
835 * supports sparse files.
836 *
837 * The check for inline data here is legal - nobody can add
838 * the feature since we have i_mutex. We must check it again
839 * after acquiring ip_alloc_sem though, as paths like mmap
840 * might have raced us to converting the inode to extents.
841 */
842 if (!(oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
843 && ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
Mark Fasheh3a0782d2007-01-17 12:53:31 -0800844 goto out_update_size;
Mark Fashehccd979b2005-12-15 14:31:24 -0800845
Mark Fasheh0effef72006-10-03 17:44:42 -0700846 /*
Mark Fasheh65ed39d2007-08-28 17:13:23 -0700847 * The alloc sem blocks people in read/write from reading our
848 * allocation until we're done changing it. We depend on
849 * i_mutex to block other extend/truncate calls while we're
850 * here.
Mark Fasheh0effef72006-10-03 17:44:42 -0700851 */
Mark Fasheh1afc32b2007-09-07 14:46:51 -0700852 down_write(&oi->ip_alloc_sem);
853
854 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
855 /*
856 * We can optimize small extends by keeping the inodes
857 * inline data.
858 */
859 if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) {
860 up_write(&oi->ip_alloc_sem);
861 goto out_update_size;
862 }
863
864 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
865 if (ret) {
866 up_write(&oi->ip_alloc_sem);
867
868 mlog_errno(ret);
Mark Fashehc934a922007-10-18 15:23:46 -0700869 goto out;
Mark Fasheh1afc32b2007-09-07 14:46:51 -0700870 }
871 }
872
873 if (!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
874 ret = ocfs2_extend_no_holes(inode, new_i_size, new_i_size);
875
876 up_write(&oi->ip_alloc_sem);
Mark Fasheh65ed39d2007-08-28 17:13:23 -0700877
Mark Fasheh0effef72006-10-03 17:44:42 -0700878 if (ret < 0) {
879 mlog_errno(ret);
Mark Fashehc934a922007-10-18 15:23:46 -0700880 goto out;
Mark Fashehccd979b2005-12-15 14:31:24 -0800881 }
882
Mark Fasheh3a0782d2007-01-17 12:53:31 -0800883out_update_size:
Mark Fasheh65ed39d2007-08-28 17:13:23 -0700884 ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
885 if (ret < 0)
886 mlog_errno(ret);
Mark Fasheh53013cb2006-05-05 19:04:03 -0700887
Mark Fashehccd979b2005-12-15 14:31:24 -0800888out:
889 return ret;
890}
891
892int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
893{
894 int status = 0, size_change;
895 struct inode *inode = dentry->d_inode;
896 struct super_block *sb = inode->i_sb;
897 struct ocfs2_super *osb = OCFS2_SB(sb);
898 struct buffer_head *bh = NULL;
Mark Fasheh1fabe142006-10-09 18:11:45 -0700899 handle_t *handle = NULL;
Jan Kara65bac572009-06-02 14:24:01 +0200900 int qtype;
901 struct dquot *transfer_from[MAXQUOTAS] = { };
902 struct dquot *transfer_to[MAXQUOTAS] = { };
Mark Fashehccd979b2005-12-15 14:31:24 -0800903
904 mlog_entry("(0x%p, '%.*s')\n", dentry,
905 dentry->d_name.len, dentry->d_name.name);
906
Sunil Mushranbc535802008-04-18 10:23:53 -0700907 /* ensuring we don't even attempt to truncate a symlink */
908 if (S_ISLNK(inode->i_mode))
909 attr->ia_valid &= ~ATTR_SIZE;
910
Mark Fashehccd979b2005-12-15 14:31:24 -0800911 if (attr->ia_valid & ATTR_MODE)
912 mlog(0, "mode change: %d\n", attr->ia_mode);
913 if (attr->ia_valid & ATTR_UID)
914 mlog(0, "uid change: %d\n", attr->ia_uid);
915 if (attr->ia_valid & ATTR_GID)
916 mlog(0, "gid change: %d\n", attr->ia_gid);
917 if (attr->ia_valid & ATTR_SIZE)
918 mlog(0, "size change...\n");
919 if (attr->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME))
920 mlog(0, "time change...\n");
921
922#define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
923 | ATTR_GID | ATTR_UID | ATTR_MODE)
924 if (!(attr->ia_valid & OCFS2_VALID_ATTRS)) {
925 mlog(0, "can't handle attrs: 0x%x\n", attr->ia_valid);
926 return 0;
927 }
928
929 status = inode_change_ok(inode, attr);
930 if (status)
931 return status;
932
933 size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
934 if (size_change) {
935 status = ocfs2_rw_lock(inode, 1);
936 if (status < 0) {
937 mlog_errno(status);
938 goto bail;
939 }
940 }
941
Mark Fashehe63aecb62007-10-18 15:30:42 -0700942 status = ocfs2_inode_lock(inode, &bh, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -0800943 if (status < 0) {
944 if (status != -ENOENT)
945 mlog_errno(status);
946 goto bail_unlock_rw;
947 }
948
949 if (size_change && attr->ia_size != i_size_read(inode)) {
Mark Fashehce76fd32007-07-20 12:02:14 -0700950 if (attr->ia_size > sb->s_maxbytes) {
951 status = -EFBIG;
952 goto bail_unlock;
953 }
954
Joel Becker2b4e30f2008-09-03 20:03:41 -0700955 if (i_size_read(inode) > attr->ia_size) {
956 if (ocfs2_should_order_data(inode)) {
957 status = ocfs2_begin_ordered_truncate(inode,
958 attr->ia_size);
959 if (status)
960 goto bail_unlock;
961 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800962 status = ocfs2_truncate_file(inode, bh, attr->ia_size);
Joel Becker2b4e30f2008-09-03 20:03:41 -0700963 } else
Mark Fasheh65ed39d2007-08-28 17:13:23 -0700964 status = ocfs2_extend_file(inode, bh, attr->ia_size);
Mark Fashehccd979b2005-12-15 14:31:24 -0800965 if (status < 0) {
966 if (status != -ENOSPC)
967 mlog_errno(status);
968 status = -ENOSPC;
969 goto bail_unlock;
970 }
971 }
972
Jan Karaa90714c2008-10-09 19:38:40 +0200973 if ((attr->ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
974 (attr->ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid)) {
Jan Kara65bac572009-06-02 14:24:01 +0200975 /*
976 * Gather pointers to quota structures so that allocation /
977 * freeing of quota structures happens here and not inside
978 * vfs_dq_transfer() where we have problems with lock ordering
979 */
Jan Karaa90714c2008-10-09 19:38:40 +0200980 if (attr->ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid
981 && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
982 OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
Jan Kara65bac572009-06-02 14:24:01 +0200983 transfer_to[USRQUOTA] = dqget(sb, attr->ia_uid,
984 USRQUOTA);
985 transfer_from[USRQUOTA] = dqget(sb, inode->i_uid,
986 USRQUOTA);
987 if (!transfer_to[USRQUOTA] || !transfer_from[USRQUOTA]) {
988 status = -ESRCH;
Jan Karaa90714c2008-10-09 19:38:40 +0200989 goto bail_unlock;
Jan Kara65bac572009-06-02 14:24:01 +0200990 }
Jan Karaa90714c2008-10-09 19:38:40 +0200991 }
992 if (attr->ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid
993 && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
994 OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
Jan Kara65bac572009-06-02 14:24:01 +0200995 transfer_to[GRPQUOTA] = dqget(sb, attr->ia_gid,
996 GRPQUOTA);
997 transfer_from[GRPQUOTA] = dqget(sb, inode->i_gid,
998 GRPQUOTA);
999 if (!transfer_to[GRPQUOTA] || !transfer_from[GRPQUOTA]) {
1000 status = -ESRCH;
Jan Karaa90714c2008-10-09 19:38:40 +02001001 goto bail_unlock;
Jan Kara65bac572009-06-02 14:24:01 +02001002 }
Jan Karaa90714c2008-10-09 19:38:40 +02001003 }
