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;
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++) {
331 tgen = EXT_GENERATION(base);
332 count = ext3_ext_calc_credits_for_insert(base, path);
334 handle = ext3_journal_start(inode, count+EXT3_ALLOC_NEEDED+1);
335 if (IS_ERR(handle)) {
336 return PTR_ERR(handle);
339 if (tgen != EXT_GENERATION(base)) {
340 /* the tree has changed. so path can be invalid at moment */
341 ext3_journal_stop(handle);
345 /* In 2.6.32 kernel, ext4_ext_walk_space()'s callback func is not
346 * protected by i_data_sem as whole. so we patch it to store
347 * generation to path and now verify the tree hasn't changed */
348 down_write((&EXT4_I(inode)->i_data_sem));
350 /* validate extent, make sure the extent tree does not changed */
351 if (EXT_GENERATION(base) != path[0].p_generation) {
352 /* cex is invalid, try again */
353 up_write(&EXT4_I(inode)->i_data_sem);
354 ext3_journal_stop(handle);
359 pblock = new_blocks(handle, base, path, cex->ec_block, &count, &err);
362 EXT_ASSERT(count <= cex->ec_len);
364 /* insert new extent */
365 nex.ee_block = cpu_to_le32(cex->ec_block);
366 ext3_ext_store_pblock(&nex, pblock);
367 nex.ee_len = cpu_to_le16(count);
368 err = fsfilt_ext3_ext_insert_extent(handle, base, path, &nex, 0);
370 /* free data blocks we just allocated */
371 /* not a good idea to call discard here directly,
372 * but otherwise we'd need to call it every free() */
373 #ifdef EXT3_MB_HINT_GROUP_ALLOC
374 ext3_mb_discard_inode_preallocations(inode);
376 ext3_free_blocks(handle, inode, ext_pblock(&nex),
377 cpu_to_le16(nex.ee_len), 0);
382 * Putting len of the actual extent we just inserted,
383 * we are asking ext3_ext_walk_space() to continue
384 * scaning after that block
386 cex->ec_len = le16_to_cpu(nex.ee_len);
387 cex->ec_start = ext_pblock(&nex);
388 BUG_ON(le16_to_cpu(nex.ee_len) == 0);
389 BUG_ON(le32_to_cpu(nex.ee_block) != cex->ec_block);
392 up_write((&EXT4_I(inode)->i_data_sem));
393 ext3_journal_stop(handle);
398 CERROR("hmm. why do we find this extent?\n");
399 CERROR("initial space: %lu:%u\n",
400 bp->start, bp->init_num);
401 #ifdef EXT3_EXT_CACHE_EXTENT
402 CERROR("current extent: %u/%u/%llu %d\n",
403 cex->ec_block, cex->ec_len,
404 (unsigned long long)cex->ec_start,
407 CERROR("current extent: %u/%u/%llu\n",
408 cex->ec_block, cex->ec_len,
409 (unsigned long long)cex->ec_start);
413 if (cex->ec_block < bp->start)
414 i = bp->start - cex->ec_block;
415 if (i >= cex->ec_len)
416 CERROR("nothing to do?! i = %d, e_num = %u\n",
418 for (; i < cex->ec_len && bp->num; i++) {
419 *(bp->blocks) = cex->ec_start + i;
420 #ifdef EXT3_EXT_CACHE_EXTENT
421 if (cex->ec_type == EXT3_EXT_CACHE_EXTENT)
423 if ((cex->ec_len != 0) && (cex->ec_start != 0))
429 /* unmap any possible underlying metadata from
430 * the block device mapping. bug 6998. */
431 ll_unmap_underlying_metadata(inode->i_sb,
443 int fsfilt_map_nblocks(struct inode *inode, unsigned long block,
444 unsigned long num, unsigned long *blocks,
445 int *created, int create)
447 struct ext3_ext_base *base = inode;
451 CDEBUG(D_OTHER, "blocks %lu-%lu requested for inode %u\n",
452 block, block + num - 1, (unsigned) inode->i_ino);
455 bp.created = created;
457 bp.init_num = bp.num = num;
460 err = fsfilt_ext3_ext_walk_space(base, block, num,
461 ext3_ext_new_extent_cb, &bp);
462 ext3_ext_invalidate_cache(base);
467 int fsfilt_ext3_map_ext_inode_pages(struct inode *inode, struct page **page,
