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
217 struct inode *filesys;
219 printk("get filesys %d\n", fsnum);
221 filesys = lafs_iget(sb, fsnum, async);
224 if (filesys->i_sb == sb) {
225 /* No mounted fs here.
226 * Need to mount one FIXME
229 return ERR_PTR(-ENOENT);
231 printk("get inode %d\n", inum);
232 rv = lafs_iget(filesys->i_sb, inum, async);
235 rv = lafs_iget(sb, inum, async);
236 atomic_inc(&sb->s_active);
242 lafs_import_inode(struct inode *ino, struct datablock *b)
244 struct la_inode *lai = map_dblock(b);
245 struct lafs_inode *li = LAFSI(ino);
248 if (lai->filetype == 0) {
255 ino->i_mode = S_IFREG;
256 ino->i_nlink = 1; /* For special file, set nlink so they
257 * never appear unlinked */
261 LAFS_BUG(ino->i_ino != b->b.fileaddr, &b->b);
262 li->cblocks = le32_to_cpu(lai->data_blocks);
263 li->pblocks = li->ablocks = 0;
264 li->vfs_inode.i_blocks = ((blkcnt_t)li->cblocks
265 << (ino->i_sb->s_blocksize_bits - 9));
266 li->ciblocks = le32_to_cpu(lai->index_blocks);
270 ino->i_generation = le16_to_cpu(lai->generation);
271 li->trunc_gen = lai->trunc_gen;
272 li->flags = lai->flags;
273 li->type = lai->filetype;
274 li->metadata_size = le16_to_cpu(lai->metadata_size);
275 li->depth = lai->depth;
277 dprintk("inode %lu type is %d\n", (unsigned long)ino->i_ino, li->type);
279 ino->i_data.a_ops = &lafs_file_aops;
285 struct fs_md *i = &li->md.fs;
286 struct fs_metadata *l = &lai->metadata[0].fs;
289 i->usagetable = le16_to_cpu(l->snapshot_usage_table);
290 decode_time(&ino->i_mtime, le64_to_cpu(l->update_time));
291 i->cblocks_used = le64_to_cpu(l->blocks_used);
292 i->pblocks_used = i->ablocks_used = 0;
293 i->blocks_allowed = le64_to_cpu(l->blocks_allowed);
294 i->blocks_unalloc = 0;
295 i->creation_age = le64_to_cpu(l->creation_age);
296 i->inodes_used = le32_to_cpu(l->inodes_used);
297 i->quota_inums[0] = le32_to_cpu(l->quota_inodes[0]);
298 i->quota_inums[1] = le32_to_cpu(l->quota_inodes[1]);
299 i->quota_inums[2] = le32_to_cpu(l->quota_inodes[2]);
300 i->quota_inodes[0] = i->quota_inodes[1]
301 = i->quota_inodes[2] = NULL;
302 nlen = li->metadata_size - offsetof(struct la_inode,
303 metadata[0].fs.name);
307 /* Need to unmap the dblock to kmalloc because
308 * the mapping makes us 'atomic'
310 unmap_dblock(b, lai);
311 i->name = kmalloc(nlen+1, GFP_KERNEL);
313 l = &lai->metadata[0].fs;
318 memcpy(i->name, l->name, nlen);
321 /* Make this look like a directory */
322 ino->i_mode = S_IFDIR;
326 ino->i_op = &lafs_subset_ino_operations;
327 ino->i_fop = &lafs_subset_file_operations;
333 struct inodemap_md *m = &li->md.inodemap;
334 struct inodemap_metadata *s = &lai->metadata[0].inodemap;
335 m->size = le32_to_cpu(s->size);
336 m->thisblock = NoBlock;
343 struct su_md *m = &li->md.segmentusage;
344 struct su_metadata *s = &lai->metadata[0].segmentusage;
345 m->table_size = le32_to_cpu(s->table_size);
351 struct quota_md *m = &li->md.quota;
352 struct quota_metadata *s = &lai->metadata[0].quota;
353 m->gracetime = le32_to_cpu(s->gracetime);
354 m->graceunits = le32_to_cpu(s->graceunits);
359 struct orphan_md *m = &li->md.orphan;
360 /* This will be set via lafs_count_orphans */
368 default: /* TypeBase or larger */
370 struct file_md *i = &li->md.file;
371 struct file_metadata *l = &lai->metadata[0].file;
372 struct dir_metadata *d = &lai->metadata[0].dir;
373 struct special_metadata *s = &lai->metadata[0].special;
375 if (li->type < TypeBase)
377 i->flags = le16_to_cpu(l->flags);
378 ino->i_mode = le16_to_cpu(l->mode);
379 ino->i_uid = le32_to_cpu(l->userid);
380 ino->i_gid = le32_to_cpu(l->groupid);
381 i->treeid = le32_to_cpu(l->treeid);
382 i->creationtime = le64_to_cpu(l->creationtime);
383 decode_time(&ino->i_mtime, le64_to_cpu(l->modifytime));
384 decode_time(&ino->i_ctime, le64_to_cpu(l->ctime));
385 decode_time(&i->i_accesstime, le64_to_cpu(l->accesstime));
386 ino->i_atime = i->i_accesstime; /* FIXME load from
388 ino->i_size = le64_to_cpu(l->size);
389 i->parent = le32_to_cpu(l->parent);
390 ino->i_nlink = le32_to_cpu(l->linkcount);
391 if (ino->i_nlink == 0 && list_empty(&b->orphans)) {
392 /* This block should already be on the orphan
393 * list, otherwise there is a filesystem
395 * Either the orphan file is wrong, or the
396 * linkcount is wrong.
397 * It is safest to assume the later - either
398 * way an FS check would be needed to fix it.
