2 * Copyright (c) 2011-2018 The DragonFly Project. All rights reserved.
4 * This code is derived from software contributed to The DragonFly Project
5 * by Matthew Dillon <dillon@dragonflybsd.org>
6 * by Venkatesh Srinivas <vsrinivas@dragonflybsd.org>
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in
16 * the documentation and/or other materials provided with the
18 * 3. Neither the name of The DragonFly Project nor the names of its
19 * contributors may be used to endorse or promote products derived
20 * from this software without specific, prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
23 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
24 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
25 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
26 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
27 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
28 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
29 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
30 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
31 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
32 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35 #include <sys/cdefs.h>
36 #include <sys/param.h>
37 #include <sys/systm.h>
38 #include <sys/types.h>
46 RB_GENERATE2(hammer2_inode_tree, hammer2_inode, rbnode, hammer2_inode_cmp,
47 hammer2_tid_t, meta.inum);
50 hammer2_inode_cmp(hammer2_inode_t *ip1, hammer2_inode_t *ip2)
52 if (ip1->meta.inum < ip2->meta.inum)
54 if (ip1->meta.inum > ip2->meta.inum)
61 hammer2_knote(struct vnode *vp, int flags)
64 KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, flags);
69 hammer2_inode_delayed_sideq(hammer2_inode_t *ip)
71 hammer2_inode_sideq_t *ipul;
72 hammer2_pfs_t *pmp = ip->pmp;
74 if ((ip->flags & HAMMER2_INODE_ONSIDEQ) == 0) {
75 ipul = kmalloc(sizeof(*ipul), pmp->minode,
78 hammer2_spin_ex(&pmp->list_spin);
79 if ((ip->flags & HAMMER2_INODE_ONSIDEQ) == 0) {
80 hammer2_inode_ref(ip);
81 atomic_set_int(&ip->flags,
82 HAMMER2_INODE_ONSIDEQ);
83 TAILQ_INSERT_TAIL(&pmp->sideq, ipul, entry);
85 hammer2_spin_unex(&pmp->list_spin);
87 hammer2_spin_unex(&pmp->list_spin);
88 kfree(ipul, pmp->minode);
96 * HAMMER2 offers shared and exclusive locks on inodes. Pass a mask of
99 * - pass HAMMER2_RESOLVE_SHARED if a shared lock is desired. The
100 * inode locking function will automatically set the RDONLY flag.
102 * - pass HAMMER2_RESOLVE_ALWAYS if you need the inode's meta-data.
103 * Most front-end inode locks do.
105 * - pass HAMMER2_RESOLVE_NEVER if you do not want to require that
106 * the inode data be resolved. This is used by the syncthr because
107 * it can run on an unresolved/out-of-sync cluster, and also by the
108 * vnode reclamation code to avoid unnecessary I/O (particularly when
109 * disposing of hundreds of thousands of cached vnodes).
111 * The inode locking function locks the inode itself, resolves any stale
112 * chains in the inode's cluster, and allocates a fresh copy of the
113 * cluster with 1 ref and all the underlying chains locked.
115 * ip->cluster will be stable while the inode is locked.
117 * NOTE: We don't combine the inode/chain lock because putting away an
118 * inode would otherwise confuse multiple lock holders of the inode.
120 * NOTE: In-memory inodes always point to hardlink targets (the actual file),
121 * and never point to a hardlink pointer.
123 * NOTE: If caller passes HAMMER2_RESOLVE_RDONLY the exclusive locking code
124 * will feel free to reduce the chain set in the cluster as an
125 * optimization. It will still be validated against the quorum if
126 * appropriate, but the optimization might be able to reduce data
127 * accesses to one node. This flag is automatically set if the inode
128 * is locked with HAMMER2_RESOLVE_SHARED.
131 hammer2_inode_lock(hammer2_inode_t *ip, int how)
133 hammer2_inode_ref(ip);
136 * Inode structure mutex
138 if (how & HAMMER2_RESOLVE_SHARED) {
139 /*how |= HAMMER2_RESOLVE_RDONLY; not used */
140 hammer2_mtx_sh(&ip->lock);
142 hammer2_mtx_ex(&ip->lock);
147 * Select a chain out of an inode's cluster and lock it.
149 * The inode does not have to be locked.
152 hammer2_inode_chain(hammer2_inode_t *ip, int clindex, int how)
154 hammer2_chain_t *chain;
155 hammer2_cluster_t *cluster;
157 hammer2_spin_sh(&ip->cluster_spin);
158 cluster = &ip->cluster;
159 if (clindex >= cluster->nchains)
162 chain = cluster->array[clindex].chain;
164 hammer2_chain_ref(chain);
165 hammer2_spin_unsh(&ip->cluster_spin);
166 hammer2_chain_lock(chain, how);
168 hammer2_spin_unsh(&ip->cluster_spin);
174 hammer2_inode_chain_and_parent(hammer2_inode_t *ip, int clindex,
175 hammer2_chain_t **parentp, int how)
177 hammer2_chain_t *chain;
178 hammer2_chain_t *parent;
181 hammer2_spin_sh(&ip->cluster_spin);
182 if (clindex >= ip->cluster.nchains)
185 chain = ip->cluster.array[clindex].chain;
187 hammer2_chain_ref(chain);
188 hammer2_spin_unsh(&ip->cluster_spin);
189 hammer2_chain_lock(chain, how);
191 hammer2_spin_unsh(&ip->cluster_spin);
195 * Get parent, lock order must be (parent, chain).
197 parent = chain->parent;
199 hammer2_chain_ref(parent);
200 hammer2_chain_unlock(chain);
201 hammer2_chain_lock(parent, how);
202 hammer2_chain_lock(chain, how);
204 if (ip->cluster.array[clindex].chain == chain &&
205 chain->parent == parent) {
212 hammer2_chain_unlock(chain);
213 hammer2_chain_drop(chain);
215 hammer2_chain_unlock(parent);
216 hammer2_chain_drop(parent);
225 hammer2_inode_unlock(hammer2_inode_t *ip)
227 hammer2_mtx_unlock(&ip->lock);
228 hammer2_inode_drop(ip);
232 * Temporarily release a lock held shared or exclusive. Caller must
233 * hold the lock shared or exclusive on call and lock will be released
236 * Restore a lock that was temporarily released.
239 hammer2_inode_lock_temp_release(hammer2_inode_t *ip)
241 return hammer2_mtx_temp_release(&ip->lock);
245 hammer2_inode_lock_temp_restore(hammer2_inode_t *ip, hammer2_mtx_state_t ostate)
247 hammer2_mtx_temp_restore(&ip->lock, ostate);
251 * Upgrade a shared inode lock to exclusive and return. If the inode lock
252 * is already held exclusively this is a NOP.
