HAMMER VFS - REDO implementation base code part 3/many
[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;
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         int maxinodes;
324         char *next_volume_ptr = NULL;
325
326         /*
327          * Accept hammer_mount_info.  mntpt is NULL for root mounts at boot.
328          */
329         if (mntpt == NULL) {
330                 bzero(&info, sizeof(info));
331                 info.asof = 0;
332                 info.hflags = 0;
333                 info.nvolumes = 1;
334
335                 next_volume_ptr = mp->mnt_stat.f_mntfromname;
336
337                 /* Count number of volumes separated by ':' */
338                 for (char *p = next_volume_ptr; *p != '\0'; ++p) {
339                         if (*p == ':') {
340                                 ++info.nvolumes;
341                         }
342                 }
343
344                 mp->mnt_flag &= ~MNT_RDONLY; /* mount R/W */
345         } else {
346                 if ((error = copyin(data, &info, sizeof(info))) != 0)
347                         return (error);
348         }
349
350         /*
351          * updating or new mount
352          */
353         if (mp->mnt_flag & MNT_UPDATE) {
354                 hmp = (void *)mp->mnt_data;
355                 KKASSERT(hmp != NULL);
356         } else {
357                 if (info.nvolumes <= 0 || info.nvolumes >= 32768)
358                         return (EINVAL);
359                 hmp = NULL;
360         }
361
362         /*
363          * master-id validation.  The master id may not be changed by a
364          * mount update.
365          */
366         if (info.hflags & HMNT_MASTERID) {
367                 if (hmp && hmp->master_id != info.master_id) {
368                         kprintf("hammer: cannot change master id "
369                                 "with mount update\n");
370                         return(EINVAL);
371                 }
372                 master_id = info.master_id;
373                 if (master_id < -1 || master_id >= HAMMER_MAX_MASTERS)
374                         return (EINVAL);
375         } else {
376                 if (hmp)
377                         master_id = hmp->master_id;
378                 else
379                         master_id = 0;
380         }
381
382         /*
383          * Interal mount data structure
384          */
385         if (hmp == NULL) {
386                 hmp = kmalloc(sizeof(*hmp), M_HAMMER, M_WAITOK | M_ZERO);
387                 mp->mnt_data = (qaddr_t)hmp;
388                 hmp->mp = mp;
389                 /*TAILQ_INIT(&hmp->recycle_list);*/
390
391                 /*
392                  * Make sure kmalloc type limits are set appropriately.  If root
393                  * increases the vnode limit you may have to do a dummy remount
394                  * to adjust the HAMMER inode limit.
395                  */
396                 kmalloc_create(&hmp->m_misc, "HAMMER-others");
397                 kmalloc_create(&hmp->m_inodes, "HAMMER-inodes");
398
399                 maxinodes = desiredvnodes + desiredvnodes / 5 +
400                             hammer_limit_reclaim * 2;
401                 kmalloc_raise_limit(hmp->m_inodes,
402                                     maxinodes * sizeof(struct hammer_inode));
403
404                 hmp->root_btree_beg.localization = 0x00000000U;
405                 hmp->root_btree_beg.obj_id = -0x8000000000000000LL;
406                 hmp->root_btree_beg.key = -0x8000000000000000LL;
407                 hmp->root_btree_beg.create_tid = 1;
408                 hmp->root_btree_beg.delete_tid = 1;
409                 hmp->root_btree_beg.rec_type = 0;
410                 hmp->root_btree_beg.obj_type = 0;
411
412                 hmp->root_btree_end.localization = 0xFFFFFFFFU;
413                 hmp->root_btree_end.obj_id = 0x7FFFFFFFFFFFFFFFLL;
414                 hmp->root_btree_end.key = 0x7FFFFFFFFFFFFFFFLL;
415                 hmp->root_btree_end.create_tid = 0xFFFFFFFFFFFFFFFFULL;
416                 hmp->root_btree_end.delete_tid = 0;   /* special case */
417                 hmp->root_btree_end.rec_type = 0xFFFFU;
418                 hmp->root_btree_end.obj_type = 0;
419
420                 hmp->krate.freq = 1;    /* maximum reporting rate (hz) */
421                 hmp->krate.count = -16; /* initial burst */
422
423                 hmp->sync_lock.refs = 1;
424                 hmp->free_lock.refs = 1;
425                 hmp->undo_lock.refs = 1;
426                 hmp->blkmap_lock.refs = 1;
427                 hmp->snapshot_lock.refs = 1;
428                 hmp->volume_lock.refs = 1;
429
430                 TAILQ_INIT(&hmp->delay_list);
431                 TAILQ_INIT(&hmp->flush_group_list);
432                 TAILQ_INIT(&hmp->objid_cache_list);
433                 TAILQ_INIT(&hmp->undo_lru_list);
434                 TAILQ_INIT(&hmp->reclaim_list);
435         }
436         hmp->hflags &= ~HMNT_USERFLAGS;
437         hmp->hflags |= info.hflags & HMNT_USERFLAGS;
438
439         hmp->master_id = master_id;
440
441         if (info.asof) {
442                 mp->mnt_flag |= MNT_RDONLY;
443                 hmp->asof = info.asof;
444         } else {
445                 hmp->asof = HAMMER_MAX_TID;
446         }
447
448         hmp->volume_to_remove = -1;
449
450         /*
451          * Re-open read-write if originally read-only, or vise-versa.
