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_always_compress
= 0; /* always try to compress */
85 int hammer2_inval_enable
= 0;
86 int hammer2_flush_pipe
= 100;
87 int hammer2_synchronous_flush
= 1;
88 int hammer2_dio_count
;
89 int hammer2_limit_dio
= 256;
90 long hammer2_chain_allocs
;
91 long hammer2_chain_frees
;
92 long hammer2_limit_dirty_chains
;
93 long hammer2_count_modified_chains
;
94 long hammer2_iod_invals
;
95 long hammer2_iod_file_read
;
96 long hammer2_iod_meta_read
;
97 long hammer2_iod_indr_read
;
98 long hammer2_iod_fmap_read
;
99 long hammer2_iod_volu_read
;
100 long hammer2_iod_file_write
;
101 long hammer2_iod_file_wembed
;
102 long hammer2_iod_file_wzero
;
103 long hammer2_iod_file_wdedup
;
104 long hammer2_iod_meta_write
;
105 long hammer2_iod_indr_write
;
106 long hammer2_iod_fmap_write
;
107 long hammer2_iod_volu_write
;
109 MALLOC_DECLARE(M_HAMMER2_CBUFFER
);
110 MALLOC_DEFINE(M_HAMMER2_CBUFFER
, "HAMMER2-compbuffer",
111 "Buffer used for compression.");
113 MALLOC_DECLARE(M_HAMMER2_DEBUFFER
);
114 MALLOC_DEFINE(M_HAMMER2_DEBUFFER
, "HAMMER2-decompbuffer",
115 "Buffer used for decompression.");
117 SYSCTL_NODE(_vfs
, OID_AUTO
, hammer2
, CTLFLAG_RW
, 0, "HAMMER2 filesystem");
119 SYSCTL_INT(_vfs_hammer2
, OID_AUTO
, debug
, CTLFLAG_RW
,
120 &hammer2_debug
, 0, "");
121 SYSCTL_INT(_vfs_hammer2
, OID_AUTO
, cluster_read
, CTLFLAG_RW
,
122 &hammer2_cluster_read
, 0, "");
123 SYSCTL_INT(_vfs_hammer2
, OID_AUTO
, cluster_write
, CTLFLAG_RW
,
124 &hammer2_cluster_write
, 0, "");
125 SYSCTL_INT(_vfs_hammer2
, OID_AUTO
, dedup_enable
, CTLFLAG_RW
,
126 &hammer2_dedup_enable
, 0, "");
127 SYSCTL_INT(_vfs_hammer2
, OID_AUTO
, always_compress
, CTLFLAG_RW
,
128 &hammer2_always_compress
, 0, "");
129 SYSCTL_INT(_vfs_hammer2
, OID_AUTO
, inval_enable
, CTLFLAG_RW
,
130 &hammer2_inval_enable
, 0, "");
131 SYSCTL_INT(_vfs_hammer2
, OID_AUTO
, flush_pipe
, CTLFLAG_RW
,
132 &hammer2_flush_pipe
, 0, "");
133 SYSCTL_INT(_vfs_hammer2
, OID_AUTO
, synchronous_flush
, CTLFLAG_RW
,
134 &hammer2_synchronous_flush
, 0, "");
135 SYSCTL_LONG(_vfs_hammer2
, OID_AUTO
, chain_allocs
, CTLFLAG_RW
,
136 &hammer2_chain_allocs
, 0, "");
137 SYSCTL_LONG(_vfs_hammer2
, OID_AUTO
, chain_frees
, CTLFLAG_RW
,
138 &hammer2_chain_frees
, 0, "");
139 SYSCTL_LONG(_vfs_hammer2
, OID_AUTO
, limit_dirty_chains
, CTLFLAG_RW
,
140 &hammer2_limit_dirty_chains
, 0, "");
141 SYSCTL_LONG(_vfs_hammer2
, OID_AUTO
, count_modified_chains
, CTLFLAG_RW
,
142 &hammer2_count_modified_chains
, 0, "");
143 SYSCTL_INT(_vfs_hammer2
, OID_AUTO
, dio_count
, CTLFLAG_RD
,
144 &hammer2_dio_count
, 0, "");
145 SYSCTL_INT(_vfs_hammer2
, OID_AUTO
, limit_dio
, CTLFLAG_RW
,
146 &hammer2_limit_dio
, 0, "");
148 SYSCTL_LONG(_vfs_hammer2
, OID_AUTO
, iod_invals
, CTLFLAG_RW
,
149 &hammer2_iod_invals
, 0, "");
150 SYSCTL_LONG(_vfs_hammer2
, OID_AUTO
, iod_file_read
, CTLFLAG_RW
,
151 &hammer2_iod_file_read
, 0, "");
152 SYSCTL_LONG(_vfs_hammer2
, OID_AUTO
, iod_meta_read
, CTLFLAG_RW
,
153 &hammer2_iod_meta_read
, 0, "");
154 SYSCTL_LONG(_vfs_hammer2
, OID_AUTO
, iod_indr_read
, CTLFLAG_RW
,
155 &hammer2_iod_indr_read
, 0, "");
156 SYSCTL_LONG(_vfs_hammer2
, OID_AUTO
, iod_fmap_read
, CTLFLAG_RW
,
157 &hammer2_iod_fmap_read
, 0, "");
158 SYSCTL_LONG(_vfs_hammer2
, OID_AUTO
, iod_volu_read
, CTLFLAG_RW
,
159 &hammer2_iod_volu_read
, 0, "");
161 SYSCTL_LONG(_vfs_hammer2
, OID_AUTO
, iod_file_write
, CTLFLAG_RW
,
162 &hammer2_iod_file_write
, 0, "");
163 SYSCTL_LONG(_vfs_hammer2
, OID_AUTO
, iod_file_wembed
, CTLFLAG_RW
,
164 &hammer2_iod_file_wembed
, 0, "");
165 SYSCTL_LONG(_vfs_hammer2
, OID_AUTO
, iod_file_wzero
, CTLFLAG_RW
,
166 &hammer2_iod_file_wzero
, 0, "");
167 SYSCTL_LONG(_vfs_hammer2
, OID_AUTO
, iod_file_wdedup
, CTLFLAG_RW
,
168 &hammer2_iod_file_wdedup
, 0, "");
169 SYSCTL_LONG(_vfs_hammer2
, OID_AUTO
, iod_meta_write
, CTLFLAG_RW
,
170 &hammer2_iod_meta_write
, 0, "");
171 SYSCTL_LONG(_vfs_hammer2
, OID_AUTO
, iod_indr_write
, CTLFLAG_RW
,
172 &hammer2_iod_indr_write
, 0, "");
173 SYSCTL_LONG(_vfs_hammer2
, OID_AUTO
, iod_fmap_write
, CTLFLAG_RW
,
174 &hammer2_iod_fmap_write
, 0, "");
175 SYSCTL_LONG(_vfs_hammer2
, OID_AUTO
, iod_volu_write
, CTLFLAG_RW
,
176 &hammer2_iod_volu_write
, 0, "");
178 long hammer2_check_icrc32
;
179 long hammer2_check_xxhash64
;
180 SYSCTL_LONG(_vfs_hammer2
, OID_AUTO
, check_icrc32
, CTLFLAG_RW
,
181 &hammer2_check_icrc32
, 0, "");
182 SYSCTL_LONG(_vfs_hammer2
, OID_AUTO
, check_xxhash64
, CTLFLAG_RW
,
183 &hammer2_check_xxhash64
, 0, "");
185 static int hammer2_vfs_init(struct vfsconf
*conf
);
186 static int hammer2_vfs_uninit(struct vfsconf
*vfsp
);
187 static int hammer2_vfs_mount(struct mount
*mp
, char *path
, caddr_t data
,
189 static int hammer2_remount(hammer2_dev_t
*, struct mount
*, char *,
190 struct vnode
*, struct ucred
*);
191 static int hammer2_recovery(hammer2_dev_t
*hmp
);
192 static int hammer2_vfs_unmount(struct mount
*mp
, int mntflags
);
193 static int hammer2_vfs_root(struct mount
*mp
, struct vnode
**vpp
);
194 static int hammer2_vfs_statfs(struct mount
*mp
, struct statfs
*sbp
,
196 static int hammer2_vfs_statvfs(struct mount
*mp
, struct statvfs
*sbp
,
198 static int hammer2_vfs_fhtovp(struct mount
*mp
, struct vnode
*rootvp
,
199 struct fid
*fhp
, struct vnode
**vpp
);
200 static int hammer2_vfs_vptofh(struct vnode
*vp
, struct fid
*fhp
);
201 static int hammer2_vfs_checkexp(struct mount
*mp
, struct sockaddr
*nam
,
202 int *exflagsp
, struct ucred
**credanonp
);
204 static int hammer2_install_volume_header(hammer2_dev_t
*hmp
);
205 static int hammer2_sync_scan2(struct mount
*mp
, struct vnode
*vp
, void *data
);
207 static void hammer2_update_pmps(hammer2_dev_t
*hmp
);
209 static void hammer2_mount_helper(struct mount
*mp
, hammer2_pfs_t
*pmp
);
210 static void hammer2_unmount_helper(struct mount
*mp
, hammer2_pfs_t
*pmp
,
214 * HAMMER2 vfs operations.
216 static struct vfsops hammer2_vfsops
= {
217 .vfs_init
= hammer2_vfs_init
,
218 .vfs_uninit
= hammer2_vfs_uninit
,
219 .vfs_sync
= hammer2_vfs_sync
,
220 .vfs_mount
= hammer2_vfs_mount
,
221 .vfs_unmount
= hammer2_vfs_unmount
,
222 .vfs_root
= hammer2_vfs_root
,
223 .vfs_statfs
= hammer2_vfs_statfs
,
224 .vfs_statvfs
= hammer2_vfs_statvfs
,
225 .vfs_vget
= hammer2_vfs_vget
,
226 .vfs_vptofh
= hammer2_vfs_vptofh
,
227 .vfs_fhtovp
= hammer2_vfs_fhtovp
,
228 .vfs_checkexp
= hammer2_vfs_checkexp
231 MALLOC_DEFINE(M_HAMMER2
, "HAMMER2-mount", "");
233 VFS_SET(hammer2_vfsops
, hammer2
, VFCF_MPSAFE
);
234 MODULE_VERSION(hammer2
, 1);
238 hammer2_vfs_init(struct vfsconf
*conf
)
240 static struct objcache_malloc_args margs_read
;
241 static struct objcache_malloc_args margs_write
;
242 static struct objcache_malloc_args margs_vop
;
249 * A large DIO cache is needed to retain dedup enablement masks.
250 * The bulkfree code clears related masks as part of the disk block
251 * recycling algorithm, preventing it from being used for a later
254 * NOTE: A large buffer cache can actually interfere with dedup
255 * operation because we dedup based on media physical buffers
256 * and not logical buffers. Try to make the DIO chace large
257 * enough to avoid this problem, but also cap it.
