hammer2 - dmsg blockdev work
[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
51 struct hammer2_sync_info {
52         int error;
53         int waitfor;
54 };
55
56 TAILQ_HEAD(hammer2_mntlist, hammer2_mount);
57 static struct hammer2_mntlist hammer2_mntlist;
58 static struct lock hammer2_mntlk;
59
60 int hammer2_debug;
61 int hammer2_cluster_enable = 1;
62 int hammer2_hardlink_enable = 1;
63 long hammer2_iod_file_read;
64 long hammer2_iod_meta_read;
65 long hammer2_iod_indr_read;
66 long hammer2_iod_file_write;
67 long hammer2_iod_meta_write;
68 long hammer2_iod_indr_write;
69 long hammer2_iod_volu_write;
70 long hammer2_ioa_file_read;
71 long hammer2_ioa_meta_read;
72 long hammer2_ioa_indr_read;
73 long hammer2_ioa_file_write;
74 long hammer2_ioa_meta_write;
75 long hammer2_ioa_indr_write;
76 long hammer2_ioa_volu_write;
77
78 SYSCTL_NODE(_vfs, OID_AUTO, hammer2, CTLFLAG_RW, 0, "HAMMER2 filesystem");
79
80 SYSCTL_INT(_vfs_hammer2, OID_AUTO, debug, CTLFLAG_RW,
81            &hammer2_debug, 0, "");
82 SYSCTL_INT(_vfs_hammer2, OID_AUTO, cluster_enable, CTLFLAG_RW,
83            &hammer2_cluster_enable, 0, "");
84 SYSCTL_INT(_vfs_hammer2, OID_AUTO, hardlink_enable, CTLFLAG_RW,
85            &hammer2_hardlink_enable, 0, "");
86 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_file_read, CTLFLAG_RW,
87            &hammer2_iod_file_read, 0, "");
88 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_meta_read, CTLFLAG_RW,
89            &hammer2_iod_meta_read, 0, "");
90 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_indr_read, CTLFLAG_RW,
91            &hammer2_iod_indr_read, 0, "");
92 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_file_write, CTLFLAG_RW,
93            &hammer2_iod_file_write, 0, "");
94 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_meta_write, CTLFLAG_RW,
95            &hammer2_iod_meta_write, 0, "");
96 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_indr_write, CTLFLAG_RW,
97            &hammer2_iod_indr_write, 0, "");
98 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_volu_write, CTLFLAG_RW,
99            &hammer2_iod_volu_write, 0, "");
100 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_file_read, CTLFLAG_RW,
101            &hammer2_ioa_file_read, 0, "");
102 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_meta_read, CTLFLAG_RW,
103            &hammer2_ioa_meta_read, 0, "");
104 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_indr_read, CTLFLAG_RW,
105            &hammer2_ioa_indr_read, 0, "");
106 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_file_write, CTLFLAG_RW,
107            &hammer2_ioa_file_write, 0, "");
108 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_meta_write, CTLFLAG_RW,
109            &hammer2_ioa_meta_write, 0, "");
110 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_indr_write, CTLFLAG_RW,
111            &hammer2_ioa_indr_write, 0, "");
112 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_volu_write, CTLFLAG_RW,
113            &hammer2_ioa_volu_write, 0, "");
114
115 static int hammer2_vfs_init(struct vfsconf *conf);
116 static int hammer2_vfs_mount(struct mount *mp, char *path, caddr_t data,
117                                 struct ucred *cred);
118 static int hammer2_remount(struct mount *, char *, struct vnode *,
119                                 struct ucred *);
120 static int hammer2_vfs_unmount(struct mount *mp, int mntflags);
121 static int hammer2_vfs_root(struct mount *mp, struct vnode **vpp);
122 static int hammer2_vfs_statfs(struct mount *mp, struct statfs *sbp,
123                                 struct ucred *cred);
124 static int hammer2_vfs_statvfs(struct mount *mp, struct statvfs *sbp,
125                                 struct ucred *cred);
126 static int hammer2_vfs_sync(struct mount *mp, int waitfor);
127 static int hammer2_vfs_vget(struct mount *mp, struct vnode *dvp,
128                                 ino_t ino, struct vnode **vpp);
129 static int hammer2_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
130                                 struct fid *fhp, struct vnode **vpp);
131 static int hammer2_vfs_vptofh(struct vnode *vp, struct fid *fhp);
132 static int hammer2_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
133                                 int *exflagsp, struct ucred **credanonp);
134
135 static int hammer2_install_volume_header(hammer2_mount_t *hmp);
136 static int hammer2_sync_scan1(struct mount *mp, struct vnode *vp, void *data);
137 static int hammer2_sync_scan2(struct mount *mp, struct vnode *vp, void *data);
138
139 static int hammer2_msg_conn_reply(kdmsg_state_t *state, kdmsg_msg_t *msg);
140 static int hammer2_msg_span_reply(kdmsg_state_t *state, kdmsg_msg_t *msg);
141 static int hammer2_msg_lnk_rcvmsg(kdmsg_msg_t *msg);
142
143 /*
144  * HAMMER2 vfs operations.
