hammer2 - Flesh out span code, API cleanups
[dragonfly.git] / sys / vfs / hammer2 / hammer2_vfsops.c
1 /*-
2  * Copyright (c) 2011, 2012 The DragonFly Project.  All rights reserved.
3  *
4  * This code is derived from software contributed to The DragonFly Project
5  * by Matthew Dillon <dillon@backplane.com>
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  *
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in
15  *    the documentation and/or other materials provided with the
16  *    distribution.
17  * 3. Neither the name of The DragonFly Project nor the names of its
18  *    contributors may be used to endorse or promote products derived
19  *    from this software without specific, prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
25  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  */
34 #include <sys/param.h>
35 #include <sys/systm.h>
36 #include <sys/kernel.h>
37 #include <sys/nlookup.h>
38 #include <sys/vnode.h>
39 #include <sys/mount.h>
40 #include <sys/fcntl.h>
41 #include <sys/buf.h>
42 #include <sys/uuid.h>
43 #include <sys/vfsops.h>
44 #include <sys/sysctl.h>
45 #include <sys/socket.h>
46
47 #include "hammer2.h"
48 #include "hammer2_disk.h"
49 #include "hammer2_mount.h"
50 #include "hammer2_network.h"
51
52 struct hammer2_sync_info {
53         int error;
54         int waitfor;
55 };
56
57 TAILQ_HEAD(hammer2_mntlist, hammer2_mount);
58 static struct hammer2_mntlist hammer2_mntlist;
59 static struct lock hammer2_mntlk;
60
61 int hammer2_debug;
62 int hammer2_cluster_enable = 1;
63 int hammer2_hardlink_enable = 1;
64 long hammer2_iod_file_read;
65 long hammer2_iod_meta_read;
66 long hammer2_iod_indr_read;
67 long hammer2_iod_file_write;
68 long hammer2_iod_meta_write;
69 long hammer2_iod_indr_write;
70 long hammer2_iod_volu_write;
71 long hammer2_ioa_file_read;
72 long hammer2_ioa_meta_read;
73 long hammer2_ioa_indr_read;
74 long hammer2_ioa_file_write;
75 long hammer2_ioa_meta_write;
76 long hammer2_ioa_indr_write;
77 long hammer2_ioa_volu_write;
78
79 SYSCTL_NODE(_vfs, OID_AUTO, hammer2, CTLFLAG_RW, 0, "HAMMER2 filesystem");
80
81 SYSCTL_INT(_vfs_hammer2, OID_AUTO, debug, CTLFLAG_RW,
82            &hammer2_debug, 0, "");
83 SYSCTL_INT(_vfs_hammer2, OID_AUTO, cluster_enable, CTLFLAG_RW,
84            &hammer2_cluster_enable, 0, "");
85 SYSCTL_INT(_vfs_hammer2, OID_AUTO, hardlink_enable, CTLFLAG_RW,
86            &hammer2_hardlink_enable, 0, "");
87 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_file_read, CTLFLAG_RW,
88            &hammer2_iod_file_read, 0, "");
89 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_meta_read, CTLFLAG_RW,
90            &hammer2_iod_meta_read, 0, "");
91 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_indr_read, CTLFLAG_RW,
92            &hammer2_iod_indr_read, 0, "");
93 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_file_write, CTLFLAG_RW,
94            &hammer2_iod_file_write, 0, "");
95 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_meta_write, CTLFLAG_RW,
96            &hammer2_iod_meta_write, 0, "");
97 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_indr_write, CTLFLAG_RW,
98            &hammer2_iod_indr_write, 0, "");
99 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_volu_write, CTLFLAG_RW,
100            &hammer2_iod_volu_write, 0, "");
101 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_file_read, CTLFLAG_RW,
102            &hammer2_ioa_file_read, 0, "");
103 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_meta_read, CTLFLAG_RW,
104            &hammer2_ioa_meta_read, 0, "");
105 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_indr_read, CTLFLAG_RW,
106            &hammer2_ioa_indr_read, 0, "");
107 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_file_write, CTLFLAG_RW,
108            &hammer2_ioa_file_write, 0, "");
109 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_meta_write, CTLFLAG_RW,
110            &hammer2_ioa_meta_write, 0, "");
111 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_indr_write, CTLFLAG_RW,
112            &hammer2_ioa_indr_write, 0, "");
113 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_volu_write, CTLFLAG_RW,
114            &hammer2_ioa_volu_write, 0, "");
115
116 static int hammer2_vfs_init(struct vfsconf *conf);
117 static int hammer2_vfs_mount(struct mount *mp, char *path, caddr_t data,
118                                 struct ucred *cred);
119 static int hammer2_remount(struct mount *, char *, struct vnode *,
120                                 struct ucred *);
121 static int hammer2_vfs_unmount(struct mount *mp, int mntflags);
122 static int hammer2_vfs_root(struct mount *mp, struct vnode **vpp);
123 static int hammer2_vfs_statfs(struct mount *mp, struct statfs *sbp,
124                                 struct ucred *cred);
125 static int hammer2_vfs_statvfs(struct mount *mp, struct statvfs *sbp,
126                                 struct ucred *cred);
127 static int hammer2_vfs_sync(struct mount *mp, int waitfor);
128 static int hammer2_vfs_vget(struct mount *mp, struct vnode *dvp,
129                                 ino_t ino, struct vnode **vpp);
130 static int hammer2_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
131                                 struct fid *fhp, struct vnode **vpp);
132 static int hammer2_vfs_vptofh(struct vnode *vp, struct fid *fhp);
133 static int hammer2_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
134                                 int *exflagsp, struct ucred **credanonp);
135
136 static int hammer2_install_volume_header(hammer2_mount_t *hmp);
137 static int hammer2_sync_scan1(struct mount *mp, struct vnode *vp, void *data);
138 static int hammer2_sync_scan2(struct mount *mp, struct vnode *vp, void *data);
139
140 static void hammer2_cluster_thread_rd(void *arg);
141 static void hammer2_cluster_thread_wr(void *arg);
142 static int hammer2_msg_conn_reply(hammer2_state_t *state, hammer2_msg_t *msg);
143 static int hammer2_msg_span_reply(hammer2_state_t *state, hammer2_msg_t *msg);
144 static int hammer2_msg_lnk_rcvmsg(hammer2_pfsmount_t *pmp, hammer2_msg_t *msg);
145
146 /*
147  * HAMMER2 vfs operations.
