sys/vfs/hammer: Remove obsolete code and comments
[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
35 #include <sys/nlookup.h>
36 #include <sys/fcntl.h>
37 #include <sys/sysctl.h>
38
39 #include "hammer.h"
40
41 /*
42  * NOTE!  Global statistics may not be MPSAFE so HAMMER never uses them
43  *        in conditionals.
44  */
45 int hammer_supported_version = HAMMER_VOL_VERSION_DEFAULT;
46 int hammer_debug_io;
47 int hammer_debug_general;
48 int hammer_debug_debug = 1;             /* medium-error panics */
49 int hammer_debug_inode;
50 int hammer_debug_locks;
51 int hammer_debug_btree;
52 int hammer_debug_tid;
53 int hammer_debug_recover;               /* -1 will disable, +1 will force */
54 int hammer_debug_recover_faults;
55 int hammer_debug_critical;              /* non-zero enter debugger on error */
56 int hammer_cluster_enable = 1;          /* enable read clustering by default */
57 int hammer_live_dedup = 0;
58 int hammer_tdmux_ticks;
59 int hammer_count_fsyncs;
60 int hammer_count_inodes;
61 int hammer_count_iqueued;
62 int hammer_count_reclaims;
63 int hammer_count_records;
64 int hammer_count_record_datas;
65 int hammer_count_volumes;
66 int hammer_count_buffers;
67 int hammer_count_nodes;
68 int64_t hammer_count_extra_space_used;
69 int64_t hammer_stats_btree_lookups;
70 int64_t hammer_stats_btree_searches;
71 int64_t hammer_stats_btree_inserts;
72 int64_t hammer_stats_btree_deletes;
73 int64_t hammer_stats_btree_elements;
74 int64_t hammer_stats_btree_splits;
75 int64_t hammer_stats_btree_iterations;
76 int64_t hammer_stats_btree_root_iterations;
77 int64_t hammer_stats_record_iterations;
78
79 int64_t hammer_stats_file_read;
80 int64_t hammer_stats_file_write;
81 int64_t hammer_stats_file_iopsr;
82 int64_t hammer_stats_file_iopsw;
83 int64_t hammer_stats_disk_read;
84 int64_t hammer_stats_disk_write;
85 int64_t hammer_stats_inode_flushes;
86 int64_t hammer_stats_commits;
87 int64_t hammer_stats_undo;
88 int64_t hammer_stats_redo;
89
90 long hammer_count_dirtybufspace;        /* global */
91 int hammer_count_refedbufs;             /* global */
92 int hammer_count_reservations;
93 long hammer_count_io_running_read;
94 long hammer_count_io_running_write;
95 int hammer_count_io_locked;
96 long hammer_limit_dirtybufspace;        /* per-mount */
97 int hammer_limit_recs;                  /* as a whole XXX */
98 int hammer_limit_inode_recs = 2048;     /* per inode */
99 int hammer_limit_reclaims;
100 int hammer_live_dedup_cache_size = DEDUP_CACHE_SIZE;
101 int hammer_limit_redo = 4096 * 1024;    /* per inode */
102 int hammer_autoflush = 500;             /* auto flush (typ on reclaim) */
103 int hammer_bio_count;
104 int hammer_verify_zone;
105 int hammer_verify_data = 1;
106 int hammer_write_mode;
107 int hammer_double_buffer;
108 int hammer_btree_full_undo = 1;
109 int hammer_yield_check = 16;
110 int hammer_fsync_mode = 3;
111 int64_t hammer_contention_count;
112 int64_t hammer_zone_limit;
113
114 /*
115  * Live dedup debug counters (sysctls are writable so that counters
116  * can be reset from userspace).
117  */
118 int64_t hammer_live_dedup_vnode_bcmps = 0;
119 int64_t hammer_live_dedup_device_bcmps = 0;
120 int64_t hammer_live_dedup_findblk_failures = 0;
121 int64_t hammer_live_dedup_bmap_saves = 0;
122
123
124 SYSCTL_NODE(_vfs, OID_AUTO, hammer, CTLFLAG_RW, 0, "HAMMER filesystem");
125
126 SYSCTL_INT(_vfs_hammer, OID_AUTO, supported_version, CTLFLAG_RD,
127            &hammer_supported_version, 0, "");
128 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_general, CTLFLAG_RW,
129            &hammer_debug_general, 0, "");
130 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_io, CTLFLAG_RW,
131            &hammer_debug_io, 0, "");
132 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_debug, CTLFLAG_RW,
133            &hammer_debug_debug, 0, "");
134 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_inode, CTLFLAG_RW,
135            &hammer_debug_inode, 0, "");
136 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_locks, CTLFLAG_RW,
137            &hammer_debug_locks, 0, "");
138 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_btree, CTLFLAG_RW,
139            &hammer_debug_btree, 0, "");
140 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_tid, CTLFLAG_RW,
141            &hammer_debug_tid, 0, "");
142 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover, CTLFLAG_RW,
143            &hammer_debug_recover, 0, "");
144 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover_faults, CTLFLAG_RW,
145            &hammer_debug_recover_faults, 0, "");
146 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_critical, CTLFLAG_RW,
147            &hammer_debug_critical, 0, "");
148 SYSCTL_INT(_vfs_hammer, OID_AUTO, cluster_enable, CTLFLAG_RW,
149            &hammer_cluster_enable, 0, "");
150 /*
151  * 0 - live dedup is disabled
152  * 1 - dedup cache is populated on reads only
153  * 2 - dedup cache is populated on both reads and writes
154  *
155  * LIVE_DEDUP IS DISABLED PERMANENTLY!  This feature appears to cause
156  * blockmap corruption over time so we've turned it off permanently.
