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 1.29 2008/02/10 09:51:01 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);
53 static hammer_off_t hammer_advance_fifo(hammer_volume_t volume,
54 hammer_off_t off, int32_t bytes);
56 static hammer_off_t hammer_alloc_fifo(hammer_mount_t hmp, int32_t rec_len,
57 int32_t data_len, struct hammer_buffer **rec_bufferp,
58 u_int16_t hdr_type, int can_cross,
59 struct hammer_buffer **data2_bufferp, int *errorp);
63 * Red-Black tree support for various structures
66 hammer_ino_rb_compare(hammer_inode_t ip1, hammer_inode_t ip2)
68 if (ip1->obj_id < ip2->obj_id)
70 if (ip1->obj_id > ip2->obj_id)
72 if (ip1->obj_asof < ip2->obj_asof)
74 if (ip1->obj_asof > ip2->obj_asof)
80 hammer_inode_info_cmp(hammer_inode_info_t info, hammer_inode_t ip)
82 if (info->obj_id < ip->obj_id)
84 if (info->obj_id > ip->obj_id)
86 if (info->obj_asof < ip->obj_asof)
88 if (info->obj_asof > ip->obj_asof)
94 hammer_vol_rb_compare(hammer_volume_t vol1, hammer_volume_t vol2)
96 if (vol1->vol_no < vol2->vol_no)
98 if (vol1->vol_no > vol2->vol_no)
104 hammer_buf_rb_compare(hammer_buffer_t buf1, hammer_buffer_t buf2)
106 if (buf1->buf_offset < buf2->buf_offset)
108 if (buf1->buf_offset > buf2->buf_offset)
114 hammer_nod_rb_compare(hammer_node_t node1, hammer_node_t node2)
116 if (node1->node_offset < node2->node_offset)
118 if (node1->node_offset > node2->node_offset)
124 * Note: The lookup function for hammer_ino_rb_tree winds up being named
125 * hammer_ino_rb_tree_RB_LOOKUP_INFO(root, info). The other lookup
126 * functions are normal, e.g. hammer_buf_rb_tree_RB_LOOKUP(root, buf_offset).
128 RB_GENERATE(hammer_ino_rb_tree, hammer_inode, rb_node, hammer_ino_rb_compare);
129 RB_GENERATE_XLOOKUP(hammer_ino_rb_tree, INFO, hammer_inode, rb_node,
130 hammer_inode_info_cmp, hammer_inode_info_t);
131 RB_GENERATE2(hammer_vol_rb_tree, hammer_volume, rb_node,
132 hammer_vol_rb_compare, int32_t, vol_no);
133 RB_GENERATE2(hammer_buf_rb_tree, hammer_buffer, rb_node,
134 hammer_buf_rb_compare, hammer_off_t, buf_offset);
135 RB_GENERATE2(hammer_nod_rb_tree, hammer_node, rb_node,
136 hammer_nod_rb_compare, hammer_off_t, node_offset);
138 /************************************************************************
140 ************************************************************************
142 * Load a HAMMER volume by name. Returns 0 on success or a positive error
143 * code on failure. Volumes must be loaded at mount time, get_volume() will
144 * not load a new volume.
146 * Calls made to hammer_load_volume() or single-threaded
149 hammer_install_volume(struct hammer_mount *hmp, const char *volname)
152 hammer_volume_t volume;
153 struct hammer_volume_ondisk *ondisk;
154 struct nlookupdata nd;
155 struct buf *bp = NULL;
160 ronly = ((mp->mnt_flag & MNT_RDONLY) ? 1 : 0);
163 * Allocate a volume structure
165 ++hammer_count_volumes;
166 volume = kmalloc(sizeof(*volume), M_HAMMER, M_WAITOK|M_ZERO);
167 volume->vol_name = kstrdup(volname, M_HAMMER);
169 hammer_io_init(&volume->io, HAMMER_STRUCTURE_VOLUME);
170 volume->io.offset = 0LL;
173 * Get the device vnode
175 error = nlookup_init(&nd, volume->vol_name, UIO_SYSSPACE, NLC_FOLLOW);
177 error = nlookup(&nd);
179 error = cache_vref(&nd.nl_nch, nd.nl_cred, &volume->devvp);
182 if (vn_isdisk(volume->devvp, &error)) {
183 error = vfs_mountedon(volume->devvp);
187 count_udev(volume->devvp->v_umajor, volume->devvp->v_uminor) > 0) {
191 vn_lock(volume->devvp, LK_EXCLUSIVE | LK_RETRY);
192 error = vinvalbuf(volume->devvp, V_SAVE, 0, 0);
194 error = VOP_OPEN(volume->devvp,
195 (ronly ? FREAD : FREAD|FWRITE),
198 vn_unlock(volume->devvp);
201 hammer_free_volume(volume);
204 volume->devvp->v_rdev->si_mountpoint = mp;
207 * Extract the volume number from the volume header and do various
210 error = bread(volume->devvp, 0LL, HAMMER_BUFSIZE, &bp);
213 ondisk = (void *)bp->b_data;
214 if (ondisk->vol_signature != HAMMER_FSBUF_VOLUME) {
215 kprintf("hammer_mount: volume %s has an invalid header\n",
220 volume->vol_no = ondisk->vol_no;
221 volume->buffer_base = ondisk->vol_buf_beg;
222 volume->vol_flags = ondisk->vol_flags;
223 volume->nblocks = ondisk->vol_nblocks;
224 volume->maxbuf_off = HAMMER_ENCODE_RAW_BUFFER(volume->vol_no,
225 ondisk->vol_buf_end - ondisk->vol_buf_beg);
226 RB_INIT(&volume->rb_bufs_root);
228 hmp->mp->mnt_stat.f_blocks += volume->nblocks;
230 if (RB_EMPTY(&hmp->rb_vols_root)) {
231 hmp->fsid = ondisk->vol_fsid;
232 } else if (bcmp(&hmp->fsid, &ondisk->vol_fsid, sizeof(uuid_t))) {
233 kprintf("hammer_mount: volume %s's fsid does not match "
234 "other volumes\n", volume->vol_name);
240 * Insert the volume structure into the red-black tree.
