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_live_dedup = 0;
67 int hammer_count_fsyncs;
68 int hammer_count_inodes;
69 int hammer_count_iqueued;
70 int hammer_count_reclaiming;
71 int hammer_count_records;
72 int hammer_count_record_datas;
73 int hammer_count_volumes;
74 int hammer_count_buffers;
75 int hammer_count_nodes;
76 int64_t hammer_count_extra_space_used;
77 int64_t hammer_stats_btree_lookups;
78 int64_t hammer_stats_btree_searches;
79 int64_t hammer_stats_btree_inserts;
80 int64_t hammer_stats_btree_deletes;
81 int64_t hammer_stats_btree_elements;
82 int64_t hammer_stats_btree_splits;
83 int64_t hammer_stats_btree_iterations;
84 int64_t hammer_stats_btree_root_iterations;
85 int64_t hammer_stats_record_iterations;
87 int64_t hammer_stats_file_read;
88 int64_t hammer_stats_file_write;
89 int64_t hammer_stats_file_iopsr;
90 int64_t hammer_stats_file_iopsw;
91 int64_t hammer_stats_disk_read;
92 int64_t hammer_stats_disk_write;
93 int64_t hammer_stats_inode_flushes;
94 int64_t hammer_stats_commits;
95 int64_t hammer_stats_undo;
96 int64_t hammer_stats_redo;
98 int hammer_count_dirtybufspace; /* global */
99 int hammer_count_refedbufs; /* global */
100 int hammer_count_reservations;
101 int hammer_count_io_running_read;
102 int hammer_count_io_running_write;
103 int hammer_count_io_locked;
104 int hammer_limit_dirtybufspace; /* per-mount */
105 int hammer_limit_running_io; /* per-mount */
106 int hammer_limit_recs; /* as a whole XXX */
107 int hammer_limit_inode_recs = 1024; /* per inode */
108 int hammer_limit_reclaim = HAMMER_RECLAIM_WAIT;
109 int hammer_live_dedup_cache_size = DEDUP_CACHE_SIZE;
110 int hammer_limit_redo = 4096 * 1024; /* per inode */
111 int hammer_autoflush = 2000; /* auto flush */
112 int hammer_bio_count;
113 int hammer_verify_zone;
114 int hammer_verify_data = 1;
115 int hammer_write_mode;
116 int hammer_yield_check = 16;
117 int hammer_fsync_mode = 3;
118 int64_t hammer_contention_count;
119 int64_t hammer_zone_limit;
122 * Live dedup debug counters (sysctls are writable so that counters
123 * can be reset from userspace).
125 int64_t hammer_live_dedup_vnode_bcmps = 0;
126 int64_t hammer_live_dedup_device_bcmps = 0;
127 int64_t hammer_live_dedup_findblk_failures = 0;
128 int64_t hammer_live_dedup_bmap_saves = 0;
131 SYSCTL_NODE(_vfs, OID_AUTO, hammer, CTLFLAG_RW, 0, "HAMMER filesystem");
133 SYSCTL_INT(_vfs_hammer, OID_AUTO, supported_version, CTLFLAG_RD,
134 &hammer_supported_version, 0, "");
135 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_general, CTLFLAG_RW,
136 &hammer_debug_general, 0, "");
137 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_io, CTLFLAG_RW,
138 &hammer_debug_io, 0, "");
139 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_debug, CTLFLAG_RW,
140 &hammer_debug_debug, 0, "");
141 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_inode, CTLFLAG_RW,
142 &hammer_debug_inode, 0, "");
143 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_locks, CTLFLAG_RW,
144 &hammer_debug_locks, 0, "");
145 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_btree, CTLFLAG_RW,
146 &hammer_debug_btree, 0, "");
147 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_tid, CTLFLAG_RW,
148 &hammer_debug_tid, 0, "");
149 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover, CTLFLAG_RW,
150 &hammer_debug_recover, 0, "");
151 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover_faults, CTLFLAG_RW,
152 &hammer_debug_recover_faults, 0, "");
153 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_critical, CTLFLAG_RW,
154 &hammer_debug_critical, 0, "");
155 SYSCTL_INT(_vfs_hammer, OID_AUTO, cluster_enable, CTLFLAG_RW,