145  */
146 static struct vfsops hammer2_vfsops = {
147         .vfs_init       = hammer2_vfs_init,
148         .vfs_sync       = hammer2_vfs_sync,
149         .vfs_mount      = hammer2_vfs_mount,
150         .vfs_unmount    = hammer2_vfs_unmount,
151         .vfs_root       = hammer2_vfs_root,
152         .vfs_statfs     = hammer2_vfs_statfs,
153         .vfs_statvfs    = hammer2_vfs_statvfs,
154         .vfs_vget       = hammer2_vfs_vget,
155         .vfs_vptofh     = hammer2_vfs_vptofh,
156         .vfs_fhtovp     = hammer2_vfs_fhtovp,
157         .vfs_checkexp   = hammer2_vfs_checkexp
158 };
159
160 MALLOC_DEFINE(M_HAMMER2, "HAMMER2-mount", "");
161
162 VFS_SET(hammer2_vfsops, hammer2, 0);
163 MODULE_VERSION(hammer2, 1);
164
165 static
166 int
167 hammer2_vfs_init(struct vfsconf *conf)
168 {
169         int error;
170
171         error = 0;
172
173         if (HAMMER2_BLOCKREF_BYTES != sizeof(struct hammer2_blockref))
174                 error = EINVAL;
175         if (HAMMER2_INODE_BYTES != sizeof(struct hammer2_inode_data))
176                 error = EINVAL;
177         if (HAMMER2_ALLOCREF_BYTES != sizeof(struct hammer2_allocref))
178                 error = EINVAL;
179         if (HAMMER2_VOLUME_BYTES != sizeof(struct hammer2_volume_data))
180                 error = EINVAL;
181
182         if (error)
183                 kprintf("HAMMER2 structure size mismatch; cannot continue.\n");
184
185         lockinit(&hammer2_mntlk, "mntlk", 0, 0);
186         TAILQ_INIT(&hammer2_mntlist);
187
188         return (error);
189 }
190
191 /*
192  * Mount or remount HAMMER2 fileystem from physical media
193  *
194  *      mountroot
195  *              mp              mount point structure
196  *              path            NULL
197  *              data            <unused>
198  *              cred            <unused>
199  *
200  *      mount
201  *              mp              mount point structure
202  *              path            path to mount point
203  *              data            pointer to argument structure in user space
204  *                      volume  volume path (device@LABEL form)
205  *                      hflags  user mount flags
206  *              cred            user credentials
207  *
208  * RETURNS:     0       Success
209  *              !0      error number
210  */
211 static
212 int
213 hammer2_vfs_mount(struct mount *mp, char *path, caddr_t data,
214                   struct ucred *cred)
215 {
216         struct hammer2_mount_info info;
217         hammer2_pfsmount_t *pmp;
218         hammer2_mount_t *hmp;
219         hammer2_key_t lhc;
220         struct vnode *devvp;
221         struct nlookupdata nd;
222         hammer2_chain_t *parent;
223         hammer2_chain_t *schain;
224         hammer2_chain_t *rchain;
225         struct file *fp;
226         char devstr[MNAMELEN];
227         size_t size;
228         size_t done;
229         char *dev;
230         char *label;
231         int ronly = 1;
232         int create_hmp;
233         int error;
234
235         hmp = NULL;
236         pmp = NULL;
237         dev = NULL;
238         label = NULL;
239         devvp = NULL;
240
241         kprintf("hammer2_mount\n");
242
243         if (path == NULL) {
244                 /*
245                  * Root mount
246                  */
247                 bzero(&info, sizeof(info));
248                 info.cluster_fd = -1;
249                 return (EOPNOTSUPP);
250         } else {
251                 /*
252                  * Non-root mount or updating a mount
253                  */
254                 error = copyin(data, &info, sizeof(info));
255                 if (error)
256                         return (error);
257
258                 error = copyinstr(info.volume, devstr, MNAMELEN - 1, &done);
259                 if (error)
260                         return (error);
261
262                 /* Extract device and label */
263                 dev = devstr;
264                 label = strchr(devstr, '@');
265                 if (label == NULL ||
266                     ((label + 1) - dev) > done) {
267                         return (EINVAL);
268                 }
269                 *label = '\0';
270                 label++;
271                 if (*label == '\0')
272                         return (EINVAL);
273
274                 if (mp->mnt_flag & MNT_UPDATE) {
275                         /* Update mount */
276                         /* HAMMER2 implements NFS export via mountctl */
277                         hmp = MPTOHMP(mp);
278                         devvp = hmp->devvp;
279                         error = hammer2_remount(mp, path, devvp, cred);
280                         return error;
281                 }
282         }
283
284         /*
285          * PFS mount
286          *
287          * Lookup name and verify it refers to a block device.
288          */
289         error = nlookup_init(&nd, dev, UIO_SYSSPACE, NLC_FOLLOW);
290         if (error == 0)
291                 error = nlookup(&nd);
292         if (error == 0)
293                 error = cache_vref(&nd.nl_nch, nd.nl_cred, &devvp);
294         nlookup_done(&nd);
295
296         if (error == 0) {
297                 if (vn_isdisk(devvp, &error))
298                         error = vfs_mountedon(devvp);
299         }
300
301         /*
302          * Determine if the device has already been mounted.  After this
303          * check hmp will be non-NULL if we are doing the second or more
304          * hammer2 mounts from the same device.
305          */
306         lockmgr(&hammer2_mntlk, LK_EXCLUSIVE);
307         TAILQ_FOREACH(hmp, &hammer2_mntlist, mntentry) {
308                 if (hmp->devvp == devvp)
309                         break;
310         }
311
312         /*
313          * Open the device if this isn't a secondary mount
314          */
315         if (hmp) {
316                 create_hmp = 0;
317         } else {
318                 create_hmp = 1;
319                 if (error == 0 && vcount(devvp) > 0)
320                         error = EBUSY;
321
322                 /*
323                  * Now open the device
324                  */
325                 if (error == 0) {
326                         ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
327                         vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
328                         error = vinvalbuf(devvp, V_SAVE, 0, 0);
329                         if (error == 0) {
330                                 error = VOP_OPEN(devvp,
331                                                  ronly ? FREAD : FREAD | FWRITE,
332                                                  FSCRED, NULL);
333                         }
334                         vn_unlock(devvp);
335                 }
336                 if (error && devvp) {
337                         vrele(devvp);
338                         devvp = NULL;
339                 }
340                 if (error) {
341                         lockmgr(&hammer2_mntlk, LK_RELEASE);
342                         return error;
343                 }
344         }
345
346         /*
347          * Block device opened successfully, finish initializing the
348          * mount structure.
