xref: /dragonfly/sys/vfs/hammer2/hammer2_vfsops.c (revision c311ab13)
1 /*-
2  * Copyright (c) 2011-2013 The DragonFly Project.  All rights reserved.
3  *
4  * This code is derived from software contributed to The DragonFly Project
5  * by Matthew Dillon <dillon@backplane.com>
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  *
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
16  *    distribution.
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.
20  *
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
32  * SUCH DAMAGE.
33  */
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>
41 #include <sys/buf.h>
42 #include <sys/uuid.h>
43 #include <sys/vfsops.h>
44 #include <sys/sysctl.h>
45 #include <sys/socket.h>
46 
47 #include "hammer2.h"
48 #include "hammer2_disk.h"
49 #include "hammer2_mount.h"
50 
51 #define REPORT_REFS_ERRORS 1	/* XXX remove me */
52 
53 struct hammer2_sync_info {
54 	hammer2_trans_t trans;
55 	int error;
56 	int waitfor;
57 };
58 
59 TAILQ_HEAD(hammer2_mntlist, hammer2_mount);
60 static struct hammer2_mntlist hammer2_mntlist;
61 static struct lock hammer2_mntlk;
62 
63 int hammer2_debug;
64 int hammer2_cluster_enable = 0;	/* XXX temporary until layout ironed out */
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;
82 
83 SYSCTL_NODE(_vfs, OID_AUTO, hammer2, CTLFLAG_RW, 0, "HAMMER2 filesystem");
84 
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, "");
119 
120 static int hammer2_vfs_init(struct vfsconf *conf);
121 static int hammer2_vfs_mount(struct mount *mp, char *path, caddr_t data,
122 				struct ucred *cred);
123 static int hammer2_remount(struct mount *, char *, struct vnode *,
124 				struct ucred *);
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,
128 				struct ucred *cred);
129 static int hammer2_vfs_statvfs(struct mount *mp, struct statvfs *sbp,
130 				struct ucred *cred);
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);
139 
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);
143 
144 static int hammer2_rcvdmsg(kdmsg_msg_t *msg);
145 static void hammer2_autodmsg(kdmsg_msg_t *msg);
146 
147 /*
148  * HAMMER2 vfs operations.
149  */
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
162 };
163 
164 MALLOC_DEFINE(M_HAMMER2, "HAMMER2-mount", "");
165 
166 VFS_SET(hammer2_vfsops, hammer2, 0);
167 MODULE_VERSION(hammer2, 1);
168 
169 static
170 int
171 hammer2_vfs_init(struct vfsconf *conf)
172 {
173 	int error;
174 
175 	error = 0;
176 
177 	if (HAMMER2_BLOCKREF_BYTES != sizeof(struct hammer2_blockref))
178 		error = EINVAL;
179 	if (HAMMER2_INODE_BYTES != sizeof(struct hammer2_inode_data))
180 		error = EINVAL;
181 	if (HAMMER2_VOLUME_BYTES != sizeof(struct hammer2_volume_data))
182 		error = EINVAL;
183 
184 	if (error)
185 		kprintf("HAMMER2 structure size mismatch; cannot continue.\n");
186 
187 	lockinit(&hammer2_mntlk, "mntlk", 0, 0);
188 	TAILQ_INIT(&hammer2_mntlist);
189 
190 	return (error);
191 }
192 
193 /*
194  * Mount or remount HAMMER2 fileystem from physical media
195  *
196  *	mountroot
197  *		mp		mount point structure
198  *		path		NULL
199  *		data		<unused>
200  *		cred		<unused>
201  *
202  *	mount
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
209  *
210  * RETURNS:	0	Success
211  *		!0	error number
212  */
213 static
214 int
215 hammer2_vfs_mount(struct mount *mp, char *path, caddr_t data,
216 		  struct ucred *cred)
217 {
218 	struct hammer2_mount_info info;
219 	hammer2_pfsmount_t *pmp;
220 	hammer2_mount_t *hmp;
221 	hammer2_key_t lhc;
222 	struct vnode *devvp;
223 	struct nlookupdata nd;
224 	hammer2_chain_t *parent;
225 	hammer2_chain_t *schain;
226 	hammer2_chain_t *rchain;
227 	struct file *fp;
228 	char devstr[MNAMELEN];
229 	size_t size;
230 	size_t done;
231 	char *dev;
232 	char *label;
233 	int ronly = 1;
234 	int create_hmp;
235 	int error;
236 
237 	hmp = NULL;
238 	pmp = NULL;
239 	dev = NULL;
240 	label = NULL;
241 	devvp = NULL;
242 
243 	kprintf("hammer2_mount\n");
244 
245 	if (path == NULL) {
246 		/*
247 		 * Root mount
248 		 */
249 		bzero(&info, sizeof(info));
250 		info.cluster_fd = -1;
251 		return (EOPNOTSUPP);
252 	} else {
253 		/*
254 		 * Non-root mount or updating a mount
255 		 */
256 		error = copyin(data, &info, sizeof(info));
257 		if (error)
258 			return (error);
259 
260 		error = copyinstr(info.volume, devstr, MNAMELEN - 1, &done);
261 		if (error)
262 			return (error);
263 
264 		/* Extract device and label */
265 		dev = devstr;
266 		label = strchr(devstr, '@');
267 		if (label == NULL ||
268 		    ((label + 1) - dev) > done) {
269 			return (EINVAL);
270 		}
271 		*label = '\0';
272 		label++;
273 		if (*label == '\0')
274 			return (EINVAL);
275 
276 		if (mp->mnt_flag & MNT_UPDATE) {
277 			/* Update mount */
278 			/* HAMMER2 implements NFS export via mountctl */
279 			hmp = MPTOHMP(mp);
280 			devvp = hmp->devvp;
281 			error = hammer2_remount(mp, path, devvp, cred);
282 			return error;
283 		}
284 	}
285 
286 	/*
287 	 * PFS mount
288 	 *
289 	 * Lookup name and verify it refers to a block device.
290 	 */
291 	error = nlookup_init(&nd, dev, UIO_SYSSPACE, NLC_FOLLOW);
292 	if (error == 0)
293 		error = nlookup(&nd);
294 	if (error == 0)
295 		error = cache_vref(&nd.nl_nch, nd.nl_cred, &devvp);
296 	nlookup_done(&nd);
297 
298 	if (error == 0) {
299 		if (vn_isdisk(devvp, &error))
300 			error = vfs_mountedon(devvp);
301 	}
302 
303 	/*
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.
