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 1.43 2008/05/13 05:04:39 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, int isnew);
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);
160 hammer_io_init(&volume->io, hmp, HAMMER_STRUCTURE_VOLUME);
161 volume->io.offset = 0LL;
164 * Get the device vnode
166 error = nlookup_init(&nd, volume->vol_name, UIO_SYSSPACE, NLC_FOLLOW);
168 error = nlookup(&nd);
170 error = cache_vref(&nd.nl_nch, nd.nl_cred, &volume->devvp);
173 if (vn_isdisk(volume->devvp, &error)) {
174 error = vfs_mountedon(volume->devvp);
178 count_udev(volume->devvp->v_umajor, volume->devvp->v_uminor) > 0) {
182 vn_lock(volume->devvp, LK_EXCLUSIVE | LK_RETRY);
183 error = vinvalbuf(volume->devvp, V_SAVE, 0, 0);
185 error = VOP_OPEN(volume->devvp,
186 (ronly ? FREAD : FREAD|FWRITE),
189 vn_unlock(volume->devvp);
192 hammer_free_volume(volume);
195 volume->devvp->v_rdev->si_mountpoint = mp;
199 * Extract the volume number from the volume header and do various
202 error = bread(volume->devvp, 0LL, HAMMER_BUFSIZE, &bp);
205 ondisk = (void *)bp->b_data;
206 if (ondisk->vol_signature != HAMMER_FSBUF_VOLUME) {
207 kprintf("hammer_mount: volume %s has an invalid header\n",
212 volume->vol_no = ondisk->vol_no;
213 volume->buffer_base = ondisk->vol_buf_beg;
214 volume->vol_flags = ondisk->vol_flags;
215 volume->nblocks = ondisk->vol_nblocks;
216 volume->maxbuf_off = HAMMER_ENCODE_RAW_BUFFER(volume->vol_no,
217 ondisk->vol_buf_end - ondisk->vol_buf_beg);
218 RB_INIT(&volume->rb_bufs_root);
220 if (RB_EMPTY(&hmp->rb_vols_root)) {
221 hmp->fsid = ondisk->vol_fsid;
222 } else if (bcmp(&hmp->fsid, &ondisk->vol_fsid, sizeof(uuid_t))) {
223 kprintf("hammer_mount: volume %s's fsid does not match "
224 "other volumes\n", volume->vol_name);
230 * Insert the volume structure into the red-black tree.
232 if (RB_INSERT(hammer_vol_rb_tree, &hmp->rb_vols_root, volume)) {
233 kprintf("hammer_mount: volume %s has a duplicate vol_no %d\n",
234 volume->vol_name, volume->vol_no);
239 * Set the root volume . HAMMER special cases rootvol the structure.
240 * We do not hold a ref because this would prevent related I/O
241 * from being flushed.
243 if (error == 0 && ondisk->vol_rootvol == ondisk->vol_no) {
244 hmp->rootvol = volume;
249 hmp->fsid_udev = dev2udev(vn_todev(volume->devvp));
250 hmp->mp->mnt_stat.f_blocks += ondisk->vol0_stat_bigblocks *
251 (HAMMER_LARGEBLOCK_SIZE / HAMMER_BUFSIZE);
257 /*vinvalbuf(volume->devvp, V_SAVE, 0, 0);*/
259 volume->devvp->v_rdev->si_mountpoint = NULL;
260 VOP_CLOSE(volume->devvp, ronly ? FREAD : FREAD|FWRITE);
261 hammer_free_volume(volume);
267 * Unload and free a HAMMER volume. Must return >= 0 to continue scan
268 * so returns -1 on failure.
271 hammer_unload_volume(hammer_volume_t volume, void *data __unused)
273 struct hammer_mount *hmp = volume->io.hmp;
274 int ronly = ((hmp->mp->mnt_flag & MNT_RDONLY) ? 1 : 0);
277 * Clean up the root volume pointer, which is held unlocked in hmp.
279 if (hmp->rootvol == volume)
285 RB_SCAN(hammer_buf_rb_tree, &volume->rb_bufs_root, NULL,
286 hammer_unload_buffer, NULL);
289 * Release our buffer and flush anything left in the buffer cache.
