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 <ldiskfs/ldiskfs_config.h>
49 #include <ext4/ext4.h>
50 #include <ext4/ext4_jbd2.h>
51 #include <linux/version.h>
52 #include <linux/bitops.h>
53 #include <linux/quota.h>
55 #include <libcfs/libcfs.h>
56 #include <lustre_fsfilt.h>
58 #include <linux/lustre_compat25.h>
59 #include <linux/lprocfs_status.h>
61 #include <ext4/ext4_extents.h>
63 /* for kernels 2.6.18 and later */
64 #define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS(sb)
66 #define fsfilt_ext3_ext_insert_extent(handle, inode, path, newext, flag) \
67 ext3_ext_insert_extent(handle, inode, path, newext, flag)
69 #define ext3_mb_discard_inode_preallocations(inode) \
70 ext3_discard_preallocations(inode)
72 #define fsfilt_log_start_commit(journal, tid) jbd2_log_start_commit(journal, tid)
73 #define fsfilt_log_wait_commit(journal, tid) jbd2_log_wait_commit(journal, tid)
75 static cfs_mem_cache_t *fcb_cache;
77 struct fsfilt_cb_data {
78 struct ext4_journal_cb_entry cb_jcb; /* private data - MUST BE FIRST */
79 fsfilt_cb_t cb_func; /* MDS/OBD completion function */
80 struct obd_device *cb_obd; /* MDS/OBD completion device */
81 __u64 cb_last_rcvd; /* MDS/OST last committed operation */
82 void *cb_data; /* MDS/OST completion function data */
85 static char *fsfilt_ext3_get_label(struct super_block *sb)
87 return EXT3_SB(sb)->s_es->s_volume_name;
90 /* kernel has ext4_blocks_for_truncate since linux-3.1.1 */
91 #ifdef HAVE_BLOCKS_FOR_TRUNCATE
92 # include <ext4/truncate.h>
94 static inline unsigned long ext4_blocks_for_truncate(struct inode *inode)
98 needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
101 if (needed > EXT4_MAX_TRANS_DATA)
102 needed = EXT4_MAX_TRANS_DATA;
103 return EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
108 * We don't currently need any additional blocks for rmdir and
109 * unlink transactions because we are storing the OST oa_id inside
110 * the inode (which we will be changing anyways as part of this
113 static void *fsfilt_ext3_start(struct inode *inode, int op, void *desc_private,
116 /* For updates to the last received file */
117 int nblocks = FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
121 if (current->journal_info) {
122 CDEBUG(D_INODE, "increasing refcount on %p\n",
123 current->journal_info);
128 case FSFILT_OP_UNLINK:
129 /* delete one file + create/update logs for each stripe */
130 nblocks += EXT3_DELETE_TRANS_BLOCKS(inode->i_sb);
131 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
132 FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
134 case FSFILT_OP_CANCEL_UNLINK:
137 /* blocks for log header bitmap update OR
138 * blocks for catalog header bitmap update + unlink of logs +
139 * blocks for delete the inode (include blocks truncating). */
140 nblocks = (LLOG_CHUNK_SIZE >> inode->i_blkbits) +
141 EXT3_DELETE_TRANS_BLOCKS(inode->i_sb) +
142 ext4_blocks_for_truncate(inode) + 3;
144 default: CERROR("unknown transaction start op %d\n", op);
148 LASSERT(current->journal_info == desc_private);
149 journal = EXT3_SB(inode->i_sb)->s_journal;
150 if (nblocks > journal->j_max_transaction_buffers) {
151 CWARN("too many credits %d for op %ux%u using %d instead\n",
152 nblocks, op, logs, journal->j_max_transaction_buffers);
153 nblocks = journal->j_max_transaction_buffers;
157 LASSERTF(nblocks > 0, "can't start %d credit transaction\n", nblocks);
