2 * Copyright (c) 2011-2015 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>
6 * by Daniel Flores (GSOC 2013 - mentored by Matthew Dillon, compression)
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in
16 * the documentation and/or other materials provided with the
18 * 3. Neither the name of The DragonFly Project nor the names of its
19 * contributors may be used to endorse or promote products derived
20 * from this software without specific, prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
23 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
24 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
25 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
26 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
27 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
28 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
29 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
30 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
31 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
32 * 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/nlookup.h>
39 #include <sys/vnode.h>
40 #include <sys/mount.h>
41 #include <sys/fcntl.h>
44 #include <sys/vfsops.h>
45 #include <sys/sysctl.h>
46 #include <sys/socket.h>
47 #include <sys/objcache.h>
50 #include <sys/namei.h>
51 #include <sys/mountctl.h>
52 #include <sys/dirent.h>
55 #include <sys/mutex.h>
56 #include <sys/mutex2.h>
59 #include "hammer2_disk.h"
60 #include "hammer2_mount.h"
61 #include "hammer2_lz4.h"
63 #include "zlib/hammer2_zlib.h"
65 #define REPORT_REFS_ERRORS 1 /* XXX remove me */
67 MALLOC_DEFINE(M_OBJCACHE, "objcache", "Object Cache");
69 struct hammer2_sync_info {
74 TAILQ_HEAD(hammer2_mntlist, hammer2_dev);
75 static struct hammer2_mntlist hammer2_mntlist;
77 struct hammer2_pfslist hammer2_pfslist;
78 struct lock hammer2_mntlk;
81 int hammer2_cluster_read = 4; /* physical read-ahead */
82 int hammer2_cluster_write = 0; /* bdwrite() so later inval works */
83 int hammer2_dedup_enable = 1;
84 int hammer2_inval_enable = 0;
85 int hammer2_flush_pipe = 100;
86 int hammer2_synchronous_flush = 1;
87 int hammer2_dio_count;
88 long hammer2_chain_allocs;
89 long hammer2_chain_frees;
90 long hammer2_limit_dirty_chains;
91 long hammer2_count_modified_chains;
92 long hammer2_iod_invals;
93 long hammer2_iod_file_read;
94 long hammer2_iod_meta_read;
95 long hammer2_iod_indr_read;
96 long hammer2_iod_fmap_read;
97 long hammer2_iod_volu_read;
98 long hammer2_iod_file_write;
99 long hammer2_iod_file_wembed;
100 long hammer2_iod_file_wzero;
101 long hammer2_iod_file_wdedup;
102 long hammer2_iod_meta_write;
103 long hammer2_iod_indr_write;
104 long hammer2_iod_fmap_write;
105 long hammer2_iod_volu_write;
107 MALLOC_DECLARE(M_HAMMER2_CBUFFER);
108 MALLOC_DEFINE(M_HAMMER2_CBUFFER, "HAMMER2-compbuffer",
109 "Buffer used for compression.");
111 MALLOC_DECLARE(M_HAMMER2_DEBUFFER);
112 MALLOC_DEFINE(M_HAMMER2_DEBUFFER, "HAMMER2-decompbuffer",
113 "Buffer used for decompression.");
115 SYSCTL_NODE(_vfs, OID_AUTO, hammer2, CTLFLAG_RW, 0, "HAMMER2 filesystem");
117 SYSCTL_INT(_vfs_hammer2, OID_AUTO, debug, CTLFLAG_RW,
118 &hammer2_debug, 0, "");
119 SYSCTL_INT(_vfs_hammer2, OID_AUTO, cluster_read, CTLFLAG_RW,
120 &hammer2_cluster_read, 0, "");
121 SYSCTL_INT(_vfs_hammer2, OID_AUTO, cluster_write, CTLFLAG_RW,
122 &hammer2_cluster_write, 0, "");
123 SYSCTL_INT(_vfs_hammer2, OID_AUTO, dedup_enable, CTLFLAG_RW,
124 &hammer2_dedup_enable, 0, "");
125 SYSCTL_INT(_vfs_hammer2, OID_AUTO, inval_enable, CTLFLAG_RW,
126 &hammer2_inval_enable, 0, "");
127 SYSCTL_INT(_vfs_hammer2, OID_AUTO, flush_pipe, CTLFLAG_RW,
128 &hammer2_flush_pipe, 0, "");
129 SYSCTL_INT(_vfs_hammer2, OID_AUTO, synchronous_flush, CTLFLAG_RW,
130 &hammer2_synchronous_flush, 0, "");
131 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, chain_allocs, CTLFLAG_RW,
132 &hammer2_chain_allocs, 0, "");
133 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, chain_frees, CTLFLAG_RW,
134 &hammer2_chain_frees, 0, "");
135 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, limit_dirty_chains, CTLFLAG_RW,
136 &hammer2_limit_dirty_chains, 0, "");
137 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, count_modified_chains, CTLFLAG_RW,
138 &hammer2_count_modified_chains, 0, "");
139 SYSCTL_INT(_vfs_hammer2, OID_AUTO, dio_count, CTLFLAG_RD,
140 &hammer2_dio_count, 0, "");
142 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_invals, CTLFLAG_RW,
143 &hammer2_iod_invals, 0, "");
144 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_file_read, CTLFLAG_RW,
145 &hammer2_iod_file_read, 0, "");
146 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_meta_read, CTLFLAG_RW,
147 &hammer2_iod_meta_read, 0, "");
148 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_indr_read, CTLFLAG_RW,
149 &hammer2_iod_indr_read, 0, "");
150 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_fmap_read, CTLFLAG_RW,
151 &hammer2_iod_fmap_read, 0, "");
152 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_volu_read, CTLFLAG_RW,
153 &hammer2_iod_volu_read, 0, "");
155 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_file_write, CTLFLAG_RW,
156 &hammer2_iod_file_write, 0, "");
157 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_file_wembed, CTLFLAG_RW,
158 &hammer2_iod_file_wembed, 0, "");
159 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_file_wzero, CTLFLAG_RW,
160 &hammer2_iod_file_wzero, 0, "");
161 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_file_wdedup, CTLFLAG_RW,
162 &hammer2_iod_file_wdedup, 0, "");
163 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_meta_write, CTLFLAG_RW,
164 &hammer2_iod_meta_write, 0, "");
165 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_indr_write, CTLFLAG_RW,
166 &hammer2_iod_indr_write, 0, "");
167 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_fmap_write, CTLFLAG_RW,
168 &hammer2_iod_fmap_write, 0, "");
169 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_volu_write, CTLFLAG_RW,
170 &hammer2_iod_volu_write, 0, "");
172 long hammer2_check_icrc32;
173 long hammer2_check_xxhash64;
174 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, check_icrc32, CTLFLAG_RW,
175 &hammer2_check_icrc32, 0, "");
176 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, check_xxhash64, CTLFLAG_RW,
177 &hammer2_check_xxhash64, 0, "");
179 static int hammer2_vfs_init(struct vfsconf *conf);
180 static int hammer2_vfs_uninit(struct vfsconf *vfsp);
181 static int hammer2_vfs_mount(struct mount *mp, char *path, caddr_t data,
183 static int hammer2_remount(hammer2_dev_t *, struct mount *, char *,
184 struct vnode *, struct ucred *);
185 static int hammer2_recovery(hammer2_dev_t *hmp);
186 static int hammer2_vfs_unmount(struct mount *mp, int mntflags);
187 static int hammer2_vfs_root(struct mount *mp, struct vnode **vpp);
188 static int hammer2_vfs_statfs(struct mount *mp, struct statfs *sbp,
190 static int hammer2_vfs_statvfs(struct mount *mp, struct statvfs *sbp,
192 static int hammer2_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
193 struct fid *fhp, struct vnode **vpp);
194 static int hammer2_vfs_vptofh(struct vnode *vp, struct fid *fhp);
195 static int hammer2_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
196 int *exflagsp, struct ucred **credanonp);
198 static int hammer2_install_volume_header(hammer2_dev_t *hmp);
199 static int hammer2_sync_scan2(struct mount *mp, struct vnode *vp, void *data);
201 static void hammer2_update_pmps(hammer2_dev_t *hmp);
203 static void hammer2_mount_helper(struct mount *mp, hammer2_pfs_t *pmp);
204 static void hammer2_unmount_helper(struct mount *mp, hammer2_pfs_t *pmp,
208 * HAMMER2 vfs operations.
