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_ondisk.c,v 1.35 2008/03/24 23:50:23 dillon Exp $
37 * Manage HAMMER's on-disk structures. These routines are primarily
38 * responsible for interfacing with the kernel's I/O subsystem and for
39 * managing in-memory structures.
43 #include <sys/fcntl.h>
44 #include <sys/nlookup.h>
48 static void hammer_free_volume(hammer_volume_t volume);
49 static int hammer_load_volume(hammer_volume_t volume);
50 static int hammer_load_buffer(hammer_buffer_t buffer, int isnew);
51 static int hammer_load_node(hammer_node_t node);
54 * Red-Black tree support for various structures
57 hammer_ino_rb_compare(hammer_inode_t ip1, hammer_inode_t ip2)
59 if (ip1->obj_id < ip2->obj_id)
61 if (ip1->obj_id > ip2->obj_id)
63 if (ip1->obj_asof < ip2->obj_asof)
65 if (ip1->obj_asof > ip2->obj_asof)
71 hammer_inode_info_cmp(hammer_inode_info_t info, hammer_inode_t ip)
73 if (info->obj_id < ip->obj_id)
75 if (info->obj_id > ip->obj_id)
77 if (info->obj_asof < ip->obj_asof)
79 if (info->obj_asof > ip->obj_asof)
85 hammer_vol_rb_compare(hammer_volume_t vol1, hammer_volume_t vol2)
87 if (vol1->vol_no < vol2->vol_no)
89 if (vol1->vol_no > vol2->vol_no)
95 hammer_buf_rb_compare(hammer_buffer_t buf1, hammer_buffer_t buf2)
97 if (buf1->zone2_offset < buf2->zone2_offset)
99 if (buf1->zone2_offset > buf2->zone2_offset)
105 hammer_nod_rb_compare(hammer_node_t node1, hammer_node_t node2)
107 if (node1->node_offset < node2->node_offset)
109 if (node1->node_offset > node2->node_offset)
115 * Note: The lookup function for hammer_ino_rb_tree winds up being named
116 * hammer_ino_rb_tree_RB_LOOKUP_INFO(root, info). The other lookup
117 * functions are normal, e.g. hammer_buf_rb_tree_RB_LOOKUP(root, zone2_offset).
119 RB_GENERATE(hammer_ino_rb_tree, hammer_inode, rb_node, hammer_ino_rb_compare);
120 RB_GENERATE_XLOOKUP(hammer_ino_rb_tree, INFO, hammer_inode, rb_node,
121 hammer_inode_info_cmp, hammer_inode_info_t);
122 RB_GENERATE2(hammer_vol_rb_tree, hammer_volume, rb_node,
123 hammer_vol_rb_compare, int32_t, vol_no);
124 RB_GENERATE2(hammer_buf_rb_tree, hammer_buffer, rb_node,
125 hammer_buf_rb_compare, hammer_off_t, zone2_offset);
126 RB_GENERATE2(hammer_nod_rb_tree, hammer_node, rb_node,
127 hammer_nod_rb_compare, hammer_off_t, node_offset);
129 /************************************************************************
131 ************************************************************************
133 * Load a HAMMER volume by name. Returns 0 on success or a positive error
134 * code on failure. Volumes must be loaded at mount time, get_volume() will
135 * not load a new volume.
