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
, NULL
,
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\" rdonly=%d\n",
929 dev
, label
, (mp
->mnt_flag
& MNT_RDONLY
));
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.
987 * Make sure its a block device. Do not check to see if it is
988 * already mounted until we determine that its a fresh H2 device.
990 if (error
== 0 && devvp
) {
991 vn_isdisk(devvp
, &error
);
995 * Determine if the device has already been mounted. After this
996 * check hmp will be non-NULL if we are doing the second or more
997 * hammer2 mounts from the same device.
999 lockmgr(&hammer2_mntlk
, LK_EXCLUSIVE
);
1002 * Match the device. Due to the way devfs works,
1003 * we may not be able to directly match the vnode pointer,
1004 * so also check to see if the underlying device matches.
1006 TAILQ_FOREACH(hmp
, &hammer2_mntlist
, mntentry
) {
1007 if (hmp
->devvp
== devvp
)
1009 if (devvp
->v_rdev
&&
1010 hmp
->devvp
->v_rdev
== devvp
->v_rdev
) {
1016 * If no match this may be a fresh H2 mount, make sure
1017 * the device is not mounted on anything else.
1020 error
= vfs_mountedon(devvp
);
1021 } else if (error
== 0) {
1023 * Match the label to a pmp already probed.
1025 TAILQ_FOREACH(pmp
, &hammer2_pfslist
, mntentry
) {
1026 for (i
= 0; i
< HAMMER2_MAXCLUSTER
; ++i
) {
1027 if (pmp
->pfs_names
[i
] &&
1028 strcmp(pmp
->pfs_names
[i
], label
) == 0) {
1029 hmp
= pmp
->pfs_hmps
[i
];
1041 * Open the device if this isn't a secondary mount and construct
1042 * the H2 device mount (hmp).
1045 hammer2_chain_t
*schain
;
1048 if (error
== 0 && vcount(devvp
) > 0) {
1049 kprintf("Primary device already has references\n");
1054 * Now open the device
1057 ronly
= ((mp
->mnt_flag
& MNT_RDONLY
) != 0);
1058 vn_lock(devvp
, LK_EXCLUSIVE
| LK_RETRY
);
1059 error
= vinvalbuf(devvp
, V_SAVE
, 0, 0);
1061 error
= VOP_OPEN(devvp
,
1062 (ronly
? FREAD
: FREAD
| FWRITE
),
1067 if (error
&& devvp
) {
1072 lockmgr(&hammer2_mntlk
, LK_RELEASE
);
1075 hmp
= kmalloc(sizeof(*hmp
), M_HAMMER2
, M_WAITOK
| M_ZERO
);
1076 ksnprintf(hmp
->devrepname
, sizeof(hmp
->devrepname
), "%s", dev
);
1079 hmp
->hflags
= info
.hflags
& HMNT2_DEVFLAGS
;
1080 kmalloc_create(&hmp
->mchain
, "HAMMER2-chains");
1081 TAILQ_INSERT_TAIL(&hammer2_mntlist
, hmp
, mntentry
);
1082 RB_INIT(&hmp
->iotree
);
1083 spin_init(&hmp
->io_spin
, "hm2mount_io");
1084 spin_init(&hmp
->list_spin
, "hm2mount_list");
1085 TAILQ_INIT(&hmp
->flushq
);
1087 lockinit(&hmp
->vollk
, "h2vol", 0, 0);
1088 lockinit(&hmp
->bulklk
, "h2bulk", 0, 0);
1089 lockinit(&hmp
->bflock
, "h2bflk", 0, 0);
1092 * vchain setup. vchain.data is embedded.
1093 * vchain.refs is initialized and will never drop to 0.
1095 * NOTE! voldata is not yet loaded.
1097 hmp
->vchain
.hmp
= hmp
;
1098 hmp
->vchain
.refs
= 1;
1099 hmp
->vchain
.data
= (void *)&hmp
->voldata
;
1100 hmp
->vchain
.bref
.type
= HAMMER2_BREF_TYPE_VOLUME
;
1101 hmp
->vchain
.bref
.data_off
= 0 | HAMMER2_PBUFRADIX
;
1102 hmp
->vchain
.bref
.mirror_tid
= hmp
->voldata
.mirror_tid
;
1104 hammer2_chain_core_init(&hmp
->vchain
);
1105 /* hmp->vchain.u.xxx is left NULL */
1108 * fchain setup. fchain.data is embedded.
1109 * fchain.refs is initialized and will never drop to 0.
1111 * The data is not used but needs to be initialized to
1112 * pass assertion muster. We use this chain primarily
1113 * as a placeholder for the freemap's top-level RBTREE
1114 * so it does not interfere with the volume's topology
1117 hmp
->fchain
.hmp
= hmp
;
1118 hmp
->fchain
.refs
= 1;
1119 hmp
->fchain
.data
= (void *)&hmp
->voldata
.freemap_blockset
;
1120 hmp
->fchain
.bref
.type
= HAMMER2_BREF_TYPE_FREEMAP
;
1121 hmp
->fchain
.bref
.data_off
= 0 | HAMMER2_PBUFRADIX
;
1122 hmp
->fchain
.bref
.mirror_tid
= hmp
->voldata
.freemap_tid
;
1123 hmp
->fchain
.bref
.methods
=
1124 HAMMER2_ENC_CHECK(HAMMER2_CHECK_FREEMAP
) |
1125 HAMMER2_ENC_COMP(HAMMER2_COMP_NONE
);
1127 hammer2_chain_core_init(&hmp
->fchain
);
1128 /* hmp->fchain.u.xxx is left NULL */
1131 * Install the volume header and initialize fields from
1134 error
= hammer2_install_volume_header(hmp
);
1136 hammer2_unmount_helper(mp
, NULL
, hmp
);
1137 lockmgr(&hammer2_mntlk
, LK_RELEASE
);
1138 hammer2_vfs_unmount(mp
, MNT_FORCE
);
1143 * Really important to get these right or flush will get
1146 hmp
->spmp
= hammer2_pfsalloc(NULL
, NULL
, 0, NULL
);
1147 kprintf("alloc spmp %p tid %016jx\n",
1148 hmp
->spmp
, hmp
->voldata
.mirror_tid
);
1152 * Dummy-up vchain and fchain's modify_tid. mirror_tid
1153 * is inherited from the volume header.
1156 hmp
->vchain
.bref
.mirror_tid
= hmp
->voldata
.mirror_tid
;
1157 hmp
->vchain
.bref
.modify_tid
= hmp
->vchain
.bref
.mirror_tid
;
1158 hmp
->vchain
.pmp
= spmp
;
1159 hmp
->fchain
.bref
.mirror_tid
= hmp
->voldata
.freemap_tid
;
1160 hmp
->fchain
.bref
.modify_tid
= hmp
->fchain
.bref
.mirror_tid
;
1161 hmp
->fchain
.pmp
= spmp
;
1164 * First locate the super-root inode, which is key 0
1165 * relative to the volume header's blockset.
1167 * Then locate the root inode by scanning the directory keyspace
1168 * represented by the label.