Jan Kara65bac572009-06-02 14:24:01 +02001004 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS +
1005 2 * ocfs2_quota_trans_credits(sb));
Jan Karaa90714c2008-10-09 19:38:40 +02001006 if (IS_ERR(handle)) {
1007 status = PTR_ERR(handle);
1008 mlog_errno(status);
1009 goto bail_unlock;
1010 }
1011 status = vfs_dq_transfer(inode, attr) ? -EDQUOT : 0;
1012 if (status < 0)
1013 goto bail_commit;
1014 } else {
1015 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1016 if (IS_ERR(handle)) {
1017 status = PTR_ERR(handle);
1018 mlog_errno(status);
1019 goto bail_unlock;
1020 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001021 }
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:
Mark Fashehe63aecb62007-10-18 15:30:42 -07001043 ocfs2_inode_unlock(inode, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -08001044bail_unlock_rw:
1045 if (size_change)
1046 ocfs2_rw_unlock(inode, 1);
1047bail:
Mark Fasheha81cb882008-10-07 14:25:16 -07001048 brelse(bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08001049
Jan Kara65bac572009-06-02 14:24:01 +02001050 /* Release quota pointers in case we acquired them */
1051 for (qtype = 0; qtype < MAXQUOTAS; qtype++) {
1052 dqput(transfer_to[qtype]);
1053 dqput(transfer_from[qtype]);
1054 }
1055
Tiger Yang060bc662008-11-14 11:17:29 +08001056 if (!status && attr->ia_valid & ATTR_MODE) {
1057 status = ocfs2_acl_chmod(inode);
1058 if (status < 0)
1059 mlog_errno(status);
1060 }
1061
Mark Fashehccd979b2005-12-15 14:31:24 -08001062 mlog_exit(status);
1063 return status;
1064}
1065
1066int ocfs2_getattr(struct vfsmount *mnt,
1067 struct dentry *dentry,
1068 struct kstat *stat)
1069{
1070 struct inode *inode = dentry->d_inode;
1071 struct super_block *sb = dentry->d_inode->i_sb;
1072 struct ocfs2_super *osb = sb->s_fs_info;
1073 int err;
1074
1075 mlog_entry_void();
1076
1077 err = ocfs2_inode_revalidate(dentry);
1078 if (err) {
1079 if (err != -ENOENT)
1080 mlog_errno(err);
1081 goto bail;
1082 }
1083
1084 generic_fillattr(inode, stat);
1085
1086 /* We set the blksize from the cluster size for performance */
1087 stat->blksize = osb->s_clustersize;
1088
1089bail:
1090 mlog_exit(err);
1091
1092 return err;
1093}
1094
Al Viroe6305c42008-07-15 21:03:57 -04001095int ocfs2_permission(struct inode *inode, int mask)
Tiger Yangd38eb8d2006-11-27 09:59:21 +08001096{
1097 int ret;
1098
1099 mlog_entry_void();
1100
Mark Fashehe63aecb62007-10-18 15:30:42 -07001101 ret = ocfs2_inode_lock(inode, NULL, 0);
Tiger Yangd38eb8d2006-11-27 09:59:21 +08001102 if (ret) {
Mark Fasheha9f5f702007-04-26 11:43:43 -07001103 if (ret != -ENOENT)
1104 mlog_errno(ret);
Tiger Yangd38eb8d2006-11-27 09:59:21 +08001105 goto out;
1106 }
1107
Tiger Yang23fc2702008-11-14 11:17:18 +08001108 ret = generic_permission(inode, mask, ocfs2_check_acl);
Tiger Yangd38eb8d2006-11-27 09:59:21 +08001109
Mark Fashehe63aecb62007-10-18 15:30:42 -07001110 ocfs2_inode_unlock(inode, 0);
Tiger Yangd38eb8d2006-11-27 09:59:21 +08001111out:
1112 mlog_exit(ret);
1113 return ret;
1114}
1115
Mark Fashehb2580102007-03-09 16:53:21 -08001116static int __ocfs2_write_remove_suid(struct inode *inode,
1117 struct buffer_head *bh)
Mark Fashehccd979b2005-12-15 14:31:24 -08001118{
1119 int ret;
Mark Fasheh1fabe142006-10-09 18:11:45 -07001120 handle_t *handle;
Mark Fashehccd979b2005-12-15 14:31:24 -08001121 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1122 struct ocfs2_dinode *di;
1123
Mark Fashehb0697052006-03-03 10:24:33 -08001124 mlog_entry("(Inode %llu, mode 0%o)\n",
Mark Fashehb2580102007-03-09 16:53:21 -08001125 (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_mode);
Mark Fashehccd979b2005-12-15 14:31:24 -08001126
Mark Fasheh65eff9c2006-10-09 17:26:22 -07001127 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
Jan Karafa38e922008-10-20 19:23:51 +02001128 if (IS_ERR(handle)) {
1129 ret = PTR_ERR(handle);
Mark Fashehccd979b2005-12-15 14:31:24 -08001130 mlog_errno(ret);
1131 goto out;
1132 }
1133
Joel Becker13723d02008-10-17 19:25:01 -07001134 ret = ocfs2_journal_access_di(handle, inode, bh,
1135 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehccd979b2005-12-15 14:31:24 -08001136 if (ret < 0) {
1137 mlog_errno(ret);
Mark Fashehb2580102007-03-09 16:53:21 -08001138 goto out_trans;
Mark Fashehccd979b2005-12-15 14:31:24 -08001139 }
1140
1141 inode->i_mode &= ~S_ISUID;
1142 if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
1143 inode->i_mode &= ~S_ISGID;
1144
1145 di = (struct ocfs2_dinode *) bh->b_data;
1146 di->i_mode = cpu_to_le16(inode->i_mode);
1147
1148 ret = ocfs2_journal_dirty(handle, bh);
1149 if (ret < 0)
1150 mlog_errno(ret);
Mark Fashehb2580102007-03-09 16:53:21 -08001151
Mark Fashehccd979b2005-12-15 14:31:24 -08001152out_trans:
Mark Fasheh02dc1af2006-10-09 16:48:10 -07001153 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -08001154out:
1155 mlog_exit(ret);
1156 return ret;
1157}
1158
Mark Fasheh9517bac2007-02-09 20:24:12 -08001159/*
1160 * Will look for holes and unwritten extents in the range starting at
1161 * pos for count bytes (inclusive).
1162 */
1163static int ocfs2_check_range_for_holes(struct inode *inode, loff_t pos,
1164 size_t count)
1165{
1166 int ret = 0;
Mark Fasheh49cb8d22007-03-09 16:21:46 -08001167 unsigned int extent_flags;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001168 u32 cpos, clusters, extent_len, phys_cpos;
1169 struct super_block *sb = inode->i_sb;
1170
1171 cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
1172 clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
1173
1174 while (clusters) {
Mark Fasheh49cb8d22007-03-09 16:21:46 -08001175 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
1176 &extent_flags);
Mark Fasheh9517bac2007-02-09 20:24:12 -08001177 if (ret < 0) {
1178 mlog_errno(ret);
1179 goto out;
1180 }
1181
Mark Fasheh49cb8d22007-03-09 16:21:46 -08001182 if (phys_cpos == 0 || (extent_flags & OCFS2_EXT_UNWRITTEN)) {
Mark Fasheh9517bac2007-02-09 20:24:12 -08001183 ret = 1;
1184 break;
1185 }
1186
1187 if (extent_len > clusters)
1188 extent_len = clusters;
1189
1190 clusters -= extent_len;
1191 cpos += extent_len;
1192 }
1193out:
1194 return ret;
1195}
1196
Mark Fashehb2580102007-03-09 16:53:21 -08001197static int ocfs2_write_remove_suid(struct inode *inode)
1198{
1199 int ret;
1200 struct buffer_head *bh = NULL;
Mark Fashehb2580102007-03-09 16:53:21 -08001201
Joel Beckerb657c952008-11-13 14:49:11 -08001202 ret = ocfs2_read_inode_block(inode, &bh);
Mark Fashehb2580102007-03-09 16:53:21 -08001203 if (ret < 0) {
1204 mlog_errno(ret);
1205 goto out;
1206 }
1207
1208 ret = __ocfs2_write_remove_suid(inode, bh);
1209out:
1210 brelse(bh);
1211 return ret;
1212}
1213
Mark Fasheh2ae99a62007-03-09 16:43:28 -08001214/*
1215 * Allocate enough extents to cover the region starting at byte offset
1216 * start for len bytes. Existing extents are skipped, any extents
1217 * added are marked as "unwritten".
