blob: 58ec54b60038ed47b1ce27f760dea386736f581c [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 Fasheh24c19ef2006-09-22 17:28:19 -070092struct inode *ocfs2_iget(struct ocfs2_super *osb, u64 blkno, int flags)
Mark Fashehccd979b2005-12-15 14:31:24 -080093{
94 struct inode *inode = NULL;
95 struct super_block *sb = osb->sb;
96 struct ocfs2_find_inode_args args;
97
Mark Fashehb06970532006-03-03 10:24:33 -080098 mlog_entry("(blkno = %llu)\n", (unsigned long long)blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -080099
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 }
108
109 args.fi_blkno = blkno;
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700110 args.fi_flags = flags;
Mark Fashehccd979b2005-12-15 14:31:24 -0800111 args.fi_ino = ino_from_blkno(sb, blkno);
112
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);
Mark Fashehccd979b2005-12-15 14:31:24 -0800131 goto bail;
132 }
133
134bail:
135 if (!IS_ERR(inode)) {
Mark Fashehb06970532006-03-03 10:24:33 -0800136 mlog(0, "returning inode with number %llu\n",
137 (unsigned long long)OCFS2_I(inode)->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -0800138 mlog_exit_ptr(inode);
Mark Fasheh6a1bd4a2007-01-03 17:06:59 -0800139 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800140
141 return inode;
142}
143
144
145/*
146 * here's how inodes get read from disk:
147 * iget5_locked -> find_actor -> OCFS2_FIND_ACTOR
148 * found? : return the in-memory inode
149 * not found? : get_new_inode -> OCFS2_INIT_LOCKED_INODE
150 */
151
152static int ocfs2_find_actor(struct inode *inode, void *opaque)
153{
154 struct ocfs2_find_inode_args *args = NULL;
155 struct ocfs2_inode_info *oi = OCFS2_I(inode);
156 int ret = 0;
157
158 mlog_entry("(0x%p, %lu, 0x%p)\n", inode, inode->i_ino, opaque);
159
160 args = opaque;
161
162 mlog_bug_on_msg(!inode, "No inode in find actor!\n");
163
164 if (oi->ip_blkno != args->fi_blkno)
165 goto bail;
166
Mark Fashehccd979b2005-12-15 14:31:24 -0800167 ret = 1;
168bail:
169 mlog_exit(ret);
170 return ret;
171}
172
173/*
174 * initialize the new inode, but don't do anything that would cause
175 * us to sleep.
176 * return 0 on success, 1 on failure
177 */
178static int ocfs2_init_locked_inode(struct inode *inode, void *opaque)
179{
180 struct ocfs2_find_inode_args *args = opaque;
181
182 mlog_entry("inode = %p, opaque = %p\n", inode, opaque);
183
184 inode->i_ino = args->fi_ino;
185 OCFS2_I(inode)->ip_blkno = args->fi_blkno;
186
187 mlog_exit(0);
188 return 0;
189}
190
191int ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe,
192 int create_ino)
193{
194 struct super_block *sb;
195 struct ocfs2_super *osb;
196 int status = -EINVAL;
197
Mark Fashehb06970532006-03-03 10:24:33 -0800198 mlog_entry("(0x%p, size:%llu)\n", inode,
199 (unsigned long long)fe->i_size);
Mark Fashehccd979b2005-12-15 14:31:24 -0800200
201 sb = inode->i_sb;
202 osb = OCFS2_SB(sb);
203
204 /* this means that read_inode cannot create a superblock inode
205 * today. change if needed. */
206 if (!OCFS2_IS_VALID_DINODE(fe) ||
207 !(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL))) {
Mark Fasheh6a1bd4a2007-01-03 17:06:59 -0800208 mlog(0, "Invalid dinode: i_ino=%lu, i_blkno=%llu, "
Mark Fashehccd979b2005-12-15 14:31:24 -0800209 "signature = %.*s, flags = 0x%x\n",
Mark Fashehb06970532006-03-03 10:24:33 -0800210 inode->i_ino,
211 (unsigned long long)le64_to_cpu(fe->i_blkno), 7,
Mark Fashehccd979b2005-12-15 14:31:24 -0800212 fe->i_signature, le32_to_cpu(fe->i_flags));
213 goto bail;
214 }
215
216 if (le32_to_cpu(fe->i_fs_generation) != osb->fs_generation) {
217 mlog(ML_ERROR, "file entry generation does not match "
218 "superblock! osb->fs_generation=%x, "
219 "fe->i_fs_generation=%x\n",
220 osb->fs_generation, le32_to_cpu(fe->i_fs_generation));
221 goto bail;
222 }
223
Mark Fasheh8110b072007-03-22 16:53:23 -0700224 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
225 OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
226
Mark Fashehccd979b2005-12-15 14:31:24 -0800227 inode->i_version = 1;
228 inode->i_generation = le32_to_cpu(fe->i_generation);
229 inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
230 inode->i_mode = le16_to_cpu(fe->i_mode);
231 inode->i_uid = le32_to_cpu(fe->i_uid);
232 inode->i_gid = le32_to_cpu(fe->i_gid);
Mark Fashehccd979b2005-12-15 14:31:24 -0800233
234 /* Fast symlinks will have i_size but no allocated clusters. */
235 if (S_ISLNK(inode->i_mode) && !fe->i_clusters)
236 inode->i_blocks = 0;
237 else
Mark Fasheh8110b072007-03-22 16:53:23 -0700238 inode->i_blocks = ocfs2_inode_sector_count(inode);
Mark Fashehccd979b2005-12-15 14:31:24 -0800239 inode->i_mapping->a_ops = &ocfs2_aops;
Mark Fashehccd979b2005-12-15 14:31:24 -0800240 inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
241 inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
242 inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
243 inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
244 inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
245 inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
246
247 if (OCFS2_I(inode)->ip_blkno != le64_to_cpu(fe->i_blkno))
248 mlog(ML_ERROR,
Mark Fashehb06970532006-03-03 10:24:33 -0800249 "ip_blkno %llu != i_blkno %llu!\n",
250 (unsigned long long)OCFS2_I(inode)->ip_blkno,
251 (unsigned long long)fe->i_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -0800252
Mark Fashehccd979b2005-12-15 14:31:24 -0800253 inode->i_nlink = le16_to_cpu(fe->i_links_count);
254
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700255 if (fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL))
