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 TAILQ_HEAD(hammer2_pfslist, hammer2_pfs);
76 static struct hammer2_mntlist hammer2_mntlist;
77 static struct hammer2_pfslist hammer2_pfslist;
78 static struct lock hammer2_mntlk;
81 int hammer2_cluster_enable = 4;
82 int hammer2_hardlink_enable = 1;
83 int hammer2_flush_pipe = 100;
84 int hammer2_synchronous_flush = 1;
85 int hammer2_dio_count;
86 long hammer2_chain_allocs;
87 long hammer2_chain_frees;
88 long hammer2_limit_dirty_chains;
89 long hammer2_count_modified_chains;
90 long hammer2_iod_file_read;
91 long hammer2_iod_meta_read;
92 long hammer2_iod_indr_read;
93 long hammer2_iod_fmap_read;
94 long hammer2_iod_volu_read;
95 long hammer2_iod_file_write;
96 long hammer2_iod_file_wembed;
97 long hammer2_iod_file_wzero;
98 long hammer2_iod_file_wdedup;
99 long hammer2_iod_meta_write;
100 long hammer2_iod_indr_write;
101 long hammer2_iod_fmap_write;
102 long hammer2_iod_volu_write;
103 long hammer2_ioa_file_read;
104 long hammer2_ioa_meta_read;
105 long hammer2_ioa_indr_read;
106 long hammer2_ioa_fmap_read;
107 long hammer2_ioa_volu_read;
108 long hammer2_ioa_fmap_write;
109 long hammer2_ioa_file_write;
110 long hammer2_ioa_meta_write;
111 long hammer2_ioa_indr_write;
112 long hammer2_ioa_volu_write;
114 MALLOC_DECLARE(M_HAMMER2_CBUFFER);
115 MALLOC_DEFINE(M_HAMMER2_CBUFFER, "HAMMER2-compbuffer",
116 "Buffer used for compression.");
118 MALLOC_DECLARE(M_HAMMER2_DEBUFFER);
119 MALLOC_DEFINE(M_HAMMER2_DEBUFFER, "HAMMER2-decompbuffer",
120 "Buffer used for decompression.");
122 SYSCTL_NODE(_vfs, OID_AUTO, hammer2, CTLFLAG_RW, 0, "HAMMER2 filesystem");
124 SYSCTL_INT(_vfs_hammer2, OID_AUTO, debug, CTLFLAG_RW,
125 &hammer2_debug, 0, "");
126 SYSCTL_INT(_vfs_hammer2, OID_AUTO, cluster_enable, CTLFLAG_RW,
127 &hammer2_cluster_enable, 0, "");
128 SYSCTL_INT(_vfs_hammer2, OID_AUTO, hardlink_enable, CTLFLAG_RW,
129 &hammer2_hardlink_enable, 0, "");
130 SYSCTL_INT(_vfs_hammer2, OID_AUTO, flush_pipe, CTLFLAG_RW,
131 &hammer2_flush_pipe, 0, "");
132 SYSCTL_INT(_vfs_hammer2, OID_AUTO, synchronous_flush, CTLFLAG_RW,
133 &hammer2_synchronous_flush, 0, "");
134 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, chain_allocs, CTLFLAG_RW,
135 &hammer2_chain_allocs, 0, "");
136 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, chain_frees, CTLFLAG_RW,
137 &hammer2_chain_frees, 0, "");
138 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, limit_dirty_chains, CTLFLAG_RW,
139 &hammer2_limit_dirty_chains, 0, "");
140 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, count_modified_chains, CTLFLAG_RW,
141 &hammer2_count_modified_chains, 0, "");
142 SYSCTL_INT(_vfs_hammer2, OID_AUTO, dio_count, CTLFLAG_RD,
143 &hammer2_dio_count, 0, "");
145 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_file_read, CTLFLAG_RW,
146 &hammer2_iod_file_read, 0, "");
147 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_meta_read, CTLFLAG_RW,
148 &hammer2_iod_meta_read, 0, "");
149 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_indr_read, CTLFLAG_RW,
150 &hammer2_iod_indr_read, 0, "");
151 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_fmap_read, CTLFLAG_RW,
152 &hammer2_iod_fmap_read, 0, "");
153 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_volu_read, CTLFLAG_RW,
154 &hammer2_iod_volu_read, 0, "");
156 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_file_write, CTLFLAG_RW,
157 &hammer2_iod_file_write, 0, "");
158 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_file_wembed, CTLFLAG_RW,
159 &hammer2_iod_file_wembed, 0, "");
160 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_file_wzero, CTLFLAG_RW,
161 &hammer2_iod_file_wzero, 0, "");
162 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_file_wdedup, CTLFLAG_RW,
163 &hammer2_iod_file_wdedup, 0, "");
164 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_meta_write, CTLFLAG_RW,
165 &hammer2_iod_meta_write, 0, "");
166 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_indr_write, CTLFLAG_RW,
167 &hammer2_iod_indr_write, 0, "");
168 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_fmap_write, CTLFLAG_RW,
169 &hammer2_iod_fmap_write, 0, "");
170 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_volu_write, CTLFLAG_RW,
171 &hammer2_iod_volu_write, 0, "");
173 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_file_read, CTLFLAG_RW,
174 &hammer2_ioa_file_read, 0, "");
175 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_meta_read, CTLFLAG_RW,
176 &hammer2_ioa_meta_read, 0, "");
177 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_indr_read, CTLFLAG_RW,
178 &hammer2_ioa_indr_read, 0, "");
179 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_fmap_read, CTLFLAG_RW,
180 &hammer2_ioa_fmap_read, 0, "");
181 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_volu_read, CTLFLAG_RW,
182 &hammer2_ioa_volu_read, 0, "");
184 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_file_write, CTLFLAG_RW,
185 &hammer2_ioa_file_write, 0, "");
186 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_meta_write, CTLFLAG_RW,
187 &hammer2_ioa_meta_write, 0, "");
188 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_indr_write, CTLFLAG_RW,
189 &hammer2_ioa_indr_write, 0, "");
190 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_fmap_write, CTLFLAG_RW,
191 &hammer2_ioa_fmap_write, 0, "");
192 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_volu_write, CTLFLAG_RW,
193 &hammer2_ioa_volu_write, 0, "");
195 long hammer2_check_icrc32;
196 long hammer2_check_xxhash64;
197 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, check_icrc32, CTLFLAG_RW,
198 &hammer2_check_icrc32, 0, "");
199 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, check_xxhash64, CTLFLAG_RW,
200 &hammer2_check_xxhash64, 0, "");
202 static int hammer2_vfs_init(struct vfsconf *conf);
203 static int hammer2_vfs_uninit(struct vfsconf *vfsp);
204 static int hammer2_vfs_mount(struct mount *mp, char *path, caddr_t data,
206 static int hammer2_remount(hammer2_dev_t *, struct mount *, char *,
207 struct vnode *, struct ucred *);
208 static int hammer2_recovery(hammer2_dev_t *hmp);
209 static int hammer2_vfs_unmount(struct mount *mp, int mntflags);
210 static int hammer2_vfs_root(struct mount *mp, struct vnode **vpp);
211 static int hammer2_vfs_statfs(struct mount *mp, struct statfs *sbp,
213 static int hammer2_vfs_statvfs(struct mount *mp, struct statvfs *sbp,
215 static int hammer2_vfs_vget(struct mount *mp, struct vnode *dvp,
216 ino_t ino, struct vnode **vpp);
217 static int hammer2_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
218 struct fid *fhp, struct vnode **vpp);
219 static int hammer2_vfs_vptofh(struct vnode *vp, struct fid *fhp);
220 static int hammer2_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
221 int *exflagsp, struct ucred **credanonp);
223 static int hammer2_install_volume_header(hammer2_dev_t *hmp);
224 static int hammer2_sync_scan2(struct mount *mp, struct vnode *vp, void *data);
226 static void hammer2_update_pmps(hammer2_dev_t *hmp);
228 static void hammer2_mount_helper(struct mount *mp, hammer2_pfs_t *pmp);
229 static void hammer2_unmount_helper(struct mount *mp, hammer2_pfs_t *pmp,
233 * HAMMER2 vfs operations.
