ocfs2_file_write_iter: keep return value and current position update in sync
[pandora-kernel.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         struct inode *inode = file->f_mapping->host;
179         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
180         struct ocfs2_inode_info *oi = OCFS2_I(inode);
181         journal_t *journal = osb->journal->j_journal;
182         int ret;
183         tid_t commit_tid;
184         bool needs_barrier = false;
185
186         trace_ocfs2_sync_file(inode, file, file->f_path.dentry,
187                               OCFS2_I(inode)->ip_blkno,
188                               file->f_path.dentry->d_name.len,
189                               file->f_path.dentry->d_name.name,
190                               (unsigned long long)datasync);
191
192         if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
193                 return -EROFS;
194
195         err = filemap_write_and_wait_range(inode->i_mapping, start, end);
196         if (err)
197                 return err;
198
199         commit_tid = datasync ? oi->i_datasync_tid : oi->i_sync_tid;
200         if (journal->j_flags & JBD2_BARRIER &&
201             !jbd2_trans_will_send_data_barrier(journal, commit_tid))
202                 needs_barrier = true;
203         err = jbd2_complete_transaction(journal, commit_tid);
204         if (needs_barrier) {
205                 ret = blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, NULL);
206                 if (!err)
207                         err = ret;
208         }
209
210         if (err)
211                 mlog_errno(err);
212
213         return (err < 0) ? -EIO : 0;
214 }
215
216 int ocfs2_should_update_atime(struct inode *inode,
217                               struct vfsmount *vfsmnt)
218 {
219         struct timespec now;
220         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
221
222         if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
223                 return 0;
224
225         if ((inode->i_flags & S_NOATIME) ||
226             ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)))
227                 return 0;
228
229         /*
230          * We can be called with no vfsmnt structure - NFSD will
231          * sometimes do this.
232          *
233          * Note that our action here is different than touch_atime() -
234          * if we can't tell whether this is a noatime mount, then we
235          * don't know whether to trust the value of s_atime_quantum.
236          */
237         if (vfsmnt == NULL)
238                 return 0;
239
240         if ((vfsmnt->mnt_flags & MNT_NOATIME) ||
241             ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
242                 return 0;
243
244         if (vfsmnt->mnt_flags & MNT_RELATIME) {
245                 if ((timespec_compare(&inode->i_atime, &inode->i_mtime) <= 0) ||
246                     (timespec_compare(&inode->i_atime, &inode->i_ctime) <= 0))
247                         return 1;
248
249                 return 0;
250         }
251
252         now = CURRENT_TIME;
253         if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum))
254                 return 0;
255         else
256                 return 1;
257 }
258
259 int ocfs2_update_inode_atime(struct inode *inode,
260                              struct buffer_head *bh)
261 {
262         int ret;
263         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
264         handle_t *handle;
265         struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data;
266
267         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
268         if (IS_ERR(handle)) {
269                 ret = PTR_ERR(handle);
270                 mlog_errno(ret);
271                 goto out;
272         }
273
274         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
275                                       OCFS2_JOURNAL_ACCESS_WRITE);
276         if (ret) {
277                 mlog_errno(ret);
278                 goto out_commit;
279         }
280
281         /*
282          * Don't use ocfs2_mark_inode_dirty() here as we don't always
283          * have i_mutex to guard against concurrent changes to other
284          * inode fields.
285          */
286         inode->i_atime = CURRENT_TIME;
287         di->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
288         di->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
289         ocfs2_update_inode_fsync_trans(handle, inode, 0);
290         ocfs2_journal_dirty(handle, bh);
291
292 out_commit:
293         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
294 out:
295         return ret;
296 }
297
298 int ocfs2_set_inode_size(handle_t *handle,
299                                 struct inode *inode,
300                                 struct buffer_head *fe_bh,
301                                 u64 new_i_size)
302 {
303         int status;
304
305         i_size_write(inode, new_i_size);
306         inode->i_blocks = ocfs2_inode_sector_count(inode);
307         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
308
309         status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
310         if (status < 0) {
311                 mlog_errno(status);
312                 goto bail;
313         }
314
315 bail:
316         return status;
317 }
318
319 int ocfs2_simple_size_update(struct inode *inode,
320                              struct buffer_head *di_bh,
321                              u64 new_i_size)
322 {
323         int ret;
324         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
325         handle_t *handle = NULL;
326
327         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
328         if (IS_ERR(handle)) {
329                 ret = PTR_ERR(handle);
330                 mlog_errno(ret);
331                 goto out;
332         }
333
334         ret = ocfs2_set_inode_size(handle, inode, di_bh,
335                                    new_i_size);
336         if (ret < 0)
337                 mlog_errno(ret);
338
339         ocfs2_update_inode_fsync_trans(handle, inode, 0);
340         ocfs2_commit_trans(osb, handle);
341 out:
342         return ret;
343 }
344
345 static int ocfs2_cow_file_pos(struct inode *inode,
346                               struct buffer_head *fe_bh,
347                               u64 offset)
348 {
349         int status;
350         u32 phys, cpos = offset >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
351         unsigned int num_clusters = 0;
352         unsigned int ext_flags = 0;
353
354         /*
355          * If the new offset is aligned to the range of the cluster, there is
356          * no space for ocfs2_zero_range_for_truncate to fill, so no need to
357          * CoW either.
358          */
359         if ((offset & (OCFS2_SB(inode->i_sb)->s_clustersize - 1)) == 0)
360                 return 0;
361
362         status = ocfs2_get_clusters(inode, cpos, &phys,
363                                     &num_clusters, &ext_flags);
364         if (status) {
365                 mlog_errno(status);
366                 goto out;
367         }
368
369         if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
370                 goto out;
371
372         return ocfs2_refcount_cow(inode, fe_bh, cpos, 1, cpos+1);
373
374 out:
375         return status;
376 }
377
378 static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
379                                      struct inode *inode,
380                                      struct buffer_head *fe_bh,
381                                      u64 new_i_size)
382 {
383         int status;
384         handle_t *handle;
385         struct ocfs2_dinode *di;
386         u64 cluster_bytes;
387
388         /*
389          * We need to CoW the cluster contains the offset if it is reflinked
390          * since we will call ocfs2_zero_range_for_truncate later which will
391          * write "0" from offset to the end of the cluster.
392          */
393         status = ocfs2_cow_file_pos(inode, fe_bh, new_i_size);
394         if (status) {
395                 mlog_errno(status);
396                 return status;
397         }
398
399         /* TODO: This needs to actually orphan the inode in this
400          * transaction. */
401
402         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
403         if (IS_ERR(handle)) {
404                 status = PTR_ERR(handle);
405                 mlog_errno(status);
406                 goto out;
407         }
408
409         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
410                                          OCFS2_JOURNAL_ACCESS_WRITE);
411         if (status < 0) {
412                 mlog_errno(status);
413                 goto out_commit;
414         }
415
416         /*
417          * Do this before setting i_size.
418          */
419         cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size);
420         status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size,
421                                                cluster_bytes);
422         if (status) {
423                 mlog_errno(status);
424                 goto out_commit;
425         }
426
427         i_size_write(inode, new_i_size);
428         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
429
430         di = (struct ocfs2_dinode *) fe_bh->b_data;
431         di->i_size = cpu_to_le64(new_i_size);
432         di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
433         di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
434         ocfs2_update_inode_fsync_trans(handle, inode, 0);
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 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         down_write(&OCFS2_I(inode)->ip_alloc_sem);
478
479         ocfs2_resv_discard(&osb->osb_la_resmap,
480                            &OCFS2_I(inode)->ip_la_data_resv);
481
482         /*
483          * The inode lock forced other nodes to sync and drop their
484          * pages, which (correctly) happens even if we have a truncate
485          * without allocation change - ocfs2 cluster sizes can be much
486          * greater than page size, so we have to truncate them
487          * anyway.
488          */
489         unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1);
490         truncate_inode_pages(inode->i_mapping, new_i_size);
491
492         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
493                 status = ocfs2_truncate_inline(inode, di_bh, new_i_size,
494                                                i_size_read(inode), 1);
495                 if (status)
496                         mlog_errno(status);
497
498                 goto bail_unlock_sem;
499         }
500
501         /* alright, we're going to need to do a full blown alloc size
502          * change. Orphan the inode so that recovery can complete the
503          * truncate if necessary. This does the task of marking
504          * i_size. */
505         status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
506         if (status < 0) {
507                 mlog_errno(status);
508                 goto bail_unlock_sem;
509         }
510
511         status = ocfs2_commit_truncate(osb, inode, di_bh);
512         if (status < 0) {
513                 mlog_errno(status);
514                 goto bail_unlock_sem;
515         }
516
517         /* TODO: orphan dir cleanup here. */
518 bail_unlock_sem:
519         up_write(&OCFS2_I(inode)->ip_alloc_sem);
520
521 bail:
522         if (!status && OCFS2_I(inode)->ip_clusters == 0)
523                 status = ocfs2_try_remove_refcount_tree(inode, di_bh);
524
525         return status;
526 }
527
528 /*
529  * extend file allocation only here.
530  * we'll update all the disk stuff, and oip->alloc_size
531  *
532  * expect stuff to be locked, a transaction started and enough data /
533  * metadata reservations in the contexts.
534  *
535  * Will return -EAGAIN, and a reason if a restart is needed.
536  * If passed in, *reason will always be set, even in error.
537  */
538 int ocfs2_add_inode_data(struct ocfs2_super *osb,
539                          struct inode *inode,
540                          u32 *logical_offset,
541                          u32 clusters_to_add,
542                          int mark_unwritten,
543                          struct buffer_head *fe_bh,
544                          handle_t *handle,
545                          struct ocfs2_alloc_context *data_ac,
546                          struct ocfs2_alloc_context *meta_ac,
547                          enum ocfs2_alloc_restarted *reason_ret)
548 {
549         int ret;
550         struct ocfs2_extent_tree et;
551
552         ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), fe_bh);
553         ret = ocfs2_add_clusters_in_btree(handle, &et, logical_offset,
554                                           clusters_to_add, mark_unwritten,
555                                           data_ac, meta_ac, reason_ret);
556
557         return ret;
558 }
559
560 static int __ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
561                                      u32 clusters_to_add, int mark_unwritten)
562 {
563         int status = 0;
564         int restart_func = 0;
565         int credits;
566         u32 prev_clusters;
567         struct buffer_head *bh = NULL;
568         struct ocfs2_dinode *fe = NULL;
569         handle_t *handle = NULL;
570         struct ocfs2_alloc_context *data_ac = NULL;
571         struct ocfs2_alloc_context *meta_ac = NULL;
572         enum ocfs2_alloc_restarted why = RESTART_NONE;
573         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
574         struct ocfs2_extent_tree et;
575         int did_quota = 0;
576
577         /*
578          * Unwritten extent only exists for file systems which
579          * support holes.
