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/jbd.h>
47 #include <linux/slab.h>
48 #include <linux/pagemap.h>
49 #include <linux/quotaops.h>
50 #include <linux/ext3_fs.h>
51 #include <linux/ext3_jbd.h>
52 #include <linux/version.h>
53 #include <linux/bitops.h>
54 #include <linux/quota.h>
55 #include <linux/quotaio_v1.h>
56 #include <linux/quotaio_v2.h>
57 #include <ext3/xattr.h>
59 #include <libcfs/libcfs.h>
60 #include <lustre_fsfilt.h>
62 #include <lustre_quota.h>
63 #include <linux/lustre_compat25.h>
64 #include <linux/lprocfs_status.h>
66 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
67 #include <linux/ext3_extents.h>
70 #include "lustre_quota_fmt.h"
72 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,15)
73 #define FSFILT_DATA_TRANS_BLOCKS(sb) EXT3_DATA_TRANS_BLOCKS
74 #define FSFILT_DELETE_TRANS_BLOCKS(sb) EXT3_DELETE_TRANS_BLOCKS
76 #define FSFILT_DATA_TRANS_BLOCKS(sb) EXT3_DATA_TRANS_BLOCKS(sb)
77 #define FSFILT_DELETE_TRANS_BLOCKS(sb) EXT3_DELETE_TRANS_BLOCKS(sb)
80 #ifdef EXT3_SINGLEDATA_TRANS_BLOCKS_HAS_SB
81 /* for kernels 2.6.18 and later */
82 #define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS(sb)
84 #define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS
87 #define fsfilt_ext3_journal_start(inode, nblocks) ext3_journal_start(inode, nblocks)
88 #define fsfilt_ext3_journal_stop(handle) ext3_journal_stop(handle)
90 static cfs_mem_cache_t *fcb_cache;
92 struct fsfilt_cb_data {
93 struct journal_callback cb_jcb; /* jbd private data - MUST BE FIRST */
94 fsfilt_cb_t cb_func; /* MDS/OBD completion function */
95 struct obd_device *cb_obd; /* MDS/OBD completion device */
96 __u64 cb_last_rcvd; /* MDS/OST last committed operation */
97 void *cb_data; /* MDS/OST completion function data */
100 #ifndef EXT3_XATTR_INDEX_TRUSTED /* temporary until we hit l28 kernel */
101 #define EXT3_XATTR_INDEX_TRUSTED 4
104 static char *fsfilt_ext3_get_label(struct super_block *sb)
106 return EXT3_SB(sb)->s_es->s_volume_name;
109 static int fsfilt_ext3_set_label(struct super_block *sb, char *label)
111 /* see e.g. fsfilt_ext3_write_record() */
116 journal = EXT3_SB(sb)->s_journal;
117 handle = journal_start(journal, 1);
118 if (IS_ERR(handle)) {
119 CERROR("can't start transaction\n");
120 return(PTR_ERR(handle));
123 err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
127 memcpy(EXT3_SB(sb)->s_es->s_volume_name, label,
128 sizeof(EXT3_SB(sb)->s_es->s_volume_name));
130 err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
133 journal_stop(handle);
138 static char *fsfilt_ext3_uuid(struct super_block *sb)
140 return EXT3_SB(sb)->s_es->s_uuid;
143 #ifdef HAVE_DISK_INODE_VERSION
145 * Get the 64-bit version for an inode.
147 static __u64 fsfilt_ext3_get_version(struct inode *inode)
149 return EXT3_I(inode)->i_fs_version;
153 * Set the 64-bit version and return the old version.
155 static __u64 fsfilt_ext3_set_version(struct inode *inode, __u64 new_version)
157 __u64 old_version = EXT3_I(inode)->i_fs_version;
159 (EXT3_I(inode))->i_fs_version = new_version;
166 * We don't currently need any additional blocks for rmdir and
167 * unlink transactions because we are storing the OST oa_id inside
168 * the inode (which we will be changing anyways as part of this
171 static void *fsfilt_ext3_start(struct inode *inode, int op, void *desc_private,
174 /* For updates to the last received file */
175 int nblocks = FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
179 if (current->journal_info) {
180 CDEBUG(D_INODE, "increasing refcount on %p\n",
181 current->journal_info);
186 case FSFILT_OP_RMDIR:
187 case FSFILT_OP_UNLINK:
188 /* delete one file + create/update logs for each stripe */
189 nblocks += FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb);
190 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
191 FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
193 case FSFILT_OP_RENAME:
194 /* modify additional directory */
195 nblocks += FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
197 case FSFILT_OP_SYMLINK:
198 /* additional block + block bitmap + GDT for long symlink */
201 case FSFILT_OP_CREATE: {
202 #if defined(EXT3_EXTENTS_FL) && defined(EXT3_INDEX_FL)
205 if (!test_opt(inode->i_sb, EXTENTS)) {
207 } else if (((EXT3_I(inode)->i_flags &
208 cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL)) ==
209 cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL))) {
210 CWARN("extent-mapped directory found - contact "
211 "http://bugzilla.lustre.org/\n");
218 case FSFILT_OP_MKDIR:
219 case FSFILT_OP_MKNOD:
220 /* modify one inode + block bitmap + GDT */
224 /* modify parent directory */
225 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
226 FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
227 /* create/update logs for each stripe */
228 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
229 FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
231 case FSFILT_OP_SETATTR:
232 /* Setattr on inode */
234 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
235 FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
236 /* quota chown log for each stripe */
237 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
238 FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
240 case FSFILT_OP_CANCEL_UNLINK:
241 /* blocks for log header bitmap update OR
242 * blocks for catalog header bitmap update + unlink of logs */
243 nblocks = (LLOG_CHUNK_SIZE >> inode->i_blkbits) +
244 FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb) * logs;
247 /* delete 2 file(file + array id) + create 1 file (array id)
248 * create/update logs for each stripe */
249 nblocks += 2 * FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb);
251 /*create array log for head file*/
253 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
254 FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb));
255 /*update head file array */
256 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
257 FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
259 default: CERROR("unknown transaction start op %d\n", op);
263 LASSERT(current->journal_info == desc_private);
264 journal = EXT3_SB(inode->i_sb)->s_journal;
265 if (nblocks > journal->j_max_transaction_buffers) {
266 CWARN("too many credits %d for op %ux%u using %d instead\n",
267 nblocks, op, logs, journal->j_max_transaction_buffers);
268 nblocks = journal->j_max_transaction_buffers;
272 LASSERTF(nblocks > 0, "can't start %d credit transaction\n", nblocks);
273 handle = fsfilt_ext3_journal_start(inode, nblocks);
276 LASSERT(current->journal_info == handle);
278 CERROR("error starting handle for op %u (%u credits): rc %ld\n",
279 op, nblocks, PTR_ERR(handle));
284 * Calculate the number of buffer credits needed to write multiple pages in
285 * a single ext3 transaction. No, this shouldn't be here, but as yet ext3
286 * doesn't have a nice API for calculating this sort of thing in advance.
288 * See comment above ext3_writepage_trans_blocks for details. We assume
289 * no data journaling is being done, but it does allow for all of the pages
290 * being non-contiguous. If we are guaranteed contiguous pages we could
291 * reduce the number of (d)indirect blocks a lot.
293 * With N blocks per page and P pages, for each inode we have at most:
295 * min(N*P, blocksize/4 + 1) dindirect blocks
298 * For the entire filesystem, we have at most:
299 * min(sum(nindir + P), ngroups) bitmap blocks (from the above)
300 * min(sum(nindir + P), gdblocks) group descriptor blocks (from the above)
301 * objcount inode blocks
303 * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quota files
305 * 1 EXT3_DATA_TRANS_BLOCKS for the last_rcvd update.