137 * Calls made to hammer_load_volume() or single-threaded
140 hammer_install_volume(struct hammer_mount *hmp, const char *volname)
143 hammer_volume_t volume;
144 struct hammer_volume_ondisk *ondisk;
145 struct nlookupdata nd;
146 struct buf *bp = NULL;
152 ronly = ((mp->mnt_flag & MNT_RDONLY) ? 1 : 0);
155 * Allocate a volume structure
157 ++hammer_count_volumes;
158 volume = kmalloc(sizeof(*volume), M_HAMMER, M_WAITOK|M_ZERO);
159 volume->vol_name = kstrdup(volname, M_HAMMER);
161 hammer_io_init(&volume->io, HAMMER_STRUCTURE_VOLUME);
162 volume->io.offset = 0LL;
165 * Get the device vnode
167 error = nlookup_init(&nd, volume->vol_name, UIO_SYSSPACE, NLC_FOLLOW);
169 error = nlookup(&nd);
171 error = cache_vref(&nd.nl_nch, nd.nl_cred, &volume->devvp);
174 if (vn_isdisk(volume->devvp, &error)) {
175 error = vfs_mountedon(volume->devvp);
179 count_udev(volume->devvp->v_umajor, volume->devvp->v_uminor) > 0) {
183 vn_lock(volume->devvp, LK_EXCLUSIVE | LK_RETRY);
184 error = vinvalbuf(volume->devvp, V_SAVE, 0, 0);
186 error = VOP_OPEN(volume->devvp,
187 (ronly ? FREAD : FREAD|FWRITE),
190 vn_unlock(volume->devvp);
193 hammer_free_volume(volume);
196 volume->devvp->v_rdev->si_mountpoint = mp;
200 * Extract the volume number from the volume header and do various
203 error = bread(volume->devvp, 0LL, HAMMER_BUFSIZE, &bp);
206 ondisk = (void *)bp->b_data;
207 if (ondisk->vol_signature != HAMMER_FSBUF_VOLUME) {
208 kprintf("hammer_mount: volume %s has an invalid header\n",
213 volume->vol_no = ondisk->vol_no;
214 volume->buffer_base = ondisk->vol_buf_beg;
215 volume->vol_flags = ondisk->vol_flags;
216 volume->nblocks = ondisk->vol_nblocks;
217 volume->maxbuf_off = HAMMER_ENCODE_RAW_BUFFER(volume->vol_no,
218 ondisk->vol_buf_end - ondisk->vol_buf_beg);
219 RB_INIT(&volume->rb_bufs_root);
221 hmp->mp->mnt_stat.f_blocks += volume->nblocks;
223 if (RB_EMPTY(&hmp->rb_vols_root)) {
224 hmp->fsid = ondisk->vol_fsid;
225 } else if (bcmp(&hmp->fsid, &ondisk->vol_fsid, sizeof(uuid_t))) {
226 kprintf("hammer_mount: volume %s's fsid does not match "
227 "other volumes\n", volume->vol_name);
233 * Insert the volume structure into the red-black tree.
235 if (RB_INSERT(hammer_vol_rb_tree, &hmp->rb_vols_root, volume)) {
236 kprintf("hammer_mount: volume %s has a duplicate vol_no %d\n",
237 volume->vol_name, volume->vol_no);
242 * Set the root volume . HAMMER special cases rootvol the structure.
243 * We do not hold a ref because this would prevent related I/O
244 * from being flushed.
246 if (error == 0 && ondisk->vol_rootvol == ondisk->vol_no) {
247 hmp->rootvol = volume;
252 hmp->fsid_udev = dev2udev(vn_todev(volume->devvp));
258 /*vinvalbuf(volume->devvp, V_SAVE, 0, 0);*/
260 volume->devvp->v_rdev->si_mountpoint = NULL;
261 VOP_CLOSE(volume->devvp, ronly ? FREAD : FREAD|FWRITE);
262 hammer_free_volume(volume);
268 * Unload and free a HAMMER volume. Must return >= 0 to continue scan
269 * so returns -1 on failure.
272 hammer_unload_volume(hammer_volume_t volume, void *data __unused)
274 struct hammer_mount *hmp = volume->hmp;
275 int ronly = ((hmp->mp->mnt_flag & MNT_RDONLY) ? 1 : 0);
278 * Sync clusters, sync volume
281 hmp->mp->mnt_stat.f_blocks -= volume->nblocks;
284 * Clean up the root volume pointer, which is held unlocked in hmp.
286 if (hmp->rootvol == volume)
292 RB_SCAN(hammer_buf_rb_tree, &volume->rb_bufs_root, NULL,
293 hammer_unload_buffer, NULL);
296 * Release our buffer and flush anything left in the buffer cache.
298 volume->io.flush = 1;
299 volume->io.waitdep = 1;
300 hammer_io_release(&volume->io);
303 * There should be no references on the volume, no clusters, and
306 KKASSERT(volume->io.lock.refs == 0);
307 KKASSERT(RB_EMPTY(&volume->rb_bufs_root));
309 volume->ondisk = NULL;
311 if (volume->devvp->v_rdev &&
312 volume->devvp->v_rdev->si_mountpoint == hmp->mp
314 volume->devvp->v_rdev->si_mountpoint = NULL;
317 vinvalbuf(volume->devvp, 0, 0, 0);
318 VOP_CLOSE(volume->devvp, FREAD);
320 vinvalbuf(volume->devvp, V_SAVE, 0, 0);
321 VOP_CLOSE(volume->devvp, FREAD|FWRITE);
326 * Destroy the structure
328 RB_REMOVE(hammer_vol_rb_tree, &hmp->rb_vols_root, volume);
329 hammer_free_volume(volume);
335 hammer_free_volume(hammer_volume_t volume)
337 if (volume->vol_name) {
338 kfree(volume->vol_name, M_HAMMER);
339 volume->vol_name = NULL;
342 vrele(volume->devvp);
343 volume->devvp = NULL;
345 --hammer_count_volumes;
346 kfree(volume, M_HAMMER);
350 * Get a HAMMER volume. The volume must already exist.