148  */
149 static struct vfsops hammer2_vfsops = {
150         .vfs_init       = hammer2_vfs_init,
151         .vfs_sync       = hammer2_vfs_sync,
152         .vfs_mount      = hammer2_vfs_mount,
153         .vfs_unmount    = hammer2_vfs_unmount,
154         .vfs_root       = hammer2_vfs_root,
155         .vfs_statfs     = hammer2_vfs_statfs,
156         .vfs_statvfs    = hammer2_vfs_statvfs,
157         .vfs_vget       = hammer2_vfs_vget,
158         .vfs_vptofh     = hammer2_vfs_vptofh,
159         .vfs_fhtovp     = hammer2_vfs_fhtovp,
160         .vfs_checkexp   = hammer2_vfs_checkexp
161 };
162
163 MALLOC_DEFINE(M_HAMMER2, "HAMMER2-mount", "");
164
165 VFS_SET(hammer2_vfsops, hammer2, 0);
166 MODULE_VERSION(hammer2, 1);
167
168 static
169 int
170 hammer2_vfs_init(struct vfsconf *conf)
171 {
172         int error;
173
174         error = 0;
175
176         if (HAMMER2_BLOCKREF_BYTES != sizeof(struct hammer2_blockref))
177                 error = EINVAL;
178         if (HAMMER2_INODE_BYTES != sizeof(struct hammer2_inode_data))
179                 error = EINVAL;
180         if (HAMMER2_ALLOCREF_BYTES != sizeof(struct hammer2_allocref))
181                 error = EINVAL;
182         if (HAMMER2_VOLUME_BYTES != sizeof(struct hammer2_volume_data))
183                 error = EINVAL;
184
185         if (error)
186                 kprintf("HAMMER2 structure size mismatch; cannot continue.\n");
187
188         lockinit(&hammer2_mntlk, "mntlk", 0, 0);
189         TAILQ_INIT(&hammer2_mntlist);
190
191         return (error);
192 }
193
194 /*
195  * Mount or remount HAMMER2 fileystem from physical media
196  *
197  *      mountroot
198  *              mp              mount point structure
199  *              path            NULL
200  *              data            <unused>
201  *              cred            <unused>
202  *
203  *      mount
204  *              mp              mount point structure
205  *              path            path to mount point
206  *              data            pointer to argument structure in user space
207  *                      volume  volume path (device@LABEL form)
208  *                      hflags  user mount flags
209  *              cred            user credentials
210  *
211  * RETURNS:     0       Success
212  *              !0      error number
213  */
214 static
215 int
216 hammer2_vfs_mount(struct mount *mp, char *path, caddr_t data,
217                   struct ucred *cred)
218 {
219         struct hammer2_mount_info info;
220         hammer2_pfsmount_t *pmp;
221         hammer2_mount_t *hmp;
222         hammer2_key_t lhc;
223         struct vnode *devvp;
224         struct nlookupdata nd;
225         hammer2_chain_t *parent;
226         hammer2_chain_t *schain;
227         hammer2_chain_t *rchain;
228         char devstr[MNAMELEN];
229         size_t size;
230         size_t done;
231         char *dev;
232         char *label;
233         int ronly = 1;
234         int create_hmp;
235         int error;
236
237         hmp = NULL;
238         pmp = NULL;
239         dev = NULL;
240         label = NULL;
241         devvp = NULL;
242
243         kprintf("hammer2_mount\n");
244
245         if (path == NULL) {
246                 /*
247                  * Root mount
248                  */
249                 bzero(&info, sizeof(info));
250                 info.cluster_fd = -1;
251                 return (EOPNOTSUPP);
252         } else {
253                 /*
254                  * Non-root mount or updating a mount
255                  */
256                 error = copyin(data, &info, sizeof(info));
257                 if (error)
258                         return (error);
259
260                 error = copyinstr(info.volume, devstr, MNAMELEN - 1, &done);
261                 if (error)
262                         return (error);
263
264                 /* Extract device and label */
265                 dev = devstr;
266                 label = strchr(devstr, '@');
267                 if (label == NULL ||
268                     ((label + 1) - dev) > done) {
269                         return (EINVAL);
270                 }
271                 *label = '\0';
272                 label++;
273                 if (*label == '\0')
274                         return (EINVAL);
275
276                 if (mp->mnt_flag & MNT_UPDATE) {
277                         /* Update mount */
278                         /* HAMMER2 implements NFS export via mountctl */
279                         hmp = MPTOHMP(mp);
280                         devvp = hmp->devvp;
281                         error = hammer2_remount(mp, path, devvp, cred);
282                         return error;
283                 }
284         }
285
286         /*
287          * PFS mount
288          *
289          * Lookup name and verify it refers to a block device.
290          */
291         error = nlookup_init(&nd, dev, UIO_SYSSPACE, NLC_FOLLOW);
292         if (error == 0)
293                 error = nlookup(&nd);
294         if (error == 0)
295                 error = cache_vref(&nd.nl_nch, nd.nl_cred, &devvp);
296         nlookup_done(&nd);
297
298         if (error == 0) {
299                 if (vn_isdisk(devvp, &error))
300                         error = vfs_mountedon(devvp);
301         }
302
303         /*
304          * Determine if the device has already been mounted.  After this
305          * check hmp will be non-NULL if we are doing the second or more
306          * hammer2 mounts from the same device.
307          */
308         lockmgr(&hammer2_mntlk, LK_EXCLUSIVE);
309         TAILQ_FOREACH(hmp, &hammer2_mntlist, mntentry) {
310                 if (hmp->devvp == devvp)
311                         break;
312         }
313
314         /*
315          * Open the device if this isn't a secondary mount
316          */
317         if (hmp) {
318                 create_hmp = 0;
319         } else {
320                 create_hmp = 1;
321                 if (error == 0 && vcount(devvp) > 0)
322                         error = EBUSY;
323
324                 /*
325                  * Now open the device
326                  */
327                 if (error == 0) {
328                         ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
329                         vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
330                         error = vinvalbuf(devvp, V_SAVE, 0, 0);
331                         if (error == 0) {
332                                 error = VOP_OPEN(devvp,
333                                                  ronly ? FREAD : FREAD | FWRITE,
334                                                  FSCRED, NULL);
335                         }
336                         vn_unlock(devvp);
337                 }
338                 if (error && devvp) {
339                         vrele(devvp);
340                         devvp = NULL;
341                 }
342                 if (error) {
343                         lockmgr(&hammer2_mntlk, LK_RELEASE);
344                         return error;
345                 }
346         }
347
348         /*
349          * Block device opened successfully, finish initializing the
350          * mount structure.
351          *
352          * From this point on we have to call hammer2_unmount() on failure.
