1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Lustre filesystem abstraction routines
6 * Copyright (C) 2002, 2003 Cluster File Systems, Inc.
7 * Author: Andreas Dilger <adilger@clusterfs.com>
9 * This file is part of Lustre, http://www.lustre.org.
11 * Lustre is free software; you can redistribute it and/or
12 * modify it under the terms of version 2 of the GNU General Public
13 * License as published by the Free Software Foundation.
15 * Lustre is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
18 * GNU General Public License for more details.
20 * You should have received a copy of the GNU General Public License
21 * along with Lustre; if not, write to the Free Software
22 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
24 #define DEBUG_SUBSYSTEM S_FILTER
26 #include <linux/init.h>
27 #include <linux/module.h>
29 #include <linux/jbd.h>
30 #include <linux/slab.h>
31 #include <linux/pagemap.h>
32 #include <linux/quotaops.h>
33 #include <linux/ext3_fs.h>
34 #include <linux/ext3_jbd.h>
35 #include <linux/ext3_extents.h>
36 #include <linux/version.h>
37 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
38 #include <linux/locks.h>
39 #include <linux/ext3_xattr.h>
40 #include <linux/module.h>
41 #include <linux/iobuf.h>
43 #include <ext3/xattr.h>
46 #include <linux/kp30.h>
47 #include <linux/lustre_fsfilt.h>
48 #include <linux/obd.h>
49 #include <linux/obd_class.h>
50 #include <linux/lustre_smfs.h>
51 #include <linux/lustre_snap.h>
53 /* For snapfs in EXT3 flags --- FIXME will find other ways to store it*/
54 #define EXT3_COW_FL 0x00100000 /* inode is snapshot cow */
55 #define EXT3_DEL_FL 0x00200000 /* inode is deleting in snapshot */
57 #define EXT3_SNAP_ATTR "@snap"
58 #define EXT3_SNAP_GENERATION "@snap_generation"
59 #define EXT3_MAX_SNAPS 10
60 #define EXT3_MAX_SNAP_DATA (sizeof(struct snap_ea))
61 #define EXT3_SNAP_INDEX EXT3_XATTR_INDEX_LUSTRE
62 #define EXT3_SNAP_COUNT "@snapcount"
65 #define SB_FEATURE_COMPAT(sb) (EXT3_SB(sb)->s_es->s_feature_compat)
67 #define SNAP_HAS_COMPAT_FEATURE(sb,mask) \
68 (SB_FEATURE_COMPAT(sb) & cpu_to_le32(mask))
70 #define EXT3_FEATURE_COMPAT_SNAPFS 0x0010
71 #define EXT3_FEATURE_COMPAT_BLOCKCOW 0x0020
72 /*snaptable info for EXT3*/
73 #define EXT3_SNAPTABLE_EA "@snaptable"
75 /* NOTE: these macros are close dependant on the structure of snap ea */
76 #define SNAP_CNT_FROM_SIZE(size) ((((size)-sizeof(ino_t)*2)/2)/sizeof(ino_t))
77 #define SNAP_EA_SIZE_FROM_INDEX(index) (sizeof(ino_t)*2 + 2*sizeof(ino_t)*((index)+1))
79 #define SNAP_EA_INO_BLOCK_SIZE(size) (((size)-sizeof(ino_t)*2)/2)
80 #define SNAP_EA_PARENT_OFFSET(size) (sizeof(ino_t)*2 + SNAP_EA_INO_BLOCK_SIZE((size)))
82 #define EXT3_JOURNAL_START(sb, handle, blocks, rc) \
85 journal = EXT3_SB(sb)->s_journal; \
87 handle = journal_start(journal, blocks); \
89 if(IS_ERR(handle)) { \
90 CERROR("can't start transaction\n"); \
91 rc = PTR_ERR(handle); \
97 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
98 static inline void double_lock_inode(struct inode *i1, struct inode *i2)
103 double_down(&i1->i_sem, &i2->i_sem);
105 static inline void double_unlock_inode(struct inode *i1, struct inode *i2)
110 double_up(&i1->i_sem, &i2->i_sem);
113 static inline void double_lock_inode(struct inode *i1, struct inode *i2)
115 struct semaphore *s1 = &i1->i_sem;
116 struct semaphore *s2 = &i2->i_sem;
119 if ((unsigned long) s1 < (unsigned long) s2) {
120 struct semaphore *tmp = s2;
128 static inline void double_unlock_inode(struct inode *i1, struct inode *i2)
130 struct semaphore *s1 = &i1->i_sem;
131 struct semaphore *s2 = &i2->i_sem;
140 /* helper functions to manipulate field 'parent' in snap_ea */
142 set_parent_ino(struct snap_ea *pea, int size, int index, ino_t val)
144 char * p = (char*) pea;
147 offset = sizeof(ino_t)*2 + (size - sizeof(ino_t)*2)/2;
148 offset += sizeof(ino_t) * index;
149 *(ino_t*)(p+offset) = val;
154 * fsfilt_ext3_get_indirect - get a specific indirect inode from a primary inode
155 * @primary: primary (direct) inode
156 * @table: table of @slot + 1 indices in reverse chronological order
157 * @slot: starting slot number to check for indirect inode number
159 * We locate an indirect inode from a primary inode using the redirection
160 * table stored in the primary inode. Because the desired inode may actually
161 * be in a "newer" slot number than the supplied slot, we are given a table
162 * of indices in chronological order to search for the correct inode number.
163 * We walk table from @slot to 0 looking for a non-zero inode to load.
165 * To only load a specific index (and fail if it does not exist), you can
166 * pass @table = NULL, and the index number in @slot. If @slot == 0, the
167 * primary inode data is returned.
169 * We return a pointer to an inode, or an error. If the indirect inode for
170 * the given index does not exist, NULL is returned.
172 static struct inode *fsfilt_ext3_get_indirect(struct inode *primary, int *table,
175 char buf[EXT3_MAX_SNAP_DATA];
176 struct snap_ea *snaps;
178 struct inode *inode = NULL;
179 int rc = 0, index = 0;
183 if (slot < 0 || slot > EXT3_MAX_SNAPS || !primary)
186 CDEBUG(D_INODE, "ino %lu, table %p, slot %d\n", primary->i_ino, table,
188 rc = ext3_xattr_get(primary, EXT3_SNAP_INDEX, EXT3_SNAP_ATTR, buf,
190 if (rc == -ENODATA) {
193 CERROR("attribute read rc=%d \n", rc);
196 snaps = (struct snap_ea *)buf;
198 /* if table is NULL and there is a slot */
199 if( !table && slot >= 0) {
201 ino = le32_to_cpu(snaps->ino[index]);
203 inode = iget(primary->i_sb, ino);
206 /* if table is not NULL */
207 while (!inode && slot >= 0 && table) {
209 ino = le32_to_cpu(snaps->ino[index]);
211 CDEBUG(D_INODE, "snap inode at slot %d is %lu\n", slot, ino);
216 inode = iget(primary->i_sb, ino);
219 if( slot == -1 && table ) {
220 CDEBUG(D_INODE, "redirector not found, using primary\n");
221 inode = iget(primary->i_sb, primary->i_ino);
227 /* Save the indirect inode in the snapshot table of the primary inode. */
228 static int fsfilt_ext3_set_indirect(struct inode *pri, int index, ino_t ind_ino,
231 char buf[EXT3_MAX_SNAP_DATA];
232 struct snap_ea *snaps;
233 int rc = 0, inlist = 1;
235 handle_t *handle = NULL;
238 CDEBUG(D_INODE, "(ino %lu, parent %lu): saving ind %lu to index %d\n",
239 pri->i_ino, parent_ino, ind_ino, index);
241 if (index < 0 || index > MAX_SNAPS || !pri)
243 /* need lock the list before get_attr() to avoid race */
244 /* read ea at first */
245 rc = ext3_xattr_get(pri, EXT3_SNAP_INDEX ,EXT3_SNAP_ATTR,
246 buf, EXT3_MAX_SNAP_DATA);
247 if (rc == -ENODATA || rc == -ENODATA) {
248 CDEBUG(D_INODE, "no extended attributes - zeroing\n");
249 memset(buf, 0, EXT3_MAX_SNAP_DATA);
251 * To judge a inode in list, we only see if it has snap ea.
