Merge branches 'hammer2' and 'master' of ssh://crater.dragonflybsd.org/repository...
[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
46 #include "hammer2.h"
47 #include "hammer2_disk.h"
48 #include "hammer2_mount.h"
49
50 struct hammer2_sync_info {
51         int error;
52         int waitfor;
53 };
54
55 TAILQ_HEAD(hammer2_mntlist, hammer2_mount);
56 static struct hammer2_mntlist hammer2_mntlist;
57 static struct lock hammer2_mntlk;
58
59 int hammer2_debug;
60 int hammer2_cluster_enable = 1;
61 int hammer2_hardlink_enable = 1;
62 long hammer2_iod_file_read;
63 long hammer2_iod_meta_read;
64 long hammer2_iod_indr_read;
65 long hammer2_iod_file_write;
66 long hammer2_iod_meta_write;
67 long hammer2_iod_indr_write;
68 long hammer2_iod_volu_write;
69 long hammer2_ioa_file_read;
70 long hammer2_ioa_meta_read;
71 long hammer2_ioa_indr_read;
72 long hammer2_ioa_file_write;
73 long hammer2_ioa_meta_write;
74 long hammer2_ioa_indr_write;
75 long hammer2_ioa_volu_write;
76
77 SYSCTL_NODE(_vfs, OID_AUTO, hammer2, CTLFLAG_RW, 0, "HAMMER2 filesystem");
78
79 SYSCTL_INT(_vfs_hammer2, OID_AUTO, debug, CTLFLAG_RW,
80            &hammer2_debug, 0, "");
81 SYSCTL_INT(_vfs_hammer2, OID_AUTO, cluster_enable, CTLFLAG_RW,
82            &hammer2_cluster_enable, 0, "");
83 SYSCTL_INT(_vfs_hammer2, OID_AUTO, hardlink_enable, CTLFLAG_RW,
84            &hammer2_hardlink_enable, 0, "");
85 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_file_read, CTLFLAG_RW,
86            &hammer2_iod_file_read, 0, "");
87 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_meta_read, CTLFLAG_RW,
88            &hammer2_iod_meta_read, 0, "");
89 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_indr_read, CTLFLAG_RW,
90            &hammer2_iod_indr_read, 0, "");
91 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_file_write, CTLFLAG_RW,
92            &hammer2_iod_file_write, 0, "");
93 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_meta_write, CTLFLAG_RW,
94            &hammer2_iod_meta_write, 0, "");
95 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_indr_write, CTLFLAG_RW,
96            &hammer2_iod_indr_write, 0, "");
97 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_volu_write, CTLFLAG_RW,
98            &hammer2_iod_volu_write, 0, "");
99 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_file_read, CTLFLAG_RW,
100            &hammer2_ioa_file_read, 0, "");
101 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_meta_read, CTLFLAG_RW,
102            &hammer2_ioa_meta_read, 0, "");
103 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_indr_read, CTLFLAG_RW,
104            &hammer2_ioa_indr_read, 0, "");
105 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_file_write, CTLFLAG_RW,
106            &hammer2_ioa_file_write, 0, "");
107 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_meta_write, CTLFLAG_RW,
108            &hammer2_ioa_meta_write, 0, "");
109 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_indr_write, CTLFLAG_RW,
110            &hammer2_ioa_indr_write, 0, "");
111 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_volu_write, CTLFLAG_RW,
112            &hammer2_ioa_volu_write, 0, "");
113
114 static int hammer2_vfs_init(struct vfsconf *conf);
115 static int hammer2_vfs_mount(struct mount *mp, char *path, caddr_t data,
116                                 struct ucred *cred);
117 static int hammer2_remount(struct mount *, char *, struct vnode *,
118                                 struct ucred *);
119 static int hammer2_vfs_unmount(struct mount *mp, int mntflags);
120 static int hammer2_vfs_root(struct mount *mp, struct vnode **vpp);
121 static int hammer2_vfs_statfs(struct mount *mp, struct statfs *sbp,
122                                 struct ucred *cred);
123 static int hammer2_vfs_statvfs(struct mount *mp, struct statvfs *sbp,
124                                 struct ucred *cred);
125 static int hammer2_vfs_sync(struct mount *mp, int waitfor);
126 static int hammer2_vfs_vget(struct mount *mp, struct vnode *dvp,
127                                 ino_t ino, struct vnode **vpp);
128 static int hammer2_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
129                                 struct fid *fhp, struct vnode **vpp);
130 static int hammer2_vfs_vptofh(struct vnode *vp, struct fid *fhp);
131 static int hammer2_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
132                                 int *exflagsp, struct ucred **credanonp);
133
134 static int hammer2_install_volume_header(hammer2_mount_t *hmp);
135 static int hammer2_sync_scan1(struct mount *mp, struct vnode *vp, void *data);
136 static int hammer2_sync_scan2(struct mount *mp, struct vnode *vp, void *data);
137
138 /*
139  * HAMMER2 vfs operations.
