4 * Copyright (C) 2005-2009
5 * Neil Brown <neilb@suse.de>
6 * Released under the GPL, version 2
8 * generic inode handling
13 #include <linux/random.h>
14 #include <linux/delay.h>
15 #include <linux/slab.h>
17 /* Supporting an async 'iget' - as required by the cleaner -
18 * is slightly non-trivial.
19 * iget*_locked will normally wait for any inode with one
20 * of the flags I_FREEING I_CLEAR I_WILL_FREE I_NEW
21 * to either be unhashed or has the flag cleared.
22 * We cannot afford that wait in the cleaner as we could deadlock.
23 * So we use iget5_locked and provide a test function that fails
24 * if it finds the inode with any of those flags set.
25 * If it does see the inode like that it clear the inum
26 * that is passed in (by reference) so that it knows to continue
27 * failing (for consistency) and so that the 'set' function
28 * we provide can know to fail the 'set'.
29 * The result of this is that if iget finds an inode it would
30 * have to wait on, the inum is cleared and NULL is returned.
31 * An unfortunate side effect is that an inode will be allocated
32 * and then destroyed to no avail.
33 * This is avoided by calling ilookup5 first. This also allows
34 * us to only allocate/load the data block if there really seems
37 #define NO_INO (~(ino_t)0)
38 static int async_itest(struct inode *inode, void *data)
44 /* found and is freeing */
46 if (inode->i_ino != inum)
48 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE|I_NEW)) {
55 static int async_iset(struct inode *inode, void *data)
60 inode->i_ino = *inump;
65 lafs_iget(struct super_block *sb, ino_t inum, int async)
67 /* find, and load if needed, this inum */
68 struct inode *ino = NULL;
70 struct datablock *b = NULL;
71 struct inode *inodefile;
75 BUG_ON(inum == NO_INO);
81 /* We cannot afford to block on 'freeing_inode'
82 * So use iget5_locked and refuse to match such
84 * If the inode is 'freeing', inum gets set to NO_INO.
85 * ilookup5 is used first to avoid an unnecessary
86 * alloc/free if the inode is locked in some way.
91 ino = ilookup5(sb, inum, async_itest, &inum2);
98 /* For async we will always want the dblock loaded,
99 * and we need to load it first as we cannot afford
100 * to fail -EAGAIN once we have an I_NEW inode.
103 b = lafs_get_block(inodefile, inum, NULL,
104 GFP_NOFS, MKREF(iget));
106 return ERR_PTR(-ENOMEM);
109 err = lafs_read_block_async(b);
112 /* Have the block, so safe to iget */
114 ino = iget5_locked(sb, inum,
115 async_itest, async_iset,
125 if (test_and_set_bit(B_Async, &b->b.flags)) {
126 putdref(b, MKREF(iget));
129 getdref(b, MKREF(async));
133 ino = iget_locked(sb, inum);
136 putdref(b, MKREF(iget));
137 return ERR_PTR(-ENOMEM);
140 if (!(ino->i_state & I_NEW)) {
141 putdref(b, MKREF(iget));
145 return ERR_PTR(-ENOENT);
148 /* Need to load block 'inum' from an inode file...
151 b = lafs_get_block(inodefile, inum, NULL, GFP_KERNEL, MKREF(iget));
155 err = lafs_read_block(b);
160 oldino = rcu_my_inode(b);
162 /* The inode is new, but the block thinks it has an
163 * old inode, so we must be in the process of destroying
165 * So fail the lookup without even looking at the content
166 * of the block (Which might not be clear yet).
168 spin_lock(&oldino->i_data.private_lock);
169 if (!test_bit(I_Deleting, &LAFSI(oldino)->iflags)) {
171 LAFSI(oldino)->dblock = NULL;
172 LAFSI(oldino)->iblock = NULL;
174 spin_unlock(&oldino->i_data.private_lock);
182 err = lafs_import_inode(ino, b);
185 printk("lafs_import_inode failed %d\n", err);
188 unlock_new_inode(ino);
190 if (b && test_and_clear_bit(B_Async, &b->b.flags)) {
191 putdref(b, MKREF(async));
192 lafs_wake_thread(fs_from_sb(sb));
194 putdref(b, MKREF(iget));
198 unlock_new_inode(ino);
205 lafs_iget_fs(struct fs *fs, int fsnum, int inum, int async)
207 struct super_block *sb;
213 /* Need to locate or load the superblock for this
214 * subordinate filesystem
216 struct inode *filesys;
217 struct super_block *sb2;
219 filesys = lafs_iget(sb, fsnum, async);
222 if (LAFSI(filesys)->type != TypeInodeFile) {
224 return ERR_PTR(-ENOENT);
226 /* FIXME can get_subset_sb be async at all?? */
227 sb2 = lafs_get_subset_sb(filesys);
230 return ERR_PTR(PTR_ERR(sb2));
232 rv = lafs_iget(sb2, inum, async);
234 deactivate_locked_super(sb2);
236 up_write(&sb2->s_umount);
238 rv = lafs_iget(sb, inum, async);
239 atomic_inc(&sb->s_active);
245 lafs_import_inode(struct inode *ino, struct datablock *b)
247 struct la_inode *lai = map_dblock(b);
248 struct lafs_inode *li = LAFSI(ino);
251 if (lai->filetype == 0) {
258 ino->i_mode = S_IFREG;
259 ino->i_nlink = 1; /* For special file, set nlink so they
260 * never appear unlinked */
264 LAFS_BUG(ino->i_ino != b->b.fileaddr, &b->b);
265 li->cblocks = le32_to_cpu(lai->data_blocks);
266 li->pblocks = li->ablocks = 0;
267 li->vfs_inode.i_blocks = ((blkcnt_t)li->cblocks
