xref: /dflybsd-src/sys/vfs/hammer2/hammer2_vfsops.c (revision 4bbca3a905b195fcafccbf8e264fbb27b05a6f6b)
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  * by Daniel Flores (GSOC 2013 - mentored by Matthew Dillon, compression)
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  *
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in
16  *    the documentation and/or other materials provided with the
17  *    distribution.
18  * 3. Neither the name of The DragonFly Project nor the names of its
19  *    contributors may be used to endorse or promote products derived
20  *    from this software without specific, prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
23  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
24  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
25  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
26  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
27  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
28  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
29  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
30  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
31  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
32  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33  * SUCH DAMAGE.
34  */
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/kernel.h>
38 #include <sys/nlookup.h>
39 #include <sys/vnode.h>
40 #include <sys/mount.h>
41 #include <sys/fcntl.h>
42 #include <sys/buf.h>
43 #include <sys/uuid.h>
44 #include <sys/vfsops.h>
45 #include <sys/sysctl.h>
46 #include <sys/socket.h>
47 #include <sys/objcache.h>
48 
49 #include <sys/proc.h>
50 #include <sys/namei.h>
51 #include <sys/mountctl.h>
52 #include <sys/dirent.h>
53 #include <sys/uio.h>
54 
55 #include <sys/mutex.h>
56 #include <sys/mutex2.h>
57 
58 #include "hammer2.h"
59 #include "hammer2_disk.h"
60 #include "hammer2_mount.h"
61 
62 #include "hammer2.h"
63 #include "hammer2_lz4.h"
64 
65 #include "zlib/hammer2_zlib.h"
66 
67 #define REPORT_REFS_ERRORS 1	/* XXX remove me */
68 
69 MALLOC_DEFINE(M_OBJCACHE, "objcache", "Object Cache");
70 
71 struct hammer2_sync_info {
72 	hammer2_trans_t trans;
73 	int error;
74 	int waitfor;
75 };
76 
77 TAILQ_HEAD(hammer2_mntlist, hammer2_mount);
78 static struct hammer2_mntlist hammer2_mntlist;
79 static struct lock hammer2_mntlk;
80 
81 int hammer2_debug;
82 int hammer2_cluster_enable = 1;
83 int hammer2_hardlink_enable = 1;
84 long hammer2_iod_file_read;
85 long hammer2_iod_meta_read;
86 long hammer2_iod_indr_read;
87 long hammer2_iod_fmap_read;
88 long hammer2_iod_volu_read;
89 long hammer2_iod_file_write;
90 long hammer2_iod_meta_write;
91 long hammer2_iod_indr_write;
92 long hammer2_iod_fmap_write;
93 long hammer2_iod_volu_write;
94 long hammer2_ioa_file_read;
95 long hammer2_ioa_meta_read;
96 long hammer2_ioa_indr_read;
97 long hammer2_ioa_fmap_read;
98 long hammer2_ioa_volu_read;
99 long hammer2_ioa_fmap_write;
100 long hammer2_ioa_file_write;
101 long hammer2_ioa_meta_write;
102 long hammer2_ioa_indr_write;
103 long hammer2_ioa_volu_write;
104 
105 MALLOC_DECLARE(C_BUFFER);
106 MALLOC_DEFINE(C_BUFFER, "compbuffer", "Buffer used for compression.");
107 
108 MALLOC_DECLARE(D_BUFFER);
109 MALLOC_DEFINE(D_BUFFER, "decompbuffer", "Buffer used for decompression.");
110 
111 MALLOC_DECLARE(W_BIOQUEUE);
112 MALLOC_DEFINE(W_BIOQUEUE, "wbioqueue", "Writing bio queue.");
113 
114 MALLOC_DECLARE(W_MTX);
115 MALLOC_DEFINE(W_MTX, "wmutex", "Mutex for write thread.");
116 
117 SYSCTL_NODE(_vfs, OID_AUTO, hammer2, CTLFLAG_RW, 0, "HAMMER2 filesystem");
118 
119 SYSCTL_INT(_vfs_hammer2, OID_AUTO, debug, CTLFLAG_RW,
120 	   &hammer2_debug, 0, "");
121 SYSCTL_INT(_vfs_hammer2, OID_AUTO, cluster_enable, CTLFLAG_RW,
122 	   &hammer2_cluster_enable, 0, "");
123 SYSCTL_INT(_vfs_hammer2, OID_AUTO, hardlink_enable, CTLFLAG_RW,
124 	   &hammer2_hardlink_enable, 0, "");
125 
126 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_file_read, CTLFLAG_RW,
127 	   &hammer2_iod_file_read, 0, "");
128 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_meta_read, CTLFLAG_RW,
129 	   &hammer2_iod_meta_read, 0, "");
130 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_indr_read, CTLFLAG_RW,
131 	   &hammer2_iod_indr_read, 0, "");
132 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_fmap_read, CTLFLAG_RW,
133 	   &hammer2_iod_fmap_read, 0, "");
134 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_volu_read, CTLFLAG_RW,
135 	   &hammer2_iod_volu_read, 0, "");
136 
137 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_file_write, CTLFLAG_RW,
138 	   &hammer2_iod_file_write, 0, "");
139 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_meta_write, CTLFLAG_RW,
140 	   &hammer2_iod_meta_write, 0, "");
141 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_indr_write, CTLFLAG_RW,
142 	   &hammer2_iod_indr_write, 0, "");
143 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_fmap_write, CTLFLAG_RW,
144 	   &hammer2_iod_fmap_write, 0, "");
145 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_volu_write, CTLFLAG_RW,
146 	   &hammer2_iod_volu_write, 0, "");
147 
148 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_file_read, CTLFLAG_RW,
149 	   &hammer2_ioa_file_read, 0, "");
150 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_meta_read, CTLFLAG_RW,
151 	   &hammer2_ioa_meta_read, 0, "");
152 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_indr_read, CTLFLAG_RW,
153 	   &hammer2_ioa_indr_read, 0, "");
154 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_fmap_read, CTLFLAG_RW,
155 	   &hammer2_ioa_fmap_read, 0, "");
156 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_volu_read, CTLFLAG_RW,
157 	   &hammer2_ioa_volu_read, 0, "");
158 
159 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_file_write, CTLFLAG_RW,
160 	   &hammer2_ioa_file_write, 0, "");
161 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_meta_write, CTLFLAG_RW,
162 	   &hammer2_ioa_meta_write, 0, "");
163 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_indr_write, CTLFLAG_RW,
164 	   &hammer2_ioa_indr_write, 0, "");
165 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_fmap_write, CTLFLAG_RW,
166 	   &hammer2_ioa_fmap_write, 0, "");
167 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_volu_write, CTLFLAG_RW,
168 	   &hammer2_ioa_volu_write, 0, "");
169 
170 static int hammer2_vfs_init(struct vfsconf *conf);
171 static int hammer2_vfs_uninit(struct vfsconf *vfsp);
172 static int hammer2_vfs_mount(struct mount *mp, char *path, caddr_t data,
173 				struct ucred *cred);
174 static int hammer2_remount(hammer2_mount_t *, char *, struct vnode *,
175 				struct ucred *);
176 static int hammer2_vfs_unmount(struct mount *mp, int mntflags);
177 static int hammer2_vfs_root(struct mount *mp, struct vnode **vpp);
178 static int hammer2_vfs_statfs(struct mount *mp, struct statfs *sbp,
179 				struct ucred *cred);
180 static int hammer2_vfs_statvfs(struct mount *mp, struct statvfs *sbp,
181 				struct ucred *cred);
182 static int hammer2_vfs_sync(struct mount *mp, int waitfor);
183 static int hammer2_vfs_vget(struct mount *mp, struct vnode *dvp,
184 				ino_t ino, struct vnode **vpp);
185 static int hammer2_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
186 				struct fid *fhp, struct vnode **vpp);
187 static int hammer2_vfs_vptofh(struct vnode *vp, struct fid *fhp);
188 static int hammer2_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
189 				int *exflagsp, struct ucred **credanonp);
190 
191 static int hammer2_install_volume_header(hammer2_mount_t *hmp);
192 static int hammer2_sync_scan1(struct mount *mp, struct vnode *vp, void *data);
193 static int hammer2_sync_scan2(struct mount *mp, struct vnode *vp, void *data);
194 
195 static void hammer2_write_thread(void *arg);
196 
197 /*
198  * Functions for compression in threads,
199  * from hammer2_vnops.c
200  */
201 static void hammer2_write_file_core(struct buf *bp, hammer2_trans_t *trans,
202 				hammer2_inode_t *ip,
203 				hammer2_inode_data_t *ipdata,
204 				hammer2_chain_t **parentp,
205 				hammer2_key_t lbase, int ioflag, int pblksize,
206 				int *errorp);
207 static void hammer2_compress_and_write(struct buf *bp, hammer2_trans_t *trans,
208 				hammer2_inode_t *ip,
209 				hammer2_inode_data_t *ipdata,
210 				hammer2_chain_t **parentp,
211 				hammer2_key_t lbase, int ioflag,
212 				int pblksize, int *errorp, int comp_algo);
213 static void hammer2_zero_check_and_write(struct buf *bp,
214 				hammer2_trans_t *trans, hammer2_inode_t *ip,
215 				hammer2_inode_data_t *ipdata,
216 				hammer2_chain_t **parentp,
217 				hammer2_key_t lbase,
218 				int ioflag, int pblksize, int *errorp);
219 static int test_block_zeros(const char *buf, size_t bytes);
220 static void zero_write(struct buf *bp, hammer2_trans_t *trans,
221 				hammer2_inode_t *ip,
222 				hammer2_inode_data_t *ipdata,
223 				hammer2_chain_t **parentp,
224 				hammer2_key_t lbase,
225 				int *errorp);
226 static void hammer2_write_bp(hammer2_chain_t *chain, struct buf *bp,
227 				int ioflag, int pblksize, int *errorp);
228 
229 static int hammer2_rcvdmsg(kdmsg_msg_t *msg);
230 static void hammer2_autodmsg(kdmsg_msg_t *msg);
231 
232 
233 /*
234  * HAMMER2 vfs operations.
