blob: 9d92c859ac949a56e171b42a646ffd2e7bf8d7d4 [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 Fashehb06970532006-03-03 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 Fashehb06970532006-03-03 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 Fashehb06970532006-03-03 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 Fashehb06970532006-03-03 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 Fashehb06970532006-03-03 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
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700463 status = ocfs2_read_block(osb, args->fi_blkno, &bh, 0,
464 can_lock ? inode : NULL);
Mark Fashehccd979b2005-12-15 14:31:24 -0800465 if (status < 0) {
466 mlog_errno(status);
Mark Fashehccd979b2005-12-15 14:31:24 -0800467 goto bail;
468 }
469
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700470 status = -EINVAL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800471 fe = (struct ocfs2_dinode *) bh->b_data;
472 if (!OCFS2_IS_VALID_DINODE(fe)) {
Mark Fasheha46043e02007-11-19 18:40:16 -0800473 mlog(0, "Invalid dinode #%llu: signature = %.*s\n",
474 (unsigned long long)args->fi_blkno, 7,
Mark Fasheh1ca1a112007-04-27 16:01:25 -0700475 fe->i_signature);
Mark Fashehccd979b2005-12-15 14:31:24 -0800476 goto bail;
477 }
478
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700479 /*
480 * This is a code bug. Right now the caller needs to
481 * understand whether it is asking for a system file inode or
482 * not so the proper lock names can be built.
483 */
484 mlog_bug_on_msg(!!(fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) !=
485 !!(args->fi_flags & OCFS2_FI_FLAG_SYSFILE),
486 "Inode %llu: system file state is ambigous\n",
487 (unsigned long long)args->fi_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -0800488
489 if (S_ISCHR(le16_to_cpu(fe->i_mode)) ||
490 S_ISBLK(le16_to_cpu(fe->i_mode)))
491 inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
492
Mark Fasheh6a1bd4a2007-01-03 17:06:59 -0800493 if (ocfs2_populate_inode(inode, fe, 0) < 0)
Mark Fashehccd979b2005-12-15 14:31:24 -0800494 goto bail;
Mark Fashehccd979b2005-12-15 14:31:24 -0800495
496 BUG_ON(args->fi_blkno != le64_to_cpu(fe->i_blkno));
497
Mark Fashehccd979b2005-12-15 14:31:24 -0800498 status = 0;
499
500bail:
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700501 if (can_lock)
Mark Fashehe63aecb62007-10-18 15:30:42 -0700502 ocfs2_inode_unlock(inode, 0);
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700503
504 if (status < 0)
505 make_bad_inode(inode);
506
Mark Fashehccd979b2005-12-15 14:31:24 -0800507 if (args && bh)
508 brelse(bh);
509
510 mlog_exit(status);
511 return status;
512}
513
514void ocfs2_sync_blockdev(struct super_block *sb)
515{
516 sync_blockdev(sb->s_bdev);
517}
518
519static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
520 struct inode *inode,
521 struct buffer_head *fe_bh)
522{
523 int status = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -0800524 struct ocfs2_truncate_context *tc = NULL;
525 struct ocfs2_dinode *fe;
Mark Fasheh60b11392007-02-16 11:46:50 -0800526 handle_t *handle = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800527
528 mlog_entry_void();
529
530 fe = (struct ocfs2_dinode *) fe_bh->b_data;
531
Mark Fasheh1afc32b2007-09-07 14:46:51 -0700532 /*
533 * This check will also skip truncate of inodes with inline
534 * data and fast symlinks.
535 */
Mark Fasheh3a0782d2007-01-17 12:53:31 -0800536 if (fe->i_clusters) {
Joel Becker2b4e30f2008-09-03 20:03:41 -0700537 if (ocfs2_should_order_data(inode))
538 ocfs2_begin_ordered_truncate(inode, 0);
539
Mark Fasheh60b11392007-02-16 11:46:50 -0800540 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
541 if (IS_ERR(handle)) {
542 status = PTR_ERR(handle);
543 mlog_errno(status);
544 goto out;
545 }
546
547 status = ocfs2_journal_access(handle, inode, fe_bh,
548 OCFS2_JOURNAL_ACCESS_WRITE);
549 if (status < 0) {
550 mlog_errno(status);
551 goto out;
552 }
553
554 i_size_write(inode, 0);
555
