kernel - fine-grained namecache and partial vnode MPSAFE work
[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 /*
51  * NOTE!  Global statistics may not be MPSAFE so HAMMER never uses them
52  *        in conditionals.
53  */
54 int hammer_supported_version = HAMMER_VOL_VERSION_DEFAULT;
55 int hammer_debug_io;
56 int hammer_debug_general;
57 int hammer_debug_debug = 1;             /* medium-error panics */ 
58 int hammer_debug_inode;
59 int hammer_debug_locks;
60 int hammer_debug_btree;
61 int hammer_debug_tid;
62 int hammer_debug_recover;               /* -1 will disable, +1 will force */
63 int hammer_debug_recover_faults;
64 int hammer_debug_critical;              /* non-zero enter debugger on error */
65 int hammer_cluster_enable = 1;          /* enable read clustering by default */
66 int hammer_count_fsyncs;
67 int hammer_count_inodes;
68 int hammer_count_iqueued;
69 int hammer_count_reclaiming;
70 int hammer_count_records;
71 int hammer_count_record_datas;
72 int hammer_count_volumes;
73 int hammer_count_buffers;
74 int hammer_count_nodes;
75 int64_t hammer_count_extra_space_used;
76 int64_t hammer_stats_btree_lookups;
77 int64_t hammer_stats_btree_searches;
78 int64_t hammer_stats_btree_inserts;
79 int64_t hammer_stats_btree_deletes;
80 int64_t hammer_stats_btree_elements;
81 int64_t hammer_stats_btree_splits;
82 int64_t hammer_stats_btree_iterations;
83 int64_t hammer_stats_btree_root_iterations;
84 int64_t hammer_stats_record_iterations;
85
86 int64_t hammer_stats_file_read;
87 int64_t hammer_stats_file_write;
88 int64_t hammer_stats_file_iopsr;
89 int64_t hammer_stats_file_iopsw;
90 int64_t hammer_stats_disk_read;
91 int64_t hammer_stats_disk_write;
92 int64_t hammer_stats_inode_flushes;
93 int64_t hammer_stats_commits;
94 int64_t hammer_stats_undo;
95
96 int hammer_count_dirtybufspace;         /* global */
97 int hammer_count_refedbufs;             /* global */
98 int hammer_count_reservations;
99 int hammer_count_io_running_read;
100 int hammer_count_io_running_write;
101 int hammer_count_io_locked;
102 int hammer_limit_dirtybufspace;         /* per-mount */
103 int hammer_limit_recs;                  /* as a whole XXX */
104 int hammer_limit_inode_recs = 1024;     /* per inode */
105 int hammer_limit_reclaim = HAMMER_RECLAIM_WAIT;
106 int hammer_autoflush = 2000;            /* auto flush */
107 int hammer_bio_count;
108 int hammer_verify_zone;
109 int hammer_verify_data = 1;
110 int hammer_write_mode;
111 int hammer_yield_check = 16;
112 int hammer_fsync_mode;
113 int64_t hammer_contention_count;
114 int64_t hammer_zone_limit;
115
116 SYSCTL_NODE(_vfs, OID_AUTO, hammer, CTLFLAG_RW, 0, "HAMMER filesystem");
117 SYSCTL_INT(_vfs_hammer, OID_AUTO, supported_version, CTLFLAG_RD,
118            &hammer_supported_version, 0, "");
119 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_general, CTLFLAG_RW,
120            &hammer_debug_general, 0, "");
121 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_io, CTLFLAG_RW,
122            &hammer_debug_io, 0, "");
123 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_debug, CTLFLAG_RW,
124            &hammer_debug_debug, 0, "");
125 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_inode, CTLFLAG_RW,
126            &hammer_debug_inode, 0, "");
127 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_locks, CTLFLAG_RW,
128            &hammer_debug_locks, 0, "");
129 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_btree, CTLFLAG_RW,
130            &hammer_debug_btree, 0, "");
131 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_tid, CTLFLAG_RW,
132            &hammer_debug_tid, 0, "");
133 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover, CTLFLAG_RW,
134            &hammer_debug_recover, 0, "");
135 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover_faults, CTLFLAG_RW,
136            &hammer_debug_recover_faults, 0, "");
137 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_critical, CTLFLAG_RW,
138            &hammer_debug_critical, 0, "");
139 SYSCTL_INT(_vfs_hammer, OID_AUTO, cluster_enable, CTLFLAG_RW,
140            &hammer_cluster_enable, 0, "");
141
142 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_dirtybufspace, CTLFLAG_RW,
143            &hammer_limit_dirtybufspace, 0, "");
144 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_recs, CTLFLAG_RW,
145            &hammer_limit_recs, 0, "");
146 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_inode_recs, CTLFLAG_RW,
147            &hammer_limit_inode_recs, 0, "");
148 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_reclaim, CTLFLAG_RW,
149            &hammer_limit_reclaim, 0, "");
150
151 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_fsyncs, CTLFLAG_RD,
152            &hammer_count_fsyncs, 0, "");
153 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_inodes, CTLFLAG_RD,
154            &hammer_count_inodes, 0, "");
155 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_iqueued, CTLFLAG_RD,
156            &hammer_count_iqueued, 0, "");
157 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reclaiming, CTLFLAG_RD,
158            &hammer_count_reclaiming, 0, "");
159 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_records, CTLFLAG_RD,
160            &hammer_count_records, 0, "");
161 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_record_datas, CTLFLAG_RD,