156 &hammer_cluster_enable, 0, "");
158 * 0 - live dedup is disabled
159 * 1 - dedup cache is populated on reads only
160 * 2 - dedup cache is populated on both reads and writes
162 SYSCTL_INT(_vfs_hammer, OID_AUTO, live_dedup, CTLFLAG_RW,
163 &hammer_live_dedup, 0, "Enable live dedup");
165 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_dirtybufspace, CTLFLAG_RW,
166 &hammer_limit_dirtybufspace, 0, "");
167 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_running_io, CTLFLAG_RW,
168 &hammer_limit_running_io, 0, "");
169 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_recs, CTLFLAG_RW,
170 &hammer_limit_recs, 0, "");
171 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_inode_recs, CTLFLAG_RW,
172 &hammer_limit_inode_recs, 0, "");
173 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_reclaim, CTLFLAG_RW,
174 &hammer_limit_reclaim, 0, "");
175 SYSCTL_INT(_vfs_hammer, OID_AUTO, live_dedup_cache_size, CTLFLAG_RW,
176 &hammer_live_dedup_cache_size, 0, "");
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_reclaiming, CTLFLAG_RD,
187 &hammer_count_reclaiming, 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_INT(_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_INT(_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_INT(_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, yield_check, CTLFLAG_RW,
279 &hammer_yield_check, 0, "");
280 SYSCTL_INT(_vfs_hammer, OID_AUTO, fsync_mode, CTLFLAG_RW,
281 &hammer_fsync_mode, 0, "");
283 KTR_INFO_MASTER(hammer);
288 static void hammer_free_hmp(struct mount *mp);
290 static int hammer_vfs_mount(struct mount *mp, char *path, caddr_t data,
292 static int hammer_vfs_unmount(struct mount *mp, int mntflags);
293 static int hammer_vfs_root(struct mount *mp, struct vnode **vpp);
294 static int hammer_vfs_statfs(struct mount *mp, struct statfs *sbp,
296 static int hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp,
298 static int hammer_vfs_sync(struct mount *mp, int waitfor);
299 static int hammer_vfs_vget(struct mount *mp, struct vnode *dvp,
300 ino_t ino, struct vnode **vpp);
301 static int hammer_vfs_init(struct vfsconf *conf);
302 static int hammer_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
303 struct fid *fhp, struct vnode **vpp);
304 static int hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp);
305 static int hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
306 int *exflagsp, struct ucred **credanonp);
309 static struct vfsops hammer_vfsops = {
310 .vfs_mount = hammer_vfs_mount,
311 .vfs_unmount = hammer_vfs_unmount,
312 .vfs_root = hammer_vfs_root,
313 .vfs_statfs = hammer_vfs_statfs,
314 .vfs_statvfs = hammer_vfs_statvfs,
315 .vfs_sync = hammer_vfs_sync,
316 .vfs_vget = hammer_vfs_vget,
317 .vfs_init = hammer_vfs_init,
318 .vfs_vptofh = hammer_vfs_vptofh,
319 .vfs_fhtovp = hammer_vfs_fhtovp,
320 .vfs_checkexp = hammer_vfs_checkexp
323 MALLOC_DEFINE(M_HAMMER, "HAMMER-mount", "");
325 VFS_SET(hammer_vfsops, hammer, 0);
326 MODULE_VERSION(hammer, 1);
329 hammer_vfs_init(struct vfsconf *conf)
333 if (hammer_limit_recs == 0) {
334 hammer_limit_recs = nbuf * 25;
335 n = kmalloc_limit(M_HAMMER) / 512;
336 if (hammer_limit_recs > n)
337 hammer_limit_recs = n;
339 if (hammer_limit_dirtybufspace == 0) {
340 hammer_limit_dirtybufspace = hidirtybufspace / 2;
341 if (hammer_limit_dirtybufspace < 100)
342 hammer_limit_dirtybufspace = 100;
346 * Set reasonable limits to maintain an I/O pipeline. This is
347 * used by the flush code which explicitly initiates I/O, and
350 * The system-driven buffer cache uses vfs.lorunningspace and
351 * vfs.hirunningspace globally.
