blob: 7aa00d5118748f392251acf5b76605350eefcd98 [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>
Mark Fashehccd979b2005-12-15 14:31:24 -080031
32#include <asm/byteorder.h>
33
34#define MLOG_MASK_PREFIX ML_INODE
35#include <cluster/masklog.h>
36
37#include "ocfs2.h"
38
39#include "alloc.h"
40#include "dlmglue.h"
41#include "extent_map.h"
42#include "file.h"
Mark Fashehb4df6ed2006-02-22 17:35:08 -080043#include "heartbeat.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080044#include "inode.h"
45#include "journal.h"
46#include "namei.h"
47#include "suballoc.h"
48#include "super.h"
49#include "symlink.h"
50#include "sysfile.h"
51#include "uptodate.h"
Tiger Yangcf1d6c72008-08-18 17:11:00 +080052#include "xattr.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080053
54#include "buffer_head_io.h"
55
Mark Fashehccd979b2005-12-15 14:31:24 -080056struct ocfs2_find_inode_args
57{
58 u64 fi_blkno;
59 unsigned long fi_ino;
60 unsigned int fi_flags;
Jan Kara5fa06132008-01-11 00:11:45 +010061 unsigned int fi_sysfile_type;
Mark Fashehccd979b2005-12-15 14:31:24 -080062};
63
Jan Kara5fa06132008-01-11 00:11:45 +010064static struct lock_class_key ocfs2_sysfile_lock_key[NUM_SYSTEM_INODES];
65
Mark Fashehccd979b2005-12-15 14:31:24 -080066static int ocfs2_read_locked_inode(struct inode *inode,
67 struct ocfs2_find_inode_args *args);
68static int ocfs2_init_locked_inode(struct inode *inode, void *opaque);
69static int ocfs2_find_actor(struct inode *inode, void *opaque);
70static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
71 struct inode *inode,
72 struct buffer_head *fe_bh);
73
Herbert Poetzlca4d1472006-07-03 17:27:12 -070074void ocfs2_set_inode_flags(struct inode *inode)
75{
76 unsigned int flags = OCFS2_I(inode)->ip_attr;
77
78 inode->i_flags &= ~(S_IMMUTABLE |
79 S_SYNC | S_APPEND | S_NOATIME | S_DIRSYNC);
80
81 if (flags & OCFS2_IMMUTABLE_FL)
82 inode->i_flags |= S_IMMUTABLE;
83
84 if (flags & OCFS2_SYNC_FL)
85 inode->i_flags |= S_SYNC;
86 if (flags & OCFS2_APPEND_FL)
87 inode->i_flags |= S_APPEND;
88 if (flags & OCFS2_NOATIME_FL)
89 inode->i_flags |= S_NOATIME;
90 if (flags & OCFS2_DIRSYNC_FL)
91 inode->i_flags |= S_DIRSYNC;
92}
93
Jan Kara6e4b0d52007-04-27 11:08:01 -070094/* Propagate flags from i_flags to OCFS2_I(inode)->ip_attr */
95void ocfs2_get_inode_flags(struct ocfs2_inode_info *oi)
96{
97 unsigned int flags = oi->vfs_inode.i_flags;
98
99 oi->ip_attr &= ~(OCFS2_SYNC_FL|OCFS2_APPEND_FL|
100 OCFS2_IMMUTABLE_FL|OCFS2_NOATIME_FL|OCFS2_DIRSYNC_FL);
101 if (flags & S_SYNC)
102 oi->ip_attr |= OCFS2_SYNC_FL;
103 if (flags & S_APPEND)
104 oi->ip_attr |= OCFS2_APPEND_FL;
105 if (flags & S_IMMUTABLE)
106 oi->ip_attr |= OCFS2_IMMUTABLE_FL;
107 if (flags & S_NOATIME)
108 oi->ip_attr |= OCFS2_NOATIME_FL;
109 if (flags & S_DIRSYNC)
110 oi->ip_attr |= OCFS2_DIRSYNC_FL;
111}
112
Jan Kara5fa06132008-01-11 00:11:45 +0100113struct inode *ocfs2_iget(struct ocfs2_super *osb, u64 blkno, unsigned flags,
114 int sysfile_type)
Mark Fashehccd979b2005-12-15 14:31:24 -0800115{
116 struct inode *inode = NULL;
117 struct super_block *sb = osb->sb;
118 struct ocfs2_find_inode_args args;
119
Mark Fashehb0697052006-03-03 10:24:33 -0800120 mlog_entry("(blkno = %llu)\n", (unsigned long long)blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -0800121
122 /* Ok. By now we've either got the offsets passed to us by the
123 * caller, or we just pulled them off the bh. Lets do some
124 * sanity checks to make sure they're OK. */
125 if (blkno == 0) {
126 inode = ERR_PTR(-EINVAL);
127 mlog_errno(PTR_ERR(inode));
128 goto bail;
129 }
130
131 args.fi_blkno = blkno;
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700132 args.fi_flags = flags;
Mark Fashehccd979b2005-12-15 14:31:24 -0800133 args.fi_ino = ino_from_blkno(sb, blkno);
Jan Kara5fa06132008-01-11 00:11:45 +0100134 args.fi_sysfile_type = sysfile_type;
Mark Fashehccd979b2005-12-15 14:31:24 -0800135
136 inode = iget5_locked(sb, args.fi_ino, ocfs2_find_actor,
137 ocfs2_init_locked_inode, &args);
138 /* inode was *not* in the inode cache. 2.6.x requires
139 * us to do our own read_inode call and unlock it
140 * afterwards. */
141 if (inode && inode->i_state & I_NEW) {
142 mlog(0, "Inode was not in inode cache, reading it.\n");
143 ocfs2_read_locked_inode(inode, &args);
144 unlock_new_inode(inode);
145 }
146 if (inode == NULL) {
147 inode = ERR_PTR(-ENOMEM);
148 mlog_errno(PTR_ERR(inode));
149 goto bail;
150 }
151 if (is_bad_inode(inode)) {
152 iput(inode);
153 inode = ERR_PTR(-ESTALE);
Mark Fashehccd979b2005-12-15 14:31:24 -0800154 goto bail;
155 }
156
157bail:
158 if (!IS_ERR(inode)) {
Mark Fashehb0697052006-03-03 10:24:33 -0800159 mlog(0, "returning inode with number %llu\n",
160 (unsigned long long)OCFS2_I(inode)->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -0800161 mlog_exit_ptr(inode);
Mark Fasheh6a1bd4a2007-01-03 17:06:59 -0800162 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800163
164 return inode;
165}
166
167
168/*
169 * here's how inodes get read from disk:
170 * iget5_locked -> find_actor -> OCFS2_FIND_ACTOR
171 * found? : return the in-memory inode
172 * not found? : get_new_inode -> OCFS2_INIT_LOCKED_INODE
173 */
174
175static int ocfs2_find_actor(struct inode *inode, void *opaque)
176{
177 struct ocfs2_find_inode_args *args = NULL;
178 struct ocfs2_inode_info *oi = OCFS2_I(inode);
179 int ret = 0;
180
181 mlog_entry("(0x%p, %lu, 0x%p)\n", inode, inode->i_ino, opaque);
182
183 args = opaque;
184
185 mlog_bug_on_msg(!inode, "No inode in find actor!\n");
186
187 if (oi->ip_blkno != args->fi_blkno)
188 goto bail;
189
Mark Fashehccd979b2005-12-15 14:31:24 -0800190 ret = 1;
191bail:
192 mlog_exit(ret);
193 return ret;
194}
195
196/*
197 * initialize the new inode, but don't do anything that would cause
198 * us to sleep.
