2 * Copyright (c) 2007 The DragonFly Project. All rights reserved.
4 * This code is derived from software contributed to The DragonFly Project
5 * by Matthew Dillon <dillon@backplane.com>
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
17 * 3. Neither the name of The DragonFly Project nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific, prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
25 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * $DragonFly: src/sys/vfs/hammer/hammer_inode.c,v 1.33 2008/03/29 20:12:54 dillon Exp $
42 * The kernel is not actively referencing this vnode but is still holding
46 hammer_vop_inactive(struct vop_inactive_args *ap)
48 struct hammer_inode *ip = VTOI(ap->a_vp);
59 * If the inode no longer has any references we recover its
60 * in-memory resources immediately.
62 if (ip->ino_rec.ino_nlinks == 0)
68 * Release the vnode association. This is typically (but not always)
69 * the last reference on the inode and will flush the inode to the
72 * XXX Currently our sync code only runs through inodes with vnode
73 * associations, so we depend on hammer_rel_inode() to sync any inode
74 * record data to the block device prior to losing the association.
75 * Otherwise transactions that the user expected to be distinct by
76 * doing a manual sync may be merged.
79 hammer_vop_reclaim(struct vop_reclaim_args *ap)
81 struct hammer_inode *ip;
86 if ((ip = vp->v_data) != NULL) {
89 hammer_rel_inode(ip, 0);
95 * Return a locked vnode for the specified inode. The inode must be
96 * referenced but NOT LOCKED on entry and will remain referenced on
100 hammer_get_vnode(struct hammer_inode *ip, int lktype, struct vnode **vpp)
106 if ((vp = ip->vp) == NULL) {
107 error = getnewvnode(VT_HAMMER, ip->hmp->mp, vpp, 0, 0);
110 hammer_lock_ex(&ip->lock);
111 if (ip->vp != NULL) {
112 hammer_unlock(&ip->lock);
117 hammer_ref(&ip->lock);
120 vp->v_type = hammer_get_vnode_type(
121 ip->ino_rec.base.base.obj_type);
123 switch(ip->ino_rec.base.base.obj_type) {
124 case HAMMER_OBJTYPE_CDEV:
125 case HAMMER_OBJTYPE_BDEV:
126 vp->v_ops = &ip->hmp->mp->mnt_vn_spec_ops;
127 addaliasu(vp, ip->ino_data.rmajor,
128 ip->ino_data.rminor);
130 case HAMMER_OBJTYPE_FIFO:
131 vp->v_ops = &ip->hmp->mp->mnt_vn_fifo_ops;
138 * Only mark as the root vnode if the ip is not
139 * historical, otherwise the VFS cache will get
140 * confused. The other half of the special handling
141 * is in hammer_vop_nlookupdotdot().
143 if (ip->obj_id == HAMMER_OBJID_ROOT &&
144 ip->obj_asof == ip->hmp->asof) {
148 vp->v_data = (void *)ip;
149 /* vnode locked by getnewvnode() */
150 /* make related vnode dirty if inode dirty? */
151 hammer_unlock(&ip->lock);
152 if (vp->v_type == VREG)
153 vinitvmio(vp, ip->ino_rec.ino_size);
158 * loop if the vget fails (aka races), or if the vp
159 * no longer matches ip->vp.
161 if (vget(vp, LK_EXCLUSIVE) == 0) {
172 * Acquire a HAMMER inode. The returned inode is not locked. These functions
173 * do not attach or detach the related vnode (use hammer_get_vnode() for
176 * The flags argument is only applied for newly created inodes, and only
177 * certain flags are inherited.
