4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
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.
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).
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
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
27 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, Whamcloud, Inc.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/lvfs/fsfilt_ext3.c
38 * Author: Andreas Dilger <adilger@clusterfs.com>
41 #define DEBUG_SUBSYSTEM S_FILTER
43 #include <linux/init.h>
44 #include <linux/module.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>
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>
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
71 #define QFMT_LUSTRE QFMT_VFS_V1
73 #define QFMT_LUSTRE QFMT_VFS_V0
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);
86 #include <libcfs/libcfs.h>
87 #include <lustre_fsfilt.h>
89 #include <lustre_quota.h>
90 #include <linux/lustre_compat25.h>
91 #include <linux/lprocfs_status.h>
93 #include <ext4/ext4_extents.h>
95 #include "lustre_quota_fmt.h"
97 /* for kernels 2.6.18 and later */
98 #define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS(sb)
100 #define fsfilt_ext3_ext_insert_extent(handle, inode, path, newext, flag) \
101 ext3_ext_insert_extent(handle, inode, path, newext, flag)
103 #define ext3_mb_discard_inode_preallocations(inode) \
104 ext3_discard_preallocations(inode)
106 #define fsfilt_log_start_commit(journal, tid) jbd2_log_start_commit(journal, tid)
107 #define fsfilt_log_wait_commit(journal, tid) jbd2_log_wait_commit(journal, tid)
109 #ifdef HAVE_EXT4_JOURNAL_CALLBACK_ADD
110 # define journal_callback ext4_journal_cb_entry
111 # define fsfilt_journal_callback_set(handle, func, jcb) \
112 ext4_journal_callback_add(handle, func, jcb)
113 #elif defined(HAVE_JBD2_JOURNAL_CALLBACK_SET)
114 # define fsfilt_journal_callback_set(handle, func, jcb) \
115 jbd2_journal_callback_set(handle, func, jcb)
116 #elif defined(HAVE_JOURNAL_CALLBACK_SET)
117 # define fsfilt_journal_callback_set(handle, func, jcb) \
118 journal_callback_set(handle, func, jcb)
120 # error missing journal commit callback
121 #endif /* HAVE_EXT4_JOURNAL_CALLBACK_ADD */
123 static cfs_mem_cache_t *fcb_cache;
125 struct fsfilt_cb_data {
126 struct journal_callback cb_jcb; /* jbd private data - MUST BE FIRST */
127 fsfilt_cb_t cb_func; /* MDS/OBD completion function */
128 struct obd_device *cb_obd; /* MDS/OBD completion device */
129 __u64 cb_last_rcvd; /* MDS/OST last committed operation */
130 void *cb_data; /* MDS/OST completion function data */
133 #ifndef ext3_find_next_bit
134 #define ext3_find_next_bit ext2_find_next_bit
137 #ifndef ext2_find_next_bit
138 #ifdef __LITTLE_ENDIAN
139 #define ext2_find_next_bit(addr, size, off) find_next_bit((unsigned long *)(addr), (size), (off))
141 error "Need implementation of find_next_bit on big-endian systems"
142 #endif /* __LITTLE_ENDIAN */
143 #endif /* !ext2_find_next_le_bit */
145 static char *fsfilt_ext3_get_label(struct super_block *sb)
147 return EXT3_SB(sb)->s_es->s_volume_name;
150 static int fsfilt_ext3_set_label(struct super_block *sb, char *label)
152 /* see e.g. fsfilt_ext3_write_record() */
157 journal = EXT3_SB(sb)->s_journal;
158 handle = ext3_journal_start_sb(sb, 1);
159 if (IS_ERR(handle)) {
160 CERROR("can't start transaction\n");
161 return(PTR_ERR(handle));
164 err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
168 memcpy(EXT3_SB(sb)->s_es->s_volume_name, label,
169 sizeof(EXT3_SB(sb)->s_es->s_volume_name));
171 err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
174 ext3_journal_stop(handle);
179 static char *fsfilt_ext3_uuid(struct super_block *sb)
181 return EXT3_SB(sb)->s_es->s_uuid;
184 #ifdef HAVE_DISK_INODE_VERSION
186 static __u64 get_i_version(struct inode *inode)
188 return EXT3_I(inode)->i_fs_version;
191 static void set_i_version(struct inode *inode, __u64 new_version)
193 (EXT3_I(inode))->i_fs_version = new_version;
197 * Get the 64-bit version for an inode.
199 static __u64 fsfilt_ext3_get_version(struct inode *inode)
201 CDEBUG(D_INFO, "Get version "LPX64" for inode %lu\n",
202 get_i_version(inode), inode->i_ino);
203 return get_i_version(inode);
207 * Set the 64-bit version and return the old version.
209 static __u64 fsfilt_ext3_set_version(struct inode *inode, __u64 new_version)
211 __u64 old_version = get_i_version(inode);
213 CDEBUG(D_INFO, "Set version "LPX64" (old "LPX64") for inode %lu\n",
214 new_version, old_version, inode->i_ino);
215 set_i_version(inode, new_version);
216 /* version is set after all inode operations are finished, so we should
217 * mark it dirty here */
218 inode->i_sb->s_op->dirty_inode(inode);
225 * We don't currently need any additional blocks for rmdir and
226 * unlink transactions because we are storing the OST oa_id inside
227 * the inode (which we will be changing anyways as part of this
230 static void *fsfilt_ext3_start(struct inode *inode, int op, void *desc_private,
233 /* For updates to the last received file */
234 int nblocks = FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
238 if (current->journal_info) {
239 CDEBUG(D_INODE, "increasing refcount on %p\n",
240 current->journal_info);
245 case FSFILT_OP_RMDIR:
246 case FSFILT_OP_UNLINK:
247 /* delete one file + create/update logs for each stripe */
248 nblocks += EXT3_DELETE_TRANS_BLOCKS(inode->i_sb);
249 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
250 FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
252 case FSFILT_OP_RENAME:
253 /* modify additional directory */
254 nblocks += FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
256 case FSFILT_OP_SYMLINK:
257 /* additional block + block bitmap + GDT for long symlink */
260 case FSFILT_OP_CREATE: {
263 case FSFILT_OP_MKDIR:
264 case FSFILT_OP_MKNOD:
265 /* modify one inode + block bitmap + GDT */
269 /* modify parent directory */
270 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
271 EXT3_DATA_TRANS_BLOCKS(inode->i_sb);
272 /* create/update logs for each stripe */
273 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
274 FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
276 case FSFILT_OP_SETATTR:
277 /* Setattr on inode */
279 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
280 EXT3_DATA_TRANS_BLOCKS(inode->i_sb);
281 /* quota chown log for each stripe */
282 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
283 FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
285 case FSFILT_OP_CANCEL_UNLINK:
286 /* blocks for log header bitmap update OR
287 * blocks for catalog header bitmap update + unlink of logs */
288 nblocks = (LLOG_CHUNK_SIZE >> inode->i_blkbits) +
289 EXT3_DELETE_TRANS_BLOCKS(inode->i_sb) * logs;
291 default: CERROR("unknown transaction start op %d\n", op);
295 LASSERT(current->journal_info == desc_private);
296 journal = EXT3_SB(inode->i_sb)->s_journal;
297 if (nblocks > journal->j_max_transaction_buffers) {
298 CWARN("too many credits %d for op %ux%u using %d instead\n",
299 nblocks, op, logs, journal->j_max_transaction_buffers);
300 nblocks = journal->j_max_transaction_buffers;
304 LASSERTF(nblocks > 0, "can't start %d credit transaction\n", nblocks);
305 handle = ext3_journal_start(inode, nblocks);
308 LASSERT(current->journal_info == handle);
310 CERROR("error starting handle for op %u (%u credits): rc %ld\n",
311 op, nblocks, PTR_ERR(handle));
316 * Calculate the number of buffer credits needed to write multiple pages in
317 * a single ext3 transaction. No, this shouldn't be here, but as yet ext3
318 * doesn't have a nice API for calculating this sort of thing in advance.
320 * See comment above ext3_writepage_trans_blocks for details. We assume
321 * no data journaling is being done, but it does allow for all of the pages
322 * being non-contiguous. If we are guaranteed contiguous pages we could
323 * reduce the number of (d)indirect blocks a lot.
325 * With N blocks per page and P pages, for each inode we have at most:
327 * min(N*P, blocksize/4 + 1) dindirect blocks
330 * For the entire filesystem, we have at most:
331 * min(sum(nindir + P), ngroups) bitmap blocks (from the above)
332 * min(sum(nindir + P), gdblocks) group descriptor blocks (from the above)
333 * objcount inode blocks
335 * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quota files
337 * 1 EXT3_DATA_TRANS_BLOCKS for the last_rcvd update.
