Root/fs/ocfs2/inode.c

1/* -*- 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/highmem.h>
29#include <linux/pagemap.h>
30#include <linux/quotaops.h>
31
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 "dir.h"
41#include "blockcheck.h"
42#include "dlmglue.h"
43#include "extent_map.h"
44#include "file.h"
45#include "heartbeat.h"
46#include "inode.h"
47#include "journal.h"
48#include "namei.h"
49#include "suballoc.h"
50#include "super.h"
51#include "symlink.h"
52#include "sysfile.h"
53#include "uptodate.h"
54#include "xattr.h"
55#include "refcounttree.h"
56
57#include "buffer_head_io.h"
58
59struct ocfs2_find_inode_args
60{
61    u64 fi_blkno;
62    unsigned long fi_ino;
63    unsigned int fi_flags;
64    unsigned int fi_sysfile_type;
65};
66
67static struct lock_class_key ocfs2_sysfile_lock_key[NUM_SYSTEM_INODES];
68
69static 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
77void 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
97/* 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
116struct 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}
127struct inode *ocfs2_iget(struct ocfs2_super *osb, u64 blkno, unsigned flags,
128             int sysfile_type)
129{
130    struct inode *inode = NULL;
131    struct super_block *sb = osb->sb;
132    struct ocfs2_find_inode_args args;
133
134    mlog_entry("(blkno = %llu)\n", (unsigned long long)blkno);
135
136    /* Ok. By now we've either got the offsets passed to us by the
137     * caller, or we just pulled them off the bh. Lets do some
138     * sanity checks to make sure they're OK. */
139    if (blkno == 0) {
140        inode = ERR_PTR(-EINVAL);
141        mlog_errno(PTR_ERR(inode));
142        goto bail;
143    }
144
145    args.fi_blkno = blkno;
146    args.fi_flags = flags;
147    args.fi_ino = ino_from_blkno(sb, blkno);
148    args.fi_sysfile_type = sysfile_type;
149
150    inode = iget5_locked(sb, args.fi_ino, ocfs2_find_actor,
151                 ocfs2_init_locked_inode, &args);
152    /* inode was *not* in the inode cache. 2.6.x requires
153     * us to do our own read_inode call and unlock it
154     * afterwards. */
155    if (inode && inode->i_state & I_NEW) {
156        mlog(0, "Inode was not in inode cache, reading it.\n");
157        ocfs2_read_locked_inode(inode, &args);
158        unlock_new_inode(inode);
159    }
160    if (inode == NULL) {
161        inode = ERR_PTR(-ENOMEM);
162        mlog_errno(PTR_ERR(inode));
163        goto bail;
164    }
165    if (is_bad_inode(inode)) {
166        iput(inode);
167        inode = ERR_PTR(-ESTALE);
168        goto bail;
169    }
170
171bail:
172    if (!IS_ERR(inode)) {
173        mlog(0, "returning inode with number %llu\n",
174             (unsigned long long)OCFS2_I(inode)->ip_blkno);
175        mlog_exit_ptr(inode);
176    }
177
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
195    mlog_entry("(0x%p, %lu, 0x%p)\n", inode, inode->i_ino, opaque);
196
197    args = opaque;
198
199    mlog_bug_on_msg(!inode, "No inode in find actor!\n");
200
201    if (oi->ip_blkno != args->fi_blkno)
202        goto bail;
203
204    ret = 1;
205bail:
206    mlog_exit(ret);
207    return ret;
208}
209
210/*
211 * initialize the new inode, but don't do anything that would cause
212 * us to sleep.
213 * return 0 on success, 1 on failure
214 */
215static int ocfs2_init_locked_inode(struct inode *inode, void *opaque)
216{
217    struct ocfs2_find_inode_args *args = opaque;
218    static struct lock_class_key ocfs2_quota_ip_alloc_sem_key,
219                     ocfs2_file_ip_alloc_sem_key;
220
221    mlog_entry("inode = %p, opaque = %p\n", inode, opaque);
222
223    inode->i_ino = args->fi_ino;
224    OCFS2_I(inode)->ip_blkno = args->fi_blkno;
225    if (args->fi_sysfile_type != 0)
226        lockdep_set_class(&inode->i_mutex,
227            &ocfs2_sysfile_lock_key[args->fi_sysfile_type]);
228    if (args->fi_sysfile_type == USER_QUOTA_SYSTEM_INODE ||
229        args->fi_sysfile_type == GROUP_QUOTA_SYSTEM_INODE ||
230        args->fi_sysfile_type == LOCAL_USER_QUOTA_SYSTEM_INODE ||
231        args->fi_sysfile_type == LOCAL_GROUP_QUOTA_SYSTEM_INODE)
232        lockdep_set_class(&OCFS2_I(inode)->ip_alloc_sem,
233                  &ocfs2_quota_ip_alloc_sem_key);
234    else
235        lockdep_set_class(&OCFS2_I(inode)->ip_alloc_sem,
236                  &ocfs2_file_ip_alloc_sem_key);
237
238    mlog_exit(0);
239    return 0;
240}
241
242void ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe,
243              int create_ino)
244{
245    struct super_block *sb;
246    struct ocfs2_super *osb;
247    int use_plocks = 1;
248
249    mlog_entry("(0x%p, size:%llu)\n", inode,
250           (unsigned long long)le64_to_cpu(fe->i_size));
251
252    sb = inode->i_sb;
253    osb = OCFS2_SB(sb);
254
255    if ((osb->s_mount_opt & OCFS2_MOUNT_LOCALFLOCKS) ||
256        ocfs2_mount_local(osb) || !ocfs2_stack_supports_plocks())
257        use_plocks = 0;
258
259    /*
260     * These have all been checked by ocfs2_read_inode_block() or set
261     * by ocfs2_mknod_locked(), so a failure is a code bug.
