?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
hammer2 - Stabilization for cluster code, features
[dragonfly.git] / sys / vfs / hammer2 / hammer2_vfsops.c
blobb47c79f627f0abd7277ff602dda8601b5ac1d7bf
1 /*
2 * Copyright (c) 2011-2015 The DragonFly Project. All rights reserved.
3 *
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)
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 *
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
17 * distribution.
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.
21 *
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
33 * SUCH DAMAGE.
34 */
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>
42 #include <sys/buf.h>
43 #include <sys/uuid.h>
44 #include <sys/vfsops.h>
45 #include <sys/sysctl.h>
46 #include <sys/socket.h>
47 #include <sys/objcache.h>
48
49 #include <sys/proc.h>
50 #include <sys/namei.h>
51 #include <sys/mountctl.h>
52 #include <sys/dirent.h>
53 #include <sys/uio.h>
54
55 #include <sys/mutex.h>
56 #include <sys/mutex2.h>
57
58 #include "hammer2.h"
59 #include "hammer2_disk.h"
60 #include "hammer2_mount.h"
61 #include "hammer2_lz4.h"
62
63 #include "zlib/hammer2_zlib.h"
64
65 #define REPORT_REFS_ERRORS 1 /* XXX remove me */
66
67 MALLOC_DEFINE(M_OBJCACHE, "objcache", "Object Cache");
68
69 struct hammer2_sync_info {
70 int error;
71 int waitfor;
72 };
73
74 TAILQ_HEAD(hammer2_mntlist, hammer2_dev);
75 TAILQ_HEAD(hammer2_pfslist, hammer2_pfs);
76 static struct hammer2_mntlist hammer2_mntlist;
77 static struct hammer2_pfslist hammer2_pfslist;
78 static struct lock hammer2_mntlk;
79
80 int hammer2_debug;
81 int hammer2_cluster_read = 4; /* physical read-ahead */
82 int hammer2_cluster_write = 0; /* bdwrite() so later inval works */
83 int hammer2_dedup_enable = 1;
84 int hammer2_inval_enable = 0;
85 int hammer2_flush_pipe = 100;
86 int hammer2_synchronous_flush = 1;
87 int hammer2_dio_count;
88 long hammer2_chain_allocs;
89 long hammer2_chain_frees;
90 long hammer2_limit_dirty_chains;
91 long hammer2_count_modified_chains;
92 long hammer2_iod_invals;
93 long hammer2_iod_file_read;
94 long hammer2_iod_meta_read;
95 long hammer2_iod_indr_read;
96 long hammer2_iod_fmap_read;
97 long hammer2_iod_volu_read;
98 long hammer2_iod_file_write;
99 long hammer2_iod_file_wembed;
100 long hammer2_iod_file_wzero;
101 long hammer2_iod_file_wdedup;
102 long hammer2_iod_meta_write;
103 long hammer2_iod_indr_write;
104 long hammer2_iod_fmap_write;
105 long hammer2_iod_volu_write;
106
107 MALLOC_DECLARE(M_HAMMER2_CBUFFER);
108 MALLOC_DEFINE(M_HAMMER2_CBUFFER, "HAMMER2-compbuffer",
109 "Buffer used for compression.");
110
111 MALLOC_DECLARE(M_HAMMER2_DEBUFFER);
112 MALLOC_DEFINE(M_HAMMER2_DEBUFFER, "HAMMER2-decompbuffer",
113 "Buffer used for decompression.");
114
115 SYSCTL_NODE(_vfs, OID_AUTO, hammer2, CTLFLAG_RW, 0, "HAMMER2 filesystem");
116
117 SYSCTL_INT(_vfs_hammer2, OID_AUTO, debug, CTLFLAG_RW,
118 &hammer2_debug, 0, "");
119 SYSCTL_INT(_vfs_hammer2, OID_AUTO, cluster_read, CTLFLAG_RW,
120 &hammer2_cluster_read, 0, "");
121 SYSCTL_INT(_vfs_hammer2, OID_AUTO, cluster_write, CTLFLAG_RW,
122 &hammer2_cluster_write, 0, "");
123 SYSCTL_INT(_vfs_hammer2, OID_AUTO, dedup_enable, CTLFLAG_RW,
124 &hammer2_dedup_enable, 0, "");
125 SYSCTL_INT(_vfs_hammer2, OID_AUTO, inval_enable, CTLFLAG_RW,
126 &hammer2_inval_enable, 0, "");
127 SYSCTL_INT(_vfs_hammer2, OID_AUTO, flush_pipe, CTLFLAG_RW,
128 &hammer2_flush_pipe, 0, "");
129 SYSCTL_INT(_vfs_hammer2, OID_AUTO, synchronous_flush, CTLFLAG_RW,
130 &hammer2_synchronous_flush, 0, "");
131 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, chain_allocs, CTLFLAG_RW,
132 &hammer2_chain_allocs, 0, "");
133 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, chain_frees, CTLFLAG_RW,
134 &hammer2_chain_frees, 0, "");
135 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, limit_dirty_chains, CTLFLAG_RW,
136 &hammer2_limit_dirty_chains, 0, "");
137 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, count_modified_chains, CTLFLAG_RW,
138 &hammer2_count_modified_chains, 0, "");
139 SYSCTL_INT(_vfs_hammer2, OID_AUTO, dio_count, CTLFLAG_RD,
140 &hammer2_dio_count, 0, "");
141
142 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_invals, CTLFLAG_RW,
143 &hammer2_iod_invals, 0, "");
144 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_file_read, CTLFLAG_RW,
145 &hammer2_iod_file_read, 0, "");
146 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_meta_read, CTLFLAG_RW,
147 &hammer2_iod_meta_read, 0, "");
148 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_indr_read, CTLFLAG_RW,
149 &hammer2_iod_indr_read, 0, "");
150 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_fmap_read, CTLFLAG_RW,
151 &hammer2_iod_fmap_read, 0, "");
152 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_volu_read, CTLFLAG_RW,
153 &hammer2_iod_volu_read, 0, "");
154
155 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_file_write, CTLFLAG_RW,
156 &hammer2_iod_file_write, 0, "");
157 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_file_wembed, CTLFLAG_RW,
158 &hammer2_iod_file_wembed, 0, "");
159 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_file_wzero, CTLFLAG_RW,
160 &hammer2_iod_file_wzero, 0, "");
161 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_file_wdedup, CTLFLAG_RW,
162 &hammer2_iod_file_wdedup, 0, "");
163 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_meta_write, CTLFLAG_RW,
164 &hammer2_iod_meta_write, 0, "");
165 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_indr_write, CTLFLAG_RW,
166 &hammer2_iod_indr_write, 0, "");
167 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_fmap_write, CTLFLAG_RW,
168 &hammer2_iod_fmap_write, 0, "");
169 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_volu_write, CTLFLAG_RW,
170 &hammer2_iod_volu_write, 0, "");
171
172 long hammer2_check_icrc32;
173 long hammer2_check_xxhash64;
174 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, check_icrc32, CTLFLAG_RW,
175 &hammer2_check_icrc32, 0, "");
176 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, check_xxhash64, CTLFLAG_RW,
177 &hammer2_check_xxhash64, 0, "");
178
179 static int hammer2_vfs_init(struct vfsconf *conf);
180 static int hammer2_vfs_uninit(struct vfsconf *vfsp);
181 static int hammer2_vfs_mount(struct mount *mp, char *path, caddr_t data,
182 struct ucred *cred);
183 static int hammer2_remount(hammer2_dev_t *, struct mount *, char *,
184 struct vnode *, struct ucred *);
185 static int hammer2_recovery(hammer2_dev_t *hmp);
186 static int hammer2_vfs_unmount(struct mount *mp, int mntflags);
187 static int hammer2_vfs_root(struct mount *mp, struct vnode **vpp);
188 static int hammer2_vfs_statfs(struct mount *mp, struct statfs *sbp,
189 struct ucred *cred);
190 static int hammer2_vfs_statvfs(struct mount *mp, struct statvfs *sbp,
191 struct ucred *cred);
192 static int hammer2_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
193 struct fid *fhp, struct vnode **vpp);
194 static int hammer2_vfs_vptofh(struct vnode *vp, struct fid *fhp);
195 static int hammer2_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
196 int *exflagsp, struct ucred **credanonp);
197
198 static int hammer2_install_volume_header(hammer2_dev_t *hmp);
199 static int hammer2_sync_scan2(struct mount *mp, struct vnode *vp, void *data);
200
201 static void hammer2_update_pmps(hammer2_dev_t *hmp);
202
203 static void hammer2_mount_helper(struct mount *mp, hammer2_pfs_t *pmp);
204 static void hammer2_unmount_helper(struct mount *mp, hammer2_pfs_t *pmp,
205 hammer2_dev_t *hmp);
206
207 /*
208 * HAMMER2 vfs operations.
