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/param.h>
36 #include <sys/systm.h>
37 #include <sys/kernel.h>
38 #include <sys/vnode.h>
39 #include <sys/mount.h>
40 #include <sys/malloc.h>
41 #include <sys/nlookup.h>
42 #include <sys/fcntl.h>
43 #include <sys/sysctl.h>
49 * NOTE! Global statistics may not be MPSAFE so HAMMER never uses them
52 int hammer_supported_version = HAMMER_VOL_VERSION_DEFAULT;
54 int hammer_debug_general;
55 int hammer_debug_debug = 1; /* medium-error panics */
56 int hammer_debug_inode;
57 int hammer_debug_locks;
58 int hammer_debug_btree;
60 int hammer_debug_recover; /* -1 will disable, +1 will force */
61 int hammer_debug_recover_faults;
62 int hammer_debug_critical; /* non-zero enter debugger on error */
63 int hammer_cluster_enable = 1; /* enable read clustering by default */
64 int hammer_live_dedup = 0;
65 int hammer_tdmux_ticks;
66 int hammer_count_fsyncs;
67 int hammer_count_inodes;
68 int hammer_count_iqueued;
69 int hammer_count_reclaims;
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 long hammer_count_dirtybufspace; /* global */
98 int hammer_count_refedbufs; /* global */
99 int hammer_count_reservations;
100 long hammer_count_io_running_read;
101 long hammer_count_io_running_write;
102 int hammer_count_io_locked;
103 long hammer_limit_dirtybufspace; /* per-mount */
104 int hammer_limit_recs; /* as a whole XXX */
105 int hammer_limit_inode_recs = 2048; /* per inode */
106 int hammer_limit_reclaims;
107 int hammer_live_dedup_cache_size = DEDUP_CACHE_SIZE;
108 int hammer_limit_redo = 4096 * 1024; /* per inode */
109 int hammer_autoflush = 500; /* auto flush (typ on reclaim) */
110 int hammer_bio_count;
111 int hammer_verify_zone;
112 int hammer_verify_data = 1;
113 int hammer_write_mode;
114 int hammer_double_buffer;
115 int hammer_yield_check = 16;
116 int hammer_fsync_mode = 3;
117 int64_t hammer_contention_count;
118 int64_t hammer_zone_limit;
121 * Live dedup debug counters (sysctls are writable so that counters
122 * can be reset from userspace).
124 int64_t hammer_live_dedup_vnode_bcmps = 0;
125 int64_t hammer_live_dedup_device_bcmps = 0;
126 int64_t hammer_live_dedup_findblk_failures = 0;
127 int64_t hammer_live_dedup_bmap_saves = 0;
130 SYSCTL_NODE(_vfs, OID_AUTO, hammer, CTLFLAG_RW, 0, "HAMMER filesystem");
132 SYSCTL_INT(_vfs_hammer, OID_AUTO, supported_version, CTLFLAG_RD,
133 &hammer_supported_version, 0, "");
134 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_general, CTLFLAG_RW,
135 &hammer_debug_general, 0, "");
136 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_io, CTLFLAG_RW,
137 &hammer_debug_io, 0, "");
138 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_debug, CTLFLAG_RW,
139 &hammer_debug_debug, 0, "");
140 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_inode, CTLFLAG_RW,
141 &hammer_debug_inode, 0, "");
142 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_locks, CTLFLAG_RW,
143 &hammer_debug_locks, 0, "");
144 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_btree, CTLFLAG_RW,
145 &hammer_debug_btree, 0, "");
146 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_tid, CTLFLAG_RW,
147 &hammer_debug_tid, 0, "");
148 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover, CTLFLAG_RW,
149 &hammer_debug_recover, 0, "");
150 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover_faults, CTLFLAG_RW,
151 &hammer_debug_recover_faults, 0, "");
152 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_critical, CTLFLAG_RW,
153 &hammer_debug_critical, 0, "");
154 SYSCTL_INT(_vfs_hammer, OID_AUTO, cluster_enable, CTLFLAG_RW,
155 &hammer_cluster_enable, 0, "");
157 * 0 - live dedup is disabled
158 * 1 - dedup cache is populated on reads only
159 * 2 - dedup cache is populated on both reads and writes
161 SYSCTL_INT(_vfs_hammer, OID_AUTO, live_dedup, CTLFLAG_RD,
162 &hammer_live_dedup, 0, "Enable live dedup (experimental)");
163 SYSCTL_INT(_vfs_hammer, OID_AUTO, tdmux_ticks, CTLFLAG_RW,
164 &hammer_tdmux_ticks, 0, "Hammer tdmux ticks");
166 SYSCTL_LONG(_vfs_hammer, OID_AUTO, limit_dirtybufspace, CTLFLAG_RW,
167 &hammer_limit_dirtybufspace, 0, "");
168 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_recs, CTLFLAG_RW,
169 &hammer_limit_recs, 0, "");
170 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_inode_recs, CTLFLAG_RW,
171 &hammer_limit_inode_recs, 0, "");
172 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_reclaims, CTLFLAG_RW,
173 &hammer_limit_reclaims, 0, "");
174 SYSCTL_INT(_vfs_hammer, OID_AUTO, live_dedup_cache_size, CTLFLAG_RW,