556 status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
557 if (status < 0) {
558 mlog_errno(status);
559 goto out;
560 }
561
562 ocfs2_commit_trans(osb, handle);
563 handle = NULL;
564
Mark Fasheh3a0782d2007-01-17 12:53:31 -0800565 status = ocfs2_prepare_truncate(osb, inode, fe_bh, &tc);
566 if (status < 0) {
567 mlog_errno(status);
568 goto out;
569 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800570
Mark Fasheh3a0782d2007-01-17 12:53:31 -0800571 status = ocfs2_commit_truncate(osb, inode, fe_bh, tc);
572 if (status < 0) {
573 mlog_errno(status);
574 goto out;
575 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800576 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800577
Mark Fasheh60b11392007-02-16 11:46:50 -0800578out:
579 if (handle)
580 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800581 mlog_exit(status);
582 return status;
583}
584
585static int ocfs2_remove_inode(struct inode *inode,
586 struct buffer_head *di_bh,
587 struct inode *orphan_dir_inode,
588 struct buffer_head *orphan_dir_bh)
589{
590 int status;
591 struct inode *inode_alloc_inode = NULL;
592 struct buffer_head *inode_alloc_bh = NULL;
Mark Fasheh1fabe142006-10-09 18:11:45 -0700593 handle_t *handle;
Mark Fashehccd979b2005-12-15 14:31:24 -0800594 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
595 struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
596
597 inode_alloc_inode =
598 ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE,
599 le16_to_cpu(di->i_suballoc_slot));
600 if (!inode_alloc_inode) {
601 status = -EEXIST;
602 mlog_errno(status);
603 goto bail;
604 }
605
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800606 mutex_lock(&inode_alloc_inode->i_mutex);
Mark Fashehe63aecb62007-10-18 15:30:42 -0700607 status = ocfs2_inode_lock(inode_alloc_inode, &inode_alloc_bh, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -0800608 if (status < 0) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800609 mutex_unlock(&inode_alloc_inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -0800610
611 mlog_errno(status);
612 goto bail;
613 }
614
Mark Fasheh65eff9c2006-10-09 17:26:22 -0700615 handle = ocfs2_start_trans(osb, OCFS2_DELETE_INODE_CREDITS);
Mark Fashehccd979b2005-12-15 14:31:24 -0800616 if (IS_ERR(handle)) {
617 status = PTR_ERR(handle);
618 mlog_errno(status);
619 goto bail_unlock;
620 }
621
622 status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
623 orphan_dir_bh);
624 if (status < 0) {
625 mlog_errno(status);
626 goto bail_commit;
627 }
628
629 /* set the inodes dtime */
630 status = ocfs2_journal_access(handle, inode, di_bh,
631 OCFS2_JOURNAL_ACCESS_WRITE);
632 if (status < 0) {
633 mlog_errno(status);
634 goto bail_commit;
635 }
636
637 di->i_dtime = cpu_to_le64(CURRENT_TIME.tv_sec);
Marcin Slusarz4092d492007-12-25 15:52:59 +0100638 di->i_flags &= cpu_to_le32(~(OCFS2_VALID_FL | OCFS2_ORPHANED_FL));
Mark Fashehccd979b2005-12-15 14:31:24 -0800639
640 status = ocfs2_journal_dirty(handle, di_bh);
641 if (status < 0) {
642 mlog_errno(status);
643 goto bail_commit;
644 }
645
646 ocfs2_remove_from_cache(inode, di_bh);
647
648 status = ocfs2_free_dinode(handle, inode_alloc_inode,
649 inode_alloc_bh, di);
650 if (status < 0)
651 mlog_errno(status);
652
653bail_commit:
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700654 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800655bail_unlock:
Mark Fashehe63aecb62007-10-18 15:30:42 -0700656 ocfs2_inode_unlock(inode_alloc_inode, 1);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800657 mutex_unlock(&inode_alloc_inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -0800658 brelse(inode_alloc_bh);
659bail:
660 iput(inode_alloc_inode);
661
662 return status;
663}
664
Mark Fashehb4df6ed2006-02-22 17:35:08 -0800665/*
666 * Serialize with orphan dir recovery. If the process doing
667 * recovery on this orphan dir does an iget() with the dir
668 * i_mutex held, we'll deadlock here. Instead we detect this
669 * and exit early - recovery will wipe this inode for us.
