xref: /openbmc/linux/fs/ocfs2/file.c (revision 87c2ce3b)
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/fs.h>
27 #include <linux/types.h>
28 #include <linux/slab.h>
29 #include <linux/highmem.h>
30 #include <linux/pagemap.h>
31 #include <linux/uio.h>
32 
33 #define MLOG_MASK_PREFIX ML_INODE
34 #include <cluster/masklog.h>
35 
36 #include "ocfs2.h"
37 
38 #include "alloc.h"
39 #include "aops.h"
40 #include "dir.h"
41 #include "dlmglue.h"
42 #include "extent_map.h"
43 #include "file.h"
44 #include "sysfile.h"
45 #include "inode.h"
46 #include "journal.h"
47 #include "mmap.h"
48 #include "suballoc.h"
49 #include "super.h"
50 
51 #include "buffer_head_io.h"
52 
53 static int ocfs2_sync_inode(struct inode *inode)
54 {
55 	filemap_fdatawrite(inode->i_mapping);
56 	return sync_mapping_buffers(inode->i_mapping);
57 }
58 
59 static int ocfs2_file_open(struct inode *inode, struct file *file)
60 {
61 	int status;
62 	int mode = file->f_flags;
63 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
64 
65 	mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file,
66 		   file->f_dentry->d_name.len, file->f_dentry->d_name.name);
67 
68 	spin_lock(&oi->ip_lock);
69 
70 	/* Check that the inode hasn't been wiped from disk by another
71 	 * node. If it hasn't then we're safe as long as we hold the
72 	 * spin lock until our increment of open count. */
73 	if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
74 		spin_unlock(&oi->ip_lock);
75 
76 		status = -ENOENT;
77 		goto leave;
78 	}
79 
80 	if (mode & O_DIRECT)
81 		oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT;
82 
83 	oi->ip_open_count++;
84 	spin_unlock(&oi->ip_lock);
85 	status = 0;
86 leave:
87 	mlog_exit(status);
88 	return status;
89 }
90 
91 static int ocfs2_file_release(struct inode *inode, struct file *file)
92 {
93 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
94 
95 	mlog_entry("(0x%p, 0x%p, '%.*s')\n", inode, file,
96 		       file->f_dentry->d_name.len,
97 		       file->f_dentry->d_name.name);
98 
99 	spin_lock(&oi->ip_lock);
100 	if (!--oi->ip_open_count)
101 		oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT;
102 	spin_unlock(&oi->ip_lock);
103 
104 	mlog_exit(0);
105 
106 	return 0;
107 }
108 
109 static int ocfs2_sync_file(struct file *file,
110 			   struct dentry *dentry,
111 			   int datasync)
112 {
113 	int err = 0;
114 	journal_t *journal;
115 	struct inode *inode = dentry->d_inode;
116 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
117 
118 	mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", file, dentry, datasync,
119 		   dentry->d_name.len, dentry->d_name.name);
120 
121 	err = ocfs2_sync_inode(dentry->d_inode);
122 	if (err)
123 		goto bail;
124 
125 	journal = osb->journal->j_journal;
126 	err = journal_force_commit(journal);
127 
128 bail:
129 	mlog_exit(err);
130 
131 	return (err < 0) ? -EIO : 0;
132 }
133 
134 int ocfs2_set_inode_size(struct ocfs2_journal_handle *handle,
135 			 struct inode *inode,
136 			 struct buffer_head *fe_bh,
137 			 u64 new_i_size)
138 {
139 	int status;
140 
141 	mlog_entry_void();
142 	i_size_write(inode, new_i_size);
143 	inode->i_blocks = ocfs2_align_bytes_to_sectors(new_i_size);
144 	inode->i_ctime = inode->i_mtime = CURRENT_TIME;
145 
146 	status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
147 	if (status < 0) {
148 		mlog_errno(status);
149 		goto bail;
150 	}
151 
152 bail:
153 	mlog_exit(status);
154 	return status;
155 }
156 
157 static int ocfs2_simple_size_update(struct inode *inode,
158 				    struct buffer_head *di_bh,
159 				    u64 new_i_size)
160 {
161 	int ret;
162 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
163 	struct ocfs2_journal_handle *handle = NULL;
164 
165 	handle = ocfs2_start_trans(osb, NULL,
166 				   OCFS2_INODE_UPDATE_CREDITS);
167 	if (handle == NULL) {
168 		ret = -ENOMEM;
169 		mlog_errno(ret);
170 		goto out;
171 	}
172 
173 	ret = ocfs2_set_inode_size(handle, inode, di_bh,
174 				   new_i_size);
175 	if (ret < 0)
176 		mlog_errno(ret);
177 
178 	ocfs2_commit_trans(handle);
179 out:
180 	return ret;
181 }