452          *
453          * When going from read-only to read-write execute the stage2
454          * recovery if it has not already been run.
455          */
456         if (mp->mnt_flag & MNT_UPDATE) {
457                 error = 0;
458                 if (hmp->ronly && (mp->mnt_kern_flag & MNTK_WANTRDWR)) {
459                         kprintf("HAMMER read-only -> read-write\n");
460                         hmp->ronly = 0;
461                         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
462                                 hammer_adjust_volume_mode, NULL);
463                         rootvol = hammer_get_root_volume(hmp, &error);
464                         if (rootvol) {
465                                 hammer_recover_flush_buffers(hmp, rootvol, 1);
466                                 error = hammer_recover_stage2(hmp, rootvol);
467                                 bcopy(rootvol->ondisk->vol0_blockmap,
468                                       hmp->blockmap,
469                                       sizeof(hmp->blockmap));
470                                 hammer_rel_volume(rootvol, 0);
471                         }
472                         RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
473                                 hammer_reload_inode, NULL);
474                         /* kernel clears MNT_RDONLY */
475                 } else if (hmp->ronly == 0 && (mp->mnt_flag & MNT_RDONLY)) {
476                         kprintf("HAMMER read-write -> read-only\n");
477                         hmp->ronly = 1; /* messy */
478                         RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
479                                 hammer_reload_inode, NULL);
480                         hmp->ronly = 0;
481                         hammer_flusher_sync(hmp);
482                         hammer_flusher_sync(hmp);
483                         hammer_flusher_sync(hmp);
484                         hmp->ronly = 1;
485                         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
486                                 hammer_adjust_volume_mode, NULL);
487                 }
488                 return(error);
489         }
490
491         RB_INIT(&hmp->rb_vols_root);
492         RB_INIT(&hmp->rb_inos_root);
493         RB_INIT(&hmp->rb_redo_root);
494         RB_INIT(&hmp->rb_nods_root);
495         RB_INIT(&hmp->rb_undo_root);
496         RB_INIT(&hmp->rb_resv_root);
497         RB_INIT(&hmp->rb_bufs_root);
498         RB_INIT(&hmp->rb_pfsm_root);
499
500         hmp->ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
501
502         TAILQ_INIT(&hmp->volu_list);
503         TAILQ_INIT(&hmp->undo_list);
504         TAILQ_INIT(&hmp->data_list);
505         TAILQ_INIT(&hmp->meta_list);
506         TAILQ_INIT(&hmp->lose_list);
507
508         /*
509          * Load volumes
510          */
511         path = objcache_get(namei_oc, M_WAITOK);
512         hmp->nvolumes = -1;
513         for (i = 0; i < info.nvolumes; ++i) {
514                 if (mntpt == NULL) {
515                         /*
516                          * Root mount.