268 << (ino->i_sb->s_blocksize_bits - 9));
269 li->ciblocks = le32_to_cpu(lai->index_blocks);
273 ino->i_generation = le16_to_cpu(lai->generation);
274 li->trunc_gen = lai->trunc_gen;
275 li->flags = lai->flags;
276 li->type = lai->filetype;
277 li->metadata_size = le16_to_cpu(lai->metadata_size);
278 li->depth = lai->depth;
280 dprintk("inode %lu type is %d\n", (unsigned long)ino->i_ino, li->type);
282 ino->i_data.a_ops = &lafs_file_aops;
288 struct fs_md *i = &li->md.fs;
289 struct fs_metadata *l = &lai->metadata[0].fs;
292 i->usagetable = le16_to_cpu(l->snapshot_usage_table);
293 decode_time(&ino->i_mtime, le64_to_cpu(l->update_time));
294 i->cblocks_used = le64_to_cpu(l->blocks_used);
295 i->pblocks_used = i->ablocks_used = 0;
296 i->blocks_allowed = le64_to_cpu(l->blocks_allowed);
297 i->blocks_unalloc = 0;
298 i->creation_age = le64_to_cpu(l->creation_age);
299 i->inodes_used = le32_to_cpu(l->inodes_used);
300 i->quota_inums[0] = le32_to_cpu(l->quota_inodes[0]);
301 i->quota_inums[1] = le32_to_cpu(l->quota_inodes[1]);
302 i->quota_inums[2] = le32_to_cpu(l->quota_inodes[2]);
303 i->quota_inodes[0] = i->quota_inodes[1]
304 = i->quota_inodes[2] = NULL;
305 nlen = li->metadata_size - offsetof(struct la_inode,
306 metadata[0].fs.name);
310 /* Need to unmap the dblock to kmalloc because
311 * the mapping makes us 'atomic'
313 unmap_dblock(b, lai);
314 i->name = kmalloc(nlen+1, GFP_KERNEL);
316 l = &lai->metadata[0].fs;
321 memcpy(i->name, l->name, nlen);
324 /* Make this look like a directory */
325 ino->i_mode = S_IFDIR;
329 ino->i_op = &lafs_subset_ino_operations;
330 ino->i_fop = &lafs_subset_file_operations;
336 struct inodemap_md *m = &li->md.inodemap;
337 struct inodemap_metadata *s = &lai->metadata[0].inodemap;
338 m->size = le32_to_cpu(s->size);
339 m->thisblock = NoBlock;
346 struct su_md *m = &li->md.segmentusage;
347 struct su_metadata *s = &lai->metadata[0].segmentusage;
348 m->table_size = le32_to_cpu(s->table_size);
354 struct quota_md *m = &li->md.quota;
355 struct quota_metadata *s = &lai->metadata[0].quota;
356 m->gracetime = le32_to_cpu(s->gracetime);
357 m->graceunits = le32_to_cpu(s->graceunits);
362 struct orphan_md *m = &li->md.orphan;
363 /* This will be set via lafs_count_orphans */
371 default: /* TypeBase or larger */
373 struct file_md *i = &li->md.file;
374 struct file_metadata *l = &lai->metadata[0].file;
375 struct dir_metadata *d = &lai->metadata[0].dir;
376 struct special_metadata *s = &lai->metadata[0].special;
378 if (li->type < TypeBase)
380 i->flags = le16_to_cpu(l->flags);
381 ino->i_mode = le16_to_cpu(l->mode);
382 ino->i_uid = le32_to_cpu(l->userid);
383 ino->i_gid = le32_to_cpu(l->groupid);
384 i->treeid = le32_to_cpu(l->treeid);
385 i->creationtime = le64_to_cpu(l->creationtime);
386 decode_time(&ino->i_mtime, le64_to_cpu(l->modifytime));
387 decode_time(&ino->i_ctime, le64_to_cpu(l->ctime));
388 decode_time(&i->i_accesstime, le64_to_cpu(l->accesstime));
389 ino->i_atime = i->i_accesstime; /* FIXME load from
391 ino->i_size = le64_to_cpu(l->size);
392 i->parent = le32_to_cpu(l->parent);
393 ino->i_nlink = le32_to_cpu(l->linkcount);
394 if (ino->i_nlink == 0 && list_empty(&b->orphans)) {
395 /* This block should already be on the orphan
396 * list, otherwise there is a filesystem
398 * Either the orphan file is wrong, or the
399 * linkcount is wrong.
400 * It is safest to assume the later - either
401 * way an FS check would be needed to fix it.
403 /* FIXME set a superblock flag requesting
404 * directory linkage checking
409 dprintk(" mode = 0%o uid %d size %lld\n",
410 ino->i_mode, ino->i_uid, ino->i_size);
413 ino->i_op = &lafs_file_ino_operations;
414 ino->i_fop = &lafs_file_file_operations;
415 ino->i_mode = (ino->i_mode & 07777) | S_IFREG;
418 i->seed = le32_to_cpu(d->hash_seed);
419 ino->i_op = &lafs_dir_ino_operations;
420 ino->i_fop = &lafs_dir_file_operations;
421 ino->i_mode = (ino->i_mode & 07777) | S_IFDIR;
424 dprintk("Hmm. %d %d %d\n",
431 ino->i_op = &lafs_link_ino_operations;
432 ino->i_mode = (ino->i_mode & 07777) | S_IFLNK;
435 /* the data had better be in the inode ... */
436 ino->i_rdev = MKDEV(le32_to_cpu(s->major),
437 le32_to_cpu(s->minor));
438 ino->i_op = &lafs_special_ino_operations;
439 init_special_inode(ino, ino->i_mode, ino->i_rdev);
446 ino->i_blkbits = ino->i_sb->s_blocksize_bits;
447 /* FIXME i_blocks and i_byte - used for quota?? */
450 /* Note: no refcount yet. Either will remove the reference to the
454 rcu_assign_pointer(b->my_inode, ino);
458 printk("inode %lu type is %d\n",
459 (unsigned long)ino->i_ino, li->type);
460 unmap_dblock(b, lai);
464 void lafs_inode_checkpin(struct inode *ino)
466 /* Make sure I_Pinned is set correctly.
467 * It should be set precisely if i_nlink is non-zero,
468 * and ->iblock is B_Pinned.
469 * When it is set, we own a reference to the inode.