1170 parent
= hammer2_chain_lookup_init(&hmp
->vchain
, 0);
1171 schain
= hammer2_chain_lookup(&parent
, &key_dummy
,
1172 HAMMER2_SROOT_KEY
, HAMMER2_SROOT_KEY
,
1174 hammer2_chain_lookup_done(parent
);
1175 if (schain
== NULL
) {
1176 kprintf("hammer2_mount: invalid super-root\n");
1177 hammer2_unmount_helper(mp
, NULL
, hmp
);
1178 lockmgr(&hammer2_mntlk
, LK_RELEASE
);
1179 hammer2_vfs_unmount(mp
, MNT_FORCE
);
1182 if (schain
->error
) {
1183 kprintf("hammer2_mount: error %s reading super-root\n",
1184 hammer2_error_str(schain
->error
));
1185 hammer2_chain_unlock(schain
);
1186 hammer2_chain_drop(schain
);
1188 hammer2_unmount_helper(mp
, NULL
, hmp
);
1189 lockmgr(&hammer2_mntlk
, LK_RELEASE
);
1190 hammer2_vfs_unmount(mp
, MNT_FORCE
);
1195 * The super-root always uses an inode_tid of 1 when
1198 spmp
->inode_tid
= 1;
1199 spmp
->modify_tid
= schain
->bref
.modify_tid
+ 1;
1202 * Sanity-check schain's pmp and finish initialization.
1203 * Any chain belonging to the super-root topology should
1204 * have a NULL pmp (not even set to spmp).
1206 ripdata
= &hammer2_chain_rdata(schain
)->ipdata
;
1207 KKASSERT(schain
->pmp
== NULL
);
1208 spmp
->pfs_clid
= ripdata
->meta
.pfs_clid
;
1211 * Replace the dummy spmp->iroot with a real one. It's
1212 * easier to just do a wholesale replacement than to try
1213 * to update the chain and fixup the iroot fields.
1215 * The returned inode is locked with the supplied cluster.
1217 cluster
= hammer2_cluster_from_chain(schain
);
1218 hammer2_inode_drop(spmp
->iroot
);
1220 spmp
->iroot
= hammer2_inode_get(spmp
, NULL
, cluster
, -1);
1221 spmp
->spmp_hmp
= hmp
;
1222 spmp
->pfs_types
[0] = ripdata
->meta
.pfs_type
;
1223 spmp
->pfs_hmps
[0] = hmp
;
1224 hammer2_inode_ref(spmp
->iroot
);
1225 hammer2_inode_unlock(spmp
->iroot
);
1226 hammer2_cluster_unlock(cluster
);
1227 hammer2_cluster_drop(cluster
);
1229 /* leave spmp->iroot with one ref */
1231 if ((mp
->mnt_flag
& MNT_RDONLY
) == 0) {
1232 error
= hammer2_recovery(hmp
);
1233 /* XXX do something with error */
1235 hammer2_update_pmps(hmp
);
1236 hammer2_iocom_init(hmp
);
1237 hammer2_bulkfree_init(hmp
);
1240 * Ref the cluster management messaging descriptor. The mount
1241 * program deals with the other end of the communications pipe.
1243 * Root mounts typically do not supply one.
1245 if (info
.cluster_fd
>= 0) {
1246 fp
= holdfp(curproc
->p_fd
, info
.cluster_fd
, -1);
1248 hammer2_cluster_reconnect(hmp
, fp
);
1250 kprintf("hammer2_mount: bad cluster_fd!\n");
1255 if (info
.hflags
& HMNT2_DEVFLAGS
) {
1256 kprintf("hammer2: Warning: mount flags pertaining "
1257 "to the whole device may only be specified "
1258 "on the first mount of the device: %08x\n",
1259 info
.hflags
& HMNT2_DEVFLAGS
);
1264 * Force local mount (disassociate all PFSs from their clusters).
1265 * Used primarily for debugging.
1267 force_local
= (hmp
->hflags
& HMNT2_LOCAL
) ? hmp
: NULL
;
1270 * Lookup the mount point under the media-localized super-root.
1271 * Scanning hammer2_pfslist doesn't help us because it represents
1272 * PFS cluster ids which can aggregate several named PFSs together.
1274 * cluster->pmp will incorrectly point to spmp and must be fixed
1277 hammer2_inode_lock(spmp
->iroot
, 0);
1278 parent
= hammer2_inode_chain(spmp
->iroot
, 0, HAMMER2_RESOLVE_ALWAYS
);
1279 lhc
= hammer2_dirhash(label
, strlen(label
));
1280 chain
= hammer2_chain_lookup(&parent
, &key_next
,
1281 lhc
, lhc
+ HAMMER2_DIRHASH_LOMASK
,
1284 if (chain
->bref
.type
== HAMMER2_BREF_TYPE_INODE
&&
1285 strcmp(label
, chain
->data
->ipdata
.filename
) == 0) {
1288 chain
= hammer2_chain_next(&parent
, chain
, &key_next
,
1290 lhc
+ HAMMER2_DIRHASH_LOMASK
,
1294 hammer2_chain_unlock(parent
);
1295 hammer2_chain_drop(parent
);
1297 hammer2_inode_unlock(spmp
->iroot
);
1300 * PFS could not be found?
1302 if (chain
== NULL
) {
1303 kprintf("hammer2_mount: PFS label not found\n");
1304 hammer2_unmount_helper(mp
, NULL
, hmp
);
1305 lockmgr(&hammer2_mntlk
, LK_RELEASE
);
1306 hammer2_vfs_unmount(mp
, MNT_FORCE
);
1312 * Acquire the pmp structure (it should have already been allocated
1313 * via hammer2_update_pmps() so do not pass cluster in to add to
1314 * available chains).
1316 * Check if the cluster has already been mounted. A cluster can
1317 * only be mounted once, use null mounts to mount additional copies.
1319 ripdata
= &chain
->data
->ipdata
;
1321 pmp
= hammer2_pfsalloc(NULL
, ripdata
,
1322 bref
.modify_tid
, force_local
);
1323 hammer2_chain_unlock(chain
);
1324 hammer2_chain_drop(chain
);
1329 kprintf("hammer2_mount hmp=%p pmp=%p\n", hmp
, pmp
);
1332 kprintf("hammer2_mount: PFS already mounted!\n");
1333 hammer2_unmount_helper(mp
, NULL
, hmp
);
1334 lockmgr(&hammer2_mntlk
, LK_RELEASE
);
1335 hammer2_vfs_unmount(mp
, MNT_FORCE
);
1340 pmp
->hflags
= info
.hflags
;
1341 mp
->mnt_flag
|= MNT_LOCAL
;
1342 mp
->mnt_kern_flag
|= MNTK_ALL_MPSAFE
; /* all entry pts are SMP */
1343 mp
->mnt_kern_flag
|= MNTK_THR_SYNC
; /* new vsyncscan semantics */
1346 * required mount structure initializations
1348 mp
->mnt_stat
.f_iosize
= HAMMER2_PBUFSIZE
;
1349 mp
->mnt_stat
.f_bsize
= HAMMER2_PBUFSIZE
;
1351 mp
->mnt_vstat
.f_frsize
= HAMMER2_PBUFSIZE
;
1352 mp
->mnt_vstat
.f_bsize
= HAMMER2_PBUFSIZE
;
1357 mp
->mnt_iosize_max
= MAXPHYS
;
1360 * Connect up mount pointers.
