Whamcloud - gitweb
ab6e403be0b241f450c46259cd3b9f6d167b0c8d
[fs/lustre-release.git] / lustre / lvfs / fsfilt_ext3.c
1 /*
2  * GPL HEADER START
3  *
4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License version 2 only,
8  * as published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope that it will be useful, but
11  * WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
13  * General Public License version 2 for more details (a copy is included
14  * in the LICENSE file that accompanied this code).
15  *
16  * You should have received a copy of the GNU General Public License
17  * version 2 along with this program; If not, see
18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
19  *
20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21  * CA 95054 USA or visit www.sun.com if you need additional information or
22  * have any questions.
23  *
24  * GPL HEADER END
25  */
26 /*
27  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
28  * Use is subject to license terms.
29  *
30  * Copyright (c) 2011, 2012, Whamcloud, Inc.
31  */
32 /*
33  * This file is part of Lustre, http://www.lustre.org/
34  * Lustre is a trademark of Sun Microsystems, Inc.
35  *
36  * lustre/lvfs/fsfilt_ext3.c
37  *
38  * Author: Andreas Dilger <adilger@clusterfs.com>
39  */
40
41 #define DEBUG_SUBSYSTEM S_FILTER
42
43 #include <linux/init.h>
44 #include <linux/module.h>
45 #include <linux/fs.h>
46 #include <linux/slab.h>
47 #include <linux/pagemap.h>
48 #include <linux/quotaops.h>
49 #ifdef HAVE_LINUX_EXPORTFS_H
50 #include <linux/exportfs.h>
51 #endif
52 #include <ext4/ext4.h>
53 #include <ext4/ext4_jbd2.h>
54 #include <linux/version.h>
55 #include <linux/bitops.h>
56 #include <linux/quota.h>
57 #ifdef HAVE_QUOTAIO_H
58 # include <linux/quotaio_v2.h>
59 #elif defined(HAVE_FS_QUOTA_QUOTAIO_H)
60 # include <quota/quotaio_v2.h>
61 # include <quota/quota_tree.h>
62 # define V2_DQTREEOFF    QT_TREEOFF
63 #elif defined(HAVE_FS_QUOTAIO_H)
64 # include <quotaio_v2.h>
65 # include <quota_tree.h>
66 # define V2_DQTREEOFF    QT_TREEOFF
67 # define V2_INITQVERSIONS_R1 V2_INITQVERSIONS
68 #endif
69
70 #ifdef QFMT_VFS_V1
71 #define QFMT_LUSTRE QFMT_VFS_V1
72 #else
73 #define QFMT_LUSTRE QFMT_VFS_V0
74 #endif
75
76 #if defined(HAVE_EXT3_XATTR_H)
77 # include <ext3/xattr.h>
78 #elif !defined(EXT3_XATTR_INDEX_TRUSTED)
79 /* ext3 xattr.h not available in rh style kernel-devel rpm */
80 /* CHAOS kernel-devel package will not include fs/ldiskfs/xattr.h */
81 # define EXT3_XATTR_INDEX_TRUSTED        4
82 extern int ext3_xattr_get(struct inode *, int, const char *, void *, size_t);
83 extern int ext3_xattr_set_handle(handle_t *, struct inode *, int, const char *, const void *, size_t, int);
84 #endif
85
86 #include <libcfs/libcfs.h>
87 #include <lustre_fsfilt.h>
88 #include <obd.h>
89 #include <lustre_quota.h>
90 #include <linux/lustre_compat25.h>
91 #include <linux/lprocfs_status.h>
92
93 #include <ext4/ext4_extents.h>
94
95 #include "lustre_quota_fmt.h"
96
97 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,15)
98 #define FSFILT_DATA_TRANS_BLOCKS(sb)      EXT3_DATA_TRANS_BLOCKS
99 #define FSFILT_DELETE_TRANS_BLOCKS(sb)    EXT3_DELETE_TRANS_BLOCKS
100 #else
101 #define FSFILT_DATA_TRANS_BLOCKS(sb)      EXT3_DATA_TRANS_BLOCKS(sb)
102 #define FSFILT_DELETE_TRANS_BLOCKS(sb)    EXT3_DELETE_TRANS_BLOCKS(sb)
103 #endif
104
105 /* for kernels 2.6.18 and later */
106 #define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS(sb)
107
108 #define fsfilt_ext3_ext_insert_extent(handle, inode, path, newext, flag) \
109                ext3_ext_insert_extent(handle, inode, path, newext, flag)
110
111 #define ext3_mb_discard_inode_preallocations(inode) \
112                  ext3_discard_preallocations(inode)
113
114 #define fsfilt_log_start_commit(journal, tid) jbd2_log_start_commit(journal, tid)
115 #define fsfilt_log_wait_commit(journal, tid) jbd2_log_wait_commit(journal, tid)
116
117 #ifdef HAVE_EXT4_JOURNAL_CALLBACK_ADD
118 # define journal_callback ext4_journal_cb_entry
119 # define fsfilt_journal_callback_set(handle, func, jcb) \
120          ext4_journal_callback_add(handle, func, jcb)
121 #elif defined(HAVE_JBD2_JOURNAL_CALLBACK_SET)
122 # define fsfilt_journal_callback_set(handle, func, jcb) \
123          jbd2_journal_callback_set(handle, func, jcb)
124 #elif defined(HAVE_JOURNAL_CALLBACK_SET)
125 # define fsfilt_journal_callback_set(handle, func, jcb) \
126          journal_callback_set(handle, func, jcb)
127 #else
128 # error missing journal commit callback
129 #endif /* HAVE_EXT4_JOURNAL_CALLBACK_ADD */
130
131 static cfs_mem_cache_t *fcb_cache;
132
133 struct fsfilt_cb_data {
134         struct journal_callback cb_jcb; /* jbd private data - MUST BE FIRST */
135         fsfilt_cb_t cb_func;            /* MDS/OBD completion function */
136         struct obd_device *cb_obd;      /* MDS/OBD completion device */
137         __u64 cb_last_rcvd;             /* MDS/OST last committed operation */
138         void *cb_data;                  /* MDS/OST completion function data */
139 };
140
141 #ifndef ext3_find_next_bit
142 #define ext3_find_next_bit           ext2_find_next_bit
143 #endif
144
145 #ifndef ext2_find_next_bit
146 #ifdef __LITTLE_ENDIAN
147 #define ext2_find_next_bit(addr, size, off) find_next_bit((unsigned long *)(addr), (size), (off))
148 #else
149 error "Need implementation of find_next_bit on big-endian systems"
150 #endif  /* __LITTLE_ENDIAN */
151 #endif  /* !ext2_find_next_le_bit */
152
153 static char *fsfilt_ext3_get_label(struct super_block *sb)
154 {
155         return EXT3_SB(sb)->s_es->s_volume_name;
156 }
157
158 static int fsfilt_ext3_set_label(struct super_block *sb, char *label)
159 {
160         /* see e.g. fsfilt_ext3_write_record() */
161         journal_t *journal;
162         handle_t *handle;
163         int err;
164
165         journal = EXT3_SB(sb)->s_journal;
166         handle = ext3_journal_start_sb(sb, 1);
167         if (IS_ERR(handle)) {
168                 CERROR("can't start transaction\n");
169                 return(PTR_ERR(handle));
170         }
171
172         err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
173         if (err)
174                 goto out;
175
176         memcpy(EXT3_SB(sb)->s_es->s_volume_name, label,
177                sizeof(EXT3_SB(sb)->s_es->s_volume_name));
178
179         err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
180
181 out:
182         ext3_journal_stop(handle);
183
184         return(err);
185 }
186
187 static char *fsfilt_ext3_uuid(struct super_block *sb)
188 {
189         return EXT3_SB(sb)->s_es->s_uuid;
190 }
191
192 #ifdef HAVE_DISK_INODE_VERSION
193
194 static __u64 get_i_version(struct inode *inode)
195 {
196         return EXT3_I(inode)->i_fs_version;
197 }
198
199 static void set_i_version(struct inode *inode, __u64 new_version)
200 {
201         (EXT3_I(inode))->i_fs_version = new_version;
202 }
203
204 /*
205  * Get the 64-bit version for an inode.
206  */
207 static __u64 fsfilt_ext3_get_version(struct inode *inode)
208 {
209         CDEBUG(D_INFO, "Get version "LPX64" for inode %lu\n",
210                get_i_version(inode), inode->i_ino);
211         return get_i_version(inode);
212 }
213
214 /*
215  * Set the 64-bit version and return the old version.
216  */
217 static __u64 fsfilt_ext3_set_version(struct inode *inode, __u64 new_version)
218 {
219         __u64 old_version = get_i_version(inode);
220
221         CDEBUG(D_INFO, "Set version "LPX64" (old "LPX64") for inode %lu\n",
222                new_version, old_version, inode->i_ino);
223         set_i_version(inode, new_version);
224         /* version is set after all inode operations are finished, so we should
225          * mark it dirty here */
226         inode->i_sb->s_op->dirty_inode(inode);
227         return old_version;
228 }
229
230 #endif
231
232 /*
233  * We don't currently need any additional blocks for rmdir and
234  * unlink transactions because we are storing the OST oa_id inside
235  * the inode (which we will be changing anyways as part of this
236  * transaction).
237  */
238 static void *fsfilt_ext3_start(struct inode *inode, int op, void *desc_private,
239                                int logs)
240 {
241         /* For updates to the last received file */
242         int nblocks = FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
243         journal_t *journal;
244         void *handle;
245
246         if (current->journal_info) {
247                 CDEBUG(D_INODE, "increasing refcount on %p\n",
248                        current->journal_info);
249                 goto journal_start;
250         }
251
252         switch(op) {
253         case FSFILT_OP_RMDIR:
254         case FSFILT_OP_UNLINK:
255                 /* delete one file + create/update logs for each stripe */
256                 nblocks += FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb);
257                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
258                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
259                 break;
260         case FSFILT_OP_RENAME:
261                 /* modify additional directory */
262                 nblocks += FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
263                 /* no break */
264         case FSFILT_OP_SYMLINK:
265                 /* additional block + block bitmap + GDT for long symlink */
266                 nblocks += 3;
267                 /* no break */
268         case FSFILT_OP_CREATE: {
269                 /* no break */
270         }
271         case FSFILT_OP_MKDIR:
272         case FSFILT_OP_MKNOD:
273                 /* modify one inode + block bitmap + GDT */
274                 nblocks += 3;
275                 /* no break */
276         case FSFILT_OP_LINK:
277                 /* modify parent directory */
278                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
279                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
280                 /* create/update logs for each stripe */
281                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
282                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
283                 break;
284         case FSFILT_OP_SETATTR:
285                 /* Setattr on inode */
286                 nblocks += 1;
287                 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
288                          FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
289                 /* quota chown log for each stripe */
290                 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
291                             FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
292                 break;
293         case FSFILT_OP_CANCEL_UNLINK:
294                 /* blocks for log header bitmap update OR
295                  * blocks for catalog header bitmap update + unlink of logs */
296                 nblocks = (LLOG_CHUNK_SIZE >> inode->i_blkbits) +
297                         FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb) * logs;
298                 break;
299         default: CERROR("unknown transaction start op %d\n", op);
300                 LBUG();
301         }
302
303         LASSERT(current->journal_info == desc_private);
304         journal = EXT3_SB(inode->i_sb)->s_journal;
305         if (nblocks > journal->j_max_transaction_buffers) {
306                 CWARN("too many credits %d for op %ux%u using %d instead\n",
307                        nblocks, op, logs, journal->j_max_transaction_buffers);
308                 nblocks = journal->j_max_transaction_buffers;
309         }
310
311  journal_start:
312         LASSERTF(nblocks > 0, "can't start %d credit transaction\n", nblocks);
313         handle = ext3_journal_start(inode, nblocks);
314
315         if (!IS_ERR(handle))
316                 LASSERT(current->journal_info == handle);
317         else
318                 CERROR("error starting handle for op %u (%u credits): rc %ld\n",
319                        op, nblocks, PTR_ERR(handle));
320         return handle;
321 }
322
323 /*
324  * Calculate the number of buffer credits needed to write multiple pages in
325  * a single ext3 transaction.  No, this shouldn't be here, but as yet ext3
326  * doesn't have a nice API for calculating this sort of thing in advance.
