2 * Copyright (c) 2011-2017 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
;
80 int hammer2_supported_version
= HAMMER2_VOL_VERSION_DEFAULT
;
82 int hammer2_cluster_meta_read
= 1; /* physical read-ahead */
83 int hammer2_cluster_data_read
= 4; /* physical read-ahead */
84 int hammer2_dedup_enable
= 1;
85 int hammer2_always_compress
= 0; /* always try to compress */
86 int hammer2_inval_enable
= 0;
87 int hammer2_flush_pipe
= 100;
88 int hammer2_synchronous_flush
= 1;
89 int hammer2_dio_count
;
90 int hammer2_limit_dio
= 256;
91 int hammer2_bulkfree_tps
= 5000;
92 long hammer2_chain_allocs
;
93 long hammer2_chain_frees
;
94 long hammer2_limit_dirty_chains
;
95 long hammer2_count_modified_chains
;
96 long hammer2_iod_invals
;
97 long hammer2_iod_file_read
;
98 long hammer2_iod_meta_read
;
99 long hammer2_iod_indr_read
;
100 long hammer2_iod_fmap_read
;
101 long hammer2_iod_volu_read
;
102 long hammer2_iod_file_write
;
103 long hammer2_iod_file_wembed
;
104 long hammer2_iod_file_wzero
;
105 long hammer2_iod_file_wdedup
;
106 long hammer2_iod_meta_write
;
107 long hammer2_iod_indr_write
;
108 long hammer2_iod_fmap_write
;
109 long hammer2_iod_volu_write
;
111 MALLOC_DECLARE(M_HAMMER2_CBUFFER
);
112 MALLOC_DEFINE(M_HAMMER2_CBUFFER
, "HAMMER2-compbuffer",
113 "Buffer used for compression.");
115 MALLOC_DECLARE(M_HAMMER2_DEBUFFER
);
116 MALLOC_DEFINE(M_HAMMER2_DEBUFFER
, "HAMMER2-decompbuffer",
117 "Buffer used for decompression.");
119 SYSCTL_NODE(_vfs
, OID_AUTO
, hammer2
, CTLFLAG_RW
, 0, "HAMMER2 filesystem");
121 SYSCTL_INT(_vfs_hammer2
, OID_AUTO
, supported_version
, CTLFLAG_RD
,
122 &hammer2_supported_version
, 0, "");
123 SYSCTL_INT(_vfs_hammer2
, OID_AUTO
, debug
, CTLFLAG_RW
,
124 &hammer2_debug
, 0, "");
125 SYSCTL_INT(_vfs_hammer2
, OID_AUTO
, cluster_meta_read
, CTLFLAG_RW
,
126 &hammer2_cluster_meta_read
, 0, "");
127 SYSCTL_INT(_vfs_hammer2
, OID_AUTO
, cluster_data_read
, CTLFLAG_RW
,
128 &hammer2_cluster_data_read
, 0, "");
129 SYSCTL_INT(_vfs_hammer2
, OID_AUTO
, dedup_enable
, CTLFLAG_RW
,
130 &hammer2_dedup_enable
, 0, "");
131 SYSCTL_INT(_vfs_hammer2
, OID_AUTO
, always_compress
, CTLFLAG_RW
,
132 &hammer2_always_compress
, 0, "");
133 SYSCTL_INT(_vfs_hammer2
, OID_AUTO
, inval_enable
, CTLFLAG_RW
,
134 &hammer2_inval_enable
, 0, "");
135 SYSCTL_INT(_vfs_hammer2
, OID_AUTO
, flush_pipe
, CTLFLAG_RW
,
136 &hammer2_flush_pipe
, 0, "");
137 SYSCTL_INT(_vfs_hammer2
, OID_AUTO
, synchronous_flush
, CTLFLAG_RW
,
138 &hammer2_synchronous_flush
, 0, "");
139 SYSCTL_INT(_vfs_hammer2
, OID_AUTO
, bulkfree_tps
, CTLFLAG_RW
,
140 &hammer2_bulkfree_tps
, 0, "");
141 SYSCTL_LONG(_vfs_hammer2
, OID_AUTO
, chain_allocs
, CTLFLAG_RW
,
142 &hammer2_chain_allocs
, 0, "");
143 SYSCTL_LONG(_vfs_hammer2
, OID_AUTO
, chain_frees
, CTLFLAG_RW
,
144 &hammer2_chain_frees
, 0, "");
145 SYSCTL_LONG(_vfs_hammer2
, OID_AUTO
, limit_dirty_chains
, CTLFLAG_RW
,
146 &hammer2_limit_dirty_chains
, 0, "");
147 SYSCTL_LONG(_vfs_hammer2
, OID_AUTO
, count_modified_chains
, CTLFLAG_RW
,
148 &hammer2_count_modified_chains
, 0, "");
149 SYSCTL_INT(_vfs_hammer2
, OID_AUTO
, dio_count
, CTLFLAG_RD
,
150 &hammer2_dio_count
, 0, "");
151 SYSCTL_INT(_vfs_hammer2
, OID_AUTO
, limit_dio
, CTLFLAG_RW
,
152 &hammer2_limit_dio
, 0, "");
154 SYSCTL_LONG(_vfs_hammer2
, OID_AUTO
, iod_invals
, CTLFLAG_RW
,
155 &hammer2_iod_invals
, 0, "");
156 SYSCTL_LONG(_vfs_hammer2
, OID_AUTO
, iod_file_read
, CTLFLAG_RW
,
157 &hammer2_iod_file_read
, 0, "");
158 SYSCTL_LONG(_vfs_hammer2
, OID_AUTO
, iod_meta_read
, CTLFLAG_RW
,
159 &hammer2_iod_meta_read
, 0, "");
160 SYSCTL_LONG(_vfs_hammer2
, OID_AUTO
, iod_indr_read
, CTLFLAG_RW
,
161 &hammer2_iod_indr_read
, 0, "");
162 SYSCTL_LONG(_vfs_hammer2
, OID_AUTO
, iod_fmap_read
, CTLFLAG_RW
,
163 &hammer2_iod_fmap_read
, 0, "");
164 SYSCTL_LONG(_vfs_hammer2
, OID_AUTO
, iod_volu_read
, CTLFLAG_RW
,
165 &hammer2_iod_volu_read
, 0, "");
167 SYSCTL_LONG(_vfs_hammer2
, OID_AUTO
, iod_file_write
, CTLFLAG_RW
,
168 &hammer2_iod_file_write
, 0, "");
169 SYSCTL_LONG(_vfs_hammer2
, OID_AUTO
, iod_file_wembed
, CTLFLAG_RW
,
170 &hammer2_iod_file_wembed
, 0, "");
171 SYSCTL_LONG(_vfs_hammer2
, OID_AUTO
, iod_file_wzero
, CTLFLAG_RW
,
172 &hammer2_iod_file_wzero
, 0, "");
173 SYSCTL_LONG(_vfs_hammer2
, OID_AUTO
, iod_file_wdedup
, CTLFLAG_RW
,
174 &hammer2_iod_file_wdedup
, 0, "");
175 SYSCTL_LONG(_vfs_hammer2
, OID_AUTO
, iod_meta_write
, CTLFLAG_RW
,
176 &hammer2_iod_meta_write
, 0, "");
177 SYSCTL_LONG(_vfs_hammer2
, OID_AUTO
, iod_indr_write
, CTLFLAG_RW
,
178 &hammer2_iod_indr_write
, 0, "");
179 SYSCTL_LONG(_vfs_hammer2
, OID_AUTO
, iod_fmap_write
, CTLFLAG_RW
,
180 &hammer2_iod_fmap_write
, 0, "");
181 SYSCTL_LONG(_vfs_hammer2
, OID_AUTO
, iod_volu_write
, CTLFLAG_RW
,
182 &hammer2_iod_volu_write
, 0, "");
184 long hammer2_check_icrc32
;
185 long hammer2_check_xxhash64
;
186 SYSCTL_LONG(_vfs_hammer2
, OID_AUTO
, check_icrc32
, CTLFLAG_RW
,
187 &hammer2_check_icrc32
, 0, "");
188 SYSCTL_LONG(_vfs_hammer2
, OID_AUTO
, check_xxhash64
, CTLFLAG_RW
,
189 &hammer2_check_xxhash64
, 0, "");
191 static int hammer2_vfs_init(struct vfsconf
*conf
);
192 static int hammer2_vfs_uninit(struct vfsconf
*vfsp
);
193 static int hammer2_vfs_mount(struct mount
*mp
, char *path
, caddr_t data
,
195 static int hammer2_remount(hammer2_dev_t
*, struct mount
*, char *,
196 struct vnode
*, struct ucred
*);
197 static int hammer2_recovery(hammer2_dev_t
*hmp
);
198 static int hammer2_vfs_unmount(struct mount
*mp
, int mntflags
);
199 static int hammer2_vfs_root(struct mount
*mp
, struct vnode
**vpp
);
200 static int hammer2_vfs_statfs(struct mount
*mp
, struct statfs
*sbp
,
202 static int hammer2_vfs_statvfs(struct mount
*mp
, struct statvfs
*sbp
,
204 static int hammer2_vfs_fhtovp(struct mount
*mp
, struct vnode
*rootvp
,
205 struct fid
*fhp
, struct vnode
**vpp
);
206 static int hammer2_vfs_vptofh(struct vnode
*vp
, struct fid
*fhp
);
207 static int hammer2_vfs_checkexp(struct mount
*mp
, struct sockaddr
*nam
,
208 int *exflagsp
, struct ucred
**credanonp
);
210 static int hammer2_install_volume_header(hammer2_dev_t
*hmp
);
211 static int hammer2_sync_scan2(struct mount
*mp
, struct vnode
*vp
, void *data
);
213 static void hammer2_update_pmps(hammer2_dev_t
*hmp
);
215 static void hammer2_mount_helper(struct mount
*mp
, hammer2_pfs_t
*pmp
);
216 static void hammer2_unmount_helper(struct mount
*mp
, hammer2_pfs_t
*pmp
,
220 * HAMMER2 vfs operations.
222 static struct vfsops hammer2_vfsops
= {
223 .vfs_init
= hammer2_vfs_init
,
224 .vfs_uninit
= hammer2_vfs_uninit
,
225 .vfs_sync
= hammer2_vfs_sync
,
226 .vfs_mount
= hammer2_vfs_mount
,
227 .vfs_unmount
= hammer2_vfs_unmount
,
228 .vfs_root
= hammer2_vfs_root
,
229 .vfs_statfs
= hammer2_vfs_statfs
,
230 .vfs_statvfs
= hammer2_vfs_statvfs
,
231 .vfs_vget
= hammer2_vfs_vget
,
232 .vfs_vptofh
= hammer2_vfs_vptofh
,
233 .vfs_fhtovp
= hammer2_vfs_fhtovp
,
234 .vfs_checkexp
= hammer2_vfs_checkexp
237 MALLOC_DEFINE(M_HAMMER2
, "HAMMER2-mount", "");
239 VFS_SET(hammer2_vfsops
, hammer2
, VFCF_MPSAFE
);
240 MODULE_VERSION(hammer2
, 1);
244 hammer2_vfs_init(struct vfsconf
*conf
)
246 static struct objcache_malloc_args margs_read
;
247 static struct objcache_malloc_args margs_write
;
248 static struct objcache_malloc_args margs_vop
;
255 * A large DIO cache is needed to retain dedup enablement masks.
256 * The bulkfree code clears related masks as part of the disk block
257 * recycling algorithm, preventing it from being used for a later
260 * NOTE: A large buffer cache can actually interfere with dedup
261 * operation because we dedup based on media physical buffers
262 * and not logical buffers. Try to make the DIO chace large
263 * enough to avoid this problem, but also cap it.