157  */
158 SYSCTL_INT(_vfs_hammer, OID_AUTO, live_dedup, CTLFLAG_RD,
159            &hammer_live_dedup, 0, "Enable live dedup (experimental)");
160 SYSCTL_INT(_vfs_hammer, OID_AUTO, tdmux_ticks, CTLFLAG_RW,
161            &hammer_tdmux_ticks, 0, "Hammer tdmux ticks");
162
163 SYSCTL_LONG(_vfs_hammer, OID_AUTO, limit_dirtybufspace, CTLFLAG_RW,
164            &hammer_limit_dirtybufspace, 0, "");
165 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_recs, CTLFLAG_RW,
166            &hammer_limit_recs, 0, "");
167 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_inode_recs, CTLFLAG_RW,
168            &hammer_limit_inode_recs, 0, "");
169 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_reclaims, CTLFLAG_RW,
170            &hammer_limit_reclaims, 0, "");
171 SYSCTL_INT(_vfs_hammer, OID_AUTO, live_dedup_cache_size, CTLFLAG_RW,
172            &hammer_live_dedup_cache_size, 0,
173            "Number of cache entries");
174 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_redo, CTLFLAG_RW,
175            &hammer_limit_redo, 0, "");
176
177 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_fsyncs, CTLFLAG_RD,
178            &hammer_count_fsyncs, 0, "");
179 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_inodes, CTLFLAG_RD,
180            &hammer_count_inodes, 0, "");
181 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_iqueued, CTLFLAG_RD,
182            &hammer_count_iqueued, 0, "");
183 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reclaims, CTLFLAG_RD,
184            &hammer_count_reclaims, 0, "");
185 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_records, CTLFLAG_RD,
186            &hammer_count_records, 0, "");
187 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_record_datas, CTLFLAG_RD,
188            &hammer_count_record_datas, 0, "");
189 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_volumes, CTLFLAG_RD,
190            &hammer_count_volumes, 0, "");
191 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_buffers, CTLFLAG_RD,
192            &hammer_count_buffers, 0, "");
193 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_nodes, CTLFLAG_RD,
194            &hammer_count_nodes, 0, "");
195 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, count_extra_space_used, CTLFLAG_RD,
196            &hammer_count_extra_space_used, 0, "");
197
198 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_searches, CTLFLAG_RD,
199            &hammer_stats_btree_searches, 0, "");
200 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_lookups, CTLFLAG_RD,
201            &hammer_stats_btree_lookups, 0, "");
202 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_inserts, CTLFLAG_RD,
203            &hammer_stats_btree_inserts, 0, "");
204 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_deletes, CTLFLAG_RD,
205            &hammer_stats_btree_deletes, 0, "");
206 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_elements, CTLFLAG_RD,
207            &hammer_stats_btree_elements, 0, "");
208 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_splits, CTLFLAG_RD,
209            &hammer_stats_btree_splits, 0, "");
210 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_iterations, CTLFLAG_RD,
211            &hammer_stats_btree_iterations, 0, "");
212 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_root_iterations, CTLFLAG_RD,
213            &hammer_stats_btree_root_iterations, 0, "");
214 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_record_iterations, CTLFLAG_RD,
215            &hammer_stats_record_iterations, 0, "");
216
217 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_read, CTLFLAG_RD,
218            &hammer_stats_file_read, 0, "");
219 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_write, CTLFLAG_RD,
220            &hammer_stats_file_write, 0, "");
221 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_iopsr, CTLFLAG_RD,
222            &hammer_stats_file_iopsr, 0, "");
223 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_iopsw, CTLFLAG_RD,
224            &hammer_stats_file_iopsw, 0, "");
225 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_disk_read, CTLFLAG_RD,
226            &hammer_stats_disk_read, 0, "");
227 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_disk_write, CTLFLAG_RD,
228            &hammer_stats_disk_write, 0, "");
229 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_inode_flushes, CTLFLAG_RD,
230            &hammer_stats_inode_flushes, 0, "");
231 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_commits, CTLFLAG_RD,
232            &hammer_stats_commits, 0, "");
233 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_undo, CTLFLAG_RD,
234            &hammer_stats_undo, 0, "");
235 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_redo, CTLFLAG_RD,
236            &hammer_stats_redo, 0, "");
237
238 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, live_dedup_vnode_bcmps, CTLFLAG_RW,
239             &hammer_live_dedup_vnode_bcmps, 0,
240             "successful vnode buffer comparisons");
241 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, live_dedup_device_bcmps, CTLFLAG_RW,
242             &hammer_live_dedup_device_bcmps, 0,
243             "successful device buffer comparisons");
244 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, live_dedup_findblk_failures, CTLFLAG_RW,
245             &hammer_live_dedup_findblk_failures, 0,
246             "block lookup failures for comparison");
247 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, live_dedup_bmap_saves, CTLFLAG_RW,
248             &hammer_live_dedup_bmap_saves, 0,
249             "useful physical block lookups");
250
251 SYSCTL_LONG(_vfs_hammer, OID_AUTO, count_dirtybufspace, CTLFLAG_RD,
252            &hammer_count_dirtybufspace, 0, "");
253 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_refedbufs, CTLFLAG_RD,
254            &hammer_count_refedbufs, 0, "");
255 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reservations, CTLFLAG_RD,
256            &hammer_count_reservations, 0, "");
257 SYSCTL_LONG(_vfs_hammer, OID_AUTO, count_io_running_read, CTLFLAG_RD,
258            &hammer_count_io_running_read, 0, "");
259 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_locked, CTLFLAG_RD,
260            &hammer_count_io_locked, 0, "");
261 SYSCTL_LONG(_vfs_hammer, OID_AUTO, count_io_running_write, CTLFLAG_RD,
262            &hammer_count_io_running_write, 0, "");
263 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, zone_limit, CTLFLAG_RW,
264            &hammer_zone_limit, 0, "");
265 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, contention_count, CTLFLAG_RW,
266            &hammer_contention_count, 0, "");
267 SYSCTL_INT(_vfs_hammer, OID_AUTO, autoflush, CTLFLAG_RW,
