GNU Linux-libre 4.9.308-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(inode);
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(inode);
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(inode);
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(inode);
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_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_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_SIZE - 1);
802         zero_to = abs_to & (PAGE_SIZE - 1);
803         if (!zero_to)
804                 zero_to = PAGE_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(inode);
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         put_page(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_MASK) + PAGE_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 = setattr_prepare(dentry, 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                 /*
1174                  * Here we should wait dio to finish before inode lock
1175                  * to avoid a deadlock between ocfs2_setattr() and
1176                  * ocfs2_dio_end_io_write()
1177                  */
1178                 inode_dio_wait(inode);
1179
1180                 status = ocfs2_rw_lock(inode, 1);
1181                 if (status < 0) {
1182                         mlog_errno(status);
1183                         goto bail;
1184                 }
1185         }
1186
1187         status = ocfs2_inode_lock(inode, &bh, 1);
1188         if (status < 0) {
1189                 if (status != -ENOENT)
1190                         mlog_errno(status);
1191                 goto bail_unlock_rw;
1192         }
1193         inode_locked = 1;
1194
1195         if (size_change) {
1196                 status = inode_newsize_ok(inode, attr->ia_size);
1197                 if (status)
1198                         goto bail_unlock;
1199
1200                 if (i_size_read(inode) >= attr->ia_size) {
1201                         if (ocfs2_should_order_data(inode)) {
1202                                 status = ocfs2_begin_ordered_truncate(inode,
1203                                                                       attr->ia_size);
1204                                 if (status)
1205                                         goto bail_unlock;
1206                         }
1207                         status = ocfs2_truncate_file(inode, bh, attr->ia_size);
1208                 } else
1209                         status = ocfs2_extend_file(inode, bh, attr->ia_size);
1210                 if (status < 0) {
1211                         if (status != -ENOSPC)
1212                                 mlog_errno(status);
1213                         status = -ENOSPC;
1214                         goto bail_unlock;
1215                 }
1216         }
1217
1218         if ((attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
1219             (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
1220                 /*
1221                  * Gather pointers to quota structures so that allocation /
1222                  * freeing of quota structures happens here and not inside
1223                  * dquot_transfer() where we have problems with lock ordering
1224                  */
1225                 if (attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)
1226                     && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1227                     OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
1228                         transfer_to[USRQUOTA] = dqget(sb, make_kqid_uid(attr->ia_uid));
1229                         if (IS_ERR(transfer_to[USRQUOTA])) {
1230                                 status = PTR_ERR(transfer_to[USRQUOTA]);
1231                                 goto bail_unlock;
1232                         }
1233                 }
1234                 if (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid)
1235                     && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1236                     OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
1237                         transfer_to[GRPQUOTA] = dqget(sb, make_kqid_gid(attr->ia_gid));
1238                         if (IS_ERR(transfer_to[GRPQUOTA])) {
1239                                 status = PTR_ERR(transfer_to[GRPQUOTA]);
1240                                 goto bail_unlock;
1241                         }
1242                 }
1243                 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1244                 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS +
1245                                            2 * ocfs2_quota_trans_credits(sb));
1246                 if (IS_ERR(handle)) {
1247                         status = PTR_ERR(handle);
1248                         mlog_errno(status);
1249                         goto bail_unlock_alloc;
1250                 }
1251                 status = __dquot_transfer(inode, transfer_to);
1252                 if (status < 0)
1253                         goto bail_commit;
1254         } else {
1255                 down_write(&OCFS2_I(inode)->ip_alloc_sem);
1256                 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1257                 if (IS_ERR(handle)) {
1258                         status = PTR_ERR(handle);
1259                         mlog_errno(status);
1260                         goto bail_unlock_alloc;
1261                 }
1262         }
1263
1264         setattr_copy(inode, attr);
1265         mark_inode_dirty(inode);
1266
1267         status = ocfs2_mark_inode_dirty(handle, inode, bh);
1268         if (status < 0)
1269                 mlog_errno(status);
1270
1271 bail_commit:
1272         ocfs2_commit_trans(osb, handle);
1273 bail_unlock_alloc:
1274         up_write(&OCFS2_I(inode)->ip_alloc_sem);
1275 bail_unlock:
1276         if (status) {
1277                 ocfs2_inode_unlock(inode, 1);
1278                 inode_locked = 0;
1279         }
1280 bail_unlock_rw:
1281         if (size_change)
1282                 ocfs2_rw_unlock(inode, 1);
1283 bail:
1284
1285         /* Release quota pointers in case we acquired them */
1286         for (qtype = 0; qtype < OCFS2_MAXQUOTAS; qtype++)
1287                 dqput(transfer_to[qtype]);
1288
1289         if (!status && attr->ia_valid & ATTR_MODE) {
1290                 status = ocfs2_acl_chmod(inode, bh);
1291                 if (status < 0)
1292                         mlog_errno(status);
1293         }
1294         if (inode_locked)
1295                 ocfs2_inode_unlock(inode, 1);
1296
1297         brelse(bh);
1298         return status;
1299 }
1300
1301 int ocfs2_getattr(struct vfsmount *mnt,
1302                   struct dentry *dentry,
1303                   struct kstat *stat)
1304 {
1305         struct inode *inode = d_inode(dentry);
1306         struct super_block *sb = dentry->d_sb;
1307         struct ocfs2_super *osb = sb->s_fs_info;
1308         int err;
1309
1310         err = ocfs2_inode_revalidate(dentry);
1311         if (err) {
1312                 if (err != -ENOENT)
1313                         mlog_errno(err);
1314                 goto bail;
1315         }
1316
1317         generic_fillattr(inode, stat);
1318         /*
1319          * If there is inline data in the inode, the inode will normally not
1320          * have data blocks allocated (it may have an external xattr block).
1321          * Report at least one sector for such files, so tools like tar, rsync,
1322          * others don't incorrectly think the file is completely sparse.