670 */
671static int ocfs2_check_orphan_recovery_state(struct ocfs2_super *osb,
672 int slot)
673{
674 int ret = 0;
675
676 spin_lock(&osb->osb_lock);
677 if (ocfs2_node_map_test_bit(osb, &osb->osb_recovering_orphan_dirs, slot)) {
678 mlog(0, "Recovery is happening on orphan dir %d, will skip "
679 "this inode\n", slot);
680 ret = -EDEADLK;
681 goto out;
682 }
683 /* This signals to the orphan recovery process that it should
684 * wait for us to handle the wipe. */
685 osb->osb_orphan_wipes[slot]++;
686out:
687 spin_unlock(&osb->osb_lock);
688 return ret;
689}
690
691static void ocfs2_signal_wipe_completion(struct ocfs2_super *osb,
692 int slot)
693{
694 spin_lock(&osb->osb_lock);
695 osb->osb_orphan_wipes[slot]--;
696 spin_unlock(&osb->osb_lock);
697
698 wake_up(&osb->osb_wipe_event);
699}
700
Mark Fashehccd979b2005-12-15 14:31:24 -0800701static int ocfs2_wipe_inode(struct inode *inode,
702 struct buffer_head *di_bh)
703{
704 int status, orphaned_slot;
705 struct inode *orphan_dir_inode = NULL;
706 struct buffer_head *orphan_dir_bh = NULL;
707 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
Tiger Yang50008632007-03-20 16:01:38 -0700708 struct ocfs2_dinode *di;
Mark Fashehccd979b2005-12-15 14:31:24 -0800709
Tiger Yang50008632007-03-20 16:01:38 -0700710 di = (struct ocfs2_dinode *) di_bh->b_data;
711 orphaned_slot = le16_to_cpu(di->i_orphaned_slot);
Mark Fashehb4df6ed2006-02-22 17:35:08 -0800712
713 status = ocfs2_check_orphan_recovery_state(osb, orphaned_slot);
714 if (status)
715 return status;
716
Mark Fashehccd979b2005-12-15 14:31:24 -0800717 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
718 ORPHAN_DIR_SYSTEM_INODE,
719 orphaned_slot);
720 if (!orphan_dir_inode) {
721 status = -EEXIST;
722 mlog_errno(status);
723 goto bail;
724 }
725
726 /* Lock the orphan dir. The lock will be held for the entire
727 * delete_inode operation. We do this now to avoid races with
728 * recovery completion on other nodes. */
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800729 mutex_lock(&orphan_dir_inode->i_mutex);
Mark Fashehe63aecb62007-10-18 15:30:42 -0700730 status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -0800731 if (status < 0) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800732 mutex_unlock(&orphan_dir_inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -0800733
734 mlog_errno(status);
735 goto bail;
736 }
737
738 /* we do this while holding the orphan dir lock because we
Mark Fasheh34d024f2007-09-24 15:56:19 -0700739 * don't want recovery being run from another node to try an
740 * inode delete underneath us -- this will result in two nodes
Mark Fashehccd979b2005-12-15 14:31:24 -0800741 * truncating the same file! */
742 status = ocfs2_truncate_for_delete(osb, inode, di_bh);
743 if (status < 0) {
744 mlog_errno(status);
745 goto bail_unlock_dir;
746 }
747
Tiger Yangcf1d6c72008-08-18 17:11:00 +0800748 /*Free extended attribute resources associated with this inode.*/
749 status = ocfs2_xattr_remove(inode, di_bh);
750 if (status < 0) {
751 mlog_errno(status);
752 goto bail_unlock_dir;
753 }
754
Mark Fashehccd979b2005-12-15 14:31:24 -0800755 status = ocfs2_remove_inode(inode, di_bh, orphan_dir_inode,
756 orphan_dir_bh);
757 if (status < 0)
758 mlog_errno(status);
759
760bail_unlock_dir:
Mark Fashehe63aecb62007-10-18 15:30:42 -0700761 ocfs2_inode_unlock(orphan_dir_inode, 1);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800762 mutex_unlock(&orphan_dir_inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -0800763 brelse(orphan_dir_bh);
764bail:
765 iput(orphan_dir_inode);
Mark Fashehb4df6ed2006-02-22 17:35:08 -0800766 ocfs2_signal_wipe_completion(osb, orphaned_slot);
Mark Fashehccd979b2005-12-15 14:31:24 -0800767
768 return status;
769}
770
771/* There is a series of simple checks that should be done before a
Mark Fasheh34d024f2007-09-24 15:56:19 -0700772 * trylock is even considered. Encapsulate those in this function. */
Mark Fashehccd979b2005-12-15 14:31:24 -0800773static int ocfs2_inode_is_valid_to_delete(struct inode *inode)
774{
775 int ret = 0;
776 struct ocfs2_inode_info *oi = OCFS2_I(inode);
777 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
778
779 /* We shouldn't be getting here for the root directory
780 * inode.. */
781 if (inode == osb->root_inode) {
782 mlog(ML_ERROR, "Skipping delete of root inode.\n");
783 goto bail;
784 }
785
Mark Fasheh34d024f2007-09-24 15:56:19 -0700786 /* If we're coming from downconvert_thread we can't go into our own
Mark Fashehccd979b2005-12-15 14:31:24 -0800787 * voting [hello, deadlock city!], so unforuntately we just
788 * have to skip deleting this guy. That's OK though because
789 * the node who's doing the actual deleting should handle it
790 * anyway. */
Mark Fasheh34d024f2007-09-24 15:56:19 -0700791 if (current == osb->dc_task) {
Mark Fashehccd979b2005-12-15 14:31:24 -0800792 mlog(0, "Skipping delete of %lu because we're currently "
Mark Fasheh34d024f2007-09-24 15:56:19 -0700793 "in downconvert\n", inode->i_ino);
Mark Fashehccd979b2005-12-15 14:31:24 -0800794 goto bail;
795 }
796
797 spin_lock(&oi->ip_lock);
798 /* OCFS2 *never* deletes system files. This should technically
799 * never get here as system file inodes should always have a
800 * positive link count. */
801 if (oi->ip_flags & OCFS2_INODE_SYSTEM_FILE) {
Mark Fashehb06970532006-03-03 10:24:33 -0800802 mlog(ML_ERROR, "Skipping delete of system file %llu\n",
803 (unsigned long long)oi->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -0800804 goto bail_unlock;
805 }
806
Mark Fasheh34d024f2007-09-24 15:56:19 -0700807 /* If we have allowd wipe of this inode for another node, it
808 * will be marked here so we can safely skip it. Recovery will
809 * cleanup any inodes we might inadvertantly skip here. */
Mark Fashehccd979b2005-12-15 14:31:24 -0800810 if (oi->ip_flags & OCFS2_INODE_SKIP_DELETE) {
811 mlog(0, "Skipping delete of %lu because another node "
812 "has done this for us.\n", inode->i_ino);
813 goto bail_unlock;
814 }
815
816 ret = 1;
817bail_unlock:
818 spin_unlock(&oi->ip_lock);
819bail:
820 return ret;
821}
822
823/* Query the cluster to determine whether we should wipe an inode from
824 * disk or not.