235  */
236 static struct vfsops hammer2_vfsops = {
237 	.vfs_init	= hammer2_vfs_init,
238 	.vfs_uninit = hammer2_vfs_uninit,
239 	.vfs_sync	= hammer2_vfs_sync,
240 	.vfs_mount	= hammer2_vfs_mount,
241 	.vfs_unmount	= hammer2_vfs_unmount,
242 	.vfs_root 	= hammer2_vfs_root,
243 	.vfs_statfs	= hammer2_vfs_statfs,
244 	.vfs_statvfs	= hammer2_vfs_statvfs,
245 	.vfs_vget	= hammer2_vfs_vget,
246 	.vfs_vptofh	= hammer2_vfs_vptofh,
247 	.vfs_fhtovp	= hammer2_vfs_fhtovp,
248 	.vfs_checkexp	= hammer2_vfs_checkexp
249 };
250 
251 MALLOC_DEFINE(M_HAMMER2, "HAMMER2-mount", "");
252 
253 VFS_SET(hammer2_vfsops, hammer2, 0);
254 MODULE_VERSION(hammer2, 1);
255 
256 static
257 int
258 hammer2_vfs_init(struct vfsconf *conf)
259 {
260 	static struct objcache_malloc_args margs_read;
261 	static struct objcache_malloc_args margs_write;
262 
263 	int error;
264 
265 	error = 0;
266 
267 	if (HAMMER2_BLOCKREF_BYTES != sizeof(struct hammer2_blockref))
268 		error = EINVAL;
269 	if (HAMMER2_INODE_BYTES != sizeof(struct hammer2_inode_data))
270 		error = EINVAL;
271 	if (HAMMER2_VOLUME_BYTES != sizeof(struct hammer2_volume_data))
272 		error = EINVAL;
273 
274 	if (error)
275 		kprintf("HAMMER2 structure size mismatch; cannot continue.\n");
276 
277 	margs_read.objsize = 65536;
278 	margs_read.mtype = D_BUFFER;
279 
280 	margs_write.objsize = 32768;
281 	margs_write.mtype = C_BUFFER;
282 
283 	cache_buffer_read = objcache_create(margs_read.mtype->ks_shortdesc,
284 				0, 1, NULL, NULL, NULL, objcache_malloc_alloc,
285 				objcache_malloc_free, &margs_read);
286 	cache_buffer_write = objcache_create(margs_write.mtype->ks_shortdesc,
287 				0, 1, NULL, NULL, NULL, objcache_malloc_alloc,
288 				objcache_malloc_free, &margs_write);
289 
290 	lockinit(&hammer2_mntlk, "mntlk", 0, 0);
291 	TAILQ_INIT(&hammer2_mntlist);
292 
293 	return (error);
294 }
295 
296 static
297 int
298 hammer2_vfs_uninit(struct vfsconf *vfsp __unused)
299 {
300 	objcache_destroy(cache_buffer_read);
301 	objcache_destroy(cache_buffer_write);
302 	return 0;
303 }
304 
305 /*
306  * Mount or remount HAMMER2 fileystem from physical media
307  *
308  *	mountroot
309  *		mp		mount point structure
310  *		path		NULL
311  *		data		<unused>
312  *		cred		<unused>
313  *
314  *	mount
315  *		mp		mount point structure
316  *		path		path to mount point
317  *		data		pointer to argument structure in user space
318  *			volume	volume path (device@LABEL form)
319  *			hflags	user mount flags
320  *		cred		user credentials
321  *
322  * RETURNS:	0	Success
323  *		!0	error number
324  */
325 static
326 int
327 hammer2_vfs_mount(struct mount *mp, char *path, caddr_t data,
328 		  struct ucred *cred)
329 {
330 	struct hammer2_mount_info info;
331 	hammer2_pfsmount_t *pmp;
332 	hammer2_mount_t *hmp;
333 	hammer2_key_t key_next;
334 	hammer2_key_t key_dummy;
335 	hammer2_key_t lhc;
336 	struct vnode *devvp;
337 	struct nlookupdata nd;
338 	hammer2_chain_t *parent;
339 	hammer2_chain_t *schain;
340 	hammer2_chain_t *rchain;
341 	struct file *fp;
342 	char devstr[MNAMELEN];
343 	size_t size;
344 	size_t done;
345 	char *dev;
346 	char *label;
347 	int ronly = 1;
348 	int error;
349 	int cache_index;
350 	int i;
351 
352 	hmp = NULL;
353 	pmp = NULL;
354 	dev = NULL;
355 	label = NULL;
356 	devvp = NULL;
357 	cache_index = -1;
358 
359 	kprintf("hammer2_mount\n");
360 
361 	if (path == NULL) {
362 		/*
363 		 * Root mount
364 		 */
365 		bzero(&info, sizeof(info));
366 		info.cluster_fd = -1;
367 		return (EOPNOTSUPP);
368 	} else {
369 		/*
370 		 * Non-root mount or updating a mount
371 		 */
372 		error = copyin(data, &info, sizeof(info));
373 		if (error)
374 			return (error);
375 
376 		error = copyinstr(info.volume, devstr, MNAMELEN - 1, &done);
377 		if (error)
378 			return (error);
379 
380 		/* Extract device and label */
381 		dev = devstr;
382 		label = strchr(devstr, '@');
383 		if (label == NULL ||
384 		    ((label + 1) - dev) > done) {
385 			return (EINVAL);
386 		}
387 		*label = '\0';
388 		label++;
389 		if (*label == '\0')
390 			return (EINVAL);
391 
392 		if (mp->mnt_flag & MNT_UPDATE) {
393 			/* Update mount */
394 			/* HAMMER2 implements NFS export via mountctl */
395 			pmp = MPTOPMP(mp);
396 			for (i = 0; i < pmp->cluster.nchains; ++i) {
397 				hmp = pmp->cluster.chains[i]->hmp;
398 				devvp = hmp->devvp;
399 				error = hammer2_remount(hmp, path, devvp, cred);
400 				if (error)
401 					break;
402 			}
403 			return error;
404 		}
405 	}
406 
407 	/*
408 	 * PFS mount
409 	 *
410 	 * Lookup name and verify it refers to a block device.
411 	 */
412 	error = nlookup_init(&nd, dev, UIO_SYSSPACE, NLC_FOLLOW);
413 	if (error == 0)
414 		error = nlookup(&nd);
415 	if (error == 0)
416 		error = cache_vref(&nd.nl_nch, nd.nl_cred, &devvp);
417 	nlookup_done(&nd);
418 
419 	if (error == 0) {
420 		if (vn_isdisk(devvp, &error))
421 			error = vfs_mountedon(devvp);
422 	}
423 
424 	/*
425 	 * Determine if the device has already been mounted.  After this
426 	 * check hmp will be non-NULL if we are doing the second or more
427 	 * hammer2 mounts from the same device.
428 	 */
429 	lockmgr(&hammer2_mntlk, LK_EXCLUSIVE);
430 	TAILQ_FOREACH(hmp, &hammer2_mntlist, mntentry) {
431 		if (hmp->devvp == devvp)
432 			break;
433 	}
434 
435 	/*
436 	 * Open the device if this isn't a secondary mount and construct
437 	 * the H2 device mount (hmp).
438 	 */
439 	if (hmp == NULL) {
440 		if (error == 0 && vcount(devvp) > 0)
441 			error = EBUSY;
442 
443 		/*
444 		 * Now open the device
445 		 */
446 		if (error == 0) {
447 			ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
448 			vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
449 			error = vinvalbuf(devvp, V_SAVE, 0, 0);
450 			if (error == 0) {
451 				error = VOP_OPEN(devvp,
452 						 ronly ? FREAD : FREAD | FWRITE,
453 						 FSCRED, NULL);
454 			}
455 			vn_unlock(devvp);
456 		}
457 		if (error && devvp) {
458 			vrele(devvp);
459 			devvp = NULL;
460 		}
461 		if (error) {
462 			lockmgr(&hammer2_mntlk, LK_RELEASE);
463 			return error;
464 		}
465 		hmp = kmalloc(sizeof(*hmp), M_HAMMER2, M_WAITOK | M_ZERO);
466 		hmp->ronly = ronly;
467 		hmp->devvp = devvp;
468 		kmalloc_create(&hmp->mchain, "HAMMER2-chains");
469 		TAILQ_INSERT_TAIL(&hammer2_mntlist, hmp, mntentry);
470 
471 		lockinit(&hmp->alloclk, "h2alloc", 0, 0);
472 		lockinit(&hmp->voldatalk, "voldata", 0, LK_CANRECURSE);
473 		TAILQ_INIT(&hmp->transq);
474 
475 		/*
476 		 * vchain setup. vchain.data is embedded.
477 		 * vchain.refs is initialized and will never drop to 0.
478 		 */
479 		hmp->vchain.hmp = hmp;
480 		hmp->vchain.refs = 1;
481 		hmp->vchain.data = (void *)&hmp->voldata;
482 		hmp->vchain.bref.type = HAMMER2_BREF_TYPE_VOLUME;
483 		hmp->vchain.bref.data_off = 0 | HAMMER2_PBUFRADIX;
484 		hmp->vchain.delete_tid = HAMMER2_MAX_TID;
485 		hammer2_chain_core_alloc(NULL, &hmp->vchain, NULL);
486 		/* hmp->vchain.u.xxx is left NULL */
487 
488 		/*
489 		 * fchain setup.  fchain.data is embedded.
490 		 * fchain.refs is initialized and will never drop to 0.
491 		 *
492 		 * The data is not used but needs to be initialized to
493 		 * pass assertion muster.  We use this chain primarily
494 		 * as a placeholder for the freemap's top-level RBTREE
495 		 * so it does not interfere with the volume's topology
496 		 * RBTREE.
497 		 */
498 		hmp->fchain.hmp = hmp;
499 		hmp->fchain.refs = 1;
500 		hmp->fchain.data = (void *)&hmp->voldata.freemap_blockset;
501 		hmp->fchain.bref.type = HAMMER2_BREF_TYPE_FREEMAP;
502 		hmp->fchain.bref.data_off = 0 | HAMMER2_PBUFRADIX;
503 		hmp->fchain.bref.methods =
504 			HAMMER2_ENC_CHECK(HAMMER2_CHECK_FREEMAP) |
505 			HAMMER2_ENC_COMP(HAMMER2_COMP_NONE);
506 		hmp->fchain.delete_tid = HAMMER2_MAX_TID;
507 
508 		hammer2_chain_core_alloc(NULL, &hmp->fchain, NULL);
509 		/* hmp->fchain.u.xxx is left NULL */
510 
511 		/*
512 		 * Install the volume header
513 		 */
514 		error = hammer2_install_volume_header(hmp);
515 		if (error) {
516 			hammer2_vfs_unmount(mp, MNT_FORCE);
517 			return error;
518 		}
519 
520 		/*
521 		 * First locate the super-root inode, which is key 0
522 		 * relative to the volume header's blockset.
523 		 *
524 		 * Then locate the root inode by scanning the directory keyspace
525 		 * represented by the label.
526 		 */
527 		parent = hammer2_chain_lookup_init(&hmp->vchain, 0);
528 		schain = hammer2_chain_lookup(&parent, &key_dummy,
529 				      HAMMER2_SROOT_KEY, HAMMER2_SROOT_KEY,
530 				      &cache_index, 0);
531 		hammer2_chain_lookup_done(parent);
532 		if (schain == NULL) {
533 			kprintf("hammer2_mount: invalid super-root\n");
534 			hammer2_vfs_unmount(mp, MNT_FORCE);
535 			return EINVAL;
536 		}
537 
538 		/*
539 		 * NOTE: inode_get sucks up schain's lock.
540 		 */
541 		atomic_set_int(&schain->flags, HAMMER2_CHAIN_PFSROOT);
542 		hmp->sroot = hammer2_inode_get(NULL, NULL, schain);
543 		hammer2_inode_ref(hmp->sroot);
544 		hammer2_inode_unlock_ex(hmp->sroot, schain);
545 		schain = NULL;
546 		/* leave hmp->sroot with one ref */
547 	}
548 
549 	/*
550 	 * Block device opened successfully, finish initializing the
551 	 * mount structure.
552 	 *
553 	 * From this point on we have to call hammer2_unmount() on failure.
