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>
50 int hammer_supported_version = HAMMER_VOL_VERSION_TWO;
52 int hammer_debug_general;
53 int hammer_debug_debug = 1; /* medium-error panics */
54 int hammer_debug_inode;
55 int hammer_debug_locks;
56 int hammer_debug_btree;
58 int hammer_debug_recover; /* -1 will disable, +1 will force */
59 int hammer_debug_recover_faults;
60 int hammer_cluster_enable = 1; /* enable read clustering by default */
61 int hammer_count_fsyncs;
62 int hammer_count_inodes;
63 int hammer_count_iqueued;
64 int hammer_count_reclaiming;
65 int hammer_count_records;
66 int hammer_count_record_datas;
67 int hammer_count_volumes;
68 int hammer_count_buffers;
69 int hammer_count_nodes;
70 int64_t hammer_count_extra_space_used;
71 int64_t hammer_stats_btree_lookups;
72 int64_t hammer_stats_btree_searches;
73 int64_t hammer_stats_btree_inserts;
74 int64_t hammer_stats_btree_deletes;
75 int64_t hammer_stats_btree_elements;
76 int64_t hammer_stats_btree_splits;
77 int64_t hammer_stats_btree_iterations;
78 int64_t hammer_stats_btree_root_iterations;
79 int64_t hammer_stats_record_iterations;
81 int64_t hammer_stats_file_read;
82 int64_t hammer_stats_file_write;
83 int64_t hammer_stats_file_iopsr;
84 int64_t hammer_stats_file_iopsw;
85 int64_t hammer_stats_disk_read;
86 int64_t hammer_stats_disk_write;
87 int64_t hammer_stats_inode_flushes;
88 int64_t hammer_stats_commits;
90 int hammer_count_dirtybufspace; /* global */
91 int hammer_count_refedbufs; /* global */
92 int hammer_count_reservations;
93 int hammer_count_io_running_read;
94 int hammer_count_io_running_write;
95 int hammer_count_io_locked;
96 int hammer_limit_dirtybufspace; /* per-mount */
97 int hammer_limit_recs; /* as a whole XXX */
98 int hammer_autoflush = 2000; /* auto flush */
100 int hammer_verify_zone;
101 int hammer_verify_data = 1;
102 int hammer_write_mode;
103 int hammer_yield_check = 16;
104 int64_t hammer_contention_count;
105 int64_t hammer_zone_limit;
107 SYSCTL_NODE(_vfs, OID_AUTO, hammer, CTLFLAG_RW, 0, "HAMMER filesystem");
108 SYSCTL_INT(_vfs_hammer, OID_AUTO, supported_version, CTLFLAG_RD,
109 &hammer_supported_version, 0, "");
110 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_general, CTLFLAG_RW,
111 &hammer_debug_general, 0, "");
112 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_io, CTLFLAG_RW,
113 &hammer_debug_io, 0, "");
114 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_debug, CTLFLAG_RW,
115 &hammer_debug_debug, 0, "");
116 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_inode, CTLFLAG_RW,
117 &hammer_debug_inode, 0, "");
118 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_locks, CTLFLAG_RW,
119 &hammer_debug_locks, 0, "");
120 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_btree, CTLFLAG_RW,
121 &hammer_debug_btree, 0, "");
122 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_tid, CTLFLAG_RW,
123 &hammer_debug_tid, 0, "");
124 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover, CTLFLAG_RW,
125 &hammer_debug_recover, 0, "");
126 SYSCTL_INT(_vfs_hammer, OID_AUTO, debug_recover_faults, CTLFLAG_RW,
127 &hammer_debug_recover_faults, 0, "");
128 SYSCTL_INT(_vfs_hammer, OID_AUTO, cluster_enable, CTLFLAG_RW,
129 &hammer_cluster_enable, 0, "");
131 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_dirtybufspace, CTLFLAG_RW,
132 &hammer_limit_dirtybufspace, 0, "");
133 SYSCTL_INT(_vfs_hammer, OID_AUTO, limit_recs, CTLFLAG_RW,
134 &hammer_limit_recs, 0, "");
136 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_fsyncs, CTLFLAG_RD,
137 &hammer_count_fsyncs, 0, "");
138 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_inodes, CTLFLAG_RD,
139 &hammer_count_inodes, 0, "");
140 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_iqueued, CTLFLAG_RD,
141 &hammer_count_iqueued, 0, "");