209 */
210 static struct vfsops hammer2_vfsops = {
211 .vfs_init = hammer2_vfs_init,
212 .vfs_uninit = hammer2_vfs_uninit,
213 .vfs_sync = hammer2_vfs_sync,
214 .vfs_mount = hammer2_vfs_mount,
215 .vfs_unmount = hammer2_vfs_unmount,
216 .vfs_root = hammer2_vfs_root,
217 .vfs_statfs = hammer2_vfs_statfs,
218 .vfs_statvfs = hammer2_vfs_statvfs,
219 .vfs_vget = hammer2_vfs_vget,
220 .vfs_vptofh = hammer2_vfs_vptofh,
221 .vfs_fhtovp = hammer2_vfs_fhtovp,
222 .vfs_checkexp = hammer2_vfs_checkexp
223 };
224
225 MALLOC_DEFINE(M_HAMMER2, "HAMMER2-mount", "");
226
227 VFS_SET(hammer2_vfsops, hammer2, VFCF_MPSAFE);
228 MODULE_VERSION(hammer2, 1);
229
230 static
231 int
232 hammer2_vfs_init(struct vfsconf *conf)
233 {
234 static struct objcache_malloc_args margs_read;
235 static struct objcache_malloc_args margs_write;
236 static struct objcache_malloc_args margs_vop;
237
238 int error;
239
240 error = 0;
241
242 if (HAMMER2_BLOCKREF_BYTES != sizeof(struct hammer2_blockref))
243 error = EINVAL;
244 if (HAMMER2_INODE_BYTES != sizeof(struct hammer2_inode_data))
245 error = EINVAL;
246 if (HAMMER2_VOLUME_BYTES != sizeof(struct hammer2_volume_data))
247 error = EINVAL;
248
249 if (error)
250 kprintf("HAMMER2 structure size mismatch; cannot continue.\n");
251
252 margs_read.objsize = 65536;
253 margs_read.mtype = M_HAMMER2_DEBUFFER;
254
255 margs_write.objsize = 32768;
256 margs_write.mtype = M_HAMMER2_CBUFFER;
257
258 margs_vop.objsize = sizeof(hammer2_xop_t);
259 margs_vop.mtype = M_HAMMER2;
260
261 /*
262 * Note thaht for the XOPS cache we want backing store allocations
263 * to use M_ZERO. This is not allowed in objcache_get() (to avoid
264 * confusion), so use the backing store function that does it. This
265 * means that initial XOPS objects are zerod but REUSED objects are
266 * not. So we are responsible for cleaning the object up sufficiently
267 * for our needs before objcache_put()ing it back (typically just the
268 * FIFO indices).
269 */
270 cache_buffer_read = objcache_create(margs_read.mtype->ks_shortdesc,
271 0, 1, NULL, NULL, NULL,
272 objcache_malloc_alloc,
273 objcache_malloc_free,
274 &margs_read);
275 cache_buffer_write = objcache_create(margs_write.mtype->ks_shortdesc,
276 0, 1, NULL, NULL, NULL,
277 objcache_malloc_alloc,
278 objcache_malloc_free,
279 &margs_write);
280 cache_xops = objcache_create(margs_vop.mtype->ks_shortdesc,
281 0, 1, NULL, NULL, NULL,
282 objcache_malloc_alloc_zero,
283 objcache_malloc_free,
284 &margs_vop);
285
286
287 lockinit(&hammer2_mntlk, "mntlk", 0, 0);
288 TAILQ_INIT(&hammer2_mntlist);
289 TAILQ_INIT(&hammer2_pfslist);
290
291 hammer2_limit_dirty_chains = maxvnodes / 10;
292 if (hammer2_limit_dirty_chains > HAMMER2_LIMIT_DIRTY_CHAINS)
293 hammer2_limit_dirty_chains = HAMMER2_LIMIT_DIRTY_CHAINS;
294
295 return (error);
296 }
297
298 static
299 int
300 hammer2_vfs_uninit(struct vfsconf *vfsp __unused)
301 {
302 objcache_destroy(cache_buffer_read);
303 objcache_destroy(cache_buffer_write);
304 objcache_destroy(cache_xops);
305 return 0;
306 }
307
308 /*
309 * Core PFS allocator. Used to allocate or reference the pmp structure
310 * for PFS cluster mounts and the spmp structure for media (hmp) structures.
311 * The pmp can be passed in or loaded by this function using the chain and
312 * inode data.
313 *
314 * pmp->modify_tid tracks new modify_tid transaction ids for front-end
315 * transactions. Note that synchronization does not use this field.
316 * (typically frontend operations and synchronization cannot run on the
317 * same PFS node at the same time).
318 *
319 * XXX check locking
320 */
321 hammer2_pfs_t *
322 hammer2_pfsalloc(hammer2_chain_t *chain,
323 const hammer2_inode_data_t *ripdata,
324 hammer2_tid_t modify_tid, hammer2_dev_t *force_local)
325 {
326 hammer2_pfs_t *pmp;
327 hammer2_inode_t *iroot;
328 int count;
329 int i;
330 int j;
331
332 pmp = NULL;
333
334 /*
335 * Locate or create the PFS based on the cluster id. If ripdata
336 * is NULL this is a spmp which is unique and is always allocated.
337 *
338 * If the device is mounted in local mode all PFSs are considered
339 * independent and not part of any cluster (for debugging only).
340 */
341 if (ripdata) {
342 TAILQ_FOREACH(pmp, &hammer2_pfslist, mntentry) {
343 if (force_local != pmp->force_local)
344 continue;
345 if (force_local == NULL &&
346 bcmp(&pmp->pfs_clid, &ripdata->meta.pfs_clid,
347 sizeof(pmp->pfs_clid)) == 0) {
348 break;
349 } else if (force_local && pmp->pfs_names[0] &&
350 strcmp(pmp->pfs_names[0], ripdata->filename) == 0) {
351 break;
352 }
353 }
354 }
355
356 if (pmp == NULL) {
357 pmp = kmalloc(sizeof(*pmp), M_HAMMER2, M_WAITOK | M_ZERO);
358 pmp->force_local = force_local;
359 hammer2_trans_manage_init(pmp);
360 kmalloc_create(&pmp->minode, "HAMMER2-inodes");
361 kmalloc_create(&pmp->mmsg, "HAMMER2-pfsmsg");
362 lockinit(&pmp->lock, "pfslk", 0, 0);
363 lockinit(&pmp->lock_nlink, "h2nlink", 0, 0);
364 spin_init(&pmp->inum_spin, "hm2pfsalloc_inum");
365 spin_init(&pmp->xop_spin, "h2xop");
366 spin_init(&pmp->lru_spin, "h2lru");
367 RB_INIT(&pmp->inum_tree);
368 TAILQ_INIT(&pmp->sideq);
369 TAILQ_INIT(&pmp->lru_list);
370 spin_init(&pmp->list_spin, "hm2pfsalloc_list");
371
372 /*
373 * Distribute backend operations to threads
374 */
375 for (i = 0; i < HAMMER2_XOPGROUPS; ++i)
376 hammer2_xop_group_init(pmp, &pmp->xop_groups[i]);
377
378 /*
379 * Save the last media transaction id for the flusher. Set
380 * initial
381 */
382 if (ripdata)
383 pmp->pfs_clid = ripdata->meta.pfs_clid;
384 TAILQ_INSERT_TAIL(&hammer2_pfslist, pmp, mntentry);
385
386 /*
387 * The synchronization thread may start too early, make
388 * sure it stays frozen until we are ready to let it go.
