2 * Copyright (c) 2007-2008 The DragonFly Project. All rights reserved.
4 * This code is derived from software contributed to The DragonFly Project
5 * by Matthew Dillon <dillon@backplane.com>
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
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
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.
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
34 * $DragonFly: src/sys/vfs/hammer/hammer_vfsops.c,v 1.74 2008/11/13 02:18:43 dillon Exp $
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>
51 * NOTE! Global statistics may not be MPSAFE so HAMMER never uses them
54 int hammer_supported_version = HAMMER_VOL_VERSION_DEFAULT;
56 int hammer_debug_general;
57 int hammer_debug_debug = 1; /* medium-error panics */
58 int hammer_debug_inode;
59 int hammer_debug_locks;
60 int hammer_debug_btree;
62 int hammer_debug_recover; /* -1 will disable, +1 will force */
63 int hammer_debug_recover_faults;
64 int hammer_debug_critical; /* non-zero enter debugger on error */
65 int hammer_cluster_enable = 1; /* enable read clustering by default */
66 int hammer_count_fsyncs;
67 int hammer_count_inodes;
68 int hammer_count_iqueued;
69 int hammer_count_reclaiming;
70 int hammer_count_records;
71 int hammer_count_record_datas;
72 int hammer_count_volumes;
73 int hammer_count_buffers;
74 int hammer_count_nodes;
75 int64_t hammer_count_extra_space_used;
76 int64_t hammer_stats_btree_lookups;
77 int64_t hammer_stats_btree_searches;
78 int64_t hammer_stats_btree_inserts;
79 int64_t hammer_stats_btree_deletes;
80 int64_t hammer_stats_btree_elements;
81 int64_t hammer_stats_btree_splits;
82 int64_t hammer_stats_btree_iterations;
83 int64_t hammer_stats_btree_root_iterations;
84 int64_t hammer_stats_record_iterations;
86 int64_t hammer_stats_file_read;
87 int64_t hammer_stats_file_write;
88 int64_t hammer_stats_file_iopsr;
89 int64_t hammer_stats_file_iopsw;
90 int64_t hammer_stats_disk_read;
91 int64_t hammer_stats_disk_write;
92 int64_t hammer_stats_inode_flushes;
93 int64_t hammer_stats_commits;
94 int64_t hammer_stats_undo;
95 int64_t hammer_stats_redo;
97 int hammer_count_dirtybufspace; /* global */
98 int hammer_count_refedbufs; /* global */
99 int hammer_count_reservations;
100 int hammer_count_io_running_read;
101 int hammer_count_io_running_write;
102 int hammer_count_io_locked;
103 int hammer_limit_dirtybufspace; /* per-mount */
104 int hammer_limit_running_io; /* per-mount */
105 int hammer_limit_recs; /* as a whole XXX */
106 int hammer_limit_inode_recs = 1024; /* per inode */
107 int hammer_limit_reclaim = HAMMER_RECLAIM_WAIT;
108 int hammer_limit_redo = 4096 * 1024; /* per inode */
109 int hammer_autoflush = 2000; /* auto flush */
110 int hammer_bio_count;
111 int hammer_verify_zone;
112 int hammer_verify_data = 1;
113 int hammer_write_mode;
114 int hammer_yield_check = 16;
115 int hammer_fsync_mode = 3;
116 int64_t hammer_contention_count;
117 int64_t hammer_zone_limit;
119 SYSCTL_NODE(_vfs, OID_AUTO, hammer, CTLFLAG_RW, 0, "HAMMER filesystem");
120 SYSCTL_INT(_vfs_hammer, OID_AUTO, supported_version, CTLFLAG_RD,
121 &hammer_supported_version, 0, "");
122 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_general, CTLFLAG_RW,
123 &hammer_debug_general, 0, "");
124 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_io, CTLFLAG_RW,
125 &hammer_debug_io, 0, "");
126 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_debug, CTLFLAG_RW,
127 &hammer_debug_debug, 0, "");
128 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_inode, CTLFLAG_RW,
129 &hammer_debug_inode, 0, "");
130 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_locks, CTLFLAG_RW,
131 &hammer_debug_locks, 0, "");
132 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_btree, CTLFLAG_RW,
133 &hammer_debug_btree, 0, "");
134 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_tid, CTLFLAG_RW,
135 &hammer_debug_tid, 0, "");
136 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover, CTLFLAG_RW,
137 &hammer_debug_recover, 0, "");
138 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover_faults, CTLFLAG_RW,
139 &hammer_debug_recover_faults, 0, "");
140 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_critical, CTLFLAG_RW,
141 &hammer_debug_critical, 0, "");
