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 int hammer_count_dirtybufspace; /* global */
98 int hammer_count_refedbufs; /* global */
99 int hammer_count_reservations;
100 int hammer_count_io_running_read;
101 int hammer_count_io_running_write;
102 int hammer_count_io_locked;
103 int hammer_limit_dirtybufspace; /* per-mount */
104 int hammer_limit_running_io; /* per-mount */
105 int hammer_limit_recs; /* as a whole XXX */
106 int hammer_limit_inode_recs = 2048; /* per inode */
107 int hammer_limit_reclaims;
108 int hammer_live_dedup_cache_size = DEDUP_CACHE_SIZE;
109 int hammer_limit_redo = 4096 * 1024; /* per inode */
110 int hammer_autoflush = 500; /* auto flush (typ on reclaim) */
111 int hammer_bio_count;
112 int hammer_verify_zone;
113 int hammer_verify_data = 1;
114 int hammer_write_mode;
115 int hammer_double_buffer;
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");
164 SYSCTL_INT(_vfs_hammer, OID_AUTO, tdmux_ticks, CTLFLAG_RW,
165 &hammer_tdmux_ticks, 0, "Hammer tdmux ticks");
167 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_dirtybufspace, CTLFLAG_RW,
168 &hammer_limit_dirtybufspace, 0, "");
169 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_running_io, CTLFLAG_RW,
170 &hammer_limit_running_io, 0, "");
171 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_recs, CTLFLAG_RW,
172 &hammer_limit_recs, 0, "");
173 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_inode_recs, CTLFLAG_RW,
174 &hammer_limit_inode_recs, 0, "");
175 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_reclaims, CTLFLAG_RW,
176 &hammer_limit_reclaims, 0, "");
177 SYSCTL_INT(_vfs_hammer, OID_AUTO, live_dedup_cache_size, CTLFLAG_RW,
178 &hammer_live_dedup_cache_size, 0,
179 "Number of cache entries");
180 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_redo, CTLFLAG_RW,
181 &hammer_limit_redo, 0, "");
183 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_fsyncs, CTLFLAG_RD,
184 &hammer_count_fsyncs, 0, "");
185 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_inodes, CTLFLAG_RD,
186 &hammer_count_inodes, 0, "");
187 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_iqueued, CTLFLAG_RD,
188 &hammer_count_iqueued, 0, "");
189 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reclaims, CTLFLAG_RD,
190 &hammer_count_reclaims, 0, "");
191 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_records, CTLFLAG_RD,
192 &hammer_count_records, 0, "");
193 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_record_datas, CTLFLAG_RD,
194 &hammer_count_record_datas, 0, "");
195 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_volumes, CTLFLAG_RD,
196 &hammer_count_volumes, 0, "");
197 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_buffers, CTLFLAG_RD,
198 &hammer_count_buffers, 0, "");
199 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_nodes, CTLFLAG_RD,
200 &hammer_count_nodes, 0, "");
201 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, count_extra_space_used, CTLFLAG_RD,
202 &hammer_count_extra_space_used, 0, "");
204 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_searches, CTLFLAG_RD,
205 &hammer_stats_btree_searches, 0, "");
206 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_lookups, CTLFLAG_RD,
207 &hammer_stats_btree_lookups, 0, "");
208 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_inserts, CTLFLAG_RD,
209 &hammer_stats_btree_inserts, 0, "");
210 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_deletes, CTLFLAG_RD,
211 &hammer_stats_btree_deletes, 0, "");
212 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_elements, CTLFLAG_RD,
213 &hammer_stats_btree_elements, 0, "");
214 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_splits, CTLFLAG_RD,
215 &hammer_stats_btree_splits, 0, "");
216 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_iterations, CTLFLAG_RD,
217 &hammer_stats_btree_iterations, 0, "");
218 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_root_iterations, CTLFLAG_RD,
219 &hammer_stats_btree_root_iterations, 0, "");
220 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_record_iterations, CTLFLAG_RD,
221 &hammer_stats_record_iterations, 0, "");
223 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_read, CTLFLAG_RD,
224 &hammer_stats_file_read, 0, "");
225 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_write, CTLFLAG_RD,
226 &hammer_stats_file_write, 0, "");
227 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_iopsr, CTLFLAG_RD,
228 &hammer_stats_file_iopsr, 0, "");
229 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_iopsw, CTLFLAG_RD,
230 &hammer_stats_file_iopsw, 0, "");
231 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_disk_read, CTLFLAG_RD,
232 &hammer_stats_disk_read, 0, "");
