Merge branch 'master' into selwakeup
[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 int64_t hammer_stats_redo;
96
97 int hammer_count_dirtybufspace;         /* global */
98 int hammer_count_refedbufs;             /* global */
99 int hammer_count_reservations;
100 int hammer_count_io_running_read;
101 int hammer_count_io_running_write;
102 int hammer_count_io_locked;
103 int hammer_limit_dirtybufspace;         /* per-mount */
104 int hammer_limit_recs;                  /* as a whole XXX */
105 int hammer_limit_inode_recs = 1024;     /* per inode */
106 int hammer_limit_reclaim = HAMMER_RECLAIM_WAIT;
107 int hammer_limit_redo = 4096 * 1024;    /* per inode */
108 int hammer_autoflush = 2000;            /* auto flush */
109 int hammer_bio_count;
110 int hammer_verify_zone;
111 int hammer_verify_data = 1;
112 int hammer_write_mode;
113 int hammer_yield_check = 16;
114 int hammer_fsync_mode = 3;
115 int64_t hammer_contention_count;
116 int64_t hammer_zone_limit;
117
118 SYSCTL_NODE(_vfs, OID_AUTO, hammer, CTLFLAG_RW, 0, "HAMMER filesystem");
119 SYSCTL_INT(_vfs_hammer, OID_AUTO, supported_version, CTLFLAG_RD,
120            &hammer_supported_version, 0, "");
121 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_general, CTLFLAG_RW,
122            &hammer_debug_general, 0, "");
123 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_io, CTLFLAG_RW,
124            &hammer_debug_io, 0, "");
125 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_debug, CTLFLAG_RW,
126            &hammer_debug_debug, 0, "");
127 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_inode, CTLFLAG_RW,
128            &hammer_debug_inode, 0, "");
129 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_locks, CTLFLAG_RW,
130            &hammer_debug_locks, 0, "");
131 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_btree, CTLFLAG_RW,
132            &hammer_debug_btree, 0, "");
133 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_tid, CTLFLAG_RW,
134            &hammer_debug_tid, 0, "");
135 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover, CTLFLAG_RW,
136            &hammer_debug_recover, 0, "");
137 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover_faults, CTLFLAG_RW,
138            &hammer_debug_recover_faults, 0, "");
139 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_critical, CTLFLAG_RW,
140            &hammer_debug_critical, 0, "");
141 SYSCTL_INT(_vfs_hammer, OID_AUTO, cluster_enable, CTLFLAG_RW,
142            &hammer_cluster_enable, 0, "");
143
144 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_dirtybufspace, CTLFLAG_RW,
145            &hammer_limit_dirtybufspace, 0, "");
146 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_recs, CTLFLAG_RW,
147            &hammer_limit_recs, 0, "");
148 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_inode_recs, CTLFLAG_RW,
149            &hammer_limit_inode_recs, 0, "");
150 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_reclaim, CTLFLAG_RW,
151            &hammer_limit_reclaim, 0, "");
152 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_redo, CTLFLAG_RW,
153            &hammer_limit_redo, 0, "");
154
155 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_fsyncs, CTLFLAG_RD,
156            &hammer_count_fsyncs, 0, "");
157 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_inodes, CTLFLAG_RD,
158            &hammer_count_inodes, 0, "");
159 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_iqueued, CTLFLAG_RD,
160            &hammer_count_iqueued, 0, "");
161 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reclaiming, CTLFLAG_RD,
162            &hammer_count_reclaiming, 0, "");
163 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_records, CTLFLAG_RD,
164            &hammer_count_records, 0, "");
165 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_record_datas, CTLFLAG_RD,
166            &hammer_count_record_datas, 0, "");
167 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_volumes, CTLFLAG_RD,
168            &hammer_count_volumes, 0, "");
169 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_buffers, CTLFLAG_RD,
170            &hammer_count_buffers, 0, "");
171 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_nodes, CTLFLAG_RD,
172            &hammer_count_nodes, 0, "");
173 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, count_extra_space_used, CTLFLAG_RD,
174            &hammer_count_extra_space_used, 0, "");
175
176 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_searches, CTLFLAG_RD,
177            &hammer_stats_btree_searches, 0, "");
178 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_lookups, CTLFLAG_RD,
179            &hammer_stats_btree_lookups, 0, "");
180 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_inserts, CTLFLAG_RD,
181            &hammer_stats_btree_inserts, 0, "");
182 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_deletes, CTLFLAG_RD,
183            &hammer_stats_btree_deletes, 0, "");
184 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_elements, CTLFLAG_RD,
185            &hammer_stats_btree_elements, 0, "");
186 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_splits, CTLFLAG_RD,
187            &hammer_stats_btree_splits, 0, "");
188 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_iterations, CTLFLAG_RD,
189            &hammer_stats_btree_iterations, 0, "");
190 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_root_iterations, CTLFLAG_RD,
191            &hammer_stats_btree_root_iterations, 0, "");
192 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_record_iterations, CTLFLAG_RD,
193            &hammer_stats_record_iterations, 0, "");
194
195 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_read, CTLFLAG_RD,
196            &hammer_stats_file_read, 0, "");
197 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_write, CTLFLAG_RD,
198            &hammer_stats_file_write, 0, "");
199 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_iopsr, CTLFLAG_RD,
