blob: b4c8bb6b8d281f74feb175580696eb1169819677 [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>
Mark Fashehccd979b2005-12-15 14:31:24 -080028#include <linux/highmem.h>
29#include <linux/pagemap.h>
Jan Karaa90714c2008-10-09 19:38:40 +020030#include <linux/quotaops.h>
Mark Fashehccd979b2005-12-15 14:31:24 -080031
32#include <asm/byteorder.h>
33
Mark Fashehccd979b2005-12-15 14:31:24 -080034#include <cluster/masklog.h>
35
36#include "ocfs2.h"
37
38#include "alloc.h"
Mark Fasheh9b7895e2008-11-12 16:27:44 -080039#include "dir.h"
Joel Beckerd6b32bb2008-10-17 14:55:01 -070040#include "blockcheck.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080041#include "dlmglue.h"
42#include "extent_map.h"
43#include "file.h"
Mark Fashehb4df6ed2006-02-22 17:35:08 -080044#include "heartbeat.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080045#include "inode.h"
46#include "journal.h"
47#include "namei.h"
48#include "suballoc.h"
49#include "super.h"
50#include "symlink.h"
51#include "sysfile.h"
52#include "uptodate.h"
Tiger Yangcf1d6c72008-08-18 17:11:00 +080053#include "xattr.h"
Tao Ma8b2c0db2009-08-18 11:43:49 +080054#include "refcounttree.h"
Tao Ma64f3b262011-02-22 22:24:57 +080055#include "ocfs2_trace.h"
Mark Fashehccd979b2005-12-15 14:31:24 -080056
57#include "buffer_head_io.h"
58
Mark Fashehccd979b2005-12-15 14:31:24 -080059struct ocfs2_find_inode_args
60{
61 u64 fi_blkno;
62 unsigned long fi_ino;
63 unsigned int fi_flags;
Jan Kara5fa06132008-01-11 00:11:45 +010064 unsigned int fi_sysfile_type;
Mark Fashehccd979b2005-12-15 14:31:24 -080065};
66
Jan Kara5fa06132008-01-11 00:11:45 +010067static struct lock_class_key ocfs2_sysfile_lock_key[NUM_SYSTEM_INODES];
68
Mark Fashehccd979b2005-12-15 14:31:24 -080069static int ocfs2_read_locked_inode(struct inode *inode,
70 struct ocfs2_find_inode_args *args);
71static int ocfs2_init_locked_inode(struct inode *inode, void *opaque);
72static int ocfs2_find_actor(struct inode *inode, void *opaque);
73static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
74 struct inode *inode,
75 struct buffer_head *fe_bh);
76
Herbert Poetzlca4d1472006-07-03 17:27:12 -070077void ocfs2_set_inode_flags(struct inode *inode)
78{
79 unsigned int flags = OCFS2_I(inode)->ip_attr;
80
81 inode->i_flags &= ~(S_IMMUTABLE |
82 S_SYNC | S_APPEND | S_NOATIME | S_DIRSYNC);
83
84 if (flags & OCFS2_IMMUTABLE_FL)
85 inode->i_flags |= S_IMMUTABLE;
86
87 if (flags & OCFS2_SYNC_FL)
88 inode->i_flags |= S_SYNC;
89 if (flags & OCFS2_APPEND_FL)
90 inode->i_flags |= S_APPEND;
91 if (flags & OCFS2_NOATIME_FL)
92 inode->i_flags |= S_NOATIME;
93 if (flags & OCFS2_DIRSYNC_FL)
94 inode->i_flags |= S_DIRSYNC;
95}
96
Jan Kara6e4b0d52007-04-27 11:08:01 -070097/* Propagate flags from i_flags to OCFS2_I(inode)->ip_attr */
98void ocfs2_get_inode_flags(struct ocfs2_inode_info *oi)
99{
100 unsigned int flags = oi->vfs_inode.i_flags;
101
102 oi->ip_attr &= ~(OCFS2_SYNC_FL|OCFS2_APPEND_FL|
103 OCFS2_IMMUTABLE_FL|OCFS2_NOATIME_FL|OCFS2_DIRSYNC_FL);
104 if (flags & S_SYNC)
105 oi->ip_attr |= OCFS2_SYNC_FL;
106 if (flags & S_APPEND)
107 oi->ip_attr |= OCFS2_APPEND_FL;
108 if (flags & S_IMMUTABLE)
109 oi->ip_attr |= OCFS2_IMMUTABLE_FL;
110 if (flags & S_NOATIME)
111 oi->ip_attr |= OCFS2_NOATIME_FL;
112 if (flags & S_DIRSYNC)
113 oi->ip_attr |= OCFS2_DIRSYNC_FL;
114}
115
wengang wang6ca497a2009-03-06 21:29:10 +0800116struct inode *ocfs2_ilookup(struct super_block *sb, u64 blkno)
117{
118 struct ocfs2_find_inode_args args;
119
120 args.fi_blkno = blkno;
121 args.fi_flags = 0;
122 args.fi_ino = ino_from_blkno(sb, blkno);
123 args.fi_sysfile_type = 0;
124
125 return ilookup5(sb, blkno, ocfs2_find_actor, &args);
126}
Jan Kara5fa06132008-01-11 00:11:45 +0100127struct inode *ocfs2_iget(struct ocfs2_super *osb, u64 blkno, unsigned flags,
128 int sysfile_type)
Mark Fashehccd979b2005-12-15 14:31:24 -0800129{
130 struct inode *inode = NULL;
131 struct super_block *sb = osb->sb;
132 struct ocfs2_find_inode_args args;
133
Tao Ma64f3b262011-02-22 22:24:57 +0800134 trace_ocfs2_iget_begin((unsigned long long)blkno, flags,
135 sysfile_type);
Mark Fashehccd979b2005-12-15 14:31:24 -0800136
137 /* Ok. By now we've either got the offsets passed to us by the
138 * caller, or we just pulled them off the bh. Lets do some
139 * sanity checks to make sure they're OK. */
140 if (blkno == 0) {
141 inode = ERR_PTR(-EINVAL);
142 mlog_errno(PTR_ERR(inode));
143 goto bail;
144 }
145
146 args.fi_blkno = blkno;
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700147 args.fi_flags = flags;
Mark Fashehccd979b2005-12-15 14:31:24 -0800148 args.fi_ino = ino_from_blkno(sb, blkno);
Jan Kara5fa06132008-01-11 00:11:45 +0100149 args.fi_sysfile_type = sysfile_type;
Mark Fashehccd979b2005-12-15 14:31:24 -0800150
151 inode = iget5_locked(sb, args.fi_ino, ocfs2_find_actor,
152 ocfs2_init_locked_inode, &args);
153 /* inode was *not* in the inode cache. 2.6.x requires
154 * us to do our own read_inode call and unlock it
155 * afterwards. */
Mark Fashehccd979b2005-12-15 14:31:24 -0800156 if (inode == NULL) {
157 inode = ERR_PTR(-ENOMEM);
158 mlog_errno(PTR_ERR(inode));
159 goto bail;
160 }
Tao Ma64f3b262011-02-22 22:24:57 +0800161 trace_ocfs2_iget5_locked(inode->i_state);
Tao Ma6218b902011-02-21 11:18:30 +0800162 if (inode->i_state & I_NEW) {
Tao Ma6218b902011-02-21 11:18:30 +0800163 ocfs2_read_locked_inode(inode, &args);
164 unlock_new_inode(inode);
165 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800166 if (is_bad_inode(inode)) {
167 iput(inode);
168 inode = ERR_PTR(-ESTALE);
Mark Fashehccd979b2005-12-15 14:31:24 -0800169 goto bail;
170 }
171
172bail:
173 if (!IS_ERR(inode)) {
Tao Ma64f3b262011-02-22 22:24:57 +0800174 trace_ocfs2_iget_end(inode,
175 (unsigned long long)OCFS2_I(inode)->ip_blkno);
Mark Fasheh6a1bd4a2007-01-03 17:06:59 -0800176 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800177
178 return inode;
179}
180
181
182/*
183 * here's how inodes get read from disk:
184 * iget5_locked -> find_actor -> OCFS2_FIND_ACTOR
185 * found? : return the in-memory inode
186 * not found? : get_new_inode -> OCFS2_INIT_LOCKED_INODE
187 */
188
189static int ocfs2_find_actor(struct inode *inode, void *opaque)
190{
191 struct ocfs2_find_inode_args *args = NULL;
192 struct ocfs2_inode_info *oi = OCFS2_I(inode);
193 int ret = 0;
194
Mark Fashehccd979b2005-12-15 14:31:24 -0800195 args = opaque;
196
197 mlog_bug_on_msg(!inode, "No inode in find actor!\n");
198
Tao Ma64f3b262011-02-22 22:24:57 +0800199 trace_ocfs2_find_actor(inode, inode->i_ino, opaque, args->fi_blkno);
200
Mark Fashehccd979b2005-12-15 14:31:24 -0800201 if (oi->ip_blkno != args->fi_blkno)
202 goto bail;
203
Mark Fashehccd979b2005-12-15 14:31:24 -0800204 ret = 1;
205bail:
Mark Fashehccd979b2005-12-15 14:31:24 -0800206 return ret;
207}
208
209/*
210 * initialize the new inode, but don't do anything that would cause
211 * us to sleep.