199 * return 0 on success, 1 on failure
200 */
201static int ocfs2_init_locked_inode(struct inode *inode, void *opaque)
202{
203 struct ocfs2_find_inode_args *args = opaque;
204
205 mlog_entry("inode = %p, opaque = %p\n", inode, opaque);
206
207 inode->i_ino = args->fi_ino;
208 OCFS2_I(inode)->ip_blkno = args->fi_blkno;
Jan Kara5fa06132008-01-11 00:11:45 +0100209 if (args->fi_sysfile_type != 0)
210 lockdep_set_class(&inode->i_mutex,
211 &ocfs2_sysfile_lock_key[args->fi_sysfile_type]);
Mark Fashehccd979b2005-12-15 14:31:24 -0800212
213 mlog_exit(0);
214 return 0;
215}
216
217int ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe,
218 int create_ino)
219{
220 struct super_block *sb;
221 struct ocfs2_super *osb;
222 int status = -EINVAL;
Mark Fasheh53da4932008-07-21 14:29:16 -0700223 int use_plocks = 1;
Mark Fashehccd979b2005-12-15 14:31:24 -0800224
Mark Fashehb0697052006-03-03 10:24:33 -0800225 mlog_entry("(0x%p, size:%llu)\n", inode,
Mark Fasheh1ca1a112007-04-27 16:01:25 -0700226 (unsigned long long)le64_to_cpu(fe->i_size));
Mark Fashehccd979b2005-12-15 14:31:24 -0800227
228 sb = inode->i_sb;
229 osb = OCFS2_SB(sb);
230
Mark Fasheh53da4932008-07-21 14:29:16 -0700231 if ((osb->s_mount_opt & OCFS2_MOUNT_LOCALFLOCKS) ||
232 ocfs2_mount_local(osb) || !ocfs2_stack_supports_plocks())
233 use_plocks = 0;
234
Mark Fashehccd979b2005-12-15 14:31:24 -0800235 /* this means that read_inode cannot create a superblock inode
236 * today. change if needed. */
237 if (!OCFS2_IS_VALID_DINODE(fe) ||
238 !(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL))) {
Mark Fasheh6a1bd4a2007-01-03 17:06:59 -0800239 mlog(0, "Invalid dinode: i_ino=%lu, i_blkno=%llu, "
Mark Fashehccd979b2005-12-15 14:31:24 -0800240 "signature = %.*s, flags = 0x%x\n",
Mark Fashehb0697052006-03-03 10:24:33 -0800241 inode->i_ino,
242 (unsigned long long)le64_to_cpu(fe->i_blkno), 7,
Mark Fashehccd979b2005-12-15 14:31:24 -0800243 fe->i_signature, le32_to_cpu(fe->i_flags));
244 goto bail;
245 }
246
247 if (le32_to_cpu(fe->i_fs_generation) != osb->fs_generation) {
248 mlog(ML_ERROR, "file entry generation does not match "
249 "superblock! osb->fs_generation=%x, "
250 "fe->i_fs_generation=%x\n",
251 osb->fs_generation, le32_to_cpu(fe->i_fs_generation));
252 goto bail;
253 }
254
Mark Fasheh8110b072007-03-22 16:53:23 -0700255 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
256 OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
Mark Fasheh15b1e362007-09-07 13:58:15 -0700257 OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features);
Mark Fasheh8110b072007-03-22 16:53:23 -0700258
Mark Fashehccd979b2005-12-15 14:31:24 -0800259 inode->i_version = 1;
260 inode->i_generation = le32_to_cpu(fe->i_generation);
261 inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
262 inode->i_mode = le16_to_cpu(fe->i_mode);
263 inode->i_uid = le32_to_cpu(fe->i_uid);
264 inode->i_gid = le32_to_cpu(fe->i_gid);
Mark Fashehccd979b2005-12-15 14:31:24 -0800265
266 /* Fast symlinks will have i_size but no allocated clusters. */
267 if (S_ISLNK(inode->i_mode) && !fe->i_clusters)
268 inode->i_blocks = 0;
269 else
Mark Fasheh8110b072007-03-22 16:53:23 -0700270 inode->i_blocks = ocfs2_inode_sector_count(inode);
Mark Fashehccd979b2005-12-15 14:31:24 -0800271 inode->i_mapping->a_ops = &ocfs2_aops;
Mark Fashehccd979b2005-12-15 14:31:24 -0800272 inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
273 inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
274 inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
275 inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
276 inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
277 inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
278
279 if (OCFS2_I(inode)->ip_blkno != le64_to_cpu(fe->i_blkno))
280 mlog(ML_ERROR,
Mark Fashehb0697052006-03-03 10:24:33 -0800281 "ip_blkno %llu != i_blkno %llu!\n",
282 (unsigned long long)OCFS2_I(inode)->ip_blkno,
Mark Fasheh1ca1a112007-04-27 16:01:25 -0700283 (unsigned long long)le64_to_cpu(fe->i_blkno));
Mark Fashehccd979b2005-12-15 14:31:24 -0800284
Mark Fashehccd979b2005-12-15 14:31:24 -0800285 inode->i_nlink = le16_to_cpu(fe->i_links_count);
286
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700287 if (fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL))
288 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SYSTEM_FILE;
289
Mark Fashehccd979b2005-12-15 14:31:24 -0800290 if (fe->i_flags & cpu_to_le32(OCFS2_LOCAL_ALLOC_FL)) {
291 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
292 mlog(0, "local alloc inode: i_ino=%lu\n", inode->i_ino);
293 } else if (fe->i_flags & cpu_to_le32(OCFS2_BITMAP_FL)) {
294 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
295 } else if (fe->i_flags & cpu_to_le32(OCFS2_SUPER_BLOCK_FL)) {
296 mlog(0, "superblock inode: i_ino=%lu\n", inode->i_ino);
297 /* we can't actually hit this as read_inode can't
298 * handle superblocks today ;-) */
299 BUG();
300 }
301
302 switch (inode->i_mode & S_IFMT) {
303 case S_IFREG:
Mark Fasheh53da4932008-07-21 14:29:16 -0700304 if (use_plocks)
305 inode->i_fop = &ocfs2_fops;
306 else
307 inode->i_fop = &ocfs2_fops_no_plocks;
Mark Fashehccd979b2005-12-15 14:31:24 -0800308 inode->i_op = &ocfs2_file_iops;
309 i_size_write(inode, le64_to_cpu(fe->i_size));
310 break;
311 case S_IFDIR:
312 inode->i_op = &ocfs2_dir_iops;
Mark Fasheh53da4932008-07-21 14:29:16 -0700313 if (use_plocks)
314 inode->i_fop = &ocfs2_dops;
315 else
316 inode->i_fop = &ocfs2_dops_no_plocks;
Mark Fashehccd979b2005-12-15 14:31:24 -0800317 i_size_write(inode, le64_to_cpu(fe->i_size));
318 break;
319 case S_IFLNK:
320 if (ocfs2_inode_is_fast_symlink(inode))
321 inode->i_op = &ocfs2_fast_symlink_inode_operations;
322 else
323 inode->i_op = &ocfs2_symlink_inode_operations;
324 i_size_write(inode, le64_to_cpu(fe->i_size));
325 break;
326 default:
327 inode->i_op = &ocfs2_special_file_iops;
328 init_special_inode(inode, inode->i_mode,
329 inode->i_rdev);
330 break;
331 }
332
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700333 if (create_ino) {
334 inode->i_ino = ino_from_blkno(inode->i_sb,
335 le64_to_cpu(fe->i_blkno));
336
337 /*
338 * If we ever want to create system files from kernel,
339 * the generation argument to
340 * ocfs2_inode_lock_res_init() will have to change.