242 if (RB_INSERT(hammer_vol_rb_tree, &hmp->rb_vols_root, volume)) {
243 kprintf("hammer_mount: volume %s has a duplicate vol_no %d\n",
244 volume->vol_name, volume->vol_no);
249 * Set the root volume . HAMMER special cases rootvol the structure.
250 * We do not hold a ref because this would prevent related I/O
251 * from being flushed.
253 if (error == 0 && ondisk->vol_rootvol == ondisk->vol_no) {
254 hmp->rootvol = volume;
259 hmp->fsid_udev = dev2udev(vn_todev(volume->devvp));
265 /*vinvalbuf(volume->devvp, V_SAVE, 0, 0);*/
266 VOP_CLOSE(volume->devvp, ronly ? FREAD : FREAD|FWRITE);
267 hammer_free_volume(volume);
273 * Unload and free a HAMMER volume. Must return >= 0 to continue scan
274 * so returns -1 on failure.
277 hammer_unload_volume(hammer_volume_t volume, void *data __unused)
279 struct hammer_mount *hmp = volume->hmp;
280 int ronly = ((hmp->mp->mnt_flag & MNT_RDONLY) ? 1 : 0);
283 * Sync clusters, sync volume
286 hmp->mp->mnt_stat.f_blocks -= volume->nblocks;
289 * Clean up the root volume pointer, which is held unlocked in hmp.
291 if (hmp->rootvol == volume)
295 * Unload clusters and super-clusters. Unloading a super-cluster
296 * also unloads related clusters, but the filesystem may not be
297 * using super-clusters so unload clusters anyway.
299 RB_SCAN(hammer_buf_rb_tree, &volume->rb_bufs_root, NULL,
300 hammer_unload_buffer, NULL);
301 hammer_io_waitdep(&volume->io);
304 * Release our buffer and flush anything left in the buffer cache.
306 hammer_io_release(&volume->io, 2);
309 * There should be no references on the volume, no clusters, and
312 KKASSERT(volume->io.lock.refs == 0);
313 KKASSERT(RB_EMPTY(&volume->rb_bufs_root));
315 volume->ondisk = NULL;
318 vinvalbuf(volume->devvp, 0, 0, 0);
319 VOP_CLOSE(volume->devvp, FREAD);
321 vinvalbuf(volume->devvp, V_SAVE, 0, 0);
322 VOP_CLOSE(volume->devvp, FREAD|FWRITE);
327 * Destroy the structure
329 RB_REMOVE(hammer_vol_rb_tree, &hmp->rb_vols_root, volume);
330 hammer_free_volume(volume);
336 hammer_free_volume(hammer_volume_t volume)
338 if (volume->vol_name) {
339 kfree(volume->vol_name, M_HAMMER);
340 volume->vol_name = NULL;
343 if (vn_isdisk(volume->devvp, NULL) &&
344 volume->devvp->v_rdev &&
345 volume->devvp->v_rdev->si_mountpoint == volume->hmp->mp
347 volume->devvp->v_rdev->si_mountpoint = NULL;
349 vrele(volume->devvp);
350 volume->devvp = NULL;
352 --hammer_count_volumes;
353 kfree(volume, M_HAMMER);
357 * Get a HAMMER volume. The volume must already exist.
360 hammer_get_volume(struct hammer_mount *hmp, int32_t vol_no, int *errorp)
362 struct hammer_volume *volume;
365 * Locate the volume structure
367 volume = RB_LOOKUP(hammer_vol_rb_tree, &hmp->rb_vols_root, vol_no);
368 if (volume == NULL) {
372 hammer_ref(&volume->io.lock);
375 * Deal with on-disk info
377 if (volume->ondisk == NULL || volume->io.loading) {
378 *errorp = hammer_load_volume(volume);
380 hammer_rel_volume(volume, 1);
390 hammer_ref_volume(hammer_volume_t volume)
394 hammer_ref(&volume->io.lock);
397 * Deal with on-disk info
399 if (volume->ondisk == NULL || volume->io.loading) {
400 error = hammer_load_volume(volume);
402 hammer_rel_volume(volume, 1);
410 hammer_get_root_volume(struct hammer_mount *hmp, int *errorp)
412 hammer_volume_t volume;
414 volume = hmp->rootvol;
415 KKASSERT(volume != NULL);
416 hammer_ref(&volume->io.lock);
419 * Deal with on-disk info
421 if (volume->ondisk == NULL || volume->io.loading) {
422 *errorp = hammer_load_volume(volume);
424 hammer_rel_volume(volume, 1);
434 * Load a volume's on-disk information. The volume must be referenced and
435 * not locked. We temporarily acquire an exclusive lock to interlock
436 * against releases or multiple get's.