235 static struct vfsops hammer2_vfsops = {
236 .vfs_init = hammer2_vfs_init,
237 .vfs_uninit = hammer2_vfs_uninit,
238 .vfs_sync = hammer2_vfs_sync,
239 .vfs_mount = hammer2_vfs_mount,
240 .vfs_unmount = hammer2_vfs_unmount,
241 .vfs_root = hammer2_vfs_root,
242 .vfs_statfs = hammer2_vfs_statfs,
243 .vfs_statvfs = hammer2_vfs_statvfs,
244 .vfs_vget = hammer2_vfs_vget,
245 .vfs_vptofh = hammer2_vfs_vptofh,
246 .vfs_fhtovp = hammer2_vfs_fhtovp,
247 .vfs_checkexp = hammer2_vfs_checkexp
250 MALLOC_DEFINE(M_HAMMER2, "HAMMER2-mount", "");
252 VFS_SET(hammer2_vfsops, hammer2, 0);
253 MODULE_VERSION(hammer2, 1);
257 hammer2_vfs_init(struct vfsconf *conf)
259 static struct objcache_malloc_args margs_read;
260 static struct objcache_malloc_args margs_write;
261 static struct objcache_malloc_args margs_vop;
267 if (HAMMER2_BLOCKREF_BYTES != sizeof(struct hammer2_blockref))
269 if (HAMMER2_INODE_BYTES != sizeof(struct hammer2_inode_data))
271 if (HAMMER2_VOLUME_BYTES != sizeof(struct hammer2_volume_data))
275 kprintf("HAMMER2 structure size mismatch; cannot continue.\n");
277 margs_read.objsize = 65536;
278 margs_read.mtype = M_HAMMER2_DEBUFFER;
280 margs_write.objsize = 32768;
281 margs_write.mtype = M_HAMMER2_CBUFFER;
283 margs_vop.objsize = sizeof(hammer2_xop_t);
284 margs_vop.mtype = M_HAMMER2;
287 * Note thaht for the XOPS cache we want backing store allocations
288 * to use M_ZERO. This is not allowed in objcache_get() (to avoid
289 * confusion), so use the backing store function that does it. This
290 * means that initial XOPS objects are zerod but REUSED objects are
291 * not. So we are responsible for cleaning the object up sufficiently
292 * for our needs before objcache_put()ing it back (typically just the
295 cache_buffer_read = objcache_create(margs_read.mtype->ks_shortdesc,
296 0, 1, NULL, NULL, NULL,
297 objcache_malloc_alloc,
298 objcache_malloc_free,
300 cache_buffer_write = objcache_create(margs_write.mtype->ks_shortdesc,
301 0, 1, NULL, NULL, NULL,
302 objcache_malloc_alloc,
303 objcache_malloc_free,
305 cache_xops = objcache_create(margs_vop.mtype->ks_shortdesc,
306 0, 1, NULL, NULL, NULL,
307 objcache_malloc_alloc_zero,
308 objcache_malloc_free,
312 lockinit(&hammer2_mntlk, "mntlk", 0, 0);
313 TAILQ_INIT(&hammer2_mntlist);
314 TAILQ_INIT(&hammer2_pfslist);
316 hammer2_limit_dirty_chains = desiredvnodes / 10;
323 hammer2_vfs_uninit(struct vfsconf *vfsp __unused)
325 objcache_destroy(cache_buffer_read);
326 objcache_destroy(cache_buffer_write);
327 objcache_destroy(cache_xops);
332 * Core PFS allocator. Used to allocate the pmp structure for PFS cluster
333 * mounts and the spmp structure for media (hmp) structures.
335 * pmp->modify_tid tracks new modify_tid transaction ids for front-end
336 * transactions. Note that synchronization does not use this field.
337 * (typically frontend operations and synchronization cannot run on the
338 * same PFS node at the same time).
343 hammer2_pfsalloc(hammer2_chain_t *chain, const hammer2_inode_data_t *ripdata,
344 hammer2_tid_t modify_tid, hammer2_dev_t *force_local)
346 hammer2_inode_t *iroot;
353 * Locate or create the PFS based on the cluster id. If ripdata
354 * is NULL this is a spmp which is unique and is always allocated.
356 * If the device is mounted in local mode all PFSs are considered
357 * independent and not part of any cluster (for debugging only).
360 TAILQ_FOREACH(pmp, &hammer2_pfslist, mntentry) {
361 if (force_local != pmp->force_local)
363 if (force_local == NULL &&
364 bcmp(&pmp->pfs_clid, &ripdata->meta.pfs_clid,
365 sizeof(pmp->pfs_clid)) == 0) {
367 } else if (force_local && pmp->pfs_names[0] &&
368 strcmp(pmp->pfs_names[0], ripdata->filename) == 0) {
377 pmp = kmalloc(sizeof(*pmp), M_HAMMER2, M_WAITOK | M_ZERO);
378 pmp->force_local = force_local;
379 hammer2_trans_manage_init(pmp);
380 kmalloc_create(&pmp->minode, "HAMMER2-inodes");
381 kmalloc_create(&pmp->mmsg, "HAMMER2-pfsmsg");
382 lockinit(&pmp->lock, "pfslk", 0, 0);
383 lockinit(&pmp->lock_nlink, "h2nlink", 0, 0);
384 spin_init(&pmp->inum_spin, "hm2pfsalloc_inum");
385 spin_init(&pmp->xop_spin, "h2xop");
386 spin_init(&pmp->lru_spin, "h2lru");
387 RB_INIT(&pmp->inum_tree);
388 TAILQ_INIT(&pmp->sideq);
389 TAILQ_INIT(&pmp->lru_list);
390 spin_init(&pmp->list_spin, "hm2pfsalloc_list");
393 * Distribute backend operations to threads
395 for (i = 0; i < HAMMER2_MAXCLUSTER; ++i) {
396 for (j = 0; j < HAMMER2_XOPGROUPS +
397 HAMMER2_SPECTHREADS; ++j) {
398 TAILQ_INIT(&pmp->xopq[i][j]);
401 for (i = 0; i < HAMMER2_XOPGROUPS; ++i)
402 hammer2_xop_group_init(pmp, &pmp->xop_groups[i]);
405 * Save the last media transaction id for the flusher. Set
409 pmp->pfs_clid = ripdata->meta.pfs_clid;
410 TAILQ_INSERT_TAIL(&hammer2_pfslist, pmp, mntentry);
413 * The synchronization thread may start too early, make
414 * sure it stays frozen until we are ready to let it go.
418 pmp->primary_thr.flags = HAMMER2_THREAD_FROZEN |
419 HAMMER2_THREAD_REMASTER;
424 * Create the PFS's root inode.
426 if ((iroot = pmp->iroot) == NULL) {
427 iroot = hammer2_inode_get(pmp, NULL, NULL, -1);
429 hammer2_inode_ref(iroot);
430 hammer2_inode_unlock(iroot);
434 * Stop here if no chain is passed in.
440 * When a chain is passed in we must add it to the PFS's root
441 * inode, update pmp->pfs_types[], and update the syncronization
444 * When forcing local mode, mark the PFS as a MASTER regardless.
446 * At the moment empty spots can develop due to removals or failures.
447 * Ultimately we want to re-fill these spots but doing so might
448 * confused running code. XXX
450 hammer2_inode_ref(iroot);
451 hammer2_mtx_ex(&iroot->lock);
452 j = iroot->cluster.nchains;
454 kprintf("add PFS to pmp %p[%d]\n", pmp, j);
456 if (j == HAMMER2_MAXCLUSTER) {
457 kprintf("hammer2_mount: cluster full!\n");
458 /* XXX fatal error? */
460 KKASSERT(chain->pmp == NULL);
462 hammer2_chain_ref(chain);
463 iroot->cluster.array[j].chain = chain;
465 pmp->pfs_types[j] = HAMMER2_PFSTYPE_MASTER;
467 pmp->pfs_types[j] = ripdata->meta.pfs_type;
468 pmp->pfs_names[j] = kstrdup(ripdata->filename, M_HAMMER2);
469 pmp->pfs_hmps[j] = chain->hmp;
472 * If the PFS is already mounted we must account
473 * for the mount_count here.
476 ++chain->hmp->mount_count;
479 * May have to fixup dirty chain tracking. Previous
480 * pmp was NULL so nothing to undo.
482 if (chain->flags & HAMMER2_CHAIN_MODIFIED)
483 hammer2_pfs_memory_inc(pmp);
486 iroot->cluster.nchains = j;
489 * Update nmasters from any PFS inode which is part of the cluster.
490 * It is possible that this will result in a value which is too
491 * high. MASTER PFSs are authoritative for pfs_nmasters and will
492 * override this value later on.
494 * (This informs us of masters that might not currently be
495 * discoverable by this mount).
497 if (ripdata && pmp->pfs_nmasters < ripdata->meta.pfs_nmasters) {
498 pmp->pfs_nmasters = ripdata->meta.pfs_nmasters;
502 * Count visible masters. Masters are usually added with
503 * ripdata->meta.pfs_nmasters set to 1. This detects when there
504 * are more (XXX and must update the master inodes).
507 for (i = 0; i < iroot->cluster.nchains; ++i) {
508 if (pmp->pfs_types[i] == HAMMER2_PFSTYPE_MASTER)
511 if (pmp->pfs_nmasters < count)
512 pmp->pfs_nmasters = count;
515 * Create missing synchronization and support threads.
517 * Single-node masters (including snapshots) have nothing to
518 * synchronize and do not require this thread.
520 * Multi-node masters or any number of soft masters, slaves, copy,
521 * or other PFS types need the thread.
523 * Each thread is responsible for its particular cluster index.
524 * We use independent threads so stalls or mismatches related to
525 * any given target do not affect other targets.
527 for (i = 0; i < iroot->cluster.nchains; ++i) {
529 * Single-node masters (including snapshots) have nothing
530 * to synchronize and will make direct xops support calls,
531 * thus they do not require this thread.
533 * Note that there can be thousands of snapshots. We do not
534 * want to create thousands of threads.