210 static struct vfsops hammer2_vfsops = {
211 .vfs_init = hammer2_vfs_init,
212 .vfs_uninit = hammer2_vfs_uninit,
213 .vfs_sync = hammer2_vfs_sync,
214 .vfs_mount = hammer2_vfs_mount,
215 .vfs_unmount = hammer2_vfs_unmount,
216 .vfs_root = hammer2_vfs_root,
217 .vfs_statfs = hammer2_vfs_statfs,
218 .vfs_statvfs = hammer2_vfs_statvfs,
219 .vfs_vget = hammer2_vfs_vget,
220 .vfs_vptofh = hammer2_vfs_vptofh,
221 .vfs_fhtovp = hammer2_vfs_fhtovp,
222 .vfs_checkexp = hammer2_vfs_checkexp
225 MALLOC_DEFINE(M_HAMMER2, "HAMMER2-mount", "");
227 VFS_SET(hammer2_vfsops, hammer2, VFCF_MPSAFE);
228 MODULE_VERSION(hammer2, 1);
232 hammer2_vfs_init(struct vfsconf *conf)
234 static struct objcache_malloc_args margs_read;
235 static struct objcache_malloc_args margs_write;
236 static struct objcache_malloc_args margs_vop;
242 if (HAMMER2_BLOCKREF_BYTES != sizeof(struct hammer2_blockref))
244 if (HAMMER2_INODE_BYTES != sizeof(struct hammer2_inode_data))
246 if (HAMMER2_VOLUME_BYTES != sizeof(struct hammer2_volume_data))
250 kprintf("HAMMER2 structure size mismatch; cannot continue.\n");
252 margs_read.objsize = 65536;
253 margs_read.mtype = M_HAMMER2_DEBUFFER;
255 margs_write.objsize = 32768;
256 margs_write.mtype = M_HAMMER2_CBUFFER;
258 margs_vop.objsize = sizeof(hammer2_xop_t);
259 margs_vop.mtype = M_HAMMER2;
262 * Note thaht for the XOPS cache we want backing store allocations
263 * to use M_ZERO. This is not allowed in objcache_get() (to avoid
264 * confusion), so use the backing store function that does it. This
265 * means that initial XOPS objects are zerod but REUSED objects are
266 * not. So we are responsible for cleaning the object up sufficiently
267 * for our needs before objcache_put()ing it back (typically just the
270 cache_buffer_read = objcache_create(margs_read.mtype->ks_shortdesc,
271 0, 1, NULL, NULL, NULL,
272 objcache_malloc_alloc,
273 objcache_malloc_free,
275 cache_buffer_write = objcache_create(margs_write.mtype->ks_shortdesc,
276 0, 1, NULL, NULL, NULL,
277 objcache_malloc_alloc,
278 objcache_malloc_free,
280 cache_xops = objcache_create(margs_vop.mtype->ks_shortdesc,
281 0, 1, NULL, NULL, NULL,
282 objcache_malloc_alloc_zero,
283 objcache_malloc_free,
287 lockinit(&hammer2_mntlk, "mntlk", 0, 0);
288 TAILQ_INIT(&hammer2_mntlist);
289 TAILQ_INIT(&hammer2_pfslist);
291 hammer2_limit_dirty_chains = maxvnodes / 10;
292 if (hammer2_limit_dirty_chains > HAMMER2_LIMIT_DIRTY_CHAINS)
293 hammer2_limit_dirty_chains = HAMMER2_LIMIT_DIRTY_CHAINS;
300 hammer2_vfs_uninit(struct vfsconf *vfsp __unused)
302 objcache_destroy(cache_buffer_read);
303 objcache_destroy(cache_buffer_write);
304 objcache_destroy(cache_xops);
309 * Core PFS allocator. Used to allocate or reference the pmp structure
310 * for PFS cluster mounts and the spmp structure for media (hmp) structures.
311 * The pmp can be passed in or loaded by this function using the chain and
314 * pmp->modify_tid tracks new modify_tid transaction ids for front-end
315 * transactions. Note that synchronization does not use this field.
316 * (typically frontend operations and synchronization cannot run on the
317 * same PFS node at the same time).
322 hammer2_pfsalloc(hammer2_chain_t *chain,
323 const hammer2_inode_data_t *ripdata,
324 hammer2_tid_t modify_tid, hammer2_dev_t *force_local)
327 hammer2_inode_t *iroot;
335 * Locate or create the PFS based on the cluster id. If ripdata
336 * is NULL this is a spmp which is unique and is always allocated.
338 * If the device is mounted in local mode all PFSs are considered
339 * independent and not part of any cluster (for debugging only).
342 TAILQ_FOREACH(pmp, &hammer2_pfslist, mntentry) {
343 if (force_local != pmp->force_local)
345 if (force_local == NULL &&
346 bcmp(&pmp->pfs_clid, &ripdata->meta.pfs_clid,
347 sizeof(pmp->pfs_clid)) == 0) {
349 } else if (force_local && pmp->pfs_names[0] &&
350 strcmp(pmp->pfs_names[0], ripdata->filename) == 0) {
357 pmp = kmalloc(sizeof(*pmp), M_HAMMER2, M_WAITOK | M_ZERO);
358 pmp->force_local = force_local;
359 hammer2_trans_manage_init(pmp);
360 kmalloc_create(&pmp->minode, "HAMMER2-inodes");
361 kmalloc_create(&pmp->mmsg, "HAMMER2-pfsmsg");
362 lockinit(&pmp->lock, "pfslk", 0, 0);
363 lockinit(&pmp->lock_nlink, "h2nlink", 0, 0);
364 spin_init(&pmp->inum_spin, "hm2pfsalloc_inum");
365 spin_init(&pmp->xop_spin, "h2xop");
366 spin_init(&pmp->lru_spin, "h2lru");
367 RB_INIT(&pmp->inum_tree);
368 TAILQ_INIT(&pmp->sideq);
369 TAILQ_INIT(&pmp->lru_list);
370 spin_init(&pmp->list_spin, "hm2pfsalloc_list");
373 * Distribute backend operations to threads
375 for (i = 0; i < HAMMER2_XOPGROUPS; ++i)
376 hammer2_xop_group_init(pmp, &pmp->xop_groups[i]);
379 * Save the last media transaction id for the flusher. Set
383 pmp->pfs_clid = ripdata->meta.pfs_clid;
384 TAILQ_INSERT_TAIL(&hammer2_pfslist, pmp, mntentry);
387 * The synchronization thread may start too early, make
388 * sure it stays frozen until we are ready to let it go.
392 pmp->primary_thr.flags = HAMMER2_THREAD_FROZEN |
393 HAMMER2_THREAD_REMASTER;
398 * Create the PFS's root inode and any missing XOP helper threads.
400 if ((iroot = pmp->iroot) == NULL) {
401 iroot = hammer2_inode_get(pmp, NULL, NULL, -1);
403 iroot->meta = ripdata->meta;
405 hammer2_inode_ref(iroot);
406 hammer2_inode_unlock(iroot);
410 * Stop here if no chain is passed in.
416 * When a chain is passed in we must add it to the PFS's root
417 * inode, update pmp->pfs_types[], and update the syncronization
420 * When forcing local mode, mark the PFS as a MASTER regardless.
422 * At the moment empty spots can develop due to removals or failures.
423 * Ultimately we want to re-fill these spots but doing so might
424 * confused running code. XXX
426 hammer2_inode_ref(iroot);
427 hammer2_mtx_ex(&iroot->lock);
428 j = iroot->cluster.nchains;
430 kprintf("add PFS to pmp %p[%d]\n", pmp, j);
432 if (j == HAMMER2_MAXCLUSTER) {
433 kprintf("hammer2_mount: cluster full!\n");
434 /* XXX fatal error? */
436 KKASSERT(chain->pmp == NULL);
438 hammer2_chain_ref(chain);
439 iroot->cluster.array[j].chain = chain;
441 pmp->pfs_types[j] = HAMMER2_PFSTYPE_MASTER;
443 pmp->pfs_types[j] = ripdata->meta.pfs_type;
444 pmp->pfs_names[j] = kstrdup(ripdata->filename, M_HAMMER2);
445 pmp->pfs_hmps[j] = chain->hmp;
448 * If the PFS is already mounted we must account
449 * for the mount_count here.
452 ++chain->hmp->mount_count;
455 * May have to fixup dirty chain tracking. Previous
456 * pmp was NULL so nothing to undo.
458 if (chain->flags & HAMMER2_CHAIN_MODIFIED)
459 hammer2_pfs_memory_inc(pmp);
462 iroot->cluster.nchains = j;
465 * Update nmasters from any PFS inode which is part of the cluster.
466 * It is possible that this will result in a value which is too
467 * high. MASTER PFSs are authoritative for pfs_nmasters and will
468 * override this value later on.
470 * (This informs us of masters that might not currently be
471 * discoverable by this mount).
473 if (ripdata && pmp->pfs_nmasters < ripdata->meta.pfs_nmasters) {
474 pmp->pfs_nmasters = ripdata->meta.pfs_nmasters;
478 * Count visible masters. Masters are usually added with
479 * ripdata->meta.pfs_nmasters set to 1. This detects when there
480 * are more (XXX and must update the master inodes).
483 for (i = 0; i < iroot->cluster.nchains; ++i) {
484 if (pmp->pfs_types[i] == HAMMER2_PFSTYPE_MASTER)
487 if (pmp->pfs_nmasters < count)
488 pmp->pfs_nmasters = count;
491 * Create missing synchronization and support threads.
493 * Single-node masters (including snapshots) have nothing to
494 * synchronize and do not require this thread.
496 * Multi-node masters or any number of soft masters, slaves, copy,
497 * or other PFS types need the thread.
499 * Each thread is responsible for its particular cluster index.
500 * We use independent threads so stalls or mismatches related to
501 * any given target do not affect other targets.
503 for (i = 0; i < iroot->cluster.nchains; ++i) {
505 * Single-node masters (including snapshots) have nothing
506 * to synchronize and will make direct xops support calls,
507 * thus they do not require this thread.
509 * Note that there can be thousands of snapshots. We do not
510 * want to create thousands of threads.
512 if (pmp->pfs_nmasters <= 1 &&
513 pmp->pfs_types[i] == HAMMER2_PFSTYPE_MASTER) {
518 * Sync support thread
520 if (pmp->sync_thrs[i].td == NULL) {
521 hammer2_thr_create(&pmp->sync_thrs[i], pmp,
523 hammer2_primary_sync_thread);
528 * Create missing Xop threads
530 * NOTE: We create helper threads for all mounted PFSs or any
531 * PFSs with 2+ nodes (so the sync thread can update them,
532 * even if not mounted).
534 if (pmp->mp || iroot->cluster.nchains >= 2)
535 hammer2_xop_helper_create(pmp);
537 hammer2_mtx_unlock(&iroot->lock);
538 hammer2_inode_drop(iroot);
544 * Deallocate an element of a probed PFS. If destroying and this is a
545 * MASTER, adjust nmasters.
547 * This function does not physically destroy the PFS element in its device
548 * under the super-root (see hammer2_ioctl_pfs_delete()).
551 hammer2_pfsdealloc(hammer2_pfs_t *pmp, int clindex, int destroying)
553 hammer2_inode_t *iroot;
554 hammer2_chain_t *chain;
558 * Cleanup our reference on iroot. iroot is (should) not be needed
566 * XXX flush after acquiring the iroot lock.
567 * XXX clean out the cluster index from all inode structures.
569 hammer2_thr_delete(&pmp->sync_thrs[clindex]);
572 * Remove the cluster index from the group. If destroying
573 * the PFS and this is a master, adjust pfs_nmasters.