200            &hammer_stats_file_iopsr, 0, "");
201 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_iopsw, CTLFLAG_RD,
202            &hammer_stats_file_iopsw, 0, "");
203 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_disk_read, CTLFLAG_RD,
204            &hammer_stats_disk_read, 0, "");
205 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_disk_write, CTLFLAG_RD,
206            &hammer_stats_disk_write, 0, "");
207 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_inode_flushes, CTLFLAG_RD,
208            &hammer_stats_inode_flushes, 0, "");
209 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_commits, CTLFLAG_RD,
210            &hammer_stats_commits, 0, "");
211 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_undo, CTLFLAG_RD,
212            &hammer_stats_undo, 0, "");
213 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_redo, CTLFLAG_RD,
214            &hammer_stats_redo, 0, "");
215
216 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_dirtybufspace, CTLFLAG_RD,
217            &hammer_count_dirtybufspace, 0, "");
218 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_refedbufs, CTLFLAG_RD,
219            &hammer_count_refedbufs, 0, "");
220 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reservations, CTLFLAG_RD,
221            &hammer_count_reservations, 0, "");
222 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_running_read, CTLFLAG_RD,
223            &hammer_count_io_running_read, 0, "");
224 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_locked, CTLFLAG_RD,
225            &hammer_count_io_locked, 0, "");
226 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_running_write, CTLFLAG_RD,
227            &hammer_count_io_running_write, 0, "");
228 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, zone_limit, CTLFLAG_RW,
229            &hammer_zone_limit, 0, "");
230 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, contention_count, CTLFLAG_RW,
231            &hammer_contention_count, 0, "");
232 SYSCTL_INT(_vfs_hammer, OID_AUTO, autoflush, CTLFLAG_RW,
233            &hammer_autoflush, 0, "");
234 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_zone, CTLFLAG_RW,
235            &hammer_verify_zone, 0, "");
236 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_data, CTLFLAG_RW,
237            &hammer_verify_data, 0, "");
238 SYSCTL_INT(_vfs_hammer, OID_AUTO, write_mode, CTLFLAG_RW,
239            &hammer_write_mode, 0, "");
240 SYSCTL_INT(_vfs_hammer, OID_AUTO, yield_check, CTLFLAG_RW,
241            &hammer_yield_check, 0, "");
242 SYSCTL_INT(_vfs_hammer, OID_AUTO, fsync_mode, CTLFLAG_RW,
243            &hammer_fsync_mode, 0, "");
244
245 KTR_INFO_MASTER(hammer);
246
247 /*
248  * VFS ABI
249  */
250 static void     hammer_free_hmp(struct mount *mp);
251
252 static int      hammer_vfs_mount(struct mount *mp, char *path, caddr_t data,
253                                 struct ucred *cred);
254 static int      hammer_vfs_unmount(struct mount *mp, int mntflags);
255 static int      hammer_vfs_root(struct mount *mp, struct vnode **vpp);
256 static int      hammer_vfs_statfs(struct mount *mp, struct statfs *sbp,
257                                 struct ucred *cred);
258 static int      hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp,
259                                 struct ucred *cred);
260 static int      hammer_vfs_sync(struct mount *mp, int waitfor);
261 static int      hammer_vfs_vget(struct mount *mp, struct vnode *dvp,
262                                 ino_t ino, struct vnode **vpp);
263 static int      hammer_vfs_init(struct vfsconf *conf);
264 static int      hammer_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
265                                 struct fid *fhp, struct vnode **vpp);
266 static int      hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp);
267 static int      hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
268                                 int *exflagsp, struct ucred **credanonp);
269
270
271 static struct vfsops hammer_vfsops = {
272         .vfs_mount      = hammer_vfs_mount,
273         .vfs_unmount    = hammer_vfs_unmount,
274         .vfs_root       = hammer_vfs_root,
275         .vfs_statfs     = hammer_vfs_statfs,
276         .vfs_statvfs    = hammer_vfs_statvfs,
277         .vfs_sync       = hammer_vfs_sync,
278         .vfs_vget       = hammer_vfs_vget,
279         .vfs_init       = hammer_vfs_init,
280         .vfs_vptofh     = hammer_vfs_vptofh,
281         .vfs_fhtovp     = hammer_vfs_fhtovp,
282         .vfs_checkexp   = hammer_vfs_checkexp
283 };
284
285 MALLOC_DEFINE(M_HAMMER, "HAMMER-mount", "");
286
287 VFS_SET(hammer_vfsops, hammer, 0);
288 MODULE_VERSION(hammer, 1);
289
290 static int
291 hammer_vfs_init(struct vfsconf *conf)
292 {
293         int n;
294
295         if (hammer_limit_recs == 0) {
296                 hammer_limit_recs = nbuf * 25;
297                 n = kmalloc_limit(M_HAMMER) / 512;
298                 if (hammer_limit_recs > n)
299                         hammer_limit_recs = n;
300         }
301         if (hammer_limit_dirtybufspace == 0) {
302                 hammer_limit_dirtybufspace = hidirtybufspace / 2;
303                 if (hammer_limit_dirtybufspace < 100)
304                         hammer_limit_dirtybufspace = 100;
305         }
306         return(0);
307 }
308
309 static int
310 hammer_vfs_mount(struct mount *mp, char *mntpt, caddr_t data,
311                  struct ucred *cred)
312 {
313         struct hammer_mount_info info;
314         hammer_mount_t hmp;
315         hammer_volume_t rootvol;
316         struct vnode *rootvp;
317         struct vnode *devvp = NULL;
318         const char *upath;      /* volume name in userspace */
319         char *path;             /* volume name in system space */
320         int error;
321         int i;
322         int master_id;
323         char *next_volume_ptr = NULL;
324
325         /*
326          * Accept hammer_mount_info.  mntpt is NULL for root mounts at boot.