1362 hammer2_mount_helper(mp
, pmp
);
1364 lockmgr(&hammer2_mntlk
, LK_RELEASE
);
1370 vfs_add_vnodeops(mp
, &hammer2_vnode_vops
, &mp
->mnt_vn_norm_ops
);
1371 vfs_add_vnodeops(mp
, &hammer2_spec_vops
, &mp
->mnt_vn_spec_ops
);
1372 vfs_add_vnodeops(mp
, &hammer2_fifo_vops
, &mp
->mnt_vn_fifo_ops
);
1375 copyinstr(info
.volume
, mp
->mnt_stat
.f_mntfromname
,
1376 MNAMELEN
- 1, &size
);
1377 bzero(mp
->mnt_stat
.f_mntfromname
+ size
, MNAMELEN
- size
);
1378 } /* else root mount, already in there */
1380 bzero(mp
->mnt_stat
.f_mntonname
, sizeof(mp
->mnt_stat
.f_mntonname
));
1382 copyinstr(path
, mp
->mnt_stat
.f_mntonname
,
1383 sizeof(mp
->mnt_stat
.f_mntonname
) - 1,
1387 mp
->mnt_stat
.f_mntonname
[0] = '/';
1391 * Initial statfs to prime mnt_stat.
1393 hammer2_vfs_statfs(mp
, &mp
->mnt_stat
, cred
);
1399 * Scan PFSs under the super-root and create hammer2_pfs structures.
1403 hammer2_update_pmps(hammer2_dev_t
*hmp
)
1405 const hammer2_inode_data_t
*ripdata
;
1406 hammer2_chain_t
*parent
;
1407 hammer2_chain_t
*chain
;
1408 hammer2_blockref_t bref
;
1409 hammer2_dev_t
*force_local
;
1410 hammer2_pfs_t
*spmp
;
1412 hammer2_key_t key_next
;
1413 int cache_index
= -1;
1416 * Force local mount (disassociate all PFSs from their clusters).
1417 * Used primarily for debugging.
1419 force_local
= (hmp
->hflags
& HMNT2_LOCAL
) ? hmp
: NULL
;
1422 * Lookup mount point under the media-localized super-root.
1424 * cluster->pmp will incorrectly point to spmp and must be fixed
1428 hammer2_inode_lock(spmp
->iroot
, 0);
1429 parent
= hammer2_inode_chain(spmp
->iroot
, 0, HAMMER2_RESOLVE_ALWAYS
);
1430 chain
= hammer2_chain_lookup(&parent
, &key_next
,
1431 HAMMER2_KEY_MIN
, HAMMER2_KEY_MAX
,
1434 if (chain
->bref
.type
!= HAMMER2_BREF_TYPE_INODE
)
1436 ripdata
= &chain
->data
->ipdata
;
1438 kprintf("ADD LOCAL PFS: %s\n", ripdata
->filename
);
1440 pmp
= hammer2_pfsalloc(chain
, ripdata
,
1441 bref
.modify_tid
, force_local
);
1442 chain
= hammer2_chain_next(&parent
, chain
, &key_next
,
1443 key_next
, HAMMER2_KEY_MAX
,
1447 hammer2_chain_unlock(parent
);
1448 hammer2_chain_drop(parent
);
1450 hammer2_inode_unlock(spmp
->iroot
);
1455 hammer2_remount(hammer2_dev_t
*hmp
, struct mount
*mp
, char *path __unused
,
1456 struct vnode
*devvp
, struct ucred
*cred
)
1460 if (hmp
->ronly
&& (mp
->mnt_kern_flag
& MNTK_WANTRDWR
)) {
1461 vn_lock(devvp
, LK_EXCLUSIVE
| LK_RETRY
);
1462 VOP_OPEN(devvp
, FREAD
| FWRITE
, FSCRED
, NULL
);
1464 error
= hammer2_recovery(hmp
);
1465 vn_lock(devvp
, LK_EXCLUSIVE
| LK_RETRY
);
1467 VOP_CLOSE(devvp
, FREAD
, NULL
);
1470 VOP_CLOSE(devvp
, FREAD
| FWRITE
, NULL
);
1481 hammer2_vfs_unmount(struct mount
*mp
, int mntflags
)
1492 lockmgr(&hammer2_mntlk
, LK_EXCLUSIVE
);
1495 * If mount initialization proceeded far enough we must flush
1496 * its vnodes and sync the underlying mount points. Three syncs
1497 * are required to fully flush the filesystem (freemap updates lag
1498 * by one flush, and one extra for safety).
1500 if (mntflags
& MNT_FORCE
)
1505 error
= vflush(mp
, 0, flags
);
1508 hammer2_vfs_sync(mp
, MNT_WAIT
);
1509 hammer2_vfs_sync(mp
, MNT_WAIT
);
1510 hammer2_vfs_sync(mp
, MNT_WAIT
);
1514 * Cleanup the frontend support XOPS threads
1516 hammer2_xop_helper_cleanup(pmp
);
1519 hammer2_unmount_helper(mp
, pmp
, NULL
);
1523 lockmgr(&hammer2_mntlk
, LK_RELEASE
);
1529 * Mount helper, hook the system mount into our PFS.
1530 * The mount lock is held.
1532 * We must bump the mount_count on related devices for any
1537 hammer2_mount_helper(struct mount
*mp
, hammer2_pfs_t
*pmp
)
1539 hammer2_cluster_t
*cluster
;
1540 hammer2_chain_t
*rchain
;
1543 mp
->mnt_data
= (qaddr_t
)pmp
;
1547 * After pmp->mp is set we have to adjust hmp->mount_count.
1549 cluster
= &pmp
->iroot
->cluster
;
1550 for (i
= 0; i
< cluster
->nchains
; ++i
) {
1551 rchain
= cluster
->array
[i
].chain
;
1554 ++rchain
->hmp
->mount_count
;
1555 kprintf("hammer2_mount hmp=%p ++mount_count=%d\n",
1556 rchain
->hmp
, rchain
->hmp
->mount_count
);
1560 * Create missing Xop threads
1562 hammer2_xop_helper_create(pmp
);
1566 * Mount helper, unhook the system mount from our PFS.
1567 * The mount lock is held.
1569 * If hmp is supplied a mount responsible for being the first to open
1570 * the block device failed and the block device and all PFSs using the
1571 * block device must be cleaned up.
1573 * If pmp is supplied multiple devices might be backing the PFS and each
1574 * must be disconnected. This might not be the last PFS using some of the
1575 * underlying devices. Also, we have to adjust our hmp->mount_count
1576 * accounting for the devices backing the pmp which is now undergoing an
1581 hammer2_unmount_helper(struct mount
*mp
, hammer2_pfs_t
*pmp
, hammer2_dev_t
*hmp
)
1583 hammer2_cluster_t
*cluster
;
1584 hammer2_chain_t
*rchain
;
1585 struct vnode
*devvp
;
1591 * If no device supplied this is a high-level unmount and we have to
1592 * to disconnect the mount, adjust mount_count, and locate devices
1593 * that might now have no mounts.
