ocfs2: fix journal commit deadlock
[firefly-linux-kernel-4.4.55.git] / fs / ocfs2 / file.c
1 /* -*- 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
26 #include <linux/capability.h>
27 #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>
33 #include <linux/sched.h>
34 #include <linux/splice.h>
35 #include <linux/mount.h>
36 #include <linux/writeback.h>
37 #include <linux/falloc.h>
38 #include <linux/quotaops.h>
39 #include <linux/blkdev.h>
40
41 #include <cluster/masklog.h>
42
43 #include "ocfs2.h"
44
45 #include "alloc.h"
46 #include "aops.h"
47 #include "dir.h"
48 #include "dlmglue.h"
49 #include "extent_map.h"
50 #include "file.h"
51 #include "sysfile.h"
52 #include "inode.h"
53 #include "ioctl.h"
54 #include "journal.h"
55 #include "locks.h"
56 #include "mmap.h"
57 #include "suballoc.h"
58 #include "super.h"
59 #include "xattr.h"
60 #include "acl.h"
61 #include "quota.h"
62 #include "refcounttree.h"
63 #include "ocfs2_trace.h"
64
65 #include "buffer_head_io.h"
66
67 static 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
83 static 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
96 static 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         trace_ocfs2_file_open(inode, file, file->f_path.dentry,
103                               (unsigned long long)OCFS2_I(inode)->ip_blkno,
104                               file->f_path.dentry->d_name.len,
105                               file->f_path.dentry->d_name.name, mode);
106
107         if (file->f_mode & FMODE_WRITE)
108                 dquot_initialize(inode);
109
110         spin_lock(&oi->ip_lock);
111
112         /* Check that the inode hasn't been wiped from disk by another
113          * node. If it hasn't then we're safe as long as we hold the
114          * spin lock until our increment of open count. */
115         if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
116                 spin_unlock(&oi->ip_lock);
117
118                 status = -ENOENT;
119                 goto leave;
120         }
121
122         if (mode & O_DIRECT)
123                 oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT;
124
125         oi->ip_open_count++;
126         spin_unlock(&oi->ip_lock);
127
128         status = ocfs2_init_file_private(inode, file);
129         if (status) {
130                 /*
131                  * We want to set open count back if we're failing the
132                  * open.
133                  */
134                 spin_lock(&oi->ip_lock);
135                 oi->ip_open_count--;
136                 spin_unlock(&oi->ip_lock);
137         }
138
139 leave:
140         return status;
141 }
142
143 static int ocfs2_file_release(struct inode *inode, struct file *file)
144 {
145         struct ocfs2_inode_info *oi = OCFS2_I(inode);
146
147         spin_lock(&oi->ip_lock);
148         if (!--oi->ip_open_count)
149                 oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT;
150
151         trace_ocfs2_file_release(inode, file, file->f_path.dentry,
152                                  oi->ip_blkno,
153                                  file->f_path.dentry->d_name.len,
154                                  file->f_path.dentry->d_name.name,
155                                  oi->ip_open_count);
156         spin_unlock(&oi->ip_lock);
157
158         ocfs2_free_file_private(inode, file);
159
160         return 0;
161 }
162
163 static int ocfs2_dir_open(struct inode *inode, struct file *file)
164 {
165         return ocfs2_init_file_private(inode, file);
166 }
167
168 static int ocfs2_dir_release(struct inode *inode, struct file *file)
169 {
170         ocfs2_free_file_private(inode, file);
171         return 0;
172 }
173
174 static int ocfs2_sync_file(struct file *file, loff_t start, loff_t end,
175                            int datasync)
176 {
177         int err = 0;
178         journal_t *journal;
179         struct inode *inode = file->f_mapping->host;
180         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
181
182         trace_ocfs2_sync_file(inode, file, file->f_path.dentry,
183                               OCFS2_I(inode)->ip_blkno,
184                               file->f_path.dentry->d_name.len,
185                               file->f_path.dentry->d_name.name,
186                               (unsigned long long)datasync);
187
188         err = filemap_write_and_wait_range(inode->i_mapping, start, end);
189         if (err)
190                 return err;
191
192         /*
193          * Probably don't need the i_mutex at all in here, just putting it here
194          * to be consistent with how fsync used to be called, someone more
195          * familiar with the fs could possibly remove it.
196          */
197         mutex_lock(&inode->i_mutex);
198         if (datasync && !(inode->i_state & I_DIRTY_DATASYNC)) {
199                 /*
200                  * We still have to flush drive's caches to get data to the
201                  * platter
202                  */
203                 if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER)
204                         blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, NULL);
205                 goto bail;
206         }
207
208         journal = osb->journal->j_journal;
209         err = jbd2_journal_force_commit(journal);
210
211 bail:
212         if (err)
213                 mlog_errno(err);
214         mutex_unlock(&inode->i_mutex);
215
216         return (err < 0) ? -EIO : 0;
217 }
218
219 int ocfs2_should_update_atime(struct inode *inode,
220                               struct vfsmount *vfsmnt)
221 {
222         struct timespec now;
223         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
224
225         if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
226                 return 0;
227
228         if ((inode->i_flags & S_NOATIME) ||
229             ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)))
230                 return 0;
231
232         /*
233          * We can be called with no vfsmnt structure - NFSD will
234          * sometimes do this.
235          *
236          * Note that our action here is different than touch_atime() -
237          * if we can't tell whether this is a noatime mount, then we
238          * don't know whether to trust the value of s_atime_quantum.
239          */
240         if (vfsmnt == NULL)
241                 return 0;
242
243         if ((vfsmnt->mnt_flags & MNT_NOATIME) ||
244             ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
245                 return 0;
246
247         if (vfsmnt->mnt_flags & MNT_RELATIME) {
248                 if ((timespec_compare(&inode->i_atime, &inode->i_mtime) <= 0) ||
249                     (timespec_compare(&inode->i_atime, &inode->i_ctime) <= 0))
250                         return 1;
251
252                 return 0;
253         }
254
255         now = CURRENT_TIME;
256         if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum))
257                 return 0;
258         else
259                 return 1;
260 }
261
262 int ocfs2_update_inode_atime(struct inode *inode,
263                              struct buffer_head *bh)
264 {
265         int ret;
266         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
267         handle_t *handle;
268         struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data;
269
270         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
271         if (IS_ERR(handle)) {
272                 ret = PTR_ERR(handle);
273                 mlog_errno(ret);
274                 goto out;
275         }
276
277         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
278                                       OCFS2_JOURNAL_ACCESS_WRITE);
279         if (ret) {
280                 mlog_errno(ret);
281                 goto out_commit;
282         }
283
284         /*
285          * Don't use ocfs2_mark_inode_dirty() here as we don't always
286          * have i_mutex to guard against concurrent changes to other
287          * inode fields.
288          */
289         inode->i_atime = CURRENT_TIME;
290         di->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
291         di->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
292         ocfs2_journal_dirty(handle, bh);
293
294 out_commit:
295         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
296 out:
297         return ret;
298 }
299
300 static int ocfs2_set_inode_size(handle_t *handle,
301                                 struct inode *inode,
302                                 struct buffer_head *fe_bh,
303                                 u64 new_i_size)
304 {
305         int status;
306
307         i_size_write(inode, new_i_size);
308         inode->i_blocks = ocfs2_inode_sector_count(inode);
309         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
310
311         status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
312         if (status < 0) {
313                 mlog_errno(status);
314                 goto bail;
315         }
316
317 bail:
318         return status;
319 }
320
321 int ocfs2_simple_size_update(struct inode *inode,
322                              struct buffer_head *di_bh,
323                              u64 new_i_size)
324 {
325         int ret;
326         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
327         handle_t *handle = NULL;
328
329         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
330         if (IS_ERR(handle)) {
331                 ret = PTR_ERR(handle);
332                 mlog_errno(ret);
333                 goto out;
334         }
335
336         ret = ocfs2_set_inode_size(handle, inode, di_bh,
337                                    new_i_size);
338         if (ret < 0)
339                 mlog_errno(ret);
340
341         ocfs2_commit_trans(osb, handle);
342 out:
343         return ret;
344 }
345
346 static int ocfs2_cow_file_pos(struct inode *inode,
347                               struct buffer_head *fe_bh,
348                               u64 offset)
349 {
350         int status;
351         u32 phys, cpos = offset >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
352         unsigned int num_clusters = 0;
353         unsigned int ext_flags = 0;
354
355         /*
356          * If the new offset is aligned to the range of the cluster, there is
357          * no space for ocfs2_zero_range_for_truncate to fill, so no need to
358          * CoW either.
359          */
360         if ((offset & (OCFS2_SB(inode->i_sb)->s_clustersize - 1)) == 0)
361                 return 0;
362
363         status = ocfs2_get_clusters(inode, cpos, &phys,
364                                     &num_clusters, &ext_flags);
365         if (status) {
366                 mlog_errno(status);
367                 goto out;
368         }
369
370         if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
371                 goto out;
372
373         return ocfs2_refcount_cow(inode, NULL, fe_bh, cpos, 1, cpos+1);
374
375 out:
376         return status;
377 }
378
379 static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
380                                      struct inode *inode,
381                                      struct buffer_head *fe_bh,
382                                      u64 new_i_size)
383 {
384         int status;
385         handle_t *handle;
386         struct ocfs2_dinode *di;
387         u64 cluster_bytes;
388
389         /*
390          * We need to CoW the cluster contains the offset if it is reflinked
391          * since we will call ocfs2_zero_range_for_truncate later which will
392          * write "0" from offset to the end of the cluster.
393          */
394         status = ocfs2_cow_file_pos(inode, fe_bh, new_i_size);
395         if (status) {
396                 mlog_errno(status);
397                 return status;
398         }
399
400         /* TODO: This needs to actually orphan the inode in this
401          * transaction. */
402
403         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
404         if (IS_ERR(handle)) {
405                 status = PTR_ERR(handle);
406                 mlog_errno(status);
407                 goto out;
408         }
409
410         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
411                                          OCFS2_JOURNAL_ACCESS_WRITE);
412         if (status < 0) {
413                 mlog_errno(status);
414                 goto out_commit;
415         }
416
417         /*
418          * Do this before setting i_size.