1218 */
1219static int ocfs2_allocate_unwritten_extents(struct inode *inode,
1220 u64 start, u64 len)
1221{
1222 int ret;
1223 u32 cpos, phys_cpos, clusters, alloc_size;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001224 u64 end = start + len;
1225 struct buffer_head *di_bh = NULL;
1226
1227 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
Joel Beckerb657c952008-11-13 14:49:11 -08001228 ret = ocfs2_read_inode_block(inode, &di_bh);
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001229 if (ret) {
1230 mlog_errno(ret);
1231 goto out;
1232 }
1233
1234 /*
1235 * Nothing to do if the requested reservation range
1236 * fits within the inode.
1237 */
1238 if (ocfs2_size_fits_inline_data(di_bh, end))
1239 goto out;
1240
1241 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1242 if (ret) {
1243 mlog_errno(ret);
1244 goto out;
1245 }
1246 }
Mark Fasheh2ae99a62007-03-09 16:43:28 -08001247
1248 /*
1249 * We consider both start and len to be inclusive.
1250 */
1251 cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
1252 clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len);
1253 clusters -= cpos;
1254
1255 while (clusters) {
1256 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1257 &alloc_size, NULL);
1258 if (ret) {
1259 mlog_errno(ret);
1260 goto out;
1261 }
1262
1263 /*
1264 * Hole or existing extent len can be arbitrary, so
1265 * cap it to our own allocation request.
1266 */
1267 if (alloc_size > clusters)
1268 alloc_size = clusters;
1269
1270 if (phys_cpos) {
1271 /*
1272 * We already have an allocation at this
1273 * region so we can safely skip it.
1274 */
1275 goto next;
1276 }
1277
1278 ret = __ocfs2_extend_allocation(inode, cpos, alloc_size, 1);
1279 if (ret) {
1280 if (ret != -ENOSPC)
1281 mlog_errno(ret);
1282 goto out;
1283 }
1284
1285next:
1286 cpos += alloc_size;
1287 clusters -= alloc_size;
1288 }
1289
1290 ret = 0;
1291out:
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001292
1293 brelse(di_bh);
Mark Fasheh2ae99a62007-03-09 16:43:28 -08001294 return ret;
1295}
1296
Mark Fasheh063c4562007-07-03 13:34:11 -07001297/*
1298 * Truncate a byte range, avoiding pages within partial clusters. This
1299 * preserves those pages for the zeroing code to write to.
1300 */
1301static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start,
1302 u64 byte_len)
1303{
1304 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1305 loff_t start, end;
1306 struct address_space *mapping = inode->i_mapping;
1307
1308 start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start);
1309 end = byte_start + byte_len;
1310 end = end & ~(osb->s_clustersize - 1);
1311
1312 if (start < end) {
1313 unmap_mapping_range(mapping, start, end - start, 0);
1314 truncate_inode_pages_range(mapping, start, end - 1);
1315 }
1316}
1317
1318static int ocfs2_zero_partial_clusters(struct inode *inode,
1319 u64 start, u64 len)
1320{
1321 int ret = 0;
1322 u64 tmpend, end = start + len;
1323 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1324 unsigned int csize = osb->s_clustersize;
1325 handle_t *handle;
1326
1327 /*
1328 * The "start" and "end" values are NOT necessarily part of
1329 * the range whose allocation is being deleted. Rather, this
1330 * is what the user passed in with the request. We must zero
1331 * partial clusters here. There's no need to worry about
1332 * physical allocation - the zeroing code knows to skip holes.
1333 */
1334 mlog(0, "byte start: %llu, end: %llu\n",
1335 (unsigned long long)start, (unsigned long long)end);
1336
1337 /*
1338 * If both edges are on a cluster boundary then there's no
1339 * zeroing required as the region is part of the allocation to
1340 * be truncated.
1341 */
1342 if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0)
1343 goto out;
1344
1345 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
Jan Karafa38e922008-10-20 19:23:51 +02001346 if (IS_ERR(handle)) {
1347 ret = PTR_ERR(handle);
Mark Fasheh063c4562007-07-03 13:34:11 -07001348 mlog_errno(ret);
1349 goto out;
1350 }
1351
1352 /*
1353 * We want to get the byte offset of the end of the 1st cluster.
1354 */
1355 tmpend = (u64)osb->s_clustersize + (start & ~(osb->s_clustersize - 1));
1356 if (tmpend > end)
1357 tmpend = end;
1358
1359 mlog(0, "1st range: start: %llu, tmpend: %llu\n",
1360 (unsigned long long)start, (unsigned long long)tmpend);
1361
1362 ret = ocfs2_zero_range_for_truncate(inode, handle, start, tmpend);
1363 if (ret)
1364 mlog_errno(ret);
1365
1366 if (tmpend < end) {
1367 /*
1368 * This may make start and end equal, but the zeroing
1369 * code will skip any work in that case so there's no
1370 * need to catch it up here.
1371 */
1372 start = end & ~(osb->s_clustersize - 1);
1373
1374 mlog(0, "2nd range: start: %llu, end: %llu\n",
1375 (unsigned long long)start, (unsigned long long)end);
1376
1377 ret = ocfs2_zero_range_for_truncate(inode, handle, start, end);
1378 if (ret)
1379 mlog_errno(ret);
1380 }
1381
1382 ocfs2_commit_trans(osb, handle);
1383out:
1384 return ret;
1385}
1386
1387static int ocfs2_remove_inode_range(struct inode *inode,
1388 struct buffer_head *di_bh, u64 byte_start,
1389 u64 byte_len)
1390{
1391 int ret = 0;
1392 u32 trunc_start, trunc_len, cpos, phys_cpos, alloc_size;
1393 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1394 struct ocfs2_cached_dealloc_ctxt dealloc;
Mark Fashehb1967d02007-11-20 11:56:39 -08001395 struct address_space *mapping = inode->i_mapping;
Mark Fashehfecc0112008-11-12 15:16:38 -08001396 struct ocfs2_extent_tree et;
Mark Fasheh063c4562007-07-03 13:34:11 -07001397
Mark Fashehfecc0112008-11-12 15:16:38 -08001398 ocfs2_init_dinode_extent_tree(&et, inode, di_bh);
Mark Fasheh063c4562007-07-03 13:34:11 -07001399 ocfs2_init_dealloc_ctxt(&dealloc);
1400
1401 if (byte_len == 0)
1402 return 0;
1403
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001404 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1405 ret = ocfs2_truncate_inline(inode, di_bh, byte_start,
Mark Fashehb1967d02007-11-20 11:56:39 -08001406 byte_start + byte_len, 0);
1407 if (ret) {
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001408 mlog_errno(ret);
Mark Fashehb1967d02007-11-20 11:56:39 -08001409 goto out;
1410 }
1411 /*
1412 * There's no need to get fancy with the page cache
1413 * truncate of an inline-data inode. We're talking
1414 * about less than a page here, which will be cached
1415 * in the dinode buffer anyway.
