blob: 9154c82d3258e5afc4b88fde94ecc68fc0aa91e2 [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"
58
59#include "buffer_head_io.h"
60
61static int ocfs2_sync_inode(struct inode *inode)
62{
63 filemap_fdatawrite(inode->i_mapping);
64 return sync_mapping_buffers(inode->i_mapping);
65}
66
Mark Fasheh53fc6222007-12-20 16:49:04 -080067static int ocfs2_init_file_private(struct inode *inode, struct file *file)
68{
69 struct ocfs2_file_private *fp;
70
71 fp = kzalloc(sizeof(struct ocfs2_file_private), GFP_KERNEL);
72 if (!fp)
73 return -ENOMEM;
74
75 fp->fp_file = file;
76 mutex_init(&fp->fp_mutex);
77 ocfs2_file_lock_res_init(&fp->fp_flock, fp);
78 file->private_data = fp;
79
80 return 0;
81}
82
83static void ocfs2_free_file_private(struct inode *inode, struct file *file)
84{
85 struct ocfs2_file_private *fp = file->private_data;
86 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
87
88 if (fp) {
89 ocfs2_simple_drop_lockres(osb, &fp->fp_flock);
90 ocfs2_lock_res_free(&fp->fp_flock);
91 kfree(fp);
92 file->private_data = NULL;
93 }
94}
95
Mark Fashehccd979b2005-12-15 14:31:24 -080096static int ocfs2_file_open(struct inode *inode, struct file *file)
97{
98 int status;
99 int mode = file->f_flags;
100 struct ocfs2_inode_info *oi = OCFS2_I(inode);
101
102 mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file,
Josef Sipekd28c9172006-12-08 02:37:25 -0800103 file->f_path.dentry->d_name.len, file->f_path.dentry->d_name.name);
Mark Fashehccd979b2005-12-15 14:31:24 -0800104
105 spin_lock(&oi->ip_lock);
106
107 /* Check that the inode hasn't been wiped from disk by another
108 * node. If it hasn't then we're safe as long as we hold the
109 * spin lock until our increment of open count. */
110 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
111 spin_unlock(&oi->ip_lock);
112
113 status = -ENOENT;
114 goto leave;
115 }
116
117 if (mode & O_DIRECT)
118 oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT;
119
120 oi->ip_open_count++;
121 spin_unlock(&oi->ip_lock);
Mark Fasheh53fc6222007-12-20 16:49:04 -0800122
123 status = ocfs2_init_file_private(inode, file);
124 if (status) {
125 /*
126 * We want to set open count back if we're failing the
127 * open.
128 */
129 spin_lock(&oi->ip_lock);
130 oi->ip_open_count--;
131 spin_unlock(&oi->ip_lock);
132 }
133
Mark Fashehccd979b2005-12-15 14:31:24 -0800134leave:
135 mlog_exit(status);
136 return status;
137}
138
139static int ocfs2_file_release(struct inode *inode, struct file *file)
140{
141 struct ocfs2_inode_info *oi = OCFS2_I(inode);
142
143 mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file,
Josef Sipekd28c9172006-12-08 02:37:25 -0800144 file->f_path.dentry->d_name.len,
145 file->f_path.dentry->d_name.name);
Mark Fashehccd979b2005-12-15 14:31:24 -0800146
147 spin_lock(&oi->ip_lock);
148 if (!--oi->ip_open_count)
149 oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT;
150 spin_unlock(&oi->ip_lock);
151
Mark Fasheh53fc6222007-12-20 16:49:04 -0800152 ocfs2_free_file_private(inode, file);
153
Mark Fashehccd979b2005-12-15 14:31:24 -0800154 mlog_exit(0);
155
156 return 0;
157}
158
Mark Fasheh53fc6222007-12-20 16:49:04 -0800159static int ocfs2_dir_open(struct inode *inode, struct file *file)
160{
161 return ocfs2_init_file_private(inode, file);
162}
163
164static int ocfs2_dir_release(struct inode *inode, struct file *file)
165{
166 ocfs2_free_file_private(inode, file);
167 return 0;
168}
169
Mark Fashehccd979b2005-12-15 14:31:24 -0800170static int ocfs2_sync_file(struct file *file,
171 struct dentry *dentry,
172 int datasync)
173{
174 int err = 0;
175 journal_t *journal;
176 struct inode *inode = dentry->d_inode;
177 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
178
179 mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", file, dentry, datasync,
180 dentry->d_name.len, dentry->d_name.name);
181
182 err = ocfs2_sync_inode(dentry->d_inode);
183 if (err)
184 goto bail;
185
186 journal = osb->journal->j_journal;
187 err = journal_force_commit(journal);
188
189bail:
190 mlog_exit(err);
191
192 return (err < 0) ? -EIO : 0;
193}
194
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800195int ocfs2_should_update_atime(struct inode *inode,
196 struct vfsmount *vfsmnt)
197{
198 struct timespec now;
199 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
200
201 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
202 return 0;
203
204 if ((inode->i_flags & S_NOATIME) ||
205 ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)))
206 return 0;
207
Mark Fasheh6c2aad02006-12-19 15:25:52 -0800208 /*
209 * We can be called with no vfsmnt structure - NFSD will
210 * sometimes do this.
211 *
212 * Note that our action here is different than touch_atime() -
213 * if we can't tell whether this is a noatime mount, then we
214 * don't know whether to trust the value of s_atime_quantum.
215 */
216 if (vfsmnt == NULL)
217 return 0;
218
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800219 if ((vfsmnt->mnt_flags & MNT_NOATIME) ||
220 ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
221 return 0;
222
Mark Fasheh7e913c52006-12-13 00:34:35 -0800223 if (vfsmnt->mnt_flags & MNT_RELATIME) {
224 if ((timespec_compare(&inode->i_atime, &inode->i_mtime) <= 0) ||
225 (timespec_compare(&inode->i_atime, &inode->i_ctime) <= 0))
226 return 1;
227
228 return 0;
229 }
230
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800231 now = CURRENT_TIME;
232 if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum))
233 return 0;
234 else
235 return 1;
236}
237
238int ocfs2_update_inode_atime(struct inode *inode,
239 struct buffer_head *bh)
240{
241 int ret;
242 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
243 handle_t *handle;
Mark Fashehc11e9fa2007-07-20 11:24:53 -0700244 struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data;
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800245
246 mlog_entry_void();
247
248 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
249 if (handle == NULL) {
250 ret = -ENOMEM;
251 mlog_errno(ret);
252 goto out;
253 }
254
Mark Fashehc11e9fa2007-07-20 11:24:53 -0700255 ret = ocfs2_journal_access(handle, inode, bh,
256 OCFS2_JOURNAL_ACCESS_WRITE);
257 if (ret) {
258 mlog_errno(ret);
259 goto out_commit;
260 }
261
262 /*
263 * Don't use ocfs2_mark_inode_dirty() here as we don't always
264 * have i_mutex to guard against concurrent changes to other
265 * inode fields.
266 */
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800267 inode->i_atime = CURRENT_TIME;
Mark Fashehc11e9fa2007-07-20 11:24:53 -0700268 di->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
269 di->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
270
271 ret = ocfs2_journal_dirty(handle, bh);
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800272 if (ret < 0)
273 mlog_errno(ret);
274
Mark Fashehc11e9fa2007-07-20 11:24:53 -0700275out_commit:
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800276 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
277out:
278 mlog_exit(ret);
279 return ret;
280}
281
Adrian Bunk6cb129f2007-04-26 00:29:35 -0700282static int ocfs2_set_inode_size(handle_t *handle,
283 struct inode *inode,
284 struct buffer_head *fe_bh,
285 u64 new_i_size)
Mark Fashehccd979b2005-12-15 14:31:24 -0800286{
287 int status;
288
289 mlog_entry_void();
290 i_size_write(inode, new_i_size);
Mark Fasheh8110b072007-03-22 16:53:23 -0700291 inode->i_blocks = ocfs2_inode_sector_count(inode);
Mark Fashehccd979b2005-12-15 14:31:24 -0800292 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
293
294 status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
295 if (status < 0) {
296 mlog_errno(status);
297 goto bail;
298 }
299
300bail:
301 mlog_exit(status);
302 return status;
303}
304
305static int ocfs2_simple_size_update(struct inode *inode,
306 struct buffer_head *di_bh,
307 u64 new_i_size)
308{
309 int ret;
310 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
Mark Fasheh1fabe142006-10-09 18:11:45 -0700311 handle_t *handle = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800312
Mark Fasheh65eff9c2006-10-09 17:26:22 -0700313 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
Mark Fashehccd979b2005-12-15 14:31:24 -0800314 if (handle == NULL) {
315 ret = -ENOMEM;
316 mlog_errno(ret);
317 goto out;
318 }
319
320 ret = ocfs2_set_inode_size(handle, inode, di_bh,
321 new_i_size);
322 if (ret < 0)
323 mlog_errno(ret);
324
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700325 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800326out:
327 return ret;
328}
329
330static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
331 struct inode *inode,
332 struct buffer_head *fe_bh,
333 u64 new_i_size)
334{
335 int status;
Mark Fasheh1fabe142006-10-09 18:11:45 -0700336 handle_t *handle;
Mark Fasheh60b11392007-02-16 11:46:50 -0800337 struct ocfs2_dinode *di;
Mark Fasheh35edec12007-07-06 14:41:18 -0700338 u64 cluster_bytes;
Mark Fashehccd979b2005-12-15 14:31:24 -0800339
340 mlog_entry_void();
341
342 /* TODO: This needs to actually orphan the inode in this
343 * transaction. */
344
Mark Fasheh65eff9c2006-10-09 17:26:22 -0700345 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
Mark Fashehccd979b2005-12-15 14:31:24 -0800346 if (IS_ERR(handle)) {
347 status = PTR_ERR(handle);
348 mlog_errno(status);
349 goto out;
350 }
351
Mark Fasheh60b11392007-02-16 11:46:50 -0800352 status = ocfs2_journal_access(handle, inode, fe_bh,
353 OCFS2_JOURNAL_ACCESS_WRITE);
354 if (status < 0) {
355 mlog_errno(status);
356 goto out_commit;
357 }
358
359 /*
360 * Do this before setting i_size.
