xref: /dflybsd-src/sys/vfs/hammer2/hammer2_vfsops.c (revision 68e609aa42b5c8082ca864b600d1683c6a82dc68)
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(struct mount *, 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 
351 	hmp = NULL;
352 	pmp = NULL;
353 	dev = NULL;
354 	label = NULL;
355 	devvp = NULL;
356 	cache_index = -1;
357 
358 	kprintf("hammer2_mount\n");
359 
360 	if (path == NULL) {
361 		/*
362 		 * Root mount
363 		 */
364 		bzero(&info, sizeof(info));
365 		info.cluster_fd = -1;
366 		return (EOPNOTSUPP);
367 	} else {
368 		/*
369 		 * Non-root mount or updating a mount
370 		 */
371 		error = copyin(data, &info, sizeof(info));
372 		if (error)
373 			return (error);
374 
375 		error = copyinstr(info.volume, devstr, MNAMELEN - 1, &done);
376 		if (error)
377 			return (error);
378 
379 		/* Extract device and label */
380 		dev = devstr;
381 		label = strchr(devstr, '@');
382 		if (label == NULL ||
383 		    ((label + 1) - dev) > done) {
384 			return (EINVAL);
385 		}
386 		*label = '\0';
387 		label++;
388 		if (*label == '\0')
389 			return (EINVAL);
390 
391 		if (mp->mnt_flag & MNT_UPDATE) {
392 			/* Update mount */
393 			/* HAMMER2 implements NFS export via mountctl */
394 			hmp = MPTOHMP(mp);
395 			devvp = hmp->devvp;
396 			error = hammer2_remount(mp, path, devvp, cred);
397 			return error;
398 		}
399 	}
400 
401 	/*
402 	 * PFS mount
403 	 *
404 	 * Lookup name and verify it refers to a block device.
405 	 */
406 	error = nlookup_init(&nd, dev, UIO_SYSSPACE, NLC_FOLLOW);
407 	if (error == 0)
408 		error = nlookup(&nd);
409 	if (error == 0)
410 		error = cache_vref(&nd.nl_nch, nd.nl_cred, &devvp);
411 	nlookup_done(&nd);
412 
413 	if (error == 0) {
414 		if (vn_isdisk(devvp, &error))
415 			error = vfs_mountedon(devvp);
416 	}
417 
418 	/*
419 	 * Determine if the device has already been mounted.  After this
420 	 * check hmp will be non-NULL if we are doing the second or more
421 	 * hammer2 mounts from the same device.
422 	 */
423 	lockmgr(&hammer2_mntlk, LK_EXCLUSIVE);
424 	TAILQ_FOREACH(hmp, &hammer2_mntlist, mntentry) {
425 		if (hmp->devvp == devvp)
426 			break;
427 	}
428 
429 	/*
430 	 * Open the device if this isn't a secondary mount and construct
431 	 * the H2 device mount (hmp).
432 	 */
433 	if (hmp == NULL) {
434 		if (error == 0 && vcount(devvp) > 0)
435 			error = EBUSY;
436 
437 		/*
438 		 * Now open the device
439 		 */
440 		if (error == 0) {
441 			ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
442 			vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
443 			error = vinvalbuf(devvp, V_SAVE, 0, 0);
444 			if (error == 0) {
445 				error = VOP_OPEN(devvp,
446 						 ronly ? FREAD : FREAD | FWRITE,
447 						 FSCRED, NULL);
448 			}
449 			vn_unlock(devvp);
450 		}
451 		if (error && devvp) {
452 			vrele(devvp);
453 			devvp = NULL;
454 		}
455 		if (error) {
456 			lockmgr(&hammer2_mntlk, LK_RELEASE);
457 			return error;
458 		}
459 		hmp = kmalloc(sizeof(*hmp), M_HAMMER2, M_WAITOK | M_ZERO);
460 		hmp->ronly = ronly;
461 		hmp->devvp = devvp;
462 		kmalloc_create(&hmp->mchain, "HAMMER2-chains");
463 		TAILQ_INSERT_TAIL(&hammer2_mntlist, hmp, mntentry);
464 
465 		lockinit(&hmp->alloclk, "h2alloc", 0, 0);
466 		lockinit(&hmp->voldatalk, "voldata", 0, LK_CANRECURSE);
467 		TAILQ_INIT(&hmp->transq);
468 
469 		/*
470 		 * vchain setup. vchain.data is embedded.
471 		 * vchain.refs is initialized and will never drop to 0.
472 		 */
473 		hmp->vchain.hmp = hmp;
474 		hmp->vchain.refs = 1;
475 		hmp->vchain.data = (void *)&hmp->voldata;
476 		hmp->vchain.bref.type = HAMMER2_BREF_TYPE_VOLUME;
477 		hmp->vchain.bref.data_off = 0 | HAMMER2_PBUFRADIX;
478 		hmp->vchain.delete_tid = HAMMER2_MAX_TID;
479 		hammer2_chain_core_alloc(NULL, &hmp->vchain, NULL);
480 		/* hmp->vchain.u.xxx is left NULL */
481 
482 		/*
483 		 * fchain setup.  fchain.data is embedded.
484 		 * fchain.refs is initialized and will never drop to 0.
485 		 *
486 		 * The data is not used but needs to be initialized to
487 		 * pass assertion muster.  We use this chain primarily
488 		 * as a placeholder for the freemap's top-level RBTREE
489 		 * so it does not interfere with the volume's topology
490 		 * RBTREE.
491 		 */
492 		hmp->fchain.hmp = hmp;
493 		hmp->fchain.refs = 1;
494 		hmp->fchain.data = (void *)&hmp->voldata.freemap_blockset;
495 		hmp->fchain.bref.type = HAMMER2_BREF_TYPE_FREEMAP;
496 		hmp->fchain.bref.data_off = 0 | HAMMER2_PBUFRADIX;
497 		hmp->fchain.bref.methods =
498 			HAMMER2_ENC_CHECK(HAMMER2_CHECK_FREEMAP) |
499 			HAMMER2_ENC_COMP(HAMMER2_COMP_NONE);
500 		hmp->fchain.delete_tid = HAMMER2_MAX_TID;
501 
502 		hammer2_chain_core_alloc(NULL, &hmp->fchain, NULL);
503 		/* hmp->fchain.u.xxx is left NULL */
504 
505 		/*
506 		 * Install the volume header
507 		 */
508 		error = hammer2_install_volume_header(hmp);
509 		if (error) {
510 			hammer2_vfs_unmount(mp, MNT_FORCE);
511 			return error;
512 		}
513 
514 		/*
515 		 * First locate the super-root inode, which is key 0
516 		 * relative to the volume header's blockset.
517 		 *
518 		 * Then locate the root inode by scanning the directory keyspace
519 		 * represented by the label.
520 		 */
521 		parent = hammer2_chain_lookup_init(&hmp->vchain, 0);
522 		schain = hammer2_chain_lookup(&parent, &key_dummy,
523 				      HAMMER2_SROOT_KEY, HAMMER2_SROOT_KEY,
524 				      &cache_index, 0);
525 		hammer2_chain_lookup_done(parent);
526 		if (schain == NULL) {
527 			kprintf("hammer2_mount: invalid super-root\n");
528 			hammer2_vfs_unmount(mp, MNT_FORCE);
529 			return EINVAL;
530 		}
531 		hammer2_chain_ref(schain);	/* for hmp->schain */
532 		hmp->schain = schain;		/* left locked for inode_get */
533 		hmp->sroot = hammer2_inode_get(NULL, NULL, schain);
534 		hammer2_inode_ref(hmp->sroot);	/* for hmp->sroot */
535 		hammer2_inode_unlock_ex(hmp->sroot, schain);
536 		schain = NULL;
537 
538 		mtx_init(&hmp->wthread_mtx);
539 		bioq_init(&hmp->wthread_bioq);
540 		hmp->wthread_destroy = 0;
541 
542 		/*
543 		 * Launch threads.
544 		 */
545 		lwkt_create(hammer2_write_thread, hmp,
546 				NULL, NULL, 0, -1, "hammer2-write");
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 	pmp->mount_cluster = kmalloc(sizeof(hammer2_cluster_t), M_HAMMER2,
557 				     M_WAITOK | M_ZERO);
558 	pmp->cluster = pmp->mount_cluster;
559 
560 	kmalloc_create(&pmp->minode, "HAMMER2-inodes");
561 	kmalloc_create(&pmp->mmsg, "HAMMER2-pfsmsg");
562 
563 	pmp->mount_cluster->hmp = hmp;
564 	spin_init(&pmp->inum_spin);
565 	RB_INIT(&pmp->inum_tree);
566 
567 	kdmsg_iocom_init(&pmp->iocom, pmp,
568 			 KDMSG_IOCOMF_AUTOCONN |
569 			 KDMSG_IOCOMF_AUTOSPAN |
570 			 KDMSG_IOCOMF_AUTOCIRC,
571 			 pmp->mmsg, hammer2_rcvdmsg);
572 
573 	ccms_domain_init(&pmp->ccms_dom);
574 	++hmp->pmp_count;
575 	lockmgr(&hammer2_mntlk, LK_RELEASE);
576 	kprintf("hammer2_mount hmp=%p pmp=%p pmpcnt=%d\n",
577 		hmp, pmp, hmp->pmp_count);
578 
579 	mp->mnt_flag = MNT_LOCAL;
580 	mp->mnt_kern_flag |= MNTK_ALL_MPSAFE;	/* all entry pts are SMP */
581 
582 	/*
583 	 * required mount structure initializations
584 	 */
585 	mp->mnt_stat.f_iosize = HAMMER2_PBUFSIZE;
586 	mp->mnt_stat.f_bsize = HAMMER2_PBUFSIZE;
587 
588 	mp->mnt_vstat.f_frsize = HAMMER2_PBUFSIZE;
589 	mp->mnt_vstat.f_bsize = HAMMER2_PBUFSIZE;
590 
591 	/*
592 	 * Optional fields
593 	 */
594 	mp->mnt_iosize_max = MAXPHYS;
595 	mp->mnt_data = (qaddr_t)pmp;
596 	pmp->mp = mp;
597 
598 	/*
599 	 * schain only has 1 ref now for its hmp->schain assignment.