158 handle = ext3_journal_start(inode, nblocks);
161 LASSERT(current->journal_info == handle);
163 CERROR("error starting handle for op %u (%u credits): rc %ld\n",
164 op, nblocks, PTR_ERR(handle));
168 static int fsfilt_ext3_commit(struct inode *inode, void *h, int force_sync)
171 handle_t *handle = h;
173 LASSERT(current->journal_info == handle);
175 handle->h_sync = 1; /* recovery likes this */
177 rc = ext3_journal_stop(handle);
182 #ifndef EXT3_EXTENTS_FL
183 #define EXT3_EXTENTS_FL 0x00080000 /* Inode uses extents */
187 #define EXT_ASSERT(cond) BUG_ON(!(cond))
190 #define EXT_GENERATION(inode) (EXT4_I(inode)->i_ext_generation)
191 #define ext3_ext_base inode
192 #define ext3_ext_base2inode(inode) (inode)
193 #define EXT_DEPTH(inode) ext_depth(inode)
194 #define fsfilt_ext3_ext_walk_space(inode, block, num, cb, cbdata) \
195 ext3_ext_walk_space(inode, block, num, cb, cbdata);
198 unsigned long *blocks;
205 static long ext3_ext_find_goal(struct inode *inode, struct ext3_ext_path *path,
206 unsigned long block, int *aflags)
208 struct ext3_inode_info *ei = EXT3_I(inode);
209 unsigned long bg_start;
210 unsigned long colour;
214 struct ext3_extent *ex;
215 depth = path->p_depth;
217 /* try to predict block placement */
218 if ((ex = path[depth].p_ext))
219 return ext_pblock(ex) + (block - le32_to_cpu(ex->ee_block));
221 /* it looks index is empty
222 * try to find starting from index itself */
223 if (path[depth].p_bh)
224 return path[depth].p_bh->b_blocknr;
227 /* OK. use inode's group */
228 bg_start = (ei->i_block_group * EXT3_BLOCKS_PER_GROUP(inode->i_sb)) +
229 le32_to_cpu(EXT3_SB(inode->i_sb)->s_es->s_first_data_block);
230 colour = (current->pid % 16) *
231 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
232 return bg_start + colour + block;
235 #define ll_unmap_underlying_metadata(sb, blocknr) \
236 unmap_underlying_metadata((sb)->s_bdev, blocknr)
238 #ifndef EXT3_MB_HINT_GROUP_ALLOC
239 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
240 struct ext3_ext_path *path, unsigned long block,
241 unsigned long *count, int *err)
243 unsigned long pblock, goal;
245 struct inode *inode = ext3_ext_base2inode(base);
247 goal = ext3_ext_find_goal(inode, path, block, &aflags);
248 aflags |= 2; /* block have been already reserved */
249 pblock = ext3_mb_new_blocks(handle, inode, goal, count, aflags, err);
254 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
255 struct ext3_ext_path *path, unsigned long block,
256 unsigned long *count, int *err)
258 struct inode *inode = ext3_ext_base2inode(base);
259 struct ext3_allocation_request ar;
260 unsigned long pblock;
263 /* find neighbour allocated blocks */
265 *err = ext3_ext_search_left(base, path, &ar.lleft, &ar.pleft);
269 *err = ext3_ext_search_right(base, path, &ar.lright, &ar.pright);
273 /* allocate new block */
274 ar.goal = ext3_ext_find_goal(inode, path, block, &aflags);
278 ar.flags = EXT3_MB_HINT_DATA;
279 pblock = ext3_mb_new_blocks(handle, &ar, err);
285 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
286 struct ext3_ext_path *path,
287 struct ext3_ext_cache *cex,
288 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
289 struct ext3_extent *ex,
293 struct bpointers *bp = cbdata;
294 struct inode *inode = ext3_ext_base2inode(base);
295 struct ext3_extent nex;
296 unsigned long pblock;
302 #ifdef EXT3_EXT_CACHE_EXTENT
303 if (cex->ec_type == EXT3_EXT_CACHE_EXTENT)
