1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * lustre/lib/fsfilt_ext3.c
5 * Lustre filesystem abstraction routines
7 * Copyright (C) 2002, 2003 Cluster File Systems, Inc.
8 * Author: Andreas Dilger <adilger@clusterfs.com>
10 * This file is part of Lustre, http://www.lustre.org.
12 * Lustre is free software; you can redistribute it and/or
13 * modify it under the terms of version 2 of the GNU General Public
14 * License as published by the Free Software Foundation.
16 * Lustre is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
19 * GNU General Public License for more details.
21 * You should have received a copy of the GNU General Public License
22 * along with Lustre; if not, write to the Free Software
23 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
26 #define DEBUG_SUBSYSTEM S_FILTER
28 #include <linux/init.h>
29 #include <linux/module.h>
31 #include <linux/jbd.h>
32 #include <linux/slab.h>
33 #include <linux/pagemap.h>
34 #include <linux/quotaops.h>
35 #include <linux/ext3_fs.h>
36 #include <linux/ext3_jbd.h>
37 #include <linux/version.h>
38 #include <linux/bitops.h>
39 #include <linux/quota.h>
40 #include <linux/quotaio_v1.h>
41 #include <linux/quotaio_v2.h>
42 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
43 #include <linux/ext3_xattr.h>
45 #include <ext3/xattr.h>
48 #include <libcfs/kp30.h>
49 #include <lustre_fsfilt.h>
51 #include <obd_class.h>
52 #include <lustre_quota.h>
53 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
54 #include <linux/iobuf.h>
56 #include <linux/lustre_compat25.h>
57 #include <linux/lprocfs_status.h>
59 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
60 #include <linux/ext3_extents.h>
63 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,15)
64 #define FSFILT_DATA_TRANS_BLOCKS(sb) EXT3_DATA_TRANS_BLOCKS
65 #define FSFILT_DELETE_TRANS_BLOCKS(sb) EXT3_DELETE_TRANS_BLOCKS
67 #define FSFILT_DATA_TRANS_BLOCKS(sb) EXT3_DATA_TRANS_BLOCKS(sb)
68 #define FSFILT_DELETE_TRANS_BLOCKS(sb) EXT3_DELETE_TRANS_BLOCKS(sb)
71 static kmem_cache_t *fcb_cache;
73 struct fsfilt_cb_data {
74 struct journal_callback cb_jcb; /* jbd private data - MUST BE FIRST */
75 fsfilt_cb_t cb_func; /* MDS/OBD completion function */
76 struct obd_device *cb_obd; /* MDS/OBD completion device */
77 __u64 cb_last_rcvd; /* MDS/OST last committed operation */
78 void *cb_data; /* MDS/OST completion function data */
81 #ifndef EXT3_XATTR_INDEX_TRUSTED /* temporary until we hit l28 kernel */
82 #define EXT3_XATTR_INDEX_TRUSTED 4
85 static char *fsfilt_ext3_get_label(struct super_block *sb)
87 return EXT3_SB(sb)->s_es->s_volume_name;
90 static int fsfilt_ext3_set_label(struct super_block *sb, char *label)
92 /* see e.g. fsfilt_ext3_write_record() */
97 journal = EXT3_SB(sb)->s_journal;
99 handle = journal_start(journal, 1);
101 if (IS_ERR(handle)) {
102 CERROR("can't start transaction\n");
103 return(PTR_ERR(handle));
106 err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
110 memcpy(EXT3_SB(sb)->s_es->s_volume_name, label,
111 sizeof(EXT3_SB(sb)->s_es->s_volume_name));
113 err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
117 journal_stop(handle);
123 static char *fsfilt_ext3_uuid(struct super_block *sb)
125 return EXT3_SB(sb)->s_es->s_uuid;
129 * We don't currently need any additional blocks for rmdir and
130 * unlink transactions because we are storing the OST oa_id inside
131 * the inode (which we will be changing anyways as part of this
134 static void *fsfilt_ext3_start(struct inode *inode, int op, void *desc_private,
137 /* For updates to the last received file */
138 int nblocks = EXT3_SINGLEDATA_TRANS_BLOCKS;
142 if (current->journal_info) {
143 CDEBUG(D_INODE, "increasing refcount on %p\n",
144 current->journal_info);
149 case FSFILT_OP_RMDIR:
150 case FSFILT_OP_UNLINK:
151 /* delete one file + create/update logs for each stripe */
152 nblocks += FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb);
153 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
154 EXT3_SINGLEDATA_TRANS_BLOCKS) * logs;
156 case FSFILT_OP_RENAME:
157 /* modify additional directory */
158 nblocks += EXT3_SINGLEDATA_TRANS_BLOCKS;
160 case FSFILT_OP_SYMLINK:
161 /* additional block + block bitmap + GDT for long symlink */
164 case FSFILT_OP_CREATE: {
165 #if defined(EXT3_EXTENTS_FL) && defined(EXT3_INDEX_FL)
168 if (!test_opt(inode->i_sb, EXTENTS)) {
170 } else if (((EXT3_I(inode)->i_flags &
171 cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL)) ==
172 cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL))) {
173 CWARN("extent-mapped directory found - contact "
174 "CFS: support@clusterfs.com\n");
181 case FSFILT_OP_MKDIR:
182 case FSFILT_OP_MKNOD:
183 /* modify one inode + block bitmap + GDT */
187 /* modify parent directory */
188 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
189 FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
190 /* create/update logs for each stripe */
191 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
192 EXT3_SINGLEDATA_TRANS_BLOCKS) * logs;
194 case FSFILT_OP_SETATTR:
195 /* Setattr on inode */
197 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
198 FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
199 /* quota chown log for each stripe */
200 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
201 EXT3_SINGLEDATA_TRANS_BLOCKS) * logs;
203 case FSFILT_OP_CANCEL_UNLINK:
204 /* blocks for log header bitmap update OR
205 * blocks for catalog header bitmap update + unlink of logs */
206 nblocks = (LLOG_CHUNK_SIZE >> inode->i_blkbits) +
207 FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb) * logs;
210 /* delete 2 file(file + array id) + create 1 file (array id)
211 * create/update logs for each stripe */
212 nblocks += 2 * FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb);
214 /*create array log for head file*/
216 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
217 EXT3_SINGLEDATA_TRANS_BLOCKS);
218 /*update head file array */
219 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
220 FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
222 default: CERROR("unknown transaction start op %d\n", op);
226 LASSERT(current->journal_info == desc_private);
227 journal = EXT3_SB(inode->i_sb)->s_journal;
228 if (nblocks > journal->j_max_transaction_buffers) {
229 CWARN("too many credits %d for op %ux%u using %d instead\n",
230 nblocks, op, logs, journal->j_max_transaction_buffers);
231 nblocks = journal->j_max_transaction_buffers;
235 LASSERTF(nblocks > 0, "can't start %d credit transaction\n", nblocks);
237 handle = journal_start(EXT3_JOURNAL(inode), nblocks);
241 LASSERT(current->journal_info == handle);
243 CERROR("error starting handle for op %u (%u credits): rc %ld\n",
244 op, nblocks, PTR_ERR(handle));
249 * Calculate the number of buffer credits needed to write multiple pages in
250 * a single ext3 transaction. No, this shouldn't be here, but as yet ext3
251 * doesn't have a nice API for calculating this sort of thing in advance.
253 * See comment above ext3_writepage_trans_blocks for details. We assume
254 * no data journaling is being done, but it does allow for all of the pages
255 * being non-contiguous. If we are guaranteed contiguous pages we could
256 * reduce the number of (d)indirect blocks a lot.