179 struct hammer_inode *
180 hammer_get_inode(hammer_transaction_t trans, struct hammer_node **cache,
181 u_int64_t obj_id, hammer_tid_t asof, int flags, int *errorp)
183 hammer_mount_t hmp = trans->hmp;
184 struct hammer_inode_info iinfo;
185 struct hammer_cursor cursor;
186 struct hammer_inode *ip;
189 * Determine if we already have an inode cached. If we do then
192 iinfo.obj_id = obj_id;
193 iinfo.obj_asof = asof;
195 ip = hammer_ino_rb_tree_RB_LOOKUP_INFO(&hmp->rb_inos_root, &iinfo);
197 hammer_ref(&ip->lock);
202 ip = kmalloc(sizeof(*ip), M_HAMMER, M_WAITOK|M_ZERO);
203 ++hammer_count_inodes;
205 ip->obj_asof = iinfo.obj_asof;
207 ip->flags = flags & HAMMER_INODE_RO;
209 ip->flags |= HAMMER_INODE_RO;
210 RB_INIT(&ip->rec_tree);
213 * Locate the on-disk inode.
216 hammer_init_cursor(trans, &cursor, cache);
217 cursor.key_beg.obj_id = ip->obj_id;
218 cursor.key_beg.key = 0;
219 cursor.key_beg.create_tid = 0;
220 cursor.key_beg.delete_tid = 0;
221 cursor.key_beg.rec_type = HAMMER_RECTYPE_INODE;
222 cursor.key_beg.obj_type = 0;
223 cursor.asof = iinfo.obj_asof;
224 cursor.flags = HAMMER_CURSOR_GET_RECORD | HAMMER_CURSOR_GET_DATA |
227 *errorp = hammer_btree_lookup(&cursor);
228 if (*errorp == EDEADLK) {
229 hammer_done_cursor(&cursor);
234 * On success the B-Tree lookup will hold the appropriate
235 * buffer cache buffers and provide a pointer to the requested
236 * information. Copy the information to the in-memory inode
237 * and cache the B-Tree node to improve future operations.
240 ip->ino_rec = cursor.record->inode;
241 ip->ino_data = cursor.data->inode;
242 ip->sync_tid = ip->ino_rec.base.base.create_tid;
243 hammer_cache_node(cursor.node, &ip->cache[0]);
245 hammer_cache_node(cursor.node, cache);
249 * On success load the inode's record and data and insert the
250 * inode into the B-Tree. It is possible to race another lookup
251 * insertion of the same inode so deal with that condition too.
253 * The cursor's locked node interlocks against others creating and
254 * destroying ip while we were blocked.
257 hammer_ref(&ip->lock);
258 if (RB_INSERT(hammer_ino_rb_tree, &hmp->rb_inos_root, ip)) {
259 hammer_uncache_node(&ip->cache[0]);
260 hammer_uncache_node(&ip->cache[1]);
261 hammer_unref(&ip->lock);
262 --hammer_count_inodes;
264 hammer_done_cursor(&cursor);
267 ip->flags |= HAMMER_INODE_ONDISK;
269 --hammer_count_inodes;
273 hammer_done_cursor(&cursor);
278 * Create a new filesystem object, returning the inode in *ipp. The
279 * returned inode will be referenced but not locked.
