2 * Copyright (c) 2007-2008 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 1.37 2008/04/24 21:20:33 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 ++volume->io.loading;
437 hammer_lock_ex(&volume->io.lock);
439 if (volume->ondisk == NULL) {
440 error = hammer_io_read(volume->devvp, &volume->io);
442 volume->ondisk = (void *)volume->io.bp->b_data;
446 --volume->io.loading;
447 hammer_unlock(&volume->io.lock);
452 * Release a volume. Call hammer_io_release on the last reference. We have
453 * to acquire an exclusive lock to interlock against volume->ondisk tests
454 * in hammer_load_volume(), and hammer_io_release() also expects an exclusive
457 * Volumes are not unloaded from memory during normal operation.
460 hammer_rel_volume(hammer_volume_t volume, int flush)
463 volume->io.flush = 1;
465 if (volume->io.lock.refs == 1) {
466 ++volume->io.loading;
467 hammer_lock_ex(&volume->io.lock);
468 if (volume->io.lock.refs == 1) {
469 volume->ondisk = NULL;
470 hammer_io_release(&volume->io);
472 --volume->io.loading;
473 hammer_unlock(&volume->io.lock);
475 hammer_unref(&volume->io.lock);
479 /************************************************************************
481 ************************************************************************
483 * Manage buffers. Currently all blockmap-backed zones are translated
484 * to zone-2 buffer offsets.
487 hammer_get_buffer(hammer_mount_t hmp, hammer_off_t buf_offset,
488 int isnew, int *errorp)
490 hammer_buffer_t buffer;
491 hammer_volume_t volume;
492 hammer_off_t zoneX_offset;
496 zoneX_offset = buf_offset;
497 zone = HAMMER_ZONE_DECODE(buf_offset);
498 if (zone >= HAMMER_ZONE_BTREE_INDEX) {
499 buf_offset = hammer_blockmap_lookup(hmp, buf_offset, errorp);
500 KKASSERT(*errorp == 0);
501 } else if (zone == HAMMER_ZONE_UNDO_INDEX) {
502 buf_offset = hammer_undo_lookup(hmp, buf_offset, errorp);
503 KKASSERT(*errorp == 0);
505 buf_offset &= ~HAMMER_BUFMASK64;
506 KKASSERT((buf_offset & HAMMER_ZONE_RAW_BUFFER) ==
507 HAMMER_ZONE_RAW_BUFFER);
508 vol_no = HAMMER_VOL_DECODE(buf_offset);
509 volume = hammer_get_volume(hmp, vol_no, errorp);
514 * NOTE: buf_offset and maxbuf_off are both full offset
517 KKASSERT(buf_offset < volume->maxbuf_off);
520 * Locate and lock the buffer structure, creating one if necessary.
523 buffer = RB_LOOKUP(hammer_buf_rb_tree, &volume->rb_bufs_root,
525 if (buffer == NULL) {
526 ++hammer_count_buffers;
527 buffer = kmalloc(sizeof(*buffer), M_HAMMER, M_WAITOK|M_ZERO);
528 buffer->zone2_offset = buf_offset;
529 buffer->volume = volume;
530 hammer_io_init(&buffer->io, HAMMER_STRUCTURE_BUFFER);
531 buffer->io.offset = volume->ondisk->vol_buf_beg +
532 (buf_offset & HAMMER_OFF_SHORT_MASK);
533 TAILQ_INIT(&buffer->clist);
534 hammer_ref(&buffer->io.lock);
537 * Insert the buffer into the RB tree and handle late
540 if (RB_INSERT(hammer_buf_rb_tree, &volume->rb_bufs_root, buffer)) {
541 hammer_unref(&buffer->io.lock);
542 --hammer_count_buffers;
543 kfree(buffer, M_HAMMER);
546 hammer_ref(&volume->io.lock);
548 hammer_ref(&buffer->io.lock);
552 * Cache the blockmap translation
554 if ((zoneX_offset & HAMMER_ZONE_RAW_BUFFER) != HAMMER_ZONE_RAW_BUFFER)
555 buffer->zoneX_offset = zoneX_offset;
558 * Deal with on-disk info
560 if (buffer->ondisk == NULL || buffer->io.loading) {
561 *errorp = hammer_load_buffer(buffer, isnew);
563 hammer_rel_buffer(buffer, 1);
569 hammer_rel_volume(volume, 0);
574 hammer_load_buffer(hammer_buffer_t buffer, int isnew)
576 hammer_volume_t volume;
580 * Load the buffer's on-disk info
582 volume = buffer->volume;
583 ++buffer->io.loading;
584 hammer_lock_ex(&buffer->io.lock);
586 if (buffer->ondisk == NULL) {
588 error = hammer_io_new(volume->devvp, &buffer->io);
590 error = hammer_io_read(volume->devvp, &buffer->io);
593 buffer->ondisk = (void *)buffer->io.bp->b_data;
595 error = hammer_io_new(volume->devvp, &buffer->io);
599 if (error == 0 && isnew) {
600 hammer_modify_buffer(NULL, buffer, NULL, 0);
601 /* additional initialization goes here */
603 --buffer->io.loading;
604 hammer_unlock(&buffer->io.lock);
609 * NOTE: Called from RB_SCAN, must return >= 0 for scan to continue.