259 hammer2_limit_dio
= nbuf
* 2;
260 if (hammer2_limit_dio
> 100000)
261 hammer2_limit_dio
= 100000;
263 if (HAMMER2_BLOCKREF_BYTES
!= sizeof(struct hammer2_blockref
))
265 if (HAMMER2_INODE_BYTES
!= sizeof(struct hammer2_inode_data
))
267 if (HAMMER2_VOLUME_BYTES
!= sizeof(struct hammer2_volume_data
))
271 kprintf("HAMMER2 structure size mismatch; cannot continue.\n");
273 margs_read
.objsize
= 65536;
274 margs_read
.mtype
= M_HAMMER2_DEBUFFER
;
276 margs_write
.objsize
= 32768;
277 margs_write
.mtype
= M_HAMMER2_CBUFFER
;
279 margs_vop
.objsize
= sizeof(hammer2_xop_t
);
280 margs_vop
.mtype
= M_HAMMER2
;
283 * Note thaht for the XOPS cache we want backing store allocations
284 * to use M_ZERO. This is not allowed in objcache_get() (to avoid
285 * confusion), so use the backing store function that does it. This
286 * means that initial XOPS objects are zerod but REUSED objects are
287 * not. So we are responsible for cleaning the object up sufficiently
288 * for our needs before objcache_put()ing it back (typically just the
291 cache_buffer_read
= objcache_create(margs_read
.mtype
->ks_shortdesc
,
292 0, 1, NULL
, NULL
, NULL
,
293 objcache_malloc_alloc
,
294 objcache_malloc_free
,
296 cache_buffer_write
= objcache_create(margs_write
.mtype
->ks_shortdesc
,
297 0, 1, NULL
, NULL
, NULL
,
298 objcache_malloc_alloc
,
299 objcache_malloc_free
,
301 cache_xops
= objcache_create(margs_vop
.mtype
->ks_shortdesc
,
302 0, 1, NULL
, NULL
, NULL
,
303 objcache_malloc_alloc_zero
,
304 objcache_malloc_free
,
308 lockinit(&hammer2_mntlk
, "mntlk", 0, 0);
309 TAILQ_INIT(&hammer2_mntlist
);
310 TAILQ_INIT(&hammer2_pfslist
);
312 hammer2_limit_dirty_chains
= maxvnodes
/ 10;
313 if (hammer2_limit_dirty_chains
> HAMMER2_LIMIT_DIRTY_CHAINS
)
314 hammer2_limit_dirty_chains
= HAMMER2_LIMIT_DIRTY_CHAINS
;
321 hammer2_vfs_uninit(struct vfsconf
*vfsp __unused
)
323 objcache_destroy(cache_buffer_read
);
324 objcache_destroy(cache_buffer_write
);
325 objcache_destroy(cache_xops
);
330 * Core PFS allocator. Used to allocate or reference the pmp structure
331 * for PFS cluster mounts and the spmp structure for media (hmp) structures.
332 * The pmp can be passed in or loaded by this function using the chain and
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
,
344 const hammer2_inode_data_t
*ripdata
,
345 hammer2_tid_t modify_tid
, hammer2_dev_t
*force_local
)
348 hammer2_inode_t
*iroot
;
356 * Locate or create the PFS based on the cluster id. If ripdata
357 * is NULL this is a spmp which is unique and is always allocated.
359 * If the device is mounted in local mode all PFSs are considered
360 * independent and not part of any cluster (for debugging only).
363 TAILQ_FOREACH(pmp
, &hammer2_pfslist
, mntentry
) {
364 if (force_local
!= pmp
->force_local
)
366 if (force_local
== NULL
&&
367 bcmp(&pmp
->pfs_clid
, &ripdata
->meta
.pfs_clid
,
368 sizeof(pmp
->pfs_clid
)) == 0) {
370 } else if (force_local
&& pmp
->pfs_names
[0] &&
371 strcmp(pmp
->pfs_names
[0], ripdata
->filename
) == 0) {
378 pmp
= kmalloc(sizeof(*pmp
), M_HAMMER2
, M_WAITOK
| M_ZERO
);
379 pmp
->force_local
= force_local
;
380 hammer2_trans_manage_init(pmp
);
381 kmalloc_create(&pmp
->minode
, "HAMMER2-inodes");
382 kmalloc_create(&pmp
->mmsg
, "HAMMER2-pfsmsg");
383 lockinit(&pmp
->lock
, "pfslk", 0, 0);
384 lockinit(&pmp
->lock_nlink
, "h2nlink", 0, 0);
385 spin_init(&pmp
->inum_spin
, "hm2pfsalloc_inum");
386 spin_init(&pmp
->xop_spin
, "h2xop");
387 spin_init(&pmp
->lru_spin
, "h2lru");
388 RB_INIT(&pmp
->inum_tree
);
389 TAILQ_INIT(&pmp
->sideq
);
390 TAILQ_INIT(&pmp
->lru_list
);
391 spin_init(&pmp
->list_spin
, "hm2pfsalloc_list");
394 * Distribute backend operations to threads
396 for (i
= 0; i
< HAMMER2_XOPGROUPS
; ++i
)
397 hammer2_xop_group_init(pmp
, &pmp
->xop_groups
[i
]);
400 * Save the last media transaction id for the flusher. Set
404 pmp
->pfs_clid
= ripdata
->meta
.pfs_clid
;
405 TAILQ_INSERT_TAIL(&hammer2_pfslist
, pmp
, mntentry
);
408 * The synchronization thread may start too early, make
409 * sure it stays frozen until we are ready to let it go.
413 pmp->primary_thr.flags = HAMMER2_THREAD_FROZEN |
414 HAMMER2_THREAD_REMASTER;
419 * Create the PFS's root inode and any missing XOP helper threads.
421 if ((iroot
= pmp
->iroot
) == NULL
) {
422 iroot
= hammer2_inode_get(pmp
, NULL
, NULL
, -1);
424 iroot
->meta
= ripdata
->meta
;
426 hammer2_inode_ref(iroot
);
427 hammer2_inode_unlock(iroot
);
431 * Stop here if no chain is passed in.
437 * When a chain is passed in we must add it to the PFS's root
438 * inode, update pmp->pfs_types[], and update the syncronization
441 * When forcing local mode, mark the PFS as a MASTER regardless.
443 * At the moment empty spots can develop due to removals or failures.
444 * Ultimately we want to re-fill these spots but doing so might
445 * confused running code. XXX
447 hammer2_inode_ref(iroot
);
448 hammer2_mtx_ex(&iroot
->lock
);
449 j
= iroot
->cluster
.nchains
;
451 kprintf("add PFS to pmp %p[%d]\n", pmp
, j
);
453 if (j
== HAMMER2_MAXCLUSTER
) {
454 kprintf("hammer2_mount: cluster full!\n");
455 /* XXX fatal error? */
457 KKASSERT(chain
->pmp
== NULL
);
459 hammer2_chain_ref(chain
);
460 iroot
->cluster
.array
[j
].chain
= chain
;
462 pmp
->pfs_types
[j
] = HAMMER2_PFSTYPE_MASTER
;
464 pmp
->pfs_types
[j
] = ripdata
->meta
.pfs_type
;
465 pmp
->pfs_names
[j
] = kstrdup(ripdata
->filename
, M_HAMMER2
);
466 pmp
->pfs_hmps
[j
] = chain
->hmp
;
469 * If the PFS is already mounted we must account
470 * for the mount_count here.
473 ++chain
->hmp
->mount_count
;
476 * May have to fixup dirty chain tracking. Previous
477 * pmp was NULL so nothing to undo.
479 if (chain
->flags
& HAMMER2_CHAIN_MODIFIED
)
480 hammer2_pfs_memory_inc(pmp
);
483 iroot
->cluster
.nchains
= j
;
486 * Update nmasters from any PFS inode which is part of the cluster.
487 * It is possible that this will result in a value which is too
488 * high. MASTER PFSs are authoritative for pfs_nmasters and will
489 * override this value later on.
491 * (This informs us of masters that might not currently be
492 * discoverable by this mount).
494 if (ripdata
&& pmp
->pfs_nmasters
< ripdata
->meta
.pfs_nmasters
) {
495 pmp
->pfs_nmasters
= ripdata
->meta
.pfs_nmasters
;
499 * Count visible masters. Masters are usually added with
500 * ripdata->meta.pfs_nmasters set to 1. This detects when there
501 * are more (XXX and must update the master inodes).
504 for (i
= 0; i
< iroot
->cluster
.nchains
; ++i
) {
505 if (pmp
->pfs_types
[i
] == HAMMER2_PFSTYPE_MASTER
)
508 if (pmp
->pfs_nmasters
< count
)
509 pmp
->pfs_nmasters
= count
;
512 * Create missing synchronization and support threads.
514 * Single-node masters (including snapshots) have nothing to
515 * synchronize and do not require this thread.
517 * Multi-node masters or any number of soft masters, slaves, copy,
518 * or other PFS types need the thread.
520 * Each thread is responsible for its particular cluster index.
521 * We use independent threads so stalls or mismatches related to
522 * any given target do not affect other targets.
524 for (i
= 0; i
< iroot
->cluster
.nchains
; ++i
) {
526 * Single-node masters (including snapshots) have nothing
527 * to synchronize and will make direct xops support calls,
528 * thus they do not require this thread.
530 * Note that there can be thousands of snapshots. We do not
531 * want to create thousands of threads.
533 if (pmp
->pfs_nmasters
<= 1 &&
534 pmp
->pfs_types
[i
] == HAMMER2_PFSTYPE_MASTER
) {
539 * Sync support thread
541 if (pmp
->sync_thrs
[i
].td
== NULL
) {
542 hammer2_thr_create(&pmp
->sync_thrs
[i
], pmp
, NULL
,
544 hammer2_primary_sync_thread
);
549 * Create missing Xop threads
551 * NOTE: We create helper threads for all mounted PFSs or any
552 * PFSs with 2+ nodes (so the sync thread can update them,
553 * even if not mounted).
555 if (pmp
->mp
|| iroot
->cluster
.nchains
>= 2)
556 hammer2_xop_helper_create(pmp
);
558 hammer2_mtx_unlock(&iroot
->lock
);
559 hammer2_inode_drop(iroot
);
565 * Deallocate an element of a probed PFS. If destroying and this is a
566 * MASTER, adjust nmasters.
568 * This function does not physically destroy the PFS element in its device
569 * under the super-root (see hammer2_ioctl_pfs_delete()).
572 hammer2_pfsdealloc(hammer2_pfs_t
*pmp
, int clindex
, int destroying
)
574 hammer2_inode_t
*iroot
;
575 hammer2_chain_t
*chain
;
579 * Cleanup our reference on iroot. iroot is (should) not be needed
587 * XXX flush after acquiring the iroot lock.
588 * XXX clean out the cluster index from all inode structures.
590 hammer2_thr_delete(&pmp
->sync_thrs
[clindex
]);
593 * Remove the cluster index from the group. If destroying
594 * the PFS and this is a master, adjust pfs_nmasters.
596 hammer2_mtx_ex(&iroot
->lock
);
597 chain
= iroot
->cluster
.array
[clindex
].chain
;
598 iroot
->cluster
.array
[clindex
].chain
= NULL
;
600 switch(pmp
->pfs_types
[clindex
]) {
601 case HAMMER2_PFSTYPE_MASTER
:
602 if (destroying
&& pmp
->pfs_nmasters
> 0)
604 /* XXX adjust ripdata->meta.pfs_nmasters */
609 pmp
->pfs_types
[clindex
] = HAMMER2_PFSTYPE_NONE
;
611 hammer2_mtx_unlock(&iroot
->lock
);
617 atomic_set_int(&chain
->flags
, HAMMER2_CHAIN_RELEASE
);
618 hammer2_chain_drop(chain
);
622 * Terminate all XOP threads for the cluster index.
