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
35 #include <sys/mountctl.h>
40 * NOTE! Global statistics may not be MPSAFE so HAMMER never uses them
43 int hammer_supported_version = HAMMER_VOL_VERSION_DEFAULT;
45 int hammer_debug_general;
46 int hammer_debug_inode;
47 int hammer_debug_locks;
48 int hammer_debug_btree;
50 int hammer_debug_recover; /* -1 will disable, +1 will force */
51 int hammer_debug_critical; /* non-zero enter debugger on error */
52 int hammer_cluster_enable = 1; /* enable read clustering by default */
53 int hammer_live_dedup = 0;
54 int hammer_tdmux_ticks;
55 int hammer_count_fsyncs;
56 int hammer_count_inodes;
57 int hammer_count_iqueued;
58 int hammer_count_reclaims;
59 int hammer_count_records;
60 int hammer_count_record_datas;
61 int hammer_count_volumes;
62 int hammer_count_buffers;
63 int hammer_count_nodes;
64 int64_t hammer_count_extra_space_used;
65 int64_t hammer_stats_btree_lookups;
66 int64_t hammer_stats_btree_searches;
67 int64_t hammer_stats_btree_inserts;
68 int64_t hammer_stats_btree_deletes;
69 int64_t hammer_stats_btree_elements;
70 int64_t hammer_stats_btree_splits;
71 int64_t hammer_stats_btree_iterations;
72 int64_t hammer_stats_btree_root_iterations;
73 int64_t hammer_stats_record_iterations;
75 int64_t hammer_stats_file_read;
76 int64_t hammer_stats_file_write;
77 int64_t hammer_stats_file_iopsr;
78 int64_t hammer_stats_file_iopsw;
79 int64_t hammer_stats_disk_read;
80 int64_t hammer_stats_disk_write;
81 int64_t hammer_stats_inode_flushes;
82 int64_t hammer_stats_commits;
83 int64_t hammer_stats_undo;
84 int64_t hammer_stats_redo;
86 long hammer_count_dirtybufspace; /* global */
87 int hammer_count_refedbufs; /* global */
88 int hammer_count_reservations;
89 long hammer_count_io_running_read;
90 long hammer_count_io_running_write;
91 int hammer_count_io_locked;
92 long hammer_limit_dirtybufspace; /* per-mount */
93 int hammer_limit_recs; /* as a whole XXX */
94 int hammer_limit_inode_recs = 2048; /* per inode */
95 int hammer_limit_reclaims;
96 int hammer_live_dedup_cache_size = DEDUP_CACHE_SIZE;
97 int hammer_limit_redo = 4096 * 1024; /* per inode */
98 int hammer_autoflush = 500; /* auto flush (typ on reclaim) */
99 int hammer_verify_zone;
100 int hammer_verify_data = 1;
101 int hammer_double_buffer;
102 int hammer_btree_full_undo = 1;
103 int hammer_yield_check = 16;
104 int hammer_fsync_mode = 3;
105 int64_t hammer_contention_count;
107 int hammer_noatime = 1;
108 TUNABLE_INT("vfs.hammer.noatime", &hammer_noatime);
111 * Live dedup debug counters (sysctls are writable so that counters
112 * can be reset from userspace).
114 int64_t hammer_live_dedup_vnode_bcmps = 0;
115 int64_t hammer_live_dedup_device_bcmps = 0;
116 int64_t hammer_live_dedup_findblk_failures = 0;
117 int64_t hammer_live_dedup_bmap_saves = 0;
120 SYSCTL_NODE(_vfs, OID_AUTO, hammer, CTLFLAG_RW, 0, "HAMMER filesystem");
122 SYSCTL_INT(_vfs_hammer, OID_AUTO, supported_version, CTLFLAG_RD,
123 &hammer_supported_version, 0, "");
124 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_general, CTLFLAG_RW,
125 &hammer_debug_general, 0, "");
126 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_io, CTLFLAG_RW,
127 &hammer_debug_io, 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_critical, CTLFLAG_RW,
139 &hammer_debug_critical, 0, "");
140 SYSCTL_INT(_vfs_hammer, OID_AUTO, cluster_enable, CTLFLAG_RW,
141 &hammer_cluster_enable, 0, "");
143 * 0 - live dedup is disabled
144 * 1 - dedup cache is populated on reads only
145 * 2 - dedup cache is populated on both reads and writes
147 * LIVE_DEDUP IS DISABLED PERMANENTLY! This feature appears to cause
148 * blockmap corruption over time so we've turned it off permanently.
