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_file_write;
70 long hammer2_iod_meta_write;
71 long hammer2_iod_indr_write;
72 long hammer2_iod_fmap_write;
73 long hammer2_iod_volu_write;
74 long hammer2_ioa_file_read;
75 long hammer2_ioa_meta_read;
76 long hammer2_ioa_indr_read;
77 long hammer2_ioa_fmap_write;
78 long hammer2_ioa_file_write;
79 long hammer2_ioa_meta_write;
80 long hammer2_ioa_indr_write;
81 long hammer2_ioa_volu_write;
83 SYSCTL_NODE(_vfs, OID_AUTO, hammer2, CTLFLAG_RW, 0, "HAMMER2 filesystem");
85 SYSCTL_INT(_vfs_hammer2, OID_AUTO, debug, CTLFLAG_RW,
86 &hammer2_debug, 0, "");
87 SYSCTL_INT(_vfs_hammer2, OID_AUTO, cluster_enable, CTLFLAG_RW,
88 &hammer2_cluster_enable, 0, "");
89 SYSCTL_INT(_vfs_hammer2, OID_AUTO, hardlink_enable, CTLFLAG_RW,
90 &hammer2_hardlink_enable, 0, "");
91 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_file_read, CTLFLAG_RW,
92 &hammer2_iod_file_read, 0, "");
93 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_meta_read, CTLFLAG_RW,
94 &hammer2_iod_meta_read, 0, "");
95 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_indr_read, CTLFLAG_RW,
96 &hammer2_iod_indr_read, 0, "");
97 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_file_write, CTLFLAG_RW,
98 &hammer2_iod_file_write, 0, "");
99 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_meta_write, CTLFLAG_RW,
100 &hammer2_iod_meta_write, 0, "");
101 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_indr_write, CTLFLAG_RW,
102 &hammer2_iod_indr_write, 0, "");
103 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_volu_write, CTLFLAG_RW,
104 &hammer2_iod_volu_write, 0, "");
105 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_file_read, CTLFLAG_RW,
106 &hammer2_ioa_file_read, 0, "");
107 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_meta_read, CTLFLAG_RW,
108 &hammer2_ioa_meta_read, 0, "");
109 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_indr_read, CTLFLAG_RW,
110 &hammer2_ioa_indr_read, 0, "");
111 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_file_write, CTLFLAG_RW,
112 &hammer2_ioa_file_write, 0, "");
113 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_meta_write, CTLFLAG_RW,
114 &hammer2_ioa_meta_write, 0, "");
115 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_indr_write, CTLFLAG_RW,
116 &hammer2_ioa_indr_write, 0, "");
117 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_volu_write, CTLFLAG_RW,
118 &hammer2_ioa_volu_write, 0, "");
120 static int hammer2_vfs_init(struct vfsconf *conf);
121 static int hammer2_vfs_mount(struct mount *mp, char *path, caddr_t data,
123 static int hammer2_remount(struct mount *, char *, struct vnode *,
125 static int hammer2_vfs_unmount(struct mount *mp, int mntflags);
126 static int hammer2_vfs_root(struct mount *mp, struct vnode **vpp);
127 static int hammer2_vfs_statfs(struct mount *mp, struct statfs *sbp,
129 static int hammer2_vfs_statvfs(struct mount *mp, struct statvfs *sbp,
131 static int hammer2_vfs_sync(struct mount *mp, int waitfor);
132 static int hammer2_vfs_vget(struct mount *mp, struct vnode *dvp,
133 ino_t ino, struct vnode **vpp);
134 static int hammer2_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
135 struct fid *fhp, struct vnode **vpp);
136 static int hammer2_vfs_vptofh(struct vnode *vp, struct fid *fhp);
137 static int hammer2_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
138 int *exflagsp, struct ucred **credanonp);
140 static int hammer2_install_volume_header(hammer2_mount_t *hmp);
141 static int hammer2_sync_scan1(struct mount *mp, struct vnode *vp, void *data);
142 static int hammer2_sync_scan2(struct mount *mp, struct vnode *vp, void *data);
144 static int hammer2_rcvdmsg(kdmsg_msg_t *msg);
145 static void hammer2_autodmsg(kdmsg_msg_t *msg);
148 * HAMMER2 vfs operations.