825 *
826 * Requires the inode to have the cluster lock. */
827static int ocfs2_query_inode_wipe(struct inode *inode,
828 struct buffer_head *di_bh,
829 int *wipe)
830{
831 int status = 0;
832 struct ocfs2_inode_info *oi = OCFS2_I(inode);
833 struct ocfs2_dinode *di;
834
835 *wipe = 0;
836
837 /* While we were waiting for the cluster lock in
838 * ocfs2_delete_inode, another node might have asked to delete
839 * the inode. Recheck our flags to catch this. */
840 if (!ocfs2_inode_is_valid_to_delete(inode)) {
Mark Fashehb06970532006-03-03 10:24:33 -0800841 mlog(0, "Skipping delete of %llu because flags changed\n",
842 (unsigned long long)oi->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -0800843 goto bail;
844 }
845
846 /* Now that we have an up to date inode, we can double check
847 * the link count. */
848 if (inode->i_nlink) {
Mark Fashehb06970532006-03-03 10:24:33 -0800849 mlog(0, "Skipping delete of %llu because nlink = %u\n",
850 (unsigned long long)oi->ip_blkno, inode->i_nlink);
Mark Fashehccd979b2005-12-15 14:31:24 -0800851 goto bail;
852 }
853
854 /* Do some basic inode verification... */
855 di = (struct ocfs2_dinode *) di_bh->b_data;
856 if (!(di->i_flags & cpu_to_le32(OCFS2_ORPHANED_FL))) {
857 /* for lack of a better error? */
858 status = -EEXIST;
859 mlog(ML_ERROR,
Mark Fashehb06970532006-03-03 10:24:33 -0800860 "Inode %llu (on-disk %llu) not orphaned! "
Mark Fashehccd979b2005-12-15 14:31:24 -0800861 "Disk flags 0x%x, inode flags 0x%x\n",
Mark Fashehb06970532006-03-03 10:24:33 -0800862 (unsigned long long)oi->ip_blkno,
Mark Fasheh1ca1a112007-04-27 16:01:25 -0700863 (unsigned long long)le64_to_cpu(di->i_blkno),
864 le32_to_cpu(di->i_flags), oi->ip_flags);
Mark Fashehccd979b2005-12-15 14:31:24 -0800865 goto bail;
866 }
867
868 /* has someone already deleted us?! baaad... */
869 if (di->i_dtime) {
870 status = -EEXIST;
871 mlog_errno(status);
872 goto bail;
873 }
874
Mark Fasheh6f16bf62007-03-20 17:17:54 -0700875 /*
876 * This is how ocfs2 determines whether an inode is still live
877 * within the cluster. Every node takes a shared read lock on
878 * the inode open lock in ocfs2_read_locked_inode(). When we
879 * get to ->delete_inode(), each node tries to convert it's
880 * lock to an exclusive. Trylocks are serialized by the inode
881 * meta data lock. If the upconvert suceeds, we know the inode
882 * is no longer live and can be deleted.
883 *
884 * Though we call this with the meta data lock held, the
885 * trylock keeps us from ABBA deadlock.