1323          */
1324         if (unlikely(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL))
1325                 stat->blocks += (stat->size + 511)>>9;
1326
1327         /* We set the blksize from the cluster size for performance */
1328         stat->blksize = osb->s_clustersize;
1329
1330 bail:
1331         return err;
1332 }
1333
1334 int ocfs2_permission(struct inode *inode, int mask)
1335 {
1336         int ret;
1337
1338         if (mask & MAY_NOT_BLOCK)
1339                 return -ECHILD;
1340
1341         ret = ocfs2_inode_lock(inode, NULL, 0);
1342         if (ret) {
1343                 if (ret != -ENOENT)
1344                         mlog_errno(ret);
1345                 goto out;
1346         }
1347
1348         ret = generic_permission(inode, mask);
1349
1350         ocfs2_inode_unlock(inode, 0);
1351 out:
1352         return ret;
1353 }
1354
1355 static int __ocfs2_write_remove_suid(struct inode *inode,
1356                                      struct buffer_head *bh)
1357 {
1358         int ret;
1359         handle_t *handle;
1360         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1361         struct ocfs2_dinode *di;
1362
1363         trace_ocfs2_write_remove_suid(
1364                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
1365                         inode->i_mode);
1366
1367         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1368         if (IS_ERR(handle)) {
1369                 ret = PTR_ERR(handle);
1370                 mlog_errno(ret);
1371                 goto out;
1372         }
1373
1374         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
1375                                       OCFS2_JOURNAL_ACCESS_WRITE);
1376         if (ret < 0) {
1377                 mlog_errno(ret);
1378                 goto out_trans;
1379         }
1380
1381         inode->i_mode &= ~S_ISUID;
1382         if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
1383                 inode->i_mode &= ~S_ISGID;
1384
1385         di = (struct ocfs2_dinode *) bh->b_data;
1386         di->i_mode = cpu_to_le16(inode->i_mode);
1387         ocfs2_update_inode_fsync_trans(handle, inode, 0);
1388
1389         ocfs2_journal_dirty(handle, bh);
1390
1391 out_trans:
1392         ocfs2_commit_trans(osb, handle);
1393 out:
1394         return ret;
1395 }
1396
1397 static int ocfs2_write_remove_suid(struct inode *inode)
1398 {
1399         int ret;
1400         struct buffer_head *bh = NULL;
1401
1402         ret = ocfs2_read_inode_block(inode, &bh);
1403         if (ret < 0) {
1404                 mlog_errno(ret);
1405                 goto out;
1406         }
1407
1408         ret =  __ocfs2_write_remove_suid(inode, bh);
1409 out:
1410         brelse(bh);
1411         return ret;
1412 }
1413
1414 /*
1415  * Allocate enough extents to cover the region starting at byte offset
1416  * start for len bytes. Existing extents are skipped, any extents
1417  * added are marked as "unwritten".
1418  */
1419 static int ocfs2_allocate_unwritten_extents(struct inode *inode,
1420                                             u64 start, u64 len)
1421 {
1422         int ret;
1423         u32 cpos, phys_cpos, clusters, alloc_size;
1424         u64 end = start + len;
1425         struct buffer_head *di_bh = NULL;
1426
1427         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1428                 ret = ocfs2_read_inode_block(inode, &di_bh);
1429                 if (ret) {
1430                         mlog_errno(ret);
1431                         goto out;
1432                 }
1433
1434                 /*
1435                  * Nothing to do if the requested reservation range
1436                  * fits within the inode.
1437                  */
1438                 if (ocfs2_size_fits_inline_data(di_bh, end))
1439                         goto out;
1440
1441                 ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1442                 if (ret) {
1443                         mlog_errno(ret);
1444                         goto out;
1445                 }
1446         }
1447
1448         /*
1449          * We consider both start and len to be inclusive.
1450          */
1451         cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
1452         clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len);
1453         clusters -= cpos;
1454
1455         while (clusters) {
1456                 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1457                                          &alloc_size, NULL);
1458                 if (ret) {
1459                         mlog_errno(ret);
1460                         goto out;
1461                 }
1462
1463                 /*
1464                  * Hole or existing extent len can be arbitrary, so
1465                  * cap it to our own allocation request.
1466                  */
1467                 if (alloc_size > clusters)
1468                         alloc_size = clusters;
1469
1470                 if (phys_cpos) {
1471                         /*
1472                          * We already have an allocation at this
1473                          * region so we can safely skip it.
1474                          */
1475                         goto next;
1476                 }
1477
1478                 ret = __ocfs2_extend_allocation(inode, cpos, alloc_size, 1);
1479                 if (ret) {
1480                         if (ret != -ENOSPC)
1481                                 mlog_errno(ret);
1482                         goto out;
1483                 }
1484
1485 next:
1486                 cpos += alloc_size;
1487                 clusters -= alloc_size;
1488         }
1489
1490         ret = 0;
1491 out:
1492
1493         brelse(di_bh);
1494         return ret;
1495 }
1496
1497 /*
1498  * Truncate a byte range, avoiding pages within partial clusters. This
1499  * preserves those pages for the zeroing code to write to.
1500  */
1501 static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start,
1502                                          u64 byte_len)
1503 {
1504         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1505         loff_t start, end;
1506         struct address_space *mapping = inode->i_mapping;
1507
1508         start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start);
1509         end = byte_start + byte_len;
1510         end = end & ~(osb->s_clustersize - 1);
1511
1512         if (start < end) {
1513                 unmap_mapping_range(mapping, start, end - start, 0);
1514                 truncate_inode_pages_range(mapping, start, end - 1);
1515         }
1516 }
1517
1518 /*
1519  * zero out partial blocks of one cluster.
1520  *
1521  * start: file offset where zero starts, will be made upper block aligned.
1522  * len: it will be trimmed to the end of current cluster if "start + len"
1523  *      is bigger than it.