252 * So take care of snap ea of primary inodes very carefully.
253 * Is it right in snapfs EXT3, check it later?
257 } else if (rc < 0 || rc > EXT3_MAX_SNAP_DATA) {
258 GOTO(out_unlock, rc);
260 EXT3_JOURNAL_START(pri->i_sb, handle, SNAP_SETIND_TRANS_BLOCKS, rc);
262 GOTO(out_unlock, rc = PTR_ERR(handle));
264 snaps = (struct snap_ea *)buf;
265 snaps->ino[index] = cpu_to_le32 (ind_ino);
266 ea_size = EXT3_MAX_SNAP_DATA;
268 set_parent_ino(snaps, ea_size, index, cpu_to_le32(parent_ino));
270 rc = ext3_xattr_set_handle(handle, pri, EXT3_SNAP_INDEX,EXT3_SNAP_ATTR,
271 buf, EXT3_MAX_SNAP_DATA, 0);
272 ext3_mark_inode_dirty(handle, pri);
273 journal_stop(handle);
278 static int ext3_set_generation(struct inode *inode, unsigned long gen)
284 EXT3_JOURNAL_START(inode->i_sb, handle, EXT3_XATTR_TRANS_BLOCKS, err);
288 err = ext3_xattr_set_handle(handle, inode, EXT3_SNAP_INDEX,
289 EXT3_SNAP_GENERATION, (char*)&gen,
292 CERROR("ino %lu, set_ext_attr err %d\n", inode->i_ino, err);
296 journal_stop(handle);
301 * Copy inode metadata from one inode to another, excluding blocks and size.
302 * FIXME do we copy EA data - ACLs and such (excluding snapshot data)?
304 static void ext3_copy_meta(handle_t *handle, struct inode *dst, struct inode *src)
308 dst->i_mode = src->i_mode;
309 dst->i_nlink = src->i_nlink;
310 dst->i_uid = src->i_uid;
311 dst->i_gid = src->i_gid;
312 dst->i_atime = src->i_atime;
313 dst->i_mtime = src->i_mtime;
314 dst->i_ctime = src->i_ctime;
315 // dst->i_version = src->i_version;
317 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
318 dst->i_attr_flags = src->i_attr_flags;
320 dst->i_generation = src->i_generation;
321 EXT3_I(dst)->i_dtime = EXT3_I(src)->i_dtime;
322 EXT3_I(dst)->i_flags = EXT3_I(src)->i_flags | EXT3_COW_FL;
323 #ifdef EXT3_FRAGMENTS
324 EXT3_I(dst)->i_faddr = EXT3_I(src)->i_faddr;
325 EXT3_I(dst)->i_frag_no = EXT3_I(src)->i_frag_no;
326 EXT3_I(dst)->i_frag_size = EXT3_I(src)->i_frag_size;
328 if ((size = ext3_xattr_list(src, NULL, 0)) > 0) {
333 if (ext3_xattr_list(src, names, 0) < 0)
336 * the list of attribute names are stored as NUL terminated
337 * strings, with a double NUL string at the end.
340 while ((namelen = strlen(name))) {
344 /* don't copy snap data */
345 if (!strcmp(name, EXT3_SNAP_ATTR)) {
346 CDEBUG(D_INFO, "skipping %s item\n", name);
349 CDEBUG(D_INODE, "copying %s item\n", name);
350 attrlen = ext3_xattr_get(src, EXT3_SNAP_INDEX,
351 EXT3_SNAP_ATTR, NULL, 0);
354 OBD_ALLOC(buf, attrlen);
360 if (ext3_xattr_get(src, EXT3_SNAP_INDEX,
361 EXT3_SNAP_ATTR, buf, attrlen) < 0)
363 if (ext3_xattr_set_handle(handle, dst, EXT3_SNAP_INDEX,
364 EXT3_SNAP_ATTR, buf, attrlen,
367 OBD_FREE(buf, attrlen);
368 name += namelen + 1; /* skip name and trailing NUL */
372 static int ext3_copy_reg_block(struct inode *dst, struct inode *src, int blk)
374 struct page *src_page, *dst_page;
375 loff_t offset = blk << src->i_sb->s_blocksize_bits;
376 unsigned long index = offset >> PAGE_CACHE_SHIFT;
380 /*read the src page*/
381 src_page = grab_cache_page(src->i_mapping, index);
382 if (src_page == NULL)
385 if (!PageUptodate(src_page)) {
386 rc = src->i_mapping->a_ops->readpage(NULL, src_page);
388 page_cache_release(src_page);
395 dst_page = grab_cache_page(dst->i_mapping, index);
396 if (dst_page == NULL)
397 GOTO(src_page_unlock, rc = -ENOMEM);
400 rc = dst->i_mapping->a_ops->prepare_write(NULL, dst_page, 0,
401 PAGE_CACHE_SIZE - 1);
403 GOTO(dst_page_unlock, rc = -EFAULT);
404 memcpy(page_address(dst_page), page_address(src_page), PAGE_CACHE_SIZE);
406 flush_dcache_page(dst_page);
408 rc = dst->i_mapping->a_ops->commit_write(NULL, dst_page, 0,
409 PAGE_CACHE_SIZE - 1);
414 unlock_page(dst_page);
415 page_cache_release(dst_page);
418 page_cache_release(src_page);
421 static int ext3_copy_dir_block(struct inode *dst, struct inode *src, int blk)
423 struct buffer_head *bh_dst = NULL, *bh_src = NULL;
425 handle_t *handle = NULL;
428 EXT3_JOURNAL_START(dst->i_sb, handle, SNAP_COPYBLOCK_TRANS_BLOCKS, rc);
432 bh_src = ext3_bread(handle, src, blk, 0, &rc);
434 CERROR("rcor for src blk %d, rcor %d\n", blk, rc);
435 GOTO(exit_relese, rc);
437 bh_dst = ext3_getblk(handle, dst, blk, 1, &rc);
439 CERROR("rcor for dst blk %d, rcor %d\n", blk, rc);
440 GOTO(exit_relese, rc);
442 CDEBUG(D_INODE, "copy block %lu to %lu (%ld bytes)\n",
443 bh_src->b_blocknr, bh_dst->b_blocknr, src->i_sb->s_blocksize);
445 ext3_journal_get_write_access(handle, bh_dst);
446 memcpy(bh_dst->b_data, bh_src->b_data, src->i_sb->s_blocksize);
447 ext3_journal_dirty_metadata(handle, bh_dst);
451 if (bh_src) brelse(bh_src);
452 if (bh_dst) brelse(bh_dst);
454 journal_stop(handle);
457 /* fsfilt_ext3_copy_block - copy one data block from inode @src to @dst.
