blob: 28ab56f2b98c50bf8300140c9002b801060c66a9 [file] [log] [blame]
Mark Fashehccd979b2005-12-15 14:31:24 -08001/* -*- 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"
Mark Fashehb4df6ed2006-02-22 17:35:08 -080044#include "heartbeat.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080045#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"
54
55#include "buffer_head_io.h"
56
Mark Fashehccd979b2005-12-15 14:31:24 -080057struct ocfs2_find_inode_args
58{
59 u64 fi_blkno;
60 unsigned long fi_ino;
61 unsigned int fi_flags;
62};
63
64static int ocfs2_read_locked_inode(struct inode *inode,
65 struct ocfs2_find_inode_args *args);
66static int ocfs2_init_locked_inode(struct inode *inode, void *opaque);
67static int ocfs2_find_actor(struct inode *inode, void *opaque);
68static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
69 struct inode *inode,
70 struct buffer_head *fe_bh);
71
Herbert Poetzlca4d1472006-07-03 17:27:12 -070072void ocfs2_set_inode_flags(struct inode *inode)
73{
74 unsigned int flags = OCFS2_I(inode)->ip_attr;
75
76 inode->i_flags &= ~(S_IMMUTABLE |
77 S_SYNC | S_APPEND | S_NOATIME | S_DIRSYNC);
78
79 if (flags & OCFS2_IMMUTABLE_FL)
80 inode->i_flags |= S_IMMUTABLE;
81
82 if (flags & OCFS2_SYNC_FL)
83 inode->i_flags |= S_SYNC;
84 if (flags & OCFS2_APPEND_FL)
85 inode->i_flags |= S_APPEND;
86 if (flags & OCFS2_NOATIME_FL)
87 inode->i_flags |= S_NOATIME;
88 if (flags & OCFS2_DIRSYNC_FL)
89 inode->i_flags |= S_DIRSYNC;
90}
91
Mark Fashehccd979b2005-12-15 14:31:24 -080092struct inode *ocfs2_ilookup_for_vote(struct ocfs2_super *osb,
93 u64 blkno,
94 int delete_vote)
95{
96 struct ocfs2_find_inode_args args;
97
98 /* ocfs2_ilookup_for_vote should *only* be called from the
99 * vote thread */
100 BUG_ON(current != osb->vote_task);
101
102 args.fi_blkno = blkno;
103 args.fi_flags = OCFS2_FI_FLAG_NOWAIT;
104 if (delete_vote)
105 args.fi_flags |= OCFS2_FI_FLAG_DELETE;
106 args.fi_ino = ino_from_blkno(osb->sb, blkno);
107 return ilookup5(osb->sb, args.fi_ino, ocfs2_find_actor, &args);
108}
109
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700110struct inode *ocfs2_iget(struct ocfs2_super *osb, u64 blkno, int flags)
Mark Fashehccd979b2005-12-15 14:31:24 -0800111{
112 struct inode *inode = NULL;
113 struct super_block *sb = osb->sb;
114 struct ocfs2_find_inode_args args;
115
Mark Fashehb06970532006-03-03 10:24:33 -0800116 mlog_entry("(blkno = %llu)\n", (unsigned long long)blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -0800117
118 /* Ok. By now we've either got the offsets passed to us by the
119 * caller, or we just pulled them off the bh. Lets do some
120 * sanity checks to make sure they're OK. */
121 if (blkno == 0) {
122 inode = ERR_PTR(-EINVAL);
123 mlog_errno(PTR_ERR(inode));
124 goto bail;
125 }
126
127 args.fi_blkno = blkno;
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700128 args.fi_flags = flags;
Mark Fashehccd979b2005-12-15 14:31:24 -0800129 args.fi_ino = ino_from_blkno(sb, blkno);
130
131 inode = iget5_locked(sb, args.fi_ino, ocfs2_find_actor,
132 ocfs2_init_locked_inode, &args);
133 /* inode was *not* in the inode cache. 2.6.x requires
134 * us to do our own read_inode call and unlock it
135 * afterwards. */
136 if (inode && inode->i_state & I_NEW) {
137 mlog(0, "Inode was not in inode cache, reading it.\n");
138 ocfs2_read_locked_inode(inode, &args);
139 unlock_new_inode(inode);
140 }
141 if (inode == NULL) {
142 inode = ERR_PTR(-ENOMEM);
143 mlog_errno(PTR_ERR(inode));
144 goto bail;
145 }
146 if (is_bad_inode(inode)) {
147 iput(inode);
148 inode = ERR_PTR(-ESTALE);
Mark Fashehccd979b2005-12-15 14:31:24 -0800149 goto bail;
150 }
151
152bail:
153 if (!IS_ERR(inode)) {
Mark Fashehb06970532006-03-03 10:24:33 -0800154 mlog(0, "returning inode with number %llu\n",
155 (unsigned long long)OCFS2_I(inode)->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -0800156 mlog_exit_ptr(inode);
Mark Fasheh6a1bd4a2007-01-03 17:06:59 -0800157 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800158
159 return inode;
160}
161
162
163/*
164 * here's how inodes get read from disk:
165 * iget5_locked -> find_actor -> OCFS2_FIND_ACTOR
166 * found? : return the in-memory inode
167 * not found? : get_new_inode -> OCFS2_INIT_LOCKED_INODE
168 */
169
170static int ocfs2_find_actor(struct inode *inode, void *opaque)
171{
172 struct ocfs2_find_inode_args *args = NULL;
173 struct ocfs2_inode_info *oi = OCFS2_I(inode);
174 int ret = 0;
175
176 mlog_entry("(0x%p, %lu, 0x%p)\n", inode, inode->i_ino, opaque);
177
178 args = opaque;
179
180 mlog_bug_on_msg(!inode, "No inode in find actor!\n");
181
182 if (oi->ip_blkno != args->fi_blkno)
183 goto bail;
184
185 /* OCFS2_FI_FLAG_NOWAIT is *only* set from
186 * ocfs2_ilookup_for_vote which won't create an inode for one
187 * that isn't found. The vote thread which doesn't want to get
188 * an inode which is in the process of going away - otherwise
189 * the call to __wait_on_freeing_inode in find_inode_fast will
190 * cause it to deadlock on an inode which may be waiting on a
191 * vote (or lock release) in delete_inode */
192 if ((args->fi_flags & OCFS2_FI_FLAG_NOWAIT) &&
193 (inode->i_state & (I_FREEING|I_CLEAR))) {
194 /* As stated above, we're not going to return an
195 * inode. In the case of a delete vote, the voting
196 * code is going to signal the other node to go
197 * ahead. Mark that state here, so this freeing inode
198 * has the state when it gets to delete_inode. */
199 if (args->fi_flags & OCFS2_FI_FLAG_DELETE) {
200 spin_lock(&oi->ip_lock);
201 ocfs2_mark_inode_remotely_deleted(inode);
202 spin_unlock(&oi->ip_lock);
203 }
204 goto bail;
205 }
206
207 ret = 1;
208bail:
209 mlog_exit(ret);
210 return ret;
211}
212
213/*
214 * initialize the new inode, but don't do anything that would cause
215 * us to sleep.