361 */
Mark Fasheh35edec12007-07-06 14:41:18 -0700362 cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size);
363 status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size,
364 cluster_bytes);
Mark Fasheh60b11392007-02-16 11:46:50 -0800365 if (status) {
366 mlog_errno(status);
367 goto out_commit;
368 }
369
370 i_size_write(inode, new_i_size);
Mark Fasheh60b11392007-02-16 11:46:50 -0800371 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
372
373 di = (struct ocfs2_dinode *) fe_bh->b_data;
374 di->i_size = cpu_to_le64(new_i_size);
375 di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
376 di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
377
378 status = ocfs2_journal_dirty(handle, fe_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800379 if (status < 0)
380 mlog_errno(status);
381
Mark Fasheh60b11392007-02-16 11:46:50 -0800382out_commit:
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700383 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800384out:
Mark Fasheh60b11392007-02-16 11:46:50 -0800385
Mark Fashehccd979b2005-12-15 14:31:24 -0800386 mlog_exit(status);
387 return status;
388}
389
390static int ocfs2_truncate_file(struct inode *inode,
391 struct buffer_head *di_bh,
392 u64 new_i_size)
393{
394 int status = 0;
395 struct ocfs2_dinode *fe = NULL;
396 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
397 struct ocfs2_truncate_context *tc = NULL;
398
Mark Fashehb06970532006-03-03 10:24:33 -0800399 mlog_entry("(inode = %llu, new_i_size = %llu\n",
400 (unsigned long long)OCFS2_I(inode)->ip_blkno,
401 (unsigned long long)new_i_size);
Mark Fashehccd979b2005-12-15 14:31:24 -0800402
Mark Fashehccd979b2005-12-15 14:31:24 -0800403 fe = (struct ocfs2_dinode *) di_bh->b_data;
404 if (!OCFS2_IS_VALID_DINODE(fe)) {
405 OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe);
406 status = -EIO;
407 goto bail;
408 }
409
410 mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode),
Mark Fashehb06970532006-03-03 10:24:33 -0800411 "Inode %llu, inode i_size = %lld != di "
412 "i_size = %llu, i_flags = 0x%x\n",
413 (unsigned long long)OCFS2_I(inode)->ip_blkno,
Mark Fashehccd979b2005-12-15 14:31:24 -0800414 i_size_read(inode),
Mark Fashehb06970532006-03-03 10:24:33 -0800415 (unsigned long long)le64_to_cpu(fe->i_size),
416 le32_to_cpu(fe->i_flags));
Mark Fashehccd979b2005-12-15 14:31:24 -0800417
418 if (new_i_size > le64_to_cpu(fe->i_size)) {
Mark Fashehb06970532006-03-03 10:24:33 -0800419 mlog(0, "asked to truncate file with size (%llu) to size (%llu)!\n",
420 (unsigned long long)le64_to_cpu(fe->i_size),
421 (unsigned long long)new_i_size);
Mark Fashehccd979b2005-12-15 14:31:24 -0800422 status = -EINVAL;
423 mlog_errno(status);
424 goto bail;
425 }
426
Mark Fashehb06970532006-03-03 10:24:33 -0800427 mlog(0, "inode %llu, i_size = %llu, new_i_size = %llu\n",
428 (unsigned long long)le64_to_cpu(fe->i_blkno),
429 (unsigned long long)le64_to_cpu(fe->i_size),
430 (unsigned long long)new_i_size);
Mark Fashehccd979b2005-12-15 14:31:24 -0800431
432 /* lets handle the simple truncate cases before doing any more
433 * cluster locking. */
434 if (new_i_size == le64_to_cpu(fe->i_size))
435 goto bail;
436
Mark Fasheh2e89b2e2007-05-09 13:40:18 -0700437 down_write(&OCFS2_I(inode)->ip_alloc_sem);
438
Mark Fashehc934a922007-10-18 15:23:46 -0700439 /*
440 * The inode lock forced other nodes to sync and drop their
441 * pages, which (correctly) happens even if we have a truncate
442 * without allocation change - ocfs2 cluster sizes can be much
443 * greater than page size, so we have to truncate them
444 * anyway.
445 */
Mark Fasheh2e89b2e2007-05-09 13:40:18 -0700446 unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1);
447 truncate_inode_pages(inode->i_mapping, new_i_size);
448
Mark Fasheh1afc32b2007-09-07 14:46:51 -0700449 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
450 status = ocfs2_truncate_inline(inode, di_bh, new_i_size,
Mark Fashehb1967d02007-11-20 11:56:39 -0800451 i_size_read(inode), 1);
Mark Fasheh1afc32b2007-09-07 14:46:51 -0700452 if (status)
453 mlog_errno(status);
454
Mark Fashehc934a922007-10-18 15:23:46 -0700455 goto bail_unlock_sem;
Mark Fasheh1afc32b2007-09-07 14:46:51 -0700456 }
457
Mark Fashehccd979b2005-12-15 14:31:24 -0800458 /* alright, we're going to need to do a full blown alloc size
459 * change. Orphan the inode so that recovery can complete the
460 * truncate if necessary. This does the task of marking
461 * i_size. */
462 status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
463 if (status < 0) {
464 mlog_errno(status);
Mark Fashehc934a922007-10-18 15:23:46 -0700465 goto bail_unlock_sem;
Mark Fashehccd979b2005-12-15 14:31:24 -0800466 }
467
468 status = ocfs2_prepare_truncate(osb, inode, di_bh, &tc);
469 if (status < 0) {
470 mlog_errno(status);
Mark Fashehc934a922007-10-18 15:23:46 -0700471 goto bail_unlock_sem;
Mark Fashehccd979b2005-12-15 14:31:24 -0800472 }
473
474 status = ocfs2_commit_truncate(osb, inode, di_bh, tc);
475 if (status < 0) {
476 mlog_errno(status);
Mark Fashehc934a922007-10-18 15:23:46 -0700477 goto bail_unlock_sem;
Mark Fashehccd979b2005-12-15 14:31:24 -0800478 }
479
480 /* TODO: orphan dir cleanup here. */
Mark Fashehc934a922007-10-18 15:23:46 -0700481bail_unlock_sem:
Mark Fasheh2e89b2e2007-05-09 13:40:18 -0700482 up_write(&OCFS2_I(inode)->ip_alloc_sem);
483
Mark Fashehccd979b2005-12-15 14:31:24 -0800484bail:
485
486 mlog_exit(status);
487 return status;
488}
489
490/*
491 * extend allocation only here.
492 * we'll update all the disk stuff, and oip->alloc_size
493 *
494 * expect stuff to be locked, a transaction started and enough data /
495 * metadata reservations in the contexts.
496 *
497 * Will return -EAGAIN, and a reason if a restart is needed.
498 * If passed in, *reason will always be set, even in error.
499 */
500int ocfs2_do_extend_allocation(struct ocfs2_super *osb,
501 struct inode *inode,
Mark Fashehdcd05382007-01-16 11:32:23 -0800502 u32 *logical_offset,
Mark Fashehccd979b2005-12-15 14:31:24 -0800503 u32 clusters_to_add,
Mark Fasheh2ae99a62007-03-09 16:43:28 -0800504 int mark_unwritten,
Mark Fashehccd979b2005-12-15 14:31:24 -0800505 struct buffer_head *fe_bh,
Mark Fasheh1fabe142006-10-09 18:11:45 -0700506 handle_t *handle,
Mark Fashehccd979b2005-12-15 14:31:24 -0800507 struct ocfs2_alloc_context *data_ac,
508 struct ocfs2_alloc_context *meta_ac,
509 enum ocfs2_alloc_restarted *reason_ret)
510{
511 int status = 0;
512 int free_extents;
513 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
514 enum ocfs2_alloc_restarted reason = RESTART_NONE;
515 u32 bit_off, num_bits;
516 u64 block;
Mark Fasheh2ae99a62007-03-09 16:43:28 -0800517 u8 flags = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -0800518
519 BUG_ON(!clusters_to_add);
520
Mark Fasheh2ae99a62007-03-09 16:43:28 -0800521 if (mark_unwritten)
522 flags = OCFS2_EXT_UNWRITTEN;
523
Mark Fashehccd979b2005-12-15 14:31:24 -0800524 free_extents = ocfs2_num_free_extents(osb, inode, fe);
525 if (free_extents < 0) {
526 status = free_extents;
527 mlog_errno(status);
528 goto leave;
529 }
530
531 /* there are two cases which could cause us to EAGAIN in the
532 * we-need-more-metadata case:
533 * 1) we haven't reserved *any*
534 * 2) we are so fragmented, we've needed to add metadata too
535 * many times. */
536 if (!free_extents && !meta_ac) {
537 mlog(0, "we haven't reserved any metadata!\n");
538 status = -EAGAIN;
539 reason = RESTART_META;
540 goto leave;
541 } else if ((!free_extents)
542 && (ocfs2_alloc_context_bits_left(meta_ac)
543 < ocfs2_extend_meta_needed(fe))) {
544 mlog(0, "filesystem is really fragmented...\n");
545 status = -EAGAIN;
546 reason = RESTART_META;
547 goto leave;
548 }
549
Mark Fasheh415cb802007-09-16 20:10:16 -0700550 status = __ocfs2_claim_clusters(osb, handle, data_ac, 1,
551 clusters_to_add, &bit_off, &num_bits);
Mark Fashehccd979b2005-12-15 14:31:24 -0800552 if (status < 0) {
553 if (status != -ENOSPC)
554 mlog_errno(status);
555 goto leave;
556 }
557
558 BUG_ON(num_bits > clusters_to_add);
559
560 /* reserve our write early -- insert_extent may update the inode */
561 status = ocfs2_journal_access(handle, inode, fe_bh,
562 OCFS2_JOURNAL_ACCESS_WRITE);
563 if (status < 0) {
564 mlog_errno(status);
565 goto leave;
566 }
567
568 block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
Mark Fashehb06970532006-03-03 10:24:33 -0800569 mlog(0, "Allocating %u clusters at block %u for inode %llu\n",
570 num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
Mark Fashehdcd05382007-01-16 11:32:23 -0800571 status = ocfs2_insert_extent(osb, handle, inode, fe_bh,
572 *logical_offset, block, num_bits,
Mark Fasheh2ae99a62007-03-09 16:43:28 -0800573 flags, meta_ac);
Mark Fashehccd979b2005-12-15 14:31:24 -0800574 if (status < 0) {
575 mlog_errno(status);
576 goto leave;
577 }
578
Mark Fashehccd979b2005-12-15 14:31:24 -0800579 status = ocfs2_journal_dirty(handle, fe_bh);
580 if (status < 0) {
581 mlog_errno(status);
582 goto leave;
583 }
584
585 clusters_to_add -= num_bits;
Mark Fashehdcd05382007-01-16 11:32:23 -0800586 *logical_offset += num_bits;
Mark Fashehccd979b2005-12-15 14:31:24 -0800587
588 if (clusters_to_add) {
589 mlog(0, "need to alloc once more, clusters = %u, wanted = "
590 "%u\n", fe->i_clusters, clusters_to_add);
591 status = -EAGAIN;
592 reason = RESTART_TRANS;
593 }
594
595leave:
596 mlog_exit(status);
597 if (reason_ret)
598 *reason_ret = reason;
599 return status;
600}
601
Mark Fashehabf8b152007-01-17 13:07:24 -0800602/*
603 * For a given allocation, determine which allocators will need to be
604 * accessed, and lock them, reserving the appropriate number of bits.
605 *
Mark Fasheh2ae99a62007-03-09 16:43:28 -0800606 * Sparse file systems call this from ocfs2_write_begin_nolock()
607 * and ocfs2_allocate_unwritten_extents().
608 *
609 * File systems which don't support holes call this from
610 * ocfs2_extend_allocation().
Mark Fashehabf8b152007-01-17 13:07:24 -0800611 */
Mark Fasheh9517bac2007-02-09 20:24:12 -0800612int ocfs2_lock_allocators(struct inode *inode, struct ocfs2_dinode *di,
Mark Fashehb27b7cb2007-06-18 11:22:56 -0700613 u32 clusters_to_add, u32 extents_to_split,
Mark Fasheh9517bac2007-02-09 20:24:12 -0800614 struct ocfs2_alloc_context **data_ac,
615 struct ocfs2_alloc_context **meta_ac)
Mark Fashehabf8b152007-01-17 13:07:24 -0800616{
Mark Fasheh063c4562007-07-03 13:34:11 -0700617 int ret = 0, num_free_extents;
Mark Fashehb27b7cb2007-06-18 11:22:56 -0700618 unsigned int max_recs_needed = clusters_to_add + 2 * extents_to_split;
Mark Fashehabf8b152007-01-17 13:07:24 -0800619 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
620
621 *meta_ac = NULL;
Mark Fasheh063c4562007-07-03 13:34:11 -0700622 if (data_ac)
623 *data_ac = NULL;
624
625 BUG_ON(clusters_to_add != 0 && data_ac == NULL);
Mark Fashehabf8b152007-01-17 13:07:24 -0800626
627 mlog(0, "extend inode %llu, i_size = %lld, di->i_clusters = %u, "
Mark Fashehb27b7cb2007-06-18 11:22:56 -0700628 "clusters_to_add = %u, extents_to_split = %u\n",
Jan Kara634bf742007-12-19 15:25:42 +0100629 (unsigned long long)OCFS2_I(inode)->ip_blkno, (long long)i_size_read(inode),
Mark Fashehb27b7cb2007-06-18 11:22:56 -0700630 le32_to_cpu(di->i_clusters), clusters_to_add, extents_to_split);
Mark Fashehabf8b152007-01-17 13:07:24 -0800631
632 num_free_extents = ocfs2_num_free_extents(osb, inode, di);
633 if (num_free_extents < 0) {
634 ret = num_free_extents;
635 mlog_errno(ret);
636 goto out;
637 }
638
639 /*
640 * Sparse allocation file systems need to be more conservative
641 * with reserving room for expansion - the actual allocation
642 * happens while we've got a journal handle open so re-taking
643 * a cluster lock (because we ran out of room for another
644 * extent) will violate ordering rules.