268            &hammer_autoflush, 0, "");
269 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_zone, CTLFLAG_RW,
270            &hammer_verify_zone, 0, "");
271 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_data, CTLFLAG_RW,
272            &hammer_verify_data, 0, "");
273 SYSCTL_INT(_vfs_hammer, OID_AUTO, write_mode, CTLFLAG_RW,
274            &hammer_write_mode, 0, "");
275 SYSCTL_INT(_vfs_hammer, OID_AUTO, double_buffer, CTLFLAG_RW,
276            &hammer_double_buffer, 0, "");
277 SYSCTL_INT(_vfs_hammer, OID_AUTO, btree_full_undo, CTLFLAG_RW,
278            &hammer_btree_full_undo, 0, "");
279 SYSCTL_INT(_vfs_hammer, OID_AUTO, yield_check, CTLFLAG_RW,
280            &hammer_yield_check, 0, "");
281 SYSCTL_INT(_vfs_hammer, OID_AUTO, fsync_mode, CTLFLAG_RW,
282            &hammer_fsync_mode, 0, "");
283
284 /* KTR_INFO_MASTER(hammer); */
285
286 /*
287  * VFS ABI
288  */
289 static void     hammer_free_hmp(struct mount *mp);
290
291 static int      hammer_vfs_mount(struct mount *mp, char *path, caddr_t data,
292                                 struct ucred *cred);
293 static int      hammer_vfs_unmount(struct mount *mp, int mntflags);
294 static int      hammer_vfs_root(struct mount *mp, struct vnode **vpp);
295 static int      hammer_vfs_statfs(struct mount *mp, struct statfs *sbp,
296                                 struct ucred *cred);
297 static int      hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp,
298                                 struct ucred *cred);
299 static int      hammer_vfs_sync(struct mount *mp, int waitfor);
300 static int      hammer_vfs_vget(struct mount *mp, struct vnode *dvp,
301                                 ino_t ino, struct vnode **vpp);
302 static int      hammer_vfs_init(struct vfsconf *conf);
303 static int      hammer_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
304                                 struct fid *fhp, struct vnode **vpp);
305 static int      hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp);
306 static int      hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
307                                 int *exflagsp, struct ucred **credanonp);
308
309
310 static struct vfsops hammer_vfsops = {
311         .vfs_mount      = hammer_vfs_mount,
312         .vfs_unmount    = hammer_vfs_unmount,
313         .vfs_root       = hammer_vfs_root,
314         .vfs_statfs     = hammer_vfs_statfs,
315         .vfs_statvfs    = hammer_vfs_statvfs,
316         .vfs_sync       = hammer_vfs_sync,
317         .vfs_vget       = hammer_vfs_vget,
318         .vfs_init       = hammer_vfs_init,
319         .vfs_vptofh     = hammer_vfs_vptofh,
320         .vfs_fhtovp     = hammer_vfs_fhtovp,
321         .vfs_checkexp   = hammer_vfs_checkexp
322 };
323
324 MALLOC_DEFINE(M_HAMMER, "HAMMER-mount", "");
325
326 VFS_SET(hammer_vfsops, hammer, 0);
327 MODULE_VERSION(hammer, 1);
328
329 static int
330 hammer_vfs_init(struct vfsconf *conf)
331 {
332         long n;
333
334         /*
335          * Wait up to this long for an exclusive deadlock to clear
336          * before acquiring a new shared lock on the ip.  The deadlock
337          * may have occured on a b-tree node related to the ip.
338          */
339         if (hammer_tdmux_ticks == 0)
340                 hammer_tdmux_ticks = hz / 5;
341
342         /*
343          * Autosize, but be careful because a hammer filesystem's
344          * reserve is partially calculated based on dirtybufspace,
345          * so we simply cannot allow it to get too large.
346          */
347         if (hammer_limit_recs == 0) {
348                 n = nbuf * 25;
349                 if (n > kmalloc_limit(M_HAMMER) / 512)
350                         n = kmalloc_limit(M_HAMMER) / 512;
351                 if (n > 2 * 1024 * 1024)
352                         n = 2 * 1024 * 1024;
353                 hammer_limit_recs = (int)n;
354         }
355         if (hammer_limit_dirtybufspace == 0) {
356                 hammer_limit_dirtybufspace = hidirtybufspace / 2;
357                 if (hammer_limit_dirtybufspace < 1L * 1024 * 1024)
358                         hammer_limit_dirtybufspace = 1024L * 1024;
359                 if (hammer_limit_dirtybufspace > 1024L * 1024 * 1024)
360                         hammer_limit_dirtybufspace = 1024L * 1024 * 1024;
361         }
362
363         /*
364          * The hammer_inode structure detaches from the vnode on reclaim.
365          * This limits the number of inodes in this state to prevent a
366          * memory pool blowout.
367          */
368         if (hammer_limit_reclaims == 0)
369                 hammer_limit_reclaims = desiredvnodes / 10;
370
371         return(0);
372 }
373
374 static int
375 hammer_vfs_mount(struct mount *mp, char *mntpt, caddr_t data,
376                  struct ucred *cred)
377 {
378         struct hammer_mount_info info;
379         hammer_mount_t hmp;
380         hammer_volume_t rootvol;
381         struct vnode *rootvp;
382         struct vnode *devvp = NULL;
383         const char *upath;      /* volume name in userspace */
384         char *path;             /* volume name in system space */
385         int error;
386         int i;
387         int master_id;
388         int nvolumes;
389         char *next_volume_ptr = NULL;
390
391         /*
392          * Accept hammer_mount_info.  mntpt is NULL for root mounts at boot.
393          */
394         if (mntpt == NULL) {
395                 bzero(&info, sizeof(info));
396                 info.asof = 0;
397                 info.hflags = 0;
398                 info.nvolumes = 1;
399
400                 next_volume_ptr = mp->mnt_stat.f_mntfromname;
401
402                 /* Count number of volumes separated by ':' */
403                 for (char *p = next_volume_ptr; *p != '\0'; ++p) {
404                         if (*p == ':') {
405                                 ++info.nvolumes;
406                         }
407                 }
408
409                 mp->mnt_flag &= ~MNT_RDONLY; /* mount R/W */
410         } else {
411                 if ((error = copyin(data, &info, sizeof(info))) != 0)
412                         return (error);
413         }
414
415         /*
416          * updating or new mount
417          */
418         if (mp->mnt_flag & MNT_UPDATE) {
419                 hmp = (void *)mp->mnt_data;
420                 KKASSERT(hmp != NULL);
421         } else {
422                 if (info.nvolumes <= 0 || info.nvolumes > HAMMER_MAX_VOLUMES)
423                         return (EINVAL);
424                 hmp = NULL;
425         }
426
427         /*
428          * master-id validation.  The master id may not be changed by a
429          * mount update.