536 if (pmp->pfs_nmasters <= 1 &&
537 pmp->pfs_types[i] == HAMMER2_PFSTYPE_MASTER) {
542 * Sync support thread
544 if (pmp->sync_thrs[i].td == NULL) {
545 hammer2_thr_create(&pmp->sync_thrs[i], pmp,
547 hammer2_primary_sync_thread);
552 * Create missing Xop threads
555 hammer2_xop_helper_create(pmp);
557 hammer2_mtx_unlock(&iroot->lock);
558 hammer2_inode_drop(iroot);
564 * Destroy a PFS, typically only occurs after the last mount on a device
568 hammer2_pfsfree(hammer2_pfs_t *pmp)
570 hammer2_inode_t *iroot;
571 hammer2_chain_t *chain;
576 * Cleanup our reference on iroot. iroot is (should) not be needed
579 TAILQ_REMOVE(&hammer2_pfslist, pmp, mntentry);
583 for (i = 0; i < iroot->cluster.nchains; ++i) {
584 hammer2_thr_delete(&pmp->sync_thrs[i]);
585 for (j = 0; j < HAMMER2_XOPGROUPS; ++j)
586 hammer2_thr_delete(&pmp->xop_groups[j].thrs[i]);
588 #if REPORT_REFS_ERRORS
589 if (pmp->iroot->refs != 1)
590 kprintf("PMP->IROOT %p REFS WRONG %d\n",
591 pmp->iroot, pmp->iroot->refs);
593 KKASSERT(pmp->iroot->refs == 1);
595 /* ref for pmp->iroot */
596 hammer2_inode_drop(pmp->iroot);
601 * Cleanup chains remaining on LRU list.
603 kprintf("pfsfree: %p lrucount=%d\n", pmp, pmp->lru_count);
604 while ((chain = TAILQ_FIRST(&pmp->lru_list)) != NULL) {
605 hammer2_chain_ref(chain);
606 atomic_set_int(&chain->flags, HAMMER2_CHAIN_RELEASE);
607 hammer2_chain_drop(chain);
611 * Free remaining pmp resources
613 kmalloc_destroy(&pmp->mmsg);
614 kmalloc_destroy(&pmp->minode);
616 kfree(pmp, M_HAMMER2);
620 * Remove all references to hmp from the pfs list. Any PFS which becomes
621 * empty is terminated and freed.
626 hammer2_pfsfree_scan(hammer2_dev_t *hmp)
629 hammer2_inode_t *iroot;
630 hammer2_chain_t *rchain;
636 TAILQ_FOREACH(pmp, &hammer2_pfslist, mntentry) {
637 if ((iroot = pmp->iroot) == NULL)
639 if (hmp->spmp == pmp) {
640 kprintf("unmount hmp %p remove spmp %p\n",
646 * Determine if this PFS is affected. If it is we must
647 * freeze all management threads and lock its iroot.
649 * Freezing a management thread forces it idle, operations
650 * in-progress will be aborted and it will have to start
651 * over again when unfrozen, or exit if told to exit.
653 for (i = 0; i < HAMMER2_MAXCLUSTER; ++i) {
654 if (pmp->pfs_hmps[i] == hmp)
657 if (i != HAMMER2_MAXCLUSTER) {
659 * Make sure all synchronization threads are locked
662 for (i = 0; i < HAMMER2_MAXCLUSTER; ++i) {
663 if (pmp->pfs_hmps[i] == NULL)
665 hammer2_thr_freeze_async(&pmp->sync_thrs[i]);
666 for (j = 0; j < HAMMER2_XOPGROUPS; ++j) {
667 hammer2_thr_freeze_async(
668 &pmp->xop_groups[j].thrs[i]);
671 for (i = 0; i < HAMMER2_MAXCLUSTER; ++i) {
672 if (pmp->pfs_hmps[i] == NULL)
674 hammer2_thr_freeze(&pmp->sync_thrs[i]);
675 for (j = 0; j < HAMMER2_XOPGROUPS; ++j) {
677 &pmp->xop_groups[j].thrs[i]);
682 * Lock the inode and clean out matching chains.
683 * Note that we cannot use hammer2_inode_lock_*()
684 * here because that would attempt to validate the
685 * cluster that we are in the middle of ripping
688 * WARNING! We are working directly on the inodes
691 hammer2_mtx_ex(&iroot->lock);
694 * Remove the chain from matching elements of the PFS.
696 for (i = 0; i < HAMMER2_MAXCLUSTER; ++i) {
697 if (pmp->pfs_hmps[i] != hmp)
699 hammer2_thr_delete(&pmp->sync_thrs[i]);
700 for (j = 0; j < HAMMER2_XOPGROUPS; ++j) {
702 &pmp->xop_groups[j].thrs[i]);
704 rchain = iroot->cluster.array[i].chain;
705 iroot->cluster.array[i].chain = NULL;
706 pmp->pfs_types[i] = 0;
707 if (pmp->pfs_names[i]) {
708 kfree(pmp->pfs_names[i], M_HAMMER2);
709 pmp->pfs_names[i] = NULL;
712 hammer2_chain_drop(rchain);
714 if (iroot->cluster.focus == rchain)
715 iroot->cluster.focus = NULL;
717 pmp->pfs_hmps[i] = NULL;
719 hammer2_mtx_unlock(&iroot->lock);
720 didfreeze = 1; /* remaster, unfreeze down below */
726 * Cleanup trailing chains. Gaps may remain.
728 for (i = HAMMER2_MAXCLUSTER - 1; i >= 0; --i) {
729 if (pmp->pfs_hmps[i])
732 iroot->cluster.nchains = i + 1;
735 * If the PMP has no elements remaining we can destroy it.
736 * (this will transition management threads from frozen->exit).
738 if (iroot->cluster.nchains == 0) {
739 kprintf("unmount hmp %p last ref to PMP=%p\n",
741 hammer2_pfsfree(pmp);
746 * If elements still remain we need to set the REMASTER
747 * flag and unfreeze it.
750 for (i = 0; i < HAMMER2_MAXCLUSTER; ++i) {
751 if (pmp->pfs_hmps[i] == NULL)
753 hammer2_thr_remaster(&pmp->sync_thrs[i]);
754 hammer2_thr_unfreeze(&pmp->sync_thrs[i]);
755 for (j = 0; j < HAMMER2_XOPGROUPS; ++j) {
756 hammer2_thr_remaster(
757 &pmp->xop_groups[j].thrs[i]);
758 hammer2_thr_unfreeze(
759 &pmp->xop_groups[j].thrs[i]);
767 * Mount or remount HAMMER2 fileystem from physical media
770 * mp mount point structure
776 * mp mount point structure
777 * path path to mount point
778 * data pointer to argument structure in user space
779 * volume volume path (device@LABEL form)
780 * hflags user mount flags
781 * cred user credentials
788 hammer2_vfs_mount(struct mount *mp, char *path, caddr_t data,
791 struct hammer2_mount_info info;
795 hammer2_dev_t *force_local;
796 hammer2_key_t key_next;
797 hammer2_key_t key_dummy;
800 struct nlookupdata nd;
801 hammer2_chain_t *parent;
802 hammer2_chain_t *chain;
803 hammer2_cluster_t *cluster;
804 const hammer2_inode_data_t *ripdata;
805 hammer2_blockref_t bref;
807 char devstr[MNAMELEN];
824 kprintf("hammer2_mount\n");
830 bzero(&info, sizeof(info));
831 info.cluster_fd = -1;
832 ksnprintf(devstr, sizeof(devstr), "%s",
833 mp->mnt_stat.f_mntfromname);
834 kprintf("hammer2_mount: root '%s'\n", devstr);
837 * Non-root mount or updating a mount
839 error = copyin(data, &info, sizeof(info));
843 error = copyinstr(info.volume, devstr, MNAMELEN - 1, &done);
848 /* Extract device and label */
850 label = strchr(devstr, '@');
852 ((label + 1) - dev) > done) {
860 if (mp->mnt_flag & MNT_UPDATE) {
862 * Update mount. Note that pmp->iroot->cluster is
863 * an inode-embedded cluster and thus cannot be
866 * XXX HAMMER2 needs to implement NFS export via
870 pmp->hflags = info.hflags;
871 cluster = &pmp->iroot->cluster;
872 for (i = 0; i < cluster->nchains; ++i) {
873 if (cluster->array[i].chain == NULL)
875 hmp = cluster->array[i].chain->hmp;
877 error = hammer2_remount(hmp, mp, path,
889 * Lookup name and verify it refers to a block device.
892 error = nlookup_init(&nd, dev, UIO_SYSSPACE, NLC_FOLLOW);
894 error = nlookup(&nd);
896 error = cache_vref(&nd.nl_nch, nd.nl_cred, &devvp);
900 cdev_t cdev = kgetdiskbyname(dev);
901 error = bdevvp(cdev, &devvp);
903 kprintf("hammer2: cannot find '%s'\n", dev);
907 if (vn_isdisk(devvp, &error))
908 error = vfs_mountedon(devvp);
912 * Determine if the device has already been mounted. After this
913 * check hmp will be non-NULL if we are doing the second or more
914 * hammer2 mounts from the same device.
916 lockmgr(&hammer2_mntlk, LK_EXCLUSIVE);
917 TAILQ_FOREACH(hmp, &hammer2_mntlist, mntentry) {
918 if (hmp->devvp == devvp)
923 * Open the device if this isn't a secondary mount and construct
924 * the H2 device mount (hmp).
927 hammer2_chain_t *schain;
930 if (error == 0 && vcount(devvp) > 0)
934 * Now open the device
937 ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
938 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
939 error = vinvalbuf(devvp, V_SAVE, 0, 0);
941 error = VOP_OPEN(devvp,
942 ronly ? FREAD : FREAD | FWRITE,
947 if (error && devvp) {
952 lockmgr(&hammer2_mntlk, LK_RELEASE);
955 hmp = kmalloc(sizeof(*hmp), M_HAMMER2, M_WAITOK | M_ZERO);
956 ksnprintf(hmp->devrepname, sizeof(hmp->devrepname), "%s", dev);
959 hmp->hflags = info.hflags & HMNT2_DEVFLAGS;
960 kmalloc_create(&hmp->mchain, "HAMMER2-chains");
961 TAILQ_INSERT_TAIL(&hammer2_mntlist, hmp, mntentry);
962 RB_INIT(&hmp->iotree);
963 spin_init(&hmp->io_spin, "hm2mount_io");
964 spin_init(&hmp->list_spin, "hm2mount_list");
965 TAILQ_INIT(&hmp->flushq);
967 lockinit(&hmp->vollk, "h2vol", 0, 0);
968 lockinit(&hmp->bulklk, "h2bulk", 0, 0);
971 * vchain setup. vchain.data is embedded.