140  */
141 static struct vfsops hammer2_vfsops = {
142         .vfs_init       = hammer2_vfs_init,
143         .vfs_sync       = hammer2_vfs_sync,
144         .vfs_mount      = hammer2_vfs_mount,
145         .vfs_unmount    = hammer2_vfs_unmount,
146         .vfs_root       = hammer2_vfs_root,
147         .vfs_statfs     = hammer2_vfs_statfs,
148         .vfs_statvfs    = hammer2_vfs_statvfs,
149         .vfs_vget       = hammer2_vfs_vget,
150         .vfs_vptofh     = hammer2_vfs_vptofh,
151         .vfs_fhtovp     = hammer2_vfs_fhtovp,
152         .vfs_checkexp   = hammer2_vfs_checkexp
153 };
154
155 MALLOC_DEFINE(M_HAMMER2, "HAMMER2-mount", "");
156
157 VFS_SET(hammer2_vfsops, hammer2, 0);
158 MODULE_VERSION(hammer2, 1);
159
160 static
161 int
162 hammer2_vfs_init(struct vfsconf *conf)
163 {
164         int error;
165
166         error = 0;
167
168         if (HAMMER2_BLOCKREF_BYTES != sizeof(struct hammer2_blockref))
169                 error = EINVAL;
170         if (HAMMER2_INODE_BYTES != sizeof(struct hammer2_inode_data))
171                 error = EINVAL;
172         if (HAMMER2_ALLOCREF_BYTES != sizeof(struct hammer2_allocref))
173                 error = EINVAL;
174         if (HAMMER2_VOLUME_BYTES != sizeof(struct hammer2_volume_data))
175                 error = EINVAL;
176
177         if (error)
178                 kprintf("HAMMER2 structure size mismatch; cannot continue.\n");
179
180         lockinit(&hammer2_mntlk, "mntlk", 0, 0);
181         TAILQ_INIT(&hammer2_mntlist);
182
183         return (error);
184 }
185
186 /*
187  * Mount or remount HAMMER2 fileystem from physical media
188  *
189  *      mountroot
190  *              mp              mount point structure
191  *              path            NULL
192  *              data            <unused>
193  *              cred            <unused>
194  *
195  *      mount
196  *              mp              mount point structure
197  *              path            path to mount point
198  *              data            pointer to argument structure in user space
199  *                      volume  volume path (device@LABEL form)
200  *                      hflags  user mount flags
201  *              cred            user credentials
202  *
203  * RETURNS:     0       Success
204  *              !0      error number
205  */
206 static
207 int
208 hammer2_vfs_mount(struct mount *mp, char *path, caddr_t data,
209               struct ucred *cred)
210 {
211         struct hammer2_mount_info info;
212         hammer2_pfsmount_t *pmp;
213         hammer2_mount_t *hmp;
214         hammer2_key_t lhc;
215         struct vnode *devvp;
216         struct nlookupdata nd;
217         hammer2_chain_t *parent;
218         hammer2_chain_t *schain;
219         hammer2_chain_t *rchain;
220         char devstr[MNAMELEN];
221         size_t size;
222         size_t done;
223         char *dev;
224         char *label;
225         int ronly = 1;
226         int create_hmp;
227         int error;
228
229         hmp = NULL;
230         pmp = NULL;
231         dev = NULL;
232         label = NULL;
233         devvp = NULL;
234
235         kprintf("hammer2_mount\n");
236
237         if (path == NULL) {
238                 /*
239                  * Root mount
240                  */
241                 return (EOPNOTSUPP);
242         } else {
243                 /*
244                  * Non-root mount or updating a mount
245                  */
246                 error = copyin(data, &info, sizeof(info));
247                 if (error)
248                         return (error);
249
250                 error = copyinstr(info.volume, devstr, MNAMELEN - 1, &done);
251                 if (error)
252                         return (error);
253
254                 /* Extract device and label */
255                 dev = devstr;
256                 label = strchr(devstr, '@');
257                 if (label == NULL ||
258                     ((label + 1) - dev) > done) {
259                         return (EINVAL);
260                 }
261                 *label = '\0';
262                 label++;
263                 if (*label == '\0')