175 &hammer_live_dedup_cache_size, 0,
176 "Number of cache entries");
177 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_redo, CTLFLAG_RW,
178 &hammer_limit_redo, 0, "");
180 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_fsyncs, CTLFLAG_RD,
181 &hammer_count_fsyncs, 0, "");
182 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_inodes, CTLFLAG_RD,
183 &hammer_count_inodes, 0, "");
184 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_iqueued, CTLFLAG_RD,
185 &hammer_count_iqueued, 0, "");
186 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reclaims, CTLFLAG_RD,
187 &hammer_count_reclaims, 0, "");
188 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_records, CTLFLAG_RD,
189 &hammer_count_records, 0, "");
190 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_record_datas, CTLFLAG_RD,
191 &hammer_count_record_datas, 0, "");
192 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_volumes, CTLFLAG_RD,
193 &hammer_count_volumes, 0, "");
194 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_buffers, CTLFLAG_RD,
195 &hammer_count_buffers, 0, "");
196 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_nodes, CTLFLAG_RD,
197 &hammer_count_nodes, 0, "");
198 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, count_extra_space_used, CTLFLAG_RD,
199 &hammer_count_extra_space_used, 0, "");
201 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_searches, CTLFLAG_RD,
202 &hammer_stats_btree_searches, 0, "");
203 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_lookups, CTLFLAG_RD,
204 &hammer_stats_btree_lookups, 0, "");
205 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_inserts, CTLFLAG_RD,
206 &hammer_stats_btree_inserts, 0, "");
207 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_deletes, CTLFLAG_RD,
208 &hammer_stats_btree_deletes, 0, "");
209 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_elements, CTLFLAG_RD,
210 &hammer_stats_btree_elements, 0, "");
211 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_splits, CTLFLAG_RD,
212 &hammer_stats_btree_splits, 0, "");
213 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_iterations, CTLFLAG_RD,
214 &hammer_stats_btree_iterations, 0, "");
215 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_root_iterations, CTLFLAG_RD,
216 &hammer_stats_btree_root_iterations, 0, "");
217 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_record_iterations, CTLFLAG_RD,
218 &hammer_stats_record_iterations, 0, "");
220 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_read, CTLFLAG_RD,
221 &hammer_stats_file_read, 0, "");
222 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_write, CTLFLAG_RD,
223 &hammer_stats_file_write, 0, "");
224 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_iopsr, CTLFLAG_RD,
225 &hammer_stats_file_iopsr, 0, "");
226 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_iopsw, CTLFLAG_RD,
227 &hammer_stats_file_iopsw, 0, "");
228 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_disk_read, CTLFLAG_RD,
229 &hammer_stats_disk_read, 0, "");
230 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_disk_write, CTLFLAG_RD,
231 &hammer_stats_disk_write, 0, "");
232 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_inode_flushes, CTLFLAG_RD,
233 &hammer_stats_inode_flushes, 0, "");
234 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_commits, CTLFLAG_RD,
235 &hammer_stats_commits, 0, "");
236 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_undo, CTLFLAG_RD,
237 &hammer_stats_undo, 0, "");
238 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_redo, CTLFLAG_RD,
239 &hammer_stats_redo, 0, "");
241 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, live_dedup_vnode_bcmps, CTLFLAG_RW,
242 &hammer_live_dedup_vnode_bcmps, 0,
243 "successful vnode buffer comparisons");
244 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, live_dedup_device_bcmps, CTLFLAG_RW,
245 &hammer_live_dedup_device_bcmps, 0,
246 "successful device buffer comparisons");
247 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, live_dedup_findblk_failures, CTLFLAG_RW,
248 &hammer_live_dedup_findblk_failures, 0,
249 "block lookup failures for comparison");
250 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, live_dedup_bmap_saves, CTLFLAG_RW,
251 &hammer_live_dedup_bmap_saves, 0,
252 "useful physical block lookups");
254 SYSCTL_LONG(_vfs_hammer, OID_AUTO, count_dirtybufspace, CTLFLAG_RD,
255 &hammer_count_dirtybufspace, 0, "");
256 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_refedbufs, CTLFLAG_RD,
257 &hammer_count_refedbufs, 0, "");