254 * The caller MUST hold the inode lock either shared or exclusive on call
255 * and will own the lock exclusively on return.
257 * Returns non-zero if the lock was already exclusive prior to the upgrade.
260 hammer2_inode_lock_upgrade(hammer2_inode_t *ip)
264 if (mtx_islocked_ex(&ip->lock)) {
267 hammer2_mtx_unlock(&ip->lock);
268 hammer2_mtx_ex(&ip->lock);
275 * Downgrade an inode lock from exclusive to shared only if the inode
276 * lock was previously shared. If the inode lock was previously exclusive,
280 hammer2_inode_lock_downgrade(hammer2_inode_t *ip, int wasexclusive)
282 if (wasexclusive == 0)
283 mtx_downgrade(&ip->lock);
287 * Lookup an inode by inode number
290 hammer2_inode_lookup(hammer2_pfs_t *pmp, hammer2_tid_t inum)
298 hammer2_spin_ex(&pmp->inum_spin);
299 ip = RB_LOOKUP(hammer2_inode_tree, &pmp->inum_tree, inum);
301 hammer2_inode_ref(ip);
302 hammer2_spin_unex(&pmp->inum_spin);
308 * Adding a ref to an inode is only legal if the inode already has at least
311 * (can be called with spinlock held)
314 hammer2_inode_ref(hammer2_inode_t *ip)
316 atomic_add_int(&ip->refs, 1);
317 if (hammer2_debug & 0x80000) {
318 kprintf("INODE+1 %p (%d->%d)\n", ip, ip->refs - 1, ip->refs);
324 * Drop an inode reference, freeing the inode when the last reference goes
328 hammer2_inode_drop(hammer2_inode_t *ip)
334 if (hammer2_debug & 0x80000) {
335 kprintf("INODE-1 %p (%d->%d)\n",
336 ip, ip->refs, ip->refs - 1);
343 * Transition to zero, must interlock with
344 * the inode inumber lookup tree (if applicable).
345 * It should not be possible for anyone to race
346 * the transition to 0.
350 hammer2_spin_ex(&pmp->inum_spin);
352 if (atomic_cmpset_int(&ip->refs, 1, 0)) {
353 KKASSERT(hammer2_mtx_refs(&ip->lock) == 0);
354 if (ip->flags & HAMMER2_INODE_ONRBTREE) {
355 atomic_clear_int(&ip->flags,
356 HAMMER2_INODE_ONRBTREE);
357 RB_REMOVE(hammer2_inode_tree,
358 &pmp->inum_tree, ip);
361 hammer2_spin_unex(&pmp->inum_spin);
366 * Cleaning out ip->cluster isn't entirely
369 hammer2_inode_repoint(ip, NULL, NULL);
371 kfree(ip, pmp->minode);
372 atomic_add_long(&pmp->inmem_inodes, -1);
373 ip = NULL; /* will terminate loop */
375 hammer2_spin_unex(&ip->pmp->inum_spin);
379 * Non zero transition
381 if (atomic_cmpset_int(&ip->refs, refs, refs - 1))
388 * Get the vnode associated with the given inode, allocating the vnode if
389 * necessary. The vnode will be returned exclusively locked.
391 * *errorp is set to a UNIX error, not a HAMMER2 error.
393 * The caller must lock the inode (shared or exclusive).
395 * Great care must be taken to avoid deadlocks and vnode acquisition/reclaim
399 hammer2_igetv(hammer2_inode_t *ip, int *errorp)
405 KKASSERT(pmp != NULL);
410 * Attempt to reuse an existing vnode assignment. It is
411 * possible to race a reclaim so the vget() may fail. The
412 * inode must be unlocked during the vget() to avoid a
413 * deadlock against a reclaim.
420 * Inode must be unlocked during the vget() to avoid
421 * possible deadlocks, but leave the ip ref intact.
423 * vnode is held to prevent destruction during the
424 * vget(). The vget() can still fail if we lost
425 * a reclaim race on the vnode.
427 hammer2_mtx_state_t ostate;
430 ostate = hammer2_inode_lock_temp_release(ip);
431 if (vget(vp, LK_EXCLUSIVE)) {
433 hammer2_inode_lock_temp_restore(ip, ostate);
436 hammer2_inode_lock_temp_restore(ip, ostate);
438 /* vp still locked and ref from vget */
440 kprintf("hammer2: igetv race %p/%p\n",
450 * No vnode exists, allocate a new vnode. Beware of
451 * allocation races. This function will return an
452 * exclusively locked and referenced vnode.
454 *errorp = getnewvnode(VT_HAMMER2, pmp->mp, &vp, 0, 0);
456 kprintf("hammer2: igetv getnewvnode failed %d\n",
463 * Lock the inode and check for an allocation race.
465 wasexclusive = hammer2_inode_lock_upgrade(ip);
466 if (ip->vp != NULL) {
469 hammer2_inode_lock_downgrade(ip, wasexclusive);
473 switch (ip->meta.type) {
474 case HAMMER2_OBJTYPE_DIRECTORY:
477 case HAMMER2_OBJTYPE_REGFILE:
479 * Regular file must use buffer cache I/O
480 * (VKVABIO cpu sync semantics supported)
483 vsetflags(vp, VKVABIO);
484 vinitvmio(vp, ip->meta.size,
486 (int)ip->meta.size & HAMMER2_LBUFMASK);
488 case HAMMER2_OBJTYPE_SOFTLINK:
490 * XXX for now we are using the generic file_read
491 * and file_write code so we need a buffer cache
494 * (VKVABIO cpu sync semantics supported)
497 vsetflags(vp, VKVABIO);
498 vinitvmio(vp, ip->meta.size,
500 (int)ip->meta.size & HAMMER2_LBUFMASK);
502 case HAMMER2_OBJTYPE_CDEV:
505 case HAMMER2_OBJTYPE_BDEV:
506 vp->v_ops = &pmp->mp->mnt_vn_spec_ops;
507 if (ip->meta.type != HAMMER2_OBJTYPE_CDEV)
513 case HAMMER2_OBJTYPE_FIFO:
515 vp->v_ops = &pmp->mp->mnt_vn_fifo_ops;
517 case HAMMER2_OBJTYPE_SOCKET:
521 panic("hammer2: unhandled objtype %d",
526 if (ip == pmp->iroot)
527 vsetflags(vp, VROOT);
531 hammer2_inode_ref(ip); /* vp association */
532 hammer2_inode_lock_downgrade(ip, wasexclusive);
537 * Return non-NULL vp and *errorp == 0, or NULL vp and *errorp != 0.
539 if (hammer2_debug & 0x0002) {
540 kprintf("igetv vp %p refs 0x%08x aux 0x%08x\n",
541 vp, vp->v_refcnt, vp->v_auxrefs);
547 * Returns the inode associated with the passed-in cluster, creating the
548 * inode if necessary and synchronizing it to the passed-in cluster otherwise.
