2 * Copyright (c) 2015 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>
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
35 * This module implements the cluster synchronizer. Basically the way
36 * it works is that a thread is created for each cluster node in a PFS.
37 * This thread is responsible for synchronizing the current node using
38 * data from other nodes.
40 * Any out of sync master or slave can get back into synchronization as
41 * long as a quorum of masters agree on the update_tid. If a quorum is
42 * not available it may still be possible to synchronize to the highest
43 * available update_tid as a way of trying to catch up as much as possible
44 * until a quorum is available.
46 * If no quorum is possible (which can happen even if all masters are
47 * available, if the update_tid does not match), then manual intervention
48 * may be required to resolve discrepancies.
52 typedef struct hammer2_deferred_ip
{
53 struct hammer2_deferred_ip
*next
;
55 } hammer2_deferred_ip_t
;
57 typedef struct hammer2_deferred_list
{
58 hammer2_deferred_ip_t
*base
;
60 } hammer2_deferred_list_t
;
63 #define HAMMER2_SYNCHRO_DEBUG 1
65 static int hammer2_sync_slaves(hammer2_thread_t
*thr
, hammer2_inode_t
*ip
,
66 hammer2_deferred_list_t
*list
, int isroot
);
68 static void hammer2_update_pfs_status(hammer2_thread_t
*thr
, uint32_t flags
);
69 nerror
= hammer2_sync_insert(
71 focus
->bref
.modify_tid
,
74 static int hammer2_sync_insert(hammer2_thread_t
*thr
,
75 hammer2_chain_t
**parentp
, hammer2_chain_t
**chainp
,
76 hammer2_tid_t modify_tid
, int idx
,
77 hammer2_chain_t
*focus
);
78 static int hammer2_sync_destroy(hammer2_thread_t
*thr
,
79 hammer2_chain_t
**parentp
, hammer2_chain_t
**chainp
,
80 hammer2_tid_t mtid
, int idx
);
81 static int hammer2_sync_replace(hammer2_thread_t
*thr
,
82 hammer2_chain_t
*parent
, hammer2_chain_t
*chain
,
83 hammer2_tid_t mtid
, int idx
,
84 hammer2_chain_t
*focus
, int isroot
);
86 /****************************************************************************
87 * HAMMER2 SYNC THREADS *
88 ****************************************************************************/
90 * Primary management thread for an element of a node. A thread will exist
91 * for each element requiring management.
93 * No management threads are needed for the SPMP or for any PMP with only
96 * On the SPMP - handles bulkfree and dedup operations
97 * On a PFS - handles remastering and synchronization
100 hammer2_primary_sync_thread(void *arg
)
102 hammer2_thread_t
*thr
= arg
;
104 hammer2_deferred_list_t list
;
105 hammer2_deferred_ip_t
*defer
;
111 bzero(&list
, sizeof(list
));
118 * Handle stop request
120 if (flags
& HAMMER2_THREAD_STOP
)
124 * Handle freeze request
126 if (flags
& HAMMER2_THREAD_FREEZE
) {
127 nflags
= (flags
& ~(HAMMER2_THREAD_FREEZE
|
128 HAMMER2_THREAD_CLIENTWAIT
)) |
129 HAMMER2_THREAD_FROZEN
;
130 if (!atomic_cmpset_int(&thr
->flags
, flags
, nflags
))
132 if (flags
& HAMMER2_THREAD_CLIENTWAIT
)
138 if (flags
& HAMMER2_THREAD_UNFREEZE
) {
139 nflags
= flags
& ~(HAMMER2_THREAD_UNFREEZE
|
140 HAMMER2_THREAD_FROZEN
|
141 HAMMER2_THREAD_CLIENTWAIT
);
142 if (!atomic_cmpset_int(&thr
->flags
, flags
, nflags
))
144 if (flags
& HAMMER2_THREAD_CLIENTWAIT
)
151 * Force idle if frozen until unfrozen or stopped.