341 */
Mark Fasheh1ca1a112007-04-27 16:01:25 -0700342 BUG_ON(le32_to_cpu(fe->i_flags) & OCFS2_SYSTEM_FL);
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700343
Mark Fashehe63aecb62007-10-18 15:30:42 -0700344 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_inode_lockres,
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700345 OCFS2_LOCK_TYPE_META, 0, inode);
Tiger Yang50008632007-03-20 16:01:38 -0700346
347 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres,
348 OCFS2_LOCK_TYPE_OPEN, 0, inode);
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700349 }
350
Mark Fashehccd979b2005-12-15 14:31:24 -0800351 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_rw_lockres,
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700352 OCFS2_LOCK_TYPE_RW, inode->i_generation,
353 inode);
354
Herbert Poetzlca4d1472006-07-03 17:27:12 -0700355 ocfs2_set_inode_flags(inode);
Herbert Poetzlca4d1472006-07-03 17:27:12 -0700356
Mark Fashehccd979b2005-12-15 14:31:24 -0800357 status = 0;
358bail:
359 mlog_exit(status);
360 return status;
361}
362
363static int ocfs2_read_locked_inode(struct inode *inode,
364 struct ocfs2_find_inode_args *args)
365{
366 struct super_block *sb;
367 struct ocfs2_super *osb;
368 struct ocfs2_dinode *fe;
369 struct buffer_head *bh = NULL;
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700370 int status, can_lock;
371 u32 generation = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -0800372
373 mlog_entry("(0x%p, 0x%p)\n", inode, args);
374
375 status = -EINVAL;
376 if (inode == NULL || inode->i_sb == NULL) {
377 mlog(ML_ERROR, "bad inode\n");
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700378 return status;
Mark Fashehccd979b2005-12-15 14:31:24 -0800379 }
380 sb = inode->i_sb;
381 osb = OCFS2_SB(sb);
382
383 if (!args) {
384 mlog(ML_ERROR, "bad inode args\n");
385 make_bad_inode(inode);
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700386 return status;
Mark Fashehccd979b2005-12-15 14:31:24 -0800387 }
388
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700389 /*
390 * To improve performance of cold-cache inode stats, we take
391 * the cluster lock here if possible.
392 *
393 * Generally, OCFS2 never trusts the contents of an inode
394 * unless it's holding a cluster lock, so taking it here isn't
395 * a correctness issue as much as it is a performance
396 * improvement.
397 *
398 * There are three times when taking the lock is not a good idea:
399 *
400 * 1) During startup, before we have initialized the DLM.
401 *
402 * 2) If we are reading certain system files which never get
403 * cluster locks (local alloc, truncate log).
404 *
405 * 3) If the process doing the iget() is responsible for
406 * orphan dir recovery. We're holding the orphan dir lock and
407 * can get into a deadlock with another process on another
408 * node in ->delete_inode().
409 *
410 * #1 and #2 can be simply solved by never taking the lock
411 * here for system files (which are the only type we read
412 * during mount). It's a heavier approach, but our main
413 * concern is user-accesible files anyway.
414 *
415 * #3 works itself out because we'll eventually take the
416 * cluster lock before trusting anything anyway.
417 */
418 can_lock = !(args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
Tiger Yang50008632007-03-20 16:01:38 -0700419 && !(args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY)
Sunil Mushranc271c5c2006-12-05 17:56:35 -0800420 && !ocfs2_mount_local(osb);
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700421
422 /*
423 * To maintain backwards compatibility with older versions of
424 * ocfs2-tools, we still store the generation value for system
425 * files. The only ones that actually matter to userspace are
426 * the journals, but it's easier and inexpensive to just flag
427 * all system files similarly.
428 */
429 if (args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
430 generation = osb->fs_generation;
431
Mark Fashehe63aecb62007-10-18 15:30:42 -0700432 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_inode_lockres,
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700433 OCFS2_LOCK_TYPE_META,
434 generation, inode);
435
Tiger Yang50008632007-03-20 16:01:38 -0700436 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres,
437 OCFS2_LOCK_TYPE_OPEN,
438 0, inode);
439
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700440 if (can_lock) {
Tiger Yang50008632007-03-20 16:01:38 -0700441 status = ocfs2_open_lock(inode);
442 if (status) {
443 make_bad_inode(inode);
444 mlog_errno(status);
445 return status;
446 }
Mark Fashehe63aecb62007-10-18 15:30:42 -0700447 status = ocfs2_inode_lock(inode, NULL, 0);
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700448 if (status) {
449 make_bad_inode(inode);
450 mlog_errno(status);
451 return status;
452 }
453 }
454
Tiger Yang50008632007-03-20 16:01:38 -0700455 if (args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY) {
456 status = ocfs2_try_open_lock(inode, 0);
457 if (status) {
458 make_bad_inode(inode);
459 return status;
460 }
461 }
462
Joel Beckerda1e9092008-10-09 17:20:29 -0700463 if (can_lock)
Joel Beckerd4a8c932008-10-09 17:20:34 -0700464 status = ocfs2_read_blocks(inode, args->fi_blkno, 1, &bh,
465 OCFS2_BH_IGNORE_CACHE);
Joel Beckerda1e9092008-10-09 17:20:29 -0700466 else
467 status = ocfs2_read_blocks_sync(osb, args->fi_blkno, 1, &bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800468 if (status < 0) {
469 mlog_errno(status);
Mark Fashehccd979b2005-12-15 14:31:24 -0800470 goto bail;
471 }
472
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700473 status = -EINVAL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800474 fe = (struct ocfs2_dinode *) bh->b_data;
475 if (!OCFS2_IS_VALID_DINODE(fe)) {
Mark Fasheha46043e2007-11-19 18:40:16 -0800476 mlog(0, "Invalid dinode #%llu: signature = %.*s\n",
477 (unsigned long long)args->fi_blkno, 7,
Mark Fasheh1ca1a112007-04-27 16:01:25 -0700478 fe->i_signature);
Mark Fashehccd979b2005-12-15 14:31:24 -0800479 goto bail;
480 }
481
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700482 /*
483 * This is a code bug. Right now the caller needs to
484 * understand whether it is asking for a system file inode or
485 * not so the proper lock names can be built.