327          */
328         if (mntpt == NULL) {
329                 bzero(&info, sizeof(info));
330                 info.asof = 0;
331                 info.hflags = 0;
332                 info.nvolumes = 1;
333
334                 next_volume_ptr = mp->mnt_stat.f_mntfromname;
335
336                 /* Count number of volumes separated by ':' */
337                 for (char *p = next_volume_ptr; *p != '\0'; ++p) {
338                         if (*p == ':') {
339                                 ++info.nvolumes;
340                         }
341                 }
342
343                 mp->mnt_flag &= ~MNT_RDONLY; /* mount R/W */
344         } else {
345                 if ((error = copyin(data, &info, sizeof(info))) != 0)
346                         return (error);
347         }
348
349         /*
350          * updating or new mount
351          */
352         if (mp->mnt_flag & MNT_UPDATE) {
353                 hmp = (void *)mp->mnt_data;
354                 KKASSERT(hmp != NULL);
355         } else {
356                 if (info.nvolumes <= 0 || info.nvolumes >= 32768)
357                         return (EINVAL);
358                 hmp = NULL;
359         }
360
361         /*
362          * master-id validation.  The master id may not be changed by a
363          * mount update.
364          */
365         if (info.hflags & HMNT_MASTERID) {
366                 if (hmp && hmp->master_id != info.master_id) {
367                         kprintf("hammer: cannot change master id "
368                                 "with mount update\n");
369                         return(EINVAL);
370                 }
371                 master_id = info.master_id;
372                 if (master_id < -1 || master_id >= HAMMER_MAX_MASTERS)
373                         return (EINVAL);
374         } else {
375                 if (hmp)
376                         master_id = hmp->master_id;
377                 else
378                         master_id = 0;
379         }
380
381         /*
382          * Interal mount data structure
383          */
384         if (hmp == NULL) {
385                 hmp = kmalloc(sizeof(*hmp), M_HAMMER, M_WAITOK | M_ZERO);
386                 mp->mnt_data = (qaddr_t)hmp;
387                 hmp->mp = mp;
388                 /*TAILQ_INIT(&hmp->recycle_list);*/
389
390                 /*
391                  * Make sure kmalloc type limits are set appropriately.
392                  *
393                  * Our inode kmalloc group is sized based on maxvnodes
394                  * (controlled by the system, not us).
395                  */
396                 kmalloc_create(&hmp->m_misc, "HAMMER-others");
397                 kmalloc_create(&hmp->m_inodes, "HAMMER-inodes");
398
399                 kmalloc_raise_limit(hmp->m_inodes, 0);  /* unlimited */
400
401                 hmp->root_btree_beg.localization = 0x00000000U;
402                 hmp->root_btree_beg.obj_id = -0x8000000000000000LL;
403                 hmp->root_btree_beg.key = -0x8000000000000000LL;
404                 hmp->root_btree_beg.create_tid = 1;
405                 hmp->root_btree_beg.delete_tid = 1;
406                 hmp->root_btree_beg.rec_type = 0;
407                 hmp->root_btree_beg.obj_type = 0;
408
409                 hmp->root_btree_end.localization = 0xFFFFFFFFU;
410                 hmp->root_btree_end.obj_id = 0x7FFFFFFFFFFFFFFFLL;
411                 hmp->root_btree_end.key = 0x7FFFFFFFFFFFFFFFLL;
412                 hmp->root_btree_end.create_tid = 0xFFFFFFFFFFFFFFFFULL;
413                 hmp->root_btree_end.delete_tid = 0;   /* special case */
414                 hmp->root_btree_end.rec_type = 0xFFFFU;
415                 hmp->root_btree_end.obj_type = 0;
416
417                 hmp->krate.freq = 1;    /* maximum reporting rate (hz) */
418                 hmp->krate.count = -16; /* initial burst */
419
420                 hmp->sync_lock.refs = 1;
421                 hmp->free_lock.refs = 1;
422                 hmp->undo_lock.refs = 1;
423                 hmp->blkmap_lock.refs = 1;
424                 hmp->snapshot_lock.refs = 1;
425                 hmp->volume_lock.refs = 1;
426
427                 TAILQ_INIT(&hmp->delay_list);
428                 TAILQ_INIT(&hmp->flush_group_list);
429                 TAILQ_INIT(&hmp->objid_cache_list);
430                 TAILQ_INIT(&hmp->undo_lru_list);
431                 TAILQ_INIT(&hmp->reclaim_list);
432         }
433         hmp->hflags &= ~HMNT_USERFLAGS;
434         hmp->hflags |= info.hflags & HMNT_USERFLAGS;
435
436         hmp->master_id = master_id;
437
438         if (info.asof) {
439                 mp->mnt_flag |= MNT_RDONLY;
440                 hmp->asof = info.asof;
441         } else {
442                 hmp->asof = HAMMER_MAX_TID;
443         }
444
445         hmp->volume_to_remove = -1;
446
447         /*
448          * Re-open read-write if originally read-only, or vise-versa.