554 	 */
555 	pmp = kmalloc(sizeof(*pmp), M_HAMMER2, M_WAITOK | M_ZERO);
556 
557 	kmalloc_create(&pmp->minode, "HAMMER2-inodes");
558 	kmalloc_create(&pmp->mmsg, "HAMMER2-pfsmsg");
559 
560 	spin_init(&pmp->inum_spin);
561 	RB_INIT(&pmp->inum_tree);
562 
563 	kdmsg_iocom_init(&pmp->iocom, pmp,
564 			 KDMSG_IOCOMF_AUTOCONN |
565 			 KDMSG_IOCOMF_AUTOSPAN |
566 			 KDMSG_IOCOMF_AUTOCIRC,
567 			 pmp->mmsg, hammer2_rcvdmsg);
568 
569 	ccms_domain_init(&pmp->ccms_dom);
570 	++hmp->pmp_count;
571 	lockmgr(&hammer2_mntlk, LK_RELEASE);
572 	kprintf("hammer2_mount hmp=%p pmp=%p pmpcnt=%d\n",
573 		hmp, pmp, hmp->pmp_count);
574 
575 	mp->mnt_flag = MNT_LOCAL;
576 	mp->mnt_kern_flag |= MNTK_ALL_MPSAFE;	/* all entry pts are SMP */
577 
578 	/*
579 	 * required mount structure initializations
580 	 */
581 	mp->mnt_stat.f_iosize = HAMMER2_PBUFSIZE;
582 	mp->mnt_stat.f_bsize = HAMMER2_PBUFSIZE;
583 
584 	mp->mnt_vstat.f_frsize = HAMMER2_PBUFSIZE;
585 	mp->mnt_vstat.f_bsize = HAMMER2_PBUFSIZE;
586 
587 	/*
588 	 * Optional fields
589 	 */
590 	mp->mnt_iosize_max = MAXPHYS;
591 	mp->mnt_data = (qaddr_t)pmp;
592 	pmp->mp = mp;
593 
594 	/*
595 	 * Lookup mount point under the media-localized super-root.
596 	 */
597 	parent = hammer2_inode_lock_ex(hmp->sroot);
598 	lhc = hammer2_dirhash(label, strlen(label));
599 	rchain = hammer2_chain_lookup(&parent, &key_next,
600 				      lhc, lhc + HAMMER2_DIRHASH_LOMASK,
601 				      &cache_index, 0);
602 	while (rchain) {
603 		if (rchain->bref.type == HAMMER2_BREF_TYPE_INODE &&
604 		    strcmp(label, rchain->data->ipdata.filename) == 0) {
605 			break;
606 		}
607 		rchain = hammer2_chain_next(&parent, rchain, &key_next,
608 					    key_next,
609 					    lhc + HAMMER2_DIRHASH_LOMASK,
610 					    &cache_index, 0);
611 	}
612 	hammer2_inode_unlock_ex(hmp->sroot, parent);
613 
614 	if (rchain == NULL) {
615 		kprintf("hammer2_mount: PFS label not found\n");
616 		--hmp->pmp_count;
617 		hammer2_vfs_unmount(mp, MNT_FORCE);
618 		return EINVAL;
619 	}
620 	if (rchain->flags & HAMMER2_CHAIN_MOUNTED) {
621 		hammer2_chain_unlock(rchain);
622 		kprintf("hammer2_mount: PFS label already mounted!\n");
623 		--hmp->pmp_count;
624 		hammer2_vfs_unmount(mp, MNT_FORCE);
625 		return EBUSY;
626 	}
627 #if 0
628 	if (rchain->flags & HAMMER2_CHAIN_RECYCLE) {
629 		kprintf("hammer2_mount: PFS label currently recycling\n");
630 		--hmp->pmp_count;
631 		hammer2_vfs_unmount(mp, MNT_FORCE);
632 		return EBUSY;
633 	}
634 #endif
635 
636 	atomic_set_int(&rchain->flags, HAMMER2_CHAIN_MOUNTED);
637 
638 	/*
639 	 * NOTE: *_get() integrates chain's lock into the inode lock.
640 	 */
641 	hammer2_chain_ref(rchain);		/* for pmp->rchain */
642 	pmp->cluster.nchains = 1;
643 	pmp->cluster.chains[0] = rchain;
644 	pmp->iroot = hammer2_inode_get(pmp, NULL, rchain);
645 	hammer2_inode_ref(pmp->iroot);		/* ref for pmp->iroot */
646 
647 	KKASSERT(rchain->pmp == NULL);		/* tracking pmp for rchain */
648 	rchain->pmp = pmp;
649 	atomic_add_long(&pmp->inmem_chains, 1);
650 
651 	hammer2_inode_unlock_ex(pmp->iroot, rchain);
652 
653 	kprintf("iroot %p\n", pmp->iroot);
654 
655 	/*
656 	 * The logical file buffer bio write thread handles things
657 	 * like physical block assignment and compression.
658 	 */
659 	mtx_init(&pmp->wthread_mtx);
660 	bioq_init(&pmp->wthread_bioq);
661 	pmp->wthread_destroy = 0;
662 	lwkt_create(hammer2_write_thread, pmp,
663 		    &pmp->wthread_td, NULL, 0, -1, "hwrite-%s", label);
664 
665 	/*
666 	 * Ref the cluster management messaging descriptor.  The mount
667 	 * program deals with the other end of the communications pipe.
668 	 */
669 	fp = holdfp(curproc->p_fd, info.cluster_fd, -1);
670 	if (fp == NULL) {
671 		kprintf("hammer2_mount: bad cluster_fd!\n");
672 		hammer2_vfs_unmount(mp, MNT_FORCE);
673 		return EBADF;
674 	}
675 	hammer2_cluster_reconnect(pmp, fp);
676 
677 	/*
678 	 * Finish setup
679 	 */
680 	vfs_getnewfsid(mp);
681 	vfs_add_vnodeops(mp, &hammer2_vnode_vops, &mp->mnt_vn_norm_ops);
682 	vfs_add_vnodeops(mp, &hammer2_spec_vops, &mp->mnt_vn_spec_ops);
683 	vfs_add_vnodeops(mp, &hammer2_fifo_vops, &mp->mnt_vn_fifo_ops);
684 
685 	copyinstr(info.volume, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, &size);
686 	bzero(mp->mnt_stat.f_mntfromname + size, MNAMELEN - size);
687 	bzero(mp->mnt_stat.f_mntonname, sizeof(mp->mnt_stat.f_mntonname));
688 	copyinstr(path, mp->mnt_stat.f_mntonname,
689 		  sizeof(mp->mnt_stat.f_mntonname) - 1,
690 		  &size);
691 
692 	/*
693 	 * Initial statfs to prime mnt_stat.
694 	 */
695 	hammer2_vfs_statfs(mp, &mp->mnt_stat, cred);
696 
697 	return 0;
698 }
699 
700 /*
701  * Handle bioq for strategy write
702  */
703 static
704 void
705 hammer2_write_thread(void *arg)
706 {
707 	hammer2_pfsmount_t *pmp;
708 	struct bio *bio;
709 	struct buf *bp;
710 	hammer2_trans_t trans;
711 	struct vnode *vp;
712 	hammer2_inode_t *last_ip;
713 	hammer2_inode_t *ip;
714 	hammer2_chain_t *parent;
715 	hammer2_chain_t **parentp;
716 	hammer2_inode_data_t *ipdata;
717 	hammer2_key_t lbase;
718 	int lblksize;
719 	int pblksize;
720 	int error;
721 
722 	pmp = arg;
723 
724 	mtx_lock(&pmp->wthread_mtx);
725 	while (pmp->wthread_destroy == 0) {
726 		if (bioq_first(&pmp->wthread_bioq) == NULL) {
727 			mtxsleep(&pmp->wthread_bioq, &pmp->wthread_mtx,
728 				 0, "h2bioqw", 0);
729 		}
730 		last_ip = NULL;
731 		parent = NULL;
732 		parentp = &parent;
733 
734 		while ((bio = bioq_takefirst(&pmp->wthread_bioq)) != NULL) {
735 			mtx_unlock(&pmp->wthread_mtx);
736 
737 			error = 0;
738 			bp = bio->bio_buf;
739 			vp = bp->b_vp;
740 			ip = VTOI(vp);
741 
742 			/*
743 			 * Cache transaction for multi-buffer flush efficiency.
744 			 * Lock the ip separately for each buffer to allow
745 			 * interleaving with frontend writes.
746 			 */
747 			if (last_ip != ip) {
748 				if (last_ip)
749 					hammer2_trans_done(&trans);
750 				hammer2_trans_init(&trans, ip->pmp,
751 						   HAMMER2_TRANS_BUFCACHE);
752 				last_ip = ip;
753 			}
754 			parent = hammer2_inode_lock_ex(ip);
755 
756 			/*
757 			 * Inode is modified, flush size and mtime changes
758 			 * to ensure that the file size remains consistent
759 			 * with the buffers being flushed.
760 			 */
761 			if (ip->flags & (HAMMER2_INODE_RESIZED |
762 					 HAMMER2_INODE_MTIME)) {
763 				hammer2_inode_fsync(&trans, ip, parentp);
764 			}
765 			ipdata = hammer2_chain_modify_ip(&trans, ip,
766 							 parentp, 0);
767 			lblksize = hammer2_calc_logical(ip, bio->bio_offset,
768 							&lbase, NULL);
769 			pblksize = hammer2_calc_physical(ip, lbase);
770 			hammer2_write_file_core(bp, &trans, ip, ipdata,
771 						parentp,
772 						lbase, IO_ASYNC,
773 						pblksize, &error);
774 			hammer2_inode_unlock_ex(ip, parent);
775 			if (error) {
776 				kprintf("hammer2: error in buffer write\n");
777 				bp->b_flags |= B_ERROR;
778 				bp->b_error = EIO;
779 			}
780 			biodone(bio);
781 			mtx_lock(&pmp->wthread_mtx);
782 		}
783 
784 		/*
785 		 * Clean out transaction cache
786 		 */
787 		if (last_ip)
788 			hammer2_trans_done(&trans);
789 	}
790 	pmp->wthread_destroy = -1;
791 	wakeup(&pmp->wthread_destroy);
792 
793 	mtx_unlock(&pmp->wthread_mtx);
794 }
795 
796 /*
797  * Return a chain suitable for I/O, creating the chain if necessary
798  * and assigning its physical block.
799  */
800 static
801 hammer2_chain_t *
802 hammer2_assign_physical(hammer2_trans_t *trans,
803 			hammer2_inode_t *ip, hammer2_chain_t **parentp,
804 			hammer2_key_t lbase, int pblksize, int *errorp)
805 {
806 	hammer2_chain_t *parent;
807 	hammer2_chain_t *chain;
808 	hammer2_off_t pbase;
809 	hammer2_key_t key_dummy;
810 	int pradix = hammer2_getradix(pblksize);
811 	int cache_index = -1;
812 
813 	/*
814 	 * Locate the chain associated with lbase, return a locked chain.
815 	 * However, do not instantiate any data reference (which utilizes a
816 	 * device buffer) because we will be using direct IO via the
817 	 * logical buffer cache buffer.