624 for (j
= 0; j
< HAMMER2_XOPGROUPS
; ++j
)
625 hammer2_thr_delete(&pmp
->xop_groups
[j
].thrs
[clindex
]);
630 * Destroy a PFS, typically only occurs after the last mount on a device
634 hammer2_pfsfree(hammer2_pfs_t
*pmp
)
636 hammer2_inode_t
*iroot
;
637 hammer2_chain_t
*chain
;
642 * Cleanup our reference on iroot. iroot is (should) not be needed
645 TAILQ_REMOVE(&hammer2_pfslist
, pmp
, mntentry
);
649 for (i
= 0; i
< iroot
->cluster
.nchains
; ++i
) {
650 hammer2_thr_delete(&pmp
->sync_thrs
[i
]);
651 for (j
= 0; j
< HAMMER2_XOPGROUPS
; ++j
)
652 hammer2_thr_delete(&pmp
->xop_groups
[j
].thrs
[i
]);
654 #if REPORT_REFS_ERRORS
655 if (pmp
->iroot
->refs
!= 1)
656 kprintf("PMP->IROOT %p REFS WRONG %d\n",
657 pmp
->iroot
, pmp
->iroot
->refs
);
659 KKASSERT(pmp
->iroot
->refs
== 1);
661 /* ref for pmp->iroot */
662 hammer2_inode_drop(pmp
->iroot
);
667 * Cleanup chains remaining on LRU list.
669 kprintf("pfsfree: %p lrucount=%d\n", pmp
, pmp
->lru_count
);
670 while ((chain
= TAILQ_FIRST(&pmp
->lru_list
)) != NULL
) {
671 hammer2_chain_ref(chain
);
672 atomic_set_int(&chain
->flags
, HAMMER2_CHAIN_RELEASE
);
673 hammer2_chain_drop(chain
);
677 * Free remaining pmp resources
679 kmalloc_destroy(&pmp
->mmsg
);
680 kmalloc_destroy(&pmp
->minode
);
682 kfree(pmp
, M_HAMMER2
);
686 * Remove all references to hmp from the pfs list. Any PFS which becomes
687 * empty is terminated and freed.
692 hammer2_pfsfree_scan(hammer2_dev_t
*hmp
)
695 hammer2_inode_t
*iroot
;
696 hammer2_chain_t
*rchain
;
702 TAILQ_FOREACH(pmp
, &hammer2_pfslist
, mntentry
) {
703 if ((iroot
= pmp
->iroot
) == NULL
)
705 if (hmp
->spmp
== pmp
) {
706 kprintf("unmount hmp %p remove spmp %p\n",
712 * Determine if this PFS is affected. If it is we must
713 * freeze all management threads and lock its iroot.
715 * Freezing a management thread forces it idle, operations
716 * in-progress will be aborted and it will have to start
717 * over again when unfrozen, or exit if told to exit.
719 for (i
= 0; i
< HAMMER2_MAXCLUSTER
; ++i
) {
720 if (pmp
->pfs_hmps
[i
] == hmp
)
723 if (i
!= HAMMER2_MAXCLUSTER
) {
725 * Make sure all synchronization threads are locked
728 for (i
= 0; i
< HAMMER2_MAXCLUSTER
; ++i
) {
729 if (pmp
->pfs_hmps
[i
] == NULL
)
731 hammer2_thr_freeze_async(&pmp
->sync_thrs
[i
]);
732 for (j
= 0; j
< HAMMER2_XOPGROUPS
; ++j
) {
733 hammer2_thr_freeze_async(
734 &pmp
->xop_groups
[j
].thrs
[i
]);
737 for (i
= 0; i
< HAMMER2_MAXCLUSTER
; ++i
) {
738 if (pmp
->pfs_hmps
[i
] == NULL
)
740 hammer2_thr_freeze(&pmp
->sync_thrs
[i
]);
741 for (j
= 0; j
< HAMMER2_XOPGROUPS
; ++j
) {
743 &pmp
->xop_groups
[j
].thrs
[i
]);
748 * Lock the inode and clean out matching chains.
749 * Note that we cannot use hammer2_inode_lock_*()
750 * here because that would attempt to validate the
751 * cluster that we are in the middle of ripping
754 * WARNING! We are working directly on the inodes
757 hammer2_mtx_ex(&iroot
->lock
);
760 * Remove the chain from matching elements of the PFS.
762 for (i
= 0; i
< HAMMER2_MAXCLUSTER
; ++i
) {
763 if (pmp
->pfs_hmps
[i
] != hmp
)
765 hammer2_thr_delete(&pmp
->sync_thrs
[i
]);
766 for (j
= 0; j
< HAMMER2_XOPGROUPS
; ++j
) {
768 &pmp
->xop_groups
[j
].thrs
[i
]);
770 rchain
= iroot
->cluster
.array
[i
].chain
;
771 iroot
->cluster
.array
[i
].chain
= NULL
;
772 pmp
->pfs_types
[i
] = 0;
773 if (pmp
->pfs_names
[i
]) {
774 kfree(pmp
->pfs_names
[i
], M_HAMMER2
);
775 pmp
->pfs_names
[i
] = NULL
;
778 hammer2_chain_drop(rchain
);
780 if (iroot
->cluster
.focus
== rchain
)
781 iroot
->cluster
.focus
= NULL
;
783 pmp
->pfs_hmps
[i
] = NULL
;
785 hammer2_mtx_unlock(&iroot
->lock
);
786 didfreeze
= 1; /* remaster, unfreeze down below */
792 * Cleanup trailing chains. Gaps may remain.
794 for (i
= HAMMER2_MAXCLUSTER
- 1; i
>= 0; --i
) {
795 if (pmp
->pfs_hmps
[i
])
798 iroot
->cluster
.nchains
= i
+ 1;
801 * If the PMP has no elements remaining we can destroy it.
802 * (this will transition management threads from frozen->exit).
804 if (iroot
->cluster
.nchains
== 0) {
805 kprintf("unmount hmp %p last ref to PMP=%p\n",
807 hammer2_pfsfree(pmp
);
812 * If elements still remain we need to set the REMASTER
813 * flag and unfreeze it.
816 for (i
= 0; i
< HAMMER2_MAXCLUSTER
; ++i
) {
817 if (pmp
->pfs_hmps
[i
] == NULL
)
819 hammer2_thr_remaster(&pmp
->sync_thrs
[i
]);
820 hammer2_thr_unfreeze(&pmp
->sync_thrs
[i
]);
821 for (j
= 0; j
< HAMMER2_XOPGROUPS
; ++j
) {
822 hammer2_thr_remaster(
823 &pmp
->xop_groups
[j
].thrs
[i
]);
824 hammer2_thr_unfreeze(
825 &pmp
->xop_groups
[j
].thrs
[i
]);
833 * Mount or remount HAMMER2 fileystem from physical media
836 * mp mount point structure
842 * mp mount point structure
843 * path path to mount point
844 * data pointer to argument structure in user space
845 * volume volume path (device@LABEL form)
846 * hflags user mount flags
847 * cred user credentials
854 hammer2_vfs_mount(struct mount
*mp
, char *path
, caddr_t data
,
857 struct hammer2_mount_info info
;
861 hammer2_dev_t
*force_local
;
862 hammer2_key_t key_next
;
863 hammer2_key_t key_dummy
;
866 struct nlookupdata nd
;
867 hammer2_chain_t
*parent
;
868 hammer2_chain_t
*chain
;
869 hammer2_cluster_t
*cluster
;
870 const hammer2_inode_data_t
*ripdata
;
871 hammer2_blockref_t bref
;
873 char devstr
[MNAMELEN
];
890 kprintf("hammer2_mount\n");
896 bzero(&info
, sizeof(info
));
897 info
.cluster_fd
= -1;
898 ksnprintf(devstr
, sizeof(devstr
), "%s",
899 mp
->mnt_stat
.f_mntfromname
);
900 kprintf("hammer2_mount: root '%s'\n", devstr
);
903 * Non-root mount or updating a mount
905 error
= copyin(data
, &info
, sizeof(info
));
909 error
= copyinstr(info
.volume
, devstr
, MNAMELEN
- 1, &done
);
915 * Extract device and label, automatically mount @BOOT, @ROOT, or @DATA
916 * if no label specified, based on the partition id. Error out if no
917 * partition id. This is strictly a convenience to match the
918 * default label created by newfs_hammer2, our preference is
919 * that a label always be specified.
922 label
= strchr(devstr
, '@');
923 if (label
&& ((label
+ 1) - dev
) > done
)
925 if (label
&& label
== devstr
)
927 if (label
== NULL
|| label
[1] == 0) {
931 label
= devstr
+ strlen(devstr
);
949 kprintf("hammer2_mount: dev=\"%s\" label=\"%s\" rdonly=%d\n",
950 dev
, label
, (mp
->mnt_flag
& MNT_RDONLY
));
952 if (mp
->mnt_flag
& MNT_UPDATE
) {
954 * Update mount. Note that pmp->iroot->cluster is
955 * an inode-embedded cluster and thus cannot be
958 * XXX HAMMER2 needs to implement NFS export via
962 pmp
->hflags
= info
.hflags
;
963 cluster
= &pmp
->iroot
->cluster
;
964 for (i
= 0; i
< cluster
->nchains
; ++i
) {
965 if (cluster
->array
[i
].chain
== NULL
)
967 hmp
= cluster
->array
[i
].chain
->hmp
;
969 error
= hammer2_remount(hmp
, mp
, path
,
981 * If a path is specified and dev is not an empty string, lookup the
982 * name and verify that it referes to a block device.
984 * If a path is specified and dev is an empty string we fall through
985 * and locate the label in the hmp search.
987 if (path
&& *dev
!= 0) {
988 error
= nlookup_init(&nd
, dev
, UIO_SYSSPACE
, NLC_FOLLOW
);
990 error
= nlookup(&nd
);
992 error
= cache_vref(&nd
.nl_nch
, nd
.nl_cred
, &devvp
);
994 } else if (path
== NULL
) {
996 cdev_t cdev
= kgetdiskbyname(dev
);
997 error
= bdevvp(cdev
, &devvp
);
999 kprintf("hammer2: cannot find '%s'\n", dev
);
1002 * We will locate the hmp using the label in the hmp loop.
1008 * Make sure its a block device. Do not check to see if it is
1009 * already mounted until we determine that its a fresh H2 device.
1011 if (error
== 0 && devvp
) {
1012 vn_isdisk(devvp
, &error
);
1016 * Determine if the device has already been mounted. After this
1017 * check hmp will be non-NULL if we are doing the second or more
1018 * hammer2 mounts from the same device.
1020 lockmgr(&hammer2_mntlk
, LK_EXCLUSIVE
);
1023 * Match the device. Due to the way devfs works,
1024 * we may not be able to directly match the vnode pointer,
1025 * so also check to see if the underlying device matches.
1027 TAILQ_FOREACH(hmp
, &hammer2_mntlist
, mntentry
) {
1028 if (hmp
->devvp
== devvp
)
1030 if (devvp
->v_rdev
&&
1031 hmp
->devvp
->v_rdev
== devvp
->v_rdev
) {
1037 * If no match this may be a fresh H2 mount, make sure
1038 * the device is not mounted on anything else.