150 SYSCTL_INT(_vfs_hammer, OID_AUTO, live_dedup, CTLFLAG_RD,
151 &hammer_live_dedup, 0, "Enable live dedup (experimental)");
152 SYSCTL_INT(_vfs_hammer, OID_AUTO, tdmux_ticks, CTLFLAG_RW,
153 &hammer_tdmux_ticks, 0, "Hammer tdmux ticks");
155 SYSCTL_LONG(_vfs_hammer, OID_AUTO, limit_dirtybufspace, CTLFLAG_RW,
156 &hammer_limit_dirtybufspace, 0, "");
157 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_recs, CTLFLAG_RW,
158 &hammer_limit_recs, 0, "");
159 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_inode_recs, CTLFLAG_RW,
160 &hammer_limit_inode_recs, 0, "");
161 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_reclaims, CTLFLAG_RW,
162 &hammer_limit_reclaims, 0, "");
163 SYSCTL_INT(_vfs_hammer, OID_AUTO, live_dedup_cache_size, CTLFLAG_RW,
164 &hammer_live_dedup_cache_size, 0,
165 "Number of cache entries");
166 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_redo, CTLFLAG_RW,
167 &hammer_limit_redo, 0, "");
169 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_fsyncs, CTLFLAG_RD,
170 &hammer_count_fsyncs, 0, "");
171 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_inodes, CTLFLAG_RD,
172 &hammer_count_inodes, 0, "");
173 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_iqueued, CTLFLAG_RD,
174 &hammer_count_iqueued, 0, "");
175 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reclaims, CTLFLAG_RD,
176 &hammer_count_reclaims, 0, "");
177 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_records, CTLFLAG_RD,
178 &hammer_count_records, 0, "");
179 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_record_datas, CTLFLAG_RD,
180 &hammer_count_record_datas, 0, "");
181 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_volumes, CTLFLAG_RD,
182 &hammer_count_volumes, 0, "");
183 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_buffers, CTLFLAG_RD,
184 &hammer_count_buffers, 0, "");
185 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_nodes, CTLFLAG_RD,
186 &hammer_count_nodes, 0, "");
187 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, count_extra_space_used, CTLFLAG_RD,
188 &hammer_count_extra_space_used, 0, "");
190 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_searches, CTLFLAG_RD,
191 &hammer_stats_btree_searches, 0, "");
192 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_lookups, CTLFLAG_RD,
193 &hammer_stats_btree_lookups, 0, "");
194 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_inserts, CTLFLAG_RD,
195 &hammer_stats_btree_inserts, 0, "");
196 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_deletes, CTLFLAG_RD,
197 &hammer_stats_btree_deletes, 0, "");
198 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_elements, CTLFLAG_RD,
199 &hammer_stats_btree_elements, 0, "");
200 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_splits, CTLFLAG_RD,
201 &hammer_stats_btree_splits, 0, "");
202 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_iterations, CTLFLAG_RD,
203 &hammer_stats_btree_iterations, 0, "");
204 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_root_iterations, CTLFLAG_RD,
205 &hammer_stats_btree_root_iterations, 0, "");
206 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_record_iterations, CTLFLAG_RD,
207 &hammer_stats_record_iterations, 0, "");
209 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_read, CTLFLAG_RD,
210 &hammer_stats_file_read, 0, "");
211 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_write, CTLFLAG_RD,
212 &hammer_stats_file_write, 0, "");
213 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_iopsr, CTLFLAG_RD,
214 &hammer_stats_file_iopsr, 0, "");
215 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_iopsw, CTLFLAG_RD,
216 &hammer_stats_file_iopsw, 0, "");
217 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_disk_read, CTLFLAG_RD,
218 &hammer_stats_disk_read, 0, "");
219 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_disk_write, CTLFLAG_RD,
220 &hammer_stats_disk_write, 0, "");
221 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_inode_flushes, CTLFLAG_RD,
222 &hammer_stats_inode_flushes, 0, "");
223 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_commits, CTLFLAG_RD,
224 &hammer_stats_commits, 0, "");
225 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_undo, CTLFLAG_RD,
226 &hammer_stats_undo, 0, "");
227 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_redo, CTLFLAG_RD,
228 &hammer_stats_redo, 0, "");
230 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, live_dedup_vnode_bcmps, CTLFLAG_RW,
231 &hammer_live_dedup_vnode_bcmps, 0,
232 "successful vnode buffer comparisons");
233 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, live_dedup_device_bcmps, CTLFLAG_RW,
234 &hammer_live_dedup_device_bcmps, 0,
235 "successful device buffer comparisons");
236 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, live_dedup_findblk_failures, CTLFLAG_RW,
237 &hammer_live_dedup_findblk_failures, 0,
238 "block lookup failures for comparison");