291 volume->io.flush = 1;
292 volume->io.waitdep = 1;
293 hammer_io_release(&volume->io);
296 * There should be no references on the volume, no clusters, and
299 KKASSERT(volume->io.lock.refs == 0);
300 KKASSERT(RB_EMPTY(&volume->rb_bufs_root));
302 volume->ondisk = NULL;
304 if (volume->devvp->v_rdev &&
305 volume->devvp->v_rdev->si_mountpoint == hmp->mp
307 volume->devvp->v_rdev->si_mountpoint = NULL;
310 vinvalbuf(volume->devvp, 0, 0, 0);
311 VOP_CLOSE(volume->devvp, FREAD);
313 vinvalbuf(volume->devvp, V_SAVE, 0, 0);
314 VOP_CLOSE(volume->devvp, FREAD|FWRITE);
319 * Destroy the structure
321 RB_REMOVE(hammer_vol_rb_tree, &hmp->rb_vols_root, volume);
322 hammer_free_volume(volume);
328 hammer_free_volume(hammer_volume_t volume)
330 if (volume->vol_name) {
331 kfree(volume->vol_name, M_HAMMER);
332 volume->vol_name = NULL;
335 vrele(volume->devvp);
336 volume->devvp = NULL;
338 --hammer_count_volumes;
339 kfree(volume, M_HAMMER);
343 * Get a HAMMER volume. The volume must already exist.
346 hammer_get_volume(struct hammer_mount *hmp, int32_t vol_no, int *errorp)
348 struct hammer_volume *volume;
351 * Locate the volume structure
353 volume = RB_LOOKUP(hammer_vol_rb_tree, &hmp->rb_vols_root, vol_no);
354 if (volume == NULL) {
358 hammer_ref(&volume->io.lock);
361 * Deal with on-disk info
363 if (volume->ondisk == NULL || volume->io.loading) {
364 *errorp = hammer_load_volume(volume);
366 hammer_rel_volume(volume, 1);
376 hammer_ref_volume(hammer_volume_t volume)
380 hammer_ref(&volume->io.lock);
383 * Deal with on-disk info
385 if (volume->ondisk == NULL || volume->io.loading) {
386 error = hammer_load_volume(volume);
388 hammer_rel_volume(volume, 1);
396 hammer_get_root_volume(struct hammer_mount *hmp, int *errorp)
398 hammer_volume_t volume;
400 volume = hmp->rootvol;
401 KKASSERT(volume != NULL);
402 hammer_ref(&volume->io.lock);
405 * Deal with on-disk info
407 if (volume->ondisk == NULL || volume->io.loading) {
408 *errorp = hammer_load_volume(volume);
410 hammer_rel_volume(volume, 1);
420 * Load a volume's on-disk information. The volume must be referenced and
421 * not locked. We temporarily acquire an exclusive lock to interlock
422 * against releases or multiple get's.
425 hammer_load_volume(hammer_volume_t volume)
429 ++volume->io.loading;
430 hammer_lock_ex(&volume->io.lock);
432 if (volume->ondisk == NULL) {
433 error = hammer_io_read(volume->devvp, &volume->io);
435 volume->ondisk = (void *)volume->io.bp->b_data;
439 --volume->io.loading;
440 hammer_unlock(&volume->io.lock);
445 * Release a volume. Call hammer_io_release on the last reference. We have
446 * to acquire an exclusive lock to interlock against volume->ondisk tests
447 * in hammer_load_volume(), and hammer_io_release() also expects an exclusive
450 * Volumes are not unloaded from memory during normal operation.
453 hammer_rel_volume(hammer_volume_t volume, int flush)
456 volume->io.flush = 1;
458 if (volume->io.lock.refs == 1) {
459 ++volume->io.loading;
460 hammer_lock_ex(&volume->io.lock);
461 if (volume->io.lock.refs == 1) {
462 volume->ondisk = NULL;
463 hammer_io_release(&volume->io);
465 --volume->io.loading;
466 hammer_unlock(&volume->io.lock);
468 hammer_unref(&volume->io.lock);
472 /************************************************************************
474 ************************************************************************
476 * Manage buffers. Currently all blockmap-backed zones are translated
477 * to zone-2 buffer offsets.