471 * This needs to be called whenever we change
472 * i_nlink, and whenever we pin or unpin an InoIdx
475 if (ino->i_nlink == 0) {
476 /* I_Pinned should not be set */
477 if (test_and_clear_bit(I_Pinned, &LAFSI(ino)->iflags))
480 /* Need to check if iblock is Pinned. */
481 struct indexblock *ib = NULL;
482 if (LAFSI(ino)->iblock) {
483 spin_lock(&ino->i_data.private_lock);
484 ib = LAFSI(ino)->iblock;
485 if (ib && !test_bit(B_Pinned, &ib->b.flags))
487 spin_unlock(&ino->i_data.private_lock);
490 if (!test_and_set_bit(I_Pinned, &LAFSI(ino)->iflags))
493 if (test_and_clear_bit(I_Pinned, &LAFSI(ino)->iflags))
499 struct datablock *lafs_inode_get_dblock(struct inode *ino, REFARG)
501 struct datablock *db;
503 spin_lock(&ino->i_data.private_lock);
504 db = LAFSI(ino)->dblock;
506 if (db->b.inode == ino)
507 getdref_locked(db, REF);
509 spin_lock_nested(&db->b.inode->i_data.private_lock, 1);
510 getdref_locked(db, REF);
511 spin_unlock(&db->b.inode->i_data.private_lock);
514 spin_unlock(&ino->i_data.private_lock);
518 struct datablock *lafs_inode_dblock(struct inode *ino, int async, REFARG)
520 struct datablock *db;
523 db = lafs_inode_get_dblock(ino, REF);
526 db = lafs_get_block(ino_from_sb(ino->i_sb), ino->i_ino, NULL,
531 LAFSI(ino)->dblock = db;
532 rcu_assign_pointer(db->my_inode, ino);
534 err = lafs_read_block_async(db);
536 err = lafs_read_block(db);
544 void lafs_inode_init(struct datablock *b, int type, int mode, struct inode *dir)
546 /* A new block has been allocated in an inode file to hold an
547 * inode. We get to fill in initial values so that when
548 * 'iget' calls lafs_import_inode, the correct inode is
551 struct fs *fs = fs_from_inode(b->b.inode);
552 struct la_inode *lai = map_dblock(b);
555 lai->data_blocks = cpu_to_le32(0);
556 lai->index_blocks = cpu_to_le32(0);
557 get_random_bytes(&lai->generation, sizeof(lai->generation));
560 lai->filetype = type;
566 struct fs_metadata *l = &lai->metadata[0].fs;
567 size = sizeof(struct fs_metadata);
570 l->blocks_allowed = 0;
571 l->creation_age = fs->wc[0].cluster_seq;
573 l->quota_inodes[0] = 0;
574 l->quota_inodes[1] = 0;
575 l->quota_inodes[2] = 0;
576 l->snapshot_usage_table = 0;
578 /* name will be zero length and not used */
583 struct inodemap_metadata *l = &lai->metadata[0].inodemap;
585 size = sizeof(struct inodemap_metadata);
589 size = sizeof(struct su_metadata);
592 size = sizeof(struct quota_metadata);
602 struct file_metadata *l = &lai->metadata[0].file;
603 struct timespec now = CURRENT_TIME;
605 l->flags = cpu_to_le16(0);
606 l->userid = cpu_to_le32(current->cred->fsuid);
607 if (dir && (dir->i_mode & S_ISGID)) {
608 l->groupid = cpu_to_le32(dir->i_gid);
612 l->groupid = cpu_to_le32(current->cred->fsgid);
613 if (dir && LAFSI(dir)->md.file.treeid)
614 l->treeid = cpu_to_le32(LAFSI(dir)->md.file.treeid);
616 l->treeid = l->userid;
618 l->mode = cpu_to_le16(mode);
619 l->creationtime = encode_time(&now);
620 l->modifytime = l->creationtime;
621 l->ctime = l->creationtime;
622 l->accesstime = l->creationtime;
624 l->parent = dir ? cpu_to_le32(dir->i_ino) : 0;
627 if (type == TypeDir) {
628 struct dir_metadata *l = &lai->metadata[0].dir;
630 get_random_bytes(&seed,
632 seed = (seed & ~7) | 1;
633 l->hash_seed = cpu_to_le32(seed);
634 size = sizeof(struct dir_metadata);
635 } else if (type == TypeSpecial) {
636 struct special_metadata *s = &lai->metadata[0].special;
637 s->major = s->minor = 0;
638 size = sizeof(struct special_metadata);
640 size = sizeof(struct file_metadata);
643 size += sizeof(struct la_inode);
644 lai->metadata_size = cpu_to_le32(size);
645 memset(((char *)lai)+size, 0, fs->blocksize-size);
646 *(u16 *)(((char *)lai)+size) = cpu_to_le16(IBLK_EXTENT);
648 unmap_dblock(b, lai);
649 set_bit(B_Valid, &b->b.flags);
650 LAFS_BUG(!test_bit(B_Pinned, &b->b.flags), &b->b);
651 lafs_dirty_dblock(b);
654 void lafs_clear_inode(struct inode *ino)
656 struct lafs_inode *li = LAFSI(ino);
657 dprintk("CLEAR INODE %d\n", (int)ino->i_ino);
661 /* Now is a good time to break the linkage between
662 * inode and dblock - but not if the file is
665 if (!test_bit(I_Deleting, &LAFSI(ino)->iflags)) {
666 struct datablock *db;
667 spin_lock(&ino->i_data.private_lock);
668 db = LAFSI(ino)->dblock;
670 struct indexblock *ib = LAFSI(ino)->iblock;
671 LAFS_BUG(ib && atomic_read(&ib->b.refcnt), &db->b);
673 LAFSI(ino)->dblock = NULL;
674 LAFSI(ino)->iblock = NULL;
676 spin_unlock(&ino->i_data.private_lock);
679 /* FIXME release quota inodes if filesystem */
682 static int inode_map_free(struct fs *fs, struct super_block *sb, u32 inum);
684 void lafs_delete_inode(struct inode *ino)
686 struct fs *fs = fs_from_inode(ino);
689 if (ino->i_mode == 0) {
690 /* There never was an inode here,
696 dprintk("DELETE INODE %d\n", (int)ino->i_ino);
698 /* Normal truncation holds an igrab, so we cannot be
699 * deleted until any truncation finishes
701 BUG_ON(test_bit(I_Trunc, &LAFSI(ino)->iflags));
703 b = lafs_inode_dblock(ino, SYNC, MKREF(delete_inode));
705 i_size_write(ino, 0);
706 truncate_inode_pages(&ino->i_data, 0);
707 LAFSI(ino)->trunc_next = 0;
708 set_bit(I_Deleting, &LAFSI(ino)->iflags);
709 set_bit(I_Trunc, &LAFSI(ino)->iflags);
713 set_bit(B_Claimed, &b->b.flags);
714 lafs_add_orphan(fs, b);
715 dprintk("PUNCH hole for %d\n", (int)b->b.fileaddr);
716 putdref(b, MKREF(delete_inode));
718 inode_map_free(fs, ino->i_sb, ino->i_ino);
723 static int prune(void *data, u32 addr, u64 paddr, int len)
725 /* This whole index block is being pruned, just account
726 * for everything and it will be cleared afterwards
728 struct indexblock *ib = data;
729 struct inode *ino = ib->b.inode;
730 struct fs *fs = fs_from_inode(ino);
731 int ph = !!test_bit(B_Phase1, &ib->b.flags);
733 dprintk("PRUNE %d for %d at %lld\n", addr, len, (long long)paddr);
734 if (paddr == 0 || len == 0)
736 for (i = 0 ; i < len ; i++)
737 lafs_summary_update(fs, ino, paddr+i, 0, 0, ph, 0);
741 static int prune_some(void *data, u32 addr, u64 paddr, int len)
743 /* Part of this index block is being pruned. Copy
744 * what addresses we can into uninc_table so that
745 * it can be 'incorporated'
746 * We should probably share some code with
747 * lafs_allocated_block??