339 static int fsfilt_ext3_credits_needed(int objcount, struct fsfilt_objinfo *fso,
340 int niocount, struct niobuf_local *nb)
342 struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
344 const int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
345 int nbitmaps = 0, ngdblocks;
346 int needed = objcount + 1; /* inodes + superblock */
349 for (i = 0, j = 0; i < objcount; i++, fso++) {
350 /* two or more dindirect blocks in case we cross boundary */
351 int ndind = (long)((nb[j + fso->fso_bufcnt - 1].offset -
353 sb->s_blocksize_bits) /
354 (EXT3_ADDR_PER_BLOCK(sb) * EXT3_ADDR_PER_BLOCK(sb));
355 nbitmaps += min(fso->fso_bufcnt, ndind > 0 ? ndind : 2);
357 /* leaf, indirect, tindirect blocks for first block */
358 nbitmaps += blockpp + 2;
360 j += fso->fso_bufcnt;
363 next_indir = nb[0].offset +
364 (EXT3_ADDR_PER_BLOCK(sb) << sb->s_blocksize_bits);
365 for (i = 1; i < niocount; i++) {
366 if (nb[i].offset >= next_indir) {
367 nbitmaps++; /* additional indirect */
368 next_indir = nb[i].offset +
369 (EXT3_ADDR_PER_BLOCK(sb)<<sb->s_blocksize_bits);
370 } else if (nb[i].offset != nb[i - 1].offset + sb->s_blocksize) {
371 nbitmaps++; /* additional indirect */
373 nbitmaps += blockpp; /* each leaf in different group? */
376 ngdblocks = nbitmaps;
377 if (nbitmaps > EXT3_SB(sb)->s_groups_count)
378 nbitmaps = EXT3_SB(sb)->s_groups_count;
379 if (ngdblocks > EXT3_SB(sb)->s_gdb_count)
380 ngdblocks = EXT3_SB(sb)->s_gdb_count;
382 needed += nbitmaps + ngdblocks;
384 /* last_rcvd update */
385 needed += EXT3_DATA_TRANS_BLOCKS(sb);
387 #if defined(CONFIG_QUOTA)
388 /* We assume that there will be 1 bit set in s_dquot.flags for each
389 * quota file that is active. This is at least true for now.
391 needed += hweight32(ll_sb_any_quota_active(sb)) *
392 FSFILT_SINGLEDATA_TRANS_BLOCKS(sb);
398 /* We have to start a huge journal transaction here to hold all of the
399 * metadata for the pages being written here. This is necessitated by
400 * the fact that we do lots of prepare_write operations before we do
401 * any of the matching commit_write operations, so even if we split
402 * up to use "smaller" transactions none of them could complete until
403 * all of them were opened. By having a single journal transaction,
404 * we eliminate duplicate reservations for common blocks like the
405 * superblock and group descriptors or bitmaps.
407 * We will start the transaction here, but each prepare_write will
408 * add a refcount to the transaction, and each commit_write will
409 * remove a refcount. The transaction will be closed when all of
410 * the pages have been written.
412 static void *fsfilt_ext3_brw_start(int objcount, struct fsfilt_objinfo *fso,
413 int niocount, struct niobuf_local *nb,
414 void *desc_private, int logs)
421 LASSERT(current->journal_info == desc_private);
422 journal = EXT3_SB(fso->fso_dentry->d_inode->i_sb)->s_journal;
423 needed = fsfilt_ext3_credits_needed(objcount, fso, niocount, nb);
425 /* The number of blocks we could _possibly_ dirty can very large.
426 * We reduce our request if it is absurd (and we couldn't get that
427 * many credits for a single handle anyways).
429 * At some point we have to limit the size of I/Os sent at one time,
430 * increase the size of the journal, or we have to calculate the
431 * actual journal requirements more carefully by checking all of
432 * the blocks instead of being maximally pessimistic. It remains to
433 * be seen if this is a real problem or not.
435 if (needed > journal->j_max_transaction_buffers) {
436 CERROR("want too many journal credits (%d) using %d instead\n",
437 needed, journal->j_max_transaction_buffers);
438 needed = journal->j_max_transaction_buffers;
441 LASSERTF(needed > 0, "can't start %d credit transaction\n", needed);
442 handle = ext3_journal_start(fso->fso_dentry->d_inode, needed);
443 if (IS_ERR(handle)) {
444 CERROR("can't get handle for %d credits: rc = %ld\n", needed,
447 LASSERT(handle->h_buffer_credits >= needed);
448 LASSERT(current->journal_info == handle);
454 static int fsfilt_ext3_extend(struct inode *inode, unsigned int nblocks,void *h)
456 handle_t *handle = h;
458 /* fsfilt_extend called with nblocks = 0 for testing in special cases */
460 handle->h_buffer_credits = 0;
461 CWARN("setting credits of handle %p to zero by request\n", h);
464 if (handle->h_buffer_credits > nblocks)
466 if (ext3_journal_extend(handle, nblocks) == 0)
469 ext3_mark_inode_dirty(handle, inode);
470 return ext3_journal_restart(handle, nblocks);
473 static int fsfilt_ext3_commit(struct inode *inode, void *h, int force_sync)
476 handle_t *handle = h;
478 LASSERT(current->journal_info == handle);
480 handle->h_sync = 1; /* recovery likes this */
482 rc = ext3_journal_stop(handle);
487 static int fsfilt_ext3_commit_async(struct inode *inode, void *h,
491 transaction_t *transaction;
492 handle_t *handle = h;
496 LASSERT(current->journal_info == handle);
498 transaction = handle->h_transaction;
499 journal = transaction->t_journal;
500 tid = transaction->t_tid;
501 /* we don't want to be blocked */
503 rc = ext3_journal_stop(handle);
505 CERROR("error while stopping transaction: %d\n", rc);
508 fsfilt_log_start_commit(journal, tid);
510 *wait_handle = (void *) tid;
511 CDEBUG(D_INODE, "commit async: %lu\n", (unsigned long) tid);
515 static int fsfilt_ext3_commit_wait(struct inode *inode, void *h)
517 journal_t *journal = EXT3_JOURNAL(inode);
518 tid_t tid = (tid_t)(long)h;
520 CDEBUG(D_INODE, "commit wait: %lu\n", (unsigned long) tid);
521 if (unlikely(is_journal_aborted(journal)))
524 fsfilt_log_wait_commit(EXT3_JOURNAL(inode), tid);
526 if (unlikely(is_journal_aborted(journal)))
531 static int fsfilt_ext3_setattr(struct dentry *dentry, void *handle,
532 struct iattr *iattr, int do_trunc)
534 struct inode *inode = dentry->d_inode;
537 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2,7,50,0)
538 /* Try to correct for a bug in 2.1.0 (LU-221) that caused negative
539 * timestamps to appear to be in the far future, due old timestamp
540 * being stored on disk as an unsigned value. This fixes up any
541 * bad values held by the client before storing them on disk,
542 * and ensures any timestamp updates are correct. LU-1042 */
543 if (unlikely(LTIME_S(inode->i_atime) == LU221_BAD_TIME &&
544 !(iattr->ia_valid & ATTR_ATIME))) {
545 iattr->ia_valid |= ATTR_ATIME;
546 LTIME_S(iattr->ia_atime) = 0;
548 if (unlikely(LTIME_S(inode->i_mtime) == LU221_BAD_TIME &&
549 !(iattr->ia_valid & ATTR_MTIME))) {
550 iattr->ia_valid |= ATTR_MTIME;
551 LTIME_S(iattr->ia_mtime) = 0;
553 if (unlikely((LTIME_S(inode->i_ctime) == LU221_BAD_TIME ||
554 LTIME_S(inode->i_ctime) == 0) &&
555 !(iattr->ia_valid & ATTR_CTIME))) {
556 iattr->ia_valid |= ATTR_CTIME;
557 LTIME_S(iattr->ia_ctime) = 0;
560 #warning "remove old LU-221/LU-1042 workaround code"
563 /* When initializating timestamps for new inodes, use the filesystem
564 * mkfs time for ctime to avoid e2fsck ibadness incorrectly thinking
565 * that this is potentially an invalid inode. Files with an old ctime
566 * migrated to a newly-formatted OST with a newer s_mkfs_time will not
567 * hit this check, since it is only for ctime == 0. LU-1010/LU-1042 */
568 if ((iattr->ia_valid & ATTR_CTIME) && LTIME_S(iattr->ia_ctime) == 0)
569 LTIME_S(iattr->ia_ctime) =
570 EXT4_SB(inode->i_sb)->s_es->s_mkfs_time;
572 /* Avoid marking the inode dirty on the superblock list unnecessarily.
