1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
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_EXT4_LDISKFS
50 #include <ext4/ext4.h>
51 #include <ext4/ext4_jbd2.h>
53 #include <linux/jbd.h>
54 #include <linux/ext3_fs.h>
55 #include <linux/ext3_jbd.h>
57 #include <linux/version.h>
58 #include <linux/bitops.h>
59 #include <linux/quota.h>
60 #ifdef HAVE_QUOTAIO_V1_H
61 # include <linux/quotaio_v1.h>
62 # include <linux/quotaio_v2.h>
64 # include <quotaio_v1.h>
65 # include <quotaio_v2.h>
66 # include <quota_tree.h>
67 # define V2_DQTREEOFF QT_TREEOFF
70 #if defined(HAVE_EXT3_XATTR_H)
71 #include <ext3/xattr.h>
73 /* ext3 xattr.h not available in rh style kernel-devel rpm */
74 extern int ext3_xattr_get(struct inode *, int, const char *, void *, size_t);
75 extern int ext3_xattr_set_handle(handle_t *, struct inode *, int, const char *, const void *, size_t, int);
78 #include <libcfs/libcfs.h>
79 #include <lustre_fsfilt.h>
81 #include <lustre_quota.h>
82 #include <linux/lustre_compat25.h>
83 #include <linux/lprocfs_status.h>
85 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
86 #ifdef HAVE_EXT4_LDISKFS
87 #include <ext4/ext4_extents.h>
89 #include <linux/ext3_extents.h>
93 #include "lustre_quota_fmt.h"
95 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,15)
96 #define FSFILT_DATA_TRANS_BLOCKS(sb) EXT3_DATA_TRANS_BLOCKS
97 #define FSFILT_DELETE_TRANS_BLOCKS(sb) EXT3_DELETE_TRANS_BLOCKS
99 #define FSFILT_DATA_TRANS_BLOCKS(sb) EXT3_DATA_TRANS_BLOCKS(sb)
100 #define FSFILT_DELETE_TRANS_BLOCKS(sb) EXT3_DELETE_TRANS_BLOCKS(sb)
103 #ifdef EXT3_SINGLEDATA_TRANS_BLOCKS_HAS_SB
104 /* for kernels 2.6.18 and later */
105 #define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS(sb)
107 #define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS
110 static cfs_mem_cache_t *fcb_cache;
112 struct fsfilt_cb_data {
113 struct journal_callback cb_jcb; /* jbd private data - MUST BE FIRST */
114 fsfilt_cb_t cb_func; /* MDS/OBD completion function */
115 struct obd_device *cb_obd; /* MDS/OBD completion device */
116 __u64 cb_last_rcvd; /* MDS/OST last committed operation */
117 void *cb_data; /* MDS/OST completion function data */
120 #ifndef EXT3_XATTR_INDEX_TRUSTED /* temporary until we hit l28 kernel */
121 #define EXT3_XATTR_INDEX_TRUSTED 4
124 #ifdef HAVE_EXT4_LDISKFS
125 #define fsfilt_log_start_commit(journal, tid) jbd2_log_start_commit(journal, tid)
126 #define fsfilt_log_wait_commit(journal, tid) jbd2_log_wait_commit(journal, tid)
127 #define fsfilt_journal_callback_set(handle, func, jcb) jbd2_journal_callback_set(handle, func, jcb)
129 #define fsfilt_log_start_commit(journal, tid) log_start_commit(journal, tid)
130 #define fsfilt_log_wait_commit(journal, tid) log_wait_commit(journal, tid)
131 #define fsfilt_journal_callback_set(handle, func, jcb) journal_callback_set(handle, func, jcb)
132 #define ext_pblock(ex) le32_to_cpu((ex)->ee_start)
133 #define ext3_ext_store_pblock(ex, pblock) ((ex)->ee_start = cpu_to_le32(pblock))
134 #define ext3_inode_bitmap(sb,desc) le32_to_cpu((desc)->bg_inode_bitmap)
137 #ifndef ext3_find_next_bit
138 #define ext3_find_next_bit ext2_find_next_bit
141 #ifndef ext2_find_next_bit
142 #ifdef __LITTLE_ENDIAN
143 #define ext2_find_next_bit(addr, size, off) find_next_bit((unsigned long *)(addr), (size), (off))
145 error "Need implementation of find_next_bit on big-endian systems"
146 #endif /* __LITTLE_ENDIAN */
147 #endif /* !ext2_find_next_le_bit */
149 static char *fsfilt_ext3_get_label(struct super_block *sb)
151 return EXT3_SB(sb)->s_es->s_volume_name;
154 static int fsfilt_ext3_set_label(struct super_block *sb, char *label)
156 /* see e.g. fsfilt_ext3_write_record() */
161 journal = EXT3_SB(sb)->s_journal;
162 handle = ext3_journal_start_sb(sb, 1);
163 if (IS_ERR(handle)) {
164 CERROR("can't start transaction\n");
165 return(PTR_ERR(handle));
168 err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
172 memcpy(EXT3_SB(sb)->s_es->s_volume_name, label,
173 sizeof(EXT3_SB(sb)->s_es->s_volume_name));
175 err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
178 ext3_journal_stop(handle);
183 static char *fsfilt_ext3_uuid(struct super_block *sb)
185 return EXT3_SB(sb)->s_es->s_uuid;
188 #ifdef HAVE_DISK_INODE_VERSION
190 static __u64 get_i_version(struct inode *inode)
192 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,27)) && defined(HAVE_EXT4_LDISKFS)
193 return inode->i_version;
195 return EXT3_I(inode)->i_fs_version;
199 static void set_i_version(struct inode *inode, __u64 new_version)
201 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,27)) && defined(HAVE_EXT4_LDISKFS)
202 inode->i_version = new_version;
204 (EXT3_I(inode))->i_fs_version = new_version;
209 * Get the 64-bit version for an inode.
211 static __u64 fsfilt_ext3_get_version(struct inode *inode)
213 CDEBUG(D_INFO, "Get version "LPX64" for inode %lu\n",
214 get_i_version(inode), inode->i_ino);
215 return get_i_version(inode);
219 * Set the 64-bit version and return the old version.
221 static __u64 fsfilt_ext3_set_version(struct inode *inode, __u64 new_version)
223 __u64 old_version = get_i_version(inode);
225 CDEBUG(D_INFO, "Set version "LPX64" (old "LPX64") for inode %lu\n",
226 new_version, old_version, inode->i_ino);
227 set_i_version(inode, new_version);
228 /* version is set after all inode operations are finished, so we should
229 * mark it dirty here */
230 inode->i_sb->s_op->dirty_inode(inode);
237 * We don't currently need any additional blocks for rmdir and
238 * unlink transactions because we are storing the OST oa_id inside
239 * the inode (which we will be changing anyways as part of this
242 static void *fsfilt_ext3_start(struct inode *inode, int op, void *desc_private,
245 /* For updates to the last received file */
246 int nblocks = FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
250 if (current->journal_info) {
251 CDEBUG(D_INODE, "increasing refcount on %p\n",
252 current->journal_info);
257 case FSFILT_OP_RMDIR:
258 case FSFILT_OP_UNLINK:
259 /* delete one file + create/update logs for each stripe */
260 nblocks += FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb);
261 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
262 FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
264 case FSFILT_OP_RENAME:
265 /* modify additional directory */
266 nblocks += FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
268 case FSFILT_OP_SYMLINK:
269 /* additional block + block bitmap + GDT for long symlink */
272 case FSFILT_OP_CREATE: {
273 #if defined(EXT3_EXTENTS_FL) && defined(EXT3_INDEX_FL) && !defined(HAVE_EXT4_LDISKFS)
276 if (!test_opt(inode->i_sb, EXTENTS)) {
278 } else if (((EXT3_I(inode)->i_flags &
279 cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL)) ==
280 cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL))) {
281 CWARN("extent-mapped directory found with "
282 "ext3-based ldiskfs - contact "
283 "http://bugzilla.lustre.org/\n");
290 case FSFILT_OP_MKDIR:
291 case FSFILT_OP_MKNOD:
292 /* modify one inode + block bitmap + GDT */
296 /* modify parent directory */
297 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
298 FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
299 /* create/update logs for each stripe */
300 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
301 FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
303 case FSFILT_OP_SETATTR:
304 /* Setattr on inode */
306 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
307 FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
308 /* quota chown log for each stripe */
309 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
310 FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
312 case FSFILT_OP_CANCEL_UNLINK:
313 /* blocks for log header bitmap update OR
314 * blocks for catalog header bitmap update + unlink of logs */
315 nblocks = (LLOG_CHUNK_SIZE >> inode->i_blkbits) +
316 FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb) * logs;
318 default: CERROR("unknown transaction start op %d\n", op);
322 LASSERT(current->journal_info == desc_private);
323 journal = EXT3_SB(inode->i_sb)->s_journal;
324 if (nblocks > journal->j_max_transaction_buffers) {
325 CWARN("too many credits %d for op %ux%u using %d instead\n",
326 nblocks, op, logs, journal->j_max_transaction_buffers);
327 nblocks = journal->j_max_transaction_buffers;
331 LASSERTF(nblocks > 0, "can't start %d credit transaction\n", nblocks);
332 handle = ext3_journal_start(inode, nblocks);
335 LASSERT(current->journal_info == handle);
337 CERROR("error starting handle for op %u (%u credits): rc %ld\n",
338 op, nblocks, PTR_ERR(handle));
343 * Calculate the number of buffer credits needed to write multiple pages in
344 * a single ext3 transaction. No, this shouldn't be here, but as yet ext3
345 * doesn't have a nice API for calculating this sort of thing in advance.
