1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/lvfs/fsfilt_ext3.c
38 * Author: Andreas Dilger <adilger@clusterfs.com>
41 #define DEBUG_SUBSYSTEM S_FILTER
43 #include <linux/init.h>
44 #include <linux/module.h>
46 #include <linux/slab.h>
47 #include <linux/pagemap.h>
48 #include <linux/quotaops.h>
49 #ifdef HAVE_EXT4_LDISKFS
50 #include <ext4/ext4.h>
51 #include <ext4/ext4_jbd2.h>
53 #include <linux/jbd.h>
54 #include <linux/ext3_fs.h>
55 #include <linux/ext3_jbd.h>
57 #include <linux/version.h>
58 #include <linux/bitops.h>
59 #include <linux/quota.h>
60 #include <linux/quotaio_v1.h>
61 #include <linux/quotaio_v2.h>
62 #include <ext3/xattr.h>
64 #include <libcfs/libcfs.h>
65 #include <lustre_fsfilt.h>
67 #include <lustre_quota.h>
68 #include <linux/lustre_compat25.h>
69 #include <linux/lprocfs_status.h>
71 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
72 #ifdef HAVE_EXT4_LDISKFS
73 #include <ext4/ext4_extents.h>
75 #include <linux/ext3_extents.h>
79 #include "lustre_quota_fmt.h"
81 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,15)
82 #define FSFILT_DATA_TRANS_BLOCKS(sb) EXT3_DATA_TRANS_BLOCKS
83 #define FSFILT_DELETE_TRANS_BLOCKS(sb) EXT3_DELETE_TRANS_BLOCKS
85 #define FSFILT_DATA_TRANS_BLOCKS(sb) EXT3_DATA_TRANS_BLOCKS(sb)
86 #define FSFILT_DELETE_TRANS_BLOCKS(sb) EXT3_DELETE_TRANS_BLOCKS(sb)
89 #ifdef EXT3_SINGLEDATA_TRANS_BLOCKS_HAS_SB
90 /* for kernels 2.6.18 and later */
91 #define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS(sb)
93 #define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS
96 #define fsfilt_ext3_journal_start(inode, nblocks) ext3_journal_start(inode, nblocks)
97 #define fsfilt_ext3_journal_stop(handle) ext3_journal_stop(handle)
99 static cfs_mem_cache_t *fcb_cache;
101 struct fsfilt_cb_data {
102 struct journal_callback cb_jcb; /* jbd private data - MUST BE FIRST */
103 fsfilt_cb_t cb_func; /* MDS/OBD completion function */
104 struct obd_device *cb_obd; /* MDS/OBD completion device */
105 __u64 cb_last_rcvd; /* MDS/OST last committed operation */
106 void *cb_data; /* MDS/OST completion function data */
109 #ifndef EXT3_XATTR_INDEX_TRUSTED /* temporary until we hit l28 kernel */
110 #define EXT3_XATTR_INDEX_TRUSTED 4
113 #ifdef HAVE_EXT4_LDISKFS
114 #define fsfilt_log_start_commit(journal, tid) jbd2_log_start_commit(journal, tid)
115 #define fsfilt_log_wait_commit(journal, tid) jbd2_log_wait_commit(journal, tid)
116 #define fsfilt_journal_callback_set(handle, func, jcb) jbd2_journal_callback_set(handle, func, jcb)
118 #define fsfilt_log_start_commit(journal, tid) log_start_commit(journal, tid)
119 #define fsfilt_log_wait_commit(journal, tid) log_wait_commit(journal, tid)
120 #define fsfilt_journal_callback_set(handle, func, jcb) journal_callback_set(handle, func, jcb)
121 #define ext_pblock(ex) le32_to_cpu((ex)->ee_start)
122 #define ext3_ext_store_pblock(ex, pblock) ((ex)->ee_start = cpu_to_le32(pblock))
123 #define ext3_inode_bitmap(sb,desc) le32_to_cpu((desc)->bg_inode_bitmap)
126 static char *fsfilt_ext3_get_label(struct super_block *sb)
128 return EXT3_SB(sb)->s_es->s_volume_name;
131 static int fsfilt_ext3_set_label(struct super_block *sb, char *label)
133 /* see e.g. fsfilt_ext3_write_record() */
138 journal = EXT3_SB(sb)->s_journal;
139 handle = ext3_journal_start_sb(sb, 1);
140 if (IS_ERR(handle)) {
141 CERROR("can't start transaction\n");
142 return(PTR_ERR(handle));
145 err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
149 memcpy(EXT3_SB(sb)->s_es->s_volume_name, label,
150 sizeof(EXT3_SB(sb)->s_es->s_volume_name));
152 err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
155 ext3_journal_stop(handle);
160 static char *fsfilt_ext3_uuid(struct super_block *sb)
162 return EXT3_SB(sb)->s_es->s_uuid;
165 #ifdef HAVE_DISK_INODE_VERSION
167 * Get the 64-bit version for an inode.
169 static __u64 fsfilt_ext3_get_version(struct inode *inode)
171 CDEBUG(D_INFO, "Get version "LPX64" for inode %lu\n",
172 EXT3_I(inode)->i_fs_version, inode->i_ino);
173 return EXT3_I(inode)->i_fs_version;
177 * Set the 64-bit version and return the old version.
179 static __u64 fsfilt_ext3_set_version(struct inode *inode, __u64 new_version)
181 __u64 old_version = EXT3_I(inode)->i_fs_version;
183 CDEBUG(D_INFO, "Set version "LPX64" (old "LPX64") for inode %lu\n",
184 new_version, old_version, inode->i_ino);
185 (EXT3_I(inode))->i_fs_version = new_version;
186 /* version is set after all inode operations are finished, so we should
187 * mark it dirty here */
188 inode->i_sb->s_op->dirty_inode(inode);
195 * We don't currently need any additional blocks for rmdir and
196 * unlink transactions because we are storing the OST oa_id inside
197 * the inode (which we will be changing anyways as part of this
200 static void *fsfilt_ext3_start(struct inode *inode, int op, void *desc_private,
203 /* For updates to the last received file */
204 int nblocks = FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
208 if (current->journal_info) {
209 CDEBUG(D_INODE, "increasing refcount on %p\n",
210 current->journal_info);
215 case FSFILT_OP_RMDIR:
216 case FSFILT_OP_UNLINK:
217 /* delete one file + create/update logs for each stripe */
218 nblocks += FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb);
219 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
220 FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
222 case FSFILT_OP_RENAME:
223 /* modify additional directory */
224 nblocks += FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
226 case FSFILT_OP_SYMLINK:
227 /* additional block + block bitmap + GDT for long symlink */
230 case FSFILT_OP_CREATE: {
231 #if defined(EXT3_EXTENTS_FL) && defined(EXT3_INDEX_FL) && !defined(HAVE_EXT4_LDISKFS)
234 if (!test_opt(inode->i_sb, EXTENTS)) {
236 } else if (((EXT3_I(inode)->i_flags &
237 cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL)) ==
238 cpu_to_le32(EXT3_EXTENTS_FL | EXT3_INDEX_FL))) {
239 CWARN("extent-mapped directory found with "
240 "ext3-based ldiskfs - contact "
241 "http://bugzilla.lustre.org/\n");
248 case FSFILT_OP_MKDIR:
249 case FSFILT_OP_MKNOD:
250 /* modify one inode + block bitmap + GDT */
254 /* modify parent directory */
255 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
256 FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
257 /* create/update logs for each stripe */
258 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
259 FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
261 case FSFILT_OP_SETATTR:
262 /* Setattr on inode */
264 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
265 FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
266 /* quota chown log for each stripe */
267 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
268 FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
270 case FSFILT_OP_CANCEL_UNLINK:
271 /* blocks for log header bitmap update OR
272 * blocks for catalog header bitmap update + unlink of logs */
273 nblocks = (LLOG_CHUNK_SIZE >> inode->i_blkbits) +
274 FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb) * logs;
277 /* delete 2 file(file + array id) + create 1 file (array id)
278 * create/update logs for each stripe */
279 nblocks += 2 * FSFILT_DELETE_TRANS_BLOCKS(inode->i_sb);
281 /*create array log for head file*/
283 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
284 FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb));
285 /*update head file array */
286 nblocks += EXT3_INDEX_EXTRA_TRANS_BLOCKS +
287 FSFILT_DATA_TRANS_BLOCKS(inode->i_sb);
289 default: CERROR("unknown transaction start op %d\n", op);
293 LASSERT(current->journal_info == desc_private);
294 journal = EXT3_SB(inode->i_sb)->s_journal;
295 if (nblocks > journal->j_max_transaction_buffers) {
296 CWARN("too many credits %d for op %ux%u using %d instead\n",
297 nblocks, op, logs, journal->j_max_transaction_buffers);
298 nblocks = journal->j_max_transaction_buffers;
302 LASSERTF(nblocks > 0, "can't start %d credit transaction\n", nblocks);
303 handle = fsfilt_ext3_journal_start(inode, nblocks);
306 LASSERT(current->journal_info == handle);
308 CERROR("error starting handle for op %u (%u credits): rc %ld\n",
309 op, nblocks, PTR_ERR(handle));
314 * Calculate the number of buffer credits needed to write multiple pages in
315 * a single ext3 transaction. No, this shouldn't be here, but as yet ext3
316 * doesn't have a nice API for calculating this sort of thing in advance.