162            &hammer_count_record_datas, 0, "");
163 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_volumes, CTLFLAG_RD,
164            &hammer_count_volumes, 0, "");
165 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_buffers, CTLFLAG_RD,
166            &hammer_count_buffers, 0, "");
167 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_nodes, CTLFLAG_RD,
168            &hammer_count_nodes, 0, "");
169 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, count_extra_space_used, CTLFLAG_RD,
170            &hammer_count_extra_space_used, 0, "");
171
172 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_searches, CTLFLAG_RD,
173            &hammer_stats_btree_searches, 0, "");
174 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_lookups, CTLFLAG_RD,
175            &hammer_stats_btree_lookups, 0, "");
176 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_inserts, CTLFLAG_RD,
177            &hammer_stats_btree_inserts, 0, "");
178 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_deletes, CTLFLAG_RD,
179            &hammer_stats_btree_deletes, 0, "");
180 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_elements, CTLFLAG_RD,
181            &hammer_stats_btree_elements, 0, "");
182 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_splits, CTLFLAG_RD,
183            &hammer_stats_btree_splits, 0, "");
184 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_iterations, CTLFLAG_RD,
185            &hammer_stats_btree_iterations, 0, "");
186 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_root_iterations, CTLFLAG_RD,
187            &hammer_stats_btree_root_iterations, 0, "");
188 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_record_iterations, CTLFLAG_RD,
189            &hammer_stats_record_iterations, 0, "");
190
191 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_read, CTLFLAG_RD,
192            &hammer_stats_file_read, 0, "");
193 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_write, CTLFLAG_RD,
194            &hammer_stats_file_write, 0, "");
195 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_iopsr, CTLFLAG_RD,
196            &hammer_stats_file_iopsr, 0, "");
197 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_iopsw, CTLFLAG_RD,
198            &hammer_stats_file_iopsw, 0, "");
199 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_disk_read, CTLFLAG_RD,
200            &hammer_stats_disk_read, 0, "");
201 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_disk_write, CTLFLAG_RD,
202            &hammer_stats_disk_write, 0, "");
203 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_inode_flushes, CTLFLAG_RD,
204            &hammer_stats_inode_flushes, 0, "");
205 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_commits, CTLFLAG_RD,
206            &hammer_stats_commits, 0, "");
207 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_undo, CTLFLAG_RD,
208            &hammer_stats_undo, 0, "");
209
210 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_dirtybufspace, CTLFLAG_RD,
211            &hammer_count_dirtybufspace, 0, "");
212 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_refedbufs, CTLFLAG_RD,
213            &hammer_count_refedbufs, 0, "");
214 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reservations, CTLFLAG_RD,
215            &hammer_count_reservations, 0, "");
216 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_running_read, CTLFLAG_RD,
217            &hammer_count_io_running_read, 0, "");
218 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_locked, CTLFLAG_RD,
219            &hammer_count_io_locked, 0, "");
220 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_running_write, CTLFLAG_RD,
221            &hammer_count_io_running_write, 0, "");
222 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, zone_limit, CTLFLAG_RW,
223            &hammer_zone_limit, 0, "");
224 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, contention_count, CTLFLAG_RW,
225            &hammer_contention_count, 0, "");
226 SYSCTL_INT(_vfs_hammer, OID_AUTO, autoflush, CTLFLAG_RW,
227            &hammer_autoflush, 0, "");
228 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_zone, CTLFLAG_RW,
229            &hammer_verify_zone, 0, "");
230 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_data, CTLFLAG_RW,
231            &hammer_verify_data, 0, "");
232 SYSCTL_INT(_vfs_hammer, OID_AUTO, write_mode, CTLFLAG_RW,
233            &hammer_write_mode, 0, "");
234 SYSCTL_INT(_vfs_hammer, OID_AUTO, yield_check, CTLFLAG_RW,
235            &hammer_yield_check, 0, "");
236 SYSCTL_INT(_vfs_hammer, OID_AUTO, fsync_mode, CTLFLAG_RW,
237            &hammer_fsync_mode, 0, "");
238
239 KTR_INFO_MASTER(hammer);
240
241 /*
242  * VFS ABI
243  */
244 static void     hammer_free_hmp(struct mount *mp);
245
246 static int      hammer_vfs_mount(struct mount *mp, char *path, caddr_t data,
247                                 struct ucred *cred);
248 static int      hammer_vfs_unmount(struct mount *mp, int mntflags);
249 static int      hammer_vfs_root(struct mount *mp, struct vnode **vpp);
250 static int      hammer_vfs_statfs(struct mount *mp, struct statfs *sbp,
251                                 struct ucred *cred);
252 static int      hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp,