1041 error
= vfs_mountedon(devvp
);
1042 } else if (error
== 0) {
1044 * Match the label to a pmp already probed.
1046 TAILQ_FOREACH(pmp
, &hammer2_pfslist
, mntentry
) {
1047 for (i
= 0; i
< HAMMER2_MAXCLUSTER
; ++i
) {
1048 if (pmp
->pfs_names
[i
] &&
1049 strcmp(pmp
->pfs_names
[i
], label
) == 0) {
1050 hmp
= pmp
->pfs_hmps
[i
];
1062 * Open the device if this isn't a secondary mount and construct
1063 * the H2 device mount (hmp).
1066 hammer2_chain_t
*schain
;
1069 if (error
== 0 && vcount(devvp
) > 0) {
1070 kprintf("Primary device already has references\n");
1075 * Now open the device
1078 ronly
= ((mp
->mnt_flag
& MNT_RDONLY
) != 0);
1079 vn_lock(devvp
, LK_EXCLUSIVE
| LK_RETRY
);
1080 error
= vinvalbuf(devvp
, V_SAVE
, 0, 0);
1082 error
= VOP_OPEN(devvp
,
1083 (ronly
? FREAD
: FREAD
| FWRITE
),
1088 if (error
&& devvp
) {
1093 lockmgr(&hammer2_mntlk
, LK_RELEASE
);
1096 hmp
= kmalloc(sizeof(*hmp
), M_HAMMER2
, M_WAITOK
| M_ZERO
);
1097 ksnprintf(hmp
->devrepname
, sizeof(hmp
->devrepname
), "%s", dev
);
1100 hmp
->hflags
= info
.hflags
& HMNT2_DEVFLAGS
;
1101 kmalloc_create(&hmp
->mchain
, "HAMMER2-chains");
1102 TAILQ_INSERT_TAIL(&hammer2_mntlist
, hmp
, mntentry
);
1103 RB_INIT(&hmp
->iotree
);
1104 spin_init(&hmp
->io_spin
, "hm2mount_io");
1105 spin_init(&hmp
->list_spin
, "hm2mount_list");
1106 TAILQ_INIT(&hmp
->flushq
);
1108 lockinit(&hmp
->vollk
, "h2vol", 0, 0);
1109 lockinit(&hmp
->bulklk
, "h2bulk", 0, 0);
1110 lockinit(&hmp
->bflock
, "h2bflk", 0, 0);
1113 * vchain setup. vchain.data is embedded.
1114 * vchain.refs is initialized and will never drop to 0.
1116 * NOTE! voldata is not yet loaded.
1118 hmp
->vchain
.hmp
= hmp
;
1119 hmp
->vchain
.refs
= 1;
1120 hmp
->vchain
.data
= (void *)&hmp
->voldata
;
1121 hmp
->vchain
.bref
.type
= HAMMER2_BREF_TYPE_VOLUME
;
1122 hmp
->vchain
.bref
.data_off
= 0 | HAMMER2_PBUFRADIX
;
1123 hmp
->vchain
.bref
.mirror_tid
= hmp
->voldata
.mirror_tid
;
1125 hammer2_chain_core_init(&hmp
->vchain
);
1126 /* hmp->vchain.u.xxx is left NULL */
1129 * fchain setup. fchain.data is embedded.
1130 * fchain.refs is initialized and will never drop to 0.
1132 * The data is not used but needs to be initialized to
1133 * pass assertion muster. We use this chain primarily
1134 * as a placeholder for the freemap's top-level RBTREE
1135 * so it does not interfere with the volume's topology
1138 hmp
->fchain
.hmp
= hmp
;
1139 hmp
->fchain
.refs
= 1;
1140 hmp
->fchain
.data
= (void *)&hmp
->voldata
.freemap_blockset
;
1141 hmp
->fchain
.bref
.type
= HAMMER2_BREF_TYPE_FREEMAP
;
1142 hmp
->fchain
.bref
.data_off
= 0 | HAMMER2_PBUFRADIX
;
1143 hmp
->fchain
.bref
.mirror_tid
= hmp
->voldata
.freemap_tid
;
1144 hmp
->fchain
.bref
.methods
=
1145 HAMMER2_ENC_CHECK(HAMMER2_CHECK_FREEMAP
) |
1146 HAMMER2_ENC_COMP(HAMMER2_COMP_NONE
);
1148 hammer2_chain_core_init(&hmp
->fchain
);
1149 /* hmp->fchain.u.xxx is left NULL */
1152 * Install the volume header and initialize fields from
1155 error
= hammer2_install_volume_header(hmp
);
1157 hammer2_unmount_helper(mp
, NULL
, hmp
);
1158 lockmgr(&hammer2_mntlk
, LK_RELEASE
);
1159 hammer2_vfs_unmount(mp
, MNT_FORCE
);
1164 * Really important to get these right or flush will get
1167 hmp
->spmp
= hammer2_pfsalloc(NULL
, NULL
, 0, NULL
);
1168 kprintf("alloc spmp %p tid %016jx\n",
1169 hmp
->spmp
, hmp
->voldata
.mirror_tid
);
1173 * Dummy-up vchain and fchain's modify_tid. mirror_tid
1174 * is inherited from the volume header.
1177 hmp
->vchain
.bref
.mirror_tid
= hmp
->voldata
.mirror_tid
;
1178 hmp
->vchain
.bref
.modify_tid
= hmp
->vchain
.bref
.mirror_tid
;
1179 hmp
->vchain
.pmp
= spmp
;
1180 hmp
->fchain
.bref
.mirror_tid
= hmp
->voldata
.freemap_tid
;
1181 hmp
->fchain
.bref
.modify_tid
= hmp
->fchain
.bref
.mirror_tid
;
1182 hmp
->fchain
.pmp
= spmp
;
1185 * First locate the super-root inode, which is key 0
1186 * relative to the volume header's blockset.
1188 * Then locate the root inode by scanning the directory keyspace
1189 * represented by the label.
1191 parent
= hammer2_chain_lookup_init(&hmp
->vchain
, 0);
1192 schain
= hammer2_chain_lookup(&parent
, &key_dummy
,
1193 HAMMER2_SROOT_KEY
, HAMMER2_SROOT_KEY
,
1195 hammer2_chain_lookup_done(parent
);
1196 if (schain
== NULL
) {
1197 kprintf("hammer2_mount: invalid super-root\n");
1198 hammer2_unmount_helper(mp
, NULL
, hmp
);
1199 lockmgr(&hammer2_mntlk
, LK_RELEASE
);
1200 hammer2_vfs_unmount(mp
, MNT_FORCE
);
1203 if (schain
->error
) {
1204 kprintf("hammer2_mount: error %s reading super-root\n",
1205 hammer2_error_str(schain
->error
));
1206 hammer2_chain_unlock(schain
);
1207 hammer2_chain_drop(schain
);
1209 hammer2_unmount_helper(mp
, NULL
, hmp
);
1210 lockmgr(&hammer2_mntlk
, LK_RELEASE
);
1211 hammer2_vfs_unmount(mp
, MNT_FORCE
);
1216 * The super-root always uses an inode_tid of 1 when
1219 spmp
->inode_tid
= 1;
1220 spmp
->modify_tid
= schain
->bref
.modify_tid
+ 1;
1223 * Sanity-check schain's pmp and finish initialization.
1224 * Any chain belonging to the super-root topology should
1225 * have a NULL pmp (not even set to spmp).
1227 ripdata
= &hammer2_chain_rdata(schain
)->ipdata
;
1228 KKASSERT(schain
->pmp
== NULL
);
1229 spmp
->pfs_clid
= ripdata
->meta
.pfs_clid
;
1232 * Replace the dummy spmp->iroot with a real one. It's
1233 * easier to just do a wholesale replacement than to try
1234 * to update the chain and fixup the iroot fields.
1236 * The returned inode is locked with the supplied cluster.
1238 cluster
= hammer2_cluster_from_chain(schain
);
1239 hammer2_inode_drop(spmp
->iroot
);
1241 spmp
->iroot
= hammer2_inode_get(spmp
, NULL
, cluster
, -1);
1242 spmp
->spmp_hmp
= hmp
;
1243 spmp
->pfs_types
[0] = ripdata
->meta
.pfs_type
;
1244 spmp
->pfs_hmps
[0] = hmp
;
1245 hammer2_inode_ref(spmp
->iroot
);
1246 hammer2_inode_unlock(spmp
->iroot
);
1247 hammer2_cluster_unlock(cluster
);
1248 hammer2_cluster_drop(cluster
);
1250 /* leave spmp->iroot with one ref */
1252 if ((mp
->mnt_flag
& MNT_RDONLY
) == 0) {
1253 error
= hammer2_recovery(hmp
);
1254 /* XXX do something with error */
1256 hammer2_update_pmps(hmp
);
1257 hammer2_iocom_init(hmp
);
1258 hammer2_bulkfree_init(hmp
);
1261 * Ref the cluster management messaging descriptor. The mount
1262 * program deals with the other end of the communications pipe.
1264 * Root mounts typically do not supply one.
1266 if (info
.cluster_fd
>= 0) {
1267 fp
= holdfp(curproc
->p_fd
, info
.cluster_fd
, -1);
1269 hammer2_cluster_reconnect(hmp
, fp
);
1271 kprintf("hammer2_mount: bad cluster_fd!\n");
1276 if (info
.hflags
& HMNT2_DEVFLAGS
) {
1277 kprintf("hammer2: Warning: mount flags pertaining "
1278 "to the whole device may only be specified "
1279 "on the first mount of the device: %08x\n",
1280 info
.hflags
& HMNT2_DEVFLAGS
);
1285 * Force local mount (disassociate all PFSs from their clusters).
1286 * Used primarily for debugging.
1288 force_local
= (hmp
->hflags
& HMNT2_LOCAL
) ? hmp
: NULL
;
1291 * Lookup the mount point under the media-localized super-root.
1292 * Scanning hammer2_pfslist doesn't help us because it represents
1293 * PFS cluster ids which can aggregate several named PFSs together.
1295 * cluster->pmp will incorrectly point to spmp and must be fixed
1298 hammer2_inode_lock(spmp
->iroot
, 0);
1299 parent
= hammer2_inode_chain(spmp
->iroot
, 0, HAMMER2_RESOLVE_ALWAYS
);
1300 lhc
= hammer2_dirhash(label
, strlen(label
));
1301 chain
= hammer2_chain_lookup(&parent
, &key_next
,
1302 lhc
, lhc
+ HAMMER2_DIRHASH_LOMASK
,
1305 if (chain
->bref
.type
== HAMMER2_BREF_TYPE_INODE
&&
1306 strcmp(label
, chain
->data
->ipdata
.filename
) == 0) {
1309 chain
= hammer2_chain_next(&parent
, chain
, &key_next
,
1311 lhc
+ HAMMER2_DIRHASH_LOMASK
,
1315 hammer2_chain_unlock(parent
);
1316 hammer2_chain_drop(parent
);
1318 hammer2_inode_unlock(spmp
->iroot
);
1321 * PFS could not be found?