347 * See comment above ext3_writepage_trans_blocks for details. We assume
348 * no data journaling is being done, but it does allow for all of the pages
349 * being non-contiguous. If we are guaranteed contiguous pages we could
350 * reduce the number of (d)indirect blocks a lot.
352 * With N blocks per page and P pages, for each inode we have at most:
354 * min(N*P, blocksize/4 + 1) dindirect blocks
357 * For the entire filesystem, we have at most:
358 * min(sum(nindir + P), ngroups) bitmap blocks (from the above)
359 * min(sum(nindir + P), gdblocks) group descriptor blocks (from the above)
360 * objcount inode blocks
362 * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quota files
364 * 1 EXT3_DATA_TRANS_BLOCKS for the last_rcvd update.
366 static int fsfilt_ext3_credits_needed(int objcount, struct fsfilt_objinfo *fso,
367 int niocount, struct niobuf_local *nb)
369 struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
371 const int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
372 int nbitmaps = 0, ngdblocks;
373 int needed = objcount + 1; /* inodes + superblock */
376 for (i = 0, j = 0; i < objcount; i++, fso++) {
377 /* two or more dindirect blocks in case we cross boundary */
378 int ndind = (long)((nb[j + fso->fso_bufcnt - 1].offset -
380 sb->s_blocksize_bits) /
381 (EXT3_ADDR_PER_BLOCK(sb) * EXT3_ADDR_PER_BLOCK(sb));
382 nbitmaps += min(fso->fso_bufcnt, ndind > 0 ? ndind : 2);
384 /* leaf, indirect, tindirect blocks for first block */
385 nbitmaps += blockpp + 2;
387 j += fso->fso_bufcnt;
390 next_indir = nb[0].offset +
391 (EXT3_ADDR_PER_BLOCK(sb) << sb->s_blocksize_bits);
392 for (i = 1; i < niocount; i++) {
393 if (nb[i].offset >= next_indir) {
394 nbitmaps++; /* additional indirect */
395 next_indir = nb[i].offset +
396 (EXT3_ADDR_PER_BLOCK(sb)<<sb->s_blocksize_bits);
397 } else if (nb[i].offset != nb[i - 1].offset + sb->s_blocksize) {
398 nbitmaps++; /* additional indirect */
400 nbitmaps += blockpp; /* each leaf in different group? */
403 ngdblocks = nbitmaps;
404 if (nbitmaps > EXT3_SB(sb)->s_groups_count)
405 nbitmaps = EXT3_SB(sb)->s_groups_count;
406 if (ngdblocks > EXT3_SB(sb)->s_gdb_count)
407 ngdblocks = EXT3_SB(sb)->s_gdb_count;
409 needed += nbitmaps + ngdblocks;
411 /* last_rcvd update */
412 needed += FSFILT_DATA_TRANS_BLOCKS(sb);
414 #if defined(CONFIG_QUOTA)
415 /* We assume that there will be 1 bit set in s_dquot.flags for each
416 * quota file that is active. This is at least true for now.
418 needed += hweight32(ll_sb_any_quota_active(sb)) *
419 FSFILT_SINGLEDATA_TRANS_BLOCKS(sb);
425 /* We have to start a huge journal transaction here to hold all of the
426 * metadata for the pages being written here. This is necessitated by
427 * the fact that we do lots of prepare_write operations before we do
428 * any of the matching commit_write operations, so even if we split
429 * up to use "smaller" transactions none of them could complete until
430 * all of them were opened. By having a single journal transaction,
431 * we eliminate duplicate reservations for common blocks like the
432 * superblock and group descriptors or bitmaps.
434 * We will start the transaction here, but each prepare_write will
435 * add a refcount to the transaction, and each commit_write will
436 * remove a refcount. The transaction will be closed when all of
437 * the pages have been written.
439 static void *fsfilt_ext3_brw_start(int objcount, struct fsfilt_objinfo *fso,
440 int niocount, struct niobuf_local *nb,
441 void *desc_private, int logs)
448 LASSERT(current->journal_info == desc_private);
449 journal = EXT3_SB(fso->fso_dentry->d_inode->i_sb)->s_journal;
450 needed = fsfilt_ext3_credits_needed(objcount, fso, niocount, nb);
452 /* The number of blocks we could _possibly_ dirty can very large.
453 * We reduce our request if it is absurd (and we couldn't get that
454 * many credits for a single handle anyways).
456 * At some point we have to limit the size of I/Os sent at one time,
457 * increase the size of the journal, or we have to calculate the
458 * actual journal requirements more carefully by checking all of
459 * the blocks instead of being maximally pessimistic. It remains to
460 * be seen if this is a real problem or not.
462 if (needed > journal->j_max_transaction_buffers) {
463 CERROR("want too many journal credits (%d) using %d instead\n",
464 needed, journal->j_max_transaction_buffers);
465 needed = journal->j_max_transaction_buffers;
468 LASSERTF(needed > 0, "can't start %d credit transaction\n", needed);
469 handle = ext3_journal_start(fso->fso_dentry->d_inode, needed);
470 if (IS_ERR(handle)) {
471 CERROR("can't get handle for %d credits: rc = %ld\n", needed,
474 LASSERT(handle->h_buffer_credits >= needed);
475 LASSERT(current->journal_info == handle);
481 static int fsfilt_ext3_extend(struct inode *inode, unsigned int nblocks,void *h)
483 handle_t *handle = h;
485 /* fsfilt_extend called with nblocks = 0 for testing in special cases */
487 handle->h_buffer_credits = 0;
488 CWARN("setting credits of handle %p to zero by request\n", h);
491 if (handle->h_buffer_credits > nblocks)
493 if (ext3_journal_extend(handle, nblocks) == 0)
496 ext3_mark_inode_dirty(handle, inode);
497 return ext3_journal_restart(handle, nblocks);
500 static int fsfilt_ext3_commit(struct inode *inode, void *h, int force_sync)
503 handle_t *handle = h;
505 LASSERT(current->journal_info == handle);
507 handle->h_sync = 1; /* recovery likes this */
509 rc = ext3_journal_stop(handle);
514 static int fsfilt_ext3_commit_async(struct inode *inode, void *h,
518 transaction_t *transaction;
519 handle_t *handle = h;
523 LASSERT(current->journal_info == handle);
525 transaction = handle->h_transaction;
526 journal = transaction->t_journal;
527 tid = transaction->t_tid;
528 /* we don't want to be blocked */
530 rc = ext3_journal_stop(handle);
532 CERROR("error while stopping transaction: %d\n", rc);
535 fsfilt_log_start_commit(journal, tid);
537 *wait_handle = (void *) tid;
538 CDEBUG(D_INODE, "commit async: %lu\n", (unsigned long) tid);
542 static int fsfilt_ext3_commit_wait(struct inode *inode, void *h)
544 journal_t *journal = EXT3_JOURNAL(inode);
545 tid_t tid = (tid_t)(long)h;
547 CDEBUG(D_INODE, "commit wait: %lu\n", (unsigned long) tid);
548 if (unlikely(is_journal_aborted(journal)))
551 fsfilt_log_wait_commit(EXT3_JOURNAL(inode), tid);
553 if (unlikely(is_journal_aborted(journal)))
558 static int fsfilt_ext3_setattr(struct dentry *dentry, void *handle,
559 struct iattr *iattr, int do_trunc)
561 struct inode *inode = dentry->d_inode;
564 /* Avoid marking the inode dirty on the superblock list unnecessarily.
