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 static char *fsfilt_ext3_get_label(struct super_block *sb)
139 return EXT3_SB(sb)->s_es->s_volume_name;
142 static int fsfilt_ext3_set_label(struct super_block *sb, char *label)
144 /* see e.g. fsfilt_ext3_write_record() */
149 journal = EXT3_SB(sb)->s_journal;
150 handle = ext3_journal_start_sb(sb, 1);
151 if (IS_ERR(handle)) {
152 CERROR("can't start transaction\n");
153 return(PTR_ERR(handle));
156 err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
160 memcpy(EXT3_SB(sb)->s_es->s_volume_name, label,
161 sizeof(EXT3_SB(sb)->s_es->s_volume_name));
163 err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
166 ext3_journal_stop(handle);
171 static char *fsfilt_ext3_uuid(struct super_block *sb)
173 return EXT3_SB(sb)->s_es->s_uuid;
176 #ifdef HAVE_DISK_INODE_VERSION
178 static __u64 get_i_version(struct inode *inode)
180 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,27)) && defined(HAVE_EXT4_LDISKFS)
181 return inode->i_version;
183 return EXT3_I(inode)->i_fs_version;
187 static void set_i_version(struct inode *inode, __u64 new_version)
189 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,27)) && defined(HAVE_EXT4_LDISKFS)
190 inode->i_version = new_version;
192 (EXT3_I(inode))->i_fs_version = new_version;
197 * Get the 64-bit version for an inode.
199 static __u64 fsfilt_ext3_get_version(struct inode *inode)
201 CDEBUG(D_INFO, "Get version "LPX64" for inode %lu\n",
202 get_i_version(inode), inode->i_ino);
203 return get_i_version(inode);
207 * Set the 64-bit version and return the old version.
209 static __u64 fsfilt_ext3_set_version(struct inode *inode, __u64 new_version)
211 __u64 old_version = get_i_version(inode);
213 CDEBUG(D_INFO, "Set version "LPX64" (old "LPX64") for inode %lu\n",
214 new_version, old_version, inode->i_ino);
215 set_i_version(inode, new_version);
216 /* version is set after all inode operations are finished, so we should
217 * mark it dirty here */
218 inode->i_sb->s_op->dirty_inode(inode);
225 * We don't currently need any additional blocks for rmdir and
226 * unlink transactions because we are storing the OST oa_id inside
227 * the inode (which we will be changing anyways as part of this
230 static void *fsfilt_ext3_start(struct inode *inode, int op, void *desc_private,
233 /* For updates to the last received file */
234 int nblocks = FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
238 if (current->journal_info) {
239 CDEBUG(D_INODE, "increasing refcount on %p\n",
240 current->journal_info);
245 case FSFILT_OP_RMDIR:
246 case FSFILT_OP_UNLINK:
247 /* delete one file + create/update logs for each stripe */
248 nblocks += FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb);
249 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
250 FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
252 case FSFILT_OP_RENAME:
253 /* modify additional directory */
254 nblocks += FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
256 case FSFILT_OP_SYMLINK:
257 /* additional block + block bitmap + GDT for long symlink */
260 case FSFILT_OP_CREATE: {
261 #if defined(EXT3_EXTENTS_FL) && defined(EXT3_INDEX_FL) && !defined(HAVE_EXT4_LDISKFS)
264 if (!test_opt(inode->i_sb, EXTENTS)) {
266 } else if (((EXT3_I(inode)->i_flags &
267 cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL)) ==
268 cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL))) {
269 CWARN("extent-mapped directory found with "
270 "ext3-based ldiskfs - contact "
271 "http://bugzilla.lustre.org/\n");
278 case FSFILT_OP_MKDIR:
279 case FSFILT_OP_MKNOD:
280 /* modify one inode + block bitmap + GDT */
284 /* modify parent directory */
285 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
286 FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
287 /* create/update logs for each stripe */
288 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
289 FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
291 case FSFILT_OP_SETATTR:
292 /* Setattr on inode */
294 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
295 FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
296 /* quota chown log for each stripe */
297 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
298 FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
300 case FSFILT_OP_CANCEL_UNLINK:
301 /* blocks for log header bitmap update OR
302 * blocks for catalog header bitmap update + unlink of logs */
303 nblocks = (LLOG_CHUNK_SIZE >> inode->i_blkbits) +
304 FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb) * logs;
307 /* delete 2 file(file + array id) + create 1 file (array id)
308 * create/update logs for each stripe */
309 nblocks += 2 * FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb);
311 /*create array log for head file*/
313 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
314 FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb));
315 /*update head file array */
316 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
317 FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
319 default: CERROR("unknown transaction start op %d\n", op);
323 LASSERT(current->journal_info == desc_private);
324 journal = EXT3_SB(inode->i_sb)->s_journal;
325 if (nblocks > journal->j_max_transaction_buffers) {
326 CWARN("too many credits %d for op %ux%u using %d instead\n",
327 nblocks, op, logs, journal->j_max_transaction_buffers);
328 nblocks = journal->j_max_transaction_buffers;
332 LASSERTF(nblocks > 0, "can't start %d credit transaction\n", nblocks);
333 handle = ext3_journal_start(inode, nblocks);
336 LASSERT(current->journal_info == handle);
338 CERROR("error starting handle for op %u (%u credits): rc %ld\n",
339 op, nblocks, PTR_ERR(handle));
344 * Calculate the number of buffer credits needed to write multiple pages in
345 * a single ext3 transaction. No, this shouldn't be here, but as yet ext3
346 * doesn't have a nice API for calculating this sort of thing in advance.
348 * See comment above ext3_writepage_trans_blocks for details. We assume
349 * no data journaling is being done, but it does allow for all of the pages
350 * being non-contiguous. If we are guaranteed contiguous pages we could
351 * reduce the number of (d)indirect blocks a lot.
353 * With N blocks per page and P pages, for each inode we have at most:
355 * min(N*P, blocksize/4 + 1) dindirect blocks
358 * For the entire filesystem, we have at most:
359 * min(sum(nindir + P), ngroups) bitmap blocks (from the above)
360 * min(sum(nindir + P), gdblocks) group descriptor blocks (from the above)
361 * objcount inode blocks
363 * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quota files
365 * 1 EXT3_DATA_TRANS_BLOCKS for the last_rcvd update.
367 static int fsfilt_ext3_credits_needed(int objcount, struct fsfilt_objinfo *fso,
368 int niocount, struct niobuf_local *nb)
370 struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
372 const int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
373 int nbitmaps = 0, ngdblocks;
374 int needed = objcount + 1; /* inodes + superblock */
377 for (i = 0, j = 0; i < objcount; i++, fso++) {
378 /* two or more dindirect blocks in case we cross boundary */
379 int ndind = (long)((nb[j + fso->fso_bufcnt - 1].offset -
381 sb->s_blocksize_bits) /
382 (EXT3_ADDR_PER_BLOCK(sb) * EXT3_ADDR_PER_BLOCK(sb));
383 nbitmaps += min(fso->fso_bufcnt, ndind > 0 ? ndind : 2);
385 /* leaf, indirect, tindirect blocks for first block */
386 nbitmaps += blockpp + 2;
388 j += fso->fso_bufcnt;
391 next_indir = nb[0].offset +
392 (EXT3_ADDR_PER_BLOCK(sb) << sb->s_blocksize_bits);
393 for (i = 1; i < niocount; i++) {
394 if (nb[i].offset >= next_indir) {
395 nbitmaps++; /* additional indirect */
396 next_indir = nb[i].offset +
397 (EXT3_ADDR_PER_BLOCK(sb)<<sb->s_blocksize_bits);
398 } else if (nb[i].offset != nb[i - 1].offset + sb->s_blocksize) {
399 nbitmaps++; /* additional indirect */
401 nbitmaps += blockpp; /* each leaf in different group? */
404 ngdblocks = nbitmaps;
405 if (nbitmaps > EXT3_SB(sb)->s_groups_count)
406 nbitmaps = EXT3_SB(sb)->s_groups_count;
407 if (ngdblocks > EXT3_SB(sb)->s_gdb_count)
408 ngdblocks = EXT3_SB(sb)->s_gdb_count;
410 needed += nbitmaps + ngdblocks;
412 /* last_rcvd update */
413 needed += FSFILT_DATA_TRANS_BLOCKS(sb);
415 #if defined(CONFIG_QUOTA)
416 /* We assume that there will be 1 bit set in s_dquot.flags for each
417 * quota file that is active. This is at least true for now.