400 /* FIXME set a superblock flag requesting
401 * directory linkage checking
406 dprintk(" mode = 0%o uid %d size %lld\n",
407 ino->i_mode, ino->i_uid, ino->i_size);
410 ino->i_op = &lafs_file_ino_operations;
411 ino->i_fop = &lafs_file_file_operations;
412 ino->i_mode = (ino->i_mode & 07777) | S_IFREG;
415 i->seed = le32_to_cpu(d->hash_seed);
416 ino->i_op = &lafs_dir_ino_operations;
417 ino->i_fop = &lafs_dir_file_operations;
418 ino->i_mode = (ino->i_mode & 07777) | S_IFDIR;
421 dprintk("Hmm. %d %d %d\n",
428 ino->i_op = &lafs_link_ino_operations;
429 ino->i_mode = (ino->i_mode & 07777) | S_IFLNK;
432 /* the data had better be in the inode ... */
433 ino->i_rdev = MKDEV(le32_to_cpu(s->major),
434 le32_to_cpu(s->minor));
435 ino->i_op = &lafs_special_ino_operations;
436 init_special_inode(ino, ino->i_mode, ino->i_rdev);
443 ino->i_blkbits = ino->i_sb->s_blocksize_bits;
444 /* FIXME i_blocks and i_byte - used for quota?? */
447 /* Note: no refcount yet. Either will remove the reference to the
451 rcu_assign_pointer(b->my_inode, ino);
455 printk("inode %lu type is %d\n",
456 (unsigned long)ino->i_ino, li->type);
457 unmap_dblock(b, lai);
461 void lafs_inode_checkpin(struct inode *ino)
463 /* Make sure I_Pinned is set correctly.
464 * It should be set precisely if i_nlink is non-zero,
465 * and ->iblock is B_Pinned.
466 * When it is set, we own a reference to the inode.
468 * This needs to be called whenever we change
469 * i_nlink, and whenever we pin or unpin an InoIdx
472 if (ino->i_nlink == 0) {
473 /* I_Pinned should not be set */
474 if (test_and_clear_bit(I_Pinned, &LAFSI(ino)->iflags))
477 /* Need to check if iblock is Pinned. */
478 struct indexblock *ib = NULL;
479 if (LAFSI(ino)->iblock) {
480 spin_lock(&ino->i_data.private_lock);
481 ib = LAFSI(ino)->iblock;
482 if (ib && !test_bit(B_Pinned, &ib->b.flags))
484 spin_unlock(&ino->i_data.private_lock);
487 if (!test_and_set_bit(I_Pinned, &LAFSI(ino)->iflags))
490 if (test_and_clear_bit(I_Pinned, &LAFSI(ino)->iflags))
496 struct datablock *lafs_inode_get_dblock(struct inode *ino, REFARG)
498 struct datablock *db;
500 spin_lock(&ino->i_data.private_lock);
501 db = LAFSI(ino)->dblock;
503 if (db->b.inode == ino)
504 getdref_locked(db, REF);
506 spin_lock_nested(&db->b.inode->i_data.private_lock, 1);
507 getdref_locked(db, REF);
508 spin_unlock(&db->b.inode->i_data.private_lock);
511 spin_unlock(&ino->i_data.private_lock);
515 struct datablock *lafs_inode_dblock(struct inode *ino, int async, REFARG)
517 struct datablock *db;
520 db = lafs_inode_get_dblock(ino, REF);
523 db = lafs_get_block(ino_from_sb(ino->i_sb), ino->i_ino, NULL,
528 LAFSI(ino)->dblock = db;
529 rcu_assign_pointer(db->my_inode, ino);
531 err = lafs_read_block_async(db);
533 err = lafs_read_block(db);
541 void lafs_inode_init(struct datablock *b, int type, int mode, struct inode *dir)
543 /* A new block has been allocated in an inode file to hold an
544 * inode. We get to fill in initial values so that when
545 * 'iget' calls lafs_import_inode, the correct inode is
548 struct fs *fs = fs_from_inode(b->b.inode);
549 struct la_inode *lai = map_dblock(b);
552 lai->data_blocks = cpu_to_le32(0);
553 lai->index_blocks = cpu_to_le32(0);
554 get_random_bytes(&lai->generation, sizeof(lai->generation));
557 lai->filetype = type;
563 struct fs_metadata *l = &lai->metadata[0].fs;
564 size = sizeof(struct fs_metadata);
567 l->blocks_allowed = 0;
568 l->creation_age = fs->wc[0].cluster_seq;
570 l->quota_inodes[0] = 0;
571 l->quota_inodes[1] = 0;
572 l->quota_inodes[2] = 0;
573 l->snapshot_usage_table = 0;
575 /* name will be zero length and not used */
580 struct inodemap_metadata *l = &lai->metadata[0].inodemap;
582 size = sizeof(struct inodemap_metadata);
586 size = sizeof(struct su_metadata);
589 size = sizeof(struct quota_metadata);
599 struct file_metadata *l = &lai->metadata[0].file;
600 struct timespec now = CURRENT_TIME;
602 l->flags = cpu_to_le16(0);
603 l->userid = cpu_to_le32(current->cred->fsuid);
604 if (dir && (dir->i_mode & S_ISGID)) {
605 l->groupid = cpu_to_le32(dir->i_gid);
609 l->groupid = cpu_to_le32(current->cred->fsgid);
610 if (dir && LAFSI(dir)->md.file.treeid)
611 l->treeid = cpu_to_le32(LAFSI(dir)->md.file.treeid);
613 l->treeid = l->userid;
615 l->mode = cpu_to_le16(mode);
616 l->creationtime = encode_time(&now);
617 l->modifytime = l->creationtime;
618 l->ctime = l->creationtime;
619 l->accesstime = l->creationtime;
621 l->parent = dir ? cpu_to_le32(dir->i_ino) : 0;
624 if (type == TypeDir) {
625 struct dir_metadata *l = &lai->metadata[0].dir;
627 get_random_bytes(&seed,
629 seed = (seed & ~7) | 1;
630 l->hash_seed = cpu_to_le32(seed);
631 size = sizeof(struct dir_metadata);
632 } else if (type == TypeSpecial) {
633 struct special_metadata *s = &lai->metadata[0].special;
634 s->major = s->minor = 0;