264                         return (EINVAL);
265
266                 if (mp->mnt_flag & MNT_UPDATE) {
267                         /* Update mount */
268                         /* HAMMER2 implements NFS export via mountctl */
269                         hmp = MPTOHMP(mp);
270                         devvp = hmp->devvp;
271                         error = hammer2_remount(mp, path, devvp, cred);
272                         return error;
273                 }
274         }
275
276         /*
277          * New non-root mount
278          */
279         /* Lookup name and verify it refers to a block device */
280         error = nlookup_init(&nd, dev, UIO_SYSSPACE, NLC_FOLLOW);
281         if (error == 0)
282                 error = nlookup(&nd);
283         if (error == 0)
284                 error = cache_vref(&nd.nl_nch, nd.nl_cred, &devvp);
285         nlookup_done(&nd);
286
287         if (error == 0) {
288                 if (vn_isdisk(devvp, &error))
289                         error = vfs_mountedon(devvp);
290         }
291
292         /*
293          * Determine if the device has already been mounted.  After this
294          * check hmp will be non-NULL if we are doing the second or more
295          * hammer2 mounts from the same device.
296          */
297         lockmgr(&hammer2_mntlk, LK_EXCLUSIVE);
298         TAILQ_FOREACH(hmp, &hammer2_mntlist, mntentry) {
299                 if (hmp->devvp == devvp)
300                         break;
301         }
302
303         /*
304          * Open the device if this isn't a secondary mount
305          */
306         if (hmp) {
307                 create_hmp = 0;
308         } else {
309                 create_hmp = 1;
310                 if (error == 0 && vcount(devvp) > 0)
311                         error = EBUSY;
312
313                 /*
314                  * Now open the device
315                  */
316                 if (error == 0) {
317                         ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
318                         vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
319                         error = vinvalbuf(devvp, V_SAVE, 0, 0);
320                         if (error == 0) {
321                                 error = VOP_OPEN(devvp,
322                                                  ronly ? FREAD : FREAD | FWRITE,
323                                                  FSCRED, NULL);
324                         }
325                         vn_unlock(devvp);
326                 }
327                 if (error && devvp) {
328                         vrele(devvp);
329                         devvp = NULL;
330                 }
331                 if (error) {
332                         lockmgr(&hammer2_mntlk, LK_RELEASE);
333                         return error;
334                 }
335         }
336
337         /*
338          * Block device opened successfully, finish initializing the
339          * mount structure.
340          *
341          * From this point on we have to call hammer2_unmount() on failure.
342          */
343         pmp = kmalloc(sizeof(*pmp), M_HAMMER2, M_WAITOK | M_ZERO);
344         mp->mnt_data = (qaddr_t)pmp;
345         pmp->mp = mp;
346
347         if (create_hmp) {
348                 hmp = kmalloc(sizeof(*hmp), M_HAMMER2, M_WAITOK | M_ZERO);
349                 hmp->ronly = ronly;
350                 hmp->devvp = devvp;
351                 kmalloc_create(&hmp->minode, "HAMMER2-inodes");
352                 kmalloc_create(&hmp->mchain, "HAMMER2-chains");
353                 TAILQ_INSERT_TAIL(&hammer2_mntlist, hmp, mntentry);
354         }
355         pmp->hmp = hmp;
356         ++hmp->pmp_count;
357         lockmgr(&hammer2_mntlk, LK_RELEASE);
358         kprintf("hammer2_mount hmp=%p pmpcnt=%d\n", hmp, hmp->pmp_count);
359         
360         mp->mnt_flag = MNT_LOCAL;
361         mp->mnt_kern_flag |= MNTK_ALL_MPSAFE;   /* all entry pts are SMP */
362
363         if (create_hmp) {
364                 /*
365                  * vchain setup. vchain.data is special cased to NULL.
366                  * vchain.refs is initialized and will never drop to 0.