153 if (flags
& HAMMER2_THREAD_FROZEN
) {
154 nflags
= flags
| HAMMER2_THREAD_WAITING
;
155 tsleep_interlock(&thr
->flags
, 0);
156 if (atomic_cmpset_int(&thr
->flags
, flags
, nflags
)) {
157 tsleep(&thr
->flags
, PINTERLOCKED
, "frozen", 0);
158 atomic_clear_int(&thr
->flags
,
159 HAMMER2_THREAD_WAITING
);
165 * Reset state on REMASTER request
167 if (thr
->flags
& HAMMER2_THREAD_REMASTER
) {
168 nflags
= flags
& ~HAMMER2_THREAD_REMASTER
;
169 if (atomic_cmpset_int(&thr
->flags
, flags
, nflags
)) {
170 /* reset state here */
176 * Synchronization scan.
178 if (hammer2_debug
& 0x8000)
179 kprintf("sync_slaves pfs %s clindex %d\n",
180 pmp
->pfs_names
[thr
->clindex
], thr
->clindex
);
181 hammer2_trans_init(pmp
, 0);
183 hammer2_inode_ref(pmp
->iroot
);
187 /* XXX lock synchronize pmp->modify_tid */
188 error
= hammer2_sync_slaves(thr
, pmp
->iroot
, &list
, 1);
189 if (hammer2_debug
& 0x8000) {
190 kprintf("sync_slaves error %d defer %p\n",
195 while ((defer
= list
.base
) != NULL
) {
196 hammer2_inode_t
*nip
;
199 error
= hammer2_sync_slaves(thr
, nip
, &list
, 0);
200 if (error
&& error
!= EAGAIN
&& error
!= ENOENT
)
202 if (hammer2_thr_break(thr
)) {
208 * If no additional defers occurred we can
209 * remove this one, otherwise keep it on
210 * the list and retry once the additional
211 * defers have completed.
213 if (defer
== list
.base
) {
215 list
.base
= defer
->next
;
216 kfree(defer
, M_HAMMER2
);
217 defer
= NULL
; /* safety */
218 hammer2_inode_drop(nip
);
223 * If the thread is being remastered, frozen, or
224 * stopped, clean up any left-over deferals.
226 if (didbreak
|| (error
&& error
!= EAGAIN
)) {
227 kprintf("didbreak\n");
228 while ((defer
= list
.base
) != NULL
) {
230 hammer2_inode_drop(defer
->ip
);
231 list
.base
= defer
->next
;
232 kfree(defer
, M_HAMMER2
);
234 if (error
== 0 || error
== EAGAIN
)
240 hammer2_inode_drop(pmp
->iroot
);
241 hammer2_trans_done(pmp
);
243 if (error
&& error
!= EINPROGRESS
)
244 kprintf("hammer2_sync_slaves: error %d\n", error
);
247 * Wait for event, or 5-second poll.
249 nflags
= flags
| HAMMER2_THREAD_WAITING
;
250 tsleep_interlock(&thr
->flags
, 0);
251 if (atomic_cmpset_int(&thr
->flags
, flags
, nflags
)) {
252 tsleep(&thr
->flags
, 0, "h2idle", hz
* 5);
253 atomic_clear_int(&thr
->flags
, HAMMER2_THREAD_WAITING
);
257 hammer2_thr_return(thr
, HAMMER2_THREAD_STOPPED
);
258 /* thr structure can go invalid after this point */
264 * Given a locked cluster created from pmp->iroot, update the PFS's
269 hammer2_update_pfs_status(hammer2_thread_t
*thr
, uint32_t flags
)
271 hammer2_pfs_t