818 	 */
819 	*errorp = 0;
820 	KKASSERT(pblksize >= HAMMER2_MIN_ALLOC);
821 retry:
822 	parent = *parentp;
823 	hammer2_chain_lock(parent, HAMMER2_RESOLVE_ALWAYS); /* extra lock */
824 	chain = hammer2_chain_lookup(&parent, &key_dummy,
825 				     lbase, lbase,
826 				     &cache_index, HAMMER2_LOOKUP_NODATA);
827 
828 	if (chain == NULL) {
829 		/*
830 		 * We found a hole, create a new chain entry.
831 		 *
832 		 * NOTE: DATA chains are created without device backing
833 		 *	 store (nor do we want any).
834 		 */
835 		*errorp = hammer2_chain_create(trans, &parent, &chain,
836 					       lbase, HAMMER2_PBUFRADIX,
837 					       HAMMER2_BREF_TYPE_DATA,
838 					       pblksize);
839 		if (chain == NULL) {
840 			hammer2_chain_lookup_done(parent);
841 			panic("hammer2_chain_create: par=%p error=%d\n",
842 				parent, *errorp);
843 			goto retry;
844 		}
845 
846 		pbase = chain->bref.data_off & ~HAMMER2_OFF_MASK_RADIX;
847 		/*ip->delta_dcount += pblksize;*/
848 	} else {
849 		switch (chain->bref.type) {
850 		case HAMMER2_BREF_TYPE_INODE:
851 			/*
852 			 * The data is embedded in the inode.  The
853 			 * caller is responsible for marking the inode
854 			 * modified and copying the data to the embedded
855 			 * area.
856 			 */
857 			pbase = NOOFFSET;
858 			break;
859 		case HAMMER2_BREF_TYPE_DATA:
860 			if (chain->bytes != pblksize) {
861 				hammer2_chain_resize(trans, ip,
862 						     parent, &chain,
863 						     pradix,
864 						     HAMMER2_MODIFY_OPTDATA);
865 			}
866 			hammer2_chain_modify(trans, &chain,
867 					     HAMMER2_MODIFY_OPTDATA);
868 			pbase = chain->bref.data_off & ~HAMMER2_OFF_MASK_RADIX;
869 			break;
870 		default:
871 			panic("hammer2_assign_physical: bad type");
872 			/* NOT REACHED */
873 			pbase = NOOFFSET;
874 			break;
875 		}
876 	}
877 
878 	/*
879 	 * Cleanup.  If chain wound up being the inode (i.e. DIRECTDATA),
880 	 * we might have to replace *parentp.
881 	 */
882 	hammer2_chain_lookup_done(parent);
883 	if (chain) {
884 		if (*parentp != chain &&
885 		    (*parentp)->core == chain->core) {
886 			parent = *parentp;
887 			*parentp = chain;		/* eats lock */
888 			hammer2_chain_unlock(parent);
889 			hammer2_chain_lock(chain, 0);	/* need another */
890 		}
891 		/* else chain already locked for return */
892 	}
893 	return (chain);
894 }
895 
896 /*
897  * From hammer2_vnops.c.
898  * The core write function which determines which path to take
899  * depending on compression settings.
900  */
901 static
902 void
903 hammer2_write_file_core(struct buf *bp, hammer2_trans_t *trans,
904 			hammer2_inode_t *ip, hammer2_inode_data_t *ipdata,
905 			hammer2_chain_t **parentp,
906 			hammer2_key_t lbase, int ioflag, int pblksize,
907 			int *errorp)
908 {
909 	hammer2_chain_t *chain;
910 
911 	switch(HAMMER2_DEC_COMP(ipdata->comp_algo)) {
912 	case HAMMER2_COMP_NONE:
913 		/*
914 		 * We have to assign physical storage to the buffer
915 		 * we intend to dirty or write now to avoid deadlocks
916 		 * in the strategy code later.
917 		 *
918 		 * This can return NOOFFSET for inode-embedded data.
919 		 * The strategy code will take care of it in that case.
920 		 */
921 		chain = hammer2_assign_physical(trans, ip, parentp,
922 						lbase, pblksize,
923 						errorp);
924 		hammer2_write_bp(chain, bp, ioflag, pblksize, errorp);
925 		if (chain)
926 			hammer2_chain_unlock(chain);
927 		break;
928 	case HAMMER2_COMP_AUTOZERO:
929 		/*
930 		 * Check for zero-fill only
931 		 */
932 		hammer2_zero_check_and_write(bp, trans, ip,
933 				    ipdata, parentp, lbase,
934 				    ioflag, pblksize, errorp);
935 		break;
936 	case HAMMER2_COMP_LZ4:
937 	case HAMMER2_COMP_ZLIB:
938 	default:
939 		/*
940 		 * Check for zero-fill and attempt compression.
941 		 */
942 		hammer2_compress_and_write(bp, trans, ip,
943 					   ipdata, parentp,
944 					   lbase, ioflag,
945 					   pblksize, errorp,
946 					   ipdata->comp_algo);
947 		break;
948 	}
949 	ipdata = &ip->chain->data->ipdata;	/* reload */
950 }
951 
952 /*
953  * From hammer2_vnops.c
954  * Generic function that will perform the compression in compression
955  * write path. The compression algorithm is determined by the settings
956  * obtained from inode.
957  */
958 static
959 void
960 hammer2_compress_and_write(struct buf *bp, hammer2_trans_t *trans,
961 	hammer2_inode_t *ip, hammer2_inode_data_t *ipdata,
962 	hammer2_chain_t **parentp,
963 	hammer2_key_t lbase, int ioflag, int pblksize,
964 	int *errorp, int comp_algo)
965 {
966 	hammer2_chain_t *chain;
967 	int comp_size;
968 	int comp_block_size;
969 	char *comp_buffer;
970 
971 	if (test_block_zeros(bp->b_data, pblksize)) {
972 		zero_write(bp, trans, ip, ipdata, parentp, lbase, errorp);
973 		return;
974 	}
975 
976 	comp_size = 0;
977 	comp_buffer = NULL;
978 
979 	KKASSERT(pblksize / 2 <= 32768);
980 
981 	if (ip->comp_heuristic < 8 || (ip->comp_heuristic & 7) == 0) {
982 		z_stream strm_compress;
983 		int comp_level;
984 		int ret;
985 
986 		switch(HAMMER2_DEC_COMP(comp_algo)) {
987 		case HAMMER2_COMP_LZ4:
988 			comp_buffer = objcache_get(cache_buffer_write,
989 						   M_INTWAIT);
990 			comp_size = LZ4_compress_limitedOutput(
991 					bp->b_data,
992 					&comp_buffer[sizeof(int)],
993 					pblksize,
994 					pblksize / 2 - sizeof(int));
995 			/*
996 			 * We need to prefix with the size, LZ4
997 			 * doesn't do it for us.  Add the related
998 			 * overhead.
999 			 */
1000 			*(int *)comp_buffer = comp_size;
1001 			if (comp_size)
1002 				comp_size += sizeof(int);
1003 			break;
1004 		case HAMMER2_COMP_ZLIB:
1005 			comp_level = HAMMER2_DEC_LEVEL(comp_algo);
1006 			if (comp_level == 0)
1007 				comp_level = 6;	/* default zlib compression */
1008 			else if (comp_level < 6)
1009 				comp_level = 6;
1010 			else if (comp_level > 9)
1011 				comp_level = 9;
1012 			ret = deflateInit(&strm_compress, comp_level);
1013 			if (ret != Z_OK) {
1014 				kprintf("HAMMER2 ZLIB: fatal error "
1015 					"on deflateInit.\n");
1016 			}
1017 
1018 			comp_buffer = objcache_get(cache_buffer_write,
1019 						   M_INTWAIT);
1020 			strm_compress.next_in = bp->b_data;
1021 			strm_compress.avail_in = pblksize;
1022 			strm_compress.next_out = comp_buffer;
1023 			strm_compress.avail_out = pblksize / 2;
1024 			ret = deflate(&strm_compress, Z_FINISH);
1025 			if (ret == Z_STREAM_END) {
1026 				comp_size = pblksize / 2 -
1027 					    strm_compress.avail_out;
1028 			} else {
1029 				comp_size = 0;
1030 			}
1031 			ret = deflateEnd(&strm_compress);
1032 			break;
1033 		default:
1034 			kprintf("Error: Unknown compression method.\n");
1035 			kprintf("Comp_method = %d.\n", comp_algo);
1036 			break;
1037 		}
1038 	}
1039 
1040 	if (comp_size == 0) {
1041 		/*
1042 		 * compression failed or turned off
1043 		 */
1044 		comp_block_size = pblksize;	/* safety */
1045 		if (++ip->comp_heuristic > 128)
1046 			ip->comp_heuristic = 8;
1047 	} else {
1048 		/*
1049 		 * compression succeeded
1050 		 */
1051 		ip->comp_heuristic = 0;
1052 		if (comp_size <= 1024) {
1053 			comp_block_size = 1024;
1054 		} else if (comp_size <= 2048) {
1055 			comp_block_size = 2048;
1056 		} else if (comp_size <= 4096) {
1057 			comp_block_size = 4096;
1058 		} else if (comp_size <= 8192) {
1059 			comp_block_size = 8192;
1060 		} else if (comp_size <= 16384) {
1061 			comp_block_size = 16384;
1062 		} else if (comp_size <= 32768) {
1063 			comp_block_size = 32768;
1064 		} else {
1065 			panic("hammer2: WRITE PATH: "
1066 			      "Weird comp_size value.");
1067 			/* NOT REACHED */
1068 			comp_block_size = pblksize;
1069 		}
1070 	}
1071 
1072 	chain = hammer2_assign_physical(trans, ip, parentp,
1073 					lbase, comp_block_size,
1074 					errorp);
1075 	ipdata = &ip->chain->data->ipdata;	/* RELOAD */
1076 
1077 	if (*errorp) {
1078 		kprintf("WRITE PATH: An error occurred while "
1079 			"assigning physical space.\n");
1080 		KKASSERT(chain == NULL);
1081 	} else {
1082 		/* Get device offset */
1083 		hammer2_off_t pbase;
1084 		hammer2_off_t pmask;
1085 		hammer2_off_t peof;
1086 		size_t boff;
1087 		size_t psize;
1088 		struct buf *dbp;
1089 		int temp_check;
1090 
1091 		KKASSERT(chain->flags & HAMMER2_CHAIN_MODIFIED);
1092 
1093 		switch(chain->bref.type) {
1094 		case HAMMER2_BREF_TYPE_INODE:
1095 			KKASSERT(chain->data->ipdata.op_flags &
1096 				 HAMMER2_OPFLAG_DIRECTDATA);
1097 			KKASSERT(bp->b_loffset == 0);
1098 			bcopy(bp->b_data, chain->data->ipdata.u.data,
1099 			      HAMMER2_EMBEDDED_BYTES);
1100 			break;
1101 		case HAMMER2_BREF_TYPE_DATA:
1102 			psize = hammer2_devblksize(chain->bytes);
1103 			pmask = (hammer2_off_t)psize - 1;
1104 			pbase = chain->bref.data_off & ~pmask;
1105 			boff = chain->bref.data_off &
1106 			       (HAMMER2_OFF_MASK & pmask);
1107 			peof = (pbase + HAMMER2_SEGMASK64) &
1108 			       ~HAMMER2_SEGMASK64;
1109 			temp_check = HAMMER2_DEC_CHECK(chain->bref.methods);
1110 
1111 			/*
1112 			 * Optimize out the read-before-write
1113 			 * if possible.