305 if ((cex->ec_len != 0) && (cex->ec_start != 0))
312 if (bp->create == 0) {
314 if (cex->ec_block < bp->start)
315 i = bp->start - cex->ec_block;
316 if (i >= cex->ec_len)
317 CERROR("nothing to do?! i = %d, e_num = %u\n",
319 for (; i < cex->ec_len && bp->num; i++) {
329 tgen = EXT_GENERATION(base);
330 count = ext3_ext_calc_credits_for_insert(base, path);
332 handle = ext3_journal_start(inode, count+EXT3_ALLOC_NEEDED+1);
333 if (IS_ERR(handle)) {
334 return PTR_ERR(handle);
337 if (tgen != EXT_GENERATION(base)) {
338 /* the tree has changed. so path can be invalid at moment */
339 ext3_journal_stop(handle);
343 /* In 2.6.32 kernel, ext4_ext_walk_space()'s callback func is not
344 * protected by i_data_sem as whole. so we patch it to store
345 * generation to path and now verify the tree hasn't changed */
346 down_write((&EXT4_I(inode)->i_data_sem));
348 /* validate extent, make sure the extent tree does not changed */
349 if (EXT_GENERATION(base) != path[0].p_generation) {
350 /* cex is invalid, try again */
351 up_write(&EXT4_I(inode)->i_data_sem);
352 ext3_journal_stop(handle);
357 pblock = new_blocks(handle, base, path, cex->ec_block, &count, &err);
360 EXT_ASSERT(count <= cex->ec_len);
362 /* insert new extent */
363 nex.ee_block = cpu_to_le32(cex->ec_block);
364 ext3_ext_store_pblock(&nex, pblock);
365 nex.ee_len = cpu_to_le16(count);
366 err = fsfilt_ext3_ext_insert_extent(handle, base, path, &nex, 0);
368 /* free data blocks we just allocated */
369 /* not a good idea to call discard here directly,
370 * but otherwise we'd need to call it every free() */
371 #ifdef EXT3_MB_HINT_GROUP_ALLOC
372 ext3_mb_discard_inode_preallocations(inode);
374 ext3_free_blocks(handle, inode, ext_pblock(&nex),
375 cpu_to_le16(nex.ee_len), 0);
380 * Putting len of the actual extent we just inserted,
381 * we are asking ext3_ext_walk_space() to continue
382 * scaning after that block
384 cex->ec_len = le16_to_cpu(nex.ee_len);
385 cex->ec_start = ext_pblock(&nex);
386 BUG_ON(le16_to_cpu(nex.ee_len) == 0);
387 BUG_ON(le32_to_cpu(nex.ee_block) != cex->ec_block);
390 up_write((&EXT4_I(inode)->i_data_sem));
391 ext3_journal_stop(handle);
396 CERROR("hmm. why do we find this extent?\n");
397 CERROR("initial space: %lu:%u\n",
398 bp->start, bp->init_num);
399 #ifdef EXT3_EXT_CACHE_EXTENT
400 CERROR("current extent: %u/%u/%llu %d\n",
401 cex->ec_block, cex->ec_len,
402 (unsigned long long)cex->ec_start,
405 CERROR("current extent: %u/%u/%llu\n",
406 cex->ec_block, cex->ec_len,
407 (unsigned long long)cex->ec_start);
411 if (cex->ec_block < bp->start)
412 i = bp->start - cex->ec_block;
413 if (i >= cex->ec_len)
414 CERROR("nothing to do?! i = %d, e_num = %u\n",
416 for (; i < cex->ec_len && bp->num; i++) {
417 *(bp->blocks) = cex->ec_start + i;
418 #ifdef EXT3_EXT_CACHE_EXTENT
419 if (cex->ec_type != EXT3_EXT_CACHE_EXTENT)
421 if ((cex->ec_len == 0) || (cex->ec_start == 0))
424 /* unmap any possible underlying metadata from
425 * the block device mapping. bug 6998. */
426 ll_unmap_underlying_metadata(inode->i_sb,
437 int fsfilt_map_nblocks(struct inode *inode, unsigned long block,
438 unsigned long num, unsigned long *blocks,
441 struct ext3_ext_base *base = inode;
445 CDEBUG(D_OTHER, "blocks %lu-%lu requested for inode %u\n",
446 block, block + num - 1, (unsigned) inode->i_ino);
450 bp.init_num = bp.num = num;