182 
183 static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
184 				     struct inode *inode,
185 				     struct buffer_head *fe_bh,
186 				     u64 new_i_size)
187 {
188 	int status;
189 	struct ocfs2_journal_handle *handle;
190 
191 	mlog_entry_void();
192 
193 	/* TODO: This needs to actually orphan the inode in this
194 	 * transaction. */
195 
196 	handle = ocfs2_start_trans(osb, NULL, OCFS2_INODE_UPDATE_CREDITS);
197 	if (IS_ERR(handle)) {
198 		status = PTR_ERR(handle);
199 		mlog_errno(status);
200 		goto out;
201 	}
202 
203 	status = ocfs2_set_inode_size(handle, inode, fe_bh, new_i_size);
204 	if (status < 0)
205 		mlog_errno(status);
206 
207 	ocfs2_commit_trans(handle);
208 out:
209 	mlog_exit(status);
210 	return status;
211 }
212 
213 static int ocfs2_truncate_file(struct inode *inode,
214 			       struct buffer_head *di_bh,
215 			       u64 new_i_size)
216 {
217 	int status = 0;
218 	struct ocfs2_dinode *fe = NULL;
219 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
220 	struct ocfs2_truncate_context *tc = NULL;
221 
222 	mlog_entry("(inode = %"MLFu64", new_i_size = %"MLFu64"\n",
223 		   OCFS2_I(inode)->ip_blkno, new_i_size);
224 
225 	truncate_inode_pages(inode->i_mapping, new_i_size);
226 
227 	fe = (struct ocfs2_dinode *) di_bh->b_data;
228 	if (!OCFS2_IS_VALID_DINODE(fe)) {
229 		OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe);
230 		status = -EIO;
231 		goto bail;
232 	}
233 
234 	mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode),
235 			"Inode %"MLFu64", inode i_size = %lld != di "
236 			"i_size = %"MLFu64", i_flags = 0x%x\n",
237 			OCFS2_I(inode)->ip_blkno,
238 			i_size_read(inode),
239 			le64_to_cpu(fe->i_size), le32_to_cpu(fe->i_flags));
240 
241 	if (new_i_size > le64_to_cpu(fe->i_size)) {
242 		mlog(0, "asked to truncate file with size (%"MLFu64") "
243 		     "to size (%"MLFu64")!\n",
244 		     le64_to_cpu(fe->i_size), new_i_size);
245 		status = -EINVAL;
246 		mlog_errno(status);
247 		goto bail;
248 	}
249 
250 	mlog(0, "inode %"MLFu64", i_size = %"MLFu64", new_i_size = %"MLFu64"\n",
251 	     le64_to_cpu(fe->i_blkno), le64_to_cpu(fe->i_size), new_i_size);
252 
253 	/* lets handle the simple truncate cases before doing any more
254 	 * cluster locking. */
255 	if (new_i_size == le64_to_cpu(fe->i_size))
256 		goto bail;
257 
258 	if (le32_to_cpu(fe->i_clusters) ==
259 	    ocfs2_clusters_for_bytes(osb->sb, new_i_size)) {
260 		mlog(0, "fe->i_clusters = %u, so we do a simple truncate\n",
261 		     fe->i_clusters);
262 		/* No allocation change is required, so lets fast path
263 		 * this truncate. */
264 		status = ocfs2_simple_size_update(inode, di_bh, new_i_size);
265 		if (status < 0)
266 			mlog_errno(status);
267 		goto bail;
268 	}
269 
270 	/* This forces other nodes to sync and drop their pages */
271 	status = ocfs2_data_lock(inode, 1);
272 	if (status < 0) {
273 		mlog_errno(status);
274 		goto bail;
275 	}
276 	ocfs2_data_unlock(inode, 1);
277 
278 	/* alright, we're going to need to do a full blown alloc size
279 	 * change. Orphan the inode so that recovery can complete the
280 	 * truncate if necessary. This does the task of marking
281 	 * i_size. */
282 	status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
283 	if (status < 0) {
284 		mlog_errno(status);
285 		goto bail;
286 	}
287 
288 	status = ocfs2_prepare_truncate(osb, inode, di_bh, &tc);
289 	if (status < 0) {
290 		mlog_errno(status);
291 		goto bail;
292 	}
293 
294 	status = ocfs2_commit_truncate(osb, inode, di_bh, tc);
295 	if (status < 0) {
296 		mlog_errno(status);
297 		goto bail;
298 	}
299 
300 	/* TODO: orphan dir cleanup here. */
301 bail:
302 
303 	mlog_exit(status);
304 	return status;
305 }
306 
307 /*
308  * extend allocation only here.
309  * we'll update all the disk stuff, and oip->alloc_size
310  *
311  * expect stuff to be locked, a transaction started and enough data /
312  * metadata reservations in the contexts.
313  *
314  * Will return -EAGAIN, and a reason if a restart is needed.
315  * If passed in, *reason will always be set, even in error.