972 * vchain.refs is initialized and will never drop to 0.
974 * NOTE! voldata is not yet loaded.
976 hmp->vchain.hmp = hmp;
977 hmp->vchain.refs = 1;
978 hmp->vchain.data = (void *)&hmp->voldata;
979 hmp->vchain.bref.type = HAMMER2_BREF_TYPE_VOLUME;
980 hmp->vchain.bref.data_off = 0 | HAMMER2_PBUFRADIX;
981 hmp->vchain.bref.mirror_tid = hmp->voldata.mirror_tid;
983 hammer2_chain_core_init(&hmp->vchain);
984 /* hmp->vchain.u.xxx is left NULL */
987 * fchain setup. fchain.data is embedded.
988 * fchain.refs is initialized and will never drop to 0.
990 * The data is not used but needs to be initialized to
991 * pass assertion muster. We use this chain primarily
992 * as a placeholder for the freemap's top-level RBTREE
993 * so it does not interfere with the volume's topology
996 hmp->fchain.hmp = hmp;
997 hmp->fchain.refs = 1;
998 hmp->fchain.data = (void *)&hmp->voldata.freemap_blockset;
999 hmp->fchain.bref.type = HAMMER2_BREF_TYPE_FREEMAP;
1000 hmp->fchain.bref.data_off = 0 | HAMMER2_PBUFRADIX;
1001 hmp->fchain.bref.mirror_tid = hmp->voldata.freemap_tid;
1002 hmp->fchain.bref.methods =
1003 HAMMER2_ENC_CHECK(HAMMER2_CHECK_FREEMAP) |
1004 HAMMER2_ENC_COMP(HAMMER2_COMP_NONE);
1006 hammer2_chain_core_init(&hmp->fchain);
1007 /* hmp->fchain.u.xxx is left NULL */
1010 * Install the volume header and initialize fields from
1013 error = hammer2_install_volume_header(hmp);
1015 hammer2_unmount_helper(mp, NULL, hmp);
1016 lockmgr(&hammer2_mntlk, LK_RELEASE);
1017 hammer2_vfs_unmount(mp, MNT_FORCE);
1022 * Really important to get these right or flush will get
1025 hmp->spmp = hammer2_pfsalloc(NULL, NULL, 0, NULL);
1026 kprintf("alloc spmp %p tid %016jx\n",
1027 hmp->spmp, hmp->voldata.mirror_tid);
1031 * Dummy-up vchain and fchain's modify_tid. mirror_tid
1032 * is inherited from the volume header.
1035 hmp->vchain.bref.mirror_tid = hmp->voldata.mirror_tid;
1036 hmp->vchain.bref.modify_tid = hmp->vchain.bref.mirror_tid;
1037 hmp->vchain.pmp = spmp;
1038 hmp->fchain.bref.mirror_tid = hmp->voldata.freemap_tid;
1039 hmp->fchain.bref.modify_tid = hmp->fchain.bref.mirror_tid;
1040 hmp->fchain.pmp = spmp;
1043 * First locate the super-root inode, which is key 0
1044 * relative to the volume header's blockset.
1046 * Then locate the root inode by scanning the directory keyspace
1047 * represented by the label.
1049 parent = hammer2_chain_lookup_init(&hmp->vchain, 0);
1050 schain = hammer2_chain_lookup(&parent, &key_dummy,
1051 HAMMER2_SROOT_KEY, HAMMER2_SROOT_KEY,
1053 hammer2_chain_lookup_done(parent);
1054 if (schain == NULL) {
1055 kprintf("hammer2_mount: invalid super-root\n");
1056 hammer2_unmount_helper(mp, NULL, hmp);
1057 lockmgr(&hammer2_mntlk, LK_RELEASE);
1058 hammer2_vfs_unmount(mp, MNT_FORCE);
1061 if (schain->error) {
1062 kprintf("hammer2_mount: error %s reading super-root\n",
1063 hammer2_error_str(schain->error));
1064 hammer2_chain_unlock(schain);
1065 hammer2_chain_drop(schain);
1067 hammer2_unmount_helper(mp, NULL, hmp);
1068 lockmgr(&hammer2_mntlk, LK_RELEASE);
1069 hammer2_vfs_unmount(mp, MNT_FORCE);
1074 * The super-root always uses an inode_tid of 1 when
1077 spmp->inode_tid = 1;
1078 spmp->modify_tid = schain->bref.modify_tid + 1;
1081 * Sanity-check schain's pmp and finish initialization.
1082 * Any chain belonging to the super-root topology should
1083 * have a NULL pmp (not even set to spmp).
1085 ripdata = &hammer2_chain_rdata(schain)->ipdata;
1086 KKASSERT(schain->pmp == NULL);
1087 spmp->pfs_clid = ripdata->meta.pfs_clid;
1090 * Replace the dummy spmp->iroot with a real one. It's
1091 * easier to just do a wholesale replacement than to try
1092 * to update the chain and fixup the iroot fields.
1094 * The returned inode is locked with the supplied cluster.
1096 cluster = hammer2_cluster_from_chain(schain);
1097 hammer2_inode_drop(spmp->iroot);
1099 spmp->iroot = hammer2_inode_get(spmp, NULL, cluster, -1);
1100 spmp->spmp_hmp = hmp;
1101 spmp->pfs_types[0] = ripdata->meta.pfs_type;
1102 spmp->pfs_hmps[0] = hmp;
1103 hammer2_inode_ref(spmp->iroot);
1104 hammer2_inode_unlock(spmp->iroot);
1105 hammer2_cluster_unlock(cluster);
1106 hammer2_cluster_drop(cluster);
1108 /* leave spmp->iroot with one ref */
1110 if ((mp->mnt_flag & MNT_RDONLY) == 0) {
1111 error = hammer2_recovery(hmp);
1112 /* XXX do something with error */
1114 hammer2_update_pmps(hmp);
1115 hammer2_iocom_init(hmp);
1118 * Ref the cluster management messaging descriptor. The mount
1119 * program deals with the other end of the communications pipe.
1121 * Root mounts typically do not supply one.
1123 if (info.cluster_fd >= 0) {
1124 fp = holdfp(curproc->p_fd, info.cluster_fd, -1);
1126 hammer2_cluster_reconnect(hmp, fp);
1128 kprintf("hammer2_mount: bad cluster_fd!\n");
1133 if (info.hflags & HMNT2_DEVFLAGS) {
1134 kprintf("hammer2: Warning: mount flags pertaining "
1135 "to the whole device may only be specified "
1136 "on the first mount of the device: %08x\n",
1137 info.hflags & HMNT2_DEVFLAGS);
1142 * Force local mount (disassociate all PFSs from their clusters).
1143 * Used primarily for debugging.
1145 force_local = (hmp->hflags & HMNT2_LOCAL) ? hmp : NULL;
1148 * Lookup the mount point under the media-localized super-root.
1149 * Scanning hammer2_pfslist doesn't help us because it represents
1150 * PFS cluster ids which can aggregate several named PFSs together.
1152 * cluster->pmp will incorrectly point to spmp and must be fixed
1155 hammer2_inode_lock(spmp->iroot, 0);
1156 parent = hammer2_inode_chain(spmp->iroot, 0, HAMMER2_RESOLVE_ALWAYS);
1157 lhc = hammer2_dirhash(label, strlen(label));
1158 chain = hammer2_chain_lookup(&parent, &key_next,
1159 lhc, lhc + HAMMER2_DIRHASH_LOMASK,
1162 if (chain->bref.type == HAMMER2_BREF_TYPE_INODE &&
1163 strcmp(label, chain->data->ipdata.filename) == 0) {
1166 chain = hammer2_chain_next(&parent, chain, &key_next,
1168 lhc + HAMMER2_DIRHASH_LOMASK,
1172 hammer2_chain_unlock(parent);
1173 hammer2_chain_drop(parent);
1175 hammer2_inode_unlock(spmp->iroot);
1178 * PFS could not be found?
1180 if (chain == NULL) {
1181 kprintf("hammer2_mount: PFS label not found\n");
1182 hammer2_unmount_helper(mp, NULL, hmp);
1183 lockmgr(&hammer2_mntlk, LK_RELEASE);
1184 hammer2_vfs_unmount(mp, MNT_FORCE);
1190 * Acquire the pmp structure (it should have already been allocated
1191 * via hammer2_update_pmps() so do not pass cluster in to add to
1192 * available chains).
1194 * Check if the cluster has already been mounted. A cluster can
1195 * only be mounted once, use null mounts to mount additional copies.