353          */
354         pmp = kmalloc(sizeof(*pmp), M_HAMMER2, M_WAITOK | M_ZERO);
355         mp->mnt_data = (qaddr_t)pmp;
356         pmp->mp = mp;
357         kmalloc_create(&pmp->mmsg, "HAMMER2-pfsmsg");
358         lockinit(&pmp->msglk, "h2msg", 0, 0);
359         TAILQ_INIT(&pmp->msgq);
360         RB_INIT(&pmp->staterd_tree);
361         RB_INIT(&pmp->statewr_tree);
362
363         if (create_hmp) {
364                 hmp = kmalloc(sizeof(*hmp), M_HAMMER2, M_WAITOK | M_ZERO);
365                 hmp->ronly = ronly;
366                 hmp->devvp = devvp;
367                 kmalloc_create(&hmp->minode, "HAMMER2-inodes");
368                 kmalloc_create(&hmp->mchain, "HAMMER2-chains");
369                 TAILQ_INSERT_TAIL(&hammer2_mntlist, hmp, mntentry);
370         }
371         ccms_domain_init(&pmp->ccms_dom);
372         pmp->hmp = hmp;
373         ++hmp->pmp_count;
374         lockmgr(&hammer2_mntlk, LK_RELEASE);
375         kprintf("hammer2_mount hmp=%p pmpcnt=%d\n", hmp, hmp->pmp_count);
376         
377         mp->mnt_flag = MNT_LOCAL;
378         mp->mnt_kern_flag |= MNTK_ALL_MPSAFE;   /* all entry pts are SMP */
379
380         if (create_hmp) {
381                 /*
382                  * vchain setup. vchain.data is special cased to NULL.
383                  * vchain.refs is initialized and will never drop to 0.
384                  */
385                 hmp->vchain.refs = 1;
386                 hmp->vchain.data = (void *)&hmp->voldata;
387                 hmp->vchain.bref.type = HAMMER2_BREF_TYPE_VOLUME;
388                 hmp->vchain.bref.data_off = 0 | HAMMER2_PBUFRADIX;
389                 hmp->vchain.bref_flush = hmp->vchain.bref;
390                 ccms_cst_init(&hmp->vchain.cst, NULL);
391                 /* hmp->vchain.u.xxx is left NULL */
392                 lockinit(&hmp->alloclk, "h2alloc", 0, 0);
393                 lockinit(&hmp->voldatalk, "voldata", 0, LK_CANRECURSE);
394
395                 /*
396                  * Install the volume header
397                  */
398                 error = hammer2_install_volume_header(hmp);
399                 if (error) {
400                         hammer2_vfs_unmount(mp, MNT_FORCE);
401                         return error;
402                 }
403         }
404
405         /*
406          * required mount structure initializations
407          */
408         mp->mnt_stat.f_iosize = HAMMER2_PBUFSIZE;
409         mp->mnt_stat.f_bsize = HAMMER2_PBUFSIZE;
410
411         mp->mnt_vstat.f_frsize = HAMMER2_PBUFSIZE;
412         mp->mnt_vstat.f_bsize = HAMMER2_PBUFSIZE;
413
414         /*
415          * Optional fields
416          */
417         mp->mnt_iosize_max = MAXPHYS;
418
419         /*
420          * First locate the super-root inode, which is key 0 relative to the
421          * volume header's blockset.
422          *
423          * Then locate the root inode by scanning the directory keyspace
424          * represented by the label.
425          */
426         if (create_hmp) {
427                 parent = &hmp->vchain;
428                 hammer2_chain_lock(hmp, parent, HAMMER2_RESOLVE_ALWAYS);
429                 schain = hammer2_chain_lookup(hmp, &parent,
430                                       HAMMER2_SROOT_KEY, HAMMER2_SROOT_KEY, 0);
431                 hammer2_chain_unlock(hmp, parent);
432                 if (schain == NULL) {
433                         kprintf("hammer2_mount: invalid super-root\n");
434                         hammer2_vfs_unmount(mp, MNT_FORCE);
435                         return EINVAL;
436                 }
437                 hammer2_chain_ref(hmp, schain); /* for hmp->schain */
438                 hmp->schain = schain;           /* left locked */
439         } else {
440                 schain = hmp->schain;
441                 hammer2_chain_lock(hmp, schain, HAMMER2_RESOLVE_ALWAYS);
442         }
443
444         parent = schain;
445         lhc = hammer2_dirhash(label, strlen(label));
446         rchain = hammer2_chain_lookup(hmp, &parent,
447                                       lhc, lhc + HAMMER2_DIRHASH_LOMASK,
448                                       0);
449         while (rchain) {
450                 if (rchain->bref.type == HAMMER2_BREF_TYPE_INODE &&
451                     rchain->u.ip &&
452                     strcmp(label, rchain->data->ipdata.filename) == 0) {
453                         break;
454                 }
455                 rchain = hammer2_chain_next(hmp, &parent, rchain,
456                                             lhc, lhc + HAMMER2_DIRHASH_LOMASK,
457                                             0);
458         }
459         hammer2_chain_unlock(hmp, parent);
460         if (rchain == NULL) {
461                 kprintf("hammer2_mount: PFS label not found\n");
462                 hammer2_vfs_unmount(mp, MNT_FORCE);
463                 return EINVAL;
464         }
465         if (rchain->flags & HAMMER2_CHAIN_MOUNTED) {
466                 hammer2_chain_unlock(hmp, rchain);
467                 kprintf("hammer2_mount: PFS label already mounted!\n");
468                 hammer2_vfs_unmount(mp, MNT_FORCE);
469                 return EBUSY;
470         }
471         atomic_set_int(&rchain->flags, HAMMER2_CHAIN_MOUNTED);
472
473         hammer2_chain_ref(hmp, rchain); /* for pmp->rchain */
474         hammer2_chain_unlock(hmp, rchain);
475         pmp->rchain = rchain;           /* left held & unlocked */
476         pmp->iroot = rchain->u.ip;      /* implied hold from rchain */
477         pmp->iroot->pmp = pmp;
478
479         kprintf("iroot %p\n", pmp->iroot);
480
481         /*
482          * Ref the cluster management messaging descriptor.  The mount
483          * program deals with the other end of the communications pipe.
484          */
485         pmp->msg_fp = holdfp(curproc->p_fd, info.cluster_fd, -1);
486         if (pmp->msg_fp == NULL) {
487                 kprintf("hammer2_mount: bad cluster_fd!\n");
488                 hammer2_vfs_unmount(mp, MNT_FORCE);
489                 return EBADF;
490         }
491         lwkt_create(hammer2_cluster_thread_rd, pmp, &pmp->msgrd_td,
492                     NULL, 0, -1, "hammer2-msgrd");
493         lwkt_create(hammer2_cluster_thread_wr, pmp, &pmp->msgwr_td,
494                     NULL, 0, -1, "hammer2-msgwr");
495
496         /*
497          * Finish setup
498          */
499         vfs_getnewfsid(mp);
500         vfs_add_vnodeops(mp, &hammer2_vnode_vops, &mp->mnt_vn_norm_ops);
501         vfs_add_vnodeops(mp, &hammer2_spec_vops, &mp->mnt_vn_spec_ops);
502         vfs_add_vnodeops(mp, &hammer2_fifo_vops, &mp->mnt_vn_fifo_ops);
503
504         copyinstr(info.volume, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, &size);
505         bzero(mp->mnt_stat.f_mntfromname + size, MNAMELEN - size);
506         bzero(mp->mnt_stat.f_mntonname, sizeof(mp->mnt_stat.f_mntonname));
507         copyinstr(path, mp->mnt_stat.f_mntonname,
508                   sizeof(mp->mnt_stat.f_mntonname) - 1,
509                   &size);
510
511         /*
512          * Initial statfs to prime mnt_stat.