1416 */
1417 unmap_mapping_range(mapping, 0, 0, 0);
1418 truncate_inode_pages(mapping, 0);
1419 goto out;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001420 }
1421
Mark Fasheh063c4562007-07-03 13:34:11 -07001422 trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start);
1423 trunc_len = (byte_start + byte_len) >> osb->s_clustersize_bits;
1424 if (trunc_len >= trunc_start)
1425 trunc_len -= trunc_start;
1426 else
1427 trunc_len = 0;
1428
1429 mlog(0, "Inode: %llu, start: %llu, len: %llu, cstart: %u, clen: %u\n",
1430 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1431 (unsigned long long)byte_start,
1432 (unsigned long long)byte_len, trunc_start, trunc_len);
1433
1434 ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len);
1435 if (ret) {
1436 mlog_errno(ret);
1437 goto out;
1438 }
1439
1440 cpos = trunc_start;
1441 while (trunc_len) {
1442 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1443 &alloc_size, NULL);
1444 if (ret) {
1445 mlog_errno(ret);
1446 goto out;
1447 }
1448
1449 if (alloc_size > trunc_len)
1450 alloc_size = trunc_len;
1451
1452 /* Only do work for non-holes */
1453 if (phys_cpos != 0) {
Mark Fashehfecc0112008-11-12 15:16:38 -08001454 ret = ocfs2_remove_btree_range(inode, &et, cpos,
1455 phys_cpos, alloc_size,
1456 &dealloc);
Mark Fasheh063c4562007-07-03 13:34:11 -07001457 if (ret) {
1458 mlog_errno(ret);
1459 goto out;
1460 }
1461 }
1462
1463 cpos += alloc_size;
1464 trunc_len -= alloc_size;
1465 }
1466
1467 ocfs2_truncate_cluster_pages(inode, byte_start, byte_len);
1468
1469out:
1470 ocfs2_schedule_truncate_log_flush(osb, 1);
1471 ocfs2_run_deallocs(osb, &dealloc);
1472
1473 return ret;
1474}
1475
Mark Fashehb2580102007-03-09 16:53:21 -08001476/*
1477 * Parts of this function taken from xfs_change_file_space()
1478 */
Mark Fasheh385820a2007-07-19 00:14:38 -07001479static int __ocfs2_change_file_space(struct file *file, struct inode *inode,
1480 loff_t f_pos, unsigned int cmd,
1481 struct ocfs2_space_resv *sr,
1482 int change_size)
Mark Fashehb2580102007-03-09 16:53:21 -08001483{
1484 int ret;
1485 s64 llen;
Mark Fasheh385820a2007-07-19 00:14:38 -07001486 loff_t size;
Mark Fashehb2580102007-03-09 16:53:21 -08001487 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1488 struct buffer_head *di_bh = NULL;
1489 handle_t *handle;
Mark Fasheha00cce32007-07-20 11:28:30 -07001490 unsigned long long max_off = inode->i_sb->s_maxbytes;
Mark Fashehb2580102007-03-09 16:53:21 -08001491
Mark Fashehb2580102007-03-09 16:53:21 -08001492 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
1493 return -EROFS;
1494
1495 mutex_lock(&inode->i_mutex);
1496
1497 /*
1498 * This prevents concurrent writes on other nodes
1499 */
1500 ret = ocfs2_rw_lock(inode, 1);
1501 if (ret) {
1502 mlog_errno(ret);
1503 goto out;
1504 }
1505
Mark Fashehe63aecb62007-10-18 15:30:42 -07001506 ret = ocfs2_inode_lock(inode, &di_bh, 1);
Mark Fashehb2580102007-03-09 16:53:21 -08001507 if (ret) {
1508 mlog_errno(ret);
1509 goto out_rw_unlock;
1510 }
1511
1512 if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
1513 ret = -EPERM;
Mark Fashehe63aecb62007-10-18 15:30:42 -07001514 goto out_inode_unlock;
Mark Fashehb2580102007-03-09 16:53:21 -08001515 }
1516
1517 switch (sr->l_whence) {
1518 case 0: /*SEEK_SET*/
1519 break;
1520 case 1: /*SEEK_CUR*/
Mark Fasheh385820a2007-07-19 00:14:38 -07001521 sr->l_start += f_pos;
Mark Fashehb2580102007-03-09 16:53:21 -08001522 break;
1523 case 2: /*SEEK_END*/
1524 sr->l_start += i_size_read(inode);
1525 break;
1526 default:
1527 ret = -EINVAL;
Mark Fashehe63aecb62007-10-18 15:30:42 -07001528 goto out_inode_unlock;
Mark Fashehb2580102007-03-09 16:53:21 -08001529 }
1530 sr->l_whence = 0;
1531
1532 llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len;
1533
1534 if (sr->l_start < 0
1535 || sr->l_start > max_off
1536 || (sr->l_start + llen) < 0
1537 || (sr->l_start + llen) > max_off) {
1538 ret = -EINVAL;
Mark Fashehe63aecb62007-10-18 15:30:42 -07001539 goto out_inode_unlock;
Mark Fashehb2580102007-03-09 16:53:21 -08001540 }
Mark Fasheh385820a2007-07-19 00:14:38 -07001541 size = sr->l_start + sr->l_len;
Mark Fashehb2580102007-03-09 16:53:21 -08001542
1543 if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) {
1544 if (sr->l_len <= 0) {
1545 ret = -EINVAL;
Mark Fashehe63aecb62007-10-18 15:30:42 -07001546 goto out_inode_unlock;
Mark Fashehb2580102007-03-09 16:53:21 -08001547 }
1548 }
1549
Mark Fasheh385820a2007-07-19 00:14:38 -07001550 if (file && should_remove_suid(file->f_path.dentry)) {
Mark Fashehb2580102007-03-09 16:53:21 -08001551 ret = __ocfs2_write_remove_suid(inode, di_bh);
1552 if (ret) {
1553 mlog_errno(ret);
Mark Fashehe63aecb62007-10-18 15:30:42 -07001554 goto out_inode_unlock;
Mark Fashehb2580102007-03-09 16:53:21 -08001555 }
1556 }
1557
1558 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1559 switch (cmd) {
1560 case OCFS2_IOC_RESVSP:
1561 case OCFS2_IOC_RESVSP64:
1562 /*
1563 * This takes unsigned offsets, but the signed ones we
1564 * pass have been checked against overflow above.