565 * We are already writing the inode to disk as part of this
566 * transaction and want to avoid a lot of extra inode writeout
567 * later on. b=9828 */
568 if (iattr->ia_valid & ATTR_SIZE && !do_trunc) {
569 /* ATTR_SIZE would invoke truncate: clear it */
570 iattr->ia_valid &= ~ATTR_SIZE;
571 EXT3_I(inode)->i_disksize = iattr->ia_size;
572 i_size_write(inode, iattr->ia_size);
574 if (iattr->ia_valid & ATTR_UID)
575 inode->i_uid = iattr->ia_uid;
576 if (iattr->ia_valid & ATTR_GID)
577 inode->i_gid = iattr->ia_gid;
578 if (iattr->ia_valid & ATTR_ATIME)
579 inode->i_atime = iattr->ia_atime;
580 if (iattr->ia_valid & ATTR_MTIME)
581 inode->i_mtime = iattr->ia_mtime;
582 if (iattr->ia_valid & ATTR_CTIME)
583 inode->i_ctime = iattr->ia_ctime;
584 if (iattr->ia_valid & ATTR_MODE) {
585 inode->i_mode = iattr->ia_mode;
587 if (!cfs_curproc_is_in_groups(inode->i_gid) &&
588 !cfs_capable(CFS_CAP_FSETID))
589 inode->i_mode &= ~S_ISGID;
592 inode->i_sb->s_op->dirty_inode(inode);
597 /* Don't allow setattr to change file type */
598 if (iattr->ia_valid & ATTR_MODE)
599 iattr->ia_mode = (inode->i_mode & S_IFMT) |
600 (iattr->ia_mode & ~S_IFMT);
602 /* We set these flags on the client, but have already checked perms
603 * so don't confuse inode_change_ok. */
604 iattr->ia_valid &= ~(ATTR_MTIME_SET | ATTR_ATIME_SET);
606 if (inode->i_op->setattr) {
607 rc = inode->i_op->setattr(dentry, iattr);
609 rc = inode_change_ok(inode, iattr);
611 rc = inode_setattr(inode, iattr);
618 static int fsfilt_ext3_iocontrol(struct inode *inode, struct file *file,
619 unsigned int cmd, unsigned long arg)
624 /* FIXME: Can't do this because of nested transaction deadlock */
625 if (cmd == EXT3_IOC_SETFLAGS && (*(int *)arg) & EXT3_JOURNAL_DATA_FL) {
626 CERROR("can't set data journal flag on file\n");
630 #ifdef HAVE_EXT4_LDISKFS
631 /* ext4_ioctl does not have a inode argument */
632 if (inode->i_fop->unlocked_ioctl)
633 rc = inode->i_fop->unlocked_ioctl(file, cmd, arg);
635 if (inode->i_fop->ioctl)
636 rc = inode->i_fop->ioctl(inode, file, cmd, arg);
644 static int fsfilt_ext3_set_md(struct inode *inode, void *handle,
645 void *lmm, int lmm_size, const char *name)
649 LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
651 rc = ext3_xattr_set_handle(handle, inode, EXT3_XATTR_INDEX_TRUSTED,
652 name, lmm, lmm_size, XATTR_NO_CTIME);
655 if (rc && rc != -EROFS)
656 CERROR("error adding MD data to inode %lu: rc = %d\n",
661 /* Must be called with i_mutex held */
662 static int fsfilt_ext3_get_md(struct inode *inode, void *lmm, int lmm_size,
667 LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
669 rc = ext3_xattr_get(inode, EXT3_XATTR_INDEX_TRUSTED,
670 name, lmm, lmm_size);
672 /* This gives us the MD size */
674 return (rc == -ENODATA) ? 0 : rc;
677 CDEBUG(D_INFO, "error getting EA %d/%s from inode %lu: rc %d\n",
678 EXT3_XATTR_INDEX_TRUSTED, name,
680 memset(lmm, 0, lmm_size);
681 return (rc == -ENODATA) ? 0 : rc;
687 static int fsfilt_ext3_send_bio(int rw, struct inode *inode, struct bio *bio)
693 static ssize_t fsfilt_ext3_readpage(struct file *file, char *buf, size_t count,
696 struct inode *inode = file->f_dentry->d_inode;
699 if (S_ISREG(inode->i_mode))
700 rc = file->f_op->read(file, buf, count, off);
702 const int blkbits = inode->i_sb->s_blocksize_bits;
703 const int blksize = inode->i_sb->s_blocksize;
705 CDEBUG(D_EXT2, "reading %lu at dir %lu+%llu\n",
706 (unsigned long)count, inode->i_ino, *off);
708 struct buffer_head *bh;
711 if (*off < i_size_read(inode)) {
714 bh = ext3_bread(NULL, inode, *off >> blkbits,
717 CDEBUG(D_EXT2, "read %u@%llu\n", blksize, *off);
720 memcpy(buf, bh->b_data, blksize);
723 /* XXX in theory we should just fake
724 * this buffer and continue like ext3,
725 * especially if this is a partial read
727 CERROR("error read dir %lu+%llu: %d\n",
728 inode->i_ino, *off, err);
733 struct ext3_dir_entry_2 *fake = (void *)buf;
735 CDEBUG(D_EXT2, "fake %u@%llu\n", blksize, *off);
736 memset(fake, 0, sizeof(*fake));
737 fake->rec_len = cpu_to_le16(blksize);
749 static void fsfilt_ext3_cb_func(struct journal_callback *jcb, int error)
751 struct fsfilt_cb_data *fcb = (struct fsfilt_cb_data *)jcb;
753 fcb->cb_func(fcb->cb_obd, fcb->cb_last_rcvd, fcb->cb_data, error);
755 OBD_SLAB_FREE(fcb, fcb_cache, sizeof *fcb);
758 static int fsfilt_ext3_add_journal_cb(struct obd_device *obd, __u64 last_rcvd,
759 void *handle, fsfilt_cb_t cb_func,
762 struct fsfilt_cb_data *fcb;
764 OBD_SLAB_ALLOC_PTR_GFP(fcb, fcb_cache, CFS_ALLOC_IO);
768 fcb->cb_func = cb_func;
770 fcb->cb_last_rcvd = last_rcvd;
771 fcb->cb_data = cb_data;
773 CDEBUG(D_EXT2, "set callback for last_rcvd: "LPD64"\n", last_rcvd);
774 fsfilt_journal_callback_set(handle, fsfilt_ext3_cb_func,
775 (struct journal_callback *)fcb);
781 * We need to hack the return value for the free inode counts because
782 * the current EA code requires one filesystem block per inode with EAs,
783 * so it is possible to run out of blocks before we run out of inodes.
785 * This can be removed when the ext3 EA code is fixed.
787 static int fsfilt_ext3_statfs(struct super_block *sb, struct obd_statfs *osfs)
792 memset(&sfs, 0, sizeof(sfs));
793 rc = ll_do_statfs(sb, &sfs);
794 if (!rc && sfs.f_bfree < sfs.f_ffree) {
795 sfs.f_files = (sfs.f_files - sfs.f_ffree) + sfs.f_bfree;
796 sfs.f_ffree = sfs.f_bfree;
799 statfs_pack(osfs, &sfs);
803 static int fsfilt_ext3_sync(struct super_block *sb)
805 return ext3_force_commit(sb);
808 #if defined(EXT3_MULTIBLOCK_ALLOCATOR) && (!defined(EXT3_EXT_CACHE_NO) || defined(EXT_CACHE_MARK))
809 #warning "kernel code has old extents/mballoc patch, disabling"
810 #undef EXT3_MULTIBLOCK_ALLOCATOR
812 #ifndef EXT3_EXTENTS_FL
813 #define EXT3_EXTENTS_FL 0x00080000 /* Inode uses extents */
816 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
817 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
818 #define fsfilt_up_truncate_sem(inode) up(&EXT3_I(inode)->truncate_sem);
819 #define fsfilt_down_truncate_sem(inode) down(&EXT3_I(inode)->truncate_sem);
821 #ifdef HAVE_EXT4_LDISKFS
822 #define fsfilt_up_truncate_sem(inode) up_write((&EXT4_I(inode)->i_data_sem));
823 #define fsfilt_down_truncate_sem(inode) down_write((&EXT4_I(inode)->i_data_sem));
825 #define fsfilt_up_truncate_sem(inode) mutex_unlock(&EXT3_I(inode)->truncate_mutex);
826 #define fsfilt_down_truncate_sem(inode) mutex_lock(&EXT3_I(inode)->truncate_mutex);
831 #define EXT_ASSERT(cond) BUG_ON(!(cond))
834 #ifdef EXT3_EXT_HAS_NO_TREE
835 /* for kernels 2.6.18 and later */
836 #ifdef HAVE_EXT4_LDISKFS
837 #define EXT_GENERATION(inode) (EXT4_I(inode)->i_ext_generation)
839 #define EXT_GENERATION(inode) ext_generation(inode)
841 #define ext3_ext_base inode
842 #define ext3_ext_base2inode(inode) (inode)
843 #define EXT_DEPTH(inode) ext_depth(inode)
844 #define fsfilt_ext3_ext_walk_space(inode, block, num, cb, cbdata) \
845 ext3_ext_walk_space(inode, block, num, cb, cbdata);
847 #define ext3_ext_base ext3_extents_tree
848 #define ext3_ext_base2inode(tree) (tree->inode)
849 #define fsfilt_ext3_ext_walk_space(tree, block, num, cb, cbdata) \
850 ext3_ext_walk_space(tree, block, num, cb);
853 #include <linux/lustre_version.h>
856 unsigned long *blocks;
864 static long ext3_ext_find_goal(struct inode *inode, struct ext3_ext_path *path,
865 unsigned long block, int *aflags)