575 hammer2_mtx_ex(&iroot->lock);
576 chain = iroot->cluster.array[clindex].chain;
577 iroot->cluster.array[clindex].chain = NULL;
579 switch(pmp->pfs_types[clindex]) {
580 case HAMMER2_PFSTYPE_MASTER:
581 if (destroying && pmp->pfs_nmasters > 0)
583 /* XXX adjust ripdata->meta.pfs_nmasters */
588 pmp->pfs_types[clindex] = HAMMER2_PFSTYPE_NONE;
590 hammer2_mtx_unlock(&iroot->lock);
596 atomic_set_int(&chain->flags, HAMMER2_CHAIN_RELEASE);
597 hammer2_chain_drop(chain);
601 * Terminate all XOP threads for the cluster index.
603 for (j = 0; j < HAMMER2_XOPGROUPS; ++j)
604 hammer2_thr_delete(&pmp->xop_groups[j].thrs[clindex]);
609 * Destroy a PFS, typically only occurs after the last mount on a device
613 hammer2_pfsfree(hammer2_pfs_t *pmp)
615 hammer2_inode_t *iroot;
616 hammer2_chain_t *chain;
621 * Cleanup our reference on iroot. iroot is (should) not be needed
624 TAILQ_REMOVE(&hammer2_pfslist, pmp, mntentry);
628 for (i = 0; i < iroot->cluster.nchains; ++i) {
629 hammer2_thr_delete(&pmp->sync_thrs[i]);
630 for (j = 0; j < HAMMER2_XOPGROUPS; ++j)
631 hammer2_thr_delete(&pmp->xop_groups[j].thrs[i]);
633 #if REPORT_REFS_ERRORS
634 if (pmp->iroot->refs != 1)
635 kprintf("PMP->IROOT %p REFS WRONG %d\n",
636 pmp->iroot, pmp->iroot->refs);
638 KKASSERT(pmp->iroot->refs == 1);
640 /* ref for pmp->iroot */
641 hammer2_inode_drop(pmp->iroot);
646 * Cleanup chains remaining on LRU list.
648 kprintf("pfsfree: %p lrucount=%d\n", pmp, pmp->lru_count);
649 while ((chain = TAILQ_FIRST(&pmp->lru_list)) != NULL) {
650 hammer2_chain_ref(chain);
651 atomic_set_int(&chain->flags, HAMMER2_CHAIN_RELEASE);
652 hammer2_chain_drop(chain);
656 * Free remaining pmp resources
658 kmalloc_destroy(&pmp->mmsg);
659 kmalloc_destroy(&pmp->minode);
661 kfree(pmp, M_HAMMER2);
665 * Remove all references to hmp from the pfs list. Any PFS which becomes
666 * empty is terminated and freed.
671 hammer2_pfsfree_scan(hammer2_dev_t *hmp)
674 hammer2_inode_t *iroot;
675 hammer2_chain_t *rchain;
681 TAILQ_FOREACH(pmp, &hammer2_pfslist, mntentry) {
682 if ((iroot = pmp->iroot) == NULL)
684 if (hmp->spmp == pmp) {
685 kprintf("unmount hmp %p remove spmp %p\n",
691 * Determine if this PFS is affected. If it is we must
692 * freeze all management threads and lock its iroot.
694 * Freezing a management thread forces it idle, operations
695 * in-progress will be aborted and it will have to start
696 * over again when unfrozen, or exit if told to exit.
698 for (i = 0; i < HAMMER2_MAXCLUSTER; ++i) {
699 if (pmp->pfs_hmps[i] == hmp)
702 if (i != HAMMER2_MAXCLUSTER) {
704 * Make sure all synchronization threads are locked
707 for (i = 0; i < HAMMER2_MAXCLUSTER; ++i) {
708 if (pmp->pfs_hmps[i] == NULL)
710 hammer2_thr_freeze_async(&pmp->sync_thrs[i]);
711 for (j = 0; j < HAMMER2_XOPGROUPS; ++j) {
712 hammer2_thr_freeze_async(
713 &pmp->xop_groups[j].thrs[i]);
716 for (i = 0; i < HAMMER2_MAXCLUSTER; ++i) {
717 if (pmp->pfs_hmps[i] == NULL)
719 hammer2_thr_freeze(&pmp->sync_thrs[i]);
720 for (j = 0; j < HAMMER2_XOPGROUPS; ++j) {
722 &pmp->xop_groups[j].thrs[i]);
727 * Lock the inode and clean out matching chains.
728 * Note that we cannot use hammer2_inode_lock_*()
729 * here because that would attempt to validate the
730 * cluster that we are in the middle of ripping
733 * WARNING! We are working directly on the inodes
736 hammer2_mtx_ex(&iroot->lock);
739 * Remove the chain from matching elements of the PFS.
741 for (i = 0; i < HAMMER2_MAXCLUSTER; ++i) {
742 if (pmp->pfs_hmps[i] != hmp)
744 hammer2_thr_delete(&pmp->sync_thrs[i]);
745 for (j = 0; j < HAMMER2_XOPGROUPS; ++j) {
747 &pmp->xop_groups[j].thrs[i]);
749 rchain = iroot->cluster.array[i].chain;
750 iroot->cluster.array[i].chain = NULL;
751 pmp->pfs_types[i] = 0;
752 if (pmp->pfs_names[i]) {
753 kfree(pmp->pfs_names[i], M_HAMMER2);
754 pmp->pfs_names[i] = NULL;
757 hammer2_chain_drop(rchain);
759 if (iroot->cluster.focus == rchain)
760 iroot->cluster.focus = NULL;
762 pmp->pfs_hmps[i] = NULL;
764 hammer2_mtx_unlock(&iroot->lock);
765 didfreeze = 1; /* remaster, unfreeze down below */
771 * Cleanup trailing chains. Gaps may remain.
773 for (i = HAMMER2_MAXCLUSTER - 1; i >= 0; --i) {
774 if (pmp->pfs_hmps[i])
777 iroot->cluster.nchains = i + 1;
780 * If the PMP has no elements remaining we can destroy it.
781 * (this will transition management threads from frozen->exit).
783 if (iroot->cluster.nchains == 0) {
784 kprintf("unmount hmp %p last ref to PMP=%p\n",
786 hammer2_pfsfree(pmp);
791 * If elements still remain we need to set the REMASTER
792 * flag and unfreeze it.
795 for (i = 0; i < HAMMER2_MAXCLUSTER; ++i) {
796 if (pmp->pfs_hmps[i] == NULL)
798 hammer2_thr_remaster(&pmp->sync_thrs[i]);
799 hammer2_thr_unfreeze(&pmp->sync_thrs[i]);
800 for (j = 0; j < HAMMER2_XOPGROUPS; ++j) {
801 hammer2_thr_remaster(
802 &pmp->xop_groups[j].thrs[i]);
803 hammer2_thr_unfreeze(
804 &pmp->xop_groups[j].thrs[i]);
812 * Mount or remount HAMMER2 fileystem from physical media
815 * mp mount point structure
821 * mp mount point structure
822 * path path to mount point
823 * data pointer to argument structure in user space
824 * volume volume path (device@LABEL form)
825 * hflags user mount flags
826 * cred user credentials
833 hammer2_vfs_mount(struct mount *mp, char *path, caddr_t data,
836 struct hammer2_mount_info info;
840 hammer2_dev_t *force_local;
841 hammer2_key_t key_next;
842 hammer2_key_t key_dummy;
845 struct nlookupdata nd;
846 hammer2_chain_t *parent;
847 hammer2_chain_t *chain;
848 hammer2_cluster_t *cluster;
849 const hammer2_inode_data_t *ripdata;
850 hammer2_blockref_t bref;
852 char devstr[MNAMELEN];
869 kprintf("hammer2_mount\n");
875 bzero(&info, sizeof(info));
876 info.cluster_fd = -1;
877 ksnprintf(devstr, sizeof(devstr), "%s",
878 mp->mnt_stat.f_mntfromname);
879 kprintf("hammer2_mount: root '%s'\n", devstr);
882 * Non-root mount or updating a mount
884 error = copyin(data, &info, sizeof(info));
888 error = copyinstr(info.volume, devstr, MNAMELEN - 1, &done);
894 * Extract device and label, automatically mount @BOOT, @ROOT, or @DATA
895 * if no label specified, based on the partition id. Error out if no
896 * partition id. This is strictly a convenience to match the
897 * default label created by newfs_hammer2, our preference is
898 * that a label always be specified.
901 label = strchr(devstr, '@');
902 if (label && ((label + 1) - dev) > done)
904 if (label && label == devstr)
906 if (label == NULL || label[1] == 0) {
910 label = devstr + strlen(devstr);
928 kprintf("hammer2_mount: dev=\"%s\" label=\"%s\"\n",
931 if (mp->mnt_flag & MNT_UPDATE) {
933 * Update mount. Note that pmp->iroot->cluster is
934 * an inode-embedded cluster and thus cannot be
937 * XXX HAMMER2 needs to implement NFS export via
941 pmp->hflags = info.hflags;
942 cluster = &pmp->iroot->cluster;
943 for (i = 0; i < cluster->nchains; ++i) {
944 if (cluster->array[i].chain == NULL)
946 hmp = cluster->array[i].chain->hmp;
948 error = hammer2_remount(hmp, mp, path,
960 * If a path is specified and dev is not an empty string, lookup the
961 * name and verify that it referes to a block device.
963 * If a path is specified and dev is an empty string we fall through
964 * and locate the label in the hmp search.
966 if (path && *dev != 0) {
967 error = nlookup_init(&nd, dev, UIO_SYSSPACE, NLC_FOLLOW);
969 error = nlookup(&nd);
971 error = cache_vref(&nd.nl_nch, nd.nl_cred, &devvp);
973 } else if (path == NULL) {
975 cdev_t cdev = kgetdiskbyname(dev);
976 error = bdevvp(cdev, &devvp);
978 kprintf("hammer2: cannot find '%s'\n", dev);
981 * We will locate the hmp using the label in the hmp loop.
986 if (error == 0 && devvp) {
987 if (vn_isdisk(devvp, &error))
988 error = vfs_mountedon(devvp);
992 * Determine if the device has already been mounted. After this
993 * check hmp will be non-NULL if we are doing the second or more
994 * hammer2 mounts from the same device.