1197 ripdata = &chain->data->ipdata;
1199 pmp = hammer2_pfsalloc(NULL, ripdata, bref.modify_tid, force_local);
1200 hammer2_chain_unlock(chain);
1201 hammer2_chain_drop(chain);
1204 kprintf("hammer2_mount: PFS already mounted!\n");
1205 hammer2_unmount_helper(mp, NULL, hmp);
1206 lockmgr(&hammer2_mntlk, LK_RELEASE);
1207 hammer2_vfs_unmount(mp, MNT_FORCE);
1215 kprintf("hammer2_mount hmp=%p pmp=%p\n", hmp, pmp);
1217 pmp->hflags = info.hflags;
1218 mp->mnt_flag = MNT_LOCAL;
1219 mp->mnt_kern_flag |= MNTK_ALL_MPSAFE; /* all entry pts are SMP */
1220 mp->mnt_kern_flag |= MNTK_THR_SYNC; /* new vsyncscan semantics */
1223 * required mount structure initializations
1225 mp->mnt_stat.f_iosize = HAMMER2_PBUFSIZE;
1226 mp->mnt_stat.f_bsize = HAMMER2_PBUFSIZE;
1228 mp->mnt_vstat.f_frsize = HAMMER2_PBUFSIZE;
1229 mp->mnt_vstat.f_bsize = HAMMER2_PBUFSIZE;
1234 mp->mnt_iosize_max = MAXPHYS;
1237 * Connect up mount pointers.
1239 hammer2_mount_helper(mp, pmp);
1241 lockmgr(&hammer2_mntlk, LK_RELEASE);
1247 vfs_add_vnodeops(mp, &hammer2_vnode_vops, &mp->mnt_vn_norm_ops);
1248 vfs_add_vnodeops(mp, &hammer2_spec_vops, &mp->mnt_vn_spec_ops);
1249 vfs_add_vnodeops(mp, &hammer2_fifo_vops, &mp->mnt_vn_fifo_ops);
1252 copyinstr(info.volume, mp->mnt_stat.f_mntfromname,
1253 MNAMELEN - 1, &size);
1254 bzero(mp->mnt_stat.f_mntfromname + size, MNAMELEN - size);
1255 } /* else root mount, already in there */
1257 bzero(mp->mnt_stat.f_mntonname, sizeof(mp->mnt_stat.f_mntonname));
1259 copyinstr(path, mp->mnt_stat.f_mntonname,
1260 sizeof(mp->mnt_stat.f_mntonname) - 1,
1264 mp->mnt_stat.f_mntonname[0] = '/';
1268 * Initial statfs to prime mnt_stat.
1270 hammer2_vfs_statfs(mp, &mp->mnt_stat, cred);
1276 * Scan PFSs under the super-root and create hammer2_pfs structures.
1280 hammer2_update_pmps(hammer2_dev_t *hmp)
1282 const hammer2_inode_data_t *ripdata;
1283 hammer2_chain_t *parent;
1284 hammer2_chain_t *chain;
1285 hammer2_blockref_t bref;
1286 hammer2_dev_t *force_local;
1287 hammer2_pfs_t *spmp;
1289 hammer2_key_t key_next;
1290 int cache_index = -1;
1293 * Force local mount (disassociate all PFSs from their clusters).
1294 * Used primarily for debugging.
1296 force_local = (hmp->hflags & HMNT2_LOCAL) ? hmp : NULL;
1299 * Lookup mount point under the media-localized super-root.
1301 * cluster->pmp will incorrectly point to spmp and must be fixed
1305 hammer2_inode_lock(spmp->iroot, 0);
1306 parent = hammer2_inode_chain(spmp->iroot, 0, HAMMER2_RESOLVE_ALWAYS);
1307 chain = hammer2_chain_lookup(&parent, &key_next,
1308 HAMMER2_KEY_MIN, HAMMER2_KEY_MAX,
1311 if (chain->bref.type != HAMMER2_BREF_TYPE_INODE)
1313 ripdata = &chain->data->ipdata;
1315 kprintf("ADD LOCAL PFS: %s\n", ripdata->filename);
1317 pmp = hammer2_pfsalloc(chain, ripdata,
1318 bref.modify_tid, force_local);
1319 chain = hammer2_chain_next(&parent, chain, &key_next,
1320 key_next, HAMMER2_KEY_MAX,
1324 hammer2_chain_unlock(parent);
1325 hammer2_chain_drop(parent);
1327 hammer2_inode_unlock(spmp->iroot);
1332 hammer2_remount(hammer2_dev_t *hmp, struct mount *mp, char *path __unused,
1333 struct vnode *devvp, struct ucred *cred)
1337 if (hmp->ronly && (mp->mnt_kern_flag & MNTK_WANTRDWR)) {
1338 error = hammer2_recovery(hmp);
1347 hammer2_vfs_unmount(struct mount *mp, int mntflags)
1358 lockmgr(&hammer2_mntlk, LK_EXCLUSIVE);
1361 * If mount initialization proceeded far enough we must flush
1362 * its vnodes and sync the underlying mount points. Three syncs
1363 * are required to fully flush the filesystem (freemap updates lag
1364 * by one flush, and one extra for safety).
1366 if (mntflags & MNT_FORCE)
1371 error = vflush(mp, 0, flags);
1374 hammer2_vfs_sync(mp, MNT_WAIT);
1375 hammer2_vfs_sync(mp, MNT_WAIT);
1376 hammer2_vfs_sync(mp, MNT_WAIT);
1380 * Cleanup the frontend support XOPS threads
1382 hammer2_xop_helper_cleanup(pmp);
1385 * Cleanup our reference on ihidden.
1388 hammer2_inode_drop(pmp->ihidden);
1389 pmp->ihidden = NULL;
1392 hammer2_unmount_helper(mp, pmp, NULL);
1396 lockmgr(&hammer2_mntlk, LK_RELEASE);
1402 * Mount helper, hook the system mount into our PFS.
1403 * The mount lock is held.
1405 * We must bump the mount_count on related devices for any
1410 hammer2_mount_helper(struct mount *mp, hammer2_pfs_t *pmp)
1412 hammer2_cluster_t *cluster;
1413 hammer2_chain_t *rchain;
1416 mp->mnt_data = (qaddr_t)pmp;
1420 * After pmp->mp is set we have to adjust hmp->mount_count.
1422 cluster = &pmp->iroot->cluster;
1423 for (i = 0; i < cluster->nchains; ++i) {
1424 rchain = cluster->array[i].chain;
1427 ++rchain->hmp->mount_count;
1428 kprintf("hammer2_mount hmp=%p ++mount_count=%d\n",
1429 rchain->hmp, rchain->hmp->mount_count);
1433 * Create missing Xop threads
1435 hammer2_xop_helper_create(pmp);
1439 * Mount helper, unhook the system mount from our PFS.
1440 * The mount lock is held.
1442 * If hmp is supplied a mount responsible for being the first to open
1443 * the block device failed and the block device and all PFSs using the
1444 * block device must be cleaned up.
1446 * If pmp is supplied multiple devices might be backing the PFS and each
1447 * must be disconnected. This might not be the last PFS using some of the
1448 * underlying devices. Also, we have to adjust our hmp->mount_count
1449 * accounting for the devices backing the pmp which is now undergoing an
1454 hammer2_unmount_helper(struct mount *mp, hammer2_pfs_t *pmp, hammer2_dev_t *hmp)
1456 hammer2_cluster_t *cluster;
1457 hammer2_chain_t *rchain;
1458 struct vnode *devvp;
1464 * If no device supplied this is a high-level unmount and we have to
1465 * to disconnect the mount, adjust mount_count, and locate devices
1466 * that might now have no mounts.
1469 KKASSERT(hmp == NULL);
1470 KKASSERT((void *)(intptr_t)mp->mnt_data == pmp);
1472 mp->mnt_data = NULL;
1475 * After pmp->mp is cleared we have to account for
1478 cluster = &pmp->iroot->cluster;
1479 for (i = 0; i < cluster->nchains; ++i) {
1480 rchain = cluster->array[i].chain;
1483 --rchain->hmp->mount_count;
1484 kprintf("hammer2_unmount hmp=%p --mount_count=%d\n",
1485 rchain->hmp, rchain->hmp->mount_count);
1486 /* scrapping hmp now may invalidate the pmp */
1489 TAILQ_FOREACH(hmp, &hammer2_mntlist, mntentry) {
1490 if (hmp->mount_count == 0) {
1491 hammer2_unmount_helper(NULL, NULL, hmp);
1499 * Try to terminate the block device. We can't terminate it if
1500 * there are still PFSs referencing it.
1502 kprintf("hammer2_unmount hmp=%p mount_count=%d\n",
1503 hmp, hmp->mount_count);
1504 if (hmp->mount_count)
1507 hammer2_pfsfree_scan(hmp);
1508 hammer2_dev_exlock(hmp); /* XXX order */
1511 * Cycle the volume data lock as a safety (probably not needed any
1512 * more). To ensure everything is out we need to flush at least
1513 * three times. (1) The running of the sideq can dirty the
1514 * filesystem, (2) A normal flush can dirty the freemap, and
1515 * (3) ensure that the freemap is fully synchronized.
1517 * The next mount's recovery scan can clean everything up but we want
1518 * to leave the filesystem in a 100% clean state on a normal unmount.