316  */
317 int ocfs2_do_extend_allocation(struct ocfs2_super *osb,
318 			       struct inode *inode,
319 			       u32 clusters_to_add,
320 			       struct buffer_head *fe_bh,
321 			       struct ocfs2_journal_handle *handle,
322 			       struct ocfs2_alloc_context *data_ac,
323 			       struct ocfs2_alloc_context *meta_ac,
324 			       enum ocfs2_alloc_restarted *reason_ret)
325 {
326 	int status = 0;
327 	int free_extents;
328 	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
329 	enum ocfs2_alloc_restarted reason = RESTART_NONE;
330 	u32 bit_off, num_bits;
331 	u64 block;
332 
333 	BUG_ON(!clusters_to_add);
334 
335 	free_extents = ocfs2_num_free_extents(osb, inode, fe);
336 	if (free_extents < 0) {
337 		status = free_extents;
338 		mlog_errno(status);
339 		goto leave;
340 	}
341 
342 	/* there are two cases which could cause us to EAGAIN in the
343 	 * we-need-more-metadata case:
344 	 * 1) we haven't reserved *any*
345 	 * 2) we are so fragmented, we've needed to add metadata too
346 	 *    many times. */
347 	if (!free_extents && !meta_ac) {
348 		mlog(0, "we haven't reserved any metadata!\n");
349 		status = -EAGAIN;
350 		reason = RESTART_META;
351 		goto leave;
352 	} else if ((!free_extents)
353 		   && (ocfs2_alloc_context_bits_left(meta_ac)
354 		       < ocfs2_extend_meta_needed(fe))) {
355 		mlog(0, "filesystem is really fragmented...\n");
356 		status = -EAGAIN;
357 		reason = RESTART_META;
358 		goto leave;
359 	}
360 
361 	status = ocfs2_claim_clusters(osb, handle, data_ac, 1,
362 				      &bit_off, &num_bits);
363 	if (status < 0) {
364 		if (status != -ENOSPC)
365 			mlog_errno(status);
366 		goto leave;
367 	}
368 
369 	BUG_ON(num_bits > clusters_to_add);
370 
371 	/* reserve our write early -- insert_extent may update the inode */
372 	status = ocfs2_journal_access(handle, inode, fe_bh,
373 				      OCFS2_JOURNAL_ACCESS_WRITE);
374 	if (status < 0) {
375 		mlog_errno(status);
376 		goto leave;
377 	}
378 
379 	block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
380 	mlog(0, "Allocating %u clusters at block %u for inode %"MLFu64"\n",
381 	     num_bits, bit_off, OCFS2_I(inode)->ip_blkno);
382 	status = ocfs2_insert_extent(osb, handle, inode, fe_bh, block,
383 				     num_bits, meta_ac);
384 	if (status < 0) {
385 		mlog_errno(status);
386 		goto leave;
387 	}
388 
389 	le32_add_cpu(&fe->i_clusters, num_bits);
390 	spin_lock(&OCFS2_I(inode)->ip_lock);
391 	OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
392 	spin_unlock(&OCFS2_I(inode)->ip_lock);
393 
394 	status = ocfs2_journal_dirty(handle, fe_bh);
395 	if (status < 0) {
396 		mlog_errno(status);
397 		goto leave;
398 	}
399 
400 	clusters_to_add -= num_bits;
401 
402 	if (clusters_to_add) {
403 		mlog(0, "need to alloc once more, clusters = %u, wanted = "
404 		     "%u\n", fe->i_clusters, clusters_to_add);
405 		status = -EAGAIN;
406 		reason = RESTART_TRANS;
407 	}
408 
409 leave:
410 	mlog_exit(status);
411 	if (reason_ret)
412 		*reason_ret = reason;
413 	return status;
414 }
415 
416 static int ocfs2_extend_allocation(struct inode *inode,
417 				   u32 clusters_to_add)
418 {
419 	int status = 0;
420 	int restart_func = 0;
421 	int drop_alloc_sem = 0;
422 	int credits, num_free_extents;
423 	u32 prev_clusters;
424 	struct buffer_head *bh = NULL;
425 	struct ocfs2_dinode *fe = NULL;
426 	struct ocfs2_journal_handle *handle = NULL;
427 	struct ocfs2_alloc_context *data_ac = NULL;
428 	struct ocfs2_alloc_context *meta_ac = NULL;
429 	enum ocfs2_alloc_restarted why;
430 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
431 
432 	mlog_entry("(clusters_to_add = %u)\n", clusters_to_add);
433 
434 	status = ocfs2_read_block(osb, OCFS2_I(inode)->ip_blkno, &bh,
435 				  OCFS2_BH_CACHED, inode);
436 	if (status < 0) {
437 		mlog_errno(status);
438 		goto leave;
439 	}
440 
441 	fe = (struct ocfs2_dinode *) bh->b_data;
442 	if (!OCFS2_IS_VALID_DINODE(fe)) {
443 		OCFS2_RO_ON_INVALID_DINODE(inode->i_sb, fe);
444 		status = -EIO;
445 		goto leave;
446 	}
447 
448 restart_all:
449 	BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
450 
451 	mlog(0, "extend inode %"MLFu64", i_size = %lld, fe->i_clusters = %u, "
452 	     "clusters_to_add = %u\n",
453 	     OCFS2_I(inode)->ip_blkno, i_size_read(inode),
454 	     fe->i_clusters, clusters_to_add);
455 
456 	handle = ocfs2_alloc_handle(osb);
457 	if (handle == NULL) {
458 		status = -ENOMEM;
459 		mlog_errno(status);
460 		goto leave;
461 	}
462 
463 	num_free_extents = ocfs2_num_free_extents(osb,
464 						  inode,
465 						  fe);
466 	if (num_free_extents < 0) {