513          */
514         hammer2_vfs_statfs(mp, &mp->mnt_stat, cred);
515
516         return 0;
517 }
518
519 static
520 int
521 hammer2_remount(struct mount *mp, char *path, struct vnode *devvp,
522                 struct ucred *cred)
523 {
524         return (0);
525 }
526
527 static
528 int
529 hammer2_vfs_unmount(struct mount *mp, int mntflags)
530 {
531         hammer2_pfsmount_t *pmp;
532         hammer2_mount_t *hmp;
533         int flags;
534         int error = 0;
535         int ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
536         struct vnode *devvp;
537
538         pmp = MPTOPMP(mp);
539         hmp = pmp->hmp;
540         flags = 0;
541
542         if (mntflags & MNT_FORCE)
543                 flags |= FORCECLOSE;
544
545         hammer2_mount_exlock(hmp);
546
547         /*
548          * If mount initialization proceeded far enough we must flush
549          * its vnodes.
550          */
551         if (pmp->iroot)
552                 error = vflush(mp, 0, flags);
553
554         if (error)
555                 return error;
556
557         lockmgr(&hammer2_mntlk, LK_EXCLUSIVE);
558         --hmp->pmp_count;
559         kprintf("hammer2_unmount hmp=%p pmpcnt=%d\n", hmp, hmp->pmp_count);
560
561         /*
562          * Flush any left over chains.  The voldata lock is only used
563          * to synchronize against HAMMER2_CHAIN_MODIFIED_AUX.
564          */
565         hammer2_voldata_lock(hmp);
566         if (hmp->vchain.flags & (HAMMER2_CHAIN_MODIFIED |
567                                  HAMMER2_CHAIN_MODIFIED_AUX |
568                                  HAMMER2_CHAIN_SUBMODIFIED)) {
569                 hammer2_voldata_unlock(hmp);
570                 hammer2_vfs_sync(mp, MNT_WAIT);
571         } else {
572                 hammer2_voldata_unlock(hmp);
573         }
574         if (hmp->pmp_count == 0) {
575                 if (hmp->vchain.flags & (HAMMER2_CHAIN_MODIFIED |
576                                          HAMMER2_CHAIN_MODIFIED_AUX |
577                                          HAMMER2_CHAIN_SUBMODIFIED)) {
578                         kprintf("hammer2_unmount: chains left over after "
579                                 "final sync\n");
580                         if (hammer2_debug & 0x0010)
581                                 Debugger("entered debugger");
582                 }
583         }
584
585         /*
586          * Cleanup the root and super-root chain elements (which should be
587          * clean).
588          */
589         pmp->iroot = NULL;
590         if (pmp->rchain) {
591                 atomic_clear_int(&pmp->rchain->flags, HAMMER2_CHAIN_MOUNTED);
592                 KKASSERT(pmp->rchain->refs == 1);
593                 hammer2_chain_drop(hmp, pmp->rchain);
594                 pmp->rchain = NULL;
595         }
596         ccms_domain_uninit(&pmp->ccms_dom);
597
598         /*
599          * Ask the cluster controller to go away
600          */
601         atomic_set_int(&pmp->msg_ctl, HAMMER2_CLUSTERCTL_KILL);
602         while (pmp->msgrd_td || pmp->msgwr_td) {
603                 wakeup(&pmp->msg_ctl);
604                 tsleep(pmp, 0, "clstrkl", hz);
605         }
606
607         /*
608          * Drop communications descriptor
609          */
610         if (pmp->msg_fp) {
611                 fdrop(pmp->msg_fp);
612                 pmp->msg_fp = NULL;
613         }
614
615         /*
616          * If no PFS's left drop the master hammer2_mount for the device.
617          */
618         if (hmp->pmp_count == 0) {
619                 if (hmp->schain) {
620                         KKASSERT(hmp->schain->refs == 1);
621                         hammer2_chain_drop(hmp, hmp->schain);
622                         hmp->schain = NULL;
623                 }
624
625                 /*
626                  * Finish up with the device vnode
627                  */
628                 if ((devvp = hmp->devvp) != NULL) {
629                         vinvalbuf(devvp, (ronly ? 0 : V_SAVE), 0, 0);
630                         hmp->devvp = NULL;
631                         VOP_CLOSE(devvp, (ronly ? FREAD : FREAD|FWRITE));
632                         vrele(devvp);
633                         devvp = NULL;
634                 }
635         }
636         hammer2_mount_unlock(hmp);
637
638         pmp->mp = NULL;
639         pmp->hmp = NULL;
640         mp->mnt_data = NULL;
641
642         kmalloc_destroy(&pmp->mmsg);
643
644         kfree(pmp, M_HAMMER2);
645         if (hmp->pmp_count == 0) {
646                 TAILQ_REMOVE(&hammer2_mntlist, hmp, mntentry);
647                 kmalloc_destroy(&hmp->minode);
648                 kmalloc_destroy(&hmp->mchain);
649                 kfree(hmp, M_HAMMER2);
650         }
651         lockmgr(&hammer2_mntlk, LK_RELEASE);
652         return (error);
653 }
654
655 static
656 int
657 hammer2_vfs_vget(struct mount *mp, struct vnode *dvp,
658              ino_t ino, struct vnode **vpp)
659 {
660         kprintf("hammer2_vget\n");
661         return (EOPNOTSUPP);
662 }
663
664 static
665 int
666 hammer2_vfs_root(struct mount *mp, struct vnode **vpp)
667 {
668         hammer2_pfsmount_t *pmp;
669         hammer2_mount_t *hmp;
670         int error;
671         struct vnode *vp;
672
673         pmp = MPTOPMP(mp);
674         hmp = pmp->hmp;
675         hammer2_mount_exlock(hmp);
676         if (pmp->iroot == NULL) {
677                 *vpp = NULL;
678                 error = EINVAL;
679         } else {
680                 hammer2_chain_lock(hmp, &pmp->iroot->chain,
681                                    HAMMER2_RESOLVE_ALWAYS |
682                                    HAMMER2_RESOLVE_SHARED);
683                 vp = hammer2_igetv(pmp->iroot, &error);
684                 hammer2_chain_unlock(hmp, &pmp->iroot->chain);
685                 *vpp = vp;
686                 if (vp == NULL)