419 needed += hweight32(ll_sb_any_quota_active(sb)) *
420 FSFILT_SINGLEDATA_TRANS_BLOCKS(sb);
426 /* We have to start a huge journal transaction here to hold all of the
427 * metadata for the pages being written here. This is necessitated by
428 * the fact that we do lots of prepare_write operations before we do
429 * any of the matching commit_write operations, so even if we split
430 * up to use "smaller" transactions none of them could complete until
431 * all of them were opened. By having a single journal transaction,
432 * we eliminate duplicate reservations for common blocks like the
433 * superblock and group descriptors or bitmaps.
435 * We will start the transaction here, but each prepare_write will
436 * add a refcount to the transaction, and each commit_write will
437 * remove a refcount. The transaction will be closed when all of
438 * the pages have been written.
440 static void *fsfilt_ext3_brw_start(int objcount, struct fsfilt_objinfo *fso,
441 int niocount, struct niobuf_local *nb,
442 void *desc_private, int logs)
449 LASSERT(current->journal_info == desc_private);
450 journal = EXT3_SB(fso->fso_dentry->d_inode->i_sb)->s_journal;
451 needed = fsfilt_ext3_credits_needed(objcount, fso, niocount, nb);
453 /* The number of blocks we could _possibly_ dirty can very large.
454 * We reduce our request if it is absurd (and we couldn't get that
455 * many credits for a single handle anyways).
457 * At some point we have to limit the size of I/Os sent at one time,
458 * increase the size of the journal, or we have to calculate the
459 * actual journal requirements more carefully by checking all of
460 * the blocks instead of being maximally pessimistic. It remains to
461 * be seen if this is a real problem or not.
463 if (needed > journal->j_max_transaction_buffers) {
464 CERROR("want too many journal credits (%d) using %d instead\n",
465 needed, journal->j_max_transaction_buffers);
466 needed = journal->j_max_transaction_buffers;
469 LASSERTF(needed > 0, "can't start %d credit transaction\n", needed);
470 handle = ext3_journal_start(fso->fso_dentry->d_inode, needed);
471 if (IS_ERR(handle)) {
472 CERROR("can't get handle for %d credits: rc = %ld\n", needed,
475 LASSERT(handle->h_buffer_credits >= needed);
476 LASSERT(current->journal_info == handle);
482 static int fsfilt_ext3_extend(struct inode *inode, unsigned int nblocks,void *h)
484 handle_t *handle = h;
486 /* fsfilt_extend called with nblocks = 0 for testing in special cases */
488 handle->h_buffer_credits = 0;
489 CWARN("setting credits of handle %p to zero by request\n", h);
492 if (handle->h_buffer_credits > nblocks)
494 if (ext3_journal_extend(handle, nblocks) == 0)
497 ext3_mark_inode_dirty(handle, inode);
498 return ext3_journal_restart(handle, nblocks);
501 static int fsfilt_ext3_commit(struct inode *inode, void *h, int force_sync)
504 handle_t *handle = h;
506 LASSERT(current->journal_info == handle);
508 handle->h_sync = 1; /* recovery likes this */
510 rc = ext3_journal_stop(handle);
515 static int fsfilt_ext3_commit_async(struct inode *inode, void *h,
519 transaction_t *transaction;
520 handle_t *handle = h;
524 LASSERT(current->journal_info == handle);
526 transaction = handle->h_transaction;
527 journal = transaction->t_journal;
528 tid = transaction->t_tid;
529 /* we don't want to be blocked */
531 rc = ext3_journal_stop(handle);
533 CERROR("error while stopping transaction: %d\n", rc);
536 fsfilt_log_start_commit(journal, tid);
538 *wait_handle = (void *) tid;
539 CDEBUG(D_INODE, "commit async: %lu\n", (unsigned long) tid);
543 static int fsfilt_ext3_commit_wait(struct inode *inode, void *h)
545 journal_t *journal = EXT3_JOURNAL(inode);
546 tid_t tid = (tid_t)(long)h;
548 CDEBUG(D_INODE, "commit wait: %lu\n", (unsigned long) tid);
549 if (unlikely(is_journal_aborted(journal)))
552 fsfilt_log_wait_commit(EXT3_JOURNAL(inode), tid);
554 if (unlikely(is_journal_aborted(journal)))
559 static int fsfilt_ext3_setattr(struct dentry *dentry, void *handle,
560 struct iattr *iattr, int do_trunc)
562 struct inode *inode = dentry->d_inode;
565 /* Avoid marking the inode dirty on the superblock list unnecessarily.
566 * We are already writing the inode to disk as part of this
567 * transaction and want to avoid a lot of extra inode writeout
568 * later on. b=9828 */
569 if (iattr->ia_valid & ATTR_SIZE && !do_trunc) {
570 /* ATTR_SIZE would invoke truncate: clear it */
571 iattr->ia_valid &= ~ATTR_SIZE;
572 EXT3_I(inode)->i_disksize = iattr->ia_size;
573 i_size_write(inode, iattr->ia_size);
575 if (iattr->ia_valid & ATTR_UID)
576 inode->i_uid = iattr->ia_uid;
577 if (iattr->ia_valid & ATTR_GID)
578 inode->i_gid = iattr->ia_gid;
579 if (iattr->ia_valid & ATTR_ATIME)
580 inode->i_atime = iattr->ia_atime;
581 if (iattr->ia_valid & ATTR_MTIME)
582 inode->i_mtime = iattr->ia_mtime;
583 if (iattr->ia_valid & ATTR_CTIME)
584 inode->i_ctime = iattr->ia_ctime;
585 if (iattr->ia_valid & ATTR_MODE) {
586 inode->i_mode = iattr->ia_mode;
588 if (!in_group_p(inode->i_gid) &&
589 !cfs_capable(CFS_CAP_FSETID))
590 inode->i_mode &= ~S_ISGID;
593 inode->i_sb->s_op->dirty_inode(inode);
598 /* Don't allow setattr to change file type */
599 if (iattr->ia_valid & ATTR_MODE)
600 iattr->ia_mode = (inode->i_mode & S_IFMT) |
601 (iattr->ia_mode & ~S_IFMT);
603 /* We set these flags on the client, but have already checked perms
604 * so don't confuse inode_change_ok. */
605 iattr->ia_valid &= ~(ATTR_MTIME_SET | ATTR_ATIME_SET);
607 if (inode->i_op->setattr) {
608 rc = inode->i_op->setattr(dentry, iattr);
610 rc = inode_change_ok(inode, iattr);
612 rc = inode_setattr(inode, iattr);
619 static int fsfilt_ext3_iocontrol(struct inode *inode, struct file *file,
620 unsigned int cmd, unsigned long arg)
625 /* FIXME: Can't do this because of nested transaction deadlock */
626 if (cmd == EXT3_IOC_SETFLAGS && (*(int *)arg) & EXT3_JOURNAL_DATA_FL) {
627 CERROR("can't set data journal flag on file\n");
631 #ifdef HAVE_EXT4_LDISKFS
632 /* ext4_ioctl does not have a inode argument */
633 if (inode->i_fop->unlocked_ioctl)
