GNU Linux-libre 4.4.295-gnu1
[releases.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 #include <linux/backing-dev.h>
41
42 #include <cluster/masklog.h>
43
44 #include "ocfs2.h"
45
46 #include "alloc.h"
47 #include "aops.h"
48 #include "dir.h"
49 #include "dlmglue.h"
50 #include "extent_map.h"
51 #include "file.h"
52 #include "sysfile.h"
53 #include "inode.h"
54 #include "ioctl.h"
55 #include "journal.h"
56 #include "locks.h"
57 #include "mmap.h"
58 #include "suballoc.h"
59 #include "super.h"
60 #include "xattr.h"
61 #include "acl.h"
62 #include "quota.h"
63 #include "refcounttree.h"
64 #include "ocfs2_trace.h"
65
66 #include "buffer_head_io.h"
67
68 static int ocfs2_init_file_private(struct inode *inode, struct file *file)
69 {
70         struct ocfs2_file_private *fp;
71
72         fp = kzalloc(sizeof(struct ocfs2_file_private), GFP_KERNEL);
73         if (!fp)
74                 return -ENOMEM;
75
76         fp->fp_file = file;
77         mutex_init(&fp->fp_mutex);
78         ocfs2_file_lock_res_init(&fp->fp_flock, fp);
79         file->private_data = fp;
80
81         return 0;
82 }
83
84 static void ocfs2_free_file_private(struct inode *inode, struct file *file)
85 {
86         struct ocfs2_file_private *fp = file->private_data;
87         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
88
89         if (fp) {
90                 ocfs2_simple_drop_lockres(osb, &fp->fp_flock);
91                 ocfs2_lock_res_free(&fp->fp_flock);
92                 kfree(fp);
93                 file->private_data = NULL;
94         }
95 }
96
97 static int ocfs2_file_open(struct inode *inode, struct file *file)
98 {
99         int status;
100         int mode = file->f_flags;
101         struct ocfs2_inode_info *oi = OCFS2_I(inode);
102
103         trace_ocfs2_file_open(inode, file, file->f_path.dentry,
104                               (unsigned long long)OCFS2_I(inode)->ip_blkno,
105                               file->f_path.dentry->d_name.len,
106                               file->f_path.dentry->d_name.name, mode);
107
108         if (file->f_mode & FMODE_WRITE) {
109                 status = dquot_initialize(inode);
110                 if (status)
111                         goto leave;
112         }
113
114         spin_lock(&oi->ip_lock);
115
116         /* Check that the inode hasn't been wiped from disk by another
117          * node. If it hasn't then we're safe as long as we hold the
118          * spin lock until our increment of open count. */
119         if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
120                 spin_unlock(&oi->ip_lock);
121
122                 status = -ENOENT;
123                 goto leave;
124         }
125
126         if (mode & O_DIRECT)
127                 oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT;
128
129         oi->ip_open_count++;
130         spin_unlock(&oi->ip_lock);
131
132         status = ocfs2_init_file_private(inode, file);
133         if (status) {
134                 /*
135                  * We want to set open count back if we're failing the
136                  * open.
137                  */
138                 spin_lock(&oi->ip_lock);
139                 oi->ip_open_count--;
140                 spin_unlock(&oi->ip_lock);
141         }
142
143 leave:
144         return status;
145 }
146
147 static int ocfs2_file_release(struct inode *inode, struct file *file)
148 {
149         struct ocfs2_inode_info *oi = OCFS2_I(inode);
150
151         spin_lock(&oi->ip_lock);
152         if (!--oi->ip_open_count)
153                 oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT;
154
155         trace_ocfs2_file_release(inode, file, file->f_path.dentry,
156                                  oi->ip_blkno,
157                                  file->f_path.dentry->d_name.len,
158                                  file->f_path.dentry->d_name.name,
159                                  oi->ip_open_count);
160         spin_unlock(&oi->ip_lock);
161
162         ocfs2_free_file_private(inode, file);
163
164         return 0;
165 }
166
167 static int ocfs2_dir_open(struct inode *inode, struct file *file)
168 {
169         return ocfs2_init_file_private(inode, file);
170 }
171
172 static int ocfs2_dir_release(struct inode *inode, struct file *file)
173 {
174         ocfs2_free_file_private(inode, file);
175         return 0;
176 }
177
178 static int ocfs2_sync_file(struct file *file, loff_t start, loff_t end,
179                            int datasync)
180 {
181         int err = 0;
182         struct inode *inode = file->f_mapping->host;
183         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
184         struct ocfs2_inode_info *oi = OCFS2_I(inode);
185         journal_t *journal = osb->journal->j_journal;
186         int ret;
187         tid_t commit_tid;
188         bool needs_barrier = false;
189
190         trace_ocfs2_sync_file(inode, file, file->f_path.dentry,
191                               OCFS2_I(inode)->ip_blkno,
192                               file->f_path.dentry->d_name.len,
193                               file->f_path.dentry->d_name.name,
194                               (unsigned long long)datasync);
195
196         if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
197                 return -EROFS;
198
199         err = filemap_write_and_wait_range(inode->i_mapping, start, end);
200         if (err)
201                 return err;
202
203         commit_tid = datasync ? oi->i_datasync_tid : oi->i_sync_tid;
204         if (journal->j_flags & JBD2_BARRIER &&
205             !jbd2_trans_will_send_data_barrier(journal, commit_tid))
206                 needs_barrier = true;
207         err = jbd2_complete_transaction(journal, commit_tid);
208         if (needs_barrier) {
209                 ret = blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, NULL);
210                 if (!err)
211                         err = ret;
212         }
213
214         if (err)
215                 mlog_errno(err);
216
217         return (err < 0) ? -EIO : 0;
218 }
219
220 int ocfs2_should_update_atime(struct inode *inode,
221                               struct vfsmount *vfsmnt)
222 {
223         struct timespec now;
224         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
225
226         if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
227                 return 0;
228
229         if ((inode->i_flags & S_NOATIME) ||
230             ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)))
231                 return 0;
232
233         /*
234          * We can be called with no vfsmnt structure - NFSD will
235          * sometimes do this.
236          *
237          * Note that our action here is different than touch_atime() -
238          * if we can't tell whether this is a noatime mount, then we
239          * don't know whether to trust the value of s_atime_quantum.
240          */
241         if (vfsmnt == NULL)
242                 return 0;
243
244         if ((vfsmnt->mnt_flags & MNT_NOATIME) ||
245             ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
246                 return 0;
247
248         if (vfsmnt->mnt_flags & MNT_RELATIME) {
249                 if ((timespec_compare(&inode->i_atime, &inode->i_mtime) <= 0) ||
250                     (timespec_compare(&inode->i_atime, &inode->i_ctime) <= 0))
251                         return 1;
252
253                 return 0;
254         }
255
256         now = CURRENT_TIME;
257         if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum))
258                 return 0;
259         else
260                 return 1;
261 }
262
263 int ocfs2_update_inode_atime(struct inode *inode,
264                              struct buffer_head *bh)
265 {
266         int ret;
267         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
268         handle_t *handle;
269         struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data;
270
271         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
272         if (IS_ERR(handle)) {
273                 ret = PTR_ERR(handle);
274                 mlog_errno(ret);
275                 goto out;
276         }
277
278         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
279                                       OCFS2_JOURNAL_ACCESS_WRITE);
280         if (ret) {
281                 mlog_errno(ret);
282                 goto out_commit;
283         }
284
285         /*
286          * Don't use ocfs2_mark_inode_dirty() here as we don't always
287          * have i_mutex to guard against concurrent changes to other
288          * inode fields.
289          */
290         inode->i_atime = CURRENT_TIME;
291         di->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
292         di->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
293         ocfs2_update_inode_fsync_trans(handle, inode, 0);
294         ocfs2_journal_dirty(handle, bh);
295
296 out_commit:
297         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
298 out:
299         return ret;
300 }
301
302 int ocfs2_set_inode_size(handle_t *handle,
303                                 struct inode *inode,
304                                 struct buffer_head *fe_bh,
305                                 u64 new_i_size)
306 {
307         int status;
308
309         i_size_write(inode, new_i_size);
310         inode->i_blocks = ocfs2_inode_sector_count(inode);
311         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
312
313         status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
314         if (status < 0) {
315                 mlog_errno(status);
316                 goto bail;
317         }
318
319 bail:
320         return status;
321 }
322
323 int ocfs2_simple_size_update(struct inode *inode,
324                              struct buffer_head *di_bh,
325                              u64 new_i_size)
326 {
327         int ret;
328         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
329         handle_t *handle = NULL;
330
331         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
332         if (IS_ERR(handle)) {
333                 ret = PTR_ERR(handle);
334                 mlog_errno(ret);
335                 goto out;
336         }
337
338         ret = ocfs2_set_inode_size(handle, inode, di_bh,
339                                    new_i_size);
340         if (ret < 0)
341                 mlog_errno(ret);
342
343         ocfs2_update_inode_fsync_trans(handle, inode, 0);
344         ocfs2_commit_trans(osb, handle);
345 out:
346         return ret;
347 }
348
349 static int ocfs2_cow_file_pos(struct inode *inode,
350                               struct buffer_head *fe_bh,
351                               u64 offset)
352 {
353         int status;
354         u32 phys, cpos = offset >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
355         unsigned int num_clusters = 0;
356         unsigned int ext_flags = 0;
357
358         /*
359          * If the new offset is aligned to the range of the cluster, there is
360          * no space for ocfs2_zero_range_for_truncate to fill, so no need to
361          * CoW either.
362          */
363         if ((offset & (OCFS2_SB(inode->i_sb)->s_clustersize - 1)) == 0)
364                 return 0;
365
366         status = ocfs2_get_clusters(inode, cpos, &phys,
367                                     &num_clusters, &ext_flags);
368         if (status) {
369                 mlog_errno(status);
370                 goto out;
371         }
372
373         if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
374                 goto out;
375
376         return ocfs2_refcount_cow(inode, fe_bh, cpos, 1, cpos+1);
377
378 out:
379         return status;
380 }
381
382 static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
383                                      struct inode *inode,
384                                      struct buffer_head *fe_bh,
385                                      u64 new_i_size)
386 {
387         int status;
388         handle_t *handle;
389         struct ocfs2_dinode *di;
390         u64 cluster_bytes;
391
392         /*
393          * We need to CoW the cluster contains the offset if it is reflinked
394          * since we will call ocfs2_zero_range_for_truncate later which will
395          * write "0" from offset to the end of the cluster.
396          */
397         status = ocfs2_cow_file_pos(inode, fe_bh, new_i_size);
398         if (status) {
399                 mlog_errno(status);
400                 return status;
401         }
402
403         /* TODO: This needs to actually orphan the inode in this
404          * transaction. */
405
406         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
407         if (IS_ERR(handle)) {
408                 status = PTR_ERR(handle);
409                 mlog_errno(status);
410                 goto out;
411         }
412
413         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
414                                          OCFS2_JOURNAL_ACCESS_WRITE);
415         if (status < 0) {
416                 mlog_errno(status);
417                 goto out_commit;
418         }
419
420         /*
421          * Do this before setting i_size.