389 * XXX
390 */
391 /*
392 pmp->primary_thr.flags = HAMMER2_THREAD_FROZEN |
393 HAMMER2_THREAD_REMASTER;
394 */
395 }
396
397 /*
398 * Create the PFS's root inode and any missing XOP helper threads.
399 */
400 if ((iroot = pmp->iroot) == NULL) {
401 iroot = hammer2_inode_get(pmp, NULL, NULL, -1);
402 pmp->iroot = iroot;
403 hammer2_inode_ref(iroot);
404 hammer2_inode_unlock(iroot);
405 }
406
407 /*
408 * Stop here if no chain is passed in.
409 */
410 if (chain == NULL)
411 goto done;
412
413 /*
414 * When a chain is passed in we must add it to the PFS's root
415 * inode, update pmp->pfs_types[], and update the syncronization
416 * threads.
417 *
418 * When forcing local mode, mark the PFS as a MASTER regardless.
419 *
420 * At the moment empty spots can develop due to removals or failures.
421 * Ultimately we want to re-fill these spots but doing so might
422 * confused running code. XXX
423 */
424 hammer2_inode_ref(iroot);
425 hammer2_mtx_ex(&iroot->lock);
426 j = iroot->cluster.nchains;
427
428 kprintf("add PFS to pmp %p[%d]\n", pmp, j);
429
430 if (j == HAMMER2_MAXCLUSTER) {
431 kprintf("hammer2_mount: cluster full!\n");
432 /* XXX fatal error? */
433 } else {
434 KKASSERT(chain->pmp == NULL);
435 chain->pmp = pmp;
436 hammer2_chain_ref(chain);
437 iroot->cluster.array[j].chain = chain;
438 if (force_local)
439 pmp->pfs_types[j] = HAMMER2_PFSTYPE_MASTER;
440 else
441 pmp->pfs_types[j] = ripdata->meta.pfs_type;
442 pmp->pfs_names[j] = kstrdup(ripdata->filename, M_HAMMER2);
443 pmp->pfs_hmps[j] = chain->hmp;
444
445 /*
446 * If the PFS is already mounted we must account
447 * for the mount_count here.
448 */
449 if (pmp->mp)
450 ++chain->hmp->mount_count;
451
452 /*
453 * May have to fixup dirty chain tracking. Previous
454 * pmp was NULL so nothing to undo.
455 */
456 if (chain->flags & HAMMER2_CHAIN_MODIFIED)
457 hammer2_pfs_memory_inc(pmp);
458 ++j;
459 }
460 iroot->cluster.nchains = j;
461
462 /*
463 * Update nmasters from any PFS inode which is part of the cluster.
464 * It is possible that this will result in a value which is too
465 * high. MASTER PFSs are authoritative for pfs_nmasters and will
466 * override this value later on.
467 *
468 * (This informs us of masters that might not currently be
469 * discoverable by this mount).
470 */
471 if (ripdata && pmp->pfs_nmasters < ripdata->meta.pfs_nmasters) {
472 pmp->pfs_nmasters = ripdata->meta.pfs_nmasters;
473 }
474
475 /*
476 * Count visible masters. Masters are usually added with
477 * ripdata->meta.pfs_nmasters set to 1. This detects when there
478 * are more (XXX and must update the master inodes).
479 */
480 count = 0;
481 for (i = 0; i < iroot->cluster.nchains; ++i) {
482 if (pmp->pfs_types[i] == HAMMER2_PFSTYPE_MASTER)
483 ++count;
484 }
485 if (pmp->pfs_nmasters < count)
486 pmp->pfs_nmasters = count;
487
488 /*
489 * Create missing synchronization and support threads.
490 *
491 * Single-node masters (including snapshots) have nothing to
492 * synchronize and do not require this thread.
493 *
494 * Multi-node masters or any number of soft masters, slaves, copy,
495 * or other PFS types need the thread.
496 *
497 * Each thread is responsible for its particular cluster index.
498 * We use independent threads so stalls or mismatches related to
499 * any given target do not affect other targets.
500 */
501 for (i = 0; i < iroot->cluster.nchains; ++i) {
502 /*
503 * Single-node masters (including snapshots) have nothing
504 * to synchronize and will make direct xops support calls,
505 * thus they do not require this thread.
506 *
507 * Note that there can be thousands of snapshots. We do not
508 * want to create thousands of threads.
509 */
510 if (pmp->pfs_nmasters <= 1 &&
511 pmp->pfs_types[i] == HAMMER2_PFSTYPE_MASTER) {
512 continue;
513 }
514
515 /*
516 * Sync support thread
517 */
518 if (pmp->sync_thrs[i].td == NULL) {
519 hammer2_thr_create(&pmp->sync_thrs[i], pmp,
520 "h2nod", i, -1,
521 hammer2_primary_sync_thread);
522 }
523 }
524
525 /*
526 * Create missing Xop threads
527 *
528 * NOTE: We create helper threads for all mounted PFSs or any
529 * PFSs with 2+ nodes (so the sync thread can update them,
530 * even if not mounted).
531 */
532 if (pmp->mp || iroot->cluster.nchains >= 2)
533 hammer2_xop_helper_create(pmp);
534
535 hammer2_mtx_unlock(&iroot->lock);
536 hammer2_inode_drop(iroot);
537 done:
538 return pmp;
539 }
540
541 /*
542 * Destroy a PFS, typically only occurs after the last mount on a device
543 * has gone away.
544 */
545 static void
546 hammer2_pfsfree(hammer2_pfs_t *pmp)
547 {
548 hammer2_inode_t *iroot;
549 hammer2_chain_t *chain;
550 int i;
551 int j;
552
553 /*
554 * Cleanup our reference on iroot. iroot is (should) not be needed
555 * by the flush code.
556 */
557 TAILQ_REMOVE(&hammer2_pfslist, pmp, mntentry);
558
559 iroot = pmp->iroot;
560 if (iroot) {
561 for (i = 0; i < iroot->cluster.nchains; ++i) {
562 hammer2_thr_delete(&pmp->sync_thrs[i]);
563 for (j = 0; j < HAMMER2_XOPGROUPS; ++j)
564 hammer2_thr_delete(&pmp->xop_groups[j].thrs[i]);
565 }
566 #if REPORT_REFS_ERRORS
567 if (pmp->iroot->refs != 1)
568 kprintf("PMP->IROOT %p REFS WRONG %d\n",
569 pmp->iroot, pmp->iroot->refs);
570 #else
571 KKASSERT(pmp->iroot->refs == 1);
572 #endif
573 /* ref for pmp->iroot */
574 hammer2_inode_drop(pmp->iroot);
575 pmp->iroot = NULL;
576 }
577
578 /*
579 * Cleanup chains remaining on LRU list.
580 */
581 kprintf("pfsfree: %p lrucount=%d\n", pmp, pmp->lru_count);
582 while ((chain = TAILQ_FIRST(&pmp->lru_list)) != NULL) {
583 hammer2_chain_ref(chain);
584 atomic_set_int(&chain->flags, HAMMER2_CHAIN_RELEASE);
585 hammer2_chain_drop(chain);
586 }
587
588 /*
589 * Free remaining pmp resources
590 */
591 kmalloc_destroy(&pmp->mmsg);
592 kmalloc_destroy(&pmp->minode);
593
594 kfree(pmp, M_HAMMER2);
595 }
596
597 /*
598 * Remove all references to hmp from the pfs list. Any PFS which becomes
599 * empty is terminated and freed.
600 *
601 * XXX inefficient.