867 struct ext3_inode_info *ei = EXT3_I(inode);
868 unsigned long bg_start;
869 unsigned long colour;
873 struct ext3_extent *ex;
874 depth = path->p_depth;
876 /* try to predict block placement */
877 if ((ex = path[depth].p_ext))
878 return ext_pblock(ex) + (block - le32_to_cpu(ex->ee_block));
880 /* it looks index is empty
881 * try to find starting from index itself */
882 if (path[depth].p_bh)
883 return path[depth].p_bh->b_blocknr;
886 /* OK. use inode's group */
887 bg_start = (ei->i_block_group * EXT3_BLOCKS_PER_GROUP(inode->i_sb)) +
888 le32_to_cpu(EXT3_SB(inode->i_sb)->s_es->s_first_data_block);
889 colour = (current->pid % 16) *
890 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
891 return bg_start + colour + block;
894 #define ll_unmap_underlying_metadata(sb, blocknr) \
895 unmap_underlying_metadata((sb)->s_bdev, blocknr)
897 #ifndef EXT3_MB_HINT_GROUP_ALLOC
898 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
899 struct ext3_ext_path *path, unsigned long block,
900 unsigned long *count, int *err)
902 unsigned long pblock, goal;
904 struct inode *inode = ext3_ext_base2inode(base);
906 goal = ext3_ext_find_goal(inode, path, block, &aflags);
907 aflags |= 2; /* block have been already reserved */
908 pblock = ext3_mb_new_blocks(handle, inode, goal, count, aflags, err);
913 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
914 struct ext3_ext_path *path, unsigned long block,
915 unsigned long *count, int *err)
917 struct inode *inode = ext3_ext_base2inode(base);
918 struct ext3_allocation_request ar;
919 unsigned long pblock;
922 /* find neighbour allocated blocks */
924 *err = ext3_ext_search_left(base, path, &ar.lleft, &ar.pleft);
928 *err = ext3_ext_search_right(base, path, &ar.lright, &ar.pright);
932 /* allocate new block */
933 ar.goal = ext3_ext_find_goal(inode, path, block, &aflags);
937 ar.flags = EXT3_MB_HINT_DATA;
938 pblock = ext3_mb_new_blocks(handle, &ar, err);
944 #ifdef EXT3_EXT_HAS_NO_TREE
945 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
946 struct ext3_ext_path *path,
947 struct ext3_ext_cache *cex,
948 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
949 struct ext3_extent *ex,
953 struct bpointers *bp = cbdata;
955 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
956 struct ext3_ext_path *path,
957 struct ext3_ext_cache *cex
958 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
959 , struct ext3_extent *ex
963 struct bpointers *bp = base->private;
965 struct inode *inode = ext3_ext_base2inode(base);
966 struct ext3_extent nex;
967 unsigned long pblock;
974 EXT_ASSERT(i == path->p_depth);
975 EXT_ASSERT(path[i].p_hdr);
977 if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
982 if (bp->create == 0) {
984 if (cex->ec_block < bp->start)
985 i = bp->start - cex->ec_block;
986 if (i >= cex->ec_len)
987 CERROR("nothing to do?! i = %d, e_num = %u\n",
989 for (; i < cex->ec_len && bp->num; i++) {
1001 tgen = EXT_GENERATION(base);
1002 count = ext3_ext_calc_credits_for_insert(base, path);
1003 fsfilt_up_truncate_sem(inode);
1005 handle = ext3_journal_start(inode, count+EXT3_ALLOC_NEEDED+1);
1006 if (IS_ERR(handle)) {
1007 fsfilt_down_truncate_sem(inode);
1008 return PTR_ERR(handle);
1011 fsfilt_down_truncate_sem(inode);
1012 if (tgen != EXT_GENERATION(base)) {
1013 /* the tree has changed. so path can be invalid at moment */
1014 ext3_journal_stop(handle);
1018 count = cex->ec_len;
1019 pblock = new_blocks(handle, base, path, cex->ec_block, &count, &err);
1022 EXT_ASSERT(count <= cex->ec_len);
1024 /* insert new extent */
1025 nex.ee_block = cpu_to_le32(cex->ec_block);
1026 ext3_ext_store_pblock(&nex, pblock);
1027 nex.ee_len = cpu_to_le16(count);
1028 err = ext3_ext_insert_extent(handle, base, path, &nex);
1030 /* free data blocks we just allocated */
1031 /* not a good idea to call discard here directly,
1032 * but otherwise we'd need to call it every free() */
1033 #ifdef EXT3_MB_HINT_GROUP_ALLOC
1034 ext3_mb_discard_inode_preallocations(inode);
1036 ext3_free_blocks(handle, inode, ext_pblock(&nex),
1037 cpu_to_le16(nex.ee_len), 0);
1042 * Putting len of the actual extent we just inserted,
1043 * we are asking ext3_ext_walk_space() to continue
1044 * scaning after that block
1046 cex->ec_len = le16_to_cpu(nex.ee_len);
1047 cex->ec_start = ext_pblock(&nex);
1048 BUG_ON(le16_to_cpu(nex.ee_len) == 0);
1049 BUG_ON(le32_to_cpu(nex.ee_block) != cex->ec_block);
1052 ext3_journal_stop(handle);
1057 CERROR("hmm. why do we find this extent?\n");
1058 CERROR("initial space: %lu:%u\n",
1059 bp->start, bp->init_num);
1060 CERROR("current extent: %u/%u/%llu %d\n",
1061 cex->ec_block, cex->ec_len,
1062 (unsigned long long)cex->ec_start,
1066 if (cex->ec_block < bp->start)
1067 i = bp->start - cex->ec_block;
1068 if (i >= cex->ec_len)
1069 CERROR("nothing to do?! i = %d, e_num = %u\n",
1071 for (; i < cex->ec_len && bp->num; i++) {
1072 *(bp->blocks) = cex->ec_start + i;
1073 if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
1077 /* unmap any possible underlying metadata from
1078 * the block device mapping. bug 6998. */
1079 ll_unmap_underlying_metadata(inode->i_sb,
1091 int fsfilt_map_nblocks(struct inode *inode, unsigned long block,
1092 unsigned long num, unsigned long *blocks,
1093 int *created, int create)
1095 #ifdef EXT3_EXT_HAS_NO_TREE
1096 struct ext3_ext_base *base = inode;
1098 struct ext3_extents_tree tree;
1099 struct ext3_ext_base *base = &tree;
1101 struct bpointers bp;
1104 CDEBUG(D_OTHER, "blocks %lu-%lu requested for inode %u\n",
1105 block, block + num - 1, (unsigned) inode->i_ino);
1107 #ifndef EXT3_EXT_HAS_NO_TREE
1108 ext3_init_tree_desc(base, inode);
1112 bp.created = created;
1114 bp.init_num = bp.num = num;
1117 fsfilt_down_truncate_sem(inode);
1118 err = fsfilt_ext3_ext_walk_space(base, block, num,
1119 ext3_ext_new_extent_cb, &bp);
1120 ext3_ext_invalidate_cache(base);
1121 fsfilt_up_truncate_sem(inode);
1126 int fsfilt_ext3_map_ext_inode_pages(struct inode *inode, struct page **page,
1127 int pages, unsigned long *blocks,
1128 int *created, int create)
1130 int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1132 struct page *fp = NULL;
1135 CDEBUG(D_OTHER, "inode %lu: map %d pages from %lu\n",
1136 inode->i_ino, pages, (*page)->index);
1138 /* pages are sorted already. so, we just have to find
1139 * contig. space and process them properly */
1142 /* start new extent */
1147 } else if (fp->index + clen == (*page)->index) {
1148 /* continue the extent */
1155 /* process found extent */
1156 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1157 clen * blocks_per_page, blocks,
1162 /* look for next extent */
1164 blocks += blocks_per_page * clen;
1165 created += blocks_per_page * clen;
1169 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1170 clen * blocks_per_page, blocks,
1175 #endif /* EXT3_MULTIBLOCK_ALLOCATOR */
1177 extern int ext3_map_inode_page(struct inode *inode, struct page *page,
1178 unsigned long *blocks, int *created, int create);
1179 int fsfilt_ext3_map_bm_inode_pages(struct inode *inode, struct page **page,
1180 int pages, unsigned long *blocks,
1181 int *created, int create)
1183 int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1187 for (i = 0, cr = created, b = blocks; i < pages; i++, page++) {
1188 rc = ext3_map_inode_page(inode, *page, b, cr, create);
1190 CERROR("ino %lu, blk %lu cr %u create %d: rc %d\n",