1596 KKASSERT(hmp
== NULL
);
1597 KKASSERT((void *)(intptr_t)mp
->mnt_data
== pmp
);
1599 mp
->mnt_data
= NULL
;
1602 * After pmp->mp is cleared we have to account for
1605 cluster
= &pmp
->iroot
->cluster
;
1606 for (i
= 0; i
< cluster
->nchains
; ++i
) {
1607 rchain
= cluster
->array
[i
].chain
;
1610 --rchain
->hmp
->mount_count
;
1611 kprintf("hammer2_unmount hmp=%p --mount_count=%d\n",
1612 rchain
->hmp
, rchain
->hmp
->mount_count
);
1613 /* scrapping hmp now may invalidate the pmp */
1616 TAILQ_FOREACH(hmp
, &hammer2_mntlist
, mntentry
) {
1617 if (hmp
->mount_count
== 0) {
1618 hammer2_unmount_helper(NULL
, NULL
, hmp
);
1626 * Try to terminate the block device. We can't terminate it if
1627 * there are still PFSs referencing it.
1629 kprintf("hammer2_unmount hmp=%p mount_count=%d\n",
1630 hmp
, hmp
->mount_count
);
1631 if (hmp
->mount_count
)
1634 hammer2_bulkfree_uninit(hmp
);
1635 hammer2_pfsfree_scan(hmp
);
1636 hammer2_dev_exlock(hmp
); /* XXX order */
1639 * Cycle the volume data lock as a safety (probably not needed any
1640 * more). To ensure everything is out we need to flush at least
1641 * three times. (1) The running of the sideq can dirty the
1642 * filesystem, (2) A normal flush can dirty the freemap, and
1643 * (3) ensure that the freemap is fully synchronized.
1645 * The next mount's recovery scan can clean everything up but we want
1646 * to leave the filesystem in a 100% clean state on a normal unmount.
1649 hammer2_voldata_lock(hmp
);
1650 hammer2_voldata_unlock(hmp
);
1652 hammer2_iocom_uninit(hmp
);
1654 if ((hmp
->vchain
.flags
| hmp
->fchain
.flags
) &
1655 HAMMER2_CHAIN_FLUSH_MASK
) {
1656 kprintf("hammer2_unmount: chains left over "
1657 "after final sync\n");
1658 kprintf(" vchain %08x\n", hmp
->vchain
.flags
);
1659 kprintf(" fchain %08x\n", hmp
->fchain
.flags
);
1661 if (hammer2_debug
& 0x0010)
1662 Debugger("entered debugger");
1665 KKASSERT(hmp
->spmp
== NULL
);
1668 * Finish up with the device vnode
1670 if ((devvp
= hmp
->devvp
) != NULL
) {
1672 vn_lock(devvp
, LK_EXCLUSIVE
| LK_RETRY
);
1673 vinvalbuf(devvp
, (ronly
? 0 : V_SAVE
), 0, 0);
1675 VOP_CLOSE(devvp
, (ronly
? FREAD
: FREAD
|FWRITE
), NULL
);
1682 * Clear vchain/fchain flags that might prevent final cleanup
1685 if (hmp
->vchain
.flags
& HAMMER2_CHAIN_MODIFIED
) {
1686 atomic_add_long(&hammer2_count_modified_chains
, -1);
1687 atomic_clear_int(&hmp
->vchain
.flags
, HAMMER2_CHAIN_MODIFIED
);
1688 hammer2_pfs_memory_wakeup(hmp
->vchain
.pmp
);
1690 if (hmp
->vchain
.flags
& HAMMER2_CHAIN_UPDATE
) {
1691 atomic_clear_int(&hmp
->vchain
.flags
, HAMMER2_CHAIN_UPDATE
);
1694 if (hmp
->fchain
.flags
& HAMMER2_CHAIN_MODIFIED
) {
1695 atomic_add_long(&hammer2_count_modified_chains
, -1);
1696 atomic_clear_int(&hmp
->fchain
.flags
, HAMMER2_CHAIN_MODIFIED
);
1697 hammer2_pfs_memory_wakeup(hmp
->fchain
.pmp
);
1699 if (hmp
->fchain
.flags
& HAMMER2_CHAIN_UPDATE
) {
1700 atomic_clear_int(&hmp
->fchain
.flags
, HAMMER2_CHAIN_UPDATE
);
1704 * Final drop of embedded freemap root chain to
1705 * clean up fchain.core (fchain structure is not
1706 * flagged ALLOCATED so it is cleaned out and then
1709 hammer2_chain_drop(&hmp
->fchain
);
1712 * Final drop of embedded volume root chain to clean
1713 * up vchain.core (vchain structure is not flagged
1714 * ALLOCATED so it is cleaned out and then left to
1718 hammer2_dump_chain(&hmp
->vchain
, 0, &dumpcnt
, 'v');
1720 hammer2_dump_chain(&hmp
->fchain
, 0, &dumpcnt
, 'f');
1721 hammer2_dev_unlock(hmp
);
1722 hammer2_chain_drop(&hmp
->vchain
);
1724 hammer2_io_cleanup(hmp
, &hmp
->iotree
);
1725 if (hmp
->iofree_count
) {
1726 kprintf("io_cleanup: %d I/O's left hanging\n",
1730 TAILQ_REMOVE(&hammer2_mntlist
, hmp
, mntentry
);
1731 kmalloc_destroy(&hmp
->mchain
);
1732 kfree(hmp
, M_HAMMER2
);
1736 hammer2_vfs_vget(struct mount
*mp
, struct vnode
*dvp
,
1737 ino_t ino
, struct vnode
**vpp
)
1739 hammer2_xop_lookup_t
*xop
;
1741 hammer2_inode_t
*ip
;
1745 inum
= (hammer2_tid_t
)ino
& HAMMER2_DIRHASH_USERMSK
;
1751 * Easy if we already have it cached
1753 ip
= hammer2_inode_lookup(pmp
, inum
);
1755 hammer2_inode_lock(ip
, HAMMER2_RESOLVE_SHARED
);
1756 *vpp
= hammer2_igetv(ip
, &error
);
1757 hammer2_inode_unlock(ip
);
1758 hammer2_inode_drop(ip
); /* from lookup */
1764 * Otherwise we have to find the inode
1766 xop
= hammer2_xop_alloc(pmp
->iroot
, 0);
1768 hammer2_xop_start(&xop
->head
, hammer2_xop_lookup
);
1769 error
= hammer2_xop_collect(&xop
->head
, 0);
1772 if (hammer2_cluster_rdata(&xop
->head
.cluster
) == NULL
) {
1773 kprintf("vget: no collect error but also no rdata\n");
1774 kprintf("xop %p\n", xop
);
1775 while ((hammer2_debug
& 0x80000) == 0) {
1776 tsleep(xop
, PCATCH
, "wait", hz
* 10);
1780 ip
= hammer2_inode_get(pmp
, NULL
, &xop
->head
.cluster
, -1);
1783 hammer2_xop_retire(&xop
->head
, HAMMER2_XOPMASK_VOP
);
1786 *vpp
= hammer2_igetv(ip
, &error
);
1787 hammer2_inode_unlock(ip
);
1797 hammer2_vfs_root(struct mount
*mp
, struct vnode
**vpp
)
1804 if (pmp
->iroot
== NULL
) {
1810 hammer2_inode_lock(pmp
->iroot
, HAMMER2_RESOLVE_SHARED
);
1812 while (pmp
->inode_tid
== 0) {
1813 hammer2_xop_ipcluster_t
*xop
;
1814 hammer2_inode_meta_t
*meta
;
1816 xop
= hammer2_xop_alloc(pmp
->iroot
, HAMMER2_XOP_MODIFYING
);
1817 hammer2_xop_start(&xop
->head
, hammer2_xop_ipcluster
);
1818 error
= hammer2_xop_collect(&xop
->head
, 0);
1821 meta
= &xop
->head
.cluster
.focus
->data
->ipdata
.meta
;
1822 pmp
->iroot
->meta
= *meta
;
1823 pmp
->inode_tid
= meta
->pfs_inum
+ 1;
1824 if (pmp
->inode_tid
< HAMMER2_INODE_START
)
1825 pmp
->inode_tid
= HAMMER2_INODE_START
;
1827 xop
->head
.cluster
.focus
->bref
.modify_tid
+ 1;
1828 kprintf("PFS: Starting inode %jd\n",
1829 (intmax_t)pmp
->inode_tid
);
1830 kprintf("PMP focus good set nextino=%ld mod=%016jx\n",
1831 pmp
->inode_tid
, pmp
->modify_tid
);
1832 wakeup(&pmp
->iroot
);
1834 hammer2_xop_retire(&xop
->head
, HAMMER2_XOPMASK_VOP
);
1837 * Prime the mount info.
