xref: /linux/fs/ocfs2/file.c (revision b2d0f5d5dc53532e6f07bc546a476a55ebdfe0f3)
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 = file_write_and_wait_range(file, 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 	unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1);
494 	truncate_inode_pages(inode->i_mapping, new_i_size);
495 
496 	if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
497 		status = ocfs2_truncate_inline(inode, di_bh, new_i_size,
498 					       i_size_read(inode), 1);
499 		if (status)
500 			mlog_errno(status);
501 
502 		goto bail_unlock_sem;
503 	}
504 
505 	/* alright, we're going to need to do a full blown alloc size
506 	 * change. Orphan the inode so that recovery can complete the
507 	 * truncate if necessary. This does the task of marking
508 	 * i_size. */
509 	status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
510 	if (status < 0) {
511 		mlog_errno(status);
512 		goto bail_unlock_sem;
513 	}
514 
515 	status = ocfs2_commit_truncate(osb, inode, di_bh);
516 	if (status < 0) {
517 		mlog_errno(status);
518 		goto bail_unlock_sem;
519 	}
520 
521 	/* TODO: orphan dir cleanup here. */
522 bail_unlock_sem:
523 	up_write(&OCFS2_I(inode)->ip_alloc_sem);
524 
525 bail:
526 	if (!status && OCFS2_I(inode)->ip_clusters == 0)
527 		status = ocfs2_try_remove_refcount_tree(inode, di_bh);
528 
529 	return status;
530 }
531 
532 /*
533  * extend file allocation only here.
534  * we'll update all the disk stuff, and oip->alloc_size
535  *
536  * expect stuff to be locked, a transaction started and enough data /
537  * metadata reservations in the contexts.
538  *
539  * Will return -EAGAIN, and a reason if a restart is needed.
540  * If passed in, *reason will always be set, even in error.
541  */
542 int ocfs2_add_inode_data(struct ocfs2_super *osb,
543 			 struct inode *inode,
544 			 u32 *logical_offset,
545 			 u32 clusters_to_add,
546 			 int mark_unwritten,
547 			 struct buffer_head *fe_bh,
548 			 handle_t *handle,
549 			 struct ocfs2_alloc_context *data_ac,
550 			 struct ocfs2_alloc_context *meta_ac,
551 			 enum ocfs2_alloc_restarted *reason_ret)
552 {
553 	int ret;
554 	struct ocfs2_extent_tree et;
555 
556 	ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), fe_bh);
557 	ret = ocfs2_add_clusters_in_btree(handle, &et, logical_offset,
558 					  clusters_to_add, mark_unwritten,
559 					  data_ac, meta_ac, reason_ret);
560 
561 	return ret;
562 }
563 
564 static int __ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
565 				     u32 clusters_to_add, int mark_unwritten)
566 {
567 	int status = 0;
568 	int restart_func = 0;
569 	int credits;
570 	u32 prev_clusters;
571 	struct buffer_head *bh = NULL;
572 	struct ocfs2_dinode *fe = NULL;
573 	handle_t *handle = NULL;
574 	struct ocfs2_alloc_context *data_ac = NULL;
575 	struct ocfs2_alloc_context *meta_ac = NULL;
576 	enum ocfs2_alloc_restarted why = RESTART_NONE;
577 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
578 	struct ocfs2_extent_tree et;
579 	int did_quota = 0;
580 
581 	/*
582 	 * Unwritten extent only exists for file systems which
583 	 * support holes.
584 	 */
585 	BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb));
586 
587 	status = ocfs2_read_inode_block(inode, &bh);
588 	if (status < 0) {
589 		mlog_errno(status);
590 		goto leave;
591 	}
592 	fe = (struct ocfs2_dinode *) bh->b_data;
593 
594 restart_all:
595 	BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
596 
597 	ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), bh);
598 	status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
599 				       &data_ac, &meta_ac);
600 	if (status) {
601 		mlog_errno(status);
602 		goto leave;
603 	}
604 
605 	credits = ocfs2_calc_extend_credits(osb->sb, &fe->id2.i_list);
606 	handle = ocfs2_start_trans(osb, credits);
607 	if (IS_ERR(handle)) {
608 		status = PTR_ERR(handle);
609 		handle = NULL;
610 		mlog_errno(status);
611 		goto leave;
612 	}
613 
614 restarted_transaction:
615 	trace_ocfs2_extend_allocation(
616 		(unsigned long long)OCFS2_I(inode)->ip_blkno,
617 		(unsigned long long)i_size_read(inode),
618 		le32_to_cpu(fe->i_clusters), clusters_to_add,
619 		why, restart_func);
620 
621 	status = dquot_alloc_space_nodirty(inode,
622 			ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
623 	if (status)
624 		goto leave;
625 	did_quota = 1;
626 
627 	/* reserve a write to the file entry early on - that we if we
628 	 * run out of credits in the allocation path, we can still
629 	 * update i_size. */
630 	status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
631 					 OCFS2_JOURNAL_ACCESS_WRITE);
632 	if (status < 0) {
633 		mlog_errno(status);
634 		goto leave;
635 	}
636 
637 	prev_clusters = OCFS2_I(inode)->ip_clusters;
638 
639 	status = ocfs2_add_inode_data(osb,
640 				      inode,
641 				      &logical_start,
642 				      clusters_to_add,
643 				      mark_unwritten,
644 				      bh,
645 				      handle,
646 				      data_ac,
647 				      meta_ac,
648 				      &why);
649 	if ((status < 0) && (status != -EAGAIN)) {
650 		if (status != -ENOSPC)
651 			mlog_errno(status);
652 		goto leave;
653 	}
654 	ocfs2_update_inode_fsync_trans(handle, inode, 1);
655 	ocfs2_journal_dirty(handle, bh);
656 
657 	spin_lock(&OCFS2_I(inode)->ip_lock);
658 	clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
659 	spin_unlock(&OCFS2_I(inode)->ip_lock);
660 	/* Release unused quota reservation */
661 	dquot_free_space(inode,
662 			ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
663 	did_quota = 0;
664 
665 	if (why != RESTART_NONE && clusters_to_add) {
666 		if (why == RESTART_META) {
667 			restart_func = 1;
668 			status = 0;
669 		} else {
670 			BUG_ON(why != RESTART_TRANS);
671 
672 			status = ocfs2_allocate_extend_trans(handle, 1);
673 			if (status < 0) {
674 				/* handle still has to be committed at
675 				 * this point. */
676 				status = -ENOMEM;
677 				mlog_errno(status);
678 				goto leave;
679 			}
680 			goto restarted_transaction;
681 		}
682 	}
683 
684 	trace_ocfs2_extend_allocation_end(OCFS2_I(inode)->ip_blkno,
685 	     le32_to_cpu(fe->i_clusters),
686 	     (unsigned long long)le64_to_cpu(fe->i_size),
687 	     OCFS2_I(inode)->ip_clusters,
688 	     (unsigned long long)i_size_read(inode));
689 
690 leave:
691 	if (status < 0 && did_quota)
692 		dquot_free_space(inode,
693 			ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
694 	if (handle) {
695 		ocfs2_commit_trans(osb, handle);
696 		handle = NULL;
697 	}
698 	if (data_ac) {
699 		ocfs2_free_alloc_context(data_ac);
700 		data_ac = NULL;
701 	}
702 	if (meta_ac) {
703 		ocfs2_free_alloc_context(meta_ac);
704 		meta_ac = NULL;
705 	}
706 	if ((!status) && restart_func) {
707 		restart_func = 0;
708 		goto restart_all;
709 	}
710 	brelse(bh);
711 	bh = NULL;
712 
713 	return status;
714 }
715 
716 /*
717  * While a write will already be ordering the data, a truncate will not.
718  * Thus, we need to explicitly order the zeroed pages.