258 * With N blocks per page and P pages, for each inode we have at most:
260 * min(N*P, blocksize/4 + 1) dindirect blocks
263 * For the entire filesystem, we have at most:
264 * min(sum(nindir + P), ngroups) bitmap blocks (from the above)
265 * min(sum(nindir + P), gdblocks) group descriptor blocks (from the above)
266 * objcount inode blocks
268 * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quota files
270 * 1 EXT3_DATA_TRANS_BLOCKS for the last_rcvd update.
272 static int fsfilt_ext3_credits_needed(int objcount, struct fsfilt_objinfo *fso,
273 int niocount, struct niobuf_local *nb)
275 struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
277 const int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
278 int nbitmaps = 0, ngdblocks;
279 int needed = objcount + 1; /* inodes + superblock */
282 for (i = 0, j = 0; i < objcount; i++, fso++) {
283 /* two or more dindirect blocks in case we cross boundary */
284 int ndind = (long)((nb[j + fso->fso_bufcnt - 1].offset -
286 sb->s_blocksize_bits) /
287 (EXT3_ADDR_PER_BLOCK(sb) * EXT3_ADDR_PER_BLOCK(sb));
288 nbitmaps += min(fso->fso_bufcnt, ndind > 0 ? ndind : 2);
290 /* leaf, indirect, tindirect blocks for first block */
291 nbitmaps += blockpp + 2;
293 j += fso->fso_bufcnt;
296 next_indir = nb[0].offset +
297 (EXT3_ADDR_PER_BLOCK(sb) << sb->s_blocksize_bits);
298 for (i = 1; i < niocount; i++) {
299 if (nb[i].offset >= next_indir) {
300 nbitmaps++; /* additional indirect */
301 next_indir = nb[i].offset +
302 (EXT3_ADDR_PER_BLOCK(sb)<<sb->s_blocksize_bits);
303 } else if (nb[i].offset != nb[i - 1].offset + sb->s_blocksize) {
304 nbitmaps++; /* additional indirect */
306 nbitmaps += blockpp; /* each leaf in different group? */
309 ngdblocks = nbitmaps;
310 if (nbitmaps > EXT3_SB(sb)->s_groups_count)
311 nbitmaps = EXT3_SB(sb)->s_groups_count;
312 if (ngdblocks > EXT3_SB(sb)->s_gdb_count)
313 ngdblocks = EXT3_SB(sb)->s_gdb_count;
315 needed += nbitmaps + ngdblocks;
317 /* last_rcvd update */
318 needed += FSFILT_DATA_TRANS_BLOCKS(sb);
320 #if defined(CONFIG_QUOTA)
321 /* We assume that there will be 1 bit set in s_dquot.flags for each
322 * quota file that is active. This is at least true for now.
324 needed += hweight32(sb_any_quota_enabled(sb)) *
325 EXT3_SINGLEDATA_TRANS_BLOCKS;
331 /* We have to start a huge journal transaction here to hold all of the
332 * metadata for the pages being written here. This is necessitated by
333 * the fact that we do lots of prepare_write operations before we do
334 * any of the matching commit_write operations, so even if we split
335 * up to use "smaller" transactions none of them could complete until
336 * all of them were opened. By having a single journal transaction,
337 * we eliminate duplicate reservations for common blocks like the
338 * superblock and group descriptors or bitmaps.
340 * We will start the transaction here, but each prepare_write will
341 * add a refcount to the transaction, and each commit_write will
342 * remove a refcount. The transaction will be closed when all of
343 * the pages have been written.
345 static void *fsfilt_ext3_brw_start(int objcount, struct fsfilt_objinfo *fso,
346 int niocount, struct niobuf_local *nb,
347 void *desc_private, int logs)
354 LASSERT(current->journal_info == desc_private);
355 journal = EXT3_SB(fso->fso_dentry->d_inode->i_sb)->s_journal;
356 needed = fsfilt_ext3_credits_needed(objcount, fso, niocount, nb);
358 /* The number of blocks we could _possibly_ dirty can very large.
359 * We reduce our request if it is absurd (and we couldn't get that
360 * many credits for a single handle anyways).
362 * At some point we have to limit the size of I/Os sent at one time,
363 * increase the size of the journal, or we have to calculate the
364 * actual journal requirements more carefully by checking all of
365 * the blocks instead of being maximally pessimistic. It remains to
366 * be seen if this is a real problem or not.
368 if (needed > journal->j_max_transaction_buffers) {
369 CERROR("want too many journal credits (%d) using %d instead\n",
370 needed, journal->j_max_transaction_buffers);
371 needed = journal->j_max_transaction_buffers;
374 LASSERTF(needed > 0, "can't start %d credit transaction\n", needed);
376 handle = journal_start(journal, needed);
378 if (IS_ERR(handle)) {
379 CERROR("can't get handle for %d credits: rc = %ld\n", needed,
382 LASSERT(handle->h_buffer_credits >= needed);
383 LASSERT(current->journal_info == handle);
389 static int fsfilt_ext3_extend(struct inode *inode, unsigned int nblocks,void *h)
391 handle_t *handle = h;
393 /* fsfilt_extend called with nblocks = 0 for testing in special cases */
395 handle->h_buffer_credits = 0;
396 CWARN("setting credits of handle %p to zero by request\n", h);
399 if (handle->h_buffer_credits > nblocks)
401 if (journal_extend(handle, nblocks) == 0)
404 ext3_mark_inode_dirty(handle, inode);
405 return journal_restart(handle, nblocks);
408 static int fsfilt_ext3_commit(struct inode *inode, void *h, int force_sync)
411 handle_t *handle = h;
413 LASSERT(current->journal_info == handle);
415 handle->h_sync = 1; /* recovery likes this */
418 rc = journal_stop(handle);
424 static int fsfilt_ext3_commit_async(struct inode *inode, void *h,
428 transaction_t *transaction;
429 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0)
432 handle_t *handle = h;
436 LASSERT(current->journal_info == handle);
439 transaction = handle->h_transaction;
440 journal = transaction->t_journal;
441 tid = transaction->t_tid;
442 /* we don't want to be blocked */
444 rc = journal_stop(handle);
446 CERROR("error while stopping transaction: %d\n", rc);
450 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0)
451 rtid = log_start_commit(journal, transaction);
453 CERROR("strange race: %lu != %lu\n",
454 (unsigned long) tid, (unsigned long) rtid);
456 log_start_commit(journal, tid);
460 *wait_handle = (void *) tid;
461 CDEBUG(D_INODE, "commit async: %lu\n", (unsigned long) tid);
465 static int fsfilt_ext3_commit_wait(struct inode *inode, void *h)
467 journal_t *journal = EXT3_JOURNAL(inode);
468 tid_t tid = (tid_t)(long)h;
470 CDEBUG(D_INODE, "commit wait: %lu\n", (unsigned long) tid);
471 if (unlikely(is_journal_aborted(journal)))
474 log_wait_commit(EXT3_JOURNAL(inode), tid);
476 if (unlikely(is_journal_aborted(journal)))
481 static int fsfilt_ext3_setattr(struct dentry *dentry, void *handle,
482 struct iattr *iattr, int do_trunc)
484 struct inode *inode = dentry->d_inode;
489 /* Avoid marking the inode dirty on the superblock list unnecessarily.
