ia64/linux-2.6.18-xen.hg

view fs/ocfs2/inode.c @ 524:7f8b544237bf

netfront: Allow netfront in domain 0.

This is useful if your physical network device is in a utility domain.

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
author Keir Fraser <keir.fraser@citrix.com>
date Tue Apr 15 15:18:58 2008 +0100 (2008-04-15)
parents 831230e53067
children
line source
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 */
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>
33 #include <asm/byteorder.h>
35 #define MLOG_MASK_PREFIX ML_INODE
36 #include <cluster/masklog.h>
38 #include "ocfs2.h"
40 #include "alloc.h"
41 #include "dlmglue.h"
42 #include "extent_map.h"
43 #include "file.h"
44 #include "heartbeat.h"
45 #include "inode.h"
46 #include "journal.h"
47 #include "namei.h"
48 #include "suballoc.h"
49 #include "super.h"
50 #include "symlink.h"
51 #include "sysfile.h"
52 #include "uptodate.h"
53 #include "vote.h"
55 #include "buffer_head_io.h"
57 #define OCFS2_FI_FLAG_NOWAIT 0x1
58 #define OCFS2_FI_FLAG_DELETE 0x2
59 struct ocfs2_find_inode_args
60 {
61 u64 fi_blkno;
62 unsigned long fi_ino;
63 unsigned int fi_flags;
64 };
66 static int ocfs2_read_locked_inode(struct inode *inode,
67 struct ocfs2_find_inode_args *args);
68 static int ocfs2_init_locked_inode(struct inode *inode, void *opaque);
69 static int ocfs2_find_actor(struct inode *inode, void *opaque);
70 static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
71 struct inode *inode,
72 struct buffer_head *fe_bh);
74 struct inode *ocfs2_ilookup_for_vote(struct ocfs2_super *osb,
75 u64 blkno,
76 int delete_vote)
77 {
78 struct ocfs2_find_inode_args args;
80 /* ocfs2_ilookup_for_vote should *only* be called from the
81 * vote thread */
82 BUG_ON(current != osb->vote_task);
84 args.fi_blkno = blkno;
85 args.fi_flags = OCFS2_FI_FLAG_NOWAIT;
86 if (delete_vote)
87 args.fi_flags |= OCFS2_FI_FLAG_DELETE;
88 args.fi_ino = ino_from_blkno(osb->sb, blkno);
89 return ilookup5(osb->sb, args.fi_ino, ocfs2_find_actor, &args);
90 }
92 struct inode *ocfs2_iget(struct ocfs2_super *osb, u64 blkno)
93 {
94 struct inode *inode = NULL;
95 struct super_block *sb = osb->sb;
96 struct ocfs2_find_inode_args args;
98 mlog_entry("(blkno = %llu)\n", (unsigned long long)blkno);
100 /* Ok. By now we've either got the offsets passed to us by the
101 * caller, or we just pulled them off the bh. Lets do some
102 * sanity checks to make sure they're OK. */
103 if (blkno == 0) {
104 inode = ERR_PTR(-EINVAL);
105 mlog_errno(PTR_ERR(inode));
106 goto bail;
107 }
109 args.fi_blkno = blkno;
110 args.fi_flags = 0;
111 args.fi_ino = ino_from_blkno(sb, blkno);
113 inode = iget5_locked(sb, args.fi_ino, ocfs2_find_actor,
114 ocfs2_init_locked_inode, &args);
115 /* inode was *not* in the inode cache. 2.6.x requires
116 * us to do our own read_inode call and unlock it
117 * afterwards. */
118 if (inode && inode->i_state & I_NEW) {
119 mlog(0, "Inode was not in inode cache, reading it.\n");
120 ocfs2_read_locked_inode(inode, &args);
121 unlock_new_inode(inode);
122 }
123 if (inode == NULL) {
124 inode = ERR_PTR(-ENOMEM);
125 mlog_errno(PTR_ERR(inode));
126 goto bail;
127 }
128 if (is_bad_inode(inode)) {
129 iput(inode);
130 inode = ERR_PTR(-ESTALE);
131 mlog_errno(PTR_ERR(inode));
132 goto bail;
133 }
135 bail:
136 if (!IS_ERR(inode)) {
137 mlog(0, "returning inode with number %llu\n",
138 (unsigned long long)OCFS2_I(inode)->ip_blkno);
139 mlog_exit_ptr(inode);
140 } else
141 mlog_errno(PTR_ERR(inode));
143 return inode;
144 }
147 /*
148 * here's how inodes get read from disk:
149 * iget5_locked -> find_actor -> OCFS2_FIND_ACTOR
150 * found? : return the in-memory inode
151 * not found? : get_new_inode -> OCFS2_INIT_LOCKED_INODE
152 */
154 static int ocfs2_find_actor(struct inode *inode, void *opaque)
155 {
156 struct ocfs2_find_inode_args *args = NULL;
157 struct ocfs2_inode_info *oi = OCFS2_I(inode);
158 int ret = 0;
160 mlog_entry("(0x%p, %lu, 0x%p)\n", inode, inode->i_ino, opaque);
162 args = opaque;
164 mlog_bug_on_msg(!inode, "No inode in find actor!\n");
166 if (oi->ip_blkno != args->fi_blkno)
167 goto bail;
169 /* OCFS2_FI_FLAG_NOWAIT is *only* set from
170 * ocfs2_ilookup_for_vote which won't create an inode for one
171 * that isn't found. The vote thread which doesn't want to get
172 * an inode which is in the process of going away - otherwise
173 * the call to __wait_on_freeing_inode in find_inode_fast will
174 * cause it to deadlock on an inode which may be waiting on a
175 * vote (or lock release) in delete_inode */
176 if ((args->fi_flags & OCFS2_FI_FLAG_NOWAIT) &&
177 (inode->i_state & (I_FREEING|I_CLEAR))) {
178 /* As stated above, we're not going to return an
179 * inode. In the case of a delete vote, the voting
180 * code is going to signal the other node to go
181 * ahead. Mark that state here, so this freeing inode
182 * has the state when it gets to delete_inode. */
183 if (args->fi_flags & OCFS2_FI_FLAG_DELETE) {
184 spin_lock(&oi->ip_lock);
185 ocfs2_mark_inode_remotely_deleted(inode);
186 spin_unlock(&oi->ip_lock);
187 }
188 goto bail;
189 }
191 ret = 1;
192 bail:
193 mlog_exit(ret);
194 return ret;
195 }
197 /*
198 * initialize the new inode, but don't do anything that would cause
199 * us to sleep.