645 *
Mark Fasheh9517bac2007-02-09 20:24:12 -0800646 * Most of the time we'll only be seeing this 1 cluster at a time
Mark Fashehabf8b152007-01-17 13:07:24 -0800647 * anyway.
Mark Fashehb27b7cb2007-06-18 11:22:56 -0700648 *
649 * Always lock for any unwritten extents - we might want to
650 * add blocks during a split.
Mark Fashehabf8b152007-01-17 13:07:24 -0800651 */
652 if (!num_free_extents ||
Mark Fashehb27b7cb2007-06-18 11:22:56 -0700653 (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed)) {
Mark Fashehabf8b152007-01-17 13:07:24 -0800654 ret = ocfs2_reserve_new_metadata(osb, di, meta_ac);
655 if (ret < 0) {
656 if (ret != -ENOSPC)
657 mlog_errno(ret);
658 goto out;
659 }
660 }
661
Mark Fasheh063c4562007-07-03 13:34:11 -0700662 if (clusters_to_add == 0)
663 goto out;
664
Mark Fashehabf8b152007-01-17 13:07:24 -0800665 ret = ocfs2_reserve_clusters(osb, clusters_to_add, data_ac);
666 if (ret < 0) {
667 if (ret != -ENOSPC)
668 mlog_errno(ret);
669 goto out;
670 }
671
672out:
673 if (ret) {
674 if (*meta_ac) {
675 ocfs2_free_alloc_context(*meta_ac);
676 *meta_ac = NULL;
677 }
678
679 /*
680 * We cannot have an error and a non null *data_ac.
681 */
682 }
683
684 return ret;
685}
686
Mark Fasheh2ae99a62007-03-09 16:43:28 -0800687static int __ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
688 u32 clusters_to_add, int mark_unwritten)
Mark Fashehccd979b2005-12-15 14:31:24 -0800689{
690 int status = 0;
691 int restart_func = 0;
Mark Fashehabf8b152007-01-17 13:07:24 -0800692 int credits;
Mark Fasheh2ae99a62007-03-09 16:43:28 -0800693 u32 prev_clusters;
Mark Fashehccd979b2005-12-15 14:31:24 -0800694 struct buffer_head *bh = NULL;
695 struct ocfs2_dinode *fe = NULL;
Mark Fasheh1fabe142006-10-09 18:11:45 -0700696 handle_t *handle = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800697 struct ocfs2_alloc_context *data_ac = NULL;
698 struct ocfs2_alloc_context *meta_ac = NULL;
699 enum ocfs2_alloc_restarted why;
700 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
701
702 mlog_entry("(clusters_to_add = %u)\n", clusters_to_add);
703
Mark Fashehdcd05382007-01-16 11:32:23 -0800704 /*
705 * This function only exists for file systems which don't
706 * support holes.
707 */
Mark Fasheh2ae99a62007-03-09 16:43:28 -0800708 BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb));
Mark Fashehdcd05382007-01-16 11:32:23 -0800709
Mark Fashehccd979b2005-12-15 14:31:24 -0800710 status = ocfs2_read_block(osb, OCFS2_I(inode)->ip_blkno, &bh,
711 OCFS2_BH_CACHED, inode);
712 if (status < 0) {
713 mlog_errno(status);
714 goto leave;
715 }
716
717 fe = (struct ocfs2_dinode *) bh->b_data;
718 if (!OCFS2_IS_VALID_DINODE(fe)) {
719 OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe);
720 status = -EIO;
721 goto leave;
722 }
723
724restart_all:
725 BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
726
Mark Fashehb27b7cb2007-06-18 11:22:56 -0700727 status = ocfs2_lock_allocators(inode, fe, clusters_to_add, 0, &data_ac,
Mark Fasheh9517bac2007-02-09 20:24:12 -0800728 &meta_ac);
729 if (status) {
730 mlog_errno(status);
731 goto leave;
732 }
733
Mark Fashehccd979b2005-12-15 14:31:24 -0800734 credits = ocfs2_calc_extend_credits(osb->sb, fe, clusters_to_add);
Mark Fasheh65eff9c2006-10-09 17:26:22 -0700735 handle = ocfs2_start_trans(osb, credits);
Mark Fashehccd979b2005-12-15 14:31:24 -0800736 if (IS_ERR(handle)) {
737 status = PTR_ERR(handle);
738 handle = NULL;
739 mlog_errno(status);
740 goto leave;
741 }
742
743restarted_transaction:
744 /* reserve a write to the file entry early on - that we if we
745 * run out of credits in the allocation path, we can still
746 * update i_size. */
747 status = ocfs2_journal_access(handle, inode, bh,
748 OCFS2_JOURNAL_ACCESS_WRITE);
749 if (status < 0) {
750 mlog_errno(status);
751 goto leave;
752 }
753
754 prev_clusters = OCFS2_I(inode)->ip_clusters;
755
756 status = ocfs2_do_extend_allocation(osb,
757 inode,
Mark Fashehdcd05382007-01-16 11:32:23 -0800758 &logical_start,
Mark Fashehccd979b2005-12-15 14:31:24 -0800759 clusters_to_add,
Mark Fasheh2ae99a62007-03-09 16:43:28 -0800760 mark_unwritten,
Mark Fashehccd979b2005-12-15 14:31:24 -0800761 bh,
762 handle,
763 data_ac,
764 meta_ac,
765 &why);
766 if ((status < 0) && (status != -EAGAIN)) {
767 if (status != -ENOSPC)
768 mlog_errno(status);
769 goto leave;
770 }
771
772 status = ocfs2_journal_dirty(handle, bh);
773 if (status < 0) {
774 mlog_errno(status);
775 goto leave;
776 }
777
778 spin_lock(&OCFS2_I(inode)->ip_lock);
779 clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
780 spin_unlock(&OCFS2_I(inode)->ip_lock);
781
782 if (why != RESTART_NONE && clusters_to_add) {
783 if (why == RESTART_META) {
784 mlog(0, "restarting function.\n");
785 restart_func = 1;
786 } else {
787 BUG_ON(why != RESTART_TRANS);
788
789 mlog(0, "restarting transaction.\n");
790 /* TODO: This can be more intelligent. */
791 credits = ocfs2_calc_extend_credits(osb->sb,
792 fe,
793 clusters_to_add);
Mark Fasheh1fabe142006-10-09 18:11:45 -0700794 status = ocfs2_extend_trans(handle, credits);
Mark Fashehccd979b2005-12-15 14:31:24 -0800795 if (status < 0) {
796 /* handle still has to be committed at
797 * this point. */
798 status = -ENOMEM;
799 mlog_errno(status);
800 goto leave;
801 }
802 goto restarted_transaction;
803 }
804 }
805
Mark Fashehb06970532006-03-03 10:24:33 -0800806 mlog(0, "fe: i_clusters = %u, i_size=%llu\n",
Mark Fasheh1ca1a112007-04-27 16:01:25 -0700807 le32_to_cpu(fe->i_clusters),
808 (unsigned long long)le64_to_cpu(fe->i_size));
Mark Fashehccd979b2005-12-15 14:31:24 -0800809 mlog(0, "inode: ip_clusters=%u, i_size=%lld\n",
Jan Kara634bf742007-12-19 15:25:42 +0100810 OCFS2_I(inode)->ip_clusters, (long long)i_size_read(inode));
Mark Fashehccd979b2005-12-15 14:31:24 -0800811
812leave:
Mark Fashehccd979b2005-12-15 14:31:24 -0800813 if (handle) {
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700814 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800815 handle = NULL;
816 }
817 if (data_ac) {
818 ocfs2_free_alloc_context(data_ac);
819 data_ac = NULL;
820 }
821 if (meta_ac) {
822 ocfs2_free_alloc_context(meta_ac);
823 meta_ac = NULL;
824 }
825 if ((!status) && restart_func) {
826 restart_func = 0;
827 goto restart_all;
828 }
829 if (bh) {
830 brelse(bh);
831 bh = NULL;
832 }
833
834 mlog_exit(status);
835 return status;
836}
837
838/* Some parts of this taken from generic_cont_expand, which turned out
839 * to be too fragile to do exactly what we need without us having to
Mark Fasheh53013cb2006-05-05 19:04:03 -0700840 * worry about recursive locking in ->prepare_write() and
841 * ->commit_write(). */
Mark Fashehccd979b2005-12-15 14:31:24 -0800842static int ocfs2_write_zero_page(struct inode *inode,
843 u64 size)
844{
845 struct address_space *mapping = inode->i_mapping;
846 struct page *page;
847 unsigned long index;
848 unsigned int offset;
Mark Fasheh1fabe142006-10-09 18:11:45 -0700849 handle_t *handle = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800850 int ret;
851
852 offset = (size & (PAGE_CACHE_SIZE-1)); /* Within page */
853 /* ugh. in prepare/commit_write, if from==to==start of block, we
854 ** skip the prepare. make sure we never send an offset for the start
855 ** of a block
856 */
857 if ((offset & (inode->i_sb->s_blocksize - 1)) == 0) {
858 offset++;
859 }
860 index = size >> PAGE_CACHE_SHIFT;
861
862 page = grab_cache_page(mapping, index);
863 if (!page) {
864 ret = -ENOMEM;
865 mlog_errno(ret);
866 goto out;
867 }
868
Mark Fasheh53013cb2006-05-05 19:04:03 -0700869 ret = ocfs2_prepare_write_nolock(inode, page, offset, offset);
Mark Fashehccd979b2005-12-15 14:31:24 -0800870 if (ret < 0) {
871 mlog_errno(ret);
872 goto out_unlock;
873 }
874
875 if (ocfs2_should_order_data(inode)) {
876 handle = ocfs2_start_walk_page_trans(inode, page, offset,
877 offset);
878 if (IS_ERR(handle)) {
879 ret = PTR_ERR(handle);
880 handle = NULL;
881 goto out_unlock;
882 }
883 }
884
885 /* must not update i_size! */
886 ret = block_commit_write(page, offset, offset);
887 if (ret < 0)
888 mlog_errno(ret);
889 else
890 ret = 0;
891
892 if (handle)
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700893 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800894out_unlock:
895 unlock_page(page);
896 page_cache_release(page);
897out:
898 return ret;
899}
900
901static int ocfs2_zero_extend(struct inode *inode,
902 u64 zero_to_size)
903{
904 int ret = 0;
905 u64 start_off;
906 struct super_block *sb = inode->i_sb;
907
908 start_off = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
909 while (start_off < zero_to_size) {
910 ret = ocfs2_write_zero_page(inode, start_off);
911 if (ret < 0) {
912 mlog_errno(ret);
913 goto out;
914 }
915
916 start_off += sb->s_blocksize;
Mark Fashehe2057c52006-10-03 17:53:05 -0700917
918 /*
919 * Very large extends have the potential to lock up
920 * the cpu for extended periods of time.
