xref: /openbmc/linux/fs/ocfs2/inode.c (revision 87c2ce3b)
1 /* -*- mode: c; c-basic-offset: 8; -*-
2  * vim: noexpandtab sw=8 ts=8 sts=0:
3  *
4  * inode.c
5  *
6  * vfs' aops, fops, dops and iops
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/smp_lock.h>
32 
33 #include <asm/byteorder.h>
34 
35 #define MLOG_MASK_PREFIX ML_INODE
36 #include <cluster/masklog.h>
37 
38 #include "ocfs2.h"
39 
40 #include "alloc.h"
41 #include "dlmglue.h"
42 #include "extent_map.h"
43 #include "file.h"
44 #include "inode.h"
45 #include "journal.h"
46 #include "namei.h"
47 #include "suballoc.h"
48 #include "super.h"
49 #include "symlink.h"
50 #include "sysfile.h"
51 #include "uptodate.h"
52 #include "vote.h"
53 
54 #include "buffer_head_io.h"
55 
56 #define OCFS2_FI_FLAG_NOWAIT	0x1
57 #define OCFS2_FI_FLAG_DELETE	0x2
58 struct ocfs2_find_inode_args
59 {
60 	u64		fi_blkno;
61 	unsigned long	fi_ino;
62 	unsigned int	fi_flags;
63 };
64 
65 static int ocfs2_read_locked_inode(struct inode *inode,
66 				   struct ocfs2_find_inode_args *args);
67 static int ocfs2_init_locked_inode(struct inode *inode, void *opaque);
68 static int ocfs2_find_actor(struct inode *inode, void *opaque);
69 static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
70 				    struct inode *inode,
71 				    struct buffer_head *fe_bh);
72 
73 struct inode *ocfs2_ilookup_for_vote(struct ocfs2_super *osb,
74 				     u64 blkno,
75 				     int delete_vote)
76 {
77 	struct ocfs2_find_inode_args args;
78 
79 	/* ocfs2_ilookup_for_vote should *only* be called from the
80 	 * vote thread */
81 	BUG_ON(current != osb->vote_task);
82 
83 	args.fi_blkno = blkno;
84 	args.fi_flags = OCFS2_FI_FLAG_NOWAIT;
85 	if (delete_vote)
86 		args.fi_flags |= OCFS2_FI_FLAG_DELETE;
87 	args.fi_ino = ino_from_blkno(osb->sb, blkno);
88 	return ilookup5(osb->sb, args.fi_ino, ocfs2_find_actor, &args);
89 }
90 
91 struct inode *ocfs2_iget(struct ocfs2_super *osb, u64 blkno)
92 {
93 	struct inode *inode = NULL;
94 	struct super_block *sb = osb->sb;
95 	struct ocfs2_find_inode_args args;
96 
97 	mlog_entry("(blkno = %"MLFu64")\n", blkno);
98 
99 	/* Ok. By now we've either got the offsets passed to us by the
100 	 * caller, or we just pulled them off the bh. Lets do some
101 	 * sanity checks to make sure they're OK. */
102 	if (blkno == 0) {
103 		inode = ERR_PTR(-EINVAL);
104 		mlog_errno(PTR_ERR(inode));
105 		goto bail;
106 	}
107 
108 	args.fi_blkno = blkno;
109 	args.fi_flags = 0;
110 	args.fi_ino = ino_from_blkno(sb, blkno);
111 
112 	inode = iget5_locked(sb, args.fi_ino, ocfs2_find_actor,
113 			     ocfs2_init_locked_inode, &args);
114 	/* inode was *not* in the inode cache. 2.6.x requires
115 	 * us to do our own read_inode call and unlock it
116 	 * afterwards. */
117 	if (inode && inode->i_state & I_NEW) {
118 		mlog(0, "Inode was not in inode cache, reading it.\n");
119 		ocfs2_read_locked_inode(inode, &args);
120 		unlock_new_inode(inode);
121 	}
122 	if (inode == NULL) {
123 		inode = ERR_PTR(-ENOMEM);
124 		mlog_errno(PTR_ERR(inode));
125 		goto bail;
126 	}
127 	if (is_bad_inode(inode)) {
128 		iput(inode);
129 		inode = ERR_PTR(-ESTALE);
130 		mlog_errno(PTR_ERR(inode));
131 		goto bail;
132 	}
133 
134 bail:
135 	if (!IS_ERR(inode)) {
136 		mlog(0, "returning inode with number %"MLFu64"\n",
137 		     OCFS2_I(inode)->ip_blkno);
138 		mlog_exit_ptr(inode);
139 	} else
140 		mlog_errno(PTR_ERR(inode));
141 
142 	return inode;
143 }
144 
145 
146 /*
147  * here's how inodes get read from disk:
148  * iget5_locked -> find_actor -> OCFS2_FIND_ACTOR
149  * found? : return the in-memory inode
150  * not found? : get_new_inode -> OCFS2_INIT_LOCKED_INODE
151  */
152 
153 static int ocfs2_find_actor(struct inode *inode, void *opaque)
154 {
155 	struct ocfs2_find_inode_args *args = NULL;
156 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
157 	int ret = 0;
158 
159 	mlog_entry("(0x%p, %lu, 0x%p)\n", inode, inode->i_ino, opaque);
160 
161 	args = opaque;
162 
163 	mlog_bug_on_msg(!inode, "No inode in find actor!\n");
164 
165 	if (oi->ip_blkno != args->fi_blkno)
166 		goto bail;
167 
168 	/* OCFS2_FI_FLAG_NOWAIT is *only* set from
169 	 * ocfs2_ilookup_for_vote which won't create an inode for one
170 	 * that isn't found. The vote thread which doesn't want to get
171 	 * an inode which is in the process of going away - otherwise
172 	 * the call to __wait_on_freeing_inode in find_inode_fast will
173 	 * cause it to deadlock on an inode which may be waiting on a
174 	 * vote (or lock release) in delete_inode */
175 	if ((args->fi_flags & OCFS2_FI_FLAG_NOWAIT) &&
176 	    (inode->i_state & (I_FREEING|I_CLEAR))) {
177 		/* As stated above, we're not going to return an
178 		 * inode.  In the case of a delete vote, the voting
179 		 * code is going to signal the other node to go
180 		 * ahead. Mark that state here, so this freeing inode
181 		 * has the state when it gets to delete_inode. */
182 		if (args->fi_flags & OCFS2_FI_FLAG_DELETE) {
183 			spin_lock(&oi->ip_lock);
184 			ocfs2_mark_inode_remotely_deleted(inode);
185 			spin_unlock(&oi->ip_lock);
186 		}
187 		goto bail;
188 	}
189 
190 	ret = 1;
191 bail:
192 	mlog_exit(ret);
193 	return ret;
194 }
195 
196 /*
197  * initialize the new inode, but don't do anything that would cause
198  * us to sleep.