468 int pages, unsigned long *blocks,
469 int *created, int create)
471 int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
473 struct page *fp = NULL;
476 CDEBUG(D_OTHER, "inode %lu: map %d pages from %lu\n",
477 inode->i_ino, pages, (*page)->index);
479 /* pages are sorted already. so, we just have to find
480 * contig. space and process them properly */
483 /* start new extent */
488 } else if (fp->index + clen == (*page)->index) {
489 /* continue the extent */
496 /* process found extent */
497 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
498 clen * blocks_per_page, blocks,
503 /* look for next extent */
505 blocks += blocks_per_page * clen;
506 created += blocks_per_page * clen;
510 rc = fsfilt_map_nblocks(inode, fp->index * blocks_per_page,
511 clen * blocks_per_page, blocks,
517 extern int ext3_map_inode_page(struct inode *inode, struct page *page,
518 unsigned long *blocks, int *created, int create);
519 int fsfilt_ext3_map_bm_inode_pages(struct inode *inode, struct page **page,
520 int pages, unsigned long *blocks,
521 int *created, int create)
523 int blocks_per_page = CFS_PAGE_SIZE >> inode->i_blkbits;
527 for (i = 0, cr = created, b = blocks; i < pages; i++, page++) {
528 rc = ext3_map_inode_page(inode, *page, b, cr, create);
530 CERROR("ino %lu, blk %lu cr %u create %d: rc %d\n",
531 inode->i_ino, *b, *cr, create, rc);
535 b += blocks_per_page;
536 cr += blocks_per_page;
541 int fsfilt_ext3_map_inode_pages(struct inode *inode, struct page **page,
542 int pages, unsigned long *blocks,
543 int *created, int create,
544 struct mutex *optional_mutex)
548 if (EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL) {
549 rc = fsfilt_ext3_map_ext_inode_pages(inode, page, pages,
550 blocks, created, create);
553 if (optional_mutex != NULL)
554 mutex_lock(optional_mutex);
555 rc = fsfilt_ext3_map_bm_inode_pages(inode, page, pages, blocks,
557 if (optional_mutex != NULL)
558 mutex_unlock(optional_mutex);
563 int fsfilt_ext3_read(struct inode *inode, void *buf, int size, loff_t *offs)
566 struct buffer_head *bh;
567 int err, blocksize, csize, boffs, osize = size;
569 /* prevent reading after eof */
570 spin_lock(&inode->i_lock);
571 if (i_size_read(inode) < *offs + size) {
572 size = i_size_read(inode) - *offs;
573 spin_unlock(&inode->i_lock);
575 CDEBUG(D_EXT2, "size %llu is too short for read @%llu\n",
576 i_size_read(inode), *offs);
578 } else if (size == 0) {
582 spin_unlock(&inode->i_lock);
585 blocksize = 1 << inode->i_blkbits;
588 block = *offs >> inode->i_blkbits;
589 boffs = *offs & (blocksize - 1);
590 csize = min(blocksize - boffs, size);
591 bh = ext3_bread(NULL, inode, block, 0, &err);
593 CERROR("can't read block: %d\n", err);
597 memcpy(buf, bh->b_data + boffs, csize);
606 EXPORT_SYMBOL(fsfilt_ext3_read);
608 static int fsfilt_ext3_read_record(struct file * file, void *buf,
609 int size, loff_t *offs)
612 rc = fsfilt_ext3_read(file->f_dentry->d_inode, buf, size, offs);
618 int fsfilt_ext3_write_handle(struct inode *inode, void *buf, int bufsize,
619 loff_t *offs, handle_t *handle)
621 struct buffer_head *bh = NULL;
622 loff_t old_size = i_size_read(inode), offset = *offs;
623 loff_t new_size = i_size_read(inode);
625 int err = 0, blocksize = 1 << inode->i_blkbits, size, boffs;
627 while (bufsize > 0) {
631 block = offset >> inode->i_blkbits;
632 boffs = offset & (blocksize - 1);
633 size = min(blocksize - boffs, bufsize);
634 bh = ext3_bread(handle, inode, block, 1, &err);
