HAMMER VFS - Take reserved space into account when reporting statvfs data
[dragonfly.git] / sys / vfs / hammer / hammer_vfsops.c
1 /*
2  * Copyright (c) 2007-2008 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  * $DragonFly: src/sys/vfs/hammer/hammer_vfsops.c,v 1.74 2008/11/13 02:18:43 dillon Exp $
35  */
36
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/kernel.h>
40 #include <sys/vnode.h>
41 #include <sys/mount.h>
42 #include <sys/malloc.h>
43 #include <sys/nlookup.h>
44 #include <sys/fcntl.h>
45 #include <sys/sysctl.h>
46 #include <sys/buf.h>
47 #include <sys/buf2.h>
48 #include "hammer.h"
49
50 int hammer_debug_io;
51 int hammer_debug_general;
52 int hammer_debug_debug = 1;             /* medium-error panics */ 
53 int hammer_debug_inode;
54 int hammer_debug_locks;
55 int hammer_debug_btree;
56 int hammer_debug_tid;
57 int hammer_debug_recover;               /* -1 will disable, +1 will force */
58 int hammer_debug_recover_faults;
59 int hammer_cluster_enable = 1;          /* enable read clustering by default */
60 int hammer_count_fsyncs;
61 int hammer_count_inodes;
62 int hammer_count_iqueued;
63 int hammer_count_reclaiming;
64 int hammer_count_records;
65 int hammer_count_record_datas;
66 int hammer_count_volumes;
67 int hammer_count_buffers;
68 int hammer_count_nodes;
69 int64_t hammer_count_extra_space_used;
70 int64_t hammer_stats_btree_lookups;
71 int64_t hammer_stats_btree_searches;
72 int64_t hammer_stats_btree_inserts;
73 int64_t hammer_stats_btree_deletes;
74 int64_t hammer_stats_btree_elements;
75 int64_t hammer_stats_btree_splits;
76 int64_t hammer_stats_btree_iterations;
77 int64_t hammer_stats_btree_root_iterations;
78 int64_t hammer_stats_record_iterations;
79
80 int64_t hammer_stats_file_read;
81 int64_t hammer_stats_file_write;
82 int64_t hammer_stats_file_iopsr;
83 int64_t hammer_stats_file_iopsw;
84 int64_t hammer_stats_disk_read;
85 int64_t hammer_stats_disk_write;
86 int64_t hammer_stats_inode_flushes;
87 int64_t hammer_stats_commits;
88
89 int hammer_count_dirtybufspace;         /* global */
90 int hammer_count_refedbufs;             /* global */
91 int hammer_count_reservations;
92 int hammer_count_io_running_read;
93 int hammer_count_io_running_write;
94 int hammer_count_io_locked;
95 int hammer_limit_dirtybufspace;         /* per-mount */
96 int hammer_limit_recs;                  /* as a whole XXX */
97 int hammer_autoflush = 2000;            /* auto flush */
98 int hammer_bio_count;
99 int hammer_verify_zone;
100 int hammer_verify_data = 1;
101 int hammer_write_mode;
102 int64_t hammer_contention_count;
103 int64_t hammer_zone_limit;
104
105 SYSCTL_NODE(_vfs, OID_AUTO, hammer, CTLFLAG_RW, 0, "HAMMER filesystem");
106 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_general, CTLFLAG_RW,
107            &hammer_debug_general, 0, "");
108 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_io, CTLFLAG_RW,
109            &hammer_debug_io, 0, "");
110 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_debug, CTLFLAG_RW,
111            &hammer_debug_debug, 0, "");
112 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_inode, CTLFLAG_RW,
113            &hammer_debug_inode, 0, "");
114 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_locks, CTLFLAG_RW,
115            &hammer_debug_locks, 0, "");
116 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_btree, CTLFLAG_RW,
117            &hammer_debug_btree, 0, "");
118 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_tid, CTLFLAG_RW,
119            &hammer_debug_tid, 0, "");
120 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover, CTLFLAG_RW,
121            &hammer_debug_recover, 0, "");
122 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover_faults, CTLFLAG_RW,
123            &hammer_debug_recover_faults, 0, "");
124 SYSCTL_INT(_vfs_hammer, OID_AUTO, cluster_enable, CTLFLAG_RW,
125            &hammer_cluster_enable, 0, "");
126
127 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_dirtybufspace, CTLFLAG_RW,
128            &hammer_limit_dirtybufspace, 0, "");
129 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_recs, CTLFLAG_RW,
130            &hammer_limit_recs, 0, "");
131
132 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_fsyncs, CTLFLAG_RD,
133            &hammer_count_fsyncs, 0, "");
134 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_inodes, CTLFLAG_RD,
135            &hammer_count_inodes, 0, "");
136 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_iqueued, CTLFLAG_RD,
137            &hammer_count_iqueued, 0, "");
138 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reclaiming, CTLFLAG_RD,
139            &hammer_count_reclaiming, 0, "");
140 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_records, CTLFLAG_RD,
141            &hammer_count_records, 0, "");
142 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_record_datas, CTLFLAG_RD,
143            &hammer_count_record_datas, 0, "");
144 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_volumes, CTLFLAG_RD,