199  * return 0 on success, 1 on failure
200  */
201 static int ocfs2_init_locked_inode(struct inode *inode, void *opaque)
202 {
203 	struct ocfs2_find_inode_args *args = opaque;
204 
205 	mlog_entry("inode = %p, opaque = %p\n", inode, opaque);
206 
207 	inode->i_ino = args->fi_ino;
208 	OCFS2_I(inode)->ip_blkno = args->fi_blkno;
209 
210 	mlog_exit(0);
211 	return 0;
212 }
213 
214 int ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe,
215 		     	 int create_ino)
216 {
217 	struct super_block *sb;
218 	struct ocfs2_super *osb;
219 	int status = -EINVAL;
220 
221 	mlog_entry("(0x%p, size:%"MLFu64")\n", inode, fe->i_size);
222 
223 	sb = inode->i_sb;
224 	osb = OCFS2_SB(sb);
225 
226 	/* this means that read_inode cannot create a superblock inode
227 	 * today.  change if needed. */
228 	if (!OCFS2_IS_VALID_DINODE(fe) ||
229 	    !(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL))) {
230 		mlog(ML_ERROR, "Invalid dinode: i_ino=%lu, i_blkno=%"MLFu64", "
231 		     "signature = %.*s, flags = 0x%x\n",
232 		     inode->i_ino, le64_to_cpu(fe->i_blkno), 7,
233 		     fe->i_signature, le32_to_cpu(fe->i_flags));
234 		goto bail;
235 	}
236 
237 	if (le32_to_cpu(fe->i_fs_generation) != osb->fs_generation) {
238 		mlog(ML_ERROR, "file entry generation does not match "
239 		     "superblock! osb->fs_generation=%x, "
240 		     "fe->i_fs_generation=%x\n",
241 		     osb->fs_generation, le32_to_cpu(fe->i_fs_generation));
242 		goto bail;
243 	}
244 
245 	inode->i_version = 1;
246 	inode->i_generation = le32_to_cpu(fe->i_generation);
247 	inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
248 	inode->i_mode = le16_to_cpu(fe->i_mode);
249 	inode->i_uid = le32_to_cpu(fe->i_uid);
250 	inode->i_gid = le32_to_cpu(fe->i_gid);
251 	inode->i_blksize = (u32)osb->s_clustersize;
252 
253 	/* Fast symlinks will have i_size but no allocated clusters. */
254 	if (S_ISLNK(inode->i_mode) && !fe->i_clusters)
255 		inode->i_blocks = 0;
256 	else
257 		inode->i_blocks =
258 			ocfs2_align_bytes_to_sectors(le64_to_cpu(fe->i_size));
259 	inode->i_mapping->a_ops = &ocfs2_aops;
260 	inode->i_flags |= S_NOATIME;
261 	inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
262 	inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
263 	inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
264 	inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
265 	inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
266 	inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
267 
268 	if (OCFS2_I(inode)->ip_blkno != le64_to_cpu(fe->i_blkno))
269 		mlog(ML_ERROR,
270 		     "ip_blkno %"MLFu64" != i_blkno %"MLFu64"!\n",
271 		     OCFS2_I(inode)->ip_blkno, fe->i_blkno);
272 
273 	OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
274 	OCFS2_I(inode)->ip_orphaned_slot = OCFS2_INVALID_SLOT;
275 
276 	if (create_ino)
277 		inode->i_ino = ino_from_blkno(inode->i_sb,
278 			       le64_to_cpu(fe->i_blkno));
279 
280 	mlog(0, "blkno = %"MLFu64", ino = %lu, create_ino = %s\n",
281 	     fe->i_blkno, inode->i_ino, create_ino ? "true" : "false");
282 
283 	inode->i_nlink = le16_to_cpu(fe->i_links_count);
284 
285 	if (fe->i_flags & cpu_to_le32(OCFS2_LOCAL_ALLOC_FL)) {
286 		OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
287 		mlog(0, "local alloc inode: i_ino=%lu\n", inode->i_ino);
288 	} else if (fe->i_flags & cpu_to_le32(OCFS2_BITMAP_FL)) {
289 		OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
290 	} else if (fe->i_flags & cpu_to_le32(OCFS2_SUPER_BLOCK_FL)) {
291 		mlog(0, "superblock inode: i_ino=%lu\n", inode->i_ino);
292 		/* we can't actually hit this as read_inode can't
293 		 * handle superblocks today ;-) */
294 		BUG();
295 	}
296 
297 	switch (inode->i_mode & S_IFMT) {
298 	    case S_IFREG:
299 		    inode->i_fop = &ocfs2_fops;
300 		    inode->i_op = &ocfs2_file_iops;
301 		    i_size_write(inode, le64_to_cpu(fe->i_size));
302 		    break;
303 	    case S_IFDIR:
304 		    inode->i_op = &ocfs2_dir_iops;
305 		    inode->i_fop = &ocfs2_dops;
306 		    i_size_write(inode, le64_to_cpu(fe->i_size));
307 		    break;
308 	    case S_IFLNK:
309 		    if (ocfs2_inode_is_fast_symlink(inode))