549 * When synchronizing, if idx >= 0, only cluster index (idx) is synchronized.
550 * Otherwise the whole cluster is synchronized.
552 * The passed-in cluster must be locked and will remain locked on return.
553 * The returned inode will be locked and the caller may dispose of both
554 * via hammer2_inode_unlock() + hammer2_inode_drop(). However, if the caller
555 * needs to resolve a hardlink it must ref/unlock/relock/drop the inode.
557 * The hammer2_inode structure regulates the interface between the high level
558 * kernel VNOPS API and the filesystem backend (the chains).
560 * On return the inode is locked with the supplied cluster.
563 hammer2_inode_get(hammer2_pfs_t *pmp, hammer2_inode_t *dip,
564 hammer2_cluster_t *cluster, int idx)
566 hammer2_inode_t *nip;
567 const hammer2_inode_data_t *iptmp;
568 const hammer2_inode_data_t *nipdata;
570 KKASSERT(cluster == NULL ||
571 hammer2_cluster_type(cluster) == HAMMER2_BREF_TYPE_INODE);
575 * Interlocked lookup/ref of the inode. This code is only needed
576 * when looking up inodes with nlinks != 0 (TODO: optimize out
577 * otherwise and test for duplicates).
579 * Cluster can be NULL during the initial pfs allocation.
583 iptmp = &hammer2_cluster_rdata(cluster)->ipdata;
584 nip = hammer2_inode_lookup(pmp, iptmp->meta.inum);
588 hammer2_mtx_ex(&nip->lock);
591 * Handle SMP race (not applicable to the super-root spmp
592 * which can't index inodes due to duplicative inode numbers).
594 if (pmp->spmp_hmp == NULL &&
595 (nip->flags & HAMMER2_INODE_ONRBTREE) == 0) {
596 hammer2_mtx_unlock(&nip->lock);
597 hammer2_inode_drop(nip);
601 hammer2_inode_repoint_one(nip, cluster, idx);
603 hammer2_inode_repoint(nip, NULL, cluster);
609 * We couldn't find the inode number, create a new inode.
611 nip = kmalloc(sizeof(*nip), pmp->minode, M_WAITOK | M_ZERO);
612 spin_init(&nip->cluster_spin, "h2clspin");
613 atomic_add_long(&pmp->inmem_inodes, 1);
614 hammer2_pfs_memory_inc(pmp);
615 hammer2_pfs_memory_wakeup(pmp);
617 nip->flags = HAMMER2_INODE_SROOT;
620 * Initialize nip's cluster. A cluster is provided for normal
621 * inodes but typically not for the super-root or PFS inodes.
623 nip->cluster.refs = 1;
624 nip->cluster.pmp = pmp;
625 nip->cluster.flags |= HAMMER2_CLUSTER_INODE;
627 nipdata = &hammer2_cluster_rdata(cluster)->ipdata;
628 nip->meta = nipdata->meta;
629 atomic_set_int(&nip->flags, HAMMER2_INODE_METAGOOD);
630 hammer2_inode_repoint(nip, NULL, cluster);
632 nip->meta.inum = 1; /* PFS inum is always 1 XXX */
633 /* mtime will be updated when a cluster is available */
634 atomic_set_int(&nip->flags, HAMMER2_INODE_METAGOOD);/*XXX*/
640 * ref and lock on nip gives it state compatible to after a
641 * hammer2_inode_lock() call.
644 hammer2_mtx_init(&nip->lock, "h2inode");
645 hammer2_mtx_ex(&nip->lock);
646 /* combination of thread lock and chain lock == inode lock */
649 * Attempt to add the inode. If it fails we raced another inode
650 * get. Undo all the work and try again.
652 if (pmp->spmp_hmp == NULL) {
653 hammer2_spin_ex(&pmp->inum_spin);
654 if (RB_INSERT(hammer2_inode_tree, &pmp->inum_tree, nip)) {
655 hammer2_spin_unex(&pmp->inum_spin);
656 hammer2_mtx_unlock(&nip->lock);
657 hammer2_inode_drop(nip);
660 atomic_set_int(&nip->flags, HAMMER2_INODE_ONRBTREE);
662 hammer2_spin_unex(&pmp->inum_spin);
671 * Create a new inode using the vattr to figure out the type. A non-zero
672 * type field overrides vattr. We need the directory to set iparent or to
673 * use when the inode is directly embedded in a directory (typically super-root
674 * entries), but note that this really only applies OBJTYPE_DIRECTORY as
675 * non-directory inodes can be hardlinked.
677 * If no error occurs the new inode is returned, otherwise NULL is returned.
678 * It is possible for an error to create a junk inode and then fail later.
679 * It will attempt to delete the junk inode and return NULL in this situation.
681 * If vap and/or cred are NULL the related fields are not set and the
682 * inode type defaults to a directory. This is used when creating PFSs
683 * under the super-root, so the inode number is set to 1 in this case.
685 * dip is not locked on entry.
687 * NOTE: This function is used to create all manners of inodes, including
688 * super-root entries for snapshots and PFSs. When used to create a
689 * snapshot the inode will be temporarily associated with the spmp.
691 * NOTE: When creating a normal file or directory the name/name_len/lhc
692 * is optional, but is typically specified to make debugging and
696 hammer2_inode_create(hammer2_inode_t *dip, hammer2_inode_t *pip,
697 struct vattr *vap, struct ucred *cred,
698 const uint8_t *name, size_t name_len, hammer2_key_t lhc,
700 uint8_t type, uint8_t target_type,
701 int flags, int *errorp)
703 hammer2_xop_create_t *xop;
704 hammer2_inode_t *nip;
710 uint8_t pip_comp_algo;
711 uint8_t pip_check_algo;
712 hammer2_tid_t pip_inum;
715 lhc = hammer2_dirhash(name, name_len);
720 * Locate the inode or indirect block to create the new
721 * entry in. At the same time check for key collisions
722 * and iterate until we don't get one.
724 * Lock the directory exclusively for now to guarantee that
725 * we can find an unused lhc for the name. Due to collisions,
726 * two different creates can end up with the same lhc so we
727 * cannot depend on the OS to prevent the collision.
729 hammer2_inode_lock(dip, 0);
731 pip_uid = pip->meta.uid;
732 pip_gid = pip->meta.gid;
733 pip_mode = pip->meta.mode;
734 pip_comp_algo = pip->meta.comp_algo;
735 pip_check_algo = pip->meta.check_algo;
736 pip_inum = (pip == pip->pmp->iroot) ? 1 : pip->meta.inum;
739 * If name specified, locate an unused key in the collision space.
740 * Otherwise use the passed-in lhc directly.