307 	 */
308 	lockmgr(&hammer2_mntlk, LK_EXCLUSIVE);
309 	TAILQ_FOREACH(hmp, &hammer2_mntlist, mntentry) {
310 		if (hmp->devvp == devvp)
311 			break;
312 	}
313 
314 	/*
315 	 * Open the device if this isn't a secondary mount
316 	 */
317 	if (hmp) {
318 		create_hmp = 0;
319 	} else {
320 		create_hmp = 1;
321 		if (error == 0 && vcount(devvp) > 0)
322 			error = EBUSY;
323 
324 		/*
325 		 * Now open the device
326 		 */
327 		if (error == 0) {
328 			ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
329 			vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
330 			error = vinvalbuf(devvp, V_SAVE, 0, 0);
331 			if (error == 0) {
332 				error = VOP_OPEN(devvp,
333 						 ronly ? FREAD : FREAD | FWRITE,
334 						 FSCRED, NULL);
335 			}
336 			vn_unlock(devvp);
337 		}
338 		if (error && devvp) {
339 			vrele(devvp);
340 			devvp = NULL;
341 		}
342 		if (error) {
343 			lockmgr(&hammer2_mntlk, LK_RELEASE);
344 			return error;
345 		}
346 	}
347 
348 	/*
349 	 * Block device opened successfully, finish initializing the
350 	 * mount structure.
351 	 *
352 	 * From this point on we have to call hammer2_unmount() on failure.
353 	 */
354 	pmp = kmalloc(sizeof(*pmp), M_HAMMER2, M_WAITOK | M_ZERO);
355 	mp->mnt_data = (qaddr_t)pmp;
356 	pmp->mp = mp;
357 
358 	spin_init(&pmp->inum_spin);
359 	RB_INIT(&pmp->inum_tree);
360 
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);
367 
368 	if (create_hmp) {
369 		hmp = kmalloc(sizeof(*hmp), M_HAMMER2, M_WAITOK | M_ZERO);
370 		hmp->ronly = ronly;
371 		hmp->devvp = devvp;
372 		kmalloc_create(&hmp->minode, "HAMMER2-inodes");
373 		kmalloc_create(&hmp->mchain, "HAMMER2-chains");
374 		TAILQ_INSERT_TAIL(&hammer2_mntlist, hmp, mntentry);
375 	}
376 	ccms_domain_init(&pmp->ccms_dom);
377 	pmp->hmp = hmp;
378 	++hmp->pmp_count;
379 	lockmgr(&hammer2_mntlk, LK_RELEASE);
380 	kprintf("hammer2_mount hmp=%p pmpcnt=%d\n", hmp, hmp->pmp_count);
381 
382 	mp->mnt_flag = MNT_LOCAL;
383 	mp->mnt_kern_flag |= MNTK_ALL_MPSAFE;	/* all entry pts are SMP */
384 
385 	if (create_hmp) {
386 		/*
387 		 * Presetup
388 		 */
389 		lockinit(&hmp->alloclk, "h2alloc", 0, 0);
390 		lockinit(&hmp->voldatalk, "voldata", 0, LK_CANRECURSE);
391 		TAILQ_INIT(&hmp->transq);
392 
393 		/*
394 		 * vchain setup. vchain.data is embedded.
395 		 * vchain.refs is initialized and will never drop to 0.
396 		 */
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 */
405 
406 		/*
407 		 * fchain setup.  fchain.data is embedded.
408 		 * fchain.refs is initialized and will never drop to 0.
409 		 *
410 		 * The data is not used but needs to be initialized to
411 		 * pass assertion muster.  We use this chain primarily
412 		 * as a placeholder for the freemap's top-level RBTREE
413 		 * so it does not interfere with the volume's topology
414 		 * RBTREE.
415 		 */
416 		hmp->fchain.hmp = hmp;
417 		hmp->fchain.refs = 1;
418 		hmp->fchain.data = (void *)&hmp->voldata.freemap_blockset;
419 		hmp->fchain.bref.type = HAMMER2_BREF_TYPE_FREEMAP;
420 		hmp->fchain.bref.data_off = 0 | HAMMER2_PBUFRADIX;
421 		hmp->fchain.bref.methods =
422 			HAMMER2_ENC_CHECK(HAMMER2_CHECK_FREEMAP) |
423 			HAMMER2_ENC_COMP(HAMMER2_COMP_NONE);
424 		hmp->fchain.delete_tid = HAMMER2_MAX_TID;
425 
426 		hammer2_chain_core_alloc(&hmp->fchain, NULL);
427 		/* hmp->fchain.u.xxx is left NULL */
428 
429 		/*
430 		 * Install the volume header
431 		 */
432 		error = hammer2_install_volume_header(hmp);
433 		if (error) {
434 			hammer2_vfs_unmount(mp, MNT_FORCE);
435 			return error;
436 		}
437 	}
438 
439 	/*
440 	 * required mount structure initializations
441 	 */
442 	mp->mnt_stat.f_iosize = HAMMER2_PBUFSIZE;
443 	mp->mnt_stat.f_bsize = HAMMER2_PBUFSIZE;
444 
445 	mp->mnt_vstat.f_frsize = HAMMER2_PBUFSIZE;
446 	mp->mnt_vstat.f_bsize = HAMMER2_PBUFSIZE;
447 
448 	/*
449 	 * Optional fields
450 	 */
451 	mp->mnt_iosize_max = MAXPHYS;
452 
453 	/*
454 	 * First locate the super-root inode, which is key 0 relative to the
455 	 * volume header's blockset.
456 	 *
457 	 * Then locate the root inode by scanning the directory keyspace
458 	 * represented by the label.