480 hammer_get_buffer(hammer_mount_t hmp, hammer_off_t buf_offset,
481 int isnew, int *errorp)
483 hammer_buffer_t buffer;
484 hammer_volume_t volume;
485 hammer_off_t zoneX_offset;
486 hammer_io_type_t iotype;
490 zoneX_offset = buf_offset;
491 zone = HAMMER_ZONE_DECODE(buf_offset);
494 * What is the buffer class?
497 case HAMMER_ZONE_LARGE_DATA_INDEX:
498 case HAMMER_ZONE_SMALL_DATA_INDEX:
499 iotype = HAMMER_STRUCTURE_DATA_BUFFER;
501 case HAMMER_ZONE_UNDO_INDEX:
502 iotype = HAMMER_STRUCTURE_UNDO_BUFFER;
505 iotype = HAMMER_STRUCTURE_META_BUFFER;
510 * Handle blockmap offset translations
512 if (zone >= HAMMER_ZONE_BTREE_INDEX) {
513 buf_offset = hammer_blockmap_lookup(hmp, buf_offset, errorp);
514 KKASSERT(*errorp == 0);
515 } else if (zone == HAMMER_ZONE_UNDO_INDEX) {
516 buf_offset = hammer_undo_lookup(hmp, buf_offset, errorp);
517 KKASSERT(*errorp == 0);
521 * Locate the buffer given its zone-2 offset.
523 buf_offset &= ~HAMMER_BUFMASK64;
524 KKASSERT((buf_offset & HAMMER_ZONE_RAW_BUFFER) ==
525 HAMMER_ZONE_RAW_BUFFER);
526 vol_no = HAMMER_VOL_DECODE(buf_offset);
527 volume = hammer_get_volume(hmp, vol_no, errorp);
532 * NOTE: buf_offset and maxbuf_off are both full offset
535 KKASSERT(buf_offset < volume->maxbuf_off);
538 * Locate and lock the buffer structure, creating one if necessary.
541 buffer = RB_LOOKUP(hammer_buf_rb_tree, &volume->rb_bufs_root,
543 if (buffer == NULL) {
544 ++hammer_count_buffers;
545 buffer = kmalloc(sizeof(*buffer), M_HAMMER, M_WAITOK|M_ZERO);
546 buffer->zone2_offset = buf_offset;
547 buffer->volume = volume;
549 hammer_io_init(&buffer->io, hmp, iotype);
550 buffer->io.offset = volume->ondisk->vol_buf_beg +
551 (buf_offset & HAMMER_OFF_SHORT_MASK);
552 TAILQ_INIT(&buffer->clist);
553 hammer_ref(&buffer->io.lock);
556 * Insert the buffer into the RB tree and handle late
559 if (RB_INSERT(hammer_buf_rb_tree, &volume->rb_bufs_root, buffer)) {
560 hammer_unref(&buffer->io.lock);
561 --hammer_count_buffers;
562 kfree(buffer, M_HAMMER);
565 hammer_ref(&volume->io.lock);
567 hammer_ref(&buffer->io.lock);
570 * The buffer is no longer loose if it has a ref.
572 if (buffer->io.mod_list == &hmp->lose_list) {
573 TAILQ_REMOVE(buffer->io.mod_list, &buffer->io,
575 buffer->io.mod_list = NULL;
577 if (buffer->io.lock.refs == 1)
578 hammer_io_reinit(&buffer->io, iotype);
580 KKASSERT(buffer->io.type == iotype);
584 * Cache the blockmap translation
586 if ((zoneX_offset & HAMMER_ZONE_RAW_BUFFER) != HAMMER_ZONE_RAW_BUFFER)
587 buffer->zoneX_offset = zoneX_offset;
590 * Deal with on-disk info
592 if (buffer->ondisk == NULL || buffer->io.loading) {
593 *errorp = hammer_load_buffer(buffer, isnew);
595 hammer_rel_buffer(buffer, 1);
601 hammer_rel_volume(volume, 0);
606 hammer_load_buffer(hammer_buffer_t buffer, int isnew)
608 hammer_volume_t volume;
612 * Load the buffer's on-disk info
614 volume = buffer->volume;
615 ++buffer->io.loading;
616 hammer_lock_ex(&buffer->io.lock);
618 if (buffer->ondisk == NULL) {
620 error = hammer_io_new(volume->devvp, &buffer->io);
622 error = hammer_io_read(volume->devvp, &buffer->io);
625 buffer->ondisk = (void *)buffer->io.bp->b_data;
627 error = hammer_io_new(volume->devvp, &buffer->io);
631 --buffer->io.loading;
632 hammer_unlock(&buffer->io.lock);
637 * NOTE: Called from RB_SCAN, must return >= 0 for scan to continue.