233 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_disk_write, CTLFLAG_RD,
234 &hammer_stats_disk_write, 0, "");
235 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_inode_flushes, CTLFLAG_RD,
236 &hammer_stats_inode_flushes, 0, "");
237 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_commits, CTLFLAG_RD,
238 &hammer_stats_commits, 0, "");
239 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_undo, CTLFLAG_RD,
240 &hammer_stats_undo, 0, "");
241 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_redo, CTLFLAG_RD,
242 &hammer_stats_redo, 0, "");
244 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, live_dedup_vnode_bcmps, CTLFLAG_RW,
245 &hammer_live_dedup_vnode_bcmps, 0,
246 "successful vnode buffer comparisons");
247 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, live_dedup_device_bcmps, CTLFLAG_RW,
248 &hammer_live_dedup_device_bcmps, 0,
249 "successful device buffer comparisons");
250 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, live_dedup_findblk_failures, CTLFLAG_RW,
251 &hammer_live_dedup_findblk_failures, 0,
252 "block lookup failures for comparison");
253 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, live_dedup_bmap_saves, CTLFLAG_RW,
254 &hammer_live_dedup_bmap_saves, 0,
255 "useful physical block lookups");
257 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_dirtybufspace, CTLFLAG_RD,
258 &hammer_count_dirtybufspace, 0, "");
259 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_refedbufs, CTLFLAG_RD,
260 &hammer_count_refedbufs, 0, "");
261 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reservations, CTLFLAG_RD,
262 &hammer_count_reservations, 0, "");
263 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_running_read, CTLFLAG_RD,
264 &hammer_count_io_running_read, 0, "");
265 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_locked, CTLFLAG_RD,
266 &hammer_count_io_locked, 0, "");
267 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_running_write, CTLFLAG_RD,
268 &hammer_count_io_running_write, 0, "");
269 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, zone_limit, CTLFLAG_RW,
270 &hammer_zone_limit, 0, "");
271 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, contention_count, CTLFLAG_RW,
272 &hammer_contention_count, 0, "");
273 SYSCTL_INT(_vfs_hammer, OID_AUTO, autoflush, CTLFLAG_RW,
274 &hammer_autoflush, 0, "");
275 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_zone, CTLFLAG_RW,
276 &hammer_verify_zone, 0, "");
277 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_data, CTLFLAG_RW,
278 &hammer_verify_data, 0, "");
279 SYSCTL_INT(_vfs_hammer, OID_AUTO, write_mode, CTLFLAG_RW,
280 &hammer_write_mode, 0, "");
281 SYSCTL_INT(_vfs_hammer, OID_AUTO, double_buffer, CTLFLAG_RW,
282 &hammer_double_buffer, 0, "");
283 SYSCTL_INT(_vfs_hammer, OID_AUTO, yield_check, CTLFLAG_RW,
284 &hammer_yield_check, 0, "");
285 SYSCTL_INT(_vfs_hammer, OID_AUTO, fsync_mode, CTLFLAG_RW,
286 &hammer_fsync_mode, 0, "");
288 /* KTR_INFO_MASTER(hammer); */
293 static void hammer_free_hmp(struct mount *mp);
295 static int hammer_vfs_mount(struct mount *mp, char *path, caddr_t data,
297 static int hammer_vfs_unmount(struct mount *mp, int mntflags);
298 static int hammer_vfs_root(struct mount *mp, struct vnode **vpp);
299 static int hammer_vfs_statfs(struct mount *mp, struct statfs *sbp,
301 static int hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp,
303 static int hammer_vfs_sync(struct mount *mp, int waitfor);
304 static int hammer_vfs_vget(struct mount *mp, struct vnode *dvp,
305 ino_t ino, struct vnode **vpp);
306 static int hammer_vfs_init(struct vfsconf *conf);
307 static int hammer_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
308 struct fid *fhp, struct vnode **vpp);
309 static int hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp);
310 static int hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
311 int *exflagsp, struct ucred **credanonp);
314 static struct vfsops hammer_vfsops = {
315 .vfs_mount = hammer_vfs_mount,
316 .vfs_unmount = hammer_vfs_unmount,
317 .vfs_root = hammer_vfs_root,
318 .vfs_statfs = hammer_vfs_statfs,
319 .vfs_statvfs = hammer_vfs_statvfs,
320 .vfs_sync = hammer_vfs_sync,
321 .vfs_vget = hammer_vfs_vget,
322 .vfs_init = hammer_vfs_init,
323 .vfs_vptofh = hammer_vfs_vptofh,
324 .vfs_fhtovp = hammer_vfs_fhtovp,
325 .vfs_checkexp = hammer_vfs_checkexp
328 MALLOC_DEFINE(M_HAMMER, "HAMMER-mount", "");
330 VFS_SET(hammer_vfsops, hammer, 0);
331 MODULE_VERSION(hammer, 1);
334 hammer_vfs_init(struct vfsconf *conf)
339 * Wait up to this long for an exclusive deadlock to clear
340 * before acquiring a new shared lock on the ip. The deadlock
341 * may have occured on a b-tree node related to the ip.