239 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, live_dedup_bmap_saves, CTLFLAG_RW,
240 &hammer_live_dedup_bmap_saves, 0,
241 "useful physical block lookups");
243 SYSCTL_LONG(_vfs_hammer, OID_AUTO, count_dirtybufspace, CTLFLAG_RD,
244 &hammer_count_dirtybufspace, 0, "");
245 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_refedbufs, CTLFLAG_RD,
246 &hammer_count_refedbufs, 0, "");
247 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reservations, CTLFLAG_RD,
248 &hammer_count_reservations, 0, "");
249 SYSCTL_LONG(_vfs_hammer, OID_AUTO, count_io_running_read, CTLFLAG_RD,
250 &hammer_count_io_running_read, 0, "");
251 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_locked, CTLFLAG_RD,
252 &hammer_count_io_locked, 0, "");
253 SYSCTL_LONG(_vfs_hammer, OID_AUTO, count_io_running_write, CTLFLAG_RD,
254 &hammer_count_io_running_write, 0, "");
255 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, contention_count, CTLFLAG_RW,
256 &hammer_contention_count, 0, "");
257 SYSCTL_INT(_vfs_hammer, OID_AUTO, autoflush, CTLFLAG_RW,
258 &hammer_autoflush, 0, "");
259 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_zone, CTLFLAG_RW,
260 &hammer_verify_zone, 0, "");
261 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_data, CTLFLAG_RW,
262 &hammer_verify_data, 0, "");
263 SYSCTL_INT(_vfs_hammer, OID_AUTO, double_buffer, CTLFLAG_RW,
264 &hammer_double_buffer, 0, "");
265 SYSCTL_INT(_vfs_hammer, OID_AUTO, btree_full_undo, CTLFLAG_RW,
266 &hammer_btree_full_undo, 0, "");
267 SYSCTL_INT(_vfs_hammer, OID_AUTO, yield_check, CTLFLAG_RW,
268 &hammer_yield_check, 0, "");
269 SYSCTL_INT(_vfs_hammer, OID_AUTO, fsync_mode, CTLFLAG_RW,
270 &hammer_fsync_mode, 0, "");
272 /* KTR_INFO_MASTER(hammer); */
277 static void hammer_free_hmp(struct mount *mp);
279 static int hammer_vfs_mount(struct mount *mp, char *path, caddr_t data,
281 static int hammer_vfs_unmount(struct mount *mp, int mntflags);
282 static int hammer_vfs_root(struct mount *mp, struct vnode **vpp);
283 static int hammer_vfs_statfs(struct mount *mp, struct statfs *sbp,
285 static int hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp,
287 static int hammer_vfs_sync(struct mount *mp, int waitfor);
288 static int hammer_vfs_vget(struct mount *mp, struct vnode *dvp,
289 ino_t ino, struct vnode **vpp);
290 static int hammer_vfs_init(struct vfsconf *conf);
291 static int hammer_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
292 struct fid *fhp, struct vnode **vpp);
293 static int hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp);
294 static int hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
295 int *exflagsp, struct ucred **credanonp);
298 static struct vfsops hammer_vfsops = {
299 .vfs_mount = hammer_vfs_mount,
300 .vfs_unmount = hammer_vfs_unmount,
301 .vfs_root = hammer_vfs_root,
302 .vfs_statfs = hammer_vfs_statfs,
303 .vfs_statvfs = hammer_vfs_statvfs,
304 .vfs_sync = hammer_vfs_sync,
305 .vfs_vget = hammer_vfs_vget,
306 .vfs_init = hammer_vfs_init,
307 .vfs_vptofh = hammer_vfs_vptofh,
308 .vfs_fhtovp = hammer_vfs_fhtovp,
309 .vfs_checkexp = hammer_vfs_checkexp
312 MALLOC_DEFINE(M_HAMMER, "HAMMER-mount", "");
314 VFS_SET(hammer_vfsops, hammer, 0);
315 MODULE_VERSION(hammer, 1);
318 hammer_vfs_init(struct vfsconf *conf)
323 * Wait up to this long for an exclusive deadlock to clear
324 * before acquiring a new shared lock on the ip. The deadlock
325 * may have occured on a b-tree node related to the ip.
327 if (hammer_tdmux_ticks == 0)
328 hammer_tdmux_ticks = hz / 5;
331 * Autosize, but be careful because a hammer filesystem's
332 * reserve is partially calculated based on dirtybufspace,
333 * so we simply cannot allow it to get too large.
335 if (hammer_limit_recs == 0) {
337 if (n > kmalloc_limit(M_HAMMER) / 512)
338 n = kmalloc_limit(M_HAMMER) / 512;
339 if (n > 2 * 1024 * 1024)
341 hammer_limit_recs = (int)n;
343 if (hammer_limit_dirtybufspace == 0) {
344 hammer_limit_dirtybufspace = hidirtybufspace / 2;
345 if (hammer_limit_dirtybufspace < 1L * 1024 * 1024)
346 hammer_limit_dirtybufspace = 1024L * 1024;
347 if (hammer_limit_dirtybufspace > 1024L * 1024 * 1024)
348 hammer_limit_dirtybufspace = 1024L * 1024 * 1024;
352 * The hammer_inode structure detaches from the vnode on reclaim.
353 * This limits the number of inodes in this state to prevent a
354 * memory pool blowout.
356 if (hammer_limit_reclaims == 0)
357 hammer_limit_reclaims = desiredvnodes / 10;
363 hammer_vfs_mount(struct mount *mp, char *mntpt, caddr_t data,
366 struct hammer_mount_info info;
368 hammer_volume_t rootvol;
369 struct vnode *rootvp;
370 struct vnode *devvp = NULL;
371 const char *upath; /* volume name in userspace */
372 char *path; /* volume name in system space */
377 char *next_volume_ptr = NULL;
379 if (hammer_noatime) {
381 mp->mnt_flag |= MNT_NOATIME;
385 * Accept hammer_mount_info. mntpt is NULL for root mounts at boot.
