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 1.15 2007/12/31 05:33:12 dillon Exp $
42 hammer_vop_inactive(struct vop_inactive_args *ap)
44 struct hammer_inode *ip = VTOI(ap->a_vp);
55 * If the inode no longer has any references we recover its
56 * in-memory resources immediately.
58 if (ip->ino_rec.ino_nlinks == 0 &&
59 (ip->hmp->mp->mnt_flag & MNT_RDONLY) == 0) {
60 hammer_sync_inode(ip, MNT_NOWAIT, 1);
66 hammer_vop_reclaim(struct vop_reclaim_args *ap)
68 struct hammer_inode *ip;
74 * Release the vnode association and ask that the inode be flushed.
76 if ((ip = vp->v_data) != NULL) {
79 hammer_rel_inode(ip, 1);
85 * Obtain a vnode for the specified inode number. An exclusively locked
89 hammer_vfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp)
91 struct hammer_mount *hmp = (void *)mp->mnt_data;
92 struct hammer_inode *ip;
96 * Get/allocate the hammer_inode structure. The structure must be
97 * unlocked while we manipulate the related vnode to avoid a
100 ip = hammer_get_inode(hmp, ino, hmp->asof, &error);
105 error = hammer_get_vnode(ip, LK_EXCLUSIVE, vpp);
106 hammer_rel_inode(ip, 0);
111 * Return a locked vnode for the specified inode. The inode must be
112 * referenced but NOT LOCKED on entry and will remain referenced on
116 hammer_get_vnode(struct hammer_inode *ip, int lktype, struct vnode **vpp)
122 if ((vp = ip->vp) == NULL) {
123 error = getnewvnode(VT_HAMMER, ip->hmp->mp, vpp, 0, 0);
126 hammer_lock_ex(&ip->lock);
127 if (ip->vp != NULL) {
128 hammer_unlock(&ip->lock);
133 hammer_ref(&ip->lock);
136 vp->v_type = hammer_get_vnode_type(
137 ip->ino_rec.base.base.obj_type);
139 switch(ip->ino_rec.base.base.obj_type) {
140 case HAMMER_OBJTYPE_CDEV:
141 case HAMMER_OBJTYPE_BDEV:
142 vp->v_ops = &ip->hmp->mp->mnt_vn_spec_ops;
143 addaliasu(vp, ip->ino_data.rmajor,
144 ip->ino_data.rminor);
146 case HAMMER_OBJTYPE_FIFO:
147 vp->v_ops = &ip->hmp->mp->mnt_vn_fifo_ops;
152 if (ip->obj_id == HAMMER_OBJID_ROOT)
155 vp->v_data = (void *)ip;
156 /* vnode locked by getnewvnode() */
157 /* make related vnode dirty if inode dirty? */
158 hammer_unlock(&ip->lock);
159 if (vp->v_type == VREG)
160 vinitvmio(vp, ip->ino_rec.ino_size);
165 * loop if the vget fails (aka races), or if the vp
166 * no longer matches ip->vp.
168 if (vget(vp, LK_EXCLUSIVE) == 0) {
179 * Acquire a HAMMER inode. The returned inode is not locked. These functions
180 * do not attach or detach the related vnode (use hammer_get_vnode() for
183 struct hammer_inode *
184 hammer_get_inode(struct hammer_mount *hmp, u_int64_t obj_id, hammer_tid_t asof,
187 struct hammer_inode_info iinfo;
188 struct hammer_cursor cursor;
189 struct hammer_inode *ip;
192 * Determine if we already have an inode cached. If we do then
195 iinfo.obj_id = obj_id;
196 iinfo.obj_asof = asof;
198 ip = hammer_ino_rb_tree_RB_LOOKUP_INFO(&hmp->rb_inos_root, &iinfo);
200 hammer_ref(&ip->lock);
205 ip = kmalloc(sizeof(*ip), M_HAMMER, M_WAITOK|M_ZERO);
206 ++hammer_count_inodes;
208 ip->obj_asof = iinfo.obj_asof;
210 RB_INIT(&ip->rec_tree);
213 * Locate the on-disk inode.