517                          */
518                         KKASSERT(next_volume_ptr != NULL);
519                         strcpy(path, "");
520                         if (*next_volume_ptr != '/') {
521                                 /* relative path */
522                                 strcpy(path, "/dev/");
523                         }
524                         int k;
525                         for (k = strlen(path); k < MAXPATHLEN-1; ++k) {
526                                 if (*next_volume_ptr == '\0') {
527                                         break;
528                                 } else if (*next_volume_ptr == ':') {
529                                         ++next_volume_ptr;
530                                         break;
531                                 } else {
532                                         path[k] = *next_volume_ptr;
533                                         ++next_volume_ptr;
534                                 }
535                         }
536                         path[k] = '\0';
537
538                         error = 0;
539                         cdev_t dev = kgetdiskbyname(path);
540                         error = bdevvp(dev, &devvp);
541                         if (error) {
542                                 kprintf("hammer_mountroot: can't find devvp\n");
543                         }
544                 } else {
545                         error = copyin(&info.volumes[i], &upath,
546                                        sizeof(char *));
547                         if (error == 0)
548                                 error = copyinstr(upath, path,
549                                                   MAXPATHLEN, NULL);
550                 }
551                 if (error == 0)
552                         error = hammer_install_volume(hmp, path, devvp);
553                 if (error)
554                         break;
555         }
556         objcache_put(namei_oc, path);
557
558         /*
559          * Make sure we found a root volume
560          */
561         if (error == 0 && hmp->rootvol == NULL) {
562                 kprintf("hammer_mount: No root volume found!\n");
563                 error = EINVAL;
564         }
565
566         /*
567          * Check that all required volumes are available
568          */
569         if (error == 0 && hammer_mountcheck_volumes(hmp)) {
570                 kprintf("hammer_mount: Missing volumes, cannot mount!\n");
571                 error = EINVAL;
572         }
573
574         if (error) {
575                 hammer_free_hmp(mp);
576                 return (error);
577         }
578
579         /*
580          * No errors, setup enough of the mount point so we can lookup the
581          * root vnode.
582          */
583         mp->mnt_iosize_max = MAXPHYS;
584         mp->mnt_kern_flag |= MNTK_FSMID;
585
586         /*
587          * MPSAFE code.  Note that VOPs and VFSops which are not MPSAFE
588          * will acquire a per-mount token prior to entry and release it
589          * on return, so even if we do not specify it we no longer get
590          * the BGL regardlless of how we are flagged.
591          */
592         mp->mnt_kern_flag |= MNTK_RD_MPSAFE | MNTK_GA_MPSAFE |
593                              MNTK_IN_MPSAFE;
594
595         /* 
596          * note: f_iosize is used by vnode_pager_haspage() when constructing
597          * its VOP_BMAP call.
598          */
599         mp->mnt_stat.f_iosize = HAMMER_BUFSIZE;
600         mp->mnt_stat.f_bsize = HAMMER_BUFSIZE;
601
602         mp->mnt_vstat.f_frsize = HAMMER_BUFSIZE;
603         mp->mnt_vstat.f_bsize = HAMMER_BUFSIZE;
604
605         mp->mnt_maxsymlinklen = 255;
606         mp->mnt_flag |= MNT_LOCAL;
607
608         vfs_add_vnodeops(mp, &hammer_vnode_vops, &mp->mnt_vn_norm_ops);
609         vfs_add_vnodeops(mp, &hammer_spec_vops, &mp->mnt_vn_spec_ops);
610         vfs_add_vnodeops(mp, &hammer_fifo_vops, &mp->mnt_vn_fifo_ops);
611
612         /*
613          * The root volume's ondisk pointer is only valid if we hold a
614          * reference to it.
615          */
616         rootvol = hammer_get_root_volume(hmp, &error);
617         if (error)
618                 goto failed;
619
620         /*
621          * Perform any necessary UNDO operations.  The recovery code does
622          * call hammer_undo_lookup() so we have to pre-cache the blockmap,
623          * and then re-copy it again after recovery is complete.
624          *
625          * If this is a read-only mount the UNDO information is retained
626          * in memory in the form of dirty buffer cache buffers, and not
627          * written back to the media.