256 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SYSTEM_FILE;
257
Mark Fashehccd979b2005-12-15 14:31:24 -0800258 if (fe->i_flags & cpu_to_le32(OCFS2_LOCAL_ALLOC_FL)) {
259 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
260 mlog(0, "local alloc inode: i_ino=%lu\n", inode->i_ino);
261 } else if (fe->i_flags & cpu_to_le32(OCFS2_BITMAP_FL)) {
262 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
263 } else if (fe->i_flags & cpu_to_le32(OCFS2_SUPER_BLOCK_FL)) {
264 mlog(0, "superblock inode: i_ino=%lu\n", inode->i_ino);
265 /* we can't actually hit this as read_inode can't
266 * handle superblocks today ;-) */
267 BUG();
268 }
269
270 switch (inode->i_mode & S_IFMT) {
271 case S_IFREG:
272 inode->i_fop = &ocfs2_fops;
273 inode->i_op = &ocfs2_file_iops;
274 i_size_write(inode, le64_to_cpu(fe->i_size));
275 break;
276 case S_IFDIR:
277 inode->i_op = &ocfs2_dir_iops;
278 inode->i_fop = &ocfs2_dops;
279 i_size_write(inode, le64_to_cpu(fe->i_size));
280 break;
281 case S_IFLNK:
282 if (ocfs2_inode_is_fast_symlink(inode))
283 inode->i_op = &ocfs2_fast_symlink_inode_operations;
284 else
285 inode->i_op = &ocfs2_symlink_inode_operations;
286 i_size_write(inode, le64_to_cpu(fe->i_size));
287 break;
288 default:
289 inode->i_op = &ocfs2_special_file_iops;
290 init_special_inode(inode, inode->i_mode,
291 inode->i_rdev);
292 break;
293 }
294
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700295 if (create_ino) {
296 inode->i_ino = ino_from_blkno(inode->i_sb,
297 le64_to_cpu(fe->i_blkno));
298
299 /*
300 * If we ever want to create system files from kernel,
301 * the generation argument to
302 * ocfs2_inode_lock_res_init() will have to change.
303 */
304 BUG_ON(fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL));
305
306 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_meta_lockres,
307 OCFS2_LOCK_TYPE_META, 0, inode);
Tiger Yang50008632007-03-20 16:01:38 -0700308
309 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres,
310 OCFS2_LOCK_TYPE_OPEN, 0, inode);
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700311 }
312
Mark Fashehccd979b2005-12-15 14:31:24 -0800313 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_rw_lockres,
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700314 OCFS2_LOCK_TYPE_RW, inode->i_generation,
315 inode);
316
Mark Fashehccd979b2005-12-15 14:31:24 -0800317 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_data_lockres,
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700318 OCFS2_LOCK_TYPE_DATA, inode->i_generation,
319 inode);
Mark Fashehccd979b2005-12-15 14:31:24 -0800320
Herbert Poetzlca4d1472006-07-03 17:27:12 -0700321 ocfs2_set_inode_flags(inode);
Herbert Poetzlca4d1472006-07-03 17:27:12 -0700322
Mark Fashehccd979b2005-12-15 14:31:24 -0800323 status = 0;
324bail:
325 mlog_exit(status);
326 return status;
327}
328
329static int ocfs2_read_locked_inode(struct inode *inode,
330 struct ocfs2_find_inode_args *args)
331{
332 struct super_block *sb;
333 struct ocfs2_super *osb;
334 struct ocfs2_dinode *fe;
335 struct buffer_head *bh = NULL;
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700336 int status, can_lock;
337 u32 generation = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -0800338
339 mlog_entry("(0x%p, 0x%p)\n", inode, args);
340
341 status = -EINVAL;
342 if (inode == NULL || inode->i_sb == NULL) {
343 mlog(ML_ERROR, "bad inode\n");
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700344 return status;
Mark Fashehccd979b2005-12-15 14:31:24 -0800345 }
346 sb = inode->i_sb;
347 osb = OCFS2_SB(sb);
348
349 if (!args) {
350 mlog(ML_ERROR, "bad inode args\n");
351 make_bad_inode(inode);
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700352 return status;
Mark Fashehccd979b2005-12-15 14:31:24 -0800353 }
354
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700355 /*
356 * To improve performance of cold-cache inode stats, we take
357 * the cluster lock here if possible.
358 *
359 * Generally, OCFS2 never trusts the contents of an inode
360 * unless it's holding a cluster lock, so taking it here isn't
361 * a correctness issue as much as it is a performance
362 * improvement.
363 *
364 * There are three times when taking the lock is not a good idea:
365 *
366 * 1) During startup, before we have initialized the DLM.
367 *
368 * 2) If we are reading certain system files which never get
369 * cluster locks (local alloc, truncate log).
370 *
371 * 3) If the process doing the iget() is responsible for
372 * orphan dir recovery. We're holding the orphan dir lock and
373 * can get into a deadlock with another process on another
374 * node in ->delete_inode().
375 *
376 * #1 and #2 can be simply solved by never taking the lock
377 * here for system files (which are the only type we read
378 * during mount). It's a heavier approach, but our main
379 * concern is user-accesible files anyway.
380 *
381 * #3 works itself out because we'll eventually take the
382 * cluster lock before trusting anything anyway.
383 */
384 can_lock = !(args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
Tiger Yang50008632007-03-20 16:01:38 -0700385 && !(args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY)
Sunil Mushranc271c5c2006-12-05 17:56:35 -0800386 && !ocfs2_mount_local(osb);
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700387
388 /*
389 * To maintain backwards compatibility with older versions of
390 * ocfs2-tools, we still store the generation value for system
391 * files. The only ones that actually matter to userspace are
392 * the journals, but it's easier and inexpensive to just flag
393 * all system files similarly.
