4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, Intel Corporation.
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 <ext4/ext4.h>
49 #include <ext4/ext4_jbd2.h>
50 #include <linux/version.h>
51 #include <linux/bitops.h>
52 #include <linux/quota.h>
54 #include <libcfs/libcfs.h>
55 #include <lustre_fsfilt.h>
57 #include <linux/lustre_compat25.h>
58 #include <linux/lprocfs_status.h>
60 #include <ext4/ext4_extents.h>
62 /* for kernels 2.6.18 and later */
63 #define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS(sb)
65 #define fsfilt_ext3_ext_insert_extent(handle, inode, path, newext, flag) \
66 ext3_ext_insert_extent(handle, inode, path, newext, flag)
68 #define ext3_mb_discard_inode_preallocations(inode) \
69 ext3_discard_preallocations(inode)
71 #define fsfilt_log_start_commit(journal, tid) jbd2_log_start_commit(journal, tid)
72 #define fsfilt_log_wait_commit(journal, tid) jbd2_log_wait_commit(journal, tid)
74 static cfs_mem_cache_t *fcb_cache;
76 struct fsfilt_cb_data {
77 struct ext4_journal_cb_entry cb_jcb; /* private data - MUST BE FIRST */
78 fsfilt_cb_t cb_func; /* MDS/OBD completion function */
79 struct obd_device *cb_obd; /* MDS/OBD completion device */
80 __u64 cb_last_rcvd; /* MDS/OST last committed operation */
81 void *cb_data; /* MDS/OST completion function data */
84 static char *fsfilt_ext3_get_label(struct super_block *sb)
86 return EXT3_SB(sb)->s_es->s_volume_name;
89 /* kernel has ext4_blocks_for_truncate since linux-3.1.1 */
90 #ifdef HAVE_BLOCKS_FOR_TRUNCATE
91 # include <ext4/truncate.h>
93 static inline unsigned long ext4_blocks_for_truncate(struct inode *inode)
97 needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
100 if (needed > EXT4_MAX_TRANS_DATA)
101 needed = EXT4_MAX_TRANS_DATA;
102 return EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
107 * We don't currently need any additional blocks for rmdir and
108 * unlink transactions because we are storing the OST oa_id inside
109 * the inode (which we will be changing anyways as part of this
112 static void *fsfilt_ext3_start(struct inode *inode, int op, void *desc_private,
115 /* For updates to the last received file */
116 int nblocks = FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
120 if (current->journal_info) {
121 CDEBUG(D_INODE, "increasing refcount on %p\n",
122 current->journal_info);
127 case FSFILT_OP_UNLINK:
128 /* delete one file + create/update logs for each stripe */
129 nblocks += EXT3_DELETE_TRANS_BLOCKS(inode->i_sb);
130 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
131 FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
133 case FSFILT_OP_CANCEL_UNLINK:
136 /* blocks for log header bitmap update OR
137 * blocks for catalog header bitmap update + unlink of logs +
138 * blocks for delete the inode (include blocks truncating). */
139 nblocks = (LLOG_CHUNK_SIZE >> inode->i_blkbits) +
140 EXT3_DELETE_TRANS_BLOCKS(inode->i_sb) +
141 ext4_blocks_for_truncate(inode) + 3;
143 default: CERROR("unknown transaction start op %d\n", op);
147 LASSERT(current->journal_info == desc_private);
148 journal = EXT3_SB(inode->i_sb)->s_journal;
149 if (nblocks > journal->j_max_transaction_buffers) {
150 CWARN("too many credits %d for op %ux%u using %d instead\n",
151 nblocks, op, logs, journal->j_max_transaction_buffers);
152 nblocks = journal->j_max_transaction_buffers;
156 LASSERTF(nblocks > 0, "can't start %d credit transaction\n", nblocks);