634 rc = inode->i_fop->unlocked_ioctl(file, cmd, arg);
636 if (inode->i_fop->ioctl)
637 rc = inode->i_fop->ioctl(inode, file, cmd, arg);
645 static int fsfilt_ext3_set_md(struct inode *inode, void *handle,
646 void *lmm, int lmm_size, const char *name)
650 LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
652 rc = ext3_xattr_set_handle(handle, inode, EXT3_XATTR_INDEX_TRUSTED,
653 name, lmm, lmm_size, XATTR_NO_CTIME);
656 if (rc && rc != -EROFS)
657 CERROR("error adding MD data to inode %lu: rc = %d\n",
662 /* Must be called with i_mutex held */
663 static int fsfilt_ext3_get_md(struct inode *inode, void *lmm, int lmm_size,
668 LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
670 rc = ext3_xattr_get(inode, EXT3_XATTR_INDEX_TRUSTED,
671 name, lmm, lmm_size);
673 /* This gives us the MD size */
675 return (rc == -ENODATA) ? 0 : rc;
678 CDEBUG(D_INFO, "error getting EA %d/%s from inode %lu: rc %d\n",
679 EXT3_XATTR_INDEX_TRUSTED, name,
681 memset(lmm, 0, lmm_size);
682 return (rc == -ENODATA) ? 0 : rc;
688 static int fsfilt_ext3_send_bio(int rw, struct inode *inode, struct bio *bio)
694 static ssize_t fsfilt_ext3_readpage(struct file *file, char *buf, size_t count,
697 struct inode *inode = file->f_dentry->d_inode;
700 if (S_ISREG(inode->i_mode))
701 rc = file->f_op->read(file, buf, count, off);
703 const int blkbits = inode->i_sb->s_blocksize_bits;
704 const int blksize = inode->i_sb->s_blocksize;
706 CDEBUG(D_EXT2, "reading %lu at dir %lu+%llu\n",
707 (unsigned long)count, inode->i_ino, *off);
709 struct buffer_head *bh;
712 if (*off < i_size_read(inode)) {
715 bh = ext3_bread(NULL, inode, *off >> blkbits,
718 CDEBUG(D_EXT2, "read %u@%llu\n", blksize, *off);
721 memcpy(buf, bh->b_data, blksize);
724 /* XXX in theory we should just fake
725 * this buffer and continue like ext3,
726 * especially if this is a partial read
728 CERROR("error read dir %lu+%llu: %d\n",
729 inode->i_ino, *off, err);
734 struct ext3_dir_entry_2 *fake = (void *)buf;
736 CDEBUG(D_EXT2, "fake %u@%llu\n", blksize, *off);
737 memset(fake, 0, sizeof(*fake));
738 fake->rec_len = cpu_to_le16(blksize);
750 static void fsfilt_ext3_cb_func(struct journal_callback *jcb, int error)
752 struct fsfilt_cb_data *fcb = (struct fsfilt_cb_data *)jcb;
754 fcb->cb_func(fcb->cb_obd, fcb->cb_last_rcvd, fcb->cb_data, error);
756 OBD_SLAB_FREE(fcb, fcb_cache, sizeof *fcb);
759 static int fsfilt_ext3_add_journal_cb(struct obd_device *obd, __u64 last_rcvd,
760 void *handle, fsfilt_cb_t cb_func,
763 struct fsfilt_cb_data *fcb;
765 OBD_SLAB_ALLOC_PTR_GFP(fcb, fcb_cache, CFS_ALLOC_IO);
769 fcb->cb_func = cb_func;
771 fcb->cb_last_rcvd = last_rcvd;
772 fcb->cb_data = cb_data;
774 CDEBUG(D_EXT2, "set callback for last_rcvd: "LPD64"\n", last_rcvd);
775 fsfilt_journal_callback_set(handle, fsfilt_ext3_cb_func,
776 (struct journal_callback *)fcb);
782 * We need to hack the return value for the free inode counts because
783 * the current EA code requires one filesystem block per inode with EAs,
784 * so it is possible to run out of blocks before we run out of inodes.
786 * This can be removed when the ext3 EA code is fixed.
788 static int fsfilt_ext3_statfs(struct super_block *sb, struct obd_statfs *osfs)
793 memset(&sfs, 0, sizeof(sfs));
794 rc = ll_do_statfs(sb, &sfs);
795 if (!rc && sfs.f_bfree < sfs.f_ffree) {
796 sfs.f_files = (sfs.f_files - sfs.f_ffree) + sfs.f_bfree;
797 sfs.f_ffree = sfs.f_bfree;
800 statfs_pack(osfs, &sfs);
804 static int fsfilt_ext3_sync(struct super_block *sb)
806 return ext3_force_commit(sb);
809 #if defined(EXT3_MULTIBLOCK_ALLOCATOR) && (!defined(EXT3_EXT_CACHE_NO) || defined(EXT_CACHE_MARK))
810 #warning "kernel code has old extents/mballoc patch, disabling"
811 #undef EXT3_MULTIBLOCK_ALLOCATOR
813 #ifndef EXT3_EXTENTS_FL
814 #define EXT3_EXTENTS_FL 0x00080000 /* Inode uses extents */
817 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
818 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
819 #define fsfilt_up_truncate_sem(inode) up(&EXT3_I(inode)->truncate_sem);
820 #define fsfilt_down_truncate_sem(inode) down(&EXT3_I(inode)->truncate_sem);
822 #ifdef HAVE_EXT4_LDISKFS
823 #define fsfilt_up_truncate_sem(inode) up_write((&EXT4_I(inode)->i_data_sem));
824 #define fsfilt_down_truncate_sem(inode) down_write((&EXT4_I(inode)->i_data_sem));
826 #define fsfilt_up_truncate_sem(inode) mutex_unlock(&EXT3_I(inode)->truncate_mutex);
827 #define fsfilt_down_truncate_sem(inode) mutex_lock(&EXT3_I(inode)->truncate_mutex);
832 #define EXT_ASSERT(cond) BUG_ON(!(cond))
835 #ifdef EXT3_EXT_HAS_NO_TREE
836 /* for kernels 2.6.18 and later */
837 #ifdef HAVE_EXT4_LDISKFS
838 #define EXT_GENERATION(inode) (EXT4_I(inode)->i_ext_generation)
840 #define EXT_GENERATION(inode) ext_generation(inode)
842 #define ext3_ext_base inode
843 #define ext3_ext_base2inode(inode) (inode)
844 #define EXT_DEPTH(inode) ext_depth(inode)
845 #define fsfilt_ext3_ext_walk_space(inode, block, num, cb, cbdata) \
846 ext3_ext_walk_space(inode, block, num, cb, cbdata);
848 #define ext3_ext_base ext3_extents_tree
849 #define ext3_ext_base2inode(tree) (tree->inode)
850 #define fsfilt_ext3_ext_walk_space(tree, block, num, cb, cbdata) \
851 ext3_ext_walk_space(tree, block, num, cb);
854 #include <linux/lustre_version.h>
857 unsigned long *blocks;
865 static long ext3_ext_find_goal(struct inode *inode, struct ext3_ext_path *path,
866 unsigned long block, int *aflags)
868 struct ext3_inode_info *ei = EXT3_I(inode);
869 unsigned long bg_start;
870 unsigned long colour;
874 struct ext3_extent *ex;
875 depth = path->p_depth;
877 /* try to predict block placement */
878 if ((ex = path[depth].p_ext))
879 return ext_pblock(ex) + (block - le32_to_cpu(ex->ee_block));
881 /* it looks index is empty
882 * try to find starting from index itself */
883 if (path[depth].p_bh)
884 return path[depth].p_bh->b_blocknr;
887 /* OK. use inode's group */
888 bg_start = (ei->i_block_group * EXT3_BLOCKS_PER_GROUP(inode->i_sb)) +
889 le32_to_cpu(EXT3_SB(inode->i_sb)->s_es->s_first_data_block);