439 hammer_load_volume(hammer_volume_t volume)
441 struct hammer_volume_ondisk *ondisk;
444 hammer_lock_ex(&volume->io.lock);
445 KKASSERT(volume->io.loading == 0);
446 volume->io.loading = 1;
448 if (volume->ondisk == NULL) {
449 error = hammer_io_read(volume->devvp, &volume->io);
451 volume->io.loading = 0;
452 hammer_unlock(&volume->io.lock);
455 volume->ondisk = ondisk = (void *)volume->io.bp->b_data;
459 volume->io.loading = 0;
460 hammer_unlock(&volume->io.lock);
465 * Release a volume. Call hammer_io_release on the last reference. We have
466 * to acquire an exclusive lock to interlock against volume->ondisk tests
467 * in hammer_load_volume(), and hammer_io_release() also expects an exclusive
470 * Volumes are not unloaded from memory during normal operation.
473 hammer_rel_volume(hammer_volume_t volume, int flush)
475 if (volume->io.lock.refs == 1) {
476 hammer_lock_ex(&volume->io.lock);
477 if (volume->io.lock.refs == 1) {
478 volume->ondisk = NULL;
479 hammer_io_release(&volume->io, flush);
481 hammer_io_flush(&volume->io);
483 hammer_unlock(&volume->io.lock);
485 hammer_unref(&volume->io.lock);
488 /************************************************************************
490 ************************************************************************
492 * Manage buffers. Currently all blockmap-backed zones are translated
493 * to zone-2 buffer offsets.
496 hammer_get_buffer(hammer_mount_t hmp, hammer_off_t buf_offset,
497 int isnew, int *errorp)
499 hammer_buffer_t buffer;
500 hammer_volume_t volume;
504 zone = HAMMER_ZONE_DECODE(buf_offset);
505 if (zone > HAMMER_ZONE_RAW_BUFFER_INDEX) {
506 buf_offset = hammer_blockmap_lookup(hmp, buf_offset, errorp);
507 KKASSERT(*errorp == 0);
509 buf_offset &= ~HAMMER_BUFMASK64;
510 KKASSERT((buf_offset & HAMMER_ZONE_RAW_BUFFER) ==
511 HAMMER_ZONE_RAW_BUFFER);
512 vol_no = HAMMER_VOL_DECODE(buf_offset);
513 volume = hammer_get_volume(hmp, vol_no, errorp);
518 * NOTE: buf_offset and maxbuf_off are both full offset
521 KKASSERT(buf_offset < volume->maxbuf_off);
524 * Locate and lock the buffer structure, creating one if necessary.
527 buffer = RB_LOOKUP(hammer_buf_rb_tree, &volume->rb_bufs_root,
529 if (buffer == NULL) {
530 ++hammer_count_buffers;
531 buffer = kmalloc(sizeof(*buffer), M_HAMMER, M_WAITOK|M_ZERO);
532 buffer->buf_offset = buf_offset;
533 buffer->volume = volume;
534 hammer_io_init(&buffer->io, HAMMER_STRUCTURE_BUFFER);
535 buffer->io.offset = volume->ondisk->vol_buf_beg +
536 (buf_offset & HAMMER_OFF_SHORT_MASK);
537 TAILQ_INIT(&buffer->clist);
538 hammer_ref(&buffer->io.lock);
541 * Insert the buffer into the RB tree and handle late
544 if (RB_INSERT(hammer_buf_rb_tree, &volume->rb_bufs_root, buffer)) {
545 hammer_unref(&buffer->io.lock);
546 --hammer_count_buffers;
547 kfree(buffer, M_HAMMER);
550 hammer_ref(&volume->io.lock);
552 hammer_ref(&buffer->io.lock);
556 * Deal with on-disk info
558 if (buffer->ondisk == NULL || buffer->io.loading) {
559 *errorp = hammer_load_buffer(buffer, isnew);
561 hammer_rel_buffer(buffer, 1);
567 hammer_rel_volume(volume, 0);
572 hammer_load_buffer(hammer_buffer_t buffer, int isnew)
574 hammer_volume_t volume;
579 * Load the buffer's on-disk info
581 volume = buffer->volume;
582 hammer_lock_ex(&buffer->io.lock);
583 KKASSERT(buffer->io.loading == 0);
584 buffer->io.loading = 1;
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->io.loading = 0;
594 hammer_unlock(&buffer->io.lock);
597 buffer->ondisk = ondisk = (void *)buffer->io.bp->b_data;
599 error = hammer_io_new(volume->devvp, &buffer->io);
603 if (error == 0 && isnew) {
604 hammer_modify_buffer(buffer, NULL, 0);
605 /* additional initialization goes here */
607 buffer->io.loading = 0;
608 hammer_unlock(&buffer->io.lock);
613 * NOTE: Called from RB_SCAN, must return >= 0 for scan to continue.
616 hammer_unload_buffer(hammer_buffer_t buffer, void *data __unused)
618 hammer_ref(&buffer->io.lock);
619 hammer_flush_buffer_nodes(buffer);
620 KKASSERT(buffer->io.lock.refs == 1);
621 hammer_rel_buffer(buffer, 2);
626 * Reference a buffer that is either already referenced or via a specially
627 * handled pointer (aka cursor->buffer).