142 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reclaiming, CTLFLAG_RD,
143 &hammer_count_reclaiming, 0, "");
144 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_records, CTLFLAG_RD,
145 &hammer_count_records, 0, "");
146 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_record_datas, CTLFLAG_RD,
147 &hammer_count_record_datas, 0, "");
148 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_volumes, CTLFLAG_RD,
149 &hammer_count_volumes, 0, "");
150 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_buffers, CTLFLAG_RD,
151 &hammer_count_buffers, 0, "");
152 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_nodes, CTLFLAG_RD,
153 &hammer_count_nodes, 0, "");
154 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, count_extra_space_used, CTLFLAG_RD,
155 &hammer_count_extra_space_used, 0, "");
157 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_searches, CTLFLAG_RD,
158 &hammer_stats_btree_searches, 0, "");
159 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_lookups, CTLFLAG_RD,
160 &hammer_stats_btree_lookups, 0, "");
161 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_inserts, CTLFLAG_RD,
162 &hammer_stats_btree_inserts, 0, "");
163 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_deletes, CTLFLAG_RD,
164 &hammer_stats_btree_deletes, 0, "");
165 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_elements, CTLFLAG_RD,
166 &hammer_stats_btree_elements, 0, "");
167 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_splits, CTLFLAG_RD,
168 &hammer_stats_btree_splits, 0, "");
169 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_iterations, CTLFLAG_RD,
170 &hammer_stats_btree_iterations, 0, "");
171 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_btree_root_iterations, CTLFLAG_RD,
172 &hammer_stats_btree_root_iterations, 0, "");
173 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_record_iterations, CTLFLAG_RD,
174 &hammer_stats_record_iterations, 0, "");
176 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_read, CTLFLAG_RD,
177 &hammer_stats_file_read, 0, "");
178 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_write, CTLFLAG_RD,
179 &hammer_stats_file_write, 0, "");
180 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_iopsr, CTLFLAG_RD,
181 &hammer_stats_file_iopsr, 0, "");
182 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_file_iopsw, CTLFLAG_RD,
183 &hammer_stats_file_iopsw, 0, "");
184 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_disk_read, CTLFLAG_RD,
185 &hammer_stats_disk_read, 0, "");
186 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_disk_write, CTLFLAG_RD,
187 &hammer_stats_disk_write, 0, "");
188 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_inode_flushes, CTLFLAG_RD,
189 &hammer_stats_inode_flushes, 0, "");
190 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, stats_commits, CTLFLAG_RD,
191 &hammer_stats_commits, 0, "");
193 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_dirtybufspace, CTLFLAG_RD,
194 &hammer_count_dirtybufspace, 0, "");
195 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_refedbufs, CTLFLAG_RD,
196 &hammer_count_refedbufs, 0, "");
197 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_reservations, CTLFLAG_RD,
198 &hammer_count_reservations, 0, "");
199 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_running_read, CTLFLAG_RD,
200 &hammer_count_io_running_read, 0, "");
201 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_locked, CTLFLAG_RD,
202 &hammer_count_io_locked, 0, "");
203 SYSCTL_INT(_vfs_hammer, OID_AUTO, count_io_running_write, CTLFLAG_RD,
204 &hammer_count_io_running_write, 0, "");
205 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, zone_limit, CTLFLAG_RW,
206 &hammer_zone_limit, 0, "");
207 SYSCTL_QUAD(_vfs_hammer, OID_AUTO, contention_count, CTLFLAG_RW,
208 &hammer_contention_count, 0, "");