580          */
581         BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb));
582
583         status = ocfs2_read_inode_block(inode, &bh);
584         if (status < 0) {
585                 mlog_errno(status);
586                 goto leave;
587         }
588         fe = (struct ocfs2_dinode *) bh->b_data;
589
590 restart_all:
591         BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
592
593         ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), bh);
594         status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
595                                        &data_ac, &meta_ac);
596         if (status) {
597                 mlog_errno(status);
598                 goto leave;
599         }
600
601         credits = ocfs2_calc_extend_credits(osb->sb, &fe->id2.i_list);
602         handle = ocfs2_start_trans(osb, credits);
603         if (IS_ERR(handle)) {
604                 status = PTR_ERR(handle);
605                 handle = NULL;
606                 mlog_errno(status);
607                 goto leave;
608         }
609
610 restarted_transaction:
611         trace_ocfs2_extend_allocation(
612                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
613                 (unsigned long long)i_size_read(inode),
614                 le32_to_cpu(fe->i_clusters), clusters_to_add,
615                 why, restart_func);
616
617         status = dquot_alloc_space_nodirty(inode,
618                         ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
619         if (status)
620                 goto leave;
621         did_quota = 1;
622
623         /* reserve a write to the file entry early on - that we if we
624          * run out of credits in the allocation path, we can still
625          * update i_size. */
626         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
627                                          OCFS2_JOURNAL_ACCESS_WRITE);
628         if (status < 0) {
629                 mlog_errno(status);
630                 goto leave;
631         }
632
633         prev_clusters = OCFS2_I(inode)->ip_clusters;
634
635         status = ocfs2_add_inode_data(osb,
636                                       inode,
637                                       &logical_start,
638                                       clusters_to_add,
639                                       mark_unwritten,
640                                       bh,
641                                       handle,
642                                       data_ac,
643                                       meta_ac,
644                                       &why);
645         if ((status < 0) && (status != -EAGAIN)) {
646                 if (status != -ENOSPC)
647                         mlog_errno(status);
648                 goto leave;
649         }
650         ocfs2_update_inode_fsync_trans(handle, inode, 1);
651         ocfs2_journal_dirty(handle, bh);
652
653         spin_lock(&OCFS2_I(inode)->ip_lock);
654         clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
655         spin_unlock(&OCFS2_I(inode)->ip_lock);
656         /* Release unused quota reservation */
657         dquot_free_space(inode,
658                         ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
659         did_quota = 0;
660
661         if (why != RESTART_NONE && clusters_to_add) {
662                 if (why == RESTART_META) {
663                         restart_func = 1;
664                         status = 0;
665                 } else {
666                         BUG_ON(why != RESTART_TRANS);
667
668                         status = ocfs2_allocate_extend_trans(handle, 1);
669                         if (status < 0) {
670                                 /* handle still has to be committed at
671                                  * this point. */
672                                 status = -ENOMEM;
673                                 mlog_errno(status);
674                                 goto leave;
675                         }
676                         goto restarted_transaction;
677                 }
678         }
679
680         trace_ocfs2_extend_allocation_end(OCFS2_I(inode)->ip_blkno,
681              le32_to_cpu(fe->i_clusters),
682              (unsigned long long)le64_to_cpu(fe->i_size),
683              OCFS2_I(inode)->ip_clusters,
684              (unsigned long long)i_size_read(inode));
685
686 leave:
687         if (status < 0 && did_quota)
688                 dquot_free_space(inode,
689                         ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
690         if (handle) {
691                 ocfs2_commit_trans(osb, handle);
692                 handle = NULL;
693         }
694         if (data_ac) {
695                 ocfs2_free_alloc_context(data_ac);
696                 data_ac = NULL;
697         }
698         if (meta_ac) {
699                 ocfs2_free_alloc_context(meta_ac);
700                 meta_ac = NULL;
701         }
702         if ((!status) && restart_func) {
703                 restart_func = 0;
704                 goto restart_all;
705         }
706         brelse(bh);
707         bh = NULL;
708
709         return status;
710 }
711
712 int ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
713                 u32 clusters_to_add, int mark_unwritten)
714 {
715         return __ocfs2_extend_allocation(inode, logical_start,
716                         clusters_to_add, mark_unwritten);
717 }
718
719 /*
720  * While a write will already be ordering the data, a truncate will not.
721  * Thus, we need to explicitly order the zeroed pages.
722  */
723 static handle_t *ocfs2_zero_start_ordered_transaction(struct inode *inode,
724                                                 struct buffer_head *di_bh)
725 {
726         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
727         handle_t *handle = NULL;
728         int ret = 0;
729
730         if (!ocfs2_should_order_data(inode))
731                 goto out;
732
733         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
734         if (IS_ERR(handle)) {
735                 ret = -ENOMEM;
736                 mlog_errno(ret);
737                 goto out;
738         }
739
740         ret = ocfs2_jbd2_file_inode(handle, inode);
741         if (ret < 0) {
742                 mlog_errno(ret);
743                 goto out;
744         }
745
746         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
747                                       OCFS2_JOURNAL_ACCESS_WRITE);
748         if (ret)
749                 mlog_errno(ret);
750         ocfs2_update_inode_fsync_trans(handle, inode, 1);
751
752 out:
753         if (ret) {
754                 if (!IS_ERR(handle))
755                         ocfs2_commit_trans(osb, handle);
756                 handle = ERR_PTR(ret);
757         }
758         return handle;
759 }
760
761 /* Some parts of this taken from generic_cont_expand, which turned out
762  * to be too fragile to do exactly what we need without us having to
763  * worry about recursive locking in ->write_begin() and ->write_end(). */
764 static int ocfs2_write_zero_page(struct inode *inode, u64 abs_from,
765                                  u64 abs_to, struct buffer_head *di_bh)
766 {
767         struct address_space *mapping = inode->i_mapping;
768         struct page *page;
769         unsigned long index = abs_from >> PAGE_CACHE_SHIFT;
770         handle_t *handle;
771         int ret = 0;
772         unsigned zero_from, zero_to, block_start, block_end;
773         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
774
775         BUG_ON(abs_from >= abs_to);
776         BUG_ON(abs_to > (((u64)index + 1) << PAGE_CACHE_SHIFT));
777         BUG_ON(abs_from & (inode->i_blkbits - 1));
778
779         handle = ocfs2_zero_start_ordered_transaction(inode, di_bh);
780         if (IS_ERR(handle)) {
781                 ret = PTR_ERR(handle);
782                 goto out;
783         }
784
785         page = find_or_create_page(mapping, index, GFP_NOFS);
786         if (!page) {
787                 ret = -ENOMEM;
788                 mlog_errno(ret);
789                 goto out_commit_trans;
790         }
791
792         /* Get the offsets within the page that we want to zero */
793         zero_from = abs_from & (PAGE_CACHE_SIZE - 1);
794         zero_to = abs_to & (PAGE_CACHE_SIZE - 1);
795         if (!zero_to)
796                 zero_to = PAGE_CACHE_SIZE;
797
798         trace_ocfs2_write_zero_page(
799                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
800                         (unsigned long long)abs_from,
801                         (unsigned long long)abs_to,
802                         index, zero_from, zero_to);
803
804         /* We know that zero_from is block aligned */
805         for (block_start = zero_from; block_start < zero_to;
806              block_start = block_end) {
807                 block_end = block_start + (1 << inode->i_blkbits);
808
809                 /*
810                  * block_start is block-aligned.  Bump it by one to force
811                  * __block_write_begin and block_commit_write to zero the
812                  * whole block.
813                  */
814                 ret = __block_write_begin(page, block_start + 1, 0,
815                                           ocfs2_get_block);
816                 if (ret < 0) {
817                         mlog_errno(ret);
818                         goto out_unlock;
819                 }
820
821
822                 /* must not update i_size! */
823                 ret = block_commit_write(page, block_start + 1,
824                                          block_start + 1);
825                 if (ret < 0)
826                         mlog_errno(ret);
827                 else
828                         ret = 0;
829         }
830
831         /*
832          * fs-writeback will release the dirty pages without page lock
833          * whose offset are over inode size, the release happens at
834          * block_write_full_page().
835          */
836         i_size_write(inode, abs_to);
837         inode->i_blocks = ocfs2_inode_sector_count(inode);
838         di->i_size = cpu_to_le64((u64)i_size_read(inode));
839         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
840         di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec);
841         di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
842         di->i_mtime_nsec = di->i_ctime_nsec;
843         if (handle) {
844                 ocfs2_journal_dirty(handle, di_bh);
845                 ocfs2_update_inode_fsync_trans(handle, inode, 1);
846         }
847
848 out_unlock:
849         unlock_page(page);
850         page_cache_release(page);
851 out_commit_trans:
852         if (handle)
853                 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
854 out:
855         return ret;
856 }
857
858 /*
859  * Find the next range to zero.  We do this in terms of bytes because
860  * that's what ocfs2_zero_extend() wants, and it is dealing with the
861  * pagecache.  We may return multiple extents.
862  *
863  * zero_start and zero_end are ocfs2_zero_extend()s current idea of what
864  * needs to be zeroed.  range_start and range_end return the next zeroing
865  * range.  A subsequent call should pass the previous range_end as its
866  * zero_start.  If range_end is 0, there's nothing to do.
867  *
868  * Unwritten extents are skipped over.  Refcounted extents are CoWd.