486 */
487 mlog_bug_on_msg(!!(fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) !=
488 !!(args->fi_flags & OCFS2_FI_FLAG_SYSFILE),
489 "Inode %llu: system file state is ambigous\n",
490 (unsigned long long)args->fi_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -0800491
492 if (S_ISCHR(le16_to_cpu(fe->i_mode)) ||
493 S_ISBLK(le16_to_cpu(fe->i_mode)))
494 inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
495
Mark Fasheh6a1bd4a2007-01-03 17:06:59 -0800496 if (ocfs2_populate_inode(inode, fe, 0) < 0)
Mark Fashehccd979b2005-12-15 14:31:24 -0800497 goto bail;
Mark Fashehccd979b2005-12-15 14:31:24 -0800498
499 BUG_ON(args->fi_blkno != le64_to_cpu(fe->i_blkno));
500
Mark Fashehccd979b2005-12-15 14:31:24 -0800501 status = 0;
502
503bail:
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700504 if (can_lock)
Mark Fashehe63aecb62007-10-18 15:30:42 -0700505 ocfs2_inode_unlock(inode, 0);
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700506
507 if (status < 0)
508 make_bad_inode(inode);
509
Mark Fashehccd979b2005-12-15 14:31:24 -0800510 if (args && bh)
511 brelse(bh);
512
513 mlog_exit(status);
514 return status;
515}
516
517void ocfs2_sync_blockdev(struct super_block *sb)
518{
519 sync_blockdev(sb->s_bdev);
520}
521
522static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
523 struct inode *inode,
524 struct buffer_head *fe_bh)
525{
526 int status = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -0800527 struct ocfs2_truncate_context *tc = NULL;
528 struct ocfs2_dinode *fe;
Mark Fasheh60b11392007-02-16 11:46:50 -0800529 handle_t *handle = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800530
531 mlog_entry_void();
532
533 fe = (struct ocfs2_dinode *) fe_bh->b_data;
534
Mark Fasheh1afc32b2007-09-07 14:46:51 -0700535 /*
536 * This check will also skip truncate of inodes with inline
537 * data and fast symlinks.
538 */
Mark Fasheh3a0782d2007-01-17 12:53:31 -0800539 if (fe->i_clusters) {
Joel Becker2b4e30f2008-09-03 20:03:41 -0700540 if (ocfs2_should_order_data(inode))
541 ocfs2_begin_ordered_truncate(inode, 0);
542
Mark Fasheh60b11392007-02-16 11:46:50 -0800543 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
544 if (IS_ERR(handle)) {
545 status = PTR_ERR(handle);
546 mlog_errno(status);
547 goto out;
548 }
549
550 status = ocfs2_journal_access(handle, inode, fe_bh,
551 OCFS2_JOURNAL_ACCESS_WRITE);
552 if (status < 0) {
553 mlog_errno(status);
554 goto out;
555 }
556
557 i_size_write(inode, 0);
558
559 status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
560 if (status < 0) {
561 mlog_errno(status);
562 goto out;
563 }
564
565 ocfs2_commit_trans(osb, handle);
566 handle = NULL;
567
Mark Fasheh3a0782d2007-01-17 12:53:31 -0800568 status = ocfs2_prepare_truncate(osb, inode, fe_bh, &tc);
569 if (status < 0) {
570 mlog_errno(status);
571 goto out;
572 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800573
Mark Fasheh3a0782d2007-01-17 12:53:31 -0800574 status = ocfs2_commit_truncate(osb, inode, fe_bh, tc);
575 if (status < 0) {
576 mlog_errno(status);
577 goto out;
578 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800579 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800580
Mark Fasheh60b11392007-02-16 11:46:50 -0800581out:
582 if (handle)
583 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800584 mlog_exit(status);
585 return status;
586}
587
588static int ocfs2_remove_inode(struct inode *inode,
589 struct buffer_head *di_bh,
590 struct inode *orphan_dir_inode,
591 struct buffer_head *orphan_dir_bh)
592{
593 int status;
594 struct inode *inode_alloc_inode = NULL;
595 struct buffer_head *inode_alloc_bh = NULL;
Mark Fasheh1fabe142006-10-09 18:11:45 -0700596 handle_t *handle;
Mark Fashehccd979b2005-12-15 14:31:24 -0800597 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
598 struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
599
600 inode_alloc_inode =
601 ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE,
602 le16_to_cpu(di->i_suballoc_slot));
603 if (!inode_alloc_inode) {
604 status = -EEXIST;
605 mlog_errno(status);
606 goto bail;
607 }
608
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800609 mutex_lock(&inode_alloc_inode->i_mutex);
Mark Fashehe63aecb62007-10-18 15:30:42 -0700610 status = ocfs2_inode_lock(inode_alloc_inode, &inode_alloc_bh, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -0800611 if (status < 0) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800612 mutex_unlock(&inode_alloc_inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -0800613
614 mlog_errno(status);
615 goto bail;
616 }
617
Mark Fasheh65eff9c2006-10-09 17:26:22 -0700618 handle = ocfs2_start_trans(osb, OCFS2_DELETE_INODE_CREDITS);
Mark Fashehccd979b2005-12-15 14:31:24 -0800619 if (IS_ERR(handle)) {
620 status = PTR_ERR(handle);
621 mlog_errno(status);
622 goto bail_unlock;
623 }
624
625 status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
626 orphan_dir_bh);
627 if (status < 0) {
628 mlog_errno(status);
629 goto bail_commit;
630 }
631
632 /* set the inodes dtime */
633 status = ocfs2_journal_access(handle, inode, di_bh,
634 OCFS2_JOURNAL_ACCESS_WRITE);
635 if (status < 0) {
636 mlog_errno(status);
637 goto bail_commit;
638 }
639
640 di->i_dtime = cpu_to_le64(CURRENT_TIME.tv_sec);
Marcin Slusarz4092d492007-12-25 15:52:59 +0100641 di->i_flags &= cpu_to_le32(~(OCFS2_VALID_FL | OCFS2_ORPHANED_FL));
Mark Fashehccd979b2005-12-15 14:31:24 -0800642
643 status = ocfs2_journal_dirty(handle, di_bh);
644 if (status < 0) {
645 mlog_errno(status);
646 goto bail_commit;
647 }
648
649 ocfs2_remove_from_cache(inode, di_bh);
650
651 status = ocfs2_free_dinode(handle, inode_alloc_inode,
652 inode_alloc_bh, di);
653 if (status < 0)
654 mlog_errno(status);
655
656bail_commit:
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700657 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800658bail_unlock:
Mark Fashehe63aecb62007-10-18 15:30:42 -0700659 ocfs2_inode_unlock(inode_alloc_inode, 1);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800660 mutex_unlock(&inode_alloc_inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -0800661 brelse(inode_alloc_bh);
662bail:
663 iput(inode_alloc_inode);
664
665 return status;
666}
667
Mark Fashehb4df6ed2006-02-22 17:35:08 -0800668/*
669 * Serialize with orphan dir recovery. If the process doing
670 * recovery on this orphan dir does an iget() with the dir
671 * i_mutex held, we'll deadlock here. Instead we detect this
672 * and exit early - recovery will wipe this inode for us.