349          *
350          * From this point on we have to call hammer2_unmount() on failure.
351          */
352         pmp = kmalloc(sizeof(*pmp), M_HAMMER2, M_WAITOK | M_ZERO);
353         mp->mnt_data = (qaddr_t)pmp;
354         pmp->mp = mp;
355
356         kmalloc_create(&pmp->mmsg, "HAMMER2-pfsmsg");
357         kdmsg_iocom_init(&pmp->iocom, pmp, pmp->mmsg,
358                          hammer2_msg_lnk_rcvmsg,
359                          hammer2_msg_dbg_rcvmsg,
360                          hammer2_msg_adhoc_input);
361
362         if (create_hmp) {
363                 hmp = kmalloc(sizeof(*hmp), M_HAMMER2, M_WAITOK | M_ZERO);
364                 hmp->ronly = ronly;
365                 hmp->devvp = devvp;
366                 kmalloc_create(&hmp->minode, "HAMMER2-inodes");
367                 kmalloc_create(&hmp->mchain, "HAMMER2-chains");
368                 TAILQ_INSERT_TAIL(&hammer2_mntlist, hmp, mntentry);
369         }
370         ccms_domain_init(&pmp->ccms_dom);
371         pmp->hmp = hmp;
372         ++hmp->pmp_count;
373         lockmgr(&hammer2_mntlk, LK_RELEASE);
374         kprintf("hammer2_mount hmp=%p pmpcnt=%d\n", hmp, hmp->pmp_count);
375         
376         mp->mnt_flag = MNT_LOCAL;
377         mp->mnt_kern_flag |= MNTK_ALL_MPSAFE;   /* all entry pts are SMP */
378
379         if (create_hmp) {
380                 /*
381                  * vchain setup. vchain.data is special cased to NULL.
382                  * vchain.refs is initialized and will never drop to 0.
383                  */
384                 hmp->vchain.refs = 1;
385                 hmp->vchain.data = (void *)&hmp->voldata;
386                 hmp->vchain.bref.type = HAMMER2_BREF_TYPE_VOLUME;
387                 hmp->vchain.bref.data_off = 0 | HAMMER2_PBUFRADIX;
388                 hmp->vchain.bref_flush = hmp->vchain.bref;
389                 ccms_cst_init(&hmp->vchain.cst, NULL);
390                 /* hmp->vchain.u.xxx is left NULL */
391                 lockinit(&hmp->alloclk, "h2alloc", 0, 0);
392                 lockinit(&hmp->voldatalk, "voldata", 0, LK_CANRECURSE);
393
394                 /*
395                  * Install the volume header
396                  */
397                 error = hammer2_install_volume_header(hmp);
398                 if (error) {
399                         hammer2_vfs_unmount(mp, MNT_FORCE);
400                         return error;
401                 }
402         }
403
404         /*
405          * required mount structure initializations
406          */
407         mp->mnt_stat.f_iosize = HAMMER2_PBUFSIZE;
408         mp->mnt_stat.f_bsize = HAMMER2_PBUFSIZE;
409
410         mp->mnt_vstat.f_frsize = HAMMER2_PBUFSIZE;
411         mp->mnt_vstat.f_bsize = HAMMER2_PBUFSIZE;
412
413         /*
414          * Optional fields
415          */
416         mp->mnt_iosize_max = MAXPHYS;
417
418         /*
419          * First locate the super-root inode, which is key 0 relative to the
420          * volume header's blockset.
421          *
422          * Then locate the root inode by scanning the directory keyspace
423          * represented by the label.
424          */
425         if (create_hmp) {
426                 parent = &hmp->vchain;
427                 hammer2_chain_lock(hmp, parent, HAMMER2_RESOLVE_ALWAYS);
428                 schain = hammer2_chain_lookup(hmp, &parent,
429                                       HAMMER2_SROOT_KEY, HAMMER2_SROOT_KEY, 0);
430                 hammer2_chain_unlock(hmp, parent);
431                 if (schain == NULL) {
432                         kprintf("hammer2_mount: invalid super-root\n");
433                         hammer2_vfs_unmount(mp, MNT_FORCE);
434                         return EINVAL;
435                 }
436                 hammer2_chain_ref(hmp, schain); /* for hmp->schain */
437                 hmp->schain = schain;           /* left locked */
438         } else {
439                 schain = hmp->schain;
440                 hammer2_chain_lock(hmp, schain, HAMMER2_RESOLVE_ALWAYS);
441         }
442
443         parent = schain;
444         lhc = hammer2_dirhash(label, strlen(label));
445         rchain = hammer2_chain_lookup(hmp, &parent,
446                                       lhc, lhc + HAMMER2_DIRHASH_LOMASK,
447                                       0);
448         while (rchain) {
449                 if (rchain->bref.type == HAMMER2_BREF_TYPE_INODE &&
450                     rchain->u.ip &&
451                     strcmp(label, rchain->data->ipdata.filename) == 0) {
452                         break;
453                 }
454                 rchain = hammer2_chain_next(hmp, &parent, rchain,
455                                             lhc, lhc + HAMMER2_DIRHASH_LOMASK,
456                                             0);
457         }
458         hammer2_chain_unlock(hmp, parent);
459         if (rchain == NULL) {
460                 kprintf("hammer2_mount: PFS label not found\n");
461                 hammer2_vfs_unmount(mp, MNT_FORCE);
462                 return EINVAL;
463         }
464         if (rchain->flags & HAMMER2_CHAIN_MOUNTED) {
465                 hammer2_chain_unlock(hmp, rchain);
466                 kprintf("hammer2_mount: PFS label already mounted!\n");
467                 hammer2_vfs_unmount(mp, MNT_FORCE);
468                 return EBUSY;
469         }
470         atomic_set_int(&rchain->flags, HAMMER2_CHAIN_MOUNTED);
471
472         hammer2_chain_ref(hmp, rchain); /* for pmp->rchain */
473         hammer2_chain_unlock(hmp, rchain);
474         pmp->rchain = rchain;           /* left held & unlocked */
475         pmp->iroot = rchain->u.ip;      /* implied hold from rchain */
476         pmp->iroot->pmp = pmp;
477
478         kprintf("iroot %p\n", pmp->iroot);
479
480         /*
481          * Ref the cluster management messaging descriptor.  The mount
482          * program deals with the other end of the communications pipe.