449          *
450          * When going from read-only to read-write execute the stage2
451          * recovery if it has not already been run.
452          */
453         if (mp->mnt_flag & MNT_UPDATE) {
454                 error = 0;
455                 if (hmp->ronly && (mp->mnt_kern_flag & MNTK_WANTRDWR)) {
456                         kprintf("HAMMER read-only -> read-write\n");
457                         hmp->ronly = 0;
458                         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
459                                 hammer_adjust_volume_mode, NULL);
460                         rootvol = hammer_get_root_volume(hmp, &error);
461                         if (rootvol) {
462                                 hammer_recover_flush_buffers(hmp, rootvol, 1);
463                                 error = hammer_recover_stage2(hmp, rootvol);
464                                 bcopy(rootvol->ondisk->vol0_blockmap,
465                                       hmp->blockmap,
466                                       sizeof(hmp->blockmap));
467                                 hammer_rel_volume(rootvol, 0);
468                         }
469                         RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
470                                 hammer_reload_inode, NULL);
471                         /* kernel clears MNT_RDONLY */
472                 } else if (hmp->ronly == 0 && (mp->mnt_flag & MNT_RDONLY)) {
473                         kprintf("HAMMER read-write -> read-only\n");
474                         hmp->ronly = 1; /* messy */
475                         RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
476                                 hammer_reload_inode, NULL);
477                         hmp->ronly = 0;
478                         hammer_flusher_sync(hmp);
479                         hammer_flusher_sync(hmp);
480                         hammer_flusher_sync(hmp);
481                         hmp->ronly = 1;
482                         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
483                                 hammer_adjust_volume_mode, NULL);
484                 }
485                 return(error);
486         }
487
488         RB_INIT(&hmp->rb_vols_root);
489         RB_INIT(&hmp->rb_inos_root);
490         RB_INIT(&hmp->rb_redo_root);
491         RB_INIT(&hmp->rb_nods_root);
492         RB_INIT(&hmp->rb_undo_root);
493         RB_INIT(&hmp->rb_resv_root);
494         RB_INIT(&hmp->rb_bufs_root);
495         RB_INIT(&hmp->rb_pfsm_root);
496
497         hmp->ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
498
499         TAILQ_INIT(&hmp->volu_list);
500         TAILQ_INIT(&hmp->undo_list);
501         TAILQ_INIT(&hmp->data_list);
502         TAILQ_INIT(&hmp->meta_list);
503         TAILQ_INIT(&hmp->lose_list);
504         TAILQ_INIT(&hmp->iorun_list);
505
506         /*
507          * Load volumes
508          */
509         path = objcache_get(namei_oc, M_WAITOK);
510         hmp->nvolumes = -1;
511         for (i = 0; i < info.nvolumes; ++i) {
512                 if (mntpt == NULL) {
513                         /*
514                          * Root mount.
515                          */
516                         KKASSERT(next_volume_ptr != NULL);
517                         strcpy(path, "");
518                         if (*next_volume_ptr != '/') {
519                                 /* relative path */
520                                 strcpy(path, "/dev/");
521                         }
522                         int k;
523                         for (k = strlen(path); k < MAXPATHLEN-1; ++k) {
524                                 if (*next_volume_ptr == '\0') {
525                                         break;
526                                 } else if (*next_volume_ptr == ':') {
527                                         ++next_volume_ptr;
528                                         break;
529                                 } else {
530                                         path[k] = *next_volume_ptr;
531                                         ++next_volume_ptr;
532                                 }
533                         }
534                         path[k] = '\0';
535
536                         error = 0;
537                         cdev_t dev = kgetdiskbyname(path);
538                         error = bdevvp(dev, &devvp);
539                         if (error) {
540                                 kprintf("hammer_mountroot: can't find devvp\n");
541                         }
542                 } else {
543                         error = copyin(&info.volumes[i], &upath,
544                                        sizeof(char *));
545                         if (error == 0)
546                                 error = copyinstr(upath, path,
547                                                   MAXPATHLEN, NULL);
548                 }
549                 if (error == 0)
550                         error = hammer_install_volume(hmp, path, devvp);
551                 if (error)
552                         break;
553         }
554         objcache_put(namei_oc, path);
555
556         /*
557          * Make sure we found a root volume
558          */
559         if (error == 0 && hmp->rootvol == NULL) {
560                 kprintf("hammer_mount: No root volume found!\n");
561                 error = EINVAL;
562         }
563
564         /*
565          * Check that all required volumes are available
566          */
567         if (error == 0 && hammer_mountcheck_volumes(hmp)) {
568                 kprintf("hammer_mount: Missing volumes, cannot mount!\n");
569                 error = EINVAL;
570         }
571
572         if (error) {
573                 hammer_free_hmp(mp);
574                 return (error);
575         }
576
577         /*
578          * No errors, setup enough of the mount point so we can lookup the
579          * root vnode.