253                                 struct ucred *cred);
254 static int      hammer_vfs_sync(struct mount *mp, int waitfor);
255 static int      hammer_vfs_vget(struct mount *mp, struct vnode *dvp,
256                                 ino_t ino, struct vnode **vpp);
257 static int      hammer_vfs_init(struct vfsconf *conf);
258 static int      hammer_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
259                                 struct fid *fhp, struct vnode **vpp);
260 static int      hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp);
261 static int      hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
262                                 int *exflagsp, struct ucred **credanonp);
263
264
265 static struct vfsops hammer_vfsops = {
266         .vfs_mount      = hammer_vfs_mount,
267         .vfs_unmount    = hammer_vfs_unmount,
268         .vfs_root       = hammer_vfs_root,
269         .vfs_statfs     = hammer_vfs_statfs,
270         .vfs_statvfs    = hammer_vfs_statvfs,
271         .vfs_sync       = hammer_vfs_sync,
272         .vfs_vget       = hammer_vfs_vget,
273         .vfs_init       = hammer_vfs_init,
274         .vfs_vptofh     = hammer_vfs_vptofh,
275         .vfs_fhtovp     = hammer_vfs_fhtovp,
276         .vfs_checkexp   = hammer_vfs_checkexp
277 };
278
279 MALLOC_DEFINE(M_HAMMER, "HAMMER-mount", "");
280
281 VFS_SET(hammer_vfsops, hammer, 0);
282 MODULE_VERSION(hammer, 1);
283
284 static int
285 hammer_vfs_init(struct vfsconf *conf)
286 {
287         int n;
288
289         if (hammer_limit_recs == 0) {
290                 hammer_limit_recs = nbuf * 25;
291                 n = kmalloc_limit(M_HAMMER) / 512;
292                 if (hammer_limit_recs > n)
293                         hammer_limit_recs = n;
294         }
295         if (hammer_limit_dirtybufspace == 0) {
296                 hammer_limit_dirtybufspace = hidirtybufspace / 2;
297                 if (hammer_limit_dirtybufspace < 100)
298                         hammer_limit_dirtybufspace = 100;
299         }
300         return(0);
301 }
302
303 static int
304 hammer_vfs_mount(struct mount *mp, char *mntpt, caddr_t data,
305                  struct ucred *cred)
306 {
307         struct hammer_mount_info info;
308         hammer_mount_t hmp;
309         hammer_volume_t rootvol;
310         struct vnode *rootvp;
311         struct vnode *devvp = NULL;
312         const char *upath;      /* volume name in userspace */
313         char *path;             /* volume name in system space */
314         int error;
315         int i;
316         int master_id;
317         int maxinodes;
318         char *next_volume_ptr = NULL;
319
320         /*
321          * Accept hammer_mount_info.  mntpt is NULL for root mounts at boot.
322          */
323         if (mntpt == NULL) {
324                 bzero(&info, sizeof(info));
325                 info.asof = 0;
326                 info.hflags = 0;
327                 info.nvolumes = 1;
328
329                 next_volume_ptr = mp->mnt_stat.f_mntfromname;
330
331                 /* Count number of volumes separated by ':' */
332                 for (char *p = next_volume_ptr; *p != '\0'; ++p) {
333                         if (*p == ':') {
334                                 ++info.nvolumes;
335                         }
336                 }
337
338                 mp->mnt_flag &= ~MNT_RDONLY; /* mount R/W */
339         } else {
340                 if ((error = copyin(data, &info, sizeof(info))) != 0)
341                         return (error);
342         }
343
344         /*
345          * updating or new mount
346          */
347         if (mp->mnt_flag & MNT_UPDATE) {
348                 hmp = (void *)mp->mnt_data;
349                 KKASSERT(hmp != NULL);
350         } else {
351                 if (info.nvolumes <= 0 || info.nvolumes >= 32768)
352                         return (EINVAL);
353                 hmp = NULL;
354         }
355
356         /*
357          * master-id validation.  The master id may not be changed by a
358          * mount update.
359          */
360         if (info.hflags & HMNT_MASTERID) {
361                 if (hmp && hmp->master_id != info.master_id) {
362                         kprintf("hammer: cannot change master id "
363                                 "with mount update\n");
364                         return(EINVAL);
365                 }
366                 master_id = info.master_id;
367                 if (master_id < -1 || master_id >= HAMMER_MAX_MASTERS)
368                         return (EINVAL);
369         } else {
370                 if (hmp)
371                         master_id = hmp->master_id;
372                 else
373                         master_id = 0;
374         }
375
376         /*
377          * Interal mount data structure
378          */
379         if (hmp == NULL) {
380                 hmp = kmalloc(sizeof(*hmp), M_HAMMER, M_WAITOK | M_ZERO);
381                 mp->mnt_data = (qaddr_t)hmp;
382                 hmp->mp = mp;
383                 /*TAILQ_INIT(&hmp->recycle_list);*/
384
385                 /*
386                  * Make sure kmalloc type limits are set appropriately.  If root
387                  * increases the vnode limit you may have to do a dummy remount
388                  * to adjust the HAMMER inode limit.
