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 (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
30 * Use is subject to license terms.
32 * Copyright (c) 2011 Whamcloud, Inc.
36 * This file is part of Lustre, http://www.lustre.org/
37 * Lustre is a trademark of Sun Microsystems, Inc.
39 * lustre/lvfs/fsfilt_ext3.c
41 * Author: Andreas Dilger <adilger@clusterfs.com>
44 #define DEBUG_SUBSYSTEM S_FILTER
46 #include <linux/init.h>
47 #include <linux/module.h>
49 #include <linux/slab.h>
50 #include <linux/pagemap.h>
51 #include <linux/quotaops.h>
52 #ifdef HAVE_LINUX_EXPORTFS_H
53 #include <linux/exportfs.h>
55 #ifdef HAVE_EXT4_LDISKFS
56 #include <ext4/ext4.h>
57 #include <ext4/ext4_jbd2.h>
59 #include <linux/jbd.h>
60 #include <linux/ext3_fs.h>
61 #include <linux/ext3_jbd.h>
63 #include <linux/version.h>
64 #include <linux/bitops.h>
65 #include <linux/quota.h>
67 # include <linux/quotaio_v2.h>
68 #elif defined(HAVE_FS_QUOTA_QUOTAIO_H)
69 # include <quota/quotaio_v2.h>
70 # include <quota/quota_tree.h>
71 # define V2_DQTREEOFF QT_TREEOFF
72 #elif defined(HAVE_FS_QUOTAIO_V1_H)
73 # include <quotaio_v2.h>
74 # include <quota_tree.h>
75 # define V2_DQTREEOFF QT_TREEOFF
76 # define V2_INITQVERSIONS_R1 V2_INITQVERSIONS
80 #define QFMT_LUSTRE QFMT_VFS_V1
82 #define QFMT_LUSTRE QFMT_VFS_V0
85 #if defined(HAVE_EXT3_XATTR_H)
86 #include <ext3/xattr.h>
88 /* ext3 xattr.h not available in rh style kernel-devel rpm */
89 extern int ext3_xattr_get(struct inode *, int, const char *, void *, size_t);
90 extern int ext3_xattr_set_handle(handle_t *, struct inode *, int, const char *, const void *, size_t, int);
93 #include <libcfs/libcfs.h>
94 #include <lustre_fsfilt.h>
96 #include <lustre_quota.h>
97 #include <linux/lustre_compat25.h>
98 #include <linux/lprocfs_status.h>
100 #ifdef HAVE_EXT4_LDISKFS
101 #include <ext4/ext4_extents.h>
103 #include <linux/ext3_extents.h>
106 #include "lustre_quota_fmt.h"
108 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,15)
109 #define FSFILT_DATA_TRANS_BLOCKS(sb) EXT3_DATA_TRANS_BLOCKS
110 #define FSFILT_DELETE_TRANS_BLOCKS(sb) EXT3_DELETE_TRANS_BLOCKS
112 #define FSFILT_DATA_TRANS_BLOCKS(sb) EXT3_DATA_TRANS_BLOCKS(sb)
113 #define FSFILT_DELETE_TRANS_BLOCKS(sb) EXT3_DELETE_TRANS_BLOCKS(sb)
116 #ifdef EXT3_SINGLEDATA_TRANS_BLOCKS_HAS_SB
117 /* for kernels 2.6.18 and later */
118 #define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS(sb)
120 #define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS
123 #ifdef EXT_INSERT_EXTENT_WITH_5ARGS
124 #define fsfilt_ext3_ext_insert_extent(handle, inode, path, newext, flag) \
125 ext3_ext_insert_extent(handle, inode, path, newext, flag)
127 #define fsfilt_ext3_ext_insert_extent(handle, inode, path, newext, flag) \
128 ext3_ext_insert_extent(handle, inode, path, newext)
131 #ifdef EXT3_DISCARD_PREALLOCATIONS
132 #define ext3_mb_discard_inode_preallocations(inode) \
133 ext3_discard_preallocations(inode)
137 static cfs_mem_cache_t *fcb_cache;
139 struct fsfilt_cb_data {
140 struct journal_callback cb_jcb; /* jbd private data - MUST BE FIRST */
141 fsfilt_cb_t cb_func; /* MDS/OBD completion function */
142 struct obd_device *cb_obd; /* MDS/OBD completion device */
143 __u64 cb_last_rcvd; /* MDS/OST last committed operation */
144 void *cb_data; /* MDS/OST completion function data */
147 #ifndef EXT3_XATTR_INDEX_TRUSTED /* temporary until we hit l28 kernel */
148 #define EXT3_XATTR_INDEX_TRUSTED 4
151 #ifdef HAVE_EXT4_LDISKFS
152 #define fsfilt_log_start_commit(journal, tid) jbd2_log_start_commit(journal, tid)
153 #define fsfilt_log_wait_commit(journal, tid) jbd2_log_wait_commit(journal, tid)
154 #define fsfilt_journal_callback_set(handle, func, jcb) jbd2_journal_callback_set(handle, func, jcb)
156 #define fsfilt_log_start_commit(journal, tid) log_start_commit(journal, tid)
157 #define fsfilt_log_wait_commit(journal, tid) log_wait_commit(journal, tid)
158 #define fsfilt_journal_callback_set(handle, func, jcb) journal_callback_set(handle, func, jcb)
159 #define ext_pblock(ex) le32_to_cpu((ex)->ee_start)
160 #define ext3_ext_store_pblock(ex, pblock) ((ex)->ee_start = cpu_to_le32(pblock))
161 #define ext3_inode_bitmap(sb,desc) le32_to_cpu((desc)->bg_inode_bitmap)
164 #ifndef ext3_find_next_bit
165 #define ext3_find_next_bit ext2_find_next_bit
168 #ifndef ext2_find_next_bit
169 #ifdef __LITTLE_ENDIAN
170 #define ext2_find_next_bit(addr, size, off) find_next_bit((unsigned long *)(addr), (size), (off))
172 error "Need implementation of find_next_bit on big-endian systems"
173 #endif /* __LITTLE_ENDIAN */
174 #endif /* !ext2_find_next_le_bit */
176 static char *fsfilt_ext3_get_label(struct super_block *sb)
178 return EXT3_SB(sb)->s_es->s_volume_name;
181 static int fsfilt_ext3_set_label(struct super_block *sb, char *label)
183 /* see e.g. fsfilt_ext3_write_record() */
188 journal = EXT3_SB(sb)->s_journal;
189 handle = ext3_journal_start_sb(sb, 1);
190 if (IS_ERR(handle)) {
191 CERROR("can't start transaction\n");
192 return(PTR_ERR(handle));
195 err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
199 memcpy(EXT3_SB(sb)->s_es->s_volume_name, label,
200 sizeof(EXT3_SB(sb)->s_es->s_volume_name));
202 err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
205 ext3_journal_stop(handle);
210 static char *fsfilt_ext3_uuid(struct super_block *sb)
212 return EXT3_SB(sb)->s_es->s_uuid;
215 #ifdef HAVE_DISK_INODE_VERSION
217 static __u64 get_i_version(struct inode *inode)
219 return EXT3_I(inode)->i_fs_version;
222 static void set_i_version(struct inode *inode, __u64 new_version)
224 (EXT3_I(inode))->i_fs_version = new_version;
228 * Get the 64-bit version for an inode.
230 static __u64 fsfilt_ext3_get_version(struct inode *inode)
232 CDEBUG(D_INFO, "Get version "LPX64" for inode %lu\n",
233 get_i_version(inode), inode->i_ino);
234 return get_i_version(inode);
238 * Set the 64-bit version and return the old version.