343 if (hammer_tdmux_ticks == 0)
344 hammer_tdmux_ticks = hz / 5;
349 if (hammer_limit_recs == 0) {
350 hammer_limit_recs = nbuf * 25;
351 n = kmalloc_limit(M_HAMMER) / 512;
352 if (hammer_limit_recs > n)
353 hammer_limit_recs = n;
355 if (hammer_limit_dirtybufspace == 0) {
356 hammer_limit_dirtybufspace = hidirtybufspace / 2;
357 if (hammer_limit_dirtybufspace < 100)
358 hammer_limit_dirtybufspace = 100;
362 * Set reasonable limits to maintain an I/O pipeline. This is
363 * used by the flush code which explicitly initiates I/O, and
366 * The system-driven buffer cache uses vfs.lorunningspace and
367 * vfs.hirunningspace globally.
369 if (hammer_limit_running_io == 0)
370 hammer_limit_running_io = hammer_limit_dirtybufspace;
372 if (hammer_limit_running_io > 10 * 1024 * 1024)
373 hammer_limit_running_io = 10 * 1024 * 1024;
376 * The hammer_inode structure detaches from the vnode on reclaim.
377 * This limits the number of inodes in this state to prevent a
378 * memory pool blowout.
380 if (hammer_limit_reclaims == 0)
381 hammer_limit_reclaims = desiredvnodes / 10;
387 hammer_vfs_mount(struct mount *mp, char *mntpt, caddr_t data,
390 struct hammer_mount_info info;
392 hammer_volume_t rootvol;
393 struct vnode *rootvp;
394 struct vnode *devvp = NULL;
395 const char *upath; /* volume name in userspace */
396 char *path; /* volume name in system space */
400 char *next_volume_ptr = NULL;
403 * Accept hammer_mount_info. mntpt is NULL for root mounts at boot.
406 bzero(&info, sizeof(info));
411 next_volume_ptr = mp->mnt_stat.f_mntfromname;
413 /* Count number of volumes separated by ':' */
414 for (char *p = next_volume_ptr; *p != '\0'; ++p) {
420 mp->mnt_flag &= ~MNT_RDONLY; /* mount R/W */
422 if ((error = copyin(data, &info, sizeof(info))) != 0)
427 * updating or new mount
429 if (mp->mnt_flag & MNT_UPDATE) {
430 hmp = (void *)mp->mnt_data;
431 KKASSERT(hmp != NULL);
433 if (info.nvolumes <= 0 || info.nvolumes >= 32768)
439 * master-id validation. The master id may not be changed by a
442 if (info.hflags & HMNT_MASTERID) {
443 if (hmp && hmp->master_id != info.master_id) {
444 kprintf("hammer: cannot change master id "
445 "with mount update\n");
448 master_id = info.master_id;
449 if (master_id < -1 || master_id >= HAMMER_MAX_MASTERS)
453 master_id = hmp->master_id;
459 * Internal mount data structure
462 hmp = kmalloc(sizeof(*hmp), M_HAMMER, M_WAITOK | M_ZERO);
463 mp->mnt_data = (qaddr_t)hmp;
465 /*TAILQ_INIT(&hmp->recycle_list);*/
468 * Make sure kmalloc type limits are set appropriately.
470 * Our inode kmalloc group is sized based on maxvnodes
471 * (controlled by the system, not us).