216 * return 0 on success, 1 on failure
217 */
218static int ocfs2_init_locked_inode(struct inode *inode, void *opaque)
219{
220 struct ocfs2_find_inode_args *args = opaque;
221
222 mlog_entry("inode = %p, opaque = %p\n", inode, opaque);
223
224 inode->i_ino = args->fi_ino;
225 OCFS2_I(inode)->ip_blkno = args->fi_blkno;
226
227 mlog_exit(0);
228 return 0;
229}
230
231int ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe,
232 int create_ino)
233{
234 struct super_block *sb;
235 struct ocfs2_super *osb;
236 int status = -EINVAL;
237
Mark Fashehb06970532006-03-03 10:24:33 -0800238 mlog_entry("(0x%p, size:%llu)\n", inode,
239 (unsigned long long)fe->i_size);
Mark Fashehccd979b2005-12-15 14:31:24 -0800240
241 sb = inode->i_sb;
242 osb = OCFS2_SB(sb);
243
244 /* this means that read_inode cannot create a superblock inode
245 * today. change if needed. */
246 if (!OCFS2_IS_VALID_DINODE(fe) ||
247 !(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL))) {
Mark Fasheh6a1bd4a2007-01-03 17:06:59 -0800248 mlog(0, "Invalid dinode: i_ino=%lu, i_blkno=%llu, "
Mark Fashehccd979b2005-12-15 14:31:24 -0800249 "signature = %.*s, flags = 0x%x\n",
Mark Fashehb06970532006-03-03 10:24:33 -0800250 inode->i_ino,
251 (unsigned long long)le64_to_cpu(fe->i_blkno), 7,
Mark Fashehccd979b2005-12-15 14:31:24 -0800252 fe->i_signature, le32_to_cpu(fe->i_flags));
253 goto bail;
254 }
255
256 if (le32_to_cpu(fe->i_fs_generation) != osb->fs_generation) {
257 mlog(ML_ERROR, "file entry generation does not match "
258 "superblock! osb->fs_generation=%x, "
259 "fe->i_fs_generation=%x\n",
260 osb->fs_generation, le32_to_cpu(fe->i_fs_generation));
261 goto bail;
262 }
263
264 inode->i_version = 1;
265 inode->i_generation = le32_to_cpu(fe->i_generation);
266 inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
267 inode->i_mode = le16_to_cpu(fe->i_mode);
268 inode->i_uid = le32_to_cpu(fe->i_uid);
269 inode->i_gid = le32_to_cpu(fe->i_gid);
Mark Fashehccd979b2005-12-15 14:31:24 -0800270
271 /* Fast symlinks will have i_size but no allocated clusters. */
272 if (S_ISLNK(inode->i_mode) && !fe->i_clusters)
273 inode->i_blocks = 0;
274 else
275 inode->i_blocks =
276 ocfs2_align_bytes_to_sectors(le64_to_cpu(fe->i_size));
277 inode->i_mapping->a_ops = &ocfs2_aops;
Mark Fashehccd979b2005-12-15 14:31:24 -0800278 inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
279 inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
280 inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
281 inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
282 inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
283 inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
284
285 if (OCFS2_I(inode)->ip_blkno != le64_to_cpu(fe->i_blkno))
286 mlog(ML_ERROR,
Mark Fashehb06970532006-03-03 10:24:33 -0800287 "ip_blkno %llu != i_blkno %llu!\n",
288 (unsigned long long)OCFS2_I(inode)->ip_blkno,
289 (unsigned long long)fe->i_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -0800290
291 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
292 OCFS2_I(inode)->ip_orphaned_slot = OCFS2_INVALID_SLOT;
Herbert Poetzlca4d1472006-07-03 17:27:12 -0700293 OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
Mark Fashehccd979b2005-12-15 14:31:24 -0800294
Mark Fashehccd979b2005-12-15 14:31:24 -0800295 inode->i_nlink = le16_to_cpu(fe->i_links_count);
296
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700297 if (fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL))
298 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SYSTEM_FILE;
299
Mark Fashehccd979b2005-12-15 14:31:24 -0800300 if (fe->i_flags & cpu_to_le32(OCFS2_LOCAL_ALLOC_FL)) {
301 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
302 mlog(0, "local alloc inode: i_ino=%lu\n", inode->i_ino);
303 } else if (fe->i_flags & cpu_to_le32(OCFS2_BITMAP_FL)) {
304 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
305 } else if (fe->i_flags & cpu_to_le32(OCFS2_SUPER_BLOCK_FL)) {
306 mlog(0, "superblock inode: i_ino=%lu\n", inode->i_ino);
307 /* we can't actually hit this as read_inode can't
308 * handle superblocks today ;-) */
309 BUG();
310 }
311
312 switch (inode->i_mode & S_IFMT) {
313 case S_IFREG:
314 inode->i_fop = &ocfs2_fops;
315 inode->i_op = &ocfs2_file_iops;
316 i_size_write(inode, le64_to_cpu(fe->i_size));
317 break;
318 case S_IFDIR:
319 inode->i_op = &ocfs2_dir_iops;
320 inode->i_fop = &ocfs2_dops;
321 i_size_write(inode, le64_to_cpu(fe->i_size));
322 break;
323 case S_IFLNK:
324 if (ocfs2_inode_is_fast_symlink(inode))
325 inode->i_op = &ocfs2_fast_symlink_inode_operations;
326 else
327 inode->i_op = &ocfs2_symlink_inode_operations;
328 i_size_write(inode, le64_to_cpu(fe->i_size));
329 break;
330 default:
331 inode->i_op = &ocfs2_special_file_iops;
332 init_special_inode(inode, inode->i_mode,
333 inode->i_rdev);
334 break;
335 }
336
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700337 if (create_ino) {
338 inode->i_ino = ino_from_blkno(inode->i_sb,
339 le64_to_cpu(fe->i_blkno));
340
341 /*
342 * If we ever want to create system files from kernel,
343 * the generation argument to
344 * ocfs2_inode_lock_res_init() will have to change.