209 SYSCTL_INT(_vfs_hammer, OID_AUTO, autoflush, CTLFLAG_RW,
210 &hammer_autoflush, 0, "");
211 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_zone, CTLFLAG_RW,
212 &hammer_verify_zone, 0, "");
213 SYSCTL_INT(_vfs_hammer, OID_AUTO, verify_data, CTLFLAG_RW,
214 &hammer_verify_data, 0, "");
215 SYSCTL_INT(_vfs_hammer, OID_AUTO, write_mode, CTLFLAG_RW,
216 &hammer_write_mode, 0, "");
217 SYSCTL_INT(_vfs_hammer, OID_AUTO, yield_check, CTLFLAG_RW,
218 &hammer_yield_check, 0, "");
220 KTR_INFO_MASTER(hammer);
225 static void hammer_free_hmp(struct mount *mp);
227 static int hammer_vfs_mount(struct mount *mp, char *path, caddr_t data,
229 static int hammer_vfs_unmount(struct mount *mp, int mntflags);
230 static int hammer_vfs_root(struct mount *mp, struct vnode **vpp);
231 static int hammer_vfs_statfs(struct mount *mp, struct statfs *sbp,
233 static int hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp,
235 static int hammer_vfs_sync(struct mount *mp, int waitfor);
236 static int hammer_vfs_vget(struct mount *mp, struct vnode *dvp,
237 ino_t ino, struct vnode **vpp);
238 static int hammer_vfs_init(struct vfsconf *conf);
239 static int hammer_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
240 struct fid *fhp, struct vnode **vpp);
241 static int hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp);
242 static int hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
243 int *exflagsp, struct ucred **credanonp);
246 static struct vfsops hammer_vfsops = {
247 .vfs_mount = hammer_vfs_mount,
248 .vfs_unmount = hammer_vfs_unmount,
249 .vfs_root = hammer_vfs_root,
250 .vfs_statfs = hammer_vfs_statfs,
251 .vfs_statvfs = hammer_vfs_statvfs,
252 .vfs_sync = hammer_vfs_sync,
253 .vfs_vget = hammer_vfs_vget,
254 .vfs_init = hammer_vfs_init,
255 .vfs_vptofh = hammer_vfs_vptofh,
256 .vfs_fhtovp = hammer_vfs_fhtovp,
257 .vfs_checkexp = hammer_vfs_checkexp
260 MALLOC_DEFINE(M_HAMMER, "HAMMER-mount", "");
262 VFS_SET(hammer_vfsops, hammer, 0);
263 MODULE_VERSION(hammer, 1);
266 hammer_vfs_init(struct vfsconf *conf)
270 if (hammer_limit_recs == 0) {
271 hammer_limit_recs = nbuf * 25;
272 n = kmalloc_limit(M_HAMMER) / 512;
273 if (hammer_limit_recs > n)
274 hammer_limit_recs = n;
276 if (hammer_limit_dirtybufspace == 0) {
277 hammer_limit_dirtybufspace = hidirtybufspace / 2;
278 if (hammer_limit_dirtybufspace < 100)
279 hammer_limit_dirtybufspace = 100;
285 hammer_vfs_mount(struct mount *mp, char *mntpt, caddr_t data,
288 struct hammer_mount_info info;
290 hammer_volume_t rootvol;
291 struct vnode *rootvp;
292 struct vnode *devvp = NULL;
293 const char *upath; /* volume name in userspace */
294 char *path; /* volume name in system space */
301 * Accept hammer_mount_info. mntpt is NULL for root mounts at boot.
304 if ((error = bdevvp(rootdev, &devvp))) {
305 kprintf("hammer_mountroot: can't find devvp\n");
308 mp->mnt_flag &= ~MNT_RDONLY; /* mount R/W */
309 bzero(&info, sizeof(info));
314 if ((error = copyin(data, &info, sizeof(info))) != 0)
319 * updating or new mount
321 if (mp->mnt_flag & MNT_UPDATE) {
322 hmp = (void *)mp->mnt_data;
323 KKASSERT(hmp != NULL);
325 if (info.nvolumes <= 0 || info.nvolumes >= 32768)
331 * master-id validation. The master id may not be changed by a
334 if (info.hflags & HMNT_MASTERID) {
335 if (hmp && hmp->master_id != info.master_id) {
336 kprintf("hammer: cannot change master id "
337 "with mount update\n");
340 master_id = info.master_id;
341 if (master_id < -1 || master_id >= HAMMER_MAX_MASTERS)
345 master_id = hmp->master_id;
351 * Interal mount data structure
354 hmp = kmalloc(sizeof(*hmp), M_HAMMER, M_WAITOK | M_ZERO);
355 mp->mnt_data = (qaddr_t)hmp;
357 /*TAILQ_INIT(&hmp->recycle_list);*/
360 * Make sure kmalloc type limits are set appropriately. If root
361 * increases the vnode limit you may have to do a dummy remount
362 * to adjust the HAMMER inode limit.