580          */
581         mp->mnt_iosize_max = MAXPHYS;
582         mp->mnt_kern_flag |= MNTK_FSMID;
583
584         /*
585          * MPSAFE code.  Note that VOPs and VFSops which are not MPSAFE
586          * will acquire a per-mount token prior to entry and release it
587          * on return, so even if we do not specify it we no longer get
588          * the BGL regardlless of how we are flagged.
589          */
590         mp->mnt_kern_flag |= MNTK_RD_MPSAFE | MNTK_GA_MPSAFE |
591                              MNTK_IN_MPSAFE;
592
593         /* 
594          * note: f_iosize is used by vnode_pager_haspage() when constructing
595          * its VOP_BMAP call.
596          */
597         mp->mnt_stat.f_iosize = HAMMER_BUFSIZE;
598         mp->mnt_stat.f_bsize = HAMMER_BUFSIZE;
599
600         mp->mnt_vstat.f_frsize = HAMMER_BUFSIZE;
601         mp->mnt_vstat.f_bsize = HAMMER_BUFSIZE;
602
603         mp->mnt_maxsymlinklen = 255;
604         mp->mnt_flag |= MNT_LOCAL;
605
606         vfs_add_vnodeops(mp, &hammer_vnode_vops, &mp->mnt_vn_norm_ops);
607         vfs_add_vnodeops(mp, &hammer_spec_vops, &mp->mnt_vn_spec_ops);
608         vfs_add_vnodeops(mp, &hammer_fifo_vops, &mp->mnt_vn_fifo_ops);
609
610         /*
611          * The root volume's ondisk pointer is only valid if we hold a
612          * reference to it.
613          */
614         rootvol = hammer_get_root_volume(hmp, &error);
615         if (error)
616                 goto failed;
617
618         /*
619          * Perform any necessary UNDO operations.  The recovery code does
620          * call hammer_undo_lookup() so we have to pre-cache the blockmap,
621          * and then re-copy it again after recovery is complete.
622          *
623          * If this is a read-only mount the UNDO information is retained
624          * in memory in the form of dirty buffer cache buffers, and not
625          * written back to the media.
626          */
627         bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
628               sizeof(hmp->blockmap));
629
630         /*
631          * Check filesystem version
632          */
633         hmp->version = rootvol->ondisk->vol_version;
634         if (hmp->version < HAMMER_VOL_VERSION_MIN ||
635             hmp->version > HAMMER_VOL_VERSION_MAX) {
636                 kprintf("HAMMER: mount unsupported fs version %d\n",
637                         hmp->version);
638                 error = ERANGE;
639                 goto done;
640         }
641
642         /*
643          * The undo_rec_limit limits the size of flush groups to avoid
644          * blowing out the UNDO FIFO.  This calculation is typically in
645          * the tens of thousands and is designed primarily when small
646          * HAMMER filesystems are created.
647          */
648         hmp->undo_rec_limit = hammer_undo_max(hmp) / 8192 + 100;
649         if (hammer_debug_general & 0x0001)
650                 kprintf("HAMMER: undo_rec_limit %d\n", hmp->undo_rec_limit);
651
652         /*
653          * NOTE: Recover stage1 not only handles meta-data recovery, it
654          *       also sets hmp->undo_seqno for HAMMER VERSION 4+ filesystems.
655          */
656         error = hammer_recover_stage1(hmp, rootvol);
657         if (error) {
658                 kprintf("Failed to recover HAMMER filesystem on mount\n");
659                 goto done;
660         }
661
662         /*
663          * Finish setup now that we have a good root volume.
664          *
665          * The top 16 bits of fsid.val[1] is a pfs id.
666          */
667         ksnprintf(mp->mnt_stat.f_mntfromname,
668                   sizeof(mp->mnt_stat.f_mntfromname), "%s",
669                   rootvol->ondisk->vol_name);
670         mp->mnt_stat.f_fsid.val[0] =
671                 crc32((char *)&rootvol->ondisk->vol_fsid + 0, 8);
672         mp->mnt_stat.f_fsid.val[1] =
673                 crc32((char *)&rootvol->ondisk->vol_fsid + 8, 8);
674         mp->mnt_stat.f_fsid.val[1] &= 0x0000FFFF;
675
676         mp->mnt_vstat.f_fsid_uuid = rootvol->ondisk->vol_fsid;
677         mp->mnt_vstat.f_fsid = crc32(&mp->mnt_vstat.f_fsid_uuid,
678                                      sizeof(mp->mnt_vstat.f_fsid_uuid));
679
680         /*
681          * Certain often-modified fields in the root volume are cached in
682          * the hammer_mount structure so we do not have to generate lots
683          * of little UNDO structures for them.