327  *
328  * See comment above ext3_writepage_trans_blocks for details.  We assume
329  * no data journaling is being done, but it does allow for all of the pages
330  * being non-contiguous.  If we are guaranteed contiguous pages we could
331  * reduce the number of (d)indirect blocks a lot.
332  *
333  * With N blocks per page and P pages, for each inode we have at most:
334  * N*P indirect
335  * min(N*P, blocksize/4 + 1) dindirect blocks
336  * niocount tindirect
337  *
338  * For the entire filesystem, we have at most:
339  * min(sum(nindir + P), ngroups) bitmap blocks (from the above)
340  * min(sum(nindir + P), gdblocks) group descriptor blocks (from the above)
341  * objcount inode blocks
342  * 1 superblock
343  * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quota files
344  *
345  * 1 EXT3_DATA_TRANS_BLOCKS for the last_rcvd update.
346  */
347 static int fsfilt_ext3_credits_needed(int objcount, struct fsfilt_objinfo *fso,
348                                       int niocount, struct niobuf_local *nb)
349 {
350         struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
351         __u64 next_indir;
352         const int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
353         int nbitmaps = 0, ngdblocks;
354         int needed = objcount + 1; /* inodes + superblock */
355         int i, j;
356
357         for (i = 0, j = 0; i < objcount; i++, fso++) {
358                 /* two or more dindirect blocks in case we cross boundary */
359                 int ndind = (long)((nb[j + fso->fso_bufcnt - 1].offset -
360                                     nb[j].offset) >>
361                                    sb->s_blocksize_bits) /
362                         (EXT3_ADDR_PER_BLOCK(sb) * EXT3_ADDR_PER_BLOCK(sb));
363                 nbitmaps += min(fso->fso_bufcnt, ndind > 0 ? ndind : 2);
364
365                 /* leaf, indirect, tindirect blocks for first block */
366                 nbitmaps += blockpp + 2;
367
368                 j += fso->fso_bufcnt;
369         }
370
371         next_indir = nb[0].offset +
372                 (EXT3_ADDR_PER_BLOCK(sb) << sb->s_blocksize_bits);
373         for (i = 1; i < niocount; i++) {
374                 if (nb[i].offset >= next_indir) {
375                         nbitmaps++;     /* additional indirect */
376                         next_indir = nb[i].offset +
377                                 (EXT3_ADDR_PER_BLOCK(sb)<<sb->s_blocksize_bits);
378                 } else if (nb[i].offset != nb[i - 1].offset + sb->s_blocksize) {
379                         nbitmaps++;     /* additional indirect */
380                 }
381                 nbitmaps += blockpp;    /* each leaf in different group? */
382         }
383
384         ngdblocks = nbitmaps;
385         if (nbitmaps > EXT3_SB(sb)->s_groups_count)
386                 nbitmaps = EXT3_SB(sb)->s_groups_count;
387         if (ngdblocks > EXT3_SB(sb)->s_gdb_count)
388                 ngdblocks = EXT3_SB(sb)->s_gdb_count;
389
390         needed += nbitmaps + ngdblocks;
391
392         /* last_rcvd update */
393         needed += FSFILT_DATA_TRANS_BLOCKS(sb);
394
395 #if defined(CONFIG_QUOTA)
396         /* We assume that there will be 1 bit set in s_dquot.flags for each
397          * quota file that is active.  This is at least true for now.
398          */
399         needed += hweight32(ll_sb_any_quota_active(sb)) *
400                 FSFILT_SINGLEDATA_TRANS_BLOCKS(sb);
401 #endif
402
403         return needed;
404 }
405
406 /* We have to start a huge journal transaction here to hold all of the
407  * metadata for the pages being written here.  This is necessitated by
408  * the fact that we do lots of prepare_write operations before we do
409  * any of the matching commit_write operations, so even if we split
410  * up to use "smaller" transactions none of them could complete until
411  * all of them were opened.  By having a single journal transaction,
412  * we eliminate duplicate reservations for common blocks like the
413  * superblock and group descriptors or bitmaps.
414  *
415  * We will start the transaction here, but each prepare_write will
416  * add a refcount to the transaction, and each commit_write will
417  * remove a refcount.  The transaction will be closed when all of
418  * the pages have been written.
419  */
420 static void *fsfilt_ext3_brw_start(int objcount, struct fsfilt_objinfo *fso,
421                                    int niocount, struct niobuf_local *nb,
422                                    void *desc_private, int logs)
423 {
424         journal_t *journal;
425         handle_t *handle;
426         int needed;
427         ENTRY;
428
429         LASSERT(current->journal_info == desc_private);
430         journal = EXT3_SB(fso->fso_dentry->d_inode->i_sb)->s_journal;
431         needed = fsfilt_ext3_credits_needed(objcount, fso, niocount, nb);
432
433         /* The number of blocks we could _possibly_ dirty can very large.
434          * We reduce our request if it is absurd (and we couldn't get that
435          * many credits for a single handle anyways).
436          *
437          * At some point we have to limit the size of I/Os sent at one time,
438          * increase the size of the journal, or we have to calculate the
439          * actual journal requirements more carefully by checking all of
440          * the blocks instead of being maximally pessimistic.  It remains to
441          * be seen if this is a real problem or not.
442          */
443         if (needed > journal->j_max_transaction_buffers) {
444                 CERROR("want too many journal credits (%d) using %d instead\n",
445                        needed, journal->j_max_transaction_buffers);
446                 needed = journal->j_max_transaction_buffers;
447         }
448
449         LASSERTF(needed > 0, "can't start %d credit transaction\n", needed);
450         handle = ext3_journal_start(fso->fso_dentry->d_inode, needed);
451         if (IS_ERR(handle)) {
452                 CERROR("can't get handle for %d credits: rc = %ld\n", needed,
453                        PTR_ERR(handle));
454         } else {
455                 LASSERT(handle->h_buffer_credits >= needed);
456                 LASSERT(current->journal_info == handle);
457         }
458
459         RETURN(handle);
460 }
461
462 static int fsfilt_ext3_extend(struct inode *inode, unsigned int nblocks,void *h)
463 {
464        handle_t *handle = h;
465
466        /* fsfilt_extend called with nblocks = 0 for testing in special cases */
467        if (nblocks == 0) {
468                handle->h_buffer_credits = 0;
469                CWARN("setting credits of handle %p to zero by request\n", h);
470        }
471
472        if (handle->h_buffer_credits > nblocks)
473                 return 0;
474        if (ext3_journal_extend(handle, nblocks) == 0)
475                 return 0;
476
477        ext3_mark_inode_dirty(handle, inode);
478        return ext3_journal_restart(handle, nblocks);
479 }
480
481 static int fsfilt_ext3_commit(struct inode *inode, void *h, int force_sync)
482 {
483         int rc;
484         handle_t *handle = h;
485
486         LASSERT(current->journal_info == handle);
487         if (force_sync)
488                 handle->h_sync = 1; /* recovery likes this */
489
490         rc = ext3_journal_stop(handle);
491
492         return rc;
493 }
494
495 static int fsfilt_ext3_commit_async(struct inode *inode, void *h,
496                                     void **wait_handle)
497 {
498         unsigned long tid;
499         transaction_t *transaction;
500         handle_t *handle = h;
501         journal_t *journal;
502         int rc;
503
504         LASSERT(current->journal_info == handle);
505
506         transaction = handle->h_transaction;
507         journal = transaction->t_journal;
508         tid = transaction->t_tid;
509         /* we don't want to be blocked */
510         handle->h_sync = 0;
511         rc = ext3_journal_stop(handle);
512         if (rc) {
513                 CERROR("error while stopping transaction: %d\n", rc);
514                 return rc;
515         }
516         fsfilt_log_start_commit(journal, tid);
517
518         *wait_handle = (void *) tid;
519         CDEBUG(D_INODE, "commit async: %lu\n", (unsigned long) tid);
520         return 0;
521 }
522
523 static int fsfilt_ext3_commit_wait(struct inode *inode, void *h)
524 {
525         journal_t *journal = EXT3_JOURNAL(inode);
526         tid_t tid = (tid_t)(long)h;
527
528         CDEBUG(D_INODE, "commit wait: %lu\n", (unsigned long) tid);
529         if (unlikely(is_journal_aborted(journal)))
530                 return -EIO;
531
532         fsfilt_log_wait_commit(EXT3_JOURNAL(inode), tid);
533
534         if (unlikely(is_journal_aborted(journal)))
535                 return -EIO;
536         return 0;
537 }
538
539 static int fsfilt_ext3_setattr(struct dentry *dentry, void *handle,
540                                struct iattr *iattr, int do_trunc)
541 {
542         struct inode *inode = dentry->d_inode;
543         int rc = 0;
544
545 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2,7,50,0)
546         /* Try to correct for a bug in 2.1.0 (LU-221) that caused negative
547          * timestamps to appear to be in the far future, due old timestamp
548          * being stored on disk as an unsigned value.  This fixes up any
549          * bad values held by the client before storing them on disk,
550          * and ensures any timestamp updates are correct.  LU-1042 */
551         if (unlikely(LTIME_S(inode->i_atime) == LU221_BAD_TIME &&
552                      !(iattr->ia_valid & ATTR_ATIME))) {
553                 iattr->ia_valid |= ATTR_ATIME;
554                 LTIME_S(iattr->ia_atime) = 0;
555         }
556         if (unlikely(LTIME_S(inode->i_mtime) == LU221_BAD_TIME &&
557                      !(iattr->ia_valid & ATTR_MTIME))) {
558                 iattr->ia_valid |= ATTR_MTIME;
559                 LTIME_S(iattr->ia_mtime) = 0;
560         }
561         if (unlikely((LTIME_S(inode->i_ctime) == LU221_BAD_TIME ||
562                       LTIME_S(inode->i_ctime) == 0) &&
563                      !(iattr->ia_valid & ATTR_CTIME))) {
564                 iattr->ia_valid |= ATTR_CTIME;
565                 LTIME_S(iattr->ia_ctime) = 0;
566         }
567 #else
568 #warning "remove old LU-221/LU-1042 workaround code"
569 #endif
570
571         /* When initializating timestamps for new inodes, use the filesystem
572          * mkfs time for ctime to avoid e2fsck ibadness incorrectly thinking
573          * that this is potentially an invalid inode.  Files with an old ctime
574          * migrated to a newly-formatted OST with a newer s_mkfs_time will not
575          * hit this check, since it is only for ctime == 0.  LU-1010/LU-1042 */
576         if ((iattr->ia_valid & ATTR_CTIME) && LTIME_S(iattr->ia_ctime) == 0)
577                 LTIME_S(iattr->ia_ctime) =
578                         EXT4_SB(inode->i_sb)->s_es->s_mkfs_time;
579
580         /* Avoid marking the inode dirty on the superblock list unnecessarily.