353 if (hammer_limit_running_io == 0)
354 hammer_limit_running_io = hammer_limit_dirtybufspace;
355 if (hammer_limit_running_io > 10 * 1024 * 1024)
356 hammer_limit_running_io = 10 * 1024 * 1024;
361 hammer_vfs_mount(struct mount *mp, char *mntpt, caddr_t data,
364 struct hammer_mount_info info;
366 hammer_volume_t rootvol;
367 struct vnode *rootvp;
368 struct vnode *devvp = NULL;
369 const char *upath; /* volume name in userspace */
370 char *path; /* volume name in system space */
374 char *next_volume_ptr = NULL;
377 * Accept hammer_mount_info. mntpt is NULL for root mounts at boot.
380 bzero(&info, sizeof(info));
385 next_volume_ptr = mp->mnt_stat.f_mntfromname;
387 /* Count number of volumes separated by ':' */
388 for (char *p = next_volume_ptr; *p != '\0'; ++p) {
394 mp->mnt_flag &= ~MNT_RDONLY; /* mount R/W */
396 if ((error = copyin(data, &info, sizeof(info))) != 0)
401 * updating or new mount
403 if (mp->mnt_flag & MNT_UPDATE) {
404 hmp = (void *)mp->mnt_data;
405 KKASSERT(hmp != NULL);
407 if (info.nvolumes <= 0 || info.nvolumes >= 32768)
413 * master-id validation. The master id may not be changed by a
416 if (info.hflags & HMNT_MASTERID) {
417 if (hmp && hmp->master_id != info.master_id) {
418 kprintf("hammer: cannot change master id "
419 "with mount update\n");
422 master_id = info.master_id;
423 if (master_id < -1 || master_id >= HAMMER_MAX_MASTERS)
427 master_id = hmp->master_id;
433 * Internal mount data structure
436 hmp = kmalloc(sizeof(*hmp), M_HAMMER, M_WAITOK | M_ZERO);
437 mp->mnt_data = (qaddr_t)hmp;
439 /*TAILQ_INIT(&hmp->recycle_list);*/
442 * Make sure kmalloc type limits are set appropriately.
444 * Our inode kmalloc group is sized based on maxvnodes
445 * (controlled by the system, not us).
447 kmalloc_create(&hmp->m_misc, "HAMMER-others");
448 kmalloc_create(&hmp->m_inodes, "HAMMER-inodes");
450 kmalloc_raise_limit(hmp->m_inodes, 0); /* unlimited */
452 hmp->root_btree_beg.localization = 0x00000000U;
453 hmp->root_btree_beg.obj_id = -0x8000000000000000LL;
454 hmp->root_btree_beg.key = -0x8000000000000000LL;
455 hmp->root_btree_beg.create_tid = 1;
456 hmp->root_btree_beg.delete_tid = 1;
457 hmp->root_btree_beg.rec_type = 0;
458 hmp->root_btree_beg.obj_type = 0;
460 hmp->root_btree_end.localization = 0xFFFFFFFFU;
461 hmp->root_btree_end.obj_id = 0x7FFFFFFFFFFFFFFFLL;
462 hmp->root_btree_end.key = 0x7FFFFFFFFFFFFFFFLL;
463 hmp->root_btree_end.create_tid = 0xFFFFFFFFFFFFFFFFULL;
464 hmp->root_btree_end.delete_tid = 0; /* special case */
465 hmp->root_btree_end.rec_type = 0xFFFFU;
466 hmp->root_btree_end.obj_type = 0;
468 hmp->krate.freq = 1; /* maximum reporting rate (hz) */
469 hmp->krate.count = -16; /* initial burst */
471 hmp->sync_lock.refs = 1;
472 hmp->free_lock.refs = 1;
473 hmp->undo_lock.refs = 1;
474 hmp->blkmap_lock.refs = 1;
475 hmp->snapshot_lock.refs = 1;
476 hmp->volume_lock.refs = 1;
478 TAILQ_INIT(&hmp->delay_list);
479 TAILQ_INIT(&hmp->flush_group_list);
480 TAILQ_INIT(&hmp->objid_cache_list);
481 TAILQ_INIT(&hmp->undo_lru_list);
482 TAILQ_INIT(&hmp->reclaim_list);
484 RB_INIT(&hmp->rb_dedup_crc_root);
485 RB_INIT(&hmp->rb_dedup_off_root);
486 TAILQ_INIT(&hmp->dedup_lru_list);
488 hmp->hflags &= ~HMNT_USERFLAGS;
489 hmp->hflags |= info.hflags & HMNT_USERFLAGS;
491 hmp->master_id = master_id;
494 mp->mnt_flag |= MNT_RDONLY;
495 hmp->asof = info.asof;
497 hmp->asof = HAMMER_MAX_TID;
500 hmp->volume_to_remove = -1;
503 * Re-open read-write if originally read-only, or vise-versa.