307 static int fsfilt_ext3_credits_needed(int objcount, struct fsfilt_objinfo *fso,
308 int niocount, struct niobuf_local *nb)
310 struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
312 const int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
313 int nbitmaps = 0, ngdblocks;
314 int needed = objcount + 1; /* inodes + superblock */
317 for (i = 0, j = 0; i < objcount; i++, fso++) {
318 /* two or more dindirect blocks in case we cross boundary */
319 int ndind = (long)((nb[j + fso->fso_bufcnt - 1].offset -
321 sb->s_blocksize_bits) /
322 (EXT3_ADDR_PER_BLOCK(sb) * EXT3_ADDR_PER_BLOCK(sb));
323 nbitmaps += min(fso->fso_bufcnt, ndind > 0 ? ndind : 2);
325 /* leaf, indirect, tindirect blocks for first block */
326 nbitmaps += blockpp + 2;
328 j += fso->fso_bufcnt;
331 next_indir = nb[0].offset +
332 (EXT3_ADDR_PER_BLOCK(sb) << sb->s_blocksize_bits);
333 for (i = 1; i < niocount; i++) {
334 if (nb[i].offset >= next_indir) {
335 nbitmaps++; /* additional indirect */
336 next_indir = nb[i].offset +
337 (EXT3_ADDR_PER_BLOCK(sb)<<sb->s_blocksize_bits);
338 } else if (nb[i].offset != nb[i - 1].offset + sb->s_blocksize) {
339 nbitmaps++; /* additional indirect */
341 nbitmaps += blockpp; /* each leaf in different group? */
344 ngdblocks = nbitmaps;
345 if (nbitmaps > EXT3_SB(sb)->s_groups_count)
346 nbitmaps = EXT3_SB(sb)->s_groups_count;
347 if (ngdblocks > EXT3_SB(sb)->s_gdb_count)
348 ngdblocks = EXT3_SB(sb)->s_gdb_count;
350 needed += nbitmaps + ngdblocks;
352 /* last_rcvd update */
353 needed += FSFILT_DATA_TRANS_BLOCKS(sb);
355 #if defined(CONFIG_QUOTA)
356 /* We assume that there will be 1 bit set in s_dquot.flags for each
357 * quota file that is active. This is at least true for now.
359 needed += hweight32(sb_any_quota_enabled(sb)) *
360 FSFILT_SINGLEDATA_TRANS_BLOCKS(sb);
366 /* We have to start a huge journal transaction here to hold all of the
367 * metadata for the pages being written here. This is necessitated by
368 * the fact that we do lots of prepare_write operations before we do
369 * any of the matching commit_write operations, so even if we split
370 * up to use "smaller" transactions none of them could complete until
371 * all of them were opened. By having a single journal transaction,
372 * we eliminate duplicate reservations for common blocks like the
373 * superblock and group descriptors or bitmaps.
375 * We will start the transaction here, but each prepare_write will
376 * add a refcount to the transaction, and each commit_write will
377 * remove a refcount. The transaction will be closed when all of
378 * the pages have been written.
380 static void *fsfilt_ext3_brw_start(int objcount, struct fsfilt_objinfo *fso,
381 int niocount, struct niobuf_local *nb,
382 void *desc_private, int logs)
389 LASSERT(current->journal_info == desc_private);
390 journal = EXT3_SB(fso->fso_dentry->d_inode->i_sb)->s_journal;
391 needed = fsfilt_ext3_credits_needed(objcount, fso, niocount, nb);
393 /* The number of blocks we could _possibly_ dirty can very large.
394 * We reduce our request if it is absurd (and we couldn't get that
395 * many credits for a single handle anyways).
397 * At some point we have to limit the size of I/Os sent at one time,
398 * increase the size of the journal, or we have to calculate the
399 * actual journal requirements more carefully by checking all of
400 * the blocks instead of being maximally pessimistic. It remains to
401 * be seen if this is a real problem or not.
403 if (needed > journal->j_max_transaction_buffers) {
404 CERROR("want too many journal credits (%d) using %d instead\n",
405 needed, journal->j_max_transaction_buffers);
406 needed = journal->j_max_transaction_buffers;
409 LASSERTF(needed > 0, "can't start %d credit transaction\n", needed);
410 handle = fsfilt_ext3_journal_start(fso->fso_dentry->d_inode, needed);
411 if (IS_ERR(handle)) {
412 CERROR("can't get handle for %d credits: rc = %ld\n", needed,
415 LASSERT(handle->h_buffer_credits >= needed);
416 LASSERT(current->journal_info == handle);
422 static int fsfilt_ext3_extend(struct inode *inode, unsigned int nblocks,void *h)
424 handle_t *handle = h;
426 /* fsfilt_extend called with nblocks = 0 for testing in special cases */
428 handle->h_buffer_credits = 0;
429 CWARN("setting credits of handle %p to zero by request\n", h);
432 if (handle->h_buffer_credits > nblocks)
434 if (journal_extend(handle, nblocks) == 0)
437 ext3_mark_inode_dirty(handle, inode);
438 return journal_restart(handle, nblocks);
441 static int fsfilt_ext3_commit(struct inode *inode, void *h, int force_sync)
444 handle_t *handle = h;
446 LASSERT(current->journal_info == handle);
448 handle->h_sync = 1; /* recovery likes this */
450 rc = fsfilt_ext3_journal_stop(handle);
455 static int fsfilt_ext3_commit_async(struct inode *inode, void *h,
459 transaction_t *transaction;
460 handle_t *handle = h;
464 LASSERT(current->journal_info == handle);
466 transaction = handle->h_transaction;
467 journal = transaction->t_journal;
468 tid = transaction->t_tid;
469 /* we don't want to be blocked */
471 rc = fsfilt_ext3_journal_stop(handle);
473 CERROR("error while stopping transaction: %d\n", rc);
476 log_start_commit(journal, tid);
478 *wait_handle = (void *) tid;
479 CDEBUG(D_INODE, "commit async: %lu\n", (unsigned long) tid);
483 static int fsfilt_ext3_commit_wait(struct inode *inode, void *h)
485 journal_t *journal = EXT3_JOURNAL(inode);
486 tid_t tid = (tid_t)(long)h;
488 CDEBUG(D_INODE, "commit wait: %lu\n", (unsigned long) tid);
489 if (unlikely(is_journal_aborted(journal)))
492 log_wait_commit(EXT3_JOURNAL(inode), tid);
494 if (unlikely(is_journal_aborted(journal)))
499 static int fsfilt_ext3_setattr(struct dentry *dentry, void *handle,
500 struct iattr *iattr, int do_trunc)
502 struct inode *inode = dentry->d_inode;
505 /* Avoid marking the inode dirty on the superblock list unnecessarily.