719  */
720 static handle_t *ocfs2_zero_start_ordered_transaction(struct inode *inode,
721 						struct buffer_head *di_bh)
722 {
723 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
724 	handle_t *handle = NULL;
725 	int ret = 0;
726 
727 	if (!ocfs2_should_order_data(inode))
728 		goto out;
729 
730 	handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
731 	if (IS_ERR(handle)) {
732 		ret = -ENOMEM;
733 		mlog_errno(ret);
734 		goto out;
735 	}
736 
737 	ret = ocfs2_jbd2_file_inode(handle, inode);
738 	if (ret < 0) {
739 		mlog_errno(ret);
740 		goto out;
741 	}
742 
743 	ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
744 				      OCFS2_JOURNAL_ACCESS_WRITE);
745 	if (ret)
746 		mlog_errno(ret);
747 	ocfs2_update_inode_fsync_trans(handle, inode, 1);
748 
749 out:
750 	if (ret) {
751 		if (!IS_ERR(handle))
752 			ocfs2_commit_trans(osb, handle);
753 		handle = ERR_PTR(ret);
754 	}
755 	return handle;
756 }
757 
758 /* Some parts of this taken from generic_cont_expand, which turned out
759  * to be too fragile to do exactly what we need without us having to
760  * worry about recursive locking in ->write_begin() and ->write_end(). */
761 static int ocfs2_write_zero_page(struct inode *inode, u64 abs_from,
762 				 u64 abs_to, struct buffer_head *di_bh)
763 {
764 	struct address_space *mapping = inode->i_mapping;
765 	struct page *page;
766 	unsigned long index = abs_from >> PAGE_SHIFT;
767 	handle_t *handle;
768 	int ret = 0;
769 	unsigned zero_from, zero_to, block_start, block_end;
770 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
771 
772 	BUG_ON(abs_from >= abs_to);
773 	BUG_ON(abs_to > (((u64)index + 1) << PAGE_SHIFT));
774 	BUG_ON(abs_from & (inode->i_blkbits - 1));
775 
776 	handle = ocfs2_zero_start_ordered_transaction(inode, di_bh);
777 	if (IS_ERR(handle)) {
778 		ret = PTR_ERR(handle);
779 		goto out;
780 	}
781 
782 	page = find_or_create_page(mapping, index, GFP_NOFS);
783 	if (!page) {
784 		ret = -ENOMEM;
785 		mlog_errno(ret);
786 		goto out_commit_trans;
787 	}
788 
789 	/* Get the offsets within the page that we want to zero */
790 	zero_from = abs_from & (PAGE_SIZE - 1);
791 	zero_to = abs_to & (PAGE_SIZE - 1);
792 	if (!zero_to)
793 		zero_to = PAGE_SIZE;
794 
795 	trace_ocfs2_write_zero_page(
796 			(unsigned long long)OCFS2_I(inode)->ip_blkno,
797 			(unsigned long long)abs_from,
798 			(unsigned long long)abs_to,
799 			index, zero_from, zero_to);
800 
801 	/* We know that zero_from is block aligned */
802 	for (block_start = zero_from; block_start < zero_to;
803 	     block_start = block_end) {
804 		block_end = block_start + i_blocksize(inode);
805 
806 		/*
807 		 * block_start is block-aligned.  Bump it by one to force
808 		 * __block_write_begin and block_commit_write to zero the
809 		 * whole block.
810 		 */
811 		ret = __block_write_begin(page, block_start + 1, 0,
812 					  ocfs2_get_block);
813 		if (ret < 0) {
814 			mlog_errno(ret);
815 			goto out_unlock;
816 		}
817 
818 
819 		/* must not update i_size! */
820 		ret = block_commit_write(page, block_start + 1,
821 					 block_start + 1);
822 		if (ret < 0)
823 			mlog_errno(ret);
824 		else
825 			ret = 0;
826 	}
827 
828 	/*
829 	 * fs-writeback will release the dirty pages without page lock
830 	 * whose offset are over inode size, the release happens at
831 	 * block_write_full_page().
832 	 */
833 	i_size_write(inode, abs_to);
834 	inode->i_blocks = ocfs2_inode_sector_count(inode);
835 	di->i_size = cpu_to_le64((u64)i_size_read(inode));
836 	inode->i_mtime = inode->i_ctime = current_time(inode);
837 	di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec);
838 	di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
839 	di->i_mtime_nsec = di->i_ctime_nsec;
840 	if (handle) {
841 		ocfs2_journal_dirty(handle, di_bh);
842 		ocfs2_update_inode_fsync_trans(handle, inode, 1);
843 	}
844 
845 out_unlock:
846 	unlock_page(page);
847 	put_page(page);
848 out_commit_trans:
849 	if (handle)
850 		ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
851 out:
852 	return ret;
853 }
854 
855 /*
856  * Find the next range to zero.  We do this in terms of bytes because
857  * that's what ocfs2_zero_extend() wants, and it is dealing with the
858  * pagecache.  We may return multiple extents.
859  *
860  * zero_start and zero_end are ocfs2_zero_extend()s current idea of what
861  * needs to be zeroed.  range_start and range_end return the next zeroing
862  * range.  A subsequent call should pass the previous range_end as its
863  * zero_start.  If range_end is 0, there's nothing to do.
864  *
865  * Unwritten extents are skipped over.  Refcounted extents are CoWd.
866  */
867 static int ocfs2_zero_extend_get_range(struct inode *inode,
868 				       struct buffer_head *di_bh,
869 				       u64 zero_start, u64 zero_end,
870 				       u64 *range_start, u64 *range_end)
871 {
872 	int rc = 0, needs_cow = 0;
873 	u32 p_cpos, zero_clusters = 0;
874 	u32 zero_cpos =
875 		zero_start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
876 	u32 last_cpos = ocfs2_clusters_for_bytes(inode->i_sb, zero_end);
877 	unsigned int num_clusters = 0;
878 	unsigned int ext_flags = 0;
879 
880 	while (zero_cpos < last_cpos) {
881 		rc = ocfs2_get_clusters(inode, zero_cpos, &p_cpos,
882 					&num_clusters, &ext_flags);
883 		if (rc) {
884 			mlog_errno(rc);
885 			goto out;
886 		}
887 
888 		if (p_cpos && !(ext_flags & OCFS2_EXT_UNWRITTEN)) {
889 			zero_clusters = num_clusters;
890 			if (ext_flags & OCFS2_EXT_REFCOUNTED)
891 				needs_cow = 1;
892 			break;
893 		}
894 
895 		zero_cpos += num_clusters;
896 	}
897 	if (!zero_clusters) {
898 		*range_end = 0;
899 		goto out;
900 	}
901 
902 	while ((zero_cpos + zero_clusters) < last_cpos) {
903 		rc = ocfs2_get_clusters(inode, zero_cpos + zero_clusters,
904 					&p_cpos, &num_clusters,
905 					&ext_flags);
906 		if (rc) {
907 			mlog_errno(rc);
908 			goto out;
909 		}
910 
911 		if (!p_cpos || (ext_flags & OCFS2_EXT_UNWRITTEN))
912 			break;
913 		if (ext_flags & OCFS2_EXT_REFCOUNTED)
914 			needs_cow = 1;
915 		zero_clusters += num_clusters;
916 	}
917 	if ((zero_cpos + zero_clusters) > last_cpos)
918 		zero_clusters = last_cpos - zero_cpos;
919 
920 	if (needs_cow) {
921 		rc = ocfs2_refcount_cow(inode, di_bh, zero_cpos,
922 					zero_clusters, UINT_MAX);
923 		if (rc) {
924 			mlog_errno(rc);
925 			goto out;
926 		}
927 	}
928 
929 	*range_start = ocfs2_clusters_to_bytes(inode->i_sb, zero_cpos);
930 	*range_end = ocfs2_clusters_to_bytes(inode->i_sb,
931 					     zero_cpos + zero_clusters);
932 
933 out:
934 	return rc;
935 }
936 
937 /*
938  * Zero one range returned from ocfs2_zero_extend_get_range().  The caller
939  * has made sure that the entire range needs zeroing.
940  */
941 static int ocfs2_zero_extend_range(struct inode *inode, u64 range_start,
942 				   u64 range_end, struct buffer_head *di_bh)
943 {
944 	int rc = 0;
945 	u64 next_pos;
946 	u64 zero_pos = range_start;
947 
948 	trace_ocfs2_zero_extend_range(
949 			(unsigned long long)OCFS2_I(inode)->ip_blkno,
950 			(unsigned long long)range_start,
951 			(unsigned long long)range_end);
952 	BUG_ON(range_start >= range_end);
953 
954 	while (zero_pos < range_end) {
955 		next_pos = (zero_pos & PAGE_MASK) + PAGE_SIZE;
956 		if (next_pos > range_end)
957 			next_pos = range_end;
958 		rc = ocfs2_write_zero_page(inode, zero_pos, next_pos, di_bh);
959 		if (rc < 0) {
960 			mlog_errno(rc);
961 			break;
962 		}
963 		zero_pos = next_pos;
964 
965 		/*
966 		 * Very large extends have the potential to lock up
967 		 * the cpu for extended periods of time.
968 		 */
969 		cond_resched();
970 	}
971 
972 	return rc;
973 }
974 
975 int ocfs2_zero_extend(struct inode *inode, struct buffer_head *di_bh,
976 		      loff_t zero_to_size)
977 {
978 	int ret = 0;
979 	u64 zero_start, range_start = 0, range_end = 0;
980 	struct super_block *sb = inode->i_sb;
981 
982 	zero_start = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
983 	trace_ocfs2_zero_extend((unsigned long long)OCFS2_I(inode)->ip_blkno,
984 				(unsigned long long)zero_start,
985 				(unsigned long long)i_size_read(inode));
986 	while (zero_start < zero_to_size) {
987 		ret = ocfs2_zero_extend_get_range(inode, di_bh, zero_start,
988 						  zero_to_size,
989 						  &range_start,
990 						  &range_end);
991 		if (ret) {
992 			mlog_errno(ret);
993 			break;
994 		}
995 		if (!range_end)
996 			break;
997 		/* Trim the ends */
998 		if (range_start < zero_start)
999 			range_start = zero_start;
1000 		if (range_end > zero_to_size)
1001 			range_end = zero_to_size;
1002 
1003 		ret = ocfs2_zero_extend_range(inode, range_start,
1004 					      range_end, di_bh);
1005 		if (ret) {
1006 			mlog_errno(ret);
1007 			break;
1008 		}
1009 		zero_start = range_end;
1010 	}
1011 
1012 	return ret;
1013 }
1014 
1015 int ocfs2_extend_no_holes(struct inode *inode, struct buffer_head *di_bh,
1016 			  u64 new_i_size, u64 zero_to)
1017 {
1018 	int ret;
1019 	u32 clusters_to_add;
1020 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
1021 
1022 	/*
1023 	 * Only quota files call this without a bh, and they can't be
1024 	 * refcounted.