612 hammer_unload_buffer(hammer_buffer_t buffer, void *data __unused)
614 hammer_ref(&buffer->io.lock);
615 hammer_flush_buffer_nodes(buffer);
616 KKASSERT(buffer->io.lock.refs == 1);
617 hammer_rel_buffer(buffer, 2);
622 * Reference a buffer that is either already referenced or via a specially
623 * handled pointer (aka cursor->buffer).
626 hammer_ref_buffer(hammer_buffer_t buffer)
630 hammer_ref(&buffer->io.lock);
631 if (buffer->ondisk == NULL || buffer->io.loading) {
632 error = hammer_load_buffer(buffer, 0);
634 hammer_rel_buffer(buffer, 1);
636 * NOTE: buffer pointer can become stale after
647 * Release a buffer. We have to deal with several places where
648 * another thread can ref the buffer.
650 * Only destroy the structure itself if the related buffer cache buffer
651 * was disassociated from it. This ties the management of the structure
652 * to the buffer cache subsystem. buffer->ondisk determines whether the
653 * embedded io is referenced or not.
656 hammer_rel_buffer(hammer_buffer_t buffer, int flush)
658 hammer_volume_t volume;
662 buffer->io.flush = 1;
664 if (buffer->io.lock.refs == 1) {
665 ++buffer->io.loading; /* force interlock check */
666 hammer_lock_ex(&buffer->io.lock);
667 if (buffer->io.lock.refs == 1) {
668 hammer_io_release(&buffer->io);
669 hammer_flush_buffer_nodes(buffer);
670 KKASSERT(TAILQ_EMPTY(&buffer->clist));
672 if (buffer->io.bp == NULL &&
673 buffer->io.lock.refs == 1) {
677 volume = buffer->volume;
678 RB_REMOVE(hammer_buf_rb_tree,
679 &volume->rb_bufs_root, buffer);
680 buffer->volume = NULL; /* sanity */
681 hammer_rel_volume(volume, 0);
685 --buffer->io.loading;
686 hammer_unlock(&buffer->io.lock);
688 hammer_unref(&buffer->io.lock);
691 --hammer_count_buffers;
692 kfree(buffer, M_HAMMER);
697 * Remove the zoneX translation cache for a buffer given its zone-2 offset.
700 hammer_uncache_buffer(hammer_mount_t hmp, hammer_off_t buf_offset)
702 hammer_volume_t volume;
703 hammer_buffer_t buffer;
707 buf_offset &= ~HAMMER_BUFMASK64;
708 KKASSERT((buf_offset & HAMMER_ZONE_RAW_BUFFER) ==
709 HAMMER_ZONE_RAW_BUFFER);
710 vol_no = HAMMER_VOL_DECODE(buf_offset);
711 volume = hammer_get_volume(hmp, vol_no, &error);
712 KKASSERT(volume != 0);
713 KKASSERT(buf_offset < volume->maxbuf_off);
715 buffer = RB_LOOKUP(hammer_buf_rb_tree, &volume->rb_bufs_root,
718 buffer->zoneX_offset = 0;
719 hammer_rel_volume(volume, 0);
723 * Access the filesystem buffer containing the specified hammer offset.