240 static __u64 fsfilt_ext3_set_version(struct inode *inode, __u64 new_version)
242 __u64 old_version = get_i_version(inode);
244 CDEBUG(D_INFO, "Set version "LPX64" (old "LPX64") for inode %lu\n",
245 new_version, old_version, inode->i_ino);
246 set_i_version(inode, new_version);
247 /* version is set after all inode operations are finished, so we should
248 * mark it dirty here */
249 inode->i_sb->s_op->dirty_inode(inode);
255 /* kernel has ext4_blocks_for_truncate since linux-3.1.1 */
256 #ifdef HAVE_BLOCKS_FOR_TRUNCATE
257 # include <ext4/truncate.h>
259 static inline unsigned long ext4_blocks_for_truncate(struct inode *inode)
263 needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
266 if (needed > EXT4_MAX_TRANS_DATA)
267 needed = EXT4_MAX_TRANS_DATA;
268 return EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
273 * We don't currently need any additional blocks for rmdir and
274 * unlink transactions because we are storing the OST oa_id inside
275 * the inode (which we will be changing anyways as part of this
278 static void *fsfilt_ext3_start(struct inode *inode, int op, void *desc_private,
281 /* For updates to the last received file */
282 int nblocks = FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
286 if (current->journal_info) {
287 CDEBUG(D_INODE, "increasing refcount on %p\n",
288 current->journal_info);
293 case FSFILT_OP_RMDIR:
294 case FSFILT_OP_UNLINK:
295 /* delete one file + create/update logs for each stripe */
296 nblocks += FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb);
297 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
298 FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
300 case FSFILT_OP_RENAME:
301 /* modify additional directory */
302 nblocks += FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
304 case FSFILT_OP_SYMLINK:
305 /* additional block + block bitmap + GDT for long symlink */
308 case FSFILT_OP_CREATE: {
309 #if defined(EXT3_EXTENTS_FL) && defined(EXT3_INDEX_FL) && !defined(HAVE_EXT4_LDISKFS)
312 if (!test_opt(inode->i_sb, EXTENTS)) {
314 } else if (((EXT3_I(inode)->i_flags &
315 cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL)) ==
316 cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL))) {
317 CWARN("extent-mapped directory found with "
318 "ext3-based ldiskfs - contact "
319 "http://bugs.whamcloud.com/\n");
326 case FSFILT_OP_MKDIR:
327 case FSFILT_OP_MKNOD:
328 /* modify one inode + block bitmap + GDT */
332 /* modify parent directory */
333 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
334 FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
335 /* create/update logs for each stripe */
336 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
337 FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
339 case FSFILT_OP_SETATTR:
340 /* Setattr on inode */
342 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
343 FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
344 /* quota chown log for each stripe */
345 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
346 FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
348 case FSFILT_OP_CANCEL_UNLINK:
351 /* blocks for log header bitmap update OR
352 * blocks for catalog header bitmap update + unlink of logs +
353 * blocks for delete the inode (include blocks truncating). */
354 nblocks = (LLOG_CHUNK_SIZE >> inode->i_blkbits) +
355 EXT3_DELETE_TRANS_BLOCKS(inode->i_sb) +
356 ext4_blocks_for_truncate(inode) + 3;
358 default: CERROR("unknown transaction start op %d\n", op);
362 LASSERT(current->journal_info == desc_private);
363 journal = EXT3_SB(inode->i_sb)->s_journal;
364 if (nblocks > journal->j_max_transaction_buffers) {
365 CWARN("too many credits %d for op %ux%u using %d instead\n",
366 nblocks, op, logs, journal->j_max_transaction_buffers);
367 nblocks = journal->j_max_transaction_buffers;
371 LASSERTF(nblocks > 0, "can't start %d credit transaction\n", nblocks);
372 handle = ext3_journal_start(inode, nblocks);
375 LASSERT(current->journal_info == handle);
377 CERROR("error starting handle for op %u (%u credits): rc %ld\n",
378 op, nblocks, PTR_ERR(handle));
383 * Calculate the number of buffer credits needed to write multiple pages in
384 * a single ext3 transaction. No, this shouldn't be here, but as yet ext3
385 * doesn't have a nice API for calculating this sort of thing in advance.
387 * See comment above ext3_writepage_trans_blocks for details. We assume
388 * no data journaling is being done, but it does allow for all of the pages
389 * being non-contiguous. If we are guaranteed contiguous pages we could
390 * reduce the number of (d)indirect blocks a lot.
392 * With N blocks per page and P pages, for each inode we have at most:
394 * min(N*P, blocksize/4 + 1) dindirect blocks
397 * For the entire filesystem, we have at most:
398 * min(sum(nindir + P), ngroups) bitmap blocks (from the above)
399 * min(sum(nindir + P), gdblocks) group descriptor blocks (from the above)
400 * objcount inode blocks
402 * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quota files
404 * 1 EXT3_DATA_TRANS_BLOCKS for the last_rcvd update.
406 static int fsfilt_ext3_credits_needed(int objcount, struct fsfilt_objinfo *fso,
407 int niocount, struct niobuf_local *nb)
409 struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
411 const int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
412 int nbitmaps = 0, ngdblocks;
413 int needed = objcount + 1; /* inodes + superblock */
416 for (i = 0, j = 0; i < objcount; i++, fso++) {
417 /* two or more dindirect blocks in case we cross boundary */
418 int ndind = (long)((nb[j + fso->fso_bufcnt - 1].offset -
420 sb->s_blocksize_bits) /
421 (EXT3_ADDR_PER_BLOCK(sb) * EXT3_ADDR_PER_BLOCK(sb));
422 nbitmaps += min(fso->fso_bufcnt, ndind > 0 ? ndind : 2);
424 /* leaf, indirect, tindirect blocks for first block */
425 nbitmaps += blockpp + 2;
427 j += fso->fso_bufcnt;
430 next_indir = nb[0].offset +
431 (EXT3_ADDR_PER_BLOCK(sb) << sb->s_blocksize_bits);
432 for (i = 1; i < niocount; i++) {
433 if (nb[i].offset >= next_indir) {
434 nbitmaps++; /* additional indirect */
435 next_indir = nb[i].offset +
436 (EXT3_ADDR_PER_BLOCK(sb)<<sb->s_blocksize_bits);
437 } else if (nb[i].offset != nb[i - 1].offset + sb->s_blocksize) {
438 nbitmaps++; /* additional indirect */
440 nbitmaps += blockpp; /* each leaf in different group? */
443 ngdblocks = nbitmaps;
444 if (nbitmaps > EXT3_SB(sb)->s_groups_count)
445 nbitmaps = EXT3_SB(sb)->s_groups_count;
446 if (ngdblocks > EXT3_SB(sb)->s_gdb_count)
447 ngdblocks = EXT3_SB(sb)->s_gdb_count;
449 needed += nbitmaps + ngdblocks;
451 /* last_rcvd update */
452 needed += FSFILT_DATA_TRANS_BLOCKS(sb);
454 #if defined(CONFIG_QUOTA)
455 /* We assume that there will be 1 bit set in s_dquot.flags for each
456 * quota file that is active. This is at least true for now.
458 needed += hweight32(ll_sb_any_quota_active(sb)) *
459 FSFILT_SINGLEDATA_TRANS_BLOCKS(sb);
465 /* We have to start a huge journal transaction here to hold all of the
466 * metadata for the pages being written here. This is necessitated by
467 * the fact that we do lots of prepare_write operations before we do
468 * any of the matching commit_write operations, so even if we split
469 * up to use "smaller" transactions none of them could complete until
470 * all of them were opened. By having a single journal transaction,
471 * we eliminate duplicate reservations for common blocks like the
472 * superblock and group descriptors or bitmaps.
474 * We will start the transaction here, but each prepare_write will
475 * add a refcount to the transaction, and each commit_write will
476 * remove a refcount. The transaction will be closed when all of
477 * the pages have been written.
479 static void *fsfilt_ext3_brw_start(int objcount, struct fsfilt_objinfo *fso,
480 int niocount, struct niobuf_local *nb,
481 void *desc_private, int logs)
488 LASSERT(current->journal_info == desc_private);
489 journal = EXT3_SB(fso->fso_dentry->d_inode->i_sb)->s_journal;
490 needed = fsfilt_ext3_credits_needed(objcount, fso, niocount, nb);
492 /* The number of blocks we could _possibly_ dirty can very large.
493 * We reduce our request if it is absurd (and we couldn't get that
494 * many credits for a single handle anyways).
496 * At some point we have to limit the size of I/Os sent at one time,
497 * increase the size of the journal, or we have to calculate the
498 * actual journal requirements more carefully by checking all of
499 * the blocks instead of being maximally pessimistic. It remains to
500 * be seen if this is a real problem or not.
502 if (needed > journal->j_max_transaction_buffers) {
503 CERROR("want too many journal credits (%d) using %d instead\n",
504 needed, journal->j_max_transaction_buffers);
505 needed = journal->j_max_transaction_buffers;
508 LASSERTF(needed > 0, "can't start %d credit transaction\n", needed);
509 handle = ext3_journal_start(fso->fso_dentry->d_inode, needed);
510 if (IS_ERR(handle)) {
511 CERROR("can't get handle for %d credits: rc = %ld\n", needed,
514 LASSERT(handle->h_buffer_credits >= needed);
515 LASSERT(current->journal_info == handle);
521 static int fsfilt_ext3_extend(struct inode *inode, unsigned int nblocks,void *h)
523 handle_t *handle = h;
525 /* fsfilt_extend called with nblocks = 0 for testing in special cases */
527 handle->h_buffer_credits = 0;
528 CWARN("setting credits of handle %p to zero by request\n", h);
531 if (handle->h_buffer_credits > nblocks)
533 if (ext3_journal_extend(handle, nblocks) == 0)
536 ext3_mark_inode_dirty(handle, inode);
537 return ext3_journal_restart(handle, nblocks);
540 static int fsfilt_ext3_commit(struct inode *inode, void *h, int force_sync)
543 handle_t *handle = h;
545 LASSERT(current->journal_info == handle);
547 handle->h_sync = 1; /* recovery likes this */
549 rc = ext3_journal_stop(handle);
554 static int fsfilt_ext3_commit_async(struct inode *inode, void *h,
558 transaction_t *transaction;
559 handle_t *handle = h;
563 LASSERT(current->journal_info == handle);
565 transaction = handle->h_transaction;
566 journal = transaction->t_journal;
567 tid = transaction->t_tid;
568 /* we don't want to be blocked */
570 rc = ext3_journal_stop(handle);
572 CERROR("error while stopping transaction: %d\n", rc);
575 fsfilt_log_start_commit(journal, tid);
577 *wait_handle = (void *) tid;
578 CDEBUG(D_INODE, "commit async: %lu\n", (unsigned long) tid);
582 static int fsfilt_ext3_commit_wait(struct inode *inode, void *h)
584 journal_t *journal = EXT3_JOURNAL(inode);
585 tid_t tid = (tid_t)(long)h;
587 CDEBUG(D_INODE, "commit wait: %lu\n", (unsigned long) tid);
588 if (unlikely(is_journal_aborted(journal)))
591 fsfilt_log_wait_commit(EXT3_JOURNAL(inode), tid);
593 if (unlikely(is_journal_aborted(journal)))
598 static int fsfilt_ext3_setattr(struct dentry *dentry, void *handle,
599 struct iattr *iattr, int do_trunc)
601 struct inode *inode = dentry->d_inode;
604 /* Avoid marking the inode dirty on the superblock list unnecessarily.