673 */
674static int ocfs2_check_orphan_recovery_state(struct ocfs2_super *osb,
675 int slot)
676{
677 int ret = 0;
678
679 spin_lock(&osb->osb_lock);
680 if (ocfs2_node_map_test_bit(osb, &osb->osb_recovering_orphan_dirs, slot)) {
681 mlog(0, "Recovery is happening on orphan dir %d, will skip "
682 "this inode\n", slot);
683 ret = -EDEADLK;
684 goto out;
685 }
686 /* This signals to the orphan recovery process that it should
687 * wait for us to handle the wipe. */
688 osb->osb_orphan_wipes[slot]++;
689out:
690 spin_unlock(&osb->osb_lock);
691 return ret;
692}
693
694static void ocfs2_signal_wipe_completion(struct ocfs2_super *osb,
695 int slot)
696{
697 spin_lock(&osb->osb_lock);
698 osb->osb_orphan_wipes[slot]--;
699 spin_unlock(&osb->osb_lock);
700
701 wake_up(&osb->osb_wipe_event);
702}
703
Mark Fashehccd979b2005-12-15 14:31:24 -0800704static int ocfs2_wipe_inode(struct inode *inode,
705 struct buffer_head *di_bh)
706{
707 int status, orphaned_slot;
708 struct inode *orphan_dir_inode = NULL;
709 struct buffer_head *orphan_dir_bh = NULL;
710 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
Tiger Yang50008632007-03-20 16:01:38 -0700711 struct ocfs2_dinode *di;
Mark Fashehccd979b2005-12-15 14:31:24 -0800712
Tiger Yang50008632007-03-20 16:01:38 -0700713 di = (struct ocfs2_dinode *) di_bh->b_data;
714 orphaned_slot = le16_to_cpu(di->i_orphaned_slot);
Mark Fashehb4df6ed2006-02-22 17:35:08 -0800715
716 status = ocfs2_check_orphan_recovery_state(osb, orphaned_slot);
717 if (status)
718 return status;
719
Mark Fashehccd979b2005-12-15 14:31:24 -0800720 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
721 ORPHAN_DIR_SYSTEM_INODE,
722 orphaned_slot);
723 if (!orphan_dir_inode) {
724 status = -EEXIST;
725 mlog_errno(status);
726 goto bail;
727 }
728
729 /* Lock the orphan dir. The lock will be held for the entire
730 * delete_inode operation. We do this now to avoid races with
731 * recovery completion on other nodes. */
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800732 mutex_lock(&orphan_dir_inode->i_mutex);
Mark Fashehe63aecb62007-10-18 15:30:42 -0700733 status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -0800734 if (status < 0) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800735 mutex_unlock(&orphan_dir_inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -0800736
737 mlog_errno(status);
738 goto bail;
739 }
740
741 /* we do this while holding the orphan dir lock because we
Mark Fasheh34d024f2007-09-24 15:56:19 -0700742 * don't want recovery being run from another node to try an
743 * inode delete underneath us -- this will result in two nodes
Mark Fashehccd979b2005-12-15 14:31:24 -0800744 * truncating the same file! */
745 status = ocfs2_truncate_for_delete(osb, inode, di_bh);
746 if (status < 0) {
747 mlog_errno(status);
748 goto bail_unlock_dir;
749 }
750
Tiger Yangcf1d6c72008-08-18 17:11:00 +0800751 /*Free extended attribute resources associated with this inode.*/
752 status = ocfs2_xattr_remove(inode, di_bh);
753 if (status < 0) {
754 mlog_errno(status);
755 goto bail_unlock_dir;
756 }
757
Mark Fashehccd979b2005-12-15 14:31:24 -0800758 status = ocfs2_remove_inode(inode, di_bh, orphan_dir_inode,
759 orphan_dir_bh);
760 if (status < 0)
761 mlog_errno(status);
762
763bail_unlock_dir:
Mark Fashehe63aecb62007-10-18 15:30:42 -0700764 ocfs2_inode_unlock(orphan_dir_inode, 1);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800765 mutex_unlock(&orphan_dir_inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -0800766 brelse(orphan_dir_bh);
767bail:
768 iput(orphan_dir_inode);
Mark Fashehb4df6ed2006-02-22 17:35:08 -0800769 ocfs2_signal_wipe_completion(osb, orphaned_slot);
Mark Fashehccd979b2005-12-15 14:31:24 -0800770
771 return status;
772}
773
774/* There is a series of simple checks that should be done before a
Mark Fasheh34d024f2007-09-24 15:56:19 -0700775 * trylock is even considered. Encapsulate those in this function. */
Mark Fashehccd979b2005-12-15 14:31:24 -0800776static int ocfs2_inode_is_valid_to_delete(struct inode *inode)
777{
778 int ret = 0;
779 struct ocfs2_inode_info *oi = OCFS2_I(inode);
780 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
781
782 /* We shouldn't be getting here for the root directory
783 * inode.. */
784 if (inode == osb->root_inode) {
785 mlog(ML_ERROR, "Skipping delete of root inode.\n");
786 goto bail;
787 }
788
Mark Fasheh34d024f2007-09-24 15:56:19 -0700789 /* If we're coming from downconvert_thread we can't go into our own
Mark Fashehccd979b2005-12-15 14:31:24 -0800790 * voting [hello, deadlock city!], so unforuntately we just
791 * have to skip deleting this guy. That's OK though because
792 * the node who's doing the actual deleting should handle it
793 * anyway. */
Mark Fasheh34d024f2007-09-24 15:56:19 -0700794 if (current == osb->dc_task) {
Mark Fashehccd979b2005-12-15 14:31:24 -0800795 mlog(0, "Skipping delete of %lu because we're currently "
Mark Fasheh34d024f2007-09-24 15:56:19 -0700796 "in downconvert\n", inode->i_ino);
Mark Fashehccd979b2005-12-15 14:31:24 -0800797 goto bail;
798 }
799
800 spin_lock(&oi->ip_lock);
801 /* OCFS2 *never* deletes system files. This should technically
802 * never get here as system file inodes should always have a
803 * positive link count. */
804 if (oi->ip_flags & OCFS2_INODE_SYSTEM_FILE) {
Mark Fashehb0697052006-03-03 10:24:33 -0800805 mlog(ML_ERROR, "Skipping delete of system file %llu\n",
806 (unsigned long long)oi->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -0800807 goto bail_unlock;
808 }
809
Mark Fasheh34d024f2007-09-24 15:56:19 -0700810 /* If we have allowd wipe of this inode for another node, it
811 * will be marked here so we can safely skip it. Recovery will
812 * cleanup any inodes we might inadvertantly skip here. */
Mark Fashehccd979b2005-12-15 14:31:24 -0800813 if (oi->ip_flags & OCFS2_INODE_SKIP_DELETE) {
814 mlog(0, "Skipping delete of %lu because another node "
815 "has done this for us.\n", inode->i_ino);
816 goto bail_unlock;
817 }
818
819 ret = 1;
820bail_unlock:
821 spin_unlock(&oi->ip_lock);
822bail:
823 return ret;
824}
825
826/* Query the cluster to determine whether we should wipe an inode from
827 * disk or not.