145            &hammer_count_volumes, 0, "");
146 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_buffers, CTLFLAG_RD,
147            &hammer_count_buffers, 0, "");
148 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_nodes, CTLFLAG_RD,
149            &hammer_count_nodes, 0, "");
150 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, count_extra_space_used, CTLFLAG_RD,
151            &hammer_count_extra_space_used, 0, "");
152
153 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_searches, CTLFLAG_RD,
154            &hammer_stats_btree_searches, 0, "");
155 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_lookups, CTLFLAG_RD,
156            &hammer_stats_btree_lookups, 0, "");
157 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_inserts, CTLFLAG_RD,
158            &hammer_stats_btree_inserts, 0, "");
159 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_deletes, CTLFLAG_RD,
160            &hammer_stats_btree_deletes, 0, "");
161 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_elements, CTLFLAG_RD,
162            &hammer_stats_btree_elements, 0, "");
163 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_splits, CTLFLAG_RD,
164            &hammer_stats_btree_splits, 0, "");
165 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_iterations, CTLFLAG_RD,
166            &hammer_stats_btree_iterations, 0, "");
167 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_root_iterations, CTLFLAG_RD,
168            &hammer_stats_btree_root_iterations, 0, "");
169 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_record_iterations, CTLFLAG_RD,
170            &hammer_stats_record_iterations, 0, "");
171
172 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_read, CTLFLAG_RD,
173            &hammer_stats_file_read, 0, "");
174 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_write, CTLFLAG_RD,
175            &hammer_stats_file_write, 0, "");
176 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_iopsr, CTLFLAG_RD,
177            &hammer_stats_file_iopsr, 0, "");
178 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_iopsw, CTLFLAG_RD,
179            &hammer_stats_file_iopsw, 0, "");
180 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_disk_read, CTLFLAG_RD,
181            &hammer_stats_disk_read, 0, "");
182 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_disk_write, CTLFLAG_RD,
183            &hammer_stats_disk_write, 0, "");
184 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_inode_flushes, CTLFLAG_RD,
185            &hammer_stats_inode_flushes, 0, "");
186 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_commits, CTLFLAG_RD,
187            &hammer_stats_commits, 0, "");
188
189 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_dirtybufspace, CTLFLAG_RD,
190            &hammer_count_dirtybufspace, 0, "");
191 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_refedbufs, CTLFLAG_RD,
192            &hammer_count_refedbufs, 0, "");
193 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reservations, CTLFLAG_RD,
194            &hammer_count_reservations, 0, "");
195 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_running_read, CTLFLAG_RD,
196            &hammer_count_io_running_read, 0, "");
197 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_locked, CTLFLAG_RD,
198            &hammer_count_io_locked, 0, "");
199 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_running_write, CTLFLAG_RD,
200            &hammer_count_io_running_write, 0, "");
201 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, zone_limit, CTLFLAG_RW,
202            &hammer_zone_limit, 0, "");
203 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, contention_count, CTLFLAG_RW,
204            &hammer_contention_count, 0, "");
205 SYSCTL_INT(_vfs_hammer, OID_AUTO, autoflush, CTLFLAG_RW,
206            &hammer_autoflush, 0, "");
207 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_zone, CTLFLAG_RW,
208            &hammer_verify_zone, 0, "");
209 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_data, CTLFLAG_RW,
210            &hammer_verify_data, 0, "");
211 SYSCTL_INT(_vfs_hammer, OID_AUTO, write_mode, CTLFLAG_RW,
212            &hammer_write_mode, 0, "");
213
214 KTR_INFO_MASTER(hammer);
215
216 /*
217  * VFS ABI
218  */
219 static void     hammer_free_hmp(struct mount *mp);
220
221 static int      hammer_vfs_mount(struct mount *mp, char *path, caddr_t data,
222                                 struct ucred *cred);
223 static int      hammer_vfs_unmount(struct mount *mp, int mntflags);
224 static int      hammer_vfs_root(struct mount *mp, struct vnode **vpp);
225 static int      hammer_vfs_statfs(struct mount *mp, struct statfs *sbp,
226                                 struct ucred *cred);
227 static int      hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp,
228                                 struct ucred *cred);
229 static int      hammer_vfs_sync(struct mount *mp, int waitfor);