200 * return 0 on success, 1 on failure
201 */
202 static int ocfs2_init_locked_inode(struct inode *inode, void *opaque)
203 {
204 struct ocfs2_find_inode_args *args = opaque;
206 mlog_entry("inode = %p, opaque = %p\n", inode, opaque);
208 inode->i_ino = args->fi_ino;
209 OCFS2_I(inode)->ip_blkno = args->fi_blkno;
211 mlog_exit(0);
212 return 0;
213 }
215 int ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe,
216 int create_ino)
217 {
218 struct super_block *sb;
219 struct ocfs2_super *osb;
220 int status = -EINVAL;
222 mlog_entry("(0x%p, size:%llu)\n", inode,
223 (unsigned long long)fe->i_size);
225 sb = inode->i_sb;
226 osb = OCFS2_SB(sb);
228 /* this means that read_inode cannot create a superblock inode
229 * today. change if needed. */
230 if (!OCFS2_IS_VALID_DINODE(fe) ||
231 !(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL))) {
232 mlog(ML_ERROR, "Invalid dinode: i_ino=%lu, i_blkno=%llu, "
233 "signature = %.*s, flags = 0x%x\n",
234 inode->i_ino,
235 (unsigned long long)le64_to_cpu(fe->i_blkno), 7,
236 fe->i_signature, le32_to_cpu(fe->i_flags));
237 goto bail;
238 }
240 if (le32_to_cpu(fe->i_fs_generation) != osb->fs_generation) {
241 mlog(ML_ERROR, "file entry generation does not match "
242 "superblock! osb->fs_generation=%x, "
243 "fe->i_fs_generation=%x\n",
244 osb->fs_generation, le32_to_cpu(fe->i_fs_generation));
245 goto bail;
246 }
248 inode->i_version = 1;
249 inode->i_generation = le32_to_cpu(fe->i_generation);
250 inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
251 inode->i_mode = le16_to_cpu(fe->i_mode);
252 inode->i_uid = le32_to_cpu(fe->i_uid);
253 inode->i_gid = le32_to_cpu(fe->i_gid);
254 inode->i_blksize = (u32)osb->s_clustersize;
256 /* Fast symlinks will have i_size but no allocated clusters. */
257 if (S_ISLNK(inode->i_mode) && !fe->i_clusters)
258 inode->i_blocks = 0;
259 else
260 inode->i_blocks =
261 ocfs2_align_bytes_to_sectors(le64_to_cpu(fe->i_size));
262 inode->i_mapping->a_ops = &ocfs2_aops;
263 inode->i_flags |= S_NOATIME;
264 inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
265 inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
266 inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
267 inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
268 inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
269 inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
271 if (OCFS2_I(inode)->ip_blkno != le64_to_cpu(fe->i_blkno))
272 mlog(ML_ERROR,
273 "ip_blkno %llu != i_blkno %llu!\n",
274 (unsigned long long)OCFS2_I(inode)->ip_blkno,
275 (unsigned long long)fe->i_blkno);
277 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
278 OCFS2_I(inode)->ip_orphaned_slot = OCFS2_INVALID_SLOT;
280 if (create_ino)
281 inode->i_ino = ino_from_blkno(inode->i_sb,
282 le64_to_cpu(fe->i_blkno));
284 mlog(0, "blkno = %llu, ino = %lu, create_ino = %s\n",
285 (unsigned long long)fe->i_blkno, inode->i_ino, create_ino ? "true" : "false");
287 inode->i_nlink = le16_to_cpu(fe->i_links_count);
289 if (fe->i_flags & cpu_to_le32(OCFS2_LOCAL_ALLOC_FL)) {
290 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
291 mlog(0, "local alloc inode: i_ino=%lu\n", inode->i_ino);
292 } else if (fe->i_flags & cpu_to_le32(OCFS2_BITMAP_FL)) {
293 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
294 } else if (fe->i_flags & cpu_to_le32(OCFS2_SUPER_BLOCK_FL)) {
295 mlog(0, "superblock inode: i_ino=%lu\n", inode->i_ino);
296 /* we can't actually hit this as read_inode can't
297 * handle superblocks today ;-) */
298 BUG();
299 }
301 switch (inode->i_mode & S_IFMT) {
302 case S_IFREG:
303 inode->i_fop = &ocfs2_fops;
304 inode->i_op = &ocfs2_file_iops;
305 i_size_write(inode, le64_to_cpu(fe->i_size));
306 break;
307 case S_IFDIR:
308 inode->i_op = &ocfs2_dir_iops;
309 inode->i_fop = &ocfs2_dops;
310 i_size_write(inode, le64_to_cpu(fe->i_size));
311 break;
312 case S_IFLNK:
313 if (ocfs2_inode_is_fast_symlink(inode))
314 inode->i_op = &ocfs2_fast_symlink_inode_operations;
315 else
316 inode->i_op = &ocfs2_symlink_inode_operations;
317 i_size_write(inode, le64_to_cpu(fe->i_size));
318 break;
319 default:
320 inode->i_op = &ocfs2_special_file_iops;
321 init_special_inode(inode, inode->i_mode,
322 inode->i_rdev);
323 break;
324 }
326 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_rw_lockres,
327 OCFS2_LOCK_TYPE_RW, inode);
328 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_meta_lockres,
329 OCFS2_LOCK_TYPE_META, inode);
330 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_data_lockres,
331 OCFS2_LOCK_TYPE_DATA, inode);
333 status = 0;
334 bail:
335 mlog_exit(status);
336 return status;
337 }
339 static int ocfs2_read_locked_inode(struct inode *inode,
340 struct ocfs2_find_inode_args *args)
341 {
342 struct super_block *sb;
343 struct ocfs2_super *osb;
344 struct ocfs2_dinode *fe;
345 struct buffer_head *bh = NULL;
346 int status;
347 int sysfile = 0;
349 mlog_entry("(0x%p, 0x%p)\n", inode, args);
351 status = -EINVAL;
352 if (inode == NULL || inode->i_sb == NULL) {
353 mlog(ML_ERROR, "bad inode\n");
354 goto bail;
355 }
356 sb = inode->i_sb;
357 osb = OCFS2_SB(sb);
359 if (!args) {
360 mlog(ML_ERROR, "bad inode args\n");
361 make_bad_inode(inode);
362 goto bail;
363 }
365 /* Read the FE off disk. This is safe because the kernel only
366 * does one read_inode2 for a new inode, and if it doesn't
367 * exist yet then nobody can be working on it! */
368 status = ocfs2_read_block(osb, args->fi_blkno, &bh, 0, NULL);
369 if (status < 0) {
370 mlog_errno(status);
371 make_bad_inode(inode);
372 goto bail;
373 }
375 fe = (struct ocfs2_dinode *) bh->b_data;
376 if (!OCFS2_IS_VALID_DINODE(fe)) {
377 mlog(ML_ERROR, "Invalid dinode #%llu: signature = %.*s\n",
378 (unsigned long long)fe->i_blkno, 7, fe->i_signature);
379 make_bad_inode(inode);
380 goto bail;
381 }
383 if (fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL))
384 sysfile = 1;
386 if (S_ISCHR(le16_to_cpu(fe->i_mode)) ||
387 S_ISBLK(le16_to_cpu(fe->i_mode)))
388 inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
390 status = -EINVAL;
391 if (ocfs2_populate_inode(inode, fe, 0) < 0) {
392 mlog(ML_ERROR, "populate failed! i_blkno=%llu, i_ino=%lu\n",
393 (unsigned long long)fe->i_blkno, inode->i_ino);
394 make_bad_inode(inode);
395 goto bail;
396 }
398 BUG_ON(args->fi_blkno != le64_to_cpu(fe->i_blkno));
400 if (sysfile)
401 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SYSTEM_FILE;
403 status = 0;
405 bail:
406 if (args && bh)
407 brelse(bh);
409 mlog_exit(status);
410 return status;
411 }
413 void ocfs2_sync_blockdev(struct super_block *sb)
414 {
415 sync_blockdev(sb->s_bdev);
416 }
418 static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
419 struct inode *inode,
420 struct buffer_head *fe_bh)
421 {
422 int status = 0;
423 struct ocfs2_journal_handle *handle = NULL;
424 struct ocfs2_truncate_context *tc = NULL;
425 struct ocfs2_dinode *fe;
427 mlog_entry_void();
429 fe = (struct ocfs2_dinode *) fe_bh->b_data;
431 /* zero allocation, zero truncate :) */
432 if (!fe->i_clusters)
433 goto bail;
435 handle = ocfs2_start_trans(osb, handle, OCFS2_INODE_UPDATE_CREDITS);
436 if (IS_ERR(handle)) {
437 status = PTR_ERR(handle);
438 handle = NULL;
439 mlog_errno(status);
440 goto bail;
441 }
443 status = ocfs2_set_inode_size(handle, inode, fe_bh, 0ULL);
444 if (status < 0) {
445 mlog_errno(status);
446 goto bail;
447 }
449 ocfs2_commit_trans(handle);
450 handle = NULL;
452 status = ocfs2_prepare_truncate(osb, inode, fe_bh, &tc);
453 if (status < 0) {
454 mlog_errno(status);
455 goto bail;
456 }
458 status = ocfs2_commit_truncate(osb, inode, fe_bh, tc);
459 if (status < 0) {
460 mlog_errno(status);
461 goto bail;
462 }
463 bail:
464 if (handle)
465 ocfs2_commit_trans(handle);
467 mlog_exit(status);
468 return status;
469 }
471 static int ocfs2_remove_inode(struct inode *inode,
472 struct buffer_head *di_bh,
473 struct inode *orphan_dir_inode,
474 struct buffer_head *orphan_dir_bh)
475 {
476 int status;
477 struct inode *inode_alloc_inode = NULL;
478 struct buffer_head *inode_alloc_bh = NULL;
479 struct ocfs2_journal_handle *handle;
480 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
481 struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
483 inode_alloc_inode =
