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, 2013, 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 #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 #ifdef HAVE_BLOCKS_FOR_TRUNCATE
95 # include <ext4/truncate.h>
97 static inline unsigned long ext4_blocks_for_truncate(struct inode *inode)
101 needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
104 if (needed > EXT4_MAX_TRANS_DATA)
105 needed = EXT4_MAX_TRANS_DATA;
106 return EXT4_DATA_TRANS_BLOCKS(inode->i_sb) + needed;
111 * We don't currently need any additional blocks for rmdir and
112 * unlink transactions because we are storing the OST oa_id inside
113 * the inode (which we will be changing anyways as part of this
116 static void *fsfilt_ext3_start(struct inode *inode, int op, void *desc_private,
119 /* For updates to the last received file */
120 int nblocks = FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb);
124 if (current->journal_info) {
125 CDEBUG(D_INODE, "increasing refcount on %p\n",
126 current->journal_info);
131 case FSFILT_OP_UNLINK:
132 /* delete one file + create/update logs for each stripe */
133 nblocks += EXT3_DELETE_TRANS_BLOCKS(inode->i_sb);
134 nblocks += (EXT3_INDEX_EXTRA_TRANS_BLOCKS +
135 FSFILT_SINGLEDATA_TRANS_BLOCKS(inode->i_sb)) * logs;
137 case FSFILT_OP_CANCEL_UNLINK:
140 /* blocks for log header bitmap update OR
141 * blocks for catalog header bitmap update + unlink of logs +
142 * blocks for delete the inode (include blocks truncating). */
143 nblocks = (LLOG_CHUNK_SIZE >> inode->i_blkbits) +
144 EXT3_DELETE_TRANS_BLOCKS(inode->i_sb) +
145 ext4_blocks_for_truncate(inode) + 3;
147 default: CERROR("unknown transaction start op %d\n", op);
151 LASSERT(current->journal_info == desc_private);
152 journal = EXT3_SB(inode->i_sb)->s_journal;
153 if (nblocks > journal->j_max_transaction_buffers) {
154 CWARN("too many credits %d for op %ux%u using %d instead\n",
155 nblocks, op, logs, journal->j_max_transaction_buffers);
156 nblocks = journal->j_max_transaction_buffers;
160 LASSERTF(nblocks > 0, "can't start %d credit transaction\n", nblocks);
161 handle = ext3_journal_start(inode, nblocks);
164 LASSERT(current->journal_info == handle);
166 CERROR("error starting handle for op %u (%u credits): rc %ld\n",
167 op, nblocks, PTR_ERR(handle));
171 static int fsfilt_ext3_commit(struct inode *inode, void *h, int force_sync)
174 handle_t *handle = h;
176 LASSERT(current->journal_info == handle);
178 handle->h_sync = 1; /* recovery likes this */
180 rc = ext3_journal_stop(handle);
185 #ifndef EXT3_EXTENTS_FL
186 #define EXT3_EXTENTS_FL 0x00080000 /* Inode uses extents */
190 #define EXT_ASSERT(cond) BUG_ON(!(cond))
193 #define EXT_GENERATION(inode) (EXT4_I(inode)->i_ext_generation)
194 #define ext3_ext_base inode
195 #define ext3_ext_base2inode(inode) (inode)
196 #define EXT_DEPTH(inode) ext_depth(inode)
197 #define fsfilt_ext3_ext_walk_space(inode, block, num, cb, cbdata) \
198 ext3_ext_walk_space(inode, block, num, cb, cbdata);
201 unsigned long *blocks;
208 static long ext3_ext_find_goal(struct inode *inode, struct ext3_ext_path *path,
209 unsigned long block, int *aflags)