*pmp
= thr
->pmp
;
273 flags
&= HAMMER2_CLUSTER_ZFLAGS
;
274 if (pmp
->cluster_flags
== flags
)
276 pmp
->cluster_flags
= flags
;
278 kprintf("pfs %p", pmp
);
279 if (flags
& HAMMER2_CLUSTER_MSYNCED
)
280 kprintf(" masters-all-good");
281 if (flags
& HAMMER2_CLUSTER_SSYNCED
)
282 kprintf(" slaves-all-good");
284 if (flags
& HAMMER2_CLUSTER_WRHARD
)
285 kprintf(" quorum/rw");
286 else if (flags
& HAMMER2_CLUSTER_RDHARD
)
287 kprintf(" quorum/ro");
289 if (flags
& HAMMER2_CLUSTER_UNHARD
)
290 kprintf(" out-of-sync-masters");
291 else if (flags
& HAMMER2_CLUSTER_NOHARD
)
292 kprintf(" no-masters-visible");
294 if (flags
& HAMMER2_CLUSTER_WRSOFT
)
296 else if (flags
& HAMMER2_CLUSTER_RDSOFT
)
299 if (flags
& HAMMER2_CLUSTER_UNSOFT
)
300 kprintf(" out-of-sync-slaves");
301 else if (flags
& HAMMER2_CLUSTER_NOSOFT
)
302 kprintf(" no-slaves-visible");
310 dumpcluster(const char *label
,
311 hammer2_cluster_t
*cparent
, hammer2_cluster_t
*cluster
)
313 hammer2_chain_t
*chain
;
316 if ((hammer2_debug
& 1) == 0)
319 kprintf("%s\t", label
);
320 KKASSERT(cparent
->nchains
== cluster
->nchains
);
321 for (i
= 0; i
< cparent
->nchains
; ++i
) {
325 if ((chain
= cparent
->array
[i
].chain
) != NULL
) {
328 ((cparent
->array
[i
].flags
&
329 HAMMER2_CITEM_INVALID
) ? "(I)" : " ")
332 kprintf(" NULL %s ", " ");
334 if ((chain
= cluster
->array
[i
].chain
) != NULL
) {
337 ((cluster
->array
[i
].flags
&
338 HAMMER2_CITEM_INVALID
) ? "(I)" : " ")
341 kprintf(" NULL %s ", " ");
349 * Each out of sync node sync-thread must issue an all-nodes XOP scan of
350 * the inode. This creates a multiplication effect since the XOP scan itself
351 * issues to all nodes. However, this is the only way we can safely
352 * synchronize nodes which might have disparate I/O bandwidths and the only
353 * way we can safely deal with stalled nodes.
357 hammer2_sync_slaves(hammer2_thread_t
*thr
, hammer2_inode_t
*ip
,
358 hammer2_deferred_list_t
*list
, int isroot
)
360 hammer2_xop_scanall_t
*xop
;
361 hammer2_chain_t
*parent
;
362 hammer2_chain_t
*chain
;
364 hammer2_key_t key_next
;
365 hammer2_tid_t sync_tid
;
366 int cache_index
= -1;
375 idx
= thr
->clindex
; /* cluster node we are responsible for */
384 * Nothing to do if all slaves are synchronized.
385 * Nothing to do if cluster not authoritatively readable.
387 if (pmp
->cluster_flags
& HAMMER2_CLUSTER_SSYNCED
)
389 if ((pmp
->cluster_flags
& HAMMER2_CLUSTER_RDHARD
) == 0)
390 return(HAMMER2_ERROR_INCOMPLETE
);
396 * Resolve the root inode of the PFS and determine if synchronization
397 * is needed by checking modify_tid.