459 	 */
460 	if (create_hmp) {
461 		parent = hammer2_chain_lookup_init(&hmp->vchain, 0);
462 		schain = hammer2_chain_lookup(&parent,
463 				      HAMMER2_SROOT_KEY, HAMMER2_SROOT_KEY, 0);
464 		hammer2_chain_lookup_done(parent);
465 		if (schain == NULL) {
466 			kprintf("hammer2_mount: invalid super-root\n");
467 			hammer2_vfs_unmount(mp, MNT_FORCE);
468 			return EINVAL;
469 		}
470 		hammer2_chain_ref(schain);	/* for hmp->schain */
471 		hmp->schain = schain;		/* left locked for inode_get */
472 		hmp->sroot = hammer2_inode_get(hmp, NULL, NULL, schain);
473 		hammer2_inode_ref(hmp->sroot);	     /* for hmp->sroot */
474 		hammer2_inode_unlock_ex(hmp->sroot, schain);
475 	} else {
476 		schain = hmp->schain;
477 	}
478 
479 	/*
480 	 * schain only has 1 ref now for its hmp->schain assignment.
481 	 * Setup for lookup (which will lock it).
482 	 */
483 	parent = hammer2_chain_lookup_init(schain, 0);
484 	lhc = hammer2_dirhash(label, strlen(label));
485 	rchain = hammer2_chain_lookup(&parent,
486 				      lhc, lhc + HAMMER2_DIRHASH_LOMASK,
487 				      0);
488 	while (rchain) {
489 		if (rchain->bref.type == HAMMER2_BREF_TYPE_INODE &&
490 		    strcmp(label, rchain->data->ipdata.filename) == 0) {
491 			break;
492 		}
493 		rchain = hammer2_chain_next(&parent, rchain,
494 					    lhc, lhc + HAMMER2_DIRHASH_LOMASK,
495 					    0);
496 	}
497 	hammer2_chain_lookup_done(parent);
498 	if (rchain == NULL) {
499 		kprintf("hammer2_mount: PFS label not found\n");
500 		hammer2_vfs_unmount(mp, MNT_FORCE);
501 		return EINVAL;
502 	}
503 	if (rchain->flags & HAMMER2_CHAIN_MOUNTED) {
504 		hammer2_chain_unlock(rchain);
505 		kprintf("hammer2_mount: PFS label already mounted!\n");
506 		hammer2_vfs_unmount(mp, MNT_FORCE);
507 		return EBUSY;
508 	}
509 	if (rchain->flags & HAMMER2_CHAIN_RECYCLE) {
510 		kprintf("hammer2_mount: PFS label currently recycling\n");
511 		hammer2_vfs_unmount(mp, MNT_FORCE);
512 		return EBUSY;
513 	}
514 
515 	atomic_set_int(&rchain->flags, HAMMER2_CHAIN_MOUNTED);
516 
517 	/*
518 	 * NOTE: *_get() integrates chain's lock into the inode lock.
519 	 */
520 	hammer2_chain_ref(rchain);		/* for pmp->rchain */
521 	pmp->rchain = rchain;			/* left held & unlocked */
522 	pmp->iroot = hammer2_inode_get(hmp, pmp, NULL, rchain);
523 	hammer2_inode_ref(pmp->iroot);		/* ref for pmp->iroot */
524 	hammer2_inode_unlock_ex(pmp->iroot, rchain);
525 
526 	kprintf("iroot %p\n", pmp->iroot);
527 
528 	/*
529 	 * Ref the cluster management messaging descriptor.  The mount
530 	 * program deals with the other end of the communications pipe.
531 	 */
532 	fp = holdfp(curproc->p_fd, info.cluster_fd, -1);
533 	if (fp == NULL) {
534 		kprintf("hammer2_mount: bad cluster_fd!\n");
535 		hammer2_vfs_unmount(mp, MNT_FORCE);
536 		return EBADF;
537 	}
538 	hammer2_cluster_reconnect(pmp, fp);
539 
540 	/*
541 	 * Finish setup
542 	 */
543 	vfs_getnewfsid(mp);
544 	vfs_add_vnodeops(mp, &hammer2_vnode_vops, &mp->mnt_vn_norm_ops);
545 	vfs_add_vnodeops(mp, &hammer2_spec_vops, &mp->mnt_vn_spec_ops);
546 	vfs_add_vnodeops(mp, &hammer2_fifo_vops, &mp->mnt_vn_fifo_ops);
547 
548 	copyinstr(info.volume, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, &size);
549 	bzero(mp->mnt_stat.f_mntfromname + size, MNAMELEN - size);
550 	bzero(mp->mnt_stat.f_mntonname, sizeof(mp->mnt_stat.f_mntonname));
551 	copyinstr(path, mp->mnt_stat.f_mntonname,
552 		  sizeof(mp->mnt_stat.f_mntonname) - 1,
553 		  &size);
554 
555 	/*
556 	 * Initial statfs to prime mnt_stat.
557 	 */
558 	hammer2_vfs_statfs(mp, &mp->mnt_stat, cred);
559 
560 	return 0;
561 }
562 
563 static
564 int
565 hammer2_remount(struct mount *mp, char *path, struct vnode *devvp,
566                 struct ucred *cred)
567 {
568 	return (0);
569 }
570 
571 static
572 int
573 hammer2_vfs_unmount(struct mount *mp, int mntflags)
574 {
575 	hammer2_pfsmount_t *pmp;
576 	hammer2_mount_t *hmp;
577 	int flags;
578 	int error = 0;
579 	int ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
580 	int dumpcnt;
581 	struct vnode *devvp;
582 
583 	pmp = MPTOPMP(mp);
584 	hmp = pmp->hmp;
585 	flags = 0;
586 
587 	if (mntflags & MNT_FORCE)
588 		flags |= FORCECLOSE;
589 
590 	hammer2_mount_exlock(hmp);
591 
592 	/*
593 	 * If mount initialization proceeded far enough we must flush
594 	 * its vnodes.
595 	 */
596 	if (pmp->iroot)
597 		error = vflush(mp, 0, flags);
598 
599 	if (error) {
600 		hammer2_mount_unlock(hmp);
601 		return error;
602 	}
603 
604 	lockmgr(&hammer2_mntlk, LK_EXCLUSIVE);
605 	--hmp->pmp_count;
606 	kprintf("hammer2_unmount hmp=%p pmpcnt=%d\n", hmp, hmp->pmp_count);
607 
608 	/*
609 	 * Flush any left over chains.  The voldata lock is only used
610 	 * to synchronize against HAMMER2_CHAIN_MODIFIED_AUX.