211 struct ext3_inode_info *ei = EXT3_I(inode);
212 unsigned long bg_start;
213 unsigned long colour;
217 struct ext3_extent *ex;
218 depth = path->p_depth;
220 /* try to predict block placement */
221 if ((ex = path[depth].p_ext))
222 return ext4_ext_pblock(ex) + (block - le32_to_cpu(ex->ee_block));
224 /* it looks index is empty
225 * try to find starting from index itself */
226 if (path[depth].p_bh)
227 return path[depth].p_bh->b_blocknr;
230 /* OK. use inode's group */
231 bg_start = (ei->i_block_group * EXT3_BLOCKS_PER_GROUP(inode->i_sb)) +
232 le32_to_cpu(EXT3_SB(inode->i_sb)->s_es->s_first_data_block);
233 colour = (current->pid % 16) *
234 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
235 return bg_start + colour + block;
238 #define ll_unmap_underlying_metadata(sb, blocknr) \
239 unmap_underlying_metadata((sb)->s_bdev, blocknr)
241 #ifndef EXT3_MB_HINT_GROUP_ALLOC
242 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
243 struct ext3_ext_path *path, unsigned long block,
244 unsigned long *count, int *err)
246 unsigned long pblock, goal;
248 struct inode *inode = ext3_ext_base2inode(base);
250 goal = ext3_ext_find_goal(inode, path, block, &aflags);
251 aflags |= 2; /* block have been already reserved */
252 pblock = ext3_mb_new_blocks(handle, inode, goal, count, aflags, err);
257 static unsigned long new_blocks(handle_t *handle, struct ext3_ext_base *base,
258 struct ext3_ext_path *path, unsigned long block,
259 unsigned long *count, int *err)
261 struct inode *inode = ext3_ext_base2inode(base);
262 struct ext3_allocation_request ar;
263 unsigned long pblock;
266 /* find neighbour allocated blocks */
268 *err = ext3_ext_search_left(base, path, &ar.lleft, &ar.pleft);
272 *err = ext3_ext_search_right(base, path, &ar.lright, &ar.pright);
276 /* allocate new block */
277 ar.goal = ext3_ext_find_goal(inode, path, block, &aflags);
281 ar.flags = EXT3_MB_HINT_DATA;
282 pblock = ext3_mb_new_blocks(handle, &ar, err);
288 static int ext3_ext_new_extent_cb(struct ext3_ext_base *base,
289 struct ext3_ext_path *path,
290 struct ext3_ext_cache *cex,
291 #ifdef HAVE_EXT_PREPARE_CB_EXTENT
292 struct ext3_extent *ex,
296 struct bpointers *bp = cbdata;
297 struct inode *inode = ext3_ext_base2inode(base);
298 struct ext3_extent nex;
299 unsigned long pblock;
305 #ifdef EXT3_EXT_CACHE_EXTENT
306 if (cex->ec_type == EXT3_EXT_CACHE_EXTENT)
308 if ((cex->ec_len != 0) && (cex->ec_start != 0))
315 if (bp->create == 0) {
317 if (cex->ec_block < bp->start)
318 i = bp->start - cex->ec_block;
319 if (i >= cex->ec_len)
320 CERROR("nothing to do?! i = %d, e_num = %u\n",
322 for (; i < cex->ec_len && bp->num; i++) {
332 tgen = EXT_GENERATION(base);
333 count = ext3_ext_calc_credits_for_insert(base, path);
335 handle = ext3_journal_start(inode, count+EXT3_ALLOC_NEEDED+1);
336 if (IS_ERR(handle)) {
337 return PTR_ERR(handle);
340 if (tgen != EXT_GENERATION(base)) {
341 /* the tree has changed. so path can be invalid at moment */
342 ext3_journal_stop(handle);
346 /* In 2.6.32 kernel, ext4_ext_walk_space()'s callback func is not
347 * protected by i_data_sem as whole. so we patch it to store