506 * We are already writing the inode to disk as part of this
507 * transaction and want to avoid a lot of extra inode writeout
508 * later on. b=9828 */
509 if (iattr->ia_valid & ATTR_SIZE && !do_trunc) {
510 /* ATTR_SIZE would invoke truncate: clear it */
511 iattr->ia_valid &= ~ATTR_SIZE;
512 EXT3_I(inode)->i_disksize = iattr->ia_size;
513 i_size_write(inode, iattr->ia_size);
515 if (iattr->ia_valid & ATTR_UID)
516 inode->i_uid = iattr->ia_uid;
517 if (iattr->ia_valid & ATTR_GID)
518 inode->i_gid = iattr->ia_gid;
519 if (iattr->ia_valid & ATTR_ATIME)
520 inode->i_atime = iattr->ia_atime;
521 if (iattr->ia_valid & ATTR_MTIME)
522 inode->i_mtime = iattr->ia_mtime;
523 if (iattr->ia_valid & ATTR_CTIME)
524 inode->i_ctime = iattr->ia_ctime;
525 if (iattr->ia_valid & ATTR_MODE) {
526 inode->i_mode = iattr->ia_mode;
528 if (!in_group_p(inode->i_gid) &&
529 !cfs_capable(CFS_CAP_FSETID))
530 inode->i_mode &= ~S_ISGID;
533 inode->i_sb->s_op->dirty_inode(inode);
538 /* Don't allow setattr to change file type */
539 if (iattr->ia_valid & ATTR_MODE)
540 iattr->ia_mode = (inode->i_mode & S_IFMT) |
541 (iattr->ia_mode & ~S_IFMT);
543 /* We set these flags on the client, but have already checked perms
544 * so don't confuse inode_change_ok. */
545 iattr->ia_valid &= ~(ATTR_MTIME_SET | ATTR_ATIME_SET);
547 if (inode->i_op->setattr) {
548 rc = inode->i_op->setattr(dentry, iattr);
550 rc = inode_change_ok(inode, iattr);
552 rc = inode_setattr(inode, iattr);
559 static int fsfilt_ext3_iocontrol(struct inode * inode, struct file *file,
560 unsigned int cmd, unsigned long arg)
565 /* FIXME: Can't do this because of nested transaction deadlock */
566 if (cmd == EXT3_IOC_SETFLAGS && (*(int *)arg) & EXT3_JOURNAL_DATA_FL) {
567 CERROR("can't set data journal flag on file\n");
571 if (inode->i_fop->ioctl)
572 rc = inode->i_fop->ioctl(inode, file, cmd, arg);
579 static int fsfilt_ext3_set_md(struct inode *inode, void *handle,
580 void *lmm, int lmm_size, const char *name)
584 LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
586 rc = ext3_xattr_set_handle(handle, inode, EXT3_XATTR_INDEX_TRUSTED,
587 name, lmm, lmm_size, XATTR_NO_CTIME);
590 if (rc && rc != -EROFS)
591 CERROR("error adding MD data to inode %lu: rc = %d\n",
596 /* Must be called with i_mutex held */
597 static int fsfilt_ext3_get_md(struct inode *inode, void *lmm, int lmm_size,
602 LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
604 rc = ext3_xattr_get(inode, EXT3_XATTR_INDEX_TRUSTED,
605 name, lmm, lmm_size);
607 /* This gives us the MD size */
609 return (rc == -ENODATA) ? 0 : rc;
612 CDEBUG(D_INFO, "error getting EA %d/%s from inode %lu: rc %d\n",
613 EXT3_XATTR_INDEX_TRUSTED, name,
615 memset(lmm, 0, lmm_size);
616 return (rc == -ENODATA) ? 0 : rc;
622 static int fsfilt_ext3_send_bio(int rw, struct inode *inode, struct bio *bio)
628 static ssize_t fsfilt_ext3_readpage(struct file *file, char *buf, size_t count,
631 struct inode *inode = file->f_dentry->d_inode;
634 if (S_ISREG(inode->i_mode))
635 rc = file->f_op->read(file, buf, count, off);
637 const int blkbits = inode->i_sb->s_blocksize_bits;
638 const int blksize = inode->i_sb->s_blocksize;
640 CDEBUG(D_EXT2, "reading "LPSZ" at dir %lu+%llu\n",
641 count, inode->i_ino, *off);
643 struct buffer_head *bh;
646 if (*off < i_size_read(inode)) {
649 bh = ext3_bread(NULL, inode, *off >> blkbits,
652 CDEBUG(D_EXT2, "read %u@%llu\n", blksize, *off);
655 memcpy(buf, bh->b_data, blksize);
658 /* XXX in theory we should just fake
659 * this buffer and continue like ext3,
660 * especially if this is a partial read
662 CERROR("error read dir %lu+%llu: %d\n",
663 inode->i_ino, *off, err);
668 struct ext3_dir_entry_2 *fake = (void *)buf;
670 CDEBUG(D_EXT2, "fake %u@%llu\n", blksize, *off);
671 memset(fake, 0, sizeof(*fake));
672 fake->rec_len = cpu_to_le16(blksize);
684 static void fsfilt_ext3_cb_func(struct journal_callback *jcb, int error)
686 struct fsfilt_cb_data *fcb = (struct fsfilt_cb_data *)jcb;
688 fcb->cb_func(fcb->cb_obd, fcb->cb_last_rcvd, fcb->cb_data, error);
690 OBD_SLAB_FREE(fcb, fcb_cache, sizeof *fcb);
693 static int fsfilt_ext3_add_journal_cb(struct obd_device *obd, __u64 last_rcvd,
694 void *handle, fsfilt_cb_t cb_func,
697 struct fsfilt_cb_data *fcb;
699 OBD_SLAB_ALLOC(fcb, fcb_cache, CFS_ALLOC_IO, sizeof *fcb);
703 fcb->cb_func = cb_func;
705 fcb->cb_last_rcvd = last_rcvd;
706 fcb->cb_data = cb_data;
708 CDEBUG(D_EXT2, "set callback for last_rcvd: "LPD64"\n", last_rcvd);
709 journal_callback_set(handle, fsfilt_ext3_cb_func,
710 (struct journal_callback *)fcb);
716 * We need to hack the return value for the free inode counts because
717 * the current EA code requires one filesystem block per inode with EAs,
718 * so it is possible to run out of blocks before we run out of inodes.
720 * This can be removed when the ext3 EA code is fixed.
722 static int fsfilt_ext3_statfs(struct super_block *sb, struct obd_statfs *osfs)
727 memset(&sfs, 0, sizeof(sfs));
728 rc = ll_do_statfs(sb, &sfs);
729 if (!rc && sfs.f_bfree < sfs.f_ffree) {
730 sfs.f_files = (sfs.f_files - sfs.f_ffree) + sfs.f_bfree;
731 sfs.f_ffree = sfs.f_bfree;
734 statfs_pack(osfs, &sfs);
738 static int fsfilt_ext3_sync(struct super_block *sb)
740 return ext3_force_commit(sb);
743 #if defined(EXT3_MULTIBLOCK_ALLOCATOR) && (!defined(EXT3_EXT_CACHE_NO) || defined(EXT_CACHE_MARK))
744 #warning "kernel code has old extents/mballoc patch, disabling"
745 #undef EXT3_MULTIBLOCK_ALLOCATOR
747 #ifndef EXT3_EXTENTS_FL
748 #define EXT3_EXTENTS_FL 0x00080000 /* Inode uses extents */
751 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
752 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
753 #define ext3_up_truncate_sem(inode) up(&EXT3_I(inode)->truncate_sem);
754 #define ext3_down_truncate_sem(inode) down(&EXT3_I(inode)->truncate_sem);
756 #define ext3_up_truncate_sem(inode) mutex_unlock(&EXT3_I(inode)->truncate_mutex);
757 #define ext3_down_truncate_sem(inode) mutex_lock(&EXT3_I(inode)->truncate_mutex);
761 #define EXT_ASSERT(cond) BUG_ON(!(cond))
764 #ifdef EXT3_EXT_HAS_NO_TREE
765 /* for kernels 2.6.18 and later */
766 #define ext3_ext_base inode
767 #define ext3_ext_base2inode(inode) (inode)
768 #define EXT_DEPTH(inode) ext_depth(inode)
769 #define EXT_GENERATION(inode) ext_generation(inode)