265 hammer2_limit_dio
= nbuf
* 2;
266 if (hammer2_limit_dio
> 100000)
267 hammer2_limit_dio
= 100000;
269 if (HAMMER2_BLOCKREF_BYTES
!= sizeof(struct hammer2_blockref
))
271 if (HAMMER2_INODE_BYTES
!= sizeof(struct hammer2_inode_data
))
273 if (HAMMER2_VOLUME_BYTES
!= sizeof(struct hammer2_volume_data
))
277 kprintf("HAMMER2 structure size mismatch; cannot continue.\n");
279 margs_read
.objsize
= 65536;
280 margs_read
.mtype
= M_HAMMER2_DEBUFFER
;
282 margs_write
.objsize
= 32768;
283 margs_write
.mtype
= M_HAMMER2_CBUFFER
;
285 margs_vop
.objsize
= sizeof(hammer2_xop_t
);
286 margs_vop
.mtype
= M_HAMMER2
;
289 * Note thaht for the XOPS cache we want backing store allocations
290 * to use M_ZERO. This is not allowed in objcache_get() (to avoid
291 * confusion), so use the backing store function that does it. This
292 * means that initial XOPS objects are zerod but REUSED objects are
293 * not. So we are responsible for cleaning the object up sufficiently
294 * for our needs before objcache_put()ing it back (typically just the
297 cache_buffer_read
= objcache_create(margs_read
.mtype
->ks_shortdesc
,
298 0, 1, NULL
, NULL
, NULL
,
299 objcache_malloc_alloc
,
300 objcache_malloc_free
,
302 cache_buffer_write
= objcache_create(margs_write
.mtype
->ks_shortdesc
,
303 0, 1, NULL
, NULL
, NULL
,
304 objcache_malloc_alloc
,
305 objcache_malloc_free
,
307 cache_xops
= objcache_create(margs_vop
.mtype
->ks_shortdesc
,
308 0, 1, NULL
, NULL
, NULL
,
309 objcache_malloc_alloc_zero
,
310 objcache_malloc_free
,
314 lockinit(&hammer2_mntlk
, "mntlk", 0, 0);
315 TAILQ_INIT(&hammer2_mntlist
);
316 TAILQ_INIT(&hammer2_pfslist
);
318 hammer2_limit_dirty_chains
= maxvnodes
/ 10;
319 if (hammer2_limit_dirty_chains
> HAMMER2_LIMIT_DIRTY_CHAINS
)
320 hammer2_limit_dirty_chains
= HAMMER2_LIMIT_DIRTY_CHAINS
;
327 hammer2_vfs_uninit(struct vfsconf
*vfsp __unused
)
329 objcache_destroy(cache_buffer_read
);
330 objcache_destroy(cache_buffer_write
);
331 objcache_destroy(cache_xops
);
336 * Core PFS allocator. Used to allocate or reference the pmp structure
337 * for PFS cluster mounts and the spmp structure for media (hmp) structures.
338 * The pmp can be passed in or loaded by this function using the chain and
341 * pmp->modify_tid tracks new modify_tid transaction ids for front-end
342 * transactions. Note that synchronization does not use this field.
343 * (typically frontend operations and synchronization cannot run on the
344 * same PFS node at the same time).
349 hammer2_pfsalloc(hammer2_chain_t
*chain
,
350 const hammer2_inode_data_t
*ripdata
,
351 hammer2_tid_t modify_tid
, hammer2_dev_t
*force_local
)
354 hammer2_inode_t
*iroot
;
362 * Locate or create the PFS based on the cluster id. If ripdata
363 * is NULL this is a spmp which is unique and is always allocated.
365 * If the device is mounted in local mode all PFSs are considered
366 * independent and not part of any cluster (for debugging only).
369 TAILQ_FOREACH(pmp
, &hammer2_pfslist
, mntentry
) {
370 if (force_local
!= pmp
->force_local
)
372 if (force_local
== NULL
&&
373 bcmp(&pmp
->pfs_clid
, &ripdata
->meta
.pfs_clid
,
374 sizeof(pmp
->pfs_clid
)) == 0) {
376 } else if (force_local
&& pmp
->pfs_names
[0] &&
377 strcmp(pmp
->pfs_names
[0], ripdata
->filename
) == 0) {
384 pmp
= kmalloc(sizeof(*pmp
), M_HAMMER2
, M_WAITOK
| M_ZERO
);
385 pmp
->force_local
= force_local
;
386 hammer2_trans_manage_init(pmp
);
387 kmalloc_create(&pmp
->minode
, "HAMMER2-inodes");
388 kmalloc_create(&pmp
->mmsg
, "HAMMER2-pfsmsg");
389 lockinit(&pmp
->lock
, "pfslk", 0, 0);
390 lockinit(&pmp
->lock_nlink
, "h2nlink", 0, 0);
391 spin_init(&pmp
->inum_spin
, "hm2pfsalloc_inum");
392 spin_init(&pmp
->xop_spin
, "h2xop");
393 spin_init(&pmp
->lru_spin
, "h2lru");
394 RB_INIT(&pmp
->inum_tree
);
395 TAILQ_INIT(&pmp
->sideq
);
396 TAILQ_INIT(&pmp
->lru_list
);
397 spin_init(&pmp
->list_spin
, "hm2pfsalloc_list");
400 * Distribute backend operations to threads
402 for (i
= 0; i
< HAMMER2_XOPGROUPS
; ++i
)
403 hammer2_xop_group_init(pmp
, &pmp
->xop_groups
[i
]);
406 * Save the last media transaction id for the flusher. Set
410 pmp
->pfs_clid
= ripdata
->meta
.pfs_clid
;
411 TAILQ_INSERT_TAIL(&hammer2_pfslist
, pmp
, mntentry
);
414 * The synchronization thread may start too early, make
415 * sure it stays frozen until we are ready to let it go.
419 pmp->primary_thr.flags = HAMMER2_THREAD_FROZEN |
420 HAMMER2_THREAD_REMASTER;
425 * Create the PFS's root inode and any missing XOP helper threads.
427 if ((iroot
= pmp
->iroot
) == NULL
) {
428 iroot
= hammer2_inode_get(pmp
, NULL
, NULL
, -1);
430 iroot
->meta
= ripdata
->meta
;
432 hammer2_inode_ref(iroot
);
433 hammer2_inode_unlock(iroot
);
437 * Stop here if no chain is passed in.
443 * When a chain is passed in we must add it to the PFS's root
444 * inode, update pmp->pfs_types[], and update the syncronization
447 * When forcing local mode, mark the PFS as a MASTER regardless.
449 * At the moment empty spots can develop due to removals or failures.
450 * Ultimately we want to re-fill these spots but doing so might
451 * confused running code. XXX
453 hammer2_inode_ref(iroot
);
454 hammer2_mtx_ex(&iroot
->lock
);
455 j
= iroot
->cluster
.nchains
;
457 if (j
== HAMMER2_MAXCLUSTER
) {
458 kprintf("hammer2_mount: cluster full!\n");
459 /* XXX fatal error? */
461 KKASSERT(chain
->pmp
== NULL
);
463 hammer2_chain_ref(chain
);
464 iroot
->cluster
.array
[j
].chain
= chain
;
466 pmp
->pfs_types
[j
] = HAMMER2_PFSTYPE_MASTER
;
468 pmp
->pfs_types
[j
] = ripdata
->meta
.pfs_type
;
469 pmp
->pfs_names
[j
] = kstrdup(ripdata
->filename
, M_HAMMER2
);
470 pmp
->pfs_hmps
[j
] = chain
->hmp
;
473 * If the PFS is already mounted we must account
474 * for the mount_count here.
477 ++chain
->hmp
->mount_count
;
480 * May have to fixup dirty chain tracking. Previous
481 * pmp was NULL so nothing to undo.
483 if (chain
->flags
& HAMMER2_CHAIN_MODIFIED
)
484 hammer2_pfs_memory_inc(pmp
);
487 iroot
->cluster
.nchains
= j
;
490 * Update nmasters from any PFS inode which is part of the cluster.
491 * It is possible that this will result in a value which is too
492 * high. MASTER PFSs are authoritative for pfs_nmasters and will
493 * override this value later on.
495 * (This informs us of masters that might not currently be
496 * discoverable by this mount).
498 if (ripdata
&& pmp
->pfs_nmasters
< ripdata
->meta
.pfs_nmasters
) {
499 pmp
->pfs_nmasters
= ripdata
->meta
.pfs_nmasters
;
503 * Count visible masters. Masters are usually added with
504 * ripdata->meta.pfs_nmasters set to 1. This detects when there
505 * are more (XXX and must update the master inodes).
508 for (i
= 0; i
< iroot
->cluster
.nchains
; ++i
) {
509 if (pmp
->pfs_types
[i
] == HAMMER2_PFSTYPE_MASTER
)
512 if (pmp
->pfs_nmasters
< count
)
513 pmp
->pfs_nmasters
= count
;
516 * Create missing synchronization and support threads.
518 * Single-node masters (including snapshots) have nothing to
519 * synchronize and do not require this thread.
521 * Multi-node masters or any number of soft masters, slaves, copy,
522 * or other PFS types need the thread.
524 * Each thread is responsible for its particular cluster index.
525 * We use independent threads so stalls or mismatches related to
526 * any given target do not affect other targets.
528 for (i
= 0; i
< iroot
->cluster
.nchains
; ++i
) {
530 * Single-node masters (including snapshots) have nothing
531 * to synchronize and will make direct xops support calls,
532 * thus they do not require this thread.
534 * Note that there can be thousands of snapshots. We do not
535 * want to create thousands of threads.
537 if (pmp
->pfs_nmasters
<= 1 &&
538 pmp
->pfs_types
[i
] == HAMMER2_PFSTYPE_MASTER
) {
543 * Sync support thread
545 if (pmp
->sync_thrs
[i
].td
== NULL
) {
546 hammer2_thr_create(&pmp
->sync_thrs
[i
], pmp
, NULL
,
548 hammer2_primary_sync_thread
);
553 * Create missing Xop threads
555 * NOTE: We create helper threads for all mounted PFSs or any
556 * PFSs with 2+ nodes (so the sync thread can update them,
557 * even if not mounted).
559 if (pmp
->mp
|| iroot
->cluster
.nchains
>= 2)
560 hammer2_xop_helper_create(pmp
);
562 hammer2_mtx_unlock(&iroot
->lock
);
563 hammer2_inode_drop(iroot
);
569 * Deallocate an element of a probed PFS. If destroying and this is a
570 * MASTER, adjust nmasters.
572 * This function does not physically destroy the PFS element in its device
573 * under the super-root (see hammer2_ioctl_pfs_delete()).
576 hammer2_pfsdealloc(hammer2_pfs_t
*pmp
, int clindex
, int destroying
)
578 hammer2_inode_t
*iroot
;
579 hammer2_chain_t
*chain
;
583 * Cleanup our reference on iroot. iroot is (should) not be needed
591 * XXX flush after acquiring the iroot lock.
592 * XXX clean out the cluster index from all inode structures.
594 hammer2_thr_delete(&pmp
->sync_thrs
[clindex
]);
597 * Remove the cluster index from the group. If destroying
598 * the PFS and this is a master, adjust pfs_nmasters.
600 hammer2_mtx_ex(&iroot
->lock
);
601 chain
= iroot
->cluster
.array
[clindex
].chain
;
602 iroot
->cluster
.array
[clindex
].chain
= NULL
;
604 switch(pmp
->pfs_types
[clindex
]) {
605 case HAMMER2_PFSTYPE_MASTER
:
606 if (destroying
&& pmp
->pfs_nmasters
> 0)
608 /* XXX adjust ripdata->meta.pfs_nmasters */
613 pmp
->pfs_types
[clindex
] = HAMMER2_PFSTYPE_NONE
;
615 hammer2_mtx_unlock(&iroot
->lock
);
621 atomic_set_int(&chain
->flags
, HAMMER2_CHAIN_RELEASE
);
622 hammer2_chain_drop(chain
);
626 * Terminate all XOP threads for the cluster index.