394 */
395 if (args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
396 generation = osb->fs_generation;
397
398 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_meta_lockres,
399 OCFS2_LOCK_TYPE_META,
400 generation, inode);
401
Tiger Yang50008632007-03-20 16:01:38 -0700402 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres,
403 OCFS2_LOCK_TYPE_OPEN,
404 0, inode);
405
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700406 if (can_lock) {
Tiger Yang50008632007-03-20 16:01:38 -0700407 status = ocfs2_open_lock(inode);
408 if (status) {
409 make_bad_inode(inode);
410 mlog_errno(status);
411 return status;
412 }
Mark Fasheh4bcec182006-10-09 16:02:40 -0700413 status = ocfs2_meta_lock(inode, NULL, 0);
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700414 if (status) {
415 make_bad_inode(inode);
416 mlog_errno(status);
417 return status;
418 }
419 }
420
Tiger Yang50008632007-03-20 16:01:38 -0700421 if (args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY) {
422 status = ocfs2_try_open_lock(inode, 0);
423 if (status) {
424 make_bad_inode(inode);
425 return status;
426 }
427 }
428
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700429 status = ocfs2_read_block(osb, args->fi_blkno, &bh, 0,
430 can_lock ? inode : NULL);
Mark Fashehccd979b2005-12-15 14:31:24 -0800431 if (status < 0) {
432 mlog_errno(status);
Mark Fashehccd979b2005-12-15 14:31:24 -0800433 goto bail;
434 }
435
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700436 status = -EINVAL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800437 fe = (struct ocfs2_dinode *) bh->b_data;
438 if (!OCFS2_IS_VALID_DINODE(fe)) {
Mark Fashehb06970532006-03-03 10:24:33 -0800439 mlog(ML_ERROR, "Invalid dinode #%llu: signature = %.*s\n",
440 (unsigned long long)fe->i_blkno, 7, fe->i_signature);
Mark Fashehccd979b2005-12-15 14:31:24 -0800441 goto bail;
442 }
443
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700444 /*
445 * This is a code bug. Right now the caller needs to
446 * understand whether it is asking for a system file inode or
447 * not so the proper lock names can be built.
448 */
449 mlog_bug_on_msg(!!(fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) !=
450 !!(args->fi_flags & OCFS2_FI_FLAG_SYSFILE),
451 "Inode %llu: system file state is ambigous\n",
452 (unsigned long long)args->fi_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -0800453
454 if (S_ISCHR(le16_to_cpu(fe->i_mode)) ||
455 S_ISBLK(le16_to_cpu(fe->i_mode)))
456 inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
457
Mark Fasheh6a1bd4a2007-01-03 17:06:59 -0800458 if (ocfs2_populate_inode(inode, fe, 0) < 0)
Mark Fashehccd979b2005-12-15 14:31:24 -0800459 goto bail;
Mark Fashehccd979b2005-12-15 14:31:24 -0800460
461 BUG_ON(args->fi_blkno != le64_to_cpu(fe->i_blkno));
462
Mark Fashehccd979b2005-12-15 14:31:24 -0800463 status = 0;
464
465bail:
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700466 if (can_lock)
467 ocfs2_meta_unlock(inode, 0);
468
469 if (status < 0)
470 make_bad_inode(inode);
471
Mark Fashehccd979b2005-12-15 14:31:24 -0800472 if (args && bh)
473 brelse(bh);
474
475 mlog_exit(status);
476 return status;
477}
478
479void ocfs2_sync_blockdev(struct super_block *sb)
480{
481 sync_blockdev(sb->s_bdev);
482}
483
484static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
485 struct inode *inode,
486 struct buffer_head *fe_bh)
487{
488 int status = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -0800489 struct ocfs2_truncate_context *tc = NULL;
490 struct ocfs2_dinode *fe;
Mark Fasheh60b11392007-02-16 11:46:50 -0800491 handle_t *handle = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800492
493 mlog_entry_void();
494
495 fe = (struct ocfs2_dinode *) fe_bh->b_data;
496
Mark Fasheh3a0782d2007-01-17 12:53:31 -0800497 if (fe->i_clusters) {
Mark Fasheh60b11392007-02-16 11:46:50 -0800498 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
499 if (IS_ERR(handle)) {
500 status = PTR_ERR(handle);
501 mlog_errno(status);
502 goto out;
503 }
504
505 status = ocfs2_journal_access(handle, inode, fe_bh,
506 OCFS2_JOURNAL_ACCESS_WRITE);
507 if (status < 0) {
508 mlog_errno(status);
509 goto out;
510 }
511
512 i_size_write(inode, 0);
513
514 status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
515 if (status < 0) {
516 mlog_errno(status);
517 goto out;
518 }
519
520 ocfs2_commit_trans(osb, handle);
521 handle = NULL;
522
Mark Fasheh3a0782d2007-01-17 12:53:31 -0800523 status = ocfs2_prepare_truncate(osb, inode, fe_bh, &tc);
524 if (status < 0) {
525 mlog_errno(status);
526 goto out;
527 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800528
Mark Fasheh3a0782d2007-01-17 12:53:31 -0800529 status = ocfs2_commit_truncate(osb, inode, fe_bh, tc);
530 if (status < 0) {
531 mlog_errno(status);
532 goto out;
533 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800534 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800535
Mark Fasheh60b11392007-02-16 11:46:50 -0800536out:
537 if (handle)
538 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800539 mlog_exit(status);
540 return status;
541}
542
543static int ocfs2_remove_inode(struct inode *inode,
544 struct buffer_head *di_bh,
545 struct inode *orphan_dir_inode,
546 struct buffer_head *orphan_dir_bh)
547{
548 int status;
549 struct inode *inode_alloc_inode = NULL;