157 handle = ext3_journal_start(inode, nblocks);
160 LASSERT(current->journal_info == handle);
162 CERROR("error starting handle for op %u (%u credits): rc %ld\n",
163 op, nblocks, PTR_ERR(handle));
167 static int fsfilt_ext3_commit(struct inode *inode, void *h, int force_sync)
170 handle_t *handle = h;
172 LASSERT(current->journal_info == handle);
174 handle->h_sync = 1; /* recovery likes this */
176 rc = ext3_journal_stop(handle);
181 #ifndef EXT3_EXTENTS_FL
182 #define EXT3_EXTENTS_FL 0x00080000 /* Inode uses extents */
186 #define EXT_ASSERT(cond) BUG_ON(!(cond))
189 #define EXT_GENERATION(inode) (EXT4_I(inode)->i_ext_generation)
190 #define ext3_ext_base inode
191 #define ext3_ext_base2inode(inode) (inode)
192 #define EXT_DEPTH(inode) ext_depth(inode)
193 #define fsfilt_ext3_ext_walk_space(inode, block, num, cb, cbdata) \
194 ext3_ext_walk_space(inode, block, num, cb, cbdata);
197 unsigned long *blocks;
204 static long ext3_ext_find_goal(struct inode *inode, struct ext3_ext_path *path,
205 unsigned long block, int *aflags)
207 struct ext3_inode_info *ei = EXT3_I(inode);
208 unsigned long bg_start;
209 unsigned long colour;
213 struct ext3_extent *ex;
214 depth = path->p_depth;
216 /* try to predict block placement */
217 if ((ex = path[depth].p_ext))
218 return ext_pblock(ex) + (block - le32_to_cpu(ex->ee_block));
220 /* it looks index is empty
221 * try to find starting from index itself */
222 if (path[depth].p_bh)
223 return path[depth].p_bh->b_blocknr;
226 /* OK. use inode's group */
227 bg_start = (ei->i_block_group * EXT3_BLOCKS_PER_GROUP(inode->i_sb)) +
228 le32_to_cpu(EXT3_SB(inode->i_sb)->s_es->s_first_data_block);
229 colour = (current->pid % 16) *
230 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
231 return bg_start + colour + block;
234 #define ll_unmap_underlying_metadata(sb, blocknr) \
235 unmap_underlying_metadata((sb)->s_bdev, blocknr)
237 #ifndef EXT3_MB_HINT_GROUP_ALLOC
238 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
239 struct ext3_ext_path *path, unsigned long block,
240 unsigned long *count, int *err)
242 unsigned long pblock, goal;
244 struct inode *inode = ext3_ext_base2inode(base);
246 goal = ext3_ext_find_goal(inode, path, block, &aflags);
247 aflags |= 2; /* block have been already reserved */
248 pblock = ext3_mb_new_blocks(handle, inode, goal, count, aflags, err);
253 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
254 struct ext3_ext_path *path, unsigned long block,
255 unsigned long *count, int *err)
257 struct inode *inode = ext3_ext_base2inode(base);
258 struct ext3_allocation_request ar;
259 unsigned long pblock;
262 /* find neighbour allocated blocks */
264 *err = ext3_ext_search_left(base, path, &ar.lleft, &ar.pleft);
268 *err = ext3_ext_search_right(base, path, &ar.lright, &ar.pright);
272 /* allocate new block */
273 ar.goal = ext3_ext_find_goal(inode, path, block, &aflags);
277 ar.flags = EXT3_MB_HINT_DATA;
278 pblock = ext3_mb_new_blocks(handle, &ar, err);
284 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
285 struct ext3_ext_path *path,
286 struct ext3_ext_cache *cex,
287 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
288 struct ext3_extent *ex,
292 struct bpointers *bp = cbdata;
293 struct inode *inode = ext3_ext_base2inode(base);
294 struct ext3_extent nex;
295 unsigned long pblock;
301 #ifdef EXT3_EXT_CACHE_EXTENT
302 if (cex->ec_type == EXT3_EXT_CACHE_EXTENT)