348 * generation to path and now verify the tree hasn't changed */
349 down_write((&EXT4_I(inode)->i_data_sem));
351 /* validate extent, make sure the extent tree does not changed */
352 if (EXT_GENERATION(base) != path[0].p_generation) {
353 /* cex is invalid, try again */
354 up_write(&EXT4_I(inode)->i_data_sem);
355 ext3_journal_stop(handle);
360 pblock = new_blocks(handle, base, path, cex->ec_block, &count, &err);
363 EXT_ASSERT(count <= cex->ec_len);
365 /* insert new extent */
366 nex.ee_block = cpu_to_le32(cex->ec_block);
367 ext3_ext_store_pblock(&nex, pblock);
368 nex.ee_len = cpu_to_le16(count);
369 err = fsfilt_ext3_ext_insert_extent(handle, base, path, &nex, 0);
371 /* free data blocks we just allocated */
372 /* not a good idea to call discard here directly,
373 * but otherwise we'd need to call it every free() */
374 #ifdef EXT3_MB_HINT_GROUP_ALLOC
375 ext3_mb_discard_inode_preallocations(inode);
377 #ifdef HAVE_EXT_FREE_BLOCK_WITH_BUFFER_HEAD /* Introduced in 2.6.32-rc7 */
378 ext3_free_blocks(handle, inode, NULL, ext4_ext_pblock(&nex),
379 cpu_to_le16(nex.ee_len), 0);
381 ext3_free_blocks(handle, inode, ext4_ext_pblock(&nex),
382 cpu_to_le16(nex.ee_len), 0);
388 * Putting len of the actual extent we just inserted,
389 * we are asking ext3_ext_walk_space() to continue
390 * scaning after that block
392 cex->ec_len = le16_to_cpu(nex.ee_len);
393 cex->ec_start = ext4_ext_pblock(&nex);
394 BUG_ON(le16_to_cpu(nex.ee_len) == 0);
395 BUG_ON(le32_to_cpu(nex.ee_block) != cex->ec_block);
398 up_write((&EXT4_I(inode)->i_data_sem));
399 ext3_journal_stop(handle);
404 CERROR("hmm. why do we find this extent?\n");
405 CERROR("initial space: %lu:%u\n",
406 bp->start, bp->init_num);
407 #ifdef EXT3_EXT_CACHE_EXTENT
408 CERROR("current extent: %u/%u/%llu %d\n",
409 cex->ec_block, cex->ec_len,
410 (unsigned long long)cex->ec_start,
413 CERROR("current extent: %u/%u/%llu\n",
414 cex->ec_block, cex->ec_len,
415 (unsigned long long)cex->ec_start);
419 if (cex->ec_block < bp->start)
420 i = bp->start - cex->ec_block;
421 if (i >= cex->ec_len)
422 CERROR("nothing to do?! i = %d, e_num = %u\n",
424 for (; i < cex->ec_len && bp->num; i++) {
425 *(bp->blocks) = cex->ec_start + i;
426 #ifdef EXT3_EXT_CACHE_EXTENT
427 if (cex->ec_type != EXT3_EXT_CACHE_EXTENT)
429 if ((cex->ec_len == 0) || (cex->ec_start == 0))
432 /* unmap any possible underlying metadata from
433 * the block device mapping. bug 6998. */
434 ll_unmap_underlying_metadata(inode->i_sb,
445 int fsfilt_map_nblocks(struct inode *inode, unsigned long block,
446 unsigned long num, unsigned long *blocks,
449 struct ext3_ext_base *base = inode;
453 CDEBUG(D_OTHER, "blocks %lu-%lu requested for inode %u\n",
454 block, block + num - 1, (unsigned) inode->i_ino);
458 bp.init_num = bp.num = num;
461 err = fsfilt_ext3_ext_walk_space(base, block, num,
462 ext3_ext_new_extent_cb, &bp);
463 ext3_ext_invalidate_cache(base);
468 int fsfilt_ext3_map_ext_inode_pages(struct inode *inode, struct page **page,
469 int pages, unsigned long *blocks,
472 int blocks_per_page = PAGE_CACHE_SIZE >> inode->i_blkbits;
474 struct page *fp = NULL;