684          *
685          * Recopy after recovery.  This also has the side effect of
686          * setting our cached undo FIFO's first_offset, which serves to
687          * placemark the FIFO start for the NEXT flush cycle while the
688          * on-disk first_offset represents the LAST flush cycle.
689          */
690         hmp->next_tid = rootvol->ondisk->vol0_next_tid;
691         hmp->flush_tid1 = hmp->next_tid;
692         hmp->flush_tid2 = hmp->next_tid;
693         bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
694               sizeof(hmp->blockmap));
695         hmp->copy_stat_freebigblocks = rootvol->ondisk->vol0_stat_freebigblocks;
696
697         hammer_flusher_create(hmp);
698
699         /*
700          * Locate the root directory using the root cluster's B-Tree as a
701          * starting point.  The root directory uses an obj_id of 1.
702          *
703          * FUTURE: Leave the root directory cached referenced but unlocked
704          * in hmp->rootvp (need to flush it on unmount).
705          */
706         error = hammer_vfs_vget(mp, NULL, 1, &rootvp);
707         if (error)
708                 goto done;
709         vput(rootvp);
710         /*vn_unlock(hmp->rootvp);*/
711         if (hmp->ronly == 0)
712                 error = hammer_recover_stage2(hmp, rootvol);
713
714 done:
715         hammer_rel_volume(rootvol, 0);
716 failed:
717         /*
718          * Cleanup and return.
719          */
720         if (error)
721                 hammer_free_hmp(mp);
722         return (error);
723 }
724
725 static int
726 hammer_vfs_unmount(struct mount *mp, int mntflags)
727 {
728 #if 0
729         struct hammer_mount *hmp = (void *)mp->mnt_data;
730 #endif
731         int flags;
732         int error;
733
734         /*
735          * Clean out the vnodes
736          */
737         flags = 0;
738         if (mntflags & MNT_FORCE)
739                 flags |= FORCECLOSE;
740         if ((error = vflush(mp, 0, flags)) != 0)
741                 return (error);
742
743         /*
744          * Clean up the internal mount structure and related entities.  This
745          * may issue I/O.
746          */
747         hammer_free_hmp(mp);
748         return(0);
749 }
750
751 /*
752  * Clean up the internal mount structure and disassociate it from the mount.
753  * This may issue I/O.
754  */
755 static void
756 hammer_free_hmp(struct mount *mp)
757 {
758         struct hammer_mount *hmp = (void *)mp->mnt_data;
759         hammer_flush_group_t flg;
760         int count;
761         int dummy;
762
763         /*
764          * Flush anything dirty.  This won't even run if the
765          * filesystem errored-out.
766          */
767         count = 0;
768         while (hammer_flusher_haswork(hmp)) {
769                 hammer_flusher_sync(hmp);
770                 ++count;
771                 if (count >= 5) {
772                         if (count == 5)
773                                 kprintf("HAMMER: umount flushing.");
774                         else
775                                 kprintf(".");
776                         tsleep(&dummy, 0, "hmrufl", hz);
777                 }
778                 if (count == 30) {
779                         kprintf("giving up\n");
780                         break;
781                 }
782         }
783         if (count >= 5 && count < 30)
784                 kprintf("\n");
785
786         /*
787          * If the mount had a critical error we have to destroy any
788          * remaining inodes before we can finish cleaning up the flusher.
789          */
790         if (hmp->flags & HAMMER_MOUNT_CRITICAL_ERROR) {
791                 RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
792                         hammer_destroy_inode_callback, NULL);
793         }
794
795         /*
796          * There shouldn't be any inodes left now and any left over
797          * flush groups should now be empty.
798          */
799         KKASSERT(RB_EMPTY(&hmp->rb_inos_root));
800         while ((flg = TAILQ_FIRST(&hmp->flush_group_list)) != NULL) {
801                 TAILQ_REMOVE(&hmp->flush_group_list, flg, flush_entry);
802                 KKASSERT(RB_EMPTY(&flg->flush_tree));
803                 if (flg->refs) {
804                         kprintf("HAMMER: Warning, flush_group %p was "
805                                 "not empty on umount!\n", flg);
806                 }
807                 kfree(flg, hmp->m_misc);
808         }
809
810         /*
811          * We can finally destroy the flusher
812          */
813         hammer_flusher_destroy(hmp);
814
815         /*
816          * We may have held recovered buffers due to a read-only mount.
817          * These must be discarded.
