blob: 781ba6c4ef852e88c058d7b3e355c07f6798c172 [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"
54#include "mmap.h"
55#include "suballoc.h"
56#include "super.h"
57
58#include "buffer_head_io.h"
59
60static int ocfs2_sync_inode(struct inode *inode)
61{
62 filemap_fdatawrite(inode->i_mapping);
63 return sync_mapping_buffers(inode->i_mapping);
64}
65
66static int ocfs2_file_open(struct inode *inode, struct file *file)
67{
68 int status;
69 int mode = file->f_flags;
70 struct ocfs2_inode_info *oi = OCFS2_I(inode);
71
72 mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file,
Josef Sipekd28c9172006-12-08 02:37:25 -080073 file->f_path.dentry->d_name.len, file->f_path.dentry->d_name.name);
Mark Fashehccd979b2005-12-15 14:31:24 -080074
75 spin_lock(&oi->ip_lock);
76
77 /* Check that the inode hasn't been wiped from disk by another
78 * node. If it hasn't then we're safe as long as we hold the
79 * spin lock until our increment of open count. */
80 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
81 spin_unlock(&oi->ip_lock);
82
83 status = -ENOENT;
84 goto leave;
85 }
86
87 if (mode & O_DIRECT)
88 oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT;
89
90 oi->ip_open_count++;
91 spin_unlock(&oi->ip_lock);
92 status = 0;
93leave:
94 mlog_exit(status);
95 return status;
96}
97
98static int ocfs2_file_release(struct inode *inode, struct file *file)
99{
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,
104 file->f_path.dentry->d_name.name);
Mark Fashehccd979b2005-12-15 14:31:24 -0800105
106 spin_lock(&oi->ip_lock);
107 if (!--oi->ip_open_count)
108 oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT;
109 spin_unlock(&oi->ip_lock);
110
111 mlog_exit(0);
112
113 return 0;
114}
115
116static int ocfs2_sync_file(struct file *file,
117 struct dentry *dentry,
118 int datasync)
119{
120 int err = 0;
121 journal_t *journal;
122 struct inode *inode = dentry->d_inode;
123 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
124
125 mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", file, dentry, datasync,
126 dentry->d_name.len, dentry->d_name.name);
127
128 err = ocfs2_sync_inode(dentry->d_inode);
129 if (err)
130 goto bail;
131
132 journal = osb->journal->j_journal;
133 err = journal_force_commit(journal);
134
135bail:
136 mlog_exit(err);
137
138 return (err < 0) ? -EIO : 0;
139}
140
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800141int ocfs2_should_update_atime(struct inode *inode,
142 struct vfsmount *vfsmnt)
143{
144 struct timespec now;
145 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
146
147 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
148 return 0;
149
150 if ((inode->i_flags & S_NOATIME) ||
151 ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)))
152 return 0;
153
Mark Fasheh6c2aad02006-12-19 15:25:52 -0800154 /*
155 * We can be called with no vfsmnt structure - NFSD will
156 * sometimes do this.
157 *
158 * Note that our action here is different than touch_atime() -
159 * if we can't tell whether this is a noatime mount, then we
160 * don't know whether to trust the value of s_atime_quantum.
161 */
162 if (vfsmnt == NULL)
163 return 0;
164
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800165 if ((vfsmnt->mnt_flags & MNT_NOATIME) ||
166 ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
167 return 0;
168
Mark Fasheh7e913c52006-12-13 00:34:35 -0800169 if (vfsmnt->mnt_flags & MNT_RELATIME) {
170 if ((timespec_compare(&inode->i_atime, &inode->i_mtime) <= 0) ||
171 (timespec_compare(&inode->i_atime, &inode->i_ctime) <= 0))
172 return 1;
173
174 return 0;
175 }
176
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800177 now = CURRENT_TIME;
178 if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum))
179 return 0;
180 else
181 return 1;
182}
183
184int ocfs2_update_inode_atime(struct inode *inode,
185 struct buffer_head *bh)
186{
187 int ret;
188 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
189 handle_t *handle;
Mark Fashehc11e9fa2007-07-20 11:24:53 -0700190 struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data;
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800191
192 mlog_entry_void();
193
194 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
195 if (handle == NULL) {
196 ret = -ENOMEM;
197 mlog_errno(ret);
198 goto out;
199 }
200
Mark Fashehc11e9fa2007-07-20 11:24:53 -0700201 ret = ocfs2_journal_access(handle, inode, bh,
202 OCFS2_JOURNAL_ACCESS_WRITE);
203 if (ret) {
204 mlog_errno(ret);
205 goto out_commit;
206 }
207
208 /*
209 * Don't use ocfs2_mark_inode_dirty() here as we don't always
210 * have i_mutex to guard against concurrent changes to other
211 * inode fields.
212 */
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800213 inode->i_atime = CURRENT_TIME;
Mark Fashehc11e9fa2007-07-20 11:24:53 -0700214 di->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
215 di->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
216
217 ret = ocfs2_journal_dirty(handle, bh);
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800218 if (ret < 0)
219 mlog_errno(ret);
220
Mark Fashehc11e9fa2007-07-20 11:24:53 -0700221out_commit:
Tiger Yang7f1a37e2006-11-15 15:48:42 +0800222 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
223out:
224 mlog_exit(ret);
225 return ret;
226}
227
Adrian Bunk6cb129f2007-04-26 00:29:35 -0700228static int ocfs2_set_inode_size(handle_t *handle,
229 struct inode *inode,
230 struct buffer_head *fe_bh,
231 u64 new_i_size)
Mark Fashehccd979b2005-12-15 14:31:24 -0800232{
233 int status;
234
235 mlog_entry_void();
236 i_size_write(inode, new_i_size);
Mark Fasheh8110b072007-03-22 16:53:23 -0700237 inode->i_blocks = ocfs2_inode_sector_count(inode);
Mark Fashehccd979b2005-12-15 14:31:24 -0800238 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
239
240 status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
241 if (status < 0) {
242 mlog_errno(status);
243 goto bail;
244 }
245
246bail:
247 mlog_exit(status);
248 return status;
249}
250
251static int ocfs2_simple_size_update(struct inode *inode,
252 struct buffer_head *di_bh,
253 u64 new_i_size)
254{
255 int ret;
256 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
Mark Fasheh1fabe142006-10-09 18:11:45 -0700257 handle_t *handle = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800258
Mark Fasheh65eff9c2006-10-09 17:26:22 -0700259 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
Mark Fashehccd979b2005-12-15 14:31:24 -0800260 if (handle == NULL) {
261 ret = -ENOMEM;
262 mlog_errno(ret);
263 goto out;
264 }
265
266 ret = ocfs2_set_inode_size(handle, inode, di_bh,
267 new_i_size);
268 if (ret < 0)
269 mlog_errno(ret);
270
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700271 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800272out:
273 return ret;
274}
275
276static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
277 struct inode *inode,
278 struct buffer_head *fe_bh,
279 u64 new_i_size)
280{
281 int status;
Mark Fasheh1fabe142006-10-09 18:11:45 -0700282 handle_t *handle;
Mark Fasheh60b11392007-02-16 11:46:50 -0800283 struct ocfs2_dinode *di;
Mark Fasheh35edec12007-07-06 14:41:18 -0700284 u64 cluster_bytes;
Mark Fashehccd979b2005-12-15 14:31:24 -0800285
286 mlog_entry_void();
287
288 /* TODO: This needs to actually orphan the inode in this
289 * transaction. */
290
Mark Fasheh65eff9c2006-10-09 17:26:22 -0700291 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
Mark Fashehccd979b2005-12-15 14:31:24 -0800292 if (IS_ERR(handle)) {
293 status = PTR_ERR(handle);
294 mlog_errno(status);
295 goto out;
296 }
297
Mark Fasheh60b11392007-02-16 11:46:50 -0800298 status = ocfs2_journal_access(handle, inode, fe_bh,
299 OCFS2_JOURNAL_ACCESS_WRITE);
300 if (status < 0) {
301 mlog_errno(status);
302 goto out_commit;
303 }
304
305 /*
306 * Do this before setting i_size.
307 */
Mark Fasheh35edec12007-07-06 14:41:18 -0700308 cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size);
309 status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size,
310 cluster_bytes);
Mark Fasheh60b11392007-02-16 11:46:50 -0800311 if (status) {
312 mlog_errno(status);
313 goto out_commit;
314 }
315
316 i_size_write(inode, new_i_size);
Mark Fasheh60b11392007-02-16 11:46:50 -0800317 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
318
319 di = (struct ocfs2_dinode *) fe_bh->b_data;
320 di->i_size = cpu_to_le64(new_i_size);
321 di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
322 di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
323
324 status = ocfs2_journal_dirty(handle, fe_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800325 if (status < 0)
326 mlog_errno(status);
327
Mark Fasheh60b11392007-02-16 11:46:50 -0800328out_commit:
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700329 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800330out:
Mark Fasheh60b11392007-02-16 11:46:50 -0800331
Mark Fashehccd979b2005-12-15 14:31:24 -0800332 mlog_exit(status);
333 return status;
334}
335
336static int ocfs2_truncate_file(struct inode *inode,
337 struct buffer_head *di_bh,
338 u64 new_i_size)
339{
340 int status = 0;
341 struct ocfs2_dinode *fe = NULL;
342 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
343 struct ocfs2_truncate_context *tc = NULL;
344
Mark Fashehb06970532006-03-03 10:24:33 -0800345 mlog_entry("(inode = %llu, new_i_size = %llu\n",
346 (unsigned long long)OCFS2_I(inode)->ip_blkno,
347 (unsigned long long)new_i_size);
Mark Fashehccd979b2005-12-15 14:31:24 -0800348
Mark Fashehccd979b2005-12-15 14:31:24 -0800349 fe = (struct ocfs2_dinode *) di_bh->b_data;
350 if (!OCFS2_IS_VALID_DINODE(fe)) {
351 OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe);
352 status = -EIO;
353 goto bail;
354 }
355
356 mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode),
Mark Fashehb06970532006-03-03 10:24:33 -0800357 "Inode %llu, inode i_size = %lld != di "
358 "i_size = %llu, i_flags = 0x%x\n",
359 (unsigned long long)OCFS2_I(inode)->ip_blkno,
Mark Fashehccd979b2005-12-15 14:31:24 -0800360 i_size_read(inode),
Mark Fashehb06970532006-03-03 10:24:33 -0800361 (unsigned long long)le64_to_cpu(fe->i_size),
362 le32_to_cpu(fe->i_flags));
Mark Fashehccd979b2005-12-15 14:31:24 -0800363
364 if (new_i_size > le64_to_cpu(fe->i_size)) {
Mark Fashehb06970532006-03-03 10:24:33 -0800365 mlog(0, "asked to truncate file with size (%llu) to size (%llu)!\n",
366 (unsigned long long)le64_to_cpu(fe->i_size),
367 (unsigned long long)new_i_size);
Mark Fashehccd979b2005-12-15 14:31:24 -0800368 status = -EINVAL;
369 mlog_errno(status);
370 goto bail;
371 }
372
Mark Fashehb06970532006-03-03 10:24:33 -0800373 mlog(0, "inode %llu, i_size = %llu, new_i_size = %llu\n",
374 (unsigned long long)le64_to_cpu(fe->i_blkno),
375 (unsigned long long)le64_to_cpu(fe->i_size),
376 (unsigned long long)new_i_size);
Mark Fashehccd979b2005-12-15 14:31:24 -0800377
378 /* lets handle the simple truncate cases before doing any more
379 * cluster locking. */
380 if (new_i_size == le64_to_cpu(fe->i_size))
381 goto bail;
382
Mark Fasheh2e89b2e2007-05-09 13:40:18 -0700383 down_write(&OCFS2_I(inode)->ip_alloc_sem);
384
Mark Fashehab0920c2006-03-16 15:06:37 -0800385 /* This forces other nodes to sync and drop their pages. Do
386 * this even if we have a truncate without allocation change -
387 * ocfs2 cluster sizes can be much greater than page size, so
388 * we have to truncate them anyway. */
389 status = ocfs2_data_lock(inode, 1);
390 if (status < 0) {
Mark Fasheh2e89b2e2007-05-09 13:40:18 -0700391 up_write(&OCFS2_I(inode)->ip_alloc_sem);
392
Mark Fashehab0920c2006-03-16 15:06:37 -0800393 mlog_errno(status);
394 goto bail;
395 }
Mark Fashehab0920c2006-03-16 15:06:37 -0800396
Mark Fasheh2e89b2e2007-05-09 13:40:18 -0700397 unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1);
398 truncate_inode_pages(inode->i_mapping, new_i_size);
399
Mark Fashehccd979b2005-12-15 14:31:24 -0800400 /* alright, we're going to need to do a full blown alloc size
401 * change. Orphan the inode so that recovery can complete the
402 * truncate if necessary. This does the task of marking
403 * i_size. */
404 status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
405 if (status < 0) {
406 mlog_errno(status);
Mark Fasheh60b11392007-02-16 11:46:50 -0800407 goto bail_unlock_data;
Mark Fashehccd979b2005-12-15 14:31:24 -0800408 }
409
410 status = ocfs2_prepare_truncate(osb, inode, di_bh, &tc);
411 if (status < 0) {
412 mlog_errno(status);
Mark Fasheh60b11392007-02-16 11:46:50 -0800413 goto bail_unlock_data;
Mark Fashehccd979b2005-12-15 14:31:24 -0800414 }
415
416 status = ocfs2_commit_truncate(osb, inode, di_bh, tc);
417 if (status < 0) {
418 mlog_errno(status);
Mark Fasheh60b11392007-02-16 11:46:50 -0800419 goto bail_unlock_data;
Mark Fashehccd979b2005-12-15 14:31:24 -0800420 }
421
422 /* TODO: orphan dir cleanup here. */
Mark Fasheh60b11392007-02-16 11:46:50 -0800423bail_unlock_data:
424 ocfs2_data_unlock(inode, 1);
425
Mark Fasheh2e89b2e2007-05-09 13:40:18 -0700426 up_write(&OCFS2_I(inode)->ip_alloc_sem);
427
Mark Fashehccd979b2005-12-15 14:31:24 -0800428bail:
429
430 mlog_exit(status);
431 return status;
432}
433
434/*
435 * extend allocation only here.