1191 inode->i_ino, *b, *cr, create, rc);
1195 b += blocks_per_page;
1196 cr += blocks_per_page;
1201 int fsfilt_ext3_map_inode_pages(struct inode *inode, struct page **page,
1202 int pages, unsigned long *blocks,
1203 int *created, int create,
1204 cfs_semaphore_t *optional_sem)
1207 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
1208 if (EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL) {
1209 rc = fsfilt_ext3_map_ext_inode_pages(inode, page, pages,
1210 blocks, created, create);
1214 if (optional_sem != NULL)
1215 cfs_down(optional_sem);
1216 rc = fsfilt_ext3_map_bm_inode_pages(inode, page, pages, blocks,
1218 if (optional_sem != NULL)
1219 cfs_up(optional_sem);
1224 int fsfilt_ext3_read(struct inode *inode, void *buf, int size, loff_t *offs)
1226 unsigned long block;
1227 struct buffer_head *bh;
1228 int err, blocksize, csize, boffs, osize = size;
1230 /* prevent reading after eof */
1232 if (i_size_read(inode) < *offs + size) {
1233 size = i_size_read(inode) - *offs;
1234 cfs_unlock_kernel();
1236 CDEBUG(D_EXT2, "size %llu is too short for read @%llu\n",
1237 i_size_read(inode), *offs);
1239 } else if (size == 0) {
1243 cfs_unlock_kernel();
1246 blocksize = 1 << inode->i_blkbits;
1249 block = *offs >> inode->i_blkbits;
1250 boffs = *offs & (blocksize - 1);
1251 csize = min(blocksize - boffs, size);
1252 bh = ext3_bread(NULL, inode, block, 0, &err);
1254 CERROR("can't read block: %d\n", err);
1258 memcpy(buf, bh->b_data + boffs, csize);
1267 EXPORT_SYMBOL(fsfilt_ext3_read);
1269 static int fsfilt_ext3_read_record(struct file * file, void *buf,
1270 int size, loff_t *offs)
1273 rc = fsfilt_ext3_read(file->f_dentry->d_inode, buf, size, offs);
1279 int fsfilt_ext3_write_handle(struct inode *inode, void *buf, int bufsize,
1280 loff_t *offs, handle_t *handle)
1282 struct buffer_head *bh = NULL;
1283 loff_t old_size = i_size_read(inode), offset = *offs;
1284 loff_t new_size = i_size_read(inode);
1285 unsigned long block;
1286 int err = 0, blocksize = 1 << inode->i_blkbits, size, boffs;
1288 while (bufsize > 0) {
1292 block = offset >> inode->i_blkbits;
1293 boffs = offset & (blocksize - 1);
1294 size = min(blocksize - boffs, bufsize);
1295 bh = ext3_bread(handle, inode, block, 1, &err);
1297 CERROR("can't read/create block: %d\n", err);
1301 err = ext3_journal_get_write_access(handle, bh);
1303 CERROR("journal_get_write_access() returned error %d\n",
1307 LASSERT(bh->b_data + boffs + size <= bh->b_data + bh->b_size);
1308 memcpy(bh->b_data + boffs, buf, size);
1309 err = ext3_journal_dirty_metadata(handle, bh);
1311 CERROR("journal_dirty_metadata() returned error %d\n",
1315 if (offset + size > new_size)
1316 new_size = offset + size;
1324 /* correct in-core and on-disk sizes */
1325 if (new_size > i_size_read(inode)) {
1327 if (new_size > i_size_read(inode))
1328 i_size_write(inode, new_size);
1329 if (i_size_read(inode) > EXT3_I(inode)->i_disksize)
1330 EXT3_I(inode)->i_disksize = i_size_read(inode);
1331 if (i_size_read(inode) > old_size)
1332 mark_inode_dirty(inode);
1333 cfs_unlock_kernel();
1340 EXPORT_SYMBOL(fsfilt_ext3_write_handle);
1342 static int fsfilt_ext3_write_record(struct file *file, void *buf, int bufsize,
1343 loff_t *offs, int force_sync)
1345 struct inode *inode = file->f_dentry->d_inode;
1347 int err, block_count = 0, blocksize;
1349 /* Determine how many transaction credits are needed */
1350 blocksize = 1 << inode->i_blkbits;
1351 block_count = (*offs & (blocksize - 1)) + bufsize;
1352 block_count = (block_count + blocksize - 1) >> inode->i_blkbits;
1354 handle = ext3_journal_start(inode,
1355 block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2);
1356 if (IS_ERR(handle)) {
1357 CERROR("can't start transaction for %d blocks (%d bytes)\n",
1358 block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2, bufsize);
1359 return PTR_ERR(handle);
1362 err = fsfilt_ext3_write_handle(inode, buf, bufsize, offs, handle);
1364 if (!err && force_sync)
1365 handle->h_sync = 1; /* recovery likes this */
1367 ext3_journal_stop(handle);
1372 static int fsfilt_ext3_setup(struct super_block *sb)
1374 #if ((LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,6)) && \
1375 defined(HAVE_QUOTA_SUPPORT)) || defined(S_PDIROPS)
1376 struct ext3_sb_info *sbi = EXT3_SB(sb);
1378 sbi->dx_lock = fsfilt_ext3_dx_lock;
1379 sbi->dx_unlock = fsfilt_ext3_dx_unlock;
1383 CWARN("Enabling PDIROPS\n");
1384 set_opt(sbi->s_mount_opt, PDIROPS);
1385 sb->s_flags |= S_PDIROPS;
1387 if (!EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1388 CWARN("filesystem doesn't have dir_index feature enabled\n");
1389 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,6)) && defined(HAVE_QUOTA_SUPPORT)
1390 /* enable journaled quota support */
1391 /* kfreed in ext3_put_super() */
1392 sbi->s_qf_names[USRQUOTA] = kstrdup("lquota.user.reserved", GFP_KERNEL);
1393 if (!sbi->s_qf_names[USRQUOTA])
1395 sbi->s_qf_names[GRPQUOTA] = kstrdup("lquota.group.reserved", GFP_KERNEL);
1396 if (!sbi->s_qf_names[GRPQUOTA]) {
1397 kfree(sbi->s_qf_names[USRQUOTA]);
1398 sbi->s_qf_names[USRQUOTA] = NULL;
1401 sbi->s_jquota_fmt = QFMT_VFS_V0;
1402 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,13))
1403 set_opt(sbi->s_mount_opt, QUOTA);
1409 /* If fso is NULL, op is FSFILT operation, otherwise op is number of fso
1410 objects. Logs is number of logfiles to update */
1411 static int fsfilt_ext3_get_op_len(int op, struct fsfilt_objinfo *fso, int logs)
1415 case FSFILT_OP_CREATE:
1416 /* directory leaf, index & indirect & EA*/
1417 return 4 + 3 * logs;
1418 case FSFILT_OP_UNLINK:
1424 struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
1425 int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
1426 int addrpp = EXT3_ADDR_PER_BLOCK(sb) * blockpp;
1427 for (i = 0; i < op; i++, fso++) {
1428 int nblocks = fso->fso_bufcnt * blockpp;
1429 int ndindirect = min(nblocks, addrpp + 1);
1430 int nindir = nblocks + ndindirect + 1;
1434 return needed + 3 * logs;
1440 #ifdef HAVE_QUOTA_SUPPORT
1441 #define DQINFO_COPY(out, in) \
1443 Q_COPY(out, in, dqi_bgrace); \
1444 Q_COPY(out, in, dqi_igrace); \
1445 Q_COPY(out, in, dqi_flags); \
1446 Q_COPY(out, in, dqi_valid); \
1449 #define DQBLK_COPY(out, in) \
1451 Q_COPY(out, in, dqb_bhardlimit); \
1452 Q_COPY(out, in, dqb_bsoftlimit); \
1453 Q_COPY(out, in, dqb_curspace); \
1454 Q_COPY(out, in, dqb_ihardlimit); \
1455 Q_COPY(out, in, dqb_isoftlimit); \
1456 Q_COPY(out, in, dqb_curinodes); \
1457 Q_COPY(out, in, dqb_btime); \
1458 Q_COPY(out, in, dqb_itime); \
1459 Q_COPY(out, in, dqb_valid); \
1462 static int fsfilt_ext3_quotactl(struct super_block *sb,
1463 struct obd_quotactl *oqc)
1465 int i, rc = 0, error = 0;
1466 struct quotactl_ops *qcop;
1467 struct if_dqinfo *info;
1468 struct if_dqblk *dqblk;
1474 OBD_ALLOC_PTR(info);
1477 OBD_ALLOC_PTR(dqblk);
1483 DQINFO_COPY(info, &oqc->qc_dqinfo);
1484 DQBLK_COPY(dqblk, &oqc->qc_dqblk);
1487 if (oqc->qc_cmd == Q_QUOTAON || oqc->qc_cmd == Q_QUOTAOFF) {
1488 for (i = 0; i < MAXQUOTAS; i++) {
1489 if (!Q_TYPESET(oqc, i))
1492 if (oqc->qc_cmd == Q_QUOTAON) {
1493 char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1495 LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1497 rc = ll_quota_on(sb, i, QFMT_VFS_V0,
1499 } else if (oqc->qc_cmd == Q_QUOTAOFF) {
1500 rc = ll_quota_off(sb, i, 0);