367                  */
368                 hmp->vchain.refs = 1;
369                 hmp->vchain.data = (void *)&hmp->voldata;
370                 hmp->vchain.bref.type = HAMMER2_BREF_TYPE_VOLUME;
371                 hmp->vchain.bref.data_off = 0 | HAMMER2_PBUFRADIX;
372                 hmp->vchain.bref_flush = hmp->vchain.bref;
373                 /* hmp->vchain.u.xxx is left NULL */
374                 lockinit(&hmp->vchain.lk, "volume", 0, LK_CANRECURSE);
375                 lockinit(&hmp->alloclk, "h2alloc", 0, 0);
376                 lockinit(&hmp->voldatalk, "voldata", 0, LK_CANRECURSE);
377
378                 /*
379                  * Install the volume header
380                  */
381                 error = hammer2_install_volume_header(hmp);
382                 if (error) {
383                         hammer2_vfs_unmount(mp, MNT_FORCE);
384                         return error;
385                 }
386         }
387
388         /*
389          * required mount structure initializations
390          */
391         mp->mnt_stat.f_iosize = HAMMER2_PBUFSIZE;
392         mp->mnt_stat.f_bsize = HAMMER2_PBUFSIZE;
393
394         mp->mnt_vstat.f_frsize = HAMMER2_PBUFSIZE;
395         mp->mnt_vstat.f_bsize = HAMMER2_PBUFSIZE;
396
397         /*
398          * Optional fields
399          */
400         mp->mnt_iosize_max = MAXPHYS;
401
402         /*
403          * First locate the super-root inode, which is key 0 relative to the
404          * volume header's blockset.
405          *
406          * Then locate the root inode by scanning the directory keyspace
407          * represented by the label.
408          */
409         if (create_hmp) {
410                 parent = &hmp->vchain;
411                 hammer2_chain_lock(hmp, parent, HAMMER2_RESOLVE_ALWAYS);
412                 schain = hammer2_chain_lookup(hmp, &parent,
413                                       HAMMER2_SROOT_KEY, HAMMER2_SROOT_KEY, 0);
414                 hammer2_chain_unlock(hmp, parent);
415                 if (schain == NULL) {
416                         kprintf("hammer2_mount: invalid super-root\n");
417                         hammer2_vfs_unmount(mp, MNT_FORCE);
418                         return EINVAL;
419                 }
420                 hammer2_chain_ref(hmp, schain); /* for hmp->schain */
421                 hmp->schain = schain;           /* left locked */
422         } else {
423                 schain = hmp->schain;
424                 hammer2_chain_lock(hmp, schain, HAMMER2_RESOLVE_ALWAYS);
425         }
426
427         parent = schain;
428         lhc = hammer2_dirhash(label, strlen(label));
429         rchain = hammer2_chain_lookup(hmp, &parent,
430                                       lhc, lhc + HAMMER2_DIRHASH_LOMASK,
431                                       0);
432         while (rchain) {
433                 if (rchain->bref.type == HAMMER2_BREF_TYPE_INODE &&
434                     rchain->u.ip &&
435                     strcmp(label, rchain->data->ipdata.filename) == 0) {
436                         break;
437                 }
438                 rchain = hammer2_chain_next(hmp, &parent, rchain,
439                                             lhc, lhc + HAMMER2_DIRHASH_LOMASK,
440                                             0);
441         }
442         hammer2_chain_unlock(hmp, parent);
443         if (rchain == NULL) {
444                 kprintf("hammer2_mount: PFS label not found\n");
445                 hammer2_vfs_unmount(mp, MNT_FORCE);
446                 return EINVAL;
447         }
448         if (rchain->flags & HAMMER2_CHAIN_MOUNTED) {
449                 hammer2_chain_unlock(hmp, rchain);