1524  */
1525 static int ocfs2_zeroout_partial_cluster(struct inode *inode,
1526                                         u64 start, u64 len)
1527 {
1528         int ret;
1529         u64 start_block, end_block, nr_blocks;
1530         u64 p_block, offset;
1531         u32 cluster, p_cluster, nr_clusters;
1532         struct super_block *sb = inode->i_sb;
1533         u64 end = ocfs2_align_bytes_to_clusters(sb, start);
1534
1535         if (start + len < end)
1536                 end = start + len;
1537
1538         start_block = ocfs2_blocks_for_bytes(sb, start);
1539         end_block = ocfs2_blocks_for_bytes(sb, end);
1540         nr_blocks = end_block - start_block;
1541         if (!nr_blocks)
1542                 return 0;
1543
1544         cluster = ocfs2_bytes_to_clusters(sb, start);
1545         ret = ocfs2_get_clusters(inode, cluster, &p_cluster,
1546                                 &nr_clusters, NULL);
1547         if (ret)
1548                 return ret;
1549         if (!p_cluster)
1550                 return 0;
1551
1552         offset = start_block - ocfs2_clusters_to_blocks(sb, cluster);
1553         p_block = ocfs2_clusters_to_blocks(sb, p_cluster) + offset;
1554         return sb_issue_zeroout(sb, p_block, nr_blocks, GFP_NOFS);
1555 }
1556
1557 static int ocfs2_zero_partial_clusters(struct inode *inode,
1558                                        u64 start, u64 len)
1559 {
1560         int ret = 0;
1561         u64 tmpend = 0;
1562         u64 end = start + len;
1563         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1564         unsigned int csize = osb->s_clustersize;
1565         handle_t *handle;
1566         loff_t isize = i_size_read(inode);
1567
1568         /*
1569          * The "start" and "end" values are NOT necessarily part of
1570          * the range whose allocation is being deleted. Rather, this
1571          * is what the user passed in with the request. We must zero
1572          * partial clusters here. There's no need to worry about
1573          * physical allocation - the zeroing code knows to skip holes.
1574          */
1575         trace_ocfs2_zero_partial_clusters(
1576                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
1577                 (unsigned long long)start, (unsigned long long)end);
1578
1579         /*
1580          * If both edges are on a cluster boundary then there's no
1581          * zeroing required as the region is part of the allocation to
1582          * be truncated.
1583          */
1584         if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0)
1585                 goto out;
1586
1587         /* No page cache for EOF blocks, issue zero out to disk. */
1588         if (end > isize) {
1589                 /*
1590                  * zeroout eof blocks in last cluster starting from
1591                  * "isize" even "start" > "isize" because it is
1592                  * complicated to zeroout just at "start" as "start"
1593                  * may be not aligned with block size, buffer write
1594                  * would be required to do that, but out of eof buffer
1595                  * write is not supported.
1596                  */
1597                 ret = ocfs2_zeroout_partial_cluster(inode, isize,
1598                                         end - isize);
1599                 if (ret) {
1600                         mlog_errno(ret);
1601                         goto out;
1602                 }
1603                 if (start >= isize)
1604                         goto out;
1605                 end = isize;
1606         }
1607         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1608         if (IS_ERR(handle)) {
1609                 ret = PTR_ERR(handle);
1610                 mlog_errno(ret);
1611                 goto out;
1612         }
1613
1614         /*
1615          * If start is on a cluster boundary and end is somewhere in another
1616          * cluster, we have not COWed the cluster starting at start, unless
1617          * end is also within the same cluster. So, in this case, we skip this
1618          * first call to ocfs2_zero_range_for_truncate() truncate and move on
1619          * to the next one.
1620          */
1621         if ((start & (csize - 1)) != 0) {
1622                 /*
1623                  * We want to get the byte offset of the end of the 1st
1624                  * cluster.
1625                  */
1626                 tmpend = (u64)osb->s_clustersize +
1627                         (start & ~(osb->s_clustersize - 1));
1628                 if (tmpend > end)
1629                         tmpend = end;
1630
1631                 trace_ocfs2_zero_partial_clusters_range1(
1632                         (unsigned long long)start,
1633                         (unsigned long long)tmpend);
1634
1635                 ret = ocfs2_zero_range_for_truncate(inode, handle, start,
1636                                                     tmpend);
1637                 if (ret)
1638                         mlog_errno(ret);
1639         }
1640
1641         if (tmpend < end) {
1642                 /*
1643                  * This may make start and end equal, but the zeroing
1644                  * code will skip any work in that case so there's no
1645                  * need to catch it up here.
1646                  */
1647                 start = end & ~(osb->s_clustersize - 1);
1648
1649                 trace_ocfs2_zero_partial_clusters_range2(
1650                         (unsigned long long)start, (unsigned long long)end);
1651
1652                 ret = ocfs2_zero_range_for_truncate(inode, handle, start, end);
1653                 if (ret)
1654                         mlog_errno(ret);
1655         }
1656         ocfs2_update_inode_fsync_trans(handle, inode, 1);
1657
1658         ocfs2_commit_trans(osb, handle);
1659 out:
1660         return ret;
1661 }
1662
1663 static int ocfs2_find_rec(struct ocfs2_extent_list *el, u32 pos)
1664 {
1665         int i;
1666         struct ocfs2_extent_rec *rec = NULL;
1667
1668         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
1669
1670                 rec = &el->l_recs[i];
1671
1672                 if (le32_to_cpu(rec->e_cpos) < pos)
1673                         break;
1674         }
1675
1676         return i;
1677 }
1678
1679 /*
1680  * Helper to calculate the punching pos and length in one run, we handle the
1681  * following three cases in order:
1682  *
1683  * - remove the entire record
1684  * - remove a partial record
1685  * - no record needs to be removed (hole-punching completed)
1686 */
1687 static void ocfs2_calc_trunc_pos(struct inode *inode,
1688                                  struct ocfs2_extent_list *el,
1689                                  struct ocfs2_extent_rec *rec,
1690                                  u32 trunc_start, u32 *trunc_cpos,
1691                                  u32 *trunc_len, u32 *trunc_end,
1692                                  u64 *blkno, int *done)
1693 {
1694         int ret = 0;
1695         u32 coff, range;
1696
1697         range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
1698
1699         if (le32_to_cpu(rec->e_cpos) >= trunc_start) {
1700                 /*
1701                  * remove an entire extent record.
1702                  */
1703                 *trunc_cpos = le32_to_cpu(rec->e_cpos);
1704                 /*
1705                  * Skip holes if any.
1706                  */
1707                 if (range < *trunc_end)
1708                         *trunc_end = range;
1709                 *trunc_len = *trunc_end - le32_to_cpu(rec->e_cpos);
1710                 *blkno = le64_to_cpu(rec->e_blkno);
1711                 *trunc_end = le32_to_cpu(rec->e_cpos);
1712         } else if (range > trunc_start) {
1713                 /*
1714                  * remove a partial extent record, which means we're
1715                  * removing the last extent record.
1716                  */
1717                 *trunc_cpos = trunc_start;
1718                 /*
1719                  * skip hole if any.