996 lockmgr(&hammer2_mntlk, LK_EXCLUSIVE);
1001 TAILQ_FOREACH(hmp, &hammer2_mntlist, mntentry) {
1002 if (hmp->devvp == devvp)
1005 } else if (error == 0) {
1007 * Match the label to a pmp already probed.
1009 TAILQ_FOREACH(pmp, &hammer2_pfslist, mntentry) {
1010 for (i = 0; i < HAMMER2_MAXCLUSTER; ++i) {
1011 if (pmp->pfs_names[i] &&
1012 strcmp(pmp->pfs_names[i], label) == 0) {
1013 hmp = pmp->pfs_hmps[i];
1025 * Open the device if this isn't a secondary mount and construct
1026 * the H2 device mount (hmp).
1029 hammer2_chain_t *schain;
1032 if (error == 0 && vcount(devvp) > 0)
1036 * Now open the device
1039 ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
1040 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
1041 error = vinvalbuf(devvp, V_SAVE, 0, 0);
1043 error = VOP_OPEN(devvp,
1044 ronly ? FREAD : FREAD | FWRITE,
1049 if (error && devvp) {
1054 lockmgr(&hammer2_mntlk, LK_RELEASE);
1057 hmp = kmalloc(sizeof(*hmp), M_HAMMER2, M_WAITOK | M_ZERO);
1058 ksnprintf(hmp->devrepname, sizeof(hmp->devrepname), "%s", dev);
1061 hmp->hflags = info.hflags & HMNT2_DEVFLAGS;
1062 kmalloc_create(&hmp->mchain, "HAMMER2-chains");
1063 TAILQ_INSERT_TAIL(&hammer2_mntlist, hmp, mntentry);
1064 RB_INIT(&hmp->iotree);
1065 spin_init(&hmp->io_spin, "hm2mount_io");
1066 spin_init(&hmp->list_spin, "hm2mount_list");
1067 TAILQ_INIT(&hmp->flushq);
1069 lockinit(&hmp->vollk, "h2vol", 0, 0);
1070 lockinit(&hmp->bulklk, "h2bulk", 0, 0);
1073 * vchain setup. vchain.data is embedded.
1074 * vchain.refs is initialized and will never drop to 0.
1076 * NOTE! voldata is not yet loaded.
1078 hmp->vchain.hmp = hmp;
1079 hmp->vchain.refs = 1;
1080 hmp->vchain.data = (void *)&hmp->voldata;
1081 hmp->vchain.bref.type = HAMMER2_BREF_TYPE_VOLUME;
1082 hmp->vchain.bref.data_off = 0 | HAMMER2_PBUFRADIX;
1083 hmp->vchain.bref.mirror_tid = hmp->voldata.mirror_tid;
1085 hammer2_chain_core_init(&hmp->vchain);
1086 /* hmp->vchain.u.xxx is left NULL */
1089 * fchain setup. fchain.data is embedded.
1090 * fchain.refs is initialized and will never drop to 0.
1092 * The data is not used but needs to be initialized to
1093 * pass assertion muster. We use this chain primarily
1094 * as a placeholder for the freemap's top-level RBTREE
1095 * so it does not interfere with the volume's topology
1098 hmp->fchain.hmp = hmp;
1099 hmp->fchain.refs = 1;
1100 hmp->fchain.data = (void *)&hmp->voldata.freemap_blockset;
1101 hmp->fchain.bref.type = HAMMER2_BREF_TYPE_FREEMAP;
1102 hmp->fchain.bref.data_off = 0 | HAMMER2_PBUFRADIX;
1103 hmp->fchain.bref.mirror_tid = hmp->voldata.freemap_tid;
1104 hmp->fchain.bref.methods =
1105 HAMMER2_ENC_CHECK(HAMMER2_CHECK_FREEMAP) |
1106 HAMMER2_ENC_COMP(HAMMER2_COMP_NONE);
1108 hammer2_chain_core_init(&hmp->fchain);
1109 /* hmp->fchain.u.xxx is left NULL */
1112 * Install the volume header and initialize fields from
1115 error = hammer2_install_volume_header(hmp);
1117 hammer2_unmount_helper(mp, NULL, hmp);
1118 lockmgr(&hammer2_mntlk, LK_RELEASE);
1119 hammer2_vfs_unmount(mp, MNT_FORCE);
1124 * Really important to get these right or flush will get
1127 hmp->spmp = hammer2_pfsalloc(NULL, NULL, 0, NULL);
1128 kprintf("alloc spmp %p tid %016jx\n",
1129 hmp->spmp, hmp->voldata.mirror_tid);
1133 * Dummy-up vchain and fchain's modify_tid. mirror_tid
1134 * is inherited from the volume header.
1137 hmp->vchain.bref.mirror_tid = hmp->voldata.mirror_tid;
1138 hmp->vchain.bref.modify_tid = hmp->vchain.bref.mirror_tid;
1139 hmp->vchain.pmp = spmp;
1140 hmp->fchain.bref.mirror_tid = hmp->voldata.freemap_tid;
1141 hmp->fchain.bref.modify_tid = hmp->fchain.bref.mirror_tid;
1142 hmp->fchain.pmp = spmp;
1145 * First locate the super-root inode, which is key 0
1146 * relative to the volume header's blockset.
1148 * Then locate the root inode by scanning the directory keyspace
1149 * represented by the label.
1151 parent = hammer2_chain_lookup_init(&hmp->vchain, 0);
1152 schain = hammer2_chain_lookup(&parent, &key_dummy,
1153 HAMMER2_SROOT_KEY, HAMMER2_SROOT_KEY,
1155 hammer2_chain_lookup_done(parent);
1156 if (schain == NULL) {
1157 kprintf("hammer2_mount: invalid super-root\n");
1158 hammer2_unmount_helper(mp, NULL, hmp);
1159 lockmgr(&hammer2_mntlk, LK_RELEASE);
1160 hammer2_vfs_unmount(mp, MNT_FORCE);
1163 if (schain->error) {
1164 kprintf("hammer2_mount: error %s reading super-root\n",
1165 hammer2_error_str(schain->error));
1166 hammer2_chain_unlock(schain);
1167 hammer2_chain_drop(schain);
1169 hammer2_unmount_helper(mp, NULL, hmp);
1170 lockmgr(&hammer2_mntlk, LK_RELEASE);
1171 hammer2_vfs_unmount(mp, MNT_FORCE);
1176 * The super-root always uses an inode_tid of 1 when
1179 spmp->inode_tid = 1;
1180 spmp->modify_tid = schain->bref.modify_tid + 1;
1183 * Sanity-check schain's pmp and finish initialization.
1184 * Any chain belonging to the super-root topology should
1185 * have a NULL pmp (not even set to spmp).
1187 ripdata = &hammer2_chain_rdata(schain)->ipdata;
1188 KKASSERT(schain->pmp == NULL);
1189 spmp->pfs_clid = ripdata->meta.pfs_clid;
1192 * Replace the dummy spmp->iroot with a real one. It's
1193 * easier to just do a wholesale replacement than to try
1194 * to update the chain and fixup the iroot fields.
1196 * The returned inode is locked with the supplied cluster.
1198 cluster = hammer2_cluster_from_chain(schain);
1199 hammer2_inode_drop(spmp->iroot);
1201 spmp->iroot = hammer2_inode_get(spmp, NULL, cluster, -1);
1202 spmp->spmp_hmp = hmp;
1203 spmp->pfs_types[0] = ripdata->meta.pfs_type;
1204 spmp->pfs_hmps[0] = hmp;
1205 hammer2_inode_ref(spmp->iroot);
1206 hammer2_inode_unlock(spmp->iroot);
1207 hammer2_cluster_unlock(cluster);
1208 hammer2_cluster_drop(cluster);
1210 /* leave spmp->iroot with one ref */
1212 if ((mp->mnt_flag & MNT_RDONLY) == 0) {
1213 error = hammer2_recovery(hmp);
1214 /* XXX do something with error */
1216 hammer2_update_pmps(hmp);
1217 hammer2_iocom_init(hmp);
1220 * Ref the cluster management messaging descriptor. The mount
1221 * program deals with the other end of the communications pipe.
1223 * Root mounts typically do not supply one.
1225 if (info.cluster_fd >= 0) {
1226 fp = holdfp(curproc->p_fd, info.cluster_fd, -1);
1228 hammer2_cluster_reconnect(hmp, fp);
1230 kprintf("hammer2_mount: bad cluster_fd!\n");
1235 if (info.hflags & HMNT2_DEVFLAGS) {
1236 kprintf("hammer2: Warning: mount flags pertaining "
1237 "to the whole device may only be specified "
1238 "on the first mount of the device: %08x\n",
1239 info.hflags & HMNT2_DEVFLAGS);
1244 * Force local mount (disassociate all PFSs from their clusters).
1245 * Used primarily for debugging.
1247 force_local = (hmp->hflags & HMNT2_LOCAL) ? hmp : NULL;
1250 * Lookup the mount point under the media-localized super-root.
1251 * Scanning hammer2_pfslist doesn't help us because it represents
1252 * PFS cluster ids which can aggregate several named PFSs together.
1254 * cluster->pmp will incorrectly point to spmp and must be fixed
1257 hammer2_inode_lock(spmp->iroot, 0);
1258 parent = hammer2_inode_chain(spmp->iroot, 0, HAMMER2_RESOLVE_ALWAYS);
1259 lhc = hammer2_dirhash(label, strlen(label));
1260 chain = hammer2_chain_lookup(&parent, &key_next,
1261 lhc, lhc + HAMMER2_DIRHASH_LOMASK,
1264 if (chain->bref.type == HAMMER2_BREF_TYPE_INODE &&
1265 strcmp(label, chain->data->ipdata.filename) == 0) {
1268 chain = hammer2_chain_next(&parent, chain, &key_next,
1270 lhc + HAMMER2_DIRHASH_LOMASK,
1274 hammer2_chain_unlock(parent);
1275 hammer2_chain_drop(parent);
1277 hammer2_inode_unlock(spmp->iroot);
1280 * PFS could not be found?