400 hammer2_xop_ipcluster_t
*xop2
;
401 hammer2_chain_t
*focus
;
403 hammer2_inode_lock(ip
, HAMMER2_RESOLVE_SHARED
);
404 xop2
= hammer2_xop_alloc(ip
, HAMMER2_XOP_MODIFYING
);
405 hammer2_xop_start_except(&xop2
->head
, hammer2_xop_ipcluster
,
407 hammer2_inode_unlock(ip
);
408 error
= hammer2_xop_collect(&xop2
->head
, 0);
409 if (error
== 0 && (focus
= xop2
->head
.cluster
.focus
) != NULL
) {
410 sync_tid
= focus
->bref
.modify_tid
; /* XXX */
411 chain
= hammer2_inode_chain_and_parent(ip
, idx
,
413 HAMMER2_RESOLVE_ALWAYS
|
414 HAMMER2_RESOLVE_SHARED
);
415 want_update
= (chain
->bref
.modify_tid
!= sync_tid
);
417 hammer2_chain_unlock(chain
);
418 hammer2_chain_drop(chain
);
422 hammer2_chain_unlock(parent
);
423 hammer2_chain_drop(parent
);
427 hammer2_xop_retire(&xop2
->head
, HAMMER2_XOPMASK_VOP
);
430 if (want_update
== 0)
434 * The inode is left unlocked during the scan. Issue a XOP
435 * that does *not* include our cluster index to iterate
436 * properly synchronized elements and resolve our cluster index
439 hammer2_inode_lock(ip
, HAMMER2_RESOLVE_SHARED
);
440 xop
= hammer2_xop_alloc(ip
, HAMMER2_XOP_MODIFYING
);
441 xop
->key_beg
= HAMMER2_KEY_MIN
;
442 xop
->key_end
= HAMMER2_KEY_MAX
;
443 xop
->resolve_flags
= HAMMER2_RESOLVE_SHARED
|
444 HAMMER2_RESOLVE_ALWAYS
;
445 xop
->lookup_flags
= HAMMER2_LOOKUP_SHARED
|
446 HAMMER2_LOOKUP_NODIRECT
|
447 HAMMER2_LOOKUP_ALWAYS
;
448 hammer2_xop_start_except(&xop
->head
, hammer2_xop_scanall
, idx
);
449 parent
= hammer2_inode_chain(ip
, idx
,
450 HAMMER2_RESOLVE_ALWAYS
|
451 HAMMER2_RESOLVE_SHARED
);
452 hammer2_inode_unlock(ip
);
454 chain
= hammer2_chain_lookup(&parent
, &key_next
,
455 HAMMER2_KEY_MIN
, HAMMER2_KEY_MAX
,
457 HAMMER2_LOOKUP_SHARED
|
458 HAMMER2_LOOKUP_NODIRECT
|
459 HAMMER2_LOOKUP_NODATA
);
460 error
= hammer2_xop_collect(&xop
->head
, 0);
461 kprintf("START_SCAN IP=%016jx chain=%p (%016jx)\n",
462 ip
->meta
.name_key
, chain
,
463 (chain
? chain
->bref
.key
: -1));
467 * We are done if our scan is done and the XOP scan is done.
468 * We are done if the XOP scan failed (that is, we don't
469 * have authoritative data to synchronize with).
471 int advance_local
= 0;
474 hammer2_chain_t
*focus
;
476 if (chain
== NULL
&& error
== ENOENT
)
478 if (error
&& error
!= ENOENT
)
484 if (chain
&& error
== ENOENT
) {
486 * If we have local chains but the XOP scan is done,
487 * the chains need to be deleted.
491 } else if (chain
== NULL
) {
493 * If our local scan is done but the XOP scan is not,
494 * we need to create the missing chain(s).
497 focus
= xop
->head
.cluster
.focus
;
500 * Otherwise compare to determine the action
503 focus
= xop
->head
.cluster
.focus
;
504 n
= hammer2_chain_cmp(chain
, focus
);
508 * Take action based on comparison results.
512 * Delete extranious local data. This will
513 * automatically advance the chain.
515 nerror
= hammer2_sync_destroy(thr
, &parent
, &chain
,
517 } else if (n
== 0 && chain
->bref
.modify_tid
!=
518 focus
->bref
.modify_tid
) {
520 * Matching key but local data or meta-data requires
521 * updating. If we will recurse, we still need to
522 * update to compatible content first but we do not
523 * synchronize modify_tid until the entire recursion
524 * has completed successfully.
526 * NOTE: Do not try to access hardlink pointers as if
527 * they were normal inodes, the inode cache will
528 * get seriously confused.
530 if (focus
->bref
.type
== HAMMER2_BREF_TYPE_INODE
&&
531 focus
->data
->ipdata
.meta
.type
!=
532 HAMMER2_OBJTYPE_HARDLINK
) {
533 nerror
= hammer2_sync_replace(
539 nerror
= hammer2_sync_replace(
541 focus
->bref
.modify_tid
,
548 * 100% match, advance both
555 * Insert missing local data.