490 * We are already writing the inode to disk as part of this
491 * transaction and want to avoid a lot of extra inode writeout
492 * later on. b=9828 */
493 if (iattr->ia_valid & ATTR_SIZE && !do_trunc) {
494 /* ATTR_SIZE would invoke truncate: clear it */
495 iattr->ia_valid &= ~ATTR_SIZE;
496 EXT3_I(inode)->i_disksize = inode->i_size = iattr->ia_size;
498 if (iattr->ia_valid & ATTR_UID)
499 inode->i_uid = iattr->ia_uid;
500 if (iattr->ia_valid & ATTR_GID)
501 inode->i_gid = iattr->ia_gid;
502 if (iattr->ia_valid & ATTR_ATIME)
503 inode->i_atime = iattr->ia_atime;
504 if (iattr->ia_valid & ATTR_MTIME)
505 inode->i_mtime = iattr->ia_mtime;
506 if (iattr->ia_valid & ATTR_CTIME)
507 inode->i_ctime = iattr->ia_ctime;
508 if (iattr->ia_valid & ATTR_MODE) {
509 inode->i_mode = iattr->ia_mode;
511 if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
512 inode->i_mode &= ~S_ISGID;
515 inode->i_sb->s_op->dirty_inode(inode);
520 /* Don't allow setattr to change file type */
521 if (iattr->ia_valid & ATTR_MODE)
522 iattr->ia_mode = (inode->i_mode & S_IFMT) |
523 (iattr->ia_mode & ~S_IFMT);
525 /* We set these flags on the client, but have already checked perms
526 * so don't confuse inode_change_ok. */
527 iattr->ia_valid &= ~(ATTR_MTIME_SET | ATTR_ATIME_SET);
529 if (inode->i_op->setattr) {
530 rc = inode->i_op->setattr(dentry, iattr);
532 rc = inode_change_ok(inode, iattr);
534 rc = inode_setattr(inode, iattr);
542 static int fsfilt_ext3_iocontrol(struct inode * inode, struct file *file,
543 unsigned int cmd, unsigned long arg)
548 /* FIXME: Can't do this because of nested transaction deadlock */
549 if (cmd == EXT3_IOC_SETFLAGS && (*(int *)arg) & EXT3_JOURNAL_DATA_FL) {
550 CERROR("can't set data journal flag on file\n");
554 if (inode->i_fop->ioctl)
555 rc = inode->i_fop->ioctl(inode, file, cmd, arg);
562 static int fsfilt_ext3_set_md(struct inode *inode, void *handle,
563 void *lmm, int lmm_size, const char *name)
567 LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
570 rc = ext3_xattr_set_handle(handle, inode, EXT3_XATTR_INDEX_TRUSTED,
571 name, lmm, lmm_size, 0);
576 CERROR("error adding MD data to inode %lu: rc = %d\n",
581 /* Must be called with i_mutex held */
582 static int fsfilt_ext3_get_md(struct inode *inode, void *lmm, int lmm_size,
587 LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
590 rc = ext3_xattr_get(inode, EXT3_XATTR_INDEX_TRUSTED,
591 name, lmm, lmm_size);
594 /* This gives us the MD size */
596 return (rc == -ENODATA) ? 0 : rc;
599 CDEBUG(D_INFO, "error getting EA %d/%s from inode %lu: rc %d\n",
600 EXT3_XATTR_INDEX_TRUSTED, name,
602 memset(lmm, 0, lmm_size);
603 return (rc == -ENODATA) ? 0 : rc;
609 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
610 static int fsfilt_ext3_send_bio(int rw, struct inode *inode, struct bio *bio)
616 static int fsfilt_ext3_send_bio(int rw, struct inode *inode, struct kiobuf *bio)
618 int rc, blk_per_page;
620 rc = brw_kiovec(rw, 1, &bio, inode->i_dev,
621 KIOBUF_GET_BLOCKS(bio), 1 << inode->i_blkbits);
623 * brw_kiovec() returns number of bytes actually written. If error
624 * occurred after something was written, error code is returned though
625 * kiobuf->errno. (See bug 6854.)
628 blk_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
630 if (rc != (1 << inode->i_blkbits) * bio->nr_pages * blk_per_page) {
631 CERROR("short write? expected %d, wrote %d (%d)\n",
632 (1 << inode->i_blkbits) * bio->nr_pages * blk_per_page,
635 if (bio->errno != 0) {
636 CERROR("IO error. Wrote %d of %d (%d)\n",
638 (1 << inode->i_blkbits) * bio->nr_pages * blk_per_page,
647 static ssize_t fsfilt_ext3_readpage(struct file *file, char *buf, size_t count,
650 struct inode *inode = file->f_dentry->d_inode;
653 if (S_ISREG(inode->i_mode))
654 rc = file->f_op->read(file, buf, count, off);
656 const int blkbits = inode->i_sb->s_blocksize_bits;
657 const int blksize = inode->i_sb->s_blocksize;
659 CDEBUG(D_EXT2, "reading "LPSZ" at dir %lu+%llu\n",
660 count, inode->i_ino, *off);
662 struct buffer_head *bh;
665 if (*off < inode->i_size) {
668 bh = ext3_bread(NULL, inode, *off >> blkbits,
671 CDEBUG(D_EXT2, "read %u@%llu\n", blksize, *off);
674 memcpy(buf, bh->b_data, blksize);
677 /* XXX in theory we should just fake
678 * this buffer and continue like ext3,
679 * especially if this is a partial read
681 CERROR("error read dir %lu+%llu: %d\n",
682 inode->i_ino, *off, err);
687 struct ext3_dir_entry_2 *fake = (void *)buf;
689 CDEBUG(D_EXT2, "fake %u@%llu\n", blksize, *off);
690 memset(fake, 0, sizeof(*fake));
691 fake->rec_len = cpu_to_le16(blksize);
703 static void fsfilt_ext3_cb_func(struct journal_callback *jcb, int error)
705 struct fsfilt_cb_data *fcb = (struct fsfilt_cb_data *)jcb;
707 fcb->cb_func(fcb->cb_obd, fcb->cb_last_rcvd, fcb->cb_data, error);
709 OBD_SLAB_FREE(fcb, fcb_cache, sizeof *fcb);
712 static int fsfilt_ext3_add_journal_cb(struct obd_device *obd, __u64 last_rcvd,
713 void *handle, fsfilt_cb_t cb_func,
716 struct fsfilt_cb_data *fcb;
718 OBD_SLAB_ALLOC(fcb, fcb_cache, GFP_NOFS, sizeof *fcb);
722 fcb->cb_func = cb_func;
724 fcb->cb_last_rcvd = last_rcvd;
725 fcb->cb_data = cb_data;
727 CDEBUG(D_EXT2, "set callback for last_rcvd: "LPD64"\n", last_rcvd);
729 journal_callback_set(handle, fsfilt_ext3_cb_func,
730 (struct journal_callback *)fcb);
737 * We need to hack the return value for the free inode counts because
738 * the current EA code requires one filesystem block per inode with EAs,
739 * so it is possible to run out of blocks before we run out of inodes.
741 * This can be removed when the ext3 EA code is fixed.