505 * When going from read-only to read-write execute the stage2
506 * recovery if it has not already been run.
508 if (mp->mnt_flag & MNT_UPDATE) {
509 lwkt_gettoken(&hmp->fs_token);
511 if (hmp->ronly && (mp->mnt_kern_flag & MNTK_WANTRDWR)) {
512 kprintf("HAMMER read-only -> read-write\n");
514 RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
515 hammer_adjust_volume_mode, NULL);
516 rootvol = hammer_get_root_volume(hmp, &error);
518 hammer_recover_flush_buffers(hmp, rootvol, 1);
519 error = hammer_recover_stage2(hmp, rootvol);
520 bcopy(rootvol->ondisk->vol0_blockmap,
522 sizeof(hmp->blockmap));
523 hammer_rel_volume(rootvol, 0);
525 RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
526 hammer_reload_inode, NULL);
527 /* kernel clears MNT_RDONLY */
528 } else if (hmp->ronly == 0 && (mp->mnt_flag & MNT_RDONLY)) {
529 kprintf("HAMMER read-write -> read-only\n");
530 hmp->ronly = 1; /* messy */
531 RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
532 hammer_reload_inode, NULL);
534 hammer_flusher_sync(hmp);
535 hammer_flusher_sync(hmp);
536 hammer_flusher_sync(hmp);
538 RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
539 hammer_adjust_volume_mode, NULL);
541 lwkt_reltoken(&hmp->fs_token);
545 RB_INIT(&hmp->rb_vols_root);
546 RB_INIT(&hmp->rb_inos_root);
547 RB_INIT(&hmp->rb_redo_root);
548 RB_INIT(&hmp->rb_nods_root);
549 RB_INIT(&hmp->rb_undo_root);
550 RB_INIT(&hmp->rb_resv_root);
551 RB_INIT(&hmp->rb_bufs_root);
552 RB_INIT(&hmp->rb_pfsm_root);
554 hmp->ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
556 TAILQ_INIT(&hmp->volu_list);
557 TAILQ_INIT(&hmp->undo_list);
558 TAILQ_INIT(&hmp->data_list);
559 TAILQ_INIT(&hmp->meta_list);
560 TAILQ_INIT(&hmp->lose_list);
561 TAILQ_INIT(&hmp->iorun_list);
563 lwkt_token_init(&hmp->fs_token, 1, "hammerfs");
564 lwkt_token_init(&hmp->io_token, 1, "hammerio");
566 lwkt_gettoken(&hmp->fs_token);
571 path = objcache_get(namei_oc, M_WAITOK);
573 for (i = 0; i < info.nvolumes; ++i) {
578 KKASSERT(next_volume_ptr != NULL);
580 if (*next_volume_ptr != '/') {
582 strcpy(path, "/dev/");
585 for (k = strlen(path); k < MAXPATHLEN-1; ++k) {
586 if (*next_volume_ptr == '\0') {
588 } else if (*next_volume_ptr == ':') {
592 path[k] = *next_volume_ptr;
599 cdev_t dev = kgetdiskbyname(path);
600 error = bdevvp(dev, &devvp);
602 kprintf("hammer_mountroot: can't find devvp\n");
605 error = copyin(&info.volumes[i], &upath,
608 error = copyinstr(upath, path,
612 error = hammer_install_volume(hmp, path, devvp);
616 objcache_put(namei_oc, path);
619 * Make sure we found a root volume
621 if (error == 0 && hmp->rootvol == NULL) {
622 kprintf("hammer_mount: No root volume found!\n");
627 * Check that all required volumes are available
629 if (error == 0 && hammer_mountcheck_volumes(hmp)) {
630 kprintf("hammer_mount: Missing volumes, cannot mount!\n");
635 /* called with fs_token held */
641 * No errors, setup enough of the mount point so we can lookup the
644 mp->mnt_iosize_max = MAXPHYS;
645 mp->mnt_kern_flag |= MNTK_FSMID;
648 * MPSAFE code. Note that VOPs and VFSops which are not MPSAFE
649 * will acquire a per-mount token prior to entry and release it
650 * on return, so even if we do not specify it we no longer get
651 * the BGL regardlless of how we are flagged.