430          */
431         if (info.hflags & HMNT_MASTERID) {
432                 if (hmp && hmp->master_id != info.master_id) {
433                         kprintf("HAMMER: cannot change master id "
434                                 "with mount update\n");
435                         return(EINVAL);
436                 }
437                 master_id = info.master_id;
438                 if (master_id < -1 || master_id >= HAMMER_MAX_MASTERS)
439                         return (EINVAL);
440         } else {
441                 if (hmp)
442                         master_id = hmp->master_id;
443                 else
444                         master_id = 0;
445         }
446
447         /*
448          * Internal mount data structure
449          */
450         if (hmp == NULL) {
451                 hmp = kmalloc(sizeof(*hmp), M_HAMMER, M_WAITOK | M_ZERO);
452                 mp->mnt_data = (qaddr_t)hmp;
453                 hmp->mp = mp;
454
455                 /*
456                  * Make sure kmalloc type limits are set appropriately.
457                  *
458                  * Our inode kmalloc group is sized based on maxvnodes
459                  * (controlled by the system, not us).
460                  */
461                 kmalloc_create(&hmp->m_misc, "HAMMER-others");
462                 kmalloc_create(&hmp->m_inodes, "HAMMER-inodes");
463
464                 kmalloc_raise_limit(hmp->m_inodes, 0);  /* unlimited */
465
466                 hmp->root_btree_beg.localization = 0x00000000U;
467                 hmp->root_btree_beg.obj_id = -0x8000000000000000LL;
468                 hmp->root_btree_beg.key = -0x8000000000000000LL;
469                 hmp->root_btree_beg.create_tid = 1;
470                 hmp->root_btree_beg.delete_tid = 1;
471                 hmp->root_btree_beg.rec_type = 0;
472                 hmp->root_btree_beg.obj_type = 0;
473                 hmp->root_btree_beg.btype = HAMMER_BTREE_TYPE_NONE;
474
475                 hmp->root_btree_end.localization = 0xFFFFFFFFU;
476                 hmp->root_btree_end.obj_id = 0x7FFFFFFFFFFFFFFFLL;
477                 hmp->root_btree_end.key = 0x7FFFFFFFFFFFFFFFLL;
478                 hmp->root_btree_end.create_tid = 0xFFFFFFFFFFFFFFFFULL;
479                 hmp->root_btree_end.delete_tid = 0;   /* special case */
480                 hmp->root_btree_end.rec_type = 0xFFFFU;
481                 hmp->root_btree_end.obj_type = 0;
482                 hmp->root_btree_end.btype = HAMMER_BTREE_TYPE_NONE;
483
484                 hmp->krate.freq = 1;    /* maximum reporting rate (hz) */
485                 hmp->krate.count = -16; /* initial burst */
486                 hmp->kdiag.freq = 1;    /* maximum reporting rate (hz) */
487                 hmp->kdiag.count = -16; /* initial burst */
488
489                 hmp->sync_lock.refs = 1;
490                 hmp->free_lock.refs = 1;
491                 hmp->undo_lock.refs = 1;
492                 hmp->blkmap_lock.refs = 1;
493                 hmp->snapshot_lock.refs = 1;
494                 hmp->volume_lock.refs = 1;
495
496                 TAILQ_INIT(&hmp->delay_list);
497                 TAILQ_INIT(&hmp->flush_group_list);
498                 TAILQ_INIT(&hmp->objid_cache_list);
499                 TAILQ_INIT(&hmp->undo_lru_list);
500                 TAILQ_INIT(&hmp->reclaim_list);
501
502                 RB_INIT(&hmp->rb_dedup_crc_root);
503                 RB_INIT(&hmp->rb_dedup_off_root);
504                 TAILQ_INIT(&hmp->dedup_lru_list);
505         }
506         hmp->hflags &= ~HMNT_USERFLAGS;
507         hmp->hflags |= info.hflags & HMNT_USERFLAGS;
508
509         hmp->master_id = master_id;
510
511         if (info.asof) {
512                 mp->mnt_flag |= MNT_RDONLY;
513                 hmp->asof = info.asof;
514         } else {
515                 hmp->asof = HAMMER_MAX_TID;
516         }
517
518         hmp->volume_to_remove = -1;
519
520         /*
521          * Re-open read-write if originally read-only, or vise-versa.
522          *
523          * When going from read-only to read-write execute the stage2
524          * recovery if it has not already been run.
525          */
526         if (mp->mnt_flag & MNT_UPDATE) {
527                 lwkt_gettoken(&hmp->fs_token);
528                 error = 0;
529                 if (hmp->ronly && (mp->mnt_kern_flag & MNTK_WANTRDWR)) {
530                         kprintf("HAMMER: read-only -> read-write\n");
531                         hmp->ronly = 0;
532                         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
533                                 hammer_adjust_volume_mode, NULL);
534                         rootvol = hammer_get_root_volume(hmp, &error);
535                         if (rootvol) {
536                                 hammer_recover_flush_buffers(hmp, rootvol, 1);
537                                 error = hammer_recover_stage2(hmp, rootvol);
538                                 bcopy(rootvol->ondisk->vol0_blockmap,
539                                       hmp->blockmap,
540                                       sizeof(hmp->blockmap));
541                                 hammer_rel_volume(rootvol, 0);
542                         }
543                         RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
544                                 hammer_reload_inode, NULL);
545                         /* kernel clears MNT_RDONLY */
546                 } else if (hmp->ronly == 0 && (mp->mnt_flag & MNT_RDONLY)) {
547                         kprintf("HAMMER: read-write -> read-only\n");
548                         hmp->ronly = 1; /* messy */
549                         RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
550                                 hammer_reload_inode, NULL);
551                         hmp->ronly = 0;
552                         hammer_flusher_sync(hmp);
553                         hammer_flusher_sync(hmp);
554                         hammer_flusher_sync(hmp);
555                         hmp->ronly = 1;
556                         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
557                                 hammer_adjust_volume_mode, NULL);
558                 }
559                 lwkt_reltoken(&hmp->fs_token);
560                 return(error);
561         }
562
563         RB_INIT(&hmp->rb_vols_root);
564         RB_INIT(&hmp->rb_inos_root);
565         RB_INIT(&hmp->rb_redo_root);
566         RB_INIT(&hmp->rb_nods_root);
567         RB_INIT(&hmp->rb_undo_root);
568         RB_INIT(&hmp->rb_resv_root);
569         RB_INIT(&hmp->rb_bufs_root);
570         RB_INIT(&hmp->rb_pfsm_root);
571
572         hmp->ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
573
574         RB_INIT(&hmp->volu_root);
575         RB_INIT(&hmp->undo_root);
576         RB_INIT(&hmp->data_root);
577         RB_INIT(&hmp->meta_root);
578         RB_INIT(&hmp->lose_root);
579         TAILQ_INIT(&hmp->iorun_list);
580
581         lwkt_token_init(&hmp->fs_token, "hammerfs");
582         lwkt_token_init(&hmp->io_token, "hammerio");
583
584         lwkt_gettoken(&hmp->fs_token);
585
586         /*
587          * Load volumes
588          */
589         path = objcache_get(namei_oc, M_WAITOK);
590         hmp->nvolumes = -1;
591         for (i = 0; i < info.nvolumes; ++i) {
592                 if (mntpt == NULL) {
593                         /*
594                          * Root mount.