142 SYSCTL_INT(_vfs_hammer, OID_AUTO, cluster_enable, CTLFLAG_RW,
143 &hammer_cluster_enable, 0, "");
145 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_dirtybufspace, CTLFLAG_RW,
146 &hammer_limit_dirtybufspace, 0, "");
147 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_running_io, CTLFLAG_RW,
148 &hammer_limit_running_io, 0, "");
149 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_recs, CTLFLAG_RW,
150 &hammer_limit_recs, 0, "");
151 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_inode_recs, CTLFLAG_RW,
152 &hammer_limit_inode_recs, 0, "");
153 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_reclaim, CTLFLAG_RW,
154 &hammer_limit_reclaim, 0, "");
155 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_redo, CTLFLAG_RW,
156 &hammer_limit_redo, 0, "");
158 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_fsyncs, CTLFLAG_RD,
159 &hammer_count_fsyncs, 0, "");
160 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_inodes, CTLFLAG_RD,
161 &hammer_count_inodes, 0, "");
162 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_iqueued, CTLFLAG_RD,
163 &hammer_count_iqueued, 0, "");
164 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reclaiming, CTLFLAG_RD,
165 &hammer_count_reclaiming, 0, "");
166 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_records, CTLFLAG_RD,
167 &hammer_count_records, 0, "");
168 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_record_datas, CTLFLAG_RD,
169 &hammer_count_record_datas, 0, "");
170 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_volumes, CTLFLAG_RD,
171 &hammer_count_volumes, 0, "");
172 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_buffers, CTLFLAG_RD,
173 &hammer_count_buffers, 0, "");
174 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_nodes, CTLFLAG_RD,
175 &hammer_count_nodes, 0, "");
176 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, count_extra_space_used, CTLFLAG_RD,
177 &hammer_count_extra_space_used, 0, "");
179 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_searches, CTLFLAG_RD,
180 &hammer_stats_btree_searches, 0, "");
181 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_lookups, CTLFLAG_RD,
182 &hammer_stats_btree_lookups, 0, "");
183 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_inserts, CTLFLAG_RD,
184 &hammer_stats_btree_inserts, 0, "");
185 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_deletes, CTLFLAG_RD,
186 &hammer_stats_btree_deletes, 0, "");
187 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_elements, CTLFLAG_RD,
188 &hammer_stats_btree_elements, 0, "");
189 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_splits, CTLFLAG_RD,
190 &hammer_stats_btree_splits, 0, "");
191 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_iterations, CTLFLAG_RD,
192 &hammer_stats_btree_iterations, 0, "");
193 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_root_iterations, CTLFLAG_RD,
194 &hammer_stats_btree_root_iterations, 0, "");
195 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_record_iterations, CTLFLAG_RD,
196 &hammer_stats_record_iterations, 0, "");
198 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_read, CTLFLAG_RD,
199 &hammer_stats_file_read, 0, "");
200 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_write, CTLFLAG_RD,
201 &hammer_stats_file_write, 0, "");
202 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_iopsr, CTLFLAG_RD,
203 &hammer_stats_file_iopsr, 0, "");
204 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_iopsw, CTLFLAG_RD,
205 &hammer_stats_file_iopsw, 0, "");
206 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_disk_read, CTLFLAG_RD,
207 &hammer_stats_disk_read, 0, "");
208 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_disk_write, CTLFLAG_RD,
209 &hammer_stats_disk_write, 0, "");
210 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_inode_flushes, CTLFLAG_RD,
211 &hammer_stats_inode_flushes, 0, "");
212 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_commits, CTLFLAG_RD,
213 &hammer_stats_commits, 0, "");
214 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_undo, CTLFLAG_RD,
215 &hammer_stats_undo, 0, "");
216 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_redo, CTLFLAG_RD,
217 &hammer_stats_redo, 0, "");