724 * buf_offset is a conglomeration of the volume number and vol_buf_beg
725 * relative buffer offset. It must also have bit 55 set to be valid.
726 * (see hammer_off_t in hammer_disk.h).
728 * Any prior buffer in *bufferp will be released and replaced by the
732 hammer_bread(hammer_mount_t hmp, hammer_off_t buf_offset, int *errorp,
733 struct hammer_buffer **bufferp)
735 hammer_buffer_t buffer;
736 int32_t xoff = (int32_t)buf_offset & HAMMER_BUFMASK;
738 buf_offset &= ~HAMMER_BUFMASK64;
739 KKASSERT((buf_offset & HAMMER_OFF_ZONE_MASK) != 0);
742 if (buffer == NULL || (buffer->zone2_offset != buf_offset &&
743 buffer->zoneX_offset != buf_offset)) {
745 hammer_rel_buffer(buffer, 0);
746 buffer = hammer_get_buffer(hmp, buf_offset, 0, errorp);
753 * Return a pointer to the buffer data.
758 return((char *)buffer->ondisk + xoff);
762 * Access the filesystem buffer containing the specified hammer offset.
763 * No disk read operation occurs. The result buffer may contain garbage.
765 * Any prior buffer in *bufferp will be released and replaced by the
769 hammer_bnew(hammer_mount_t hmp, hammer_off_t buf_offset, int *errorp,
770 struct hammer_buffer **bufferp)
772 hammer_buffer_t buffer;
773 int32_t xoff = (int32_t)buf_offset & HAMMER_BUFMASK;
775 buf_offset &= ~HAMMER_BUFMASK64;
778 if (buffer == NULL || (buffer->zone2_offset != buf_offset &&
779 buffer->zoneX_offset != buf_offset)) {
781 hammer_rel_buffer(buffer, 0);
782 buffer = hammer_get_buffer(hmp, buf_offset, 1, errorp);
789 * Return a pointer to the buffer data.
794 return((char *)buffer->ondisk + xoff);
797 /************************************************************************
799 ************************************************************************
801 * Manage B-Tree nodes. B-Tree nodes represent the primary indexing
802 * method used by the HAMMER filesystem.
804 * Unlike other HAMMER structures, a hammer_node can be PASSIVELY
805 * associated with its buffer, and will only referenced the buffer while
806 * the node itself is referenced.
808 * A hammer_node can also be passively associated with other HAMMER
809 * structures, such as inodes, while retaining 0 references. These
810 * associations can be cleared backwards using a pointer-to-pointer in
813 * This allows the HAMMER implementation to cache hammer_nodes long-term
814 * and short-cut a great deal of the infrastructure's complexity. In
815 * most cases a cached node can be reacquired without having to dip into
816 * either the buffer or cluster management code.
818 * The caller must pass a referenced cluster on call and will retain
819 * ownership of the reference on return. The node will acquire its own
820 * additional references, if necessary.
823 hammer_get_node(hammer_mount_t hmp, hammer_off_t node_offset, int *errorp)
827 KKASSERT((node_offset & HAMMER_OFF_ZONE_MASK) == HAMMER_ZONE_BTREE);
830 * Locate the structure, allocating one if necessary.
833 node = RB_LOOKUP(hammer_nod_rb_tree, &hmp->rb_nods_root, node_offset);
835 ++hammer_count_nodes;
836 node = kmalloc(sizeof(*node), M_HAMMER, M_WAITOK|M_ZERO);
837 node->node_offset = node_offset;
839 if (RB_INSERT(hammer_nod_rb_tree, &hmp->rb_nods_root, node)) {
840 --hammer_count_nodes;
841 kfree(node, M_HAMMER);
845 hammer_ref(&node->lock);
849 *errorp = hammer_load_node(node);
851 hammer_rel_node(node);
858 * Reference an already-referenced node.