1839 hammer2_vfs_statfs(mp
, &mp
->mnt_stat
, NULL
);
1846 hammer2_xop_retire(&xop
->head
, HAMMER2_XOPMASK_VOP
);
1847 hammer2_inode_unlock(pmp
->iroot
);
1848 error
= tsleep(&pmp
->iroot
, PCATCH
, "h2root", hz
);
1849 hammer2_inode_lock(pmp
->iroot
, HAMMER2_RESOLVE_SHARED
);
1855 hammer2_inode_unlock(pmp
->iroot
);
1858 vp
= hammer2_igetv(pmp
->iroot
, &error
);
1859 hammer2_inode_unlock(pmp
->iroot
);
1869 * XXX incorporate ipdata->meta.inode_quota and data_quota
1873 hammer2_vfs_statfs(struct mount
*mp
, struct statfs
*sbp
, struct ucred
*cred
)
1877 hammer2_blockref_t bref
;
1881 * NOTE: iroot might not have validated the cluster yet.
1885 mp
->mnt_stat
.f_files
= 0;
1886 mp
->mnt_stat
.f_ffree
= 0;
1887 mp
->mnt_stat
.f_blocks
= 0;
1888 mp
->mnt_stat
.f_bfree
= 0;
1889 mp
->mnt_stat
.f_bavail
= 0;
1891 for (i
= 0; i
< pmp
->iroot
->cluster
.nchains
; ++i
) {
1892 hmp
= pmp
->pfs_hmps
[i
];
1895 if (pmp
->iroot
->cluster
.array
[i
].chain
)
1896 bref
= pmp
->iroot
->cluster
.array
[i
].chain
->bref
;
1898 bzero(&bref
, sizeof(bref
));
1900 mp
->mnt_stat
.f_files
= bref
.embed
.stats
.inode_count
;
1901 mp
->mnt_stat
.f_ffree
= 0;
1902 mp
->mnt_stat
.f_blocks
= hmp
->voldata
.allocator_size
/
1903 mp
->mnt_vstat
.f_bsize
;
1904 mp
->mnt_stat
.f_bfree
= hmp
->voldata
.allocator_free
/
1905 mp
->mnt_vstat
.f_bsize
;
1906 mp
->mnt_stat
.f_bavail
= mp
->mnt_stat
.f_bfree
;
1908 *sbp
= mp
->mnt_stat
;
1915 hammer2_vfs_statvfs(struct mount
*mp
, struct statvfs
*sbp
, struct ucred
*cred
)
1919 hammer2_blockref_t bref
;
1923 * NOTE: iroot might not have validated the cluster yet.
1927 mp
->mnt_vstat
.f_bsize
= 0;
1928 mp
->mnt_vstat
.f_files
= 0;
1929 mp
->mnt_vstat
.f_ffree
= 0;
1930 mp
->mnt_vstat
.f_blocks
= 0;
1931 mp
->mnt_vstat
.f_bfree
= 0;
1932 mp
->mnt_vstat
.f_bavail
= 0;
1934 for (i
= 0; i
< pmp
->iroot
->cluster
.nchains
; ++i
) {
1935 hmp
= pmp
->pfs_hmps
[i
];
1938 if (pmp
->iroot
->cluster
.array
[i
].chain
)
1939 bref
= pmp
->iroot
->cluster
.array
[i
].chain
->bref
;
1941 bzero(&bref
, sizeof(bref
));
1943 mp
->mnt_vstat
.f_bsize
= HAMMER2_PBUFSIZE
;
1944 mp
->mnt_vstat
.f_files
= bref
.embed
.stats
.inode_count
;
1945 mp
->mnt_vstat
.f_ffree
= 0;
1946 mp
->mnt_vstat
.f_blocks
= hmp
->voldata
.allocator_size
/
1947 mp
->mnt_vstat
.f_bsize
;
1948 mp
->mnt_vstat
.f_bfree
= hmp
->voldata
.allocator_free
/
1949 mp
->mnt_vstat
.f_bsize
;
1950 mp
->mnt_vstat
.f_bavail
= mp
->mnt_vstat
.f_bfree
;
1952 *sbp
= mp
->mnt_vstat
;
1958 * Mount-time recovery (RW mounts)
1960 * Updates to the free block table are allowed to lag flushes by one
1961 * transaction. In case of a crash, then on a fresh mount we must do an
1962 * incremental scan of the last committed transaction id and make sure that
1963 * all related blocks have been marked allocated.
1965 * The super-root topology and each PFS has its own transaction id domain,
1966 * so we must track PFS boundary transitions.