353 hammer_get_volume(struct hammer_mount *hmp, int32_t vol_no, int *errorp)
355 struct hammer_volume *volume;
358 * Locate the volume structure
360 volume = RB_LOOKUP(hammer_vol_rb_tree, &hmp->rb_vols_root, vol_no);
361 if (volume == NULL) {
365 hammer_ref(&volume->io.lock);
368 * Deal with on-disk info
370 if (volume->ondisk == NULL || volume->io.loading) {
371 *errorp = hammer_load_volume(volume);
373 hammer_rel_volume(volume, 1);
383 hammer_ref_volume(hammer_volume_t volume)
387 hammer_ref(&volume->io.lock);
390 * Deal with on-disk info
392 if (volume->ondisk == NULL || volume->io.loading) {
393 error = hammer_load_volume(volume);
395 hammer_rel_volume(volume, 1);
403 hammer_get_root_volume(struct hammer_mount *hmp, int *errorp)
405 hammer_volume_t volume;
407 volume = hmp->rootvol;
408 KKASSERT(volume != NULL);
409 hammer_ref(&volume->io.lock);
412 * Deal with on-disk info
414 if (volume->ondisk == NULL || volume->io.loading) {
415 *errorp = hammer_load_volume(volume);
417 hammer_rel_volume(volume, 1);
427 * Load a volume's on-disk information. The volume must be referenced and
428 * not locked. We temporarily acquire an exclusive lock to interlock
429 * against releases or multiple get's.
432 hammer_load_volume(hammer_volume_t volume)
436 hammer_lock_ex(&volume->io.lock);
437 KKASSERT(volume->io.loading == 0);
438 ++volume->io.loading;
440 if (volume->ondisk == NULL) {
441 error = hammer_io_read(volume->devvp, &volume->io);
443 volume->ondisk = (void *)volume->io.bp->b_data;
447 --volume->io.loading;
448 hammer_unlock(&volume->io.lock);
453 * Release a volume. Call hammer_io_release on the last reference. We have
454 * to acquire an exclusive lock to interlock against volume->ondisk tests
455 * in hammer_load_volume(), and hammer_io_release() also expects an exclusive
458 * Volumes are not unloaded from memory during normal operation.
461 hammer_rel_volume(hammer_volume_t volume, int flush)
464 volume->io.flush = 1;
466 if (volume->io.lock.refs == 1) {
467 ++volume->io.loading;
468 hammer_lock_ex(&volume->io.lock);
469 if (volume->io.lock.refs == 1) {
470 volume->ondisk = NULL;
471 hammer_io_release(&volume->io);
473 --volume->io.loading;
474 hammer_unlock(&volume->io.lock);
476 hammer_unref(&volume->io.lock);
480 /************************************************************************
482 ************************************************************************
484 * Manage buffers. Currently all blockmap-backed zones are translated
485 * to zone-2 buffer offsets.
488 hammer_get_buffer(hammer_mount_t hmp, hammer_off_t buf_offset,
489 int isnew, int *errorp)
491 hammer_buffer_t buffer;
492 hammer_volume_t volume;
493 hammer_off_t zoneX_offset;
497 zoneX_offset = buf_offset;
498 zone = HAMMER_ZONE_DECODE(buf_offset);
499 if (zone >= HAMMER_ZONE_BTREE_INDEX) {
500 buf_offset = hammer_blockmap_lookup(hmp, buf_offset, errorp);
501 KKASSERT(*errorp == 0);
502 } else if (zone == HAMMER_ZONE_UNDO_INDEX) {
503 buf_offset = hammer_undo_lookup(hmp, buf_offset, errorp);
504 KKASSERT(*errorp == 0);
506 buf_offset &= ~HAMMER_BUFMASK64;
507 KKASSERT((buf_offset & HAMMER_ZONE_RAW_BUFFER) ==
508 HAMMER_ZONE_RAW_BUFFER);
509 vol_no = HAMMER_VOL_DECODE(buf_offset);
510 volume = hammer_get_volume(hmp, vol_no, errorp);
515 * NOTE: buf_offset and maxbuf_off are both full offset
518 KKASSERT(buf_offset < volume->maxbuf_off);
521 * Locate and lock the buffer structure, creating one if necessary.