318 * See comment above ext3_writepage_trans_blocks for details. We assume
319 * no data journaling is being done, but it does allow for all of the pages
320 * being non-contiguous. If we are guaranteed contiguous pages we could
321 * reduce the number of (d)indirect blocks a lot.
323 * With N blocks per page and P pages, for each inode we have at most:
325 * min(N*P, blocksize/4 + 1) dindirect blocks
328 * For the entire filesystem, we have at most:
329 * min(sum(nindir + P), ngroups) bitmap blocks (from the above)
330 * min(sum(nindir + P), gdblocks) group descriptor blocks (from the above)
331 * objcount inode blocks
333 * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quota files
335 * 1 EXT3_DATA_TRANS_BLOCKS for the last_rcvd update.
337 static int fsfilt_ext3_credits_needed(int objcount, struct fsfilt_objinfo *fso,
338 int niocount, struct niobuf_local *nb)
340 struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
342 const int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
343 int nbitmaps = 0, ngdblocks;
344 int needed = objcount + 1; /* inodes + superblock */
347 for (i = 0, j = 0; i < objcount; i++, fso++) {
348 /* two or more dindirect blocks in case we cross boundary */
349 int ndind = (long)((nb[j + fso->fso_bufcnt - 1].offset -
351 sb->s_blocksize_bits) /
352 (EXT3_ADDR_PER_BLOCK(sb) * EXT3_ADDR_PER_BLOCK(sb));
353 nbitmaps += min(fso->fso_bufcnt, ndind > 0 ? ndind : 2);
355 /* leaf, indirect, tindirect blocks for first block */
356 nbitmaps += blockpp + 2;
358 j += fso->fso_bufcnt;
361 next_indir = nb[0].offset +
362 (EXT3_ADDR_PER_BLOCK(sb) << sb->s_blocksize_bits);
363 for (i = 1; i < niocount; i++) {
364 if (nb[i].offset >= next_indir) {
365 nbitmaps++; /* additional indirect */
366 next_indir = nb[i].offset +
367 (EXT3_ADDR_PER_BLOCK(sb)<<sb->s_blocksize_bits);
368 } else if (nb[i].offset != nb[i - 1].offset + sb->s_blocksize) {
369 nbitmaps++; /* additional indirect */
371 nbitmaps += blockpp; /* each leaf in different group? */
374 ngdblocks = nbitmaps;
375 if (nbitmaps > EXT3_SB(sb)->s_groups_count)
376 nbitmaps = EXT3_SB(sb)->s_groups_count;
377 if (ngdblocks > EXT3_SB(sb)->s_gdb_count)
378 ngdblocks = EXT3_SB(sb)->s_gdb_count;
380 needed += nbitmaps + ngdblocks;
382 /* last_rcvd update */
383 needed += FSFILT_DATA_TRANS_BLOCKS(sb);
385 #if defined(CONFIG_QUOTA)
386 /* We assume that there will be 1 bit set in s_dquot.flags for each
387 * quota file that is active. This is at least true for now.
389 needed += hweight32(sb_any_quota_enabled(sb)) *
390 FSFILT_SINGLEDATA_TRANS_BLOCKS(sb);
396 /* We have to start a huge journal transaction here to hold all of the
397 * metadata for the pages being written here. This is necessitated by
398 * the fact that we do lots of prepare_write operations before we do
399 * any of the matching commit_write operations, so even if we split
400 * up to use "smaller" transactions none of them could complete until
401 * all of them were opened. By having a single journal transaction,
402 * we eliminate duplicate reservations for common blocks like the
403 * superblock and group descriptors or bitmaps.
405 * We will start the transaction here, but each prepare_write will
406 * add a refcount to the transaction, and each commit_write will
407 * remove a refcount. The transaction will be closed when all of
408 * the pages have been written.
410 static void *fsfilt_ext3_brw_start(int objcount, struct fsfilt_objinfo *fso,
411 int niocount, struct niobuf_local *nb,
412 void *desc_private, int logs)
419 LASSERT(current->journal_info == desc_private);
420 journal = EXT3_SB(fso->fso_dentry->d_inode->i_sb)->s_journal;
421 needed = fsfilt_ext3_credits_needed(objcount, fso, niocount, nb);
423 /* The number of blocks we could _possibly_ dirty can very large.
424 * We reduce our request if it is absurd (and we couldn't get that
425 * many credits for a single handle anyways).
427 * At some point we have to limit the size of I/Os sent at one time,
428 * increase the size of the journal, or we have to calculate the
429 * actual journal requirements more carefully by checking all of
430 * the blocks instead of being maximally pessimistic. It remains to
431 * be seen if this is a real problem or not.
433 if (needed > journal->j_max_transaction_buffers) {
434 CERROR("want too many journal credits (%d) using %d instead\n",
435 needed, journal->j_max_transaction_buffers);
436 needed = journal->j_max_transaction_buffers;
439 LASSERTF(needed > 0, "can't start %d credit transaction\n", needed);
440 handle = fsfilt_ext3_journal_start(fso->fso_dentry->d_inode, needed);
441 if (IS_ERR(handle)) {
442 CERROR("can't get handle for %d credits: rc = %ld\n", needed,
445 LASSERT(handle->h_buffer_credits >= needed);
446 LASSERT(current->journal_info == handle);
452 static int fsfilt_ext3_extend(struct inode *inode, unsigned int nblocks,void *h)
454 handle_t *handle = h;
456 /* fsfilt_extend called with nblocks = 0 for testing in special cases */
458 handle->h_buffer_credits = 0;
459 CWARN("setting credits of handle %p to zero by request\n", h);
462 if (handle->h_buffer_credits > nblocks)
464 if (ext3_journal_extend(handle, nblocks) == 0)
467 ext3_mark_inode_dirty(handle, inode);
468 return ext3_journal_restart(handle, nblocks);
471 static int fsfilt_ext3_commit(struct inode *inode, void *h, int force_sync)
474 handle_t *handle = h;
476 LASSERT(current->journal_info == handle);
478 handle->h_sync = 1; /* recovery likes this */
480 rc = fsfilt_ext3_journal_stop(handle);
485 static int fsfilt_ext3_commit_async(struct inode *inode, void *h,
489 transaction_t *transaction;
490 handle_t *handle = h;
494 LASSERT(current->journal_info == handle);
496 transaction = handle->h_transaction;
497 journal = transaction->t_journal;
498 tid = transaction->t_tid;
499 /* we don't want to be blocked */
501 rc = fsfilt_ext3_journal_stop(handle);
503 CERROR("error while stopping transaction: %d\n", rc);
506 fsfilt_log_start_commit(journal, tid);
508 *wait_handle = (void *) tid;
509 CDEBUG(D_INODE, "commit async: %lu\n", (unsigned long) tid);
513 static int fsfilt_ext3_commit_wait(struct inode *inode, void *h)
515 journal_t *journal = EXT3_JOURNAL(inode);
516 tid_t tid = (tid_t)(long)h;
518 CDEBUG(D_INODE, "commit wait: %lu\n", (unsigned long) tid);
519 if (unlikely(is_journal_aborted(journal)))
522 fsfilt_log_wait_commit(EXT3_JOURNAL(inode), tid);
524 if (unlikely(is_journal_aborted(journal)))
529 static int fsfilt_ext3_setattr(struct dentry *dentry, void *handle,
530 struct iattr *iattr, int do_trunc)
532 struct inode *inode = dentry->d_inode;
535 /* Avoid marking the inode dirty on the superblock list unnecessarily.