1968 struct hammer2_recovery_elm
{
1969 TAILQ_ENTRY(hammer2_recovery_elm
) entry
;
1970 hammer2_chain_t
*chain
;
1971 hammer2_tid_t sync_tid
;
1974 TAILQ_HEAD(hammer2_recovery_list
, hammer2_recovery_elm
);
1976 struct hammer2_recovery_info
{
1977 struct hammer2_recovery_list list
;
1982 static int hammer2_recovery_scan(hammer2_dev_t
*hmp
,
1983 hammer2_chain_t
*parent
,
1984 struct hammer2_recovery_info
*info
,
1985 hammer2_tid_t sync_tid
);
1987 #define HAMMER2_RECOVERY_MAXDEPTH 10
1991 hammer2_recovery(hammer2_dev_t
*hmp
)
1993 struct hammer2_recovery_info info
;
1994 struct hammer2_recovery_elm
*elm
;
1995 hammer2_chain_t
*parent
;
1996 hammer2_tid_t sync_tid
;
1997 hammer2_tid_t mirror_tid
;
1999 int cumulative_error
= 0;
2001 hammer2_trans_init(hmp
->spmp
, 0);
2003 sync_tid
= hmp
->voldata
.freemap_tid
;
2004 mirror_tid
= hmp
->voldata
.mirror_tid
;
2006 kprintf("hammer2 mount \"%s\": ", hmp
->devrepname
);
2007 if (sync_tid
>= mirror_tid
) {
2008 kprintf(" no recovery needed\n");
2010 kprintf(" freemap recovery %016jx-%016jx\n",
2011 sync_tid
+ 1, mirror_tid
);
2014 TAILQ_INIT(&info
.list
);
2016 parent
= hammer2_chain_lookup_init(&hmp
->vchain
, 0);
2017 cumulative_error
= hammer2_recovery_scan(hmp
, parent
, &info
, sync_tid
);
2018 hammer2_chain_lookup_done(parent
);
2020 while ((elm
= TAILQ_FIRST(&info
.list
)) != NULL
) {
2021 TAILQ_REMOVE(&info
.list
, elm
, entry
);
2022 parent
= elm
->chain
;
2023 sync_tid
= elm
->sync_tid
;
2024 kfree(elm
, M_HAMMER2
);
2026 hammer2_chain_lock(parent
, HAMMER2_RESOLVE_ALWAYS
);
2027 error
= hammer2_recovery_scan(hmp
, parent
, &info
,
2028 hmp
->voldata
.freemap_tid
);
2029 hammer2_chain_unlock(parent
);
2030 hammer2_chain_drop(parent
); /* drop elm->chain ref */
2032 cumulative_error
= error
;
2034 hammer2_trans_done(hmp
->spmp
);
2036 return cumulative_error
;
2041 hammer2_recovery_scan(hammer2_dev_t
*hmp
, hammer2_chain_t
*parent
,
2042 struct hammer2_recovery_info
*info
,
2043 hammer2_tid_t sync_tid
)
2045 const hammer2_inode_data_t
*ripdata
;
2046 hammer2_chain_t
*chain
;
2047 hammer2_blockref_t bref
;
2049 int cumulative_error
= 0;
2054 * Adjust freemap to ensure that the block(s) are marked allocated.
2056 if (parent
->bref
.type
!= HAMMER2_BREF_TYPE_VOLUME
) {
2057 hammer2_freemap_adjust(hmp
, &parent
->bref
,
2058 HAMMER2_FREEMAP_DORECOVER
);
2062 * Check type for recursive scan
2064 switch(parent
->bref
.type
) {
2065 case HAMMER2_BREF_TYPE_VOLUME
:
2066 /* data already instantiated */
2068 case HAMMER2_BREF_TYPE_INODE
:
2070 * Must instantiate data for DIRECTDATA test and also
2073 hammer2_chain_lock(parent
, HAMMER2_RESOLVE_ALWAYS
);
2074 ripdata
= &hammer2_chain_rdata(parent
)->ipdata
;
2075 if (ripdata
->meta
.op_flags
& HAMMER2_OPFLAG_DIRECTDATA
) {
2076 /* not applicable to recovery scan */
2077 hammer2_chain_unlock(parent
);
2080 hammer2_chain_unlock(parent
);
2082 case HAMMER2_BREF_TYPE_INDIRECT
:
2084 * Must instantiate data for recursion
2086 hammer2_chain_lock(parent
, HAMMER2_RESOLVE_ALWAYS
);
2087 hammer2_chain_unlock(parent
);
2089 case HAMMER2_BREF_TYPE_DIRENT
:
2090 case HAMMER2_BREF_TYPE_DATA
:
2091 case HAMMER2_BREF_TYPE_FREEMAP
:
2092 case HAMMER2_BREF_TYPE_FREEMAP_NODE
:
2093 case HAMMER2_BREF_TYPE_FREEMAP_LEAF
:
2094 /* not applicable to recovery scan */
2102 * Defer operation if depth limit reached or if we are crossing a
2105 if (info
->depth
>= HAMMER2_RECOVERY_MAXDEPTH
) {
2106 struct hammer2_recovery_elm
*elm
;
2108 elm
= kmalloc(sizeof(*elm
), M_HAMMER2
, M_ZERO
| M_WAITOK
);
2109 elm
->chain
= parent
;
2110 elm
->sync_tid
= sync_tid
;
2111 hammer2_chain_ref(parent
);
2112 TAILQ_INSERT_TAIL(&info
->list
, elm
, entry
);
2113 /* unlocked by caller */
2120 * Recursive scan of the last flushed transaction only. We are
2121 * doing this without pmp assignments so don't leave the chains
2122 * hanging around after we are done with them.
2128 while (hammer2_chain_scan(parent
, &chain
, &bref
,
2129 &first
, &cache_index
,
2130 HAMMER2_LOOKUP_NODATA
) != NULL
) {
2134 if (chain
== NULL
) {
2135 if (bref
.mirror_tid
> sync_tid
) {
2136 hammer2_freemap_adjust(hmp
, &bref
,
2137 HAMMER2_FREEMAP_DORECOVER
);
2143 * This may or may not be a recursive node.
2145 atomic_set_int(&chain
->flags
, HAMMER2_CHAIN_RELEASE
);
2146 if (bref
.mirror_tid
> sync_tid
) {
2148 error
= hammer2_recovery_scan(hmp
, chain
,
2152 cumulative_error
= error
;
2156 * Flush the recovery at the PFS boundary to stage it for
2157 * the final flush of the super-root topology.
2159 if ((bref
.flags
& HAMMER2_BREF_FLAG_PFSROOT
) &&
2160 (chain
->flags
& HAMMER2_CHAIN_ONFLUSH
)) {
2161 hammer2_flush(chain
, HAMMER2_FLUSH_TOP
);
2165 return cumulative_error
;
2169 * Sync a mount point; this is called on a per-mount basis from the
2170 * filesystem syncer process periodically and whenever a user issues
2174 hammer2_vfs_sync(struct mount
*mp
, int waitfor
)
2176 hammer2_xop_flush_t
*xop
;
2177 struct hammer2_sync_info info
;
2178 hammer2_inode_t
*iroot
;
2186 KKASSERT(iroot
->pmp
== pmp
);
2189 * We can't acquire locks on existing vnodes while in a transaction
2190 * without risking a deadlock. This assumes that vfsync() can be
2191 * called without the vnode locked (which it can in DragonFly).
2192 * Otherwise we'd have to implement a multi-pass or flag the lock
2193 * failures and retry.
2195 * The reclamation code interlocks with the sync list's token
2196 * (by removing the vnode from the scan list) before unlocking
2197 * the inode, giving us time to ref the inode.
2199 /*flags = VMSC_GETVP;*/
2201 if (waitfor
& MNT_LAZY
)
2202 flags
|= VMSC_ONEPASS
;
2205 * Preflush the vnodes using a normal transaction before interlocking
2206 * with a flush transaction. We do this to try to run as much of
2207 * the compression as possible outside the flush transaction.
2209 * For efficiency do an async pass before making sure with a
2210 * synchronous pass on all related buffer cache buffers.
