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/bitops.h>
52 #include <linux/quota.h>
54 #include <linux/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 #ifdef HAVE_EXT_PBLOCK /* Name changed to ext4_ext_pblock for kernel 2.6.35 */
63 #define ext3_ext_pblock(ex) ext_pblock((ex))
66 /* for kernels 2.6.18 and later */
67 #define FSFILT_SINGLEDATA_TRANS_BLOCKS(sb) EXT3_SINGLEDATA_TRANS_BLOCKS(sb)
69 #define fsfilt_ext3_ext_insert_extent(handle, inode, path, newext, flag) \
70 ext3_ext_insert_extent(handle, inode, path, newext, flag)
72 #define ext3_mb_discard_inode_preallocations(inode) \
73 ext3_discard_preallocations(inode)
75 #define fsfilt_log_start_commit(journal, tid) jbd2_log_start_commit(journal, tid)
76 #define fsfilt_log_wait_commit(journal, tid) jbd2_log_wait_commit(journal, tid)
78 static struct kmem_cache *fcb_cache;
80 struct fsfilt_cb_data {
81 struct ext4_journal_cb_entry cb_jcb; /* private data - MUST BE FIRST */
82 fsfilt_cb_t cb_func; /* MDS/OBD completion function */
83 struct obd_device *cb_obd; /* MDS/OBD completion device */
84 __u64 cb_last_rcvd; /* MDS/OST last committed operation */
85 void *cb_data; /* MDS/OST completion function data */
88 static char *fsfilt_ext3_get_label(struct super_block *sb)
90 return EXT3_SB(sb)->s_es->s_volume_name;
93 /* kernel has ext4_blocks_for_truncate since linux-3.1.1 */
94 # include <ext4/truncate.h>
97 * We don't currently need any additional blocks for rmdir and
98 * unlink transactions because we are storing the OST oa_id inside
99 * the inode (which we will be changing anyways as part of this
102 static void *fsfilt_ext3_start(struct inode *inode, int op, void *desc_private,
105 /* For updates to the last received file */
106 int nblocks = FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
110 if (current->journal_info) {
111 CDEBUG(D_INODE, "increasing refcount on %p\n",
112 current->journal_info);
117 case FSFILT_OP_UNLINK:
118 /* delete one file + create/update logs for each stripe */
119 nblocks += EXT3_DELETE_TRANS_BLOCKS(inode->i_sb);
120 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
121 FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
123 case FSFILT_OP_CANCEL_UNLINK:
126 /* blocks for log header bitmap update OR
127 * blocks for catalog header bitmap update + unlink of logs +
128 * blocks for delete the inode (include blocks truncating). */
129 nblocks = (LLOG_CHUNK_SIZE >> inode->i_blkbits) +
130 EXT3_DELETE_TRANS_BLOCKS(inode->i_sb) +
131 ext4_blocks_for_truncate(inode) + 3;
133 default: CERROR("unknown transaction start op %d\n", op);
137 LASSERT(current->journal_info == desc_private);
138 journal = EXT3_SB(inode->i_sb)->s_journal;
139 if (nblocks > journal->j_max_transaction_buffers) {
140 CWARN("too many credits %d for op %ux%u using %d instead\n",
141 nblocks, op, logs, journal->j_max_transaction_buffers);
142 nblocks = journal->j_max_transaction_buffers;
146 LASSERTF(nblocks > 0, "can't start %d credit transaction\n", nblocks);
147 handle = ext3_journal_start(inode, nblocks);
150 LASSERT(current->journal_info == handle);
152 CERROR("error starting handle for op %u (%u credits): rc %ld\n",
153 op, nblocks, PTR_ERR(handle));
157 static int fsfilt_ext3_commit(struct inode *inode, void *h, int force_sync)
160 handle_t *handle = h;
162 LASSERT(current->journal_info == handle);