1565 */
1566 ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start,
1567 sr->l_len);
1568 break;
1569 case OCFS2_IOC_UNRESVSP:
1570 case OCFS2_IOC_UNRESVSP64:
1571 ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start,
1572 sr->l_len);
1573 break;
1574 default:
1575 ret = -EINVAL;
1576 }
1577 up_write(&OCFS2_I(inode)->ip_alloc_sem);
1578 if (ret) {
1579 mlog_errno(ret);
Mark Fashehe63aecb62007-10-18 15:30:42 -07001580 goto out_inode_unlock;
Mark Fashehb2580102007-03-09 16:53:21 -08001581 }
1582
1583 /*
1584 * We update c/mtime for these changes
1585 */
1586 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1587 if (IS_ERR(handle)) {
1588 ret = PTR_ERR(handle);
1589 mlog_errno(ret);
Mark Fashehe63aecb62007-10-18 15:30:42 -07001590 goto out_inode_unlock;
Mark Fashehb2580102007-03-09 16:53:21 -08001591 }
1592
Mark Fasheh385820a2007-07-19 00:14:38 -07001593 if (change_size && i_size_read(inode) < size)
1594 i_size_write(inode, size);
1595
Mark Fashehb2580102007-03-09 16:53:21 -08001596 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1597 ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
1598 if (ret < 0)
1599 mlog_errno(ret);
1600
1601 ocfs2_commit_trans(osb, handle);
1602
Mark Fashehe63aecb62007-10-18 15:30:42 -07001603out_inode_unlock:
Mark Fashehb2580102007-03-09 16:53:21 -08001604 brelse(di_bh);
Mark Fashehe63aecb62007-10-18 15:30:42 -07001605 ocfs2_inode_unlock(inode, 1);
Mark Fashehb2580102007-03-09 16:53:21 -08001606out_rw_unlock:
1607 ocfs2_rw_unlock(inode, 1);
1608
Mark Fashehb2580102007-03-09 16:53:21 -08001609out:
Julia Lawallc259ae52008-07-21 09:59:15 +02001610 mutex_unlock(&inode->i_mutex);
Mark Fashehb2580102007-03-09 16:53:21 -08001611 return ret;
1612}
1613
Mark Fasheh385820a2007-07-19 00:14:38 -07001614int ocfs2_change_file_space(struct file *file, unsigned int cmd,
1615 struct ocfs2_space_resv *sr)
1616{
1617 struct inode *inode = file->f_path.dentry->d_inode;
Fernando Carrijoc19a28e2009-01-07 18:09:08 -08001618 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
Mark Fasheh385820a2007-07-19 00:14:38 -07001619
1620 if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) &&
1621 !ocfs2_writes_unwritten_extents(osb))
1622 return -ENOTTY;
1623 else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) &&
1624 !ocfs2_sparse_alloc(osb))
1625 return -ENOTTY;
1626
1627 if (!S_ISREG(inode->i_mode))
1628 return -EINVAL;
1629
1630 if (!(file->f_mode & FMODE_WRITE))
1631 return -EBADF;
1632
1633 return __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0);
1634}
1635
1636static long ocfs2_fallocate(struct inode *inode, int mode, loff_t offset,
1637 loff_t len)
1638{
1639 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1640 struct ocfs2_space_resv sr;
1641 int change_size = 1;
1642
1643 if (!ocfs2_writes_unwritten_extents(osb))
1644 return -EOPNOTSUPP;
1645
1646 if (S_ISDIR(inode->i_mode))
1647 return -ENODEV;
1648
1649 if (mode & FALLOC_FL_KEEP_SIZE)
1650 change_size = 0;
1651
1652 sr.l_whence = 0;
1653 sr.l_start = (s64)offset;
1654 sr.l_len = (s64)len;
1655
1656 return __ocfs2_change_file_space(NULL, inode, offset,
1657 OCFS2_IOC_RESVSP64, &sr, change_size);
1658}
1659
Tiger Yang8659ac22006-10-17 18:29:52 -07001660static int ocfs2_prepare_inode_for_write(struct dentry *dentry,
1661 loff_t *ppos,
1662 size_t count,
Mark Fasheh9517bac2007-02-09 20:24:12 -08001663 int appending,
1664 int *direct_io)
Mark Fashehccd979b2005-12-15 14:31:24 -08001665{
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001666 int ret = 0, meta_level = 0;
Tiger Yang8659ac22006-10-17 18:29:52 -07001667 struct inode *inode = dentry->d_inode;
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001668 loff_t saved_pos, end;
Mark Fashehccd979b2005-12-15 14:31:24 -08001669
Mark Fashehccd979b2005-12-15 14:31:24 -08001670 /*
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001671 * We start with a read level meta lock and only jump to an ex
1672 * if we need to make modifications here.
Mark Fashehccd979b2005-12-15 14:31:24 -08001673 */
Mark Fashehccd979b2005-12-15 14:31:24 -08001674 for(;;) {
Mark Fashehe63aecb62007-10-18 15:30:42 -07001675 ret = ocfs2_inode_lock(inode, NULL, meta_level);
Mark Fashehccd979b2005-12-15 14:31:24 -08001676 if (ret < 0) {
1677 meta_level = -1;
1678 mlog_errno(ret);
1679 goto out;
1680 }
1681
1682 /* Clear suid / sgid if necessary. We do this here
1683 * instead of later in the write path because
1684 * remove_suid() calls ->setattr without any hint that
1685 * we may have already done our cluster locking. Since
1686 * ocfs2_setattr() *must* take cluster locks to
1687 * proceeed, this will lead us to recursively lock the
1688 * inode. There's also the dinode i_size state which
1689 * can be lost via setattr during extending writes (we
1690 * set inode->i_size at the end of a write. */
Tiger Yang8659ac22006-10-17 18:29:52 -07001691 if (should_remove_suid(dentry)) {
Mark Fashehccd979b2005-12-15 14:31:24 -08001692 if (meta_level == 0) {
Mark Fashehe63aecb62007-10-18 15:30:42 -07001693 ocfs2_inode_unlock(inode, meta_level);
Mark Fashehccd979b2005-12-15 14:31:24 -08001694 meta_level = 1;
1695 continue;
1696 }
1697
1698 ret = ocfs2_write_remove_suid(inode);
1699 if (ret < 0) {
1700 mlog_errno(ret);
Tiger Yang8659ac22006-10-17 18:29:52 -07001701 goto out_unlock;
Mark Fashehccd979b2005-12-15 14:31:24 -08001702 }
1703 }
1704
1705 /* work on a copy of ppos until we're sure that we won't have
1706 * to recalculate it due to relocking. */
Tiger Yang8659ac22006-10-17 18:29:52 -07001707 if (appending) {
Mark Fashehccd979b2005-12-15 14:31:24 -08001708 saved_pos = i_size_read(inode);
1709 mlog(0, "O_APPEND: inode->i_size=%llu\n", saved_pos);
1710 } else {
Tiger Yang8659ac22006-10-17 18:29:52 -07001711 saved_pos = *ppos;
Mark Fashehccd979b2005-12-15 14:31:24 -08001712 }
Mark Fasheh3a0782d2007-01-17 12:53:31 -08001713
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001714 end = saved_pos + count;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001715
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001716 /*
1717 * Skip the O_DIRECT checks if we don't need
1718 * them.
1719 */
1720 if (!direct_io || !(*direct_io))
1721 break;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001722
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001723 /*
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001724 * There's no sane way to do direct writes to an inode
1725 * with inline data.
1726 */
1727 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1728 *direct_io = 0;
1729 break;
1730 }
1731
1732 /*
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001733 * Allowing concurrent direct writes means
1734 * i_size changes wouldn't be synchronized, so
1735 * one node could wind up truncating another
1736 * nodes writes.
1737 */
1738 if (end > i_size_read(inode)) {
1739 *direct_io = 0;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001740 break;
1741 }
1742
Mark Fasheh3a0782d2007-01-17 12:53:31 -08001743 /*
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001744 * We don't fill holes during direct io, so
1745 * check for them here. If any are found, the
1746 * caller will have to retake some cluster
1747 * locks and initiate the io as buffered.