150 static struct vfsops hammer2_vfsops = {
151 .vfs_init = hammer2_vfs_init,
152 .vfs_sync = hammer2_vfs_sync,
153 .vfs_mount = hammer2_vfs_mount,
154 .vfs_unmount = hammer2_vfs_unmount,
155 .vfs_root = hammer2_vfs_root,
156 .vfs_statfs = hammer2_vfs_statfs,
157 .vfs_statvfs = hammer2_vfs_statvfs,
158 .vfs_vget = hammer2_vfs_vget,
159 .vfs_vptofh = hammer2_vfs_vptofh,
160 .vfs_fhtovp = hammer2_vfs_fhtovp,
161 .vfs_checkexp = hammer2_vfs_checkexp
164 MALLOC_DEFINE(M_HAMMER2, "HAMMER2-mount", "");
166 VFS_SET(hammer2_vfsops, hammer2, 0);
167 MODULE_VERSION(hammer2, 1);
171 hammer2_vfs_init(struct vfsconf *conf)
177 if (HAMMER2_BLOCKREF_BYTES != sizeof(struct hammer2_blockref))
179 if (HAMMER2_INODE_BYTES != sizeof(struct hammer2_inode_data))
181 if (HAMMER2_VOLUME_BYTES != sizeof(struct hammer2_volume_data))
185 kprintf("HAMMER2 structure size mismatch; cannot continue.\n");
187 lockinit(&hammer2_mntlk, "mntlk", 0, 0);
188 TAILQ_INIT(&hammer2_mntlist);
194 * Mount or remount HAMMER2 fileystem from physical media
197 * mp mount point structure
203 * mp mount point structure
204 * path path to mount point
205 * data pointer to argument structure in user space
206 * volume volume path (device@LABEL form)
207 * hflags user mount flags
208 * cred user credentials
215 hammer2_vfs_mount(struct mount *mp, char *path, caddr_t data,
218 struct hammer2_mount_info info;
219 hammer2_pfsmount_t *pmp;
220 hammer2_mount_t *hmp;
223 struct nlookupdata nd;
224 hammer2_chain_t *parent;
225 hammer2_chain_t *schain;
226 hammer2_chain_t *rchain;
228 char devstr[MNAMELEN];
243 kprintf("hammer2_mount\n");
249 bzero(&info, sizeof(info));
250 info.cluster_fd = -1;
254 * Non-root mount or updating a mount
256 error = copyin(data, &info, sizeof(info));
260 error = copyinstr(info.volume, devstr, MNAMELEN - 1, &done);
264 /* Extract device and label */
266 label = strchr(devstr, '@');
268 ((label + 1) - dev) > done) {
276 if (mp->mnt_flag & MNT_UPDATE) {
278 /* HAMMER2 implements NFS export via mountctl */
281 error = hammer2_remount(mp, path, devvp, cred);
289 * Lookup name and verify it refers to a block device.
291 error = nlookup_init(&nd, dev, UIO_SYSSPACE, NLC_FOLLOW);
293 error = nlookup(&nd);
295 error = cache_vref(&nd.nl_nch, nd.nl_cred, &devvp);
299 if (vn_isdisk(devvp, &error))
300 error = vfs_mountedon(devvp);
304 * Determine if the device has already been mounted. After this
305 * check hmp will be non-NULL if we are doing the second or more
306 * hammer2 mounts from the same device.
308 lockmgr(&hammer2_mntlk, LK_EXCLUSIVE);
309 TAILQ_FOREACH(hmp, &hammer2_mntlist, mntentry) {
310 if (hmp->devvp == devvp)
315 * Open the device if this isn't a secondary mount
321 if (error == 0 && vcount(devvp) > 0)
325 * Now open the device
328 ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
329 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
330 error = vinvalbuf(devvp, V_SAVE, 0, 0);
332 error = VOP_OPEN(devvp,
333 ronly ? FREAD : FREAD | FWRITE,
338 if (error && devvp) {
343 lockmgr(&hammer2_mntlk, LK_RELEASE);
349 * Block device opened successfully, finish initializing the
352 * From this point on we have to call hammer2_unmount() on failure.
354 pmp = kmalloc(sizeof(*pmp), M_HAMMER2, M_WAITOK | M_ZERO);
355 mp->mnt_data = (qaddr_t)pmp;
358 spin_init(&pmp->inum_spin);
359 RB_INIT(&pmp->inum_tree);
361 kmalloc_create(&pmp->mmsg, "HAMMER2-pfsmsg");
362 kdmsg_iocom_init(&pmp->iocom, pmp,
363 KDMSG_IOCOMF_AUTOCONN |
364 KDMSG_IOCOMF_AUTOSPAN |
365 KDMSG_IOCOMF_AUTOCIRC,
366 pmp->mmsg, hammer2_rcvdmsg);
369 hmp = kmalloc(sizeof(*hmp), M_HAMMER2, M_WAITOK | M_ZERO);
372 kmalloc_create(&hmp->minode, "HAMMER2-inodes");
373 kmalloc_create(&hmp->mchain, "HAMMER2-chains");
374 TAILQ_INSERT_TAIL(&hammer2_mntlist, hmp, mntentry);
376 ccms_domain_init(&pmp->ccms_dom);
379 lockmgr(&hammer2_mntlk, LK_RELEASE);
380 kprintf("hammer2_mount hmp=%p pmpcnt=%d\n", hmp, hmp->pmp_count);
382 mp->mnt_flag = MNT_LOCAL;
383 mp->mnt_kern_flag |= MNTK_ALL_MPSAFE; /* all entry pts are SMP */
389 lockinit(&hmp->alloclk, "h2alloc", 0, 0);
390 lockinit(&hmp->voldatalk, "voldata", 0, LK_CANRECURSE);
391 TAILQ_INIT(&hmp->transq);
394 * vchain setup. vchain.data is special cased to NULL.