1025 	 */
1026 	BUG_ON(!di_bh && ocfs2_is_refcount_inode(inode));
1027 	BUG_ON(!di_bh && !(oi->ip_flags & OCFS2_INODE_SYSTEM_FILE));
1028 
1029 	clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size);
1030 	if (clusters_to_add < oi->ip_clusters)
1031 		clusters_to_add = 0;
1032 	else
1033 		clusters_to_add -= oi->ip_clusters;
1034 
1035 	if (clusters_to_add) {
1036 		ret = __ocfs2_extend_allocation(inode, oi->ip_clusters,
1037 						clusters_to_add, 0);
1038 		if (ret) {
1039 			mlog_errno(ret);
1040 			goto out;
1041 		}
1042 	}
1043 
1044 	/*
1045 	 * Call this even if we don't add any clusters to the tree. We
1046 	 * still need to zero the area between the old i_size and the
1047 	 * new i_size.
1048 	 */
1049 	ret = ocfs2_zero_extend(inode, di_bh, zero_to);
1050 	if (ret < 0)
1051 		mlog_errno(ret);
1052 
1053 out:
1054 	return ret;
1055 }
1056 
1057 static int ocfs2_extend_file(struct inode *inode,
1058 			     struct buffer_head *di_bh,
1059 			     u64 new_i_size)
1060 {
1061 	int ret = 0;
1062 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
1063 
1064 	BUG_ON(!di_bh);
1065 
1066 	/* setattr sometimes calls us like this. */
1067 	if (new_i_size == 0)
1068 		goto out;
1069 
1070 	if (i_size_read(inode) == new_i_size)
1071 		goto out;
1072 	BUG_ON(new_i_size < i_size_read(inode));
1073 
1074 	/*
1075 	 * The alloc sem blocks people in read/write from reading our
1076 	 * allocation until we're done changing it. We depend on
1077 	 * i_mutex to block other extend/truncate calls while we're
1078 	 * here.  We even have to hold it for sparse files because there
1079 	 * might be some tail zeroing.
1080 	 */
1081 	down_write(&oi->ip_alloc_sem);
1082 
1083 	if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1084 		/*
1085 		 * We can optimize small extends by keeping the inodes
1086 		 * inline data.
1087 		 */
1088 		if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) {
1089 			up_write(&oi->ip_alloc_sem);
1090 			goto out_update_size;
1091 		}
1092 
1093 		ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1094 		if (ret) {
1095 			up_write(&oi->ip_alloc_sem);
1096 			mlog_errno(ret);
1097 			goto out;
1098 		}
1099 	}
1100 
1101 	if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
1102 		ret = ocfs2_zero_extend(inode, di_bh, new_i_size);
1103 	else
1104 		ret = ocfs2_extend_no_holes(inode, di_bh, new_i_size,
1105 					    new_i_size);
1106 
1107 	up_write(&oi->ip_alloc_sem);
1108 
1109 	if (ret < 0) {
1110 		mlog_errno(ret);
1111 		goto out;
1112 	}
1113 
1114 out_update_size:
1115 	ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
1116 	if (ret < 0)
1117 		mlog_errno(ret);
1118 
1119 out:
1120 	return ret;
1121 }
1122 
1123 int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
1124 {
1125 	int status = 0, size_change;
1126 	int inode_locked = 0;
1127 	struct inode *inode = d_inode(dentry);
1128 	struct super_block *sb = inode->i_sb;
1129 	struct ocfs2_super *osb = OCFS2_SB(sb);
1130 	struct buffer_head *bh = NULL;
1131 	handle_t *handle = NULL;
1132 	struct dquot *transfer_to[MAXQUOTAS] = { };
1133 	int qtype;
1134 	int had_lock;
1135 	struct ocfs2_lock_holder oh;
1136 
1137 	trace_ocfs2_setattr(inode, dentry,
1138 			    (unsigned long long)OCFS2_I(inode)->ip_blkno,
1139 			    dentry->d_name.len, dentry->d_name.name,
1140 			    attr->ia_valid, attr->ia_mode,
1141 			    from_kuid(&init_user_ns, attr->ia_uid),
1142 			    from_kgid(&init_user_ns, attr->ia_gid));
1143 
1144 	/* ensuring we don't even attempt to truncate a symlink */
1145 	if (S_ISLNK(inode->i_mode))
1146 		attr->ia_valid &= ~ATTR_SIZE;
1147 
1148 #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
1149 			   | ATTR_GID | ATTR_UID | ATTR_MODE)
1150 	if (!(attr->ia_valid & OCFS2_VALID_ATTRS))
1151 		return 0;
1152 
1153 	status = setattr_prepare(dentry, attr);
1154 	if (status)
1155 		return status;
1156 
1157 	if (is_quota_modification(inode, attr)) {
1158 		status = dquot_initialize(inode);
1159 		if (status)
1160 			return status;
1161 	}
1162 	size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
1163 	if (size_change) {
1164 		status = ocfs2_rw_lock(inode, 1);
1165 		if (status < 0) {
1166 			mlog_errno(status);
1167 			goto bail;
1168 		}
1169 	}
1170 
1171 	had_lock = ocfs2_inode_lock_tracker(inode, &bh, 1, &oh);
1172 	if (had_lock < 0) {
1173 		status = had_lock;
1174 		goto bail_unlock_rw;
1175 	} else if (had_lock) {
1176 		/*
1177 		 * As far as we know, ocfs2_setattr() could only be the first
1178 		 * VFS entry point in the call chain of recursive cluster
1179 		 * locking issue.
1180 		 *
1181 		 * For instance:
1182 		 * chmod_common()
1183 		 *  notify_change()
1184 		 *   ocfs2_setattr()
1185 		 *    posix_acl_chmod()
1186 		 *     ocfs2_iop_get_acl()
1187 		 *
1188 		 * But, we're not 100% sure if it's always true, because the
1189 		 * ordering of the VFS entry points in the call chain is out
1190 		 * of our control. So, we'd better dump the stack here to
1191 		 * catch the other cases of recursive locking.
1192 		 */
1193 		mlog(ML_ERROR, "Another case of recursive locking:\n");
1194 		dump_stack();
1195 	}
1196 	inode_locked = 1;
1197 
1198 	if (size_change) {
1199 		status = inode_newsize_ok(inode, attr->ia_size);
1200 		if (status)
1201 			goto bail_unlock;
1202 
1203 		inode_dio_wait(inode);
1204 
1205 		if (i_size_read(inode) >= attr->ia_size) {
1206 			if (ocfs2_should_order_data(inode)) {
1207 				status = ocfs2_begin_ordered_truncate(inode,
1208 								      attr->ia_size);
1209 				if (status)
1210 					goto bail_unlock;
1211 			}
1212 			status = ocfs2_truncate_file(inode, bh, attr->ia_size);
1213 		} else
1214 			status = ocfs2_extend_file(inode, bh, attr->ia_size);
1215 		if (status < 0) {
1216 			if (status != -ENOSPC)
1217 				mlog_errno(status);
1218 			status = -ENOSPC;
1219 			goto bail_unlock;
1220 		}
1221 	}
1222 
1223 	if ((attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
1224 	    (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
1225 		/*
1226 		 * Gather pointers to quota structures so that allocation /
1227 		 * freeing of quota structures happens here and not inside
1228 		 * dquot_transfer() where we have problems with lock ordering
1229 		 */
1230 		if (attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)
1231 		    && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1232 		    OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
1233 			transfer_to[USRQUOTA] = dqget(sb, make_kqid_uid(attr->ia_uid));
1234 			if (IS_ERR(transfer_to[USRQUOTA])) {
1235 				status = PTR_ERR(transfer_to[USRQUOTA]);
1236 				goto bail_unlock;
1237 			}
1238 		}
1239 		if (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid)
1240 		    && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
1241 		    OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
1242 			transfer_to[GRPQUOTA] = dqget(sb, make_kqid_gid(attr->ia_gid));
1243 			if (IS_ERR(transfer_to[GRPQUOTA])) {
1244 				status = PTR_ERR(transfer_to[GRPQUOTA]);
1245 				goto bail_unlock;
1246 			}
1247 		}
1248 		handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS +
1249 					   2 * ocfs2_quota_trans_credits(sb));
1250 		if (IS_ERR(handle)) {
1251 			status = PTR_ERR(handle);
1252 			mlog_errno(status);
1253 			goto bail_unlock;
1254 		}
1255 		status = __dquot_transfer(inode, transfer_to);
1256 		if (status < 0)
1257 			goto bail_commit;
1258 	} else {
1259 		handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1260 		if (IS_ERR(handle)) {
1261 			status = PTR_ERR(handle);
1262 			mlog_errno(status);
1263 			goto bail_unlock;
1264 		}
1265 	}
1266 
1267 	setattr_copy(inode, attr);
1268 	mark_inode_dirty(inode);
1269 
1270 	status = ocfs2_mark_inode_dirty(handle, inode, bh);
1271 	if (status < 0)
1272 		mlog_errno(status);
1273 
1274 bail_commit:
1275 	ocfs2_commit_trans(osb, handle);
1276 bail_unlock:
1277 	if (status && inode_locked) {
1278 		ocfs2_inode_unlock_tracker(inode, 1, &oh, had_lock);
1279 		inode_locked = 0;
1280 	}
1281 bail_unlock_rw:
1282 	if (size_change)
1283 		ocfs2_rw_unlock(inode, 1);
1284 bail:
1285 
1286 	/* Release quota pointers in case we acquired them */
1287 	for (qtype = 0; qtype < OCFS2_MAXQUOTAS; qtype++)
1288 		dqput(transfer_to[qtype]);
1289 
1290 	if (!status && attr->ia_valid & ATTR_MODE) {
1291 		status = ocfs2_acl_chmod(inode, bh);
1292 		if (status < 0)
1293 			mlog_errno(status);
1294 	}
1295 	if (inode_locked)
1296 		ocfs2_inode_unlock_tracker(inode, 1, &oh, had_lock);
1297 
1298 	brelse(bh);
1299 	return status;
1300 }
1301 
1302 int ocfs2_getattr(const struct path *path, struct kstat *stat,
1303 		  u32 request_mask, unsigned int flags)
1304 {
1305 	struct inode *inode = d_inode(path->dentry);
1306 	struct super_block *sb = path->dentry->d_sb;
1307 	struct ocfs2_super *osb = sb->s_fs_info;
1308 	int err;
1309 
1310 	err = ocfs2_inode_revalidate(path->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, had_lock;
1337 	struct ocfs2_lock_holder oh;
1338 
1339 	if (mask & MAY_NOT_BLOCK)
1340 		return -ECHILD;
1341 
1342 	had_lock = ocfs2_inode_lock_tracker(inode, NULL, 0, &oh);
1343 	if (had_lock < 0) {
1344 		ret = had_lock;
1345 		goto out;
1346 	} else if (had_lock) {
1347 		/* See comments in ocfs2_setattr() for details.