230 static int      hammer_vfs_vget(struct mount *mp, ino_t ino,
231                                 struct vnode **vpp);
232 static int      hammer_vfs_init(struct vfsconf *conf);
233 static int      hammer_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
234                                 struct fid *fhp, struct vnode **vpp);
235 static int      hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp);
236 static int      hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
237                                 int *exflagsp, struct ucred **credanonp);
238
239
240 static struct vfsops hammer_vfsops = {
241         .vfs_mount      = hammer_vfs_mount,
242         .vfs_unmount    = hammer_vfs_unmount,
243         .vfs_root       = hammer_vfs_root,
244         .vfs_statfs     = hammer_vfs_statfs,
245         .vfs_statvfs    = hammer_vfs_statvfs,
246         .vfs_sync       = hammer_vfs_sync,
247         .vfs_vget       = hammer_vfs_vget,
248         .vfs_init       = hammer_vfs_init,
249         .vfs_vptofh     = hammer_vfs_vptofh,
250         .vfs_fhtovp     = hammer_vfs_fhtovp,
251         .vfs_checkexp   = hammer_vfs_checkexp
252 };
253
254 MALLOC_DEFINE(M_HAMMER, "HAMMER-mount", "");
255
256 VFS_SET(hammer_vfsops, hammer, 0);
257 MODULE_VERSION(hammer, 1);
258
259 static int
260 hammer_vfs_init(struct vfsconf *conf)
261 {
262         int n;
263
264         if (hammer_limit_recs == 0) {
265                 hammer_limit_recs = nbuf * 25;
266                 n = kmalloc_limit(M_HAMMER) / 512;
267                 if (hammer_limit_recs > n)
268                         hammer_limit_recs = n;
269         }
270         if (hammer_limit_dirtybufspace == 0) {
271                 hammer_limit_dirtybufspace = hidirtybufspace / 2;
272                 if (hammer_limit_dirtybufspace < 100)
273                         hammer_limit_dirtybufspace = 100;
274         }
275         return(0);
276 }
277
278 static int
279 hammer_vfs_mount(struct mount *mp, char *mntpt, caddr_t data,
280                  struct ucred *cred)
281 {
282         struct hammer_mount_info info;
283         hammer_mount_t hmp;
284         hammer_volume_t rootvol;
285         struct vnode *rootvp;
286         struct vnode *devvp = NULL;
287         const char *upath;      /* volume name in userspace */
288         char *path;             /* volume name in system space */
289         int error;
290         int i;
291         int master_id;
292         int maxinodes;
293
294         /*
295          * Accept hammer_mount_info.  mntpt is NULL for root mounts at boot.
296          */
297         if (mntpt == NULL) {
298                 if ((error = bdevvp(rootdev, &devvp))) {
299                         kprintf("hammer_mountroot: can't find devvp\n");
300                         return (error);
301                 }
302                 mp->mnt_flag &= ~MNT_RDONLY; /* mount R/W */
303                 bzero(&info, sizeof(info));
304                 info.asof = 0;
305                 info.hflags = 0;
306                 info.nvolumes = 1;
307         } else {
308                 if ((error = copyin(data, &info, sizeof(info))) != 0)
309                         return (error);
310         }
311
312         /*
313          * updating or new mount
314          */
315         if (mp->mnt_flag & MNT_UPDATE) {
316                 hmp = (void *)mp->mnt_data;
317                 KKASSERT(hmp != NULL);
318         } else {
319                 if (info.nvolumes <= 0 || info.nvolumes >= 32768)
320                         return (EINVAL);
321                 hmp = NULL;
322         }
323
324         /*
325          * master-id validation.  The master id may not be changed by a
326          * mount update.
327          */
328         if (info.hflags & HMNT_MASTERID) {
329                 if (hmp && hmp->master_id != info.master_id) {
330                         kprintf("hammer: cannot change master id "
331                                 "with mount update\n");
332                         return(EINVAL);
333                 }
334                 master_id = info.master_id;
335                 if (master_id < -1 || master_id >= HAMMER_MAX_MASTERS)
336                         return (EINVAL);
337         } else {
338                 if (hmp)
339                         master_id = hmp->master_id;
340                 else
341                         master_id = 0;
342         }
343
344         /*
345          * Interal mount data structure
346          */
347         if (hmp == NULL) {
348                 hmp = kmalloc(sizeof(*hmp), M_HAMMER, M_WAITOK | M_ZERO);
349                 mp->mnt_data = (qaddr_t)hmp;
350                 hmp->mp = mp;
351                 /*TAILQ_INIT(&hmp->recycle_list);*/
352
353                 /*
354                  * Make sure kmalloc type limits are set appropriately.  If root
355                  * increases the vnode limit you may have to do a dummy remount
356                  * to adjust the HAMMER inode limit.