653 mp->mnt_kern_flag |= MNTK_ALL_MPSAFE;
654 /*MNTK_RD_MPSAFE | MNTK_GA_MPSAFE | MNTK_IN_MPSAFE;*/
657 * note: f_iosize is used by vnode_pager_haspage() when constructing
660 mp->mnt_stat.f_iosize = HAMMER_BUFSIZE;
661 mp->mnt_stat.f_bsize = HAMMER_BUFSIZE;
663 mp->mnt_vstat.f_frsize = HAMMER_BUFSIZE;
664 mp->mnt_vstat.f_bsize = HAMMER_BUFSIZE;
666 mp->mnt_maxsymlinklen = 255;
667 mp->mnt_flag |= MNT_LOCAL;
669 vfs_add_vnodeops(mp, &hammer_vnode_vops, &mp->mnt_vn_norm_ops);
670 vfs_add_vnodeops(mp, &hammer_spec_vops, &mp->mnt_vn_spec_ops);
671 vfs_add_vnodeops(mp, &hammer_fifo_vops, &mp->mnt_vn_fifo_ops);
674 * The root volume's ondisk pointer is only valid if we hold a
677 rootvol = hammer_get_root_volume(hmp, &error);
682 * Perform any necessary UNDO operations. The recovery code does
683 * call hammer_undo_lookup() so we have to pre-cache the blockmap,
684 * and then re-copy it again after recovery is complete.
686 * If this is a read-only mount the UNDO information is retained
687 * in memory in the form of dirty buffer cache buffers, and not
688 * written back to the media.
690 bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
691 sizeof(hmp->blockmap));
694 * Check filesystem version
696 hmp->version = rootvol->ondisk->vol_version;
697 if (hmp->version < HAMMER_VOL_VERSION_MIN ||
698 hmp->version > HAMMER_VOL_VERSION_MAX) {
699 kprintf("HAMMER: mount unsupported fs version %d\n",
706 * The undo_rec_limit limits the size of flush groups to avoid
707 * blowing out the UNDO FIFO. This calculation is typically in
708 * the tens of thousands and is designed primarily when small
709 * HAMMER filesystems are created.
711 hmp->undo_rec_limit = hammer_undo_max(hmp) / 8192 + 100;
712 if (hammer_debug_general & 0x0001)
713 kprintf("HAMMER: undo_rec_limit %d\n", hmp->undo_rec_limit);
716 * NOTE: Recover stage1 not only handles meta-data recovery, it
717 * also sets hmp->undo_seqno for HAMMER VERSION 4+ filesystems.
719 error = hammer_recover_stage1(hmp, rootvol);
721 kprintf("Failed to recover HAMMER filesystem on mount\n");
726 * Finish setup now that we have a good root volume.
728 * The top 16 bits of fsid.val[1] is a pfs id.
730 ksnprintf(mp->mnt_stat.f_mntfromname,
731 sizeof(mp->mnt_stat.f_mntfromname), "%s",
732 rootvol->ondisk->vol_name);
733 mp->mnt_stat.f_fsid.val[0] =
734 crc32((char *)&rootvol->ondisk->vol_fsid + 0, 8);
735 mp->mnt_stat.f_fsid.val[1] =
736 crc32((char *)&rootvol->ondisk->vol_fsid + 8, 8);
737 mp->mnt_stat.f_fsid.val[1] &= 0x0000FFFF;
739 mp->mnt_vstat.f_fsid_uuid = rootvol->ondisk->vol_fsid;
740 mp->mnt_vstat.f_fsid = crc32(&mp->mnt_vstat.f_fsid_uuid,
741 sizeof(mp->mnt_vstat.f_fsid_uuid));
744 * Certain often-modified fields in the root volume are cached in
745 * the hammer_mount structure so we do not have to generate lots
746 * of little UNDO structures for them.
