2 * Copyright (c) 2011-2013 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 #include <sys/param.h>
35 #include <sys/systm.h>
36 #include <sys/kernel.h>
37 #include <sys/nlookup.h>
38 #include <sys/vnode.h>
39 #include <sys/mount.h>
40 #include <sys/fcntl.h>
43 #include <sys/vfsops.h>
44 #include <sys/sysctl.h>
45 #include <sys/socket.h>
48 #include "hammer2_disk.h"
49 #include "hammer2_mount.h"
51 #define REPORT_REFS_ERRORS 1 /* XXX remove me */
53 struct hammer2_sync_info {
54 hammer2_trans_t trans;
59 TAILQ_HEAD(hammer2_mntlist, hammer2_mount);
60 static struct hammer2_mntlist hammer2_mntlist;
61 static struct lock hammer2_mntlk;
64 int hammer2_cluster_enable = 1;
65 int hammer2_hardlink_enable = 1;
66 long hammer2_iod_file_read;
67 long hammer2_iod_meta_read;
68 long hammer2_iod_indr_read;
69 long hammer2_iod_fmap_read;
70 long hammer2_iod_volu_read;
71 long hammer2_iod_file_write;
72 long hammer2_iod_meta_write;
73 long hammer2_iod_indr_write;
74 long hammer2_iod_fmap_write;
75 long hammer2_iod_volu_write;
76 long hammer2_ioa_file_read;
77 long hammer2_ioa_meta_read;
78 long hammer2_ioa_indr_read;
79 long hammer2_ioa_fmap_read;
80 long hammer2_ioa_volu_read;
81 long hammer2_ioa_fmap_write;
82 long hammer2_ioa_file_write;
83 long hammer2_ioa_meta_write;
84 long hammer2_ioa_indr_write;
85 long hammer2_ioa_volu_write;
87 SYSCTL_NODE(_vfs, OID_AUTO, hammer2, CTLFLAG_RW, 0, "HAMMER2 filesystem");
89 SYSCTL_INT(_vfs_hammer2, OID_AUTO, debug, CTLFLAG_RW,
90 &hammer2_debug, 0, "");
91 SYSCTL_INT(_vfs_hammer2, OID_AUTO, cluster_enable, CTLFLAG_RW,
92 &hammer2_cluster_enable, 0, "");
93 SYSCTL_INT(_vfs_hammer2, OID_AUTO, hardlink_enable, CTLFLAG_RW,
94 &hammer2_hardlink_enable, 0, "");
96 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_file_read, CTLFLAG_RW,
97 &hammer2_iod_file_read, 0, "");
98 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_meta_read, CTLFLAG_RW,
99 &hammer2_iod_meta_read, 0, "");
100 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_indr_read, CTLFLAG_RW,
101 &hammer2_iod_indr_read, 0, "");
102 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_fmap_read, CTLFLAG_RW,
103 &hammer2_iod_fmap_read, 0, "");
104 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_volu_read, CTLFLAG_RW,
105 &hammer2_iod_volu_read, 0, "");
107 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_file_write, CTLFLAG_RW,
108 &hammer2_iod_file_write, 0, "");
109 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_meta_write, CTLFLAG_RW,
110 &hammer2_iod_meta_write, 0, "");
111 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_indr_write, CTLFLAG_RW,
112 &hammer2_iod_indr_write, 0, "");
113 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_fmap_write, CTLFLAG_RW,
114 &hammer2_iod_fmap_write, 0, "");
115 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_volu_write, CTLFLAG_RW,
116 &hammer2_iod_volu_write, 0, "");
118 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_file_read, CTLFLAG_RW,
119 &hammer2_ioa_file_read, 0, "");
120 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_meta_read, CTLFLAG_RW,
121 &hammer2_ioa_meta_read, 0, "");
122 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_indr_read, CTLFLAG_RW,
123 &hammer2_ioa_indr_read, 0, "");
124 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_fmap_read, CTLFLAG_RW,
125 &hammer2_ioa_fmap_read, 0, "");
126 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_volu_read, CTLFLAG_RW,
127 &hammer2_ioa_volu_read, 0, "");
129 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_file_write, CTLFLAG_RW,
130 &hammer2_ioa_file_write, 0, "");
131 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_meta_write, CTLFLAG_RW,
132 &hammer2_ioa_meta_write, 0, "");
133 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_indr_write, CTLFLAG_RW,
134 &hammer2_ioa_indr_write, 0, "");
135 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_fmap_write, CTLFLAG_RW,
136 &hammer2_ioa_fmap_write, 0, "");
137 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_volu_write, CTLFLAG_RW,
138 &hammer2_ioa_volu_write, 0, "");
140 static int hammer2_vfs_init(struct vfsconf *conf);
141 static int hammer2_vfs_mount(struct mount *mp, char *path, caddr_t data,
143 static int hammer2_remount(struct mount *, char *, struct vnode *,
145 static int hammer2_vfs_unmount(struct mount *mp, int mntflags);
146 static int hammer2_vfs_root(struct mount *mp, struct vnode **vpp);
147 static int hammer2_vfs_statfs(struct mount *mp, struct statfs *sbp,
149 static int hammer2_vfs_statvfs(struct mount *mp, struct statvfs *sbp,
151 static int hammer2_vfs_sync(struct mount *mp, int waitfor);
152 static int hammer2_vfs_vget(struct mount *mp, struct vnode *dvp,
153 ino_t ino, struct vnode **vpp);
154 static int hammer2_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
155 struct fid *fhp, struct vnode **vpp);
156 static int hammer2_vfs_vptofh(struct vnode *vp, struct fid *fhp);
157 static int hammer2_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
158 int *exflagsp, struct ucred **credanonp);
160 static int hammer2_install_volume_header(hammer2_mount_t *hmp);
161 static int hammer2_sync_scan1(struct mount *mp, struct vnode *vp, void *data);
162 static int hammer2_sync_scan2(struct mount *mp, struct vnode *vp, void *data);
164 static int hammer2_rcvdmsg(kdmsg_msg_t *msg);
165 static void hammer2_autodmsg(kdmsg_msg_t *msg);
168 * HAMMER2 vfs operations.