550 struct buffer_head *inode_alloc_bh = NULL;
Mark Fasheh1fabe142006-10-09 18:11:45 -0700551 handle_t *handle;
Mark Fashehccd979b2005-12-15 14:31:24 -0800552 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
553 struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
554
555 inode_alloc_inode =
556 ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE,
557 le16_to_cpu(di->i_suballoc_slot));
558 if (!inode_alloc_inode) {
559 status = -EEXIST;
560 mlog_errno(status);
561 goto bail;
562 }
563
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800564 mutex_lock(&inode_alloc_inode->i_mutex);
Mark Fasheh4bcec182006-10-09 16:02:40 -0700565 status = ocfs2_meta_lock(inode_alloc_inode, &inode_alloc_bh, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -0800566 if (status < 0) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800567 mutex_unlock(&inode_alloc_inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -0800568
569 mlog_errno(status);
570 goto bail;
571 }
572
Mark Fasheh65eff9c2006-10-09 17:26:22 -0700573 handle = ocfs2_start_trans(osb, OCFS2_DELETE_INODE_CREDITS);
Mark Fashehccd979b2005-12-15 14:31:24 -0800574 if (IS_ERR(handle)) {
575 status = PTR_ERR(handle);
576 mlog_errno(status);
577 goto bail_unlock;
578 }
579
580 status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
581 orphan_dir_bh);
582 if (status < 0) {
583 mlog_errno(status);
584 goto bail_commit;
585 }
586
587 /* set the inodes dtime */
588 status = ocfs2_journal_access(handle, inode, di_bh,
589 OCFS2_JOURNAL_ACCESS_WRITE);
590 if (status < 0) {
591 mlog_errno(status);
592 goto bail_commit;
593 }
594
595 di->i_dtime = cpu_to_le64(CURRENT_TIME.tv_sec);
596 le32_and_cpu(&di->i_flags, ~(OCFS2_VALID_FL | OCFS2_ORPHANED_FL));
597
598 status = ocfs2_journal_dirty(handle, di_bh);
599 if (status < 0) {
600 mlog_errno(status);
601 goto bail_commit;
602 }
603
604 ocfs2_remove_from_cache(inode, di_bh);
605
606 status = ocfs2_free_dinode(handle, inode_alloc_inode,
607 inode_alloc_bh, di);
608 if (status < 0)
609 mlog_errno(status);
610
611bail_commit:
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700612 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800613bail_unlock:
614 ocfs2_meta_unlock(inode_alloc_inode, 1);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800615 mutex_unlock(&inode_alloc_inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -0800616 brelse(inode_alloc_bh);
617bail:
618 iput(inode_alloc_inode);
619
620 return status;
621}
622
Mark Fashehb4df6ed2006-02-22 17:35:08 -0800623/*
624 * Serialize with orphan dir recovery. If the process doing
625 * recovery on this orphan dir does an iget() with the dir
626 * i_mutex held, we'll deadlock here. Instead we detect this
627 * and exit early - recovery will wipe this inode for us.
628 */
629static int ocfs2_check_orphan_recovery_state(struct ocfs2_super *osb,
630 int slot)
631{
632 int ret = 0;
633
634 spin_lock(&osb->osb_lock);
635 if (ocfs2_node_map_test_bit(osb, &osb->osb_recovering_orphan_dirs, slot)) {
636 mlog(0, "Recovery is happening on orphan dir %d, will skip "
637 "this inode\n", slot);
638 ret = -EDEADLK;
639 goto out;
640 }
641 /* This signals to the orphan recovery process that it should
642 * wait for us to handle the wipe. */
643 osb->osb_orphan_wipes[slot]++;
644out:
645 spin_unlock(&osb->osb_lock);
646 return ret;
647}
648
649static void ocfs2_signal_wipe_completion(struct ocfs2_super *osb,
650 int slot)
651{
652 spin_lock(&osb->osb_lock);
653 osb->osb_orphan_wipes[slot]--;
654 spin_unlock(&osb->osb_lock);
655
656 wake_up(&osb->osb_wipe_event);
657}
658
Mark Fashehccd979b2005-12-15 14:31:24 -0800659static int ocfs2_wipe_inode(struct inode *inode,
660 struct buffer_head *di_bh)
661{
662 int status, orphaned_slot;
663 struct inode *orphan_dir_inode = NULL;
664 struct buffer_head *orphan_dir_bh = NULL;
665 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
Tiger Yang50008632007-03-20 16:01:38 -0700666 struct ocfs2_dinode *di;
Mark Fashehccd979b2005-12-15 14:31:24 -0800667
Tiger Yang50008632007-03-20 16:01:38 -0700668 di = (struct ocfs2_dinode *) di_bh->b_data;
669 orphaned_slot = le16_to_cpu(di->i_orphaned_slot);
Mark Fashehb4df6ed2006-02-22 17:35:08 -0800670
671 status = ocfs2_check_orphan_recovery_state(osb, orphaned_slot);
672 if (status)
673 return status;
674
Mark Fashehccd979b2005-12-15 14:31:24 -0800675 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
676 ORPHAN_DIR_SYSTEM_INODE,
677 orphaned_slot);
678 if (!orphan_dir_inode) {
679 status = -EEXIST;
680 mlog_errno(status);
681 goto bail;
682 }
683
684 /* Lock the orphan dir. The lock will be held for the entire
685 * delete_inode operation. We do this now to avoid races with
686 * recovery completion on other nodes. */
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800687 mutex_lock(&orphan_dir_inode->i_mutex);
Mark Fasheh4bcec182006-10-09 16:02:40 -0700688 status = ocfs2_meta_lock(orphan_dir_inode, &orphan_dir_bh, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -0800689 if (status < 0) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800690 mutex_unlock(&orphan_dir_inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -0800691
692 mlog_errno(status);
693 goto bail;
694 }
695
696 /* we do this while holding the orphan dir lock because we