389                  */
390                 kmalloc_create(&hmp->m_misc, "HAMMER-others");
391                 kmalloc_create(&hmp->m_inodes, "HAMMER-inodes");
392
393                 maxinodes = desiredvnodes + desiredvnodes / 5 +
394                             hammer_limit_reclaim * 2;
395                 kmalloc_raise_limit(hmp->m_inodes,
396                                     maxinodes * sizeof(struct hammer_inode));
397
398                 hmp->root_btree_beg.localization = 0x00000000U;
399                 hmp->root_btree_beg.obj_id = -0x8000000000000000LL;
400                 hmp->root_btree_beg.key = -0x8000000000000000LL;
401                 hmp->root_btree_beg.create_tid = 1;
402                 hmp->root_btree_beg.delete_tid = 1;
403                 hmp->root_btree_beg.rec_type = 0;
404                 hmp->root_btree_beg.obj_type = 0;
405
406                 hmp->root_btree_end.localization = 0xFFFFFFFFU;
407                 hmp->root_btree_end.obj_id = 0x7FFFFFFFFFFFFFFFLL;
408                 hmp->root_btree_end.key = 0x7FFFFFFFFFFFFFFFLL;
409                 hmp->root_btree_end.create_tid = 0xFFFFFFFFFFFFFFFFULL;
410                 hmp->root_btree_end.delete_tid = 0;   /* special case */
411                 hmp->root_btree_end.rec_type = 0xFFFFU;
412                 hmp->root_btree_end.obj_type = 0;
413
414                 hmp->krate.freq = 1;    /* maximum reporting rate (hz) */
415                 hmp->krate.count = -16; /* initial burst */
416
417                 hmp->sync_lock.refs = 1;
418                 hmp->free_lock.refs = 1;
419                 hmp->undo_lock.refs = 1;
420                 hmp->blkmap_lock.refs = 1;
421                 hmp->snapshot_lock.refs = 1;
422                 hmp->volume_lock.refs = 1;
423
424                 TAILQ_INIT(&hmp->delay_list);
425                 TAILQ_INIT(&hmp->flush_group_list);
426                 TAILQ_INIT(&hmp->objid_cache_list);
427                 TAILQ_INIT(&hmp->undo_lru_list);
428                 TAILQ_INIT(&hmp->reclaim_list);
429         }
430         hmp->hflags &= ~HMNT_USERFLAGS;
431         hmp->hflags |= info.hflags & HMNT_USERFLAGS;
432
433         hmp->master_id = master_id;
434
435         if (info.asof) {
436                 mp->mnt_flag |= MNT_RDONLY;
437                 hmp->asof = info.asof;
438         } else {
439                 hmp->asof = HAMMER_MAX_TID;
440         }
441
442         hmp->volume_to_remove = -1;
443
444         /*
445          * Re-open read-write if originally read-only, or vise-versa.
446          *
447          * When going from read-only to read-write execute the stage2
448          * recovery if it has not already been run.
449          */
450         if (mp->mnt_flag & MNT_UPDATE) {
451                 error = 0;
452                 if (hmp->ronly && (mp->mnt_kern_flag & MNTK_WANTRDWR)) {
453                         kprintf("HAMMER read-only -> read-write\n");
454                         hmp->ronly = 0;
455                         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
456                                 hammer_adjust_volume_mode, NULL);
457                         rootvol = hammer_get_root_volume(hmp, &error);
458                         if (rootvol) {
459                                 hammer_recover_flush_buffers(hmp, rootvol, 1);
460                                 error = hammer_recover_stage2(hmp, rootvol);
461                                 bcopy(rootvol->ondisk->vol0_blockmap,
462                                       hmp->blockmap,
463                                       sizeof(hmp->blockmap));
464                                 hammer_rel_volume(rootvol, 0);
465                         }
466                         RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
467                                 hammer_reload_inode, NULL);
468                         /* kernel clears MNT_RDONLY */
469                 } else if (hmp->ronly == 0 && (mp->mnt_flag & MNT_RDONLY)) {
470                         kprintf("HAMMER read-write -> read-only\n");
471                         hmp->ronly = 1; /* messy */
472                         RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
473                                 hammer_reload_inode, NULL);
474                         hmp->ronly = 0;
475                         hammer_flusher_sync(hmp);
476                         hammer_flusher_sync(hmp);
477                         hammer_flusher_sync(hmp);
478                         hmp->ronly = 1;
479                         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
480                                 hammer_adjust_volume_mode, NULL);
481                 }
482                 return(error);
483         }
484
485         RB_INIT(&hmp->rb_vols_root);
486         RB_INIT(&hmp->rb_inos_root);
487         RB_INIT(&hmp->rb_nods_root);
488         RB_INIT(&hmp->rb_undo_root);
489         RB_INIT(&hmp->rb_resv_root);
490         RB_INIT(&hmp->rb_bufs_root);
491         RB_INIT(&hmp->rb_pfsm_root);
492
493         hmp->ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
494
495         TAILQ_INIT(&hmp->volu_list);
496         TAILQ_INIT(&hmp->undo_list);
497         TAILQ_INIT(&hmp->data_list);
498         TAILQ_INIT(&hmp->meta_list);
499         TAILQ_INIT(&hmp->lose_list);
500
501         /*
502          * Load volumes
503          */
504         path = objcache_get(namei_oc, M_WAITOK);
505         hmp->nvolumes = -1;
506         for (i = 0; i < info.nvolumes; ++i) {
507                 if (mntpt == NULL) {
508                         /*
509                          * Root mount.