212 * return 0 on success, 1 on failure
213 */
214static int ocfs2_init_locked_inode(struct inode *inode, void *opaque)
215{
216 struct ocfs2_find_inode_args *args = opaque;
Jan Karacb257972009-06-04 15:26:50 +0200217 static struct lock_class_key ocfs2_quota_ip_alloc_sem_key,
218 ocfs2_file_ip_alloc_sem_key;
Mark Fashehccd979b2005-12-15 14:31:24 -0800219
Mark Fashehccd979b2005-12-15 14:31:24 -0800220 inode->i_ino = args->fi_ino;
221 OCFS2_I(inode)->ip_blkno = args->fi_blkno;
Jan Kara5fa06132008-01-11 00:11:45 +0100222 if (args->fi_sysfile_type != 0)
223 lockdep_set_class(&inode->i_mutex,
224 &ocfs2_sysfile_lock_key[args->fi_sysfile_type]);
Jan Karacb257972009-06-04 15:26:50 +0200225 if (args->fi_sysfile_type == USER_QUOTA_SYSTEM_INODE ||
226 args->fi_sysfile_type == GROUP_QUOTA_SYSTEM_INODE ||
227 args->fi_sysfile_type == LOCAL_USER_QUOTA_SYSTEM_INODE ||
228 args->fi_sysfile_type == LOCAL_GROUP_QUOTA_SYSTEM_INODE)
229 lockdep_set_class(&OCFS2_I(inode)->ip_alloc_sem,
230 &ocfs2_quota_ip_alloc_sem_key);
231 else
232 lockdep_set_class(&OCFS2_I(inode)->ip_alloc_sem,
233 &ocfs2_file_ip_alloc_sem_key);
Mark Fashehccd979b2005-12-15 14:31:24 -0800234
Mark Fashehccd979b2005-12-15 14:31:24 -0800235 return 0;
236}
237
Joel Beckerb657c952008-11-13 14:49:11 -0800238void ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe,
239 int create_ino)
Mark Fashehccd979b2005-12-15 14:31:24 -0800240{
241 struct super_block *sb;
242 struct ocfs2_super *osb;
Mark Fasheh53da4932008-07-21 14:29:16 -0700243 int use_plocks = 1;
Mark Fashehccd979b2005-12-15 14:31:24 -0800244
Mark Fashehccd979b2005-12-15 14:31:24 -0800245 sb = inode->i_sb;
246 osb = OCFS2_SB(sb);
247
Mark Fasheh53da4932008-07-21 14:29:16 -0700248 if ((osb->s_mount_opt & OCFS2_MOUNT_LOCALFLOCKS) ||
249 ocfs2_mount_local(osb) || !ocfs2_stack_supports_plocks())
250 use_plocks = 0;
251
Joel Beckerb657c952008-11-13 14:49:11 -0800252 /*
253 * These have all been checked by ocfs2_read_inode_block() or set
254 * by ocfs2_mknod_locked(), so a failure is a code bug.
255 */
256 BUG_ON(!OCFS2_IS_VALID_DINODE(fe)); /* This means that read_inode
257 cannot create a superblock
258 inode today. change if
259 that is needed. */
260 BUG_ON(!(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL)));
261 BUG_ON(le32_to_cpu(fe->i_fs_generation) != osb->fs_generation);
Mark Fashehccd979b2005-12-15 14:31:24 -0800262
Mark Fashehccd979b2005-12-15 14:31:24 -0800263
Mark Fasheh8110b072007-03-22 16:53:23 -0700264 OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
265 OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
Mark Fasheh15b1e362007-09-07 13:58:15 -0700266 OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features);
Mark Fasheh8110b072007-03-22 16:53:23 -0700267
Mark Fashehccd979b2005-12-15 14:31:24 -0800268 inode->i_version = 1;
269 inode->i_generation = le32_to_cpu(fe->i_generation);
270 inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
271 inode->i_mode = le16_to_cpu(fe->i_mode);
272 inode->i_uid = le32_to_cpu(fe->i_uid);
273 inode->i_gid = le32_to_cpu(fe->i_gid);
Mark Fashehccd979b2005-12-15 14:31:24 -0800274
275 /* Fast symlinks will have i_size but no allocated clusters. */
276 if (S_ISLNK(inode->i_mode) && !fe->i_clusters)
277 inode->i_blocks = 0;
278 else
Mark Fasheh8110b072007-03-22 16:53:23 -0700279 inode->i_blocks = ocfs2_inode_sector_count(inode);
Mark Fashehccd979b2005-12-15 14:31:24 -0800280 inode->i_mapping->a_ops = &ocfs2_aops;
Mark Fashehccd979b2005-12-15 14:31:24 -0800281 inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
282 inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
283 inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
284 inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
285 inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
286 inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
287
288 if (OCFS2_I(inode)->ip_blkno != le64_to_cpu(fe->i_blkno))
289 mlog(ML_ERROR,
Mark Fashehb0697052006-03-03 10:24:33 -0800290 "ip_blkno %llu != i_blkno %llu!\n",
291 (unsigned long long)OCFS2_I(inode)->ip_blkno,
Mark Fasheh1ca1a112007-04-27 16:01:25 -0700292 (unsigned long long)le64_to_cpu(fe->i_blkno));
Mark Fashehccd979b2005-12-15 14:31:24 -0800293
Mark Fasheh198a1ca2008-11-20 17:54:57 -0800294 inode->i_nlink = ocfs2_read_links_count(fe);
Mark Fashehccd979b2005-12-15 14:31:24 -0800295
Tao Ma64f3b262011-02-22 22:24:57 +0800296 trace_ocfs2_populate_inode(OCFS2_I(inode)->ip_blkno,
297 le32_to_cpu(fe->i_flags));
Jan Karabbbd0eb2008-08-21 18:22:30 +0200298 if (fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) {
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700299 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SYSTEM_FILE;
Jan Karabbbd0eb2008-08-21 18:22:30 +0200300 inode->i_flags |= S_NOQUOTA;
301 }
Tao Ma64f3b262011-02-22 22:24:57 +0800302
Mark Fashehccd979b2005-12-15 14:31:24 -0800303 if (fe->i_flags & cpu_to_le32(OCFS2_LOCAL_ALLOC_FL)) {
304 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
Mark Fashehccd979b2005-12-15 14:31:24 -0800305 } else if (fe->i_flags & cpu_to_le32(OCFS2_BITMAP_FL)) {
306 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
Jan Kara1a224ad2008-08-20 15:43:36 +0200307 } else if (fe->i_flags & cpu_to_le32(OCFS2_QUOTA_FL)) {
308 inode->i_flags |= S_NOQUOTA;
Mark Fashehccd979b2005-12-15 14:31:24 -0800309 } else if (fe->i_flags & cpu_to_le32(OCFS2_SUPER_BLOCK_FL)) {
Mark Fashehccd979b2005-12-15 14:31:24 -0800310 /* we can't actually hit this as read_inode can't
311 * handle superblocks today ;-) */
312 BUG();
313 }
314
315 switch (inode->i_mode & S_IFMT) {
316 case S_IFREG:
Mark Fasheh53da4932008-07-21 14:29:16 -0700317 if (use_plocks)
318 inode->i_fop = &ocfs2_fops;
319 else
320 inode->i_fop = &ocfs2_fops_no_plocks;
Mark Fashehccd979b2005-12-15 14:31:24 -0800321 inode->i_op = &ocfs2_file_iops;
322 i_size_write(inode, le64_to_cpu(fe->i_size));
323 break;
324 case S_IFDIR:
325 inode->i_op = &ocfs2_dir_iops;
Mark Fasheh53da4932008-07-21 14:29:16 -0700326 if (use_plocks)
327 inode->i_fop = &ocfs2_dops;
328 else
329 inode->i_fop = &ocfs2_dops_no_plocks;
Mark Fashehccd979b2005-12-15 14:31:24 -0800330 i_size_write(inode, le64_to_cpu(fe->i_size));
Goldwyn Rodrigues5e98d492010-06-28 10:04:32 -0500331 OCFS2_I(inode)->ip_dir_lock_gen = 1;
Mark Fashehccd979b2005-12-15 14:31:24 -0800332 break;
333 case S_IFLNK:
334 if (ocfs2_inode_is_fast_symlink(inode))
335 inode->i_op = &ocfs2_fast_symlink_inode_operations;
336 else
337 inode->i_op = &ocfs2_symlink_inode_operations;
338 i_size_write(inode, le64_to_cpu(fe->i_size));
339 break;
340 default:
341 inode->i_op = &ocfs2_special_file_iops;
342 init_special_inode(inode, inode->i_mode,
343 inode->i_rdev);
344 break;
345 }
346
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700347 if (create_ino) {
348 inode->i_ino = ino_from_blkno(inode->i_sb,
349 le64_to_cpu(fe->i_blkno));
350
351 /*
352 * If we ever want to create system files from kernel,
353 * the generation argument to
354 * ocfs2_inode_lock_res_init() will have to change.