635 size = sizeof(struct special_metadata);
637 size = sizeof(struct file_metadata);
640 size += sizeof(struct la_inode);
641 lai->metadata_size = cpu_to_le32(size);
642 memset(((char *)lai)+size, 0, fs->blocksize-size);
643 *(u16 *)(((char *)lai)+size) = cpu_to_le16(IBLK_EXTENT);
645 unmap_dblock(b, lai);
646 set_bit(B_Valid, &b->b.flags);
647 LAFS_BUG(!test_bit(B_Pinned, &b->b.flags), &b->b);
648 lafs_dirty_dblock(b);
651 void lafs_clear_inode(struct inode *ino)
653 struct lafs_inode *li = LAFSI(ino);
654 dprintk("CLEAR INODE %d\n", (int)ino->i_ino);
658 /* Now is a good time to break the linkage between
659 * inode and dblock - but not if the file is
662 if (!test_bit(I_Deleting, &LAFSI(ino)->iflags)) {
663 struct datablock *db;
664 spin_lock(&ino->i_data.private_lock);
665 db = LAFSI(ino)->dblock;
667 struct indexblock *ib = LAFSI(ino)->iblock;
668 LAFS_BUG(ib && atomic_read(&ib->b.refcnt), &db->b);
670 LAFSI(ino)->dblock = NULL;
671 LAFSI(ino)->iblock = NULL;
673 spin_unlock(&ino->i_data.private_lock);
676 /* FIXME release quota inodes if filesystem */
679 static int inode_map_free(struct fs *fs, struct super_block *sb, u32 inum);
681 void lafs_delete_inode(struct inode *ino)
683 struct fs *fs = fs_from_inode(ino);
686 if (ino->i_mode == 0) {
687 /* There never was an inode here,
693 dprintk("DELETE INODE %d\n", (int)ino->i_ino);
695 /* Normal truncation holds an igrab, so we cannot be
696 * deleted until any truncation finishes
698 BUG_ON(test_bit(I_Trunc, &LAFSI(ino)->iflags));
700 b = lafs_inode_dblock(ino, SYNC, MKREF(delete_inode));
702 i_size_write(ino, 0);
703 truncate_inode_pages(&ino->i_data, 0);
704 LAFSI(ino)->trunc_next = 0;
705 set_bit(I_Trunc, &LAFSI(ino)->iflags);
707 set_bit(I_Deleting, &LAFSI(ino)->iflags);
709 set_bit(B_Claimed, &b->b.flags);
710 lafs_add_orphan(fs, b);
711 dprintk("PUNCH hole for %d\n", (int)b->b.fileaddr);
712 putdref(b, MKREF(delete_inode));
714 inode_map_free(fs, ino->i_sb, ino->i_ino);
719 static int prune(void *data, u32 addr, u64 paddr, int len)
721 /* This whole index block is being pruned, just account
722 * for everything and it will be cleared afterwards
724 struct indexblock *ib = data;
725 struct inode *ino = ib->b.inode;
726 struct fs *fs = fs_from_inode(ino);
727 int ph = !!test_bit(B_Phase1, &ib->b.flags);
729 dprintk("PRUNE %d for %d at %lld\n", addr, len, (long long)paddr);
730 if (paddr == 0 || len == 0)
732 for (i = 0 ; i < len ; i++)
733 lafs_summary_update(fs, ino, paddr+i, 0, 0, ph, 0);
737 static int prune_some(void *data, u32 addr, u64 paddr, int len)
739 /* Part of this index block is being pruned. Copy
740 * what addresses we can into uninc_table so that
741 * it can be 'incorporated'
742 * We should probably share some code with
743 * lafs_allocated_block??
745 struct indexblock *ib = data;
746 struct inode *ino = ib->b.inode;
747 struct fs *fs = fs_from_inode(ino);
748 int ph = !!test_bit(B_Phase1, &ib->b.flags);
751 if (paddr == 0 || len == 0)
753 dprintk("PRUNE2 %d for %d at %lld\n", addr, len, (long long)paddr);
754 for (i = 0 ; i < len ; i++) {
755 /* FIXME should allow longer truncation ranges in uninc_table
756 * as they are easy to handle.
759 if (addr + i < LAFSI(ino)->trunc_next)
761 spin_lock(&ino->i_data.private_lock);
762 a = &ib->uninc_table.pending_addr
763 [ib->uninc_table.pending_cnt - 1];
764 if (ib->uninc_table.pending_cnt <
765 ARRAY_SIZE(ib->uninc_table.pending_addr)) {
767 a->fileaddr = addr + i;
770 LAFS_BUG(!test_bit(B_Pinned, &ib->b.flags), &ib->b);
771 ib->uninc_table.pending_cnt++;
773 spin_unlock(&ino->i_data.private_lock);
776 spin_unlock(&ino->i_data.private_lock);
777 lafs_summary_update(fs, ino, paddr+i, 0, 0, ph, 0);
782 int lafs_inode_handle_orphan(struct datablock *b)
784 /* Don't need rcu protection for my_inode run_orphan
787 struct indexblock *ib, *ib2;
788 struct inode *ino = b->my_inode;
789 struct fs *fs = fs_from_inode(ino);
790 u32 trunc_next, next_trunc;
794 if (!test_bit(I_Trunc, &LAFSI(ino)->iflags)) {
795 if (test_bit(I_Deleting, &LAFSI(ino)->iflags)) {
796 LAFS_BUG(ino->i_nlink, &b->b);
797 if (LAFSI(ino)->cblocks +
798 LAFSI(ino)->pblocks +
799 LAFSI(ino)->ablocks +
800 LAFSI(ino)->ciblocks +
801 LAFSI(ino)->piblocks)
802 printk("Deleting inode %lu: %ld+%ld+%ld %ld+%ld\n",
807 LAFSI(ino)->ciblocks,
808 LAFSI(ino)->piblocks);
809 BUG_ON(LAFSI(ino)->cblocks +
810 LAFSI(ino)->pblocks +
811 LAFSI(ino)->ablocks +
812 LAFSI(ino)->ciblocks +
813 LAFSI(ino)->piblocks);
814 if (lafs_erase_dblock_async(b))
815 lafs_orphan_release(fs, b, NULL);
816 } else if (ino->i_nlink || LAFSI(ino)->type == 0)
817 lafs_orphan_release(fs, b, NULL);
819 lafs_orphan_forget(fs, b);
823 ib = lafs_make_iblock(ino, ADOPT, SYNC, MKREF(inode_handle_orphan));
827 /* Here is the guts of 'truncate'. We find the next leaf index
828 * block and discard all the addresses there-in.