890 colour = (current->pid % 16) *
891 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
892 return bg_start + colour + block;
895 #define ll_unmap_underlying_metadata(sb, blocknr) \
896 unmap_underlying_metadata((sb)->s_bdev, blocknr)
898 #ifndef EXT3_MB_HINT_GROUP_ALLOC
899 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
900 struct ext3_ext_path *path, unsigned long block,
901 unsigned long *count, int *err)
903 unsigned long pblock, goal;
905 struct inode *inode = ext3_ext_base2inode(base);
907 goal = ext3_ext_find_goal(inode, path, block, &aflags);
908 aflags |= 2; /* block have been already reserved */
909 pblock = ext3_mb_new_blocks(handle, inode, goal, count, aflags, err);
914 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
915 struct ext3_ext_path *path, unsigned long block,
916 unsigned long *count, int *err)
918 struct inode *inode = ext3_ext_base2inode(base);
919 struct ext3_allocation_request ar;
920 unsigned long pblock;
923 /* find neighbour allocated blocks */
925 *err = ext3_ext_search_left(base, path, &ar.lleft, &ar.pleft);
929 *err = ext3_ext_search_right(base, path, &ar.lright, &ar.pright);
933 /* allocate new block */
934 ar.goal = ext3_ext_find_goal(inode, path, block, &aflags);
938 ar.flags = EXT3_MB_HINT_DATA;
939 pblock = ext3_mb_new_blocks(handle, &ar, err);
945 #ifdef EXT3_EXT_HAS_NO_TREE
946 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
947 struct ext3_ext_path *path,
948 struct ext3_ext_cache *cex,
949 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
950 struct ext3_extent *ex,
954 struct bpointers *bp = cbdata;
956 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
957 struct ext3_ext_path *path,
958 struct ext3_ext_cache *cex
959 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
960 , struct ext3_extent *ex
964 struct bpointers *bp = base->private;
966 struct inode *inode = ext3_ext_base2inode(base);
967 struct ext3_extent nex;
968 unsigned long pblock;
975 EXT_ASSERT(i == path->p_depth);
976 EXT_ASSERT(path[i].p_hdr);
978 if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
983 if (bp->create == 0) {
985 if (cex->ec_block < bp->start)
986 i = bp->start - cex->ec_block;
987 if (i >= cex->ec_len)
988 CERROR("nothing to do?! i = %d, e_num = %u\n",
990 for (; i < cex->ec_len && bp->num; i++) {
1002 tgen = EXT_GENERATION(base);
1003 count = ext3_ext_calc_credits_for_insert(base, path);
1004 fsfilt_up_truncate_sem(inode);
1006 handle = ext3_journal_start(inode, count+EXT3_ALLOC_NEEDED+1);
1007 if (IS_ERR(handle)) {
1008 fsfilt_down_truncate_sem(inode);
1009 return PTR_ERR(handle);
1012 fsfilt_down_truncate_sem(inode);
1013 if (tgen != EXT_GENERATION(base)) {
1014 /* the tree has changed. so path can be invalid at moment */
1015 ext3_journal_stop(handle);
1019 count = cex->ec_len;
1020 pblock = new_blocks(handle, base, path, cex->ec_block, &count, &err);
1023 EXT_ASSERT(count <= cex->ec_len);
1025 /* insert new extent */
1026 nex.ee_block = cpu_to_le32(cex->ec_block);
1027 ext3_ext_store_pblock(&nex, pblock);
1028 nex.ee_len = cpu_to_le16(count);
1029 err = ext3_ext_insert_extent(handle, base, path, &nex);
1031 /* free data blocks we just allocated */
1032 /* not a good idea to call discard here directly,
1033 * but otherwise we'd need to call it every free() */
1034 #ifdef EXT3_MB_HINT_GROUP_ALLOC
1035 ext3_mb_discard_inode_preallocations(inode);
1037 ext3_free_blocks(handle, inode, ext_pblock(&nex),
1038 cpu_to_le16(nex.ee_len), 0);
1043 * Putting len of the actual extent we just inserted,
1044 * we are asking ext3_ext_walk_space() to continue
1045 * scaning after that block
1047 cex->ec_len = le16_to_cpu(nex.ee_len);
1048 cex->ec_start = ext_pblock(&nex);
1049 BUG_ON(le16_to_cpu(nex.ee_len) == 0);
1050 BUG_ON(le32_to_cpu(nex.ee_block) != cex->ec_block);
1053 ext3_journal_stop(handle);
1058 CERROR("hmm. why do we find this extent?\n");
1059 CERROR("initial space: %lu:%u\n",
1060 bp->start, bp->init_num);
1061 CERROR("current extent: %u/%u/%llu %d\n",
1062 cex->ec_block, cex->ec_len,
1063 (unsigned long long)cex->ec_start,
1067 if (cex->ec_block < bp->start)
1068 i = bp->start - cex->ec_block;
1069 if (i >= cex->ec_len)
1070 CERROR("nothing to do?! i = %d, e_num = %u\n",
1072 for (; i < cex->ec_len && bp->num; i++) {
1073 *(bp->blocks) = cex->ec_start + i;
1074 if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
1078 /* unmap any possible underlying metadata from
1079 * the block device mapping. bug 6998. */
1080 ll_unmap_underlying_metadata(inode->i_sb,
1092 int fsfilt_map_nblocks(struct inode *inode, unsigned long block,
1093 unsigned long num, unsigned long *blocks,
1094 int *created, int create)
1096 #ifdef EXT3_EXT_HAS_NO_TREE
1097 struct ext3_ext_base *base = inode;
1099 struct ext3_extents_tree tree;
1100 struct ext3_ext_base *base = &tree;
1102 struct bpointers bp;
1105 CDEBUG(D_OTHER, "blocks %lu-%lu requested for inode %u\n",
1106 block, block + num - 1, (unsigned) inode->i_ino);
1108 #ifndef EXT3_EXT_HAS_NO_TREE
1109 ext3_init_tree_desc(base, inode);
1113 bp.created = created;
1115 bp.init_num = bp.num = num;
1118 fsfilt_down_truncate_sem(inode);
1119 err = fsfilt_ext3_ext_walk_space(base, block, num,
1120 ext3_ext_new_extent_cb, &bp);
1121 ext3_ext_invalidate_cache(base);
1122 fsfilt_up_truncate_sem(inode);
1127 int fsfilt_ext3_map_ext_inode_pages(struct inode *inode, struct page **page,
1128 int pages, unsigned long *blocks,
1129 int *created, int create)
1131 int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1133 struct page *fp = NULL;
1136 CDEBUG(D_OTHER, "inode %lu: map %d pages from %lu\n",
1137 inode->i_ino, pages, (*page)->index);
1139 /* pages are sorted already. so, we just have to find
1140 * contig. space and process them properly */
1143 /* start new extent */
1148 } else if (fp->index + clen == (*page)->index) {
1149 /* continue the extent */
1156 /* process found extent */
1157 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1158 clen * blocks_per_page, blocks,
1163 /* look for next extent */
1165 blocks += blocks_per_page * clen;