170 static struct vfsops hammer2_vfsops = {
171 .vfs_init = hammer2_vfs_init,
172 .vfs_sync = hammer2_vfs_sync,
173 .vfs_mount = hammer2_vfs_mount,
174 .vfs_unmount = hammer2_vfs_unmount,
175 .vfs_root = hammer2_vfs_root,
176 .vfs_statfs = hammer2_vfs_statfs,
177 .vfs_statvfs = hammer2_vfs_statvfs,
178 .vfs_vget = hammer2_vfs_vget,
179 .vfs_vptofh = hammer2_vfs_vptofh,
180 .vfs_fhtovp = hammer2_vfs_fhtovp,
181 .vfs_checkexp = hammer2_vfs_checkexp
184 MALLOC_DEFINE(M_HAMMER2, "HAMMER2-mount", "");
186 VFS_SET(hammer2_vfsops, hammer2, 0);
187 MODULE_VERSION(hammer2, 1);
191 hammer2_vfs_init(struct vfsconf *conf)
197 if (HAMMER2_BLOCKREF_BYTES != sizeof(struct hammer2_blockref))
199 if (HAMMER2_INODE_BYTES != sizeof(struct hammer2_inode_data))
201 if (HAMMER2_VOLUME_BYTES != sizeof(struct hammer2_volume_data))
205 kprintf("HAMMER2 structure size mismatch; cannot continue.\n");
207 lockinit(&hammer2_mntlk, "mntlk", 0, 0);
208 TAILQ_INIT(&hammer2_mntlist);
214 * Mount or remount HAMMER2 fileystem from physical media
217 * mp mount point structure
223 * mp mount point structure
224 * path path to mount point
225 * data pointer to argument structure in user space
226 * volume volume path (device@LABEL form)
227 * hflags user mount flags
228 * cred user credentials
235 hammer2_vfs_mount(struct mount *mp, char *path, caddr_t data,
238 struct hammer2_mount_info info;
239 hammer2_pfsmount_t *pmp;
240 hammer2_mount_t *hmp;
243 struct nlookupdata nd;
244 hammer2_chain_t *parent;
245 hammer2_chain_t *schain;
246 hammer2_chain_t *rchain;
248 char devstr[MNAMELEN];
263 kprintf("hammer2_mount\n");
269 bzero(&info, sizeof(info));
270 info.cluster_fd = -1;
274 * Non-root mount or updating a mount
276 error = copyin(data, &info, sizeof(info));
280 error = copyinstr(info.volume, devstr, MNAMELEN - 1, &done);
284 /* Extract device and label */
286 label = strchr(devstr, '@');
288 ((label + 1) - dev) > done) {
296 if (mp->mnt_flag & MNT_UPDATE) {
298 /* HAMMER2 implements NFS export via mountctl */
301 error = hammer2_remount(mp, path, devvp, cred);
309 * Lookup name and verify it refers to a block device.
311 error = nlookup_init(&nd, dev, UIO_SYSSPACE, NLC_FOLLOW);
313 error = nlookup(&nd);
315 error = cache_vref(&nd.nl_nch, nd.nl_cred, &devvp);
319 if (vn_isdisk(devvp, &error))
320 error = vfs_mountedon(devvp);
324 * Determine if the device has already been mounted. After this
325 * check hmp will be non-NULL if we are doing the second or more
326 * hammer2 mounts from the same device.
328 lockmgr(&hammer2_mntlk, LK_EXCLUSIVE);
329 TAILQ_FOREACH(hmp, &hammer2_mntlist, mntentry) {
330 if (hmp->devvp == devvp)
335 * Open the device if this isn't a secondary mount
341 if (error == 0 && vcount(devvp) > 0)
345 * Now open the device
348 ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
349 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
350 error = vinvalbuf(devvp, V_SAVE, 0, 0);
352 error = VOP_OPEN(devvp,
353 ronly ? FREAD : FREAD | FWRITE,
358 if (error && devvp) {
363 lockmgr(&hammer2_mntlk, LK_RELEASE);
369 * Block device opened successfully, finish initializing the
372 * From this point on we have to call hammer2_unmount() on failure.