458 No lock here. User should do the lock.
459 User should check the return value to see if the result is correct.
461 1: The block has been copied successfully
462 0: No block is copied, usually this is because src has no such blk
466 static int fsfilt_ext3_copy_block (struct inode *dst, struct inode *src, int blk)
470 CDEBUG(D_INODE, "copy blk %d from %lu to %lu \n", blk, src->i_ino,
473 * ext3_getblk() require handle!=NULL
475 if (S_ISREG(src->i_mode)) {
476 rc = ext3_copy_reg_block(dst, src, blk);
478 rc = ext3_copy_dir_block(dst, src, blk);
484 static inline int ext3_has_ea(struct inode *inode)
486 return (EXT3_I(inode)->i_file_acl != 0);
488 /* XXXThis function has a very bad effect to
489 * the performance of filesystem,
490 * will find another way to fix it
492 static void fs_flushinval_pages(handle_t *handle, struct inode* inode)
494 if (inode->i_blocks > 0 && inode->i_mapping) {
495 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
496 fsync_inode_data_buffers(inode);
498 truncate_inode_pages(inode->i_mapping, 0);
501 /* ext3_migrate_data:
502 * MOVE all the data blocks from inode src to inode dst as well as
503 * COPY all attributes(meta data) from inode src to inode dst.
504 * For extended attributes(EA), we COPY all the EAs but skip the Snap EA from
505 * src to dst. If the dst has Snap EA, then we CAN'T overwrite it. We CAN'T
506 * copy the src Snap EA. XXX for EA, can we change it to MOVE all the EAs
507 * (exclude Snap EA) to dst and copy it back to src ? This is for LAN free
510 static int ext3_migrate_data(handle_t *handle, struct inode *dst,
513 unsigned long err = 0;
514 /* 512 byte disk blocks per inode block */
515 int bpib = src->i_sb->s_blocksize >> 9;
522 if (dst->i_ino == src->i_ino)
525 fs_flushinval_pages(handle, src);
527 ext3_copy_meta(handle, dst, src);
529 CDEBUG(D_INODE, "migrating data blocks from %lu to %lu\n",
530 src->i_ino, dst->i_ino);
531 /* Can't check blocks in case of EAs */
533 memcpy(EXT3_I(dst)->i_data, EXT3_I(src)->i_data,
534 sizeof(EXT3_I(src)->i_data));
535 memset(EXT3_I(src)->i_data, 0, sizeof(EXT3_I(src)->i_data));
537 ext3_discard_prealloc(src);
539 dst->i_size = EXT3_I(dst)->i_disksize = EXT3_I(src)->i_disksize;
540 src->i_size = EXT3_I(src)->i_disksize = 0;
542 dst->i_blocks = src->i_blocks;
544 /* Check EA blocks here to modify i_blocks correctly */
545 if(ext3_has_ea (src)) {
546 src->i_blocks += bpib;
547 if( ! ext3_has_ea (dst) )
548 if( dst->i_blocks >= bpib )
549 dst->i_blocks -= bpib;
551 if( ext3_has_ea (dst))
552 dst->i_blocks += bpib;
555 CDEBUG(D_INODE, "migrate data from ino %lu to ino %lu\n", src->i_ino,
557 ext3_mark_inode_dirty(handle, src);
558 ext3_mark_inode_dirty(handle, dst);
562 static handle_t * ext3_copy_data(handle_t *handle, struct inode *dst,
563 struct inode *src, int *has_orphan)
565 unsigned long blocks, blk, cur_blks;
566 int low_credits, save_ref;
570 blocks =(src->i_size + src->i_sb->s_blocksize-1) >>
571 src->i_sb->s_blocksize_bits;
572 low_credits = handle->h_buffer_credits - SNAP_BIGCOPY_TRANS_BLOCKS;
574 CDEBUG(D_INODE, "%lu blocks need to be copied,low credits limit %d\n",
575 blocks, low_credits);
577 for (blk = 0, cur_blks= dst->i_blocks; blk < blocks; blk++) {
578 if (!ext3_bmap(src->i_mapping, blk))
580 if(handle->h_buffer_credits <= low_credits) {
581 int needed = (blocks - blk) * EXT3_DATA_TRANS_BLOCKS;
582 if (needed > 4 * SNAP_COPYBLOCK_TRANS_BLOCKS)
583 needed = 4 * SNAP_COPYBLOCK_TRANS_BLOCKS;
584 if (journal_extend(handle, needed)) {
585 CDEBUG(D_INFO, "create_indirect:fail to extend "
586 "journal, restart trans\n");
589 CDEBUG(D_INODE, "add orphan ino %lu"
590 "nlink %d to orphan list \n",
591 dst->i_ino, dst->i_nlink);
592 ext3_orphan_add(handle, dst);
595 EXT3_I(dst)->i_disksize =
596 blk * dst->i_sb->s_blocksize;
597 dst->i_blocks = cur_blks;
598 dst->i_mtime = CURRENT_TIME;
599 ext3_mark_inode_dirty(handle, dst);
601 * We can be sure the last handle was stoped
602 * ONLY if the handle's reference count is 1
604 save_ref = handle->h_ref;
606 if(journal_stop(handle) ){
607 CERROR("fail to stop journal\n");
611 EXT3_JOURNAL_START(dst->i_sb, handle,
612 low_credits + needed, err);
614 handle->h_ref = save_ref;
617 if (fsfilt_ext3_copy_block( dst, src, blk) < 0 )
619 cur_blks += dst->i_sb->s_blocksize / 512;
622 dst->i_size = EXT3_I(dst)->i_disksize = src->i_size;
625 /*Here delete the data of that pri inode
626 *FIXME later, should throw the blocks of
627 *primary inode directly
629 static int ext3_throw_inode_data(handle_t *handle, struct inode *inode)
631 struct inode *tmp = NULL;
633 tmp = ext3_new_inode(handle, inode, (int)inode->i_mode, 0);
635 CERROR("ext3_new_inode error\n");
638 double_lock_inode(inode, tmp);
639 ext3_migrate_data(handle, tmp, inode);
640 double_unlock_inode(inode, tmp);
646 * fsfilt_ext3_create_indirect - copy data, attributes from primary to new indir inode
647 * @pri: primary (source) inode
648 * @index: index in snapshot table where indirect inode should be stored
649 * @delete: flag that the primary inode is being deleted
651 * We copy all of the data blocks from the @*src inode to the @*dst inode, as
652 * well as copying the attributes from @*src to @*dst. If @delete == 1, then
653 * the primary inode will only be a redirector and will appear deleted.
655 * FIXME do we move EAs, only non-snap EAs, what?