258 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reservations, CTLFLAG_RD,
259 &hammer_count_reservations, 0, "");
260 SYSCTL_LONG(_vfs_hammer, OID_AUTO, count_io_running_read, CTLFLAG_RD,
261 &hammer_count_io_running_read, 0, "");
262 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_locked, CTLFLAG_RD,
263 &hammer_count_io_locked, 0, "");
264 SYSCTL_LONG(_vfs_hammer, OID_AUTO, count_io_running_write, CTLFLAG_RD,
265 &hammer_count_io_running_write, 0, "");
266 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, zone_limit, CTLFLAG_RW,
267 &hammer_zone_limit, 0, "");
268 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, contention_count, CTLFLAG_RW,
269 &hammer_contention_count, 0, "");
270 SYSCTL_INT(_vfs_hammer, OID_AUTO, autoflush, CTLFLAG_RW,
271 &hammer_autoflush, 0, "");
272 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_zone, CTLFLAG_RW,
273 &hammer_verify_zone, 0, "");
274 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_data, CTLFLAG_RW,
275 &hammer_verify_data, 0, "");
276 SYSCTL_INT(_vfs_hammer, OID_AUTO, write_mode, CTLFLAG_RW,
277 &hammer_write_mode, 0, "");
278 SYSCTL_INT(_vfs_hammer, OID_AUTO, double_buffer, CTLFLAG_RW,
279 &hammer_double_buffer, 0, "");
280 SYSCTL_INT(_vfs_hammer, OID_AUTO, yield_check, CTLFLAG_RW,
281 &hammer_yield_check, 0, "");
282 SYSCTL_INT(_vfs_hammer, OID_AUTO, fsync_mode, CTLFLAG_RW,
283 &hammer_fsync_mode, 0, "");
285 /* KTR_INFO_MASTER(hammer); */
290 static void hammer_free_hmp(struct mount *mp);
292 static int hammer_vfs_mount(struct mount *mp, char *path, caddr_t data,
294 static int hammer_vfs_unmount(struct mount *mp, int mntflags);
295 static int hammer_vfs_root(struct mount *mp, struct vnode **vpp);
296 static int hammer_vfs_statfs(struct mount *mp, struct statfs *sbp,
298 static int hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp,
300 static int hammer_vfs_sync(struct mount *mp, int waitfor);
301 static int hammer_vfs_vget(struct mount *mp, struct vnode *dvp,
302 ino_t ino, struct vnode **vpp);
303 static int hammer_vfs_init(struct vfsconf *conf);
304 static int hammer_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
305 struct fid *fhp, struct vnode **vpp);
306 static int hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp);
307 static int hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
308 int *exflagsp, struct ucred **credanonp);
311 static struct vfsops hammer_vfsops = {
312 .vfs_mount = hammer_vfs_mount,
313 .vfs_unmount = hammer_vfs_unmount,
314 .vfs_root = hammer_vfs_root,
315 .vfs_statfs = hammer_vfs_statfs,
316 .vfs_statvfs = hammer_vfs_statvfs,
317 .vfs_sync = hammer_vfs_sync,
318 .vfs_vget = hammer_vfs_vget,
319 .vfs_init = hammer_vfs_init,
320 .vfs_vptofh = hammer_vfs_vptofh,
321 .vfs_fhtovp = hammer_vfs_fhtovp,
322 .vfs_checkexp = hammer_vfs_checkexp
325 MALLOC_DEFINE(M_HAMMER, "HAMMER-mount", "");
327 VFS_SET(hammer_vfsops, hammer, 0);
328 MODULE_VERSION(hammer, 1);
331 hammer_vfs_init(struct vfsconf *conf)
336 * Wait up to this long for an exclusive deadlock to clear
337 * before acquiring a new shared lock on the ip. The deadlock
338 * may have occured on a b-tree node related to the ip.
340 if (hammer_tdmux_ticks == 0)
341 hammer_tdmux_ticks = hz / 5;
344 * Autosize, but be careful because a hammer filesystem's
345 * reserve is partially calculated based on dirtybufspace,
346 * so we simply cannot allow it to get too large.
348 if (hammer_limit_recs == 0) {
350 if (n > kmalloc_limit(M_HAMMER) / 512)
351 n = kmalloc_limit(M_HAMMER) / 512;
352 if (n > 2 * 1024 * 1024)
354 hammer_limit_recs = (int)n;
356 if (hammer_limit_dirtybufspace == 0) {
357 hammer_limit_dirtybufspace = hidirtybufspace / 2;
358 if (hammer_limit_dirtybufspace < 1L * 1024 * 1024)
359 hammer_limit_dirtybufspace = 1024L * 1024;
360 if (hammer_limit_dirtybufspace > 1024L * 1024 * 1024)
361 hammer_limit_dirtybufspace = 1024L * 1024 * 1024;
365 * The hammer_inode structure detaches from the vnode on reclaim.
366 * This limits the number of inodes in this state to prevent a
367 * memory pool blowout.
369 if (hammer_limit_reclaims == 0)
370 hammer_limit_reclaims = desiredvnodes / 10;
376 hammer_vfs_mount(struct mount *mp, char *mntpt, caddr_t data,
379 struct hammer_mount_info info;
381 hammer_volume_t rootvol;
382 struct vnode *rootvp;
383 struct vnode *devvp = NULL;
384 const char *upath; /* volume name in userspace */
385 char *path; /* volume name in system space */
389 char *next_volume_ptr = NULL;
392 * Accept hammer_mount_info. mntpt is NULL for root mounts at boot.