1323 if (chain
== NULL
) {
1324 kprintf("hammer2_mount: PFS label not found\n");
1325 hammer2_unmount_helper(mp
, NULL
, hmp
);
1326 lockmgr(&hammer2_mntlk
, LK_RELEASE
);
1327 hammer2_vfs_unmount(mp
, MNT_FORCE
);
1333 * Acquire the pmp structure (it should have already been allocated
1334 * via hammer2_update_pmps() so do not pass cluster in to add to
1335 * available chains).
1337 * Check if the cluster has already been mounted. A cluster can
1338 * only be mounted once, use null mounts to mount additional copies.
1340 ripdata
= &chain
->data
->ipdata
;
1342 pmp
= hammer2_pfsalloc(NULL
, ripdata
,
1343 bref
.modify_tid
, force_local
);
1344 hammer2_chain_unlock(chain
);
1345 hammer2_chain_drop(chain
);
1350 kprintf("hammer2_mount hmp=%p pmp=%p\n", hmp
, pmp
);
1353 kprintf("hammer2_mount: PFS already mounted!\n");
1354 hammer2_unmount_helper(mp
, NULL
, hmp
);
1355 lockmgr(&hammer2_mntlk
, LK_RELEASE
);
1356 hammer2_vfs_unmount(mp
, MNT_FORCE
);
1361 pmp
->hflags
= info
.hflags
;
1362 mp
->mnt_flag
|= MNT_LOCAL
;
1363 mp
->mnt_kern_flag
|= MNTK_ALL_MPSAFE
; /* all entry pts are SMP */
1364 mp
->mnt_kern_flag
|= MNTK_THR_SYNC
; /* new vsyncscan semantics */
1367 * required mount structure initializations
1369 mp
->mnt_stat
.f_iosize
= HAMMER2_PBUFSIZE
;
1370 mp
->mnt_stat
.f_bsize
= HAMMER2_PBUFSIZE
;
1372 mp
->mnt_vstat
.f_frsize
= HAMMER2_PBUFSIZE
;
1373 mp
->mnt_vstat
.f_bsize
= HAMMER2_PBUFSIZE
;
1378 mp
->mnt_iosize_max
= MAXPHYS
;
1381 * Connect up mount pointers.
1383 hammer2_mount_helper(mp
, pmp
);
1385 lockmgr(&hammer2_mntlk
, LK_RELEASE
);
1391 vfs_add_vnodeops(mp
, &hammer2_vnode_vops
, &mp
->mnt_vn_norm_ops
);
1392 vfs_add_vnodeops(mp
, &hammer2_spec_vops
, &mp
->mnt_vn_spec_ops
);
1393 vfs_add_vnodeops(mp
, &hammer2_fifo_vops
, &mp
->mnt_vn_fifo_ops
);
1396 copyinstr(info
.volume
, mp
->mnt_stat
.f_mntfromname
,
1397 MNAMELEN
- 1, &size
);
1398 bzero(mp
->mnt_stat
.f_mntfromname
+ size
, MNAMELEN
- size
);
1399 } /* else root mount, already in there */
1401 bzero(mp
->mnt_stat
.f_mntonname
, sizeof(mp
->mnt_stat
.f_mntonname
));
1403 copyinstr(path
, mp
->mnt_stat
.f_mntonname
,
1404 sizeof(mp
->mnt_stat
.f_mntonname
) - 1,
1408 mp
->mnt_stat
.f_mntonname
[0] = '/';
1412 * Initial statfs to prime mnt_stat.
1414 hammer2_vfs_statfs(mp
, &mp
->mnt_stat
, cred
);
1420 * Scan PFSs under the super-root and create hammer2_pfs structures.
1424 hammer2_update_pmps(hammer2_dev_t
*hmp
)
1426 const hammer2_inode_data_t
*ripdata
;
1427 hammer2_chain_t
*parent
;
1428 hammer2_chain_t
*chain
;
1429 hammer2_blockref_t bref
;
1430 hammer2_dev_t
*force_local
;
1431 hammer2_pfs_t
*spmp
;
1433 hammer2_key_t key_next
;
1434 int cache_index
= -1;
1437 * Force local mount (disassociate all PFSs from their clusters).
1438 * Used primarily for debugging.
1440 force_local
= (hmp
->hflags
& HMNT2_LOCAL
) ? hmp
: NULL
;
1443 * Lookup mount point under the media-localized super-root.
1445 * cluster->pmp will incorrectly point to spmp and must be fixed
1449 hammer2_inode_lock(spmp
->iroot
, 0);
1450 parent
= hammer2_inode_chain(spmp
->iroot
, 0, HAMMER2_RESOLVE_ALWAYS
);
1451 chain
= hammer2_chain_lookup(&parent
, &key_next
,
1452 HAMMER2_KEY_MIN
, HAMMER2_KEY_MAX
,
1455 if (chain
->bref
.type
!= HAMMER2_BREF_TYPE_INODE
)
1457 ripdata
= &chain
->data
->ipdata
;
1459 kprintf("ADD LOCAL PFS: %s\n", ripdata
->filename
);
1461 pmp
= hammer2_pfsalloc(chain
, ripdata
,
1462 bref
.modify_tid
, force_local
);
1463 chain
= hammer2_chain_next(&parent
, chain
, &key_next
,
1464 key_next
, HAMMER2_KEY_MAX
,
1468 hammer2_chain_unlock(parent
);
1469 hammer2_chain_drop(parent
);
1471 hammer2_inode_unlock(spmp
->iroot
);
1476 hammer2_remount(hammer2_dev_t
*hmp
, struct mount
*mp
, char *path __unused
,
1477 struct vnode
*devvp
, struct ucred
*cred
)
1481 if (hmp
->ronly
&& (mp
->mnt_kern_flag
& MNTK_WANTRDWR
)) {
1482 vn_lock(devvp
, LK_EXCLUSIVE
| LK_RETRY
);
1483 VOP_OPEN(devvp
, FREAD
| FWRITE
, FSCRED
, NULL
);
1485 error
= hammer2_recovery(hmp
);
1486 vn_lock(devvp
, LK_EXCLUSIVE
| LK_RETRY
);
1488 VOP_CLOSE(devvp
, FREAD
, NULL
);
1491 VOP_CLOSE(devvp
, FREAD
| FWRITE
, NULL
);
1502 hammer2_vfs_unmount(struct mount
*mp
, int mntflags
)
1513 lockmgr(&hammer2_mntlk
, LK_EXCLUSIVE
);
1516 * If mount initialization proceeded far enough we must flush
1517 * its vnodes and sync the underlying mount points. Three syncs
1518 * are required to fully flush the filesystem (freemap updates lag
1519 * by one flush, and one extra for safety).
1521 if (mntflags
& MNT_FORCE
)
1526 error
= vflush(mp
, 0, flags
);
1529 hammer2_vfs_sync(mp
, MNT_WAIT
);
1530 hammer2_vfs_sync(mp
, MNT_WAIT
);
1531 hammer2_vfs_sync(mp
, MNT_WAIT
);
1535 * Cleanup the frontend support XOPS threads
1537 hammer2_xop_helper_cleanup(pmp
);
1540 hammer2_unmount_helper(mp
, pmp
, NULL
);
1544 lockmgr(&hammer2_mntlk
, LK_RELEASE
);
1550 * Mount helper, hook the system mount into our PFS.
1551 * The mount lock is held.
1553 * We must bump the mount_count on related devices for any
1558 hammer2_mount_helper(struct mount
*mp
, hammer2_pfs_t
*pmp
)
1560 hammer2_cluster_t
*cluster
;
1561 hammer2_chain_t
*rchain
;
1564 mp
->mnt_data
= (qaddr_t
)pmp
;
1568 * After pmp->mp is set we have to adjust hmp->mount_count.
1570 cluster
= &pmp
->iroot
->cluster
;
1571 for (i
= 0; i
< cluster
->nchains
; ++i
) {
1572 rchain
= cluster
->array
[i
].chain
;
1575 ++rchain
->hmp
->mount_count
;
1576 kprintf("hammer2_mount hmp=%p ++mount_count=%d\n",
1577 rchain
->hmp
, rchain
->hmp
->mount_count
);
1581 * Create missing Xop threads
1583 hammer2_xop_helper_create(pmp
);
1587 * Mount helper, unhook the system mount from our PFS.
1588 * The mount lock is held.
1590 * If hmp is supplied a mount responsible for being the first to open
1591 * the block device failed and the block device and all PFSs using the
1592 * block device must be cleaned up.
1594 * If pmp is supplied multiple devices might be backing the PFS and each
1595 * must be disconnected. This might not be the last PFS using some of the
1596 * underlying devices. Also, we have to adjust our hmp->mount_count
1597 * accounting for the devices backing the pmp which is now undergoing an
1602 hammer2_unmount_helper(struct mount
*mp
, hammer2_pfs_t
*pmp
, hammer2_dev_t
*hmp
)
1604 hammer2_cluster_t
*cluster
;
1605 hammer2_chain_t
*rchain
;
1606 struct vnode
*devvp
;
1612 * If no device supplied this is a high-level unmount and we have to
1613 * to disconnect the mount, adjust mount_count, and locate devices
1614 * that might now have no mounts.
1617 KKASSERT(hmp
== NULL
);
1618 KKASSERT((void *)(intptr_t)mp
->mnt_data
== pmp
);
1620 mp
->mnt_data
= NULL
;
1623 * After pmp->mp is cleared we have to account for
1626 cluster
= &pmp
->iroot
->cluster
;
1627 for (i
= 0; i
< cluster
->nchains
; ++i
) {
1628 rchain
= cluster
->array
[i
].chain
;
1631 --rchain
->hmp
->mount_count
;
1632 kprintf("hammer2_unmount hmp=%p --mount_count=%d\n",
1633 rchain
->hmp
, rchain
->hmp
->mount_count
);
1634 /* scrapping hmp now may invalidate the pmp */
1637 TAILQ_FOREACH(hmp
, &hammer2_mntlist
, mntentry
) {
1638 if (hmp
->mount_count
== 0) {
1639 hammer2_unmount_helper(NULL
, NULL
, hmp
);
1647 * Try to terminate the block device. We can't terminate it if
1648 * there are still PFSs referencing it.
1650 kprintf("hammer2_unmount hmp=%p mount_count=%d\n",
1651 hmp
, hmp
->mount_count
);
1652 if (hmp
->mount_count
)
1655 hammer2_bulkfree_uninit(hmp
);
1656 hammer2_pfsfree_scan(hmp
);
1657 hammer2_dev_exlock(hmp
); /* XXX order */
1660 * Cycle the volume data lock as a safety (probably not needed any
1661 * more). To ensure everything is out we need to flush at least
1662 * three times. (1) The running of the sideq can dirty the
1663 * filesystem, (2) A normal flush can dirty the freemap, and
1664 * (3) ensure that the freemap is fully synchronized.
1666 * The next mount's recovery scan can clean everything up but we want
1667 * to leave the filesystem in a 100% clean state on a normal unmount.