345 */
346 BUG_ON(fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL));
347
348 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_meta_lockres,
349 OCFS2_LOCK_TYPE_META, 0, inode);
350 }
351
Mark Fashehccd979b2005-12-15 14:31:24 -0800352 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_rw_lockres,
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700353 OCFS2_LOCK_TYPE_RW, inode->i_generation,
354 inode);
355
Mark Fashehccd979b2005-12-15 14:31:24 -0800356 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_data_lockres,
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700357 OCFS2_LOCK_TYPE_DATA, inode->i_generation,
358 inode);
Mark Fashehccd979b2005-12-15 14:31:24 -0800359
Herbert Poetzlca4d1472006-07-03 17:27:12 -0700360 ocfs2_set_inode_flags(inode);
Herbert Poetzlca4d1472006-07-03 17:27:12 -0700361
Mark Fashehccd979b2005-12-15 14:31:24 -0800362 status = 0;
363bail:
364 mlog_exit(status);
365 return status;
366}
367
368static int ocfs2_read_locked_inode(struct inode *inode,
369 struct ocfs2_find_inode_args *args)
370{
371 struct super_block *sb;
372 struct ocfs2_super *osb;
373 struct ocfs2_dinode *fe;
374 struct buffer_head *bh = NULL;
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700375 int status, can_lock;
376 u32 generation = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -0800377
378 mlog_entry("(0x%p, 0x%p)\n", inode, args);
379
380 status = -EINVAL;
381 if (inode == NULL || inode->i_sb == NULL) {
382 mlog(ML_ERROR, "bad inode\n");
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700383 return status;
Mark Fashehccd979b2005-12-15 14:31:24 -0800384 }
385 sb = inode->i_sb;
386 osb = OCFS2_SB(sb);
387
388 if (!args) {
389 mlog(ML_ERROR, "bad inode args\n");
390 make_bad_inode(inode);
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700391 return status;
Mark Fashehccd979b2005-12-15 14:31:24 -0800392 }
393
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700394 /*
395 * To improve performance of cold-cache inode stats, we take
396 * the cluster lock here if possible.
397 *
398 * Generally, OCFS2 never trusts the contents of an inode
399 * unless it's holding a cluster lock, so taking it here isn't
400 * a correctness issue as much as it is a performance
401 * improvement.
402 *
403 * There are three times when taking the lock is not a good idea:
404 *
405 * 1) During startup, before we have initialized the DLM.
406 *
407 * 2) If we are reading certain system files which never get
408 * cluster locks (local alloc, truncate log).
409 *
410 * 3) If the process doing the iget() is responsible for
411 * orphan dir recovery. We're holding the orphan dir lock and
412 * can get into a deadlock with another process on another
413 * node in ->delete_inode().
414 *
415 * #1 and #2 can be simply solved by never taking the lock
416 * here for system files (which are the only type we read
417 * during mount). It's a heavier approach, but our main
418 * concern is user-accesible files anyway.
419 *
420 * #3 works itself out because we'll eventually take the
421 * cluster lock before trusting anything anyway.
422 */
423 can_lock = !(args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
Sunil Mushranc271c5c2006-12-05 17:56:35 -0800424 && !(args->fi_flags & OCFS2_FI_FLAG_NOLOCK)
425 && !ocfs2_mount_local(osb);
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700426
427 /*
428 * To maintain backwards compatibility with older versions of
429 * ocfs2-tools, we still store the generation value for system
430 * files. The only ones that actually matter to userspace are
431 * the journals, but it's easier and inexpensive to just flag
432 * all system files similarly.
433 */
434 if (args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
435 generation = osb->fs_generation;
436
437 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_meta_lockres,
438 OCFS2_LOCK_TYPE_META,
439 generation, inode);
440
441 if (can_lock) {
Mark Fasheh4bcec182006-10-09 16:02:40 -0700442 status = ocfs2_meta_lock(inode, NULL, 0);
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700443 if (status) {
444 make_bad_inode(inode);
445 mlog_errno(status);
446 return status;
447 }
448 }
449
450 status = ocfs2_read_block(osb, args->fi_blkno, &bh, 0,
451 can_lock ? inode : NULL);
Mark Fashehccd979b2005-12-15 14:31:24 -0800452 if (status < 0) {
453 mlog_errno(status);
Mark Fashehccd979b2005-12-15 14:31:24 -0800454 goto bail;
455 }
456
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700457 status = -EINVAL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800458 fe = (struct ocfs2_dinode *) bh->b_data;
459 if (!OCFS2_IS_VALID_DINODE(fe)) {
Mark Fashehb06970532006-03-03 10:24:33 -0800460 mlog(ML_ERROR, "Invalid dinode #%llu: signature = %.*s\n",
461 (unsigned long long)fe->i_blkno, 7, fe->i_signature);
Mark Fashehccd979b2005-12-15 14:31:24 -0800462 goto bail;
463 }
464
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700465 /*
466 * This is a code bug. Right now the caller needs to
467 * understand whether it is asking for a system file inode or
468 * not so the proper lock names can be built.