581          * We are already writing the inode to disk as part of this
582          * transaction and want to avoid a lot of extra inode writeout
583          * later on. b=9828 */
584         if (iattr->ia_valid & ATTR_SIZE && !do_trunc) {
585                 /* ATTR_SIZE would invoke truncate: clear it */
586                 iattr->ia_valid &= ~ATTR_SIZE;
587                 EXT3_I(inode)->i_disksize = iattr->ia_size;
588                 i_size_write(inode, iattr->ia_size);
589
590                 if (iattr->ia_valid & ATTR_UID)
591                         inode->i_uid = iattr->ia_uid;
592                 if (iattr->ia_valid & ATTR_GID)
593                         inode->i_gid = iattr->ia_gid;
594                 if (iattr->ia_valid & ATTR_ATIME)
595                         inode->i_atime = iattr->ia_atime;
596                 if (iattr->ia_valid & ATTR_MTIME)
597                         inode->i_mtime = iattr->ia_mtime;
598                 if (iattr->ia_valid & ATTR_CTIME)
599                         inode->i_ctime = iattr->ia_ctime;
600                 if (iattr->ia_valid & ATTR_MODE) {
601                         inode->i_mode = iattr->ia_mode;
602
603                         if (!cfs_curproc_is_in_groups(inode->i_gid) &&
604                             !cfs_capable(CFS_CAP_FSETID))
605                                 inode->i_mode &= ~S_ISGID;
606                 }
607
608                 inode->i_sb->s_op->dirty_inode(inode);
609
610                 goto out;
611         }
612
613         /* Don't allow setattr to change file type */
614         if (iattr->ia_valid & ATTR_MODE)
615                 iattr->ia_mode = (inode->i_mode & S_IFMT) |
616                                  (iattr->ia_mode & ~S_IFMT);
617
618         /* We set these flags on the client, but have already checked perms
619          * so don't confuse inode_change_ok. */
620         iattr->ia_valid &= ~TIMES_SET_FLAGS;
621
622         if (inode->i_op->setattr) {
623                 rc = inode->i_op->setattr(dentry, iattr);
624         } else {
625 #ifndef HAVE_SIMPLE_SETATTR /* simple_setattr() already call it */
626                 rc = inode_change_ok(inode, iattr);
627                 if (!rc)
628 #endif
629                         rc = simple_setattr(dentry, iattr);
630         }
631
632  out:
633         RETURN(rc);
634 }
635
636 static int fsfilt_ext3_iocontrol(struct inode *inode, struct file *file,
637                                  unsigned int cmd, unsigned long arg)
638 {
639         int rc = 0;
640         ENTRY;
641
642         /* FIXME: Can't do this because of nested transaction deadlock */
643         if (cmd == EXT3_IOC_SETFLAGS) {
644                 /* We can't enable data journaling on OST objects, because
645                 * this forces the transaction to be closed in order to
646                 * flush the journal, but the caller will already have a
647                 * compound transaction open to update the last_rcvd file,
648                 * and this thread would deadlock trying to set the flag. */
649                 if ((*(int *)arg) & EXT3_JOURNAL_DATA_FL) {
650                         CERROR("can't set data journal flag on file\n");
651                         RETURN(-EPERM);
652                 }
653                 /* Because the MDS does not see the EXTENTS_FL set on the
654                  * OST objects, mask this flag into all set flags.  It is
655                  * not legal to clear this flag in any case, so we are not
656                  * changing the functionality by doing this.  b=22911 */
657                 *(int *)arg |= EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL;
658         }
659
660         /* ext4_ioctl does not have a inode argument */
661         if (inode->i_fop->unlocked_ioctl)
662                 rc = inode->i_fop->unlocked_ioctl(file, cmd, arg);
663         else
664                 RETURN(-ENOTTY);
665
666         RETURN(rc);
667 }
668
669 static int fsfilt_ext3_set_md(struct inode *inode, void *handle,
670                               void *lmm, int lmm_size, const char *name)
671 {
672         int rc;
673
674         LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
675
676         rc = ext3_xattr_set_handle(handle, inode, EXT3_XATTR_INDEX_TRUSTED,
677                                    name, lmm, lmm_size, XATTR_NO_CTIME);
678
679
680         if (rc && rc != -EROFS)
681                 CERROR("error adding MD data to inode %lu: rc = %d\n",
682                        inode->i_ino, rc);
683         return rc;
684 }
685
686 /* Must be called with i_mutex held */
687 static int fsfilt_ext3_get_md(struct inode *inode, void *lmm, int lmm_size,
688                               const char *name)
689 {
690         int rc;
691
692         LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
693
694         rc = ext3_xattr_get(inode, EXT3_XATTR_INDEX_TRUSTED,
695                             name, lmm, lmm_size);
696
697         /* This gives us the MD size */
698         if (lmm == NULL)
699                 return (rc == -ENODATA) ? 0 : rc;
700
701         if (rc < 0) {
702                 CDEBUG(D_INFO, "error getting EA %d/%s from inode %lu: rc %d\n",
703                        EXT3_XATTR_INDEX_TRUSTED, name,
704                        inode->i_ino, rc);
705                 memset(lmm, 0, lmm_size);
706                 return (rc == -ENODATA) ? 0 : rc;
707         }
708
709         return rc;
710 }
711
712 static int fsfilt_ext3_send_bio(int rw, struct inode *inode, struct bio *bio)
713 {
714         submit_bio(rw, bio);
715         return 0;
716 }
717
718 static ssize_t fsfilt_ext3_readpage(struct file *file, char *buf, size_t count,
719                                     loff_t *off)
720 {
721         struct inode *inode = file->f_dentry->d_inode;
722         int rc = 0;
723
724         if (S_ISREG(inode->i_mode))
725                 rc = file->f_op->read(file, buf, count, off);
726         else {
727                 const int blkbits = inode->i_sb->s_blocksize_bits;
728                 const int blksize = inode->i_sb->s_blocksize;
729
730                 CDEBUG(D_EXT2, "reading %lu at dir %lu+%llu\n",
731                        (unsigned long)count, inode->i_ino, *off);
732                 while (count > 0) {
733                         struct buffer_head *bh;
734
735                         bh = NULL;
736                         if (*off < i_size_read(inode)) {
737                                 int err = 0;
738
739                                 bh = ext3_bread(NULL, inode, *off >> blkbits,
740                                                 0, &err);
741
742                                 CDEBUG(D_EXT2, "read %u@%llu\n", blksize, *off);
743
744                                 if (bh) {
745                                         memcpy(buf, bh->b_data, blksize);
746                                         brelse(bh);
747                                 } else if (err) {
748                                         /* XXX in theory we should just fake
749                                          * this buffer and continue like ext3,
750                                          * especially if this is a partial read
751                                          */
752                                         CERROR("error read dir %lu+%llu: %d\n",
753                                                inode->i_ino, *off, err);
754                                         RETURN(err);
755                                 }
756                         }
757                         if (!bh) {
758                                 struct ext3_dir_entry_2 *fake = (void *)buf;
759
760                                 CDEBUG(D_EXT2, "fake %u@%llu\n", blksize, *off);
761                                 memset(fake, 0, sizeof(*fake));
762                                 fake->rec_len = cpu_to_le16(blksize);
763                         }
764                         count -= blksize;
765                         buf += blksize;
766                         *off += blksize;
767                         rc += blksize;
768                 }
769         }
770
771         return rc;
772 }
773
774 #ifdef HAVE_EXT4_JOURNAL_CALLBACK_ADD
775 static void fsfilt_ext3_cb_func(struct super_block *sb,
776                                 struct journal_callback *jcb, int error)
777 #else
778 static void fsfilt_ext3_cb_func(struct journal_callback *jcb, int error)
779 #endif
780 {
781         struct fsfilt_cb_data *fcb = container_of(jcb, typeof(*fcb), cb_jcb);
782
783         fcb->cb_func(fcb->cb_obd, fcb->cb_last_rcvd, fcb->cb_data, error);
784
785         OBD_SLAB_FREE(fcb, fcb_cache, sizeof *fcb);
786 }
787
788 static int fsfilt_ext3_add_journal_cb(struct obd_device *obd, __u64 last_rcvd,
789                                       void *handle, fsfilt_cb_t cb_func,
790                                       void *cb_data)
791 {
792         struct fsfilt_cb_data *fcb;
793
794         OBD_SLAB_ALLOC_PTR_GFP(fcb, fcb_cache, CFS_ALLOC_IO);
795         if (fcb == NULL)
796                 RETURN(-ENOMEM);
797
798         fcb->cb_func = cb_func;
799         fcb->cb_obd = obd;
800         fcb->cb_last_rcvd = last_rcvd;
801         fcb->cb_data = cb_data;
802
803         CDEBUG(D_EXT2, "set callback for last_rcvd: "LPD64"\n", last_rcvd);
804         fsfilt_journal_callback_set(handle, fsfilt_ext3_cb_func, &fcb->cb_jcb);
805
806         return 0;
807 }
808
809 static int fsfilt_ext3_statfs(struct super_block *sb, struct obd_statfs *osfs)
810 {
811         struct kstatfs sfs;
812         int rc;
813
814         memset(&sfs, 0, sizeof(sfs));
815         rc = ll_do_statfs(sb, &sfs);
816         statfs_pack(osfs, &sfs);
817         return rc;
818 }
819
820 static int fsfilt_ext3_sync(struct super_block *sb)
821 {
822         return ext3_force_commit(sb);
823 }
824
825 #ifndef EXT3_EXTENTS_FL
826 #define EXT3_EXTENTS_FL                 0x00080000 /* Inode uses extents */
827 #endif
828
829 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
830 # define fsfilt_up_truncate_sem(inode)  up(&LDISKFS_I(inode)->truncate_sem);
831 # define fsfilt_down_truncate_sem(inode)  down(&LDISKFS_I(inode)->truncate_sem);
832 #else
833 # define fsfilt_up_truncate_sem(inode) do{ }while(0)
834 # define fsfilt_down_truncate_sem(inode) do{ }while(0)
835 #endif
836
837 #ifndef EXT_ASSERT
838 #define EXT_ASSERT(cond)  BUG_ON(!(cond))
839 #endif
840
841 #ifdef EXT3_EXT_HAS_NO_TREE
842 /* for kernels 2.6.18 and later */
843 #define EXT_GENERATION(inode)           (EXT4_I(inode)->i_ext_generation)
844 #define ext3_ext_base                   inode
845 #define ext3_ext_base2inode(inode)      (inode)
846 #define EXT_DEPTH(inode)                ext_depth(inode)
847 #define fsfilt_ext3_ext_walk_space(inode, block, num, cb, cbdata) \
848                         ext3_ext_walk_space(inode, block, num, cb, cbdata);
849 #else
850 #define ext3_ext_base                   ext3_extents_tree
851 #define ext3_ext_base2inode(tree)       (tree->inode)
852 #define fsfilt_ext3_ext_walk_space(tree, block, num, cb, cbdata) \
853                         ext3_ext_walk_space(tree, block, num, cb);
854 #endif
855
856 struct bpointers {
857         unsigned long *blocks;
858         int *created;
859         unsigned long start;
860         int num;
861         int init_num;
862         int create;
863 };
864
865 static long ext3_ext_find_goal(struct inode *inode, struct ext3_ext_path *path,
866                                unsigned long block, int *aflags)
867 {
868         struct ext3_inode_info *ei = EXT3_I(inode);
869         unsigned long bg_start;
870         unsigned long colour;
871         int depth;
872
873         if (path) {
874                 struct ext3_extent *ex;
875                 depth = path->p_depth;
876
877                 /* try to predict block placement */
878                 if ((ex = path[depth].p_ext))
879                         return ext_pblock(ex) + (block - le32_to_cpu(ex->ee_block));
880
881                 /* it looks index is empty
882                  * try to find starting from index itself */
883                 if (path[depth].p_bh)
884                         return path[depth].p_bh->b_blocknr;
885         }
886
887         /* OK. use inode's group */
888         bg_start = (ei->i_block_group * EXT3_BLOCKS_PER_GROUP(inode->i_sb)) +
889                 le32_to_cpu(EXT3_SB(inode->i_sb)->s_es->s_first_data_block);
890         colour = (current->pid % 16) *
891                 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