467 		status = num_free_extents;
468 		mlog_errno(status);
469 		goto leave;
470 	}
471 
472 	if (!num_free_extents) {
473 		status = ocfs2_reserve_new_metadata(osb,
474 						    handle,
475 						    fe,
476 						    &meta_ac);
477 		if (status < 0) {
478 			if (status != -ENOSPC)
479 				mlog_errno(status);
480 			goto leave;
481 		}
482 	}
483 
484 	status = ocfs2_reserve_clusters(osb,
485 					handle,
486 					clusters_to_add,
487 					&data_ac);
488 	if (status < 0) {
489 		if (status != -ENOSPC)
490 			mlog_errno(status);
491 		goto leave;
492 	}
493 
494 	/* blocks peope in read/write from reading our allocation
495 	 * until we're done changing it. We depend on i_mutex to block
496 	 * other extend/truncate calls while we're here. Ordering wrt
497 	 * start_trans is important here -- always do it before! */
498 	down_write(&OCFS2_I(inode)->ip_alloc_sem);
499 	drop_alloc_sem = 1;
500 
501 	credits = ocfs2_calc_extend_credits(osb->sb, fe, clusters_to_add);
502 	handle = ocfs2_start_trans(osb, handle, credits);
503 	if (IS_ERR(handle)) {
504 		status = PTR_ERR(handle);
505 		handle = NULL;
506 		mlog_errno(status);
507 		goto leave;
508 	}
509 
510 restarted_transaction:
511 	/* reserve a write to the file entry early on - that we if we
512 	 * run out of credits in the allocation path, we can still
513 	 * update i_size. */
514 	status = ocfs2_journal_access(handle, inode, bh,
515 				      OCFS2_JOURNAL_ACCESS_WRITE);
516 	if (status < 0) {
517 		mlog_errno(status);
518 		goto leave;
519 	}
520 
521 	prev_clusters = OCFS2_I(inode)->ip_clusters;
522 
523 	status = ocfs2_do_extend_allocation(osb,
524 					    inode,
525 					    clusters_to_add,
526 					    bh,
527 					    handle,
528 					    data_ac,
529 					    meta_ac,
530 					    &why);
531 	if ((status < 0) && (status != -EAGAIN)) {
532 		if (status != -ENOSPC)
533 			mlog_errno(status);
534 		goto leave;
535 	}
536 
537 	status = ocfs2_journal_dirty(handle, bh);
538 	if (status < 0) {
539 		mlog_errno(status);
540 		goto leave;
541 	}
542 
543 	spin_lock(&OCFS2_I(inode)->ip_lock);
544 	clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
545 	spin_unlock(&OCFS2_I(inode)->ip_lock);
546 
547 	if (why != RESTART_NONE && clusters_to_add) {
548 		if (why == RESTART_META) {
549 			mlog(0, "restarting function.\n");
550 			restart_func = 1;
551 		} else {
552 			BUG_ON(why != RESTART_TRANS);
553 
554 			mlog(0, "restarting transaction.\n");
555 			/* TODO: This can be more intelligent. */
556 			credits = ocfs2_calc_extend_credits(osb->sb,
557 							    fe,
558 							    clusters_to_add);
559 			status = ocfs2_extend_trans(handle, credits);
560 			if (status < 0) {
561 				/* handle still has to be committed at
562 				 * this point. */
563 				status = -ENOMEM;
564 				mlog_errno(status);
565 				goto leave;
566 			}
567 			goto restarted_transaction;
568 		}
569 	}
570 
571 	mlog(0, "fe: i_clusters = %u, i_size=%"MLFu64"\n",
572 	     fe->i_clusters, fe->i_size);
573 	mlog(0, "inode: ip_clusters=%u, i_size=%lld\n",
574 	     OCFS2_I(inode)->ip_clusters, i_size_read(inode));
575 
576 leave:
577 	if (drop_alloc_sem) {
578 		up_write(&OCFS2_I(inode)->ip_alloc_sem);
579 		drop_alloc_sem = 0;
580 	}
581 	if (handle) {
582 		ocfs2_commit_trans(handle);
583 		handle = NULL;
584 	}
585 	if (data_ac) {
586 		ocfs2_free_alloc_context(data_ac);
587 		data_ac = NULL;
588 	}
589 	if (meta_ac) {
590 		ocfs2_free_alloc_context(meta_ac);
591 		meta_ac = NULL;
592 	}
593 	if ((!status) && restart_func) {
594 		restart_func = 0;
595 		goto restart_all;
596 	}
597 	if (bh) {
598 		brelse(bh);
599 		bh = NULL;
600 	}
601 
602 	mlog_exit(status);
603 	return status;
604 }
605 
606 /* Some parts of this taken from generic_cont_expand, which turned out
607  * to be too fragile to do exactly what we need without us having to
608  * worry about recursive locking in ->commit_write(). */
609 static int ocfs2_write_zero_page(struct inode *inode,
610 				 u64 size)
611 {
612 	struct address_space *mapping = inode->i_mapping;
613 	struct page *page;
614 	unsigned long index;
615 	unsigned int offset;
616 	struct ocfs2_journal_handle *handle = NULL;
617 	int ret;
618 
619 	offset = (size & (PAGE_CACHE_SIZE-1)); /* Within page */
620 	/* ugh.  in prepare/commit_write, if from==to==start of block, we
621 	** skip the prepare.  make sure we never send an offset for the start
622 	** of a block
623 	*/
624 	if ((offset & (inode->i_sb->s_blocksize - 1)) == 0) {
625 		offset++;
626 	}
627 	index = size >> PAGE_CACHE_SHIFT;
628 
629 	page = grab_cache_page(mapping, index);
630 	if (!page) {
631 		ret = -ENOMEM;