1348 		 * The call chain of this case could be:
1349 		 * do_sys_open()
1350 		 *  may_open()
1351 		 *   inode_permission()
1352 		 *    ocfs2_permission()
1353 		 *     ocfs2_iop_get_acl()
1354 		 */
1355 		mlog(ML_ERROR, "Another case of recursive locking:\n");
1356 		dump_stack();
1357 	}
1358 
1359 	ret = generic_permission(inode, mask);
1360 
1361 	ocfs2_inode_unlock_tracker(inode, 0, &oh, had_lock);
1362 out:
1363 	return ret;
1364 }
1365 
1366 static int __ocfs2_write_remove_suid(struct inode *inode,
1367 				     struct buffer_head *bh)
1368 {
1369 	int ret;
1370 	handle_t *handle;
1371 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1372 	struct ocfs2_dinode *di;
1373 
1374 	trace_ocfs2_write_remove_suid(
1375 			(unsigned long long)OCFS2_I(inode)->ip_blkno,
1376 			inode->i_mode);
1377 
1378 	handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1379 	if (IS_ERR(handle)) {
1380 		ret = PTR_ERR(handle);
1381 		mlog_errno(ret);
1382 		goto out;
1383 	}
1384 
1385 	ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
1386 				      OCFS2_JOURNAL_ACCESS_WRITE);
1387 	if (ret < 0) {
1388 		mlog_errno(ret);
1389 		goto out_trans;
1390 	}
1391 
1392 	inode->i_mode &= ~S_ISUID;
1393 	if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
1394 		inode->i_mode &= ~S_ISGID;
1395 
1396 	di = (struct ocfs2_dinode *) bh->b_data;
1397 	di->i_mode = cpu_to_le16(inode->i_mode);
1398 	ocfs2_update_inode_fsync_trans(handle, inode, 0);
1399 
1400 	ocfs2_journal_dirty(handle, bh);
1401 
1402 out_trans:
1403 	ocfs2_commit_trans(osb, handle);
1404 out:
1405 	return ret;
1406 }
1407 
1408 static int ocfs2_write_remove_suid(struct inode *inode)
1409 {
1410 	int ret;
1411 	struct buffer_head *bh = NULL;
1412 
1413 	ret = ocfs2_read_inode_block(inode, &bh);
1414 	if (ret < 0) {
1415 		mlog_errno(ret);
1416 		goto out;
1417 	}
1418 
1419 	ret =  __ocfs2_write_remove_suid(inode, bh);
1420 out:
1421 	brelse(bh);
1422 	return ret;
1423 }
1424 
1425 /*
1426  * Allocate enough extents to cover the region starting at byte offset
1427  * start for len bytes. Existing extents are skipped, any extents
1428  * added are marked as "unwritten".
1429  */
1430 static int ocfs2_allocate_unwritten_extents(struct inode *inode,
1431 					    u64 start, u64 len)
1432 {
1433 	int ret;
1434 	u32 cpos, phys_cpos, clusters, alloc_size;
1435 	u64 end = start + len;
1436 	struct buffer_head *di_bh = NULL;
1437 
1438 	if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1439 		ret = ocfs2_read_inode_block(inode, &di_bh);
1440 		if (ret) {
1441 			mlog_errno(ret);
1442 			goto out;
1443 		}
1444 
1445 		/*
1446 		 * Nothing to do if the requested reservation range
1447 		 * fits within the inode.
1448 		 */
1449 		if (ocfs2_size_fits_inline_data(di_bh, end))
1450 			goto out;
1451 
1452 		ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
1453 		if (ret) {
1454 			mlog_errno(ret);
1455 			goto out;
1456 		}
1457 	}
1458 
1459 	/*
1460 	 * We consider both start and len to be inclusive.
1461 	 */
1462 	cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
1463 	clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len);
1464 	clusters -= cpos;
1465 
1466 	while (clusters) {
1467 		ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
1468 					 &alloc_size, NULL);
1469 		if (ret) {
1470 			mlog_errno(ret);
1471 			goto out;
1472 		}
1473 
1474 		/*
1475 		 * Hole or existing extent len can be arbitrary, so
1476 		 * cap it to our own allocation request.
1477 		 */
1478 		if (alloc_size > clusters)
1479 			alloc_size = clusters;
1480 
1481 		if (phys_cpos) {
1482 			/*
1483 			 * We already have an allocation at this
1484 			 * region so we can safely skip it.
1485 			 */
1486 			goto next;
1487 		}
1488 
1489 		ret = __ocfs2_extend_allocation(inode, cpos, alloc_size, 1);
1490 		if (ret) {
1491 			if (ret != -ENOSPC)
1492 				mlog_errno(ret);
1493 			goto out;
1494 		}
1495 
1496 next:
1497 		cpos += alloc_size;
1498 		clusters -= alloc_size;
1499 	}
1500 
1501 	ret = 0;
1502 out:
1503 
1504 	brelse(di_bh);
1505 	return ret;
1506 }
1507 
1508 /*
1509  * Truncate a byte range, avoiding pages within partial clusters. This
1510  * preserves those pages for the zeroing code to write to.
1511  */
1512 static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start,
1513 					 u64 byte_len)
1514 {
1515 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1516 	loff_t start, end;
1517 	struct address_space *mapping = inode->i_mapping;
1518 
1519 	start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start);
1520 	end = byte_start + byte_len;
1521 	end = end & ~(osb->s_clustersize - 1);
1522 
1523 	if (start < end) {
1524 		unmap_mapping_range(mapping, start, end - start, 0);
1525 		truncate_inode_pages_range(mapping, start, end - 1);
1526 	}
1527 }
1528 
1529 static int ocfs2_zero_partial_clusters(struct inode *inode,
1530 				       u64 start, u64 len)
1531 {
1532 	int ret = 0;
1533 	u64 tmpend = 0;
1534 	u64 end = start + len;
1535 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1536 	unsigned int csize = osb->s_clustersize;
1537 	handle_t *handle;
1538 
1539 	/*
1540 	 * The "start" and "end" values are NOT necessarily part of
1541 	 * the range whose allocation is being deleted. Rather, this
1542 	 * is what the user passed in with the request. We must zero
1543 	 * partial clusters here. There's no need to worry about
1544 	 * physical allocation - the zeroing code knows to skip holes.
1545 	 */
1546 	trace_ocfs2_zero_partial_clusters(
1547 		(unsigned long long)OCFS2_I(inode)->ip_blkno,
1548 		(unsigned long long)start, (unsigned long long)end);
1549 
1550 	/*
1551 	 * If both edges are on a cluster boundary then there's no
1552 	 * zeroing required as the region is part of the allocation to
1553 	 * be truncated.
1554 	 */
1555 	if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0)
1556 		goto out;
1557 
1558 	handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1559 	if (IS_ERR(handle)) {
1560 		ret = PTR_ERR(handle);
1561 		mlog_errno(ret);
1562 		goto out;
1563 	}
1564 
1565 	/*
1566 	 * If start is on a cluster boundary and end is somewhere in another
1567 	 * cluster, we have not COWed the cluster starting at start, unless
1568 	 * end is also within the same cluster. So, in this case, we skip this
1569 	 * first call to ocfs2_zero_range_for_truncate() truncate and move on
1570 	 * to the next one.
1571 	 */
1572 	if ((start & (csize - 1)) != 0) {
1573 		/*
1574 		 * We want to get the byte offset of the end of the 1st
1575 		 * cluster.