628 for (j
= 0; j
< HAMMER2_XOPGROUPS
; ++j
)
629 hammer2_thr_delete(&pmp
->xop_groups
[j
].thrs
[clindex
]);
634 * Destroy a PFS, typically only occurs after the last mount on a device
638 hammer2_pfsfree(hammer2_pfs_t
*pmp
)
640 hammer2_inode_t
*iroot
;
641 hammer2_chain_t
*chain
;
646 * Cleanup our reference on iroot. iroot is (should) not be needed
649 TAILQ_REMOVE(&hammer2_pfslist
, pmp
, mntentry
);
653 for (i
= 0; i
< iroot
->cluster
.nchains
; ++i
) {
654 hammer2_thr_delete(&pmp
->sync_thrs
[i
]);
655 for (j
= 0; j
< HAMMER2_XOPGROUPS
; ++j
)
656 hammer2_thr_delete(&pmp
->xop_groups
[j
].thrs
[i
]);
658 #if REPORT_REFS_ERRORS
659 if (pmp
->iroot
->refs
!= 1)
660 kprintf("PMP->IROOT %p REFS WRONG %d\n",
661 pmp
->iroot
, pmp
->iroot
->refs
);
663 KKASSERT(pmp
->iroot
->refs
== 1);
665 /* ref for pmp->iroot */
666 hammer2_inode_drop(pmp
->iroot
);
671 * Cleanup chains remaining on LRU list.
673 while ((chain
= TAILQ_FIRST(&pmp
->lru_list
)) != NULL
) {
674 hammer2_chain_ref(chain
);
675 atomic_set_int(&chain
->flags
, HAMMER2_CHAIN_RELEASE
);
676 hammer2_chain_drop(chain
);
680 * Free remaining pmp resources
682 kmalloc_destroy(&pmp
->mmsg
);
683 kmalloc_destroy(&pmp
->minode
);
685 kfree(pmp
, M_HAMMER2
);
689 * Remove all references to hmp from the pfs list. Any PFS which becomes
690 * empty is terminated and freed.
695 hammer2_pfsfree_scan(hammer2_dev_t
*hmp
)
698 hammer2_inode_t
*iroot
;
699 hammer2_chain_t
*rchain
;
705 TAILQ_FOREACH(pmp
, &hammer2_pfslist
, mntentry
) {
706 if ((iroot
= pmp
->iroot
) == NULL
)
708 hammer2_trans_init(pmp
, HAMMER2_TRANS_ISFLUSH
);
709 hammer2_inode_run_sideq(pmp
, 1);
710 hammer2_bioq_sync(pmp
);
711 hammer2_trans_done(pmp
);
712 if (hmp
->spmp
== pmp
) {
714 hmp
->vchain
.pmp
= NULL
;
715 hmp
->fchain
.pmp
= NULL
;
719 * Determine if this PFS is affected. If it is we must
720 * freeze all management threads and lock its iroot.
722 * Freezing a management thread forces it idle, operations
723 * in-progress will be aborted and it will have to start
724 * over again when unfrozen, or exit if told to exit.
726 for (i
= 0; i
< HAMMER2_MAXCLUSTER
; ++i
) {
727 if (pmp
->pfs_hmps
[i
] == hmp
)
730 if (i
!= HAMMER2_MAXCLUSTER
) {
732 * Make sure all synchronization threads are locked
735 for (i
= 0; i
< HAMMER2_MAXCLUSTER
; ++i
) {
736 if (pmp
->pfs_hmps
[i
] == NULL
)
738 hammer2_thr_freeze_async(&pmp
->sync_thrs
[i
]);
739 for (j
= 0; j
< HAMMER2_XOPGROUPS
; ++j
) {
740 hammer2_thr_freeze_async(
741 &pmp
->xop_groups
[j
].thrs
[i
]);
744 for (i
= 0; i
< HAMMER2_MAXCLUSTER
; ++i
) {
745 if (pmp
->pfs_hmps
[i
] == NULL
)
747 hammer2_thr_freeze(&pmp
->sync_thrs
[i
]);
748 for (j
= 0; j
< HAMMER2_XOPGROUPS
; ++j
) {
750 &pmp
->xop_groups
[j
].thrs
[i
]);
755 * Lock the inode and clean out matching chains.
756 * Note that we cannot use hammer2_inode_lock_*()
757 * here because that would attempt to validate the
758 * cluster that we are in the middle of ripping
761 * WARNING! We are working directly on the inodes
764 hammer2_mtx_ex(&iroot
->lock
);
767 * Remove the chain from matching elements of the PFS.
769 for (i
= 0; i
< HAMMER2_MAXCLUSTER
; ++i
) {
770 if (pmp
->pfs_hmps
[i
] != hmp
)
772 hammer2_thr_delete(&pmp
->sync_thrs
[i
]);
773 for (j
= 0; j
< HAMMER2_XOPGROUPS
; ++j
) {
775 &pmp
->xop_groups
[j
].thrs
[i
]);
777 rchain
= iroot
->cluster
.array
[i
].chain
;
778 iroot
->cluster
.array
[i
].chain
= NULL
;
779 pmp
->pfs_types
[i
] = 0;
780 if (pmp
->pfs_names
[i
]) {
781 kfree(pmp
->pfs_names
[i
], M_HAMMER2
);
782 pmp
->pfs_names
[i
] = NULL
;
785 hammer2_chain_drop(rchain
);
787 if (iroot
->cluster
.focus
== rchain
)
788 iroot
->cluster
.focus
= NULL
;
790 pmp
->pfs_hmps
[i
] = NULL
;
792 hammer2_mtx_unlock(&iroot
->lock
);
793 didfreeze
= 1; /* remaster, unfreeze down below */
799 * Cleanup trailing chains. Gaps may remain.
801 for (i
= HAMMER2_MAXCLUSTER
- 1; i
>= 0; --i
) {
802 if (pmp
->pfs_hmps
[i
])
805 iroot
->cluster
.nchains
= i
+ 1;
808 * If the PMP has no elements remaining we can destroy it.
809 * (this will transition management threads from frozen->exit).
811 if (iroot
->cluster
.nchains
== 0) {
812 hammer2_pfsfree(pmp
);
817 * If elements still remain we need to set the REMASTER
818 * flag and unfreeze it.
821 for (i
= 0; i
< HAMMER2_MAXCLUSTER
; ++i
) {
822 if (pmp
->pfs_hmps
[i
] == NULL
)
824 hammer2_thr_remaster(&pmp
->sync_thrs
[i
]);
825 hammer2_thr_unfreeze(&pmp
->sync_thrs
[i
]);
826 for (j
= 0; j
< HAMMER2_XOPGROUPS
; ++j
) {
827 hammer2_thr_remaster(
828 &pmp
->xop_groups
[j
].thrs
[i
]);
829 hammer2_thr_unfreeze(
830 &pmp
->xop_groups
[j
].thrs
[i
]);
838 * Mount or remount HAMMER2 fileystem from physical media
841 * mp mount point structure
847 * mp mount point structure
848 * path path to mount point
849 * data pointer to argument structure in user space
850 * volume volume path (device@LABEL form)
851 * hflags user mount flags
852 * cred user credentials
859 hammer2_vfs_mount(struct mount
*mp
, char *path
, caddr_t data
,
862 struct hammer2_mount_info info
;
866 hammer2_dev_t
*force_local
;
867 hammer2_key_t key_next
;
868 hammer2_key_t key_dummy
;
871 struct nlookupdata nd
;
872 hammer2_chain_t
*parent
;
873 hammer2_chain_t
*chain
;
874 hammer2_cluster_t
*cluster
;
875 const hammer2_inode_data_t
*ripdata
;
876 hammer2_blockref_t bref
;
878 char devstr
[MNAMELEN
];
893 kprintf("hammer2_mount\n");
899 bzero(&info
, sizeof(info
));
900 info
.cluster_fd
= -1;
901 ksnprintf(devstr
, sizeof(devstr
), "%s",
902 mp
->mnt_stat
.f_mntfromname
);
903 kprintf("hammer2_mount: root '%s'\n", devstr
);
906 * Non-root mount or updating a mount
908 error
= copyin(data
, &info
, sizeof(info
));
912 error
= copyinstr(info
.volume
, devstr
, MNAMELEN
- 1, &done
);
918 * Extract device and label, automatically mount @BOOT, @ROOT, or @DATA
919 * if no label specified, based on the partition id. Error out if no
920 * label or device (with partition id) is specified. This is strictly
921 * a convenience to match the default label created by newfs_hammer2,
922 * our preference is that a label always be specified.
924 * NOTE: We allow 'mount @LABEL <blah>'... that is, a mount command
925 * that does not specify a device, as long as some H2 label
926 * has already been mounted from that device. This makes
927 * mounting snapshots a lot easier.
930 label
= strchr(devstr
, '@');
931 if (label
&& ((label
+ 1) - dev
) > done
)
933 if (label
== NULL
|| label
[1] == 0) {
937 label
= devstr
+ strlen(devstr
);
955 kprintf("hammer2_mount: dev=\"%s\" label=\"%s\" rdonly=%d\n",
956 dev
, label
, (mp
->mnt_flag
& MNT_RDONLY
));
958 if (mp
->mnt_flag
& MNT_UPDATE
) {
960 * Update mount. Note that pmp->iroot->cluster is
961 * an inode-embedded cluster and thus cannot be
964 * XXX HAMMER2 needs to implement NFS export via
968 pmp
->hflags
= info
.hflags
;
969 cluster
= &pmp
->iroot
->cluster
;
970 for (i
= 0; i
< cluster
->nchains
; ++i
) {
971 if (cluster
->array
[i
].chain
== NULL
)
973 hmp
= cluster
->array
[i
].chain
->hmp
;
975 error
= hammer2_remount(hmp
, mp
, path
,
987 * If a path is specified and dev is not an empty string, lookup the
988 * name and verify that it referes to a block device.
990 * If a path is specified and dev is an empty string we fall through
991 * and locate the label in the hmp search.
993 if (path
&& *dev
!= 0) {
994 error
= nlookup_init(&nd
, dev
, UIO_SYSSPACE
, NLC_FOLLOW
);
996 error
= nlookup(&nd
);
998 error
= cache_vref(&nd
.nl_nch
, nd
.nl_cred
, &devvp
);
1000 } else if (path
== NULL
) {
1002 cdev_t cdev
= kgetdiskbyname(dev
);
1003 error
= bdevvp(cdev
, &devvp
);
1005 kprintf("hammer2: cannot find '%s'\n", dev
);
1008 * We will locate the hmp using the label in the hmp loop.
1014 * Make sure its a block device. Do not check to see if it is
1015 * already mounted until we determine that its a fresh H2 device.
1017 if (error
== 0 && devvp
) {
1018 vn_isdisk(devvp
, &error
);
1022 * Determine if the device has already been mounted. After this
1023 * check hmp will be non-NULL if we are doing the second or more
1024 * hammer2 mounts from the same device.
1026 lockmgr(&hammer2_mntlk
, LK_EXCLUSIVE
);
1029 * Match the device. Due to the way devfs works,
1030 * we may not be able to directly match the vnode pointer,
1031 * so also check to see if the underlying device matches.
1033 TAILQ_FOREACH(hmp
, &hammer2_mntlist
, mntentry
) {
1034 if (hmp
->devvp
== devvp
)
1036 if (devvp
->v_rdev
&&
1037 hmp
->devvp
->v_rdev
== devvp
->v_rdev
) {
1043 * If no match this may be a fresh H2 mount, make sure
1044 * the device is not mounted on anything else.
1047 error
= vfs_mountedon(devvp
);
1048 } else if (error
== 0) {
1050 * Match the label to a pmp already probed.
1052 TAILQ_FOREACH(pmp
, &hammer2_pfslist
, mntentry
) {
1053 for (i
= 0; i
< HAMMER2_MAXCLUSTER
; ++i
) {
1054 if (pmp
->pfs_names
[i
] &&
1055 strcmp(pmp
->pfs_names
[i
], label
) == 0) {
1056 hmp
= pmp
->pfs_hmps
[i
];
1068 * Open the device if this isn't a secondary mount and construct
1069 * the H2 device mount (hmp).