374 pmp = kmalloc(sizeof(*pmp), M_HAMMER2, M_WAITOK | M_ZERO);
375 mp->mnt_data = (qaddr_t)pmp;
378 spin_init(&pmp->inum_spin);
379 RB_INIT(&pmp->inum_tree);
381 kmalloc_create(&pmp->mmsg, "HAMMER2-pfsmsg");
382 kdmsg_iocom_init(&pmp->iocom, pmp,
383 KDMSG_IOCOMF_AUTOCONN |
384 KDMSG_IOCOMF_AUTOSPAN |
385 KDMSG_IOCOMF_AUTOCIRC,
386 pmp->mmsg, hammer2_rcvdmsg);
389 hmp = kmalloc(sizeof(*hmp), M_HAMMER2, M_WAITOK | M_ZERO);
392 kmalloc_create(&hmp->minode, "HAMMER2-inodes");
393 kmalloc_create(&hmp->mchain, "HAMMER2-chains");
394 TAILQ_INSERT_TAIL(&hammer2_mntlist, hmp, mntentry);
396 ccms_domain_init(&pmp->ccms_dom);
399 lockmgr(&hammer2_mntlk, LK_RELEASE);
400 kprintf("hammer2_mount hmp=%p pmpcnt=%d\n", hmp, hmp->pmp_count);
402 mp->mnt_flag = MNT_LOCAL;
403 mp->mnt_kern_flag |= MNTK_ALL_MPSAFE; /* all entry pts are SMP */
409 lockinit(&hmp->alloclk, "h2alloc", 0, 0);
410 lockinit(&hmp->voldatalk, "voldata", 0, LK_CANRECURSE);
411 TAILQ_INIT(&hmp->transq);
414 * vchain setup. vchain.data is embedded.
415 * vchain.refs is initialized and will never drop to 0.
417 hmp->vchain.hmp = hmp;
418 hmp->vchain.refs = 1;
419 hmp->vchain.data = (void *)&hmp->voldata;
420 hmp->vchain.bref.type = HAMMER2_BREF_TYPE_VOLUME;
421 hmp->vchain.bref.data_off = 0 | HAMMER2_PBUFRADIX;
422 hmp->vchain.delete_tid = HAMMER2_MAX_TID;
423 hammer2_chain_core_alloc(&hmp->vchain, NULL);
424 /* hmp->vchain.u.xxx is left NULL */
427 * fchain setup. fchain.data is embedded.
428 * fchain.refs is initialized and will never drop to 0.
430 * The data is not used but needs to be initialized to
431 * pass assertion muster. We use this chain primarily
432 * as a placeholder for the freemap's top-level RBTREE
433 * so it does not interfere with the volume's topology
436 hmp->fchain.hmp = hmp;
437 hmp->fchain.refs = 1;
438 hmp->fchain.data = (void *)&hmp->voldata.freemap_blockset;
439 hmp->fchain.bref.type = HAMMER2_BREF_TYPE_FREEMAP;
440 hmp->fchain.bref.data_off = 0 | HAMMER2_PBUFRADIX;
441 hmp->fchain.bref.methods =
442 HAMMER2_ENC_CHECK(HAMMER2_CHECK_FREEMAP) |
443 HAMMER2_ENC_COMP(HAMMER2_COMP_NONE);
444 hmp->fchain.delete_tid = HAMMER2_MAX_TID;
446 hammer2_chain_core_alloc(&hmp->fchain, NULL);
447 /* hmp->fchain.u.xxx is left NULL */
450 * Install the volume header
452 error = hammer2_install_volume_header(hmp);
454 hammer2_vfs_unmount(mp, MNT_FORCE);
460 * required mount structure initializations
462 mp->mnt_stat.f_iosize = HAMMER2_PBUFSIZE;
463 mp->mnt_stat.f_bsize = HAMMER2_PBUFSIZE;
465 mp->mnt_vstat.f_frsize = HAMMER2_PBUFSIZE;
466 mp->mnt_vstat.f_bsize = HAMMER2_PBUFSIZE;
471 mp->mnt_iosize_max = MAXPHYS;
474 * First locate the super-root inode, which is key 0 relative to the
475 * volume header's blockset.
477 * Then locate the root inode by scanning the directory keyspace
478 * represented by the label.
481 parent = hammer2_chain_lookup_init(&hmp->vchain, 0);
482 schain = hammer2_chain_lookup(&parent,
483 HAMMER2_SROOT_KEY, HAMMER2_SROOT_KEY, 0);
484 hammer2_chain_lookup_done(parent);
485 if (schain == NULL) {
486 kprintf("hammer2_mount: invalid super-root\n");
487 hammer2_vfs_unmount(mp, MNT_FORCE);
490 hammer2_chain_ref(schain); /* for hmp->schain */
491 hmp->schain = schain; /* left locked for inode_get */
492 hmp->sroot = hammer2_inode_get(hmp, NULL, NULL, schain);
493 hammer2_inode_ref(hmp->sroot); /* for hmp->sroot */
494 hammer2_inode_unlock_ex(hmp->sroot, schain);
496 schain = hmp->schain;
500 * schain only has 1 ref now for its hmp->schain assignment.
501 * Setup for lookup (which will lock it).