628          */
629         bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
630               sizeof(hmp->blockmap));
631
632         /*
633          * Check filesystem version
634          */
635         hmp->version = rootvol->ondisk->vol_version;
636         if (hmp->version < HAMMER_VOL_VERSION_MIN ||
637             hmp->version > HAMMER_VOL_VERSION_MAX) {
638                 kprintf("HAMMER: mount unsupported fs version %d\n",
639                         hmp->version);
640                 error = ERANGE;
641                 goto done;
642         }
643
644         /*
645          * The undo_rec_limit limits the size of flush groups to avoid
646          * blowing out the UNDO FIFO.  This calculation is typically in
647          * the tens of thousands and is designed primarily when small
648          * HAMMER filesystems are created.
649          */
650         hmp->undo_rec_limit = hammer_undo_max(hmp) / 8192 + 100;
651         if (hammer_debug_general & 0x0001)
652                 kprintf("HAMMER: undo_rec_limit %d\n", hmp->undo_rec_limit);
653
654         /*
655          * NOTE: Recover stage1 not only handles meta-data recovery, it
656          *       also sets hmp->undo_seqno for HAMMER VERSION 4+ filesystems.
657          */
658         error = hammer_recover_stage1(hmp, rootvol);
659         if (error) {
660                 kprintf("Failed to recover HAMMER filesystem on mount\n");
661                 goto done;
662         }
663
664         /*
665          * Finish setup now that we have a good root volume.
666          *
667          * The top 16 bits of fsid.val[1] is a pfs id.
668          */
669         ksnprintf(mp->mnt_stat.f_mntfromname,
670                   sizeof(mp->mnt_stat.f_mntfromname), "%s",
671                   rootvol->ondisk->vol_name);
672         mp->mnt_stat.f_fsid.val[0] =
673                 crc32((char *)&rootvol->ondisk->vol_fsid + 0, 8);
674         mp->mnt_stat.f_fsid.val[1] =
675                 crc32((char *)&rootvol->ondisk->vol_fsid + 8, 8);
676         mp->mnt_stat.f_fsid.val[1] &= 0x0000FFFF;
677
678         mp->mnt_vstat.f_fsid_uuid = rootvol->ondisk->vol_fsid;
679         mp->mnt_vstat.f_fsid = crc32(&mp->mnt_vstat.f_fsid_uuid,
680                                      sizeof(mp->mnt_vstat.f_fsid_uuid));
681
682         /*
683          * Certain often-modified fields in the root volume are cached in
684          * the hammer_mount structure so we do not have to generate lots
685          * of little UNDO structures for them.
686          *
687          * Recopy after recovery.  This also has the side effect of
688          * setting our cached undo FIFO's first_offset, which serves to
689          * placemark the FIFO start for the NEXT flush cycle while the
690          * on-disk first_offset represents the LAST flush cycle.
691          */
692         hmp->next_tid = rootvol->ondisk->vol0_next_tid;
693         hmp->flush_tid1 = hmp->next_tid;
694         hmp->flush_tid2 = hmp->next_tid;
695         bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
696               sizeof(hmp->blockmap));
697         hmp->copy_stat_freebigblocks = rootvol->ondisk->vol0_stat_freebigblocks;
698
699         hammer_flusher_create(hmp);
700
701         /*
702          * Locate the root directory using the root cluster's B-Tree as a
703          * starting point.  The root directory uses an obj_id of 1.
704          *
705          * FUTURE: Leave the root directory cached referenced but unlocked
706          * in hmp->rootvp (need to flush it on unmount).
707          */
708         error = hammer_vfs_vget(mp, NULL, 1, &rootvp);
709         if (error)
710                 goto done;
711         vput(rootvp);
712         /*vn_unlock(hmp->rootvp);*/
713         if (hmp->ronly == 0)
714                 error = hammer_recover_stage2(hmp, rootvol);
715
716 done:
717         hammer_rel_volume(rootvol, 0);
718 failed:
719         /*
720          * Cleanup and return.
721          */
722         if (error)
723                 hammer_free_hmp(mp);
724         return (error);
725 }
726
727 static int
728 hammer_vfs_unmount(struct mount *mp, int mntflags)
729 {
730 #if 0
731         struct hammer_mount *hmp = (void *)mp->mnt_data;
732 #endif
733         int flags;
734         int error;
735
736         /*
737          * Clean out the vnodes
738          */
739         flags = 0;
740         if (mntflags & MNT_FORCE)
741                 flags |= FORCECLOSE;
742         if ((error = vflush(mp, 0, flags)) != 0)
743                 return (error);
744
745         /*
746          * Clean up the internal mount structure and related entities.  This
747          * may issue I/O.