743 hammer2_xop_scanlhc_t *sxop;
744 hammer2_key_t lhcbase;
747 sxop = hammer2_xop_alloc(dip, HAMMER2_XOP_MODIFYING);
749 hammer2_xop_start(&sxop->head, hammer2_xop_scanlhc);
750 while ((error = hammer2_xop_collect(&sxop->head, 0)) == 0) {
751 if (lhc != sxop->head.cluster.focus->bref.key)
755 hammer2_xop_retire(&sxop->head, HAMMER2_XOPMASK_VOP);
758 if (error != HAMMER2_ERROR_ENOENT)
763 if ((lhcbase ^ lhc) & ~HAMMER2_DIRHASH_LOMASK) {
764 error = HAMMER2_ERROR_ENOSPC;
770 * Create the inode with the lhc as the key.
772 xop = hammer2_xop_alloc(dip, HAMMER2_XOP_MODIFYING);
775 bzero(&xop->meta, sizeof(xop->meta));
778 xop->meta.type = hammer2_get_obj_type(vap->va_type);
780 switch (xop->meta.type) {
781 case HAMMER2_OBJTYPE_CDEV:
782 case HAMMER2_OBJTYPE_BDEV:
783 xop->meta.rmajor = vap->va_rmajor;
784 xop->meta.rminor = vap->va_rminor;
789 type = xop->meta.type;
791 xop->meta.type = type;
792 xop->meta.target_type = target_type;
794 xop->meta.inum = inum;
795 xop->meta.iparent = pip_inum;
797 /* Inherit parent's inode compression mode. */
798 xop->meta.comp_algo = pip_comp_algo;
799 xop->meta.check_algo = pip_check_algo;
800 xop->meta.version = HAMMER2_INODE_VERSION_ONE;
801 hammer2_update_time(&xop->meta.ctime);
802 xop->meta.mtime = xop->meta.ctime;
804 xop->meta.mode = vap->va_mode;
805 xop->meta.nlinks = 1;
808 xuid = hammer2_to_unix_xid(&pip_uid);
809 xuid = vop_helper_create_uid(dip->pmp->mp,
815 /* super-root has no dip and/or pmp */
818 if (vap->va_vaflags & VA_UID_UUID_VALID)
819 xop->meta.uid = vap->va_uid_uuid;
820 else if (vap->va_uid != (uid_t)VNOVAL)
821 hammer2_guid_to_uuid(&xop->meta.uid, vap->va_uid);
823 hammer2_guid_to_uuid(&xop->meta.uid, xuid);
825 if (vap->va_vaflags & VA_GID_UUID_VALID)
826 xop->meta.gid = vap->va_gid_uuid;
827 else if (vap->va_gid != (gid_t)VNOVAL)
828 hammer2_guid_to_uuid(&xop->meta.gid, vap->va_gid);
830 xop->meta.gid = pip_gid;
834 * Regular files and softlinks allow a small amount of data to be
835 * directly embedded in the inode. This flag will be cleared if
836 * the size is extended past the embedded limit.
838 if (xop->meta.type == HAMMER2_OBJTYPE_REGFILE ||
839 xop->meta.type == HAMMER2_OBJTYPE_SOFTLINK) {
840 xop->meta.op_flags |= HAMMER2_OPFLAG_DIRECTDATA;
843 hammer2_xop_setname(&xop->head, name, name_len);
845 name_len = hammer2_xop_setname_inum(&xop->head, inum);
846 KKASSERT(lhc == inum);
848 xop->meta.name_len = name_len;
849 xop->meta.name_key = lhc;
850 KKASSERT(name_len < HAMMER2_INODE_MAXNAME);
852 hammer2_xop_start(&xop->head, hammer2_inode_xop_create);
854 error = hammer2_xop_collect(&xop->head, 0);
856 kprintf("CREATE INODE %*.*s\n",
857 (int)name_len, (int)name_len, name);
866 * Set up the new inode if not a hardlink pointer.
868 * NOTE: *_get() integrates chain's lock into the inode lock.
870 * NOTE: Only one new inode can currently be created per
871 * transaction. If the need arises we can adjust
872 * hammer2_trans_init() to allow more.
874 * NOTE: nipdata will have chain's blockset data.
876 nip = hammer2_inode_get(dip->pmp, dip, &xop->head.cluster, -1);
877 nip->comp_heuristic = 0;
879 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
881 hammer2_inode_unlock(dip);
887 * Create a directory entry under dip with the specified name, inode number,
888 * and OBJTYPE (type).
890 * This returns a UNIX errno code, not a HAMMER2_ERROR_* code.
893 hammer2_dirent_create(hammer2_inode_t *dip, const char *name, size_t name_len,
894 hammer2_key_t inum, uint8_t type)
896 hammer2_xop_mkdirent_t *xop;
903 KKASSERT(name != NULL);
904 lhc = hammer2_dirhash(name, name_len);
907 * Locate the inode or indirect block to create the new
908 * entry in. At the same time check for key collisions
909 * and iterate until we don't get one.
911 * Lock the directory exclusively for now to guarantee that
912 * we can find an unused lhc for the name. Due to collisions,
913 * two different creates can end up with the same lhc so we
914 * cannot depend on the OS to prevent the collision.
916 hammer2_inode_lock(dip, 0);
919 * If name specified, locate an unused key in the collision space.
920 * Otherwise use the passed-in lhc directly.
923 hammer2_xop_scanlhc_t *sxop;
924 hammer2_key_t lhcbase;
927 sxop = hammer2_xop_alloc(dip, HAMMER2_XOP_MODIFYING);
929 hammer2_xop_start(&sxop->head, hammer2_xop_scanlhc);
930 while ((error = hammer2_xop_collect(&sxop->head, 0)) == 0) {
931 if (lhc != sxop->head.cluster.focus->bref.key)
935 hammer2_xop_retire(&sxop->head, HAMMER2_XOPMASK_VOP);
938 if (error != HAMMER2_ERROR_ENOENT)
943 if ((lhcbase ^ lhc) & ~HAMMER2_DIRHASH_LOMASK) {
944 error = HAMMER2_ERROR_ENOSPC;
950 * Create the directory entry with the lhc as the key.
952 xop = hammer2_xop_alloc(dip, HAMMER2_XOP_MODIFYING);
954 bzero(&xop->dirent, sizeof(xop->dirent));
955 xop->dirent.inum = inum;
956 xop->dirent.type = type;
957 xop->dirent.namlen = name_len;
959 KKASSERT(name_len < HAMMER2_INODE_MAXNAME);
960 hammer2_xop_setname(&xop->head, name, name_len);
962 hammer2_xop_start(&xop->head, hammer2_inode_xop_mkdirent);
964 error = hammer2_xop_collect(&xop->head, 0);
966 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
968 error = hammer2_error_to_errno(error);
969 hammer2_inode_unlock(dip);
975 * Repoint ip->cluster's chains to cluster's chains and fixup the default
976 * focus. All items, valid or invalid, are repointed. hammer2_xop_start()
977 * filters out invalid or non-matching elements.