573 * We are already writing the inode to disk as part of this
574 * transaction and want to avoid a lot of extra inode writeout
575 * later on. b=9828 */
576 if (iattr->ia_valid & ATTR_SIZE && !do_trunc) {
577 /* ATTR_SIZE would invoke truncate: clear it */
578 iattr->ia_valid &= ~ATTR_SIZE;
579 EXT3_I(inode)->i_disksize = iattr->ia_size;
580 i_size_write(inode, iattr->ia_size);
582 if (iattr->ia_valid & ATTR_UID)
583 inode->i_uid = iattr->ia_uid;
584 if (iattr->ia_valid & ATTR_GID)
585 inode->i_gid = iattr->ia_gid;
586 if (iattr->ia_valid & ATTR_ATIME)
587 inode->i_atime = iattr->ia_atime;
588 if (iattr->ia_valid & ATTR_MTIME)
589 inode->i_mtime = iattr->ia_mtime;
590 if (iattr->ia_valid & ATTR_CTIME)
591 inode->i_ctime = iattr->ia_ctime;
592 if (iattr->ia_valid & ATTR_MODE) {
593 inode->i_mode = iattr->ia_mode;
595 if (!cfs_curproc_is_in_groups(inode->i_gid) &&
596 !cfs_capable(CFS_CAP_FSETID))
597 inode->i_mode &= ~S_ISGID;
600 inode->i_sb->s_op->dirty_inode(inode);
605 /* Don't allow setattr to change file type */
606 if (iattr->ia_valid & ATTR_MODE)
607 iattr->ia_mode = (inode->i_mode & S_IFMT) |
608 (iattr->ia_mode & ~S_IFMT);
610 /* We set these flags on the client, but have already checked perms
611 * so don't confuse inode_change_ok. */
612 iattr->ia_valid &= ~TIMES_SET_FLAGS;
614 if (inode->i_op->setattr) {
615 rc = inode->i_op->setattr(dentry, iattr);
617 #ifndef HAVE_SIMPLE_SETATTR /* simple_setattr() already call it */
618 rc = inode_change_ok(inode, iattr);
621 rc = simple_setattr(dentry, iattr);
628 static int fsfilt_ext3_iocontrol(struct inode *inode, struct file *file,
629 unsigned int cmd, unsigned long arg)
634 /* FIXME: Can't do this because of nested transaction deadlock */
635 if (cmd == EXT3_IOC_SETFLAGS) {
636 /* We can't enable data journaling on OST objects, because
637 * this forces the transaction to be closed in order to
638 * flush the journal, but the caller will already have a
639 * compound transaction open to update the last_rcvd file,
640 * and this thread would deadlock trying to set the flag. */
641 if ((*(int *)arg) & EXT3_JOURNAL_DATA_FL) {
642 CERROR("can't set data journal flag on file\n");
645 /* Because the MDS does not see the EXTENTS_FL set on the
646 * OST objects, mask this flag into all set flags. It is
647 * not legal to clear this flag in any case, so we are not
648 * changing the functionality by doing this. b=22911 */
649 *(int *)arg |= EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL;
652 /* ext4_ioctl does not have a inode argument */
653 if (inode->i_fop->unlocked_ioctl)
654 rc = inode->i_fop->unlocked_ioctl(file, cmd, arg);
661 static int fsfilt_ext3_set_md(struct inode *inode, void *handle,
662 void *lmm, int lmm_size, const char *name)
666 LASSERT(mutex_trylock(&inode->i_mutex) == 0);
668 rc = ext3_xattr_set_handle(handle, inode, EXT3_XATTR_INDEX_TRUSTED,
669 name, lmm, lmm_size, XATTR_NO_CTIME);
671 if (rc && rc != -EROFS)
672 CERROR("error adding MD data to inode %lu: rc = %d\n",
677 /* Must be called with i_mutex held */
678 static int fsfilt_ext3_get_md(struct inode *inode, void *lmm, int lmm_size,
683 LASSERT(mutex_trylock(&inode->i_mutex) == 0);
685 rc = ext3_xattr_get(inode, EXT3_XATTR_INDEX_TRUSTED,
686 name, lmm, lmm_size);
688 /* This gives us the MD size */
690 return (rc == -ENODATA) ? 0 : rc;
693 CDEBUG(D_INFO, "error getting EA %d/%s from inode %lu: rc %d\n",
694 EXT3_XATTR_INDEX_TRUSTED, name,
696 memset(lmm, 0, lmm_size);
697 return (rc == -ENODATA) ? 0 : rc;
703 static int fsfilt_ext3_send_bio(int rw, struct inode *inode, struct bio *bio)
709 static ssize_t fsfilt_ext3_readpage(struct file *file, char *buf, size_t count,
712 struct inode *inode = file->f_dentry->d_inode;
715 if (S_ISREG(inode->i_mode))
716 rc = file->f_op->read(file, buf, count, off);
718 const int blkbits = inode->i_sb->s_blocksize_bits;
719 const int blksize = inode->i_sb->s_blocksize;
721 CDEBUG(D_EXT2, "reading %lu at dir %lu+%llu\n",
722 (unsigned long)count, inode->i_ino, *off);
724 struct buffer_head *bh;
727 if (*off < i_size_read(inode)) {
730 bh = ext3_bread(NULL, inode, *off >> blkbits,
733 CDEBUG(D_EXT2, "read %u@%llu\n", blksize, *off);
736 memcpy(buf, bh->b_data, blksize);
739 /* XXX in theory we should just fake
740 * this buffer and continue like ext3,
741 * especially if this is a partial read
743 CERROR("error read dir %lu+%llu: %d\n",
744 inode->i_ino, *off, err);
749 struct ext3_dir_entry_2 *fake = (void *)buf;
751 CDEBUG(D_EXT2, "fake %u@%llu\n", blksize, *off);
752 memset(fake, 0, sizeof(*fake));
753 fake->rec_len = cpu_to_le16(blksize);
765 #ifdef HAVE_EXT4_JOURNAL_CALLBACK_ADD
766 static void fsfilt_ext3_cb_func(struct super_block *sb,
767 struct journal_callback *jcb, int error)
769 static void fsfilt_ext3_cb_func(struct journal_callback *jcb, int error)
772 struct fsfilt_cb_data *fcb = container_of(jcb, typeof(*fcb), cb_jcb);
774 fcb->cb_func(fcb->cb_obd, fcb->cb_last_rcvd, fcb->cb_data, error);
776 OBD_SLAB_FREE(fcb, fcb_cache, sizeof *fcb);
779 static int fsfilt_ext3_add_journal_cb(struct obd_device *obd, __u64 last_rcvd,
780 void *handle, fsfilt_cb_t cb_func,
783 struct fsfilt_cb_data *fcb;
785 OBD_SLAB_ALLOC_PTR_GFP(fcb, fcb_cache, CFS_ALLOC_IO);
789 fcb->cb_func = cb_func;
791 fcb->cb_last_rcvd = last_rcvd;
792 fcb->cb_data = cb_data;
794 CDEBUG(D_EXT2, "set callback for last_rcvd: "LPD64"\n", last_rcvd);
795 fsfilt_journal_callback_set(handle, fsfilt_ext3_cb_func, &fcb->cb_jcb);
800 static int fsfilt_ext3_statfs(struct super_block *sb, struct obd_statfs *osfs)
805 memset(&sfs, 0, sizeof(sfs));
806 rc = ll_do_statfs(sb, &sfs);
807 statfs_pack(osfs, &sfs);
811 static int fsfilt_ext3_sync(struct super_block *sb)
813 return ext3_force_commit(sb);
816 #ifndef EXT3_EXTENTS_FL
817 #define EXT3_EXTENTS_FL 0x00080000 /* Inode uses extents */
821 #define EXT_ASSERT(cond) BUG_ON(!(cond))
824 #ifdef EXT3_EXT_HAS_NO_TREE
825 /* for kernels 2.6.18 and later */
826 #define EXT_GENERATION(inode) (EXT4_I(inode)->i_ext_generation)
827 #define ext3_ext_base inode
828 #define ext3_ext_base2inode(inode) (inode)
829 #define EXT_DEPTH(inode) ext_depth(inode)
830 #define fsfilt_ext3_ext_walk_space(inode, block, num, cb, cbdata) \
831 ext3_ext_walk_space(inode, block, num, cb, cbdata);
833 #define ext3_ext_base ext3_extents_tree
834 #define ext3_ext_base2inode(tree) (tree->inode)
835 #define fsfilt_ext3_ext_walk_space(tree, block, num, cb, cbdata) \
836 ext3_ext_walk_space(tree, block, num, cb);
840 unsigned long *blocks;
848 static long ext3_ext_find_goal(struct inode *inode, struct ext3_ext_path *path,
849 unsigned long block, int *aflags)
851 struct ext3_inode_info *ei = EXT3_I(inode);
852 unsigned long bg_start;
853 unsigned long colour;
857 struct ext3_extent *ex;
858 depth = path->p_depth;
860 /* try to predict block placement */
861 if ((ex = path[depth].p_ext))
862 return ext_pblock(ex) + (block - le32_to_cpu(ex->ee_block));
864 /* it looks index is empty
865 * try to find starting from index itself */