640 hammer_unload_buffer(hammer_buffer_t buffer, void *data __unused)
642 hammer_ref(&buffer->io.lock);
643 hammer_flush_buffer_nodes(buffer);
644 KKASSERT(buffer->io.lock.refs == 1);
645 hammer_rel_buffer(buffer, 2);
650 * Reference a buffer that is either already referenced or via a specially
651 * handled pointer (aka cursor->buffer).
654 hammer_ref_buffer(hammer_buffer_t buffer)
658 hammer_ref(&buffer->io.lock);
663 if (buffer->io.mod_list == &buffer->io.hmp->lose_list) {
664 TAILQ_REMOVE(buffer->io.mod_list, &buffer->io, mod_entry);
665 buffer->io.mod_list = NULL;
668 if (buffer->ondisk == NULL || buffer->io.loading) {
669 error = hammer_load_buffer(buffer, 0);
671 hammer_rel_buffer(buffer, 1);
673 * NOTE: buffer pointer can become stale after
684 * Release a buffer. We have to deal with several places where
685 * another thread can ref the buffer.
687 * Only destroy the structure itself if the related buffer cache buffer
688 * was disassociated from it. This ties the management of the structure
689 * to the buffer cache subsystem. buffer->ondisk determines whether the
690 * embedded io is referenced or not.
693 hammer_rel_buffer(hammer_buffer_t buffer, int flush)
695 hammer_volume_t volume;
699 buffer->io.flush = 1;
701 if (buffer->io.lock.refs == 1) {
702 ++buffer->io.loading; /* force interlock check */
703 hammer_lock_ex(&buffer->io.lock);
704 if (buffer->io.lock.refs == 1) {
705 hammer_io_release(&buffer->io);
706 hammer_flush_buffer_nodes(buffer);
707 KKASSERT(TAILQ_EMPTY(&buffer->clist));
709 if (buffer->io.bp == NULL &&
710 buffer->io.lock.refs == 1) {
714 volume = buffer->volume;
715 RB_REMOVE(hammer_buf_rb_tree,
716 &volume->rb_bufs_root, buffer);
717 buffer->volume = NULL; /* sanity */
718 hammer_rel_volume(volume, 0);
722 --buffer->io.loading;
723 hammer_unlock(&buffer->io.lock);
725 hammer_unref(&buffer->io.lock);
728 KKASSERT(buffer->io.mod_list == NULL);
729 --hammer_count_buffers;
730 kfree(buffer, M_HAMMER);
735 * Remove the zoneX translation cache for a buffer given its zone-2 offset.
738 hammer_uncache_buffer(hammer_mount_t hmp, hammer_off_t buf_offset)
740 hammer_volume_t volume;
741 hammer_buffer_t buffer;
745 buf_offset &= ~HAMMER_BUFMASK64;
746 KKASSERT((buf_offset & HAMMER_ZONE_RAW_BUFFER) ==
747 HAMMER_ZONE_RAW_BUFFER);
748 vol_no = HAMMER_VOL_DECODE(buf_offset);
749 volume = hammer_get_volume(hmp, vol_no, &error);
750 KKASSERT(volume != 0);
751 KKASSERT(buf_offset < volume->maxbuf_off);
753 buffer = RB_LOOKUP(hammer_buf_rb_tree, &volume->rb_bufs_root,
756 buffer->zoneX_offset = 0;
757 hammer_rel_volume(volume, 0);
761 * Access the filesystem buffer containing the specified hammer offset.