600 	 * Setup for lookup (which will lock it).
601 	 */
602 	parent = hammer2_chain_lookup_init(hmp->schain, 0);
603 	lhc = hammer2_dirhash(label, strlen(label));
604 	rchain = hammer2_chain_lookup(&parent, &key_next,
605 				      lhc, lhc + HAMMER2_DIRHASH_LOMASK,
606 				      &cache_index, 0);
607 	while (rchain) {
608 		if (rchain->bref.type == HAMMER2_BREF_TYPE_INODE &&
609 		    strcmp(label, rchain->data->ipdata.filename) == 0) {
610 			break;
611 		}
612 		rchain = hammer2_chain_next(&parent, rchain, &key_next,
613 					    key_next,
614 					    lhc + HAMMER2_DIRHASH_LOMASK,
615 					    &cache_index, 0);
616 	}
617 	hammer2_chain_lookup_done(parent);
618 	if (rchain == NULL) {
619 		kprintf("hammer2_mount: PFS label not found\n");
620 		hammer2_vfs_unmount(mp, MNT_FORCE);
621 		return EINVAL;
622 	}
623 	if (rchain->flags & HAMMER2_CHAIN_MOUNTED) {
624 		hammer2_chain_unlock(rchain);
625 		kprintf("hammer2_mount: PFS label already mounted!\n");
626 		hammer2_vfs_unmount(mp, MNT_FORCE);
627 		return EBUSY;
628 	}
629 	if (rchain->flags & HAMMER2_CHAIN_RECYCLE) {
630 		kprintf("hammer2_mount: PFS label currently recycling\n");
631 		hammer2_vfs_unmount(mp, MNT_FORCE);
632 		return EBUSY;
633 	}
634 
635 	atomic_set_int(&rchain->flags, HAMMER2_CHAIN_MOUNTED);
636 
637 	/*
638 	 * NOTE: *_get() integrates chain's lock into the inode lock.
639 	 */
640 	hammer2_chain_ref(rchain);		/* for pmp->rchain */
641 	pmp->mount_cluster->rchain = rchain;	/* left held & unlocked */
642 	pmp->iroot = hammer2_inode_get(pmp, NULL, rchain);
643 	hammer2_inode_ref(pmp->iroot);		/* ref for pmp->iroot */
644 
645 	KKASSERT(rchain->pmp == NULL);		/* tracking pmp for rchain */
646 	rchain->pmp = pmp;
647 	atomic_add_long(&pmp->inmem_chains, 1);
648 
649 	hammer2_inode_unlock_ex(pmp->iroot, rchain);
650 
651 	kprintf("iroot %p\n", pmp->iroot);
652 
653 	/*
654 	 * Ref the cluster management messaging descriptor.  The mount
655 	 * program deals with the other end of the communications pipe.
656 	 */
657 	fp = holdfp(curproc->p_fd, info.cluster_fd, -1);
658 	if (fp == NULL) {
659 		kprintf("hammer2_mount: bad cluster_fd!\n");
660 		hammer2_vfs_unmount(mp, MNT_FORCE);
661 		return EBADF;
662 	}
663 	hammer2_cluster_reconnect(pmp, fp);
664 
665 	/*
666 	 * Finish setup
667 	 */
668 	vfs_getnewfsid(mp);
669 	vfs_add_vnodeops(mp, &hammer2_vnode_vops, &mp->mnt_vn_norm_ops);
670 	vfs_add_vnodeops(mp, &hammer2_spec_vops, &mp->mnt_vn_spec_ops);
671 	vfs_add_vnodeops(mp, &hammer2_fifo_vops, &mp->mnt_vn_fifo_ops);
672 
673 	copyinstr(info.volume, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, &size);
674 	bzero(mp->mnt_stat.f_mntfromname + size, MNAMELEN - size);
675 	bzero(mp->mnt_stat.f_mntonname, sizeof(mp->mnt_stat.f_mntonname));
676 	copyinstr(path, mp->mnt_stat.f_mntonname,
677 		  sizeof(mp->mnt_stat.f_mntonname) - 1,
678 		  &size);
679 
680 	/*
681 	 * Initial statfs to prime mnt_stat.
682 	 */
683 	hammer2_vfs_statfs(mp, &mp->mnt_stat, cred);
684 
685 	return 0;
686 }
687 
688 /*
689  * Handle bioq for strategy write
690  */
691 static
692 void
693 hammer2_write_thread(void *arg)
694 {
695 	hammer2_mount_t* hmp;
696 	struct bio *bio;
697 	struct buf *bp;
698 	hammer2_trans_t trans;
699 	struct vnode *vp;
700 	hammer2_inode_t *last_ip;
701 	hammer2_inode_t *ip;
702 	hammer2_chain_t *parent;
703 	hammer2_chain_t **parentp;
704 	hammer2_inode_data_t *ipdata;
705 	hammer2_key_t lbase;
706 	int lblksize;
707 	int pblksize;
708 	int error;
709 
710 	hmp = arg;
711 
712 	mtx_lock(&hmp->wthread_mtx);
713 	while (hmp->wthread_destroy == 0) {
714 		if (bioq_first(&hmp->wthread_bioq) == NULL) {
715 			mtxsleep(&hmp->wthread_bioq, &hmp->wthread_mtx,
716 				 0, "h2bioqw", 0);
717 		}
718 		last_ip = NULL;
719 		parent = NULL;
720 		parentp = &parent;
721 
722 		while ((bio = bioq_takefirst(&hmp->wthread_bioq)) != NULL) {
723 			mtx_unlock(&hmp->wthread_mtx);
724 
725 			error = 0;
726 			bp = bio->bio_buf;
727 			vp = bp->b_vp;
728 			ip = VTOI(vp);
729 
730 			/*
731 			 * Cache transaction for multi-buffer flush efficiency.
732 			 * Lock the ip separately for each buffer to allow
733 			 * interleaving with frontend writes.
734 			 */
735 			if (last_ip != ip) {
736 				if (last_ip)
737 					hammer2_trans_done(&trans);
738 				hammer2_trans_init(&trans, ip->pmp,
739 						   HAMMER2_TRANS_BUFCACHE);
740 				last_ip = ip;
741 			}
742 			parent = hammer2_inode_lock_ex(ip);
743 
744 			/*
745 			 * Inode is modified, flush size and mtime changes
746 			 * to ensure that the file size remains consistent
747 			 * with the buffers being flushed.
748 			 */
749 			if (ip->flags & (HAMMER2_INODE_RESIZED |
750 					 HAMMER2_INODE_MTIME)) {
751 				hammer2_inode_fsync(&trans, ip, parentp);
752 			}
753 			ipdata = hammer2_chain_modify_ip(&trans, ip,
754 							 parentp, 0);
755 			lblksize = hammer2_calc_logical(ip, bio->bio_offset,
756 							&lbase, NULL);
757 			pblksize = hammer2_calc_physical(ip, lbase);
758 			hammer2_write_file_core(bp, &trans, ip, ipdata,
759 						parentp,
760 						lbase, IO_ASYNC,
761 						pblksize, &error);
762 			hammer2_inode_unlock_ex(ip, parent);
763 			if (error) {
764 				kprintf("hammer2: error in buffer write\n");
765 				bp->b_flags |= B_ERROR;
766 				bp->b_error = EIO;
767 			}
768 			biodone(bio);
769 			mtx_lock(&hmp->wthread_mtx);
770 		}
771 
772 		/*
773 		 * Clean out transaction cache
774 		 */
775 		if (last_ip)
776 			hammer2_trans_done(&trans);
777 	}
778 	hmp->wthread_destroy = -1;
779 	wakeup(&hmp->wthread_destroy);
780 
781 	mtx_unlock(&hmp->wthread_mtx);
782 }
783 
784 /*
785  * Return a chain suitable for I/O, creating the chain if necessary
786  * and assigning its physical block.
787  */
788 static
789 hammer2_chain_t *
790 hammer2_assign_physical(hammer2_trans_t *trans,
791 			hammer2_inode_t *ip, hammer2_chain_t **parentp,
792 			hammer2_key_t lbase, int pblksize, int *errorp)
793 {
794 	hammer2_chain_t *parent;
795 	hammer2_chain_t *chain;
796 	hammer2_off_t pbase;
797 	hammer2_key_t key_dummy;
798 	int pradix = hammer2_getradix(pblksize);
799 	int cache_index = -1;
800 
801 	/*
802 	 * Locate the chain associated with lbase, return a locked chain.
803 	 * However, do not instantiate any data reference (which utilizes a
804 	 * device buffer) because we will be using direct IO via the
805 	 * logical buffer cache buffer.