656 * FIXME we could do readpage/writepage, but we would have to handle block
657 * allocation then, and it ruins sparse files for 1k/2k filesystems,
658 * at the expense of doing a memcpy.
660 static struct inode* fsfilt_ext3_create_indirect(struct inode *pri, int index,
662 struct inode* parent,
665 struct inode *ind = NULL;
666 handle_t *handle = NULL;
671 if( pri == EXT3_SB(pri->i_sb)->s_journal_inode ){
672 CERROR("TRY TO COW JOUNRAL\n");
673 RETURN(ERR_PTR(-EINVAL));
675 CDEBUG(D_INODE, "creating indirect inode for %lu at index %d, %s pri\n",
676 pri->i_ino, index, del ? "deleting" : "preserve");
678 ind = fsfilt_ext3_get_indirect(pri, NULL, index);
680 EXT3_JOURNAL_START(pri->i_sb, handle, SNAP_CREATEIND_TRANS_BLOCKS,
683 RETURN(ERR_PTR(err));
684 /* XXX ? We should pass an err argument to get_indirect and precisely
685 * detect the errors, for some errors, we should exit right away.
688 /* if the option is SNAP_DEL_PRI_WITH_IND and there is an indirect,
689 * we just free the primary data blocks and mark this inode delete
691 if((del) && ind && !IS_ERR(ind)) {
692 /* for directory, we don't free the data blocks,
693 * or ext3_rmdir will report errors "bad dir, no data blocks"
695 CDEBUG(D_INODE, "del==SNAP_DEL_PRI_WITH_IND && ind\n");
696 if(!S_ISDIR(pri->i_mode)) {
697 err = ext3_throw_inode_data(handle, pri);
702 EXT3_I(pri)->i_dtime = LTIME_S(CURRENT_TIME);
703 ext3_mark_inode_dirty(handle, pri);
707 if (ind && !IS_ERR(ind)) {
708 CDEBUG(D_INODE, "existing indirect ino %lu for %lu: index %d\n",
709 ind->i_ino, pri->i_ino, index);
714 /* XXX: check this, ext3_new_inode, the first arg should be "dir" */
715 ind = ext3_new_inode(handle, pri, (int)pri->i_mode, 0);
719 CDEBUG(D_INODE, "got new inode %lu\n", ind->i_ino);
720 ind->i_rdev = pri->i_rdev;
721 ind->i_op = pri->i_op;
724 memcpy(ind->i_op, pri->i_op, sizeof(*pri->i_op));
725 memcpy(ind->i_fop, pri->i_fop, sizeof(*pri->i_fop));
726 memcpy(ind->i_mapping->a_ops, pri->i_mapping->a_ops,
727 sizeof(*pri->i_mapping->a_ops));
729 ext3_set_generation(ind, (unsigned long)gen);
730 /* If we are deleting the primary inode, we want to ensure that it is
731 * written to disk with a non-zero link count, otherwise the next iget
732 * and iput will mark the inode as free (which we don't want, we want
733 * it to stay a redirector). We fix this in ext3_destroy_indirect()
734 * when the last indirect inode is removed.
736 * We then do what ext3_delete_inode() does so that the metadata will
737 * appear the same as a deleted inode, and we can detect it later.
740 CDEBUG(D_INODE, "deleting primary inode\n");
743 err = ext3_migrate_data(handle, ind, pri);
745 GOTO(exit_unlock, err);
747 err = fsfilt_ext3_set_indirect(pri, index, ind->i_ino, parent->i_ino);
749 GOTO(exit_unlock, err);
751 /* XXX for directory, we copy the block back
752 * or ext3_rmdir will report errors "bad dir, no data blocks"
754 if( S_ISDIR(pri->i_mode)) {
755 handle = ext3_copy_data(handle, pri, ind, &has_orphan);
757 GOTO(exit_unlock, err= -EINVAL);
760 EXT3_I(pri)->i_flags |= EXT3_DEL_FL;
761 EXT3_I(ind)->i_flags |= EXT3_COW_FL;
762 if(S_ISREG(pri->i_mode)) pri->i_nlink = 1;
763 EXT3_I(pri)->i_dtime = LTIME_S(CURRENT_TIME);
764 //EXT3_I(pri)->i_generation++;
765 ext3_mark_inode_dirty(handle, pri);
766 ext3_mark_inode_dirty(handle, ind);
770 err = ext3_migrate_data(handle, ind, pri);
774 /* for regular files we do blocklevel COW's maybe */
775 if (EXT3_HAS_COMPAT_FEATURE(pri->i_sb, EXT3_FEATURE_COMPAT_BLOCKCOW)
776 && S_ISREG(pri->i_mode)) {
778 CDEBUG(D_INODE, "ino %lu, do block cow\n", pri->i_ino);
779 /* because after migrate_data , pri->i_size is 0 */
780 pri->i_size = ind->i_size;
783 int bpib = pri->i_sb->s_blocksize >> 9;
784 CDEBUG(D_INODE, "ino %lu, do file cow\n", pri->i_ino);
786 /* XXX: can we do this better?
787 * If it's a fast symlink, we should copy i_data back!
788 * The criteria to determine a fast symlink is:
789 * 1) it's a link and its i_blocks is 0
790 * 2) it's a link and its i_blocks is bpib ( the case
791 * it has been cowed and has ea )
793 if( S_ISLNK(ind->i_mode) && ((ind->i_blocks == 0) ||
794 (ext3_has_ea(ind) && ind->i_blocks == bpib))) {
795 CDEBUG(D_INODE, "ino %lu is fast symlink\n", pri->i_ino);
796 memcpy(EXT3_I(pri)->i_data, EXT3_I(ind)->i_data,
797 sizeof(EXT3_I(ind)->i_data));
798 pri->i_size = ind->i_size;
801 handle = ext3_copy_data(handle, pri, ind, &has_orphan);
803 GOTO(exit_unlock, err);
806 /* set cow flag for ind */
807 EXT3_I(ind)->i_flags |= EXT3_COW_FL;
808 EXT3_I(pri)->i_flags &= ~EXT3_COW_FL;
810 ext3_mark_inode_dirty(handle, pri);
811 ext3_mark_inode_dirty(handle, ind);
813 err = fsfilt_ext3_set_indirect(pri, index, ind->i_ino, parent->i_ino);
815 GOTO(exit_unlock, err);
819 if (!EXT3_HAS_COMPAT_FEATURE(pri->i_sb,
820 EXT3_FEATURE_COMPAT_SNAPFS)) {
821 lock_super(pri->i_sb);
822 ext3_journal_get_write_access(handle, EXT3_SB(pri->i_sb)->s_sbh);
823 EXT3_SB(pri->i_sb)->s_es->s_feature_compat |=
824 cpu_to_le32(EXT3_FEATURE_COMPAT_SNAPFS);
825 ext3_journal_dirty_metadata(handle, EXT3_SB(pri->i_sb)->s_sbh);
826 pri->i_sb->s_dirt = 1;
827 unlock_super(pri->i_sb);
830 CDEBUG(D_INODE, "del %lu nlink %d from orphan list\n",
831 ind->i_ino, ind->i_nlink);
832 ext3_orphan_del(handle, ind);
834 journal_stop(handle);
843 CDEBUG(D_INODE, "del %lu nlink %d from orphan list\n",
844 ind->i_ino, ind->i_nlink);
845 ext3_orphan_del(handle, ind);
848 journal_stop(handle);
850 RETURN(ERR_PTR(err));
853 static int fsfilt_ext3_snap_feature (struct super_block *sb, int feature, int op) {
860 case SNAP_SET_FEATURE:
861 case SNAP_CLEAR_FEATURE:
862 EXT3_JOURNAL_START(sb, handle, 1, rc);
866 ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
867 if (op == SNAP_SET_FEATURE)
868 SB_FEATURE_COMPAT(sb) |= cpu_to_le32(feature);
870 SB_FEATURE_COMPAT(sb) &= ~cpu_to_le32(feature);
872 ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
874 journal_stop(handle);
876 case SNAP_HAS_FEATURE:
877 /*FIXME should lock super or not*/
878 rc = SNAP_HAS_COMPAT_FEATURE(sb, feature);
886 * is_redirector - determines if a primary inode is a redirector
887 * @inode: primary inode to test
889 * Returns 1 if the inode is a redirector, 0 otherwise.