1508 GOTO(out, rc ?: error);
1511 switch (oqc->qc_cmd) {
1514 if (!qcop->get_info)
1515 GOTO(out, rc = -ENOSYS);
1516 rc = qcop->get_info(sb, oqc->qc_type, info);
1520 if (!qcop->set_dqblk)
1521 GOTO(out, rc = -ENOSYS);
1522 rc = qcop->set_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1526 if (!qcop->get_dqblk)
1527 GOTO(out, rc = -ENOSYS);
1528 rc = qcop->get_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1530 dqblk->dqb_valid = QIF_LIMITS | QIF_USAGE;
1533 if (!sb->s_qcop->quota_sync)
1534 GOTO(out, rc = -ENOSYS);
1535 qcop->quota_sync(sb, oqc->qc_type);
1538 CDEBUG(D_WARNING, "invalidating operational quota files\n");
1539 for (i = 0; i < MAXQUOTAS; i++) {
1541 char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1543 LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1545 if (!Q_TYPESET(oqc, i))
1548 fp = filp_open(name[i], O_CREAT | O_TRUNC | O_RDWR, 0644);
1551 CERROR("error invalidating operational quota file"
1552 " %s (rc:%d)\n", name[i], rc);
1560 CERROR("unsupported quotactl command: %d\n", oqc->qc_cmd);
1564 DQINFO_COPY(&oqc->qc_dqinfo, info);
1565 DQBLK_COPY(&oqc->qc_dqblk, dqblk);
1568 OBD_FREE_PTR(dqblk);
1571 CDEBUG(D_QUOTA, "quotactl command %#x, id %u, type %u "
1573 oqc->qc_cmd, oqc->qc_id, oqc->qc_type, rc);
1578 cfs_hlist_node_t dqb_hash; /** quotacheck hash */
1579 cfs_list_t dqb_list; /** in list also */
1580 qid_t dqb_id; /** uid/gid */
1581 short dqb_type; /** USRQUOTA/GRPQUOTA */
1582 qsize_t dqb_bhardlimit; /** block hard limit */
1583 qsize_t dqb_bsoftlimit; /** block soft limit */
1584 qsize_t dqb_curspace; /** current space */
1585 qsize_t dqb_ihardlimit; /** inode hard limit */
1586 qsize_t dqb_isoftlimit; /** inode soft limit */
1587 qsize_t dqb_curinodes; /** current inodes */
1588 __u64 dqb_btime; /** block grace time */
1589 __u64 dqb_itime; /** inode grace time */
1590 __u32 dqb_valid; /** flag for above fields */
1593 static inline unsigned int chkquot_hash(qid_t id, int type)
1594 __attribute__((__const__));
1596 static inline unsigned int chkquot_hash(qid_t id, int type)
1598 return (id * (MAXQUOTAS - type)) % NR_DQHASH;
1601 static inline struct chk_dqblk *
1602 find_chkquot(cfs_hlist_head_t *head, qid_t id, int type)
1604 cfs_hlist_node_t *node;
1605 struct chk_dqblk *cdqb;
1607 cfs_hlist_for_each(node, head) {
1608 cdqb = cfs_hlist_entry(node, struct chk_dqblk, dqb_hash);
1609 if (cdqb->dqb_id == id && cdqb->dqb_type == type)
1616 static struct chk_dqblk *alloc_chkquot(qid_t id, int type)
1618 struct chk_dqblk *cdqb;
1620 OBD_ALLOC_PTR(cdqb);
1622 CFS_INIT_HLIST_NODE(&cdqb->dqb_hash);
1623 CFS_INIT_LIST_HEAD(&cdqb->dqb_list);
1625 cdqb->dqb_type = type;
1631 static struct chk_dqblk *
1632 cqget(struct super_block *sb, cfs_hlist_head_t *hash,
1633 cfs_list_t *list, qid_t id, int type, int first_check)
1635 cfs_hlist_head_t *head = hash + chkquot_hash(id, type);
1636 struct if_dqblk dqb;
1637 struct chk_dqblk *cdqb;
1640 cdqb = find_chkquot(head, id, type);
1644 cdqb = alloc_chkquot(id, type);
1649 rc = sb->s_qcop->get_dqblk(sb, type, id, &dqb);
1651 CERROR("get_dqblk of id %u, type %d failed: %d\n",
1654 DQBLK_COPY(cdqb, &dqb);
1655 cdqb->dqb_curspace = 0;
1656 cdqb->dqb_curinodes = 0;
1660 cfs_hlist_add_head(&cdqb->dqb_hash, head);
1661 cfs_list_add_tail(&cdqb->dqb_list, list);
1666 static inline int quota_onoff(struct super_block *sb, int cmd, int type, int qfmt)
1668 struct obd_quotactl *oqctl;
1671 OBD_ALLOC_PTR(oqctl);
1675 oqctl->qc_cmd = cmd;
1676 oqctl->qc_id = qfmt;
1677 oqctl->qc_type = type;
1678 rc = fsfilt_ext3_quotactl(sb, oqctl);
1680 OBD_FREE_PTR(oqctl);
1684 static inline int read_old_dqinfo(struct super_block *sb, int type,
1685 struct if_dqinfo *dqinfo)
1687 struct obd_quotactl *oqctl;
1691 OBD_ALLOC_PTR(oqctl);
1695 oqctl->qc_cmd = Q_GETINFO;
1696 oqctl->qc_type = type;
1697 rc = fsfilt_ext3_quotactl(sb, oqctl);
1699 ((struct obd_dqinfo *)dqinfo)[type] = oqctl->qc_dqinfo;
1701 OBD_FREE_PTR(oqctl);
1705 #ifndef HAVE_EXT4_LDISKFS
1706 static inline struct ext3_group_desc *
1707 get_group_desc(struct super_block *sb, int group, struct buffer_head **bh)
1709 unsigned long desc_block, desc;
1710 struct ext3_group_desc *gdp;
1712 desc_block = group / EXT3_DESC_PER_BLOCK(sb);
1713 desc = group % EXT3_DESC_PER_BLOCK(sb);
1714 gdp = (struct ext3_group_desc *)
1715 EXT3_SB(sb)->s_group_desc[desc_block]->b_data;
1720 static inline struct buffer_head *
1721 ext3_read_inode_bitmap(struct super_block *sb, unsigned long group)
1723 struct ext3_group_desc *desc;
1724 struct buffer_head *bh;
1726 desc = get_group_desc(sb, group, NULL);
1727 bh = sb_bread(sb, ext3_inode_bitmap(sb, desc));
1731 static __u32 ext3_itable_unused_count(struct super_block *sb,
1732 struct ext3_group_desc *bg) {
1733 return le16_to_cpu(bg->bg_itable_unused);
1736 #define get_group_desc ext3_get_group_desc
1740 cfs_hlist_head_t qckt_hash[NR_DQHASH]; /* quotacheck hash */
1741 cfs_list_t qckt_list; /* quotacheck list */
1742 int qckt_first_check[MAXQUOTAS]; /* 1 if no old quotafile */
1743 struct if_dqinfo qckt_dqinfo[MAXQUOTAS]; /* old dqinfo */
1746 static int add_inode_quota(struct inode *inode, struct qchk_ctxt *qctxt,
1747 struct obd_quotactl *oqc)
1749 struct chk_dqblk *cdqb[MAXQUOTAS] = { NULL, };
1751 qid_t qid[MAXQUOTAS];
1757 qid[USRQUOTA] = inode->i_uid;
1758 qid[GRPQUOTA] = inode->i_gid;
1760 if (S_ISDIR(inode->i_mode) ||
1761 S_ISREG(inode->i_mode) ||
1762 S_ISLNK(inode->i_mode))
1763 size = inode_get_bytes(inode);
1765 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1766 if (!Q_TYPESET(oqc, cnt))
1769 cdqb[cnt] = cqget(inode->i_sb, qctxt->qckt_hash,
1770 &qctxt->qckt_list, qid[cnt], cnt,
1771 qctxt->qckt_first_check[cnt]);
1777 cdqb[cnt]->dqb_curspace += size;
1778 cdqb[cnt]->dqb_curinodes++;
1782 for (i = 0; i < cnt; i++) {
1783 if (!Q_TYPESET(oqc, i))
1786 cdqb[i]->dqb_curspace -= size;
1787 cdqb[i]->dqb_curinodes--;
1794 /* write dqinfo struct in a new quota file */
1795 static int v3_write_dqinfo(struct file *f, int type, struct if_dqinfo *info)
1797 struct v2_disk_dqinfo dqinfo;
1798 __u32 blocks = V2_DQTREEOFF + 1;
1799 loff_t offset = V2_DQINFOOFF;
1802 dqinfo.dqi_bgrace = cpu_to_le32(info->dqi_bgrace);
1803 dqinfo.dqi_igrace = cpu_to_le32(info->dqi_igrace);
1804 dqinfo.dqi_flags = cpu_to_le32(info->dqi_flags & DQF_MASK &
1807 dqinfo.dqi_bgrace = cpu_to_le32(MAX_DQ_TIME);
1808 dqinfo.dqi_igrace = cpu_to_le32(MAX_IQ_TIME);
1809 dqinfo.dqi_flags = 0;
1812 dqinfo.dqi_blocks = cpu_to_le32(blocks);
1813 dqinfo.dqi_free_blk = 0;
1814 dqinfo.dqi_free_entry = 0;
1816 return cfs_user_write(f, (char *)&dqinfo, sizeof(dqinfo), &offset);
1819 static int v3_write_dqheader(struct file *f, int type)
1821 static const __u32 quota_magics[] = V2_INITQMAGICS;
1822 static const __u32 quota_versions[] = V2_INITQVERSIONS_R1;
1823 struct v2_disk_dqheader dqhead;
1826 CLASSERT(ARRAY_SIZE(quota_magics) == ARRAY_SIZE(quota_versions));
1827 LASSERT(0 <= type && type < ARRAY_SIZE(quota_magics));
1829 dqhead.dqh_magic = cpu_to_le32(quota_magics[type]);
1830 dqhead.dqh_version = cpu_to_le32(quota_versions[type]);
1832 return cfs_user_write(f, (char *)&dqhead, sizeof(dqhead), &offset);
1835 static int create_new_quota_files(struct qchk_ctxt *qctxt,
1836 struct obd_quotactl *oqc)
1841 for (i = 0; i < MAXQUOTAS; i++) {
1842 struct if_dqinfo *info = qctxt->qckt_first_check[i]?