1072 hammer2_chain_t
*schain
;
1075 if (error
== 0 && vcount(devvp
) > 0) {
1076 kprintf("Primary device already has references\n");
1081 * Now open the device
1084 ronly
= ((mp
->mnt_flag
& MNT_RDONLY
) != 0);
1085 vn_lock(devvp
, LK_EXCLUSIVE
| LK_RETRY
);
1086 error
= vinvalbuf(devvp
, V_SAVE
, 0, 0);
1088 error
= VOP_OPEN(devvp
,
1089 (ronly
? FREAD
: FREAD
| FWRITE
),
1094 if (error
&& devvp
) {
1099 lockmgr(&hammer2_mntlk
, LK_RELEASE
);
1102 hmp
= kmalloc(sizeof(*hmp
), M_HAMMER2
, M_WAITOK
| M_ZERO
);
1103 ksnprintf(hmp
->devrepname
, sizeof(hmp
->devrepname
), "%s", dev
);
1106 hmp
->hflags
= info
.hflags
& HMNT2_DEVFLAGS
;
1107 kmalloc_create(&hmp
->mchain
, "HAMMER2-chains");
1108 TAILQ_INSERT_TAIL(&hammer2_mntlist
, hmp
, mntentry
);
1109 RB_INIT(&hmp
->iotree
);
1110 spin_init(&hmp
->io_spin
, "hm2mount_io");
1111 spin_init(&hmp
->list_spin
, "hm2mount_list");
1112 TAILQ_INIT(&hmp
->flushq
);
1114 lockinit(&hmp
->vollk
, "h2vol", 0, 0);
1115 lockinit(&hmp
->bulklk
, "h2bulk", 0, 0);
1116 lockinit(&hmp
->bflock
, "h2bflk", 0, 0);
1119 * vchain setup. vchain.data is embedded.
1120 * vchain.refs is initialized and will never drop to 0.
1122 * NOTE! voldata is not yet loaded.
1124 hmp
->vchain
.hmp
= hmp
;
1125 hmp
->vchain
.refs
= 1;
1126 hmp
->vchain
.data
= (void *)&hmp
->voldata
;
1127 hmp
->vchain
.bref
.type
= HAMMER2_BREF_TYPE_VOLUME
;
1128 hmp
->vchain
.bref
.data_off
= 0 | HAMMER2_PBUFRADIX
;
1129 hmp
->vchain
.bref
.mirror_tid
= hmp
->voldata
.mirror_tid
;
1131 hammer2_chain_core_init(&hmp
->vchain
);
1132 /* hmp->vchain.u.xxx is left NULL */
1135 * fchain setup. fchain.data is embedded.
1136 * fchain.refs is initialized and will never drop to 0.
1138 * The data is not used but needs to be initialized to
1139 * pass assertion muster. We use this chain primarily
1140 * as a placeholder for the freemap's top-level RBTREE
1141 * so it does not interfere with the volume's topology
1144 hmp
->fchain
.hmp
= hmp
;
1145 hmp
->fchain
.refs
= 1;
1146 hmp
->fchain
.data
= (void *)&hmp
->voldata
.freemap_blockset
;
1147 hmp
->fchain
.bref
.type
= HAMMER2_BREF_TYPE_FREEMAP
;
1148 hmp
->fchain
.bref
.data_off
= 0 | HAMMER2_PBUFRADIX
;
1149 hmp
->fchain
.bref
.mirror_tid
= hmp
->voldata
.freemap_tid
;
1150 hmp
->fchain
.bref
.methods
=
1151 HAMMER2_ENC_CHECK(HAMMER2_CHECK_FREEMAP
) |
1152 HAMMER2_ENC_COMP(HAMMER2_COMP_NONE
);
1154 hammer2_chain_core_init(&hmp
->fchain
);
1155 /* hmp->fchain.u.xxx is left NULL */
1158 * Install the volume header and initialize fields from
1161 error
= hammer2_install_volume_header(hmp
);
1163 hammer2_unmount_helper(mp
, NULL
, hmp
);
1164 lockmgr(&hammer2_mntlk
, LK_RELEASE
);
1165 hammer2_vfs_unmount(mp
, MNT_FORCE
);
1170 * Really important to get these right or flush will get
1173 hmp
->spmp
= hammer2_pfsalloc(NULL
, NULL
, 0, NULL
);
1174 kprintf("alloc spmp %p tid %016jx\n",
1175 hmp
->spmp
, hmp
->voldata
.mirror_tid
);
1179 * Dummy-up vchain and fchain's modify_tid. mirror_tid
1180 * is inherited from the volume header.
1183 hmp
->vchain
.bref
.mirror_tid
= hmp
->voldata
.mirror_tid
;
1184 hmp
->vchain
.bref
.modify_tid
= hmp
->vchain
.bref
.mirror_tid
;
1185 hmp
->vchain
.pmp
= spmp
;
1186 hmp
->fchain
.bref
.mirror_tid
= hmp
->voldata
.freemap_tid
;
1187 hmp
->fchain
.bref
.modify_tid
= hmp
->fchain
.bref
.mirror_tid
;
1188 hmp
->fchain
.pmp
= spmp
;
1191 * First locate the super-root inode, which is key 0
1192 * relative to the volume header's blockset.
1194 * Then locate the root inode by scanning the directory keyspace
1195 * represented by the label.
1197 parent
= hammer2_chain_lookup_init(&hmp
->vchain
, 0);
1198 schain
= hammer2_chain_lookup(&parent
, &key_dummy
,
1199 HAMMER2_SROOT_KEY
, HAMMER2_SROOT_KEY
,
1201 hammer2_chain_lookup_done(parent
);
1202 if (schain
== NULL
) {
1203 kprintf("hammer2_mount: invalid super-root\n");
1204 hammer2_unmount_helper(mp
, NULL
, hmp
);
1205 lockmgr(&hammer2_mntlk
, LK_RELEASE
);
1206 hammer2_vfs_unmount(mp
, MNT_FORCE
);
1209 if (schain
->error
) {
1210 kprintf("hammer2_mount: error %s reading super-root\n",
1211 hammer2_error_str(schain
->error
));
1212 hammer2_chain_unlock(schain
);
1213 hammer2_chain_drop(schain
);
1215 hammer2_unmount_helper(mp
, NULL
, hmp
);
1216 lockmgr(&hammer2_mntlk
, LK_RELEASE
);
1217 hammer2_vfs_unmount(mp
, MNT_FORCE
);
1222 * The super-root always uses an inode_tid of 1 when
1225 spmp
->inode_tid
= 1;
1226 spmp
->modify_tid
= schain
->bref
.modify_tid
+ 1;
1229 * Sanity-check schain's pmp and finish initialization.
1230 * Any chain belonging to the super-root topology should
1231 * have a NULL pmp (not even set to spmp).
1233 ripdata
= &hammer2_chain_rdata(schain
)->ipdata
;
1234 KKASSERT(schain
->pmp
== NULL
);
1235 spmp
->pfs_clid
= ripdata
->meta
.pfs_clid
;
1238 * Replace the dummy spmp->iroot with a real one. It's
1239 * easier to just do a wholesale replacement than to try
1240 * to update the chain and fixup the iroot fields.
1242 * The returned inode is locked with the supplied cluster.
1244 cluster
= hammer2_cluster_from_chain(schain
);
1245 hammer2_inode_drop(spmp
->iroot
);
1247 spmp
->iroot
= hammer2_inode_get(spmp
, NULL
, cluster
, -1);
1248 spmp
->spmp_hmp
= hmp
;
1249 spmp
->pfs_types
[0] = ripdata
->meta
.pfs_type
;
1250 spmp
->pfs_hmps
[0] = hmp
;
1251 hammer2_inode_ref(spmp
->iroot
);
1252 hammer2_inode_unlock(spmp
->iroot
);
1253 hammer2_cluster_unlock(cluster
);
1254 hammer2_cluster_drop(cluster
);
1256 /* leave spmp->iroot with one ref */
1258 if ((mp
->mnt_flag
& MNT_RDONLY
) == 0) {
1259 error
= hammer2_recovery(hmp
);
1260 /* XXX do something with error */
1262 hammer2_update_pmps(hmp
);
1263 hammer2_iocom_init(hmp
);
1264 hammer2_bulkfree_init(hmp
);
1267 * Ref the cluster management messaging descriptor. The mount
1268 * program deals with the other end of the communications pipe.
1270 * Root mounts typically do not supply one.
1272 if (info
.cluster_fd
>= 0) {
1273 fp
= holdfp(curproc
->p_fd
, info
.cluster_fd
, -1);
1275 hammer2_cluster_reconnect(hmp
, fp
);
1277 kprintf("hammer2_mount: bad cluster_fd!\n");
1282 if (info
.hflags
& HMNT2_DEVFLAGS
) {
1283 kprintf("hammer2: Warning: mount flags pertaining "
1284 "to the whole device may only be specified "
1285 "on the first mount of the device: %08x\n",
1286 info
.hflags
& HMNT2_DEVFLAGS
);
1291 * Force local mount (disassociate all PFSs from their clusters).
1292 * Used primarily for debugging.
1294 force_local
= (hmp
->hflags
& HMNT2_LOCAL
) ? hmp
: NULL
;
1297 * Lookup the mount point under the media-localized super-root.
1298 * Scanning hammer2_pfslist doesn't help us because it represents
1299 * PFS cluster ids which can aggregate several named PFSs together.
1301 * cluster->pmp will incorrectly point to spmp and must be fixed
1304 hammer2_inode_lock(spmp
->iroot
, 0);
1305 parent
= hammer2_inode_chain(spmp
->iroot
, 0, HAMMER2_RESOLVE_ALWAYS
);
1306 lhc
= hammer2_dirhash(label
, strlen(label
));
1307 chain
= hammer2_chain_lookup(&parent
, &key_next
,
1308 lhc
, lhc
+ HAMMER2_DIRHASH_LOMASK
,
1311 if (chain
->bref
.type
== HAMMER2_BREF_TYPE_INODE
&&
1312 strcmp(label
, chain
->data
->ipdata
.filename
) == 0) {
1315 chain
= hammer2_chain_next(&parent
, chain
, &key_next
,
1317 lhc
+ HAMMER2_DIRHASH_LOMASK
,
1321 hammer2_chain_unlock(parent
);
1322 hammer2_chain_drop(parent
);
1324 hammer2_inode_unlock(spmp
->iroot
);
1327 * PFS could not be found?
1329 if (chain
== NULL
) {
1331 kprintf("hammer2_mount: PFS label I/O error\n");
1333 kprintf("hammer2_mount: PFS label not found\n");
1334 hammer2_unmount_helper(mp
, NULL
, hmp
);
1335 lockmgr(&hammer2_mntlk
, LK_RELEASE
);
1336 hammer2_vfs_unmount(mp
, MNT_FORCE
);
1342 * Acquire the pmp structure (it should have already been allocated
1343 * via hammer2_update_pmps() so do not pass cluster in to add to
1344 * available chains).
1346 * Check if the cluster has already been mounted. A cluster can
1347 * only be mounted once, use null mounts to mount additional copies.