436 * we'll update all the disk stuff, and oip->alloc_size
437 *
438 * expect stuff to be locked, a transaction started and enough data /
439 * metadata reservations in the contexts.
440 *
441 * Will return -EAGAIN, and a reason if a restart is needed.
442 * If passed in, *reason will always be set, even in error.
443 */
444int ocfs2_do_extend_allocation(struct ocfs2_super *osb,
445 struct inode *inode,
Mark Fashehdcd05382007-01-16 11:32:23 -0800446 u32 *logical_offset,
Mark Fashehccd979b2005-12-15 14:31:24 -0800447 u32 clusters_to_add,
Mark Fasheh2ae99a62007-03-09 16:43:28 -0800448 int mark_unwritten,
Mark Fashehccd979b2005-12-15 14:31:24 -0800449 struct buffer_head *fe_bh,
Mark Fasheh1fabe142006-10-09 18:11:45 -0700450 handle_t *handle,
Mark Fashehccd979b2005-12-15 14:31:24 -0800451 struct ocfs2_alloc_context *data_ac,
452 struct ocfs2_alloc_context *meta_ac,
453 enum ocfs2_alloc_restarted *reason_ret)
454{
455 int status = 0;
456 int free_extents;
457 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
458 enum ocfs2_alloc_restarted reason = RESTART_NONE;
459 u32 bit_off, num_bits;
460 u64 block;
Mark Fasheh2ae99a62007-03-09 16:43:28 -0800461 u8 flags = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -0800462
463 BUG_ON(!clusters_to_add);
464
Mark Fasheh2ae99a62007-03-09 16:43:28 -0800465 if (mark_unwritten)
466 flags = OCFS2_EXT_UNWRITTEN;
467
Mark Fashehccd979b2005-12-15 14:31:24 -0800468 free_extents = ocfs2_num_free_extents(osb, inode, fe);
469 if (free_extents < 0) {
470 status = free_extents;
471 mlog_errno(status);
472 goto leave;
473 }
474
475 /* there are two cases which could cause us to EAGAIN in the
476 * we-need-more-metadata case:
477 * 1) we haven't reserved *any*
478 * 2) we are so fragmented, we've needed to add metadata too
479 * many times. */
480 if (!free_extents && !meta_ac) {
481 mlog(0, "we haven't reserved any metadata!\n");
482 status = -EAGAIN;
483 reason = RESTART_META;
484 goto leave;
485 } else if ((!free_extents)
486 && (ocfs2_alloc_context_bits_left(meta_ac)
487 < ocfs2_extend_meta_needed(fe))) {
488 mlog(0, "filesystem is really fragmented...\n");
489 status = -EAGAIN;
490 reason = RESTART_META;
491 goto leave;
492 }
493
Mark Fasheh415cb802007-09-16 20:10:16 -0700494 status = __ocfs2_claim_clusters(osb, handle, data_ac, 1,
495 clusters_to_add, &bit_off, &num_bits);
Mark Fashehccd979b2005-12-15 14:31:24 -0800496 if (status < 0) {
497 if (status != -ENOSPC)
498 mlog_errno(status);
499 goto leave;
500 }
501
502 BUG_ON(num_bits > clusters_to_add);
503
504 /* reserve our write early -- insert_extent may update the inode */
505 status = ocfs2_journal_access(handle, inode, fe_bh,
506 OCFS2_JOURNAL_ACCESS_WRITE);
507 if (status < 0) {
508 mlog_errno(status);
509 goto leave;
510 }
511
512 block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
Mark Fashehb06970532006-03-03 10:24:33 -0800513 mlog(0, "Allocating %u clusters at block %u for inode %llu\n",
514 num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
Mark Fashehdcd05382007-01-16 11:32:23 -0800515 status = ocfs2_insert_extent(osb, handle, inode, fe_bh,
516 *logical_offset, block, num_bits,
Mark Fasheh2ae99a62007-03-09 16:43:28 -0800517 flags, meta_ac);
Mark Fashehccd979b2005-12-15 14:31:24 -0800518 if (status < 0) {
519 mlog_errno(status);
520 goto leave;
521 }
522
Mark Fashehccd979b2005-12-15 14:31:24 -0800523 status = ocfs2_journal_dirty(handle, fe_bh);
524 if (status < 0) {
525 mlog_errno(status);
526 goto leave;
527 }
528
529 clusters_to_add -= num_bits;
Mark Fashehdcd05382007-01-16 11:32:23 -0800530 *logical_offset += num_bits;
Mark Fashehccd979b2005-12-15 14:31:24 -0800531
532 if (clusters_to_add) {
533 mlog(0, "need to alloc once more, clusters = %u, wanted = "
534 "%u\n", fe->i_clusters, clusters_to_add);
535 status = -EAGAIN;
536 reason = RESTART_TRANS;
537 }
538
539leave:
540 mlog_exit(status);
541 if (reason_ret)
542 *reason_ret = reason;
543 return status;
544}
545
Mark Fashehabf8b152007-01-17 13:07:24 -0800546/*
547 * For a given allocation, determine which allocators will need to be
548 * accessed, and lock them, reserving the appropriate number of bits.
549 *
Mark Fasheh2ae99a62007-03-09 16:43:28 -0800550 * Sparse file systems call this from ocfs2_write_begin_nolock()
551 * and ocfs2_allocate_unwritten_extents().
552 *
553 * File systems which don't support holes call this from
554 * ocfs2_extend_allocation().
Mark Fashehabf8b152007-01-17 13:07:24 -0800555 */
Mark Fasheh9517bac2007-02-09 20:24:12 -0800556int ocfs2_lock_allocators(struct inode *inode, struct ocfs2_dinode *di,
Mark Fashehb27b7cb2007-06-18 11:22:56 -0700557 u32 clusters_to_add, u32 extents_to_split,
Mark Fasheh9517bac2007-02-09 20:24:12 -0800558 struct ocfs2_alloc_context **data_ac,
559 struct ocfs2_alloc_context **meta_ac)
Mark Fashehabf8b152007-01-17 13:07:24 -0800560{
Mark Fasheh063c4562007-07-03 13:34:11 -0700561 int ret = 0, num_free_extents;
Mark Fashehb27b7cb2007-06-18 11:22:56 -0700562 unsigned int max_recs_needed = clusters_to_add + 2 * extents_to_split;
Mark Fashehabf8b152007-01-17 13:07:24 -0800563 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
564
565 *meta_ac = NULL;
Mark Fasheh063c4562007-07-03 13:34:11 -0700566 if (data_ac)
567 *data_ac = NULL;
568
569 BUG_ON(clusters_to_add != 0 && data_ac == NULL);
Mark Fashehabf8b152007-01-17 13:07:24 -0800570
571 mlog(0, "extend inode %llu, i_size = %lld, di->i_clusters = %u, "
Mark Fashehb27b7cb2007-06-18 11:22:56 -0700572 "clusters_to_add = %u, extents_to_split = %u\n",
Mark Fashehabf8b152007-01-17 13:07:24 -0800573 (unsigned long long)OCFS2_I(inode)->ip_blkno, i_size_read(inode),
Mark Fashehb27b7cb2007-06-18 11:22:56 -0700574 le32_to_cpu(di->i_clusters), clusters_to_add, extents_to_split);
Mark Fashehabf8b152007-01-17 13:07:24 -0800575
576 num_free_extents = ocfs2_num_free_extents(osb, inode, di);
577 if (num_free_extents < 0) {
578 ret = num_free_extents;
579 mlog_errno(ret);
580 goto out;
581 }
582
583 /*
584 * Sparse allocation file systems need to be more conservative
585 * with reserving room for expansion - the actual allocation
586 * happens while we've got a journal handle open so re-taking
587 * a cluster lock (because we ran out of room for another
588 * extent) will violate ordering rules.
589 *
Mark Fasheh9517bac2007-02-09 20:24:12 -0800590 * Most of the time we'll only be seeing this 1 cluster at a time
Mark Fashehabf8b152007-01-17 13:07:24 -0800591 * anyway.
Mark Fashehb27b7cb2007-06-18 11:22:56 -0700592 *
593 * Always lock for any unwritten extents - we might want to
594 * add blocks during a split.
Mark Fashehabf8b152007-01-17 13:07:24 -0800595 */
596 if (!num_free_extents ||
Mark Fashehb27b7cb2007-06-18 11:22:56 -0700597 (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed)) {
Mark Fashehabf8b152007-01-17 13:07:24 -0800598 ret = ocfs2_reserve_new_metadata(osb, di, meta_ac);
599 if (ret < 0) {
600 if (ret != -ENOSPC)
601 mlog_errno(ret);
602 goto out;
603 }
604 }
605
Mark Fasheh063c4562007-07-03 13:34:11 -0700606 if (clusters_to_add == 0)
607 goto out;
608
Mark Fashehabf8b152007-01-17 13:07:24 -0800609 ret = ocfs2_reserve_clusters(osb, clusters_to_add, data_ac);
610 if (ret < 0) {
611 if (ret != -ENOSPC)
612 mlog_errno(ret);
613 goto out;
614 }
615
616out:
617 if (ret) {
618 if (*meta_ac) {
619 ocfs2_free_alloc_context(*meta_ac);
620 *meta_ac = NULL;
621 }
622
623 /*
624 * We cannot have an error and a non null *data_ac.