806 	 */
807 	*errorp = 0;
808 	KKASSERT(pblksize >= HAMMER2_MIN_ALLOC);
809 retry:
810 	parent = *parentp;
811 	hammer2_chain_lock(parent, HAMMER2_RESOLVE_ALWAYS); /* extra lock */
812 	chain = hammer2_chain_lookup(&parent, &key_dummy,
813 				     lbase, lbase,
814 				     &cache_index, HAMMER2_LOOKUP_NODATA);
815 
816 	if (chain == NULL) {
817 		/*
818 		 * We found a hole, create a new chain entry.
819 		 *
820 		 * NOTE: DATA chains are created without device backing
821 		 *	 store (nor do we want any).
822 		 */
823 		*errorp = hammer2_chain_create(trans, &parent, &chain,
824 					       lbase, HAMMER2_PBUFRADIX,
825 					       HAMMER2_BREF_TYPE_DATA,
826 					       pblksize);
827 		if (chain == NULL) {
828 			hammer2_chain_lookup_done(parent);
829 			panic("hammer2_chain_create: par=%p error=%d\n",
830 				parent, *errorp);
831 			goto retry;
832 		}
833 
834 		pbase = chain->bref.data_off & ~HAMMER2_OFF_MASK_RADIX;
835 		/*ip->delta_dcount += pblksize;*/
836 	} else {
837 		switch (chain->bref.type) {
838 		case HAMMER2_BREF_TYPE_INODE:
839 			/*
840 			 * The data is embedded in the inode.  The
841 			 * caller is responsible for marking the inode
842 			 * modified and copying the data to the embedded
843 			 * area.
844 			 */
845 			pbase = NOOFFSET;
846 			break;
847 		case HAMMER2_BREF_TYPE_DATA:
848 			if (chain->bytes != pblksize) {
849 				hammer2_chain_resize(trans, ip,
850 						     parent, &chain,
851 						     pradix,
852 						     HAMMER2_MODIFY_OPTDATA);
853 			}
854 			hammer2_chain_modify(trans, &chain,
855 					     HAMMER2_MODIFY_OPTDATA);
856 			pbase = chain->bref.data_off & ~HAMMER2_OFF_MASK_RADIX;
857 			break;
858 		default:
859 			panic("hammer2_assign_physical: bad type");
860 			/* NOT REACHED */
861 			pbase = NOOFFSET;
862 			break;
863 		}
864 	}
865 
866 	/*
867 	 * Cleanup.  If chain wound up being the inode (i.e. DIRECTDATA),
868 	 * we might have to replace *parentp.
869 	 */
870 	hammer2_chain_lookup_done(parent);
871 	if (chain) {
872 		if (*parentp != chain &&
873 		    (*parentp)->core == chain->core) {
874 			parent = *parentp;
875 			*parentp = chain;		/* eats lock */
876 			hammer2_chain_unlock(parent);
877 			hammer2_chain_lock(chain, 0);	/* need another */
878 		}
879 		/* else chain already locked for return */
880 	}
881 	return (chain);
882 }
883 
884 /*
885  * From hammer2_vnops.c.
886  * The core write function which determines which path to take
887  * depending on compression settings.
888  */
889 static
890 void
891 hammer2_write_file_core(struct buf *bp, hammer2_trans_t *trans,
892 			hammer2_inode_t *ip, hammer2_inode_data_t *ipdata,
893 			hammer2_chain_t **parentp,
894 			hammer2_key_t lbase, int ioflag, int pblksize,
895 			int *errorp)
896 {
897 	hammer2_chain_t *chain;
898 
899 	switch(HAMMER2_DEC_COMP(ipdata->comp_algo)) {
900 	case HAMMER2_COMP_NONE:
901 		/*
902 		 * We have to assign physical storage to the buffer
903 		 * we intend to dirty or write now to avoid deadlocks
904 		 * in the strategy code later.
905 		 *
906 		 * This can return NOOFFSET for inode-embedded data.
907 		 * The strategy code will take care of it in that case.
908 		 */
909 		chain = hammer2_assign_physical(trans, ip, parentp,
910 						lbase, pblksize,
911 						errorp);
912 		hammer2_write_bp(chain, bp, ioflag, pblksize, errorp);
913 		if (chain)
914 			hammer2_chain_unlock(chain);
915 		break;
916 	case HAMMER2_COMP_AUTOZERO:
917 		/*
918 		 * Check for zero-fill only
919 		 */
920 		hammer2_zero_check_and_write(bp, trans, ip,
921 				    ipdata, parentp, lbase,
922 				    ioflag, pblksize, errorp);
923 		break;
924 	case HAMMER2_COMP_LZ4:
925 	case HAMMER2_COMP_ZLIB:
926 	default:
927 		/*
928 		 * Check for zero-fill and attempt compression.
929 		 */
930 		hammer2_compress_and_write(bp, trans, ip,
931 					   ipdata, parentp,
932 					   lbase, ioflag,
933 					   pblksize, errorp,
934 					   ipdata->comp_algo);
935 		break;
936 	}
937 	ipdata = &ip->chain->data->ipdata;	/* reload */
938 }
939 
940 /*
941  * From hammer2_vnops.c
942  * Generic function that will perform the compression in compression
943  * write path. The compression algorithm is determined by the settings
944  * obtained from inode.
945  */
946 static
947 void
948 hammer2_compress_and_write(struct buf *bp, hammer2_trans_t *trans,
949 	hammer2_inode_t *ip, hammer2_inode_data_t *ipdata,
950 	hammer2_chain_t **parentp,
951 	hammer2_key_t lbase, int ioflag, int pblksize,
952 	int *errorp, int comp_algo)
953 {
954 	hammer2_chain_t *chain;
955 	int comp_size;
956 	int comp_block_size;
957 	char *comp_buffer;
958 
959 	if (test_block_zeros(bp->b_data, pblksize)) {
960 		zero_write(bp, trans, ip, ipdata, parentp, lbase, errorp);
961 		return;
962 	}
963 
964 	comp_size = 0;
965 	comp_buffer = NULL;
966 
967 	KKASSERT(pblksize / 2 <= 32768);
968 
969 	if (ip->comp_heuristic < 8 || (ip->comp_heuristic & 7) == 0) {
970 		z_stream strm_compress;
971 		int comp_level;
972 		int ret;
973 
974 		switch(HAMMER2_DEC_COMP(comp_algo)) {
975 		case HAMMER2_COMP_LZ4:
976 			comp_buffer = objcache_get(cache_buffer_write,
977 						   M_INTWAIT);
978 			comp_size = LZ4_compress_limitedOutput(
979 					bp->b_data,
980 					&comp_buffer[sizeof(int)],
981 					pblksize,
982 					pblksize / 2 - sizeof(int));
983 			/*
984 			 * We need to prefix with the size, LZ4
985 			 * doesn't do it for us.  Add the related
986 			 * overhead.
987 			 */
988 			*(int *)comp_buffer = comp_size;
989 			if (comp_size)
990 				comp_size += sizeof(int);
991 			break;
992 		case HAMMER2_COMP_ZLIB:
993 			comp_level = HAMMER2_DEC_LEVEL(comp_algo);
994 			if (comp_level == 0)
995 				comp_level = 6;	/* default zlib compression */
996 			else if (comp_level < 6)
997 				comp_level = 6;
998 			else if (comp_level > 9)
999 				comp_level = 9;
1000 			ret = deflateInit(&strm_compress, comp_level);
1001 			if (ret != Z_OK) {
1002 				kprintf("HAMMER2 ZLIB: fatal error "
1003 					"on deflateInit.\n");
1004 			}
1005 
1006 			comp_buffer = objcache_get(cache_buffer_write,
1007 						   M_INTWAIT);
1008 			strm_compress.next_in = bp->b_data;
1009 			strm_compress.avail_in = pblksize;
1010 			strm_compress.next_out = comp_buffer;
1011 			strm_compress.avail_out = pblksize / 2;
1012 			ret = deflate(&strm_compress, Z_FINISH);
1013 			if (ret == Z_STREAM_END) {
1014 				comp_size = pblksize / 2 -
1015 					    strm_compress.avail_out;
1016 			} else {
1017 				comp_size = 0;
1018 			}
1019 			ret = deflateEnd(&strm_compress);
1020 			break;
1021 		default:
1022 			kprintf("Error: Unknown compression method.\n");
1023 			kprintf("Comp_method = %d.\n", comp_algo);
1024 			break;
1025 		}
1026 	}
1027 
1028 	if (comp_size == 0) {
1029 		/*
1030 		 * compression failed or turned off
1031 		 */
1032 		comp_block_size = pblksize;	/* safety */
1033 		if (++ip->comp_heuristic > 128)
1034 			ip->comp_heuristic = 8;
1035 	} else {
1036 		/*
1037 		 * compression succeeded
1038 		 */
1039 		ip->comp_heuristic = 0;
1040 		if (comp_size <= 1024) {
1041 			comp_block_size = 1024;
1042 		} else if (comp_size <= 2048) {
1043 			comp_block_size = 2048;
1044 		} else if (comp_size <= 4096) {
1045 			comp_block_size = 4096;
1046 		} else if (comp_size <= 8192) {
1047 			comp_block_size = 8192;
1048 		} else if (comp_size <= 16384) {
1049 			comp_block_size = 16384;
1050 		} else if (comp_size <= 32768) {
1051 			comp_block_size = 32768;
1052 		} else {
1053 			panic("hammer2: WRITE PATH: "
1054 			      "Weird comp_size value.");
1055 			/* NOT REACHED */
1056 			comp_block_size = pblksize;
1057 		}
1058 	}
1059 
1060 	chain = hammer2_assign_physical(trans, ip, parentp,
1061 					lbase, comp_block_size,
1062 					errorp);
1063 	ipdata = &ip->chain->data->ipdata;	/* RELOAD */
1064 
1065 	if (*errorp) {
1066 		kprintf("WRITE PATH: An error occurred while "
1067 			"assigning physical space.\n");
1068 		KKASSERT(chain == NULL);
1069 	} else {
1070 		/* Get device offset */
1071 		hammer2_off_t pbase;
1072 		hammer2_off_t pmask;
1073 		hammer2_off_t peof;
1074 		size_t boff;
1075 		size_t psize;
1076 		struct buf *dbp;
1077 		int temp_check;
1078 
1079 		KKASSERT(chain->flags & HAMMER2_CHAIN_MODIFIED);
1080 
1081 		switch(chain->bref.type) {
1082 		case HAMMER2_BREF_TYPE_INODE:
1083 			KKASSERT(chain->data->ipdata.op_flags &
1084 				 HAMMER2_OPFLAG_DIRECTDATA);
1085 			KKASSERT(bp->b_loffset == 0);
1086 			bcopy(bp->b_data, chain->data->ipdata.u.data,
1087 			      HAMMER2_EMBEDDED_BYTES);
1088 			break;
1089 		case HAMMER2_BREF_TYPE_DATA:
1090 			psize = hammer2_devblksize(chain->bytes);
1091 			pmask = (hammer2_off_t)psize - 1;
1092 			pbase = chain->bref.data_off & ~pmask;
1093 			boff = chain->bref.data_off &
1094 			       (HAMMER2_OFF_MASK & pmask);
1095 			peof = (pbase + HAMMER2_SEGMASK64) &
1096 			       ~HAMMER2_SEGMASK64;
1097 			temp_check = HAMMER2_DEC_CHECK(chain->bref.methods);
1098 
1099 			/*
1100 			 * Optimize out the read-before-write
1101 			 * if possible.