1843 NULL : &qctxt->qckt_dqinfo[i];
1845 const char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1847 if (!Q_TYPESET(oqc, i))
1850 LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1852 file = filp_open(name[i], O_RDWR | O_CREAT | O_TRUNC, 0644);
1855 CERROR("can't create %s file: rc = %d\n",
1860 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
1861 CERROR("file %s is not regular", name[i]);
1862 filp_close(file, 0);
1863 GOTO(out, rc = -EINVAL);
1866 DQUOT_DROP(file->f_dentry->d_inode);
1868 rc = v3_write_dqheader(file, i);
1870 filp_close(file, 0);
1874 rc = v3_write_dqinfo(file, i, info);
1875 filp_close(file, 0);
1885 static int commit_chkquot(struct super_block *sb, struct qchk_ctxt *qctxt,
1886 struct chk_dqblk *cdqb)
1888 struct obd_quotactl *oqc;
1897 now = cfs_time_current_sec();
1899 if (cdqb->dqb_bsoftlimit &&
1900 toqb(cdqb->dqb_curspace) >= cdqb->dqb_bsoftlimit &&
1903 now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_bgrace;
1905 if (cdqb->dqb_isoftlimit &&
1906 cdqb->dqb_curinodes >= cdqb->dqb_isoftlimit &&
1909 now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_igrace;
1911 cdqb->dqb_valid = QIF_ALL;
1913 oqc->qc_cmd = Q_SETQUOTA;
1914 oqc->qc_type = cdqb->dqb_type;
1915 oqc->qc_id = cdqb->dqb_id;
1916 DQBLK_COPY(&oqc->qc_dqblk, cdqb);
1918 rc = fsfilt_ext3_quotactl(sb, oqc);
1923 static int prune_chkquots(struct super_block *sb,
1924 struct qchk_ctxt *qctxt, int error)
1926 struct chk_dqblk *cdqb, *tmp;
1929 cfs_list_for_each_entry_safe(cdqb, tmp, &qctxt->qckt_list, dqb_list) {
1931 rc = commit_chkquot(sb, qctxt, cdqb);
1935 cfs_hlist_del_init(&cdqb->dqb_hash);
1936 cfs_list_del(&cdqb->dqb_list);
1943 #ifndef EXT3_FEATURE_RO_COMPAT_GDT_CSUM
1944 #define EXT3_FEATURE_RO_COMPAT_GDT_CSUM 0x0010
1947 static int fsfilt_ext3_quotacheck(struct super_block *sb,
1948 struct obd_quotactl *oqc)
1950 struct ext3_sb_info *sbi = EXT3_SB(sb);
1951 int i, group, uninit_feat = 0;
1952 struct qchk_ctxt *qctxt;
1953 struct buffer_head *bitmap_bh = NULL;
1954 unsigned long ino, inode_inuse;
1955 struct inode *inode;
1959 /* turn on quota and read dqinfo if existed */
1960 OBD_ALLOC_PTR(qctxt);
1962 oqc->qc_stat = -ENOMEM;
1966 for (i = 0; i < NR_DQHASH; i++)
1967 CFS_INIT_HLIST_HEAD(&qctxt->qckt_hash[i]);
1968 CFS_INIT_LIST_HEAD(&qctxt->qckt_list);
1970 for (i = 0; i < MAXQUOTAS; i++) {
1971 if (!Q_TYPESET(oqc, i))
1974 rc = quota_onoff(sb, Q_QUOTAON, i, oqc->qc_id);
1975 if (!rc || rc == -EBUSY) {
1976 rc = read_old_dqinfo(sb, i, qctxt->qckt_dqinfo);
1979 } else if (rc == -ENOENT || rc == -EINVAL || rc == -EEXIST) {
1980 qctxt->qckt_first_check[i] = 1;
1985 if (EXT3_HAS_RO_COMPAT_FEATURE(sb, EXT3_FEATURE_RO_COMPAT_GDT_CSUM))
1986 /* This filesystem supports the uninit group feature */
1989 /* number of inodes that have been allocated */
1990 inode_inuse = sbi->s_inodes_per_group * sbi->s_groups_count -
1991 percpu_counter_sum(&sbi->s_freeinodes_counter);
1993 /* check quota and update in hash */
1994 for (group = 0; group < sbi->s_groups_count && inode_inuse > 0;
1996 unsigned long used_count = sbi->s_inodes_per_group;
1999 struct ext3_group_desc *desc;
2000 desc = get_group_desc(sb, group, NULL);
2004 /* we don't really need to take the group lock here,
2005 * but it may be useful if one day we support online
2007 spin_lock(sb_bgl_lock(sbi, group));
2008 if (desc->bg_flags & cpu_to_le16(EXT3_BG_INODE_UNINIT)) {
2009 /* no inode in use in this group, just skip it */
2010 spin_unlock(sb_bgl_lock(sbi, group));
2013 used_count -= ext3_itable_unused_count(sb, desc);
2014 spin_unlock(sb_bgl_lock(sbi, group));
2017 ino = group * sbi->s_inodes_per_group + 1;
2018 bitmap_bh = ext3_read_inode_bitmap(sb, group);
2020 CERROR("%s: ext3_read_inode_bitmap group %d failed\n",
2026 while (i < used_count &&
2027 (i = ext3_find_next_bit(bitmap_bh->b_data,
2028 used_count, i)) < used_count) {
2031 ino = i + group * sbi->s_inodes_per_group;
2032 if (ino < sbi->s_first_ino)
2034 #if defined(HAVE_EXT4_LDISKFS) || !defined(HAVE_READ_INODE_IN_SBOPS)
2035 inode = ext3_iget(sb, ino);
2037 inode = iget(sb, ino);
2039 if (!inode || IS_ERR(inode))
2042 rc = add_inode_quota(inode, qctxt, oqc);
2052 /* read old quota limits from old quota file. (only for the user
2053 * has limits but hasn't file) */
2054 #ifdef HAVE_QUOTA_SUPPORT
2055 for (i = 0; i < MAXQUOTAS; i++) {
2057 struct dquot_id *dqid, *tmp;
2059 if (!Q_TYPESET(oqc, i))
2062 if (qctxt->qckt_first_check[i])
2066 LASSERT(sb_dqopt(sb)->files[i] != NULL);
2067 CFS_INIT_LIST_HEAD(&id_list);
2068 #ifndef KERNEL_SUPPORTS_QUOTA_READ
2069 rc = lustre_get_qids(sb_dqopt(sb)->files[i], NULL, i, &id_list);
2071 rc = lustre_get_qids(NULL, sb_dqopt(sb)->files[i], i, &id_list);
2074 CERROR("read old limits failed. (rc:%d)\n", rc);
2076 cfs_list_for_each_entry_safe(dqid, tmp, &id_list, di_link) {
2077 cfs_list_del_init(&dqid->di_link);
2080 cqget(sb, qctxt->qckt_hash, &qctxt->qckt_list,
2082 qctxt->qckt_first_check[i]);
2087 /* turn off quota cause we are to dump chk_dqblk to files */
2088 quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type, oqc->qc_id);
2090 rc = create_new_quota_files(qctxt, oqc);
2094 /* we use vfs functions to set dqblk, so turn quota on */
2095 rc = quota_onoff(sb, Q_QUOTAON, oqc->qc_type, oqc->qc_id);
2097 /* dump and free chk_dqblk */
2098 rc = prune_chkquots(sb, qctxt, rc);
2099 OBD_FREE_PTR(qctxt);
2101 /* turn off quota, `lfs quotacheck` will turn on when all
2102 * nodes quotacheck finish. */
2103 quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type, oqc->qc_id);
2107 CERROR("quotacheck failed: rc = %d\n", rc);
2112 static int fsfilt_ext3_quotainfo(struct lustre_quota_info *lqi, int type,
2118 if (lqi->qi_files[type] == NULL) {