262     */
263    BUG_ON(!OCFS2_IS_VALID_DINODE(fe)); /* This means that read_inode
264                        cannot create a superblock
265                        inode today. change if
266                        that is needed. */
267    BUG_ON(!(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL)));
268    BUG_ON(le32_to_cpu(fe->i_fs_generation) != osb->fs_generation);
269
270
271    OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
272    OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
273    OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features);
274
275    inode->i_version = 1;
276    inode->i_generation = le32_to_cpu(fe->i_generation);
277    inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
278    inode->i_mode = le16_to_cpu(fe->i_mode);
279    inode->i_uid = le32_to_cpu(fe->i_uid);
280    inode->i_gid = le32_to_cpu(fe->i_gid);
281
282    /* Fast symlinks will have i_size but no allocated clusters. */
283    if (S_ISLNK(inode->i_mode) && !fe->i_clusters)
284        inode->i_blocks = 0;
285    else
286        inode->i_blocks = ocfs2_inode_sector_count(inode);
287    inode->i_mapping->a_ops = &ocfs2_aops;
288    inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
289    inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
290    inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
291    inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
292    inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
293    inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
294
295    if (OCFS2_I(inode)->ip_blkno != le64_to_cpu(fe->i_blkno))
296        mlog(ML_ERROR,
297             "ip_blkno %llu != i_blkno %llu!\n",
298             (unsigned long long)OCFS2_I(inode)->ip_blkno,
299             (unsigned long long)le64_to_cpu(fe->i_blkno));
300
301    inode->i_nlink = ocfs2_read_links_count(fe);
302
303    if (fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) {
304        OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SYSTEM_FILE;
305        inode->i_flags |= S_NOQUOTA;
306    }
307
308    if (fe->i_flags & cpu_to_le32(OCFS2_LOCAL_ALLOC_FL)) {
309        OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
310        mlog(0, "local alloc inode: i_ino=%lu\n", inode->i_ino);
311    } else if (fe->i_flags & cpu_to_le32(OCFS2_BITMAP_FL)) {
312        OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
313    } else if (fe->i_flags & cpu_to_le32(OCFS2_QUOTA_FL)) {
314        inode->i_flags |= S_NOQUOTA;
315    } else if (fe->i_flags & cpu_to_le32(OCFS2_SUPER_BLOCK_FL)) {
316        mlog(0, "superblock inode: i_ino=%lu\n", inode->i_ino);
317        /* we can't actually hit this as read_inode can't
318         * handle superblocks today ;-) */
319        BUG();
320    }
321
322    switch (inode->i_mode & S_IFMT) {
323        case S_IFREG:
324            if (use_plocks)
325                inode->i_fop = &ocfs2_fops;
326            else
327                inode->i_fop = &ocfs2_fops_no_plocks;
328            inode->i_op = &ocfs2_file_iops;
329            i_size_write(inode, le64_to_cpu(fe->i_size));
330            break;
331        case S_IFDIR:
332            inode->i_op = &ocfs2_dir_iops;
333            if (use_plocks)
334                inode->i_fop = &ocfs2_dops;
335            else
336                inode->i_fop = &ocfs2_dops_no_plocks;
337            i_size_write(inode, le64_to_cpu(fe->i_size));
338            break;
339        case S_IFLNK:
340            if (ocfs2_inode_is_fast_symlink(inode))
341            inode->i_op = &ocfs2_fast_symlink_inode_operations;
342            else
343            inode->i_op = &ocfs2_symlink_inode_operations;
344            i_size_write(inode, le64_to_cpu(fe->i_size));
345            break;
346        default:
347            inode->i_op = &ocfs2_special_file_iops;
348            init_special_inode(inode, inode->i_mode,
349                       inode->i_rdev);
350            break;
351    }
352
353    if (create_ino) {
354        inode->i_ino = ino_from_blkno(inode->i_sb,
355                   le64_to_cpu(fe->i_blkno));
356
357        /*
358         * If we ever want to create system files from kernel,
359         * the generation argument to
360         * ocfs2_inode_lock_res_init() will have to change.
361         */
362        BUG_ON(le32_to_cpu(fe->i_flags) & OCFS2_SYSTEM_FL);
363
364        ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_inode_lockres,
365                      OCFS2_LOCK_TYPE_META, 0, inode);
366
367        ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres,
368                      OCFS2_LOCK_TYPE_OPEN, 0, inode);
369    }
370
371    ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_rw_lockres,
372                  OCFS2_LOCK_TYPE_RW, inode->i_generation,
373                  inode);
374
375    ocfs2_set_inode_flags(inode);
376
377    OCFS2_I(inode)->ip_last_used_slot = 0;
378    OCFS2_I(inode)->ip_last_used_group = 0;
379    mlog_exit_void();
380}
381
382static int ocfs2_read_locked_inode(struct inode *inode,
383                   struct ocfs2_find_inode_args *args)
384{
385    struct super_block *sb;
386    struct ocfs2_super *osb;
387    struct ocfs2_dinode *fe;
388    struct buffer_head *bh = NULL;
389    int status, can_lock;
390    u32 generation = 0;
391
392    mlog_entry("(0x%p, 0x%p)\n", inode, args);
393
394    status = -EINVAL;
395    if (inode == NULL || inode->i_sb == NULL) {
396        mlog(ML_ERROR, "bad inode\n");
397        return status;
398    }
399    sb = inode->i_sb;
400    osb = OCFS2_SB(sb);
401
402    if (!args) {
403        mlog(ML_ERROR, "bad inode args\n");
404        make_bad_inode(inode);
405        return status;
406    }
407
408    /*
409     * To improve performance of cold-cache inode stats, we take
410     * the cluster lock here if possible.
411     *
412     * Generally, OCFS2 never trusts the contents of an inode
413     * unless it's holding a cluster lock, so taking it here isn't
414     * a correctness issue as much as it is a performance
415     * improvement.
416     *
417     * There are three times when taking the lock is not a good idea:
418     *
419     * 1) During startup, before we have initialized the DLM.
420     *
421     * 2) If we are reading certain system files which never get
422     * cluster locks (local alloc, truncate log).
423     *
424     * 3) If the process doing the iget() is responsible for
425     * orphan dir recovery. We're holding the orphan dir lock and
426     * can get into a deadlock with another process on another
427     * node in ->delete_inode().
428     *
429     * #1 and #2 can be simply solved by never taking the lock
430     * here for system files (which are the only type we read
431     * during mount). It's a heavier approach, but our main
432     * concern is user-accesible files anyway.
433     *
434     * #3 works itself out because we'll eventually take the
435     * cluster lock before trusting anything anyway.