886 */
887 status = ocfs2_try_open_lock(inode, 1);
Tiger Yang50008632007-03-20 16:01:38 -0700888 if (status == -EAGAIN) {
Mark Fashehccd979b2005-12-15 14:31:24 -0800889 status = 0;
Joe Perches27592362007-11-19 17:53:34 -0800890 mlog(0, "Skipping delete of %llu because it is in use on "
Mark Fashehb06970532006-03-03 10:24:33 -0800891 "other nodes\n", (unsigned long long)oi->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -0800892 goto bail;
893 }
894 if (status < 0) {
895 mlog_errno(status);
896 goto bail;
897 }
898
Tiger Yang50008632007-03-20 16:01:38 -0700899 *wipe = 1;
900 mlog(0, "Inode %llu is ok to wipe from orphan dir %u\n",
901 (unsigned long long)oi->ip_blkno,
902 le16_to_cpu(di->i_orphaned_slot));
Mark Fashehccd979b2005-12-15 14:31:24 -0800903
904bail:
905 return status;
906}
907
908/* Support function for ocfs2_delete_inode. Will help us keep the
909 * inode data in a consistent state for clear_inode. Always truncates
910 * pages, optionally sync's them first. */
911static void ocfs2_cleanup_delete_inode(struct inode *inode,
912 int sync_data)
913{
Mark Fashehb06970532006-03-03 10:24:33 -0800914 mlog(0, "Cleanup inode %llu, sync = %d\n",
915 (unsigned long long)OCFS2_I(inode)->ip_blkno, sync_data);
Mark Fashehccd979b2005-12-15 14:31:24 -0800916 if (sync_data)
917 write_inode_now(inode, 1);
918 truncate_inode_pages(&inode->i_data, 0);
919}
920
921void ocfs2_delete_inode(struct inode *inode)
922{
923 int wipe, status;
924 sigset_t blocked, oldset;
925 struct buffer_head *di_bh = NULL;
926
927 mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
928
929 if (is_bad_inode(inode)) {
930 mlog(0, "Skipping delete of bad inode\n");
931 goto bail;
932 }
933
934 if (!ocfs2_inode_is_valid_to_delete(inode)) {
935 /* It's probably not necessary to truncate_inode_pages
936 * here but we do it for safety anyway (it will most
937 * likely be a no-op anyway) */
938 ocfs2_cleanup_delete_inode(inode, 0);
939 goto bail;
940 }
941
942 /* We want to block signals in delete_inode as the lock and
943 * messaging paths may return us -ERESTARTSYS. Which would
944 * cause us to exit early, resulting in inodes being orphaned
945 * forever. */
946 sigfillset(&blocked);
947 status = sigprocmask(SIG_BLOCK, &blocked, &oldset);
948 if (status < 0) {
949 mlog_errno(status);
950 ocfs2_cleanup_delete_inode(inode, 1);
951 goto bail;
952 }
953
954 /* Lock down the inode. This gives us an up to date view of
955 * it's metadata (for verification), and allows us to
Mark Fasheh34d024f2007-09-24 15:56:19 -0700956 * serialize delete_inode on multiple nodes.
Mark Fashehccd979b2005-12-15 14:31:24 -0800957 *
958 * Even though we might be doing a truncate, we don't take the
959 * allocation lock here as it won't be needed - nobody will
960 * have the file open.
961 */
Mark Fashehe63aecb62007-10-18 15:30:42 -0700962 status = ocfs2_inode_lock(inode, &di_bh, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -0800963 if (status < 0) {
964 if (status != -ENOENT)
965 mlog_errno(status);
966 ocfs2_cleanup_delete_inode(inode, 0);
967 goto bail_unblock;
968 }
969
970 /* Query the cluster. This will be the final decision made
971 * before we go ahead and wipe the inode. */
972 status = ocfs2_query_inode_wipe(inode, di_bh, &wipe);
973 if (!wipe || status < 0) {
Mark Fasheh34d024f2007-09-24 15:56:19 -0700974 /* Error and remote inode busy both mean we won't be
Mark Fashehccd979b2005-12-15 14:31:24 -0800975 * removing the inode, so they take almost the same
976 * path. */
977 if (status < 0)
978 mlog_errno(status);
979
Mark Fasheh34d024f2007-09-24 15:56:19 -0700980 /* Someone in the cluster has disallowed a wipe of
981 * this inode, or it was never completely
982 * orphaned. Write out the pages and exit now. */
Mark Fashehccd979b2005-12-15 14:31:24 -0800983 ocfs2_cleanup_delete_inode(inode, 1);
984 goto bail_unlock_inode;
985 }
986
987 ocfs2_cleanup_delete_inode(inode, 0);
988
989 status = ocfs2_wipe_inode(inode, di_bh);
990 if (status < 0) {
Mark Fashehb4df6ed2006-02-22 17:35:08 -0800991 if (status != -EDEADLK)
992 mlog_errno(status);
Mark Fashehccd979b2005-12-15 14:31:24 -0800993 goto bail_unlock_inode;
994 }
995
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700996 /*
997 * Mark the inode as successfully deleted.
998 *
999 * This is important for ocfs2_clear_inode() as it will check
1000 * this flag and skip any checkpointing work
1001 *
1002 * ocfs2_stuff_meta_lvb() also uses this flag to invalidate
1003 * the LVB for other nodes.