419          */
420         cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size);
421         status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size,
422                                                cluster_bytes);
423         if (status) {
424                 mlog_errno(status);
425                 goto out_commit;
426         }
427
428         i_size_write(inode, new_i_size);
429         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
430
431         di = (struct ocfs2_dinode *) fe_bh->b_data;
432         di->i_size = cpu_to_le64(new_i_size);
433         di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
434         di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
435
436         ocfs2_journal_dirty(handle, fe_bh);
437
438 out_commit:
439         ocfs2_commit_trans(osb, handle);
440 out:
441         return status;
442 }
443
444 static int ocfs2_truncate_file(struct inode *inode,
445                                struct buffer_head *di_bh,
446                                u64 new_i_size)
447 {
448         int status = 0;
449         struct ocfs2_dinode *fe = NULL;
450         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
451
452         /* We trust di_bh because it comes from ocfs2_inode_lock(), which
453          * already validated it */
454         fe = (struct ocfs2_dinode *) di_bh->b_data;
455
456         trace_ocfs2_truncate_file((unsigned long long)OCFS2_I(inode)->ip_blkno,
457                                   (unsigned long long)le64_to_cpu(fe->i_size),
458                                   (unsigned long long)new_i_size);
459
460         mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode),
461                         "Inode %llu, inode i_size = %lld != di "
462                         "i_size = %llu, i_flags = 0x%x\n",
463                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
464                         i_size_read(inode),
465                         (unsigned long long)le64_to_cpu(fe->i_size),
466                         le32_to_cpu(fe->i_flags));
467
468         if (new_i_size > le64_to_cpu(fe->i_size)) {
469                 trace_ocfs2_truncate_file_error(
470                         (unsigned long long)le64_to_cpu(fe->i_size),
471                         (unsigned long long)new_i_size);
472                 status = -EINVAL;
473                 mlog_errno(status);
474                 goto bail;
475         }
476
477         /* lets handle the simple truncate cases before doing any more
478          * cluster locking. */
479         if (new_i_size == le64_to_cpu(fe->i_size))
480                 goto bail;
481
482         down_write(&OCFS2_I(inode)->ip_alloc_sem);
483
484         ocfs2_resv_discard(&osb->osb_la_resmap,
485                            &OCFS2_I(inode)->ip_la_data_resv);
486
487         /*
488          * The inode lock forced other nodes to sync and drop their
489          * pages, which (correctly) happens even if we have a truncate
490          * without allocation change - ocfs2 cluster sizes can be much
491          * greater than page size, so we have to truncate them
492          * anyway.
493          */
494         unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1);
495         truncate_inode_pages(inode->i_mapping, new_i_size);
496
497         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
498                 status = ocfs2_truncate_inline(inode, di_bh, new_i_size,
499                                                i_size_read(inode), 1);
500                 if (status)
501                         mlog_errno(status);
502
503                 goto bail_unlock_sem;
504         }
505
506         /* alright, we're going to need to do a full blown alloc size
507          * change. Orphan the inode so that recovery can complete the
508          * truncate if necessary. This does the task of marking
509          * i_size. */
510         status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
511         if (status < 0) {
512                 mlog_errno(status);
513                 goto bail_unlock_sem;
514         }
515
516         status = ocfs2_commit_truncate(osb, inode, di_bh);
517         if (status < 0) {
518                 mlog_errno(status);
519                 goto bail_unlock_sem;
520         }
521
522         /* TODO: orphan dir cleanup here. */
523 bail_unlock_sem:
524         up_write(&OCFS2_I(inode)->ip_alloc_sem);
525
526 bail:
527         if (!status && OCFS2_I(inode)->ip_clusters == 0)
528                 status = ocfs2_try_remove_refcount_tree(inode, di_bh);
529
530         return status;
531 }
532
533 /*
534  * extend file allocation only here.
535  * we'll update all the disk stuff, and oip->alloc_size
536  *
537  * expect stuff to be locked, a transaction started and enough data /
538  * metadata reservations in the contexts.
539  *
540  * Will return -EAGAIN, and a reason if a restart is needed.
541  * If passed in, *reason will always be set, even in error.
542  */
543 int ocfs2_add_inode_data(struct ocfs2_super *osb,
544                          struct inode *inode,
545                          u32 *logical_offset,
546                          u32 clusters_to_add,
547                          int mark_unwritten,
548                          struct buffer_head *fe_bh,
549                          handle_t *handle,
550                          struct ocfs2_alloc_context *data_ac,
551                          struct ocfs2_alloc_context *meta_ac,
552                          enum ocfs2_alloc_restarted *reason_ret)
553 {
554         int ret;
555         struct ocfs2_extent_tree et;
556
557         ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), fe_bh);
558         ret = ocfs2_add_clusters_in_btree(handle, &et, logical_offset,
559                                           clusters_to_add, mark_unwritten,
560                                           data_ac, meta_ac, reason_ret);
561
562         return ret;
563 }
564
565 static int __ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
566                                      u32 clusters_to_add, int mark_unwritten)
567 {
568         int status = 0;
569         int restart_func = 0;
570         int credits;
571         u32 prev_clusters;
572         struct buffer_head *bh = NULL;
573         struct ocfs2_dinode *fe = NULL;
574         handle_t *handle = NULL;
575         struct ocfs2_alloc_context *data_ac = NULL;
576         struct ocfs2_alloc_context *meta_ac = NULL;
577         enum ocfs2_alloc_restarted why;
578         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
579         struct ocfs2_extent_tree et;
580         int did_quota = 0;
581
582         /*
583          * This function only exists for file systems which don't
584          * support holes.
585          */
586         BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb));
587
588         status = ocfs2_read_inode_block(inode, &bh);
589         if (status < 0) {
590                 mlog_errno(status);
591                 goto leave;
592         }
593         fe = (struct ocfs2_dinode *) bh->b_data;
594
595 restart_all:
596         BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
597
598         ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), bh);
599         status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
600                                        &data_ac, &meta_ac);
601         if (status) {
602                 mlog_errno(status);
603                 goto leave;
604         }
605
606         credits = ocfs2_calc_extend_credits(osb->sb, &fe->id2.i_list,
607                                             clusters_to_add);
608         handle = ocfs2_start_trans(osb, credits);
609         if (IS_ERR(handle)) {
610                 status = PTR_ERR(handle);
611                 handle = NULL;
612                 mlog_errno(status);
613                 goto leave;
614         }
615
616 restarted_transaction:
617         trace_ocfs2_extend_allocation(
618                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
619                 (unsigned long long)i_size_read(inode),
620                 le32_to_cpu(fe->i_clusters), clusters_to_add,
621                 why, restart_func);
622
623         status = dquot_alloc_space_nodirty(inode,
624                         ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
625         if (status)
626                 goto leave;
627         did_quota = 1;
628
629         /* reserve a write to the file entry early on - that we if we
630          * run out of credits in the allocation path, we can still
631          * update i_size. */
632         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
633                                          OCFS2_JOURNAL_ACCESS_WRITE);
634         if (status < 0) {
635                 mlog_errno(status);
636                 goto leave;
637         }
638
639         prev_clusters = OCFS2_I(inode)->ip_clusters;
640
641         status = ocfs2_add_inode_data(osb,
642                                       inode,
643                                       &logical_start,
644                                       clusters_to_add,
645                                       mark_unwritten,
646                                       bh,
647                                       handle,
648                                       data_ac,
649                                       meta_ac,
650                                       &why);
651         if ((status < 0) && (status != -EAGAIN)) {
652                 if (status != -ENOSPC)
653                         mlog_errno(status);
654                 goto leave;
655         }
656
657         ocfs2_journal_dirty(handle, bh);
658
659         spin_lock(&OCFS2_I(inode)->ip_lock);
660         clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
661         spin_unlock(&OCFS2_I(inode)->ip_lock);
662         /* Release unused quota reservation */
663         dquot_free_space(inode,
664                         ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
665         did_quota = 0;
666
667         if (why != RESTART_NONE && clusters_to_add) {
668                 if (why == RESTART_META) {
669                         restart_func = 1;
670                         status = 0;
671                 } else {
672                         BUG_ON(why != RESTART_TRANS);
673
674                         /* TODO: This can be more intelligent. */
675                         credits = ocfs2_calc_extend_credits(osb->sb,
676                                                             &fe->id2.i_list,
677                                                             clusters_to_add);
678                         status = ocfs2_extend_trans(handle, credits);
679                         if (status < 0) {
680                                 /* handle still has to be committed at
681                                  * this point. */
682                                 status = -ENOMEM;
683                                 mlog_errno(status);
684                                 goto leave;
685                         }
686                         goto restarted_transaction;
687                 }
688         }
689
690         trace_ocfs2_extend_allocation_end(OCFS2_I(inode)->ip_blkno,
691              le32_to_cpu(fe->i_clusters),
692              (unsigned long long)le64_to_cpu(fe->i_size),
693              OCFS2_I(inode)->ip_clusters,
694              (unsigned long long)i_size_read(inode));
695
696 leave:
697         if (status < 0 && did_quota)
698                 dquot_free_space(inode,
699                         ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
700         if (handle) {
701                 ocfs2_commit_trans(osb, handle);
702                 handle = NULL;
703         }
704         if (data_ac) {
705                 ocfs2_free_alloc_context(data_ac);
706                 data_ac = NULL;
707         }
708         if (meta_ac) {
709                 ocfs2_free_alloc_context(meta_ac);
710                 meta_ac = NULL;
711         }
712         if ((!status) && restart_func) {
713                 restart_func = 0;
714                 goto restart_all;
715         }
716         brelse(bh);
717         bh = NULL;
718
719         return status;
720 }
721
722 /*
723  * While a write will already be ordering the data, a truncate will not.
724  * Thus, we need to explicitly order the zeroed pages.
725  */
726 static handle_t *ocfs2_zero_start_ordered_transaction(struct inode *inode)
727 {
728         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
729         handle_t *handle = NULL;
730         int ret = 0;
731
732         if (!ocfs2_should_order_data(inode))
733                 goto out;
734
735         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
736         if (IS_ERR(handle)) {
737                 ret = -ENOMEM;
738                 mlog_errno(ret);
739                 goto out;
740         }
741
742         ret = ocfs2_jbd2_file_inode(handle, inode);
743         if (ret < 0)
744                 mlog_errno(ret);
745
746 out:
747         if (ret) {
748                 if (!IS_ERR(handle))
749                         ocfs2_commit_trans(osb, handle);
750                 handle = ERR_PTR(ret);
751         }
752         return handle;
753 }
754
755 /* Some parts of this taken from generic_cont_expand, which turned out
756  * to be too fragile to do exactly what we need without us having to
757  * worry about recursive locking in ->write_begin() and ->write_end(). */
758 static int ocfs2_write_zero_page(struct inode *inode, u64 abs_from,
759                                  u64 abs_to)
760 {
761         struct address_space *mapping = inode->i_mapping;
762         struct page *page;
763         unsigned long index = abs_from >> PAGE_CACHE_SHIFT;
764         handle_t *handle = NULL;
765         int ret = 0;
766         unsigned zero_from, zero_to, block_start, block_end;
767
768         BUG_ON(abs_from >= abs_to);
769         BUG_ON(abs_to > (((u64)index + 1) << PAGE_CACHE_SHIFT));
770         BUG_ON(abs_from & (inode->i_blkbits - 1));
771
772         page = find_or_create_page(mapping, index, GFP_NOFS);
773         if (!page) {
774                 ret = -ENOMEM;
775                 mlog_errno(ret);
776                 goto out;
777         }
778
779         /* Get the offsets within the page that we want to zero */
780         zero_from = abs_from & (PAGE_CACHE_SIZE - 1);
781         zero_to = abs_to & (PAGE_CACHE_SIZE - 1);
782         if (!zero_to)
783                 zero_to = PAGE_CACHE_SIZE;
784
785         trace_ocfs2_write_zero_page(
786                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
787                         (unsigned long long)abs_from,
788                         (unsigned long long)abs_to,
789                         index, zero_from, zero_to);
790
791         /* We know that zero_from is block aligned */
792         for (block_start = zero_from; block_start < zero_to;
793              block_start = block_end) {
794                 block_end = block_start + (1 << inode->i_blkbits);
795
796                 /*
797                  * block_start is block-aligned.  Bump it by one to force
798                  * __block_write_begin and block_commit_write to zero the
799                  * whole block.