602 */
603 static void
604 hammer2_pfsfree_scan(hammer2_dev_t *hmp)
605 {
606 hammer2_pfs_t *pmp;
607 hammer2_inode_t *iroot;
608 hammer2_chain_t *rchain;
609 int didfreeze;
610 int i;
611 int j;
612
613 again:
614 TAILQ_FOREACH(pmp, &hammer2_pfslist, mntentry) {
615 if ((iroot = pmp->iroot) == NULL)
616 continue;
617 if (hmp->spmp == pmp) {
618 kprintf("unmount hmp %p remove spmp %p\n",
619 hmp, pmp);
620 hmp->spmp = NULL;
621 }
622
623 /*
624 * Determine if this PFS is affected. If it is we must
625 * freeze all management threads and lock its iroot.
626 *
627 * Freezing a management thread forces it idle, operations
628 * in-progress will be aborted and it will have to start
629 * over again when unfrozen, or exit if told to exit.
630 */
631 for (i = 0; i < HAMMER2_MAXCLUSTER; ++i) {
632 if (pmp->pfs_hmps[i] == hmp)
633 break;
634 }
635 if (i != HAMMER2_MAXCLUSTER) {
636 /*
637 * Make sure all synchronization threads are locked
638 * down.
639 */
640 for (i = 0; i < HAMMER2_MAXCLUSTER; ++i) {
641 if (pmp->pfs_hmps[i] == NULL)
642 continue;
643 hammer2_thr_freeze_async(&pmp->sync_thrs[i]);
644 for (j = 0; j < HAMMER2_XOPGROUPS; ++j) {
645 hammer2_thr_freeze_async(
646 &pmp->xop_groups[j].thrs[i]);
647 }
648 }
649 for (i = 0; i < HAMMER2_MAXCLUSTER; ++i) {
650 if (pmp->pfs_hmps[i] == NULL)
651 continue;
652 hammer2_thr_freeze(&pmp->sync_thrs[i]);
653 for (j = 0; j < HAMMER2_XOPGROUPS; ++j) {
654 hammer2_thr_freeze(
655 &pmp->xop_groups[j].thrs[i]);
656 }
657 }
658
659 /*
660 * Lock the inode and clean out matching chains.
661 * Note that we cannot use hammer2_inode_lock_*()
662 * here because that would attempt to validate the
663 * cluster that we are in the middle of ripping
664 * apart.
665 *
666 * WARNING! We are working directly on the inodes
667 * embedded cluster.
668 */
669 hammer2_mtx_ex(&iroot->lock);
670
671 /*
672 * Remove the chain from matching elements of the PFS.
673 */
674 for (i = 0; i < HAMMER2_MAXCLUSTER; ++i) {
675 if (pmp->pfs_hmps[i] != hmp)
676 continue;
677 hammer2_thr_delete(&pmp->sync_thrs[i]);
678 for (j = 0; j < HAMMER2_XOPGROUPS; ++j) {
679 hammer2_thr_delete(
680 &pmp->xop_groups[j].thrs[i]);
681 }
682 rchain = iroot->cluster.array[i].chain;
683 iroot->cluster.array[i].chain = NULL;
684 pmp->pfs_types[i] = 0;
685 if (pmp->pfs_names[i]) {
686 kfree(pmp->pfs_names[i], M_HAMMER2);
687 pmp->pfs_names[i] = NULL;
688 }
689 if (rchain) {
690 hammer2_chain_drop(rchain);
691 /* focus hint */
692 if (iroot->cluster.focus == rchain)
693 iroot->cluster.focus = NULL;
694 }
695 pmp->pfs_hmps[i] = NULL;
696 }
697 hammer2_mtx_unlock(&iroot->lock);
698 didfreeze = 1; /* remaster, unfreeze down below */
699 } else {
700 didfreeze = 0;
701 }
702
703 /*
704 * Cleanup trailing chains. Gaps may remain.
705 */
706 for (i = HAMMER2_MAXCLUSTER - 1; i >= 0; --i) {
707 if (pmp->pfs_hmps[i])
708 break;
709 }
710 iroot->cluster.nchains = i + 1;
711
712 /*
713 * If the PMP has no elements remaining we can destroy it.
714 * (this will transition management threads from frozen->exit).
715 */
716 if (iroot->cluster.nchains == 0) {
717 kprintf("unmount hmp %p last ref to PMP=%p\n",
718 hmp, pmp);
719 hammer2_pfsfree(pmp);
720 goto again;
721 }
722
723 /*
724 * If elements still remain we need to set the REMASTER
725 * flag and unfreeze it.
726 */
727 if (didfreeze) {
728 for (i = 0; i < HAMMER2_MAXCLUSTER; ++i) {
729 if (pmp->pfs_hmps[i] == NULL)
730 continue;
731 hammer2_thr_remaster(&pmp->sync_thrs[i]);
732 hammer2_thr_unfreeze(&pmp->sync_thrs[i]);
733 for (j = 0; j < HAMMER2_XOPGROUPS; ++j) {
734 hammer2_thr_remaster(
735 &pmp->xop_groups[j].thrs[i]);
736 hammer2_thr_unfreeze(
737 &pmp->xop_groups[j].thrs[i]);
738 }
739 }
740 }
741 }
742 }
743
744 /*
745 * Mount or remount HAMMER2 fileystem from physical media
746 *
747 * mountroot
748 * mp mount point structure
749 * path NULL
750 * data <unused>
751 * cred <unused>
752 *
753 * mount
754 * mp mount point structure
755 * path path to mount point
756 * data pointer to argument structure in user space
757 * volume volume path (device@LABEL form)
758 * hflags user mount flags
759 * cred user credentials
760 *
761 * RETURNS: 0 Success
762 * !0 error number
763 */
764 static
765 int
766 hammer2_vfs_mount(struct mount *mp, char *path, caddr_t data,
767 struct ucred *cred)
768 {
769 struct hammer2_mount_info info;
770 hammer2_pfs_t *pmp;
771 hammer2_pfs_t *spmp;
772 hammer2_dev_t *hmp;
773 hammer2_dev_t *force_local;
774 hammer2_key_t key_next;
775 hammer2_key_t key_dummy;
776 hammer2_key_t lhc;
777 struct vnode *devvp;
778 struct nlookupdata nd;
779 hammer2_chain_t *parent;
780 hammer2_chain_t *chain;
781 hammer2_cluster_t *cluster;
782 const hammer2_inode_data_t *ripdata;
783 hammer2_blockref_t bref;
784 struct file *fp;
785 char devstr[MNAMELEN];
786 size_t size;
787 size_t done;
788 char *dev;
789 char *label;
790 int ronly = 1;
791 int error;
792 int cache_index;
793 int i;
794
795 hmp = NULL;
796 pmp = NULL;
797 dev = NULL;
798 label = NULL;
799 devvp = NULL;
800 cache_index = -1;
801
802 kprintf("hammer2_mount\n");
803
804 if (path == NULL) {
805 /*
806 * Root mount
807 */
808 bzero(&info, sizeof(info));
809 info.cluster_fd = -1;
810 ksnprintf(devstr, sizeof(devstr), "%s",
811 mp->mnt_stat.f_mntfromname);
812 kprintf("hammer2_mount: root '%s'\n", devstr);
813 } else {
814 /*
815 * Non-root mount or updating a mount
816 */
817 error = copyin(data, &info, sizeof(info));
818 if (error)
819 return (error);
820
821 error = copyinstr(info.volume, devstr, MNAMELEN - 1, &done);
822 if (error)
823 return (error);
824 }
825
826 /* Extract device and label */
827 dev = devstr;
828 label = strchr(devstr, '@');
829 if (label == NULL ||
830 ((label + 1) - dev) > done) {
831 return (EINVAL);
832 }
833 *label = '\0';
834 label++;
835 if (*label == '\0')
836 return (EINVAL);
837
838 kprintf("hammer2_mount: dev=\"%s\" label=\"%s\"\n",
839 dev, label);
840
841 if (mp->mnt_flag & MNT_UPDATE) {
842 /*
843 * Update mount. Note that pmp->iroot->cluster is
844 * an inode-embedded cluster and thus cannot be
845 * directly locked.
846 *
847 * XXX HAMMER2 needs to implement NFS export via
848 * mountctl.