357                  */
358                 kmalloc_create(&hmp->m_misc, "HAMMER-others");
359                 kmalloc_create(&hmp->m_inodes, "HAMMER-inodes");
360
361                 maxinodes = desiredvnodes + desiredvnodes / 5 +
362                             HAMMER_RECLAIM_WAIT;
363                 kmalloc_raise_limit(hmp->m_inodes,
364                                     maxinodes * sizeof(struct hammer_inode));
365
366                 hmp->root_btree_beg.localization = 0x00000000U;
367                 hmp->root_btree_beg.obj_id = -0x8000000000000000LL;
368                 hmp->root_btree_beg.key = -0x8000000000000000LL;
369                 hmp->root_btree_beg.create_tid = 1;
370                 hmp->root_btree_beg.delete_tid = 1;
371                 hmp->root_btree_beg.rec_type = 0;
372                 hmp->root_btree_beg.obj_type = 0;
373
374                 hmp->root_btree_end.localization = 0xFFFFFFFFU;
375                 hmp->root_btree_end.obj_id = 0x7FFFFFFFFFFFFFFFLL;
376                 hmp->root_btree_end.key = 0x7FFFFFFFFFFFFFFFLL;
377                 hmp->root_btree_end.create_tid = 0xFFFFFFFFFFFFFFFFULL;
378                 hmp->root_btree_end.delete_tid = 0;   /* special case */
379                 hmp->root_btree_end.rec_type = 0xFFFFU;
380                 hmp->root_btree_end.obj_type = 0;
381
382                 hmp->krate.freq = 1;    /* maximum reporting rate (hz) */
383                 hmp->krate.count = -16; /* initial burst */
384
385                 hmp->sync_lock.refs = 1;
386                 hmp->free_lock.refs = 1;
387                 hmp->undo_lock.refs = 1;
388                 hmp->blkmap_lock.refs = 1;
389
390                 TAILQ_INIT(&hmp->delay_list);
391                 TAILQ_INIT(&hmp->flush_group_list);
392                 TAILQ_INIT(&hmp->objid_cache_list);
393                 TAILQ_INIT(&hmp->undo_lru_list);
394                 TAILQ_INIT(&hmp->reclaim_list);
395         }
396         hmp->hflags &= ~HMNT_USERFLAGS;
397         hmp->hflags |= info.hflags & HMNT_USERFLAGS;
398
399         hmp->master_id = master_id;
400
401         if (info.asof) {
402                 mp->mnt_flag |= MNT_RDONLY;
403                 hmp->asof = info.asof;
404         } else {
405                 hmp->asof = HAMMER_MAX_TID;
406         }
407
408         /*
409          * Re-open read-write if originally read-only, or vise-versa.
410          */
411         if (mp->mnt_flag & MNT_UPDATE) {
412                 error = 0;
413                 if (hmp->ronly && (mp->mnt_kern_flag & MNTK_WANTRDWR)) {
414                         kprintf("HAMMER read-only -> read-write\n");
415                         hmp->ronly = 0;
416                         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
417                                 hammer_adjust_volume_mode, NULL);
418                         rootvol = hammer_get_root_volume(hmp, &error);
419                         if (rootvol) {
420                                 hammer_recover_flush_buffers(hmp, rootvol, 1);
421                                 bcopy(rootvol->ondisk->vol0_blockmap,
422                                       hmp->blockmap,
423                                       sizeof(hmp->blockmap));
424                                 hammer_rel_volume(rootvol, 0);
425                         }
426                         RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
427                                 hammer_reload_inode, NULL);
428                         /* kernel clears MNT_RDONLY */
429                 } else if (hmp->ronly == 0 && (mp->mnt_flag & MNT_RDONLY)) {
430                         kprintf("HAMMER read-write -> read-only\n");
431                         hmp->ronly = 1; /* messy */
432                         RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
433                                 hammer_reload_inode, NULL);
434                         hmp->ronly = 0;
435                         hammer_flusher_sync(hmp);
436                         hammer_flusher_sync(hmp);
437                         hammer_flusher_sync(hmp);
438                         hmp->ronly = 1;
439                         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
440                                 hammer_adjust_volume_mode, NULL);
441                 }
442                 return(error);
443         }
444
445         RB_INIT(&hmp->rb_vols_root);
446         RB_INIT(&hmp->rb_inos_root);
447         RB_INIT(&hmp->rb_nods_root);
448         RB_INIT(&hmp->rb_undo_root);
449         RB_INIT(&hmp->rb_resv_root);
450         RB_INIT(&hmp->rb_bufs_root);
451         RB_INIT(&hmp->rb_pfsm_root);
452
453         hmp->ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
454
455         TAILQ_INIT(&hmp->volu_list);
456         TAILQ_INIT(&hmp->undo_list);
457         TAILQ_INIT(&hmp->data_list);
458         TAILQ_INIT(&hmp->meta_list);
459         TAILQ_INIT(&hmp->lose_list);
460
461         /*
462          * Load volumes
463          */
464         path = objcache_get(namei_oc, M_WAITOK);
465         hmp->nvolumes = -1;
466         for (i = 0; i < info.nvolumes; ++i) {
467                 if (mntpt == NULL) {
468                         /*
469                          * Root mount.