828 *
829 * Requires the inode to have the cluster lock. */
830static int ocfs2_query_inode_wipe(struct inode *inode,
831 struct buffer_head *di_bh,
832 int *wipe)
833{
834 int status = 0;
835 struct ocfs2_inode_info *oi = OCFS2_I(inode);
836 struct ocfs2_dinode *di;
837
838 *wipe = 0;
839
840 /* While we were waiting for the cluster lock in
841 * ocfs2_delete_inode, another node might have asked to delete
842 * the inode. Recheck our flags to catch this. */
843 if (!ocfs2_inode_is_valid_to_delete(inode)) {
Mark Fashehb0697052006-03-03 10:24:33 -0800844 mlog(0, "Skipping delete of %llu because flags changed\n",
845 (unsigned long long)oi->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -0800846 goto bail;
847 }
848
849 /* Now that we have an up to date inode, we can double check
850 * the link count. */
851 if (inode->i_nlink) {
Mark Fashehb0697052006-03-03 10:24:33 -0800852 mlog(0, "Skipping delete of %llu because nlink = %u\n",
853 (unsigned long long)oi->ip_blkno, inode->i_nlink);
Mark Fashehccd979b2005-12-15 14:31:24 -0800854 goto bail;
855 }
856
857 /* Do some basic inode verification... */
858 di = (struct ocfs2_dinode *) di_bh->b_data;
859 if (!(di->i_flags & cpu_to_le32(OCFS2_ORPHANED_FL))) {
860 /* for lack of a better error? */
861 status = -EEXIST;
862 mlog(ML_ERROR,
Mark Fashehb0697052006-03-03 10:24:33 -0800863 "Inode %llu (on-disk %llu) not orphaned! "
Mark Fashehccd979b2005-12-15 14:31:24 -0800864 "Disk flags 0x%x, inode flags 0x%x\n",
Mark Fashehb0697052006-03-03 10:24:33 -0800865 (unsigned long long)oi->ip_blkno,
Mark Fasheh1ca1a112007-04-27 16:01:25 -0700866 (unsigned long long)le64_to_cpu(di->i_blkno),
867 le32_to_cpu(di->i_flags), oi->ip_flags);
Mark Fashehccd979b2005-12-15 14:31:24 -0800868 goto bail;
869 }
870
871 /* has someone already deleted us?! baaad... */
872 if (di->i_dtime) {
873 status = -EEXIST;
874 mlog_errno(status);
875 goto bail;
876 }
877
Mark Fasheh6f16bf62007-03-20 17:17:54 -0700878 /*
879 * This is how ocfs2 determines whether an inode is still live
880 * within the cluster. Every node takes a shared read lock on
881 * the inode open lock in ocfs2_read_locked_inode(). When we
882 * get to ->delete_inode(), each node tries to convert it's
883 * lock to an exclusive. Trylocks are serialized by the inode
884 * meta data lock. If the upconvert suceeds, we know the inode
885 * is no longer live and can be deleted.
886 *
887 * Though we call this with the meta data lock held, the
888 * trylock keeps us from ABBA deadlock.
889 */
890 status = ocfs2_try_open_lock(inode, 1);
Tiger Yang50008632007-03-20 16:01:38 -0700891 if (status == -EAGAIN) {
Mark Fashehccd979b2005-12-15 14:31:24 -0800892 status = 0;
Joe Perches27592362007-11-19 17:53:34 -0800893 mlog(0, "Skipping delete of %llu because it is in use on "
Mark Fashehb0697052006-03-03 10:24:33 -0800894 "other nodes\n", (unsigned long long)oi->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -0800895 goto bail;
896 }
897 if (status < 0) {
898 mlog_errno(status);
899 goto bail;
900 }
901
Tiger Yang50008632007-03-20 16:01:38 -0700902 *wipe = 1;
903 mlog(0, "Inode %llu is ok to wipe from orphan dir %u\n",
904 (unsigned long long)oi->ip_blkno,
905 le16_to_cpu(di->i_orphaned_slot));
Mark Fashehccd979b2005-12-15 14:31:24 -0800906
907bail:
908 return status;
909}
910
911/* Support function for ocfs2_delete_inode. Will help us keep the
912 * inode data in a consistent state for clear_inode. Always truncates
913 * pages, optionally sync's them first. */
914static void ocfs2_cleanup_delete_inode(struct inode *inode,
915 int sync_data)
916{
Mark Fashehb0697052006-03-03 10:24:33 -0800917 mlog(0, "Cleanup inode %llu, sync = %d\n",
918 (unsigned long long)OCFS2_I(inode)->ip_blkno, sync_data);
Mark Fashehccd979b2005-12-15 14:31:24 -0800919 if (sync_data)
920 write_inode_now(inode, 1);
921 truncate_inode_pages(&inode->i_data, 0);
922}
923
924void ocfs2_delete_inode(struct inode *inode)
925{
926 int wipe, status;
927 sigset_t blocked, oldset;
928 struct buffer_head *di_bh = NULL;
929
930 mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
931
932 if (is_bad_inode(inode)) {
933 mlog(0, "Skipping delete of bad inode\n");
934 goto bail;
935 }
936
937 if (!ocfs2_inode_is_valid_to_delete(inode)) {
938 /* It's probably not necessary to truncate_inode_pages
939 * here but we do it for safety anyway (it will most
940 * likely be a no-op anyway) */
941 ocfs2_cleanup_delete_inode(inode, 0);
942 goto bail;
943 }
944
945 /* We want to block signals in delete_inode as the lock and
946 * messaging paths may return us -ERESTARTSYS. Which would
947 * cause us to exit early, resulting in inodes being orphaned
948 * forever. */
949 sigfillset(&blocked);
950 status = sigprocmask(SIG_BLOCK, &blocked, &oldset);
951 if (status < 0) {
952 mlog_errno(status);
953 ocfs2_cleanup_delete_inode(inode, 1);
954 goto bail;
955 }
956
957 /* Lock down the inode. This gives us an up to date view of
958 * it's metadata (for verification), and allows us to
Mark Fasheh34d024f2007-09-24 15:56:19 -0700959 * serialize delete_inode on multiple nodes.
Mark Fashehccd979b2005-12-15 14:31:24 -0800960 *
961 * Even though we might be doing a truncate, we don't take the
962 * allocation lock here as it won't be needed - nobody will
963 * have the file open.
964 */
Mark Fashehe63aecb62007-10-18 15:30:42 -0700965 status = ocfs2_inode_lock(inode, &di_bh, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -0800966 if (status < 0) {
967 if (status != -ENOENT)
968 mlog_errno(status);
969 ocfs2_cleanup_delete_inode(inode, 0);
970 goto bail_unblock;
971 }
972
973 /* Query the cluster. This will be the final decision made
974 * before we go ahead and wipe the inode. */
975 status = ocfs2_query_inode_wipe(inode, di_bh, &wipe);
976 if (!wipe || status < 0) {
Mark Fasheh34d024f2007-09-24 15:56:19 -0700977 /* Error and remote inode busy both mean we won't be
Mark Fashehccd979b2005-12-15 14:31:24 -0800978 * removing the inode, so they take almost the same
979 * path. */
980 if (status < 0)
981 mlog_errno(status);
982
Mark Fasheh34d024f2007-09-24 15:56:19 -0700983 /* Someone in the cluster has disallowed a wipe of
984 * this inode, or it was never completely
985 * orphaned. Write out the pages and exit now. */
Mark Fashehccd979b2005-12-15 14:31:24 -0800986 ocfs2_cleanup_delete_inode(inode, 1);
987 goto bail_unlock_inode;
988 }
989
990 ocfs2_cleanup_delete_inode(inode, 0);
991
992 status = ocfs2_wipe_inode(inode, di_bh);
993 if (status < 0) {
Mark Fashehb4df6ed2006-02-22 17:35:08 -0800994 if (status != -EDEADLK)
995 mlog_errno(status);
Mark Fashehccd979b2005-12-15 14:31:24 -0800996 goto bail_unlock_inode;
997 }
998
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700999 /*
1000 * Mark the inode as successfully deleted.