304 if ((cex->ec_len != 0) && (cex->ec_start != 0))
311 if (bp->create == 0) {
313 if (cex->ec_block < bp->start)
314 i = bp->start - cex->ec_block;
315 if (i >= cex->ec_len)
316 CERROR("nothing to do?! i = %d, e_num = %u\n",
318 for (; i < cex->ec_len && bp->num; i++) {
328 tgen = EXT_GENERATION(base);
329 count = ext3_ext_calc_credits_for_insert(base, path);
331 handle = ext3_journal_start(inode, count+EXT3_ALLOC_NEEDED+1);
332 if (IS_ERR(handle)) {
333 return PTR_ERR(handle);
336 if (tgen != EXT_GENERATION(base)) {
337 /* the tree has changed. so path can be invalid at moment */
338 ext3_journal_stop(handle);
342 /* In 2.6.32 kernel, ext4_ext_walk_space()'s callback func is not
343 * protected by i_data_sem as whole. so we patch it to store
344 * generation to path and now verify the tree hasn't changed */
345 down_write((&EXT4_I(inode)->i_data_sem));
347 /* validate extent, make sure the extent tree does not changed */
348 if (EXT_GENERATION(base) != path[0].p_generation) {
349 /* cex is invalid, try again */
350 up_write(&EXT4_I(inode)->i_data_sem);
351 ext3_journal_stop(handle);
356 pblock = new_blocks(handle, base, path, cex->ec_block, &count, &err);
359 EXT_ASSERT(count <= cex->ec_len);
361 /* insert new extent */
362 nex.ee_block = cpu_to_le32(cex->ec_block);
363 ext3_ext_store_pblock(&nex, pblock);
364 nex.ee_len = cpu_to_le16(count);
365 err = fsfilt_ext3_ext_insert_extent(handle, base, path, &nex, 0);
367 /* free data blocks we just allocated */
368 /* not a good idea to call discard here directly,
369 * but otherwise we'd need to call it every free() */
370 #ifdef EXT3_MB_HINT_GROUP_ALLOC
371 ext3_mb_discard_inode_preallocations(inode);
373 ext3_free_blocks(handle, inode, ext_pblock(&nex),
374 cpu_to_le16(nex.ee_len), 0);
379 * Putting len of the actual extent we just inserted,
380 * we are asking ext3_ext_walk_space() to continue
381 * scaning after that block
383 cex->ec_len = le16_to_cpu(nex.ee_len);
384 cex->ec_start = ext_pblock(&nex);
385 BUG_ON(le16_to_cpu(nex.ee_len) == 0);
386 BUG_ON(le32_to_cpu(nex.ee_block) != cex->ec_block);
389 up_write((&EXT4_I(inode)->i_data_sem));
390 ext3_journal_stop(handle);
395 CERROR("hmm. why do we find this extent?\n");
396 CERROR("initial space: %lu:%u\n",
397 bp->start, bp->init_num);
398 #ifdef EXT3_EXT_CACHE_EXTENT
399 CERROR("current extent: %u/%u/%llu %d\n",
400 cex->ec_block, cex->ec_len,
401 (unsigned long long)cex->ec_start,
404 CERROR("current extent: %u/%u/%llu\n",
405 cex->ec_block, cex->ec_len,
406 (unsigned long long)cex->ec_start);
410 if (cex->ec_block < bp->start)
411 i = bp->start - cex->ec_block;
412 if (i >= cex->ec_len)
413 CERROR("nothing to do?! i = %d, e_num = %u\n",
415 for (; i < cex->ec_len && bp->num; i++) {
416 *(bp->blocks) = cex->ec_start + i;
417 #ifdef EXT3_EXT_CACHE_EXTENT
418 if (cex->ec_type != EXT3_EXT_CACHE_EXTENT)
420 if ((cex->ec_len == 0) || (cex->ec_start == 0))
423 /* unmap any possible underlying metadata from
424 * the block device mapping. bug 6998. */
425 ll_unmap_underlying_metadata(inode->i_sb,
436 int fsfilt_map_nblocks(struct inode *inode, unsigned long block,
437 unsigned long num, unsigned long *blocks,
440 struct ext3_ext_base *base = inode;
444 CDEBUG(D_OTHER, "blocks %lu-%lu requested for inode %u\n",
445 block, block + num - 1, (unsigned) inode->i_ino);
449 bp.init_num = bp.num = num;
452 err = fsfilt_ext3_ext_walk_space(base, block, num,