830 trunc_next = LAFSI(ino)->trunc_next;
832 if (trunc_next == 0xFFFFFFFF) {
833 /* truncate has finished in that all data blocks
834 * have been removed and all index block are either
835 * gone or pending incorporation at which point they will
837 * If we hit a phase change, we will need to postpone
838 * the rest of the cleaning until it completes.
839 * If there is a checkpoint happening, then all the work
840 * that we can do now, it will do for us. So just
843 struct indexblock *tmp;
844 struct indexblock *next;
847 if (!test_bit(B_Pinned, &ib->b.flags)) {
848 /* must be finished */
849 LAFS_BUG(test_bit(B_Dirty, &ib->b.flags), &ib->b);
850 clear_bit(I_Trunc, &LAFSI(ino)->iflags);
851 if (!test_bit(I_Deleting, &LAFSI(ino)->iflags))
853 wake_up(&fs->trunc_wait);
857 if (fs->checkpointing) {
858 /* This cannot happen with current code,
859 * but leave it in case we ever have
860 * orphan handling parallel with checkpoints
862 err = -EBUSY; /* Try again after the checkpoint */
866 lastaddr = (i_size_read(ino) +
868 >> fs->blocksize_bits;
869 /* Find a Pinned descendent of ib which has no
870 * Pinned descendents and no PrimaryRef dependent
871 * (so take the last).
872 * Prefer blocks that are beyond EOF (again, take the last).
873 * If there are none, descend the last block that
874 * is not after EOF and look at its children.
877 spin_lock(&ib->b.inode->i_data.private_lock);
881 list_for_each_entry(tmp, &ib2->children, b.siblings) {
882 if (!test_bit(B_Index, &tmp->b.flags) ||
883 !test_bit(B_Pinned, &tmp->b.flags))
886 tmp->b.fileaddr > next->b.fileaddr)
890 if (ib2->b.fileaddr < lastaddr) {
891 /* Must be all done */
892 spin_unlock(&ib->b.inode->i_data.private_lock);
893 clear_bit(I_Trunc, &LAFSI(ino)->iflags);
894 if (!test_bit(I_Deleting, &LAFSI(ino)->iflags))
896 wake_up(&fs->trunc_wait);
900 getiref(ib2, MKREF(inode_handle_orphan2));
901 spin_unlock(&ib->b.inode->i_data.private_lock);
903 /* ib2 is an index block beyond EOF with no
905 * Incorporating it should unpin it.
907 if (!list_empty(&ib2->children)) {
908 lafs_print_tree(&ib2->b, 3);
909 LAFS_BUG(1, &ib2->b);
912 if (!lafs_iolock_written_async(&ib2->b)) {
913 putiref(ib2, MKREF(inode_handle_orphan2));
917 while (ib2->uninc_table.pending_cnt || ib2->uninc)
918 lafs_incorporate(fs, ib2);
920 if (test_bit(B_Dirty, &ib2->b.flags) ||
921 test_bit(B_Realloc, &ib2->b.flags))
922 lafs_cluster_allocate(&ib2->b, 0);
924 lafs_iounlock_block(&ib2->b);
926 if (!list_empty(&ib2->b.siblings)) {
927 printk("looping on %s\n", strblk(&ib2->b));
932 putiref(ib2, MKREF(inode_handle_orphan2));
935 if (lafs_iolock_written_async(&ib->b)) {
937 lafs_incorporate(fs, ib);
938 lafs_iounlock_block(&ib->b);
943 putiref(ib, MKREF(inode_handle_orphan));
947 putiref(ib, MKREF(inode_handle_orphan));
949 ib = lafs_leaf_find(ino, trunc_next, ADOPT, &next_trunc,
950 ASYNC, MKREF(inode_handle_orphan3));
953 /* now hold an iolock on ib */
955 /* Ok, trunc_next seems to refer to a block that exists.
956 * We need to erase it..
958 * So we open up the index block ourselves, call
959 * lafs_summary_update with each block address, and then
963 if (LAFSI(ino)->depth == 0) {
964 /* Nothing to truncate */
965 clear_bit(I_Trunc, &LAFSI(ino)->iflags);
966 if (!test_bit(I_Deleting, &LAFSI(ino)->iflags))
968 if (test_bit(B_Pinned, &ib->b.flags))
969 /* Need to move the dirtiness which keeps this
970 * pinned to the data block.
972 lafs_cluster_allocate(&ib->b, 0);
974 lafs_iounlock_block(&ib->b);
979 lafs_checkpoint_lock(fs);
980 err = lafs_reserve_block(&ib->b, ReleaseSpace);
984 if (!test_bit(B_Valid, &ib->b.flags) &&
985 test_bit(B_InoIdx, &ib->b.flags)) {
986 /* still invalid, just re-erase to remove
988 LAFSI(ino)->trunc_next = next_trunc;
989 lafs_cluster_allocate(&ib->b, 0);
994 lafs_pin_block(&ib->b);
996 /* It might be that this can happen, in which case
997 * we simply update trunc_next and loop. But I'd like
998 * to be sure before I implement that
1000 if (!test_bit(B_Valid, &ib->b.flags)) {
1001 printk("Not Valid: %s\n", strblk(&ib->b));
1002 printk("depth = %d\n", LAFSI(ino)->depth);
1003 if (test_bit(B_InoIdx, &ib->b.flags))
1004 printk("DB: %s\n", strblk(&LAFSI(ib->b.inode)->dblock->b));
1005 LAFSI(ino)->trunc_next = next_trunc;
1006 //BUG_ON(!test_bit(B_Valid, &ib->b.flags));
1011 if (ib->b.fileaddr < trunc_next &&
1012 lafs_leaf_next(ib, 0) < trunc_next) {
1013 /* We only want to truncate part of this index block.