1720                  */
1721                 if (range < *trunc_end)
1722                         *trunc_end = range;
1723                 *trunc_len = *trunc_end - trunc_start;
1724                 coff = trunc_start - le32_to_cpu(rec->e_cpos);
1725                 *blkno = le64_to_cpu(rec->e_blkno) +
1726                                 ocfs2_clusters_to_blocks(inode->i_sb, coff);
1727                 *trunc_end = trunc_start;
1728         } else {
1729                 /*
1730                  * It may have two following possibilities:
1731                  *
1732                  * - last record has been removed
1733                  * - trunc_start was within a hole
1734                  *
1735                  * both two cases mean the completion of hole punching.
1736                  */
1737                 ret = 1;
1738         }
1739
1740         *done = ret;
1741 }
1742
1743 static int ocfs2_remove_inode_range(struct inode *inode,
1744                                     struct buffer_head *di_bh, u64 byte_start,
1745                                     u64 byte_len)
1746 {
1747         int ret = 0, flags = 0, done = 0, i;
1748         u32 trunc_start, trunc_len, trunc_end, trunc_cpos, phys_cpos;
1749         u32 cluster_in_el;
1750         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1751         struct ocfs2_cached_dealloc_ctxt dealloc;
1752         struct address_space *mapping = inode->i_mapping;
1753         struct ocfs2_extent_tree et;
1754         struct ocfs2_path *path = NULL;
1755         struct ocfs2_extent_list *el = NULL;
1756         struct ocfs2_extent_rec *rec = NULL;
1757         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1758         u64 blkno, refcount_loc = le64_to_cpu(di->i_refcount_loc);
1759
1760         ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
1761         ocfs2_init_dealloc_ctxt(&dealloc);
1762
1763         trace_ocfs2_remove_inode_range(
1764                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
1765                         (unsigned long long)byte_start,
1766                         (unsigned long long)byte_len);
1767
1768         if (byte_len == 0)
1769                 return 0;
1770
1771         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1772                 ret = ocfs2_truncate_inline(inode, di_bh, byte_start,
1773                                             byte_start + byte_len, 0);
1774                 if (ret) {
1775                         mlog_errno(ret);
1776                         goto out;
1777                 }
1778                 /*
1779                  * There's no need to get fancy with the page cache
1780                  * truncate of an inline-data inode. We're talking
1781                  * about less than a page here, which will be cached
1782                  * in the dinode buffer anyway.
1783                  */
1784                 unmap_mapping_range(mapping, 0, 0, 0);
1785                 truncate_inode_pages(mapping, 0);
1786                 goto out;
1787         }
1788
1789         /*
1790          * For reflinks, we may need to CoW 2 clusters which might be
1791          * partially zero'd later, if hole's start and end offset were
1792          * within one cluster(means is not exactly aligned to clustersize).
1793          */
1794
1795         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) {
1796
1797                 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start);
1798                 if (ret) {
1799                         mlog_errno(ret);
1800                         goto out;
1801                 }
1802
1803                 ret = ocfs2_cow_file_pos(inode, di_bh, byte_start + byte_len);
1804                 if (ret) {
1805                         mlog_errno(ret);
1806                         goto out;
1807                 }
1808         }
1809
1810         trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start);
1811         trunc_end = (byte_start + byte_len) >> osb->s_clustersize_bits;
1812         cluster_in_el = trunc_end;
1813
1814         ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len);
1815         if (ret) {
1816                 mlog_errno(ret);
1817                 goto out;
1818         }
1819
1820         path = ocfs2_new_path_from_et(&et);
1821         if (!path) {
1822                 ret = -ENOMEM;
1823                 mlog_errno(ret);
1824                 goto out;
1825         }
1826
1827         while (trunc_end > trunc_start) {
1828
1829                 ret = ocfs2_find_path(INODE_CACHE(inode), path,
1830                                       cluster_in_el);
1831                 if (ret) {
1832                         mlog_errno(ret);
1833                         goto out;
1834                 }
1835
1836                 el = path_leaf_el(path);
1837
1838                 i = ocfs2_find_rec(el, trunc_end);
1839                 /*
1840                  * Need to go to previous extent block.
1841                  */
1842                 if (i < 0) {
1843                         if (path->p_tree_depth == 0)
1844                                 break;
1845
1846                         ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb,
1847                                                             path,
1848                                                             &cluster_in_el);
1849                         if (ret) {
1850                                 mlog_errno(ret);
1851                                 goto out;
1852                         }
1853
1854                         /*
1855                          * We've reached the leftmost extent block,
1856                          * it's safe to leave.
1857                          */
1858                         if (cluster_in_el == 0)
1859                                 break;
1860
1861                         /*
1862                          * The 'pos' searched for previous extent block is
1863                          * always one cluster less than actual trunc_end.