866 if (path[depth].p_bh)
867 return path[depth].p_bh->b_blocknr;
870 /* OK. use inode's group */
871 bg_start = (ei->i_block_group * EXT3_BLOCKS_PER_GROUP(inode->i_sb)) +
872 le32_to_cpu(EXT3_SB(inode->i_sb)->s_es->s_first_data_block);
873 colour = (current->pid % 16) *
874 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
875 return bg_start + colour + block;
878 #define ll_unmap_underlying_metadata(sb, blocknr) \
879 unmap_underlying_metadata((sb)->s_bdev, blocknr)
881 #ifndef EXT3_MB_HINT_GROUP_ALLOC
882 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
883 struct ext3_ext_path *path, unsigned long block,
884 unsigned long *count, int *err)
886 unsigned long pblock, goal;
888 struct inode *inode = ext3_ext_base2inode(base);
890 goal = ext3_ext_find_goal(inode, path, block, &aflags);
891 aflags |= 2; /* block have been already reserved */
892 pblock = ext3_mb_new_blocks(handle, inode, goal, count, aflags, err);
897 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
898 struct ext3_ext_path *path, unsigned long block,
899 unsigned long *count, int *err)
901 struct inode *inode = ext3_ext_base2inode(base);
902 struct ext3_allocation_request ar;
903 unsigned long pblock;
906 /* find neighbour allocated blocks */
908 *err = ext3_ext_search_left(base, path, &ar.lleft, &ar.pleft);
912 *err = ext3_ext_search_right(base, path, &ar.lright, &ar.pright);
916 /* allocate new block */
917 ar.goal = ext3_ext_find_goal(inode, path, block, &aflags);
921 ar.flags = EXT3_MB_HINT_DATA;
922 pblock = ext3_mb_new_blocks(handle, &ar, err);
928 #ifdef EXT3_EXT_HAS_NO_TREE
929 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
930 struct ext3_ext_path *path,
931 struct ext3_ext_cache *cex,
932 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
933 struct ext3_extent *ex,
937 struct bpointers *bp = cbdata;
939 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
940 struct ext3_ext_path *path,
941 struct ext3_ext_cache *cex
942 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
943 , struct ext3_extent *ex
947 struct bpointers *bp = base->private;
949 struct inode *inode = ext3_ext_base2inode(base);
950 struct ext3_extent nex;
951 unsigned long pblock;
957 if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
962 if (bp->create == 0) {
964 if (cex->ec_block < bp->start)
965 i = bp->start - cex->ec_block;
966 if (i >= cex->ec_len)
967 CERROR("nothing to do?! i = %d, e_num = %u\n",
969 for (; i < cex->ec_len && bp->num; i++) {
981 tgen = EXT_GENERATION(base);
982 count = ext3_ext_calc_credits_for_insert(base, path);
984 handle = ext3_journal_start(inode, count+EXT3_ALLOC_NEEDED+1);
985 if (IS_ERR(handle)) {
986 return PTR_ERR(handle);
989 if (tgen != EXT_GENERATION(base)) {
990 /* the tree has changed. so path can be invalid at moment */
991 ext3_journal_stop(handle);
995 /* In 2.6.32 kernel, ext4_ext_walk_space()'s callback func is not
996 * protected by i_data_sem as whole. so we patch it to store
997 * generation to path and now verify the tree hasn't changed */
998 down_write((&EXT4_I(inode)->i_data_sem));
1000 /* validate extent, make sure the extent tree does not changed */
1001 if (EXT_GENERATION(base) != path[0].p_generation) {
1002 /* cex is invalid, try again */
1003 up_write(&EXT4_I(inode)->i_data_sem);
1004 ext3_journal_stop(handle);
1008 count = cex->ec_len;
1009 pblock = new_blocks(handle, base, path, cex->ec_block, &count, &err);
1012 EXT_ASSERT(count <= cex->ec_len);
1014 /* insert new extent */
1015 nex.ee_block = cpu_to_le32(cex->ec_block);
1016 ext3_ext_store_pblock(&nex, pblock);
1017 nex.ee_len = cpu_to_le16(count);
1018 err = fsfilt_ext3_ext_insert_extent(handle, base, path, &nex, 0);
1020 /* free data blocks we just allocated */
1021 /* not a good idea to call discard here directly,
1022 * but otherwise we'd need to call it every free() */
1023 #ifdef EXT3_MB_HINT_GROUP_ALLOC
1024 ext3_mb_discard_inode_preallocations(inode);
1026 ext3_free_blocks(handle, inode, ext_pblock(&nex),
1027 cpu_to_le16(nex.ee_len), 0);
1032 * Putting len of the actual extent we just inserted,
1033 * we are asking ext3_ext_walk_space() to continue
1034 * scaning after that block
1036 cex->ec_len = le16_to_cpu(nex.ee_len);
1037 cex->ec_start = ext_pblock(&nex);
1038 BUG_ON(le16_to_cpu(nex.ee_len) == 0);
1039 BUG_ON(le32_to_cpu(nex.ee_block) != cex->ec_block);
1042 up_write((&EXT4_I(inode)->i_data_sem));
1043 ext3_journal_stop(handle);
1048 CERROR("hmm. why do we find this extent?\n");
1049 CERROR("initial space: %lu:%u\n",
1050 bp->start, bp->init_num);
1051 CERROR("current extent: %u/%u/%llu %d\n",
1052 cex->ec_block, cex->ec_len,
1053 (unsigned long long)cex->ec_start,
1057 if (cex->ec_block < bp->start)
1058 i = bp->start - cex->ec_block;
1059 if (i >= cex->ec_len)
1060 CERROR("nothing to do?! i = %d, e_num = %u\n",
1062 for (; i < cex->ec_len && bp->num; i++) {
1063 *(bp->blocks) = cex->ec_start + i;
1064 if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
1068 /* unmap any possible underlying metadata from
1069 * the block device mapping. bug 6998. */
1070 ll_unmap_underlying_metadata(inode->i_sb,
1082 int fsfilt_map_nblocks(struct inode *inode, unsigned long block,
1083 unsigned long num, unsigned long *blocks,
1084 int *created, int create)
1086 #ifdef EXT3_EXT_HAS_NO_TREE
1087 struct ext3_ext_base *base = inode;
1089 struct ext3_extents_tree tree;
1090 struct ext3_ext_base *base = &tree;
1092 struct bpointers bp;
1095 CDEBUG(D_OTHER, "blocks %lu-%lu requested for inode %u\n",
1096 block, block + num - 1, (unsigned) inode->i_ino);
1098 #ifndef EXT3_EXT_HAS_NO_TREE
1099 ext3_init_tree_desc(base, inode);
1103 bp.created = created;
1105 bp.init_num = bp.num = num;
1108 err = fsfilt_ext3_ext_walk_space(base, block, num,
1109 ext3_ext_new_extent_cb, &bp);
1110 ext3_ext_invalidate_cache(base);
1115 int fsfilt_ext3_map_ext_inode_pages(struct inode *inode, struct page **page,
1116 int pages, unsigned long *blocks,
1117 int *created, int create)
1119 int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1121 struct page *fp = NULL;
1124 CDEBUG(D_OTHER, "inode %lu: map %d pages from %lu\n",
1125 inode->i_ino, pages, (*page)->index);
1127 /* pages are sorted already. so, we just have to find
1128 * contig. space and process them properly */
1131 /* start new extent */
1136 } else if (fp->index + clen == (*page)->index) {
1137 /* continue the extent */
1144 /* process found extent */
1145 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1146 clen * blocks_per_page, blocks,
1151 /* look for next extent */
1153 blocks += blocks_per_page * clen;
1154 created += blocks_per_page * clen;
1158 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1159 clen * blocks_per_page, blocks,
1165 extern int ext3_map_inode_page(struct inode *inode, struct page *page,
1166 unsigned long *blocks, int *created, int create);
1167 int fsfilt_ext3_map_bm_inode_pages(struct inode *inode, struct page **page,
1168 int pages, unsigned long *blocks,
1169 int *created, int create)