762 * buf_offset is a conglomeration of the volume number and vol_buf_beg
763 * relative buffer offset. It must also have bit 55 set to be valid.
764 * (see hammer_off_t in hammer_disk.h).
766 * Any prior buffer in *bufferp will be released and replaced by the
770 hammer_bread(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;
777 KKASSERT((buf_offset & HAMMER_OFF_ZONE_MASK) != 0);
780 if (buffer == NULL || (buffer->zone2_offset != buf_offset &&
781 buffer->zoneX_offset != buf_offset)) {
783 hammer_rel_buffer(buffer, 0);
784 buffer = hammer_get_buffer(hmp, buf_offset, 0, errorp);
791 * Return a pointer to the buffer data.
796 return((char *)buffer->ondisk + xoff);
800 * Access the filesystem buffer containing the specified hammer offset.
801 * No disk read operation occurs. The result buffer may contain garbage.
803 * Any prior buffer in *bufferp will be released and replaced by the
806 * This function marks the buffer dirty but does not increment its
810 hammer_bnew(hammer_mount_t hmp, hammer_off_t buf_offset, int *errorp,
811 struct hammer_buffer **bufferp)
813 hammer_buffer_t buffer;
814 int32_t xoff = (int32_t)buf_offset & HAMMER_BUFMASK;
816 buf_offset &= ~HAMMER_BUFMASK64;
819 if (buffer == NULL || (buffer->zone2_offset != buf_offset &&
820 buffer->zoneX_offset != buf_offset)) {
822 hammer_rel_buffer(buffer, 0);
823 buffer = hammer_get_buffer(hmp, buf_offset, 1, errorp);
830 * Return a pointer to the buffer data.
835 return((char *)buffer->ondisk + xoff);
838 /************************************************************************
840 ************************************************************************
842 * Manage B-Tree nodes. B-Tree nodes represent the primary indexing
843 * method used by the HAMMER filesystem.
845 * Unlike other HAMMER structures, a hammer_node can be PASSIVELY
846 * associated with its buffer, and will only referenced the buffer while
847 * the node itself is referenced.
849 * A hammer_node can also be passively associated with other HAMMER
850 * structures, such as inodes, while retaining 0 references. These
851 * associations can be cleared backwards using a pointer-to-pointer in
854 * This allows the HAMMER implementation to cache hammer_nodes long-term
855 * and short-cut a great deal of the infrastructure's complexity. In
856 * most cases a cached node can be reacquired without having to dip into
857 * either the buffer or cluster management code.
859 * The caller must pass a referenced cluster on call and will retain
860 * ownership of the reference on return. The node will acquire its own
861 * additional references, if necessary.
864 hammer_get_node(hammer_mount_t hmp, hammer_off_t node_offset,
865 int isnew, int *errorp)
869 KKASSERT((node_offset & HAMMER_OFF_ZONE_MASK) == HAMMER_ZONE_BTREE);
872 * Locate the structure, allocating one if necessary.
875 node = RB_LOOKUP(hammer_nod_rb_tree, &hmp->rb_nods_root, node_offset);
877 ++hammer_count_nodes;
878 node = kmalloc(sizeof(*node), M_HAMMER, M_WAITOK|M_ZERO);
879 node->node_offset = node_offset;
881 if (RB_INSERT(hammer_nod_rb_tree, &hmp->rb_nods_root, node)) {
882 --hammer_count_nodes;
883 kfree(node, M_HAMMER);
887 hammer_ref(&node->lock);
891 *errorp = hammer_load_node(node, isnew);
893 hammer_rel_node(node);
900 * Reference an already-referenced node.
903 hammer_ref_node(hammer_node_t node)
905 KKASSERT(node->lock.refs > 0 && node->ondisk != NULL);
906 hammer_ref(&node->lock);
910 * Load a node's on-disk data reference.
913 hammer_load_node(hammer_node_t node, int isnew)
915 hammer_buffer_t buffer;
920 hammer_lock_ex(&node->lock);
921 if (node->ondisk == NULL) {
923 * This is a little confusing but the jist is that
924 * node->buffer determines whether the node is on
925 * the buffer's clist and node->ondisk determines
926 * whether the buffer is referenced.