869  */
870 static int ocfs2_zero_extend_get_range(struct inode *inode,
871                                        struct buffer_head *di_bh,
872                                        u64 zero_start, u64 zero_end,
873                                        u64 *range_start, u64 *range_end)
874 {
875         int rc = 0, needs_cow = 0;
876         u32 p_cpos, zero_clusters = 0;
877         u32 zero_cpos =
878                 zero_start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
879         u32 last_cpos = ocfs2_clusters_for_bytes(inode->i_sb, zero_end);
880         unsigned int num_clusters = 0;
881         unsigned int ext_flags = 0;
882
883         while (zero_cpos < last_cpos) {
884                 rc = ocfs2_get_clusters(inode, zero_cpos, &p_cpos,
885                                         &num_clusters, &ext_flags);
886                 if (rc) {
887                         mlog_errno(rc);
888                         goto out;
889                 }
890
891                 if (p_cpos && !(ext_flags & OCFS2_EXT_UNWRITTEN)) {
892                         zero_clusters = num_clusters;
893                         if (ext_flags & OCFS2_EXT_REFCOUNTED)
894                                 needs_cow = 1;
895                         break;
896                 }
897
898                 zero_cpos += num_clusters;
899         }
900         if (!zero_clusters) {
901                 *range_end = 0;
902                 goto out;
903         }
904
905         while ((zero_cpos + zero_clusters) < last_cpos) {
906                 rc = ocfs2_get_clusters(inode, zero_cpos + zero_clusters,
907                                         &p_cpos, &num_clusters,
908                                         &ext_flags);
909                 if (rc) {
910                         mlog_errno(rc);
911                         goto out;
912                 }
913
914                 if (!p_cpos || (ext_flags & OCFS2_EXT_UNWRITTEN))
915                         break;
916                 if (ext_flags & OCFS2_EXT_REFCOUNTED)
917                         needs_cow = 1;
918                 zero_clusters += num_clusters;
919         }
920         if ((zero_cpos + zero_clusters) > last_cpos)
921                 zero_clusters = last_cpos - zero_cpos;
922
923         if (needs_cow) {
924                 rc = ocfs2_refcount_cow(inode, di_bh, zero_cpos,
925                                         zero_clusters, UINT_MAX);
926                 if (rc) {
927                         mlog_errno(rc);
928                         goto out;
929                 }
930         }
931
932         *range_start = ocfs2_clusters_to_bytes(inode->i_sb, zero_cpos);
933         *range_end = ocfs2_clusters_to_bytes(inode->i_sb,
934                                              zero_cpos + zero_clusters);
935
936 out:
937         return rc;
938 }
939
940 /*
941  * Zero one range returned from ocfs2_zero_extend_get_range().  The caller
942  * has made sure that the entire range needs zeroing.
943  */
944 static int ocfs2_zero_extend_range(struct inode *inode, u64 range_start,
945                                    u64 range_end, struct buffer_head *di_bh)
946 {
947         int rc = 0;
948         u64 next_pos;
949         u64 zero_pos = range_start;
950
951         trace_ocfs2_zero_extend_range(
952                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
953                         (unsigned long long)range_start,
954                         (unsigned long long)range_end);
955         BUG_ON(range_start >= range_end);
956
957         while (zero_pos < range_end) {
958                 next_pos = (zero_pos & PAGE_CACHE_MASK) + PAGE_CACHE_SIZE;
959                 if (next_pos > range_end)
960                         next_pos = range_end;
961                 rc = ocfs2_write_zero_page(inode, zero_pos, next_pos, di_bh);
962                 if (rc < 0) {
963                         mlog_errno(rc);
964                         break;
965                 }
966                 zero_pos = next_pos;
967
968                 /*
969                  * Very large extends have the potential to lock up
970                  * the cpu for extended periods of time.
971                  */
972                 cond_resched();
973         }
974
975         return rc;
976 }
977
978 int ocfs2_zero_extend(struct inode *inode, struct buffer_head *di_bh,
979                       loff_t zero_to_size)
980 {
981         int ret = 0;
982         u64 zero_start, range_start = 0, range_end = 0;
983         struct super_block *sb = inode->i_sb;
984
985         zero_start = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
986         trace_ocfs2_zero_extend((unsigned long long)OCFS2_I(inode)->ip_blkno,
987                                 (unsigned long long)zero_start,
988                                 (unsigned long long)i_size_read(inode));
989         while (zero_start < zero_to_size) {
990                 ret = ocfs2_zero_extend_get_range(inode, di_bh, zero_start,
991                                                   zero_to_size,
992                                                   &range_start,
993                                                   &range_end);
994                 if (ret) {
995                         mlog_errno(ret);
996                         break;
997                 }
998                 if (!range_end)
999                         break;
1000                 /* Trim the ends */
1001                 if (range_start < zero_start)
1002                         range_start = zero_start;
1003                 if (range_end > zero_to_size)
1004                         range_end = zero_to_size;
1005
1006                 ret = ocfs2_zero_extend_range(inode, range_start,
1007                                               range_end, di_bh);
1008                 if (ret) {
1009                         mlog_errno(ret);
1010                         break;
1011                 }
1012                 zero_start = range_end;
1013         }
1014
1015         return ret;
1016 }
1017
1018 int ocfs2_extend_no_holes(struct inode *inode, struct buffer_head *di_bh,
1019                           u64 new_i_size, u64 zero_to)
1020 {
1021         int ret;
1022         u32 clusters_to_add;
1023         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1024
1025         /*
1026          * Only quota files call this without a bh, and they can't be
1027          * refcounted.
1028          */
1029         BUG_ON(!di_bh && (oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
1030         BUG_ON(!di_bh && !(oi->ip_flags & OCFS2_INODE_SYSTEM_FILE));
1031
1032         clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size);
1033         if (clusters_to_add < oi->ip_clusters)
1034                 clusters_to_add = 0;
1035         else
1036                 clusters_to_add -= oi->ip_clusters;
1037
1038         if (clusters_to_add) {
1039                 ret = __ocfs2_extend_allocation(inode, oi->ip_clusters,
1040                                                 clusters_to_add, 0);
1041                 if (ret) {
1042                         mlog_errno(ret);
1043                         goto out;
1044                 }
1045         }
1046
1047         /*
1048          * Call this even if we don't add any clusters to the tree. We
1049          * still need to zero the area between the old i_size and the
1050          * new i_size.
1051          */
1052         ret = ocfs2_zero_extend(inode, di_bh, zero_to);
1053         if (ret < 0)
1054                 mlog_errno(ret);
1055
1056 out:
1057         return ret;
1058 }
1059
1060 static int ocfs2_extend_file(struct inode *inode,
1061                              struct buffer_head *di_bh,
1062                              u64 new_i_size)
1063 {
1064         int ret = 0;
1065         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1066
1067         BUG_ON(!di_bh);
1068
1069         /* setattr sometimes calls us like this. */
1070         if (new_i_size == 0)
1071                 goto out;
1072
1073         if (i_size_read(inode) == new_i_size)
1074                 goto out;
1075         BUG_ON(new_i_size < i_size_read(inode));
1076
1077         /*
1078          * The alloc sem blocks people in read/write from reading our
1079          * allocation until we're done changing it. We depend on
1080          * i_mutex to block other extend/truncate calls while we're
1081          * here.  We even have to hold it for sparse files because there
1082          * might be some tail zeroing.
1083          */
1084         down_write(&oi->ip_alloc_sem);
1085
1086         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1087                 /*
1088                  * We can optimize small extends by keeping the inodes
1089                  * inline data.
1090                  */
1091                 if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) {
1092                         up_write(&oi->ip_alloc_sem);
1093                         goto out_update_size;
1094                 }
1095
1096                 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1097                 if (ret) {
1098                         up_write(&oi->ip_alloc_sem);
1099                         mlog_errno(ret);
1100                         goto out;
1101                 }
1102         }
1103
1104         if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
1105                 ret = ocfs2_zero_extend(inode, di_bh, new_i_size);
1106         else
1107                 ret = ocfs2_extend_no_holes(inode, di_bh, new_i_size,
1108                                             new_i_size);
1109
1110         up_write(&oi->ip_alloc_sem);
1111
1112         if (ret < 0) {
1113                 mlog_errno(ret);
1114                 goto out;
1115         }
1116
1117 out_update_size:
1118         ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
1119         if (ret < 0)
1120                 mlog_errno(ret);
1121
1122 out:
1123         return ret;
1124 }
1125
1126 int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
1127 {
1128         int status = 0, size_change;
1129         struct inode *inode = dentry->d_inode;
1130         struct super_block *sb = inode->i_sb;
1131         struct ocfs2_super *osb = OCFS2_SB(sb);
1132         struct buffer_head *bh = NULL;
1133         handle_t *handle = NULL;
1134         struct dquot *transfer_to[MAXQUOTAS] = { };
1135         int qtype;
1136
1137         trace_ocfs2_setattr(inode, dentry,
1138                             (unsigned long long)OCFS2_I(inode)->ip_blkno,
1139                             dentry->d_name.len, dentry->d_name.name,
1140                             attr->ia_valid, attr->ia_mode,
1141                             from_kuid(&init_user_ns, attr->ia_uid),
1142                             from_kgid(&init_user_ns, attr->ia_gid));
1143
1144         /* ensuring we don't even attempt to truncate a symlink */
1145         if (S_ISLNK(inode->i_mode))
1146                 attr->ia_valid &= ~ATTR_SIZE;
1147
1148 #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
1149                            | ATTR_GID | ATTR_UID | ATTR_MODE)
1150         if (!(attr->ia_valid & OCFS2_VALID_ATTRS))
1151                 return 0;
1152
1153         status = inode_change_ok(inode, attr);
1154         if (status)
1155                 return status;
1156
1157         if (is_quota_modification(inode, attr))
1158                 dquot_initialize(inode);
1159         size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
1160         if (size_change) {
1161                 status = ocfs2_rw_lock(inode, 1);
1162                 if (status < 0) {
1163                         mlog_errno(status);
1164                         goto bail;
1165                 }
1166         }
1167
1168         status = ocfs2_inode_lock(inode, &bh, 1);
1169         if (status < 0) {
1170                 if (status != -ENOENT)
1171                         mlog_errno(status);
1172                 goto bail_unlock_rw;
1173         }
1174
1175         if (size_change) {
1176                 status = inode_newsize_ok(inode, attr->ia_size);
1177                 if (status)
1178                         goto bail_unlock;
1179
1180                 inode_dio_wait(inode);
1181
1182                 if (i_size_read(inode) >= attr->ia_size) {
1183                         if (ocfs2_should_order_data(inode)) {
1184                                 status = ocfs2_begin_ordered_truncate(inode,
1185                                                                       attr->ia_size);
1186                                 if (status)
1187                                         goto bail_unlock;
1188                         }
1189                         status = ocfs2_truncate_file(inode, bh, attr->ia_size);
1190                 } else
1191                         status = ocfs2_extend_file(inode, bh, attr->ia_size);
1192                 if (status < 0) {
1193                         if (status != -ENOSPC)
1194                                 mlog_errno(status);