436     */
437    can_lock = !(args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
438        && !(args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY)
439        && !ocfs2_mount_local(osb);
440
441    /*
442     * To maintain backwards compatibility with older versions of
443     * ocfs2-tools, we still store the generation value for system
444     * files. The only ones that actually matter to userspace are
445     * the journals, but it's easier and inexpensive to just flag
446     * all system files similarly.
447     */
448    if (args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
449        generation = osb->fs_generation;
450
451    ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_inode_lockres,
452                  OCFS2_LOCK_TYPE_META,
453                  generation, inode);
454
455    ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres,
456                  OCFS2_LOCK_TYPE_OPEN,
457                  0, inode);
458
459    if (can_lock) {
460        status = ocfs2_open_lock(inode);
461        if (status) {
462            make_bad_inode(inode);
463            mlog_errno(status);
464            return status;
465        }
466        status = ocfs2_inode_lock(inode, NULL, 0);
467        if (status) {
468            make_bad_inode(inode);
469            mlog_errno(status);
470            return status;
471        }
472    }
473
474    if (args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY) {
475        status = ocfs2_try_open_lock(inode, 0);
476        if (status) {
477            make_bad_inode(inode);
478            return status;
479        }
480    }
481
482    if (can_lock) {
483        status = ocfs2_read_inode_block_full(inode, &bh,
484                             OCFS2_BH_IGNORE_CACHE);
485    } else {
486        status = ocfs2_read_blocks_sync(osb, args->fi_blkno, 1, &bh);
487        if (!status)
488            status = ocfs2_validate_inode_block(osb->sb, bh);
489    }
490    if (status < 0) {
491        mlog_errno(status);
492        goto bail;
493    }
494
495    status = -EINVAL;
496    fe = (struct ocfs2_dinode *) bh->b_data;
497
498    /*
499     * This is a code bug. Right now the caller needs to
500     * understand whether it is asking for a system file inode or
501     * not so the proper lock names can be built.
502     */
503    mlog_bug_on_msg(!!(fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) !=
504            !!(args->fi_flags & OCFS2_FI_FLAG_SYSFILE),
505            "Inode %llu: system file state is ambigous\n",
506            (unsigned long long)args->fi_blkno);
507
508    if (S_ISCHR(le16_to_cpu(fe->i_mode)) ||
509        S_ISBLK(le16_to_cpu(fe->i_mode)))
510        inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
511
512    ocfs2_populate_inode(inode, fe, 0);
513
514    BUG_ON(args->fi_blkno != le64_to_cpu(fe->i_blkno));
515
516    status = 0;
517
518bail:
519    if (can_lock)
520        ocfs2_inode_unlock(inode, 0);
521
522    if (status < 0)
523        make_bad_inode(inode);
524
525    if (args && bh)
526        brelse(bh);
527
528    mlog_exit(status);
529    return status;
530}
531
532void ocfs2_sync_blockdev(struct super_block *sb)
533{
534    sync_blockdev(sb->s_bdev);
535}
536
537static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
538                     struct inode *inode,
539                     struct buffer_head *fe_bh)
540{
541    int status = 0;
542    struct ocfs2_truncate_context *tc = NULL;
543    struct ocfs2_dinode *fe;
544    handle_t *handle = NULL;
545
546    mlog_entry_void();
547
548    fe = (struct ocfs2_dinode *) fe_bh->b_data;
549
550    /*
551     * This check will also skip truncate of inodes with inline
552     * data and fast symlinks.
553     */
554    if (fe->i_clusters) {
555        if (ocfs2_should_order_data(inode))
556            ocfs2_begin_ordered_truncate(inode, 0);
557
558        handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
559        if (IS_ERR(handle)) {
560            status = PTR_ERR(handle);
561            mlog_errno(status);
562            goto out;
563        }
564
565        status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
566                         fe_bh,
567                         OCFS2_JOURNAL_ACCESS_WRITE);
568        if (status < 0) {
569            mlog_errno(status);
570            goto out;
571        }
572
573        i_size_write(inode, 0);
574
575        status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
576        if (status < 0) {
577            mlog_errno(status);
578            goto out;
579        }
580
581        ocfs2_commit_trans(osb, handle);
582        handle = NULL;
583
584        status = ocfs2_prepare_truncate(osb, inode, fe_bh, &tc);
585        if (status < 0) {
586            mlog_errno(status);
587            goto out;
588        }
589
590        status = ocfs2_commit_truncate(osb, inode, fe_bh, tc);
591        if (status < 0) {
592            mlog_errno(status);
593            goto out;
594        }
595    }
596
597out:
598    if (handle)
599        ocfs2_commit_trans(osb, handle);
600    mlog_exit(status);
601    return status;
602}
603
604static int ocfs2_remove_inode(struct inode *inode,
605                  struct buffer_head *di_bh,
606                  struct inode *orphan_dir_inode,
607                  struct buffer_head *orphan_dir_bh)
608{
609    int status;
610    struct inode *inode_alloc_inode = NULL;
611    struct buffer_head *inode_alloc_bh = NULL;
612    handle_t *handle;
613    struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
614    struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
615
616    inode_alloc_inode =
617        ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE,
618                        le16_to_cpu(di->i_suballoc_slot));
619    if (!inode_alloc_inode) {
620        status = -EEXIST;
621        mlog_errno(status);
622        goto bail;
623    }
624
625    mutex_lock(&inode_alloc_inode->i_mutex);
626    status = ocfs2_inode_lock(inode_alloc_inode, &inode_alloc_bh, 1);
627    if (status < 0) {
628        mutex_unlock(&inode_alloc_inode->i_mutex);
629
630        mlog_errno(status);
631        goto bail;
632    }
633
634    handle = ocfs2_start_trans(osb, OCFS2_DELETE_INODE_CREDITS +
635                   ocfs2_quota_trans_credits(inode->i_sb));
636    if (IS_ERR(handle)) {
637        status = PTR_ERR(handle);
638        mlog_errno(status);
639        goto bail_unlock;
640    }
641
642    status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
643                  orphan_dir_bh);
644    if (status < 0) {
645        mlog_errno(status);
646        goto bail_commit;
647    }
648
649    /* set the inodes dtime */
650    status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
651                     OCFS2_JOURNAL_ACCESS_WRITE);
652    if (status < 0) {
653        mlog_errno(status);
654        goto bail_commit;
655    }
656
657    di->i_dtime = cpu_to_le64(CURRENT_TIME.tv_sec);
658    di->i_flags &= cpu_to_le32(~(OCFS2_VALID_FL | OCFS2_ORPHANED_FL));
659
660    status = ocfs2_journal_dirty(handle, di_bh);
661    if (status < 0) {
662        mlog_errno(status);
663        goto bail_commit;
664    }
665
666    ocfs2_remove_from_cache(INODE_CACHE(inode), di_bh);
667    dquot_free_inode(inode);
668
669    status = ocfs2_free_dinode(handle, inode_alloc_inode,
670                   inode_alloc_bh, di);
671    if (status < 0)
672        mlog_errno(status);
673
674bail_commit:
675    ocfs2_commit_trans(osb, handle);
676bail_unlock:
677    ocfs2_inode_unlock(inode_alloc_inode, 1);
678    mutex_unlock(&inode_alloc_inode->i_mutex);
679    brelse(inode_alloc_bh);
680bail:
681    iput(inode_alloc_inode);
682
683    return status;
684}
685
686/*
687 * Serialize with orphan dir recovery. If the process doing
688 * recovery on this orphan dir does an iget() with the dir
689 * i_mutex held, we'll deadlock here. Instead we detect this
690 * and exit early - recovery will wipe this inode for us.