891 static int fsfilt_ext3_is_redirector(struct inode *inode)
893 int is_redirector = 0;
897 rc = ext3_xattr_get(inode, EXT3_SNAP_INDEX ,EXT3_SNAP_ATTR,
899 if (rc > 0 && rc <= MAX_SNAP_DATA)
901 CDEBUG(D_INODE, "inode %lu %s redirector\n", inode->i_ino,
902 is_redirector ? "is" : "isn't");
903 RETURN(is_redirector);
905 /*if it's indirect inode or not */
906 static int fsfilt_ext3_is_indirect(struct inode *inode)
908 if (EXT3_I(inode)->i_flags |= EXT3_COW_FL)
914 /* get the indirect ino at index of the primary inode
915 * return value: postive: indirect ino number
916 * negative or 0: error
918 static ino_t fsfilt_ext3_get_indirect_ino(struct super_block *sb,
919 ino_t primary_ino, int index)
921 char buf[EXT3_MAX_SNAP_DATA];
922 struct inode *primary = NULL;
923 struct snap_ea *snaps;
927 if (index < 0 || index > EXT3_MAX_SNAPS || !primary)
929 primary = iget(sb, primary_ino);
933 CERROR("attribute read error=%d", err);
934 GOTO (err_free, ino = err);
936 err = ext3_xattr_get(primary, EXT3_SNAP_INDEX, EXT3_SNAP_ATTR,
937 buf, EXT3_MAX_SNAP_DATA);
938 if (err == -ENODATA) {
939 GOTO(err_free, ino = -ENODATA);
940 } else if (err < 0) {
941 CERROR(" attribute read error err=%d\n", err);
942 GOTO(err_free, ino = err);
944 snaps = (struct snap_ea *)buf;
945 ino = le32_to_cpu (snaps->ino[index]);
946 CDEBUG(D_INODE, "snap ino for %ld at index %d is %lu\n",
947 primary->i_ino, index, ino);
955 /* The following functions are used by destroy_indirect */
956 #define inode_bmap(inode, nr) (EXT3_I(inode)->i_data[(nr)])
957 #define inode_setbmap(inode, nr, physical) (EXT3_I(inode)->i_data[(nr)]=(physical))
958 static inline int block_bmap(struct buffer_head * bh, int nr)
964 tmp = le32_to_cpu(((u32 *) bh->b_data)[nr]);
969 static inline int block_setbmap(handle_t *handle, struct buffer_head * bh,
970 int nr, int physical)
975 ext3_journal_get_write_access(handle, bh);
976 ((u32 *) bh->b_data)[nr] = cpu_to_le32(physical);
977 ext3_journal_dirty_metadata(handle, bh);
982 static int ext3_migrate_block(handle_t *handle, struct inode * dst,
983 struct inode *src, int block)
985 int i1_d=0, i1_s=0, i2_d=0, i2_s=0, i3_d=0, i3_s=0;
986 int addr_per_block = EXT3_ADDR_PER_BLOCK(src->i_sb);
987 int addr_per_block_bits = EXT3_ADDR_PER_BLOCK_BITS(src->i_sb);
992 CWARN("ext3_migrate_block block < 0 %p \n", src->i_sb);
995 if (block >= EXT3_NDIR_BLOCKS + addr_per_block +
996 (1 << (addr_per_block_bits * 2)) +
997 ((1 << (addr_per_block_bits * 2)) << addr_per_block_bits)) {
998 CWARN("ext3_migrate_block block > big %p \n", src->i_sb);
1001 /* EXT3_NDIR_BLOCK */
1002 if (block < EXT3_NDIR_BLOCKS) {
1003 if(inode_bmap(dst, block))
1006 if( (physical = inode_bmap(src, block)) ) {
1007 inode_setbmap (dst, block, physical);
1008 inode_setbmap (src, block, 0);
1015 /* EXT3_IND_BLOCK */
1016 block -= EXT3_NDIR_BLOCKS;
1017 if (block < addr_per_block) {
1018 i1_d = inode_bmap (dst, EXT3_IND_BLOCK);
1020 physical = inode_bmap(src, EXT3_IND_BLOCK);
1022 inode_setbmap (dst, EXT3_IND_BLOCK, physical);
1023 inode_setbmap (src, EXT3_IND_BLOCK, 0);
1029 if(block_bmap(sb_bread(dst->i_sb, i1_d), block))
1032 i1_s = inode_bmap (src, EXT3_IND_BLOCK);
1033 if( !i1_s) RETURN(0);
1035 physical = block_bmap(sb_bread(src->i_sb, i1_s), block);
1038 block_setbmap(handle, sb_bread(dst->i_sb, i1_d),block,
1040 block_setbmap(handle, sb_bread(src->i_sb, i1_s),block,0);
1046 /* EXT3_DIND_BLOCK */
1047 block -= addr_per_block;
1048 if (block < (1 << (addr_per_block_bits * 2))) {
1049 i1_d = inode_bmap (dst, EXT3_DIND_BLOCK);
1050 i1_s = inode_bmap (src, EXT3_DIND_BLOCK);
1052 if( (physical = inode_bmap(src, EXT3_DIND_BLOCK)) ) {
1053 inode_setbmap (dst, EXT3_DIND_BLOCK, physical);
1054 inode_setbmap (src, EXT3_DIND_BLOCK, 0);
1060 i2_d = block_bmap (sb_bread (dst->i_sb, i1_d),
1061 block >> addr_per_block_bits);
1065 if(!i1_s) RETURN(0);
1067 physical = block_bmap(sb_bread (src->i_sb, i1_s),
1068 block >> addr_per_block_bits);
1070 block_setbmap(handle, sb_bread(dst->i_sb, i1_d),
1071 block >> addr_per_block_bits,
1073 block_setbmap(handle, sb_bread(src->i_sb, i1_s),
1074 block >> addr_per_block_bits, 0);
1080 physical = block_bmap(sb_bread(dst->i_sb, i2_d),
1081 block & (addr_per_block - 1));
1085 i2_s = block_bmap (sb_bread(src->i_sb, i1_s),
1086 block >> addr_per_block_bits);
1087 if(!i2_s) RETURN(0);
1089 physical = block_bmap(sb_bread(src->i_sb, i2_s),
1090 block & (addr_per_block - 1));
1092 block_setbmap(handle, sb_bread(dst->i_sb, i2_d),
1093 block & (addr_per_block - 1), physical);
1094 block_setbmap(handle, sb_bread(src->i_sb, i2_s),
1095 block & (addr_per_block - 1), 0);
1103 /* EXT3_TIND_BLOCK */
1104 block -= (1 << (addr_per_block_bits * 2));
1105 i1_d = inode_bmap (dst, EXT3_TIND_BLOCK);
1106 i1_s = inode_bmap (src, EXT3_TIND_BLOCK);
1108 if((physical = inode_bmap(src, EXT3_TIND_BLOCK)) )
1109 inode_setbmap (dst, EXT3_TIND_BLOCK, physical);
1113 i2_d = block_bmap(sb_bread (dst->i_sb, i1_d),
1114 block >> (addr_per_block_bits * 2));
1116 if(i1_s) i2_s = block_bmap(sb_bread(src->i_sb, i1_s),
1117 block >> (addr_per_block_bits * 2));
1120 if( !i1_s) RETURN(0);
1122 physical = block_bmap(sb_bread (src->i_sb, i1_s),