557 * If we will recurse, we still need to update to
558 * compatible content first but we do not synchronize
559 * modify_tid until the entire recursion has
560 * completed successfully.
562 * NOTE: Do not try to access hardlink pointers as if
563 * they were normal inodes, the inode cache will
564 * get seriously confused.
566 if (focus
->bref
.type
== HAMMER2_BREF_TYPE_INODE
&&
567 focus
->data
->ipdata
.meta
.type
!=
568 HAMMER2_OBJTYPE_HARDLINK
) {
569 nerror
= hammer2_sync_insert(
570 thr
, &parent
, &chain
,
575 nerror
= hammer2_sync_insert(
576 thr
, &parent
, &chain
,
577 focus
->bref
.modify_tid
,
585 * We cannot recurse depth-first because the XOP is still
586 * running in node threads for this scan. Create a placemarker
587 * by obtaining and record the hammer2_inode.
589 * We excluded our node from the XOP so we must temporarily
590 * add it to xop->head.cluster so it is properly incorporated
593 * The deferral is pushed onto a LIFO list for bottom-up
596 if (error
== 0 && dodefer
) {
597 hammer2_inode_t
*nip
;
598 hammer2_deferred_ip_t
*defer
;
600 KKASSERT(focus
->bref
.type
== HAMMER2_BREF_TYPE_INODE
);
602 defer
= kmalloc(sizeof(*defer
), M_HAMMER2
,
604 KKASSERT(xop
->head
.cluster
.array
[idx
].chain
== NULL
);
605 xop
->head
.cluster
.array
[idx
].flags
=
606 HAMMER2_CITEM_INVALID
;
607 xop
->head
.cluster
.array
[idx
].chain
= chain
;
608 nip
= hammer2_inode_get(pmp
, ip
,
609 &xop
->head
.cluster
, idx
);
610 xop
->head
.cluster
.array
[idx
].chain
= NULL
;
612 hammer2_inode_ref(nip
);
613 hammer2_inode_unlock(nip
);
615 defer
->next
= list
->base
;
623 * If at least one deferral was added and the deferral
624 * list has grown too large, stop adding more. This
625 * will trigger an EAGAIN return.
627 if (needrescan
&& list
->count
> 1000)
631 * Advancements for iteration.
634 error
= hammer2_xop_collect(&xop
->head
, 0);
637 chain
= hammer2_chain_next(&parent
, chain
, &key_next
,
638 key_next
, HAMMER2_KEY_MAX
,
640 HAMMER2_LOOKUP_SHARED
|
641 HAMMER2_LOOKUP_NODIRECT
|
642 HAMMER2_LOOKUP_NODATA
);
645 hammer2_xop_retire(&xop
->head
, HAMMER2_XOPMASK_VOP
);
647 hammer2_chain_unlock(chain
);
648 hammer2_chain_drop(chain
);
651 hammer2_chain_unlock(parent
);
652 hammer2_chain_drop(parent
);
656 * If we added deferrals we want the caller to synchronize them
657 * and then call us again.
659 * NOTE: In this situation we do not yet want to synchronize our
660 * inode, setting the error code also has that effect.
662 if ((error
== 0 || error
== ENOENT
) && needrescan
)
666 * If no error occurred we can synchronize the inode meta-data
667 * and modify_tid. Only limited changes are made to PFSROOTs.