748 * Recopy after recovery. This also has the side effect of
749 * setting our cached undo FIFO's first_offset, which serves to
750 * placemark the FIFO start for the NEXT flush cycle while the
751 * on-disk first_offset represents the LAST flush cycle.
753 hmp->next_tid = rootvol->ondisk->vol0_next_tid;
754 hmp->flush_tid1 = hmp->next_tid;
755 hmp->flush_tid2 = hmp->next_tid;
756 bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
757 sizeof(hmp->blockmap));
758 hmp->copy_stat_freebigblocks = rootvol->ondisk->vol0_stat_freebigblocks;
760 hammer_flusher_create(hmp);
763 * Locate the root directory using the root cluster's B-Tree as a
764 * starting point. The root directory uses an obj_id of 1.
766 * FUTURE: Leave the root directory cached referenced but unlocked
767 * in hmp->rootvp (need to flush it on unmount).
769 error = hammer_vfs_vget(mp, NULL, 1, &rootvp);
773 /*vn_unlock(hmp->rootvp);*/
775 error = hammer_recover_stage2(hmp, rootvol);
778 * If the stage2 recovery fails be sure to clean out all cached
779 * vnodes before throwing away the mount structure or bad things
786 hammer_rel_volume(rootvol, 0);
789 * Cleanup and return.
792 /* called with fs_token held */
795 lwkt_reltoken(&hmp->fs_token);
801 hammer_vfs_unmount(struct mount *mp, int mntflags)
803 hammer_mount_t hmp = (void *)mp->mnt_data;
808 * Clean out the vnodes
810 lwkt_gettoken(&hmp->fs_token);
812 if (mntflags & MNT_FORCE)
814 error = vflush(mp, 0, flags);
817 * Clean up the internal mount structure and related entities. This
821 /* called with fs_token held */
824 lwkt_reltoken(&hmp->fs_token);
830 * Clean up the internal mount structure and disassociate it from the mount.
831 * This may issue I/O.
833 * Called with fs_token held.
836 hammer_free_hmp(struct mount *mp)
838 hammer_mount_t hmp = (void *)mp->mnt_data;
839 hammer_flush_group_t flg;
844 * Flush anything dirty. This won't even run if the
845 * filesystem errored-out.
848 while (hammer_flusher_haswork(hmp)) {
849 hammer_flusher_sync(hmp);
853 kprintf("HAMMER: umount flushing.");
856 tsleep(&dummy, 0, "hmrufl", hz);
859 kprintf("giving up\n");
863 if (count >= 5 && count < 30)
867 * If the mount had a critical error we have to destroy any
868 * remaining inodes before we can finish cleaning up the flusher.
870 if (hmp->flags & HAMMER_MOUNT_CRITICAL_ERROR) {
871 RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
872 hammer_destroy_inode_callback, NULL);
876 * There shouldn't be any inodes left now and any left over
877 * flush groups should now be empty.
879 KKASSERT(RB_EMPTY(&hmp->rb_inos_root));
880 while ((flg = TAILQ_FIRST(&hmp->flush_group_list)) != NULL) {
881 TAILQ_REMOVE(&hmp->flush_group_list, flg, flush_entry);
882 KKASSERT(RB_EMPTY(&flg->flush_tree));
884 kprintf("HAMMER: Warning, flush_group %p was "
885 "not empty on umount!\n", flg);
887 kfree(flg, hmp->m_misc);
891 * We can finally destroy the flusher
893 hammer_flusher_destroy(hmp);
896 * We may have held recovered buffers due to a read-only mount.
897 * These must be discarded.
900 hammer_recover_flush_buffers(hmp, NULL, -1);
903 * Unload buffers and then volumes
905 RB_SCAN(hammer_buf_rb_tree, &hmp->rb_bufs_root, NULL,
906 hammer_unload_buffer, NULL);
907 RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
908 hammer_unload_volume, NULL);
911 mp->mnt_flag &= ~MNT_LOCAL;
913 hammer_destroy_objid_cache(hmp);
914 hammer_destroy_dedup_cache(hmp);
915 if (hmp->dedup_free_cache != NULL) {
916 kfree(hmp->dedup_free_cache, hmp->m_misc);
917 hmp->dedup_free_cache = NULL;
919 kmalloc_destroy(&hmp->m_misc);
920 kmalloc_destroy(&hmp->m_inodes);
921 lwkt_reltoken(&hmp->fs_token);
922 kfree(hmp, M_HAMMER);
926 * Report critical errors. ip may be NULL.