749 struct indexblock *ib = data;
750 struct inode *ino = ib->b.inode;
751 struct fs *fs = fs_from_inode(ino);
752 int ph = !!test_bit(B_Phase1, &ib->b.flags);
755 if (paddr == 0 || len == 0)
757 dprintk("PRUNE2 %d for %d at %lld\n", addr, len, (long long)paddr);
758 for (i = 0 ; i < len ; i++) {
759 /* FIXME should allow longer truncation ranges in uninc_table
760 * as they are easy to handle.
763 if (addr + i < LAFSI(ino)->trunc_next)
765 spin_lock(&ino->i_data.private_lock);
766 a = &ib->uninc_table.pending_addr
767 [ib->uninc_table.pending_cnt - 1];
768 if (ib->uninc_table.pending_cnt <
769 ARRAY_SIZE(ib->uninc_table.pending_addr)) {
771 a->fileaddr = addr + i;
774 LAFS_BUG(!test_bit(B_Pinned, &ib->b.flags), &ib->b);
775 ib->uninc_table.pending_cnt++;
777 spin_unlock(&ino->i_data.private_lock);
780 spin_unlock(&ino->i_data.private_lock);
781 lafs_summary_update(fs, ino, paddr+i, 0, 0, ph, 0);
786 int lafs_inode_handle_orphan(struct datablock *b)
788 /* Don't need rcu protection for my_inode run_orphan
791 struct indexblock *ib, *ib2;
792 struct inode *ino = b->my_inode;
793 struct fs *fs = fs_from_inode(ino);
794 u32 trunc_next, next_trunc;
798 if (!test_bit(I_Trunc, &LAFSI(ino)->iflags)) {
799 if (test_bit(I_Deleting, &LAFSI(ino)->iflags)) {
800 LAFS_BUG(ino->i_nlink, &b->b);
801 if (LAFSI(ino)->cblocks +
802 LAFSI(ino)->pblocks +
803 LAFSI(ino)->ablocks +
804 LAFSI(ino)->ciblocks +
805 LAFSI(ino)->piblocks)
806 printk("Deleting inode %lu: %ld+%ld+%ld %ld+%ld\n",
811 LAFSI(ino)->ciblocks,
812 LAFSI(ino)->piblocks);
813 BUG_ON(LAFSI(ino)->cblocks +
814 LAFSI(ino)->pblocks +
815 LAFSI(ino)->ablocks +
816 LAFSI(ino)->ciblocks +
817 LAFSI(ino)->piblocks);
818 if (lafs_erase_dblock_async(b))
819 lafs_orphan_release(fs, b);
820 } else if (ino->i_nlink || LAFSI(ino)->type == 0)
821 lafs_orphan_release(fs, b);
823 lafs_orphan_forget(fs, b);
827 ib = lafs_make_iblock(ino, ADOPT, SYNC, MKREF(inode_handle_orphan));
831 /* Here is the guts of 'truncate'. We find the next leaf index
832 * block and discard all the addresses there-in.
834 trunc_next = LAFSI(ino)->trunc_next;
836 if (trunc_next == 0xFFFFFFFF) {
837 /* truncate has finished in that all data blocks
838 * have been removed and all index block are either
839 * gone or pending incorporation at which point they will
841 * If we hit a phase change, we will need to postpone
842 * the rest of the cleaning until it completes.
843 * If there is a checkpoint happening, then all the work
844 * that we can do now, it will do for us. So just
847 struct indexblock *tmp;
848 struct indexblock *next;
851 if (!test_bit(B_Pinned, &ib->b.flags)) {
852 /* must be finished */
853 LAFS_BUG(test_bit(B_Dirty, &ib->b.flags), &ib->b);
854 clear_bit(I_Trunc, &LAFSI(ino)->iflags);
856 wake_up(&fs->trunc_wait);
860 if (fs->checkpointing) {
861 /* This cannot happen with current code,
862 * but leave it in case we ever have
863 * orphan handling parallel with checkpoints
865 err = -EBUSY; /* Try again after the checkpoint */
869 lastaddr = (i_size_read(ino) +
871 >> fs->blocksize_bits;
872 /* Find a Pinned descendent of ib which has no
873 * Pinned descendents and no PrimaryRef dependent
874 * (so take the last).
875 * Prefer blocks that are beyond EOF (again, take the last).
876 * If there are none, descend the last block that
877 * is not after EOF and look at its children.
880 spin_lock(&ib->b.inode->i_data.private_lock);
884 list_for_each_entry(tmp, &ib2->children, b.siblings) {
885 if (!test_bit(B_Index, &tmp->b.flags) ||
886 !test_bit(B_Pinned, &tmp->b.flags))
889 tmp->b.fileaddr > next->b.fileaddr)
893 if (ib2->b.fileaddr < lastaddr) {
894 /* Must be all done */
895 spin_unlock(&ib->b.inode->i_data.private_lock);
896 clear_bit(I_Trunc, &LAFSI(ino)->iflags);
898 wake_up(&fs->trunc_wait);
902 getiref(ib2, MKREF(inode_handle_orphan2));
903 spin_unlock(&ib->b.inode->i_data.private_lock);
905 /* ib2 is an index block beyond EOF with no
907 * Incorporating it should unpin it.
909 if (!list_empty(&ib2->children)) {
910 lafs_print_tree(&ib2->b, 3);
911 LAFS_BUG(1, &ib2->b);
914 if (!lafs_iolock_written_async(&ib2->b)) {
915 putiref(ib2, MKREF(inode_handle_orphan2));
919 while (ib2->uninc_table.pending_cnt || ib2->uninc)
920 lafs_incorporate(fs, ib2);
922 if (test_bit(B_Dirty, &ib2->b.flags) ||
923 test_bit(B_Realloc, &ib2->b.flags))
924 lafs_cluster_allocate(&ib2->b, 0);
926 lafs_iounlock_block(&ib2->b);
928 if (!list_empty(&ib2->b.siblings)) {
929 printk("looping on %s\n", strblk(&ib2->b));
934 putiref(ib2, MKREF(inode_handle_orphan2));
937 if (lafs_iolock_written_async(&ib->b)) {
939 lafs_incorporate(fs, ib);
940 lafs_iounlock_block(&ib->b);
945 putiref(ib, MKREF(inode_handle_orphan));
949 putiref(ib, MKREF(inode_handle_orphan));
951 ib = lafs_leaf_find(ino, trunc_next, ADOPT, &next_trunc,
952 ASYNC, MKREF(inode_handle_orphan3));
955 /* now hold an iolock on ib */
957 /* Ok, trunc_next seems to refer to a block that exists.