1123 block >> (addr_per_block_bits * 2));
1125 block_setbmap(handle, sb_bread (dst->i_sb, i1_d),
1126 block >> (addr_per_block_bits * 2), physical);
1127 block_setbmap(handle, sb_bread (src->i_sb, i1_s),
1128 block >> (addr_per_block_bits * 2), 0);
1134 i3_d = block_bmap (sb_bread (dst->i_sb, i2_d),
1135 (block >> addr_per_block_bits) & (addr_per_block - 1));
1136 if( i2_s) i3_s = block_bmap (sb_bread (src->i_sb, i2_s),
1137 (block >> addr_per_block_bits) & (addr_per_block - 1));
1140 if (!i2_s) RETURN(0);
1141 physical = block_bmap (sb_bread (src->i_sb, i2_s),
1142 (block >> addr_per_block_bits) & (addr_per_block - 1));
1144 block_setbmap (handle, sb_bread (dst->i_sb, i2_d),
1145 (block >> addr_per_block_bits) &
1146 (addr_per_block - 1), physical);
1147 block_setbmap (handle, sb_bread (src->i_sb, i2_s),
1148 (block >> addr_per_block_bits) &
1149 (addr_per_block - 1),0);
1155 physical = block_bmap (sb_bread (dst->i_sb, i3_d),
1156 block & (addr_per_block - 1)) ;
1162 physical = block_bmap(sb_bread(src->i_sb, i3_s),
1163 block & (addr_per_block - 1));
1165 block_setbmap (handle, sb_bread (dst->i_sb, i3_d),
1166 block & (addr_per_block - 1), physical);
1167 block_setbmap (handle, sb_bread (src->i_sb, i3_s),
1168 block & (addr_per_block - 1), 0);
1176 /* Generate i_blocks from blocks for an inode .
1177 * We also calculate EA block here.
1179 static unsigned long calculate_i_blocks(struct inode *inode, int blocks)
1181 /* 512 byte disk blocks per inode block */
1182 int bpib = inode->i_sb->s_blocksize >> 9;
1183 int addr_per_block = EXT3_ADDR_PER_BLOCK(inode->i_sb);
1184 unsigned long i_blocks = 0;
1185 int i=0, j=0, meta_blocks = 0;
1191 /* re-calculate blocks here */
1192 blocks = (inode->i_size + inode->i_sb->s_blocksize-1)
1193 >> inode->i_sb->s_blocksize_bits;
1196 /* calculate data blocks */
1197 for(i = 0; i < blocks; i++) {
1198 if(ext3_bmap(inode->i_mapping, i))
1201 /* calculate meta blocks */
1202 blocks -= EXT3_NDIR_BLOCKS;
1205 blocks -= addr_per_block;
1207 if( blocks > 0 ) meta_blocks++;
1210 while( (blocks > 0) && (i < addr_per_block) ) {
1212 blocks -= addr_per_block;
1216 if ( blocks > 0 ) meta_blocks += 2;
1219 while( blocks > 0) {
1221 blocks -= addr_per_block;
1223 if(i >= addr_per_block ) {
1227 if( j >= addr_per_block) {
1232 /* calculate EA blocks */
1233 if(ext3_has_ea(inode))
1236 i_blocks += meta_blocks * bpib;
1237 CDEBUG(D_INODE, "ino %lu, get i_blocks %lu\n", inode->i_ino, i_blocks);
1243 * fsfilt_ext3_destroy_indirect - delete an indirect inode from the table
1244 * @pri: primary inode
1245 * @ind: indirect inode
1246 * @index: index of inode that should be deleted
1248 * We delete the @*ind inode, and remove it from the snapshot table. If @*ind
1249 * is NULL, we use the inode at @index.
1251 static int fsfilt_ext3_destroy_indirect(struct inode *pri, int index,
1252 struct inode *next_ind)
1254 char buf[EXT3_MAX_SNAP_DATA];
1255 struct snap_ea *snaps;
1257 int save = 0, i=0, err = 0;
1258 handle_t *handle=NULL;
1261 if (index < 0 || index > EXT3_MAX_SNAPS)
1264 if( pri == EXT3_SB(pri->i_sb)->s_journal_inode ){
1265 CERROR("TRY TO DESTROY JOURNAL'S IND\n");
1269 err = ext3_xattr_get(pri, EXT3_SNAP_INDEX, EXT3_SNAP_ATTR,
1270 buf, EXT3_MAX_SNAP_DATA);
1272 CERROR("inode %lu attribute read error\n", pri->i_ino);
1276 snaps = (struct snap_ea *)buf;
1277 if ( !snaps->ino[index] ) {
1278 CERROR("for pri ino %lu, index %d, redirect ino is 0\n",
1283 CDEBUG(D_INODE, "for pri ino %lu, reading inode %lu at index %d\n",
1284 pri->i_ino, (ulong)le32_to_cpu(snaps->ino[index]), index);
1286 ind = iget(pri->i_sb, le32_to_cpu (snaps->ino[index]));
1288 if ( !ind || IS_ERR(ind) || is_bad_inode(ind))
1291 CDEBUG(D_INODE, "iget ind %lu, ref count = %d\n",
1292 ind->i_ino, atomic_read(&ind->i_count));
1294 EXT3_JOURNAL_START(pri->i_sb, handle, SNAP_DESTROY_TRANS_BLOCKS, err);
1299 /* if it's block level cow, first copy the blocks back */
1300 if (EXT3_HAS_COMPAT_FEATURE(pri->i_sb, EXT3_FEATURE_COMPAT_BLOCKCOW) &&
1301 S_ISREG(pri->i_mode)) {
1308 double_lock_inode(next_ind, ind);
1310 blocks = (next_ind->i_size + next_ind->i_sb->s_blocksize-1)
1311 >> next_ind->i_sb->s_blocksize_bits;
1313 CDEBUG(D_INODE, "migrate block back from ino %lu to %lu\n",
1314 ind->i_ino, next_ind->i_ino);
1316 for(i = 0; i < blocks; i++) {
1317 if( ext3_bmap(next_ind->i_mapping, i) )
1319 if( !ext3_bmap(ind->i_mapping, i) )
1321 ext3_migrate_block(handle, next_ind, ind, i) ;
1323 /* Now re-compute the i_blocks */
1324 /* XXX shall we take care of ind here? probably not */
1325 next_ind->i_blocks = calculate_i_blocks( next_ind, blocks);
1326 ext3_mark_inode_dirty(handle, next_ind);
1328 if (next_ind == pri)
1331 double_unlock_inode(next_ind, ind);
1334 CDEBUG(D_INODE, "delete indirect ino %lu\n", ind->i_ino);
1335 CDEBUG(D_INODE, "iput ind %lu, ref count = %d\n", ind->i_ino,
1336 atomic_read(&ind->i_count));
1341 snaps->ino[index] = cpu_to_le32(0);
1342 for (i = 0; i < EXT3_MAX_SNAPS; i++)
1343 save += snaps->ino[i];
1346 /*Should we remove snap feature here*/
1348 * If we are deleting the last indirect inode, and the primary inode
1349 * has already been deleted, then mark the primary for deletion also.