395 bzero(&info, sizeof(info));
400 next_volume_ptr = mp->mnt_stat.f_mntfromname;
402 /* Count number of volumes separated by ':' */
403 for (char *p = next_volume_ptr; *p != '\0'; ++p) {
409 mp->mnt_flag &= ~MNT_RDONLY; /* mount R/W */
411 if ((error = copyin(data, &info, sizeof(info))) != 0)
416 * updating or new mount
418 if (mp->mnt_flag & MNT_UPDATE) {
419 hmp = (void *)mp->mnt_data;
420 KKASSERT(hmp != NULL);
422 if (info.nvolumes <= 0 || info.nvolumes >= 32768)
428 * master-id validation. The master id may not be changed by a
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");
437 master_id = info.master_id;
438 if (master_id < -1 || master_id >= HAMMER_MAX_MASTERS)
442 master_id = hmp->master_id;
448 * Internal mount data structure
451 hmp = kmalloc(sizeof(*hmp), M_HAMMER, M_WAITOK | M_ZERO);
452 mp->mnt_data = (qaddr_t)hmp;
454 /*TAILQ_INIT(&hmp->recycle_list);*/
457 * Make sure kmalloc type limits are set appropriately.
459 * Our inode kmalloc group is sized based on maxvnodes
460 * (controlled by the system, not us).
462 kmalloc_create(&hmp->m_misc, "HAMMER-others");
463 kmalloc_create(&hmp->m_inodes, "HAMMER-inodes");
465 kmalloc_raise_limit(hmp->m_inodes, 0); /* unlimited */
467 hmp->root_btree_beg.localization = 0x00000000U;
468 hmp->root_btree_beg.obj_id = -0x8000000000000000LL;
469 hmp->root_btree_beg.key = -0x8000000000000000LL;
470 hmp->root_btree_beg.create_tid = 1;
471 hmp->root_btree_beg.delete_tid = 1;
472 hmp->root_btree_beg.rec_type = 0;
473 hmp->root_btree_beg.obj_type = 0;
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;
483 hmp->krate.freq = 1; /* maximum reporting rate (hz) */
484 hmp->krate.count = -16; /* initial burst */
486 hmp->sync_lock.refs = 1;
487 hmp->free_lock.refs = 1;
488 hmp->undo_lock.refs = 1;
489 hmp->blkmap_lock.refs = 1;
490 hmp->snapshot_lock.refs = 1;
491 hmp->volume_lock.refs = 1;
493 TAILQ_INIT(&hmp->delay_list);
494 TAILQ_INIT(&hmp->flush_group_list);
495 TAILQ_INIT(&hmp->objid_cache_list);
496 TAILQ_INIT(&hmp->undo_lru_list);
497 TAILQ_INIT(&hmp->reclaim_list);
499 RB_INIT(&hmp->rb_dedup_crc_root);
500 RB_INIT(&hmp->rb_dedup_off_root);
501 TAILQ_INIT(&hmp->dedup_lru_list);
503 hmp->hflags &= ~HMNT_USERFLAGS;
504 hmp->hflags |= info.hflags & HMNT_USERFLAGS;
506 hmp->master_id = master_id;
509 mp->mnt_flag |= MNT_RDONLY;
510 hmp->asof = info.asof;
512 hmp->asof = HAMMER_MAX_TID;
515 hmp->volume_to_remove = -1;
518 * Re-open read-write if originally read-only, or vise-versa.
520 * When going from read-only to read-write execute the stage2
521 * recovery if it has not already been run.