611 	 */
612 	hammer2_voldata_lock(hmp);
613 	if (hmp->vchain.flags & (HAMMER2_CHAIN_MODIFIED |
614 				 HAMMER2_CHAIN_SUBMODIFIED)) {
615 		hammer2_voldata_unlock(hmp, 0);
616 		hammer2_vfs_sync(mp, MNT_WAIT);
617 	} else {
618 		hammer2_voldata_unlock(hmp, 0);
619 	}
620 	if (hmp->pmp_count == 0) {
621 		if (hmp->vchain.flags & (HAMMER2_CHAIN_MODIFIED |
622 					 HAMMER2_CHAIN_SUBMODIFIED)) {
623 			kprintf("hammer2_unmount: chains left over after "
624 				"final sync\n");
625 			if (hammer2_debug & 0x0010)
626 				Debugger("entered debugger");
627 		}
628 	}
629 
630 	/*
631 	 * Cleanup the root and super-root chain elements (which should be
632 	 * clean).
633 	 */
634 	if (pmp->iroot) {
635 #if REPORT_REFS_ERRORS
636 		if (pmp->iroot->refs != 1)
637 			kprintf("PMP->IROOT %p REFS WRONG %d\n",
638 				pmp->iroot, pmp->iroot->refs);
639 #else
640 		KKASSERT(pmp->iroot->refs == 1);
641 #endif
642 		hammer2_inode_drop(pmp->iroot);	    /* ref for pmp->iroot */
643 		pmp->iroot = NULL;
644 	}
645 	if (pmp->rchain) {
646 		atomic_clear_int(&pmp->rchain->flags, HAMMER2_CHAIN_MOUNTED);
647 #if REPORT_REFS_ERRORS
648 		if (pmp->rchain->refs != 1)
649 			kprintf("PMP->RCHAIN %p REFS WRONG %d\n",
650 				pmp->rchain, pmp->rchain->refs);
651 #else
652 		KKASSERT(pmp->rchain->refs == 1);
653 #endif
654 		hammer2_chain_drop(pmp->rchain);
655 		pmp->rchain = NULL;
656 	}
657 	ccms_domain_uninit(&pmp->ccms_dom);
658 
659 	/*
660 	 * Kill cluster controller
661 	 */
662 	kdmsg_iocom_uninit(&pmp->iocom);
663 
664 	/*
665 	 * If no PFS's left drop the master hammer2_mount for the device.
666 	 */
667 	if (hmp->pmp_count == 0) {
668 		if (hmp->sroot) {
669 			hammer2_inode_drop(hmp->sroot);
670 			hmp->sroot = NULL;
671 		}
672 		if (hmp->schain) {
673 #if REPORT_REFS_ERRORS
674 			if (hmp->schain->refs != 1)
675 				kprintf("HMP->SCHAIN %p REFS WRONG %d\n",
676 					hmp->schain, hmp->schain->refs);
677 #else
678 			KKASSERT(hmp->schain->refs == 1);
679 #endif
680 			hammer2_chain_drop(hmp->schain);
681 			hmp->schain = NULL;
682 		}
683 
684 		/*
685 		 * Finish up with the device vnode
686 		 */
687 		if ((devvp = hmp->devvp) != NULL) {
688 			vinvalbuf(devvp, (ronly ? 0 : V_SAVE), 0, 0);
689 			hmp->devvp = NULL;
690 			VOP_CLOSE(devvp, (ronly ? FREAD : FREAD|FWRITE));
691 			vrele(devvp);
692 			devvp = NULL;
693 		}
694 
695 		/*
696 		 * Final drop of embedded freemap root chain to clean up
697 		 * fchain.core (fchain structure is not flagged ALLOCATED
698 		 * so it is cleaned out and then left to rot).
699 		 */
700 		hammer2_chain_drop(&hmp->fchain);
701 
702 		/*
703 		 * Final drop of embedded volume root chain to clean up
704 		 * vchain.core (vchain structure is not flagged ALLOCATED
705 		 * so it is cleaned out and then left to rot).
706 		 */
707 		dumpcnt = 50;
708 		hammer2_dump_chain(&hmp->vchain, 0, &dumpcnt);
709 		hammer2_mount_unlock(hmp);
710 		hammer2_chain_drop(&hmp->vchain);
711 	} else {
712 		hammer2_mount_unlock(hmp);
713 	}
714 
715 	pmp->mp = NULL;
716 	pmp->hmp = NULL;
717 	mp->mnt_data = NULL;
718 
719 	kmalloc_destroy(&pmp->mmsg);
720 
721 	kfree(pmp, M_HAMMER2);
722 	if (hmp->pmp_count == 0) {
723 		TAILQ_REMOVE(&hammer2_mntlist, hmp, mntentry);
724 		kmalloc_destroy(&hmp->minode);
725 		kmalloc_destroy(&hmp->mchain);
726 		kfree(hmp, M_HAMMER2);
727 	}
728 	lockmgr(&hammer2_mntlk, LK_RELEASE);
729 	return (error);
730 }
731 
732 static
733 int
734 hammer2_vfs_vget(struct mount *mp, struct vnode *dvp,
735 	     ino_t ino, struct vnode **vpp)
736 {
737 	kprintf("hammer2_vget\n");
738 	return (EOPNOTSUPP);
739 }
740 
741 static
742 int
743 hammer2_vfs_root(struct mount *mp, struct vnode **vpp)
744 {
745 	hammer2_pfsmount_t *pmp;
746 	hammer2_mount_t *hmp;
747 	hammer2_chain_t *parent;
748 	int error;
749 	struct vnode *vp;
750 
751 	pmp = MPTOPMP(mp);
752 	hmp = pmp->hmp;
753 	hammer2_mount_exlock(hmp);
754 	if (pmp->iroot == NULL) {
755 		*vpp = NULL;
756 		error = EINVAL;
757 	} else {
758 		parent = hammer2_inode_lock_sh(pmp->iroot);
759 		vp = hammer2_igetv(pmp->iroot, &error);
760 		hammer2_inode_unlock_sh(pmp->iroot, parent);
761 		*vpp = vp;
762 		if (vp == NULL)
763 			kprintf("vnodefail\n");
764 	}
765 	hammer2_mount_unlock(hmp);
766 
767 	return (error);
768 }
769 
770 /*
771  * Filesystem status
772  *
773  * XXX incorporate ipdata->inode_quota and data_quota
774  */
775 static
776 int
777 hammer2_vfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
778 {
779 	hammer2_pfsmount_t *pmp;
780 	hammer2_mount_t *hmp;
781 
782 	pmp = MPTOPMP(mp);
783 	hmp = MPTOHMP(mp);
784 
785 	mp->mnt_stat.f_files = pmp->inode_count;
786 	mp->mnt_stat.f_ffree = 0;
787 	mp->mnt_stat.f_blocks = hmp->voldata.allocator_size / HAMMER2_PBUFSIZE;
788 	mp->mnt_stat.f_bfree = (hmp->voldata.allocator_size -
789 				hmp->voldata.allocator_beg) / HAMMER2_PBUFSIZE;
790 	mp->mnt_stat.f_bavail = mp->mnt_stat.f_bfree;
791 
792 	*sbp = mp->mnt_stat;
793 	return (0);
794 }
795 
796 static
797 int
798 hammer2_vfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred)
799 {
800 	hammer2_pfsmount_t *pmp;
801 	hammer2_mount_t *hmp;
802 
803 	pmp = MPTOPMP(mp);
804 	hmp = MPTOHMP(mp);
805 
806 	mp->mnt_vstat.f_bsize = HAMMER2_PBUFSIZE;
807 	mp->mnt_vstat.f_files = pmp->inode_count;
808 	mp->mnt_vstat.f_ffree = 0;
809 	mp->mnt_vstat.f_blocks = hmp->voldata.allocator_size / HAMMER2_PBUFSIZE;
810 	mp->mnt_vstat.f_bfree = (hmp->voldata.allocator_size -
811 				 hmp->voldata.allocator_beg) / HAMMER2_PBUFSIZE;
812 	mp->mnt_vstat.f_bavail = mp->mnt_vstat.f_bfree;
813 
814 	*sbp = mp->mnt_vstat;
815 	return (0);
816 }
817 
818 /*
819  * Sync the entire filesystem; this is called from the filesystem syncer
820  * process periodically and whenever a user calls sync(1) on the hammer
821  * mountpoint.