1521 hammer2_voldata_lock(hmp);
1522 hammer2_voldata_unlock(hmp);
1524 hammer2_iocom_uninit(hmp);
1526 if ((hmp->vchain.flags | hmp->fchain.flags) &
1527 HAMMER2_CHAIN_FLUSH_MASK) {
1528 kprintf("hammer2_unmount: chains left over "
1529 "after final sync\n");
1530 kprintf(" vchain %08x\n", hmp->vchain.flags);
1531 kprintf(" fchain %08x\n", hmp->fchain.flags);
1533 if (hammer2_debug & 0x0010)
1534 Debugger("entered debugger");
1537 KKASSERT(hmp->spmp == NULL);
1540 * Finish up with the device vnode
1542 if ((devvp = hmp->devvp) != NULL) {
1543 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
1544 vinvalbuf(devvp, (ronly ? 0 : V_SAVE), 0, 0);
1546 VOP_CLOSE(devvp, (ronly ? FREAD : FREAD|FWRITE), NULL);
1553 * Clear vchain/fchain flags that might prevent final cleanup
1556 if (hmp->vchain.flags & HAMMER2_CHAIN_MODIFIED) {
1557 atomic_add_long(&hammer2_count_modified_chains, -1);
1558 atomic_clear_int(&hmp->vchain.flags, HAMMER2_CHAIN_MODIFIED);
1559 hammer2_pfs_memory_wakeup(hmp->vchain.pmp);
1561 if (hmp->vchain.flags & HAMMER2_CHAIN_UPDATE) {
1562 atomic_clear_int(&hmp->vchain.flags, HAMMER2_CHAIN_UPDATE);
1565 if (hmp->fchain.flags & HAMMER2_CHAIN_MODIFIED) {
1566 atomic_add_long(&hammer2_count_modified_chains, -1);
1567 atomic_clear_int(&hmp->fchain.flags, HAMMER2_CHAIN_MODIFIED);
1568 hammer2_pfs_memory_wakeup(hmp->fchain.pmp);
1570 if (hmp->fchain.flags & HAMMER2_CHAIN_UPDATE) {
1571 atomic_clear_int(&hmp->fchain.flags, HAMMER2_CHAIN_UPDATE);
1575 * Final drop of embedded freemap root chain to
1576 * clean up fchain.core (fchain structure is not
1577 * flagged ALLOCATED so it is cleaned out and then
1580 hammer2_chain_drop(&hmp->fchain);
1583 * Final drop of embedded volume root chain to clean
1584 * up vchain.core (vchain structure is not flagged
1585 * ALLOCATED so it is cleaned out and then left to
1589 hammer2_dump_chain(&hmp->vchain, 0, &dumpcnt, 'v');
1591 hammer2_dump_chain(&hmp->fchain, 0, &dumpcnt, 'f');
1592 hammer2_dev_unlock(hmp);
1593 hammer2_chain_drop(&hmp->vchain);
1595 hammer2_io_cleanup(hmp, &hmp->iotree);
1596 if (hmp->iofree_count) {
1597 kprintf("io_cleanup: %d I/O's left hanging\n",
1601 TAILQ_REMOVE(&hammer2_mntlist, hmp, mntentry);
1602 kmalloc_destroy(&hmp->mchain);
1603 kfree(hmp, M_HAMMER2);
1608 hammer2_vfs_vget(struct mount *mp, struct vnode *dvp,
1609 ino_t ino, struct vnode **vpp)
1611 kprintf("hammer2_vget\n");
1612 return (EOPNOTSUPP);
1617 hammer2_vfs_root(struct mount *mp, struct vnode **vpp)
1624 if (pmp->iroot == NULL) {
1630 hammer2_inode_lock(pmp->iroot, HAMMER2_RESOLVE_SHARED);
1632 while (pmp->inode_tid == 0) {
1633 hammer2_xop_ipcluster_t *xop;
1634 hammer2_inode_meta_t *meta;
1636 xop = hammer2_xop_alloc(pmp->iroot, HAMMER2_XOP_MODIFYING);
1637 hammer2_xop_start(&xop->head, hammer2_xop_ipcluster);
1638 error = hammer2_xop_collect(&xop->head, 0);
1641 meta = &xop->head.cluster.focus->data->ipdata.meta;
1642 pmp->iroot->meta = *meta;
1643 pmp->inode_tid = meta->pfs_inum + 1;
1644 if (pmp->inode_tid < HAMMER2_INODE_START)
1645 pmp->inode_tid = HAMMER2_INODE_START;
1647 xop->head.cluster.focus->bref.modify_tid + 1;
1648 kprintf("PFS: Starting inode %jd\n",
1649 (intmax_t)pmp->inode_tid);
1650 kprintf("PMP focus good set nextino=%ld mod=%016jx\n",
1651 pmp->inode_tid, pmp->modify_tid);
1652 wakeup(&pmp->iroot);
1654 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1657 * Prime the mount info.
1659 hammer2_vfs_statfs(mp, &mp->mnt_stat, NULL);
1662 * With the cluster operational, check for and
1663 * install ihidden if needed. The install_hidden
1664 * code needs to get a transaction so we must unlock
1667 * This is only applicable PFS mounts, there is no
1668 * hidden directory in the spmp.
1670 hammer2_inode_unlock(pmp->iroot);
1671 hammer2_inode_install_hidden(pmp);
1672 hammer2_inode_lock(pmp->iroot, HAMMER2_RESOLVE_SHARED);
1680 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1681 hammer2_inode_unlock(pmp->iroot);
1682 error = tsleep(&pmp->iroot, PCATCH, "h2root", hz);
1683 hammer2_inode_lock(pmp->iroot, HAMMER2_RESOLVE_SHARED);
1689 hammer2_inode_unlock(pmp->iroot);
1692 vp = hammer2_igetv(pmp->iroot, &error);
1693 hammer2_inode_unlock(pmp->iroot);
1703 * XXX incorporate ipdata->meta.inode_quota and data_quota
1707 hammer2_vfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
1711 hammer2_blockref_t bref;
1715 * NOTE: iroot might not have validated the cluster yet.
1719 mp->mnt_stat.f_files = 0;
1720 mp->mnt_stat.f_ffree = 0;
1721 mp->mnt_stat.f_blocks = 0;
1722 mp->mnt_stat.f_bfree = 0;
1723 mp->mnt_stat.f_bavail = 0;
1725 for (i = 0; i < pmp->iroot->cluster.nchains; ++i) {
1726 hmp = pmp->pfs_hmps[i];
1729 if (pmp->iroot->cluster.array[i].chain)
1730 bref = pmp->iroot->cluster.array[i].chain->bref;
1732 bzero(&bref, sizeof(bref));
1734 mp->mnt_stat.f_files = bref.inode_count;
1735 mp->mnt_stat.f_ffree = 0;
1736 mp->mnt_stat.f_blocks = (bref.data_count +
1737 hmp->voldata.allocator_free) /
1738 mp->mnt_vstat.f_bsize;
1739 mp->mnt_stat.f_bfree = hmp->voldata.allocator_free /
1740 mp->mnt_vstat.f_bsize;
1741 mp->mnt_stat.f_bavail = mp->mnt_stat.f_bfree;
1743 *sbp = mp->mnt_stat;
1750 hammer2_vfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred)
1754 hammer2_blockref_t bref;
1758 * NOTE: iroot might not have validated the cluster yet.
1762 mp->mnt_vstat.f_bsize = 0;
1763 mp->mnt_vstat.f_files = 0;
1764 mp->mnt_vstat.f_ffree = 0;
1765 mp->mnt_vstat.f_blocks = 0;
1766 mp->mnt_vstat.f_bfree = 0;
1767 mp->mnt_vstat.f_bavail = 0;
1769 for (i = 0; i < pmp->iroot->cluster.nchains; ++i) {
1770 hmp = pmp->pfs_hmps[i];
1773 if (pmp->iroot->cluster.array[i].chain)
1774 bref = pmp->iroot->cluster.array[i].chain->bref;
1776 bzero(&bref, sizeof(bref));
1778 mp->mnt_vstat.f_bsize = HAMMER2_PBUFSIZE;
1779 mp->mnt_vstat.f_files = bref.inode_count;
1780 mp->mnt_vstat.f_ffree = 0;
1781 mp->mnt_vstat.f_blocks = (bref.data_count +
1782 hmp->voldata.allocator_free) /
1783 mp->mnt_vstat.f_bsize;
1784 mp->mnt_vstat.f_bfree = hmp->voldata.allocator_free /
1785 mp->mnt_vstat.f_bsize;
1786 mp->mnt_vstat.f_bavail = mp->mnt_vstat.f_bfree;
1788 *sbp = mp->mnt_vstat;
1794 * Mount-time recovery (RW mounts)
1796 * Updates to the free block table are allowed to lag flushes by one
1797 * transaction. In case of a crash, then on a fresh mount we must do an
1798 * incremental scan of the last committed transaction id and make sure that
1799 * all related blocks have been marked allocated.
1801 * The super-root topology and each PFS has its own transaction id domain,
1802 * so we must track PFS boundary transitions.