364 kmalloc_create(&hmp->m_misc, "HAMMER-others");
365 kmalloc_create(&hmp->m_inodes, "HAMMER-inodes");
367 maxinodes = desiredvnodes + desiredvnodes / 5 +
369 kmalloc_raise_limit(hmp->m_inodes,
370 maxinodes * sizeof(struct hammer_inode));
372 hmp->root_btree_beg.localization = 0x00000000U;
373 hmp->root_btree_beg.obj_id = -0x8000000000000000LL;
374 hmp->root_btree_beg.key = -0x8000000000000000LL;
375 hmp->root_btree_beg.create_tid = 1;
376 hmp->root_btree_beg.delete_tid = 1;
377 hmp->root_btree_beg.rec_type = 0;
378 hmp->root_btree_beg.obj_type = 0;
380 hmp->root_btree_end.localization = 0xFFFFFFFFU;
381 hmp->root_btree_end.obj_id = 0x7FFFFFFFFFFFFFFFLL;
382 hmp->root_btree_end.key = 0x7FFFFFFFFFFFFFFFLL;
383 hmp->root_btree_end.create_tid = 0xFFFFFFFFFFFFFFFFULL;
384 hmp->root_btree_end.delete_tid = 0; /* special case */
385 hmp->root_btree_end.rec_type = 0xFFFFU;
386 hmp->root_btree_end.obj_type = 0;
388 hmp->krate.freq = 1; /* maximum reporting rate (hz) */
389 hmp->krate.count = -16; /* initial burst */
391 hmp->sync_lock.refs = 1;
392 hmp->free_lock.refs = 1;
393 hmp->undo_lock.refs = 1;
394 hmp->blkmap_lock.refs = 1;
396 TAILQ_INIT(&hmp->delay_list);
397 TAILQ_INIT(&hmp->flush_group_list);
398 TAILQ_INIT(&hmp->objid_cache_list);
399 TAILQ_INIT(&hmp->undo_lru_list);
400 TAILQ_INIT(&hmp->reclaim_list);
402 hmp->hflags &= ~HMNT_USERFLAGS;
403 hmp->hflags |= info.hflags & HMNT_USERFLAGS;
405 hmp->master_id = master_id;
408 mp->mnt_flag |= MNT_RDONLY;
409 hmp->asof = info.asof;
411 hmp->asof = HAMMER_MAX_TID;
415 * Re-open read-write if originally read-only, or vise-versa.
417 if (mp->mnt_flag & MNT_UPDATE) {
419 if (hmp->ronly && (mp->mnt_kern_flag & MNTK_WANTRDWR)) {
420 kprintf("HAMMER read-only -> read-write\n");
422 RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
423 hammer_adjust_volume_mode, NULL);
424 rootvol = hammer_get_root_volume(hmp, &error);
426 hammer_recover_flush_buffers(hmp, rootvol, 1);
427 bcopy(rootvol->ondisk->vol0_blockmap,
429 sizeof(hmp->blockmap));
430 hammer_rel_volume(rootvol, 0);
432 RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
433 hammer_reload_inode, NULL);
434 /* kernel clears MNT_RDONLY */
435 } else if (hmp->ronly == 0 && (mp->mnt_flag & MNT_RDONLY)) {
436 kprintf("HAMMER read-write -> read-only\n");
437 hmp->ronly = 1; /* messy */
438 RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
439 hammer_reload_inode, NULL);
441 hammer_flusher_sync(hmp);
442 hammer_flusher_sync(hmp);
443 hammer_flusher_sync(hmp);
445 RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
446 hammer_adjust_volume_mode, NULL);
451 RB_INIT(&hmp->rb_vols_root);
452 RB_INIT(&hmp->rb_inos_root);
453 RB_INIT(&hmp->rb_nods_root);
454 RB_INIT(&hmp->rb_undo_root);
455 RB_INIT(&hmp->rb_resv_root);
456 RB_INIT(&hmp->rb_bufs_root);
457 RB_INIT(&hmp->rb_pfsm_root);
459 hmp->ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
461 TAILQ_INIT(&hmp->volu_list);
462 TAILQ_INIT(&hmp->undo_list);
463 TAILQ_INIT(&hmp->data_list);
464 TAILQ_INIT(&hmp->meta_list);
465 TAILQ_INIT(&hmp->lose_list);
470 path = objcache_get(namei_oc, M_WAITOK);
472 for (i = 0; i < info.nvolumes; ++i) {
476 * Only one volume; and no need for copyin.