743 static int fsfilt_ext3_statfs(struct super_block *sb, struct obd_statfs *osfs)
748 memset(&sfs, 0, sizeof(sfs));
749 rc = ll_do_statfs(sb,&sfs);
751 if (!rc && sfs.f_bfree < sfs.f_ffree) {
752 sfs.f_files = (sfs.f_files - sfs.f_ffree) + sfs.f_bfree;
753 sfs.f_ffree = sfs.f_bfree;
756 statfs_pack(osfs, &sfs);
760 static int fsfilt_ext3_sync(struct super_block *sb)
762 return ext3_force_commit(sb);
765 #if defined(EXT3_MULTIBLOCK_ALLOCATOR) && (!defined(EXT3_EXT_CACHE_NO) || defined(EXT_CACHE_MARK))
766 #warning "kernel code has old extents/mballoc patch, disabling"
767 #undef EXT3_MULTIBLOCK_ALLOCATOR
769 #ifndef EXT3_EXTENTS_FL
770 #define EXT3_EXTENTS_FL 0x00080000 /* Inode uses extents */
773 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
774 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
775 #define ext3_up_truncate_sem(inode) up_write(&EXT3_I(inode)->truncate_sem);
776 #define ext3_down_truncate_sem(inode) down_write(&EXT3_I(inode)->truncate_sem);
777 #elif (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
778 #define ext3_up_truncate_sem(inode) up(&EXT3_I(inode)->truncate_sem);
779 #define ext3_down_truncate_sem(inode) down(&EXT3_I(inode)->truncate_sem);
781 #define ext3_up_truncate_sem(inode) mutex_unlock(&EXT3_I(inode)->truncate_mutex);
782 #define ext3_down_truncate_sem(inode) mutex_lock(&EXT3_I(inode)->truncate_mutex);
785 #include <linux/lustre_version.h>
786 #if EXT3_EXT_MAGIC == 0xf301
787 #define ee_start e_start
788 #define ee_block e_block
791 #ifndef EXT3_BB_MAX_BLOCKS
792 #define ext3_mb_new_blocks(handle, inode, goal, count, aflags, err) \
793 ext3_new_blocks(handle, inode, count, goal, err)
797 unsigned long *blocks;
805 static int ext3_ext_find_goal(struct inode *inode, struct ext3_ext_path *path,
806 unsigned long block, int *aflags)
808 struct ext3_inode_info *ei = EXT3_I(inode);
809 unsigned long bg_start;
810 unsigned long colour;
814 struct ext3_extent *ex;
815 depth = path->p_depth;
817 /* try to predict block placement */
818 if ((ex = path[depth].p_ext)) {
820 /* This prefers to eat into a contiguous extent
821 * rather than find an extent that the whole
822 * request will fit into. This can fragment data
823 * block allocation and prevents our lovely 1M I/Os
824 * from reaching the disk intact. */
825 if (ex->ee_block + ex->ee_len == block)
828 return ex->ee_start + (block - ex->ee_block);
831 /* it looks index is empty
832 * try to find starting from index itself */
833 if (path[depth].p_bh)
834 return path[depth].p_bh->b_blocknr;
837 /* OK. use inode's group */
838 bg_start = (ei->i_block_group * EXT3_BLOCKS_PER_GROUP(inode->i_sb)) +
839 le32_to_cpu(EXT3_SB(inode->i_sb)->s_es->s_first_data_block);
840 colour = (current->pid % 16) *
841 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
842 return bg_start + colour + block;
845 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
846 #include <linux/locks.h>
847 static void ll_unmap_underlying_metadata(struct super_block *sb,
848 unsigned long blocknr)
850 struct buffer_head *old_bh;
852 old_bh = get_hash_table(sb->s_dev, blocknr, sb->s_blocksize);
854 mark_buffer_clean(old_bh);
855 wait_on_buffer(old_bh);
856 clear_bit(BH_Req, &old_bh->b_state);
861 #define ll_unmap_underlying_metadata(sb, blocknr) \
862 unmap_underlying_metadata((sb)->s_bdev, blocknr)
865 static int ext3_ext_new_extent_cb(struct ext3_extents_tree *tree,
866 struct ext3_ext_path *path,
867 struct ext3_ext_cache *cex)
869 struct inode *inode = tree->inode;
870 struct bpointers *bp = tree->private;
871 struct ext3_extent nex;
872 int count, err, goal;
873 unsigned long pblock;
879 EXT_ASSERT(i == path->p_depth);
880 EXT_ASSERT(path[i].p_hdr);
882 if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
887 if (bp->create == 0) {
889 if (cex->ec_block < bp->start)
890 i = bp->start - cex->ec_block;
891 if (i >= cex->ec_len)
892 CERROR("nothing to do?! i = %d, e_num = %u\n",
894 for (; i < cex->ec_len && bp->num; i++) {
906 tgen = EXT_GENERATION(tree);
907 count = ext3_ext_calc_credits_for_insert(tree, path);
908 ext3_up_truncate_sem(inode);
911 handle = journal_start(EXT3_JOURNAL(inode), count+EXT3_ALLOC_NEEDED+1);
913 if (IS_ERR(handle)) {
914 ext3_down_truncate_sem(inode);
915 return PTR_ERR(handle);
918 ext3_down_truncate_sem(inode);
919 if (tgen != EXT_GENERATION(tree)) {
920 /* the tree has changed. so path can be invalid at moment */
922 journal_stop(handle);
928 goal = ext3_ext_find_goal(inode, path, cex->ec_block, &aflags);
929 aflags |= 2; /* block have been already reserved */
931 pblock = ext3_mb_new_blocks(handle, inode, goal, &count, aflags, &err);
935 EXT_ASSERT(count <= cex->ec_len);
937 /* insert new extent */
938 nex.ee_block = cex->ec_block;
939 nex.ee_start = pblock;
941 err = ext3_ext_insert_extent(handle, tree, path, &nex);
946 * Putting len of the actual extent we just inserted,
947 * we are asking ext3_ext_walk_space() to continue
948 * scaning after that block
950 cex->ec_len = nex.ee_len;
951 cex->ec_start = nex.ee_start;
952 BUG_ON(nex.ee_len == 0);
953 BUG_ON(nex.ee_block != cex->ec_block);
957 journal_stop(handle);
963 CERROR("hmm. why do we find this extent?\n");
964 CERROR("initial space: %lu:%u\n",
965 bp->start, bp->init_num);
966 CERROR("current extent: %u/%u/%u %d\n",
967 cex->ec_block, cex->ec_len,
968 cex->ec_start, cex->ec_type);
971 if (cex->ec_block < bp->start)
972 i = bp->start - cex->ec_block;
973 if (i >= cex->ec_len)
974 CERROR("nothing to do?! i = %d, e_num = %u\n",
976 for (; i < cex->ec_len && bp->num; i++) {
977 *(bp->blocks) = cex->ec_start + i;
978 if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
982 /* unmap any possible underlying metadata from
983 * the block device mapping. bug 6998. */
984 ll_unmap_underlying_metadata(inode->i_sb,