849 */
850 pmp = MPTOPMP(mp);
851 pmp->hflags = info.hflags;
852 cluster = &pmp->iroot->cluster;
853 for (i = 0; i < cluster->nchains; ++i) {
854 if (cluster->array[i].chain == NULL)
855 continue;
856 hmp = cluster->array[i].chain->hmp;
857 devvp = hmp->devvp;
858 error = hammer2_remount(hmp, mp, path,
859 devvp, cred);
860 if (error)
861 break;
862 }
863
864 return error;
865 }
866
867 /*
868 * HMP device mount
869 *
870 * If a path is specified and dev is not an empty string, lookup the
871 * name and verify that it referes to a block device.
872 *
873 * If a path is specified and dev is an empty string we fall through
874 * and locate the label in the hmp search.
875 */
876 if (path && *dev != 0) {
877 error = nlookup_init(&nd, dev, UIO_SYSSPACE, NLC_FOLLOW);
878 if (error == 0)
879 error = nlookup(&nd);
880 if (error == 0)
881 error = cache_vref(&nd.nl_nch, nd.nl_cred, &devvp);
882 nlookup_done(&nd);
883 } else if (path == NULL) {
884 /* root mount */
885 cdev_t cdev = kgetdiskbyname(dev);
886 error = bdevvp(cdev, &devvp);
887 if (error)
888 kprintf("hammer2: cannot find '%s'\n", dev);
889 } else {
890 /*
891 * We will locate the hmp using the label in the hmp loop.
892 */
893 error = 0;
894 }
895
896 if (error == 0 && devvp) {
897 if (vn_isdisk(devvp, &error))
898 error = vfs_mountedon(devvp);
899 }
900
901 /*
902 * Determine if the device has already been mounted. After this
903 * check hmp will be non-NULL if we are doing the second or more
904 * hammer2 mounts from the same device.
905 */
906 lockmgr(&hammer2_mntlk, LK_EXCLUSIVE);
907 if (devvp) {
908 /*
909 * Match the device
910 */
911 TAILQ_FOREACH(hmp, &hammer2_mntlist, mntentry) {
912 if (hmp->devvp == devvp)
913 break;
914 }
915 } else if (error == 0) {
916 /*
917 * Match the label to a pmp already probed.
918 */
919 TAILQ_FOREACH(pmp, &hammer2_pfslist, mntentry) {
920 for (i = 0; i < HAMMER2_MAXCLUSTER; ++i) {
921 if (pmp->pfs_names[i] &&
922 strcmp(pmp->pfs_names[i], label) == 0) {
923 hmp = pmp->pfs_hmps[i];
924 break;
925 }
926 }
927 if (hmp)
928 break;
929 }
930 if (hmp == NULL)
931 error = ENOENT;
932 }
933
934 /*
935 * Open the device if this isn't a secondary mount and construct
936 * the H2 device mount (hmp).
937 */
938 if (hmp == NULL) {
939 hammer2_chain_t *schain;
940 hammer2_xid_t xid;
941
942 if (error == 0 && vcount(devvp) > 0)
943 error = EBUSY;
944
945 /*
946 * Now open the device
947 */
948 if (error == 0) {
949 ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
950 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
951 error = vinvalbuf(devvp, V_SAVE, 0, 0);
952 if (error == 0) {
953 error = VOP_OPEN(devvp,
954 ronly ? FREAD : FREAD | FWRITE,
955 FSCRED, NULL);
956 }
957 vn_unlock(devvp);
958 }
959 if (error && devvp) {
960 vrele(devvp);
961 devvp = NULL;
962 }
963 if (error) {
964 lockmgr(&hammer2_mntlk, LK_RELEASE);
965 return error;
966 }
967 hmp = kmalloc(sizeof(*hmp), M_HAMMER2, M_WAITOK | M_ZERO);
968 ksnprintf(hmp->devrepname, sizeof(hmp->devrepname), "%s", dev);
969 hmp->ronly = ronly;
970 hmp->devvp = devvp;
971 hmp->hflags = info.hflags & HMNT2_DEVFLAGS;
972 kmalloc_create(&hmp->mchain, "HAMMER2-chains");
973 TAILQ_INSERT_TAIL(&hammer2_mntlist, hmp, mntentry);
974 RB_INIT(&hmp->iotree);
975 spin_init(&hmp->io_spin, "hm2mount_io");
976 spin_init(&hmp->list_spin, "hm2mount_list");
977 TAILQ_INIT(&hmp->flushq);
978
979 lockinit(&hmp->vollk, "h2vol", 0, 0);
980 lockinit(&hmp->bulklk, "h2bulk", 0, 0);
981
982 /*
983 * vchain setup. vchain.data is embedded.
984 * vchain.refs is initialized and will never drop to 0.
985 *
986 * NOTE! voldata is not yet loaded.
987 */
988 hmp->vchain.hmp = hmp;
989 hmp->vchain.refs = 1;
990 hmp->vchain.data = (void *)&hmp->voldata;
991 hmp->vchain.bref.type = HAMMER2_BREF_TYPE_VOLUME;
992 hmp->vchain.bref.data_off = 0 | HAMMER2_PBUFRADIX;
993 hmp->vchain.bref.mirror_tid = hmp->voldata.mirror_tid;
994
995 hammer2_chain_core_init(&hmp->vchain);
996 /* hmp->vchain.u.xxx is left NULL */
997
998 /*
999 * fchain setup. fchain.data is embedded.
1000 * fchain.refs is initialized and will never drop to 0.
1001 *
1002 * The data is not used but needs to be initialized to
1003 * pass assertion muster. We use this chain primarily
1004 * as a placeholder for the freemap's top-level RBTREE
1005 * so it does not interfere with the volume's topology
1006 * RBTREE.
1007 */
1008 hmp->fchain.hmp = hmp;
1009 hmp->fchain.refs = 1;
1010 hmp->fchain.data = (void *)&hmp->voldata.freemap_blockset;
1011 hmp->fchain.bref.type = HAMMER2_BREF_TYPE_FREEMAP;
1012 hmp->fchain.bref.data_off = 0 | HAMMER2_PBUFRADIX;
1013 hmp->fchain.bref.mirror_tid = hmp->voldata.freemap_tid;
1014 hmp->fchain.bref.methods =
1015 HAMMER2_ENC_CHECK(HAMMER2_CHECK_FREEMAP) |
1016 HAMMER2_ENC_COMP(HAMMER2_COMP_NONE);
1017
1018 hammer2_chain_core_init(&hmp->fchain);
1019 /* hmp->fchain.u.xxx is left NULL */
1020
1021 /*
1022 * Install the volume header and initialize fields from
1023 * voldata.
1024 */
1025 error = hammer2_install_volume_header(hmp);
1026 if (error) {
1027 hammer2_unmount_helper(mp, NULL, hmp);
1028 lockmgr(&hammer2_mntlk, LK_RELEASE);
1029 hammer2_vfs_unmount(mp, MNT_FORCE);
1030 return error;
1031 }
1032
1033 /*
1034 * Really important to get these right or flush will get
1035 * confused.
1036 */
1037 hmp->spmp = hammer2_pfsalloc(NULL, NULL, 0, NULL);
1038 kprintf("alloc spmp %p tid %016jx\n",
1039 hmp->spmp, hmp->voldata.mirror_tid);
1040 spmp = hmp->spmp;
1041
1042 /*
1043 * Dummy-up vchain and fchain's modify_tid. mirror_tid
1044 * is inherited from the volume header.
1045 */
1046 xid = 0;
1047 hmp->vchain.bref.mirror_tid = hmp->voldata.mirror_tid;
1048 hmp->vchain.bref.modify_tid = hmp->vchain.bref.mirror_tid;
1049 hmp->vchain.pmp = spmp;
1050 hmp->fchain.bref.mirror_tid = hmp->voldata.freemap_tid;
1051 hmp->fchain.bref.modify_tid = hmp->fchain.bref.mirror_tid;
1052 hmp->fchain.pmp = spmp;
1053
1054 /*
1055 * First locate the super-root inode, which is key 0
1056 * relative to the volume header's blockset.