395 * vchain.refs is initialized and will never drop to 0.
397 hmp->vchain.hmp = hmp;
398 hmp->vchain.refs = 1;
399 hmp->vchain.data = (void *)&hmp->voldata;
400 hmp->vchain.bref.type = HAMMER2_BREF_TYPE_VOLUME;
401 hmp->vchain.bref.data_off = 0 | HAMMER2_PBUFRADIX;
402 hmp->vchain.delete_tid = HAMMER2_MAX_TID;
403 hammer2_chain_core_alloc(&hmp->vchain, NULL);
404 /* hmp->vchain.u.xxx is left NULL */
407 * Install the volume header
409 error = hammer2_install_volume_header(hmp);
411 hammer2_vfs_unmount(mp, MNT_FORCE);
417 * required mount structure initializations
419 mp->mnt_stat.f_iosize = HAMMER2_PBUFSIZE;
420 mp->mnt_stat.f_bsize = HAMMER2_PBUFSIZE;
422 mp->mnt_vstat.f_frsize = HAMMER2_PBUFSIZE;
423 mp->mnt_vstat.f_bsize = HAMMER2_PBUFSIZE;
428 mp->mnt_iosize_max = MAXPHYS;
431 * First locate the super-root inode, which is key 0 relative to the
432 * volume header's blockset.
434 * Then locate the root inode by scanning the directory keyspace
435 * represented by the label.
438 parent = hammer2_chain_lookup_init(&hmp->vchain, 0);
439 schain = hammer2_chain_lookup(&parent,
440 HAMMER2_SROOT_KEY, HAMMER2_SROOT_KEY, 0);
441 hammer2_chain_lookup_done(parent);
442 if (schain == NULL) {
443 kprintf("hammer2_mount: invalid super-root\n");
444 hammer2_vfs_unmount(mp, MNT_FORCE);
447 hammer2_chain_ref(schain); /* for hmp->schain */
448 hmp->schain = schain; /* left locked for inode_get */
449 hmp->sroot = hammer2_inode_get(hmp, NULL, NULL, schain);
450 hammer2_inode_ref(hmp->sroot); /* for hmp->sroot */
451 hammer2_inode_unlock_ex(hmp->sroot, schain);
453 schain = hmp->schain;
457 * schain only has 1 ref now for its hmp->schain assignment.
458 * Setup for lookup (which will lock it).
460 parent = hammer2_chain_lookup_init(schain, 0);
461 lhc = hammer2_dirhash(label, strlen(label));
462 rchain = hammer2_chain_lookup(&parent,
463 lhc, lhc + HAMMER2_DIRHASH_LOMASK,
466 if (rchain->bref.type == HAMMER2_BREF_TYPE_INODE &&
467 strcmp(label, rchain->data->ipdata.filename) == 0) {
470 rchain = hammer2_chain_next(&parent, rchain,
471 lhc, lhc + HAMMER2_DIRHASH_LOMASK,
474 hammer2_chain_lookup_done(parent);
475 if (rchain == NULL) {
476 kprintf("hammer2_mount: PFS label not found\n");
477 hammer2_vfs_unmount(mp, MNT_FORCE);
480 if (rchain->flags & HAMMER2_CHAIN_MOUNTED) {
481 hammer2_chain_unlock(rchain);
482 kprintf("hammer2_mount: PFS label already mounted!\n");
483 hammer2_vfs_unmount(mp, MNT_FORCE);
486 if (rchain->flags & HAMMER2_CHAIN_RECYCLE) {
487 kprintf("hammer2_mount: PFS label currently recycling\n");
488 hammer2_vfs_unmount(mp, MNT_FORCE);
492 atomic_set_int(&rchain->flags, HAMMER2_CHAIN_MOUNTED);
495 * NOTE: *_get() integrates chain's lock into the inode lock.
497 hammer2_chain_ref(rchain); /* for pmp->rchain */
498 pmp->rchain = rchain; /* left held & unlocked */
499 pmp->iroot = hammer2_inode_get(hmp, pmp, NULL, rchain);
500 hammer2_inode_ref(pmp->iroot); /* ref for pmp->iroot */
501 hammer2_inode_unlock_ex(pmp->iroot, rchain);
503 kprintf("iroot %p\n", pmp->iroot);
506 * Ref the cluster management messaging descriptor. The mount
507 * program deals with the other end of the communications pipe.