1576 		 */
1577 		tmpend = (u64)osb->s_clustersize +
1578 			(start & ~(osb->s_clustersize - 1));
1579 		if (tmpend > end)
1580 			tmpend = end;
1581 
1582 		trace_ocfs2_zero_partial_clusters_range1(
1583 			(unsigned long long)start,
1584 			(unsigned long long)tmpend);
1585 
1586 		ret = ocfs2_zero_range_for_truncate(inode, handle, start,
1587 						    tmpend);
1588 		if (ret)
1589 			mlog_errno(ret);
1590 	}
1591 
1592 	if (tmpend < end) {
1593 		/*
1594 		 * This may make start and end equal, but the zeroing
1595 		 * code will skip any work in that case so there's no
1596 		 * need to catch it up here.
1597 		 */
1598 		start = end & ~(osb->s_clustersize - 1);
1599 
1600 		trace_ocfs2_zero_partial_clusters_range2(
1601 			(unsigned long long)start, (unsigned long long)end);
1602 
1603 		ret = ocfs2_zero_range_for_truncate(inode, handle, start, end);
1604 		if (ret)
1605 			mlog_errno(ret);
1606 	}
1607 	ocfs2_update_inode_fsync_trans(handle, inode, 1);
1608 
1609 	ocfs2_commit_trans(osb, handle);
1610 out:
1611 	return ret;
1612 }
1613 
1614 static int ocfs2_find_rec(struct ocfs2_extent_list *el, u32 pos)
1615 {
1616 	int i;
1617 	struct ocfs2_extent_rec *rec = NULL;
1618 
1619 	for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
1620 
1621 		rec = &el->l_recs[i];
1622 
1623 		if (le32_to_cpu(rec->e_cpos) < pos)
1624 			break;
1625 	}
1626 
1627 	return i;
1628 }
1629 
1630 /*
1631  * Helper to calculate the punching pos and length in one run, we handle the
1632  * following three cases in order:
1633  *
1634  * - remove the entire record
1635  * - remove a partial record
1636  * - no record needs to be removed (hole-punching completed)
1637 */
1638 static void ocfs2_calc_trunc_pos(struct inode *inode,
1639 				 struct ocfs2_extent_list *el,
1640 				 struct ocfs2_extent_rec *rec,
1641 				 u32 trunc_start, u32 *trunc_cpos,
1642 				 u32 *trunc_len, u32 *trunc_end,
1643 				 u64 *blkno, int *done)
1644 {
1645 	int ret = 0;
1646 	u32 coff, range;
1647 
1648 	range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
1649 
1650 	if (le32_to_cpu(rec->e_cpos) >= trunc_start) {
1651 		/*
1652 		 * remove an entire extent record.
1653 		 */
1654 		*trunc_cpos = le32_to_cpu(rec->e_cpos);
1655 		/*
1656 		 * Skip holes if any.
1657 		 */
1658 		if (range < *trunc_end)
1659 			*trunc_end = range;
1660 		*trunc_len = *trunc_end - le32_to_cpu(rec->e_cpos);
1661 		*blkno = le64_to_cpu(rec->e_blkno);
1662 		*trunc_end = le32_to_cpu(rec->e_cpos);
1663 	} else if (range > trunc_start) {
1664 		/*
1665 		 * remove a partial extent record, which means we're
1666 		 * removing the last extent record.
1667 		 */
1668 		*trunc_cpos = trunc_start;
1669 		/*
1670 		 * skip hole if any.
1671 		 */
1672 		if (range < *trunc_end)
1673 			*trunc_end = range;
1674 		*trunc_len = *trunc_end - trunc_start;
1675 		coff = trunc_start - le32_to_cpu(rec->e_cpos);
1676 		*blkno = le64_to_cpu(rec->e_blkno) +
1677 				ocfs2_clusters_to_blocks(inode->i_sb, coff);
1678 		*trunc_end = trunc_start;
1679 	} else {
1680 		/*
1681 		 * It may have two following possibilities:
1682 		 *
1683 		 * - last record has been removed
1684 		 * - trunc_start was within a hole
1685 		 *
1686 		 * both two cases mean the completion of hole punching.
1687 		 */
1688 		ret = 1;
1689 	}
1690 
1691 	*done = ret;
1692 }
1693 
1694 int ocfs2_remove_inode_range(struct inode *inode,
1695 			     struct buffer_head *di_bh, u64 byte_start,
1696 			     u64 byte_len)
1697 {
1698 	int ret = 0, flags = 0, done = 0, i;
1699 	u32 trunc_start, trunc_len, trunc_end, trunc_cpos, phys_cpos;
1700 	u32 cluster_in_el;
1701 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1702 	struct ocfs2_cached_dealloc_ctxt dealloc;
1703 	struct address_space *mapping = inode->i_mapping;
1704 	struct ocfs2_extent_tree et;
1705 	struct ocfs2_path *path = NULL;
1706 	struct ocfs2_extent_list *el = NULL;
1707 	struct ocfs2_extent_rec *rec = NULL;
1708 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1709 	u64 blkno, refcount_loc = le64_to_cpu(di->i_refcount_loc);
1710 
1711 	ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
1712 	ocfs2_init_dealloc_ctxt(&dealloc);
1713 
1714 	trace_ocfs2_remove_inode_range(
1715 			(unsigned long long)OCFS2_I(inode)->ip_blkno,
1716 			(unsigned long long)byte_start,
1717 			(unsigned long long)byte_len);
1718 
1719 	if (byte_len == 0)
1720 		return 0;
1721 
1722 	if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1723 		ret = ocfs2_truncate_inline(inode, di_bh, byte_start,
1724 					    byte_start + byte_len, 0);
1725 		if (ret) {
1726 			mlog_errno(ret);
1727 			goto out;
1728 		}
1729 		/*
1730 		 * There's no need to get fancy with the page cache
1731 		 * truncate of an inline-data inode. We're talking
1732 		 * about less than a page here, which will be cached
1733 		 * in the dinode buffer anyway.
1734 		 */
1735 		unmap_mapping_range(mapping, 0, 0, 0);
1736 		truncate_inode_pages(mapping, 0);
1737 		goto out;
1738 	}
1739 
1740 	/*
1741 	 * For reflinks, we may need to CoW 2 clusters which might be
1742 	 * partially zero'd later, if hole's start and end offset were
1743 	 * within one cluster(means is not exactly aligned to clustersize).
1744 	 */
1745 
1746 	if (ocfs2_is_refcount_inode(inode)) {
1747 		ret = ocfs2_cow_file_pos(inode, di_bh, byte_start);
1748 		if (ret) {
1749 			mlog_errno(ret);
1750 			goto out;
1751 		}
1752 
1753 		ret = ocfs2_cow_file_pos(inode, di_bh, byte_start + byte_len);
1754 		if (ret) {
1755 			mlog_errno(ret);
1756 			goto out;
1757 		}
1758 	}
1759 
1760 	trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start);
1761 	trunc_end = (byte_start + byte_len) >> osb->s_clustersize_bits;
1762 	cluster_in_el = trunc_end;
1763 
1764 	ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len);
1765 	if (ret) {
1766 		mlog_errno(ret);
1767 		goto out;
1768 	}
1769 
1770 	path = ocfs2_new_path_from_et(&et);
1771 	if (!path) {
1772 		ret = -ENOMEM;
1773 		mlog_errno(ret);
1774 		goto out;
1775 	}
1776 
1777 	while (trunc_end > trunc_start) {
1778 
1779 		ret = ocfs2_find_path(INODE_CACHE(inode), path,
1780 				      cluster_in_el);
1781 		if (ret) {
1782 			mlog_errno(ret);
1783 			goto out;
1784 		}
1785 
1786 		el = path_leaf_el(path);
1787 
1788 		i = ocfs2_find_rec(el, trunc_end);
1789 		/*
1790 		 * Need to go to previous extent block.
1791 		 */
1792 		if (i < 0) {
1793 			if (path->p_tree_depth == 0)
1794 				break;
1795 
1796 			ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb,
1797 							    path,
1798 							    &cluster_in_el);
1799 			if (ret) {
1800 				mlog_errno(ret);
1801 				goto out;
1802 			}
1803 
1804 			/*
1805 			 * We've reached the leftmost extent block,
1806 			 * it's safe to leave.
1807 			 */
1808 			if (cluster_in_el == 0)
1809 				break;
1810 
1811 			/*
1812 			 * The 'pos' searched for previous extent block is
1813 			 * always one cluster less than actual trunc_end.