473 kmalloc_create(&hmp->m_misc, "HAMMER-others");
474 kmalloc_create(&hmp->m_inodes, "HAMMER-inodes");
476 kmalloc_raise_limit(hmp->m_inodes, 0); /* unlimited */
478 hmp->root_btree_beg.localization = 0x00000000U;
479 hmp->root_btree_beg.obj_id = -0x8000000000000000LL;
480 hmp->root_btree_beg.key = -0x8000000000000000LL;
481 hmp->root_btree_beg.create_tid = 1;
482 hmp->root_btree_beg.delete_tid = 1;
483 hmp->root_btree_beg.rec_type = 0;
484 hmp->root_btree_beg.obj_type = 0;
486 hmp->root_btree_end.localization = 0xFFFFFFFFU;
487 hmp->root_btree_end.obj_id = 0x7FFFFFFFFFFFFFFFLL;
488 hmp->root_btree_end.key = 0x7FFFFFFFFFFFFFFFLL;
489 hmp->root_btree_end.create_tid = 0xFFFFFFFFFFFFFFFFULL;
490 hmp->root_btree_end.delete_tid = 0; /* special case */
491 hmp->root_btree_end.rec_type = 0xFFFFU;
492 hmp->root_btree_end.obj_type = 0;
494 hmp->krate.freq = 1; /* maximum reporting rate (hz) */
495 hmp->krate.count = -16; /* initial burst */
497 hmp->sync_lock.refs = 1;
498 hmp->free_lock.refs = 1;
499 hmp->undo_lock.refs = 1;
500 hmp->blkmap_lock.refs = 1;
501 hmp->snapshot_lock.refs = 1;
502 hmp->volume_lock.refs = 1;
504 TAILQ_INIT(&hmp->delay_list);
505 TAILQ_INIT(&hmp->flush_group_list);
506 TAILQ_INIT(&hmp->objid_cache_list);
507 TAILQ_INIT(&hmp->undo_lru_list);
508 TAILQ_INIT(&hmp->reclaim_list);
510 RB_INIT(&hmp->rb_dedup_crc_root);
511 RB_INIT(&hmp->rb_dedup_off_root);
512 TAILQ_INIT(&hmp->dedup_lru_list);
514 hmp->hflags &= ~HMNT_USERFLAGS;
515 hmp->hflags |= info.hflags & HMNT_USERFLAGS;
517 hmp->master_id = master_id;
520 mp->mnt_flag |= MNT_RDONLY;
521 hmp->asof = info.asof;
523 hmp->asof = HAMMER_MAX_TID;
526 hmp->volume_to_remove = -1;
529 * Re-open read-write if originally read-only, or vise-versa.
531 * When going from read-only to read-write execute the stage2
532 * recovery if it has not already been run.
534 if (mp->mnt_flag & MNT_UPDATE) {
535 lwkt_gettoken(&hmp->fs_token);
537 if (hmp->ronly && (mp->mnt_kern_flag & MNTK_WANTRDWR)) {
538 kprintf("HAMMER read-only -> read-write\n");
540 RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
541 hammer_adjust_volume_mode, NULL);
542 rootvol = hammer_get_root_volume(hmp, &error);
544 hammer_recover_flush_buffers(hmp, rootvol, 1);
545 error = hammer_recover_stage2(hmp, rootvol);
546 bcopy(rootvol->ondisk->vol0_blockmap,
548 sizeof(hmp->blockmap));
549 hammer_rel_volume(rootvol, 0);
551 RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
552 hammer_reload_inode, NULL);
553 /* kernel clears MNT_RDONLY */
554 } else if (hmp->ronly == 0 && (mp->mnt_flag & MNT_RDONLY)) {
555 kprintf("HAMMER read-write -> read-only\n");
556 hmp->ronly = 1; /* messy */
557 RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
558 hammer_reload_inode, NULL);
560 hammer_flusher_sync(hmp);
561 hammer_flusher_sync(hmp);
562 hammer_flusher_sync(hmp);
564 RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
565 hammer_adjust_volume_mode, NULL);
567 lwkt_reltoken(&hmp->fs_token);
571 RB_INIT(&hmp->rb_vols_root);
572 RB_INIT(&hmp->rb_inos_root);
573 RB_INIT(&hmp->rb_redo_root);
574 RB_INIT(&hmp->rb_nods_root);
575 RB_INIT(&hmp->rb_undo_root);
576 RB_INIT(&hmp->rb_resv_root);
577 RB_INIT(&hmp->rb_bufs_root);
578 RB_INIT(&hmp->rb_pfsm_root);
580 hmp->ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
582 RB_INIT(&hmp->volu_root);
583 RB_INIT(&hmp->undo_root);
584 RB_INIT(&hmp->data_root);
585 RB_INIT(&hmp->meta_root);
586 RB_INIT(&hmp->lose_root);
587 TAILQ_INIT(&hmp->iorun_list);
589 lwkt_token_init(&hmp->fs_token, "hammerfs");