1102 			 */
1103 			if (comp_block_size == psize) {
1104 				dbp = getblk(chain->hmp->devvp, pbase,
1105 					     psize, 0, 0);
1106 			} else {
1107 				*errorp = bread(chain->hmp->devvp,
1108 						pbase, psize, &dbp);
1109 				if (*errorp) {
1110 					kprintf("hammer2: WRITE PATH: "
1111 						"dbp bread error\n");
1112 					break;
1113 				}
1114 			}
1115 
1116 			/*
1117 			 * When loading the block make sure we don't
1118 			 * leave garbage after the compressed data.
1119 			 */
1120 			if (comp_size) {
1121 				chain->bref.methods =
1122 					HAMMER2_ENC_COMP(comp_algo) +
1123 					HAMMER2_ENC_CHECK(temp_check);
1124 				bcopy(comp_buffer, dbp->b_data + boff,
1125 				      comp_size);
1126 				if (comp_size != comp_block_size) {
1127 					bzero(dbp->b_data + boff +
1128 						comp_size,
1129 					      comp_block_size -
1130 						comp_size);
1131 				}
1132 			} else {
1133 				chain->bref.methods =
1134 					HAMMER2_ENC_COMP(
1135 						HAMMER2_COMP_NONE) +
1136 					HAMMER2_ENC_CHECK(temp_check);
1137 				bcopy(bp->b_data, dbp->b_data + boff,
1138 				      pblksize);
1139 			}
1140 
1141 			/*
1142 			 * Device buffer is now valid, chain is no
1143 			 * longer in the initial state.
1144 			 */
1145 			atomic_clear_int(&chain->flags,
1146 					 HAMMER2_CHAIN_INITIAL);
1147 
1148 			/* Now write the related bdp. */
1149 			if (ioflag & IO_SYNC) {
1150 				/*
1151 				 * Synchronous I/O requested.
1152 				 */
1153 				bwrite(dbp);
1154 			/*
1155 			} else if ((ioflag & IO_DIRECT) &&
1156 				   loff + n == pblksize) {
1157 				bdwrite(dbp);
1158 			*/
1159 			} else if (ioflag & IO_ASYNC) {
1160 				bawrite(dbp);
1161 			} else if (hammer2_cluster_enable) {
1162 				cluster_write(dbp, peof,
1163 					      HAMMER2_PBUFSIZE,
1164 					      4/*XXX*/);
1165 			} else {
1166 				bdwrite(dbp);
1167 			}
1168 			break;
1169 		default:
1170 			panic("hammer2_write_bp: bad chain type %d\n",
1171 				chain->bref.type);
1172 			/* NOT REACHED */
1173 			break;
1174 		}
1175 
1176 		hammer2_chain_unlock(chain);
1177 	}
1178 	if (comp_buffer)
1179 		objcache_put(cache_buffer_write, comp_buffer);
1180 }
1181 
1182 /*
1183  * Function that performs zero-checking and writing without compression,
1184  * it corresponds to default zero-checking path.
1185  */
1186 static
1187 void
1188 hammer2_zero_check_and_write(struct buf *bp, hammer2_trans_t *trans,
1189 	hammer2_inode_t *ip, hammer2_inode_data_t *ipdata,
1190 	hammer2_chain_t **parentp,
1191 	hammer2_key_t lbase, int ioflag, int pblksize, int *errorp)
1192 {
1193 	hammer2_chain_t *chain;
1194 
1195 	if (test_block_zeros(bp->b_data, pblksize)) {
1196 		zero_write(bp, trans, ip, ipdata, parentp, lbase, errorp);
1197 	} else {
1198 		chain = hammer2_assign_physical(trans, ip, parentp,
1199 						lbase, pblksize, errorp);
1200 		hammer2_write_bp(chain, bp, ioflag, pblksize, errorp);
1201 		if (chain)
1202 			hammer2_chain_unlock(chain);
1203 	}
1204 }
1205 
1206 /*
1207  * A function to test whether a block of data contains only zeros,
1208  * returns TRUE (non-zero) if the block is all zeros.
1209  */
1210 static
1211 int
1212 test_block_zeros(const char *buf, size_t bytes)
1213 {
1214 	size_t i;
1215 
1216 	for (i = 0; i < bytes; i += sizeof(long)) {
1217 		if (*(const long *)(buf + i) != 0)
1218 			return (0);
1219 	}
1220 	return (1);
1221 }
1222 
1223 /*
1224  * Function to "write" a block that contains only zeros.
1225  */
1226 static
1227 void
1228 zero_write(struct buf *bp, hammer2_trans_t *trans, hammer2_inode_t *ip,
1229 	hammer2_inode_data_t *ipdata, hammer2_chain_t **parentp,
1230 	hammer2_key_t lbase, int *errorp __unused)
1231 {
1232 	hammer2_chain_t *parent;
1233 	hammer2_chain_t *chain;
1234 	hammer2_key_t key_dummy;
1235 	int cache_index = -1;
1236 
1237 	parent = hammer2_chain_lookup_init(*parentp, 0);
1238 
1239 	chain = hammer2_chain_lookup(&parent, &key_dummy, lbase, lbase,
1240 				     &cache_index, HAMMER2_LOOKUP_NODATA);
1241 	if (chain) {
1242 		if (chain->bref.type == HAMMER2_BREF_TYPE_INODE) {
1243 			bzero(chain->data->ipdata.u.data,
1244 			      HAMMER2_EMBEDDED_BYTES);
1245 		} else {
1246 			hammer2_chain_delete(trans, chain, 0);
1247 		}
1248 		hammer2_chain_unlock(chain);
1249 	}
1250 	hammer2_chain_lookup_done(parent);
1251 }
1252 
1253 /*
1254  * Function to write the data as it is, without performing any sort of
1255  * compression. This function is used in path without compression and
1256  * default zero-checking path.
1257  */
1258 static
1259 void
1260 hammer2_write_bp(hammer2_chain_t *chain, struct buf *bp, int ioflag,
1261 				int pblksize, int *errorp)
1262 {
1263 	hammer2_off_t pbase;
1264 	hammer2_off_t pmask;
1265 	hammer2_off_t peof;
1266 	struct buf *dbp;
1267 	size_t boff;
1268 	size_t psize;
1269 	int error;
1270 	int temp_check = HAMMER2_DEC_CHECK(chain->bref.methods);
1271 
1272 	KKASSERT(chain->flags & HAMMER2_CHAIN_MODIFIED);
1273 
1274 	switch(chain->bref.type) {
1275 	case HAMMER2_BREF_TYPE_INODE:
1276 		KKASSERT(chain->data->ipdata.op_flags &
1277 			 HAMMER2_OPFLAG_DIRECTDATA);
1278 		KKASSERT(bp->b_loffset == 0);
1279 		bcopy(bp->b_data, chain->data->ipdata.u.data,
1280 		      HAMMER2_EMBEDDED_BYTES);
1281 		error = 0;
1282 		break;
1283 	case HAMMER2_BREF_TYPE_DATA:
1284 		psize = hammer2_devblksize(chain->bytes);
1285 		pmask = (hammer2_off_t)psize - 1;
1286 		pbase = chain->bref.data_off & ~pmask;
1287 		boff = chain->bref.data_off & (HAMMER2_OFF_MASK & pmask);
1288 		peof = (pbase + HAMMER2_SEGMASK64) & ~HAMMER2_SEGMASK64;
1289 
1290 		if (psize == pblksize) {
1291 			dbp = getblk(chain->hmp->devvp, pbase,
1292 				     psize, 0, 0);
1293 			error = 0;
1294 		} else {
1295 			error = bread(chain->hmp->devvp, pbase, psize, &dbp);
1296 			if (error) {
1297 				kprintf("hammer2: WRITE PATH: "
1298 					"dbp bread error\n");
1299 				break;
1300 			}
1301 		}
1302 
1303 		chain->bref.methods = HAMMER2_ENC_COMP(HAMMER2_COMP_NONE) +
1304 				      HAMMER2_ENC_CHECK(temp_check);
1305 		bcopy(bp->b_data, dbp->b_data + boff, chain->bytes);
1306 
1307 		/*
1308 		 * Device buffer is now valid, chain is no
1309 		 * longer in the initial state.