503 parent = hammer2_chain_lookup_init(schain, 0);
504 lhc = hammer2_dirhash(label, strlen(label));
505 rchain = hammer2_chain_lookup(&parent,
506 lhc, lhc + HAMMER2_DIRHASH_LOMASK,
509 if (rchain->bref.type == HAMMER2_BREF_TYPE_INODE &&
510 strcmp(label, rchain->data->ipdata.filename) == 0) {
513 rchain = hammer2_chain_next(&parent, rchain,
514 lhc, lhc + HAMMER2_DIRHASH_LOMASK,
517 hammer2_chain_lookup_done(parent);
518 if (rchain == NULL) {
519 kprintf("hammer2_mount: PFS label not found\n");
520 hammer2_vfs_unmount(mp, MNT_FORCE);
523 if (rchain->flags & HAMMER2_CHAIN_MOUNTED) {
524 hammer2_chain_unlock(rchain);
525 kprintf("hammer2_mount: PFS label already mounted!\n");
526 hammer2_vfs_unmount(mp, MNT_FORCE);
529 if (rchain->flags & HAMMER2_CHAIN_RECYCLE) {
530 kprintf("hammer2_mount: PFS label currently recycling\n");
531 hammer2_vfs_unmount(mp, MNT_FORCE);
535 atomic_set_int(&rchain->flags, HAMMER2_CHAIN_MOUNTED);
538 * NOTE: *_get() integrates chain's lock into the inode lock.
540 hammer2_chain_ref(rchain); /* for pmp->rchain */
541 pmp->rchain = rchain; /* left held & unlocked */
542 pmp->iroot = hammer2_inode_get(hmp, pmp, NULL, rchain);
543 hammer2_inode_ref(pmp->iroot); /* ref for pmp->iroot */
544 hammer2_inode_unlock_ex(pmp->iroot, rchain);
546 kprintf("iroot %p\n", pmp->iroot);
549 * Ref the cluster management messaging descriptor. The mount
550 * program deals with the other end of the communications pipe.
552 fp = holdfp(curproc->p_fd, info.cluster_fd, -1);
554 kprintf("hammer2_mount: bad cluster_fd!\n");
555 hammer2_vfs_unmount(mp, MNT_FORCE);
558 hammer2_cluster_reconnect(pmp, fp);
564 vfs_add_vnodeops(mp, &hammer2_vnode_vops, &mp->mnt_vn_norm_ops);
565 vfs_add_vnodeops(mp, &hammer2_spec_vops, &mp->mnt_vn_spec_ops);
566 vfs_add_vnodeops(mp, &hammer2_fifo_vops, &mp->mnt_vn_fifo_ops);
568 copyinstr(info.volume, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, &size);
569 bzero(mp->mnt_stat.f_mntfromname + size, MNAMELEN - size);
570 bzero(mp->mnt_stat.f_mntonname, sizeof(mp->mnt_stat.f_mntonname));
571 copyinstr(path, mp->mnt_stat.f_mntonname,
572 sizeof(mp->mnt_stat.f_mntonname) - 1,
576 * Initial statfs to prime mnt_stat.
578 hammer2_vfs_statfs(mp, &mp->mnt_stat, cred);
585 hammer2_remount(struct mount *mp, char *path, struct vnode *devvp,
593 hammer2_vfs_unmount(struct mount *mp, int mntflags)
595 hammer2_pfsmount_t *pmp;
596 hammer2_mount_t *hmp;
599 int ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
607 if (mntflags & MNT_FORCE)
610 hammer2_mount_exlock(hmp);
613 * If mount initialization proceeded far enough we must flush
617 error = vflush(mp, 0, flags);
620 hammer2_mount_unlock(hmp);
624 lockmgr(&hammer2_mntlk, LK_EXCLUSIVE);
626 kprintf("hammer2_unmount hmp=%p pmpcnt=%d\n", hmp, hmp->pmp_count);
629 * Flush any left over chains. The voldata lock is only used
630 * to synchronize against HAMMER2_CHAIN_MODIFIED_AUX.
632 hammer2_voldata_lock(hmp);
633 if (hmp->vchain.flags & (HAMMER2_CHAIN_MODIFIED |
634 HAMMER2_CHAIN_SUBMODIFIED)) {
635 hammer2_voldata_unlock(hmp, 0);
636 hammer2_vfs_sync(mp, MNT_WAIT);
638 hammer2_voldata_unlock(hmp, 0);
640 if (hmp->pmp_count == 0) {
641 if (hmp->vchain.flags & (HAMMER2_CHAIN_MODIFIED |
642 HAMMER2_CHAIN_SUBMODIFIED)) {
643 kprintf("hammer2_unmount: chains left over after "
645 if (hammer2_debug & 0x0010)
646 Debugger("entered debugger");
651 * Cleanup the root and super-root chain elements (which should be
655 #if REPORT_REFS_ERRORS
656 if (pmp->iroot->refs != 1)
657 kprintf("PMP->IROOT %p REFS WRONG %d\n",
658 pmp->iroot, pmp->iroot->refs);
660 KKASSERT(pmp->iroot->refs == 1);
662 hammer2_inode_drop(pmp->iroot); /* ref for pmp->iroot */
666 atomic_clear_int(&pmp->rchain->flags, HAMMER2_CHAIN_MOUNTED);
667 #if REPORT_REFS_ERRORS
668 if (pmp->rchain->refs != 1)
669 kprintf("PMP->RCHAIN %p REFS WRONG %d\n",
670 pmp->rchain, pmp->rchain->refs);
672 KKASSERT(pmp->rchain->refs == 1);
674 hammer2_chain_drop(pmp->rchain);
677 ccms_domain_uninit(&pmp->ccms_dom);
680 * Kill cluster controller
682 kdmsg_iocom_uninit(&pmp->iocom);
685 * If no PFS's left drop the master hammer2_mount for the device.