219 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_dirtybufspace, CTLFLAG_RD,
220 &hammer_count_dirtybufspace, 0, "");
221 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_refedbufs, CTLFLAG_RD,
222 &hammer_count_refedbufs, 0, "");
223 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reservations, CTLFLAG_RD,
224 &hammer_count_reservations, 0, "");
225 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_running_read, CTLFLAG_RD,
226 &hammer_count_io_running_read, 0, "");
227 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_locked, CTLFLAG_RD,
228 &hammer_count_io_locked, 0, "");
229 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_running_write, CTLFLAG_RD,
230 &hammer_count_io_running_write, 0, "");
231 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, zone_limit, CTLFLAG_RW,
232 &hammer_zone_limit, 0, "");
233 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, contention_count, CTLFLAG_RW,
234 &hammer_contention_count, 0, "");
235 SYSCTL_INT(_vfs_hammer, OID_AUTO, autoflush, CTLFLAG_RW,
236 &hammer_autoflush, 0, "");
237 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_zone, CTLFLAG_RW,
238 &hammer_verify_zone, 0, "");
239 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_data, CTLFLAG_RW,
240 &hammer_verify_data, 0, "");
241 SYSCTL_INT(_vfs_hammer, OID_AUTO, write_mode, CTLFLAG_RW,
242 &hammer_write_mode, 0, "");
243 SYSCTL_INT(_vfs_hammer, OID_AUTO, yield_check, CTLFLAG_RW,
244 &hammer_yield_check, 0, "");
245 SYSCTL_INT(_vfs_hammer, OID_AUTO, fsync_mode, CTLFLAG_RW,
246 &hammer_fsync_mode, 0, "");
248 KTR_INFO_MASTER(hammer);
253 static void hammer_free_hmp(struct mount *mp);
255 static int hammer_vfs_mount(struct mount *mp, char *path, caddr_t data,
257 static int hammer_vfs_unmount(struct mount *mp, int mntflags);
258 static int hammer_vfs_root(struct mount *mp, struct vnode **vpp);
259 static int hammer_vfs_statfs(struct mount *mp, struct statfs *sbp,
261 static int hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp,
263 static int hammer_vfs_sync(struct mount *mp, int waitfor);
264 static int hammer_vfs_vget(struct mount *mp, struct vnode *dvp,
265 ino_t ino, struct vnode **vpp);
266 static int hammer_vfs_init(struct vfsconf *conf);
267 static int hammer_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
268 struct fid *fhp, struct vnode **vpp);
269 static int hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp);
270 static int hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
271 int *exflagsp, struct ucred **credanonp);
274 static struct vfsops hammer_vfsops = {
275 .vfs_mount = hammer_vfs_mount,
276 .vfs_unmount = hammer_vfs_unmount,
277 .vfs_root = hammer_vfs_root,
278 .vfs_statfs = hammer_vfs_statfs,
279 .vfs_statvfs = hammer_vfs_statvfs,
280 .vfs_sync = hammer_vfs_sync,
281 .vfs_vget = hammer_vfs_vget,
282 .vfs_init = hammer_vfs_init,
283 .vfs_vptofh = hammer_vfs_vptofh,
284 .vfs_fhtovp = hammer_vfs_fhtovp,
285 .vfs_checkexp = hammer_vfs_checkexp
288 MALLOC_DEFINE(M_HAMMER, "HAMMER-mount", "");
290 VFS_SET(hammer_vfsops, hammer, 0);
291 MODULE_VERSION(hammer, 1);
294 hammer_vfs_init(struct vfsconf *conf)
298 if (hammer_limit_recs == 0) {
299 hammer_limit_recs = nbuf * 25;
300 n = kmalloc_limit(M_HAMMER) / 512;
301 if (hammer_limit_recs > n)
302 hammer_limit_recs = n;
304 if (hammer_limit_dirtybufspace == 0) {
305 hammer_limit_dirtybufspace = hidirtybufspace / 2;
306 if (hammer_limit_dirtybufspace < 100)
307 hammer_limit_dirtybufspace = 100;
311 * Set reasonable limits to maintain an I/O pipeline. This is
312 * used by the flush code which explicitly initiates I/O, and
315 * The system-driven buffer cache uses vfs.lorunningspace and
316 * vfs.hirunningspace globally.
318 if (hammer_limit_running_io == 0)
319 hammer_limit_running_io = hammer_limit_dirtybufspace;
320 if (hammer_limit_running_io > 10 * 1024 * 1024)
321 hammer_limit_running_io = 10 * 1024 * 1024;
326 hammer_vfs_mount(struct mount *mp, char *mntpt, caddr_t data,
329 struct hammer_mount_info info;
331 hammer_volume_t rootvol;
332 struct vnode *rootvp;
333 struct vnode *devvp = NULL;
334 const char *upath; /* volume name in userspace */
335 char *path; /* volume name in system space */
339 char *next_volume_ptr = NULL;
342 * Accept hammer_mount_info. mntpt is NULL for root mounts at boot.