1282 if (chain == NULL) {
1283 kprintf("hammer2_mount: PFS label not found\n");
1284 hammer2_unmount_helper(mp, NULL, hmp);
1285 lockmgr(&hammer2_mntlk, LK_RELEASE);
1286 hammer2_vfs_unmount(mp, MNT_FORCE);
1292 * Acquire the pmp structure (it should have already been allocated
1293 * via hammer2_update_pmps() so do not pass cluster in to add to
1294 * available chains).
1296 * Check if the cluster has already been mounted. A cluster can
1297 * only be mounted once, use null mounts to mount additional copies.
1299 ripdata = &chain->data->ipdata;
1301 pmp = hammer2_pfsalloc(NULL, ripdata,
1302 bref.modify_tid, force_local);
1303 hammer2_chain_unlock(chain);
1304 hammer2_chain_drop(chain);
1309 kprintf("hammer2_mount hmp=%p pmp=%p\n", hmp, pmp);
1312 kprintf("hammer2_mount: PFS already mounted!\n");
1313 hammer2_unmount_helper(mp, NULL, hmp);
1314 lockmgr(&hammer2_mntlk, LK_RELEASE);
1315 hammer2_vfs_unmount(mp, MNT_FORCE);
1320 pmp->hflags = info.hflags;
1321 mp->mnt_flag = MNT_LOCAL;
1322 mp->mnt_kern_flag |= MNTK_ALL_MPSAFE; /* all entry pts are SMP */
1323 mp->mnt_kern_flag |= MNTK_THR_SYNC; /* new vsyncscan semantics */
1326 * required mount structure initializations
1328 mp->mnt_stat.f_iosize = HAMMER2_PBUFSIZE;
1329 mp->mnt_stat.f_bsize = HAMMER2_PBUFSIZE;
1331 mp->mnt_vstat.f_frsize = HAMMER2_PBUFSIZE;
1332 mp->mnt_vstat.f_bsize = HAMMER2_PBUFSIZE;
1337 mp->mnt_iosize_max = MAXPHYS;
1340 * Connect up mount pointers.
1342 hammer2_mount_helper(mp, pmp);
1344 lockmgr(&hammer2_mntlk, LK_RELEASE);
1350 vfs_add_vnodeops(mp, &hammer2_vnode_vops, &mp->mnt_vn_norm_ops);
1351 vfs_add_vnodeops(mp, &hammer2_spec_vops, &mp->mnt_vn_spec_ops);
1352 vfs_add_vnodeops(mp, &hammer2_fifo_vops, &mp->mnt_vn_fifo_ops);
1355 copyinstr(info.volume, mp->mnt_stat.f_mntfromname,
1356 MNAMELEN - 1, &size);
1357 bzero(mp->mnt_stat.f_mntfromname + size, MNAMELEN - size);
1358 } /* else root mount, already in there */
1360 bzero(mp->mnt_stat.f_mntonname, sizeof(mp->mnt_stat.f_mntonname));
1362 copyinstr(path, mp->mnt_stat.f_mntonname,
1363 sizeof(mp->mnt_stat.f_mntonname) - 1,
1367 mp->mnt_stat.f_mntonname[0] = '/';
1371 * Initial statfs to prime mnt_stat.
1373 hammer2_vfs_statfs(mp, &mp->mnt_stat, cred);
1379 * Scan PFSs under the super-root and create hammer2_pfs structures.
1383 hammer2_update_pmps(hammer2_dev_t *hmp)
1385 const hammer2_inode_data_t *ripdata;
1386 hammer2_chain_t *parent;
1387 hammer2_chain_t *chain;
1388 hammer2_blockref_t bref;
1389 hammer2_dev_t *force_local;
1390 hammer2_pfs_t *spmp;
1392 hammer2_key_t key_next;
1393 int cache_index = -1;
1396 * Force local mount (disassociate all PFSs from their clusters).
1397 * Used primarily for debugging.
1399 force_local = (hmp->hflags & HMNT2_LOCAL) ? hmp : NULL;
1402 * Lookup mount point under the media-localized super-root.
1404 * cluster->pmp will incorrectly point to spmp and must be fixed
1408 hammer2_inode_lock(spmp->iroot, 0);
1409 parent = hammer2_inode_chain(spmp->iroot, 0, HAMMER2_RESOLVE_ALWAYS);
1410 chain = hammer2_chain_lookup(&parent, &key_next,
1411 HAMMER2_KEY_MIN, HAMMER2_KEY_MAX,
1414 if (chain->bref.type != HAMMER2_BREF_TYPE_INODE)
1416 ripdata = &chain->data->ipdata;
1418 kprintf("ADD LOCAL PFS: %s\n", ripdata->filename);
1420 pmp = hammer2_pfsalloc(chain, ripdata,
1421 bref.modify_tid, force_local);
1422 chain = hammer2_chain_next(&parent, chain, &key_next,
1423 key_next, HAMMER2_KEY_MAX,
1427 hammer2_chain_unlock(parent);
1428 hammer2_chain_drop(parent);
1430 hammer2_inode_unlock(spmp->iroot);
1435 hammer2_remount(hammer2_dev_t *hmp, struct mount *mp, char *path __unused,
1436 struct vnode *devvp, struct ucred *cred)
1440 if (hmp->ronly && (mp->mnt_kern_flag & MNTK_WANTRDWR)) {
1441 error = hammer2_recovery(hmp);
1450 hammer2_vfs_unmount(struct mount *mp, int mntflags)
1461 lockmgr(&hammer2_mntlk, LK_EXCLUSIVE);
1464 * If mount initialization proceeded far enough we must flush
1465 * its vnodes and sync the underlying mount points. Three syncs
1466 * are required to fully flush the filesystem (freemap updates lag
1467 * by one flush, and one extra for safety).
1469 if (mntflags & MNT_FORCE)
1474 error = vflush(mp, 0, flags);
1477 hammer2_vfs_sync(mp, MNT_WAIT);
1478 hammer2_vfs_sync(mp, MNT_WAIT);
1479 hammer2_vfs_sync(mp, MNT_WAIT);
1483 * Cleanup the frontend support XOPS threads
1485 hammer2_xop_helper_cleanup(pmp);
1488 hammer2_unmount_helper(mp, pmp, NULL);
1492 lockmgr(&hammer2_mntlk, LK_RELEASE);
1498 * Mount helper, hook the system mount into our PFS.
1499 * The mount lock is held.
1501 * We must bump the mount_count on related devices for any
1506 hammer2_mount_helper(struct mount *mp, hammer2_pfs_t *pmp)
1508 hammer2_cluster_t *cluster;
1509 hammer2_chain_t *rchain;
1512 mp->mnt_data = (qaddr_t)pmp;
1516 * After pmp->mp is set we have to adjust hmp->mount_count.
1518 cluster = &pmp->iroot->cluster;
1519 for (i = 0; i < cluster->nchains; ++i) {
1520 rchain = cluster->array[i].chain;
1523 ++rchain->hmp->mount_count;
1524 kprintf("hammer2_mount hmp=%p ++mount_count=%d\n",
1525 rchain->hmp, rchain->hmp->mount_count);
1529 * Create missing Xop threads
1531 hammer2_xop_helper_create(pmp);
1535 * Mount helper, unhook the system mount from our PFS.
1536 * The mount lock is held.
1538 * If hmp is supplied a mount responsible for being the first to open
1539 * the block device failed and the block device and all PFSs using the
1540 * block device must be cleaned up.
1542 * If pmp is supplied multiple devices might be backing the PFS and each
1543 * must be disconnected. This might not be the last PFS using some of the
1544 * underlying devices. Also, we have to adjust our hmp->mount_count
1545 * accounting for the devices backing the pmp which is now undergoing an
1550 hammer2_unmount_helper(struct mount *mp, hammer2_pfs_t *pmp, hammer2_dev_t *hmp)
1552 hammer2_cluster_t *cluster;
1553 hammer2_chain_t *rchain;
1554 struct vnode *devvp;
1560 * If no device supplied this is a high-level unmount and we have to
1561 * to disconnect the mount, adjust mount_count, and locate devices
1562 * that might now have no mounts.
1565 KKASSERT(hmp == NULL);
1566 KKASSERT((void *)(intptr_t)mp->mnt_data == pmp);
1568 mp->mnt_data = NULL;
1571 * After pmp->mp is cleared we have to account for
1574 cluster = &pmp->iroot->cluster;
1575 for (i = 0; i < cluster->nchains; ++i) {
1576 rchain = cluster->array[i].chain;
1579 --rchain->hmp->mount_count;
1580 kprintf("hammer2_unmount hmp=%p --mount_count=%d\n",
1581 rchain->hmp, rchain->hmp->mount_count);
1582 /* scrapping hmp now may invalidate the pmp */
1585 TAILQ_FOREACH(hmp, &hammer2_mntlist, mntentry) {
1586 if (hmp->mount_count == 0) {
1587 hammer2_unmount_helper(NULL, NULL, hmp);
1595 * Try to terminate the block device. We can't terminate it if
1596 * there are still PFSs referencing it.
1598 kprintf("hammer2_unmount hmp=%p mount_count=%d\n",
1599 hmp, hmp->mount_count);
1600 if (hmp->mount_count)
1603 hammer2_pfsfree_scan(hmp);
1604 hammer2_dev_exlock(hmp); /* XXX order */
1607 * Cycle the volume data lock as a safety (probably not needed any
1608 * more). To ensure everything is out we need to flush at least
1609 * three times. (1) The running of the sideq can dirty the
1610 * filesystem, (2) A normal flush can dirty the freemap, and
1611 * (3) ensure that the freemap is fully synchronized.
1613 * The next mount's recovery scan can clean everything up but we want
1614 * to leave the filesystem in a 100% clean state on a normal unmount.