892         return bg_start + colour + block;
893 }
894
895 #define ll_unmap_underlying_metadata(sb, blocknr) \
896         unmap_underlying_metadata((sb)->s_bdev, blocknr)
897
898 #ifndef EXT3_MB_HINT_GROUP_ALLOC
899 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
900                                 struct ext3_ext_path *path, unsigned long block,
901                                 unsigned long *count, int *err)
902 {
903         unsigned long pblock, goal;
904         int aflags = 0;
905         struct inode *inode = ext3_ext_base2inode(base);
906
907         goal = ext3_ext_find_goal(inode, path, block, &aflags);
908         aflags |= 2; /* block have been already reserved */
909         pblock = ext3_mb_new_blocks(handle, inode, goal, count, aflags, err);
910         return pblock;
911
912 }
913 #else
914 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
915                                 struct ext3_ext_path *path, unsigned long block,
916                                 unsigned long *count, int *err)
917 {
918         struct inode *inode = ext3_ext_base2inode(base);
919         struct ext3_allocation_request ar;
920         unsigned long pblock;
921         int aflags;
922
923         /* find neighbour allocated blocks */
924         ar.lleft = block;
925         *err = ext3_ext_search_left(base, path, &ar.lleft, &ar.pleft);
926         if (*err)
927                 return 0;
928         ar.lright = block;
929         *err = ext3_ext_search_right(base, path, &ar.lright, &ar.pright);
930         if (*err)
931                 return 0;
932
933         /* allocate new block */
934         ar.goal = ext3_ext_find_goal(inode, path, block, &aflags);
935         ar.inode = inode;
936         ar.logical = block;
937         ar.len = *count;
938         ar.flags = EXT3_MB_HINT_DATA;
939         pblock = ext3_mb_new_blocks(handle, &ar, err);
940         *count = ar.len;
941         return pblock;
942 }
943 #endif
944
945 #ifdef EXT3_EXT_HAS_NO_TREE
946 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
947                                   struct ext3_ext_path *path,
948                                   struct ext3_ext_cache *cex,
949 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
950                                    struct ext3_extent *ex,
951 #endif
952                                   void *cbdata)
953 {
954         struct bpointers *bp = cbdata;
955 #else
956 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
957                                   struct ext3_ext_path *path,
958                                   struct ext3_ext_cache *cex
959 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
960                                   , struct ext3_extent *ex
961 #endif
962                                  )
963 {
964         struct bpointers *bp = base->private;
965 #endif
966         struct inode *inode = ext3_ext_base2inode(base);
967         struct ext3_extent nex;
968         unsigned long pblock;
969         unsigned long tgen;
970         int err, i;
971         unsigned long count;
972         handle_t *handle;
973
974         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
975                 err = EXT_CONTINUE;
976                 goto map;
977         }
978
979         if (bp->create == 0) {
980                 i = 0;
981                 if (cex->ec_block < bp->start)
982                         i = bp->start - cex->ec_block;
983                 if (i >= cex->ec_len)
984                         CERROR("nothing to do?! i = %d, e_num = %u\n",
985                                         i, cex->ec_len);
986                 for (; i < cex->ec_len && bp->num; i++) {
987                         *(bp->created) = 0;
988                         bp->created++;
989                         *(bp->blocks) = 0;
990                         bp->blocks++;
991                         bp->num--;
992                         bp->start++;
993                 }
994
995                 return EXT_CONTINUE;
996         }
997
998         tgen = EXT_GENERATION(base);
999         count = ext3_ext_calc_credits_for_insert(base, path);
1000         fsfilt_up_truncate_sem(inode);
1001
1002         handle = ext3_journal_start(inode, count+EXT3_ALLOC_NEEDED+1);
1003         if (IS_ERR(handle)) {
1004                 fsfilt_down_truncate_sem(inode);
1005                 return PTR_ERR(handle);
1006         }
1007
1008         fsfilt_down_truncate_sem(inode);
1009         if (tgen != EXT_GENERATION(base)) {
1010                 /* the tree has changed. so path can be invalid at moment */
1011                 ext3_journal_stop(handle);
1012                 return EXT_REPEAT;
1013         }
1014
1015         /* In 2.6.32 kernel, ext4_ext_walk_space()'s callback func is not
1016          * protected by i_data_sem as whole. so we patch it to store
1017          * generation to path and now verify the tree hasn't changed */
1018         down_write((&EXT4_I(inode)->i_data_sem));
1019
1020         /* validate extent, make sure the extent tree does not changed */
1021         if (EXT_GENERATION(base) != path[0].p_generation) {
1022                 /* cex is invalid, try again */
1023                 up_write(&EXT4_I(inode)->i_data_sem);
1024                 ext3_journal_stop(handle);
1025                 return EXT_REPEAT;
1026         }
1027
1028         count = cex->ec_len;
1029         pblock = new_blocks(handle, base, path, cex->ec_block, &count, &err);
1030         if (!pblock)
1031                 goto out;
1032         EXT_ASSERT(count <= cex->ec_len);
1033
1034         /* insert new extent */
1035         nex.ee_block = cpu_to_le32(cex->ec_block);
1036         ext3_ext_store_pblock(&nex, pblock);
1037         nex.ee_len = cpu_to_le16(count);
1038         err = fsfilt_ext3_ext_insert_extent(handle, base, path, &nex, 0);
1039         if (err) {
1040                 /* free data blocks we just allocated */
1041                 /* not a good idea to call discard here directly,
1042                  * but otherwise we'd need to call it every free() */
1043 #ifdef EXT3_MB_HINT_GROUP_ALLOC
1044                 ext3_mb_discard_inode_preallocations(inode);
1045 #endif
1046                 ext3_free_blocks(handle, inode, ext_pblock(&nex),
1047                                  cpu_to_le16(nex.ee_len), 0);
1048                 goto out;
1049         }
1050
1051         /*
1052          * Putting len of the actual extent we just inserted,
1053          * we are asking ext3_ext_walk_space() to continue
1054          * scaning after that block
1055          */
1056         cex->ec_len = le16_to_cpu(nex.ee_len);
1057         cex->ec_start = ext_pblock(&nex);
1058         BUG_ON(le16_to_cpu(nex.ee_len) == 0);
1059         BUG_ON(le32_to_cpu(nex.ee_block) != cex->ec_block);
1060
1061 out:
1062         up_write((&EXT4_I(inode)->i_data_sem));
1063         ext3_journal_stop(handle);
1064 map:
1065         if (err >= 0) {
1066                 /* map blocks */
1067                 if (bp->num == 0) {
1068                         CERROR("hmm. why do we find this extent?\n");
1069                         CERROR("initial space: %lu:%u\n",
1070                                 bp->start, bp->init_num);
1071                         CERROR("current extent: %u/%u/%llu %d\n",
1072                                 cex->ec_block, cex->ec_len,
1073                                 (unsigned long long)cex->ec_start,
1074                                 cex->ec_type);
1075                 }
1076                 i = 0;
1077                 if (cex->ec_block < bp->start)
1078                         i = bp->start - cex->ec_block;
1079                 if (i >= cex->ec_len)
1080                         CERROR("nothing to do?! i = %d, e_num = %u\n",
1081                                         i, cex->ec_len);
1082                 for (; i < cex->ec_len && bp->num; i++) {
1083                         *(bp->blocks) = cex->ec_start + i;
1084                         if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
1085                                 *(bp->created) = 0;
1086                         } else {
1087                                 *(bp->created) = 1;
1088                                 /* unmap any possible underlying metadata from
1089                                  * the block device mapping.  bug 6998. */
1090                                 ll_unmap_underlying_metadata(inode->i_sb,
1091                                                              *(bp->blocks));
1092                         }
1093                         bp->created++;
1094                         bp->blocks++;
1095                         bp->num--;
1096                         bp->start++;
1097                 }
1098         }
1099         return err;
1100 }
1101
1102 int fsfilt_map_nblocks(struct inode *inode, unsigned long block,
1103                        unsigned long num, unsigned long *blocks,
1104                        int *created, int create)
1105 {
1106 #ifdef EXT3_EXT_HAS_NO_TREE
1107         struct ext3_ext_base *base = inode;
1108 #else
1109         struct ext3_extents_tree tree;
1110         struct ext3_ext_base *base = &tree;
1111 #endif
1112         struct bpointers bp;
1113         int err;
1114
1115         CDEBUG(D_OTHER, "blocks %lu-%lu requested for inode %u\n",
1116                block, block + num - 1, (unsigned) inode->i_ino);
1117
1118 #ifndef EXT3_EXT_HAS_NO_TREE
1119         ext3_init_tree_desc(base, inode);
1120         tree.private = &bp;
1121 #endif
1122         bp.blocks = blocks;
1123         bp.created = created;
1124         bp.start = block;
1125         bp.init_num = bp.num = num;
1126         bp.create = create;
1127
1128         fsfilt_down_truncate_sem(inode);
1129         err = fsfilt_ext3_ext_walk_space(base, block, num,
1130                                          ext3_ext_new_extent_cb, &bp);
1131         ext3_ext_invalidate_cache(base);
1132         fsfilt_up_truncate_sem(inode);
1133
1134         return err;
1135 }
1136
1137 int fsfilt_ext3_map_ext_inode_pages(struct inode *inode, struct page **page,
1138                                     int pages, unsigned long *blocks,
1139                                     int *created, int create)
1140 {
1141         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1142         int rc = 0, i = 0;
1143         struct page *fp = NULL;
1144         int clen = 0;
1145
1146         CDEBUG(D_OTHER, "inode %lu: map %d pages from %lu\n",
1147                 inode->i_ino, pages, (*page)->index);
1148
1149         /* pages are sorted already. so, we just have to find
1150          * contig. space and process them properly */
1151         while (i < pages) {
1152                 if (fp == NULL) {
1153                         /* start new extent */
1154                         fp = *page++;
1155                         clen = 1;
1156                         i++;
1157                         continue;
1158                 } else if (fp->index + clen == (*page)->index) {
1159                         /* continue the extent */
1160                         page++;
1161                         clen++;
1162                         i++;
1163                         continue;
1164                 }
1165
1166                 /* process found extent */
1167                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1168                                         clen * blocks_per_page, blocks,
1169                                         created, create);
1170                 if (rc)
1171                         GOTO(cleanup, rc);
1172
1173                 /* look for next extent */
1174                 fp = NULL;
1175                 blocks += blocks_per_page * clen;
1176                 created += blocks_per_page * clen;
1177         }
1178
1179         if (fp)
1180                 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1181                                         clen * blocks_per_page, blocks,
1182                                         created, create);
1183 cleanup:
1184         return rc;
1185 }
1186
1187 extern int ext3_map_inode_page(struct inode *inode, struct page *page,
1188                                unsigned long *blocks, int *created, int create);
1189 int fsfilt_ext3_map_bm_inode_pages(struct inode *inode, struct page **page,
1190                                    int pages, unsigned long *blocks,
1191                                    int *created, int create)
1192 {
1193         int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1194         unsigned long *b;