958 * We need to erase it..
960 * So we open up the index block ourselves, call
961 * lafs_summary_update with each block address, and then
965 if (LAFSI(ino)->depth == 0) {
966 /* Nothing to truncate */
967 clear_bit(I_Trunc, &LAFSI(ino)->iflags);
969 if (test_bit(B_Pinned, &ib->b.flags))
970 /* Need to move the dirtiness which keeps this
971 * pinned to the data block.
973 lafs_cluster_allocate(&ib->b, 0);
975 lafs_iounlock_block(&ib->b);
980 lafs_checkpoint_lock(fs);
981 err = lafs_reserve_block(&ib->b, ReleaseSpace);
985 if (!test_bit(B_Valid, &ib->b.flags) &&
986 test_bit(B_InoIdx, &ib->b.flags)) {
987 /* still invalid, just re-erase to remove
989 LAFSI(ino)->trunc_next = next_trunc;
990 lafs_cluster_allocate(&ib->b, 0);
995 lafs_pin_block(&ib->b);
997 /* It might be that this can happen, in which case
998 * we simply update trunc_next and loop. But I'd like
999 * to be sure before I implement that
1001 if (!test_bit(B_Valid, &ib->b.flags)) {
1002 printk("Not Valid: %s\n", strblk(&ib->b));
1003 printk("depth = %d\n", LAFSI(ino)->depth);
1004 if (test_bit(B_InoIdx, &ib->b.flags))
1005 printk("DB: %s\n", strblk(&LAFSI(ib->b.inode)->dblock->b));
1006 LAFSI(ino)->trunc_next = next_trunc;
1007 //BUG_ON(!test_bit(B_Valid, &ib->b.flags));
1012 if (ib->b.fileaddr < trunc_next &&
1013 lafs_leaf_next(ib, 0) < trunc_next) {
1014 /* We only want to truncate part of this index block.
1015 * So we copy addresses into uninc_table and then
1016 * call lafs_incorporate.
1017 * This might cause the index tree to grow, so we
1018 * cannot trust next_trunc
1020 if (ib->uninc_table.pending_cnt == 0 &&
1021 ib->uninc == NULL) {
1022 lafs_dirty_iblock(ib, 0);
1023 /* FIXME this just removes 8 blocks at a time,
1024 * which is not enough
1026 lafs_walk_leaf_index(ib, prune_some, ib);
1028 if (test_bit(B_Dirty, &ib->b.flags))
1029 lafs_incorporate(fs, ib);
1033 LAFSI(ino)->trunc_next = next_trunc;
1035 while (ib->uninc_table.pending_cnt || ib->uninc) {
1036 /* There should be no Realloc data blocks here
1037 * but index blocks might be realloc still.
1039 LAFS_BUG(!test_bit(B_Dirty, &ib->b.flags) &&
1040 !test_bit(B_Realloc, &ib->b.flags), &ib->b);
1041 lafs_incorporate(fs, ib);
1043 if (test_bit(B_InoIdx, &ib->b.flags) ||
1044 !test_bit(B_PhysValid, &ib->b.flags) ||
1045 ib->b.physaddr != 0) {
1046 lafs_walk_leaf_index(ib, prune, ib);
1047 lafs_clear_index(ib);
1048 lafs_dirty_iblock(ib, 0);
1050 if (test_bit(B_Dirty, &ib->b.flags))
1051 lafs_incorporate(fs, ib);
1052 if (!list_empty(&ib->children))
1053 lafs_print_tree(&ib->b, 2);
1054 LAFS_BUG(!list_empty(&ib->children), &ib->b);
1057 lafs_iounlock_block(&ib->b);
1059 lafs_checkpoint_unlock(fs);
1061 putiref(ib, MKREF(inode_handle_orphan3));
1065 void lafs_dirty_inode(struct inode *ino)
1067 /* this is called in one of three cases:
1068 * 1/ by lafs internally when dblock or iblock is pinned and
1069 * ready to be dirtied
1070 * 2/ by writeout before requesting a write - to update mtime
1071 * 3/ by read to update atime
1073 * As we don't know which, there is not much we can do.
1074 * We mustn't update the data block as it could be in
1075 * writeout and we cannot always wait safely.
1076 * So require that anyone who really cares, dirties the datablock
1077 * or a child themselves.
1078 * When cluster_allocate eventually gets called, it will update
1079 * the datablock from the inode.
1080 * If an update has to wait for the next phase, lock_dblock
1081 * (e.g. in setattr) will do that.
1083 * We also use this opportunity to update the filesystem modify time.
1085 struct timespec now;
1086 struct inode *filesys;
1087 set_bit(I_Dirty, &LAFSI(ino)->iflags);
1088 ino->i_sb->s_dirt = 1;
1090 now = current_fs_time(ino->i_sb);
1091 filesys = ino_from_sb(ino->i_sb);
1092 if (!timespec_equal(&filesys->i_mtime, &now)) {
1093 filesys->i_mtime = now;
1094 set_bit(I_Dirty, &LAFSI(filesys)->iflags);
1098 int lafs_sync_inode(struct inode *ino, int wait)
1100 /* fsync has been called on this file so we need
1101 * to sync any inode updates to the next cluster.
1103 * If we cannot create an update record,
1104 * we wait for a phase change, which writes everything
1107 struct datablock *b;
1108 struct fs *fs = fs_from_inode(ino);
1109 struct update_handle uh;
1113 if (LAFSI(ino)->update_cluster > 1)
1114 lafs_cluster_wait(fs, LAFSI(ino)->update_cluster);
1115 if (LAFSI(ino)->update_cluster == 1) {
1116 lafs_checkpoint_lock(fs);
1117 lafs_checkpoint_unlock_wait(fs);
1122 LAFSI(ino)->update_cluster = 0;
1123 if (!test_bit(I_Dirty, &LAFSI(ino)->iflags))
1125 b = lafs_inode_dblock(ino, SYNC, MKREF(write_inode));
1129 lafs_iolock_written(&b->b);
1130 lafs_inode_fillblock(ino);
1131 lafs_iounlock_block(&b->b);
1133 err = lafs_cluster_update_prepare(&uh, fs, LAFS_INODE_LOG_SIZE);
1135 lafs_cluster_update_abort(&uh);
1137 lafs_checkpoint_lock(fs);
1138 if (lafs_cluster_update_pin(&uh) == 0) {
1139 if (test_and_clear_bit(B_Dirty, &b->b.flags))
1140 lafs_space_return(fs, 1);
1141 LAFSI(ino)->update_cluster =
1142 lafs_cluster_update_commit
1143 (&uh, b, LAFS_INODE_LOG_START,
1144 LAFS_INODE_LOG_SIZE);
1146 lafs_cluster_update_abort(&uh);
1147 lafs_checkpoint_unlock(fs);
1149 if (test_bit(B_Dirty, &b->b.flags)) {
1150 /* FIXME need to write out the data block...