483          */
484         fp = holdfp(curproc->p_fd, info.cluster_fd, -1);
485         if (fp == NULL) {
486                 kprintf("hammer2_mount: bad cluster_fd!\n");
487                 hammer2_vfs_unmount(mp, MNT_FORCE);
488                 return EBADF;
489         }
490         hammer2_cluster_reconnect(pmp, fp);
491
492         /*
493          * Finish setup
494          */
495         vfs_getnewfsid(mp);
496         vfs_add_vnodeops(mp, &hammer2_vnode_vops, &mp->mnt_vn_norm_ops);
497         vfs_add_vnodeops(mp, &hammer2_spec_vops, &mp->mnt_vn_spec_ops);
498         vfs_add_vnodeops(mp, &hammer2_fifo_vops, &mp->mnt_vn_fifo_ops);
499
500         copyinstr(info.volume, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, &size);
501         bzero(mp->mnt_stat.f_mntfromname + size, MNAMELEN - size);
502         bzero(mp->mnt_stat.f_mntonname, sizeof(mp->mnt_stat.f_mntonname));
503         copyinstr(path, mp->mnt_stat.f_mntonname,
504                   sizeof(mp->mnt_stat.f_mntonname) - 1,
505                   &size);
506
507         /*
508          * Initial statfs to prime mnt_stat.
509          */
510         hammer2_vfs_statfs(mp, &mp->mnt_stat, cred);
511
512         return 0;
513 }
514
515 static
516 int
517 hammer2_remount(struct mount *mp, char *path, struct vnode *devvp,
518                 struct ucred *cred)
519 {
520         return (0);
521 }
522
523 static
524 int
525 hammer2_vfs_unmount(struct mount *mp, int mntflags)
526 {
527         hammer2_pfsmount_t *pmp;
528         hammer2_mount_t *hmp;
529         int flags;
530         int error = 0;
531         int ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
532         struct vnode *devvp;
533
534         pmp = MPTOPMP(mp);
535         hmp = pmp->hmp;
536         flags = 0;
537
538         if (mntflags & MNT_FORCE)
539                 flags |= FORCECLOSE;
540
541         hammer2_mount_exlock(hmp);
542
543         /*
544          * If mount initialization proceeded far enough we must flush
545          * its vnodes.
546          */
547         if (pmp->iroot)
548                 error = vflush(mp, 0, flags);
549
550         if (error)
551                 return error;
552
553         lockmgr(&hammer2_mntlk, LK_EXCLUSIVE);
554         --hmp->pmp_count;
555         kprintf("hammer2_unmount hmp=%p pmpcnt=%d\n", hmp, hmp->pmp_count);
556
557         /*
558          * Flush any left over chains.  The voldata lock is only used
559          * to synchronize against HAMMER2_CHAIN_MODIFIED_AUX.
560          */
561         hammer2_voldata_lock(hmp);
562         if (hmp->vchain.flags & (HAMMER2_CHAIN_MODIFIED |
563                                  HAMMER2_CHAIN_MODIFIED_AUX |
564                                  HAMMER2_CHAIN_SUBMODIFIED)) {
565                 hammer2_voldata_unlock(hmp);
566                 hammer2_vfs_sync(mp, MNT_WAIT);
567         } else {
568                 hammer2_voldata_unlock(hmp);
569         }
570         if (hmp->pmp_count == 0) {
571                 if (hmp->vchain.flags & (HAMMER2_CHAIN_MODIFIED |
572                                          HAMMER2_CHAIN_MODIFIED_AUX |
573                                          HAMMER2_CHAIN_SUBMODIFIED)) {
574                         kprintf("hammer2_unmount: chains left over after "
575                                 "final sync\n");
576                         if (hammer2_debug & 0x0010)
577                                 Debugger("entered debugger");
578                 }
579         }
580
581         /*
582          * Cleanup the root and super-root chain elements (which should be
583          * clean).
584          */
585         pmp->iroot = NULL;
586         if (pmp->rchain) {
587                 atomic_clear_int(&pmp->rchain->flags, HAMMER2_CHAIN_MOUNTED);
588                 KKASSERT(pmp->rchain->refs == 1);
589                 hammer2_chain_drop(hmp, pmp->rchain);
590                 pmp->rchain = NULL;
591         }
592         ccms_domain_uninit(&pmp->ccms_dom);
593
594         /*
595          * Kill cluster controller
596          */
597         kdmsg_iocom_uninit(&pmp->iocom);
598
599         /*
600          * If no PFS's left drop the master hammer2_mount for the device.