1166 created += blocks_per_page * clen;
1170 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1171 clen * blocks_per_page, blocks,
1176 #endif /* EXT3_MULTIBLOCK_ALLOCATOR */
1178 extern int ext3_map_inode_page(struct inode *inode, struct page *page,
1179 unsigned long *blocks, int *created, int create);
1180 int fsfilt_ext3_map_bm_inode_pages(struct inode *inode, struct page **page,
1181 int pages, unsigned long *blocks,
1182 int *created, int create)
1184 int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1188 for (i = 0, cr = created, b = blocks; i < pages; i++, page++) {
1189 rc = ext3_map_inode_page(inode, *page, b, cr, create);
1191 CERROR("ino %lu, blk %lu cr %u create %d: rc %d\n",
1192 inode->i_ino, *b, *cr, create, rc);
1196 b += blocks_per_page;
1197 cr += blocks_per_page;
1202 int fsfilt_ext3_map_inode_pages(struct inode *inode, struct page **page,
1203 int pages, unsigned long *blocks,
1204 int *created, int create,
1205 struct semaphore *optional_sem)
1208 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
1209 if (EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL) {
1210 rc = fsfilt_ext3_map_ext_inode_pages(inode, page, pages,
1211 blocks, created, create);
1215 if (optional_sem != NULL)
1217 rc = fsfilt_ext3_map_bm_inode_pages(inode, page, pages, blocks,
1219 if (optional_sem != NULL)
1225 int fsfilt_ext3_read(struct inode *inode, void *buf, int size, loff_t *offs)
1227 unsigned long block;
1228 struct buffer_head *bh;
1229 int err, blocksize, csize, boffs, osize = size;
1231 /* prevent reading after eof */
1233 if (i_size_read(inode) < *offs + size) {
1234 size = i_size_read(inode) - *offs;
1237 CERROR("size %llu is too short for read %u@%llu\n",
1238 i_size_read(inode), size, *offs);
1240 } else if (size == 0) {
1247 blocksize = 1 << inode->i_blkbits;
1250 block = *offs >> inode->i_blkbits;
1251 boffs = *offs & (blocksize - 1);
1252 csize = min(blocksize - boffs, size);
1253 bh = ext3_bread(NULL, inode, block, 0, &err);
1255 CERROR("can't read block: %d\n", err);
1259 memcpy(buf, bh->b_data + boffs, csize);
1268 EXPORT_SYMBOL(fsfilt_ext3_read);
1270 static int fsfilt_ext3_read_record(struct file * file, void *buf,
1271 int size, loff_t *offs)
1274 rc = fsfilt_ext3_read(file->f_dentry->d_inode, buf, size, offs);
1280 int fsfilt_ext3_write_handle(struct inode *inode, void *buf, int bufsize,
1281 loff_t *offs, handle_t *handle)
1283 struct buffer_head *bh = NULL;
1284 loff_t old_size = i_size_read(inode), offset = *offs;
1285 loff_t new_size = i_size_read(inode);
1286 unsigned long block;
1287 int err = 0, blocksize = 1 << inode->i_blkbits, size, boffs;
1289 while (bufsize > 0) {
1293 block = offset >> inode->i_blkbits;
1294 boffs = offset & (blocksize - 1);
1295 size = min(blocksize - boffs, bufsize);
1296 bh = ext3_bread(handle, inode, block, 1, &err);
1298 CERROR("can't read/create block: %d\n", err);
1302 err = ext3_journal_get_write_access(handle, bh);
1304 CERROR("journal_get_write_access() returned error %d\n",
1308 LASSERT(bh->b_data + boffs + size <= bh->b_data + bh->b_size);
1309 memcpy(bh->b_data + boffs, buf, size);
1310 err = ext3_journal_dirty_metadata(handle, bh);
1312 CERROR("journal_dirty_metadata() returned error %d\n",
1316 if (offset + size > new_size)
1317 new_size = offset + size;
1325 /* correct in-core and on-disk sizes */
1326 if (new_size > i_size_read(inode)) {
1328 if (new_size > i_size_read(inode))
1329 i_size_write(inode, new_size);
1330 if (i_size_read(inode) > EXT3_I(inode)->i_disksize)
1331 EXT3_I(inode)->i_disksize = i_size_read(inode);
1332 if (i_size_read(inode) > old_size)
1333 mark_inode_dirty(inode);
1341 EXPORT_SYMBOL(fsfilt_ext3_write_handle);
1343 static int fsfilt_ext3_write_record(struct file *file, void *buf, int bufsize,
1344 loff_t *offs, int force_sync)
1346 struct inode *inode = file->f_dentry->d_inode;
1348 int err, block_count = 0, blocksize;
1350 /* Determine how many transaction credits are needed */
1351 blocksize = 1 << inode->i_blkbits;
1352 block_count = (*offs & (blocksize - 1)) + bufsize;
1353 block_count = (block_count + blocksize - 1) >> inode->i_blkbits;
1355 handle = ext3_journal_start(inode,
1356 block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2);
1357 if (IS_ERR(handle)) {
1358 CERROR("can't start transaction for %d blocks (%d bytes)\n",
1359 block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2, bufsize);
1360 return PTR_ERR(handle);
1363 err = fsfilt_ext3_write_handle(inode, buf, bufsize, offs, handle);
1365 if (!err && force_sync)
1366 handle->h_sync = 1; /* recovery likes this */
1368 ext3_journal_stop(handle);
1373 static int fsfilt_ext3_setup(struct super_block *sb)
1375 #if ((LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,6)) && \
1376 defined(HAVE_QUOTA_SUPPORT)) || defined(S_PDIROPS)
1377 struct ext3_sb_info *sbi = EXT3_SB(sb);
1379 sbi->dx_lock = fsfilt_ext3_dx_lock;
1380 sbi->dx_unlock = fsfilt_ext3_dx_unlock;
1384 CWARN("Enabling PDIROPS\n");
1385 set_opt(sbi->s_mount_opt, PDIROPS);
1386 sb->s_flags |= S_PDIROPS;
1388 if (!EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1389 CWARN("filesystem doesn't have dir_index feature enabled\n");
1390 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,6)) && defined(HAVE_QUOTA_SUPPORT)
1391 /* enable journaled quota support */
1392 /* kfreed in ext3_put_super() */
1393 sbi->s_qf_names[USRQUOTA] = kstrdup("lquota.user.reserved", GFP_KERNEL);
1394 if (!sbi->s_qf_names[USRQUOTA])
1396 sbi->s_qf_names[GRPQUOTA] = kstrdup("lquota.group.reserved", GFP_KERNEL);
1397 if (!sbi->s_qf_names[GRPQUOTA]) {
1398 kfree(sbi->s_qf_names[USRQUOTA]);
1399 sbi->s_qf_names[USRQUOTA] = NULL;
1402 sbi->s_jquota_fmt = QFMT_VFS_V0;
1403 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,13))
1404 set_opt(sbi->s_mount_opt, QUOTA);
1410 /* If fso is NULL, op is FSFILT operation, otherwise op is number of fso
1411 objects. Logs is number of logfiles to update */
1412 static int fsfilt_ext3_get_op_len(int op, struct fsfilt_objinfo *fso, int logs)
1416 case FSFILT_OP_CREATE:
1417 /* directory leaf, index & indirect & EA*/
1418 return 4 + 3 * logs;
1419 case FSFILT_OP_UNLINK:
1425 struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
1426 int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
1427 int addrpp = EXT3_ADDR_PER_BLOCK(sb) * blockpp;
1428 for (i = 0; i < op; i++, fso++) {
1429 int nblocks = fso->fso_bufcnt * blockpp;
1430 int ndindirect = min(nblocks, addrpp + 1);
1431 int nindir = nblocks + ndindirect + 1;
1435 return needed + 3 * logs;
1441 #ifdef HAVE_QUOTA_SUPPORT
1442 #define DQINFO_COPY(out, in) \
1444 Q_COPY(out, in, dqi_bgrace); \
1445 Q_COPY(out, in, dqi_igrace); \
1446 Q_COPY(out, in, dqi_flags); \
1447 Q_COPY(out, in, dqi_valid); \
1450 #define DQBLK_COPY(out, in) \
1452 Q_COPY(out, in, dqb_bhardlimit); \
1453 Q_COPY(out, in, dqb_bsoftlimit); \
1454 Q_COPY(out, in, dqb_curspace); \
1455 Q_COPY(out, in, dqb_ihardlimit); \
1456 Q_COPY(out, in, dqb_isoftlimit); \
1457 Q_COPY(out, in, dqb_curinodes); \
1458 Q_COPY(out, in, dqb_btime); \
1459 Q_COPY(out, in, dqb_itime); \
1460 Q_COPY(out, in, dqb_valid); \
1463 static int fsfilt_ext3_quotactl(struct super_block *sb,
1464 struct obd_quotactl *oqc)
1466 int i, rc = 0, error = 0;
1467 struct quotactl_ops *qcop;
1468 struct if_dqinfo *info;
1469 struct if_dqblk *dqblk;
1475 OBD_ALLOC_PTR(info);
1478 OBD_ALLOC_PTR(dqblk);
1484 DQINFO_COPY(info, &oqc->qc_dqinfo);
1485 DQBLK_COPY(dqblk, &oqc->qc_dqblk);
1488 if (oqc->qc_cmd == Q_QUOTAON || oqc->qc_cmd == Q_QUOTAOFF) {
1489 for (i = 0; i < MAXQUOTAS; i++) {
1490 if (!Q_TYPESET(oqc, i))
1493 if (oqc->qc_cmd == Q_QUOTAON) {
1494 char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1496 LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1498 rc = ll_quota_on(sb, i, QFMT_VFS_V0,
1500 } else if (oqc->qc_cmd == Q_QUOTAOFF) {
1501 rc = ll_quota_off(sb, i, 0);
1509 GOTO(out, rc ?: error);
1512 switch (oqc->qc_cmd) {
1515 if (!qcop->get_info)
1516 GOTO(out, rc = -ENOSYS);
1517 rc = qcop->get_info(sb, oqc->qc_type, info);
1521 if (!qcop->set_dqblk)
1522 GOTO(out, rc = -ENOSYS);
1523 rc = qcop->set_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1527 if (!qcop->get_dqblk)
1528 GOTO(out, rc = -ENOSYS);
1529 rc = qcop->get_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1531 dqblk->dqb_valid = QIF_LIMITS | QIF_USAGE;
1534 if (!sb->s_qcop->quota_sync)
1535 GOTO(out, rc = -ENOSYS);
1536 qcop->quota_sync(sb, oqc->qc_type);
1539 CDEBUG(D_WARNING, "invalidating operational quota files\n");
1540 for (i = 0; i < MAXQUOTAS; i++) {
1542 char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1544 LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1546 if (!Q_TYPESET(oqc, i))
1549 fp = filp_open(name[i], O_CREAT | O_TRUNC | O_RDWR, 0644);
1552 CERROR("error invalidating operational quota file"
1553 " %s (rc:%d)\n", name[i], rc);
1561 CERROR("unsupported quotactl command: %d\n", oqc->qc_cmd);
1565 DQINFO_COPY(&oqc->qc_dqinfo, info);
1566 DQBLK_COPY(&oqc->qc_dqblk, dqblk);
1569 OBD_FREE_PTR(dqblk);
1572 CDEBUG(D_QUOTA, "quotactl command %#x, id %u, type %u "
1574 oqc->qc_cmd, oqc->qc_id, oqc->qc_type, rc);
1579 struct hlist_node dqb_hash; /** quotacheck hash */
1580 struct list_head dqb_list; /** in list also */
1581 qid_t dqb_id; /** uid/gid */
1582 short dqb_type; /** USRQUOTA/GRPQUOTA */
1583 qsize_t dqb_bhardlimit; /** block hard limit */
1584 qsize_t dqb_bsoftlimit; /** block soft limit */
1585 qsize_t dqb_curspace; /** current space */
1586 qsize_t dqb_ihardlimit; /** inode hard limit */
1587 qsize_t dqb_isoftlimit; /** inode soft limit */
1588 qsize_t dqb_curinodes; /** current inodes */
1589 __u64 dqb_btime; /** block grace time */
1590 __u64 dqb_itime; /** inode grace time */
1591 __u32 dqb_valid; /** flag for above fields */
1594 static inline unsigned int chkquot_hash(qid_t id, int type)
1595 __attribute__((__const__));
1597 static inline unsigned int chkquot_hash(qid_t id, int type)
1599 return (id * (MAXQUOTAS - type)) % NR_DQHASH;
1602 static inline struct chk_dqblk *
1603 find_chkquot(struct hlist_head *head, qid_t id, int type)
1605 struct hlist_node *node;
1606 struct chk_dqblk *cdqb;
1608 hlist_for_each(node, head) {
1609 cdqb = hlist_entry(node, struct chk_dqblk, dqb_hash);
1610 if (cdqb->dqb_id == id && cdqb->dqb_type == type)
1617 static struct chk_dqblk *alloc_chkquot(qid_t id, int type)
1619 struct chk_dqblk *cdqb;
1621 OBD_ALLOC_PTR(cdqb);
1623 INIT_HLIST_NODE(&cdqb->dqb_hash);
1624 INIT_LIST_HEAD(&cdqb->dqb_list);
1626 cdqb->dqb_type = type;
1632 static struct chk_dqblk *
1633 cqget(struct super_block *sb, struct hlist_head *hash, struct list_head *list,
1634 qid_t id, int type, int first_check)
1636 struct hlist_head *head = hash + chkquot_hash(id, type);
1637 struct if_dqblk dqb;
1638 struct chk_dqblk *cdqb;
1641 cdqb = find_chkquot(head, id, type);
1645 cdqb = alloc_chkquot(id, type);
1650 rc = sb->s_qcop->get_dqblk(sb, type, id, &dqb);
1652 CERROR("get_dqblk of id %u, type %d failed: %d\n",
1655 DQBLK_COPY(cdqb, &dqb);
1656 cdqb->dqb_curspace = 0;
1657 cdqb->dqb_curinodes = 0;
1661 hlist_add_head(&cdqb->dqb_hash, head);
1662 list_add_tail(&cdqb->dqb_list, list);
1667 static inline int quota_onoff(struct super_block *sb, int cmd, int type, int qfmt)
1669 struct obd_quotactl *oqctl;
1672 OBD_ALLOC_PTR(oqctl);
1676 oqctl->qc_cmd = cmd;
1677 oqctl->qc_id = qfmt;
1678 oqctl->qc_type = type;
1679 rc = fsfilt_ext3_quotactl(sb, oqctl);
1681 OBD_FREE_PTR(oqctl);
1685 static inline int read_old_dqinfo(struct super_block *sb, int type,
1686 struct if_dqinfo *dqinfo)
1688 struct obd_quotactl *oqctl;
1692 OBD_ALLOC_PTR(oqctl);
1696 oqctl->qc_cmd = Q_GETINFO;
1697 oqctl->qc_type = type;
1698 rc = fsfilt_ext3_quotactl(sb, oqctl);
1700 ((struct obd_dqinfo *)dqinfo)[type] = oqctl->qc_dqinfo;
1702 OBD_FREE_PTR(oqctl);
1706 static inline struct ext3_group_desc *
1707 get_group_desc(struct super_block *sb, int group)
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;
1721 #ifndef HAVE_EXT4_LDISKFS
1722 static inline struct buffer_head *
1723 ext3_read_inode_bitmap(struct super_block *sb, unsigned long group)