625 */
626 }
627
628 return ret;
629}
630
Mark Fasheh2ae99a62007-03-09 16:43:28 -0800631static int __ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
632 u32 clusters_to_add, int mark_unwritten)
Mark Fashehccd979b2005-12-15 14:31:24 -0800633{
634 int status = 0;
635 int restart_func = 0;
Mark Fashehabf8b152007-01-17 13:07:24 -0800636 int credits;
Mark Fasheh2ae99a62007-03-09 16:43:28 -0800637 u32 prev_clusters;
Mark Fashehccd979b2005-12-15 14:31:24 -0800638 struct buffer_head *bh = NULL;
639 struct ocfs2_dinode *fe = NULL;
Mark Fasheh1fabe142006-10-09 18:11:45 -0700640 handle_t *handle = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800641 struct ocfs2_alloc_context *data_ac = NULL;
642 struct ocfs2_alloc_context *meta_ac = NULL;
643 enum ocfs2_alloc_restarted why;
644 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
645
646 mlog_entry("(clusters_to_add = %u)\n", clusters_to_add);
647
Mark Fashehdcd05382007-01-16 11:32:23 -0800648 /*
649 * This function only exists for file systems which don't
650 * support holes.
651 */
Mark Fasheh2ae99a62007-03-09 16:43:28 -0800652 BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb));
Mark Fashehdcd05382007-01-16 11:32:23 -0800653
Mark Fashehccd979b2005-12-15 14:31:24 -0800654 status = ocfs2_read_block(osb, OCFS2_I(inode)->ip_blkno, &bh,
655 OCFS2_BH_CACHED, inode);
656 if (status < 0) {
657 mlog_errno(status);
658 goto leave;
659 }
660
661 fe = (struct ocfs2_dinode *) bh->b_data;
662 if (!OCFS2_IS_VALID_DINODE(fe)) {
663 OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe);
664 status = -EIO;
665 goto leave;
666 }
667
668restart_all:
669 BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
670
Mark Fashehb27b7cb2007-06-18 11:22:56 -0700671 status = ocfs2_lock_allocators(inode, fe, clusters_to_add, 0, &data_ac,
Mark Fasheh9517bac2007-02-09 20:24:12 -0800672 &meta_ac);
673 if (status) {
674 mlog_errno(status);
675 goto leave;
676 }
677
Mark Fashehccd979b2005-12-15 14:31:24 -0800678 credits = ocfs2_calc_extend_credits(osb->sb, fe, clusters_to_add);
Mark Fasheh65eff9c2006-10-09 17:26:22 -0700679 handle = ocfs2_start_trans(osb, credits);
Mark Fashehccd979b2005-12-15 14:31:24 -0800680 if (IS_ERR(handle)) {
681 status = PTR_ERR(handle);
682 handle = NULL;
683 mlog_errno(status);
684 goto leave;
685 }
686
687restarted_transaction:
688 /* reserve a write to the file entry early on - that we if we
689 * run out of credits in the allocation path, we can still
690 * update i_size. */
691 status = ocfs2_journal_access(handle, inode, bh,
692 OCFS2_JOURNAL_ACCESS_WRITE);
693 if (status < 0) {
694 mlog_errno(status);
695 goto leave;
696 }
697
698 prev_clusters = OCFS2_I(inode)->ip_clusters;
699
700 status = ocfs2_do_extend_allocation(osb,
701 inode,
Mark Fashehdcd05382007-01-16 11:32:23 -0800702 &logical_start,
Mark Fashehccd979b2005-12-15 14:31:24 -0800703 clusters_to_add,
Mark Fasheh2ae99a62007-03-09 16:43:28 -0800704 mark_unwritten,
Mark Fashehccd979b2005-12-15 14:31:24 -0800705 bh,
706 handle,
707 data_ac,
708 meta_ac,
709 &why);
710 if ((status < 0) && (status != -EAGAIN)) {
711 if (status != -ENOSPC)
712 mlog_errno(status);
713 goto leave;
714 }
715
716 status = ocfs2_journal_dirty(handle, bh);
717 if (status < 0) {
718 mlog_errno(status);
719 goto leave;
720 }
721
722 spin_lock(&OCFS2_I(inode)->ip_lock);
723 clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
724 spin_unlock(&OCFS2_I(inode)->ip_lock);
725
726 if (why != RESTART_NONE && clusters_to_add) {
727 if (why == RESTART_META) {
728 mlog(0, "restarting function.\n");
729 restart_func = 1;
730 } else {
731 BUG_ON(why != RESTART_TRANS);
732
733 mlog(0, "restarting transaction.\n");
734 /* TODO: This can be more intelligent. */
735 credits = ocfs2_calc_extend_credits(osb->sb,
736 fe,
737 clusters_to_add);
Mark Fasheh1fabe142006-10-09 18:11:45 -0700738 status = ocfs2_extend_trans(handle, credits);
Mark Fashehccd979b2005-12-15 14:31:24 -0800739 if (status < 0) {
740 /* handle still has to be committed at
741 * this point. */
742 status = -ENOMEM;
743 mlog_errno(status);
744 goto leave;
745 }
746 goto restarted_transaction;
747 }
748 }
749
Mark Fashehb06970532006-03-03 10:24:33 -0800750 mlog(0, "fe: i_clusters = %u, i_size=%llu\n",
Mark Fasheh1ca1a112007-04-27 16:01:25 -0700751 le32_to_cpu(fe->i_clusters),
752 (unsigned long long)le64_to_cpu(fe->i_size));
Mark Fashehccd979b2005-12-15 14:31:24 -0800753 mlog(0, "inode: ip_clusters=%u, i_size=%lld\n",
754 OCFS2_I(inode)->ip_clusters, i_size_read(inode));
755
756leave:
Mark Fashehccd979b2005-12-15 14:31:24 -0800757 if (handle) {
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700758 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800759 handle = NULL;
760 }
761 if (data_ac) {
762 ocfs2_free_alloc_context(data_ac);
763 data_ac = NULL;
764 }
765 if (meta_ac) {
766 ocfs2_free_alloc_context(meta_ac);
767 meta_ac = NULL;
768 }
769 if ((!status) && restart_func) {
770 restart_func = 0;
771 goto restart_all;
772 }
773 if (bh) {
774 brelse(bh);
775 bh = NULL;
776 }
777
778 mlog_exit(status);
779 return status;
780}
781
782/* Some parts of this taken from generic_cont_expand, which turned out
783 * to be too fragile to do exactly what we need without us having to
Mark Fasheh53013cb2006-05-05 19:04:03 -0700784 * worry about recursive locking in ->prepare_write() and
785 * ->commit_write(). */
Mark Fashehccd979b2005-12-15 14:31:24 -0800786static int ocfs2_write_zero_page(struct inode *inode,
787 u64 size)
788{
789 struct address_space *mapping = inode->i_mapping;
790 struct page *page;
791 unsigned long index;
792 unsigned int offset;
Mark Fasheh1fabe142006-10-09 18:11:45 -0700793 handle_t *handle = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800794 int ret;
795
796 offset = (size & (PAGE_CACHE_SIZE-1)); /* Within page */
797 /* ugh. in prepare/commit_write, if from==to==start of block, we
798 ** skip the prepare. make sure we never send an offset for the start
799 ** of a block
800 */
801 if ((offset & (inode->i_sb->s_blocksize - 1)) == 0) {
802 offset++;
803 }
804 index = size >> PAGE_CACHE_SHIFT;
805
806 page = grab_cache_page(mapping, index);
807 if (!page) {
808 ret = -ENOMEM;
809 mlog_errno(ret);
810 goto out;
811 }
812
Mark Fasheh53013cb2006-05-05 19:04:03 -0700813 ret = ocfs2_prepare_write_nolock(inode, page, offset, offset);
Mark Fashehccd979b2005-12-15 14:31:24 -0800814 if (ret < 0) {
815 mlog_errno(ret);
816 goto out_unlock;
817 }
818
819 if (ocfs2_should_order_data(inode)) {
820 handle = ocfs2_start_walk_page_trans(inode, page, offset,
821 offset);
822 if (IS_ERR(handle)) {
823 ret = PTR_ERR(handle);
824 handle = NULL;
825 goto out_unlock;
826 }
827 }
828
829 /* must not update i_size! */
830 ret = block_commit_write(page, offset, offset);
831 if (ret < 0)
832 mlog_errno(ret);
833 else
834 ret = 0;
835
836 if (handle)
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700837 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800838out_unlock:
839 unlock_page(page);
840 page_cache_release(page);
841out:
842 return ret;
843}
844
845static int ocfs2_zero_extend(struct inode *inode,
846 u64 zero_to_size)
847{
848 int ret = 0;
849 u64 start_off;
850 struct super_block *sb = inode->i_sb;
851
852 start_off = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
853 while (start_off < zero_to_size) {
854 ret = ocfs2_write_zero_page(inode, start_off);
855 if (ret < 0) {
856 mlog_errno(ret);
857 goto out;
858 }
859
860 start_off += sb->s_blocksize;
Mark Fashehe2057c52006-10-03 17:53:05 -0700861
862 /*
863 * Very large extends have the potential to lock up
864 * the cpu for extended periods of time.
865 */
866 cond_resched();
Mark Fashehccd979b2005-12-15 14:31:24 -0800867 }
868
869out:
870 return ret;
871}
872
Mark Fasheh65ed39d2007-08-28 17:13:23 -0700873int ocfs2_extend_no_holes(struct inode *inode, u64 new_i_size, u64 zero_to)
874{
875 int ret;
876 u32 clusters_to_add;
877 struct ocfs2_inode_info *oi = OCFS2_I(inode);
878
879 clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size);
880 if (clusters_to_add < oi->ip_clusters)
881 clusters_to_add = 0;
882 else
883 clusters_to_add -= oi->ip_clusters;
884
885 if (clusters_to_add) {
886 ret = __ocfs2_extend_allocation(inode, oi->ip_clusters,
887 clusters_to_add, 0);
888 if (ret) {
889 mlog_errno(ret);
890 goto out;
891 }
892 }
893
894 /*
895 * Call this even if we don't add any clusters to the tree. We
896 * still need to zero the area between the old i_size and the
897 * new i_size.
898 */
899 ret = ocfs2_zero_extend(inode, zero_to);
900 if (ret < 0)
901 mlog_errno(ret);
902
903out:
904 return ret;
905}
906
Mark Fashehccd979b2005-12-15 14:31:24 -0800907static int ocfs2_extend_file(struct inode *inode,
908 struct buffer_head *di_bh,
Mark Fasheh65ed39d2007-08-28 17:13:23 -0700909 u64 new_i_size)
Mark Fashehccd979b2005-12-15 14:31:24 -0800910{
911 int ret = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -0800912
Mark Fasheh65ed39d2007-08-28 17:13:23 -0700913 BUG_ON(!di_bh);
Mark Fasheh53013cb2006-05-05 19:04:03 -0700914
Mark Fashehccd979b2005-12-15 14:31:24 -0800915 /* setattr sometimes calls us like this. */
916 if (new_i_size == 0)
917 goto out;
918
919 if (i_size_read(inode) == new_i_size)
920 goto out;
921 BUG_ON(new_i_size < i_size_read(inode));
922
Mark Fasheh65ed39d2007-08-28 17:13:23 -0700923 if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
Mark Fasheh3a0782d2007-01-17 12:53:31 -0800924 goto out_update_size;
Mark Fashehccd979b2005-12-15 14:31:24 -0800925
Mark Fasheh0effef72006-10-03 17:44:42 -0700926 /*
927 * protect the pages that ocfs2_zero_extend is going to be
928 * pulling into the page cache.. we do this before the
929 * metadata extend so that we don't get into the situation
930 * where we've extended the metadata but can't get the data
931 * lock to zero.
932 */
933 ret = ocfs2_data_lock(inode, 1);
934 if (ret < 0) {
935 mlog_errno(ret);
936 goto out;
937 }
Mark Fasheh53013cb2006-05-05 19:04:03 -0700938
Mark Fasheh0effef72006-10-03 17:44:42 -0700939 /*
Mark Fasheh65ed39d2007-08-28 17:13:23 -0700940 * The alloc sem blocks people in read/write from reading our
941 * allocation until we're done changing it. We depend on
942 * i_mutex to block other extend/truncate calls while we're
943 * here.