450                 kprintf("hammer2_mount: PFS label already mounted!\n");
451                 hammer2_vfs_unmount(mp, MNT_FORCE);
452                 return EBUSY;
453         }
454         atomic_set_int(&rchain->flags, HAMMER2_CHAIN_MOUNTED);
455
456         hammer2_chain_ref(hmp, rchain); /* for pmp->rchain */
457         hammer2_chain_unlock(hmp, rchain);
458         pmp->rchain = rchain;           /* left held & unlocked */
459         pmp->iroot = rchain->u.ip;      /* implied hold from rchain */
460         pmp->iroot->pmp = pmp;
461         kprintf("iroot %p\n", pmp->iroot);
462
463         vfs_getnewfsid(mp);
464         vfs_add_vnodeops(mp, &hammer2_vnode_vops, &mp->mnt_vn_norm_ops);
465         vfs_add_vnodeops(mp, &hammer2_spec_vops, &mp->mnt_vn_spec_ops);
466         vfs_add_vnodeops(mp, &hammer2_fifo_vops, &mp->mnt_vn_fifo_ops);
467
468         copyinstr(info.volume, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, &size);
469         bzero(mp->mnt_stat.f_mntfromname + size, MNAMELEN - size);
470         bzero(mp->mnt_stat.f_mntonname, sizeof(mp->mnt_stat.f_mntonname));
471         copyinstr(path, mp->mnt_stat.f_mntonname,
472                   sizeof(mp->mnt_stat.f_mntonname) - 1,
473                   &size);
474
475         hammer2_vfs_statfs(mp, &mp->mnt_stat, cred);
476
477         return 0;
478 }
479
480 static
481 int
482 hammer2_remount(struct mount *mp, char *path, struct vnode *devvp,
483                 struct ucred *cred)
484 {
485         return (0);
486 }
487
488 static
489 int
490 hammer2_vfs_unmount(struct mount *mp, int mntflags)
491 {
492         hammer2_pfsmount_t *pmp;
493         hammer2_mount_t *hmp;
494         int flags;
495         int error = 0;
496         int ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
497         struct vnode *devvp;
498
499         pmp = MPTOPMP(mp);
500         hmp = pmp->hmp;
501         flags = 0;
502
503         if (mntflags & MNT_FORCE)
504                 flags |= FORCECLOSE;
505
506         hammer2_mount_exlock(hmp);
507
508         /*
509          * If mount initialization proceeded far enough we must flush
510          * its vnodes.
511          */
512         if (pmp->iroot)
513                 error = vflush(mp, 0, flags);
514
515         if (error)
516                 return error;
517
518         lockmgr(&hammer2_mntlk, LK_EXCLUSIVE);
519         --hmp->pmp_count;
520         kprintf("hammer2_unmount hmp=%p pmpcnt=%d\n", hmp, hmp->pmp_count);
521
522         /*
523          * Flush any left over chains.  The voldata lock is only used
524          * to synchronize against HAMMER2_CHAIN_MODIFIED_AUX.
525          */
526         hammer2_voldata_lock(hmp);
527         if (hmp->vchain.flags & (HAMMER2_CHAIN_MODIFIED |
528                                  HAMMER2_CHAIN_MODIFIED_AUX |
529                                  HAMMER2_CHAIN_SUBMODIFIED)) {
530                 hammer2_voldata_unlock(hmp);
531                 hammer2_vfs_sync(mp, MNT_WAIT);
532         } else {
533                 hammer2_voldata_unlock(hmp);
534         }
535         if (hmp->pmp_count == 0) {
536                 if (hmp->vchain.flags & (HAMMER2_CHAIN_MODIFIED |
537                                          HAMMER2_CHAIN_MODIFIED_AUX |
538                                          HAMMER2_CHAIN_SUBMODIFIED)) {
539                         kprintf("hammer2_unmount: chains left over after "
540                                 "final sync\n");
541                         if (hammer2_debug & 0x0010)
542                                 Debugger("entered debugger");
543                 }
544         }
545
546         /*
547          * Cleanup the root and super-root chain elements (which should be
548          * clean).