422          */
423         cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size);
424         status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size,
425                                                cluster_bytes);
426         if (status) {
427                 mlog_errno(status);
428                 goto out_commit;
429         }
430
431         i_size_write(inode, new_i_size);
432         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
433
434         di = (struct ocfs2_dinode *) fe_bh->b_data;
435         di->i_size = cpu_to_le64(new_i_size);
436         di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
437         di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
438         ocfs2_update_inode_fsync_trans(handle, inode, 0);
439
440         ocfs2_journal_dirty(handle, fe_bh);
441
442 out_commit:
443         ocfs2_commit_trans(osb, handle);
444 out:
445         return status;
446 }
447
448 int ocfs2_truncate_file(struct inode *inode,
449                                struct buffer_head *di_bh,
450                                u64 new_i_size)
451 {
452         int status = 0;
453         struct ocfs2_dinode *fe = NULL;
454         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
455
456         /* We trust di_bh because it comes from ocfs2_inode_lock(), which
457          * already validated it */
458         fe = (struct ocfs2_dinode *) di_bh->b_data;
459
460         trace_ocfs2_truncate_file((unsigned long long)OCFS2_I(inode)->ip_blkno,
461                                   (unsigned long long)le64_to_cpu(fe->i_size),
462                                   (unsigned long long)new_i_size);
463
464         mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode),
465                         "Inode %llu, inode i_size = %lld != di "
466                         "i_size = %llu, i_flags = 0x%x\n",
467                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
468                         i_size_read(inode),
469                         (unsigned long long)le64_to_cpu(fe->i_size),
470                         le32_to_cpu(fe->i_flags));
471
472         if (new_i_size > le64_to_cpu(fe->i_size)) {
473                 trace_ocfs2_truncate_file_error(
474                         (unsigned long long)le64_to_cpu(fe->i_size),
475                         (unsigned long long)new_i_size);
476                 status = -EINVAL;
477                 mlog_errno(status);
478                 goto bail;
479         }
480
481         down_write(&OCFS2_I(inode)->ip_alloc_sem);
482
483         ocfs2_resv_discard(&osb->osb_la_resmap,
484                            &OCFS2_I(inode)->ip_la_data_resv);
485
486         /*
487          * The inode lock forced other nodes to sync and drop their
488          * pages, which (correctly) happens even if we have a truncate
489          * without allocation change - ocfs2 cluster sizes can be much
490          * greater than page size, so we have to truncate them
491          * anyway.
492          */
493
494         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
495                 unmap_mapping_range(inode->i_mapping,
496                                     new_i_size + PAGE_SIZE - 1, 0, 1);
497                 truncate_inode_pages(inode->i_mapping, new_i_size);
498                 status = ocfs2_truncate_inline(inode, di_bh, new_i_size,
499                                                i_size_read(inode), 1);
500                 if (status)
501                         mlog_errno(status);
502
503                 goto bail_unlock_sem;
504         }
505
506         /* alright, we're going to need to do a full blown alloc size
507          * change. Orphan the inode so that recovery can complete the
508          * truncate if necessary. This does the task of marking
509          * i_size. */
510         status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
511         if (status < 0) {
512                 mlog_errno(status);
513                 goto bail_unlock_sem;
514         }
515
516         unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1);
517         truncate_inode_pages(inode->i_mapping, new_i_size);
518
519         status = ocfs2_commit_truncate(osb, inode, di_bh);
520         if (status < 0) {
521                 mlog_errno(status);
522                 goto bail_unlock_sem;
523         }
524
525         /* TODO: orphan dir cleanup here. */
526 bail_unlock_sem:
527         up_write(&OCFS2_I(inode)->ip_alloc_sem);
528
529 bail:
530         if (!status && OCFS2_I(inode)->ip_clusters == 0)
531                 status = ocfs2_try_remove_refcount_tree(inode, di_bh);
532
533         return status;
534 }
535
536 /*
537  * extend file allocation only here.
538  * we'll update all the disk stuff, and oip->alloc_size
539  *
540  * expect stuff to be locked, a transaction started and enough data /
541  * metadata reservations in the contexts.
542  *
543  * Will return -EAGAIN, and a reason if a restart is needed.
544  * If passed in, *reason will always be set, even in error.
545  */
546 int ocfs2_add_inode_data(struct ocfs2_super *osb,
547                          struct inode *inode,
548                          u32 *logical_offset,
549                          u32 clusters_to_add,
550                          int mark_unwritten,
551                          struct buffer_head *fe_bh,
552                          handle_t *handle,
553                          struct ocfs2_alloc_context *data_ac,
554                          struct ocfs2_alloc_context *meta_ac,
555                          enum ocfs2_alloc_restarted *reason_ret)
556 {
557         int ret;
558         struct ocfs2_extent_tree et;
559
560         ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), fe_bh);
561         ret = ocfs2_add_clusters_in_btree(handle, &et, logical_offset,
562                                           clusters_to_add, mark_unwritten,
563                                           data_ac, meta_ac, reason_ret);
564
565         return ret;
566 }
567
568 static int __ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
569                                      u32 clusters_to_add, int mark_unwritten)
570 {
571         int status = 0;
572         int restart_func = 0;
573         int credits;
574         u32 prev_clusters;
575         struct buffer_head *bh = NULL;
576         struct ocfs2_dinode *fe = NULL;
577         handle_t *handle = NULL;
578         struct ocfs2_alloc_context *data_ac = NULL;
579         struct ocfs2_alloc_context *meta_ac = NULL;
580         enum ocfs2_alloc_restarted why = RESTART_NONE;
581         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
582         struct ocfs2_extent_tree et;
583         int did_quota = 0;
584
585         /*
586          * Unwritten extent only exists for file systems which
587          * support holes.
588          */
589         BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb));
590
591         status = ocfs2_read_inode_block(inode, &bh);
592         if (status < 0) {
593                 mlog_errno(status);
594                 goto leave;
595         }
596         fe = (struct ocfs2_dinode *) bh->b_data;
597
598 restart_all:
599         BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
600
601         ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), bh);
602         status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
603                                        &data_ac, &meta_ac);
604         if (status) {
605                 mlog_errno(status);
606                 goto leave;
607         }
608
609         credits = ocfs2_calc_extend_credits(osb->sb, &fe->id2.i_list);
610         handle = ocfs2_start_trans(osb, credits);
611         if (IS_ERR(handle)) {
612                 status = PTR_ERR(handle);
613                 handle = NULL;
614                 mlog_errno(status);
615                 goto leave;
616         }
617
618 restarted_transaction:
619         trace_ocfs2_extend_allocation(
620                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
621                 (unsigned long long)i_size_read(inode),
622                 le32_to_cpu(fe->i_clusters), clusters_to_add,
623                 why, restart_func);
624
625         status = dquot_alloc_space_nodirty(inode,
626                         ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
627         if (status)
628                 goto leave;
629         did_quota = 1;
630
631         /* reserve a write to the file entry early on - that we if we
632          * run out of credits in the allocation path, we can still
633          * update i_size. */
634         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
635                                          OCFS2_JOURNAL_ACCESS_WRITE);
636         if (status < 0) {
637                 mlog_errno(status);
638                 goto leave;
639         }
640
641         prev_clusters = OCFS2_I(inode)->ip_clusters;
642
643         status = ocfs2_add_inode_data(osb,
644                                       inode,
645                                       &logical_start,
646                                       clusters_to_add,
647                                       mark_unwritten,
648                                       bh,
649                                       handle,
650                                       data_ac,
651                                       meta_ac,
652                                       &why);
653         if ((status < 0) && (status != -EAGAIN)) {
654                 if (status != -ENOSPC)
655                         mlog_errno(status);
656                 goto leave;
657         }
658         ocfs2_update_inode_fsync_trans(handle, inode, 1);
659         ocfs2_journal_dirty(handle, bh);
660
661         spin_lock(&OCFS2_I(inode)->ip_lock);
662         clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
663         spin_unlock(&OCFS2_I(inode)->ip_lock);
664         /* Release unused quota reservation */
665         dquot_free_space(inode,
666                         ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
667         did_quota = 0;
668
669         if (why != RESTART_NONE && clusters_to_add) {
670                 if (why == RESTART_META) {
671                         restart_func = 1;
672                         status = 0;
673                 } else {
674                         BUG_ON(why != RESTART_TRANS);
675
676                         status = ocfs2_allocate_extend_trans(handle, 1);
677                         if (status < 0) {
678                                 /* handle still has to be committed at
679                                  * this point. */
680                                 status = -ENOMEM;
681                                 mlog_errno(status);
682                                 goto leave;
683                         }
684                         goto restarted_transaction;
685                 }
686         }
687
688         trace_ocfs2_extend_allocation_end(OCFS2_I(inode)->ip_blkno,
689              le32_to_cpu(fe->i_clusters),
690              (unsigned long long)le64_to_cpu(fe->i_size),
691              OCFS2_I(inode)->ip_clusters,
692              (unsigned long long)i_size_read(inode));
693
694 leave:
695         if (status < 0 && did_quota)
696                 dquot_free_space(inode,
697                         ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
698         if (handle) {
699                 ocfs2_commit_trans(osb, handle);
700                 handle = NULL;
701         }
702         if (data_ac) {
703                 ocfs2_free_alloc_context(data_ac);
704                 data_ac = NULL;
705         }
706         if (meta_ac) {
707                 ocfs2_free_alloc_context(meta_ac);
708                 meta_ac = NULL;
709         }
710         if ((!status) && restart_func) {
711                 restart_func = 0;
712                 goto restart_all;
713         }
714         brelse(bh);
715         bh = NULL;
716
717         return status;
718 }
719
720 int ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
721                 u32 clusters_to_add, int mark_unwritten)
722 {
723         return __ocfs2_extend_allocation(inode, logical_start,
724                         clusters_to_add, mark_unwritten);
725 }
726
727 /*
728  * While a write will already be ordering the data, a truncate will not.
729  * Thus, we need to explicitly order the zeroed pages.