770 #define fsfilt_ext3_ext_walk_space(inode, block, num, cb, cbdata) \
771 ext3_ext_walk_space(inode, block, num, cb, cbdata);
773 #define ext3_ext_base ext3_extents_tree
774 #define ext3_ext_base2inode(tree) (tree->inode)
775 #define fsfilt_ext3_ext_walk_space(tree, block, num, cb, cbdata) \
776 ext3_ext_walk_space(tree, block, num, cb);
779 #include <linux/lustre_version.h>
780 #if EXT3_EXT_MAGIC == 0xf301
781 #define ee_start e_start
782 #define ee_block e_block
785 #ifndef EXT3_BB_MAX_BLOCKS
786 #define ext3_mb_new_blocks(handle, inode, goal, count, aflags, err) \
787 ext3_new_blocks(handle, inode, count, goal, err)
791 unsigned long *blocks;
799 static int ext3_ext_find_goal(struct inode *inode, struct ext3_ext_path *path,
800 unsigned long block, int *aflags)
802 struct ext3_inode_info *ei = EXT3_I(inode);
803 unsigned long bg_start;
804 unsigned long colour;
808 struct ext3_extent *ex;
809 depth = path->p_depth;
811 /* try to predict block placement */
812 if ((ex = path[depth].p_ext)) {
814 /* This prefers to eat into a contiguous extent
815 * rather than find an extent that the whole
816 * request will fit into. This can fragment data
817 * block allocation and prevents our lovely 1M I/Os
818 * from reaching the disk intact. */
819 if (ex->ee_block + ex->ee_len == block)
822 return ex->ee_start + (block - ex->ee_block);
825 /* it looks index is empty
826 * try to find starting from index itself */
827 if (path[depth].p_bh)
828 return path[depth].p_bh->b_blocknr;
831 /* OK. use inode's group */
832 bg_start = (ei->i_block_group * EXT3_BLOCKS_PER_GROUP(inode->i_sb)) +
833 le32_to_cpu(EXT3_SB(inode->i_sb)->s_es->s_first_data_block);
834 colour = (current->pid % 16) *
835 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
836 return bg_start + colour + block;
839 #define ll_unmap_underlying_metadata(sb, blocknr) \
840 unmap_underlying_metadata((sb)->s_bdev, blocknr)
842 #ifndef EXT3_MB_HINT_GROUP_ALLOC
843 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
844 struct ext3_ext_path *path, unsigned long block,
845 unsigned long *count, int *err)
847 unsigned long pblock, goal;
849 struct inode *inode = ext3_ext_base2inode(base);
851 goal = ext3_ext_find_goal(inode, path, block, &aflags);
852 aflags |= 2; /* block have been already reserved */
853 pblock = ext3_mb_new_blocks(handle, inode, goal, count, aflags, err);
858 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
859 struct ext3_ext_path *path, unsigned long block,
860 unsigned long *count, int *err)
862 struct inode *inode = ext3_ext_base2inode(base);
863 struct ext3_allocation_request ar;
864 unsigned long pblock;
867 /* find neighbour allocated blocks */
869 *err = ext3_ext_search_left(base, path, &ar.lleft, &ar.pleft);
873 *err = ext3_ext_search_right(base, path, &ar.lright, &ar.pright);
877 /* allocate new block */
878 ar.goal = ext3_ext_find_goal(inode, path, block, &aflags);
882 ar.flags = EXT3_MB_HINT_DATA;
883 pblock = ext3_mb_new_blocks(handle, &ar, err);
889 #ifdef EXT3_EXT_HAS_NO_TREE
890 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
891 struct ext3_ext_path *path,
892 struct ext3_ext_cache *cex,
893 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
894 struct ext3_extent *ex,
898 struct bpointers *bp = cbdata;
900 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
901 struct ext3_ext_path *path,
902 struct ext3_ext_cache *cex
903 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
904 , struct ext3_extent *ex
908 struct bpointers *bp = base->private;
910 struct inode *inode = ext3_ext_base2inode(base);
911 struct ext3_extent nex;
912 unsigned long pblock;
919 EXT_ASSERT(i == path->p_depth);
920 EXT_ASSERT(path[i].p_hdr);
922 if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
927 if (bp->create == 0) {
929 if (cex->ec_block < bp->start)
930 i = bp->start - cex->ec_block;
931 if (i >= cex->ec_len)
932 CERROR("nothing to do?! i = %d, e_num = %u\n",
934 for (; i < cex->ec_len && bp->num; i++) {
946 tgen = EXT_GENERATION(base);
947 count = ext3_ext_calc_credits_for_insert(base, path);
948 ext3_up_truncate_sem(inode);
950 handle = fsfilt_ext3_journal_start(inode, count+EXT3_ALLOC_NEEDED+1);
951 if (IS_ERR(handle)) {
952 ext3_down_truncate_sem(inode);
953 return PTR_ERR(handle);
956 ext3_down_truncate_sem(inode);
957 if (tgen != EXT_GENERATION(base)) {
958 /* the tree has changed. so path can be invalid at moment */
959 fsfilt_ext3_journal_stop(handle);
964 pblock = new_blocks(handle, base, path, cex->ec_block, &count, &err);
967 EXT_ASSERT(count <= cex->ec_len);
969 /* insert new extent */
970 nex.ee_block = cex->ec_block;
971 nex.ee_start = pblock;
973 err = ext3_ext_insert_extent(handle, base, path, &nex);
975 /* free data blocks we just allocated */
976 /* not a good idea to call discard here directly,
977 * but otherwise we'd need to call it every free() */
978 #ifdef EXT3_MB_HINT_GROUP_ALLOC
979 ext3_mb_discard_inode_preallocations(inode);
981 ext3_free_blocks(handle, inode, nex.ee_start, nex.ee_len, 0);
986 * Putting len of the actual extent we just inserted,
987 * we are asking ext3_ext_walk_space() to continue
988 * scaning after that block
990 cex->ec_len = nex.ee_len;
991 cex->ec_start = nex.ee_start;
992 BUG_ON(nex.ee_len == 0);
993 BUG_ON(nex.ee_block != cex->ec_block);
996 fsfilt_ext3_journal_stop(handle);
1001 CERROR("hmm. why do we find this extent?\n");
1002 CERROR("initial space: %lu:%u\n",
1003 bp->start, bp->init_num);
1004 CERROR("current extent: %u/%u/%u %d\n",
1005 cex->ec_block, cex->ec_len,
1006 cex->ec_start, cex->ec_type);
1009 if (cex->ec_block < bp->start)
1010 i = bp->start - cex->ec_block;
1011 if (i >= cex->ec_len)
1012 CERROR("nothing to do?! i = %d, e_num = %u\n",
1014 for (; i < cex->ec_len && bp->num; i++) {
1015 *(bp->blocks) = cex->ec_start + i;
1016 if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
1020 /* unmap any possible underlying metadata from
1021 * the block device mapping. bug 6998. */
1022 ll_unmap_underlying_metadata(inode->i_sb,
1034 int fsfilt_map_nblocks(struct inode *inode, unsigned long block,
1035 unsigned long num, unsigned long *blocks,
1036 int *created, int create)
1038 #ifdef EXT3_EXT_HAS_NO_TREE
1039 struct ext3_ext_base *base = inode;
1041 struct ext3_extents_tree tree;
1042 struct ext3_ext_base *base = &tree;
1044 struct bpointers bp;