601          */
602         if (hmp->pmp_count == 0) {
603                 if (hmp->schain) {
604                         KKASSERT(hmp->schain->refs == 1);
605                         hammer2_chain_drop(hmp, hmp->schain);
606                         hmp->schain = NULL;
607                 }
608
609                 /*
610                  * Finish up with the device vnode
611                  */
612                 if ((devvp = hmp->devvp) != NULL) {
613                         vinvalbuf(devvp, (ronly ? 0 : V_SAVE), 0, 0);
614                         hmp->devvp = NULL;
615                         VOP_CLOSE(devvp, (ronly ? FREAD : FREAD|FWRITE));
616                         vrele(devvp);
617                         devvp = NULL;
618                 }
619         }
620         hammer2_mount_unlock(hmp);
621
622         pmp->mp = NULL;
623         pmp->hmp = NULL;
624         mp->mnt_data = NULL;
625
626         kmalloc_destroy(&pmp->mmsg);
627
628         kfree(pmp, M_HAMMER2);
629         if (hmp->pmp_count == 0) {
630                 TAILQ_REMOVE(&hammer2_mntlist, hmp, mntentry);
631                 kmalloc_destroy(&hmp->minode);
632                 kmalloc_destroy(&hmp->mchain);
633                 kfree(hmp, M_HAMMER2);
634         }
635         lockmgr(&hammer2_mntlk, LK_RELEASE);
636         return (error);
637 }
638
639 static
640 int
641 hammer2_vfs_vget(struct mount *mp, struct vnode *dvp,
642              ino_t ino, struct vnode **vpp)
643 {
644         kprintf("hammer2_vget\n");
645         return (EOPNOTSUPP);
646 }
647
648 static
649 int
650 hammer2_vfs_root(struct mount *mp, struct vnode **vpp)
651 {
652         hammer2_pfsmount_t *pmp;
653         hammer2_mount_t *hmp;
654         int error;
655         struct vnode *vp;
656
657         pmp = MPTOPMP(mp);
658         hmp = pmp->hmp;
659         hammer2_mount_exlock(hmp);
660         if (pmp->iroot == NULL) {
661                 *vpp = NULL;
662                 error = EINVAL;
663         } else {
664                 hammer2_chain_lock(hmp, &pmp->iroot->chain,
665                                    HAMMER2_RESOLVE_ALWAYS |
666                                    HAMMER2_RESOLVE_SHARED);
667                 vp = hammer2_igetv(pmp->iroot, &error);
668                 hammer2_chain_unlock(hmp, &pmp->iroot->chain);
669                 *vpp = vp;
670                 if (vp == NULL)
671                         kprintf("vnodefail\n");
672         }
673         hammer2_mount_unlock(hmp);
674
675         return (error);
676 }
677
678 /*
679  * Filesystem status
680  *
681  * XXX incorporate pmp->iroot->ip_data.inode_quota and data_quota
682  */
683 static
684 int
685 hammer2_vfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
686 {
687         hammer2_pfsmount_t *pmp;
688         hammer2_mount_t *hmp;
689
690         pmp = MPTOPMP(mp);
691         hmp = MPTOHMP(mp);
692
693         mp->mnt_stat.f_files = pmp->iroot->ip_data.inode_count +
694                                pmp->iroot->delta_icount;
695         mp->mnt_stat.f_ffree = 0;
696         mp->mnt_stat.f_blocks = hmp->voldata.allocator_size / HAMMER2_PBUFSIZE;
697         mp->mnt_stat.f_bfree = (hmp->voldata.allocator_size -
698                                 hmp->voldata.allocator_beg) / HAMMER2_PBUFSIZE;
699         mp->mnt_stat.f_bavail = mp->mnt_stat.f_bfree;
700
701         *sbp = mp->mnt_stat;
702         return (0);
703 }
704
705 static
706 int
707 hammer2_vfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred)
708 {
709         hammer2_pfsmount_t *pmp;
710         hammer2_mount_t *hmp;
711
712         pmp = MPTOPMP(mp);
713         hmp = MPTOHMP(mp);
714
715         mp->mnt_vstat.f_bsize = HAMMER2_PBUFSIZE;
716         mp->mnt_vstat.f_files = pmp->iroot->ip_data.inode_count +
717                                 pmp->iroot->delta_icount;
718         mp->mnt_vstat.f_ffree = 0;
719         mp->mnt_vstat.f_blocks = hmp->voldata.allocator_size / HAMMER2_PBUFSIZE;
720         mp->mnt_vstat.f_bfree = (hmp->voldata.allocator_size -
721                                  hmp->voldata.allocator_beg) / HAMMER2_PBUFSIZE;
722         mp->mnt_vstat.f_bavail = mp->mnt_vstat.f_bfree;
723
724         *sbp = mp->mnt_vstat;
725         return (0);
726 }
727
728 /*
729  * Sync the entire filesystem; this is called from the filesystem syncer
730  * process periodically and whenever a user calls sync(1) on the hammer
731  * mountpoint.
732  *
733  * Currently is actually called from the syncer! \o/
734  *
735  * This task will have to snapshot the state of the dirty inode chain.
736  * From that, it will have to make sure all of the inodes on the dirty
737  * chain have IO initiated. We make sure that io is initiated for the root
738  * block.
739  *
740  * If waitfor is set, we wait for media to acknowledge the new rootblock.
741  *
742  * THINKS: side A vs side B, to have sync not stall all I/O?