Mark Fasheh3a0782d2007-01-17 12:53:31 -08001748 */
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001749 ret = ocfs2_check_range_for_holes(inode, saved_pos, count);
1750 if (ret == 1) {
1751 *direct_io = 0;
1752 ret = 0;
1753 } else if (ret < 0)
1754 mlog_errno(ret);
Mark Fashehccd979b2005-12-15 14:31:24 -08001755 break;
1756 }
1757
Tiger Yang8659ac22006-10-17 18:29:52 -07001758 if (appending)
1759 *ppos = saved_pos;
1760
1761out_unlock:
Mark Fashehe63aecb62007-10-18 15:30:42 -07001762 ocfs2_inode_unlock(inode, meta_level);
Tiger Yang8659ac22006-10-17 18:29:52 -07001763
1764out:
1765 return ret;
1766}
1767
1768static ssize_t ocfs2_file_aio_write(struct kiocb *iocb,
1769 const struct iovec *iov,
1770 unsigned long nr_segs,
1771 loff_t pos)
1772{
Mark Fasheh9517bac2007-02-09 20:24:12 -08001773 int ret, direct_io, appending, rw_level, have_alloc_sem = 0;
Nick Pigginb6af1bc2007-10-16 01:25:24 -07001774 int can_do_direct;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001775 ssize_t written = 0;
1776 size_t ocount; /* original count */
1777 size_t count; /* after file limit checks */
Mark Fasheh9ea2d322007-10-18 14:14:45 -07001778 loff_t old_size, *ppos = &iocb->ki_pos;
1779 u32 old_clusters;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001780 struct file *file = iocb->ki_filp;
1781 struct inode *inode = file->f_path.dentry->d_inode;
Mark Fasheh9ea2d322007-10-18 14:14:45 -07001782 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
Tiger Yang8659ac22006-10-17 18:29:52 -07001783
Mark Fasheh9517bac2007-02-09 20:24:12 -08001784 mlog_entry("(0x%p, %u, '%.*s')\n", file,
Tiger Yang8659ac22006-10-17 18:29:52 -07001785 (unsigned int)nr_segs,
Mark Fasheh9517bac2007-02-09 20:24:12 -08001786 file->f_path.dentry->d_name.len,
1787 file->f_path.dentry->d_name.name);
Tiger Yang8659ac22006-10-17 18:29:52 -07001788
Tiger Yang8659ac22006-10-17 18:29:52 -07001789 if (iocb->ki_left == 0)
1790 return 0;
1791
Mark Fasheh9517bac2007-02-09 20:24:12 -08001792 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
1793
1794 appending = file->f_flags & O_APPEND ? 1 : 0;
1795 direct_io = file->f_flags & O_DIRECT ? 1 : 0;
1796
Tiger Yang8659ac22006-10-17 18:29:52 -07001797 mutex_lock(&inode->i_mutex);
Mark Fasheh9517bac2007-02-09 20:24:12 -08001798
1799relock:
Tiger Yang8659ac22006-10-17 18:29:52 -07001800 /* to match setattr's i_mutex -> i_alloc_sem -> rw_lock ordering */
Mark Fasheh9517bac2007-02-09 20:24:12 -08001801 if (direct_io) {
Tiger Yang8659ac22006-10-17 18:29:52 -07001802 down_read(&inode->i_alloc_sem);
Mark Fasheh9517bac2007-02-09 20:24:12 -08001803 have_alloc_sem = 1;
Tiger Yang8659ac22006-10-17 18:29:52 -07001804 }
1805
1806 /* concurrent O_DIRECT writes are allowed */
Mark Fasheh9517bac2007-02-09 20:24:12 -08001807 rw_level = !direct_io;
Tiger Yang8659ac22006-10-17 18:29:52 -07001808 ret = ocfs2_rw_lock(inode, rw_level);
1809 if (ret < 0) {
Mark Fasheh9517bac2007-02-09 20:24:12 -08001810 mlog_errno(ret);
1811 goto out_sems;
1812 }
1813
1814 can_do_direct = direct_io;
1815 ret = ocfs2_prepare_inode_for_write(file->f_path.dentry, ppos,
1816 iocb->ki_left, appending,
1817 &can_do_direct);
1818 if (ret < 0) {
Tiger Yang8659ac22006-10-17 18:29:52 -07001819 mlog_errno(ret);
1820 goto out;
1821 }
1822
Mark Fasheh9517bac2007-02-09 20:24:12 -08001823 /*
1824 * We can't complete the direct I/O as requested, fall back to
1825 * buffered I/O.
1826 */
1827 if (direct_io && !can_do_direct) {
1828 ocfs2_rw_unlock(inode, rw_level);
1829 up_read(&inode->i_alloc_sem);
1830
1831 have_alloc_sem = 0;
1832 rw_level = -1;
1833
1834 direct_io = 0;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001835 goto relock;
Tiger Yang8659ac22006-10-17 18:29:52 -07001836 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001837
Mark Fasheh9ea2d322007-10-18 14:14:45 -07001838 /*
1839 * To later detect whether a journal commit for sync writes is
1840 * necessary, we sample i_size, and cluster count here.
1841 */
1842 old_size = i_size_read(inode);
1843 old_clusters = OCFS2_I(inode)->ip_clusters;
1844
Mark Fashehccd979b2005-12-15 14:31:24 -08001845 /* communicate with ocfs2_dio_end_io */
Mark Fasheh7cdfc3a2007-04-16 17:28:51 -07001846 ocfs2_iocb_set_rw_locked(iocb, rw_level);
Mark Fashehccd979b2005-12-15 14:31:24 -08001847
Mark Fasheh9517bac2007-02-09 20:24:12 -08001848 if (direct_io) {
Nick Pigginb6af1bc2007-10-16 01:25:24 -07001849 ret = generic_segment_checks(iov, &nr_segs, &ocount,
1850 VERIFY_READ);
1851 if (ret)
1852 goto out_dio;
1853
1854 ret = generic_write_checks(file, ppos, &count,
1855 S_ISBLK(inode->i_mode));
1856 if (ret)
1857 goto out_dio;
1858
Mark Fasheh9517bac2007-02-09 20:24:12 -08001859 written = generic_file_direct_write(iocb, iov, &nr_segs, *ppos,
1860 ppos, count, ocount);
1861 if (written < 0) {
Dmitri Monakhovc4354002008-10-27 13:01:49 -07001862 /*
1863 * direct write may have instantiated a few
1864 * blocks outside i_size. Trim these off again.
1865 * Don't need i_size_read because we hold i_mutex.
1866 */
1867 if (*ppos + count > inode->i_size)
1868 vmtruncate(inode, inode->i_size);
Mark Fasheh9517bac2007-02-09 20:24:12 -08001869 ret = written;
1870 goto out_dio;
1871 }
1872 } else {
Nick Pigginb6af1bc2007-10-16 01:25:24 -07001873 written = generic_file_aio_write_nolock(iocb, iov, nr_segs,
1874 *ppos);
Mark Fasheh9517bac2007-02-09 20:24:12 -08001875 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001876
Mark Fasheh9517bac2007-02-09 20:24:12 -08001877out_dio:
Mark Fashehccd979b2005-12-15 14:31:24 -08001878 /* buffered aio wouldn't have proper lock coverage today */
Mark Fasheh9517bac2007-02-09 20:24:12 -08001879 BUG_ON(ret == -EIOCBQUEUED && !(file->f_flags & O_DIRECT));
Mark Fashehccd979b2005-12-15 14:31:24 -08001880
Mark Fasheh9ea2d322007-10-18 14:14:45 -07001881 if ((file->f_flags & O_SYNC && !direct_io) || IS_SYNC(inode)) {
1882 /*
1883 * The generic write paths have handled getting data
1884 * to disk, but since we don't make use of the dirty
1885 * inode list, a manual journal commit is necessary
1886 * here.
1887 */
1888 if (old_size != i_size_read(inode) ||
1889 old_clusters != OCFS2_I(inode)->ip_clusters) {
Joel Becker2b4e30f2008-09-03 20:03:41 -07001890 ret = jbd2_journal_force_commit(osb->journal->j_journal);
Mark Fasheh9ea2d322007-10-18 14:14:45 -07001891 if (ret < 0)
1892 written = ret;
1893 }
1894 }
1895
Mark Fashehccd979b2005-12-15 14:31:24 -08001896 /*
1897 * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
1898 * function pointer which is called when o_direct io completes so that
1899 * it can unlock our rw lock. (it's the clustered equivalent of
1900 * i_alloc_sem; protects truncate from racing with pending ios).
1901 * Unfortunately there are error cases which call end_io and others
1902 * that don't. so we don't have to unlock the rw_lock if either an
1903 * async dio is going to do it in the future or an end_io after an
1904 * error has already done it.