630 hammer_ref_buffer(hammer_buffer_t buffer)
634 hammer_ref(&buffer->io.lock);
635 if (buffer->ondisk == NULL || buffer->io.loading) {
636 error = hammer_load_buffer(buffer, 0);
638 hammer_rel_buffer(buffer, 1);
640 * NOTE: buffer pointer can become stale after
651 * Release a buffer. We have to deal with several places where
652 * another thread can ref the buffer.
654 * Only destroy the structure itself if the related buffer cache buffer
655 * was disassociated from it. This ties the management of the structure
656 * to the buffer cache subsystem. buffer->ondisk determines whether the
657 * embedded io is referenced or not.
660 hammer_rel_buffer(hammer_buffer_t buffer, int flush)
662 hammer_volume_t volume;
664 if (buffer->io.lock.refs == 1) {
665 hammer_lock_ex(&buffer->io.lock);
666 if (buffer->io.lock.refs == 1) {
667 hammer_io_release(&buffer->io, flush);
669 if (buffer->io.bp == NULL &&
670 buffer->io.lock.refs == 1) {
671 hammer_flush_buffer_nodes(buffer);
672 KKASSERT(TAILQ_EMPTY(&buffer->clist));
673 volume = buffer->volume;
674 RB_REMOVE(hammer_buf_rb_tree,
675 &volume->rb_bufs_root, buffer);
676 buffer->volume = NULL; /* sanity */
677 --hammer_count_buffers;
678 kfree(buffer, M_HAMMER);
679 hammer_rel_volume(volume, 0);
683 hammer_io_flush(&buffer->io);
685 hammer_unlock(&buffer->io.lock);
687 hammer_unref(&buffer->io.lock);
691 * Access the filesystem buffer containing the specified hammer offset.
692 * buf_offset is a conglomeration of the volume number and vol_buf_beg
693 * relative buffer offset. It must also have bit 55 set to be valid.
694 * (see hammer_off_t in hammer_disk.h).
696 * Any prior buffer in *bufferp will be released and replaced by the
700 hammer_bread(hammer_mount_t hmp, hammer_off_t buf_offset, int *errorp,
701 struct hammer_buffer **bufferp)
703 hammer_buffer_t buffer;
704 int32_t xoff = (int32_t)buf_offset & HAMMER_BUFMASK;
706 buf_offset &= ~HAMMER_BUFMASK64;
709 if (buffer == NULL || buffer->buf_offset != buf_offset) {
711 hammer_rel_buffer(buffer, 0);
712 buffer = hammer_get_buffer(hmp, buf_offset, 0, errorp);
719 * Return a pointer to the buffer data.
724 return((char *)buffer->ondisk + xoff);
728 * Access the filesystem buffer containing the specified hammer offset.
729 * No disk read operation occurs. The result buffer may contain garbage.
731 * Any prior buffer in *bufferp will be released and replaced by the
735 hammer_bnew(hammer_mount_t hmp, hammer_off_t buf_offset, int *errorp,
736 struct hammer_buffer **bufferp)
738 hammer_buffer_t buffer;
739 int32_t xoff = (int32_t)buf_offset & HAMMER_BUFMASK;
741 buf_offset &= ~HAMMER_BUFMASK64;
744 if (buffer == NULL || buffer->buf_offset != buf_offset) {
746 hammer_rel_buffer(buffer, 0);
747 buffer = hammer_get_buffer(hmp, buf_offset, 1, errorp);
754 * Return a pointer to the buffer data.
759 return((char *)buffer->ondisk + xoff);
762 /************************************************************************
764 ************************************************************************
766 * Manage B-Tree nodes. B-Tree nodes represent the primary indexing
767 * method used by the HAMMER filesystem.
769 * Unlike other HAMMER structures, a hammer_node can be PASSIVELY
770 * associated with its buffer, and will only referenced the buffer while
771 * the node itself is referenced.
773 * A hammer_node can also be passively associated with other HAMMER
774 * structures, such as inodes, while retaining 0 references. These
775 * associations can be cleared backwards using a pointer-to-pointer in
778 * This allows the HAMMER implementation to cache hammer_nodes long-term
779 * and short-cut a great deal of the infrastructure's complexity. In
780 * most cases a cached node can be reacquired without having to dip into
781 * either the buffer or cluster management code.
783 * The caller must pass a referenced cluster on call and will retain
784 * ownership of the reference on return. The node will acquire its own
785 * additional references, if necessary.
788 hammer_get_node(hammer_mount_t hmp, hammer_off_t node_offset, int *errorp)
792 KKASSERT((node_offset & HAMMER_OFF_ZONE_MASK) == HAMMER_ZONE_BTREE);
795 * Locate the structure, allocating one if necessary.
798 node = RB_LOOKUP(hammer_nod_rb_tree, &hmp->rb_nods_root, node_offset);
800 ++hammer_count_nodes;
801 node = kmalloc(sizeof(*node), M_HAMMER, M_WAITOK|M_ZERO);
802 node->node_offset = node_offset;
804 if (RB_INSERT(hammer_nod_rb_tree, &hmp->rb_nods_root, node)) {
805 --hammer_count_nodes;
806 kfree(node, M_HAMMER);
810 hammer_ref(&node->lock);
811 *errorp = hammer_load_node(node);
813 hammer_rel_node(node);
820 * Reference an already-referenced node.