281 * The inode is created in-memory and will be delay-synchronized to the
285 hammer_create_inode(hammer_transaction_t trans, struct vattr *vap,
286 struct ucred *cred, hammer_inode_t dip,
287 struct hammer_inode **ipp)
294 ip = kmalloc(sizeof(*ip), M_HAMMER, M_WAITOK|M_ZERO);
295 ++hammer_count_inodes;
296 ip->obj_id = hammer_alloc_tid(trans);
297 KKASSERT(ip->obj_id != 0);
298 ip->obj_asof = hmp->asof;
300 ip->flags = HAMMER_INODE_DDIRTY | HAMMER_INODE_RDIRTY |
301 HAMMER_INODE_ITIMES | HAMMER_INODE_TIDLOCKED;
302 ip->last_tid = trans->tid;
304 RB_INIT(&ip->rec_tree);
306 ip->ino_rec.ino_atime = trans->tid;
307 ip->ino_rec.ino_mtime = trans->tid;
308 ip->ino_rec.ino_size = 0;
309 ip->ino_rec.ino_nlinks = 0;
311 ip->ino_rec.base.base.btype = HAMMER_BTREE_TYPE_RECORD;
312 ip->ino_rec.base.base.obj_id = ip->obj_id;
313 ip->ino_rec.base.base.key = 0;
314 ip->ino_rec.base.base.create_tid = trans->tid;
315 ip->ino_rec.base.base.delete_tid = 0;
316 ip->ino_rec.base.base.rec_type = HAMMER_RECTYPE_INODE;
317 ip->ino_rec.base.base.obj_type = hammer_get_obj_type(vap->va_type);
319 ip->ino_data.version = HAMMER_INODE_DATA_VERSION;
320 ip->ino_data.mode = vap->va_mode;
321 ip->ino_data.ctime = trans->tid;
322 ip->ino_data.parent_obj_id = (dip) ? dip->ino_rec.base.base.obj_id : 0;
324 switch(ip->ino_rec.base.base.obj_type) {
325 case HAMMER_OBJTYPE_CDEV:
326 case HAMMER_OBJTYPE_BDEV:
327 ip->ino_data.rmajor = vap->va_rmajor;
328 ip->ino_data.rminor = vap->va_rminor;
335 * Calculate default uid/gid and overwrite with information from
338 xuid = hammer_to_unix_xid(&dip->ino_data.uid);
339 ip->ino_data.gid = dip->ino_data.gid;
340 xuid = vop_helper_create_uid(hmp->mp, dip->ino_data.mode, xuid, cred,
342 ip->ino_data.mode = vap->va_mode;
344 if (vap->va_vaflags & VA_UID_UUID_VALID)
345 ip->ino_data.uid = vap->va_uid_uuid;
346 else if (vap->va_uid != (uid_t)VNOVAL)
347 hammer_guid_to_uuid(&ip->ino_data.uid, xuid);
348 if (vap->va_vaflags & VA_GID_UUID_VALID)
349 ip->ino_data.gid = vap->va_gid_uuid;
350 else if (vap->va_gid != (gid_t)VNOVAL)
351 hammer_guid_to_uuid(&ip->ino_data.gid, vap->va_gid);
353 hammer_ref(&ip->lock);
354 if (RB_INSERT(hammer_ino_rb_tree, &hmp->rb_inos_root, ip)) {
355 hammer_unref(&ip->lock);
356 panic("hammer_create_inode: duplicate obj_id %llx", ip->obj_id);
363 * Called by hammer_sync_inode().
366 hammer_update_inode(hammer_transaction_t trans, hammer_inode_t ip)
368 struct hammer_cursor cursor;
369 hammer_record_t record;
373 * Locate the record on-disk and mark it as deleted. Both the B-Tree
374 * node and the record must be marked deleted. The record may or
375 * may not be physically deleted, depending on the retention policy.
377 * If the inode has already been deleted on-disk we have nothing
380 * XXX Update the inode record and data in-place if the retention
386 if ((ip->flags & (HAMMER_INODE_ONDISK|HAMMER_INODE_DELONDISK)) ==
387 HAMMER_INODE_ONDISK) {
388 hammer_init_cursor(trans, &cursor, &ip->cache[0]);
389 cursor.key_beg.obj_id = ip->obj_id;
390 cursor.key_beg.key = 0;
391 cursor.key_beg.create_tid = 0;
392 cursor.key_beg.delete_tid = 0;
393 cursor.key_beg.rec_type = HAMMER_RECTYPE_INODE;
394 cursor.key_beg.obj_type = 0;
395 cursor.asof = ip->obj_asof;
396 cursor.flags |= HAMMER_CURSOR_GET_RECORD | HAMMER_CURSOR_ASOF;
398 error = hammer_btree_lookup(&cursor);
401 error = hammer_ip_delete_record(&cursor, trans->tid);
403 ip->flags |= HAMMER_INODE_DELONDISK;
404 hammer_cache_node(cursor.node, &ip->cache[0]);
406 hammer_done_cursor(&cursor);
407 if (error == EDEADLK)
412 * Write out a new record if the in-memory inode is not marked
413 * as having been deleted. Update our inode statistics if this
414 * is the first application of the inode on-disk.