1004 */
Mark Fashehccd979b2005-12-15 14:31:24 -08001005 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_DELETED;
1006
1007bail_unlock_inode:
Mark Fashehe63aecb62007-10-18 15:30:42 -07001008 ocfs2_inode_unlock(inode, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -08001009 brelse(di_bh);
1010bail_unblock:
1011 status = sigprocmask(SIG_SETMASK, &oldset, NULL);
1012 if (status < 0)
1013 mlog_errno(status);
1014bail:
1015 clear_inode(inode);
1016 mlog_exit_void();
1017}
1018
1019void ocfs2_clear_inode(struct inode *inode)
1020{
1021 int status;
1022 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1023
1024 mlog_entry_void();
1025
1026 if (!inode)
1027 goto bail;
1028
Mark Fashehb06970532006-03-03 10:24:33 -08001029 mlog(0, "Clearing inode: %llu, nlink = %u\n",
1030 (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_nlink);
Mark Fashehccd979b2005-12-15 14:31:24 -08001031
1032 mlog_bug_on_msg(OCFS2_SB(inode->i_sb) == NULL,
1033 "Inode=%lu\n", inode->i_ino);
1034
Mark Fasheh34d024f2007-09-24 15:56:19 -07001035 /* To preven remote deletes we hold open lock before, now it
1036 * is time to unlock PR and EX open locks. */
Tiger Yang50008632007-03-20 16:01:38 -07001037 ocfs2_open_unlock(inode);
1038
Mark Fashehccd979b2005-12-15 14:31:24 -08001039 /* Do these before all the other work so that we don't bounce
Mark Fasheh34d024f2007-09-24 15:56:19 -07001040 * the downconvert thread while waiting to destroy the locks. */
Mark Fashehccd979b2005-12-15 14:31:24 -08001041 ocfs2_mark_lockres_freeing(&oi->ip_rw_lockres);
Mark Fashehe63aecb62007-10-18 15:30:42 -07001042 ocfs2_mark_lockres_freeing(&oi->ip_inode_lockres);
Tiger Yang50008632007-03-20 16:01:38 -07001043 ocfs2_mark_lockres_freeing(&oi->ip_open_lockres);
Mark Fashehccd979b2005-12-15 14:31:24 -08001044
1045 /* We very well may get a clear_inode before all an inodes
1046 * metadata has hit disk. Of course, we can't drop any cluster
1047 * locks until the journal has finished with it. The only
1048 * exception here are successfully wiped inodes - their
1049 * metadata can now be considered to be part of the system
1050 * inodes from which it came. */
1051 if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED))
1052 ocfs2_checkpoint_inode(inode);
1053
1054 mlog_bug_on_msg(!list_empty(&oi->ip_io_markers),
Mark Fashehb06970532006-03-03 10:24:33 -08001055 "Clear inode of %llu, inode has io markers\n",
1056 (unsigned long long)oi->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08001057
Mark Fasheh83418972007-04-23 18:53:12 -07001058 ocfs2_extent_map_trunc(inode, 0);
1059
Mark Fashehccd979b2005-12-15 14:31:24 -08001060 status = ocfs2_drop_inode_locks(inode);
1061 if (status < 0)
1062 mlog_errno(status);
1063
1064 ocfs2_lock_res_free(&oi->ip_rw_lockres);
Mark Fashehe63aecb62007-10-18 15:30:42 -07001065 ocfs2_lock_res_free(&oi->ip_inode_lockres);
Tiger Yang50008632007-03-20 16:01:38 -07001066 ocfs2_lock_res_free(&oi->ip_open_lockres);
Mark Fashehccd979b2005-12-15 14:31:24 -08001067
1068 ocfs2_metadata_cache_purge(inode);
1069
1070 mlog_bug_on_msg(oi->ip_metadata_cache.ci_num_cached,
Mark Fashehb06970532006-03-03 10:24:33 -08001071 "Clear inode of %llu, inode has %u cache items\n",
1072 (unsigned long long)oi->ip_blkno, oi->ip_metadata_cache.ci_num_cached);
Mark Fashehccd979b2005-12-15 14:31:24 -08001073
1074 mlog_bug_on_msg(!(oi->ip_flags & OCFS2_INODE_CACHE_INLINE),
Mark Fashehb06970532006-03-03 10:24:33 -08001075 "Clear inode of %llu, inode has a bad flag\n",
1076 (unsigned long long)oi->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08001077
1078 mlog_bug_on_msg(spin_is_locked(&oi->ip_lock),
Mark Fashehb06970532006-03-03 10:24:33 -08001079 "Clear inode of %llu, inode is locked\n",
1080 (unsigned long long)oi->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08001081