453 err = fsfilt_ext3_ext_walk_space(base, block, num,
454 ext3_ext_new_extent_cb, &bp);
455 ext3_ext_invalidate_cache(base);
460 int fsfilt_ext3_map_ext_inode_pages(struct inode *inode, struct page **page,
461 int pages, unsigned long *blocks,
464 int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
466 struct page *fp = NULL;
469 CDEBUG(D_OTHER, "inode %lu: map %d pages from %lu\n",
470 inode->i_ino, pages, (*page)->index);
472 /* pages are sorted already. so, we just have to find
473 * contig. space and process them properly */
476 /* start new extent */
481 } else if (fp->index + clen == (*page)->index) {
482 /* continue the extent */
489 /* process found extent */
490 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
491 clen * blocks_per_page, blocks,
496 /* look for next extent */
498 blocks += blocks_per_page * clen;
502 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
503 clen * blocks_per_page, blocks,
509 int fsfilt_ext3_map_bm_inode_pages(struct inode *inode, struct page **page,
510 int pages, unsigned long *blocks,
513 int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
517 for (i = 0, b = blocks; i < pages; i++, page++) {
518 rc = ext3_map_inode_page(inode, *page, b, create);
520 CERROR("ino %lu, blk %lu create %d: rc %d\n",
521 inode->i_ino, *b, create, rc);
525 b += blocks_per_page;
530 int fsfilt_ext3_map_inode_pages(struct inode *inode, struct page **page,
531 int pages, unsigned long *blocks,
532 int create, struct mutex *optional_mutex)
536 if (EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL) {
537 rc = fsfilt_ext3_map_ext_inode_pages(inode, page, pages,
541 if (optional_mutex != NULL)
542 mutex_lock(optional_mutex);
543 rc = fsfilt_ext3_map_bm_inode_pages(inode, page, pages, blocks, create);
544 if (optional_mutex != NULL)
545 mutex_unlock(optional_mutex);
550 int fsfilt_ext3_read(struct inode *inode, void *buf, int size, loff_t *offs)
553 struct buffer_head *bh;
554 int err, blocksize, csize, boffs, osize = size;
556 /* prevent reading after eof */
557 spin_lock(&inode->i_lock);
558 if (i_size_read(inode) < *offs + size) {
559 size = i_size_read(inode) - *offs;
560 spin_unlock(&inode->i_lock);
562 CDEBUG(D_EXT2, "size %llu is too short for read @%llu\n",
563 i_size_read(inode), *offs);
565 } else if (size == 0) {
569 spin_unlock(&inode->i_lock);
572 blocksize = 1 << inode->i_blkbits;
575 block = *offs >> inode->i_blkbits;
576 boffs = *offs & (blocksize - 1);
577 csize = min(blocksize - boffs, size);
578 bh = ext3_bread(NULL, inode, block, 0, &err);
580 CERROR("can't read block: %d\n", err);
584 memcpy(buf, bh->b_data + boffs, csize);
593 EXPORT_SYMBOL(fsfilt_ext3_read);
595 static int fsfilt_ext3_read_record(struct file * file, void *buf,
596 int size, loff_t *offs)
599 rc = fsfilt_ext3_read(file->f_dentry->d_inode, buf, size, offs);
605 int fsfilt_ext3_write_handle(struct inode *inode, void *buf, int bufsize,
606 loff_t *offs, handle_t *handle)
608 struct buffer_head *bh = NULL;
609 loff_t old_size = i_size_read(inode), offset = *offs;
610 loff_t new_size = i_size_read(inode);
612 int err = 0, blocksize = 1 << inode->i_blkbits, size, boffs;
614 while (bufsize > 0) {
618 block = offset >> inode->i_blkbits;
619 boffs = offset & (blocksize - 1);
620 size = min(blocksize - boffs, bufsize);
621 bh = ext3_bread(handle, inode, block, 1, &err);
623 CERROR("can't read/create block: %d\n", err);