1195         int rc = 0, i, *cr;
1196
1197         for (i = 0, cr = created, b = blocks; i < pages; i++, page++) {
1198                 rc = ext3_map_inode_page(inode, *page, b, cr, create);
1199                 if (rc) {
1200                         CERROR("ino %lu, blk %lu cr %u create %d: rc %d\n",
1201                                inode->i_ino, *b, *cr, create, rc);
1202                         break;
1203                 }
1204
1205                 b += blocks_per_page;
1206                 cr += blocks_per_page;
1207         }
1208         return rc;
1209 }
1210
1211 int fsfilt_ext3_map_inode_pages(struct inode *inode, struct page **page,
1212                                 int pages, unsigned long *blocks,
1213                                 int *created, int create,
1214                                 cfs_mutex_t *optional_mutex)
1215 {
1216         int rc;
1217
1218         if (EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL) {
1219                 rc = fsfilt_ext3_map_ext_inode_pages(inode, page, pages,
1220                                                      blocks, created, create);
1221                 return rc;
1222         }
1223         if (optional_mutex != NULL)
1224                 cfs_mutex_lock(optional_mutex);
1225         rc = fsfilt_ext3_map_bm_inode_pages(inode, page, pages, blocks,
1226                                             created, create);
1227         if (optional_mutex != NULL)
1228                 cfs_mutex_unlock(optional_mutex);
1229
1230         return rc;
1231 }
1232
1233 int fsfilt_ext3_read(struct inode *inode, void *buf, int size, loff_t *offs)
1234 {
1235         unsigned long block;
1236         struct buffer_head *bh;
1237         int err, blocksize, csize, boffs, osize = size;
1238
1239         /* prevent reading after eof */
1240         spin_lock(&inode->i_lock);
1241         if (i_size_read(inode) < *offs + size) {
1242                 size = i_size_read(inode) - *offs;
1243                 spin_unlock(&inode->i_lock);
1244                 if (size < 0) {
1245                         CDEBUG(D_EXT2, "size %llu is too short for read @%llu\n",
1246                                i_size_read(inode), *offs);
1247                         return -EBADR;
1248                 } else if (size == 0) {
1249                         return 0;
1250                 }
1251         } else {
1252                 spin_unlock(&inode->i_lock);
1253         }
1254
1255         blocksize = 1 << inode->i_blkbits;
1256
1257         while (size > 0) {
1258                 block = *offs >> inode->i_blkbits;
1259                 boffs = *offs & (blocksize - 1);
1260                 csize = min(blocksize - boffs, size);
1261                 bh = ext3_bread(NULL, inode, block, 0, &err);
1262                 if (!bh) {
1263                         CERROR("can't read block: %d\n", err);
1264                         return err;
1265                 }
1266
1267                 memcpy(buf, bh->b_data + boffs, csize);
1268                 brelse(bh);
1269
1270                 *offs += csize;
1271                 buf += csize;
1272                 size -= csize;
1273         }
1274         return osize;
1275 }
1276 EXPORT_SYMBOL(fsfilt_ext3_read);
1277
1278 static int fsfilt_ext3_read_record(struct file * file, void *buf,
1279                                    int size, loff_t *offs)
1280 {
1281         int rc;
1282         rc = fsfilt_ext3_read(file->f_dentry->d_inode, buf, size, offs);
1283         if (rc > 0)
1284                 rc = 0;
1285         return rc;
1286 }
1287
1288 int fsfilt_ext3_write_handle(struct inode *inode, void *buf, int bufsize,
1289                                 loff_t *offs, handle_t *handle)
1290 {
1291         struct buffer_head *bh = NULL;
1292         loff_t old_size = i_size_read(inode), offset = *offs;
1293         loff_t new_size = i_size_read(inode);
1294         unsigned long block;
1295         int err = 0, blocksize = 1 << inode->i_blkbits, size, boffs;
1296
1297         while (bufsize > 0) {
1298                 if (bh != NULL)
1299                         brelse(bh);
1300
1301                 block = offset >> inode->i_blkbits;
1302                 boffs = offset & (blocksize - 1);
1303                 size = min(blocksize - boffs, bufsize);
1304                 bh = ext3_bread(handle, inode, block, 1, &err);
1305                 if (!bh) {
1306                         CERROR("can't read/create block: %d\n", err);
1307                         break;
1308                 }
1309
1310                 err = ext3_journal_get_write_access(handle, bh);
1311                 if (err) {
1312                         CERROR("journal_get_write_access() returned error %d\n",
1313                                err);
1314                         break;
1315                 }
1316                 LASSERT(bh->b_data + boffs + size <= bh->b_data + bh->b_size);
1317                 memcpy(bh->b_data + boffs, buf, size);
1318                 err = ext3_journal_dirty_metadata(handle, bh);
1319                 if (err) {
1320                         CERROR("journal_dirty_metadata() returned error %d\n",
1321                                err);
1322                         break;
1323                 }
1324                 if (offset + size > new_size)
1325                         new_size = offset + size;
1326                 offset += size;
1327                 bufsize -= size;
1328                 buf += size;
1329         }
1330         if (bh)
1331                 brelse(bh);
1332
1333         /* correct in-core and on-disk sizes */
1334         if (new_size > i_size_read(inode)) {
1335                 spin_lock(&inode->i_lock);
1336                 if (new_size > i_size_read(inode))
1337                         i_size_write(inode, new_size);
1338                 if (i_size_read(inode) > EXT3_I(inode)->i_disksize)
1339                         EXT3_I(inode)->i_disksize = i_size_read(inode);
1340                 if (i_size_read(inode) > old_size) {
1341                         spin_unlock(&inode->i_lock);
1342                         mark_inode_dirty(inode);
1343                 } else {
1344                         spin_unlock(&inode->i_lock);
1345                 }
1346         }
1347
1348         if (err == 0)
1349                 *offs = offset;
1350         return err;
1351 }
1352 EXPORT_SYMBOL(fsfilt_ext3_write_handle);
1353
1354 static int fsfilt_ext3_write_record(struct file *file, void *buf, int bufsize,
1355                                     loff_t *offs, int force_sync)
1356 {
1357         struct inode *inode = file->f_dentry->d_inode;
1358         handle_t *handle;
1359         int err, block_count = 0, blocksize;
1360
1361         /* Determine how many transaction credits are needed */
1362         blocksize = 1 << inode->i_blkbits;
1363         block_count = (*offs & (blocksize - 1)) + bufsize;
1364         block_count = (block_count + blocksize - 1) >> inode->i_blkbits;
1365
1366         handle = ext3_journal_start(inode,
1367                                block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2);
1368         if (IS_ERR(handle)) {
1369                 CERROR("can't start transaction for %d blocks (%d bytes)\n",
1370                        block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2, bufsize);
1371                 return PTR_ERR(handle);
1372         }
1373
1374         err = fsfilt_ext3_write_handle(inode, buf, bufsize, offs, handle);
1375
1376         if (!err && force_sync)
1377                 handle->h_sync = 1; /* recovery likes this */
1378
1379         ext3_journal_stop(handle);
1380
1381         return err;
1382 }
1383
1384 static int fsfilt_ext3_setup(struct super_block *sb)
1385 {
1386 #if ((LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,6)) && \
1387      defined(HAVE_QUOTA_SUPPORT)) || defined(S_PDIROPS)
1388         struct ext3_sb_info *sbi = EXT3_SB(sb);
1389 #if 0
1390         sbi->dx_lock = fsfilt_ext3_dx_lock;
1391         sbi->dx_unlock = fsfilt_ext3_dx_unlock;
1392 #endif
1393 #endif
1394         if (!EXT3_HAS_COMPAT_FEATURE(sb,
1395                                 EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1396                 CERROR("ext3 mounted without journal\n");
1397                 return -EINVAL;
1398         }
1399
1400 #ifdef S_PDIROPS
1401         CWARN("Enabling PDIROPS\n");
1402         set_opt(sbi->s_mount_opt, PDIROPS);
1403         sb->s_flags |= S_PDIROPS;
1404 #endif
1405         if (!EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1406                 CWARN("filesystem doesn't have dir_index feature enabled\n");
1407 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,6)) && defined(HAVE_QUOTA_SUPPORT)
1408         /* enable journaled quota support */
1409         /* kfreed in ext3_put_super() */
1410         sbi->s_qf_names[USRQUOTA] = kstrdup("lquota.user.reserved", GFP_KERNEL);
1411         if (!sbi->s_qf_names[USRQUOTA])
1412                 return -ENOMEM;
1413         sbi->s_qf_names[GRPQUOTA] = kstrdup("lquota.group.reserved", GFP_KERNEL);
1414         if (!sbi->s_qf_names[GRPQUOTA]) {
1415                 kfree(sbi->s_qf_names[USRQUOTA]);
1416                 sbi->s_qf_names[USRQUOTA] = NULL;
1417                 return -ENOMEM;
1418         }
1419         sbi->s_jquota_fmt = QFMT_LUSTRE;
1420 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,13))
1421         set_opt(sbi->s_mount_opt, QUOTA);
1422 #endif
1423 #endif
1424         return 0;
1425 }
1426
1427 /* If fso is NULL, op is FSFILT operation, otherwise op is number of fso
1428    objects. Logs is number of logfiles to update */
1429 static int fsfilt_ext3_get_op_len(int op, struct fsfilt_objinfo *fso, int logs)
1430 {
1431         if ( !fso ) {
1432                 switch(op) {
1433                 case FSFILT_OP_CREATE:
1434                                  /* directory leaf, index & indirect & EA*/
1435                         return 4 + 3 * logs;
1436                 case FSFILT_OP_UNLINK:
1437                         return 3 * logs;
1438                 }
1439         } else {
1440                 int i;
1441                 int needed = 0;
1442                 struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
1443                 int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
1444                 int addrpp = EXT3_ADDR_PER_BLOCK(sb) * blockpp;
1445                 for (i = 0; i < op; i++, fso++) {
1446                         int nblocks = fso->fso_bufcnt * blockpp;
1447                         int ndindirect = min(nblocks, addrpp + 1);
1448                         int nindir = nblocks + ndindirect + 1;
1449
1450                         needed += nindir;
1451                 }
1452                 return needed + 3 * logs;
1453         }
1454
1455         return 0;
1456 }
1457
1458 #ifdef HAVE_QUOTA_SUPPORT
1459 #define DQINFO_COPY(out, in)                    \
1460 do {                                            \
1461         Q_COPY(out, in, dqi_bgrace);            \
1462         Q_COPY(out, in, dqi_igrace);            \
1463         Q_COPY(out, in, dqi_flags);             \
1464         Q_COPY(out, in, dqi_valid);             \
1465 } while (0)
1466
1467 #define DQBLK_COPY(out, in)                     \
1468 do {                                            \
1469         Q_COPY(out, in, dqb_bhardlimit);        \
1470         Q_COPY(out, in, dqb_bsoftlimit);        \
1471         Q_COPY(out, in, dqb_curspace);          \
1472         Q_COPY(out, in, dqb_ihardlimit);        \
1473         Q_COPY(out, in, dqb_isoftlimit);        \
1474         Q_COPY(out, in, dqb_curinodes);         \
1475         Q_COPY(out, in, dqb_btime);             \
1476         Q_COPY(out, in, dqb_itime);             \
1477         Q_COPY(out, in, dqb_valid);             \
1478 } while (0)
1479
1480 static int fsfilt_ext3_quotactl(struct super_block *sb,
1481                                 struct obd_quotactl *oqc)
1482 {
1483         int i, rc = 0, error = 0;
1484         const struct quotactl_ops *qcop;
1485         struct if_dqinfo *info;
1486         struct if_dqblk *dqblk;
1487         ENTRY;
1488
1489         if (!sb->s_qcop)
1490                 RETURN(-ENOSYS);
1491
1492         OBD_ALLOC_PTR(info);
1493         if (!info)
1494                 RETURN(-ENOMEM);
1495         OBD_ALLOC_PTR(dqblk);
1496         if (!dqblk) {
1497                 OBD_FREE_PTR(info);
1498                 RETURN(-ENOMEM);
1499         }
1500
1501         DQINFO_COPY(info, &oqc->qc_dqinfo);
1502         DQBLK_COPY(dqblk, &oqc->qc_dqblk);