164 handle->h_sync = 1; /* recovery likes this */
166 rc = ext3_journal_stop(handle);
171 #ifndef EXT3_EXTENTS_FL
172 #define EXT3_EXTENTS_FL 0x00080000 /* Inode uses extents */
176 #define EXT_ASSERT(cond) BUG_ON(!(cond))
179 #define EXT_GENERATION(inode) (EXT4_I(inode)->i_ext_generation)
180 #define ext3_ext_base inode
181 #define ext3_ext_base2inode(inode) (inode)
182 #define EXT_DEPTH(inode) ext_depth(inode)
183 #define fsfilt_ext3_ext_walk_space(inode, block, num, cb, cbdata) \
184 ext3_ext_walk_space(inode, block, num, cb, cbdata);
187 unsigned long *blocks;
194 static long ext3_ext_find_goal(struct inode *inode, struct ext3_ext_path *path,
195 unsigned long block, int *aflags)
197 struct ext3_inode_info *ei = EXT3_I(inode);
198 unsigned long bg_start;
199 unsigned long colour;
203 struct ext3_extent *ex;
204 depth = path->p_depth;
206 /* try to predict block placement */
207 if ((ex = path[depth].p_ext))
208 return ext4_ext_pblock(ex) + (block - le32_to_cpu(ex->ee_block));
210 /* it looks index is empty
211 * try to find starting from index itself */
212 if (path[depth].p_bh)
213 return path[depth].p_bh->b_blocknr;
216 /* OK. use inode's group */
217 bg_start = (ei->i_block_group * EXT3_BLOCKS_PER_GROUP(inode->i_sb)) +
218 le32_to_cpu(EXT3_SB(inode->i_sb)->s_es->s_first_data_block);
219 colour = (current->pid % 16) *
220 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
221 return bg_start + colour + block;
224 #define ll_unmap_underlying_metadata(sb, blocknr) \
225 unmap_underlying_metadata((sb)->s_bdev, blocknr)
227 #ifndef EXT3_MB_HINT_GROUP_ALLOC
228 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
229 struct ext3_ext_path *path, unsigned long block,
230 unsigned long *count, int *err)
232 unsigned long pblock, goal;
234 struct inode *inode = ext3_ext_base2inode(base);
236 goal = ext3_ext_find_goal(inode, path, block, &aflags);
237 aflags |= 2; /* block have been already reserved */
238 pblock = ext3_mb_new_blocks(handle, inode, goal, count, aflags, err);
243 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
244 struct ext3_ext_path *path, unsigned long block,
245 unsigned long *count, int *err)
247 struct inode *inode = ext3_ext_base2inode(base);
248 struct ext3_allocation_request ar;
249 unsigned long pblock;
252 /* find neighbour allocated blocks */
254 *err = ext3_ext_search_left(base, path, &ar.lleft, &ar.pleft);
258 *err = ext3_ext_search_right(base, path, &ar.lright, &ar.pright);
262 /* allocate new block */
263 ar.goal = ext3_ext_find_goal(inode, path, block, &aflags);
267 ar.flags = EXT3_MB_HINT_DATA;
268 pblock = ext3_mb_new_blocks(handle, &ar, err);
274 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
275 struct ext3_ext_path *path,
276 struct ext3_ext_cache *cex,
277 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
278 struct ext3_extent *ex,
282 struct bpointers *bp = cbdata;
283 struct inode *inode = ext3_ext_base2inode(base);
284 struct ext3_extent nex;
285 unsigned long pblock;
291 #ifdef EXT3_EXT_CACHE_EXTENT
292 if (cex->ec_type == EXT3_EXT_CACHE_EXTENT)
294 if ((cex->ec_len != 0) && (cex->ec_start != 0))
301 if (bp->create == 0) {
303 if (cex->ec_block < bp->start)
304 i = bp->start - cex->ec_block;
305 if (i >= cex->ec_len)
306 CERROR("nothing to do?! i = %d, e_num = %u\n",
308 for (; i < cex->ec_len && bp->num; i++) {
318 tgen = EXT_GENERATION(base);