691 */
692static int ocfs2_check_orphan_recovery_state(struct ocfs2_super *osb,
693                         int slot)
694{
695    int ret = 0;
696
697    spin_lock(&osb->osb_lock);
698    if (ocfs2_node_map_test_bit(osb, &osb->osb_recovering_orphan_dirs, slot)) {
699        mlog(0, "Recovery is happening on orphan dir %d, will skip "
700             "this inode\n", slot);
701        ret = -EDEADLK;
702        goto out;
703    }
704    /* This signals to the orphan recovery process that it should
705     * wait for us to handle the wipe. */
706    osb->osb_orphan_wipes[slot]++;
707out:
708    spin_unlock(&osb->osb_lock);
709    return ret;
710}
711
712static void ocfs2_signal_wipe_completion(struct ocfs2_super *osb,
713                     int slot)
714{
715    spin_lock(&osb->osb_lock);
716    osb->osb_orphan_wipes[slot]--;
717    spin_unlock(&osb->osb_lock);
718
719    wake_up(&osb->osb_wipe_event);
720}
721
722static int ocfs2_wipe_inode(struct inode *inode,
723                struct buffer_head *di_bh)
724{
725    int status, orphaned_slot;
726    struct inode *orphan_dir_inode = NULL;
727    struct buffer_head *orphan_dir_bh = NULL;
728    struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
729    struct ocfs2_dinode *di;
730
731    di = (struct ocfs2_dinode *) di_bh->b_data;
732    orphaned_slot = le16_to_cpu(di->i_orphaned_slot);
733
734    status = ocfs2_check_orphan_recovery_state(osb, orphaned_slot);
735    if (status)
736        return status;
737
738    orphan_dir_inode = ocfs2_get_system_file_inode(osb,
739                               ORPHAN_DIR_SYSTEM_INODE,
740                               orphaned_slot);
741    if (!orphan_dir_inode) {
742        status = -EEXIST;
743        mlog_errno(status);
744        goto bail;
745    }
746
747    /* Lock the orphan dir. The lock will be held for the entire
748     * delete_inode operation. We do this now to avoid races with
749     * recovery completion on other nodes. */
750    mutex_lock(&orphan_dir_inode->i_mutex);
751    status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
752    if (status < 0) {
753        mutex_unlock(&orphan_dir_inode->i_mutex);
754
755        mlog_errno(status);
756        goto bail;
757    }
758
759    /* we do this while holding the orphan dir lock because we
760     * don't want recovery being run from another node to try an
761     * inode delete underneath us -- this will result in two nodes
762     * truncating the same file! */
763    status = ocfs2_truncate_for_delete(osb, inode, di_bh);
764    if (status < 0) {
765        mlog_errno(status);
766        goto bail_unlock_dir;
767    }
768
769    /* Remove any dir index tree */
770    if (S_ISDIR(inode->i_mode)) {
771        status = ocfs2_dx_dir_truncate(inode, di_bh);
772        if (status) {
773            mlog_errno(status);
774            goto bail_unlock_dir;
775        }
776    }
777
778    /*Free extended attribute resources associated with this inode.*/
779    status = ocfs2_xattr_remove(inode, di_bh);
780    if (status < 0) {
781        mlog_errno(status);
782        goto bail_unlock_dir;
783    }
784
785    status = ocfs2_remove_refcount_tree(inode, di_bh);
786    if (status < 0) {
787        mlog_errno(status);
788        goto bail_unlock_dir;
789    }
790
791    status = ocfs2_remove_inode(inode, di_bh, orphan_dir_inode,
792                    orphan_dir_bh);
793    if (status < 0)
794        mlog_errno(status);
795
796bail_unlock_dir:
797    ocfs2_inode_unlock(orphan_dir_inode, 1);
798    mutex_unlock(&orphan_dir_inode->i_mutex);
799    brelse(orphan_dir_bh);
800bail:
801    iput(orphan_dir_inode);
802    ocfs2_signal_wipe_completion(osb, orphaned_slot);
803
804    return status;
805}
806
807/* There is a series of simple checks that should be done before a
808 * trylock is even considered. Encapsulate those in this function. */
809static int ocfs2_inode_is_valid_to_delete(struct inode *inode)
810{
811    int ret = 0;
812    struct ocfs2_inode_info *oi = OCFS2_I(inode);
813    struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
814
815    /* We shouldn't be getting here for the root directory
816     * inode.. */
817    if (inode == osb->root_inode) {
818        mlog(ML_ERROR, "Skipping delete of root inode.\n");
819        goto bail;
820    }
821
822    /* If we're coming from downconvert_thread we can't go into our own
823     * voting [hello, deadlock city!], so unforuntately we just
824     * have to skip deleting this guy. That's OK though because
825     * the node who's doing the actual deleting should handle it
826     * anyway. */
827    if (current == osb->dc_task) {
828        mlog(0, "Skipping delete of %lu because we're currently "
829             "in downconvert\n", inode->i_ino);
830        goto bail;
831    }
832
833    spin_lock(&oi->ip_lock);
834    /* OCFS2 *never* deletes system files. This should technically
835     * never get here as system file inodes should always have a
836     * positive link count. */
837    if (oi->ip_flags & OCFS2_INODE_SYSTEM_FILE) {
838        mlog(ML_ERROR, "Skipping delete of system file %llu\n",
839             (unsigned long long)oi->ip_blkno);
840        goto bail_unlock;
841    }
842
843    /* If we have allowd wipe of this inode for another node, it
844     * will be marked here so we can safely skip it. Recovery will
845     * cleanup any inodes we might inadvertantly skip here. */
846    if (oi->ip_flags & OCFS2_INODE_SKIP_DELETE) {
847        mlog(0, "Skipping delete of %lu because another node "
848             "has done this for us.\n", inode->i_ino);
849        goto bail_unlock;
850    }
851
852    ret = 1;
853bail_unlock:
854    spin_unlock(&oi->ip_lock);
855bail:
856    return ret;
857}
858
859/* Query the cluster to determine whether we should wipe an inode from
860 * disk or not.