416 * If the inode has been deleted permanently, HAMMER_INODE_DELONDISK
417 * will remain set and prevent further updates.
419 if (error == 0 && (ip->flags & HAMMER_INODE_DELETED) == 0) {
420 record = hammer_alloc_mem_record(ip);
421 record->rec.inode = ip->ino_rec;
422 record->rec.inode.base.base.create_tid = trans->tid;
423 record->rec.inode.base.data_len = sizeof(ip->ino_data);
424 record->data = (void *)&ip->ino_data;
425 error = hammer_ip_sync_record(trans, record);
426 record->flags |= HAMMER_RECF_DELETED;
427 hammer_rel_mem_record(record);
429 ip->flags &= ~(HAMMER_INODE_RDIRTY |
430 HAMMER_INODE_DDIRTY |
431 HAMMER_INODE_DELONDISK |
432 HAMMER_INODE_ITIMES);
433 if ((ip->flags & HAMMER_INODE_ONDISK) == 0) {
434 hammer_modify_volume(trans, ip->hmp->rootvol,
436 ++ip->hmp->rootvol->ondisk->vol0_stat_inodes;
437 ip->flags |= HAMMER_INODE_ONDISK;
441 * Unlock the sync TID if it was locked, now that
442 * we have written it out to disk.
444 ip->flags &= ~HAMMER_INODE_TIDLOCKED;
445 ip->sync_tid = trans->tid;
452 * Update only the itimes fields. This is done no-historically. The
453 * record is updated in-place on the disk.
456 hammer_update_itimes(hammer_transaction_t trans, hammer_inode_t ip)
458 struct hammer_cursor cursor;
459 struct hammer_inode_record *rec;
464 if ((ip->flags & (HAMMER_INODE_ONDISK|HAMMER_INODE_DELONDISK)) ==
465 HAMMER_INODE_ONDISK) {
466 hammer_init_cursor(trans, &cursor, &ip->cache[0]);
467 cursor.key_beg.obj_id = ip->obj_id;
468 cursor.key_beg.key = 0;
469 cursor.key_beg.create_tid = 0;
470 cursor.key_beg.delete_tid = 0;
471 cursor.key_beg.rec_type = HAMMER_RECTYPE_INODE;
472 cursor.key_beg.obj_type = 0;
473 cursor.asof = ip->obj_asof;
474 cursor.flags |= HAMMER_CURSOR_GET_RECORD | HAMMER_CURSOR_ASOF;
476 error = hammer_btree_lookup(&cursor);
478 rec = &cursor.record->inode;
479 hammer_modify_buffer(cursor.trans, cursor.record_buffer,
481 rec->ino_atime = ip->ino_rec.ino_atime;
482 rec->ino_mtime = ip->ino_rec.ino_mtime;
483 ip->flags &= ~HAMMER_INODE_ITIMES;
484 /* XXX recalculate crc */
485 hammer_cache_node(cursor.node, &ip->cache[0]);
487 hammer_done_cursor(&cursor);
488 if (error == EDEADLK)
495 * Release a reference on an inode. If asked to flush the last release
496 * will flush the inode.
499 hammer_rel_inode(struct hammer_inode *ip, int flush)
501 hammer_unref(&ip->lock);
503 ip->flags |= HAMMER_INODE_FLUSH;
504 if (ip->lock.refs == 0) {
505 KKASSERT(ip->cursor_ip_refs == 0);
506 if (ip->flags & HAMMER_INODE_FLUSH)
507 hammer_unload_inode(ip, (void *)MNT_WAIT);
509 hammer_unload_inode(ip, (void *)MNT_NOWAIT);
514 * Unload and destroy the specified inode.