1001 *
1002 * This is important for ocfs2_clear_inode() as it will check
1003 * this flag and skip any checkpointing work
1004 *
1005 * ocfs2_stuff_meta_lvb() also uses this flag to invalidate
1006 * the LVB for other nodes.
1007 */
Mark Fashehccd979b2005-12-15 14:31:24 -08001008 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_DELETED;
1009
1010bail_unlock_inode:
Mark Fashehe63aecb62007-10-18 15:30:42 -07001011 ocfs2_inode_unlock(inode, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -08001012 brelse(di_bh);
1013bail_unblock:
1014 status = sigprocmask(SIG_SETMASK, &oldset, NULL);
1015 if (status < 0)
1016 mlog_errno(status);
1017bail:
1018 clear_inode(inode);
1019 mlog_exit_void();
1020}
1021
1022void ocfs2_clear_inode(struct inode *inode)
1023{
1024 int status;
1025 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1026
1027 mlog_entry_void();
1028
1029 if (!inode)
1030 goto bail;
1031
Mark Fashehb0697052006-03-03 10:24:33 -08001032 mlog(0, "Clearing inode: %llu, nlink = %u\n",
1033 (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_nlink);
Mark Fashehccd979b2005-12-15 14:31:24 -08001034
1035 mlog_bug_on_msg(OCFS2_SB(inode->i_sb) == NULL,
1036 "Inode=%lu\n", inode->i_ino);
1037
Mark Fasheh34d024f2007-09-24 15:56:19 -07001038 /* To preven remote deletes we hold open lock before, now it
1039 * is time to unlock PR and EX open locks. */
Tiger Yang50008632007-03-20 16:01:38 -07001040 ocfs2_open_unlock(inode);
1041
Mark Fashehccd979b2005-12-15 14:31:24 -08001042 /* Do these before all the other work so that we don't bounce
Mark Fasheh34d024f2007-09-24 15:56:19 -07001043 * the downconvert thread while waiting to destroy the locks. */
Mark Fashehccd979b2005-12-15 14:31:24 -08001044 ocfs2_mark_lockres_freeing(&oi->ip_rw_lockres);
Mark Fashehe63aecb62007-10-18 15:30:42 -07001045 ocfs2_mark_lockres_freeing(&oi->ip_inode_lockres);
Tiger Yang50008632007-03-20 16:01:38 -07001046 ocfs2_mark_lockres_freeing(&oi->ip_open_lockres);
Mark Fashehccd979b2005-12-15 14:31:24 -08001047
1048 /* We very well may get a clear_inode before all an inodes
1049 * metadata has hit disk. Of course, we can't drop any cluster
1050 * locks until the journal has finished with it. The only
1051 * exception here are successfully wiped inodes - their
1052 * metadata can now be considered to be part of the system
1053 * inodes from which it came. */
1054 if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED))
1055 ocfs2_checkpoint_inode(inode);
1056
1057 mlog_bug_on_msg(!list_empty(&oi->ip_io_markers),
Mark Fashehb0697052006-03-03 10:24:33 -08001058 "Clear inode of %llu, inode has io markers\n",
1059 (unsigned long long)oi->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08001060
Mark Fasheh83418972007-04-23 18:53:12 -07001061 ocfs2_extent_map_trunc(inode, 0);
1062
Mark Fashehccd979b2005-12-15 14:31:24 -08001063 status = ocfs2_drop_inode_locks(inode);
1064 if (status < 0)
1065 mlog_errno(status);
1066
1067 ocfs2_lock_res_free(&oi->ip_rw_lockres);
Mark Fashehe63aecb62007-10-18 15:30:42 -07001068 ocfs2_lock_res_free(&oi->ip_inode_lockres);
Tiger Yang50008632007-03-20 16:01:38 -07001069 ocfs2_lock_res_free(&oi->ip_open_lockres);
Mark Fashehccd979b2005-12-15 14:31:24 -08001070
1071 ocfs2_metadata_cache_purge(inode);
1072
1073 mlog_bug_on_msg(oi->ip_metadata_cache.ci_num_cached,
Mark Fashehb0697052006-03-03 10:24:33 -08001074 "Clear inode of %llu, inode has %u cache items\n",
1075 (unsigned long long)oi->ip_blkno, oi->ip_metadata_cache.ci_num_cached);
Mark Fashehccd979b2005-12-15 14:31:24 -08001076
1077 mlog_bug_on_msg(!(oi->ip_flags & OCFS2_INODE_CACHE_INLINE),
Mark Fashehb0697052006-03-03 10:24:33 -08001078 "Clear inode of %llu, inode has a bad flag\n",
1079 (unsigned long long)oi->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08001080
1081 mlog_bug_on_msg(spin_is_locked(&oi->ip_lock),
Mark Fashehb0697052006-03-03 10:24:33 -08001082 "Clear inode of %llu, inode is locked\n",
1083 (unsigned long long)oi->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08001084
Mark Fasheh251b6ec2006-01-10 15:41:43 -08001085 mlog_bug_on_msg(!mutex_trylock(&oi->ip_io_mutex),
Mark Fashehb0697052006-03-03 10:24:33 -08001086 "Clear inode of %llu, io_mutex is locked\n",
1087 (unsigned long long)oi->ip_blkno);
Mark Fasheh251b6ec2006-01-10 15:41:43 -08001088 mutex_unlock(&oi->ip_io_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -08001089
1090 /*
1091 * down_trylock() returns 0, down_write_trylock() returns 1
1092 * kernel 1, world 0
1093 */
1094 mlog_bug_on_msg(!down_write_trylock(&oi->ip_alloc_sem),
Mark Fashehb0697052006-03-03 10:24:33 -08001095 "Clear inode of %llu, alloc_sem is locked\n",
1096 (unsigned long long)oi->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08001097 up_write(&oi->ip_alloc_sem);
1098
1099 mlog_bug_on_msg(oi->ip_open_count,
Mark Fashehb0697052006-03-03 10:24:33 -08001100 "Clear inode of %llu has open count %d\n",
1101 (unsigned long long)oi->ip_blkno, oi->ip_open_count);
Mark Fashehccd979b2005-12-15 14:31:24 -08001102
1103 /* Clear all other flags. */
1104 oi->ip_flags = OCFS2_INODE_CACHE_INLINE;
1105 oi->ip_created_trans = 0;
1106 oi->ip_last_trans = 0;
1107 oi->ip_dir_start_lookup = 0;
1108 oi->ip_blkno = 0ULL;
Sunil Mushranae0dff62008-10-22 13:24:29 -07001109
1110 /*
1111 * ip_jinode is used to track txns against this inode. We ensure that
1112 * the journal is flushed before journal shutdown. Thus it is safe to
1113 * have inodes get cleaned up after journal shutdown.