823 hammer_ref_node(hammer_node_t node)
827 KKASSERT(node->lock.refs > 0);
828 hammer_ref(&node->lock);
829 if ((error = hammer_load_node(node)) != 0)
830 hammer_rel_node(node);
835 * Load a node's on-disk data reference.
838 hammer_load_node(hammer_node_t node)
840 hammer_buffer_t buffer;
846 hammer_lock_ex(&node->lock);
847 if (node->ondisk == NULL) {
849 * This is a little confusing but the jist is that
850 * node->buffer determines whether the node is on
851 * the buffer's clist and node->ondisk determines
852 * whether the buffer is referenced.
854 if ((buffer = node->buffer) != NULL) {
855 error = hammer_ref_buffer(buffer);
857 buffer = hammer_get_buffer(node->hmp,
858 node->node_offset, 0,
861 KKASSERT(error == 0);
862 TAILQ_INSERT_TAIL(&buffer->clist,
864 node->buffer = buffer;
868 node->ondisk = (void *)((char *)buffer->ondisk +
869 (node->node_offset & HAMMER_BUFMASK));
872 hammer_unlock(&node->lock);
877 * Safely reference a node, interlock against flushes via the IO subsystem.
880 hammer_ref_node_safe(struct hammer_mount *hmp, struct hammer_node **cache,
885 if ((node = *cache) != NULL)
886 hammer_ref(&node->lock);
888 *errorp = hammer_load_node(node);
890 hammer_rel_node(node);
900 * Release a hammer_node. On the last release the node dereferences
901 * its underlying buffer and may or may not be destroyed.
904 hammer_rel_node(hammer_node_t node)
906 hammer_buffer_t buffer;
909 * If this isn't the last ref just decrement the ref count and
912 if (node->lock.refs > 1) {
913 hammer_unref(&node->lock);
918 * If there is no ondisk info or no buffer the node failed to load,
919 * remove the last reference and destroy the node.
921 if (node->ondisk == NULL) {
922 hammer_unref(&node->lock);
923 hammer_flush_node(node);
924 /* node is stale now */
929 * Do final cleanups and then either destroy the node and leave it
930 * passively cached. The buffer reference is removed regardless.
932 buffer = node->buffer;
935 if ((node->flags & (HAMMER_NODE_DELETED|HAMMER_NODE_FLUSH)) == 0) {
936 hammer_unref(&node->lock);
937 hammer_rel_buffer(buffer, 0);
942 * Destroy the node if it has been marked for deletion. We mark
943 * it as being free. Note that the disk space is physically
944 * freed when the fifo cycles back through the node.
946 if (node->flags & HAMMER_NODE_DELETED) {
947 hammer_blockmap_free(node->hmp, node->node_offset,
948 sizeof(*node->ondisk));
952 * Destroy the node. Record pertainant data because the node
953 * becomes stale the instant we flush it.
955 hammer_unref(&node->lock);
956 hammer_flush_node(node);
958 hammer_rel_buffer(buffer, 0);
962 * Passively cache a referenced hammer_node in *cache. The caller may
963 * release the node on return.
966 hammer_cache_node(hammer_node_t node, struct hammer_node **cache)
971 * If the node is being deleted, don't cache it!
973 if (node->flags & HAMMER_NODE_DELETED)
977 * Cache the node. If we previously cached a different node we
978 * have to give HAMMER a chance to destroy it.
981 if (node->cache1 != cache) {
982 if (node->cache2 != cache) {
983 if ((old = *cache) != NULL) {
984 KKASSERT(node->lock.refs != 0);
985 hammer_uncache_node(cache);
989 *node->cache2 = NULL;
990 node->cache2 = node->cache1;
991 node->cache1 = cache;
994 struct hammer_node **tmp;
996 node->cache1 = node->cache2;
1003 hammer_uncache_node(struct hammer_node **cache)
1007 if ((node = *cache) != NULL) {
1009 if (node->cache1 == cache) {
1010 node->cache1 = node->cache2;
1011 node->cache2 = NULL;
1012 } else if (node->cache2 == cache) {
1013 node->cache2 = NULL;
1015 panic("hammer_uncache_node: missing cache linkage");
1017 if (node->cache1 == NULL && node->cache2 == NULL)
1018 hammer_flush_node(node);
1023 * Remove a node's cache references and destroy the node if it has no
1024 * other references or backing store.
1027 hammer_flush_node(hammer_node_t node)
1029 hammer_buffer_t buffer;
1032 *node->cache1 = NULL;
1034 *node->cache2 = NULL;
1035 if (node->lock.refs == 0 && node->ondisk == NULL) {
1036 RB_REMOVE(hammer_nod_rb_tree, &node->hmp->rb_nods_root, node);
1037 if ((buffer = node->buffer) != NULL) {
1038 node->buffer = NULL;
1039 TAILQ_REMOVE(&buffer->clist, node, entry);
1040 /* buffer is unreferenced because ondisk is NULL */
1042 --hammer_count_nodes;
1043 kfree(node, M_HAMMER);
1048 * Flush passively cached B-Tree nodes associated with this buffer.