Mark Fasheh0effef72006-10-03 17:44:42 -0700944 */
Mark Fasheh65ed39d2007-08-28 17:13:23 -0700945 down_write(&OCFS2_I(inode)->ip_alloc_sem);
946 ret = ocfs2_extend_no_holes(inode, new_i_size, new_i_size);
947 up_write(&OCFS2_I(inode)->ip_alloc_sem);
948
Mark Fasheh0effef72006-10-03 17:44:42 -0700949 if (ret < 0) {
950 mlog_errno(ret);
951 goto out_unlock;
Mark Fashehccd979b2005-12-15 14:31:24 -0800952 }
953
Mark Fasheh3a0782d2007-01-17 12:53:31 -0800954out_update_size:
Mark Fasheh65ed39d2007-08-28 17:13:23 -0700955 ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
956 if (ret < 0)
957 mlog_errno(ret);
Mark Fasheh53013cb2006-05-05 19:04:03 -0700958
959out_unlock:
Mark Fasheh3a0782d2007-01-17 12:53:31 -0800960 if (!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
961 ocfs2_data_unlock(inode, 1);
Mark Fasheh53013cb2006-05-05 19:04:03 -0700962
Mark Fashehccd979b2005-12-15 14:31:24 -0800963out:
964 return ret;
965}
966
967int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
968{
969 int status = 0, size_change;
970 struct inode *inode = dentry->d_inode;
971 struct super_block *sb = inode->i_sb;
972 struct ocfs2_super *osb = OCFS2_SB(sb);
973 struct buffer_head *bh = NULL;
Mark Fasheh1fabe142006-10-09 18:11:45 -0700974 handle_t *handle = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800975
976 mlog_entry("(0x%p, '%.*s')\n", dentry,
977 dentry->d_name.len, dentry->d_name.name);
978
979 if (attr->ia_valid & ATTR_MODE)
980 mlog(0, "mode change: %d\n", attr->ia_mode);
981 if (attr->ia_valid & ATTR_UID)
982 mlog(0, "uid change: %d\n", attr->ia_uid);
983 if (attr->ia_valid & ATTR_GID)
984 mlog(0, "gid change: %d\n", attr->ia_gid);
985 if (attr->ia_valid & ATTR_SIZE)
986 mlog(0, "size change...\n");
987 if (attr->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME))
988 mlog(0, "time change...\n");
989
990#define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
991 | ATTR_GID | ATTR_UID | ATTR_MODE)
992 if (!(attr->ia_valid & OCFS2_VALID_ATTRS)) {
993 mlog(0, "can't handle attrs: 0x%x\n", attr->ia_valid);
994 return 0;
995 }
996
997 status = inode_change_ok(inode, attr);
998 if (status)
999 return status;
1000
1001 size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
1002 if (size_change) {
1003 status = ocfs2_rw_lock(inode, 1);
1004 if (status < 0) {
1005 mlog_errno(status);
1006 goto bail;
1007 }
1008 }
1009
Mark Fasheh4bcec182006-10-09 16:02:40 -07001010 status = ocfs2_meta_lock(inode, &bh, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -08001011 if (status < 0) {
1012 if (status != -ENOENT)
1013 mlog_errno(status);
1014 goto bail_unlock_rw;
1015 }
1016
1017 if (size_change && attr->ia_size != i_size_read(inode)) {
Mark Fashehce76fd32007-07-20 12:02:14 -07001018 if (attr->ia_size > sb->s_maxbytes) {
1019 status = -EFBIG;
1020 goto bail_unlock;
1021 }
1022
Mark Fashehccd979b2005-12-15 14:31:24 -08001023 if (i_size_read(inode) > attr->ia_size)
1024 status = ocfs2_truncate_file(inode, bh, attr->ia_size);
1025 else
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001026 status = ocfs2_extend_file(inode, bh, attr->ia_size);
Mark Fashehccd979b2005-12-15 14:31:24 -08001027 if (status < 0) {
1028 if (status != -ENOSPC)
1029 mlog_errno(status);
1030 status = -ENOSPC;
1031 goto bail_unlock;
1032 }
1033 }
1034
Mark Fasheh65eff9c2006-10-09 17:26:22 -07001035 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
Mark Fashehccd979b2005-12-15 14:31:24 -08001036 if (IS_ERR(handle)) {
1037 status = PTR_ERR(handle);
1038 mlog_errno(status);
1039 goto bail_unlock;
1040 }
1041
Mark Fasheh7307de82007-05-09 15:16:19 -07001042 /*
1043 * This will intentionally not wind up calling vmtruncate(),
1044 * since all the work for a size change has been done above.
1045 * Otherwise, we could get into problems with truncate as
1046 * ip_alloc_sem is used there to protect against i_size
1047 * changes.
1048 */
Mark Fashehccd979b2005-12-15 14:31:24 -08001049 status = inode_setattr(inode, attr);
1050 if (status < 0) {
1051 mlog_errno(status);
1052 goto bail_commit;
1053 }
1054
1055 status = ocfs2_mark_inode_dirty(handle, inode, bh);
1056 if (status < 0)
1057 mlog_errno(status);
1058
1059bail_commit:
Mark Fasheh02dc1af2006-10-09 16:48:10 -07001060 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -08001061bail_unlock:
1062 ocfs2_meta_unlock(inode, 1);
1063bail_unlock_rw:
1064 if (size_change)
1065 ocfs2_rw_unlock(inode, 1);
1066bail:
1067 if (bh)
1068 brelse(bh);
1069
1070 mlog_exit(status);
1071 return status;
1072}
1073
1074int ocfs2_getattr(struct vfsmount *mnt,
1075 struct dentry *dentry,
1076 struct kstat *stat)
1077{
1078 struct inode *inode = dentry->d_inode;
1079 struct super_block *sb = dentry->d_inode->i_sb;
1080 struct ocfs2_super *osb = sb->s_fs_info;
1081 int err;
1082
1083 mlog_entry_void();
1084
1085 err = ocfs2_inode_revalidate(dentry);
1086 if (err) {
1087 if (err != -ENOENT)
1088 mlog_errno(err);
1089 goto bail;
1090 }
1091
1092 generic_fillattr(inode, stat);
1093
1094 /* We set the blksize from the cluster size for performance */
1095 stat->blksize = osb->s_clustersize;
1096
1097bail:
1098 mlog_exit(err);
1099
1100 return err;
1101}
1102
Tiger Yangd38eb8d2006-11-27 09:59:21 +08001103int ocfs2_permission(struct inode *inode, int mask, struct nameidata *nd)
1104{
1105 int ret;
1106
1107 mlog_entry_void();
1108
1109 ret = ocfs2_meta_lock(inode, NULL, 0);
1110 if (ret) {
Mark Fasheha9f5f702007-04-26 11:43:43 -07001111 if (ret != -ENOENT)
1112 mlog_errno(ret);
Tiger Yangd38eb8d2006-11-27 09:59:21 +08001113 goto out;
1114 }
1115
1116 ret = generic_permission(inode, mask, NULL);
Tiger Yangd38eb8d2006-11-27 09:59:21 +08001117
1118 ocfs2_meta_unlock(inode, 0);
1119out:
1120 mlog_exit(ret);
1121 return ret;
1122}
1123
Mark Fashehb2580102007-03-09 16:53:21 -08001124static int __ocfs2_write_remove_suid(struct inode *inode,
1125 struct buffer_head *bh)
Mark Fashehccd979b2005-12-15 14:31:24 -08001126{
1127 int ret;
Mark Fasheh1fabe142006-10-09 18:11:45 -07001128 handle_t *handle;
Mark Fashehccd979b2005-12-15 14:31:24 -08001129 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1130 struct ocfs2_dinode *di;
1131
Mark Fashehb06970532006-03-03 10:24:33 -08001132 mlog_entry("(Inode %llu, mode 0%o)\n",
Mark Fashehb2580102007-03-09 16:53:21 -08001133 (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_mode);
Mark Fashehccd979b2005-12-15 14:31:24 -08001134
Mark Fasheh65eff9c2006-10-09 17:26:22 -07001135 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
Mark Fashehccd979b2005-12-15 14:31:24 -08001136 if (handle == NULL) {
1137 ret = -ENOMEM;
1138 mlog_errno(ret);
1139 goto out;
1140 }
1141
Mark Fashehccd979b2005-12-15 14:31:24 -08001142 ret = ocfs2_journal_access(handle, inode, bh,
1143 OCFS2_JOURNAL_ACCESS_WRITE);
1144 if (ret < 0) {
1145 mlog_errno(ret);
Mark Fashehb2580102007-03-09 16:53:21 -08001146 goto out_trans;
Mark Fashehccd979b2005-12-15 14:31:24 -08001147 }
1148
1149 inode->i_mode &= ~S_ISUID;
1150 if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
1151 inode->i_mode &= ~S_ISGID;
1152
1153 di = (struct ocfs2_dinode *) bh->b_data;
1154 di->i_mode = cpu_to_le16(inode->i_mode);
1155
1156 ret = ocfs2_journal_dirty(handle, bh);
1157 if (ret < 0)
1158 mlog_errno(ret);
Mark Fashehb2580102007-03-09 16:53:21 -08001159
Mark Fashehccd979b2005-12-15 14:31:24 -08001160out_trans:
Mark Fasheh02dc1af2006-10-09 16:48:10 -07001161 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -08001162out:
1163 mlog_exit(ret);
1164 return ret;
1165}
1166
Mark Fasheh9517bac2007-02-09 20:24:12 -08001167/*
1168 * Will look for holes and unwritten extents in the range starting at
1169 * pos for count bytes (inclusive).
1170 */
1171static int ocfs2_check_range_for_holes(struct inode *inode, loff_t pos,
1172 size_t count)
1173{
1174 int ret = 0;
Mark Fasheh49cb8d22007-03-09 16:21:46 -08001175 unsigned int extent_flags;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001176 u32 cpos, clusters, extent_len, phys_cpos;
1177 struct super_block *sb = inode->i_sb;
1178
1179 cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
1180 clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
1181
1182 while (clusters) {
Mark Fasheh49cb8d22007-03-09 16:21:46 -08001183 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
1184 &extent_flags);
Mark Fasheh9517bac2007-02-09 20:24:12 -08001185 if (ret < 0) {
1186 mlog_errno(ret);
1187 goto out;
1188 }
1189
Mark Fasheh49cb8d22007-03-09 16:21:46 -08001190 if (phys_cpos == 0 || (extent_flags & OCFS2_EXT_UNWRITTEN)) {
Mark Fasheh9517bac2007-02-09 20:24:12 -08001191 ret = 1;
1192 break;
1193 }
1194
1195 if (extent_len > clusters)
1196 extent_len = clusters;
1197
1198 clusters -= extent_len;
1199 cpos += extent_len;
1200 }
1201out:
1202 return ret;
1203}
1204
Mark Fashehb2580102007-03-09 16:53:21 -08001205static int ocfs2_write_remove_suid(struct inode *inode)
1206{
1207 int ret;
1208 struct buffer_head *bh = NULL;
1209 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1210
1211 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
1212 oi->ip_blkno, &bh, OCFS2_BH_CACHED, inode);
1213 if (ret < 0) {
1214 mlog_errno(ret);
1215 goto out;
1216 }
1217
1218 ret = __ocfs2_write_remove_suid(inode, bh);
1219out:
1220 brelse(bh);
1221 return ret;
1222}
1223
Mark Fasheh2ae99a62007-03-09 16:43:28 -08001224/*
1225 * Allocate enough extents to cover the region starting at byte offset
1226 * start for len bytes. Existing extents are skipped, any extents
1227 * added are marked as "unwritten".