388 bzero(&info, sizeof(info));
393 next_volume_ptr = mp->mnt_stat.f_mntfromname;
395 /* Count number of volumes separated by ':' */
396 for (char *p = next_volume_ptr; *p != '\0'; ++p) {
402 mp->mnt_flag &= ~MNT_RDONLY; /* mount R/W */
404 if ((error = copyin(data, &info, sizeof(info))) != 0)
409 * updating or new mount
411 if (mp->mnt_flag & MNT_UPDATE) {
412 hmp = (void *)mp->mnt_data;
413 KKASSERT(hmp != NULL);
415 if (info.nvolumes <= 0 || info.nvolumes > HAMMER_MAX_VOLUMES)
421 * master-id validation. The master id may not be changed by a
424 if (info.hflags & HMNT_MASTERID || info.hflags & HMNT_NOMIRROR) {
425 if (hmp && hmp->master_id != info.master_id) {
426 hkprintf("cannot change master id with mount update\n");
429 master_id = info.master_id;
430 if (master_id < -1 || master_id >= HAMMER_MAX_MASTERS)
434 master_id = hmp->master_id;
440 * Internal mount data structure
443 hmp = kmalloc(sizeof(*hmp), M_HAMMER, M_WAITOK | M_ZERO);
444 mp->mnt_data = (qaddr_t)hmp;
448 * Make sure kmalloc type limits are set appropriately.
450 * Our inode kmalloc group is sized based on maxvnodes
451 * (controlled by the system, not us).
453 kmalloc_create(&hmp->m_misc, "HAMMER-others");
454 kmalloc_create(&hmp->m_inodes, "HAMMER-inodes");
456 kmalloc_raise_limit(hmp->m_inodes, 0); /* unlimited */
458 hmp->root_btree_beg.localization = 0x00000000U;
459 hmp->root_btree_beg.obj_id = -0x8000000000000000LL;
460 hmp->root_btree_beg.key = -0x8000000000000000LL;
461 hmp->root_btree_beg.create_tid = 1;
462 hmp->root_btree_beg.delete_tid = 1;
463 hmp->root_btree_beg.rec_type = 0;
464 hmp->root_btree_beg.obj_type = 0;
465 hmp->root_btree_beg.btype = HAMMER_BTREE_TYPE_NONE;
467 hmp->root_btree_end.localization = 0xFFFFFFFFU;
468 hmp->root_btree_end.obj_id = 0x7FFFFFFFFFFFFFFFLL;
469 hmp->root_btree_end.key = 0x7FFFFFFFFFFFFFFFLL;
470 hmp->root_btree_end.create_tid = 0xFFFFFFFFFFFFFFFFULL;
471 hmp->root_btree_end.delete_tid = 0; /* special case */
472 hmp->root_btree_end.rec_type = 0xFFFFU;
473 hmp->root_btree_end.obj_type = 0;
474 hmp->root_btree_end.btype = HAMMER_BTREE_TYPE_NONE;
476 hmp->krate.freq = 1; /* maximum reporting rate (hz) */
477 hmp->krate.count = -16; /* initial burst */
478 hmp->kdiag.freq = 1; /* maximum reporting rate (hz) */
479 hmp->kdiag.count = -16; /* initial burst */
481 hmp->sync_lock.refs = 1;
482 hmp->free_lock.refs = 1;
483 hmp->undo_lock.refs = 1;
484 hmp->blkmap_lock.refs = 1;
485 hmp->snapshot_lock.refs = 1;
486 hmp->volume_lock.refs = 1;
488 TAILQ_INIT(&hmp->delay_list);
489 TAILQ_INIT(&hmp->flush_group_list);
490 TAILQ_INIT(&hmp->objid_cache_list);
491 TAILQ_INIT(&hmp->undo_lru_list);
492 TAILQ_INIT(&hmp->reclaim_list);
494 RB_INIT(&hmp->rb_dedup_crc_root);
495 RB_INIT(&hmp->rb_dedup_off_root);
496 TAILQ_INIT(&hmp->dedup_lru_list);
498 hmp->hflags &= ~HMNT_USERFLAGS;
499 hmp->hflags |= info.hflags & HMNT_USERFLAGS;
501 hmp->master_id = master_id;
504 mp->mnt_flag |= MNT_RDONLY;
505 hmp->asof = info.asof;
507 hmp->asof = HAMMER_MAX_TID;
510 hmp->volume_to_remove = -1;
513 * Re-open read-write if originally read-only, or vise-versa.
515 * When going from read-only to read-write execute the stage2
516 * recovery if it has not already been run.