1670 hammer2_voldata_lock(hmp
);
1671 hammer2_voldata_unlock(hmp
);
1673 hammer2_iocom_uninit(hmp
);
1675 if ((hmp
->vchain
.flags
| hmp
->fchain
.flags
) &
1676 HAMMER2_CHAIN_FLUSH_MASK
) {
1677 kprintf("hammer2_unmount: chains left over "
1678 "after final sync\n");
1679 kprintf(" vchain %08x\n", hmp
->vchain
.flags
);
1680 kprintf(" fchain %08x\n", hmp
->fchain
.flags
);
1682 if (hammer2_debug
& 0x0010)
1683 Debugger("entered debugger");
1686 KKASSERT(hmp
->spmp
== NULL
);
1689 * Finish up with the device vnode
1691 if ((devvp
= hmp
->devvp
) != NULL
) {
1693 vn_lock(devvp
, LK_EXCLUSIVE
| LK_RETRY
);
1694 kprintf("hammer2_unmount(A): devvp %s rbdirty %p ronly=%d\n",
1695 hmp
->devrepname
, RB_ROOT(&devvp
->v_rbdirty_tree
),
1697 vinvalbuf(devvp
, (ronly
? 0 : V_SAVE
), 0, 0);
1698 kprintf("hammer2_unmount(B): devvp %s rbdirty %p\n",
1699 hmp
->devrepname
, RB_ROOT(&devvp
->v_rbdirty_tree
));
1701 VOP_CLOSE(devvp
, (ronly
? FREAD
: FREAD
|FWRITE
), NULL
);
1708 * Clear vchain/fchain flags that might prevent final cleanup
1711 if (hmp
->vchain
.flags
& HAMMER2_CHAIN_MODIFIED
) {
1712 atomic_add_long(&hammer2_count_modified_chains
, -1);
1713 atomic_clear_int(&hmp
->vchain
.flags
, HAMMER2_CHAIN_MODIFIED
);
1714 hammer2_pfs_memory_wakeup(hmp
->vchain
.pmp
);
1716 if (hmp
->vchain
.flags
& HAMMER2_CHAIN_UPDATE
) {
1717 atomic_clear_int(&hmp
->vchain
.flags
, HAMMER2_CHAIN_UPDATE
);
1720 if (hmp
->fchain
.flags
& HAMMER2_CHAIN_MODIFIED
) {
1721 atomic_add_long(&hammer2_count_modified_chains
, -1);
1722 atomic_clear_int(&hmp
->fchain
.flags
, HAMMER2_CHAIN_MODIFIED
);
1723 hammer2_pfs_memory_wakeup(hmp
->fchain
.pmp
);
1725 if (hmp
->fchain
.flags
& HAMMER2_CHAIN_UPDATE
) {
1726 atomic_clear_int(&hmp
->fchain
.flags
, HAMMER2_CHAIN_UPDATE
);
1730 * Final drop of embedded freemap root chain to
1731 * clean up fchain.core (fchain structure is not
1732 * flagged ALLOCATED so it is cleaned out and then
1735 hammer2_chain_drop(&hmp
->fchain
);
1738 * Final drop of embedded volume root chain to clean
1739 * up vchain.core (vchain structure is not flagged
1740 * ALLOCATED so it is cleaned out and then left to
1744 hammer2_dump_chain(&hmp
->vchain
, 0, &dumpcnt
, 'v');
1746 hammer2_dump_chain(&hmp
->fchain
, 0, &dumpcnt
, 'f');
1747 hammer2_dev_unlock(hmp
);
1748 hammer2_chain_drop(&hmp
->vchain
);
1750 hammer2_io_cleanup(hmp
, &hmp
->iotree
);
1751 if (hmp
->iofree_count
) {
1752 kprintf("io_cleanup: %d I/O's left hanging\n",
1756 TAILQ_REMOVE(&hammer2_mntlist
, hmp
, mntentry
);
1757 kmalloc_destroy(&hmp
->mchain
);
1758 kfree(hmp
, M_HAMMER2
);
1762 hammer2_vfs_vget(struct mount
*mp
, struct vnode
*dvp
,
1763 ino_t ino
, struct vnode
**vpp
)
1765 hammer2_xop_lookup_t
*xop
;
1767 hammer2_inode_t
*ip
;
1771 inum
= (hammer2_tid_t
)ino
& HAMMER2_DIRHASH_USERMSK
;
1777 * Easy if we already have it cached
1779 ip
= hammer2_inode_lookup(pmp
, inum
);
1781 hammer2_inode_lock(ip
, HAMMER2_RESOLVE_SHARED
);
1782 *vpp
= hammer2_igetv(ip
, &error
);
1783 hammer2_inode_unlock(ip
);
1784 hammer2_inode_drop(ip
); /* from lookup */
1790 * Otherwise we have to find the inode
1792 xop
= hammer2_xop_alloc(pmp
->iroot
, 0);
1794 hammer2_xop_start(&xop
->head
, hammer2_xop_lookup
);
1795 error
= hammer2_xop_collect(&xop
->head
, 0);
1798 if (hammer2_cluster_rdata(&xop
->head
.cluster
) == NULL
) {
1799 kprintf("vget: no collect error but also no rdata\n");
1800 kprintf("xop %p\n", xop
);
1801 while ((hammer2_debug
& 0x80000) == 0) {
1802 tsleep(xop
, PCATCH
, "wait", hz
* 10);
1806 ip
= hammer2_inode_get(pmp
, NULL
, &xop
->head
.cluster
, -1);
1809 hammer2_xop_retire(&xop
->head
, HAMMER2_XOPMASK_VOP
);
1812 *vpp
= hammer2_igetv(ip
, &error
);
1813 hammer2_inode_unlock(ip
);
1823 hammer2_vfs_root(struct mount
*mp
, struct vnode
**vpp
)
1830 if (pmp
->iroot
== NULL
) {
1836 hammer2_inode_lock(pmp
->iroot
, HAMMER2_RESOLVE_SHARED
);
1838 while (pmp
->inode_tid
== 0) {
1839 hammer2_xop_ipcluster_t
*xop
;
1840 hammer2_inode_meta_t
*meta
;
1842 xop
= hammer2_xop_alloc(pmp
->iroot
, HAMMER2_XOP_MODIFYING
);
1843 hammer2_xop_start(&xop
->head
, hammer2_xop_ipcluster
);
1844 error
= hammer2_xop_collect(&xop
->head
, 0);
1847 meta
= &xop
->head
.cluster
.focus
->data
->ipdata
.meta
;
1848 pmp
->iroot
->meta
= *meta
;
1849 pmp
->inode_tid
= meta
->pfs_inum
+ 1;
1850 if (pmp
->inode_tid
< HAMMER2_INODE_START
)
1851 pmp
->inode_tid
= HAMMER2_INODE_START
;
1853 xop
->head
.cluster
.focus
->bref
.modify_tid
+ 1;
1854 kprintf("PFS: Starting inode %jd\n",
1855 (intmax_t)pmp
->inode_tid
);
1856 kprintf("PMP focus good set nextino=%ld mod=%016jx\n",
1857 pmp
->inode_tid
, pmp
->modify_tid
);
1858 wakeup(&pmp
->iroot
);
1860 hammer2_xop_retire(&xop
->head
, HAMMER2_XOPMASK_VOP
);
1863 * Prime the mount info.
1865 hammer2_vfs_statfs(mp
, &mp
->mnt_stat
, NULL
);
1872 hammer2_xop_retire(&xop
->head
, HAMMER2_XOPMASK_VOP
);
1873 hammer2_inode_unlock(pmp
->iroot
);
1874 error
= tsleep(&pmp
->iroot
, PCATCH
, "h2root", hz
);
1875 hammer2_inode_lock(pmp
->iroot
, HAMMER2_RESOLVE_SHARED
);
1881 hammer2_inode_unlock(pmp
->iroot
);
1884 vp
= hammer2_igetv(pmp
->iroot
, &error
);
1885 hammer2_inode_unlock(pmp
->iroot
);
1895 * XXX incorporate ipdata->meta.inode_quota and data_quota
1899 hammer2_vfs_statfs(struct mount
*mp
, struct statfs
*sbp
, struct ucred
*cred
)
1903 hammer2_blockref_t bref
;
1907 * NOTE: iroot might not have validated the cluster yet.
1911 mp
->mnt_stat
.f_files
= 0;
1912 mp
->mnt_stat
.f_ffree
= 0;
1913 mp
->mnt_stat
.f_blocks
= 0;
1914 mp
->mnt_stat
.f_bfree
= 0;
1915 mp
->mnt_stat
.f_bavail
= 0;
1917 for (i
= 0; i
< pmp
->iroot
->cluster
.nchains
; ++i
) {
1918 hmp
= pmp
->pfs_hmps
[i
];
1921 if (pmp
->iroot
->cluster
.array
[i
].chain
)
1922 bref
= pmp
->iroot
->cluster
.array
[i
].chain
->bref
;
1924 bzero(&bref
, sizeof(bref
));
1926 mp
->mnt_stat
.f_files
= bref
.embed
.stats
.inode_count
;
1927 mp
->mnt_stat
.f_ffree
= 0;
1928 mp
->mnt_stat
.f_blocks
= hmp
->voldata
.allocator_size
/
1929 mp
->mnt_vstat
.f_bsize
;
1930 mp
->mnt_stat
.f_bfree
= hmp
->voldata
.allocator_free
/
1931 mp
->mnt_vstat
.f_bsize
;
1932 mp
->mnt_stat
.f_bavail
= mp
->mnt_stat
.f_bfree
;
1934 *sbp
= mp
->mnt_stat
;
1941 hammer2_vfs_statvfs(struct mount
*mp
, struct statvfs
*sbp
, struct ucred
*cred
)
1945 hammer2_blockref_t bref
;
1949 * NOTE: iroot might not have validated the cluster yet.
1953 mp
->mnt_vstat
.f_bsize
= 0;
1954 mp
->mnt_vstat
.f_files
= 0;
1955 mp
->mnt_vstat
.f_ffree
= 0;
1956 mp
->mnt_vstat
.f_blocks
= 0;
1957 mp
->mnt_vstat
.f_bfree
= 0;
1958 mp
->mnt_vstat
.f_bavail
= 0;
1960 for (i
= 0; i
< pmp
->iroot
->cluster
.nchains
; ++i
) {
1961 hmp
= pmp
->pfs_hmps
[i
];
1964 if (pmp
->iroot
->cluster
.array
[i
].chain
)
1965 bref
= pmp
->iroot
->cluster
.array
[i
].chain
->bref
;
1967 bzero(&bref
, sizeof(bref
));
1969 mp
->mnt_vstat
.f_bsize
= HAMMER2_PBUFSIZE
;
1970 mp
->mnt_vstat
.f_files
= bref
.embed
.stats
.inode_count
;
1971 mp
->mnt_vstat
.f_ffree
= 0;
1972 mp
->mnt_vstat
.f_blocks
= hmp
->voldata
.allocator_size
/
1973 mp
->mnt_vstat
.f_bsize
;
1974 mp
->mnt_vstat
.f_bfree
= hmp
->voldata
.allocator_free
/
1975 mp
->mnt_vstat
.f_bsize
;
1976 mp
->mnt_vstat
.f_bavail
= mp
->mnt_vstat
.f_bfree
;
1978 *sbp
= mp
->mnt_vstat
;
1984 * Mount-time recovery (RW mounts)
1986 * Updates to the free block table are allowed to lag flushes by one
1987 * transaction. In case of a crash, then on a fresh mount we must do an
1988 * incremental scan of the last committed transaction id and make sure that
1989 * all related blocks have been marked allocated.
1991 * The super-root topology and each PFS has its own transaction id domain,
1992 * so we must track PFS boundary transitions.