1047 CDEBUG(D_OTHER, "blocks %lu-%lu requested for inode %u\n",
1048 block, block + num - 1, (unsigned) inode->i_ino);
1050 #ifndef EXT3_EXT_HAS_NO_TREE
1051 ext3_init_tree_desc(base, inode);
1055 bp.created = created;
1057 bp.init_num = bp.num = num;
1060 ext3_down_truncate_sem(inode);
1061 err = fsfilt_ext3_ext_walk_space(base, block, num,
1062 ext3_ext_new_extent_cb, &bp);
1063 ext3_ext_invalidate_cache(base);
1064 ext3_up_truncate_sem(inode);
1069 int fsfilt_ext3_map_ext_inode_pages(struct inode *inode, struct page **page,
1070 int pages, unsigned long *blocks,
1071 int *created, int create)
1073 int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1075 struct page *fp = NULL;
1078 CDEBUG(D_OTHER, "inode %lu: map %d pages from %lu\n",
1079 inode->i_ino, pages, (*page)->index);
1081 /* pages are sorted already. so, we just have to find
1082 * contig. space and process them properly */
1085 /* start new extent */
1090 } else if (fp->index + clen == (*page)->index) {
1091 /* continue the extent */
1098 /* process found extent */
1099 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1100 clen * blocks_per_page, blocks,
1105 /* look for next extent */
1107 blocks += blocks_per_page * clen;
1108 created += blocks_per_page * clen;
1112 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1113 clen * blocks_per_page, blocks,
1118 #endif /* EXT3_MULTIBLOCK_ALLOCATOR */
1120 extern int ext3_map_inode_page(struct inode *inode, struct page *page,
1121 unsigned long *blocks, int *created, int create);
1122 int fsfilt_ext3_map_bm_inode_pages(struct inode *inode, struct page **page,
1123 int pages, unsigned long *blocks,
1124 int *created, int create)
1126 int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1130 for (i = 0, cr = created, b = blocks; i < pages; i++, page++) {
1131 rc = ext3_map_inode_page(inode, *page, b, cr, create);
1133 CERROR("ino %lu, blk %lu cr %u create %d: rc %d\n",
1134 inode->i_ino, *b, *cr, create, rc);
1138 b += blocks_per_page;
1139 cr += blocks_per_page;
1144 int fsfilt_ext3_map_inode_pages(struct inode *inode, struct page **page,
1145 int pages, unsigned long *blocks,
1146 int *created, int create,
1147 struct semaphore *optional_sem)
1150 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
1151 if (EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL) {
1152 rc = fsfilt_ext3_map_ext_inode_pages(inode, page, pages,
1153 blocks, created, create);
1157 if (optional_sem != NULL)
1159 rc = fsfilt_ext3_map_bm_inode_pages(inode, page, pages, blocks,
1161 if (optional_sem != NULL)
1167 int fsfilt_ext3_read(struct inode *inode, void *buf, int size, loff_t *offs)
1169 unsigned long block;
1170 struct buffer_head *bh;
1171 int err, blocksize, csize, boffs, osize = size;
1173 /* prevent reading after eof */
1175 if (i_size_read(inode) < *offs + size) {
1176 size = i_size_read(inode) - *offs;
1179 CERROR("size %llu is too short for read %u@%llu\n",
1180 i_size_read(inode), size, *offs);
1182 } else if (size == 0) {
1189 blocksize = 1 << inode->i_blkbits;
1192 block = *offs >> inode->i_blkbits;
1193 boffs = *offs & (blocksize - 1);
1194 csize = min(blocksize - boffs, size);
1195 bh = ext3_bread(NULL, inode, block, 0, &err);
1197 CERROR("can't read block: %d\n", err);
1201 memcpy(buf, bh->b_data + boffs, csize);
1210 EXPORT_SYMBOL(fsfilt_ext3_read);
1212 static int fsfilt_ext3_read_record(struct file * file, void *buf,
1213 int size, loff_t *offs)
1216 rc = fsfilt_ext3_read(file->f_dentry->d_inode, buf, size, offs);
1222 int fsfilt_ext3_write_handle(struct inode *inode, void *buf, int bufsize,
1223 loff_t *offs, handle_t *handle)
1225 struct buffer_head *bh = NULL;
1226 loff_t old_size = i_size_read(inode), offset = *offs;
1227 loff_t new_size = i_size_read(inode);
1228 unsigned long block;
1229 int err = 0, blocksize = 1 << inode->i_blkbits, size, boffs;
1231 while (bufsize > 0) {
1235 block = offset >> inode->i_blkbits;
1236 boffs = offset & (blocksize - 1);
1237 size = min(blocksize - boffs, bufsize);
1238 bh = ext3_bread(handle, inode, block, 1, &err);
1240 CERROR("can't read/create block: %d\n", err);
1244 err = ext3_journal_get_write_access(handle, bh);
1246 CERROR("journal_get_write_access() returned error %d\n",
1250 LASSERT(bh->b_data + boffs + size <= bh->b_data + bh->b_size);
1251 memcpy(bh->b_data + boffs, buf, size);
1252 err = ext3_journal_dirty_metadata(handle, bh);
1254 CERROR("journal_dirty_metadata() returned error %d\n",
1258 if (offset + size > new_size)
1259 new_size = offset + size;
1267 /* correct in-core and on-disk sizes */
1268 if (new_size > i_size_read(inode)) {
1270 if (new_size > i_size_read(inode))
1271 i_size_write(inode, new_size);
1272 if (i_size_read(inode) > EXT3_I(inode)->i_disksize)
1273 EXT3_I(inode)->i_disksize = i_size_read(inode);
1274 if (i_size_read(inode) > old_size)
1275 mark_inode_dirty(inode);
1283 EXPORT_SYMBOL(fsfilt_ext3_write_handle);
1285 static int fsfilt_ext3_write_record(struct file *file, void *buf, int bufsize,
1286 loff_t *offs, int force_sync)
1288 struct inode *inode = file->f_dentry->d_inode;
1290 int err, block_count = 0, blocksize;
1292 /* Determine how many transaction credits are needed */
1293 blocksize = 1 << inode->i_blkbits;
1294 block_count = (*offs & (blocksize - 1)) + bufsize;
1295 block_count = (block_count + blocksize - 1) >> inode->i_blkbits;
1297 handle = fsfilt_ext3_journal_start(inode,
1298 block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2);
1299 if (IS_ERR(handle)) {
1300 CERROR("can't start transaction for %d blocks (%d bytes)\n",
1301 block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2, bufsize);
1302 return PTR_ERR(handle);
1305 err = fsfilt_ext3_write_handle(inode, buf, bufsize, offs, handle);
1307 if (!err && force_sync)
1308 handle->h_sync = 1; /* recovery likes this */
1310 fsfilt_ext3_journal_stop(handle);
1315 static int fsfilt_ext3_setup(struct super_block *sb)
1317 #if ((LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,6)) && \
1318 defined(HAVE_QUOTA_SUPPORT)) || defined(S_PDIROPS)
1319 struct ext3_sb_info *sbi = EXT3_SB(sb);
1321 sbi->dx_lock = fsfilt_ext3_dx_lock;
1322 sbi->dx_unlock = fsfilt_ext3_dx_unlock;
1326 CWARN("Enabling PDIROPS\n");
1327 set_opt(sbi->s_mount_opt, PDIROPS);
1328 sb->s_flags |= S_PDIROPS;