822  *
823  * Currently is actually called from the syncer! \o/
824  *
825  * This task will have to snapshot the state of the dirty inode chain.
826  * From that, it will have to make sure all of the inodes on the dirty
827  * chain have IO initiated. We make sure that io is initiated for the root
828  * block.
829  *
830  * If waitfor is set, we wait for media to acknowledge the new rootblock.
831  *
832  * THINKS: side A vs side B, to have sync not stall all I/O?
833  */
834 static
835 int
836 hammer2_vfs_sync(struct mount *mp, int waitfor)
837 {
838 	struct hammer2_sync_info info;
839 	hammer2_mount_t *hmp;
840 	int flags;
841 	int error;
842 	int i;
843 #if 0
844 	int dumpcnt;
845 #endif
846 
847 	hmp = MPTOHMP(mp);
848 #if 0
849 	if ((waitfor & MNT_LAZY) == 0) {
850 		dumpcnt = 50;
851 		hammer2_dump_chain(&hmp->vchain, 0, &dumpcnt);
852 	}
853 #endif
854 
855 	flags = VMSC_GETVP;
856 	if (waitfor & MNT_LAZY)
857 		flags |= VMSC_ONEPASS;
858 
859 	hammer2_trans_init(&info.trans, hmp, NULL, HAMMER2_TRANS_ISFLUSH);
860 
861 	info.error = 0;
862 	info.waitfor = MNT_NOWAIT;
863 	vmntvnodescan(mp, flags | VMSC_NOWAIT,
864 		      hammer2_sync_scan1,
865 		      hammer2_sync_scan2, &info);
866 	if (info.error == 0 && (waitfor & MNT_WAIT)) {
867 		info.waitfor = waitfor;
868 		    vmntvnodescan(mp, flags,
869 				  hammer2_sync_scan1,
870 				  hammer2_sync_scan2, &info);
871 
872 	}
873 #if 0
874 	if (waitfor == MNT_WAIT) {
875 		/* XXX */
876 	} else {
877 		/* XXX */
878 	}
879 #endif
880 	hammer2_chain_lock(&hmp->vchain, HAMMER2_RESOLVE_ALWAYS);
881 	if (hmp->vchain.flags & (HAMMER2_CHAIN_MODIFIED |
882 				  HAMMER2_CHAIN_SUBMODIFIED)) {
883 		hammer2_chain_flush(&info.trans, &hmp->vchain);
884 	}
885 	hammer2_chain_unlock(&hmp->vchain);
886 
887 #if 0
888 	/*
889 	 * Rollup flush.  The fsyncs above basically just flushed
890 	 * data blocks.  The flush below gets all the meta-data.
891 	 */
892 	hammer2_chain_lock(&hmp->fchain, HAMMER2_RESOLVE_ALWAYS);
893 	if (hmp->fchain.flags & (HAMMER2_CHAIN_MODIFIED |
894 				 HAMMER2_CHAIN_SUBMODIFIED)) {
895 		/* this will modify vchain as a side effect */
896 		hammer2_chain_flush(&info.trans, &hmp->fchain);
897 	}
898 	hammer2_chain_unlock(&hmp->fchain);
899 #endif
900 
901 
902 	error = 0;
903 
904 	/*
905 	 * We can't safely flush the volume header until we have
906 	 * flushed any device buffers which have built up.
907 	 *
908 	 * XXX this isn't being incremental
909 	 */
910 	vn_lock(hmp->devvp, LK_EXCLUSIVE | LK_RETRY);
911 	error = VOP_FSYNC(hmp->devvp, MNT_WAIT, 0);
912 	vn_unlock(hmp->devvp);
913 
914 	/*
915 	 * The flush code sets CHAIN_VOLUMESYNC to indicate that the
916 	 * volume header needs synchronization via hmp->volsync.
917 	 *
918 	 * XXX synchronize the flag & data with only this flush XXX
919 	 */
920 	if (error == 0 && (hmp->vchain.flags & HAMMER2_CHAIN_VOLUMESYNC)) {
921 		struct buf *bp;
922 
923 		/*
924 		 * Synchronize the disk before flushing the volume
925 		 * header.