1994 struct hammer2_recovery_elm
{
1995 TAILQ_ENTRY(hammer2_recovery_elm
) entry
;
1996 hammer2_chain_t
*chain
;
1997 hammer2_tid_t sync_tid
;
2000 TAILQ_HEAD(hammer2_recovery_list
, hammer2_recovery_elm
);
2002 struct hammer2_recovery_info
{
2003 struct hammer2_recovery_list list
;
2008 static int hammer2_recovery_scan(hammer2_dev_t
*hmp
,
2009 hammer2_chain_t
*parent
,
2010 struct hammer2_recovery_info
*info
,
2011 hammer2_tid_t sync_tid
);
2013 #define HAMMER2_RECOVERY_MAXDEPTH 10
2017 hammer2_recovery(hammer2_dev_t
*hmp
)
2019 struct hammer2_recovery_info info
;
2020 struct hammer2_recovery_elm
*elm
;
2021 hammer2_chain_t
*parent
;
2022 hammer2_tid_t sync_tid
;
2023 hammer2_tid_t mirror_tid
;
2026 hammer2_trans_init(hmp
->spmp
, 0);
2028 sync_tid
= hmp
->voldata
.freemap_tid
;
2029 mirror_tid
= hmp
->voldata
.mirror_tid
;
2031 kprintf("hammer2 mount \"%s\": ", hmp
->devrepname
);
2032 if (sync_tid
>= mirror_tid
) {
2033 kprintf(" no recovery needed\n");
2035 kprintf(" freemap recovery %016jx-%016jx\n",
2036 sync_tid
+ 1, mirror_tid
);
2039 TAILQ_INIT(&info
.list
);
2041 parent
= hammer2_chain_lookup_init(&hmp
->vchain
, 0);
2042 error
= hammer2_recovery_scan(hmp
, parent
, &info
, sync_tid
);
2043 hammer2_chain_lookup_done(parent
);
2045 while ((elm
= TAILQ_FIRST(&info
.list
)) != NULL
) {
2046 TAILQ_REMOVE(&info
.list
, elm
, entry
);
2047 parent
= elm
->chain
;
2048 sync_tid
= elm
->sync_tid
;
2049 kfree(elm
, M_HAMMER2
);
2051 hammer2_chain_lock(parent
, HAMMER2_RESOLVE_ALWAYS
);
2052 error
|= hammer2_recovery_scan(hmp
, parent
, &info
,
2053 hmp
->voldata
.freemap_tid
);
2054 hammer2_chain_unlock(parent
);
2055 hammer2_chain_drop(parent
); /* drop elm->chain ref */
2057 hammer2_trans_done(hmp
->spmp
);
2064 hammer2_recovery_scan(hammer2_dev_t
*hmp
, hammer2_chain_t
*parent
,
2065 struct hammer2_recovery_info
*info
,
2066 hammer2_tid_t sync_tid
)
2068 const hammer2_inode_data_t
*ripdata
;
2069 hammer2_chain_t
*chain
;
2070 hammer2_blockref_t bref
;
2078 * Adjust freemap to ensure that the block(s) are marked allocated.
2080 if (parent
->bref
.type
!= HAMMER2_BREF_TYPE_VOLUME
) {
2081 hammer2_freemap_adjust(hmp
, &parent
->bref
,
2082 HAMMER2_FREEMAP_DORECOVER
);
2086 * Check type for recursive scan
2088 switch(parent
->bref
.type
) {
2089 case HAMMER2_BREF_TYPE_VOLUME
:
2090 /* data already instantiated */
2092 case HAMMER2_BREF_TYPE_INODE
:
2094 * Must instantiate data for DIRECTDATA test and also
2097 hammer2_chain_lock(parent
, HAMMER2_RESOLVE_ALWAYS
);
2098 ripdata
= &hammer2_chain_rdata(parent
)->ipdata
;
2099 if (ripdata
->meta
.op_flags
& HAMMER2_OPFLAG_DIRECTDATA
) {
2100 /* not applicable to recovery scan */
2101 hammer2_chain_unlock(parent
);
2104 hammer2_chain_unlock(parent
);
2106 case HAMMER2_BREF_TYPE_INDIRECT
:
2108 * Must instantiate data for recursion
2110 hammer2_chain_lock(parent
, HAMMER2_RESOLVE_ALWAYS
);
2111 hammer2_chain_unlock(parent
);
2113 case HAMMER2_BREF_TYPE_DIRENT
:
2114 case HAMMER2_BREF_TYPE_DATA
:
2115 case HAMMER2_BREF_TYPE_FREEMAP
:
2116 case HAMMER2_BREF_TYPE_FREEMAP_NODE
:
2117 case HAMMER2_BREF_TYPE_FREEMAP_LEAF
:
2118 /* not applicable to recovery scan */
2122 return HAMMER2_ERROR_BADBREF
;
2126 * Defer operation if depth limit reached or if we are crossing a
2129 if (info
->depth
>= HAMMER2_RECOVERY_MAXDEPTH
) {
2130 struct hammer2_recovery_elm
*elm
;
2132 elm
= kmalloc(sizeof(*elm
), M_HAMMER2
, M_ZERO
| M_WAITOK
);
2133 elm
->chain
= parent
;
2134 elm
->sync_tid
= sync_tid
;
2135 hammer2_chain_ref(parent
);
2136 TAILQ_INSERT_TAIL(&info
->list
, elm
, entry
);
2137 /* unlocked by caller */
2144 * Recursive scan of the last flushed transaction only. We are
2145 * doing this without pmp assignments so don't leave the chains
2146 * hanging around after we are done with them.
2148 * error Cumulative error this level only
2149 * rup_error Cumulative error for recursion
2150 * tmp_error Specific non-cumulative recursion error
2159 error
|= hammer2_chain_scan(parent
, &chain
, &bref
,
2160 &first
, &cache_index
,
2161 HAMMER2_LOOKUP_NODATA
);
2164 * Problem during scan or EOF
2172 if (chain
== NULL
) {
2173 if (bref
.mirror_tid
> sync_tid
) {
2174 hammer2_freemap_adjust(hmp
, &bref
,
2175 HAMMER2_FREEMAP_DORECOVER
);
2181 * This may or may not be a recursive node.
2183 atomic_set_int(&chain
->flags
, HAMMER2_CHAIN_RELEASE
);
2184 if (bref
.mirror_tid
> sync_tid
) {
2186 tmp_error
= hammer2_recovery_scan(hmp
, chain
,
2194 * Flush the recovery at the PFS boundary to stage it for
2195 * the final flush of the super-root topology.
2197 if (tmp_error
== 0 &&
2198 (bref
.flags
& HAMMER2_BREF_FLAG_PFSROOT
) &&
2199 (chain
->flags
& HAMMER2_CHAIN_ONFLUSH
)) {
2200 hammer2_flush(chain
, HAMMER2_FLUSH_TOP
);
2202 rup_error
|= tmp_error
;
2204 return ((error
| rup_error
) & ~HAMMER2_ERROR_EOF
);
2208 * Sync a mount point; this is called on a per-mount basis from the
2209 * filesystem syncer process periodically and whenever a user issues
2213 hammer2_vfs_sync(struct mount
*mp
, int waitfor
)
2215 hammer2_xop_flush_t
*xop
;
2216 struct hammer2_sync_info info
;
2217 hammer2_inode_t
*iroot
;
2225 KKASSERT(iroot
->pmp
== pmp
);
2228 * We can't acquire locks on existing vnodes while in a transaction
2229 * without risking a deadlock. This assumes that vfsync() can be
2230 * called without the vnode locked (which it can in DragonFly).
2231 * Otherwise we'd have to implement a multi-pass or flag the lock
2232 * failures and retry.
2234 * The reclamation code interlocks with the sync list's token
2235 * (by removing the vnode from the scan list) before unlocking
2236 * the inode, giving us time to ref the inode.
2238 /*flags = VMSC_GETVP;*/
2240 if (waitfor
& MNT_LAZY
)
2241 flags
|= VMSC_ONEPASS
;
2244 * Preflush the vnodes using a normal transaction before interlocking
2245 * with a flush transaction. We do this to try to run as much of
2246 * the compression as possible outside the flush transaction.
2248 * For efficiency do an async pass before making sure with a
2249 * synchronous pass on all related buffer cache buffers.
2251 hammer2_trans_init(pmp
, 0);
2253 info
.waitfor
= MNT_NOWAIT
;
2254 vsyncscan(mp
, flags
| VMSC_NOWAIT
, hammer2_sync_scan2
, &info
);
2255 info
.waitfor
= MNT_WAIT
;
2256 vsyncscan(mp
, flags
, hammer2_sync_scan2
, &info
);
2257 hammer2_trans_done(pmp
);
2260 * Start our flush transaction. This does not return until all
2261 * concurrent transactions have completed and will prevent any
2262 * new transactions from running concurrently, except for the
2263 * buffer cache transactions.
2265 * NOTE! It is still possible for the paging code to push pages
2266 * out via a UIO_NOCOPY hammer2_vop_write() during the main
2269 hammer2_trans_init(pmp
, HAMMER2_TRANS_ISFLUSH
);
2270 hammer2_inode_run_sideq(pmp
);
2273 info
.waitfor
= MNT_NOWAIT
;
2274 vsyncscan(mp
, flags
| VMSC_NOWAIT
, hammer2_sync_scan2
, &info
);
2275 info
.waitfor
= MNT_WAIT
;
2276 vsyncscan(mp
, flags
, hammer2_sync_scan2
, &info
);
2277 hammer2_bioq_sync(pmp
);
2280 * Use the XOP interface to concurrently flush all nodes to
2281 * synchronize the PFSROOT subtopology to the media. A standard
2282 * end-of-scan ENOENT error indicates cluster sufficiency.
2284 * Note that this flush will not be visible on crash recovery until
2285 * we flush the super-root topology in the next loop.
2287 * XXX For now wait for all flushes to complete.
2290 xop
= hammer2_xop_alloc(iroot
, HAMMER2_XOP_MODIFYING
);
2291 hammer2_xop_start(&xop
->head
, hammer2_inode_xop_flush
);
2292 error
= hammer2_xop_collect(&xop
->head
,
2293 HAMMER2_XOP_COLLECT_WAITALL
);
2294 hammer2_xop_retire(&xop
->head
, HAMMER2_XOPMASK_VOP
);
2295 if (error
== ENOENT
)
2300 hammer2_trans_done(pmp
);
2308 * Note that we ignore the tranasction mtid we got above. Instead,
2309 * each vfsync below will ultimately get its own via TRANS_BUFCACHE
2313 hammer2_sync_scan2(struct mount
*mp
, struct vnode
*vp
, void *data
)
2315 struct hammer2_sync_info
*info
= data
;
2316 hammer2_inode_t
*ip
;
2320 * Degenerate cases. Note that ip == NULL typically means the
2321 * syncer vnode itself and we don't want to vclrisdirty() in that
2328 if (vp
->v_type
== VNON
|| vp
->v_type
== VBAD
) {
2334 * VOP_FSYNC will start a new transaction so replicate some code
2335 * here to do it inline (see hammer2_vop_fsync()).
2337 * WARNING: The vfsync interacts with the buffer cache and might
2338 * block, we can't hold the inode lock at that time.
2339 * However, we MUST ref ip before blocking to ensure that
2340 * it isn't ripped out from under us (since we do not
2341 * hold a lock on the vnode).