1171 int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1175 for (i = 0, cr = created, b = blocks; i < pages; i++, page++) {
1176 rc = ext3_map_inode_page(inode, *page, b, cr, create);
1178 CERROR("ino %lu, blk %lu cr %u create %d: rc %d\n",
1179 inode->i_ino, *b, *cr, create, rc);
1183 b += blocks_per_page;
1184 cr += blocks_per_page;
1189 int fsfilt_ext3_map_inode_pages(struct inode *inode, struct page **page,
1190 int pages, unsigned long *blocks,
1191 int *created, int create,
1192 cfs_mutex_t *optional_mutex)
1196 if (EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL) {
1197 rc = fsfilt_ext3_map_ext_inode_pages(inode, page, pages,
1198 blocks, created, create);
1201 if (optional_mutex != NULL)
1202 cfs_mutex_lock(optional_mutex);
1203 rc = fsfilt_ext3_map_bm_inode_pages(inode, page, pages, blocks,
1205 if (optional_mutex != NULL)
1206 cfs_mutex_unlock(optional_mutex);
1211 int fsfilt_ext3_read(struct inode *inode, void *buf, int size, loff_t *offs)
1213 unsigned long block;
1214 struct buffer_head *bh;
1215 int err, blocksize, csize, boffs, osize = size;
1217 /* prevent reading after eof */
1218 spin_lock(&inode->i_lock);
1219 if (i_size_read(inode) < *offs + size) {
1220 size = i_size_read(inode) - *offs;
1221 spin_unlock(&inode->i_lock);
1223 CDEBUG(D_EXT2, "size %llu is too short for read @%llu\n",
1224 i_size_read(inode), *offs);
1226 } else if (size == 0) {
1230 spin_unlock(&inode->i_lock);
1233 blocksize = 1 << inode->i_blkbits;
1236 block = *offs >> inode->i_blkbits;
1237 boffs = *offs & (blocksize - 1);
1238 csize = min(blocksize - boffs, size);
1239 bh = ext3_bread(NULL, inode, block, 0, &err);
1241 CERROR("can't read block: %d\n", err);
1245 memcpy(buf, bh->b_data + boffs, csize);
1254 EXPORT_SYMBOL(fsfilt_ext3_read);
1256 static int fsfilt_ext3_read_record(struct file * file, void *buf,
1257 int size, loff_t *offs)
1260 rc = fsfilt_ext3_read(file->f_dentry->d_inode, buf, size, offs);
1266 int fsfilt_ext3_write_handle(struct inode *inode, void *buf, int bufsize,
1267 loff_t *offs, handle_t *handle)
1269 struct buffer_head *bh = NULL;
1270 loff_t old_size = i_size_read(inode), offset = *offs;
1271 loff_t new_size = i_size_read(inode);
1272 unsigned long block;
1273 int err = 0, blocksize = 1 << inode->i_blkbits, size, boffs;
1275 while (bufsize > 0) {
1279 block = offset >> inode->i_blkbits;
1280 boffs = offset & (blocksize - 1);
1281 size = min(blocksize - boffs, bufsize);
1282 bh = ext3_bread(handle, inode, block, 1, &err);
1284 CERROR("can't read/create block: %d\n", err);
1288 err = ext3_journal_get_write_access(handle, bh);
1290 CERROR("journal_get_write_access() returned error %d\n",
1294 LASSERT(bh->b_data + boffs + size <= bh->b_data + bh->b_size);
1295 memcpy(bh->b_data + boffs, buf, size);
1296 err = ext3_journal_dirty_metadata(handle, bh);
1298 CERROR("journal_dirty_metadata() returned error %d\n",
1302 if (offset + size > new_size)
1303 new_size = offset + size;
1311 /* correct in-core and on-disk sizes */
1312 if (new_size > i_size_read(inode)) {
1313 spin_lock(&inode->i_lock);
1314 if (new_size > i_size_read(inode))
1315 i_size_write(inode, new_size);
1316 if (i_size_read(inode) > EXT3_I(inode)->i_disksize)
1317 EXT3_I(inode)->i_disksize = i_size_read(inode);
1318 if (i_size_read(inode) > old_size) {
1319 spin_unlock(&inode->i_lock);
1320 mark_inode_dirty(inode);
1322 spin_unlock(&inode->i_lock);
1330 EXPORT_SYMBOL(fsfilt_ext3_write_handle);
1332 static int fsfilt_ext3_write_record(struct file *file, void *buf, int bufsize,
1333 loff_t *offs, int force_sync)
1335 struct inode *inode = file->f_dentry->d_inode;
1337 int err, block_count = 0, blocksize;
1339 /* Determine how many transaction credits are needed */
1340 blocksize = 1 << inode->i_blkbits;
1341 block_count = (*offs & (blocksize - 1)) + bufsize;
1342 block_count = (block_count + blocksize - 1) >> inode->i_blkbits;
1344 handle = ext3_journal_start(inode,
1345 block_count * EXT3_DATA_TRANS_BLOCKS(inode->i_sb) + 2);
1346 if (IS_ERR(handle)) {
1347 CERROR("can't start transaction for %d blocks (%d bytes)\n",
1348 block_count * EXT3_DATA_TRANS_BLOCKS(inode->i_sb) + 2,
1350 return PTR_ERR(handle);
1353 err = fsfilt_ext3_write_handle(inode, buf, bufsize, offs, handle);
1355 if (!err && force_sync)
1356 handle->h_sync = 1; /* recovery likes this */
1358 ext3_journal_stop(handle);
1363 static int fsfilt_ext3_setup(struct super_block *sb)
1365 #if defined(HAVE_QUOTA_SUPPORT) || defined(S_PDIROPS)
1366 struct ext3_sb_info *sbi = EXT3_SB(sb);
1368 sbi->dx_lock = fsfilt_ext3_dx_lock;
1369 sbi->dx_unlock = fsfilt_ext3_dx_unlock;
1372 if (!EXT3_HAS_COMPAT_FEATURE(sb,
1373 EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
1374 CERROR("ext3 mounted without journal\n");
1379 CWARN("Enabling PDIROPS\n");
1380 set_opt(sbi->s_mount_opt, PDIROPS);
1381 sb->s_flags |= S_PDIROPS;
1383 if (!EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1384 CWARN("filesystem doesn't have dir_index feature enabled\n");
1385 #ifdef HAVE_QUOTA_SUPPORT
1386 /* enable journaled quota support */
1387 /* kfreed in ext3_put_super() */
1388 sbi->s_qf_names[USRQUOTA] = kstrdup("lquota.user.reserved", GFP_KERNEL);
1389 if (!sbi->s_qf_names[USRQUOTA])
1391 sbi->s_qf_names[GRPQUOTA] = kstrdup("lquota.group.reserved", GFP_KERNEL);
1392 if (!sbi->s_qf_names[GRPQUOTA]) {
1393 kfree(sbi->s_qf_names[USRQUOTA]);
1394 sbi->s_qf_names[USRQUOTA] = NULL;
1397 sbi->s_jquota_fmt = QFMT_LUSTRE;
1398 set_opt(sbi->s_mount_opt, QUOTA);
1403 /* If fso is NULL, op is FSFILT operation, otherwise op is number of fso
1404 objects. Logs is number of logfiles to update */
1405 static int fsfilt_ext3_get_op_len(int op, struct fsfilt_objinfo *fso, int logs)
1409 case FSFILT_OP_CREATE:
1410 /* directory leaf, index & indirect & EA*/
1411 return 4 + 3 * logs;
1412 case FSFILT_OP_UNLINK:
1418 struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
1419 int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
1420 int addrpp = EXT3_ADDR_PER_BLOCK(sb) * blockpp;
1421 for (i = 0; i < op; i++, fso++) {
1422 int nblocks = fso->fso_bufcnt * blockpp;
1423 int ndindirect = min(nblocks, addrpp + 1);
1424 int nindir = nblocks + ndindirect + 1;
1428 return needed + 3 * logs;
1434 #ifdef HAVE_QUOTA_SUPPORT
1435 #define DQINFO_COPY(out, in) \
1437 Q_COPY(out, in, dqi_bgrace); \
1438 Q_COPY(out, in, dqi_igrace); \
1439 Q_COPY(out, in, dqi_flags); \
1440 Q_COPY(out, in, dqi_valid); \
1443 #define DQBLK_COPY(out, in) \
1445 Q_COPY(out, in, dqb_bhardlimit); \
1446 Q_COPY(out, in, dqb_bsoftlimit); \
1447 Q_COPY(out, in, dqb_curspace); \
1448 Q_COPY(out, in, dqb_ihardlimit); \
1449 Q_COPY(out, in, dqb_isoftlimit); \
1450 Q_COPY(out, in, dqb_curinodes); \
1451 Q_COPY(out, in, dqb_btime); \
1452 Q_COPY(out, in, dqb_itime); \
1453 Q_COPY(out, in, dqb_valid); \
1456 static int fsfilt_ext3_quotactl(struct super_block *sb,
1457 struct obd_quotactl *oqc)
1459 int i, rc = 0, error = 0;
1460 const struct quotactl_ops *qcop;
1461 struct if_dqinfo *info;
1462 struct if_dqblk *dqblk;
1468 OBD_ALLOC_PTR(info);