687 if (hmp->pmp_count == 0) {
689 hammer2_inode_drop(hmp->sroot);
693 #if REPORT_REFS_ERRORS
694 if (hmp->schain->refs != 1)
695 kprintf("HMP->SCHAIN %p REFS WRONG %d\n",
696 hmp->schain, hmp->schain->refs);
698 KKASSERT(hmp->schain->refs == 1);
700 hammer2_chain_drop(hmp->schain);
705 * Finish up with the device vnode
707 if ((devvp = hmp->devvp) != NULL) {
708 vinvalbuf(devvp, (ronly ? 0 : V_SAVE), 0, 0);
710 VOP_CLOSE(devvp, (ronly ? FREAD : FREAD|FWRITE));
716 * Final drop of embedded freemap root chain to clean up
717 * fchain.core (fchain structure is not flagged ALLOCATED
718 * so it is cleaned out and then left to rot).
720 hammer2_chain_drop(&hmp->fchain);
723 * Final drop of embedded volume root chain to clean up
724 * vchain.core (vchain structure is not flagged ALLOCATED
725 * so it is cleaned out and then left to rot).
728 hammer2_dump_chain(&hmp->vchain, 0, &dumpcnt);
729 hammer2_mount_unlock(hmp);
730 hammer2_chain_drop(&hmp->vchain);
732 hammer2_mount_unlock(hmp);
739 kmalloc_destroy(&pmp->mmsg);
741 kfree(pmp, M_HAMMER2);
742 if (hmp->pmp_count == 0) {
743 TAILQ_REMOVE(&hammer2_mntlist, hmp, mntentry);
744 kmalloc_destroy(&hmp->minode);
745 kmalloc_destroy(&hmp->mchain);
746 kfree(hmp, M_HAMMER2);
748 lockmgr(&hammer2_mntlk, LK_RELEASE);
754 hammer2_vfs_vget(struct mount *mp, struct vnode *dvp,
755 ino_t ino, struct vnode **vpp)
757 kprintf("hammer2_vget\n");
763 hammer2_vfs_root(struct mount *mp, struct vnode **vpp)
765 hammer2_pfsmount_t *pmp;
766 hammer2_mount_t *hmp;
767 hammer2_chain_t *parent;
773 hammer2_mount_exlock(hmp);
774 if (pmp->iroot == NULL) {
778 parent = hammer2_inode_lock_sh(pmp->iroot);
779 vp = hammer2_igetv(pmp->iroot, &error);
780 hammer2_inode_unlock_sh(pmp->iroot, parent);
783 kprintf("vnodefail\n");
785 hammer2_mount_unlock(hmp);
793 * XXX incorporate ipdata->inode_quota and data_quota
797 hammer2_vfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
799 hammer2_pfsmount_t *pmp;
800 hammer2_mount_t *hmp;
805 mp->mnt_stat.f_files = pmp->inode_count;
806 mp->mnt_stat.f_ffree = 0;
807 mp->mnt_stat.f_blocks = hmp->voldata.allocator_size / HAMMER2_PBUFSIZE;
808 mp->mnt_stat.f_bfree = (hmp->voldata.allocator_size -
809 hmp->voldata.allocator_beg) / HAMMER2_PBUFSIZE;
810 mp->mnt_stat.f_bavail = mp->mnt_stat.f_bfree;
818 hammer2_vfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred)
820 hammer2_pfsmount_t *pmp;
821 hammer2_mount_t *hmp;
826 mp->mnt_vstat.f_bsize = HAMMER2_PBUFSIZE;
827 mp->mnt_vstat.f_files = pmp->inode_count;
828 mp->mnt_vstat.f_ffree = 0;
829 mp->mnt_vstat.f_blocks = hmp->voldata.allocator_size / HAMMER2_PBUFSIZE;
830 mp->mnt_vstat.f_bfree = (hmp->voldata.allocator_size -
831 hmp->voldata.allocator_beg) / HAMMER2_PBUFSIZE;
832 mp->mnt_vstat.f_bavail = mp->mnt_vstat.f_bfree;
834 *sbp = mp->mnt_vstat;
839 * Sync the entire filesystem; this is called from the filesystem syncer
840 * process periodically and whenever a user calls sync(1) on the hammer
843 * Currently is actually called from the syncer! \o/
845 * This task will have to snapshot the state of the dirty inode chain.
846 * From that, it will have to make sure all of the inodes on the dirty
847 * chain have IO initiated. We make sure that io is initiated for the root
850 * If waitfor is set, we wait for media to acknowledge the new rootblock.
852 * THINKS: side A vs side B, to have sync not stall all I/O?
856 hammer2_vfs_sync(struct mount *mp, int waitfor)
858 struct hammer2_sync_info info;
859 hammer2_mount_t *hmp;
869 if ((waitfor & MNT_LAZY) == 0) {
871 hammer2_dump_chain(&hmp->vchain, 0, &dumpcnt);
876 if (waitfor & MNT_LAZY)
877 flags |= VMSC_ONEPASS;
879 hammer2_trans_init(&info.trans, hmp, NULL, HAMMER2_TRANS_ISFLUSH);
882 info.waitfor = MNT_NOWAIT;
883 vmntvnodescan(mp, flags | VMSC_NOWAIT,
885 hammer2_sync_scan2, &info);
886 if (info.error == 0 && (waitfor & MNT_WAIT)) {
887 info.waitfor = waitfor;
888 vmntvnodescan(mp, flags,
890 hammer2_sync_scan2, &info);
894 if (waitfor == MNT_WAIT) {
900 hammer2_chain_lock(&hmp->vchain, HAMMER2_RESOLVE_ALWAYS);
901 if (hmp->vchain.flags & (HAMMER2_CHAIN_MODIFIED |
902 HAMMER2_CHAIN_SUBMODIFIED)) {
903 hammer2_chain_flush(&info.trans, &hmp->vchain);
905 hammer2_chain_unlock(&hmp->vchain);
909 * Rollup flush. The fsyncs above basically just flushed
910 * data blocks. The flush below gets all the meta-data.