595                          */
596                         KKASSERT(next_volume_ptr != NULL);
597                         strcpy(path, "");
598                         if (*next_volume_ptr != '/') {
599                                 /* relative path */
600                                 strcpy(path, "/dev/");
601                         }
602                         int k;
603                         for (k = strlen(path); k < MAXPATHLEN-1; ++k) {
604                                 if (*next_volume_ptr == '\0') {
605                                         break;
606                                 } else if (*next_volume_ptr == ':') {
607                                         ++next_volume_ptr;
608                                         break;
609                                 } else {
610                                         path[k] = *next_volume_ptr;
611                                         ++next_volume_ptr;
612                                 }
613                         }
614                         path[k] = '\0';
615
616                         error = 0;
617                         cdev_t dev = kgetdiskbyname(path);
618                         error = bdevvp(dev, &devvp);
619                         if (error) {
620                                 kprintf("hammer_mount: can't find devvp\n");
621                         }
622                 } else {
623                         error = copyin(&info.volumes[i], &upath,
624                                        sizeof(char *));
625                         if (error == 0)
626                                 error = copyinstr(upath, path,
627                                                   MAXPATHLEN, NULL);
628                 }
629                 if (error == 0)
630                         error = hammer_install_volume(hmp, path, devvp, NULL);
631                 if (error)
632                         break;
633         }
634         objcache_put(namei_oc, path);
635
636         /*
637          * Make sure we found a root volume
638          */
639         if (hmp->rootvol == NULL) {
640                 kprintf("hammer_mount: No root volume found!\n");
641                 error = EINVAL;
642                 goto failed;
643         }
644
645         /*
646          * Check that all required volumes are available
647          */
648         if (error == 0 && hammer_mountcheck_volumes(hmp)) {
649                 kprintf("hammer_mount: Missing volumes, cannot mount!\n");
650                 error = EINVAL;
651                 goto failed;
652         }
653
654         /*
655          * Other errors
656          */
657         if (error) {
658                 kprintf("hammer_mount: Failed to load volumes!\n");
659                 goto failed;
660         }
661
662         nvolumes = hammer_get_installed_volumes(hmp);
663         if (hmp->nvolumes != nvolumes) {
664                 kprintf("hammer_mount: volume header says %d volumes, "
665                         "but %d installed\n",
666                         hmp->nvolumes, nvolumes);
667                 error = EINVAL;
668                 goto failed;
669         }
670
671         /*
672          * No errors, setup enough of the mount point so we can lookup the
673          * root vnode.
674          */
675         mp->mnt_iosize_max = MAXPHYS;
676         mp->mnt_kern_flag |= MNTK_FSMID;
677         mp->mnt_kern_flag |= MNTK_THR_SYNC;     /* new vsyncscan semantics */
678
679         /*
680          * MPSAFE code.  Note that VOPs and VFSops which are not MPSAFE
681          * will acquire a per-mount token prior to entry and release it
682          * on return.
683          */
684         mp->mnt_kern_flag |= MNTK_ALL_MPSAFE;
685
686         /*
687          * note: f_iosize is used by vnode_pager_haspage() when constructing
688          * its VOP_BMAP call.
689          */
690         mp->mnt_stat.f_iosize = HAMMER_BUFSIZE;
691         mp->mnt_stat.f_bsize = HAMMER_BUFSIZE;
692
693         mp->mnt_vstat.f_frsize = HAMMER_BUFSIZE;
694         mp->mnt_vstat.f_bsize = HAMMER_BUFSIZE;
695
696         mp->mnt_maxsymlinklen = 255;
697         mp->mnt_flag |= MNT_LOCAL;
698
699         vfs_add_vnodeops(mp, &hammer_vnode_vops, &mp->mnt_vn_norm_ops);
700         vfs_add_vnodeops(mp, &hammer_spec_vops, &mp->mnt_vn_spec_ops);
701         vfs_add_vnodeops(mp, &hammer_fifo_vops, &mp->mnt_vn_fifo_ops);
702
703         /*
704          * The root volume's ondisk pointer is only valid if we hold a
705          * reference to it.
706          */
707         rootvol = hammer_get_root_volume(hmp, &error);
708         if (error)
709                 goto failed;
710
711         /*
712          * Perform any necessary UNDO operations.  The recovery code does
713          * call hammer_undo_lookup() so we have to pre-cache the blockmap,
714          * and then re-copy it again after recovery is complete.
715          *
716          * If this is a read-only mount the UNDO information is retained
717          * in memory in the form of dirty buffer cache buffers, and not
718          * written back to the media.
719          */
720         bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
721               sizeof(hmp->blockmap));
722
723         /*
724          * Check filesystem version
725          */
726         hmp->version = rootvol->ondisk->vol_version;
727         if (hmp->version < HAMMER_VOL_VERSION_MIN ||
728             hmp->version > HAMMER_VOL_VERSION_MAX) {
729                 kprintf("HAMMER: mount unsupported fs version %d\n",
730                         hmp->version);
731                 error = ERANGE;
732                 goto done;
733         }
734
735         /*
736          * The undo_rec_limit limits the size of flush groups to avoid
737          * blowing out the UNDO FIFO.  This calculation is typically in
738          * the tens of thousands and is designed primarily when small
739          * HAMMER filesystems are created.