1114 			 */
1115 			if (comp_block_size == psize) {
1116 				dbp = getblk(chain->hmp->devvp, pbase,
1117 					     psize, 0, 0);
1118 			} else {
1119 				*errorp = bread(chain->hmp->devvp,
1120 						pbase, psize, &dbp);
1121 				if (*errorp) {
1122 					kprintf("hammer2: WRITE PATH: "
1123 						"dbp bread error\n");
1124 					break;
1125 				}
1126 			}
1127 
1128 			/*
1129 			 * When loading the block make sure we don't
1130 			 * leave garbage after the compressed data.
1131 			 */
1132 			if (comp_size) {
1133 				chain->bref.methods =
1134 					HAMMER2_ENC_COMP(comp_algo) +
1135 					HAMMER2_ENC_CHECK(temp_check);
1136 				bcopy(comp_buffer, dbp->b_data + boff,
1137 				      comp_size);
1138 				if (comp_size != comp_block_size) {
1139 					bzero(dbp->b_data + boff +
1140 						comp_size,
1141 					      comp_block_size -
1142 						comp_size);
1143 				}
1144 			} else {
1145 				chain->bref.methods =
1146 					HAMMER2_ENC_COMP(
1147 						HAMMER2_COMP_NONE) +
1148 					HAMMER2_ENC_CHECK(temp_check);
1149 				bcopy(bp->b_data, dbp->b_data + boff,
1150 				      pblksize);
1151 			}
1152 
1153 			/*
1154 			 * Device buffer is now valid, chain is no
1155 			 * longer in the initial state.
1156 			 */
1157 			atomic_clear_int(&chain->flags,
1158 					 HAMMER2_CHAIN_INITIAL);
1159 
1160 			/* Now write the related bdp. */
1161 			if (ioflag & IO_SYNC) {
1162 				/*
1163 				 * Synchronous I/O requested.
1164 				 */
1165 				bwrite(dbp);
1166 			/*
1167 			} else if ((ioflag & IO_DIRECT) &&
1168 				   loff + n == pblksize) {
1169 				bdwrite(dbp);
1170 			*/
1171 			} else if (ioflag & IO_ASYNC) {
1172 				bawrite(dbp);
1173 			} else if (hammer2_cluster_enable) {
1174 				cluster_write(dbp, peof,
1175 					      HAMMER2_PBUFSIZE,
1176 					      4/*XXX*/);
1177 			} else {
1178 				bdwrite(dbp);
1179 			}
1180 			break;
1181 		default:
1182 			panic("hammer2_write_bp: bad chain type %d\n",
1183 				chain->bref.type);
1184 			/* NOT REACHED */
1185 			break;
1186 		}
1187 
1188 		hammer2_chain_unlock(chain);
1189 	}
1190 	if (comp_buffer)
1191 		objcache_put(cache_buffer_write, comp_buffer);
1192 }
1193 
1194 /*
1195  * Function that performs zero-checking and writing without compression,
1196  * it corresponds to default zero-checking path.
1197  */
1198 static
1199 void
1200 hammer2_zero_check_and_write(struct buf *bp, hammer2_trans_t *trans,
1201 	hammer2_inode_t *ip, hammer2_inode_data_t *ipdata,
1202 	hammer2_chain_t **parentp,
1203 	hammer2_key_t lbase, int ioflag, int pblksize, int *errorp)
1204 {
1205 	hammer2_chain_t *chain;
1206 
1207 	if (test_block_zeros(bp->b_data, pblksize)) {
1208 		zero_write(bp, trans, ip, ipdata, parentp, lbase, errorp);
1209 	} else {
1210 		chain = hammer2_assign_physical(trans, ip, parentp,
1211 						lbase, pblksize, errorp);
1212 		hammer2_write_bp(chain, bp, ioflag, pblksize, errorp);
1213 		if (chain)
1214 			hammer2_chain_unlock(chain);
1215 	}
1216 }
1217 
1218 /*
1219  * A function to test whether a block of data contains only zeros,
1220  * returns TRUE (non-zero) if the block is all zeros.
1221  */
1222 static
1223 int
1224 test_block_zeros(const char *buf, size_t bytes)
1225 {
1226 	size_t i;
1227 
1228 	for (i = 0; i < bytes; i += sizeof(long)) {
1229 		if (*(const long *)(buf + i) != 0)
1230 			return (0);
1231 	}
1232 	return (1);
1233 }
1234 
1235 /*
1236  * Function to "write" a block that contains only zeros.
1237  */
1238 static
1239 void
1240 zero_write(struct buf *bp, hammer2_trans_t *trans, hammer2_inode_t *ip,
1241 	hammer2_inode_data_t *ipdata, hammer2_chain_t **parentp,
1242 	hammer2_key_t lbase, int *errorp __unused)
1243 {
1244 	hammer2_chain_t *parent;
1245 	hammer2_chain_t *chain;
1246 	hammer2_key_t key_dummy;
1247 	int cache_index = -1;
1248 
1249 	parent = hammer2_chain_lookup_init(*parentp, 0);
1250 
1251 	chain = hammer2_chain_lookup(&parent, &key_dummy, lbase, lbase,
1252 				     &cache_index, HAMMER2_LOOKUP_NODATA);
1253 	if (chain) {
1254 		if (chain->bref.type == HAMMER2_BREF_TYPE_INODE) {
1255 			bzero(chain->data->ipdata.u.data,
1256 			      HAMMER2_EMBEDDED_BYTES);
1257 		} else {
1258 			hammer2_chain_delete(trans, chain, 0);
1259 		}
1260 		hammer2_chain_unlock(chain);
1261 	}
1262 	hammer2_chain_lookup_done(parent);
1263 }
1264 
1265 /*
1266  * Function to write the data as it is, without performing any sort of
1267  * compression. This function is used in path without compression and
1268  * default zero-checking path.
1269  */
1270 static
1271 void
1272 hammer2_write_bp(hammer2_chain_t *chain, struct buf *bp, int ioflag,
1273 				int pblksize, int *errorp)
1274 {
1275 	hammer2_off_t pbase;
1276 	hammer2_off_t pmask;
1277 	hammer2_off_t peof;
1278 	struct buf *dbp;
1279 	size_t boff;
1280 	size_t psize;
1281 	int error;
1282 	int temp_check = HAMMER2_DEC_CHECK(chain->bref.methods);
1283 
1284 	KKASSERT(chain->flags & HAMMER2_CHAIN_MODIFIED);
1285 
1286 	switch(chain->bref.type) {
1287 	case HAMMER2_BREF_TYPE_INODE:
1288 		KKASSERT(chain->data->ipdata.op_flags &
1289 			 HAMMER2_OPFLAG_DIRECTDATA);
1290 		KKASSERT(bp->b_loffset == 0);
1291 		bcopy(bp->b_data, chain->data->ipdata.u.data,
1292 		      HAMMER2_EMBEDDED_BYTES);
1293 		error = 0;
1294 		break;
1295 	case HAMMER2_BREF_TYPE_DATA:
1296 		psize = hammer2_devblksize(chain->bytes);
1297 		pmask = (hammer2_off_t)psize - 1;
1298 		pbase = chain->bref.data_off & ~pmask;
1299 		boff = chain->bref.data_off & (HAMMER2_OFF_MASK & pmask);
1300 		peof = (pbase + HAMMER2_SEGMASK64) & ~HAMMER2_SEGMASK64;
1301 
1302 		if (psize == pblksize) {
1303 			dbp = getblk(chain->hmp->devvp, pbase,
1304 				     psize, 0, 0);
1305 			error = 0;
1306 		} else {
1307 			error = bread(chain->hmp->devvp, pbase, psize, &dbp);
1308 			if (error) {
1309 				kprintf("hammer2: WRITE PATH: "
1310 					"dbp bread error\n");
1311 				break;
1312 			}
1313 		}
1314 
1315 		chain->bref.methods = HAMMER2_ENC_COMP(HAMMER2_COMP_NONE) +
1316 				      HAMMER2_ENC_CHECK(temp_check);
1317 		bcopy(bp->b_data, dbp->b_data + boff, chain->bytes);
1318 
1319 		/*
1320 		 * Device buffer is now valid, chain is no
1321 		 * longer in the initial state.
1322 		 */
1323 		atomic_clear_int(&chain->flags, HAMMER2_CHAIN_INITIAL);
1324 
1325 		if (ioflag & IO_SYNC) {
1326 			/*
1327 			 * Synchronous I/O requested.
1328 			 */
1329 			bwrite(dbp);
1330 		/*
1331 		} else if ((ioflag & IO_DIRECT) && loff + n == pblksize) {
1332 			bdwrite(dbp);
1333 		*/
1334 		} else if (ioflag & IO_ASYNC) {
1335 			bawrite(dbp);
1336 		} else if (hammer2_cluster_enable) {
1337 			cluster_write(dbp, peof, HAMMER2_PBUFSIZE, 4/*XXX*/);
1338 		} else {
1339 			bdwrite(dbp);
1340 		}
1341 		break;
1342 	default:
1343 		panic("hammer2_write_bp: bad chain type %d\n",
1344 		      chain->bref.type);
1345 		/* NOT REACHED */
1346 		error = 0;
1347 		break;
1348 	}
1349 	*errorp = error;
1350 }
1351 
1352 static
1353 int
1354 hammer2_remount(hammer2_mount_t *hmp, char *path, struct vnode *devvp,
1355                 struct ucred *cred)
1356 {
1357 	return (0);
1358 }
1359 
1360 static
1361 int
1362 hammer2_vfs_unmount(struct mount *mp, int mntflags)
1363 {
1364 	hammer2_pfsmount_t *pmp;
1365 	hammer2_mount_t *hmp;
1366 	hammer2_chain_t *rchain;
1367 	int flags;
1368 	int error = 0;
1369 	int ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
1370 	int dumpcnt;
1371 	int i;
1372 	struct vnode *devvp;
1373 
1374 	pmp = MPTOPMP(mp);
1375 
1376 	ccms_domain_uninit(&pmp->ccms_dom);
1377 	kdmsg_iocom_uninit(&pmp->iocom);	/* XXX chain dependency */
1378 
1379 	lockmgr(&hammer2_mntlk, LK_EXCLUSIVE);
1380 
1381 	/*
1382 	 * If mount initialization proceeded far enough we must flush
1383 	 * its vnodes.
1384 	 */
1385 	if (mntflags & MNT_FORCE)
1386 		flags = FORCECLOSE;
1387 	else
1388 		flags = 0;
1389 	if (pmp->iroot) {
1390 		error = vflush(mp, 0, flags);
1391 		if (error)
1392 			goto failed;
1393 	}
1394 
1395 	if (pmp->wthread_td) {
1396 		mtx_lock(&pmp->wthread_mtx);
1397 		pmp->wthread_destroy = 1;
1398 		wakeup(&pmp->wthread_bioq);
1399 		while (pmp->wthread_destroy != -1) {
1400 			mtxsleep(&pmp->wthread_destroy,
1401 				&pmp->wthread_mtx, 0,
1402 				"umount-sleep",	0);
1403 		}
1404 		mtx_unlock(&pmp->wthread_mtx);
1405 		pmp->wthread_td = NULL;
1406 	}
1407 
1408 	for (i = 0; i < pmp->cluster.nchains; ++i) {
1409 		hmp = pmp->cluster.chains[i]->hmp;
1410 
1411 		hammer2_mount_exlock(hmp);
1412 
1413 		--hmp->pmp_count;
1414 		kprintf("hammer2_unmount hmp=%p pmpcnt=%d\n",
1415 			hmp, hmp->pmp_count);
1416 
1417 		/*
1418 		 * Flush any left over chains.  The voldata lock is only used
1419 		 * to synchronize against HAMMER2_CHAIN_MODIFIED_AUX.