510                          */
511                         KKASSERT(next_volume_ptr != NULL);
512                         strcpy(path, "");
513                         if (*next_volume_ptr != '/') {
514                                 /* relative path */
515                                 strcpy(path, "/dev/");
516                         }
517                         int k;
518                         for (k = strlen(path); k < MAXPATHLEN-1; ++k) {
519                                 if (*next_volume_ptr == '\0') {
520                                         break;
521                                 } else if (*next_volume_ptr == ':') {
522                                         ++next_volume_ptr;
523                                         break;
524                                 } else {
525                                         path[k] = *next_volume_ptr;
526                                         ++next_volume_ptr;
527                                 }
528                         }
529                         path[k] = '\0';
530
531                         error = 0;
532                         cdev_t dev = kgetdiskbyname(path);
533                         error = bdevvp(dev, &devvp);
534                         if (error) {
535                                 kprintf("hammer_mountroot: can't find devvp\n");
536                         }
537                 } else {
538                         error = copyin(&info.volumes[i], &upath,
539                                        sizeof(char *));
540                         if (error == 0)
541                                 error = copyinstr(upath, path,
542                                                   MAXPATHLEN, NULL);
543                 }
544                 if (error == 0)
545                         error = hammer_install_volume(hmp, path, devvp);
546                 if (error)
547                         break;
548         }
549         objcache_put(namei_oc, path);
550
551         /*
552          * Make sure we found a root volume
553          */
554         if (error == 0 && hmp->rootvol == NULL) {
555                 kprintf("hammer_mount: No root volume found!\n");
556                 error = EINVAL;
557         }
558
559         /*
560          * Check that all required volumes are available
561          */
562         if (error == 0 && hammer_mountcheck_volumes(hmp)) {
563                 kprintf("hammer_mount: Missing volumes, cannot mount!\n");
564                 error = EINVAL;
565         }
566
567         if (error) {
568                 hammer_free_hmp(mp);
569                 return (error);
570         }
571
572         /*
573          * No errors, setup enough of the mount point so we can lookup the
574          * root vnode.
575          */
576         mp->mnt_iosize_max = MAXPHYS;
577         mp->mnt_kern_flag |= MNTK_FSMID;
578
579         /*
580          * MPSAFE code.  Note that VOPs and VFSops which are not MPSAFE
581          * will acquire a per-mount token prior to entry and release it
582          * on return, so even if we do not specify it we no longer get
583          * the BGL regardlless of how we are flagged.
584          */
585         mp->mnt_kern_flag |= MNTK_RD_MPSAFE | MNTK_GA_MPSAFE |
586                              MNTK_IN_MPSAFE;
587
588         /* 
589          * note: f_iosize is used by vnode_pager_haspage() when constructing
590          * its VOP_BMAP call.
591          */
592         mp->mnt_stat.f_iosize = HAMMER_BUFSIZE;
593         mp->mnt_stat.f_bsize = HAMMER_BUFSIZE;
594
595         mp->mnt_vstat.f_frsize = HAMMER_BUFSIZE;
596         mp->mnt_vstat.f_bsize = HAMMER_BUFSIZE;
597
598         mp->mnt_maxsymlinklen = 255;
599         mp->mnt_flag |= MNT_LOCAL;
600
601         vfs_add_vnodeops(mp, &hammer_vnode_vops, &mp->mnt_vn_norm_ops);
602         vfs_add_vnodeops(mp, &hammer_spec_vops, &mp->mnt_vn_spec_ops);
603         vfs_add_vnodeops(mp, &hammer_fifo_vops, &mp->mnt_vn_fifo_ops);
604
605         /*
606          * The root volume's ondisk pointer is only valid if we hold a
607          * reference to it.
608          */
609         rootvol = hammer_get_root_volume(hmp, &error);
610         if (error)
611                 goto failed;
612
613         /*
614          * Perform any necessary UNDO operations.  The recovery code does
615          * call hammer_undo_lookup() so we have to pre-cache the blockmap,
616          * and then re-copy it again after recovery is complete.
617          *
618          * If this is a read-only mount the UNDO information is retained
619          * in memory in the form of dirty buffer cache buffers, and not
620          * written back to the media.
621          */
622         bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
623               sizeof(hmp->blockmap));
624
625         /*
626          * Check filesystem version
627          */
628         hmp->version = rootvol->ondisk->vol_version;
629         if (hmp->version < HAMMER_VOL_VERSION_MIN ||
630             hmp->version > HAMMER_VOL_VERSION_MAX) {
631                 kprintf("HAMMER: mount unsupported fs version %d\n",
632                         hmp->version);
633                 error = ERANGE;
634                 goto done;
635         }
636
637         /*
638          * The undo_rec_limit limits the size of flush groups to avoid
639          * blowing out the UNDO FIFO.  This calculation is typically in
640          * the tens of thousands and is designed primarily when small
641          * HAMMER filesystems are created.
642          */
643         hmp->undo_rec_limit = hammer_undo_max(hmp) / 8192 + 100;
644         if (hammer_debug_general & 0x0001)
645                 kprintf("HAMMER: undo_rec_limit %d\n", hmp->undo_rec_limit);
646
647         /*
648          * NOTE: Recover stage1 not only handles meta-data recovery, it
649          *       also sets hmp->undo_seqno for HAMMER VERSION 4+ filesystems.
650          */
651         error = hammer_recover_stage1(hmp, rootvol);
652         if (error) {
653                 kprintf("Failed to recover HAMMER filesystem on mount\n");
654                 goto done;
655         }
656
657         /*
658          * Finish setup now that we have a good root volume.