470                          * Only one volume; and no need for copyin.
471                          */
472                         KKASSERT(info.nvolumes == 1);
473                         ksnprintf(path, MAXPATHLEN, "/dev/%s",
474                                   mp->mnt_stat.f_mntfromname);  
475                         error = 0;
476                 } else {
477                         error = copyin(&info.volumes[i], &upath,
478                                        sizeof(char *));
479                         if (error == 0)
480                                 error = copyinstr(upath, path,
481                                                   MAXPATHLEN, NULL);
482                 }
483                 if (error == 0)
484                         error = hammer_install_volume(hmp, path, devvp);
485                 if (error)
486                         break;
487         }
488         objcache_put(namei_oc, path);
489
490         /*
491          * Make sure we found a root volume
492          */
493         if (error == 0 && hmp->rootvol == NULL) {
494                 kprintf("hammer_mount: No root volume found!\n");
495                 error = EINVAL;
496         }
497
498         /*
499          * Check that all required volumes are available
500          */
501         if (error == 0 && hammer_mountcheck_volumes(hmp)) {
502                 kprintf("hammer_mount: Missing volumes, cannot mount!\n");
503                 error = EINVAL;
504         }
505
506         if (error) {
507                 hammer_free_hmp(mp);
508                 return (error);
509         }
510
511         /*
512          * No errors, setup enough of the mount point so we can lookup the
513          * root vnode.
514          */
515         mp->mnt_iosize_max = MAXPHYS;
516         mp->mnt_kern_flag |= MNTK_FSMID;
517
518         /* 
519          * note: f_iosize is used by vnode_pager_haspage() when constructing
520          * its VOP_BMAP call.
521          */
522         mp->mnt_stat.f_iosize = HAMMER_BUFSIZE;
523         mp->mnt_stat.f_bsize = HAMMER_BUFSIZE;
524
525         mp->mnt_vstat.f_frsize = HAMMER_BUFSIZE;
526         mp->mnt_vstat.f_bsize = HAMMER_BUFSIZE;
527
528         mp->mnt_maxsymlinklen = 255;
529         mp->mnt_flag |= MNT_LOCAL;
530
531         vfs_add_vnodeops(mp, &hammer_vnode_vops, &mp->mnt_vn_norm_ops);
532         vfs_add_vnodeops(mp, &hammer_spec_vops, &mp->mnt_vn_spec_ops);
533         vfs_add_vnodeops(mp, &hammer_fifo_vops, &mp->mnt_vn_fifo_ops);
534
535         /*
536          * The root volume's ondisk pointer is only valid if we hold a
537          * reference to it.
538          */
539         rootvol = hammer_get_root_volume(hmp, &error);
540         if (error)
541                 goto failed;
542
543         /*
544          * Perform any necessary UNDO operations.  The recovery code does
545          * call hammer_undo_lookup() so we have to pre-cache the blockmap,
546          * and then re-copy it again after recovery is complete.
547          *
548          * If this is a read-only mount the UNDO information is retained
549          * in memory in the form of dirty buffer cache buffers, and not
550          * written back to the media.
551          */
552         bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
553               sizeof(hmp->blockmap));
554
555         /*
556          * Check filesystem version
557          */
558         hmp->version = rootvol->ondisk->vol_version;
559         if (hmp->version < HAMMER_VOL_VERSION_MIN ||
560             hmp->version > HAMMER_VOL_VERSION_MAX) {
561                 kprintf("HAMMER: mount unsupported fs version %d\n",
562                         hmp->version);
563                 error = ERANGE;
564                 goto done;
565         }
566
567         /*
568          * The undo_rec_limit limits the size of flush groups to avoid
569          * blowing out the UNDO FIFO.  This calculation is typically in
570          * the tens of thousands and is designed primarily when small
571          * HAMMER filesystems are created.
572          */
573         hmp->undo_rec_limit = hammer_undo_max(hmp) / 8192 + 100;
574         if (hammer_debug_general & 0x0001)
575                 kprintf("HAMMER: undo_rec_limit %d\n", hmp->undo_rec_limit);
576
577         error = hammer_recover(hmp, rootvol);
578         if (error) {
579                 kprintf("Failed to recover HAMMER filesystem on mount\n");
580                 goto done;
581         }
582
583         /*
584          * Finish setup now that we have a good root volume.
585          *
586          * The top 16 bits of fsid.val[1] is a pfs id.