996 int fsfilt_map_nblocks(struct inode *inode, unsigned long block,
997 unsigned long num, unsigned long *blocks,
998 int *created, int create)
1000 struct ext3_extents_tree tree;
1001 struct bpointers bp;
1004 CDEBUG(D_OTHER, "blocks %lu-%lu requested for inode %u\n",
1005 block, block + num - 1, (unsigned) inode->i_ino);
1007 ext3_init_tree_desc(&tree, inode);
1010 bp.created = created;
1012 bp.init_num = bp.num = num;
1015 ext3_down_truncate_sem(inode);
1016 err = ext3_ext_walk_space(&tree, block, num, ext3_ext_new_extent_cb);
1017 ext3_ext_invalidate_cache(&tree);
1018 ext3_up_truncate_sem(inode);
1023 int fsfilt_ext3_map_ext_inode_pages(struct inode *inode, struct page **page,
1024 int pages, unsigned long *blocks,
1025 int *created, int create)
1027 int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1029 struct page *fp = NULL;
1032 CDEBUG(D_OTHER, "inode %lu: map %d pages from %lu\n",
1033 inode->i_ino, pages, (*page)->index);
1035 /* pages are sorted already. so, we just have to find
1036 * contig. space and process them properly */
1039 /* start new extent */
1044 } else if (fp->index + clen == (*page)->index) {
1045 /* continue the extent */
1052 /* process found extent */
1053 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1054 clen * blocks_per_page, blocks,
1059 /* look for next extent */
1061 blocks += blocks_per_page * clen;
1062 created += blocks_per_page * clen;
1066 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1067 clen * blocks_per_page, blocks,
1072 #endif /* EXT3_MULTIBLOCK_ALLOCATOR */
1074 extern int ext3_map_inode_page(struct inode *inode, struct page *page,
1075 unsigned long *blocks, int *created, int create);
1076 int fsfilt_ext3_map_bm_inode_pages(struct inode *inode, struct page **page,
1077 int pages, unsigned long *blocks,
1078 int *created, int create)
1080 int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1084 for (i = 0, cr = created, b = blocks; i < pages; i++, page++) {
1085 rc = ext3_map_inode_page(inode, *page, b, cr, create);
1087 CERROR("ino %lu, blk %lu cr %u create %d: rc %d\n",
1088 inode->i_ino, *b, *cr, create, rc);
1092 b += blocks_per_page;
1093 cr += blocks_per_page;
1098 int fsfilt_ext3_map_inode_pages(struct inode *inode, struct page **page,
1099 int pages, unsigned long *blocks,
1100 int *created, int create,
1101 struct semaphore *optional_sem)
1104 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
1105 if (EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL) {
1106 rc = fsfilt_ext3_map_ext_inode_pages(inode, page, pages,
1107 blocks, created, create);
1111 if (optional_sem != NULL)
1113 rc = fsfilt_ext3_map_bm_inode_pages(inode, page, pages, blocks,
1115 if (optional_sem != NULL)
1121 extern int ext3_prep_san_write(struct inode *inode, long *blocks,
1122 int nblocks, loff_t newsize);
1123 static int fsfilt_ext3_prep_san_write(struct inode *inode, long *blocks,
1124 int nblocks, loff_t newsize)
1126 return ext3_prep_san_write(inode, blocks, nblocks, newsize);
1129 static int fsfilt_ext3_read_record(struct file * file, void *buf,
1130 int size, loff_t *offs)
1132 struct inode *inode = file->f_dentry->d_inode;
1133 unsigned long block;
1134 struct buffer_head *bh;
1135 int err, blocksize, csize, boffs;
1137 /* prevent reading after eof */
1139 if (inode->i_size < *offs + size) {
1140 size = inode->i_size - *offs;
1143 CERROR("size %llu is too short for read %u@%llu\n",
1144 inode->i_size, size, *offs);
1146 } else if (size == 0) {
1153 blocksize = 1 << inode->i_blkbits;
1156 block = *offs >> inode->i_blkbits;
1157 boffs = *offs & (blocksize - 1);
1158 csize = min(blocksize - boffs, size);
1159 bh = ext3_bread(NULL, inode, block, 0, &err);
1161 CERROR("can't read block: %d\n", err);
1165 memcpy(buf, bh->b_data + boffs, csize);
1175 static int fsfilt_ext3_write_record(struct file *file, void *buf, int bufsize,
1176 loff_t *offs, int force_sync)
1178 struct buffer_head *bh = NULL;
1179 unsigned long block;
1180 struct inode *inode = file->f_dentry->d_inode;
1181 loff_t old_size = inode->i_size, offset = *offs;
1182 loff_t new_size = inode->i_size;
1185 int err, block_count = 0, blocksize, size, boffs;
1187 /* Determine how many transaction credits are needed */
1188 blocksize = 1 << inode->i_blkbits;
1189 block_count = (*offs & (blocksize - 1)) + bufsize;
1190 block_count = (block_count + blocksize - 1) >> inode->i_blkbits;
1192 journal = EXT3_SB(inode->i_sb)->s_journal;
1194 handle = journal_start(journal,
1195 block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2);
1197 if (IS_ERR(handle)) {
1198 CERROR("can't start transaction for %d blocks (%d bytes)\n",
1199 block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2, bufsize);
1200 return PTR_ERR(handle);
1203 while (bufsize > 0) {
1207 block = offset >> inode->i_blkbits;
1208 boffs = offset & (blocksize - 1);
1209 size = min(blocksize - boffs, bufsize);
1210 bh = ext3_bread(handle, inode, block, 1, &err);
1212 CERROR("can't read/create block: %d\n", err);
1216 err = ext3_journal_get_write_access(handle, bh);
1218 CERROR("journal_get_write_access() returned error %d\n",
1222 LASSERT(bh->b_data + boffs + size <= bh->b_data + bh->b_size);
1223 memcpy(bh->b_data + boffs, buf, size);
1224 err = ext3_journal_dirty_metadata(handle, bh);
1226 CERROR("journal_dirty_metadata() returned error %d\n",
1230 if (offset + size > new_size)
1231 new_size = offset + size;
1238 handle->h_sync = 1; /* recovery likes this */
1243 /* correct in-core and on-disk sizes */
1244 if (new_size > inode->i_size) {
1246 if (new_size > inode->i_size)
1247 inode->i_size = new_size;
1248 if (inode->i_size > EXT3_I(inode)->i_disksize)
1249 EXT3_I(inode)->i_disksize = inode->i_size;
1250 if (inode->i_size > old_size)