697 * don't want recovery being run from another node to vote for
698 * an inode delete on us -- this will result in two nodes
699 * truncating the same file! */
700 status = ocfs2_truncate_for_delete(osb, inode, di_bh);
701 if (status < 0) {
702 mlog_errno(status);
703 goto bail_unlock_dir;
704 }
705
706 status = ocfs2_remove_inode(inode, di_bh, orphan_dir_inode,
707 orphan_dir_bh);
708 if (status < 0)
709 mlog_errno(status);
710
711bail_unlock_dir:
712 ocfs2_meta_unlock(orphan_dir_inode, 1);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800713 mutex_unlock(&orphan_dir_inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -0800714 brelse(orphan_dir_bh);
715bail:
716 iput(orphan_dir_inode);
Mark Fashehb4df6ed2006-02-22 17:35:08 -0800717 ocfs2_signal_wipe_completion(osb, orphaned_slot);
Mark Fashehccd979b2005-12-15 14:31:24 -0800718
719 return status;
720}
721
722/* There is a series of simple checks that should be done before a
723 * vote is even considered. Encapsulate those in this function. */
724static int ocfs2_inode_is_valid_to_delete(struct inode *inode)
725{
726 int ret = 0;
727 struct ocfs2_inode_info *oi = OCFS2_I(inode);
728 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
729
730 /* We shouldn't be getting here for the root directory
731 * inode.. */
732 if (inode == osb->root_inode) {
733 mlog(ML_ERROR, "Skipping delete of root inode.\n");
734 goto bail;
735 }
736
737 /* If we're coming from process_vote we can't go into our own
738 * voting [hello, deadlock city!], so unforuntately we just
739 * have to skip deleting this guy. That's OK though because
740 * the node who's doing the actual deleting should handle it
741 * anyway. */
742 if (current == osb->vote_task) {
743 mlog(0, "Skipping delete of %lu because we're currently "
744 "in process_vote\n", inode->i_ino);
745 goto bail;
746 }
747
748 spin_lock(&oi->ip_lock);
749 /* OCFS2 *never* deletes system files. This should technically
750 * never get here as system file inodes should always have a
751 * positive link count. */
752 if (oi->ip_flags & OCFS2_INODE_SYSTEM_FILE) {
Mark Fashehb06970532006-03-03 10:24:33 -0800753 mlog(ML_ERROR, "Skipping delete of system file %llu\n",
754 (unsigned long long)oi->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -0800755 goto bail_unlock;
756 }
757
758 /* If we have voted "yes" on the wipe of this inode for
759 * another node, it will be marked here so we can safely skip
760 * it. Recovery will cleanup any inodes we might inadvertantly
761 * skip here. */
762 if (oi->ip_flags & OCFS2_INODE_SKIP_DELETE) {
763 mlog(0, "Skipping delete of %lu because another node "
764 "has done this for us.\n", inode->i_ino);
765 goto bail_unlock;
766 }
767
768 ret = 1;
769bail_unlock:
770 spin_unlock(&oi->ip_lock);
771bail:
772 return ret;
773}
774
775/* Query the cluster to determine whether we should wipe an inode from
776 * disk or not.
777 *
778 * Requires the inode to have the cluster lock. */
779static int ocfs2_query_inode_wipe(struct inode *inode,
780 struct buffer_head *di_bh,
781 int *wipe)
782{
783 int status = 0;
784 struct ocfs2_inode_info *oi = OCFS2_I(inode);
785 struct ocfs2_dinode *di;
786
787 *wipe = 0;
788
789 /* While we were waiting for the cluster lock in
790 * ocfs2_delete_inode, another node might have asked to delete
791 * the inode. Recheck our flags to catch this. */
792 if (!ocfs2_inode_is_valid_to_delete(inode)) {
Mark Fashehb06970532006-03-03 10:24:33 -0800793 mlog(0, "Skipping delete of %llu because flags changed\n",
794 (unsigned long long)oi->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -0800795 goto bail;
796 }
797
798 /* Now that we have an up to date inode, we can double check
799 * the link count. */
800 if (inode->i_nlink) {
Mark Fashehb06970532006-03-03 10:24:33 -0800801 mlog(0, "Skipping delete of %llu because nlink = %u\n",
802 (unsigned long long)oi->ip_blkno, inode->i_nlink);
Mark Fashehccd979b2005-12-15 14:31:24 -0800803 goto bail;
804 }
805
806 /* Do some basic inode verification... */
807 di = (struct ocfs2_dinode *) di_bh->b_data;
808 if (!(di->i_flags & cpu_to_le32(OCFS2_ORPHANED_FL))) {
809 /* for lack of a better error? */
810 status = -EEXIST;
811 mlog(ML_ERROR,
Mark Fashehb06970532006-03-03 10:24:33 -0800812 "Inode %llu (on-disk %llu) not orphaned! "
Mark Fashehccd979b2005-12-15 14:31:24 -0800813 "Disk flags 0x%x, inode flags 0x%x\n",
Mark Fashehb06970532006-03-03 10:24:33 -0800814 (unsigned long long)oi->ip_blkno,
815 (unsigned long long)di->i_blkno, di->i_flags,
816 oi->ip_flags);
Mark Fashehccd979b2005-12-15 14:31:24 -0800817 goto bail;
818 }
819
820 /* has someone already deleted us?! baaad... */
821 if (di->i_dtime) {
822 status = -EEXIST;
823 mlog_errno(status);
824 goto bail;
825 }
826
Mark Fasheh6f16bf62007-03-20 17:17:54 -0700827 /*
828 * This is how ocfs2 determines whether an inode is still live
829 * within the cluster. Every node takes a shared read lock on
830 * the inode open lock in ocfs2_read_locked_inode(). When we
831 * get to ->delete_inode(), each node tries to convert it's
832 * lock to an exclusive. Trylocks are serialized by the inode
833 * meta data lock. If the upconvert suceeds, we know the inode
834 * is no longer live and can be deleted.
835 *
836 * Though we call this with the meta data lock held, the
837 * trylock keeps us from ABBA deadlock.