1195                         status = -ENOSPC;
1196                         goto bail_unlock;
1197                 }
1198         }
1199
1200         if ((attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
1201             (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
1202                 /*
1203                  * Gather pointers to quota structures so that allocation /
1204                  * freeing of quota structures happens here and not inside
1205                  * dquot_transfer() where we have problems with lock ordering
1206                  */
1207                 if (attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)
1208                     && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1209                     OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
1210                         transfer_to[USRQUOTA] = dqget(sb, make_kqid_uid(attr->ia_uid));
1211                         if (!transfer_to[USRQUOTA]) {
1212                                 status = -ESRCH;
1213                                 goto bail_unlock;
1214                         }
1215                 }
1216                 if (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid)
1217                     && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1218                     OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
1219                         transfer_to[GRPQUOTA] = dqget(sb, make_kqid_gid(attr->ia_gid));
1220                         if (!transfer_to[GRPQUOTA]) {
1221                                 status = -ESRCH;
1222                                 goto bail_unlock;
1223                         }
1224                 }
1225                 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS +
1226                                            2 * ocfs2_quota_trans_credits(sb));
1227                 if (IS_ERR(handle)) {
1228                         status = PTR_ERR(handle);
1229                         mlog_errno(status);
1230                         goto bail_unlock;
1231                 }
1232                 status = __dquot_transfer(inode, transfer_to);
1233                 if (status < 0)
1234                         goto bail_commit;
1235         } else {
1236                 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1237                 if (IS_ERR(handle)) {
1238                         status = PTR_ERR(handle);
1239                         mlog_errno(status);
1240                         goto bail_unlock;
1241                 }
1242         }
1243
1244         setattr_copy(inode, attr);
1245         mark_inode_dirty(inode);
1246
1247         status = ocfs2_mark_inode_dirty(handle, inode, bh);
1248         if (status < 0)
1249                 mlog_errno(status);
1250
1251 bail_commit:
1252         ocfs2_commit_trans(osb, handle);
1253 bail_unlock:
1254         ocfs2_inode_unlock(inode, 1);
1255 bail_unlock_rw:
1256         if (size_change)
1257                 ocfs2_rw_unlock(inode, 1);
1258 bail:
1259         brelse(bh);
1260
1261         /* Release quota pointers in case we acquired them */
1262         for (qtype = 0; qtype < OCFS2_MAXQUOTAS; qtype++)
1263                 dqput(transfer_to[qtype]);
1264
1265         if (!status && attr->ia_valid & ATTR_MODE) {
1266                 status = posix_acl_chmod(inode, inode->i_mode);
1267                 if (status < 0)
1268                         mlog_errno(status);
1269         }
1270
1271         return status;
1272 }
1273
1274 int ocfs2_getattr(struct vfsmount *mnt,
1275                   struct dentry *dentry,
1276                   struct kstat *stat)
1277 {
1278         struct inode *inode = dentry->d_inode;
1279         struct super_block *sb = dentry->d_inode->i_sb;
1280         struct ocfs2_super *osb = sb->s_fs_info;
1281         int err;
1282
1283         err = ocfs2_inode_revalidate(dentry);
1284         if (err) {
1285                 if (err != -ENOENT)
1286                         mlog_errno(err);
1287                 goto bail;
1288         }
1289
1290         generic_fillattr(inode, stat);
1291
1292         /* We set the blksize from the cluster size for performance */
1293         stat->blksize = osb->s_clustersize;
1294
1295 bail:
1296         return err;
1297 }
1298
1299 int ocfs2_permission(struct inode *inode, int mask)
1300 {
1301         int ret;
1302
1303         if (mask & MAY_NOT_BLOCK)
1304                 return -ECHILD;
1305
1306         ret = ocfs2_inode_lock(inode, NULL, 0);
1307         if (ret) {
1308                 if (ret != -ENOENT)
1309                         mlog_errno(ret);
1310                 goto out;
1311         }
1312
1313         ret = generic_permission(inode, mask);
1314
1315         ocfs2_inode_unlock(inode, 0);
1316 out:
1317         return ret;
1318 }
1319
1320 static int __ocfs2_write_remove_suid(struct inode *inode,
1321                                      struct buffer_head *bh)
1322 {
1323         int ret;
1324         handle_t *handle;
1325         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1326         struct ocfs2_dinode *di;
1327
1328         trace_ocfs2_write_remove_suid(
1329                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
1330                         inode->i_mode);
1331
1332         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1333         if (IS_ERR(handle)) {
1334                 ret = PTR_ERR(handle);
1335                 mlog_errno(ret);
1336                 goto out;
1337         }
1338
1339         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
1340                                       OCFS2_JOURNAL_ACCESS_WRITE);
1341         if (ret < 0) {
1342                 mlog_errno(ret);
1343                 goto out_trans;
1344         }
1345
1346         inode->i_mode &= ~S_ISUID;
1347         if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
1348                 inode->i_mode &= ~S_ISGID;
1349
1350         di = (struct ocfs2_dinode *) bh->b_data;
1351         di->i_mode = cpu_to_le16(inode->i_mode);
1352         ocfs2_update_inode_fsync_trans(handle, inode, 0);
1353
1354         ocfs2_journal_dirty(handle, bh);
1355
1356 out_trans:
1357         ocfs2_commit_trans(osb, handle);
1358 out:
1359         return ret;
1360 }
1361
1362 /*
1363  * Will look for holes and unwritten extents in the range starting at
1364  * pos for count bytes (inclusive).
1365  */
1366 static int ocfs2_check_range_for_holes(struct inode *inode, loff_t pos,
1367                                        size_t count)
1368 {
1369         int ret = 0;
1370         unsigned int extent_flags;
1371         u32 cpos, clusters, extent_len, phys_cpos;
1372         struct super_block *sb = inode->i_sb;
1373
1374         cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
1375         clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
1376
1377         while (clusters) {
1378                 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
1379                                          &extent_flags);
1380                 if (ret < 0) {
1381                         mlog_errno(ret);
1382                         goto out;
1383                 }
1384
1385                 if (phys_cpos == 0 || (extent_flags & OCFS2_EXT_UNWRITTEN)) {
1386                         ret = 1;
1387                         break;
1388                 }
1389
1390                 if (extent_len > clusters)
1391                         extent_len = clusters;
1392
1393                 clusters -= extent_len;
1394                 cpos += extent_len;
1395         }
1396 out:
1397         return ret;
1398 }
1399
1400 static int ocfs2_write_remove_suid(struct inode *inode)
1401 {
1402         int ret;
1403         struct buffer_head *bh = NULL;
1404
1405         ret = ocfs2_read_inode_block(inode, &bh);
1406         if (ret < 0) {
1407                 mlog_errno(ret);
1408                 goto out;
1409         }
1410
1411         ret =  __ocfs2_write_remove_suid(inode, bh);
1412 out:
1413         brelse(bh);
1414         return ret;
1415 }
1416
1417 /*
1418  * Allocate enough extents to cover the region starting at byte offset
1419  * start for len bytes. Existing extents are skipped, any extents
1420  * added are marked as "unwritten".
1421  */
1422 static int ocfs2_allocate_unwritten_extents(struct inode *inode,
1423                                             u64 start, u64 len)
1424 {
1425         int ret;
1426         u32 cpos, phys_cpos, clusters, alloc_size;
1427         u64 end = start + len;
1428         struct buffer_head *di_bh = NULL;
1429
1430         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1431                 ret = ocfs2_read_inode_block(inode, &di_bh);
1432                 if (ret) {
1433                         mlog_errno(ret);
1434                         goto out;
1435                 }
1436
1437                 /*
1438                  * Nothing to do if the requested reservation range
1439                  * fits within the inode.
1440                  */
1441                 if (ocfs2_size_fits_inline_data(di_bh, end))
1442                         goto out;
1443
1444                 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1445                 if (ret) {
1446                         mlog_errno(ret);
1447                         goto out;
1448                 }
1449         }
1450
1451         /*
1452          * We consider both start and len to be inclusive.
1453          */
1454         cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
1455         clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len);
1456         clusters -= cpos;
1457
1458         while (clusters) {
1459                 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1460                                          &alloc_size, NULL);
1461                 if (ret) {
1462                         mlog_errno(ret);
1463                         goto out;
1464                 }
1465
1466                 /*
1467                  * Hole or existing extent len can be arbitrary, so
1468                  * cap it to our own allocation request.
1469                  */
1470                 if (alloc_size > clusters)
1471                         alloc_size = clusters;
1472
1473                 if (phys_cpos) {
1474                         /*
1475                          * We already have an allocation at this
1476                          * region so we can safely skip it.
1477                          */
1478                         goto next;
1479                 }
1480
1481                 ret = __ocfs2_extend_allocation(inode, cpos, alloc_size, 1);
1482                 if (ret) {
1483                         if (ret != -ENOSPC)
1484                                 mlog_errno(ret);
1485                         goto out;
1486                 }
1487
1488 next:
1489                 cpos += alloc_size;
1490                 clusters -= alloc_size;
1491         }
1492
1493         ret = 0;
1494 out:
1495
1496         brelse(di_bh);
1497         return ret;
1498 }
1499
1500 /*
1501  * Truncate a byte range, avoiding pages within partial clusters. This
1502  * preserves those pages for the zeroing code to write to.
1503  */
1504 static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start,
1505                                          u64 byte_len)
1506 {
1507         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1508         loff_t start, end;
1509         struct address_space *mapping = inode->i_mapping;
1510
1511         start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start);
1512         end = byte_start + byte_len;
1513         end = end & ~(osb->s_clustersize - 1);
1514
1515         if (start < end) {
1516                 unmap_mapping_range(mapping, start, end - start, 0);
1517                 truncate_inode_pages_range(mapping, start, end - 1);
1518         }
1519 }
1520
1521 static int ocfs2_zero_partial_clusters(struct inode *inode,
1522                                        u64 start, u64 len)
1523 {
1524         int ret = 0;
1525         u64 tmpend, end = start + len;
1526         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1527         unsigned int csize = osb->s_clustersize;
1528         handle_t *handle;
1529
1530         /*
1531          * The "start" and "end" values are NOT necessarily part of
1532          * the range whose allocation is being deleted. Rather, this
1533          * is what the user passed in with the request. We must zero
1534          * partial clusters here. There's no need to worry about
1535          * physical allocation - the zeroing code knows to skip holes.
1536          */
1537         trace_ocfs2_zero_partial_clusters(
1538                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1539                 (unsigned long long)start, (unsigned long long)end);
1540
1541         /*
1542          * If both edges are on a cluster boundary then there's no
1543          * zeroing required as the region is part of the allocation to
1544          * be truncated.