800                  */
801                 ret = __block_write_begin(page, block_start + 1, 0,
802                                           ocfs2_get_block);
803                 if (ret < 0) {
804                         mlog_errno(ret);
805                         goto out_unlock;
806                 }
807
808                 if (!handle) {
809                         handle = ocfs2_zero_start_ordered_transaction(inode);
810                         if (IS_ERR(handle)) {
811                                 ret = PTR_ERR(handle);
812                                 handle = NULL;
813                                 break;
814                         }
815                 }
816
817                 /* must not update i_size! */
818                 ret = block_commit_write(page, block_start + 1,
819                                          block_start + 1);
820                 if (ret < 0)
821                         mlog_errno(ret);
822                 else
823                         ret = 0;
824         }
825
826         if (handle)
827                 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
828
829 out_unlock:
830         unlock_page(page);
831         page_cache_release(page);
832 out:
833         return ret;
834 }
835
836 /*
837  * Find the next range to zero.  We do this in terms of bytes because
838  * that's what ocfs2_zero_extend() wants, and it is dealing with the
839  * pagecache.  We may return multiple extents.
840  *
841  * zero_start and zero_end are ocfs2_zero_extend()s current idea of what
842  * needs to be zeroed.  range_start and range_end return the next zeroing
843  * range.  A subsequent call should pass the previous range_end as its
844  * zero_start.  If range_end is 0, there's nothing to do.
845  *
846  * Unwritten extents are skipped over.  Refcounted extents are CoWd.
847  */
848 static int ocfs2_zero_extend_get_range(struct inode *inode,
849                                        struct buffer_head *di_bh,
850                                        u64 zero_start, u64 zero_end,
851                                        u64 *range_start, u64 *range_end)
852 {
853         int rc = 0, needs_cow = 0;
854         u32 p_cpos, zero_clusters = 0;
855         u32 zero_cpos =
856                 zero_start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
857         u32 last_cpos = ocfs2_clusters_for_bytes(inode->i_sb, zero_end);
858         unsigned int num_clusters = 0;
859         unsigned int ext_flags = 0;
860
861         while (zero_cpos < last_cpos) {
862                 rc = ocfs2_get_clusters(inode, zero_cpos, &p_cpos,
863                                         &num_clusters, &ext_flags);
864                 if (rc) {
865                         mlog_errno(rc);
866                         goto out;
867                 }
868
869                 if (p_cpos && !(ext_flags & OCFS2_EXT_UNWRITTEN)) {
870                         zero_clusters = num_clusters;
871                         if (ext_flags & OCFS2_EXT_REFCOUNTED)
872                                 needs_cow = 1;
873                         break;
874                 }
875
876                 zero_cpos += num_clusters;
877         }
878         if (!zero_clusters) {
879                 *range_end = 0;
880                 goto out;
881         }
882
883         while ((zero_cpos + zero_clusters) < last_cpos) {
884                 rc = ocfs2_get_clusters(inode, zero_cpos + zero_clusters,
885                                         &p_cpos, &num_clusters,
886                                         &ext_flags);
887                 if (rc) {
888                         mlog_errno(rc);
889                         goto out;
890                 }
891
892                 if (!p_cpos || (ext_flags & OCFS2_EXT_UNWRITTEN))
893                         break;
894                 if (ext_flags & OCFS2_EXT_REFCOUNTED)
895                         needs_cow = 1;
896                 zero_clusters += num_clusters;
897         }
898         if ((zero_cpos + zero_clusters) > last_cpos)
899                 zero_clusters = last_cpos - zero_cpos;
900
901         if (needs_cow) {
902                 rc = ocfs2_refcount_cow(inode, NULL, di_bh, zero_cpos,
903                                         zero_clusters, UINT_MAX);
904                 if (rc) {
905                         mlog_errno(rc);
906                         goto out;
907                 }
908         }
909
910         *range_start = ocfs2_clusters_to_bytes(inode->i_sb, zero_cpos);
911         *range_end = ocfs2_clusters_to_bytes(inode->i_sb,
912                                              zero_cpos + zero_clusters);
913
914 out:
915         return rc;
916 }
917
918 /*
919  * Zero one range returned from ocfs2_zero_extend_get_range().  The caller
920  * has made sure that the entire range needs zeroing.
921  */
922 static int ocfs2_zero_extend_range(struct inode *inode, u64 range_start,
923                                    u64 range_end)
924 {
925         int rc = 0;
926         u64 next_pos;
927         u64 zero_pos = range_start;
928
929         trace_ocfs2_zero_extend_range(
930                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
931                         (unsigned long long)range_start,
932                         (unsigned long long)range_end);
933         BUG_ON(range_start >= range_end);
934
935         while (zero_pos < range_end) {
936                 next_pos = (zero_pos & PAGE_CACHE_MASK) + PAGE_CACHE_SIZE;
937                 if (next_pos > range_end)
938                         next_pos = range_end;
939                 rc = ocfs2_write_zero_page(inode, zero_pos, next_pos);
940                 if (rc < 0) {
941                         mlog_errno(rc);
942                         break;
943                 }
944                 zero_pos = next_pos;
945
946                 /*
947                  * Very large extends have the potential to lock up
948                  * the cpu for extended periods of time.
949                  */
950                 cond_resched();
951         }
952
953         return rc;
954 }
955
956 int ocfs2_zero_extend(struct inode *inode, struct buffer_head *di_bh,
957                       loff_t zero_to_size)
958 {
959         int ret = 0;
960         u64 zero_start, range_start = 0, range_end = 0;
961         struct super_block *sb = inode->i_sb;
962
963         zero_start = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
964         trace_ocfs2_zero_extend((unsigned long long)OCFS2_I(inode)->ip_blkno,
965                                 (unsigned long long)zero_start,
966                                 (unsigned long long)i_size_read(inode));
967         while (zero_start < zero_to_size) {
968                 ret = ocfs2_zero_extend_get_range(inode, di_bh, zero_start,
969                                                   zero_to_size,
970                                                   &range_start,
971                                                   &range_end);
972                 if (ret) {
973                         mlog_errno(ret);
974                         break;
975                 }
976                 if (!range_end)
977                         break;
978                 /* Trim the ends */
979                 if (range_start < zero_start)
980                         range_start = zero_start;
981                 if (range_end > zero_to_size)
982                         range_end = zero_to_size;
983
984                 ret = ocfs2_zero_extend_range(inode, range_start,
985                                               range_end);
986                 if (ret) {
987                         mlog_errno(ret);
988                         break;
989                 }
990                 zero_start = range_end;
991         }
992
993         return ret;
994 }
995
996 int ocfs2_extend_no_holes(struct inode *inode, struct buffer_head *di_bh,
997                           u64 new_i_size, u64 zero_to)
998 {
999         int ret;
1000         u32 clusters_to_add;
1001         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1002
1003         /*
1004          * Only quota files call this without a bh, and they can't be
1005          * refcounted.
1006          */
1007         BUG_ON(!di_bh && (oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
1008         BUG_ON(!di_bh && !(oi->ip_flags & OCFS2_INODE_SYSTEM_FILE));
1009
1010         clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size);
1011         if (clusters_to_add < oi->ip_clusters)
1012                 clusters_to_add = 0;
1013         else
1014                 clusters_to_add -= oi->ip_clusters;
1015
1016         if (clusters_to_add) {
1017                 ret = __ocfs2_extend_allocation(inode, oi->ip_clusters,
1018                                                 clusters_to_add, 0);
1019                 if (ret) {
1020                         mlog_errno(ret);
1021                         goto out;
1022                 }
1023         }
1024
1025         /*
1026          * Call this even if we don't add any clusters to the tree. We
1027          * still need to zero the area between the old i_size and the
1028          * new i_size.
1029          */
1030         ret = ocfs2_zero_extend(inode, di_bh, zero_to);
1031         if (ret < 0)
1032                 mlog_errno(ret);
1033
1034 out:
1035         return ret;
1036 }
1037
1038 static int ocfs2_extend_file(struct inode *inode,
1039                              struct buffer_head *di_bh,
1040                              u64 new_i_size)
1041 {
1042         int ret = 0;
1043         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1044
1045         BUG_ON(!di_bh);
1046
1047         /* setattr sometimes calls us like this. */
1048         if (new_i_size == 0)
1049                 goto out;
1050
1051         if (i_size_read(inode) == new_i_size)
1052                 goto out;
1053         BUG_ON(new_i_size < i_size_read(inode));
1054
1055         /*
1056          * The alloc sem blocks people in read/write from reading our
1057          * allocation until we're done changing it. We depend on
1058          * i_mutex to block other extend/truncate calls while we're
1059          * here.  We even have to hold it for sparse files because there
1060          * might be some tail zeroing.
1061          */
1062         down_write(&oi->ip_alloc_sem);
1063
1064         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1065                 /*
1066                  * We can optimize small extends by keeping the inodes
1067                  * inline data.