1804 struct hammer2_recovery_elm {
1805 TAILQ_ENTRY(hammer2_recovery_elm) entry;
1806 hammer2_chain_t *chain;
1807 hammer2_tid_t sync_tid;
1810 TAILQ_HEAD(hammer2_recovery_list, hammer2_recovery_elm);
1812 struct hammer2_recovery_info {
1813 struct hammer2_recovery_list list;
1818 static int hammer2_recovery_scan(hammer2_dev_t *hmp,
1819 hammer2_chain_t *parent,
1820 struct hammer2_recovery_info *info,
1821 hammer2_tid_t sync_tid);
1823 #define HAMMER2_RECOVERY_MAXDEPTH 10
1827 hammer2_recovery(hammer2_dev_t *hmp)
1829 struct hammer2_recovery_info info;
1830 struct hammer2_recovery_elm *elm;
1831 hammer2_chain_t *parent;
1832 hammer2_tid_t sync_tid;
1833 hammer2_tid_t mirror_tid;
1835 int cumulative_error = 0;
1837 hammer2_trans_init(hmp->spmp, 0);
1839 sync_tid = hmp->voldata.freemap_tid;
1840 mirror_tid = hmp->voldata.mirror_tid;
1842 kprintf("hammer2 mount \"%s\": ", hmp->devrepname);
1843 if (sync_tid >= mirror_tid) {
1844 kprintf(" no recovery needed\n");
1846 kprintf(" freemap recovery %016jx-%016jx\n",
1847 sync_tid + 1, mirror_tid);
1850 TAILQ_INIT(&info.list);
1852 parent = hammer2_chain_lookup_init(&hmp->vchain, 0);
1853 cumulative_error = hammer2_recovery_scan(hmp, parent, &info, sync_tid);
1854 hammer2_chain_lookup_done(parent);
1856 while ((elm = TAILQ_FIRST(&info.list)) != NULL) {
1857 TAILQ_REMOVE(&info.list, elm, entry);
1858 parent = elm->chain;
1859 sync_tid = elm->sync_tid;
1860 kfree(elm, M_HAMMER2);
1862 hammer2_chain_lock(parent, HAMMER2_RESOLVE_ALWAYS);
1863 error = hammer2_recovery_scan(hmp, parent, &info,
1864 hmp->voldata.freemap_tid);
1865 hammer2_chain_unlock(parent);
1866 hammer2_chain_drop(parent); /* drop elm->chain ref */
1868 cumulative_error = error;
1870 hammer2_trans_done(hmp->spmp);
1872 return cumulative_error;
1877 hammer2_recovery_scan(hammer2_dev_t *hmp, hammer2_chain_t *parent,
1878 struct hammer2_recovery_info *info,
1879 hammer2_tid_t sync_tid)
1881 const hammer2_inode_data_t *ripdata;
1882 hammer2_chain_t *chain;
1883 hammer2_blockref_t bref;
1885 int cumulative_error = 0;
1890 * Adjust freemap to ensure that the block(s) are marked allocated.
1892 if (parent->bref.type != HAMMER2_BREF_TYPE_VOLUME) {
1893 hammer2_freemap_adjust(hmp, &parent->bref,
1894 HAMMER2_FREEMAP_DORECOVER);
1898 * Check type for recursive scan
1900 switch(parent->bref.type) {
1901 case HAMMER2_BREF_TYPE_VOLUME:
1902 /* data already instantiated */
1904 case HAMMER2_BREF_TYPE_INODE:
1906 * Must instantiate data for DIRECTDATA test and also
1909 hammer2_chain_lock(parent, HAMMER2_RESOLVE_ALWAYS);
1910 ripdata = &hammer2_chain_rdata(parent)->ipdata;
1911 if (ripdata->meta.op_flags & HAMMER2_OPFLAG_DIRECTDATA) {
1912 /* not applicable to recovery scan */
1913 hammer2_chain_unlock(parent);
1916 hammer2_chain_unlock(parent);
1918 case HAMMER2_BREF_TYPE_INDIRECT:
1920 * Must instantiate data for recursion
1922 hammer2_chain_lock(parent, HAMMER2_RESOLVE_ALWAYS);
1923 hammer2_chain_unlock(parent);
1925 case HAMMER2_BREF_TYPE_DATA:
1926 case HAMMER2_BREF_TYPE_FREEMAP:
1927 case HAMMER2_BREF_TYPE_FREEMAP_NODE:
1928 case HAMMER2_BREF_TYPE_FREEMAP_LEAF:
1929 /* not applicable to recovery scan */
1937 * Defer operation if depth limit reached or if we are crossing a
1940 if (info->depth >= HAMMER2_RECOVERY_MAXDEPTH) {
1941 struct hammer2_recovery_elm *elm;
1943 elm = kmalloc(sizeof(*elm), M_HAMMER2, M_ZERO | M_WAITOK);
1944 elm->chain = parent;
1945 elm->sync_tid = sync_tid;
1946 hammer2_chain_ref(parent);
1947 TAILQ_INSERT_TAIL(&info->list, elm, entry);
1948 /* unlocked by caller */
1955 * Recursive scan of the last flushed transaction only. We are
1956 * doing this without pmp assignments so don't leave the chains
1957 * hanging around after we are done with them.
1963 while (hammer2_chain_scan(parent, &chain, &bref,
1964 &first, &cache_index,
1965 HAMMER2_LOOKUP_NODATA) != NULL) {
1969 if (chain == NULL) {
1970 if (bref.mirror_tid > sync_tid) {
1971 hammer2_freemap_adjust(hmp, &bref,
1972 HAMMER2_FREEMAP_DORECOVER);
1978 * This may or may not be a recursive node.
1980 atomic_set_int(&chain->flags, HAMMER2_CHAIN_RELEASE);
1981 if (bref.mirror_tid > sync_tid) {
1983 error = hammer2_recovery_scan(hmp, chain,
1987 cumulative_error = error;
1991 * Flush the recovery at the PFS boundary to stage it for
1992 * the final flush of the super-root topology.
1994 if ((bref.flags & HAMMER2_BREF_FLAG_PFSROOT) &&
1995 (chain->flags & HAMMER2_CHAIN_ONFLUSH)) {
1996 hammer2_flush(chain, HAMMER2_FLUSH_TOP);
2000 return cumulative_error;
2004 * Sync a mount point; this is called on a per-mount basis from the
2005 * filesystem syncer process periodically and whenever a user issues
2009 hammer2_vfs_sync(struct mount *mp, int waitfor)
2011 hammer2_xop_flush_t *xop;
2012 struct hammer2_sync_info info;
2013 hammer2_inode_t *iroot;
2021 KKASSERT(iroot->pmp == pmp);
2024 * We can't acquire locks on existing vnodes while in a transaction
2025 * without risking a deadlock. This assumes that vfsync() can be
2026 * called without the vnode locked (which it can in DragonFly).
2027 * Otherwise we'd have to implement a multi-pass or flag the lock
2028 * failures and retry.
2030 * The reclamation code interlocks with the sync list's token
2031 * (by removing the vnode from the scan list) before unlocking
2032 * the inode, giving us time to ref the inode.
2034 /*flags = VMSC_GETVP;*/
2036 if (waitfor & MNT_LAZY)
2037 flags |= VMSC_ONEPASS;
2041 * Preflush the vnodes using a normal transaction before interlocking
2042 * with a flush transaction.
2044 hammer2_trans_init(pmp, 0);
2046 info.waitfor = MNT_NOWAIT;
2047 vsyncscan(mp, flags | VMSC_NOWAIT, hammer2_sync_scan2, &info);
2048 hammer2_trans_done(pmp);
2052 * Start our flush transaction. This does not return until all
2053 * concurrent transactions have completed and will prevent any
2054 * new transactions from running concurrently, except for the
2055 * buffer cache transactions.
2057 * For efficiency do an async pass before making sure with a
2058 * synchronous pass on all related buffer cache buffers. It
2059 * should theoretically not be possible for any new file buffers
2060 * to be instantiated during this sequence.
2062 hammer2_trans_init(pmp, HAMMER2_TRANS_ISFLUSH |
2063 HAMMER2_TRANS_PREFLUSH);
2064 hammer2_inode_run_sideq(pmp);
2067 info.waitfor = MNT_NOWAIT;
2068 vsyncscan(mp, flags | VMSC_NOWAIT, hammer2_sync_scan2, &info);
2069 info.waitfor = MNT_WAIT;
2070 vsyncscan(mp, flags, hammer2_sync_scan2, &info);
2073 * Clear PREFLUSH. This prevents (or asserts on) any new logical
2074 * buffer cache flushes which occur during the flush. Device buffers
2077 hammer2_bioq_sync(pmp);
2078 hammer2_trans_clear_preflush(pmp);
2081 * Use the XOP interface to concurrently flush all nodes to
2082 * synchronize the PFSROOT subtopology to the media. A standard
2083 * end-of-scan ENOENT error indicates cluster sufficiency.
2085 * Note that this flush will not be visible on crash recovery until
2086 * we flush the super-root topology in the next loop.
2088 * XXX For now wait for all flushes to complete.
2091 xop = hammer2_xop_alloc(iroot, HAMMER2_XOP_MODIFYING);
2092 hammer2_xop_start(&xop->head, hammer2_inode_xop_flush);
2093 error = hammer2_xop_collect(&xop->head,
2094 HAMMER2_XOP_COLLECT_WAITALL);
2095 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
2096 if (error == ENOENT)
2101 hammer2_trans_done(pmp);
2109 * Note that we ignore the tranasction mtid we got above. Instead,
2110 * each vfsync below will ultimately get its own via TRANS_BUFCACHE
2114 hammer2_sync_scan2(struct mount *mp, struct vnode *vp, void *data)
2116 struct hammer2_sync_info *info = data;
2117 hammer2_inode_t *ip;
2121 * Degenerate cases. Note that ip == NULL typically means the
2122 * syncer vnode itself and we don't want to vclrisdirty() in that
2129 if (vp->v_type == VNON || vp->v_type == VBAD) {
2135 * VOP_FSYNC will start a new transaction so replicate some code
2136 * here to do it inline (see hammer2_vop_fsync()).
2138 * WARNING: The vfsync interacts with the buffer cache and might
2139 * block, we can't hold the inode lock at that time.