659          *
660          * The top 16 bits of fsid.val[1] is a pfs id.
661          */
662         ksnprintf(mp->mnt_stat.f_mntfromname,
663                   sizeof(mp->mnt_stat.f_mntfromname), "%s",
664                   rootvol->ondisk->vol_name);
665         mp->mnt_stat.f_fsid.val[0] =
666                 crc32((char *)&rootvol->ondisk->vol_fsid + 0, 8);
667         mp->mnt_stat.f_fsid.val[1] =
668                 crc32((char *)&rootvol->ondisk->vol_fsid + 8, 8);
669         mp->mnt_stat.f_fsid.val[1] &= 0x0000FFFF;
670
671         mp->mnt_vstat.f_fsid_uuid = rootvol->ondisk->vol_fsid;
672         mp->mnt_vstat.f_fsid = crc32(&mp->mnt_vstat.f_fsid_uuid,
673                                      sizeof(mp->mnt_vstat.f_fsid_uuid));
674
675         /*
676          * Certain often-modified fields in the root volume are cached in
677          * the hammer_mount structure so we do not have to generate lots
678          * of little UNDO structures for them.
679          *
680          * Recopy after recovery.  This also has the side effect of
681          * setting our cached undo FIFO's first_offset, which serves to
682          * placemark the FIFO start for the NEXT flush cycle while the
683          * on-disk first_offset represents the LAST flush cycle.
684          */
685         hmp->next_tid = rootvol->ondisk->vol0_next_tid;
686         hmp->flush_tid1 = hmp->next_tid;
687         hmp->flush_tid2 = hmp->next_tid;
688         bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
689               sizeof(hmp->blockmap));
690         hmp->copy_stat_freebigblocks = rootvol->ondisk->vol0_stat_freebigblocks;
691
692         hammer_flusher_create(hmp);
693
694         /*
695          * Locate the root directory using the root cluster's B-Tree as a
696          * starting point.  The root directory uses an obj_id of 1.
697          *
698          * FUTURE: Leave the root directory cached referenced but unlocked
699          * in hmp->rootvp (need to flush it on unmount).
700          */
701         error = hammer_vfs_vget(mp, NULL, 1, &rootvp);
702         if (error)
703                 goto done;
704         vput(rootvp);
705         /*vn_unlock(hmp->rootvp);*/
706         if (hmp->ronly == 0)
707                 error = hammer_recover_stage2(hmp, rootvol);
708
709 done:
710         hammer_rel_volume(rootvol, 0);
711 failed:
712         /*
713          * Cleanup and return.
714          */
715         if (error)
716                 hammer_free_hmp(mp);
717         return (error);
718 }
719
720 static int
721 hammer_vfs_unmount(struct mount *mp, int mntflags)
722 {
723 #if 0
724         struct hammer_mount *hmp = (void *)mp->mnt_data;
725 #endif
726         int flags;
727         int error;
728
729         /*
730          * Clean out the vnodes
731          */
732         flags = 0;
733         if (mntflags & MNT_FORCE)
734                 flags |= FORCECLOSE;
735         if ((error = vflush(mp, 0, flags)) != 0)
736                 return (error);
737
738         /*
739          * Clean up the internal mount structure and related entities.  This
740          * may issue I/O.
741          */
742         hammer_free_hmp(mp);
743         return(0);
744 }
745
746 /*
747  * Clean up the internal mount structure and disassociate it from the mount.
748  * This may issue I/O.
749  */
750 static void
751 hammer_free_hmp(struct mount *mp)
752 {
753         struct hammer_mount *hmp = (void *)mp->mnt_data;
754         hammer_flush_group_t flg;
755         int count;
756         int dummy;
757
758         /*
759          * Flush anything dirty.  This won't even run if the
760          * filesystem errored-out.
761          */
762         count = 0;
763         while (hammer_flusher_haswork(hmp)) {
764                 hammer_flusher_sync(hmp);
765                 ++count;
766                 if (count >= 5) {
767                         if (count == 5)
768                                 kprintf("HAMMER: umount flushing.");
769                         else
770                                 kprintf(".");
771                         tsleep(&dummy, 0, "hmrufl", hz);
772                 }
773                 if (count == 30) {
774                         kprintf("giving up\n");
775                         break;
776                 }
777         }
778         if (count >= 5 && count < 30)
779                 kprintf("\n");
780
781         /*
782          * If the mount had a critical error we have to destroy any
783          * remaining inodes before we can finish cleaning up the flusher.
784          */
785         if (hmp->flags & HAMMER_MOUNT_CRITICAL_ERROR) {
786                 RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
787                         hammer_destroy_inode_callback, NULL);
788         }
789
790         /*
791          * There shouldn't be any inodes left now and any left over
792          * flush groups should now be empty.
793          */
794         KKASSERT(RB_EMPTY(&hmp->rb_inos_root));
795         while ((flg = TAILQ_FIRST(&hmp->flush_group_list)) != NULL) {
796                 TAILQ_REMOVE(&hmp->flush_group_list, flg, flush_entry);
797                 KKASSERT(RB_EMPTY(&flg->flush_tree));
798                 if (flg->refs) {
799                         kprintf("HAMMER: Warning, flush_group %p was "
800                                 "not empty on umount!\n", flg);
801                 }
802                 kfree(flg, hmp->m_misc);
803         }
804
805         /*
806          * We can finally destroy the flusher
807          */
808         hammer_flusher_destroy(hmp);
809
810         /*
811          * We may have held recovered buffers due to a read-only mount.