1814 			 */
1815 			trunc_end = cluster_in_el + 1;
1816 
1817 			ocfs2_reinit_path(path, 1);
1818 
1819 			continue;
1820 
1821 		} else
1822 			rec = &el->l_recs[i];
1823 
1824 		ocfs2_calc_trunc_pos(inode, el, rec, trunc_start, &trunc_cpos,
1825 				     &trunc_len, &trunc_end, &blkno, &done);
1826 		if (done)
1827 			break;
1828 
1829 		flags = rec->e_flags;
1830 		phys_cpos = ocfs2_blocks_to_clusters(inode->i_sb, blkno);
1831 
1832 		ret = ocfs2_remove_btree_range(inode, &et, trunc_cpos,
1833 					       phys_cpos, trunc_len, flags,
1834 					       &dealloc, refcount_loc, false);
1835 		if (ret < 0) {
1836 			mlog_errno(ret);
1837 			goto out;
1838 		}
1839 
1840 		cluster_in_el = trunc_end;
1841 
1842 		ocfs2_reinit_path(path, 1);
1843 	}
1844 
1845 	ocfs2_truncate_cluster_pages(inode, byte_start, byte_len);
1846 
1847 out:
1848 	ocfs2_free_path(path);
1849 	ocfs2_schedule_truncate_log_flush(osb, 1);
1850 	ocfs2_run_deallocs(osb, &dealloc);
1851 
1852 	return ret;
1853 }
1854 
1855 /*
1856  * Parts of this function taken from xfs_change_file_space()
1857  */
1858 static int __ocfs2_change_file_space(struct file *file, struct inode *inode,
1859 				     loff_t f_pos, unsigned int cmd,
1860 				     struct ocfs2_space_resv *sr,
1861 				     int change_size)
1862 {
1863 	int ret;
1864 	s64 llen;
1865 	loff_t size;
1866 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1867 	struct buffer_head *di_bh = NULL;
1868 	handle_t *handle;
1869 	unsigned long long max_off = inode->i_sb->s_maxbytes;
1870 
1871 	if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
1872 		return -EROFS;
1873 
1874 	inode_lock(inode);
1875 
1876 	/*
1877 	 * This prevents concurrent writes on other nodes
1878 	 */
1879 	ret = ocfs2_rw_lock(inode, 1);
1880 	if (ret) {
1881 		mlog_errno(ret);
1882 		goto out;
1883 	}
1884 
1885 	ret = ocfs2_inode_lock(inode, &di_bh, 1);
1886 	if (ret) {
1887 		mlog_errno(ret);
1888 		goto out_rw_unlock;
1889 	}
1890 
1891 	if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
1892 		ret = -EPERM;
1893 		goto out_inode_unlock;
1894 	}
1895 
1896 	switch (sr->l_whence) {
1897 	case 0: /*SEEK_SET*/
1898 		break;
1899 	case 1: /*SEEK_CUR*/
1900 		sr->l_start += f_pos;
1901 		break;
1902 	case 2: /*SEEK_END*/
1903 		sr->l_start += i_size_read(inode);
1904 		break;
1905 	default:
1906 		ret = -EINVAL;
1907 		goto out_inode_unlock;
1908 	}
1909 	sr->l_whence = 0;
1910 
1911 	llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len;
1912 
1913 	if (sr->l_start < 0
1914 	    || sr->l_start > max_off
1915 	    || (sr->l_start + llen) < 0
1916 	    || (sr->l_start + llen) > max_off) {
1917 		ret = -EINVAL;
1918 		goto out_inode_unlock;
1919 	}
1920 	size = sr->l_start + sr->l_len;
1921 
1922 	if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64 ||
1923 	    cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) {
1924 		if (sr->l_len <= 0) {
1925 			ret = -EINVAL;
1926 			goto out_inode_unlock;
1927 		}
1928 	}
1929 
1930 	if (file && should_remove_suid(file->f_path.dentry)) {
1931 		ret = __ocfs2_write_remove_suid(inode, di_bh);
1932 		if (ret) {
1933 			mlog_errno(ret);
1934 			goto out_inode_unlock;
1935 		}
1936 	}
1937 
1938 	down_write(&OCFS2_I(inode)->ip_alloc_sem);
1939 	switch (cmd) {
1940 	case OCFS2_IOC_RESVSP:
1941 	case OCFS2_IOC_RESVSP64:
1942 		/*
1943 		 * This takes unsigned offsets, but the signed ones we
1944 		 * pass have been checked against overflow above.
1945 		 */
1946 		ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start,
1947 						       sr->l_len);
1948 		break;
1949 	case OCFS2_IOC_UNRESVSP:
1950 	case OCFS2_IOC_UNRESVSP64:
1951 		ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start,
1952 					       sr->l_len);
1953 		break;
1954 	default:
1955 		ret = -EINVAL;
1956 	}
1957 	up_write(&OCFS2_I(inode)->ip_alloc_sem);
1958 	if (ret) {
1959 		mlog_errno(ret);
1960 		goto out_inode_unlock;
1961 	}
1962 
1963 	/*
1964 	 * We update c/mtime for these changes
1965 	 */
1966 	handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
1967 	if (IS_ERR(handle)) {
1968 		ret = PTR_ERR(handle);
1969 		mlog_errno(ret);
1970 		goto out_inode_unlock;
1971 	}
1972 
1973 	if (change_size && i_size_read(inode) < size)
1974 		i_size_write(inode, size);
1975 
1976 	inode->i_ctime = inode->i_mtime = current_time(inode);
1977 	ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
1978 	if (ret < 0)
1979 		mlog_errno(ret);
1980 
1981 	if (file && (file->f_flags & O_SYNC))
1982 		handle->h_sync = 1;
1983 
1984 	ocfs2_commit_trans(osb, handle);
1985 
1986 out_inode_unlock:
1987 	brelse(di_bh);
1988 	ocfs2_inode_unlock(inode, 1);
1989 out_rw_unlock:
1990 	ocfs2_rw_unlock(inode, 1);
1991 
1992 out:
1993 	inode_unlock(inode);
1994 	return ret;
1995 }
1996 
1997 int ocfs2_change_file_space(struct file *file, unsigned int cmd,
1998 			    struct ocfs2_space_resv *sr)
1999 {
2000 	struct inode *inode = file_inode(file);
2001 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2002 	int ret;
2003 
2004 	if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) &&
2005 	    !ocfs2_writes_unwritten_extents(osb))
2006 		return -ENOTTY;
2007 	else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) &&
2008 		 !ocfs2_sparse_alloc(osb))
2009 		return -ENOTTY;
2010 
2011 	if (!S_ISREG(inode->i_mode))
2012 		return -EINVAL;
2013 
2014 	if (!(file->f_mode & FMODE_WRITE))
2015 		return -EBADF;
2016 
2017 	ret = mnt_want_write_file(file);
2018 	if (ret)
2019 		return ret;
2020 	ret = __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0);
2021 	mnt_drop_write_file(file);
2022 	return ret;
2023 }
2024 
2025 static long ocfs2_fallocate(struct file *file, int mode, loff_t offset,
2026 			    loff_t len)
2027 {
2028 	struct inode *inode = file_inode(file);
2029 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2030 	struct ocfs2_space_resv sr;
2031 	int change_size = 1;
2032 	int cmd = OCFS2_IOC_RESVSP64;
2033 
2034 	if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2035 		return -EOPNOTSUPP;
2036 	if (!ocfs2_writes_unwritten_extents(osb))
2037 		return -EOPNOTSUPP;
2038 
2039 	if (mode & FALLOC_FL_KEEP_SIZE)
2040 		change_size = 0;
2041 
2042 	if (mode & FALLOC_FL_PUNCH_HOLE)
2043 		cmd = OCFS2_IOC_UNRESVSP64;
2044 
2045 	sr.l_whence = 0;
2046 	sr.l_start = (s64)offset;
2047 	sr.l_len = (s64)len;
2048 
2049 	return __ocfs2_change_file_space(NULL, inode, offset, cmd, &sr,
2050 					 change_size);
2051 }
2052 
2053 int ocfs2_check_range_for_refcount(struct inode *inode, loff_t pos,
2054 				   size_t count)
2055 {
2056 	int ret = 0;
2057 	unsigned int extent_flags;
2058 	u32 cpos, clusters, extent_len, phys_cpos;
2059 	struct super_block *sb = inode->i_sb;
2060 
2061 	if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)) ||
2062 	    !ocfs2_is_refcount_inode(inode) ||
2063 	    OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
2064 		return 0;
2065 
2066 	cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
2067 	clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
2068 
2069 	while (clusters) {
2070 		ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
2071 					 &extent_flags);
2072 		if (ret < 0) {
2073 			mlog_errno(ret);
2074 			goto out;
2075 		}
2076 
2077 		if (phys_cpos && (extent_flags & OCFS2_EXT_REFCOUNTED)) {
2078 			ret = 1;
2079 			break;
2080 		}
2081 
2082 		if (extent_len > clusters)
2083 			extent_len = clusters;
2084 
2085 		clusters -= extent_len;
2086 		cpos += extent_len;
2087 	}
2088 out:
2089 	return ret;
2090 }
2091 
2092 static int ocfs2_is_io_unaligned(struct inode *inode, size_t count, loff_t pos)
2093 {
2094 	int blockmask = inode->i_sb->s_blocksize - 1;
2095 	loff_t final_size = pos + count;
2096 
2097 	if ((pos & blockmask) || (final_size & blockmask))
2098 		return 1;
2099 	return 0;
2100 }
2101 
2102 static int ocfs2_prepare_inode_for_refcount(struct inode *inode,
2103 					    struct file *file,
2104 					    loff_t pos, size_t count,
2105 					    int *meta_level)
2106 {
2107 	int ret;
2108 	struct buffer_head *di_bh = NULL;
2109 	u32 cpos = pos >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
2110 	u32 clusters =
2111 		ocfs2_clusters_for_bytes(inode->i_sb, pos + count) - cpos;
2112 
2113 	ret = ocfs2_inode_lock(inode, &di_bh, 1);
2114 	if (ret) {
2115 		mlog_errno(ret);
2116 		goto out;
2117 	}
2118 
2119 	*meta_level = 1;
2120 
2121 	ret = ocfs2_refcount_cow(inode, di_bh, cpos, clusters, UINT_MAX);
2122 	if (ret)
2123 		mlog_errno(ret);
2124 out:
2125 	brelse(di_bh);
2126 	return ret;
2127 }
2128 
2129 static int ocfs2_prepare_inode_for_write(struct file *file,
2130 					 loff_t pos,
2131 					 size_t count)
2132 {
2133 	int ret = 0, meta_level = 0;
2134 	struct dentry *dentry = file->f_path.dentry;
2135 	struct inode *inode = d_inode(dentry);
2136 	loff_t end;
2137 
2138 	/*
2139 	 * We start with a read level meta lock and only jump to an ex
2140 	 * if we need to make modifications here.