1310 		 */
1311 		atomic_clear_int(&chain->flags, HAMMER2_CHAIN_INITIAL);
1312 
1313 		if (ioflag & IO_SYNC) {
1314 			/*
1315 			 * Synchronous I/O requested.
1316 			 */
1317 			bwrite(dbp);
1318 		/*
1319 		} else if ((ioflag & IO_DIRECT) && loff + n == pblksize) {
1320 			bdwrite(dbp);
1321 		*/
1322 		} else if (ioflag & IO_ASYNC) {
1323 			bawrite(dbp);
1324 		} else if (hammer2_cluster_enable) {
1325 			cluster_write(dbp, peof, HAMMER2_PBUFSIZE, 4/*XXX*/);
1326 		} else {
1327 			bdwrite(dbp);
1328 		}
1329 		break;
1330 	default:
1331 		panic("hammer2_write_bp: bad chain type %d\n",
1332 		      chain->bref.type);
1333 		/* NOT REACHED */
1334 		error = 0;
1335 		break;
1336 	}
1337 	*errorp = error;
1338 }
1339 
1340 static
1341 int
1342 hammer2_remount(struct mount *mp, char *path, struct vnode *devvp,
1343                 struct ucred *cred)
1344 {
1345 	return (0);
1346 }
1347 
1348 static
1349 int
1350 hammer2_vfs_unmount(struct mount *mp, int mntflags)
1351 {
1352 	hammer2_pfsmount_t *pmp;
1353 	hammer2_mount_t *hmp;
1354 	hammer2_cluster_t *cluster;
1355 	int flags;
1356 	int error = 0;
1357 	int ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
1358 	int dumpcnt;
1359 	struct vnode *devvp;
1360 
1361 	pmp = MPTOPMP(mp);
1362 	cluster = pmp->mount_cluster;
1363 	hmp = cluster->hmp;
1364 	flags = 0;
1365 
1366 	if (mntflags & MNT_FORCE)
1367 		flags |= FORCECLOSE;
1368 
1369 	hammer2_mount_exlock(hmp);
1370 
1371 	/*
1372 	 * If mount initialization proceeded far enough we must flush
1373 	 * its vnodes.
1374 	 */
1375 	if (pmp->iroot)
1376 		error = vflush(mp, 0, flags);
1377 
1378 	if (error) {
1379 		hammer2_mount_unlock(hmp);
1380 		return error;
1381 	}
1382 
1383 	lockmgr(&hammer2_mntlk, LK_EXCLUSIVE);
1384 	--hmp->pmp_count;
1385 	kprintf("hammer2_unmount hmp=%p pmpcnt=%d\n", hmp, hmp->pmp_count);
1386 
1387 	/*
1388 	 * Flush any left over chains.  The voldata lock is only used
1389 	 * to synchronize against HAMMER2_CHAIN_MODIFIED_AUX.
1390 	 */
1391 	hammer2_voldata_lock(hmp);
1392 	if ((hmp->vchain.flags | hmp->fchain.flags) &
1393 	    (HAMMER2_CHAIN_MODIFIED | HAMMER2_CHAIN_SUBMODIFIED)) {
1394 		hammer2_voldata_unlock(hmp, 0);
1395 		hammer2_vfs_sync(mp, MNT_WAIT);
1396 		hammer2_vfs_sync(mp, MNT_WAIT);
1397 	} else {
1398 		hammer2_voldata_unlock(hmp, 0);
1399 	}
1400 	if (hmp->pmp_count == 0) {
1401 		if (hmp->vchain.flags & (HAMMER2_CHAIN_MODIFIED |
1402 					 HAMMER2_CHAIN_SUBMODIFIED)) {
1403 			kprintf("hammer2_unmount: chains left over after "
1404 				"final sync\n");
1405 			if (hammer2_debug & 0x0010)
1406 				Debugger("entered debugger");
1407 		}
1408 	}
1409 
1410 	/*
1411 	 * Cleanup the root and super-root chain elements (which should be
1412 	 * clean).
1413 	 */
1414 	if (pmp->iroot) {
1415 #if REPORT_REFS_ERRORS
1416 		if (pmp->iroot->refs != 1)
1417 			kprintf("PMP->IROOT %p REFS WRONG %d\n",
1418 				pmp->iroot, pmp->iroot->refs);
1419 #else
1420 		KKASSERT(pmp->iroot->refs == 1);
1421 #endif
1422 		hammer2_inode_drop(pmp->iroot);	    /* ref for pmp->iroot */
1423 		pmp->iroot = NULL;
1424 	}
1425 	if (cluster->rchain) {
1426 		atomic_clear_int(&cluster->rchain->flags,
1427 				 HAMMER2_CHAIN_MOUNTED);
1428 #if REPORT_REFS_ERRORS
1429 		if (cluster->rchain->refs != 1)
1430 			kprintf("PMP->RCHAIN %p REFS WRONG %d\n",
1431 				cluster->rchain, cluster->rchain->refs);
1432 #else
1433 		KKASSERT(cluster->rchain->refs == 1);
1434 #endif
1435 		hammer2_chain_drop(cluster->rchain);
1436 		cluster->rchain = NULL;
1437 	}
1438 	ccms_domain_uninit(&pmp->ccms_dom);
1439 
1440 	/*
1441 	 * Kill cluster controller
1442 	 */
1443 	kdmsg_iocom_uninit(&pmp->iocom);
1444 
1445 	/*
1446 	 * If no PFS's left drop the master hammer2_mount for the device.
1447 	 */
1448 	if (hmp->pmp_count == 0) {
1449 		if (hmp->sroot) {
1450 			hammer2_inode_drop(hmp->sroot);
1451 			hmp->sroot = NULL;
1452 		}
1453 		if (hmp->schain) {
1454 #if REPORT_REFS_ERRORS
1455 			if (hmp->schain->refs != 1)
1456 				kprintf("HMP->SCHAIN %p REFS WRONG %d\n",
1457 					hmp->schain, hmp->schain->refs);
1458 #else
1459 			KKASSERT(hmp->schain->refs == 1);
1460 #endif
1461 			hammer2_chain_drop(hmp->schain);
1462 			hmp->schain = NULL;
1463 		}
1464 
1465 		/*
1466 		 * Finish up with the device vnode
1467 		 */
1468 		if ((devvp = hmp->devvp) != NULL) {
1469 			vinvalbuf(devvp, (ronly ? 0 : V_SAVE), 0, 0);
1470 			hmp->devvp = NULL;
1471 			VOP_CLOSE(devvp, (ronly ? FREAD : FREAD|FWRITE));
1472 			vrele(devvp);
1473 			devvp = NULL;
1474 		}
1475 
1476 		/*
1477 		 * Final drop of embedded freemap root chain to clean up
1478 		 * fchain.core (fchain structure is not flagged ALLOCATED
1479 		 * so it is cleaned out and then left to rot).
1480 		 */
1481 		hammer2_chain_drop(&hmp->fchain);
1482 
1483 		/*
1484 		 * Final drop of embedded volume root chain to clean up
1485 		 * vchain.core (vchain structure is not flagged ALLOCATED
1486 		 * so it is cleaned out and then left to rot).