536 * We are already writing the inode to disk as part of this
537 * transaction and want to avoid a lot of extra inode writeout
538 * later on. b=9828 */
539 if (iattr->ia_valid & ATTR_SIZE && !do_trunc) {
540 /* ATTR_SIZE would invoke truncate: clear it */
541 iattr->ia_valid &= ~ATTR_SIZE;
542 EXT3_I(inode)->i_disksize = iattr->ia_size;
543 i_size_write(inode, iattr->ia_size);
545 if (iattr->ia_valid & ATTR_UID)
546 inode->i_uid = iattr->ia_uid;
547 if (iattr->ia_valid & ATTR_GID)
548 inode->i_gid = iattr->ia_gid;
549 if (iattr->ia_valid & ATTR_ATIME)
550 inode->i_atime = iattr->ia_atime;
551 if (iattr->ia_valid & ATTR_MTIME)
552 inode->i_mtime = iattr->ia_mtime;
553 if (iattr->ia_valid & ATTR_CTIME)
554 inode->i_ctime = iattr->ia_ctime;
555 if (iattr->ia_valid & ATTR_MODE) {
556 inode->i_mode = iattr->ia_mode;
558 if (!in_group_p(inode->i_gid) &&
559 !cfs_capable(CFS_CAP_FSETID))
560 inode->i_mode &= ~S_ISGID;
563 inode->i_sb->s_op->dirty_inode(inode);
568 /* Don't allow setattr to change file type */
569 if (iattr->ia_valid & ATTR_MODE)
570 iattr->ia_mode = (inode->i_mode & S_IFMT) |
571 (iattr->ia_mode & ~S_IFMT);
573 /* We set these flags on the client, but have already checked perms
574 * so don't confuse inode_change_ok. */
575 iattr->ia_valid &= ~(ATTR_MTIME_SET | ATTR_ATIME_SET);
577 if (inode->i_op->setattr) {
578 rc = inode->i_op->setattr(dentry, iattr);
580 rc = inode_change_ok(inode, iattr);
582 rc = inode_setattr(inode, iattr);
589 static int fsfilt_ext3_iocontrol(struct inode *inode, struct file *file,
590 unsigned int cmd, unsigned long arg)
595 /* FIXME: Can't do this because of nested transaction deadlock */
596 if (cmd == EXT3_IOC_SETFLAGS && (*(int *)arg) & EXT3_JOURNAL_DATA_FL) {
597 CERROR("can't set data journal flag on file\n");
601 #ifdef HAVE_EXT4_LDISKFS
602 /* ext4_ioctl does not have a inode argument */
603 if (inode->i_fop->unlocked_ioctl)
604 rc = inode->i_fop->unlocked_ioctl(file, cmd, arg);
606 if (inode->i_fop->ioctl)
607 rc = inode->i_fop->ioctl(inode, file, cmd, arg);
615 static int fsfilt_ext3_set_md(struct inode *inode, void *handle,
616 void *lmm, int lmm_size, const char *name)
620 LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
622 rc = ext3_xattr_set_handle(handle, inode, EXT3_XATTR_INDEX_TRUSTED,
623 name, lmm, lmm_size, XATTR_NO_CTIME);
626 if (rc && rc != -EROFS)
627 CERROR("error adding MD data to inode %lu: rc = %d\n",
632 /* Must be called with i_mutex held */
633 static int fsfilt_ext3_get_md(struct inode *inode, void *lmm, int lmm_size,
638 LASSERT(TRYLOCK_INODE_MUTEX(inode) == 0);
640 rc = ext3_xattr_get(inode, EXT3_XATTR_INDEX_TRUSTED,
641 name, lmm, lmm_size);
643 /* This gives us the MD size */
645 return (rc == -ENODATA) ? 0 : rc;
648 CDEBUG(D_INFO, "error getting EA %d/%s from inode %lu: rc %d\n",
649 EXT3_XATTR_INDEX_TRUSTED, name,
651 memset(lmm, 0, lmm_size);
652 return (rc == -ENODATA) ? 0 : rc;
658 static int fsfilt_ext3_send_bio(int rw, struct inode *inode, struct bio *bio)
664 static ssize_t fsfilt_ext3_readpage(struct file *file, char *buf, size_t count,
667 struct inode *inode = file->f_dentry->d_inode;
670 if (S_ISREG(inode->i_mode))
671 rc = file->f_op->read(file, buf, count, off);
673 const int blkbits = inode->i_sb->s_blocksize_bits;
674 const int blksize = inode->i_sb->s_blocksize;
676 CDEBUG(D_EXT2, "reading "LPSZ" at dir %lu+%llu\n",
677 count, inode->i_ino, *off);
679 struct buffer_head *bh;
682 if (*off < i_size_read(inode)) {
685 bh = ext3_bread(NULL, inode, *off >> blkbits,
688 CDEBUG(D_EXT2, "read %u@%llu\n", blksize, *off);
691 memcpy(buf, bh->b_data, blksize);
694 /* XXX in theory we should just fake
695 * this buffer and continue like ext3,
696 * especially if this is a partial read
698 CERROR("error read dir %lu+%llu: %d\n",
699 inode->i_ino, *off, err);
704 struct ext3_dir_entry_2 *fake = (void *)buf;
706 CDEBUG(D_EXT2, "fake %u@%llu\n", blksize, *off);
707 memset(fake, 0, sizeof(*fake));
708 fake->rec_len = cpu_to_le16(blksize);
720 static void fsfilt_ext3_cb_func(struct journal_callback *jcb, int error)
722 struct fsfilt_cb_data *fcb = (struct fsfilt_cb_data *)jcb;
724 fcb->cb_func(fcb->cb_obd, fcb->cb_last_rcvd, fcb->cb_data, error);
726 OBD_SLAB_FREE(fcb, fcb_cache, sizeof *fcb);
729 static int fsfilt_ext3_add_journal_cb(struct obd_device *obd, __u64 last_rcvd,
730 void *handle, fsfilt_cb_t cb_func,
733 struct fsfilt_cb_data *fcb;
735 OBD_SLAB_ALLOC_PTR_GFP(fcb, fcb_cache, CFS_ALLOC_IO);
739 fcb->cb_func = cb_func;
741 fcb->cb_last_rcvd = last_rcvd;
742 fcb->cb_data = cb_data;
744 CDEBUG(D_EXT2, "set callback for last_rcvd: "LPD64"\n", last_rcvd);
745 fsfilt_journal_callback_set(handle, fsfilt_ext3_cb_func,
746 (struct journal_callback *)fcb);
752 * We need to hack the return value for the free inode counts because
753 * the current EA code requires one filesystem block per inode with EAs,
754 * so it is possible to run out of blocks before we run out of inodes.
756 * This can be removed when the ext3 EA code is fixed.