687                         kprintf("vnodefail\n");
688         }
689         hammer2_mount_unlock(hmp);
690
691         return (error);
692 }
693
694 /*
695  * Filesystem status
696  *
697  * XXX incorporate pmp->iroot->ip_data.inode_quota and data_quota
698  */
699 static
700 int
701 hammer2_vfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
702 {
703         hammer2_pfsmount_t *pmp;
704         hammer2_mount_t *hmp;
705
706         pmp = MPTOPMP(mp);
707         hmp = MPTOHMP(mp);
708
709         mp->mnt_stat.f_files = pmp->iroot->ip_data.inode_count +
710                                pmp->iroot->delta_icount;
711         mp->mnt_stat.f_ffree = 0;
712         mp->mnt_stat.f_blocks = hmp->voldata.allocator_size / HAMMER2_PBUFSIZE;
713         mp->mnt_stat.f_bfree = (hmp->voldata.allocator_size -
714                                 hmp->voldata.allocator_beg) / HAMMER2_PBUFSIZE;
715         mp->mnt_stat.f_bavail = mp->mnt_stat.f_bfree;
716
717         *sbp = mp->mnt_stat;
718         return (0);
719 }
720
721 static
722 int
723 hammer2_vfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred)
724 {
725         hammer2_pfsmount_t *pmp;
726         hammer2_mount_t *hmp;
727
728         pmp = MPTOPMP(mp);
729         hmp = MPTOHMP(mp);
730
731         mp->mnt_vstat.f_bsize = HAMMER2_PBUFSIZE;
732         mp->mnt_vstat.f_files = pmp->iroot->ip_data.inode_count +
733                                 pmp->iroot->delta_icount;
734         mp->mnt_vstat.f_ffree = 0;
735         mp->mnt_vstat.f_blocks = hmp->voldata.allocator_size / HAMMER2_PBUFSIZE;
736         mp->mnt_vstat.f_bfree = (hmp->voldata.allocator_size -
737                                  hmp->voldata.allocator_beg) / HAMMER2_PBUFSIZE;
738         mp->mnt_vstat.f_bavail = mp->mnt_vstat.f_bfree;
739
740         *sbp = mp->mnt_vstat;
741         return (0);
742 }
743
744 /*
745  * Sync the entire filesystem; this is called from the filesystem syncer
746  * process periodically and whenever a user calls sync(1) on the hammer
747  * mountpoint.
748  *
749  * Currently is actually called from the syncer! \o/
750  *
751  * This task will have to snapshot the state of the dirty inode chain.
752  * From that, it will have to make sure all of the inodes on the dirty
753  * chain have IO initiated. We make sure that io is initiated for the root
754  * block.
755  *
756  * If waitfor is set, we wait for media to acknowledge the new rootblock.
757  *
758  * THINKS: side A vs side B, to have sync not stall all I/O?
759  */
760 static
761 int
762 hammer2_vfs_sync(struct mount *mp, int waitfor)
763 {
764         struct hammer2_sync_info info;
765         hammer2_mount_t *hmp;
766         int flags;
767         int error;
768         int haswork;
769
770         hmp = MPTOHMP(mp);
771
772         flags = VMSC_GETVP;
773         if (waitfor & MNT_LAZY)
774                 flags |= VMSC_ONEPASS;
775
776         info.error = 0;
777         info.waitfor = MNT_NOWAIT;
778         vmntvnodescan(mp, flags | VMSC_NOWAIT,
779                       hammer2_sync_scan1,
780                       hammer2_sync_scan2, &info);
781         if (info.error == 0 && (waitfor & MNT_WAIT)) {
782                 info.waitfor = waitfor;
783                     vmntvnodescan(mp, flags,
784                                   hammer2_sync_scan1,
785                                   hammer2_sync_scan2, &info);
786
787         }
788 #if 0
789         if (waitfor == MNT_WAIT) {
790                 /* XXX */
791         } else {
792                 /* XXX */
793         }
794 #endif
795         hammer2_chain_lock(hmp, &hmp->vchain, HAMMER2_RESOLVE_ALWAYS);
796         if (hmp->vchain.flags & (HAMMER2_CHAIN_MODIFIED |
797                                  HAMMER2_CHAIN_MODIFIED_AUX |
798                                  HAMMER2_CHAIN_SUBMODIFIED)) {
799                 hammer2_chain_flush(hmp, &hmp->vchain, 0);
800                 haswork = 1;
801         } else {
802                 haswork = 0;
803         }
804         hammer2_chain_unlock(hmp, &hmp->vchain);
805
806         error = 0;
807
808         if ((waitfor & MNT_LAZY) == 0) {
809                 waitfor = MNT_NOWAIT;
810                 vn_lock(hmp->devvp, LK_EXCLUSIVE | LK_RETRY);
811                 error = VOP_FSYNC(hmp->devvp, waitfor, 0);
812                 vn_unlock(hmp->devvp);
813         }
814
815         if (error == 0 && haswork) {
816                 struct buf *bp;
817
818                 /*
819                  * Synchronize the disk before flushing the volume
820                  * header.
821                  */
822                 bp = getpbuf(NULL);
823                 bp->b_bio1.bio_offset = 0;
824                 bp->b_bufsize = 0;
825                 bp->b_bcount = 0;
826                 bp->b_cmd = BUF_CMD_FLUSH;
827                 bp->b_bio1.bio_done = biodone_sync;
828                 bp->b_bio1.bio_flags |= BIO_SYNC;
829                 vn_strategy(hmp->devvp, &bp->b_bio1);
830                 biowait(&bp->b_bio1, "h2vol");
831                 relpbuf(bp, NULL);
832
833                 /*
834                  * Then we can safely flush the volume header.  Volume
835                  * data is locked separately to prevent ioctl functions
836                  * from deadlocking due to a configuration issue.
837                  */
838                 bp = getblk(hmp->devvp, 0, HAMMER2_PBUFSIZE, 0, 0);
839                 hammer2_voldata_lock(hmp);
840                 bcopy(&hmp->voldata, bp->b_data, HAMMER2_PBUFSIZE);
841                 hammer2_voldata_unlock(hmp);
842                 bawrite(bp);
843         }
844         return (error);
845 }
846
847 /*
848  * Sync passes.
849  *
850  * NOTE: We don't test SUBMODIFIED or MOVED here because the fsync code
851  *       won't flush on those flags.  The syncer code above will do a
852  *       general meta-data flush globally that will catch these flags.