1014 * So we copy addresses into uninc_table and then
1015 * call lafs_incorporate.
1016 * This might cause the index tree to grow, so we
1017 * cannot trust next_trunc
1019 if (ib->uninc_table.pending_cnt == 0 &&
1020 ib->uninc == NULL) {
1021 lafs_dirty_iblock(ib, 0);
1022 /* FIXME this just removes 8 blocks at a time,
1023 * which is not enough
1025 lafs_walk_leaf_index(ib, prune_some, ib);
1027 if (test_bit(B_Dirty, &ib->b.flags))
1028 lafs_incorporate(fs, ib);
1032 LAFSI(ino)->trunc_next = next_trunc;
1034 while (ib->uninc_table.pending_cnt || ib->uninc) {
1035 /* There should be no Realloc data blocks here
1036 * but index blocks might be realloc still.
1038 LAFS_BUG(!test_bit(B_Dirty, &ib->b.flags) &&
1039 !test_bit(B_Realloc, &ib->b.flags), &ib->b);
1040 lafs_incorporate(fs, ib);
1042 if (test_bit(B_InoIdx, &ib->b.flags) ||
1043 !test_bit(B_PhysValid, &ib->b.flags) ||
1044 ib->b.physaddr != 0) {
1045 lafs_walk_leaf_index(ib, prune, ib);
1046 lafs_clear_index(ib);
1047 lafs_dirty_iblock(ib, 0);
1049 if (test_bit(B_Dirty, &ib->b.flags))
1050 lafs_incorporate(fs, ib);
1051 if (!list_empty(&ib->children))
1052 lafs_print_tree(&ib->b, 2);
1053 LAFS_BUG(!list_empty(&ib->children), &ib->b);
1056 lafs_iounlock_block(&ib->b);
1058 lafs_checkpoint_unlock(fs);
1060 putiref(ib, MKREF(inode_handle_orphan3));
1064 void lafs_dirty_inode(struct inode *ino)
1066 /* this is called in one of three cases:
1067 * 1/ by lafs internally when dblock or iblock is pinned and
1068 * ready to be dirtied
1069 * 2/ by writeout before requesting a write - to update mtime
1070 * 3/ by read to update atime
1072 * As we don't know which, there is not much we can do.
1073 * We mustn't update the data block as it could be in
1074 * writeout and we cannot always wait safely.
1075 * So require that anyone who really cares, dirties the datablock
1076 * or a child themselves.
1077 * When cluster_allocate eventually gets called, it will update
1078 * the datablock from the inode.
1079 * If an update has to wait for the next phase, lock_dblock
1080 * (e.g. in setattr) will do that.
1082 * We also use this opportunity to update the filesystem modify time.
1084 struct timespec now;
1085 struct inode *filesys;
1086 set_bit(I_Dirty, &LAFSI(ino)->iflags);
1087 ino->i_sb->s_dirt = 1;
1089 now = current_fs_time(ino->i_sb);
1090 filesys = ino_from_sb(ino->i_sb);
1091 if (!timespec_equal(&filesys->i_mtime, &now)) {
1092 filesys->i_mtime = now;
1093 set_bit(I_Dirty, &LAFSI(filesys)->iflags);
1097 int lafs_sync_inode(struct inode *ino, int wait)
1099 /* fsync has been called on this file so we need
1100 * to sync any inode updates to the next cluster.
1102 * If we cannot create an update record,
1103 * we wait for a phase change, which writes everything
1106 struct datablock *b;
1107 struct fs *fs = fs_from_inode(ino);
1108 struct update_handle uh;
1112 if (LAFSI(ino)->update_cluster > 1)
1113 lafs_cluster_wait(fs, LAFSI(ino)->update_cluster);
1114 if (LAFSI(ino)->update_cluster == 1) {
1115 lafs_checkpoint_lock(fs);
1116 lafs_checkpoint_unlock_wait(fs);
1121 LAFSI(ino)->update_cluster = 0;
1122 if (!test_bit(I_Dirty, &LAFSI(ino)->iflags))
1124 b = lafs_inode_dblock(ino, SYNC, MKREF(write_inode));
1128 lafs_iolock_written(&b->b);
1129 lafs_inode_fillblock(ino);
1130 lafs_iounlock_block(&b->b);
1132 err = lafs_cluster_update_prepare(&uh, fs, LAFS_INODE_LOG_SIZE);
1134 lafs_cluster_update_abort(&uh);
1136 lafs_checkpoint_lock(fs);
1137 if (lafs_cluster_update_pin(&uh) == 0) {
1138 if (test_and_clear_bit(B_Dirty, &b->b.flags))
1139 lafs_space_return(fs, 1);
1140 LAFSI(ino)->update_cluster =
1141 lafs_cluster_update_commit
1142 (&uh, b, LAFS_INODE_LOG_START,
1143 LAFS_INODE_LOG_SIZE);
1145 lafs_cluster_update_abort(&uh);
1146 lafs_checkpoint_unlock(fs);
1148 if (test_bit(B_Dirty, &b->b.flags)) {
1149 /* FIXME need to write out the data block...
1150 * Is that just lafs_cluster_allocate ?
1154 if (LAFSI(ino)->update_cluster == 0) {
1155 lafs_checkpoint_lock(fs);
1156 if (test_bit(B_Dirty, &b->b.flags))
1157 LAFSI(ino)->update_cluster = 1;
1158 lafs_checkpoint_start(fs);
1159 lafs_checkpoint_unlock(fs);
1161 putdref(b, MKREF(write_inode));
1162 return 0; /* FIXME should I return some error message??? */
1165 void lafs_inode_fillblock(struct inode *ino)
1167 /* copy data from ino into the related data block */
1169 struct lafs_inode *li = LAFSI(ino);
1170 struct datablock *db = li->dblock;
1171 struct la_inode *lai;
1173 clear_bit(I_Dirty, &LAFSI(ino)->iflags);
1175 lai = map_dblock(db);
1176 lai->data_blocks = cpu_to_le32(li->cblocks);
1177 lai->index_blocks = cpu_to_le32(li->ciblocks);
1178 lai->generation = cpu_to_le16(ino->i_generation);
1179 lai->trunc_gen = li->trunc_gen;
1180 lai->flags = li->flags;
1181 lai->filetype = li->type;
1182 if (lai->metadata_size != cpu_to_le16(li->metadata_size)) {
1183 /* Changing metadata size is wierd.