2343 hammer2_inode_ref(ip
);
2344 if ((ip
->flags
& HAMMER2_INODE_MODIFIED
) ||
2345 !RB_EMPTY(&vp
->v_rbdirty_tree
)) {
2346 vfsync(vp
, info
->waitfor
, 1, NULL
, NULL
);
2347 if (ip
->flags
& (HAMMER2_INODE_RESIZED
|
2348 HAMMER2_INODE_MODIFIED
)) {
2349 hammer2_inode_lock(ip
, 0);
2350 if (ip
->flags
& (HAMMER2_INODE_RESIZED
|
2351 HAMMER2_INODE_MODIFIED
)) {
2352 hammer2_inode_chain_sync(ip
);
2354 hammer2_inode_unlock(ip
);
2357 if ((ip
->flags
& HAMMER2_INODE_MODIFIED
) == 0 &&
2358 RB_EMPTY(&vp
->v_rbdirty_tree
)) {
2362 hammer2_inode_drop(ip
);
2366 info
->error
= error
;
2373 hammer2_vfs_vptofh(struct vnode
*vp
, struct fid
*fhp
)
2375 hammer2_inode_t
*ip
;
2377 KKASSERT(MAXFIDSZ
>= 16);
2379 fhp
->fid_len
= offsetof(struct fid
, fid_data
[16]);
2381 ((hammer2_tid_t
*)fhp
->fid_data
)[0] = ip
->meta
.inum
;
2382 ((hammer2_tid_t
*)fhp
->fid_data
)[1] = 0;
2389 hammer2_vfs_fhtovp(struct mount
*mp
, struct vnode
*rootvp
,
2390 struct fid
*fhp
, struct vnode
**vpp
)
2397 inum
= ((hammer2_tid_t
*)fhp
->fid_data
)[0] & HAMMER2_DIRHASH_USERMSK
;
2400 error
= hammer2_vfs_root(mp
, vpp
);
2402 error
= hammer2_vfs_vget(mp
, NULL
, inum
, vpp
);
2407 kprintf("fhtovp: %016jx -> %p, %d\n", inum
, *vpp
, error
);
2413 hammer2_vfs_checkexp(struct mount
*mp
, struct sockaddr
*nam
,
2414 int *exflagsp
, struct ucred
**credanonp
)
2421 np
= vfs_export_lookup(mp
, &pmp
->export
, nam
);
2423 *exflagsp
= np
->netc_exflags
;
2424 *credanonp
= &np
->netc_anon
;
2433 * Support code for hammer2_vfs_mount(). Read, verify, and install the volume
2434 * header into the HMP
2436 * XXX read four volhdrs and use the one with the highest TID whos CRC
2441 * XXX For filesystems w/ less than 4 volhdrs, make sure to not write to
2442 * nonexistant locations.
2444 * XXX Record selected volhdr and ring updates to each of 4 volhdrs
2448 hammer2_install_volume_header(hammer2_dev_t
*hmp
)
2450 hammer2_volume_data_t
*vd
;
2452 hammer2_crc32_t crc0
, crc
, bcrc0
, bcrc
;
2464 * There are up to 4 copies of the volume header (syncs iterate
2465 * between them so there is no single master). We don't trust the
2466 * volu_size field so we don't know precisely how large the filesystem
2467 * is, so depend on the OS to return an error if we go beyond the
2468 * block device's EOF.
2470 for (i
= 0; i
< HAMMER2_NUM_VOLHDRS
; i
++) {
2471 error
= bread(hmp
->devvp
, i
* HAMMER2_ZONE_BYTES64
,
2472 HAMMER2_VOLUME_BYTES
, &bp
);
2479 vd
= (struct hammer2_volume_data
*) bp
->b_data
;
2480 if ((vd
->magic
!= HAMMER2_VOLUME_ID_HBO
) &&
2481 (vd
->magic
!= HAMMER2_VOLUME_ID_ABO
)) {
2487 if (vd
->magic
== HAMMER2_VOLUME_ID_ABO
) {
2488 /* XXX: Reversed-endianness filesystem */
2489 kprintf("hammer2: reverse-endian filesystem detected");
2495 crc
= vd
->icrc_sects
[HAMMER2_VOL_ICRC_SECT0
];
2496 crc0
= hammer2_icrc32(bp
->b_data
+ HAMMER2_VOLUME_ICRC0_OFF
,
2497 HAMMER2_VOLUME_ICRC0_SIZE
);
2498 bcrc
= vd
->icrc_sects
[HAMMER2_VOL_ICRC_SECT1
];
2499 bcrc0
= hammer2_icrc32(bp
->b_data
+ HAMMER2_VOLUME_ICRC1_OFF
,
2500 HAMMER2_VOLUME_ICRC1_SIZE
);
2501 if ((crc0
!= crc
) || (bcrc0
!= bcrc
)) {
2502 kprintf("hammer2 volume header crc "
2503 "mismatch copy #%d %08x/%08x\n",
2510 if (valid
== 0 || hmp
->voldata
.mirror_tid
< vd
->mirror_tid
) {
2519 hmp
->volsync
= hmp
->voldata
;
2521 if (error_reported
|| bootverbose
|| 1) { /* 1/DEBUG */
2522 kprintf("hammer2: using volume header #%d\n",
2527 kprintf("hammer2: no valid volume headers found!\n");
2533 * This handles hysteresis on regular file flushes. Because the BIOs are
2534 * routed to a thread it is possible for an excessive number to build up
2535 * and cause long front-end stalls long before the runningbuffspace limit
2536 * is hit, so we implement hammer2_flush_pipe to control the
2539 * This is a particular problem when compression is used.
2542 hammer2_lwinprog_ref(hammer2_pfs_t
*pmp
)
2544 atomic_add_int(&pmp
->count_lwinprog
, 1);
2548 hammer2_lwinprog_drop(hammer2_pfs_t
*pmp
)
2552 lwinprog
= atomic_fetchadd_int(&pmp
->count_lwinprog
, -1);
2553 if ((lwinprog
& HAMMER2_LWINPROG_WAITING
) &&
2554 (lwinprog
& HAMMER2_LWINPROG_MASK
) <= hammer2_flush_pipe
* 2 / 3) {
2555 atomic_clear_int(&pmp
->count_lwinprog
,
2556 HAMMER2_LWINPROG_WAITING
);
2557 wakeup(&pmp
->count_lwinprog
);
2559 if ((lwinprog
& HAMMER2_LWINPROG_WAITING0
) &&
2560 (lwinprog
& HAMMER2_LWINPROG_MASK
) <= 0) {
2561 atomic_clear_int(&pmp
->count_lwinprog
,
2562 HAMMER2_LWINPROG_WAITING0
);
2563 wakeup(&pmp
->count_lwinprog
);
2568 hammer2_lwinprog_wait(hammer2_pfs_t
*pmp
, int flush_pipe
)
2571 int lwflag
= (flush_pipe
) ? HAMMER2_LWINPROG_WAITING
:
2572 HAMMER2_LWINPROG_WAITING0
;
2575 lwinprog
= pmp
->count_lwinprog
;
2577 if ((lwinprog
& HAMMER2_LWINPROG_MASK
) <= flush_pipe
)
2579 tsleep_interlock(&pmp
->count_lwinprog
, 0);
2580 atomic_set_int(&pmp
->count_lwinprog
, lwflag
);
2581 lwinprog
= pmp
->count_lwinprog
;
2582 if ((lwinprog
& HAMMER2_LWINPROG_MASK
) <= flush_pipe
)
2584 tsleep(&pmp
->count_lwinprog
, PINTERLOCKED
, "h2wpipe", hz
);
2589 * Manage excessive memory resource use for chain and related
2593 hammer2_pfs_memory_wait(hammer2_pfs_t
*pmp
)
2603 * Atomic check condition and wait. Also do an early speedup of
2604 * the syncer to try to avoid hitting the wait.
2607 waiting
= pmp
->inmem_dirty_chains
;
2609 count
= waiting
& HAMMER2_DIRTYCHAIN_MASK
;
2611 limit
= pmp
->mp
->mnt_nvnodelistsize
/ 10;
2612 if (limit
< hammer2_limit_dirty_chains
)
2613 limit
= hammer2_limit_dirty_chains
;
2618 if ((int)(ticks
- zzticks
) > hz
) {
2620 kprintf("count %ld %ld\n", count
, limit
);
2625 * Block if there are too many dirty chains present, wait
2626 * for the flush to clean some out.
2628 if (count
> limit
) {
2629 tsleep_interlock(&pmp
->inmem_dirty_chains
, 0);
2630 if (atomic_cmpset_int(&pmp
->inmem_dirty_chains
,
2632 waiting
| HAMMER2_DIRTYCHAIN_WAITING
)) {
2633 speedup_syncer(pmp
->mp
);
2634 tsleep(&pmp
->inmem_dirty_chains
, PINTERLOCKED
,
2637 continue; /* loop on success or fail */
2641 * Try to start an early flush before we are forced to block.
2643 if (count
> limit
* 7 / 10)
2644 speedup_syncer(pmp
->mp
);
2650 hammer2_pfs_memory_inc(hammer2_pfs_t
*pmp
)
2653 atomic_add_int(&pmp
->inmem_dirty_chains
, 1);
2658 hammer2_pfs_memory_wakeup(hammer2_pfs_t
*pmp
)
2666 waiting
= pmp
->inmem_dirty_chains
;
2668 if (atomic_cmpset_int(&pmp
->inmem_dirty_chains
,
2671 ~HAMMER2_DIRTYCHAIN_WAITING
)) {
2676 if (waiting
& HAMMER2_DIRTYCHAIN_WAITING
)
2677 wakeup(&pmp
->inmem_dirty_chains
);
2684 hammer2_dump_chain(hammer2_chain_t
*chain
, int tab
, int *countp
, char pfx
)
2686 hammer2_chain_t
*scan
;
2687 hammer2_chain_t
*parent
;
2691 kprintf("%*.*s...\n", tab
, tab
, "");
2696 kprintf("%*.*s%c-chain %p.%d %016jx/%d mir=%016jx\n",
2698 chain
, chain
->bref
.type
,
2699 chain
->bref
.key
, chain
->bref
.keybits
,
2700 chain
->bref
.mirror_tid
);
2702 kprintf("%*.*s [%08x] (%s) refs=%d",
2705 ((chain
->bref
.type
== HAMMER2_BREF_TYPE_INODE
&&
2706 chain
->data
) ? (char *)chain
->data
->ipdata
.filename
: "?"),
2709 parent
= chain
->parent
;
2711 kprintf("\n%*.*s p=%p [pflags %08x prefs %d",
2713 parent
, parent
->flags
, parent
->refs
);
2714 if (RB_EMPTY(&chain
->core
.rbtree
)) {
2718 RB_FOREACH(scan
, hammer2_chain_tree
, &chain
->core
.rbtree
)
2719 hammer2_dump_chain(scan
, tab
+ 4, countp
, 'a');
2720 if (chain
->bref
.type
== HAMMER2_BREF_TYPE_INODE
&& chain
->data
)
2721 kprintf("%*.*s}(%s)\n", tab
, tab
, "",
2722 chain
->data
->ipdata
.filename
);
2724 kprintf("%*.*s}\n", tab
, tab
, "");