Mark Fasheh251b6ec2006-01-10 15:41:43 -08001082 mlog_bug_on_msg(!mutex_trylock(&oi->ip_io_mutex),
Mark Fashehb06970532006-03-03 10:24:33 -08001083 "Clear inode of %llu, io_mutex is locked\n",
1084 (unsigned long long)oi->ip_blkno);
Mark Fasheh251b6ec2006-01-10 15:41:43 -08001085 mutex_unlock(&oi->ip_io_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -08001086
1087 /*
1088 * down_trylock() returns 0, down_write_trylock() returns 1
1089 * kernel 1, world 0
1090 */
1091 mlog_bug_on_msg(!down_write_trylock(&oi->ip_alloc_sem),
Mark Fashehb06970532006-03-03 10:24:33 -08001092 "Clear inode of %llu, alloc_sem is locked\n",
1093 (unsigned long long)oi->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08001094 up_write(&oi->ip_alloc_sem);
1095
1096 mlog_bug_on_msg(oi->ip_open_count,
Mark Fashehb06970532006-03-03 10:24:33 -08001097 "Clear inode of %llu has open count %d\n",
1098 (unsigned long long)oi->ip_blkno, oi->ip_open_count);
Mark Fashehccd979b2005-12-15 14:31:24 -08001099
1100 /* Clear all other flags. */
1101 oi->ip_flags = OCFS2_INODE_CACHE_INLINE;
1102 oi->ip_created_trans = 0;
1103 oi->ip_last_trans = 0;
1104 oi->ip_dir_start_lookup = 0;
1105 oi->ip_blkno = 0ULL;
Joel Becker2b4e30f2008-09-03 20:03:41 -07001106 jbd2_journal_release_jbd_inode(OCFS2_SB(inode->i_sb)->journal->j_journal,
1107 &oi->ip_jinode);
Mark Fashehccd979b2005-12-15 14:31:24 -08001108
1109bail:
1110 mlog_exit_void();
1111}
1112
1113/* Called under inode_lock, with no more references on the
1114 * struct inode, so it's safe here to check the flags field
1115 * and to manipulate i_nlink without any other locks. */
1116void ocfs2_drop_inode(struct inode *inode)
1117{
1118 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1119
1120 mlog_entry_void();
1121
Mark Fashehb06970532006-03-03 10:24:33 -08001122 mlog(0, "Drop inode %llu, nlink = %u, ip_flags = 0x%x\n",
1123 (unsigned long long)oi->ip_blkno, inode->i_nlink, oi->ip_flags);
Mark Fashehccd979b2005-12-15 14:31:24 -08001124
Mark Fasheh379dfe92006-09-08 14:21:03 -07001125 if (oi->ip_flags & OCFS2_INODE_MAYBE_ORPHANED)
1126 generic_delete_inode(inode);
1127 else
1128 generic_drop_inode(inode);
Mark Fashehccd979b2005-12-15 14:31:24 -08001129
1130 mlog_exit_void();
1131}
1132
1133/*
1134 * TODO: this should probably be merged into ocfs2_get_block
1135 *
1136 * However, you now need to pay attention to the cont_prepare_write()
1137 * stuff in ocfs2_get_block (that is, ocfs2_get_block pretty much
1138 * expects never to extend).
1139 */
1140struct buffer_head *ocfs2_bread(struct inode *inode,
1141 int block, int *err, int reada)
1142{
1143 struct buffer_head *bh = NULL;
1144 int tmperr;
1145 u64 p_blkno;
1146 int readflags = OCFS2_BH_CACHED;
1147
Mark Fashehccd979b2005-12-15 14:31:24 -08001148 if (reada)
1149 readflags |= OCFS2_BH_READAHEAD;
Mark Fashehccd979b2005-12-15 14:31:24 -08001150
1151 if (((u64)block << inode->i_sb->s_blocksize_bits) >=
1152 i_size_read(inode)) {
1153 BUG_ON(!reada);
1154 return NULL;
1155 }
1156
Joel Beckeree19a772007-03-28 18:27:07 -07001157 down_read(&OCFS2_I(inode)->ip_alloc_sem);
Mark Fasheh49cb8d22007-03-09 16:21:46 -08001158 tmperr = ocfs2_extent_map_get_blocks(inode, block, &p_blkno, NULL,
1159 NULL);
Joel Beckeree19a772007-03-28 18:27:07 -07001160 up_read(&OCFS2_I(inode)->ip_alloc_sem);
Mark Fashehccd979b2005-12-15 14:31:24 -08001161 if (tmperr < 0) {
1162 mlog_errno(tmperr);
1163 goto fail;
1164 }
1165
1166 tmperr = ocfs2_read_block(OCFS2_SB(inode->i_sb), p_blkno, &bh,
1167 readflags, inode);
1168 if (tmperr < 0)
1169 goto fail;
1170
1171 tmperr = 0;
1172
1173 *err = 0;
1174 return bh;
1175
1176fail:
1177 if (bh) {
1178 brelse(bh);
1179 bh = NULL;
1180 }
1181 *err = -EIO;
1182 return NULL;
1183}
1184
1185/*
1186 * This is called from our getattr.