590 lwkt_token_init(&hmp->io_token, "hammerio");
592 lwkt_gettoken(&hmp->fs_token);
597 path = objcache_get(namei_oc, M_WAITOK);
599 for (i = 0; i < info.nvolumes; ++i) {
604 KKASSERT(next_volume_ptr != NULL);
606 if (*next_volume_ptr != '/') {
608 strcpy(path, "/dev/");
611 for (k = strlen(path); k < MAXPATHLEN-1; ++k) {
612 if (*next_volume_ptr == '\0') {
614 } else if (*next_volume_ptr == ':') {
618 path[k] = *next_volume_ptr;
625 cdev_t dev = kgetdiskbyname(path);
626 error = bdevvp(dev, &devvp);
628 kprintf("hammer_mountroot: can't find devvp\n");
631 error = copyin(&info.volumes[i], &upath,
634 error = copyinstr(upath, path,
638 error = hammer_install_volume(hmp, path, devvp);
642 objcache_put(namei_oc, path);
645 * Make sure we found a root volume
647 if (error == 0 && hmp->rootvol == NULL) {
648 kprintf("hammer_mount: No root volume found!\n");
653 * Check that all required volumes are available
655 if (error == 0 && hammer_mountcheck_volumes(hmp)) {
656 kprintf("hammer_mount: Missing volumes, cannot mount!\n");
661 /* called with fs_token held */
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;
674 * MPSAFE code. Note that VOPs and VFSops which are not MPSAFE
675 * will acquire a per-mount token prior to entry and release it
676 * on return, so even if we do not specify it we no longer get
677 * the BGL regardlless of how we are flagged.
679 mp->mnt_kern_flag |= MNTK_ALL_MPSAFE;
680 /*MNTK_RD_MPSAFE | MNTK_GA_MPSAFE | MNTK_IN_MPSAFE;*/
683 * note: f_iosize is used by vnode_pager_haspage() when constructing
686 mp->mnt_stat.f_iosize = HAMMER_BUFSIZE;
687 mp->mnt_stat.f_bsize = HAMMER_BUFSIZE;
689 mp->mnt_vstat.f_frsize = HAMMER_BUFSIZE;
690 mp->mnt_vstat.f_bsize = HAMMER_BUFSIZE;
692 mp->mnt_maxsymlinklen = 255;
693 mp->mnt_flag |= MNT_LOCAL;
695 vfs_add_vnodeops(mp, &hammer_vnode_vops, &mp->mnt_vn_norm_ops);
696 vfs_add_vnodeops(mp, &hammer_spec_vops, &mp->mnt_vn_spec_ops);
697 vfs_add_vnodeops(mp, &hammer_fifo_vops, &mp->mnt_vn_fifo_ops);
700 * The root volume's ondisk pointer is only valid if we hold a
703 rootvol = hammer_get_root_volume(hmp, &error);
708 * Perform any necessary UNDO operations. The recovery code does
709 * call hammer_undo_lookup() so we have to pre-cache the blockmap,
710 * and then re-copy it again after recovery is complete.
712 * If this is a read-only mount the UNDO information is retained
713 * in memory in the form of dirty buffer cache buffers, and not
714 * written back to the media.
716 bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
717 sizeof(hmp->blockmap));
720 * Check filesystem version
722 hmp->version = rootvol->ondisk->vol_version;
723 if (hmp->version < HAMMER_VOL_VERSION_MIN ||
724 hmp->version > HAMMER_VOL_VERSION_MAX) {
725 kprintf("HAMMER: mount unsupported fs version %d\n",
732 * The undo_rec_limit limits the size of flush groups to avoid
733 * blowing out the UNDO FIFO. This calculation is typically in
734 * the tens of thousands and is designed primarily when small
735 * HAMMER filesystems are created.
737 hmp->undo_rec_limit = hammer_undo_max(hmp) / 8192 + 100;
738 if (hammer_debug_general & 0x0001)
739 kprintf("HAMMER: undo_rec_limit %d\n", hmp->undo_rec_limit);
742 * NOTE: Recover stage1 not only handles meta-data recovery, it
743 * also sets hmp->undo_seqno for HAMMER VERSION 4+ filesystems.
745 error = hammer_recover_stage1(hmp, rootvol);
747 kprintf("Failed to recover HAMMER filesystem on mount\n");
752 * Finish setup now that we have a good root volume.
754 * The top 16 bits of fsid.val[1] is a pfs id.