605 * We are already writing the inode to disk as part of this
606 * transaction and want to avoid a lot of extra inode writeout
607 * later on. b=9828 */
608 if (iattr->ia_valid & ATTR_SIZE && !do_trunc) {
609 /* ATTR_SIZE would invoke truncate: clear it */
610 iattr->ia_valid &= ~ATTR_SIZE;
611 EXT3_I(inode)->i_disksize = iattr->ia_size;
612 i_size_write(inode, iattr->ia_size);
614 if (iattr->ia_valid & ATTR_UID)
615 inode->i_uid = iattr->ia_uid;
616 if (iattr->ia_valid & ATTR_GID)
617 inode->i_gid = iattr->ia_gid;
618 if (iattr->ia_valid & ATTR_ATIME)
619 inode->i_atime = iattr->ia_atime;
620 if (iattr->ia_valid & ATTR_MTIME)
621 inode->i_mtime = iattr->ia_mtime;
622 if (iattr->ia_valid & ATTR_CTIME)
623 inode->i_ctime = iattr->ia_ctime;
624 if (iattr->ia_valid & ATTR_MODE) {
625 inode->i_mode = iattr->ia_mode;
627 if (!cfs_curproc_is_in_groups(inode->i_gid) &&
628 !cfs_capable(CFS_CAP_FSETID))
629 inode->i_mode &= ~S_ISGID;
632 inode->i_sb->s_op->dirty_inode(inode);
637 /* Don't allow setattr to change file type */
638 if (iattr->ia_valid & ATTR_MODE)
639 iattr->ia_mode = (inode->i_mode & S_IFMT) |
640 (iattr->ia_mode & ~S_IFMT);
642 /* We set these flags on the client, but have already checked perms
643 * so don't confuse inode_change_ok. */
644 iattr->ia_valid &= ~(ATTR_MTIME_SET | ATTR_ATIME_SET);
646 if (inode->i_op->setattr) {
647 rc = inode->i_op->setattr(dentry, iattr);
649 rc = inode_change_ok(inode, iattr);
651 rc = inode_setattr(inode, iattr);
658 static int fsfilt_ext3_iocontrol(struct inode *inode, struct file *file,
659 unsigned int cmd, unsigned long arg)
664 /* FIXME: Can't do this because of nested transaction deadlock */
665 if (cmd == EXT3_IOC_SETFLAGS) {
666 /* We can't enable data journaling on OST objects, because
667 * this forces the transaction to be closed in order to
668 * flush the journal, but the caller will already have a
669 * compound transaction open to update the last_rcvd file,
670 * and this thread would deadlock trying to set the flag. */
671 if ((*(int *)arg) & EXT3_JOURNAL_DATA_FL) {
672 CERROR("can't set data journal flag on file\n");
675 /* Because the MDS does not see the EXTENTS_FL set on the
676 * OST objects, mask this flag into all set flags. It is
677 * not legal to clear this flag in any case, so we are not
678 * changing the functionality by doing this. b=22911 */
679 *(int *)arg |= EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL;
682 #ifdef HAVE_EXT4_LDISKFS
683 /* ext4_ioctl does not have a inode argument */
684 if (inode->i_fop->unlocked_ioctl)
685 rc = inode->i_fop->unlocked_ioctl(file, cmd, arg);
687 if (inode->i_fop->ioctl)
688 rc = inode->i_fop->ioctl(inode, file, cmd, arg);
696 static int fsfilt_ext3_set_md(struct inode *inode, void *handle,
697 void *lmm, int lmm_size, const char *name)
701 LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
703 rc = ext3_xattr_set_handle(handle, inode, EXT3_XATTR_INDEX_TRUSTED,
704 name, lmm, lmm_size, XATTR_NO_CTIME);
707 if (rc && rc != -EROFS)
708 CERROR("error adding MD data to inode %lu: rc = %d\n",
713 /* Must be called with i_mutex held */
714 static int fsfilt_ext3_get_md(struct inode *inode, void *lmm, int lmm_size,
719 LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
721 rc = ext3_xattr_get(inode, EXT3_XATTR_INDEX_TRUSTED,
722 name, lmm, lmm_size);
724 /* This gives us the MD size */
726 return (rc == -ENODATA) ? 0 : rc;
729 CDEBUG(D_INFO, "error getting EA %d/%s from inode %lu: rc %d\n",
730 EXT3_XATTR_INDEX_TRUSTED, name,
732 memset(lmm, 0, lmm_size);
733 return (rc == -ENODATA) ? 0 : rc;
739 static int fsfilt_ext3_send_bio(int rw, struct inode *inode, struct bio *bio)
745 static ssize_t fsfilt_ext3_readpage(struct file *file, char *buf, size_t count,
748 struct inode *inode = file->f_dentry->d_inode;
751 if (S_ISREG(inode->i_mode))
752 rc = file->f_op->read(file, buf, count, off);
754 const int blkbits = inode->i_sb->s_blocksize_bits;
755 const int blksize = inode->i_sb->s_blocksize;
757 CDEBUG(D_EXT2, "reading %lu at dir %lu+%llu\n",
758 (unsigned long)count, inode->i_ino, *off);
760 struct buffer_head *bh;
763 if (*off < i_size_read(inode)) {
766 bh = ext3_bread(NULL, inode, *off >> blkbits,
769 CDEBUG(D_EXT2, "read %u@%llu\n", blksize, *off);
772 memcpy(buf, bh->b_data, blksize);
775 /* XXX in theory we should just fake
776 * this buffer and continue like ext3,
777 * especially if this is a partial read
779 CERROR("error read dir %lu+%llu: %d\n",
780 inode->i_ino, *off, err);
785 struct ext3_dir_entry_2 *fake = (void *)buf;
787 CDEBUG(D_EXT2, "fake %u@%llu\n", blksize, *off);
788 memset(fake, 0, sizeof(*fake));
789 fake->rec_len = cpu_to_le16(blksize);
801 static void fsfilt_ext3_cb_func(struct journal_callback *jcb, int error)
803 struct fsfilt_cb_data *fcb = (struct fsfilt_cb_data *)jcb;
805 fcb->cb_func(fcb->cb_obd, fcb->cb_last_rcvd, fcb->cb_data, error);
807 OBD_SLAB_FREE(fcb, fcb_cache, sizeof *fcb);
810 static int fsfilt_ext3_add_journal_cb(struct obd_device *obd, __u64 last_rcvd,
811 void *handle, fsfilt_cb_t cb_func,
814 struct fsfilt_cb_data *fcb;
816 OBD_SLAB_ALLOC_PTR_GFP(fcb, fcb_cache, CFS_ALLOC_IO);
820 fcb->cb_func = cb_func;
822 fcb->cb_last_rcvd = last_rcvd;
823 fcb->cb_data = cb_data;
825 CDEBUG(D_EXT2, "set callback for last_rcvd: "LPD64"\n", last_rcvd);
826 fsfilt_journal_callback_set(handle, fsfilt_ext3_cb_func,
827 (struct journal_callback *)fcb);
832 static int fsfilt_ext3_statfs(struct super_block *sb, struct obd_statfs *osfs)
837 memset(&sfs, 0, sizeof(sfs));
838 rc = ll_do_statfs(sb, &sfs);
839 statfs_pack(osfs, &sfs);
843 static int fsfilt_ext3_sync(struct super_block *sb)
845 return ext3_force_commit(sb);
848 #ifndef EXT3_EXTENTS_FL
849 #define EXT3_EXTENTS_FL 0x00080000 /* Inode uses extents */
852 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
853 # define fsfilt_up_truncate_sem(inode) up(&LDISKFS_I(inode)->truncate_sem);
854 # define fsfilt_down_truncate_sem(inode) down(&LDISKFS_I(inode)->truncate_sem);
856 # ifdef HAVE_EXT4_LDISKFS
857 # ifdef WALK_SPACE_HAS_DATA_SEM /* We only use it in fsfilt_map_nblocks() for now */
858 # define fsfilt_up_truncate_sem(inode) do{ }while(0)
859 # define fsfilt_down_truncate_sem(inode) do{ }while(0)
861 # define fsfilt_up_truncate_sem(inode) up_write((&EXT4_I(inode)->i_data_sem))
862 # define fsfilt_down_truncate_sem(inode) down_write((&EXT4_I(inode)->i_data_sem))
865 # define fsfilt_up_truncate_sem(inode) mutex_unlock(&EXT3_I(inode)->truncate_mutex)
866 # define fsfilt_down_truncate_sem(inode) mutex_lock(&EXT3_I(inode)->truncate_mutex)
871 #define EXT_ASSERT(cond) BUG_ON(!(cond))
874 #ifdef EXT3_EXT_HAS_NO_TREE
875 /* for kernels 2.6.18 and later */
876 #ifdef HAVE_EXT4_LDISKFS
877 #define EXT_GENERATION(inode) (EXT4_I(inode)->i_ext_generation)
879 #define EXT_GENERATION(inode) ext_generation(inode)
881 #define ext3_ext_base inode
882 #define ext3_ext_base2inode(inode) (inode)
883 #define EXT_DEPTH(inode) ext_depth(inode)
884 #define fsfilt_ext3_ext_walk_space(inode, block, num, cb, cbdata) \
885 ext3_ext_walk_space(inode, block, num, cb, cbdata);
887 #define ext3_ext_base ext3_extents_tree
888 #define ext3_ext_base2inode(tree) (tree->inode)
889 #define fsfilt_ext3_ext_walk_space(tree, block, num, cb, cbdata) \
890 ext3_ext_walk_space(tree, block, num, cb);
893 #ifdef EXT_INSERT_EXTENT_WITH_5ARGS
894 #define fsfilt_ext3_ext_insert_extent(handle, inode, path, newext, flag) \
895 ext3_ext_insert_extent(handle, inode, path, newext, flag)
897 #define fsfilt_ext3_ext_insert_extent(handle, inode, path, newext, flag) \
898 ext3_ext_insert_extent(handle, inode, path, newext)