1617 hammer2_voldata_lock(hmp);
1618 hammer2_voldata_unlock(hmp);
1620 hammer2_iocom_uninit(hmp);
1622 if ((hmp->vchain.flags | hmp->fchain.flags) &
1623 HAMMER2_CHAIN_FLUSH_MASK) {
1624 kprintf("hammer2_unmount: chains left over "
1625 "after final sync\n");
1626 kprintf(" vchain %08x\n", hmp->vchain.flags);
1627 kprintf(" fchain %08x\n", hmp->fchain.flags);
1629 if (hammer2_debug & 0x0010)
1630 Debugger("entered debugger");
1633 KKASSERT(hmp->spmp == NULL);
1636 * Finish up with the device vnode
1638 if ((devvp = hmp->devvp) != NULL) {
1639 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
1640 vinvalbuf(devvp, (ronly ? 0 : V_SAVE), 0, 0);
1642 VOP_CLOSE(devvp, (ronly ? FREAD : FREAD|FWRITE), NULL);
1649 * Clear vchain/fchain flags that might prevent final cleanup
1652 if (hmp->vchain.flags & HAMMER2_CHAIN_MODIFIED) {
1653 atomic_add_long(&hammer2_count_modified_chains, -1);
1654 atomic_clear_int(&hmp->vchain.flags, HAMMER2_CHAIN_MODIFIED);
1655 hammer2_pfs_memory_wakeup(hmp->vchain.pmp);
1657 if (hmp->vchain.flags & HAMMER2_CHAIN_UPDATE) {
1658 atomic_clear_int(&hmp->vchain.flags, HAMMER2_CHAIN_UPDATE);
1661 if (hmp->fchain.flags & HAMMER2_CHAIN_MODIFIED) {
1662 atomic_add_long(&hammer2_count_modified_chains, -1);
1663 atomic_clear_int(&hmp->fchain.flags, HAMMER2_CHAIN_MODIFIED);
1664 hammer2_pfs_memory_wakeup(hmp->fchain.pmp);
1666 if (hmp->fchain.flags & HAMMER2_CHAIN_UPDATE) {
1667 atomic_clear_int(&hmp->fchain.flags, HAMMER2_CHAIN_UPDATE);
1671 * Final drop of embedded freemap root chain to
1672 * clean up fchain.core (fchain structure is not
1673 * flagged ALLOCATED so it is cleaned out and then
1676 hammer2_chain_drop(&hmp->fchain);
1679 * Final drop of embedded volume root chain to clean
1680 * up vchain.core (vchain structure is not flagged
1681 * ALLOCATED so it is cleaned out and then left to
1685 hammer2_dump_chain(&hmp->vchain, 0, &dumpcnt, 'v');
1687 hammer2_dump_chain(&hmp->fchain, 0, &dumpcnt, 'f');
1688 hammer2_dev_unlock(hmp);
1689 hammer2_chain_drop(&hmp->vchain);
1691 hammer2_io_cleanup(hmp, &hmp->iotree);
1692 if (hmp->iofree_count) {
1693 kprintf("io_cleanup: %d I/O's left hanging\n",
1697 TAILQ_REMOVE(&hammer2_mntlist, hmp, mntentry);
1698 kmalloc_destroy(&hmp->mchain);
1699 kfree(hmp, M_HAMMER2);
1703 hammer2_vfs_vget(struct mount *mp, struct vnode *dvp,
1704 ino_t ino, struct vnode **vpp)
1706 hammer2_xop_lookup_t *xop;
1708 hammer2_inode_t *ip;
1712 inum = (hammer2_tid_t)ino & HAMMER2_DIRHASH_USERMSK;
1718 * Easy if we already have it cached
1720 ip = hammer2_inode_lookup(pmp, inum);
1722 hammer2_inode_lock(ip, HAMMER2_RESOLVE_SHARED);
1723 *vpp = hammer2_igetv(ip, &error);
1724 hammer2_inode_unlock(ip);
1725 hammer2_inode_drop(ip); /* from lookup */
1731 * Otherwise we have to find the inode
1733 xop = hammer2_xop_alloc(pmp->iroot, 0);
1735 hammer2_xop_start(&xop->head, hammer2_xop_lookup);
1736 error = hammer2_xop_collect(&xop->head, 0);
1739 if (hammer2_cluster_rdata(&xop->head.cluster) == NULL) {
1740 kprintf("vget: no collect error but also no rdata\n");
1741 kprintf("xop %p\n", xop);
1742 while ((hammer2_debug & 0x80000) == 0) {
1743 tsleep(xop, PCATCH, "wait", hz * 10);
1747 ip = hammer2_inode_get(pmp, NULL, &xop->head.cluster, -1);
1750 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1753 *vpp = hammer2_igetv(ip, &error);
1754 hammer2_inode_unlock(ip);
1764 hammer2_vfs_root(struct mount *mp, struct vnode **vpp)
1771 if (pmp->iroot == NULL) {
1777 hammer2_inode_lock(pmp->iroot, HAMMER2_RESOLVE_SHARED);
1779 while (pmp->inode_tid == 0) {
1780 hammer2_xop_ipcluster_t *xop;
1781 hammer2_inode_meta_t *meta;
1783 xop = hammer2_xop_alloc(pmp->iroot, HAMMER2_XOP_MODIFYING);
1784 hammer2_xop_start(&xop->head, hammer2_xop_ipcluster);
1785 error = hammer2_xop_collect(&xop->head, 0);
1788 meta = &xop->head.cluster.focus->data->ipdata.meta;
1789 pmp->iroot->meta = *meta;
1790 pmp->inode_tid = meta->pfs_inum + 1;
1791 if (pmp->inode_tid < HAMMER2_INODE_START)
1792 pmp->inode_tid = HAMMER2_INODE_START;
1794 xop->head.cluster.focus->bref.modify_tid + 1;
1795 kprintf("PFS: Starting inode %jd\n",
1796 (intmax_t)pmp->inode_tid);
1797 kprintf("PMP focus good set nextino=%ld mod=%016jx\n",
1798 pmp->inode_tid, pmp->modify_tid);
1799 wakeup(&pmp->iroot);
1801 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1804 * Prime the mount info.
1806 hammer2_vfs_statfs(mp, &mp->mnt_stat, NULL);
1813 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1814 hammer2_inode_unlock(pmp->iroot);
1815 error = tsleep(&pmp->iroot, PCATCH, "h2root", hz);
1816 hammer2_inode_lock(pmp->iroot, HAMMER2_RESOLVE_SHARED);
1822 hammer2_inode_unlock(pmp->iroot);
1825 vp = hammer2_igetv(pmp->iroot, &error);
1826 hammer2_inode_unlock(pmp->iroot);
1836 * XXX incorporate ipdata->meta.inode_quota and data_quota
1840 hammer2_vfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
1844 hammer2_blockref_t bref;
1848 * NOTE: iroot might not have validated the cluster yet.
1852 mp->mnt_stat.f_files = 0;
1853 mp->mnt_stat.f_ffree = 0;
1854 mp->mnt_stat.f_blocks = 0;
1855 mp->mnt_stat.f_bfree = 0;
1856 mp->mnt_stat.f_bavail = 0;
1858 for (i = 0; i < pmp->iroot->cluster.nchains; ++i) {
1859 hmp = pmp->pfs_hmps[i];
1862 if (pmp->iroot->cluster.array[i].chain)
1863 bref = pmp->iroot->cluster.array[i].chain->bref;
1865 bzero(&bref, sizeof(bref));
1867 mp->mnt_stat.f_files = bref.embed.stats.inode_count;
1868 mp->mnt_stat.f_ffree = 0;
1869 mp->mnt_stat.f_blocks = hmp->voldata.allocator_size /
1870 mp->mnt_vstat.f_bsize;
1871 mp->mnt_stat.f_bfree = hmp->voldata.allocator_free /
1872 mp->mnt_vstat.f_bsize;
1873 mp->mnt_stat.f_bavail = mp->mnt_stat.f_bfree;
1875 *sbp = mp->mnt_stat;
1882 hammer2_vfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred)
1886 hammer2_blockref_t bref;
1890 * NOTE: iroot might not have validated the cluster yet.
1894 mp->mnt_vstat.f_bsize = 0;
1895 mp->mnt_vstat.f_files = 0;
1896 mp->mnt_vstat.f_ffree = 0;
1897 mp->mnt_vstat.f_blocks = 0;
1898 mp->mnt_vstat.f_bfree = 0;
1899 mp->mnt_vstat.f_bavail = 0;
1901 for (i = 0; i < pmp->iroot->cluster.nchains; ++i) {
1902 hmp = pmp->pfs_hmps[i];
1905 if (pmp->iroot->cluster.array[i].chain)
1906 bref = pmp->iroot->cluster.array[i].chain->bref;
1908 bzero(&bref, sizeof(bref));
1910 mp->mnt_vstat.f_bsize = HAMMER2_PBUFSIZE;
1911 mp->mnt_vstat.f_files = bref.embed.stats.inode_count;
1912 mp->mnt_vstat.f_ffree = 0;
1913 mp->mnt_vstat.f_blocks = hmp->voldata.allocator_size /
1914 mp->mnt_vstat.f_bsize;
1915 mp->mnt_vstat.f_bfree = hmp->voldata.allocator_free /
1916 mp->mnt_vstat.f_bsize;
1917 mp->mnt_vstat.f_bavail = mp->mnt_vstat.f_bfree;
1919 *sbp = mp->mnt_vstat;
1925 * Mount-time recovery (RW mounts)
1927 * Updates to the free block table are allowed to lag flushes by one
1928 * transaction. In case of a crash, then on a fresh mount we must do an
1929 * incremental scan of the last committed transaction id and make sure that
1930 * all related blocks have been marked allocated.
1932 * The super-root topology and each PFS has its own transaction id domain,
1933 * so we must track PFS boundary transitions.