929 hammer_critical_error(hammer_mount_t hmp, hammer_inode_t ip,
930 int error, const char *msg)
932 hmp->flags |= HAMMER_MOUNT_CRITICAL_ERROR;
934 krateprintf(&hmp->krate,
935 "HAMMER(%s): Critical error inode=%jd error=%d %s\n",
936 hmp->mp->mnt_stat.f_mntfromname,
937 (intmax_t)(ip ? ip->obj_id : -1),
940 if (hmp->ronly == 0) {
941 hmp->ronly = 2; /* special errored read-only mode */
942 hmp->mp->mnt_flag |= MNT_RDONLY;
943 kprintf("HAMMER(%s): Forcing read-only mode\n",
944 hmp->mp->mnt_stat.f_mntfromname);
947 if (hammer_debug_critical)
948 Debugger("Entering debugger");
953 * Obtain a vnode for the specified inode number. An exclusively locked
957 hammer_vfs_vget(struct mount *mp, struct vnode *dvp,
958 ino_t ino, struct vnode **vpp)
960 struct hammer_transaction trans;
961 struct hammer_mount *hmp = (void *)mp->mnt_data;
962 struct hammer_inode *ip;
964 u_int32_t localization;
966 lwkt_gettoken(&hmp->fs_token);
967 hammer_simple_transaction(&trans, hmp);
970 * If a directory vnode is supplied (mainly NFS) then we can acquire
971 * the PFS domain from it. Otherwise we would only be able to vget
972 * inodes in the root PFS.
975 localization = HAMMER_DEF_LOCALIZATION +
976 VTOI(dvp)->obj_localization;
978 localization = HAMMER_DEF_LOCALIZATION;
982 * Lookup the requested HAMMER inode. The structure must be
983 * left unlocked while we manipulate the related vnode to avoid
986 ip = hammer_get_inode(&trans, NULL, ino,
987 hmp->asof, localization,
992 error = hammer_get_vnode(ip, vpp);
993 hammer_rel_inode(ip, 0);
995 hammer_done_transaction(&trans);
996 lwkt_reltoken(&hmp->fs_token);
1001 * Return the root vnode for the filesystem.
1003 * HAMMER stores the root vnode in the hammer_mount structure so
1004 * getting it is easy.
1007 hammer_vfs_root(struct mount *mp, struct vnode **vpp)
1011 error = hammer_vfs_vget(mp, NULL, 1, vpp);
1016 hammer_vfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
1018 struct hammer_mount *hmp = (void *)mp->mnt_data;
1019 hammer_volume_t volume;
1020 hammer_volume_ondisk_t ondisk;
1025 lwkt_gettoken(&hmp->fs_token);
1026 volume = hammer_get_root_volume(hmp, &error);
1028 lwkt_reltoken(&hmp->fs_token);
1031 ondisk = volume->ondisk;
1036 _hammer_checkspace(hmp, HAMMER_CHKSPC_WRITE, &breserved);
1037 mp->mnt_stat.f_files = ondisk->vol0_stat_inodes;
1038 bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE;
1039 hammer_rel_volume(volume, 0);
1041 mp->mnt_stat.f_bfree = (bfree - breserved) / HAMMER_BUFSIZE;
1042 mp->mnt_stat.f_bavail = mp->mnt_stat.f_bfree;
1043 if (mp->mnt_stat.f_files < 0)
1044 mp->mnt_stat.f_files = 0;
1046 *sbp = mp->mnt_stat;
1047 lwkt_reltoken(&hmp->fs_token);
1052 hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred)
1054 struct hammer_mount *hmp = (void *)mp->mnt_data;
1055 hammer_volume_t volume;
1056 hammer_volume_ondisk_t ondisk;
1061 lwkt_gettoken(&hmp->fs_token);
1062 volume = hammer_get_root_volume(hmp, &error);
1064 lwkt_reltoken(&hmp->fs_token);
1067 ondisk = volume->ondisk;
1072 _hammer_checkspace(hmp, HAMMER_CHKSPC_WRITE, &breserved);
1073 mp->mnt_vstat.f_files = ondisk->vol0_stat_inodes;
1074 bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE;
1075 hammer_rel_volume(volume, 0);
1077 mp->mnt_vstat.f_bfree = (bfree - breserved) / HAMMER_BUFSIZE;
1078 mp->mnt_vstat.f_bavail = mp->mnt_vstat.f_bfree;
1079 if (mp->mnt_vstat.f_files < 0)
1080 mp->mnt_vstat.f_files = 0;
1081 *sbp = mp->mnt_vstat;
1082 lwkt_reltoken(&hmp->fs_token);
1087 * Sync the filesystem. Currently we have to run it twice, the second
1088 * one will advance the undo start index to the end index, so if a crash
1089 * occurs no undos will be run on mount.