214 * If we do not have an inode cached search the HAMMER on-disk B-Tree
218 hammer_init_cursor_hmp(&cursor, hmp);
219 cursor.key_beg.obj_id = ip->obj_id;
220 cursor.key_beg.key = 0;
221 cursor.key_beg.create_tid = iinfo.obj_asof;
222 cursor.key_beg.delete_tid = 0;
223 cursor.key_beg.rec_type = HAMMER_RECTYPE_INODE;
224 cursor.key_beg.obj_type = 0;
225 cursor.flags = HAMMER_CURSOR_GET_RECORD | HAMMER_CURSOR_GET_DATA;
227 *errorp = hammer_btree_lookup(&cursor);
230 * On success the B-Tree lookup will hold the appropriate
231 * buffer cache buffers and provide a pointer to the requested
232 * information. Copy the information to the in-memory inode.
235 ip->ino_rec = cursor.record->inode;
236 ip->ino_data = cursor.data->inode;
237 } else if (cursor.node) {
238 hammer_cache_node(cursor.node, &ip->cache);
242 * On success load the inode's record and data and insert the
243 * inode into the B-Tree. It is possible to race another lookup
244 * insertion of the same inode so deal with that condition too.
246 * The cursor's locked node interlocks against others creating and
247 * destroying ip while we were blocked.
250 hammer_ref(&ip->lock);
251 if (RB_INSERT(hammer_ino_rb_tree, &hmp->rb_inos_root, ip)) {
252 hammer_uncache_node(&ip->cache);
253 hammer_unref(&ip->lock);
254 --hammer_count_inodes;
256 hammer_done_cursor(&cursor);
259 ip->flags |= HAMMER_INODE_ONDISK;
261 --hammer_count_inodes;
265 hammer_done_cursor(&cursor);
270 * Create a new filesystem object, returning the inode in *ipp. The
271 * returned inode will be referenced but not locked.
273 * The inode is created in-memory and will be delay-synchronized to the
277 hammer_create_inode(hammer_transaction_t trans, struct vattr *vap,
278 struct ucred *cred, hammer_inode_t dip,
279 struct hammer_inode **ipp)
286 ip = kmalloc(sizeof(*ip), M_HAMMER, M_WAITOK|M_ZERO);
287 ++hammer_count_inodes;
288 ip->obj_id = hammer_alloc_tid(trans);
289 KKASSERT(ip->obj_id != 0);
290 ip->obj_asof = hmp->asof;
292 ip->flags = HAMMER_INODE_DDIRTY | HAMMER_INODE_RDIRTY |
294 ip->last_tid = trans->tid;
296 RB_INIT(&ip->rec_tree);
298 ip->ino_rec.ino_atime = trans->tid;
299 ip->ino_rec.ino_mtime = trans->tid;
300 ip->ino_rec.ino_size = 0;
301 ip->ino_rec.ino_nlinks = 0;
303 ip->ino_rec.base.rec_id = hammer_alloc_recid(trans);
304 KKASSERT(ip->ino_rec.base.rec_id != 0);
305 ip->ino_rec.base.base.obj_id = ip->obj_id;
306 ip->ino_rec.base.base.key = 0;
307 ip->ino_rec.base.base.create_tid = trans->tid;
308 ip->ino_rec.base.base.delete_tid = 0;
309 ip->ino_rec.base.base.rec_type = HAMMER_RECTYPE_INODE;
310 ip->ino_rec.base.base.obj_type = hammer_get_obj_type(vap->va_type);
312 ip->ino_data.version = HAMMER_INODE_DATA_VERSION;
313 ip->ino_data.mode = vap->va_mode;
314 ip->ino_data.ctime = trans->tid;
315 ip->ino_data.parent_obj_id = (dip) ? dip->ino_rec.base.base.obj_id : 0;
317 switch(ip->ino_rec.base.base.obj_type) {
318 case HAMMER_OBJTYPE_CDEV:
319 case HAMMER_OBJTYPE_BDEV:
320 ip->ino_data.rmajor = vap->va_rmajor;
321 ip->ino_data.rminor = vap->va_rminor;
328 * Calculate default uid/gid and overwrite with information from