1420 		 */
1421 		hammer2_voldata_lock(hmp);
1422 		if ((hmp->vchain.flags | hmp->fchain.flags) &
1423 		    (HAMMER2_CHAIN_MODIFIED | HAMMER2_CHAIN_SUBMODIFIED)) {
1424 			hammer2_voldata_unlock(hmp, 0);
1425 			hammer2_vfs_sync(mp, MNT_WAIT);
1426 			hammer2_vfs_sync(mp, MNT_WAIT);
1427 		} else {
1428 			hammer2_voldata_unlock(hmp, 0);
1429 		}
1430 		if (hmp->pmp_count == 0) {
1431 			if (hmp->vchain.flags & (HAMMER2_CHAIN_MODIFIED |
1432 						 HAMMER2_CHAIN_SUBMODIFIED)) {
1433 				kprintf("hammer2_unmount: chains left over "
1434 					"after final sync\n");
1435 				if (hammer2_debug & 0x0010)
1436 					Debugger("entered debugger");
1437 			}
1438 		}
1439 
1440 		/*
1441 		 * Cleanup the root and super-root chain elements
1442 		 * (which should be clean).
1443 		 */
1444 		if (pmp->iroot) {
1445 #if REPORT_REFS_ERRORS
1446 			if (pmp->iroot->refs != 1)
1447 				kprintf("PMP->IROOT %p REFS WRONG %d\n",
1448 					pmp->iroot, pmp->iroot->refs);
1449 #else
1450 			KKASSERT(pmp->iroot->refs == 1);
1451 #endif
1452 			/* ref for pmp->iroot */
1453 			hammer2_inode_drop(pmp->iroot);
1454 			pmp->iroot = NULL;
1455 		}
1456 
1457 		rchain = pmp->cluster.chains[i];
1458 		if (rchain) {
1459 			atomic_clear_int(&rchain->flags, HAMMER2_CHAIN_MOUNTED);
1460 #if REPORT_REFS_ERRORS
1461 			if (rchain->refs != 1)
1462 				kprintf("PMP->RCHAIN %p REFS WRONG %d\n",
1463 					rchain, rchain->refs);
1464 #else
1465 			KKASSERT(rchain->refs == 1);
1466 #endif
1467 			hammer2_chain_drop(rchain);
1468 			pmp->cluster.chains[i] = NULL;
1469 		}
1470 
1471 		/*
1472 		 * If no PFS's left drop the master hammer2_mount for the
1473 		 * device.
1474 		 */
1475 		if (hmp->pmp_count == 0) {
1476 			if (hmp->sroot) {
1477 				hammer2_inode_drop(hmp->sroot);
1478 				hmp->sroot = NULL;
1479 			}
1480 
1481 			/*
1482 			 * Finish up with the device vnode
1483 			 */
1484 			if ((devvp = hmp->devvp) != NULL) {
1485 				vinvalbuf(devvp, (ronly ? 0 : V_SAVE), 0, 0);
1486 				hmp->devvp = NULL;
1487 				VOP_CLOSE(devvp,
1488 					  (ronly ? FREAD : FREAD|FWRITE));
1489 				vrele(devvp);
1490 				devvp = NULL;
1491 			}
1492 
1493 			/*
1494 			 * Final drop of embedded freemap root chain to
1495 			 * clean up fchain.core (fchain structure is not
1496 			 * flagged ALLOCATED so it is cleaned out and then
1497 			 * left to rot).
1498 			 */
1499 			hammer2_chain_drop(&hmp->fchain);
1500 
1501 			/*
1502 			 * Final drop of embedded volume root chain to clean
1503 			 * up vchain.core (vchain structure is not flagged
1504 			 * ALLOCATED so it is cleaned out and then left to
1505 			 * rot).
1506 			 */
1507 			dumpcnt = 50;
1508 			hammer2_dump_chain(&hmp->vchain, 0, &dumpcnt);
1509 			hammer2_mount_unlock(hmp);
1510 			hammer2_chain_drop(&hmp->vchain);
1511 
1512 			TAILQ_REMOVE(&hammer2_mntlist, hmp, mntentry);
1513 			kmalloc_destroy(&hmp->mchain);
1514 			kfree(hmp, M_HAMMER2);
1515 		} else {
1516 			hammer2_mount_unlock(hmp);
1517 		}
1518 	}
1519 
1520 	pmp->mp = NULL;
1521 	mp->mnt_data = NULL;
1522 
1523 	kmalloc_destroy(&pmp->mmsg);
1524 	kmalloc_destroy(&pmp->minode);
1525 
1526 	kfree(pmp, M_HAMMER2);
1527 	error = 0;
1528 
1529 failed:
1530 	lockmgr(&hammer2_mntlk, LK_RELEASE);
1531 
1532 	return (error);
1533 }
1534 
1535 static
1536 int
1537 hammer2_vfs_vget(struct mount *mp, struct vnode *dvp,
1538 	     ino_t ino, struct vnode **vpp)
1539 {
1540 	kprintf("hammer2_vget\n");
1541 	return (EOPNOTSUPP);
1542 }
1543 
1544 static
1545 int
1546 hammer2_vfs_root(struct mount *mp, struct vnode **vpp)
1547 {
1548 	hammer2_pfsmount_t *pmp;
1549 	hammer2_chain_t *parent;
1550 	int error;
1551 	struct vnode *vp;
1552 
1553 	pmp = MPTOPMP(mp);
1554 	if (pmp->iroot == NULL) {
1555 		*vpp = NULL;
1556 		error = EINVAL;
1557 	} else {
1558 		parent = hammer2_inode_lock_sh(pmp->iroot);
1559 		vp = hammer2_igetv(pmp->iroot, &error);
1560 		hammer2_inode_unlock_sh(pmp->iroot, parent);
1561 		*vpp = vp;
1562 		if (vp == NULL)
1563 			kprintf("vnodefail\n");
1564 	}
1565 
1566 	return (error);
1567 }
1568 
1569 /*
1570  * Filesystem status
1571  *
1572  * XXX incorporate ipdata->inode_quota and data_quota
1573  */
1574 static
1575 int
1576 hammer2_vfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
1577 {
1578 	hammer2_pfsmount_t *pmp;
1579 	hammer2_mount_t *hmp;
1580 
1581 	pmp = MPTOPMP(mp);
1582 	KKASSERT(pmp->cluster.nchains >= 1);
1583 	hmp = pmp->cluster.chains[0]->hmp;	/* XXX */
1584 
1585 	mp->mnt_stat.f_files = pmp->inode_count;
1586 	mp->mnt_stat.f_ffree = 0;
1587 	mp->mnt_stat.f_blocks = hmp->voldata.allocator_size / HAMMER2_PBUFSIZE;
1588 	mp->mnt_stat.f_bfree =  hmp->voldata.allocator_free / HAMMER2_PBUFSIZE;
1589 	mp->mnt_stat.f_bavail = mp->mnt_stat.f_bfree;
1590 
1591 	*sbp = mp->mnt_stat;
1592 	return (0);
1593 }
1594 
1595 static
1596 int
1597 hammer2_vfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred)
1598 {
1599 	hammer2_pfsmount_t *pmp;
1600 	hammer2_mount_t *hmp;
1601 
1602 	pmp = MPTOPMP(mp);
1603 	KKASSERT(pmp->cluster.nchains >= 1);
1604 	hmp = pmp->cluster.chains[0]->hmp;	/* XXX */
1605 
1606 	mp->mnt_vstat.f_bsize = HAMMER2_PBUFSIZE;
1607 	mp->mnt_vstat.f_files = pmp->inode_count;
1608 	mp->mnt_vstat.f_ffree = 0;
1609 	mp->mnt_vstat.f_blocks = hmp->voldata.allocator_size / HAMMER2_PBUFSIZE;
1610 	mp->mnt_vstat.f_bfree =  hmp->voldata.allocator_free / HAMMER2_PBUFSIZE;
1611 	mp->mnt_vstat.f_bavail = mp->mnt_vstat.f_bfree;
1612 
1613 	*sbp = mp->mnt_vstat;
1614 	return (0);
1615 }
1616 
1617 /*
1618  * Sync the entire filesystem; this is called from the filesystem syncer
1619  * process periodically and whenever a user calls sync(1) on the hammer
1620  * mountpoint.
1621  *
1622  * Currently is actually called from the syncer! \o/
1623  *
1624  * This task will have to snapshot the state of the dirty inode chain.
1625  * From that, it will have to make sure all of the inodes on the dirty
1626  * chain have IO initiated. We make sure that io is initiated for the root
1627  * block.
1628  *
1629  * If waitfor is set, we wait for media to acknowledge the new rootblock.
1630  *
1631  * THINKS: side A vs side B, to have sync not stall all I/O?
1632  */
1633 static
1634 int
1635 hammer2_vfs_sync(struct mount *mp, int waitfor)
1636 {
1637 	struct hammer2_sync_info info;
1638 	hammer2_pfsmount_t *pmp;
1639 	hammer2_mount_t *hmp;
1640 	int flags;
1641 	int error;
1642 	int total_error;
1643 	int i;
1644 
1645 	pmp = MPTOPMP(mp);
1646 
1647 	/*
1648 	 * We can't acquire locks on existing vnodes while in a transaction
1649 	 * without risking a deadlock.  This assumes that vfsync() can be
1650 	 * called without the vnode locked (which it can in DragonFly).
1651 	 * Otherwise we'd have to implement a multi-pass or flag the lock
1652 	 * failures and retry.
1653 	 */
1654 	/*flags = VMSC_GETVP;*/
1655 	flags = 0;
1656 	if (waitfor & MNT_LAZY)
1657 		flags |= VMSC_ONEPASS;
1658 
1659 	hammer2_trans_init(&info.trans, pmp, HAMMER2_TRANS_ISFLUSH);
1660 
1661 	/*
1662 	 * vfsync the vnodes. XXX
1663 	 */
1664 	info.error = 0;
1665 	info.waitfor = MNT_NOWAIT;
1666 	vmntvnodescan(mp, flags | VMSC_NOWAIT,
1667 		      hammer2_sync_scan1,
1668 		      hammer2_sync_scan2, &info);
1669 	if (info.error == 0 && (waitfor & MNT_WAIT)) {
1670 		info.waitfor = waitfor;
1671 		    vmntvnodescan(mp, flags,
1672 				  hammer2_sync_scan1,
1673 				  hammer2_sync_scan2, &info);
1674 
1675 	}
1676 #if 0
1677 	if (waitfor == MNT_WAIT) {
1678 		/* XXX */
1679 	} else {
1680 		/* XXX */
1681 	}
1682 #endif
1683 
1684 	total_error = 0;
1685 	for (i = 0; i < pmp->cluster.nchains; ++i) {
1686 		hmp = pmp->cluster.chains[i]->hmp;
1687 
1688 		/*
1689 		 * Media mounts have two 'roots', vchain for the topology
1690 		 * and fchain for the free block table.  Flush both.