2119 CERROR("operate qinfo before it's enabled!\n");
2125 rc = lustre_check_quota_file(lqi, type);
2128 rc = lustre_read_quota_info(lqi, type);
2131 rc = lustre_write_quota_info(lqi, type);
2133 case QFILE_INIT_INFO:
2134 rc = lustre_init_quota_info(lqi, type);
2138 CERROR("quota CONVERT command is not supported\n");
2142 CERROR("Unsupported admin quota file cmd %d\n"
2143 "Are lquota.ko and fsfilt_ldiskfs.ko modules in sync?\n",
2150 static int fsfilt_ext3_qids(struct file *file, struct inode *inode, int type,
2153 return lustre_get_qids(file, inode, type, list);
2156 static int fsfilt_ext3_dquot(struct lustre_dquot *dquot, int cmd)
2161 if (dquot->dq_info->qi_files[dquot->dq_type] == NULL) {
2162 CERROR("operate dquot before it's enabled!\n");
2167 case QFILE_RD_DQUOT:
2168 rc = lustre_read_dquot(dquot);
2170 case QFILE_WR_DQUOT:
2171 if (dquot->dq_dqb.dqb_ihardlimit ||
2172 dquot->dq_dqb.dqb_isoftlimit ||
2173 dquot->dq_dqb.dqb_bhardlimit ||
2174 dquot->dq_dqb.dqb_bsoftlimit)
2175 cfs_clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2177 cfs_set_bit(DQ_FAKE_B, &dquot->dq_flags);
2179 rc = lustre_commit_dquot(dquot);
2184 CERROR("Unsupported admin quota file cmd %d\n", cmd);
2191 static int fsfilt_ext3_get_mblk(struct super_block *sb, int *count,
2192 struct inode *inode, int frags)
2194 #ifdef EXT3_EXT_HAS_NO_TREE
2195 struct ext3_ext_base *base = inode;
2197 struct ext3_extents_tree tree;
2198 struct ext3_ext_base *base = &tree;
2200 ext3_init_tree_desc(base, inode);
2202 /* for an ost_write request, it needs <#fragments> * <tree depth + 1>
2203 * metablocks at maxium b=16542 */
2204 *count = frags * (EXT_DEPTH(base) + 1) * EXT3_BLOCK_SIZE(sb);
2210 lvfs_sbdev_type fsfilt_ext3_journal_sbdev(struct super_block *sb)
2212 return (EXT3_SB(sb)->journal_bdev);
2214 EXPORT_SYMBOL(fsfilt_ext3_journal_sbdev);
2216 ssize_t lustre_read_quota(struct file *f, struct inode *inode, int type,
2217 char *buf, int count, loff_t pos)
2222 /* Support for both adm and op quota files must be provided */
2224 rc = fsfilt_ext3_read_record(f, buf, count, &p);
2225 rc = rc < 0 ? rc : p - pos;
2227 struct super_block *sb = inode->i_sb;
2228 rc = sb->s_op->quota_read(sb, type, buf, count, pos);
2233 ssize_t lustre_write_quota(struct file *f, char *buf, int count, loff_t pos)
2238 /* Only adm quota files are supported, op updates are handled by vfs */
2239 rc = fsfilt_ext3_write_record(f, buf, count, &p, 0);
2240 rc = rc < 0 ? rc : p - pos;
2245 void *lustre_quota_journal_start(struct inode *inode, int delete)
2248 unsigned block_count;
2251 /* each indirect block (+4) may become free, attaching to the
2252 * header list of free blocks (+1); the data block (+1) may
2253 * become a free block (+0) or a block with free dqentries (+0) */
2254 block_count = (4 + 1) + 1;
2255 handle = ext3_journal_start(inode,
2256 block_count*FSFILT_DATA_TRANS_BLOCKS(inode->i_sb)+2);
2258 /* indirect blocks are touched (+4), each causes file expansion (+0) or
2259 * freeblk reusage with a header update (+1); dqentry is either reused
2260 * causing update of the entry block (+1), prev (+1) and next (+1) or
2261 * a new block allocation (+1) with a header update (+1) */
2262 block_count = (4 + 1) + 3;
2263 handle = ext3_journal_start(inode,
2264 block_count*FSFILT_DATA_TRANS_BLOCKS(inode->i_sb)+2);
2271 void lustre_quota_journal_stop(void *handle)
2273 ext3_journal_stop((handle_t *)handle);
2276 static struct fsfilt_operations fsfilt_ext3_ops = {
2278 .fs_owner = THIS_MODULE,
2279 .fs_getlabel = fsfilt_ext3_get_label,
2280 .fs_setlabel = fsfilt_ext3_set_label,
2281 .fs_uuid = fsfilt_ext3_uuid,
2282 .fs_start = fsfilt_ext3_start,
2283 .fs_brw_start = fsfilt_ext3_brw_start,
2284 .fs_extend = fsfilt_ext3_extend,
2285 .fs_commit = fsfilt_ext3_commit,
2286 .fs_commit_async = fsfilt_ext3_commit_async,
2287 .fs_commit_wait = fsfilt_ext3_commit_wait,
2288 .fs_setattr = fsfilt_ext3_setattr,
2289 .fs_iocontrol = fsfilt_ext3_iocontrol,
2290 .fs_set_md = fsfilt_ext3_set_md,
2291 .fs_get_md = fsfilt_ext3_get_md,
2292 .fs_readpage = fsfilt_ext3_readpage,
2293 .fs_add_journal_cb = fsfilt_ext3_add_journal_cb,
2294 .fs_statfs = fsfilt_ext3_statfs,
2295 .fs_sync = fsfilt_ext3_sync,
2296 .fs_map_inode_pages = fsfilt_ext3_map_inode_pages,
2297 .fs_write_record = fsfilt_ext3_write_record,
2298 .fs_read_record = fsfilt_ext3_read_record,
2299 .fs_setup = fsfilt_ext3_setup,
2300 .fs_send_bio = fsfilt_ext3_send_bio,
2301 .fs_get_op_len = fsfilt_ext3_get_op_len,
2302 #ifdef HAVE_DISK_INODE_VERSION
2303 .fs_get_version = fsfilt_ext3_get_version,
2304 .fs_set_version = fsfilt_ext3_set_version,
2306 #ifdef HAVE_QUOTA_SUPPORT
2307 .fs_quotactl = fsfilt_ext3_quotactl,
2308 .fs_quotacheck = fsfilt_ext3_quotacheck,
2309 .fs_quotainfo = fsfilt_ext3_quotainfo,
2310 .fs_qids = fsfilt_ext3_qids,
2311 .fs_dquot = fsfilt_ext3_dquot,
2312 .fs_get_mblk = fsfilt_ext3_get_mblk,
2314 .fs_journal_sbdev = fsfilt_ext3_journal_sbdev,
2317 static int __init fsfilt_ext3_init(void)
2321 fcb_cache = cfs_mem_cache_create("fsfilt_ext3_fcb",
2322 sizeof(struct fsfilt_cb_data), 0, 0);
2324 CERROR("error allocating fsfilt journal callback cache\n");
2325 GOTO(out, rc = -ENOMEM);
2328 rc = fsfilt_register_ops(&fsfilt_ext3_ops);
2331 int err = cfs_mem_cache_destroy(fcb_cache);
2332 LASSERTF(err == 0, "error destroying new cache: rc %d\n", err);
2338 static void __exit fsfilt_ext3_exit(void)
2342 fsfilt_unregister_ops(&fsfilt_ext3_ops);
2343 rc = cfs_mem_cache_destroy(fcb_cache);
2344 LASSERTF(rc == 0, "couldn't destroy fcb_cache slab\n");
2347 module_init(fsfilt_ext3_init);
2348 module_exit(fsfilt_ext3_exit);
2350 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2351 MODULE_DESCRIPTION("Lustre ext3 Filesystem Helper v0.1");
2352 MODULE_LICENSE("GPL");