1184 * We will need to handle this somehow for xattrs
1185 * For now we just want to cope with
1186 * Dir -> InodeFile changes, and that guarantees us
1187 * there is no index info - so just clear the index
1190 u16 *s = (u16*)(((char*)lai) + li->metadata_size);
1191 BUG_ON(li->type != TypeInodeFile);
1192 lai->metadata_size = cpu_to_le16(li->metadata_size);
1193 memset(s, 0, ino->i_sb->s_blocksize - li->metadata_size);
1194 *s = cpu_to_le16(IBLK_INDIRECT);
1196 lai->depth = li->depth;
1201 struct fs_md *i = &li->md.fs;
1202 struct fs_metadata *l = &lai->metadata[0].fs;
1205 l->snapshot_usage_table = cpu_to_le16(i->usagetable);
1206 l->update_time = cpu_to_le64(encode_time(&ino->i_mtime));
1207 l->blocks_used = cpu_to_le64(i->cblocks_used);
1208 l->blocks_allowed = cpu_to_le64(i->blocks_allowed);
1209 l->creation_age = cpu_to_le64(i->creation_age);
1210 l->inodes_used = cpu_to_le32(i->inodes_used);
1211 l->quota_inodes[0] = cpu_to_le32(i->quota_inums[0]);
1212 l->quota_inodes[1] = cpu_to_le32(i->quota_inums[1]);
1213 l->quota_inodes[2] = cpu_to_le32(i->quota_inums[2]);
1214 nlen = lai->metadata_size - offsetof(struct la_inode,
1215 metadata[0].fs.name);
1216 memset(l->name, 0, nlen);
1217 if (i->name == NULL)
1219 else if (strlen(i->name) < nlen)
1220 nlen = strlen(i->name);
1221 memcpy(l->name, i->name, nlen);
1227 struct inodemap_md *m = &li->md.inodemap;
1228 struct inodemap_metadata *s = &lai->metadata[0].inodemap;
1229 s->size = cpu_to_le32(m->size);
1233 case TypeSegmentMap:
1235 struct su_md *m = &li->md.segmentusage;
1236 struct su_metadata *s = &lai->metadata[0].segmentusage;
1237 s->table_size = cpu_to_le32(m->table_size);
1243 struct quota_md *m = &li->md.quota;
1244 struct quota_metadata *s = &lai->metadata[0].quota;
1245 s->gracetime = cpu_to_le32(m->gracetime);
1246 s->graceunits = cpu_to_le32(m->graceunits);
1249 case TypeOrphanList:
1250 case TypeAccessTime:
1253 default: /* TypeBase or larger */
1255 struct file_md *i = &li->md.file;
1256 struct file_metadata *l = &lai->metadata[0].file;
1257 struct dir_metadata *d = &lai->metadata[0].dir;
1258 struct special_metadata *s = &lai->metadata[0].special;
1260 if (li->type < TypeBase)
1262 l->flags = cpu_to_le16(i->flags);
1263 l->mode = cpu_to_le16(ino->i_mode);
1264 l->userid = cpu_to_le32(ino->i_uid);
1265 l->groupid = cpu_to_le32(ino->i_gid);
1266 l->treeid = cpu_to_le32(i->treeid);
1267 l->creationtime = cpu_to_le64(i->creationtime);
1268 l->modifytime = cpu_to_le64(encode_time(&ino->i_mtime));
1269 l->ctime = cpu_to_le64(encode_time(&ino->i_ctime));
1270 l->accesstime = cpu_to_le64(encode_time(&ino->i_atime));
1271 /* FIXME write 0 to accesstime file */
1272 l->size = cpu_to_le64(ino->i_size);
1273 l->parent = cpu_to_le32(i->parent);
1274 l->linkcount = cpu_to_le32(ino->i_nlink);
1280 d->hash_seed = cpu_to_le32(i->seed);
1285 s->major = cpu_to_le32(MAJOR(ino->i_rdev));
1286 s->minor = cpu_to_le32(MINOR(ino->i_rdev));
1291 unmap_dblock(db, lai);
1294 /*-----------------------------------------------------------------------
1295 * Inode allocate map handling.
1296 * Inode 1 of each fileset is a bitmap of free inode numbers.
1297 * Whenever the file is extended in size, new bits are set to one. They
1298 * are then cleared when the inode is allocated. When a block becomes
1299 * full of zeros, we don't need to store it any more.
1301 * We don't clear the bit until we are committed to creating an inode
1302 * This means we cannot clear it straight away, so two different threads
1303 * might see the same inode number as being available. We have two
1304 * approaches to guard against this.
1305 * Firstly we have a 'current' pointer into the inodemap file and
1306 * increase that past the inode we return. This discourages multiple
1307 * hits but as the pointer would need to be rewound occasionally it
1308 * isn't a guarantee. The guarantee against multiple allocations is done
1309 * via a flag in the block representing an inode. This is set
1310 * while an inode is being allocated.