743  */
744 static
745 int
746 hammer2_vfs_sync(struct mount *mp, int waitfor)
747 {
748         struct hammer2_sync_info info;
749         hammer2_mount_t *hmp;
750         int flags;
751         int error;
752         int haswork;
753
754         hmp = MPTOHMP(mp);
755
756         flags = VMSC_GETVP;
757         if (waitfor & MNT_LAZY)
758                 flags |= VMSC_ONEPASS;
759
760         info.error = 0;
761         info.waitfor = MNT_NOWAIT;
762         vmntvnodescan(mp, flags | VMSC_NOWAIT,
763                       hammer2_sync_scan1,
764                       hammer2_sync_scan2, &info);
765         if (info.error == 0 && (waitfor & MNT_WAIT)) {
766                 info.waitfor = waitfor;
767                     vmntvnodescan(mp, flags,
768                                   hammer2_sync_scan1,
769                                   hammer2_sync_scan2, &info);
770
771         }
772 #if 0
773         if (waitfor == MNT_WAIT) {
774                 /* XXX */
775         } else {
776                 /* XXX */
777         }
778 #endif
779         hammer2_chain_lock(hmp, &hmp->vchain, HAMMER2_RESOLVE_ALWAYS);
780         if (hmp->vchain.flags & (HAMMER2_CHAIN_MODIFIED |
781                                  HAMMER2_CHAIN_MODIFIED_AUX |
782                                  HAMMER2_CHAIN_SUBMODIFIED)) {
783                 hammer2_chain_flush(hmp, &hmp->vchain, 0);
784                 haswork = 1;
785         } else {
786                 haswork = 0;
787         }
788         hammer2_chain_unlock(hmp, &hmp->vchain);
789
790         error = 0;
791
792         if ((waitfor & MNT_LAZY) == 0) {
793                 waitfor = MNT_NOWAIT;
794                 vn_lock(hmp->devvp, LK_EXCLUSIVE | LK_RETRY);
795                 error = VOP_FSYNC(hmp->devvp, waitfor, 0);
796                 vn_unlock(hmp->devvp);
797         }
798
799         if (error == 0 && haswork) {
800                 struct buf *bp;
801
802                 /*
803                  * Synchronize the disk before flushing the volume
804                  * header.
805                  */
806                 bp = getpbuf(NULL);
807                 bp->b_bio1.bio_offset = 0;
808                 bp->b_bufsize = 0;
809                 bp->b_bcount = 0;
810                 bp->b_cmd = BUF_CMD_FLUSH;
811                 bp->b_bio1.bio_done = biodone_sync;
812                 bp->b_bio1.bio_flags |= BIO_SYNC;
813                 vn_strategy(hmp->devvp, &bp->b_bio1);
814                 biowait(&bp->b_bio1, "h2vol");
815                 relpbuf(bp, NULL);
816
817                 /*
818                  * Then we can safely flush the volume header.  Volume
819                  * data is locked separately to prevent ioctl functions
820                  * from deadlocking due to a configuration issue.
821                  */
822                 bp = getblk(hmp->devvp, 0, HAMMER2_PBUFSIZE, 0, 0);
823                 hammer2_voldata_lock(hmp);
824                 bcopy(&hmp->voldata, bp->b_data, HAMMER2_PBUFSIZE);
825                 hammer2_voldata_unlock(hmp);
826                 bawrite(bp);
827         }
828         return (error);
829 }
830
831 /*
832  * Sync passes.
833  *
834  * NOTE: We don't test SUBMODIFIED or MOVED here because the fsync code
835  *       won't flush on those flags.  The syncer code above will do a
836  *       general meta-data flush globally that will catch these flags.
837  */
838 static int
839 hammer2_sync_scan1(struct mount *mp, struct vnode *vp, void *data)
840 {
841         hammer2_inode_t *ip;
842
843         ip = VTOI(vp);
844         if (vp->v_type == VNON || ip == NULL ||
845             ((ip->chain.flags & (HAMMER2_CHAIN_MODIFIED |
846                                  HAMMER2_CHAIN_DIRTYEMBED)) == 0 &&
847              RB_EMPTY(&vp->v_rbdirty_tree))) {
848                 return(-1);
849         }
850         return(0);
851 }
852
853 static int
854 hammer2_sync_scan2(struct mount *mp, struct vnode *vp, void *data)
855 {
856         struct hammer2_sync_info *info = data;
857         hammer2_inode_t *ip;
858         int error;
859
860         ip = VTOI(vp);
861         if (vp->v_type == VNON || vp->v_type == VBAD ||
862             ((ip->chain.flags & (HAMMER2_CHAIN_MODIFIED |
863                                  HAMMER2_CHAIN_DIRTYEMBED)) == 0 &&
864             RB_EMPTY(&vp->v_rbdirty_tree))) {
865                 return(0);
866         }
867         error = VOP_FSYNC(vp, MNT_NOWAIT, 0);
868         if (error)
869                 info->error = error;
870         return(0);
871 }
872
873 static
874 int
875 hammer2_vfs_vptofh(struct vnode *vp, struct fid *fhp)
876 {
877         return (0);
878 }
879
880 static
881 int
882 hammer2_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
883                struct fid *fhp, struct vnode **vpp)
884 {
885         return (0);
886 }
887
888 static
889 int
890 hammer2_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
891                  int *exflagsp, struct ucred **credanonp)
892 {
893         return (0);
894 }
895
896 /*
897  * Support code for hammer2_mount().  Read, verify, and install the volume
898  * header into the HMP
899  *
900  * XXX read four volhdrs and use the one with the highest TID whos CRC
901  *     matches.
902  *
903  * XXX check iCRCs.
904  *
905  * XXX For filesystems w/ less than 4 volhdrs, make sure to not write to
906  *     nonexistant locations.