979 * Caller must hold the inode and cluster exclusive locked, if not NULL,
980 * must also be locked.
982 * Cluster may be NULL to clean out any chains in ip->cluster.
985 hammer2_inode_repoint(hammer2_inode_t *ip, hammer2_inode_t *pip,
986 hammer2_cluster_t *cluster)
988 hammer2_chain_t *dropch[HAMMER2_MAXCLUSTER];
989 hammer2_chain_t *ochain;
990 hammer2_chain_t *nchain;
993 bzero(dropch, sizeof(dropch));
996 * Replace chains in ip->cluster with chains from cluster and
997 * adjust the focus if necessary.
999 * NOTE: nchain and/or ochain can be NULL due to gaps
1000 * in the cluster arrays.
1002 hammer2_spin_ex(&ip->cluster_spin);
1003 for (i = 0; cluster && i < cluster->nchains; ++i) {
1005 * Do not replace elements which are the same. Also handle
1006 * element count discrepancies.
1008 nchain = cluster->array[i].chain;
1009 if (i < ip->cluster.nchains) {
1010 ochain = ip->cluster.array[i].chain;
1011 if (ochain == nchain)
1020 ip->cluster.array[i].chain = nchain;
1021 ip->cluster.array[i].flags &= ~HAMMER2_CITEM_INVALID;
1022 ip->cluster.array[i].flags |= cluster->array[i].flags &
1023 HAMMER2_CITEM_INVALID;
1025 hammer2_chain_ref(nchain);
1030 * Release any left-over chains in ip->cluster.
1032 while (i < ip->cluster.nchains) {
1033 nchain = ip->cluster.array[i].chain;
1035 ip->cluster.array[i].chain = NULL;
1036 ip->cluster.array[i].flags |= HAMMER2_CITEM_INVALID;
1043 * Fixup fields. Note that the inode-embedded cluster is never
1047 ip->cluster.nchains = cluster->nchains;
1048 ip->cluster.focus = cluster->focus;
1049 ip->cluster.flags = cluster->flags & ~HAMMER2_CLUSTER_LOCKED;
1051 ip->cluster.nchains = 0;
1052 ip->cluster.focus = NULL;
1053 ip->cluster.flags &= ~HAMMER2_CLUSTER_ZFLAGS;
1056 hammer2_spin_unex(&ip->cluster_spin);
1059 * Cleanup outside of spinlock
1063 hammer2_chain_drop(dropch[i]);
1068 * Repoint a single element from the cluster to the ip. Used by the
1069 * synchronization threads to piecemeal update inodes. Does not change
1070 * focus and requires inode to be re-locked to clean-up flags (XXX).
1073 hammer2_inode_repoint_one(hammer2_inode_t *ip, hammer2_cluster_t *cluster,
1076 hammer2_chain_t *ochain;
1077 hammer2_chain_t *nchain;
1080 hammer2_spin_ex(&ip->cluster_spin);
1081 KKASSERT(idx < cluster->nchains);
1082 if (idx < ip->cluster.nchains) {
1083 ochain = ip->cluster.array[idx].chain;
1084 nchain = cluster->array[idx].chain;
1087 nchain = cluster->array[idx].chain;
1088 for (i = ip->cluster.nchains; i <= idx; ++i) {
1089 bzero(&ip->cluster.array[i],
1090 sizeof(ip->cluster.array[i]));
1091 ip->cluster.array[i].flags |= HAMMER2_CITEM_INVALID;
1093 ip->cluster.nchains = idx + 1;
1095 if (ochain != nchain) {
1099 ip->cluster.array[idx].chain = nchain;
1100 ip->cluster.array[idx].flags &= ~HAMMER2_CITEM_INVALID;
1101 ip->cluster.array[idx].flags |= cluster->array[idx].flags &
1102 HAMMER2_CITEM_INVALID;
1104 hammer2_spin_unex(&ip->cluster_spin);
1105 if (ochain != nchain) {
1107 hammer2_chain_ref(nchain);
1109 hammer2_chain_drop(ochain);
1114 * Called with a locked inode to finish unlinking an inode after xop_unlink
1115 * had been run. This function is responsible for decrementing nlinks.
1117 * We don't bother decrementing nlinks if the file is not open and this was
1120 * If the inode is a hardlink target it's chain has not yet been deleted,
1121 * otherwise it's chain has been deleted.
1123 * If isopen then any prior deletion was not permanent and the inode is
1124 * left intact with nlinks == 0;
1127 hammer2_inode_unlink_finisher(hammer2_inode_t *ip, int isopen)
1135 * Decrement nlinks. If this is the last link and the file is
1136 * not open we can just delete the inode and not bother dropping
1137 * nlinks to 0 (avoiding unnecessary block updates).
1139 if (ip->meta.nlinks == 1) {
1140 atomic_set_int(&ip->flags, HAMMER2_INODE_ISUNLINKED);
1145 hammer2_inode_modify(ip);
1147 if ((int64_t)ip->meta.nlinks < 0)
1148 ip->meta.nlinks = 0; /* safety */
1151 * If nlinks is not zero we are done. However, this should only be
1152 * possible with a hardlink target. If the inode is an embedded
1153 * hardlink nlinks should have dropped to zero, warn and proceed
1154 * with the next step.
1156 if (ip->meta.nlinks) {
1157 if ((ip->meta.name_key & HAMMER2_DIRHASH_VISIBLE) == 0)
1159 kprintf("hammer2_inode_unlink: nlinks was not 0 (%jd)\n",
1160 (intmax_t)ip->meta.nlinks);
1165 hammer2_knote(ip->vp, NOTE_DELETE);
1168 * nlinks is now an implied zero, delete the inode if not open.
1169 * We avoid unnecessary media updates by not bothering to actually
1170 * decrement nlinks for the 1->0 transition
1172 * Put the inode on the sideq to ensure that any disconnected chains
1173 * get properly flushed (so they can be freed).
1176 hammer2_xop_destroy_t *xop;
1179 hammer2_inode_delayed_sideq(ip);
1180 atomic_set_int(&ip->flags, HAMMER2_INODE_ISDELETED);
1181 xop = hammer2_xop_alloc(ip, HAMMER2_XOP_MODIFYING);
1182 hammer2_xop_start(&xop->head, hammer2_inode_xop_destroy);
1183 error = hammer2_xop_collect(&xop->head, 0);
1184 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1186 error = 0; /* XXX */
1192 * Mark an inode as being modified, meaning that the caller will modify
1195 * If a vnode is present we set the vnode dirty and the nominal filesystem
1196 * sync will also handle synchronizing the inode meta-data. If no vnode
1197 * is present we must ensure that the inode is on pmp->sideq.