1545          */
1546         if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0)
1547                 goto out;
1548
1549         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1550         if (IS_ERR(handle)) {
1551                 ret = PTR_ERR(handle);
1552                 mlog_errno(ret);
1553                 goto out;
1554         }
1555
1556         /*
1557          * We want to get the byte offset of the end of the 1st cluster.
1558          */
1559         tmpend = (u64)osb->s_clustersize + (start & ~(osb->s_clustersize - 1));
1560         if (tmpend > end)
1561                 tmpend = end;
1562
1563         trace_ocfs2_zero_partial_clusters_range1((unsigned long long)start,
1564                                                  (unsigned long long)tmpend);
1565
1566         ret = ocfs2_zero_range_for_truncate(inode, handle, start, tmpend);
1567         if (ret)
1568                 mlog_errno(ret);
1569
1570         if (tmpend < end) {
1571                 /*
1572                  * This may make start and end equal, but the zeroing
1573                  * code will skip any work in that case so there's no
1574                  * need to catch it up here.
1575                  */
1576                 start = end & ~(osb->s_clustersize - 1);
1577
1578                 trace_ocfs2_zero_partial_clusters_range2(
1579                         (unsigned long long)start, (unsigned long long)end);
1580
1581                 ret = ocfs2_zero_range_for_truncate(inode, handle, start, end);
1582                 if (ret)
1583                         mlog_errno(ret);
1584         }
1585         ocfs2_update_inode_fsync_trans(handle, inode, 1);
1586
1587         ocfs2_commit_trans(osb, handle);
1588 out:
1589         return ret;
1590 }
1591
1592 static int ocfs2_find_rec(struct ocfs2_extent_list *el, u32 pos)
1593 {
1594         int i;
1595         struct ocfs2_extent_rec *rec = NULL;
1596
1597         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
1598
1599                 rec = &el->l_recs[i];
1600
1601                 if (le32_to_cpu(rec->e_cpos) < pos)
1602                         break;
1603         }
1604
1605         return i;
1606 }
1607
1608 /*
1609  * Helper to calculate the punching pos and length in one run, we handle the
1610  * following three cases in order:
1611  *
1612  * - remove the entire record
1613  * - remove a partial record
1614  * - no record needs to be removed (hole-punching completed)
1615 */
1616 static void ocfs2_calc_trunc_pos(struct inode *inode,
1617                                  struct ocfs2_extent_list *el,
1618                                  struct ocfs2_extent_rec *rec,
1619                                  u32 trunc_start, u32 *trunc_cpos,
1620                                  u32 *trunc_len, u32 *trunc_end,
1621                                  u64 *blkno, int *done)
1622 {
1623         int ret = 0;
1624         u32 coff, range;
1625
1626         range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
1627
1628         if (le32_to_cpu(rec->e_cpos) >= trunc_start) {
1629                 /*
1630                  * remove an entire extent record.
1631                  */
1632                 *trunc_cpos = le32_to_cpu(rec->e_cpos);
1633                 /*
1634                  * Skip holes if any.
1635                  */
1636                 if (range < *trunc_end)
1637                         *trunc_end = range;
1638                 *trunc_len = *trunc_end - le32_to_cpu(rec->e_cpos);
1639                 *blkno = le64_to_cpu(rec->e_blkno);
1640                 *trunc_end = le32_to_cpu(rec->e_cpos);
1641         } else if (range > trunc_start) {
1642                 /*
1643                  * remove a partial extent record, which means we're
1644                  * removing the last extent record.
1645                  */
1646                 *trunc_cpos = trunc_start;
1647                 /*
1648                  * skip hole if any.
1649                  */
1650                 if (range < *trunc_end)
1651                         *trunc_end = range;
1652                 *trunc_len = *trunc_end - trunc_start;
1653                 coff = trunc_start - le32_to_cpu(rec->e_cpos);
1654                 *blkno = le64_to_cpu(rec->e_blkno) +
1655                                 ocfs2_clusters_to_blocks(inode->i_sb, coff);
1656                 *trunc_end = trunc_start;
1657         } else {
1658                 /*
1659                  * It may have two following possibilities:
1660                  *
1661                  * - last record has been removed
1662                  * - trunc_start was within a hole
1663                  *
1664                  * both two cases mean the completion of hole punching.
1665                  */
1666                 ret = 1;
1667         }
1668
1669         *done = ret;
1670 }
1671
1672 static int ocfs2_remove_inode_range(struct inode *inode,
1673                                     struct buffer_head *di_bh, u64 byte_start,
1674                                     u64 byte_len)
1675 {
1676         int ret = 0, flags = 0, done = 0, i;
1677         u32 trunc_start, trunc_len, trunc_end, trunc_cpos, phys_cpos;
1678         u32 cluster_in_el;
1679         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1680         struct ocfs2_cached_dealloc_ctxt dealloc;
1681         struct address_space *mapping = inode->i_mapping;
1682         struct ocfs2_extent_tree et;
1683         struct ocfs2_path *path = NULL;
1684         struct ocfs2_extent_list *el = NULL;
1685         struct ocfs2_extent_rec *rec = NULL;
1686         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1687         u64 blkno, refcount_loc = le64_to_cpu(di->i_refcount_loc);
1688
1689         ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
1690         ocfs2_init_dealloc_ctxt(&dealloc);
1691
1692         trace_ocfs2_remove_inode_range(
1693                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
1694                         (unsigned long long)byte_start,
1695                         (unsigned long long)byte_len);
1696
1697         if (byte_len == 0)
1698                 return 0;
1699
1700         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1701                 ret = ocfs2_truncate_inline(inode, di_bh, byte_start,
1702                                             byte_start + byte_len, 0);
1703                 if (ret) {
1704                         mlog_errno(ret);
1705                         goto out;
1706                 }
1707                 /*
1708                  * There's no need to get fancy with the page cache
1709                  * truncate of an inline-data inode. We're talking
1710                  * about less than a page here, which will be cached
1711                  * in the dinode buffer anyway.
1712                  */
1713                 unmap_mapping_range(mapping, 0, 0, 0);
1714                 truncate_inode_pages(mapping, 0);
1715                 goto out;
1716         }
1717
1718         /*
1719          * For reflinks, we may need to CoW 2 clusters which might be
1720          * partially zero'd later, if hole's start and end offset were
1721          * within one cluster(means is not exactly aligned to clustersize).
1722          */
1723
1724         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) {
1725
1726                 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start);
1727                 if (ret) {
1728                         mlog_errno(ret);
1729                         goto out;
1730                 }
1731
1732                 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start + byte_len);
1733                 if (ret) {
1734                         mlog_errno(ret);
1735                         goto out;
1736                 }
1737         }
1738
1739         trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start);
1740         trunc_end = (byte_start + byte_len) >> osb->s_clustersize_bits;
1741         cluster_in_el = trunc_end;
1742
1743         ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len);
1744         if (ret) {
1745                 mlog_errno(ret);
1746                 goto out;
1747         }
1748
1749         path = ocfs2_new_path_from_et(&et);
1750         if (!path) {
1751                 ret = -ENOMEM;
1752                 mlog_errno(ret);
1753                 goto out;
1754         }
1755
1756         while (trunc_end > trunc_start) {
1757
1758                 ret = ocfs2_find_path(INODE_CACHE(inode), path,
1759                                       cluster_in_el);
1760                 if (ret) {
1761                         mlog_errno(ret);
1762                         goto out;
1763                 }
1764
1765                 el = path_leaf_el(path);
1766
1767                 i = ocfs2_find_rec(el, trunc_end);
1768                 /*
1769                  * Need to go to previous extent block.
1770                  */
1771                 if (i < 0) {
1772                         if (path->p_tree_depth == 0)
1773                                 break;
1774
1775                         ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb,
1776                                                             path,
1777                                                             &cluster_in_el);
1778                         if (ret) {
1779                                 mlog_errno(ret);
1780                                 goto out;
1781                         }
1782
1783                         /*
1784                          * We've reached the leftmost extent block,
1785                          * it's safe to leave.
1786                          */
1787                         if (cluster_in_el == 0)
1788                                 break;
1789
1790                         /*
1791                          * The 'pos' searched for previous extent block is
1792                          * always one cluster less than actual trunc_end.
1793                          */
1794                         trunc_end = cluster_in_el + 1;
1795
1796                         ocfs2_reinit_path(path, 1);
1797
1798                         continue;
1799
1800                 } else
1801                         rec = &el->l_recs[i];
1802
1803                 ocfs2_calc_trunc_pos(inode, el, rec, trunc_start, &trunc_cpos,
1804                                      &trunc_len, &trunc_end, &blkno, &done);
1805                 if (done)
1806                         break;
1807
1808                 flags = rec->e_flags;
1809                 phys_cpos = ocfs2_blocks_to_clusters(inode->i_sb, blkno);
1810
1811                 ret = ocfs2_remove_btree_range(inode, &et, trunc_cpos,
1812                                                phys_cpos, trunc_len, flags,
1813                                                &dealloc, refcount_loc, false);
1814                 if (ret < 0) {
1815                         mlog_errno(ret);
1816                         goto out;
1817                 }
1818
1819                 cluster_in_el = trunc_end;
1820
1821                 ocfs2_reinit_path(path, 1);
1822         }
1823
1824         ocfs2_truncate_cluster_pages(inode, byte_start, byte_len);
1825
1826 out:
1827         ocfs2_free_path(path);
1828         ocfs2_schedule_truncate_log_flush(osb, 1);
1829         ocfs2_run_deallocs(osb, &dealloc);
1830
1831         return ret;
1832 }
1833
1834 /*
1835  * Parts of this function taken from xfs_change_file_space()
1836  */
1837 static int __ocfs2_change_file_space(struct file *file, struct inode *inode,
1838                                      loff_t f_pos, unsigned int cmd,
1839                                      struct ocfs2_space_resv *sr,
1840                                      int change_size)
1841 {
1842         int ret;
1843         s64 llen;
1844         loff_t size;
1845         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1846         struct buffer_head *di_bh = NULL;
1847         handle_t *handle;
1848         unsigned long long max_off = inode->i_sb->s_maxbytes;
1849
1850         if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
1851                 return -EROFS;
1852
1853         mutex_lock(&inode->i_mutex);
1854
1855         /*
1856          * This prevents concurrent writes on other nodes
1857          */
1858         ret = ocfs2_rw_lock(inode, 1);
1859         if (ret) {
1860                 mlog_errno(ret);
1861                 goto out;
1862         }
1863
1864         ret = ocfs2_inode_lock(inode, &di_bh, 1);
1865         if (ret) {
1866                 mlog_errno(ret);
1867                 goto out_rw_unlock;
1868         }
1869
1870         if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
1871                 ret = -EPERM;
1872                 goto out_inode_unlock;
1873         }
1874
1875         switch (sr->l_whence) {
1876         case 0: /*SEEK_SET*/
1877                 break;
1878         case 1: /*SEEK_CUR*/
1879                 sr->l_start += f_pos;
1880                 break;
1881         case 2: /*SEEK_END*/
1882                 sr->l_start += i_size_read(inode);
1883                 break;
1884         default:
1885                 ret = -EINVAL;
1886                 goto out_inode_unlock;
1887         }
1888         sr->l_whence = 0;
1889
1890         llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len;
1891
1892         if (sr->l_start < 0
1893             || sr->l_start > max_off
1894             || (sr->l_start + llen) < 0
1895             || (sr->l_start + llen) > max_off) {
1896                 ret = -EINVAL;
1897                 goto out_inode_unlock;
1898         }
1899         size = sr->l_start + sr->l_len;
1900
1901         if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64 ||
1902             cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) {
1903                 if (sr->l_len <= 0) {
1904                         ret = -EINVAL;
1905                         goto out_inode_unlock;
1906                 }
1907         }
1908
1909         if (file && should_remove_suid(file->f_path.dentry)) {
1910                 ret = __ocfs2_write_remove_suid(inode, di_bh);
1911                 if (ret) {
1912                         mlog_errno(ret);
1913                         goto out_inode_unlock;
1914                 }
1915         }
1916
1917         down_write(&OCFS2_I(inode)->ip_alloc_sem);
1918         switch (cmd) {
1919         case OCFS2_IOC_RESVSP:
1920         case OCFS2_IOC_RESVSP64:
1921                 /*
1922                  * This takes unsigned offsets, but the signed ones we
1923                  * pass have been checked against overflow above.