1487 		 */
1488 		dumpcnt = 50;
1489 		hammer2_dump_chain(&hmp->vchain, 0, &dumpcnt);
1490 		hammer2_mount_unlock(hmp);
1491 		hammer2_chain_drop(&hmp->vchain);
1492 	} else {
1493 		hammer2_mount_unlock(hmp);
1494 	}
1495 
1496 	pmp->mp = NULL;
1497 	mp->mnt_data = NULL;
1498 
1499 	pmp->mount_cluster = NULL;
1500 	pmp->cluster = NULL;		/* XXX */
1501 
1502 	kmalloc_destroy(&pmp->mmsg);
1503 	kmalloc_destroy(&pmp->minode);
1504 
1505 	cluster->hmp = NULL;
1506 
1507 	kfree(cluster, M_HAMMER2);
1508 	kfree(pmp, M_HAMMER2);
1509 	if (hmp->pmp_count == 0) {
1510 		mtx_lock(&hmp->wthread_mtx);
1511 		hmp->wthread_destroy = 1;
1512 		wakeup(&hmp->wthread_bioq);
1513 		while (hmp->wthread_destroy != -1) {
1514 			mtxsleep(&hmp->wthread_destroy, &hmp->wthread_mtx, 0,
1515 				"umount-sleep",	0);
1516 		}
1517 		mtx_unlock(&hmp->wthread_mtx);
1518 
1519 		TAILQ_REMOVE(&hammer2_mntlist, hmp, mntentry);
1520 		kmalloc_destroy(&hmp->mchain);
1521 		kfree(hmp, M_HAMMER2);
1522 	}
1523 	lockmgr(&hammer2_mntlk, LK_RELEASE);
1524 
1525 	return (error);
1526 }
1527 
1528 static
1529 int
1530 hammer2_vfs_vget(struct mount *mp, struct vnode *dvp,
1531 	     ino_t ino, struct vnode **vpp)
1532 {
1533 	kprintf("hammer2_vget\n");
1534 	return (EOPNOTSUPP);
1535 }
1536 
1537 static
1538 int
1539 hammer2_vfs_root(struct mount *mp, struct vnode **vpp)
1540 {
1541 	hammer2_pfsmount_t *pmp;
1542 	hammer2_chain_t *parent;
1543 	int error;
1544 	struct vnode *vp;
1545 
1546 	pmp = MPTOPMP(mp);
1547 	if (pmp->iroot == NULL) {
1548 		*vpp = NULL;
1549 		error = EINVAL;
1550 	} else {
1551 		parent = hammer2_inode_lock_sh(pmp->iroot);
1552 		vp = hammer2_igetv(pmp->iroot, &error);
1553 		hammer2_inode_unlock_sh(pmp->iroot, parent);
1554 		*vpp = vp;
1555 		if (vp == NULL)
1556 			kprintf("vnodefail\n");
1557 	}
1558 
1559 	return (error);
1560 }
1561 
1562 /*
1563  * Filesystem status
1564  *
1565  * XXX incorporate ipdata->inode_quota and data_quota
1566  */
1567 static
1568 int
1569 hammer2_vfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
1570 {
1571 	hammer2_pfsmount_t *pmp;
1572 	hammer2_mount_t *hmp;
1573 
1574 	pmp = MPTOPMP(mp);
1575 	hmp = MPTOHMP(mp);
1576 
1577 	mp->mnt_stat.f_files = pmp->inode_count;
1578 	mp->mnt_stat.f_ffree = 0;
1579 	mp->mnt_stat.f_blocks = hmp->voldata.allocator_size / HAMMER2_PBUFSIZE;
1580 	mp->mnt_stat.f_bfree =  hmp->voldata.allocator_free / HAMMER2_PBUFSIZE;
1581 	mp->mnt_stat.f_bavail = mp->mnt_stat.f_bfree;
1582 
1583 	*sbp = mp->mnt_stat;
1584 	return (0);
1585 }
1586 
1587 static
1588 int
1589 hammer2_vfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred)
1590 {
1591 	hammer2_pfsmount_t *pmp;
1592 	hammer2_mount_t *hmp;
1593 
1594 	pmp = MPTOPMP(mp);
1595 	hmp = MPTOHMP(mp);
1596 
1597 	mp->mnt_vstat.f_bsize = HAMMER2_PBUFSIZE;
1598 	mp->mnt_vstat.f_files = pmp->inode_count;
1599 	mp->mnt_vstat.f_ffree = 0;
1600 	mp->mnt_vstat.f_blocks = hmp->voldata.allocator_size / HAMMER2_PBUFSIZE;
1601 	mp->mnt_vstat.f_bfree =  hmp->voldata.allocator_free / HAMMER2_PBUFSIZE;
1602 	mp->mnt_vstat.f_bavail = mp->mnt_vstat.f_bfree;
1603 
1604 	*sbp = mp->mnt_vstat;
1605 	return (0);
1606 }
1607 
1608 /*
1609  * Sync the entire filesystem; this is called from the filesystem syncer
1610  * process periodically and whenever a user calls sync(1) on the hammer
1611  * mountpoint.
1612  *
1613  * Currently is actually called from the syncer! \o/
1614  *
1615  * This task will have to snapshot the state of the dirty inode chain.
1616  * From that, it will have to make sure all of the inodes on the dirty
1617  * chain have IO initiated. We make sure that io is initiated for the root
1618  * block.
1619  *
1620  * If waitfor is set, we wait for media to acknowledge the new rootblock.
1621  *
1622  * THINKS: side A vs side B, to have sync not stall all I/O?
1623  */
1624 static
1625 int
1626 hammer2_vfs_sync(struct mount *mp, int waitfor)
1627 {
1628 	struct hammer2_sync_info info;
1629 	hammer2_pfsmount_t *pmp;
1630 	hammer2_cluster_t *cluster;
1631 	hammer2_mount_t *hmp;
1632 	int flags;
1633 	int error;
1634 	int i;
1635 
1636 	pmp = MPTOPMP(mp);
1637 
1638 	/*
1639 	 * We can't acquire locks on existing vnodes while in a transaction
1640 	 * without risking a deadlock.  This assumes that vfsync() can be
1641 	 * called without the vnode locked (which it can in DragonFly).
1642 	 * Otherwise we'd have to implement a multi-pass or flag the lock
1643 	 * failures and retry.
1644 	 */
1645 	/*flags = VMSC_GETVP;*/
1646 	flags = 0;
1647 	if (waitfor & MNT_LAZY)
1648 		flags |= VMSC_ONEPASS;
1649 
1650 	hammer2_trans_init(&info.trans, pmp, HAMMER2_TRANS_ISFLUSH);
1651 
1652 	info.error = 0;
1653 	info.waitfor = MNT_NOWAIT;
1654 	vmntvnodescan(mp, flags | VMSC_NOWAIT,
1655 		      hammer2_sync_scan1,
1656 		      hammer2_sync_scan2, &info);
1657 	if (info.error == 0 && (waitfor & MNT_WAIT)) {
1658 		info.waitfor = waitfor;
1659 		    vmntvnodescan(mp, flags,
1660 				  hammer2_sync_scan1,
1661 				  hammer2_sync_scan2, &info);
1662 
1663 	}
1664 #if 0
1665 	if (waitfor == MNT_WAIT) {
1666 		/* XXX */
1667 	} else {
1668 		/* XXX */
1669 	}
1670 #endif
1671 
1672 	cluster = pmp->cluster;
1673 	hmp = cluster->hmp;
1674 
1675 	hammer2_chain_lock(&hmp->vchain, HAMMER2_RESOLVE_ALWAYS);
1676 	if (hmp->vchain.flags & (HAMMER2_CHAIN_MODIFIED |
1677 				  HAMMER2_CHAIN_SUBMODIFIED)) {
1678 		hammer2_chain_flush(&info.trans, &hmp->vchain);
1679 	}
1680 	hammer2_chain_unlock(&hmp->vchain);
1681 
1682 #if 1
1683 	/*
1684 	 * Rollup flush.  The fsyncs above basically just flushed
1685 	 * data blocks.  The flush below gets all the meta-data.
1686 	 */
1687 	hammer2_chain_lock(&hmp->fchain, HAMMER2_RESOLVE_ALWAYS);
1688 	if (hmp->fchain.flags & (HAMMER2_CHAIN_MODIFIED |
1689 				 HAMMER2_CHAIN_SUBMODIFIED)) {
1690 		/* this will modify vchain as a side effect */
1691 		hammer2_chain_flush(&info.trans, &hmp->fchain);
1692 	}
1693 	hammer2_chain_unlock(&hmp->fchain);
1694 #endif
1695 
1696 
1697 	error = 0;
1698 
1699 	/*
1700 	 * We can't safely flush the volume header until we have
1701 	 * flushed any device buffers which have built up.
1702 	 *
1703 	 * XXX this isn't being incremental
1704 	 */
1705 	vn_lock(hmp->devvp, LK_EXCLUSIVE | LK_RETRY);
1706 	error = VOP_FSYNC(hmp->devvp, MNT_WAIT, 0);
1707 	vn_unlock(hmp->devvp);
1708 
1709 	/*
1710 	 * The flush code sets CHAIN_VOLUMESYNC to indicate that the
1711 	 * volume header needs synchronization via hmp->volsync.
1712 	 *
1713 	 * XXX synchronize the flag & data with only this flush XXX
1714 	 */
1715 	if (error == 0 && (hmp->vchain.flags & HAMMER2_CHAIN_VOLUMESYNC)) {
1716 		struct buf *bp;
1717 
1718 		/*
1719 		 * Synchronize the disk before flushing the volume
1720 		 * header.
1721 		 */
1722 		bp = getpbuf(NULL);
1723 		bp->b_bio1.bio_offset = 0;
1724 		bp->b_bufsize = 0;
1725 		bp->b_bcount = 0;
1726 		bp->b_cmd = BUF_CMD_FLUSH;
1727 		bp->b_bio1.bio_done = biodone_sync;
1728 		bp->b_bio1.bio_flags |= BIO_SYNC;
1729 		vn_strategy(hmp->devvp, &bp->b_bio1);
1730 		biowait(&bp->b_bio1, "h2vol");
1731 		relpbuf(bp, NULL);
1732 
1733 		/*
1734 		 * Then we can safely flush the version of the volume header
1735 		 * synchronized by the flush code.