1313 /* inode number allocation has the prealloc/pin/commit/abort structure
1314 * so it can be committed effectively
1318 choose_free_inum(struct fs *fs, struct super_block *sb, u32 *inump,
1319 struct datablock **bp, int *restarted)
1321 struct inode *im = lafs_iget(sb, 1, SYNC);
1323 struct datablock *b;
1329 struct inode *i = (*bp)->b.inode;
1330 putdref(*bp, MKREF(cfi_map));
1335 mutex_lock_nested(&im->i_mutex, I_MUTEX_QUOTA);
1337 bnum = LAFSI(im)->md.inodemap.thisblock;
1339 if (bnum == NoBlock ||
1340 LAFSI(im)->md.inodemap.nextbit >= (fs->blocksize<<3)) {
1341 if (bnum == NoBlock)
1342 bnum = LAFSI(im)->md.inodemap.size;
1344 if (bnum+1 < LAFSI(im)->md.inodemap.size)
1346 else if (!*restarted) {
1350 /* Need to add a new block to the file */
1351 bnum = LAFSI(im)->md.inodemap.size;
1352 b = lafs_get_block(im, bnum, NULL, GFP_KERNEL,
1357 lafs_iolock_written(&b->b);
1358 set_bit(B_PinPending, &b->b.flags);
1359 lafs_iounlock_block(&b->b);
1361 lafs_checkpoint_lock(fs);
1362 err = lafs_pin_dblock(b, NewSpace);
1363 if (err == -EAGAIN) {
1364 lafs_checkpoint_unlock_wait(fs);
1370 buf = map_dblock(b);
1371 /* Set block to "all are free" */
1372 memset(buf, 0xff, fs->blocksize);
1373 unmap_dblock(b, buf);
1374 set_bit(B_Valid, &b->b.flags);
1375 LAFSI(im)->md.inodemap.size = bnum+1;
1376 lafs_dirty_inode(im);
1377 lafs_dirty_dblock(b);
1378 lafs_checkpoint_unlock(fs);
1379 putdref(b, MKREF(cfi_map));
1382 err = lafs_find_next(im, &bnum);
1388 LAFSI(im)->md.inodemap.nextbit = 0;
1389 LAFSI(im)->md.inodemap.thisblock = bnum;
1392 b = lafs_get_block(im, bnum, NULL, GFP_KERNEL, MKREF(cfi_map));
1396 err = lafs_find_block(b, NOADOPT);
1399 if (b->b.physaddr == 0 && !test_bit(B_Valid, &b->b.flags)) {
1400 LAFSI(im)->md.inodemap.nextbit =
1401 (fs->blocksize<<3) + 1;
1402 putdref(b,MKREF(cfi_map));
1405 err = lafs_read_block(b);
1409 bit = LAFSI(im)->md.inodemap.nextbit;
1410 LAFSI(im)->md.inodemap.thisblock = bnum;
1411 buf = map_dblock(b);
1412 while (bnum == 0 && bit < 16) {
1413 /* Never return an inum below 16 - they are special */
1414 if (!generic_test_le_bit(bit, (unsigned long *)buf))
1415 generic___clear_le_bit(bit, (unsigned long *)buf);
1419 bit = generic_find_next_le_bit((unsigned long *)buf,
1420 fs->blocksize<<3, bit);
1421 unmap_dblock(b, buf);
1422 LAFSI(im)->md.inodemap.nextbit = bit+1;
1423 if (bit >= fs->blocksize<<3) {
1424 putdref(b,MKREF(cfi_map));
1427 mutex_unlock(&im->i_mutex);
1429 *inump = bit + (bnum << (im->i_blkbits + 3));
1433 lafs_checkpoint_unlock(fs);
1435 putdref(b, MKREF(cfi_map));
1437 mutex_unlock(&im->i_mutex);
1442 struct inode_map_new_info {
1443 struct datablock *ib, *mb;
1447 inode_map_new_prepare(struct fs *fs, int inum, struct super_block *sb,
1448 struct inode_map_new_info *imni)
1453 struct datablock *b;
1455 imni->ib = imni->mb = NULL;
1458 /* choose a possibly-free inode number */
1459 err = choose_free_inum(fs, sb, &choice,
1460 &imni->mb, &restarted);
1464 b = lafs_get_block(ino_from_sb(sb), choice, NULL, GFP_KERNEL,
1469 if (test_and_set_bit(B_Claimed, &b->b.flags)) {
1470 putdref(b, MKREF(cfi_ino));
1476 lafs_iolock_written(&imni->mb->b);
1477 set_bit(B_PinPending, &imni->mb->b.flags);
1478 lafs_iounlock_block(&imni->mb->b);
1480 set_bit(B_PinPending, &b->b.flags);
1487 inode_map_new_pin(struct inode_map_new_info *imni)
1491 err = lafs_pin_dblock(imni->mb, NewSpace);
1492 err = err ?: lafs_pin_dblock(imni->ib, NewSpace);
1497 inode_map_new_commit(struct inode_map_new_info *imni)
1502 int blksize = imni->ib->b.inode->i_sb->s_blocksize;
1503 int bit = imni->ib->b.fileaddr & (blksize*8 - 1);
1505 struct inode *ino = imni->mb->b.inode;
1507 mutex_lock_nested(&ino->i_mutex, I_MUTEX_QUOTA);
1508 buf = map_dblock(imni->mb);
1509 generic___clear_le_bit(bit, buf);
1510 if (buf[blksize/sizeof(*buf)-1] == 0 &&
1511 generic_find_next_le_bit(buf, blksize*8, 0) == blksize*8)
1512 /* block is empty, punch a hole */
1515 unmap_dblock(imni->mb, buf);
1517 lafs_erase_dblock(imni->mb);
1519 lafs_dirty_dblock(imni->mb);
1521 putdref(imni->mb, MKREF(cfi_map));
1522 mutex_unlock(&ino->i_mutex);
1525 putdref(imni->ib, MKREF(cfi_ino));
1529 inode_map_new_abort(struct inode_map_new_info *imni)
1532 clear_bit(B_Claimed, &imni->ib->b.flags);
1533 clear_bit(B_PinPending, &imni->ib->b.flags);
1534 lafs_orphan_release(fs_from_inode(imni->ib->b.inode),
1537 putdref(imni->ib, MKREF(cfi_ino));
1539 struct inode *ino = imni->mb->b.inode;
1540 putdref(imni->mb, MKREF(cfi_map));
1546 lafs_new_inode(struct fs *fs, struct super_block *sb, struct inode *dir,
1547 int type, int inum, int mode, struct datablock **inodbp)
1549 /* allocate and instantiate a new inode. If inum is non-zero,
1550 * choose any number, otherwise we are creating a special inode
1551 * and have to use the given number.