926 		 */
927 		bp = getpbuf(NULL);
928 		bp->b_bio1.bio_offset = 0;
929 		bp->b_bufsize = 0;
930 		bp->b_bcount = 0;
931 		bp->b_cmd = BUF_CMD_FLUSH;
932 		bp->b_bio1.bio_done = biodone_sync;
933 		bp->b_bio1.bio_flags |= BIO_SYNC;
934 		vn_strategy(hmp->devvp, &bp->b_bio1);
935 		biowait(&bp->b_bio1, "h2vol");
936 		relpbuf(bp, NULL);
937 
938 		/*
939 		 * Then we can safely flush the version of the volume header
940 		 * synchronized by the flush code.
941 		 */
942 		i = hmp->volhdrno + 1;
943 		if (i >= HAMMER2_NUM_VOLHDRS)
944 			i = 0;
945 		if (i * HAMMER2_ZONE_BYTES64 + HAMMER2_SEGSIZE >
946 		    hmp->volsync.volu_size) {
947 			i = 0;
948 		}
949 		kprintf("sync volhdr %d %jd\n",
950 			i, (intmax_t)hmp->volsync.volu_size);
951 		bp = getblk(hmp->devvp, i * HAMMER2_ZONE_BYTES64,
952 			    HAMMER2_PBUFSIZE, 0, 0);
953 		atomic_clear_int(&hmp->vchain.flags, HAMMER2_CHAIN_VOLUMESYNC);
954 		bcopy(&hmp->volsync, bp->b_data, HAMMER2_PBUFSIZE);
955 		bawrite(bp);
956 		hmp->volhdrno = i;
957 	}
958 	hammer2_trans_done(&info.trans);
959 	return (error);
960 }
961 
962 /*
963  * Sync passes.
964  *
965  * NOTE: We don't test SUBMODIFIED or MOVED here because the fsync code
966  *	 won't flush on those flags.  The syncer code above will do a
967  *	 general meta-data flush globally that will catch these flags.
968  */
969 static int
970 hammer2_sync_scan1(struct mount *mp, struct vnode *vp, void *data)
971 {
972 	hammer2_inode_t *ip;
973 
974 	ip = VTOI(vp);
975 	if (vp->v_type == VNON || ip == NULL ||
976 	    ((ip->flags & HAMMER2_INODE_MODIFIED) == 0 &&
977 	     RB_EMPTY(&vp->v_rbdirty_tree))) {
978 		return(-1);
979 	}
980 	return(0);
981 }
982 
983 static int
984 hammer2_sync_scan2(struct mount *mp, struct vnode *vp, void *data)
985 {
986 	struct hammer2_sync_info *info = data;
987 	hammer2_inode_t *ip;
988 	hammer2_chain_t *parent;
989 	int error;
990 
991 	ip = VTOI(vp);
992 	if (vp->v_type == VNON || vp->v_type == VBAD ||
993 	    ((ip->flags & HAMMER2_INODE_MODIFIED) == 0 &&
994 	     RB_EMPTY(&vp->v_rbdirty_tree))) {
995 		return(0);
996 	}
997 
998 	/*
999 	 * VOP_FSYNC will start a new transaction so replicate some code
1000 	 * here to do it inline (see hammer2_vop_fsync()).
1001 	 */
1002 	parent = hammer2_inode_lock_ex(ip);
1003 	atomic_clear_int(&ip->flags, HAMMER2_INODE_MODIFIED);
1004 	if (ip->vp)
1005 		vfsync(ip->vp, MNT_NOWAIT, 1, NULL, NULL);
1006 	hammer2_chain_flush(&info->trans, parent);
1007 	hammer2_inode_unlock_ex(ip, parent);
1008 	error = 0;
1009 #if 0
1010 	error = VOP_FSYNC(vp, MNT_NOWAIT, 0);
1011 #endif
1012 	if (error)
1013 		info->error = error;
1014 	return(0);
1015 }
1016 
1017 static
1018 int
1019 hammer2_vfs_vptofh(struct vnode *vp, struct fid *fhp)
1020 {
1021 	return (0);
1022 }
1023 
1024 static
1025 int
1026 hammer2_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
1027 	       struct fid *fhp, struct vnode **vpp)
1028 {
1029 	return (0);
1030 }
1031 
1032 static
1033 int
1034 hammer2_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
1035 		 int *exflagsp, struct ucred **credanonp)
1036 {
1037 	return (0);
1038 }
1039 
1040 /*
1041  * Support code for hammer2_mount().  Read, verify, and install the volume
1042  * header into the HMP
1043  *
1044  * XXX read four volhdrs and use the one with the highest TID whos CRC
1045  *     matches.
1046  *
1047  * XXX check iCRCs.
1048  *
1049  * XXX For filesystems w/ less than 4 volhdrs, make sure to not write to
1050  *     nonexistant locations.
1051  *
1052  * XXX Record selected volhdr and ring updates to each of 4 volhdrs
1053  */
1054 static
1055 int
1056 hammer2_install_volume_header(hammer2_mount_t *hmp)
1057 {
1058 	hammer2_volume_data_t *vd;
1059 	struct buf *bp;
1060 	hammer2_crc32_t crc0, crc, bcrc0, bcrc;
1061 	int error_reported;
1062 	int error;
1063 	int valid;
1064 	int i;
1065 
1066 	error_reported = 0;
1067 	error = 0;
1068 	valid = 0;
1069 	bp = NULL;
1070 
1071 	/*
1072 	 * There are up to 4 copies of the volume header (syncs iterate
1073 	 * between them so there is no single master).  We don't trust the
1074 	 * volu_size field so we don't know precisely how large the filesystem
1075 	 * is, so depend on the OS to return an error if we go beyond the
1076 	 * block device's EOF.