912 hammer2_chain_lock(&hmp->fchain, HAMMER2_RESOLVE_ALWAYS);
913 if (hmp->fchain.flags & (HAMMER2_CHAIN_MODIFIED |
914 HAMMER2_CHAIN_SUBMODIFIED)) {
915 /* this will modify vchain as a side effect */
916 hammer2_chain_flush(&info.trans, &hmp->fchain);
918 hammer2_chain_unlock(&hmp->fchain);
925 * We can't safely flush the volume header until we have
926 * flushed any device buffers which have built up.
928 * XXX this isn't being incremental
930 vn_lock(hmp->devvp, LK_EXCLUSIVE | LK_RETRY);
931 error = VOP_FSYNC(hmp->devvp, MNT_WAIT, 0);
932 vn_unlock(hmp->devvp);
935 * The flush code sets CHAIN_VOLUMESYNC to indicate that the
936 * volume header needs synchronization via hmp->volsync.
938 * XXX synchronize the flag & data with only this flush XXX
940 if (error == 0 && (hmp->vchain.flags & HAMMER2_CHAIN_VOLUMESYNC)) {
944 * Synchronize the disk before flushing the volume
948 bp->b_bio1.bio_offset = 0;
951 bp->b_cmd = BUF_CMD_FLUSH;
952 bp->b_bio1.bio_done = biodone_sync;
953 bp->b_bio1.bio_flags |= BIO_SYNC;
954 vn_strategy(hmp->devvp, &bp->b_bio1);
955 biowait(&bp->b_bio1, "h2vol");
959 * Then we can safely flush the version of the volume header
960 * synchronized by the flush code.
962 i = hmp->volhdrno + 1;
963 if (i >= HAMMER2_NUM_VOLHDRS)
965 if (i * HAMMER2_ZONE_BYTES64 + HAMMER2_SEGSIZE >
966 hmp->volsync.volu_size) {
969 kprintf("sync volhdr %d %jd\n",
970 i, (intmax_t)hmp->volsync.volu_size);
971 bp = getblk(hmp->devvp, i * HAMMER2_ZONE_BYTES64,
972 HAMMER2_PBUFSIZE, 0, 0);
973 atomic_clear_int(&hmp->vchain.flags, HAMMER2_CHAIN_VOLUMESYNC);
974 bcopy(&hmp->volsync, bp->b_data, HAMMER2_PBUFSIZE);
978 hammer2_trans_done(&info.trans);
985 * NOTE: We don't test SUBMODIFIED or MOVED here because the fsync code
986 * won't flush on those flags. The syncer code above will do a
987 * general meta-data flush globally that will catch these flags.
990 hammer2_sync_scan1(struct mount *mp, struct vnode *vp, void *data)
995 if (vp->v_type == VNON || ip == NULL ||
996 ((ip->flags & HAMMER2_INODE_MODIFIED) == 0 &&
997 RB_EMPTY(&vp->v_rbdirty_tree))) {
1004 hammer2_sync_scan2(struct mount *mp, struct vnode *vp, void *data)
1006 struct hammer2_sync_info *info = data;
1007 hammer2_inode_t *ip;
1008 hammer2_chain_t *parent;
1012 if (vp->v_type == VNON || vp->v_type == VBAD ||
1013 ((ip->flags & HAMMER2_INODE_MODIFIED) == 0 &&
1014 RB_EMPTY(&vp->v_rbdirty_tree))) {
1019 * VOP_FSYNC will start a new transaction so replicate some code
1020 * here to do it inline (see hammer2_vop_fsync()).
1022 parent = hammer2_inode_lock_ex(ip);
1023 atomic_clear_int(&ip->flags, HAMMER2_INODE_MODIFIED);
1025 vfsync(ip->vp, MNT_NOWAIT, 1, NULL, NULL);
1026 hammer2_chain_flush(&info->trans, parent);
1027 hammer2_inode_unlock_ex(ip, parent);
1030 error = VOP_FSYNC(vp, MNT_NOWAIT, 0);
1033 info->error = error;
1039 hammer2_vfs_vptofh(struct vnode *vp, struct fid *fhp)
1046 hammer2_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
1047 struct fid *fhp, struct vnode **vpp)
1054 hammer2_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
1055 int *exflagsp, struct ucred **credanonp)
1061 * Support code for hammer2_mount(). Read, verify, and install the volume
1062 * header into the HMP
1064 * XXX read four volhdrs and use the one with the highest TID whos CRC
1069 * XXX For filesystems w/ less than 4 volhdrs, make sure to not write to
1070 * nonexistant locations.