549          */
550         pmp->iroot = NULL;
551         if (pmp->rchain) {
552                 atomic_clear_int(&pmp->rchain->flags, HAMMER2_CHAIN_MOUNTED);
553                 KKASSERT(pmp->rchain->refs == 1);
554                 hammer2_chain_drop(hmp, pmp->rchain);
555                 pmp->rchain = NULL;
556         }
557         if (hmp->pmp_count == 0) {
558                 if (hmp->schain) {
559                         KKASSERT(hmp->schain->refs == 1);
560                         hammer2_chain_drop(hmp, hmp->schain);
561                         hmp->schain = NULL;
562                 }
563
564                 /*
565                  * Finish up with the device vnode
566                  */
567                 if ((devvp = hmp->devvp) != NULL) {
568                         vinvalbuf(devvp, (ronly ? 0 : V_SAVE), 0, 0);
569                         hmp->devvp = NULL;
570                         VOP_CLOSE(devvp, (ronly ? FREAD : FREAD|FWRITE));
571                         vrele(devvp);
572                         devvp = NULL;
573                 }
574         }
575         hammer2_mount_unlock(hmp);
576
577         pmp->mp = NULL;
578         pmp->hmp = NULL;
579         mp->mnt_data = NULL;
580
581         kfree(pmp, M_HAMMER2);
582         if (hmp->pmp_count == 0) {
583                 TAILQ_REMOVE(&hammer2_mntlist, hmp, mntentry);
584                 kmalloc_destroy(&hmp->minode);
585                 kmalloc_destroy(&hmp->mchain);
586                 kfree(hmp, M_HAMMER2);
587         }
588         lockmgr(&hammer2_mntlk, LK_RELEASE);
589         return (error);
590 }
591
592 static
593 int
594 hammer2_vfs_vget(struct mount *mp, struct vnode *dvp,
595              ino_t ino, struct vnode **vpp)
596 {
597         kprintf("hammer2_vget\n");
598         return (EOPNOTSUPP);
599 }
600
601 static
602 int
603 hammer2_vfs_root(struct mount *mp, struct vnode **vpp)
604 {
605         hammer2_pfsmount_t *pmp;
606         int error;
607         struct vnode *vp;
608
609         pmp = MPTOPMP(mp);
610         hammer2_mount_exlock(pmp->hmp);
611         if (pmp->iroot == NULL) {
612                 *vpp = NULL;
613                 error = EINVAL;
614         } else {
615                 vp = hammer2_igetv(pmp->iroot, &error);
616                 *vpp = vp;
617                 if (vp == NULL)
618                         kprintf("vnodefail\n");
619         }
620         hammer2_mount_unlock(pmp->hmp);
621
622         return (error);
623 }
624
625 /*
626  * Filesystem status
627  *
628  * XXX incorporate pmp->iroot->ip_data.inode_quota and data_quota
629  */
630 static
631 int
632 hammer2_vfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
633 {
634         hammer2_pfsmount_t *pmp;
635         hammer2_mount_t *hmp;
636
637         pmp = MPTOPMP(mp);
638         hmp = MPTOHMP(mp);
639
640         mp->mnt_stat.f_files = pmp->iroot->ip_data.inode_count +
641                                pmp->iroot->delta_icount;
642         mp->mnt_stat.f_ffree = 0;
643         mp->mnt_stat.f_blocks = hmp->voldata.allocator_size / HAMMER2_PBUFSIZE;
644         mp->mnt_stat.f_bfree = (hmp->voldata.allocator_size -
645                                 hmp->voldata.allocator_beg) / HAMMER2_PBUFSIZE;
646         mp->mnt_stat.f_bavail = mp->mnt_stat.f_bfree;
647
648         *sbp = mp->mnt_stat;
649         return (0);
650 }
651
652 static
653 int
654 hammer2_vfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred)
655 {
656         hammer2_pfsmount_t *pmp;
657         hammer2_mount_t *hmp;
658
659         pmp = MPTOPMP(mp);
660         hmp = MPTOHMP(mp);
661
662         mp->mnt_vstat.f_bsize = HAMMER2_PBUFSIZE;
663         mp->mnt_vstat.f_files = pmp->iroot->ip_data.inode_count +
664                                 pmp->iroot->delta_icount;
665         mp->mnt_vstat.f_ffree = 0;
666         mp->mnt_vstat.f_blocks = hmp->voldata.allocator_size / HAMMER2_PBUFSIZE;
667         mp->mnt_vstat.f_bfree = (hmp->voldata.allocator_size -
668                                  hmp->voldata.allocator_beg) / HAMMER2_PBUFSIZE;
669         mp->mnt_vstat.f_bavail = mp->mnt_vstat.f_bfree;
670
671         *sbp = mp->mnt_vstat;
672         return (0);
673 }
674
675 /*
676  * Sync the entire filesystem; this is called from the filesystem syncer
677  * process periodically and whenever a user calls sync(1) on the hammer
678  * mountpoint.
679  *
680  * Currently is actually called from the syncer! \o/
681  *
682  * This task will have to snapshot the state of the dirty inode chain.
683  * From that, it will have to make sure all of the inodes on the dirty
684  * chain have IO initiated. We make sure that io is initiated for the root
685  * block.
686  *
687  * If waitfor is set, we wait for media to acknowledge the new rootblock.
688  *
689  * THINKS: side A vs side B, to have sync not stall all I/O?