758 static int fsfilt_ext3_statfs(struct super_block *sb, struct obd_statfs *osfs)
763 memset(&sfs, 0, sizeof(sfs));
764 rc = ll_do_statfs(sb, &sfs);
765 if (!rc && sfs.f_bfree < sfs.f_ffree) {
766 sfs.f_files = (sfs.f_files - sfs.f_ffree) + sfs.f_bfree;
767 sfs.f_ffree = sfs.f_bfree;
770 statfs_pack(osfs, &sfs);
774 static int fsfilt_ext3_sync(struct super_block *sb)
776 return ext3_force_commit(sb);
779 #if defined(EXT3_MULTIBLOCK_ALLOCATOR) && (!defined(EXT3_EXT_CACHE_NO) || defined(EXT_CACHE_MARK))
780 #warning "kernel code has old extents/mballoc patch, disabling"
781 #undef EXT3_MULTIBLOCK_ALLOCATOR
783 #ifndef EXT3_EXTENTS_FL
784 #define EXT3_EXTENTS_FL 0x00080000 /* Inode uses extents */
787 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
788 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,17))
789 #define fsfilt_up_truncate_sem(inode) up(&EXT3_I(inode)->truncate_sem);
790 #define fsfilt_down_truncate_sem(inode) down(&EXT3_I(inode)->truncate_sem);
792 #ifdef HAVE_EXT4_LDISKFS
793 #define fsfilt_up_truncate_sem(inode) up_write((&EXT4_I(inode)->i_data_sem));
794 #define fsfilt_down_truncate_sem(inode) down_write((&EXT4_I(inode)->i_data_sem));
796 #define fsfilt_up_truncate_sem(inode) mutex_unlock(&EXT3_I(inode)->truncate_mutex);
797 #define fsfilt_down_truncate_sem(inode) mutex_lock(&EXT3_I(inode)->truncate_mutex);
802 #define EXT_ASSERT(cond) BUG_ON(!(cond))
805 #ifdef EXT3_EXT_HAS_NO_TREE
806 /* for kernels 2.6.18 and later */
807 #ifdef HAVE_EXT4_LDISKFS
808 #define EXT_GENERATION(inode) (EXT4_I(inode)->i_ext_generation)
810 #define EXT_GENERATION(inode) ext_generation(inode)
812 #define ext3_ext_base inode
813 #define ext3_ext_base2inode(inode) (inode)
814 #define EXT_DEPTH(inode) ext_depth(inode)
815 #define fsfilt_ext3_ext_walk_space(inode, block, num, cb, cbdata) \
816 ext3_ext_walk_space(inode, block, num, cb, cbdata);
818 #define ext3_ext_base ext3_extents_tree
819 #define ext3_ext_base2inode(tree) (tree->inode)
820 #define fsfilt_ext3_ext_walk_space(tree, block, num, cb, cbdata) \
821 ext3_ext_walk_space(tree, block, num, cb);
824 #include <linux/lustre_version.h>
827 unsigned long *blocks;
835 static int ext3_ext_find_goal(struct inode *inode, struct ext3_ext_path *path,
836 unsigned long block, int *aflags)
838 struct ext3_inode_info *ei = EXT3_I(inode);
839 unsigned long bg_start;
840 unsigned long colour;
844 struct ext3_extent *ex;
845 depth = path->p_depth;
847 /* try to predict block placement */
848 if ((ex = path[depth].p_ext))
849 return ext_pblock(ex) + (block - le32_to_cpu(ex->ee_block));
851 /* it looks index is empty
852 * try to find starting from index itself */
853 if (path[depth].p_bh)
854 return path[depth].p_bh->b_blocknr;
857 /* OK. use inode's group */
858 bg_start = (ei->i_block_group * EXT3_BLOCKS_PER_GROUP(inode->i_sb)) +
859 le32_to_cpu(EXT3_SB(inode->i_sb)->s_es->s_first_data_block);
860 colour = (current->pid % 16) *
861 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
862 return bg_start + colour + block;
865 #define ll_unmap_underlying_metadata(sb, blocknr) \
866 unmap_underlying_metadata((sb)->s_bdev, blocknr)
868 #ifndef EXT3_MB_HINT_GROUP_ALLOC
869 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
870 struct ext3_ext_path *path, unsigned long block,
871 unsigned long *count, int *err)
873 unsigned long pblock, goal;
875 struct inode *inode = ext3_ext_base2inode(base);
877 goal = ext3_ext_find_goal(inode, path, block, &aflags);
878 aflags |= 2; /* block have been already reserved */
879 pblock = ext3_mb_new_blocks(handle, inode, goal, count, aflags, err);
884 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
885 struct ext3_ext_path *path, unsigned long block,
886 unsigned long *count, int *err)
888 struct inode *inode = ext3_ext_base2inode(base);
889 struct ext3_allocation_request ar;
890 unsigned long pblock;
893 /* find neighbour allocated blocks */
895 *err = ext3_ext_search_left(base, path, &ar.lleft, &ar.pleft);
899 *err = ext3_ext_search_right(base, path, &ar.lright, &ar.pright);
903 /* allocate new block */
904 ar.goal = ext3_ext_find_goal(inode, path, block, &aflags);
908 ar.flags = EXT3_MB_HINT_DATA;
909 pblock = ext3_mb_new_blocks(handle, &ar, err);
915 #ifdef EXT3_EXT_HAS_NO_TREE
916 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
917 struct ext3_ext_path *path,
918 struct ext3_ext_cache *cex,
919 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
920 struct ext3_extent *ex,
924 struct bpointers *bp = cbdata;
926 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
927 struct ext3_ext_path *path,
928 struct ext3_ext_cache *cex
929 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
930 , struct ext3_extent *ex
934 struct bpointers *bp = base->private;
936 struct inode *inode = ext3_ext_base2inode(base);
937 struct ext3_extent nex;
938 unsigned long pblock;
945 EXT_ASSERT(i == path->p_depth);
946 EXT_ASSERT(path[i].p_hdr);
948 if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
953 if (bp->create == 0) {
955 if (cex->ec_block < bp->start)
956 i = bp->start - cex->ec_block;
957 if (i >= cex->ec_len)
958 CERROR("nothing to do?! i = %d, e_num = %u\n",
960 for (; i < cex->ec_len && bp->num; i++) {
972 tgen = EXT_GENERATION(base);
973 count = ext3_ext_calc_credits_for_insert(base, path);
974 fsfilt_up_truncate_sem(inode);
976 handle = fsfilt_ext3_journal_start(inode, count+EXT3_ALLOC_NEEDED+1);
977 if (IS_ERR(handle)) {
978 fsfilt_down_truncate_sem(inode);
979 return PTR_ERR(handle);
982 fsfilt_down_truncate_sem(inode);
983 if (tgen != EXT_GENERATION(base)) {
984 /* the tree has changed. so path can be invalid at moment */
985 fsfilt_ext3_journal_stop(handle);
990 pblock = new_blocks(handle, base, path, cex->ec_block, &count, &err);
993 EXT_ASSERT(count <= cex->ec_len);
995 /* insert new extent */
996 nex.ee_block = cpu_to_le32(cex->ec_block);
997 ext3_ext_store_pblock(&nex, pblock);
998 nex.ee_len = cpu_to_le16(count);
999 err = ext3_ext_insert_extent(handle, base, path, &nex);
1001 /* free data blocks we just allocated */
1002 /* not a good idea to call discard here directly,
1003 * but otherwise we'd need to call it every free() */
1004 #ifdef EXT3_MB_HINT_GROUP_ALLOC
1005 ext3_mb_discard_inode_preallocations(inode);
1007 ext3_free_blocks(handle, inode, ext_pblock(&nex),
1008 cpu_to_le16(nex.ee_len), 0);
1013 * Putting len of the actual extent we just inserted,
1014 * we are asking ext3_ext_walk_space() to continue
1015 * scaning after that block
1017 cex->ec_len = le16_to_cpu(nex.ee_len);
1018 cex->ec_start = ext_pblock(&nex);
1019 BUG_ON(le16_to_cpu(nex.ee_len) == 0);