818          */
819         if (hmp->ronly)
820                 hammer_recover_flush_buffers(hmp, NULL, -1);
821
822         /*
823          * Unload buffers and then volumes
824          */
825         RB_SCAN(hammer_buf_rb_tree, &hmp->rb_bufs_root, NULL,
826                 hammer_unload_buffer, NULL);
827         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
828                 hammer_unload_volume, NULL);
829
830         mp->mnt_data = NULL;
831         mp->mnt_flag &= ~MNT_LOCAL;
832         hmp->mp = NULL;
833         hammer_destroy_objid_cache(hmp);
834         kmalloc_destroy(&hmp->m_misc);
835         kmalloc_destroy(&hmp->m_inodes);
836         kfree(hmp, M_HAMMER);
837 }
838
839 /*
840  * Report critical errors.  ip may be NULL.
841  */
842 void
843 hammer_critical_error(hammer_mount_t hmp, hammer_inode_t ip,
844                       int error, const char *msg)
845 {
846         hmp->flags |= HAMMER_MOUNT_CRITICAL_ERROR;
847
848         krateprintf(&hmp->krate,
849                     "HAMMER(%s): Critical error inode=%jd error=%d %s\n",
850                     hmp->mp->mnt_stat.f_mntfromname,
851                     (intmax_t)(ip ? ip->obj_id : -1),
852                     error, msg);
853
854         if (hmp->ronly == 0) {
855                 hmp->ronly = 2;         /* special errored read-only mode */
856                 hmp->mp->mnt_flag |= MNT_RDONLY;
857                 kprintf("HAMMER(%s): Forcing read-only mode\n",
858                         hmp->mp->mnt_stat.f_mntfromname);
859         }
860         hmp->error = error;
861         if (hammer_debug_critical)
862                 Debugger("Entering debugger");
863 }
864
865
866 /*
867  * Obtain a vnode for the specified inode number.  An exclusively locked
868  * vnode is returned.
869  */
870 int
871 hammer_vfs_vget(struct mount *mp, struct vnode *dvp,
872                 ino_t ino, struct vnode **vpp)
873 {
874         struct hammer_transaction trans;
875         struct hammer_mount *hmp = (void *)mp->mnt_data;
876         struct hammer_inode *ip;
877         int error;
878         u_int32_t localization;
879
880         hammer_simple_transaction(&trans, hmp);
881
882         /*
883          * If a directory vnode is supplied (mainly NFS) then we can acquire
884          * the PFS domain from it.  Otherwise we would only be able to vget
885          * inodes in the root PFS.
886          */
887         if (dvp) {
888                 localization = HAMMER_DEF_LOCALIZATION +
889                                 VTOI(dvp)->obj_localization;
890         } else {
891                 localization = HAMMER_DEF_LOCALIZATION;
892         }
893
894         /*
895          * Lookup the requested HAMMER inode.  The structure must be
896          * left unlocked while we manipulate the related vnode to avoid
897          * a deadlock.
898          */
899         ip = hammer_get_inode(&trans, NULL, ino,
900                               hmp->asof, localization,
901                               0, &error);
902         if (ip == NULL) {
903                 *vpp = NULL;
904                 hammer_done_transaction(&trans);
905                 return(error);
906         }
907         error = hammer_get_vnode(ip, vpp);
908         hammer_rel_inode(ip, 0);
909         hammer_done_transaction(&trans);
910         return (error);
911 }
912
913 /*
914  * Return the root vnode for the filesystem.
915  *
916  * HAMMER stores the root vnode in the hammer_mount structure so
917  * getting it is easy.
918  */
919 static int
920 hammer_vfs_root(struct mount *mp, struct vnode **vpp)
921 {
922 #if 0
923         struct hammer_mount *hmp = (void *)mp->mnt_data;
924 #endif
925         int error;
926
927         error = hammer_vfs_vget(mp, NULL, 1, vpp);
928         return (error);
929 }
930
931 static int
932 hammer_vfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
933 {
934         struct hammer_mount *hmp = (void *)mp->mnt_data;
935         hammer_volume_t volume;
936         hammer_volume_ondisk_t ondisk;
937         int error;
938         int64_t bfree;
939         int64_t breserved;
940
941         volume = hammer_get_root_volume(hmp, &error);
942         if (error)
943                 return(error);
944         ondisk = volume->ondisk;
945
946         /*
947          * Basic stats
948          */
949         _hammer_checkspace(hmp, HAMMER_CHKSPC_WRITE, &breserved);
950         mp->mnt_stat.f_files = ondisk->vol0_stat_inodes;
951         bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE;
952         hammer_rel_volume(volume, 0);
953
954         mp->mnt_stat.f_bfree = (bfree - breserved) / HAMMER_BUFSIZE;
955         mp->mnt_stat.f_bavail = mp->mnt_stat.f_bfree;
956         if (mp->mnt_stat.f_files < 0)
957                 mp->mnt_stat.f_files = 0;
958
959         *sbp = mp->mnt_stat;
960         return(0);
961 }
962
963 static int
964 hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred)
965 {
966         struct hammer_mount *hmp = (void *)mp->mnt_data;
967         hammer_volume_t volume;
968         hammer_volume_ondisk_t ondisk;
969         int error;
970         int64_t bfree;
971         int64_t breserved;
972
973         volume = hammer_get_root_volume(hmp, &error);
974         if (error)
975                 return(error);
976         ondisk = volume->ondisk;
977
978         /*
979          * Basic stats
980          */
981         _hammer_checkspace(hmp, HAMMER_CHKSPC_WRITE, &breserved);
982         mp->mnt_vstat.f_files = ondisk->vol0_stat_inodes;
983         bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE;
984         hammer_rel_volume(volume, 0);
985
986         mp->mnt_vstat.f_bfree = (bfree - breserved) / HAMMER_BUFSIZE;
987         mp->mnt_vstat.f_bavail = mp->mnt_vstat.f_bfree;
988         if (mp->mnt_vstat.f_files < 0)
989                 mp->mnt_vstat.f_files = 0;
990         *sbp = mp->mnt_vstat;
991         return(0);
992 }
993
994 /*
995  * Sync the filesystem.  Currently we have to run it twice, the second
996  * one will advance the undo start index to the end index, so if a crash
997  * occurs no undos will be run on mount.