1552 * This creation is committed independently of any name that might
1553 * subsequently be given to the inode. So we register it as an
1554 * orphan so that it will be cleaned up if the name isn't
1555 * successfully created
1559 struct datablock *b;
1560 struct inode_map_new_info imni;
1561 struct update_handle ui;
1564 err = inode_map_new_prepare(fs, inum, sb, &imni);
1565 err = lafs_cluster_update_prepare(&ui, fs, sizeof(struct la_inode))
1568 err = lafs_make_orphan(fs, imni.ib, NULL);
1572 lafs_checkpoint_lock(fs);
1574 err = inode_map_new_pin(&imni);
1576 if (err == -EAGAIN) {
1577 lafs_checkpoint_unlock_wait(fs);
1583 b = getdref(imni.ib, MKREF(inode_new));
1585 lafs_iolock_block(&b->b); /* make sure we don't race with the cleaner
1586 * and zero this inode while trying to load it
1588 lafs_inode_init(b, type, mode, dir);
1589 lafs_iounlock_block(&b->b);
1591 inode_map_new_commit(&imni);
1592 ino = lafs_iget(sb, b->b.fileaddr, SYNC);
1594 lafs_cluster_update_abort(&ui);
1597 lafs_cluster_update_commit(&ui, b, 0,
1598 LAFSI(ino)->metadata_size);
1599 LAFS_BUG(LAFSI(ino)->dblock != b, &b->b);
1600 LAFS_BUG(b->my_inode != ino, &b->b);
1601 lafs_checkpoint_unlock(fs);
1606 putdref(b, MKREF(inode_new));
1610 lafs_checkpoint_unlock(fs);
1613 inode_map_new_abort(&imni);
1614 lafs_cluster_update_abort(&ui);
1615 dprintk("After abort %d: %s\n", err, strblk(&imni.ib->b));
1616 return ERR_PTR(err);
1619 static int inode_map_free(struct fs *fs, struct super_block *sb, u32 inum)
1621 struct inode *im = lafs_iget(sb, 1, SYNC);
1624 struct datablock *b;
1628 mutex_lock_nested(&im->i_mutex, I_MUTEX_QUOTA);
1630 bnum = inum >> (3 + sb->s_blocksize_bits);
1631 bit = inum - (bnum << (3 + sb->s_blocksize_bits));
1632 b = lafs_get_block(im, bnum, NULL, GFP_KERNEL, MKREF(inode_map_free));
1634 mutex_unlock(&im->i_mutex);
1638 err = lafs_read_block(b);
1640 putdref(b, MKREF(inode_map_free));
1641 mutex_unlock(&im->i_mutex);
1645 lafs_iolock_written(&b->b);
1646 set_bit(B_PinPending, &b->b.flags);
1647 lafs_iounlock_block(&b->b);
1649 lafs_checkpoint_lock(fs);
1650 err = lafs_pin_dblock(b, ReleaseSpace);
1651 if (err == -EAGAIN) {
1652 lafs_checkpoint_unlock_wait(fs);
1656 buf = map_dblock(b);
1657 generic___set_le_bit(bit, buf);
1658 unmap_dblock(b, buf);
1659 lafs_dirty_dblock(b);
1660 putdref(b, MKREF(inode_map_free));
1661 lafs_checkpoint_unlock(fs);
1662 mutex_unlock(&im->i_mutex);
1667 int lafs_setattr(struct dentry *dentry, struct iattr *attr)
1670 struct inode *ino = dentry->d_inode;
1671 struct fs *fs = fs_from_inode(ino);
1672 struct datablock *db;
1674 err = inode_change_ok(ino, attr);
1675 db = lafs_inode_dblock(ino, SYNC, MKREF(setattr));
1681 /* We don't need iolock_written here as we don't
1682 * actually change the inode block yet
1684 lafs_iolock_block(&db->b);
1685 set_bit(B_PinPending, &db->b.flags);
1686 lafs_iounlock_block(&db->b);
1688 /* FIXME quota stuff */
1691 lafs_checkpoint_lock(fs);
1692 err = lafs_pin_dblock(db, ReleaseSpace);
1693 if (err == -EAGAIN) {
1694 lafs_checkpoint_unlock_wait(fs);
1697 /* inode_setattr calls lafs_dirty_inode, which sets
1698 * I_Dirty so the dblock will get updated.
1700 err = err ?: inode_setattr(ino, attr);
1702 lafs_dirty_dblock(db);
1703 clear_bit(B_PinPending, &db->b.flags);
1704 putdref(db, MKREF(setattr));
1705 lafs_checkpoint_unlock(fs);
1710 void lafs_truncate(struct inode *ino)
1712 /* Want to truncate this file.
1713 * i_size has already been changed, and the address space
1714 * has been cleaned up.
1715 * So just start the background truncate
1717 struct fs *fs = fs_from_inode(ino);
1718 struct datablock *db = lafs_inode_dblock(ino, SYNC, MKREF(trunc));
1725 trunc_block = ((i_size_read(ino) + fs->blocksize - 1)
1726 >> fs->blocksize_bits);
1727 /* We hold i_mutex, so regular orphan processing cannot
1728 * contine - we have to push it forward ourselves.
1730 while (test_bit(I_Trunc, &LAFSI(ino)->iflags) &&
1731 LAFSI(ino)->trunc_next < trunc_block) {
1732 prepare_to_wait(&fs->async_complete, &wq,
1733 TASK_UNINTERRUPTIBLE);
1734 lafs_inode_handle_orphan(db);
1735 if (test_bit(B_Orphan, &db->b.flags))
1738 finish_wait(&fs->async_complete, &wq);
1740 /* There is nothing we can do about errors here. The
1741 * most likely are ENOMEM which itself is very unlikely.
1742 * If this doesn't get registered as an orphan .... maybe
1743 * it will have to wait until something else truncates it.
1745 lafs_make_orphan(fs, db, NULL);
1747 if (!test_and_set_bit(I_Trunc, &LAFSI(ino)->iflags))
1749 if (trunc_block == 0)
1750 LAFSI(ino)->trunc_gen++;
1751 LAFSI(ino)->trunc_next = trunc_block;
1752 putdref(db, MKREF(trunc));
1755 const struct inode_operations lafs_special_ino_operations = {
1756 .setattr = lafs_setattr,
1757 .truncate = lafs_truncate,