524 buffer = RB_LOOKUP(hammer_buf_rb_tree, &volume->rb_bufs_root,
526 if (buffer == NULL) {
527 ++hammer_count_buffers;
528 buffer = kmalloc(sizeof(*buffer), M_HAMMER, M_WAITOK|M_ZERO);
529 buffer->zone2_offset = buf_offset;
530 buffer->volume = volume;
531 hammer_io_init(&buffer->io, HAMMER_STRUCTURE_BUFFER);
532 buffer->io.offset = volume->ondisk->vol_buf_beg +
533 (buf_offset & HAMMER_OFF_SHORT_MASK);
534 TAILQ_INIT(&buffer->clist);
535 hammer_ref(&buffer->io.lock);
538 * Insert the buffer into the RB tree and handle late
541 if (RB_INSERT(hammer_buf_rb_tree, &volume->rb_bufs_root, buffer)) {
542 hammer_unref(&buffer->io.lock);
543 --hammer_count_buffers;
544 kfree(buffer, M_HAMMER);
547 hammer_ref(&volume->io.lock);
549 hammer_ref(&buffer->io.lock);
553 * Cache the blockmap translation
555 if ((zoneX_offset & HAMMER_ZONE_RAW_BUFFER) != HAMMER_ZONE_RAW_BUFFER)
556 buffer->zoneX_offset = zoneX_offset;
559 * Deal with on-disk info
561 if (buffer->ondisk == NULL || buffer->io.loading) {
562 *errorp = hammer_load_buffer(buffer, isnew);
564 hammer_rel_buffer(buffer, 1);
570 hammer_rel_volume(volume, 0);
575 hammer_load_buffer(hammer_buffer_t buffer, int isnew)
577 hammer_volume_t volume;
581 * Load the buffer's on-disk info
583 volume = buffer->volume;
584 ++buffer->io.loading;
585 hammer_lock_ex(&buffer->io.lock);
587 if (buffer->ondisk == NULL) {
589 error = hammer_io_new(volume->devvp, &buffer->io);
591 error = hammer_io_read(volume->devvp, &buffer->io);
594 buffer->ondisk = (void *)buffer->io.bp->b_data;
596 error = hammer_io_new(volume->devvp, &buffer->io);
600 if (error == 0 && isnew) {
601 hammer_modify_buffer(NULL, buffer, NULL, 0);
602 /* additional initialization goes here */
604 --buffer->io.loading;
605 hammer_unlock(&buffer->io.lock);
610 * NOTE: Called from RB_SCAN, must return >= 0 for scan to continue.
613 hammer_unload_buffer(hammer_buffer_t buffer, void *data __unused)
615 hammer_ref(&buffer->io.lock);
616 hammer_flush_buffer_nodes(buffer);
617 KKASSERT(buffer->io.lock.refs == 1);
618 hammer_rel_buffer(buffer, 2);
623 * Reference a buffer that is either already referenced or via a specially
624 * handled pointer (aka cursor->buffer).
627 hammer_ref_buffer(hammer_buffer_t buffer)
631 hammer_ref(&buffer->io.lock);
632 if (buffer->ondisk == NULL || buffer->io.loading) {
633 error = hammer_load_buffer(buffer, 0);
635 hammer_rel_buffer(buffer, 1);
637 * NOTE: buffer pointer can become stale after
648 * Release a buffer. We have to deal with several places where
649 * another thread can ref the buffer.
651 * Only destroy the structure itself if the related buffer cache buffer
652 * was disassociated from it. This ties the management of the structure
653 * to the buffer cache subsystem. buffer->ondisk determines whether the
654 * embedded io is referenced or not.