861 hammer_ref_node(hammer_node_t node)
863 KKASSERT(node->lock.refs > 0 && node->ondisk != NULL);
864 hammer_ref(&node->lock);
868 * Load a node's on-disk data reference.
871 hammer_load_node(hammer_node_t node)
873 hammer_buffer_t buffer;
878 hammer_lock_ex(&node->lock);
879 if (node->ondisk == NULL) {
881 * This is a little confusing but the jist is that
882 * node->buffer determines whether the node is on
883 * the buffer's clist and node->ondisk determines
884 * whether the buffer is referenced.
886 * We could be racing a buffer release, in which case
887 * node->buffer may become NULL while we are blocked
888 * referencing the buffer.
890 if ((buffer = node->buffer) != NULL) {
891 error = hammer_ref_buffer(buffer);
892 if (error == 0 && node->buffer == NULL) {
893 TAILQ_INSERT_TAIL(&buffer->clist,
895 node->buffer = buffer;
898 buffer = hammer_get_buffer(node->hmp,
899 node->node_offset, 0,
902 KKASSERT(error == 0);
903 TAILQ_INSERT_TAIL(&buffer->clist,
905 node->buffer = buffer;
909 node->ondisk = (void *)((char *)buffer->ondisk +
910 (node->node_offset & HAMMER_BUFMASK));
914 hammer_unlock(&node->lock);
919 * Safely reference a node, interlock against flushes via the IO subsystem.
922 hammer_ref_node_safe(struct hammer_mount *hmp, struct hammer_node **cache,
929 hammer_ref(&node->lock);
933 *errorp = hammer_load_node(node);
935 hammer_rel_node(node);
945 * Release a hammer_node. On the last release the node dereferences
946 * its underlying buffer and may or may not be destroyed.
949 hammer_rel_node(hammer_node_t node)
951 hammer_buffer_t buffer;
954 * If this isn't the last ref just decrement the ref count and
957 if (node->lock.refs > 1) {
958 hammer_unref(&node->lock);
963 * If there is no ondisk info or no buffer the node failed to load,
964 * remove the last reference and destroy the node.
966 if (node->ondisk == NULL) {
967 hammer_unref(&node->lock);
968 hammer_flush_node(node);
969 /* node is stale now */
974 * Do final cleanups and then either destroy the node and leave it
975 * passively cached. The buffer reference is removed regardless.
977 buffer = node->buffer;
980 if ((node->flags & HAMMER_NODE_FLUSH) == 0) {
981 hammer_unref(&node->lock);
982 hammer_rel_buffer(buffer, 0);
989 hammer_unref(&node->lock);
990 hammer_flush_node(node);
992 hammer_rel_buffer(buffer, 0);
1000 hammer_delete_node(hammer_transaction_t trans, hammer_node_t node)
1002 node->flags |= HAMMER_NODE_DELETED;
1003 hammer_blockmap_free(trans, node->node_offset, sizeof(*node->ondisk));
1007 * Passively cache a referenced hammer_node in *cache. The caller may
1008 * release the node on return.
1011 hammer_cache_node(hammer_node_t node, struct hammer_node **cache)
1016 * If the node is being deleted, don't cache it!
1018 if (node->flags & HAMMER_NODE_DELETED)
1022 * Cache the node. If we previously cached a different node we
1023 * have to give HAMMER a chance to destroy it.
1026 if (node->cache1 != cache) {
1027 if (node->cache2 != cache) {
1028 if ((old = *cache) != NULL) {
1029 KKASSERT(node->lock.refs != 0);
1030 hammer_uncache_node(cache);
1034 *node->cache2 = NULL;
1035 node->cache2 = node->cache1;
1036 node->cache1 = cache;
1039 struct hammer_node **tmp;
1041 node->cache1 = node->cache2;
1048 hammer_uncache_node(struct hammer_node **cache)
1052 if ((node = *cache) != NULL) {
1054 if (node->cache1 == cache) {
1055 node->cache1 = node->cache2;
1056 node->cache2 = NULL;
1057 } else if (node->cache2 == cache) {
1058 node->cache2 = NULL;
1060 panic("hammer_uncache_node: missing cache linkage");
1062 if (node->cache1 == NULL && node->cache2 == NULL)
1063 hammer_flush_node(node);
1068 * Remove a node's cache references and destroy the node if it has no
1069 * other references or backing store.