509 fp = holdfp(curproc->p_fd, info.cluster_fd, -1);
511 kprintf("hammer2_mount: bad cluster_fd!\n");
512 hammer2_vfs_unmount(mp, MNT_FORCE);
515 hammer2_cluster_reconnect(pmp, fp);
521 vfs_add_vnodeops(mp, &hammer2_vnode_vops, &mp->mnt_vn_norm_ops);
522 vfs_add_vnodeops(mp, &hammer2_spec_vops, &mp->mnt_vn_spec_ops);
523 vfs_add_vnodeops(mp, &hammer2_fifo_vops, &mp->mnt_vn_fifo_ops);
525 copyinstr(info.volume, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, &size);
526 bzero(mp->mnt_stat.f_mntfromname + size, MNAMELEN - size);
527 bzero(mp->mnt_stat.f_mntonname, sizeof(mp->mnt_stat.f_mntonname));
528 copyinstr(path, mp->mnt_stat.f_mntonname,
529 sizeof(mp->mnt_stat.f_mntonname) - 1,
533 * Initial statfs to prime mnt_stat.
535 hammer2_vfs_statfs(mp, &mp->mnt_stat, cred);
542 hammer2_remount(struct mount *mp, char *path, struct vnode *devvp,
550 hammer2_vfs_unmount(struct mount *mp, int mntflags)
552 hammer2_pfsmount_t *pmp;
553 hammer2_mount_t *hmp;
554 hammer2_chain_t *parent;
557 int ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
565 if (mntflags & MNT_FORCE)
568 hammer2_mount_exlock(hmp);
571 * If mount initialization proceeded far enough we must flush
575 error = vflush(mp, 0, flags);
578 hammer2_mount_unlock(hmp);
582 lockmgr(&hammer2_mntlk, LK_EXCLUSIVE);
584 kprintf("hammer2_unmount hmp=%p pmpcnt=%d\n", hmp, hmp->pmp_count);
587 * Flush any left over chains. The voldata lock is only used
588 * to synchronize against HAMMER2_CHAIN_MODIFIED_AUX.
590 hammer2_voldata_lock(hmp);
591 if (hmp->vchain.flags & (HAMMER2_CHAIN_MODIFIED |
592 HAMMER2_CHAIN_SUBMODIFIED)) {
593 hammer2_voldata_unlock(hmp, 0);
594 hammer2_vfs_sync(mp, MNT_WAIT);
596 hammer2_voldata_unlock(hmp, 0);
598 if (hmp->pmp_count == 0) {
599 if (hmp->vchain.flags & (HAMMER2_CHAIN_MODIFIED |
600 HAMMER2_CHAIN_SUBMODIFIED)) {
601 kprintf("hammer2_unmount: chains left over after "
603 if (hammer2_debug & 0x0010)
604 Debugger("entered debugger");
609 * Cleanup the root and super-root chain elements (which should be
613 parent = hammer2_inode_lock_ex(pmp->iroot);
614 hammer2_inode_put(pmp->iroot, parent);
615 /* lock destroyed by the put */
616 #if REPORT_REFS_ERRORS
617 if (pmp->iroot->refs != 1)
618 kprintf("PMP->IROOT %p REFS WRONG %d\n",
619 pmp->iroot, pmp->iroot->refs);
621 KKASSERT(pmp->iroot->refs == 1);
623 hammer2_inode_drop(pmp->iroot); /* ref for pmp->iroot */
627 atomic_clear_int(&pmp->rchain->flags, HAMMER2_CHAIN_MOUNTED);
628 #if REPORT_REFS_ERRORS
629 if (pmp->rchain->refs != 1)
630 kprintf("PMP->RCHAIN %p REFS WRONG %d\n",
631 pmp->rchain, pmp->rchain->refs);
633 KKASSERT(pmp->rchain->refs == 1);
635 hammer2_chain_drop(pmp->rchain);
638 ccms_domain_uninit(&pmp->ccms_dom);
641 * Kill cluster controller
643 kdmsg_iocom_uninit(&pmp->iocom);
646 * If no PFS's left drop the master hammer2_mount for the device.
648 if (hmp->pmp_count == 0) {
650 hammer2_inode_drop(hmp->sroot);
654 #if REPORT_REFS_ERRORS
655 if (hmp->schain->refs != 1)
656 kprintf("HMP->SCHAIN %p REFS WRONG %d\n",
657 hmp->schain, hmp->schain->refs);
659 KKASSERT(hmp->schain->refs == 1);
661 hammer2_chain_drop(hmp->schain);
666 * Finish up with the device vnode
668 if ((devvp = hmp->devvp) != NULL) {
669 vinvalbuf(devvp, (ronly ? 0 : V_SAVE), 0, 0);
671 VOP_CLOSE(devvp, (ronly ? FREAD : FREAD|FWRITE));
677 * Final drop of embedded volume root chain to clean up
678 * vchain.core (vchain structure is not flagged ALLOCATED
679 * so it is cleaned out and then left).