523 if (mp->mnt_flag & MNT_UPDATE) {
524 lwkt_gettoken(&hmp->fs_token);
526 if (hmp->ronly && (mp->mnt_kern_flag & MNTK_WANTRDWR)) {
527 kprintf("HAMMER read-only -> read-write\n");
529 RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
530 hammer_adjust_volume_mode, NULL);
531 rootvol = hammer_get_root_volume(hmp, &error);
533 hammer_recover_flush_buffers(hmp, rootvol, 1);
534 error = hammer_recover_stage2(hmp, rootvol);
535 bcopy(rootvol->ondisk->vol0_blockmap,
537 sizeof(hmp->blockmap));
538 hammer_rel_volume(rootvol, 0);
540 RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
541 hammer_reload_inode, NULL);
542 /* kernel clears MNT_RDONLY */
543 } else if (hmp->ronly == 0 && (mp->mnt_flag & MNT_RDONLY)) {
544 kprintf("HAMMER read-write -> read-only\n");
545 hmp->ronly = 1; /* messy */
546 RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
547 hammer_reload_inode, NULL);
549 hammer_flusher_sync(hmp);
550 hammer_flusher_sync(hmp);
551 hammer_flusher_sync(hmp);
553 RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
554 hammer_adjust_volume_mode, NULL);
556 lwkt_reltoken(&hmp->fs_token);
560 RB_INIT(&hmp->rb_vols_root);
561 RB_INIT(&hmp->rb_inos_root);
562 RB_INIT(&hmp->rb_redo_root);
563 RB_INIT(&hmp->rb_nods_root);
564 RB_INIT(&hmp->rb_undo_root);
565 RB_INIT(&hmp->rb_resv_root);
566 RB_INIT(&hmp->rb_bufs_root);
567 RB_INIT(&hmp->rb_pfsm_root);
569 hmp->ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
571 RB_INIT(&hmp->volu_root);
572 RB_INIT(&hmp->undo_root);
573 RB_INIT(&hmp->data_root);
574 RB_INIT(&hmp->meta_root);
575 RB_INIT(&hmp->lose_root);
576 TAILQ_INIT(&hmp->iorun_list);
578 lwkt_token_init(&hmp->fs_token, "hammerfs");
579 lwkt_token_init(&hmp->io_token, "hammerio");
581 lwkt_gettoken(&hmp->fs_token);
586 path = objcache_get(namei_oc, M_WAITOK);
588 for (i = 0; i < info.nvolumes; ++i) {
593 KKASSERT(next_volume_ptr != NULL);
595 if (*next_volume_ptr != '/') {
597 strcpy(path, "/dev/");
600 for (k = strlen(path); k < MAXPATHLEN-1; ++k) {
601 if (*next_volume_ptr == '\0') {
603 } else if (*next_volume_ptr == ':') {
607 path[k] = *next_volume_ptr;
614 cdev_t dev = kgetdiskbyname(path);
615 error = bdevvp(dev, &devvp);
617 kprintf("hammer_mountroot: can't find devvp\n");
620 error = copyin(&info.volumes[i], &upath,
623 error = copyinstr(upath, path,
627 error = hammer_install_volume(hmp, path, devvp);
631 objcache_put(namei_oc, path);
634 * Make sure we found a root volume
636 if (error == 0 && hmp->rootvol == NULL) {
637 kprintf("hammer_mount: No root volume found!\n");
642 * Check that all required volumes are available
644 if (error == 0 && hammer_mountcheck_volumes(hmp)) {
645 kprintf("hammer_mount: Missing volumes, cannot mount!\n");
650 /* called with fs_token held */
656 * No errors, setup enough of the mount point so we can lookup the
659 mp->mnt_iosize_max = MAXPHYS;
660 mp->mnt_kern_flag |= MNTK_FSMID;
663 * MPSAFE code. Note that VOPs and VFSops which are not MPSAFE
664 * will acquire a per-mount token prior to entry and release it
665 * on return, so even if we do not specify it we no longer get
666 * the BGL regardlless of how we are flagged.
668 mp->mnt_kern_flag |= MNTK_ALL_MPSAFE;
669 /*MNTK_RD_MPSAFE | MNTK_GA_MPSAFE | MNTK_IN_MPSAFE;*/
672 * note: f_iosize is used by vnode_pager_haspage() when constructing
675 mp->mnt_stat.f_iosize = HAMMER_BUFSIZE;
676 mp->mnt_stat.f_bsize = HAMMER_BUFSIZE;
678 mp->mnt_vstat.f_frsize = HAMMER_BUFSIZE;
679 mp->mnt_vstat.f_bsize = HAMMER_BUFSIZE;
681 mp->mnt_maxsymlinklen = 255;
682 mp->mnt_flag |= MNT_LOCAL;
684 vfs_add_vnodeops(mp, &hammer_vnode_vops, &mp->mnt_vn_norm_ops);
685 vfs_add_vnodeops(mp, &hammer_spec_vops, &mp->mnt_vn_spec_ops);
686 vfs_add_vnodeops(mp, &hammer_fifo_vops, &mp->mnt_vn_fifo_ops);
689 * The root volume's ondisk pointer is only valid if we hold a
692 rootvol = hammer_get_root_volume(hmp, &error);
697 * Perform any necessary UNDO operations. The recovery code does
698 * call hammer_undo_lookup() so we have to pre-cache the blockmap,
699 * and then re-copy it again after recovery is complete.
701 * If this is a read-only mount the UNDO information is retained
702 * in memory in the form of dirty buffer cache buffers, and not
703 * written back to the media.
705 bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
706 sizeof(hmp->blockmap));
709 * Check filesystem version
711 hmp->version = rootvol->ondisk->vol_version;
712 if (hmp->version < HAMMER_VOL_VERSION_MIN ||
713 hmp->version > HAMMER_VOL_VERSION_MAX) {
714 kprintf("HAMMER: mount unsupported fs version %d\n",
721 * The undo_rec_limit limits the size of flush groups to avoid
722 * blowing out the UNDO FIFO. This calculation is typically in
723 * the tens of thousands and is designed primarily when small
724 * HAMMER filesystems are created.