469 */
470 mlog_bug_on_msg(!!(fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) !=
471 !!(args->fi_flags & OCFS2_FI_FLAG_SYSFILE),
472 "Inode %llu: system file state is ambigous\n",
473 (unsigned long long)args->fi_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -0800474
475 if (S_ISCHR(le16_to_cpu(fe->i_mode)) ||
476 S_ISBLK(le16_to_cpu(fe->i_mode)))
477 inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
478
Mark Fasheh6a1bd4a2007-01-03 17:06:59 -0800479 if (ocfs2_populate_inode(inode, fe, 0) < 0)
Mark Fashehccd979b2005-12-15 14:31:24 -0800480 goto bail;
Mark Fashehccd979b2005-12-15 14:31:24 -0800481
482 BUG_ON(args->fi_blkno != le64_to_cpu(fe->i_blkno));
483
Mark Fashehccd979b2005-12-15 14:31:24 -0800484 status = 0;
485
486bail:
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700487 if (can_lock)
488 ocfs2_meta_unlock(inode, 0);
489
490 if (status < 0)
491 make_bad_inode(inode);
492
Mark Fashehccd979b2005-12-15 14:31:24 -0800493 if (args && bh)
494 brelse(bh);
495
496 mlog_exit(status);
497 return status;
498}
499
500void ocfs2_sync_blockdev(struct super_block *sb)
501{
502 sync_blockdev(sb->s_bdev);
503}
504
505static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
506 struct inode *inode,
507 struct buffer_head *fe_bh)
508{
509 int status = 0;
Mark Fasheh1fabe142006-10-09 18:11:45 -0700510 handle_t *handle = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800511 struct ocfs2_truncate_context *tc = NULL;
512 struct ocfs2_dinode *fe;
513
514 mlog_entry_void();
515
516 fe = (struct ocfs2_dinode *) fe_bh->b_data;
517
518 /* zero allocation, zero truncate :) */
519 if (!fe->i_clusters)
520 goto bail;
521
Mark Fasheh65eff9c2006-10-09 17:26:22 -0700522 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
Mark Fashehccd979b2005-12-15 14:31:24 -0800523 if (IS_ERR(handle)) {
524 status = PTR_ERR(handle);
525 handle = NULL;
526 mlog_errno(status);
527 goto bail;
528 }
529
530 status = ocfs2_set_inode_size(handle, inode, fe_bh, 0ULL);
531 if (status < 0) {
532 mlog_errno(status);
533 goto bail;
534 }
535
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700536 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800537 handle = NULL;
538
539 status = ocfs2_prepare_truncate(osb, inode, fe_bh, &tc);
540 if (status < 0) {
541 mlog_errno(status);
542 goto bail;
543 }
544
545 status = ocfs2_commit_truncate(osb, inode, fe_bh, tc);
546 if (status < 0) {
547 mlog_errno(status);
548 goto bail;
549 }
550bail:
551 if (handle)
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700552 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800553
554 mlog_exit(status);
555 return status;
556}
557
558static int ocfs2_remove_inode(struct inode *inode,
559 struct buffer_head *di_bh,
560 struct inode *orphan_dir_inode,
561 struct buffer_head *orphan_dir_bh)
562{
563 int status;
564 struct inode *inode_alloc_inode = NULL;
565 struct buffer_head *inode_alloc_bh = NULL;
Mark Fasheh1fabe142006-10-09 18:11:45 -0700566 handle_t *handle;
Mark Fashehccd979b2005-12-15 14:31:24 -0800567 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
568 struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
569
570 inode_alloc_inode =
571 ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE,
572 le16_to_cpu(di->i_suballoc_slot));
573 if (!inode_alloc_inode) {
574 status = -EEXIST;
575 mlog_errno(status);
576 goto bail;
577 }
578
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800579 mutex_lock(&inode_alloc_inode->i_mutex);
Mark Fasheh4bcec182006-10-09 16:02:40 -0700580 status = ocfs2_meta_lock(inode_alloc_inode, &inode_alloc_bh, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -0800581 if (status < 0) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800582 mutex_unlock(&inode_alloc_inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -0800583
584 mlog_errno(status);
585 goto bail;
586 }
587
Mark Fasheh65eff9c2006-10-09 17:26:22 -0700588 handle = ocfs2_start_trans(osb, OCFS2_DELETE_INODE_CREDITS);
Mark Fashehccd979b2005-12-15 14:31:24 -0800589 if (IS_ERR(handle)) {
590 status = PTR_ERR(handle);
591 mlog_errno(status);
592 goto bail_unlock;
593 }
594
595 status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
596 orphan_dir_bh);
597 if (status < 0) {
598 mlog_errno(status);
599 goto bail_commit;
600 }
601
602 /* set the inodes dtime */
603 status = ocfs2_journal_access(handle, inode, di_bh,
604 OCFS2_JOURNAL_ACCESS_WRITE);
605 if (status < 0) {
606 mlog_errno(status);
607 goto bail_commit;
608 }
609
610 di->i_dtime = cpu_to_le64(CURRENT_TIME.tv_sec);
611 le32_and_cpu(&di->i_flags, ~(OCFS2_VALID_FL | OCFS2_ORPHANED_FL));
612
613 status = ocfs2_journal_dirty(handle, di_bh);
614 if (status < 0) {
615 mlog_errno(status);
616 goto bail_commit;
617 }
618
619 ocfs2_remove_from_cache(inode, di_bh);
620
621 status = ocfs2_free_dinode(handle, inode_alloc_inode,
622 inode_alloc_bh, di);
623 if (status < 0)
624 mlog_errno(status);
625
626bail_commit:
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700627 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800628bail_unlock:
629 ocfs2_meta_unlock(inode_alloc_inode, 1);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800630 mutex_unlock(&inode_alloc_inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -0800631 brelse(inode_alloc_bh);
632bail:
633 iput(inode_alloc_inode);
634
635 return status;
636}
637
Mark Fashehb4df6ed2006-02-22 17:35:08 -0800638/*
639 * Serialize with orphan dir recovery. If the process doing
640 * recovery on this orphan dir does an iget() with the dir
641 * i_mutex held, we'll deadlock here. Instead we detect this
642 * and exit early - recovery will wipe this inode for us.