682 hammer2_dump_chain(&hmp->vchain, 0, &dumpcnt);
683 hammer2_mount_unlock(hmp);
684 hammer2_chain_drop(&hmp->vchain);
686 hammer2_mount_unlock(hmp);
693 kmalloc_destroy(&pmp->mmsg);
695 kfree(pmp, M_HAMMER2);
696 if (hmp->pmp_count == 0) {
697 TAILQ_REMOVE(&hammer2_mntlist, hmp, mntentry);
698 kmalloc_destroy(&hmp->minode);
699 kmalloc_destroy(&hmp->mchain);
700 kfree(hmp, M_HAMMER2);
702 lockmgr(&hammer2_mntlk, LK_RELEASE);
708 hammer2_vfs_vget(struct mount *mp, struct vnode *dvp,
709 ino_t ino, struct vnode **vpp)
711 kprintf("hammer2_vget\n");
717 hammer2_vfs_root(struct mount *mp, struct vnode **vpp)
719 hammer2_pfsmount_t *pmp;
720 hammer2_mount_t *hmp;
721 hammer2_chain_t *parent;
727 hammer2_mount_exlock(hmp);
728 if (pmp->iroot == NULL) {
732 parent = hammer2_inode_lock_sh(pmp->iroot);
733 vp = hammer2_igetv(pmp->iroot, &error);
734 hammer2_inode_unlock_sh(pmp->iroot, parent);
737 kprintf("vnodefail\n");
739 hammer2_mount_unlock(hmp);
747 * XXX incorporate ipdata->inode_quota and data_quota
751 hammer2_vfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
753 hammer2_pfsmount_t *pmp;
754 hammer2_mount_t *hmp;
759 mp->mnt_stat.f_files = pmp->inode_count;
760 mp->mnt_stat.f_ffree = 0;
761 mp->mnt_stat.f_blocks = hmp->voldata.allocator_size / HAMMER2_PBUFSIZE;
762 mp->mnt_stat.f_bfree = (hmp->voldata.allocator_size -
763 hmp->voldata.allocator_beg) / HAMMER2_PBUFSIZE;
764 mp->mnt_stat.f_bavail = mp->mnt_stat.f_bfree;
772 hammer2_vfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred)
774 hammer2_pfsmount_t *pmp;
775 hammer2_mount_t *hmp;
780 mp->mnt_vstat.f_bsize = HAMMER2_PBUFSIZE;
781 mp->mnt_vstat.f_files = pmp->inode_count;
782 mp->mnt_vstat.f_ffree = 0;
783 mp->mnt_vstat.f_blocks = hmp->voldata.allocator_size / HAMMER2_PBUFSIZE;
784 mp->mnt_vstat.f_bfree = (hmp->voldata.allocator_size -
785 hmp->voldata.allocator_beg) / HAMMER2_PBUFSIZE;
786 mp->mnt_vstat.f_bavail = mp->mnt_vstat.f_bfree;
788 *sbp = mp->mnt_vstat;
793 * Sync the entire filesystem; this is called from the filesystem syncer
794 * process periodically and whenever a user calls sync(1) on the hammer
797 * Currently is actually called from the syncer! \o/
799 * This task will have to snapshot the state of the dirty inode chain.
800 * From that, it will have to make sure all of the inodes on the dirty
801 * chain have IO initiated. We make sure that io is initiated for the root
804 * If waitfor is set, we wait for media to acknowledge the new rootblock.
806 * THINKS: side A vs side B, to have sync not stall all I/O?
810 hammer2_vfs_sync(struct mount *mp, int waitfor)
812 struct hammer2_sync_info info;
813 hammer2_mount_t *hmp;
823 if ((waitfor & MNT_LAZY) == 0) {
825 hammer2_dump_chain(&hmp->vchain, 0, &dumpcnt);
830 if (waitfor & MNT_LAZY)
831 flags |= VMSC_ONEPASS;
833 hammer2_trans_init(hmp, &info.trans, HAMMER2_TRANS_ISFLUSH);
836 info.waitfor = MNT_NOWAIT;
837 vmntvnodescan(mp, flags | VMSC_NOWAIT,
839 hammer2_sync_scan2, &info);
840 if (info.error == 0 && (waitfor & MNT_WAIT)) {
841 info.waitfor = waitfor;
842 vmntvnodescan(mp, flags,
844 hammer2_sync_scan2, &info);
848 if (waitfor == MNT_WAIT) {
855 * Rollup flush. The fsyncs above basically just flushed
856 * data blocks. The flush below gets all the meta-data.
858 hammer2_chain_lock(&hmp->vchain, HAMMER2_RESOLVE_ALWAYS);
859 if (hmp->vchain.flags & (HAMMER2_CHAIN_MODIFIED |
860 HAMMER2_CHAIN_SUBMODIFIED)) {
861 hammer2_chain_flush(&info.trans, &hmp->vchain);
863 hammer2_chain_unlock(&hmp->vchain);
868 * We can't safely flush the volume header until we have
869 * flushed any device buffers which have built up.