345 bzero(&info, sizeof(info));
350 next_volume_ptr = mp->mnt_stat.f_mntfromname;
352 /* Count number of volumes separated by ':' */
353 for (char *p = next_volume_ptr; *p != '\0'; ++p) {
359 mp->mnt_flag &= ~MNT_RDONLY; /* mount R/W */
361 if ((error = copyin(data, &info, sizeof(info))) != 0)
366 * updating or new mount
368 if (mp->mnt_flag & MNT_UPDATE) {
369 hmp = (void *)mp->mnt_data;
370 KKASSERT(hmp != NULL);
372 if (info.nvolumes <= 0 || info.nvolumes >= 32768)
378 * master-id validation. The master id may not be changed by a
381 if (info.hflags & HMNT_MASTERID) {
382 if (hmp && hmp->master_id != info.master_id) {
383 kprintf("hammer: cannot change master id "
384 "with mount update\n");
387 master_id = info.master_id;
388 if (master_id < -1 || master_id >= HAMMER_MAX_MASTERS)
392 master_id = hmp->master_id;
398 * Interal mount data structure
401 hmp = kmalloc(sizeof(*hmp), M_HAMMER, M_WAITOK | M_ZERO);
402 mp->mnt_data = (qaddr_t)hmp;
404 /*TAILQ_INIT(&hmp->recycle_list);*/
407 * Make sure kmalloc type limits are set appropriately.
409 * Our inode kmalloc group is sized based on maxvnodes
410 * (controlled by the system, not us).
412 kmalloc_create(&hmp->m_misc, "HAMMER-others");
413 kmalloc_create(&hmp->m_inodes, "HAMMER-inodes");
415 kmalloc_raise_limit(hmp->m_inodes, 0); /* unlimited */
417 hmp->root_btree_beg.localization = 0x00000000U;
418 hmp->root_btree_beg.obj_id = -0x8000000000000000LL;
419 hmp->root_btree_beg.key = -0x8000000000000000LL;
420 hmp->root_btree_beg.create_tid = 1;
421 hmp->root_btree_beg.delete_tid = 1;
422 hmp->root_btree_beg.rec_type = 0;
423 hmp->root_btree_beg.obj_type = 0;
425 hmp->root_btree_end.localization = 0xFFFFFFFFU;
426 hmp->root_btree_end.obj_id = 0x7FFFFFFFFFFFFFFFLL;
427 hmp->root_btree_end.key = 0x7FFFFFFFFFFFFFFFLL;
428 hmp->root_btree_end.create_tid = 0xFFFFFFFFFFFFFFFFULL;
429 hmp->root_btree_end.delete_tid = 0; /* special case */
430 hmp->root_btree_end.rec_type = 0xFFFFU;
431 hmp->root_btree_end.obj_type = 0;
433 hmp->krate.freq = 1; /* maximum reporting rate (hz) */
434 hmp->krate.count = -16; /* initial burst */
436 hmp->sync_lock.refs = 1;
437 hmp->free_lock.refs = 1;
438 hmp->undo_lock.refs = 1;
439 hmp->blkmap_lock.refs = 1;
440 hmp->snapshot_lock.refs = 1;
441 hmp->volume_lock.refs = 1;
443 TAILQ_INIT(&hmp->delay_list);
444 TAILQ_INIT(&hmp->flush_group_list);
445 TAILQ_INIT(&hmp->objid_cache_list);
446 TAILQ_INIT(&hmp->undo_lru_list);
447 TAILQ_INIT(&hmp->reclaim_list);
449 hmp->hflags &= ~HMNT_USERFLAGS;
450 hmp->hflags |= info.hflags & HMNT_USERFLAGS;
452 hmp->master_id = master_id;
455 mp->mnt_flag |= MNT_RDONLY;
456 hmp->asof = info.asof;
458 hmp->asof = HAMMER_MAX_TID;
461 hmp->volume_to_remove = -1;
464 * Re-open read-write if originally read-only, or vise-versa.
466 * When going from read-only to read-write execute the stage2
467 * recovery if it has not already been run.