1330 if (!EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1331 CWARN("filesystem doesn't have dir_index feature enabled\n");
1332 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,6)) && defined(HAVE_QUOTA_SUPPORT)
1333 /* enable journaled quota support */
1334 /* kfreed in ext3_put_super() */
1335 sbi->s_qf_names[USRQUOTA] = kstrdup("lquota.user.reserved", GFP_KERNEL);
1336 if (!sbi->s_qf_names[USRQUOTA])
1338 sbi->s_qf_names[GRPQUOTA] = kstrdup("lquota.group.reserved", GFP_KERNEL);
1339 if (!sbi->s_qf_names[GRPQUOTA]) {
1340 kfree(sbi->s_qf_names[USRQUOTA]);
1341 sbi->s_qf_names[USRQUOTA] = NULL;
1344 sbi->s_jquota_fmt = QFMT_VFS_V0;
1345 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,13))
1346 set_opt(sbi->s_mount_opt, QUOTA);
1352 /* If fso is NULL, op is FSFILT operation, otherwise op is number of fso
1353 objects. Logs is number of logfiles to update */
1354 static int fsfilt_ext3_get_op_len(int op, struct fsfilt_objinfo *fso, int logs)
1358 case FSFILT_OP_CREATE:
1359 /* directory leaf, index & indirect & EA*/
1360 return 4 + 3 * logs;
1361 case FSFILT_OP_UNLINK:
1367 struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
1368 int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
1369 int addrpp = EXT3_ADDR_PER_BLOCK(sb) * blockpp;
1370 for (i = 0; i < op; i++, fso++) {
1371 int nblocks = fso->fso_bufcnt * blockpp;
1372 int ndindirect = min(nblocks, addrpp + 1);
1373 int nindir = nblocks + ndindirect + 1;
1377 return needed + 3 * logs;
1383 #ifdef HAVE_QUOTA_SUPPORT
1384 #define DQINFO_COPY(out, in) \
1386 Q_COPY(out, in, dqi_bgrace); \
1387 Q_COPY(out, in, dqi_igrace); \
1388 Q_COPY(out, in, dqi_flags); \
1389 Q_COPY(out, in, dqi_valid); \
1392 #define DQBLK_COPY(out, in) \
1394 Q_COPY(out, in, dqb_bhardlimit); \
1395 Q_COPY(out, in, dqb_bsoftlimit); \
1396 Q_COPY(out, in, dqb_curspace); \
1397 Q_COPY(out, in, dqb_ihardlimit); \
1398 Q_COPY(out, in, dqb_isoftlimit); \
1399 Q_COPY(out, in, dqb_curinodes); \
1400 Q_COPY(out, in, dqb_btime); \
1401 Q_COPY(out, in, dqb_itime); \
1402 Q_COPY(out, in, dqb_valid); \
1405 static int fsfilt_ext3_quotactl(struct super_block *sb,
1406 struct obd_quotactl *oqc)
1408 int i, rc = 0, error = 0;
1409 struct quotactl_ops *qcop;
1410 struct if_dqinfo *info;
1411 struct if_dqblk *dqblk;
1417 OBD_ALLOC_PTR(info);
1420 OBD_ALLOC_PTR(dqblk);
1426 DQINFO_COPY(info, &oqc->qc_dqinfo);
1427 DQBLK_COPY(dqblk, &oqc->qc_dqblk);
1430 if (oqc->qc_cmd == Q_QUOTAON || oqc->qc_cmd == Q_QUOTAOFF) {
1431 for (i = 0; i < MAXQUOTAS; i++) {
1432 if (!Q_TYPESET(oqc, i))
1435 if (oqc->qc_cmd == Q_QUOTAON) {
1436 char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1438 LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1440 if (!qcop->quota_on)
1441 GOTO(out, rc = -ENOSYS);
1443 rc = qcop->quota_on(sb, i, QFMT_VFS_V0,
1445 } else if (oqc->qc_cmd == Q_QUOTAOFF) {
1446 if (!qcop->quota_off)
1447 GOTO(out, rc = -ENOSYS);
1448 rc = qcop->quota_off(sb, i);
1456 GOTO(out, rc ?: error);
1459 switch (oqc->qc_cmd) {
1462 if (!qcop->get_info)
1463 GOTO(out, rc = -ENOSYS);
1464 rc = qcop->get_info(sb, oqc->qc_type, info);
1468 if (!qcop->set_dqblk)
1469 GOTO(out, rc = -ENOSYS);
1470 rc = qcop->set_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1474 if (!qcop->get_dqblk)
1475 GOTO(out, rc = -ENOSYS);
1476 rc = qcop->get_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1478 dqblk->dqb_valid = QIF_LIMITS | QIF_USAGE;
1481 if (!sb->s_qcop->quota_sync)
1482 GOTO(out, rc = -ENOSYS);
1483 qcop->quota_sync(sb, oqc->qc_type);
1486 CDEBUG(D_WARNING, "invalidating operational quota files\n");
1487 for (i = 0; i < MAXQUOTAS; i++) {
1489 char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1491 LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1493 if (!Q_TYPESET(oqc, i))
1496 fp = filp_open(name[i], O_CREAT | O_TRUNC | O_RDWR, 0644);
1499 CERROR("error invalidating operational quota file"
1500 " %s (rc:%d)\n", name[i], rc);
1508 CERROR("unsupported quotactl command: %d\n", oqc->qc_cmd);
1512 DQINFO_COPY(&oqc->qc_dqinfo, info);
1513 DQBLK_COPY(&oqc->qc_dqblk, dqblk);
1516 OBD_FREE_PTR(dqblk);
1519 CDEBUG(D_QUOTA, "quotactl command %#x, id %u, type %u "
1521 oqc->qc_cmd, oqc->qc_id, oqc->qc_type, rc);
1526 struct hlist_node dqb_hash; /** quotacheck hash */
1527 struct list_head dqb_list; /** in list also */
1528 qid_t dqb_id; /** uid/gid */
1529 short dqb_type; /** USRQUOTA/GRPQUOTA */
1530 qsize_t dqb_bhardlimit; /** block hard limit */
1531 qsize_t dqb_bsoftlimit; /** block soft limit */
1532 qsize_t dqb_curspace; /** current space */
1533 qsize_t dqb_ihardlimit; /** inode hard limit */
1534 qsize_t dqb_isoftlimit; /** inode soft limit */
1535 qsize_t dqb_curinodes; /** current inodes */
1536 __u64 dqb_btime; /** block grace time */
1537 __u64 dqb_itime; /** inode grace time */
1538 __u32 dqb_valid; /** flag for above fields */
1541 static inline unsigned int chkquot_hash(qid_t id, int type)
1542 __attribute__((__const__));
1544 static inline unsigned int chkquot_hash(qid_t id, int type)
1546 return (id * (MAXQUOTAS - type)) % NR_DQHASH;
1549 static inline struct chk_dqblk *
1550 find_chkquot(struct hlist_head *head, qid_t id, int type)
1552 struct hlist_node *node;
1553 struct chk_dqblk *cdqb;
1555 hlist_for_each(node, head) {
1556 cdqb = hlist_entry(node, struct chk_dqblk, dqb_hash);
1557 if (cdqb->dqb_id == id && cdqb->dqb_type == type)
1564 static struct chk_dqblk *alloc_chkquot(qid_t id, int type)
1566 struct chk_dqblk *cdqb;
1568 OBD_ALLOC_PTR(cdqb);
1570 INIT_HLIST_NODE(&cdqb->dqb_hash);
1571 INIT_LIST_HEAD(&cdqb->dqb_list);
1573 cdqb->dqb_type = type;
1579 static struct chk_dqblk *
1580 cqget(struct super_block *sb, struct hlist_head *hash, struct list_head *list,
1581 qid_t id, int type, int first_check)
1583 struct hlist_head *head = hash + chkquot_hash(id, type);
1584 struct if_dqblk dqb;
1585 struct chk_dqblk *cdqb;
1588 cdqb = find_chkquot(head, id, type);
1592 cdqb = alloc_chkquot(id, type);
1597 rc = sb->s_qcop->get_dqblk(sb, type, id, &dqb);
1599 CERROR("get_dqblk of id %u, type %d failed: %d\n",
1602 DQBLK_COPY(cdqb, &dqb);
1603 cdqb->dqb_curspace = 0;
1604 cdqb->dqb_curinodes = 0;
1608 hlist_add_head(&cdqb->dqb_hash, head);
1609 list_add_tail(&cdqb->dqb_list, list);
1614 static inline int quota_onoff(struct super_block *sb, int cmd, int type, int qfmt)
1616 struct obd_quotactl *oqctl;