756 ksnprintf(mp->mnt_stat.f_mntfromname,
757 sizeof(mp->mnt_stat.f_mntfromname), "%s",
758 rootvol->ondisk->vol_name);
759 mp->mnt_stat.f_fsid.val[0] =
760 crc32((char *)&rootvol->ondisk->vol_fsid + 0, 8);
761 mp->mnt_stat.f_fsid.val[1] =
762 crc32((char *)&rootvol->ondisk->vol_fsid + 8, 8);
763 mp->mnt_stat.f_fsid.val[1] &= 0x0000FFFF;
765 mp->mnt_vstat.f_fsid_uuid = rootvol->ondisk->vol_fsid;
766 mp->mnt_vstat.f_fsid = crc32(&mp->mnt_vstat.f_fsid_uuid,
767 sizeof(mp->mnt_vstat.f_fsid_uuid));
770 * Certain often-modified fields in the root volume are cached in
771 * the hammer_mount structure so we do not have to generate lots
772 * of little UNDO structures for them.
774 * Recopy after recovery. This also has the side effect of
775 * setting our cached undo FIFO's first_offset, which serves to
776 * placemark the FIFO start for the NEXT flush cycle while the
777 * on-disk first_offset represents the LAST flush cycle.
779 hmp->next_tid = rootvol->ondisk->vol0_next_tid;
780 hmp->flush_tid1 = hmp->next_tid;
781 hmp->flush_tid2 = hmp->next_tid;
782 bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
783 sizeof(hmp->blockmap));
784 hmp->copy_stat_freebigblocks = rootvol->ondisk->vol0_stat_freebigblocks;
786 hammer_flusher_create(hmp);
789 * Locate the root directory using the root cluster's B-Tree as a
790 * starting point. The root directory uses an obj_id of 1.
792 * FUTURE: Leave the root directory cached referenced but unlocked
793 * in hmp->rootvp (need to flush it on unmount).
795 error = hammer_vfs_vget(mp, NULL, 1, &rootvp);
799 /*vn_unlock(hmp->rootvp);*/
801 error = hammer_recover_stage2(hmp, rootvol);
804 * If the stage2 recovery fails be sure to clean out all cached
805 * vnodes before throwing away the mount structure or bad things
812 hammer_rel_volume(rootvol, 0);
815 * Cleanup and return.
818 /* called with fs_token held */
821 lwkt_reltoken(&hmp->fs_token);
827 hammer_vfs_unmount(struct mount *mp, int mntflags)
829 hammer_mount_t hmp = (void *)mp->mnt_data;
834 * Clean out the vnodes
836 lwkt_gettoken(&hmp->fs_token);
838 if (mntflags & MNT_FORCE)
840 error = vflush(mp, 0, flags);
843 * Clean up the internal mount structure and related entities. This
847 /* called with fs_token held */
850 lwkt_reltoken(&hmp->fs_token);
856 * Clean up the internal mount structure and disassociate it from the mount.
857 * This may issue I/O.
859 * Called with fs_token held.
862 hammer_free_hmp(struct mount *mp)
864 hammer_mount_t hmp = (void *)mp->mnt_data;
865 hammer_flush_group_t flg;
870 * Flush anything dirty. This won't even run if the
871 * filesystem errored-out.
874 while (hammer_flusher_haswork(hmp)) {
875 hammer_flusher_sync(hmp);
879 kprintf("HAMMER: umount flushing.");
882 tsleep(&dummy, 0, "hmrufl", hz);
885 kprintf("giving up\n");
889 if (count >= 5 && count < 30)
893 * If the mount had a critical error we have to destroy any
894 * remaining inodes before we can finish cleaning up the flusher.
896 if (hmp->flags & HAMMER_MOUNT_CRITICAL_ERROR) {
897 RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
898 hammer_destroy_inode_callback, NULL);
902 * There shouldn't be any inodes left now and any left over
903 * flush groups should now be empty.
905 KKASSERT(RB_EMPTY(&hmp->rb_inos_root));
906 while ((flg = TAILQ_FIRST(&hmp->flush_group_list)) != NULL) {
907 TAILQ_REMOVE(&hmp->flush_group_list, flg, flush_entry);
908 KKASSERT(RB_EMPTY(&flg->flush_tree));
910 kprintf("HAMMER: Warning, flush_group %p was "
911 "not empty on umount!\n", flg);
913 kfree(flg, hmp->m_misc);
917 * We can finally destroy the flusher
919 hammer_flusher_destroy(hmp);
922 * We may have held recovered buffers due to a read-only mount.
923 * These must be discarded.