518 if (mp->mnt_flag & MNT_UPDATE) {
519 lwkt_gettoken(&hmp->fs_token);
521 if (hmp->ronly && (mp->mnt_kern_flag & MNTK_WANTRDWR)) {
522 hkprintf("read-only -> read-write\n");
524 RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
525 hammer_adjust_volume_mode, NULL);
526 rootvol = hammer_get_root_volume(hmp, &error);
528 hammer_recover_flush_buffers(hmp, rootvol, 1);
529 error = hammer_recover_stage2(hmp, rootvol);
530 bcopy(rootvol->ondisk->vol0_blockmap,
532 sizeof(hmp->blockmap));
533 hammer_rel_volume(rootvol, 0);
535 RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
536 hammer_reload_inode, NULL);
537 /* kernel clears MNT_RDONLY */
538 } else if (hmp->ronly == 0 && (mp->mnt_flag & MNT_RDONLY)) {
539 hkprintf("read-write -> read-only\n");
540 hmp->ronly = 1; /* messy */
541 RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
542 hammer_reload_inode, NULL);
544 hammer_flusher_sync(hmp);
545 hammer_flusher_sync(hmp);
546 hammer_flusher_sync(hmp);
548 RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
549 hammer_adjust_volume_mode, NULL);
551 lwkt_reltoken(&hmp->fs_token);
555 RB_INIT(&hmp->rb_vols_root);
556 RB_INIT(&hmp->rb_inos_root);
557 RB_INIT(&hmp->rb_redo_root);
558 RB_INIT(&hmp->rb_nods_root);
559 RB_INIT(&hmp->rb_undo_root);
560 RB_INIT(&hmp->rb_resv_root);
561 RB_INIT(&hmp->rb_bufs_root);
562 RB_INIT(&hmp->rb_pfsm_root);
564 hmp->ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
566 RB_INIT(&hmp->volu_root);
567 RB_INIT(&hmp->undo_root);
568 RB_INIT(&hmp->data_root);
569 RB_INIT(&hmp->meta_root);
570 RB_INIT(&hmp->lose_root);
571 TAILQ_INIT(&hmp->iorun_list);
573 lwkt_token_init(&hmp->fs_token, "hammerfs");
574 lwkt_token_init(&hmp->io_token, "hammerio");
576 lwkt_gettoken(&hmp->fs_token);
581 path = objcache_get(namei_oc, M_WAITOK);
583 for (i = 0; i < info.nvolumes; ++i) {
588 KKASSERT(next_volume_ptr != NULL);
590 if (*next_volume_ptr != '/') {
592 strcpy(path, "/dev/");
595 for (k = strlen(path); k < MAXPATHLEN-1; ++k) {
596 if (*next_volume_ptr == '\0') {
598 } else if (*next_volume_ptr == ':') {
602 path[k] = *next_volume_ptr;
609 cdev_t dev = kgetdiskbyname(path);
610 error = bdevvp(dev, &devvp);
612 hdkprintf("can't find devvp\n");
615 error = copyin(&info.volumes[i], &upath,
618 error = copyinstr(upath, path,
622 error = hammer_install_volume(hmp, path, devvp, NULL);
626 objcache_put(namei_oc, path);
629 * Make sure we found a root volume
631 if (hmp->rootvol == NULL) {
632 if (error == EBUSY) {
633 hdkprintf("The volumes are probably mounted\n");
635 hdkprintf("No root volume found!\n");
642 * Check that all required volumes are available
644 if (error == 0 && hammer_mountcheck_volumes(hmp)) {
645 hdkprintf("Missing volumes, cannot mount!\n");
654 hdkprintf("Failed to load volumes!\n");
658 nvolumes = hammer_get_installed_volumes(hmp);
659 if (hmp->nvolumes != nvolumes) {
660 hdkprintf("volume header says %d volumes, but %d installed\n",
661 hmp->nvolumes, nvolumes);
667 * No errors, setup enough of the mount point so we can lookup the
670 mp->mnt_iosize_max = MAXPHYS;
671 mp->mnt_kern_flag |= MNTK_FSMID;
672 mp->mnt_kern_flag |= MNTK_THR_SYNC; /* new vsyncscan semantics */
675 * MPSAFE code. Note that VOPs and VFSops which are not MPSAFE
676 * will acquire a per-mount token prior to entry and release it
679 mp->mnt_kern_flag |= MNTK_ALL_MPSAFE;
682 * note: f_iosize is used by vnode_pager_haspage() when constructing
685 mp->mnt_stat.f_iosize = HAMMER_BUFSIZE;
686 mp->mnt_stat.f_bsize = HAMMER_BUFSIZE;
688 mp->mnt_vstat.f_frsize = HAMMER_BUFSIZE;
689 mp->mnt_vstat.f_bsize = HAMMER_BUFSIZE;
691 mp->mnt_maxsymlinklen = 255;
692 mp->mnt_flag |= MNT_LOCAL;
694 vfs_add_vnodeops(mp, &hammer_vnode_vops, &mp->mnt_vn_norm_ops);
695 vfs_add_vnodeops(mp, &hammer_spec_vops, &mp->mnt_vn_spec_ops);
696 vfs_add_vnodeops(mp, &hammer_fifo_vops, &mp->mnt_vn_fifo_ops);
699 * The root volume's ondisk pointer is only valid if we hold a
702 rootvol = hammer_get_root_volume(hmp, &error);
707 * Perform any necessary UNDO operations. The recovery code does
708 * call hammer_undo_lookup() so we have to pre-cache the blockmap,
709 * and then re-copy it again after recovery is complete.