1471 OBD_ALLOC_PTR(dqblk);
1477 DQINFO_COPY(info, &oqc->qc_dqinfo);
1478 DQBLK_COPY(dqblk, &oqc->qc_dqblk);
1481 if (oqc->qc_cmd == Q_QUOTAON || oqc->qc_cmd == Q_QUOTAOFF) {
1482 for (i = 0; i < MAXQUOTAS; i++) {
1483 if (!Q_TYPESET(oqc, i))
1486 if (oqc->qc_cmd == Q_QUOTAON) {
1487 char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1489 LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1491 rc = ll_quota_on(sb, i, QFMT_LUSTRE,
1493 } else if (oqc->qc_cmd == Q_QUOTAOFF) {
1494 rc = ll_quota_off(sb, i, 0);
1502 GOTO(out, rc ?: error);
1505 switch (oqc->qc_cmd) {
1508 if (!qcop->get_info)
1509 GOTO(out, rc = -ENOSYS);
1510 rc = qcop->get_info(sb, oqc->qc_type, info);
1514 if (!qcop->set_dqblk)
1515 GOTO(out, rc = -ENOSYS);
1516 rc = qcop->set_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1520 if (!qcop->get_dqblk)
1521 GOTO(out, rc = -ENOSYS);
1522 rc = qcop->get_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1524 dqblk->dqb_valid = QIF_LIMITS | QIF_USAGE;
1527 if (!sb->s_qcop->quota_sync)
1528 GOTO(out, rc = -ENOSYS);
1529 qcop->quota_sync(sb, oqc->qc_type);
1532 CDEBUG(D_WARNING, "invalidating operational quota files\n");
1533 for (i = 0; i < MAXQUOTAS; i++) {
1535 char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1537 LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1539 if (!Q_TYPESET(oqc, i))
1542 fp = filp_open(name[i], O_CREAT | O_TRUNC | O_RDWR, 0644);
1545 CERROR("error invalidating operational quota file"
1546 " %s (rc:%d)\n", name[i], rc);
1554 CERROR("unsupported quotactl command: %d\n", oqc->qc_cmd);
1558 DQINFO_COPY(&oqc->qc_dqinfo, info);
1559 DQBLK_COPY(&oqc->qc_dqblk, dqblk);
1562 OBD_FREE_PTR(dqblk);
1565 CDEBUG(D_QUOTA, "quotactl command %#x, id %u, type %u "
1567 oqc->qc_cmd, oqc->qc_id, oqc->qc_type, rc);
1572 cfs_hlist_node_t dqb_hash; /** quotacheck hash */
1573 cfs_list_t dqb_list; /** in list also */
1574 qid_t dqb_id; /** uid/gid */
1575 short dqb_type; /** USRQUOTA/GRPQUOTA */
1576 qsize_t dqb_bhardlimit; /** block hard limit */
1577 qsize_t dqb_bsoftlimit; /** block soft limit */
1578 qsize_t dqb_curspace; /** current space */
1579 qsize_t dqb_ihardlimit; /** inode hard limit */
1580 qsize_t dqb_isoftlimit; /** inode soft limit */
1581 qsize_t dqb_curinodes; /** current inodes */
1582 obd_time dqb_btime; /** block grace time */
1583 obd_time dqb_itime; /** inode grace time */
1584 __u32 dqb_valid; /** flag for above fields */
1587 static inline unsigned int chkquot_hash(qid_t id, int type)
1588 __attribute__((__const__));
1590 static inline unsigned int chkquot_hash(qid_t id, int type)
1592 return (id * (MAXQUOTAS - type)) % NR_DQHASH;
1595 static inline struct chk_dqblk *
1596 find_chkquot(cfs_hlist_head_t *head, qid_t id, int type)
1598 cfs_hlist_node_t *node;
1599 struct chk_dqblk *cdqb;
1601 cfs_hlist_for_each(node, head) {
1602 cdqb = cfs_hlist_entry(node, struct chk_dqblk, dqb_hash);
1603 if (cdqb->dqb_id == id && cdqb->dqb_type == type)
1610 static struct chk_dqblk *alloc_chkquot(qid_t id, int type)
1612 struct chk_dqblk *cdqb;
1614 OBD_ALLOC_PTR(cdqb);
1616 CFS_INIT_HLIST_NODE(&cdqb->dqb_hash);
1617 CFS_INIT_LIST_HEAD(&cdqb->dqb_list);
1619 cdqb->dqb_type = type;
1625 static struct chk_dqblk *
1626 cqget(struct super_block *sb, cfs_hlist_head_t *hash,
1627 cfs_list_t *list, qid_t id, int type, int first_check)
1629 cfs_hlist_head_t *head = hash + chkquot_hash(id, type);
1630 struct if_dqblk dqb;
1631 struct chk_dqblk *cdqb;
1634 cdqb = find_chkquot(head, id, type);
1638 cdqb = alloc_chkquot(id, type);
1643 rc = sb->s_qcop->get_dqblk(sb, type, id, &dqb);
1645 CERROR("get_dqblk of id %u, type %d failed: %d\n",
1648 DQBLK_COPY(cdqb, &dqb);
1649 cdqb->dqb_curspace = 0;
1650 cdqb->dqb_curinodes = 0;
1654 cfs_hlist_add_head(&cdqb->dqb_hash, head);
1655 cfs_list_add_tail(&cdqb->dqb_list, list);
1660 static inline int quota_onoff(struct super_block *sb, int cmd, int type, int qfmt)
1662 struct obd_quotactl *oqctl;
1665 OBD_ALLOC_PTR(oqctl);
1669 oqctl->qc_cmd = cmd;
1670 oqctl->qc_id = qfmt;
1671 oqctl->qc_type = type;
1672 rc = fsfilt_ext3_quotactl(sb, oqctl);
1674 OBD_FREE_PTR(oqctl);
1678 static inline int read_old_dqinfo(struct super_block *sb, int type,
1679 struct if_dqinfo *dqinfo)
1681 struct obd_quotactl *oqctl;
1685 OBD_ALLOC_PTR(oqctl);
1689 oqctl->qc_cmd = Q_GETINFO;
1690 oqctl->qc_type = type;
1691 rc = fsfilt_ext3_quotactl(sb, oqctl);
1693 ((struct obd_dqinfo *)dqinfo)[type] = oqctl->qc_dqinfo;
1695 OBD_FREE_PTR(oqctl);
1700 cfs_hlist_head_t qckt_hash[NR_DQHASH]; /* quotacheck hash */
1701 cfs_list_t qckt_list; /* quotacheck list */
1702 int qckt_first_check[MAXQUOTAS]; /* 1 if no old quotafile */
1703 struct if_dqinfo qckt_dqinfo[MAXQUOTAS]; /* old dqinfo */
1706 static int add_inode_quota(struct inode *inode, struct qchk_ctxt *qctxt,
1707 struct obd_quotactl *oqc)
1709 struct chk_dqblk *cdqb[MAXQUOTAS] = { NULL, };
1711 qid_t qid[MAXQUOTAS];
1717 qid[USRQUOTA] = inode->i_uid;
1718 qid[GRPQUOTA] = inode->i_gid;
1720 if (S_ISDIR(inode->i_mode) ||
1721 S_ISREG(inode->i_mode) ||
1722 S_ISLNK(inode->i_mode))
1723 size = inode_get_bytes(inode);
1725 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1726 if (!Q_TYPESET(oqc, cnt))
1729 cdqb[cnt] = cqget(inode->i_sb, qctxt->qckt_hash,
1730 &qctxt->qckt_list, qid[cnt], cnt,
1731 qctxt->qckt_first_check[cnt]);
1737 cdqb[cnt]->dqb_curspace += size;
1738 cdqb[cnt]->dqb_curinodes++;
1742 for (i = 0; i < cnt; i++) {
1743 if (!Q_TYPESET(oqc, i))
1746 cdqb[i]->dqb_curspace -= size;
1747 cdqb[i]->dqb_curinodes--;
1754 /* write dqinfo struct in a new quota file */
1755 static int v3_write_dqinfo(struct file *f, int type, struct if_dqinfo *info)
1757 struct v2_disk_dqinfo dqinfo;
1758 __u32 blocks = V2_DQTREEOFF + 1;
1759 loff_t offset = V2_DQINFOOFF;
1762 dqinfo.dqi_bgrace = cpu_to_le32(info->dqi_bgrace);
1763 dqinfo.dqi_igrace = cpu_to_le32(info->dqi_igrace);
1764 dqinfo.dqi_flags = cpu_to_le32(info->dqi_flags & DQF_MASK &
1767 dqinfo.dqi_bgrace = cpu_to_le32(MAX_DQ_TIME);
1768 dqinfo.dqi_igrace = cpu_to_le32(MAX_IQ_TIME);
1769 dqinfo.dqi_flags = 0;
1772 dqinfo.dqi_blocks = cpu_to_le32(blocks);
1773 dqinfo.dqi_free_blk = 0;
1774 dqinfo.dqi_free_entry = 0;
1776 return cfs_user_write(f, (char *)&dqinfo, sizeof(dqinfo), &offset);
1779 static int v3_write_dqheader(struct file *f, int type)
1781 static const __u32 quota_magics[] = V2_INITQMAGICS;
1782 static const __u32 quota_versions[] = LUSTRE_INITQVERSIONS_V2;
1783 struct v2_disk_dqheader dqhead;
1786 CLASSERT(ARRAY_SIZE(quota_magics) == ARRAY_SIZE(quota_versions));
1787 LASSERT(0 <= type && type < ARRAY_SIZE(quota_magics));
1789 dqhead.dqh_magic = cpu_to_le32(quota_magics[type]);
1790 dqhead.dqh_version = cpu_to_le32(quota_versions[type]);
1792 return cfs_user_write(f, (char *)&dqhead, sizeof(dqhead), &offset);
1795 static int create_new_quota_files(struct qchk_ctxt *qctxt,
1796 struct obd_quotactl *oqc)
1801 for (i = 0; i < MAXQUOTAS; i++) {
1802 struct if_dqinfo *info = qctxt->qckt_first_check[i]?