627 err = ext3_journal_get_write_access(handle, bh);
629 CERROR("journal_get_write_access() returned error %d\n",
633 LASSERT(bh->b_data + boffs + size <= bh->b_data + bh->b_size);
634 memcpy(bh->b_data + boffs, buf, size);
635 err = ext3_journal_dirty_metadata(handle, bh);
637 CERROR("journal_dirty_metadata() returned error %d\n",
641 if (offset + size > new_size)
642 new_size = offset + size;
650 /* correct in-core and on-disk sizes */
651 if (new_size > i_size_read(inode)) {
652 spin_lock(&inode->i_lock);
653 if (new_size > i_size_read(inode))
654 i_size_write(inode, new_size);
655 if (i_size_read(inode) > EXT3_I(inode)->i_disksize)
656 EXT3_I(inode)->i_disksize = i_size_read(inode);
657 if (i_size_read(inode) > old_size) {
658 spin_unlock(&inode->i_lock);
659 mark_inode_dirty(inode);
661 spin_unlock(&inode->i_lock);
669 EXPORT_SYMBOL(fsfilt_ext3_write_handle);
671 static int fsfilt_ext3_write_record(struct file *file, void *buf, int bufsize,
672 loff_t *offs, int force_sync)
674 struct inode *inode = file->f_dentry->d_inode;
676 int err, block_count = 0, blocksize;
678 /* Determine how many transaction credits are needed */
679 blocksize = 1 << inode->i_blkbits;
680 block_count = (*offs & (blocksize - 1)) + bufsize;
681 block_count = (block_count + blocksize - 1) >> inode->i_blkbits;
683 handle = ext3_journal_start(inode,
684 block_count * EXT3_DATA_TRANS_BLOCKS(inode->i_sb) + 2);
685 if (IS_ERR(handle)) {
686 CERROR("can't start transaction for %d blocks (%d bytes)\n",
687 block_count * EXT3_DATA_TRANS_BLOCKS(inode->i_sb) + 2,
689 return PTR_ERR(handle);
692 err = fsfilt_ext3_write_handle(inode, buf, bufsize, offs, handle);
694 if (!err && force_sync)
695 handle->h_sync = 1; /* recovery likes this */
697 ext3_journal_stop(handle);
702 static int fsfilt_ext3_setup(struct super_block *sb)
704 if (!EXT3_HAS_COMPAT_FEATURE(sb,
705 EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
706 CERROR("ext3 mounted without journal\n");
711 CWARN("Enabling PDIROPS\n");
712 set_opt(EXT3_SB(sb)->s_mount_opt, PDIROPS);
713 sb->s_flags |= S_PDIROPS;
715 if (!EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
716 CWARN("filesystem doesn't have dir_index feature enabled\n");
719 static struct fsfilt_operations fsfilt_ext3_ops = {
721 .fs_owner = THIS_MODULE,
722 .fs_getlabel = fsfilt_ext3_get_label,
723 .fs_start = fsfilt_ext3_start,
724 .fs_commit = fsfilt_ext3_commit,
725 .fs_map_inode_pages = fsfilt_ext3_map_inode_pages,
726 .fs_write_record = fsfilt_ext3_write_record,
727 .fs_read_record = fsfilt_ext3_read_record,
728 .fs_setup = fsfilt_ext3_setup,
731 static int __init fsfilt_ext3_init(void)
735 fcb_cache = cfs_mem_cache_create("fsfilt_ext3_fcb",
736 sizeof(struct fsfilt_cb_data), 0, 0);
738 CERROR("error allocating fsfilt journal callback cache\n");
739 GOTO(out, rc = -ENOMEM);
742 rc = fsfilt_register_ops(&fsfilt_ext3_ops);
745 int err = cfs_mem_cache_destroy(fcb_cache);
746 LASSERTF(err == 0, "error destroying new cache: rc %d\n", err);
752 static void __exit fsfilt_ext3_exit(void)
756 fsfilt_unregister_ops(&fsfilt_ext3_ops);
757 rc = cfs_mem_cache_destroy(fcb_cache);
758 LASSERTF(rc == 0, "couldn't destroy fcb_cache slab\n");
761 module_init(fsfilt_ext3_init);
762 module_exit(fsfilt_ext3_exit);
764 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
765 MODULE_DESCRIPTION("Lustre ext3 Filesystem Helper v0.1");
766 MODULE_LICENSE("GPL");