355 */
Mark Fasheh1ca1a112007-04-27 16:01:25 -0700356 BUG_ON(le32_to_cpu(fe->i_flags) & OCFS2_SYSTEM_FL);
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700357
Mark Fashehe63aecb62007-10-18 15:30:42 -0700358 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_inode_lockres,
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700359 OCFS2_LOCK_TYPE_META, 0, inode);
Tiger Yang50008632007-03-20 16:01:38 -0700360
361 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres,
362 OCFS2_LOCK_TYPE_OPEN, 0, inode);
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700363 }
364
Mark Fashehccd979b2005-12-15 14:31:24 -0800365 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_rw_lockres,
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700366 OCFS2_LOCK_TYPE_RW, inode->i_generation,
367 inode);
368
Herbert Poetzlca4d1472006-07-03 17:27:12 -0700369 ocfs2_set_inode_flags(inode);
Herbert Poetzlca4d1472006-07-03 17:27:12 -0700370
Tao Ma13821152009-02-25 00:53:23 +0800371 OCFS2_I(inode)->ip_last_used_slot = 0;
372 OCFS2_I(inode)->ip_last_used_group = 0;
Mark Fashehe3b4a972009-12-07 13:16:07 -0800373
374 if (S_ISDIR(inode->i_mode))
375 ocfs2_resv_set_type(&OCFS2_I(inode)->ip_la_data_resv,
376 OCFS2_RESV_FLAG_DIR);
Mark Fashehccd979b2005-12-15 14:31:24 -0800377}
378
379static int ocfs2_read_locked_inode(struct inode *inode,
380 struct ocfs2_find_inode_args *args)
381{
382 struct super_block *sb;
383 struct ocfs2_super *osb;
384 struct ocfs2_dinode *fe;
385 struct buffer_head *bh = NULL;
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700386 int status, can_lock;
387 u32 generation = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -0800388
Mark Fashehccd979b2005-12-15 14:31:24 -0800389 status = -EINVAL;
390 if (inode == NULL || inode->i_sb == NULL) {
391 mlog(ML_ERROR, "bad inode\n");
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700392 return status;
Mark Fashehccd979b2005-12-15 14:31:24 -0800393 }
394 sb = inode->i_sb;
395 osb = OCFS2_SB(sb);
396
397 if (!args) {
398 mlog(ML_ERROR, "bad inode args\n");
399 make_bad_inode(inode);
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700400 return status;
Mark Fashehccd979b2005-12-15 14:31:24 -0800401 }
402
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700403 /*
404 * To improve performance of cold-cache inode stats, we take
405 * the cluster lock here if possible.
406 *
407 * Generally, OCFS2 never trusts the contents of an inode
408 * unless it's holding a cluster lock, so taking it here isn't
409 * a correctness issue as much as it is a performance
410 * improvement.
411 *
412 * There are three times when taking the lock is not a good idea:
413 *
414 * 1) During startup, before we have initialized the DLM.
415 *
416 * 2) If we are reading certain system files which never get
417 * cluster locks (local alloc, truncate log).
418 *
419 * 3) If the process doing the iget() is responsible for
420 * orphan dir recovery. We're holding the orphan dir lock and
421 * can get into a deadlock with another process on another
422 * node in ->delete_inode().
423 *
424 * #1 and #2 can be simply solved by never taking the lock
425 * here for system files (which are the only type we read
426 * during mount). It's a heavier approach, but our main
Uwe Kleine-Königb5950762010-11-01 15:38:34 -0400427 * concern is user-accessible files anyway.
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700428 *
429 * #3 works itself out because we'll eventually take the
430 * cluster lock before trusting anything anyway.
431 */
432 can_lock = !(args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
Tiger Yang50008632007-03-20 16:01:38 -0700433 && !(args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY)
Sunil Mushranc271c5c2006-12-05 17:56:35 -0800434 && !ocfs2_mount_local(osb);
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700435
Tao Ma64f3b262011-02-22 22:24:57 +0800436 trace_ocfs2_read_locked_inode(
437 (unsigned long long)OCFS2_I(inode)->ip_blkno, can_lock);
438
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700439 /*
440 * To maintain backwards compatibility with older versions of
441 * ocfs2-tools, we still store the generation value for system
442 * files. The only ones that actually matter to userspace are
443 * the journals, but it's easier and inexpensive to just flag
444 * all system files similarly.
445 */
446 if (args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
447 generation = osb->fs_generation;
448
Mark Fashehe63aecb62007-10-18 15:30:42 -0700449 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_inode_lockres,
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700450 OCFS2_LOCK_TYPE_META,
451 generation, inode);
452
Tiger Yang50008632007-03-20 16:01:38 -0700453 ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres,
454 OCFS2_LOCK_TYPE_OPEN,
455 0, inode);
456
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700457 if (can_lock) {
Tiger Yang50008632007-03-20 16:01:38 -0700458 status = ocfs2_open_lock(inode);
459 if (status) {
460 make_bad_inode(inode);
461 mlog_errno(status);
462 return status;
463 }
Mark Fashehe63aecb62007-10-18 15:30:42 -0700464 status = ocfs2_inode_lock(inode, NULL, 0);
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700465 if (status) {
466 make_bad_inode(inode);
467 mlog_errno(status);
468 return status;
469 }
470 }
471
Tiger Yang50008632007-03-20 16:01:38 -0700472 if (args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY) {
473 status = ocfs2_try_open_lock(inode, 0);
474 if (status) {
Sunil Mushran2bd63212010-01-25 16:57:38 -0800475 make_bad_inode(inode);
Tiger Yang50008632007-03-20 16:01:38 -0700476 return status;
477 }
478 }
479
Joel Beckerb657c952008-11-13 14:49:11 -0800480 if (can_lock) {
481 status = ocfs2_read_inode_block_full(inode, &bh,
482 OCFS2_BH_IGNORE_CACHE);
483 } else {
Joel Beckerda1e9092008-10-09 17:20:29 -0700484 status = ocfs2_read_blocks_sync(osb, args->fi_blkno, 1, &bh);
Sunil Mushranf5ce5a02010-08-12 16:24:26 -0700485 /*
486 * If buffer is in jbd, then its checksum may not have been
487 * computed as yet.
488 */
489 if (!status && !buffer_jbd(bh))
Joel Beckerb657c952008-11-13 14:49:11 -0800490 status = ocfs2_validate_inode_block(osb->sb, bh);
491 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800492 if (status < 0) {
493 mlog_errno(status);
Mark Fashehccd979b2005-12-15 14:31:24 -0800494 goto bail;
495 }
496
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700497 status = -EINVAL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800498 fe = (struct ocfs2_dinode *) bh->b_data;
Mark Fashehccd979b2005-12-15 14:31:24 -0800499
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700500 /*
501 * This is a code bug. Right now the caller needs to
502 * understand whether it is asking for a system file inode or
503 * not so the proper lock names can be built.
504 */
505 mlog_bug_on_msg(!!(fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) !=
506 !!(args->fi_flags & OCFS2_FI_FLAG_SYSFILE),
507 "Inode %llu: system file state is ambigous\n",
508 (unsigned long long)args->fi_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -0800509
510 if (S_ISCHR(le16_to_cpu(fe->i_mode)) ||
511 S_ISBLK(le16_to_cpu(fe->i_mode)))
Joel Beckerb657c952008-11-13 14:49:11 -0800512 inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
Mark Fashehccd979b2005-12-15 14:31:24 -0800513
Joel Beckerb657c952008-11-13 14:49:11 -0800514 ocfs2_populate_inode(inode, fe, 0);
Mark Fashehccd979b2005-12-15 14:31:24 -0800515
516 BUG_ON(args->fi_blkno != le64_to_cpu(fe->i_blkno));
517
Mark Fashehccd979b2005-12-15 14:31:24 -0800518 status = 0;
519
520bail:
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700521 if (can_lock)
Mark Fashehe63aecb62007-10-18 15:30:42 -0700522 ocfs2_inode_unlock(inode, 0);
Mark Fasheh24c19ef2006-09-22 17:28:19 -0700523
524 if (status < 0)
525 make_bad_inode(inode);
526
Mark Fashehccd979b2005-12-15 14:31:24 -0800527 if (args && bh)
528 brelse(bh);
529
Mark Fashehccd979b2005-12-15 14:31:24 -0800530 return status;
531}
532
533void ocfs2_sync_blockdev(struct super_block *sb)
534{
535 sync_blockdev(sb->s_bdev);
536}
537
538static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
539 struct inode *inode,
540 struct buffer_head *fe_bh)
541{
542 int status = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -0800543 struct ocfs2_dinode *fe;
Mark Fasheh60b11392007-02-16 11:46:50 -0800544 handle_t *handle = NULL;
Mark Fashehccd979b2005-12-15 14:31:24 -0800545
Mark Fashehccd979b2005-12-15 14:31:24 -0800546 fe = (struct ocfs2_dinode *) fe_bh->b_data;
547
Mark Fasheh1afc32b2007-09-07 14:46:51 -0700548 /*
549 * This check will also skip truncate of inodes with inline
550 * data and fast symlinks.