643 */
644static int ocfs2_check_orphan_recovery_state(struct ocfs2_super *osb,
645 int slot)
646{
647 int ret = 0;
648
649 spin_lock(&osb->osb_lock);
650 if (ocfs2_node_map_test_bit(osb, &osb->osb_recovering_orphan_dirs, slot)) {
651 mlog(0, "Recovery is happening on orphan dir %d, will skip "
652 "this inode\n", slot);
653 ret = -EDEADLK;
654 goto out;
655 }
656 /* This signals to the orphan recovery process that it should
657 * wait for us to handle the wipe. */
658 osb->osb_orphan_wipes[slot]++;
659out:
660 spin_unlock(&osb->osb_lock);
661 return ret;
662}
663
664static void ocfs2_signal_wipe_completion(struct ocfs2_super *osb,
665 int slot)
666{
667 spin_lock(&osb->osb_lock);
668 osb->osb_orphan_wipes[slot]--;
669 spin_unlock(&osb->osb_lock);
670
671 wake_up(&osb->osb_wipe_event);
672}
673
Mark Fashehccd979b2005-12-15 14:31:24 -0800674static int ocfs2_wipe_inode(struct inode *inode,
675 struct buffer_head *di_bh)
676{
677 int status, orphaned_slot;
678 struct inode *orphan_dir_inode = NULL;
679 struct buffer_head *orphan_dir_bh = NULL;
680 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
681
682 /* We've already voted on this so it should be readonly - no
683 * spinlock needed. */
684 orphaned_slot = OCFS2_I(inode)->ip_orphaned_slot;
Mark Fashehb4df6ed2006-02-22 17:35:08 -0800685
686 status = ocfs2_check_orphan_recovery_state(osb, orphaned_slot);
687 if (status)
688 return status;
689
Mark Fashehccd979b2005-12-15 14:31:24 -0800690 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
691 ORPHAN_DIR_SYSTEM_INODE,
692 orphaned_slot);
693 if (!orphan_dir_inode) {
694 status = -EEXIST;
695 mlog_errno(status);
696 goto bail;
697 }
698
699 /* Lock the orphan dir. The lock will be held for the entire
700 * delete_inode operation. We do this now to avoid races with
701 * recovery completion on other nodes. */
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800702 mutex_lock(&orphan_dir_inode->i_mutex);
Mark Fasheh4bcec182006-10-09 16:02:40 -0700703 status = ocfs2_meta_lock(orphan_dir_inode, &orphan_dir_bh, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -0800704 if (status < 0) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800705 mutex_unlock(&orphan_dir_inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -0800706
707 mlog_errno(status);
708 goto bail;
709 }
710
711 /* we do this while holding the orphan dir lock because we
712 * don't want recovery being run from another node to vote for
713 * an inode delete on us -- this will result in two nodes
714 * truncating the same file! */
715 status = ocfs2_truncate_for_delete(osb, inode, di_bh);
716 if (status < 0) {
717 mlog_errno(status);
718 goto bail_unlock_dir;
719 }
720
721 status = ocfs2_remove_inode(inode, di_bh, orphan_dir_inode,
722 orphan_dir_bh);
723 if (status < 0)
724 mlog_errno(status);
725
726bail_unlock_dir:
727 ocfs2_meta_unlock(orphan_dir_inode, 1);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800728 mutex_unlock(&orphan_dir_inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -0800729 brelse(orphan_dir_bh);
730bail:
731 iput(orphan_dir_inode);
Mark Fashehb4df6ed2006-02-22 17:35:08 -0800732 ocfs2_signal_wipe_completion(osb, orphaned_slot);
Mark Fashehccd979b2005-12-15 14:31:24 -0800733
734 return status;
735}
736
737/* There is a series of simple checks that should be done before a
738 * vote is even considered. Encapsulate those in this function. */
739static int ocfs2_inode_is_valid_to_delete(struct inode *inode)
740{
741 int ret = 0;
742 struct ocfs2_inode_info *oi = OCFS2_I(inode);
743 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
744
745 /* We shouldn't be getting here for the root directory
746 * inode.. */
747 if (inode == osb->root_inode) {
748 mlog(ML_ERROR, "Skipping delete of root inode.\n");
749 goto bail;
750 }
751
752 /* If we're coming from process_vote we can't go into our own
753 * voting [hello, deadlock city!], so unforuntately we just
754 * have to skip deleting this guy. That's OK though because
755 * the node who's doing the actual deleting should handle it
756 * anyway. */
757 if (current == osb->vote_task) {
758 mlog(0, "Skipping delete of %lu because we're currently "
759 "in process_vote\n", inode->i_ino);
760 goto bail;
761 }
762
763 spin_lock(&oi->ip_lock);
764 /* OCFS2 *never* deletes system files. This should technically
765 * never get here as system file inodes should always have a
766 * positive link count. */
767 if (oi->ip_flags & OCFS2_INODE_SYSTEM_FILE) {
Mark Fashehb06970532006-03-03 10:24:33 -0800768 mlog(ML_ERROR, "Skipping delete of system file %llu\n",
769 (unsigned long long)oi->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -0800770 goto bail_unlock;
771 }
772
773 /* If we have voted "yes" on the wipe of this inode for
774 * another node, it will be marked here so we can safely skip
775 * it. Recovery will cleanup any inodes we might inadvertantly
776 * skip here. */
777 if (oi->ip_flags & OCFS2_INODE_SKIP_DELETE) {
778 mlog(0, "Skipping delete of %lu because another node "
779 "has done this for us.\n", inode->i_ino);
780 goto bail_unlock;
781 }
782
783 ret = 1;
784bail_unlock:
785 spin_unlock(&oi->ip_lock);
786bail:
787 return ret;
788}
789
790/* Query the cluster to determine whether we should wipe an inode from
791 * disk or not.