1020 BUG_ON(le32_to_cpu(nex.ee_block) != cex->ec_block);
1023 fsfilt_ext3_journal_stop(handle);
1028 CERROR("hmm. why do we find this extent?\n");
1029 CERROR("initial space: %lu:%u\n",
1030 bp->start, bp->init_num);
1031 CERROR("current extent: %u/%u/%llu %d\n",
1032 cex->ec_block, cex->ec_len,
1033 (unsigned long long)cex->ec_start,
1037 if (cex->ec_block < bp->start)
1038 i = bp->start - cex->ec_block;
1039 if (i >= cex->ec_len)
1040 CERROR("nothing to do?! i = %d, e_num = %u\n",
1042 for (; i < cex->ec_len && bp->num; i++) {
1043 *(bp->blocks) = cex->ec_start + i;
1044 if (cex->ec_type == EXT3_EXT_CACHE_EXTENT) {
1048 /* unmap any possible underlying metadata from
1049 * the block device mapping. bug 6998. */
1050 ll_unmap_underlying_metadata(inode->i_sb,
1062 int fsfilt_map_nblocks(struct inode *inode, unsigned long block,
1063 unsigned long num, unsigned long *blocks,
1064 int *created, int create)
1066 #ifdef EXT3_EXT_HAS_NO_TREE
1067 struct ext3_ext_base *base = inode;
1069 struct ext3_extents_tree tree;
1070 struct ext3_ext_base *base = &tree;
1072 struct bpointers bp;
1075 CDEBUG(D_OTHER, "blocks %lu-%lu requested for inode %u\n",
1076 block, block + num - 1, (unsigned) inode->i_ino);
1078 #ifndef EXT3_EXT_HAS_NO_TREE
1079 ext3_init_tree_desc(base, inode);
1083 bp.created = created;
1085 bp.init_num = bp.num = num;
1088 fsfilt_down_truncate_sem(inode);
1089 err = fsfilt_ext3_ext_walk_space(base, block, num,
1090 ext3_ext_new_extent_cb, &bp);
1091 ext3_ext_invalidate_cache(base);
1092 fsfilt_up_truncate_sem(inode);
1097 int fsfilt_ext3_map_ext_inode_pages(struct inode *inode, struct page **page,
1098 int pages, unsigned long *blocks,
1099 int *created, int create)
1101 int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1103 struct page *fp = NULL;
1106 CDEBUG(D_OTHER, "inode %lu: map %d pages from %lu\n",
1107 inode->i_ino, pages, (*page)->index);
1109 /* pages are sorted already. so, we just have to find
1110 * contig. space and process them properly */
1113 /* start new extent */
1118 } else if (fp->index + clen == (*page)->index) {
1119 /* continue the extent */
1126 /* process found extent */
1127 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1128 clen * blocks_per_page, blocks,
1133 /* look for next extent */
1135 blocks += blocks_per_page * clen;
1136 created += blocks_per_page * clen;
1140 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
1141 clen * blocks_per_page, blocks,
1146 #endif /* EXT3_MULTIBLOCK_ALLOCATOR */
1148 extern int ext3_map_inode_page(struct inode *inode, struct page *page,
1149 unsigned long *blocks, int *created, int create);
1150 int fsfilt_ext3_map_bm_inode_pages(struct inode *inode, struct page **page,
1151 int pages, unsigned long *blocks,
1152 int *created, int create)
1154 int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
1158 for (i = 0, cr = created, b = blocks; i < pages; i++, page++) {
1159 rc = ext3_map_inode_page(inode, *page, b, cr, create);
1161 CERROR("ino %lu, blk %lu cr %u create %d: rc %d\n",
1162 inode->i_ino, *b, *cr, create, rc);
1166 b += blocks_per_page;
1167 cr += blocks_per_page;
1172 int fsfilt_ext3_map_inode_pages(struct inode *inode, struct page **page,
1173 int pages, unsigned long *blocks,
1174 int *created, int create,
1175 struct semaphore *optional_sem)
1178 #ifdef EXT3_MULTIBLOCK_ALLOCATOR
1179 if (EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL) {
1180 rc = fsfilt_ext3_map_ext_inode_pages(inode, page, pages,
1181 blocks, created, create);
1185 if (optional_sem != NULL)
1187 rc = fsfilt_ext3_map_bm_inode_pages(inode, page, pages, blocks,
1189 if (optional_sem != NULL)
1195 int fsfilt_ext3_read(struct inode *inode, void *buf, int size, loff_t *offs)
1197 unsigned long block;
1198 struct buffer_head *bh;
1199 int err, blocksize, csize, boffs, osize = size;
1201 /* prevent reading after eof */
1203 if (i_size_read(inode) < *offs + size) {
1204 size = i_size_read(inode) - *offs;
1207 CERROR("size %llu is too short for read %u@%llu\n",
1208 i_size_read(inode), size, *offs);
1210 } else if (size == 0) {
1217 blocksize = 1 << inode->i_blkbits;
1220 block = *offs >> inode->i_blkbits;
1221 boffs = *offs & (blocksize - 1);
1222 csize = min(blocksize - boffs, size);
1223 bh = ext3_bread(NULL, inode, block, 0, &err);
1225 CERROR("can't read block: %d\n", err);
1229 memcpy(buf, bh->b_data + boffs, csize);
1238 EXPORT_SYMBOL(fsfilt_ext3_read);
1240 static int fsfilt_ext3_read_record(struct file * file, void *buf,
1241 int size, loff_t *offs)
1244 rc = fsfilt_ext3_read(file->f_dentry->d_inode, buf, size, offs);
1250 int fsfilt_ext3_write_handle(struct inode *inode, void *buf, int bufsize,
1251 loff_t *offs, handle_t *handle)
1253 struct buffer_head *bh = NULL;
1254 loff_t old_size = i_size_read(inode), offset = *offs;
1255 loff_t new_size = i_size_read(inode);
1256 unsigned long block;
1257 int err = 0, blocksize = 1 << inode->i_blkbits, size, boffs;
1259 while (bufsize > 0) {
1263 block = offset >> inode->i_blkbits;
1264 boffs = offset & (blocksize - 1);
1265 size = min(blocksize - boffs, bufsize);
1266 bh = ext3_bread(handle, inode, block, 1, &err);
1268 CERROR("can't read/create block: %d\n", err);
1272 err = ext3_journal_get_write_access(handle, bh);
1274 CERROR("journal_get_write_access() returned error %d\n",
1278 LASSERT(bh->b_data + boffs + size <= bh->b_data + bh->b_size);
1279 memcpy(bh->b_data + boffs, buf, size);
1280 err = ext3_journal_dirty_metadata(handle, bh);
1282 CERROR("journal_dirty_metadata() returned error %d\n",
1286 if (offset + size > new_size)
1287 new_size = offset + size;
1295 /* correct in-core and on-disk sizes */
1296 if (new_size > i_size_read(inode)) {
1298 if (new_size > i_size_read(inode))
1299 i_size_write(inode, new_size);
1300 if (i_size_read(inode) > EXT3_I(inode)->i_disksize)
1301 EXT3_I(inode)->i_disksize = i_size_read(inode);
1302 if (i_size_read(inode) > old_size)
1303 mark_inode_dirty(inode);
1311 EXPORT_SYMBOL(fsfilt_ext3_write_handle);
1313 static int fsfilt_ext3_write_record(struct file *file, void *buf, int bufsize,
1314 loff_t *offs, int force_sync)
1316 struct inode *inode = file->f_dentry->d_inode;
1318 int err, block_count = 0, blocksize;
1320 /* Determine how many transaction credits are needed */
1321 blocksize = 1 << inode->i_blkbits;
1322 block_count = (*offs & (blocksize - 1)) + bufsize;
1323 block_count = (block_count + blocksize - 1) >> inode->i_blkbits;
1325 handle = fsfilt_ext3_journal_start(inode,
1326 block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2);
1327 if (IS_ERR(handle)) {
1328 CERROR("can't start transaction for %d blocks (%d bytes)\n",