484 ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE,
485 le16_to_cpu(di->i_suballoc_slot));
486 if (!inode_alloc_inode) {
487 status = -EEXIST;
488 mlog_errno(status);
489 goto bail;
490 }
492 mutex_lock(&inode_alloc_inode->i_mutex);
493 status = ocfs2_meta_lock(inode_alloc_inode, NULL, &inode_alloc_bh, 1);
494 if (status < 0) {
495 mutex_unlock(&inode_alloc_inode->i_mutex);
497 mlog_errno(status);
498 goto bail;
499 }
501 handle = ocfs2_start_trans(osb, NULL, OCFS2_DELETE_INODE_CREDITS);
502 if (IS_ERR(handle)) {
503 status = PTR_ERR(handle);
504 mlog_errno(status);
505 goto bail_unlock;
506 }
508 status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
509 orphan_dir_bh);
510 if (status < 0) {
511 mlog_errno(status);
512 goto bail_commit;
513 }
515 /* set the inodes dtime */
516 status = ocfs2_journal_access(handle, inode, di_bh,
517 OCFS2_JOURNAL_ACCESS_WRITE);
518 if (status < 0) {
519 mlog_errno(status);
520 goto bail_commit;
521 }
523 di->i_dtime = cpu_to_le64(CURRENT_TIME.tv_sec);
524 le32_and_cpu(&di->i_flags, ~(OCFS2_VALID_FL | OCFS2_ORPHANED_FL));
526 status = ocfs2_journal_dirty(handle, di_bh);
527 if (status < 0) {
528 mlog_errno(status);
529 goto bail_commit;
530 }
532 ocfs2_remove_from_cache(inode, di_bh);
534 status = ocfs2_free_dinode(handle, inode_alloc_inode,
535 inode_alloc_bh, di);
536 if (status < 0)
537 mlog_errno(status);
539 bail_commit:
540 ocfs2_commit_trans(handle);
541 bail_unlock:
542 ocfs2_meta_unlock(inode_alloc_inode, 1);
543 mutex_unlock(&inode_alloc_inode->i_mutex);
544 brelse(inode_alloc_bh);
545 bail:
546 iput(inode_alloc_inode);
548 return status;
549 }
551 /*
552 * Serialize with orphan dir recovery. If the process doing
553 * recovery on this orphan dir does an iget() with the dir
554 * i_mutex held, we'll deadlock here. Instead we detect this
555 * and exit early - recovery will wipe this inode for us.
556 */
557 static int ocfs2_check_orphan_recovery_state(struct ocfs2_super *osb,
558 int slot)
559 {
560 int ret = 0;
562 spin_lock(&osb->osb_lock);
563 if (ocfs2_node_map_test_bit(osb, &osb->osb_recovering_orphan_dirs, slot)) {
564 mlog(0, "Recovery is happening on orphan dir %d, will skip "
565 "this inode\n", slot);
566 ret = -EDEADLK;
567 goto out;
568 }
569 /* This signals to the orphan recovery process that it should
570 * wait for us to handle the wipe. */
571 osb->osb_orphan_wipes[slot]++;
572 out:
573 spin_unlock(&osb->osb_lock);
574 return ret;
575 }
577 static void ocfs2_signal_wipe_completion(struct ocfs2_super *osb,
578 int slot)
579 {
580 spin_lock(&osb->osb_lock);
581 osb->osb_orphan_wipes[slot]--;
582 spin_unlock(&osb->osb_lock);
584 wake_up(&osb->osb_wipe_event);
585 }
587 static int ocfs2_wipe_inode(struct inode *inode,
588 struct buffer_head *di_bh)
589 {
590 int status, orphaned_slot;
591 struct inode *orphan_dir_inode = NULL;
592 struct buffer_head *orphan_dir_bh = NULL;
593 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
595 /* We've already voted on this so it should be readonly - no
596 * spinlock needed. */
597 orphaned_slot = OCFS2_I(inode)->ip_orphaned_slot;
599 status = ocfs2_check_orphan_recovery_state(osb, orphaned_slot);
600 if (status)
601 return status;
603 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
604 ORPHAN_DIR_SYSTEM_INODE,
605 orphaned_slot);
606 if (!orphan_dir_inode) {
607 status = -EEXIST;
608 mlog_errno(status);
609 goto bail;
610 }
612 /* Lock the orphan dir. The lock will be held for the entire
613 * delete_inode operation. We do this now to avoid races with
614 * recovery completion on other nodes. */
615 mutex_lock(&orphan_dir_inode->i_mutex);
616 status = ocfs2_meta_lock(orphan_dir_inode, NULL, &orphan_dir_bh, 1);
617 if (status < 0) {
618 mutex_unlock(&orphan_dir_inode->i_mutex);
620 mlog_errno(status);
621 goto bail;
622 }
624 /* we do this while holding the orphan dir lock because we
625 * don't want recovery being run from another node to vote for
626 * an inode delete on us -- this will result in two nodes
627 * truncating the same file! */
628 status = ocfs2_truncate_for_delete(osb, inode, di_bh);
629 if (status < 0) {
630 mlog_errno(status);