477 CDEBUG(D_OTHER, "inode %lu: map %d pages from %lu\n",
478 inode->i_ino, pages, (*page)->index);
480 /* pages are sorted already. so, we just have to find
481 * contig. space and process them properly */
484 /* start new extent */
489 } else if (fp->index + clen == (*page)->index) {
490 /* continue the extent */
497 /* process found extent */
498 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
499 clen * blocks_per_page, blocks,
504 /* look for next extent */
506 blocks += blocks_per_page * clen;
510 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
511 clen * blocks_per_page, blocks,
517 int fsfilt_ext3_map_bm_inode_pages(struct inode *inode, struct page **page,
518 int pages, unsigned long *blocks,
521 int blocks_per_page = PAGE_CACHE_SIZE >> inode->i_blkbits;
525 for (i = 0, b = blocks; i < pages; i++, page++) {
526 rc = ext3_map_inode_page(inode, *page, b, create);
528 CERROR("ino %lu, blk %lu create %d: rc %d\n",
529 inode->i_ino, *b, create, rc);
533 b += blocks_per_page;
538 int fsfilt_ext3_map_inode_pages(struct inode *inode, struct page **page,
539 int pages, unsigned long *blocks,
540 int create, struct mutex *optional_mutex)
544 if (EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL) {
545 rc = fsfilt_ext3_map_ext_inode_pages(inode, page, pages,
549 if (optional_mutex != NULL)
550 mutex_lock(optional_mutex);
551 rc = fsfilt_ext3_map_bm_inode_pages(inode, page, pages, blocks, create);
552 if (optional_mutex != NULL)
553 mutex_unlock(optional_mutex);
558 int fsfilt_ext3_read(struct inode *inode, void *buf, int size, loff_t *offs)
561 struct buffer_head *bh;
562 int err, blocksize, csize, boffs, osize = size;
564 /* prevent reading after eof */
565 spin_lock(&inode->i_lock);
566 if (i_size_read(inode) < *offs + size) {
567 size = i_size_read(inode) - *offs;
568 spin_unlock(&inode->i_lock);
570 CDEBUG(D_EXT2, "size %llu is too short for read @%llu\n",
571 i_size_read(inode), *offs);
573 } else if (size == 0) {
577 spin_unlock(&inode->i_lock);
580 blocksize = 1 << inode->i_blkbits;
583 block = *offs >> inode->i_blkbits;
584 boffs = *offs & (blocksize - 1);
585 csize = min(blocksize - boffs, size);
586 bh = ext3_bread(NULL, inode, block, 0, &err);
588 CERROR("can't read block: %d\n", err);
592 memcpy(buf, bh->b_data + boffs, csize);
601 EXPORT_SYMBOL(fsfilt_ext3_read);
603 static int fsfilt_ext3_read_record(struct file * file, void *buf,
604 int size, loff_t *offs)
607 rc = fsfilt_ext3_read(file->f_dentry->d_inode, buf, size, offs);
613 int fsfilt_ext3_write_handle(struct inode *inode, void *buf, int bufsize,
614 loff_t *offs, handle_t *handle)
616 struct buffer_head *bh = NULL;
617 loff_t old_size = i_size_read(inode), offset = *offs;
618 loff_t new_size = i_size_read(inode);
620 int err = 0, blocksize = 1 << inode->i_blkbits, size, boffs;
622 while (bufsize > 0) {
626 block = offset >> inode->i_blkbits;
627 boffs = offset & (blocksize - 1);
628 size = min(blocksize - boffs, bufsize);
629 bh = ext3_bread(handle, inode, block, 1, &err);
631 CERROR("can't read/create block: %d\n", err);
635 err = ext3_journal_get_write_access(handle, bh);
637 CERROR("journal_get_write_access() returned error %d\n",