516 * (typically called via RB_SCAN)
519 hammer_unload_inode(struct hammer_inode *ip, void *data)
523 KASSERT(ip->lock.refs == 0,
524 ("hammer_unload_inode: %d refs\n", ip->lock.refs));
525 KKASSERT(ip->vp == NULL);
526 hammer_ref(&ip->lock);
528 error = hammer_sync_inode(ip, (int)data, 1);
530 kprintf("hammer_sync_inode failed error %d\n", error);
531 if (ip->lock.refs == 1) {
532 KKASSERT(RB_EMPTY(&ip->rec_tree));
533 RB_REMOVE(hammer_ino_rb_tree, &ip->hmp->rb_inos_root, ip);
535 hammer_uncache_node(&ip->cache[0]);
536 hammer_uncache_node(&ip->cache[1]);
537 --hammer_count_inodes;
540 hammer_unref(&ip->lock);
546 * A transaction has modified an inode, requiring updates as specified by
549 * HAMMER_INODE_RDIRTY: Inode record has been updated
550 * HAMMER_INODE_DDIRTY: Inode data has been updated
551 * HAMMER_INODE_DELETED: Inode record/data must be deleted
552 * HAMMER_INODE_ITIMES: mtime/atime has been updated
554 * last_tid is the TID to use to generate the correct TID when the inode
555 * is synced to disk. The first inode record laid out on disk must match
556 * the transaction id of the related directory entry so only update last_tid
557 * if that has already occured.
560 hammer_modify_inode(struct hammer_transaction *trans,
561 struct hammer_inode *ip, int flags)
563 KKASSERT ((ip->flags & HAMMER_INODE_RO) == 0 ||
564 (HAMMER_INODE_RDIRTY|HAMMER_INODE_DDIRTY|
565 HAMMER_INODE_DELETED|HAMMER_INODE_ITIMES) == 0);
568 (HAMMER_INODE_RDIRTY|HAMMER_INODE_DDIRTY|HAMMER_INODE_DELETED)) {
569 if (hammer_debug_tid) {
570 kprintf("hammer_modify_inode: %016llx (%08x)\n",
571 trans->tid, (int)(trans->tid / 1000000000LL));
575 * Update the inode sync transaction id unless it's locked
576 * due to some prior required synchroznization. Locking the
577 * tid in the new flags overrides this (used by rename).
579 if ((ip->flags & HAMMER_INODE_TIDLOCKED) == 0)
580 ip->last_tid = trans->tid;
581 else if (flags & HAMMER_INODE_TIDLOCKED)
582 ip->last_tid = trans->tid;
588 * Sync any dirty buffers and records associated with an inode. The
589 * inode's last_tid field is used as the transaction id for the sync,
590 * overriding any intermediate TIDs that were used for records. Note
591 * that the dirty buffer cache buffers do not have any knowledge of
592 * the transaction id they were modified under.
595 hammer_sync_inode_callback(hammer_record_t rec, void *data)
597 hammer_transaction_t trans = data;
600 hammer_ref(&rec->lock);
601 error = hammer_ip_sync_record(trans, rec);
602 hammer_rel_mem_record(rec);
606 if (error != -ENOSPC) {
607 kprintf("hammer_sync_inode_callback: sync failed rec "
608 "%p, error %d\n", rec, error);
618 hammer_sync_inode(hammer_inode_t ip, int waitfor, int handle_delete)
620 struct hammer_transaction trans;
623 if ((ip->flags & HAMMER_INODE_MODMASK) == 0) {
627 hammer_lock_ex(&ip->lock);
630 * Use the transaction id of the last operation to sync. But we
631 * can't reuse a previous sync TID.
633 if (ip->last_tid && ip->last_tid != ip->sync_tid)
634 hammer_start_transaction_tid(&trans, ip->hmp, ip->last_tid);
636 hammer_start_transaction(&trans, ip->hmp);
639 * If the inode has been deleted (nlinks == 0), and the OS no longer
640 * has any references to it (handle_delete != 0), clean up in-memory
643 * NOTE: We do not set the RDIRTY flag when updating the delete_tid,
644 * setting HAMMER_INODE_DELETED takes care of it.