631 goto bail_unlock_dir;
632 }
634 status = ocfs2_remove_inode(inode, di_bh, orphan_dir_inode,
635 orphan_dir_bh);
636 if (status < 0)
637 mlog_errno(status);
639 bail_unlock_dir:
640 ocfs2_meta_unlock(orphan_dir_inode, 1);
641 mutex_unlock(&orphan_dir_inode->i_mutex);
642 brelse(orphan_dir_bh);
643 bail:
644 iput(orphan_dir_inode);
645 ocfs2_signal_wipe_completion(osb, orphaned_slot);
647 return status;
648 }
650 /* There is a series of simple checks that should be done before a
651 * vote is even considered. Encapsulate those in this function. */
652 static int ocfs2_inode_is_valid_to_delete(struct inode *inode)
653 {
654 int ret = 0;
655 struct ocfs2_inode_info *oi = OCFS2_I(inode);
656 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
658 /* We shouldn't be getting here for the root directory
659 * inode.. */
660 if (inode == osb->root_inode) {
661 mlog(ML_ERROR, "Skipping delete of root inode.\n");
662 goto bail;
663 }
665 /* If we're coming from process_vote we can't go into our own
666 * voting [hello, deadlock city!], so unforuntately we just
667 * have to skip deleting this guy. That's OK though because
668 * the node who's doing the actual deleting should handle it
669 * anyway. */
670 if (current == osb->vote_task) {
671 mlog(0, "Skipping delete of %lu because we're currently "
672 "in process_vote\n", inode->i_ino);
673 goto bail;
674 }
676 spin_lock(&oi->ip_lock);
677 /* OCFS2 *never* deletes system files. This should technically
678 * never get here as system file inodes should always have a
679 * positive link count. */
680 if (oi->ip_flags & OCFS2_INODE_SYSTEM_FILE) {
681 mlog(ML_ERROR, "Skipping delete of system file %llu\n",
682 (unsigned long long)oi->ip_blkno);
683 goto bail_unlock;
684 }
686 /* If we have voted "yes" on the wipe of this inode for
687 * another node, it will be marked here so we can safely skip
688 * it. Recovery will cleanup any inodes we might inadvertantly
689 * skip here. */
690 if (oi->ip_flags & OCFS2_INODE_SKIP_DELETE) {
691 mlog(0, "Skipping delete of %lu because another node "
692 "has done this for us.\n", inode->i_ino);
693 goto bail_unlock;
694 }
696 ret = 1;
697 bail_unlock:
698 spin_unlock(&oi->ip_lock);
699 bail:
700 return ret;
701 }
703 /* Query the cluster to determine whether we should wipe an inode from
704 * disk or not.
705 *
706 * Requires the inode to have the cluster lock. */
707 static int ocfs2_query_inode_wipe(struct inode *inode,
708 struct buffer_head *di_bh,
709 int *wipe)
710 {
711 int status = 0;
712 struct ocfs2_inode_info *oi = OCFS2_I(inode);
713 struct ocfs2_dinode *di;
715 *wipe = 0;
717 /* While we were waiting for the cluster lock in
718 * ocfs2_delete_inode, another node might have asked to delete
719 * the inode. Recheck our flags to catch this. */
720 if (!ocfs2_inode_is_valid_to_delete(inode)) {
721 mlog(0, "Skipping delete of %llu because flags changed\n",
722 (unsigned long long)oi->ip_blkno);
723 goto bail;
724 }
726 /* Now that we have an up to date inode, we can double check
727 * the link count. */
728 if (inode->i_nlink) {
729 mlog(0, "Skipping delete of %llu because nlink = %u\n",
730 (unsigned long long)oi->ip_blkno, inode->i_nlink);
731 goto bail;
732 }
734 /* Do some basic inode verification... */
735 di = (struct ocfs2_dinode *) di_bh->b_data;
736 if (!(di->i_flags & cpu_to_le32(OCFS2_ORPHANED_FL))) {
737 /* for lack of a better error? */
738 status = -EEXIST;
739 mlog(ML_ERROR,
740 "Inode %llu (on-disk %llu) not orphaned! "
741 "Disk flags 0x%x, inode flags 0x%x\n",
742 (unsigned long long)oi->ip_blkno,
743 (unsigned long long)di->i_blkno, di->i_flags,
744 oi->ip_flags);
745 goto bail;
746 }
748 /* has someone already deleted us?! baaad... */
749 if (di->i_dtime) {
750 status = -EEXIST;
751 mlog_errno(status);
752 goto bail;
753 }
755 status = ocfs2_request_delete_vote(inode);
756 /* -EBUSY means that other nodes are still using the
757 * inode. We're done here though, so avoid doing anything on
758 * disk and let them worry about deleting it. */
759 if (status == -EBUSY) {
760 status = 0;
761 mlog(0, "Skipping delete of %llu because it is in use on"