1329 block_count * FSFILT_DATA_TRANS_BLOCKS(inode->i_sb) + 2, bufsize);
1330 return PTR_ERR(handle);
1333 err = fsfilt_ext3_write_handle(inode, buf, bufsize, offs, handle);
1335 if (!err && force_sync)
1336 handle->h_sync = 1; /* recovery likes this */
1338 fsfilt_ext3_journal_stop(handle);
1343 static int fsfilt_ext3_setup(struct super_block *sb)
1345 #if ((LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,6)) && \
1346 defined(HAVE_QUOTA_SUPPORT)) || defined(S_PDIROPS)
1347 struct ext3_sb_info *sbi = EXT3_SB(sb);
1349 sbi->dx_lock = fsfilt_ext3_dx_lock;
1350 sbi->dx_unlock = fsfilt_ext3_dx_unlock;
1354 CWARN("Enabling PDIROPS\n");
1355 set_opt(sbi->s_mount_opt, PDIROPS);
1356 sb->s_flags |= S_PDIROPS;
1358 if (!EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1359 CWARN("filesystem doesn't have dir_index feature enabled\n");
1360 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,6)) && defined(HAVE_QUOTA_SUPPORT)
1361 /* enable journaled quota support */
1362 /* kfreed in ext3_put_super() */
1363 sbi->s_qf_names[USRQUOTA] = kstrdup("lquota.user.reserved", GFP_KERNEL);
1364 if (!sbi->s_qf_names[USRQUOTA])
1366 sbi->s_qf_names[GRPQUOTA] = kstrdup("lquota.group.reserved", GFP_KERNEL);
1367 if (!sbi->s_qf_names[GRPQUOTA]) {
1368 kfree(sbi->s_qf_names[USRQUOTA]);
1369 sbi->s_qf_names[USRQUOTA] = NULL;
1372 sbi->s_jquota_fmt = QFMT_VFS_V0;
1373 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,13))
1374 set_opt(sbi->s_mount_opt, QUOTA);
1380 /* If fso is NULL, op is FSFILT operation, otherwise op is number of fso
1381 objects. Logs is number of logfiles to update */
1382 static int fsfilt_ext3_get_op_len(int op, struct fsfilt_objinfo *fso, int logs)
1386 case FSFILT_OP_CREATE:
1387 /* directory leaf, index & indirect & EA*/
1388 return 4 + 3 * logs;
1389 case FSFILT_OP_UNLINK:
1395 struct super_block *sb = fso->fso_dentry->d_inode->i_sb;
1396 int blockpp = 1 << (CFS_PAGE_SHIFT - sb->s_blocksize_bits);
1397 int addrpp = EXT3_ADDR_PER_BLOCK(sb) * blockpp;
1398 for (i = 0; i < op; i++, fso++) {
1399 int nblocks = fso->fso_bufcnt * blockpp;
1400 int ndindirect = min(nblocks, addrpp + 1);
1401 int nindir = nblocks + ndindirect + 1;
1405 return needed + 3 * logs;
1411 #ifdef HAVE_QUOTA_SUPPORT
1412 #define DQINFO_COPY(out, in) \
1414 Q_COPY(out, in, dqi_bgrace); \
1415 Q_COPY(out, in, dqi_igrace); \
1416 Q_COPY(out, in, dqi_flags); \
1417 Q_COPY(out, in, dqi_valid); \
1420 #define DQBLK_COPY(out, in) \
1422 Q_COPY(out, in, dqb_bhardlimit); \
1423 Q_COPY(out, in, dqb_bsoftlimit); \
1424 Q_COPY(out, in, dqb_curspace); \
1425 Q_COPY(out, in, dqb_ihardlimit); \
1426 Q_COPY(out, in, dqb_isoftlimit); \
1427 Q_COPY(out, in, dqb_curinodes); \
1428 Q_COPY(out, in, dqb_btime); \
1429 Q_COPY(out, in, dqb_itime); \
1430 Q_COPY(out, in, dqb_valid); \
1433 static int fsfilt_ext3_quotactl(struct super_block *sb,
1434 struct obd_quotactl *oqc)
1436 int i, rc = 0, error = 0;
1437 struct quotactl_ops *qcop;
1438 struct if_dqinfo *info;
1439 struct if_dqblk *dqblk;
1445 OBD_ALLOC_PTR(info);
1448 OBD_ALLOC_PTR(dqblk);
1454 DQINFO_COPY(info, &oqc->qc_dqinfo);
1455 DQBLK_COPY(dqblk, &oqc->qc_dqblk);
1458 if (oqc->qc_cmd == Q_QUOTAON || oqc->qc_cmd == Q_QUOTAOFF) {
1459 for (i = 0; i < MAXQUOTAS; i++) {
1460 if (!Q_TYPESET(oqc, i))
1463 if (oqc->qc_cmd == Q_QUOTAON) {
1464 char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1466 LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1468 if (!qcop->quota_on)
1469 GOTO(out, rc = -ENOSYS);
1471 rc = qcop->quota_on(sb, i, QFMT_VFS_V0,
1473 } else if (oqc->qc_cmd == Q_QUOTAOFF) {
1474 if (!qcop->quota_off)
1475 GOTO(out, rc = -ENOSYS);
1476 rc = qcop->quota_off(sb, i);
1484 GOTO(out, rc ?: error);
1487 switch (oqc->qc_cmd) {
1490 if (!qcop->get_info)
1491 GOTO(out, rc = -ENOSYS);
1492 rc = qcop->get_info(sb, oqc->qc_type, info);
1496 if (!qcop->set_dqblk)
1497 GOTO(out, rc = -ENOSYS);
1498 rc = qcop->set_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1502 if (!qcop->get_dqblk)
1503 GOTO(out, rc = -ENOSYS);
1504 rc = qcop->get_dqblk(sb, oqc->qc_type, oqc->qc_id, dqblk);
1506 dqblk->dqb_valid = QIF_LIMITS | QIF_USAGE;
1509 if (!sb->s_qcop->quota_sync)
1510 GOTO(out, rc = -ENOSYS);
1511 qcop->quota_sync(sb, oqc->qc_type);
1514 CDEBUG(D_WARNING, "invalidating operational quota files\n");
1515 for (i = 0; i < MAXQUOTAS; i++) {
1517 char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1519 LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1521 if (!Q_TYPESET(oqc, i))
1524 fp = filp_open(name[i], O_CREAT | O_TRUNC | O_RDWR, 0644);
1527 CERROR("error invalidating operational quota file"
1528 " %s (rc:%d)\n", name[i], rc);
1536 CERROR("unsupported quotactl command: %d\n", oqc->qc_cmd);
1540 DQINFO_COPY(&oqc->qc_dqinfo, info);
1541 DQBLK_COPY(&oqc->qc_dqblk, dqblk);
1544 OBD_FREE_PTR(dqblk);
1547 CDEBUG(D_QUOTA, "quotactl command %#x, id %u, type %u "
1549 oqc->qc_cmd, oqc->qc_id, oqc->qc_type, rc);
1554 struct hlist_node dqb_hash; /** quotacheck hash */
1555 struct list_head dqb_list; /** in list also */
1556 qid_t dqb_id; /** uid/gid */
1557 short dqb_type; /** USRQUOTA/GRPQUOTA */
1558 qsize_t dqb_bhardlimit; /** block hard limit */
1559 qsize_t dqb_bsoftlimit; /** block soft limit */
1560 qsize_t dqb_curspace; /** current space */
1561 qsize_t dqb_ihardlimit; /** inode hard limit */
1562 qsize_t dqb_isoftlimit; /** inode soft limit */
1563 qsize_t dqb_curinodes; /** current inodes */
1564 __u64 dqb_btime; /** block grace time */
1565 __u64 dqb_itime; /** inode grace time */
1566 __u32 dqb_valid; /** flag for above fields */
1569 static inline unsigned int chkquot_hash(qid_t id, int type)
1570 __attribute__((__const__));
1572 static inline unsigned int chkquot_hash(qid_t id, int type)
1574 return (id * (MAXQUOTAS - type)) % NR_DQHASH;
1577 static inline struct chk_dqblk *
1578 find_chkquot(struct hlist_head *head, qid_t id, int type)
1580 struct hlist_node *node;
1581 struct chk_dqblk *cdqb;
1583 hlist_for_each(node, head) {
1584 cdqb = hlist_entry(node, struct chk_dqblk, dqb_hash);
1585 if (cdqb->dqb_id == id && cdqb->dqb_type == type)
1592 static struct chk_dqblk *alloc_chkquot(qid_t id, int type)
1594 struct chk_dqblk *cdqb;
1596 OBD_ALLOC_PTR(cdqb);
1598 INIT_HLIST_NODE(&cdqb->dqb_hash);
1599 INIT_LIST_HEAD(&cdqb->dqb_list);
1601 cdqb->dqb_type = type;
1607 static struct chk_dqblk *
1608 cqget(struct super_block *sb, struct hlist_head *hash, struct list_head *list,
1609 qid_t id, int type, int first_check)
1611 struct hlist_head *head = hash + chkquot_hash(id, type);
1612 struct if_dqblk dqb;
1613 struct chk_dqblk *cdqb;