1228 */
1229static int ocfs2_allocate_unwritten_extents(struct inode *inode,
1230 u64 start, u64 len)
1231{
1232 int ret;
1233 u32 cpos, phys_cpos, clusters, alloc_size;
1234
1235 /*
1236 * We consider both start and len to be inclusive.
1237 */
1238 cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
1239 clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len);
1240 clusters -= cpos;
1241
1242 while (clusters) {
1243 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1244 &alloc_size, NULL);
1245 if (ret) {
1246 mlog_errno(ret);
1247 goto out;
1248 }
1249
1250 /*
1251 * Hole or existing extent len can be arbitrary, so
1252 * cap it to our own allocation request.
1253 */
1254 if (alloc_size > clusters)
1255 alloc_size = clusters;
1256
1257 if (phys_cpos) {
1258 /*
1259 * We already have an allocation at this
1260 * region so we can safely skip it.
1261 */
1262 goto next;
1263 }
1264
1265 ret = __ocfs2_extend_allocation(inode, cpos, alloc_size, 1);
1266 if (ret) {
1267 if (ret != -ENOSPC)
1268 mlog_errno(ret);
1269 goto out;
1270 }
1271
1272next:
1273 cpos += alloc_size;
1274 clusters -= alloc_size;
1275 }
1276
1277 ret = 0;
1278out:
1279 return ret;
1280}
1281
Mark Fasheh063c4562007-07-03 13:34:11 -07001282static int __ocfs2_remove_inode_range(struct inode *inode,
1283 struct buffer_head *di_bh,
1284 u32 cpos, u32 phys_cpos, u32 len,
1285 struct ocfs2_cached_dealloc_ctxt *dealloc)
1286{
1287 int ret;
1288 u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
1289 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1290 struct inode *tl_inode = osb->osb_tl_inode;
1291 handle_t *handle;
1292 struct ocfs2_alloc_context *meta_ac = NULL;
1293 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1294
1295 ret = ocfs2_lock_allocators(inode, di, 0, 1, NULL, &meta_ac);
1296 if (ret) {
1297 mlog_errno(ret);
1298 return ret;
1299 }
1300
1301 mutex_lock(&tl_inode->i_mutex);
1302
1303 if (ocfs2_truncate_log_needs_flush(osb)) {
1304 ret = __ocfs2_flush_truncate_log(osb);
1305 if (ret < 0) {
1306 mlog_errno(ret);
1307 goto out;
1308 }
1309 }
1310
1311 handle = ocfs2_start_trans(osb, OCFS2_REMOVE_EXTENT_CREDITS);
1312 if (handle == NULL) {
1313 ret = -ENOMEM;
1314 mlog_errno(ret);
1315 goto out;
1316 }
1317
1318 ret = ocfs2_journal_access(handle, inode, di_bh,
1319 OCFS2_JOURNAL_ACCESS_WRITE);
1320 if (ret) {
1321 mlog_errno(ret);
1322 goto out;
1323 }
1324
1325 ret = ocfs2_remove_extent(inode, di_bh, cpos, len, handle, meta_ac,
1326 dealloc);
1327 if (ret) {
1328 mlog_errno(ret);
1329 goto out_commit;
1330 }
1331
1332 OCFS2_I(inode)->ip_clusters -= len;
1333 di->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters);
1334
1335 ret = ocfs2_journal_dirty(handle, di_bh);
1336 if (ret) {
1337 mlog_errno(ret);
1338 goto out_commit;
1339 }
1340
1341 ret = ocfs2_truncate_log_append(osb, handle, phys_blkno, len);
1342 if (ret)
1343 mlog_errno(ret);
1344
1345out_commit:
1346 ocfs2_commit_trans(osb, handle);
1347out:
1348 mutex_unlock(&tl_inode->i_mutex);
1349
1350 if (meta_ac)
1351 ocfs2_free_alloc_context(meta_ac);
1352
1353 return ret;
1354}
1355
1356/*
1357 * Truncate a byte range, avoiding pages within partial clusters. This
1358 * preserves those pages for the zeroing code to write to.
1359 */
1360static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start,
1361 u64 byte_len)
1362{
1363 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1364 loff_t start, end;
1365 struct address_space *mapping = inode->i_mapping;
1366
1367 start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start);
1368 end = byte_start + byte_len;
1369 end = end & ~(osb->s_clustersize - 1);
1370
1371 if (start < end) {
1372 unmap_mapping_range(mapping, start, end - start, 0);
1373 truncate_inode_pages_range(mapping, start, end - 1);
1374 }
1375}
1376
1377static int ocfs2_zero_partial_clusters(struct inode *inode,
1378 u64 start, u64 len)
1379{
1380 int ret = 0;
1381 u64 tmpend, end = start + len;
1382 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1383 unsigned int csize = osb->s_clustersize;
1384 handle_t *handle;
1385
1386 /*
1387 * The "start" and "end" values are NOT necessarily part of
1388 * the range whose allocation is being deleted. Rather, this
1389 * is what the user passed in with the request. We must zero
1390 * partial clusters here. There's no need to worry about
1391 * physical allocation - the zeroing code knows to skip holes.
1392 */
1393 mlog(0, "byte start: %llu, end: %llu\n",
1394 (unsigned long long)start, (unsigned long long)end);
1395
1396 /*
1397 * If both edges are on a cluster boundary then there's no
1398 * zeroing required as the region is part of the allocation to
1399 * be truncated.
1400 */
1401 if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0)
1402 goto out;
1403
1404 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1405 if (handle == NULL) {
1406 ret = -ENOMEM;
1407 mlog_errno(ret);
1408 goto out;
1409 }
1410
1411 /*
1412 * We want to get the byte offset of the end of the 1st cluster.
1413 */
1414 tmpend = (u64)osb->s_clustersize + (start & ~(osb->s_clustersize - 1));
1415 if (tmpend > end)
1416 tmpend = end;
1417
1418 mlog(0, "1st range: start: %llu, tmpend: %llu\n",
1419 (unsigned long long)start, (unsigned long long)tmpend);
1420
1421 ret = ocfs2_zero_range_for_truncate(inode, handle, start, tmpend);
1422 if (ret)
1423 mlog_errno(ret);
1424
1425 if (tmpend < end) {
1426 /*
1427 * This may make start and end equal, but the zeroing
1428 * code will skip any work in that case so there's no
1429 * need to catch it up here.
1430 */
1431 start = end & ~(osb->s_clustersize - 1);
1432
1433 mlog(0, "2nd range: start: %llu, end: %llu\n",
1434 (unsigned long long)start, (unsigned long long)end);
1435
1436 ret = ocfs2_zero_range_for_truncate(inode, handle, start, end);
1437 if (ret)
1438 mlog_errno(ret);
1439 }
1440
1441 ocfs2_commit_trans(osb, handle);
1442out:
1443 return ret;
1444}
1445
1446static int ocfs2_remove_inode_range(struct inode *inode,
1447 struct buffer_head *di_bh, u64 byte_start,
1448 u64 byte_len)
1449{
1450 int ret = 0;
1451 u32 trunc_start, trunc_len, cpos, phys_cpos, alloc_size;
1452 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1453 struct ocfs2_cached_dealloc_ctxt dealloc;
1454
1455 ocfs2_init_dealloc_ctxt(&dealloc);
1456
1457 if (byte_len == 0)
1458 return 0;
1459
1460 trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start);
1461 trunc_len = (byte_start + byte_len) >> osb->s_clustersize_bits;
1462 if (trunc_len >= trunc_start)
1463 trunc_len -= trunc_start;
1464 else
1465 trunc_len = 0;
1466
1467 mlog(0, "Inode: %llu, start: %llu, len: %llu, cstart: %u, clen: %u\n",
1468 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1469 (unsigned long long)byte_start,
1470 (unsigned long long)byte_len, trunc_start, trunc_len);
1471
1472 ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len);
1473 if (ret) {
1474 mlog_errno(ret);
1475 goto out;
1476 }
1477
1478 cpos = trunc_start;
1479 while (trunc_len) {
1480 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1481 &alloc_size, NULL);
1482 if (ret) {
1483 mlog_errno(ret);
1484 goto out;
1485 }
1486
1487 if (alloc_size > trunc_len)
1488 alloc_size = trunc_len;
1489
1490 /* Only do work for non-holes */
1491 if (phys_cpos != 0) {
1492 ret = __ocfs2_remove_inode_range(inode, di_bh, cpos,
1493 phys_cpos, alloc_size,
1494 &dealloc);
1495 if (ret) {
1496 mlog_errno(ret);
1497 goto out;
1498 }
1499 }
1500
1501 cpos += alloc_size;
1502 trunc_len -= alloc_size;
1503 }
1504
1505 ocfs2_truncate_cluster_pages(inode, byte_start, byte_len);
1506
1507out:
1508 ocfs2_schedule_truncate_log_flush(osb, 1);
1509 ocfs2_run_deallocs(osb, &dealloc);
1510
1511 return ret;
1512}
1513
Mark Fashehb2580102007-03-09 16:53:21 -08001514/*
1515 * Parts of this function taken from xfs_change_file_space()
1516 */
Mark Fasheh385820a2007-07-19 00:14:38 -07001517static int __ocfs2_change_file_space(struct file *file, struct inode *inode,
1518 loff_t f_pos, unsigned int cmd,
1519 struct ocfs2_space_resv *sr,
1520 int change_size)
Mark Fashehb2580102007-03-09 16:53:21 -08001521{
1522 int ret;
1523 s64 llen;
Mark Fasheh385820a2007-07-19 00:14:38 -07001524 loff_t size;
Mark Fashehb2580102007-03-09 16:53:21 -08001525 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1526 struct buffer_head *di_bh = NULL;
1527 handle_t *handle;
Mark Fasheha00cce32007-07-20 11:28:30 -07001528 unsigned long long max_off = inode->i_sb->s_maxbytes;
Mark Fashehb2580102007-03-09 16:53:21 -08001529
Mark Fashehb2580102007-03-09 16:53:21 -08001530 if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
1531 return -EROFS;
1532
1533 mutex_lock(&inode->i_mutex);
1534
1535 /*
1536 * This prevents concurrent writes on other nodes
1537 */
1538 ret = ocfs2_rw_lock(inode, 1);
1539 if (ret) {
1540 mlog_errno(ret);
1541 goto out;
1542 }
1543
1544 ret = ocfs2_meta_lock(inode, &di_bh, 1);
1545 if (ret) {
1546 mlog_errno(ret);
1547 goto out_rw_unlock;
1548 }
1549
1550 if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
1551 ret = -EPERM;
1552 goto out_meta_unlock;
1553 }
1554
1555 switch (sr->l_whence) {
1556 case 0: /*SEEK_SET*/
1557 break;
1558 case 1: /*SEEK_CUR*/
Mark Fasheh385820a2007-07-19 00:14:38 -07001559 sr->l_start += f_pos;
Mark Fashehb2580102007-03-09 16:53:21 -08001560 break;
1561 case 2: /*SEEK_END*/
1562 sr->l_start += i_size_read(inode);
1563 break;
1564 default:
1565 ret = -EINVAL;
1566 goto out_meta_unlock;
1567 }
1568 sr->l_whence = 0;
1569
1570 llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len;
1571
1572 if (sr->l_start < 0
1573 || sr->l_start > max_off
1574 || (sr->l_start + llen) < 0
1575 || (sr->l_start + llen) > max_off) {
1576 ret = -EINVAL;
1577 goto out_meta_unlock;
1578 }
Mark Fasheh385820a2007-07-19 00:14:38 -07001579 size = sr->l_start + sr->l_len;
Mark Fashehb2580102007-03-09 16:53:21 -08001580
1581 if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) {
1582 if (sr->l_len <= 0) {
1583 ret = -EINVAL;
1584 goto out_meta_unlock;
1585 }
1586 }
1587
Mark Fasheh385820a2007-07-19 00:14:38 -07001588 if (file && should_remove_suid(file->f_path.dentry)) {
Mark Fashehb2580102007-03-09 16:53:21 -08001589 ret = __ocfs2_write_remove_suid(inode, di_bh);
1590 if (ret) {
1591 mlog_errno(ret);
1592 goto out_meta_unlock;
1593 }
1594 }
1595
1596 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1597 switch (cmd) {
1598 case OCFS2_IOC_RESVSP:
1599 case OCFS2_IOC_RESVSP64:
1600 /*
1601 * This takes unsigned offsets, but the signed ones we
1602 * pass have been checked against overflow above.
