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