1905 */
1906 if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
1907 rw_level = -1;
1908 have_alloc_sem = 0;
1909 }
1910
1911out:
Mark Fasheh9517bac2007-02-09 20:24:12 -08001912 if (rw_level != -1)
1913 ocfs2_rw_unlock(inode, rw_level);
1914
1915out_sems:
Mark Fashehccd979b2005-12-15 14:31:24 -08001916 if (have_alloc_sem)
1917 up_read(&inode->i_alloc_sem);
Mark Fasheh9517bac2007-02-09 20:24:12 -08001918
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001919 mutex_unlock(&inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -08001920
1921 mlog_exit(ret);
Mark Fasheh9517bac2007-02-09 20:24:12 -08001922 return written ? written : ret;
Mark Fashehccd979b2005-12-15 14:31:24 -08001923}
1924
Miklos Szeredi328eaab2009-04-14 19:48:39 +02001925static int ocfs2_splice_to_file(struct pipe_inode_info *pipe,
1926 struct file *out,
1927 struct splice_desc *sd)
1928{
1929 int ret;
1930
1931 ret = ocfs2_prepare_inode_for_write(out->f_path.dentry, &sd->pos,
1932 sd->total_len, 0, NULL);
1933 if (ret < 0) {
1934 mlog_errno(ret);
1935 return ret;
1936 }
1937
1938 return splice_from_pipe_feed(pipe, sd, pipe_to_file);
1939}
1940
Tiger Yang8659ac22006-10-17 18:29:52 -07001941static ssize_t ocfs2_file_splice_write(struct pipe_inode_info *pipe,
1942 struct file *out,
1943 loff_t *ppos,
1944 size_t len,
1945 unsigned int flags)
1946{
1947 int ret;
Miklos Szeredi328eaab2009-04-14 19:48:39 +02001948 struct address_space *mapping = out->f_mapping;
1949 struct inode *inode = mapping->host;
1950 struct splice_desc sd = {
1951 .total_len = len,
1952 .flags = flags,
1953 .pos = *ppos,
1954 .u.file = out,
1955 };
Tiger Yang8659ac22006-10-17 18:29:52 -07001956
1957 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", out, pipe,
1958 (unsigned int)len,
Josef Sipekd28c9172006-12-08 02:37:25 -08001959 out->f_path.dentry->d_name.len,
1960 out->f_path.dentry->d_name.name);
Tiger Yang8659ac22006-10-17 18:29:52 -07001961
Miklos Szeredi7bfac9e2009-04-06 17:41:00 +02001962 if (pipe->inode)
Miklos Szeredi328eaab2009-04-14 19:48:39 +02001963 mutex_lock_nested(&pipe->inode->i_mutex, I_MUTEX_PARENT);
1964
1965 splice_from_pipe_begin(&sd);
1966 do {
1967 ret = splice_from_pipe_next(pipe, &sd);
1968 if (ret <= 0)
1969 break;
1970
1971 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
1972 ret = ocfs2_rw_lock(inode, 1);
1973 if (ret < 0)
1974 mlog_errno(ret);
1975 else {
1976 ret = ocfs2_splice_to_file(pipe, out, &sd);
1977 ocfs2_rw_unlock(inode, 1);
1978 }
1979 mutex_unlock(&inode->i_mutex);
1980 } while (ret > 0);
1981 splice_from_pipe_end(pipe, &sd);
1982
Miklos Szeredi7bfac9e2009-04-06 17:41:00 +02001983 if (pipe->inode)
1984 mutex_unlock(&pipe->inode->i_mutex);
Tiger Yang8659ac22006-10-17 18:29:52 -07001985
Miklos Szeredi328eaab2009-04-14 19:48:39 +02001986 if (sd.num_spliced)
1987 ret = sd.num_spliced;
1988
1989 if (ret > 0) {
1990 unsigned long nr_pages;
1991
1992 *ppos += ret;
1993 nr_pages = (ret + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1994
1995 /*
1996 * If file or inode is SYNC and we actually wrote some data,
1997 * sync it.
1998 */
1999 if (unlikely((out->f_flags & O_SYNC) || IS_SYNC(inode))) {
2000 int err;
2001
2002 mutex_lock(&inode->i_mutex);
2003 err = ocfs2_rw_lock(inode, 1);
2004 if (err < 0) {
2005 mlog_errno(err);
2006 } else {
2007 err = generic_osync_inode(inode, mapping,
2008 OSYNC_METADATA|OSYNC_DATA);
2009 ocfs2_rw_unlock(inode, 1);
2010 }
2011 mutex_unlock(&inode->i_mutex);
2012
2013 if (err)
2014 ret = err;
2015 }
2016 balance_dirty_pages_ratelimited_nr(mapping, nr_pages);
2017 }
Tiger Yang8659ac22006-10-17 18:29:52 -07002018
2019 mlog_exit(ret);
2020 return ret;
2021}
2022
2023static ssize_t ocfs2_file_splice_read(struct file *in,
2024 loff_t *ppos,
2025 struct pipe_inode_info *pipe,
2026 size_t len,
2027 unsigned int flags)
2028{
Tao Ma1962f392009-06-19 15:36:52 +08002029 int ret = 0, lock_level = 0;
Josef Sipekd28c9172006-12-08 02:37:25 -08002030 struct inode *inode = in->f_path.dentry->d_inode;
Tiger Yang8659ac22006-10-17 18:29:52 -07002031
2032 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", in, pipe,
2033 (unsigned int)len,
Josef Sipekd28c9172006-12-08 02:37:25 -08002034 in->f_path.dentry->d_name.len,
2035 in->f_path.dentry->d_name.name);
Tiger Yang8659ac22006-10-17 18:29:52 -07002036
2037 /*
2038 * See the comment in ocfs2_file_aio_read()
2039 */
Tao Ma1962f392009-06-19 15:36:52 +08002040 ret = ocfs2_inode_lock_atime(inode, in->f_vfsmnt, &lock_level);
Tiger Yang8659ac22006-10-17 18:29:52 -07002041 if (ret < 0) {
2042 mlog_errno(ret);
2043 goto bail;
2044 }
Tao Ma1962f392009-06-19 15:36:52 +08002045 ocfs2_inode_unlock(inode, lock_level);
Tiger Yang8659ac22006-10-17 18:29:52 -07002046
2047 ret = generic_file_splice_read(in, ppos, pipe, len, flags);
2048
2049bail:
2050 mlog_exit(ret);
2051 return ret;
2052}
2053
Mark Fashehccd979b2005-12-15 14:31:24 -08002054static ssize_t ocfs2_file_aio_read(struct kiocb *iocb,
Badari Pulavarty027445c2006-09-30 23:28:46 -07002055 const struct iovec *iov,
2056 unsigned long nr_segs,
Mark Fashehccd979b2005-12-15 14:31:24 -08002057 loff_t pos)
2058{
Tiger Yang25899de2006-11-15 15:49:02 +08002059 int ret = 0, rw_level = -1, have_alloc_sem = 0, lock_level = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -08002060 struct file *filp = iocb->ki_filp;
Josef Sipekd28c9172006-12-08 02:37:25 -08002061 struct inode *inode = filp->f_path.dentry->d_inode;
Mark Fashehccd979b2005-12-15 14:31:24 -08002062
Badari Pulavarty027445c2006-09-30 23:28:46 -07002063 mlog_entry("(0x%p, %u, '%.*s')\n", filp,
2064 (unsigned int)nr_segs,
Josef Sipekd28c9172006-12-08 02:37:25 -08002065 filp->f_path.dentry->d_name.len,
2066 filp->f_path.dentry->d_name.name);
Mark Fashehccd979b2005-12-15 14:31:24 -08002067
2068 if (!inode) {
2069 ret = -EINVAL;
2070 mlog_errno(ret);
2071 goto bail;
2072 }
2073
Mark Fashehccd979b2005-12-15 14:31:24 -08002074 /*
2075 * buffered reads protect themselves in ->readpage(). O_DIRECT reads
2076 * need locks to protect pending reads from racing with truncate.
2077 */
2078 if (filp->f_flags & O_DIRECT) {
2079 down_read(&inode->i_alloc_sem);
2080 have_alloc_sem = 1;
2081
2082 ret = ocfs2_rw_lock(inode, 0);
2083 if (ret < 0) {
2084 mlog_errno(ret);
2085 goto bail;
2086 }
2087 rw_level = 0;
2088 /* communicate with ocfs2_dio_end_io */
Mark Fasheh7cdfc3a2007-04-16 17:28:51 -07002089 ocfs2_iocb_set_rw_locked(iocb, rw_level);
Mark Fashehccd979b2005-12-15 14:31:24 -08002090 }
2091
Mark Fashehc4374f82006-05-05 19:04:35 -07002092 /*
2093 * We're fine letting folks race truncates and extending
2094 * writes with read across the cluster, just like they can
2095 * locally. Hence no rw_lock during read.
