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