2140 * However, we MUST ref ip before blocking to ensure that
2141 * it isn't ripped out from under us (since we do not
2142 * hold a lock on the vnode).
2144 hammer2_inode_ref(ip);
2145 if ((ip->flags & HAMMER2_INODE_MODIFIED) ||
2146 !RB_EMPTY(&vp->v_rbdirty_tree)) {
2147 vfsync(vp, info->waitfor, 1, NULL, NULL);
2148 if (ip->flags & (HAMMER2_INODE_RESIZED |
2149 HAMMER2_INODE_MODIFIED)) {
2150 hammer2_inode_lock(ip, 0);
2151 hammer2_inode_chain_sync(ip);
2152 hammer2_inode_unlock(ip);
2155 if ((ip->flags & HAMMER2_INODE_MODIFIED) == 0 &&
2156 RB_EMPTY(&vp->v_rbdirty_tree)) {
2160 hammer2_inode_drop(ip);
2164 info->error = error;
2171 hammer2_vfs_vptofh(struct vnode *vp, struct fid *fhp)
2178 hammer2_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
2179 struct fid *fhp, struct vnode **vpp)
2186 hammer2_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
2187 int *exflagsp, struct ucred **credanonp)
2193 * Support code for hammer2_vfs_mount(). Read, verify, and install the volume
2194 * header into the HMP
2196 * XXX read four volhdrs and use the one with the highest TID whos CRC
2201 * XXX For filesystems w/ less than 4 volhdrs, make sure to not write to
2202 * nonexistant locations.
2204 * XXX Record selected volhdr and ring updates to each of 4 volhdrs
2208 hammer2_install_volume_header(hammer2_dev_t *hmp)
2210 hammer2_volume_data_t *vd;
2212 hammer2_crc32_t crc0, crc, bcrc0, bcrc;
2224 * There are up to 4 copies of the volume header (syncs iterate
2225 * between them so there is no single master). We don't trust the
2226 * volu_size field so we don't know precisely how large the filesystem
2227 * is, so depend on the OS to return an error if we go beyond the
2228 * block device's EOF.
2230 for (i = 0; i < HAMMER2_NUM_VOLHDRS; i++) {
2231 error = bread(hmp->devvp, i * HAMMER2_ZONE_BYTES64,
2232 HAMMER2_VOLUME_BYTES, &bp);
2239 vd = (struct hammer2_volume_data *) bp->b_data;
2240 if ((vd->magic != HAMMER2_VOLUME_ID_HBO) &&
2241 (vd->magic != HAMMER2_VOLUME_ID_ABO)) {
2247 if (vd->magic == HAMMER2_VOLUME_ID_ABO) {
2248 /* XXX: Reversed-endianness filesystem */
2249 kprintf("hammer2: reverse-endian filesystem detected");
2255 crc = vd->icrc_sects[HAMMER2_VOL_ICRC_SECT0];
2256 crc0 = hammer2_icrc32(bp->b_data + HAMMER2_VOLUME_ICRC0_OFF,
2257 HAMMER2_VOLUME_ICRC0_SIZE);
2258 bcrc = vd->icrc_sects[HAMMER2_VOL_ICRC_SECT1];
2259 bcrc0 = hammer2_icrc32(bp->b_data + HAMMER2_VOLUME_ICRC1_OFF,
2260 HAMMER2_VOLUME_ICRC1_SIZE);
2261 if ((crc0 != crc) || (bcrc0 != bcrc)) {
2262 kprintf("hammer2 volume header crc "
2263 "mismatch copy #%d %08x/%08x\n",
2270 if (valid == 0 || hmp->voldata.mirror_tid < vd->mirror_tid) {
2279 hmp->volsync = hmp->voldata;
2281 if (error_reported || bootverbose || 1) { /* 1/DEBUG */
2282 kprintf("hammer2: using volume header #%d\n",
2287 kprintf("hammer2: no valid volume headers found!\n");
2293 * This handles hysteresis on regular file flushes. Because the BIOs are
2294 * routed to a thread it is possible for an excessive number to build up
2295 * and cause long front-end stalls long before the runningbuffspace limit
2296 * is hit, so we implement hammer2_flush_pipe to control the
2299 * This is a particular problem when compression is used.
2302 hammer2_lwinprog_ref(hammer2_pfs_t *pmp)
2304 atomic_add_int(&pmp->count_lwinprog, 1);
2308 hammer2_lwinprog_drop(hammer2_pfs_t *pmp)
2312 lwinprog = atomic_fetchadd_int(&pmp->count_lwinprog, -1);
2313 if ((lwinprog & HAMMER2_LWINPROG_WAITING) &&
2314 (lwinprog & HAMMER2_LWINPROG_MASK) <= hammer2_flush_pipe * 2 / 3) {
2315 atomic_clear_int(&pmp->count_lwinprog,
2316 HAMMER2_LWINPROG_WAITING);
2317 wakeup(&pmp->count_lwinprog);
2319 if ((lwinprog & HAMMER2_LWINPROG_WAITING0) &&
2320 (lwinprog & HAMMER2_LWINPROG_MASK) <= 0) {
2321 atomic_clear_int(&pmp->count_lwinprog,
2322 HAMMER2_LWINPROG_WAITING0);
2323 wakeup(&pmp->count_lwinprog);
2328 hammer2_lwinprog_wait(hammer2_pfs_t *pmp, int flush_pipe)
2331 int lwflag = (flush_pipe) ? HAMMER2_LWINPROG_WAITING :
2332 HAMMER2_LWINPROG_WAITING0;
2335 lwinprog = pmp->count_lwinprog;
2337 if ((lwinprog & HAMMER2_LWINPROG_MASK) <= flush_pipe)
2339 tsleep_interlock(&pmp->count_lwinprog, 0);
2340 atomic_set_int(&pmp->count_lwinprog, lwflag);
2341 lwinprog = pmp->count_lwinprog;
2342 if ((lwinprog & HAMMER2_LWINPROG_MASK) <= flush_pipe)
2344 tsleep(&pmp->count_lwinprog, PINTERLOCKED, "h2wpipe", hz);
2349 * Manage excessive memory resource use for chain and related
2353 hammer2_pfs_memory_wait(hammer2_pfs_t *pmp)
2363 * Atomic check condition and wait. Also do an early speedup of
2364 * the syncer to try to avoid hitting the wait.
2367 waiting = pmp->inmem_dirty_chains;
2369 count = waiting & HAMMER2_DIRTYCHAIN_MASK;
2371 limit = pmp->mp->mnt_nvnodelistsize / 10;
2372 if (limit < hammer2_limit_dirty_chains)
2373 limit = hammer2_limit_dirty_chains;
2378 if ((int)(ticks - zzticks) > hz) {
2380 kprintf("count %ld %ld\n", count, limit);
2385 * Block if there are too many dirty chains present, wait
2386 * for the flush to clean some out.
2388 if (count > limit) {
2389 tsleep_interlock(&pmp->inmem_dirty_chains, 0);
2390 if (atomic_cmpset_int(&pmp->inmem_dirty_chains,
2392 waiting | HAMMER2_DIRTYCHAIN_WAITING)) {
2393 speedup_syncer(pmp->mp);
2394 tsleep(&pmp->inmem_dirty_chains, PINTERLOCKED,
2397 continue; /* loop on success or fail */
2401 * Try to start an early flush before we are forced to block.
2403 if (count > limit * 7 / 10)
2404 speedup_syncer(pmp->mp);
2410 hammer2_pfs_memory_inc(hammer2_pfs_t *pmp)
2413 atomic_add_int(&pmp->inmem_dirty_chains, 1);
2418 hammer2_pfs_memory_wakeup(hammer2_pfs_t *pmp)
2426 waiting = pmp->inmem_dirty_chains;
2428 if (atomic_cmpset_int(&pmp->inmem_dirty_chains,
2431 ~HAMMER2_DIRTYCHAIN_WAITING)) {
2436 if (waiting & HAMMER2_DIRTYCHAIN_WAITING)
2437 wakeup(&pmp->inmem_dirty_chains);
2444 hammer2_dump_chain(hammer2_chain_t *chain, int tab, int *countp, char pfx)
2446 hammer2_chain_t *scan;
2447 hammer2_chain_t *parent;
2451 kprintf("%*.*s...\n", tab, tab, "");
2456 kprintf("%*.*s%c-chain %p.%d %016jx/%d mir=%016jx\n",
2458 chain, chain->bref.type,
2459 chain->bref.key, chain->bref.keybits,
2460 chain->bref.mirror_tid);
2462 kprintf("%*.*s [%08x] (%s) refs=%d",
2465 ((chain->bref.type == HAMMER2_BREF_TYPE_INODE &&
2466 chain->data) ? (char *)chain->data->ipdata.filename : "?"),
2469 parent = chain->parent;
2471 kprintf("\n%*.*s p=%p [pflags %08x prefs %d",
2473 parent, parent->flags, parent->refs);
2474 if (RB_EMPTY(&chain->core.rbtree)) {
2478 RB_FOREACH(scan, hammer2_chain_tree, &chain->core.rbtree)
2479 hammer2_dump_chain(scan, tab + 4, countp, 'a');
2480 if (chain->bref.type == HAMMER2_BREF_TYPE_INODE && chain->data)
2481 kprintf("%*.*s}(%s)\n", tab, tab, "",
2482 chain->data->ipdata.filename);
2484 kprintf("%*.*s}\n", tab, tab, "");