1072 * XXX Record selected volhdr and ring updates to each of 4 volhdrs
1076 hammer2_install_volume_header(hammer2_mount_t *hmp)
1078 hammer2_volume_data_t *vd;
1080 hammer2_crc32_t crc0, crc, bcrc0, bcrc;
1092 * There are up to 4 copies of the volume header (syncs iterate
1093 * between them so there is no single master). We don't trust the
1094 * volu_size field so we don't know precisely how large the filesystem
1095 * is, so depend on the OS to return an error if we go beyond the
1096 * block device's EOF.
1098 for (i = 0; i < HAMMER2_NUM_VOLHDRS; i++) {
1099 error = bread(hmp->devvp, i * HAMMER2_ZONE_BYTES64,
1100 HAMMER2_VOLUME_BYTES, &bp);
1107 vd = (struct hammer2_volume_data *) bp->b_data;
1108 if ((vd->magic != HAMMER2_VOLUME_ID_HBO) &&
1109 (vd->magic != HAMMER2_VOLUME_ID_ABO)) {
1115 if (vd->magic == HAMMER2_VOLUME_ID_ABO) {
1116 /* XXX: Reversed-endianness filesystem */
1117 kprintf("hammer2: reverse-endian filesystem detected");
1123 crc = vd->icrc_sects[HAMMER2_VOL_ICRC_SECT0];
1124 crc0 = hammer2_icrc32(bp->b_data + HAMMER2_VOLUME_ICRC0_OFF,
1125 HAMMER2_VOLUME_ICRC0_SIZE);
1126 bcrc = vd->icrc_sects[HAMMER2_VOL_ICRC_SECT1];
1127 bcrc0 = hammer2_icrc32(bp->b_data + HAMMER2_VOLUME_ICRC1_OFF,
1128 HAMMER2_VOLUME_ICRC1_SIZE);
1129 if ((crc0 != crc) || (bcrc0 != bcrc)) {
1130 kprintf("hammer2 volume header crc "
1131 "mismatch copy #%d %08x/%08x\n",
1138 if (valid == 0 || hmp->voldata.mirror_tid < vd->mirror_tid) {
1147 hmp->volsync = hmp->voldata;
1149 if (error_reported || bootverbose || 1) { /* 1/DEBUG */
1150 kprintf("hammer2: using volume header #%d\n",
1155 kprintf("hammer2: no valid volume headers found!\n");
1161 * Reconnect using the passed file pointer. The caller must ref the
1165 hammer2_cluster_reconnect(hammer2_pfsmount_t *pmp, struct file *fp)
1167 hammer2_inode_data_t *ipdata;
1168 hammer2_chain_t *parent;
1172 * Closes old comm descriptor, kills threads, cleans up
1173 * states, then installs the new descriptor and creates
1176 kdmsg_iocom_reconnect(&pmp->iocom, fp, "hammer2");
1179 * Setup LNK_CONN fields for autoinitiated state machine
1181 parent = hammer2_inode_lock_ex(pmp->iroot);
1182 ipdata = &parent->data->ipdata;
1183 pmp->iocom.auto_lnk_conn.pfs_clid = ipdata->pfs_clid;
1184 pmp->iocom.auto_lnk_conn.pfs_fsid = ipdata->pfs_fsid;
1185 pmp->iocom.auto_lnk_conn.pfs_type = ipdata->pfs_type;
1186 pmp->iocom.auto_lnk_conn.proto_version = DMSG_SPAN_PROTO_1;
1187 pmp->iocom.auto_lnk_conn.peer_type = pmp->hmp->voldata.peer_type;
1190 * Filter adjustment. Clients do not need visibility into other
1191 * clients (otherwise millions of clients would present a serious
1192 * problem). The fs_label also serves to restrict the namespace.
1194 pmp->iocom.auto_lnk_conn.peer_mask = 1LLU << HAMMER2_PEER_HAMMER2;
1195 pmp->iocom.auto_lnk_conn.pfs_mask = (uint64_t)-1;
1196 switch (ipdata->pfs_type) {
1197 case DMSG_PFSTYPE_CLIENT:
1198 pmp->iocom.auto_lnk_conn.peer_mask &=
1199 ~(1LLU << DMSG_PFSTYPE_CLIENT);
1205 name_len = ipdata->name_len;
1206 if (name_len >= sizeof(pmp->iocom.auto_lnk_conn.fs_label))
1207 name_len = sizeof(pmp->iocom.auto_lnk_conn.fs_label) - 1;
1208 bcopy(ipdata->filename,
1209 pmp->iocom.auto_lnk_conn.fs_label,
1211 pmp->iocom.auto_lnk_conn.fs_label[name_len] = 0;
1214 * Setup LNK_SPAN fields for autoinitiated state machine
1216 pmp->iocom.auto_lnk_span.pfs_clid = ipdata->pfs_clid;
1217 pmp->iocom.auto_lnk_span.pfs_fsid = ipdata->pfs_fsid;
1218 pmp->iocom.auto_lnk_span.pfs_type = ipdata->pfs_type;
1219 pmp->iocom.auto_lnk_span.peer_type = pmp->hmp->voldata.peer_type;
1220 pmp->iocom.auto_lnk_span.proto_version = DMSG_SPAN_PROTO_1;
1221 name_len = ipdata->name_len;
1222 if (name_len >= sizeof(pmp->iocom.auto_lnk_span.fs_label))
1223 name_len = sizeof(pmp->iocom.auto_lnk_span.fs_label) - 1;
1224 bcopy(ipdata->filename,
1225 pmp->iocom.auto_lnk_span.fs_label,
1227 pmp->iocom.auto_lnk_span.fs_label[name_len] = 0;
1228 hammer2_inode_unlock_ex(pmp->iroot, parent);
1230 kdmsg_iocom_autoinitiate(&pmp->iocom, hammer2_autodmsg);
1234 hammer2_rcvdmsg(kdmsg_msg_t *msg)
1236 switch(msg->any.head.cmd & DMSGF_TRANSMASK) {
1237 case DMSG_DBG_SHELL:
1240 * Execute shell command (not supported atm)
1242 kdmsg_msg_reply(msg, DMSG_ERR_NOSUPP);
1244 case DMSG_DBG_SHELL | DMSGF_REPLY:
1248 if (msg->aux_data) {
1249 msg->aux_data[msg->aux_size - 1] = 0;
1250 kprintf("HAMMER2 DBG: %s\n", msg->aux_data);
1255 * Unsupported message received. We only need to
1256 * reply if it's a transaction in order to close our end.