657 hammer_rel_buffer(hammer_buffer_t buffer, int flush)
659 hammer_volume_t volume;
663 buffer->io.flush = 1;
665 if (buffer->io.lock.refs == 1) {
666 ++buffer->io.loading; /* force interlock check */
667 hammer_lock_ex(&buffer->io.lock);
668 if (buffer->io.lock.refs == 1) {
669 hammer_io_release(&buffer->io);
670 hammer_flush_buffer_nodes(buffer);
671 KKASSERT(TAILQ_EMPTY(&buffer->clist));
673 if (buffer->io.bp == NULL &&
674 buffer->io.lock.refs == 1) {
678 volume = buffer->volume;
679 RB_REMOVE(hammer_buf_rb_tree,
680 &volume->rb_bufs_root, buffer);
681 buffer->volume = NULL; /* sanity */
682 hammer_rel_volume(volume, 0);
686 --buffer->io.loading;
687 hammer_unlock(&buffer->io.lock);
689 hammer_unref(&buffer->io.lock);
692 --hammer_count_buffers;
693 kfree(buffer, M_HAMMER);
698 * Remove the zoneX translation cache for a buffer given its zone-2 offset.
701 hammer_uncache_buffer(hammer_mount_t hmp, hammer_off_t buf_offset)
703 hammer_volume_t volume;
704 hammer_buffer_t buffer;
708 buf_offset &= ~HAMMER_BUFMASK64;
709 KKASSERT((buf_offset & HAMMER_ZONE_RAW_BUFFER) ==
710 HAMMER_ZONE_RAW_BUFFER);
711 vol_no = HAMMER_VOL_DECODE(buf_offset);
712 volume = hammer_get_volume(hmp, vol_no, &error);
713 KKASSERT(volume != 0);
714 KKASSERT(buf_offset < volume->maxbuf_off);
716 buffer = RB_LOOKUP(hammer_buf_rb_tree, &volume->rb_bufs_root,
719 buffer->zoneX_offset = 0;
720 hammer_rel_volume(volume, 0);
724 * Access the filesystem buffer containing the specified hammer offset.
725 * buf_offset is a conglomeration of the volume number and vol_buf_beg
726 * relative buffer offset. It must also have bit 55 set to be valid.
727 * (see hammer_off_t in hammer_disk.h).
729 * Any prior buffer in *bufferp will be released and replaced by the
733 hammer_bread(hammer_mount_t hmp, hammer_off_t buf_offset, int *errorp,
734 struct hammer_buffer **bufferp)
736 hammer_buffer_t buffer;
737 int32_t xoff = (int32_t)buf_offset & HAMMER_BUFMASK;
739 buf_offset &= ~HAMMER_BUFMASK64;
740 KKASSERT((buf_offset & HAMMER_OFF_ZONE_MASK) != 0);
743 if (buffer == NULL || (buffer->zone2_offset != buf_offset &&
744 buffer->zoneX_offset != buf_offset)) {
746 hammer_rel_buffer(buffer, 0);
747 buffer = hammer_get_buffer(hmp, buf_offset, 0, errorp);
754 * Return a pointer to the buffer data.
759 return((char *)buffer->ondisk + xoff);
763 * Access the filesystem buffer containing the specified hammer offset.
764 * No disk read operation occurs. The result buffer may contain garbage.
766 * Any prior buffer in *bufferp will be released and replaced by the
770 hammer_bnew(hammer_mount_t hmp, hammer_off_t buf_offset, int *errorp,
771 struct hammer_buffer **bufferp)
773 hammer_buffer_t buffer;
774 int32_t xoff = (int32_t)buf_offset & HAMMER_BUFMASK;
776 buf_offset &= ~HAMMER_BUFMASK64;
779 if (buffer == NULL || (buffer->zone2_offset != buf_offset &&
780 buffer->zoneX_offset != buf_offset)) {
782 hammer_rel_buffer(buffer, 0);
783 buffer = hammer_get_buffer(hmp, buf_offset, 1, errorp);
790 * Return a pointer to the buffer data.
795 return((char *)buffer->ondisk + xoff);
798 /************************************************************************
800 ************************************************************************
802 * Manage B-Tree nodes. B-Tree nodes represent the primary indexing
803 * method used by the HAMMER filesystem.
805 * Unlike other HAMMER structures, a hammer_node can be PASSIVELY
806 * associated with its buffer, and will only referenced the buffer while
807 * the node itself is referenced.
809 * A hammer_node can also be passively associated with other HAMMER
810 * structures, such as inodes, while retaining 0 references. These
811 * associations can be cleared backwards using a pointer-to-pointer in
814 * This allows the HAMMER implementation to cache hammer_nodes long-term
815 * and short-cut a great deal of the infrastructure's complexity. In
816 * most cases a cached node can be reacquired without having to dip into
817 * either the buffer or cluster management code.