1691 		 *
1692 		 * Note that the topology and free block table are handled
1693 		 * independently, so the free block table can wind up being
1694 		 * ahead of the topology.  We depend on the bulk free scan
1695 		 * code to deal with any loose ends.
1696 		 */
1697 		hammer2_chain_lock(&hmp->vchain, HAMMER2_RESOLVE_ALWAYS);
1698 		if (hmp->vchain.flags & (HAMMER2_CHAIN_MODIFIED |
1699 					  HAMMER2_CHAIN_SUBMODIFIED)) {
1700 			hammer2_chain_flush(&info.trans, &hmp->vchain);
1701 		}
1702 		hammer2_chain_unlock(&hmp->vchain);
1703 
1704 		hammer2_chain_lock(&hmp->fchain, HAMMER2_RESOLVE_ALWAYS);
1705 		if (hmp->fchain.flags & (HAMMER2_CHAIN_MODIFIED |
1706 					 HAMMER2_CHAIN_SUBMODIFIED)) {
1707 			/* this will also modify vchain as a side effect */
1708 			hammer2_chain_flush(&info.trans, &hmp->fchain);
1709 		}
1710 		hammer2_chain_unlock(&hmp->fchain);
1711 
1712 		error = 0;
1713 
1714 		/*
1715 		 * We can't safely flush the volume header until we have
1716 		 * flushed any device buffers which have built up.
1717 		 *
1718 		 * XXX this isn't being incremental
1719 		 */
1720 		vn_lock(hmp->devvp, LK_EXCLUSIVE | LK_RETRY);
1721 		error = VOP_FSYNC(hmp->devvp, MNT_WAIT, 0);
1722 		vn_unlock(hmp->devvp);
1723 
1724 		/*
1725 		 * The flush code sets CHAIN_VOLUMESYNC to indicate that the
1726 		 * volume header needs synchronization via hmp->volsync.
1727 		 *
1728 		 * XXX synchronize the flag & data with only this flush XXX
1729 		 */
1730 		if (error == 0 &&
1731 		    (hmp->vchain.flags & HAMMER2_CHAIN_VOLUMESYNC)) {
1732 			struct buf *bp;
1733 
1734 			/*
1735 			 * Synchronize the disk before flushing the volume
1736 			 * header.
1737 			 */
1738 			bp = getpbuf(NULL);
1739 			bp->b_bio1.bio_offset = 0;
1740 			bp->b_bufsize = 0;
1741 			bp->b_bcount = 0;
1742 			bp->b_cmd = BUF_CMD_FLUSH;
1743 			bp->b_bio1.bio_done = biodone_sync;
1744 			bp->b_bio1.bio_flags |= BIO_SYNC;
1745 			vn_strategy(hmp->devvp, &bp->b_bio1);
1746 			biowait(&bp->b_bio1, "h2vol");
1747 			relpbuf(bp, NULL);
1748 
1749 			/*
1750 			 * Then we can safely flush the version of the
1751 			 * volume header synchronized by the flush code.
1752 			 */
1753 			i = hmp->volhdrno + 1;
1754 			if (i >= HAMMER2_NUM_VOLHDRS)
1755 				i = 0;
1756 			if (i * HAMMER2_ZONE_BYTES64 + HAMMER2_SEGSIZE >
1757 			    hmp->volsync.volu_size) {
1758 				i = 0;
1759 			}
1760 			kprintf("sync volhdr %d %jd\n",
1761 				i, (intmax_t)hmp->volsync.volu_size);
1762 			bp = getblk(hmp->devvp, i * HAMMER2_ZONE_BYTES64,
1763 				    HAMMER2_PBUFSIZE, 0, 0);
1764 			atomic_clear_int(&hmp->vchain.flags,
1765 					 HAMMER2_CHAIN_VOLUMESYNC);
1766 			bcopy(&hmp->volsync, bp->b_data, HAMMER2_PBUFSIZE);
1767 			bawrite(bp);
1768 			hmp->volhdrno = i;
1769 		}
1770 		if (error)
1771 			total_error = error;
1772 	}
1773 
1774 	hammer2_trans_done(&info.trans);
1775 	return (total_error);
1776 }
1777 
1778 /*
1779  * Sync passes.
1780  *
1781  * NOTE: We don't test SUBMODIFIED or MOVED here because the fsync code
1782  *	 won't flush on those flags.  The syncer code above will do a
1783  *	 general meta-data flush globally that will catch these flags.
1784  */
1785 static int
1786 hammer2_sync_scan1(struct mount *mp, struct vnode *vp, void *data)
1787 {
1788 	hammer2_inode_t *ip;
1789 
1790 	ip = VTOI(vp);
1791 	if (vp->v_type == VNON || ip == NULL ||
1792 	    ((ip->flags & HAMMER2_INODE_MODIFIED) == 0 &&
1793 	     RB_EMPTY(&vp->v_rbdirty_tree))) {
1794 		return(-1);
1795 	}
1796 	return(0);
1797 }
1798 
1799 static int
1800 hammer2_sync_scan2(struct mount *mp, struct vnode *vp, void *data)
1801 {
1802 	struct hammer2_sync_info *info = data;
1803 	hammer2_inode_t *ip;
1804 	hammer2_chain_t *parent;
1805 	int error;
1806 
1807 	ip = VTOI(vp);
1808 	if (vp->v_type == VNON || vp->v_type == VBAD ||
1809 	    ((ip->flags & HAMMER2_INODE_MODIFIED) == 0 &&
1810 	     RB_EMPTY(&vp->v_rbdirty_tree))) {
1811 		return(0);
1812 	}
1813 
1814 	/*
1815 	 * VOP_FSYNC will start a new transaction so replicate some code
1816 	 * here to do it inline (see hammer2_vop_fsync()).
1817 	 *
1818 	 * WARNING: The vfsync interacts with the buffer cache and might
1819 	 *          block, we can't hold the inode lock at that time.
1820 	 */
1821 	atomic_clear_int(&ip->flags, HAMMER2_INODE_MODIFIED);
1822 	if (ip->vp)
1823 		vfsync(ip->vp, MNT_NOWAIT, 1, NULL, NULL);
1824 	parent = hammer2_inode_lock_ex(ip);
1825 	hammer2_chain_flush(&info->trans, parent);
1826 	hammer2_inode_unlock_ex(ip, parent);
1827 	error = 0;
1828 #if 0
1829 	error = VOP_FSYNC(vp, MNT_NOWAIT, 0);
1830 #endif
1831 	if (error)
1832 		info->error = error;
1833 	return(0);
1834 }
1835 
1836 static
1837 int
1838 hammer2_vfs_vptofh(struct vnode *vp, struct fid *fhp)
1839 {
1840 	return (0);
1841 }
1842 
1843 static
1844 int
1845 hammer2_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
1846 	       struct fid *fhp, struct vnode **vpp)
1847 {
1848 	return (0);
1849 }
1850 
1851 static
1852 int
1853 hammer2_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
1854 		 int *exflagsp, struct ucred **credanonp)
1855 {
1856 	return (0);
1857 }
1858 
1859 /*
1860  * Support code for hammer2_mount().  Read, verify, and install the volume
1861  * header into the HMP
1862  *
1863  * XXX read four volhdrs and use the one with the highest TID whos CRC
1864  *     matches.
1865  *
1866  * XXX check iCRCs.
1867  *
1868  * XXX For filesystems w/ less than 4 volhdrs, make sure to not write to
1869  *     nonexistant locations.
1870  *
1871  * XXX Record selected volhdr and ring updates to each of 4 volhdrs
1872  */
1873 static
1874 int
1875 hammer2_install_volume_header(hammer2_mount_t *hmp)
1876 {
1877 	hammer2_volume_data_t *vd;
1878 	struct buf *bp;
1879 	hammer2_crc32_t crc0, crc, bcrc0, bcrc;
1880 	int error_reported;
1881 	int error;
1882 	int valid;
1883 	int i;
1884 
1885 	error_reported = 0;
1886 	error = 0;
1887 	valid = 0;
1888 	bp = NULL;
1889 
1890 	/*
1891 	 * There are up to 4 copies of the volume header (syncs iterate
1892 	 * between them so there is no single master).  We don't trust the
1893 	 * volu_size field so we don't know precisely how large the filesystem
1894 	 * is, so depend on the OS to return an error if we go beyond the
1895 	 * block device's EOF.
1896 	 */
1897 	for (i = 0; i < HAMMER2_NUM_VOLHDRS; i++) {
1898 		error = bread(hmp->devvp, i * HAMMER2_ZONE_BYTES64,
1899 			      HAMMER2_VOLUME_BYTES, &bp);
1900 		if (error) {
1901 			brelse(bp);
1902 			bp = NULL;
1903 			continue;
1904 		}
1905 
1906 		vd = (struct hammer2_volume_data *) bp->b_data;
1907 		if ((vd->magic != HAMMER2_VOLUME_ID_HBO) &&
1908 		    (vd->magic != HAMMER2_VOLUME_ID_ABO)) {
1909 			brelse(bp);
1910 			bp = NULL;
1911 			continue;
1912 		}
1913 
1914 		if (vd->magic == HAMMER2_VOLUME_ID_ABO) {
1915 			/* XXX: Reversed-endianness filesystem */
1916 			kprintf("hammer2: reverse-endian filesystem detected");
1917 			brelse(bp);
1918 			bp = NULL;
1919 			continue;
1920 		}
1921 
1922 		crc = vd->icrc_sects[HAMMER2_VOL_ICRC_SECT0];
1923 		crc0 = hammer2_icrc32(bp->b_data + HAMMER2_VOLUME_ICRC0_OFF,
1924 				      HAMMER2_VOLUME_ICRC0_SIZE);
1925 		bcrc = vd->icrc_sects[HAMMER2_VOL_ICRC_SECT1];
1926 		bcrc0 = hammer2_icrc32(bp->b_data + HAMMER2_VOLUME_ICRC1_OFF,
1927 				       HAMMER2_VOLUME_ICRC1_SIZE);
1928 		if ((crc0 != crc) || (bcrc0 != bcrc)) {
1929 			kprintf("hammer2 volume header crc "
1930 				"mismatch copy #%d %08x/%08x\n",
1931 				i, crc0, crc);
1932 			error_reported = 1;
1933 			brelse(bp);
1934 			bp = NULL;
1935 			continue;
1936 		}
1937 		if (valid == 0 || hmp->voldata.mirror_tid < vd->mirror_tid) {
1938 			valid = 1;
1939 			hmp->voldata = *vd;
1940 			hmp->volhdrno = i;
1941 		}
1942 		brelse(bp);
1943 		bp = NULL;
1944 	}
1945 	if (valid) {
1946 		hmp->volsync = hmp->voldata;
1947 		error = 0;
1948 		if (error_reported || bootverbose || 1) { /* 1/DEBUG */
1949 			kprintf("hammer2: using volume header #%d\n",
1950 				hmp->volhdrno);
1951 		}
1952 	} else {
1953 		error = EINVAL;
1954 		kprintf("hammer2: no valid volume headers found!\n");
1955 	}
1956 	return (error);
1957 }
1958 
1959 /*
1960  * Reconnect using the passed file pointer.  The caller must ref the
1961  * fp for us.