1736 		 */
1737 		i = hmp->volhdrno + 1;
1738 		if (i >= HAMMER2_NUM_VOLHDRS)
1739 			i = 0;
1740 		if (i * HAMMER2_ZONE_BYTES64 + HAMMER2_SEGSIZE >
1741 		    hmp->volsync.volu_size) {
1742 			i = 0;
1743 		}
1744 		kprintf("sync volhdr %d %jd\n",
1745 			i, (intmax_t)hmp->volsync.volu_size);
1746 		bp = getblk(hmp->devvp, i * HAMMER2_ZONE_BYTES64,
1747 			    HAMMER2_PBUFSIZE, 0, 0);
1748 		atomic_clear_int(&hmp->vchain.flags, HAMMER2_CHAIN_VOLUMESYNC);
1749 		bcopy(&hmp->volsync, bp->b_data, HAMMER2_PBUFSIZE);
1750 		bawrite(bp);
1751 		hmp->volhdrno = i;
1752 	}
1753 	hammer2_trans_done(&info.trans);
1754 	return (error);
1755 }
1756 
1757 /*
1758  * Sync passes.
1759  *
1760  * NOTE: We don't test SUBMODIFIED or MOVED here because the fsync code
1761  *	 won't flush on those flags.  The syncer code above will do a
1762  *	 general meta-data flush globally that will catch these flags.
1763  */
1764 static int
1765 hammer2_sync_scan1(struct mount *mp, struct vnode *vp, void *data)
1766 {
1767 	hammer2_inode_t *ip;
1768 
1769 	ip = VTOI(vp);
1770 	if (vp->v_type == VNON || ip == NULL ||
1771 	    ((ip->flags & HAMMER2_INODE_MODIFIED) == 0 &&
1772 	     RB_EMPTY(&vp->v_rbdirty_tree))) {
1773 		return(-1);
1774 	}
1775 	return(0);
1776 }
1777 
1778 static int
1779 hammer2_sync_scan2(struct mount *mp, struct vnode *vp, void *data)
1780 {
1781 	struct hammer2_sync_info *info = data;
1782 	hammer2_inode_t *ip;
1783 	hammer2_chain_t *parent;
1784 	int error;
1785 
1786 	ip = VTOI(vp);
1787 	if (vp->v_type == VNON || vp->v_type == VBAD ||
1788 	    ((ip->flags & HAMMER2_INODE_MODIFIED) == 0 &&
1789 	     RB_EMPTY(&vp->v_rbdirty_tree))) {
1790 		return(0);
1791 	}
1792 
1793 	/*
1794 	 * VOP_FSYNC will start a new transaction so replicate some code
1795 	 * here to do it inline (see hammer2_vop_fsync()).
1796 	 *
1797 	 * WARNING: The vfsync interacts with the buffer cache and might
1798 	 *          block, we can't hold the inode lock at that time.
1799 	 */
1800 	atomic_clear_int(&ip->flags, HAMMER2_INODE_MODIFIED);
1801 	if (ip->vp)
1802 		vfsync(ip->vp, MNT_NOWAIT, 1, NULL, NULL);
1803 	parent = hammer2_inode_lock_ex(ip);
1804 	hammer2_chain_flush(&info->trans, parent);
1805 	hammer2_inode_unlock_ex(ip, parent);
1806 	error = 0;
1807 #if 0
1808 	error = VOP_FSYNC(vp, MNT_NOWAIT, 0);
1809 #endif
1810 	if (error)
1811 		info->error = error;
1812 	return(0);
1813 }
1814 
1815 static
1816 int
1817 hammer2_vfs_vptofh(struct vnode *vp, struct fid *fhp)
1818 {
1819 	return (0);
1820 }
1821 
1822 static
1823 int
1824 hammer2_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
1825 	       struct fid *fhp, struct vnode **vpp)
1826 {
1827 	return (0);
1828 }
1829 
1830 static
1831 int
1832 hammer2_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
1833 		 int *exflagsp, struct ucred **credanonp)
1834 {
1835 	return (0);
1836 }
1837 
1838 /*
1839  * Support code for hammer2_mount().  Read, verify, and install the volume
1840  * header into the HMP
1841  *
1842  * XXX read four volhdrs and use the one with the highest TID whos CRC
1843  *     matches.
1844  *
1845  * XXX check iCRCs.
1846  *
1847  * XXX For filesystems w/ less than 4 volhdrs, make sure to not write to
1848  *     nonexistant locations.
1849  *
1850  * XXX Record selected volhdr and ring updates to each of 4 volhdrs
1851  */
1852 static
1853 int
1854 hammer2_install_volume_header(hammer2_mount_t *hmp)
1855 {
1856 	hammer2_volume_data_t *vd;
1857 	struct buf *bp;
1858 	hammer2_crc32_t crc0, crc, bcrc0, bcrc;
1859 	int error_reported;
1860 	int error;
1861 	int valid;
1862 	int i;
1863 
1864 	error_reported = 0;
1865 	error = 0;
1866 	valid = 0;
1867 	bp = NULL;
1868 
1869 	/*
1870 	 * There are up to 4 copies of the volume header (syncs iterate
1871 	 * between them so there is no single master).  We don't trust the
1872 	 * volu_size field so we don't know precisely how large the filesystem
1873 	 * is, so depend on the OS to return an error if we go beyond the
1874 	 * block device's EOF.
1875 	 */
1876 	for (i = 0; i < HAMMER2_NUM_VOLHDRS; i++) {
1877 		error = bread(hmp->devvp, i * HAMMER2_ZONE_BYTES64,
1878 			      HAMMER2_VOLUME_BYTES, &bp);
1879 		if (error) {
1880 			brelse(bp);
1881 			bp = NULL;
1882 			continue;
1883 		}
1884 
1885 		vd = (struct hammer2_volume_data *) bp->b_data;
1886 		if ((vd->magic != HAMMER2_VOLUME_ID_HBO) &&
1887 		    (vd->magic != HAMMER2_VOLUME_ID_ABO)) {
1888 			brelse(bp);
1889 			bp = NULL;
1890 			continue;
1891 		}
1892 
1893 		if (vd->magic == HAMMER2_VOLUME_ID_ABO) {
1894 			/* XXX: Reversed-endianness filesystem */
1895 			kprintf("hammer2: reverse-endian filesystem detected");
1896 			brelse(bp);
1897 			bp = NULL;
1898 			continue;
1899 		}
1900 
1901 		crc = vd->icrc_sects[HAMMER2_VOL_ICRC_SECT0];
1902 		crc0 = hammer2_icrc32(bp->b_data + HAMMER2_VOLUME_ICRC0_OFF,
1903 				      HAMMER2_VOLUME_ICRC0_SIZE);
1904 		bcrc = vd->icrc_sects[HAMMER2_VOL_ICRC_SECT1];
1905 		bcrc0 = hammer2_icrc32(bp->b_data + HAMMER2_VOLUME_ICRC1_OFF,
1906 				       HAMMER2_VOLUME_ICRC1_SIZE);
1907 		if ((crc0 != crc) || (bcrc0 != bcrc)) {
1908 			kprintf("hammer2 volume header crc "
1909 				"mismatch copy #%d %08x/%08x\n",
1910 				i, crc0, crc);
1911 			error_reported = 1;
1912 			brelse(bp);
1913 			bp = NULL;
1914 			continue;
1915 		}
1916 		if (valid == 0 || hmp->voldata.mirror_tid < vd->mirror_tid) {
1917 			valid = 1;
1918 			hmp->voldata = *vd;
1919 			hmp->volhdrno = i;
1920 		}
1921 		brelse(bp);
1922 		bp = NULL;
1923 	}
1924 	if (valid) {
1925 		hmp->volsync = hmp->voldata;
1926 		error = 0;
1927 		if (error_reported || bootverbose || 1) { /* 1/DEBUG */
1928 			kprintf("hammer2: using volume header #%d\n",
1929 				hmp->volhdrno);
1930 		}
1931 	} else {
1932 		error = EINVAL;
1933 		kprintf("hammer2: no valid volume headers found!\n");
1934 	}
1935 	return (error);
1936 }
1937 
1938 /*
1939  * Reconnect using the passed file pointer.  The caller must ref the
1940  * fp for us.
1941  */
1942 void
1943 hammer2_cluster_reconnect(hammer2_pfsmount_t *pmp, struct file *fp)
1944 {
1945 	hammer2_inode_data_t *ipdata;
1946 	hammer2_chain_t *parent;
1947 	hammer2_mount_t *hmp;
1948 	size_t name_len;
1949 
1950 	hmp = pmp->mount_cluster->hmp;
1951 
1952 	/*
1953 	 * Closes old comm descriptor, kills threads, cleans up
1954 	 * states, then installs the new descriptor and creates
1955 	 * new threads.
1956 	 */
1957 	kdmsg_iocom_reconnect(&pmp->iocom, fp, "hammer2");
1958 
1959 	/*
1960 	 * Setup LNK_CONN fields for autoinitiated state machine
1961 	 */
1962 	parent = hammer2_inode_lock_ex(pmp->iroot);
1963 	ipdata = &parent->data->ipdata;
1964 	pmp->iocom.auto_lnk_conn.pfs_clid = ipdata->pfs_clid;
1965 	pmp->iocom.auto_lnk_conn.pfs_fsid = ipdata->pfs_fsid;
1966 	pmp->iocom.auto_lnk_conn.pfs_type = ipdata->pfs_type;
1967 	pmp->iocom.auto_lnk_conn.proto_version = DMSG_SPAN_PROTO_1;
1968 	pmp->iocom.auto_lnk_conn.peer_type = hmp->voldata.peer_type;
1969 
1970 	/*
1971 	 * Filter adjustment.  Clients do not need visibility into other
1972 	 * clients (otherwise millions of clients would present a serious
1973 	 * problem).  The fs_label also serves to restrict the namespace.