928 * We could be racing a buffer release, in which case
929 * node->buffer may become NULL while we are blocked
930 * referencing the buffer.
932 if ((buffer = node->buffer) != NULL) {
933 error = hammer_ref_buffer(buffer);
934 if (error == 0 && node->buffer == NULL) {
935 TAILQ_INSERT_TAIL(&buffer->clist,
937 node->buffer = buffer;
940 buffer = hammer_get_buffer(node->hmp,
941 node->node_offset, 0,
944 KKASSERT(error == 0);
945 TAILQ_INSERT_TAIL(&buffer->clist,
947 node->buffer = buffer;
951 node->ondisk = (void *)((char *)buffer->ondisk +
952 (node->node_offset & HAMMER_BUFMASK));
954 hammer_crc_test_btree(node->ondisk) == 0) {
955 Debugger("CRC FAILED: B-TREE NODE");
960 hammer_unlock(&node->lock);
965 * Safely reference a node, interlock against flushes via the IO subsystem.
968 hammer_ref_node_safe(struct hammer_mount *hmp, struct hammer_node **cache,
975 hammer_ref(&node->lock);
979 *errorp = hammer_load_node(node, 0);
981 hammer_rel_node(node);
991 * Release a hammer_node. On the last release the node dereferences
992 * its underlying buffer and may or may not be destroyed.
995 hammer_rel_node(hammer_node_t node)
997 hammer_buffer_t buffer;
1000 * If this isn't the last ref just decrement the ref count and
1003 if (node->lock.refs > 1) {
1004 hammer_unref(&node->lock);
1009 * If there is no ondisk info or no buffer the node failed to load,
1010 * remove the last reference and destroy the node.
1012 if (node->ondisk == NULL) {
1013 hammer_unref(&node->lock);
1014 hammer_flush_node(node);
1015 /* node is stale now */
1020 * Do final cleanups and then either destroy the node and leave it
1021 * passively cached. The buffer reference is removed regardless.
1023 buffer = node->buffer;
1024 node->ondisk = NULL;
1026 if ((node->flags & HAMMER_NODE_FLUSH) == 0) {
1027 hammer_unref(&node->lock);
1028 hammer_rel_buffer(buffer, 0);
1035 hammer_unref(&node->lock);
1036 hammer_flush_node(node);
1038 hammer_rel_buffer(buffer, 0);
1042 * Free space on-media associated with a B-Tree node.
1045 hammer_delete_node(hammer_transaction_t trans, hammer_node_t node)
1047 KKASSERT((node->flags & HAMMER_NODE_DELETED) == 0);
1048 node->flags |= HAMMER_NODE_DELETED;
1049 hammer_blockmap_free(trans, node->node_offset, sizeof(*node->ondisk));
1053 * Passively cache a referenced hammer_node in *cache. The caller may
1054 * release the node on return.
1057 hammer_cache_node(hammer_node_t node, struct hammer_node **cache)
1062 * If the node is being deleted, don't cache it!
1064 if (node->flags & HAMMER_NODE_DELETED)
1068 * Cache the node. If we previously cached a different node we
1069 * have to give HAMMER a chance to destroy it.
1072 if (node->cache1 != cache) {
1073 if (node->cache2 != cache) {
1074 if ((old = *cache) != NULL) {
1075 KKASSERT(node->lock.refs != 0);
1076 hammer_uncache_node(cache);
1080 *node->cache2 = NULL;
1081 node->cache2 = node->cache1;
1082 node->cache1 = cache;
1085 struct hammer_node **tmp;
1087 node->cache1 = node->cache2;
1094 hammer_uncache_node(struct hammer_node **cache)
1098 if ((node = *cache) != NULL) {
1100 if (node->cache1 == cache) {
1101 node->cache1 = node->cache2;
1102 node->cache2 = NULL;
1103 } else if (node->cache2 == cache) {
1104 node->cache2 = NULL;
1106 panic("hammer_uncache_node: missing cache linkage");
1108 if (node->cache1 == NULL && node->cache2 == NULL)
1109 hammer_flush_node(node);
1114 * Remove a node's cache references and destroy the node if it has no
1115 * other references or backing store.