331 xuid = hammer_to_unix_xid(&dip->ino_data.uid);
332 ip->ino_data.gid = dip->ino_data.gid;
333 xuid = vop_helper_create_uid(hmp->mp, dip->ino_data.mode, xuid, cred,
335 ip->ino_data.mode = vap->va_mode;
337 if (vap->va_vaflags & VA_UID_UUID_VALID)
338 ip->ino_data.uid = vap->va_uid_uuid;
339 else if (vap->va_uid != (uid_t)VNOVAL)
340 hammer_guid_to_uuid(&ip->ino_data.uid, xuid);
341 if (vap->va_vaflags & VA_GID_UUID_VALID)
342 ip->ino_data.gid = vap->va_gid_uuid;
343 else if (vap->va_gid != (gid_t)VNOVAL)
344 hammer_guid_to_uuid(&ip->ino_data.gid, vap->va_gid);
346 hammer_ref(&ip->lock);
347 if (RB_INSERT(hammer_ino_rb_tree, &hmp->rb_inos_root, ip)) {
348 hammer_unref(&ip->lock);
349 panic("hammer_create_inode: duplicate obj_id %llx", ip->obj_id);
356 hammer_update_inode(hammer_inode_t ip)
358 struct hammer_cursor cursor;
359 struct hammer_cursor *spike = NULL;
360 hammer_record_t record;
364 * Locate the record on-disk and mark it as deleted. Both the B-Tree
365 * node and the record must be marked deleted. The record may or
366 * may not be physically deleted, depending on the retention policy.
368 * If the inode has already been deleted on-disk we have nothing
371 * XXX Update the inode record and data in-place if the retention
377 if ((ip->flags & (HAMMER_INODE_ONDISK|HAMMER_INODE_DELONDISK)) ==
378 HAMMER_INODE_ONDISK) {
379 hammer_init_cursor_ip(&cursor, ip);
380 cursor.key_beg.obj_id = ip->obj_id;
381 cursor.key_beg.key = 0;
382 cursor.key_beg.create_tid = ip->obj_asof;
383 cursor.key_beg.delete_tid = 0;
384 cursor.key_beg.rec_type = HAMMER_RECTYPE_INODE;
385 cursor.key_beg.obj_type = 0;
386 cursor.flags = HAMMER_CURSOR_GET_RECORD;
388 error = hammer_btree_lookup(&cursor);
391 error = hammer_ip_delete_record(&cursor, ip->last_tid);
393 ip->flags |= HAMMER_INODE_DELONDISK;
395 hammer_cache_node(cursor.node, &ip->cache);
396 hammer_done_cursor(&cursor);
400 * Write out a new record if the in-memory inode is not marked
401 * as having been deleted. Update our inode statistics if this
402 * is the first application of the inode on-disk.
404 * If the inode has been deleted permanently, HAMMER_INODE_DELONDISK
405 * will remain set and prevent further updates.
407 if (error == 0 && (ip->flags & HAMMER_INODE_DELETED) == 0) {
408 record = hammer_alloc_mem_record(ip);
409 record->rec.inode = ip->ino_rec;
410 record->rec.inode.base.base.create_tid = ip->last_tid;
411 record->rec.inode.base.data_len = sizeof(ip->ino_data);
412 record->data = (void *)&ip->ino_data;
413 error = hammer_ip_sync_record(record, &spike);
414 record->flags |= HAMMER_RECF_DELETED;
415 hammer_rel_mem_record(record);
416 if (error == ENOSPC) {
417 error = hammer_spike(&spike);
421 KKASSERT(spike == NULL);
423 ip->flags &= ~(HAMMER_INODE_RDIRTY|HAMMER_INODE_DDIRTY|
424 HAMMER_INODE_DELONDISK);
425 if ((ip->flags & HAMMER_INODE_ONDISK) == 0) {
426 hammer_modify_volume(ip->hmp->rootvol);
427 ++ip->hmp->rootvol->ondisk->vol0_stat_inodes;
428 hammer_modify_volume_done(ip->hmp->rootvol);
429 ip->flags |= HAMMER_INODE_ONDISK;
434 ip->flags &= ~HAMMER_INODE_TID;
439 * Release a reference on an inode and unload it if told to flush.