551 */
Mark Fasheh3a0782d2007-01-17 12:53:31 -0800552 if (fe->i_clusters) {
Joel Becker2b4e30f2008-09-03 20:03:41 -0700553 if (ocfs2_should_order_data(inode))
554 ocfs2_begin_ordered_truncate(inode, 0);
555
Mark Fasheh60b11392007-02-16 11:46:50 -0800556 handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
557 if (IS_ERR(handle)) {
558 status = PTR_ERR(handle);
Dan Carpenter0350cb02010-04-22 11:39:29 +0200559 handle = NULL;
Mark Fasheh60b11392007-02-16 11:46:50 -0800560 mlog_errno(status);
561 goto out;
562 }
563
Joel Becker0cf2f762009-02-12 16:41:25 -0800564 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
565 fe_bh,
Joel Becker13723d02008-10-17 19:25:01 -0700566 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fasheh60b11392007-02-16 11:46:50 -0800567 if (status < 0) {
568 mlog_errno(status);
569 goto out;
570 }
571
572 i_size_write(inode, 0);
573
574 status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
575 if (status < 0) {
576 mlog_errno(status);
577 goto out;
578 }
579
580 ocfs2_commit_trans(osb, handle);
581 handle = NULL;
582
Tristan Ye78f94672010-05-11 17:54:42 +0800583 status = ocfs2_commit_truncate(osb, inode, fe_bh);
Mark Fasheh3a0782d2007-01-17 12:53:31 -0800584 if (status < 0) {
585 mlog_errno(status);
586 goto out;
587 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800588 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800589
Mark Fasheh60b11392007-02-16 11:46:50 -0800590out:
591 if (handle)
592 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800593 return status;
594}
595
596static int ocfs2_remove_inode(struct inode *inode,
597 struct buffer_head *di_bh,
598 struct inode *orphan_dir_inode,
599 struct buffer_head *orphan_dir_bh)
600{
601 int status;
602 struct inode *inode_alloc_inode = NULL;
603 struct buffer_head *inode_alloc_bh = NULL;
Mark Fasheh1fabe142006-10-09 18:11:45 -0700604 handle_t *handle;
Mark Fashehccd979b2005-12-15 14:31:24 -0800605 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
606 struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
607
608 inode_alloc_inode =
609 ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE,
610 le16_to_cpu(di->i_suballoc_slot));
611 if (!inode_alloc_inode) {
612 status = -EEXIST;
613 mlog_errno(status);
614 goto bail;
615 }
616
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800617 mutex_lock(&inode_alloc_inode->i_mutex);
Mark Fashehe63aecb62007-10-18 15:30:42 -0700618 status = ocfs2_inode_lock(inode_alloc_inode, &inode_alloc_bh, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -0800619 if (status < 0) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800620 mutex_unlock(&inode_alloc_inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -0800621
622 mlog_errno(status);
623 goto bail;
624 }
625
Jan Karaa90714c2008-10-09 19:38:40 +0200626 handle = ocfs2_start_trans(osb, OCFS2_DELETE_INODE_CREDITS +
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800627 ocfs2_quota_trans_credits(inode->i_sb));
Mark Fashehccd979b2005-12-15 14:31:24 -0800628 if (IS_ERR(handle)) {
629 status = PTR_ERR(handle);
630 mlog_errno(status);
631 goto bail_unlock;
632 }
633
Li Dongyangd4cd1872010-04-22 16:11:19 +0800634 if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR)) {
635 status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
636 orphan_dir_bh);
637 if (status < 0) {
638 mlog_errno(status);
639 goto bail_commit;
640 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800641 }
642
643 /* set the inodes dtime */
Joel Becker0cf2f762009-02-12 16:41:25 -0800644 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
Joel Becker13723d02008-10-17 19:25:01 -0700645 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehccd979b2005-12-15 14:31:24 -0800646 if (status < 0) {
647 mlog_errno(status);
648 goto bail_commit;
649 }
650
651 di->i_dtime = cpu_to_le64(CURRENT_TIME.tv_sec);
Marcin Slusarz4092d492007-12-25 15:52:59 +0100652 di->i_flags &= cpu_to_le32(~(OCFS2_VALID_FL | OCFS2_ORPHANED_FL));
Joel Beckerec20cec2010-03-19 14:13:52 -0700653 ocfs2_journal_dirty(handle, di_bh);
Mark Fashehccd979b2005-12-15 14:31:24 -0800654
Joel Becker8cb471e2009-02-10 20:00:41 -0800655 ocfs2_remove_from_cache(INODE_CACHE(inode), di_bh);
Christoph Hellwig63936dd2010-03-03 09:05:01 -0500656 dquot_free_inode(inode);
Mark Fashehccd979b2005-12-15 14:31:24 -0800657
658 status = ocfs2_free_dinode(handle, inode_alloc_inode,
659 inode_alloc_bh, di);
660 if (status < 0)
661 mlog_errno(status);
662
663bail_commit:
Mark Fasheh02dc1af2006-10-09 16:48:10 -0700664 ocfs2_commit_trans(osb, handle);
Mark Fashehccd979b2005-12-15 14:31:24 -0800665bail_unlock:
Mark Fashehe63aecb62007-10-18 15:30:42 -0700666 ocfs2_inode_unlock(inode_alloc_inode, 1);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800667 mutex_unlock(&inode_alloc_inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -0800668 brelse(inode_alloc_bh);
669bail:
670 iput(inode_alloc_inode);
671
672 return status;
673}
674
Sunil Mushran2bd63212010-01-25 16:57:38 -0800675/*
Mark Fashehb4df6ed2006-02-22 17:35:08 -0800676 * Serialize with orphan dir recovery. If the process doing
677 * recovery on this orphan dir does an iget() with the dir
678 * i_mutex held, we'll deadlock here. Instead we detect this
679 * and exit early - recovery will wipe this inode for us.
680 */
681static int ocfs2_check_orphan_recovery_state(struct ocfs2_super *osb,
682 int slot)
683{
684 int ret = 0;
685
686 spin_lock(&osb->osb_lock);
687 if (ocfs2_node_map_test_bit(osb, &osb->osb_recovering_orphan_dirs, slot)) {
Mark Fashehb4df6ed2006-02-22 17:35:08 -0800688 ret = -EDEADLK;
689 goto out;
690 }
691 /* This signals to the orphan recovery process that it should
692 * wait for us to handle the wipe. */
693 osb->osb_orphan_wipes[slot]++;
694out:
695 spin_unlock(&osb->osb_lock);
Tao Ma64f3b262011-02-22 22:24:57 +0800696 trace_ocfs2_check_orphan_recovery_state(slot, ret);
Mark Fashehb4df6ed2006-02-22 17:35:08 -0800697 return ret;
698}
699
700static void ocfs2_signal_wipe_completion(struct ocfs2_super *osb,
701 int slot)
702{
703 spin_lock(&osb->osb_lock);
704 osb->osb_orphan_wipes[slot]--;
705 spin_unlock(&osb->osb_lock);
706
707 wake_up(&osb->osb_wipe_event);
708}
709
Mark Fashehccd979b2005-12-15 14:31:24 -0800710static int ocfs2_wipe_inode(struct inode *inode,
711 struct buffer_head *di_bh)
712{
Joel Beckerd5776322010-05-03 19:15:49 -0700713 int status, orphaned_slot = -1;
Mark Fashehccd979b2005-12-15 14:31:24 -0800714 struct inode *orphan_dir_inode = NULL;
715 struct buffer_head *orphan_dir_bh = NULL;
716 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
Li Dongyangd4cd1872010-04-22 16:11:19 +0800717 struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
Mark Fashehccd979b2005-12-15 14:31:24 -0800718
Li Dongyangd4cd1872010-04-22 16:11:19 +0800719 if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR)) {
720 orphaned_slot = le16_to_cpu(di->i_orphaned_slot);
Mark Fashehb4df6ed2006-02-22 17:35:08 -0800721
Li Dongyangd4cd1872010-04-22 16:11:19 +0800722 status = ocfs2_check_orphan_recovery_state(osb, orphaned_slot);
723 if (status)
724 return status;
Mark Fashehb4df6ed2006-02-22 17:35:08 -0800725
Li Dongyangd4cd1872010-04-22 16:11:19 +0800726 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
727 ORPHAN_DIR_SYSTEM_INODE,
728 orphaned_slot);
729 if (!orphan_dir_inode) {
730 status = -EEXIST;
731 mlog_errno(status);
732 goto bail;
733 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800734
Li Dongyangd4cd1872010-04-22 16:11:19 +0800735 /* Lock the orphan dir. The lock will be held for the entire
736 * delete_inode operation. We do this now to avoid races with
737 * recovery completion on other nodes. */
738 mutex_lock(&orphan_dir_inode->i_mutex);
739 status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
740 if (status < 0) {
741 mutex_unlock(&orphan_dir_inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -0800742
Li Dongyangd4cd1872010-04-22 16:11:19 +0800743 mlog_errno(status);