1077 	 */
1078 	for (i = 0; i < HAMMER2_NUM_VOLHDRS; i++) {
1079 		error = bread(hmp->devvp, i * HAMMER2_ZONE_BYTES64,
1080 			      HAMMER2_VOLUME_BYTES, &bp);
1081 		if (error) {
1082 			brelse(bp);
1083 			bp = NULL;
1084 			continue;
1085 		}
1086 
1087 		vd = (struct hammer2_volume_data *) bp->b_data;
1088 		if ((vd->magic != HAMMER2_VOLUME_ID_HBO) &&
1089 		    (vd->magic != HAMMER2_VOLUME_ID_ABO)) {
1090 			brelse(bp);
1091 			bp = NULL;
1092 			continue;
1093 		}
1094 
1095 		if (vd->magic == HAMMER2_VOLUME_ID_ABO) {
1096 			/* XXX: Reversed-endianness filesystem */
1097 			kprintf("hammer2: reverse-endian filesystem detected");
1098 			brelse(bp);
1099 			bp = NULL;
1100 			continue;
1101 		}
1102 
1103 		crc = vd->icrc_sects[HAMMER2_VOL_ICRC_SECT0];
1104 		crc0 = hammer2_icrc32(bp->b_data + HAMMER2_VOLUME_ICRC0_OFF,
1105 				      HAMMER2_VOLUME_ICRC0_SIZE);
1106 		bcrc = vd->icrc_sects[HAMMER2_VOL_ICRC_SECT1];
1107 		bcrc0 = hammer2_icrc32(bp->b_data + HAMMER2_VOLUME_ICRC1_OFF,
1108 				       HAMMER2_VOLUME_ICRC1_SIZE);
1109 		if ((crc0 != crc) || (bcrc0 != bcrc)) {
1110 			kprintf("hammer2 volume header crc "
1111 				"mismatch copy #%d %08x/%08x\n",
1112 				i, crc0, crc);
1113 			error_reported = 1;
1114 			brelse(bp);
1115 			bp = NULL;
1116 			continue;
1117 		}
1118 		if (valid == 0 || hmp->voldata.mirror_tid < vd->mirror_tid) {
1119 			valid = 1;
1120 			hmp->voldata = *vd;
1121 			hmp->volhdrno = i;
1122 		}
1123 		brelse(bp);
1124 		bp = NULL;
1125 	}
1126 	if (valid) {
1127 		hmp->volsync = hmp->voldata;
1128 		error = 0;
1129 		if (error_reported || bootverbose || 1) { /* 1/DEBUG */
1130 			kprintf("hammer2: using volume header #%d\n",
1131 				hmp->volhdrno);
1132 		}
1133 	} else {
1134 		error = EINVAL;
1135 		kprintf("hammer2: no valid volume headers found!\n");
1136 	}
1137 	return (error);
1138 }
1139 
1140 /*
1141  * Reconnect using the passed file pointer.  The caller must ref the
1142  * fp for us.
1143  */
1144 void
1145 hammer2_cluster_reconnect(hammer2_pfsmount_t *pmp, struct file *fp)
1146 {
1147 	hammer2_inode_data_t *ipdata;
1148 	hammer2_chain_t *parent;
1149 	size_t name_len;
1150 
1151 	/*
1152 	 * Closes old comm descriptor, kills threads, cleans up
1153 	 * states, then installs the new descriptor and creates
1154 	 * new threads.
1155 	 */
1156 	kdmsg_iocom_reconnect(&pmp->iocom, fp, "hammer2");
1157 
1158 	/*
1159 	 * Setup LNK_CONN fields for autoinitiated state machine
1160 	 */
1161 	parent = hammer2_inode_lock_ex(pmp->iroot);
1162 	ipdata = &parent->data->ipdata;
1163 	pmp->iocom.auto_lnk_conn.pfs_clid = ipdata->pfs_clid;
1164 	pmp->iocom.auto_lnk_conn.pfs_fsid = ipdata->pfs_fsid;
1165 	pmp->iocom.auto_lnk_conn.pfs_type = ipdata->pfs_type;
1166 	pmp->iocom.auto_lnk_conn.proto_version = DMSG_SPAN_PROTO_1;
1167 	pmp->iocom.auto_lnk_conn.peer_type = pmp->hmp->voldata.peer_type;
1168 
1169 	/*
1170 	 * Filter adjustment.  Clients do not need visibility into other
1171 	 * clients (otherwise millions of clients would present a serious
1172 	 * problem).  The fs_label also serves to restrict the namespace.
1173 	 */
1174 	pmp->iocom.auto_lnk_conn.peer_mask = 1LLU << HAMMER2_PEER_HAMMER2;
1175 	pmp->iocom.auto_lnk_conn.pfs_mask = (uint64_t)-1;
1176 	switch (ipdata->pfs_type) {
1177 	case DMSG_PFSTYPE_CLIENT:
1178 		pmp->iocom.auto_lnk_conn.peer_mask &=
1179 				~(1LLU << DMSG_PFSTYPE_CLIENT);
1180 		break;
1181 	default:
1182 		break;
1183 	}
1184 
1185 	name_len = ipdata->name_len;
1186 	if (name_len >= sizeof(pmp->iocom.auto_lnk_conn.fs_label))
1187 		name_len = sizeof(pmp->iocom.auto_lnk_conn.fs_label) - 1;
1188 	bcopy(ipdata->filename,
1189 	      pmp->iocom.auto_lnk_conn.fs_label,
1190 	      name_len);
1191 	pmp->iocom.auto_lnk_conn.fs_label[name_len] = 0;
1192 
1193 	/*
1194 	 * Setup LNK_SPAN fields for autoinitiated state machine
1195 	 */
1196 	pmp->iocom.auto_lnk_span.pfs_clid = ipdata->pfs_clid;
1197 	pmp->iocom.auto_lnk_span.pfs_fsid = ipdata->pfs_fsid;
1198 	pmp->iocom.auto_lnk_span.pfs_type = ipdata->pfs_type;
1199 	pmp->iocom.auto_lnk_span.peer_type = pmp->hmp->voldata.peer_type;
1200 	pmp->iocom.auto_lnk_span.proto_version = DMSG_SPAN_PROTO_1;
1201 	name_len = ipdata->name_len;
1202 	if (name_len >= sizeof(pmp->iocom.auto_lnk_span.fs_label))
1203 		name_len = sizeof(pmp->iocom.auto_lnk_span.fs_label) - 1;
1204 	bcopy(ipdata->filename,
1205 	      pmp->iocom.auto_lnk_span.fs_label,
1206 	      name_len);
1207 	pmp->iocom.auto_lnk_span.fs_label[name_len] = 0;
1208 	hammer2_inode_unlock_ex(pmp->iroot, parent);
1209 
1210 	kdmsg_iocom_autoinitiate(&pmp->iocom, hammer2_autodmsg);
1211 }
1212 
1213 static int
1214 hammer2_rcvdmsg(kdmsg_msg_t *msg)
1215 {
1216 	switch(msg->any.head.cmd & DMSGF_TRANSMASK) {
1217 	case DMSG_DBG_SHELL:
1218 		/*
1219 		 * (non-transaction)
1220 		 * Execute shell command (not supported atm)
1221 		 */
1222 		kdmsg_msg_reply(msg, DMSG_ERR_NOSUPP);
1223 		break;
1224 	case DMSG_DBG_SHELL | DMSGF_REPLY:
1225 		/*
1226 		 * (non-transaction)
1227 		 */
1228 		if (msg->aux_data) {
1229 			msg->aux_data[msg->aux_size - 1] = 0;
1230 			kprintf("HAMMER2 DBG: %s\n", msg->aux_data);
1231 		}
1232 		break;
1233 	default:
1234 		/*
1235 		 * Unsupported message received.  We only need to
1236 		 * reply if it's a transaction in order to close our end.