1503
1504         qcop = sb->s_qcop;
1505         if (oqc->qc_cmd == Q_QUOTAON || oqc->qc_cmd == Q_QUOTAOFF) {
1506                 for (i = 0; i < MAXQUOTAS; i++) {
1507                         if (!Q_TYPESET(oqc, i))
1508                                 continue;
1509
1510                         if (oqc->qc_cmd == Q_QUOTAON) {
1511                                 char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1512
1513                                 LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1514
1515                                 rc = ll_quota_on(sb, i, QFMT_LUSTRE,
1516                                                  name[i], 0);
1517                         } else if (oqc->qc_cmd == Q_QUOTAOFF) {
1518                                 rc = ll_quota_off(sb, i, 0);
1519                         }
1520
1521                         if (rc == -EBUSY)
1522                                 error = rc;
1523                         else if (rc)
1524                                 GOTO(out, rc);
1525                 }
1526                 GOTO(out, rc ?: error);
1527         }
1528
1529         switch (oqc->qc_cmd) {
1530         case Q_GETOINFO:
1531         case Q_GETINFO:
1532                 if (!qcop->get_info)
1533                         GOTO(out, rc = -ENOSYS);
1534                 rc = qcop->get_info(sb, oqc->qc_type, info);
1535                 break;
1536         case Q_SETQUOTA:
1537         case Q_INITQUOTA:
1538                 if (!qcop->set_dqblk)
1539                         GOTO(out, rc = -ENOSYS);
1540                 rc = qcop->set_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1541                 break;
1542         case Q_GETOQUOTA:
1543         case Q_GETQUOTA:
1544                 if (!qcop->get_dqblk)
1545                         GOTO(out, rc = -ENOSYS);
1546                 rc = qcop->get_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1547                 if (!rc)
1548                         dqblk->dqb_valid = QIF_LIMITS | QIF_USAGE;
1549                 break;
1550         case Q_SYNC:
1551                 if (!sb->s_qcop->quota_sync)
1552                         GOTO(out, rc = -ENOSYS);
1553                 qcop->quota_sync(sb, oqc->qc_type);
1554                 break;
1555         case Q_FINVALIDATE:
1556                 CDEBUG(D_WARNING, "invalidating operational quota files\n");
1557                 for (i = 0; i < MAXQUOTAS; i++) {
1558                         struct file *fp;
1559                         char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1560
1561                         LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1562
1563                         if (!Q_TYPESET(oqc, i))
1564                                 continue;
1565
1566                         fp = filp_open(name[i], O_CREAT | O_TRUNC | O_RDWR, 0644);
1567                         if (IS_ERR(fp)) {
1568                                 rc = PTR_ERR(fp);
1569                                 CERROR("error invalidating operational quota file"
1570                                        " %s (rc:%d)\n", name[i], rc);
1571                         } else {
1572                                 filp_close(fp, 0);
1573                         }
1574
1575                 }
1576                 break;
1577         default:
1578                 CERROR("unsupported quotactl command: %d\n", oqc->qc_cmd);
1579                 LBUG();
1580         }
1581 out:
1582         DQINFO_COPY(&oqc->qc_dqinfo, info);
1583         DQBLK_COPY(&oqc->qc_dqblk, dqblk);
1584
1585         OBD_FREE_PTR(info);
1586         OBD_FREE_PTR(dqblk);
1587
1588         if (rc)
1589                 CDEBUG(D_QUOTA, "quotactl command %#x, id %u, type %u "
1590                                 "failed: %d\n",
1591                        oqc->qc_cmd, oqc->qc_id, oqc->qc_type, rc);
1592         RETURN(rc);
1593 }
1594
1595 struct chk_dqblk{
1596         cfs_hlist_node_t        dqb_hash;        /** quotacheck hash */
1597         cfs_list_t              dqb_list;        /** in list also */
1598         qid_t                   dqb_id;          /** uid/gid */
1599         short                   dqb_type;        /** USRQUOTA/GRPQUOTA */
1600         qsize_t                 dqb_bhardlimit;  /** block hard limit */
1601         qsize_t                 dqb_bsoftlimit;  /** block soft limit */
1602         qsize_t                 dqb_curspace;    /** current space */
1603         qsize_t                 dqb_ihardlimit;  /** inode hard limit */
1604         qsize_t                 dqb_isoftlimit;  /** inode soft limit */
1605         qsize_t                 dqb_curinodes;   /** current inodes */
1606         obd_time                dqb_btime;       /** block grace time */
1607         obd_time                dqb_itime;       /** inode grace time */
1608         __u32                   dqb_valid;       /** flag for above fields */
1609 };
1610
1611 static inline unsigned int chkquot_hash(qid_t id, int type)
1612                                         __attribute__((__const__));
1613
1614 static inline unsigned int chkquot_hash(qid_t id, int type)
1615 {
1616         return (id * (MAXQUOTAS - type)) % NR_DQHASH;
1617 }
1618
1619 static inline struct chk_dqblk *
1620 find_chkquot(cfs_hlist_head_t *head, qid_t id, int type)
1621 {
1622         cfs_hlist_node_t *node;
1623         struct chk_dqblk *cdqb;
1624
1625         cfs_hlist_for_each(node, head) {
1626                 cdqb = cfs_hlist_entry(node, struct chk_dqblk, dqb_hash);
1627                 if (cdqb->dqb_id == id && cdqb->dqb_type == type)
1628                         return cdqb;
1629         }
1630
1631         return NULL;
1632 }
1633
1634 static struct chk_dqblk *alloc_chkquot(qid_t id, int type)
1635 {
1636         struct chk_dqblk *cdqb;
1637
1638         OBD_ALLOC_PTR(cdqb);
1639         if (cdqb) {
1640                 CFS_INIT_HLIST_NODE(&cdqb->dqb_hash);
1641                 CFS_INIT_LIST_HEAD(&cdqb->dqb_list);
1642                 cdqb->dqb_id = id;
1643                 cdqb->dqb_type = type;
1644         }
1645
1646         return cdqb;
1647 }
1648
1649 static struct chk_dqblk *
1650 cqget(struct super_block *sb, cfs_hlist_head_t *hash,
1651       cfs_list_t *list, qid_t id, int type, int first_check)
1652 {
1653         cfs_hlist_head_t *head = hash + chkquot_hash(id, type);
1654         struct if_dqblk dqb;
1655         struct chk_dqblk *cdqb;
1656         int rc;
1657
1658         cdqb = find_chkquot(head, id, type);
1659         if (cdqb)
1660                 return cdqb;
1661
1662         cdqb = alloc_chkquot(id, type);
1663         if (!cdqb)
1664                 return NULL;
1665
1666         if (!first_check) {
1667                 rc = sb->s_qcop->get_dqblk(sb, type, id, &dqb);
1668                 if (rc) {
1669                         CERROR("get_dqblk of id %u, type %d failed: %d\n",
1670                                id, type, rc);
1671                 } else {
1672                         DQBLK_COPY(cdqb, &dqb);
1673                         cdqb->dqb_curspace = 0;
1674                         cdqb->dqb_curinodes = 0;
1675                 }
1676         }
1677
1678         cfs_hlist_add_head(&cdqb->dqb_hash, head);
1679         cfs_list_add_tail(&cdqb->dqb_list, list);
1680
1681         return cdqb;
1682 }
1683
1684 static inline int quota_onoff(struct super_block *sb, int cmd, int type, int qfmt)
1685 {
1686         struct obd_quotactl *oqctl;
1687         int rc;
1688
1689         OBD_ALLOC_PTR(oqctl);
1690         if (!oqctl)
1691                 RETURN(-ENOMEM);
1692
1693         oqctl->qc_cmd = cmd;
1694         oqctl->qc_id = qfmt;
1695         oqctl->qc_type = type;
1696         rc = fsfilt_ext3_quotactl(sb, oqctl);
1697
1698         OBD_FREE_PTR(oqctl);
1699         return rc;
1700 }
1701
1702 static inline int read_old_dqinfo(struct super_block *sb, int type,
1703                                   struct if_dqinfo *dqinfo)
1704 {
1705         struct obd_quotactl *oqctl;
1706         int rc;
1707         ENTRY;
1708
1709         OBD_ALLOC_PTR(oqctl);
1710         if (!oqctl)
1711                 RETURN(-ENOMEM);
1712
1713         oqctl->qc_cmd = Q_GETINFO;
1714         oqctl->qc_type = type;
1715         rc = fsfilt_ext3_quotactl(sb, oqctl);
1716         if (!rc)
1717                 ((struct obd_dqinfo *)dqinfo)[type] = oqctl->qc_dqinfo;
1718
1719         OBD_FREE_PTR(oqctl);
1720         RETURN(rc);
1721 }
1722
1723 struct qchk_ctxt {
1724         cfs_hlist_head_t        qckt_hash[NR_DQHASH];      /* quotacheck hash */
1725         cfs_list_t              qckt_list;                 /* quotacheck list */
1726         int                     qckt_first_check[MAXQUOTAS]; /* 1 if no old quotafile */
1727         struct if_dqinfo        qckt_dqinfo[MAXQUOTAS];    /* old dqinfo */
1728 };
1729
1730 static int add_inode_quota(struct inode *inode, struct qchk_ctxt *qctxt,
1731                            struct obd_quotactl *oqc)
1732 {
1733         struct chk_dqblk *cdqb[MAXQUOTAS] = { NULL, };
1734         loff_t size = 0;
1735         qid_t qid[MAXQUOTAS];
1736         int cnt, i, rc = 0;
1737
1738         if (!inode)
1739                 return 0;
1740
1741         qid[USRQUOTA] = inode->i_uid;
1742         qid[GRPQUOTA] = inode->i_gid;
1743
1744         if (S_ISDIR(inode->i_mode) ||
1745             S_ISREG(inode->i_mode) ||
1746             S_ISLNK(inode->i_mode))
1747                 size = inode_get_bytes(inode);
1748
1749         for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1750                 if (!Q_TYPESET(oqc, cnt))
1751                         continue;
1752
1753                 cdqb[cnt] = cqget(inode->i_sb, qctxt->qckt_hash,
1754                                 &qctxt->qckt_list, qid[cnt], cnt,
1755                                 qctxt->qckt_first_check[cnt]);
1756                 if (!cdqb[cnt]) {
1757                         rc = -ENOMEM;
1758                         break;
1759                 }
1760
1761                 cdqb[cnt]->dqb_curspace += size;
1762                 cdqb[cnt]->dqb_curinodes++;
1763         }
1764
1765         if (rc) {
1766                 for (i = 0; i < cnt; i++) {
1767                         if (!Q_TYPESET(oqc, i))
1768                                 continue;
1769                         LASSERT(cdqb[i]);
1770                         cdqb[i]->dqb_curspace -= size;
1771                         cdqb[i]->dqb_curinodes--;
1772                 }
1773         }
1774
1775         return rc;
1776 }
1777
1778 /* write dqinfo struct in a new quota file */
1779 static int v3_write_dqinfo(struct file *f, int type, struct if_dqinfo *info)
1780 {
1781         struct v2_disk_dqinfo dqinfo;
1782         __u32 blocks = V2_DQTREEOFF + 1;
1783         loff_t offset = V2_DQINFOOFF;
1784
1785         if (info) {
1786                 dqinfo.dqi_bgrace = cpu_to_le32(info->dqi_bgrace);
1787                 dqinfo.dqi_igrace = cpu_to_le32(info->dqi_igrace);
1788                 dqinfo.dqi_flags = cpu_to_le32(info->dqi_flags & DQF_MASK &
1789                                                ~DQF_INFO_DIRTY);
1790         } else {
1791                 dqinfo.dqi_bgrace = cpu_to_le32(MAX_DQ_TIME);
1792                 dqinfo.dqi_igrace = cpu_to_le32(MAX_IQ_TIME);
1793                 dqinfo.dqi_flags = 0;
1794         }
1795
1796         dqinfo.dqi_blocks = cpu_to_le32(blocks);
1797         dqinfo.dqi_free_blk = 0;
1798         dqinfo.dqi_free_entry = 0;
1799
1800         return cfs_user_write(f, (char *)&dqinfo, sizeof(dqinfo), &offset);
1801 }
1802
1803 static int v3_write_dqheader(struct file *f, int type)
1804 {
1805         static const __u32 quota_magics[] = V2_INITQMAGICS;
1806         static const __u32 quota_versions[] = LUSTRE_INITQVERSIONS_V2;
1807         struct v2_disk_dqheader dqhead;
1808         loff_t offset = 0;
1809
1810         CLASSERT(ARRAY_SIZE(quota_magics) == ARRAY_SIZE(quota_versions));
1811         LASSERT(0 <= type && type < ARRAY_SIZE(quota_magics));
1812
1813         dqhead.dqh_magic = cpu_to_le32(quota_magics[type]);
1814         dqhead.dqh_version = cpu_to_le32(quota_versions[type]);
1815
1816         return cfs_user_write(f, (char *)&dqhead, sizeof(dqhead), &offset);
1817 }
1818
1819 static int create_new_quota_files(struct qchk_ctxt *qctxt,
1820                                   struct obd_quotactl *oqc)
1821 {
1822         int i, rc = 0;
1823         ENTRY;
1824
1825         for (i = 0; i < MAXQUOTAS; i++) {
1826                 struct if_dqinfo *info = qctxt->qckt_first_check[i]?