921 */
922 cond_resched();
Mark Fashehccd979b2005-12-15 14:31:24 -0800923 }
924
925out:
926 return ret;
927}
928
Mark Fasheh65ed39d2007-08-28 17:13:23 -0700929int ocfs2_extend_no_holes(struct inode *inode, u64 new_i_size, u64 zero_to)
930{
931 int ret;
932 u32 clusters_to_add;
933 struct ocfs2_inode_info *oi = OCFS2_I(inode);
934
935 clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size);
936 if (clusters_to_add < oi->ip_clusters)
937 clusters_to_add = 0;
938 else
939 clusters_to_add -= oi->ip_clusters;
940
941 if (clusters_to_add) {
942 ret = __ocfs2_extend_allocation(inode, oi->ip_clusters,
943 clusters_to_add, 0);
944 if (ret) {
945 mlog_errno(ret);
946 goto out;
947 }
948 }
949
950 /*
951 * Call this even if we don't add any clusters to the tree. We
952 * still need to zero the area between the old i_size and the
953 * new i_size.
954 */
955 ret = ocfs2_zero_extend(inode, zero_to);
956 if (ret < 0)
957 mlog_errno(ret);
958
959out:
960 return ret;
961}
962
Mark Fashehccd979b2005-12-15 14:31:24 -0800963static int ocfs2_extend_file(struct inode *inode,
964 struct buffer_head *di_bh,
Mark Fasheh65ed39d2007-08-28 17:13:23 -0700965 u64 new_i_size)
Mark Fashehccd979b2005-12-15 14:31:24 -0800966{
Mark Fashehc934a922007-10-18 15:23:46 -0700967 int ret = 0;
Mark Fasheh1afc32b2007-09-07 14:46:51 -0700968 struct ocfs2_inode_info *oi = OCFS2_I(inode);
Mark Fashehccd979b2005-12-15 14:31:24 -0800969
Mark Fasheh65ed39d2007-08-28 17:13:23 -0700970 BUG_ON(!di_bh);
Mark Fasheh53013cb2006-05-05 19:04:03 -0700971
Mark Fashehccd979b2005-12-15 14:31:24 -0800972 /* setattr sometimes calls us like this. */
973 if (new_i_size == 0)
974 goto out;
975
976 if (i_size_read(inode) == new_i_size)
977 goto out;
978 BUG_ON(new_i_size < i_size_read(inode));
979
Mark Fasheh1afc32b2007-09-07 14:46:51 -0700980 /*
981 * Fall through for converting inline data, even if the fs
982 * supports sparse files.
983 *
984 * The check for inline data here is legal - nobody can add
985 * the feature since we have i_mutex. We must check it again
986 * after acquiring ip_alloc_sem though, as paths like mmap
987 * might have raced us to converting the inode to extents.
988 */
989 if (!(oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
990 && ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
Mark Fasheh3a0782d2007-01-17 12:53:31 -0800991 goto out_update_size;
Mark Fashehccd979b2005-12-15 14:31:24 -0800992
Mark Fasheh0effef72006-10-03 17:44:42 -0700993 /*
Mark Fasheh65ed39d2007-08-28 17:13:23 -0700994 * The alloc sem blocks people in read/write from reading our
995 * allocation until we're done changing it. We depend on
996 * i_mutex to block other extend/truncate calls while we're
997 * here.
Mark Fasheh0effef72006-10-03 17:44:42 -0700998 */
Mark Fasheh1afc32b2007-09-07 14:46:51 -0700999 down_write(&oi->ip_alloc_sem);
1000
1001 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1002 /*
1003 * We can optimize small extends by keeping the inodes
1004 * inline data.
1005 */
1006 if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) {
1007 up_write(&oi->ip_alloc_sem);
1008 goto out_update_size;
1009 }
1010
1011 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1012 if (ret) {
1013 up_write(&oi->ip_alloc_sem);
1014
1015 mlog_errno(ret);
Mark Fashehc934a922007-10-18 15:23:46 -07001016 goto out;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001017 }
1018 }
1019
1020 if (!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
1021 ret = ocfs2_extend_no_holes(inode, new_i_size, new_i_size);
1022
1023 up_write(&oi->ip_alloc_sem);
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001024
Mark Fasheh0effef72006-10-03 17:44:42 -07001025 if (ret < 0) {
1026 mlog_errno(ret);
Mark Fashehc934a922007-10-18 15:23:46 -07001027 goto out;
Mark Fashehccd979b2005-12-15 14:31:24 -08001028 }
1029
Mark Fasheh3a0782d2007-01-17 12:53:31 -08001030out_update_size:
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001031 ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
1032 if (ret < 0)
1033 mlog_errno(ret);
Mark Fasheh53013cb2006-05-05 19:04:03 -07001034
Mark Fashehccd979b2005-12-15 14:31:24 -08001035out:
1036 return ret;
1037}
1038
1039int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
1040{
1041 int status = 0, size_change;
1042 struct inode *inode = dentry->d_inode;
1043 struct super_block *sb = inode->i_sb;
1044 struct ocfs2_super *osb = OCFS2_SB(sb);
1045 struct buffer_head *bh = NULL;
Mark Fasheh1fabe142006-10-09 18:11:45 -07001046 handle_t *handle = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -08001047
1048 mlog_entry("(0x%p, '%.*s')\n", dentry,
1049 dentry->d_name.len, dentry->d_name.name);
1050
1051 if (attr->ia_valid & ATTR_MODE)
1052 mlog(0, "mode change: %d\n", attr->ia_mode);
1053 if (attr->ia_valid & ATTR_UID)
1054 mlog(0, "uid change: %d\n", attr->ia_uid);
1055 if (attr->ia_valid & ATTR_GID)
1056 mlog(0, "gid change: %d\n", attr->ia_gid);
1057 if (attr->ia_valid & ATTR_SIZE)
1058 mlog(0, "size change...\n");
1059 if (attr->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME))
1060 mlog(0, "time change...\n");
1061
1062#define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
1063 | ATTR_GID | ATTR_UID | ATTR_MODE)
1064 if (!(attr->ia_valid & OCFS2_VALID_ATTRS)) {
1065 mlog(0, "can't handle attrs: 0x%x\n", attr->ia_valid);
1066 return 0;
1067 }
1068
1069 status = inode_change_ok(inode, attr);
1070 if (status)
1071 return status;
1072
1073 size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
1074 if (size_change) {
1075 status = ocfs2_rw_lock(inode, 1);
1076 if (status < 0) {
1077 mlog_errno(status);
1078 goto bail;
1079 }
1080 }
1081
Mark Fashehe63aecb62007-10-18 15:30:42 -07001082 status = ocfs2_inode_lock(inode, &bh, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -08001083 if (status < 0) {
1084 if (status != -ENOENT)
1085 mlog_errno(status);
1086 goto bail_unlock_rw;
1087 }
1088
1089 if (size_change && attr->ia_size != i_size_read(inode)) {
Mark Fashehce76fd32007-07-20 12:02:14 -07001090 if (attr->ia_size > sb->s_maxbytes) {
1091 status = -EFBIG;
1092 goto bail_unlock;
1093 }
1094
Mark Fashehccd979b2005-12-15 14:31:24 -08001095 if (i_size_read(inode) > attr->ia_size)
1096 status = ocfs2_truncate_file(inode, bh, attr->ia_size);
1097 else
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001098 status = ocfs2_extend_file(inode, bh, attr->ia_size);
Mark Fashehccd979b2005-12-15 14:31:24 -08001099 if (status < 0) {
1100 if (status != -ENOSPC)
1101 mlog_errno(status);
1102 status = -ENOSPC;
1103 goto bail_unlock;
1104 }
1105 }
1106
Mark Fasheh65eff9c2006-10-09 17:26:22 -07001107 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
Mark Fashehccd979b2005-12-15 14:31:24 -08001108 if (IS_ERR(handle)) {
1109 status = PTR_ERR(handle);
1110 mlog_errno(status);
1111 goto bail_unlock;
1112 }
1113
Mark Fasheh7307de82007-05-09 15:16:19 -07001114 /*
1115 * This will intentionally not wind up calling vmtruncate(),
1116 * since all the work for a size change has been done above.
1117 * Otherwise, we could get into problems with truncate as
1118 * ip_alloc_sem is used there to protect against i_size
1119 * changes.
1120 */
Mark Fashehccd979b2005-12-15 14:31:24 -08001121 status = inode_setattr(inode, attr);
1122 if (status < 0) {
1123 mlog_errno(status);
1124 goto bail_commit;
1125 }
1126
1127 status = ocfs2_mark_inode_dirty(handle, inode, bh);
1128 if (status < 0)
1129 mlog_errno(status);
1130
1131bail_commit:
Mark Fasheh02dc1af2006-10-09 16:48:10 -07001132 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -08001133bail_unlock:
Mark Fashehe63aecb62007-10-18 15:30:42 -07001134 ocfs2_inode_unlock(inode, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -08001135bail_unlock_rw:
1136 if (size_change)
1137 ocfs2_rw_unlock(inode, 1);
1138bail:
1139 if (bh)
1140 brelse(bh);
1141
1142 mlog_exit(status);
1143 return status;
1144}
1145
1146int ocfs2_getattr(struct vfsmount *mnt,
1147 struct dentry *dentry,
1148 struct kstat *stat)
1149{
1150 struct inode *inode = dentry->d_inode;
1151 struct super_block *sb = dentry->d_inode->i_sb;
1152 struct ocfs2_super *osb = sb->s_fs_info;
1153 int err;
1154
1155 mlog_entry_void();
1156
1157 err = ocfs2_inode_revalidate(dentry);
1158 if (err) {
1159 if (err != -ENOENT)
1160 mlog_errno(err);
1161 goto bail;
1162 }
1163
1164 generic_fillattr(inode, stat);
1165
1166 /* We set the blksize from the cluster size for performance */
1167 stat->blksize = osb->s_clustersize;
1168
1169bail:
1170 mlog_exit(err);
1171
1172 return err;
1173}
1174
Tiger Yangd38eb8d2006-11-27 09:59:21 +08001175int ocfs2_permission(struct inode *inode, int mask, struct nameidata *nd)
1176{
1177 int ret;
1178
1179 mlog_entry_void();
1180
Mark Fashehe63aecb62007-10-18 15:30:42 -07001181 ret = ocfs2_inode_lock(inode, NULL, 0);
Tiger Yangd38eb8d2006-11-27 09:59:21 +08001182 if (ret) {
Mark Fasheha9f5f702007-04-26 11:43:43 -07001183 if (ret != -ENOENT)
1184 mlog_errno(ret);
Tiger Yangd38eb8d2006-11-27 09:59:21 +08001185 goto out;
1186 }
1187
1188 ret = generic_permission(inode, mask, NULL);
Tiger Yangd38eb8d2006-11-27 09:59:21 +08001189
Mark Fashehe63aecb62007-10-18 15:30:42 -07001190 ocfs2_inode_unlock(inode, 0);
Tiger Yangd38eb8d2006-11-27 09:59:21 +08001191out:
1192 mlog_exit(ret);
1193 return ret;
1194}
1195
Mark Fashehb2580102007-03-09 16:53:21 -08001196static int __ocfs2_write_remove_suid(struct inode *inode,
1197 struct buffer_head *bh)
Mark Fashehccd979b2005-12-15 14:31:24 -08001198{
1199 int ret;
Mark Fasheh1fabe142006-10-09 18:11:45 -07001200 handle_t *handle;
Mark Fashehccd979b2005-12-15 14:31:24 -08001201 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1202 struct ocfs2_dinode *di;
1203
Mark Fashehb06970532006-03-03 10:24:33 -08001204 mlog_entry("(Inode %llu, mode 0%o)\n",
Mark Fashehb2580102007-03-09 16:53:21 -08001205 (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_mode);
Mark Fashehccd979b2005-12-15 14:31:24 -08001206
Mark Fasheh65eff9c2006-10-09 17:26:22 -07001207 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
Mark Fashehccd979b2005-12-15 14:31:24 -08001208 if (handle == NULL) {
1209 ret = -ENOMEM;
1210 mlog_errno(ret);
1211 goto out;
1212 }
1213
Mark Fashehccd979b2005-12-15 14:31:24 -08001214 ret = ocfs2_journal_access(handle, inode, bh,
1215 OCFS2_JOURNAL_ACCESS_WRITE);
1216 if (ret < 0) {
1217 mlog_errno(ret);
Mark Fashehb2580102007-03-09 16:53:21 -08001218 goto out_trans;
Mark Fashehccd979b2005-12-15 14:31:24 -08001219 }
1220
1221 inode->i_mode &= ~S_ISUID;
1222 if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
1223 inode->i_mode &= ~S_ISGID;
1224
1225 di = (struct ocfs2_dinode *) bh->b_data;
1226 di->i_mode = cpu_to_le16(inode->i_mode);
1227
1228 ret = ocfs2_journal_dirty(handle, bh);
1229 if (ret < 0)
1230 mlog_errno(ret);
Mark Fashehb2580102007-03-09 16:53:21 -08001231
Mark Fashehccd979b2005-12-15 14:31:24 -08001232out_trans:
Mark Fasheh02dc1af2006-10-09 16:48:10 -07001233 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -08001234out:
1235 mlog_exit(ret);
1236 return ret;
1237}
1238
Mark Fasheh9517bac2007-02-09 20:24:12 -08001239/*
1240 * Will look for holes and unwritten extents in the range starting at
1241 * pos for count bytes (inclusive).