690  */
691 static
692 int
693 hammer2_vfs_sync(struct mount *mp, int waitfor)
694 {
695         struct hammer2_sync_info info;
696         hammer2_mount_t *hmp;
697         int flags;
698         int error;
699         int haswork;
700
701         hmp = MPTOHMP(mp);
702
703         flags = VMSC_GETVP;
704         if (waitfor & MNT_LAZY)
705                 flags |= VMSC_ONEPASS;
706
707         info.error = 0;
708         info.waitfor = MNT_NOWAIT;
709         vmntvnodescan(mp, flags | VMSC_NOWAIT,
710                       hammer2_sync_scan1,
711                       hammer2_sync_scan2, &info);
712         if (info.error == 0 && (waitfor & MNT_WAIT)) {
713                 info.waitfor = waitfor;
714                     vmntvnodescan(mp, flags,
715                                   hammer2_sync_scan1,
716                                   hammer2_sync_scan2, &info);
717
718         }
719 #if 0
720         if (waitfor == MNT_WAIT) {
721                 /* XXX */
722         } else {
723                 /* XXX */
724         }
725 #endif
726         hammer2_chain_lock(hmp, &hmp->vchain, HAMMER2_RESOLVE_ALWAYS);
727         if (hmp->vchain.flags & (HAMMER2_CHAIN_MODIFIED |
728                                  HAMMER2_CHAIN_MODIFIED_AUX |
729                                  HAMMER2_CHAIN_SUBMODIFIED)) {
730                 hammer2_chain_flush(hmp, &hmp->vchain, 0);
731                 haswork = 1;
732         } else {
733                 haswork = 0;
734         }
735         hammer2_chain_unlock(hmp, &hmp->vchain);
736
737         error = 0;
738
739         if ((waitfor & MNT_LAZY) == 0) {
740                 waitfor = MNT_NOWAIT;
741                 vn_lock(hmp->devvp, LK_EXCLUSIVE | LK_RETRY);
742                 error = VOP_FSYNC(hmp->devvp, waitfor, 0);
743                 vn_unlock(hmp->devvp);
744         }
745
746         if (error == 0 && haswork) {
747                 struct buf *bp;
748
749                 /*
750                  * Synchronize the disk before flushing the volume
751                  * header.
752                  */
753                 bp = getpbuf(NULL);
754                 bp->b_bio1.bio_offset = 0;
755                 bp->b_bufsize = 0;
756                 bp->b_bcount = 0;
757                 bp->b_cmd = BUF_CMD_FLUSH;
758                 bp->b_bio1.bio_done = biodone_sync;
759                 bp->b_bio1.bio_flags |= BIO_SYNC;
760                 vn_strategy(hmp->devvp, &bp->b_bio1);
761                 biowait(&bp->b_bio1, "h2vol");
762                 relpbuf(bp, NULL);
763
764                 /*
765                  * Then we can safely flush the volume header.  Volume
766                  * data is locked separately to prevent ioctl functions
767                  * from deadlocking due to a configuration issue.
768                  */
769                 bp = getblk(hmp->devvp, 0, HAMMER2_PBUFSIZE, 0, 0);
770                 hammer2_voldata_lock(hmp);
771                 bcopy(&hmp->voldata, bp->b_data, HAMMER2_PBUFSIZE);
772                 hammer2_voldata_unlock(hmp);
773                 bawrite(bp);
774         }
775         return (error);
776 }
777
778 /*
779  * Sync passes.
780  *
781  * NOTE: We don't test SUBMODIFIED or MOVED here because the fsync code
782  *       won't flush on those flags.  The syncer code above will do a
783  *       general meta-data flush globally that will catch these flags.
784  */
785 static int
786 hammer2_sync_scan1(struct mount *mp, struct vnode *vp, void *data)
787 {
788         hammer2_inode_t *ip;
789
790         ip = VTOI(vp);
791         if (vp->v_type == VNON || ip == NULL ||
792             ((ip->chain.flags & (HAMMER2_CHAIN_MODIFIED |
793                                  HAMMER2_CHAIN_DIRTYEMBED)) == 0 &&
794              RB_EMPTY(&vp->v_rbdirty_tree))) {
795                 return(-1);
796         }
797         return(0);
798 }
799
800 static int
801 hammer2_sync_scan2(struct mount *mp, struct vnode *vp, void *data)
802 {
803         struct hammer2_sync_info *info = data;
804         hammer2_inode_t *ip;
805         int error;
806
807         ip = VTOI(vp);
808         if (vp->v_type == VNON || vp->v_type == VBAD ||
809             ((ip->chain.flags & (HAMMER2_CHAIN_MODIFIED |
810                                  HAMMER2_CHAIN_DIRTYEMBED)) == 0 &&
811             RB_EMPTY(&vp->v_rbdirty_tree))) {
812                 return(0);
813         }
814         error = VOP_FSYNC(vp, MNT_NOWAIT, 0);
815         if (error)
816                 info->error = error;
817         return(0);
818 }
819
820 static
821 int
822 hammer2_vfs_vptofh(struct vnode *vp, struct fid *fhp)
823 {
824         return (0);
825 }
826
827 static
828 int
829 hammer2_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
830                struct fid *fhp, struct vnode **vpp)
831 {
832         return (0);
833 }
834
835 static
836 int
837 hammer2_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
838                  int *exflagsp, struct ucred **credanonp)
839 {
840         return (0);
841 }
842
843 /*
844  * Support code for hammer2_mount().  Read, verify, and install the volume
845  * header into the HMP
846  *
847  * XXX read four volhdrs and use the one with the highest TID whos CRC
848  *     matches.