861 *
862 * Requires the inode to have the cluster lock. */
863static int ocfs2_query_inode_wipe(struct inode *inode,
864                  struct buffer_head *di_bh,
865                  int *wipe)
866{
867    int status = 0;
868    struct ocfs2_inode_info *oi = OCFS2_I(inode);
869    struct ocfs2_dinode *di;
870
871    *wipe = 0;
872
873    /* While we were waiting for the cluster lock in
874     * ocfs2_delete_inode, another node might have asked to delete
875     * the inode. Recheck our flags to catch this. */
876    if (!ocfs2_inode_is_valid_to_delete(inode)) {
877        mlog(0, "Skipping delete of %llu because flags changed\n",
878             (unsigned long long)oi->ip_blkno);
879        goto bail;
880    }
881
882    /* Now that we have an up to date inode, we can double check
883     * the link count. */
884    if (inode->i_nlink) {
885        mlog(0, "Skipping delete of %llu because nlink = %u\n",
886             (unsigned long long)oi->ip_blkno, inode->i_nlink);
887        goto bail;
888    }
889
890    /* Do some basic inode verification... */
891    di = (struct ocfs2_dinode *) di_bh->b_data;
892    if (!(di->i_flags & cpu_to_le32(OCFS2_ORPHANED_FL))) {
893        /*
894         * Inodes in the orphan dir must have ORPHANED_FL. The only
895         * inodes that come back out of the orphan dir are reflink
896         * targets. A reflink target may be moved out of the orphan
897         * dir between the time we scan the directory and the time we
898         * process it. This would lead to HAS_REFCOUNT_FL being set but
899         * ORPHANED_FL not.
900         */
901        if (di->i_dyn_features & cpu_to_le16(OCFS2_HAS_REFCOUNT_FL)) {
902            mlog(0, "Reflinked inode %llu is no longer orphaned. "
903                 "it shouldn't be deleted\n",
904                 (unsigned long long)oi->ip_blkno);
905            goto bail;
906        }
907
908        /* for lack of a better error? */
909        status = -EEXIST;
910        mlog(ML_ERROR,
911             "Inode %llu (on-disk %llu) not orphaned! "
912             "Disk flags 0x%x, inode flags 0x%x\n",
913             (unsigned long long)oi->ip_blkno,
914             (unsigned long long)le64_to_cpu(di->i_blkno),
915             le32_to_cpu(di->i_flags), oi->ip_flags);
916        goto bail;
917    }
918
919    /* has someone already deleted us?! baaad... */
920    if (di->i_dtime) {
921        status = -EEXIST;
922        mlog_errno(status);
923        goto bail;
924    }
925
926    /*
927     * This is how ocfs2 determines whether an inode is still live
928     * within the cluster. Every node takes a shared read lock on
929     * the inode open lock in ocfs2_read_locked_inode(). When we
930     * get to ->delete_inode(), each node tries to convert it's
931     * lock to an exclusive. Trylocks are serialized by the inode
932     * meta data lock. If the upconvert suceeds, we know the inode
933     * is no longer live and can be deleted.
934     *
935     * Though we call this with the meta data lock held, the
936     * trylock keeps us from ABBA deadlock.
937     */
938    status = ocfs2_try_open_lock(inode, 1);
939    if (status == -EAGAIN) {
940        status = 0;
941        mlog(0, "Skipping delete of %llu because it is in use on "
942             "other nodes\n", (unsigned long long)oi->ip_blkno);
943        goto bail;
944    }
945    if (status < 0) {
946        mlog_errno(status);
947        goto bail;
948    }
949
950    *wipe = 1;
951    mlog(0, "Inode %llu is ok to wipe from orphan dir %u\n",
952         (unsigned long long)oi->ip_blkno,
953         le16_to_cpu(di->i_orphaned_slot));
954
955bail:
956    return status;
957}
958
959/* Support function for ocfs2_delete_inode. Will help us keep the
960 * inode data in a consistent state for clear_inode. Always truncates
961 * pages, optionally sync's them first. */
962static void ocfs2_cleanup_delete_inode(struct inode *inode,
963                       int sync_data)
964{
965    mlog(0, "Cleanup inode %llu, sync = %d\n",
966         (unsigned long long)OCFS2_I(inode)->ip_blkno, sync_data);
967    if (sync_data)
968        write_inode_now(inode, 1);
969    truncate_inode_pages(&inode->i_data, 0);
970}
971
972void ocfs2_delete_inode(struct inode *inode)
973{
974    int wipe, status;
975    sigset_t blocked, oldset;
976    struct buffer_head *di_bh = NULL;
977
978    mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
979
980    /* When we fail in read_inode() we mark inode as bad. The second test
981     * catches the case when inode allocation fails before allocating
982     * a block for inode. */
983    if (is_bad_inode(inode) || !OCFS2_I(inode)->ip_blkno) {
984        mlog(0, "Skipping delete of bad inode\n");
985        goto bail;
986    }
987
988    dquot_initialize(inode);
989
990    if (!ocfs2_inode_is_valid_to_delete(inode)) {
991        /* It's probably not necessary to truncate_inode_pages
992         * here but we do it for safety anyway (it will most
993         * likely be a no-op anyway) */
994        ocfs2_cleanup_delete_inode(inode, 0);
995        goto bail;
996    }
997
998    /* We want to block signals in delete_inode as the lock and
999     * messaging paths may return us -ERESTARTSYS. Which would
1000     * cause us to exit early, resulting in inodes being orphaned
1001     * forever. */
1002    sigfillset(&blocked);
1003    status = sigprocmask(SIG_BLOCK, &blocked, &oldset);
1004    if (status < 0) {
1005        mlog_errno(status);
1006        ocfs2_cleanup_delete_inode(inode, 1);
1007        goto bail;
1008    }
1009
1010    /*
1011     * Synchronize us against ocfs2_get_dentry. We take this in
1012     * shared mode so that all nodes can still concurrently
1013     * process deletes.