901 #include <linux/lustre_version.h>
904 unsigned long *blocks;
912 static long ext3_ext_find_goal(struct inode *inode, struct ext3_ext_path *path,
913 unsigned long block, int *aflags)
915 struct ext3_inode_info *ei = EXT3_I(inode);
916 unsigned long bg_start;
917 unsigned long colour;
921 struct ext3_extent *ex;
922 depth = path->p_depth;
924 /* try to predict block placement */
925 if ((ex = path[depth].p_ext))
926 return ext_pblock(ex) + (block - le32_to_cpu(ex->ee_block));
928 /* it looks index is empty
929 * try to find starting from index itself */
930 if (path[depth].p_bh)
931 return path[depth].p_bh->b_blocknr;
934 /* OK. use inode's group */
935 bg_start = (ei->i_block_group * EXT3_BLOCKS_PER_GROUP(inode->i_sb)) +
936 le32_to_cpu(EXT3_SB(inode->i_sb)->s_es->s_first_data_block);
937 colour = (current->pid % 16) *
938 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
939 return bg_start + colour + block;
942 #define ll_unmap_underlying_metadata(sb, blocknr) \
943 unmap_underlying_metadata((sb)->s_bdev, blocknr)
945 #ifndef EXT3_MB_HINT_GROUP_ALLOC
946 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
947 struct ext3_ext_path *path, unsigned long block,
948 unsigned long *count, int *err)
950 unsigned long pblock, goal;
952 struct inode *inode = ext3_ext_base2inode(base);
954 goal = ext3_ext_find_goal(inode, path, block, &aflags);
955 aflags |= 2; /* block have been already reserved */
956 pblock = ext3_mb_new_blocks(handle, inode, goal, count, aflags, err);
961 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
962 struct ext3_ext_path *path, unsigned long block,
963 unsigned long *count, int *err)
965 struct inode *inode = ext3_ext_base2inode(base);
966 struct ext3_allocation_request ar;
967 unsigned long pblock;
970 /* find neighbour allocated blocks */
972 *err = ext3_ext_search_left(base, path, &ar.lleft, &ar.pleft);
976 *err = ext3_ext_search_right(base, path, &ar.lright, &ar.pright);
980 /* allocate new block */
981 ar.goal = ext3_ext_find_goal(inode, path, block, &aflags);
985 ar.flags = EXT3_MB_HINT_DATA;
986 pblock = ext3_mb_new_blocks(handle, &ar, err);
992 #ifdef EXT3_EXT_HAS_NO_TREE
993 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
994 struct ext3_ext_path *path,
995 struct ext3_ext_cache *cex,
996 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
997 struct ext3_extent *ex,
1001 struct bpointers *bp = cbdata;
1003 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
1004 struct ext3_ext_path *path,
1005 struct ext3_ext_cache *cex
1006 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
1007 , struct ext3_extent *ex
1011 struct bpointers *bp = base->private;
1013 struct inode *inode = ext3_ext_base2inode(base);
1014 struct ext3_extent nex;
1015 unsigned long pblock;
1018 unsigned long count;
1021 if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
1026 if (bp->create == 0) {
1028 if (cex->ec_block < bp->start)
1029 i = bp->start - cex->ec_block;
1030 if (i >= cex->ec_len)
1031 CERROR("nothing to do?! i = %d, e_num = %u\n",
1033 for (; i < cex->ec_len && bp->num; i++) {
1042 return EXT_CONTINUE;
1045 tgen = EXT_GENERATION(base);
1046 count = ext3_ext_calc_credits_for_insert(base, path);
1047 fsfilt_up_truncate_sem(inode);
1049 handle = ext3_journal_start(inode, count+EXT3_ALLOC_NEEDED+1);
1050 if (IS_ERR(handle)) {
1051 fsfilt_down_truncate_sem(inode);
1052 return PTR_ERR(handle);
1055 fsfilt_down_truncate_sem(inode);
1056 if (tgen != EXT_GENERATION(base)) {
1057 /* the tree has changed. so path can be invalid at moment */
1058 ext3_journal_stop(handle);
1062 #if defined(HAVE_EXT4_LDISKFS) && defined(WALK_SPACE_HAS_DATA_SEM)
1063 /* In 2.6.32 kernel, ext4_ext_walk_space()'s callback func is not
1064 * protected by i_data_sem as whole. so we patch it to store
1065 * generation to path and now verify the tree hasn't changed */
1066 down_write((&EXT4_I(inode)->i_data_sem));
1068 /* validate extent, make sure the extent tree does not changed */
1069 if (EXT_GENERATION(base) != path[0].p_generation) {
1070 /* cex is invalid, try again */
1071 up_write(&EXT4_I(inode)->i_data_sem);
1072 ext3_journal_stop(handle);
1077 count = cex->ec_len;
1078 pblock = new_blocks(handle, base, path, cex->ec_block, &count, &err);
1081 EXT_ASSERT(count <= cex->ec_len);
1083 /* insert new extent */
1084 nex.ee_block = cpu_to_le32(cex->ec_block);
1085 ext3_ext_store_pblock(&nex, pblock);
1086 nex.ee_len = cpu_to_le16(count);
1087 err = fsfilt_ext3_ext_insert_extent(handle, base, path, &nex, 0);
1089 /* free data blocks we just allocated */
1090 /* not a good idea to call discard here directly,
1091 * but otherwise we'd need to call it every free() */
1092 #ifdef EXT3_MB_HINT_GROUP_ALLOC
1093 ext3_mb_discard_inode_preallocations(inode);
1095 ext3_free_blocks(handle, inode, ext_pblock(&nex),
1096 cpu_to_le16(nex.ee_len), 0);
1101 * Putting len of the actual extent we just inserted,
1102 * we are asking ext3_ext_walk_space() to continue
1103 * scaning after that block
1105 cex->ec_len = le16_to_cpu(nex.ee_len);
1106 cex->ec_start = ext_pblock(&nex);
1107 BUG_ON(le16_to_cpu(nex.ee_len) == 0);
1108 BUG_ON(le32_to_cpu(nex.ee_block) != cex->ec_block);
1111 #if defined(HAVE_EXT4_LDISKFS) && defined(WALK_SPACE_HAS_DATA_SEM)
1112 up_write((&EXT4_I(inode)->i_data_sem));
1114 ext3_journal_stop(handle);
1119 CERROR("hmm. why do we find this extent?\n");
1120 CERROR("initial space: %lu:%u\n",
1121 bp->start, bp->init_num);
1122 CERROR("current extent: %u/%u/%llu %d\n",
1123 cex->ec_block, cex->ec_len,
1124 (unsigned long long)cex->ec_start,
1128 if (cex->ec_block < bp->start)
1129 i = bp->start - cex->ec_block;
1130 if (i >= cex->ec_len)
1131 CERROR("nothing to do?! i = %d, e_num = %u\n",
1133 for (; i < cex->ec_len && bp->num; i++) {
1134 *(bp->blocks) = cex->ec_start + i;
1135 if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
1139 /* unmap any possible underlying metadata from
1140 * the block device mapping. bug 6998. */
1141 ll_unmap_underlying_metadata(inode->i_sb,
1153 int fsfilt_map_nblocks(struct inode *inode, unsigned long block,
1154 unsigned long num, unsigned long *blocks,
1155 int *created, int create)
1157 #ifdef EXT3_EXT_HAS_NO_TREE
1158 struct ext3_ext_base *base = inode;
1160 struct ext3_extents_tree tree;
1161 struct ext3_ext_base *base = &tree;
1163 struct bpointers bp;
1166 CDEBUG(D_OTHER, "blocks %lu-%lu requested for inode %u\n",
1167 block, block + num - 1, (unsigned) inode->i_ino);
1169 #ifndef EXT3_EXT_HAS_NO_TREE
1170 ext3_init_tree_desc(base, inode);
1174 bp.created = created;
1176 bp.init_num = bp.num = num;
1179 fsfilt_down_truncate_sem(inode);
1180 err = fsfilt_ext3_ext_walk_space(base, block, num,
1181 ext3_ext_new_extent_cb, &bp);
1182 ext3_ext_invalidate_cache(base);
1183 fsfilt_up_truncate_sem(inode);
1188 int fsfilt_ext3_map_ext_inode_pages(struct inode *inode, struct page **page,
1189 int pages, unsigned long *blocks,
1190 int *created, int create)
1192 int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1194 struct page *fp = NULL;
1197 CDEBUG(D_OTHER, "inode %lu: map %d pages from %lu\n",
1198 inode->i_ino, pages, (*page)->index);
1200 /* pages are sorted already. so, we just have to find
1201 * contig. space and process them properly */
1204 /* start new extent */
1209 } else if (fp->index + clen == (*page)->index) {
1210 /* continue the extent */
1217 /* process found extent */
1218 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1219 clen * blocks_per_page, blocks,
1224 /* look for next extent */
1226 blocks += blocks_per_page * clen;
1227 created += blocks_per_page * clen;