1251 mark_inode_dirty(inode);
1256 journal_stop(handle);
1264 static int fsfilt_ext3_setup(struct super_block *sb)
1267 EXT3_SB(sb)->dx_lock = fsfilt_ext3_dx_lock;
1268 EXT3_SB(sb)->dx_unlock = fsfilt_ext3_dx_unlock;
1271 CWARN("Enabling PDIROPS\n");
1272 set_opt(EXT3_SB(sb)->s_mount_opt, PDIROPS);
1273 sb->s_flags |= S_PDIROPS;
1275 if (!EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1276 CWARN("filesystem doesn't have dir_index feature enabled\n");
1280 /* If fso is NULL, op is FSFILT operation, otherwise op is number of fso
1281 objects. Logs is number of logfiles to update */
1282 static int fsfilt_ext3_get_op_len(int op, struct fsfilt_objinfo *fso, int logs)
1286 case FSFILT_OP_CREATE:
1287 /* directory leaf, index & indirect & EA*/
1288 return 4 + 3 * logs;
1289 case FSFILT_OP_UNLINK:
1295 struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
1296 int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
1297 int addrpp = EXT3_ADDR_PER_BLOCK(sb) * blockpp;
1298 for (i = 0; i < op; i++, fso++) {
1299 int nblocks = fso->fso_bufcnt * blockpp;
1300 int ndindirect = min(nblocks, addrpp + 1);
1301 int nindir = nblocks + ndindirect + 1;
1305 return needed + 3 * logs;
1311 static const char *op_quotafile[] = { "lquota.user", "lquota.group" };
1313 #define DQINFO_COPY(out, in) \
1315 Q_COPY(out, in, dqi_bgrace); \
1316 Q_COPY(out, in, dqi_igrace); \
1317 Q_COPY(out, in, dqi_flags); \
1318 Q_COPY(out, in, dqi_valid); \
1321 #define DQBLK_COPY(out, in) \
1323 Q_COPY(out, in, dqb_bhardlimit); \
1324 Q_COPY(out, in, dqb_bsoftlimit); \
1325 Q_COPY(out, in, dqb_curspace); \
1326 Q_COPY(out, in, dqb_ihardlimit); \
1327 Q_COPY(out, in, dqb_isoftlimit); \
1328 Q_COPY(out, in, dqb_curinodes); \
1329 Q_COPY(out, in, dqb_btime); \
1330 Q_COPY(out, in, dqb_itime); \
1331 Q_COPY(out, in, dqb_valid); \
1336 static int fsfilt_ext3_quotactl(struct super_block *sb,
1337 struct obd_quotactl *oqc)
1339 int i, rc = 0, error = 0;
1340 struct quotactl_ops *qcop;
1341 struct if_dqinfo *info;
1342 struct if_dqblk *dqblk;
1348 OBD_ALLOC_PTR(info);
1351 OBD_ALLOC_PTR(dqblk);
1357 DQINFO_COPY(info, &oqc->qc_dqinfo);
1358 DQBLK_COPY(dqblk, &oqc->qc_dqblk);
1361 if (oqc->qc_cmd == Q_QUOTAON || oqc->qc_cmd == Q_QUOTAOFF) {
1362 for (i = 0; i < MAXQUOTAS; i++) {
1363 if (!Q_TYPESET(oqc, i))
1366 if (oqc->qc_cmd == Q_QUOTAON) {
1367 if (!qcop->quota_on)
1368 GOTO(out, rc = -ENOSYS);
1369 rc = qcop->quota_on(sb, i, oqc->qc_id,
1370 (char *)op_quotafile[i]);
1371 } else if (oqc->qc_cmd == Q_QUOTAOFF) {
1372 if (!qcop->quota_off)
1373 GOTO(out, rc = -ENOSYS);
1374 rc = qcop->quota_off(sb, i);
1382 GOTO(out, rc ?: error);
1385 switch (oqc->qc_cmd) {
1388 if (!qcop->get_info)
1389 GOTO(out, rc = -ENOSYS);
1390 rc = qcop->get_info(sb, oqc->qc_type, info);
1394 if (!qcop->set_dqblk)
1395 GOTO(out, rc = -ENOSYS);
1396 rc = qcop->set_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1400 if (!qcop->get_dqblk)
1401 GOTO(out, rc = -ENOSYS);
1402 rc = qcop->get_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1405 if (!sb->s_qcop->quota_sync)
1406 GOTO(out, rc = -ENOSYS);
1407 qcop->quota_sync(sb, oqc->qc_type);
1410 CERROR("unsupported quotactl command: %d", oqc->qc_cmd);
1414 DQINFO_COPY(&oqc->qc_dqinfo, info);
1415 DQBLK_COPY(&oqc->qc_dqblk, dqblk);
1418 OBD_FREE_PTR(dqblk);
1421 CDEBUG(D_QUOTA, "quotactl command %#x, id %u, type %d "
1423 oqc->qc_cmd, oqc->qc_id, oqc->qc_type, rc);
1428 struct hlist_node dqb_hash; /* quotacheck hash */
1429 struct list_head dqb_list; /* in list also */
1430 qid_t dqb_id; /* uid/gid */
1431 short dqb_type; /* USRQUOTA/GRPQUOTA */
1432 __u32 dqb_bhardlimit; /* block hard limit */
1433 __u32 dqb_bsoftlimit; /* block soft limit */
1434 qsize_t dqb_curspace; /* current space */
1435 __u32 dqb_ihardlimit; /* inode hard limit */
1436 __u32 dqb_isoftlimit; /* inode soft limit */
1437 __u32 dqb_curinodes; /* current inodes */
1438 __u64 dqb_btime; /* block grace time */
1439 __u64 dqb_itime; /* inode grace time */
1440 __u32 dqb_valid; /* flag for above fields */
1443 static inline unsigned int chkquot_hash(qid_t id, int type)
1444 __attribute__((__const__));
1446 static inline unsigned int chkquot_hash(qid_t id, int type)
1448 return (id * (MAXQUOTAS - type)) % NR_DQHASH;
1451 static inline struct chk_dqblk *
1452 find_chkquot(struct hlist_head *head, qid_t id, int type)
1454 struct hlist_node *node;
1455 struct chk_dqblk *cdqb;
1457 hlist_for_each(node, head) {
1458 cdqb = hlist_entry(node, struct chk_dqblk, dqb_hash);
1459 if (cdqb->dqb_id == id && cdqb->dqb_type == type)
1466 static struct chk_dqblk *alloc_chkquot(qid_t id, int type)
1468 struct chk_dqblk *cdqb;
1470 OBD_ALLOC_PTR(cdqb);
1472 INIT_HLIST_NODE(&cdqb->dqb_hash);
1473 INIT_LIST_HEAD(&cdqb->dqb_list);
1475 cdqb->dqb_type = type;
1481 static struct chk_dqblk *
1482 cqget(struct super_block *sb, struct hlist_head *hash, struct list_head *list,
1483 qid_t id, int type, int first_check)
1485 struct hlist_head *head = hash + chkquot_hash(id, type);
1486 struct if_dqblk dqb;
1487 struct chk_dqblk *cdqb;
1490 cdqb = find_chkquot(head, id, type);
1494 cdqb = alloc_chkquot(id, type);
1499 rc = sb->s_qcop->get_dqblk(sb, type, id, &dqb);
1501 CERROR("get_dqblk of id %u, type %d failed: %d\n",
1504 DQBLK_COPY(cdqb, &dqb);
1505 cdqb->dqb_curspace = 0;
1506 cdqb->dqb_curinodes = 0;
1510 hlist_add_head(&cdqb->dqb_hash, head);
1511 list_add_tail(&cdqb->dqb_list, list);
1516 static inline int quota_onoff(struct super_block *sb, int cmd, int type)
1518 struct obd_quotactl *oqctl;
1521 OBD_ALLOC_PTR(oqctl);
1525 oqctl->qc_cmd = cmd;
1526 oqctl->qc_id = QFMT_LDISKFS;
1527 oqctl->qc_type = type;
1528 rc = fsfilt_ext3_quotactl(sb, oqctl);
1530 OBD_FREE_PTR(oqctl);
1534 static inline int read_old_dqinfo(struct super_block *sb, int type,