711 * If this is a read-only mount the UNDO information is retained
712 * in memory in the form of dirty buffer cache buffers, and not
713 * written back to the media.
715 bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
716 sizeof(hmp->blockmap));
719 * Check filesystem version
721 hmp->version = rootvol->ondisk->vol_version;
722 if (hmp->version < HAMMER_VOL_VERSION_MIN ||
723 hmp->version > HAMMER_VOL_VERSION_MAX) {
724 hkprintf("mount unsupported fs version %d\n", hmp->version);
730 * The undo_rec_limit limits the size of flush groups to avoid
731 * blowing out the UNDO FIFO. This calculation is typically in
732 * the tens of thousands and is designed primarily when small
733 * HAMMER filesystems are created.
735 hmp->undo_rec_limit = hammer_undo_max(hmp) / 8192 + 100;
736 if (hammer_debug_general & 0x0001)
737 hkprintf("undo_rec_limit %d\n", hmp->undo_rec_limit);
740 * NOTE: Recover stage1 not only handles meta-data recovery, it
741 * also sets hmp->undo_seqno for HAMMER VERSION 4+ filesystems.
743 error = hammer_recover_stage1(hmp, rootvol);
745 kprintf("Failed to recover HAMMER filesystem on mount\n");
750 * Finish setup now that we have a good root volume.
751 * vol_name is a filesystem label string.
753 ksnprintf(mp->mnt_stat.f_mntfromname,
754 sizeof(mp->mnt_stat.f_mntfromname), "%s",
755 rootvol->ondisk->vol_name);
756 mp->mnt_stat.f_fsid.val[0] =
757 crc32((char *)&rootvol->ondisk->vol_fsid + 0, 8);
758 mp->mnt_stat.f_fsid.val[1] =
759 crc32((char *)&rootvol->ondisk->vol_fsid + 8, 8);
760 mp->mnt_stat.f_fsid.val[1] &= HAMMER_LOCALIZE_MASK;
762 mp->mnt_vstat.f_fsid_uuid = rootvol->ondisk->vol_fsid;
763 mp->mnt_vstat.f_fsid = crc32(&mp->mnt_vstat.f_fsid_uuid,
764 sizeof(mp->mnt_vstat.f_fsid_uuid));
767 * Certain often-modified fields in the root volume are cached in
768 * the hammer_mount structure so we do not have to generate lots
769 * of little UNDO structures for them.
771 * Recopy after recovery. This also has the side effect of
772 * setting our cached undo FIFO's first_offset, which serves to
773 * placemark the FIFO start for the NEXT flush cycle while the
774 * on-disk first_offset represents the LAST flush cycle.
776 hmp->next_tid = rootvol->ondisk->vol0_next_tid;
777 hmp->flush_tid1 = hmp->next_tid;
778 hmp->flush_tid2 = hmp->next_tid;
779 bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
780 sizeof(hmp->blockmap));
781 hmp->copy_stat_freebigblocks = rootvol->ondisk->vol0_stat_freebigblocks;
783 hammer_flusher_create(hmp);
786 * Locate the root directory with an obj_id of 1.
788 error = hammer_vfs_root(mp, &rootvp);
793 error = hammer_recover_stage2(hmp, rootvol);
796 * If the stage2 recovery fails be sure to clean out all cached
797 * vnodes before throwing away the mount structure or bad things
804 if ((mp->mnt_flag & MNT_UPDATE) == 0) {
807 /* Populate info for mount point (NULL pad)*/
808 bzero(mp->mnt_stat.f_mntonname, MNAMELEN);
811 copyinstr(mntpt, mp->mnt_stat.f_mntonname,
813 } else { /* Root mount */
814 mp->mnt_stat.f_mntonname[0] = '/';
817 (void)VFS_STATFS(mp, &mp->mnt_stat, cred);
818 hammer_rel_volume(rootvol, 0);
821 * Cleanup and return.
824 /* called with fs_token held */
827 lwkt_reltoken(&hmp->fs_token);
833 hammer_vfs_unmount(struct mount *mp, int mntflags)
835 hammer_mount_t hmp = (void *)mp->mnt_data;
840 * Clean out the vnodes
842 lwkt_gettoken(&hmp->fs_token);
844 if (mntflags & MNT_FORCE)
846 error = vflush(mp, 0, flags);
849 * Clean up the internal mount structure and related entities. This
853 /* called with fs_token held */
856 lwkt_reltoken(&hmp->fs_token);
862 * Clean up the internal mount structure and disassociate it from the mount.
863 * This may issue I/O.
865 * Called with fs_token held.
868 hammer_free_hmp(struct mount *mp)
870 hammer_mount_t hmp = (void *)mp->mnt_data;
871 hammer_flush_group_t flg;
874 * Flush anything dirty. This won't even run if the
875 * filesystem errored-out.
877 hammer_flush_dirty(hmp, 30);
880 * If the mount had a critical error we have to destroy any
881 * remaining inodes before we can finish cleaning up the flusher.