1199 * NOTE: No mtid (modify_tid) is passed into this routine. The caller is
1200 * only modifying the in-memory inode. A modify_tid is synchronized
1201 * later when the inode gets flushed.
1203 * NOTE: As an exception to the general rule, the inode MAY be locked
1204 * shared for this particular call.
1207 hammer2_inode_modify(hammer2_inode_t *ip)
1209 atomic_set_int(&ip->flags, HAMMER2_INODE_MODIFIED);
1211 vsetisdirty(ip->vp);
1212 } else if (ip->pmp && (ip->flags & HAMMER2_INODE_NOSIDEQ) == 0) {
1213 hammer2_inode_delayed_sideq(ip);
1218 * Synchronize the inode's frontend state with the chain state prior
1219 * to any explicit flush of the inode or any strategy write call. This
1220 * does not flush the inode's chain or its sub-topology to media (higher
1221 * level layers are responsible for doing that).
1223 * Called with a locked inode inside a normal transaction.
1225 * inode must be locked.
1228 hammer2_inode_chain_sync(hammer2_inode_t *ip)
1233 if (ip->flags & (HAMMER2_INODE_RESIZED | HAMMER2_INODE_MODIFIED)) {
1234 hammer2_xop_fsync_t *xop;
1236 xop = hammer2_xop_alloc(ip, HAMMER2_XOP_MODIFYING);
1237 xop->clear_directdata = 0;
1238 if (ip->flags & HAMMER2_INODE_RESIZED) {
1239 if ((ip->meta.op_flags & HAMMER2_OPFLAG_DIRECTDATA) &&
1240 ip->meta.size > HAMMER2_EMBEDDED_BYTES) {
1241 ip->meta.op_flags &= ~HAMMER2_OPFLAG_DIRECTDATA;
1242 xop->clear_directdata = 1;
1244 xop->osize = ip->osize;
1246 xop->osize = ip->meta.size; /* safety */
1248 xop->ipflags = ip->flags;
1249 xop->meta = ip->meta;
1251 atomic_clear_int(&ip->flags, HAMMER2_INODE_RESIZED |
1252 HAMMER2_INODE_MODIFIED);
1253 hammer2_xop_start(&xop->head, hammer2_inode_xop_chain_sync);
1254 error = hammer2_xop_collect(&xop->head, 0);
1255 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1256 if (error == HAMMER2_ERROR_ENOENT)
1259 kprintf("hammer2: unable to fsync inode %p\n", ip);
1261 atomic_set_int(&ip->flags,
1262 xop->ipflags & (HAMMER2_INODE_RESIZED |
1263 HAMMER2_INODE_MODIFIED));
1265 /* XXX return error somehow? */
1272 * Flushes the inode's chain and its sub-topology to media. Interlocks
1273 * HAMMER2_INODE_DIRTYDATA by clearing it prior to the flush. Any strategy
1274 * function creating or modifying a chain under this inode will re-set the
1277 * inode must be locked.
1280 hammer2_inode_chain_flush(hammer2_inode_t *ip)
1282 hammer2_xop_fsync_t *xop;
1285 atomic_clear_int(&ip->flags, HAMMER2_INODE_DIRTYDATA);
1286 xop = hammer2_xop_alloc(ip, HAMMER2_XOP_MODIFYING |
1287 HAMMER2_XOP_INODE_STOP);
1288 hammer2_xop_start(&xop->head, hammer2_inode_xop_flush);
1289 error = hammer2_xop_collect(&xop->head, HAMMER2_XOP_COLLECT_WAITALL);
1290 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1291 if (error == HAMMER2_ERROR_ENOENT)
1298 * The normal filesystem sync no longer has visibility to an inode structure
1299 * after its vnode has been reclaimed. In this situation a dirty inode may
1300 * require additional processing to synchronize ip->meta to its underlying
1303 * In particular, reclaims can occur in almost any state (for example, when
1304 * doing operations on unrelated vnodes) and flushing the reclaimed inode
1305 * in the reclaim path itself is a non-starter.
1307 * Caller must be in a transaction.
1310 hammer2_inode_run_sideq(hammer2_pfs_t *pmp, int doall)
1312 hammer2_xop_destroy_t *xop;
1313 hammer2_inode_sideq_t *ipul;
1314 hammer2_inode_t *ip;
1318 * Nothing to do if sideq is empty or (if doall == 0) there just
1319 * aren't very many sideq entries.
1321 if (TAILQ_EMPTY(&pmp->sideq))
1324 if (pmp->sideq_count > (pmp->inum_count >> 3)) {
1325 if (hammer2_debug & 0x0001) {
1326 kprintf("hammer2: flush sideq %ld/%ld\n",
1327 pmp->sideq_count, pmp->inum_count);
1332 if (doall == 0 && pmp->sideq_count <= (pmp->inum_count >> 3))
1335 hammer2_spin_ex(&pmp->list_spin);
1336 while ((ipul = TAILQ_FIRST(&pmp->sideq)) != NULL) {
1337 TAILQ_REMOVE(&pmp->sideq, ipul, entry);
1340 KKASSERT(ip->flags & HAMMER2_INODE_ONSIDEQ);
1341 atomic_clear_int(&ip->flags, HAMMER2_INODE_ONSIDEQ);
1342 hammer2_spin_unex(&pmp->list_spin);
1343 kfree(ipul, pmp->minode);
1345 hammer2_inode_lock(ip, 0);
1346 if (ip->flags & HAMMER2_INODE_ISDELETED) {
1348 * The inode has already been deleted. This is a
1349 * fairly rare circumstance. For now we don't rock
1350 * the boat and synchronize it normally.
1352 hammer2_inode_chain_sync(ip);
1353 hammer2_inode_chain_flush(ip);
1354 } else if (ip->flags & HAMMER2_INODE_ISUNLINKED) {
1356 * The inode was unlinked while open. The inode must
1357 * be deleted and destroyed.
1359 xop = hammer2_xop_alloc(ip, HAMMER2_XOP_MODIFYING);
1360 hammer2_xop_start(&xop->head,
1361 hammer2_inode_xop_destroy);
1362 error = hammer2_xop_collect(&xop->head, 0);
1363 /* XXX error handling */
1364 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1367 * The inode was dirty as-of the reclaim, requiring
1368 * synchronization of ip->meta with its underlying
1371 hammer2_inode_chain_sync(ip);
1372 hammer2_inode_chain_flush(ip);
1375 hammer2_inode_unlock(ip);
1376 hammer2_inode_drop(ip); /* ipul ref */
1378 hammer2_spin_ex(&pmp->list_spin);
1381 * If doall is 0 the original sideq_count was greater than
1382 * 1/8 the inode count. Add some hysteresis in the loop,
1383 * don't stop flushing until sideq_count drops below 1/16.