998  *
999  * We do not sync the filesystem if we are called from a panic.  If we did
1000  * we might end up blowing up a sync that was already in progress.
1001  */
1002 static int
1003 hammer_vfs_sync(struct mount *mp, int waitfor)
1004 {
1005         struct hammer_mount *hmp = (void *)mp->mnt_data;
1006         int error;
1007
1008         if (panicstr == NULL) {
1009                 error = hammer_sync_hmp(hmp, waitfor);
1010         } else {
1011                 error = EIO;
1012         }
1013         return (error);
1014 }
1015
1016 /*
1017  * Convert a vnode to a file handle.
1018  */
1019 static int
1020 hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp)
1021 {
1022         hammer_inode_t ip;
1023
1024         KKASSERT(MAXFIDSZ >= 16);
1025         ip = VTOI(vp);
1026         fhp->fid_len = offsetof(struct fid, fid_data[16]);
1027         fhp->fid_ext = ip->obj_localization >> 16;
1028         bcopy(&ip->obj_id, fhp->fid_data + 0, sizeof(ip->obj_id));
1029         bcopy(&ip->obj_asof, fhp->fid_data + 8, sizeof(ip->obj_asof));
1030         return(0);
1031 }
1032
1033
1034 /*
1035  * Convert a file handle back to a vnode.
1036  *
1037  * Use rootvp to enforce PFS isolation when a PFS is exported via a
1038  * null mount.
1039  */
1040 static int
1041 hammer_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
1042                   struct fid *fhp, struct vnode **vpp)
1043 {
1044         struct hammer_transaction trans;
1045         struct hammer_inode *ip;
1046         struct hammer_inode_info info;
1047         int error;
1048         u_int32_t localization;
1049
1050         bcopy(fhp->fid_data + 0, &info.obj_id, sizeof(info.obj_id));
1051         bcopy(fhp->fid_data + 8, &info.obj_asof, sizeof(info.obj_asof));
1052         if (rootvp)
1053                 localization = VTOI(rootvp)->obj_localization;
1054         else
1055                 localization = (u_int32_t)fhp->fid_ext << 16;
1056
1057         hammer_simple_transaction(&trans, (void *)mp->mnt_data);
1058
1059         /*
1060          * Get/allocate the hammer_inode structure.  The structure must be
1061          * unlocked while we manipulate the related vnode to avoid a
1062          * deadlock.
1063          */
1064         ip = hammer_get_inode(&trans, NULL, info.obj_id,
1065                               info.obj_asof, localization, 0, &error);
1066         if (ip) {
1067                 error = hammer_get_vnode(ip, vpp);
1068                 hammer_rel_inode(ip, 0);
1069         } else {
1070                 *vpp = NULL;
1071         }
1072         hammer_done_transaction(&trans);
1073         return (error);
1074 }
1075
1076 static int
1077 hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
1078                     int *exflagsp, struct ucred **credanonp)
1079 {
1080         hammer_mount_t hmp = (void *)mp->mnt_data;
1081         struct netcred *np;
1082         int error;
1083
1084         np = vfs_export_lookup(mp, &hmp->export, nam);
1085         if (np) {
1086                 *exflagsp = np->netc_exflags;
1087                 *credanonp = &np->netc_anon;
1088                 error = 0;
1089         } else {
1090                 error = EACCES;
1091         }
1092         return (error);
1093
1094 }
1095
1096 int
1097 hammer_vfs_export(struct mount *mp, int op, const struct export_args *export)
1098 {
1099         hammer_mount_t hmp = (void *)mp->mnt_data;
1100         int error;
1101
1102         switch(op) {
1103         case MOUNTCTL_SET_EXPORT:
1104                 error = vfs_export(mp, &hmp->export, export);
1105                 break;
1106         default:
1107                 error = EOPNOTSUPP;
1108                 break;
1109         }
1110         return(error);
1111 }
1112