1057 *
1058 * Then locate the root inode by scanning the directory keyspace
1059 * represented by the label.
1060 */
1061 parent = hammer2_chain_lookup_init(&hmp->vchain, 0);
1062 schain = hammer2_chain_lookup(&parent, &key_dummy,
1063 HAMMER2_SROOT_KEY, HAMMER2_SROOT_KEY,
1064 &cache_index, 0);
1065 hammer2_chain_lookup_done(parent);
1066 if (schain == NULL) {
1067 kprintf("hammer2_mount: invalid super-root\n");
1068 hammer2_unmount_helper(mp, NULL, hmp);
1069 lockmgr(&hammer2_mntlk, LK_RELEASE);
1070 hammer2_vfs_unmount(mp, MNT_FORCE);
1071 return EINVAL;
1072 }
1073 if (schain->error) {
1074 kprintf("hammer2_mount: error %s reading super-root\n",
1075 hammer2_error_str(schain->error));
1076 hammer2_chain_unlock(schain);
1077 hammer2_chain_drop(schain);
1078 schain = NULL;
1079 hammer2_unmount_helper(mp, NULL, hmp);
1080 lockmgr(&hammer2_mntlk, LK_RELEASE);
1081 hammer2_vfs_unmount(mp, MNT_FORCE);
1082 return EINVAL;
1083 }
1084
1085 /*
1086 * The super-root always uses an inode_tid of 1 when
1087 * creating PFSs.
1088 */
1089 spmp->inode_tid = 1;
1090 spmp->modify_tid = schain->bref.modify_tid + 1;
1091
1092 /*
1093 * Sanity-check schain's pmp and finish initialization.
1094 * Any chain belonging to the super-root topology should
1095 * have a NULL pmp (not even set to spmp).
1096 */
1097 ripdata = &hammer2_chain_rdata(schain)->ipdata;
1098 KKASSERT(schain->pmp == NULL);
1099 spmp->pfs_clid = ripdata->meta.pfs_clid;
1100
1101 /*
1102 * Replace the dummy spmp->iroot with a real one. It's
1103 * easier to just do a wholesale replacement than to try
1104 * to update the chain and fixup the iroot fields.
1105 *
1106 * The returned inode is locked with the supplied cluster.
1107 */
1108 cluster = hammer2_cluster_from_chain(schain);
1109 hammer2_inode_drop(spmp->iroot);
1110 spmp->iroot = NULL;
1111 spmp->iroot = hammer2_inode_get(spmp, NULL, cluster, -1);
1112 spmp->spmp_hmp = hmp;
1113 spmp->pfs_types[0] = ripdata->meta.pfs_type;
1114 spmp->pfs_hmps[0] = hmp;
1115 hammer2_inode_ref(spmp->iroot);
1116 hammer2_inode_unlock(spmp->iroot);
1117 hammer2_cluster_unlock(cluster);
1118 hammer2_cluster_drop(cluster);
1119 schain = NULL;
1120 /* leave spmp->iroot with one ref */
1121
1122 if ((mp->mnt_flag & MNT_RDONLY) == 0) {
1123 error = hammer2_recovery(hmp);
1124 /* XXX do something with error */
1125 }
1126 hammer2_update_pmps(hmp);
1127 hammer2_iocom_init(hmp);
1128
1129 /*
1130 * Ref the cluster management messaging descriptor. The mount
1131 * program deals with the other end of the communications pipe.
1132 *
1133 * Root mounts typically do not supply one.
1134 */
1135 if (info.cluster_fd >= 0) {
1136 fp = holdfp(curproc->p_fd, info.cluster_fd, -1);
1137 if (fp) {
1138 hammer2_cluster_reconnect(hmp, fp);
1139 } else {
1140 kprintf("hammer2_mount: bad cluster_fd!\n");
1141 }
1142 }
1143 } else {
1144 spmp = hmp->spmp;
1145 if (info.hflags & HMNT2_DEVFLAGS) {
1146 kprintf("hammer2: Warning: mount flags pertaining "
1147 "to the whole device may only be specified "
1148 "on the first mount of the device: %08x\n",
1149 info.hflags & HMNT2_DEVFLAGS);
1150 }
1151 }
1152
1153 /*
1154 * Force local mount (disassociate all PFSs from their clusters).
1155 * Used primarily for debugging.
1156 */
1157 force_local = (hmp->hflags & HMNT2_LOCAL) ? hmp : NULL;
1158
1159 /*
1160 * Lookup the mount point under the media-localized super-root.
1161 * Scanning hammer2_pfslist doesn't help us because it represents
1162 * PFS cluster ids which can aggregate several named PFSs together.
1163 *
1164 * cluster->pmp will incorrectly point to spmp and must be fixed
1165 * up later on.
1166 */
1167 hammer2_inode_lock(spmp->iroot, 0);
1168 parent = hammer2_inode_chain(spmp->iroot, 0, HAMMER2_RESOLVE_ALWAYS);
1169 lhc = hammer2_dirhash(label, strlen(label));
1170 chain = hammer2_chain_lookup(&parent, &key_next,
1171 lhc, lhc + HAMMER2_DIRHASH_LOMASK,
1172 &cache_index, 0);
1173 while (chain) {
1174 if (chain->bref.type == HAMMER2_BREF_TYPE_INODE &&
1175 strcmp(label, chain->data->ipdata.filename) == 0) {
1176 break;
1177 }
1178 chain = hammer2_chain_next(&parent, chain, &key_next,
1179 key_next,
1180 lhc + HAMMER2_DIRHASH_LOMASK,
1181 &cache_index, 0);
1182 }
1183 if (parent) {
1184 hammer2_chain_unlock(parent);
1185 hammer2_chain_drop(parent);
1186 }
1187 hammer2_inode_unlock(spmp->iroot);
1188
1189 /*
1190 * PFS could not be found?
1191 */
1192 if (chain == NULL) {
1193 kprintf("hammer2_mount: PFS label not found\n");
1194 hammer2_unmount_helper(mp, NULL, hmp);
1195 lockmgr(&hammer2_mntlk, LK_RELEASE);
1196 hammer2_vfs_unmount(mp, MNT_FORCE);
1197
1198 return EINVAL;
1199 }
1200
1201 /*
1202 * Acquire the pmp structure (it should have already been allocated
1203 * via hammer2_update_pmps() so do not pass cluster in to add to
1204 * available chains).
1205 *
1206 * Check if the cluster has already been mounted. A cluster can
1207 * only be mounted once, use null mounts to mount additional copies.
1208 */
1209 ripdata = &chain->data->ipdata;
1210 bref = chain->bref;
1211 pmp = hammer2_pfsalloc(NULL, ripdata,
1212 bref.modify_tid, force_local);
1213 hammer2_chain_unlock(chain);
1214 hammer2_chain_drop(chain);
1215
1216 /*
1217 * Finish the mount
1218 */
1219 kprintf("hammer2_mount hmp=%p pmp=%p\n", hmp, pmp);
1220
1221 if (pmp->mp) {
1222 kprintf("hammer2_mount: PFS already mounted!\n");
1223 hammer2_unmount_helper(mp, NULL, hmp);
1224 lockmgr(&hammer2_mntlk, LK_RELEASE);
1225 hammer2_vfs_unmount(mp, MNT_FORCE);
1226
1227 return EBUSY;
1228 }
1229
1230 pmp->hflags = info.hflags;
1231 mp->mnt_flag = MNT_LOCAL;
1232 mp->mnt_kern_flag |= MNTK_ALL_MPSAFE; /* all entry pts are SMP */
1233 mp->mnt_kern_flag |= MNTK_THR_SYNC; /* new vsyncscan semantics */
1234
1235 /*
1236 * required mount structure initializations
1237 */
1238 mp->mnt_stat.f_iosize = HAMMER2_PBUFSIZE;
1239 mp->mnt_stat.f_bsize = HAMMER2_PBUFSIZE;
1240
1241 mp->mnt_vstat.f_frsize = HAMMER2_PBUFSIZE;
1242 mp->mnt_vstat.f_bsize = HAMMER2_PBUFSIZE;
1243
1244 /*
1245 * Optional fields
1246 */
1247 mp->mnt_iosize_max = MAXPHYS;
1248
1249 /*
1250 * Connect up mount pointers.