469 if (mp->mnt_flag & MNT_UPDATE) {
471 if (hmp->ronly && (mp->mnt_kern_flag & MNTK_WANTRDWR)) {
472 kprintf("HAMMER read-only -> read-write\n");
474 RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
475 hammer_adjust_volume_mode, NULL);
476 rootvol = hammer_get_root_volume(hmp, &error);
478 hammer_recover_flush_buffers(hmp, rootvol, 1);
479 error = hammer_recover_stage2(hmp, rootvol);
480 bcopy(rootvol->ondisk->vol0_blockmap,
482 sizeof(hmp->blockmap));
483 hammer_rel_volume(rootvol, 0);
485 RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
486 hammer_reload_inode, NULL);
487 /* kernel clears MNT_RDONLY */
488 } else if (hmp->ronly == 0 && (mp->mnt_flag & MNT_RDONLY)) {
489 kprintf("HAMMER read-write -> read-only\n");
490 hmp->ronly = 1; /* messy */
491 RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
492 hammer_reload_inode, NULL);
494 hammer_flusher_sync(hmp);
495 hammer_flusher_sync(hmp);
496 hammer_flusher_sync(hmp);
498 RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
499 hammer_adjust_volume_mode, NULL);
504 RB_INIT(&hmp->rb_vols_root);
505 RB_INIT(&hmp->rb_inos_root);
506 RB_INIT(&hmp->rb_redo_root);
507 RB_INIT(&hmp->rb_nods_root);
508 RB_INIT(&hmp->rb_undo_root);
509 RB_INIT(&hmp->rb_resv_root);
510 RB_INIT(&hmp->rb_bufs_root);
511 RB_INIT(&hmp->rb_pfsm_root);
513 hmp->ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
515 TAILQ_INIT(&hmp->volu_list);
516 TAILQ_INIT(&hmp->undo_list);
517 TAILQ_INIT(&hmp->data_list);
518 TAILQ_INIT(&hmp->meta_list);
519 TAILQ_INIT(&hmp->lose_list);
520 TAILQ_INIT(&hmp->iorun_list);
525 path = objcache_get(namei_oc, M_WAITOK);
527 for (i = 0; i < info.nvolumes; ++i) {
532 KKASSERT(next_volume_ptr != NULL);
534 if (*next_volume_ptr != '/') {
536 strcpy(path, "/dev/");
539 for (k = strlen(path); k < MAXPATHLEN-1; ++k) {
540 if (*next_volume_ptr == '\0') {
542 } else if (*next_volume_ptr == ':') {
546 path[k] = *next_volume_ptr;
553 cdev_t dev = kgetdiskbyname(path);
554 error = bdevvp(dev, &devvp);
556 kprintf("hammer_mountroot: can't find devvp\n");
559 error = copyin(&info.volumes[i], &upath,
562 error = copyinstr(upath, path,
566 error = hammer_install_volume(hmp, path, devvp);
570 objcache_put(namei_oc, path);
573 * Make sure we found a root volume
575 if (error == 0 && hmp->rootvol == NULL) {
576 kprintf("hammer_mount: No root volume found!\n");
581 * Check that all required volumes are available
583 if (error == 0 && hammer_mountcheck_volumes(hmp)) {
584 kprintf("hammer_mount: Missing volumes, cannot mount!\n");
594 * No errors, setup enough of the mount point so we can lookup the
597 mp->mnt_iosize_max = MAXPHYS;
598 mp->mnt_kern_flag |= MNTK_FSMID;
601 * MPSAFE code. Note that VOPs and VFSops which are not MPSAFE
602 * will acquire a per-mount token prior to entry and release it
603 * on return, so even if we do not specify it we no longer get
604 * the BGL regardlless of how we are flagged.
606 mp->mnt_kern_flag |= MNTK_RD_MPSAFE | MNTK_GA_MPSAFE |
610 * note: f_iosize is used by vnode_pager_haspage() when constructing
613 mp->mnt_stat.f_iosize = HAMMER_BUFSIZE;
614 mp->mnt_stat.f_bsize = HAMMER_BUFSIZE;
616 mp->mnt_vstat.f_frsize = HAMMER_BUFSIZE;
617 mp->mnt_vstat.f_bsize = HAMMER_BUFSIZE;
619 mp->mnt_maxsymlinklen = 255;
620 mp->mnt_flag |= MNT_LOCAL;
622 vfs_add_vnodeops(mp, &hammer_vnode_vops, &mp->mnt_vn_norm_ops);
623 vfs_add_vnodeops(mp, &hammer_spec_vops, &mp->mnt_vn_spec_ops);
624 vfs_add_vnodeops(mp, &hammer_fifo_vops, &mp->mnt_vn_fifo_ops);
627 * The root volume's ondisk pointer is only valid if we hold a
630 rootvol = hammer_get_root_volume(hmp, &error);
635 * Perform any necessary UNDO operations. The recovery code does
636 * call hammer_undo_lookup() so we have to pre-cache the blockmap,
637 * and then re-copy it again after recovery is complete.