1350 kprintf("hammer2_mount: PFS label I/O error\n");
1352 ripdata
= &chain
->data
->ipdata
;
1354 pmp
= hammer2_pfsalloc(NULL
, ripdata
,
1355 bref
.modify_tid
, force_local
);
1357 hammer2_chain_unlock(chain
);
1358 hammer2_chain_drop(chain
);
1363 kprintf("hammer2_mount hmp=%p pmp=%p\n", hmp
, pmp
);
1366 kprintf("hammer2_mount: PFS already mounted!\n");
1367 hammer2_unmount_helper(mp
, NULL
, hmp
);
1368 lockmgr(&hammer2_mntlk
, LK_RELEASE
);
1369 hammer2_vfs_unmount(mp
, MNT_FORCE
);
1374 pmp
->hflags
= info
.hflags
;
1375 mp
->mnt_flag
|= MNT_LOCAL
;
1376 mp
->mnt_kern_flag
|= MNTK_ALL_MPSAFE
; /* all entry pts are SMP */
1377 mp
->mnt_kern_flag
|= MNTK_THR_SYNC
; /* new vsyncscan semantics */
1380 * required mount structure initializations
1382 mp
->mnt_stat
.f_iosize
= HAMMER2_PBUFSIZE
;
1383 mp
->mnt_stat
.f_bsize
= HAMMER2_PBUFSIZE
;
1385 mp
->mnt_vstat
.f_frsize
= HAMMER2_PBUFSIZE
;
1386 mp
->mnt_vstat
.f_bsize
= HAMMER2_PBUFSIZE
;
1391 mp
->mnt_iosize_max
= MAXPHYS
;
1394 * Connect up mount pointers.
1396 hammer2_mount_helper(mp
, pmp
);
1398 lockmgr(&hammer2_mntlk
, LK_RELEASE
);
1404 vfs_add_vnodeops(mp
, &hammer2_vnode_vops
, &mp
->mnt_vn_norm_ops
);
1405 vfs_add_vnodeops(mp
, &hammer2_spec_vops
, &mp
->mnt_vn_spec_ops
);
1406 vfs_add_vnodeops(mp
, &hammer2_fifo_vops
, &mp
->mnt_vn_fifo_ops
);
1409 copyinstr(info
.volume
, mp
->mnt_stat
.f_mntfromname
,
1410 MNAMELEN
- 1, &size
);
1411 bzero(mp
->mnt_stat
.f_mntfromname
+ size
, MNAMELEN
- size
);
1412 } /* else root mount, already in there */
1414 bzero(mp
->mnt_stat
.f_mntonname
, sizeof(mp
->mnt_stat
.f_mntonname
));
1416 copyinstr(path
, mp
->mnt_stat
.f_mntonname
,
1417 sizeof(mp
->mnt_stat
.f_mntonname
) - 1,
1421 mp
->mnt_stat
.f_mntonname
[0] = '/';
1425 * Initial statfs to prime mnt_stat.
1427 hammer2_vfs_statfs(mp
, &mp
->mnt_stat
, cred
);
1433 * Scan PFSs under the super-root and create hammer2_pfs structures.
1437 hammer2_update_pmps(hammer2_dev_t
*hmp
)
1439 const hammer2_inode_data_t
*ripdata
;
1440 hammer2_chain_t
*parent
;
1441 hammer2_chain_t
*chain
;
1442 hammer2_blockref_t bref
;
1443 hammer2_dev_t
*force_local
;
1444 hammer2_pfs_t
*spmp
;
1446 hammer2_key_t key_next
;
1450 * Force local mount (disassociate all PFSs from their clusters).
1451 * Used primarily for debugging.
1453 force_local
= (hmp
->hflags
& HMNT2_LOCAL
) ? hmp
: NULL
;
1456 * Lookup mount point under the media-localized super-root.
1458 * cluster->pmp will incorrectly point to spmp and must be fixed
1462 hammer2_inode_lock(spmp
->iroot
, 0);
1463 parent
= hammer2_inode_chain(spmp
->iroot
, 0, HAMMER2_RESOLVE_ALWAYS
);
1464 chain
= hammer2_chain_lookup(&parent
, &key_next
,
1465 HAMMER2_KEY_MIN
, HAMMER2_KEY_MAX
,
1468 if (chain
->bref
.type
!= HAMMER2_BREF_TYPE_INODE
)
1471 kprintf("I/O error scanning PFS labels\n");
1473 ripdata
= &chain
->data
->ipdata
;
1476 pmp
= hammer2_pfsalloc(chain
, ripdata
,
1477 bref
.modify_tid
, force_local
);
1479 chain
= hammer2_chain_next(&parent
, chain
, &key_next
,
1480 key_next
, HAMMER2_KEY_MAX
,
1484 hammer2_chain_unlock(parent
);
1485 hammer2_chain_drop(parent
);
1487 hammer2_inode_unlock(spmp
->iroot
);
1492 hammer2_remount(hammer2_dev_t
*hmp
, struct mount
*mp
, char *path __unused
,
1493 struct vnode
*devvp
, struct ucred
*cred
)
1497 if (hmp
->ronly
&& (mp
->mnt_kern_flag
& MNTK_WANTRDWR
)) {
1498 vn_lock(devvp
, LK_EXCLUSIVE
| LK_RETRY
);
1499 VOP_OPEN(devvp
, FREAD
| FWRITE
, FSCRED
, NULL
);
1501 error
= hammer2_recovery(hmp
);
1502 vn_lock(devvp
, LK_EXCLUSIVE
| LK_RETRY
);
1504 VOP_CLOSE(devvp
, FREAD
, NULL
);
1507 VOP_CLOSE(devvp
, FREAD
| FWRITE
, NULL
);
1518 hammer2_vfs_unmount(struct mount
*mp
, int mntflags
)
1529 lockmgr(&hammer2_mntlk
, LK_EXCLUSIVE
);
1532 * If mount initialization proceeded far enough we must flush
1533 * its vnodes and sync the underlying mount points. Three syncs
1534 * are required to fully flush the filesystem (freemap updates lag
1535 * by one flush, and one extra for safety).
1537 if (mntflags
& MNT_FORCE
)
1542 error
= vflush(mp
, 0, flags
);
1545 hammer2_vfs_sync(mp
, MNT_WAIT
);
1546 hammer2_vfs_sync(mp
, MNT_WAIT
);
1547 hammer2_vfs_sync(mp
, MNT_WAIT
);
1551 * Cleanup the frontend support XOPS threads
1553 hammer2_xop_helper_cleanup(pmp
);
1556 hammer2_unmount_helper(mp
, pmp
, NULL
);
1560 lockmgr(&hammer2_mntlk
, LK_RELEASE
);
1566 * Mount helper, hook the system mount into our PFS.
1567 * The mount lock is held.
1569 * We must bump the mount_count on related devices for any
1574 hammer2_mount_helper(struct mount
*mp
, hammer2_pfs_t
*pmp
)
1576 hammer2_cluster_t
*cluster
;
1577 hammer2_chain_t
*rchain
;
1580 mp
->mnt_data
= (qaddr_t
)pmp
;
1584 * After pmp->mp is set we have to adjust hmp->mount_count.
1586 cluster
= &pmp
->iroot
->cluster
;
1587 for (i
= 0; i
< cluster
->nchains
; ++i
) {
1588 rchain
= cluster
->array
[i
].chain
;
1591 ++rchain
->hmp
->mount_count
;
1592 kprintf("hammer2_mount hmp=%p ++mount_count=%d\n",
1593 rchain
->hmp
, rchain
->hmp
->mount_count
);
1597 * Create missing Xop threads
1599 hammer2_xop_helper_create(pmp
);
1603 * Mount helper, unhook the system mount from our PFS.
1604 * The mount lock is held.
1606 * If hmp is supplied a mount responsible for being the first to open
1607 * the block device failed and the block device and all PFSs using the
1608 * block device must be cleaned up.
1610 * If pmp is supplied multiple devices might be backing the PFS and each
1611 * must be disconnected. This might not be the last PFS using some of the
1612 * underlying devices. Also, we have to adjust our hmp->mount_count
1613 * accounting for the devices backing the pmp which is now undergoing an
1618 hammer2_unmount_helper(struct mount
*mp
, hammer2_pfs_t
*pmp
, hammer2_dev_t
*hmp
)
1620 hammer2_cluster_t
*cluster
;
1621 hammer2_chain_t
*rchain
;
1622 struct vnode
*devvp
;
1628 * If no device supplied this is a high-level unmount and we have to
1629 * to disconnect the mount, adjust mount_count, and locate devices
1630 * that might now have no mounts.
1633 KKASSERT(hmp
== NULL
);
1634 KKASSERT((void *)(intptr_t)mp
->mnt_data
== pmp
);
1636 mp
->mnt_data
= NULL
;
1639 * After pmp->mp is cleared we have to account for
1642 cluster
= &pmp
->iroot
->cluster
;
1643 for (i
= 0; i
< cluster
->nchains
; ++i
) {
1644 rchain
= cluster
->array
[i
].chain
;
1647 --rchain
->hmp
->mount_count
;
1648 /* scrapping hmp now may invalidate the pmp */
1651 TAILQ_FOREACH(hmp
, &hammer2_mntlist
, mntentry
) {
1652 if (hmp
->mount_count
== 0) {
1653 hammer2_unmount_helper(NULL
, NULL
, hmp
);
1661 * Try to terminate the block device. We can't terminate it if
1662 * there are still PFSs referencing it.
1664 if (hmp
->mount_count
)
1668 * Decomission the network before we start messing with the
1671 hammer2_iocom_uninit(hmp
);
1673 hammer2_bulkfree_uninit(hmp
);
1674 hammer2_pfsfree_scan(hmp
);
1675 hammer2_dev_exlock(hmp
); /* XXX order */
1678 * Cycle the volume data lock as a safety (probably not needed any
1679 * more). To ensure everything is out we need to flush at least
1680 * three times. (1) The running of the sideq can dirty the
1681 * filesystem, (2) A normal flush can dirty the freemap, and
1682 * (3) ensure that the freemap is fully synchronized.
1684 * The next mount's recovery scan can clean everything up but we want
1685 * to leave the filesystem in a 100% clean state on a normal unmount.
1688 hammer2_voldata_lock(hmp
);
1689 hammer2_voldata_unlock(hmp
);
1693 * Flush whatever is left. Unmounted but modified PFS's might still
1694 * have some dirty chains on them.