587          */
588         ksnprintf(mp->mnt_stat.f_mntfromname,
589                   sizeof(mp->mnt_stat.f_mntfromname), "%s",
590                   rootvol->ondisk->vol_name);
591         mp->mnt_stat.f_fsid.val[0] =
592                 crc32((char *)&rootvol->ondisk->vol_fsid + 0, 8);
593         mp->mnt_stat.f_fsid.val[1] =
594                 crc32((char *)&rootvol->ondisk->vol_fsid + 8, 8);
595         mp->mnt_stat.f_fsid.val[1] &= 0x0000FFFF;
596
597         mp->mnt_vstat.f_fsid_uuid = rootvol->ondisk->vol_fsid;
598         mp->mnt_vstat.f_fsid = crc32(&mp->mnt_vstat.f_fsid_uuid,
599                                      sizeof(mp->mnt_vstat.f_fsid_uuid));
600
601         /*
602          * Certain often-modified fields in the root volume are cached in
603          * the hammer_mount structure so we do not have to generate lots
604          * of little UNDO structures for them.
605          *
606          * Recopy after recovery.  This also has the side effect of
607          * setting our cached undo FIFO's first_offset, which serves to
608          * placemark the FIFO start for the NEXT flush cycle while the
609          * on-disk first_offset represents the LAST flush cycle.
610          */
611         hmp->next_tid = rootvol->ondisk->vol0_next_tid;
612         hmp->flush_tid1 = hmp->next_tid;
613         hmp->flush_tid2 = hmp->next_tid;
614         bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
615               sizeof(hmp->blockmap));
616         hmp->copy_stat_freebigblocks = rootvol->ondisk->vol0_stat_freebigblocks;
617
618         hammer_flusher_create(hmp);
619
620         /*
621          * Locate the root directory using the root cluster's B-Tree as a
622          * starting point.  The root directory uses an obj_id of 1.
623          *
624          * FUTURE: Leave the root directory cached referenced but unlocked
625          * in hmp->rootvp (need to flush it on unmount).
626          */
627         error = hammer_vfs_vget(mp, 1, &rootvp);
628         if (error)
629                 goto done;
630         vput(rootvp);
631         /*vn_unlock(hmp->rootvp);*/
632
633 done:
634         hammer_rel_volume(rootvol, 0);
635 failed:
636         /*
637          * Cleanup and return.
638          */
639         if (error)
640                 hammer_free_hmp(mp);
641         return (error);
642 }
643
644 static int
645 hammer_vfs_unmount(struct mount *mp, int mntflags)
646 {
647 #if 0
648         struct hammer_mount *hmp = (void *)mp->mnt_data;
649 #endif
650         int flags;
651         int error;
652
653         /*
654          * Clean out the vnodes
655          */
656         flags = 0;
657         if (mntflags & MNT_FORCE)
658                 flags |= FORCECLOSE;
659         if ((error = vflush(mp, 0, flags)) != 0)
660                 return (error);
661
662         /*
663          * Clean up the internal mount structure and related entities.  This
664          * may issue I/O.
665          */
666         hammer_free_hmp(mp);
667         return(0);
668 }
669
670 /*
671  * Clean up the internal mount structure and disassociate it from the mount.
672  * This may issue I/O.
673  */
674 static void
675 hammer_free_hmp(struct mount *mp)
676 {
677         struct hammer_mount *hmp = (void *)mp->mnt_data;
678         hammer_flush_group_t flg;
679         int count;
680         int dummy;
681
682         /*
683          * Flush anything dirty.  This won't even run if the
684          * filesystem errored-out.
685          */
686         count = 0;
687         while (hammer_flusher_haswork(hmp)) {
688                 hammer_flusher_sync(hmp);
689                 ++count;
690                 if (count >= 5) {
691                         if (count == 5)
692                                 kprintf("HAMMER: umount flushing.");
693                         else
694                                 kprintf(".");
695                         tsleep(&dummy, 0, "hmrufl", hz);
696                 }
697                 if (count == 30) {
698                         kprintf("giving up\n");
699                         break;
700                 }
701         }
702         if (count >= 5 && count < 30)
703                 kprintf("\n");
704
705         /*
706          * If the mount had a critical error we have to destroy any
707          * remaining inodes before we can finish cleaning up the flusher.
708          */
709         if (hmp->flags & HAMMER_MOUNT_CRITICAL_ERROR) {
710                 RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
711                         hammer_destroy_inode_callback, NULL);
712         }
713
714         /*
715          * There shouldn't be any inodes left now and any left over
716          * flush groups should now be empty.