819 * The caller must pass a referenced cluster on call and will retain
820 * ownership of the reference on return. The node will acquire its own
821 * additional references, if necessary.
824 hammer_get_node(hammer_mount_t hmp, hammer_off_t node_offset, int *errorp)
828 KKASSERT((node_offset & HAMMER_OFF_ZONE_MASK) == HAMMER_ZONE_BTREE);
831 * Locate the structure, allocating one if necessary.
834 node = RB_LOOKUP(hammer_nod_rb_tree, &hmp->rb_nods_root, node_offset);
836 ++hammer_count_nodes;
837 node = kmalloc(sizeof(*node), M_HAMMER, M_WAITOK|M_ZERO);
838 node->node_offset = node_offset;
840 if (RB_INSERT(hammer_nod_rb_tree, &hmp->rb_nods_root, node)) {
841 --hammer_count_nodes;
842 kfree(node, M_HAMMER);
846 hammer_ref(&node->lock);
847 *errorp = hammer_load_node(node);
849 hammer_rel_node(node);
856 * Reference an already-referenced node.
859 hammer_ref_node(hammer_node_t node)
863 KKASSERT(node->lock.refs > 0);
864 hammer_ref(&node->lock);
865 if ((error = hammer_load_node(node)) != 0)
866 hammer_rel_node(node);
871 * Load a node's on-disk data reference.
874 hammer_load_node(hammer_node_t node)
876 hammer_buffer_t buffer;
882 hammer_lock_ex(&node->lock);
883 if (node->ondisk == NULL) {
885 * This is a little confusing but the jist is that
886 * node->buffer determines whether the node is on
887 * the buffer's clist and node->ondisk determines
888 * whether the buffer is referenced.
890 if ((buffer = node->buffer) != NULL) {
891 error = hammer_ref_buffer(buffer);
893 buffer = hammer_get_buffer(node->hmp,
894 node->node_offset, 0,
897 KKASSERT(error == 0);
898 TAILQ_INSERT_TAIL(&buffer->clist,
900 node->buffer = buffer;
904 node->ondisk = (void *)((char *)buffer->ondisk +
905 (node->node_offset & HAMMER_BUFMASK));
908 hammer_unlock(&node->lock);
913 * Safely reference a node, interlock against flushes via the IO subsystem.
916 hammer_ref_node_safe(struct hammer_mount *hmp, struct hammer_node **cache,
921 if ((node = *cache) != NULL)
922 hammer_ref(&node->lock);
924 *errorp = hammer_load_node(node);
926 hammer_rel_node(node);
936 * Release a hammer_node. On the last release the node dereferences
937 * its underlying buffer and may or may not be destroyed.
940 hammer_rel_node(hammer_node_t node)
942 hammer_buffer_t buffer;
945 * If this isn't the last ref just decrement the ref count and
948 if (node->lock.refs > 1) {
949 hammer_unref(&node->lock);
954 * If there is no ondisk info or no buffer the node failed to load,
955 * remove the last reference and destroy the node.
957 if (node->ondisk == NULL) {
958 hammer_unref(&node->lock);
959 hammer_flush_node(node);
960 /* node is stale now */
965 * Do final cleanups and then either destroy the node and leave it
966 * passively cached. The buffer reference is removed regardless.
968 buffer = node->buffer;
971 if ((node->flags & HAMMER_NODE_FLUSH) == 0) {
972 hammer_unref(&node->lock);
973 hammer_rel_buffer(buffer, 0);
980 hammer_unref(&node->lock);
981 hammer_flush_node(node);
983 hammer_rel_buffer(buffer, 0);
991 hammer_delete_node(hammer_transaction_t trans, hammer_node_t node)
993 node->flags |= HAMMER_NODE_DELETED;
994 hammer_blockmap_free(trans, node->node_offset, sizeof(*node->ondisk));
998 * Passively cache a referenced hammer_node in *cache. The caller may
999 * release the node on return.
1002 hammer_cache_node(hammer_node_t node, struct hammer_node **cache)
1007 * If the node is being deleted, don't cache it!
1009 if (node->flags & HAMMER_NODE_DELETED)
1013 * Cache the node. If we previously cached a different node we
1014 * have to give HAMMER a chance to destroy it.