1827                                          NULL : &qctxt->qckt_dqinfo[i];
1828                 struct file *file;
1829                 const char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1830
1831                 if (!Q_TYPESET(oqc, i))
1832                         continue;
1833
1834                 LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1835
1836                 file = filp_open(name[i], O_RDWR | O_CREAT | O_TRUNC, 0644);
1837                 if (IS_ERR(file)) {
1838                         rc = PTR_ERR(file);
1839                         CERROR("can't create %s file: rc = %d\n",
1840                                name[i], rc);
1841                         GOTO(out, rc);
1842                 }
1843
1844                 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
1845                         CERROR("file %s is not regular", name[i]);
1846                         filp_close(file, 0);
1847                         GOTO(out, rc = -EINVAL);
1848                 }
1849
1850                 ll_vfs_dq_drop(file->f_dentry->d_inode);
1851
1852                 rc = v3_write_dqheader(file, i);
1853                 if (rc) {
1854                         filp_close(file, 0);
1855                         GOTO(out, rc);
1856                 }
1857
1858                 rc = v3_write_dqinfo(file, i, info);
1859                 filp_close(file, 0);
1860                 if (rc)
1861                         GOTO(out, rc);
1862         }
1863
1864 out:
1865         RETURN(rc);
1866 }
1867
1868
1869 static int commit_chkquot(struct super_block *sb, struct qchk_ctxt *qctxt,
1870                           struct chk_dqblk *cdqb)
1871 {
1872         struct obd_quotactl *oqc;
1873         long now;
1874         int rc;
1875         ENTRY;
1876
1877         OBD_ALLOC_PTR(oqc);
1878         if (!oqc)
1879                 RETURN(-ENOMEM);
1880
1881         now = cfs_time_current_sec();
1882
1883         if (cdqb->dqb_bsoftlimit &&
1884             toqb(cdqb->dqb_curspace) >= cdqb->dqb_bsoftlimit &&
1885             !cdqb->dqb_btime)
1886                 cdqb->dqb_btime =
1887                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_bgrace;
1888
1889         if (cdqb->dqb_isoftlimit &&
1890             cdqb->dqb_curinodes >= cdqb->dqb_isoftlimit &&
1891             !cdqb->dqb_itime)
1892                 cdqb->dqb_itime =
1893                         now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_igrace;
1894
1895         cdqb->dqb_valid = QIF_ALL;
1896
1897         oqc->qc_cmd = Q_SETQUOTA;
1898         oqc->qc_type = cdqb->dqb_type;
1899         oqc->qc_id = cdqb->dqb_id;
1900         DQBLK_COPY(&oqc->qc_dqblk, cdqb);
1901
1902         rc = fsfilt_ext3_quotactl(sb, oqc);
1903         OBD_FREE_PTR(oqc);
1904         RETURN(rc);
1905 }
1906
1907 static int prune_chkquots(struct super_block *sb,
1908                           struct qchk_ctxt *qctxt, int error)
1909 {
1910         struct chk_dqblk *cdqb, *tmp;
1911         int rc;
1912
1913         cfs_list_for_each_entry_safe(cdqb, tmp, &qctxt->qckt_list, dqb_list) {
1914                 if (!error) {
1915                         rc = commit_chkquot(sb, qctxt, cdqb);
1916                         if (rc)
1917                                 error = rc;
1918                 }
1919                 cfs_hlist_del_init(&cdqb->dqb_hash);
1920                 cfs_list_del(&cdqb->dqb_list);
1921                 OBD_FREE_PTR(cdqb);
1922         }
1923
1924         return error;
1925 }
1926
1927 #ifndef EXT3_FEATURE_RO_COMPAT_GDT_CSUM
1928 #define EXT3_FEATURE_RO_COMPAT_GDT_CSUM 0x0010
1929 #endif
1930
1931 static int fsfilt_ext3_quotacheck(struct super_block *sb,
1932                                   struct obd_quotactl *oqc)
1933 {
1934         struct ext3_sb_info *sbi = EXT3_SB(sb);
1935         int i, group, uninit_feat = 0;
1936         struct qchk_ctxt *qctxt;
1937         struct buffer_head *bitmap_bh = NULL;
1938         unsigned long ino, inode_inuse;
1939         struct inode *inode;
1940         int rc = 0;
1941         ENTRY;
1942
1943         /* turn on quota and read dqinfo if existed */
1944         OBD_ALLOC_PTR(qctxt);
1945         if (!qctxt) {
1946                 oqc->qc_stat = -ENOMEM;
1947                 RETURN(-ENOMEM);
1948         }
1949
1950         for (i = 0; i < NR_DQHASH; i++)
1951                 CFS_INIT_HLIST_HEAD(&qctxt->qckt_hash[i]);
1952         CFS_INIT_LIST_HEAD(&qctxt->qckt_list);
1953
1954         for (i = 0; i < MAXQUOTAS; i++) {
1955                 if (!Q_TYPESET(oqc, i))
1956                         continue;
1957
1958                 rc = quota_onoff(sb, Q_QUOTAON, i, oqc->qc_id);
1959                 if (!rc || rc == -EBUSY) {
1960                         rc = read_old_dqinfo(sb, i, qctxt->qckt_dqinfo);
1961                         if (rc)
1962                                 GOTO(out, rc);
1963                 } else if (rc == -ENOENT || rc == -EINVAL || rc == -EEXIST) {
1964                         qctxt->qckt_first_check[i] = 1;
1965                 } else if (rc) {
1966                         GOTO(out, rc);
1967                 }
1968         }
1969         if (EXT3_HAS_RO_COMPAT_FEATURE(sb, EXT3_FEATURE_RO_COMPAT_GDT_CSUM))
1970                 /* This filesystem supports the uninit group feature */
1971                 uninit_feat = 1;
1972
1973         /* number of inodes that have been allocated */
1974         inode_inuse = sbi->s_inodes_per_group * sbi->s_groups_count -
1975                       percpu_counter_sum(&sbi->s_freeinodes_counter);
1976
1977         /* check quota and update in hash */
1978         for (group = 0; group < sbi->s_groups_count && inode_inuse > 0;
1979              group++) {
1980                 unsigned long used_count = sbi->s_inodes_per_group;
1981
1982                 if (uninit_feat) {
1983                         struct ext3_group_desc *desc;
1984                         desc = ext3_get_group_desc(sb, group, NULL);
1985                         if (!desc)
1986                                 GOTO(out, -EIO);
1987
1988                         /* we don't really need to take the group lock here,
1989                          * but it may be useful if one day we support online
1990                          * quotacheck */
1991                         ext4_lock_group(sb, group);
1992                         if (desc->bg_flags & cpu_to_le16(EXT3_BG_INODE_UNINIT)) {
1993                                 /* no inode in use in this group, just skip it */
1994                                 ext3_unlock_group(sb, group);
1995                                 continue;
1996                         }
1997
1998                         used_count -= ext3_itable_unused_count(sb, desc);
1999                         ext3_unlock_group(sb, group);
2000                 }
2001
2002                 ino = group * sbi->s_inodes_per_group + 1;
2003                 bitmap_bh = ext3_read_inode_bitmap(sb, group);
2004                 if (!bitmap_bh) {
2005                         CERROR("%s: ext3_read_inode_bitmap group %d failed\n",
2006                                sb->s_id, group);
2007                         GOTO(out, -EIO);
2008                 }
2009
2010                 i = 0;
2011                 while (i < used_count &&
2012                        (i = ext3_find_next_bit(bitmap_bh->b_data,
2013                                                used_count, i)) < used_count) {
2014                         inode_inuse--;
2015                         i++;
2016                         ino = i + group * sbi->s_inodes_per_group;
2017                         if (ino < sbi->s_first_ino)
2018                                 continue;
2019                         inode = ext3_iget(sb, ino);
2020                         if (!inode || IS_ERR(inode))
2021                                 continue;
2022
2023                         rc = add_inode_quota(inode, qctxt, oqc);
2024                         iput(inode);
2025                         if (rc) {
2026                                 brelse(bitmap_bh);
2027                                 GOTO(out, rc);
2028                         }
2029                 }
2030                 brelse(bitmap_bh);
2031         }
2032
2033         /* read old quota limits from old quota file. (only for the user
2034          * has limits but hasn't file) */
2035 #ifdef HAVE_QUOTA_SUPPORT
2036         for (i = 0; i < MAXQUOTAS; i++) {
2037                 cfs_list_t id_list;
2038                 struct dquot_id *dqid, *tmp;
2039
2040                 if (!Q_TYPESET(oqc, i))
2041                         continue;
2042
2043                 if (qctxt->qckt_first_check[i])
2044                         continue;
2045
2046
2047                 LASSERT(sb_dqopt(sb)->files[i] != NULL);
2048                 CFS_INIT_LIST_HEAD(&id_list);
2049 #ifndef KERNEL_SUPPORTS_QUOTA_READ
2050                 rc = lustre_get_qids(sb_dqopt(sb)->files[i], NULL, i, &id_list);
2051 #else
2052                 rc = lustre_get_qids(NULL, sb_dqopt(sb)->files[i], i, &id_list);
2053 #endif
2054                 if (rc)
2055                         CERROR("read old limits failed. (rc:%d)\n", rc);
2056
2057                 cfs_list_for_each_entry_safe(dqid, tmp, &id_list, di_link) {
2058                         cfs_list_del_init(&dqid->di_link);
2059
2060                         if (!rc)
2061                                 cqget(sb, qctxt->qckt_hash, &qctxt->qckt_list,
2062                                       dqid->di_id, i,
2063                                       qctxt->qckt_first_check[i]);
2064                         OBD_FREE_PTR(dqid);
2065                 }
2066         }
2067 #endif
2068         /* turn off quota cause we are to dump chk_dqblk to files */
2069         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type, oqc->qc_id);
2070
2071         rc = create_new_quota_files(qctxt, oqc);
2072         if (rc)
2073                 GOTO(out, rc);
2074
2075         /* we use vfs functions to set dqblk, so turn quota on */
2076         rc = quota_onoff(sb, Q_QUOTAON, oqc->qc_type, oqc->qc_id);
2077 out:
2078         /* dump and free chk_dqblk */
2079         rc = prune_chkquots(sb, qctxt, rc);
2080         OBD_FREE_PTR(qctxt);
2081
2082         /* turn off quota, `lfs quotacheck` will turn on when all
2083          * nodes quotacheck finish. */
2084         quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type, oqc->qc_id);
2085
2086         oqc->qc_stat = rc;
2087         if (rc)
2088                 CERROR("quotacheck failed: rc = %d\n", rc);
2089
2090         RETURN(rc);
2091 }
2092
2093 static int fsfilt_ext3_quotainfo(struct lustre_quota_info *lqi, int type,
2094                                  int cmd)
2095 {
2096         int rc = 0;
2097         ENTRY;
2098
2099         if (lqi->qi_files[type] == NULL) {
2100                 CERROR("operate qinfo before it's enabled!\n");
2101                 RETURN(-ESRCH);
2102         }
2103
2104         switch (cmd) {
2105         case QFILE_CHK:
2106                 rc = lustre_check_quota_file(lqi, type);
2107                 break;
2108         case QFILE_RD_INFO:
2109                 rc = lustre_read_quota_info(lqi, type);
2110                 break;
2111         case QFILE_WR_INFO:
2112                 rc = lustre_write_quota_info(lqi, type);
2113                 break;
2114         case QFILE_INIT_INFO:
2115                 rc = lustre_init_quota_info(lqi, type);
2116                 break;
2117         case QFILE_CONVERT:
2118                 rc = -ENOTSUPP;