726 hmp->undo_rec_limit = hammer_undo_max(hmp) / 8192 + 100;
727 if (hammer_debug_general & 0x0001)
728 kprintf("HAMMER: undo_rec_limit %d\n", hmp->undo_rec_limit);
731 * NOTE: Recover stage1 not only handles meta-data recovery, it
732 * also sets hmp->undo_seqno for HAMMER VERSION 4+ filesystems.
734 error = hammer_recover_stage1(hmp, rootvol);
736 kprintf("Failed to recover HAMMER filesystem on mount\n");
741 * Finish setup now that we have a good root volume.
743 * The top 16 bits of fsid.val[1] is a pfs id.
745 ksnprintf(mp->mnt_stat.f_mntfromname,
746 sizeof(mp->mnt_stat.f_mntfromname), "%s",
747 rootvol->ondisk->vol_name);
748 mp->mnt_stat.f_fsid.val[0] =
749 crc32((char *)&rootvol->ondisk->vol_fsid + 0, 8);
750 mp->mnt_stat.f_fsid.val[1] =
751 crc32((char *)&rootvol->ondisk->vol_fsid + 8, 8);
752 mp->mnt_stat.f_fsid.val[1] &= 0x0000FFFF;
754 mp->mnt_vstat.f_fsid_uuid = rootvol->ondisk->vol_fsid;
755 mp->mnt_vstat.f_fsid = crc32(&mp->mnt_vstat.f_fsid_uuid,
756 sizeof(mp->mnt_vstat.f_fsid_uuid));
759 * Certain often-modified fields in the root volume are cached in
760 * the hammer_mount structure so we do not have to generate lots
761 * of little UNDO structures for them.
763 * Recopy after recovery. This also has the side effect of
764 * setting our cached undo FIFO's first_offset, which serves to
765 * placemark the FIFO start for the NEXT flush cycle while the
766 * on-disk first_offset represents the LAST flush cycle.
768 hmp->next_tid = rootvol->ondisk->vol0_next_tid;
769 hmp->flush_tid1 = hmp->next_tid;
770 hmp->flush_tid2 = hmp->next_tid;
771 bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
772 sizeof(hmp->blockmap));
773 hmp->copy_stat_freebigblocks = rootvol->ondisk->vol0_stat_freebigblocks;
775 hammer_flusher_create(hmp);
778 * Locate the root directory using the root cluster's B-Tree as a
779 * starting point. The root directory uses an obj_id of 1.
781 * FUTURE: Leave the root directory cached referenced but unlocked
782 * in hmp->rootvp (need to flush it on unmount).
784 error = hammer_vfs_vget(mp, NULL, 1, &rootvp);
788 /*vn_unlock(hmp->rootvp);*/
790 error = hammer_recover_stage2(hmp, rootvol);
793 * If the stage2 recovery fails be sure to clean out all cached
794 * vnodes before throwing away the mount structure or bad things
801 if ((mp->mnt_flag & MNT_UPDATE) == 0) {
804 /* Populate info for mount point (NULL pad)*/
805 bzero(mp->mnt_stat.f_mntonname, MNAMELEN);
808 copyinstr(mntpt, mp->mnt_stat.f_mntonname,
810 } else { /* Root mount */
811 mp->mnt_stat.f_mntonname[0] = '/';
814 (void)VFS_STATFS(mp, &mp->mnt_stat, cred);
815 hammer_rel_volume(rootvol, 0);
818 * Cleanup and return.
821 /* called with fs_token held */
824 lwkt_reltoken(&hmp->fs_token);
830 hammer_vfs_unmount(struct mount *mp, int mntflags)
832 hammer_mount_t hmp = (void *)mp->mnt_data;
837 * Clean out the vnodes
839 lwkt_gettoken(&hmp->fs_token);
841 if (mntflags & MNT_FORCE)
843 error = vflush(mp, 0, flags);
846 * Clean up the internal mount structure and related entities. This
850 /* called with fs_token held */
853 lwkt_reltoken(&hmp->fs_token);
859 * Clean up the internal mount structure and disassociate it from the mount.
860 * This may issue I/O.
862 * Called with fs_token held.
865 hammer_free_hmp(struct mount *mp)
867 hammer_mount_t hmp = (void *)mp->mnt_data;
868 hammer_flush_group_t flg;
873 * Flush anything dirty. This won't even run if the
874 * filesystem errored-out.
877 while (hammer_flusher_haswork(hmp)) {
878 hammer_flusher_sync(hmp);
882 kprintf("HAMMER: umount flushing.");
885 tsleep(&dummy, 0, "hmrufl", hz);
888 kprintf("giving up\n");
892 if (count >= 5 && count < 30)
896 * If the mount had a critical error we have to destroy any
897 * remaining inodes before we can finish cleaning up the flusher.
899 if (hmp->flags & HAMMER_MOUNT_CRITICAL_ERROR) {
900 RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
901 hammer_destroy_inode_callback, NULL);
905 * There shouldn't be any inodes left now and any left over
906 * flush groups should now be empty.