310 			inode->i_op = &ocfs2_fast_symlink_inode_operations;
311 		    else
312 			inode->i_op = &ocfs2_symlink_inode_operations;
313 		    i_size_write(inode, le64_to_cpu(fe->i_size));
314 		    break;
315 	    default:
316 		    inode->i_op = &ocfs2_special_file_iops;
317 		    init_special_inode(inode, inode->i_mode,
318 				       inode->i_rdev);
319 		    break;
320 	}
321 
322 	ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_rw_lockres,
323 				  OCFS2_LOCK_TYPE_RW, inode);
324 	ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_meta_lockres,
325 				  OCFS2_LOCK_TYPE_META, inode);
326 	ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_data_lockres,
327 				  OCFS2_LOCK_TYPE_DATA, inode);
328 
329 	status = 0;
330 bail:
331 	mlog_exit(status);
332 	return status;
333 }
334 
335 static int ocfs2_read_locked_inode(struct inode *inode,
336 				   struct ocfs2_find_inode_args *args)
337 {
338 	struct super_block *sb;
339 	struct ocfs2_super *osb;
340 	struct ocfs2_dinode *fe;
341 	struct buffer_head *bh = NULL;
342 	int status;
343 	int sysfile = 0;
344 
345 	mlog_entry("(0x%p, 0x%p)\n", inode, args);
346 
347 	status = -EINVAL;
348 	if (inode == NULL || inode->i_sb == NULL) {
349 		mlog(ML_ERROR, "bad inode\n");
350 		goto bail;
351 	}
352 	sb = inode->i_sb;
353 	osb = OCFS2_SB(sb);
354 
355 	if (!args) {
356 		mlog(ML_ERROR, "bad inode args\n");
357 		make_bad_inode(inode);
358 		goto bail;
359 	}
360 
361 	/* Read the FE off disk. This is safe because the kernel only
362 	 * does one read_inode2 for a new inode, and if it doesn't
363 	 * exist yet then nobody can be working on it! */
364 	status = ocfs2_read_block(osb, args->fi_blkno, &bh, 0, NULL);
365 	if (status < 0) {
366 		mlog_errno(status);
367 		make_bad_inode(inode);
368 		goto bail;
369 	}
370 
371 	fe = (struct ocfs2_dinode *) bh->b_data;
372 	if (!OCFS2_IS_VALID_DINODE(fe)) {
373 		mlog(ML_ERROR, "Invalid dinode #%"MLFu64": signature = %.*s\n",
374 		     fe->i_blkno, 7, fe->i_signature);
375 		make_bad_inode(inode);
376 		goto bail;
377 	}
378 
379 	if (fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL))
380 		sysfile = 1;
381 
382 	if (S_ISCHR(le16_to_cpu(fe->i_mode)) ||
383 	    S_ISBLK(le16_to_cpu(fe->i_mode)))
384     		inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
385 
386 	status = -EINVAL;
387 	if (ocfs2_populate_inode(inode, fe, 0) < 0) {
388 		mlog(ML_ERROR, "populate inode failed! i_blkno=%"MLFu64", "
389 		     "i_ino=%lu\n", fe->i_blkno, inode->i_ino);
390 		make_bad_inode(inode);
391 		goto bail;
392 	}
393 
394 	BUG_ON(args->fi_blkno != le64_to_cpu(fe->i_blkno));
395 
396 	if (sysfile)
397 	       OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SYSTEM_FILE;
398 
399 	status = 0;
400 
401 bail:
402 	if (args && bh)
403 		brelse(bh);
404 
405 	mlog_exit(status);
406 	return status;
407 }
408 
409 void ocfs2_sync_blockdev(struct super_block *sb)
410 {
411 	sync_blockdev(sb->s_bdev);
412 }
413 
414 static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
415 				     struct inode *inode,
416 				     struct buffer_head *fe_bh)
417 {
418 	int status = 0;
419 	struct ocfs2_journal_handle *handle = NULL;
420 	struct ocfs2_truncate_context *tc = NULL;
421 	struct ocfs2_dinode *fe;
422 
423 	mlog_entry_void();
424 
425 	fe = (struct ocfs2_dinode *) fe_bh->b_data;
426 
427 	/* zero allocation, zero truncate :) */
428 	if (!fe->i_clusters)
429 		goto bail;
430 
431 	handle = ocfs2_start_trans(osb, handle, OCFS2_INODE_UPDATE_CREDITS);
432 	if (IS_ERR(handle)) {
433 		status = PTR_ERR(handle);
434 		handle = NULL;
435 		mlog_errno(status);
436 		goto bail;
437 	}
438 
439 	status = ocfs2_set_inode_size(handle, inode, fe_bh, 0ULL);
440 	if (status < 0) {
441 		mlog_errno(status);
442 		goto bail;
443 	}
444 
445 	ocfs2_commit_trans(handle);
446 	handle = NULL;
447 
448 	status = ocfs2_prepare_truncate(osb, inode, fe_bh, &tc);
449 	if (status < 0) {
450 		mlog_errno(status);
451 		goto bail;
452 	}
453 
454 	status = ocfs2_commit_truncate(osb, inode, fe_bh, tc);
455 	if (status < 0) {
456 		mlog_errno(status);
457 		goto bail;
458 	}
459 bail:
460 	if (handle)