1350 * Otherwise, if we are deleting the last indirect inode remove the
1351 * snaptable from the inode. XXX
1353 if (!save && EXT3_I(pri)->i_dtime) {
1354 CDEBUG(D_INODE, "deleting primary %lu\n", pri->i_ino);
1356 /* reset err to 0 now */
1359 CDEBUG(D_INODE, "%s redirector table\n",
1360 save ? "saving" : "deleting");
1361 err = ext3_xattr_set_handle(handle, pri, EXT3_SNAP_INDEX,
1362 EXT3_SNAP_ATTR, save ? buf : NULL,
1363 EXT3_MAX_SNAP_DATA, 0);
1364 ext3_mark_inode_dirty(handle, pri);
1366 journal_stop(handle);
1371 /* restore a primary inode with the indirect inode at index */
1372 static int fsfilt_ext3_restore_indirect(struct inode *pri, int index)
1376 handle_t *handle = NULL;
1379 if (index < 0 || index > EXT3_MAX_SNAPS)
1382 if( pri == EXT3_SB(pri->i_sb)->s_journal_inode ){
1383 CERROR("TRY TO RESTORE JOURNAL\n");
1386 CDEBUG(D_INODE, "pri ino %lu, index %d\n", pri->i_ino, index);
1388 ind = fsfilt_ext3_get_indirect(pri, NULL, index);
1393 CDEBUG(D_INODE, "restore ino %lu to %lu\n", pri->i_ino, ind->i_ino);
1395 EXT3_JOURNAL_START(pri->i_sb, handle, SNAP_RESTORE_TRANS_BLOCKS, err);
1398 /* first destroy all the data blocks in primary inode */
1399 /* XXX: check this, ext3_new_inode, the first arg should be "dir" */
1400 err = ext3_throw_inode_data(handle, pri);
1402 CERROR("restore_indirect, new_inode err\n");
1405 double_lock_inode(pri, ind);
1406 ext3_migrate_data(handle, pri, ind);
1407 EXT3_I(pri)->i_flags &= ~EXT3_COW_FL;
1408 ext3_mark_inode_dirty(handle, pri);
1409 double_unlock_inode(pri, ind);
1412 //fsfilt_ext3_destroy_indirect(pri, index);
1413 journal_stop(handle);
1419 * ext3_snap_iterate - iterate through all of the inodes
1420 * @sb: filesystem superblock
1421 * @repeat: pointer to function called on each valid inode
1422 * @start: inode to start iterating at
1423 * @priv: private data to the caller/repeat function
1425 * If @start is NULL, then we do not return an inode pointer. If @*start is
1426 * NULL, then we start at the beginning of the filesystem, and iterate over
1427 * all of the inodes in the system. If @*start is non-NULL, then we start
1428 * iterating at this inode.
1430 * We call the repeat function for each inode that is in use. The repeat
1431 * function must check if this is a redirector (with is_redirector) if it
1432 * only wants to operate on redirector inodes. If there is an error or
1433 * the repeat function returns non-zero, we return the last inode operated
1434 * on in the @*start parameter. This allows the caller to restart the
1435 * iteration at this inode if desired, by returning a positive value.
1436 * Negative return values indicate an error.
1438 * NOTE we cannot simply traverse the existing filesystem tree from the root
1439 * inode, as there may be disconnected trees from deleted files/dirs
1441 * FIXME If there was a list of inodes with EAs, we could simply walk the list
1442 * intead of reading every inode. This is an internal implementation issue.
1445 static int ext3_iterate_all(struct super_block *sb,
1446 int (*repeat)(struct inode *inode,void *priv),
1447 struct inode **start, void *priv)
1449 struct inode *tmp = NULL;
1450 int gstart, gnum, err = 0;
1451 ino_t istart, ibase;
1457 *start = iget(sb, EXT3_ROOT_INO);
1459 GOTO(exit, err = -ENOMEM);
1461 if (is_bad_inode(*start))
1462 GOTO(exit, err = -EIO);
1464 if ((*start)->i_ino > le32_to_cpu(EXT3_SB(sb)->s_es->s_inodes_count)) {
1465 CERROR("invalid starting inode %ld\n",(*start)->i_ino);
1466 GOTO(exit, err = -EINVAL);
1468 if ((*start)->i_ino < EXT3_FIRST_INO(sb)) {
1469 if ((err = (*repeat)(*start, priv) != 0))
1472 *start = iget(sb, EXT3_FIRST_INO(sb));
1474 GOTO(exit, err = -ENOMEM);
1475 if (is_bad_inode(*start))
1476 GOTO(exit, err = -EIO);
1479 gstart = ((*start)->i_ino - 1) / EXT3_INODES_PER_GROUP(sb);
1480 istart = ((*start)->i_ino - 1) % EXT3_INODES_PER_GROUP(sb);
1481 ibase = gstart * EXT3_INODES_PER_GROUP(sb);
1482 for (gnum = gstart; gnum < EXT3_SB(sb)->s_groups_count;
1483 gnum++, ibase += EXT3_INODES_PER_GROUP(sb)) {
1484 struct buffer_head *bitmap_bh = NULL;
1485 struct ext3_group_desc * gdp;
1488 gdp = ext3_get_group_desc (sb, gnum, NULL);
1489 if (!gdp || le16_to_cpu(gdp->bg_free_inodes_count) ==
1490 EXT3_INODES_PER_GROUP(sb))
1492 bitmap_bh = read_inode_bitmap(sb, gnum);
1498 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
1499 ino = find_next_bit((unsigned long *)bitmap_bh->b_data,
1500 EXT3_INODES_PER_GROUP(sb), ino);
1502 ino = find_next_bit((unsigned long *)bitmap_bh->b_data,
1503 EXT3_INODES_PER_GROUP(sb), ino);
1504 #warning"FIXME-WANGDI need to port find_next_bit to 2.4"
1506 if (ino < EXT3_INODES_PER_GROUP(sb)) {
1507 ino_t inum = ino + gnum * EXT3_INODES_PER_GROUP(sb) + 1;