2119                 CERROR("quota CONVERT command is not supported\n");
2120                 break;
2121         default:
2122                 rc = -ENOTSUPP;
2123                 CERROR("Unsupported admin quota file cmd %d\n"
2124                        "Are lquota.ko and fsfilt_ldiskfs.ko modules in sync?\n",
2125                        cmd);
2126                 break;
2127         }
2128         RETURN(rc);
2129 }
2130
2131 static int fsfilt_ext3_qids(struct file *file, struct inode *inode, int type,
2132                             cfs_list_t *list)
2133 {
2134         return lustre_get_qids(file, inode, type, list);
2135 }
2136
2137 static int fsfilt_ext3_dquot(struct lustre_dquot *dquot, int cmd)
2138 {
2139         int rc = 0;
2140         ENTRY;
2141
2142         if (dquot->dq_info->qi_files[dquot->dq_type] == NULL) {
2143                 CERROR("operate dquot before it's enabled!\n");
2144                 RETURN(-ESRCH);
2145         }
2146
2147         switch (cmd) {
2148         case QFILE_RD_DQUOT:
2149                 rc = lustre_read_dquot(dquot);
2150                 break;
2151         case QFILE_WR_DQUOT:
2152                 if (dquot->dq_dqb.dqb_ihardlimit ||
2153                     dquot->dq_dqb.dqb_isoftlimit ||
2154                     dquot->dq_dqb.dqb_bhardlimit ||
2155                     dquot->dq_dqb.dqb_bsoftlimit)
2156                         cfs_clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2157                 else
2158                         cfs_set_bit(DQ_FAKE_B, &dquot->dq_flags);
2159
2160                 rc = lustre_commit_dquot(dquot);
2161                 if (rc >= 0)
2162                         rc = 0;
2163                 break;
2164         default:
2165                 CERROR("Unsupported admin quota file cmd %d\n", cmd);
2166                 LBUG();
2167                 break;
2168         }
2169         RETURN(rc);
2170 }
2171
2172 static int fsfilt_ext3_get_mblk(struct super_block *sb, int *count,
2173                                 struct inode *inode, int frags)
2174 {
2175 #ifdef EXT3_EXT_HAS_NO_TREE
2176         struct ext3_ext_base *base = inode;
2177 #else
2178         struct ext3_extents_tree tree;
2179         struct ext3_ext_base *base = &tree;
2180
2181         ext3_init_tree_desc(base, inode);
2182 #endif
2183         /* for an ost_write request, it needs <#fragments> * <tree depth + 1>
2184          * metablocks at maxium b=16542 */
2185         *count = frags * (EXT_DEPTH(base) + 1) * EXT3_BLOCK_SIZE(sb);
2186         return 0;
2187 }
2188
2189 #endif
2190
2191 lvfs_sbdev_type fsfilt_ext3_journal_sbdev(struct super_block *sb)
2192 {
2193         return (EXT3_SB(sb)->journal_bdev);
2194 }
2195 EXPORT_SYMBOL(fsfilt_ext3_journal_sbdev);
2196
2197 ssize_t lustre_read_quota(struct file *f, struct inode *inode, int type,
2198                           char *buf, int count, loff_t pos)
2199 {
2200         loff_t p = pos;
2201         int rc;
2202
2203         if (!f && !inode) {
2204                 CERROR("lustre_read_quota failed for no quota file!\n");
2205                 libcfs_debug_dumpstack(NULL);
2206                 return -EINVAL;
2207         }
2208
2209         /* Support for both adm and op quota files must be provided */
2210         if (f) {
2211                 rc = fsfilt_ext3_read_record(f, buf, count, &p);
2212                 rc = rc < 0 ? rc : p - pos;
2213         } else {
2214                 struct super_block *sb = inode->i_sb;
2215                 rc = sb->s_op->quota_read(sb, type, buf, count, pos);
2216         }
2217         return rc;
2218 }
2219
2220 ssize_t lustre_write_quota(struct file *f, char *buf, int count, loff_t pos)
2221 {
2222         loff_t p = pos;
2223         int rc;
2224
2225         /* Only adm quota files are supported, op updates are handled by vfs */
2226         rc = fsfilt_ext3_write_record(f, buf, count, &p, 0);
2227         rc = rc < 0 ? rc : p - pos;
2228
2229         return rc;
2230 }
2231
2232 void *lustre_quota_journal_start(struct inode *inode, int delete)
2233 {
2234         handle_t *handle;
2235         unsigned block_count;
2236
2237         if (delete) {
2238                 /* each indirect block (+4) may become free, attaching to the
2239                  * header list of free blocks (+1); the data block (+1) may
2240                  * become a free block (+0) or a block with free dqentries (+0) */
2241                 block_count = (4 + 1) + 1;
2242                 handle = ext3_journal_start(inode,
2243                             block_count*FSFILT_DATA_TRANS_BLOCKS(inode->i_sb)+2);
2244         } else {
2245                 /* indirect blocks are touched (+4), each causes file expansion (+0) or
2246                  * freeblk reusage with a header update (+1); dqentry is either reused
2247                  * causing update of the entry block (+1), prev (+1) and next (+1) or
2248                  * a new block allocation (+1) with a header update (+1)              */
2249                 block_count = (4 + 1) + 3;
2250                 handle = ext3_journal_start(inode,
2251                              block_count*FSFILT_DATA_TRANS_BLOCKS(inode->i_sb)+2);
2252
2253         }
2254
2255         return handle;
2256 }
2257
2258 void lustre_quota_journal_stop(void *handle)
2259 {
2260         ext3_journal_stop((handle_t *)handle);
2261 }
2262
2263 static int ll_decode_fh_accept(void *context, struct dentry *de)
2264 {
2265         return 1;
2266 }
2267
2268 #ifdef HAVE_EXPORTFS_DECODE_FH
2269 # define ll_exportfs_decode_fh(mnt, fid, len, type, acceptable, context) \
2270          exportfs_decode_fh(mnt, (struct fid*)(fid), len, type,          \
2271                             acceptable, context)
2272 #else
2273 # define ll_exportfs_decode_fh(mnt, fid, len, type, acceptable, context) \
2274          export_op_default.decode_fh((mnt)->mnt_sb, &(fid)->ino, len,    \
2275                                      type, acceptable, context)
2276 # define FILEID_INO32_GEN 1
2277 extern struct export_operations export_op_default;
2278 #endif
2279
2280 struct dentry *fsfilt_ext3_fid2dentry(struct vfsmount *mnt,
2281                                       struct fsfilt_fid *fid, int ignore_gen)
2282 {
2283         struct inode  *inode;
2284         struct dentry *result;
2285         
2286         result = ll_exportfs_decode_fh(mnt, fid, 2, FILEID_INO32_GEN,
2287                                        ll_decode_fh_accept, NULL);
2288         if (IS_ERR(result)) {
2289                 CDEBUG(D_DENTRY, "%s of %u/%u failed %ld\n", __func__,
2290                        fid->ino, fid->gen, PTR_ERR(result));
2291                 return result;
2292         }
2293
2294         CDEBUG(D_DENTRY, "%s of %u/%u succeeded\n", __func__,
2295                fid->ino, fid->gen);
2296         inode = result->d_inode;
2297         if (inode == NULL)
2298                 goto err_out;
2299
2300         if (inode->i_nlink == 0 &&
2301             inode->i_mode == 0 && LTIME_S(inode->i_ctime) == 0) {
2302                 LCONSOLE_WARN("Found inode with zero nlink, mode and"
2303                               " ctime -- this may indicate disk "
2304                               "corruption (inode: %lu, link: %lu, "
2305                               "count: %d)\n", inode->i_ino,
2306                               (unsigned long)inode->i_nlink,
2307                               atomic_read(&inode->i_count));
2308                 goto err_out;
2309         }
2310         if (fid->gen && inode->i_generation != fid->gen) {
2311                 /* we didn't find the right inode.. */
2312                 CDEBUG(D_INODE, "found wrong generation: inode %lu, link: %lu, "
2313                        "count: %d, generation %u/%u\n",
2314                        inode->i_ino, (unsigned long)inode->i_nlink,
2315                        atomic_read(&inode->i_count), inode->i_generation,
2316                        fid->gen);
2317                 goto err_out;
2318         }
2319
2320         return result;
2321 err_out:
2322         l_dput(result);
2323         return ERR_PTR(-ENOENT);
2324 }
2325
2326 static struct fsfilt_operations fsfilt_ext3_ops = {
2327         .fs_type                = "ext3",
2328         .fs_owner               = THIS_MODULE,
2329         .fs_getlabel            = fsfilt_ext3_get_label,
2330         .fs_setlabel            = fsfilt_ext3_set_label,
2331         .fs_uuid                = fsfilt_ext3_uuid,
2332         .fs_start               = fsfilt_ext3_start,
2333         .fs_brw_start           = fsfilt_ext3_brw_start,
2334         .fs_extend              = fsfilt_ext3_extend,
2335         .fs_commit              = fsfilt_ext3_commit,
2336         .fs_commit_async        = fsfilt_ext3_commit_async,
2337         .fs_commit_wait         = fsfilt_ext3_commit_wait,
2338         .fs_setattr             = fsfilt_ext3_setattr,
2339         .fs_iocontrol           = fsfilt_ext3_iocontrol,
2340         .fs_set_md              = fsfilt_ext3_set_md,
2341         .fs_get_md              = fsfilt_ext3_get_md,
2342         .fs_readpage            = fsfilt_ext3_readpage,
2343         .fs_add_journal_cb      = fsfilt_ext3_add_journal_cb,
2344         .fs_statfs              = fsfilt_ext3_statfs,
2345         .fs_sync                = fsfilt_ext3_sync,
2346         .fs_map_inode_pages     = fsfilt_ext3_map_inode_pages,
2347         .fs_write_record        = fsfilt_ext3_write_record,
2348         .fs_read_record         = fsfilt_ext3_read_record,
2349         .fs_setup               = fsfilt_ext3_setup,
2350         .fs_send_bio            = fsfilt_ext3_send_bio,
2351         .fs_get_op_len          = fsfilt_ext3_get_op_len,
2352 #ifdef HAVE_DISK_INODE_VERSION
2353         .fs_get_version         = fsfilt_ext3_get_version,
2354         .fs_set_version         = fsfilt_ext3_set_version,
2355 #endif
2356 #ifdef HAVE_QUOTA_SUPPORT
2357         .fs_quotactl            = fsfilt_ext3_quotactl,
2358         .fs_quotacheck          = fsfilt_ext3_quotacheck,
2359         .fs_quotainfo           = fsfilt_ext3_quotainfo,
2360         .fs_qids                = fsfilt_ext3_qids,
2361         .fs_dquot               = fsfilt_ext3_dquot,
2362         .fs_get_mblk            = fsfilt_ext3_get_mblk,
2363 #endif
2364         .fs_journal_sbdev       = fsfilt_ext3_journal_sbdev,
2365         .fs_fid2dentry          = fsfilt_ext3_fid2dentry,
2366 };
2367
2368 static int __init fsfilt_ext3_init(void)
2369 {
2370         int rc;
2371
2372         fcb_cache = cfs_mem_cache_create("fsfilt_ext3_fcb",
2373                                          sizeof(struct fsfilt_cb_data), 0, 0);
2374         if (!fcb_cache) {
2375                 CERROR("error allocating fsfilt journal callback cache\n");
2376                 GOTO(out, rc = -ENOMEM);
2377         }
2378
2379         rc = fsfilt_register_ops(&fsfilt_ext3_ops);
2380
2381         if (rc) {
2382                 int err = cfs_mem_cache_destroy(fcb_cache);
2383                 LASSERTF(err == 0, "error destroying new cache: rc %d\n", err);
2384         }
2385 out:
2386         return rc;
2387 }
2388
2389 static void __exit fsfilt_ext3_exit(void)
2390 {
2391         int rc;
2392
2393         fsfilt_unregister_ops(&fsfilt_ext3_ops);
2394         rc = cfs_mem_cache_destroy(fcb_cache);
2395         LASSERTF(rc == 0, "couldn't destroy fcb_cache slab\n");
2396 }
2397
2398 module_init(fsfilt_ext3_init);
2399 module_exit(fsfilt_ext3_exit);
2400
2401 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2402 MODULE_DESCRIPTION("Lustre ext3 Filesystem Helper v0.1");
2403 MODULE_LICENSE("GPL");