740          */
741         hmp->undo_rec_limit = hammer_undo_max(hmp) / 8192 + 100;
742         if (hammer_debug_general & 0x0001)
743                 kprintf("HAMMER: undo_rec_limit %d\n", hmp->undo_rec_limit);
744
745         /*
746          * NOTE: Recover stage1 not only handles meta-data recovery, it
747          *       also sets hmp->undo_seqno for HAMMER VERSION 4+ filesystems.
748          */
749         error = hammer_recover_stage1(hmp, rootvol);
750         if (error) {
751                 kprintf("Failed to recover HAMMER filesystem on mount\n");
752                 goto done;
753         }
754
755         /*
756          * Finish setup now that we have a good root volume.
757          *
758          * The top 16 bits of fsid.val[1] is a pfs id.
759          */
760         ksnprintf(mp->mnt_stat.f_mntfromname,
761                   sizeof(mp->mnt_stat.f_mntfromname), "%s",
762                   rootvol->ondisk->vol_name);
763         mp->mnt_stat.f_fsid.val[0] =
764                 crc32((char *)&rootvol->ondisk->vol_fsid + 0, 8);
765         mp->mnt_stat.f_fsid.val[1] =
766                 crc32((char *)&rootvol->ondisk->vol_fsid + 8, 8);
767         mp->mnt_stat.f_fsid.val[1] &= 0x0000FFFF;
768
769         mp->mnt_vstat.f_fsid_uuid = rootvol->ondisk->vol_fsid;
770         mp->mnt_vstat.f_fsid = crc32(&mp->mnt_vstat.f_fsid_uuid,
771                                      sizeof(mp->mnt_vstat.f_fsid_uuid));
772
773         /*
774          * Certain often-modified fields in the root volume are cached in
775          * the hammer_mount structure so we do not have to generate lots
776          * of little UNDO structures for them.
777          *
778          * Recopy after recovery.  This also has the side effect of
779          * setting our cached undo FIFO's first_offset, which serves to
780          * placemark the FIFO start for the NEXT flush cycle while the
781          * on-disk first_offset represents the LAST flush cycle.
782          */
783         hmp->next_tid = rootvol->ondisk->vol0_next_tid;
784         hmp->flush_tid1 = hmp->next_tid;
785         hmp->flush_tid2 = hmp->next_tid;
786         bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
787               sizeof(hmp->blockmap));
788         hmp->copy_stat_freebigblocks = rootvol->ondisk->vol0_stat_freebigblocks;
789
790         hammer_flusher_create(hmp);
791
792         /*
793          * Locate the root directory with an obj_id of 1.
794          */
795         error = hammer_vfs_vget(mp, NULL, HAMMER_OBJID_ROOT, &rootvp);
796         if (error)
797                 goto done;
798         vput(rootvp);
799         if (hmp->ronly == 0)
800                 error = hammer_recover_stage2(hmp, rootvol);
801
802         /*
803          * If the stage2 recovery fails be sure to clean out all cached
804          * vnodes before throwing away the mount structure or bad things
805          * will happen.
806          */
807         if (error)
808                 vflush(mp, 0, 0);
809
810 done:
811         if ((mp->mnt_flag & MNT_UPDATE) == 0) {
812                 /* New mount */
813
814                 /* Populate info for mount point (NULL pad)*/
815                 bzero(mp->mnt_stat.f_mntonname, MNAMELEN);
816                 size_t size;
817                 if (mntpt) {
818                         copyinstr(mntpt, mp->mnt_stat.f_mntonname,
819                                                         MNAMELEN -1, &size);
820                 } else { /* Root mount */
821                         mp->mnt_stat.f_mntonname[0] = '/';
822                 }
823         }
824         (void)VFS_STATFS(mp, &mp->mnt_stat, cred);
825         hammer_rel_volume(rootvol, 0);
826 failed:
827         /*
828          * Cleanup and return.
829          */
830         if (error) {
831                 /* called with fs_token held */
832                 hammer_free_hmp(mp);
833         } else {
834                 lwkt_reltoken(&hmp->fs_token);
835         }
836         return (error);
837 }
838
839 static int
840 hammer_vfs_unmount(struct mount *mp, int mntflags)
841 {
842         hammer_mount_t hmp = (void *)mp->mnt_data;
843         int flags;
844         int error;
845
846         /*
847          * Clean out the vnodes
848          */
849         lwkt_gettoken(&hmp->fs_token);
850         flags = 0;
851         if (mntflags & MNT_FORCE)
852                 flags |= FORCECLOSE;
853         error = vflush(mp, 0, flags);
854
855         /*
856          * Clean up the internal mount structure and related entities.  This
857          * may issue I/O.
858          */
859         if (error == 0) {
860                 /* called with fs_token held */
861                 hammer_free_hmp(mp);
862         } else {
863                 lwkt_reltoken(&hmp->fs_token);
864         }
865         return(error);
866 }
867
868 /*
869  * Clean up the internal mount structure and disassociate it from the mount.
870  * This may issue I/O.
871  *
872  * Called with fs_token held.
873  */
874 static void
875 hammer_free_hmp(struct mount *mp)
876 {
877         hammer_mount_t hmp = (void *)mp->mnt_data;
878         hammer_flush_group_t flg;
879
880         /*
881          * Flush anything dirty.  This won't even run if the
882          * filesystem errored-out.
883          */
884         hammer_flush_dirty(hmp, 30);
885
886         /*
887          * If the mount had a critical error we have to destroy any
888          * remaining inodes before we can finish cleaning up the flusher.
889          */
890         if (hmp->flags & HAMMER_MOUNT_CRITICAL_ERROR) {
891                 RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
892                         hammer_destroy_inode_callback, NULL);
893         }
894
895         /*
896          * There shouldn't be any inodes left now and any left over
897          * flush groups should now be empty.