744 goto bail;
745 }
Mark Fashehccd979b2005-12-15 14:31:24 -0800746 }
747
748 /* we do this while holding the orphan dir lock because we
Mark Fasheh34d024f2007-09-24 15:56:19 -0700749 * don't want recovery being run from another node to try an
750 * inode delete underneath us -- this will result in two nodes
Mark Fashehccd979b2005-12-15 14:31:24 -0800751 * truncating the same file! */
752 status = ocfs2_truncate_for_delete(osb, inode, di_bh);
753 if (status < 0) {
754 mlog_errno(status);
755 goto bail_unlock_dir;
756 }
757
Mark Fasheh9b7895e2008-11-12 16:27:44 -0800758 /* Remove any dir index tree */
759 if (S_ISDIR(inode->i_mode)) {
760 status = ocfs2_dx_dir_truncate(inode, di_bh);
761 if (status) {
762 mlog_errno(status);
763 goto bail_unlock_dir;
764 }
765 }
766
Tiger Yangcf1d6c72008-08-18 17:11:00 +0800767 /*Free extended attribute resources associated with this inode.*/
768 status = ocfs2_xattr_remove(inode, di_bh);
769 if (status < 0) {
770 mlog_errno(status);
771 goto bail_unlock_dir;
772 }
773
Tao Ma8b2c0db2009-08-18 11:43:49 +0800774 status = ocfs2_remove_refcount_tree(inode, di_bh);
775 if (status < 0) {
776 mlog_errno(status);
777 goto bail_unlock_dir;
778 }
779
Mark Fashehccd979b2005-12-15 14:31:24 -0800780 status = ocfs2_remove_inode(inode, di_bh, orphan_dir_inode,
781 orphan_dir_bh);
782 if (status < 0)
783 mlog_errno(status);
784
785bail_unlock_dir:
Li Dongyangd4cd1872010-04-22 16:11:19 +0800786 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR)
787 return status;
788
Mark Fashehe63aecb62007-10-18 15:30:42 -0700789 ocfs2_inode_unlock(orphan_dir_inode, 1);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800790 mutex_unlock(&orphan_dir_inode->i_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -0800791 brelse(orphan_dir_bh);
792bail:
793 iput(orphan_dir_inode);
Mark Fashehb4df6ed2006-02-22 17:35:08 -0800794 ocfs2_signal_wipe_completion(osb, orphaned_slot);
Mark Fashehccd979b2005-12-15 14:31:24 -0800795
796 return status;
797}
798
799/* There is a series of simple checks that should be done before a
Mark Fasheh34d024f2007-09-24 15:56:19 -0700800 * trylock is even considered. Encapsulate those in this function. */
Mark Fashehccd979b2005-12-15 14:31:24 -0800801static int ocfs2_inode_is_valid_to_delete(struct inode *inode)
802{
803 int ret = 0;
804 struct ocfs2_inode_info *oi = OCFS2_I(inode);
805 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
806
Tao Ma64f3b262011-02-22 22:24:57 +0800807 trace_ocfs2_inode_is_valid_to_delete(current, osb->dc_task,
808 (unsigned long long)oi->ip_blkno,
809 oi->ip_flags);
810
Mark Fashehccd979b2005-12-15 14:31:24 -0800811 /* We shouldn't be getting here for the root directory
812 * inode.. */
813 if (inode == osb->root_inode) {
814 mlog(ML_ERROR, "Skipping delete of root inode.\n");
815 goto bail;
816 }
817
Mark Fasheh34d024f2007-09-24 15:56:19 -0700818 /* If we're coming from downconvert_thread we can't go into our own
Mark Fashehccd979b2005-12-15 14:31:24 -0800819 * voting [hello, deadlock city!], so unforuntately we just
820 * have to skip deleting this guy. That's OK though because
821 * the node who's doing the actual deleting should handle it
822 * anyway. */
Tao Ma64f3b262011-02-22 22:24:57 +0800823 if (current == osb->dc_task)
Mark Fashehccd979b2005-12-15 14:31:24 -0800824 goto bail;
Mark Fashehccd979b2005-12-15 14:31:24 -0800825
826 spin_lock(&oi->ip_lock);
827 /* OCFS2 *never* deletes system files. This should technically
828 * never get here as system file inodes should always have a
829 * positive link count. */
830 if (oi->ip_flags & OCFS2_INODE_SYSTEM_FILE) {
Mark Fashehb0697052006-03-03 10:24:33 -0800831 mlog(ML_ERROR, "Skipping delete of system file %llu\n",
832 (unsigned long long)oi->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -0800833 goto bail_unlock;
834 }
835
Mark Fasheh34d024f2007-09-24 15:56:19 -0700836 /* If we have allowd wipe of this inode for another node, it
837 * will be marked here so we can safely skip it. Recovery will
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300838 * cleanup any inodes we might inadvertently skip here. */
Tao Ma64f3b262011-02-22 22:24:57 +0800839 if (oi->ip_flags & OCFS2_INODE_SKIP_DELETE)
Mark Fashehccd979b2005-12-15 14:31:24 -0800840 goto bail_unlock;
Mark Fashehccd979b2005-12-15 14:31:24 -0800841
842 ret = 1;
843bail_unlock:
844 spin_unlock(&oi->ip_lock);
845bail:
846 return ret;
847}
848
849/* Query the cluster to determine whether we should wipe an inode from
850 * disk or not.
851 *
852 * Requires the inode to have the cluster lock. */
853static int ocfs2_query_inode_wipe(struct inode *inode,
854 struct buffer_head *di_bh,
855 int *wipe)
856{
Tao Ma64f3b262011-02-22 22:24:57 +0800857 int status = 0, reason = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -0800858 struct ocfs2_inode_info *oi = OCFS2_I(inode);
859 struct ocfs2_dinode *di;
860
861 *wipe = 0;
862
Tao Ma64f3b262011-02-22 22:24:57 +0800863 trace_ocfs2_query_inode_wipe_begin((unsigned long long)oi->ip_blkno,
864 inode->i_nlink);
865
Mark Fashehccd979b2005-12-15 14:31:24 -0800866 /* While we were waiting for the cluster lock in
867 * ocfs2_delete_inode, another node might have asked to delete
868 * the inode. Recheck our flags to catch this. */
869 if (!ocfs2_inode_is_valid_to_delete(inode)) {
Tao Ma64f3b262011-02-22 22:24:57 +0800870 reason = 1;
Mark Fashehccd979b2005-12-15 14:31:24 -0800871 goto bail;
872 }
873
874 /* Now that we have an up to date inode, we can double check
875 * the link count. */
Tao Ma64f3b262011-02-22 22:24:57 +0800876 if (inode->i_nlink)
Mark Fashehccd979b2005-12-15 14:31:24 -0800877 goto bail;
Mark Fashehccd979b2005-12-15 14:31:24 -0800878
879 /* Do some basic inode verification... */
880 di = (struct ocfs2_dinode *) di_bh->b_data;
Li Dongyangd4cd1872010-04-22 16:11:19 +0800881 if (!(di->i_flags & cpu_to_le32(OCFS2_ORPHANED_FL)) &&
882 !(oi->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR)) {
Tristan Yeb54c2ca2010-03-19 09:21:10 +0800883 /*
884 * Inodes in the orphan dir must have ORPHANED_FL. The only
885 * inodes that come back out of the orphan dir are reflink
886 * targets. A reflink target may be moved out of the orphan
887 * dir between the time we scan the directory and the time we
888 * process it. This would lead to HAS_REFCOUNT_FL being set but
889 * ORPHANED_FL not.
890 */
891 if (di->i_dyn_features & cpu_to_le16(OCFS2_HAS_REFCOUNT_FL)) {
Tao Ma64f3b262011-02-22 22:24:57 +0800892 reason = 2;
Tristan Yeb54c2ca2010-03-19 09:21:10 +0800893 goto bail;
894 }
895
Mark Fashehccd979b2005-12-15 14:31:24 -0800896 /* for lack of a better error? */
897 status = -EEXIST;
898 mlog(ML_ERROR,
Mark Fashehb0697052006-03-03 10:24:33 -0800899 "Inode %llu (on-disk %llu) not orphaned! "
Mark Fashehccd979b2005-12-15 14:31:24 -0800900 "Disk flags 0x%x, inode flags 0x%x\n",
Mark Fashehb0697052006-03-03 10:24:33 -0800901 (unsigned long long)oi->ip_blkno,
Mark Fasheh1ca1a112007-04-27 16:01:25 -0700902 (unsigned long long)le64_to_cpu(di->i_blkno),
903 le32_to_cpu(di->i_flags), oi->ip_flags);
Mark Fashehccd979b2005-12-15 14:31:24 -0800904 goto bail;
905 }
906
907 /* has someone already deleted us?! baaad... */
908 if (di->i_dtime) {
909 status = -EEXIST;
910 mlog_errno(status);
911 goto bail;
912 }
913
Mark Fasheh6f16bf62007-03-20 17:17:54 -0700914 /*
915 * This is how ocfs2 determines whether an inode is still live
916 * within the cluster. Every node takes a shared read lock on
917 * the inode open lock in ocfs2_read_locked_inode(). When we
918 * get to ->delete_inode(), each node tries to convert it's
919 * lock to an exclusive. Trylocks are serialized by the inode
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300920 * meta data lock. If the upconvert succeeds, we know the inode
Mark Fasheh6f16bf62007-03-20 17:17:54 -0700921 * is no longer live and can be deleted.
922 *
923 * Though we call this with the meta data lock held, the
924 * trylock keeps us from ABBA deadlock.