1864                          */
1865                         trunc_end = cluster_in_el + 1;
1866
1867                         ocfs2_reinit_path(path, 1);
1868
1869                         continue;
1870
1871                 } else
1872                         rec = &el->l_recs[i];
1873
1874                 ocfs2_calc_trunc_pos(inode, el, rec, trunc_start, &trunc_cpos,
1875                                      &trunc_len, &trunc_end, &blkno, &done);
1876                 if (done)
1877                         break;
1878
1879                 flags = rec->e_flags;
1880                 phys_cpos = ocfs2_blocks_to_clusters(inode->i_sb, blkno);
1881
1882                 ret = ocfs2_remove_btree_range(inode, &et, trunc_cpos,
1883                                                phys_cpos, trunc_len, flags,
1884                                                &dealloc, refcount_loc, false);
1885                 if (ret < 0) {
1886                         mlog_errno(ret);
1887                         goto out;
1888                 }
1889
1890                 cluster_in_el = trunc_end;
1891
1892                 ocfs2_reinit_path(path, 1);
1893         }
1894
1895         ocfs2_truncate_cluster_pages(inode, byte_start, byte_len);
1896
1897 out:
1898         ocfs2_free_path(path);
1899         ocfs2_schedule_truncate_log_flush(osb, 1);
1900         ocfs2_run_deallocs(osb, &dealloc);
1901
1902         return ret;
1903 }
1904
1905 /*
1906  * Parts of this function taken from xfs_change_file_space()
1907  */
1908 static int __ocfs2_change_file_space(struct file *file, struct inode *inode,
1909                                      loff_t f_pos, unsigned int cmd,
1910                                      struct ocfs2_space_resv *sr,
1911                                      int change_size)
1912 {
1913         int ret;
1914         s64 llen;
1915         loff_t size, orig_isize;
1916         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1917         struct buffer_head *di_bh = NULL;
1918         handle_t *handle;
1919         unsigned long long max_off = inode->i_sb->s_maxbytes;
1920
1921         if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
1922                 return -EROFS;
1923
1924         inode_lock(inode);
1925
1926         /*
1927          * This prevents concurrent writes on other nodes
1928          */
1929         ret = ocfs2_rw_lock(inode, 1);
1930         if (ret) {
1931                 mlog_errno(ret);
1932                 goto out;
1933         }
1934
1935         ret = ocfs2_inode_lock(inode, &di_bh, 1);
1936         if (ret) {
1937                 mlog_errno(ret);
1938                 goto out_rw_unlock;
1939         }
1940
1941         if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
1942                 ret = -EPERM;
1943                 goto out_inode_unlock;
1944         }
1945
1946         switch (sr->l_whence) {
1947         case 0: /*SEEK_SET*/
1948                 break;
1949         case 1: /*SEEK_CUR*/
1950                 sr->l_start += f_pos;
1951                 break;
1952         case 2: /*SEEK_END*/
1953                 sr->l_start += i_size_read(inode);
1954                 break;
1955         default:
1956                 ret = -EINVAL;
1957                 goto out_inode_unlock;
1958         }
1959         sr->l_whence = 0;
1960
1961         llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len;
1962
1963         if (sr->l_start < 0
1964             || sr->l_start > max_off
1965             || (sr->l_start + llen) < 0
1966             || (sr->l_start + llen) > max_off) {
1967                 ret = -EINVAL;
1968                 goto out_inode_unlock;
1969         }
1970         size = sr->l_start + sr->l_len;
1971
1972         if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64 ||
1973             cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) {
1974                 if (sr->l_len <= 0) {
1975                         ret = -EINVAL;
1976                         goto out_inode_unlock;
1977                 }
1978         }
1979
1980         if (file && should_remove_suid(file->f_path.dentry)) {
1981                 ret = __ocfs2_write_remove_suid(inode, di_bh);
1982                 if (ret) {
1983                         mlog_errno(ret);
1984                         goto out_inode_unlock;
1985                 }
1986         }
1987
1988         down_write(&OCFS2_I(inode)->ip_alloc_sem);
1989         switch (cmd) {
1990         case OCFS2_IOC_RESVSP:
1991         case OCFS2_IOC_RESVSP64:
1992                 /*
1993                  * This takes unsigned offsets, but the signed ones we
1994                  * pass have been checked against overflow above.
1995                  */
1996                 ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start,
1997                                                        sr->l_len);
1998                 break;
1999         case OCFS2_IOC_UNRESVSP:
2000         case OCFS2_IOC_UNRESVSP64:
2001                 ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start,
2002                                                sr->l_len);
2003                 break;
2004         default:
2005                 ret = -EINVAL;
2006         }
2007
2008         orig_isize = i_size_read(inode);
2009         /* zeroout eof blocks in the cluster. */
2010         if (!ret && change_size && orig_isize < size) {
2011                 ret = ocfs2_zeroout_partial_cluster(inode, orig_isize,
2012                                         size - orig_isize);
2013                 if (!ret)
2014                         i_size_write(inode, size);
2015         }
2016         up_write(&OCFS2_I(inode)->ip_alloc_sem);
2017         if (ret) {
2018                 mlog_errno(ret);
2019                 goto out_inode_unlock;
2020         }
2021
2022         /*
2023          * We update c/mtime for these changes
2024          */
2025         handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
2026         if (IS_ERR(handle)) {
2027                 ret = PTR_ERR(handle);
2028                 mlog_errno(ret);
2029                 goto out_inode_unlock;
2030         }
2031
2032         inode->i_ctime = inode->i_mtime = current_time(inode);
2033         ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
2034         if (ret < 0)
2035                 mlog_errno(ret);
2036
2037         if (file && (file->f_flags & O_SYNC))
2038                 handle->h_sync = 1;
2039
2040         ocfs2_commit_trans(osb, handle);
2041
2042 out_inode_unlock:
2043         brelse(di_bh);
2044         ocfs2_inode_unlock(inode, 1);
2045 out_rw_unlock:
2046         ocfs2_rw_unlock(inode, 1);
2047
2048 out:
2049         inode_unlock(inode);
2050         return ret;
2051 }
2052
2053 int ocfs2_change_file_space(struct file *file, unsigned int cmd,
2054                             struct ocfs2_space_resv *sr)
2055 {
2056         struct inode *inode = file_inode(file);
2057         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2058         int ret;
2059
2060         if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) &&
2061             !ocfs2_writes_unwritten_extents(osb))
2062                 return -ENOTTY;
2063         else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) &&
2064                  !ocfs2_sparse_alloc(osb))
2065                 return -ENOTTY;
2066
2067         if (!S_ISREG(inode->i_mode))
2068                 return -EINVAL;
2069
2070         if (!(file->f_mode & FMODE_WRITE))
2071                 return -EBADF;
2072
2073         ret = mnt_want_write_file(file);
2074         if (ret)
2075                 return ret;
2076         ret = __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0);
2077         mnt_drop_write_file(file);
2078         return ret;
2079 }
2080
2081 static long ocfs2_fallocate(struct file *file, int mode, loff_t offset,
2082                             loff_t len)
2083 {
2084         struct inode *inode = file_inode(file);
2085         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2086         struct ocfs2_space_resv sr;
2087         int change_size = 1;