461 		ocfs2_commit_trans(handle);
462 
463 	mlog_exit(status);
464 	return status;
465 }
466 
467 static int ocfs2_remove_inode(struct inode *inode,
468 			      struct buffer_head *di_bh,
469 			      struct inode *orphan_dir_inode,
470 			      struct buffer_head *orphan_dir_bh)
471 {
472 	int status;
473 	struct inode *inode_alloc_inode = NULL;
474 	struct buffer_head *inode_alloc_bh = NULL;
475 	struct ocfs2_journal_handle *handle;
476 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
477 	struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
478 
479 	inode_alloc_inode =
480 		ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE,
481 					    le16_to_cpu(di->i_suballoc_slot));
482 	if (!inode_alloc_inode) {
483 		status = -EEXIST;
484 		mlog_errno(status);
485 		goto bail;
486 	}
487 
488 	mutex_lock(&inode_alloc_inode->i_mutex);
489 	status = ocfs2_meta_lock(inode_alloc_inode, NULL, &inode_alloc_bh, 1);
490 	if (status < 0) {
491 		mutex_unlock(&inode_alloc_inode->i_mutex);
492 
493 		mlog_errno(status);
494 		goto bail;
495 	}
496 
497 	handle = ocfs2_start_trans(osb, NULL, OCFS2_DELETE_INODE_CREDITS);
498 	if (IS_ERR(handle)) {
499 		status = PTR_ERR(handle);
500 		mlog_errno(status);
501 		goto bail_unlock;
502 	}
503 
504 	status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
505 				  orphan_dir_bh);
506 	if (status < 0) {
507 		mlog_errno(status);
508 		goto bail_commit;
509 	}
510 
511 	/* set the inodes dtime */
512 	status = ocfs2_journal_access(handle, inode, di_bh,
513 				      OCFS2_JOURNAL_ACCESS_WRITE);
514 	if (status < 0) {
515 		mlog_errno(status);
516 		goto bail_commit;
517 	}
518 
519 	di->i_dtime = cpu_to_le64(CURRENT_TIME.tv_sec);
520 	le32_and_cpu(&di->i_flags, ~(OCFS2_VALID_FL | OCFS2_ORPHANED_FL));
521 
522 	status = ocfs2_journal_dirty(handle, di_bh);
523 	if (status < 0) {
524 		mlog_errno(status);
525 		goto bail_commit;
526 	}
527 
528 	ocfs2_remove_from_cache(inode, di_bh);
529 
530 	status = ocfs2_free_dinode(handle, inode_alloc_inode,
531 				   inode_alloc_bh, di);
532 	if (status < 0)
533 		mlog_errno(status);
534 
535 bail_commit:
536 	ocfs2_commit_trans(handle);
537 bail_unlock:
538 	ocfs2_meta_unlock(inode_alloc_inode, 1);
539 	mutex_unlock(&inode_alloc_inode->i_mutex);
540 	brelse(inode_alloc_bh);
541 bail:
542 	iput(inode_alloc_inode);
543 
544 	return status;
545 }
546 
547 static int ocfs2_wipe_inode(struct inode *inode,
548 			    struct buffer_head *di_bh)
549 {
550 	int status, orphaned_slot;
551 	struct inode *orphan_dir_inode = NULL;
552 	struct buffer_head *orphan_dir_bh = NULL;
553 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
554 
555 	/* We've already voted on this so it should be readonly - no
556 	 * spinlock needed. */
557 	orphaned_slot = OCFS2_I(inode)->ip_orphaned_slot;
558 	orphan_dir_inode = ocfs2_get_system_file_inode(osb,
559 						       ORPHAN_DIR_SYSTEM_INODE,
560 						       orphaned_slot);
561 	if (!orphan_dir_inode) {
562 		status = -EEXIST;
563 		mlog_errno(status);
564 		goto bail;
565 	}
566 
567 	/* Lock the orphan dir. The lock will be held for the entire
568 	 * delete_inode operation. We do this now to avoid races with
569 	 * recovery completion on other nodes. */
570 	mutex_lock(&orphan_dir_inode->i_mutex);
571 	status = ocfs2_meta_lock(orphan_dir_inode, NULL, &orphan_dir_bh, 1);
572 	if (status < 0) {
573 		mutex_unlock(&orphan_dir_inode->i_mutex);
574 
575 		mlog_errno(status);
576 		goto bail;
577 	}
578 
579 	/* we do this while holding the orphan dir lock because we
580 	 * don't want recovery being run from another node to vote for
581 	 * an inode delete on us -- this will result in two nodes
582 	 * truncating the same file! */
583 	status = ocfs2_truncate_for_delete(osb, inode, di_bh);
584 	if (status < 0) {
585 		mlog_errno(status);
586 		goto bail_unlock_dir;
587 	}
588 
589 	status = ocfs2_remove_inode(inode, di_bh, orphan_dir_inode,
590 				    orphan_dir_bh);
591 	if (status < 0)
592 		mlog_errno(status);
593 
594 bail_unlock_dir:
595 	ocfs2_meta_unlock(orphan_dir_inode, 1);
596 	mutex_unlock(&orphan_dir_inode->i_mutex);
597 	brelse(orphan_dir_bh);
598 bail:
599 	iput(orphan_dir_inode);
600 
601 	return status;
602 }
603 
604 /* There is a series of simple checks that should be done before a
605  * vote is even considered. Encapsulate those in this function. */
606 static int ocfs2_inode_is_valid_to_delete(struct inode *inode)
607 {
608 	int ret = 0;
609 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
610 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
611 
612 	/* We shouldn't be getting here for the root directory
613 	 * inode.. */
614 	if (inode == osb->root_inode) {
615 		mlog(ML_ERROR, "Skipping delete of root inode.\n");
616 		goto bail;
617 	}
618 
619 	/* If we're coming from process_vote we can't go into our own
620 	 * voting [hello, deadlock city!], so unforuntately we just
621 	 * have to skip deleting this guy. That's OK though because
622 	 * the node who's doing the actual deleting should handle it
623 	 * anyway. */
624 	if (current == osb->vote_task) {
625 		mlog(0, "Skipping delete of %lu because we're currently "
626 		     "in process_vote\n", inode->i_ino);
627 		goto bail;
628 	}
629 
630 	spin_lock(&oi->ip_lock);
631 	/* OCFS2 *never* deletes system files. This should technically
632 	 * never get here as system file inodes should always have a
633 	 * positive link count. */
634 	if (oi->ip_flags & OCFS2_INODE_SYSTEM_FILE) {
635 		mlog(ML_ERROR, "Skipping delete of system file %"MLFu64".\n",
636 		     oi->ip_blkno);
637 		goto bail_unlock;
638 	}
639 
640 	/* If we have voted "yes" on the wipe of this inode for
641 	 * another node, it will be marked here so we can safely skip
642 	 * it. Recovery will cleanup any inodes we might inadvertantly
643 	 * skip here. */
644 	if (oi->ip_flags & OCFS2_INODE_SKIP_DELETE) {
645 		mlog(0, "Skipping delete of %lu because another node "
646 		     "has done this for us.\n", inode->i_ino);
647 		goto bail_unlock;
648 	}
649 
650 	ret = 1;
651 bail_unlock:
652 	spin_unlock(&oi->ip_lock);
653 bail:
654 	return ret;
655 }
656 
657 /* Query the cluster to determine whether we should wipe an inode from
658  * disk or not.
659  *
660  * Requires the inode to have the cluster lock. */
661 static int ocfs2_query_inode_wipe(struct inode *inode,
662 				  struct buffer_head *di_bh,
663 				  int *wipe)
664 {
665 	int status = 0;
666 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
667 	struct ocfs2_dinode *di;
668 
669 	*wipe = 0;
670 
671 	/* While we were waiting for the cluster lock in
672 	 * ocfs2_delete_inode, another node might have asked to delete
673 	 * the inode. Recheck our flags to catch this. */
674 	if (!ocfs2_inode_is_valid_to_delete(inode)) {
675 		mlog(0, "Skipping delete of %"MLFu64" because flags changed\n",
676 		     oi->ip_blkno);
677 		goto bail;
678 	}
679 
680 	/* Now that we have an up to date inode, we can double check
681 	 * the link count. */
682 	if (inode->i_nlink) {
683 		mlog(0, "Skipping delete of %"MLFu64" because nlink = %u\n",
684 		     oi->ip_blkno, inode->i_nlink);
685 		goto bail;
686 	}
687 
688 	/* Do some basic inode verification... */
689 	di = (struct ocfs2_dinode *) di_bh->b_data;
690 	if (!(di->i_flags & cpu_to_le32(OCFS2_ORPHANED_FL))) {
691 		/* for lack of a better error? */
692 		status = -EEXIST;
693 		mlog(ML_ERROR,
694 		     "Inode %"MLFu64" (on-disk %"MLFu64") not orphaned! "
695 		     "Disk flags  0x%x, inode flags 0x%x\n",
696 		     oi->ip_blkno, di->i_blkno, di->i_flags, oi->ip_flags);
697 		goto bail;
698 	}
699 
700 	/* has someone already deleted us?! baaad... */
701 	if (di->i_dtime) {
702 		status = -EEXIST;
703 		mlog_errno(status);
704 		goto bail;
705 	}
706 
707 	status = ocfs2_request_delete_vote(inode);
708 	/* -EBUSY means that other nodes are still using the
709 	 * inode. We're done here though, so avoid doing anything on
710 	 * disk and let them worry about deleting it. */
711 	if (status == -EBUSY) {
712 		status = 0;
713 		mlog(0, "Skipping delete of %"MLFu64" because it is in use on"
714 		     "other nodes\n", oi->ip_blkno);
715 		goto bail;
716 	}
717 	if (status < 0) {
718 		mlog_errno(status);
719 		goto bail;
720 	}
721 
722 	spin_lock(&oi->ip_lock);
723 	if (oi->ip_orphaned_slot == OCFS2_INVALID_SLOT) {