812          * These must be discarded.
813          */
814         if (hmp->ronly)
815                 hammer_recover_flush_buffers(hmp, NULL, -1);
816
817         /*
818          * Unload buffers and then volumes
819          */
820         RB_SCAN(hammer_buf_rb_tree, &hmp->rb_bufs_root, NULL,
821                 hammer_unload_buffer, NULL);
822         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
823                 hammer_unload_volume, NULL);
824
825         mp->mnt_data = NULL;
826         mp->mnt_flag &= ~MNT_LOCAL;
827         hmp->mp = NULL;
828         hammer_destroy_objid_cache(hmp);
829         kmalloc_destroy(&hmp->m_misc);
830         kmalloc_destroy(&hmp->m_inodes);
831         kfree(hmp, M_HAMMER);
832 }
833
834 /*
835  * Report critical errors.  ip may be NULL.
836  */
837 void
838 hammer_critical_error(hammer_mount_t hmp, hammer_inode_t ip,
839                       int error, const char *msg)
840 {
841         hmp->flags |= HAMMER_MOUNT_CRITICAL_ERROR;
842
843         krateprintf(&hmp->krate,
844                     "HAMMER(%s): Critical error inode=%jd error=%d %s\n",
845                     hmp->mp->mnt_stat.f_mntfromname,
846                     (intmax_t)(ip ? ip->obj_id : -1),
847                     error, msg);
848
849         if (hmp->ronly == 0) {
850                 hmp->ronly = 2;         /* special errored read-only mode */
851                 hmp->mp->mnt_flag |= MNT_RDONLY;
852                 kprintf("HAMMER(%s): Forcing read-only mode\n",
853                         hmp->mp->mnt_stat.f_mntfromname);
854         }
855         hmp->error = error;
856         if (hammer_debug_critical)
857                 Debugger("Entering debugger");
858 }
859
860
861 /*
862  * Obtain a vnode for the specified inode number.  An exclusively locked
863  * vnode is returned.
864  */
865 int
866 hammer_vfs_vget(struct mount *mp, struct vnode *dvp,
867                 ino_t ino, struct vnode **vpp)
868 {
869         struct hammer_transaction trans;
870         struct hammer_mount *hmp = (void *)mp->mnt_data;
871         struct hammer_inode *ip;
872         int error;
873         u_int32_t localization;
874
875         hammer_simple_transaction(&trans, hmp);
876
877         /*
878          * If a directory vnode is supplied (mainly NFS) then we can acquire
879          * the PFS domain from it.  Otherwise we would only be able to vget
880          * inodes in the root PFS.
881          */
882         if (dvp) {
883                 localization = HAMMER_DEF_LOCALIZATION +
884                                 VTOI(dvp)->obj_localization;
885         } else {
886                 localization = HAMMER_DEF_LOCALIZATION;
887         }
888
889         /*
890          * Lookup the requested HAMMER inode.  The structure must be
891          * left unlocked while we manipulate the related vnode to avoid
892          * a deadlock.
893          */
894         ip = hammer_get_inode(&trans, NULL, ino,
895                               hmp->asof, localization,
896                               0, &error);
897         if (ip == NULL) {
898                 *vpp = NULL;
899                 hammer_done_transaction(&trans);
900                 return(error);
901         }
902         error = hammer_get_vnode(ip, vpp);
903         hammer_rel_inode(ip, 0);
904         hammer_done_transaction(&trans);
905         return (error);
906 }
907
908 /*
909  * Return the root vnode for the filesystem.
910  *
911  * HAMMER stores the root vnode in the hammer_mount structure so
912  * getting it is easy.