1803 NULL : &qctxt->qckt_dqinfo[i];
1805 const char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1807 if (!Q_TYPESET(oqc, i))
1810 LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1812 file = filp_open(name[i], O_RDWR | O_CREAT | O_TRUNC, 0644);
1815 CERROR("can't create %s file: rc = %d\n",
1820 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
1821 CERROR("file %s is not regular", name[i]);
1822 filp_close(file, 0);
1823 GOTO(out, rc = -EINVAL);
1826 ll_vfs_dq_drop(file->f_dentry->d_inode);
1828 rc = v3_write_dqheader(file, i);
1830 filp_close(file, 0);
1834 rc = v3_write_dqinfo(file, i, info);
1835 filp_close(file, 0);
1845 static int commit_chkquot(struct super_block *sb, struct qchk_ctxt *qctxt,
1846 struct chk_dqblk *cdqb)
1848 struct obd_quotactl *oqc;
1857 now = cfs_time_current_sec();
1859 if (cdqb->dqb_bsoftlimit &&
1860 toqb(cdqb->dqb_curspace) >= cdqb->dqb_bsoftlimit &&
1863 now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_bgrace;
1865 if (cdqb->dqb_isoftlimit &&
1866 cdqb->dqb_curinodes >= cdqb->dqb_isoftlimit &&
1869 now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_igrace;
1871 cdqb->dqb_valid = QIF_ALL;
1873 oqc->qc_cmd = Q_SETQUOTA;
1874 oqc->qc_type = cdqb->dqb_type;
1875 oqc->qc_id = cdqb->dqb_id;
1876 DQBLK_COPY(&oqc->qc_dqblk, cdqb);
1878 rc = fsfilt_ext3_quotactl(sb, oqc);
1883 static int prune_chkquots(struct super_block *sb,
1884 struct qchk_ctxt *qctxt, int error)
1886 struct chk_dqblk *cdqb, *tmp;
1889 cfs_list_for_each_entry_safe(cdqb, tmp, &qctxt->qckt_list, dqb_list) {
1891 rc = commit_chkquot(sb, qctxt, cdqb);
1895 cfs_hlist_del_init(&cdqb->dqb_hash);
1896 cfs_list_del(&cdqb->dqb_list);
1903 #ifndef EXT3_FEATURE_RO_COMPAT_GDT_CSUM
1904 #define EXT3_FEATURE_RO_COMPAT_GDT_CSUM 0x0010
1907 static int fsfilt_ext3_quotacheck(struct super_block *sb,
1908 struct obd_quotactl *oqc)
1910 struct ext3_sb_info *sbi = EXT3_SB(sb);
1911 int i, group, uninit_feat = 0;
1912 struct qchk_ctxt *qctxt;
1913 struct buffer_head *bitmap_bh = NULL;
1914 unsigned long ino, inode_inuse;
1915 struct inode *inode;
1919 /* turn on quota and read dqinfo if existed */
1920 OBD_ALLOC_PTR(qctxt);
1922 oqc->qc_stat = -ENOMEM;
1926 for (i = 0; i < NR_DQHASH; i++)
1927 CFS_INIT_HLIST_HEAD(&qctxt->qckt_hash[i]);
1928 CFS_INIT_LIST_HEAD(&qctxt->qckt_list);
1930 for (i = 0; i < MAXQUOTAS; i++) {
1931 if (!Q_TYPESET(oqc, i))
1934 rc = quota_onoff(sb, Q_QUOTAON, i, oqc->qc_id);
1935 if (!rc || rc == -EBUSY) {
1936 rc = read_old_dqinfo(sb, i, qctxt->qckt_dqinfo);
1939 } else if (rc == -ENOENT || rc == -EINVAL || rc == -EEXIST) {
1940 qctxt->qckt_first_check[i] = 1;
1945 if (EXT3_HAS_RO_COMPAT_FEATURE(sb, EXT3_FEATURE_RO_COMPAT_GDT_CSUM))
1946 /* This filesystem supports the uninit group feature */
1949 /* number of inodes that have been allocated */
1950 inode_inuse = sbi->s_inodes_per_group * sbi->s_groups_count -
1951 percpu_counter_sum(&sbi->s_freeinodes_counter);
1953 /* check quota and update in hash */
1954 for (group = 0; group < sbi->s_groups_count && inode_inuse > 0;
1956 unsigned long used_count = sbi->s_inodes_per_group;
1959 struct ext3_group_desc *desc;
1960 desc = ext3_get_group_desc(sb, group, NULL);
1964 /* we don't really need to take the group lock here,
1965 * but it may be useful if one day we support online
1967 ext4_lock_group(sb, group);
1968 if (desc->bg_flags & cpu_to_le16(EXT3_BG_INODE_UNINIT)) {
1969 /* no inode in use in this group, just skip it */
1970 ext3_unlock_group(sb, group);
1974 used_count -= ext3_itable_unused_count(sb, desc);
1975 ext3_unlock_group(sb, group);
1978 ino = group * sbi->s_inodes_per_group + 1;
1979 bitmap_bh = ext3_read_inode_bitmap(sb, group);
1981 CERROR("%s: ext3_read_inode_bitmap group %d failed\n",
1987 while (i < used_count &&
1988 (i = ext3_find_next_bit(bitmap_bh->b_data,
1989 used_count, i)) < used_count) {
1992 ino = i + group * sbi->s_inodes_per_group;
1993 if (ino < sbi->s_first_ino)
1995 inode = ext3_iget(sb, ino);
1996 if (!inode || IS_ERR(inode))
1999 rc = add_inode_quota(inode, qctxt, oqc);
2009 /* read old quota limits from old quota file. (only for the user
2010 * has limits but hasn't file) */
2011 #ifdef HAVE_QUOTA_SUPPORT
2012 for (i = 0; i < MAXQUOTAS; i++) {
2014 struct dquot_id *dqid, *tmp;
2016 if (!Q_TYPESET(oqc, i))
2019 if (qctxt->qckt_first_check[i])
2023 LASSERT(sb_dqopt(sb)->files[i] != NULL);
2024 CFS_INIT_LIST_HEAD(&id_list);
2025 rc = lustre_get_qids(NULL, sb_dqopt(sb)->files[i], i, &id_list);
2027 CERROR("read old limits failed. (rc:%d)\n", rc);
2029 cfs_list_for_each_entry_safe(dqid, tmp, &id_list, di_link) {
2030 cfs_list_del_init(&dqid->di_link);
2033 cqget(sb, qctxt->qckt_hash, &qctxt->qckt_list,
2035 qctxt->qckt_first_check[i]);
2040 /* turn off quota cause we are to dump chk_dqblk to files */
2041 quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type, oqc->qc_id);
2043 rc = create_new_quota_files(qctxt, oqc);
2047 /* we use vfs functions to set dqblk, so turn quota on */
2048 rc = quota_onoff(sb, Q_QUOTAON, oqc->qc_type, oqc->qc_id);
2050 /* dump and free chk_dqblk */
2051 rc = prune_chkquots(sb, qctxt, rc);
2052 OBD_FREE_PTR(qctxt);
2054 /* turn off quota, `lfs quotacheck` will turn on when all
2055 * nodes quotacheck finish. */
2056 quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type, oqc->qc_id);
2060 CERROR("quotacheck failed: rc = %d\n", rc);
2065 static int fsfilt_ext3_quotainfo(struct lustre_quota_info *lqi, int type,
2071 if (lqi->qi_files[type] == NULL) {
2072 CERROR("operate qinfo before it's enabled!\n");
2078 rc = lustre_check_quota_file(lqi, type);
2081 rc = lustre_read_quota_info(lqi, type);
2084 rc = lustre_write_quota_info(lqi, type);
2086 case QFILE_INIT_INFO:
2087 rc = lustre_init_quota_info(lqi, type);
2091 CERROR("quota CONVERT command is not supported\n");
2095 CERROR("Unsupported admin quota file cmd %d\n"
2096 "Are lquota.ko and fsfilt_ldiskfs.ko modules in sync?\n",
2103 static int fsfilt_ext3_qids(struct file *file, struct inode *inode, int type,
2106 return lustre_get_qids(file, inode, type, list);
2109 static int fsfilt_ext3_dquot(struct lustre_dquot *dquot, int cmd)
2114 if (dquot->dq_info->qi_files[dquot->dq_type] == NULL) {
2115 CERROR("operate dquot before it's enabled!\n");
2120 case QFILE_RD_DQUOT:
2121 rc = lustre_read_dquot(dquot);
2123 case QFILE_WR_DQUOT:
2124 if (dquot->dq_dqb.dqb_ihardlimit ||
2125 dquot->dq_dqb.dqb_isoftlimit ||
2126 dquot->dq_dqb.dqb_bhardlimit ||
2127 dquot->dq_dqb.dqb_bsoftlimit)
2128 cfs_clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2130 cfs_set_bit(DQ_FAKE_B, &dquot->dq_flags);