898          */
899         KKASSERT(RB_EMPTY(&hmp->rb_inos_root));
900         while ((flg = TAILQ_FIRST(&hmp->flush_group_list)) != NULL) {
901                 TAILQ_REMOVE(&hmp->flush_group_list, flg, flush_entry);
902                 KKASSERT(RB_EMPTY(&flg->flush_tree));
903                 if (flg->refs) {
904                         kprintf("HAMMER: Warning, flush_group %p was "
905                                 "not empty on umount!\n", flg);
906                 }
907                 kfree(flg, hmp->m_misc);
908         }
909
910         /*
911          * We can finally destroy the flusher
912          */
913         hammer_flusher_destroy(hmp);
914
915         /*
916          * We may have held recovered buffers due to a read-only mount.
917          * These must be discarded.
918          */
919         if (hmp->ronly)
920                 hammer_recover_flush_buffers(hmp, NULL, -1);
921
922         /*
923          * Unload buffers and then volumes
924          */
925         RB_SCAN(hammer_buf_rb_tree, &hmp->rb_bufs_root, NULL,
926                 hammer_unload_buffer, NULL);
927         RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
928                 hammer_unload_volume, NULL);
929
930         mp->mnt_data = NULL;
931         mp->mnt_flag &= ~MNT_LOCAL;
932         hmp->mp = NULL;
933         hammer_destroy_objid_cache(hmp);
934         hammer_destroy_dedup_cache(hmp);
935         if (hmp->dedup_free_cache != NULL) {
936                 kfree(hmp->dedup_free_cache, hmp->m_misc);
937                 hmp->dedup_free_cache = NULL;
938         }
939         kmalloc_destroy(&hmp->m_misc);
940         kmalloc_destroy(&hmp->m_inodes);
941         lwkt_reltoken(&hmp->fs_token);
942         kfree(hmp, M_HAMMER);
943 }
944
945 /*
946  * Report critical errors.  ip may be NULL.
947  */
948 void
949 hammer_critical_error(hammer_mount_t hmp, hammer_inode_t ip,
950                       int error, const char *msg)
951 {
952         hmp->flags |= HAMMER_MOUNT_CRITICAL_ERROR;
953
954         krateprintf(&hmp->krate,
955                     "HAMMER(%s): Critical error inode=%jd error=%d %s\n",
956                     hmp->mp->mnt_stat.f_mntfromname,
957                     (intmax_t)(ip ? ip->obj_id : -1),
958                     error, msg);
959
960         if (hmp->ronly == 0) {
961                 hmp->ronly = 2;         /* special errored read-only mode */
962                 hmp->mp->mnt_flag |= MNT_RDONLY;
963                 RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
964                         hammer_adjust_volume_mode, NULL);
965                 kprintf("HAMMER(%s): Forcing read-only mode\n",
966                         hmp->mp->mnt_stat.f_mntfromname);
967         }
968         hmp->error = error;
969         if (hammer_debug_critical)
970                 Debugger("Entering debugger");
971 }
972
973
974 /*
975  * Obtain a vnode for the specified inode number.  An exclusively locked
976  * vnode is returned.
977  */
978 int
979 hammer_vfs_vget(struct mount *mp, struct vnode *dvp,
980                 ino_t ino, struct vnode **vpp)
981 {
982         struct hammer_transaction trans;
983         struct hammer_mount *hmp = (void *)mp->mnt_data;
984         struct hammer_inode *ip;
985         int error;
986         u_int32_t localization;
987
988         lwkt_gettoken(&hmp->fs_token);
989         hammer_simple_transaction(&trans, hmp);
990
991         /*
992          * If a directory vnode is supplied (mainly NFS) then we can acquire
993          * the PFS domain from it.  Otherwise we would only be able to vget
994          * inodes in the root PFS.
995          */
996         if (dvp) {
997                 localization = HAMMER_DEF_LOCALIZATION +
998                                 VTOI(dvp)->obj_localization;
999         } else {
1000                 localization = HAMMER_DEF_LOCALIZATION;
1001         }
1002
1003         /*
1004          * Lookup the requested HAMMER inode.  The structure must be
1005          * left unlocked while we manipulate the related vnode to avoid
1006          * a deadlock.
1007          */
1008         ip = hammer_get_inode(&trans, NULL, ino,
1009                               hmp->asof, localization,
1010                               0, &error);
1011         if (ip == NULL) {
1012                 *vpp = NULL;
1013         } else {
1014                 error = hammer_get_vnode(ip, vpp);
1015                 hammer_rel_inode(ip, 0);
1016         }
1017         hammer_done_transaction(&trans);
1018         lwkt_reltoken(&hmp->fs_token);
1019         return (error);
1020 }
1021
1022 /*
1023  * Return the root vnode for the filesystem.
1024  *
1025  * HAMMER stores the root vnode in the hammer_mount structure so
1026  * getting it is easy.