838 */
839 status = ocfs2_try_open_lock(inode, 1);
Tiger Yang50008632007-03-20 16:01:38 -0700840 if (status == -EAGAIN) {
Mark Fashehccd979b2005-12-15 14:31:24 -0800841 status = 0;
Mark Fashehb06970532006-03-03 10:24:33 -0800842 mlog(0, "Skipping delete of %llu because it is in use on"
843 "other nodes\n", (unsigned long long)oi->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -0800844 goto bail;
845 }
846 if (status < 0) {
847 mlog_errno(status);
848 goto bail;
849 }
850
Tiger Yang50008632007-03-20 16:01:38 -0700851 *wipe = 1;
852 mlog(0, "Inode %llu is ok to wipe from orphan dir %u\n",
853 (unsigned long long)oi->ip_blkno,
854 le16_to_cpu(di->i_orphaned_slot));
Mark Fashehccd979b2005-12-15 14:31:24 -0800855
856bail:
857 return status;
858}
859
860/* Support function for ocfs2_delete_inode. Will help us keep the
861 * inode data in a consistent state for clear_inode. Always truncates
862 * pages, optionally sync's them first. */
863static void ocfs2_cleanup_delete_inode(struct inode *inode,
864 int sync_data)
865{
Mark Fashehb06970532006-03-03 10:24:33 -0800866 mlog(0, "Cleanup inode %llu, sync = %d\n",
867 (unsigned long long)OCFS2_I(inode)->ip_blkno, sync_data);
Mark Fashehccd979b2005-12-15 14:31:24 -0800868 if (sync_data)
869 write_inode_now(inode, 1);
870 truncate_inode_pages(&inode->i_data, 0);
871}
872
873void ocfs2_delete_inode(struct inode *inode)
874{
875 int wipe, status;
876 sigset_t blocked, oldset;
877 struct buffer_head *di_bh = NULL;
878
879 mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
880
881 if (is_bad_inode(inode)) {
882 mlog(0, "Skipping delete of bad inode\n");
883 goto bail;
884 }
885
886 if (!ocfs2_inode_is_valid_to_delete(inode)) {
887 /* It's probably not necessary to truncate_inode_pages
888 * here but we do it for safety anyway (it will most
889 * likely be a no-op anyway) */
890 ocfs2_cleanup_delete_inode(inode, 0);
891 goto bail;
892 }
893
894 /* We want to block signals in delete_inode as the lock and
895 * messaging paths may return us -ERESTARTSYS. Which would
896 * cause us to exit early, resulting in inodes being orphaned
897 * forever. */
898 sigfillset(&blocked);
899 status = sigprocmask(SIG_BLOCK, &blocked, &oldset);
900 if (status < 0) {
901 mlog_errno(status);
902 ocfs2_cleanup_delete_inode(inode, 1);
903 goto bail;
904 }
905
906 /* Lock down the inode. This gives us an up to date view of
907 * it's metadata (for verification), and allows us to
908 * serialize delete_inode votes.
909 *
910 * Even though we might be doing a truncate, we don't take the
911 * allocation lock here as it won't be needed - nobody will
912 * have the file open.
913 */
Mark Fasheh4bcec182006-10-09 16:02:40 -0700914 status = ocfs2_meta_lock(inode, &di_bh, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -0800915 if (status < 0) {
916 if (status != -ENOENT)
917 mlog_errno(status);
918 ocfs2_cleanup_delete_inode(inode, 0);
919 goto bail_unblock;
920 }
921
922 /* Query the cluster. This will be the final decision made
923 * before we go ahead and wipe the inode. */
924 status = ocfs2_query_inode_wipe(inode, di_bh, &wipe);
925 if (!wipe || status < 0) {
926 /* Error and inode busy vote both mean we won't be
927 * removing the inode, so they take almost the same
928 * path. */
929 if (status < 0)
930 mlog_errno(status);
931
932 /* Someone in the cluster has voted to not wipe this
933 * inode, or it was never completely orphaned. Write
934 * out the pages and exit now. */
935 ocfs2_cleanup_delete_inode(inode, 1);
936 goto bail_unlock_inode;
937 }
938
939 ocfs2_cleanup_delete_inode(inode, 0);
940
941 status = ocfs2_wipe_inode(inode, di_bh);
942 if (status < 0) {
Mark Fashehb4df6ed2006-02-22 17:35:08 -0800943 if (status != -EDEADLK)
944 mlog_errno(status);
Mark Fashehccd979b2005-12-15 14:31:24 -0800945 goto bail_unlock_inode;
946 }
947
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700948 /*
949 * Mark the inode as successfully deleted.
950 *
951 * This is important for ocfs2_clear_inode() as it will check
952 * this flag and skip any checkpointing work
953 *
954 * ocfs2_stuff_meta_lvb() also uses this flag to invalidate
955 * the LVB for other nodes.
956 */
Mark Fashehccd979b2005-12-15 14:31:24 -0800957 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_DELETED;
958
959bail_unlock_inode:
960 ocfs2_meta_unlock(inode, 1);
961 brelse(di_bh);
962bail_unblock:
963 status = sigprocmask(SIG_SETMASK, &oldset, NULL);
964 if (status < 0)
965 mlog_errno(status);
966bail:
967 clear_inode(inode);
968 mlog_exit_void();
969}
970
971void ocfs2_clear_inode(struct inode *inode)
972{
973 int status;
974 struct ocfs2_inode_info *oi = OCFS2_I(inode);
975
976 mlog_entry_void();
977
978 if (!inode)
979 goto bail;
980
Mark Fashehb06970532006-03-03 10:24:33 -0800981 mlog(0, "Clearing inode: %llu, nlink = %u\n",
982 (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_nlink);
Mark Fashehccd979b2005-12-15 14:31:24 -0800983
984 mlog_bug_on_msg(OCFS2_SB(inode->i_sb) == NULL,
985 "Inode=%lu\n", inode->i_ino);
986
Tiger Yang50008632007-03-20 16:01:38 -0700987 /* For remove delete_inode vote, we hold open lock before,
988 * now it is time to unlock PR and EX open locks. */
989 ocfs2_open_unlock(inode);
990
Mark Fashehccd979b2005-12-15 14:31:24 -0800991 /* Do these before all the other work so that we don't bounce
992 * the vote thread while waiting to destroy the locks. */
993 ocfs2_mark_lockres_freeing(&oi->ip_rw_lockres);
994 ocfs2_mark_lockres_freeing(&oi->ip_meta_lockres);
995 ocfs2_mark_lockres_freeing(&oi->ip_data_lockres);
Tiger Yang50008632007-03-20 16:01:38 -0700996 ocfs2_mark_lockres_freeing(&oi->ip_open_lockres);