2096 *
2097 * Take and drop the meta data lock to update inode fields
2098 * like i_size. This allows the checks down below
2099 * generic_file_aio_read() a chance of actually working.
2100 */
Mark Fashehe63aecb62007-10-18 15:30:42 -07002101 ret = ocfs2_inode_lock_atime(inode, filp->f_vfsmnt, &lock_level);
Mark Fashehc4374f82006-05-05 19:04:35 -07002102 if (ret < 0) {
2103 mlog_errno(ret);
2104 goto bail;
2105 }
Mark Fashehe63aecb62007-10-18 15:30:42 -07002106 ocfs2_inode_unlock(inode, lock_level);
Mark Fashehc4374f82006-05-05 19:04:35 -07002107
Badari Pulavarty027445c2006-09-30 23:28:46 -07002108 ret = generic_file_aio_read(iocb, iov, nr_segs, iocb->ki_pos);
Mark Fashehccd979b2005-12-15 14:31:24 -08002109 if (ret == -EINVAL)
Sunil Mushran56753bd2008-06-09 11:24:41 -07002110 mlog(0, "generic_file_aio_read returned -EINVAL\n");
Mark Fashehccd979b2005-12-15 14:31:24 -08002111
2112 /* buffered aio wouldn't have proper lock coverage today */
2113 BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT));
2114
2115 /* see ocfs2_file_aio_write */
2116 if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2117 rw_level = -1;
2118 have_alloc_sem = 0;
2119 }
2120
2121bail:
2122 if (have_alloc_sem)
2123 up_read(&inode->i_alloc_sem);
2124 if (rw_level != -1)
2125 ocfs2_rw_unlock(inode, rw_level);
2126 mlog_exit(ret);
2127
2128 return ret;
2129}
2130
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002131const struct inode_operations ocfs2_file_iops = {
Mark Fashehccd979b2005-12-15 14:31:24 -08002132 .setattr = ocfs2_setattr,
2133 .getattr = ocfs2_getattr,
Tiger Yangd38eb8d2006-11-27 09:59:21 +08002134 .permission = ocfs2_permission,
Tiger Yangcf1d6c72008-08-18 17:11:00 +08002135 .setxattr = generic_setxattr,
2136 .getxattr = generic_getxattr,
2137 .listxattr = ocfs2_listxattr,
2138 .removexattr = generic_removexattr,
Mark Fasheh385820a2007-07-19 00:14:38 -07002139 .fallocate = ocfs2_fallocate,
Mark Fasheh00dc4172008-10-03 17:32:11 -04002140 .fiemap = ocfs2_fiemap,
Mark Fashehccd979b2005-12-15 14:31:24 -08002141};
2142
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002143const struct inode_operations ocfs2_special_file_iops = {
Mark Fashehccd979b2005-12-15 14:31:24 -08002144 .setattr = ocfs2_setattr,
2145 .getattr = ocfs2_getattr,
Tiger Yangd38eb8d2006-11-27 09:59:21 +08002146 .permission = ocfs2_permission,
Mark Fashehccd979b2005-12-15 14:31:24 -08002147};
2148
Mark Fasheh53da4932008-07-21 14:29:16 -07002149/*
2150 * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with
2151 * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks!
2152 */
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08002153const struct file_operations ocfs2_fops = {
Jan Kara32c3c0e2007-12-19 15:24:52 +01002154 .llseek = generic_file_llseek,
Mark Fashehccd979b2005-12-15 14:31:24 -08002155 .read = do_sync_read,
2156 .write = do_sync_write,
Mark Fashehccd979b2005-12-15 14:31:24 -08002157 .mmap = ocfs2_mmap,
2158 .fsync = ocfs2_sync_file,
2159 .release = ocfs2_file_release,
2160 .open = ocfs2_file_open,
2161 .aio_read = ocfs2_file_aio_read,
2162 .aio_write = ocfs2_file_aio_write,
Andi Kleenc9ec1482008-01-27 03:17:17 +01002163 .unlocked_ioctl = ocfs2_ioctl,
Mark Fasheh586d2322007-03-09 15:56:28 -08002164#ifdef CONFIG_COMPAT
2165 .compat_ioctl = ocfs2_compat_ioctl,
2166#endif
Mark Fasheh53da4932008-07-21 14:29:16 -07002167 .lock = ocfs2_lock,
2168 .flock = ocfs2_flock,
2169 .splice_read = ocfs2_file_splice_read,
2170 .splice_write = ocfs2_file_splice_write,
2171};
2172
2173const struct file_operations ocfs2_dops = {
2174 .llseek = generic_file_llseek,
2175 .read = generic_read_dir,
2176 .readdir = ocfs2_readdir,
2177 .fsync = ocfs2_sync_file,
2178 .release = ocfs2_dir_release,
2179 .open = ocfs2_dir_open,
2180 .unlocked_ioctl = ocfs2_ioctl,
2181#ifdef CONFIG_COMPAT
2182 .compat_ioctl = ocfs2_compat_ioctl,
2183#endif
2184 .lock = ocfs2_lock,
2185 .flock = ocfs2_flock,
2186};
2187
2188/*
2189 * POSIX-lockless variants of our file_operations.
2190 *
2191 * These will be used if the underlying cluster stack does not support
2192 * posix file locking, if the user passes the "localflocks" mount
2193 * option, or if we have a local-only fs.
2194 *
2195 * ocfs2_flock is in here because all stacks handle UNIX file locks,
2196 * so we still want it in the case of no stack support for
2197 * plocks. Internally, it will do the right thing when asked to ignore
2198 * the cluster.
2199 */
2200const struct file_operations ocfs2_fops_no_plocks = {
2201 .llseek = generic_file_llseek,
2202 .read = do_sync_read,
2203 .write = do_sync_write,
2204 .mmap = ocfs2_mmap,
2205 .fsync = ocfs2_sync_file,
2206 .release = ocfs2_file_release,
2207 .open = ocfs2_file_open,
2208 .aio_read = ocfs2_file_aio_read,
2209 .aio_write = ocfs2_file_aio_write,
2210 .unlocked_ioctl = ocfs2_ioctl,
2211#ifdef CONFIG_COMPAT
2212 .compat_ioctl = ocfs2_compat_ioctl,
2213#endif
Mark Fasheh53fc6222007-12-20 16:49:04 -08002214 .flock = ocfs2_flock,
Tiger Yang8659ac22006-10-17 18:29:52 -07002215 .splice_read = ocfs2_file_splice_read,
2216 .splice_write = ocfs2_file_splice_write,
Mark Fashehccd979b2005-12-15 14:31:24 -08002217};
2218
Mark Fasheh53da4932008-07-21 14:29:16 -07002219const struct file_operations ocfs2_dops_no_plocks = {
Jan Kara32c3c0e2007-12-19 15:24:52 +01002220 .llseek = generic_file_llseek,
Mark Fashehccd979b2005-12-15 14:31:24 -08002221 .read = generic_read_dir,
2222 .readdir = ocfs2_readdir,
2223 .fsync = ocfs2_sync_file,
Mark Fasheh53fc6222007-12-20 16:49:04 -08002224 .release = ocfs2_dir_release,
2225 .open = ocfs2_dir_open,
Andi Kleenc9ec1482008-01-27 03:17:17 +01002226 .unlocked_ioctl = ocfs2_ioctl,
Mark Fasheh586d2322007-03-09 15:56:28 -08002227#ifdef CONFIG_COMPAT
2228 .compat_ioctl = ocfs2_compat_ioctl,
2229#endif
Mark Fasheh53fc6222007-12-20 16:49:04 -08002230 .flock = ocfs2_flock,
Mark Fashehccd979b2005-12-15 14:31:24 -08002231};