1251 */
1252 hammer2_mount_helper(mp, pmp);
1253
1254 lockmgr(&hammer2_mntlk, LK_RELEASE);
1255
1256 /*
1257 * Finish setup
1258 */
1259 vfs_getnewfsid(mp);
1260 vfs_add_vnodeops(mp, &hammer2_vnode_vops, &mp->mnt_vn_norm_ops);
1261 vfs_add_vnodeops(mp, &hammer2_spec_vops, &mp->mnt_vn_spec_ops);
1262 vfs_add_vnodeops(mp, &hammer2_fifo_vops, &mp->mnt_vn_fifo_ops);
1263
1264 if (path) {
1265 copyinstr(info.volume, mp->mnt_stat.f_mntfromname,
1266 MNAMELEN - 1, &size);
1267 bzero(mp->mnt_stat.f_mntfromname + size, MNAMELEN - size);
1268 } /* else root mount, already in there */
1269
1270 bzero(mp->mnt_stat.f_mntonname, sizeof(mp->mnt_stat.f_mntonname));
1271 if (path) {
1272 copyinstr(path, mp->mnt_stat.f_mntonname,
1273 sizeof(mp->mnt_stat.f_mntonname) - 1,
1274 &size);
1275 } else {
1276 /* root mount */
1277 mp->mnt_stat.f_mntonname[0] = '/';
1278 }
1279
1280 /*
1281 * Initial statfs to prime mnt_stat.
1282 */
1283 hammer2_vfs_statfs(mp, &mp->mnt_stat, cred);
1284
1285 return 0;
1286 }
1287
1288 /*
1289 * Scan PFSs under the super-root and create hammer2_pfs structures.
1290 */
1291 static
1292 void
1293 hammer2_update_pmps(hammer2_dev_t *hmp)
1294 {
1295 const hammer2_inode_data_t *ripdata;
1296 hammer2_chain_t *parent;
1297 hammer2_chain_t *chain;
1298 hammer2_blockref_t bref;
1299 hammer2_dev_t *force_local;
1300 hammer2_pfs_t *spmp;
1301 hammer2_pfs_t *pmp;
1302 hammer2_key_t key_next;
1303 int cache_index = -1;
1304
1305 /*
1306 * Force local mount (disassociate all PFSs from their clusters).
1307 * Used primarily for debugging.
1308 */
1309 force_local = (hmp->hflags & HMNT2_LOCAL) ? hmp : NULL;
1310
1311 /*
1312 * Lookup mount point under the media-localized super-root.
1313 *
1314 * cluster->pmp will incorrectly point to spmp and must be fixed
1315 * up later on.
1316 */
1317 spmp = hmp->spmp;
1318 hammer2_inode_lock(spmp->iroot, 0);
1319 parent = hammer2_inode_chain(spmp->iroot, 0, HAMMER2_RESOLVE_ALWAYS);
1320 chain = hammer2_chain_lookup(&parent, &key_next,
1321 HAMMER2_KEY_MIN, HAMMER2_KEY_MAX,
1322 &cache_index, 0);
1323 while (chain) {
1324 if (chain->bref.type != HAMMER2_BREF_TYPE_INODE)
1325 continue;
1326 ripdata = &chain->data->ipdata;
1327 bref = chain->bref;
1328 kprintf("ADD LOCAL PFS: %s\n", ripdata->filename);
1329
1330 pmp = hammer2_pfsalloc(chain, ripdata,
1331 bref.modify_tid, force_local);
1332 chain = hammer2_chain_next(&parent, chain, &key_next,
1333 key_next, HAMMER2_KEY_MAX,
1334 &cache_index, 0);
1335 }
1336 if (parent) {
1337 hammer2_chain_unlock(parent);
1338 hammer2_chain_drop(parent);
1339 }
1340 hammer2_inode_unlock(spmp->iroot);
1341 }
1342
1343 static
1344 int
1345 hammer2_remount(hammer2_dev_t *hmp, struct mount *mp, char *path __unused,
1346 struct vnode *devvp, struct ucred *cred)
1347 {
1348 int error;
1349
1350 if (hmp->ronly && (mp->mnt_kern_flag & MNTK_WANTRDWR)) {
1351 error = hammer2_recovery(hmp);
1352 } else {
1353 error = 0;
1354 }
1355 return error;
1356 }
1357
1358 static
1359 int
1360 hammer2_vfs_unmount(struct mount *mp, int mntflags)
1361 {
1362 hammer2_pfs_t *pmp;
1363 int flags;
1364 int error = 0;
1365
1366 pmp = MPTOPMP(mp);
1367
1368 if (pmp == NULL)
1369 return(0);
1370
1371 lockmgr(&hammer2_mntlk, LK_EXCLUSIVE);
1372
1373 /*
1374 * If mount initialization proceeded far enough we must flush
1375 * its vnodes and sync the underlying mount points. Three syncs
1376 * are required to fully flush the filesystem (freemap updates lag
1377 * by one flush, and one extra for safety).
1378 */
1379 if (mntflags & MNT_FORCE)
1380 flags = FORCECLOSE;
1381 else
1382 flags = 0;
1383 if (pmp->iroot) {
1384 error = vflush(mp, 0, flags);
1385 if (error)
1386 goto failed;
1387 hammer2_vfs_sync(mp, MNT_WAIT);
1388 hammer2_vfs_sync(mp, MNT_WAIT);
1389 hammer2_vfs_sync(mp, MNT_WAIT);
1390 }
1391
1392 /*
1393 * Cleanup the frontend support XOPS threads
1394 */
1395 hammer2_xop_helper_cleanup(pmp);
1396
1397 if (pmp->mp)
1398 hammer2_unmount_helper(mp, pmp, NULL);
1399
1400 error = 0;
1401 failed:
1402 lockmgr(&hammer2_mntlk, LK_RELEASE);
1403
1404 return (error);
1405 }
1406
1407 /*
1408 * Mount helper, hook the system mount into our PFS.
1409 * The mount lock is held.
1410 *
1411 * We must bump the mount_count on related devices for any
1412 * mounted PFSs.
1413 */
1414 static
1415 void
1416 hammer2_mount_helper(struct mount *mp, hammer2_pfs_t *pmp)
1417 {
1418 hammer2_cluster_t *cluster;
1419 hammer2_chain_t *rchain;
1420 int i;
1421
1422 mp->mnt_data = (qaddr_t)pmp;
1423 pmp->mp = mp;
1424
1425 /*
1426 * After pmp->mp is set we have to adjust hmp->mount_count.
1427 */
1428 cluster = &pmp->iroot->cluster;
1429 for (i = 0; i < cluster->nchains; ++i) {
1430 rchain = cluster->array[i].chain;
1431 if (rchain == NULL)
1432 continue;
1433 ++rchain->hmp->mount_count;
1434 kprintf("hammer2_mount hmp=%p ++mount_count=%d\n",
1435 rchain->hmp, rchain->hmp->mount_count);
1436 }
1437
1438 /*
1439 * Create missing Xop threads
1440 */
1441 hammer2_xop_helper_create(pmp);
1442 }
1443
1444 /*
1445 * Mount helper, unhook the system mount from our PFS.
1446 * The mount lock is held.
1447 *
1448 * If hmp is supplied a mount responsible for being the first to open
1449 * the block device failed and the block device and all PFSs using the
1450 * block device must be cleaned up.
1451 *
1452 * If pmp is supplied multiple devices might be backing the PFS and each
1453 * must be disconnected. This might not be the last PFS using some of the
1454 * underlying devices. Also, we have to adjust our hmp->mount_count
1455 * accounting for the devices backing the pmp which is now undergoing an
1456 * unmount.