907  *
908  * XXX Record selected volhdr and ring updates to each of 4 volhdrs
909  */
910 static
911 int
912 hammer2_install_volume_header(hammer2_mount_t *hmp)
913 {
914         hammer2_volume_data_t *vd;
915         struct buf *bp;
916         hammer2_crc32_t crc0, crc, bcrc0, bcrc;
917         int error_reported;
918         int error;
919         int valid;
920         int i;
921
922         error_reported = 0;
923         error = 0;
924         valid = 0;
925         bp = NULL;
926
927         /*
928          * There are up to 4 copies of the volume header (syncs iterate
929          * between them so there is no single master).  We don't trust the
930          * volu_size field so we don't know precisely how large the filesystem
931          * is, so depend on the OS to return an error if we go beyond the
932          * block device's EOF.
933          */
934         for (i = 0; i < HAMMER2_NUM_VOLHDRS; i++) {
935                 error = bread(hmp->devvp, i * HAMMER2_ZONE_BYTES64,
936                               HAMMER2_VOLUME_BYTES, &bp);
937                 if (error) {
938                         brelse(bp);
939                         bp = NULL;
940                         continue;
941                 }
942
943                 vd = (struct hammer2_volume_data *) bp->b_data;
944                 if ((vd->magic != HAMMER2_VOLUME_ID_HBO) &&
945                     (vd->magic != HAMMER2_VOLUME_ID_ABO)) {
946                         brelse(bp);
947                         bp = NULL;
948                         continue;
949                 }
950
951                 if (vd->magic == HAMMER2_VOLUME_ID_ABO) {
952                         /* XXX: Reversed-endianness filesystem */
953                         kprintf("hammer2: reverse-endian filesystem detected");
954                         brelse(bp);
955                         bp = NULL;
956                         continue;
957                 }
958
959                 crc = vd->icrc_sects[HAMMER2_VOL_ICRC_SECT0];
960                 crc0 = hammer2_icrc32(bp->b_data + HAMMER2_VOLUME_ICRC0_OFF,
961                                       HAMMER2_VOLUME_ICRC0_SIZE);
962                 bcrc = vd->icrc_sects[HAMMER2_VOL_ICRC_SECT1];
963                 bcrc0 = hammer2_icrc32(bp->b_data + HAMMER2_VOLUME_ICRC1_OFF,
964                                        HAMMER2_VOLUME_ICRC1_SIZE);
965                 if ((crc0 != crc) || (bcrc0 != bcrc)) {
966                         kprintf("hammer2 volume header crc "
967                                 "mismatch copy #%d\t%08x %08x",
968                                 i, crc0, crc);
969                         error_reported = 1;
970                         brelse(bp);
971                         bp = NULL;
972                         continue;
973                 }
974                 if (valid == 0 || hmp->voldata.mirror_tid < vd->mirror_tid) {
975                         valid = 1;
976                         hmp->voldata = *vd;
977                 }
978                 brelse(bp);
979                 bp = NULL;
980         }
981         if (valid) {
982                 error = 0;
983                 if (error_reported)
984                         kprintf("hammer2: a valid volume header was found\n");
985         } else {
986                 error = EINVAL;
987                 kprintf("hammer2: no valid volume headers found!\n");
988         }
989         return (error);
990 }
991
992 /*
993  * Reconnect using the passed file pointer.  The caller must ref the
994  * fp for us.
995  */
996 void
997 hammer2_cluster_reconnect(hammer2_pfsmount_t *pmp, struct file *fp)
998 {
999         kdmsg_msg_t *msg;
1000         size_t name_len;
1001
1002         /*
1003          * Closes old comm descriptor, kills threads, cleans up
1004          * states, then installs the new descriptor and creates
1005          * new threads.
1006          */
1007         kdmsg_iocom_reconnect(&pmp->iocom, fp, "hammer2");
1008
1009         /*
1010          * Open a LNK_CONN transaction indicating that we want to take part
1011          * in the spanning tree algorithm.  Filter explicitly on the PFS
1012          * info in the iroot.
1013          *
1014          * We do not transmit our (only) LNK_SPAN until the other end has
1015          * acknowledged our link connection request.
1016          *
1017          * The transaction remains fully open for the duration of the
1018          * connection.
1019          */
1020         msg = kdmsg_msg_alloc(&pmp->iocom.router, DMSG_LNK_CONN | DMSGF_CREATE,
1021                                 hammer2_msg_conn_reply, pmp);
1022         msg->any.lnk_conn.pfs_clid = pmp->iroot->ip_data.pfs_clid;
1023         msg->any.lnk_conn.pfs_fsid = pmp->iroot->ip_data.pfs_fsid;
1024         msg->any.lnk_conn.pfs_type = pmp->iroot->ip_data.pfs_type;
1025         msg->any.lnk_conn.proto_version = DMSG_SPAN_PROTO_1;
1026         msg->any.lnk_conn.peer_type = pmp->hmp->voldata.peer_type;
1027         msg->any.lnk_conn.peer_mask = 1LLU << HAMMER2_PEER_HAMMER2;
1028         name_len = pmp->iroot->ip_data.name_len;
1029         if (name_len >= sizeof(msg->any.lnk_conn.fs_label))
1030                 name_len = sizeof(msg->any.lnk_conn.fs_label) - 1;
1031         bcopy(pmp->iroot->ip_data.filename,
1032               msg->any.lnk_conn.fs_label,
1033               name_len);
1034         pmp->iocom.conn_state = msg->state;
1035         msg->any.lnk_conn.fs_label[name_len] = 0;
1036         kdmsg_msg_write(msg);
1037 }
1038
1039 static int
1040 hammer2_msg_lnk_rcvmsg(kdmsg_msg_t *msg)
1041 {
1042         switch(msg->any.head.cmd & DMSGF_TRANSMASK) {
1043         case DMSG_LNK_CONN | DMSGF_CREATE:
1044                 /*
1045                  * reply & leave trans open
1046                  */
1047                 kprintf("CONN RECEIVE - (just ignore it)\n");
1048                 kdmsg_msg_result(msg, 0);
1049                 break;
1050         case DMSG_LNK_SPAN | DMSGF_CREATE:
1051                 kprintf("SPAN RECEIVE - ADDED FROM CLUSTER\n");
1052                 break;
1053         case DMSG_LNK_SPAN | DMSGF_DELETE:
1054                 kprintf("SPAN RECEIVE - DELETED FROM CLUSTER\n");
1055                 break;
1056         default:
1057                 break;
1058         }
1059         return(0);
1060 }
1061
1062 /*
1063  * This function is called when the other end replies to our LNK_CONN
1064  * request.