478 KKASSERT(info.nvolumes == 1);
479 ksnprintf(path, MAXPATHLEN, "/dev/%s",
480 mp->mnt_stat.f_mntfromname);
483 error = copyin(&info.volumes[i], &upath,
486 error = copyinstr(upath, path,
490 error = hammer_install_volume(hmp, path, devvp);
494 objcache_put(namei_oc, path);
497 * Make sure we found a root volume
499 if (error == 0 && hmp->rootvol == NULL) {
500 kprintf("hammer_mount: No root volume found!\n");
505 * Check that all required volumes are available
507 if (error == 0 && hammer_mountcheck_volumes(hmp)) {
508 kprintf("hammer_mount: Missing volumes, cannot mount!\n");
518 * No errors, setup enough of the mount point so we can lookup the
521 mp->mnt_iosize_max = MAXPHYS;
522 mp->mnt_kern_flag |= MNTK_FSMID;
525 * note: f_iosize is used by vnode_pager_haspage() when constructing
528 mp->mnt_stat.f_iosize = HAMMER_BUFSIZE;
529 mp->mnt_stat.f_bsize = HAMMER_BUFSIZE;
531 mp->mnt_vstat.f_frsize = HAMMER_BUFSIZE;
532 mp->mnt_vstat.f_bsize = HAMMER_BUFSIZE;
534 mp->mnt_maxsymlinklen = 255;
535 mp->mnt_flag |= MNT_LOCAL;
537 vfs_add_vnodeops(mp, &hammer_vnode_vops, &mp->mnt_vn_norm_ops);
538 vfs_add_vnodeops(mp, &hammer_spec_vops, &mp->mnt_vn_spec_ops);
539 vfs_add_vnodeops(mp, &hammer_fifo_vops, &mp->mnt_vn_fifo_ops);
542 * The root volume's ondisk pointer is only valid if we hold a
545 rootvol = hammer_get_root_volume(hmp, &error);
550 * Perform any necessary UNDO operations. The recovery code does
551 * call hammer_undo_lookup() so we have to pre-cache the blockmap,
552 * and then re-copy it again after recovery is complete.
554 * If this is a read-only mount the UNDO information is retained
555 * in memory in the form of dirty buffer cache buffers, and not
556 * written back to the media.
558 bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
559 sizeof(hmp->blockmap));
562 * Check filesystem version
564 hmp->version = rootvol->ondisk->vol_version;
565 if (hmp->version < HAMMER_VOL_VERSION_MIN ||
566 hmp->version > HAMMER_VOL_VERSION_MAX) {
567 kprintf("HAMMER: mount unsupported fs version %d\n",
574 * The undo_rec_limit limits the size of flush groups to avoid
575 * blowing out the UNDO FIFO. This calculation is typically in
576 * the tens of thousands and is designed primarily when small
577 * HAMMER filesystems are created.
579 hmp->undo_rec_limit = hammer_undo_max(hmp) / 8192 + 100;
580 if (hammer_debug_general & 0x0001)
581 kprintf("HAMMER: undo_rec_limit %d\n", hmp->undo_rec_limit);
583 error = hammer_recover(hmp, rootvol);
585 kprintf("Failed to recover HAMMER filesystem on mount\n");
590 * Finish setup now that we have a good root volume.