1616 cdqb = find_chkquot(head, id, type);
1620 cdqb = alloc_chkquot(id, type);
1625 rc = sb->s_qcop->get_dqblk(sb, type, id, &dqb);
1627 CERROR("get_dqblk of id %u, type %d failed: %d\n",
1630 DQBLK_COPY(cdqb, &dqb);
1631 cdqb->dqb_curspace = 0;
1632 cdqb->dqb_curinodes = 0;
1636 hlist_add_head(&cdqb->dqb_hash, head);
1637 list_add_tail(&cdqb->dqb_list, list);
1642 static inline int quota_onoff(struct super_block *sb, int cmd, int type, int qfmt)
1644 struct obd_quotactl *oqctl;
1647 OBD_ALLOC_PTR(oqctl);
1651 oqctl->qc_cmd = cmd;
1652 oqctl->qc_id = qfmt;
1653 oqctl->qc_type = type;
1654 rc = fsfilt_ext3_quotactl(sb, oqctl);
1656 OBD_FREE_PTR(oqctl);
1660 static inline int read_old_dqinfo(struct super_block *sb, int type,
1661 struct if_dqinfo *dqinfo)
1663 struct obd_quotactl *oqctl;
1667 OBD_ALLOC_PTR(oqctl);
1671 oqctl->qc_cmd = Q_GETINFO;
1672 oqctl->qc_type = type;
1673 rc = fsfilt_ext3_quotactl(sb, oqctl);
1675 ((struct obd_dqinfo *)dqinfo)[type] = oqctl->qc_dqinfo;
1677 OBD_FREE_PTR(oqctl);
1681 static inline struct ext3_group_desc *
1682 get_group_desc(struct super_block *sb, int group)
1684 unsigned long desc_block, desc;
1685 struct ext3_group_desc *gdp;
1687 desc_block = group / EXT3_DESC_PER_BLOCK(sb);
1688 desc = group % EXT3_DESC_PER_BLOCK(sb);
1689 gdp = (struct ext3_group_desc *)
1690 EXT3_SB(sb)->s_group_desc[desc_block]->b_data;
1696 #ifndef HAVE_EXT4_LDISKFS
1697 static inline struct buffer_head *
1698 ext3_read_inode_bitmap(struct super_block *sb, unsigned long group)
1700 struct ext3_group_desc *desc;
1701 struct buffer_head *bh;
1703 desc = get_group_desc(sb, group);
1704 bh = sb_bread(sb, ext3_inode_bitmap(sb, desc));
1709 static inline struct inode *ext3_iget_inuse(struct super_block *sb,
1710 struct buffer_head *bitmap_bh,
1711 int index, unsigned long ino)
1713 struct inode *inode = NULL;
1716 if (ext3_test_bit(index, bitmap_bh->b_data))
1717 #ifdef HAVE_EXT4_LDISKFS
1718 inode = ext4_iget(sb, ino);
1720 /* Newer kernels return an error instead of a NULL pointer */
1723 inode = iget(sb, ino);
1729 struct hlist_head qckt_hash[NR_DQHASH]; /* quotacheck hash */
1730 struct list_head qckt_list; /* quotacheck list */
1731 int qckt_first_check[MAXQUOTAS]; /* 1 if no old quotafile */
1732 struct if_dqinfo qckt_dqinfo[MAXQUOTAS]; /* old dqinfo */
1735 static int add_inode_quota(struct inode *inode, struct qchk_ctxt *qctxt,
1736 struct obd_quotactl *oqc)
1738 struct chk_dqblk *cdqb[MAXQUOTAS] = { NULL, };
1740 qid_t qid[MAXQUOTAS];
1746 qid[USRQUOTA] = inode->i_uid;
1747 qid[GRPQUOTA] = inode->i_gid;
1749 if (S_ISDIR(inode->i_mode) ||
1750 S_ISREG(inode->i_mode) ||
1751 S_ISLNK(inode->i_mode))
1752 size = inode_get_bytes(inode);
1754 for (cnt = 0; cnt < MAXQUOTAS; cnt++) {
1755 if (!Q_TYPESET(oqc, cnt))
1758 cdqb[cnt] = cqget(inode->i_sb, qctxt->qckt_hash,
1759 &qctxt->qckt_list, qid[cnt], cnt,
1760 qctxt->qckt_first_check[cnt]);
1766 cdqb[cnt]->dqb_curspace += size;
1767 cdqb[cnt]->dqb_curinodes++;
1771 for (i = 0; i < cnt; i++) {
1772 if (!Q_TYPESET(oqc, i))
1775 cdqb[i]->dqb_curspace -= size;
1776 cdqb[i]->dqb_curinodes--;
1783 /* write dqinfo struct in a new quota file */
1784 static int v3_write_dqinfo(struct file *f, int type, struct if_dqinfo *info)
1786 struct v2_disk_dqinfo dqinfo;
1787 __u32 blocks = V2_DQTREEOFF + 1;
1788 loff_t offset = V2_DQINFOOFF;
1791 dqinfo.dqi_bgrace = cpu_to_le32(info->dqi_bgrace);
1792 dqinfo.dqi_igrace = cpu_to_le32(info->dqi_igrace);
1793 dqinfo.dqi_flags = cpu_to_le32(info->dqi_flags & DQF_MASK &
1796 dqinfo.dqi_bgrace = cpu_to_le32(MAX_DQ_TIME);
1797 dqinfo.dqi_igrace = cpu_to_le32(MAX_IQ_TIME);
1798 dqinfo.dqi_flags = 0;
1801 dqinfo.dqi_blocks = cpu_to_le32(blocks);
1802 dqinfo.dqi_free_blk = 0;
1803 dqinfo.dqi_free_entry = 0;
1805 return cfs_user_write(f, (char *)&dqinfo, sizeof(dqinfo), &offset);
1808 static int v3_write_dqheader(struct file *f, int type)
1810 static const __u32 quota_magics[] = V2_INITQMAGICS;
1811 static const __u32 quota_versions[] = V2_INITQVERSIONS_R1;
1812 struct v2_disk_dqheader dqhead;
1815 CLASSERT(ARRAY_SIZE(quota_magics) == ARRAY_SIZE(quota_versions));
1816 LASSERT(0 <= type && type < ARRAY_SIZE(quota_magics));
1818 dqhead.dqh_magic = cpu_to_le32(quota_magics[type]);
1819 dqhead.dqh_version = cpu_to_le32(quota_versions[type]);
1821 return cfs_user_write(f, (char *)&dqhead, sizeof(dqhead), &offset);
1824 static int create_new_quota_files(struct qchk_ctxt *qctxt,
1825 struct obd_quotactl *oqc)
1830 for (i = 0; i < MAXQUOTAS; i++) {
1831 struct if_dqinfo *info = qctxt->qckt_first_check[i]?
1832 NULL : &qctxt->qckt_dqinfo[i];
1834 const char *name[MAXQUOTAS] = LUSTRE_OPQFILES_NAMES_V2;
1836 if (!Q_TYPESET(oqc, i))
1839 LASSERT(oqc->qc_id == LUSTRE_QUOTA_V2);
1841 file = filp_open(name[i], O_RDWR | O_CREAT | O_TRUNC, 0644);
1844 CERROR("can't create %s file: rc = %d\n",
1849 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
1850 CERROR("file %s is not regular", name[i]);
1851 filp_close(file, 0);
1852 GOTO(out, rc = -EINVAL);
1855 DQUOT_DROP(file->f_dentry->d_inode);
1857 rc = v3_write_dqheader(file, i);
1859 filp_close(file, 0);
1863 rc = v3_write_dqinfo(file, i, info);
1864 filp_close(file, 0);
1874 static int commit_chkquot(struct super_block *sb, struct qchk_ctxt *qctxt,
1875 struct chk_dqblk *cdqb)
1877 struct obd_quotactl *oqc;
1886 now = cfs_time_current_sec();
1888 if (cdqb->dqb_bsoftlimit &&
1889 toqb(cdqb->dqb_curspace) >= cdqb->dqb_bsoftlimit &&
1892 now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_bgrace;
1894 if (cdqb->dqb_isoftlimit &&
1895 cdqb->dqb_curinodes >= cdqb->dqb_isoftlimit &&
1898 now + qctxt->qckt_dqinfo[cdqb->dqb_type].dqi_igrace;
1900 cdqb->dqb_valid = QIF_ALL;
1902 oqc->qc_cmd = Q_SETQUOTA;
1903 oqc->qc_type = cdqb->dqb_type;
1904 oqc->qc_id = cdqb->dqb_id;
1905 DQBLK_COPY(&oqc->qc_dqblk, cdqb);
1907 rc = fsfilt_ext3_quotactl(sb, oqc);
1912 static int prune_chkquots(struct super_block *sb,
1913 struct qchk_ctxt *qctxt, int error)
1915 struct chk_dqblk *cdqb, *tmp;
1918 list_for_each_entry_safe(cdqb, tmp, &qctxt->qckt_list, dqb_list) {
1920 rc = commit_chkquot(sb, qctxt, cdqb);
1924 hlist_del_init(&cdqb->dqb_hash);
1925 list_del(&cdqb->dqb_list);
1932 static int fsfilt_ext3_quotacheck(struct super_block *sb,
1933 struct obd_quotactl *oqc)
1935 struct ext3_sb_info *sbi = EXT3_SB(sb);
1937 struct qchk_ctxt *qctxt;
1938 struct buffer_head *bitmap_bh = NULL;
1940 struct inode *inode;
1944 /* turn on quota and read dqinfo if existed */
1945 OBD_ALLOC_PTR(qctxt);
1947 oqc->qc_stat = -ENOMEM;
1951 for (i = 0; i < NR_DQHASH; i++)
1952 INIT_HLIST_HEAD(&qctxt->qckt_hash[i]);