883 if (hmp->flags & HAMMER_MOUNT_CRITICAL_ERROR) {
884 RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
885 hammer_destroy_inode_callback, NULL);
889 * There shouldn't be any inodes left now and any left over
890 * flush groups should now be empty.
892 KKASSERT(RB_EMPTY(&hmp->rb_inos_root));
893 while ((flg = TAILQ_FIRST(&hmp->flush_group_list)) != NULL) {
894 TAILQ_REMOVE(&hmp->flush_group_list, flg, flush_entry);
895 KKASSERT(RB_EMPTY(&flg->flush_tree));
897 hkprintf("Warning, flush_group %p was "
898 "not empty on umount!\n", flg);
900 kfree(flg, hmp->m_misc);
904 * We can finally destroy the flusher
906 hammer_flusher_destroy(hmp);
909 * We may have held recovered buffers due to a read-only mount.
910 * These must be discarded.
913 hammer_recover_flush_buffers(hmp, NULL, -1);
916 * Unload buffers and then volumes
918 RB_SCAN(hammer_buf_rb_tree, &hmp->rb_bufs_root, NULL,
919 hammer_unload_buffer, NULL);
920 RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
921 hammer_unload_volume, NULL);
924 mp->mnt_flag &= ~MNT_LOCAL;
926 hammer_destroy_objid_cache(hmp);
927 hammer_destroy_dedup_cache(hmp);
928 if (hmp->dedup_free_cache != NULL) {
929 kfree(hmp->dedup_free_cache, hmp->m_misc);
930 hmp->dedup_free_cache = NULL;
932 kmalloc_destroy(&hmp->m_misc);
933 kmalloc_destroy(&hmp->m_inodes);
934 lwkt_reltoken(&hmp->fs_token);
935 kfree(hmp, M_HAMMER);
939 * Report critical errors. ip may be NULL.
942 hammer_critical_error(hammer_mount_t hmp, hammer_inode_t ip,
943 int error, const char *msg)
945 hmp->flags |= HAMMER_MOUNT_CRITICAL_ERROR;
947 hmkrateprintf(&hmp->krate, hmp,
948 "Critical error inode=%jd error=%d %s\n",
949 (intmax_t)(ip ? ip->obj_id : -1),
952 if (hmp->ronly == 0) {
953 hmp->ronly = 2; /* special errored read-only mode */
954 hmp->mp->mnt_flag |= MNT_RDONLY;
955 RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
956 hammer_adjust_volume_mode, NULL);
957 hmkprintf(hmp, "Forcing read-only mode\n");
960 if (hammer_debug_critical)
961 Debugger("Entering debugger");
966 * Obtain a vnode for the specified inode number. An exclusively locked
970 hammer_vfs_vget(struct mount *mp, struct vnode *dvp,
971 ino_t ino, struct vnode **vpp)
973 struct hammer_transaction trans;
974 struct hammer_mount *hmp = (void *)mp->mnt_data;
975 struct hammer_inode *ip;
977 uint32_t localization;
979 lwkt_gettoken(&hmp->fs_token);
980 hammer_simple_transaction(&trans, hmp);
983 * If a directory vnode is supplied (mainly NFS) then we can acquire
984 * the PFS domain from it. Otherwise we would only be able to vget
985 * inodes in the root PFS.
988 localization = HAMMER_DEF_LOCALIZATION |
989 VTOI(dvp)->obj_localization;
991 localization = HAMMER_DEF_LOCALIZATION;
995 * Lookup the requested HAMMER inode. The structure must be
996 * left unlocked while we manipulate the related vnode to avoid
999 ip = hammer_get_inode(&trans, NULL, ino,
1000 hmp->asof, localization,
1005 error = hammer_get_vnode(ip, vpp);
1006 hammer_rel_inode(ip, 0);
1008 hammer_done_transaction(&trans);
1009 lwkt_reltoken(&hmp->fs_token);
1014 * Return the root vnode for the filesystem.
1016 * HAMMER stores the root vnode in the hammer_mount structure so
1017 * getting it is easy.