632 		mlog_errno(ret);
633 		goto out;
634 	}
635 
636 	ret = ocfs2_prepare_write(NULL, page, offset, offset);
637 	if (ret < 0) {
638 		mlog_errno(ret);
639 		goto out_unlock;
640 	}
641 
642 	if (ocfs2_should_order_data(inode)) {
643 		handle = ocfs2_start_walk_page_trans(inode, page, offset,
644 						     offset);
645 		if (IS_ERR(handle)) {
646 			ret = PTR_ERR(handle);
647 			handle = NULL;
648 			goto out_unlock;
649 		}
650 	}
651 
652 	/* must not update i_size! */
653 	ret = block_commit_write(page, offset, offset);
654 	if (ret < 0)
655 		mlog_errno(ret);
656 	else
657 		ret = 0;
658 
659 	if (handle)
660 		ocfs2_commit_trans(handle);
661 out_unlock:
662 	unlock_page(page);
663 	page_cache_release(page);
664 out:
665 	return ret;
666 }
667 
668 static int ocfs2_zero_extend(struct inode *inode,
669 			     u64 zero_to_size)
670 {
671 	int ret = 0;
672 	u64 start_off;
673 	struct super_block *sb = inode->i_sb;
674 
675 	start_off = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
676 	while (start_off < zero_to_size) {
677 		ret = ocfs2_write_zero_page(inode, start_off);
678 		if (ret < 0) {
679 			mlog_errno(ret);
680 			goto out;
681 		}
682 
683 		start_off += sb->s_blocksize;
684 	}
685 
686 out:
687 	return ret;
688 }
689 
690 static int ocfs2_extend_file(struct inode *inode,
691 			     struct buffer_head *di_bh,
692 			     u64 new_i_size)
693 {
694 	int ret = 0;
695 	u32 clusters_to_add;
696 
697 	/* setattr sometimes calls us like this. */
698 	if (new_i_size == 0)
699 		goto out;
700 
701 	if (i_size_read(inode) == new_i_size)
702   		goto out;
703 	BUG_ON(new_i_size < i_size_read(inode));
704 
705 	clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size) -
706 		OCFS2_I(inode)->ip_clusters;
707 
708 	if (clusters_to_add) {
709 		ret = ocfs2_extend_allocation(inode, clusters_to_add);
710 		if (ret < 0) {
711 			mlog_errno(ret);
712 			goto out;
713 		}
714 
715 		ret = ocfs2_zero_extend(inode, new_i_size);
716 		if (ret < 0) {
717 			mlog_errno(ret);
718 			goto out;
719 		}
720 	}
721 
722 	/* No allocation required, we just use this helper to
723 	 * do a trivial update of i_size. */
724 	ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
725 	if (ret < 0) {
726 		mlog_errno(ret);
727 		goto out;
728 	}
729 
730 out:
731 	return ret;
732 }
733 
734 int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
735 {
736 	int status = 0, size_change;
737 	struct inode *inode = dentry->d_inode;
738 	struct super_block *sb = inode->i_sb;
739 	struct ocfs2_super *osb = OCFS2_SB(sb);
740 	struct buffer_head *bh = NULL;
741 	struct ocfs2_journal_handle *handle = NULL;
742 
743 	mlog_entry("(0x%p, '%.*s')\n", dentry,
744 	           dentry->d_name.len, dentry->d_name.name);
745 
746 	if (attr->ia_valid & ATTR_MODE)
747 		mlog(0, "mode change: %d\n", attr->ia_mode);
748 	if (attr->ia_valid & ATTR_UID)
749 		mlog(0, "uid change: %d\n", attr->ia_uid);
750 	if (attr->ia_valid & ATTR_GID)
751 		mlog(0, "gid change: %d\n", attr->ia_gid);
752 	if (attr->ia_valid & ATTR_SIZE)
753 		mlog(0, "size change...\n");
754 	if (attr->ia_valid & (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME))
755 		mlog(0, "time change...\n");
756 
757 #define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
758 			   | ATTR_GID | ATTR_UID | ATTR_MODE)
759 	if (!(attr->ia_valid & OCFS2_VALID_ATTRS)) {
760 		mlog(0, "can't handle attrs: 0x%x\n", attr->ia_valid);
761 		return 0;
762 	}
763 
764 	status = inode_change_ok(inode, attr);
765 	if (status)
766 		return status;
767 
768 	size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
769 	if (size_change) {
770 		status = ocfs2_rw_lock(inode, 1);
771 		if (status < 0) {
772 			mlog_errno(status);
773 			goto bail;
774 		}
775 	}
776 
777 	status = ocfs2_meta_lock(inode, NULL, &bh, 1);
778 	if (status < 0) {
779 		if (status != -ENOENT)
780 			mlog_errno(status);
781 		goto bail_unlock_rw;
782 	}
783 
784 	if (size_change && attr->ia_size != i_size_read(inode)) {
785 		if (i_size_read(inode) > attr->ia_size)
786 			status = ocfs2_truncate_file(inode, bh, attr->ia_size);
787 		else
788 			status = ocfs2_extend_file(inode, bh, attr->ia_size);
789 		if (status < 0) {
790 			if (status != -ENOSPC)
791 				mlog_errno(status);
792 			status = -ENOSPC;
793 			goto bail_unlock;
794 		}
795 	}
796 
797 	handle = ocfs2_start_trans(osb, NULL, OCFS2_INODE_UPDATE_CREDITS);
798 	if (IS_ERR(handle)) {
799 		status = PTR_ERR(handle);
800 		mlog_errno(status);
801 		goto bail_unlock;
802 	}
803 
804 	status = inode_setattr(inode, attr);
805 	if (status < 0) {
806 		mlog_errno(status);
807 		goto bail_commit;