1935 struct hammer2_recovery_elm {
1936 TAILQ_ENTRY(hammer2_recovery_elm) entry;
1937 hammer2_chain_t *chain;
1938 hammer2_tid_t sync_tid;
1941 TAILQ_HEAD(hammer2_recovery_list, hammer2_recovery_elm);
1943 struct hammer2_recovery_info {
1944 struct hammer2_recovery_list list;
1949 static int hammer2_recovery_scan(hammer2_dev_t *hmp,
1950 hammer2_chain_t *parent,
1951 struct hammer2_recovery_info *info,
1952 hammer2_tid_t sync_tid);
1954 #define HAMMER2_RECOVERY_MAXDEPTH 10
1958 hammer2_recovery(hammer2_dev_t *hmp)
1960 struct hammer2_recovery_info info;
1961 struct hammer2_recovery_elm *elm;
1962 hammer2_chain_t *parent;
1963 hammer2_tid_t sync_tid;
1964 hammer2_tid_t mirror_tid;
1966 int cumulative_error = 0;
1968 hammer2_trans_init(hmp->spmp, 0);
1970 sync_tid = hmp->voldata.freemap_tid;
1971 mirror_tid = hmp->voldata.mirror_tid;
1973 kprintf("hammer2 mount \"%s\": ", hmp->devrepname);
1974 if (sync_tid >= mirror_tid) {
1975 kprintf(" no recovery needed\n");
1977 kprintf(" freemap recovery %016jx-%016jx\n",
1978 sync_tid + 1, mirror_tid);
1981 TAILQ_INIT(&info.list);
1983 parent = hammer2_chain_lookup_init(&hmp->vchain, 0);
1984 cumulative_error = hammer2_recovery_scan(hmp, parent, &info, sync_tid);
1985 hammer2_chain_lookup_done(parent);
1987 while ((elm = TAILQ_FIRST(&info.list)) != NULL) {
1988 TAILQ_REMOVE(&info.list, elm, entry);
1989 parent = elm->chain;
1990 sync_tid = elm->sync_tid;
1991 kfree(elm, M_HAMMER2);
1993 hammer2_chain_lock(parent, HAMMER2_RESOLVE_ALWAYS);
1994 error = hammer2_recovery_scan(hmp, parent, &info,
1995 hmp->voldata.freemap_tid);
1996 hammer2_chain_unlock(parent);
1997 hammer2_chain_drop(parent); /* drop elm->chain ref */
1999 cumulative_error = error;
2001 hammer2_trans_done(hmp->spmp);
2003 return cumulative_error;
2008 hammer2_recovery_scan(hammer2_dev_t *hmp, hammer2_chain_t *parent,
2009 struct hammer2_recovery_info *info,
2010 hammer2_tid_t sync_tid)
2012 const hammer2_inode_data_t *ripdata;
2013 hammer2_chain_t *chain;
2014 hammer2_blockref_t bref;
2016 int cumulative_error = 0;
2021 * Adjust freemap to ensure that the block(s) are marked allocated.
2023 if (parent->bref.type != HAMMER2_BREF_TYPE_VOLUME) {
2024 hammer2_freemap_adjust(hmp, &parent->bref,
2025 HAMMER2_FREEMAP_DORECOVER);
2029 * Check type for recursive scan
2031 switch(parent->bref.type) {
2032 case HAMMER2_BREF_TYPE_VOLUME:
2033 /* data already instantiated */
2035 case HAMMER2_BREF_TYPE_INODE:
2037 * Must instantiate data for DIRECTDATA test and also
2040 hammer2_chain_lock(parent, HAMMER2_RESOLVE_ALWAYS);
2041 ripdata = &hammer2_chain_rdata(parent)->ipdata;
2042 if (ripdata->meta.op_flags & HAMMER2_OPFLAG_DIRECTDATA) {
2043 /* not applicable to recovery scan */
2044 hammer2_chain_unlock(parent);
2047 hammer2_chain_unlock(parent);
2049 case HAMMER2_BREF_TYPE_INDIRECT:
2051 * Must instantiate data for recursion
2053 hammer2_chain_lock(parent, HAMMER2_RESOLVE_ALWAYS);
2054 hammer2_chain_unlock(parent);
2056 case HAMMER2_BREF_TYPE_DIRENT:
2057 case HAMMER2_BREF_TYPE_DATA:
2058 case HAMMER2_BREF_TYPE_FREEMAP:
2059 case HAMMER2_BREF_TYPE_FREEMAP_NODE:
2060 case HAMMER2_BREF_TYPE_FREEMAP_LEAF:
2061 /* not applicable to recovery scan */
2069 * Defer operation if depth limit reached or if we are crossing a
2072 if (info->depth >= HAMMER2_RECOVERY_MAXDEPTH) {
2073 struct hammer2_recovery_elm *elm;
2075 elm = kmalloc(sizeof(*elm), M_HAMMER2, M_ZERO | M_WAITOK);
2076 elm->chain = parent;
2077 elm->sync_tid = sync_tid;
2078 hammer2_chain_ref(parent);
2079 TAILQ_INSERT_TAIL(&info->list, elm, entry);
2080 /* unlocked by caller */
2087 * Recursive scan of the last flushed transaction only. We are
2088 * doing this without pmp assignments so don't leave the chains
2089 * hanging around after we are done with them.
2095 while (hammer2_chain_scan(parent, &chain, &bref,
2096 &first, &cache_index,
2097 HAMMER2_LOOKUP_NODATA) != NULL) {
2101 if (chain == NULL) {
2102 if (bref.mirror_tid > sync_tid) {
2103 hammer2_freemap_adjust(hmp, &bref,
2104 HAMMER2_FREEMAP_DORECOVER);
2110 * This may or may not be a recursive node.
2112 atomic_set_int(&chain->flags, HAMMER2_CHAIN_RELEASE);
2113 if (bref.mirror_tid > sync_tid) {
2115 error = hammer2_recovery_scan(hmp, chain,
2119 cumulative_error = error;
2123 * Flush the recovery at the PFS boundary to stage it for
2124 * the final flush of the super-root topology.
2126 if ((bref.flags & HAMMER2_BREF_FLAG_PFSROOT) &&
2127 (chain->flags & HAMMER2_CHAIN_ONFLUSH)) {
2128 hammer2_flush(chain, HAMMER2_FLUSH_TOP);
2132 return cumulative_error;
2136 * Sync a mount point; this is called on a per-mount basis from the
2137 * filesystem syncer process periodically and whenever a user issues
2141 hammer2_vfs_sync(struct mount *mp, int waitfor)
2143 hammer2_xop_flush_t *xop;
2144 struct hammer2_sync_info info;
2145 hammer2_inode_t *iroot;
2153 KKASSERT(iroot->pmp == pmp);
2156 * We can't acquire locks on existing vnodes while in a transaction
2157 * without risking a deadlock. This assumes that vfsync() can be
2158 * called without the vnode locked (which it can in DragonFly).
2159 * Otherwise we'd have to implement a multi-pass or flag the lock
2160 * failures and retry.
2162 * The reclamation code interlocks with the sync list's token
2163 * (by removing the vnode from the scan list) before unlocking
2164 * the inode, giving us time to ref the inode.
2166 /*flags = VMSC_GETVP;*/
2168 if (waitfor & MNT_LAZY)
2169 flags |= VMSC_ONEPASS;
2172 * Preflush the vnodes using a normal transaction before interlocking
2173 * with a flush transaction. We do this to try to run as much of
2174 * the compression as possible outside the flush transaction.
2176 * For efficiency do an async pass before making sure with a
2177 * synchronous pass on all related buffer cache buffers.
2179 hammer2_trans_init(pmp, 0);
2181 info.waitfor = MNT_NOWAIT;
2182 vsyncscan(mp, flags | VMSC_NOWAIT, hammer2_sync_scan2, &info);
2183 info.waitfor = MNT_WAIT;
2184 vsyncscan(mp, flags, hammer2_sync_scan2, &info);
2185 hammer2_trans_done(pmp);
2188 * Start our flush transaction. This does not return until all
2189 * concurrent transactions have completed and will prevent any
2190 * new transactions from running concurrently, except for the
2191 * buffer cache transactions.
2193 * NOTE! It is still possible for the paging code to push pages
2194 * out via a UIO_NOCOPY hammer2_vop_write() during the main
2197 hammer2_trans_init(pmp, HAMMER2_TRANS_ISFLUSH);
2198 hammer2_inode_run_sideq(pmp);
2201 info.waitfor = MNT_NOWAIT;
2202 vsyncscan(mp, flags | VMSC_NOWAIT, hammer2_sync_scan2, &info);
2203 info.waitfor = MNT_WAIT;
2204 vsyncscan(mp, flags, hammer2_sync_scan2, &info);
2205 hammer2_bioq_sync(pmp);
2208 * Use the XOP interface to concurrently flush all nodes to
2209 * synchronize the PFSROOT subtopology to the media. A standard
2210 * end-of-scan ENOENT error indicates cluster sufficiency.
2212 * Note that this flush will not be visible on crash recovery until
2213 * we flush the super-root topology in the next loop.
2215 * XXX For now wait for all flushes to complete.
2218 xop = hammer2_xop_alloc(iroot, HAMMER2_XOP_MODIFYING);
2219 hammer2_xop_start(&xop->head, hammer2_inode_xop_flush);
2220 error = hammer2_xop_collect(&xop->head,
2221 HAMMER2_XOP_COLLECT_WAITALL);
2222 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
2223 if (error == ENOENT)
2228 hammer2_trans_done(pmp);
2236 * Note that we ignore the tranasction mtid we got above. Instead,
2237 * each vfsync below will ultimately get its own via TRANS_BUFCACHE
2241 hammer2_sync_scan2(struct mount *mp, struct vnode *vp, void *data)
2243 struct hammer2_sync_info *info = data;
2244 hammer2_inode_t *ip;
2248 * Degenerate cases. Note that ip == NULL typically means the
2249 * syncer vnode itself and we don't want to vclrisdirty() in that
2256 if (vp->v_type == VNON || vp->v_type == VBAD) {
2262 * VOP_FSYNC will start a new transaction so replicate some code
2263 * here to do it inline (see hammer2_vop_fsync()).
2265 * WARNING: The vfsync interacts with the buffer cache and might
2266 * block, we can't hold the inode lock at that time.
2267 * However, we MUST ref ip before blocking to ensure that
2268 * it isn't ripped out from under us (since we do not
2269 * hold a lock on the vnode).