1068                  */
1069                 if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) {
1070                         up_write(&oi->ip_alloc_sem);
1071                         goto out_update_size;
1072                 }
1073
1074                 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1075                 if (ret) {
1076                         up_write(&oi->ip_alloc_sem);
1077                         mlog_errno(ret);
1078                         goto out;
1079                 }
1080         }
1081
1082         if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
1083                 ret = ocfs2_zero_extend(inode, di_bh, new_i_size);
1084         else
1085                 ret = ocfs2_extend_no_holes(inode, di_bh, new_i_size,
1086                                             new_i_size);
1087
1088         up_write(&oi->ip_alloc_sem);
1089
1090         if (ret < 0) {
1091                 mlog_errno(ret);
1092                 goto out;
1093         }
1094
1095 out_update_size:
1096         ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
1097         if (ret < 0)
1098                 mlog_errno(ret);
1099
1100 out:
1101         return ret;
1102 }
1103
1104 int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
1105 {
1106         int status = 0, size_change;
1107         struct inode *inode = dentry->d_inode;
1108         struct super_block *sb = inode->i_sb;
1109         struct ocfs2_super *osb = OCFS2_SB(sb);
1110         struct buffer_head *bh = NULL;
1111         handle_t *handle = NULL;
1112         struct dquot *transfer_to[MAXQUOTAS] = { };
1113         int qtype;
1114
1115         trace_ocfs2_setattr(inode, dentry,
1116                             (unsigned long long)OCFS2_I(inode)->ip_blkno,
1117                             dentry->d_name.len, dentry->d_name.name,
1118                             attr->ia_valid, attr->ia_mode,
1119                             from_kuid(&init_user_ns, attr->ia_uid),
1120                             from_kgid(&init_user_ns, attr->ia_gid));
1121
1122         /* ensuring we don't even attempt to truncate a symlink */
1123         if (S_ISLNK(inode->i_mode))
1124                 attr->ia_valid &= ~ATTR_SIZE;
1125
1126 #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
1127                            | ATTR_GID | ATTR_UID | ATTR_MODE)
1128         if (!(attr->ia_valid & OCFS2_VALID_ATTRS))
1129                 return 0;
1130
1131         status = inode_change_ok(inode, attr);
1132         if (status)
1133                 return status;
1134
1135         if (is_quota_modification(inode, attr))
1136                 dquot_initialize(inode);
1137         size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
1138         if (size_change) {
1139                 status = ocfs2_rw_lock(inode, 1);
1140                 if (status < 0) {
1141                         mlog_errno(status);
1142                         goto bail;
1143                 }
1144         }
1145
1146         status = ocfs2_inode_lock(inode, &bh, 1);
1147         if (status < 0) {
1148                 if (status != -ENOENT)
1149                         mlog_errno(status);
1150                 goto bail_unlock_rw;
1151         }
1152
1153         if (size_change && attr->ia_size != i_size_read(inode)) {
1154                 status = inode_newsize_ok(inode, attr->ia_size);
1155                 if (status)
1156                         goto bail_unlock;
1157
1158                 inode_dio_wait(inode);
1159
1160                 if (i_size_read(inode) > attr->ia_size) {
1161                         if (ocfs2_should_order_data(inode)) {
1162                                 status = ocfs2_begin_ordered_truncate(inode,
1163                                                                       attr->ia_size);
1164                                 if (status)
1165                                         goto bail_unlock;
1166                         }
1167                         status = ocfs2_truncate_file(inode, bh, attr->ia_size);
1168                 } else
1169                         status = ocfs2_extend_file(inode, bh, attr->ia_size);
1170                 if (status < 0) {
1171                         if (status != -ENOSPC)
1172                                 mlog_errno(status);
1173                         status = -ENOSPC;
1174                         goto bail_unlock;
1175                 }
1176         }
1177
1178         if ((attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
1179             (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
1180                 /*
1181                  * Gather pointers to quota structures so that allocation /
1182                  * freeing of quota structures happens here and not inside
1183                  * dquot_transfer() where we have problems with lock ordering
1184                  */
1185                 if (attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)
1186                     && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1187                     OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
1188                         transfer_to[USRQUOTA] = dqget(sb, make_kqid_uid(attr->ia_uid));
1189                         if (!transfer_to[USRQUOTA]) {
1190                                 status = -ESRCH;
1191                                 goto bail_unlock;
1192                         }
1193                 }
1194                 if (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid)
1195                     && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1196                     OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
1197                         transfer_to[GRPQUOTA] = dqget(sb, make_kqid_gid(attr->ia_gid));
1198                         if (!transfer_to[GRPQUOTA]) {
1199                                 status = -ESRCH;
1200                                 goto bail_unlock;
1201                         }
1202                 }
1203                 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS +
1204                                            2 * ocfs2_quota_trans_credits(sb));
1205                 if (IS_ERR(handle)) {
1206                         status = PTR_ERR(handle);
1207                         mlog_errno(status);
1208                         goto bail_unlock;
1209                 }
1210                 status = __dquot_transfer(inode, transfer_to);
1211                 if (status < 0)
1212                         goto bail_commit;
1213         } else {
1214                 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1215                 if (IS_ERR(handle)) {
1216                         status = PTR_ERR(handle);
1217                         mlog_errno(status);
1218                         goto bail_unlock;
1219                 }
1220         }
1221
1222         setattr_copy(inode, attr);
1223         mark_inode_dirty(inode);
1224
1225         status = ocfs2_mark_inode_dirty(handle, inode, bh);
1226         if (status < 0)
1227                 mlog_errno(status);
1228
1229 bail_commit:
1230         ocfs2_commit_trans(osb, handle);
1231 bail_unlock:
1232         ocfs2_inode_unlock(inode, 1);
1233 bail_unlock_rw:
1234         if (size_change)
1235                 ocfs2_rw_unlock(inode, 1);
1236 bail:
1237         brelse(bh);
1238
1239         /* Release quota pointers in case we acquired them */
1240         for (qtype = 0; qtype < MAXQUOTAS; qtype++)
1241                 dqput(transfer_to[qtype]);
1242
1243         if (!status && attr->ia_valid & ATTR_MODE) {
1244                 status = ocfs2_acl_chmod(inode);
1245                 if (status < 0)
1246                         mlog_errno(status);
1247         }
1248
1249         return status;
1250 }
1251
1252 int ocfs2_getattr(struct vfsmount *mnt,
1253                   struct dentry *dentry,
1254                   struct kstat *stat)
1255 {
1256         struct inode *inode = dentry->d_inode;
1257         struct super_block *sb = dentry->d_inode->i_sb;
1258         struct ocfs2_super *osb = sb->s_fs_info;
1259         int err;
1260
1261         err = ocfs2_inode_revalidate(dentry);
1262         if (err) {
1263                 if (err != -ENOENT)
1264                         mlog_errno(err);
1265                 goto bail;
1266         }
1267
1268         generic_fillattr(inode, stat);
1269
1270         /* We set the blksize from the cluster size for performance */
1271         stat->blksize = osb->s_clustersize;
1272
1273 bail:
1274         return err;
1275 }
1276
1277 int ocfs2_permission(struct inode *inode, int mask)
1278 {
1279         int ret;
1280
1281         if (mask & MAY_NOT_BLOCK)
1282                 return -ECHILD;
1283
1284         ret = ocfs2_inode_lock(inode, NULL, 0);
1285         if (ret) {
1286                 if (ret != -ENOENT)
1287                         mlog_errno(ret);
1288                 goto out;
1289         }
1290
1291         ret = generic_permission(inode, mask);
1292
1293         ocfs2_inode_unlock(inode, 0);
1294 out:
1295         return ret;
1296 }
1297
1298 static int __ocfs2_write_remove_suid(struct inode *inode,
1299                                      struct buffer_head *bh)
1300 {
1301         int ret;
1302         handle_t *handle;
1303         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1304         struct ocfs2_dinode *di;
1305
1306         trace_ocfs2_write_remove_suid(
1307                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
1308                         inode->i_mode);
1309
1310         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1311         if (IS_ERR(handle)) {
1312                 ret = PTR_ERR(handle);
1313                 mlog_errno(ret);
1314                 goto out;
1315         }
1316
1317         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
1318                                       OCFS2_JOURNAL_ACCESS_WRITE);
1319         if (ret < 0) {
1320                 mlog_errno(ret);
1321                 goto out_trans;
1322         }
1323
1324         inode->i_mode &= ~S_ISUID;
1325         if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
1326                 inode->i_mode &= ~S_ISGID;
1327
1328         di = (struct ocfs2_dinode *) bh->b_data;
1329         di->i_mode = cpu_to_le16(inode->i_mode);
1330
1331         ocfs2_journal_dirty(handle, bh);
1332
1333 out_trans:
1334         ocfs2_commit_trans(osb, handle);
1335 out:
1336         return ret;
1337 }
1338
1339 /*
1340  * Will look for holes and unwritten extents in the range starting at
1341  * pos for count bytes (inclusive).
1342  */
1343 static int ocfs2_check_range_for_holes(struct inode *inode, loff_t pos,
1344                                        size_t count)
1345 {
1346         int ret = 0;
1347         unsigned int extent_flags;
1348         u32 cpos, clusters, extent_len, phys_cpos;
1349         struct super_block *sb = inode->i_sb;
1350
1351         cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
1352         clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
1353
1354         while (clusters) {
1355                 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
1356                                          &extent_flags);
1357                 if (ret < 0) {
1358                         mlog_errno(ret);
1359                         goto out;
1360                 }
1361
1362                 if (phys_cpos == 0 || (extent_flags & OCFS2_EXT_UNWRITTEN)) {
1363                         ret = 1;
1364                         break;
1365                 }
1366
1367                 if (extent_len > clusters)
1368                         extent_len = clusters;
1369
1370                 clusters -= extent_len;
1371                 cpos += extent_len;
1372         }
1373 out:
1374         return ret;
1375 }
1376
1377 static int ocfs2_write_remove_suid(struct inode *inode)
1378 {
1379         int ret;
1380         struct buffer_head *bh = NULL;
1381
1382         ret = ocfs2_read_inode_block(inode, &bh);
1383         if (ret < 0) {
1384                 mlog_errno(ret);
1385                 goto out;
1386         }
1387
1388         ret =  __ocfs2_write_remove_suid(inode, bh);
1389 out:
1390         brelse(bh);
1391         return ret;
1392 }
1393
1394 /*
1395  * Allocate enough extents to cover the region starting at byte offset
1396  * start for len bytes. Existing extents are skipped, any extents
1397  * added are marked as "unwritten".
1398  */
1399 static int ocfs2_allocate_unwritten_extents(struct inode *inode,
1400                                             u64 start, u64 len)
1401 {
1402         int ret;
1403         u32 cpos, phys_cpos, clusters, alloc_size;
1404         u64 end = start + len;
1405         struct buffer_head *di_bh = NULL;
1406
1407         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1408                 ret = ocfs2_read_inode_block(inode, &di_bh);
1409                 if (ret) {
1410                         mlog_errno(ret);
1411                         goto out;
1412                 }
1413
1414                 /*
1415                  * Nothing to do if the requested reservation range
1416                  * fits within the inode.
1417                  */
1418                 if (ocfs2_size_fits_inline_data(di_bh, end))
1419                         goto out;
1420
1421                 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1422                 if (ret) {
1423                         mlog_errno(ret);
1424                         goto out;
1425                 }
1426         }
1427
1428         /*
1429          * We consider both start and len to be inclusive.
1430          */
1431         cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
1432         clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len);
1433         clusters -= cpos;
1434
1435         while (clusters) {
1436                 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1437                                          &alloc_size, NULL);
1438                 if (ret) {
1439                         mlog_errno(ret);
1440                         goto out;
1441                 }
1442
1443                 /*
1444                  * Hole or existing extent len can be arbitrary, so
1445                  * cap it to our own allocation request.
1446                  */
1447                 if (alloc_size > clusters)
1448                         alloc_size = clusters;
1449
1450                 if (phys_cpos) {
1451                         /*
1452                          * We already have an allocation at this
1453                          * region so we can safely skip it.
1454                          */
1455                         goto next;
1456                 }
1457
1458                 ret = __ocfs2_extend_allocation(inode, cpos, alloc_size, 1);
1459                 if (ret) {
1460                         if (ret != -ENOSPC)
1461                                 mlog_errno(ret);
1462                         goto out;
1463                 }
1464
1465 next:
1466                 cpos += alloc_size;
1467                 clusters -= alloc_size;
1468         }
1469
1470         ret = 0;
1471 out:
1472
1473         brelse(di_bh);
1474         return ret;
1475 }
1476
1477 /*
1478  * Truncate a byte range, avoiding pages within partial clusters. This
1479  * preserves those pages for the zeroing code to write to.