2132 rc = lustre_commit_dquot(dquot);
2137 CERROR("Unsupported admin quota file cmd %d\n", cmd);
2144 static int fsfilt_ext3_get_mblk(struct super_block *sb, int *count,
2145 struct inode *inode, int frags)
2147 #ifdef EXT3_EXT_HAS_NO_TREE
2148 struct ext3_ext_base *base = inode;
2150 struct ext3_extents_tree tree;
2151 struct ext3_ext_base *base = &tree;
2153 ext3_init_tree_desc(base, inode);
2155 /* for an ost_write request, it needs <#fragments> * <tree depth + 1>
2156 * metablocks at maxium b=16542 */
2157 *count = frags * (EXT_DEPTH(base) + 1) * EXT3_BLOCK_SIZE(sb);
2163 lvfs_sbdev_type fsfilt_ext3_journal_sbdev(struct super_block *sb)
2165 return (EXT3_SB(sb)->journal_bdev);
2167 EXPORT_SYMBOL(fsfilt_ext3_journal_sbdev);
2169 ssize_t lustre_read_quota(struct file *f, struct inode *inode, int type,
2170 char *buf, int count, loff_t pos)
2176 CERROR("lustre_read_quota failed for no quota file!\n");
2177 libcfs_debug_dumpstack(NULL);
2181 /* Support for both adm and op quota files must be provided */
2183 rc = fsfilt_ext3_read_record(f, buf, count, &p);
2184 rc = rc < 0 ? rc : p - pos;
2186 struct super_block *sb = inode->i_sb;
2187 rc = sb->s_op->quota_read(sb, type, buf, count, pos);
2192 ssize_t lustre_write_quota(struct file *f, char *buf, int count, loff_t pos)
2197 /* Only adm quota files are supported, op updates are handled by vfs */
2198 rc = fsfilt_ext3_write_record(f, buf, count, &p, 0);
2199 rc = rc < 0 ? rc : p - pos;
2204 void *lustre_quota_journal_start(struct inode *inode, int delete)
2207 unsigned block_count;
2210 /* each indirect block (+4) may become free, attaching to the
2211 * header list of free blocks (+1); the data block (+1) may
2212 * become a free block (+0) or a block with free dqentries (+0) */
2213 block_count = (4 + 1) + 1;
2214 handle = ext3_journal_start(inode,
2215 block_count*EXT3_DATA_TRANS_BLOCKS(inode->i_sb)+2);
2217 /* indirect blocks are touched (+4), each causes file expansion (+0) or
2218 * freeblk reusage with a header update (+1); dqentry is either reused
2219 * causing update of the entry block (+1), prev (+1) and next (+1) or
2220 * a new block allocation (+1) with a header update (+1) */
2221 block_count = (4 + 1) + 3;
2222 handle = ext3_journal_start(inode,
2223 block_count*EXT3_DATA_TRANS_BLOCKS(inode->i_sb)+2);
2230 void lustre_quota_journal_stop(void *handle)
2232 ext3_journal_stop((handle_t *)handle);
2235 static int ll_decode_fh_accept(void *context, struct dentry *de)
2240 #ifdef HAVE_EXPORTFS_DECODE_FH
2241 # define ll_exportfs_decode_fh(mnt, fid, len, type, acceptable, context) \
2242 exportfs_decode_fh(mnt, (struct fid*)(fid), len, type, \
2243 acceptable, context)
2245 # define ll_exportfs_decode_fh(mnt, fid, len, type, acceptable, context) \
2246 export_op_default.decode_fh((mnt)->mnt_sb, &(fid)->ino, len, \
2247 type, acceptable, context)
2248 # define FILEID_INO32_GEN 1
2249 extern struct export_operations export_op_default;
2252 struct dentry *fsfilt_ext3_fid2dentry(struct vfsmount *mnt,
2253 struct fsfilt_fid *fid, int ignore_gen)
2255 struct inode *inode;
2256 struct dentry *result;
2258 result = ll_exportfs_decode_fh(mnt, fid, 2, FILEID_INO32_GEN,
2259 ll_decode_fh_accept, NULL);
2260 if (IS_ERR(result)) {
2261 CDEBUG(D_DENTRY, "%s of %u/%u failed %ld\n", __func__,
2262 fid->ino, fid->gen, PTR_ERR(result));
2266 CDEBUG(D_DENTRY, "%s of %u/%u succeeded\n", __func__,
2267 fid->ino, fid->gen);
2268 inode = result->d_inode;
2272 if (inode->i_nlink == 0 &&
2273 inode->i_mode == 0 && LTIME_S(inode->i_ctime) == 0) {
2274 LCONSOLE_WARN("Found inode with zero nlink, mode and"
2275 " ctime -- this may indicate disk "
2276 "corruption (inode: %lu, link: %lu, "
2277 "count: %d)\n", inode->i_ino,
2278 (unsigned long)inode->i_nlink,
2279 atomic_read(&inode->i_count));
2282 if (fid->gen && inode->i_generation != fid->gen) {
2283 /* we didn't find the right inode.. */
2284 CDEBUG(D_INODE, "found wrong generation: inode %lu, link: %lu, "
2285 "count: %d, generation %u/%u\n",
2286 inode->i_ino, (unsigned long)inode->i_nlink,
2287 atomic_read(&inode->i_count), inode->i_generation,
2295 return ERR_PTR(-ENOENT);
2298 static struct fsfilt_operations fsfilt_ext3_ops = {
2300 .fs_owner = THIS_MODULE,
2301 .fs_getlabel = fsfilt_ext3_get_label,
2302 .fs_setlabel = fsfilt_ext3_set_label,
2303 .fs_uuid = fsfilt_ext3_uuid,
2304 .fs_start = fsfilt_ext3_start,
2305 .fs_brw_start = fsfilt_ext3_brw_start,
2306 .fs_extend = fsfilt_ext3_extend,
2307 .fs_commit = fsfilt_ext3_commit,
2308 .fs_commit_async = fsfilt_ext3_commit_async,
2309 .fs_commit_wait = fsfilt_ext3_commit_wait,
2310 .fs_setattr = fsfilt_ext3_setattr,
2311 .fs_iocontrol = fsfilt_ext3_iocontrol,
2312 .fs_set_md = fsfilt_ext3_set_md,
2313 .fs_get_md = fsfilt_ext3_get_md,
2314 .fs_readpage = fsfilt_ext3_readpage,
2315 .fs_add_journal_cb = fsfilt_ext3_add_journal_cb,
2316 .fs_statfs = fsfilt_ext3_statfs,
2317 .fs_sync = fsfilt_ext3_sync,
2318 .fs_map_inode_pages = fsfilt_ext3_map_inode_pages,
2319 .fs_write_record = fsfilt_ext3_write_record,
2320 .fs_read_record = fsfilt_ext3_read_record,
2321 .fs_setup = fsfilt_ext3_setup,
2322 .fs_send_bio = fsfilt_ext3_send_bio,
2323 .fs_get_op_len = fsfilt_ext3_get_op_len,
2324 #ifdef HAVE_DISK_INODE_VERSION
2325 .fs_get_version = fsfilt_ext3_get_version,
2326 .fs_set_version = fsfilt_ext3_set_version,
2328 #ifdef HAVE_QUOTA_SUPPORT
2329 .fs_quotactl = fsfilt_ext3_quotactl,
2330 .fs_quotacheck = fsfilt_ext3_quotacheck,
2331 .fs_quotainfo = fsfilt_ext3_quotainfo,
2332 .fs_qids = fsfilt_ext3_qids,
2333 .fs_dquot = fsfilt_ext3_dquot,
2334 .fs_get_mblk = fsfilt_ext3_get_mblk,
2336 .fs_journal_sbdev = fsfilt_ext3_journal_sbdev,
2337 .fs_fid2dentry = fsfilt_ext3_fid2dentry,
2340 static int __init fsfilt_ext3_init(void)
2344 fcb_cache = cfs_mem_cache_create("fsfilt_ext3_fcb",
2345 sizeof(struct fsfilt_cb_data), 0, 0);
2347 CERROR("error allocating fsfilt journal callback cache\n");
2348 GOTO(out, rc = -ENOMEM);
2351 rc = fsfilt_register_ops(&fsfilt_ext3_ops);
2354 int err = cfs_mem_cache_destroy(fcb_cache);
2355 LASSERTF(err == 0, "error destroying new cache: rc %d\n", err);
2361 static void __exit fsfilt_ext3_exit(void)
2365 fsfilt_unregister_ops(&fsfilt_ext3_ops);
2366 rc = cfs_mem_cache_destroy(fcb_cache);
2367 LASSERTF(rc == 0, "couldn't destroy fcb_cache slab\n");
2370 module_init(fsfilt_ext3_init);
2371 module_exit(fsfilt_ext3_exit);
2373 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2374 MODULE_DESCRIPTION("Lustre ext3 Filesystem Helper v0.1");
2375 MODULE_LICENSE("GPL");