730  */
731 static handle_t *ocfs2_zero_start_ordered_transaction(struct inode *inode,
732                                                 struct buffer_head *di_bh)
733 {
734         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
735         handle_t *handle = NULL;
736         int ret = 0;
737
738         if (!ocfs2_should_order_data(inode))
739                 goto out;
740
741         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
742         if (IS_ERR(handle)) {
743                 ret = -ENOMEM;
744                 mlog_errno(ret);
745                 goto out;
746         }
747
748         ret = ocfs2_jbd2_file_inode(handle, inode);
749         if (ret < 0) {
750                 mlog_errno(ret);
751                 goto out;
752         }
753
754         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
755                                       OCFS2_JOURNAL_ACCESS_WRITE);
756         if (ret)
757                 mlog_errno(ret);
758         ocfs2_update_inode_fsync_trans(handle, inode, 1);
759
760 out:
761         if (ret) {
762                 if (!IS_ERR(handle))
763                         ocfs2_commit_trans(osb, handle);
764                 handle = ERR_PTR(ret);
765         }
766         return handle;
767 }
768
769 /* Some parts of this taken from generic_cont_expand, which turned out
770  * to be too fragile to do exactly what we need without us having to
771  * worry about recursive locking in ->write_begin() and ->write_end(). */
772 static int ocfs2_write_zero_page(struct inode *inode, u64 abs_from,
773                                  u64 abs_to, struct buffer_head *di_bh)
774 {
775         struct address_space *mapping = inode->i_mapping;
776         struct page *page;
777         unsigned long index = abs_from >> PAGE_CACHE_SHIFT;
778         handle_t *handle;
779         int ret = 0;
780         unsigned zero_from, zero_to, block_start, block_end;
781         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
782
783         BUG_ON(abs_from >= abs_to);
784         BUG_ON(abs_to > (((u64)index + 1) << PAGE_CACHE_SHIFT));
785         BUG_ON(abs_from & (inode->i_blkbits - 1));
786
787         handle = ocfs2_zero_start_ordered_transaction(inode, di_bh);
788         if (IS_ERR(handle)) {
789                 ret = PTR_ERR(handle);
790                 goto out;
791         }
792
793         page = find_or_create_page(mapping, index, GFP_NOFS);
794         if (!page) {
795                 ret = -ENOMEM;
796                 mlog_errno(ret);
797                 goto out_commit_trans;
798         }
799
800         /* Get the offsets within the page that we want to zero */
801         zero_from = abs_from & (PAGE_CACHE_SIZE - 1);
802         zero_to = abs_to & (PAGE_CACHE_SIZE - 1);
803         if (!zero_to)
804                 zero_to = PAGE_CACHE_SIZE;
805
806         trace_ocfs2_write_zero_page(
807                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
808                         (unsigned long long)abs_from,
809                         (unsigned long long)abs_to,
810                         index, zero_from, zero_to);
811
812         /* We know that zero_from is block aligned */
813         for (block_start = zero_from; block_start < zero_to;
814              block_start = block_end) {
815                 block_end = block_start + i_blocksize(inode);
816
817                 /*
818                  * block_start is block-aligned.  Bump it by one to force
819                  * __block_write_begin and block_commit_write to zero the
820                  * whole block.
821                  */
822                 ret = __block_write_begin(page, block_start + 1, 0,
823                                           ocfs2_get_block);
824                 if (ret < 0) {
825                         mlog_errno(ret);
826                         goto out_unlock;
827                 }
828
829
830                 /* must not update i_size! */
831                 ret = block_commit_write(page, block_start + 1,
832                                          block_start + 1);
833                 if (ret < 0)
834                         mlog_errno(ret);
835                 else
836                         ret = 0;
837         }
838
839         /*
840          * fs-writeback will release the dirty pages without page lock
841          * whose offset are over inode size, the release happens at
842          * block_write_full_page().
843          */
844         i_size_write(inode, abs_to);
845         inode->i_blocks = ocfs2_inode_sector_count(inode);
846         di->i_size = cpu_to_le64((u64)i_size_read(inode));
847         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
848         di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec);
849         di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
850         di->i_mtime_nsec = di->i_ctime_nsec;
851         if (handle) {
852                 ocfs2_journal_dirty(handle, di_bh);
853                 ocfs2_update_inode_fsync_trans(handle, inode, 1);
854         }
855
856 out_unlock:
857         unlock_page(page);
858         page_cache_release(page);
859 out_commit_trans:
860         if (handle)
861                 ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
862 out:
863         return ret;
864 }
865
866 /*
867  * Find the next range to zero.  We do this in terms of bytes because
868  * that's what ocfs2_zero_extend() wants, and it is dealing with the
869  * pagecache.  We may return multiple extents.
870  *
871  * zero_start and zero_end are ocfs2_zero_extend()s current idea of what
872  * needs to be zeroed.  range_start and range_end return the next zeroing
873  * range.  A subsequent call should pass the previous range_end as its
874  * zero_start.  If range_end is 0, there's nothing to do.
875  *
876  * Unwritten extents are skipped over.  Refcounted extents are CoWd.
877  */
878 static int ocfs2_zero_extend_get_range(struct inode *inode,
879                                        struct buffer_head *di_bh,
880                                        u64 zero_start, u64 zero_end,
881                                        u64 *range_start, u64 *range_end)
882 {
883         int rc = 0, needs_cow = 0;
884         u32 p_cpos, zero_clusters = 0;
885         u32 zero_cpos =
886                 zero_start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
887         u32 last_cpos = ocfs2_clusters_for_bytes(inode->i_sb, zero_end);
888         unsigned int num_clusters = 0;
889         unsigned int ext_flags = 0;
890
891         while (zero_cpos < last_cpos) {
892                 rc = ocfs2_get_clusters(inode, zero_cpos, &p_cpos,
893                                         &num_clusters, &ext_flags);
894                 if (rc) {
895                         mlog_errno(rc);
896                         goto out;
897                 }
898
899                 if (p_cpos && !(ext_flags & OCFS2_EXT_UNWRITTEN)) {
900                         zero_clusters = num_clusters;
901                         if (ext_flags & OCFS2_EXT_REFCOUNTED)
902                                 needs_cow = 1;
903                         break;
904                 }
905
906                 zero_cpos += num_clusters;
907         }
908         if (!zero_clusters) {
909                 *range_end = 0;
910                 goto out;
911         }
912
913         while ((zero_cpos + zero_clusters) < last_cpos) {
914                 rc = ocfs2_get_clusters(inode, zero_cpos + zero_clusters,
915                                         &p_cpos, &num_clusters,
916                                         &ext_flags);
917                 if (rc) {
918                         mlog_errno(rc);
919                         goto out;
920                 }
921
922                 if (!p_cpos || (ext_flags & OCFS2_EXT_UNWRITTEN))
923                         break;
924                 if (ext_flags & OCFS2_EXT_REFCOUNTED)
925                         needs_cow = 1;
926                 zero_clusters += num_clusters;
927         }
928         if ((zero_cpos + zero_clusters) > last_cpos)
929                 zero_clusters = last_cpos - zero_cpos;
930
931         if (needs_cow) {
932                 rc = ocfs2_refcount_cow(inode, di_bh, zero_cpos,
933                                         zero_clusters, UINT_MAX);
934                 if (rc) {
935                         mlog_errno(rc);
936                         goto out;
937                 }
938         }
939
940         *range_start = ocfs2_clusters_to_bytes(inode->i_sb, zero_cpos);
941         *range_end = ocfs2_clusters_to_bytes(inode->i_sb,
942                                              zero_cpos + zero_clusters);
943
944 out:
945         return rc;
946 }
947
948 /*
949  * Zero one range returned from ocfs2_zero_extend_get_range().  The caller
950  * has made sure that the entire range needs zeroing.
951  */
952 static int ocfs2_zero_extend_range(struct inode *inode, u64 range_start,
953                                    u64 range_end, struct buffer_head *di_bh)
954 {
955         int rc = 0;
956         u64 next_pos;
957         u64 zero_pos = range_start;
958
959         trace_ocfs2_zero_extend_range(
960                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
961                         (unsigned long long)range_start,
962                         (unsigned long long)range_end);
963         BUG_ON(range_start >= range_end);
964
965         while (zero_pos < range_end) {
966                 next_pos = (zero_pos & PAGE_CACHE_MASK) + PAGE_CACHE_SIZE;
967                 if (next_pos > range_end)
968                         next_pos = range_end;
969                 rc = ocfs2_write_zero_page(inode, zero_pos, next_pos, di_bh);
970                 if (rc < 0) {
971                         mlog_errno(rc);
972                         break;
973                 }
974                 zero_pos = next_pos;
975
976                 /*
977                  * Very large extends have the potential to lock up
978                  * the cpu for extended periods of time.
979                  */
980                 cond_resched();
981         }
982
983         return rc;
984 }
985
986 int ocfs2_zero_extend(struct inode *inode, struct buffer_head *di_bh,
987                       loff_t zero_to_size)
988 {
989         int ret = 0;
990         u64 zero_start, range_start = 0, range_end = 0;
991         struct super_block *sb = inode->i_sb;
992
993         zero_start = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
994         trace_ocfs2_zero_extend((unsigned long long)OCFS2_I(inode)->ip_blkno,
995                                 (unsigned long long)zero_start,
996                                 (unsigned long long)i_size_read(inode));
997         while (zero_start < zero_to_size) {
998                 ret = ocfs2_zero_extend_get_range(inode, di_bh, zero_start,
999                                                   zero_to_size,
1000                                                   &range_start,
1001                                                   &range_end);
1002                 if (ret) {
1003                         mlog_errno(ret);
1004                         break;
1005                 }
1006                 if (!range_end)
1007                         break;
1008                 /* Trim the ends */
1009                 if (range_start < zero_start)
1010                         range_start = zero_start;
1011                 if (range_end > zero_to_size)
1012                         range_end = zero_to_size;
1013
1014                 ret = ocfs2_zero_extend_range(inode, range_start,
1015                                               range_end, di_bh);
1016                 if (ret) {
1017                         mlog_errno(ret);
1018                         break;
1019                 }
1020                 zero_start = range_end;
1021         }
1022
1023         return ret;
1024 }
1025
1026 int ocfs2_extend_no_holes(struct inode *inode, struct buffer_head *di_bh,
1027                           u64 new_i_size, u64 zero_to)
1028 {
1029         int ret;
1030         u32 clusters_to_add;
1031         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1032
1033         /*
1034          * Only quota files call this without a bh, and they can't be
1035          * refcounted.
1036          */
1037         BUG_ON(!di_bh && (oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
1038         BUG_ON(!di_bh && !(oi->ip_flags & OCFS2_INODE_SYSTEM_FILE));
1039
1040         clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size);
1041         if (clusters_to_add < oi->ip_clusters)
1042                 clusters_to_add = 0;
1043         else
1044                 clusters_to_add -= oi->ip_clusters;
1045
1046         if (clusters_to_add) {
1047                 ret = __ocfs2_extend_allocation(inode, oi->ip_clusters,
1048                                                 clusters_to_add, 0);
1049                 if (ret) {
1050                         mlog_errno(ret);
1051                         goto out;
1052                 }
1053         }
1054
1055         /*
1056          * Call this even if we don't add any clusters to the tree. We
1057          * still need to zero the area between the old i_size and the
1058          * new i_size.
1059          */
1060         ret = ocfs2_zero_extend(inode, di_bh, zero_to);
1061         if (ret < 0)
1062                 mlog_errno(ret);
1063
1064 out:
1065         return ret;
1066 }
1067
1068 static int ocfs2_extend_file(struct inode *inode,
1069                              struct buffer_head *di_bh,
1070                              u64 new_i_size)
1071 {
1072         int ret = 0;
1073         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1074
1075         BUG_ON(!di_bh);
1076
1077         /* setattr sometimes calls us like this. */
1078         if (new_i_size == 0)
1079                 goto out;
1080
1081         if (i_size_read(inode) == new_i_size)
1082                 goto out;
1083         BUG_ON(new_i_size < i_size_read(inode));
1084
1085         /*
1086          * The alloc sem blocks people in read/write from reading our
1087          * allocation until we're done changing it. We depend on
1088          * i_mutex to block other extend/truncate calls while we're
1089          * here.  We even have to hold it for sparse files because there
1090          * might be some tail zeroing.
1091          */
1092         down_write(&oi->ip_alloc_sem);
1093
1094         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1095                 /*
1096                  * We can optimize small extends by keeping the inodes
1097                  * inline data.