1696 hammer2_chain_lock(&hmp
->vchain
, HAMMER2_RESOLVE_ALWAYS
);
1697 hammer2_chain_lock(&hmp
->fchain
, HAMMER2_RESOLVE_ALWAYS
);
1698 hammer2_flush(&hmp
->fchain
, HAMMER2_FLUSH_TOP
| HAMMER2_FLUSH_ALL
);
1699 hammer2_chain_unlock(&hmp
->fchain
);
1700 hammer2_flush(&hmp
->vchain
, HAMMER2_FLUSH_TOP
| HAMMER2_FLUSH_ALL
);
1701 hammer2_chain_unlock(&hmp
->vchain
);
1703 if ((hmp
->vchain
.flags
| hmp
->fchain
.flags
) &
1704 HAMMER2_CHAIN_FLUSH_MASK
) {
1705 kprintf("hammer2_unmount: chains left over "
1706 "after final sync\n");
1707 kprintf(" vchain %08x\n", hmp
->vchain
.flags
);
1708 kprintf(" fchain %08x\n", hmp
->fchain
.flags
);
1710 if (hammer2_debug
& 0x0010)
1711 Debugger("entered debugger");
1714 KKASSERT(hmp
->spmp
== NULL
);
1717 * Finish up with the device vnode
1719 if ((devvp
= hmp
->devvp
) != NULL
) {
1721 vn_lock(devvp
, LK_EXCLUSIVE
| LK_RETRY
);
1722 kprintf("hammer2_unmount(A): devvp %s rbdirty %p ronly=%d\n",
1723 hmp
->devrepname
, RB_ROOT(&devvp
->v_rbdirty_tree
),
1725 vinvalbuf(devvp
, (ronly
? 0 : V_SAVE
), 0, 0);
1726 kprintf("hammer2_unmount(B): devvp %s rbdirty %p\n",
1727 hmp
->devrepname
, RB_ROOT(&devvp
->v_rbdirty_tree
));
1729 VOP_CLOSE(devvp
, (ronly
? FREAD
: FREAD
|FWRITE
), NULL
);
1736 * Clear vchain/fchain flags that might prevent final cleanup
1739 if (hmp
->vchain
.flags
& HAMMER2_CHAIN_MODIFIED
) {
1740 atomic_add_long(&hammer2_count_modified_chains
, -1);
1741 atomic_clear_int(&hmp
->vchain
.flags
, HAMMER2_CHAIN_MODIFIED
);
1742 hammer2_pfs_memory_wakeup(hmp
->vchain
.pmp
);
1744 if (hmp
->vchain
.flags
& HAMMER2_CHAIN_UPDATE
) {
1745 atomic_clear_int(&hmp
->vchain
.flags
, HAMMER2_CHAIN_UPDATE
);
1748 if (hmp
->fchain
.flags
& HAMMER2_CHAIN_MODIFIED
) {
1749 atomic_add_long(&hammer2_count_modified_chains
, -1);
1750 atomic_clear_int(&hmp
->fchain
.flags
, HAMMER2_CHAIN_MODIFIED
);
1751 hammer2_pfs_memory_wakeup(hmp
->fchain
.pmp
);
1753 if (hmp
->fchain
.flags
& HAMMER2_CHAIN_UPDATE
) {
1754 atomic_clear_int(&hmp
->fchain
.flags
, HAMMER2_CHAIN_UPDATE
);
1758 * Final drop of embedded freemap root chain to
1759 * clean up fchain.core (fchain structure is not
1760 * flagged ALLOCATED so it is cleaned out and then
1763 hammer2_chain_drop(&hmp
->fchain
);
1766 * Final drop of embedded volume root chain to clean
1767 * up vchain.core (vchain structure is not flagged
1768 * ALLOCATED so it is cleaned out and then left to
1772 hammer2_dump_chain(&hmp
->vchain
, 0, &dumpcnt
, 'v');
1774 hammer2_dump_chain(&hmp
->fchain
, 0, &dumpcnt
, 'f');
1775 hammer2_dev_unlock(hmp
);
1776 hammer2_chain_drop(&hmp
->vchain
);
1778 hammer2_io_cleanup(hmp
, &hmp
->iotree
);
1779 if (hmp
->iofree_count
) {
1780 kprintf("io_cleanup: %d I/O's left hanging\n",
1784 TAILQ_REMOVE(&hammer2_mntlist
, hmp
, mntentry
);
1785 kmalloc_destroy(&hmp
->mchain
);
1786 kfree(hmp
, M_HAMMER2
);
1790 hammer2_vfs_vget(struct mount
*mp
, struct vnode
*dvp
,
1791 ino_t ino
, struct vnode
**vpp
)
1793 hammer2_xop_lookup_t
*xop
;
1795 hammer2_inode_t
*ip
;
1799 inum
= (hammer2_tid_t
)ino
& HAMMER2_DIRHASH_USERMSK
;
1805 * Easy if we already have it cached
1807 ip
= hammer2_inode_lookup(pmp
, inum
);
1809 hammer2_inode_lock(ip
, HAMMER2_RESOLVE_SHARED
);
1810 *vpp
= hammer2_igetv(ip
, &error
);
1811 hammer2_inode_unlock(ip
);
1812 hammer2_inode_drop(ip
); /* from lookup */
1818 * Otherwise we have to find the inode
1820 xop
= hammer2_xop_alloc(pmp
->iroot
, 0);
1822 hammer2_xop_start(&xop
->head
, hammer2_xop_lookup
);
1823 error
= hammer2_xop_collect(&xop
->head
, 0);
1826 if (hammer2_cluster_rdata(&xop
->head
.cluster
) == NULL
) {
1827 kprintf("vget: no collect error but also no rdata\n");
1828 kprintf("xop %p\n", xop
);
1829 while ((hammer2_debug
& 0x80000) == 0) {
1830 tsleep(xop
, PCATCH
, "wait", hz
* 10);
1834 ip
= hammer2_inode_get(pmp
, NULL
, &xop
->head
.cluster
, -1);
1837 hammer2_xop_retire(&xop
->head
, HAMMER2_XOPMASK_VOP
);
1840 *vpp
= hammer2_igetv(ip
, &error
);
1841 hammer2_inode_unlock(ip
);
1851 hammer2_vfs_root(struct mount
*mp
, struct vnode
**vpp
)
1858 if (pmp
->iroot
== NULL
) {
1864 hammer2_inode_lock(pmp
->iroot
, HAMMER2_RESOLVE_SHARED
);
1866 while (pmp
->inode_tid
== 0) {
1867 hammer2_xop_ipcluster_t
*xop
;
1868 hammer2_inode_meta_t
*meta
;
1870 xop
= hammer2_xop_alloc(pmp
->iroot
, HAMMER2_XOP_MODIFYING
);
1871 hammer2_xop_start(&xop
->head
, hammer2_xop_ipcluster
);
1872 error
= hammer2_xop_collect(&xop
->head
, 0);
1875 meta
= &xop
->head
.cluster
.focus
->data
->ipdata
.meta
;
1876 pmp
->iroot
->meta
= *meta
;
1877 pmp
->inode_tid
= meta
->pfs_inum
+ 1;
1878 if (pmp
->inode_tid
< HAMMER2_INODE_START
)
1879 pmp
->inode_tid
= HAMMER2_INODE_START
;
1881 xop
->head
.cluster
.focus
->bref
.modify_tid
+ 1;
1882 kprintf("PFS: Starting inode %jd\n",
1883 (intmax_t)pmp
->inode_tid
);
1884 kprintf("PMP focus good set nextino=%ld mod=%016jx\n",
1885 pmp
->inode_tid
, pmp
->modify_tid
);
1886 wakeup(&pmp
->iroot
);
1888 hammer2_xop_retire(&xop
->head
, HAMMER2_XOPMASK_VOP
);
1891 * Prime the mount info.
1893 hammer2_vfs_statfs(mp
, &mp
->mnt_stat
, NULL
);
1900 hammer2_xop_retire(&xop
->head
, HAMMER2_XOPMASK_VOP
);
1901 hammer2_inode_unlock(pmp
->iroot
);
1902 error
= tsleep(&pmp
->iroot
, PCATCH
, "h2root", hz
);
1903 hammer2_inode_lock(pmp
->iroot
, HAMMER2_RESOLVE_SHARED
);
1909 hammer2_inode_unlock(pmp
->iroot
);
1912 vp
= hammer2_igetv(pmp
->iroot
, &error
);
1913 hammer2_inode_unlock(pmp
->iroot
);
1923 * XXX incorporate ipdata->meta.inode_quota and data_quota
1927 hammer2_vfs_statfs(struct mount
*mp
, struct statfs
*sbp
, struct ucred
*cred
)
1931 hammer2_blockref_t bref
;
1936 * NOTE: iroot might not have validated the cluster yet.
1940 bzero(&tmp
, sizeof(tmp
));
1942 for (i
= 0; i
< pmp
->iroot
->cluster
.nchains
; ++i
) {
1943 hmp
= pmp
->pfs_hmps
[i
];
1946 if (pmp
->iroot
->cluster
.array
[i
].chain
)
1947 bref
= pmp
->iroot
->cluster
.array
[i
].chain
->bref
;
1949 bzero(&bref
, sizeof(bref
));
1951 tmp
.f_files
= bref
.embed
.stats
.inode_count
;
1953 tmp
.f_blocks
= hmp
->voldata
.allocator_size
/
1954 mp
->mnt_vstat
.f_bsize
;
1955 tmp
.f_bfree
= hmp
->voldata
.allocator_free
/
1956 mp
->mnt_vstat
.f_bsize
;
1957 tmp
.f_bavail
= tmp
.f_bfree
;
1959 if (cred
&& cred
->cr_uid
!= 0) {
1963 adj
= hmp
->free_reserved
/ mp
->mnt_vstat
.f_bsize
;
1964 tmp
.f_blocks
-= adj
;
1966 tmp
.f_bavail
-= adj
;
1969 mp
->mnt_stat
.f_blocks
= tmp
.f_blocks
;
1970 mp
->mnt_stat
.f_bfree
= tmp
.f_bfree
;
1971 mp
->mnt_stat
.f_bavail
= tmp
.f_bavail
;
1972 mp
->mnt_stat
.f_files
= tmp
.f_files
;
1973 mp
->mnt_stat
.f_ffree
= tmp
.f_ffree
;
1975 *sbp
= mp
->mnt_stat
;
1982 hammer2_vfs_statvfs(struct mount
*mp
, struct statvfs
*sbp
, struct ucred
*cred
)
1986 hammer2_blockref_t bref
;
1991 * NOTE: iroot might not have validated the cluster yet.
1994 bzero(&tmp
, sizeof(tmp
));
1996 for (i
= 0; i
< pmp
->iroot
->cluster
.nchains
; ++i
) {
1997 hmp
= pmp
->pfs_hmps
[i
];
2000 if (pmp
->iroot
->cluster
.array
[i
].chain
)
2001 bref
= pmp
->iroot
->cluster
.array
[i
].chain
->bref
;
2003 bzero(&bref
, sizeof(bref
));
2005 tmp
.f_files
= bref
.embed
.stats
.inode_count
;
2007 tmp
.f_blocks
= hmp
->voldata
.allocator_size
/
2008 mp
->mnt_vstat
.f_bsize
;
2009 tmp
.f_bfree
= hmp
->voldata
.allocator_free
/
2010 mp
->mnt_vstat
.f_bsize
;
2011 tmp
.f_bavail
= tmp
.f_bfree
;
2013 if (cred
&& cred
->cr_uid
!= 0) {
2017 adj
= hmp
->free_reserved
/ mp
->mnt_vstat
.f_bsize
;
2018 tmp
.f_blocks
-= adj
;
2020 tmp
.f_bavail
-= adj
;
2023 mp
->mnt_vstat
.f_blocks
= tmp
.f_blocks
;
2024 mp
->mnt_vstat
.f_bfree
= tmp
.f_bfree
;
2025 mp
->mnt_vstat
.f_bavail
= tmp
.f_bavail
;
2026 mp
->mnt_vstat
.f_files
= tmp
.f_files
;
2027 mp
->mnt_vstat
.f_ffree
= tmp
.f_ffree
;
2029 *sbp
= mp
->mnt_vstat
;
2035 * Mount-time recovery (RW mounts)
2037 * Updates to the free block table are allowed to lag flushes by one
2038 * transaction. In case of a crash, then on a fresh mount we must do an
2039 * incremental scan of the last committed transaction id and make sure that
2040 * all related blocks have been marked allocated.
2042 * The super-root topology and each PFS has its own transaction id domain,
2043 * so we must track PFS boundary transitions.