453 ext3_ext_new_extent_cb, &bp);
454 ext3_ext_invalidate_cache(base);
459 int fsfilt_ext3_map_ext_inode_pages(struct inode *inode, struct page **page,
460 int pages, unsigned long *blocks,
463 int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
465 struct page *fp = NULL;
468 CDEBUG(D_OTHER, "inode %lu: map %d pages from %lu\n",
469 inode->i_ino, pages, (*page)->index);
471 /* pages are sorted already. so, we just have to find
472 * contig. space and process them properly */
475 /* start new extent */
480 } else if (fp->index + clen == (*page)->index) {
481 /* continue the extent */
488 /* process found extent */
489 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
490 clen * blocks_per_page, blocks,
495 /* look for next extent */
497 blocks += blocks_per_page * clen;
501 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
502 clen * blocks_per_page, blocks,
508 extern int ext3_map_inode_page(struct inode *inode, struct page *page,
509 unsigned long *blocks, int create);
510 int fsfilt_ext3_map_bm_inode_pages(struct inode *inode, struct page **page,
511 int pages, unsigned long *blocks,
514 int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
518 for (i = 0, b = blocks; i < pages; i++, page++) {
519 rc = ext3_map_inode_page(inode, *page, b, create);
521 CERROR("ino %lu, blk %lu create %d: rc %d\n",
522 inode->i_ino, *b, create, rc);
526 b += blocks_per_page;
531 int fsfilt_ext3_map_inode_pages(struct inode *inode, struct page **page,
532 int pages, unsigned long *blocks,
533 int create, struct mutex *optional_mutex)
537 if (EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL) {
538 rc = fsfilt_ext3_map_ext_inode_pages(inode, page, pages,
542 if (optional_mutex != NULL)
543 mutex_lock(optional_mutex);
544 rc = fsfilt_ext3_map_bm_inode_pages(inode, page, pages, blocks, create);
545 if (optional_mutex != NULL)
546 mutex_unlock(optional_mutex);
551 int fsfilt_ext3_read(struct inode *inode, void *buf, int size, loff_t *offs)
554 struct buffer_head *bh;
555 int err, blocksize, csize, boffs, osize = size;
557 /* prevent reading after eof */
558 spin_lock(&inode->i_lock);
559 if (i_size_read(inode) < *offs + size) {
560 size = i_size_read(inode) - *offs;
561 spin_unlock(&inode->i_lock);
563 CDEBUG(D_EXT2, "size %llu is too short for read @%llu\n",
564 i_size_read(inode), *offs);
566 } else if (size == 0) {
570 spin_unlock(&inode->i_lock);
573 blocksize = 1 << inode->i_blkbits;
576 block = *offs >> inode->i_blkbits;
577 boffs = *offs & (blocksize - 1);
578 csize = min(blocksize - boffs, size);
579 bh = ext3_bread(NULL, inode, block, 0, &err);
581 CERROR("can't read block: %d\n", err);
585 memcpy(buf, bh->b_data + boffs, csize);
594 EXPORT_SYMBOL(fsfilt_ext3_read);
596 static int fsfilt_ext3_read_record(struct file * file, void *buf,
597 int size, loff_t *offs)
600 rc = fsfilt_ext3_read(file->f_dentry->d_inode, buf, size, offs);
606 int fsfilt_ext3_write_handle(struct inode *inode, void *buf, int bufsize,
607 loff_t *offs, handle_t *handle)
609 struct buffer_head *bh = NULL;
610 loff_t old_size = i_size_read(inode), offset = *offs;
611 loff_t new_size = i_size_read(inode);
613 int err = 0, blocksize = 1 << inode->i_blkbits, size, boffs;
615 while (bufsize > 0) {
619 block = offset >> inode->i_blkbits;
620 boffs = offset & (blocksize - 1);
621 size = min(blocksize - boffs, bufsize);
622 bh = ext3_bread(handle, inode, block, 1, &err);
624 CERROR("can't read/create block: %d\n", err);