1603 */
1604 ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start,
1605 sr->l_len);
1606 break;
1607 case OCFS2_IOC_UNRESVSP:
1608 case OCFS2_IOC_UNRESVSP64:
1609 ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start,
1610 sr->l_len);
1611 break;
1612 default:
1613 ret = -EINVAL;
1614 }
1615 up_write(&OCFS2_I(inode)->ip_alloc_sem);
1616 if (ret) {
1617 mlog_errno(ret);
1618 goto out_meta_unlock;
1619 }
1620
1621 /*
1622 * We update c/mtime for these changes
1623 */
1624 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1625 if (IS_ERR(handle)) {
1626 ret = PTR_ERR(handle);
1627 mlog_errno(ret);
1628 goto out_meta_unlock;
1629 }
1630
Mark Fasheh385820a2007-07-19 00:14:38 -07001631 if (change_size && i_size_read(inode) < size)
1632 i_size_write(inode, size);
1633
Mark Fashehb2580102007-03-09 16:53:21 -08001634 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1635 ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
1636 if (ret < 0)
1637 mlog_errno(ret);
1638
1639 ocfs2_commit_trans(osb, handle);
1640
1641out_meta_unlock:
1642 brelse(di_bh);
1643 ocfs2_meta_unlock(inode, 1);
1644out_rw_unlock:
1645 ocfs2_rw_unlock(inode, 1);
1646
1647 mutex_unlock(&inode->i_mutex);
1648out:
1649 return ret;
1650}
1651
Mark Fasheh385820a2007-07-19 00:14:38 -07001652int ocfs2_change_file_space(struct file *file, unsigned int cmd,
1653 struct ocfs2_space_resv *sr)
1654{
1655 struct inode *inode = file->f_path.dentry->d_inode;
1656 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);;
1657
1658 if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) &&
1659 !ocfs2_writes_unwritten_extents(osb))
1660 return -ENOTTY;
1661 else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) &&
1662 !ocfs2_sparse_alloc(osb))
1663 return -ENOTTY;
1664
1665 if (!S_ISREG(inode->i_mode))
1666 return -EINVAL;
1667
1668 if (!(file->f_mode & FMODE_WRITE))
1669 return -EBADF;
1670
1671 return __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0);
1672}
1673
1674static long ocfs2_fallocate(struct inode *inode, int mode, loff_t offset,
1675 loff_t len)
1676{
1677 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1678 struct ocfs2_space_resv sr;
1679 int change_size = 1;
1680
1681 if (!ocfs2_writes_unwritten_extents(osb))
1682 return -EOPNOTSUPP;
1683
1684 if (S_ISDIR(inode->i_mode))
1685 return -ENODEV;
1686
1687 if (mode & FALLOC_FL_KEEP_SIZE)
1688 change_size = 0;
1689
1690 sr.l_whence = 0;
1691 sr.l_start = (s64)offset;
1692 sr.l_len = (s64)len;
1693
1694 return __ocfs2_change_file_space(NULL, inode, offset,
1695 OCFS2_IOC_RESVSP64, &sr, change_size);
1696}
1697
Tiger Yang8659ac22006-10-17 18:29:52 -07001698static int ocfs2_prepare_inode_for_write(struct dentry *dentry,
1699 loff_t *ppos,
1700 size_t count,
Mark Fasheh9517bac2007-02-09 20:24:12 -08001701 int appending,
1702 int *direct_io)
Mark Fashehccd979b2005-12-15 14:31:24 -08001703{
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001704 int ret = 0, meta_level = 0;
Tiger Yang8659ac22006-10-17 18:29:52 -07001705 struct inode *inode = dentry->d_inode;
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001706 loff_t saved_pos, end;
Mark Fashehccd979b2005-12-15 14:31:24 -08001707
Mark Fashehccd979b2005-12-15 14:31:24 -08001708 /*
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001709 * We start with a read level meta lock and only jump to an ex
1710 * if we need to make modifications here.
Mark Fashehccd979b2005-12-15 14:31:24 -08001711 */
Mark Fashehccd979b2005-12-15 14:31:24 -08001712 for(;;) {
Mark Fasheh4bcec182006-10-09 16:02:40 -07001713 ret = ocfs2_meta_lock(inode, NULL, meta_level);
Mark Fashehccd979b2005-12-15 14:31:24 -08001714 if (ret < 0) {
1715 meta_level = -1;
1716 mlog_errno(ret);
1717 goto out;
1718 }
1719
1720 /* Clear suid / sgid if necessary. We do this here
1721 * instead of later in the write path because
1722 * remove_suid() calls ->setattr without any hint that
1723 * we may have already done our cluster locking. Since
1724 * ocfs2_setattr() *must* take cluster locks to
1725 * proceeed, this will lead us to recursively lock the
1726 * inode. There's also the dinode i_size state which
1727 * can be lost via setattr during extending writes (we
1728 * set inode->i_size at the end of a write. */
Tiger Yang8659ac22006-10-17 18:29:52 -07001729 if (should_remove_suid(dentry)) {
Mark Fashehccd979b2005-12-15 14:31:24 -08001730 if (meta_level == 0) {
1731 ocfs2_meta_unlock(inode, meta_level);
1732 meta_level = 1;
1733 continue;
1734 }
1735
1736 ret = ocfs2_write_remove_suid(inode);
1737 if (ret < 0) {
1738 mlog_errno(ret);
Tiger Yang8659ac22006-10-17 18:29:52 -07001739 goto out_unlock;
Mark Fashehccd979b2005-12-15 14:31:24 -08001740 }
1741 }
1742
1743 /* work on a copy of ppos until we're sure that we won't have
1744 * to recalculate it due to relocking. */
Tiger Yang8659ac22006-10-17 18:29:52 -07001745 if (appending) {
Mark Fashehccd979b2005-12-15 14:31:24 -08001746 saved_pos = i_size_read(inode);
1747 mlog(0, "O_APPEND: inode->i_size=%llu\n", saved_pos);
1748 } else {
Tiger Yang8659ac22006-10-17 18:29:52 -07001749 saved_pos = *ppos;
Mark Fashehccd979b2005-12-15 14:31:24 -08001750 }
Mark Fasheh3a0782d2007-01-17 12:53:31 -08001751
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001752 end = saved_pos + count;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001753
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001754 /*
1755 * Skip the O_DIRECT checks if we don't need
1756 * them.
1757 */
1758 if (!direct_io || !(*direct_io))
1759 break;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001760
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001761 /*
1762 * Allowing concurrent direct writes means
1763 * i_size changes wouldn't be synchronized, so
1764 * one node could wind up truncating another
1765 * nodes writes.
1766 */
1767 if (end > i_size_read(inode)) {
1768 *direct_io = 0;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001769 break;
1770 }
1771
Mark Fasheh3a0782d2007-01-17 12:53:31 -08001772 /*
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001773 * We don't fill holes during direct io, so
1774 * check for them here. If any are found, the
1775 * caller will have to retake some cluster
1776 * locks and initiate the io as buffered.
Mark Fasheh3a0782d2007-01-17 12:53:31 -08001777 */
Mark Fasheh65ed39d2007-08-28 17:13:23 -07001778 ret = ocfs2_check_range_for_holes(inode, saved_pos, count);
1779 if (ret == 1) {
1780 *direct_io = 0;
1781 ret = 0;
1782 } else if (ret < 0)
1783 mlog_errno(ret);
Mark Fashehccd979b2005-12-15 14:31:24 -08001784 break;
1785 }
1786
Tiger Yang8659ac22006-10-17 18:29:52 -07001787 if (appending)
1788 *ppos = saved_pos;
1789
1790out_unlock:
Mark Fashehccd979b2005-12-15 14:31:24 -08001791 ocfs2_meta_unlock(inode, meta_level);
Tiger Yang8659ac22006-10-17 18:29:52 -07001792
1793out:
1794 return ret;
1795}
1796
Mark Fasheh9517bac2007-02-09 20:24:12 -08001797static inline void
1798ocfs2_set_next_iovec(const struct iovec **iovp, size_t *basep, size_t bytes)
1799{
1800 const struct iovec *iov = *iovp;
1801 size_t base = *basep;
1802
1803 do {
1804 int copy = min(bytes, iov->iov_len - base);
1805
1806 bytes -= copy;
1807 base += copy;
1808 if (iov->iov_len == base) {
1809 iov++;
1810 base = 0;
1811 }
1812 } while (bytes);
1813 *iovp = iov;
1814 *basep = base;
1815}
1816
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001817static struct page * ocfs2_get_write_source(char **ret_src_buf,
Mark Fasheh9517bac2007-02-09 20:24:12 -08001818 const struct iovec *cur_iov,
1819 size_t iov_offset)
1820{
1821 int ret;
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001822 char *buf = cur_iov->iov_base + iov_offset;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001823 struct page *src_page = NULL;
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001824 unsigned long off;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001825
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001826 off = (unsigned long)(buf) & ~PAGE_CACHE_MASK;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001827
1828 if (!segment_eq(get_fs(), KERNEL_DS)) {
1829 /*
1830 * Pull in the user page. We want to do this outside
1831 * of the meta data locks in order to preserve locking
1832 * order in case of page fault.
1833 */
1834 ret = get_user_pages(current, current->mm,
1835 (unsigned long)buf & PAGE_CACHE_MASK, 1,
1836 0, 0, &src_page, NULL);
1837 if (ret == 1)
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001838 *ret_src_buf = kmap(src_page) + off;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001839 else
1840 src_page = ERR_PTR(-EFAULT);
1841 } else {
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001842 *ret_src_buf = buf;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001843 }
1844
1845 return src_page;
1846}
1847
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001848static void ocfs2_put_write_source(struct page *page)
Mark Fasheh9517bac2007-02-09 20:24:12 -08001849{
1850 if (page) {
1851 kunmap(page);
1852 page_cache_release(page);
1853 }
1854}
1855
1856static ssize_t ocfs2_file_buffered_write(struct file *file, loff_t *ppos,
1857 const struct iovec *iov,
1858 unsigned long nr_segs,
1859 size_t count,
1860 ssize_t o_direct_written)
1861{
1862 int ret = 0;
1863 ssize_t copied, total = 0;
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001864 size_t iov_offset = 0, bytes;
1865 loff_t pos;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001866 const struct iovec *cur_iov = iov;
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001867 struct page *user_page, *page;
Jeff Garzik8e1c0912007-07-17 05:40:59 -04001868 char * uninitialized_var(buf);
1869 char *dst;
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001870 void *fsdata;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001871
1872 /*
1873 * handle partial DIO write. Adjust cur_iov if needed.