1014     */
1015    status = ocfs2_nfs_sync_lock(OCFS2_SB(inode->i_sb), 0);
1016    if (status < 0) {
1017        mlog(ML_ERROR, "getting nfs sync lock(PR) failed %d\n", status);
1018        ocfs2_cleanup_delete_inode(inode, 0);
1019        goto bail_unblock;
1020    }
1021    /* Lock down the inode. This gives us an up to date view of
1022     * it's metadata (for verification), and allows us to
1023     * serialize delete_inode on multiple nodes.
1024     *
1025     * Even though we might be doing a truncate, we don't take the
1026     * allocation lock here as it won't be needed - nobody will
1027     * have the file open.
1028     */
1029    status = ocfs2_inode_lock(inode, &di_bh, 1);
1030    if (status < 0) {
1031        if (status != -ENOENT)
1032            mlog_errno(status);
1033        ocfs2_cleanup_delete_inode(inode, 0);
1034        goto bail_unlock_nfs_sync;
1035    }
1036
1037    /* Query the cluster. This will be the final decision made
1038     * before we go ahead and wipe the inode. */
1039    status = ocfs2_query_inode_wipe(inode, di_bh, &wipe);
1040    if (!wipe || status < 0) {
1041        /* Error and remote inode busy both mean we won't be
1042         * removing the inode, so they take almost the same
1043         * path. */
1044        if (status < 0)
1045            mlog_errno(status);
1046
1047        /* Someone in the cluster has disallowed a wipe of
1048         * this inode, or it was never completely
1049         * orphaned. Write out the pages and exit now. */
1050        ocfs2_cleanup_delete_inode(inode, 1);
1051        goto bail_unlock_inode;
1052    }
1053
1054    ocfs2_cleanup_delete_inode(inode, 0);
1055
1056    status = ocfs2_wipe_inode(inode, di_bh);
1057    if (status < 0) {
1058        if (status != -EDEADLK)
1059            mlog_errno(status);
1060        goto bail_unlock_inode;
1061    }
1062
1063    /*
1064     * Mark the inode as successfully deleted.
1065     *
1066     * This is important for ocfs2_clear_inode() as it will check
1067     * this flag and skip any checkpointing work
1068     *
1069     * ocfs2_stuff_meta_lvb() also uses this flag to invalidate
1070     * the LVB for other nodes.
1071     */
1072    OCFS2_I(inode)->ip_flags |= OCFS2_INODE_DELETED;
1073
1074bail_unlock_inode:
1075    ocfs2_inode_unlock(inode, 1);
1076    brelse(di_bh);
1077
1078bail_unlock_nfs_sync:
1079    ocfs2_nfs_sync_unlock(OCFS2_SB(inode->i_sb), 0);
1080
1081bail_unblock:
1082    status = sigprocmask(SIG_SETMASK, &oldset, NULL);
1083    if (status < 0)
1084        mlog_errno(status);
1085bail:
1086    clear_inode(inode);
1087    mlog_exit_void();
1088}
1089
1090void ocfs2_clear_inode(struct inode *inode)
1091{
1092    int status;
1093    struct ocfs2_inode_info *oi = OCFS2_I(inode);
1094
1095    mlog_entry_void();
1096
1097    if (!inode)
1098        goto bail;
1099
1100    mlog(0, "Clearing inode: %llu, nlink = %u\n",
1101         (unsigned long long)OCFS2_I(inode)->ip_blkno, inode->i_nlink);
1102
1103    mlog_bug_on_msg(OCFS2_SB(inode->i_sb) == NULL,
1104            "Inode=%lu\n", inode->i_ino);
1105
1106    dquot_drop(inode);
1107
1108    /* To preven remote deletes we hold open lock before, now it
1109     * is time to unlock PR and EX open locks. */
1110    ocfs2_open_unlock(inode);
1111
1112    /* Do these before all the other work so that we don't bounce
1113     * the downconvert thread while waiting to destroy the locks. */
1114    ocfs2_mark_lockres_freeing(&oi->ip_rw_lockres);
1115    ocfs2_mark_lockres_freeing(&oi->ip_inode_lockres);
1116    ocfs2_mark_lockres_freeing(&oi->ip_open_lockres);
1117
1118    /* We very well may get a clear_inode before all an inodes
1119     * metadata has hit disk. Of course, we can't drop any cluster
1120     * locks until the journal has finished with it. The only
1121     * exception here are successfully wiped inodes - their
1122     * metadata can now be considered to be part of the system
1123     * inodes from which it came. */
1124    if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED))
1125        ocfs2_checkpoint_inode(inode);
1126
1127    mlog_bug_on_msg(!list_empty(&oi->ip_io_markers),
1128            "Clear inode of %llu, inode has io markers\n",
1129            (unsigned long long)oi->ip_blkno);
1130
1131    ocfs2_extent_map_trunc(inode, 0);
1132
1133    status = ocfs2_drop_inode_locks(inode);
1134    if (status < 0)
1135        mlog_errno(status);
1136
1137    ocfs2_lock_res_free(&oi->ip_rw_lockres);
1138    ocfs2_lock_res_free(&oi->ip_inode_lockres);
1139    ocfs2_lock_res_free(&oi->ip_open_lockres);
1140
1141    ocfs2_metadata_cache_exit(INODE_CACHE(inode));
1142
1143    mlog_bug_on_msg(INODE_CACHE(inode)->ci_num_cached,
1144            "Clear inode of %llu, inode has %u cache items\n",
1145            (unsigned long long)oi->ip_blkno,
1146            INODE_CACHE(inode)->ci_num_cached);
1147
1148    mlog_bug_on_msg(!(INODE_CACHE(inode)->ci_flags & OCFS2_CACHE_FL_INLINE),
1149            "Clear inode of %llu, inode has a bad flag\n",
1150            (unsigned long long)oi->ip_blkno);
1151
1152    mlog_bug_on_msg(spin_is_locked(&oi->ip_lock),