2212 hammer2_trans_init(pmp
, 0);
2214 info
.waitfor
= MNT_NOWAIT
;
2215 vsyncscan(mp
, flags
| VMSC_NOWAIT
, hammer2_sync_scan2
, &info
);
2216 info
.waitfor
= MNT_WAIT
;
2217 vsyncscan(mp
, flags
, hammer2_sync_scan2
, &info
);
2218 hammer2_trans_done(pmp
);
2221 * Start our flush transaction. This does not return until all
2222 * concurrent transactions have completed and will prevent any
2223 * new transactions from running concurrently, except for the
2224 * buffer cache transactions.
2226 * NOTE! It is still possible for the paging code to push pages
2227 * out via a UIO_NOCOPY hammer2_vop_write() during the main
2230 hammer2_trans_init(pmp
, HAMMER2_TRANS_ISFLUSH
);
2231 hammer2_inode_run_sideq(pmp
);
2234 info
.waitfor
= MNT_NOWAIT
;
2235 vsyncscan(mp
, flags
| VMSC_NOWAIT
, hammer2_sync_scan2
, &info
);
2236 info
.waitfor
= MNT_WAIT
;
2237 vsyncscan(mp
, flags
, hammer2_sync_scan2
, &info
);
2238 hammer2_bioq_sync(pmp
);
2241 * Use the XOP interface to concurrently flush all nodes to
2242 * synchronize the PFSROOT subtopology to the media. A standard
2243 * end-of-scan ENOENT error indicates cluster sufficiency.
2245 * Note that this flush will not be visible on crash recovery until
2246 * we flush the super-root topology in the next loop.
2248 * XXX For now wait for all flushes to complete.
2251 xop
= hammer2_xop_alloc(iroot
, HAMMER2_XOP_MODIFYING
);
2252 hammer2_xop_start(&xop
->head
, hammer2_inode_xop_flush
);
2253 error
= hammer2_xop_collect(&xop
->head
,
2254 HAMMER2_XOP_COLLECT_WAITALL
);
2255 hammer2_xop_retire(&xop
->head
, HAMMER2_XOPMASK_VOP
);
2256 if (error
== ENOENT
)
2261 hammer2_trans_done(pmp
);
2269 * Note that we ignore the tranasction mtid we got above. Instead,
2270 * each vfsync below will ultimately get its own via TRANS_BUFCACHE
2274 hammer2_sync_scan2(struct mount
*mp
, struct vnode
*vp
, void *data
)
2276 struct hammer2_sync_info
*info
= data
;
2277 hammer2_inode_t
*ip
;
2281 * Degenerate cases. Note that ip == NULL typically means the
2282 * syncer vnode itself and we don't want to vclrisdirty() in that
2289 if (vp
->v_type
== VNON
|| vp
->v_type
== VBAD
) {
2295 * VOP_FSYNC will start a new transaction so replicate some code
2296 * here to do it inline (see hammer2_vop_fsync()).
2298 * WARNING: The vfsync interacts with the buffer cache and might
2299 * block, we can't hold the inode lock at that time.
2300 * However, we MUST ref ip before blocking to ensure that
2301 * it isn't ripped out from under us (since we do not
2302 * hold a lock on the vnode).
2304 hammer2_inode_ref(ip
);
2305 if ((ip
->flags
& HAMMER2_INODE_MODIFIED
) ||
2306 !RB_EMPTY(&vp
->v_rbdirty_tree
)) {
2307 vfsync(vp
, info
->waitfor
, 1, NULL
, NULL
);
2308 if (ip
->flags
& (HAMMER2_INODE_RESIZED
|
2309 HAMMER2_INODE_MODIFIED
)) {
2310 hammer2_inode_lock(ip
, 0);
2311 hammer2_inode_chain_sync(ip
);
2312 hammer2_inode_unlock(ip
);
2315 if ((ip
->flags
& HAMMER2_INODE_MODIFIED
) == 0 &&
2316 RB_EMPTY(&vp
->v_rbdirty_tree
)) {
2320 hammer2_inode_drop(ip
);
2324 info
->error
= error
;
2331 hammer2_vfs_vptofh(struct vnode
*vp
, struct fid
*fhp
)
2333 hammer2_inode_t
*ip
;
2335 KKASSERT(MAXFIDSZ
>= 16);
2337 fhp
->fid_len
= offsetof(struct fid
, fid_data
[16]);
2339 ((hammer2_tid_t
*)fhp
->fid_data
)[0] = ip
->meta
.inum
;
2340 ((hammer2_tid_t
*)fhp
->fid_data
)[1] = 0;
2347 hammer2_vfs_fhtovp(struct mount
*mp
, struct vnode
*rootvp
,
2348 struct fid
*fhp
, struct vnode
**vpp
)
2355 inum
= ((hammer2_tid_t
*)fhp
->fid_data
)[0] & HAMMER2_DIRHASH_USERMSK
;
2358 error
= hammer2_vfs_root(mp
, vpp
);
2360 error
= hammer2_vfs_vget(mp
, NULL
, inum
, vpp
);
2365 kprintf("fhtovp: %016jx -> %p, %d\n", inum
, *vpp
, error
);
2371 hammer2_vfs_checkexp(struct mount
*mp
, struct sockaddr
*nam
,
2372 int *exflagsp
, struct ucred
**credanonp
)
2379 np
= vfs_export_lookup(mp
, &pmp
->export
, nam
);
2381 *exflagsp
= np
->netc_exflags
;
2382 *credanonp
= &np
->netc_anon
;
2391 * Support code for hammer2_vfs_mount(). Read, verify, and install the volume
2392 * header into the HMP
2394 * XXX read four volhdrs and use the one with the highest TID whos CRC
2399 * XXX For filesystems w/ less than 4 volhdrs, make sure to not write to
2400 * nonexistant locations.
2402 * XXX Record selected volhdr and ring updates to each of 4 volhdrs
2406 hammer2_install_volume_header(hammer2_dev_t
*hmp
)
2408 hammer2_volume_data_t
*vd
;
2410 hammer2_crc32_t crc0
, crc
, bcrc0
, bcrc
;
2422 * There are up to 4 copies of the volume header (syncs iterate
2423 * between them so there is no single master). We don't trust the
2424 * volu_size field so we don't know precisely how large the filesystem
2425 * is, so depend on the OS to return an error if we go beyond the
2426 * block device's EOF.