1874 */
1875 ocfs2_set_next_iovec(&cur_iov, &iov_offset, o_direct_written);
1876
1877 do {
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001878 pos = *ppos;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001879
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001880 user_page = ocfs2_get_write_source(&buf, cur_iov, iov_offset);
1881 if (IS_ERR(user_page)) {
1882 ret = PTR_ERR(user_page);
Mark Fasheh9517bac2007-02-09 20:24:12 -08001883 goto out;
1884 }
1885
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001886 /* Stay within our page boundaries */
1887 bytes = min((PAGE_CACHE_SIZE - ((unsigned long)pos & ~PAGE_CACHE_MASK)),
1888 (PAGE_CACHE_SIZE - ((unsigned long)buf & ~PAGE_CACHE_MASK)));
1889 /* Stay within the vector boundary */
1890 bytes = min_t(size_t, bytes, cur_iov->iov_len - iov_offset);
1891 /* Stay within count */
1892 bytes = min(bytes, count);
Mark Fasheh9517bac2007-02-09 20:24:12 -08001893
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001894 page = NULL;
1895 ret = ocfs2_write_begin(file, file->f_mapping, pos, bytes, 0,
1896 &page, &fsdata);
1897 if (ret) {
1898 mlog_errno(ret);
1899 goto out;
1900 }
Mark Fasheh9517bac2007-02-09 20:24:12 -08001901
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001902 dst = kmap_atomic(page, KM_USER0);
Mark Fasheh7c08d702007-07-20 11:58:36 -07001903 memcpy(dst + (pos & (loff_t)(PAGE_CACHE_SIZE - 1)), buf, bytes);
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001904 kunmap_atomic(dst, KM_USER0);
1905 flush_dcache_page(page);
1906 ocfs2_put_write_source(user_page);
1907
1908 copied = ocfs2_write_end(file, file->f_mapping, pos, bytes,
1909 bytes, page, fsdata);
Mark Fasheh9517bac2007-02-09 20:24:12 -08001910 if (copied < 0) {
1911 mlog_errno(copied);
1912 ret = copied;
1913 goto out;
1914 }
1915
1916 total += copied;
Mark Fasheh3a307ff2007-05-08 17:47:32 -07001917 *ppos = pos + copied;
Mark Fasheh9517bac2007-02-09 20:24:12 -08001918 count -= copied;
1919
1920 ocfs2_set_next_iovec(&cur_iov, &iov_offset, copied);
1921 } while(count);
1922
1923out:
1924 return total ? total : ret;
1925}
1926
Tiger Yang8659ac22006-10-17 18:29:52 -07001927static ssize_t ocfs2_file_aio_write(struct kiocb *iocb,
1928 const struct iovec *iov,
1929 unsigned long nr_segs,
1930 loff_t pos)
1931{
Mark Fasheh9517bac2007-02-09 20:24:12 -08001932 int ret, direct_io, appending, rw_level, have_alloc_sem = 0;
1933 int can_do_direct, sync = 0;
1934 ssize_t written = 0;
1935 size_t ocount; /* original count */
1936 size_t count; /* after file limit checks */
1937 loff_t *ppos = &iocb->ki_pos;
1938 struct file *file = iocb->ki_filp;
1939 struct inode *inode = file->f_path.dentry->d_inode;
Tiger Yang8659ac22006-10-17 18:29:52 -07001940
Mark Fasheh9517bac2007-02-09 20:24:12 -08001941 mlog_entry("(0x%p, %u, '%.*s')\n", file,
Tiger Yang8659ac22006-10-17 18:29:52 -07001942 (unsigned int)nr_segs,
Mark Fasheh9517bac2007-02-09 20:24:12 -08001943 file->f_path.dentry->d_name.len,
1944 file->f_path.dentry->d_name.name);
Tiger Yang8659ac22006-10-17 18:29:52 -07001945
Tiger Yang8659ac22006-10-17 18:29:52 -07001946 if (iocb->ki_left == 0)
1947 return 0;
1948
Christoph Hellwigd9b08b92007-05-18 13:12:40 +02001949 ret = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
Mark Fasheh9517bac2007-02-09 20:24:12 -08001950 if (ret)
1951 return ret;
1952
1953 count = ocount;
1954
1955 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
1956
1957 appending = file->f_flags & O_APPEND ? 1 : 0;
1958 direct_io = file->f_flags & O_DIRECT ? 1 : 0;
1959
Tiger Yang8659ac22006-10-17 18:29:52 -07001960 mutex_lock(&inode->i_mutex);
Mark Fasheh9517bac2007-02-09 20:24:12 -08001961
1962relock:
Tiger Yang8659ac22006-10-17 18:29:52 -07001963 /* to match setattr's i_mutex -> i_alloc_sem -> rw_lock ordering */
Mark Fasheh9517bac2007-02-09 20:24:12 -08001964 if (direct_io) {
Tiger Yang8659ac22006-10-17 18:29:52 -07001965 down_read(&inode->i_alloc_sem);
Mark Fasheh9517bac2007-02-09 20:24:12 -08001966 have_alloc_sem = 1;
Tiger Yang8659ac22006-10-17 18:29:52 -07001967 }
1968
1969 /* concurrent O_DIRECT writes are allowed */
Mark Fasheh9517bac2007-02-09 20:24:12 -08001970 rw_level = !direct_io;
Tiger Yang8659ac22006-10-17 18:29:52 -07001971 ret = ocfs2_rw_lock(inode, rw_level);
1972 if (ret < 0) {
Mark Fasheh9517bac2007-02-09 20:24:12 -08001973 mlog_errno(ret);
1974 goto out_sems;
1975 }
1976
1977 can_do_direct = direct_io;
1978 ret = ocfs2_prepare_inode_for_write(file->f_path.dentry, ppos,
1979 iocb->ki_left, appending,
1980 &can_do_direct);
1981 if (ret < 0) {
Tiger Yang8659ac22006-10-17 18:29:52 -07001982 mlog_errno(ret);
1983 goto out;
1984 }
1985
Mark Fasheh9517bac2007-02-09 20:24:12 -08001986 /*
1987 * We can't complete the direct I/O as requested, fall back to
1988 * buffered I/O.
1989 */
1990 if (direct_io && !can_do_direct) {
1991 ocfs2_rw_unlock(inode, rw_level);
1992 up_read(&inode->i_alloc_sem);
1993
1994 have_alloc_sem = 0;
1995 rw_level = -1;
1996
1997 direct_io = 0;
1998 sync = 1;
1999 goto relock;
Tiger Yang8659ac22006-10-17 18:29:52 -07002000 }
Mark Fashehccd979b2005-12-15 14:31:24 -08002001
Mark Fasheh9517bac2007-02-09 20:24:12 -08002002 if (!sync && ((file->f_flags & O_SYNC) || IS_SYNC(inode)))
2003 sync = 1;
2004
2005 /*
2006 * XXX: Is it ok to execute these checks a second time?
2007 */
2008 ret = generic_write_checks(file, ppos, &count, S_ISBLK(inode->i_mode));
2009 if (ret)
2010 goto out;
2011
2012 /*
2013 * Set pos so that sync_page_range_nolock() below understands
2014 * where to start from. We might've moved it around via the
2015 * calls above. The range we want to actually sync starts from
2016 * *ppos here.
2017 *
2018 */
2019 pos = *ppos;
2020
Mark Fashehccd979b2005-12-15 14:31:24 -08002021 /* communicate with ocfs2_dio_end_io */
Mark Fasheh7cdfc3a2007-04-16 17:28:51 -07002022 ocfs2_iocb_set_rw_locked(iocb, rw_level);
Mark Fashehccd979b2005-12-15 14:31:24 -08002023
Mark Fasheh9517bac2007-02-09 20:24:12 -08002024 if (direct_io) {
2025 written = generic_file_direct_write(iocb, iov, &nr_segs, *ppos,
2026 ppos, count, ocount);
2027 if (written < 0) {
2028 ret = written;
2029 goto out_dio;
2030 }
2031 } else {
2032 written = ocfs2_file_buffered_write(file, ppos, iov, nr_segs,
2033 count, written);
2034 if (written < 0) {
2035 ret = written;
2036 if (ret != -EFAULT || ret != -ENOSPC)
2037 mlog_errno(ret);
2038 goto out;
2039 }
2040 }
Mark Fashehccd979b2005-12-15 14:31:24 -08002041
Mark Fasheh9517bac2007-02-09 20:24:12 -08002042out_dio:
Mark Fashehccd979b2005-12-15 14:31:24 -08002043 /* buffered aio wouldn't have proper lock coverage today */
Mark Fasheh9517bac2007-02-09 20:24:12 -08002044 BUG_ON(ret == -EIOCBQUEUED && !(file->f_flags & O_DIRECT));
Mark Fashehccd979b2005-12-15 14:31:24 -08002045
2046 /*
2047 * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
2048 * function pointer which is called when o_direct io completes so that
2049 * it can unlock our rw lock. (it's the clustered equivalent of
2050 * i_alloc_sem; protects truncate from racing with pending ios).
2051 * Unfortunately there are error cases which call end_io and others
2052 * that don't. so we don't have to unlock the rw_lock if either an
2053 * async dio is going to do it in the future or an end_io after an
2054 * error has already done it.