639 * If this is a read-only mount the UNDO information is retained
640 * in memory in the form of dirty buffer cache buffers, and not
641 * written back to the media.
643 bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
644 sizeof(hmp->blockmap));
647 * Check filesystem version
649 hmp->version = rootvol->ondisk->vol_version;
650 if (hmp->version < HAMMER_VOL_VERSION_MIN ||
651 hmp->version > HAMMER_VOL_VERSION_MAX) {
652 kprintf("HAMMER: mount unsupported fs version %d\n",
659 * The undo_rec_limit limits the size of flush groups to avoid
660 * blowing out the UNDO FIFO. This calculation is typically in
661 * the tens of thousands and is designed primarily when small
662 * HAMMER filesystems are created.
664 hmp->undo_rec_limit = hammer_undo_max(hmp) / 8192 + 100;
665 if (hammer_debug_general & 0x0001)
666 kprintf("HAMMER: undo_rec_limit %d\n", hmp->undo_rec_limit);
669 * NOTE: Recover stage1 not only handles meta-data recovery, it
670 * also sets hmp->undo_seqno for HAMMER VERSION 4+ filesystems.
672 error = hammer_recover_stage1(hmp, rootvol);
674 kprintf("Failed to recover HAMMER filesystem on mount\n");
679 * Finish setup now that we have a good root volume.
681 * The top 16 bits of fsid.val[1] is a pfs id.
683 ksnprintf(mp->mnt_stat.f_mntfromname,
684 sizeof(mp->mnt_stat.f_mntfromname), "%s",
685 rootvol->ondisk->vol_name);
686 mp->mnt_stat.f_fsid.val[0] =
687 crc32((char *)&rootvol->ondisk->vol_fsid + 0, 8);
688 mp->mnt_stat.f_fsid.val[1] =
689 crc32((char *)&rootvol->ondisk->vol_fsid + 8, 8);
690 mp->mnt_stat.f_fsid.val[1] &= 0x0000FFFF;
692 mp->mnt_vstat.f_fsid_uuid = rootvol->ondisk->vol_fsid;
693 mp->mnt_vstat.f_fsid = crc32(&mp->mnt_vstat.f_fsid_uuid,
694 sizeof(mp->mnt_vstat.f_fsid_uuid));
697 * Certain often-modified fields in the root volume are cached in
698 * the hammer_mount structure so we do not have to generate lots
699 * of little UNDO structures for them.
701 * Recopy after recovery. This also has the side effect of
702 * setting our cached undo FIFO's first_offset, which serves to
703 * placemark the FIFO start for the NEXT flush cycle while the
704 * on-disk first_offset represents the LAST flush cycle.
706 hmp->next_tid = rootvol->ondisk->vol0_next_tid;
707 hmp->flush_tid1 = hmp->next_tid;
708 hmp->flush_tid2 = hmp->next_tid;
709 bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
710 sizeof(hmp->blockmap));
711 hmp->copy_stat_freebigblocks = rootvol->ondisk->vol0_stat_freebigblocks;
713 hammer_flusher_create(hmp);
716 * Locate the root directory using the root cluster's B-Tree as a
717 * starting point. The root directory uses an obj_id of 1.
719 * FUTURE: Leave the root directory cached referenced but unlocked
720 * in hmp->rootvp (need to flush it on unmount).
722 error = hammer_vfs_vget(mp, NULL, 1, &rootvp);
726 /*vn_unlock(hmp->rootvp);*/
728 error = hammer_recover_stage2(hmp, rootvol);
731 * If the stage2 recovery fails be sure to clean out all cached
732 * vnodes before throwing away the mount structure or bad things
739 hammer_rel_volume(rootvol, 0);
742 * Cleanup and return.
750 hammer_vfs_unmount(struct mount *mp, int mntflags)
753 struct hammer_mount *hmp = (void *)mp->mnt_data;
759 * Clean out the vnodes
762 if (mntflags & MNT_FORCE)
764 if ((error = vflush(mp, 0, flags)) != 0)
768 * Clean up the internal mount structure and related entities. This
776 * Clean up the internal mount structure and disassociate it from the mount.
777 * This may issue I/O.
780 hammer_free_hmp(struct mount *mp)
782 struct hammer_mount *hmp = (void *)mp->mnt_data;
783 hammer_flush_group_t flg;
788 * Flush anything dirty. This won't even run if the
789 * filesystem errored-out.