1480  */
1481 static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start,
1482                                          u64 byte_len)
1483 {
1484         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1485         loff_t start, end;
1486         struct address_space *mapping = inode->i_mapping;
1487
1488         start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start);
1489         end = byte_start + byte_len;
1490         end = end & ~(osb->s_clustersize - 1);
1491
1492         if (start < end) {
1493                 unmap_mapping_range(mapping, start, end - start, 0);
1494                 truncate_inode_pages_range(mapping, start, end - 1);
1495         }
1496 }
1497
1498 static int ocfs2_zero_partial_clusters(struct inode *inode,
1499                                        u64 start, u64 len)
1500 {
1501         int ret = 0;
1502         u64 tmpend, end = start + len;
1503         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1504         unsigned int csize = osb->s_clustersize;
1505         handle_t *handle;
1506
1507         /*
1508          * The "start" and "end" values are NOT necessarily part of
1509          * the range whose allocation is being deleted. Rather, this
1510          * is what the user passed in with the request. We must zero
1511          * partial clusters here. There's no need to worry about
1512          * physical allocation - the zeroing code knows to skip holes.
1513          */
1514         trace_ocfs2_zero_partial_clusters(
1515                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1516                 (unsigned long long)start, (unsigned long long)end);
1517
1518         /*
1519          * If both edges are on a cluster boundary then there's no
1520          * zeroing required as the region is part of the allocation to
1521          * be truncated.
1522          */
1523         if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0)
1524                 goto out;
1525
1526         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1527         if (IS_ERR(handle)) {
1528                 ret = PTR_ERR(handle);
1529                 mlog_errno(ret);
1530                 goto out;
1531         }
1532
1533         /*
1534          * We want to get the byte offset of the end of the 1st cluster.
1535          */
1536         tmpend = (u64)osb->s_clustersize + (start & ~(osb->s_clustersize - 1));
1537         if (tmpend > end)
1538                 tmpend = end;
1539
1540         trace_ocfs2_zero_partial_clusters_range1((unsigned long long)start,
1541                                                  (unsigned long long)tmpend);
1542
1543         ret = ocfs2_zero_range_for_truncate(inode, handle, start, tmpend);
1544         if (ret)
1545                 mlog_errno(ret);
1546
1547         if (tmpend < end) {
1548                 /*
1549                  * This may make start and end equal, but the zeroing
1550                  * code will skip any work in that case so there's no
1551                  * need to catch it up here.
1552                  */
1553                 start = end & ~(osb->s_clustersize - 1);
1554
1555                 trace_ocfs2_zero_partial_clusters_range2(
1556                         (unsigned long long)start, (unsigned long long)end);
1557
1558                 ret = ocfs2_zero_range_for_truncate(inode, handle, start, end);
1559                 if (ret)
1560                         mlog_errno(ret);
1561         }
1562
1563         ocfs2_commit_trans(osb, handle);
1564 out:
1565         return ret;
1566 }
1567
1568 static int ocfs2_find_rec(struct ocfs2_extent_list *el, u32 pos)
1569 {
1570         int i;
1571         struct ocfs2_extent_rec *rec = NULL;
1572
1573         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
1574
1575                 rec = &el->l_recs[i];
1576
1577                 if (le32_to_cpu(rec->e_cpos) < pos)
1578                         break;
1579         }
1580
1581         return i;
1582 }
1583
1584 /*
1585  * Helper to calculate the punching pos and length in one run, we handle the
1586  * following three cases in order:
1587  *
1588  * - remove the entire record
1589  * - remove a partial record
1590  * - no record needs to be removed (hole-punching completed)
1591 */
1592 static void ocfs2_calc_trunc_pos(struct inode *inode,
1593                                  struct ocfs2_extent_list *el,
1594                                  struct ocfs2_extent_rec *rec,
1595                                  u32 trunc_start, u32 *trunc_cpos,
1596                                  u32 *trunc_len, u32 *trunc_end,
1597                                  u64 *blkno, int *done)
1598 {
1599         int ret = 0;
1600         u32 coff, range;
1601
1602         range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
1603
1604         if (le32_to_cpu(rec->e_cpos) >= trunc_start) {
1605                 /*
1606                  * remove an entire extent record.
1607                  */
1608                 *trunc_cpos = le32_to_cpu(rec->e_cpos);
1609                 /*
1610                  * Skip holes if any.
1611                  */
1612                 if (range < *trunc_end)
1613                         *trunc_end = range;
1614                 *trunc_len = *trunc_end - le32_to_cpu(rec->e_cpos);
1615                 *blkno = le64_to_cpu(rec->e_blkno);
1616                 *trunc_end = le32_to_cpu(rec->e_cpos);
1617         } else if (range > trunc_start) {
1618                 /*
1619                  * remove a partial extent record, which means we're
1620                  * removing the last extent record.
1621                  */
1622                 *trunc_cpos = trunc_start;
1623                 /*
1624                  * skip hole if any.
1625                  */
1626                 if (range < *trunc_end)
1627                         *trunc_end = range;
1628                 *trunc_len = *trunc_end - trunc_start;
1629                 coff = trunc_start - le32_to_cpu(rec->e_cpos);
1630                 *blkno = le64_to_cpu(rec->e_blkno) +
1631                                 ocfs2_clusters_to_blocks(inode->i_sb, coff);
1632                 *trunc_end = trunc_start;
1633         } else {
1634                 /*
1635                  * It may have two following possibilities:
1636                  *
1637                  * - last record has been removed
1638                  * - trunc_start was within a hole
1639                  *
1640                  * both two cases mean the completion of hole punching.
1641                  */
1642                 ret = 1;
1643         }
1644
1645         *done = ret;
1646 }
1647
1648 static int ocfs2_remove_inode_range(struct inode *inode,
1649                                     struct buffer_head *di_bh, u64 byte_start,
1650                                     u64 byte_len)
1651 {
1652         int ret = 0, flags = 0, done = 0, i;
1653         u32 trunc_start, trunc_len, trunc_end, trunc_cpos, phys_cpos;
1654         u32 cluster_in_el;
1655         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1656         struct ocfs2_cached_dealloc_ctxt dealloc;
1657         struct address_space *mapping = inode->i_mapping;
1658         struct ocfs2_extent_tree et;
1659         struct ocfs2_path *path = NULL;
1660         struct ocfs2_extent_list *el = NULL;
1661         struct ocfs2_extent_rec *rec = NULL;
1662         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1663         u64 blkno, refcount_loc = le64_to_cpu(di->i_refcount_loc);
1664
1665         ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
1666         ocfs2_init_dealloc_ctxt(&dealloc);
1667
1668         trace_ocfs2_remove_inode_range(
1669                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
1670                         (unsigned long long)byte_start,
1671                         (unsigned long long)byte_len);
1672
1673         if (byte_len == 0)
1674                 return 0;
1675
1676         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1677                 ret = ocfs2_truncate_inline(inode, di_bh, byte_start,
1678                                             byte_start + byte_len, 0);
1679                 if (ret) {
1680                         mlog_errno(ret);
1681                         goto out;
1682                 }
1683                 /*
1684                  * There's no need to get fancy with the page cache
1685                  * truncate of an inline-data inode. We're talking
1686                  * about less than a page here, which will be cached
1687                  * in the dinode buffer anyway.
1688                  */
1689                 unmap_mapping_range(mapping, 0, 0, 0);
1690                 truncate_inode_pages(mapping, 0);
1691                 goto out;
1692         }
1693
1694         /*
1695          * For reflinks, we may need to CoW 2 clusters which might be
1696          * partially zero'd later, if hole's start and end offset were
1697          * within one cluster(means is not exactly aligned to clustersize).
1698          */
1699
1700         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) {
1701
1702                 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start);
1703                 if (ret) {
1704                         mlog_errno(ret);
1705                         goto out;
1706                 }
1707
1708                 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start + byte_len);
1709                 if (ret) {
1710                         mlog_errno(ret);
1711                         goto out;
1712                 }
1713         }
1714
1715         trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start);
1716         trunc_end = (byte_start + byte_len) >> osb->s_clustersize_bits;
1717         cluster_in_el = trunc_end;
1718
1719         ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len);
1720         if (ret) {
1721                 mlog_errno(ret);
1722                 goto out;
1723         }
1724
1725         path = ocfs2_new_path_from_et(&et);
1726         if (!path) {
1727                 ret = -ENOMEM;
1728                 mlog_errno(ret);
1729                 goto out;
1730         }
1731
1732         while (trunc_end > trunc_start) {
1733
1734                 ret = ocfs2_find_path(INODE_CACHE(inode), path,
1735                                       cluster_in_el);
1736                 if (ret) {
1737                         mlog_errno(ret);
1738                         goto out;
1739                 }
1740
1741                 el = path_leaf_el(path);
1742
1743                 i = ocfs2_find_rec(el, trunc_end);
1744                 /*
1745                  * Need to go to previous extent block.
1746                  */
1747                 if (i < 0) {
1748                         if (path->p_tree_depth == 0)
1749                                 break;
1750
1751                         ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb,
1752                                                             path,
1753                                                             &cluster_in_el);
1754                         if (ret) {
1755                                 mlog_errno(ret);
1756                                 goto out;
1757                         }
1758
1759                         /*
1760                          * We've reached the leftmost extent block,
1761                          * it's safe to leave.
1762                          */
1763                         if (cluster_in_el == 0)
1764                                 break;
1765
1766                         /*
1767                          * The 'pos' searched for previous extent block is
1768                          * always one cluster less than actual trunc_end.