2045 struct hammer2_recovery_elm
{
2046 TAILQ_ENTRY(hammer2_recovery_elm
) entry
;
2047 hammer2_chain_t
*chain
;
2048 hammer2_tid_t sync_tid
;
2051 TAILQ_HEAD(hammer2_recovery_list
, hammer2_recovery_elm
);
2053 struct hammer2_recovery_info
{
2054 struct hammer2_recovery_list list
;
2059 static int hammer2_recovery_scan(hammer2_dev_t
*hmp
,
2060 hammer2_chain_t
*parent
,
2061 struct hammer2_recovery_info
*info
,
2062 hammer2_tid_t sync_tid
);
2064 #define HAMMER2_RECOVERY_MAXDEPTH 10
2068 hammer2_recovery(hammer2_dev_t
*hmp
)
2070 struct hammer2_recovery_info info
;
2071 struct hammer2_recovery_elm
*elm
;
2072 hammer2_chain_t
*parent
;
2073 hammer2_tid_t sync_tid
;
2074 hammer2_tid_t mirror_tid
;
2077 hammer2_trans_init(hmp
->spmp
, 0);
2079 sync_tid
= hmp
->voldata
.freemap_tid
;
2080 mirror_tid
= hmp
->voldata
.mirror_tid
;
2082 kprintf("hammer2 mount \"%s\": ", hmp
->devrepname
);
2083 if (sync_tid
>= mirror_tid
) {
2084 kprintf(" no recovery needed\n");
2086 kprintf(" freemap recovery %016jx-%016jx\n",
2087 sync_tid
+ 1, mirror_tid
);
2090 TAILQ_INIT(&info
.list
);
2092 parent
= hammer2_chain_lookup_init(&hmp
->vchain
, 0);
2093 error
= hammer2_recovery_scan(hmp
, parent
, &info
, sync_tid
);
2094 hammer2_chain_lookup_done(parent
);
2096 while ((elm
= TAILQ_FIRST(&info
.list
)) != NULL
) {
2097 TAILQ_REMOVE(&info
.list
, elm
, entry
);
2098 parent
= elm
->chain
;
2099 sync_tid
= elm
->sync_tid
;
2100 kfree(elm
, M_HAMMER2
);
2102 hammer2_chain_lock(parent
, HAMMER2_RESOLVE_ALWAYS
);
2103 error
|= hammer2_recovery_scan(hmp
, parent
, &info
,
2104 hmp
->voldata
.freemap_tid
);
2105 hammer2_chain_unlock(parent
);
2106 hammer2_chain_drop(parent
); /* drop elm->chain ref */
2108 hammer2_trans_done(hmp
->spmp
);
2115 hammer2_recovery_scan(hammer2_dev_t
*hmp
, hammer2_chain_t
*parent
,
2116 struct hammer2_recovery_info
*info
,
2117 hammer2_tid_t sync_tid
)
2119 const hammer2_inode_data_t
*ripdata
;
2120 hammer2_chain_t
*chain
;
2121 hammer2_blockref_t bref
;
2128 * Adjust freemap to ensure that the block(s) are marked allocated.
2130 if (parent
->bref
.type
!= HAMMER2_BREF_TYPE_VOLUME
) {
2131 hammer2_freemap_adjust(hmp
, &parent
->bref
,
2132 HAMMER2_FREEMAP_DORECOVER
);
2136 * Check type for recursive scan
2138 switch(parent
->bref
.type
) {
2139 case HAMMER2_BREF_TYPE_VOLUME
:
2140 /* data already instantiated */
2142 case HAMMER2_BREF_TYPE_INODE
:
2144 * Must instantiate data for DIRECTDATA test and also
2147 hammer2_chain_lock(parent
, HAMMER2_RESOLVE_ALWAYS
);
2148 ripdata
= &hammer2_chain_rdata(parent
)->ipdata
;
2149 if (ripdata
->meta
.op_flags
& HAMMER2_OPFLAG_DIRECTDATA
) {
2150 /* not applicable to recovery scan */
2151 hammer2_chain_unlock(parent
);
2154 hammer2_chain_unlock(parent
);
2156 case HAMMER2_BREF_TYPE_INDIRECT
:
2158 * Must instantiate data for recursion
2160 hammer2_chain_lock(parent
, HAMMER2_RESOLVE_ALWAYS
);
2161 hammer2_chain_unlock(parent
);
2163 case HAMMER2_BREF_TYPE_DIRENT
:
2164 case HAMMER2_BREF_TYPE_DATA
:
2165 case HAMMER2_BREF_TYPE_FREEMAP
:
2166 case HAMMER2_BREF_TYPE_FREEMAP_NODE
:
2167 case HAMMER2_BREF_TYPE_FREEMAP_LEAF
:
2168 /* not applicable to recovery scan */
2172 return HAMMER2_ERROR_BADBREF
;
2176 * Defer operation if depth limit reached or if we are crossing a
2179 if (info
->depth
>= HAMMER2_RECOVERY_MAXDEPTH
) {
2180 struct hammer2_recovery_elm
*elm
;
2182 elm
= kmalloc(sizeof(*elm
), M_HAMMER2
, M_ZERO
| M_WAITOK
);
2183 elm
->chain
= parent
;
2184 elm
->sync_tid
= sync_tid
;
2185 hammer2_chain_ref(parent
);
2186 TAILQ_INSERT_TAIL(&info
->list
, elm
, entry
);
2187 /* unlocked by caller */
2194 * Recursive scan of the last flushed transaction only. We are
2195 * doing this without pmp assignments so don't leave the chains
2196 * hanging around after we are done with them.
2198 * error Cumulative error this level only
2199 * rup_error Cumulative error for recursion
2200 * tmp_error Specific non-cumulative recursion error
2208 error
|= hammer2_chain_scan(parent
, &chain
, &bref
,
2210 HAMMER2_LOOKUP_NODATA
);
2213 * Problem during scan or EOF
2221 if (chain
== NULL
) {
2222 if (bref
.mirror_tid
> sync_tid
) {
2223 hammer2_freemap_adjust(hmp
, &bref
,
2224 HAMMER2_FREEMAP_DORECOVER
);
2230 * This may or may not be a recursive node.
2232 atomic_set_int(&chain
->flags
, HAMMER2_CHAIN_RELEASE
);
2233 if (bref
.mirror_tid
> sync_tid
) {
2235 tmp_error
= hammer2_recovery_scan(hmp
, chain
,
2243 * Flush the recovery at the PFS boundary to stage it for
2244 * the final flush of the super-root topology.
2246 if (tmp_error
== 0 &&
2247 (bref
.flags
& HAMMER2_BREF_FLAG_PFSROOT
) &&
2248 (chain
->flags
& HAMMER2_CHAIN_ONFLUSH
)) {
2249 hammer2_flush(chain
, HAMMER2_FLUSH_TOP
);
2251 rup_error
|= tmp_error
;
2253 return ((error
| rup_error
) & ~HAMMER2_ERROR_EOF
);
2257 * Sync a mount point; this is called on a per-mount basis from the
2258 * filesystem syncer process periodically and whenever a user issues
2262 hammer2_vfs_sync(struct mount
*mp
, int waitfor
)
2264 hammer2_xop_flush_t
*xop
;
2265 struct hammer2_sync_info info
;
2266 hammer2_inode_t
*iroot
;
2274 KKASSERT(iroot
->pmp
== pmp
);
2277 * We can't acquire locks on existing vnodes while in a transaction
2278 * without risking a deadlock. This assumes that vfsync() can be
2279 * called without the vnode locked (which it can in DragonFly).
2280 * Otherwise we'd have to implement a multi-pass or flag the lock
2281 * failures and retry.
2283 * The reclamation code interlocks with the sync list's token
2284 * (by removing the vnode from the scan list) before unlocking
2285 * the inode, giving us time to ref the inode.
2287 /*flags = VMSC_GETVP;*/
2289 if (waitfor
& MNT_LAZY
)
2290 flags
|= VMSC_ONEPASS
;
2293 * Preflush the vnodes using a normal transaction before interlocking
2294 * with a flush transaction. We do this to try to run as much of
2295 * the compression as possible outside the flush transaction.
2297 * For efficiency do an async pass before making sure with a
2298 * synchronous pass on all related buffer cache buffers.
2300 hammer2_trans_init(pmp
, 0);
2302 info
.waitfor
= MNT_NOWAIT
;
2303 vsyncscan(mp
, flags
| VMSC_NOWAIT
, hammer2_sync_scan2
, &info
);
2304 info
.waitfor
= MNT_WAIT
;
2305 vsyncscan(mp
, flags
, hammer2_sync_scan2
, &info
);
2306 hammer2_trans_done(pmp
);
2309 * Start our flush transaction. This does not return until all
2310 * concurrent transactions have completed and will prevent any
2311 * new transactions from running concurrently, except for the
2312 * buffer cache transactions.
2314 * (1) vfsync() all dirty vnodes via vfsyncscan().
2316 * (2) Flush any remaining dirty inodes (the sideq), including any
2317 * which may have been created during or raced against the
2318 * vfsync(). To catch all cases this must be done after the
2321 * (3) Wait for any pending BIO I/O to complete (hammer2_bioq_sync()).
2323 * NOTE! It is still possible for the paging code to push pages
2324 * out via a UIO_NOCOPY hammer2_vop_write() during the main
2327 hammer2_trans_init(pmp
, HAMMER2_TRANS_ISFLUSH
);
2330 info
.waitfor
= MNT_NOWAIT
;
2331 vsyncscan(mp
, flags
| VMSC_NOWAIT
, hammer2_sync_scan2
, &info
);
2332 info
.waitfor
= MNT_WAIT
;
2333 vsyncscan(mp
, flags
, hammer2_sync_scan2
, &info
);
2334 hammer2_inode_run_sideq(pmp
, 1);
2335 hammer2_bioq_sync(pmp
);
2338 * Use the XOP interface to concurrently flush all nodes to
2339 * synchronize the PFSROOT subtopology to the media. A standard
2340 * end-of-scan ENOENT error indicates cluster sufficiency.
2342 * Note that this flush will not be visible on crash recovery until
2343 * we flush the super-root topology in the next loop.
2345 * XXX For now wait for all flushes to complete.
2348 xop
= hammer2_xop_alloc(iroot
, HAMMER2_XOP_MODIFYING
);
2349 hammer2_xop_start(&xop
->head
, hammer2_inode_xop_flush
);
2350 error
= hammer2_xop_collect(&xop
->head
,
2351 HAMMER2_XOP_COLLECT_WAITALL
);
2352 hammer2_xop_retire(&xop
->head
, HAMMER2_XOPMASK_VOP
);
2353 if (error
== HAMMER2_ERROR_ENOENT
)
2356 error
= hammer2_error_to_errno(error
);
2360 hammer2_trans_done(pmp
);
2368 * Note that we ignore the tranasction mtid we got above. Instead,
2369 * each vfsync below will ultimately get its own via TRANS_BUFCACHE
2373 hammer2_sync_scan2(struct mount
*mp
, struct vnode
*vp
, void *data
)
2375 struct hammer2_sync_info
*info
= data
;
2376 hammer2_inode_t
*ip
;
2380 * Degenerate cases. Note that ip == NULL typically means the
2381 * syncer vnode itself and we don't want to vclrisdirty() in that
2388 if (vp
->v_type
== VNON
|| vp
->v_type
== VBAD
) {
2394 * VOP_FSYNC will start a new transaction so replicate some code
2395 * here to do it inline (see hammer2_vop_fsync()).
2397 * WARNING: The vfsync interacts with the buffer cache and might
2398 * block, we can't hold the inode lock at that time.
2399 * However, we MUST ref ip before blocking to ensure that
2400 * it isn't ripped out from under us (since we do not
2401 * hold a lock on the vnode).
2403 hammer2_inode_ref(ip
);
2404 if ((ip
->flags
& HAMMER2_INODE_MODIFIED
) ||
2405 !RB_EMPTY(&vp
->v_rbdirty_tree
)) {
2406 vfsync(vp
, info
->waitfor
, 1, NULL
, NULL
);
2407 if (ip
->flags
& (HAMMER2_INODE_RESIZED
|
2408 HAMMER2_INODE_MODIFIED
)) {
2409 hammer2_inode_lock(ip
, 0);
2410 if (ip
->flags
& (HAMMER2_INODE_RESIZED
|
2411 HAMMER2_INODE_MODIFIED
)) {
2412 hammer2_inode_chain_sync(ip
);
2414 hammer2_inode_unlock(ip
);
2417 if ((ip
->flags
& HAMMER2_INODE_MODIFIED
) == 0 &&
2418 RB_EMPTY(&vp
->v_rbdirty_tree
)) {
2422 hammer2_inode_drop(ip
);
2426 info
->error
= error
;
2433 hammer2_vfs_vptofh(struct vnode
*vp
, struct fid
*fhp
)
2435 hammer2_inode_t
*ip
;
2437 KKASSERT(MAXFIDSZ
>= 16);
2439 fhp
->fid_len
= offsetof(struct fid
, fid_data
[16]);
2441 ((hammer2_tid_t
*)fhp
->fid_data
)[0] = ip
->meta
.inum
;
2442 ((hammer2_tid_t
*)fhp
->fid_data
)[1] = 0;
2449 hammer2_vfs_fhtovp(struct mount
*mp
, struct vnode
*rootvp
,
2450 struct fid
*fhp
, struct vnode
**vpp
)
2457 inum
= ((hammer2_tid_t
*)fhp
->fid_data
)[0] & HAMMER2_DIRHASH_USERMSK
;
2460 error
= hammer2_vfs_root(mp
, vpp
);
2462 error
= hammer2_vfs_vget(mp
, NULL
, inum
, vpp
);
2467 kprintf("fhtovp: %016jx -> %p, %d\n", inum
, *vpp
, error
);
2473 hammer2_vfs_checkexp(struct mount
*mp
, struct sockaddr
*nam
,
2474 int *exflagsp
, struct ucred
**credanonp
)
2481 np
= vfs_export_lookup(mp
, &pmp
->export
, nam
);
2483 *exflagsp
= np
->netc_exflags
;
2484 *credanonp
= &np
->netc_anon
;
2493 * Support code for hammer2_vfs_mount(). Read, verify, and install the volume
2494 * header into the HMP
2496 * XXX read four volhdrs and use the one with the highest TID whos CRC
2501 * XXX For filesystems w/ less than 4 volhdrs, make sure to not write to
2502 * nonexistant locations.