1118 hammer_flush_node(hammer_node_t node)
1120 hammer_buffer_t buffer;
1123 *node->cache1 = NULL;
1125 *node->cache2 = NULL;
1126 if (node->lock.refs == 0 && node->ondisk == NULL) {
1127 RB_REMOVE(hammer_nod_rb_tree, &node->hmp->rb_nods_root, node);
1128 if ((buffer = node->buffer) != NULL) {
1129 node->buffer = NULL;
1130 TAILQ_REMOVE(&buffer->clist, node, entry);
1131 /* buffer is unreferenced because ondisk is NULL */
1133 --hammer_count_nodes;
1134 kfree(node, M_HAMMER);
1139 * Flush passively cached B-Tree nodes associated with this buffer.
1140 * This is only called when the buffer is about to be destroyed, so
1141 * none of the nodes should have any references. The buffer is locked.
1143 * We may be interlocked with the buffer.
1146 hammer_flush_buffer_nodes(hammer_buffer_t buffer)
1150 while ((node = TAILQ_FIRST(&buffer->clist)) != NULL) {
1151 KKASSERT(node->ondisk == NULL);
1153 if (node->lock.refs == 0) {
1154 hammer_ref(&node->lock);
1155 node->flags |= HAMMER_NODE_FLUSH;
1156 hammer_rel_node(node);
1158 KKASSERT(node->loading != 0);
1159 KKASSERT(node->buffer != NULL);
1160 buffer = node->buffer;
1161 node->buffer = NULL;
1162 TAILQ_REMOVE(&buffer->clist, node, entry);
1163 /* buffer is unreferenced because ondisk is NULL */
1169 /************************************************************************
1171 ************************************************************************/
1174 * Allocate a B-Tree node.
1177 hammer_alloc_btree(hammer_transaction_t trans, int *errorp)
1179 hammer_buffer_t buffer = NULL;
1180 hammer_node_t node = NULL;
1181 hammer_off_t node_offset;
1183 node_offset = hammer_blockmap_alloc(trans, HAMMER_ZONE_BTREE_INDEX,
1184 sizeof(struct hammer_node_ondisk),
1187 node = hammer_get_node(trans->hmp, node_offset, 1, errorp);
1188 hammer_modify_node_noundo(trans, node);
1189 bzero(node->ondisk, sizeof(*node->ondisk));
1190 hammer_modify_node_done(node);
1193 hammer_rel_buffer(buffer, 0);
1200 * The returned buffers are already appropriately marked as being modified.
1201 * If the caller marks them again unnecessary undo records may be generated.
1203 * In-band data is indicated by data_bufferp == NULL. Pass a data_len of 0
1204 * for zero-fill (caller modifies data_len afterwords).
1206 * If the caller is responsible for calling hammer_modify_*() prior to making
1207 * any additional modifications to either the returned record buffer or the
1208 * returned data buffer.
1211 hammer_alloc_record(hammer_transaction_t trans,
1212 hammer_off_t *rec_offp, u_int16_t rec_type,
1213 struct hammer_buffer **rec_bufferp,
1214 int32_t data_len, void **datap,
1215 hammer_off_t *data_offp,
1216 struct hammer_buffer **data_bufferp, int *errorp)
1218 hammer_record_ondisk_t rec;
1219 hammer_off_t rec_offset;
1220 hammer_off_t data_offset;
1227 * Allocate the record
1229 rec_offset = hammer_blockmap_alloc(trans, HAMMER_ZONE_RECORD_INDEX,
1230 HAMMER_RECORD_SIZE, errorp);
1240 if (data_bufferp == NULL) {
1242 case HAMMER_RECTYPE_DATA:
1243 reclen = offsetof(struct hammer_data_record,
1246 case HAMMER_RECTYPE_DIRENTRY:
1247 reclen = offsetof(struct hammer_entry_record,
1251 panic("hammer_alloc_record: illegal "
1257 KKASSERT(reclen + data_len <= HAMMER_RECORD_SIZE);
1258 data_offset = rec_offset + reclen;
1259 } else if (data_len < HAMMER_BUFSIZE) {
1260 data_offset = hammer_blockmap_alloc(trans,
1261 HAMMER_ZONE_SMALL_DATA_INDEX,
1263 *data_offp = data_offset;
1265 data_offset = hammer_blockmap_alloc(trans,
1266 HAMMER_ZONE_LARGE_DATA_INDEX,
1268 *data_offp = data_offset;
1274 hammer_blockmap_free(trans, rec_offset, HAMMER_RECORD_SIZE);
1279 * Basic return values.