1049 * This is only called when the buffer is about to be destroyed, so
1050 * none of the nodes should have any references.
1053 hammer_flush_buffer_nodes(hammer_buffer_t buffer)
1057 while ((node = TAILQ_FIRST(&buffer->clist)) != NULL) {
1058 KKASSERT(node->lock.refs == 0 && node->ondisk == NULL);
1059 hammer_ref(&node->lock);
1060 node->flags |= HAMMER_NODE_FLUSH;
1061 hammer_rel_node(node);
1066 /************************************************************************
1068 ************************************************************************/
1071 * Allocate a B-Tree node.
1074 hammer_alloc_btree(hammer_mount_t hmp, int *errorp)
1076 hammer_buffer_t buffer = NULL;
1077 hammer_node_t node = NULL;
1078 hammer_off_t node_offset;
1080 node_offset = hammer_blockmap_alloc(hmp, HAMMER_ZONE_BTREE_INDEX,
1081 sizeof(struct hammer_node_ondisk),
1084 node = hammer_get_node(hmp, node_offset, errorp);
1085 hammer_modify_node(node);
1086 bzero(node->ondisk, sizeof(*node->ondisk));
1089 hammer_rel_buffer(buffer, 0);
1094 * The returned buffers are already appropriately marked as being modified.
1095 * If the caller marks them again unnecessary undo records may be generated.
1097 * In-band data is indicated by data_bufferp == NULL. Pass a data_len of 0
1098 * for zero-fill (caller modifies data_len afterwords).
1101 hammer_alloc_record(hammer_mount_t hmp,
1102 hammer_off_t *rec_offp, u_int8_t rec_type,
1103 struct hammer_buffer **rec_bufferp,
1104 int32_t data_len, void **datap,
1105 struct hammer_buffer **data_bufferp, int *errorp)
1107 hammer_record_ondisk_t rec;
1108 hammer_off_t rec_offset;
1109 hammer_off_t data_offset;
1116 * Allocate the record
1118 rec_offset = hammer_blockmap_alloc(hmp, HAMMER_ZONE_RECORD_INDEX,
1119 HAMMER_RECORD_SIZE, errorp);
1127 if (data_bufferp == NULL) {
1129 case HAMMER_RECTYPE_DATA:
1130 reclen = offsetof(struct hammer_data_record,
1133 case HAMMER_RECTYPE_DIRENTRY:
1134 reclen = offsetof(struct hammer_entry_record,
1138 panic("hammer_alloc_record: illegal "
1144 KKASSERT(reclen + data_len <= HAMMER_RECORD_SIZE);
1145 data_offset = rec_offset + reclen;
1146 } else if (data_len < HAMMER_BUFSIZE) {
1147 data_offset = hammer_blockmap_alloc(hmp,
1148 HAMMER_ZONE_SMALL_DATA_INDEX,
1151 data_offset = hammer_blockmap_alloc(hmp,
1152 HAMMER_ZONE_LARGE_DATA_INDEX,
1159 hammer_blockmap_free(hmp, rec_offset, HAMMER_RECORD_SIZE);
1164 * Basic return values.
1166 *rec_offp = rec_offset;
1167 rec = hammer_bread(hmp, rec_offset, errorp, rec_bufferp);
1168 KKASSERT(*errorp == 0);
1169 rec->base.data_off = data_offset;
1170 rec->base.data_len = data_len;
1171 hammer_modify_buffer(*rec_bufferp, NULL, 0);
1175 *datap = hammer_bread(hmp, data_offset, errorp,
1177 KKASSERT(*errorp == 0);
1178 hammer_modify_buffer(*data_bufferp, NULL, 0);
1182 } else if (data_len) {
1183 KKASSERT(data_offset + data_len - rec_offset <=
1184 HAMMER_RECORD_SIZE);
1186 *datap = (void *)((char *)rec +
1187 (int32_t)(data_offset - rec_offset));
1190 KKASSERT(datap == NULL);
1192 KKASSERT(*errorp == 0);
1197 * Generate an undo fifo entry and return the buffer to the caller (XXX).
1198 * The caller must create a dependancy to ensure that the undo record is
1199 * flushed before the modified buffer is flushed.
1202 hammer_generate_undo(hammer_mount_t hmp, hammer_off_t off, void *base, int len)
1206 hammer_off_t rec_offset;
1207 hammer_fifo_undo_t undo;
1208 hammer_buffer_t buffer = NULL;
1211 rec_offset = hammer_alloc_fifo(hmp, sizeof(*undo), len,
1212 &buffer, HAMMER_HEAD_TYPE_UNDO,
1215 undo = (void *)((char *)buffer->ondisk +
1216 ((int32_t)rec_offset & HAMMER_BUFMASK));
1217 undo->undo_offset = off;
1218 bcopy(base, undo + 1, len);
1221 hammer_rel_buffer(buffer, 0);
1229 * Allocate space from the FIFO. The first rec_len bytes will be zero'd.
1230 * The entire space is marked modified (the caller should not remark it as
1231 * that will cause unnecessary undo records to be added).