Mark Fashehccd979b2005-12-15 14:31:24 -0800997
998 /* We very well may get a clear_inode before all an inodes
999 * metadata has hit disk. Of course, we can't drop any cluster
1000 * locks until the journal has finished with it. The only
1001 * exception here are successfully wiped inodes - their
1002 * metadata can now be considered to be part of the system
1003 * inodes from which it came. */
1004 if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED))
1005 ocfs2_checkpoint_inode(inode);
1006
1007 mlog_bug_on_msg(!list_empty(&oi->ip_io_markers),
Mark Fashehb06970532006-03-03 10:24:33 -08001008 "Clear inode of %llu, inode has io markers\n",
1009 (unsigned long long)oi->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08001010
Mark Fasheh83418972007-04-23 18:53:12 -07001011 ocfs2_extent_map_trunc(inode, 0);
1012
Mark Fashehccd979b2005-12-15 14:31:24 -08001013 status = ocfs2_drop_inode_locks(inode);
1014 if (status < 0)
1015 mlog_errno(status);
1016
1017 ocfs2_lock_res_free(&oi->ip_rw_lockres);
1018 ocfs2_lock_res_free(&oi->ip_meta_lockres);
1019 ocfs2_lock_res_free(&oi->ip_data_lockres);
Tiger Yang50008632007-03-20 16:01:38 -07001020 ocfs2_lock_res_free(&oi->ip_open_lockres);
Mark Fashehccd979b2005-12-15 14:31:24 -08001021
1022 ocfs2_metadata_cache_purge(inode);
1023
1024 mlog_bug_on_msg(oi->ip_metadata_cache.ci_num_cached,
Mark Fashehb06970532006-03-03 10:24:33 -08001025 "Clear inode of %llu, inode has %u cache items\n",
1026 (unsigned long long)oi->ip_blkno, oi->ip_metadata_cache.ci_num_cached);
Mark Fashehccd979b2005-12-15 14:31:24 -08001027
1028 mlog_bug_on_msg(!(oi->ip_flags & OCFS2_INODE_CACHE_INLINE),
Mark Fashehb06970532006-03-03 10:24:33 -08001029 "Clear inode of %llu, inode has a bad flag\n",
1030 (unsigned long long)oi->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08001031
1032 mlog_bug_on_msg(spin_is_locked(&oi->ip_lock),
Mark Fashehb06970532006-03-03 10:24:33 -08001033 "Clear inode of %llu, inode is locked\n",
1034 (unsigned long long)oi->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08001035
Mark Fasheh251b6ec2006-01-10 15:41:43 -08001036 mlog_bug_on_msg(!mutex_trylock(&oi->ip_io_mutex),
Mark Fashehb06970532006-03-03 10:24:33 -08001037 "Clear inode of %llu, io_mutex is locked\n",
1038 (unsigned long long)oi->ip_blkno);
Mark Fasheh251b6ec2006-01-10 15:41:43 -08001039 mutex_unlock(&oi->ip_io_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -08001040
1041 /*
1042 * down_trylock() returns 0, down_write_trylock() returns 1
1043 * kernel 1, world 0
1044 */
1045 mlog_bug_on_msg(!down_write_trylock(&oi->ip_alloc_sem),
Mark Fashehb06970532006-03-03 10:24:33 -08001046 "Clear inode of %llu, alloc_sem is locked\n",
1047 (unsigned long long)oi->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08001048 up_write(&oi->ip_alloc_sem);
1049
1050 mlog_bug_on_msg(oi->ip_open_count,
Mark Fashehb06970532006-03-03 10:24:33 -08001051 "Clear inode of %llu has open count %d\n",
1052 (unsigned long long)oi->ip_blkno, oi->ip_open_count);
Mark Fashehccd979b2005-12-15 14:31:24 -08001053
1054 /* Clear all other flags. */
1055 oi->ip_flags = OCFS2_INODE_CACHE_INLINE;
1056 oi->ip_created_trans = 0;
1057 oi->ip_last_trans = 0;
1058 oi->ip_dir_start_lookup = 0;
1059 oi->ip_blkno = 0ULL;
1060
1061bail:
1062 mlog_exit_void();
1063}
1064
1065/* Called under inode_lock, with no more references on the
1066 * struct inode, so it's safe here to check the flags field
1067 * and to manipulate i_nlink without any other locks. */
1068void ocfs2_drop_inode(struct inode *inode)
1069{
1070 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1071
1072 mlog_entry_void();
1073
Mark Fashehb06970532006-03-03 10:24:33 -08001074 mlog(0, "Drop inode %llu, nlink = %u, ip_flags = 0x%x\n",
1075 (unsigned long long)oi->ip_blkno, inode->i_nlink, oi->ip_flags);
Mark Fashehccd979b2005-12-15 14:31:24 -08001076
Mark Fasheh379dfe92006-09-08 14:21:03 -07001077 if (oi->ip_flags & OCFS2_INODE_MAYBE_ORPHANED)
1078 generic_delete_inode(inode);
1079 else
1080 generic_drop_inode(inode);
Mark Fashehccd979b2005-12-15 14:31:24 -08001081
1082 mlog_exit_void();
1083}
1084
1085/*
1086 * TODO: this should probably be merged into ocfs2_get_block
1087 *
1088 * However, you now need to pay attention to the cont_prepare_write()
1089 * stuff in ocfs2_get_block (that is, ocfs2_get_block pretty much
1090 * expects never to extend).
1091 */
1092struct buffer_head *ocfs2_bread(struct inode *inode,
1093 int block, int *err, int reada)
1094{
1095 struct buffer_head *bh = NULL;
1096 int tmperr;
1097 u64 p_blkno;
1098 int readflags = OCFS2_BH_CACHED;
1099
Mark Fashehccd979b2005-12-15 14:31:24 -08001100 if (reada)
1101 readflags |= OCFS2_BH_READAHEAD;
Mark Fashehccd979b2005-12-15 14:31:24 -08001102
1103 if (((u64)block << inode->i_sb->s_blocksize_bits) >=
1104 i_size_read(inode)) {
1105 BUG_ON(!reada);
1106 return NULL;
1107 }
1108
Joel Beckeree19a772007-03-28 18:27:07 -07001109 down_read(&OCFS2_I(inode)->ip_alloc_sem);
Mark Fasheh49cb8d22007-03-09 16:21:46 -08001110 tmperr = ocfs2_extent_map_get_blocks(inode, block, &p_blkno, NULL,
1111 NULL);
Joel Beckeree19a772007-03-28 18:27:07 -07001112 up_read(&OCFS2_I(inode)->ip_alloc_sem);
Mark Fashehccd979b2005-12-15 14:31:24 -08001113 if (tmperr < 0) {
1114 mlog_errno(tmperr);
1115 goto fail;
1116 }
1117
1118 tmperr = ocfs2_read_block(OCFS2_SB(inode->i_sb), p_blkno, &bh,
1119 readflags, inode);
1120 if (tmperr < 0)
1121 goto fail;
1122
1123 tmperr = 0;
1124
1125 *err = 0;
1126 return bh;
1127
1128fail:
1129 if (bh) {
1130 brelse(bh);
1131 bh = NULL;
1132 }
1133 *err = -EIO;
1134 return NULL;
1135}
1136
1137/*
1138 * This is called from our getattr.