2141 	 */
2142 	for(;;) {
2143 		ret = ocfs2_inode_lock(inode, NULL, meta_level);
2144 		if (ret < 0) {
2145 			meta_level = -1;
2146 			mlog_errno(ret);
2147 			goto out;
2148 		}
2149 
2150 		/* Clear suid / sgid if necessary. We do this here
2151 		 * instead of later in the write path because
2152 		 * remove_suid() calls ->setattr without any hint that
2153 		 * we may have already done our cluster locking. Since
2154 		 * ocfs2_setattr() *must* take cluster locks to
2155 		 * proceed, this will lead us to recursively lock the
2156 		 * inode. There's also the dinode i_size state which
2157 		 * can be lost via setattr during extending writes (we
2158 		 * set inode->i_size at the end of a write. */
2159 		if (should_remove_suid(dentry)) {
2160 			if (meta_level == 0) {
2161 				ocfs2_inode_unlock(inode, meta_level);
2162 				meta_level = 1;
2163 				continue;
2164 			}
2165 
2166 			ret = ocfs2_write_remove_suid(inode);
2167 			if (ret < 0) {
2168 				mlog_errno(ret);
2169 				goto out_unlock;
2170 			}
2171 		}
2172 
2173 		end = pos + count;
2174 
2175 		ret = ocfs2_check_range_for_refcount(inode, pos, count);
2176 		if (ret == 1) {
2177 			ocfs2_inode_unlock(inode, meta_level);
2178 			meta_level = -1;
2179 
2180 			ret = ocfs2_prepare_inode_for_refcount(inode,
2181 							       file,
2182 							       pos,
2183 							       count,
2184 							       &meta_level);
2185 		}
2186 
2187 		if (ret < 0) {
2188 			mlog_errno(ret);
2189 			goto out_unlock;
2190 		}
2191 
2192 		break;
2193 	}
2194 
2195 out_unlock:
2196 	trace_ocfs2_prepare_inode_for_write(OCFS2_I(inode)->ip_blkno,
2197 					    pos, count);
2198 
2199 	if (meta_level >= 0)
2200 		ocfs2_inode_unlock(inode, meta_level);
2201 
2202 out:
2203 	return ret;
2204 }
2205 
2206 static ssize_t ocfs2_file_write_iter(struct kiocb *iocb,
2207 				    struct iov_iter *from)
2208 {
2209 	int direct_io, rw_level;
2210 	ssize_t written = 0;
2211 	ssize_t ret;
2212 	size_t count = iov_iter_count(from);
2213 	struct file *file = iocb->ki_filp;
2214 	struct inode *inode = file_inode(file);
2215 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2216 	int full_coherency = !(osb->s_mount_opt &
2217 			       OCFS2_MOUNT_COHERENCY_BUFFERED);
2218 	void *saved_ki_complete = NULL;
2219 	int append_write = ((iocb->ki_pos + count) >=
2220 			i_size_read(inode) ? 1 : 0);
2221 
2222 	trace_ocfs2_file_aio_write(inode, file, file->f_path.dentry,
2223 		(unsigned long long)OCFS2_I(inode)->ip_blkno,
2224 		file->f_path.dentry->d_name.len,
2225 		file->f_path.dentry->d_name.name,
2226 		(unsigned int)from->nr_segs);	/* GRRRRR */
2227 
2228 	if (count == 0)
2229 		return 0;
2230 
2231 	direct_io = iocb->ki_flags & IOCB_DIRECT ? 1 : 0;
2232 
2233 	inode_lock(inode);
2234 
2235 	/*
2236 	 * Concurrent O_DIRECT writes are allowed with
2237 	 * mount_option "coherency=buffered".
2238 	 * For append write, we must take rw EX.
2239 	 */
2240 	rw_level = (!direct_io || full_coherency || append_write);
2241 
2242 	ret = ocfs2_rw_lock(inode, rw_level);
2243 	if (ret < 0) {
2244 		mlog_errno(ret);
2245 		goto out_mutex;
2246 	}
2247 
2248 	/*
2249 	 * O_DIRECT writes with "coherency=full" need to take EX cluster
2250 	 * inode_lock to guarantee coherency.
2251 	 */
2252 	if (direct_io && full_coherency) {
2253 		/*
2254 		 * We need to take and drop the inode lock to force
2255 		 * other nodes to drop their caches.  Buffered I/O
2256 		 * already does this in write_begin().
2257 		 */
2258 		ret = ocfs2_inode_lock(inode, NULL, 1);
2259 		if (ret < 0) {
2260 			mlog_errno(ret);
2261 			goto out;
2262 		}
2263 
2264 		ocfs2_inode_unlock(inode, 1);
2265 	}
2266 
2267 	ret = generic_write_checks(iocb, from);
2268 	if (ret <= 0) {
2269 		if (ret)
2270 			mlog_errno(ret);
2271 		goto out;
2272 	}
2273 	count = ret;
2274 
2275 	ret = ocfs2_prepare_inode_for_write(file, iocb->ki_pos, count);
2276 	if (ret < 0) {
2277 		mlog_errno(ret);
2278 		goto out;
2279 	}
2280 
2281 	if (direct_io && !is_sync_kiocb(iocb) &&
2282 	    ocfs2_is_io_unaligned(inode, count, iocb->ki_pos)) {
2283 		/*
2284 		 * Make it a sync io if it's an unaligned aio.
2285 		 */
2286 		saved_ki_complete = xchg(&iocb->ki_complete, NULL);
2287 	}
2288 
2289 	/* communicate with ocfs2_dio_end_io */
2290 	ocfs2_iocb_set_rw_locked(iocb, rw_level);
2291 
2292 	written = __generic_file_write_iter(iocb, from);
2293 	/* buffered aio wouldn't have proper lock coverage today */
2294 	BUG_ON(written == -EIOCBQUEUED && !(iocb->ki_flags & IOCB_DIRECT));
2295 
2296 	/*
2297 	 * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
2298 	 * function pointer which is called when o_direct io completes so that
2299 	 * it can unlock our rw lock.
2300 	 * Unfortunately there are error cases which call end_io and others
2301 	 * that don't.  so we don't have to unlock the rw_lock if either an
2302 	 * async dio is going to do it in the future or an end_io after an
2303 	 * error has already done it.
2304 	 */
2305 	if ((written == -EIOCBQUEUED) || (!ocfs2_iocb_is_rw_locked(iocb))) {
2306 		rw_level = -1;
2307 	}
2308 
2309 	if (unlikely(written <= 0))
2310 		goto out;
2311 
2312 	if (((file->f_flags & O_DSYNC) && !direct_io) ||
2313 	    IS_SYNC(inode)) {
2314 		ret = filemap_fdatawrite_range(file->f_mapping,
2315 					       iocb->ki_pos - written,
2316 					       iocb->ki_pos - 1);
2317 		if (ret < 0)
2318 			written = ret;
2319 
2320 		if (!ret) {
2321 			ret = jbd2_journal_force_commit(osb->journal->j_journal);
2322 			if (ret < 0)
2323 				written = ret;
2324 		}
2325 
2326 		if (!ret)
2327 			ret = filemap_fdatawait_range(file->f_mapping,
2328 						      iocb->ki_pos - written,
2329 						      iocb->ki_pos - 1);
2330 	}
2331 
2332 out:
2333 	if (saved_ki_complete)
2334 		xchg(&iocb->ki_complete, saved_ki_complete);
2335 
2336 	if (rw_level != -1)
2337 		ocfs2_rw_unlock(inode, rw_level);
2338 
2339 out_mutex:
2340 	inode_unlock(inode);
2341 
2342 	if (written)
2343 		ret = written;
2344 	return ret;
2345 }
2346 
2347 static ssize_t ocfs2_file_read_iter(struct kiocb *iocb,
2348 				   struct iov_iter *to)
2349 {
2350 	int ret = 0, rw_level = -1, lock_level = 0;
2351 	struct file *filp = iocb->ki_filp;
2352 	struct inode *inode = file_inode(filp);
2353 
2354 	trace_ocfs2_file_aio_read(inode, filp, filp->f_path.dentry,
2355 			(unsigned long long)OCFS2_I(inode)->ip_blkno,
2356 			filp->f_path.dentry->d_name.len,
2357 			filp->f_path.dentry->d_name.name,
2358 			to->nr_segs);	/* GRRRRR */
2359 
2360 
2361 	if (!inode) {
2362 		ret = -EINVAL;
2363 		mlog_errno(ret);
2364 		goto bail;
2365 	}
2366 
2367 	/*
2368 	 * buffered reads protect themselves in ->readpage().  O_DIRECT reads
2369 	 * need locks to protect pending reads from racing with truncate.