319 count = ext3_ext_calc_credits_for_insert(base, path);
321 handle = ext3_journal_start(inode, count+EXT3_ALLOC_NEEDED+1);
322 if (IS_ERR(handle)) {
323 return PTR_ERR(handle);
326 if (tgen != EXT_GENERATION(base)) {
327 /* the tree has changed. so path can be invalid at moment */
328 ext3_journal_stop(handle);
332 /* In 2.6.32 kernel, ext4_ext_walk_space()'s callback func is not
333 * protected by i_data_sem as whole. so we patch it to store
334 * generation to path and now verify the tree hasn't changed */
335 down_write((&EXT4_I(inode)->i_data_sem));
337 /* validate extent, make sure the extent tree does not changed */
338 if (EXT_GENERATION(base) != path[0].p_generation) {
339 /* cex is invalid, try again */
340 up_write(&EXT4_I(inode)->i_data_sem);
341 ext3_journal_stop(handle);
346 pblock = new_blocks(handle, base, path, cex->ec_block, &count, &err);
349 EXT_ASSERT(count <= cex->ec_len);
351 /* insert new extent */
352 nex.ee_block = cpu_to_le32(cex->ec_block);
353 ext3_ext_store_pblock(&nex, pblock);
354 nex.ee_len = cpu_to_le16(count);
355 err = fsfilt_ext3_ext_insert_extent(handle, base, path, &nex, 0);
357 /* free data blocks we just allocated */
358 /* not a good idea to call discard here directly,
359 * but otherwise we'd need to call it every free() */
360 #ifdef EXT3_MB_HINT_GROUP_ALLOC
361 ext3_mb_discard_inode_preallocations(inode);
363 #ifdef HAVE_EXT_FREE_BLOCK_WITH_BUFFER_HEAD /* Introduced in 2.6.32-rc7 */
364 ext3_free_blocks(handle, inode, NULL, ext4_ext_pblock(&nex),
365 cpu_to_le16(nex.ee_len), 0);
367 ext3_free_blocks(handle, inode, ext4_ext_pblock(&nex),
368 cpu_to_le16(nex.ee_len), 0);
374 * Putting len of the actual extent we just inserted,
375 * we are asking ext3_ext_walk_space() to continue
376 * scaning after that block
378 cex->ec_len = le16_to_cpu(nex.ee_len);
379 cex->ec_start = ext4_ext_pblock(&nex);
380 BUG_ON(le16_to_cpu(nex.ee_len) == 0);
381 BUG_ON(le32_to_cpu(nex.ee_block) != cex->ec_block);
384 up_write((&EXT4_I(inode)->i_data_sem));
385 ext3_journal_stop(handle);
390 CERROR("hmm. why do we find this extent?\n");
391 CERROR("initial space: %lu:%u\n",
392 bp->start, bp->init_num);
393 #ifdef EXT3_EXT_CACHE_EXTENT
394 CERROR("current extent: %u/%u/%llu %d\n",
395 cex->ec_block, cex->ec_len,
396 (unsigned long long)cex->ec_start,
399 CERROR("current extent: %u/%u/%llu\n",
400 cex->ec_block, cex->ec_len,
401 (unsigned long long)cex->ec_start);
405 if (cex->ec_block < bp->start)
406 i = bp->start - cex->ec_block;
407 if (i >= cex->ec_len)
408 CERROR("nothing to do?! i = %d, e_num = %u\n",
410 for (; i < cex->ec_len && bp->num; i++) {
411 *(bp->blocks) = cex->ec_start + i;
412 #ifdef EXT3_EXT_CACHE_EXTENT
413 if (cex->ec_type != EXT3_EXT_CACHE_EXTENT)
415 if ((cex->ec_len == 0) || (cex->ec_start == 0))
418 /* unmap any possible underlying metadata from
419 * the block device mapping. bug 6998. */
420 ll_unmap_underlying_metadata(inode->i_sb,
431 int fsfilt_map_nblocks(struct inode *inode, unsigned long block,
432 unsigned long num, unsigned long *blocks,
435 struct ext3_ext_base *base = inode;
439 CDEBUG(D_OTHER, "blocks %lu-%lu requested for inode %u\n",
440 block, block + num - 1, (unsigned) inode->i_ino);
444 bp.init_num = bp.num = num;
447 err = fsfilt_ext3_ext_walk_space(base, block, num,
448 ext3_ext_new_extent_cb, &bp);