669 * XXX inode lock was lost
671 if (error
== 0 || error
== ENOENT
) {
672 hammer2_xop_ipcluster_t
*xop2
;
673 hammer2_chain_t
*focus
;
675 hammer2_inode_lock(ip
, HAMMER2_RESOLVE_SHARED
);
676 xop2
= hammer2_xop_alloc(ip
, HAMMER2_XOP_MODIFYING
);
677 hammer2_xop_start_except(&xop2
->head
, hammer2_xop_ipcluster
,
679 hammer2_inode_unlock(ip
);
680 error
= hammer2_xop_collect(&xop2
->head
, 0);
682 focus
= xop2
->head
.cluster
.focus
;
683 kprintf("syncthr: update inode %p (%s)\n",
686 (char *)focus
->data
->ipdata
.filename
: "?"));
687 chain
= hammer2_inode_chain_and_parent(ip
, idx
,
689 HAMMER2_RESOLVE_ALWAYS
|
690 HAMMER2_RESOLVE_SHARED
);
692 KKASSERT(parent
!= NULL
);
693 nerror
= hammer2_sync_replace(
697 hammer2_chain_unlock(chain
);
698 hammer2_chain_drop(chain
);
699 hammer2_chain_unlock(parent
);
700 hammer2_chain_drop(parent
);
703 hammer2_xop_retire(&xop2
->head
, HAMMER2_XOPMASK_VOP
);
710 * Create a missing chain by copying the focus from another device.
712 * On entry *parentp and focus are both locked shared. The chain will be
713 * created and returned in *chainp also locked shared.
717 hammer2_sync_insert(hammer2_thread_t
*thr
,
718 hammer2_chain_t
**parentp
, hammer2_chain_t
**chainp
,
719 hammer2_tid_t mtid
, int idx
, hammer2_chain_t
*focus
)
721 hammer2_chain_t
*chain
;
723 int cache_index
= -1;
725 #if HAMMER2_SYNCHRO_DEBUG
726 if (hammer2_debug
& 1)
727 kprintf("insert rec par=%p/%d.%016jx slave %d %d.%016jx mod=%016jx\n",
729 (*parentp
)->bref
.type
,
730 (*parentp
)->bref
.key
,
732 focus
->bref
.type
, focus
->bref
.key
, mtid
);
736 * Parent requires an exclusive lock for the insertion.
737 * We must unlock the child to avoid deadlocks while
738 * relocking the parent.
741 hammer2_chain_unlock(*chainp
);
742 hammer2_chain_drop(*chainp
);
745 hammer2_chain_unlock(*parentp
);
746 hammer2_chain_lock(*parentp
, HAMMER2_RESOLVE_ALWAYS
);
749 * We must reissue the lookup to properly position (*parentp)
752 chain
= hammer2_chain_lookup(parentp
, &dummy
,
753 focus
->bref
.key
, focus
->bref
.key
,
755 HAMMER2_LOOKUP_NODIRECT
|
756 HAMMER2_LOOKUP_ALWAYS
);
757 KKASSERT(chain
== NULL
);
760 hammer2_chain_create(parentp
, &chain
,
761 thr
->pmp
, focus
->bref
.methods
,
762 focus
->bref
.key
, focus
->bref
.keybits
,
763 focus
->bref
.type
, focus
->bytes
,
765 hammer2_chain_modify(chain
, mtid
, 0, 0);
768 * Copy focus to new chain
771 /* type already set */
772 chain
->bref
.methods
= focus
->bref
.methods
;
773 /* keybits already set */
774 chain
->bref
.vradix
= focus
->bref
.vradix
;
775 /* mirror_tid set by flush */
776 KKASSERT(chain
->bref
.modify_tid
== mtid
);
777 chain
->bref
.flags
= focus
->bref
.flags
;
778 /* key already present */
779 /* check code will be recalculated */
784 switch(chain
->bref
.type
) {
785 case HAMMER2_BREF_TYPE_INODE
:
786 if ((focus
->data
->ipdata
.meta
.op_flags
&
787 HAMMER2_OPFLAG_DIRECTDATA
) == 0) {
788 /* do not copy block table */
789 bcopy(focus
->data
, chain
->data
,
790 offsetof(hammer2_inode_data_t
, u
));
793 /* fall through copy whole thing */
794 case HAMMER2_BREF_TYPE_DATA
:
795 bcopy(focus
->data
, chain
->data
, chain
->bytes
);
796 hammer2_chain_setcheck(chain
, chain
->data
);
803 hammer2_chain_unlock(chain
); /* unlock, leave ref */
804 *chainp
= chain
; /* will be returned locked */
807 * Avoid ordering deadlock when relocking shared.