641 LASSERT(bh->b_data + boffs + size <= bh->b_data + bh->b_size);
642 memcpy(bh->b_data + boffs, buf, size);
643 err = ext3_journal_dirty_metadata(handle, bh);
645 CERROR("journal_dirty_metadata() returned error %d\n",
649 if (offset + size > new_size)
650 new_size = offset + size;
658 /* correct in-core and on-disk sizes */
659 if (new_size > i_size_read(inode)) {
660 spin_lock(&inode->i_lock);
661 if (new_size > i_size_read(inode))
662 i_size_write(inode, new_size);
663 if (i_size_read(inode) > EXT3_I(inode)->i_disksize)
664 EXT3_I(inode)->i_disksize = i_size_read(inode);
665 if (i_size_read(inode) > old_size) {
666 spin_unlock(&inode->i_lock);
667 mark_inode_dirty(inode);
669 spin_unlock(&inode->i_lock);
677 EXPORT_SYMBOL(fsfilt_ext3_write_handle);
679 static int fsfilt_ext3_write_record(struct file *file, void *buf, int bufsize,
680 loff_t *offs, int force_sync)
682 struct inode *inode = file->f_dentry->d_inode;
684 int err, block_count = 0, blocksize;
686 /* Determine how many transaction credits are needed */
687 blocksize = 1 << inode->i_blkbits;
688 block_count = (*offs & (blocksize - 1)) + bufsize;
689 block_count = (block_count + blocksize - 1) >> inode->i_blkbits;
691 handle = ext3_journal_start(inode,
692 block_count * EXT3_DATA_TRANS_BLOCKS(inode->i_sb) + 2);
693 if (IS_ERR(handle)) {
694 CERROR("can't start transaction for %d blocks (%d bytes)\n",
695 block_count * EXT3_DATA_TRANS_BLOCKS(inode->i_sb) + 2,
697 return PTR_ERR(handle);
700 err = fsfilt_ext3_write_handle(inode, buf, bufsize, offs, handle);
702 if (!err && force_sync)
703 handle->h_sync = 1; /* recovery likes this */
705 ext3_journal_stop(handle);
710 static int fsfilt_ext3_setup(struct super_block *sb)
712 if (!EXT3_HAS_COMPAT_FEATURE(sb,
713 EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
714 CERROR("ext3 mounted without journal\n");
719 CWARN("Enabling PDIROPS\n");
720 set_opt(EXT3_SB(sb)->s_mount_opt, PDIROPS);
721 sb->s_flags |= S_PDIROPS;
723 if (!EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
724 CWARN("filesystem doesn't have dir_index feature enabled\n");
727 static struct fsfilt_operations fsfilt_ext3_ops = {
729 .fs_owner = THIS_MODULE,
730 .fs_getlabel = fsfilt_ext3_get_label,
731 .fs_start = fsfilt_ext3_start,
732 .fs_commit = fsfilt_ext3_commit,
733 .fs_map_inode_pages = fsfilt_ext3_map_inode_pages,
734 .fs_write_record = fsfilt_ext3_write_record,
735 .fs_read_record = fsfilt_ext3_read_record,
736 .fs_setup = fsfilt_ext3_setup,
739 static int __init fsfilt_ext3_init(void)
743 fcb_cache = kmem_cache_create("fsfilt_ext3_fcb",
744 sizeof(struct fsfilt_cb_data),
747 CERROR("error allocating fsfilt journal callback cache\n");
748 GOTO(out, rc = -ENOMEM);
751 rc = fsfilt_register_ops(&fsfilt_ext3_ops);
754 kmem_cache_destroy(fcb_cache);
759 static void __exit fsfilt_ext3_exit(void)
761 fsfilt_unregister_ops(&fsfilt_ext3_ops);
762 kmem_cache_destroy(fcb_cache);
765 module_init(fsfilt_ext3_init);
766 module_exit(fsfilt_ext3_exit);
768 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
769 MODULE_DESCRIPTION("Lustre ext3 Filesystem Helper v0.1");
770 MODULE_LICENSE("GPL");