1098                  */
1099                 if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) {
1100                         up_write(&oi->ip_alloc_sem);
1101                         goto out_update_size;
1102                 }
1103
1104                 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1105                 if (ret) {
1106                         up_write(&oi->ip_alloc_sem);
1107                         mlog_errno(ret);
1108                         goto out;
1109                 }
1110         }
1111
1112         if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
1113                 ret = ocfs2_zero_extend(inode, di_bh, new_i_size);
1114         else
1115                 ret = ocfs2_extend_no_holes(inode, di_bh, new_i_size,
1116                                             new_i_size);
1117
1118         up_write(&oi->ip_alloc_sem);
1119
1120         if (ret < 0) {
1121                 mlog_errno(ret);
1122                 goto out;
1123         }
1124
1125 out_update_size:
1126         ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
1127         if (ret < 0)
1128                 mlog_errno(ret);
1129
1130 out:
1131         return ret;
1132 }
1133
1134 int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
1135 {
1136         int status = 0, size_change;
1137         int inode_locked = 0;
1138         struct inode *inode = d_inode(dentry);
1139         struct super_block *sb = inode->i_sb;
1140         struct ocfs2_super *osb = OCFS2_SB(sb);
1141         struct buffer_head *bh = NULL;
1142         handle_t *handle = NULL;
1143         struct dquot *transfer_to[MAXQUOTAS] = { };
1144         int qtype;
1145
1146         trace_ocfs2_setattr(inode, dentry,
1147                             (unsigned long long)OCFS2_I(inode)->ip_blkno,
1148                             dentry->d_name.len, dentry->d_name.name,
1149                             attr->ia_valid, attr->ia_mode,
1150                             from_kuid(&init_user_ns, attr->ia_uid),
1151                             from_kgid(&init_user_ns, attr->ia_gid));
1152
1153         /* ensuring we don't even attempt to truncate a symlink */
1154         if (S_ISLNK(inode->i_mode))
1155                 attr->ia_valid &= ~ATTR_SIZE;
1156
1157 #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
1158                            | ATTR_GID | ATTR_UID | ATTR_MODE)
1159         if (!(attr->ia_valid & OCFS2_VALID_ATTRS))
1160                 return 0;
1161
1162         status = inode_change_ok(inode, attr);
1163         if (status)
1164                 return status;
1165
1166         if (is_quota_modification(inode, attr)) {
1167                 status = dquot_initialize(inode);
1168                 if (status)
1169                         return status;
1170         }
1171         size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
1172         if (size_change) {
1173                 status = ocfs2_rw_lock(inode, 1);
1174                 if (status < 0) {
1175                         mlog_errno(status);
1176                         goto bail;
1177                 }
1178         }
1179
1180         status = ocfs2_inode_lock(inode, &bh, 1);
1181         if (status < 0) {
1182                 if (status != -ENOENT)
1183                         mlog_errno(status);
1184                 goto bail_unlock_rw;
1185         }
1186         inode_locked = 1;
1187
1188         if (size_change) {
1189                 status = inode_newsize_ok(inode, attr->ia_size);
1190                 if (status)
1191                         goto bail_unlock;
1192
1193                 inode_dio_wait(inode);
1194
1195                 if (i_size_read(inode) >= attr->ia_size) {
1196                         if (ocfs2_should_order_data(inode)) {
1197                                 status = ocfs2_begin_ordered_truncate(inode,
1198                                                                       attr->ia_size);
1199                                 if (status)
1200                                         goto bail_unlock;
1201                         }
1202                         status = ocfs2_truncate_file(inode, bh, attr->ia_size);
1203                 } else
1204                         status = ocfs2_extend_file(inode, bh, attr->ia_size);
1205                 if (status < 0) {
1206                         if (status != -ENOSPC)
1207                                 mlog_errno(status);
1208                         status = -ENOSPC;
1209                         goto bail_unlock;
1210                 }
1211         }
1212
1213         if ((attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
1214             (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
1215                 /*
1216                  * Gather pointers to quota structures so that allocation /
1217                  * freeing of quota structures happens here and not inside
1218                  * dquot_transfer() where we have problems with lock ordering
1219                  */
1220                 if (attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)
1221                     && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1222                     OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
1223                         transfer_to[USRQUOTA] = dqget(sb, make_kqid_uid(attr->ia_uid));
1224                         if (IS_ERR(transfer_to[USRQUOTA])) {
1225                                 status = PTR_ERR(transfer_to[USRQUOTA]);
1226                                 goto bail_unlock;
1227                         }
1228                 }
1229                 if (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid)
1230                     && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1231                     OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
1232                         transfer_to[GRPQUOTA] = dqget(sb, make_kqid_gid(attr->ia_gid));
1233                         if (IS_ERR(transfer_to[GRPQUOTA])) {
1234                                 status = PTR_ERR(transfer_to[GRPQUOTA]);
1235                                 goto bail_unlock;
1236                         }
1237                 }
1238                 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS +
1239                                            2 * ocfs2_quota_trans_credits(sb));
1240                 if (IS_ERR(handle)) {
1241                         status = PTR_ERR(handle);
1242                         mlog_errno(status);
1243                         goto bail_unlock;
1244                 }
1245                 status = __dquot_transfer(inode, transfer_to);
1246                 if (status < 0)
1247                         goto bail_commit;
1248         } else {
1249                 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1250                 if (IS_ERR(handle)) {
1251                         status = PTR_ERR(handle);
1252                         mlog_errno(status);
1253                         goto bail_unlock;
1254                 }
1255         }
1256
1257         setattr_copy(inode, attr);
1258         mark_inode_dirty(inode);
1259
1260         status = ocfs2_mark_inode_dirty(handle, inode, bh);
1261         if (status < 0)
1262                 mlog_errno(status);
1263
1264 bail_commit:
1265         ocfs2_commit_trans(osb, handle);
1266 bail_unlock:
1267         if (status) {
1268                 ocfs2_inode_unlock(inode, 1);
1269                 inode_locked = 0;
1270         }
1271 bail_unlock_rw:
1272         if (size_change)
1273                 ocfs2_rw_unlock(inode, 1);
1274 bail:
1275
1276         /* Release quota pointers in case we acquired them */
1277         for (qtype = 0; qtype < OCFS2_MAXQUOTAS; qtype++)
1278                 dqput(transfer_to[qtype]);
1279
1280         if (!status && attr->ia_valid & ATTR_MODE) {
1281                 status = ocfs2_acl_chmod(inode, bh);
1282                 if (status < 0)
1283                         mlog_errno(status);
1284         }
1285         if (inode_locked)
1286                 ocfs2_inode_unlock(inode, 1);
1287
1288         brelse(bh);
1289         return status;
1290 }
1291
1292 int ocfs2_getattr(struct vfsmount *mnt,
1293                   struct dentry *dentry,
1294                   struct kstat *stat)
1295 {
1296         struct inode *inode = d_inode(dentry);
1297         struct super_block *sb = d_inode(dentry)->i_sb;
1298         struct ocfs2_super *osb = sb->s_fs_info;
1299         int err;
1300
1301         err = ocfs2_inode_revalidate(dentry);
1302         if (err) {
1303                 if (err != -ENOENT)
1304                         mlog_errno(err);
1305                 goto bail;
1306         }
1307
1308         generic_fillattr(inode, stat);
1309
1310         /* We set the blksize from the cluster size for performance */
1311         stat->blksize = osb->s_clustersize;
1312
1313 bail:
1314         return err;
1315 }
1316
1317 int ocfs2_permission(struct inode *inode, int mask)
1318 {
1319         int ret;
1320
1321         if (mask & MAY_NOT_BLOCK)
1322                 return -ECHILD;
1323
1324         ret = ocfs2_inode_lock(inode, NULL, 0);
1325         if (ret) {
1326                 if (ret != -ENOENT)
1327                         mlog_errno(ret);
1328                 goto out;
1329         }
1330
1331         ret = generic_permission(inode, mask);
1332
1333         ocfs2_inode_unlock(inode, 0);
1334 out:
1335         return ret;
1336 }
1337
1338 static int __ocfs2_write_remove_suid(struct inode *inode,
1339                                      struct buffer_head *bh)
1340 {
1341         int ret;
1342         handle_t *handle;
1343         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1344         struct ocfs2_dinode *di;
1345
1346         trace_ocfs2_write_remove_suid(
1347                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
1348                         inode->i_mode);
1349
1350         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1351         if (IS_ERR(handle)) {
1352                 ret = PTR_ERR(handle);
1353                 mlog_errno(ret);
1354                 goto out;
1355         }
1356
1357         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
1358                                       OCFS2_JOURNAL_ACCESS_WRITE);
1359         if (ret < 0) {
1360                 mlog_errno(ret);
1361                 goto out_trans;
1362         }
1363
1364         inode->i_mode &= ~S_ISUID;
1365         if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
1366                 inode->i_mode &= ~S_ISGID;
1367
1368         di = (struct ocfs2_dinode *) bh->b_data;
1369         di->i_mode = cpu_to_le16(inode->i_mode);
1370         ocfs2_update_inode_fsync_trans(handle, inode, 0);
1371
1372         ocfs2_journal_dirty(handle, bh);
1373
1374 out_trans:
1375         ocfs2_commit_trans(osb, handle);
1376 out:
1377         return ret;
1378 }
1379
1380 /*
1381  * Will look for holes and unwritten extents in the range starting at
1382  * pos for count bytes (inclusive).
1383  */
1384 static int ocfs2_check_range_for_holes(struct inode *inode, loff_t pos,
1385                                        size_t count)
1386 {
1387         int ret = 0;
1388         unsigned int extent_flags;
1389         u32 cpos, clusters, extent_len, phys_cpos;
1390         struct super_block *sb = inode->i_sb;
1391
1392         cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
1393         clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
1394
1395         while (clusters) {
1396                 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
1397                                          &extent_flags);
1398                 if (ret < 0) {
1399                         mlog_errno(ret);
1400                         goto out;
1401                 }
1402
1403                 if (phys_cpos == 0 || (extent_flags & OCFS2_EXT_UNWRITTEN)) {
1404                         ret = 1;
1405                         break;
1406                 }
1407
1408                 if (extent_len > clusters)
1409                         extent_len = clusters;
1410
1411                 clusters -= extent_len;
1412                 cpos += extent_len;
1413         }
1414 out:
1415         return ret;
1416 }
1417
1418 static int ocfs2_write_remove_suid(struct inode *inode)
1419 {
1420         int ret;
1421         struct buffer_head *bh = NULL;
1422
1423         ret = ocfs2_read_inode_block(inode, &bh);
1424         if (ret < 0) {
1425                 mlog_errno(ret);
1426                 goto out;
1427         }
1428
1429         ret =  __ocfs2_write_remove_suid(inode, bh);
1430 out:
1431         brelse(bh);
1432         return ret;
1433 }
1434
1435 /*
1436  * Allocate enough extents to cover the region starting at byte offset
1437  * start for len bytes. Existing extents are skipped, any extents
1438  * added are marked as "unwritten".
1439  */
1440 static int ocfs2_allocate_unwritten_extents(struct inode *inode,
1441                                             u64 start, u64 len)
1442 {
1443         int ret;
1444         u32 cpos, phys_cpos, clusters, alloc_size;
1445         u64 end = start + len;
1446         struct buffer_head *di_bh = NULL;
1447
1448         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1449                 ret = ocfs2_read_inode_block(inode, &di_bh);
1450                 if (ret) {
1451                         mlog_errno(ret);
1452                         goto out;
1453                 }
1454
1455                 /*
1456                  * Nothing to do if the requested reservation range
1457                  * fits within the inode.