442 hammer_rel_inode(struct hammer_inode *ip, int flush)
444 hammer_unref(&ip->lock);
445 if (flush || ip->ino_rec.ino_nlinks == 0)
446 ip->flags |= HAMMER_INODE_FLUSH;
447 if (ip->lock.refs == 0 && (ip->flags & HAMMER_INODE_FLUSH))
448 hammer_unload_inode(ip, NULL);
452 * Unload and destroy the specified inode.
454 * (called via RB_SCAN)
457 hammer_unload_inode(struct hammer_inode *ip, void *data __unused)
461 KASSERT(ip->lock.refs == 0,
462 ("hammer_unload_inode: %d refs\n", ip->lock.refs));
463 KKASSERT(ip->vp == NULL);
464 hammer_ref(&ip->lock);
466 error = hammer_sync_inode(ip, MNT_WAIT, 1);
468 kprintf("hammer_sync_inode failed error %d\n", error);
469 KKASSERT(RB_EMPTY(&ip->rec_tree));
470 RB_REMOVE(hammer_ino_rb_tree, &ip->hmp->rb_inos_root, ip);
472 hammer_uncache_node(&ip->cache);
473 --hammer_count_inodes;
479 * A transaction has modified an inode, requiring a new record and possibly
480 * also data to be written out.
482 * last_tid is the TID to use for the disk sync.
485 hammer_modify_inode(struct hammer_transaction *trans,
486 struct hammer_inode *ip, int flags)
488 if ((flags & HAMMER_INODE_TID) && (ip->flags & HAMMER_INODE_TID) == 0) {
489 ip->last_tid = trans->tid;
495 * Sync any dirty buffers and records associated with an inode. The
496 * inode's last_tid field is used as the transaction id for the sync,
497 * overriding any intermediate TIDs that were used for records. Note
498 * that the dirty buffer cache buffers do not have any knowledge of
499 * the transaction id they were modified under.
501 * If we can't sync due to a cluster becoming full the spike structure
502 * will be filled in and ENOSPC returned. We must return -ENOSPC to
503 * terminate the RB_SCAN.
506 hammer_sync_inode_callback(hammer_record_t rec, void *data)
508 struct hammer_cursor **spike = data;
511 hammer_ref(&rec->lock);
512 error = hammer_ip_sync_record(rec, spike);
513 hammer_rel_mem_record(rec);
517 if (error != -ENOSPC) {
518 kprintf("hammer_sync_inode_callback: sync failed rec "
519 "%p, error %d\n", rec, error);
529 hammer_sync_inode(hammer_inode_t ip, int waitfor, int handle_delete)
531 struct hammer_transaction trans;
532 struct hammer_cursor *spike = NULL;
535 hammer_lock_ex(&ip->lock);
536 hammer_start_transaction(&trans, ip->hmp);
539 * If the inode has been deleted (nlinks == 0), and the OS no longer
540 * has any references to it (handle_delete != 0), clean up in-memory
543 * NOTE: We do not set the RDIRTY flag when updating the delete_tid,
544 * setting HAMMER_INODE_DELETED takes care of it.
546 * NOTE: Because we may sync records within this new transaction,
547 * force the inode update later on to use our transaction id or
548 * the delete_tid of the inode may be less then the create_tid of
549 * the inode update. XXX shouldn't happen but don't take the chance.
551 * NOTE: The call to hammer_ip_delete_range() cannot return ENOSPC
552 * so we can pass a NULL spike structure, because no partial data
553 * deletion can occur (yet).