925 */
926 status = ocfs2_try_open_lock(inode, 1);
Tiger Yang50008632007-03-20 16:01:38 -0700927 if (status == -EAGAIN) {
Mark Fashehccd979b2005-12-15 14:31:24 -0800928 status = 0;
Tao Ma64f3b262011-02-22 22:24:57 +0800929 reason = 3;
Mark Fashehccd979b2005-12-15 14:31:24 -0800930 goto bail;
931 }
932 if (status < 0) {
933 mlog_errno(status);
934 goto bail;
935 }
936
Tiger Yang50008632007-03-20 16:01:38 -0700937 *wipe = 1;
Tao Ma64f3b262011-02-22 22:24:57 +0800938 trace_ocfs2_query_inode_wipe_succ(le16_to_cpu(di->i_orphaned_slot));
Mark Fashehccd979b2005-12-15 14:31:24 -0800939
940bail:
Tao Ma64f3b262011-02-22 22:24:57 +0800941 trace_ocfs2_query_inode_wipe_end(status, reason);
Mark Fashehccd979b2005-12-15 14:31:24 -0800942 return status;
943}
944
945/* Support function for ocfs2_delete_inode. Will help us keep the
946 * inode data in a consistent state for clear_inode. Always truncates
947 * pages, optionally sync's them first. */
948static void ocfs2_cleanup_delete_inode(struct inode *inode,
949 int sync_data)
950{
Tao Ma64f3b262011-02-22 22:24:57 +0800951 trace_ocfs2_cleanup_delete_inode(
952 (unsigned long long)OCFS2_I(inode)->ip_blkno, sync_data);
Mark Fashehccd979b2005-12-15 14:31:24 -0800953 if (sync_data)
954 write_inode_now(inode, 1);
955 truncate_inode_pages(&inode->i_data, 0);
956}
957
Al Viro066d92d2010-06-08 21:28:10 -0400958static void ocfs2_delete_inode(struct inode *inode)
Mark Fashehccd979b2005-12-15 14:31:24 -0800959{
960 int wipe, status;
Joel Beckere4b963f2009-09-02 17:17:36 -0700961 sigset_t oldset;
Mark Fashehccd979b2005-12-15 14:31:24 -0800962 struct buffer_head *di_bh = NULL;
963
Tao Ma64f3b262011-02-22 22:24:57 +0800964 trace_ocfs2_delete_inode(inode->i_ino,
965 (unsigned long long)OCFS2_I(inode)->ip_blkno,
966 is_bad_inode(inode));
Mark Fashehccd979b2005-12-15 14:31:24 -0800967
Jan Karaa90714c2008-10-09 19:38:40 +0200968 /* When we fail in read_inode() we mark inode as bad. The second test
969 * catches the case when inode allocation fails before allocating
970 * a block for inode. */
Tao Ma64f3b262011-02-22 22:24:57 +0800971 if (is_bad_inode(inode) || !OCFS2_I(inode)->ip_blkno)
Mark Fashehccd979b2005-12-15 14:31:24 -0800972 goto bail;
Mark Fashehccd979b2005-12-15 14:31:24 -0800973
Christoph Hellwig871a2932010-03-03 09:05:07 -0500974 dquot_initialize(inode);
Christoph Hellwig907f4552010-03-03 09:05:06 -0500975
Mark Fashehccd979b2005-12-15 14:31:24 -0800976 if (!ocfs2_inode_is_valid_to_delete(inode)) {
977 /* It's probably not necessary to truncate_inode_pages
978 * here but we do it for safety anyway (it will most
979 * likely be a no-op anyway) */
980 ocfs2_cleanup_delete_inode(inode, 0);
981 goto bail;
982 }
983
984 /* We want to block signals in delete_inode as the lock and
985 * messaging paths may return us -ERESTARTSYS. Which would
986 * cause us to exit early, resulting in inodes being orphaned
987 * forever. */
Joel Beckere4b963f2009-09-02 17:17:36 -0700988 ocfs2_block_signals(&oldset);
Mark Fashehccd979b2005-12-15 14:31:24 -0800989
wengang wang6ca497a2009-03-06 21:29:10 +0800990 /*
991 * Synchronize us against ocfs2_get_dentry. We take this in
992 * shared mode so that all nodes can still concurrently
993 * process deletes.
994 */
995 status = ocfs2_nfs_sync_lock(OCFS2_SB(inode->i_sb), 0);
996 if (status < 0) {
997 mlog(ML_ERROR, "getting nfs sync lock(PR) failed %d\n", status);
998 ocfs2_cleanup_delete_inode(inode, 0);
999 goto bail_unblock;
1000 }
Mark Fashehccd979b2005-12-15 14:31:24 -08001001 /* Lock down the inode. This gives us an up to date view of
1002 * it's metadata (for verification), and allows us to
Mark Fasheh34d024f2007-09-24 15:56:19 -07001003 * serialize delete_inode on multiple nodes.
Mark Fashehccd979b2005-12-15 14:31:24 -08001004 *
1005 * Even though we might be doing a truncate, we don't take the
1006 * allocation lock here as it won't be needed - nobody will
1007 * have the file open.
1008 */
Mark Fashehe63aecb62007-10-18 15:30:42 -07001009 status = ocfs2_inode_lock(inode, &di_bh, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -08001010 if (status < 0) {
1011 if (status != -ENOENT)
1012 mlog_errno(status);
1013 ocfs2_cleanup_delete_inode(inode, 0);
wengang wang6ca497a2009-03-06 21:29:10 +08001014 goto bail_unlock_nfs_sync;
Mark Fashehccd979b2005-12-15 14:31:24 -08001015 }
1016
1017 /* Query the cluster. This will be the final decision made
1018 * before we go ahead and wipe the inode. */
1019 status = ocfs2_query_inode_wipe(inode, di_bh, &wipe);
1020 if (!wipe || status < 0) {
Mark Fasheh34d024f2007-09-24 15:56:19 -07001021 /* Error and remote inode busy both mean we won't be
Mark Fashehccd979b2005-12-15 14:31:24 -08001022 * removing the inode, so they take almost the same
1023 * path. */
1024 if (status < 0)
1025 mlog_errno(status);
1026
Mark Fasheh34d024f2007-09-24 15:56:19 -07001027 /* Someone in the cluster has disallowed a wipe of
1028 * this inode, or it was never completely
1029 * orphaned. Write out the pages and exit now. */
Mark Fashehccd979b2005-12-15 14:31:24 -08001030 ocfs2_cleanup_delete_inode(inode, 1);
1031 goto bail_unlock_inode;
1032 }
1033
1034 ocfs2_cleanup_delete_inode(inode, 0);
1035
1036 status = ocfs2_wipe_inode(inode, di_bh);
1037 if (status < 0) {
Mark Fashehb4df6ed2006-02-22 17:35:08 -08001038 if (status != -EDEADLK)
1039 mlog_errno(status);
Mark Fashehccd979b2005-12-15 14:31:24 -08001040 goto bail_unlock_inode;
1041 }
1042
Mark Fasheh24c19ef2006-09-22 17:28:19 -07001043 /*
1044 * Mark the inode as successfully deleted.
1045 *
1046 * This is important for ocfs2_clear_inode() as it will check
1047 * this flag and skip any checkpointing work
1048 *
1049 * ocfs2_stuff_meta_lvb() also uses this flag to invalidate
1050 * the LVB for other nodes.
1051 */
Mark Fashehccd979b2005-12-15 14:31:24 -08001052 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_DELETED;
1053
1054bail_unlock_inode:
Mark Fashehe63aecb62007-10-18 15:30:42 -07001055 ocfs2_inode_unlock(inode, 1);
Mark Fashehccd979b2005-12-15 14:31:24 -08001056 brelse(di_bh);
wengang wang6ca497a2009-03-06 21:29:10 +08001057
1058bail_unlock_nfs_sync:
1059 ocfs2_nfs_sync_unlock(OCFS2_SB(inode->i_sb), 0);
1060
Mark Fashehccd979b2005-12-15 14:31:24 -08001061bail_unblock:
Joel Beckere4b963f2009-09-02 17:17:36 -07001062 ocfs2_unblock_signals(&oldset);
Mark Fashehccd979b2005-12-15 14:31:24 -08001063bail:
Tao Mac1e8d352011-03-07 16:43:21 +08001064 return;
Mark Fashehccd979b2005-12-15 14:31:24 -08001065}
1066
Al Viro066d92d2010-06-08 21:28:10 -04001067static void ocfs2_clear_inode(struct inode *inode)
Mark Fashehccd979b2005-12-15 14:31:24 -08001068{
1069 int status;
1070 struct ocfs2_inode_info *oi = OCFS2_I(inode);
1071
Al Viro066d92d2010-06-08 21:28:10 -04001072 end_writeback(inode);
Tao Ma64f3b262011-02-22 22:24:57 +08001073 trace_ocfs2_clear_inode((unsigned long long)oi->ip_blkno,
1074 inode->i_nlink);
Mark Fashehccd979b2005-12-15 14:31:24 -08001075
1076 mlog_bug_on_msg(OCFS2_SB(inode->i_sb) == NULL,
1077 "Inode=%lu\n", inode->i_ino);
1078
Christoph Hellwig9f754752010-03-03 09:05:05 -05001079 dquot_drop(inode);
Christoph Hellwig257ba152010-03-03 09:05:04 -05001080
Mark Fasheh34d024f2007-09-24 15:56:19 -07001081 /* To preven remote deletes we hold open lock before, now it
1082 * is time to unlock PR and EX open locks. */
Tiger Yang50008632007-03-20 16:01:38 -07001083 ocfs2_open_unlock(inode);
1084
Mark Fashehccd979b2005-12-15 14:31:24 -08001085 /* Do these before all the other work so that we don't bounce
Mark Fasheh34d024f2007-09-24 15:56:19 -07001086 * the downconvert thread while waiting to destroy the locks. */