1065  *
1066  * We transmit our (single) SPAN on the initial reply, leaving that
1067  * transaction open too.
1068  */
1069 static int
1070 hammer2_msg_conn_reply(kdmsg_state_t *state, kdmsg_msg_t *msg)
1071 {
1072         hammer2_pfsmount_t *pmp = state->any.pmp;
1073         hammer2_mount_t *hmp = pmp->hmp;
1074         kdmsg_msg_t *rmsg;
1075         size_t name_len;
1076         int copyid;
1077
1078         kprintf("LNK_CONN REPLY RECEIVED CMD %08x\n", msg->any.head.cmd);
1079
1080         if (msg->any.head.cmd & DMSGF_CREATE) {
1081                 kprintf("LNK_CONN transaction replied to, initiate SPAN\n");
1082                 rmsg = kdmsg_msg_alloc(&pmp->iocom.router,
1083                                        DMSG_LNK_SPAN | DMSGF_CREATE,
1084                                        hammer2_msg_span_reply, pmp);
1085                 rmsg->any.lnk_span.pfs_clid = pmp->iroot->ip_data.pfs_clid;
1086                 rmsg->any.lnk_span.pfs_fsid = pmp->iroot->ip_data.pfs_fsid;
1087                 rmsg->any.lnk_span.pfs_type = pmp->iroot->ip_data.pfs_type;
1088                 rmsg->any.lnk_span.peer_type = pmp->hmp->voldata.peer_type;
1089                 rmsg->any.lnk_span.proto_version = DMSG_SPAN_PROTO_1;
1090                 name_len = pmp->iroot->ip_data.name_len;
1091                 if (name_len >= sizeof(rmsg->any.lnk_span.fs_label))
1092                         name_len = sizeof(rmsg->any.lnk_span.fs_label) - 1;
1093                 bcopy(pmp->iroot->ip_data.filename,
1094                       rmsg->any.lnk_span.fs_label,
1095                       name_len);
1096                 rmsg->any.lnk_span.fs_label[name_len] = 0;
1097                 kdmsg_msg_write(rmsg);
1098
1099                 /*
1100                  * Dump the configuration stored in the volume header
1101                  */
1102                 hammer2_voldata_lock(hmp);
1103                 for (copyid = 0; copyid < HAMMER2_COPYID_COUNT; ++copyid) {
1104                         if (hmp->voldata.copyinfo[copyid].copyid == 0)
1105                                 continue;
1106                         hammer2_volconf_update(pmp, copyid);
1107                 }
1108                 hammer2_voldata_unlock(hmp);
1109         }
1110         if ((state->txcmd & DMSGF_DELETE) == 0 &&
1111             (msg->any.head.cmd & DMSGF_DELETE)) {
1112                 kprintf("LNK_CONN transaction terminated by remote\n");
1113                 pmp->iocom.conn_state = NULL;
1114                 kdmsg_msg_reply(msg, 0);
1115         }
1116         return(0);
1117 }
1118
1119 /*
1120  * Remote terminated our span transaction.  We have to terminate our side.
1121  */
1122 static int
1123 hammer2_msg_span_reply(kdmsg_state_t *state, kdmsg_msg_t *msg)
1124 {
1125         /*hammer2_pfsmount_t *pmp = state->any.pmp;*/
1126
1127         kprintf("SPAN REPLY - Our sent span was terminated by the "
1128                 "remote %08x state %p\n", msg->any.head.cmd, state);
1129         if ((state->txcmd & DMSGF_DELETE) == 0 &&
1130             (msg->any.head.cmd & DMSGF_DELETE)) {
1131                 kdmsg_msg_reply(msg, 0);
1132         }
1133         return(0);
1134 }
1135
1136 /*
1137  * Volume configuration updates are passed onto the userland service
1138  * daemon via the open LNK_CONN transaction.
1139  */
1140 void
1141 hammer2_volconf_update(hammer2_pfsmount_t *pmp, int index)
1142 {
1143         hammer2_mount_t *hmp = pmp->hmp;
1144         kdmsg_msg_t *msg;
1145
1146         /* XXX interlock against connection state termination */
1147         kprintf("volconf update %p\n", pmp->iocom.conn_state);
1148         if (pmp->iocom.conn_state) {
1149                 kprintf("TRANSMIT VOLCONF VIA OPEN CONN TRANSACTION\n");
1150                 msg = kdmsg_msg_alloc(&pmp->iocom.router, DMSG_LNK_VOLCONF,
1151                                         NULL, NULL);
1152                 msg->state = pmp->iocom.conn_state;
1153                 msg->any.lnk_volconf.copy = hmp->voldata.copyinfo[index];
1154                 msg->any.lnk_volconf.mediaid = hmp->voldata.fsid;
1155                 msg->any.lnk_volconf.index = index;
1156                 kdmsg_msg_write(msg);
1157         }
1158 }