792 *
793 * Requires the inode to have the cluster lock. */
794static int ocfs2_query_inode_wipe(struct inode *inode,
795 struct buffer_head *di_bh,
796 int *wipe)
797{
798 int status = 0;
799 struct ocfs2_inode_info *oi = OCFS2_I(inode);
800 struct ocfs2_dinode *di;
801
802 *wipe = 0;
803
804 /* While we were waiting for the cluster lock in
805 * ocfs2_delete_inode, another node might have asked to delete
806 * the inode. Recheck our flags to catch this. */
807 if (!ocfs2_inode_is_valid_to_delete(inode)) {
Mark Fashehb06970532006-03-03 10:24:33 -0800808 mlog(0, "Skipping delete of %llu because flags changed\n",
809 (unsigned long long)oi->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -0800810 goto bail;
811 }
812
813 /* Now that we have an up to date inode, we can double check
814 * the link count. */
815 if (inode->i_nlink) {
Mark Fashehb06970532006-03-03 10:24:33 -0800816 mlog(0, "Skipping delete of %llu because nlink = %u\n",
817 (unsigned long long)oi->ip_blkno, inode->i_nlink);
Mark Fashehccd979b2005-12-15 14:31:24 -0800818 goto bail;
819 }
820
821 /* Do some basic inode verification... */
822 di = (struct ocfs2_dinode *) di_bh->b_data;
823 if (!(di->i_flags & cpu_to_le32(OCFS2_ORPHANED_FL))) {
824 /* for lack of a better error? */
825 status = -EEXIST;
826 mlog(ML_ERROR,
Mark Fashehb06970532006-03-03 10:24:33 -0800827 "Inode %llu (on-disk %llu) not orphaned! "
Mark Fashehccd979b2005-12-15 14:31:24 -0800828 "Disk flags 0x%x, inode flags 0x%x\n",
Mark Fashehb06970532006-03-03 10:24:33 -0800829 (unsigned long long)oi->ip_blkno,
830 (unsigned long long)di->i_blkno, di->i_flags,
831 oi->ip_flags);
Mark Fashehccd979b2005-12-15 14:31:24 -0800832 goto bail;
833 }
834
835 /* has someone already deleted us?! baaad... */
836 if (di->i_dtime) {
837 status = -EEXIST;
838 mlog_errno(status);
839 goto bail;
840 }
841
842 status = ocfs2_request_delete_vote(inode);
843 /* -EBUSY means that other nodes are still using the
844 * inode. We're done here though, so avoid doing anything on
845 * disk and let them worry about deleting it. */
846 if (status == -EBUSY) {
847 status = 0;
Mark Fashehb06970532006-03-03 10:24:33 -0800848 mlog(0, "Skipping delete of %llu because it is in use on"
849 "other nodes\n", (unsigned long long)oi->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -0800850 goto bail;
851 }
852 if (status < 0) {
853 mlog_errno(status);
854 goto bail;
855 }
856
857 spin_lock(&oi->ip_lock);
858 if (oi->ip_orphaned_slot == OCFS2_INVALID_SLOT) {
859 /* Nobody knew which slot this inode was orphaned
860 * into. This may happen during node death and
861 * recovery knows how to clean it up so we can safely
862 * ignore this inode for now on. */
Mark Fashehb06970532006-03-03 10:24:33 -0800863 mlog(0, "Nobody knew where inode %llu was orphaned!\n",
864 (unsigned long long)oi->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -0800865 } else {
866 *wipe = 1;
867
Mark Fashehb06970532006-03-03 10:24:33 -0800868 mlog(0, "Inode %llu is ok to wipe from orphan dir %d\n",
869 (unsigned long long)oi->ip_blkno, oi->ip_orphaned_slot);
Mark Fashehccd979b2005-12-15 14:31:24 -0800870 }
871 spin_unlock(&oi->ip_lock);
872
873bail:
874 return status;
875}
876
877/* Support function for ocfs2_delete_inode. Will help us keep the
878 * inode data in a consistent state for clear_inode. Always truncates
879 * pages, optionally sync's them first. */
880static void ocfs2_cleanup_delete_inode(struct inode *inode,
881 int sync_data)
882{
Mark Fashehb06970532006-03-03 10:24:33 -0800883 mlog(0, "Cleanup inode %llu, sync = %d\n",
884 (unsigned long long)OCFS2_I(inode)->ip_blkno, sync_data);
Mark Fashehccd979b2005-12-15 14:31:24 -0800885 if (sync_data)
886 write_inode_now(inode, 1);
887 truncate_inode_pages(&inode->i_data, 0);
888}
889
890void ocfs2_delete_inode(struct inode *inode)
891{
892 int wipe, status;
893 sigset_t blocked, oldset;
894 struct buffer_head *di_bh = NULL;
895
896 mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
897
898 if (is_bad_inode(inode)) {
899 mlog(0, "Skipping delete of bad inode\n");
900 goto bail;
901 }
902
903 if (!ocfs2_inode_is_valid_to_delete(inode)) {
904 /* It's probably not necessary to truncate_inode_pages
905 * here but we do it for safety anyway (it will most
906 * likely be a no-op anyway) */
907 ocfs2_cleanup_delete_inode(inode, 0);
908 goto bail;
909 }
910
911 /* We want to block signals in delete_inode as the lock and
912 * messaging paths may return us -ERESTARTSYS. Which would
913 * cause us to exit early, resulting in inodes being orphaned
914 * forever. */
915 sigfillset(&blocked);
916 status = sigprocmask(SIG_BLOCK, &blocked, &oldset);
917 if (status < 0) {
918 mlog_errno(status);
919 ocfs2_cleanup_delete_inode(inode, 1);
920 goto bail;
921 }
922
923 /* Lock down the inode. This gives us an up to date view of
924 * it's metadata (for verification), and allows us to
925 * serialize delete_inode votes.
926 *
927 * Even though we might be doing a truncate, we don't take the
928 * allocation lock here as it won't be needed - nobody will
929 * have the file open.
930 */
Mark Fasheh4bcec182006-10-09 16:02:40 -0700931 status = ocfs2_meta_lock(inode, &di_bh, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -0800932 if (status < 0) {
933 if (status != -ENOENT)
934 mlog_errno(status);
935 ocfs2_cleanup_delete_inode(inode, 0);
936 goto bail_unblock;
937 }
938
939 /* Query the cluster. This will be the final decision made
940 * before we go ahead and wipe the inode. */
941 status = ocfs2_query_inode_wipe(inode, di_bh, &wipe);
942 if (!wipe || status < 0) {
943 /* Error and inode busy vote both mean we won't be
944 * removing the inode, so they take almost the same
945 * path. */
946 if (status < 0)
947 mlog_errno(status);
948
949 /* Someone in the cluster has voted to not wipe this
950 * inode, or it was never completely orphaned. Write
951 * out the pages and exit now. */
952 ocfs2_cleanup_delete_inode(inode, 1);
953 goto bail_unlock_inode;
954 }
955
956 ocfs2_cleanup_delete_inode(inode, 0);
957
958 status = ocfs2_wipe_inode(inode, di_bh);
959 if (status < 0) {
Mark Fashehb4df6ed2006-02-22 17:35:08 -0800960 if (status != -EDEADLK)
961 mlog_errno(status);
Mark Fashehccd979b2005-12-15 14:31:24 -0800962 goto bail_unlock_inode;
963 }
964
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700965 /*
966 * Mark the inode as successfully deleted.
967 *
968 * This is important for ocfs2_clear_inode() as it will check
969 * this flag and skip any checkpointing work
970 *
971 * ocfs2_stuff_meta_lvb() also uses this flag to invalidate
972 * the LVB for other nodes.