762 "other nodes\n", (unsigned long long)oi->ip_blkno);
763 goto bail;
764 }
765 if (status < 0) {
766 mlog_errno(status);
767 goto bail;
768 }
770 spin_lock(&oi->ip_lock);
771 if (oi->ip_orphaned_slot == OCFS2_INVALID_SLOT) {
772 /* Nobody knew which slot this inode was orphaned
773 * into. This may happen during node death and
774 * recovery knows how to clean it up so we can safely
775 * ignore this inode for now on. */
776 mlog(0, "Nobody knew where inode %llu was orphaned!\n",
777 (unsigned long long)oi->ip_blkno);
778 } else {
779 *wipe = 1;
781 mlog(0, "Inode %llu is ok to wipe from orphan dir %d\n",
782 (unsigned long long)oi->ip_blkno, oi->ip_orphaned_slot);
783 }
784 spin_unlock(&oi->ip_lock);
786 bail:
787 return status;
788 }
790 /* Support function for ocfs2_delete_inode. Will help us keep the
791 * inode data in a consistent state for clear_inode. Always truncates
792 * pages, optionally sync's them first. */
793 static void ocfs2_cleanup_delete_inode(struct inode *inode,
794 int sync_data)
795 {
796 mlog(0, "Cleanup inode %llu, sync = %d\n",
797 (unsigned long long)OCFS2_I(inode)->ip_blkno, sync_data);
798 if (sync_data)
799 write_inode_now(inode, 1);
800 truncate_inode_pages(&inode->i_data, 0);
801 }
803 void ocfs2_delete_inode(struct inode *inode)
804 {
805 int wipe, status;
806 sigset_t blocked, oldset;
807 struct buffer_head *di_bh = NULL;
809 mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
811 if (is_bad_inode(inode)) {
812 mlog(0, "Skipping delete of bad inode\n");
813 goto bail;
814 }
816 if (!ocfs2_inode_is_valid_to_delete(inode)) {
817 /* It's probably not necessary to truncate_inode_pages
818 * here but we do it for safety anyway (it will most
819 * likely be a no-op anyway) */
820 ocfs2_cleanup_delete_inode(inode, 0);
821 goto bail;
822 }
824 /* We want to block signals in delete_inode as the lock and
825 * messaging paths may return us -ERESTARTSYS. Which would
826 * cause us to exit early, resulting in inodes being orphaned
827 * forever. */
828 sigfillset(&blocked);
829 status = sigprocmask(SIG_BLOCK, &blocked, &oldset);
830 if (status < 0) {
831 mlog_errno(status);
832 ocfs2_cleanup_delete_inode(inode, 1);
833 goto bail;
834 }
836 /* Lock down the inode. This gives us an up to date view of
837 * it's metadata (for verification), and allows us to
838 * serialize delete_inode votes.
839 *
840 * Even though we might be doing a truncate, we don't take the
841 * allocation lock here as it won't be needed - nobody will
842 * have the file open.
843 */
844 status = ocfs2_meta_lock(inode, NULL, &di_bh, 1);
845 if (status < 0) {
846 if (status != -ENOENT)
847 mlog_errno(status);
848 ocfs2_cleanup_delete_inode(inode, 0);
849 goto bail_unblock;
850 }
852 /* Query the cluster. This will be the final decision made
853 * before we go ahead and wipe the inode. */
854 status = ocfs2_query_inode_wipe(inode, di_bh, &wipe);
855 if (!wipe || status < 0) {
856 /* Error and inode busy vote both mean we won't be
857 * removing the inode, so they take almost the same
858 * path. */
859 if (status < 0)
860 mlog_errno(status);
862 /* Someone in the cluster has voted to not wipe this
863 * inode, or it was never completely orphaned. Write
864 * out the pages and exit now. */
865 ocfs2_cleanup_delete_inode(inode, 1);
866 goto bail_unlock_inode;
867 }
869 ocfs2_cleanup_delete_inode(inode, 0);
871 status = ocfs2_wipe_inode(inode, di_bh);
872 if (status < 0) {
873 if (status != -EDEADLK)
874 mlog_errno(status);
875 goto bail_unlock_inode;
876 }
878 /* Mark the inode as successfully deleted. This is important
879 * for ocfs2_clear_inode as it will check this flag and skip
880 * any checkpointing work */
881 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_DELETED;
883 bail_unlock_inode:
884 ocfs2_meta_unlock(inode, 1);
885 brelse(di_bh);
886 bail_unblock:
887 status = sigprocmask(SIG_SETMASK, &oldset, NULL);
888 if (status < 0)
889 mlog_errno(status);
890 bail:
891 clear_inode(inode);
892 mlog_exit_void();
893 }
895 void ocfs2_clear_inode(struct inode *inode)
896 {
897 int status;
898 struct ocfs2_inode_info *oi = OCFS2_I(inode);