849  *
850  * XXX check iCRCs.
851  *
852  * XXX For filesystems w/ less than 4 volhdrs, make sure to not write to
853  *     nonexistant locations.
854  *
855  * XXX Record selected volhdr and ring updates to each of 4 volhdrs
856  */
857 static
858 int
859 hammer2_install_volume_header(hammer2_mount_t *hmp)
860 {
861         hammer2_volume_data_t *vd;
862         struct buf *bp;
863         hammer2_crc32_t crc0, crc, bcrc0, bcrc;
864         int error_reported;
865         int error;
866         int valid;
867         int i;
868
869         error_reported = 0;
870         error = 0;
871         valid = 0;
872         bp = NULL;
873
874         /*
875          * There are up to 4 copies of the volume header (syncs iterate
876          * between them so there is no single master).  We don't trust the
877          * volu_size field so we don't know precisely how large the filesystem
878          * is, so depend on the OS to return an error if we go beyond the
879          * block device's EOF.
880          */
881         for (i = 0; i < HAMMER2_NUM_VOLHDRS; i++) {
882                 error = bread(hmp->devvp, i * HAMMER2_ZONE_BYTES64,
883                               HAMMER2_VOLUME_BYTES, &bp);
884                 if (error) {
885                         brelse(bp);
886                         bp = NULL;
887                         continue;
888                 }
889
890                 vd = (struct hammer2_volume_data *) bp->b_data;
891                 if ((vd->magic != HAMMER2_VOLUME_ID_HBO) &&
892                     (vd->magic != HAMMER2_VOLUME_ID_ABO)) {
893                         brelse(bp);
894                         bp = NULL;
895                         continue;
896                 }
897
898                 if (vd->magic == HAMMER2_VOLUME_ID_ABO) {
899                         /* XXX: Reversed-endianness filesystem */
900                         kprintf("hammer2: reverse-endian filesystem detected");
901                         brelse(bp);
902                         bp = NULL;
903                         continue;
904                 }
905
906                 crc = vd->icrc_sects[HAMMER2_VOL_ICRC_SECT0];
907                 crc0 = hammer2_icrc32(bp->b_data + HAMMER2_VOLUME_ICRC0_OFF,
908                                       HAMMER2_VOLUME_ICRC0_SIZE);
909                 bcrc = vd->icrc_sects[HAMMER2_VOL_ICRC_SECT1];
910                 bcrc0 = hammer2_icrc32(bp->b_data + HAMMER2_VOLUME_ICRC1_OFF,
911                                        HAMMER2_VOLUME_ICRC1_SIZE);
912                 if ((crc0 != crc) || (bcrc0 != bcrc)) {
913                         kprintf("hammer2 volume header crc "
914                                 "mismatch copy #%d\t%08x %08x",
915                                 i, crc0, crc);
916                         error_reported = 1;
917                         brelse(bp);
918                         bp = NULL;
919                         continue;
920                 }
921                 if (valid == 0 || hmp->voldata.mirror_tid < vd->mirror_tid) {
922                         valid = 1;
923                         hmp->voldata = *vd;
924                 }
925                 brelse(bp);
926                 bp = NULL;
927         }
928         if (valid) {
929                 error = 0;
930                 if (error_reported)
931                         kprintf("hammer2: a valid volume header was found\n");
932         } else {
933                 error = EINVAL;
934                 kprintf("hammer2: no valid volume headers found!\n");
935         }
936         return (error);
937 }
938