973 */
Mark Fashehccd979b2005-12-15 14:31:24 -0800974 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_DELETED;
975
976bail_unlock_inode:
977 ocfs2_meta_unlock(inode, 1);
978 brelse(di_bh);
979bail_unblock:
980 status = sigprocmask(SIG_SETMASK, &oldset, NULL);
981 if (status < 0)
982 mlog_errno(status);
983bail:
984 clear_inode(inode);
985 mlog_exit_void();
986}
987
988void ocfs2_clear_inode(struct inode *inode)
989{
990 int status;
991 struct ocfs2_inode_info *oi = OCFS2_I(inode);
992
993 mlog_entry_void();
994
995 if (!inode)
996 goto bail;
997
Mark Fashehb06970532006-03-03 10:24:33 -0800998 mlog(0, "Clearing inode: %llu, nlink = %u\n",
999 (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_nlink);
Mark Fashehccd979b2005-12-15 14:31:24 -08001000
1001 mlog_bug_on_msg(OCFS2_SB(inode->i_sb) == NULL,
1002 "Inode=%lu\n", inode->i_ino);
1003
1004 /* Do these before all the other work so that we don't bounce
1005 * the vote thread while waiting to destroy the locks. */
1006 ocfs2_mark_lockres_freeing(&oi->ip_rw_lockres);
1007 ocfs2_mark_lockres_freeing(&oi->ip_meta_lockres);
1008 ocfs2_mark_lockres_freeing(&oi->ip_data_lockres);
1009
1010 /* We very well may get a clear_inode before all an inodes
1011 * metadata has hit disk. Of course, we can't drop any cluster
1012 * locks until the journal has finished with it. The only
1013 * exception here are successfully wiped inodes - their
1014 * metadata can now be considered to be part of the system
1015 * inodes from which it came. */
1016 if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED))
1017 ocfs2_checkpoint_inode(inode);
1018
1019 mlog_bug_on_msg(!list_empty(&oi->ip_io_markers),
Mark Fashehb06970532006-03-03 10:24:33 -08001020 "Clear inode of %llu, inode has io markers\n",
1021 (unsigned long long)oi->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08001022
1023 ocfs2_extent_map_drop(inode, 0);
1024 ocfs2_extent_map_init(inode);
1025
1026 status = ocfs2_drop_inode_locks(inode);
1027 if (status < 0)
1028 mlog_errno(status);
1029
1030 ocfs2_lock_res_free(&oi->ip_rw_lockres);
1031 ocfs2_lock_res_free(&oi->ip_meta_lockres);
1032 ocfs2_lock_res_free(&oi->ip_data_lockres);
1033
1034 ocfs2_metadata_cache_purge(inode);
1035
1036 mlog_bug_on_msg(oi->ip_metadata_cache.ci_num_cached,
Mark Fashehb06970532006-03-03 10:24:33 -08001037 "Clear inode of %llu, inode has %u cache items\n",
1038 (unsigned long long)oi->ip_blkno, oi->ip_metadata_cache.ci_num_cached);
Mark Fashehccd979b2005-12-15 14:31:24 -08001039
1040 mlog_bug_on_msg(!(oi->ip_flags & OCFS2_INODE_CACHE_INLINE),
Mark Fashehb06970532006-03-03 10:24:33 -08001041 "Clear inode of %llu, inode has a bad flag\n",
1042 (unsigned long long)oi->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08001043
1044 mlog_bug_on_msg(spin_is_locked(&oi->ip_lock),
Mark Fashehb06970532006-03-03 10:24:33 -08001045 "Clear inode of %llu, inode is locked\n",
1046 (unsigned long long)oi->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08001047
Mark Fasheh251b6ec2006-01-10 15:41:43 -08001048 mlog_bug_on_msg(!mutex_trylock(&oi->ip_io_mutex),
Mark Fashehb06970532006-03-03 10:24:33 -08001049 "Clear inode of %llu, io_mutex is locked\n",
1050 (unsigned long long)oi->ip_blkno);
Mark Fasheh251b6ec2006-01-10 15:41:43 -08001051 mutex_unlock(&oi->ip_io_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -08001052
1053 /*
1054 * down_trylock() returns 0, down_write_trylock() returns 1
1055 * kernel 1, world 0
1056 */
1057 mlog_bug_on_msg(!down_write_trylock(&oi->ip_alloc_sem),
Mark Fashehb06970532006-03-03 10:24:33 -08001058 "Clear inode of %llu, alloc_sem is locked\n",
1059 (unsigned long long)oi->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08001060 up_write(&oi->ip_alloc_sem);
1061
1062 mlog_bug_on_msg(oi->ip_open_count,
Mark Fashehb06970532006-03-03 10:24:33 -08001063 "Clear inode of %llu has open count %d\n",
1064 (unsigned long long)oi->ip_blkno, oi->ip_open_count);
Mark Fashehccd979b2005-12-15 14:31:24 -08001065
1066 /* Clear all other flags. */
1067 oi->ip_flags = OCFS2_INODE_CACHE_INLINE;
1068 oi->ip_created_trans = 0;
1069 oi->ip_last_trans = 0;
1070 oi->ip_dir_start_lookup = 0;
1071 oi->ip_blkno = 0ULL;
1072
1073bail:
1074 mlog_exit_void();
1075}
1076
1077/* Called under inode_lock, with no more references on the
1078 * struct inode, so it's safe here to check the flags field
1079 * and to manipulate i_nlink without any other locks. */
1080void ocfs2_drop_inode(struct inode *inode)
1081{
1082 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1083
1084 mlog_entry_void();
1085
Mark Fashehb06970532006-03-03 10:24:33 -08001086 mlog(0, "Drop inode %llu, nlink = %u, ip_flags = 0x%x\n",
1087 (unsigned long long)oi->ip_blkno, inode->i_nlink, oi->ip_flags);
Mark Fashehccd979b2005-12-15 14:31:24 -08001088
1089 /* Testing ip_orphaned_slot here wouldn't work because we may
1090 * not have gotten a delete_inode vote from any other nodes
1091 * yet. */
Mark Fasheh379dfe92006-09-08 14:21:03 -07001092 if (oi->ip_flags & OCFS2_INODE_MAYBE_ORPHANED)
1093 generic_delete_inode(inode);
1094 else
1095 generic_drop_inode(inode);
Mark Fashehccd979b2005-12-15 14:31:24 -08001096
1097 mlog_exit_void();
1098}
1099
1100/*
1101 * TODO: this should probably be merged into ocfs2_get_block
1102 *
1103 * However, you now need to pay attention to the cont_prepare_write()
1104 * stuff in ocfs2_get_block (that is, ocfs2_get_block pretty much
1105 * expects never to extend).