871 * XXX this isn't being incremental
873 vn_lock(hmp->devvp, LK_EXCLUSIVE | LK_RETRY);
874 error = VOP_FSYNC(hmp->devvp, MNT_WAIT, 0);
875 vn_unlock(hmp->devvp);
878 * The flush code sets CHAIN_VOLUMESYNC to indicate that the
879 * volume header needs synchronization via hmp->volsync.
881 * XXX synchronize the flag & data with only this flush XXX
883 if (error == 0 && (hmp->vchain.flags & HAMMER2_CHAIN_VOLUMESYNC)) {
887 * Synchronize the disk before flushing the volume
891 bp->b_bio1.bio_offset = 0;
894 bp->b_cmd = BUF_CMD_FLUSH;
895 bp->b_bio1.bio_done = biodone_sync;
896 bp->b_bio1.bio_flags |= BIO_SYNC;
897 vn_strategy(hmp->devvp, &bp->b_bio1);
898 biowait(&bp->b_bio1, "h2vol");
902 * Then we can safely flush the version of the volume header
903 * synchronized by the flush code.
905 i = hmp->volhdrno + 1;
906 if (i >= HAMMER2_NUM_VOLHDRS)
908 if (i * HAMMER2_ZONE_BYTES64 + HAMMER2_SEGSIZE >
909 hmp->volsync.volu_size) {
912 kprintf("sync volhdr %d %jd\n",
913 i, (intmax_t)hmp->volsync.volu_size);
914 bp = getblk(hmp->devvp, i * HAMMER2_ZONE_BYTES64,
915 HAMMER2_PBUFSIZE, 0, 0);
916 atomic_clear_int(&hmp->vchain.flags, HAMMER2_CHAIN_VOLUMESYNC);
917 bcopy(&hmp->volsync, bp->b_data, HAMMER2_PBUFSIZE);
921 hammer2_trans_done(&info.trans);
928 * NOTE: We don't test SUBMODIFIED or MOVED here because the fsync code
929 * won't flush on those flags. The syncer code above will do a
930 * general meta-data flush globally that will catch these flags.
933 hammer2_sync_scan1(struct mount *mp, struct vnode *vp, void *data)
938 if (vp->v_type == VNON || ip == NULL ||
939 ((ip->flags & HAMMER2_INODE_MODIFIED) == 0 &&
940 RB_EMPTY(&vp->v_rbdirty_tree))) {
947 hammer2_sync_scan2(struct mount *mp, struct vnode *vp, void *data)
949 struct hammer2_sync_info *info = data;
951 hammer2_chain_t *parent;
955 if (vp->v_type == VNON || vp->v_type == VBAD ||
956 ((ip->flags & HAMMER2_INODE_MODIFIED) == 0 &&
957 RB_EMPTY(&vp->v_rbdirty_tree))) {
962 * VOP_FSYNC will start a new transaction so replicate some code
963 * here to do it inline (see hammer2_vop_fsync()).
965 parent = hammer2_inode_lock_ex(ip);
966 atomic_clear_int(&ip->flags, HAMMER2_INODE_MODIFIED);
968 vfsync(ip->vp, MNT_NOWAIT, 1, NULL, NULL);
969 hammer2_chain_flush(&info->trans, parent);
970 hammer2_inode_unlock_ex(ip, parent);
973 error = VOP_FSYNC(vp, MNT_NOWAIT, 0);
982 hammer2_vfs_vptofh(struct vnode *vp, struct fid *fhp)
989 hammer2_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
990 struct fid *fhp, struct vnode **vpp)
997 hammer2_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
998 int *exflagsp, struct ucred **credanonp)
1004 * Support code for hammer2_mount(). Read, verify, and install the volume
1005 * header into the HMP
1007 * XXX read four volhdrs and use the one with the highest TID whos CRC
1012 * XXX For filesystems w/ less than 4 volhdrs, make sure to not write to
1013 * nonexistant locations.
1015 * XXX Record selected volhdr and ring updates to each of 4 volhdrs
1019 hammer2_install_volume_header(hammer2_mount_t *hmp)
1021 hammer2_volume_data_t *vd;
1023 hammer2_crc32_t crc0, crc, bcrc0, bcrc;
1035 * There are up to 4 copies of the volume header (syncs iterate
1036 * between them so there is no single master). We don't trust the
1037 * volu_size field so we don't know precisely how large the filesystem
1038 * is, so depend on the OS to return an error if we go beyond the
1039 * block device's EOF.