748          */
749         hammer_free_hmp(mp);
750         return(0);
751 }
752
753 /*
754  * Clean up the internal mount structure and disassociate it from the mount.
755  * This may issue I/O.
756  */
757 static void
758 hammer_free_hmp(struct mount *mp)
759 {
760         struct hammer_mount *hmp = (void *)mp->mnt_data;
761         hammer_flush_group_t flg;
762         int count;
763         int dummy;
764
765         /*
766          * Flush anything dirty.  This won't even run if the
767          * filesystem errored-out.
768          */
769         count = 0;
770         while (hammer_flusher_haswork(hmp)) {
771                 hammer_flusher_sync(hmp);
772                 ++count;
773                 if (count >= 5) {
774                         if (count == 5)
775                                 kprintf("HAMMER: umount flushing.");
776                         else
777                                 kprintf(".");
778                         tsleep(&dummy, 0, "hmrufl", hz);
779                 }
780                 if (count == 30) {
781                         kprintf("giving up\n");
782                         break;
783                 }
784         }
785         if (count >= 5 && count < 30)
786                 kprintf("\n");
787
788         /*
789          * If the mount had a critical error we have to destroy any
790          * remaining inodes before we can finish cleaning up the flusher.
791          */
792         if (hmp->flags & HAMMER_MOUNT_CRITICAL_ERROR) {
793                 RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
794                         hammer_destroy_inode_callback, NULL);
795         }
796
797         /*
798          * There shouldn't be any inodes left now and any left over
799          * flush groups should now be empty.
800          */
801         KKASSERT(RB_EMPTY(&hmp->rb_inos_root));
802         while ((flg = TAILQ_FIRST(&hmp->flush_group_list)) != NULL) {
803                 TAILQ_REMOVE(&hmp->flush_group_list, flg, flush_entry);
804                 KKASSERT(RB_EMPTY(&flg->flush_tree));
805                 if (flg->refs) {
806                         kprintf("HAMMER: Warning, flush_group %p was "
807                                 "not empty on umount!\n", flg);
808                 }
809                 kfree(flg, hmp->m_misc);
810         }
811
812         /*
813          * We can finally destroy the flusher
814          */
815         hammer_flusher_destroy(hmp);
816
817         /*
818          * We may have held recovered buffers due to a read-only mount.
819          * These must be discarded.
820          */
821         if (hmp->ronly)
822                 hammer_recover_flush_buffers(hmp, NULL, -1);
823
824         /*
825          * Unload buffers and then volumes
826          */
827         RB_SCAN(hammer_buf_rb_tree, &hmp->rb_bufs_root, NULL,
828                 hammer_unload_buffer, NULL);
829         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
830                 hammer_unload_volume, NULL);
831
832         mp->mnt_data = NULL;
833         mp->mnt_flag &= ~MNT_LOCAL;
834         hmp->mp = NULL;
835         hammer_destroy_objid_cache(hmp);
836         kmalloc_destroy(&hmp->m_misc);
837         kmalloc_destroy(&hmp->m_inodes);
838         kfree(hmp, M_HAMMER);
839 }
840
841 /*
842  * Report critical errors.  ip may be NULL.
843  */
844 void
845 hammer_critical_error(hammer_mount_t hmp, hammer_inode_t ip,
846                       int error, const char *msg)
847 {
848         hmp->flags |= HAMMER_MOUNT_CRITICAL_ERROR;
849
850         krateprintf(&hmp->krate,
851                     "HAMMER(%s): Critical error inode=%jd error=%d %s\n",
852                     hmp->mp->mnt_stat.f_mntfromname,
853                     (intmax_t)(ip ? ip->obj_id : -1),
854                     error, msg);
855
856         if (hmp->ronly == 0) {
857                 hmp->ronly = 2;         /* special errored read-only mode */
858                 hmp->mp->mnt_flag |= MNT_RDONLY;
859                 kprintf("HAMMER(%s): Forcing read-only mode\n",
860                         hmp->mp->mnt_stat.f_mntfromname);
861         }
862         hmp->error = error;
863         if (hammer_debug_critical)
864                 Debugger("Entering debugger");
865 }
866
867
868 /*
869  * Obtain a vnode for the specified inode number.  An exclusively locked
870  * vnode is returned.