555 if (ip->ino_rec.ino_nlinks == 0 && handle_delete) {
557 vtruncbuf(ip->vp, 0, HAMMER_BUFSIZE);
558 error = hammer_ip_delete_range_all(&trans, ip);
559 KKASSERT(RB_EMPTY(&ip->rec_tree));
560 ip->flags &= ~HAMMER_INODE_TID;
561 ip->ino_rec.base.base.delete_tid = trans.tid;
562 hammer_modify_inode(&trans, ip,
563 HAMMER_INODE_DELETED | HAMMER_INODE_TID);
564 hammer_modify_volume(ip->hmp->rootvol);
565 --ip->hmp->rootvol->ondisk->vol0_stat_inodes;
566 hammer_modify_volume_done(ip->hmp->rootvol);
570 * Sync the buffer cache
573 error = vfsync(ip->vp, waitfor, 1, NULL, NULL);
578 * Now sync related records
581 error = RB_SCAN(hammer_rec_rb_tree, &ip->rec_tree, NULL,
582 hammer_sync_inode_callback, &spike);
583 KKASSERT(error <= 0);
586 if (error == ENOSPC) {
587 error = hammer_spike(&spike);
595 * Now update the inode's on-disk inode-data and/or on-disk record.
597 switch(ip->flags & (HAMMER_INODE_DELETED|HAMMER_INODE_ONDISK)) {
598 case HAMMER_INODE_DELETED|HAMMER_INODE_ONDISK:
600 * If deleted and on-disk, don't set any additional flags.
601 * the delete flag takes care of things.
604 case HAMMER_INODE_DELETED:
606 * Take care of the case where a deleted inode was never
607 * flushed to the disk in the first place.
609 ip->flags &= ~(HAMMER_INODE_RDIRTY|HAMMER_INODE_DDIRTY);
610 while (RB_ROOT(&ip->rec_tree)) {
611 hammer_record_t rec = RB_ROOT(&ip->rec_tree);
612 hammer_ref(&rec->lock);
613 rec->flags |= HAMMER_RECF_DELETED;
614 hammer_rel_mem_record(rec);
617 case HAMMER_INODE_ONDISK:
619 * If already on-disk, do not set any additional flags.
624 * If not on-disk and not deleted, set both dirty flags
625 * to force an initial record to be written.
627 ip->flags |= HAMMER_INODE_RDIRTY | HAMMER_INODE_DDIRTY;
632 * If RDIRTY or DDIRTY is set, write out a new record. If the
633 * inode is already on-disk, the old record is marked as deleted.
635 if (ip->flags & (HAMMER_INODE_RDIRTY | HAMMER_INODE_DDIRTY |
636 HAMMER_INODE_DELETED)) {
637 error = hammer_update_inode(ip);
639 hammer_commit_transaction(&trans);
640 hammer_unlock(&ip->lock);
645 * Access the filesystem buffer containing the cluster-relative byte
646 * offset, validate the buffer type, load *bufferp and return a
647 * pointer to the requested data. The buffer is reference and locked on
650 * If buf_type is 0 the buffer is assumed to be a pure-data buffer and
651 * no type or crc check is performed.
653 * If *bufferp is not NULL on entry it is assumed to contain a locked
654 * and referenced buffer which will then be replaced.
656 * If the caller is holding another unrelated buffer locked it must be
657 * passed in reorderbuf so we can properly order buffer locks.
659 * XXX add a flag for the buffer type and check the CRC here XXX
662 hammer_bread(hammer_cluster_t cluster, int32_t cloff,
663 u_int64_t buf_type, int *errorp,
664 struct hammer_buffer **bufferp)
666 hammer_buffer_t buffer;
671 * Load the correct filesystem buffer, replacing *bufferp.
673 buf_no = cloff / HAMMER_BUFSIZE;
675 if (buffer == NULL || buffer->cluster != cluster ||
676 buffer->buf_no != buf_no) {
678 /*hammer_unlock(&buffer->io.lock);*/
679 hammer_rel_buffer(buffer, 0);
681 buffer = hammer_get_buffer(cluster, buf_no, 0, errorp);
685 /*hammer_lock_ex(&buffer->io.lock);*/
689 * Validate the buffer type
691 buf_off = cloff & HAMMER_BUFMASK;
693 if (buf_type != buffer->ondisk->head.buf_type) {
694 kprintf("BUFFER HEAD TYPE MISMATCH %llx %llx\n",
695 buf_type, buffer->ondisk->head.buf_type);
699 if (buf_off < sizeof(buffer->ondisk->head)) {
700 kprintf("BUFFER OFFSET TOO LOW %d\n", buf_off);
707 * Return a pointer to the buffer data.
710 return((char *)buffer->ondisk + buf_off);