592 * The top 16 bits of fsid.val[1] is a pfs id.
594 ksnprintf(mp->mnt_stat.f_mntfromname,
595 sizeof(mp->mnt_stat.f_mntfromname), "%s",
596 rootvol->ondisk->vol_name);
597 mp->mnt_stat.f_fsid.val[0] =
598 crc32((char *)&rootvol->ondisk->vol_fsid + 0, 8);
599 mp->mnt_stat.f_fsid.val[1] =
600 crc32((char *)&rootvol->ondisk->vol_fsid + 8, 8);
601 mp->mnt_stat.f_fsid.val[1] &= 0x0000FFFF;
603 mp->mnt_vstat.f_fsid_uuid = rootvol->ondisk->vol_fsid;
604 mp->mnt_vstat.f_fsid = crc32(&mp->mnt_vstat.f_fsid_uuid,
605 sizeof(mp->mnt_vstat.f_fsid_uuid));
608 * Certain often-modified fields in the root volume are cached in
609 * the hammer_mount structure so we do not have to generate lots
610 * of little UNDO structures for them.
612 * Recopy after recovery. This also has the side effect of
613 * setting our cached undo FIFO's first_offset, which serves to
614 * placemark the FIFO start for the NEXT flush cycle while the
615 * on-disk first_offset represents the LAST flush cycle.
617 hmp->next_tid = rootvol->ondisk->vol0_next_tid;
618 hmp->flush_tid1 = hmp->next_tid;
619 hmp->flush_tid2 = hmp->next_tid;
620 bcopy(rootvol->ondisk->vol0_blockmap, hmp->blockmap,
621 sizeof(hmp->blockmap));
622 hmp->copy_stat_freebigblocks = rootvol->ondisk->vol0_stat_freebigblocks;
624 hammer_flusher_create(hmp);
627 * Locate the root directory using the root cluster's B-Tree as a
628 * starting point. The root directory uses an obj_id of 1.
630 * FUTURE: Leave the root directory cached referenced but unlocked
631 * in hmp->rootvp (need to flush it on unmount).
633 error = hammer_vfs_vget(mp, NULL, 1, &rootvp);
637 /*vn_unlock(hmp->rootvp);*/
640 hammer_rel_volume(rootvol, 0);
643 * Cleanup and return.
651 hammer_vfs_unmount(struct mount *mp, int mntflags)
654 struct hammer_mount *hmp = (void *)mp->mnt_data;
660 * Clean out the vnodes
663 if (mntflags & MNT_FORCE)
665 if ((error = vflush(mp, 0, flags)) != 0)
669 * Clean up the internal mount structure and related entities. This
677 * Clean up the internal mount structure and disassociate it from the mount.
678 * This may issue I/O.
681 hammer_free_hmp(struct mount *mp)
683 struct hammer_mount *hmp = (void *)mp->mnt_data;
684 hammer_flush_group_t flg;
689 * Flush anything dirty. This won't even run if the
690 * filesystem errored-out.
693 while (hammer_flusher_haswork(hmp)) {
694 hammer_flusher_sync(hmp);
698 kprintf("HAMMER: umount flushing.");
701 tsleep(&dummy, 0, "hmrufl", hz);
704 kprintf("giving up\n");
708 if (count >= 5 && count < 30)
712 * If the mount had a critical error we have to destroy any
713 * remaining inodes before we can finish cleaning up the flusher.
715 if (hmp->flags & HAMMER_MOUNT_CRITICAL_ERROR) {
716 RB_SCAN(hammer_ino_rb_tree, &hmp->rb_inos_root, NULL,
717 hammer_destroy_inode_callback, NULL);
721 * There shouldn't be any inodes left now and any left over
722 * flush groups should now be empty.