871  */
872 int
873 hammer_vfs_vget(struct mount *mp, struct vnode *dvp,
874                 ino_t ino, struct vnode **vpp)
875 {
876         struct hammer_transaction trans;
877         struct hammer_mount *hmp = (void *)mp->mnt_data;
878         struct hammer_inode *ip;
879         int error;
880         u_int32_t localization;
881
882         hammer_simple_transaction(&trans, hmp);
883
884         /*
885          * If a directory vnode is supplied (mainly NFS) then we can acquire
886          * the PFS domain from it.  Otherwise we would only be able to vget
887          * inodes in the root PFS.
888          */
889         if (dvp) {
890                 localization = HAMMER_DEF_LOCALIZATION +
891                                 VTOI(dvp)->obj_localization;
892         } else {
893                 localization = HAMMER_DEF_LOCALIZATION;
894         }
895
896         /*
897          * Lookup the requested HAMMER inode.  The structure must be
898          * left unlocked while we manipulate the related vnode to avoid
899          * a deadlock.
900          */
901         ip = hammer_get_inode(&trans, NULL, ino,
902                               hmp->asof, localization,
903                               0, &error);
904         if (ip == NULL) {
905                 *vpp = NULL;
906                 hammer_done_transaction(&trans);
907                 return(error);
908         }
909         error = hammer_get_vnode(ip, vpp);
910         hammer_rel_inode(ip, 0);
911         hammer_done_transaction(&trans);
912         return (error);
913 }
914
915 /*
916  * Return the root vnode for the filesystem.
917  *
918  * HAMMER stores the root vnode in the hammer_mount structure so
919  * getting it is easy.
920  */
921 static int
922 hammer_vfs_root(struct mount *mp, struct vnode **vpp)
923 {
924 #if 0
925         struct hammer_mount *hmp = (void *)mp->mnt_data;
926 #endif
927         int error;
928
929         error = hammer_vfs_vget(mp, NULL, 1, vpp);
930         return (error);
931 }
932
933 static int
934 hammer_vfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
935 {
936         struct hammer_mount *hmp = (void *)mp->mnt_data;
937         hammer_volume_t volume;
938         hammer_volume_ondisk_t ondisk;
939         int error;
940         int64_t bfree;
941         int64_t breserved;
942
943         volume = hammer_get_root_volume(hmp, &error);
944         if (error)
945                 return(error);
946         ondisk = volume->ondisk;
947
948         /*
949          * Basic stats
950          */
951         _hammer_checkspace(hmp, HAMMER_CHKSPC_WRITE, &breserved);
952         mp->mnt_stat.f_files = ondisk->vol0_stat_inodes;
953         bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE;
954         hammer_rel_volume(volume, 0);
955
956         mp->mnt_stat.f_bfree = (bfree - breserved) / HAMMER_BUFSIZE;
957         mp->mnt_stat.f_bavail = mp->mnt_stat.f_bfree;
958         if (mp->mnt_stat.f_files < 0)
959                 mp->mnt_stat.f_files = 0;
960
961         *sbp = mp->mnt_stat;
962         return(0);
963 }
964
965 static int
966 hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred)
967 {
968         struct hammer_mount *hmp = (void *)mp->mnt_data;
969         hammer_volume_t volume;
970         hammer_volume_ondisk_t ondisk;
971         int error;
972         int64_t bfree;
973         int64_t breserved;
974
975         volume = hammer_get_root_volume(hmp, &error);
976         if (error)
977                 return(error);
978         ondisk = volume->ondisk;
979
980         /*
981          * Basic stats
982          */
983         _hammer_checkspace(hmp, HAMMER_CHKSPC_WRITE, &breserved);
984         mp->mnt_vstat.f_files = ondisk->vol0_stat_inodes;
985         bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE;
986         hammer_rel_volume(volume, 0);
987
988         mp->mnt_vstat.f_bfree = (bfree - breserved) / HAMMER_BUFSIZE;
989         mp->mnt_vstat.f_bavail = mp->mnt_vstat.f_bfree;
990         if (mp->mnt_vstat.f_files < 0)
991                 mp->mnt_vstat.f_files = 0;
992         *sbp = mp->mnt_vstat;
993         return(0);
994 }
995
996 /*
997  * Sync the filesystem.  Currently we have to run it twice, the second
998  * one will advance the undo start index to the end index, so if a crash
999  * occurs no undos will be run on mount.