926 hammer_recover_flush_buffers(hmp, NULL, -1);
929 * Unload buffers and then volumes
931 RB_SCAN(hammer_buf_rb_tree, &hmp->rb_bufs_root, NULL,
932 hammer_unload_buffer, NULL);
933 RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
934 hammer_unload_volume, NULL);
937 mp->mnt_flag &= ~MNT_LOCAL;
939 hammer_destroy_objid_cache(hmp);
940 hammer_destroy_dedup_cache(hmp);
941 if (hmp->dedup_free_cache != NULL) {
942 kfree(hmp->dedup_free_cache, hmp->m_misc);
943 hmp->dedup_free_cache = NULL;
945 kmalloc_destroy(&hmp->m_misc);
946 kmalloc_destroy(&hmp->m_inodes);
947 lwkt_reltoken(&hmp->fs_token);
948 kfree(hmp, M_HAMMER);
952 * Report critical errors. ip may be NULL.
955 hammer_critical_error(hammer_mount_t hmp, hammer_inode_t ip,
956 int error, const char *msg)
958 hmp->flags |= HAMMER_MOUNT_CRITICAL_ERROR;
960 krateprintf(&hmp->krate,
961 "HAMMER(%s): Critical error inode=%jd error=%d %s\n",
962 hmp->mp->mnt_stat.f_mntfromname,
963 (intmax_t)(ip ? ip->obj_id : -1),
966 if (hmp->ronly == 0) {
967 hmp->ronly = 2; /* special errored read-only mode */
968 hmp->mp->mnt_flag |= MNT_RDONLY;
969 RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
970 hammer_adjust_volume_mode, NULL);
971 kprintf("HAMMER(%s): Forcing read-only mode\n",
972 hmp->mp->mnt_stat.f_mntfromname);
975 if (hammer_debug_critical)
976 Debugger("Entering debugger");
981 * Obtain a vnode for the specified inode number. An exclusively locked
985 hammer_vfs_vget(struct mount *mp, struct vnode *dvp,
986 ino_t ino, struct vnode **vpp)
988 struct hammer_transaction trans;
989 struct hammer_mount *hmp = (void *)mp->mnt_data;
990 struct hammer_inode *ip;
992 u_int32_t localization;
994 lwkt_gettoken(&hmp->fs_token);
995 hammer_simple_transaction(&trans, hmp);
998 * If a directory vnode is supplied (mainly NFS) then we can acquire
999 * the PFS domain from it. Otherwise we would only be able to vget
1000 * inodes in the root PFS.
1003 localization = HAMMER_DEF_LOCALIZATION +
1004 VTOI(dvp)->obj_localization;
1006 localization = HAMMER_DEF_LOCALIZATION;
1010 * Lookup the requested HAMMER inode. The structure must be
1011 * left unlocked while we manipulate the related vnode to avoid
1014 ip = hammer_get_inode(&trans, NULL, ino,
1015 hmp->asof, localization,
1020 error = hammer_get_vnode(ip, vpp);
1021 hammer_rel_inode(ip, 0);
1023 hammer_done_transaction(&trans);
1024 lwkt_reltoken(&hmp->fs_token);
1029 * Return the root vnode for the filesystem.
1031 * HAMMER stores the root vnode in the hammer_mount structure so
1032 * getting it is easy.
1035 hammer_vfs_root(struct mount *mp, struct vnode **vpp)
1039 error = hammer_vfs_vget(mp, NULL, 1, vpp);
1044 hammer_vfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
1046 struct hammer_mount *hmp = (void *)mp->mnt_data;
1047 hammer_volume_t volume;
1048 hammer_volume_ondisk_t ondisk;
1053 lwkt_gettoken(&hmp->fs_token);
1054 volume = hammer_get_root_volume(hmp, &error);
1056 lwkt_reltoken(&hmp->fs_token);
1059 ondisk = volume->ondisk;
1064 _hammer_checkspace(hmp, HAMMER_CHKSPC_WRITE, &breserved);
1065 mp->mnt_stat.f_files = ondisk->vol0_stat_inodes;
1066 bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE;
1067 hammer_rel_volume(volume, 0);
1069 mp->mnt_stat.f_bfree = (bfree - breserved) / HAMMER_BUFSIZE;
1070 mp->mnt_stat.f_bavail = mp->mnt_stat.f_bfree;
1071 if (mp->mnt_stat.f_files < 0)
1072 mp->mnt_stat.f_files = 0;
1074 *sbp = mp->mnt_stat;
1075 lwkt_reltoken(&hmp->fs_token);
1080 hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred)
1082 struct hammer_mount *hmp = (void *)mp->mnt_data;
1083 hammer_volume_t volume;
1084 hammer_volume_ondisk_t ondisk;
1089 lwkt_gettoken(&hmp->fs_token);
1090 volume = hammer_get_root_volume(hmp, &error);
1092 lwkt_reltoken(&hmp->fs_token);
1095 ondisk = volume->ondisk;
1100 _hammer_checkspace(hmp, HAMMER_CHKSPC_WRITE, &breserved);
1101 mp->mnt_vstat.f_files = ondisk->vol0_stat_inodes;
1102 bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE;
1103 hammer_rel_volume(volume, 0);
1105 mp->mnt_vstat.f_bfree = (bfree - breserved) / HAMMER_BUFSIZE;
1106 mp->mnt_vstat.f_bavail = mp->mnt_vstat.f_bfree;
1107 if (mp->mnt_vstat.f_files < 0)
1108 mp->mnt_vstat.f_files = 0;
1109 *sbp = mp->mnt_vstat;
1110 lwkt_reltoken(&hmp->fs_token);
1115 * Sync the filesystem. Currently we have to run it twice, the second
1116 * one will advance the undo start index to the end index, so if a crash
1117 * occurs no undos will be run on mount.