2428 for (i
= 0; i
< HAMMER2_NUM_VOLHDRS
; i
++) {
2429 error
= bread(hmp
->devvp
, i
* HAMMER2_ZONE_BYTES64
,
2430 HAMMER2_VOLUME_BYTES
, &bp
);
2437 vd
= (struct hammer2_volume_data
*) bp
->b_data
;
2438 if ((vd
->magic
!= HAMMER2_VOLUME_ID_HBO
) &&
2439 (vd
->magic
!= HAMMER2_VOLUME_ID_ABO
)) {
2445 if (vd
->magic
== HAMMER2_VOLUME_ID_ABO
) {
2446 /* XXX: Reversed-endianness filesystem */
2447 kprintf("hammer2: reverse-endian filesystem detected");
2453 crc
= vd
->icrc_sects
[HAMMER2_VOL_ICRC_SECT0
];
2454 crc0
= hammer2_icrc32(bp
->b_data
+ HAMMER2_VOLUME_ICRC0_OFF
,
2455 HAMMER2_VOLUME_ICRC0_SIZE
);
2456 bcrc
= vd
->icrc_sects
[HAMMER2_VOL_ICRC_SECT1
];
2457 bcrc0
= hammer2_icrc32(bp
->b_data
+ HAMMER2_VOLUME_ICRC1_OFF
,
2458 HAMMER2_VOLUME_ICRC1_SIZE
);
2459 if ((crc0
!= crc
) || (bcrc0
!= bcrc
)) {
2460 kprintf("hammer2 volume header crc "
2461 "mismatch copy #%d %08x/%08x\n",
2468 if (valid
== 0 || hmp
->voldata
.mirror_tid
< vd
->mirror_tid
) {
2477 hmp
->volsync
= hmp
->voldata
;
2479 if (error_reported
|| bootverbose
|| 1) { /* 1/DEBUG */
2480 kprintf("hammer2: using volume header #%d\n",
2485 kprintf("hammer2: no valid volume headers found!\n");
2491 * This handles hysteresis on regular file flushes. Because the BIOs are
2492 * routed to a thread it is possible for an excessive number to build up
2493 * and cause long front-end stalls long before the runningbuffspace limit
2494 * is hit, so we implement hammer2_flush_pipe to control the
2497 * This is a particular problem when compression is used.
2500 hammer2_lwinprog_ref(hammer2_pfs_t
*pmp
)
2502 atomic_add_int(&pmp
->count_lwinprog
, 1);
2506 hammer2_lwinprog_drop(hammer2_pfs_t
*pmp
)
2510 lwinprog
= atomic_fetchadd_int(&pmp
->count_lwinprog
, -1);
2511 if ((lwinprog
& HAMMER2_LWINPROG_WAITING
) &&
2512 (lwinprog
& HAMMER2_LWINPROG_MASK
) <= hammer2_flush_pipe
* 2 / 3) {
2513 atomic_clear_int(&pmp
->count_lwinprog
,
2514 HAMMER2_LWINPROG_WAITING
);
2515 wakeup(&pmp
->count_lwinprog
);
2517 if ((lwinprog
& HAMMER2_LWINPROG_WAITING0
) &&
2518 (lwinprog
& HAMMER2_LWINPROG_MASK
) <= 0) {
2519 atomic_clear_int(&pmp
->count_lwinprog
,
2520 HAMMER2_LWINPROG_WAITING0
);
2521 wakeup(&pmp
->count_lwinprog
);
2526 hammer2_lwinprog_wait(hammer2_pfs_t
*pmp
, int flush_pipe
)
2529 int lwflag
= (flush_pipe
) ? HAMMER2_LWINPROG_WAITING
:
2530 HAMMER2_LWINPROG_WAITING0
;
2533 lwinprog
= pmp
->count_lwinprog
;
2535 if ((lwinprog
& HAMMER2_LWINPROG_MASK
) <= flush_pipe
)
2537 tsleep_interlock(&pmp
->count_lwinprog
, 0);
2538 atomic_set_int(&pmp
->count_lwinprog
, lwflag
);
2539 lwinprog
= pmp
->count_lwinprog
;
2540 if ((lwinprog
& HAMMER2_LWINPROG_MASK
) <= flush_pipe
)
2542 tsleep(&pmp
->count_lwinprog
, PINTERLOCKED
, "h2wpipe", hz
);
2547 * Manage excessive memory resource use for chain and related
2551 hammer2_pfs_memory_wait(hammer2_pfs_t
*pmp
)
2561 * Atomic check condition and wait. Also do an early speedup of
2562 * the syncer to try to avoid hitting the wait.
2565 waiting
= pmp
->inmem_dirty_chains
;
2567 count
= waiting
& HAMMER2_DIRTYCHAIN_MASK
;
2569 limit
= pmp
->mp
->mnt_nvnodelistsize
/ 10;
2570 if (limit
< hammer2_limit_dirty_chains
)
2571 limit
= hammer2_limit_dirty_chains
;
2576 if ((int)(ticks
- zzticks
) > hz
) {
2578 kprintf("count %ld %ld\n", count
, limit
);
2583 * Block if there are too many dirty chains present, wait
2584 * for the flush to clean some out.
2586 if (count
> limit
) {
2587 tsleep_interlock(&pmp
->inmem_dirty_chains
, 0);
2588 if (atomic_cmpset_int(&pmp
->inmem_dirty_chains
,
2590 waiting
| HAMMER2_DIRTYCHAIN_WAITING
)) {
2591 speedup_syncer(pmp
->mp
);
2592 tsleep(&pmp
->inmem_dirty_chains
, PINTERLOCKED
,
2595 continue; /* loop on success or fail */
2599 * Try to start an early flush before we are forced to block.
2601 if (count
> limit
* 7 / 10)
2602 speedup_syncer(pmp
->mp
);
2608 hammer2_pfs_memory_inc(hammer2_pfs_t
*pmp
)
2611 atomic_add_int(&pmp
->inmem_dirty_chains
, 1);
2616 hammer2_pfs_memory_wakeup(hammer2_pfs_t
*pmp
)
2624 waiting
= pmp
->inmem_dirty_chains
;
2626 if (atomic_cmpset_int(&pmp
->inmem_dirty_chains
,
2629 ~HAMMER2_DIRTYCHAIN_WAITING
)) {
2634 if (waiting
& HAMMER2_DIRTYCHAIN_WAITING
)
2635 wakeup(&pmp
->inmem_dirty_chains
);
2642 hammer2_dump_chain(hammer2_chain_t
*chain
, int tab
, int *countp
, char pfx
)
2644 hammer2_chain_t
*scan
;
2645 hammer2_chain_t
*parent
;
2649 kprintf("%*.*s...\n", tab
, tab
, "");
2654 kprintf("%*.*s%c-chain %p.%d %016jx/%d mir=%016jx\n",
2656 chain
, chain
->bref
.type
,
2657 chain
->bref
.key
, chain
->bref
.keybits
,
2658 chain
->bref
.mirror_tid
);
2660 kprintf("%*.*s [%08x] (%s) refs=%d",
2663 ((chain
->bref
.type
== HAMMER2_BREF_TYPE_INODE
&&
2664 chain
->data
) ? (char *)chain
->data
->ipdata
.filename
: "?"),
2667 parent
= chain
->parent
;
2669 kprintf("\n%*.*s p=%p [pflags %08x prefs %d",
2671 parent
, parent
->flags
, parent
->refs
);
2672 if (RB_EMPTY(&chain
->core
.rbtree
)) {
2676 RB_FOREACH(scan
, hammer2_chain_tree
, &chain
->core
.rbtree
)
2677 hammer2_dump_chain(scan
, tab
+ 4, countp
, 'a');
2678 if (chain
->bref
.type
== HAMMER2_BREF_TYPE_INODE
&& chain
->data
)
2679 kprintf("%*.*s}(%s)\n", tab
, tab
, "",
2680 chain
->data
->ipdata
.filename
);
2682 kprintf("%*.*s}\n", tab
, tab
, "");