913  */
914 static int
915 hammer_vfs_root(struct mount *mp, struct vnode **vpp)
916 {
917 #if 0
918         struct hammer_mount *hmp = (void *)mp->mnt_data;
919 #endif
920         int error;
921
922         error = hammer_vfs_vget(mp, NULL, 1, vpp);
923         return (error);
924 }
925
926 static int
927 hammer_vfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
928 {
929         struct hammer_mount *hmp = (void *)mp->mnt_data;
930         hammer_volume_t volume;
931         hammer_volume_ondisk_t ondisk;
932         int error;
933         int64_t bfree;
934         int64_t breserved;
935
936         volume = hammer_get_root_volume(hmp, &error);
937         if (error)
938                 return(error);
939         ondisk = volume->ondisk;
940
941         /*
942          * Basic stats
943          */
944         _hammer_checkspace(hmp, HAMMER_CHKSPC_WRITE, &breserved);
945         mp->mnt_stat.f_files = ondisk->vol0_stat_inodes;
946         bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE;
947         hammer_rel_volume(volume, 0);
948
949         mp->mnt_stat.f_bfree = (bfree - breserved) / HAMMER_BUFSIZE;
950         mp->mnt_stat.f_bavail = mp->mnt_stat.f_bfree;
951         if (mp->mnt_stat.f_files < 0)
952                 mp->mnt_stat.f_files = 0;
953
954         *sbp = mp->mnt_stat;
955         return(0);
956 }
957
958 static int
959 hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred)
960 {
961         struct hammer_mount *hmp = (void *)mp->mnt_data;
962         hammer_volume_t volume;
963         hammer_volume_ondisk_t ondisk;
964         int error;
965         int64_t bfree;
966         int64_t breserved;
967
968         volume = hammer_get_root_volume(hmp, &error);
969         if (error)
970                 return(error);
971         ondisk = volume->ondisk;
972
973         /*
974          * Basic stats
975          */
976         _hammer_checkspace(hmp, HAMMER_CHKSPC_WRITE, &breserved);
977         mp->mnt_vstat.f_files = ondisk->vol0_stat_inodes;
978         bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE;
979         hammer_rel_volume(volume, 0);
980
981         mp->mnt_vstat.f_bfree = (bfree - breserved) / HAMMER_BUFSIZE;
982         mp->mnt_vstat.f_bavail = mp->mnt_vstat.f_bfree;
983         if (mp->mnt_vstat.f_files < 0)
984                 mp->mnt_vstat.f_files = 0;
985         *sbp = mp->mnt_vstat;
986         return(0);
987 }
988
989 /*
990  * Sync the filesystem.  Currently we have to run it twice, the second
991  * one will advance the undo start index to the end index, so if a crash
992  * occurs no undos will be run on mount.
993  *
994  * We do not sync the filesystem if we are called from a panic.  If we did
995  * we might end up blowing up a sync that was already in progress.
996  */
997 static int
998 hammer_vfs_sync(struct mount *mp, int waitfor)
999 {
1000         struct hammer_mount *hmp = (void *)mp->mnt_data;
1001         int error;
1002
1003         if (panicstr == NULL) {
1004                 error = hammer_sync_hmp(hmp, waitfor);
1005         } else {
1006                 error = EIO;
1007         }
1008         return (error);
1009 }
1010
1011 /*
1012  * Convert a vnode to a file handle.
1013  */
1014 static int
1015 hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp)
1016 {
1017         hammer_inode_t ip;
1018
1019         KKASSERT(MAXFIDSZ >= 16);
1020         ip = VTOI(vp);
1021         fhp->fid_len = offsetof(struct fid, fid_data[16]);
1022         fhp->fid_ext = ip->obj_localization >> 16;
1023         bcopy(&ip->obj_id, fhp->fid_data + 0, sizeof(ip->obj_id));
1024         bcopy(&ip->obj_asof, fhp->fid_data + 8, sizeof(ip->obj_asof));
1025         return(0);
1026 }
1027
1028
1029 /*
1030  * Convert a file handle back to a vnode.
1031  *
1032  * Use rootvp to enforce PFS isolation when a PFS is exported via a
1033  * null mount.
1034  */
1035 static int
1036 hammer_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
1037                   struct fid *fhp, struct vnode **vpp)
1038 {
1039         struct hammer_transaction trans;
1040         struct hammer_inode *ip;
1041         struct hammer_inode_info info;
1042         int error;
1043         u_int32_t localization;
1044
1045         bcopy(fhp->fid_data + 0, &info.obj_id, sizeof(info.obj_id));
1046         bcopy(fhp->fid_data + 8, &info.obj_asof, sizeof(info.obj_asof));
1047         if (rootvp)
1048                 localization = VTOI(rootvp)->obj_localization;
1049         else
1050                 localization = (u_int32_t)fhp->fid_ext << 16;
1051
1052         hammer_simple_transaction(&trans, (void *)mp->mnt_data);
1053
1054         /*
1055          * Get/allocate the hammer_inode structure.  The structure must be
1056          * unlocked while we manipulate the related vnode to avoid a
1057          * deadlock.
1058          */
1059         ip = hammer_get_inode(&trans, NULL, info.obj_id,
1060                               info.obj_asof, localization, 0, &error);
1061         if (ip == NULL) {
1062                 *vpp = NULL;
1063                 return(error);
1064         }
1065         error = hammer_get_vnode(ip, vpp);
1066         hammer_rel_inode(ip, 0);
1067         hammer_done_transaction(&trans);
1068         return (error);
1069 }
1070
1071 static int
1072 hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
1073                     int *exflagsp, struct ucred **credanonp)
1074 {
1075         hammer_mount_t hmp = (void *)mp->mnt_data;
1076         struct netcred *np;
1077         int error;
1078
1079         np = vfs_export_lookup(mp, &hmp->export, nam);
1080         if (np) {
1081                 *exflagsp = np->netc_exflags;
1082                 *credanonp = &np->netc_anon;
1083                 error = 0;
1084         } else {
1085                 error = EACCES;
1086         }
1087         return (error);
1088
1089 }
1090
1091 int
1092 hammer_vfs_export(struct mount *mp, int op, const struct export_args *export)
1093 {
1094         hammer_mount_t hmp = (void *)mp->mnt_data;
1095         int error;
1096
1097         switch(op) {
1098         case MOUNTCTL_SET_EXPORT:
1099                 error = vfs_export(mp, &hmp->export, export);
1100                 break;
1101         default:
1102                 error = EOPNOTSUPP;
1103                 break;
1104         }
1105         return(error);
1106 }
1107