1242 */
1243static int ocfs2_check_range_for_holes(struct inode *inode, loff_t pos,
1244 size_t count)
1245{
1246 int ret = 0;
Mark Fasheh49cb8d22007-03-09 16:21:46 -08001247 unsigned int extent_flags;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001248 u32 cpos, clusters, extent_len, phys_cpos;
1249 struct super_block *sb = inode->i_sb;
1250
1251 cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
1252 clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
1253
1254 while (clusters) {
Mark Fasheh49cb8d22007-03-09 16:21:46 -08001255 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
1256 &extent_flags);
Mark Fasheh9517bac2007-02-09 20:24:12 -08001257 if (ret < 0) {
1258 mlog_errno(ret);
1259 goto out;
1260 }
1261
Mark Fasheh49cb8d22007-03-09 16:21:46 -08001262 if (phys_cpos == 0 || (extent_flags & OCFS2_EXT_UNWRITTEN)) {
Mark Fasheh9517bac2007-02-09 20:24:12 -08001263 ret = 1;
1264 break;
1265 }
1266
1267 if (extent_len > clusters)
1268 extent_len = clusters;
1269
1270 clusters -= extent_len;
1271 cpos += extent_len;
1272 }
1273out:
1274 return ret;
1275}
1276
Mark Fashehb2580102007-03-09 16:53:21 -08001277static int ocfs2_write_remove_suid(struct inode *inode)
1278{
1279 int ret;
1280 struct buffer_head *bh = NULL;
1281 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1282
1283 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
1284 oi->ip_blkno, &bh, OCFS2_BH_CACHED, inode);
1285 if (ret < 0) {
1286 mlog_errno(ret);
1287 goto out;
1288 }
1289
1290 ret = __ocfs2_write_remove_suid(inode, bh);
1291out:
1292 brelse(bh);
1293 return ret;
1294}
1295
Mark Fasheh2ae99a62007-03-09 16:43:28 -08001296/*
1297 * Allocate enough extents to cover the region starting at byte offset
1298 * start for len bytes. Existing extents are skipped, any extents
1299 * added are marked as "unwritten".
1300 */
1301static int ocfs2_allocate_unwritten_extents(struct inode *inode,
1302 u64 start, u64 len)
1303{
1304 int ret;
1305 u32 cpos, phys_cpos, clusters, alloc_size;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001306 u64 end = start + len;
1307 struct buffer_head *di_bh = NULL;
1308
1309 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1310 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
1311 OCFS2_I(inode)->ip_blkno, &di_bh,
1312 OCFS2_BH_CACHED, inode);
1313 if (ret) {
1314 mlog_errno(ret);
1315 goto out;
1316 }
1317
1318 /*
1319 * Nothing to do if the requested reservation range
1320 * fits within the inode.
1321 */
1322 if (ocfs2_size_fits_inline_data(di_bh, end))
1323 goto out;
1324
1325 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1326 if (ret) {
1327 mlog_errno(ret);
1328 goto out;
1329 }
1330 }
Mark Fasheh2ae99a62007-03-09 16:43:28 -08001331
1332 /*
1333 * We consider both start and len to be inclusive.
1334 */
1335 cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
1336 clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len);
1337 clusters -= cpos;
1338
1339 while (clusters) {
1340 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1341 &alloc_size, NULL);
1342 if (ret) {
1343 mlog_errno(ret);
1344 goto out;
1345 }
1346
1347 /*
1348 * Hole or existing extent len can be arbitrary, so
1349 * cap it to our own allocation request.
1350 */
1351 if (alloc_size > clusters)
1352 alloc_size = clusters;
1353
1354 if (phys_cpos) {
1355 /*
1356 * We already have an allocation at this
1357 * region so we can safely skip it.
1358 */
1359 goto next;
1360 }
1361
1362 ret = __ocfs2_extend_allocation(inode, cpos, alloc_size, 1);
1363 if (ret) {
1364 if (ret != -ENOSPC)
1365 mlog_errno(ret);
1366 goto out;
1367 }
1368
1369next:
1370 cpos += alloc_size;
1371 clusters -= alloc_size;
1372 }
1373
1374 ret = 0;
1375out:
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001376
1377 brelse(di_bh);
Mark Fasheh2ae99a62007-03-09 16:43:28 -08001378 return ret;
1379}
1380
Mark Fasheh063c4562007-07-03 13:34:11 -07001381static int __ocfs2_remove_inode_range(struct inode *inode,
1382 struct buffer_head *di_bh,
1383 u32 cpos, u32 phys_cpos, u32 len,
1384 struct ocfs2_cached_dealloc_ctxt *dealloc)
1385{
1386 int ret;
1387 u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
1388 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1389 struct inode *tl_inode = osb->osb_tl_inode;
1390 handle_t *handle;
1391 struct ocfs2_alloc_context *meta_ac = NULL;
1392 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1393
1394 ret = ocfs2_lock_allocators(inode, di, 0, 1, NULL, &meta_ac);
1395 if (ret) {
1396 mlog_errno(ret);
1397 return ret;
1398 }
1399
1400 mutex_lock(&tl_inode->i_mutex);
1401
1402 if (ocfs2_truncate_log_needs_flush(osb)) {
1403 ret = __ocfs2_flush_truncate_log(osb);
1404 if (ret < 0) {
1405 mlog_errno(ret);
1406 goto out;
1407 }
1408 }
1409
1410 handle = ocfs2_start_trans(osb, OCFS2_REMOVE_EXTENT_CREDITS);
1411 if (handle == NULL) {
1412 ret = -ENOMEM;
1413 mlog_errno(ret);
1414 goto out;
1415 }
1416
1417 ret = ocfs2_journal_access(handle, inode, di_bh,
1418 OCFS2_JOURNAL_ACCESS_WRITE);
1419 if (ret) {
1420 mlog_errno(ret);
1421 goto out;
1422 }
1423
1424 ret = ocfs2_remove_extent(inode, di_bh, cpos, len, handle, meta_ac,
1425 dealloc);
1426 if (ret) {
1427 mlog_errno(ret);
1428 goto out_commit;
1429 }
1430
1431 OCFS2_I(inode)->ip_clusters -= len;
1432 di->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters);
1433
1434 ret = ocfs2_journal_dirty(handle, di_bh);
1435 if (ret) {
1436 mlog_errno(ret);
1437 goto out_commit;
1438 }
1439
1440 ret = ocfs2_truncate_log_append(osb, handle, phys_blkno, len);
1441 if (ret)
1442 mlog_errno(ret);
1443
1444out_commit:
1445 ocfs2_commit_trans(osb, handle);
1446out:
1447 mutex_unlock(&tl_inode->i_mutex);
1448
1449 if (meta_ac)
1450 ocfs2_free_alloc_context(meta_ac);
1451
1452 return ret;
1453}
1454
1455/*
1456 * Truncate a byte range, avoiding pages within partial clusters. This
1457 * preserves those pages for the zeroing code to write to.
1458 */
1459static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start,
1460 u64 byte_len)
1461{
1462 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1463 loff_t start, end;
1464 struct address_space *mapping = inode->i_mapping;
1465
1466 start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start);
1467 end = byte_start + byte_len;
1468 end = end & ~(osb->s_clustersize - 1);
1469
1470 if (start < end) {
1471 unmap_mapping_range(mapping, start, end - start, 0);
1472 truncate_inode_pages_range(mapping, start, end - 1);
1473 }
1474}
1475
1476static int ocfs2_zero_partial_clusters(struct inode *inode,
1477 u64 start, u64 len)
1478{
1479 int ret = 0;
1480 u64 tmpend, end = start + len;
1481 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1482 unsigned int csize = osb->s_clustersize;
1483 handle_t *handle;
1484
1485 /*
1486 * The "start" and "end" values are NOT necessarily part of
1487 * the range whose allocation is being deleted. Rather, this
1488 * is what the user passed in with the request. We must zero
1489 * partial clusters here. There's no need to worry about
1490 * physical allocation - the zeroing code knows to skip holes.
1491 */
1492 mlog(0, "byte start: %llu, end: %llu\n",
1493 (unsigned long long)start, (unsigned long long)end);
1494
1495 /*
1496 * If both edges are on a cluster boundary then there's no
1497 * zeroing required as the region is part of the allocation to
1498 * be truncated.
1499 */
1500 if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0)
1501 goto out;
1502
1503 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1504 if (handle == NULL) {
1505 ret = -ENOMEM;
1506 mlog_errno(ret);
1507 goto out;
1508 }
1509
1510 /*
1511 * We want to get the byte offset of the end of the 1st cluster.
1512 */
1513 tmpend = (u64)osb->s_clustersize + (start & ~(osb->s_clustersize - 1));
1514 if (tmpend > end)
1515 tmpend = end;
1516
1517 mlog(0, "1st range: start: %llu, tmpend: %llu\n",
1518 (unsigned long long)start, (unsigned long long)tmpend);
1519
1520 ret = ocfs2_zero_range_for_truncate(inode, handle, start, tmpend);
1521 if (ret)
1522 mlog_errno(ret);
1523
1524 if (tmpend < end) {
1525 /*
1526 * This may make start and end equal, but the zeroing
1527 * code will skip any work in that case so there's no
1528 * need to catch it up here.