724 		/* Nobody knew which slot this inode was orphaned
725 		 * into. This may happen during node death and
726 		 * recovery knows how to clean it up so we can safely
727 		 * ignore this inode for now on. */
728 		mlog(0, "Nobody knew where inode %"MLFu64" was orphaned!\n",
729 		     oi->ip_blkno);
730 	} else {
731 		*wipe = 1;
732 
733 		mlog(0, "Inode %"MLFu64" is ok to wipe from orphan dir %d\n",
734 		     oi->ip_blkno, oi->ip_orphaned_slot);
735 	}
736 	spin_unlock(&oi->ip_lock);
737 
738 bail:
739 	return status;
740 }
741 
742 /* Support function for ocfs2_delete_inode. Will help us keep the
743  * inode data in a consistent state for clear_inode. Always truncates
744  * pages, optionally sync's them first. */
745 static void ocfs2_cleanup_delete_inode(struct inode *inode,
746 				       int sync_data)
747 {
748 	mlog(0, "Cleanup inode %"MLFu64", sync = %d\n",
749 	     OCFS2_I(inode)->ip_blkno, sync_data);
750 	if (sync_data)
751 		write_inode_now(inode, 1);
752 	truncate_inode_pages(&inode->i_data, 0);
753 }
754 
755 void ocfs2_delete_inode(struct inode *inode)
756 {
757 	int wipe, status;
758 	sigset_t blocked, oldset;
759 	struct buffer_head *di_bh = NULL;
760 
761 	mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
762 
763 	if (is_bad_inode(inode)) {
764 		mlog(0, "Skipping delete of bad inode\n");
765 		goto bail;
766 	}
767 
768 	if (!ocfs2_inode_is_valid_to_delete(inode)) {
769 		/* It's probably not necessary to truncate_inode_pages
770 		 * here but we do it for safety anyway (it will most
771 		 * likely be a no-op anyway) */
772 		ocfs2_cleanup_delete_inode(inode, 0);
773 		goto bail;
774 	}
775 
776 	/* We want to block signals in delete_inode as the lock and
777 	 * messaging paths may return us -ERESTARTSYS. Which would
778 	 * cause us to exit early, resulting in inodes being orphaned
779 	 * forever. */
780 	sigfillset(&blocked);
781 	status = sigprocmask(SIG_BLOCK, &blocked, &oldset);
782 	if (status < 0) {
783 		mlog_errno(status);
784 		ocfs2_cleanup_delete_inode(inode, 1);
785 		goto bail;
786 	}
787 
788 	/* Lock down the inode. This gives us an up to date view of
789 	 * it's metadata (for verification), and allows us to
790 	 * serialize delete_inode votes.
791 	 *
792 	 * Even though we might be doing a truncate, we don't take the
793 	 * allocation lock here as it won't be needed - nobody will
794 	 * have the file open.
795 	 */
796 	status = ocfs2_meta_lock(inode, NULL, &di_bh, 1);
797 	if (status < 0) {
798 		if (status != -ENOENT)
799 			mlog_errno(status);
800 		ocfs2_cleanup_delete_inode(inode, 0);
801 		goto bail_unblock;
802 	}
803 
804 	/* Query the cluster. This will be the final decision made
805 	 * before we go ahead and wipe the inode. */
806 	status = ocfs2_query_inode_wipe(inode, di_bh, &wipe);
807 	if (!wipe || status < 0) {
808 		/* Error and inode busy vote both mean we won't be
809 		 * removing the inode, so they take almost the same
810 		 * path. */
811 		if (status < 0)
812 			mlog_errno(status);
813 
814 		/* Someone in the cluster has voted to not wipe this
815 		 * inode, or it was never completely orphaned. Write
816 		 * out the pages and exit now. */
817 		ocfs2_cleanup_delete_inode(inode, 1);
818 		goto bail_unlock_inode;
819 	}
820 
821 	ocfs2_cleanup_delete_inode(inode, 0);
822 
823 	status = ocfs2_wipe_inode(inode, di_bh);
824 	if (status < 0) {
825 		mlog_errno(status);
826 		goto bail_unlock_inode;
827 	}
828 
829 	/* Mark the inode as successfully deleted. This is important
830 	 * for ocfs2_clear_inode as it will check this flag and skip
831 	 * any checkpointing work */
832 	OCFS2_I(inode)->ip_flags |= OCFS2_INODE_DELETED;
833 
834 bail_unlock_inode:
835 	ocfs2_meta_unlock(inode, 1);
836 	brelse(di_bh);
837 bail_unblock:
838 	status = sigprocmask(SIG_SETMASK, &oldset, NULL);
839 	if (status < 0)
840 		mlog_errno(status);
841 bail:
842 	clear_inode(inode);
843 	mlog_exit_void();
844 }
845 
846 void ocfs2_clear_inode(struct inode *inode)
847 {
848 	int status;
849 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
850 
851 	mlog_entry_void();
852 
853 	if (!inode)
854 		goto bail;