908 KKASSERT(RB_EMPTY(&hmp->rb_inos_root));
909 while ((flg = TAILQ_FIRST(&hmp->flush_group_list)) != NULL) {
910 TAILQ_REMOVE(&hmp->flush_group_list, flg, flush_entry);
911 KKASSERT(RB_EMPTY(&flg->flush_tree));
913 kprintf("HAMMER: Warning, flush_group %p was "
914 "not empty on umount!\n", flg);
916 kfree(flg, hmp->m_misc);
920 * We can finally destroy the flusher
922 hammer_flusher_destroy(hmp);
925 * We may have held recovered buffers due to a read-only mount.
926 * These must be discarded.
929 hammer_recover_flush_buffers(hmp, NULL, -1);
932 * Unload buffers and then volumes
934 RB_SCAN(hammer_buf_rb_tree, &hmp->rb_bufs_root, NULL,
935 hammer_unload_buffer, NULL);
936 RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
937 hammer_unload_volume, NULL);
940 mp->mnt_flag &= ~MNT_LOCAL;
942 hammer_destroy_objid_cache(hmp);
943 hammer_destroy_dedup_cache(hmp);
944 if (hmp->dedup_free_cache != NULL) {
945 kfree(hmp->dedup_free_cache, hmp->m_misc);
946 hmp->dedup_free_cache = NULL;
948 kmalloc_destroy(&hmp->m_misc);
949 kmalloc_destroy(&hmp->m_inodes);
950 lwkt_reltoken(&hmp->fs_token);
951 kfree(hmp, M_HAMMER);
955 * Report critical errors. ip may be NULL.
958 hammer_critical_error(hammer_mount_t hmp, hammer_inode_t ip,
959 int error, const char *msg)
961 hmp->flags |= HAMMER_MOUNT_CRITICAL_ERROR;
963 krateprintf(&hmp->krate,
964 "HAMMER(%s): Critical error inode=%jd error=%d %s\n",
965 hmp->mp->mnt_stat.f_mntfromname,
966 (intmax_t)(ip ? ip->obj_id : -1),
969 if (hmp->ronly == 0) {
970 hmp->ronly = 2; /* special errored read-only mode */
971 hmp->mp->mnt_flag |= MNT_RDONLY;
972 RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
973 hammer_adjust_volume_mode, NULL);
974 kprintf("HAMMER(%s): Forcing read-only mode\n",
975 hmp->mp->mnt_stat.f_mntfromname);
978 if (hammer_debug_critical)
979 Debugger("Entering debugger");
984 * Obtain a vnode for the specified inode number. An exclusively locked
988 hammer_vfs_vget(struct mount *mp, struct vnode *dvp,
989 ino_t ino, struct vnode **vpp)
991 struct hammer_transaction trans;
992 struct hammer_mount *hmp = (void *)mp->mnt_data;
993 struct hammer_inode *ip;
995 u_int32_t localization;
997 lwkt_gettoken(&hmp->fs_token);
998 hammer_simple_transaction(&trans, hmp);
1001 * If a directory vnode is supplied (mainly NFS) then we can acquire
1002 * the PFS domain from it. Otherwise we would only be able to vget
1003 * inodes in the root PFS.
1006 localization = HAMMER_DEF_LOCALIZATION +
1007 VTOI(dvp)->obj_localization;
1009 localization = HAMMER_DEF_LOCALIZATION;
1013 * Lookup the requested HAMMER inode. The structure must be
1014 * left unlocked while we manipulate the related vnode to avoid
1017 ip = hammer_get_inode(&trans, NULL, ino,
1018 hmp->asof, localization,
1023 error = hammer_get_vnode(ip, vpp);
1024 hammer_rel_inode(ip, 0);
1026 hammer_done_transaction(&trans);
1027 lwkt_reltoken(&hmp->fs_token);
1032 * Return the root vnode for the filesystem.
1034 * HAMMER stores the root vnode in the hammer_mount structure so
1035 * getting it is easy.