1458                  */
1459                 if (ocfs2_size_fits_inline_data(di_bh, end))
1460                         goto out;
1461
1462                 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1463                 if (ret) {
1464                         mlog_errno(ret);
1465                         goto out;
1466                 }
1467         }
1468
1469         /*
1470          * We consider both start and len to be inclusive.
1471          */
1472         cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
1473         clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len);
1474         clusters -= cpos;
1475
1476         while (clusters) {
1477                 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1478                                          &alloc_size, NULL);
1479                 if (ret) {
1480                         mlog_errno(ret);
1481                         goto out;
1482                 }
1483
1484                 /*
1485                  * Hole or existing extent len can be arbitrary, so
1486                  * cap it to our own allocation request.
1487                  */
1488                 if (alloc_size > clusters)
1489                         alloc_size = clusters;
1490
1491                 if (phys_cpos) {
1492                         /*
1493                          * We already have an allocation at this
1494                          * region so we can safely skip it.
1495                          */
1496                         goto next;
1497                 }
1498
1499                 ret = __ocfs2_extend_allocation(inode, cpos, alloc_size, 1);
1500                 if (ret) {
1501                         if (ret != -ENOSPC)
1502                                 mlog_errno(ret);
1503                         goto out;
1504                 }
1505
1506 next:
1507                 cpos += alloc_size;
1508                 clusters -= alloc_size;
1509         }
1510
1511         ret = 0;
1512 out:
1513
1514         brelse(di_bh);
1515         return ret;
1516 }
1517
1518 /*
1519  * Truncate a byte range, avoiding pages within partial clusters. This
1520  * preserves those pages for the zeroing code to write to.
1521  */
1522 static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start,
1523                                          u64 byte_len)
1524 {
1525         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1526         loff_t start, end;
1527         struct address_space *mapping = inode->i_mapping;
1528
1529         start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start);
1530         end = byte_start + byte_len;
1531         end = end & ~(osb->s_clustersize - 1);
1532
1533         if (start < end) {
1534                 unmap_mapping_range(mapping, start, end - start, 0);
1535                 truncate_inode_pages_range(mapping, start, end - 1);
1536         }
1537 }
1538
1539 /*
1540  * zero out partial blocks of one cluster.
1541  *
1542  * start: file offset where zero starts, will be made upper block aligned.
1543  * len: it will be trimmed to the end of current cluster if "start + len"
1544  *      is bigger than it.
1545  */
1546 static int ocfs2_zeroout_partial_cluster(struct inode *inode,
1547                                         u64 start, u64 len)
1548 {
1549         int ret;
1550         u64 start_block, end_block, nr_blocks;
1551         u64 p_block, offset;
1552         u32 cluster, p_cluster, nr_clusters;
1553         struct super_block *sb = inode->i_sb;
1554         u64 end = ocfs2_align_bytes_to_clusters(sb, start);
1555
1556         if (start + len < end)
1557                 end = start + len;
1558
1559         start_block = ocfs2_blocks_for_bytes(sb, start);
1560         end_block = ocfs2_blocks_for_bytes(sb, end);
1561         nr_blocks = end_block - start_block;
1562         if (!nr_blocks)
1563                 return 0;
1564
1565         cluster = ocfs2_bytes_to_clusters(sb, start);
1566         ret = ocfs2_get_clusters(inode, cluster, &p_cluster,
1567                                 &nr_clusters, NULL);
1568         if (ret)
1569                 return ret;
1570         if (!p_cluster)
1571                 return 0;
1572
1573         offset = start_block - ocfs2_clusters_to_blocks(sb, cluster);
1574         p_block = ocfs2_clusters_to_blocks(sb, p_cluster) + offset;
1575         return sb_issue_zeroout(sb, p_block, nr_blocks, GFP_NOFS);
1576 }
1577
1578 static int ocfs2_zero_partial_clusters(struct inode *inode,
1579                                        u64 start, u64 len)
1580 {
1581         int ret = 0;
1582         u64 tmpend = 0;
1583         u64 end = start + len;
1584         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1585         unsigned int csize = osb->s_clustersize;
1586         handle_t *handle;
1587         loff_t isize = i_size_read(inode);
1588
1589         /*
1590          * The "start" and "end" values are NOT necessarily part of
1591          * the range whose allocation is being deleted. Rather, this
1592          * is what the user passed in with the request. We must zero
1593          * partial clusters here. There's no need to worry about
1594          * physical allocation - the zeroing code knows to skip holes.
1595          */
1596         trace_ocfs2_zero_partial_clusters(
1597                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1598                 (unsigned long long)start, (unsigned long long)end);
1599
1600         /*
1601          * If both edges are on a cluster boundary then there's no
1602          * zeroing required as the region is part of the allocation to
1603          * be truncated.
1604          */
1605         if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0)
1606                 goto out;
1607
1608         /* No page cache for EOF blocks, issue zero out to disk. */
1609         if (end > isize) {
1610                 /*
1611                  * zeroout eof blocks in last cluster starting from
1612                  * "isize" even "start" > "isize" because it is
1613                  * complicated to zeroout just at "start" as "start"
1614                  * may be not aligned with block size, buffer write
1615                  * would be required to do that, but out of eof buffer
1616                  * write is not supported.
1617                  */
1618                 ret = ocfs2_zeroout_partial_cluster(inode, isize,
1619                                         end - isize);
1620                 if (ret) {
1621                         mlog_errno(ret);
1622                         goto out;
1623                 }
1624                 if (start >= isize)
1625                         goto out;
1626                 end = isize;
1627         }
1628         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1629         if (IS_ERR(handle)) {
1630                 ret = PTR_ERR(handle);
1631                 mlog_errno(ret);
1632                 goto out;
1633         }
1634
1635         /*
1636          * If start is on a cluster boundary and end is somewhere in another
1637          * cluster, we have not COWed the cluster starting at start, unless
1638          * end is also within the same cluster. So, in this case, we skip this
1639          * first call to ocfs2_zero_range_for_truncate() truncate and move on
1640          * to the next one.
1641          */
1642         if ((start & (csize - 1)) != 0) {
1643                 /*
1644                  * We want to get the byte offset of the end of the 1st
1645                  * cluster.
1646                  */
1647                 tmpend = (u64)osb->s_clustersize +
1648                         (start & ~(osb->s_clustersize - 1));
1649                 if (tmpend > end)
1650                         tmpend = end;
1651
1652                 trace_ocfs2_zero_partial_clusters_range1(
1653                         (unsigned long long)start,
1654                         (unsigned long long)tmpend);
1655
1656                 ret = ocfs2_zero_range_for_truncate(inode, handle, start,
1657                                                     tmpend);
1658                 if (ret)
1659                         mlog_errno(ret);
1660         }
1661
1662         if (tmpend < end) {
1663                 /*
1664                  * This may make start and end equal, but the zeroing
1665                  * code will skip any work in that case so there's no
1666                  * need to catch it up here.
1667                  */
1668                 start = end & ~(osb->s_clustersize - 1);
1669
1670                 trace_ocfs2_zero_partial_clusters_range2(
1671                         (unsigned long long)start, (unsigned long long)end);
1672
1673                 ret = ocfs2_zero_range_for_truncate(inode, handle, start, end);
1674                 if (ret)
1675                         mlog_errno(ret);
1676         }
1677         ocfs2_update_inode_fsync_trans(handle, inode, 1);
1678
1679         ocfs2_commit_trans(osb, handle);
1680 out:
1681         return ret;
1682 }
1683
1684 static int ocfs2_find_rec(struct ocfs2_extent_list *el, u32 pos)
1685 {
1686         int i;
1687         struct ocfs2_extent_rec *rec = NULL;
1688
1689         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
1690
1691                 rec = &el->l_recs[i];
1692
1693                 if (le32_to_cpu(rec->e_cpos) < pos)
1694                         break;
1695         }
1696
1697         return i;
1698 }
1699
1700 /*
1701  * Helper to calculate the punching pos and length in one run, we handle the
1702  * following three cases in order:
1703  *
1704  * - remove the entire record
1705  * - remove a partial record
1706  * - no record needs to be removed (hole-punching completed)
1707 */
1708 static void ocfs2_calc_trunc_pos(struct inode *inode,
1709                                  struct ocfs2_extent_list *el,
1710                                  struct ocfs2_extent_rec *rec,
1711                                  u32 trunc_start, u32 *trunc_cpos,
1712                                  u32 *trunc_len, u32 *trunc_end,
1713                                  u64 *blkno, int *done)
1714 {
1715         int ret = 0;
1716         u32 coff, range;
1717
1718         range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
1719
1720         if (le32_to_cpu(rec->e_cpos) >= trunc_start) {
1721                 /*
1722                  * remove an entire extent record.
1723                  */
1724                 *trunc_cpos = le32_to_cpu(rec->e_cpos);
1725                 /*
1726                  * Skip holes if any.
1727                  */
1728                 if (range < *trunc_end)
1729                         *trunc_end = range;
1730                 *trunc_len = *trunc_end - le32_to_cpu(rec->e_cpos);
1731                 *blkno = le64_to_cpu(rec->e_blkno);
1732                 *trunc_end = le32_to_cpu(rec->e_cpos);
1733         } else if (range > trunc_start) {
1734                 /*
1735                  * remove a partial extent record, which means we're
1736                  * removing the last extent record.
1737                  */
1738                 *trunc_cpos = trunc_start;
1739                 /*
1740                  * skip hole if any.
1741                  */
1742                 if (range < *trunc_end)
1743                         *trunc_end = range;
1744                 *trunc_len = *trunc_end - trunc_start;
1745                 coff = trunc_start - le32_to_cpu(rec->e_cpos);
1746                 *blkno = le64_to_cpu(rec->e_blkno) +
1747                                 ocfs2_clusters_to_blocks(inode->i_sb, coff);
1748                 *trunc_end = trunc_start;
1749         } else {
1750                 /*
1751                  * It may have two following possibilities:
1752                  *
1753                  * - last record has been removed
1754                  * - trunc_start was within a hole
1755                  *
1756                  * both two cases mean the completion of hole punching.
1757                  */
1758                 ret = 1;
1759         }
1760
1761         *done = ret;
1762 }
1763
1764 static int ocfs2_remove_inode_range(struct inode *inode,
1765                                     struct buffer_head *di_bh, u64 byte_start,
1766                                     u64 byte_len)
1767 {
1768         int ret = 0, flags = 0, done = 0, i;
1769         u32 trunc_start, trunc_len, trunc_end, trunc_cpos, phys_cpos;
1770         u32 cluster_in_el;
1771         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1772         struct ocfs2_cached_dealloc_ctxt dealloc;
1773         struct address_space *mapping = inode->i_mapping;
1774         struct ocfs2_extent_tree et;
1775         struct ocfs2_path *path = NULL;
1776         struct ocfs2_extent_list *el = NULL;
1777         struct ocfs2_extent_rec *rec = NULL;
1778         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1779         u64 blkno, refcount_loc = le64_to_cpu(di->i_refcount_loc);
1780
1781         ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
1782         ocfs2_init_dealloc_ctxt(&dealloc);
1783
1784         trace_ocfs2_remove_inode_range(
1785                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
1786                         (unsigned long long)byte_start,
1787                         (unsigned long long)byte_len);
1788
1789         if (byte_len == 0)
1790                 return 0;
1791
1792         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1793                 ret = ocfs2_truncate_inline(inode, di_bh, byte_start,
1794                                             byte_start + byte_len, 0);
1795                 if (ret) {
1796                         mlog_errno(ret);
1797                         goto out;
1798                 }
1799                 /*
1800                  * There's no need to get fancy with the page cache
1801                  * truncate of an inline-data inode. We're talking
1802                  * about less than a page here, which will be cached
1803                  * in the dinode buffer anyway.