1000  *
1001  * We do not sync the filesystem if we are called from a panic.  If we did
1002  * we might end up blowing up a sync that was already in progress.
1003  */
1004 static int
1005 hammer_vfs_sync(struct mount *mp, int waitfor)
1006 {
1007         struct hammer_mount *hmp = (void *)mp->mnt_data;
1008         int error;
1009
1010         if (panicstr == NULL) {
1011                 error = hammer_sync_hmp(hmp, waitfor);
1012         } else {
1013                 error = EIO;
1014         }
1015         return (error);
1016 }
1017
1018 /*
1019  * Convert a vnode to a file handle.
1020  */
1021 static int
1022 hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp)
1023 {
1024         hammer_inode_t ip;
1025
1026         KKASSERT(MAXFIDSZ >= 16);
1027         ip = VTOI(vp);
1028         fhp->fid_len = offsetof(struct fid, fid_data[16]);
1029         fhp->fid_ext = ip->obj_localization >> 16;
1030         bcopy(&ip->obj_id, fhp->fid_data + 0, sizeof(ip->obj_id));
1031         bcopy(&ip->obj_asof, fhp->fid_data + 8, sizeof(ip->obj_asof));
1032         return(0);
1033 }
1034
1035
1036 /*
1037  * Convert a file handle back to a vnode.
1038  *
1039  * Use rootvp to enforce PFS isolation when a PFS is exported via a
1040  * null mount.
1041  */
1042 static int
1043 hammer_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
1044                   struct fid *fhp, struct vnode **vpp)
1045 {
1046         struct hammer_transaction trans;
1047         struct hammer_inode *ip;
1048         struct hammer_inode_info info;
1049         int error;
1050         u_int32_t localization;
1051
1052         bcopy(fhp->fid_data + 0, &info.obj_id, sizeof(info.obj_id));
1053         bcopy(fhp->fid_data + 8, &info.obj_asof, sizeof(info.obj_asof));
1054         if (rootvp)
1055                 localization = VTOI(rootvp)->obj_localization;
1056         else
1057                 localization = (u_int32_t)fhp->fid_ext << 16;
1058
1059         hammer_simple_transaction(&trans, (void *)mp->mnt_data);
1060
1061         /*
1062          * Get/allocate the hammer_inode structure.  The structure must be
1063          * unlocked while we manipulate the related vnode to avoid a
1064          * deadlock.
1065          */
1066         ip = hammer_get_inode(&trans, NULL, info.obj_id,
1067                               info.obj_asof, localization, 0, &error);
1068         if (ip) {
1069                 error = hammer_get_vnode(ip, vpp);
1070                 hammer_rel_inode(ip, 0);
1071         } else {
1072                 *vpp = NULL;
1073         }
1074         hammer_done_transaction(&trans);
1075         return (error);
1076 }
1077
1078 static int
1079 hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
1080                     int *exflagsp, struct ucred **credanonp)
1081 {
1082         hammer_mount_t hmp = (void *)mp->mnt_data;
1083         struct netcred *np;
1084         int error;
1085
1086         np = vfs_export_lookup(mp, &hmp->export, nam);
1087         if (np) {
1088                 *exflagsp = np->netc_exflags;
1089                 *credanonp = &np->netc_anon;
1090                 error = 0;
1091         } else {
1092                 error = EACCES;
1093         }
1094         return (error);
1095
1096 }
1097
1098 int
1099 hammer_vfs_export(struct mount *mp, int op, const struct export_args *export)
1100 {
1101         hammer_mount_t hmp = (void *)mp->mnt_data;
1102         int error;
1103
1104         switch(op) {
1105         case MOUNTCTL_SET_EXPORT:
1106                 error = vfs_export(mp, &hmp->export, export);
1107                 break;
1108         default:
1109                 error = EOPNOTSUPP;
1110                 break;
1111         }
1112         return(error);
1113 }
1114