1072 hammer_flush_node(hammer_node_t node)
1074 hammer_buffer_t buffer;
1077 *node->cache1 = NULL;
1079 *node->cache2 = NULL;
1080 if (node->lock.refs == 0 && node->ondisk == NULL) {
1081 RB_REMOVE(hammer_nod_rb_tree, &node->hmp->rb_nods_root, node);
1082 if ((buffer = node->buffer) != NULL) {
1083 node->buffer = NULL;
1084 TAILQ_REMOVE(&buffer->clist, node, entry);
1085 /* buffer is unreferenced because ondisk is NULL */
1087 --hammer_count_nodes;
1088 kfree(node, M_HAMMER);
1093 * Flush passively cached B-Tree nodes associated with this buffer.
1094 * This is only called when the buffer is about to be destroyed, so
1095 * none of the nodes should have any references. The buffer is locked.
1097 * We may be interlocked with the buffer.
1100 hammer_flush_buffer_nodes(hammer_buffer_t buffer)
1104 while ((node = TAILQ_FIRST(&buffer->clist)) != NULL) {
1105 KKASSERT(node->ondisk == NULL);
1107 if (node->lock.refs == 0) {
1108 hammer_ref(&node->lock);
1109 node->flags |= HAMMER_NODE_FLUSH;
1110 hammer_rel_node(node);
1112 KKASSERT(node->loading != 0);
1113 KKASSERT(node->buffer != NULL);
1114 buffer = node->buffer;
1115 node->buffer = NULL;
1116 TAILQ_REMOVE(&buffer->clist, node, entry);
1117 /* buffer is unreferenced because ondisk is NULL */
1123 /************************************************************************
1125 ************************************************************************/
1128 * Allocate a B-Tree node.
1131 hammer_alloc_btree(hammer_transaction_t trans, int *errorp)
1133 hammer_buffer_t buffer = NULL;
1134 hammer_node_t node = NULL;
1135 hammer_off_t node_offset;
1137 node_offset = hammer_blockmap_alloc(trans, HAMMER_ZONE_BTREE_INDEX,
1138 sizeof(struct hammer_node_ondisk),
1141 node = hammer_get_node(trans->hmp, node_offset, errorp);
1142 hammer_modify_node_noundo(trans, node);
1143 bzero(node->ondisk, sizeof(*node->ondisk));
1146 hammer_rel_buffer(buffer, 0);
1151 * The returned buffers are already appropriately marked as being modified.
1152 * If the caller marks them again unnecessary undo records may be generated.
1154 * In-band data is indicated by data_bufferp == NULL. Pass a data_len of 0
1155 * for zero-fill (caller modifies data_len afterwords).
1158 hammer_alloc_record(hammer_transaction_t trans,
1159 hammer_off_t *rec_offp, u_int16_t rec_type,
1160 struct hammer_buffer **rec_bufferp,
1161 int32_t data_len, void **datap,
1162 struct hammer_buffer **data_bufferp, int *errorp)
1164 hammer_record_ondisk_t rec;
1165 hammer_off_t rec_offset;
1166 hammer_off_t data_offset;
1173 * Allocate the record
1175 rec_offset = hammer_blockmap_alloc(trans, HAMMER_ZONE_RECORD_INDEX,
1176 HAMMER_RECORD_SIZE, errorp);
1184 if (data_bufferp == NULL) {
1186 case HAMMER_RECTYPE_DATA:
1187 reclen = offsetof(struct hammer_data_record,
1190 case HAMMER_RECTYPE_DIRENTRY:
1191 reclen = offsetof(struct hammer_entry_record,
1195 panic("hammer_alloc_record: illegal "
1201 KKASSERT(reclen + data_len <= HAMMER_RECORD_SIZE);
1202 data_offset = rec_offset + reclen;
1203 } else if (data_len < HAMMER_BUFSIZE) {
1204 data_offset = hammer_blockmap_alloc(trans,
1205 HAMMER_ZONE_SMALL_DATA_INDEX,
1208 data_offset = hammer_blockmap_alloc(trans,
1209 HAMMER_ZONE_LARGE_DATA_INDEX,
1216 hammer_blockmap_free(trans, rec_offset, HAMMER_RECORD_SIZE);
1221 * Basic return values.