1231 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1232 clen * blocks_per_page, blocks,
1238 extern int ext3_map_inode_page(struct inode *inode, struct page *page,
1239 unsigned long *blocks, int *created, int create);
1240 int fsfilt_ext3_map_bm_inode_pages(struct inode *inode, struct page **page,
1241 int pages, unsigned long *blocks,
1242 int *created, int create)
1244 int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1248 for (i = 0, cr = created, b = blocks; i < pages; i++, page++) {
1249 rc = ext3_map_inode_page(inode, *page, b, cr, create);
1251 CERROR("ino %lu, blk %lu cr %u create %d: rc %d\n",
1252 inode->i_ino, *b, *cr, create, rc);
1256 b += blocks_per_page;
1257 cr += blocks_per_page;
1262 int fsfilt_ext3_map_inode_pages(struct inode *inode, struct page **page,
1263 int pages, unsigned long *blocks,
1264 int *created, int create,
1265 cfs_semaphore_t *optional_sem)
1269 if (EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL) {
1270 rc = fsfilt_ext3_map_ext_inode_pages(inode, page, pages,
1271 blocks, created, create);
1274 if (optional_sem != NULL)
1275 cfs_down(optional_sem);
1276 rc = fsfilt_ext3_map_bm_inode_pages(inode, page, pages, blocks,
1278 if (optional_sem != NULL)
1279 cfs_up(optional_sem);
1284 int fsfilt_ext3_read(struct inode *inode, void *buf, int size, loff_t *offs)
1286 unsigned long block;
1287 struct buffer_head *bh;
1288 int err, blocksize, csize, boffs, osize = size;
1290 /* prevent reading after eof */
1292 if (i_size_read(inode) < *offs + size) {
1293 size = i_size_read(inode) - *offs;
1294 cfs_unlock_kernel();
1296 CDEBUG(D_EXT2, "size %llu is too short for read @%llu\n",
1297 i_size_read(inode), *offs);
1299 } else if (size == 0) {
1303 cfs_unlock_kernel();
1306 blocksize = 1 << inode->i_blkbits;
1309 block = *offs >> inode->i_blkbits;
1310 boffs = *offs & (blocksize - 1);
1311 csize = min(blocksize - boffs, size);
1312 bh = ext3_bread(NULL, inode, block, 0, &err);
1314 CERROR("can't read block: %d\n", err);
1318 memcpy(buf, bh->b_data + boffs, csize);
1327 EXPORT_SYMBOL(fsfilt_ext3_read);
1329 static int fsfilt_ext3_read_record(struct file * file, void *buf,
1330 int size, loff_t *offs)
1333 rc = fsfilt_ext3_read(file->f_dentry->d_inode, buf, size, offs);
1339 int fsfilt_ext3_write_handle(struct inode *inode, void *buf, int bufsize,
1340 loff_t *offs, handle_t *handle)
1342 struct buffer_head *bh = NULL;
1343 loff_t old_size = i_size_read(inode), offset = *offs;
1344 loff_t new_size = i_size_read(inode);
1345 unsigned long block;
1346 int err = 0, blocksize = 1 << inode->i_blkbits, size, boffs;
1348 while (bufsize > 0) {
1352 block = offset >> inode->i_blkbits;
1353 boffs = offset & (blocksize - 1);
1354 size = min(blocksize - boffs, bufsize);
1355 bh = ext3_bread(handle, inode, block, 1, &err);
1357 CERROR("can't read/create block: %d\n", err);
1361 err = ext3_journal_get_write_access(handle, bh);
1363 CERROR("journal_get_write_access() returned error %d\n",
1367 LASSERT(bh->b_data + boffs + size <= bh->b_data + bh->b_size);
1368 memcpy(bh->b_data + boffs, buf, size);
1369 err = ext3_journal_dirty_metadata(handle, bh);
1371 CERROR("journal_dirty_metadata() returned error %d\n",
1375 if (offset + size > new_size)
1376 new_size = offset + size;
1384 /* correct in-core and on-disk sizes */
1385 if (new_size > i_size_read(inode)) {
1387 if (new_size > i_size_read(inode))
1388 i_size_write(inode, new_size);
1389 if (i_size_read(inode) > EXT3_I(inode)->i_disksize)
1390 EXT3_I(inode)->i_disksize = i_size_read(inode);
1391 if (i_size_read(inode) > old_size)
1392 mark_inode_dirty(inode);
1393 cfs_unlock_kernel();
1400 EXPORT_SYMBOL(fsfilt_ext3_write_handle);
1402 static int fsfilt_ext3_write_record(struct file *file, void *buf, int bufsize,
1403 loff_t *offs, int force_sync)
1405 struct inode *inode = file->f_dentry->d_inode;
1407 int err, block_count = 0, blocksize;
1409 /* Determine how many transaction credits are needed */
1410 blocksize = 1 << inode->i_blkbits;
1411 block_count = (*offs & (blocksize - 1)) + bufsize;
1412 block_count = (block_count + blocksize - 1) >> inode->i_blkbits;
1414 handle = ext3_journal_start(inode,
1415 block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2);
1416 if (IS_ERR(handle)) {
1417 CERROR("can't start transaction for %d blocks (%d bytes)\n",
1418 block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2, bufsize);
1419 return PTR_ERR(handle);
1422 err = fsfilt_ext3_write_handle(inode, buf, bufsize, offs, handle);
1424 if (!err && force_sync)
1425 handle->h_sync = 1; /* recovery likes this */
1427 ext3_journal_stop(handle);
1432 static int fsfilt_ext3_setup(struct super_block *sb)
1434 #if ((LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,6)) && \
1435 defined(HAVE_QUOTA_SUPPORT)) || defined(S_PDIROPS)
1436 struct ext3_sb_info *sbi = EXT3_SB(sb);
1438 sbi->dx_lock = fsfilt_ext3_dx_lock;
1439 sbi->dx_unlock = fsfilt_ext3_dx_unlock;
1443 CWARN("Enabling PDIROPS\n");
1444 set_opt(sbi->s_mount_opt, PDIROPS);
1445 sb->s_flags |= S_PDIROPS;
1447 if (!EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1448 CWARN("filesystem doesn't have dir_index feature enabled\n");
1449 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,6)) && defined(HAVE_QUOTA_SUPPORT)
1450 /* enable journaled quota support */
1451 /* kfreed in ext3_put_super() */
1452 sbi->s_qf_names[USRQUOTA] = kstrdup("lquota.user.reserved", GFP_KERNEL);
1453 if (!sbi->s_qf_names[USRQUOTA])
1455 sbi->s_qf_names[GRPQUOTA] = kstrdup("lquota.group.reserved", GFP_KERNEL);
1456 if (!sbi->s_qf_names[GRPQUOTA]) {
1457 kfree(sbi->s_qf_names[USRQUOTA]);
1458 sbi->s_qf_names[USRQUOTA] = NULL;
1461 sbi->s_jquota_fmt = QFMT_LUSTRE;
1462 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,13))
1463 set_opt(sbi->s_mount_opt, QUOTA);
1469 /* If fso is NULL, op is FSFILT operation, otherwise op is number of fso
1470 objects. Logs is number of logfiles to update */
1471 static int fsfilt_ext3_get_op_len(int op, struct fsfilt_objinfo *fso, int logs)
1475 case FSFILT_OP_CREATE:
1476 /* directory leaf, index & indirect & EA*/
1477 return 4 + 3 * logs;
1478 case FSFILT_OP_UNLINK:
1484 struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
1485 int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
1486 int addrpp = EXT3_ADDR_PER_BLOCK(sb) * blockpp;
1487 for (i = 0; i < op; i++, fso++) {
1488 int nblocks = fso->fso_bufcnt * blockpp;
1489 int ndindirect = min(nblocks, addrpp + 1);
1490 int nindir = nblocks + ndindirect + 1;
1494 return needed + 3 * logs;
1500 #ifdef HAVE_QUOTA_SUPPORT
1501 #define DQINFO_COPY(out, in) \
1503 Q_COPY(out, in, dqi_bgrace); \
1504 Q_COPY(out, in, dqi_igrace); \
1505 Q_COPY(out, in, dqi_flags); \
1506 Q_COPY(out, in, dqi_valid); \
1509 #define DQBLK_COPY(out, in) \
1511 Q_COPY(out, in, dqb_bhardlimit); \
1512 Q_COPY(out, in, dqb_bsoftlimit); \
1513 Q_COPY(out, in, dqb_curspace); \
1514 Q_COPY(out, in, dqb_ihardlimit); \
1515 Q_COPY(out, in, dqb_isoftlimit); \
1516 Q_COPY(out, in, dqb_curinodes); \
1517 Q_COPY(out, in, dqb_btime); \
1518 Q_COPY(out, in, dqb_itime); \
1519 Q_COPY(out, in, dqb_valid); \
1522 static int fsfilt_ext3_quotactl(struct super_block *sb,
1523 struct obd_quotactl *oqc)
1525 int i, rc = 0, error = 0;
1526 const struct quotactl_ops *qcop;
1527 struct if_dqinfo *info;
1528 struct if_dqblk *dqblk;
1534 OBD_ALLOC_PTR(info);
1537 OBD_ALLOC_PTR(dqblk);
1543 DQINFO_COPY(info, &oqc->qc_dqinfo);
1544 DQBLK_COPY(dqblk, &oqc->qc_dqblk);
1547 if (oqc->qc_cmd == Q_QUOTAON || oqc->qc_cmd == Q_QUOTAOFF) {
1548 for (i = 0; i < MAXQUOTAS; i++) {
1549 if (!Q_TYPESET(oqc, i))
1552 if (oqc->qc_cmd == Q_QUOTAON) {