1535 struct if_dqinfo *dqinfo)
1537 struct obd_quotactl *oqctl;
1541 OBD_ALLOC_PTR(oqctl);
1545 oqctl->qc_cmd = Q_GETINFO;
1546 oqctl->qc_type = type;
1547 rc = fsfilt_ext3_quotactl(sb, oqctl);
1549 ((struct obd_dqinfo *)dqinfo)[type] = oqctl->qc_dqinfo;
1551 OBD_FREE_PTR(oqctl);
1555 static inline struct ext3_group_desc *
1556 get_group_desc(struct super_block *sb, int group)
1558 unsigned long desc_block, desc;
1559 struct ext3_group_desc *gdp;
1561 desc_block = group / EXT3_DESC_PER_BLOCK(sb);
1562 desc = group % EXT3_DESC_PER_BLOCK(sb);
1563 gdp = (struct ext3_group_desc *)
1564 EXT3_SB(sb)->s_group_desc[desc_block]->b_data;
1569 static inline struct buffer_head *
1570 read_inode_bitmap(struct super_block *sb, unsigned long group)
1572 struct ext3_group_desc *desc;
1573 struct buffer_head *bh;
1575 desc = get_group_desc(sb, group);
1576 bh = sb_bread(sb, le32_to_cpu(desc->bg_inode_bitmap));
1581 static inline struct inode *ext3_iget_inuse(struct super_block *sb,
1582 struct buffer_head *bitmap_bh,
1583 int index, unsigned long ino)
1585 struct inode *inode = NULL;
1587 if (ext3_test_bit(index, bitmap_bh->b_data))
1588 inode = iget(sb, ino);
1594 struct hlist_head qckt_hash[NR_DQHASH]; /* quotacheck hash */
1595 struct list_head qckt_list; /* quotacheck list */
1596 int qckt_first_check[MAXQUOTAS]; /* 1 if no old quotafile */
1597 struct if_dqinfo qckt_dqinfo[MAXQUOTAS]; /* old dqinfo */
1600 static int add_inode_quota(struct inode *inode, struct qchk_ctxt *qctxt,
1601 struct obd_quotactl *oqc)
1603 struct chk_dqblk *cdqb[MAXQUOTAS] = { NULL, };
1605 qid_t qid[MAXQUOTAS];
1611 qid[USRQUOTA] = inode->i_uid;
1612 qid[GRPQUOTA] = inode->i_gid;
1614 if (S_ISDIR(inode->i_mode) ||
1615 S_ISREG(inode->i_mode) ||
1616 S_ISLNK(inode->i_mode))
1617 size = inode_get_bytes(inode);
1619 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1620 if (!Q_TYPESET(oqc, cnt))
1623 cdqb[cnt] = cqget(inode->i_sb, qctxt->qckt_hash,
1624 &qctxt->qckt_list, qid[cnt], cnt,
1625 qctxt->qckt_first_check[cnt]);
1631 cdqb[cnt]->dqb_curspace += size;
1632 cdqb[cnt]->dqb_curinodes++;
1636 for (i = 0; i < cnt; i++) {
1637 if (!Q_TYPESET(oqc, i))
1640 cdqb[i]->dqb_curspace -= size;
1641 cdqb[i]->dqb_curinodes--;
1648 static int v2_write_dqheader(struct file *f, int type)
1650 static const __u32 quota_magics[] = V2_INITQMAGICS;
1651 static const __u32 quota_versions[] = V2_INITQVERSIONS;
1652 struct v2_disk_dqheader dqhead;
1655 CLASSERT(ARRAY_SIZE(quota_magics) == ARRAY_SIZE(quota_versions));
1656 LASSERT(0 <= type && type < ARRAY_SIZE(quota_magics));
1658 dqhead.dqh_magic = cpu_to_le32(quota_magics[type]);
1659 dqhead.dqh_version = cpu_to_le32(quota_versions[type]);
1661 return cfs_user_write(f, (char *)&dqhead, sizeof(dqhead), &offset);
1664 /* write dqinfo struct in a new quota file */
1665 static int v2_write_dqinfo(struct file *f, int type, struct if_dqinfo *info)
1667 struct v2_disk_dqinfo dqinfo;
1668 __u32 blocks = V2_DQTREEOFF + 1;
1669 loff_t offset = V2_DQINFOOFF;
1672 dqinfo.dqi_bgrace = cpu_to_le32(info->dqi_bgrace);
1673 dqinfo.dqi_igrace = cpu_to_le32(info->dqi_igrace);
1674 dqinfo.dqi_flags = cpu_to_le32(info->dqi_flags & DQF_MASK &
1677 dqinfo.dqi_bgrace = cpu_to_le32(MAX_DQ_TIME);
1678 dqinfo.dqi_igrace = cpu_to_le32(MAX_IQ_TIME);
1679 dqinfo.dqi_flags = 0;
1682 dqinfo.dqi_blocks = cpu_to_le32(blocks);
1683 dqinfo.dqi_free_blk = 0;
1684 dqinfo.dqi_free_entry = 0;
1686 return cfs_user_write(f, (char *)&dqinfo, sizeof(dqinfo), &offset);
1689 static int create_new_quota_files(struct qchk_ctxt *qctxt,
1690 struct obd_quotactl *oqc)
1695 for (i = 0; i < MAXQUOTAS; i++) {
1696 struct if_dqinfo *info = qctxt->qckt_first_check[i]?
1697 NULL : &qctxt->qckt_dqinfo[i];
1700 if (!Q_TYPESET(oqc, i))
1703 file = filp_open(op_quotafile[i], O_RDWR | O_CREAT | O_TRUNC,
1707 CERROR("can't create %s file: rc = %d\n",
1708 op_quotafile[i], rc);
1712 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
1713 CERROR("file %s is not regular", op_quotafile[i]);
1714 filp_close(file, 0);
1715 GOTO(out, rc = -EINVAL);
1718 rc = v2_write_dqheader(file, i);
1720 filp_close(file, 0);
1724 rc = v2_write_dqinfo(file, i, info);
1725 filp_close(file, 0);
1735 static int commit_chkquot(struct super_block *sb, struct qchk_ctxt *qctxt,
1736 struct chk_dqblk *cdqb)
1738 struct obd_quotactl *oqc;
1747 now = CURRENT_SECONDS;
1749 if (cdqb->dqb_bsoftlimit &&
1750 toqb(cdqb->dqb_curspace) >= cdqb->dqb_bsoftlimit &&
1753 now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_bgrace;
1755 if (cdqb->dqb_isoftlimit &&
1756 cdqb->dqb_curinodes >= cdqb->dqb_isoftlimit &&
1759 now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_igrace;
1761 cdqb->dqb_valid = QIF_ALL;
1763 oqc->qc_cmd = Q_SETQUOTA;
1764 oqc->qc_type = cdqb->dqb_type;
1765 oqc->qc_id = cdqb->dqb_id;
1766 DQBLK_COPY(&oqc->qc_dqblk, cdqb);
1768 rc = fsfilt_ext3_quotactl(sb, oqc);
1773 static int prune_chkquots(struct super_block *sb,
1774 struct qchk_ctxt *qctxt, int error)
1776 struct chk_dqblk *cdqb, *tmp;
1779 list_for_each_entry_safe(cdqb, tmp, &qctxt->qckt_list, dqb_list) {
1781 rc = commit_chkquot(sb, qctxt, cdqb);
1785 hlist_del_init(&cdqb->dqb_hash);
1786 list_del(&cdqb->dqb_list);
1793 static int fsfilt_ext3_quotacheck(struct super_block *sb,
1794 struct obd_quotactl *oqc)
1796 struct ext3_sb_info *sbi = EXT3_SB(sb);
1798 struct qchk_ctxt *qctxt;
1799 struct buffer_head *bitmap_bh = NULL;
1801 struct inode *inode;
1805 /* turn on quota and read dqinfo if existed */
1806 OBD_ALLOC_PTR(qctxt);
1808 oqc->qc_stat = -ENOMEM;
1812 for (i = 0; i < NR_DQHASH; i++)
1813 INIT_HLIST_HEAD(&qctxt->qckt_hash[i]);
1814 INIT_LIST_HEAD(&qctxt->qckt_list);
1816 for (i = 0; i < MAXQUOTAS; i++) {
1817 if (!Q_TYPESET(oqc, i))