1529 */
1530 start = end & ~(osb->s_clustersize - 1);
1531
1532 mlog(0, "2nd range: start: %llu, end: %llu\n",
1533 (unsigned long long)start, (unsigned long long)end);
1534
1535 ret = ocfs2_zero_range_for_truncate(inode, handle, start, end);
1536 if (ret)
1537 mlog_errno(ret);
1538 }
1539
1540 ocfs2_commit_trans(osb, handle);
1541out:
1542 return ret;
1543}
1544
1545static int ocfs2_remove_inode_range(struct inode *inode,
1546 struct buffer_head *di_bh, u64 byte_start,
1547 u64 byte_len)
1548{
1549 int ret = 0;
1550 u32 trunc_start, trunc_len, cpos, phys_cpos, alloc_size;
1551 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1552 struct ocfs2_cached_dealloc_ctxt dealloc;
Mark Fashehb1967d02007-11-20 11:56:39 -08001553 struct address_space *mapping = inode->i_mapping;
Mark Fasheh063c4562007-07-03 13:34:11 -07001554
1555 ocfs2_init_dealloc_ctxt(&dealloc);
1556
1557 if (byte_len == 0)
1558 return 0;
1559
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001560 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1561 ret = ocfs2_truncate_inline(inode, di_bh, byte_start,
Mark Fashehb1967d02007-11-20 11:56:39 -08001562 byte_start + byte_len, 0);
1563 if (ret) {
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001564 mlog_errno(ret);
Mark Fashehb1967d02007-11-20 11:56:39 -08001565 goto out;
1566 }
1567 /*
1568 * There's no need to get fancy with the page cache
1569 * truncate of an inline-data inode. We're talking
1570 * about less than a page here, which will be cached
1571 * in the dinode buffer anyway.
1572 */
1573 unmap_mapping_range(mapping, 0, 0, 0);
1574 truncate_inode_pages(mapping, 0);
1575 goto out;
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001576 }
1577
Mark Fasheh063c4562007-07-03 13:34:11 -07001578 trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start);
1579 trunc_len = (byte_start + byte_len) >> osb->s_clustersize_bits;
1580 if (trunc_len >= trunc_start)
1581 trunc_len -= trunc_start;
1582 else
1583 trunc_len = 0;
1584
1585 mlog(0, "Inode: %llu, start: %llu, len: %llu, cstart: %u, clen: %u\n",
1586 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1587 (unsigned long long)byte_start,
1588 (unsigned long long)byte_len, trunc_start, trunc_len);
1589
1590 ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len);
1591 if (ret) {
1592 mlog_errno(ret);
1593 goto out;
1594 }
1595
1596 cpos = trunc_start;
1597 while (trunc_len) {
1598 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1599 &alloc_size, NULL);
1600 if (ret) {
1601 mlog_errno(ret);
1602 goto out;
1603 }
1604
1605 if (alloc_size > trunc_len)
1606 alloc_size = trunc_len;
1607
1608 /* Only do work for non-holes */
1609 if (phys_cpos != 0) {
1610 ret = __ocfs2_remove_inode_range(inode, di_bh, cpos,
1611 phys_cpos, alloc_size,
1612 &dealloc);
1613 if (ret) {
1614 mlog_errno(ret);
1615 goto out;
1616 }
1617 }
1618
1619 cpos += alloc_size;
1620 trunc_len -= alloc_size;
1621 }
1622
1623 ocfs2_truncate_cluster_pages(inode, byte_start, byte_len);
1624
1625out:
1626 ocfs2_schedule_truncate_log_flush(osb, 1);
1627 ocfs2_run_deallocs(osb, &dealloc);
1628
1629 return ret;
1630}
1631
Mark Fashehb2580102007-03-09 16:53:21 -08001632/*
1633 * Parts of this function taken from xfs_change_file_space()
1634 */
Mark Fasheh385820a2007-07-19 00:14:38 -07001635static int __ocfs2_change_file_space(struct file *file, struct inode *inode,
1636 loff_t f_pos, unsigned int cmd,
1637 struct ocfs2_space_resv *sr,
1638 int change_size)
Mark Fashehb2580102007-03-09 16:53:21 -08001639{
1640 int ret;
1641 s64 llen;
Mark Fasheh385820a2007-07-19 00:14:38 -07001642 loff_t size;
Mark Fashehb2580102007-03-09 16:53:21 -08001643 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1644 struct buffer_head *di_bh = NULL;
1645 handle_t *handle;
Mark Fasheha00cce32007-07-20 11:28:30 -07001646 unsigned long long max_off = inode->i_sb->s_maxbytes;
Mark Fashehb2580102007-03-09 16:53:21 -08001647
Mark Fashehb2580102007-03-09 16:53:21 -08001648 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
1649 return -EROFS;
1650
1651 mutex_lock(&inode->i_mutex);
1652
1653 /*
1654 * This prevents concurrent writes on other nodes
1655 */
1656 ret = ocfs2_rw_lock(inode, 1);
1657 if (ret) {
1658 mlog_errno(ret);
1659 goto out;
1660 }
1661
Mark Fashehe63aecb62007-10-18 15:30:42 -07001662 ret = ocfs2_inode_lock(inode, &di_bh, 1);
Mark Fashehb2580102007-03-09 16:53:21 -08001663 if (ret) {
1664 mlog_errno(ret);
1665 goto out_rw_unlock;
1666 }
1667
1668 if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
1669 ret = -EPERM;
Mark Fashehe63aecb62007-10-18 15:30:42 -07001670 goto out_inode_unlock;
Mark Fashehb2580102007-03-09 16:53:21 -08001671 }
1672
1673 switch (sr->l_whence) {
1674 case 0: /*SEEK_SET*/
1675 break;
1676 case 1: /*SEEK_CUR*/
Mark Fasheh385820a2007-07-19 00:14:38 -07001677 sr->l_start += f_pos;
Mark Fashehb2580102007-03-09 16:53:21 -08001678 break;
1679 case 2: /*SEEK_END*/
1680 sr->l_start += i_size_read(inode);
1681 break;
1682 default:
1683 ret = -EINVAL;
Mark Fashehe63aecb62007-10-18 15:30:42 -07001684 goto out_inode_unlock;
Mark Fashehb2580102007-03-09 16:53:21 -08001685 }
1686 sr->l_whence = 0;
1687
1688 llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len;
1689
1690 if (sr->l_start < 0
1691 || sr->l_start > max_off
1692 || (sr->l_start + llen) < 0
1693 || (sr->l_start + llen) > max_off) {
1694 ret = -EINVAL;
Mark Fashehe63aecb62007-10-18 15:30:42 -07001695 goto out_inode_unlock;
Mark Fashehb2580102007-03-09 16:53:21 -08001696 }
Mark Fasheh385820a2007-07-19 00:14:38 -07001697 size = sr->l_start + sr->l_len;
Mark Fashehb2580102007-03-09 16:53:21 -08001698
1699 if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) {
1700 if (sr->l_len <= 0) {
1701 ret = -EINVAL;
Mark Fashehe63aecb62007-10-18 15:30:42 -07001702 goto out_inode_unlock;
Mark Fashehb2580102007-03-09 16:53:21 -08001703 }
1704 }
1705
Mark Fasheh385820a2007-07-19 00:14:38 -07001706 if (file && should_remove_suid(file->f_path.dentry)) {
Mark Fashehb2580102007-03-09 16:53:21 -08001707 ret = __ocfs2_write_remove_suid(inode, di_bh);
1708 if (ret) {
1709 mlog_errno(ret);
Mark Fashehe63aecb62007-10-18 15:30:42 -07001710 goto out_inode_unlock;
Mark Fashehb2580102007-03-09 16:53:21 -08001711 }
1712 }
1713
1714 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1715 switch (cmd) {
1716 case OCFS2_IOC_RESVSP:
1717 case OCFS2_IOC_RESVSP64:
1718 /*
1719 * This takes unsigned offsets, but the signed ones we
1720 * pass have been checked against overflow above.
1721 */
1722 ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start,
1723 sr->l_len);
1724 break;
1725 case OCFS2_IOC_UNRESVSP:
1726 case OCFS2_IOC_UNRESVSP64:
1727 ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start,
1728 sr->l_len);
1729 break;
1730 default:
1731 ret = -EINVAL;
1732 }
1733 up_write(&OCFS2_I(inode)->ip_alloc_sem);
1734 if (ret) {
1735 mlog_errno(ret);
Mark Fashehe63aecb62007-10-18 15:30:42 -07001736 goto out_inode_unlock;
Mark Fashehb2580102007-03-09 16:53:21 -08001737 }
1738
1739 /*
1740 * We update c/mtime for these changes
1741 */
1742 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1743 if (IS_ERR(handle)) {
1744 ret = PTR_ERR(handle);
1745 mlog_errno(ret);
Mark Fashehe63aecb62007-10-18 15:30:42 -07001746 goto out_inode_unlock;
Mark Fashehb2580102007-03-09 16:53:21 -08001747 }
1748
Mark Fasheh385820a2007-07-19 00:14:38 -07001749 if (change_size && i_size_read(inode) < size)
1750 i_size_write(inode, size);
1751
Mark Fashehb2580102007-03-09 16:53:21 -08001752 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1753 ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
1754 if (ret < 0)
1755 mlog_errno(ret);
1756
1757 ocfs2_commit_trans(osb, handle);
1758
Mark Fashehe63aecb62007-10-18 15:30:42 -07001759out_inode_unlock:
Mark Fashehb2580102007-03-09 16:53:21 -08001760 brelse(di_bh);
Mark Fashehe63aecb62007-10-18 15:30:42 -07001761 ocfs2_inode_unlock(inode, 1);
Mark Fashehb2580102007-03-09 16:53:21 -08001762out_rw_unlock:
1763 ocfs2_rw_unlock(inode, 1);
1764
1765 mutex_unlock(&inode->i_mutex);
1766out:
1767 return ret;
1768}
1769
Mark Fasheh385820a2007-07-19 00:14:38 -07001770int ocfs2_change_file_space(struct file *file, unsigned int cmd,
1771 struct ocfs2_space_resv *sr)
1772{
1773 struct inode *inode = file->f_path.dentry->d_inode;
1774 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);;
1775
1776 if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) &&
1777 !ocfs2_writes_unwritten_extents(osb))
1778 return -ENOTTY;
1779 else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) &&
1780 !ocfs2_sparse_alloc(osb))
1781 return -ENOTTY;
1782
1783 if (!S_ISREG(inode->i_mode))
1784 return -EINVAL;
1785
1786 if (!(file->f_mode & FMODE_WRITE))
1787 return -EBADF;
1788
1789 return __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0);
1790}
1791
1792static long ocfs2_fallocate(struct inode *inode, int mode, loff_t offset,
1793 loff_t len)
1794{
1795 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1796 struct ocfs2_space_resv sr;
1797 int change_size = 1;
1798
1799 if (!ocfs2_writes_unwritten_extents(osb))
1800 return -EOPNOTSUPP;
1801
1802 if (S_ISDIR(inode->i_mode))
1803 return -ENODEV;
1804
1805 if (mode & FALLOC_FL_KEEP_SIZE)
1806 change_size = 0;
1807
1808 sr.l_whence = 0;
1809 sr.l_start = (s64)offset;
1810 sr.l_len = (s64)len;
1811
1812 return __ocfs2_change_file_space(NULL, inode, offset,
1813 OCFS2_IOC_RESVSP64, &sr, change_size);
1814}
1815
Tiger Yang8659ac22006-10-17 18:29:52 -07001816static int ocfs2_prepare_inode_for_write(struct dentry *dentry,
1817 loff_t *ppos,
1818 size_t count,
Mark Fasheh9517bac2007-02-09 20:24:12 -08001819 int appending,
1820 int *direct_io)
Mark Fashehccd979b2005-12-15 14:31:24 -08001821{
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001822 int ret = 0, meta_level = 0;
Tiger Yang8659ac22006-10-17 18:29:52 -07001823 struct inode *inode = dentry->d_inode;
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001824 loff_t saved_pos, end;
Mark Fashehccd979b2005-12-15 14:31:24 -08001825
Mark Fashehccd979b2005-12-15 14:31:24 -08001826 /*
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001827 * We start with a read level meta lock and only jump to an ex
1828 * if we need to make modifications here.