853  */
854 static int
855 hammer2_sync_scan1(struct mount *mp, struct vnode *vp, void *data)
856 {
857         hammer2_inode_t *ip;
858
859         ip = VTOI(vp);
860         if (vp->v_type == VNON || ip == NULL ||
861             ((ip->chain.flags & (HAMMER2_CHAIN_MODIFIED |
862                                  HAMMER2_CHAIN_DIRTYEMBED)) == 0 &&
863              RB_EMPTY(&vp->v_rbdirty_tree))) {
864                 return(-1);
865         }
866         return(0);
867 }
868
869 static int
870 hammer2_sync_scan2(struct mount *mp, struct vnode *vp, void *data)
871 {
872         struct hammer2_sync_info *info = data;
873         hammer2_inode_t *ip;
874         int error;
875
876         ip = VTOI(vp);
877         if (vp->v_type == VNON || vp->v_type == VBAD ||
878             ((ip->chain.flags & (HAMMER2_CHAIN_MODIFIED |
879                                  HAMMER2_CHAIN_DIRTYEMBED)) == 0 &&
880             RB_EMPTY(&vp->v_rbdirty_tree))) {
881                 return(0);
882         }
883         error = VOP_FSYNC(vp, MNT_NOWAIT, 0);
884         if (error)
885                 info->error = error;
886         return(0);
887 }
888
889 static
890 int
891 hammer2_vfs_vptofh(struct vnode *vp, struct fid *fhp)
892 {
893         return (0);
894 }
895
896 static
897 int
898 hammer2_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
899                struct fid *fhp, struct vnode **vpp)
900 {
901         return (0);
902 }
903
904 static
905 int
906 hammer2_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
907                  int *exflagsp, struct ucred **credanonp)
908 {
909         return (0);
910 }
911
912 /*
913  * Support code for hammer2_mount().  Read, verify, and install the volume
914  * header into the HMP
915  *
916  * XXX read four volhdrs and use the one with the highest TID whos CRC
917  *     matches.
918  *
919  * XXX check iCRCs.
920  *
921  * XXX For filesystems w/ less than 4 volhdrs, make sure to not write to
922  *     nonexistant locations.
923  *
924  * XXX Record selected volhdr and ring updates to each of 4 volhdrs
925  */
926 static
927 int
928 hammer2_install_volume_header(hammer2_mount_t *hmp)
929 {
930         hammer2_volume_data_t *vd;
931         struct buf *bp;
932         hammer2_crc32_t crc0, crc, bcrc0, bcrc;
933         int error_reported;
934         int error;
935         int valid;
936         int i;
937
938         error_reported = 0;
939         error = 0;
940         valid = 0;
941         bp = NULL;
942
943         /*
944          * There are up to 4 copies of the volume header (syncs iterate
945          * between them so there is no single master).  We don't trust the
946          * volu_size field so we don't know precisely how large the filesystem
947          * is, so depend on the OS to return an error if we go beyond the
948          * block device's EOF.
949          */
950         for (i = 0; i < HAMMER2_NUM_VOLHDRS; i++) {
951                 error = bread(hmp->devvp, i * HAMMER2_ZONE_BYTES64,
952                               HAMMER2_VOLUME_BYTES, &bp);
953                 if (error) {
954                         brelse(bp);
955                         bp = NULL;
956                         continue;
957                 }
958
959                 vd = (struct hammer2_volume_data *) bp->b_data;
960                 if ((vd->magic != HAMMER2_VOLUME_ID_HBO) &&
961                     (vd->magic != HAMMER2_VOLUME_ID_ABO)) {
962                         brelse(bp);
963                         bp = NULL;
964                         continue;
965                 }
966
967                 if (vd->magic == HAMMER2_VOLUME_ID_ABO) {
968                         /* XXX: Reversed-endianness filesystem */
969                         kprintf("hammer2: reverse-endian filesystem detected");
970                         brelse(bp);
971                         bp = NULL;
972                         continue;
973                 }
974
975                 crc = vd->icrc_sects[HAMMER2_VOL_ICRC_SECT0];
976                 crc0 = hammer2_icrc32(bp->b_data + HAMMER2_VOLUME_ICRC0_OFF,
977                                       HAMMER2_VOLUME_ICRC0_SIZE);
978                 bcrc = vd->icrc_sects[HAMMER2_VOL_ICRC_SECT1];
979                 bcrc0 = hammer2_icrc32(bp->b_data + HAMMER2_VOLUME_ICRC1_OFF,
980                                        HAMMER2_VOLUME_ICRC1_SIZE);
981                 if ((crc0 != crc) || (bcrc0 != bcrc)) {
982                         kprintf("hammer2 volume header crc "
983                                 "mismatch copy #%d\t%08x %08x",
984                                 i, crc0, crc);
985                         error_reported = 1;
986                         brelse(bp);
987                         bp = NULL;
988                         continue;
989                 }
990                 if (valid == 0 || hmp->voldata.mirror_tid < vd->mirror_tid) {
991                         valid = 1;
992                         hmp->voldata = *vd;
993                 }
994                 brelse(bp);
995                 bp = NULL;
996         }
997         if (valid) {
998                 error = 0;
999                 if (error_reported)
1000                         kprintf("hammer2: a valid volume header was found\n");
1001         } else {
1002                 error = EINVAL;
1003                 kprintf("hammer2: no valid volume headers found!\n");
1004         }
1005         return (error);
1006 }
1007
1008 /*
1009  * Cluster controller thread.  Perform messaging functions.  We have one
1010  * thread for the reader and one for the writer.  The writer handles
1011  * shutdown requests (which should break the reader thread).
1012  */
1013 static
1014 void
1015 hammer2_cluster_thread_rd(void *arg)
1016 {
1017         hammer2_pfsmount_t *pmp = arg;
1018         hammer2_msg_hdr_t hdr;
1019         hammer2_msg_t *msg;
1020         hammer2_state_t *state;
1021         size_t hbytes;
1022         int error = 0;
1023
1024         while ((pmp->msg_ctl & HAMMER2_CLUSTERCTL_KILL) == 0) {
1025                 /*
1026                  * Retrieve the message from the pipe or socket.