1151 * Is that just lafs_cluster_allocate ?
1155 if (LAFSI(ino)->update_cluster == 0) {
1156 lafs_checkpoint_lock(fs);
1157 if (test_bit(B_Dirty, &b->b.flags))
1158 LAFSI(ino)->update_cluster = 1;
1159 lafs_checkpoint_start(fs);
1160 lafs_checkpoint_unlock(fs);
1162 putdref(b, MKREF(write_inode));
1163 return 0; /* FIXME should I return some error message??? */
1166 void lafs_inode_fillblock(struct inode *ino)
1168 /* copy data from ino into the related data block */
1170 struct lafs_inode *li = LAFSI(ino);
1171 struct datablock *db = li->dblock;
1172 struct la_inode *lai;
1174 clear_bit(I_Dirty, &LAFSI(ino)->iflags);
1176 lai = map_dblock(db);
1177 lai->data_blocks = cpu_to_le32(li->cblocks);
1178 lai->index_blocks = cpu_to_le32(li->ciblocks);
1179 lai->generation = cpu_to_le16(ino->i_generation);
1180 lai->trunc_gen = li->trunc_gen;
1181 lai->flags = li->flags;
1182 lai->filetype = li->type;
1183 if (lai->metadata_size != cpu_to_le16(li->metadata_size)) {
1184 /* Changing metadata size is wierd.
1185 * We will need to handle this somehow for xattrs
1186 * For now we just want to cope with
1187 * Dir -> InodeFile changes, and that guarantees us
1188 * there is no index info - so just clear the index
1191 u16 *s = (u16*)(((char*)lai) + li->metadata_size);
1192 BUG_ON(li->type != TypeInodeFile);
1193 lai->metadata_size = cpu_to_le16(li->metadata_size);
1194 memset(s, 0, ino->i_sb->s_blocksize - li->metadata_size);
1195 *s = cpu_to_le16(IBLK_INDIRECT);
1197 lai->depth = li->depth;
1202 struct fs_md *i = &li->md.fs;
1203 struct fs_metadata *l = &lai->metadata[0].fs;
1206 l->snapshot_usage_table = cpu_to_le16(i->usagetable);
1207 l->update_time = cpu_to_le64(encode_time(&ino->i_mtime));
1208 l->blocks_used = cpu_to_le64(i->cblocks_used);
1209 l->blocks_allowed = cpu_to_le64(i->blocks_allowed);
1210 l->creation_age = cpu_to_le64(i->creation_age);
1211 l->inodes_used = cpu_to_le32(i->inodes_used);
1212 l->quota_inodes[0] = cpu_to_le32(i->quota_inums[0]);
1213 l->quota_inodes[1] = cpu_to_le32(i->quota_inums[1]);
1214 l->quota_inodes[2] = cpu_to_le32(i->quota_inums[2]);
1215 nlen = lai->metadata_size - offsetof(struct la_inode,
1216 metadata[0].fs.name);
1217 memset(l->name, 0, nlen);
1218 if (i->name == NULL)
1220 else if (strlen(i->name) < nlen)
1221 nlen = strlen(i->name);
1222 memcpy(l->name, i->name, nlen);
1228 struct inodemap_md *m = &li->md.inodemap;
1229 struct inodemap_metadata *s = &lai->metadata[0].inodemap;
1230 s->size = cpu_to_le32(m->size);
1234 case TypeSegmentMap:
1236 struct su_md *m = &li->md.segmentusage;
1237 struct su_metadata *s = &lai->metadata[0].segmentusage;
1238 s->table_size = cpu_to_le32(m->table_size);
1244 struct quota_md *m = &li->md.quota;
1245 struct quota_metadata *s = &lai->metadata[0].quota;
1246 s->gracetime = cpu_to_le32(m->gracetime);
1247 s->graceunits = cpu_to_le32(m->graceunits);
1250 case TypeOrphanList:
1251 case TypeAccessTime:
1254 default: /* TypeBase or larger */
1256 struct file_md *i = &li->md.file;
1257 struct file_metadata *l = &lai->metadata[0].file;
1258 struct dir_metadata *d = &lai->metadata[0].dir;
1259 struct special_metadata *s = &lai->metadata[0].special;
1261 if (li->type < TypeBase)
1263 l->flags = cpu_to_le16(i->flags);
1264 l->mode = cpu_to_le16(ino->i_mode);
1265 l->userid = cpu_to_le32(ino->i_uid);
1266 l->groupid = cpu_to_le32(ino->i_gid);
1267 l->treeid = cpu_to_le32(i->treeid);
1268 l->creationtime = cpu_to_le64(i->creationtime);
1269 l->modifytime = cpu_to_le64(encode_time(&ino->i_mtime));
1270 l->ctime = cpu_to_le64(encode_time(&ino->i_ctime));
1271 l->accesstime = cpu_to_le64(encode_time(&ino->i_atime));
1272 /* FIXME write 0 to accesstime file */
1273 l->size = cpu_to_le64(ino->i_size);
1274 l->parent = cpu_to_le32(i->parent);
1275 l->linkcount = cpu_to_le32(ino->i_nlink);
1281 d->hash_seed = cpu_to_le32(i->seed);
1286 s->major = cpu_to_le32(MAJOR(ino->i_rdev));
1287 s->minor = cpu_to_le32(MINOR(ino->i_rdev));
1292 unmap_dblock(db, lai);
1295 /*-----------------------------------------------------------------------
1296 * Inode allocate map handling.
1297 * Inode 1 of each fileset is a bitmap of free inode numbers.
1298 * Whenever the file is extended in size, new bits are set to one. They
1299 * are then cleared when the inode is allocated. When a block becomes
1300 * full of zeros, we don't need to store it any more.
1302 * We don't clear the bit until we are committed to creating an inode
1303 * This means we cannot clear it straight away, so two different threads
1304 * might see the same inode number as being available. We have two
1305 * approaches to guard against this.