809 hammer2_chain_unlock(*parentp
);
810 hammer2_chain_lock(*parentp
, HAMMER2_RESOLVE_SHARED
|
811 HAMMER2_RESOLVE_ALWAYS
);
812 hammer2_chain_lock(chain
, HAMMER2_RESOLVE_SHARED
|
813 HAMMER2_RESOLVE_ALWAYS
);
819 * Destroy an extranious chain.
821 * Both *parentp and *chainp are locked shared.
823 * On return, *chainp will be adjusted to point to the next element in the
824 * iteration and locked shared.
828 hammer2_sync_destroy(hammer2_thread_t
*thr
,
829 hammer2_chain_t
**parentp
, hammer2_chain_t
**chainp
,
830 hammer2_tid_t mtid
, int idx
)
832 hammer2_chain_t
*chain
;
833 hammer2_chain_t
*parent
;
834 hammer2_key_t key_next
;
835 hammer2_key_t save_key
;
836 int cache_index
= -1;
840 #if HAMMER2_SYNCHRO_DEBUG
841 if (hammer2_debug
& 1)
842 kprintf("destroy rec %p/%p slave %d %d.%016jx\n",
844 idx
, chain
->bref
.type
, chain
->bref
.key
);
847 save_key
= chain
->bref
.key
;
848 if (save_key
!= HAMMER2_KEY_MAX
)
852 * Try to avoid unnecessary I/O.
854 * XXX accounting not propagated up properly. We might have to do
855 * a RESOLVE_MAYBE here and pass 0 for the flags.
857 hammer2_chain_unlock(chain
); /* relock exclusive */
858 hammer2_chain_unlock(*parentp
);
859 hammer2_chain_lock(*parentp
, HAMMER2_RESOLVE_ALWAYS
);
860 hammer2_chain_lock(chain
, HAMMER2_RESOLVE_NEVER
);
862 hammer2_chain_delete(*parentp
, chain
, mtid
, HAMMER2_DELETE_PERMANENT
);
863 hammer2_chain_unlock(chain
);
864 hammer2_chain_drop(chain
);
865 chain
= NULL
; /* safety */
867 hammer2_chain_unlock(*parentp
); /* relock shared */
868 hammer2_chain_lock(*parentp
, HAMMER2_RESOLVE_SHARED
|
869 HAMMER2_RESOLVE_ALWAYS
);
870 *chainp
= hammer2_chain_lookup(&parent
, &key_next
,
871 save_key
, HAMMER2_KEY_MAX
,
873 HAMMER2_LOOKUP_SHARED
|
874 HAMMER2_LOOKUP_NODIRECT
|
875 HAMMER2_LOOKUP_NODATA
);
880 * cparent is locked exclusively, with an extra ref, cluster is not locked.
881 * Replace element [i] in the cluster.
885 hammer2_sync_replace(hammer2_thread_t
*thr
,
886 hammer2_chain_t
*parent
, hammer2_chain_t
*chain
,
887 hammer2_tid_t mtid
, int idx
,
888 hammer2_chain_t
*focus
, int isroot
)
893 #if HAMMER2_SYNCHRO_DEBUG
894 if (hammer2_debug
& 1)
895 kprintf("replace rec %p slave %d %d.%016jx mod=%016jx\n",
898 focus
->bref
.type
, focus
->bref
.key
, mtid
);
900 hammer2_chain_unlock(chain
);
901 hammer2_chain_lock(chain
, HAMMER2_RESOLVE_ALWAYS
);
902 if (chain
->bytes
!= focus
->bytes
) {
903 /* XXX what if compressed? */
904 nradix
= hammer2_getradix(chain
->bytes
);
905 hammer2_chain_resize(NULL
, parent
, chain
,
909 hammer2_chain_modify(chain
, mtid
, 0, 0);
910 otype
= chain
->bref
.type
;
911 chain
->bref
.type
= focus
->bref
.type
;
912 chain
->bref
.methods
= focus
->bref
.methods
;
913 chain
->bref
.keybits
= focus
->bref
.keybits
;
914 chain
->bref
.vradix
= focus
->bref
.vradix
;