1237 		 * Ignore any one-way messages are any further messages
1238 		 * associated with the transaction.
1239 		 *
1240 		 * NOTE: This case also includes DMSG_LNK_ERROR messages
1241 		 *	 which might be one-way, replying to those would
1242 		 *	 cause an infinite ping-pong.
1243 		 */
1244 		if (msg->any.head.cmd & DMSGF_CREATE)
1245 			kdmsg_msg_reply(msg, DMSG_ERR_NOSUPP);
1246 		break;
1247 	}
1248 	return(0);
1249 }
1250 
1251 /*
1252  * This function is called after KDMSG has automatically handled processing
1253  * of a LNK layer message (typically CONN, SPAN, or CIRC).
1254  *
1255  * We tag off the LNK_CONN to trigger our LNK_VOLCONF messages which
1256  * advertises all available hammer2 super-root volumes.
1257  */
1258 static void
1259 hammer2_autodmsg(kdmsg_msg_t *msg)
1260 {
1261 	hammer2_pfsmount_t *pmp = msg->iocom->handle;
1262 	hammer2_mount_t *hmp = pmp->hmp;
1263 	int copyid;
1264 
1265 	/*
1266 	 * We only care about replies to our LNK_CONN auto-request.  kdmsg
1267 	 * has already processed the reply, we use this calback as a shim
1268 	 * to know when we can advertise available super-root volumes.
1269 	 */
1270 	if ((msg->any.head.cmd & DMSGF_TRANSMASK) !=
1271 	    (DMSG_LNK_CONN | DMSGF_CREATE | DMSGF_REPLY) ||
1272 	    msg->state == NULL) {
1273 		return;
1274 	}
1275 
1276 	kprintf("LNK_CONN REPLY RECEIVED CMD %08x\n", msg->any.head.cmd);
1277 
1278 	if (msg->any.head.cmd & DMSGF_CREATE) {
1279 		kprintf("HAMMER2: VOLDATA DUMP\n");
1280 
1281 		/*
1282 		 * Dump the configuration stored in the volume header
1283 		 */
1284 		hammer2_voldata_lock(hmp);
1285 		for (copyid = 0; copyid < HAMMER2_COPYID_COUNT; ++copyid) {
1286 			if (hmp->voldata.copyinfo[copyid].copyid == 0)
1287 				continue;
1288 			hammer2_volconf_update(pmp, copyid);
1289 		}
1290 		hammer2_voldata_unlock(hmp, 0);
1291 	}
1292 	if ((msg->any.head.cmd & DMSGF_DELETE) &&
1293 	    msg->state && (msg->state->txcmd & DMSGF_DELETE) == 0) {
1294 		kprintf("HAMMER2: CONN WAS TERMINATED\n");
1295 	}
1296 }
1297 
1298 /*
1299  * Volume configuration updates are passed onto the userland service
1300  * daemon via the open LNK_CONN transaction.
1301  */
1302 void
1303 hammer2_volconf_update(hammer2_pfsmount_t *pmp, int index)
1304 {
1305 	hammer2_mount_t *hmp = pmp->hmp;
1306 	kdmsg_msg_t *msg;
1307 
1308 	/* XXX interlock against connection state termination */
1309 	kprintf("volconf update %p\n", pmp->iocom.conn_state);
1310 	if (pmp->iocom.conn_state) {
1311 		kprintf("TRANSMIT VOLCONF VIA OPEN CONN TRANSACTION\n");
1312 		msg = kdmsg_msg_alloc_state(pmp->iocom.conn_state,
1313 					    DMSG_LNK_VOLCONF, NULL, NULL);
1314 		msg->any.lnk_volconf.copy = hmp->voldata.copyinfo[index];
1315 		msg->any.lnk_volconf.mediaid = hmp->voldata.fsid;
1316 		msg->any.lnk_volconf.index = index;
1317 		kdmsg_msg_write(msg);
1318 	}
1319 }
1320 
1321 void
1322 hammer2_dump_chain(hammer2_chain_t *chain, int tab, int *countp)
1323 {
1324 	hammer2_chain_t *scan;
1325 
1326 	--*countp;
1327 	if (*countp == 0) {
1328 		kprintf("%*.*s...\n", tab, tab, "");
1329 		return;
1330 	}
1331 	if (*countp < 0)
1332 		return;
1333 	kprintf("%*.*schain[%d] %p.%d [%08x][core=%p] (%s) dl=%p dt=%s refs=%d",
1334 		tab, tab, "",
1335 		chain->index, chain, chain->bref.type, chain->flags,
1336 		chain->core,
1337 		((chain->bref.type == HAMMER2_BREF_TYPE_INODE &&
1338 		chain->data) ?  (char *)chain->data->ipdata.filename : "?"),
1339 		chain->next_parent,
1340 		(chain->delete_tid == HAMMER2_MAX_TID ? "max" : "fls"),
1341 		chain->refs);
1342 	if (chain->core == NULL || RB_EMPTY(&chain->core->rbtree))
1343 		kprintf("\n");
1344 	else
1345 		kprintf(" {\n");
1346 	RB_FOREACH(scan, hammer2_chain_tree, &chain->core->rbtree) {
1347 		hammer2_dump_chain(scan, tab + 4, countp);
1348 	}
1349 	if (chain->core && !RB_EMPTY(&chain->core->rbtree)) {
1350 		if (chain->bref.type == HAMMER2_BREF_TYPE_INODE && chain->data)
1351 			kprintf("%*.*s}(%s)\n", tab, tab, "",
1352 				chain->data->ipdata.filename);
1353 		else
1354 			kprintf("%*.*s}\n", tab, tab, "");
1355 	}
1356 }
1357