636 CERROR("can't read/create block: %d\n", err);
640 err = ext3_journal_get_write_access(handle, bh);
642 CERROR("journal_get_write_access() returned error %d\n",
646 LASSERT(bh->b_data + boffs + size <= bh->b_data + bh->b_size);
647 memcpy(bh->b_data + boffs, buf, size);
648 err = ext3_journal_dirty_metadata(handle, bh);
650 CERROR("journal_dirty_metadata() returned error %d\n",
654 if (offset + size > new_size)
655 new_size = offset + size;
663 /* correct in-core and on-disk sizes */
664 if (new_size > i_size_read(inode)) {
665 spin_lock(&inode->i_lock);
666 if (new_size > i_size_read(inode))
667 i_size_write(inode, new_size);
668 if (i_size_read(inode) > EXT3_I(inode)->i_disksize)
669 EXT3_I(inode)->i_disksize = i_size_read(inode);
670 if (i_size_read(inode) > old_size) {
671 spin_unlock(&inode->i_lock);
672 mark_inode_dirty(inode);
674 spin_unlock(&inode->i_lock);
682 EXPORT_SYMBOL(fsfilt_ext3_write_handle);
684 static int fsfilt_ext3_write_record(struct file *file, void *buf, int bufsize,
685 loff_t *offs, int force_sync)
687 struct inode *inode = file->f_dentry->d_inode;
689 int err, block_count = 0, blocksize;
691 /* Determine how many transaction credits are needed */
692 blocksize = 1 << inode->i_blkbits;
693 block_count = (*offs & (blocksize - 1)) + bufsize;
694 block_count = (block_count + blocksize - 1) >> inode->i_blkbits;
696 handle = ext3_journal_start(inode,
697 block_count * EXT3_DATA_TRANS_BLOCKS(inode->i_sb) + 2);
698 if (IS_ERR(handle)) {
699 CERROR("can't start transaction for %d blocks (%d bytes)\n",
700 block_count * EXT3_DATA_TRANS_BLOCKS(inode->i_sb) + 2,
702 return PTR_ERR(handle);
705 err = fsfilt_ext3_write_handle(inode, buf, bufsize, offs, handle);
707 if (!err && force_sync)
708 handle->h_sync = 1; /* recovery likes this */
710 ext3_journal_stop(handle);
715 static int fsfilt_ext3_setup(struct super_block *sb)
717 if (!EXT3_HAS_COMPAT_FEATURE(sb,
718 EXT3_FEATURE_COMPAT_HAS_JOURNAL)) {
719 CERROR("ext3 mounted without journal\n");
724 CWARN("Enabling PDIROPS\n");
725 set_opt(EXT3_SB(sb)->s_mount_opt, PDIROPS);
726 sb->s_flags |= S_PDIROPS;
728 if (!EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
729 CWARN("filesystem doesn't have dir_index feature enabled\n");
732 static struct fsfilt_operations fsfilt_ext3_ops = {
734 .fs_owner = THIS_MODULE,
735 .fs_getlabel = fsfilt_ext3_get_label,
736 .fs_start = fsfilt_ext3_start,
737 .fs_commit = fsfilt_ext3_commit,
738 .fs_map_inode_pages = fsfilt_ext3_map_inode_pages,
739 .fs_write_record = fsfilt_ext3_write_record,
740 .fs_read_record = fsfilt_ext3_read_record,
741 .fs_setup = fsfilt_ext3_setup,
744 static int __init fsfilt_ext3_init(void)
748 fcb_cache = cfs_mem_cache_create("fsfilt_ext3_fcb",
749 sizeof(struct fsfilt_cb_data), 0, 0);
751 CERROR("error allocating fsfilt journal callback cache\n");
752 GOTO(out, rc = -ENOMEM);
755 rc = fsfilt_register_ops(&fsfilt_ext3_ops);
758 int err = cfs_mem_cache_destroy(fcb_cache);
759 LASSERTF(err == 0, "error destroying new cache: rc %d\n", err);
765 static void __exit fsfilt_ext3_exit(void)
769 fsfilt_unregister_ops(&fsfilt_ext3_ops);
770 rc = cfs_mem_cache_destroy(fcb_cache);
771 LASSERTF(rc == 0, "couldn't destroy fcb_cache slab\n");
774 module_init(fsfilt_ext3_init);
775 module_exit(fsfilt_ext3_exit);
777 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
778 MODULE_DESCRIPTION("Lustre ext3 Filesystem Helper v0.1");
779 MODULE_LICENSE("GPL");