1619 OBD_ALLOC_PTR(oqctl);
1623 oqctl->qc_cmd = cmd;
1624 oqctl->qc_id = qfmt;
1625 oqctl->qc_type = type;
1626 rc = fsfilt_ext3_quotactl(sb, oqctl);
1628 OBD_FREE_PTR(oqctl);
1632 static inline int read_old_dqinfo(struct super_block *sb, int type,
1633 struct if_dqinfo *dqinfo)
1635 struct obd_quotactl *oqctl;
1639 OBD_ALLOC_PTR(oqctl);
1643 oqctl->qc_cmd = Q_GETINFO;
1644 oqctl->qc_type = type;
1645 rc = fsfilt_ext3_quotactl(sb, oqctl);
1647 ((struct obd_dqinfo *)dqinfo)[type] = oqctl->qc_dqinfo;
1649 OBD_FREE_PTR(oqctl);
1653 static inline struct ext3_group_desc *
1654 get_group_desc(struct super_block *sb, int group)
1656 unsigned long desc_block, desc;
1657 struct ext3_group_desc *gdp;
1659 desc_block = group / EXT3_DESC_PER_BLOCK(sb);
1660 desc = group % EXT3_DESC_PER_BLOCK(sb);
1661 gdp = (struct ext3_group_desc *)
1662 EXT3_SB(sb)->s_group_desc[desc_block]->b_data;
1667 static inline struct buffer_head *
1668 read_inode_bitmap(struct super_block *sb, unsigned long group)
1670 struct ext3_group_desc *desc;
1671 struct buffer_head *bh;
1673 desc = get_group_desc(sb, group);
1674 bh = sb_bread(sb, le32_to_cpu(desc->bg_inode_bitmap));
1679 static inline struct inode *ext3_iget_inuse(struct super_block *sb,
1680 struct buffer_head *bitmap_bh,
1681 int index, unsigned long ino)
1683 struct inode *inode = NULL;
1685 if (ext3_test_bit(index, bitmap_bh->b_data))
1686 inode = iget(sb, ino);
1692 struct hlist_head qckt_hash[NR_DQHASH]; /* quotacheck hash */
1693 struct list_head qckt_list; /* quotacheck list */
1694 int qckt_first_check[MAXQUOTAS]; /* 1 if no old quotafile */
1695 struct if_dqinfo qckt_dqinfo[MAXQUOTAS]; /* old dqinfo */
1698 static int add_inode_quota(struct inode *inode, struct qchk_ctxt *qctxt,
1699 struct obd_quotactl *oqc)
1701 struct chk_dqblk *cdqb[MAXQUOTAS] = { NULL, };
1703 qid_t qid[MAXQUOTAS];
1709 qid[USRQUOTA] = inode->i_uid;
1710 qid[GRPQUOTA] = inode->i_gid;
1712 if (S_ISDIR(inode->i_mode) ||
1713 S_ISREG(inode->i_mode) ||
1714 S_ISLNK(inode->i_mode))
1715 size = inode_get_bytes(inode);
1717 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1718 if (!Q_TYPESET(oqc, cnt))
1721 cdqb[cnt] = cqget(inode->i_sb, qctxt->qckt_hash,
1722 &qctxt->qckt_list, qid[cnt], cnt,
1723 qctxt->qckt_first_check[cnt]);
1729 cdqb[cnt]->dqb_curspace += size;
1730 cdqb[cnt]->dqb_curinodes++;
1734 for (i = 0; i < cnt; i++) {
1735 if (!Q_TYPESET(oqc, i))
1738 cdqb[i]->dqb_curspace -= size;
1739 cdqb[i]->dqb_curinodes--;
1746 /* write dqinfo struct in a new quota file */
1747 static int v3_write_dqinfo(struct file *f, int type, struct if_dqinfo *info)
1749 struct v2_disk_dqinfo dqinfo;
1750 __u32 blocks = V2_DQTREEOFF + 1;
1751 loff_t offset = V2_DQINFOOFF;
1754 dqinfo.dqi_bgrace = cpu_to_le32(info->dqi_bgrace);
1755 dqinfo.dqi_igrace = cpu_to_le32(info->dqi_igrace);
1756 dqinfo.dqi_flags = cpu_to_le32(info->dqi_flags & DQF_MASK &
1759 dqinfo.dqi_bgrace = cpu_to_le32(MAX_DQ_TIME);
1760 dqinfo.dqi_igrace = cpu_to_le32(MAX_IQ_TIME);
1761 dqinfo.dqi_flags = 0;
1764 dqinfo.dqi_blocks = cpu_to_le32(blocks);
1765 dqinfo.dqi_free_blk = 0;
1766 dqinfo.dqi_free_entry = 0;
1768 return cfs_user_write(f, (char *)&dqinfo, sizeof(dqinfo), &offset);
1771 static int v3_write_dqheader(struct file *f, int type)
1773 static const __u32 quota_magics[] = V2_INITQMAGICS;
1774 static const __u32 quota_versions[] = V2_INITQVERSIONS_R1;
1775 struct v2_disk_dqheader dqhead;
1778 CLASSERT(ARRAY_SIZE(quota_magics) == ARRAY_SIZE(quota_versions));
1779 LASSERT(0 <= type && type < ARRAY_SIZE(quota_magics));
1781 dqhead.dqh_magic = cpu_to_le32(quota_magics[type]);
1782 dqhead.dqh_version = cpu_to_le32(quota_versions[type]);
1784 return cfs_user_write(f, (char *)&dqhead, sizeof(dqhead), &offset);
1787 static int create_new_quota_files(struct qchk_ctxt *qctxt,
1788 struct obd_quotactl *oqc)
1793 for (i = 0; i < MAXQUOTAS; i++) {
1794 struct if_dqinfo *info = qctxt->qckt_first_check[i]?
1795 NULL : &qctxt->qckt_dqinfo[i];
1797 const char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1799 if (!Q_TYPESET(oqc, i))
1802 LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1804 file = filp_open(name[i], O_RDWR | O_CREAT | O_TRUNC, 0644);
1807 CERROR("can't create %s file: rc = %d\n",
1812 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
1813 CERROR("file %s is not regular", name[i]);
1814 filp_close(file, 0);
1815 GOTO(out, rc = -EINVAL);
1818 DQUOT_DROP(file->f_dentry->d_inode);
1820 rc = v3_write_dqheader(file, i);
1822 filp_close(file, 0);
1826 rc = v3_write_dqinfo(file, i, info);
1827 filp_close(file, 0);
1837 static int commit_chkquot(struct super_block *sb, struct qchk_ctxt *qctxt,
1838 struct chk_dqblk *cdqb)
1840 struct obd_quotactl *oqc;
1849 now = cfs_time_current_sec();
1851 if (cdqb->dqb_bsoftlimit &&
1852 toqb(cdqb->dqb_curspace) >= cdqb->dqb_bsoftlimit &&
1855 now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_bgrace;
1857 if (cdqb->dqb_isoftlimit &&
1858 cdqb->dqb_curinodes >= cdqb->dqb_isoftlimit &&
1861 now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_igrace;
1863 cdqb->dqb_valid = QIF_ALL;
1865 oqc->qc_cmd = Q_SETQUOTA;
1866 oqc->qc_type = cdqb->dqb_type;
1867 oqc->qc_id = cdqb->dqb_id;
1868 DQBLK_COPY(&oqc->qc_dqblk, cdqb);
1870 rc = fsfilt_ext3_quotactl(sb, oqc);
1875 static int prune_chkquots(struct super_block *sb,
1876 struct qchk_ctxt *qctxt, int error)
1878 struct chk_dqblk *cdqb, *tmp;
1881 list_for_each_entry_safe(cdqb, tmp, &qctxt->qckt_list, dqb_list) {
1883 rc = commit_chkquot(sb, qctxt, cdqb);
1887 hlist_del_init(&cdqb->dqb_hash);
1888 list_del(&cdqb->dqb_list);
1895 static int fsfilt_ext3_quotacheck(struct super_block *sb,
1896 struct obd_quotactl *oqc)
1898 struct ext3_sb_info *sbi = EXT3_SB(sb);
1900 struct qchk_ctxt *qctxt;
1901 struct buffer_head *bitmap_bh = NULL;
1903 struct inode *inode;
1907 /* turn on quota and read dqinfo if existed */
1908 OBD_ALLOC_PTR(qctxt);
1910 oqc->qc_stat = -ENOMEM;
1914 for (i = 0; i < NR_DQHASH; i++)
1915 INIT_HLIST_HEAD(&qctxt->qckt_hash[i]);
1916 INIT_LIST_HEAD(&qctxt->qckt_list);
1918 for (i = 0; i < MAXQUOTAS; i++) {
1919 if (!Q_TYPESET(oqc, i))
1922 rc = quota_onoff(sb, Q_QUOTAON, i, oqc->qc_id);
1923 if (!rc || rc == -EBUSY) {
1924 rc = read_old_dqinfo(sb, i, qctxt->qckt_dqinfo);
1927 } else if (rc == -ENOENT || rc == -EINVAL || rc == -EEXIST) {