1041 for (i = 0; i < HAMMER2_NUM_VOLHDRS; i++) {
1042 error = bread(hmp->devvp, i * HAMMER2_ZONE_BYTES64,
1043 HAMMER2_VOLUME_BYTES, &bp);
1050 vd = (struct hammer2_volume_data *) bp->b_data;
1051 if ((vd->magic != HAMMER2_VOLUME_ID_HBO) &&
1052 (vd->magic != HAMMER2_VOLUME_ID_ABO)) {
1058 if (vd->magic == HAMMER2_VOLUME_ID_ABO) {
1059 /* XXX: Reversed-endianness filesystem */
1060 kprintf("hammer2: reverse-endian filesystem detected");
1066 crc = vd->icrc_sects[HAMMER2_VOL_ICRC_SECT0];
1067 crc0 = hammer2_icrc32(bp->b_data + HAMMER2_VOLUME_ICRC0_OFF,
1068 HAMMER2_VOLUME_ICRC0_SIZE);
1069 bcrc = vd->icrc_sects[HAMMER2_VOL_ICRC_SECT1];
1070 bcrc0 = hammer2_icrc32(bp->b_data + HAMMER2_VOLUME_ICRC1_OFF,
1071 HAMMER2_VOLUME_ICRC1_SIZE);
1072 if ((crc0 != crc) || (bcrc0 != bcrc)) {
1073 kprintf("hammer2 volume header crc "
1074 "mismatch copy #%d %08x/%08x\n",
1081 if (valid == 0 || hmp->voldata.mirror_tid < vd->mirror_tid) {
1090 hmp->volsync = hmp->voldata;
1092 if (error_reported || bootverbose || 1) { /* 1/DEBUG */
1093 kprintf("hammer2: using volume header #%d\n",
1098 kprintf("hammer2: no valid volume headers found!\n");
1104 * Reconnect using the passed file pointer. The caller must ref the
1108 hammer2_cluster_reconnect(hammer2_pfsmount_t *pmp, struct file *fp)
1110 hammer2_inode_data_t *ipdata;
1111 hammer2_chain_t *parent;
1115 * Closes old comm descriptor, kills threads, cleans up
1116 * states, then installs the new descriptor and creates
1119 kdmsg_iocom_reconnect(&pmp->iocom, fp, "hammer2");
1122 * Setup LNK_CONN fields for autoinitiated state machine
1124 parent = hammer2_inode_lock_ex(pmp->iroot);
1125 ipdata = &parent->data->ipdata;
1126 pmp->iocom.auto_lnk_conn.pfs_clid = ipdata->pfs_clid;
1127 pmp->iocom.auto_lnk_conn.pfs_fsid = ipdata->pfs_fsid;
1128 pmp->iocom.auto_lnk_conn.pfs_type = ipdata->pfs_type;
1129 pmp->iocom.auto_lnk_conn.proto_version = DMSG_SPAN_PROTO_1;
1130 pmp->iocom.auto_lnk_conn.peer_type = pmp->hmp->voldata.peer_type;
1133 * Filter adjustment. Clients do not need visibility into other
1134 * clients (otherwise millions of clients would present a serious
1135 * problem). The fs_label also serves to restrict the namespace.
1137 pmp->iocom.auto_lnk_conn.peer_mask = 1LLU << HAMMER2_PEER_HAMMER2;
1138 pmp->iocom.auto_lnk_conn.pfs_mask = (uint64_t)-1;
1139 switch (ipdata->pfs_type) {
1140 case DMSG_PFSTYPE_CLIENT:
1141 pmp->iocom.auto_lnk_conn.peer_mask &=
1142 ~(1LLU << DMSG_PFSTYPE_CLIENT);
1148 name_len = ipdata->name_len;
1149 if (name_len >= sizeof(pmp->iocom.auto_lnk_conn.fs_label))
1150 name_len = sizeof(pmp->iocom.auto_lnk_conn.fs_label) - 1;
1151 bcopy(ipdata->filename,
1152 pmp->iocom.auto_lnk_conn.fs_label,
1154 pmp->iocom.auto_lnk_conn.fs_label[name_len] = 0;
1157 * Setup LNK_SPAN fields for autoinitiated state machine
1159 pmp->iocom.auto_lnk_span.pfs_clid = ipdata->pfs_clid;
1160 pmp->iocom.auto_lnk_span.pfs_fsid = ipdata->pfs_fsid;
1161 pmp->iocom.auto_lnk_span.pfs_type = ipdata->pfs_type;
1162 pmp->iocom.auto_lnk_span.peer_type = pmp->hmp->voldata.peer_type;
1163 pmp->iocom.auto_lnk_span.proto_version = DMSG_SPAN_PROTO_1;
1164 name_len = ipdata->name_len;
1165 if (name_len >= sizeof(pmp->iocom.auto_lnk_span.fs_label))
1166 name_len = sizeof(pmp->iocom.auto_lnk_span.fs_label) - 1;
1167 bcopy(ipdata->filename,
1168 pmp->iocom.auto_lnk_span.fs_label,
1170 pmp->iocom.auto_lnk_span.fs_label[name_len] = 0;
1171 hammer2_inode_unlock_ex(pmp->iroot, parent);
1173 kdmsg_iocom_autoinitiate(&pmp->iocom, hammer2_autodmsg);
1177 hammer2_rcvdmsg(kdmsg_msg_t *msg)
1179 switch(msg->any.head.cmd & DMSGF_TRANSMASK) {
1180 case DMSG_DBG_SHELL:
1183 * Execute shell command (not supported atm)
1185 kdmsg_msg_reply(msg, DMSG_ERR_NOSUPP);
1187 case DMSG_DBG_SHELL | DMSGF_REPLY:
1191 if (msg->aux_data) {
1192 msg->aux_data[msg->aux_size - 1] = 0;
1193 kprintf("HAMMER2 DBG: %s\n", msg->aux_data);
1198 * Unsupported message received. We only need to
1199 * reply if it's a transaction in order to close our end.