1281 * Note that because this is a 'new' buffer, there is no need to
1282 * generate UNDO records for it.
1284 *rec_offp = rec_offset;
1285 rec = hammer_bread(trans->hmp, rec_offset, errorp, rec_bufferp);
1286 hammer_modify_buffer(trans, *rec_bufferp, NULL, 0);
1287 bzero(rec, sizeof(*rec));
1288 KKASSERT(*errorp == 0);
1289 rec->base.data_off = data_offset;
1290 rec->base.data_len = data_len;
1291 hammer_modify_buffer_done(*rec_bufferp);
1295 *datap = hammer_bread(trans->hmp, data_offset, errorp,
1297 KKASSERT(*errorp == 0);
1301 } else if (data_len) {
1302 KKASSERT(data_offset + data_len - rec_offset <=
1303 HAMMER_RECORD_SIZE);
1305 *datap = (void *)((char *)rec +
1306 (int32_t)(data_offset - rec_offset));
1309 KKASSERT(datap == NULL);
1311 KKASSERT(*errorp == 0);
1318 * Allocate data. If the address of a data buffer is supplied then
1319 * any prior non-NULL *data_bufferp will be released and *data_bufferp
1320 * will be set to the related buffer. The caller must release it when
1321 * finally done. The initial *data_bufferp should be set to NULL by
1324 * The caller is responsible for making hammer_modify*() calls on the
1328 hammer_alloc_data(hammer_transaction_t trans, int32_t data_len,
1329 hammer_off_t *data_offsetp,
1330 struct hammer_buffer **data_bufferp, int *errorp)
1338 if (data_len < HAMMER_BUFSIZE) {
1339 *data_offsetp = hammer_blockmap_alloc(trans,
1340 HAMMER_ZONE_SMALL_DATA_INDEX,
1343 *data_offsetp = hammer_blockmap_alloc(trans,
1344 HAMMER_ZONE_LARGE_DATA_INDEX,
1350 if (*errorp == 0 && data_bufferp) {
1352 data = hammer_bread(trans->hmp, *data_offsetp, errorp,
1354 KKASSERT(*errorp == 0);
1361 KKASSERT(*errorp == 0);
1366 * Sync dirty buffers to the media and clean-up any loose ends.
1368 static int hammer_sync_scan1(struct mount *mp, struct vnode *vp, void *data);
1369 static int hammer_sync_scan2(struct mount *mp, struct vnode *vp, void *data);
1372 hammer_sync_hmp(hammer_mount_t hmp, int waitfor)
1374 struct hammer_sync_info info;
1377 info.waitfor = waitfor;
1379 vmntvnodescan(hmp->mp, VMSC_GETVP|VMSC_NOWAIT,
1380 hammer_sync_scan1, hammer_sync_scan2, &info);
1381 if (waitfor == MNT_WAIT)
1382 hammer_flusher_sync(hmp);
1384 hammer_flusher_async(hmp);
1390 hammer_sync_scan1(struct mount *mp, struct vnode *vp, void *data)
1392 struct hammer_inode *ip;
1395 if (vp->v_type == VNON || ip == NULL ||
1396 ((ip->flags & HAMMER_INODE_MODMASK) == 0 &&
1397 RB_EMPTY(&vp->v_rbdirty_tree))) {
1404 hammer_sync_scan2(struct mount *mp, struct vnode *vp, void *data)
1406 struct hammer_sync_info *info = data;
1407 struct hammer_inode *ip;
1411 if (vp->v_type == VNON || vp->v_type == VBAD ||
1412 ((ip->flags & HAMMER_INODE_MODMASK) == 0 &&
1413 RB_EMPTY(&vp->v_rbdirty_tree))) {
1416 error = VOP_FSYNC(vp, info->waitfor);
1418 info->error = error;