808 	}
809 
810 	status = ocfs2_mark_inode_dirty(handle, inode, bh);
811 	if (status < 0)
812 		mlog_errno(status);
813 
814 bail_commit:
815 	ocfs2_commit_trans(handle);
816 bail_unlock:
817 	ocfs2_meta_unlock(inode, 1);
818 bail_unlock_rw:
819 	if (size_change)
820 		ocfs2_rw_unlock(inode, 1);
821 bail:
822 	if (bh)
823 		brelse(bh);
824 
825 	mlog_exit(status);
826 	return status;
827 }
828 
829 int ocfs2_getattr(struct vfsmount *mnt,
830 		  struct dentry *dentry,
831 		  struct kstat *stat)
832 {
833 	struct inode *inode = dentry->d_inode;
834 	struct super_block *sb = dentry->d_inode->i_sb;
835 	struct ocfs2_super *osb = sb->s_fs_info;
836 	int err;
837 
838 	mlog_entry_void();
839 
840 	err = ocfs2_inode_revalidate(dentry);
841 	if (err) {
842 		if (err != -ENOENT)
843 			mlog_errno(err);
844 		goto bail;
845 	}
846 
847 	generic_fillattr(inode, stat);
848 
849 	/* We set the blksize from the cluster size for performance */
850 	stat->blksize = osb->s_clustersize;
851 
852 bail:
853 	mlog_exit(err);
854 
855 	return err;
856 }
857 
858 static int ocfs2_write_remove_suid(struct inode *inode)
859 {
860 	int ret;
861 	struct buffer_head *bh = NULL;
862 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
863 	struct ocfs2_journal_handle *handle;
864 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
865 	struct ocfs2_dinode *di;
866 
867 	mlog_entry("(Inode %"MLFu64", mode 0%o)\n", oi->ip_blkno,
868 		   inode->i_mode);
869 
870 	handle = ocfs2_start_trans(osb, NULL, OCFS2_INODE_UPDATE_CREDITS);
871 	if (handle == NULL) {
872 		ret = -ENOMEM;
873 		mlog_errno(ret);
874 		goto out;
875 	}
876 
877 	ret = ocfs2_read_block(osb, oi->ip_blkno, &bh, OCFS2_BH_CACHED, inode);
878 	if (ret < 0) {
879 		mlog_errno(ret);
880 		goto out_trans;
881 	}
882 
883 	ret = ocfs2_journal_access(handle, inode, bh,
884 				   OCFS2_JOURNAL_ACCESS_WRITE);
885 	if (ret < 0) {
886 		mlog_errno(ret);
887 		goto out_bh;
888 	}
889 
890 	inode->i_mode &= ~S_ISUID;
891 	if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
892 		inode->i_mode &= ~S_ISGID;
893 
894 	di = (struct ocfs2_dinode *) bh->b_data;
895 	di->i_mode = cpu_to_le16(inode->i_mode);
896 
897 	ret = ocfs2_journal_dirty(handle, bh);
898 	if (ret < 0)
899 		mlog_errno(ret);
900 out_bh:
901 	brelse(bh);
902 out_trans:
903 	ocfs2_commit_trans(handle);
904 out:
905 	mlog_exit(ret);
906 	return ret;
907 }
908 
909 static inline int ocfs2_write_should_remove_suid(struct inode *inode)
910 {
911 	mode_t mode = inode->i_mode;
912 
913 	if (!capable(CAP_FSETID)) {
914 		if (unlikely(mode & S_ISUID))
915 			return 1;
916 
917 		if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
918 			return 1;
919 	}
920 	return 0;
921 }
922 
923 static ssize_t ocfs2_file_aio_write(struct kiocb *iocb,
924 				    const char __user *buf,
925 				    size_t count,
926 				    loff_t pos)
927 {
928 	struct iovec local_iov = { .iov_base = (void __user *)buf,
929 				   .iov_len = count };
930 	int ret, rw_level = -1, meta_level = -1, have_alloc_sem = 0;
931 	u32 clusters;
932 	struct file *filp = iocb->ki_filp;
933 	struct inode *inode = filp->f_dentry->d_inode;
934 	loff_t newsize, saved_pos;
935 #ifdef OCFS2_ORACORE_WORKAROUNDS
936 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
937 #endif
938 
939 	mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", filp, buf,
940 		   (unsigned int)count,
941 		   filp->f_dentry->d_name.len,
942 		   filp->f_dentry->d_name.name);
943 
944 	/* happy write of zero bytes */
945 	if (count == 0)
946 		return 0;
947 
948 	if (!inode) {
949 		mlog(0, "bad inode\n");
950 		return -EIO;
951 	}
952 
953 #ifdef OCFS2_ORACORE_WORKAROUNDS
954 	/* ugh, work around some applications which open everything O_DIRECT +
955 	 * O_APPEND and really don't mean to use O_DIRECT. */
956 	if (osb->s_mount_opt & OCFS2_MOUNT_COMPAT_OCFS &&
957 	    (filp->f_flags & O_APPEND) && (filp->f_flags & O_DIRECT))
958 		filp->f_flags &= ~O_DIRECT;
959 #endif
960 
961 	mutex_lock(&inode->i_mutex);
962 	/* to match setattr's i_mutex -> i_alloc_sem -> rw_lock ordering */
963 	if (filp->f_flags & O_DIRECT) {
964 		have_alloc_sem = 1;
965 		down_read(&inode->i_alloc_sem);
966 	}
967 
968 	/* concurrent O_DIRECT writes are allowed */
969 	rw_level = (filp->f_flags & O_DIRECT) ? 0 : 1;
970 	ret = ocfs2_rw_lock(inode, rw_level);
971 	if (ret < 0) {
972 		rw_level = -1;
973 		mlog_errno(ret);
974 		goto out;
975 	}
976 
977 	/*
978 	 * We sample i_size under a read level meta lock to see if our write
979 	 * is extending the file, if it is we back off and get a write level
980 	 * meta lock.