1924                  */
1925                 ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start,
1926                                                        sr->l_len);
1927                 break;
1928         case OCFS2_IOC_UNRESVSP:
1929         case OCFS2_IOC_UNRESVSP64:
1930                 ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start,
1931                                                sr->l_len);
1932                 break;
1933         default:
1934                 ret = -EINVAL;
1935         }
1936         up_write(&OCFS2_I(inode)->ip_alloc_sem);
1937         if (ret) {
1938                 mlog_errno(ret);
1939                 goto out_inode_unlock;
1940         }
1941
1942         /*
1943          * We update c/mtime for these changes
1944          */
1945         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1946         if (IS_ERR(handle)) {
1947                 ret = PTR_ERR(handle);
1948                 mlog_errno(ret);
1949                 goto out_inode_unlock;
1950         }
1951
1952         if (change_size && i_size_read(inode) < size)
1953                 i_size_write(inode, size);
1954
1955         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1956         ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
1957         if (ret < 0)
1958                 mlog_errno(ret);
1959
1960         if (file && (file->f_flags & O_SYNC))
1961                 handle->h_sync = 1;
1962
1963         ocfs2_commit_trans(osb, handle);
1964
1965 out_inode_unlock:
1966         brelse(di_bh);
1967         ocfs2_inode_unlock(inode, 1);
1968 out_rw_unlock:
1969         ocfs2_rw_unlock(inode, 1);
1970
1971 out:
1972         mutex_unlock(&inode->i_mutex);
1973         return ret;
1974 }
1975
1976 int ocfs2_change_file_space(struct file *file, unsigned int cmd,
1977                             struct ocfs2_space_resv *sr)
1978 {
1979         struct inode *inode = file_inode(file);
1980         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1981         int ret;
1982
1983         if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) &&
1984             !ocfs2_writes_unwritten_extents(osb))
1985                 return -ENOTTY;
1986         else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) &&
1987                  !ocfs2_sparse_alloc(osb))
1988                 return -ENOTTY;
1989
1990         if (!S_ISREG(inode->i_mode))
1991                 return -EINVAL;
1992
1993         if (!(file->f_mode & FMODE_WRITE))
1994                 return -EBADF;
1995
1996         ret = mnt_want_write_file(file);
1997         if (ret)
1998                 return ret;
1999         ret = __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0);
2000         mnt_drop_write_file(file);
2001         return ret;
2002 }
2003
2004 static long ocfs2_fallocate(struct file *file, int mode, loff_t offset,
2005                             loff_t len)
2006 {
2007         struct inode *inode = file_inode(file);
2008         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2009         struct ocfs2_space_resv sr;
2010         int change_size = 1;
2011         int cmd = OCFS2_IOC_RESVSP64;
2012
2013         if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2014                 return -EOPNOTSUPP;
2015         if (!ocfs2_writes_unwritten_extents(osb))
2016                 return -EOPNOTSUPP;
2017
2018         if (mode & FALLOC_FL_KEEP_SIZE)
2019                 change_size = 0;
2020
2021         if (mode & FALLOC_FL_PUNCH_HOLE)
2022                 cmd = OCFS2_IOC_UNRESVSP64;
2023
2024         sr.l_whence = 0;
2025         sr.l_start = (s64)offset;
2026         sr.l_len = (s64)len;
2027
2028         return __ocfs2_change_file_space(NULL, inode, offset, cmd, &sr,
2029                                          change_size);
2030 }
2031
2032 int ocfs2_check_range_for_refcount(struct inode *inode, loff_t pos,
2033                                    size_t count)
2034 {
2035         int ret = 0;
2036         unsigned int extent_flags;
2037         u32 cpos, clusters, extent_len, phys_cpos;
2038         struct super_block *sb = inode->i_sb;
2039
2040         if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)) ||
2041             !(OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) ||
2042             OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
2043                 return 0;
2044
2045         cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
2046         clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
2047
2048         while (clusters) {
2049                 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
2050                                          &extent_flags);
2051                 if (ret < 0) {
2052                         mlog_errno(ret);
2053                         goto out;
2054                 }
2055
2056                 if (phys_cpos && (extent_flags & OCFS2_EXT_REFCOUNTED)) {
2057                         ret = 1;
2058                         break;
2059                 }
2060
2061                 if (extent_len > clusters)
2062                         extent_len = clusters;
2063
2064                 clusters -= extent_len;
2065                 cpos += extent_len;
2066         }
2067 out:
2068         return ret;
2069 }
2070
2071 static int ocfs2_is_io_unaligned(struct inode *inode, size_t count, loff_t pos)
2072 {
2073         int blockmask = inode->i_sb->s_blocksize - 1;
2074         loff_t final_size = pos + count;
2075
2076         if ((pos & blockmask) || (final_size & blockmask))
2077                 return 1;
2078         return 0;
2079 }
2080
2081 static int ocfs2_prepare_inode_for_refcount(struct inode *inode,
2082                                             struct file *file,
2083                                             loff_t pos, size_t count,
2084                                             int *meta_level)
2085 {
2086         int ret;
2087         struct buffer_head *di_bh = NULL;
2088         u32 cpos = pos >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
2089         u32 clusters =
2090                 ocfs2_clusters_for_bytes(inode->i_sb, pos + count) - cpos;
2091
2092         ret = ocfs2_inode_lock(inode, &di_bh, 1);
2093         if (ret) {
2094                 mlog_errno(ret);
2095                 goto out;
2096         }
2097
2098         *meta_level = 1;
2099
2100         ret = ocfs2_refcount_cow(inode, di_bh, cpos, clusters, UINT_MAX);
2101         if (ret)
2102                 mlog_errno(ret);
2103 out:
2104         brelse(di_bh);
2105         return ret;
2106 }
2107
2108 static int ocfs2_prepare_inode_for_write(struct file *file,
2109                                          loff_t *ppos,
2110                                          size_t count,
2111                                          int appending,
2112                                          int *direct_io,
2113                                          int *has_refcount)
2114 {
2115         int ret = 0, meta_level = 0;
2116         struct dentry *dentry = file->f_path.dentry;
2117         struct inode *inode = dentry->d_inode;
2118         loff_t saved_pos = 0, end;
2119         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2120         int full_coherency = !(osb->s_mount_opt &
2121                 OCFS2_MOUNT_COHERENCY_BUFFERED);
2122
2123         /*
2124          * We start with a read level meta lock and only jump to an ex
2125          * if we need to make modifications here.
2126          */
2127         for(;;) {
2128                 ret = ocfs2_inode_lock(inode, NULL, meta_level);
2129                 if (ret < 0) {
2130                         meta_level = -1;
2131                         mlog_errno(ret);
2132                         goto out;
2133                 }
2134
2135                 /* Clear suid / sgid if necessary. We do this here
2136                  * instead of later in the write path because
2137                  * remove_suid() calls ->setattr without any hint that
2138                  * we may have already done our cluster locking. Since
2139                  * ocfs2_setattr() *must* take cluster locks to
2140                  * proceed, this will lead us to recursively lock the
2141                  * inode. There's also the dinode i_size state which
2142                  * can be lost via setattr during extending writes (we
2143                  * set inode->i_size at the end of a write. */
2144                 if (should_remove_suid(dentry)) {
2145                         if (meta_level == 0) {
2146                                 ocfs2_inode_unlock(inode, meta_level);
2147                                 meta_level = 1;
2148                                 continue;
2149                         }
2150
2151                         ret = ocfs2_write_remove_suid(inode);
2152                         if (ret < 0) {
2153                                 mlog_errno(ret);
2154                                 goto out_unlock;
2155                         }
2156                 }
2157
2158                 /* work on a copy of ppos until we're sure that we won't have
2159                  * to recalculate it due to relocking. */
2160                 if (appending)
2161                         saved_pos = i_size_read(inode);
2162                 else
2163                         saved_pos = *ppos;
2164
2165                 end = saved_pos + count;
2166
2167                 ret = ocfs2_check_range_for_refcount(inode, saved_pos, count);
2168                 if (ret == 1) {
2169                         ocfs2_inode_unlock(inode, meta_level);
2170                         meta_level = -1;
2171
2172                         ret = ocfs2_prepare_inode_for_refcount(inode,
2173                                                                file,
2174                                                                saved_pos,
2175                                                                count,
2176                                                                &meta_level);
2177                         if (has_refcount)
2178                                 *has_refcount = 1;
2179                         if (direct_io)
2180                                 *direct_io = 0;
2181                 }
2182
2183                 if (ret < 0) {
2184                         mlog_errno(ret);
2185                         goto out_unlock;
2186                 }
2187
2188                 /*
2189                  * Skip the O_DIRECT checks if we don't need
2190                  * them.