1027                  */
1028                 error = fp_read(pmp->msg_fp, &hdr, sizeof(hdr),
1029                                 NULL, 1, UIO_SYSSPACE);
1030                 if (error)
1031                         break;
1032                 if (hdr.magic != HAMMER2_MSGHDR_MAGIC) {
1033                         kprintf("hammer2: msgrd: bad magic: %04x\n",
1034                                 hdr.magic);
1035                         error = EINVAL;
1036                         break;
1037                 }
1038                 hbytes = (hdr.cmd & HAMMER2_MSGF_SIZE) * HAMMER2_MSG_ALIGN;
1039                 if (hbytes < sizeof(hdr) || hbytes > HAMMER2_MSGAUX_MAX) {
1040                         kprintf("hammer2: msgrd: bad header size %zd\n",
1041                                 hbytes);
1042                         error = EINVAL;
1043                         break;
1044                 }
1045                 msg = kmalloc(offsetof(struct hammer2_msg, any) + hbytes,
1046                               pmp->mmsg, M_WAITOK | M_ZERO);
1047                 msg->any.head = hdr;
1048                 msg->hdr_size = hbytes;
1049                 if (hbytes > sizeof(hdr)) {
1050                         error = fp_read(pmp->msg_fp, &msg->any.head + 1,
1051                                         hbytes - sizeof(hdr),
1052                                         NULL, 1, UIO_SYSSPACE);
1053                         if (error) {
1054                                 kprintf("hammer2: short msg received\n");
1055                                 error = EINVAL;
1056                                 break;
1057                         }
1058                 }
1059                 msg->aux_size = hdr.aux_bytes * HAMMER2_MSG_ALIGN;
1060                 if (msg->aux_size > HAMMER2_MSGAUX_MAX) {
1061                         kprintf("hammer2: illegal msg payload size %zd\n",
1062                                 msg->aux_size);
1063                         error = EINVAL;
1064                         break;
1065                 }
1066                 if (msg->aux_size) {
1067                         msg->aux_data = kmalloc(msg->aux_size, pmp->mmsg,
1068                                                 M_WAITOK | M_ZERO);
1069                         error = fp_read(pmp->msg_fp, msg->aux_data,
1070                                         msg->aux_size,
1071                                         NULL, 1, UIO_SYSSPACE);
1072                         if (error) {
1073                                 kprintf("hammer2: short msg "
1074                                         "payload received\n");
1075                                 break;
1076                         }
1077                 }
1078
1079                 /*
1080                  * State machine tracking, state assignment for msg,
1081                  * returns error and discard status.  Errors are fatal
1082                  * to the connection except for EALREADY which forces
1083                  * a discard without execution.
1084                  */
1085                 error = hammer2_state_msgrx(pmp, msg);
1086                 if (error) {
1087                         /*
1088                          * Raw protocol or connection error
1089                          */
1090                         hammer2_msg_free(pmp, msg);
1091                         if (error == EALREADY)
1092                                 error = 0;
1093                 } else if (msg->state && msg->state->func) {
1094                         /*
1095                          * Message related to state which already has a
1096                          * handling function installed for it.
1097                          */
1098                         error = msg->state->func(msg->state, msg);
1099                         hammer2_state_cleanuprx(pmp, msg);
1100                 } else if ((msg->any.head.cmd & HAMMER2_MSGF_PROTOS) ==
1101                            HAMMER2_MSG_PROTO_LNK) {
1102                         /*
1103                          * Message related to the LNK protocol set
1104                          */
1105                         error = hammer2_msg_lnk_rcvmsg(pmp, msg);
1106                         hammer2_state_cleanuprx(pmp, msg);
1107                 } else if ((msg->any.head.cmd & HAMMER2_MSGF_PROTOS) ==
1108                            HAMMER2_MSG_PROTO_DBG) {
1109                         /*
1110                          * Message related to the DBG protocol set
1111                          */
1112                         error = hammer2_msg_dbg_rcvmsg(pmp, msg);
1113                         hammer2_state_cleanuprx(pmp, msg);
1114                 } else {
1115                         /*
1116                          * Other higher-level messages (e.g. vnops)
1117                          */
1118                         error = hammer2_msg_adhoc_input(pmp, msg);
1119                         hammer2_state_cleanuprx(pmp, msg);
1120                 }
1121                 msg = NULL;
1122         }
1123
1124         if (error)
1125                 kprintf("hammer2: msg read failed error %d\n", error);
1126
1127         lockmgr(&pmp->msglk, LK_EXCLUSIVE);
1128         if (msg) {
1129                 if (msg->state && msg->state->msg == msg)
1130                         msg->state->msg = NULL;
1131                 hammer2_msg_free(pmp, msg);
1132         }
1133
1134         if ((state = pmp->freerd_state) != NULL) {
1135                 pmp->freerd_state = NULL;
1136                 hammer2_state_free(state);
1137         }
1138
1139         while ((state = RB_ROOT(&pmp->staterd_tree)) != NULL) {
1140                 RB_REMOVE(hammer2_state_tree, &pmp->staterd_tree, state);
1141                 hammer2_state_free(state);
1142         }
1143         lockmgr(&pmp->msglk, LK_RELEASE);
1144
1145         fp_shutdown(pmp->msg_fp, SHUT_RDWR);
1146         pmp->msgrd_td = NULL;
1147         /* pmp can be ripped out from under us at this point */
1148         wakeup(pmp);
1149         lwkt_exit();
1150 }
1151
1152 static
1153 void
1154 hammer2_cluster_thread_wr(void *arg)
1155 {
1156         hammer2_pfsmount_t *pmp = arg;
1157         hammer2_msg_t *msg = NULL;
1158         hammer2_state_t *state;
1159         ssize_t res;
1160         size_t name_len;
1161         int error = 0;
1162
1163         /*
1164          * Open a LNK_CONN transaction indicating that we want to take part
1165          * in the spanning tree algorithm.  Filter explicitly on the PFS
1166          * info in the iroot.
1167          *
1168          * We do not transmit our (only) LNK_SPAN until the other end has
1169          * acknowledged our link connection request.
1170          *
1171          * The transaction remains fully open for the duration of the
1172          * connection.
1173          */
1174         msg = hammer2_msg_alloc(pmp, 0, HAMMER2_LNK_CONN | HAMMER2_MSGF_CREATE);
1175         msg->any.lnk_conn.pfs_clid = pmp->iroot->ip_data.pfs_clid;
1176         msg->any.lnk_conn.pfs_fsid = pmp->iroot->ip_data.pfs_fsid;
1177         msg->any.lnk_conn.pfs_type = pmp->iroot->ip_data.pfs_type;
1178         msg->any.lnk_conn.proto_version = HAMMER2_SPAN_PROTO_1;
1179         name_len = pmp->iroot->ip_data.name_len;
1180         if (name_len >= sizeof(msg->any.lnk_conn.label))
1181                 name_len = sizeof(msg->any.lnk_conn.label) - 1;
1182         bcopy(pmp->iroot->ip_data.filename, msg->any.lnk_conn.label, name_len);
1183         msg->any.lnk_conn.label[name_len] = 0;
1184         hammer2_msg_write(pmp, msg, hammer2_msg_conn_reply, pmp);
1185
1186         /*
1187          * Transmit loop
1188          */
1189         msg = NULL;
1190         lockmgr(&pmp->msglk, LK_EXCLUSIVE);
1191
1192         while ((pmp->msg_ctl & HAMMER2_CLUSTERCTL_KILL) == 0 && error == 0) {
1193                 lksleep(&pmp->msg_ctl, &pmp->msglk, 0, "msgwr", hz);
1194                 while ((msg = TAILQ_FIRST(&pmp->msgq)) != NULL) {
1195                         /*
1196                          * Remove msg from the transmit queue and do
1197                          * persist and half-closed state handling.