1804                  */
1805                 unmap_mapping_range(mapping, 0, 0, 0);
1806                 truncate_inode_pages(mapping, 0);
1807                 goto out;
1808         }
1809
1810         /*
1811          * For reflinks, we may need to CoW 2 clusters which might be
1812          * partially zero'd later, if hole's start and end offset were
1813          * within one cluster(means is not exactly aligned to clustersize).
1814          */
1815
1816         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) {
1817
1818                 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start);
1819                 if (ret) {
1820                         mlog_errno(ret);
1821                         goto out;
1822                 }
1823
1824                 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start + byte_len);
1825                 if (ret) {
1826                         mlog_errno(ret);
1827                         goto out;
1828                 }
1829         }
1830
1831         trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start);
1832         trunc_end = (byte_start + byte_len) >> osb->s_clustersize_bits;
1833         cluster_in_el = trunc_end;
1834
1835         ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len);
1836         if (ret) {
1837                 mlog_errno(ret);
1838                 goto out;
1839         }
1840
1841         path = ocfs2_new_path_from_et(&et);
1842         if (!path) {
1843                 ret = -ENOMEM;
1844                 mlog_errno(ret);
1845                 goto out;
1846         }
1847
1848         while (trunc_end > trunc_start) {
1849
1850                 ret = ocfs2_find_path(INODE_CACHE(inode), path,
1851                                       cluster_in_el);
1852                 if (ret) {
1853                         mlog_errno(ret);
1854                         goto out;
1855                 }
1856
1857                 el = path_leaf_el(path);
1858
1859                 i = ocfs2_find_rec(el, trunc_end);
1860                 /*
1861                  * Need to go to previous extent block.
1862                  */
1863                 if (i < 0) {
1864                         if (path->p_tree_depth == 0)
1865                                 break;
1866
1867                         ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb,
1868                                                             path,
1869                                                             &cluster_in_el);
1870                         if (ret) {
1871                                 mlog_errno(ret);
1872                                 goto out;
1873                         }
1874
1875                         /*
1876                          * We've reached the leftmost extent block,
1877                          * it's safe to leave.
1878                          */
1879                         if (cluster_in_el == 0)
1880                                 break;
1881
1882                         /*
1883                          * The 'pos' searched for previous extent block is
1884                          * always one cluster less than actual trunc_end.
1885                          */
1886                         trunc_end = cluster_in_el + 1;
1887
1888                         ocfs2_reinit_path(path, 1);
1889
1890                         continue;
1891
1892                 } else
1893                         rec = &el->l_recs[i];
1894
1895                 ocfs2_calc_trunc_pos(inode, el, rec, trunc_start, &trunc_cpos,
1896                                      &trunc_len, &trunc_end, &blkno, &done);
1897                 if (done)
1898                         break;
1899
1900                 flags = rec->e_flags;
1901                 phys_cpos = ocfs2_blocks_to_clusters(inode->i_sb, blkno);
1902
1903                 ret = ocfs2_remove_btree_range(inode, &et, trunc_cpos,
1904                                                phys_cpos, trunc_len, flags,
1905                                                &dealloc, refcount_loc, false);
1906                 if (ret < 0) {
1907                         mlog_errno(ret);
1908                         goto out;
1909                 }
1910
1911                 cluster_in_el = trunc_end;
1912
1913                 ocfs2_reinit_path(path, 1);
1914         }
1915
1916         ocfs2_truncate_cluster_pages(inode, byte_start, byte_len);
1917
1918 out:
1919         ocfs2_free_path(path);
1920         ocfs2_schedule_truncate_log_flush(osb, 1);
1921         ocfs2_run_deallocs(osb, &dealloc);
1922
1923         return ret;
1924 }
1925
1926 /*
1927  * Parts of this function taken from xfs_change_file_space()
1928  */
1929 static int __ocfs2_change_file_space(struct file *file, struct inode *inode,
1930                                      loff_t f_pos, unsigned int cmd,
1931                                      struct ocfs2_space_resv *sr,
1932                                      int change_size)
1933 {
1934         int ret;
1935         s64 llen;
1936         loff_t size, orig_isize;
1937         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1938         struct buffer_head *di_bh = NULL;
1939         handle_t *handle;
1940         unsigned long long max_off = inode->i_sb->s_maxbytes;
1941
1942         if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
1943                 return -EROFS;
1944
1945         mutex_lock(&inode->i_mutex);
1946
1947         /*
1948          * This prevents concurrent writes on other nodes
1949          */
1950         ret = ocfs2_rw_lock(inode, 1);
1951         if (ret) {
1952                 mlog_errno(ret);
1953                 goto out;
1954         }
1955
1956         ret = ocfs2_inode_lock(inode, &di_bh, 1);
1957         if (ret) {
1958                 mlog_errno(ret);
1959                 goto out_rw_unlock;
1960         }
1961
1962         if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
1963                 ret = -EPERM;
1964                 goto out_inode_unlock;
1965         }
1966
1967         switch (sr->l_whence) {
1968         case 0: /*SEEK_SET*/
1969                 break;
1970         case 1: /*SEEK_CUR*/
1971                 sr->l_start += f_pos;
1972                 break;
1973         case 2: /*SEEK_END*/
1974                 sr->l_start += i_size_read(inode);
1975                 break;
1976         default:
1977                 ret = -EINVAL;
1978                 goto out_inode_unlock;
1979         }
1980         sr->l_whence = 0;
1981
1982         llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len;
1983
1984         if (sr->l_start < 0
1985             || sr->l_start > max_off
1986             || (sr->l_start + llen) < 0
1987             || (sr->l_start + llen) > max_off) {
1988                 ret = -EINVAL;
1989                 goto out_inode_unlock;
1990         }
1991         size = sr->l_start + sr->l_len;
1992
1993         if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64 ||
1994             cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) {
1995                 if (sr->l_len <= 0) {
1996                         ret = -EINVAL;
1997                         goto out_inode_unlock;
1998                 }
1999         }
2000
2001         if (file && should_remove_suid(file->f_path.dentry)) {
2002                 ret = __ocfs2_write_remove_suid(inode, di_bh);
2003                 if (ret) {
2004                         mlog_errno(ret);
2005                         goto out_inode_unlock;
2006                 }
2007         }
2008
2009         down_write(&OCFS2_I(inode)->ip_alloc_sem);
2010         switch (cmd) {
2011         case OCFS2_IOC_RESVSP:
2012         case OCFS2_IOC_RESVSP64:
2013                 /*
2014                  * This takes unsigned offsets, but the signed ones we
2015                  * pass have been checked against overflow above.
2016                  */
2017                 ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start,
2018                                                        sr->l_len);
2019                 break;
2020         case OCFS2_IOC_UNRESVSP:
2021         case OCFS2_IOC_UNRESVSP64:
2022                 ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start,
2023                                                sr->l_len);
2024                 break;
2025         default:
2026                 ret = -EINVAL;
2027         }
2028
2029         orig_isize = i_size_read(inode);
2030         /* zeroout eof blocks in the cluster. */
2031         if (!ret && change_size && orig_isize < size) {
2032                 ret = ocfs2_zeroout_partial_cluster(inode, orig_isize,
2033                                         size - orig_isize);
2034                 if (!ret)
2035                         i_size_write(inode, size);
2036         }
2037         up_write(&OCFS2_I(inode)->ip_alloc_sem);
2038         if (ret) {
2039                 mlog_errno(ret);
2040                 goto out_inode_unlock;
2041         }
2042
2043         /*
2044          * We update c/mtime for these changes
2045          */
2046         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
2047         if (IS_ERR(handle)) {
2048                 ret = PTR_ERR(handle);
2049                 mlog_errno(ret);
2050                 goto out_inode_unlock;
2051         }
2052
2053         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
2054         ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
2055         if (ret < 0)
2056                 mlog_errno(ret);
2057
2058         if (file && (file->f_flags & O_SYNC))
2059                 handle->h_sync = 1;
2060
2061         ocfs2_commit_trans(osb, handle);
2062
2063 out_inode_unlock:
2064         brelse(di_bh);
2065         ocfs2_inode_unlock(inode, 1);
2066 out_rw_unlock:
2067         ocfs2_rw_unlock(inode, 1);
2068
2069 out:
2070         mutex_unlock(&inode->i_mutex);
2071         return ret;
2072 }
2073
2074 int ocfs2_change_file_space(struct file *file, unsigned int cmd,
2075                             struct ocfs2_space_resv *sr)
2076 {
2077         struct inode *inode = file_inode(file);
2078         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2079         int ret;
2080
2081         if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) &&
2082             !ocfs2_writes_unwritten_extents(osb))
2083                 return -ENOTTY;
2084         else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) &&
2085                  !ocfs2_sparse_alloc(osb))
2086                 return -ENOTTY;
2087
2088         if (!S_ISREG(inode->i_mode))
2089                 return -EINVAL;
2090
2091         if (!(file->f_mode & FMODE_WRITE))
2092                 return -EBADF;
2093
2094         ret = mnt_want_write_file(file);
2095         if (ret)
2096                 return ret;
2097         ret = __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0);
2098         mnt_drop_write_file(file);
2099         return ret;
2100 }
2101
2102 static long ocfs2_fallocate(struct file *file, int mode, loff_t offset,
2103                             loff_t len)
2104 {
2105         struct inode *inode = file_inode(file);
2106         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2107         struct ocfs2_space_resv sr;
2108         int change_size = 1;
2109         int cmd = OCFS2_IOC_RESVSP64;
2110
2111         if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2112                 return -EOPNOTSUPP;
2113         if (!ocfs2_writes_unwritten_extents(osb))
2114                 return -EOPNOTSUPP;
2115
2116         if (mode & FALLOC_FL_KEEP_SIZE)
2117                 change_size = 0;
2118
2119         if (mode & FALLOC_FL_PUNCH_HOLE)
2120                 cmd = OCFS2_IOC_UNRESVSP64;
2121
2122         sr.l_whence = 0;
2123         sr.l_start = (s64)offset;
2124         sr.l_len = (s64)len;
2125
2126         return __ocfs2_change_file_space(NULL, inode, offset, cmd, &sr,
2127                                          change_size);
2128 }
2129
2130 int ocfs2_check_range_for_refcount(struct inode *inode, loff_t pos,
2131                                    size_t count)
2132 {
2133         int ret = 0;
2134         unsigned int extent_flags;
2135         u32 cpos, clusters, extent_len, phys_cpos;
2136         struct super_block *sb = inode->i_sb;
2137
2138         if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)) ||
2139             !(OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) ||
2140             OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
2141                 return 0;
2142
2143         cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
2144         clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
2145
2146         while (clusters) {
2147                 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
2148                                          &extent_flags);
2149                 if (ret < 0) {
2150                         mlog_errno(ret);
2151                         goto out;
2152                 }
2153
2154                 if (phys_cpos && (extent_flags & OCFS2_EXT_REFCOUNTED)) {
2155                         ret = 1;
2156                         break;
2157                 }
2158
2159                 if (extent_len > clusters)
2160                         extent_len = clusters;
2161
2162                 clusters -= extent_len;
2163                 cpos += extent_len;
2164         }
2165 out:
2166         return ret;
2167 }
2168
2169 static int ocfs2_is_io_unaligned(struct inode *inode, size_t count, loff_t pos)
2170 {
2171         int blockmask = inode->i_sb->s_blocksize - 1;
2172         loff_t final_size = pos + count;
2173
2174         if ((pos & blockmask) || (final_size & blockmask))
2175                 return 1;
2176         return 0;
2177 }
2178
2179 static int ocfs2_prepare_inode_for_refcount(struct inode *inode,
2180                                             struct file *file,
2181                                             loff_t pos, size_t count,
2182                                             int *meta_level)
2183 {
2184         int ret;
2185         struct buffer_head *di_bh = NULL;
2186         u32 cpos = pos >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
2187         u32 clusters =
2188                 ocfs2_clusters_for_bytes(inode->i_sb, pos + count) - cpos;
2189
2190         ret = ocfs2_inode_lock(inode, &di_bh, 1);
2191         if (ret) {
2192                 mlog_errno(ret);
2193                 goto out;
2194         }
2195
2196         *meta_level = 1;
2197
2198         ret = ocfs2_refcount_cow(inode, di_bh, cpos, clusters, UINT_MAX);
2199         if (ret)
2200                 mlog_errno(ret);
2201 out:
2202         brelse(di_bh);
2203         return ret;
2204 }
2205
2206 static int ocfs2_prepare_inode_for_write(struct file *file,
2207                                          loff_t pos,
2208                                          size_t count,
2209                                          int appending,
2210                                          int *direct_io,
2211                                          int *has_refcount)
2212 {
2213         int ret = 0, meta_level = 0;
2214         struct dentry *dentry = file->f_path.dentry;
2215         struct inode *inode = d_inode(dentry);
2216         loff_t end;
2217         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2218         int full_coherency = !(osb->s_mount_opt &
2219                 OCFS2_MOUNT_COHERENCY_BUFFERED);
2220
2221         /*
2222          * We start with a read level meta lock and only jump to an ex
2223          * if we need to make modifications here.