1553 char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1555 LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1557 rc = ll_quota_on(sb, i, QFMT_LUSTRE,
1559 } else if (oqc->qc_cmd == Q_QUOTAOFF) {
1560 rc = ll_quota_off(sb, i, 0);
1568 GOTO(out, rc ?: error);
1571 switch (oqc->qc_cmd) {
1574 if (!qcop->get_info)
1575 GOTO(out, rc = -ENOSYS);
1576 rc = qcop->get_info(sb, oqc->qc_type, info);
1580 if (!qcop->set_dqblk)
1581 GOTO(out, rc = -ENOSYS);
1582 rc = qcop->set_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1586 if (!qcop->get_dqblk)
1587 GOTO(out, rc = -ENOSYS);
1588 rc = qcop->get_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1590 dqblk->dqb_valid = QIF_LIMITS | QIF_USAGE;
1593 if (!sb->s_qcop->quota_sync)
1594 GOTO(out, rc = -ENOSYS);
1595 qcop->quota_sync(sb, oqc->qc_type);
1598 CDEBUG(D_WARNING, "invalidating operational quota files\n");
1599 for (i = 0; i < MAXQUOTAS; i++) {
1601 char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1603 LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1605 if (!Q_TYPESET(oqc, i))
1608 fp = filp_open(name[i], O_CREAT | O_TRUNC | O_RDWR, 0644);
1611 CERROR("error invalidating operational quota file"
1612 " %s (rc:%d)\n", name[i], rc);
1620 CERROR("unsupported quotactl command: %d\n", oqc->qc_cmd);
1624 DQINFO_COPY(&oqc->qc_dqinfo, info);
1625 DQBLK_COPY(&oqc->qc_dqblk, dqblk);
1628 OBD_FREE_PTR(dqblk);
1631 CDEBUG(D_QUOTA, "quotactl command %#x, id %u, type %u "
1633 oqc->qc_cmd, oqc->qc_id, oqc->qc_type, rc);
1638 cfs_hlist_node_t dqb_hash; /** quotacheck hash */
1639 cfs_list_t dqb_list; /** in list also */
1640 qid_t dqb_id; /** uid/gid */
1641 short dqb_type; /** USRQUOTA/GRPQUOTA */
1642 qsize_t dqb_bhardlimit; /** block hard limit */
1643 qsize_t dqb_bsoftlimit; /** block soft limit */
1644 qsize_t dqb_curspace; /** current space */
1645 qsize_t dqb_ihardlimit; /** inode hard limit */
1646 qsize_t dqb_isoftlimit; /** inode soft limit */
1647 qsize_t dqb_curinodes; /** current inodes */
1648 obd_time dqb_btime; /** block grace time */
1649 obd_time dqb_itime; /** inode grace time */
1650 __u32 dqb_valid; /** flag for above fields */
1653 static inline unsigned int chkquot_hash(qid_t id, int type)
1654 __attribute__((__const__));
1656 static inline unsigned int chkquot_hash(qid_t id, int type)
1658 return (id * (MAXQUOTAS - type)) % NR_DQHASH;
1661 static inline struct chk_dqblk *
1662 find_chkquot(cfs_hlist_head_t *head, qid_t id, int type)
1664 cfs_hlist_node_t *node;
1665 struct chk_dqblk *cdqb;
1667 cfs_hlist_for_each(node, head) {
1668 cdqb = cfs_hlist_entry(node, struct chk_dqblk, dqb_hash);
1669 if (cdqb->dqb_id == id && cdqb->dqb_type == type)
1676 static struct chk_dqblk *alloc_chkquot(qid_t id, int type)
1678 struct chk_dqblk *cdqb;
1680 OBD_ALLOC_PTR(cdqb);
1682 CFS_INIT_HLIST_NODE(&cdqb->dqb_hash);
1683 CFS_INIT_LIST_HEAD(&cdqb->dqb_list);
1685 cdqb->dqb_type = type;
1691 static struct chk_dqblk *
1692 cqget(struct super_block *sb, cfs_hlist_head_t *hash,
1693 cfs_list_t *list, qid_t id, int type, int first_check)
1695 cfs_hlist_head_t *head = hash + chkquot_hash(id, type);
1696 struct if_dqblk dqb;
1697 struct chk_dqblk *cdqb;
1700 cdqb = find_chkquot(head, id, type);
1704 cdqb = alloc_chkquot(id, type);
1709 rc = sb->s_qcop->get_dqblk(sb, type, id, &dqb);
1711 CERROR("get_dqblk of id %u, type %d failed: %d\n",
1714 DQBLK_COPY(cdqb, &dqb);
1715 cdqb->dqb_curspace = 0;
1716 cdqb->dqb_curinodes = 0;
1720 cfs_hlist_add_head(&cdqb->dqb_hash, head);
1721 cfs_list_add_tail(&cdqb->dqb_list, list);
1726 static inline int quota_onoff(struct super_block *sb, int cmd, int type, int qfmt)
1728 struct obd_quotactl *oqctl;
1731 OBD_ALLOC_PTR(oqctl);
1735 oqctl->qc_cmd = cmd;
1736 oqctl->qc_id = qfmt;
1737 oqctl->qc_type = type;
1738 rc = fsfilt_ext3_quotactl(sb, oqctl);
1740 OBD_FREE_PTR(oqctl);
1744 static inline int read_old_dqinfo(struct super_block *sb, int type,
1745 struct if_dqinfo *dqinfo)
1747 struct obd_quotactl *oqctl;
1751 OBD_ALLOC_PTR(oqctl);
1755 oqctl->qc_cmd = Q_GETINFO;
1756 oqctl->qc_type = type;
1757 rc = fsfilt_ext3_quotactl(sb, oqctl);
1759 ((struct obd_dqinfo *)dqinfo)[type] = oqctl->qc_dqinfo;
1761 OBD_FREE_PTR(oqctl);
1765 #ifndef HAVE_EXT4_LDISKFS
1766 static inline struct ext3_group_desc *
1767 get_group_desc(struct super_block *sb, int group, struct buffer_head **bh)
1769 unsigned long desc_block, desc;
1770 struct ext3_group_desc *gdp;
1772 desc_block = group / EXT3_DESC_PER_BLOCK(sb);
1773 desc = group % EXT3_DESC_PER_BLOCK(sb);
1774 gdp = (struct ext3_group_desc *)
1775 EXT3_SB(sb)->s_group_desc[desc_block]->b_data;
1780 static inline struct buffer_head *
1781 ext3_read_inode_bitmap(struct super_block *sb, unsigned long group)
1783 struct ext3_group_desc *desc;
1784 struct buffer_head *bh;
1786 desc = get_group_desc(sb, group, NULL);
1787 bh = sb_bread(sb, ext3_inode_bitmap(sb, desc));
1791 static __u32 ext3_itable_unused_count(struct super_block *sb,
1792 struct ext3_group_desc *bg) {
1793 return le16_to_cpu(bg->bg_itable_unused);
1796 #define get_group_desc ext3_get_group_desc
1800 cfs_hlist_head_t qckt_hash[NR_DQHASH]; /* quotacheck hash */
1801 cfs_list_t qckt_list; /* quotacheck list */
1802 int qckt_first_check[MAXQUOTAS]; /* 1 if no old quotafile */
1803 struct if_dqinfo qckt_dqinfo[MAXQUOTAS]; /* old dqinfo */
1806 static int add_inode_quota(struct inode *inode, struct qchk_ctxt *qctxt,
1807 struct obd_quotactl *oqc)
1809 struct chk_dqblk *cdqb[MAXQUOTAS] = { NULL, };
1811 qid_t qid[MAXQUOTAS];
1817 qid[USRQUOTA] = inode->i_uid;
1818 qid[GRPQUOTA] = inode->i_gid;
1820 if (S_ISDIR(inode->i_mode) ||
1821 S_ISREG(inode->i_mode) ||
1822 S_ISLNK(inode->i_mode))
1823 size = inode_get_bytes(inode);
1825 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1826 if (!Q_TYPESET(oqc, cnt))
1829 cdqb[cnt] = cqget(inode->i_sb, qctxt->qckt_hash,
1830 &qctxt->qckt_list, qid[cnt], cnt,
1831 qctxt->qckt_first_check[cnt]);
1837 cdqb[cnt]->dqb_curspace += size;
1838 cdqb[cnt]->dqb_curinodes++;
1842 for (i = 0; i < cnt; i++) {
1843 if (!Q_TYPESET(oqc, i))
1846 cdqb[i]->dqb_curspace -= size;
1847 cdqb[i]->dqb_curinodes--;
1854 /* write dqinfo struct in a new quota file */
1855 static int v3_write_dqinfo(struct file *f, int type, struct if_dqinfo *info)
1857 struct v2_disk_dqinfo dqinfo;
1858 __u32 blocks = V2_DQTREEOFF + 1;
1859 loff_t offset = V2_DQINFOOFF;
1862 dqinfo.dqi_bgrace = cpu_to_le32(info->dqi_bgrace);
1863 dqinfo.dqi_igrace = cpu_to_le32(info->dqi_igrace);
1864 dqinfo.dqi_flags = cpu_to_le32(info->dqi_flags & DQF_MASK &
1867 dqinfo.dqi_bgrace = cpu_to_le32(MAX_DQ_TIME);
1868 dqinfo.dqi_igrace = cpu_to_le32(MAX_IQ_TIME);
1869 dqinfo.dqi_flags = 0;
1872 dqinfo.dqi_blocks = cpu_to_le32(blocks);
1873 dqinfo.dqi_free_blk = 0;
1874 dqinfo.dqi_free_entry = 0;
1876 return cfs_user_write(f, (char *)&dqinfo, sizeof(dqinfo), &offset);
1879 static int v3_write_dqheader(struct file *f, int type)
1881 static const __u32 quota_magics[] = V2_INITQMAGICS;
1882 static const __u32 quota_versions[] = LUSTRE_INITQVERSIONS_V2;
1883 struct v2_disk_dqheader dqhead;
1886 CLASSERT(ARRAY_SIZE(quota_magics) == ARRAY_SIZE(quota_versions));
1887 LASSERT(0 <= type && type < ARRAY_SIZE(quota_magics));
1889 dqhead.dqh_magic = cpu_to_le32(quota_magics[type]);
1890 dqhead.dqh_version = cpu_to_le32(quota_versions[type]);
1892 return cfs_user_write(f, (char *)&dqhead, sizeof(dqhead), &offset);
1895 static int create_new_quota_files(struct qchk_ctxt *qctxt,
1896 struct obd_quotactl *oqc)
1901 for (i = 0; i < MAXQUOTAS; i++) {
1902 struct if_dqinfo *info = qctxt->qckt_first_check[i]?