1509 if (inum < (*start)->i_ino)
1512 *start = iget(sb, inum);
1514 GOTO(exit, err = -ENOMEM);
1515 if (is_bad_inode(*start))
1516 GOTO(exit, err = -EIO);
1518 if ((err = (*repeat)(*start, priv)) != 0)
1522 if (++ino < EXT3_INODES_PER_GROUP(sb))
1532 static int fsfilt_ext3_iterate(struct super_block *sb,
1533 int (*repeat)(struct inode *inode, void *priv),
1534 struct inode **start, void *priv, int flag)
1537 case SNAP_ITERATE_ALL_INODE:
1538 return ext3_iterate_all (sb, repeat, start, priv);
1544 static int fsfilt_ext3_get_snap_info(struct inode *inode, void *key,
1545 __u32 keylen, void *val,
1551 if (!vallen || !val) {
1552 CERROR("val and val_size is 0!\n");
1555 if (keylen >= strlen(MAX_SNAPTABLE_COUNT)
1556 && strcmp(key, MAX_SNAPTABLE_COUNT) == 0) {
1557 /*FIXME should get it from the EA_size*/
1558 *((__u32 *)val) = EXT3_MAX_SNAPS;
1559 *vallen = sizeof(int);
1561 } else if (keylen >= strlen(SNAPTABLE_INFO)
1562 && strcmp(key, SNAPTABLE_INFO) == 0) {
1563 rc = ext3_xattr_get(inode, EXT3_SNAP_INDEX,
1564 EXT3_SNAPTABLE_EA, val, *vallen);
1566 } else if (keylen >= strlen(SNAP_GENERATION)
1567 && strcmp(key, SNAP_GENERATION) == 0) {
1569 rc = ext3_xattr_get(inode, EXT3_SNAP_INDEX,
1570 EXT3_SNAP_GENERATION, (char *)val, *vallen);
1571 if (rc == -ENODATA) {
1572 *((__u32 *)val) = 0;
1573 *vallen = sizeof(int);
1577 } else if (keylen >= strlen(SNAP_COUNT) &&
1578 strcmp(key, SNAP_COUNT) == 0) {
1579 rc = ext3_xattr_get(inode, EXT3_SNAP_INDEX,
1580 EXT3_SNAP_COUNT, val, *vallen);
1581 if (rc == -ENODATA) {
1582 *((__u32 *)val) = 0;
1583 *vallen = sizeof(int);
1592 static int fsfilt_ext3_set_snap_info(struct inode *inode, void *key,
1593 __u32 keylen, void *val,
1599 if (!vallen || !val) {
1600 CERROR("val and val_size is 0!\n");
1604 if (keylen >= strlen(SNAPTABLE_INFO)
1605 && strcmp(key, SNAPTABLE_INFO) == 0) {
1607 EXT3_JOURNAL_START(inode->i_sb, handle, EXT3_XATTR_TRANS_BLOCKS,
1611 rc = ext3_xattr_set_handle(handle, inode, EXT3_SNAP_INDEX,
1612 EXT3_SNAPTABLE_EA, val, *vallen, 0);
1613 journal_stop(handle);
1616 } else if (keylen >= strlen(SNAP_GENERATION)
1617 && strcmp(key, SNAP_GENERATION) == 0) {
1619 rc = ext3_set_generation(inode, *(int*)val);
1622 } else if (keylen >= strlen(SNAP_COUNT) &&
1623 (strcmp(key, SNAP_COUNT) == 0)) {
1625 EXT3_JOURNAL_START(inode->i_sb, handle,
1626 EXT3_XATTR_TRANS_BLOCKS, rc);
1629 rc = ext3_xattr_set_handle(handle, inode, EXT3_SNAP_INDEX,
1630 EXT3_SNAP_COUNT, val, *vallen, 0);
1631 journal_stop(handle);
1634 } else if (keylen >= strlen(SNAP_ROOT_INO) &&
1635 (strcmp(key, SNAP_ROOT_INO) == 0)) {
1644 static int fsfilt_ext3_dir_ent_size(char *name)
1647 return EXT3_DIR_REC_LEN(strlen(name));
1652 static int fsfilt_ext3_set_dir_ent(struct super_block *sb, char *name,
1653 char *buf, int buf_off, int nlen, size_t count)
1657 if (buf_off == 0 && nlen == 0) {
1658 struct ext3_dir_entry_2 *de = (struct ext3_dir_entry_2 *)buf;
1659 LASSERT(count == PAGE_CACHE_SIZE);
1660 de->rec_len = count;
1664 struct ext3_dir_entry_2 *de, *de1;
1665 de = (struct ext3_dir_entry_2 *)(buf + buf_off - nlen);
1666 de1 = (struct ext3_dir_entry_2 *)(buf + buf_off);
1669 LASSERT(nlen == EXT3_DIR_REC_LEN_DE(de));
1671 rlen = le16_to_cpu(de->rec_len);
1672 de->rec_len = cpu_to_le16(nlen);
1674 de1->rec_len = cpu_to_le16(rlen - nlen);
1675 de1->name_len = strlen(name);
1676 memcpy (de1->name, name, de->name_len);
1677 nlen = EXT3_DIR_REC_LEN_DE(de1);
1682 struct fsfilt_operations fsfilt_ext3_snap_ops = {
1683 .fs_type = "ext3_snap",
1684 .fs_owner = THIS_MODULE,
1685 .fs_create_indirect = fsfilt_ext3_create_indirect,
1686 .fs_get_indirect = fsfilt_ext3_get_indirect,
1687 .fs_set_indirect = fsfilt_ext3_set_indirect,
1688 .fs_snap_feature = fsfilt_ext3_snap_feature,
1689 .fs_is_redirector = fsfilt_ext3_is_redirector,
1690 .fs_is_indirect = fsfilt_ext3_is_indirect,
1691 .fs_get_indirect_ino = fsfilt_ext3_get_indirect_ino,
1692 .fs_destroy_indirect = fsfilt_ext3_destroy_indirect,
1693 .fs_restore_indirect = fsfilt_ext3_restore_indirect,
1694 .fs_iterate = fsfilt_ext3_iterate,
1695 .fs_copy_block = fsfilt_ext3_copy_block,
1696 .fs_set_snap_info = fsfilt_ext3_set_snap_info,
1697 .fs_get_snap_info = fsfilt_ext3_get_snap_info,
1698 .fs_dir_ent_size = fsfilt_ext3_dir_ent_size,
1699 .fs_set_dir_ent = fsfilt_ext3_set_dir_ent,
1703 static int __init fsfilt_ext3_snap_init(void)
1707 rc = fsfilt_register_ops(&fsfilt_ext3_snap_ops);
1712 static void __exit fsfilt_ext3_snap_exit(void)
1715 fsfilt_unregister_ops(&fsfilt_ext3_snap_ops);
1718 module_init(fsfilt_ext3_snap_init);
1719 module_exit(fsfilt_ext3_snap_exit);
1721 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
1722 MODULE_DESCRIPTION("Lustre ext3 Filesystem Helper v0.1");
1723 MODULE_LICENSE("GPL");