1953 INIT_LIST_HEAD(&qctxt->qckt_list);
1955 for (i = 0; i < MAXQUOTAS; i++) {
1956 if (!Q_TYPESET(oqc, i))
1959 rc = quota_onoff(sb, Q_QUOTAON, i, oqc->qc_id);
1960 if (!rc || rc == -EBUSY) {
1961 rc = read_old_dqinfo(sb, i, qctxt->qckt_dqinfo);
1964 } else if (rc == -ENOENT || rc == -EINVAL || rc == -EEXIST) {
1965 qctxt->qckt_first_check[i] = 1;
1971 /* check quota and update in hash */
1972 for (group = 0; group < sbi->s_groups_count; group++) {
1973 ino = group * sbi->s_inodes_per_group + 1;
1974 bitmap_bh = ext3_read_inode_bitmap(sb, group);
1976 CERROR("ext3_read_inode_bitmap group %d failed", group);
1977 GOTO(out, rc = -EIO);
1980 for (i = 0; i < sbi->s_inodes_per_group; i++, ino++) {
1981 if (ino < sbi->s_first_ino)
1984 inode = ext3_iget_inuse(sb, bitmap_bh, i, ino);
1985 rc = add_inode_quota(inode, qctxt, oqc);
1996 /* read old quota limits from old quota file. (only for the user
1997 * has limits but hasn't file) */
1998 #ifdef HAVE_QUOTA_SUPPORT
1999 for (i = 0; i < MAXQUOTAS; i++) {
2000 struct list_head id_list;
2001 struct dquot_id *dqid, *tmp;
2003 if (!Q_TYPESET(oqc, i))
2006 if (qctxt->qckt_first_check[i])
2010 LASSERT(sb_dqopt(sb)->files[i] != NULL);
2011 INIT_LIST_HEAD(&id_list);
2012 #ifndef KERNEL_SUPPORTS_QUOTA_READ
2013 rc = lustre_get_qids(sb_dqopt(sb)->files[i], NULL, i, &id_list);
2015 rc = lustre_get_qids(NULL, sb_dqopt(sb)->files[i], i, &id_list);
2018 CERROR("read old limits failed. (rc:%d)\n", rc);
2020 list_for_each_entry_safe(dqid, tmp, &id_list, di_link) {
2021 list_del_init(&dqid->di_link);
2024 cqget(sb, qctxt->qckt_hash, &qctxt->qckt_list,
2026 qctxt->qckt_first_check[i]);
2031 /* turn off quota cause we are to dump chk_dqblk to files */
2032 quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type, oqc->qc_id);
2034 rc = create_new_quota_files(qctxt, oqc);
2038 /* we use vfs functions to set dqblk, so turn quota on */
2039 rc = quota_onoff(sb, Q_QUOTAON, oqc->qc_type, oqc->qc_id);
2041 /* dump and free chk_dqblk */
2042 rc = prune_chkquots(sb, qctxt, rc);
2043 OBD_FREE_PTR(qctxt);
2045 /* turn off quota, `lfs quotacheck` will turn on when all
2046 * nodes quotacheck finish. */
2047 quota_onoff(sb, Q_QUOTAOFF, oqc->qc_type, oqc->qc_id);
2051 CERROR("quotacheck failed: rc = %d\n", rc);
2056 static int fsfilt_ext3_quotainfo(struct lustre_quota_info *lqi, int type,
2062 if (lqi->qi_files[type] == NULL) {
2063 CERROR("operate qinfo before it's enabled!\n");
2069 rc = lustre_check_quota_file(lqi, type);
2072 rc = lustre_read_quota_info(lqi, type);
2075 rc = lustre_write_quota_info(lqi, type);
2077 case QFILE_INIT_INFO:
2078 rc = lustre_init_quota_info(lqi, type);
2082 CERROR("quota CONVERT command is not supported\n");
2086 CERROR("Unsupported admin quota file cmd %d\n"
2087 "Are lquota.ko and fsfilt_ldiskfs.ko modules in sync?\n",
2094 static int fsfilt_ext3_qids(struct file *file, struct inode *inode, int type,
2095 struct list_head *list)
2097 return lustre_get_qids(file, inode, type, list);
2100 static int fsfilt_ext3_dquot(struct lustre_dquot *dquot, int cmd)
2105 if (dquot->dq_info->qi_files[dquot->dq_type] == NULL) {
2106 CERROR("operate dquot before it's enabled!\n");
2111 case QFILE_RD_DQUOT:
2112 rc = lustre_read_dquot(dquot);
2114 case QFILE_WR_DQUOT:
2115 if (dquot->dq_dqb.dqb_ihardlimit ||
2116 dquot->dq_dqb.dqb_isoftlimit ||
2117 dquot->dq_dqb.dqb_bhardlimit ||
2118 dquot->dq_dqb.dqb_bsoftlimit)
2119 clear_bit(DQ_FAKE_B, &dquot->dq_flags);
2121 set_bit(DQ_FAKE_B, &dquot->dq_flags);
2123 rc = lustre_commit_dquot(dquot);
2128 CERROR("Unsupported admin quota file cmd %d\n", cmd);
2135 static int fsfilt_ext3_get_mblk(struct super_block *sb, int *count,
2136 struct inode *inode, int frags)
2138 #ifdef EXT3_EXT_HAS_NO_TREE
2139 struct ext3_ext_base *base = inode;
2141 struct ext3_extents_tree tree;
2142 struct ext3_ext_base *base = &tree;
2144 ext3_init_tree_desc(base, inode);
2146 /* for an ost_write request, it needs <#fragments> * <tree depth + 1>
2147 * metablocks at maxium b=16542 */
2148 *count = frags * (EXT_DEPTH(base) + 1) * EXT3_BLOCK_SIZE(sb);
2154 lvfs_sbdev_type fsfilt_ext3_journal_sbdev(struct super_block *sb)
2156 return (EXT3_SB(sb)->journal_bdev);
2158 EXPORT_SYMBOL(fsfilt_ext3_journal_sbdev);
2160 static struct fsfilt_operations fsfilt_ext3_ops = {
2162 .fs_owner = THIS_MODULE,
2163 .fs_getlabel = fsfilt_ext3_get_label,
2164 .fs_setlabel = fsfilt_ext3_set_label,
2165 .fs_uuid = fsfilt_ext3_uuid,
2166 .fs_start = fsfilt_ext3_start,
2167 .fs_brw_start = fsfilt_ext3_brw_start,
2168 .fs_extend = fsfilt_ext3_extend,
2169 .fs_commit = fsfilt_ext3_commit,
2170 .fs_commit_async = fsfilt_ext3_commit_async,
2171 .fs_commit_wait = fsfilt_ext3_commit_wait,
2172 .fs_setattr = fsfilt_ext3_setattr,
2173 .fs_iocontrol = fsfilt_ext3_iocontrol,
2174 .fs_set_md = fsfilt_ext3_set_md,
2175 .fs_get_md = fsfilt_ext3_get_md,
2176 .fs_readpage = fsfilt_ext3_readpage,
2177 .fs_add_journal_cb = fsfilt_ext3_add_journal_cb,
2178 .fs_statfs = fsfilt_ext3_statfs,
2179 .fs_sync = fsfilt_ext3_sync,
2180 .fs_map_inode_pages = fsfilt_ext3_map_inode_pages,
2181 .fs_write_record = fsfilt_ext3_write_record,
2182 .fs_read_record = fsfilt_ext3_read_record,
2183 .fs_setup = fsfilt_ext3_setup,
2184 .fs_send_bio = fsfilt_ext3_send_bio,
2185 .fs_get_op_len = fsfilt_ext3_get_op_len,
2186 #ifdef HAVE_DISK_INODE_VERSION
2187 .fs_get_version = fsfilt_ext3_get_version,
2188 .fs_set_version = fsfilt_ext3_set_version,
2190 #ifdef HAVE_QUOTA_SUPPORT
2191 .fs_quotactl = fsfilt_ext3_quotactl,
2192 .fs_quotacheck = fsfilt_ext3_quotacheck,
2193 .fs_quotainfo = fsfilt_ext3_quotainfo,
2194 .fs_qids = fsfilt_ext3_qids,
2195 .fs_dquot = fsfilt_ext3_dquot,
2196 .fs_get_mblk = fsfilt_ext3_get_mblk,
2198 .fs_journal_sbdev = fsfilt_ext3_journal_sbdev,
2201 static int __init fsfilt_ext3_init(void)
2205 fcb_cache = cfs_mem_cache_create("fsfilt_ext3_fcb",
2206 sizeof(struct fsfilt_cb_data), 0, 0);
2208 CERROR("error allocating fsfilt journal callback cache\n");
2209 GOTO(out, rc = -ENOMEM);
2212 rc = fsfilt_register_ops(&fsfilt_ext3_ops);
2215 int err = cfs_mem_cache_destroy(fcb_cache);
2216 LASSERTF(err == 0, "error destroying new cache: rc %d\n", err);
2222 static void __exit fsfilt_ext3_exit(void)
2226 fsfilt_unregister_ops(&fsfilt_ext3_ops);
2227 rc = cfs_mem_cache_destroy(fcb_cache);
2228 LASSERTF(rc == 0, "couldn't destroy fcb_cache slab\n");
2231 module_init(fsfilt_ext3_init);
2232 module_exit(fsfilt_ext3_exit);
2234 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2235 MODULE_DESCRIPTION("Lustre ext3 Filesystem Helper v0.1");
2236 MODULE_LICENSE("GPL");