724 KKASSERT(RB_EMPTY(&hmp->rb_inos_root));
725 while ((flg = TAILQ_FIRST(&hmp->flush_group_list)) != NULL) {
726 TAILQ_REMOVE(&hmp->flush_group_list, flg, flush_entry);
727 KKASSERT(TAILQ_EMPTY(&flg->flush_list));
729 kprintf("HAMMER: Warning, flush_group %p was "
730 "not empty on umount!\n", flg);
732 kfree(flg, hmp->m_misc);
736 * We can finally destroy the flusher
738 hammer_flusher_destroy(hmp);
741 * We may have held recovered buffers due to a read-only mount.
742 * These must be discarded.
745 hammer_recover_flush_buffers(hmp, NULL, -1);
748 * Unload buffers and then volumes
750 RB_SCAN(hammer_buf_rb_tree, &hmp->rb_bufs_root, NULL,
751 hammer_unload_buffer, NULL);
752 RB_SCAN(hammer_vol_rb_tree, &hmp->rb_vols_root, NULL,
753 hammer_unload_volume, NULL);
756 mp->mnt_flag &= ~MNT_LOCAL;
758 hammer_destroy_objid_cache(hmp);
759 kmalloc_destroy(&hmp->m_misc);
760 kmalloc_destroy(&hmp->m_inodes);
761 kfree(hmp, M_HAMMER);
765 * Report critical errors. ip may be NULL.
768 hammer_critical_error(hammer_mount_t hmp, hammer_inode_t ip,
769 int error, const char *msg)
771 hmp->flags |= HAMMER_MOUNT_CRITICAL_ERROR;
772 krateprintf(&hmp->krate,
773 "HAMMER(%s): Critical error inode=%lld %s\n",
774 hmp->mp->mnt_stat.f_mntfromname,
775 (long long)(ip ? ip->obj_id : -1), msg);
776 if (hmp->ronly == 0) {
777 hmp->ronly = 2; /* special errored read-only mode */
778 hmp->mp->mnt_flag |= MNT_RDONLY;
779 kprintf("HAMMER(%s): Forcing read-only mode\n",
780 hmp->mp->mnt_stat.f_mntfromname);
787 * Obtain a vnode for the specified inode number. An exclusively locked
791 hammer_vfs_vget(struct mount *mp, struct vnode *dvp,
792 ino_t ino, struct vnode **vpp)
794 struct hammer_transaction trans;
795 struct hammer_mount *hmp = (void *)mp->mnt_data;
796 struct hammer_inode *ip;
798 u_int32_t localization;
800 hammer_simple_transaction(&trans, hmp);
803 * If a directory vnode is supplied (mainly NFS) then we can acquire
804 * the PFS domain from it. Otherwise we would only be able to vget
805 * inodes in the root PFS.
808 localization = HAMMER_DEF_LOCALIZATION +
809 VTOI(dvp)->obj_localization;
811 localization = HAMMER_DEF_LOCALIZATION;
815 * Lookup the requested HAMMER inode. The structure must be
816 * left unlocked while we manipulate the related vnode to avoid
819 ip = hammer_get_inode(&trans, NULL, ino,
820 hmp->asof, localization,
824 hammer_done_transaction(&trans);
827 error = hammer_get_vnode(ip, vpp);
828 hammer_rel_inode(ip, 0);
829 hammer_done_transaction(&trans);
834 * Return the root vnode for the filesystem.
836 * HAMMER stores the root vnode in the hammer_mount structure so
837 * getting it is easy.