2055 */
2056 if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2057 rw_level = -1;
2058 have_alloc_sem = 0;
2059 }
2060
2061out:
Mark Fasheh9517bac2007-02-09 20:24:12 -08002062 if (rw_level != -1)
2063 ocfs2_rw_unlock(inode, rw_level);
2064
2065out_sems:
Mark Fashehccd979b2005-12-15 14:31:24 -08002066 if (have_alloc_sem)
2067 up_read(&inode->i_alloc_sem);
Mark Fasheh9517bac2007-02-09 20:24:12 -08002068
2069 if (written > 0 && sync) {
2070 ssize_t err;
2071
2072 err = sync_page_range_nolock(inode, file->f_mapping, pos, count);
2073 if (err < 0)
2074 written = err;
2075 }
2076
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002077 mutex_unlock(&inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -08002078
2079 mlog_exit(ret);
Mark Fasheh9517bac2007-02-09 20:24:12 -08002080 return written ? written : ret;
Mark Fashehccd979b2005-12-15 14:31:24 -08002081}
2082
Mark Fasheh6af67d82007-03-06 17:24:46 -08002083static int ocfs2_splice_write_actor(struct pipe_inode_info *pipe,
2084 struct pipe_buffer *buf,
2085 struct splice_desc *sd)
2086{
Mark Fasheh3a307ff2007-05-08 17:47:32 -07002087 int ret, count;
Mark Fasheh6af67d82007-03-06 17:24:46 -08002088 ssize_t copied = 0;
Mark Fasheh3a307ff2007-05-08 17:47:32 -07002089 struct file *file = sd->u.file;
2090 unsigned int offset;
2091 struct page *page = NULL;
2092 void *fsdata;
2093 char *src, *dst;
Mark Fasheh6af67d82007-03-06 17:24:46 -08002094
Jens Axboecac36bb02007-06-14 13:10:48 +02002095 ret = buf->ops->confirm(pipe, buf);
Mark Fasheh6af67d82007-03-06 17:24:46 -08002096 if (ret)
2097 goto out;
2098
Mark Fasheh3a307ff2007-05-08 17:47:32 -07002099 offset = sd->pos & ~PAGE_CACHE_MASK;
Mark Fasheh6af67d82007-03-06 17:24:46 -08002100 count = sd->len;
Mark Fasheh3a307ff2007-05-08 17:47:32 -07002101 if (count + offset > PAGE_CACHE_SIZE)
2102 count = PAGE_CACHE_SIZE - offset;
Mark Fasheh6af67d82007-03-06 17:24:46 -08002103
Mark Fasheh3a307ff2007-05-08 17:47:32 -07002104 ret = ocfs2_write_begin(file, file->f_mapping, sd->pos, count, 0,
2105 &page, &fsdata);
2106 if (ret) {
2107 mlog_errno(ret);
2108 goto out;
2109 }
Mark Fasheh6af67d82007-03-06 17:24:46 -08002110
Mark Fasheh3a307ff2007-05-08 17:47:32 -07002111 src = buf->ops->map(pipe, buf, 1);
2112 dst = kmap_atomic(page, KM_USER1);
2113 memcpy(dst + offset, src + buf->offset, count);
Jens Axboe3836df62007-07-24 10:17:50 +02002114 kunmap_atomic(dst, KM_USER1);
Mark Fasheh3a307ff2007-05-08 17:47:32 -07002115 buf->ops->unmap(pipe, buf, src);
Mark Fasheh6af67d82007-03-06 17:24:46 -08002116
Mark Fasheh3a307ff2007-05-08 17:47:32 -07002117 copied = ocfs2_write_end(file, file->f_mapping, sd->pos, count, count,
2118 page, fsdata);
2119 if (copied < 0) {
2120 mlog_errno(copied);
2121 ret = copied;
2122 goto out;
2123 }
Mark Fasheh6af67d82007-03-06 17:24:46 -08002124out:
2125
Mark Fasheh3a307ff2007-05-08 17:47:32 -07002126 return copied ? copied : ret;
Mark Fasheh6af67d82007-03-06 17:24:46 -08002127}
2128
2129static ssize_t __ocfs2_file_splice_write(struct pipe_inode_info *pipe,
2130 struct file *out,
2131 loff_t *ppos,
2132 size_t len,
2133 unsigned int flags)
2134{
2135 int ret, err;
2136 struct address_space *mapping = out->f_mapping;
2137 struct inode *inode = mapping->host;
Jens Axboec66ab6f2007-06-12 21:17:17 +02002138 struct splice_desc sd = {
2139 .total_len = len,
2140 .flags = flags,
2141 .pos = *ppos,
Jens Axboe6a14b902007-06-14 13:08:55 +02002142 .u.file = out,
Jens Axboec66ab6f2007-06-12 21:17:17 +02002143 };
Mark Fasheh6af67d82007-03-06 17:24:46 -08002144
Jens Axboec66ab6f2007-06-12 21:17:17 +02002145 ret = __splice_from_pipe(pipe, &sd, ocfs2_splice_write_actor);
Mark Fasheh6af67d82007-03-06 17:24:46 -08002146 if (ret > 0) {
2147 *ppos += ret;
2148
2149 if (unlikely((out->f_flags & O_SYNC) || IS_SYNC(inode))) {
2150 err = generic_osync_inode(inode, mapping,
2151 OSYNC_METADATA|OSYNC_DATA);
2152 if (err)
2153 ret = err;
2154 }
2155 }
2156
2157 return ret;
2158}
2159
Tiger Yang8659ac22006-10-17 18:29:52 -07002160static ssize_t ocfs2_file_splice_write(struct pipe_inode_info *pipe,
2161 struct file *out,
2162 loff_t *ppos,
2163 size_t len,
2164 unsigned int flags)
2165{
2166 int ret;
Josef Sipekd28c9172006-12-08 02:37:25 -08002167 struct inode *inode = out->f_path.dentry->d_inode;
Tiger Yang8659ac22006-10-17 18:29:52 -07002168
2169 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", out, pipe,
2170 (unsigned int)len,
Josef Sipekd28c9172006-12-08 02:37:25 -08002171 out->f_path.dentry->d_name.len,
2172 out->f_path.dentry->d_name.name);
Tiger Yang8659ac22006-10-17 18:29:52 -07002173
2174 inode_double_lock(inode, pipe->inode);
2175
2176 ret = ocfs2_rw_lock(inode, 1);
2177 if (ret < 0) {
2178 mlog_errno(ret);
2179 goto out;
2180 }
2181
Mark Fasheh9517bac2007-02-09 20:24:12 -08002182 ret = ocfs2_prepare_inode_for_write(out->f_path.dentry, ppos, len, 0,
2183 NULL);
Tiger Yang8659ac22006-10-17 18:29:52 -07002184 if (ret < 0) {
2185 mlog_errno(ret);
2186 goto out_unlock;
2187 }
2188
2189 /* ok, we're done with i_size and alloc work */
Mark Fasheh6af67d82007-03-06 17:24:46 -08002190 ret = __ocfs2_file_splice_write(pipe, out, ppos, len, flags);
Tiger Yang8659ac22006-10-17 18:29:52 -07002191
2192out_unlock:
2193 ocfs2_rw_unlock(inode, 1);
2194out:
2195 inode_double_unlock(inode, pipe->inode);
2196
2197 mlog_exit(ret);
2198 return ret;
2199}
2200
2201static ssize_t ocfs2_file_splice_read(struct file *in,
2202 loff_t *ppos,
2203 struct pipe_inode_info *pipe,
2204 size_t len,
2205 unsigned int flags)
2206{
2207 int ret = 0;
Josef Sipekd28c9172006-12-08 02:37:25 -08002208 struct inode *inode = in->f_path.dentry->d_inode;
Tiger Yang8659ac22006-10-17 18:29:52 -07002209
2210 mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", in, pipe,
2211 (unsigned int)len,
Josef Sipekd28c9172006-12-08 02:37:25 -08002212 in->f_path.dentry->d_name.len,
2213 in->f_path.dentry->d_name.name);
Tiger Yang8659ac22006-10-17 18:29:52 -07002214
2215 /*
2216 * See the comment in ocfs2_file_aio_read()
2217 */
2218 ret = ocfs2_meta_lock(inode, NULL, 0);
2219 if (ret < 0) {
2220 mlog_errno(ret);
2221 goto bail;
2222 }
2223 ocfs2_meta_unlock(inode, 0);
2224
2225 ret = generic_file_splice_read(in, ppos, pipe, len, flags);
2226
2227bail:
2228 mlog_exit(ret);
2229 return ret;
2230}
2231
Mark Fashehccd979b2005-12-15 14:31:24 -08002232static ssize_t ocfs2_file_aio_read(struct kiocb *iocb,
Badari Pulavarty027445c2006-09-30 23:28:46 -07002233 const struct iovec *iov,
2234 unsigned long nr_segs,
Mark Fashehccd979b2005-12-15 14:31:24 -08002235 loff_t pos)
2236{
Tiger Yang25899de2006-11-15 15:49:02 +08002237 int ret = 0, rw_level = -1, have_alloc_sem = 0, lock_level = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -08002238 struct file *filp = iocb->ki_filp;
Josef Sipekd28c9172006-12-08 02:37:25 -08002239 struct inode *inode = filp->f_path.dentry->d_inode;
Mark Fashehccd979b2005-12-15 14:31:24 -08002240
Badari Pulavarty027445c2006-09-30 23:28:46 -07002241 mlog_entry("(0x%p, %u, '%.*s')\n", filp,
2242 (unsigned int)nr_segs,
Josef Sipekd28c9172006-12-08 02:37:25 -08002243 filp->f_path.dentry->d_name.len,
2244 filp->f_path.dentry->d_name.name);
Mark Fashehccd979b2005-12-15 14:31:24 -08002245
2246 if (!inode) {
2247 ret = -EINVAL;
2248 mlog_errno(ret);
2249 goto bail;
2250 }
2251
Mark Fashehccd979b2005-12-15 14:31:24 -08002252 /*
2253 * buffered reads protect themselves in ->readpage(). O_DIRECT reads
2254 * need locks to protect pending reads from racing with truncate.
2255 */
2256 if (filp->f_flags & O_DIRECT) {
2257 down_read(&inode->i_alloc_sem);
2258 have_alloc_sem = 1;
2259
2260 ret = ocfs2_rw_lock(inode, 0);
2261 if (ret < 0) {
2262 mlog_errno(ret);
2263 goto bail;
2264 }
2265 rw_level = 0;
2266 /* communicate with ocfs2_dio_end_io */
Mark Fasheh7cdfc3a2007-04-16 17:28:51 -07002267 ocfs2_iocb_set_rw_locked(iocb, rw_level);
Mark Fashehccd979b2005-12-15 14:31:24 -08002268 }
2269
Mark Fashehc4374f82006-05-05 19:04:35 -07002270 /*
2271 * We're fine letting folks race truncates and extending
2272 * writes with read across the cluster, just like they can
2273 * locally. Hence no rw_lock during read.
2274 *
2275 * Take and drop the meta data lock to update inode fields
2276 * like i_size. This allows the checks down below
2277 * generic_file_aio_read() a chance of actually working.
2278 */
Tiger Yang25899de2006-11-15 15:49:02 +08002279 ret = ocfs2_meta_lock_atime(inode, filp->f_vfsmnt, &lock_level);
Mark Fashehc4374f82006-05-05 19:04:35 -07002280 if (ret < 0) {
2281 mlog_errno(ret);
2282 goto bail;
2283 }
Tiger Yang25899de2006-11-15 15:49:02 +08002284 ocfs2_meta_unlock(inode, lock_level);
Mark Fashehc4374f82006-05-05 19:04:35 -07002285
Badari Pulavarty027445c2006-09-30 23:28:46 -07002286 ret = generic_file_aio_read(iocb, iov, nr_segs, iocb->ki_pos);
Mark Fashehccd979b2005-12-15 14:31:24 -08002287 if (ret == -EINVAL)
2288 mlog(ML_ERROR, "generic_file_aio_read returned -EINVAL\n");
2289
2290 /* buffered aio wouldn't have proper lock coverage today */
2291 BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT));
2292
2293 /* see ocfs2_file_aio_write */
2294 if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2295 rw_level = -1;
2296 have_alloc_sem = 0;
2297 }
2298
2299bail:
2300 if (have_alloc_sem)
2301 up_read(&inode->i_alloc_sem);
2302 if (rw_level != -1)
2303 ocfs2_rw_unlock(inode, rw_level);
2304 mlog_exit(ret);
2305
2306 return ret;
2307}
2308
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002309const struct inode_operations ocfs2_file_iops = {
Mark Fashehccd979b2005-12-15 14:31:24 -08002310 .setattr = ocfs2_setattr,
2311 .getattr = ocfs2_getattr,
Tiger Yangd38eb8d2006-11-27 09:59:21 +08002312 .permission = ocfs2_permission,
Mark Fasheh385820a2007-07-19 00:14:38 -07002313 .fallocate = ocfs2_fallocate,
Mark Fashehccd979b2005-12-15 14:31:24 -08002314};
2315
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002316const struct inode_operations ocfs2_special_file_iops = {
Mark Fashehccd979b2005-12-15 14:31:24 -08002317 .setattr = ocfs2_setattr,
2318 .getattr = ocfs2_getattr,
Tiger Yangd38eb8d2006-11-27 09:59:21 +08002319 .permission = ocfs2_permission,
Mark Fashehccd979b2005-12-15 14:31:24 -08002320};
2321
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08002322const struct file_operations ocfs2_fops = {
Mark Fashehccd979b2005-12-15 14:31:24 -08002323 .read = do_sync_read,
2324 .write = do_sync_write,
Mark Fashehccd979b2005-12-15 14:31:24 -08002325 .mmap = ocfs2_mmap,
2326 .fsync = ocfs2_sync_file,
2327 .release = ocfs2_file_release,
2328 .open = ocfs2_file_open,
2329 .aio_read = ocfs2_file_aio_read,
2330 .aio_write = ocfs2_file_aio_write,
Herbert Poetzlca4d1472006-07-03 17:27:12 -07002331 .ioctl = ocfs2_ioctl,
Mark Fasheh586d2322007-03-09 15:56:28 -08002332#ifdef CONFIG_COMPAT
2333 .compat_ioctl = ocfs2_compat_ioctl,
2334#endif
Tiger Yang8659ac22006-10-17 18:29:52 -07002335 .splice_read = ocfs2_file_splice_read,
2336 .splice_write = ocfs2_file_splice_write,
Mark Fashehccd979b2005-12-15 14:31:24 -08002337};
2338
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08002339const struct file_operations ocfs2_dops = {
Mark Fashehccd979b2005-12-15 14:31:24 -08002340 .read = generic_read_dir,
2341 .readdir = ocfs2_readdir,
2342 .fsync = ocfs2_sync_file,
Herbert Poetzlca4d1472006-07-03 17:27:12 -07002343 .ioctl = ocfs2_ioctl,
Mark Fasheh586d2322007-03-09 15:56:28 -08002344#ifdef CONFIG_COMPAT
2345 .compat_ioctl = ocfs2_compat_ioctl,
2346#endif
Mark Fashehccd979b2005-12-15 14:31:24 -08002347};