2088         int cmd = OCFS2_IOC_RESVSP64;
2089
2090         if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2091                 return -EOPNOTSUPP;
2092         if (!ocfs2_writes_unwritten_extents(osb))
2093                 return -EOPNOTSUPP;
2094
2095         if (mode & FALLOC_FL_KEEP_SIZE)
2096                 change_size = 0;
2097
2098         if (mode & FALLOC_FL_PUNCH_HOLE)
2099                 cmd = OCFS2_IOC_UNRESVSP64;
2100
2101         sr.l_whence = 0;
2102         sr.l_start = (s64)offset;
2103         sr.l_len = (s64)len;
2104
2105         return __ocfs2_change_file_space(NULL, inode, offset, cmd, &sr,
2106                                          change_size);
2107 }
2108
2109 int ocfs2_check_range_for_refcount(struct inode *inode, loff_t pos,
2110                                    size_t count)
2111 {
2112         int ret = 0;
2113         unsigned int extent_flags;
2114         u32 cpos, clusters, extent_len, phys_cpos;
2115         struct super_block *sb = inode->i_sb;
2116
2117         if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)) ||
2118             !(OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) ||
2119             OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
2120                 return 0;
2121
2122         cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
2123         clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
2124
2125         while (clusters) {
2126                 ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
2127                                          &extent_flags);
2128                 if (ret < 0) {
2129                         mlog_errno(ret);
2130                         goto out;
2131                 }
2132
2133                 if (phys_cpos && (extent_flags & OCFS2_EXT_REFCOUNTED)) {
2134                         ret = 1;
2135                         break;
2136                 }
2137
2138                 if (extent_len > clusters)
2139                         extent_len = clusters;
2140
2141                 clusters -= extent_len;
2142                 cpos += extent_len;
2143         }
2144 out:
2145         return ret;
2146 }
2147
2148 static int ocfs2_is_io_unaligned(struct inode *inode, size_t count, loff_t pos)
2149 {
2150         int blockmask = inode->i_sb->s_blocksize - 1;
2151         loff_t final_size = pos + count;
2152
2153         if ((pos & blockmask) || (final_size & blockmask))
2154                 return 1;
2155         return 0;
2156 }
2157
2158 static int ocfs2_prepare_inode_for_refcount(struct inode *inode,
2159                                             struct file *file,
2160                                             loff_t pos, size_t count,
2161                                             int *meta_level)
2162 {
2163         int ret;
2164         struct buffer_head *di_bh = NULL;
2165         u32 cpos = pos >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
2166         u32 clusters =
2167                 ocfs2_clusters_for_bytes(inode->i_sb, pos + count) - cpos;
2168
2169         ret = ocfs2_inode_lock(inode, &di_bh, 1);
2170         if (ret) {
2171                 mlog_errno(ret);
2172                 goto out;
2173         }
2174
2175         *meta_level = 1;
2176
2177         ret = ocfs2_refcount_cow(inode, di_bh, cpos, clusters, UINT_MAX);
2178         if (ret)
2179                 mlog_errno(ret);
2180 out:
2181         brelse(di_bh);
2182         return ret;
2183 }
2184
2185 static int ocfs2_prepare_inode_for_write(struct file *file,
2186                                          loff_t pos,
2187                                          size_t count)
2188 {
2189         int ret = 0, meta_level = 0;
2190         struct dentry *dentry = file->f_path.dentry;
2191         struct inode *inode = d_inode(dentry);
2192         loff_t end;
2193
2194         /*
2195          * We start with a read level meta lock and only jump to an ex
2196          * if we need to make modifications here.
2197          */
2198         for(;;) {
2199                 ret = ocfs2_inode_lock(inode, NULL, meta_level);
2200                 if (ret < 0) {
2201                         meta_level = -1;
2202                         mlog_errno(ret);
2203                         goto out;
2204                 }
2205
2206                 /* Clear suid / sgid if necessary. We do this here
2207                  * instead of later in the write path because
2208                  * remove_suid() calls ->setattr without any hint that
2209                  * we may have already done our cluster locking. Since
2210                  * ocfs2_setattr() *must* take cluster locks to
2211                  * proceed, this will lead us to recursively lock the
2212                  * inode. There's also the dinode i_size state which
2213                  * can be lost via setattr during extending writes (we
2214                  * set inode->i_size at the end of a write. */
2215                 if (should_remove_suid(dentry)) {
2216                         if (meta_level == 0) {
2217                                 ocfs2_inode_unlock(inode, meta_level);
2218                                 meta_level = 1;
2219                                 continue;
2220                         }
2221
2222                         ret = ocfs2_write_remove_suid(inode);
2223                         if (ret < 0) {
2224                                 mlog_errno(ret);
2225                                 goto out_unlock;
2226                         }
2227                 }
2228
2229                 end = pos + count;
2230
2231                 ret = ocfs2_check_range_for_refcount(inode, pos, count);
2232                 if (ret == 1) {
2233                         ocfs2_inode_unlock(inode, meta_level);
2234                         meta_level = -1;
2235
2236                         ret = ocfs2_prepare_inode_for_refcount(inode,
2237                                                                file,
2238                                                                pos,
2239                                                                count,
2240                                                                &meta_level);
2241                 }
2242
2243                 if (ret < 0) {
2244                         mlog_errno(ret);
2245                         goto out_unlock;
2246                 }
2247
2248                 break;
2249         }
2250
2251 out_unlock:
2252         trace_ocfs2_prepare_inode_for_write(OCFS2_I(inode)->ip_blkno,
2253                                             pos, count);
2254
2255         if (meta_level >= 0)
2256                 ocfs2_inode_unlock(inode, meta_level);
2257
2258 out:
2259         return ret;
2260 }
2261
2262 static ssize_t ocfs2_file_write_iter(struct kiocb *iocb,
2263                                     struct iov_iter *from)
2264 {
2265         int direct_io, rw_level;
2266         ssize_t written = 0;
2267         ssize_t ret;
2268         size_t count = iov_iter_count(from);
2269         struct file *file = iocb->ki_filp;
2270         struct inode *inode = file_inode(file);
2271         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2272         int full_coherency = !(osb->s_mount_opt &
2273                                OCFS2_MOUNT_COHERENCY_BUFFERED);
2274         void *saved_ki_complete = NULL;
2275         int append_write = ((iocb->ki_pos + count) >=
2276                         i_size_read(inode) ? 1 : 0);
2277
2278         trace_ocfs2_file_aio_write(inode, file, file->f_path.dentry,
2279                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
2280                 file->f_path.dentry->d_name.len,
2281                 file->f_path.dentry->d_name.name,
2282                 (unsigned int)from->nr_segs);   /* GRRRRR */
2283
2284         if (count == 0)
2285                 return 0;
2286
2287         direct_io = iocb->ki_flags & IOCB_DIRECT ? 1 : 0;
2288
2289         inode_lock(inode);
2290
2291         /*
2292          * Concurrent O_DIRECT writes are allowed with
2293          * mount_option "coherency=buffered".