855 
856 	mlog(0, "Clearing inode: %"MLFu64", nlink = %u\n",
857 	     OCFS2_I(inode)->ip_blkno, inode->i_nlink);
858 
859 	mlog_bug_on_msg(OCFS2_SB(inode->i_sb) == NULL,
860 			"Inode=%lu\n", inode->i_ino);
861 
862 	/* Do these before all the other work so that we don't bounce
863 	 * the vote thread while waiting to destroy the locks. */
864 	ocfs2_mark_lockres_freeing(&oi->ip_rw_lockres);
865 	ocfs2_mark_lockres_freeing(&oi->ip_meta_lockres);
866 	ocfs2_mark_lockres_freeing(&oi->ip_data_lockres);
867 
868 	/* We very well may get a clear_inode before all an inodes
869 	 * metadata has hit disk. Of course, we can't drop any cluster
870 	 * locks until the journal has finished with it. The only
871 	 * exception here are successfully wiped inodes - their
872 	 * metadata can now be considered to be part of the system
873 	 * inodes from which it came. */
874 	if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED))
875 		ocfs2_checkpoint_inode(inode);
876 
877 	mlog_bug_on_msg(!list_empty(&oi->ip_io_markers),
878 			"Clear inode of %"MLFu64", inode has io markers\n",
879 			oi->ip_blkno);
880 
881 	ocfs2_extent_map_drop(inode, 0);
882 	ocfs2_extent_map_init(inode);
883 
884 	status = ocfs2_drop_inode_locks(inode);
885 	if (status < 0)
886 		mlog_errno(status);
887 
888 	ocfs2_lock_res_free(&oi->ip_rw_lockres);
889 	ocfs2_lock_res_free(&oi->ip_meta_lockres);
890 	ocfs2_lock_res_free(&oi->ip_data_lockres);
891 
892 	ocfs2_metadata_cache_purge(inode);
893 
894 	mlog_bug_on_msg(oi->ip_metadata_cache.ci_num_cached,
895 			"Clear inode of %"MLFu64", inode has %u cache items\n",
896 			oi->ip_blkno, oi->ip_metadata_cache.ci_num_cached);
897 
898 	mlog_bug_on_msg(!(oi->ip_flags & OCFS2_INODE_CACHE_INLINE),
899 			"Clear inode of %"MLFu64", inode has a bad flag\n",
900 			oi->ip_blkno);
901 
902 	mlog_bug_on_msg(spin_is_locked(&oi->ip_lock),
903 			"Clear inode of %"MLFu64", inode is locked\n",
904 			oi->ip_blkno);
905 
906 	mlog_bug_on_msg(down_trylock(&oi->ip_io_sem),
907 			"Clear inode of %"MLFu64", io_sem is locked\n",
908 			oi->ip_blkno);
909 	up(&oi->ip_io_sem);
910 
911 	/*
912 	 * down_trylock() returns 0, down_write_trylock() returns 1
913 	 * kernel 1, world 0
914 	 */
915 	mlog_bug_on_msg(!down_write_trylock(&oi->ip_alloc_sem),
916 			"Clear inode of %"MLFu64", alloc_sem is locked\n",
917 			oi->ip_blkno);
918 	up_write(&oi->ip_alloc_sem);
919 
920 	mlog_bug_on_msg(oi->ip_open_count,
921 			"Clear inode of %"MLFu64" has open count %d\n",
922 			oi->ip_blkno, oi->ip_open_count);
923 	mlog_bug_on_msg(!list_empty(&oi->ip_handle_list),
924 			"Clear inode of %"MLFu64" has non empty handle list\n",
925 			oi->ip_blkno);
926 	mlog_bug_on_msg(oi->ip_handle,
927 			"Clear inode of %"MLFu64" has non empty handle pointer\n",
928 			oi->ip_blkno);
929 
930 	/* Clear all other flags. */
931 	oi->ip_flags = OCFS2_INODE_CACHE_INLINE;
932 	oi->ip_created_trans = 0;
933 	oi->ip_last_trans = 0;
934 	oi->ip_dir_start_lookup = 0;
935 	oi->ip_blkno = 0ULL;
936 
937 bail:
938 	mlog_exit_void();
939 }
940 
941 /* Called under inode_lock, with no more references on the
942  * struct inode, so it's safe here to check the flags field
943  * and to manipulate i_nlink without any other locks. */
944 void ocfs2_drop_inode(struct inode *inode)
945 {
946 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
947 
948 	mlog_entry_void();
949 
950 	mlog(0, "Drop inode %"MLFu64", nlink = %u, ip_flags = 0x%x\n",
951 	     oi->ip_blkno, inode->i_nlink, oi->ip_flags);
952 
953 	/* Testing ip_orphaned_slot here wouldn't work because we may
954 	 * not have gotten a delete_inode vote from any other nodes
955 	 * yet. */
956 	if (oi->ip_flags & OCFS2_INODE_MAYBE_ORPHANED) {
957 		mlog(0, "Inode was orphaned on another node, clearing nlink.\n");
958 		inode->i_nlink = 0;
959 	}
960 
961 	generic_drop_inode(inode);
962 
963 	mlog_exit_void();
964 }
965 
966 /*
967  * TODO: this should probably be merged into ocfs2_get_block
968  *
969  * However, you now need to pay attention to the cont_prepare_write()
970  * stuff in ocfs2_get_block (that is, ocfs2_get_block pretty much
971  * expects never to extend).