1820 rc = quota_onoff(sb, Q_QUOTAON, i);
1821 if (!rc || rc == -EBUSY) {
1822 rc = read_old_dqinfo(sb, i, qctxt->qckt_dqinfo);
1825 } else if (rc == -ENOENT) {
1826 qctxt->qckt_first_check[i] = 1;
1832 /* check quota and update in hash */
1833 for (group = 0; group < sbi->s_groups_count; group++) {
1834 ino = group * sbi->s_inodes_per_group + 1;
1835 bitmap_bh = read_inode_bitmap(sb, group);
1837 CERROR("read_inode_bitmap group %d failed", group);
1841 for (i = 0; i < sbi->s_inodes_per_group; i++, ino++) {
1842 if (ino < sbi->s_first_ino)
1845 inode = ext3_iget_inuse(sb, bitmap_bh, i, ino);
1846 rc = add_inode_quota(inode, qctxt, oqc);
1857 /* read old quota limits from old quota file. (only for the user
1858 * has limits but hasn't file) */
1859 #ifdef HAVE_QUOTA_SUPPORT
1860 for (i = 0; i < MAXQUOTAS; i++) {
1861 struct list_head id_list;
1862 struct dquot_id *dqid, *tmp;
1864 if (!Q_TYPESET(oqc, i))
1867 if (qctxt->qckt_first_check[i])
1871 LASSERT(sb_dqopt(sb)->files[i] != NULL);
1872 INIT_LIST_HEAD(&id_list);
1873 #ifndef KERNEL_SUPPORTS_QUOTA_READ
1874 rc = lustre_get_qids(sb_dqopt(sb)->files[i], NULL, i, &id_list);
1876 rc = lustre_get_qids(NULL, sb_dqopt(sb)->files[i], i, &id_list);
1879 CERROR("read old limits failed. (rc:%d)\n", rc);
1881 list_for_each_entry_safe(dqid, tmp, &id_list, di_link) {
1882 list_del_init(&dqid->di_link);
1885 cqget(sb, qctxt->qckt_hash, &qctxt->qckt_list,
1887 qctxt->qckt_first_check[i]);
1892 /* turn off quota cause we are to dump chk_dqblk to files */
1893 quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type);
1895 rc = create_new_quota_files(qctxt, oqc);
1899 /* we use vfs functions to set dqblk, so turn quota on */
1900 rc = quota_onoff(sb, Q_QUOTAON, oqc->qc_type);
1902 /* dump and free chk_dqblk */
1903 rc = prune_chkquots(sb, qctxt, rc);
1904 OBD_FREE_PTR(qctxt);
1906 /* turn off quota, `lfs quotacheck` will turn on when all
1907 * nodes quotacheck finish. */
1908 quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type);
1912 CERROR("quotacheck failed: rc = %d\n", rc);
1917 #ifdef HAVE_QUOTA_SUPPORT
1918 static int fsfilt_ext3_quotainfo(struct lustre_quota_info *lqi, int type,
1924 if (lqi->qi_files[type] == NULL) {
1925 CERROR("operate qinfo before it's enabled!\n");
1931 rc = lustre_check_quota_file(lqi, type);
1934 rc = lustre_read_quota_info(lqi, type);
1937 rc = lustre_write_quota_info(lqi, type);
1939 case QFILE_INIT_INFO:
1940 rc = lustre_init_quota_info(lqi, type);
1943 CERROR("Unsupported admin quota file cmd %d\n", cmd);
1950 static int fsfilt_ext3_qids(struct file *file, struct inode *inode, int type,
1951 struct list_head *list)
1953 return lustre_get_qids(file, inode, type, list);
1956 static int fsfilt_ext3_dquot(struct lustre_dquot *dquot, int cmd)
1961 if (dquot->dq_info->qi_files[dquot->dq_type] == NULL) {
1962 CERROR("operate dquot before it's enabled!\n");
1967 case QFILE_RD_DQUOT:
1968 rc = lustre_read_dquot(dquot);
1970 case QFILE_WR_DQUOT:
1971 if (dquot->dq_dqb.dqb_ihardlimit ||
1972 dquot->dq_dqb.dqb_isoftlimit ||
1973 dquot->dq_dqb.dqb_bhardlimit ||
1974 dquot->dq_dqb.dqb_bsoftlimit)
1975 clear_bit(DQ_FAKE_B, &dquot->dq_flags);
1977 set_bit(DQ_FAKE_B, &dquot->dq_flags);
1979 rc = lustre_commit_dquot(dquot);
1984 CERROR("Unsupported admin quota file cmd %d\n", cmd);
1992 static lvfs_sbdev_type fsfilt_ext3_journal_sbdev(struct super_block *sb)
1994 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
1995 return (EXT3_SB(sb)->journal_bdev);
1997 return kdev_t_to_nr(EXT3_SB(sb)->s_journal->j_dev);
2001 static struct fsfilt_operations fsfilt_ext3_ops = {
2003 .fs_owner = THIS_MODULE,
2004 .fs_getlabel = fsfilt_ext3_get_label,
2005 .fs_setlabel = fsfilt_ext3_set_label,
2006 .fs_uuid = fsfilt_ext3_uuid,
2007 .fs_start = fsfilt_ext3_start,
2008 .fs_brw_start = fsfilt_ext3_brw_start,
2009 .fs_extend = fsfilt_ext3_extend,
2010 .fs_commit = fsfilt_ext3_commit,
2011 .fs_commit_async = fsfilt_ext3_commit_async,
2012 .fs_commit_wait = fsfilt_ext3_commit_wait,
2013 .fs_setattr = fsfilt_ext3_setattr,
2014 .fs_iocontrol = fsfilt_ext3_iocontrol,
2015 .fs_set_md = fsfilt_ext3_set_md,
2016 .fs_get_md = fsfilt_ext3_get_md,
2017 .fs_readpage = fsfilt_ext3_readpage,
2018 .fs_add_journal_cb = fsfilt_ext3_add_journal_cb,
2019 .fs_statfs = fsfilt_ext3_statfs,
2020 .fs_sync = fsfilt_ext3_sync,
2021 .fs_map_inode_pages = fsfilt_ext3_map_inode_pages,
2022 .fs_prep_san_write = fsfilt_ext3_prep_san_write,
2023 .fs_write_record = fsfilt_ext3_write_record,
2024 .fs_read_record = fsfilt_ext3_read_record,
2025 .fs_setup = fsfilt_ext3_setup,
2026 .fs_send_bio = fsfilt_ext3_send_bio,
2027 .fs_get_op_len = fsfilt_ext3_get_op_len,
2028 .fs_quotactl = fsfilt_ext3_quotactl,
2029 .fs_quotacheck = fsfilt_ext3_quotacheck,
2030 #ifdef HAVE_QUOTA_SUPPORT
2031 .fs_quotainfo = fsfilt_ext3_quotainfo,
2032 .fs_qids = fsfilt_ext3_qids,
2033 .fs_dquot = fsfilt_ext3_dquot,
2035 .fs_journal_sbdev = fsfilt_ext3_journal_sbdev,
2038 static int __init fsfilt_ext3_init(void)
2042 fcb_cache = kmem_cache_create("fsfilt_ext3_fcb",
2043 sizeof(struct fsfilt_cb_data), 0,
2046 CERROR("error allocating fsfilt journal callback cache\n");
2047 GOTO(out, rc = -ENOMEM);
2050 rc = fsfilt_register_ops(&fsfilt_ext3_ops);
2053 int err = kmem_cache_destroy(fcb_cache);
2054 LASSERTF(err == 0, "error destroying new cache: rc %d\n", err);
2060 static void __exit fsfilt_ext3_exit(void)
2064 fsfilt_unregister_ops(&fsfilt_ext3_ops);
2065 rc = kmem_cache_destroy(fcb_cache);
2066 LASSERTF(rc == 0, "couldn't destroy fcb_cache slab\n");
2069 module_init(fsfilt_ext3_init);
2070 module_exit(fsfilt_ext3_exit);
2072 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
2073 MODULE_DESCRIPTION("Lustre ext3 Filesystem Helper v0.1");
2074 MODULE_LICENSE("GPL");