1223 *rec_offp = rec_offset;
1224 rec = hammer_bread(trans->hmp, rec_offset, errorp, rec_bufferp);
1225 hammer_modify_buffer(trans, *rec_bufferp, NULL, 0);
1226 bzero(rec, sizeof(*rec));
1227 KKASSERT(*errorp == 0);
1228 rec->base.data_off = data_offset;
1229 rec->base.data_len = data_len;
1233 *datap = hammer_bread(trans->hmp, data_offset, errorp,
1235 KKASSERT(*errorp == 0);
1236 hammer_modify_buffer(trans, *data_bufferp, NULL, 0);
1240 } else if (data_len) {
1241 KKASSERT(data_offset + data_len - rec_offset <=
1242 HAMMER_RECORD_SIZE);
1244 *datap = (void *)((char *)rec +
1245 (int32_t)(data_offset - rec_offset));
1248 KKASSERT(datap == NULL);
1250 KKASSERT(*errorp == 0);
1255 hammer_alloc_data(hammer_transaction_t trans, int32_t data_len,
1256 hammer_off_t *data_offsetp,
1257 struct hammer_buffer **data_bufferp, int *errorp)
1265 if (data_len < HAMMER_BUFSIZE) {
1266 *data_offsetp = hammer_blockmap_alloc(trans,
1267 HAMMER_ZONE_SMALL_DATA_INDEX,
1270 *data_offsetp = hammer_blockmap_alloc(trans,
1271 HAMMER_ZONE_LARGE_DATA_INDEX,
1277 if (*errorp == 0 && data_bufferp) {
1279 data = hammer_bread(trans->hmp, *data_offsetp, errorp,
1281 KKASSERT(*errorp == 0);
1282 hammer_modify_buffer(trans, *data_bufferp, NULL, 0);
1289 KKASSERT(*errorp == 0);
1295 * Sync dirty buffers to the media
1298 static int hammer_sync_scan1(struct mount *mp, struct vnode *vp, void *data);
1299 static int hammer_sync_scan2(struct mount *mp, struct vnode *vp, void *data);
1302 hammer_sync_hmp(hammer_mount_t hmp, int waitfor)
1304 struct hammer_sync_info info;
1307 info.waitfor = waitfor;
1309 vmntvnodescan(hmp->mp, VMSC_GETVP|VMSC_NOWAIT,
1310 hammer_sync_scan1, hammer_sync_scan2, &info);
1312 RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
1313 hammer_sync_volume, &info);
1318 hammer_sync_scan1(struct mount *mp, struct vnode *vp, void *data)
1320 struct hammer_inode *ip;
1323 if (vp->v_type == VNON || ip == NULL ||
1324 ((ip->flags & HAMMER_INODE_MODMASK) == 0 &&
1325 RB_EMPTY(&vp->v_rbdirty_tree))) {
1332 hammer_sync_scan2(struct mount *mp, struct vnode *vp, void *data)
1334 struct hammer_sync_info *info = data;
1335 struct hammer_inode *ip;
1339 if (vp->v_type == VNON || vp->v_type == VBAD ||
1340 ((ip->flags & HAMMER_INODE_MODMASK) == 0 &&
1341 RB_EMPTY(&vp->v_rbdirty_tree))) {
1344 error = VOP_FSYNC(vp, info->waitfor);
1346 info->error = error;
1351 hammer_sync_volume(hammer_volume_t volume, void *data)
1353 struct hammer_sync_info *info = data;
1355 hammer_ref(&volume->io.lock);
1356 RB_SCAN(hammer_buf_rb_tree, &volume->rb_bufs_root, NULL,
1357 hammer_sync_buffer, info);
1358 hammer_rel_volume(volume, 1);
1363 hammer_sync_buffer(hammer_buffer_t buffer, void *data __unused)
1365 hammer_ref(&buffer->io.lock);
1366 hammer_rel_buffer(buffer, 1);