1027  */
1028 static int
1029 hammer_vfs_root(struct mount *mp, struct vnode **vpp)
1030 {
1031         int error;
1032
1033         error = hammer_vfs_vget(mp, NULL, HAMMER_OBJID_ROOT, vpp);
1034         return (error);
1035 }
1036
1037 static int
1038 hammer_vfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
1039 {
1040         struct hammer_mount *hmp = (void *)mp->mnt_data;
1041         hammer_volume_t volume;
1042         hammer_volume_ondisk_t ondisk;
1043         int error;
1044         int64_t bfree;
1045         int64_t breserved;
1046
1047         lwkt_gettoken(&hmp->fs_token);
1048         volume = hammer_get_root_volume(hmp, &error);
1049         if (error) {
1050                 lwkt_reltoken(&hmp->fs_token);
1051                 return(error);
1052         }
1053         ondisk = volume->ondisk;
1054
1055         /*
1056          * Basic stats
1057          */
1058         _hammer_checkspace(hmp, HAMMER_CHKSPC_WRITE, &breserved);
1059         mp->mnt_stat.f_files = ondisk->vol0_stat_inodes;
1060         bfree = ondisk->vol0_stat_freebigblocks * HAMMER_BIGBLOCK_SIZE;
1061         hammer_rel_volume(volume, 0);
1062
1063         mp->mnt_stat.f_bfree = (bfree - breserved) / HAMMER_BUFSIZE;
1064         mp->mnt_stat.f_bavail = mp->mnt_stat.f_bfree;
1065         if (mp->mnt_stat.f_files < 0)
1066                 mp->mnt_stat.f_files = 0;
1067
1068         *sbp = mp->mnt_stat;
1069         lwkt_reltoken(&hmp->fs_token);
1070         return(0);
1071 }
1072
1073 static int
1074 hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred)
1075 {
1076         struct hammer_mount *hmp = (void *)mp->mnt_data;
1077         hammer_volume_t volume;
1078         hammer_volume_ondisk_t ondisk;
1079         int error;
1080         int64_t bfree;
1081         int64_t breserved;
1082
1083         lwkt_gettoken(&hmp->fs_token);
1084         volume = hammer_get_root_volume(hmp, &error);
1085         if (error) {
1086                 lwkt_reltoken(&hmp->fs_token);
1087                 return(error);
1088         }
1089         ondisk = volume->ondisk;
1090
1091         /*
1092          * Basic stats
1093          */
1094         _hammer_checkspace(hmp, HAMMER_CHKSPC_WRITE, &breserved);
1095         mp->mnt_vstat.f_files = ondisk->vol0_stat_inodes;
1096         bfree = ondisk->vol0_stat_freebigblocks * HAMMER_BIGBLOCK_SIZE;
1097         hammer_rel_volume(volume, 0);
1098
1099         mp->mnt_vstat.f_bfree = (bfree - breserved) / HAMMER_BUFSIZE;
1100         mp->mnt_vstat.f_bavail = mp->mnt_vstat.f_bfree;
1101         if (mp->mnt_vstat.f_files < 0)
1102                 mp->mnt_vstat.f_files = 0;
1103         *sbp = mp->mnt_vstat;
1104         lwkt_reltoken(&hmp->fs_token);
1105         return(0);
1106 }
1107
1108 /*
1109  * Sync the filesystem.  Currently we have to run it twice, the second
1110  * one will advance the undo start index to the end index, so if a crash
1111  * occurs no undos will be run on mount.
1112  *
1113  * We do not sync the filesystem if we are called from a panic.  If we did
1114  * we might end up blowing up a sync that was already in progress.
1115  */
1116 static int
1117 hammer_vfs_sync(struct mount *mp, int waitfor)
1118 {
1119         struct hammer_mount *hmp = (void *)mp->mnt_data;
1120         int error;
1121
1122         lwkt_gettoken(&hmp->fs_token);
1123         if (panicstr == NULL) {
1124                 error = hammer_sync_hmp(hmp, waitfor);
1125         } else {
1126                 error = EIO;
1127         }
1128         lwkt_reltoken(&hmp->fs_token);
1129         return (error);
1130 }
1131
1132 /*
1133  * Convert a vnode to a file handle.
1134  *
1135  * Accesses read-only fields on already-referenced structures so
1136  * no token is needed.
1137  */
1138 static int
1139 hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp)
1140 {
1141         hammer_inode_t ip;
1142
1143         KKASSERT(MAXFIDSZ >= 16);
1144         ip = VTOI(vp);
1145         fhp->fid_len = offsetof(struct fid, fid_data[16]);
1146         fhp->fid_ext = ip->obj_localization >> 16;
1147         bcopy(&ip->obj_id, fhp->fid_data + 0, sizeof(ip->obj_id));
1148         bcopy(&ip->obj_asof, fhp->fid_data + 8, sizeof(ip->obj_asof));
1149         return(0);
1150 }
1151
1152
1153 /*
1154  * Convert a file handle back to a vnode.
1155  *
1156  * Use rootvp to enforce PFS isolation when a PFS is exported via a
1157  * null mount.
1158  */
1159 static int
1160 hammer_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
1161                   struct fid *fhp, struct vnode **vpp)
1162 {
1163         hammer_mount_t hmp = (void *)mp->mnt_data;
1164         struct hammer_transaction trans;
1165         struct hammer_inode *ip;
1166         struct hammer_inode_info info;
1167         int error;
1168         u_int32_t localization;
1169
1170         bcopy(fhp->fid_data + 0, &info.obj_id, sizeof(info.obj_id));
1171         bcopy(fhp->fid_data + 8, &info.obj_asof, sizeof(info.obj_asof));
1172         if (rootvp)
1173                 localization = VTOI(rootvp)->obj_localization;
1174         else
1175                 localization = (u_int32_t)fhp->fid_ext << 16;
1176
1177         lwkt_gettoken(&hmp->fs_token);
1178         hammer_simple_transaction(&trans, hmp);
1179
1180         /*
1181          * Get/allocate the hammer_inode structure.  The structure must be
1182          * unlocked while we manipulate the related vnode to avoid a
1183          * deadlock.
1184          */
1185         ip = hammer_get_inode(&trans, NULL, info.obj_id,
1186                               info.obj_asof, localization, 0, &error);
1187         if (ip) {
1188                 error = hammer_get_vnode(ip, vpp);
1189                 hammer_rel_inode(ip, 0);
1190         } else {
1191                 *vpp = NULL;
1192         }
1193         hammer_done_transaction(&trans);
1194         lwkt_reltoken(&hmp->fs_token);
1195         return (error);
1196 }
1197
1198 static int
1199 hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
1200                     int *exflagsp, struct ucred **credanonp)
1201 {
1202         hammer_mount_t hmp = (void *)mp->mnt_data;
1203         struct netcred *np;
1204         int error;
1205
1206         lwkt_gettoken(&hmp->fs_token);
1207         np = vfs_export_lookup(mp, &hmp->export, nam);
1208         if (np) {
1209                 *exflagsp = np->netc_exflags;
1210                 *credanonp = &np->netc_anon;
1211                 error = 0;
1212         } else {
1213                 error = EACCES;
1214         }
1215         lwkt_reltoken(&hmp->fs_token);
1216         return (error);
1217
1218 }
1219
1220 int
1221 hammer_vfs_export(struct mount *mp, int op, const struct export_args *export)
1222 {
1223         hammer_mount_t hmp = (void *)mp->mnt_data;
1224         int error;
1225
1226         lwkt_gettoken(&hmp->fs_token);
1227
1228         switch(op) {
1229         case MOUNTCTL_SET_EXPORT:
1230                 error = vfs_export(mp, &hmp->export, export);
1231                 break;
1232         default:
1233                 error = EOPNOTSUPP;
1234                 break;
1235         }
1236         lwkt_reltoken(&hmp->fs_token);
1237
1238         return(error);
1239 }
1240