1725 struct ext3_group_desc *desc;
1726 struct buffer_head *bh;
1728 desc = get_group_desc(sb, group);
1729 bh = sb_bread(sb, ext3_inode_bitmap(sb, desc));
1734 static inline struct inode *ext3_iget_inuse(struct super_block *sb,
1735 struct buffer_head *bitmap_bh,
1736 int index, unsigned long ino)
1738 struct inode *inode = NULL;
1741 if (ext3_test_bit(index, bitmap_bh->b_data))
1742 #ifdef HAVE_EXT4_LDISKFS
1743 inode = ext4_iget(sb, ino);
1745 /* Newer kernels return an error instead of a NULL pointer */
1748 inode = iget(sb, ino);
1754 struct hlist_head qckt_hash[NR_DQHASH]; /* quotacheck hash */
1755 struct list_head qckt_list; /* quotacheck list */
1756 int qckt_first_check[MAXQUOTAS]; /* 1 if no old quotafile */
1757 struct if_dqinfo qckt_dqinfo[MAXQUOTAS]; /* old dqinfo */
1760 static int add_inode_quota(struct inode *inode, struct qchk_ctxt *qctxt,
1761 struct obd_quotactl *oqc)
1763 struct chk_dqblk *cdqb[MAXQUOTAS] = { NULL, };
1765 qid_t qid[MAXQUOTAS];
1771 qid[USRQUOTA] = inode->i_uid;
1772 qid[GRPQUOTA] = inode->i_gid;
1774 if (S_ISDIR(inode->i_mode) ||
1775 S_ISREG(inode->i_mode) ||
1776 S_ISLNK(inode->i_mode))
1777 size = inode_get_bytes(inode);
1779 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1780 if (!Q_TYPESET(oqc, cnt))
1783 cdqb[cnt] = cqget(inode->i_sb, qctxt->qckt_hash,
1784 &qctxt->qckt_list, qid[cnt], cnt,
1785 qctxt->qckt_first_check[cnt]);
1791 cdqb[cnt]->dqb_curspace += size;
1792 cdqb[cnt]->dqb_curinodes++;
1796 for (i = 0; i < cnt; i++) {
1797 if (!Q_TYPESET(oqc, i))
1800 cdqb[i]->dqb_curspace -= size;
1801 cdqb[i]->dqb_curinodes--;
1808 /* write dqinfo struct in a new quota file */
1809 static int v3_write_dqinfo(struct file *f, int type, struct if_dqinfo *info)
1811 struct v2_disk_dqinfo dqinfo;
1812 __u32 blocks = V2_DQTREEOFF + 1;
1813 loff_t offset = V2_DQINFOOFF;
1816 dqinfo.dqi_bgrace = cpu_to_le32(info->dqi_bgrace);
1817 dqinfo.dqi_igrace = cpu_to_le32(info->dqi_igrace);
1818 dqinfo.dqi_flags = cpu_to_le32(info->dqi_flags & DQF_MASK &
1821 dqinfo.dqi_bgrace = cpu_to_le32(MAX_DQ_TIME);
1822 dqinfo.dqi_igrace = cpu_to_le32(MAX_IQ_TIME);
1823 dqinfo.dqi_flags = 0;
1826 dqinfo.dqi_blocks = cpu_to_le32(blocks);
1827 dqinfo.dqi_free_blk = 0;
1828 dqinfo.dqi_free_entry = 0;
1830 return cfs_user_write(f, (char *)&dqinfo, sizeof(dqinfo), &offset);
1833 static int v3_write_dqheader(struct file *f, int type)
1835 static const __u32 quota_magics[] = V2_INITQMAGICS;
1836 static const __u32 quota_versions[] = V2_INITQVERSIONS_R1;
1837 struct v2_disk_dqheader dqhead;
1840 CLASSERT(ARRAY_SIZE(quota_magics) == ARRAY_SIZE(quota_versions));
1841 LASSERT(0 <= type && type < ARRAY_SIZE(quota_magics));
1843 dqhead.dqh_magic = cpu_to_le32(quota_magics[type]);
1844 dqhead.dqh_version = cpu_to_le32(quota_versions[type]);
1846 return cfs_user_write(f, (char *)&dqhead, sizeof(dqhead), &offset);
1849 static int create_new_quota_files(struct qchk_ctxt *qctxt,
1850 struct obd_quotactl *oqc)
1855 for (i = 0; i < MAXQUOTAS; i++) {
1856 struct if_dqinfo *info = qctxt->qckt_first_check[i]?
1857 NULL : &qctxt->qckt_dqinfo[i];
1859 const char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1861 if (!Q_TYPESET(oqc, i))
1864 LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1866 file = filp_open(name[i], O_RDWR | O_CREAT | O_TRUNC, 0644);
1869 CERROR("can't create %s file: rc = %d\n",
1874 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
1875 CERROR("file %s is not regular", name[i]);
1876 filp_close(file, 0);
1877 GOTO(out, rc = -EINVAL);
1880 DQUOT_DROP(file->f_dentry->d_inode);
1882 rc = v3_write_dqheader(file, i);
1884 filp_close(file, 0);
1888 rc = v3_write_dqinfo(file, i, info);
1889 filp_close(file, 0);
1899 static int commit_chkquot(struct super_block *sb, struct qchk_ctxt *qctxt,
1900 struct chk_dqblk *cdqb)
1902 struct obd_quotactl *oqc;
1911 now = cfs_time_current_sec();
1913 if (cdqb->dqb_bsoftlimit &&
1914 toqb(cdqb->dqb_curspace) >= cdqb->dqb_bsoftlimit &&
1917 now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_bgrace;
1919 if (cdqb->dqb_isoftlimit &&
1920 cdqb->dqb_curinodes >= cdqb->dqb_isoftlimit &&
1923 now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_igrace;
1925 cdqb->dqb_valid = QIF_ALL;
1927 oqc->qc_cmd = Q_SETQUOTA;
1928 oqc->qc_type = cdqb->dqb_type;
1929 oqc->qc_id = cdqb->dqb_id;
1930 DQBLK_COPY(&oqc->qc_dqblk, cdqb);
1932 rc = fsfilt_ext3_quotactl(sb, oqc);
1937 static int prune_chkquots(struct super_block *sb,
1938 struct qchk_ctxt *qctxt, int error)
1940 struct chk_dqblk *cdqb, *tmp;
1943 list_for_each_entry_safe(cdqb, tmp, &qctxt->qckt_list, dqb_list) {
1945 rc = commit_chkquot(sb, qctxt, cdqb);
1949 hlist_del_init(&cdqb->dqb_hash);
1950 list_del(&cdqb->dqb_list);
1957 static int fsfilt_ext3_quotacheck(struct super_block *sb,
1958 struct obd_quotactl *oqc)
1960 struct ext3_sb_info *sbi = EXT3_SB(sb);
1962 struct qchk_ctxt *qctxt;
1963 struct buffer_head *bitmap_bh = NULL;
1965 struct inode *inode;
1969 /* turn on quota and read dqinfo if existed */
1970 OBD_ALLOC_PTR(qctxt);
1972 oqc->qc_stat = -ENOMEM;
1976 for (i = 0; i < NR_DQHASH; i++)
1977 INIT_HLIST_HEAD(&qctxt->qckt_hash[i]);
1978 INIT_LIST_HEAD(&qctxt->qckt_list);
1980 for (i = 0; i < MAXQUOTAS; i++) {
1981 if (!Q_TYPESET(oqc, i))
1984 rc = quota_onoff(sb, Q_QUOTAON, i, oqc->qc_id);
1985 if (!rc || rc == -EBUSY) {
1986 rc = read_old_dqinfo(sb, i, qctxt->qckt_dqinfo);
1989 } else if (rc == -ENOENT || rc == -EINVAL || rc == -EEXIST) {
1990 qctxt->qckt_first_check[i] = 1;
1996 /* check quota and update in hash */
1997 for (group = 0; group < sbi->s_groups_count; group++) {
1998 struct ext3_group_desc *desc;
1999 desc = get_group_desc(sb, group);
2003 spin_lock(sb_bgl_lock(sbi, group));
2004 if (desc->bg_flags & cpu_to_le16(EXT3_BG_INODE_UNINIT)) {
2005 /* no inode in use in this group, just skip it */
2006 spin_unlock(sb_bgl_lock(sbi, group));
2009 spin_unlock(sb_bgl_lock(sbi, group));
2011 ino = group * sbi->s_inodes_per_group + 1;
2012 bitmap_bh = ext3_read_inode_bitmap(sb, group);