1769                          */
1770                         trunc_end = cluster_in_el + 1;
1771
1772                         ocfs2_reinit_path(path, 1);
1773
1774                         continue;
1775
1776                 } else
1777                         rec = &el->l_recs[i];
1778
1779                 ocfs2_calc_trunc_pos(inode, el, rec, trunc_start, &trunc_cpos,
1780                                      &trunc_len, &trunc_end, &blkno, &done);
1781                 if (done)
1782                         break;
1783
1784                 flags = rec->e_flags;
1785                 phys_cpos = ocfs2_blocks_to_clusters(inode->i_sb, blkno);
1786
1787                 ret = ocfs2_remove_btree_range(inode, &et, trunc_cpos,
1788                                                phys_cpos, trunc_len, flags,
1789                                                &dealloc, refcount_loc);
1790                 if (ret < 0) {
1791                         mlog_errno(ret);
1792                         goto out;
1793                 }
1794
1795                 cluster_in_el = trunc_end;
1796
1797                 ocfs2_reinit_path(path, 1);
1798         }
1799
1800         ocfs2_truncate_cluster_pages(inode, byte_start, byte_len);
1801
1802 out:
1803         ocfs2_schedule_truncate_log_flush(osb, 1);
1804         ocfs2_run_deallocs(osb, &dealloc);
1805
1806         return ret;
1807 }
1808
1809 /*
1810  * Parts of this function taken from xfs_change_file_space()
1811  */
1812 static int __ocfs2_change_file_space(struct file *file, struct inode *inode,
1813                                      loff_t f_pos, unsigned int cmd,
1814                                      struct ocfs2_space_resv *sr,
1815                                      int change_size)
1816 {
1817         int ret;
1818         s64 llen;
1819         loff_t size;
1820         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1821         struct buffer_head *di_bh = NULL;
1822         handle_t *handle;
1823         unsigned long long max_off = inode->i_sb->s_maxbytes;
1824
1825         if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
1826                 return -EROFS;
1827
1828         mutex_lock(&inode->i_mutex);
1829
1830         /*
1831          * This prevents concurrent writes on other nodes
1832          */
1833         ret = ocfs2_rw_lock(inode, 1);
1834         if (ret) {
1835                 mlog_errno(ret);
1836                 goto out;
1837         }
1838
1839         ret = ocfs2_inode_lock(inode, &di_bh, 1);
1840         if (ret) {
1841                 mlog_errno(ret);
1842                 goto out_rw_unlock;
1843         }
1844
1845         if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
1846                 ret = -EPERM;
1847                 goto out_inode_unlock;
1848         }
1849
1850         switch (sr->l_whence) {
1851         case 0: /*SEEK_SET*/
1852                 break;
1853         case 1: /*SEEK_CUR*/
1854                 sr->l_start += f_pos;
1855                 break;
1856         case 2: /*SEEK_END*/
1857                 sr->l_start += i_size_read(inode);
1858                 break;
1859         default:
1860                 ret = -EINVAL;
1861                 goto out_inode_unlock;
1862         }
1863         sr->l_whence = 0;
1864
1865         llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len;
1866
1867         if (sr->l_start < 0
1868             || sr->l_start > max_off
1869             || (sr->l_start + llen) < 0
1870             || (sr->l_start + llen) > max_off) {
1871                 ret = -EINVAL;
1872                 goto out_inode_unlock;
1873         }
1874         size = sr->l_start + sr->l_len;
1875
1876         if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) {
1877                 if (sr->l_len <= 0) {
1878                         ret = -EINVAL;
1879                         goto out_inode_unlock;
1880                 }
1881         }
1882
1883         if (file && should_remove_suid(file->f_path.dentry)) {
1884                 ret = __ocfs2_write_remove_suid(inode, di_bh);
1885                 if (ret) {
1886                         mlog_errno(ret);
1887                         goto out_inode_unlock;
1888                 }
1889         }
1890
1891         down_write(&OCFS2_I(inode)->ip_alloc_sem);
1892         switch (cmd) {
1893         case OCFS2_IOC_RESVSP:
1894         case OCFS2_IOC_RESVSP64:
1895                 /*
1896                  * This takes unsigned offsets, but the signed ones we
1897                  * pass have been checked against overflow above.
1898                  */
1899                 ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start,
1900                                                        sr->l_len);
1901                 break;
1902         case OCFS2_IOC_UNRESVSP:
1903         case OCFS2_IOC_UNRESVSP64:
1904                 ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start,
1905                                                sr->l_len);
1906                 break;
1907         default:
1908                 ret = -EINVAL;
1909         }
1910         up_write(&OCFS2_I(inode)->ip_alloc_sem);
1911         if (ret) {
1912                 mlog_errno(ret);
1913                 goto out_inode_unlock;
1914         }
1915
1916         /*
1917          * We update c/mtime for these changes
1918          */
1919         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1920         if (IS_ERR(handle)) {
1921                 ret = PTR_ERR(handle);
1922                 mlog_errno(ret);
1923                 goto out_inode_unlock;
1924         }
1925
1926         if (change_size && i_size_read(inode) < size)
1927                 i_size_write(inode, size);
1928
1929         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1930         ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
1931         if (ret < 0)
1932                 mlog_errno(ret);
1933
1934         if (file && (file->f_flags & O_SYNC))
1935                 handle->h_sync = 1;
1936
1937         ocfs2_commit_trans(osb, handle);
1938
1939 out_inode_unlock:
1940         brelse(di_bh);
1941         ocfs2_inode_unlock(inode, 1);
1942 out_rw_unlock:
1943         ocfs2_rw_unlock(inode, 1);
1944
1945 out:
1946         mutex_unlock(&inode->i_mutex);
1947         return ret;
1948 }
1949
1950 int ocfs2_change_file_space(struct file *file, unsigned int cmd,
1951                             struct ocfs2_space_resv *sr)
1952 {
1953         struct inode *inode = file_inode(file);
1954         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1955         int ret;
1956
1957         if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) &&
1958             !ocfs2_writes_unwritten_extents(osb))
1959                 return -ENOTTY;
1960         else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) &&
1961                  !ocfs2_sparse_alloc(osb))
1962                 return -ENOTTY;
1963
1964         if (!S_ISREG(inode->i_mode))
1965                 return -EINVAL;
1966
1967         if (!(file->f_mode & FMODE_WRITE))
1968                 return -EBADF;
1969
1970         ret = mnt_want_write_file(file);
1971         if (ret)
1972                 return ret;
1973         ret = __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0);
1974         mnt_drop_write_file(file);
1975         return ret;
1976 }
1977
1978 static long ocfs2_fallocate(struct file *file, int mode, loff_t offset,
1979                             loff_t len)
1980 {
1981         struct inode *inode = file_inode(file);
1982         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1983         struct ocfs2_space_resv sr;
1984         int change_size = 1;
1985         int cmd = OCFS2_IOC_RESVSP64;
1986
1987         if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
1988                 return -EOPNOTSUPP;
1989         if (!ocfs2_writes_unwritten_extents(osb))
1990                 return -EOPNOTSUPP;
1991
1992         if (mode & FALLOC_FL_KEEP_SIZE)
1993                 change_size = 0;
1994
1995         if (mode & FALLOC_FL_PUNCH_HOLE)
1996                 cmd = OCFS2_IOC_UNRESVSP64;
1997
1998         sr.l_whence = 0;
1999         sr.l_start = (s64)offset;
2000         sr.l_len = (s64)len;
2001
2002         return __ocfs2_change_file_space(NULL, inode, offset, cmd, &sr,
2003                                          change_size);
2004 }
2005
2006 int ocfs2_check_range_for_refcount(struct inode *inode, loff_t pos,
2007                                    size_t count)
2008 {
2009         int ret = 0;
2010         unsigned int extent_flags;
2011         u32 cpos, clusters, extent_len, phys_cpos;
2012         struct super_block *sb = inode->i_sb;
2013
2014         if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)) ||
2015             !(OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) ||
2016             OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
2017                 return 0;
2018
2019         cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
2020         clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
2021
2022         while (clusters) {
2023                 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
2024                                          &extent_flags);
2025                 if (ret < 0) {
2026                         mlog_errno(ret);
2027                         goto out;
2028                 }
2029
2030                 if (phys_cpos && (extent_flags & OCFS2_EXT_REFCOUNTED)) {
2031                         ret = 1;
2032                         break;
2033                 }
2034
2035                 if (extent_len > clusters)
2036                         extent_len = clusters;
2037
2038                 clusters -= extent_len;
2039                 cpos += extent_len;
2040         }
2041 out:
2042         return ret;
2043 }
2044
2045 static void ocfs2_aiodio_wait(struct inode *inode)
2046 {
2047         wait_queue_head_t *wq = ocfs2_ioend_wq(inode);
2048
2049         wait_event(*wq, (atomic_read(&OCFS2_I(inode)->ip_unaligned_aio) == 0));
2050 }
2051
2052 static int ocfs2_is_io_unaligned(struct inode *inode, size_t count, loff_t pos)
2053 {
2054         int blockmask = inode->i_sb->s_blocksize - 1;
2055         loff_t final_size = pos + count;
2056
2057         if ((pos & blockmask) || (final_size & blockmask))
2058                 return 1;
2059         return 0;
2060 }
2061
2062 static int ocfs2_prepare_inode_for_refcount(struct inode *inode,
2063                                             struct file *file,
2064                                             loff_t pos, size_t count,
2065                                             int *meta_level)
2066 {
2067         int ret;
2068         struct buffer_head *di_bh = NULL;
2069         u32 cpos = pos >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
2070         u32 clusters =
2071                 ocfs2_clusters_for_bytes(inode->i_sb, pos + count) - cpos;
2072
2073         ret = ocfs2_inode_lock(inode, &di_bh, 1);
2074         if (ret) {
2075                 mlog_errno(ret);
2076                 goto out;
2077         }
2078
2079         *meta_level = 1;
2080
2081         ret = ocfs2_refcount_cow(inode, file, di_bh, cpos, clusters, UINT_MAX);
2082         if (ret)
2083                 mlog_errno(ret);
2084 out:
2085         brelse(di_bh);
2086         return ret;
2087 }
2088
2089 static int ocfs2_prepare_inode_for_write(struct file *file,
2090                                          loff_t *ppos,
2091                                          size_t count,
2092                                          int appending,
2093                                          int *direct_io,
2094                                          int *has_refcount)
2095 {
2096         int ret = 0, meta_level = 0;
2097         struct dentry *dentry = file->f_path.dentry;
2098         struct inode *inode = dentry->d_inode;
2099         loff_t saved_pos = 0, end;
2100
2101         /*
2102          * We start with a read level meta lock and only jump to an ex
2103          * if we need to make modifications here.
2104          */
2105         for(;;) {
2106                 ret = ocfs2_inode_lock(inode, NULL, meta_level);
2107                 if (ret < 0) {
2108                         meta_level = -1;
2109                         mlog_errno(ret);
2110                         goto out;
2111                 }
2112
2113                 /* Clear suid / sgid if necessary. We do this here
2114                  * instead of later in the write path because
2115                  * remove_suid() calls ->setattr without any hint that
2116                  * we may have already done our cluster locking. Since
2117                  * ocfs2_setattr() *must* take cluster locks to
2118                  * proceed, this will lead us to recursively lock the
2119                  * inode. There's also the dinode i_size state which
2120                  * can be lost via setattr during extending writes (we
2121                  * set inode->i_size at the end of a write. */
2122                 if (should_remove_suid(dentry)) {
2123                         if (meta_level == 0) {
2124                                 ocfs2_inode_unlock(inode, meta_level);
2125                                 meta_level = 1;
2126                                 continue;
2127                         }
2128
2129                         ret = ocfs2_write_remove_suid(inode);
2130                         if (ret < 0) {
2131                                 mlog_errno(ret);
2132                                 goto out_unlock;
2133                         }
2134                 }
2135
2136                 /* work on a copy of ppos until we're sure that we won't have
2137                  * to recalculate it due to relocking. */
2138                 if (appending)
2139                         saved_pos = i_size_read(inode);
2140                 else
2141                         saved_pos = *ppos;
2142
2143                 end = saved_pos + count;
2144
2145                 ret = ocfs2_check_range_for_refcount(inode, saved_pos, count);
2146                 if (ret == 1) {
2147                         ocfs2_inode_unlock(inode, meta_level);
2148                         meta_level = -1;
2149
2150                         ret = ocfs2_prepare_inode_for_refcount(inode,
2151                                                                file,
2152                                                                saved_pos,
2153                                                                count,
2154                                                                &meta_level);
2155                         if (has_refcount)
2156                                 *has_refcount = 1;
2157                         if (direct_io)
2158                                 *direct_io = 0;
2159                 }
2160
2161                 if (ret < 0) {
2162                         mlog_errno(ret);
2163                         goto out_unlock;
2164                 }
2165
2166                 /*
2167                  * Skip the O_DIRECT checks if we don't need
2168                  * them.