981 	 */
982 	meta_level = (filp->f_flags & O_APPEND) ? 1 : 0;
983 	for(;;) {
984 		ret = ocfs2_meta_lock(inode, NULL, NULL, meta_level);
985 		if (ret < 0) {
986 			meta_level = -1;
987 			mlog_errno(ret);
988 			goto out;
989 		}
990 
991 		/* Clear suid / sgid if necessary. We do this here
992 		 * instead of later in the write path because
993 		 * remove_suid() calls ->setattr without any hint that
994 		 * we may have already done our cluster locking. Since
995 		 * ocfs2_setattr() *must* take cluster locks to
996 		 * proceeed, this will lead us to recursively lock the
997 		 * inode. There's also the dinode i_size state which
998 		 * can be lost via setattr during extending writes (we
999 		 * set inode->i_size at the end of a write. */
1000 		if (ocfs2_write_should_remove_suid(inode)) {
1001 			if (meta_level == 0) {
1002 				ocfs2_meta_unlock(inode, meta_level);
1003 				meta_level = 1;
1004 				continue;
1005 			}
1006 
1007 			ret = ocfs2_write_remove_suid(inode);
1008 			if (ret < 0) {
1009 				mlog_errno(ret);
1010 				goto out;
1011 			}
1012 		}
1013 
1014 		/* work on a copy of ppos until we're sure that we won't have
1015 		 * to recalculate it due to relocking. */
1016 		if (filp->f_flags & O_APPEND) {
1017 			saved_pos = i_size_read(inode);
1018 			mlog(0, "O_APPEND: inode->i_size=%llu\n", saved_pos);
1019 		} else {
1020 			saved_pos = iocb->ki_pos;
1021 		}
1022 		newsize = count + saved_pos;
1023 
1024 		mlog(0, "pos=%lld newsize=%"MLFu64" cursize=%lld\n",
1025 		     saved_pos, newsize, i_size_read(inode));
1026 
1027 		/* No need for a higher level metadata lock if we're
1028 		 * never going past i_size. */
1029 		if (newsize <= i_size_read(inode))
1030 			break;
1031 
1032 		if (meta_level == 0) {
1033 			ocfs2_meta_unlock(inode, meta_level);
1034 			meta_level = 1;
1035 			continue;
1036 		}
1037 
1038 		spin_lock(&OCFS2_I(inode)->ip_lock);
1039 		clusters = ocfs2_clusters_for_bytes(inode->i_sb, newsize) -
1040 			OCFS2_I(inode)->ip_clusters;
1041 		spin_unlock(&OCFS2_I(inode)->ip_lock);
1042 
1043 		mlog(0, "Writing at EOF, may need more allocation: "
1044 		     "i_size = %lld, newsize = %"MLFu64", need %u clusters\n",
1045 		     i_size_read(inode), newsize, clusters);
1046 
1047 		/* We only want to continue the rest of this loop if
1048 		 * our extend will actually require more
1049 		 * allocation. */
1050 		if (!clusters)
1051 			break;
1052 
1053 		ret = ocfs2_extend_allocation(inode, clusters);
1054 		if (ret < 0) {
1055 			if (ret != -ENOSPC)
1056 				mlog_errno(ret);
1057 			goto out;
1058 		}
1059 
1060 		/* Fill any holes which would've been created by this
1061 		 * write. If we're O_APPEND, this will wind up
1062 		 * (correctly) being a noop. */
1063 		ret = ocfs2_zero_extend(inode, (u64) newsize - count);
1064 		if (ret < 0) {
1065 			mlog_errno(ret);
1066 			goto out;
1067 		}
1068 		break;
1069 	}
1070 
1071 	/* ok, we're done with i_size and alloc work */
1072 	iocb->ki_pos = saved_pos;
1073 	ocfs2_meta_unlock(inode, meta_level);
1074 	meta_level = -1;
1075 
1076 	/* communicate with ocfs2_dio_end_io */
1077 	ocfs2_iocb_set_rw_locked(iocb);
1078 
1079 #ifdef OCFS2_ORACORE_WORKAROUNDS
1080 	if (osb->s_mount_opt & OCFS2_MOUNT_COMPAT_OCFS &&
1081 	    filp->f_flags & O_DIRECT) {
1082 		unsigned int saved_flags = filp->f_flags;
1083 		int sector_size = 1 << osb->s_sectsize_bits;
1084 
1085 		if ((saved_pos & (sector_size - 1)) ||
1086 		    (count & (sector_size - 1)) ||
1087 		    ((unsigned long)buf & (sector_size - 1))) {
1088 			filp->f_flags |= O_SYNC;
1089 			filp->f_flags &= ~O_DIRECT;
1090 		}
1091 
1092 		ret = generic_file_aio_write_nolock(iocb, &local_iov, 1,
1093 						    &iocb->ki_pos);
1094 
1095 		filp->f_flags = saved_flags;
1096 	} else
1097 #endif
1098 		ret = generic_file_aio_write_nolock(iocb, &local_iov, 1,
1099 						    &iocb->ki_pos);
1100 
1101 	/* buffered aio wouldn't have proper lock coverage today */
1102 	BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT));
1103 
1104 	/*
1105 	 * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
1106 	 * function pointer which is called when o_direct io completes so that
1107 	 * it can unlock our rw lock.  (it's the clustered equivalent of
1108 	 * i_alloc_sem; protects truncate from racing with pending ios).