1903 NULL : &qctxt->qckt_dqinfo[i];
1905 const char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1907 if (!Q_TYPESET(oqc, i))
1910 LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1912 file = filp_open(name[i], O_RDWR | O_CREAT | O_TRUNC, 0644);
1915 CERROR("can't create %s file: rc = %d\n",
1920 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
1921 CERROR("file %s is not regular", name[i]);
1922 filp_close(file, 0);
1923 GOTO(out, rc = -EINVAL);
1926 ll_vfs_dq_drop(file->f_dentry->d_inode);
1928 rc = v3_write_dqheader(file, i);
1930 filp_close(file, 0);
1934 rc = v3_write_dqinfo(file, i, info);
1935 filp_close(file, 0);
1945 static int commit_chkquot(struct super_block *sb, struct qchk_ctxt *qctxt,
1946 struct chk_dqblk *cdqb)
1948 struct obd_quotactl *oqc;
1957 now = cfs_time_current_sec();
1959 if (cdqb->dqb_bsoftlimit &&
1960 toqb(cdqb->dqb_curspace) >= cdqb->dqb_bsoftlimit &&
1963 now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_bgrace;
1965 if (cdqb->dqb_isoftlimit &&
1966 cdqb->dqb_curinodes >= cdqb->dqb_isoftlimit &&
1969 now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_igrace;
1971 cdqb->dqb_valid = QIF_ALL;
1973 oqc->qc_cmd = Q_SETQUOTA;
1974 oqc->qc_type = cdqb->dqb_type;
1975 oqc->qc_id = cdqb->dqb_id;
1976 DQBLK_COPY(&oqc->qc_dqblk, cdqb);
1978 rc = fsfilt_ext3_quotactl(sb, oqc);
1983 static int prune_chkquots(struct super_block *sb,
1984 struct qchk_ctxt *qctxt, int error)
1986 struct chk_dqblk *cdqb, *tmp;
1989 cfs_list_for_each_entry_safe(cdqb, tmp, &qctxt->qckt_list, dqb_list) {
1991 rc = commit_chkquot(sb, qctxt, cdqb);
1995 cfs_hlist_del_init(&cdqb->dqb_hash);
1996 cfs_list_del(&cdqb->dqb_list);
2003 #ifndef EXT3_FEATURE_RO_COMPAT_GDT_CSUM
2004 #define EXT3_FEATURE_RO_COMPAT_GDT_CSUM 0x0010
2007 static int fsfilt_ext3_quotacheck(struct super_block *sb,
2008 struct obd_quotactl *oqc)
2010 struct ext3_sb_info *sbi = EXT3_SB(sb);
2011 int i, group, uninit_feat = 0;
2012 struct qchk_ctxt *qctxt;
2013 struct buffer_head *bitmap_bh = NULL;
2014 unsigned long ino, inode_inuse;
2015 struct inode *inode;
2019 /* turn on quota and read dqinfo if existed */
2020 OBD_ALLOC_PTR(qctxt);
2022 oqc->qc_stat = -ENOMEM;
2026 for (i = 0; i < NR_DQHASH; i++)
2027 CFS_INIT_HLIST_HEAD(&qctxt->qckt_hash[i]);
2028 CFS_INIT_LIST_HEAD(&qctxt->qckt_list);
2030 for (i = 0; i < MAXQUOTAS; i++) {
2031 if (!Q_TYPESET(oqc, i))
2034 rc = quota_onoff(sb, Q_QUOTAON, i, oqc->qc_id);
2035 if (!rc || rc == -EBUSY) {
2036 rc = read_old_dqinfo(sb, i, qctxt->qckt_dqinfo);
2039 } else if (rc == -ENOENT || rc == -EINVAL || rc == -EEXIST) {
2040 qctxt->qckt_first_check[i] = 1;
2045 if (EXT3_HAS_RO_COMPAT_FEATURE(sb, EXT3_FEATURE_RO_COMPAT_GDT_CSUM))
2046 /* This filesystem supports the uninit group feature */
2049 /* number of inodes that have been allocated */
2050 inode_inuse = sbi->s_inodes_per_group * sbi->s_groups_count -
2051 percpu_counter_sum(&sbi->s_freeinodes_counter);
2053 /* check quota and update in hash */
2054 for (group = 0; group < sbi->s_groups_count && inode_inuse > 0;
2056 unsigned long used_count = sbi->s_inodes_per_group;
2059 struct ext3_group_desc *desc;
2060 desc = get_group_desc(sb, group, NULL);
2064 /* we don't really need to take the group lock here,
2065 * but it may be useful if one day we support online
2067 #ifdef HAVE_EXT4_LDISKFS
2068 ext4_lock_group(sb, group);
2070 spin_lock(sb_bgl_lock(sbi, group));
2072 if (desc->bg_flags & cpu_to_le16(EXT3_BG_INODE_UNINIT)) {
2073 /* no inode in use in this group, just skip it */
2074 #ifdef HAVE_EXT4_LDISKFS
2075 ext3_unlock_group(sb, group);
2077 spin_unlock(sb_bgl_lock(sbi, group));
2082 used_count -= ext3_itable_unused_count(sb, desc);
2083 #ifdef HAVE_EXT4_LDISKFS
2084 ext3_unlock_group(sb, group);
2086 spin_unlock(sb_bgl_lock(sbi, group));
2090 ino = group * sbi->s_inodes_per_group + 1;
2091 bitmap_bh = ext3_read_inode_bitmap(sb, group);
2093 CERROR("%s: ext3_read_inode_bitmap group %d failed\n",
2099 while (i < used_count &&
2100 (i = ext3_find_next_bit(bitmap_bh->b_data,
2101 used_count, i)) < used_count) {
2104 ino = i + group * sbi->s_inodes_per_group;
2105 if (ino < sbi->s_first_ino)
2107 #if defined(HAVE_EXT4_LDISKFS) || !defined(HAVE_READ_INODE_IN_SBOPS)
2108 inode = ext3_iget(sb, ino);
2110 inode = iget(sb, ino);
2112 if (!inode || IS_ERR(inode))
2115 rc = add_inode_quota(inode, qctxt, oqc);
2125 /* read old quota limits from old quota file. (only for the user
2126 * has limits but hasn't file) */
2127 #ifdef HAVE_QUOTA_SUPPORT
2128 for (i = 0; i < MAXQUOTAS; i++) {
2130 struct dquot_id *dqid, *tmp;
2132 if (!Q_TYPESET(oqc, i))
2135 if (qctxt->qckt_first_check[i])
2139 LASSERT(sb_dqopt(sb)->files[i] != NULL);
2140 CFS_INIT_LIST_HEAD(&id_list);
2141 #ifndef KERNEL_SUPPORTS_QUOTA_READ
2142 rc = lustre_get_qids(sb_dqopt(sb)->files[i], NULL, i, &id_list);
2144 rc = lustre_get_qids(NULL, sb_dqopt(sb)->files[i], i, &id_list);
2147 CERROR("read old limits failed. (rc:%d)\n", rc);
2149 cfs_list_for_each_entry_safe(dqid, tmp, &id_list, di_link) {
2150 cfs_list_del_init(&dqid->di_link);
2153 cqget(sb, qctxt->qckt_hash, &qctxt->qckt_list,
2155 qctxt->qckt_first_check[i]);
2160 /* turn off quota cause we are to dump chk_dqblk to files */
2161 quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type, oqc->qc_id);
2163 rc = create_new_quota_files(qctxt, oqc);
2167 /* we use vfs functions to set dqblk, so turn quota on */
2168 rc = quota_onoff(sb, Q_QUOTAON, oqc->qc_type, oqc->qc_id);
2170 /* dump and free chk_dqblk */
2171 rc = prune_chkquots(sb, qctxt, rc);
2172 OBD_FREE_PTR(qctxt);
2174 /* turn off quota, `lfs quotacheck` will turn on when all
2175 * nodes quotacheck finish. */
2176 quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type, oqc->qc_id);
2180 CERROR("quotacheck failed: rc = %d\n", rc);
2185 static int fsfilt_ext3_quotainfo(struct lustre_quota_info *lqi, int type,
2191 if (lqi->qi_files[type] == NULL) {
2192 CERROR("operate qinfo before it's enabled!\n");
2198 rc = lustre_check_quota_file(lqi, type);
2201 rc = lustre_read_quota_info(lqi, type);
2204 rc = lustre_write_quota_info(lqi, type);
2206 case QFILE_INIT_INFO:
2207 rc = lustre_init_quota_info(lqi, type);
2211 CERROR("quota CONVERT command is not supported\n");
2215 CERROR("Unsupported admin quota file cmd %d\n"
2216 "Are lquota.ko and fsfilt_ldiskfs.ko modules in sync?\n",
2223 static int fsfilt_ext3_qids(struct file *file, struct inode *inode, int type,
2226 return lustre_get_qids(file, inode, type, list);
2229 static int fsfilt_ext3_dquot(struct lustre_dquot *dquot, int cmd)
2234 if (dquot->dq_info->qi_files[dquot->dq_type] == NULL) {
2235 CERROR("operate dquot before it's enabled!\n");
2240 case QFILE_RD_DQUOT:
2241 rc = lustre_read_dquot(dquot);
2243 case QFILE_WR_DQUOT:
2244 if (dquot->dq_dqb.dqb_ihardlimit ||
2245 dquot->dq_dqb.dqb_isoftlimit ||
2246 dquot->dq_dqb.dqb_bhardlimit ||
2247 dquot->dq_dqb.dqb_bsoftlimit)
2248 cfs_clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2250 cfs_set_bit(DQ_FAKE_B, &dquot->dq_flags);
2252 rc = lustre_commit_dquot(dquot);
2257 CERROR("Unsupported admin quota file cmd %d\n", cmd);
2264 static int fsfilt_ext3_get_mblk(struct super_block *sb, int *count,
2265 struct inode *inode, int frags)
2267 #ifdef EXT3_EXT_HAS_NO_TREE
2268 struct ext3_ext_base *base = inode;
2270 struct ext3_extents_tree tree;
2271 struct ext3_ext_base *base = &tree;
2273 ext3_init_tree_desc(base, inode);
2275 /* for an ost_write request, it needs <#fragments> * <tree depth + 1>
2276 * metablocks at maxium b=16542 */
2277 *count = frags * (EXT_DEPTH(base) + 1) * EXT3_BLOCK_SIZE(sb);
2283 lvfs_sbdev_type fsfilt_ext3_journal_sbdev(struct super_block *sb)
2285 return (EXT3_SB(sb)->journal_bdev);
2287 EXPORT_SYMBOL(fsfilt_ext3_journal_sbdev);
2289 ssize_t lustre_read_quota(struct file *f, struct inode *inode, int type,
2290 char *buf, int count, loff_t pos)
2296 CERROR("lustre_read_quota failed for no quota file!\n");
2297 libcfs_debug_dumpstack(NULL);
2301 /* Support for both adm and op quota files must be provided */
2303 rc = fsfilt_ext3_read_record(f, buf, count, &p);
2304 rc = rc < 0 ? rc : p - pos;
2306 struct super_block *sb = inode->i_sb;
2307 rc = sb->s_op->quota_read(sb, type, buf, count, pos);
2312 ssize_t lustre_write_quota(struct file *f, char *buf, int count, loff_t pos)
2317 /* Only adm quota files are supported, op updates are handled by vfs */
2318 rc = fsfilt_ext3_write_record(f, buf, count, &p, 0);
2319 rc = rc < 0 ? rc : p - pos;
2324 void *lustre_quota_journal_start(struct inode *inode, int delete)
2327 unsigned block_count;
2330 /* each indirect block (+4) may become free, attaching to the
2331 * header list of free blocks (+1); the data block (+1) may
2332 * become a free block (+0) or a block with free dqentries (+0) */
2333 block_count = (4 + 1) + 1;
2334 handle = ext3_journal_start(inode,
2335 block_count*FSFILT_DATA_TRANS_BLOCKS(inode->i_sb)+2);
2337 /* indirect blocks are touched (+4), each causes file expansion (+0) or
2338 * freeblk reusage with a header update (+1); dqentry is either reused
2339 * causing update of the entry block (+1), prev (+1) and next (+1) or
2340 * a new block allocation (+1) with a header update (+1) */
2341 block_count = (4 + 1) + 3;
2342 handle = ext3_journal_start(inode,
2343 block_count*FSFILT_DATA_TRANS_BLOCKS(inode->i_sb)+2);
2350 void lustre_quota_journal_stop(void *handle)
2352 ext3_journal_stop((handle_t *)handle);
2355 static int ll_decode_fh_accept(void *context, struct dentry *de)
2360 #ifdef HAVE_EXPORTFS_DECODE_FH
2361 # define ll_exportfs_decode_fh(mnt, fid, len, type, acceptable, context) \
2362 exportfs_decode_fh(mnt, (struct fid*)(fid), len, type, \
2363 acceptable, context)
2365 # define ll_exportfs_decode_fh(mnt, fid, len, type, acceptable, context) \
2366 export_op_default.decode_fh((mnt)->mnt_sb, &(fid)->ino, len, \
2367 type, acceptable, context)
2368 # define FILEID_INO32_GEN 1
2369 extern struct export_operations export_op_default;
2372 struct dentry *fsfilt_ext3_fid2dentry(struct vfsmount *mnt,
2373 struct fsfilt_fid *fid, int ignore_gen)
2375 struct inode *inode;
2376 struct dentry *result;
2378 result = ll_exportfs_decode_fh(mnt, fid, 2, FILEID_INO32_GEN,
2379 ll_decode_fh_accept, NULL);
2380 if (IS_ERR(result)) {
2381 CDEBUG(D_DENTRY, "%s of %u/%u failed %ld\n", __func__,
2382 fid->ino, fid->gen, PTR_ERR(result));
2386 CDEBUG(D_DENTRY, "%s of %u/%u succeeded\n", __func__,
2387 fid->ino, fid->gen);
2388 inode = result->d_inode;
2392 if (inode->i_nlink == 0 &&
2393 inode->i_mode == 0 && LTIME_S(inode->i_ctime) == 0) {
2394 LCONSOLE_WARN("Found inode with zero nlink, mode and"
2395 " ctime -- this may indicate disk "
2396 "corruption (inode: %lu, link: %lu, "
2397 "count: %d)\n", inode->i_ino,
2398 (unsigned long)inode->i_nlink,
2399 atomic_read(&inode->i_count));
2402 if (fid->gen && inode->i_generation != fid->gen) {
2403 /* we didn't find the right inode.. */
2404 CDEBUG(D_INODE, "found wrong generation: inode %lu, link: %lu, "
2405 "count: %d, generation %u/%u\n",
2406 inode->i_ino, (unsigned long)inode->i_nlink,
2407 atomic_read(&inode->i_count), inode->i_generation,
2415 return ERR_PTR(-ENOENT);
2418 static struct fsfilt_operations fsfilt_ext3_ops = {
2420 .fs_owner = THIS_MODULE,
2421 .fs_getlabel = fsfilt_ext3_get_label,
2422 .fs_setlabel = fsfilt_ext3_set_label,
2423 .fs_uuid = fsfilt_ext3_uuid,
2424 .fs_start = fsfilt_ext3_start,
2425 .fs_brw_start = fsfilt_ext3_brw_start,
2426 .fs_extend = fsfilt_ext3_extend,
2427 .fs_commit = fsfilt_ext3_commit,
2428 .fs_commit_async = fsfilt_ext3_commit_async,
2429 .fs_commit_wait = fsfilt_ext3_commit_wait,
2430 .fs_setattr = fsfilt_ext3_setattr,
2431 .fs_iocontrol = fsfilt_ext3_iocontrol,
2432 .fs_set_md = fsfilt_ext3_set_md,
2433 .fs_get_md = fsfilt_ext3_get_md,
2434 .fs_readpage = fsfilt_ext3_readpage,
2435 .fs_add_journal_cb = fsfilt_ext3_add_journal_cb,
2436 .fs_statfs = fsfilt_ext3_statfs,
2437 .fs_sync = fsfilt_ext3_sync,
2438 .fs_map_inode_pages = fsfilt_ext3_map_inode_pages,
2439 .fs_write_record = fsfilt_ext3_write_record,
2440 .fs_read_record = fsfilt_ext3_read_record,
2441 .fs_setup = fsfilt_ext3_setup,
2442 .fs_send_bio = fsfilt_ext3_send_bio,
2443 .fs_get_op_len = fsfilt_ext3_get_op_len,
2444 #ifdef HAVE_DISK_INODE_VERSION
2445 .fs_get_version = fsfilt_ext3_get_version,
2446 .fs_set_version = fsfilt_ext3_set_version,
2448 #ifdef HAVE_QUOTA_SUPPORT
2449 .fs_quotactl = fsfilt_ext3_quotactl,
2450 .fs_quotacheck = fsfilt_ext3_quotacheck,
2451 .fs_quotainfo = fsfilt_ext3_quotainfo,
2452 .fs_qids = fsfilt_ext3_qids,
2453 .fs_dquot = fsfilt_ext3_dquot,
2454 .fs_get_mblk = fsfilt_ext3_get_mblk,
2456 .fs_journal_sbdev = fsfilt_ext3_journal_sbdev,
2457 .fs_fid2dentry = fsfilt_ext3_fid2dentry,
2460 static int __init fsfilt_ext3_init(void)
2464 fcb_cache = cfs_mem_cache_create("fsfilt_ext3_fcb",
2465 sizeof(struct fsfilt_cb_data), 0, 0);
2467 CERROR("error allocating fsfilt journal callback cache\n");
2468 GOTO(out, rc = -ENOMEM);
2471 rc = fsfilt_register_ops(&fsfilt_ext3_ops);
2474 int err = cfs_mem_cache_destroy(fcb_cache);
2475 LASSERTF(err == 0, "error destroying new cache: rc %d\n", err);
2481 static void __exit fsfilt_ext3_exit(void)
2485 fsfilt_unregister_ops(&fsfilt_ext3_ops);
2486 rc = cfs_mem_cache_destroy(fcb_cache);
2487 LASSERTF(rc == 0, "couldn't destroy fcb_cache slab\n");
2490 module_init(fsfilt_ext3_init);
2491 module_exit(fsfilt_ext3_exit);
2493 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2494 MODULE_DESCRIPTION("Lustre ext3 Filesystem Helper v0.1");
2495 MODULE_LICENSE("GPL");