1198                          */
1199                         TAILQ_REMOVE(&pmp->msgq, msg, qentry);
1200                         lockmgr(&pmp->msglk, LK_RELEASE);
1201
1202                         error = hammer2_state_msgtx(pmp, msg);
1203                         if (error == EALREADY) {
1204                                 error = 0;
1205                                 hammer2_msg_free(pmp, msg);
1206                                 lockmgr(&pmp->msglk, LK_EXCLUSIVE);
1207                                 continue;
1208                         }
1209                         if (error)
1210                                 break;
1211
1212                         /*
1213                          * Dump the message to the pipe or socket.
1214                          */
1215                         error = fp_write(pmp->msg_fp, &msg->any, msg->hdr_size,
1216                                          &res, UIO_SYSSPACE);
1217                         if (error || res != msg->hdr_size) {
1218                                 if (error == 0)
1219                                         error = EINVAL;
1220                                 lockmgr(&pmp->msglk, LK_EXCLUSIVE);
1221                                 break;
1222                         }
1223                         if (msg->aux_size) {
1224                                 error = fp_write(pmp->msg_fp,
1225                                                  msg->aux_data, msg->aux_size,
1226                                                  &res, UIO_SYSSPACE);
1227                                 if (error || res != msg->aux_size) {
1228                                         if (error == 0)
1229                                                 error = EINVAL;
1230                                         lockmgr(&pmp->msglk, LK_EXCLUSIVE);
1231                                         break;
1232                                 }
1233                         }
1234                         hammer2_state_cleanuptx(pmp, msg);
1235                         lockmgr(&pmp->msglk, LK_EXCLUSIVE);
1236                 }
1237         }
1238
1239         /*
1240          * Cleanup messages pending transmission and release msgq lock.
1241          */
1242         if (error)
1243                 kprintf("hammer2: msg write failed error %d\n", error);
1244
1245         if (msg) {
1246                 if (msg->state && msg->state->msg == msg)
1247                         msg->state->msg = NULL;
1248                 hammer2_msg_free(pmp, msg);
1249         }
1250
1251         while ((msg = TAILQ_FIRST(&pmp->msgq)) != NULL) {
1252                 TAILQ_REMOVE(&pmp->msgq, msg, qentry);
1253                 if (msg->state && msg->state->msg == msg)
1254                         msg->state->msg = NULL;
1255                 hammer2_msg_free(pmp, msg);
1256         }
1257
1258         if ((state = pmp->freewr_state) != NULL) {
1259                 pmp->freewr_state = NULL;
1260                 hammer2_state_free(state);
1261         }
1262
1263         while ((state = RB_ROOT(&pmp->statewr_tree)) != NULL) {
1264                 RB_REMOVE(hammer2_state_tree, &pmp->statewr_tree, state);
1265                 hammer2_state_free(state);
1266         }
1267         lockmgr(&pmp->msglk, LK_RELEASE);
1268
1269         /*
1270          * Cleanup descriptor, be sure the read size is shutdown so the
1271          * (probably blocked) read operations returns an error.
1272          *
1273          * pmp can be ripped out from under us once msgwr_td is set to NULL.
1274          */
1275         fp_shutdown(pmp->msg_fp, SHUT_RDWR);
1276         pmp->msgwr_td = NULL;
1277         wakeup(pmp);
1278         lwkt_exit();
1279 }
1280
1281 static int
1282 hammer2_msg_lnk_rcvmsg(hammer2_pfsmount_t *pmp, hammer2_msg_t *msg)
1283 {
1284         switch(msg->any.head.cmd & HAMMER2_MSGF_TRANSMASK) {
1285         case HAMMER2_LNK_CONN | HAMMER2_MSGF_CREATE:
1286                 kprintf("CONN RECEIVE - (just ignore it)\n");
1287                 hammer2_msg_result(pmp, msg, 0);
1288                 break;
1289         case HAMMER2_LNK_SPAN | HAMMER2_MSGF_CREATE:
1290                 kprintf("SPAN RECEIVE - ADDED FROM CLUSTER\n");
1291                 break;
1292         case HAMMER2_LNK_SPAN | HAMMER2_MSGF_DELETE:
1293                 kprintf("SPAN RECEIVE - DELETED FROM CLUSTER\n");
1294                 break;
1295         default:
1296                 break;
1297         }
1298         return(0);
1299 }
1300
1301 /*
1302  * This function is called when the other end replies to our LNK_CONN
1303  * request.
1304  *
1305  * We transmit our (single) SPAN on the initial reply, leaving that
1306  * transaction open too.
1307  */
1308 static int
1309 hammer2_msg_conn_reply(hammer2_state_t *state, hammer2_msg_t *msg)
1310 {
1311         hammer2_pfsmount_t *pmp = state->any.pmp;
1312         size_t name_len;
1313
1314         if (msg->any.head.cmd & HAMMER2_MSGF_CREATE) {
1315                 kprintf("LNK_CONN transaction replied to, initiate SPAN\n");
1316                 msg = hammer2_msg_alloc(pmp, 0, HAMMER2_LNK_SPAN |
1317                                                 HAMMER2_MSGF_CREATE);
1318                 msg->any.lnk_span.pfs_clid = pmp->iroot->ip_data.pfs_clid;
1319                 msg->any.lnk_span.pfs_fsid = pmp->iroot->ip_data.pfs_fsid;
1320                 msg->any.lnk_span.pfs_type = pmp->iroot->ip_data.pfs_type;
1321                 msg->any.lnk_span.proto_version = HAMMER2_SPAN_PROTO_1;
1322                 name_len = pmp->iroot->ip_data.name_len;
1323                 if (name_len >= sizeof(msg->any.lnk_span.label))
1324                         name_len = sizeof(msg->any.lnk_span.label) - 1;
1325                 bcopy(pmp->iroot->ip_data.filename,
1326                       msg->any.lnk_span.label,
1327                       name_len);
1328                 msg->any.lnk_span.label[name_len] = 0;
1329                 hammer2_msg_write(pmp, msg, hammer2_msg_span_reply, pmp);
1330         }
1331         if (msg->any.head.cmd & HAMMER2_MSGF_DELETE) {
1332                 kprintf("LNK_CONN transaction terminated by remote\n");
1333                 hammer2_msg_reply(pmp, msg, 0);
1334         }
1335         return(0);
1336 }
1337
1338 static int
1339 hammer2_msg_span_reply(hammer2_state_t *state, hammer2_msg_t *msg)
1340 {
1341         hammer2_pfsmount_t *pmp = state->any.pmp;
1342
1343         kprintf("SPAN REPLY - Our span was terminated? %p\n", pmp);
1344         return(0);
1345 }