2224          */
2225         for(;;) {
2226                 ret = ocfs2_inode_lock(inode, NULL, meta_level);
2227                 if (ret < 0) {
2228                         meta_level = -1;
2229                         mlog_errno(ret);
2230                         goto out;
2231                 }
2232
2233                 /* Clear suid / sgid if necessary. We do this here
2234                  * instead of later in the write path because
2235                  * remove_suid() calls ->setattr without any hint that
2236                  * we may have already done our cluster locking. Since
2237                  * ocfs2_setattr() *must* take cluster locks to
2238                  * proceed, this will lead us to recursively lock the
2239                  * inode. There's also the dinode i_size state which
2240                  * can be lost via setattr during extending writes (we
2241                  * set inode->i_size at the end of a write. */
2242                 if (should_remove_suid(dentry)) {
2243                         if (meta_level == 0) {
2244                                 ocfs2_inode_unlock(inode, meta_level);
2245                                 meta_level = 1;
2246                                 continue;
2247                         }
2248
2249                         ret = ocfs2_write_remove_suid(inode);
2250                         if (ret < 0) {
2251                                 mlog_errno(ret);
2252                                 goto out_unlock;
2253                         }
2254                 }
2255
2256                 end = pos + count;
2257
2258                 ret = ocfs2_check_range_for_refcount(inode, pos, count);
2259                 if (ret == 1) {
2260                         ocfs2_inode_unlock(inode, meta_level);
2261                         meta_level = -1;
2262
2263                         ret = ocfs2_prepare_inode_for_refcount(inode,
2264                                                                file,
2265                                                                pos,
2266                                                                count,
2267                                                                &meta_level);
2268                         if (has_refcount)
2269                                 *has_refcount = 1;
2270                         if (direct_io)
2271                                 *direct_io = 0;
2272                 }
2273
2274                 if (ret < 0) {
2275                         mlog_errno(ret);
2276                         goto out_unlock;
2277                 }
2278
2279                 /*
2280                  * Skip the O_DIRECT checks if we don't need
2281                  * them.
2282                  */
2283                 if (!direct_io || !(*direct_io))
2284                         break;
2285
2286                 /*
2287                  * There's no sane way to do direct writes to an inode
2288                  * with inline data.
2289                  */
2290                 if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2291                         *direct_io = 0;
2292                         break;
2293                 }
2294
2295                 /*
2296                  * Allowing concurrent direct writes means
2297                  * i_size changes wouldn't be synchronized, so
2298                  * one node could wind up truncating another
2299                  * nodes writes.
2300                  */
2301                 if (end > i_size_read(inode) && !full_coherency) {
2302                         *direct_io = 0;
2303                         break;
2304                 }
2305
2306                 /*
2307                  * Fallback to old way if the feature bit is not set.
2308                  */
2309                 if (end > i_size_read(inode) &&
2310                                 !ocfs2_supports_append_dio(osb)) {
2311                         *direct_io = 0;
2312                         break;
2313                 }
2314
2315                 /*
2316                  * We don't fill holes during direct io, so
2317                  * check for them here. If any are found, the
2318                  * caller will have to retake some cluster
2319                  * locks and initiate the io as buffered.
2320                  */
2321                 ret = ocfs2_check_range_for_holes(inode, pos, count);
2322                 if (ret == 1) {
2323                         /*
2324                          * Fallback to old way if the feature bit is not set.
2325                          * Otherwise try dio first and then complete the rest
2326                          * request through buffer io.
2327                          */
2328                         if (!ocfs2_supports_append_dio(osb))
2329                                 *direct_io = 0;
2330                         ret = 0;
2331                 } else if (ret < 0)
2332                         mlog_errno(ret);
2333                 break;
2334         }
2335
2336 out_unlock:
2337         trace_ocfs2_prepare_inode_for_write(OCFS2_I(inode)->ip_blkno,
2338                                             pos, appending, count,
2339                                             direct_io, has_refcount);
2340
2341         if (meta_level >= 0)
2342                 ocfs2_inode_unlock(inode, meta_level);
2343
2344 out:
2345         return ret;
2346 }
2347
2348 static ssize_t ocfs2_file_write_iter(struct kiocb *iocb,
2349                                     struct iov_iter *from)
2350 {
2351         int direct_io, appending, rw_level;
2352         int can_do_direct, has_refcount = 0;
2353         ssize_t written = 0;
2354         ssize_t ret;
2355         size_t count = iov_iter_count(from), orig_count;
2356         struct file *file = iocb->ki_filp;
2357         struct inode *inode = file_inode(file);
2358         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2359         int full_coherency = !(osb->s_mount_opt &
2360                                OCFS2_MOUNT_COHERENCY_BUFFERED);
2361         int unaligned_dio = 0;
2362         int dropped_dio = 0;
2363         int append_write = ((iocb->ki_pos + count) >=
2364                         i_size_read(inode) ? 1 : 0);
2365
2366         trace_ocfs2_file_aio_write(inode, file, file->f_path.dentry,
2367                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
2368                 file->f_path.dentry->d_name.len,
2369                 file->f_path.dentry->d_name.name,
2370                 (unsigned int)from->nr_segs);   /* GRRRRR */
2371
2372         if (count == 0)
2373                 return 0;
2374
2375         appending = iocb->ki_flags & IOCB_APPEND ? 1 : 0;
2376         direct_io = iocb->ki_flags & IOCB_DIRECT ? 1 : 0;
2377
2378         mutex_lock(&inode->i_mutex);
2379
2380 relock:
2381         /*
2382          * Concurrent O_DIRECT writes are allowed with
2383          * mount_option "coherency=buffered".
2384          * For append write, we must take rw EX.
2385          */
2386         rw_level = (!direct_io || full_coherency || append_write);
2387
2388         ret = ocfs2_rw_lock(inode, rw_level);
2389         if (ret < 0) {
2390                 mlog_errno(ret);
2391                 goto out_mutex;
2392         }
2393
2394         /*
2395          * O_DIRECT writes with "coherency=full" need to take EX cluster
2396          * inode_lock to guarantee coherency.
2397          */
2398         if (direct_io && full_coherency) {
2399                 /*
2400                  * We need to take and drop the inode lock to force
2401                  * other nodes to drop their caches.  Buffered I/O
2402                  * already does this in write_begin().
2403                  */
2404                 ret = ocfs2_inode_lock(inode, NULL, 1);
2405                 if (ret < 0) {
2406                         mlog_errno(ret);
2407                         goto out;
2408                 }
2409
2410                 ocfs2_inode_unlock(inode, 1);
2411         }
2412
2413         orig_count = iov_iter_count(from);
2414         ret = generic_write_checks(iocb, from);
2415         if (ret <= 0) {
2416                 if (ret)
2417                         mlog_errno(ret);
2418                 goto out;
2419         }
2420         count = ret;
2421
2422         can_do_direct = direct_io;
2423         ret = ocfs2_prepare_inode_for_write(file, iocb->ki_pos, count, appending,
2424                                             &can_do_direct, &has_refcount);
2425         if (ret < 0) {
2426                 mlog_errno(ret);
2427                 goto out;
2428         }
2429
2430         if (direct_io && !is_sync_kiocb(iocb))
2431                 unaligned_dio = ocfs2_is_io_unaligned(inode, count, iocb->ki_pos);
2432
2433         /*
2434          * We can't complete the direct I/O as requested, fall back to
2435          * buffered I/O.
2436          */
2437         if (direct_io && !can_do_direct) {
2438                 ocfs2_rw_unlock(inode, rw_level);
2439
2440                 rw_level = -1;
2441
2442                 direct_io = 0;
2443                 iocb->ki_flags &= ~IOCB_DIRECT;
2444                 iov_iter_reexpand(from, orig_count);
2445                 dropped_dio = 1;
2446                 goto relock;
2447         }
2448
2449         if (unaligned_dio) {
2450                 /*
2451                  * Wait on previous unaligned aio to complete before
2452                  * proceeding.
2453                  */
2454                 mutex_lock(&OCFS2_I(inode)->ip_unaligned_aio);
2455                 /* Mark the iocb as needing an unlock in ocfs2_dio_end_io */
2456                 ocfs2_iocb_set_unaligned_aio(iocb);
2457         }
2458
2459         /* communicate with ocfs2_dio_end_io */
2460         ocfs2_iocb_set_rw_locked(iocb, rw_level);
2461
2462         written = __generic_file_write_iter(iocb, from);
2463         /* buffered aio wouldn't have proper lock coverage today */
2464         BUG_ON(written == -EIOCBQUEUED && !(iocb->ki_flags & IOCB_DIRECT));
2465
2466         /*
2467          * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
2468          * function pointer which is called when o_direct io completes so that
2469          * it can unlock our rw lock.
2470          * Unfortunately there are error cases which call end_io and others
2471          * that don't.  so we don't have to unlock the rw_lock if either an
2472          * async dio is going to do it in the future or an end_io after an
2473          * error has already done it.