Mark Fashehccd979b2005-12-15 14:31:24 -08001829 */
Mark Fashehccd979b2005-12-15 14:31:24 -08001830 for(;;) {
Mark Fashehe63aecb62007-10-18 15:30:42 -07001831 ret = ocfs2_inode_lock(inode, NULL, meta_level);
Mark Fashehccd979b2005-12-15 14:31:24 -08001832 if (ret < 0) {
1833 meta_level = -1;
1834 mlog_errno(ret);
1835 goto out;
1836 }
1837
1838 /* Clear suid / sgid if necessary. We do this here
1839 * instead of later in the write path because
1840 * remove_suid() calls ->setattr without any hint that
1841 * we may have already done our cluster locking. Since
1842 * ocfs2_setattr() *must* take cluster locks to
1843 * proceeed, this will lead us to recursively lock the
1844 * inode. There's also the dinode i_size state which
1845 * can be lost via setattr during extending writes (we
1846 * set inode->i_size at the end of a write. */
Tiger Yang8659ac22006-10-17 18:29:52 -07001847 if (should_remove_suid(dentry)) {
Mark Fashehccd979b2005-12-15 14:31:24 -08001848 if (meta_level == 0) {
Mark Fashehe63aecb62007-10-18 15:30:42 -07001849 ocfs2_inode_unlock(inode, meta_level);
Mark Fashehccd979b2005-12-15 14:31:24 -08001850 meta_level = 1;
1851 continue;
1852 }
1853
1854 ret = ocfs2_write_remove_suid(inode);
1855 if (ret < 0) {
1856 mlog_errno(ret);
Tiger Yang8659ac22006-10-17 18:29:52 -07001857 goto out_unlock;
Mark Fashehccd979b2005-12-15 14:31:24 -08001858 }
1859 }
1860
1861 /* work on a copy of ppos until we're sure that we won't have
1862 * to recalculate it due to relocking. */
Tiger Yang8659ac22006-10-17 18:29:52 -07001863 if (appending) {
Mark Fashehccd979b2005-12-15 14:31:24 -08001864 saved_pos = i_size_read(inode);
1865 mlog(0, "O_APPEND: inode->i_size=%llu\n", saved_pos);
1866 } else {
Tiger Yang8659ac22006-10-17 18:29:52 -07001867 saved_pos = *ppos;
Mark Fashehccd979b2005-12-15 14:31:24 -08001868 }
Mark Fasheh3a0782d2007-01-17 12:53:31 -08001869
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001870 end = saved_pos + count;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001871
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001872 /*
1873 * Skip the O_DIRECT checks if we don't need
1874 * them.
1875 */
1876 if (!direct_io || !(*direct_io))
1877 break;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001878
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001879 /*
Mark Fasheh1afc32b2007-09-07 14:46:51 -07001880 * There's no sane way to do direct writes to an inode
1881 * with inline data.
1882 */
1883 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1884 *direct_io = 0;
1885 break;
1886 }
1887
1888 /*
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001889 * Allowing concurrent direct writes means
1890 * i_size changes wouldn't be synchronized, so
1891 * one node could wind up truncating another
1892 * nodes writes.
1893 */
1894 if (end > i_size_read(inode)) {
1895 *direct_io = 0;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001896 break;
1897 }
1898
Mark Fasheh3a0782d2007-01-17 12:53:31 -08001899 /*
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001900 * We don't fill holes during direct io, so
1901 * check for them here. If any are found, the
1902 * caller will have to retake some cluster
1903 * locks and initiate the io as buffered.
Mark Fasheh3a0782d2007-01-17 12:53:31 -08001904 */
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001905 ret = ocfs2_check_range_for_holes(inode, saved_pos, count);
1906 if (ret == 1) {
1907 *direct_io = 0;
1908 ret = 0;
1909 } else if (ret < 0)
1910 mlog_errno(ret);
Mark Fashehccd979b2005-12-15 14:31:24 -08001911 break;
1912 }
1913
Tiger Yang8659ac22006-10-17 18:29:52 -07001914 if (appending)
1915 *ppos = saved_pos;
1916
1917out_unlock:
Mark Fashehe63aecb62007-10-18 15:30:42 -07001918 ocfs2_inode_unlock(inode, meta_level);
Tiger Yang8659ac22006-10-17 18:29:52 -07001919
1920out:
1921 return ret;
1922}
1923
1924static ssize_t ocfs2_file_aio_write(struct kiocb *iocb,
1925 const struct iovec *iov,
1926 unsigned long nr_segs,
1927 loff_t pos)
1928{
Mark Fasheh9517bac2007-02-09 20:24:12 -08001929 int ret, direct_io, appending, rw_level, have_alloc_sem = 0;
Nick Pigginb6af1bc2007-10-16 01:25:24 -07001930 int can_do_direct;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001931 ssize_t written = 0;
1932 size_t ocount; /* original count */
1933 size_t count; /* after file limit checks */
Mark Fasheh9ea2d322007-10-18 14:14:45 -07001934 loff_t old_size, *ppos = &iocb->ki_pos;
1935 u32 old_clusters;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001936 struct file *file = iocb->ki_filp;
1937 struct inode *inode = file->f_path.dentry->d_inode;
Mark Fasheh9ea2d322007-10-18 14:14:45 -07001938 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
Tiger Yang8659ac22006-10-17 18:29:52 -07001939
Mark Fasheh9517bac2007-02-09 20:24:12 -08001940 mlog_entry("(0x%p, %u, '%.*s')\n", file,
Tiger Yang8659ac22006-10-17 18:29:52 -07001941 (unsigned int)nr_segs,
Mark Fasheh9517bac2007-02-09 20:24:12 -08001942 file->f_path.dentry->d_name.len,
1943 file->f_path.dentry->d_name.name);
Tiger Yang8659ac22006-10-17 18:29:52 -07001944
Tiger Yang8659ac22006-10-17 18:29:52 -07001945 if (iocb->ki_left == 0)
1946 return 0;
1947
Mark Fasheh9517bac2007-02-09 20:24:12 -08001948 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
1949
1950 appending = file->f_flags & O_APPEND ? 1 : 0;
1951 direct_io = file->f_flags & O_DIRECT ? 1 : 0;
1952
Tiger Yang8659ac22006-10-17 18:29:52 -07001953 mutex_lock(&inode->i_mutex);
Mark Fasheh9517bac2007-02-09 20:24:12 -08001954
1955relock:
Tiger Yang8659ac22006-10-17 18:29:52 -07001956 /* to match setattr's i_mutex -> i_alloc_sem -> rw_lock ordering */
Mark Fasheh9517bac2007-02-09 20:24:12 -08001957 if (direct_io) {
Tiger Yang8659ac22006-10-17 18:29:52 -07001958 down_read(&inode->i_alloc_sem);
Mark Fasheh9517bac2007-02-09 20:24:12 -08001959 have_alloc_sem = 1;
Tiger Yang8659ac22006-10-17 18:29:52 -07001960 }
1961
1962 /* concurrent O_DIRECT writes are allowed */
Mark Fasheh9517bac2007-02-09 20:24:12 -08001963 rw_level = !direct_io;
Tiger Yang8659ac22006-10-17 18:29:52 -07001964 ret = ocfs2_rw_lock(inode, rw_level);
1965 if (ret < 0) {
Mark Fasheh9517bac2007-02-09 20:24:12 -08001966 mlog_errno(ret);
1967 goto out_sems;
1968 }
1969
1970 can_do_direct = direct_io;
1971 ret = ocfs2_prepare_inode_for_write(file->f_path.dentry, ppos,
1972 iocb->ki_left, appending,
1973 &can_do_direct);
1974 if (ret < 0) {
Tiger Yang8659ac22006-10-17 18:29:52 -07001975 mlog_errno(ret);
1976 goto out;
1977 }
1978
Mark Fasheh9517bac2007-02-09 20:24:12 -08001979 /*
1980 * We can't complete the direct I/O as requested, fall back to
1981 * buffered I/O.
1982 */
1983 if (direct_io && !can_do_direct) {
1984 ocfs2_rw_unlock(inode, rw_level);
1985 up_read(&inode->i_alloc_sem);
1986
1987 have_alloc_sem = 0;
1988 rw_level = -1;
1989
1990 direct_io = 0;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001991 goto relock;
Tiger Yang8659ac22006-10-17 18:29:52 -07001992 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001993
Mark Fasheh9ea2d322007-10-18 14:14:45 -07001994 /*
1995 * To later detect whether a journal commit for sync writes is
1996 * necessary, we sample i_size, and cluster count here.
1997 */
1998 old_size = i_size_read(inode);
1999 old_clusters = OCFS2_I(inode)->ip_clusters;
2000
Mark Fashehccd979b2005-12-15 14:31:24 -08002001 /* communicate with ocfs2_dio_end_io */
Mark Fasheh7cdfc3a2007-04-16 17:28:51 -07002002 ocfs2_iocb_set_rw_locked(iocb, rw_level);
Mark Fashehccd979b2005-12-15 14:31:24 -08002003
Mark Fasheh9517bac2007-02-09 20:24:12 -08002004 if (direct_io) {
Nick Pigginb6af1bc2007-10-16 01:25:24 -07002005 ret = generic_segment_checks(iov, &nr_segs, &ocount,
2006 VERIFY_READ);
2007 if (ret)
2008 goto out_dio;
2009
2010 ret = generic_write_checks(file, ppos, &count,
2011 S_ISBLK(inode->i_mode));
2012 if (ret)
2013 goto out_dio;
2014
Mark Fasheh9517bac2007-02-09 20:24:12 -08002015 written = generic_file_direct_write(iocb, iov, &nr_segs, *ppos,
2016 ppos, count, ocount);
2017 if (written < 0) {
2018 ret = written;
2019 goto out_dio;
2020 }
2021 } else {
Nick Pigginb6af1bc2007-10-16 01:25:24 -07002022 written = generic_file_aio_write_nolock(iocb, iov, nr_segs,
2023 *ppos);
Mark Fasheh9517bac2007-02-09 20:24:12 -08002024 }
Mark Fashehccd979b2005-12-15 14:31:24 -08002025
Mark Fasheh9517bac2007-02-09 20:24:12 -08002026out_dio:
Mark Fashehccd979b2005-12-15 14:31:24 -08002027 /* buffered aio wouldn't have proper lock coverage today */
Mark Fasheh9517bac2007-02-09 20:24:12 -08002028 BUG_ON(ret == -EIOCBQUEUED && !(file->f_flags & O_DIRECT));
Mark Fashehccd979b2005-12-15 14:31:24 -08002029
Mark Fasheh9ea2d322007-10-18 14:14:45 -07002030 if ((file->f_flags & O_SYNC && !direct_io) || IS_SYNC(inode)) {
2031 /*
2032 * The generic write paths have handled getting data
2033 * to disk, but since we don't make use of the dirty
2034 * inode list, a manual journal commit is necessary
2035 * here.
2036 */
2037 if (old_size != i_size_read(inode) ||
2038 old_clusters != OCFS2_I(inode)->ip_clusters) {
2039 ret = journal_force_commit(osb->journal->j_journal);
2040 if (ret < 0)
2041 written = ret;
2042 }
2043 }
2044
Mark Fashehccd979b2005-12-15 14:31:24 -08002045 /*
2046 * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
2047 * function pointer which is called when o_direct io completes so that
2048 * it can unlock our rw lock. (it's the clustered equivalent of
2049 * i_alloc_sem; protects truncate from racing with pending ios).