1153            "Clear inode of %llu, inode is locked\n",
1154            (unsigned long long)oi->ip_blkno);
1155
1156    mlog_bug_on_msg(!mutex_trylock(&oi->ip_io_mutex),
1157            "Clear inode of %llu, io_mutex is locked\n",
1158            (unsigned long long)oi->ip_blkno);
1159    mutex_unlock(&oi->ip_io_mutex);
1160
1161    /*
1162     * down_trylock() returns 0, down_write_trylock() returns 1
1163     * kernel 1, world 0
1164     */
1165    mlog_bug_on_msg(!down_write_trylock(&oi->ip_alloc_sem),
1166            "Clear inode of %llu, alloc_sem is locked\n",
1167            (unsigned long long)oi->ip_blkno);
1168    up_write(&oi->ip_alloc_sem);
1169
1170    mlog_bug_on_msg(oi->ip_open_count,
1171            "Clear inode of %llu has open count %d\n",
1172            (unsigned long long)oi->ip_blkno, oi->ip_open_count);
1173
1174    /* Clear all other flags. */
1175    oi->ip_flags = 0;
1176    oi->ip_dir_start_lookup = 0;
1177    oi->ip_blkno = 0ULL;
1178
1179    /*
1180     * ip_jinode is used to track txns against this inode. We ensure that
1181     * the journal is flushed before journal shutdown. Thus it is safe to
1182     * have inodes get cleaned up after journal shutdown.
1183     */
1184    jbd2_journal_release_jbd_inode(OCFS2_SB(inode->i_sb)->journal->j_journal,
1185                       &oi->ip_jinode);
1186
1187bail:
1188    mlog_exit_void();
1189}
1190
1191/* Called under inode_lock, with no more references on the
1192 * struct inode, so it's safe here to check the flags field
1193 * and to manipulate i_nlink without any other locks. */
1194void ocfs2_drop_inode(struct inode *inode)
1195{
1196    struct ocfs2_inode_info *oi = OCFS2_I(inode);
1197
1198    mlog_entry_void();
1199
1200    mlog(0, "Drop inode %llu, nlink = %u, ip_flags = 0x%x\n",
1201         (unsigned long long)oi->ip_blkno, inode->i_nlink, oi->ip_flags);
1202
1203    if (oi->ip_flags & OCFS2_INODE_MAYBE_ORPHANED)
1204        generic_delete_inode(inode);
1205    else
1206        generic_drop_inode(inode);
1207
1208    mlog_exit_void();
1209}
1210
1211/*
1212 * This is called from our getattr.
1213 */
1214int ocfs2_inode_revalidate(struct dentry *dentry)
1215{
1216    struct inode *inode = dentry->d_inode;
1217    int status = 0;
1218
1219    mlog_entry("(inode = 0x%p, ino = %llu)\n", inode,
1220           inode ? (unsigned long long)OCFS2_I(inode)->ip_blkno : 0ULL);
1221
1222    if (!inode) {
1223        mlog(0, "eep, no inode!\n");
1224        status = -ENOENT;
1225        goto bail;
1226    }
1227
1228    spin_lock(&OCFS2_I(inode)->ip_lock);
1229    if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
1230        spin_unlock(&OCFS2_I(inode)->ip_lock);
1231        mlog(0, "inode deleted!\n");
1232        status = -ENOENT;
1233        goto bail;
1234    }
1235    spin_unlock(&OCFS2_I(inode)->ip_lock);
1236
1237    /* Let ocfs2_inode_lock do the work of updating our struct
1238     * inode for us. */
1239    status = ocfs2_inode_lock(inode, NULL, 0);
1240    if (status < 0) {
1241        if (status != -ENOENT)
1242            mlog_errno(status);
1243        goto bail;
1244    }
1245    ocfs2_inode_unlock(inode, 0);
1246bail:
1247    mlog_exit(status);
1248
1249    return status;
1250}
1251
1252/*
1253 * Updates a disk inode from a
1254 * struct inode.
1255 * Only takes ip_lock.
1256 */
1257int ocfs2_mark_inode_dirty(handle_t *handle,
1258               struct inode *inode,
1259               struct buffer_head *bh)
1260{
1261    int status;
1262    struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
1263
1264    mlog_entry("(inode %llu)\n",
1265           (unsigned long long)OCFS2_I(inode)->ip_blkno);
1266
1267    status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
1268                     OCFS2_JOURNAL_ACCESS_WRITE);
1269    if (status < 0) {
1270        mlog_errno(status);
1271        goto leave;
1272    }
1273
1274    spin_lock(&OCFS2_I(inode)->ip_lock);
1275    fe->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters);
1276    ocfs2_get_inode_flags(OCFS2_I(inode));
1277    fe->i_attr = cpu_to_le32(OCFS2_I(inode)->ip_attr);
1278    fe->i_dyn_features = cpu_to_le16(OCFS2_I(inode)->ip_dyn_features);
1279    spin_unlock(&OCFS2_I(inode)->ip_lock);
1280
1281    fe->i_size = cpu_to_le64(i_size_read(inode));
1282    ocfs2_set_links_count(fe, inode->i_nlink);
1283    fe->i_uid = cpu_to_le32(inode->i_uid);
1284    fe->i_gid = cpu_to_le32(inode->i_gid);
1285    fe->i_mode = cpu_to_le16(inode->i_mode);
1286    fe->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
1287    fe->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
1288    fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
1289    fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
1290    fe->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
1291    fe->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
1292
1293    status = ocfs2_journal_dirty(handle, bh);
1294    if (status < 0)
1295        mlog_errno(status);
1296
1297    status = 0;
1298leave:
1299
1300    mlog_exit(status);
1301    return status;
1302}
1303
1304/*
1305 *
1306 * Updates a struct inode from a disk inode.