1306 * Firstly we have a 'current' pointer into the inodemap file and
1307 * increase that past the inode we return. This discourages multiple
1308 * hits but as the pointer would need to be rewound occasionally it
1309 * isn't a guarantee. The guarantee against multiple allocations is done
1310 * via a flag in the block representing an inode. This is set
1311 * while an inode is being allocated.
1314 /* inode number allocation has the prealloc/pin/commit/abort structure
1315 * so it can be committed effectively
1319 choose_free_inum(struct fs *fs, struct super_block *sb, u32 *inump,
1320 struct datablock **bp, int *restarted)
1322 struct inode *im = lafs_iget(sb, 1, SYNC);
1324 struct datablock *b;
1330 struct inode *i = (*bp)->b.inode;
1331 putdref(*bp, MKREF(cfi_map));
1336 mutex_lock_nested(&im->i_mutex, I_MUTEX_QUOTA);
1338 bnum = LAFSI(im)->md.inodemap.thisblock;
1340 if (bnum == NoBlock ||
1341 LAFSI(im)->md.inodemap.nextbit >= (fs->blocksize<<3)) {
1342 if (bnum == NoBlock)
1343 bnum = LAFSI(im)->md.inodemap.size;
1345 if (bnum+1 < LAFSI(im)->md.inodemap.size)
1347 else if (!*restarted) {
1351 /* Need to add a new block to the file */
1352 bnum = LAFSI(im)->md.inodemap.size;
1353 b = lafs_get_block(im, bnum, NULL, GFP_KERNEL,
1358 lafs_iolock_written(&b->b);
1359 set_bit(B_PinPending, &b->b.flags);
1360 lafs_iounlock_block(&b->b);
1362 lafs_checkpoint_lock(fs);
1363 err = lafs_pin_dblock(b, NewSpace);
1364 if (err == -EAGAIN) {
1365 lafs_checkpoint_unlock_wait(fs);
1371 buf = map_dblock(b);
1372 /* Set block to "all are free" */
1373 memset(buf, 0xff, fs->blocksize);
1374 unmap_dblock(b, buf);
1375 set_bit(B_Valid, &b->b.flags);
1376 LAFSI(im)->md.inodemap.size = bnum+1;
1377 lafs_dirty_inode(im);
1378 lafs_dirty_dblock(b);
1379 lafs_checkpoint_unlock(fs);
1380 putdref(b, MKREF(cfi_map));
1383 err = lafs_find_next(im, &bnum);
1389 LAFSI(im)->md.inodemap.nextbit = 0;
1390 LAFSI(im)->md.inodemap.thisblock = bnum;
1393 b = lafs_get_block(im, bnum, NULL, GFP_KERNEL, MKREF(cfi_map));
1397 err = lafs_find_block(b, NOADOPT);
1400 if (b->b.physaddr == 0 && !test_bit(B_Valid, &b->b.flags)) {
1401 LAFSI(im)->md.inodemap.nextbit =
1402 (fs->blocksize<<3) + 1;
1403 putdref(b,MKREF(cfi_map));
1406 err = lafs_read_block(b);
1410 bit = LAFSI(im)->md.inodemap.nextbit;
1411 LAFSI(im)->md.inodemap.thisblock = bnum;
1412 buf = map_dblock(b);
1413 while (bnum == 0 && bit < 16) {
1414 /* Never return an inum below 16 - they are special */
1415 if (!generic_test_le_bit(bit, (unsigned long *)buf))
1416 generic___clear_le_bit(bit, (unsigned long *)buf);
1420 bit = generic_find_next_le_bit((unsigned long *)buf,
1421 fs->blocksize<<3, bit);
1422 unmap_dblock(b, buf);
1423 LAFSI(im)->md.inodemap.nextbit = bit+1;
1424 if (bit >= fs->blocksize<<3) {
1425 putdref(b,MKREF(cfi_map));
1428 mutex_unlock(&im->i_mutex);
1430 *inump = bit + (bnum << (im->i_blkbits + 3));
1434 lafs_checkpoint_unlock(fs);
1436 putdref(b, MKREF(cfi_map));
1438 mutex_unlock(&im->i_mutex);
1443 struct inode_map_new_info {
1444 struct datablock *ib, *mb;
1448 inode_map_new_prepare(struct fs *fs, int inum, struct super_block *sb,
1449 struct inode_map_new_info *imni)
1454 struct datablock *b;
1456 imni->ib = imni->mb = NULL;
1459 /* choose a possibly-free inode number */
1460 err = choose_free_inum(fs, sb, &choice,
1461 &imni->mb, &restarted);
1465 b = lafs_get_block(ino_from_sb(sb), choice, NULL, GFP_KERNEL,
1470 if (test_and_set_bit(B_Claimed, &b->b.flags)) {
1471 putdref(b, MKREF(cfi_ino));
1477 lafs_iolock_written(&imni->mb->b);
1478 set_bit(B_PinPending, &imni->mb->b.flags);
1479 lafs_iounlock_block(&imni->mb->b);
1481 set_bit(B_PinPending, &b->b.flags);
1488 inode_map_new_pin(struct inode_map_new_info *imni)
1492 err = lafs_pin_dblock(imni->mb, NewSpace);
1493 err = err ?: lafs_pin_dblock(imni->ib, NewSpace);
1498 inode_map_new_commit(struct inode_map_new_info *imni)
1503 int blksize = imni->ib->b.inode->i_sb->s_blocksize;
1504 int bit = imni->ib->b.fileaddr & (blksize*8 - 1);
1506 struct inode *ino = imni->mb->b.inode;
1508 mutex_lock_nested(&ino->i_mutex, I_MUTEX_QUOTA);
1509 buf = map_dblock(imni->mb);
1510 generic___clear_le_bit(bit, buf);
1511 if (buf[blksize/sizeof(*buf)-1] == 0 &&
1512 generic_find_next_le_bit(buf, blksize*8, 0) == blksize*8)
1513 /* block is empty, punch a hole */
1516 unmap_dblock(imni->mb, buf);
1518 lafs_erase_dblock(imni->mb);
1520 lafs_dirty_dblock(imni->mb);
1522 putdref(imni->mb, MKREF(cfi_map));
1523 mutex_unlock(&ino->i_mutex);
1526 putdref(imni->ib, MKREF(cfi_ino));
1530 inode_map_new_abort(struct inode_map_new_info *imni)
1533 clear_bit(B_Claimed, &imni->ib->b.flags);
1534 clear_bit(B_PinPending, &imni->ib->b.flags);
1535 lafs_orphan_release(fs_from_inode(imni->ib->b.inode),
1538 putdref(imni->ib, MKREF(cfi_ino));
1540 struct inode *ino = imni->mb->b.inode;
1541 putdref(imni->mb, MKREF(cfi_map));
1547 lafs_new_inode(struct fs *fs, struct super_block *sb, struct inode *dir,
1548 int type, int inum, int mode, struct datablock **inodbp)
1550 /* allocate and instantiate a new inode. If inum is non-zero,
1551 * choose any number, otherwise we are creating a special inode
1552 * and have to use the given number.