717          */
718         KKASSERT(RB_EMPTY(&hmp->rb_inos_root));
719         while ((flg = TAILQ_FIRST(&hmp->flush_group_list)) != NULL) {
720                 TAILQ_REMOVE(&hmp->flush_group_list, flg, flush_entry);
721                 KKASSERT(TAILQ_EMPTY(&flg->flush_list));
722                 if (flg->refs) {
723                         kprintf("HAMMER: Warning, flush_group %p was "
724                                 "not empty on umount!\n", flg);
725                 }
726                 kfree(flg, hmp->m_misc);
727         }
728
729         /*
730          * We can finally destroy the flusher
731          */
732         hammer_flusher_destroy(hmp);
733
734         /*
735          * We may have held recovered buffers due to a read-only mount.
736          * These must be discarded.
737          */
738         if (hmp->ronly)
739                 hammer_recover_flush_buffers(hmp, NULL, -1);
740
741         /*
742          * Unload buffers and then volumes
743          */
744         RB_SCAN(hammer_buf_rb_tree, &hmp->rb_bufs_root, NULL,
745                 hammer_unload_buffer, NULL);
746         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
747                 hammer_unload_volume, NULL);
748
749         mp->mnt_data = NULL;
750         mp->mnt_flag &= ~MNT_LOCAL;
751         hmp->mp = NULL;
752         hammer_destroy_objid_cache(hmp);
753         kmalloc_destroy(&hmp->m_misc);
754         kmalloc_destroy(&hmp->m_inodes);
755         kfree(hmp, M_HAMMER);
756 }
757
758 /*
759  * Report critical errors.  ip may be NULL.
760  */
761 void
762 hammer_critical_error(hammer_mount_t hmp, hammer_inode_t ip,
763                       int error, const char *msg)
764 {
765         hmp->flags |= HAMMER_MOUNT_CRITICAL_ERROR;
766         krateprintf(&hmp->krate,
767                 "HAMMER(%s): Critical error inode=%lld %s\n",
768                 hmp->mp->mnt_stat.f_mntfromname,
769                 (long long)(ip ? ip->obj_id : -1), msg);
770         if (hmp->ronly == 0) {
771                 hmp->ronly = 2;         /* special errored read-only mode */
772                 hmp->mp->mnt_flag |= MNT_RDONLY;
773                 kprintf("HAMMER(%s): Forcing read-only mode\n",
774                         hmp->mp->mnt_stat.f_mntfromname);
775         }
776         hmp->error = error;
777 }
778
779
780 /*
781  * Obtain a vnode for the specified inode number.  An exclusively locked
782  * vnode is returned.
783  */
784 int
785 hammer_vfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp)
786 {
787         struct hammer_transaction trans;
788         struct hammer_mount *hmp = (void *)mp->mnt_data;
789         struct hammer_inode *ip;
790         int error;
791
792         hammer_simple_transaction(&trans, hmp);
793
794         /*
795          * Lookup the requested HAMMER inode.  The structure must be
796          * left unlocked while we manipulate the related vnode to avoid
797          * a deadlock.
798          */
799         ip = hammer_get_inode(&trans, NULL, ino,
800                               hmp->asof, HAMMER_DEF_LOCALIZATION, 
801                               0, &error);
802         if (ip == NULL) {
803                 *vpp = NULL;
804                 hammer_done_transaction(&trans);
805                 return(error);
806         }
807         error = hammer_get_vnode(ip, vpp);
808         hammer_rel_inode(ip, 0);
809         hammer_done_transaction(&trans);
810         return (error);
811 }
812
813 /*
814  * Return the root vnode for the filesystem.
815  *
816  * HAMMER stores the root vnode in the hammer_mount structure so
817  * getting it is easy.
818  */
819 static int
820 hammer_vfs_root(struct mount *mp, struct vnode **vpp)
821 {
822 #if 0
823         struct hammer_mount *hmp = (void *)mp->mnt_data;
824 #endif
825         int error;
826
827         error = hammer_vfs_vget(mp, 1, vpp);
828         return (error);
829 }
830
831 static int
832 hammer_vfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
833 {
834         struct hammer_mount *hmp = (void *)mp->mnt_data;
835         hammer_volume_t volume;
836         hammer_volume_ondisk_t ondisk;
837         int error;
838         int64_t bfree;
839
840         volume = hammer_get_root_volume(hmp, &error);
841         if (error)
842                 return(error);
843         ondisk = volume->ondisk;
844
845         /*
846          * Basic stats
847          */
848         mp->mnt_stat.f_files = ondisk->vol0_stat_inodes;
849         bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE;
850         hammer_rel_volume(volume, 0);
851
852         mp->mnt_stat.f_bfree = bfree / HAMMER_BUFSIZE;
853         mp->mnt_stat.f_bavail = mp->mnt_stat.f_bfree;
854         if (mp->mnt_stat.f_files < 0)
855                 mp->mnt_stat.f_files = 0;
856
857         *sbp = mp->mnt_stat;
858         return(0);
859 }
860
861 static int
862 hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred)
863 {
864         struct hammer_mount *hmp = (void *)mp->mnt_data;
865         hammer_volume_t volume;
866         hammer_volume_ondisk_t ondisk;
867         int error;
868         int64_t bfree;
869         int64_t breserved;
870
871         volume = hammer_get_root_volume(hmp, &error);
872         if (error)
873                 return(error);
874         ondisk = volume->ondisk;
875
876         /*
877          * Basic stats
878          */
879         _hammer_checkspace(hmp, HAMMER_CHKSPC_WRITE, &breserved);
880         mp->mnt_vstat.f_files = ondisk->vol0_stat_inodes;
881         bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE;
882         hammer_rel_volume(volume, 0);
883
884         mp->mnt_vstat.f_bfree = (bfree - breserved) / HAMMER_BUFSIZE;
885         mp->mnt_vstat.f_bavail = mp->mnt_stat.f_bfree;
886         if (mp->mnt_vstat.f_files < 0)
887                 mp->mnt_vstat.f_files = 0;
888         *sbp = mp->mnt_vstat;
889         return(0);
890 }
891
892 /*
893  * Sync the filesystem.  Currently we have to run it twice, the second
894  * one will advance the undo start index to the end index, so if a crash
895  * occurs no undos will be run on mount.