2191                  */
2192                 if (!direct_io || !(*direct_io))
2193                         break;
2194
2195                 /*
2196                  * There's no sane way to do direct writes to an inode
2197                  * with inline data.
2198                  */
2199                 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2200                         *direct_io = 0;
2201                         break;
2202                 }
2203
2204                 /*
2205                  * Allowing concurrent direct writes means
2206                  * i_size changes wouldn't be synchronized, so
2207                  * one node could wind up truncating another
2208                  * nodes writes.
2209                  */
2210                 if (end > i_size_read(inode) && !full_coherency) {
2211                         *direct_io = 0;
2212                         break;
2213                 }
2214
2215                 /*
2216                  * Fallback to old way if the feature bit is not set.
2217                  */
2218                 if (end > i_size_read(inode) &&
2219                                 !ocfs2_supports_append_dio(osb)) {
2220                         *direct_io = 0;
2221                         break;
2222                 }
2223
2224                 /*
2225                  * We don't fill holes during direct io, so
2226                  * check for them here. If any are found, the
2227                  * caller will have to retake some cluster
2228                  * locks and initiate the io as buffered.
2229                  */
2230                 ret = ocfs2_check_range_for_holes(inode, saved_pos, count);
2231                 if (ret == 1) {
2232                         /*
2233                          * Fallback to old way if the feature bit is not set.
2234                          * Otherwise try dio first and then complete the rest
2235                          * request through buffer io.
2236                          */
2237                         if (!ocfs2_supports_append_dio(osb))
2238                                 *direct_io = 0;
2239                         ret = 0;
2240                 } else if (ret < 0)
2241                         mlog_errno(ret);
2242                 break;
2243         }
2244
2245         if (appending)
2246                 *ppos = saved_pos;
2247
2248 out_unlock:
2249         trace_ocfs2_prepare_inode_for_write(OCFS2_I(inode)->ip_blkno,
2250                                             saved_pos, appending, count,
2251                                             direct_io, has_refcount);
2252
2253         if (meta_level >= 0)
2254                 ocfs2_inode_unlock(inode, meta_level);
2255
2256 out:
2257         return ret;
2258 }
2259
2260 static ssize_t ocfs2_file_write_iter(struct kiocb *iocb,
2261                                     struct iov_iter *from)
2262 {
2263         int ret, direct_io, appending, rw_level, have_alloc_sem  = 0;
2264         int can_do_direct, has_refcount = 0;
2265         ssize_t written = 0;
2266         size_t count = iov_iter_count(from);
2267         loff_t old_size, *ppos = &iocb->ki_pos;
2268         u32 old_clusters;
2269         struct file *file = iocb->ki_filp;
2270         struct inode *inode = file_inode(file);
2271         struct address_space *mapping = file->f_mapping;
2272         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2273         int full_coherency = !(osb->s_mount_opt &
2274                                OCFS2_MOUNT_COHERENCY_BUFFERED);
2275         int unaligned_dio = 0;
2276
2277         trace_ocfs2_file_aio_write(inode, file, file->f_path.dentry,
2278                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
2279                 file->f_path.dentry->d_name.len,
2280                 file->f_path.dentry->d_name.name,
2281                 (unsigned int)from->nr_segs);   /* GRRRRR */
2282
2283         if (iocb->ki_nbytes == 0)
2284                 return 0;
2285
2286         appending = file->f_flags & O_APPEND ? 1 : 0;
2287         direct_io = file->f_flags & O_DIRECT ? 1 : 0;
2288
2289         mutex_lock(&inode->i_mutex);
2290
2291         ocfs2_iocb_clear_sem_locked(iocb);
2292
2293 relock:
2294         /* to match setattr's i_mutex -> rw_lock ordering */
2295         if (direct_io) {
2296                 have_alloc_sem = 1;
2297                 /* communicate with ocfs2_dio_end_io */
2298                 ocfs2_iocb_set_sem_locked(iocb);
2299         }
2300
2301         /*
2302          * Concurrent O_DIRECT writes are allowed with
2303          * mount_option "coherency=buffered".
2304          */
2305         rw_level = (!direct_io || full_coherency);
2306
2307         ret = ocfs2_rw_lock(inode, rw_level);
2308         if (ret < 0) {
2309                 mlog_errno(ret);
2310                 goto out_sems;
2311         }
2312
2313         /*
2314          * O_DIRECT writes with "coherency=full" need to take EX cluster
2315          * inode_lock to guarantee coherency.
2316          */
2317         if (direct_io && full_coherency) {
2318                 /*
2319                  * We need to take and drop the inode lock to force
2320                  * other nodes to drop their caches.  Buffered I/O
2321                  * already does this in write_begin().
2322                  */
2323                 ret = ocfs2_inode_lock(inode, NULL, 1);
2324                 if (ret < 0) {
2325                         mlog_errno(ret);
2326                         goto out;
2327                 }
2328
2329                 ocfs2_inode_unlock(inode, 1);
2330         }
2331
2332         can_do_direct = direct_io;
2333         ret = ocfs2_prepare_inode_for_write(file, ppos,
2334                                             iocb->ki_nbytes, appending,
2335                                             &can_do_direct, &has_refcount);
2336         if (ret < 0) {
2337                 mlog_errno(ret);
2338                 goto out;
2339         }
2340
2341         if (direct_io && !is_sync_kiocb(iocb))
2342                 unaligned_dio = ocfs2_is_io_unaligned(inode, iocb->ki_nbytes,
2343                                                       *ppos);
2344
2345         /*
2346          * We can't complete the direct I/O as requested, fall back to
2347          * buffered I/O.
2348          */
2349         if (direct_io && !can_do_direct) {
2350                 ocfs2_rw_unlock(inode, rw_level);
2351
2352                 have_alloc_sem = 0;
2353                 rw_level = -1;
2354
2355                 direct_io = 0;
2356                 goto relock;
2357         }
2358
2359         if (unaligned_dio) {
2360                 /*
2361                  * Wait on previous unaligned aio to complete before
2362                  * proceeding.
2363                  */
2364                 mutex_lock(&OCFS2_I(inode)->ip_unaligned_aio);
2365                 /* Mark the iocb as needing an unlock in ocfs2_dio_end_io */
2366                 ocfs2_iocb_set_unaligned_aio(iocb);
2367         }
2368
2369         /*
2370          * To later detect whether a journal commit for sync writes is
2371          * necessary, we sample i_size, and cluster count here.
2372          */
2373         old_size = i_size_read(inode);
2374         old_clusters = OCFS2_I(inode)->ip_clusters;
2375
2376         /* communicate with ocfs2_dio_end_io */
2377         ocfs2_iocb_set_rw_locked(iocb, rw_level);
2378
2379         ret = generic_write_checks(file, ppos, &count,
2380                                    S_ISBLK(inode->i_mode));
2381         if (ret)
2382                 goto out_dio;
2383
2384         iov_iter_truncate(from, count);
2385         if (direct_io) {
2386                 loff_t endbyte;
2387                 ssize_t written_buffered;
2388                 written = generic_file_direct_write(iocb, from, *ppos);
2389                 if (written < 0 || written == count) {
2390                         ret = written;
2391                         goto out_dio;
2392                 }
2393
2394                 /*
2395                  * for completing the rest of the request.
2396                  */
2397                 count -= written;
2398                 written_buffered = generic_perform_write(file, from, *ppos);
2399                 /*
2400                  * If generic_file_buffered_write() returned a synchronous error
2401                  * then we want to return the number of bytes which were
2402                  * direct-written, or the error code if that was zero. Note
2403                  * that this differs from normal direct-io semantics, which
2404                  * will return -EFOO even if some bytes were written.
2405                  */
2406                 if (written_buffered < 0) {
2407                         ret = written_buffered;
2408                         goto out_dio;
2409                 }
2410
2411                 /* We need to ensure that the page cache pages are written to
2412                  * disk and invalidated to preserve the expected O_DIRECT
2413                  * semantics.
2414                  */
2415                 endbyte = *ppos + written_buffered - 1;
2416                 ret = filemap_write_and_wait_range(file->f_mapping, *ppos,
2417                                 endbyte);
2418                 if (ret == 0) {
2419                         iocb->ki_pos = *ppos + written_buffered;
2420                         written += written_buffered;
2421                         invalidate_mapping_pages(mapping,
2422                                         *ppos >> PAGE_CACHE_SHIFT,
2423                                         endbyte >> PAGE_CACHE_SHIFT);
2424                 } else {
2425                         /*
2426                          * We don't know how much we wrote, so just return
2427                          * the number of bytes which were direct-written
2428                          */
2429                 }
2430         } else {
2431                 current->backing_dev_info = inode_to_bdi(inode);
2432                 written = generic_perform_write(file, from, *ppos);
2433                 if (likely(written >= 0))
2434                         iocb->ki_pos = *ppos + written;
2435                 current->backing_dev_info = NULL;
2436         }
2437
2438 out_dio:
2439         /* buffered aio wouldn't have proper lock coverage today */
2440         BUG_ON(ret == -EIOCBQUEUED && !(file->f_flags & O_DIRECT));
2441
2442         if (((file->f_flags & O_DSYNC) && !direct_io) || IS_SYNC(inode) ||
2443             ((file->f_flags & O_DIRECT) && !direct_io)) {
2444                 ret = filemap_fdatawrite_range(file->f_mapping, *ppos,
2445                                                *ppos + count - 1);
2446                 if (ret < 0)
2447                         written = ret;
2448
2449                 if (!ret) {
2450                         ret = jbd2_journal_force_commit(osb->journal->j_journal);
2451                         if (ret < 0)
2452                                 written = ret;
2453                 }
2454
2455                 if (!ret)
2456                         ret = filemap_fdatawait_range(file->f_mapping, *ppos,
2457                                                       *ppos + count - 1);
2458         }
2459
2460         /*
2461          * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
2462          * function pointer which is called when o_direct io completes so that
2463          * it can unlock our rw lock.
2464          * Unfortunately there are error cases which call end_io and others
2465          * that don't.  so we don't have to unlock the rw_lock if either an
2466          * async dio is going to do it in the future or an end_io after an
2467          * error has already done it.