2370 	 */
2371 	if (iocb->ki_flags & IOCB_DIRECT) {
2372 		ret = ocfs2_rw_lock(inode, 0);
2373 		if (ret < 0) {
2374 			mlog_errno(ret);
2375 			goto bail;
2376 		}
2377 		rw_level = 0;
2378 		/* communicate with ocfs2_dio_end_io */
2379 		ocfs2_iocb_set_rw_locked(iocb, rw_level);
2380 	}
2381 
2382 	/*
2383 	 * We're fine letting folks race truncates and extending
2384 	 * writes with read across the cluster, just like they can
2385 	 * locally. Hence no rw_lock during read.
2386 	 *
2387 	 * Take and drop the meta data lock to update inode fields
2388 	 * like i_size. This allows the checks down below
2389 	 * generic_file_aio_read() a chance of actually working.
2390 	 */
2391 	ret = ocfs2_inode_lock_atime(inode, filp->f_path.mnt, &lock_level);
2392 	if (ret < 0) {
2393 		mlog_errno(ret);
2394 		goto bail;
2395 	}
2396 	ocfs2_inode_unlock(inode, lock_level);
2397 
2398 	ret = generic_file_read_iter(iocb, to);
2399 	trace_generic_file_aio_read_ret(ret);
2400 
2401 	/* buffered aio wouldn't have proper lock coverage today */
2402 	BUG_ON(ret == -EIOCBQUEUED && !(iocb->ki_flags & IOCB_DIRECT));
2403 
2404 	/* see ocfs2_file_write_iter */
2405 	if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
2406 		rw_level = -1;
2407 	}
2408 
2409 bail:
2410 	if (rw_level != -1)
2411 		ocfs2_rw_unlock(inode, rw_level);
2412 
2413 	return ret;
2414 }
2415 
2416 /* Refer generic_file_llseek_unlocked() */
2417 static loff_t ocfs2_file_llseek(struct file *file, loff_t offset, int whence)
2418 {
2419 	struct inode *inode = file->f_mapping->host;
2420 	int ret = 0;
2421 
2422 	inode_lock(inode);
2423 
2424 	switch (whence) {
2425 	case SEEK_SET:
2426 		break;
2427 	case SEEK_END:
2428 		/* SEEK_END requires the OCFS2 inode lock for the file
2429 		 * because it references the file's size.
2430 		 */
2431 		ret = ocfs2_inode_lock(inode, NULL, 0);
2432 		if (ret < 0) {
2433 			mlog_errno(ret);
2434 			goto out;
2435 		}
2436 		offset += i_size_read(inode);
2437 		ocfs2_inode_unlock(inode, 0);
2438 		break;
2439 	case SEEK_CUR:
2440 		if (offset == 0) {
2441 			offset = file->f_pos;
2442 			goto out;
2443 		}
2444 		offset += file->f_pos;
2445 		break;
2446 	case SEEK_DATA:
2447 	case SEEK_HOLE:
2448 		ret = ocfs2_seek_data_hole_offset(file, &offset, whence);
2449 		if (ret)
2450 			goto out;
2451 		break;
2452 	default:
2453 		ret = -EINVAL;
2454 		goto out;
2455 	}
2456 
2457 	offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
2458 
2459 out:
2460 	inode_unlock(inode);
2461 	if (ret)
2462 		return ret;
2463 	return offset;
2464 }
2465 
2466 static int ocfs2_file_clone_range(struct file *file_in,
2467 				  loff_t pos_in,
2468 				  struct file *file_out,
2469 				  loff_t pos_out,
2470 				  u64 len)
2471 {
2472 	return ocfs2_reflink_remap_range(file_in, pos_in, file_out, pos_out,
2473 					 len, false);
2474 }
2475 
2476 static ssize_t ocfs2_file_dedupe_range(struct file *src_file,
2477 				       u64 loff,
2478 				       u64 len,
2479 				       struct file *dst_file,
2480 				       u64 dst_loff)
2481 {
2482 	int error;
2483 
2484 	error = ocfs2_reflink_remap_range(src_file, loff, dst_file, dst_loff,
2485 					  len, true);
2486 	if (error)
2487 		return error;
2488 	return len;
2489 }
2490 
2491 const struct inode_operations ocfs2_file_iops = {
2492 	.setattr	= ocfs2_setattr,
2493 	.getattr	= ocfs2_getattr,
2494 	.permission	= ocfs2_permission,
2495 	.listxattr	= ocfs2_listxattr,
2496 	.fiemap		= ocfs2_fiemap,
2497 	.get_acl	= ocfs2_iop_get_acl,
2498 	.set_acl	= ocfs2_iop_set_acl,
2499 };
2500 
2501 const struct inode_operations ocfs2_special_file_iops = {
2502 	.setattr	= ocfs2_setattr,
2503 	.getattr	= ocfs2_getattr,
2504 	.permission	= ocfs2_permission,
2505 	.get_acl	= ocfs2_iop_get_acl,
2506 	.set_acl	= ocfs2_iop_set_acl,
2507 };
2508 
2509 /*
2510  * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with
2511  * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks!
2512  */
2513 const struct file_operations ocfs2_fops = {
2514 	.llseek		= ocfs2_file_llseek,
2515 	.mmap		= ocfs2_mmap,
2516 	.fsync		= ocfs2_sync_file,
2517 	.release	= ocfs2_file_release,
2518 	.open		= ocfs2_file_open,
2519 	.read_iter	= ocfs2_file_read_iter,
2520 	.write_iter	= ocfs2_file_write_iter,
2521 	.unlocked_ioctl	= ocfs2_ioctl,
2522 #ifdef CONFIG_COMPAT
2523 	.compat_ioctl   = ocfs2_compat_ioctl,
2524 #endif
2525 	.lock		= ocfs2_lock,
2526 	.flock		= ocfs2_flock,
2527 	.splice_read	= generic_file_splice_read,
2528 	.splice_write	= iter_file_splice_write,
2529 	.fallocate	= ocfs2_fallocate,
2530 	.clone_file_range = ocfs2_file_clone_range,
2531 	.dedupe_file_range = ocfs2_file_dedupe_range,
2532 };
2533 
2534 const struct file_operations ocfs2_dops = {
2535 	.llseek		= generic_file_llseek,
2536 	.read		= generic_read_dir,
2537 	.iterate	= ocfs2_readdir,
2538 	.fsync		= ocfs2_sync_file,
2539 	.release	= ocfs2_dir_release,
2540 	.open		= ocfs2_dir_open,
2541 	.unlocked_ioctl	= ocfs2_ioctl,
2542 #ifdef CONFIG_COMPAT
2543 	.compat_ioctl   = ocfs2_compat_ioctl,
2544 #endif
2545 	.lock		= ocfs2_lock,
2546 	.flock		= ocfs2_flock,
2547 };
2548 
2549 /*
2550  * POSIX-lockless variants of our file_operations.
2551  *
2552  * These will be used if the underlying cluster stack does not support
2553  * posix file locking, if the user passes the "localflocks" mount
2554  * option, or if we have a local-only fs.
2555  *
2556  * ocfs2_flock is in here because all stacks handle UNIX file locks,
2557  * so we still want it in the case of no stack support for
2558  * plocks. Internally, it will do the right thing when asked to ignore
2559  * the cluster.
2560  */
2561 const struct file_operations ocfs2_fops_no_plocks = {
2562 	.llseek		= ocfs2_file_llseek,
2563 	.mmap		= ocfs2_mmap,
2564 	.fsync		= ocfs2_sync_file,
2565 	.release	= ocfs2_file_release,
2566 	.open		= ocfs2_file_open,
2567 	.read_iter	= ocfs2_file_read_iter,
2568 	.write_iter	= ocfs2_file_write_iter,
2569 	.unlocked_ioctl	= ocfs2_ioctl,
2570 #ifdef CONFIG_COMPAT
2571 	.compat_ioctl   = ocfs2_compat_ioctl,
2572 #endif
2573 	.flock		= ocfs2_flock,
2574 	.splice_read	= generic_file_splice_read,
2575 	.splice_write	= iter_file_splice_write,
2576 	.fallocate	= ocfs2_fallocate,
2577 	.clone_file_range = ocfs2_file_clone_range,
2578 	.dedupe_file_range = ocfs2_file_dedupe_range,
2579 };
2580 
2581 const struct file_operations ocfs2_dops_no_plocks = {
2582 	.llseek		= generic_file_llseek,
2583 	.read		= generic_read_dir,
2584 	.iterate	= ocfs2_readdir,
2585 	.fsync		= ocfs2_sync_file,
2586 	.release	= ocfs2_dir_release,
2587 	.open		= ocfs2_dir_open,
2588 	.unlocked_ioctl	= ocfs2_ioctl,
2589 #ifdef CONFIG_COMPAT
2590 	.compat_ioctl   = ocfs2_compat_ioctl,
2591 #endif
2592 	.flock		= ocfs2_flock,
2593 };
2594