1017 if (node->cache1 != cache) {
1018 if (node->cache2 != cache) {
1019 if ((old = *cache) != NULL) {
1020 KKASSERT(node->lock.refs != 0);
1021 hammer_uncache_node(cache);
1025 *node->cache2 = NULL;
1026 node->cache2 = node->cache1;
1027 node->cache1 = cache;
1030 struct hammer_node **tmp;
1032 node->cache1 = node->cache2;
1039 hammer_uncache_node(struct hammer_node **cache)
1043 if ((node = *cache) != NULL) {
1045 if (node->cache1 == cache) {
1046 node->cache1 = node->cache2;
1047 node->cache2 = NULL;
1048 } else if (node->cache2 == cache) {
1049 node->cache2 = NULL;
1051 panic("hammer_uncache_node: missing cache linkage");
1053 if (node->cache1 == NULL && node->cache2 == NULL)
1054 hammer_flush_node(node);
1059 * Remove a node's cache references and destroy the node if it has no
1060 * other references or backing store.
1063 hammer_flush_node(hammer_node_t node)
1065 hammer_buffer_t buffer;
1068 *node->cache1 = NULL;
1070 *node->cache2 = NULL;
1071 if (node->lock.refs == 0 && node->ondisk == NULL) {
1072 RB_REMOVE(hammer_nod_rb_tree, &node->hmp->rb_nods_root, node);
1073 if ((buffer = node->buffer) != NULL) {
1074 node->buffer = NULL;
1075 TAILQ_REMOVE(&buffer->clist, node, entry);
1076 /* buffer is unreferenced because ondisk is NULL */
1078 --hammer_count_nodes;
1079 kfree(node, M_HAMMER);
1084 * Flush passively cached B-Tree nodes associated with this buffer.
1085 * This is only called when the buffer is about to be destroyed, so
1086 * none of the nodes should have any references.
1089 hammer_flush_buffer_nodes(hammer_buffer_t buffer)
1093 while ((node = TAILQ_FIRST(&buffer->clist)) != NULL) {
1094 KKASSERT(node->lock.refs == 0 && node->ondisk == NULL);
1095 hammer_ref(&node->lock);
1096 node->flags |= HAMMER_NODE_FLUSH;
1097 hammer_rel_node(node);
1102 /************************************************************************
1104 ************************************************************************/
1107 * Allocate a B-Tree node.
1110 hammer_alloc_btree(hammer_transaction_t trans, int *errorp)
1112 hammer_buffer_t buffer = NULL;
1113 hammer_node_t node = NULL;
1114 hammer_off_t node_offset;
1116 node_offset = hammer_blockmap_alloc(trans, HAMMER_ZONE_BTREE_INDEX,
1117 sizeof(struct hammer_node_ondisk),
1120 node = hammer_get_node(trans->hmp, node_offset, errorp);
1121 hammer_modify_node_noundo(trans, node);
1122 bzero(node->ondisk, sizeof(*node->ondisk));
1125 hammer_rel_buffer(buffer, 0);
1130 * The returned buffers are already appropriately marked as being modified.
1131 * If the caller marks them again unnecessary undo records may be generated.
1133 * In-band data is indicated by data_bufferp == NULL. Pass a data_len of 0
1134 * for zero-fill (caller modifies data_len afterwords).
1137 hammer_alloc_record(hammer_transaction_t trans,
1138 hammer_off_t *rec_offp, u_int16_t rec_type,
1139 struct hammer_buffer **rec_bufferp,
1140 int32_t data_len, void **datap,
1141 struct hammer_buffer **data_bufferp, int *errorp)
1143 hammer_record_ondisk_t rec;
1144 hammer_off_t rec_offset;
1145 hammer_off_t data_offset;
1152 * Allocate the record
1154 rec_offset = hammer_blockmap_alloc(trans, HAMMER_ZONE_RECORD_INDEX,
1155 HAMMER_RECORD_SIZE, errorp);
1163 if (data_bufferp == NULL) {
1165 case HAMMER_RECTYPE_DATA:
1166 reclen = offsetof(struct hammer_data_record,
1169 case HAMMER_RECTYPE_DIRENTRY:
1170 reclen = offsetof(struct hammer_entry_record,
1174 panic("hammer_alloc_record: illegal "
1180 KKASSERT(reclen + data_len <= HAMMER_RECORD_SIZE);
1181 data_offset = rec_offset + reclen;
1182 } else if (data_len < HAMMER_BUFSIZE) {
1183 data_offset = hammer_blockmap_alloc(trans,
1184 HAMMER_ZONE_SMALL_DATA_INDEX,
1187 data_offset = hammer_blockmap_alloc(trans,
1188 HAMMER_ZONE_LARGE_DATA_INDEX,
1195 hammer_blockmap_free(trans, rec_offset, HAMMER_RECORD_SIZE);
1200 * Basic return values.