2169                  */
2170                 if (!direct_io || !(*direct_io))
2171                         break;
2172
2173                 /*
2174                  * There's no sane way to do direct writes to an inode
2175                  * with inline data.
2176                  */
2177                 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2178                         *direct_io = 0;
2179                         break;
2180                 }
2181
2182                 /*
2183                  * Allowing concurrent direct writes means
2184                  * i_size changes wouldn't be synchronized, so
2185                  * one node could wind up truncating another
2186                  * nodes writes.
2187                  */
2188                 if (end > i_size_read(inode)) {
2189                         *direct_io = 0;
2190                         break;
2191                 }
2192
2193                 /*
2194                  * We don't fill holes during direct io, so
2195                  * check for them here. If any are found, the
2196                  * caller will have to retake some cluster
2197                  * locks and initiate the io as buffered.
2198                  */
2199                 ret = ocfs2_check_range_for_holes(inode, saved_pos, count);
2200                 if (ret == 1) {
2201                         *direct_io = 0;
2202                         ret = 0;
2203                 } else if (ret < 0)
2204                         mlog_errno(ret);
2205                 break;
2206         }
2207
2208         if (appending)
2209                 *ppos = saved_pos;
2210
2211 out_unlock:
2212         trace_ocfs2_prepare_inode_for_write(OCFS2_I(inode)->ip_blkno,
2213                                             saved_pos, appending, count,
2214                                             direct_io, has_refcount);
2215
2216         if (meta_level >= 0)
2217                 ocfs2_inode_unlock(inode, meta_level);
2218
2219 out:
2220         return ret;
2221 }
2222
2223 static ssize_t ocfs2_file_aio_write(struct kiocb *iocb,
2224                                     const struct iovec *iov,
2225                                     unsigned long nr_segs,
2226                                     loff_t pos)
2227 {
2228         int ret, direct_io, appending, rw_level, have_alloc_sem  = 0;
2229         int can_do_direct, has_refcount = 0;
2230         ssize_t written = 0;
2231         size_t ocount;          /* original count */
2232         size_t count;           /* after file limit checks */
2233         loff_t old_size, *ppos = &iocb->ki_pos;
2234         u32 old_clusters;
2235         struct file *file = iocb->ki_filp;
2236         struct inode *inode = file_inode(file);
2237         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2238         int full_coherency = !(osb->s_mount_opt &
2239                                OCFS2_MOUNT_COHERENCY_BUFFERED);
2240         int unaligned_dio = 0;
2241
2242         trace_ocfs2_file_aio_write(inode, file, file->f_path.dentry,
2243                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
2244                 file->f_path.dentry->d_name.len,
2245                 file->f_path.dentry->d_name.name,
2246                 (unsigned int)nr_segs);
2247
2248         if (iocb->ki_left == 0)
2249                 return 0;
2250
2251         appending = file->f_flags & O_APPEND ? 1 : 0;
2252         direct_io = file->f_flags & O_DIRECT ? 1 : 0;
2253
2254         mutex_lock(&inode->i_mutex);
2255
2256         ocfs2_iocb_clear_sem_locked(iocb);
2257
2258 relock:
2259         /* to match setattr's i_mutex -> rw_lock ordering */
2260         if (direct_io) {
2261                 have_alloc_sem = 1;
2262                 /* communicate with ocfs2_dio_end_io */
2263                 ocfs2_iocb_set_sem_locked(iocb);
2264         }
2265
2266         /*
2267          * Concurrent O_DIRECT writes are allowed with
2268          * mount_option "coherency=buffered".
2269          */
2270         rw_level = (!direct_io || full_coherency);
2271
2272         ret = ocfs2_rw_lock(inode, rw_level);
2273         if (ret < 0) {
2274                 mlog_errno(ret);
2275                 goto out_sems;
2276         }
2277
2278         /*
2279          * O_DIRECT writes with "coherency=full" need to take EX cluster
2280          * inode_lock to guarantee coherency.
2281          */
2282         if (direct_io && full_coherency) {
2283                 /*
2284                  * We need to take and drop the inode lock to force
2285                  * other nodes to drop their caches.  Buffered I/O
2286                  * already does this in write_begin().
2287                  */
2288                 ret = ocfs2_inode_lock(inode, NULL, 1);
2289                 if (ret < 0) {
2290                         mlog_errno(ret);
2291                         goto out;
2292                 }
2293
2294                 ocfs2_inode_unlock(inode, 1);
2295         }
2296
2297         can_do_direct = direct_io;
2298         ret = ocfs2_prepare_inode_for_write(file, ppos,
2299                                             iocb->ki_left, appending,
2300                                             &can_do_direct, &has_refcount);
2301         if (ret < 0) {
2302                 mlog_errno(ret);
2303                 goto out;
2304         }
2305
2306         if (direct_io && !is_sync_kiocb(iocb))
2307                 unaligned_dio = ocfs2_is_io_unaligned(inode, iocb->ki_left,
2308                                                       *ppos);
2309
2310         /*
2311          * We can't complete the direct I/O as requested, fall back to
2312          * buffered I/O.
2313          */
2314         if (direct_io && !can_do_direct) {
2315                 ocfs2_rw_unlock(inode, rw_level);
2316
2317                 have_alloc_sem = 0;
2318                 rw_level = -1;
2319
2320                 direct_io = 0;
2321                 goto relock;
2322         }
2323
2324         if (unaligned_dio) {
2325                 /*
2326                  * Wait on previous unaligned aio to complete before
2327                  * proceeding.
2328                  */
2329                 ocfs2_aiodio_wait(inode);
2330
2331                 /* Mark the iocb as needing a decrement in ocfs2_dio_end_io */
2332                 atomic_inc(&OCFS2_I(inode)->ip_unaligned_aio);
2333                 ocfs2_iocb_set_unaligned_aio(iocb);
2334         }
2335
2336         /*
2337          * To later detect whether a journal commit for sync writes is
2338          * necessary, we sample i_size, and cluster count here.
2339          */
2340         old_size = i_size_read(inode);
2341         old_clusters = OCFS2_I(inode)->ip_clusters;
2342
2343         /* communicate with ocfs2_dio_end_io */
2344         ocfs2_iocb_set_rw_locked(iocb, rw_level);
2345
2346         ret = generic_segment_checks(iov, &nr_segs, &ocount,
2347                                      VERIFY_READ);
2348         if (ret)
2349                 goto out_dio;
2350
2351         count = ocount;
2352         ret = generic_write_checks(file, ppos, &count,
2353                                    S_ISBLK(inode->i_mode));
2354         if (ret)
2355                 goto out_dio;
2356
2357         if (direct_io) {
2358                 written = generic_file_direct_write(iocb, iov, &nr_segs, *ppos,
2359                                                     ppos, count, ocount);
2360                 if (written < 0) {
2361                         ret = written;
2362                         goto out_dio;
2363                 }
2364         } else {
2365                 current->backing_dev_info = file->f_mapping->backing_dev_info;
2366                 written = generic_file_buffered_write(iocb, iov, nr_segs, *ppos,
2367                                                       ppos, count, 0);
2368                 current->backing_dev_info = NULL;
2369         }
2370
2371 out_dio:
2372         /* buffered aio wouldn't have proper lock coverage today */
2373         BUG_ON(ret == -EIOCBQUEUED && !(file->f_flags & O_DIRECT));
2374
2375         if (((file->f_flags & O_DSYNC) && !direct_io) || IS_SYNC(inode) ||
2376             ((file->f_flags & O_DIRECT) && !direct_io)) {
2377                 ret = filemap_fdatawrite_range(file->f_mapping, *ppos,
2378                                                *ppos + count - 1);
2379                 if (ret < 0)
2380                         written = ret;
2381
2382                 if (!ret && ((old_size != i_size_read(inode)) ||
2383                              (old_clusters != OCFS2_I(inode)->ip_clusters) ||
2384                              has_refcount)) {
2385                         ret = jbd2_journal_force_commit(osb->journal->j_journal);
2386                         if (ret < 0)
2387                                 written = ret;
2388                 }
2389
2390                 if (!ret)
2391                         ret = filemap_fdatawait_range(file->f_mapping, *ppos,
2392                                                       *ppos + count - 1);
2393         }
2394
2395         /*
2396          * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
2397          * function pointer which is called when o_direct io completes so that
2398          * it can unlock our rw lock.
2399          * Unfortunately there are error cases which call end_io and others
2400          * that don't.  so we don't have to unlock the rw_lock if either an
2401          * async dio is going to do it in the future or an end_io after an
2402          * error has already done it.