1106 */
1107struct buffer_head *ocfs2_bread(struct inode *inode,
1108 int block, int *err, int reada)
1109{
1110 struct buffer_head *bh = NULL;
1111 int tmperr;
1112 u64 p_blkno;
1113 int readflags = OCFS2_BH_CACHED;
1114
Mark Fashehccd979b2005-12-15 14:31:24 -08001115 if (reada)
1116 readflags |= OCFS2_BH_READAHEAD;
Mark Fashehccd979b2005-12-15 14:31:24 -08001117
1118 if (((u64)block << inode->i_sb->s_blocksize_bits) >=
1119 i_size_read(inode)) {
1120 BUG_ON(!reada);
1121 return NULL;
1122 }
1123
1124 tmperr = ocfs2_extent_map_get_blocks(inode, block, 1,
1125 &p_blkno, NULL);
1126 if (tmperr < 0) {
1127 mlog_errno(tmperr);
1128 goto fail;
1129 }
1130
1131 tmperr = ocfs2_read_block(OCFS2_SB(inode->i_sb), p_blkno, &bh,
1132 readflags, inode);
1133 if (tmperr < 0)
1134 goto fail;
1135
1136 tmperr = 0;
1137
1138 *err = 0;
1139 return bh;
1140
1141fail:
1142 if (bh) {
1143 brelse(bh);
1144 bh = NULL;
1145 }
1146 *err = -EIO;
1147 return NULL;
1148}
1149
1150/*
1151 * This is called from our getattr.
1152 */
1153int ocfs2_inode_revalidate(struct dentry *dentry)
1154{
1155 struct inode *inode = dentry->d_inode;
1156 int status = 0;
1157
Mark Fashehb06970532006-03-03 10:24:33 -08001158 mlog_entry("(inode = 0x%p, ino = %llu)\n", inode,
1159 inode ? (unsigned long long)OCFS2_I(inode)->ip_blkno : 0ULL);
Mark Fashehccd979b2005-12-15 14:31:24 -08001160
1161 if (!inode) {
1162 mlog(0, "eep, no inode!\n");
1163 status = -ENOENT;
1164 goto bail;
1165 }
1166
1167 spin_lock(&OCFS2_I(inode)->ip_lock);
1168 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
1169 spin_unlock(&OCFS2_I(inode)->ip_lock);
1170 mlog(0, "inode deleted!\n");
1171 status = -ENOENT;
1172 goto bail;
1173 }
1174 spin_unlock(&OCFS2_I(inode)->ip_lock);
1175
1176 /* Let ocfs2_meta_lock do the work of updating our struct
1177 * inode for us. */
Mark Fasheh4bcec182006-10-09 16:02:40 -07001178 status = ocfs2_meta_lock(inode, NULL, 0);
Mark Fashehccd979b2005-12-15 14:31:24 -08001179 if (status < 0) {
1180 if (status != -ENOENT)
1181 mlog_errno(status);
1182 goto bail;
1183 }
1184 ocfs2_meta_unlock(inode, 0);
1185bail:
1186 mlog_exit(status);
1187
1188 return status;
1189}
1190
1191/*
1192 * Updates a disk inode from a
1193 * struct inode.
1194 * Only takes ip_lock.
1195 */
Mark Fasheh1fabe142006-10-09 18:11:45 -07001196int ocfs2_mark_inode_dirty(handle_t *handle,
Mark Fashehccd979b2005-12-15 14:31:24 -08001197 struct inode *inode,
1198 struct buffer_head *bh)
1199{
1200 int status;
1201 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
1202
Mark Fashehb06970532006-03-03 10:24:33 -08001203 mlog_entry("(inode %llu)\n",
1204 (unsigned long long)OCFS2_I(inode)->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08001205
1206 status = ocfs2_journal_access(handle, inode, bh,
1207 OCFS2_JOURNAL_ACCESS_WRITE);
1208 if (status < 0) {
1209 mlog_errno(status);
1210 goto leave;
1211 }
1212
1213 spin_lock(&OCFS2_I(inode)->ip_lock);
1214 fe->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters);
Herbert Poetzlca4d1472006-07-03 17:27:12 -07001215 fe->i_attr = cpu_to_le32(OCFS2_I(inode)->ip_attr);
Mark Fashehccd979b2005-12-15 14:31:24 -08001216 spin_unlock(&OCFS2_I(inode)->ip_lock);
1217
1218 fe->i_size = cpu_to_le64(i_size_read(inode));
1219 fe->i_links_count = cpu_to_le16(inode->i_nlink);
1220 fe->i_uid = cpu_to_le32(inode->i_uid);
1221 fe->i_gid = cpu_to_le32(inode->i_gid);
1222 fe->i_mode = cpu_to_le16(inode->i_mode);
1223 fe->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
1224 fe->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
1225 fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
1226 fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
1227 fe->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
1228 fe->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
1229
1230 status = ocfs2_journal_dirty(handle, bh);
1231 if (status < 0)
1232 mlog_errno(status);
1233
1234 status = 0;
1235leave:
1236
1237 mlog_exit(status);
1238 return status;
1239}
1240
1241/*
1242 *
1243 * Updates a struct inode from a disk inode.
1244 * does no i/o, only takes ip_lock.
1245 */
1246void ocfs2_refresh_inode(struct inode *inode,
1247 struct ocfs2_dinode *fe)
1248{
Mark Fashehccd979b2005-12-15 14:31:24 -08001249 spin_lock(&OCFS2_I(inode)->ip_lock);
1250
1251 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
Herbert Poetzlca4d1472006-07-03 17:27:12 -07001252 OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
1253 ocfs2_set_inode_flags(inode);
Mark Fashehccd979b2005-12-15 14:31:24 -08001254 i_size_write(inode, le64_to_cpu(fe->i_size));
1255 inode->i_nlink = le16_to_cpu(fe->i_links_count);
1256 inode->i_uid = le32_to_cpu(fe->i_uid);
1257 inode->i_gid = le32_to_cpu(fe->i_gid);
1258 inode->i_mode = le16_to_cpu(fe->i_mode);
Mark Fashehccd979b2005-12-15 14:31:24 -08001259 if (S_ISLNK(inode->i_mode) && le32_to_cpu(fe->i_clusters) == 0)
1260 inode->i_blocks = 0;
1261 else
1262 inode->i_blocks = ocfs2_align_bytes_to_sectors(i_size_read(inode));
1263 inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
1264 inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
1265 inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
1266 inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
1267 inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
1268 inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
1269
1270 spin_unlock(&OCFS2_I(inode)->ip_lock);
1271}