972  */
973 struct buffer_head *ocfs2_bread(struct inode *inode,
974 				int block, int *err, int reada)
975 {
976 	struct buffer_head *bh = NULL;
977 	int tmperr;
978 	u64 p_blkno;
979 	int readflags = OCFS2_BH_CACHED;
980 
981 #if 0
982 	/* only turn this on if we know we can deal with read_block
983 	 * returning nothing */
984 	if (reada)
985 		readflags |= OCFS2_BH_READAHEAD;
986 #endif
987 
988 	if (((u64)block << inode->i_sb->s_blocksize_bits) >=
989 	    i_size_read(inode)) {
990 		BUG_ON(!reada);
991 		return NULL;
992 	}
993 
994 	tmperr = ocfs2_extent_map_get_blocks(inode, block, 1,
995 					     &p_blkno, NULL);
996 	if (tmperr < 0) {
997 		mlog_errno(tmperr);
998 		goto fail;
999 	}
1000 
1001 	tmperr = ocfs2_read_block(OCFS2_SB(inode->i_sb), p_blkno, &bh,
1002 				  readflags, inode);
1003 	if (tmperr < 0)
1004 		goto fail;
1005 
1006 	tmperr = 0;
1007 
1008 	*err = 0;
1009 	return bh;
1010 
1011 fail:
1012 	if (bh) {
1013 		brelse(bh);
1014 		bh = NULL;
1015 	}
1016 	*err = -EIO;
1017 	return NULL;
1018 }
1019 
1020 /*
1021  * This is called from our getattr.
1022  */
1023 int ocfs2_inode_revalidate(struct dentry *dentry)
1024 {
1025 	struct inode *inode = dentry->d_inode;
1026 	int status = 0;
1027 
1028 	mlog_entry("(inode = 0x%p, ino = %"MLFu64")\n", inode,
1029 		   inode ? OCFS2_I(inode)->ip_blkno : 0ULL);
1030 
1031 	if (!inode) {
1032 		mlog(0, "eep, no inode!\n");
1033 		status = -ENOENT;
1034 		goto bail;
1035 	}
1036 
1037 	spin_lock(&OCFS2_I(inode)->ip_lock);
1038 	if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
1039 		spin_unlock(&OCFS2_I(inode)->ip_lock);
1040 		mlog(0, "inode deleted!\n");
1041 		status = -ENOENT;
1042 		goto bail;
1043 	}
1044 	spin_unlock(&OCFS2_I(inode)->ip_lock);
1045 
1046 	/* Let ocfs2_meta_lock do the work of updating our struct
1047 	 * inode for us. */
1048 	status = ocfs2_meta_lock(inode, NULL, NULL, 0);
1049 	if (status < 0) {
1050 		if (status != -ENOENT)
1051 			mlog_errno(status);
1052 		goto bail;
1053 	}
1054 	ocfs2_meta_unlock(inode, 0);
1055 bail:
1056 	mlog_exit(status);
1057 
1058 	return status;
1059 }
1060 
1061 /*
1062  * Updates a disk inode from a
1063  * struct inode.
1064  * Only takes ip_lock.
1065  */
1066 int ocfs2_mark_inode_dirty(struct ocfs2_journal_handle *handle,
1067 			   struct inode *inode,
1068 			   struct buffer_head *bh)
1069 {
1070 	int status;
1071 	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
1072 
1073 	mlog_entry("(inode %"MLFu64")\n", OCFS2_I(inode)->ip_blkno);
1074 
1075 	status = ocfs2_journal_access(handle, inode, bh,
1076 				      OCFS2_JOURNAL_ACCESS_WRITE);
1077 	if (status < 0) {
1078 		mlog_errno(status);
1079 		goto leave;
1080 	}
1081 
1082 	spin_lock(&OCFS2_I(inode)->ip_lock);
1083 	fe->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters);
1084 	spin_unlock(&OCFS2_I(inode)->ip_lock);
1085 
1086 	fe->i_size = cpu_to_le64(i_size_read(inode));
1087 	fe->i_links_count = cpu_to_le16(inode->i_nlink);
1088 	fe->i_uid = cpu_to_le32(inode->i_uid);
1089 	fe->i_gid = cpu_to_le32(inode->i_gid);
1090 	fe->i_mode = cpu_to_le16(inode->i_mode);
1091 	fe->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
1092 	fe->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
1093 	fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
1094 	fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
1095 	fe->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
1096 	fe->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
1097 
1098 	status = ocfs2_journal_dirty(handle, bh);
1099 	if (status < 0)
1100 		mlog_errno(status);
1101 
1102 	status = 0;
1103 leave:
1104 
1105 	mlog_exit(status);
1106 	return status;
1107 }
1108 
1109 /*
1110  *
1111  * Updates a struct inode from a disk inode.
1112  * does no i/o, only takes ip_lock.
1113  */
1114 void ocfs2_refresh_inode(struct inode *inode,
1115 			 struct ocfs2_dinode *fe)
1116 {
1117 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1118 
1119 	spin_lock(&OCFS2_I(inode)->ip_lock);
1120 
1121 	OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
1122 	i_size_write(inode, le64_to_cpu(fe->i_size));
1123 	inode->i_nlink = le16_to_cpu(fe->i_links_count);
1124 	inode->i_uid = le32_to_cpu(fe->i_uid);
1125 	inode->i_gid = le32_to_cpu(fe->i_gid);
1126 	inode->i_mode = le16_to_cpu(fe->i_mode);
1127 	inode->i_blksize = (u32) osb->s_clustersize;
1128 	if (S_ISLNK(inode->i_mode) && le32_to_cpu(fe->i_clusters) == 0)
1129 		inode->i_blocks = 0;
1130 	else
1131 		inode->i_blocks = ocfs2_align_bytes_to_sectors(i_size_read(inode));
1132 	inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
1133 	inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
1134 	inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
1135 	inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
1136 	inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
1137 	inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
1138 
1139 	spin_unlock(&OCFS2_I(inode)->ip_lock);
1140 }
1141