1385 if (doall == 0 && pmp->sideq_count <= (pmp->inum_count >> 4)) {
1386 if (hammer2_debug & 0x0001) {
1387 kprintf("hammer2: flush sideq %ld/%ld (end)\n",
1388 pmp->sideq_count, pmp->inum_count);
1393 hammer2_spin_unex(&pmp->list_spin);
1397 * Helper to create a directory entry.
1400 hammer2_inode_xop_mkdirent(hammer2_thread_t *thr, hammer2_xop_t *arg)
1402 hammer2_xop_mkdirent_t *xop = &arg->xop_mkdirent;
1403 hammer2_chain_t *parent;
1404 hammer2_chain_t *chain;
1405 hammer2_key_t key_next;
1409 if (hammer2_debug & 0x0001)
1410 kprintf("dirent_create lhc %016jx clindex %d\n",
1411 xop->lhc, thr->clindex);
1413 parent = hammer2_inode_chain(xop->head.ip1, thr->clindex,
1414 HAMMER2_RESOLVE_ALWAYS);
1415 if (parent == NULL) {
1416 error = HAMMER2_ERROR_EIO;
1420 chain = hammer2_chain_lookup(&parent, &key_next,
1424 error = HAMMER2_ERROR_EEXIST;
1429 * We may be able to embed the directory entry directly in the
1432 if (xop->dirent.namlen <= sizeof(chain->bref.check.buf))
1435 data_len = HAMMER2_ALLOC_MIN;
1437 error = hammer2_chain_create(&parent, &chain,
1438 xop->head.ip1->pmp, HAMMER2_METH_DEFAULT,
1440 HAMMER2_BREF_TYPE_DIRENT,
1442 xop->head.mtid, 0, 0);
1445 * WARNING: chain->data->buf is sized to chain->bytes,
1446 * do not use sizeof(chain->data->buf), which
1447 * will be much larger.
1449 error = hammer2_chain_modify(chain, xop->head.mtid, 0, 0);
1451 chain->bref.embed.dirent = xop->dirent;
1452 if (xop->dirent.namlen <= sizeof(chain->bref.check.buf))
1453 bcopy(xop->head.name1, chain->bref.check.buf,
1454 xop->dirent.namlen);
1456 bcopy(xop->head.name1, chain->data->buf,
1457 xop->dirent.namlen);
1462 hammer2_chain_unlock(parent);
1463 hammer2_chain_drop(parent);
1465 hammer2_xop_feed(&xop->head, chain, thr->clindex, error);
1467 hammer2_chain_unlock(chain);
1468 hammer2_chain_drop(chain);
1473 * Inode create helper (threaded, backend)
1475 * Used by ncreate, nmknod, nsymlink, nmkdir.
1476 * Used by nlink and rename to create HARDLINK pointers.
1478 * Frontend holds the parent directory ip locked exclusively. We
1479 * create the inode and feed the exclusively locked chain to the
1483 hammer2_inode_xop_create(hammer2_thread_t *thr, hammer2_xop_t *arg)
1485 hammer2_xop_create_t *xop = &arg->xop_create;
1486 hammer2_chain_t *parent;
1487 hammer2_chain_t *chain;
1488 hammer2_key_t key_next;
1491 if (hammer2_debug & 0x0001)
1492 kprintf("inode_create lhc %016jx clindex %d\n",
1493 xop->lhc, thr->clindex);
1495 parent = hammer2_inode_chain(xop->head.ip1, thr->clindex,
1496 HAMMER2_RESOLVE_ALWAYS);
1497 if (parent == NULL) {
1498 error = HAMMER2_ERROR_EIO;
1502 chain = hammer2_chain_lookup(&parent, &key_next,
1506 error = HAMMER2_ERROR_EEXIST;
1510 error = hammer2_chain_create(&parent, &chain,
1511 xop->head.ip1->pmp, HAMMER2_METH_DEFAULT,
1513 HAMMER2_BREF_TYPE_INODE,
1514 HAMMER2_INODE_BYTES,
1515 xop->head.mtid, 0, xop->flags);
1517 error = hammer2_chain_modify(chain, xop->head.mtid, 0, 0);
1519 chain->data->ipdata.meta = xop->meta;
1520 if (xop->head.name1) {
1521 bcopy(xop->head.name1,
1522 chain->data->ipdata.filename,
1523 xop->head.name1_len);
1524 chain->data->ipdata.meta.name_len =
1525 xop->head.name1_len;
1527 chain->data->ipdata.meta.name_key = xop->lhc;
1532 hammer2_chain_unlock(parent);
1533 hammer2_chain_drop(parent);
1535 hammer2_xop_feed(&xop->head, chain, thr->clindex, error);
1537 hammer2_chain_unlock(chain);
1538 hammer2_chain_drop(chain);
1543 * Inode delete helper (backend, threaded)
1545 * Generally used by hammer2_run_sideq()
1548 hammer2_inode_xop_destroy(hammer2_thread_t *thr, hammer2_xop_t *arg)
1550 hammer2_xop_destroy_t *xop = &arg->xop_destroy;
1552 hammer2_chain_t *parent;
1553 hammer2_chain_t *chain;
1554 hammer2_inode_t *ip;
1558 * We need the precise parent chain to issue the deletion.
1563 chain = hammer2_inode_chain(ip, thr->clindex, HAMMER2_RESOLVE_ALWAYS);
1564 if (chain == NULL) {
1566 error = HAMMER2_ERROR_EIO;
1569 parent = hammer2_chain_getparent(chain, HAMMER2_RESOLVE_ALWAYS);
1570 if (parent == NULL) {
1571 error = HAMMER2_ERROR_EIO;
1574 KKASSERT(chain->parent == parent);
1577 * We have the correct parent, we can issue the deletion.
1579 hammer2_chain_delete(parent, chain, xop->head.mtid, 0);
1582 hammer2_xop_feed(&xop->head, NULL, thr->clindex, error);
1584 hammer2_chain_unlock(parent);
1585 hammer2_chain_drop(parent);
1588 hammer2_chain_unlock(chain);
1589 hammer2_chain_drop(chain);
1594 hammer2_inode_xop_unlinkall(hammer2_thread_t *thr, hammer2_xop_t *arg)
1596 hammer2_xop_unlinkall_t *xop = &arg->xop_unlinkall;
1597 hammer2_chain_t *parent;
1598 hammer2_chain_t *chain;
1599 hammer2_key_t key_next;
1603 * We need the precise parent chain to issue the deletion.