2014 CERROR("ext3_read_inode_bitmap group %d failed", group);
2015 GOTO(out, rc = -EIO);
2018 for (i = 0; i < sbi->s_inodes_per_group; i++, ino++) {
2019 if (ino < sbi->s_first_ino)
2022 inode = ext3_iget_inuse(sb, bitmap_bh, i, ino);
2023 rc = add_inode_quota(inode, qctxt, oqc);
2034 /* read old quota limits from old quota file. (only for the user
2035 * has limits but hasn't file) */
2036 #ifdef HAVE_QUOTA_SUPPORT
2037 for (i = 0; i < MAXQUOTAS; i++) {
2038 struct list_head id_list;
2039 struct dquot_id *dqid, *tmp;
2041 if (!Q_TYPESET(oqc, i))
2044 if (qctxt->qckt_first_check[i])
2048 LASSERT(sb_dqopt(sb)->files[i] != NULL);
2049 INIT_LIST_HEAD(&id_list);
2050 #ifndef KERNEL_SUPPORTS_QUOTA_READ
2051 rc = lustre_get_qids(sb_dqopt(sb)->files[i], NULL, i, &id_list);
2053 rc = lustre_get_qids(NULL, sb_dqopt(sb)->files[i], i, &id_list);
2056 CERROR("read old limits failed. (rc:%d)\n", rc);
2058 list_for_each_entry_safe(dqid, tmp, &id_list, di_link) {
2059 list_del_init(&dqid->di_link);
2062 cqget(sb, qctxt->qckt_hash, &qctxt->qckt_list,
2064 qctxt->qckt_first_check[i]);
2069 /* turn off quota cause we are to dump chk_dqblk to files */
2070 quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type, oqc->qc_id);
2072 rc = create_new_quota_files(qctxt, oqc);
2076 /* we use vfs functions to set dqblk, so turn quota on */
2077 rc = quota_onoff(sb, Q_QUOTAON, oqc->qc_type, oqc->qc_id);
2079 /* dump and free chk_dqblk */
2080 rc = prune_chkquots(sb, qctxt, rc);
2081 OBD_FREE_PTR(qctxt);
2083 /* turn off quota, `lfs quotacheck` will turn on when all
2084 * nodes quotacheck finish. */
2085 quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type, oqc->qc_id);
2089 CERROR("quotacheck failed: rc = %d\n", rc);
2094 static int fsfilt_ext3_quotainfo(struct lustre_quota_info *lqi, int type,
2100 if (lqi->qi_files[type] == NULL) {
2101 CERROR("operate qinfo before it's enabled!\n");
2107 rc = lustre_check_quota_file(lqi, type);
2110 rc = lustre_read_quota_info(lqi, type);
2113 rc = lustre_write_quota_info(lqi, type);
2115 case QFILE_INIT_INFO:
2116 rc = lustre_init_quota_info(lqi, type);
2120 CERROR("quota CONVERT command is not supported\n");
2124 CERROR("Unsupported admin quota file cmd %d\n"
2125 "Are lquota.ko and fsfilt_ldiskfs.ko modules in sync?\n",
2132 static int fsfilt_ext3_qids(struct file *file, struct inode *inode, int type,
2133 struct list_head *list)
2135 return lustre_get_qids(file, inode, type, list);
2138 static int fsfilt_ext3_dquot(struct lustre_dquot *dquot, int cmd)
2143 if (dquot->dq_info->qi_files[dquot->dq_type] == NULL) {
2144 CERROR("operate dquot before it's enabled!\n");
2149 case QFILE_RD_DQUOT:
2150 rc = lustre_read_dquot(dquot);
2152 case QFILE_WR_DQUOT:
2153 if (dquot->dq_dqb.dqb_ihardlimit ||
2154 dquot->dq_dqb.dqb_isoftlimit ||
2155 dquot->dq_dqb.dqb_bhardlimit ||
2156 dquot->dq_dqb.dqb_bsoftlimit)
2157 clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2159 set_bit(DQ_FAKE_B, &dquot->dq_flags);
2161 rc = lustre_commit_dquot(dquot);
2166 CERROR("Unsupported admin quota file cmd %d\n", cmd);
2173 static int fsfilt_ext3_get_mblk(struct super_block *sb, int *count,
2174 struct inode *inode, int frags)
2176 #ifdef EXT3_EXT_HAS_NO_TREE
2177 struct ext3_ext_base *base = inode;
2179 struct ext3_extents_tree tree;
2180 struct ext3_ext_base *base = &tree;
2182 ext3_init_tree_desc(base, inode);
2184 /* for an ost_write request, it needs <#fragments> * <tree depth + 1>
2185 * metablocks at maxium b=16542 */
2186 *count = frags * (EXT_DEPTH(base) + 1) * EXT3_BLOCK_SIZE(sb);
2192 lvfs_sbdev_type fsfilt_ext3_journal_sbdev(struct super_block *sb)
2194 return (EXT3_SB(sb)->journal_bdev);
2196 EXPORT_SYMBOL(fsfilt_ext3_journal_sbdev);
2198 static struct fsfilt_operations fsfilt_ext3_ops = {
2200 .fs_owner = THIS_MODULE,
2201 .fs_getlabel = fsfilt_ext3_get_label,
2202 .fs_setlabel = fsfilt_ext3_set_label,
2203 .fs_uuid = fsfilt_ext3_uuid,
2204 .fs_start = fsfilt_ext3_start,
2205 .fs_brw_start = fsfilt_ext3_brw_start,
2206 .fs_extend = fsfilt_ext3_extend,
2207 .fs_commit = fsfilt_ext3_commit,
2208 .fs_commit_async = fsfilt_ext3_commit_async,
2209 .fs_commit_wait = fsfilt_ext3_commit_wait,
2210 .fs_setattr = fsfilt_ext3_setattr,
2211 .fs_iocontrol = fsfilt_ext3_iocontrol,
2212 .fs_set_md = fsfilt_ext3_set_md,
2213 .fs_get_md = fsfilt_ext3_get_md,
2214 .fs_readpage = fsfilt_ext3_readpage,
2215 .fs_add_journal_cb = fsfilt_ext3_add_journal_cb,
2216 .fs_statfs = fsfilt_ext3_statfs,
2217 .fs_sync = fsfilt_ext3_sync,
2218 .fs_map_inode_pages = fsfilt_ext3_map_inode_pages,
2219 .fs_write_record = fsfilt_ext3_write_record,
2220 .fs_read_record = fsfilt_ext3_read_record,
2221 .fs_setup = fsfilt_ext3_setup,
2222 .fs_send_bio = fsfilt_ext3_send_bio,
2223 .fs_get_op_len = fsfilt_ext3_get_op_len,
2224 #ifdef HAVE_DISK_INODE_VERSION
2225 .fs_get_version = fsfilt_ext3_get_version,
2226 .fs_set_version = fsfilt_ext3_set_version,
2228 #ifdef HAVE_QUOTA_SUPPORT
2229 .fs_quotactl = fsfilt_ext3_quotactl,
2230 .fs_quotacheck = fsfilt_ext3_quotacheck,
2231 .fs_quotainfo = fsfilt_ext3_quotainfo,
2232 .fs_qids = fsfilt_ext3_qids,
2233 .fs_dquot = fsfilt_ext3_dquot,
2234 .fs_get_mblk = fsfilt_ext3_get_mblk,
2236 .fs_journal_sbdev = fsfilt_ext3_journal_sbdev,
2239 static int __init fsfilt_ext3_init(void)
2243 fcb_cache = cfs_mem_cache_create("fsfilt_ext3_fcb",
2244 sizeof(struct fsfilt_cb_data), 0, 0);
2246 CERROR("error allocating fsfilt journal callback cache\n");
2247 GOTO(out, rc = -ENOMEM);
2250 rc = fsfilt_register_ops(&fsfilt_ext3_ops);
2253 int err = cfs_mem_cache_destroy(fcb_cache);
2254 LASSERTF(err == 0, "error destroying new cache: rc %d\n", err);
2260 static void __exit fsfilt_ext3_exit(void)
2264 fsfilt_unregister_ops(&fsfilt_ext3_ops);
2265 rc = cfs_mem_cache_destroy(fcb_cache);
2266 LASSERTF(rc == 0, "couldn't destroy fcb_cache slab\n");
2269 module_init(fsfilt_ext3_init);
2270 module_exit(fsfilt_ext3_exit);
2272 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2273 MODULE_DESCRIPTION("Lustre ext3 Filesystem Helper v0.1");
2274 MODULE_LICENSE("GPL");