915 /* mirror_tid updated by flush */
916 KKASSERT(mtid
== 0 || chain
->bref
.modify_tid
== mtid
);
917 chain
->bref
.flags
= focus
->bref
.flags
;
918 /* key already present */
919 /* check code will be recalculated */
925 switch(chain
->bref
.type
) {
926 case HAMMER2_BREF_TYPE_INODE
:
928 * Special case PFSROOTs, only limited changes can be made
929 * since the meta-data contains miscellanious distinguishing
933 chain
->data
->ipdata
.meta
.uflags
=
934 focus
->data
->ipdata
.meta
.uflags
;
935 chain
->data
->ipdata
.meta
.rmajor
=
936 focus
->data
->ipdata
.meta
.rmajor
;
937 chain
->data
->ipdata
.meta
.rminor
=
938 focus
->data
->ipdata
.meta
.rminor
;
939 chain
->data
->ipdata
.meta
.ctime
=
940 focus
->data
->ipdata
.meta
.ctime
;
941 chain
->data
->ipdata
.meta
.mtime
=
942 focus
->data
->ipdata
.meta
.mtime
;
943 chain
->data
->ipdata
.meta
.atime
=
944 focus
->data
->ipdata
.meta
.atime
;
946 chain
->data
->ipdata
.meta
.uid
=
947 focus
->data
->ipdata
.meta
.uid
;
948 chain
->data
->ipdata
.meta
.gid
=
949 focus
->data
->ipdata
.meta
.gid
;
950 chain
->data
->ipdata
.meta
.mode
=
951 focus
->data
->ipdata
.meta
.mode
;
952 chain
->data
->ipdata
.meta
.ncopies
=
953 focus
->data
->ipdata
.meta
.ncopies
;
954 chain
->data
->ipdata
.meta
.comp_algo
=
955 focus
->data
->ipdata
.meta
.comp_algo
;
956 chain
->data
->ipdata
.meta
.check_algo
=
957 focus
->data
->ipdata
.meta
.check_algo
;
958 chain
->data
->ipdata
.meta
.data_quota
=
959 focus
->data
->ipdata
.meta
.data_quota
;
960 chain
->data
->ipdata
.meta
.inode_quota
=
961 focus
->data
->ipdata
.meta
.inode_quota
;
964 * last snapshot tid controls overwrite
966 if (chain
->data
->ipdata
.meta
.pfs_lsnap_tid
<
967 focus
->data
->ipdata
.meta
.pfs_lsnap_tid
) {
968 chain
->data
->ipdata
.meta
.pfs_lsnap_tid
=
969 focus
->data
->ipdata
.meta
.pfs_lsnap_tid
;
972 hammer2_chain_setcheck(chain
, chain
->data
);
977 * Normal replacement.
979 if ((focus
->data
->ipdata
.meta
.op_flags
&
980 HAMMER2_OPFLAG_DIRECTDATA
) == 0) {
982 * If DIRECTDATA is transitioning to 0 or the old
983 * chain is not an inode we have to initialize
986 if (otype
!= HAMMER2_BREF_TYPE_INODE
||
987 (chain
->data
->ipdata
.meta
.op_flags
&
988 HAMMER2_OPFLAG_DIRECTDATA
)) {
989 kprintf("chain inode trans away from dd\n");
990 bzero(&chain
->data
->ipdata
.u
,
991 sizeof(chain
->data
->ipdata
.u
));
993 bcopy(focus
->data
, chain
->data
,
994 offsetof(hammer2_inode_data_t
, u
));
995 /* XXX setcheck on inode should not be needed */
996 hammer2_chain_setcheck(chain
, chain
->data
);
1000 case HAMMER2_BREF_TYPE_DATA
:
1001 bcopy(focus
->data
, chain
->data
, chain
->bytes
);
1002 hammer2_chain_setcheck(chain
, chain
->data
);
1009 hammer2_chain_unlock(chain
);
1010 hammer2_chain_lock(chain
, HAMMER2_RESOLVE_SHARED
|
1011 HAMMER2_RESOLVE_MAYBE
);