449 ext3_ext_invalidate_cache(base);
454 int fsfilt_ext3_map_ext_inode_pages(struct inode *inode, struct page **page,
455 int pages, unsigned long *blocks,
458 int blocks_per_page = PAGE_CACHE_SIZE >> inode->i_blkbits;
460 struct page *fp = NULL;
463 CDEBUG(D_OTHER, "inode %lu: map %d pages from %lu\n",
464 inode->i_ino, pages, (*page)->index);
466 /* pages are sorted already. so, we just have to find
467 * contig. space and process them properly */
470 /* start new extent */
475 } else if (fp->index + clen == (*page)->index) {
476 /* continue the extent */
483 /* process found extent */
484 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
485 clen * blocks_per_page, blocks,
490 /* look for next extent */
492 blocks += blocks_per_page * clen;
496 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
497 clen * blocks_per_page, blocks,
503 int fsfilt_ext3_map_bm_inode_pages(struct inode *inode, struct page **page,
504 int pages, unsigned long *blocks,
507 int blocks_per_page = PAGE_CACHE_SIZE >> inode->i_blkbits;
511 for (i = 0, b = blocks; i < pages; i++, page++) {
512 rc = ext3_map_inode_page(inode, *page, b, create);
514 CERROR("ino %lu, blk %lu create %d: rc %d\n",
515 inode->i_ino, *b, create, rc);
519 b += blocks_per_page;
524 int fsfilt_ext3_map_inode_pages(struct inode *inode, struct page **page,
525 int pages, unsigned long *blocks,
526 int create, struct mutex *optional_mutex)
530 if (EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL) {
531 rc = fsfilt_ext3_map_ext_inode_pages(inode, page, pages,
535 if (optional_mutex != NULL)
536 mutex_lock(optional_mutex);
537 rc = fsfilt_ext3_map_bm_inode_pages(inode, page, pages, blocks, create);
538 if (optional_mutex != NULL)
539 mutex_unlock(optional_mutex);
544 int fsfilt_ext3_read(struct inode *inode, void *buf, int size, loff_t *offs)
547 struct buffer_head *bh;
548 int err, blocksize, csize, boffs, osize = size;
550 /* prevent reading after eof */
551 spin_lock(&inode->i_lock);
552 if (i_size_read(inode) < *offs + size) {
553 size = i_size_read(inode) - *offs;
554 spin_unlock(&inode->i_lock);
556 CDEBUG(D_EXT2, "size %llu is too short for read @%llu\n",
557 i_size_read(inode), *offs);
559 } else if (size == 0) {
563 spin_unlock(&inode->i_lock);
566 blocksize = 1 << inode->i_blkbits;
569 block = *offs >> inode->i_blkbits;
570 boffs = *offs & (blocksize - 1);
571 csize = min(blocksize - boffs, size);
572 bh = ext3_bread(NULL, inode, block, 0, &err);
574 CERROR("can't read block: %d\n", err);
578 memcpy(buf, bh->b_data + boffs, csize);
587 EXPORT_SYMBOL(fsfilt_ext3_read);
589 static int fsfilt_ext3_read_record(struct file * file, void *buf,
590 int size, loff_t *offs)
593 rc = fsfilt_ext3_read(file->f_dentry->d_inode, buf, size, offs);
599 int fsfilt_ext3_write_handle(struct inode *inode, void *buf, int bufsize,
600 loff_t *offs, handle_t *handle)
602 struct buffer_head *bh = NULL;
603 loff_t old_size = i_size_read(inode), offset = *offs;
604 loff_t new_size = i_size_read(inode);
606 int err = 0, blocksize = 1 << inode->i_blkbits, size, boffs;
608 while (bufsize > 0) {
612 block = offset >> inode->i_blkbits;
613 boffs = offset & (blocksize - 1);
614 size = min(blocksize - boffs, bufsize);
615 bh = ext3_bread(handle, inode, block, 1, &err);
617 CERROR("can't read/create block: %d\n", err);
621 err = ext3_journal_get_write_access(handle, bh);