1235 hammer_alloc_fifo(hammer_mount_t hmp, int32_t rec_len, int32_t data_len,
1236 struct hammer_buffer **rec_bufferp, u_int16_t hdr_type,
1238 struct hammer_buffer **data2_bufferp, int *errorp)
1240 hammer_volume_t root_volume;
1241 hammer_volume_t end_volume;
1242 hammer_volume_ondisk_t ondisk;
1243 hammer_fifo_head_t head;
1244 hammer_fifo_tail_t tail;
1245 hammer_off_t end_off = 0;
1246 hammer_off_t tmp_off = 0;
1250 int32_t aligned_bytes;
1253 aligned_bytes = (rec_len + data_len + HAMMER_TAIL_ONDISK_SIZE +
1254 HAMMER_HEAD_ALIGN_MASK) & ~HAMMER_HEAD_ALIGN_MASK;
1256 root_volume = hammer_get_root_volume(hmp, errorp);
1258 hammer_modify_volume(root_volume, NULL, 0);
1260 while (root_volume) {
1261 ondisk = root_volume->ondisk;
1263 end_off = ondisk->vol0_fifo_end;
1264 end_vol_no = HAMMER_VOL_DECODE(end_off);
1266 end_volume = hammer_get_volume(hmp, end_vol_no, errorp);
1271 * Check to see if we ran out of space. Include some extra
1274 * vol0_fifo_end cannot be advanced into the same buffer
1275 * that vol0_fifo_beg resides in. This allows us to
1276 * instantiate a new buffer without reading it in.
1280 tmp_off = ondisk->vol0_fifo_beg & ~HAMMER_BUFMASK64;
1281 tmp_vol_no = HAMMER_VOL_DECODE(tmp_off);
1282 if ((tmp_off & HAMMER_OFF_SHORT_MASK) == 0) {
1283 if (end_vol_no + 1 == tmp_vol_no) {
1284 tmp_vol_no = end_vol_no;
1285 tmp_off = end_volume->maxbuf_off;
1286 } else if (end_vol_no + 1 == hmp->nvolumes &&
1288 tmp_vol_no = end_vol_no;
1289 tmp_off = end_volume->maxbuf_off;
1292 hammer_rel_volume(end_volume, 0);
1295 * XXX dummy head at end of fifo
1297 if (end_vol_no == tmp_vol_no &&
1298 end_off < tmp_off &&
1299 end_off + aligned_bytes + sizeof(*head) >= tmp_off) {
1304 if ((int32_t)end_off & HAMMER_BUFMASK)
1305 head = hammer_bread(hmp, end_off, errorp, rec_bufferp);
1307 head = hammer_bnew(hmp, end_off, errorp, rec_bufferp);
1312 * Load the buffer, retry if someone else squeeked in
1313 * while we were blocked.
1316 if (ondisk->vol0_fifo_end != end_off)
1320 * Ok, we're gonna do something. Modify the buffer
1322 hammer_modify_buffer(*rec_bufferp, NULL, 0);
1323 if (ondisk->vol0_fifo_end != end_off)
1325 xoff = (int32_t)end_off & HAMMER_BUFMASK;
1328 * The non-data portion of the fifo record cannot cross
1329 * a buffer boundary.
1331 * The entire record cannot cross a buffer boundary if
1334 * The entire record cannot cover more then two whole buffers
1335 * regardless. Even if the data portion is 16K, this case
1336 * can occur due to the addition of the fifo_tail.
1338 * It is illegal for a record to cross a volume boundary.
1340 * It is illegal for a record to cross a recovery boundary
1341 * (this is so recovery code is guaranteed a record rather
1342 * then data at certain points).
1344 * Add a pad record and loop if it does.
1347 if (xoff + rec_len > HAMMER_BUFSIZE)
1349 if (can_cross == 0) {
1350 if (xoff + aligned_bytes > HAMMER_BUFSIZE)
1353 if (xoff + aligned_bytes > HAMMER_BUFSIZE &&
1354 (end_off + aligned_bytes) >=
1355 (*rec_bufferp)->volume->maxbuf_off) {
1358 if ((end_off ^ (end_off + aligned_bytes)) &
1359 HAMMER_OFF_SHORT_REC_MASK) {
1362 if (xoff + aligned_bytes - HAMMER_BUFSIZE >
1364 KKASSERT(xoff != 0);
1370 * Pad to end of the buffer if necessary. PADs can be
1371 * squeezed into as little as 8 bytes (hence our alignment
1372 * requirement). The crc, reserved, and sequence number
1373 * fields are not used, but initialize them anyway if there
1377 xoff = HAMMER_BUFSIZE - xoff;
1378 head->hdr_signature = HAMMER_HEAD_SIGNATURE;
1379 head->hdr_type = HAMMER_HEAD_TYPE_PAD;
1380 head->hdr_size = xoff;
1381 if (xoff >= HAMMER_HEAD_ONDISK_SIZE +
1382 HAMMER_TAIL_ONDISK_SIZE) {
1384 head->hdr_reserved02 = 0;
1388 tail = (void *)((char *)head + xoff -
1389 HAMMER_TAIL_ONDISK_SIZE);
1390 if ((void *)head != (void *)tail) {
1391 tail->tail_signature = HAMMER_TAIL_SIGNATURE;
1392 tail->tail_type = HAMMER_HEAD_TYPE_PAD;
1393 tail->tail_size = xoff;
1395 KKASSERT((xoff & HAMMER_HEAD_ALIGN_MASK) == 0);
1396 ondisk->vol0_fifo_end =
1397 hammer_advance_fifo((*rec_bufferp)->volume,
1402 if (xoff + aligned_bytes > HAMMER_BUFSIZE) {
1403 xoff = xoff + aligned_bytes - HAMMER_BUFSIZE;
1405 KKASSERT(xoff <= HAMMER_BUFSIZE);
1406 tail = hammer_bnew(hmp, end_off + aligned_bytes -
1407 HAMMER_TAIL_ONDISK_SIZE,
1408 errorp, data2_bufferp);
1409 hammer_modify_buffer(*data2_bufferp, NULL, 0);
1414 * Retry if someone else appended to the fifo while
1417 if (ondisk->vol0_fifo_end != end_off)
1420 tail = (void *)((char *)head + aligned_bytes -
1421 HAMMER_TAIL_ONDISK_SIZE);
1424 bzero(head, rec_len);
1425 head->hdr_signature = HAMMER_HEAD_SIGNATURE;
1426 head->hdr_type = hdr_type;
1427 head->hdr_size = aligned_bytes;
1429 head->hdr_seq = root_volume->ondisk->vol0_next_seq++;
1431 tail->tail_signature = HAMMER_TAIL_SIGNATURE;
1432 tail->tail_type = hdr_type;
1433 tail->tail_size = aligned_bytes;
1435 ondisk->vol0_fifo_end =
1436 hammer_advance_fifo((*rec_bufferp)->volume,
1437 end_off, aligned_bytes);
1439 hammer_rel_volume(root_volume, 0);
1448 * Mark a fifo record as having been freed. XXX needs undo.