792 while (hammer_flusher_haswork(hmp)) {
793 hammer_flusher_sync(hmp);
797 kprintf("HAMMER: umount flushing.");
800 tsleep(&dummy, 0, "hmrufl", hz);
803 kprintf("giving up\n");
807 if (count >= 5 && count < 30)
811 * If the mount had a critical error we have to destroy any
812 * remaining inodes before we can finish cleaning up the flusher.
814 if (hmp->flags & HAMMER_MOUNT_CRITICAL_ERROR) {
815 RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
816 hammer_destroy_inode_callback, NULL);
820 * There shouldn't be any inodes left now and any left over
821 * flush groups should now be empty.
823 KKASSERT(RB_EMPTY(&hmp->rb_inos_root));
824 while ((flg = TAILQ_FIRST(&hmp->flush_group_list)) != NULL) {
825 TAILQ_REMOVE(&hmp->flush_group_list, flg, flush_entry);
826 KKASSERT(RB_EMPTY(&flg->flush_tree));
828 kprintf("HAMMER: Warning, flush_group %p was "
829 "not empty on umount!\n", flg);
831 kfree(flg, hmp->m_misc);
835 * We can finally destroy the flusher
837 hammer_flusher_destroy(hmp);
840 * We may have held recovered buffers due to a read-only mount.
841 * These must be discarded.
844 hammer_recover_flush_buffers(hmp, NULL, -1);
847 * Unload buffers and then volumes
849 RB_SCAN(hammer_buf_rb_tree, &hmp->rb_bufs_root, NULL,
850 hammer_unload_buffer, NULL);
851 RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
852 hammer_unload_volume, NULL);
855 mp->mnt_flag &= ~MNT_LOCAL;
857 hammer_destroy_objid_cache(hmp);
858 kmalloc_destroy(&hmp->m_misc);
859 kmalloc_destroy(&hmp->m_inodes);
860 kfree(hmp, M_HAMMER);
864 * Report critical errors. ip may be NULL.
867 hammer_critical_error(hammer_mount_t hmp, hammer_inode_t ip,
868 int error, const char *msg)
870 hmp->flags |= HAMMER_MOUNT_CRITICAL_ERROR;
872 krateprintf(&hmp->krate,
873 "HAMMER(%s): Critical error inode=%jd error=%d %s\n",
874 hmp->mp->mnt_stat.f_mntfromname,
875 (intmax_t)(ip ? ip->obj_id : -1),
878 if (hmp->ronly == 0) {
879 hmp->ronly = 2; /* special errored read-only mode */
880 hmp->mp->mnt_flag |= MNT_RDONLY;
881 kprintf("HAMMER(%s): Forcing read-only mode\n",
882 hmp->mp->mnt_stat.f_mntfromname);
885 if (hammer_debug_critical)
886 Debugger("Entering debugger");
891 * Obtain a vnode for the specified inode number. An exclusively locked
895 hammer_vfs_vget(struct mount *mp, struct vnode *dvp,
896 ino_t ino, struct vnode **vpp)
898 struct hammer_transaction trans;
899 struct hammer_mount *hmp = (void *)mp->mnt_data;
900 struct hammer_inode *ip;
902 u_int32_t localization;
904 hammer_simple_transaction(&trans, hmp);
907 * If a directory vnode is supplied (mainly NFS) then we can acquire
908 * the PFS domain from it. Otherwise we would only be able to vget
909 * inodes in the root PFS.
912 localization = HAMMER_DEF_LOCALIZATION +
913 VTOI(dvp)->obj_localization;
915 localization = HAMMER_DEF_LOCALIZATION;
919 * Lookup the requested HAMMER inode. The structure must be
920 * left unlocked while we manipulate the related vnode to avoid
923 ip = hammer_get_inode(&trans, NULL, ino,
924 hmp->asof, localization,
928 hammer_done_transaction(&trans);
931 error = hammer_get_vnode(ip, vpp);
932 hammer_rel_inode(ip, 0);
933 hammer_done_transaction(&trans);
938 * Return the root vnode for the filesystem.
940 * HAMMER stores the root vnode in the hammer_mount structure so
941 * getting it is easy.