840 hammer_vfs_root(struct mount *mp, struct vnode **vpp)
843 struct hammer_mount *hmp = (void *)mp->mnt_data;
847 error = hammer_vfs_vget(mp, NULL, 1, vpp);
852 hammer_vfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
854 struct hammer_mount *hmp = (void *)mp->mnt_data;
855 hammer_volume_t volume;
856 hammer_volume_ondisk_t ondisk;
861 volume = hammer_get_root_volume(hmp, &error);
864 ondisk = volume->ondisk;
869 _hammer_checkspace(hmp, HAMMER_CHKSPC_WRITE, &breserved);
870 mp->mnt_stat.f_files = ondisk->vol0_stat_inodes;
871 bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE;
872 hammer_rel_volume(volume, 0);
874 mp->mnt_stat.f_bfree = (bfree - breserved) / HAMMER_BUFSIZE;
875 mp->mnt_stat.f_bavail = mp->mnt_stat.f_bfree;
876 if (mp->mnt_stat.f_files < 0)
877 mp->mnt_stat.f_files = 0;
884 hammer_vfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred)
886 struct hammer_mount *hmp = (void *)mp->mnt_data;
887 hammer_volume_t volume;
888 hammer_volume_ondisk_t ondisk;
893 volume = hammer_get_root_volume(hmp, &error);
896 ondisk = volume->ondisk;
901 _hammer_checkspace(hmp, HAMMER_CHKSPC_WRITE, &breserved);
902 mp->mnt_vstat.f_files = ondisk->vol0_stat_inodes;
903 bfree = ondisk->vol0_stat_freebigblocks * HAMMER_LARGEBLOCK_SIZE;
904 hammer_rel_volume(volume, 0);
906 mp->mnt_vstat.f_bfree = (bfree - breserved) / HAMMER_BUFSIZE;
907 mp->mnt_vstat.f_bavail = mp->mnt_stat.f_bfree;
908 if (mp->mnt_vstat.f_files < 0)
909 mp->mnt_vstat.f_files = 0;
910 *sbp = mp->mnt_vstat;
915 * Sync the filesystem. Currently we have to run it twice, the second
916 * one will advance the undo start index to the end index, so if a crash
917 * occurs no undos will be run on mount.
919 * We do not sync the filesystem if we are called from a panic. If we did
920 * we might end up blowing up a sync that was already in progress.
923 hammer_vfs_sync(struct mount *mp, int waitfor)
925 struct hammer_mount *hmp = (void *)mp->mnt_data;
928 if (panicstr == NULL) {
929 error = hammer_sync_hmp(hmp, waitfor);
937 * Convert a vnode to a file handle.
940 hammer_vfs_vptofh(struct vnode *vp, struct fid *fhp)
944 KKASSERT(MAXFIDSZ >= 16);
946 fhp->fid_len = offsetof(struct fid, fid_data[16]);
947 fhp->fid_ext = ip->obj_localization >> 16;
948 bcopy(&ip->obj_id, fhp->fid_data + 0, sizeof(ip->obj_id));
949 bcopy(&ip->obj_asof, fhp->fid_data + 8, sizeof(ip->obj_asof));
955 * Convert a file handle back to a vnode.
957 * Use rootvp to enforce PFS isolation when a PFS is exported via a
961 hammer_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
962 struct fid *fhp, struct vnode **vpp)
964 struct hammer_transaction trans;
965 struct hammer_inode *ip;
966 struct hammer_inode_info info;
968 u_int32_t localization;
970 bcopy(fhp->fid_data + 0, &info.obj_id, sizeof(info.obj_id));
971 bcopy(fhp->fid_data + 8, &info.obj_asof, sizeof(info.obj_asof));
973 localization = VTOI(rootvp)->obj_localization;
975 localization = (u_int32_t)fhp->fid_ext << 16;
977 hammer_simple_transaction(&trans, (void *)mp->mnt_data);
980 * Get/allocate the hammer_inode structure. The structure must be
981 * unlocked while we manipulate the related vnode to avoid a
984 ip = hammer_get_inode(&trans, NULL, info.obj_id,
985 info.obj_asof, localization, 0, &error);
990 error = hammer_get_vnode(ip, vpp);
991 hammer_rel_inode(ip, 0);
992 hammer_done_transaction(&trans);
997 hammer_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
998 int *exflagsp, struct ucred **credanonp)
1000 hammer_mount_t hmp = (void *)mp->mnt_data;
1004 np = vfs_export_lookup(mp, &hmp->export, nam);
1006 *exflagsp = np->netc_exflags;
1007 *credanonp = &np->netc_anon;
1017 hammer_vfs_export(struct mount *mp, int op, const struct export_args *export)
1019 hammer_mount_t hmp = (void *)mp->mnt_data;
1023 case MOUNTCTL_SET_EXPORT:
1024 error = vfs_export(mp, &hmp->export, export);