1307 * does no i/o, only takes ip_lock.
1308 */
1309void ocfs2_refresh_inode(struct inode *inode,
1310             struct ocfs2_dinode *fe)
1311{
1312    spin_lock(&OCFS2_I(inode)->ip_lock);
1313
1314    OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
1315    OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
1316    OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features);
1317    ocfs2_set_inode_flags(inode);
1318    i_size_write(inode, le64_to_cpu(fe->i_size));
1319    inode->i_nlink = ocfs2_read_links_count(fe);
1320    inode->i_uid = le32_to_cpu(fe->i_uid);
1321    inode->i_gid = le32_to_cpu(fe->i_gid);
1322    inode->i_mode = le16_to_cpu(fe->i_mode);
1323    if (S_ISLNK(inode->i_mode) && le32_to_cpu(fe->i_clusters) == 0)
1324        inode->i_blocks = 0;
1325    else
1326        inode->i_blocks = ocfs2_inode_sector_count(inode);
1327    inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
1328    inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
1329    inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
1330    inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
1331    inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
1332    inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
1333
1334    spin_unlock(&OCFS2_I(inode)->ip_lock);
1335}
1336
1337int ocfs2_validate_inode_block(struct super_block *sb,
1338                   struct buffer_head *bh)
1339{
1340    int rc;
1341    struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
1342
1343    mlog(0, "Validating dinode %llu\n",
1344         (unsigned long long)bh->b_blocknr);
1345
1346    BUG_ON(!buffer_uptodate(bh));
1347
1348    /*
1349     * If the ecc fails, we return the error but otherwise
1350     * leave the filesystem running. We know any error is
1351     * local to this block.
1352     */
1353    rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &di->i_check);
1354    if (rc) {
1355        mlog(ML_ERROR, "Checksum failed for dinode %llu\n",
1356             (unsigned long long)bh->b_blocknr);
1357        goto bail;
1358    }
1359
1360    /*
1361     * Errors after here are fatal.
1362     */
1363
1364    rc = -EINVAL;
1365
1366    if (!OCFS2_IS_VALID_DINODE(di)) {
1367        ocfs2_error(sb, "Invalid dinode #%llu: signature = %.*s\n",
1368                (unsigned long long)bh->b_blocknr, 7,
1369                di->i_signature);
1370        goto bail;
1371    }
1372
1373    if (le64_to_cpu(di->i_blkno) != bh->b_blocknr) {
1374        ocfs2_error(sb, "Invalid dinode #%llu: i_blkno is %llu\n",
1375                (unsigned long long)bh->b_blocknr,
1376                (unsigned long long)le64_to_cpu(di->i_blkno));
1377        goto bail;
1378    }
1379
1380    if (!(di->i_flags & cpu_to_le32(OCFS2_VALID_FL))) {
1381        ocfs2_error(sb,
1382                "Invalid dinode #%llu: OCFS2_VALID_FL not set\n",
1383                (unsigned long long)bh->b_blocknr);
1384        goto bail;
1385    }
1386
1387    if (le32_to_cpu(di->i_fs_generation) !=
1388        OCFS2_SB(sb)->fs_generation) {
1389        ocfs2_error(sb,
1390                "Invalid dinode #%llu: fs_generation is %u\n",
1391                (unsigned long long)bh->b_blocknr,
1392                le32_to_cpu(di->i_fs_generation));
1393        goto bail;
1394    }
1395
1396    rc = 0;
1397
1398bail:
1399    return rc;
1400}
1401
1402int ocfs2_read_inode_block_full(struct inode *inode, struct buffer_head **bh,
1403                int flags)
1404{
1405    int rc;
1406    struct buffer_head *tmp = *bh;
1407
1408    rc = ocfs2_read_blocks(INODE_CACHE(inode), OCFS2_I(inode)->ip_blkno,
1409                   1, &tmp, flags, ocfs2_validate_inode_block);
1410
1411    /* If ocfs2_read_blocks() got us a new bh, pass it up. */
1412    if (!rc && !*bh)
1413        *bh = tmp;
1414
1415    return rc;
1416}
1417
1418int ocfs2_read_inode_block(struct inode *inode, struct buffer_head **bh)
1419{
1420    return ocfs2_read_inode_block_full(inode, bh, 0);
1421}
1422
1423
1424static u64 ocfs2_inode_cache_owner(struct ocfs2_caching_info *ci)
1425{
1426    struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1427
1428    return oi->ip_blkno;
1429}
1430
1431static struct super_block *ocfs2_inode_cache_get_super(struct ocfs2_caching_info *ci)
1432{
1433    struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1434
1435    return oi->vfs_inode.i_sb;
1436}
1437
1438static void ocfs2_inode_cache_lock(struct ocfs2_caching_info *ci)
1439{
1440    struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1441
1442    spin_lock(&oi->ip_lock);
1443}
1444
1445static void ocfs2_inode_cache_unlock(struct ocfs2_caching_info *ci)
1446{
1447    struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1448
1449    spin_unlock(&oi->ip_lock);
1450}
1451
1452static void ocfs2_inode_cache_io_lock(struct ocfs2_caching_info *ci)
1453{
1454    struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1455
1456    mutex_lock(&oi->ip_io_mutex);
1457}
1458
1459static void ocfs2_inode_cache_io_unlock(struct ocfs2_caching_info *ci)
1460{
1461    struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
1462
1463    mutex_unlock(&oi->ip_io_mutex);
1464}
1465
1466const struct ocfs2_caching_operations ocfs2_inode_caching_ops = {
1467    .co_owner = ocfs2_inode_cache_owner,
1468    .co_get_super = ocfs2_inode_cache_get_super,
1469    .co_cache_lock = ocfs2_inode_cache_lock,
1470    .co_cache_unlock = ocfs2_inode_cache_unlock,
1471    .co_io_lock = ocfs2_inode_cache_io_lock,
1472    .co_io_unlock = ocfs2_inode_cache_io_unlock,
1473};
1474
1475

Archive Download this file



interactive