944 hammer_vfs_root(struct mount *mp, struct vnode **vpp)
947 struct hammer_mount *hmp = (void *)mp->mnt_data;
951 error = hammer_vfs_vget(mp, NULL, 1, vpp);
956 hammer_vfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
958 struct hammer_mount *hmp = (void *)mp->mnt_data;
959 hammer_volume_t volume;
960 hammer_volume_ondisk_t ondisk;
965 volume = hammer_get_root_volume(hmp, &error);
968 ondisk = volume->ondisk;
973 _hammer_checkspace(hmp, HAMMER_CHKSPC_WRITE, &breserved);
974 mp->mnt_stat.f_files = ondisk->vol0_stat_inodes;
975 bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE;
976 hammer_rel_volume(volume, 0);
978 mp->mnt_stat.f_bfree = (bfree - breserved) / HAMMER_BUFSIZE;
979 mp->mnt_stat.f_bavail = mp->mnt_stat.f_bfree;
980 if (mp->mnt_stat.f_files < 0)
981 mp->mnt_stat.f_files = 0;
988 hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred)
990 struct hammer_mount *hmp = (void *)mp->mnt_data;
991 hammer_volume_t volume;
992 hammer_volume_ondisk_t ondisk;
997 volume = hammer_get_root_volume(hmp, &error);
1000 ondisk = volume->ondisk;
1005 _hammer_checkspace(hmp, HAMMER_CHKSPC_WRITE, &breserved);
1006 mp->mnt_vstat.f_files = ondisk->vol0_stat_inodes;
1007 bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE;
1008 hammer_rel_volume(volume, 0);
1010 mp->mnt_vstat.f_bfree = (bfree - breserved) / HAMMER_BUFSIZE;
1011 mp->mnt_vstat.f_bavail = mp->mnt_vstat.f_bfree;
1012 if (mp->mnt_vstat.f_files < 0)
1013 mp->mnt_vstat.f_files = 0;
1014 *sbp = mp->mnt_vstat;
1019 * Sync the filesystem. Currently we have to run it twice, the second
1020 * one will advance the undo start index to the end index, so if a crash
1021 * occurs no undos will be run on mount.
1023 * We do not sync the filesystem if we are called from a panic. If we did
1024 * we might end up blowing up a sync that was already in progress.
1027 hammer_vfs_sync(struct mount *mp, int waitfor)
1029 struct hammer_mount *hmp = (void *)mp->mnt_data;
1032 if (panicstr == NULL) {
1033 error = hammer_sync_hmp(hmp, waitfor);
1041 * Convert a vnode to a file handle.
1044 hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp)
1048 KKASSERT(MAXFIDSZ >= 16);
1050 fhp->fid_len = offsetof(struct fid, fid_data[16]);
1051 fhp->fid_ext = ip->obj_localization >> 16;
1052 bcopy(&ip->obj_id, fhp->fid_data + 0, sizeof(ip->obj_id));
1053 bcopy(&ip->obj_asof, fhp->fid_data + 8, sizeof(ip->obj_asof));
1059 * Convert a file handle back to a vnode.
1061 * Use rootvp to enforce PFS isolation when a PFS is exported via a
1065 hammer_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
1066 struct fid *fhp, struct vnode **vpp)
1068 struct hammer_transaction trans;
1069 struct hammer_inode *ip;
1070 struct hammer_inode_info info;
1072 u_int32_t localization;
1074 bcopy(fhp->fid_data + 0, &info.obj_id, sizeof(info.obj_id));
1075 bcopy(fhp->fid_data + 8, &info.obj_asof, sizeof(info.obj_asof));
1077 localization = VTOI(rootvp)->obj_localization;
1079 localization = (u_int32_t)fhp->fid_ext << 16;
1081 hammer_simple_transaction(&trans, (void *)mp->mnt_data);
1084 * Get/allocate the hammer_inode structure. The structure must be
1085 * unlocked while we manipulate the related vnode to avoid a
1088 ip = hammer_get_inode(&trans, NULL, info.obj_id,
1089 info.obj_asof, localization, 0, &error);
1091 error = hammer_get_vnode(ip, vpp);
1092 hammer_rel_inode(ip, 0);
1096 hammer_done_transaction(&trans);
1101 hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
1102 int *exflagsp, struct ucred **credanonp)
1104 hammer_mount_t hmp = (void *)mp->mnt_data;
1108 np = vfs_export_lookup(mp, &hmp->export, nam);
1110 *exflagsp = np->netc_exflags;
1111 *credanonp = &np->netc_anon;
1121 hammer_vfs_export(struct mount *mp, int op, const struct export_args *export)
1123 hammer_mount_t hmp = (void *)mp->mnt_data;
1127 case MOUNTCTL_SET_EXPORT:
1128 error = vfs_export(mp, &hmp->export, export);