2294          * For append write, we must take rw EX.
2295          */
2296         rw_level = (!direct_io || full_coherency || append_write);
2297
2298         ret = ocfs2_rw_lock(inode, rw_level);
2299         if (ret < 0) {
2300                 mlog_errno(ret);
2301                 goto out_mutex;
2302         }
2303
2304         /*
2305          * O_DIRECT writes with "coherency=full" need to take EX cluster
2306          * inode_lock to guarantee coherency.
2307          */
2308         if (direct_io && full_coherency) {
2309                 /*
2310                  * We need to take and drop the inode lock to force
2311                  * other nodes to drop their caches.  Buffered I/O
2312                  * already does this in write_begin().
2313                  */
2314                 ret = ocfs2_inode_lock(inode, NULL, 1);
2315                 if (ret < 0) {
2316                         mlog_errno(ret);
2317                         goto out;
2318                 }
2319
2320                 ocfs2_inode_unlock(inode, 1);
2321         }
2322
2323         ret = generic_write_checks(iocb, from);
2324         if (ret <= 0) {
2325                 if (ret)
2326                         mlog_errno(ret);
2327                 goto out;
2328         }
2329         count = ret;
2330
2331         ret = ocfs2_prepare_inode_for_write(file, iocb->ki_pos, count);
2332         if (ret < 0) {
2333                 mlog_errno(ret);
2334                 goto out;
2335         }
2336
2337         if (direct_io && !is_sync_kiocb(iocb) &&
2338             ocfs2_is_io_unaligned(inode, count, iocb->ki_pos)) {
2339                 /*
2340                  * Make it a sync io if it's an unaligned aio.
2341                  */
2342                 saved_ki_complete = xchg(&iocb->ki_complete, NULL);
2343         }
2344
2345         /* communicate with ocfs2_dio_end_io */
2346         ocfs2_iocb_set_rw_locked(iocb, rw_level);
2347
2348         written = __generic_file_write_iter(iocb, from);
2349         /* buffered aio wouldn't have proper lock coverage today */
2350         BUG_ON(written == -EIOCBQUEUED && !(iocb->ki_flags & IOCB_DIRECT));
2351
2352         /*
2353          * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
2354          * function pointer which is called when o_direct io completes so that
2355          * it can unlock our rw lock.
2356          * Unfortunately there are error cases which call end_io and others
2357          * that don't.  so we don't have to unlock the rw_lock if either an
2358          * async dio is going to do it in the future or an end_io after an
2359          * error has already done it.
2360          */
2361         if ((written == -EIOCBQUEUED) || (!ocfs2_iocb_is_rw_locked(iocb))) {
2362                 rw_level = -1;
2363         }
2364
2365         if (unlikely(written <= 0))
2366                 goto out;
2367
2368         if (((file->f_flags & O_DSYNC) && !direct_io) ||
2369             IS_SYNC(inode)) {
2370                 ret = filemap_fdatawrite_range(file->f_mapping,
2371                                                iocb->ki_pos - written,
2372                                                iocb->ki_pos - 1);
2373                 if (ret < 0)
2374                         written = ret;
2375
2376                 if (!ret) {
2377                         ret = jbd2_journal_force_commit(osb->journal->j_journal);
2378                         if (ret < 0)
2379                                 written = ret;
2380                 }
2381
2382                 if (!ret)
2383                         ret = filemap_fdatawait_range(file->f_mapping,
2384                                                       iocb->ki_pos - written,
2385                                                       iocb->ki_pos - 1);
2386         }
2387
2388 out:
2389         if (saved_ki_complete)
2390                 xchg(&iocb->ki_complete, saved_ki_complete);
2391
2392         if (rw_level != -1)
2393                 ocfs2_rw_unlock(inode, rw_level);
2394
2395 out_mutex:
2396         inode_unlock(inode);
2397
2398         if (written)
2399                 ret = written;
2400         return ret;
2401 }
2402
2403 static ssize_t ocfs2_file_read_iter(struct kiocb *iocb,
2404                                    struct iov_iter *to)
2405 {
2406         int ret = 0, rw_level = -1, lock_level = 0;
2407         struct file *filp = iocb->ki_filp;
2408         struct inode *inode = file_inode(filp);
2409
2410         trace_ocfs2_file_aio_read(inode, filp, filp->f_path.dentry,
2411                         (unsigned long long)OCFS2_I(inode)->ip_blkno,
2412                         filp->f_path.dentry->d_name.len,
2413                         filp->f_path.dentry->d_name.name,
2414                         to->nr_segs);   /* GRRRRR */
2415
2416
2417         if (!inode) {
2418                 ret = -EINVAL;
2419                 mlog_errno(ret);
2420                 goto bail;
2421         }
2422
2423         /*
2424          * buffered reads protect themselves in ->readpage().  O_DIRECT reads
2425          * need locks to protect pending reads from racing with truncate.
2426          */
2427         if (iocb->ki_flags & IOCB_DIRECT) {
2428                 ret = ocfs2_rw_lock(inode, 0);
2429                 if (ret < 0) {
2430                         mlog_errno(ret);
2431                         goto bail;
2432                 }
2433                 rw_level = 0;
2434                 /* communicate with ocfs2_dio_end_io */
2435                 ocfs2_iocb_set_rw_locked(iocb, rw_level);
2436         }
2437
2438         /*
2439          * We're fine letting folks race truncates and extending
2440          * writes with read across the cluster, just like they can
2441          * locally. Hence no rw_lock during read.
2442          *
2443          * Take and drop the meta data lock to update inode fields
2444          * like i_size. This allows the checks down below
2445          * generic_file_aio_read() a chance of actually working.