1139 */
1140int ocfs2_inode_revalidate(struct dentry *dentry)
1141{
1142 struct inode *inode = dentry->d_inode;
1143 int status = 0;
1144
Mark Fashehb06970532006-03-03 10:24:33 -08001145 mlog_entry("(inode = 0x%p, ino = %llu)\n", inode,
1146 inode ? (unsigned long long)OCFS2_I(inode)->ip_blkno : 0ULL);
Mark Fashehccd979b2005-12-15 14:31:24 -08001147
1148 if (!inode) {
1149 mlog(0, "eep, no inode!\n");
1150 status = -ENOENT;
1151 goto bail;
1152 }
1153
1154 spin_lock(&OCFS2_I(inode)->ip_lock);
1155 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
1156 spin_unlock(&OCFS2_I(inode)->ip_lock);
1157 mlog(0, "inode deleted!\n");
1158 status = -ENOENT;
1159 goto bail;
1160 }
1161 spin_unlock(&OCFS2_I(inode)->ip_lock);
1162
1163 /* Let ocfs2_meta_lock do the work of updating our struct
1164 * inode for us. */
Mark Fasheh4bcec182006-10-09 16:02:40 -07001165 status = ocfs2_meta_lock(inode, NULL, 0);
Mark Fashehccd979b2005-12-15 14:31:24 -08001166 if (status < 0) {
1167 if (status != -ENOENT)
1168 mlog_errno(status);
1169 goto bail;
1170 }
1171 ocfs2_meta_unlock(inode, 0);
1172bail:
1173 mlog_exit(status);
1174
1175 return status;
1176}
1177
1178/*
1179 * Updates a disk inode from a
1180 * struct inode.
1181 * Only takes ip_lock.
1182 */
Mark Fasheh1fabe142006-10-09 18:11:45 -07001183int ocfs2_mark_inode_dirty(handle_t *handle,
Mark Fashehccd979b2005-12-15 14:31:24 -08001184 struct inode *inode,
1185 struct buffer_head *bh)
1186{
1187 int status;
1188 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
1189
Mark Fashehb06970532006-03-03 10:24:33 -08001190 mlog_entry("(inode %llu)\n",
1191 (unsigned long long)OCFS2_I(inode)->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08001192
1193 status = ocfs2_journal_access(handle, inode, bh,
1194 OCFS2_JOURNAL_ACCESS_WRITE);
1195 if (status < 0) {
1196 mlog_errno(status);
1197 goto leave;
1198 }
1199
1200 spin_lock(&OCFS2_I(inode)->ip_lock);
1201 fe->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters);
Herbert Poetzlca4d1472006-07-03 17:27:12 -07001202 fe->i_attr = cpu_to_le32(OCFS2_I(inode)->ip_attr);
Mark Fashehccd979b2005-12-15 14:31:24 -08001203 spin_unlock(&OCFS2_I(inode)->ip_lock);
1204
1205 fe->i_size = cpu_to_le64(i_size_read(inode));
1206 fe->i_links_count = cpu_to_le16(inode->i_nlink);
1207 fe->i_uid = cpu_to_le32(inode->i_uid);
1208 fe->i_gid = cpu_to_le32(inode->i_gid);
1209 fe->i_mode = cpu_to_le16(inode->i_mode);
1210 fe->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
1211 fe->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
1212 fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
1213 fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
1214 fe->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
1215 fe->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
1216
1217 status = ocfs2_journal_dirty(handle, bh);
1218 if (status < 0)
1219 mlog_errno(status);
1220
1221 status = 0;
1222leave:
1223
1224 mlog_exit(status);
1225 return status;
1226}
1227
1228/*
1229 *
1230 * Updates a struct inode from a disk inode.
1231 * does no i/o, only takes ip_lock.
1232 */
1233void ocfs2_refresh_inode(struct inode *inode,
1234 struct ocfs2_dinode *fe)
1235{
Mark Fashehccd979b2005-12-15 14:31:24 -08001236 spin_lock(&OCFS2_I(inode)->ip_lock);
1237
1238 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
Herbert Poetzlca4d1472006-07-03 17:27:12 -07001239 OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
1240 ocfs2_set_inode_flags(inode);
Mark Fashehccd979b2005-12-15 14:31:24 -08001241 i_size_write(inode, le64_to_cpu(fe->i_size));
1242 inode->i_nlink = le16_to_cpu(fe->i_links_count);
1243 inode->i_uid = le32_to_cpu(fe->i_uid);
1244 inode->i_gid = le32_to_cpu(fe->i_gid);
1245 inode->i_mode = le16_to_cpu(fe->i_mode);
Mark Fashehccd979b2005-12-15 14:31:24 -08001246 if (S_ISLNK(inode->i_mode) && le32_to_cpu(fe->i_clusters) == 0)
1247 inode->i_blocks = 0;
1248 else
Mark Fasheh8110b072007-03-22 16:53:23 -07001249 inode->i_blocks = ocfs2_inode_sector_count(inode);
Mark Fashehccd979b2005-12-15 14:31:24 -08001250 inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
1251 inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
1252 inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
1253 inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
1254 inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
1255 inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
1256
1257 spin_unlock(&OCFS2_I(inode)->ip_lock);
1258}