Mark Fashehccd979b2005-12-15 14:31:24 -08001087 ocfs2_mark_lockres_freeing(&oi->ip_rw_lockres);
Mark Fashehe63aecb62007-10-18 15:30:42 -07001088 ocfs2_mark_lockres_freeing(&oi->ip_inode_lockres);
Tiger Yang50008632007-03-20 16:01:38 -07001089 ocfs2_mark_lockres_freeing(&oi->ip_open_lockres);
Mark Fashehccd979b2005-12-15 14:31:24 -08001090
Mark Fasheh4fe370a2009-12-07 13:15:40 -08001091 ocfs2_resv_discard(&OCFS2_SB(inode->i_sb)->osb_la_resmap,
1092 &oi->ip_la_data_resv);
1093 ocfs2_resv_init_once(&oi->ip_la_data_resv);
1094
Mark Fashehccd979b2005-12-15 14:31:24 -08001095 /* We very well may get a clear_inode before all an inodes
1096 * metadata has hit disk. Of course, we can't drop any cluster
1097 * locks until the journal has finished with it. The only
1098 * exception here are successfully wiped inodes - their
1099 * metadata can now be considered to be part of the system
1100 * inodes from which it came. */
1101 if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED))
1102 ocfs2_checkpoint_inode(inode);
1103
1104 mlog_bug_on_msg(!list_empty(&oi->ip_io_markers),
Mark Fashehb0697052006-03-03 10:24:33 -08001105 "Clear inode of %llu, inode has io markers\n",
1106 (unsigned long long)oi->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08001107
Mark Fasheh83418972007-04-23 18:53:12 -07001108 ocfs2_extent_map_trunc(inode, 0);
1109
Mark Fashehccd979b2005-12-15 14:31:24 -08001110 status = ocfs2_drop_inode_locks(inode);
1111 if (status < 0)
1112 mlog_errno(status);
1113
1114 ocfs2_lock_res_free(&oi->ip_rw_lockres);
Mark Fashehe63aecb62007-10-18 15:30:42 -07001115 ocfs2_lock_res_free(&oi->ip_inode_lockres);
Tiger Yang50008632007-03-20 16:01:38 -07001116 ocfs2_lock_res_free(&oi->ip_open_lockres);
Mark Fashehccd979b2005-12-15 14:31:24 -08001117
Joel Becker66fb3452009-02-12 15:24:40 -08001118 ocfs2_metadata_cache_exit(INODE_CACHE(inode));
Mark Fashehccd979b2005-12-15 14:31:24 -08001119
Joel Becker8cb471e2009-02-10 20:00:41 -08001120 mlog_bug_on_msg(INODE_CACHE(inode)->ci_num_cached,
Mark Fashehb0697052006-03-03 10:24:33 -08001121 "Clear inode of %llu, inode has %u cache items\n",
Joel Becker8cb471e2009-02-10 20:00:41 -08001122 (unsigned long long)oi->ip_blkno,
1123 INODE_CACHE(inode)->ci_num_cached);
Mark Fashehccd979b2005-12-15 14:31:24 -08001124
Joel Becker8cb471e2009-02-10 20:00:41 -08001125 mlog_bug_on_msg(!(INODE_CACHE(inode)->ci_flags & OCFS2_CACHE_FL_INLINE),
Mark Fashehb0697052006-03-03 10:24:33 -08001126 "Clear inode of %llu, inode has a bad flag\n",
1127 (unsigned long long)oi->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08001128
1129 mlog_bug_on_msg(spin_is_locked(&oi->ip_lock),
Mark Fashehb0697052006-03-03 10:24:33 -08001130 "Clear inode of %llu, inode is locked\n",
1131 (unsigned long long)oi->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08001132
Mark Fasheh251b6ec2006-01-10 15:41:43 -08001133 mlog_bug_on_msg(!mutex_trylock(&oi->ip_io_mutex),
Mark Fashehb0697052006-03-03 10:24:33 -08001134 "Clear inode of %llu, io_mutex is locked\n",
1135 (unsigned long long)oi->ip_blkno);
Mark Fasheh251b6ec2006-01-10 15:41:43 -08001136 mutex_unlock(&oi->ip_io_mutex);
Mark Fashehccd979b2005-12-15 14:31:24 -08001137
1138 /*
1139 * down_trylock() returns 0, down_write_trylock() returns 1
1140 * kernel 1, world 0
1141 */
1142 mlog_bug_on_msg(!down_write_trylock(&oi->ip_alloc_sem),
Mark Fashehb0697052006-03-03 10:24:33 -08001143 "Clear inode of %llu, alloc_sem is locked\n",
1144 (unsigned long long)oi->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08001145 up_write(&oi->ip_alloc_sem);
1146
1147 mlog_bug_on_msg(oi->ip_open_count,
Mark Fashehb0697052006-03-03 10:24:33 -08001148 "Clear inode of %llu has open count %d\n",
1149 (unsigned long long)oi->ip_blkno, oi->ip_open_count);
Mark Fashehccd979b2005-12-15 14:31:24 -08001150
1151 /* Clear all other flags. */
Joel Becker47460d62009-02-10 16:05:07 -08001152 oi->ip_flags = 0;
Mark Fashehccd979b2005-12-15 14:31:24 -08001153 oi->ip_dir_start_lookup = 0;
1154 oi->ip_blkno = 0ULL;
Sunil Mushranae0dff62008-10-22 13:24:29 -07001155
1156 /*
1157 * ip_jinode is used to track txns against this inode. We ensure that
1158 * the journal is flushed before journal shutdown. Thus it is safe to
1159 * have inodes get cleaned up after journal shutdown.
1160 */
Joel Becker2b4e30f2008-09-03 20:03:41 -07001161 jbd2_journal_release_jbd_inode(OCFS2_SB(inode->i_sb)->journal->j_journal,
1162 &oi->ip_jinode);
Mark Fashehccd979b2005-12-15 14:31:24 -08001163}
1164
Al Viro066d92d2010-06-08 21:28:10 -04001165void ocfs2_evict_inode(struct inode *inode)
1166{
1167 if (!inode->i_nlink ||
1168 (OCFS2_I(inode)->ip_flags & OCFS2_INODE_MAYBE_ORPHANED)) {
1169 ocfs2_delete_inode(inode);
1170 } else {
1171 truncate_inode_pages(&inode->i_data, 0);
1172 }
1173 ocfs2_clear_inode(inode);
1174}
1175
Mark Fashehccd979b2005-12-15 14:31:24 -08001176/* Called under inode_lock, with no more references on the
1177 * struct inode, so it's safe here to check the flags field
1178 * and to manipulate i_nlink without any other locks. */
Al Viro45321ac2010-06-07 13:43:19 -04001179int ocfs2_drop_inode(struct inode *inode)
Mark Fashehccd979b2005-12-15 14:31:24 -08001180{
1181 struct ocfs2_inode_info *oi = OCFS2_I(inode);
Al Viro45321ac2010-06-07 13:43:19 -04001182 int res;
Mark Fashehccd979b2005-12-15 14:31:24 -08001183
Tao Ma64f3b262011-02-22 22:24:57 +08001184 trace_ocfs2_drop_inode((unsigned long long)oi->ip_blkno,
1185 inode->i_nlink, oi->ip_flags);
Mark Fashehccd979b2005-12-15 14:31:24 -08001186
Mark Fasheh379dfe92006-09-08 14:21:03 -07001187 if (oi->ip_flags & OCFS2_INODE_MAYBE_ORPHANED)
Al Viro45321ac2010-06-07 13:43:19 -04001188 res = 1;
Mark Fasheh379dfe92006-09-08 14:21:03 -07001189 else
Al Viro45321ac2010-06-07 13:43:19 -04001190 res = generic_drop_inode(inode);
Mark Fashehccd979b2005-12-15 14:31:24 -08001191
Al Viro45321ac2010-06-07 13:43:19 -04001192 return res;
Mark Fashehccd979b2005-12-15 14:31:24 -08001193}
1194
1195/*
Mark Fashehccd979b2005-12-15 14:31:24 -08001196 * This is called from our getattr.
1197 */
1198int ocfs2_inode_revalidate(struct dentry *dentry)
1199{
1200 struct inode *inode = dentry->d_inode;
1201 int status = 0;
1202
Tao Ma64f3b262011-02-22 22:24:57 +08001203 trace_ocfs2_inode_revalidate(inode,
1204 inode ? (unsigned long long)OCFS2_I(inode)->ip_blkno : 0ULL,
1205 inode ? (unsigned long long)OCFS2_I(inode)->ip_flags : 0);
Mark Fashehccd979b2005-12-15 14:31:24 -08001206
1207 if (!inode) {
Mark Fashehccd979b2005-12-15 14:31:24 -08001208 status = -ENOENT;
1209 goto bail;
1210 }
1211
1212 spin_lock(&OCFS2_I(inode)->ip_lock);
1213 if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
1214 spin_unlock(&OCFS2_I(inode)->ip_lock);
Mark Fashehccd979b2005-12-15 14:31:24 -08001215 status = -ENOENT;
1216 goto bail;
1217 }
1218 spin_unlock(&OCFS2_I(inode)->ip_lock);
1219
Mark Fashehe63aecb62007-10-18 15:30:42 -07001220 /* Let ocfs2_inode_lock do the work of updating our struct
Mark Fashehccd979b2005-12-15 14:31:24 -08001221 * inode for us. */
Mark Fashehe63aecb62007-10-18 15:30:42 -07001222 status = ocfs2_inode_lock(inode, NULL, 0);
Mark Fashehccd979b2005-12-15 14:31:24 -08001223 if (status < 0) {
1224 if (status != -ENOENT)
1225 mlog_errno(status);
1226 goto bail;
1227 }
Mark Fashehe63aecb62007-10-18 15:30:42 -07001228 ocfs2_inode_unlock(inode, 0);
Mark Fashehccd979b2005-12-15 14:31:24 -08001229bail:
Mark Fashehccd979b2005-12-15 14:31:24 -08001230 return status;
1231}
1232
1233/*
1234 * Updates a disk inode from a
1235 * struct inode.