2271 hammer2_inode_ref(ip);
2272 if ((ip->flags & HAMMER2_INODE_MODIFIED) ||
2273 !RB_EMPTY(&vp->v_rbdirty_tree)) {
2274 vfsync(vp, info->waitfor, 1, NULL, NULL);
2275 if (ip->flags & (HAMMER2_INODE_RESIZED |
2276 HAMMER2_INODE_MODIFIED)) {
2277 hammer2_inode_lock(ip, 0);
2278 hammer2_inode_chain_sync(ip);
2279 hammer2_inode_unlock(ip);
2282 if ((ip->flags & HAMMER2_INODE_MODIFIED) == 0 &&
2283 RB_EMPTY(&vp->v_rbdirty_tree)) {
2287 hammer2_inode_drop(ip);
2291 info->error = error;
2298 hammer2_vfs_vptofh(struct vnode *vp, struct fid *fhp)
2300 hammer2_inode_t *ip;
2302 KKASSERT(MAXFIDSZ >= 16);
2304 fhp->fid_len = offsetof(struct fid, fid_data[16]);
2306 ((hammer2_tid_t *)fhp->fid_data)[0] = ip->meta.inum;
2307 ((hammer2_tid_t *)fhp->fid_data)[1] = 0;
2314 hammer2_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
2315 struct fid *fhp, struct vnode **vpp)
2322 inum = ((hammer2_tid_t *)fhp->fid_data)[0] & HAMMER2_DIRHASH_USERMSK;
2325 error = hammer2_vfs_root(mp, vpp);
2327 error = hammer2_vfs_vget(mp, NULL, inum, vpp);
2332 kprintf("fhtovp: %016jx -> %p, %d\n", inum, *vpp, error);
2338 hammer2_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
2339 int *exflagsp, struct ucred **credanonp)
2346 np = vfs_export_lookup(mp, &pmp->export, nam);
2348 *exflagsp = np->netc_exflags;
2349 *credanonp = &np->netc_anon;
2358 * Support code for hammer2_vfs_mount(). Read, verify, and install the volume
2359 * header into the HMP
2361 * XXX read four volhdrs and use the one with the highest TID whos CRC
2366 * XXX For filesystems w/ less than 4 volhdrs, make sure to not write to
2367 * nonexistant locations.
2369 * XXX Record selected volhdr and ring updates to each of 4 volhdrs
2373 hammer2_install_volume_header(hammer2_dev_t *hmp)
2375 hammer2_volume_data_t *vd;
2377 hammer2_crc32_t crc0, crc, bcrc0, bcrc;
2389 * There are up to 4 copies of the volume header (syncs iterate
2390 * between them so there is no single master). We don't trust the
2391 * volu_size field so we don't know precisely how large the filesystem
2392 * is, so depend on the OS to return an error if we go beyond the
2393 * block device's EOF.
2395 for (i = 0; i < HAMMER2_NUM_VOLHDRS; i++) {
2396 error = bread(hmp->devvp, i * HAMMER2_ZONE_BYTES64,
2397 HAMMER2_VOLUME_BYTES, &bp);
2404 vd = (struct hammer2_volume_data *) bp->b_data;
2405 if ((vd->magic != HAMMER2_VOLUME_ID_HBO) &&
2406 (vd->magic != HAMMER2_VOLUME_ID_ABO)) {
2412 if (vd->magic == HAMMER2_VOLUME_ID_ABO) {
2413 /* XXX: Reversed-endianness filesystem */
2414 kprintf("hammer2: reverse-endian filesystem detected");
2420 crc = vd->icrc_sects[HAMMER2_VOL_ICRC_SECT0];
2421 crc0 = hammer2_icrc32(bp->b_data + HAMMER2_VOLUME_ICRC0_OFF,
2422 HAMMER2_VOLUME_ICRC0_SIZE);
2423 bcrc = vd->icrc_sects[HAMMER2_VOL_ICRC_SECT1];
2424 bcrc0 = hammer2_icrc32(bp->b_data + HAMMER2_VOLUME_ICRC1_OFF,
2425 HAMMER2_VOLUME_ICRC1_SIZE);
2426 if ((crc0 != crc) || (bcrc0 != bcrc)) {
2427 kprintf("hammer2 volume header crc "
2428 "mismatch copy #%d %08x/%08x\n",
2435 if (valid == 0 || hmp->voldata.mirror_tid < vd->mirror_tid) {
2444 hmp->volsync = hmp->voldata;
2446 if (error_reported || bootverbose || 1) { /* 1/DEBUG */
2447 kprintf("hammer2: using volume header #%d\n",
2452 kprintf("hammer2: no valid volume headers found!\n");
2458 * This handles hysteresis on regular file flushes. Because the BIOs are
2459 * routed to a thread it is possible for an excessive number to build up
2460 * and cause long front-end stalls long before the runningbuffspace limit
2461 * is hit, so we implement hammer2_flush_pipe to control the
2464 * This is a particular problem when compression is used.
2467 hammer2_lwinprog_ref(hammer2_pfs_t *pmp)
2469 atomic_add_int(&pmp->count_lwinprog, 1);
2473 hammer2_lwinprog_drop(hammer2_pfs_t *pmp)
2477 lwinprog = atomic_fetchadd_int(&pmp->count_lwinprog, -1);
2478 if ((lwinprog & HAMMER2_LWINPROG_WAITING) &&
2479 (lwinprog & HAMMER2_LWINPROG_MASK) <= hammer2_flush_pipe * 2 / 3) {
2480 atomic_clear_int(&pmp->count_lwinprog,
2481 HAMMER2_LWINPROG_WAITING);
2482 wakeup(&pmp->count_lwinprog);
2484 if ((lwinprog & HAMMER2_LWINPROG_WAITING0) &&
2485 (lwinprog & HAMMER2_LWINPROG_MASK) <= 0) {
2486 atomic_clear_int(&pmp->count_lwinprog,
2487 HAMMER2_LWINPROG_WAITING0);
2488 wakeup(&pmp->count_lwinprog);
2493 hammer2_lwinprog_wait(hammer2_pfs_t *pmp, int flush_pipe)
2496 int lwflag = (flush_pipe) ? HAMMER2_LWINPROG_WAITING :
2497 HAMMER2_LWINPROG_WAITING0;
2500 lwinprog = pmp->count_lwinprog;
2502 if ((lwinprog & HAMMER2_LWINPROG_MASK) <= flush_pipe)
2504 tsleep_interlock(&pmp->count_lwinprog, 0);
2505 atomic_set_int(&pmp->count_lwinprog, lwflag);
2506 lwinprog = pmp->count_lwinprog;
2507 if ((lwinprog & HAMMER2_LWINPROG_MASK) <= flush_pipe)
2509 tsleep(&pmp->count_lwinprog, PINTERLOCKED, "h2wpipe", hz);
2514 * Manage excessive memory resource use for chain and related
2518 hammer2_pfs_memory_wait(hammer2_pfs_t *pmp)
2528 * Atomic check condition and wait. Also do an early speedup of
2529 * the syncer to try to avoid hitting the wait.
2532 waiting = pmp->inmem_dirty_chains;
2534 count = waiting & HAMMER2_DIRTYCHAIN_MASK;
2536 limit = pmp->mp->mnt_nvnodelistsize / 10;
2537 if (limit < hammer2_limit_dirty_chains)
2538 limit = hammer2_limit_dirty_chains;
2543 if ((int)(ticks - zzticks) > hz) {
2545 kprintf("count %ld %ld\n", count, limit);
2550 * Block if there are too many dirty chains present, wait
2551 * for the flush to clean some out.
2553 if (count > limit) {
2554 tsleep_interlock(&pmp->inmem_dirty_chains, 0);
2555 if (atomic_cmpset_int(&pmp->inmem_dirty_chains,
2557 waiting | HAMMER2_DIRTYCHAIN_WAITING)) {
2558 speedup_syncer(pmp->mp);
2559 tsleep(&pmp->inmem_dirty_chains, PINTERLOCKED,
2562 continue; /* loop on success or fail */
2566 * Try to start an early flush before we are forced to block.
2568 if (count > limit * 7 / 10)
2569 speedup_syncer(pmp->mp);
2575 hammer2_pfs_memory_inc(hammer2_pfs_t *pmp)
2578 atomic_add_int(&pmp->inmem_dirty_chains, 1);
2583 hammer2_pfs_memory_wakeup(hammer2_pfs_t *pmp)
2591 waiting = pmp->inmem_dirty_chains;
2593 if (atomic_cmpset_int(&pmp->inmem_dirty_chains,
2596 ~HAMMER2_DIRTYCHAIN_WAITING)) {
2601 if (waiting & HAMMER2_DIRTYCHAIN_WAITING)
2602 wakeup(&pmp->inmem_dirty_chains);
2609 hammer2_dump_chain(hammer2_chain_t *chain, int tab, int *countp, char pfx)
2611 hammer2_chain_t *scan;
2612 hammer2_chain_t *parent;
2616 kprintf("%*.*s...\n", tab, tab, "");
2621 kprintf("%*.*s%c-chain %p.%d %016jx/%d mir=%016jx\n",
2623 chain, chain->bref.type,
2624 chain->bref.key, chain->bref.keybits,
2625 chain->bref.mirror_tid);
2627 kprintf("%*.*s [%08x] (%s) refs=%d",
2630 ((chain->bref.type == HAMMER2_BREF_TYPE_INODE &&
2631 chain->data) ? (char *)chain->data->ipdata.filename : "?"),
2634 parent = chain->parent;
2636 kprintf("\n%*.*s p=%p [pflags %08x prefs %d",
2638 parent, parent->flags, parent->refs);
2639 if (RB_EMPTY(&chain->core.rbtree)) {
2643 RB_FOREACH(scan, hammer2_chain_tree, &chain->core.rbtree)
2644 hammer2_dump_chain(scan, tab + 4, countp, 'a');
2645 if (chain->bref.type == HAMMER2_BREF_TYPE_INODE && chain->data)
2646 kprintf("%*.*s}(%s)\n", tab, tab, "",
2647 chain->data->ipdata.filename);
2649 kprintf("%*.*s}\n", tab, tab, "");