1257 * Ignore any one-way messages are any further messages
1258 * associated with the transaction.
1260 * NOTE: This case also includes DMSG_LNK_ERROR messages
1261 * which might be one-way, replying to those would
1262 * cause an infinite ping-pong.
1264 if (msg->any.head.cmd & DMSGF_CREATE)
1265 kdmsg_msg_reply(msg, DMSG_ERR_NOSUPP);
1272 * This function is called after KDMSG has automatically handled processing
1273 * of a LNK layer message (typically CONN, SPAN, or CIRC).
1275 * We tag off the LNK_CONN to trigger our LNK_VOLCONF messages which
1276 * advertises all available hammer2 super-root volumes.
1279 hammer2_autodmsg(kdmsg_msg_t *msg)
1281 hammer2_pfsmount_t *pmp = msg->iocom->handle;
1282 hammer2_mount_t *hmp = pmp->hmp;
1286 * We only care about replies to our LNK_CONN auto-request. kdmsg
1287 * has already processed the reply, we use this calback as a shim
1288 * to know when we can advertise available super-root volumes.
1290 if ((msg->any.head.cmd & DMSGF_TRANSMASK) !=
1291 (DMSG_LNK_CONN | DMSGF_CREATE | DMSGF_REPLY) ||
1292 msg->state == NULL) {
1296 kprintf("LNK_CONN REPLY RECEIVED CMD %08x\n", msg->any.head.cmd);
1298 if (msg->any.head.cmd & DMSGF_CREATE) {
1299 kprintf("HAMMER2: VOLDATA DUMP\n");
1302 * Dump the configuration stored in the volume header
1304 hammer2_voldata_lock(hmp);
1305 for (copyid = 0; copyid < HAMMER2_COPYID_COUNT; ++copyid) {
1306 if (hmp->voldata.copyinfo[copyid].copyid == 0)
1308 hammer2_volconf_update(pmp, copyid);
1310 hammer2_voldata_unlock(hmp, 0);
1312 if ((msg->any.head.cmd & DMSGF_DELETE) &&
1313 msg->state && (msg->state->txcmd & DMSGF_DELETE) == 0) {
1314 kprintf("HAMMER2: CONN WAS TERMINATED\n");
1319 * Volume configuration updates are passed onto the userland service
1320 * daemon via the open LNK_CONN transaction.
1323 hammer2_volconf_update(hammer2_pfsmount_t *pmp, int index)
1325 hammer2_mount_t *hmp = pmp->hmp;
1328 /* XXX interlock against connection state termination */
1329 kprintf("volconf update %p\n", pmp->iocom.conn_state);
1330 if (pmp->iocom.conn_state) {
1331 kprintf("TRANSMIT VOLCONF VIA OPEN CONN TRANSACTION\n");
1332 msg = kdmsg_msg_alloc_state(pmp->iocom.conn_state,
1333 DMSG_LNK_VOLCONF, NULL, NULL);
1334 msg->any.lnk_volconf.copy = hmp->voldata.copyinfo[index];
1335 msg->any.lnk_volconf.mediaid = hmp->voldata.fsid;
1336 msg->any.lnk_volconf.index = index;
1337 kdmsg_msg_write(msg);
1342 hammer2_dump_chain(hammer2_chain_t *chain, int tab, int *countp)
1344 hammer2_chain_t *scan;
1348 kprintf("%*.*s...\n", tab, tab, "");
1353 kprintf("%*.*schain[%d] %p.%d [%08x][core=%p] (%s) dl=%p dt=%s refs=%d",
1355 chain->index, chain, chain->bref.type, chain->flags,
1357 ((chain->bref.type == HAMMER2_BREF_TYPE_INODE &&
1358 chain->data) ? (char *)chain->data->ipdata.filename : "?"),
1360 (chain->delete_tid == HAMMER2_MAX_TID ? "max" : "fls"),
1362 if (chain->core == NULL || RB_EMPTY(&chain->core->rbtree))
1366 RB_FOREACH(scan, hammer2_chain_tree, &chain->core->rbtree) {
1367 hammer2_dump_chain(scan, tab + 4, countp);
1369 if (chain->core && !RB_EMPTY(&chain->core->rbtree)) {
1370 if (chain->bref.type == HAMMER2_BREF_TYPE_INODE && chain->data)
1371 kprintf("%*.*s}(%s)\n", tab, tab, "",
1372 chain->data->ipdata.filename);
1374 kprintf("%*.*s}\n", tab, tab, "");