1236 * Only takes ip_lock.
1237 */
Mark Fasheh1fabe142006-10-09 18:11:45 -07001238int ocfs2_mark_inode_dirty(handle_t *handle,
Mark Fashehccd979b2005-12-15 14:31:24 -08001239 struct inode *inode,
1240 struct buffer_head *bh)
1241{
1242 int status;
1243 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
1244
Tao Ma64f3b262011-02-22 22:24:57 +08001245 trace_ocfs2_mark_inode_dirty((unsigned long long)OCFS2_I(inode)->ip_blkno);
Mark Fashehccd979b2005-12-15 14:31:24 -08001246
Joel Becker0cf2f762009-02-12 16:41:25 -08001247 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
Joel Becker13723d02008-10-17 19:25:01 -07001248 OCFS2_JOURNAL_ACCESS_WRITE);
Mark Fashehccd979b2005-12-15 14:31:24 -08001249 if (status < 0) {
1250 mlog_errno(status);
1251 goto leave;
1252 }
1253
1254 spin_lock(&OCFS2_I(inode)->ip_lock);
1255 fe->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters);
Jan Kara6e4b0d52007-04-27 11:08:01 -07001256 ocfs2_get_inode_flags(OCFS2_I(inode));
Herbert Poetzlca4d1472006-07-03 17:27:12 -07001257 fe->i_attr = cpu_to_le32(OCFS2_I(inode)->ip_attr);
Mark Fasheh15b1e362007-09-07 13:58:15 -07001258 fe->i_dyn_features = cpu_to_le16(OCFS2_I(inode)->ip_dyn_features);
Mark Fashehccd979b2005-12-15 14:31:24 -08001259 spin_unlock(&OCFS2_I(inode)->ip_lock);
1260
1261 fe->i_size = cpu_to_le64(i_size_read(inode));
Mark Fasheh198a1ca2008-11-20 17:54:57 -08001262 ocfs2_set_links_count(fe, inode->i_nlink);
Mark Fashehccd979b2005-12-15 14:31:24 -08001263 fe->i_uid = cpu_to_le32(inode->i_uid);
1264 fe->i_gid = cpu_to_le32(inode->i_gid);
1265 fe->i_mode = cpu_to_le16(inode->i_mode);
1266 fe->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
1267 fe->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
1268 fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
1269 fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
1270 fe->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
1271 fe->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
1272
Joel Beckerec20cec2010-03-19 14:13:52 -07001273 ocfs2_journal_dirty(handle, bh);
Mark Fashehccd979b2005-12-15 14:31:24 -08001274leave:
Mark Fashehccd979b2005-12-15 14:31:24 -08001275 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));
Mark Fasheh198a1ca2008-11-20 17:54:57 -08001293 inode->i_nlink = ocfs2_read_links_count(fe);
Mark Fashehccd979b2005-12-15 14:31:24 -08001294 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}
Joel Beckerb657c952008-11-13 14:49:11 -08001310
1311int ocfs2_validate_inode_block(struct super_block *sb,
1312 struct buffer_head *bh)
1313{
Joel Beckerd6b32bb2008-10-17 14:55:01 -07001314 int rc;
Joel Beckerb657c952008-11-13 14:49:11 -08001315 struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
1316
Tao Ma64f3b262011-02-22 22:24:57 +08001317 trace_ocfs2_validate_inode_block((unsigned long long)bh->b_blocknr);
Joel Becker970e4932008-11-13 14:49:19 -08001318
Joel Beckerb657c952008-11-13 14:49:11 -08001319 BUG_ON(!buffer_uptodate(bh));
1320
Joel Beckerd6b32bb2008-10-17 14:55:01 -07001321 /*
1322 * If the ecc fails, we return the error but otherwise
1323 * leave the filesystem running. We know any error is
1324 * local to this block.
1325 */
1326 rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &di->i_check);
Joel Becker13723d02008-10-17 19:25:01 -07001327 if (rc) {
1328 mlog(ML_ERROR, "Checksum failed for dinode %llu\n",
1329 (unsigned long long)bh->b_blocknr);
Joel Beckerd6b32bb2008-10-17 14:55:01 -07001330 goto bail;
Joel Becker13723d02008-10-17 19:25:01 -07001331 }
Joel Beckerd6b32bb2008-10-17 14:55:01 -07001332
1333 /*
1334 * Errors after here are fatal.
1335 */
1336
1337 rc = -EINVAL;
1338
Joel Beckerb657c952008-11-13 14:49:11 -08001339 if (!OCFS2_IS_VALID_DINODE(di)) {
1340 ocfs2_error(sb, "Invalid dinode #%llu: signature = %.*s\n",
1341 (unsigned long long)bh->b_blocknr, 7,
1342 di->i_signature);
1343 goto bail;
1344 }
1345
1346 if (le64_to_cpu(di->i_blkno) != bh->b_blocknr) {
1347 ocfs2_error(sb, "Invalid dinode #%llu: i_blkno is %llu\n",
1348 (unsigned long long)bh->b_blocknr,
1349 (unsigned long long)le64_to_cpu(di->i_blkno));
1350 goto bail;
1351 }
1352
1353 if (!(di->i_flags & cpu_to_le32(OCFS2_VALID_FL))) {
1354 ocfs2_error(sb,
1355 "Invalid dinode #%llu: OCFS2_VALID_FL not set\n",
1356 (unsigned long long)bh->b_blocknr);
1357 goto bail;
1358 }
1359
1360 if (le32_to_cpu(di->i_fs_generation) !=
1361 OCFS2_SB(sb)->fs_generation) {
1362 ocfs2_error(sb,
1363 "Invalid dinode #%llu: fs_generation is %u\n",
1364 (unsigned long long)bh->b_blocknr,
1365 le32_to_cpu(di->i_fs_generation));
1366 goto bail;
1367 }
1368
1369 rc = 0;
1370
1371bail:
1372 return rc;
1373}
1374
1375int ocfs2_read_inode_block_full(struct inode *inode, struct buffer_head **bh,
1376 int flags)
1377{
1378 int rc;
1379 struct buffer_head *tmp = *bh;
1380
Joel Becker8cb471e2009-02-10 20:00:41 -08001381 rc = ocfs2_read_blocks(INODE_CACHE(inode), OCFS2_I(inode)->ip_blkno,
1382 1, &tmp, flags, ocfs2_validate_inode_block);
Joel Beckerb657c952008-11-13 14:49:11 -08001383
1384 /* If ocfs2_read_blocks() got us a new bh, pass it up. */
Joel Becker970e4932008-11-13 14:49:19 -08001385 if (!rc && !*bh)
Joel Beckerb657c952008-11-13 14:49:11 -08001386 *bh = tmp;
1387
Joel Beckerb657c952008-11-13 14:49:11 -08001388 return rc;
1389}
1390
1391int ocfs2_read_inode_block(struct inode *inode, struct buffer_head **bh)
1392{
1393 return ocfs2_read_inode_block_full(inode, bh, 0);
1394}
Joel Becker6e5a3d72009-02-10 19:00:37 -08001395
Joel Becker6e5a3d72009-02-10 19:00:37 -08001396
1397static u64 ocfs2_inode_cache_owner(struct ocfs2_caching_info *ci)
1398{
1399 struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1400
1401 return oi->ip_blkno;
1402}
1403
Joel Becker8cb471e2009-02-10 20:00:41 -08001404static struct super_block *ocfs2_inode_cache_get_super(struct ocfs2_caching_info *ci)
1405{
1406 struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1407
1408 return oi->vfs_inode.i_sb;
1409}
1410
Joel Becker6e5a3d72009-02-10 19:00:37 -08001411static void ocfs2_inode_cache_lock(struct ocfs2_caching_info *ci)
1412{
1413 struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1414
1415 spin_lock(&oi->ip_lock);
1416}
1417
1418static void ocfs2_inode_cache_unlock(struct ocfs2_caching_info *ci)
1419{
1420 struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1421
1422 spin_unlock(&oi->ip_lock);
1423}
1424
1425static void ocfs2_inode_cache_io_lock(struct ocfs2_caching_info *ci)
1426{
1427 struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1428
1429 mutex_lock(&oi->ip_io_mutex);
1430}
1431
1432static void ocfs2_inode_cache_io_unlock(struct ocfs2_caching_info *ci)
1433{
1434 struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1435
1436 mutex_unlock(&oi->ip_io_mutex);
1437}
1438
1439const struct ocfs2_caching_operations ocfs2_inode_caching_ops = {
1440 .co_owner = ocfs2_inode_cache_owner,
Joel Becker8cb471e2009-02-10 20:00:41 -08001441 .co_get_super = ocfs2_inode_cache_get_super,
Joel Becker6e5a3d72009-02-10 19:00:37 -08001442 .co_cache_lock = ocfs2_inode_cache_lock,
1443 .co_cache_unlock = ocfs2_inode_cache_unlock,
1444 .co_io_lock = ocfs2_inode_cache_io_lock,
1445 .co_io_unlock = ocfs2_inode_cache_io_unlock,
1446};
1447