1114 */
Joel Becker2b4e30f2008-09-03 20:03:41 -07001115 jbd2_journal_release_jbd_inode(OCFS2_SB(inode->i_sb)->journal->j_journal,
1116 &oi->ip_jinode);
Mark Fashehccd979b2005-12-15 14:31:24 -08001117
1118bail:
1119 mlog_exit_void();
1120}
1121
1122/* Called under inode_lock, with no more references on the
1123 * struct inode, so it's safe here to check the flags field
1124 * and to manipulate i_nlink without any other locks. */
1125void ocfs2_drop_inode(struct inode *inode)
1126{
1127 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1128
1129 mlog_entry_void();
1130
Mark Fashehb0697052006-03-03 10:24:33 -08001131 mlog(0, "Drop inode %llu, nlink = %u, ip_flags = 0x%x\n",
1132 (unsigned long long)oi->ip_blkno, inode->i_nlink, oi->ip_flags);
Mark Fashehccd979b2005-12-15 14:31:24 -08001133
Mark Fasheh379dfe92006-09-08 14:21:03 -07001134 if (oi->ip_flags & OCFS2_INODE_MAYBE_ORPHANED)
1135 generic_delete_inode(inode);
1136 else
1137 generic_drop_inode(inode);
Mark Fashehccd979b2005-12-15 14:31:24 -08001138
1139 mlog_exit_void();
1140}
1141
1142/*
Mark Fashehccd979b2005-12-15 14:31:24 -08001143 * This is called from our getattr.
1144 */
1145int ocfs2_inode_revalidate(struct dentry *dentry)
1146{
1147 struct inode *inode = dentry->d_inode;
1148 int status = 0;
1149
Mark Fashehb0697052006-03-03 10:24:33 -08001150 mlog_entry("(inode = 0x%p, ino = %llu)\n", inode,
1151 inode ? (unsigned long long)OCFS2_I(inode)->ip_blkno : 0ULL);
Mark Fashehccd979b2005-12-15 14:31:24 -08001152
1153 if (!inode) {
1154 mlog(0, "eep, no inode!\n");
1155 status = -ENOENT;
1156 goto bail;
1157 }
1158
1159 spin_lock(&OCFS2_I(inode)->ip_lock);
1160 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
1161 spin_unlock(&OCFS2_I(inode)->ip_lock);
1162 mlog(0, "inode deleted!\n");
1163 status = -ENOENT;
1164 goto bail;
1165 }
1166 spin_unlock(&OCFS2_I(inode)->ip_lock);
1167
Mark Fashehe63aecb62007-10-18 15:30:42 -07001168 /* Let ocfs2_inode_lock do the work of updating our struct
Mark Fashehccd979b2005-12-15 14:31:24 -08001169 * inode for us. */
Mark Fashehe63aecb62007-10-18 15:30:42 -07001170 status = ocfs2_inode_lock(inode, NULL, 0);
Mark Fashehccd979b2005-12-15 14:31:24 -08001171 if (status < 0) {
1172 if (status != -ENOENT)
1173 mlog_errno(status);
1174 goto bail;
1175 }
Mark Fashehe63aecb62007-10-18 15:30:42 -07001176 ocfs2_inode_unlock(inode, 0);
Mark Fashehccd979b2005-12-15 14:31:24 -08001177bail:
1178 mlog_exit(status);
1179
1180 return status;
1181}
1182
1183/*
1184 * Updates a disk inode from a
1185 * struct inode.
1186 * Only takes ip_lock.
1187 */
Mark Fasheh1fabe142006-10-09 18:11:45 -07001188int ocfs2_mark_inode_dirty(handle_t *handle,
Mark Fashehccd979b2005-12-15 14:31:24 -08001189 struct inode *inode,
1190 struct buffer_head *bh)
1191{
1192 int status;
1193 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
1194
Mark Fashehb0697052006-03-03 10:24:33 -08001195 mlog_entry("(inode %llu)\n",
1196 (unsigned long long)OCFS2_I(inode)->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08001197
1198 status = ocfs2_journal_access(handle, inode, bh,
1199 OCFS2_JOURNAL_ACCESS_WRITE);
1200 if (status < 0) {
1201 mlog_errno(status);
1202 goto leave;
1203 }
1204
1205 spin_lock(&OCFS2_I(inode)->ip_lock);
1206 fe->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters);
Jan Kara6e4b0d52007-04-27 11:08:01 -07001207 ocfs2_get_inode_flags(OCFS2_I(inode));
Herbert Poetzlca4d1472006-07-03 17:27:12 -07001208 fe->i_attr = cpu_to_le32(OCFS2_I(inode)->ip_attr);
Mark Fasheh15b1e362007-09-07 13:58:15 -07001209 fe->i_dyn_features = cpu_to_le16(OCFS2_I(inode)->ip_dyn_features);
Mark Fashehccd979b2005-12-15 14:31:24 -08001210 spin_unlock(&OCFS2_I(inode)->ip_lock);
1211
1212 fe->i_size = cpu_to_le64(i_size_read(inode));
1213 fe->i_links_count = cpu_to_le16(inode->i_nlink);
1214 fe->i_uid = cpu_to_le32(inode->i_uid);
1215 fe->i_gid = cpu_to_le32(inode->i_gid);
1216 fe->i_mode = cpu_to_le16(inode->i_mode);
1217 fe->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
1218 fe->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
1219 fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
1220 fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
1221 fe->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
1222 fe->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
1223
1224 status = ocfs2_journal_dirty(handle, bh);
1225 if (status < 0)
1226 mlog_errno(status);
1227
1228 status = 0;
1229leave:
1230
1231 mlog_exit(status);
1232 return status;
1233}
1234
1235/*
1236 *
1237 * Updates a struct inode from a disk inode.
1238 * does no i/o, only takes ip_lock.
1239 */
1240void ocfs2_refresh_inode(struct inode *inode,
1241 struct ocfs2_dinode *fe)
1242{
Mark Fashehccd979b2005-12-15 14:31:24 -08001243 spin_lock(&OCFS2_I(inode)->ip_lock);
1244
1245 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
Herbert Poetzlca4d1472006-07-03 17:27:12 -07001246 OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
Mark Fasheh15b1e362007-09-07 13:58:15 -07001247 OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features);
Herbert Poetzlca4d1472006-07-03 17:27:12 -07001248 ocfs2_set_inode_flags(inode);
Mark Fashehccd979b2005-12-15 14:31:24 -08001249 i_size_write(inode, le64_to_cpu(fe->i_size));
1250 inode->i_nlink = le16_to_cpu(fe->i_links_count);
1251 inode->i_uid = le32_to_cpu(fe->i_uid);
1252 inode->i_gid = le32_to_cpu(fe->i_gid);
1253 inode->i_mode = le16_to_cpu(fe->i_mode);
Mark Fashehccd979b2005-12-15 14:31:24 -08001254 if (S_ISLNK(inode->i_mode) && le32_to_cpu(fe->i_clusters) == 0)
1255 inode->i_blocks = 0;
1256 else
Mark Fasheh8110b072007-03-22 16:53:23 -07001257 inode->i_blocks = ocfs2_inode_sector_count(inode);
Mark Fashehccd979b2005-12-15 14:31:24 -08001258 inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
1259 inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
1260 inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
1261 inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
1262 inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
1263 inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
1264
1265 spin_unlock(&OCFS2_I(inode)->ip_lock);
1266}