1451 hammer_free_fifo(hammer_mount_t hmp, hammer_off_t fifo_offset)
1453 hammer_buffer_t buffer = NULL;
1454 hammer_fifo_head_t head;
1457 head = hammer_bread(hmp, fifo_offset, &error, &buffer);
1459 hammer_modify_buffer(buffer, &head->hdr_type,
1460 sizeof(head->hdr_type));
1461 head->hdr_type |= HAMMER_HEAD_FLAG_FREE;
1464 hammer_rel_buffer(buffer, 0);
1468 * Attempt to rewind the FIFO
1470 * This routine is allowed to do nothing.
1473 hammer_unwind_fifo(hammer_mount_t hmp, hammer_off_t rec_offset)
1478 * Advance the FIFO a certain number of bytes.
1482 hammer_advance_fifo(hammer_volume_t volume, hammer_off_t off, int32_t bytes)
1487 KKASSERT(off <= volume->maxbuf_off);
1488 KKASSERT((off & HAMMER_OFF_ZONE_MASK) == HAMMER_ZONE_RAW_BUFFER);
1489 if (off == volume->maxbuf_off) {
1490 vol_no = volume->vol_no + 1;
1491 if (vol_no == volume->hmp->nvolumes)
1493 off = HAMMER_ENCODE_RAW_BUFFER(vol_no, 0);
1500 * Sync dirty buffers to the media
1503 static int hammer_sync_scan1(struct mount *mp, struct vnode *vp, void *data);
1504 static int hammer_sync_scan2(struct mount *mp, struct vnode *vp, void *data);
1507 hammer_sync_hmp(hammer_mount_t hmp, int waitfor)
1509 struct hammer_sync_info info;
1512 info.waitfor = waitfor;
1514 vmntvnodescan(hmp->mp, VMSC_GETVP|VMSC_NOWAIT,
1515 hammer_sync_scan1, hammer_sync_scan2, &info);
1517 RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
1518 hammer_sync_volume, &info);
1523 hammer_sync_scan1(struct mount *mp, struct vnode *vp, void *data)
1525 struct hammer_inode *ip;
1528 if (vp->v_type == VNON || ip == NULL ||
1529 ((ip->flags & HAMMER_INODE_MODMASK) == 0 &&
1530 RB_EMPTY(&vp->v_rbdirty_tree))) {
1537 hammer_sync_scan2(struct mount *mp, struct vnode *vp, void *data)
1539 struct hammer_sync_info *info = data;
1540 struct hammer_inode *ip;
1544 if (vp->v_type == VNON || vp->v_type == VBAD ||
1545 ((ip->flags & HAMMER_INODE_MODMASK) == 0 &&
1546 RB_EMPTY(&vp->v_rbdirty_tree))) {
1549 error = VOP_FSYNC(vp, info->waitfor);
1551 info->error = error;
1556 hammer_sync_volume(hammer_volume_t volume, void *data)
1558 struct hammer_sync_info *info = data;
1560 hammer_ref(&volume->io.lock);
1561 RB_SCAN(hammer_buf_rb_tree, &volume->rb_bufs_root, NULL,
1562 hammer_sync_buffer, info);
1563 hammer_rel_volume(volume, 1);
1568 hammer_sync_buffer(hammer_buffer_t buffer, void *data __unused)
1570 hammer_ref(&buffer->io.lock);
1571 hammer_rel_buffer(buffer, 1);
1577 * Generic buffer initialization. Initialize the A-list into an all-allocated
1578 * state with the free block limit properly set.
1580 * Note that alloc_new_buffer() will free the appropriate block range via
1581 * the appropriate cluster alist, so the free count is properly propogated.
1584 hammer_init_fifo(hammer_fifo_head_t head, u_int16_t type)
1586 head->hdr_signature = HAMMER_HEAD_SIGNATURE;
1587 head->hdr_type = type;