2468          */
2469         if ((ret == -EIOCBQUEUED) || (!ocfs2_iocb_is_rw_locked(iocb))) {
2470                 rw_level = -1;
2471                 have_alloc_sem = 0;
2472                 unaligned_dio = 0;
2473         }
2474
2475         if (unaligned_dio) {
2476                 ocfs2_iocb_clear_unaligned_aio(iocb);
2477                 mutex_unlock(&OCFS2_I(inode)->ip_unaligned_aio);
2478         }
2479
2480 out:
2481         if (rw_level != -1)
2482                 ocfs2_rw_unlock(inode, rw_level);
2483
2484 out_sems:
2485         if (have_alloc_sem)
2486                 ocfs2_iocb_clear_sem_locked(iocb);
2487
2488         mutex_unlock(&inode->i_mutex);
2489
2490         if (written)
2491                 ret = written;
2492         return ret;
2493 }
2494
2495 static ssize_t ocfs2_file_splice_read(struct file *in,
2496                                       loff_t *ppos,
2497                                       struct pipe_inode_info *pipe,
2498                                       size_t len,
2499                                       unsigned int flags)
2500 {
2501         int ret = 0, lock_level = 0;
2502         struct inode *inode = file_inode(in);
2503
2504         trace_ocfs2_file_splice_read(inode, in, in->f_path.dentry,
2505                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
2506                         in->f_path.dentry->d_name.len,
2507                         in->f_path.dentry->d_name.name, len);
2508
2509         /*
2510          * See the comment in ocfs2_file_read_iter()
2511          */
2512         ret = ocfs2_inode_lock_atime(inode, in->f_path.mnt, &lock_level);
2513         if (ret < 0) {
2514                 mlog_errno(ret);
2515                 goto bail;
2516         }
2517         ocfs2_inode_unlock(inode, lock_level);
2518
2519         ret = generic_file_splice_read(in, ppos, pipe, len, flags);
2520
2521 bail:
2522         return ret;
2523 }
2524
2525 static ssize_t ocfs2_file_read_iter(struct kiocb *iocb,
2526                                    struct iov_iter *to)
2527 {
2528         int ret = 0, rw_level = -1, have_alloc_sem = 0, lock_level = 0;
2529         struct file *filp = iocb->ki_filp;
2530         struct inode *inode = file_inode(filp);
2531
2532         trace_ocfs2_file_aio_read(inode, filp, filp->f_path.dentry,
2533                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
2534                         filp->f_path.dentry->d_name.len,
2535                         filp->f_path.dentry->d_name.name,
2536                         to->nr_segs);   /* GRRRRR */
2537
2538
2539         if (!inode) {
2540                 ret = -EINVAL;
2541                 mlog_errno(ret);
2542                 goto bail;
2543         }
2544
2545         ocfs2_iocb_clear_sem_locked(iocb);
2546
2547         /*
2548          * buffered reads protect themselves in ->readpage().  O_DIRECT reads
2549          * need locks to protect pending reads from racing with truncate.
2550          */
2551         if (filp->f_flags & O_DIRECT) {
2552                 have_alloc_sem = 1;
2553                 ocfs2_iocb_set_sem_locked(iocb);
2554
2555                 ret = ocfs2_rw_lock(inode, 0);
2556                 if (ret < 0) {
2557                         mlog_errno(ret);
2558                         goto bail;
2559                 }
2560                 rw_level = 0;
2561                 /* communicate with ocfs2_dio_end_io */
2562                 ocfs2_iocb_set_rw_locked(iocb, rw_level);
2563         }
2564
2565         /*
2566          * We're fine letting folks race truncates and extending
2567          * writes with read across the cluster, just like they can
2568          * locally. Hence no rw_lock during read.
2569          *
2570          * Take and drop the meta data lock to update inode fields
2571          * like i_size. This allows the checks down below
2572          * generic_file_aio_read() a chance of actually working.
2573          */
2574         ret = ocfs2_inode_lock_atime(inode, filp->f_path.mnt, &lock_level);
2575         if (ret < 0) {
2576                 mlog_errno(ret);
2577                 goto bail;
2578         }
2579         ocfs2_inode_unlock(inode, lock_level);
2580
2581         ret = generic_file_read_iter(iocb, to);
2582         trace_generic_file_aio_read_ret(ret);
2583
2584         /* buffered aio wouldn't have proper lock coverage today */
2585         BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT));
2586
2587         /* see ocfs2_file_write_iter */
2588         if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2589                 rw_level = -1;
2590                 have_alloc_sem = 0;
2591         }
2592
2593 bail:
2594         if (have_alloc_sem)
2595                 ocfs2_iocb_clear_sem_locked(iocb);
2596
2597         if (rw_level != -1)
2598                 ocfs2_rw_unlock(inode, rw_level);
2599
2600         return ret;
2601 }
2602
2603 /* Refer generic_file_llseek_unlocked() */
2604 static loff_t ocfs2_file_llseek(struct file *file, loff_t offset, int whence)
2605 {
2606         struct inode *inode = file->f_mapping->host;
2607         int ret = 0;
2608
2609         mutex_lock(&inode->i_mutex);
2610
2611         switch (whence) {
2612         case SEEK_SET:
2613                 break;
2614         case SEEK_END:
2615                 /* SEEK_END requires the OCFS2 inode lock for the file
2616                  * because it references the file's size.
2617                  */
2618                 ret = ocfs2_inode_lock(inode, NULL, 0);
2619                 if (ret < 0) {
2620                         mlog_errno(ret);
2621                         goto out;
2622                 }
2623                 offset += i_size_read(inode);
2624                 ocfs2_inode_unlock(inode, 0);
2625                 break;
2626         case SEEK_CUR:
2627                 if (offset == 0) {
2628                         offset = file->f_pos;
2629                         goto out;
2630                 }
2631                 offset += file->f_pos;
2632                 break;
2633         case SEEK_DATA:
2634         case SEEK_HOLE:
2635                 ret = ocfs2_seek_data_hole_offset(file, &offset, whence);
2636                 if (ret)
2637                         goto out;
2638                 break;
2639         default:
2640                 ret = -EINVAL;
2641                 goto out;
2642         }
2643
2644         offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
2645
2646 out:
2647         mutex_unlock(&inode->i_mutex);
2648         if (ret)
2649                 return ret;
2650         return offset;
2651 }
2652
2653 const struct inode_operations ocfs2_file_iops = {
2654         .setattr        = ocfs2_setattr,
2655         .getattr        = ocfs2_getattr,
2656         .permission     = ocfs2_permission,
2657         .setxattr       = generic_setxattr,
2658         .getxattr       = generic_getxattr,
2659         .listxattr      = ocfs2_listxattr,
2660         .removexattr    = generic_removexattr,
2661         .fiemap         = ocfs2_fiemap,
2662         .get_acl        = ocfs2_iop_get_acl,
2663         .set_acl        = ocfs2_iop_set_acl,
2664 };
2665
2666 const struct inode_operations ocfs2_special_file_iops = {
2667         .setattr        = ocfs2_setattr,
2668         .getattr        = ocfs2_getattr,
2669         .permission     = ocfs2_permission,
2670         .get_acl        = ocfs2_iop_get_acl,
2671         .set_acl        = ocfs2_iop_set_acl,
2672 };
2673
2674 /*
2675  * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with
2676  * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks!
2677  */
2678 const struct file_operations ocfs2_fops = {
2679         .llseek         = ocfs2_file_llseek,
2680         .read           = new_sync_read,
2681         .write          = new_sync_write,
2682         .mmap           = ocfs2_mmap,
2683         .fsync          = ocfs2_sync_file,
2684         .release        = ocfs2_file_release,
2685         .open           = ocfs2_file_open,
2686         .read_iter      = ocfs2_file_read_iter,
2687         .write_iter     = ocfs2_file_write_iter,
2688         .unlocked_ioctl = ocfs2_ioctl,
2689 #ifdef CONFIG_COMPAT
2690         .compat_ioctl   = ocfs2_compat_ioctl,
2691 #endif
2692         .lock           = ocfs2_lock,
2693         .flock          = ocfs2_flock,
2694         .splice_read    = ocfs2_file_splice_read,
2695         .splice_write   = iter_file_splice_write,
2696         .fallocate      = ocfs2_fallocate,
2697 };
2698
2699 const struct file_operations ocfs2_dops = {
2700         .llseek         = generic_file_llseek,
2701         .read           = generic_read_dir,
2702         .iterate        = ocfs2_readdir,
2703         .fsync          = ocfs2_sync_file,
2704         .release        = ocfs2_dir_release,
2705         .open           = ocfs2_dir_open,
2706         .unlocked_ioctl = ocfs2_ioctl,
2707 #ifdef CONFIG_COMPAT
2708         .compat_ioctl   = ocfs2_compat_ioctl,
2709 #endif
2710         .lock           = ocfs2_lock,
2711         .flock          = ocfs2_flock,
2712 };
2713
2714 /*
2715  * POSIX-lockless variants of our file_operations.
2716  *
2717  * These will be used if the underlying cluster stack does not support
2718  * posix file locking, if the user passes the "localflocks" mount
2719  * option, or if we have a local-only fs.
2720  *
2721  * ocfs2_flock is in here because all stacks handle UNIX file locks,
2722  * so we still want it in the case of no stack support for
2723  * plocks. Internally, it will do the right thing when asked to ignore
2724  * the cluster.
2725  */
2726 const struct file_operations ocfs2_fops_no_plocks = {
2727         .llseek         = ocfs2_file_llseek,
2728         .read           = new_sync_read,
2729         .write          = new_sync_write,
2730         .mmap           = ocfs2_mmap,
2731         .fsync          = ocfs2_sync_file,
2732         .release        = ocfs2_file_release,
2733         .open           = ocfs2_file_open,
2734         .read_iter      = ocfs2_file_read_iter,
2735         .write_iter     = ocfs2_file_write_iter,
2736         .unlocked_ioctl = ocfs2_ioctl,
2737 #ifdef CONFIG_COMPAT
2738         .compat_ioctl   = ocfs2_compat_ioctl,
2739 #endif
2740         .flock          = ocfs2_flock,
2741         .splice_read    = ocfs2_file_splice_read,
2742         .splice_write   = iter_file_splice_write,
2743         .fallocate      = ocfs2_fallocate,
2744 };
2745
2746 const struct file_operations ocfs2_dops_no_plocks = {
2747         .llseek         = generic_file_llseek,
2748         .read           = generic_read_dir,
2749         .iterate        = ocfs2_readdir,
2750         .fsync          = ocfs2_sync_file,
2751         .release        = ocfs2_dir_release,
2752         .open           = ocfs2_dir_open,
2753         .unlocked_ioctl = ocfs2_ioctl,
2754 #ifdef CONFIG_COMPAT
2755         .compat_ioctl   = ocfs2_compat_ioctl,
2756 #endif
2757         .flock          = ocfs2_flock,
2758 };