1038 hammer_vfs_root(struct mount *mp, struct vnode **vpp)
1042 error = hammer_vfs_vget(mp, NULL, 1, vpp);
1047 hammer_vfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
1049 struct hammer_mount *hmp = (void *)mp->mnt_data;
1050 hammer_volume_t volume;
1051 hammer_volume_ondisk_t ondisk;
1056 lwkt_gettoken(&hmp->fs_token);
1057 volume = hammer_get_root_volume(hmp, &error);
1059 lwkt_reltoken(&hmp->fs_token);
1062 ondisk = volume->ondisk;
1067 _hammer_checkspace(hmp, HAMMER_CHKSPC_WRITE, &breserved);
1068 mp->mnt_stat.f_files = ondisk->vol0_stat_inodes;
1069 bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE;
1070 hammer_rel_volume(volume, 0);
1072 mp->mnt_stat.f_bfree = (bfree - breserved) / HAMMER_BUFSIZE;
1073 mp->mnt_stat.f_bavail = mp->mnt_stat.f_bfree;
1074 if (mp->mnt_stat.f_files < 0)
1075 mp->mnt_stat.f_files = 0;
1077 *sbp = mp->mnt_stat;
1078 lwkt_reltoken(&hmp->fs_token);
1083 hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred)
1085 struct hammer_mount *hmp = (void *)mp->mnt_data;
1086 hammer_volume_t volume;
1087 hammer_volume_ondisk_t ondisk;
1092 lwkt_gettoken(&hmp->fs_token);
1093 volume = hammer_get_root_volume(hmp, &error);
1095 lwkt_reltoken(&hmp->fs_token);
1098 ondisk = volume->ondisk;
1103 _hammer_checkspace(hmp, HAMMER_CHKSPC_WRITE, &breserved);
1104 mp->mnt_vstat.f_files = ondisk->vol0_stat_inodes;
1105 bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE;
1106 hammer_rel_volume(volume, 0);
1108 mp->mnt_vstat.f_bfree = (bfree - breserved) / HAMMER_BUFSIZE;
1109 mp->mnt_vstat.f_bavail = mp->mnt_vstat.f_bfree;
1110 if (mp->mnt_vstat.f_files < 0)
1111 mp->mnt_vstat.f_files = 0;
1112 *sbp = mp->mnt_vstat;
1113 lwkt_reltoken(&hmp->fs_token);
1118 * Sync the filesystem. Currently we have to run it twice, the second
1119 * one will advance the undo start index to the end index, so if a crash
1120 * occurs no undos will be run on mount.
1122 * We do not sync the filesystem if we are called from a panic. If we did
1123 * we might end up blowing up a sync that was already in progress.
1126 hammer_vfs_sync(struct mount *mp, int waitfor)
1128 struct hammer_mount *hmp = (void *)mp->mnt_data;
1131 lwkt_gettoken(&hmp->fs_token);
1132 if (panicstr == NULL) {
1133 error = hammer_sync_hmp(hmp, waitfor);
1137 lwkt_reltoken(&hmp->fs_token);
1142 * Convert a vnode to a file handle.
1144 * Accesses read-only fields on already-referenced structures so
1145 * no token is needed.
1148 hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp)
1152 KKASSERT(MAXFIDSZ >= 16);
1154 fhp->fid_len = offsetof(struct fid, fid_data[16]);
1155 fhp->fid_ext = ip->obj_localization >> 16;
1156 bcopy(&ip->obj_id, fhp->fid_data + 0, sizeof(ip->obj_id));
1157 bcopy(&ip->obj_asof, fhp->fid_data + 8, sizeof(ip->obj_asof));
1163 * Convert a file handle back to a vnode.
1165 * Use rootvp to enforce PFS isolation when a PFS is exported via a
1169 hammer_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
1170 struct fid *fhp, struct vnode **vpp)
1172 hammer_mount_t hmp = (void *)mp->mnt_data;
1173 struct hammer_transaction trans;
1174 struct hammer_inode *ip;
1175 struct hammer_inode_info info;
1177 u_int32_t localization;
1179 bcopy(fhp->fid_data + 0, &info.obj_id, sizeof(info.obj_id));
1180 bcopy(fhp->fid_data + 8, &info.obj_asof, sizeof(info.obj_asof));
1182 localization = VTOI(rootvp)->obj_localization;
1184 localization = (u_int32_t)fhp->fid_ext << 16;
1186 lwkt_gettoken(&hmp->fs_token);
1187 hammer_simple_transaction(&trans, hmp);
1190 * Get/allocate the hammer_inode structure. The structure must be
1191 * unlocked while we manipulate the related vnode to avoid a
1194 ip = hammer_get_inode(&trans, NULL, info.obj_id,
1195 info.obj_asof, localization, 0, &error);
1197 error = hammer_get_vnode(ip, vpp);
1198 hammer_rel_inode(ip, 0);
1202 hammer_done_transaction(&trans);
1203 lwkt_reltoken(&hmp->fs_token);
1208 hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
1209 int *exflagsp, struct ucred **credanonp)
1211 hammer_mount_t hmp = (void *)mp->mnt_data;
1215 lwkt_gettoken(&hmp->fs_token);
1216 np = vfs_export_lookup(mp, &hmp->export, nam);
1218 *exflagsp = np->netc_exflags;
1219 *credanonp = &np->netc_anon;
1224 lwkt_reltoken(&hmp->fs_token);
1230 hammer_vfs_export(struct mount *mp, int op, const struct export_args *export)
1232 hammer_mount_t hmp = (void *)mp->mnt_data;
1235 lwkt_gettoken(&hmp->fs_token);
1238 case MOUNTCTL_SET_EXPORT:
1239 error = vfs_export(mp, &hmp->export, export);
1245 lwkt_reltoken(&hmp->fs_token);