1200 * Ignore any one-way messages are any further messages
1201 * associated with the transaction.
1203 * NOTE: This case also includes DMSG_LNK_ERROR messages
1204 * which might be one-way, replying to those would
1205 * cause an infinite ping-pong.
1207 if (msg->any.head.cmd & DMSGF_CREATE)
1208 kdmsg_msg_reply(msg, DMSG_ERR_NOSUPP);
1215 * This function is called after KDMSG has automatically handled processing
1216 * of a LNK layer message (typically CONN, SPAN, or CIRC).
1218 * We tag off the LNK_CONN to trigger our LNK_VOLCONF messages which
1219 * advertises all available hammer2 super-root volumes.
1222 hammer2_autodmsg(kdmsg_msg_t *msg)
1224 hammer2_pfsmount_t *pmp = msg->iocom->handle;
1225 hammer2_mount_t *hmp = pmp->hmp;
1229 * We only care about replies to our LNK_CONN auto-request. kdmsg
1230 * has already processed the reply, we use this calback as a shim
1231 * to know when we can advertise available super-root volumes.
1233 if ((msg->any.head.cmd & DMSGF_TRANSMASK) !=
1234 (DMSG_LNK_CONN | DMSGF_CREATE | DMSGF_REPLY) ||
1235 msg->state == NULL) {
1239 kprintf("LNK_CONN REPLY RECEIVED CMD %08x\n", msg->any.head.cmd);
1241 if (msg->any.head.cmd & DMSGF_CREATE) {
1242 kprintf("HAMMER2: VOLDATA DUMP\n");
1245 * Dump the configuration stored in the volume header
1247 hammer2_voldata_lock(hmp);
1248 for (copyid = 0; copyid < HAMMER2_COPYID_COUNT; ++copyid) {
1249 if (hmp->voldata.copyinfo[copyid].copyid == 0)
1251 hammer2_volconf_update(pmp, copyid);
1253 hammer2_voldata_unlock(hmp, 0);
1255 if ((msg->any.head.cmd & DMSGF_DELETE) &&
1256 msg->state && (msg->state->txcmd & DMSGF_DELETE) == 0) {
1257 kprintf("HAMMER2: CONN WAS TERMINATED\n");
1262 * Volume configuration updates are passed onto the userland service
1263 * daemon via the open LNK_CONN transaction.
1266 hammer2_volconf_update(hammer2_pfsmount_t *pmp, int index)
1268 hammer2_mount_t *hmp = pmp->hmp;
1271 /* XXX interlock against connection state termination */
1272 kprintf("volconf update %p\n", pmp->iocom.conn_state);
1273 if (pmp->iocom.conn_state) {
1274 kprintf("TRANSMIT VOLCONF VIA OPEN CONN TRANSACTION\n");
1275 msg = kdmsg_msg_alloc_state(pmp->iocom.conn_state,
1276 DMSG_LNK_VOLCONF, NULL, NULL);
1277 msg->any.lnk_volconf.copy = hmp->voldata.copyinfo[index];
1278 msg->any.lnk_volconf.mediaid = hmp->voldata.fsid;
1279 msg->any.lnk_volconf.index = index;
1280 kdmsg_msg_write(msg);
1285 hammer2_dump_chain(hammer2_chain_t *chain, int tab, int *countp)
1287 hammer2_chain_t *scan;
1291 kprintf("%*.*s...\n", tab, tab, "");
1296 kprintf("%*.*schain[%d] %p.%d [%08x][core=%p] (%s) dl=%p refs=%d",
1298 chain->index, chain, chain->bref.type, chain->flags,
1300 ((chain->bref.type == HAMMER2_BREF_TYPE_INODE &&
1301 chain->data) ? (char *)chain->data->ipdata.filename : "?"),
1302 chain->next_parent, chain->refs);
1303 if (chain->core == NULL || RB_EMPTY(&chain->core->rbtree))
1307 RB_FOREACH(scan, hammer2_chain_tree, &chain->core->rbtree) {
1308 hammer2_dump_chain(scan, tab + 4, countp);
1310 if (chain->core && !RB_EMPTY(&chain->core->rbtree)) {
1311 if (chain->bref.type == HAMMER2_BREF_TYPE_INODE && chain->data)
1312 kprintf("%*.*s}(%s)\n", tab, tab, "",
1313 chain->data->ipdata.filename);
1315 kprintf("%*.*s}\n", tab, tab, "");