1605 parent = hammer2_inode_chain(xop->head.ip1, thr->clindex,
1606 HAMMER2_RESOLVE_ALWAYS);
1608 if (parent == NULL) {
1612 chain = hammer2_chain_lookup(&parent, &key_next,
1613 xop->key_beg, xop->key_end,
1614 &error, HAMMER2_LOOKUP_ALWAYS);
1616 hammer2_chain_delete(parent, chain,
1617 xop->head.mtid, HAMMER2_DELETE_PERMANENT);
1618 hammer2_xop_feed(&xop->head, chain, thr->clindex, chain->error);
1619 /* depend on function to unlock the shared lock */
1620 chain = hammer2_chain_next(&parent, chain, &key_next,
1621 key_next, xop->key_end,
1623 HAMMER2_LOOKUP_ALWAYS);
1627 error = HAMMER2_ERROR_ENOENT;
1628 hammer2_xop_feed(&xop->head, NULL, thr->clindex, error);
1630 hammer2_chain_unlock(parent);
1631 hammer2_chain_drop(parent);
1634 hammer2_chain_unlock(chain);
1635 hammer2_chain_drop(chain);
1640 hammer2_inode_xop_connect(hammer2_thread_t *thr, hammer2_xop_t *arg)
1642 hammer2_xop_connect_t *xop = &arg->xop_connect;
1643 hammer2_inode_data_t *wipdata;
1644 hammer2_chain_t *parent;
1645 hammer2_chain_t *chain;
1647 hammer2_key_t key_dummy;
1651 * Get directory, then issue a lookup to prime the parent chain
1652 * for the create. The lookup is expected to fail.
1654 pmp = xop->head.ip1->pmp;
1655 parent = hammer2_inode_chain(xop->head.ip1, thr->clindex,
1656 HAMMER2_RESOLVE_ALWAYS);
1657 if (parent == NULL) {
1659 error = HAMMER2_ERROR_EIO;
1662 chain = hammer2_chain_lookup(&parent, &key_dummy,
1666 hammer2_chain_unlock(chain);
1667 hammer2_chain_drop(chain);
1669 error = HAMMER2_ERROR_EEXIST;
1676 * Adjust the filename in the inode, set the name key.
1678 * NOTE: Frontend must also adjust ip2->meta on success, we can't
1681 chain = hammer2_inode_chain(xop->head.ip2, thr->clindex,
1682 HAMMER2_RESOLVE_ALWAYS);
1683 error = hammer2_chain_modify(chain, xop->head.mtid, 0, 0);
1687 wipdata = &chain->data->ipdata;
1689 hammer2_inode_modify(xop->head.ip2);
1690 if (xop->head.name1) {
1691 bzero(wipdata->filename, sizeof(wipdata->filename));
1692 bcopy(xop->head.name1, wipdata->filename, xop->head.name1_len);
1693 wipdata->meta.name_len = xop->head.name1_len;
1695 wipdata->meta.name_key = xop->lhc;
1698 * Reconnect the chain to the new parent directory
1700 error = hammer2_chain_create(&parent, &chain,
1701 pmp, HAMMER2_METH_DEFAULT,
1703 HAMMER2_BREF_TYPE_INODE,
1704 HAMMER2_INODE_BYTES,
1705 xop->head.mtid, 0, 0);
1711 hammer2_xop_feed(&xop->head, NULL, thr->clindex, error);
1713 hammer2_chain_unlock(parent);
1714 hammer2_chain_drop(parent);
1717 hammer2_chain_unlock(chain);
1718 hammer2_chain_drop(chain);
1723 * Synchronize the in-memory inode with the chain. This does not flush
1724 * the chain to disk. Instead, it makes front-end inode changes visible
1725 * in the chain topology, thus visible to the backend. This is done in an
1726 * ad-hoc manner outside of the filesystem vfs_sync, and in a controlled
1727 * manner inside the vfs_sync.
1730 hammer2_inode_xop_chain_sync(hammer2_thread_t *thr, hammer2_xop_t *arg)
1732 hammer2_xop_fsync_t *xop = &arg->xop_fsync;
1733 hammer2_chain_t *parent;
1734 hammer2_chain_t *chain;
1737 parent = hammer2_inode_chain(xop->head.ip1, thr->clindex,
1738 HAMMER2_RESOLVE_ALWAYS);
1740 if (parent == NULL) {
1741 error = HAMMER2_ERROR_EIO;
1744 if (parent->error) {
1745 error = parent->error;
1751 if ((xop->ipflags & HAMMER2_INODE_RESIZED) == 0) {
1752 /* osize must be ignored */
1753 } else if (xop->meta.size < xop->osize) {
1755 * We must delete any chains beyond the EOF. The chain
1756 * straddling the EOF will be pending in the bioq.
1758 hammer2_key_t lbase;
1759 hammer2_key_t key_next;
1761 lbase = (xop->meta.size + HAMMER2_PBUFMASK64) &
1762 ~HAMMER2_PBUFMASK64;
1763 chain = hammer2_chain_lookup(&parent, &key_next,
1764 lbase, HAMMER2_KEY_MAX,
1766 HAMMER2_LOOKUP_NODATA |
1767 HAMMER2_LOOKUP_NODIRECT);
1770 * Degenerate embedded case, nothing to loop on
1772 switch (chain->bref.type) {
1773 case HAMMER2_BREF_TYPE_DIRENT:
1774 case HAMMER2_BREF_TYPE_INODE:
1777 case HAMMER2_BREF_TYPE_DATA:
1778 hammer2_chain_delete(parent, chain,
1780 HAMMER2_DELETE_PERMANENT);
1783 chain = hammer2_chain_next(&parent, chain, &key_next,
1784 key_next, HAMMER2_KEY_MAX,
1786 HAMMER2_LOOKUP_NODATA |
1787 HAMMER2_LOOKUP_NODIRECT);
1791 * Reset to point at inode for following code, if necessary.
1793 if (parent->bref.type != HAMMER2_BREF_TYPE_INODE) {
1794 hammer2_chain_unlock(parent);
1795 hammer2_chain_drop(parent);
1796 parent = hammer2_inode_chain(xop->head.ip1,
1798 HAMMER2_RESOLVE_ALWAYS);
1799 kprintf("hammer2: TRUNCATE RESET on '%s'\n",
1800 parent->data->ipdata.filename);
1805 * Sync the inode meta-data, potentially clear the blockset area
1806 * of direct data so it can be used for blockrefs.
1809 error = hammer2_chain_modify(parent, xop->head.mtid, 0, 0);
1811 parent->data->ipdata.meta = xop->meta;
1812 if (xop->clear_directdata) {
1813 bzero(&parent->data->ipdata.u.blockset,
1814 sizeof(parent->data->ipdata.u.blockset));
1820 hammer2_chain_unlock(chain);
1821 hammer2_chain_drop(chain);
1824 hammer2_chain_unlock(parent);
1825 hammer2_chain_drop(parent);
1827 hammer2_xop_feed(&xop->head, NULL, thr->clindex, error);