2474          */
2475         if ((written == -EIOCBQUEUED) || (!ocfs2_iocb_is_rw_locked(iocb))) {
2476                 rw_level = -1;
2477                 unaligned_dio = 0;
2478         }
2479
2480         if (unlikely(written <= 0))
2481                 goto no_sync;
2482
2483         if (((file->f_flags & O_DSYNC) && !direct_io) ||
2484             IS_SYNC(inode) || dropped_dio) {
2485                 ret = filemap_fdatawrite_range(file->f_mapping,
2486                                                iocb->ki_pos - written,
2487                                                iocb->ki_pos - 1);
2488                 if (ret < 0)
2489                         written = ret;
2490
2491                 if (!ret) {
2492                         ret = jbd2_journal_force_commit(osb->journal->j_journal);
2493                         if (ret < 0)
2494                                 written = ret;
2495                 }
2496
2497                 if (!ret)
2498                         ret = filemap_fdatawait_range(file->f_mapping,
2499                                                       iocb->ki_pos - written,
2500                                                       iocb->ki_pos - 1);
2501         }
2502
2503 no_sync:
2504         if (unaligned_dio && ocfs2_iocb_is_unaligned_aio(iocb)) {
2505                 ocfs2_iocb_clear_unaligned_aio(iocb);
2506                 mutex_unlock(&OCFS2_I(inode)->ip_unaligned_aio);
2507         }
2508
2509 out:
2510         if (rw_level != -1)
2511                 ocfs2_rw_unlock(inode, rw_level);
2512
2513 out_mutex:
2514         mutex_unlock(&inode->i_mutex);
2515
2516         if (written)
2517                 ret = written;
2518         return ret;
2519 }
2520
2521 static ssize_t ocfs2_file_splice_read(struct file *in,
2522                                       loff_t *ppos,
2523                                       struct pipe_inode_info *pipe,
2524                                       size_t len,
2525                                       unsigned int flags)
2526 {
2527         int ret = 0, lock_level = 0;
2528         struct inode *inode = file_inode(in);
2529
2530         trace_ocfs2_file_splice_read(inode, in, in->f_path.dentry,
2531                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
2532                         in->f_path.dentry->d_name.len,
2533                         in->f_path.dentry->d_name.name, len);
2534
2535         /*
2536          * See the comment in ocfs2_file_read_iter()
2537          */
2538         ret = ocfs2_inode_lock_atime(inode, in->f_path.mnt, &lock_level);
2539         if (ret < 0) {
2540                 mlog_errno(ret);
2541                 goto bail;
2542         }
2543         ocfs2_inode_unlock(inode, lock_level);
2544
2545         ret = generic_file_splice_read(in, ppos, pipe, len, flags);
2546
2547 bail:
2548         return ret;
2549 }
2550
2551 static ssize_t ocfs2_file_read_iter(struct kiocb *iocb,
2552                                    struct iov_iter *to)
2553 {
2554         int ret = 0, rw_level = -1, lock_level = 0;
2555         struct file *filp = iocb->ki_filp;
2556         struct inode *inode = file_inode(filp);
2557
2558         trace_ocfs2_file_aio_read(inode, filp, filp->f_path.dentry,
2559                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
2560                         filp->f_path.dentry->d_name.len,
2561                         filp->f_path.dentry->d_name.name,
2562                         to->nr_segs);   /* GRRRRR */
2563
2564
2565         if (!inode) {
2566                 ret = -EINVAL;
2567                 mlog_errno(ret);
2568                 goto bail;
2569         }
2570
2571         /*
2572          * buffered reads protect themselves in ->readpage().  O_DIRECT reads
2573          * need locks to protect pending reads from racing with truncate.
2574          */
2575         if (iocb->ki_flags & IOCB_DIRECT) {
2576                 ret = ocfs2_rw_lock(inode, 0);
2577                 if (ret < 0) {
2578                         mlog_errno(ret);
2579                         goto bail;
2580                 }
2581                 rw_level = 0;
2582                 /* communicate with ocfs2_dio_end_io */
2583                 ocfs2_iocb_set_rw_locked(iocb, rw_level);
2584         }
2585
2586         /*
2587          * We're fine letting folks race truncates and extending
2588          * writes with read across the cluster, just like they can
2589          * locally. Hence no rw_lock during read.
2590          *
2591          * Take and drop the meta data lock to update inode fields
2592          * like i_size. This allows the checks down below
2593          * generic_file_aio_read() a chance of actually working.
2594          */
2595         ret = ocfs2_inode_lock_atime(inode, filp->f_path.mnt, &lock_level);
2596         if (ret < 0) {
2597                 mlog_errno(ret);
2598                 goto bail;
2599         }
2600         ocfs2_inode_unlock(inode, lock_level);
2601
2602         ret = generic_file_read_iter(iocb, to);
2603         trace_generic_file_aio_read_ret(ret);
2604
2605         /* buffered aio wouldn't have proper lock coverage today */
2606         BUG_ON(ret == -EIOCBQUEUED && !(iocb->ki_flags & IOCB_DIRECT));
2607
2608         /* see ocfs2_file_write_iter */
2609         if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2610                 rw_level = -1;
2611         }
2612
2613 bail:
2614         if (rw_level != -1)
2615                 ocfs2_rw_unlock(inode, rw_level);
2616
2617         return ret;
2618 }
2619
2620 /* Refer generic_file_llseek_unlocked() */
2621 static loff_t ocfs2_file_llseek(struct file *file, loff_t offset, int whence)
2622 {
2623         struct inode *inode = file->f_mapping->host;
2624         int ret = 0;
2625
2626         mutex_lock(&inode->i_mutex);
2627
2628         switch (whence) {
2629         case SEEK_SET:
2630                 break;
2631         case SEEK_END:
2632                 /* SEEK_END requires the OCFS2 inode lock for the file
2633                  * because it references the file's size.
2634                  */
2635                 ret = ocfs2_inode_lock(inode, NULL, 0);
2636                 if (ret < 0) {
2637                         mlog_errno(ret);
2638                         goto out;
2639                 }
2640                 offset += i_size_read(inode);
2641                 ocfs2_inode_unlock(inode, 0);
2642                 break;
2643         case SEEK_CUR:
2644                 if (offset == 0) {
2645                         offset = file->f_pos;
2646                         goto out;
2647                 }
2648                 offset += file->f_pos;
2649                 break;
2650         case SEEK_DATA:
2651         case SEEK_HOLE:
2652                 ret = ocfs2_seek_data_hole_offset(file, &offset, whence);
2653                 if (ret)
2654                         goto out;
2655                 break;
2656         default:
2657                 ret = -EINVAL;
2658                 goto out;
2659         }
2660
2661         offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
2662
2663 out:
2664         mutex_unlock(&inode->i_mutex);
2665         if (ret)
2666                 return ret;
2667         return offset;
2668 }
2669
2670 const struct inode_operations ocfs2_file_iops = {
2671         .setattr        = ocfs2_setattr,
2672         .getattr        = ocfs2_getattr,
2673         .permission     = ocfs2_permission,
2674         .setxattr       = generic_setxattr,
2675         .getxattr       = generic_getxattr,
2676         .listxattr      = ocfs2_listxattr,
2677         .removexattr    = generic_removexattr,
2678         .fiemap         = ocfs2_fiemap,
2679         .get_acl        = ocfs2_iop_get_acl,
2680         .set_acl        = ocfs2_iop_set_acl,
2681 };
2682
2683 const struct inode_operations ocfs2_special_file_iops = {
2684         .setattr        = ocfs2_setattr,
2685         .getattr        = ocfs2_getattr,
2686         .permission     = ocfs2_permission,
2687         .get_acl        = ocfs2_iop_get_acl,
2688         .set_acl        = ocfs2_iop_set_acl,
2689 };
2690
2691 /*
2692  * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with
2693  * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks!
2694  */
2695 const struct file_operations ocfs2_fops = {
2696         .llseek         = ocfs2_file_llseek,
2697         .mmap           = ocfs2_mmap,
2698         .fsync          = ocfs2_sync_file,
2699         .release        = ocfs2_file_release,
2700         .open           = ocfs2_file_open,
2701         .read_iter      = ocfs2_file_read_iter,
2702         .write_iter     = ocfs2_file_write_iter,
2703         .unlocked_ioctl = ocfs2_ioctl,
2704 #ifdef CONFIG_COMPAT
2705         .compat_ioctl   = ocfs2_compat_ioctl,
2706 #endif
2707         .lock           = ocfs2_lock,
2708         .flock          = ocfs2_flock,
2709         .splice_read    = ocfs2_file_splice_read,
2710         .splice_write   = iter_file_splice_write,
2711         .fallocate      = ocfs2_fallocate,
2712 };
2713
2714 const struct file_operations ocfs2_dops = {
2715         .llseek         = generic_file_llseek,
2716         .read           = generic_read_dir,
2717         .iterate        = ocfs2_readdir,
2718         .fsync          = ocfs2_sync_file,
2719         .release        = ocfs2_dir_release,
2720         .open           = ocfs2_dir_open,
2721         .unlocked_ioctl = ocfs2_ioctl,
2722 #ifdef CONFIG_COMPAT
2723         .compat_ioctl   = ocfs2_compat_ioctl,
2724 #endif
2725         .lock           = ocfs2_lock,
2726         .flock          = ocfs2_flock,
2727 };
2728
2729 /*
2730  * POSIX-lockless variants of our file_operations.
2731  *
2732  * These will be used if the underlying cluster stack does not support
2733  * posix file locking, if the user passes the "localflocks" mount
2734  * option, or if we have a local-only fs.
2735  *
2736  * ocfs2_flock is in here because all stacks handle UNIX file locks,
2737  * so we still want it in the case of no stack support for
2738  * plocks. Internally, it will do the right thing when asked to ignore
2739  * the cluster.
2740  */
2741 const struct file_operations ocfs2_fops_no_plocks = {
2742         .llseek         = ocfs2_file_llseek,
2743         .mmap           = ocfs2_mmap,
2744         .fsync          = ocfs2_sync_file,
2745         .release        = ocfs2_file_release,
2746         .open           = ocfs2_file_open,
2747         .read_iter      = ocfs2_file_read_iter,
2748         .write_iter     = ocfs2_file_write_iter,
2749         .unlocked_ioctl = ocfs2_ioctl,
2750 #ifdef CONFIG_COMPAT
2751         .compat_ioctl   = ocfs2_compat_ioctl,
2752 #endif
2753         .flock          = ocfs2_flock,
2754         .splice_read    = ocfs2_file_splice_read,
2755         .splice_write   = iter_file_splice_write,
2756         .fallocate      = ocfs2_fallocate,
2757 };
2758
2759 const struct file_operations ocfs2_dops_no_plocks = {
2760         .llseek         = generic_file_llseek,
2761         .read           = generic_read_dir,
2762         .iterate        = ocfs2_readdir,
2763         .fsync          = ocfs2_sync_file,
2764         .release        = ocfs2_dir_release,
2765         .open           = ocfs2_dir_open,
2766         .unlocked_ioctl = ocfs2_ioctl,
2767 #ifdef CONFIG_COMPAT
2768         .compat_ioctl   = ocfs2_compat_ioctl,
2769 #endif
2770         .flock          = ocfs2_flock,
2771 };