2446          */
2447         ret = ocfs2_inode_lock_atime(inode, filp->f_path.mnt, &lock_level);
2448         if (ret < 0) {
2449                 mlog_errno(ret);
2450                 goto bail;
2451         }
2452         ocfs2_inode_unlock(inode, lock_level);
2453
2454         ret = generic_file_read_iter(iocb, to);
2455         trace_generic_file_aio_read_ret(ret);
2456
2457         /* buffered aio wouldn't have proper lock coverage today */
2458         BUG_ON(ret == -EIOCBQUEUED && !(iocb->ki_flags & IOCB_DIRECT));
2459
2460         /* see ocfs2_file_write_iter */
2461         if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2462                 rw_level = -1;
2463         }
2464
2465 bail:
2466         if (rw_level != -1)
2467                 ocfs2_rw_unlock(inode, rw_level);
2468
2469         return ret;
2470 }
2471
2472 /* Refer generic_file_llseek_unlocked() */
2473 static loff_t ocfs2_file_llseek(struct file *file, loff_t offset, int whence)
2474 {
2475         struct inode *inode = file->f_mapping->host;
2476         int ret = 0;
2477
2478         inode_lock(inode);
2479
2480         switch (whence) {
2481         case SEEK_SET:
2482                 break;
2483         case SEEK_END:
2484                 /* SEEK_END requires the OCFS2 inode lock for the file
2485                  * because it references the file's size.
2486                  */
2487                 ret = ocfs2_inode_lock(inode, NULL, 0);
2488                 if (ret < 0) {
2489                         mlog_errno(ret);
2490                         goto out;
2491                 }
2492                 offset += i_size_read(inode);
2493                 ocfs2_inode_unlock(inode, 0);
2494                 break;
2495         case SEEK_CUR:
2496                 if (offset == 0) {
2497                         offset = file->f_pos;
2498                         goto out;
2499                 }
2500                 offset += file->f_pos;
2501                 break;
2502         case SEEK_DATA:
2503         case SEEK_HOLE:
2504                 ret = ocfs2_seek_data_hole_offset(file, &offset, whence);
2505                 if (ret)
2506                         goto out;
2507                 break;
2508         default:
2509                 ret = -EINVAL;
2510                 goto out;
2511         }
2512
2513         offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
2514
2515 out:
2516         inode_unlock(inode);
2517         if (ret)
2518                 return ret;
2519         return offset;
2520 }
2521
2522 const struct inode_operations ocfs2_file_iops = {
2523         .setattr        = ocfs2_setattr,
2524         .getattr        = ocfs2_getattr,
2525         .permission     = ocfs2_permission,
2526         .listxattr      = ocfs2_listxattr,
2527         .fiemap         = ocfs2_fiemap,
2528         .get_acl        = ocfs2_iop_get_acl,
2529         .set_acl        = ocfs2_iop_set_acl,
2530 };
2531
2532 const struct inode_operations ocfs2_special_file_iops = {
2533         .setattr        = ocfs2_setattr,
2534         .getattr        = ocfs2_getattr,
2535         .permission     = ocfs2_permission,
2536         .get_acl        = ocfs2_iop_get_acl,
2537         .set_acl        = ocfs2_iop_set_acl,
2538 };
2539
2540 /*
2541  * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with
2542  * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks!
2543  */
2544 const struct file_operations ocfs2_fops = {
2545         .llseek         = ocfs2_file_llseek,
2546         .mmap           = ocfs2_mmap,
2547         .fsync          = ocfs2_sync_file,
2548         .release        = ocfs2_file_release,
2549         .open           = ocfs2_file_open,
2550         .read_iter      = ocfs2_file_read_iter,
2551         .write_iter     = ocfs2_file_write_iter,
2552         .unlocked_ioctl = ocfs2_ioctl,
2553 #ifdef CONFIG_COMPAT
2554         .compat_ioctl   = ocfs2_compat_ioctl,
2555 #endif
2556         .lock           = ocfs2_lock,
2557         .flock          = ocfs2_flock,
2558         .splice_read    = generic_file_splice_read,
2559         .splice_write   = iter_file_splice_write,
2560         .fallocate      = ocfs2_fallocate,
2561 };
2562
2563 const struct file_operations ocfs2_dops = {
2564         .llseek         = generic_file_llseek,
2565         .read           = generic_read_dir,
2566         .iterate        = ocfs2_readdir,
2567         .fsync          = ocfs2_sync_file,
2568         .release        = ocfs2_dir_release,
2569         .open           = ocfs2_dir_open,
2570         .unlocked_ioctl = ocfs2_ioctl,
2571 #ifdef CONFIG_COMPAT
2572         .compat_ioctl   = ocfs2_compat_ioctl,
2573 #endif
2574         .lock           = ocfs2_lock,
2575         .flock          = ocfs2_flock,
2576 };
2577
2578 /*
2579  * POSIX-lockless variants of our file_operations.
2580  *
2581  * These will be used if the underlying cluster stack does not support
2582  * posix file locking, if the user passes the "localflocks" mount
2583  * option, or if we have a local-only fs.
2584  *
2585  * ocfs2_flock is in here because all stacks handle UNIX file locks,
2586  * so we still want it in the case of no stack support for
2587  * plocks. Internally, it will do the right thing when asked to ignore
2588  * the cluster.
2589  */
2590 const struct file_operations ocfs2_fops_no_plocks = {
2591         .llseek         = ocfs2_file_llseek,
2592         .mmap           = ocfs2_mmap,
2593         .fsync          = ocfs2_sync_file,
2594         .release        = ocfs2_file_release,
2595         .open           = ocfs2_file_open,
2596         .read_iter      = ocfs2_file_read_iter,
2597         .write_iter     = ocfs2_file_write_iter,
2598         .unlocked_ioctl = ocfs2_ioctl,
2599 #ifdef CONFIG_COMPAT
2600         .compat_ioctl   = ocfs2_compat_ioctl,
2601 #endif
2602         .flock          = ocfs2_flock,
2603         .splice_read    = generic_file_splice_read,
2604         .splice_write   = iter_file_splice_write,
2605         .fallocate      = ocfs2_fallocate,
2606 };
2607
2608 const struct file_operations ocfs2_dops_no_plocks = {
2609         .llseek         = generic_file_llseek,
2610         .read           = generic_read_dir,
2611         .iterate        = ocfs2_readdir,
2612         .fsync          = ocfs2_sync_file,
2613         .release        = ocfs2_dir_release,
2614         .open           = ocfs2_dir_open,
2615         .unlocked_ioctl = ocfs2_ioctl,
2616 #ifdef CONFIG_COMPAT
2617         .compat_ioctl   = ocfs2_compat_ioctl,
2618 #endif
2619         .flock          = ocfs2_flock,
2620 };