1962  */
1963 void
1964 hammer2_cluster_reconnect(hammer2_pfsmount_t *pmp, struct file *fp)
1965 {
1966 	hammer2_inode_data_t *ipdata;
1967 	hammer2_chain_t *parent;
1968 	hammer2_mount_t *hmp;
1969 	size_t name_len;
1970 
1971 	hmp = pmp->cluster.chains[0]->hmp;	/* XXX */
1972 
1973 	/*
1974 	 * Closes old comm descriptor, kills threads, cleans up
1975 	 * states, then installs the new descriptor and creates
1976 	 * new threads.
1977 	 */
1978 	kdmsg_iocom_reconnect(&pmp->iocom, fp, "hammer2");
1979 
1980 	/*
1981 	 * Setup LNK_CONN fields for autoinitiated state machine
1982 	 */
1983 	parent = hammer2_inode_lock_ex(pmp->iroot);
1984 	ipdata = &parent->data->ipdata;
1985 	pmp->iocom.auto_lnk_conn.pfs_clid = ipdata->pfs_clid;
1986 	pmp->iocom.auto_lnk_conn.pfs_fsid = ipdata->pfs_fsid;
1987 	pmp->iocom.auto_lnk_conn.pfs_type = ipdata->pfs_type;
1988 	pmp->iocom.auto_lnk_conn.proto_version = DMSG_SPAN_PROTO_1;
1989 	pmp->iocom.auto_lnk_conn.peer_type = hmp->voldata.peer_type;
1990 
1991 	/*
1992 	 * Filter adjustment.  Clients do not need visibility into other
1993 	 * clients (otherwise millions of clients would present a serious
1994 	 * problem).  The fs_label also serves to restrict the namespace.
1995 	 */
1996 	pmp->iocom.auto_lnk_conn.peer_mask = 1LLU << HAMMER2_PEER_HAMMER2;
1997 	pmp->iocom.auto_lnk_conn.pfs_mask = (uint64_t)-1;
1998 	switch (ipdata->pfs_type) {
1999 	case DMSG_PFSTYPE_CLIENT:
2000 		pmp->iocom.auto_lnk_conn.peer_mask &=
2001 				~(1LLU << DMSG_PFSTYPE_CLIENT);
2002 		break;
2003 	default:
2004 		break;
2005 	}
2006 
2007 	name_len = ipdata->name_len;
2008 	if (name_len >= sizeof(pmp->iocom.auto_lnk_conn.fs_label))
2009 		name_len = sizeof(pmp->iocom.auto_lnk_conn.fs_label) - 1;
2010 	bcopy(ipdata->filename,
2011 	      pmp->iocom.auto_lnk_conn.fs_label,
2012 	      name_len);
2013 	pmp->iocom.auto_lnk_conn.fs_label[name_len] = 0;
2014 
2015 	/*
2016 	 * Setup LNK_SPAN fields for autoinitiated state machine
2017 	 */
2018 	pmp->iocom.auto_lnk_span.pfs_clid = ipdata->pfs_clid;
2019 	pmp->iocom.auto_lnk_span.pfs_fsid = ipdata->pfs_fsid;
2020 	pmp->iocom.auto_lnk_span.pfs_type = ipdata->pfs_type;
2021 	pmp->iocom.auto_lnk_span.peer_type = hmp->voldata.peer_type;
2022 	pmp->iocom.auto_lnk_span.proto_version = DMSG_SPAN_PROTO_1;
2023 	name_len = ipdata->name_len;
2024 	if (name_len >= sizeof(pmp->iocom.auto_lnk_span.fs_label))
2025 		name_len = sizeof(pmp->iocom.auto_lnk_span.fs_label) - 1;
2026 	bcopy(ipdata->filename,
2027 	      pmp->iocom.auto_lnk_span.fs_label,
2028 	      name_len);
2029 	pmp->iocom.auto_lnk_span.fs_label[name_len] = 0;
2030 	hammer2_inode_unlock_ex(pmp->iroot, parent);
2031 
2032 	kdmsg_iocom_autoinitiate(&pmp->iocom, hammer2_autodmsg);
2033 }
2034 
2035 static int
2036 hammer2_rcvdmsg(kdmsg_msg_t *msg)
2037 {
2038 	switch(msg->any.head.cmd & DMSGF_TRANSMASK) {
2039 	case DMSG_DBG_SHELL:
2040 		/*
2041 		 * (non-transaction)
2042 		 * Execute shell command (not supported atm)
2043 		 */
2044 		kdmsg_msg_reply(msg, DMSG_ERR_NOSUPP);
2045 		break;
2046 	case DMSG_DBG_SHELL | DMSGF_REPLY:
2047 		/*
2048 		 * (non-transaction)
2049 		 */
2050 		if (msg->aux_data) {
2051 			msg->aux_data[msg->aux_size - 1] = 0;
2052 			kprintf("HAMMER2 DBG: %s\n", msg->aux_data);
2053 		}
2054 		break;
2055 	default:
2056 		/*
2057 		 * Unsupported message received.  We only need to
2058 		 * reply if it's a transaction in order to close our end.
2059 		 * Ignore any one-way messages are any further messages
2060 		 * associated with the transaction.
2061 		 *
2062 		 * NOTE: This case also includes DMSG_LNK_ERROR messages
2063 		 *	 which might be one-way, replying to those would
2064 		 *	 cause an infinite ping-pong.
2065 		 */
2066 		if (msg->any.head.cmd & DMSGF_CREATE)
2067 			kdmsg_msg_reply(msg, DMSG_ERR_NOSUPP);
2068 		break;
2069 	}
2070 	return(0);
2071 }
2072 
2073 /*
2074  * This function is called after KDMSG has automatically handled processing
2075  * of a LNK layer message (typically CONN, SPAN, or CIRC).
2076  *
2077  * We tag off the LNK_CONN to trigger our LNK_VOLCONF messages which
2078  * advertises all available hammer2 super-root volumes.
2079  */
2080 static void
2081 hammer2_autodmsg(kdmsg_msg_t *msg)
2082 {
2083 	hammer2_pfsmount_t *pmp = msg->iocom->handle;
2084 	hammer2_mount_t *hmp = pmp->cluster.chains[0]->hmp; /* XXX */
2085 	int copyid;
2086 
2087 	/*
2088 	 * We only care about replies to our LNK_CONN auto-request.  kdmsg
2089 	 * has already processed the reply, we use this calback as a shim
2090 	 * to know when we can advertise available super-root volumes.
2091 	 */
2092 	if ((msg->any.head.cmd & DMSGF_TRANSMASK) !=
2093 	    (DMSG_LNK_CONN | DMSGF_CREATE | DMSGF_REPLY) ||
2094 	    msg->state == NULL) {
2095 		return;
2096 	}
2097 
2098 	kprintf("LNK_CONN REPLY RECEIVED CMD %08x\n", msg->any.head.cmd);
2099 
2100 	if (msg->any.head.cmd & DMSGF_CREATE) {
2101 		kprintf("HAMMER2: VOLDATA DUMP\n");
2102 
2103 		/*
2104 		 * Dump the configuration stored in the volume header
2105 		 */
2106 		hammer2_voldata_lock(hmp);
2107 		for (copyid = 0; copyid < HAMMER2_COPYID_COUNT; ++copyid) {
2108 			if (hmp->voldata.copyinfo[copyid].copyid == 0)
2109 				continue;
2110 			hammer2_volconf_update(pmp, copyid);
2111 		}
2112 		hammer2_voldata_unlock(hmp, 0);
2113 	}
2114 	if ((msg->any.head.cmd & DMSGF_DELETE) &&
2115 	    msg->state && (msg->state->txcmd & DMSGF_DELETE) == 0) {
2116 		kprintf("HAMMER2: CONN WAS TERMINATED\n");
2117 	}
2118 }
2119 
2120 /*
2121  * Volume configuration updates are passed onto the userland service
2122  * daemon via the open LNK_CONN transaction.
2123  */
2124 void
2125 hammer2_volconf_update(hammer2_pfsmount_t *pmp, int index)
2126 {
2127 	hammer2_mount_t *hmp = pmp->cluster.chains[0]->hmp;	/* XXX */
2128 	kdmsg_msg_t *msg;
2129 
2130 	/* XXX interlock against connection state termination */
2131 	kprintf("volconf update %p\n", pmp->iocom.conn_state);
2132 	if (pmp->iocom.conn_state) {
2133 		kprintf("TRANSMIT VOLCONF VIA OPEN CONN TRANSACTION\n");
2134 		msg = kdmsg_msg_alloc_state(pmp->iocom.conn_state,
2135 					    DMSG_LNK_VOLCONF, NULL, NULL);
2136 		msg->any.lnk_volconf.copy = hmp->voldata.copyinfo[index];
2137 		msg->any.lnk_volconf.mediaid = hmp->voldata.fsid;
2138 		msg->any.lnk_volconf.index = index;
2139 		kdmsg_msg_write(msg);
2140 	}
2141 }
2142 
2143 void
2144 hammer2_dump_chain(hammer2_chain_t *chain, int tab, int *countp)
2145 {
2146 	hammer2_chain_layer_t *layer;
2147 	hammer2_chain_t *scan;
2148 	hammer2_chain_t *first_parent;
2149 
2150 	--*countp;
2151 	if (*countp == 0) {
2152 		kprintf("%*.*s...\n", tab, tab, "");
2153 		return;
2154 	}
2155 	if (*countp < 0)
2156 		return;
2157 	first_parent = chain->core ? TAILQ_FIRST(&chain->core->ownerq) : NULL;
2158 	kprintf("%*.*schain %p.%d [%08x][core=%p fp=%p] (%s) np=%p dt=%s refs=%d",
2159 		tab, tab, "",
2160 		chain, chain->bref.type, chain->flags,
2161 		chain->core,
2162 		first_parent,
2163 		((chain->bref.type == HAMMER2_BREF_TYPE_INODE &&
2164 		chain->data) ?  (char *)chain->data->ipdata.filename : "?"),
2165 		(first_parent ? TAILQ_NEXT(chain, core_entry) : NULL),
2166 		(chain->delete_tid == HAMMER2_MAX_TID ? "max" : "fls"),
2167 		chain->refs);
2168 	if (first_parent)
2169 		kprintf(" [fpflags %08x fprefs %d\n",
2170 			first_parent->flags,
2171 			first_parent->refs);
2172 	if (chain->core == NULL || TAILQ_EMPTY(&chain->core->layerq))
2173 		kprintf("\n");
2174 	else
2175 		kprintf(" {\n");
2176 	TAILQ_FOREACH(layer, &chain->core->layerq, entry) {
2177 		RB_FOREACH(scan, hammer2_chain_tree, &layer->rbtree) {
2178 			hammer2_dump_chain(scan, tab + 4, countp);
2179 		}
2180 	}
2181 	if (chain->core && !TAILQ_EMPTY(&chain->core->layerq)) {
2182 		if (chain->bref.type == HAMMER2_BREF_TYPE_INODE && chain->data)
2183 			kprintf("%*.*s}(%s)\n", tab, tab, "",
2184 				chain->data->ipdata.filename);
2185 		else
2186 			kprintf("%*.*s}\n", tab, tab, "");
2187 	}
2188 }
2189