628 err = ext3_journal_get_write_access(handle, bh);
630 CERROR("journal_get_write_access() returned error %d\n",
634 LASSERT(bh->b_data + boffs + size <= bh->b_data + bh->b_size);
635 memcpy(bh->b_data + boffs, buf, size);
636 err = ext3_journal_dirty_metadata(handle, bh);
638 CERROR("journal_dirty_metadata() returned error %d\n",
642 if (offset + size > new_size)
643 new_size = offset + size;
651 /* correct in-core and on-disk sizes */
652 if (new_size > i_size_read(inode)) {
653 spin_lock(&inode->i_lock);
654 if (new_size > i_size_read(inode))
655 i_size_write(inode, new_size);
656 if (i_size_read(inode) > EXT3_I(inode)->i_disksize)
657 EXT3_I(inode)->i_disksize = i_size_read(inode);
658 if (i_size_read(inode) > old_size) {
659 spin_unlock(&inode->i_lock);
660 mark_inode_dirty(inode);
662 spin_unlock(&inode->i_lock);
670 EXPORT_SYMBOL(fsfilt_ext3_write_handle);
672 static int fsfilt_ext3_write_record(struct file *file, void *buf, int bufsize,
673 loff_t *offs, int force_sync)
675 struct inode *inode = file->f_dentry->d_inode;
677 int err, block_count = 0, blocksize;
679 /* Determine how many transaction credits are needed */
680 blocksize = 1 << inode->i_blkbits;
681 block_count = (*offs & (blocksize - 1)) + bufsize;
682 block_count = (block_count + blocksize - 1) >> inode->i_blkbits;
684 handle = ext3_journal_start(inode,
685 block_count * EXT3_DATA_TRANS_BLOCKS(inode->i_sb) + 2);
686 if (IS_ERR(handle)) {
687 CERROR("can't start transaction for %d blocks (%d bytes)\n",
688 block_count * EXT3_DATA_TRANS_BLOCKS(inode->i_sb) + 2,
690 return PTR_ERR(handle);
693 err = fsfilt_ext3_write_handle(inode, buf, bufsize, offs, handle);
695 if (!err && force_sync)
696 handle->h_sync = 1; /* recovery likes this */
698 ext3_journal_stop(handle);
703 static int fsfilt_ext3_setup(struct super_block *sb)
705 if (!EXT3_HAS_COMPAT_FEATURE(sb,
706 EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
707 CERROR("ext3 mounted without journal\n");
712 CWARN("Enabling PDIROPS\n");
713 set_opt(EXT3_SB(sb)->s_mount_opt, PDIROPS);
714 sb->s_flags |= S_PDIROPS;
716 if (!EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
717 CWARN("filesystem doesn't have dir_index feature enabled\n");
720 static struct fsfilt_operations fsfilt_ext3_ops = {
722 .fs_owner = THIS_MODULE,
723 .fs_getlabel = fsfilt_ext3_get_label,
724 .fs_start = fsfilt_ext3_start,
725 .fs_commit = fsfilt_ext3_commit,
726 .fs_map_inode_pages = fsfilt_ext3_map_inode_pages,
727 .fs_write_record = fsfilt_ext3_write_record,
728 .fs_read_record = fsfilt_ext3_read_record,
729 .fs_setup = fsfilt_ext3_setup,
732 static int __init fsfilt_ext3_init(void)
736 fcb_cache = cfs_mem_cache_create("fsfilt_ext3_fcb",
737 sizeof(struct fsfilt_cb_data), 0, 0);
739 CERROR("error allocating fsfilt journal callback cache\n");
740 GOTO(out, rc = -ENOMEM);
743 rc = fsfilt_register_ops(&fsfilt_ext3_ops);
746 int err = cfs_mem_cache_destroy(fcb_cache);
747 LASSERTF(err == 0, "error destroying new cache: rc %d\n", err);
753 static void __exit fsfilt_ext3_exit(void)
757 fsfilt_unregister_ops(&fsfilt_ext3_ops);
758 rc = cfs_mem_cache_destroy(fcb_cache);
759 LASSERTF(rc == 0, "couldn't destroy fcb_cache slab\n");
762 module_init(fsfilt_ext3_init);
763 module_exit(fsfilt_ext3_exit);
765 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
766 MODULE_DESCRIPTION("Lustre ext3 Filesystem Helper v0.1");
767 MODULE_LICENSE("GPL");