900 mlog_entry_void();
902 if (!inode)
903 goto bail;
905 mlog(0, "Clearing inode: %llu, nlink = %u\n",
906 (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_nlink);
908 mlog_bug_on_msg(OCFS2_SB(inode->i_sb) == NULL,
909 "Inode=%lu\n", inode->i_ino);
911 /* Do these before all the other work so that we don't bounce
912 * the vote thread while waiting to destroy the locks. */
913 ocfs2_mark_lockres_freeing(&oi->ip_rw_lockres);
914 ocfs2_mark_lockres_freeing(&oi->ip_meta_lockres);
915 ocfs2_mark_lockres_freeing(&oi->ip_data_lockres);
917 /* We very well may get a clear_inode before all an inodes
918 * metadata has hit disk. Of course, we can't drop any cluster
919 * locks until the journal has finished with it. The only
920 * exception here are successfully wiped inodes - their
921 * metadata can now be considered to be part of the system
922 * inodes from which it came. */
923 if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED))
924 ocfs2_checkpoint_inode(inode);
926 mlog_bug_on_msg(!list_empty(&oi->ip_io_markers),
927 "Clear inode of %llu, inode has io markers\n",
928 (unsigned long long)oi->ip_blkno);
930 ocfs2_extent_map_drop(inode, 0);
931 ocfs2_extent_map_init(inode);
933 status = ocfs2_drop_inode_locks(inode);
934 if (status < 0)
935 mlog_errno(status);
937 ocfs2_lock_res_free(&oi->ip_rw_lockres);
938 ocfs2_lock_res_free(&oi->ip_meta_lockres);
939 ocfs2_lock_res_free(&oi->ip_data_lockres);
941 ocfs2_metadata_cache_purge(inode);
943 mlog_bug_on_msg(oi->ip_metadata_cache.ci_num_cached,
944 "Clear inode of %llu, inode has %u cache items\n",
945 (unsigned long long)oi->ip_blkno, oi->ip_metadata_cache.ci_num_cached);
947 mlog_bug_on_msg(!(oi->ip_flags & OCFS2_INODE_CACHE_INLINE),
948 "Clear inode of %llu, inode has a bad flag\n",
949 (unsigned long long)oi->ip_blkno);
951 mlog_bug_on_msg(spin_is_locked(&oi->ip_lock),
952 "Clear inode of %llu, inode is locked\n",
953 (unsigned long long)oi->ip_blkno);
955 mlog_bug_on_msg(!mutex_trylock(&oi->ip_io_mutex),
956 "Clear inode of %llu, io_mutex is locked\n",
957 (unsigned long long)oi->ip_blkno);
958 mutex_unlock(&oi->ip_io_mutex);
960 /*
961 * down_trylock() returns 0, down_write_trylock() returns 1
962 * kernel 1, world 0
963 */
964 mlog_bug_on_msg(!down_write_trylock(&oi->ip_alloc_sem),
965 "Clear inode of %llu, alloc_sem is locked\n",
966 (unsigned long long)oi->ip_blkno);
967 up_write(&oi->ip_alloc_sem);
969 mlog_bug_on_msg(oi->ip_open_count,
970 "Clear inode of %llu has open count %d\n",
971 (unsigned long long)oi->ip_blkno, oi->ip_open_count);
972 mlog_bug_on_msg(!list_empty(&oi->ip_handle_list),
973 "Clear inode of %llu has non empty handle list\n",
974 (unsigned long long)oi->ip_blkno);
975 mlog_bug_on_msg(oi->ip_handle,
976 "Clear inode of %llu has non empty handle pointer\n",
977 (unsigned long long)oi->ip_blkno);
979 /* Clear all other flags. */
980 oi->ip_flags = OCFS2_INODE_CACHE_INLINE;
981 oi->ip_created_trans = 0;
982 oi->ip_last_trans = 0;
983 oi->ip_dir_start_lookup = 0;
984 oi->ip_blkno = 0ULL;
986 bail:
987 mlog_exit_void();
988 }
990 /* Called under inode_lock, with no more references on the
991 * struct inode, so it's safe here to check the flags field
992 * and to manipulate i_nlink without any other locks. */
993 void ocfs2_drop_inode(struct inode *inode)
994 {
995 struct ocfs2_inode_info *oi = OCFS2_I(inode);
997 mlog_entry_void();
999 mlog(0, "Drop inode %llu, nlink = %u, ip_flags = 0x%x\n",
1000 (unsigned long long)oi->ip_blkno, inode->i_nlink, oi->ip_flags);
1002 /* Testing ip_orphaned_slot here wouldn't work because we may
1003 * not have gotten a delete_inode vote from any other nodes
1004 * yet. */
1005 if (oi->ip_flags & OCFS2_INODE_MAYBE_ORPHANED) {
1006 mlog(0, "Inode was orphaned on another node, clearing nlink.\n");
1007 inode->i_nlink = 0;
1010 generic_drop_inode(inode);
1012 mlog_exit_void();
1015 /*
1016 * TODO: this should probably be merged into ocfs2_get_block
1018 * However, you now need to pay attention to the cont_prepare_write()
1019 * stuff in ocfs2_get_block (that is, ocfs2_get_block pretty much
1020 * expects never to extend).