2504 * XXX Record selected volhdr and ring updates to each of 4 volhdrs
2508 hammer2_install_volume_header(hammer2_dev_t
*hmp
)
2510 hammer2_volume_data_t
*vd
;
2512 hammer2_crc32_t crc0
, crc
, bcrc0
, bcrc
;
2524 * There are up to 4 copies of the volume header (syncs iterate
2525 * between them so there is no single master). We don't trust the
2526 * volu_size field so we don't know precisely how large the filesystem
2527 * is, so depend on the OS to return an error if we go beyond the
2528 * block device's EOF.
2530 for (i
= 0; i
< HAMMER2_NUM_VOLHDRS
; i
++) {
2531 error
= bread(hmp
->devvp
, i
* HAMMER2_ZONE_BYTES64
,
2532 HAMMER2_VOLUME_BYTES
, &bp
);
2539 vd
= (struct hammer2_volume_data
*) bp
->b_data
;
2540 if ((vd
->magic
!= HAMMER2_VOLUME_ID_HBO
) &&
2541 (vd
->magic
!= HAMMER2_VOLUME_ID_ABO
)) {
2547 if (vd
->magic
== HAMMER2_VOLUME_ID_ABO
) {
2548 /* XXX: Reversed-endianness filesystem */
2549 kprintf("hammer2: reverse-endian filesystem detected");
2555 crc
= vd
->icrc_sects
[HAMMER2_VOL_ICRC_SECT0
];
2556 crc0
= hammer2_icrc32(bp
->b_data
+ HAMMER2_VOLUME_ICRC0_OFF
,
2557 HAMMER2_VOLUME_ICRC0_SIZE
);
2558 bcrc
= vd
->icrc_sects
[HAMMER2_VOL_ICRC_SECT1
];
2559 bcrc0
= hammer2_icrc32(bp
->b_data
+ HAMMER2_VOLUME_ICRC1_OFF
,
2560 HAMMER2_VOLUME_ICRC1_SIZE
);
2561 if ((crc0
!= crc
) || (bcrc0
!= bcrc
)) {
2562 kprintf("hammer2 volume header crc "
2563 "mismatch copy #%d %08x/%08x\n",
2570 if (valid
== 0 || hmp
->voldata
.mirror_tid
< vd
->mirror_tid
) {
2579 hmp
->volsync
= hmp
->voldata
;
2580 hmp
->free_reserved
= hmp
->voldata
.allocator_size
/ 20;
2582 if (error_reported
|| bootverbose
|| 1) { /* 1/DEBUG */
2583 kprintf("hammer2: using volume header #%d\n",
2588 kprintf("hammer2: no valid volume headers found!\n");
2594 * This handles hysteresis on regular file flushes. Because the BIOs are
2595 * routed to a thread it is possible for an excessive number to build up
2596 * and cause long front-end stalls long before the runningbuffspace limit
2597 * is hit, so we implement hammer2_flush_pipe to control the
2600 * This is a particular problem when compression is used.
2603 hammer2_lwinprog_ref(hammer2_pfs_t
*pmp
)
2605 atomic_add_int(&pmp
->count_lwinprog
, 1);
2609 hammer2_lwinprog_drop(hammer2_pfs_t
*pmp
)
2613 lwinprog
= atomic_fetchadd_int(&pmp
->count_lwinprog
, -1);
2614 if ((lwinprog
& HAMMER2_LWINPROG_WAITING
) &&
2615 (lwinprog
& HAMMER2_LWINPROG_MASK
) <= hammer2_flush_pipe
* 2 / 3) {
2616 atomic_clear_int(&pmp
->count_lwinprog
,
2617 HAMMER2_LWINPROG_WAITING
);
2618 wakeup(&pmp
->count_lwinprog
);
2620 if ((lwinprog
& HAMMER2_LWINPROG_WAITING0
) &&
2621 (lwinprog
& HAMMER2_LWINPROG_MASK
) <= 0) {
2622 atomic_clear_int(&pmp
->count_lwinprog
,
2623 HAMMER2_LWINPROG_WAITING0
);
2624 wakeup(&pmp
->count_lwinprog
);
2629 hammer2_lwinprog_wait(hammer2_pfs_t
*pmp
, int flush_pipe
)
2632 int lwflag
= (flush_pipe
) ? HAMMER2_LWINPROG_WAITING
:
2633 HAMMER2_LWINPROG_WAITING0
;
2636 lwinprog
= pmp
->count_lwinprog
;
2638 if ((lwinprog
& HAMMER2_LWINPROG_MASK
) <= flush_pipe
)
2640 tsleep_interlock(&pmp
->count_lwinprog
, 0);
2641 atomic_set_int(&pmp
->count_lwinprog
, lwflag
);
2642 lwinprog
= pmp
->count_lwinprog
;
2643 if ((lwinprog
& HAMMER2_LWINPROG_MASK
) <= flush_pipe
)
2645 tsleep(&pmp
->count_lwinprog
, PINTERLOCKED
, "h2wpipe", hz
);
2650 * Manage excessive memory resource use for chain and related
2654 hammer2_pfs_memory_wait(hammer2_pfs_t
*pmp
)
2664 * Atomic check condition and wait. Also do an early speedup of
2665 * the syncer to try to avoid hitting the wait.
2668 waiting
= pmp
->inmem_dirty_chains
;
2670 count
= waiting
& HAMMER2_DIRTYCHAIN_MASK
;
2672 limit
= pmp
->mp
->mnt_nvnodelistsize
/ 10;
2673 if (limit
< hammer2_limit_dirty_chains
)
2674 limit
= hammer2_limit_dirty_chains
;
2679 if ((int)(ticks
- zzticks
) > hz
) {
2681 kprintf("count %ld %ld\n", count
, limit
);
2686 * Block if there are too many dirty chains present, wait
2687 * for the flush to clean some out.
2689 if (count
> limit
) {
2690 tsleep_interlock(&pmp
->inmem_dirty_chains
, 0);
2691 if (atomic_cmpset_int(&pmp
->inmem_dirty_chains
,
2693 waiting
| HAMMER2_DIRTYCHAIN_WAITING
)) {
2694 speedup_syncer(pmp
->mp
);
2695 tsleep(&pmp
->inmem_dirty_chains
, PINTERLOCKED
,
2698 continue; /* loop on success or fail */
2702 * Try to start an early flush before we are forced to block.
2704 if (count
> limit
* 7 / 10)
2705 speedup_syncer(pmp
->mp
);
2711 hammer2_pfs_memory_inc(hammer2_pfs_t
*pmp
)
2714 atomic_add_int(&pmp
->inmem_dirty_chains
, 1);
2719 hammer2_pfs_memory_wakeup(hammer2_pfs_t
*pmp
)
2724 waiting
= atomic_fetchadd_int(&pmp
->inmem_dirty_chains
, -1);
2725 /* don't need --waiting to test flag */
2726 if (waiting
& HAMMER2_DIRTYCHAIN_WAITING
) {
2727 atomic_clear_int(&pmp
->inmem_dirty_chains
,
2728 HAMMER2_DIRTYCHAIN_WAITING
);
2729 wakeup(&pmp
->inmem_dirty_chains
);
2735 * Returns 0 if the filesystem has tons of free space
2736 * Returns 1 if the filesystem has less than 10% remaining
2737 * Returns 2 if the filesystem has less than 2%/5% (user/root) remaining.
2740 hammer2_vfs_enospace(hammer2_inode_t
*ip
, off_t bytes
, struct ucred
*cred
)
2744 hammer2_off_t free_reserved
;
2745 hammer2_off_t free_nominal
;
2750 if (pmp
->free_ticks
== 0 || pmp
->free_ticks
!= ticks
) {
2751 free_reserved
= HAMMER2_SEGSIZE
;
2752 free_nominal
= 0x7FFFFFFFFFFFFFFFLLU
;
2753 for (i
= 0; i
< pmp
->iroot
->cluster
.nchains
; ++i
) {
2754 hmp
= pmp
->pfs_hmps
[i
];
2757 if (pmp
->pfs_types
[i
] != HAMMER2_PFSTYPE_MASTER
&&
2758 pmp
->pfs_types
[i
] != HAMMER2_PFSTYPE_SOFT_MASTER
)
2761 if (free_nominal
> hmp
->voldata
.allocator_free
)
2762 free_nominal
= hmp
->voldata
.allocator_free
;
2763 if (free_reserved
< hmp
->free_reserved
)
2764 free_reserved
= hmp
->free_reserved
;
2770 pmp
->free_reserved
= free_reserved
;
2771 pmp
->free_nominal
= free_nominal
;
2772 pmp
->free_ticks
= ticks
;
2774 free_reserved
= pmp
->free_reserved
;
2775 free_nominal
= pmp
->free_nominal
;
2777 if (cred
&& cred
->cr_uid
!= 0) {
2778 if ((int64_t)(free_nominal
- bytes
) <
2779 (int64_t)free_reserved
) {
2783 if ((int64_t)(free_nominal
- bytes
) <
2784 (int64_t)free_reserved
/ 2) {
2788 if ((int64_t)(free_nominal
- bytes
) < (int64_t)free_reserved
* 2)
2797 hammer2_dump_chain(hammer2_chain_t
*chain
, int tab
, int *countp
, char pfx
)
2799 hammer2_chain_t
*scan
;
2800 hammer2_chain_t
*parent
;
2804 kprintf("%*.*s...\n", tab
, tab
, "");
2809 kprintf("%*.*s%c-chain %p.%d %016jx/%d mir=%016jx\n",
2811 chain
, chain
->bref
.type
,
2812 chain
->bref
.key
, chain
->bref
.keybits
,
2813 chain
->bref
.mirror_tid
);
2815 kprintf("%*.*s [%08x] (%s) refs=%d",
2818 ((chain
->bref
.type
== HAMMER2_BREF_TYPE_INODE
&&
2819 chain
->data
) ? (char *)chain
->data
->ipdata
.filename
: "?"),
2822 parent
= chain
->parent
;
2824 kprintf("\n%*.*s p=%p [pflags %08x prefs %d",
2826 parent
, parent
->flags
, parent
->refs
);
2827 if (RB_EMPTY(&chain
->core
.rbtree
)) {
2831 RB_FOREACH(scan
, hammer2_chain_tree
, &chain
->core
.rbtree
)
2832 hammer2_dump_chain(scan
, tab
+ 4, countp
, 'a');
2833 if (chain
->bref
.type
== HAMMER2_BREF_TYPE_INODE
&& chain
->data
)
2834 kprintf("%*.*s}(%s)\n", tab
, tab
, "",
2835 chain
->data
->ipdata
.filename
);
2837 kprintf("%*.*s}\n", tab
, tab
, "");