1109 	 * Unfortunately there are error cases which call end_io and others
1110 	 * that don't.  so we don't have to unlock the rw_lock if either an
1111 	 * async dio is going to do it in the future or an end_io after an
1112 	 * error has already done it.
1113 	 */
1114 	if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
1115 		rw_level = -1;
1116 		have_alloc_sem = 0;
1117 	}
1118 
1119 out:
1120 	if (meta_level != -1)
1121 		ocfs2_meta_unlock(inode, meta_level);
1122 	if (have_alloc_sem)
1123 		up_read(&inode->i_alloc_sem);
1124 	if (rw_level != -1)
1125 		ocfs2_rw_unlock(inode, rw_level);
1126 	mutex_unlock(&inode->i_mutex);
1127 
1128 	mlog_exit(ret);
1129 	return ret;
1130 }
1131 
1132 static ssize_t ocfs2_file_aio_read(struct kiocb *iocb,
1133 				   char __user *buf,
1134 				   size_t count,
1135 				   loff_t pos)
1136 {
1137 	int ret = 0, rw_level = -1, have_alloc_sem = 0;
1138 	struct file *filp = iocb->ki_filp;
1139 	struct inode *inode = filp->f_dentry->d_inode;
1140 #ifdef OCFS2_ORACORE_WORKAROUNDS
1141 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1142 #endif
1143 
1144 	mlog_entry("(0x%p, 0x%p, %u, '%.*s')\n", filp, buf,
1145 		   (unsigned int)count,
1146 		   filp->f_dentry->d_name.len,
1147 		   filp->f_dentry->d_name.name);
1148 
1149 	if (!inode) {
1150 		ret = -EINVAL;
1151 		mlog_errno(ret);
1152 		goto bail;
1153 	}
1154 
1155 #ifdef OCFS2_ORACORE_WORKAROUNDS
1156 	if (osb->s_mount_opt & OCFS2_MOUNT_COMPAT_OCFS) {
1157 		if (filp->f_flags & O_DIRECT) {
1158 			int sector_size = 1 << osb->s_sectsize_bits;
1159 
1160 			if ((pos & (sector_size - 1)) ||
1161 			    (count & (sector_size - 1)) ||
1162 			    ((unsigned long)buf & (sector_size - 1)) ||
1163 			    (i_size_read(inode) & (sector_size -1))) {
1164 				filp->f_flags &= ~O_DIRECT;
1165 			}
1166 		}
1167 	}
1168 #endif
1169 
1170 	/*
1171 	 * buffered reads protect themselves in ->readpage().  O_DIRECT reads
1172 	 * need locks to protect pending reads from racing with truncate.
1173 	 */
1174 	if (filp->f_flags & O_DIRECT) {
1175 		down_read(&inode->i_alloc_sem);
1176 		have_alloc_sem = 1;
1177 
1178 		ret = ocfs2_rw_lock(inode, 0);
1179 		if (ret < 0) {
1180 			mlog_errno(ret);
1181 			goto bail;
1182 		}
1183 		rw_level = 0;
1184 		/* communicate with ocfs2_dio_end_io */
1185 		ocfs2_iocb_set_rw_locked(iocb);
1186 	}
1187 
1188 	ret = generic_file_aio_read(iocb, buf, count, iocb->ki_pos);
1189 	if (ret == -EINVAL)
1190 		mlog(ML_ERROR, "generic_file_aio_read returned -EINVAL\n");
1191 
1192 	/* buffered aio wouldn't have proper lock coverage today */
1193 	BUG_ON(ret == -EIOCBQUEUED && !(filp->f_flags & O_DIRECT));
1194 
1195 	/* see ocfs2_file_aio_write */
1196 	if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
1197 		rw_level = -1;
1198 		have_alloc_sem = 0;
1199 	}
1200 
1201 bail:
1202 	if (have_alloc_sem)
1203 		up_read(&inode->i_alloc_sem);
1204 	if (rw_level != -1)
1205 		ocfs2_rw_unlock(inode, rw_level);
1206 	mlog_exit(ret);
1207 
1208 	return ret;
1209 }
1210 
1211 struct inode_operations ocfs2_file_iops = {
1212 	.setattr	= ocfs2_setattr,
1213 	.getattr	= ocfs2_getattr,
1214 };
1215 
1216 struct inode_operations ocfs2_special_file_iops = {
1217 	.setattr	= ocfs2_setattr,
1218 	.getattr	= ocfs2_getattr,
1219 };
1220 
1221 struct file_operations ocfs2_fops = {
1222 	.read		= do_sync_read,
1223 	.write		= do_sync_write,
1224 	.sendfile	= generic_file_sendfile,
1225 	.mmap		= ocfs2_mmap,
1226 	.fsync		= ocfs2_sync_file,
1227 	.release	= ocfs2_file_release,
1228 	.open		= ocfs2_file_open,
1229 	.aio_read	= ocfs2_file_aio_read,
1230 	.aio_write	= ocfs2_file_aio_write,
1231 };
1232 
1233 struct file_operations ocfs2_dops = {
1234 	.read		= generic_read_dir,
1235 	.readdir	= ocfs2_readdir,
1236 	.fsync		= ocfs2_sync_file,
1237 };
1238