896  *
897  * We do not sync the filesystem if we are called from a panic.  If we did
898  * we might end up blowing up a sync that was already in progress.
899  */
900 static int
901 hammer_vfs_sync(struct mount *mp, int waitfor)
902 {
903         struct hammer_mount *hmp = (void *)mp->mnt_data;
904         int error;
905
906         if (panicstr == NULL) {
907                 error = hammer_sync_hmp(hmp, waitfor);
908         } else {
909                 error = EIO;
910         }
911         return (error);
912 }
913
914 /*
915  * Convert a vnode to a file handle.
916  */
917 static int
918 hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp)
919 {
920         hammer_inode_t ip;
921
922         KKASSERT(MAXFIDSZ >= 16);
923         ip = VTOI(vp);
924         fhp->fid_len = offsetof(struct fid, fid_data[16]);
925         fhp->fid_ext = ip->obj_localization >> 16;
926         bcopy(&ip->obj_id, fhp->fid_data + 0, sizeof(ip->obj_id));
927         bcopy(&ip->obj_asof, fhp->fid_data + 8, sizeof(ip->obj_asof));
928         return(0);
929 }
930
931
932 /*
933  * Convert a file handle back to a vnode.
934  *
935  * Use rootvp to enforce PFS isolation when a PFS is exported via a
936  * null mount.
937  */
938 static int
939 hammer_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
940                   struct fid *fhp, struct vnode **vpp)
941 {
942         struct hammer_transaction trans;
943         struct hammer_inode *ip;
944         struct hammer_inode_info info;
945         int error;
946         u_int32_t localization;
947
948         bcopy(fhp->fid_data + 0, &info.obj_id, sizeof(info.obj_id));
949         bcopy(fhp->fid_data + 8, &info.obj_asof, sizeof(info.obj_asof));
950         if (rootvp)
951                 localization = VTOI(rootvp)->obj_localization;
952         else
953                 localization = (u_int32_t)fhp->fid_ext << 16;
954
955         hammer_simple_transaction(&trans, (void *)mp->mnt_data);
956
957         /*
958          * Get/allocate the hammer_inode structure.  The structure must be
959          * unlocked while we manipulate the related vnode to avoid a
960          * deadlock.
961          */
962         ip = hammer_get_inode(&trans, NULL, info.obj_id,
963                               info.obj_asof, localization, 0, &error);
964         if (ip == NULL) {
965                 *vpp = NULL;
966                 return(error);
967         }
968         error = hammer_get_vnode(ip, vpp);
969         hammer_rel_inode(ip, 0);
970         hammer_done_transaction(&trans);
971         return (error);
972 }
973
974 static int
975 hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
976                     int *exflagsp, struct ucred **credanonp)
977 {
978         hammer_mount_t hmp = (void *)mp->mnt_data;
979         struct netcred *np;
980         int error;
981
982         np = vfs_export_lookup(mp, &hmp->export, nam);
983         if (np) {
984                 *exflagsp = np->netc_exflags;
985                 *credanonp = &np->netc_anon;
986                 error = 0;
987         } else {
988                 error = EACCES;
989         }
990         return (error);
991
992 }
993
994 int
995 hammer_vfs_export(struct mount *mp, int op, const struct export_args *export)
996 {
997         hammer_mount_t hmp = (void *)mp->mnt_data;
998         int error;
999
1000         switch(op) {
1001         case MOUNTCTL_SET_EXPORT:
1002                 error = vfs_export(mp, &hmp->export, export);
1003                 break;
1004         default:
1005                 error = EOPNOTSUPP;
1006                 break;
1007         }
1008         return(error);
1009 }
1010