2403          */
2404         if ((ret == -EIOCBQUEUED) || (!ocfs2_iocb_is_rw_locked(iocb))) {
2405                 rw_level = -1;
2406                 have_alloc_sem = 0;
2407                 unaligned_dio = 0;
2408         }
2409
2410         if (unaligned_dio) {
2411                 ocfs2_iocb_clear_unaligned_aio(iocb);
2412                 atomic_dec(&OCFS2_I(inode)->ip_unaligned_aio);
2413         }
2414
2415 out:
2416         if (rw_level != -1)
2417                 ocfs2_rw_unlock(inode, rw_level);
2418
2419 out_sems:
2420         if (have_alloc_sem)
2421                 ocfs2_iocb_clear_sem_locked(iocb);
2422
2423         mutex_unlock(&inode->i_mutex);
2424
2425         if (written)
2426                 ret = written;
2427         return ret;
2428 }
2429
2430 static int ocfs2_splice_to_file(struct pipe_inode_info *pipe,
2431                                 struct file *out,
2432                                 struct splice_desc *sd)
2433 {
2434         int ret;
2435
2436         ret = ocfs2_prepare_inode_for_write(out, &sd->pos,
2437                                             sd->total_len, 0, NULL, NULL);
2438         if (ret < 0) {
2439                 mlog_errno(ret);
2440                 return ret;
2441         }
2442
2443         return splice_from_pipe_feed(pipe, sd, pipe_to_file);
2444 }
2445
2446 static ssize_t ocfs2_file_splice_write(struct pipe_inode_info *pipe,
2447                                        struct file *out,
2448                                        loff_t *ppos,
2449                                        size_t len,
2450                                        unsigned int flags)
2451 {
2452         int ret;
2453         struct address_space *mapping = out->f_mapping;
2454         struct inode *inode = mapping->host;
2455         struct splice_desc sd = {
2456                 .total_len = len,
2457                 .flags = flags,
2458                 .pos = *ppos,
2459                 .u.file = out,
2460         };
2461
2462
2463         trace_ocfs2_file_splice_write(inode, out, out->f_path.dentry,
2464                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
2465                         out->f_path.dentry->d_name.len,
2466                         out->f_path.dentry->d_name.name, len);
2467
2468         pipe_lock(pipe);
2469
2470         splice_from_pipe_begin(&sd);
2471         do {
2472                 ret = splice_from_pipe_next(pipe, &sd);
2473                 if (ret <= 0)
2474                         break;
2475
2476                 mutex_lock_nested(&inode->i_mutex, I_MUTEX_CHILD);
2477                 ret = ocfs2_rw_lock(inode, 1);
2478                 if (ret < 0)
2479                         mlog_errno(ret);
2480                 else {
2481                         ret = ocfs2_splice_to_file(pipe, out, &sd);
2482                         ocfs2_rw_unlock(inode, 1);
2483                 }
2484                 mutex_unlock(&inode->i_mutex);
2485         } while (ret > 0);
2486         splice_from_pipe_end(pipe, &sd);
2487
2488         pipe_unlock(pipe);
2489
2490         if (sd.num_spliced)
2491                 ret = sd.num_spliced;
2492
2493         if (ret > 0) {
2494                 int err;
2495
2496                 err = generic_write_sync(out, *ppos, ret);
2497                 if (err)
2498                         ret = err;
2499                 else
2500                         *ppos += ret;
2501
2502                 balance_dirty_pages_ratelimited(mapping);
2503         }
2504
2505         return ret;
2506 }
2507
2508 static ssize_t ocfs2_file_splice_read(struct file *in,
2509                                       loff_t *ppos,
2510                                       struct pipe_inode_info *pipe,
2511                                       size_t len,
2512                                       unsigned int flags)
2513 {
2514         int ret = 0, lock_level = 0;
2515         struct inode *inode = file_inode(in);
2516
2517         trace_ocfs2_file_splice_read(inode, in, in->f_path.dentry,
2518                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
2519                         in->f_path.dentry->d_name.len,
2520                         in->f_path.dentry->d_name.name, len);
2521
2522         /*
2523          * See the comment in ocfs2_file_aio_read()
2524          */
2525         ret = ocfs2_inode_lock_atime(inode, in->f_path.mnt, &lock_level);
2526         if (ret < 0) {
2527                 mlog_errno(ret);
2528                 goto bail;
2529         }
2530         ocfs2_inode_unlock(inode, lock_level);
2531
2532         ret = generic_file_splice_read(in, ppos, pipe, len, flags);
2533
2534 bail:
2535         return ret;
2536 }
2537
2538 static ssize_t ocfs2_file_aio_read(struct kiocb *iocb,
2539                                    const struct iovec *iov,
2540                                    unsigned long nr_segs,
2541                                    loff_t pos)
2542 {
2543         int ret = 0, rw_level = -1, have_alloc_sem = 0, lock_level = 0;
2544         struct file *filp = iocb->ki_filp;
2545         struct inode *inode = file_inode(filp);
2546
2547         trace_ocfs2_file_aio_read(inode, filp, filp->f_path.dentry,
2548                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
2549                         filp->f_path.dentry->d_name.len,
2550                         filp->f_path.dentry->d_name.name, nr_segs);
2551
2552
2553         if (!inode) {
2554                 ret = -EINVAL;
2555                 mlog_errno(ret);
2556                 goto bail;
2557         }
2558
2559         ocfs2_iocb_clear_sem_locked(iocb);
2560
2561         /*
2562          * buffered reads protect themselves in ->readpage().  O_DIRECT reads
2563          * need locks to protect pending reads from racing with truncate.
2564          */
2565         if (filp->f_flags & O_DIRECT) {
2566                 have_alloc_sem = 1;
2567                 ocfs2_iocb_set_sem_locked(iocb);
2568
2569                 ret = ocfs2_rw_lock(inode, 0);
2570                 if (ret < 0) {
2571                         mlog_errno(ret);
2572                         goto bail;
2573                 }
2574                 rw_level = 0;
2575                 /* communicate with ocfs2_dio_end_io */
2576                 ocfs2_iocb_set_rw_locked(iocb, rw_level);
2577         }
2578
2579         /*
2580          * We're fine letting folks race truncates and extending
2581          * writes with read across the cluster, just like they can
2582          * locally. Hence no rw_lock during read.
2583          *
2584          * Take and drop the meta data lock to update inode fields
2585          * like i_size. This allows the checks down below
2586          * generic_file_aio_read() a chance of actually working.
2587          */
2588         ret = ocfs2_inode_lock_atime(inode, filp->f_path.mnt, &lock_level);
2589         if (ret < 0) {
2590                 mlog_errno(ret);
2591                 goto bail;
2592         }
2593         ocfs2_inode_unlock(inode, lock_level);
2594
2595         ret = generic_file_aio_read(iocb, iov, nr_segs, iocb->ki_pos);
2596         trace_generic_file_aio_read_ret(ret);
2597
2598         /* buffered aio wouldn't have proper lock coverage today */
2599         BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT));
2600
2601         /* see ocfs2_file_aio_write */
2602         if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2603                 rw_level = -1;
2604                 have_alloc_sem = 0;
2605         }
2606
2607 bail:
2608         if (have_alloc_sem)
2609                 ocfs2_iocb_clear_sem_locked(iocb);
2610
2611         if (rw_level != -1)
2612                 ocfs2_rw_unlock(inode, rw_level);
2613
2614         return ret;
2615 }
2616
2617 /* Refer generic_file_llseek_unlocked() */
2618 static loff_t ocfs2_file_llseek(struct file *file, loff_t offset, int whence)
2619 {
2620         struct inode *inode = file->f_mapping->host;
2621         int ret = 0;
2622
2623         mutex_lock(&inode->i_mutex);
2624
2625         switch (whence) {
2626         case SEEK_SET:
2627                 break;
2628         case SEEK_END:
2629                 offset += inode->i_size;
2630                 break;
2631         case SEEK_CUR:
2632                 if (offset == 0) {
2633                         offset = file->f_pos;
2634                         goto out;
2635                 }
2636                 offset += file->f_pos;
2637                 break;
2638         case SEEK_DATA:
2639         case SEEK_HOLE:
2640                 ret = ocfs2_seek_data_hole_offset(file, &offset, whence);
2641                 if (ret)
2642                         goto out;
2643                 break;
2644         default:
2645                 ret = -EINVAL;
2646                 goto out;
2647         }
2648
2649         if (offset < 0 && !(file->f_mode & FMODE_UNSIGNED_OFFSET))
2650                 ret = -EINVAL;
2651         if (!ret && offset > inode->i_sb->s_maxbytes)
2652                 ret = -EINVAL;
2653         if (ret)
2654                 goto out;
2655
2656         if (offset != file->f_pos) {
2657                 file->f_pos = offset;
2658                 file->f_version = 0;
2659         }
2660
2661 out:
2662         mutex_unlock(&inode->i_mutex);
2663         if (ret)
2664                 return ret;
2665         return offset;
2666 }
2667
2668 const struct inode_operations ocfs2_file_iops = {
2669         .setattr        = ocfs2_setattr,
2670         .getattr        = ocfs2_getattr,
2671         .permission     = ocfs2_permission,
2672         .setxattr       = generic_setxattr,
2673         .getxattr       = generic_getxattr,
2674         .listxattr      = ocfs2_listxattr,
2675         .removexattr    = generic_removexattr,
2676         .fiemap         = ocfs2_fiemap,
2677         .get_acl        = ocfs2_iop_get_acl,
2678 };
2679
2680 const struct inode_operations ocfs2_special_file_iops = {
2681         .setattr        = ocfs2_setattr,
2682         .getattr        = ocfs2_getattr,
2683         .permission     = ocfs2_permission,
2684         .get_acl        = ocfs2_iop_get_acl,
2685 };
2686
2687 /*
2688  * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with
2689  * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks!
2690  */
2691 const struct file_operations ocfs2_fops = {
2692         .llseek         = ocfs2_file_llseek,
2693         .read           = do_sync_read,
2694         .write          = do_sync_write,
2695         .mmap           = ocfs2_mmap,
2696         .fsync          = ocfs2_sync_file,
2697         .release        = ocfs2_file_release,
2698         .open           = ocfs2_file_open,
2699         .aio_read       = ocfs2_file_aio_read,
2700         .aio_write      = ocfs2_file_aio_write,
2701         .unlocked_ioctl = ocfs2_ioctl,
2702 #ifdef CONFIG_COMPAT
2703         .compat_ioctl   = ocfs2_compat_ioctl,
2704 #endif
2705         .lock           = ocfs2_lock,
2706         .flock          = ocfs2_flock,
2707         .splice_read    = ocfs2_file_splice_read,
2708         .splice_write   = ocfs2_file_splice_write,
2709         .fallocate      = ocfs2_fallocate,
2710 };
2711
2712 const struct file_operations ocfs2_dops = {
2713         .llseek         = generic_file_llseek,
2714         .read           = generic_read_dir,
2715         .readdir        = ocfs2_readdir,
2716         .fsync          = ocfs2_sync_file,
2717         .release        = ocfs2_dir_release,
2718         .open           = ocfs2_dir_open,
2719         .unlocked_ioctl = ocfs2_ioctl,
2720 #ifdef CONFIG_COMPAT
2721         .compat_ioctl   = ocfs2_compat_ioctl,
2722 #endif
2723         .lock           = ocfs2_lock,
2724         .flock          = ocfs2_flock,
2725 };
2726
2727 /*
2728  * POSIX-lockless variants of our file_operations.
2729  *
2730  * These will be used if the underlying cluster stack does not support
2731  * posix file locking, if the user passes the "localflocks" mount
2732  * option, or if we have a local-only fs.
2733  *
2734  * ocfs2_flock is in here because all stacks handle UNIX file locks,
2735  * so we still want it in the case of no stack support for
2736  * plocks. Internally, it will do the right thing when asked to ignore
2737  * the cluster.
2738  */
2739 const struct file_operations ocfs2_fops_no_plocks = {
2740         .llseek         = ocfs2_file_llseek,
2741         .read           = do_sync_read,
2742         .write          = do_sync_write,
2743         .mmap           = ocfs2_mmap,
2744         .fsync          = ocfs2_sync_file,
2745         .release        = ocfs2_file_release,
2746         .open           = ocfs2_file_open,
2747         .aio_read       = ocfs2_file_aio_read,
2748         .aio_write      = ocfs2_file_aio_write,
2749         .unlocked_ioctl = ocfs2_ioctl,
2750 #ifdef CONFIG_COMPAT
2751         .compat_ioctl   = ocfs2_compat_ioctl,
2752 #endif
2753         .flock          = ocfs2_flock,
2754         .splice_read    = ocfs2_file_splice_read,
2755         .splice_write   = ocfs2_file_splice_write,
2756         .fallocate      = ocfs2_fallocate,
2757 };
2758
2759 const struct file_operations ocfs2_dops_no_plocks = {
2760         .llseek         = generic_file_llseek,
2761         .read           = generic_read_dir,
2762         .readdir        = ocfs2_readdir,
2763         .fsync          = ocfs2_sync_file,
2764         .release        = ocfs2_dir_release,
2765         .open           = ocfs2_dir_open,
2766         .unlocked_ioctl = ocfs2_ioctl,
2767 #ifdef CONFIG_COMPAT
2768         .compat_ioctl   = ocfs2_compat_ioctl,
2769 #endif
2770         .flock          = ocfs2_flock,
2771 };