1187 */
1188int ocfs2_inode_revalidate(struct dentry *dentry)
1189{
1190 struct inode *inode = dentry->d_inode;
1191 int status = 0;
1192
Mark Fashehb06970532006-03-03 10:24:33 -08001193 mlog_entry("(inode = 0x%p, ino = %llu)\n", inode,
1194 inode ? (unsigned long long)OCFS2_I(inode)->ip_blkno : 0ULL);
Mark Fashehccd979b2005-12-15 14:31:24 -08001195
1196 if (!inode) {
1197 mlog(0, "eep, no inode!\n");
1198 status = -ENOENT;
1199 goto bail;
1200 }
1201
1202 spin_lock(&OCFS2_I(inode)->ip_lock);
1203 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
1204 spin_unlock(&OCFS2_I(inode)->ip_lock);
1205 mlog(0, "inode deleted!\n");
1206 status = -ENOENT;
1207 goto bail;
1208 }
1209 spin_unlock(&OCFS2_I(inode)->ip_lock);
1210
Mark Fashehe63aecb62007-10-18 15:30:42 -07001211 /* Let ocfs2_inode_lock do the work of updating our struct
Mark Fashehccd979b2005-12-15 14:31:24 -08001212 * inode for us. */
Mark Fashehe63aecb62007-10-18 15:30:42 -07001213 status = ocfs2_inode_lock(inode, NULL, 0);
Mark Fashehccd979b2005-12-15 14:31:24 -08001214 if (status < 0) {
1215 if (status != -ENOENT)
1216 mlog_errno(status);
1217 goto bail;
1218 }
Mark Fashehe63aecb62007-10-18 15:30:42 -07001219 ocfs2_inode_unlock(inode, 0);
Mark Fashehccd979b2005-12-15 14:31:24 -08001220bail:
1221 mlog_exit(status);
1222
1223 return status;
1224}
1225
1226/*
1227 * Updates a disk inode from a
1228 * struct inode.
1229 * Only takes ip_lock.
1230 */
Mark Fasheh1fabe142006-10-09 18:11:45 -07001231int ocfs2_mark_inode_dirty(handle_t *handle,
Mark Fashehccd979b2005-12-15 14:31:24 -08001232 struct inode *inode,
1233 struct buffer_head *bh)
1234{
1235 int status;
1236 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
1237
Mark Fashehb06970532006-03-03 10:24:33 -08001238 mlog_entry("(inode %llu)\n",
1239 (unsigned long long)OCFS2_I(inode)->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08001240
1241 status = ocfs2_journal_access(handle, inode, bh,
1242 OCFS2_JOURNAL_ACCESS_WRITE);
1243 if (status < 0) {
1244 mlog_errno(status);
1245 goto leave;
1246 }
1247
1248 spin_lock(&OCFS2_I(inode)->ip_lock);
1249 fe->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters);
Jan Kara6e4b0d52007-04-27 11:08:01 -07001250 ocfs2_get_inode_flags(OCFS2_I(inode));
Herbert Poetzlca4d1472006-07-03 17:27:12 -07001251 fe->i_attr = cpu_to_le32(OCFS2_I(inode)->ip_attr);
Mark Fasheh15b1e362007-09-07 13:58:15 -07001252 fe->i_dyn_features = cpu_to_le16(OCFS2_I(inode)->ip_dyn_features);
Mark Fashehccd979b2005-12-15 14:31:24 -08001253 spin_unlock(&OCFS2_I(inode)->ip_lock);
1254
1255 fe->i_size = cpu_to_le64(i_size_read(inode));
1256 fe->i_links_count = cpu_to_le16(inode->i_nlink);
1257 fe->i_uid = cpu_to_le32(inode->i_uid);
1258 fe->i_gid = cpu_to_le32(inode->i_gid);
1259 fe->i_mode = cpu_to_le16(inode->i_mode);
1260 fe->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
1261 fe->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
1262 fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
1263 fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
1264 fe->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
1265 fe->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
1266
1267 status = ocfs2_journal_dirty(handle, bh);
1268 if (status < 0)
1269 mlog_errno(status);
1270
1271 status = 0;
1272leave:
1273
1274 mlog_exit(status);
1275 return status;
1276}
1277
1278/*
1279 *
1280 * Updates a struct inode from a disk inode.
1281 * does no i/o, only takes ip_lock.
1282 */
1283void ocfs2_refresh_inode(struct inode *inode,
1284 struct ocfs2_dinode *fe)
1285{
Mark Fashehccd979b2005-12-15 14:31:24 -08001286 spin_lock(&OCFS2_I(inode)->ip_lock);
1287
1288 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
Herbert Poetzlca4d1472006-07-03 17:27:12 -07001289 OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
Mark Fasheh15b1e362007-09-07 13:58:15 -07001290 OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features);
Herbert Poetzlca4d1472006-07-03 17:27:12 -07001291 ocfs2_set_inode_flags(inode);
Mark Fashehccd979b2005-12-15 14:31:24 -08001292 i_size_write(inode, le64_to_cpu(fe->i_size));
1293 inode->i_nlink = le16_to_cpu(fe->i_links_count);
1294 inode->i_uid = le32_to_cpu(fe->i_uid);
1295 inode->i_gid = le32_to_cpu(fe->i_gid);
1296 inode->i_mode = le16_to_cpu(fe->i_mode);
Mark Fashehccd979b2005-12-15 14:31:24 -08001297 if (S_ISLNK(inode->i_mode) && le32_to_cpu(fe->i_clusters) == 0)
1298 inode->i_blocks = 0;
1299 else
Mark Fasheh8110b072007-03-22 16:53:23 -07001300 inode->i_blocks = ocfs2_inode_sector_count(inode);
Mark Fashehccd979b2005-12-15 14:31:24 -08001301 inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
1302 inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
1303 inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
1304 inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
1305 inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
1306 inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
1307
1308 spin_unlock(&OCFS2_I(inode)->ip_lock);
1309}