1091 * We do not sync the filesystem if we are called from a panic. If we did
1092 * we might end up blowing up a sync that was already in progress.
1095 hammer_vfs_sync(struct mount *mp, int waitfor)
1097 struct hammer_mount *hmp = (void *)mp->mnt_data;
1100 lwkt_gettoken(&hmp->fs_token);
1101 if (panicstr == NULL) {
1102 error = hammer_sync_hmp(hmp, waitfor);
1106 lwkt_reltoken(&hmp->fs_token);
1111 * Convert a vnode to a file handle.
1113 * Accesses read-only fields on already-referenced structures so
1114 * no token is needed.
1117 hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp)
1121 KKASSERT(MAXFIDSZ >= 16);
1123 fhp->fid_len = offsetof(struct fid, fid_data[16]);
1124 fhp->fid_ext = ip->obj_localization >> 16;
1125 bcopy(&ip->obj_id, fhp->fid_data + 0, sizeof(ip->obj_id));
1126 bcopy(&ip->obj_asof, fhp->fid_data + 8, sizeof(ip->obj_asof));
1132 * Convert a file handle back to a vnode.
1134 * Use rootvp to enforce PFS isolation when a PFS is exported via a
1138 hammer_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
1139 struct fid *fhp, struct vnode **vpp)
1141 hammer_mount_t hmp = (void *)mp->mnt_data;
1142 struct hammer_transaction trans;
1143 struct hammer_inode *ip;
1144 struct hammer_inode_info info;
1146 u_int32_t localization;
1148 bcopy(fhp->fid_data + 0, &info.obj_id, sizeof(info.obj_id));
1149 bcopy(fhp->fid_data + 8, &info.obj_asof, sizeof(info.obj_asof));
1151 localization = VTOI(rootvp)->obj_localization;
1153 localization = (u_int32_t)fhp->fid_ext << 16;
1155 lwkt_gettoken(&hmp->fs_token);
1156 hammer_simple_transaction(&trans, hmp);
1159 * Get/allocate the hammer_inode structure. The structure must be
1160 * unlocked while we manipulate the related vnode to avoid a
1163 ip = hammer_get_inode(&trans, NULL, info.obj_id,
1164 info.obj_asof, localization, 0, &error);
1166 error = hammer_get_vnode(ip, vpp);
1167 hammer_rel_inode(ip, 0);
1171 hammer_done_transaction(&trans);
1172 lwkt_reltoken(&hmp->fs_token);
1177 hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
1178 int *exflagsp, struct ucred **credanonp)
1180 hammer_mount_t hmp = (void *)mp->mnt_data;
1184 lwkt_gettoken(&hmp->fs_token);
1185 np = vfs_export_lookup(mp, &hmp->export, nam);
1187 *exflagsp = np->netc_exflags;
1188 *credanonp = &np->netc_anon;
1193 lwkt_reltoken(&hmp->fs_token);
1199 hammer_vfs_export(struct mount *mp, int op, const struct export_args *export)
1201 hammer_mount_t hmp = (void *)mp->mnt_data;
1204 lwkt_gettoken(&hmp->fs_token);
1207 case MOUNTCTL_SET_EXPORT:
1208 error = vfs_export(mp, &hmp->export, export);
1214 lwkt_reltoken(&hmp->fs_token);