1021 */
1022 struct buffer_head *ocfs2_bread(struct inode *inode,
1023 int block, int *err, int reada)
1025 struct buffer_head *bh = NULL;
1026 int tmperr;
1027 u64 p_blkno;
1028 int readflags = OCFS2_BH_CACHED;
1030 #if 0
1031 /* only turn this on if we know we can deal with read_block
1032 * returning nothing */
1033 if (reada)
1034 readflags |= OCFS2_BH_READAHEAD;
1035 #endif
1037 if (((u64)block << inode->i_sb->s_blocksize_bits) >=
1038 i_size_read(inode)) {
1039 BUG_ON(!reada);
1040 return NULL;
1043 tmperr = ocfs2_extent_map_get_blocks(inode, block, 1,
1044 &p_blkno, NULL);
1045 if (tmperr < 0) {
1046 mlog_errno(tmperr);
1047 goto fail;
1050 tmperr = ocfs2_read_block(OCFS2_SB(inode->i_sb), p_blkno, &bh,
1051 readflags, inode);
1052 if (tmperr < 0)
1053 goto fail;
1055 tmperr = 0;
1057 *err = 0;
1058 return bh;
1060 fail:
1061 if (bh) {
1062 brelse(bh);
1063 bh = NULL;
1065 *err = -EIO;
1066 return NULL;
1069 /*
1070 * This is called from our getattr.
1071 */
1072 int ocfs2_inode_revalidate(struct dentry *dentry)
1074 struct inode *inode = dentry->d_inode;
1075 int status = 0;
1077 mlog_entry("(inode = 0x%p, ino = %llu)\n", inode,
1078 inode ? (unsigned long long)OCFS2_I(inode)->ip_blkno : 0ULL);
1080 if (!inode) {
1081 mlog(0, "eep, no inode!\n");
1082 status = -ENOENT;
1083 goto bail;
1086 spin_lock(&OCFS2_I(inode)->ip_lock);
1087 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
1088 spin_unlock(&OCFS2_I(inode)->ip_lock);
1089 mlog(0, "inode deleted!\n");
1090 status = -ENOENT;
1091 goto bail;
1093 spin_unlock(&OCFS2_I(inode)->ip_lock);
1095 /* Let ocfs2_meta_lock do the work of updating our struct
1096 * inode for us. */
1097 status = ocfs2_meta_lock(inode, NULL, NULL, 0);
1098 if (status < 0) {
1099 if (status != -ENOENT)
1100 mlog_errno(status);
1101 goto bail;
1103 ocfs2_meta_unlock(inode, 0);
1104 bail:
1105 mlog_exit(status);
1107 return status;
1110 /*
1111 * Updates a disk inode from a
1112 * struct inode.
1113 * Only takes ip_lock.
1114 */
1115 int ocfs2_mark_inode_dirty(struct ocfs2_journal_handle *handle,
1116 struct inode *inode,
1117 struct buffer_head *bh)
1119 int status;
1120 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
1122 mlog_entry("(inode %llu)\n",
1123 (unsigned long long)OCFS2_I(inode)->ip_blkno);
1125 status = ocfs2_journal_access(handle, inode, bh,
1126 OCFS2_JOURNAL_ACCESS_WRITE);
1127 if (status < 0) {
1128 mlog_errno(status);
1129 goto leave;
1132 spin_lock(&OCFS2_I(inode)->ip_lock);
1133 fe->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters);
1134 spin_unlock(&OCFS2_I(inode)->ip_lock);
1136 fe->i_size = cpu_to_le64(i_size_read(inode));
1137 fe->i_links_count = cpu_to_le16(inode->i_nlink);
1138 fe->i_uid = cpu_to_le32(inode->i_uid);
1139 fe->i_gid = cpu_to_le32(inode->i_gid);
1140 fe->i_mode = cpu_to_le16(inode->i_mode);
1141 fe->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
1142 fe->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
1143 fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
1144 fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
1145 fe->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
1146 fe->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
1148 status = ocfs2_journal_dirty(handle, bh);
1149 if (status < 0)
1150 mlog_errno(status);
1152 status = 0;
1153 leave:
1155 mlog_exit(status);
1156 return status;
1159 /*
1161 * Updates a struct inode from a disk inode.
1162 * does no i/o, only takes ip_lock.
1163 */
1164 void ocfs2_refresh_inode(struct inode *inode,
1165 struct ocfs2_dinode *fe)
1167 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1169 spin_lock(&OCFS2_I(inode)->ip_lock);
1171 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
1172 i_size_write(inode, le64_to_cpu(fe->i_size));
1173 inode->i_nlink = le16_to_cpu(fe->i_links_count);
1174 inode->i_uid = le32_to_cpu(fe->i_uid);
1175 inode->i_gid = le32_to_cpu(fe->i_gid);
1176 inode->i_mode = le16_to_cpu(fe->i_mode);
1177 inode->i_blksize = (u32) osb->s_clustersize;
1178 if (S_ISLNK(inode->i_mode) && le32_to_cpu(fe->i_clusters) == 0)
1179 inode->i_blocks = 0;
1180 else
1181 inode->i_blocks = ocfs2_align_bytes_to_sectors(i_size_read(inode));
1182 inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
1183 inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
1184 inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
1185 inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
1186 inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
1187 inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
1189 spin_unlock(&OCFS2_I(inode)->ip_lock);