2050 * Unfortunately there are error cases which call end_io and others
2051 * that don't. so we don't have to unlock the rw_lock if either an
2052 * async dio is going to do it in the future or an end_io after an
2053 * error has already done it.
2054 */
2055 if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2056 rw_level = -1;
2057 have_alloc_sem = 0;
2058 }
2059
2060out:
Mark Fasheh9517bac2007-02-09 20:24:12 -08002061 if (rw_level != -1)
2062 ocfs2_rw_unlock(inode, rw_level);
2063
2064out_sems:
Mark Fashehccd979b2005-12-15 14:31:24 -08002065 if (have_alloc_sem)
2066 up_read(&inode->i_alloc_sem);
Mark Fasheh9517bac2007-02-09 20:24:12 -08002067
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002068 mutex_unlock(&inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -08002069
2070 mlog_exit(ret);
Mark Fasheh9517bac2007-02-09 20:24:12 -08002071 return written ? written : ret;
Mark Fashehccd979b2005-12-15 14:31:24 -08002072}
2073
Tiger Yang8659ac22006-10-17 18:29:52 -07002074static ssize_t ocfs2_file_splice_write(struct pipe_inode_info *pipe,
2075 struct file *out,
2076 loff_t *ppos,
2077 size_t len,
2078 unsigned int flags)
2079{
2080 int ret;
Josef Sipekd28c9172006-12-08 02:37:25 -08002081 struct inode *inode = out->f_path.dentry->d_inode;
Tiger Yang8659ac22006-10-17 18:29:52 -07002082
2083 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", out, pipe,
2084 (unsigned int)len,
Josef Sipekd28c9172006-12-08 02:37:25 -08002085 out->f_path.dentry->d_name.len,
2086 out->f_path.dentry->d_name.name);
Tiger Yang8659ac22006-10-17 18:29:52 -07002087
2088 inode_double_lock(inode, pipe->inode);
2089
2090 ret = ocfs2_rw_lock(inode, 1);
2091 if (ret < 0) {
2092 mlog_errno(ret);
2093 goto out;
2094 }
2095
Mark Fasheh9517bac2007-02-09 20:24:12 -08002096 ret = ocfs2_prepare_inode_for_write(out->f_path.dentry, ppos, len, 0,
2097 NULL);
Tiger Yang8659ac22006-10-17 18:29:52 -07002098 if (ret < 0) {
2099 mlog_errno(ret);
2100 goto out_unlock;
2101 }
2102
Nick Pigginb6af1bc2007-10-16 01:25:24 -07002103 ret = generic_file_splice_write_nolock(pipe, out, ppos, len, flags);
Tiger Yang8659ac22006-10-17 18:29:52 -07002104
2105out_unlock:
2106 ocfs2_rw_unlock(inode, 1);
2107out:
2108 inode_double_unlock(inode, pipe->inode);
2109
2110 mlog_exit(ret);
2111 return ret;
2112}
2113
2114static ssize_t ocfs2_file_splice_read(struct file *in,
2115 loff_t *ppos,
2116 struct pipe_inode_info *pipe,
2117 size_t len,
2118 unsigned int flags)
2119{
2120 int ret = 0;
Josef Sipekd28c9172006-12-08 02:37:25 -08002121 struct inode *inode = in->f_path.dentry->d_inode;
Tiger Yang8659ac22006-10-17 18:29:52 -07002122
2123 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", in, pipe,
2124 (unsigned int)len,
Josef Sipekd28c9172006-12-08 02:37:25 -08002125 in->f_path.dentry->d_name.len,
2126 in->f_path.dentry->d_name.name);
Tiger Yang8659ac22006-10-17 18:29:52 -07002127
2128 /*
2129 * See the comment in ocfs2_file_aio_read()
2130 */
Mark Fashehe63aecb62007-10-18 15:30:42 -07002131 ret = ocfs2_inode_lock(inode, NULL, 0);
Tiger Yang8659ac22006-10-17 18:29:52 -07002132 if (ret < 0) {
2133 mlog_errno(ret);
2134 goto bail;
2135 }
Mark Fashehe63aecb62007-10-18 15:30:42 -07002136 ocfs2_inode_unlock(inode, 0);
Tiger Yang8659ac22006-10-17 18:29:52 -07002137
2138 ret = generic_file_splice_read(in, ppos, pipe, len, flags);
2139
2140bail:
2141 mlog_exit(ret);
2142 return ret;
2143}
2144
Mark Fashehccd979b2005-12-15 14:31:24 -08002145static ssize_t ocfs2_file_aio_read(struct kiocb *iocb,
Badari Pulavarty027445c2006-09-30 23:28:46 -07002146 const struct iovec *iov,
2147 unsigned long nr_segs,
Mark Fashehccd979b2005-12-15 14:31:24 -08002148 loff_t pos)
2149{
Tiger Yang25899de2006-11-15 15:49:02 +08002150 int ret = 0, rw_level = -1, have_alloc_sem = 0, lock_level = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -08002151 struct file *filp = iocb->ki_filp;
Josef Sipekd28c9172006-12-08 02:37:25 -08002152 struct inode *inode = filp->f_path.dentry->d_inode;
Mark Fashehccd979b2005-12-15 14:31:24 -08002153
Badari Pulavarty027445c2006-09-30 23:28:46 -07002154 mlog_entry("(0x%p, %u, '%.*s')\n", filp,
2155 (unsigned int)nr_segs,
Josef Sipekd28c9172006-12-08 02:37:25 -08002156 filp->f_path.dentry->d_name.len,
2157 filp->f_path.dentry->d_name.name);
Mark Fashehccd979b2005-12-15 14:31:24 -08002158
2159 if (!inode) {
2160 ret = -EINVAL;
2161 mlog_errno(ret);
2162 goto bail;
2163 }
2164
Mark Fashehccd979b2005-12-15 14:31:24 -08002165 /*
2166 * buffered reads protect themselves in ->readpage(). O_DIRECT reads
2167 * need locks to protect pending reads from racing with truncate.
2168 */
2169 if (filp->f_flags & O_DIRECT) {
2170 down_read(&inode->i_alloc_sem);
2171 have_alloc_sem = 1;
2172
2173 ret = ocfs2_rw_lock(inode, 0);
2174 if (ret < 0) {
2175 mlog_errno(ret);
2176 goto bail;
2177 }
2178 rw_level = 0;
2179 /* communicate with ocfs2_dio_end_io */
Mark Fasheh7cdfc3a2007-04-16 17:28:51 -07002180 ocfs2_iocb_set_rw_locked(iocb, rw_level);
Mark Fashehccd979b2005-12-15 14:31:24 -08002181 }
2182
Mark Fashehc4374f82006-05-05 19:04:35 -07002183 /*
2184 * We're fine letting folks race truncates and extending
2185 * writes with read across the cluster, just like they can
2186 * locally. Hence no rw_lock during read.
2187 *
2188 * Take and drop the meta data lock to update inode fields
2189 * like i_size. This allows the checks down below
2190 * generic_file_aio_read() a chance of actually working.
2191 */
Mark Fashehe63aecb62007-10-18 15:30:42 -07002192 ret = ocfs2_inode_lock_atime(inode, filp->f_vfsmnt, &lock_level);
Mark Fashehc4374f82006-05-05 19:04:35 -07002193 if (ret < 0) {
2194 mlog_errno(ret);
2195 goto bail;
2196 }
Mark Fashehe63aecb62007-10-18 15:30:42 -07002197 ocfs2_inode_unlock(inode, lock_level);
Mark Fashehc4374f82006-05-05 19:04:35 -07002198
Badari Pulavarty027445c2006-09-30 23:28:46 -07002199 ret = generic_file_aio_read(iocb, iov, nr_segs, iocb->ki_pos);
Mark Fashehccd979b2005-12-15 14:31:24 -08002200 if (ret == -EINVAL)
2201 mlog(ML_ERROR, "generic_file_aio_read returned -EINVAL\n");
2202
2203 /* buffered aio wouldn't have proper lock coverage today */
2204 BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT));
2205
2206 /* see ocfs2_file_aio_write */
2207 if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2208 rw_level = -1;
2209 have_alloc_sem = 0;
2210 }
2211
2212bail:
2213 if (have_alloc_sem)
2214 up_read(&inode->i_alloc_sem);
2215 if (rw_level != -1)
2216 ocfs2_rw_unlock(inode, rw_level);
2217 mlog_exit(ret);
2218
2219 return ret;
2220}
2221
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002222const struct inode_operations ocfs2_file_iops = {
Mark Fashehccd979b2005-12-15 14:31:24 -08002223 .setattr = ocfs2_setattr,
2224 .getattr = ocfs2_getattr,
Tiger Yangd38eb8d2006-11-27 09:59:21 +08002225 .permission = ocfs2_permission,
Mark Fasheh385820a2007-07-19 00:14:38 -07002226 .fallocate = ocfs2_fallocate,
Mark Fashehccd979b2005-12-15 14:31:24 -08002227};
2228
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002229const struct inode_operations ocfs2_special_file_iops = {
Mark Fashehccd979b2005-12-15 14:31:24 -08002230 .setattr = ocfs2_setattr,
2231 .getattr = ocfs2_getattr,
Tiger Yangd38eb8d2006-11-27 09:59:21 +08002232 .permission = ocfs2_permission,
Mark Fashehccd979b2005-12-15 14:31:24 -08002233};
2234
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08002235const struct file_operations ocfs2_fops = {
Jan Kara32c3c0e2007-12-19 15:24:52 +01002236 .llseek = generic_file_llseek,
Mark Fashehccd979b2005-12-15 14:31:24 -08002237 .read = do_sync_read,
2238 .write = do_sync_write,
Mark Fashehccd979b2005-12-15 14:31:24 -08002239 .mmap = ocfs2_mmap,
2240 .fsync = ocfs2_sync_file,
2241 .release = ocfs2_file_release,
2242 .open = ocfs2_file_open,
2243 .aio_read = ocfs2_file_aio_read,
2244 .aio_write = ocfs2_file_aio_write,
Andi Kleenc9ec1482008-01-27 03:17:17 +01002245 .unlocked_ioctl = ocfs2_ioctl,
Mark Fasheh586d2322007-03-09 15:56:28 -08002246#ifdef CONFIG_COMPAT
2247 .compat_ioctl = ocfs2_compat_ioctl,
2248#endif
Mark Fasheh53fc6222007-12-20 16:49:04 -08002249 .flock = ocfs2_flock,
Tiger Yang8659ac22006-10-17 18:29:52 -07002250 .splice_read = ocfs2_file_splice_read,
2251 .splice_write = ocfs2_file_splice_write,
Mark Fashehccd979b2005-12-15 14:31:24 -08002252};
2253
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08002254const struct file_operations ocfs2_dops = {
Jan Kara32c3c0e2007-12-19 15:24:52 +01002255 .llseek = generic_file_llseek,
Mark Fashehccd979b2005-12-15 14:31:24 -08002256 .read = generic_read_dir,
2257 .readdir = ocfs2_readdir,
2258 .fsync = ocfs2_sync_file,
Mark Fasheh53fc6222007-12-20 16:49:04 -08002259 .release = ocfs2_dir_release,
2260 .open = ocfs2_dir_open,
Andi Kleenc9ec1482008-01-27 03:17:17 +01002261 .unlocked_ioctl = ocfs2_ioctl,
Mark Fasheh586d2322007-03-09 15:56:28 -08002262#ifdef CONFIG_COMPAT
2263 .compat_ioctl = ocfs2_compat_ioctl,
2264#endif
Mark Fasheh53fc6222007-12-20 16:49:04 -08002265 .flock = ocfs2_flock,
Mark Fashehccd979b2005-12-15 14:31:24 -08002266};