646 * NOTE: Because we may sync records within this new transaction,
647 * force the inode update later on to use our transaction id or
648 * the delete_tid of the inode may be less then the create_tid of
649 * the inode update. XXX shouldn't happen but don't take the chance.
651 if (ip->ino_rec.ino_nlinks == 0 && handle_delete &&
652 (ip->flags & HAMMER_INODE_GONE) == 0) {
653 ip->flags |= HAMMER_INODE_GONE;
655 vtruncbuf(ip->vp, 0, HAMMER_BUFSIZE);
656 error = hammer_ip_delete_range_all(&trans, ip);
657 KKASSERT(RB_EMPTY(&ip->rec_tree));
658 ip->ino_rec.base.base.delete_tid = trans.tid;
659 hammer_modify_inode(&trans, ip, HAMMER_INODE_DELETED);
660 hammer_modify_volume(&trans, ip->hmp->rootvol, NULL, 0);
661 --ip->hmp->rootvol->ondisk->vol0_stat_inodes;
665 * Sync the buffer cache.
667 if (ip->vp != NULL) {
668 error = vfsync(ip->vp, waitfor, 1, NULL, NULL);
669 if (RB_ROOT(&ip->vp->v_rbdirty_tree) == NULL)
670 ip->flags &= ~HAMMER_INODE_BUFS;
677 * Now sync related records
680 error = RB_SCAN(hammer_rec_rb_tree, &ip->rec_tree, NULL,
681 hammer_sync_inode_callback, &trans);
682 KKASSERT(error <= 0);
687 if (RB_EMPTY(&ip->rec_tree))
688 ip->flags &= ~HAMMER_INODE_XDIRTY;
691 * Now update the inode's on-disk inode-data and/or on-disk record.
693 switch(ip->flags & (HAMMER_INODE_DELETED|HAMMER_INODE_ONDISK)) {
694 case HAMMER_INODE_DELETED|HAMMER_INODE_ONDISK:
696 * If deleted and on-disk, don't set any additional flags.
697 * the delete flag takes care of things.
700 case HAMMER_INODE_DELETED:
702 * Take care of the case where a deleted inode was never
703 * flushed to the disk in the first place.
705 ip->flags &= ~(HAMMER_INODE_RDIRTY|HAMMER_INODE_DDIRTY|
706 HAMMER_INODE_XDIRTY|HAMMER_INODE_ITIMES);
707 while (RB_ROOT(&ip->rec_tree)) {
708 hammer_record_t rec = RB_ROOT(&ip->rec_tree);
709 hammer_ref(&rec->lock);
710 rec->flags |= HAMMER_RECF_DELETED;
711 hammer_rel_mem_record(rec);
714 case HAMMER_INODE_ONDISK:
716 * If already on-disk, do not set any additional flags.
721 * If not on-disk and not deleted, set both dirty flags
722 * to force an initial record to be written.
724 ip->flags |= HAMMER_INODE_RDIRTY | HAMMER_INODE_DDIRTY;
729 * If RDIRTY or DDIRTY is set, write out a new record. If the inode
730 * is already on-disk the old record is marked as deleted.
732 * If DELETED is set hammer_update_inode() will delete the existing
733 * record without writing out a new one.
735 * If *ONLY* the ITIMES flag is set we can update the record in-place.
737 if ((ip->flags & (HAMMER_INODE_RDIRTY | HAMMER_INODE_DDIRTY |
738 HAMMER_INODE_ITIMES | HAMMER_INODE_DELETED)) ==
739 HAMMER_INODE_ITIMES) {
740 error = hammer_update_itimes(&trans, ip);
742 if (ip->flags & (HAMMER_INODE_RDIRTY | HAMMER_INODE_DDIRTY |
743 HAMMER_INODE_ITIMES | HAMMER_INODE_DELETED)) {
744 error = hammer_update_inode(&trans, ip);
746 hammer_commit_transaction(&trans);
747 hammer_unlock(&ip->lock);