1553 * This creation is committed independently of any name that might
1554 * subsequently be given to the inode. So we register it as an
1555 * orphan so that it will be cleaned up if the name isn't
1556 * successfully created
1560 struct datablock *b;
1561 struct inode_map_new_info imni;
1562 struct update_handle ui;
1565 err = inode_map_new_prepare(fs, inum, sb, &imni);
1566 err = lafs_cluster_update_prepare(&ui, fs, sizeof(struct la_inode))
1569 err = lafs_make_orphan(fs, imni.ib);
1573 lafs_checkpoint_lock(fs);
1575 err = inode_map_new_pin(&imni);
1577 if (err == -EAGAIN) {
1578 lafs_checkpoint_unlock_wait(fs);
1584 b = getdref(imni.ib, MKREF(inode_new));
1586 lafs_iolock_block(&b->b); /* make sure we don't race with the cleaner
1587 * and zero this inode while trying to load it
1589 lafs_inode_init(b, type, mode, dir);
1590 lafs_iounlock_block(&b->b);
1592 inode_map_new_commit(&imni);
1593 ino = lafs_iget(sb, b->b.fileaddr, SYNC);
1595 lafs_cluster_update_abort(&ui);
1598 lafs_cluster_update_commit(&ui, b, 0,
1599 LAFSI(ino)->metadata_size);
1600 LAFS_BUG(LAFSI(ino)->dblock != b, &b->b);
1601 LAFS_BUG(b->my_inode != ino, &b->b);
1602 lafs_checkpoint_unlock(fs);
1607 putdref(b, MKREF(inode_new));
1611 lafs_checkpoint_unlock(fs);
1614 inode_map_new_abort(&imni);
1615 lafs_cluster_update_abort(&ui);
1616 dprintk("After abort %d: %s\n", err, strblk(&imni.ib->b));
1617 return ERR_PTR(err);
1620 static int inode_map_free(struct fs *fs, struct super_block *sb, u32 inum)
1622 struct inode *im = lafs_iget(sb, 1, SYNC);
1625 struct datablock *b;
1629 mutex_lock_nested(&im->i_mutex, I_MUTEX_QUOTA);
1631 bnum = inum >> (3 + sb->s_blocksize_bits);
1632 bit = inum - (bnum << (3 + sb->s_blocksize_bits));
1633 b = lafs_get_block(im, bnum, NULL, GFP_KERNEL, MKREF(inode_map_free));
1635 mutex_unlock(&im->i_mutex);
1639 err = lafs_read_block(b);
1641 putdref(b, MKREF(inode_map_free));
1642 mutex_unlock(&im->i_mutex);
1646 lafs_iolock_written(&b->b);
1647 set_bit(B_PinPending, &b->b.flags);
1648 lafs_iounlock_block(&b->b);
1650 lafs_checkpoint_lock(fs);
1651 err = lafs_pin_dblock(b, ReleaseSpace);
1652 if (err == -EAGAIN) {
1653 lafs_checkpoint_unlock_wait(fs);
1657 buf = map_dblock(b);
1658 generic___set_le_bit(bit, buf);
1659 unmap_dblock(b, buf);
1660 lafs_dirty_dblock(b);
1661 putdref(b, MKREF(inode_map_free));
1662 lafs_checkpoint_unlock(fs);
1663 mutex_unlock(&im->i_mutex);
1668 int lafs_setattr(struct dentry *dentry, struct iattr *attr)
1671 struct inode *ino = dentry->d_inode;
1672 struct fs *fs = fs_from_inode(ino);
1673 struct datablock *db;
1675 err = inode_change_ok(ino, attr);
1676 db = lafs_inode_dblock(ino, SYNC, MKREF(setattr));
1682 /* We don't need iolock_written here as we don't
1683 * actually change the inode block yet
1685 lafs_iolock_block(&db->b);
1686 set_bit(B_PinPending, &db->b.flags);
1687 lafs_iounlock_block(&db->b);
1689 /* FIXME quota stuff */
1692 lafs_checkpoint_lock(fs);
1693 err = lafs_pin_dblock(db, ReleaseSpace);
1694 if (err == -EAGAIN) {
1695 lafs_checkpoint_unlock_wait(fs);
1698 /* inode_setattr calls lafs_dirty_inode, which sets
1699 * I_Dirty so the dblock will get updated.
1701 err = err ?: inode_setattr(ino, attr);
1703 lafs_dirty_dblock(db);
1704 clear_bit(B_PinPending, &db->b.flags);
1705 putdref(db, MKREF(setattr));
1706 lafs_checkpoint_unlock(fs);
1711 void lafs_truncate(struct inode *ino)
1713 /* Want to truncate this file.
1714 * i_size has already been changed, and the address space
1715 * has been cleaned up.
1716 * So just start the background truncate
1718 struct fs *fs = fs_from_inode(ino);
1719 struct datablock *db = lafs_inode_dblock(ino, SYNC, MKREF(trunc));
1726 trunc_block = ((i_size_read(ino) + fs->blocksize - 1)
1727 >> fs->blocksize_bits);
1728 /* We hold i_mutex, so regular orphan processing cannot
1729 * contine - we have to push it forward ourselves.
1731 while (test_bit(I_Trunc, &LAFSI(ino)->iflags) &&
1732 LAFSI(ino)->trunc_next < trunc_block) {
1733 prepare_to_wait(&fs->async_complete, &wq,
1734 TASK_UNINTERRUPTIBLE);
1735 lafs_inode_handle_orphan(db);
1736 if (test_bit(B_Orphan, &db->b.flags))
1739 finish_wait(&fs->async_complete, &wq);
1741 /* There is nothing we can do about errors here. The
1742 * most likely are ENOMEM which itself is very unlikely.
1743 * If this doesn't get registered as an orphan .... maybe
1744 * it will have to wait until something else truncates it.
1746 lafs_make_orphan(fs, db);
1748 if (!test_and_set_bit(I_Trunc, &LAFSI(ino)->iflags))
1750 if (trunc_block == 0)
1751 LAFSI(ino)->trunc_gen++;
1752 LAFSI(ino)->trunc_next = trunc_block;
1753 putdref(db, MKREF(trunc));
1756 const struct inode_operations lafs_special_ino_operations = {
1757 .setattr = lafs_setattr,
1758 .truncate = lafs_truncate,