623 CERROR("journal_get_write_access() returned error %d\n",
627 LASSERT(bh->b_data + boffs + size <= bh->b_data + bh->b_size);
628 memcpy(bh->b_data + boffs, buf, size);
629 err = ext3_journal_dirty_metadata(handle, bh);
631 CERROR("journal_dirty_metadata() returned error %d\n",
635 if (offset + size > new_size)
636 new_size = offset + size;
644 /* correct in-core and on-disk sizes */
645 if (new_size > i_size_read(inode)) {
646 spin_lock(&inode->i_lock);
647 if (new_size > i_size_read(inode))
648 i_size_write(inode, new_size);
649 if (i_size_read(inode) > EXT3_I(inode)->i_disksize)
650 EXT3_I(inode)->i_disksize = i_size_read(inode);
651 if (i_size_read(inode) > old_size) {
652 spin_unlock(&inode->i_lock);
653 mark_inode_dirty(inode);
655 spin_unlock(&inode->i_lock);
663 EXPORT_SYMBOL(fsfilt_ext3_write_handle);
665 static int fsfilt_ext3_write_record(struct file *file, void *buf, int bufsize,
666 loff_t *offs, int force_sync)
668 struct inode *inode = file->f_dentry->d_inode;
670 int err, block_count = 0, blocksize;
672 /* Determine how many transaction credits are needed */
673 blocksize = 1 << inode->i_blkbits;
674 block_count = (*offs & (blocksize - 1)) + bufsize;
675 block_count = (block_count + blocksize - 1) >> inode->i_blkbits;
677 handle = ext3_journal_start(inode,
678 block_count * EXT3_DATA_TRANS_BLOCKS(inode->i_sb) + 2);
679 if (IS_ERR(handle)) {
680 CERROR("can't start transaction for %d blocks (%d bytes)\n",
681 block_count * EXT3_DATA_TRANS_BLOCKS(inode->i_sb) + 2,
683 return PTR_ERR(handle);
686 err = fsfilt_ext3_write_handle(inode, buf, bufsize, offs, handle);
688 if (!err && force_sync)
689 handle->h_sync = 1; /* recovery likes this */
691 ext3_journal_stop(handle);
696 static int fsfilt_ext3_setup(struct super_block *sb)
698 if (!EXT3_HAS_COMPAT_FEATURE(sb,
699 EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
700 CERROR("ext3 mounted without journal\n");
705 CWARN("Enabling PDIROPS\n");
706 set_opt(EXT3_SB(sb)->s_mount_opt, PDIROPS);
707 sb->s_flags |= S_PDIROPS;
709 if (!EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
710 CWARN("filesystem doesn't have dir_index feature enabled\n");
713 static struct fsfilt_operations fsfilt_ext3_ops = {
715 .fs_owner = THIS_MODULE,
716 .fs_getlabel = fsfilt_ext3_get_label,
717 .fs_start = fsfilt_ext3_start,
718 .fs_commit = fsfilt_ext3_commit,
719 .fs_map_inode_pages = fsfilt_ext3_map_inode_pages,
720 .fs_write_record = fsfilt_ext3_write_record,
721 .fs_read_record = fsfilt_ext3_read_record,
722 .fs_setup = fsfilt_ext3_setup,
725 static int __init fsfilt_ext3_init(void)
729 fcb_cache = kmem_cache_create("fsfilt_ext3_fcb",
730 sizeof(struct fsfilt_cb_data), 0, 0);
732 CERROR("error allocating fsfilt journal callback cache\n");
733 GOTO(out, rc = -ENOMEM);
736 rc = fsfilt_register_ops(&fsfilt_ext3_ops);
739 int err = kmem_cache_destroy(fcb_cache);
740 LASSERTF(err == 0, "error destroying new cache: rc %d\n", err);
746 static void __exit fsfilt_ext3_exit(void)
750 fsfilt_unregister_ops(&fsfilt_ext3_ops);
751 rc = kmem_cache_destroy(fcb_cache);
752 LASSERTF(rc == 0, "couldn't destroy fcb_cache slab\n");
755 module_init(fsfilt_ext3_init);
756 module_exit(fsfilt_ext3_exit);
758 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
759 MODULE_DESCRIPTION("Lustre ext3 Filesystem Helper v0.1");
760 MODULE_LICENSE("GPL");