1928 qctxt->qckt_first_check[i] = 1;
1934 /* check quota and update in hash */
1935 for (group = 0; group < sbi->s_groups_count; group++) {
1936 ino = group * sbi->s_inodes_per_group + 1;
1937 bitmap_bh = read_inode_bitmap(sb, group);
1939 CERROR("read_inode_bitmap group %d failed", group);
1940 GOTO(out, rc = -EIO);
1943 for (i = 0; i < sbi->s_inodes_per_group; i++, ino++) {
1944 if (ino < sbi->s_first_ino)
1947 inode = ext3_iget_inuse(sb, bitmap_bh, i, ino);
1948 rc = add_inode_quota(inode, qctxt, oqc);
1959 /* read old quota limits from old quota file. (only for the user
1960 * has limits but hasn't file) */
1961 #ifdef HAVE_QUOTA_SUPPORT
1962 for (i = 0; i < MAXQUOTAS; i++) {
1963 struct list_head id_list;
1964 struct dquot_id *dqid, *tmp;
1966 if (!Q_TYPESET(oqc, i))
1969 if (qctxt->qckt_first_check[i])
1973 LASSERT(sb_dqopt(sb)->files[i] != NULL);
1974 INIT_LIST_HEAD(&id_list);
1975 #ifndef KERNEL_SUPPORTS_QUOTA_READ
1976 rc = lustre_get_qids(sb_dqopt(sb)->files[i], NULL, i, &id_list);
1978 rc = lustre_get_qids(NULL, sb_dqopt(sb)->files[i], i, &id_list);
1981 CERROR("read old limits failed. (rc:%d)\n", rc);
1983 list_for_each_entry_safe(dqid, tmp, &id_list, di_link) {
1984 list_del_init(&dqid->di_link);
1987 cqget(sb, qctxt->qckt_hash, &qctxt->qckt_list,
1989 qctxt->qckt_first_check[i]);
1994 /* turn off quota cause we are to dump chk_dqblk to files */
1995 quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type, oqc->qc_id);
1997 rc = create_new_quota_files(qctxt, oqc);
2001 /* we use vfs functions to set dqblk, so turn quota on */
2002 rc = quota_onoff(sb, Q_QUOTAON, oqc->qc_type, oqc->qc_id);
2004 /* dump and free chk_dqblk */
2005 rc = prune_chkquots(sb, qctxt, rc);
2006 OBD_FREE_PTR(qctxt);
2008 /* turn off quota, `lfs quotacheck` will turn on when all
2009 * nodes quotacheck finish. */
2010 quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type, oqc->qc_id);
2014 CERROR("quotacheck failed: rc = %d\n", rc);
2019 static int fsfilt_ext3_quotainfo(struct lustre_quota_info *lqi, int type,
2025 if (lqi->qi_files[type] == NULL) {
2026 CERROR("operate qinfo before it's enabled!\n");
2032 rc = lustre_check_quota_file(lqi, type);
2035 rc = lustre_read_quota_info(lqi, type);
2038 rc = lustre_write_quota_info(lqi, type);
2040 case QFILE_INIT_INFO:
2041 rc = lustre_init_quota_info(lqi, type);
2045 CERROR("quota CONVERT command is not supported\n");
2049 CERROR("Unsupported admin quota file cmd %d\n"
2050 "Are lquota.ko and fsfilt_ldiskfs.ko modules in sync?\n",
2057 static int fsfilt_ext3_qids(struct file *file, struct inode *inode, int type,
2058 struct list_head *list)
2060 return lustre_get_qids(file, inode, type, list);
2063 static int fsfilt_ext3_dquot(struct lustre_dquot *dquot, int cmd)
2068 if (dquot->dq_info->qi_files[dquot->dq_type] == NULL) {
2069 CERROR("operate dquot before it's enabled!\n");
2074 case QFILE_RD_DQUOT:
2075 rc = lustre_read_dquot(dquot);
2077 case QFILE_WR_DQUOT:
2078 if (dquot->dq_dqb.dqb_ihardlimit ||
2079 dquot->dq_dqb.dqb_isoftlimit ||
2080 dquot->dq_dqb.dqb_bhardlimit ||
2081 dquot->dq_dqb.dqb_bsoftlimit)
2082 clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2084 set_bit(DQ_FAKE_B, &dquot->dq_flags);
2086 rc = lustre_commit_dquot(dquot);
2091 CERROR("Unsupported admin quota file cmd %d\n", cmd);
2098 static int fsfilt_ext3_get_mblk(struct super_block *sb, int *count,
2099 struct inode *inode, int frags)
2101 #ifdef EXT3_EXT_HAS_NO_TREE
2102 struct ext3_ext_base *base = inode;
2104 struct ext3_extents_tree tree;
2105 struct ext3_ext_base *base = &tree;
2107 ext3_init_tree_desc(base, inode);
2109 /* for an ost_write request, it needs <#fragments> * <tree depth + 1>
2110 * metablocks at maxium b=16542 */
2111 *count = frags * (EXT_DEPTH(base) + 1) * EXT3_BLOCK_SIZE(sb);
2117 lvfs_sbdev_type fsfilt_ext3_journal_sbdev(struct super_block *sb)
2119 return (EXT3_SB(sb)->journal_bdev);
2121 EXPORT_SYMBOL(fsfilt_ext3_journal_sbdev);
2123 static struct fsfilt_operations fsfilt_ext3_ops = {
2125 .fs_owner = THIS_MODULE,
2126 .fs_getlabel = fsfilt_ext3_get_label,
2127 .fs_setlabel = fsfilt_ext3_set_label,
2128 .fs_uuid = fsfilt_ext3_uuid,
2129 .fs_start = fsfilt_ext3_start,
2130 .fs_brw_start = fsfilt_ext3_brw_start,
2131 .fs_extend = fsfilt_ext3_extend,
2132 .fs_commit = fsfilt_ext3_commit,
2133 .fs_commit_async = fsfilt_ext3_commit_async,
2134 .fs_commit_wait = fsfilt_ext3_commit_wait,
2135 .fs_setattr = fsfilt_ext3_setattr,
2136 .fs_iocontrol = fsfilt_ext3_iocontrol,
2137 .fs_set_md = fsfilt_ext3_set_md,
2138 .fs_get_md = fsfilt_ext3_get_md,
2139 .fs_readpage = fsfilt_ext3_readpage,
2140 .fs_add_journal_cb = fsfilt_ext3_add_journal_cb,
2141 .fs_statfs = fsfilt_ext3_statfs,
2142 .fs_sync = fsfilt_ext3_sync,
2143 .fs_map_inode_pages = fsfilt_ext3_map_inode_pages,
2144 .fs_write_record = fsfilt_ext3_write_record,
2145 .fs_read_record = fsfilt_ext3_read_record,
2146 .fs_setup = fsfilt_ext3_setup,
2147 .fs_send_bio = fsfilt_ext3_send_bio,
2148 .fs_get_op_len = fsfilt_ext3_get_op_len,
2149 #ifdef HAVE_DISK_INODE_VERSION
2150 .fs_get_version = fsfilt_ext3_get_version,
2151 .fs_set_version = fsfilt_ext3_set_version,
2153 #ifdef HAVE_QUOTA_SUPPORT
2154 .fs_quotactl = fsfilt_ext3_quotactl,
2155 .fs_quotacheck = fsfilt_ext3_quotacheck,
2156 .fs_quotainfo = fsfilt_ext3_quotainfo,
2157 .fs_qids = fsfilt_ext3_qids,
2158 .fs_dquot = fsfilt_ext3_dquot,
2159 .fs_get_mblk = fsfilt_ext3_get_mblk,
2161 .fs_journal_sbdev = fsfilt_ext3_journal_sbdev,
2164 static int __init fsfilt_ext3_init(void)
2168 fcb_cache = cfs_mem_cache_create("fsfilt_ext3_fcb",
2169 sizeof(struct fsfilt_cb_data), 0, 0);
2171 CERROR("error allocating fsfilt journal callback cache\n");
2172 GOTO(out, rc = -ENOMEM);
2175 rc = fsfilt_register_ops(&fsfilt_ext3_ops);
2178 int err = cfs_mem_cache_destroy(fcb_cache);
2179 LASSERTF(err == 0, "error destroying new cache: rc %d\n", err);
2185 static void __exit fsfilt_ext3_exit(void)
2189 fsfilt_unregister_ops(&fsfilt_ext3_ops);
2190 rc = cfs_mem_cache_destroy(fcb_cache);
2191 LASSERTF(rc == 0, "couldn't destroy fcb_cache slab\n");
2194 module_init(fsfilt_ext3_init);
2195 module_exit(fsfilt_ext3_exit);
2197 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2198 MODULE_DESCRIPTION("Lustre ext3 Filesystem Helper v0.1");
2199 MODULE_LICENSE("GPL");