1020 hammer_vfs_root(struct mount *mp, struct vnode **vpp)
1024 error = hammer_vfs_vget(mp, NULL, HAMMER_OBJID_ROOT, vpp);
1029 hammer_vfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
1031 struct hammer_mount *hmp = (void *)mp->mnt_data;
1032 hammer_volume_t volume;
1033 hammer_volume_ondisk_t ondisk;
1038 lwkt_gettoken(&hmp->fs_token);
1039 volume = hammer_get_root_volume(hmp, &error);
1041 lwkt_reltoken(&hmp->fs_token);
1044 ondisk = volume->ondisk;
1049 _hammer_checkspace(hmp, HAMMER_CHKSPC_WRITE, &breserved);
1050 mp->mnt_stat.f_files = ondisk->vol0_stat_inodes;
1051 bfree = ondisk->vol0_stat_freebigblocks * HAMMER_BIGBLOCK_SIZE;
1052 hammer_rel_volume(volume, 0);
1054 mp->mnt_stat.f_bfree = (bfree - breserved) / HAMMER_BUFSIZE;
1055 mp->mnt_stat.f_bavail = mp->mnt_stat.f_bfree;
1056 if (mp->mnt_stat.f_files < 0)
1057 mp->mnt_stat.f_files = 0;
1059 *sbp = mp->mnt_stat;
1060 lwkt_reltoken(&hmp->fs_token);
1065 hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred)
1067 struct hammer_mount *hmp = (void *)mp->mnt_data;
1068 hammer_volume_t volume;
1069 hammer_volume_ondisk_t ondisk;
1074 lwkt_gettoken(&hmp->fs_token);
1075 volume = hammer_get_root_volume(hmp, &error);
1077 lwkt_reltoken(&hmp->fs_token);
1080 ondisk = volume->ondisk;
1085 _hammer_checkspace(hmp, HAMMER_CHKSPC_WRITE, &breserved);
1086 mp->mnt_vstat.f_files = ondisk->vol0_stat_inodes;
1087 bfree = ondisk->vol0_stat_freebigblocks * HAMMER_BIGBLOCK_SIZE;
1088 hammer_rel_volume(volume, 0);
1090 mp->mnt_vstat.f_bfree = (bfree - breserved) / HAMMER_BUFSIZE;
1091 mp->mnt_vstat.f_bavail = mp->mnt_vstat.f_bfree;
1092 if (mp->mnt_vstat.f_files < 0)
1093 mp->mnt_vstat.f_files = 0;
1094 *sbp = mp->mnt_vstat;
1095 lwkt_reltoken(&hmp->fs_token);
1100 * Sync the filesystem. Currently we have to run it twice, the second
1101 * one will advance the undo start index to the end index, so if a crash
1102 * occurs no undos will be run on mount.
1104 * We do not sync the filesystem if we are called from a panic. If we did
1105 * we might end up blowing up a sync that was already in progress.
1108 hammer_vfs_sync(struct mount *mp, int waitfor)
1110 struct hammer_mount *hmp = (void *)mp->mnt_data;
1113 lwkt_gettoken(&hmp->fs_token);
1114 if (panicstr == NULL) {
1115 error = hammer_sync_hmp(hmp, waitfor);
1119 lwkt_reltoken(&hmp->fs_token);
1124 * Convert a vnode to a file handle.
1126 * Accesses read-only fields on already-referenced structures so
1127 * no token is needed.
1130 hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp)
1134 KKASSERT(MAXFIDSZ >= 16);
1136 fhp->fid_len = offsetof(struct fid, fid_data[16]);
1137 fhp->fid_ext = ip->obj_localization >> 16;
1138 bcopy(&ip->obj_id, fhp->fid_data + 0, sizeof(ip->obj_id));
1139 bcopy(&ip->obj_asof, fhp->fid_data + 8, sizeof(ip->obj_asof));
1145 * Convert a file handle back to a vnode.
1147 * Use rootvp to enforce PFS isolation when a PFS is exported via a
1151 hammer_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
1152 struct fid *fhp, struct vnode **vpp)
1154 hammer_mount_t hmp = (void *)mp->mnt_data;
1155 struct hammer_transaction trans;
1156 struct hammer_inode *ip;
1157 struct hammer_inode_info info;
1159 uint32_t localization;
1161 bcopy(fhp->fid_data + 0, &info.obj_id, sizeof(info.obj_id));
1162 bcopy(fhp->fid_data + 8, &info.obj_asof, sizeof(info.obj_asof));
1164 localization = VTOI(rootvp)->obj_localization;
1166 localization = (uint32_t)fhp->fid_ext << 16;
1168 lwkt_gettoken(&hmp->fs_token);
1169 hammer_simple_transaction(&trans, hmp);
1172 * Get/allocate the hammer_inode structure. The structure must be
1173 * unlocked while we manipulate the related vnode to avoid a
1176 ip = hammer_get_inode(&trans, NULL, info.obj_id,
1177 info.obj_asof, localization, 0, &error);
1179 error = hammer_get_vnode(ip, vpp);
1180 hammer_rel_inode(ip, 0);
1184 hammer_done_transaction(&trans);
1185 lwkt_reltoken(&hmp->fs_token);
1190 hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
1191 int *exflagsp, struct ucred **credanonp)
1193 hammer_mount_t hmp = (void *)mp->mnt_data;
1197 lwkt_gettoken(&hmp->fs_token);
1198 np = vfs_export_lookup(mp, &hmp->export, nam);
1200 *exflagsp = np->netc_exflags;
1201 *credanonp = &np->netc_anon;
1206 lwkt_reltoken(&hmp->fs_token);
1212 hammer_vfs_export(struct mount *mp, int op, const struct export_args *export)
1214 hammer_mount_t hmp = (void *)mp->mnt_data;
1217 lwkt_gettoken(&hmp->fs_token);
1220 case MOUNTCTL_SET_EXPORT:
1221 error = vfs_export(mp, &hmp->export, export);
1227 lwkt_reltoken(&hmp->fs_token);