1974 	 */
1975 	pmp->iocom.auto_lnk_conn.peer_mask = 1LLU << HAMMER2_PEER_HAMMER2;
1976 	pmp->iocom.auto_lnk_conn.pfs_mask = (uint64_t)-1;
1977 	switch (ipdata->pfs_type) {
1978 	case DMSG_PFSTYPE_CLIENT:
1979 		pmp->iocom.auto_lnk_conn.peer_mask &=
1980 				~(1LLU << DMSG_PFSTYPE_CLIENT);
1981 		break;
1982 	default:
1983 		break;
1984 	}
1985 
1986 	name_len = ipdata->name_len;
1987 	if (name_len >= sizeof(pmp->iocom.auto_lnk_conn.fs_label))
1988 		name_len = sizeof(pmp->iocom.auto_lnk_conn.fs_label) - 1;
1989 	bcopy(ipdata->filename,
1990 	      pmp->iocom.auto_lnk_conn.fs_label,
1991 	      name_len);
1992 	pmp->iocom.auto_lnk_conn.fs_label[name_len] = 0;
1993 
1994 	/*
1995 	 * Setup LNK_SPAN fields for autoinitiated state machine
1996 	 */
1997 	pmp->iocom.auto_lnk_span.pfs_clid = ipdata->pfs_clid;
1998 	pmp->iocom.auto_lnk_span.pfs_fsid = ipdata->pfs_fsid;
1999 	pmp->iocom.auto_lnk_span.pfs_type = ipdata->pfs_type;
2000 	pmp->iocom.auto_lnk_span.peer_type = hmp->voldata.peer_type;
2001 	pmp->iocom.auto_lnk_span.proto_version = DMSG_SPAN_PROTO_1;
2002 	name_len = ipdata->name_len;
2003 	if (name_len >= sizeof(pmp->iocom.auto_lnk_span.fs_label))
2004 		name_len = sizeof(pmp->iocom.auto_lnk_span.fs_label) - 1;
2005 	bcopy(ipdata->filename,
2006 	      pmp->iocom.auto_lnk_span.fs_label,
2007 	      name_len);
2008 	pmp->iocom.auto_lnk_span.fs_label[name_len] = 0;
2009 	hammer2_inode_unlock_ex(pmp->iroot, parent);
2010 
2011 	kdmsg_iocom_autoinitiate(&pmp->iocom, hammer2_autodmsg);
2012 }
2013 
2014 static int
2015 hammer2_rcvdmsg(kdmsg_msg_t *msg)
2016 {
2017 	switch(msg->any.head.cmd & DMSGF_TRANSMASK) {
2018 	case DMSG_DBG_SHELL:
2019 		/*
2020 		 * (non-transaction)
2021 		 * Execute shell command (not supported atm)
2022 		 */
2023 		kdmsg_msg_reply(msg, DMSG_ERR_NOSUPP);
2024 		break;
2025 	case DMSG_DBG_SHELL | DMSGF_REPLY:
2026 		/*
2027 		 * (non-transaction)
2028 		 */
2029 		if (msg->aux_data) {
2030 			msg->aux_data[msg->aux_size - 1] = 0;
2031 			kprintf("HAMMER2 DBG: %s\n", msg->aux_data);
2032 		}
2033 		break;
2034 	default:
2035 		/*
2036 		 * Unsupported message received.  We only need to
2037 		 * reply if it's a transaction in order to close our end.
2038 		 * Ignore any one-way messages are any further messages
2039 		 * associated with the transaction.
2040 		 *
2041 		 * NOTE: This case also includes DMSG_LNK_ERROR messages
2042 		 *	 which might be one-way, replying to those would
2043 		 *	 cause an infinite ping-pong.
2044 		 */
2045 		if (msg->any.head.cmd & DMSGF_CREATE)
2046 			kdmsg_msg_reply(msg, DMSG_ERR_NOSUPP);
2047 		break;
2048 	}
2049 	return(0);
2050 }
2051 
2052 /*
2053  * This function is called after KDMSG has automatically handled processing
2054  * of a LNK layer message (typically CONN, SPAN, or CIRC).
2055  *
2056  * We tag off the LNK_CONN to trigger our LNK_VOLCONF messages which
2057  * advertises all available hammer2 super-root volumes.
2058  */
2059 static void
2060 hammer2_autodmsg(kdmsg_msg_t *msg)
2061 {
2062 	hammer2_pfsmount_t *pmp = msg->iocom->handle;
2063 	hammer2_mount_t *hmp = pmp->mount_cluster->hmp;
2064 	int copyid;
2065 
2066 	/*
2067 	 * We only care about replies to our LNK_CONN auto-request.  kdmsg
2068 	 * has already processed the reply, we use this calback as a shim
2069 	 * to know when we can advertise available super-root volumes.
2070 	 */
2071 	if ((msg->any.head.cmd & DMSGF_TRANSMASK) !=
2072 	    (DMSG_LNK_CONN | DMSGF_CREATE | DMSGF_REPLY) ||
2073 	    msg->state == NULL) {
2074 		return;
2075 	}
2076 
2077 	kprintf("LNK_CONN REPLY RECEIVED CMD %08x\n", msg->any.head.cmd);
2078 
2079 	if (msg->any.head.cmd & DMSGF_CREATE) {
2080 		kprintf("HAMMER2: VOLDATA DUMP\n");
2081 
2082 		/*
2083 		 * Dump the configuration stored in the volume header
2084 		 */
2085 		hammer2_voldata_lock(hmp);
2086 		for (copyid = 0; copyid < HAMMER2_COPYID_COUNT; ++copyid) {
2087 			if (hmp->voldata.copyinfo[copyid].copyid == 0)
2088 				continue;
2089 			hammer2_volconf_update(pmp, copyid);
2090 		}
2091 		hammer2_voldata_unlock(hmp, 0);
2092 	}
2093 	if ((msg->any.head.cmd & DMSGF_DELETE) &&
2094 	    msg->state && (msg->state->txcmd & DMSGF_DELETE) == 0) {
2095 		kprintf("HAMMER2: CONN WAS TERMINATED\n");
2096 	}
2097 }
2098 
2099 /*
2100  * Volume configuration updates are passed onto the userland service
2101  * daemon via the open LNK_CONN transaction.
2102  */
2103 void
2104 hammer2_volconf_update(hammer2_pfsmount_t *pmp, int index)
2105 {
2106 	hammer2_mount_t *hmp = pmp->mount_cluster->hmp;
2107 	kdmsg_msg_t *msg;
2108 
2109 	/* XXX interlock against connection state termination */
2110 	kprintf("volconf update %p\n", pmp->iocom.conn_state);
2111 	if (pmp->iocom.conn_state) {
2112 		kprintf("TRANSMIT VOLCONF VIA OPEN CONN TRANSACTION\n");
2113 		msg = kdmsg_msg_alloc_state(pmp->iocom.conn_state,
2114 					    DMSG_LNK_VOLCONF, NULL, NULL);
2115 		msg->any.lnk_volconf.copy = hmp->voldata.copyinfo[index];
2116 		msg->any.lnk_volconf.mediaid = hmp->voldata.fsid;
2117 		msg->any.lnk_volconf.index = index;
2118 		kdmsg_msg_write(msg);
2119 	}
2120 }
2121 
2122 void
2123 hammer2_dump_chain(hammer2_chain_t *chain, int tab, int *countp)
2124 {
2125 	hammer2_chain_layer_t *layer;
2126 	hammer2_chain_t *scan;
2127 	hammer2_chain_t *first_parent;
2128 
2129 	--*countp;
2130 	if (*countp == 0) {
2131 		kprintf("%*.*s...\n", tab, tab, "");
2132 		return;
2133 	}
2134 	if (*countp < 0)
2135 		return;
2136 	first_parent = chain->core ? TAILQ_FIRST(&chain->core->ownerq) : NULL;
2137 	kprintf("%*.*schain %p.%d [%08x][core=%p fp=%p] (%s) np=%p dt=%s refs=%d",
2138 		tab, tab, "",
2139 		chain, chain->bref.type, chain->flags,
2140 		chain->core,
2141 		first_parent,
2142 		((chain->bref.type == HAMMER2_BREF_TYPE_INODE &&
2143 		chain->data) ?  (char *)chain->data->ipdata.filename : "?"),
2144 		(first_parent ? TAILQ_NEXT(chain, core_entry) : NULL),
2145 		(chain->delete_tid == HAMMER2_MAX_TID ? "max" : "fls"),
2146 		chain->refs);
2147 	if (first_parent)
2148 		kprintf(" [fpflags %08x fprefs %d\n",
2149 			first_parent->flags,
2150 			first_parent->refs);
2151 	if (chain->core == NULL || TAILQ_EMPTY(&chain->core->layerq))
2152 		kprintf("\n");
2153 	else
2154 		kprintf(" {\n");
2155 	TAILQ_FOREACH(layer, &chain->core->layerq, entry) {
2156 		RB_FOREACH(scan, hammer2_chain_tree, &layer->rbtree) {
2157 			hammer2_dump_chain(scan, tab + 4, countp);
2158 		}
2159 	}
2160 	if (chain->core && !TAILQ_EMPTY(&chain->core->layerq)) {
2161 		if (chain->bref.type == HAMMER2_BREF_TYPE_INODE && chain->data)
2162 			kprintf("%*.*s}(%s)\n", tab, tab, "",
2163 				chain->data->ipdata.filename);
2164 		else
2165 			kprintf("%*.*s}\n", tab, tab, "");
2166 	}
2167 }
2168