1119 * We do not sync the filesystem if we are called from a panic. If we did
1120 * we might end up blowing up a sync that was already in progress.
1123 hammer_vfs_sync(struct mount *mp, int waitfor)
1125 struct hammer_mount *hmp = (void *)mp->mnt_data;
1128 lwkt_gettoken(&hmp->fs_token);
1129 if (panicstr == NULL) {
1130 error = hammer_sync_hmp(hmp, waitfor);
1134 lwkt_reltoken(&hmp->fs_token);
1139 * Convert a vnode to a file handle.
1141 * Accesses read-only fields on already-referenced structures so
1142 * no token is needed.
1145 hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp)
1149 KKASSERT(MAXFIDSZ >= 16);
1151 fhp->fid_len = offsetof(struct fid, fid_data[16]);
1152 fhp->fid_ext = ip->obj_localization >> 16;
1153 bcopy(&ip->obj_id, fhp->fid_data + 0, sizeof(ip->obj_id));
1154 bcopy(&ip->obj_asof, fhp->fid_data + 8, sizeof(ip->obj_asof));
1160 * Convert a file handle back to a vnode.
1162 * Use rootvp to enforce PFS isolation when a PFS is exported via a
1166 hammer_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
1167 struct fid *fhp, struct vnode **vpp)
1169 hammer_mount_t hmp = (void *)mp->mnt_data;
1170 struct hammer_transaction trans;
1171 struct hammer_inode *ip;
1172 struct hammer_inode_info info;
1174 u_int32_t localization;
1176 bcopy(fhp->fid_data + 0, &info.obj_id, sizeof(info.obj_id));
1177 bcopy(fhp->fid_data + 8, &info.obj_asof, sizeof(info.obj_asof));
1179 localization = VTOI(rootvp)->obj_localization;
1181 localization = (u_int32_t)fhp->fid_ext << 16;
1183 lwkt_gettoken(&hmp->fs_token);
1184 hammer_simple_transaction(&trans, hmp);
1187 * Get/allocate the hammer_inode structure. The structure must be
1188 * unlocked while we manipulate the related vnode to avoid a
1191 ip = hammer_get_inode(&trans, NULL, info.obj_id,
1192 info.obj_asof, localization, 0, &error);
1194 error = hammer_get_vnode(ip, vpp);
1195 hammer_rel_inode(ip, 0);
1199 hammer_done_transaction(&trans);
1200 lwkt_reltoken(&hmp->fs_token);
1205 hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
1206 int *exflagsp, struct ucred **credanonp)
1208 hammer_mount_t hmp = (void *)mp->mnt_data;
1212 lwkt_gettoken(&hmp->fs_token);
1213 np = vfs_export_lookup(mp, &hmp->export, nam);
1215 *exflagsp = np->netc_exflags;
1216 *credanonp = &np->netc_anon;
1221 lwkt_reltoken(&hmp->fs_token);
1227 hammer_vfs_export(struct mount *mp, int op, const struct export_args *export)
1229 hammer_mount_t hmp = (void *)mp->mnt_data;
1232 lwkt_gettoken(&hmp->fs_token);
1235 case MOUNTCTL_SET_EXPORT:
1236 error = vfs_export(mp, &hmp->export, export);
1242 lwkt_reltoken(&hmp->fs_token);