1202 *rec_offp = rec_offset;
1203 rec = hammer_bread(trans->hmp, rec_offset, errorp, rec_bufferp);
1204 hammer_modify_buffer(trans, *rec_bufferp, NULL, 0);
1205 bzero(rec, sizeof(*rec));
1206 KKASSERT(*errorp == 0);
1207 rec->base.data_off = data_offset;
1208 rec->base.data_len = data_len;
1212 *datap = hammer_bread(trans->hmp, data_offset, errorp,
1214 KKASSERT(*errorp == 0);
1215 hammer_modify_buffer(trans, *data_bufferp, NULL, 0);
1219 } else if (data_len) {
1220 KKASSERT(data_offset + data_len - rec_offset <=
1221 HAMMER_RECORD_SIZE);
1223 *datap = (void *)((char *)rec +
1224 (int32_t)(data_offset - rec_offset));
1227 KKASSERT(datap == NULL);
1229 KKASSERT(*errorp == 0);
1234 hammer_alloc_data(hammer_transaction_t trans, int32_t data_len,
1235 hammer_off_t *data_offsetp,
1236 struct hammer_buffer **data_bufferp, int *errorp)
1244 if (data_len < HAMMER_BUFSIZE) {
1245 *data_offsetp = hammer_blockmap_alloc(trans,
1246 HAMMER_ZONE_SMALL_DATA_INDEX,
1249 *data_offsetp = hammer_blockmap_alloc(trans,
1250 HAMMER_ZONE_LARGE_DATA_INDEX,
1256 if (*errorp == 0 && data_bufferp) {
1258 data = hammer_bread(trans->hmp, *data_offsetp, errorp,
1260 KKASSERT(*errorp == 0);
1261 hammer_modify_buffer(trans, *data_bufferp, NULL, 0);
1268 KKASSERT(*errorp == 0);
1274 * Sync dirty buffers to the media
1277 static int hammer_sync_scan1(struct mount *mp, struct vnode *vp, void *data);
1278 static int hammer_sync_scan2(struct mount *mp, struct vnode *vp, void *data);
1281 hammer_sync_hmp(hammer_mount_t hmp, int waitfor)
1283 struct hammer_sync_info info;
1286 info.waitfor = waitfor;
1288 vmntvnodescan(hmp->mp, VMSC_GETVP|VMSC_NOWAIT,
1289 hammer_sync_scan1, hammer_sync_scan2, &info);
1291 RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
1292 hammer_sync_volume, &info);
1297 hammer_sync_scan1(struct mount *mp, struct vnode *vp, void *data)
1299 struct hammer_inode *ip;
1302 if (vp->v_type == VNON || ip == NULL ||
1303 ((ip->flags & HAMMER_INODE_MODMASK) == 0 &&
1304 RB_EMPTY(&vp->v_rbdirty_tree))) {
1311 hammer_sync_scan2(struct mount *mp, struct vnode *vp, void *data)
1313 struct hammer_sync_info *info = data;
1314 struct hammer_inode *ip;
1318 if (vp->v_type == VNON || vp->v_type == VBAD ||
1319 ((ip->flags & HAMMER_INODE_MODMASK) == 0 &&
1320 RB_EMPTY(&vp->v_rbdirty_tree))) {
1323 error = VOP_FSYNC(vp, info->waitfor);
1325 info->error = error;
1330 hammer_sync_volume(hammer_volume_t volume, void *data)
1332 struct hammer_sync_info *info = data;
1334 hammer_ref(&volume->io.lock);
1335 RB_SCAN(hammer_buf_rb_tree, &volume->rb_bufs_root, NULL,
1336 hammer_sync_buffer, info);
1337 hammer_rel_volume(volume, 1);
1342 hammer_sync_buffer(hammer_buffer_t buffer, void *data __unused)
1344 hammer_ref(&buffer->io.lock);
1345 hammer_rel_buffer(buffer, 1);