1457 */
1458 static
1459 void
1460 hammer2_unmount_helper(struct mount *mp, hammer2_pfs_t *pmp, hammer2_dev_t *hmp)
1461 {
1462 hammer2_cluster_t *cluster;
1463 hammer2_chain_t *rchain;
1464 struct vnode *devvp;
1465 int dumpcnt;
1466 int ronly = 0;
1467 int i;
1468
1469 /*
1470 * If no device supplied this is a high-level unmount and we have to
1471 * to disconnect the mount, adjust mount_count, and locate devices
1472 * that might now have no mounts.
1473 */
1474 if (pmp) {
1475 KKASSERT(hmp == NULL);
1476 KKASSERT((void *)(intptr_t)mp->mnt_data == pmp);
1477 pmp->mp = NULL;
1478 mp->mnt_data = NULL;
1479
1480 /*
1481 * After pmp->mp is cleared we have to account for
1482 * mount_count.
1483 */
1484 cluster = &pmp->iroot->cluster;
1485 for (i = 0; i < cluster->nchains; ++i) {
1486 rchain = cluster->array[i].chain;
1487 if (rchain == NULL)
1488 continue;
1489 --rchain->hmp->mount_count;
1490 kprintf("hammer2_unmount hmp=%p --mount_count=%d\n",
1491 rchain->hmp, rchain->hmp->mount_count);
1492 /* scrapping hmp now may invalidate the pmp */
1493 }
1494 again:
1495 TAILQ_FOREACH(hmp, &hammer2_mntlist, mntentry) {
1496 if (hmp->mount_count == 0) {
1497 hammer2_unmount_helper(NULL, NULL, hmp);
1498 goto again;
1499 }
1500 }
1501 return;
1502 }
1503
1504 /*
1505 * Try to terminate the block device. We can't terminate it if
1506 * there are still PFSs referencing it.
1507 */
1508 kprintf("hammer2_unmount hmp=%p mount_count=%d\n",
1509 hmp, hmp->mount_count);
1510 if (hmp->mount_count)
1511 return;
1512
1513 hammer2_pfsfree_scan(hmp);
1514 hammer2_dev_exlock(hmp); /* XXX order */
1515
1516 /*
1517 * Cycle the volume data lock as a safety (probably not needed any
1518 * more). To ensure everything is out we need to flush at least
1519 * three times. (1) The running of the sideq can dirty the
1520 * filesystem, (2) A normal flush can dirty the freemap, and
1521 * (3) ensure that the freemap is fully synchronized.
1522 *
1523 * The next mount's recovery scan can clean everything up but we want
1524 * to leave the filesystem in a 100% clean state on a normal unmount.
1525 */
1526 #if 0
1527 hammer2_voldata_lock(hmp);
1528 hammer2_voldata_unlock(hmp);
1529 #endif
1530 hammer2_iocom_uninit(hmp);
1531
1532 if ((hmp->vchain.flags | hmp->fchain.flags) &
1533 HAMMER2_CHAIN_FLUSH_MASK) {
1534 kprintf("hammer2_unmount: chains left over "
1535 "after final sync\n");
1536 kprintf(" vchain %08x\n", hmp->vchain.flags);
1537 kprintf(" fchain %08x\n", hmp->fchain.flags);
1538
1539 if (hammer2_debug & 0x0010)
1540 Debugger("entered debugger");
1541 }
1542
1543 KKASSERT(hmp->spmp == NULL);
1544
1545 /*
1546 * Finish up with the device vnode
1547 */
1548 if ((devvp = hmp->devvp) != NULL) {
1549 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
1550 vinvalbuf(devvp, (ronly ? 0 : V_SAVE), 0, 0);
1551 hmp->devvp = NULL;
1552 VOP_CLOSE(devvp, (ronly ? FREAD : FREAD|FWRITE), NULL);
1553 vn_unlock(devvp);
1554 vrele(devvp);
1555 devvp = NULL;
1556 }
1557
1558 /*
1559 * Clear vchain/fchain flags that might prevent final cleanup
1560 * of these chains.
1561 */
1562 if (hmp->vchain.flags & HAMMER2_CHAIN_MODIFIED) {
1563 atomic_add_long(&hammer2_count_modified_chains, -1);
1564 atomic_clear_int(&hmp->vchain.flags, HAMMER2_CHAIN_MODIFIED);
1565 hammer2_pfs_memory_wakeup(hmp->vchain.pmp);
1566 }
1567 if (hmp->vchain.flags & HAMMER2_CHAIN_UPDATE) {
1568 atomic_clear_int(&hmp->vchain.flags, HAMMER2_CHAIN_UPDATE);
1569 }
1570
1571 if (hmp->fchain.flags & HAMMER2_CHAIN_MODIFIED) {
1572 atomic_add_long(&hammer2_count_modified_chains, -1);
1573 atomic_clear_int(&hmp->fchain.flags, HAMMER2_CHAIN_MODIFIED);
1574 hammer2_pfs_memory_wakeup(hmp->fchain.pmp);
1575 }
1576 if (hmp->fchain.flags & HAMMER2_CHAIN_UPDATE) {
1577 atomic_clear_int(&hmp->fchain.flags, HAMMER2_CHAIN_UPDATE);
1578 }
1579
1580 /*
1581 * Final drop of embedded freemap root chain to
1582 * clean up fchain.core (fchain structure is not
1583 * flagged ALLOCATED so it is cleaned out and then
1584 * left to rot).
1585 */
1586 hammer2_chain_drop(&hmp->fchain);
1587
1588 /*
1589 * Final drop of embedded volume root chain to clean
1590 * up vchain.core (vchain structure is not flagged
1591 * ALLOCATED so it is cleaned out and then left to
1592 * rot).
1593 */
1594 dumpcnt = 50;
1595 hammer2_dump_chain(&hmp->vchain, 0, &dumpcnt, 'v');
1596 dumpcnt = 50;
1597 hammer2_dump_chain(&hmp->fchain, 0, &dumpcnt, 'f');
1598 hammer2_dev_unlock(hmp);
1599 hammer2_chain_drop(&hmp->vchain);
1600
1601 hammer2_io_cleanup(hmp, &hmp->iotree);
1602 if (hmp->iofree_count) {
1603 kprintf("io_cleanup: %d I/O's left hanging\n",
1604 hmp->iofree_count);
1605 }
1606
1607 TAILQ_REMOVE(&hammer2_mntlist, hmp, mntentry);
1608 kmalloc_destroy(&hmp->mchain);
1609 kfree(hmp, M_HAMMER2);
1610 }
1611
1612 int
1613 hammer2_vfs_vget(struct mount *mp, struct vnode *dvp,
1614 ino_t ino, struct vnode **vpp)
1615 {
1616 hammer2_xop_lookup_t *xop;
1617 hammer2_pfs_t *pmp;
1618 hammer2_inode_t *ip;
1619 hammer2_tid_t inum;
1620 int error;
1621
1622 inum = (hammer2_tid_t)ino & HAMMER2_DIRHASH_USERMSK;
1623
1624 error = 0;
1625 pmp = MPTOPMP(mp);
1626
1627 /*
1628 * Easy if we already have it cached
1629 */
1630 ip = hammer2_inode_lookup(pmp, inum);
1631 if (ip) {
1632 hammer2_inode_lock(ip, HAMMER2_RESOLVE_SHARED);
1633 *vpp = hammer2_igetv(ip, &error);
1634 hammer2_inode_unlock(ip);
1635 hammer2_inode_drop(ip); /* from lookup */
1636
1637 return error;
1638 }
1639
1640 /*
1641 * Otherwise we have to find the inode
1642 */
1643 xop = hammer2_xop_alloc(pmp->iroot, 0);
1644 xop->lhc = inum;
1645 hammer2_xop_start(&xop->head, hammer2_xop_lookup);
1646 error = hammer2_xop_collect(&xop->head, 0);
1647
1648 if (error == 0) {
1649 if (hammer2_cluster_rdata(&xop->head.cluster) == NULL) {
1650 kprintf("vget: no collect error but also no rdata\n");
1651 kprintf("xop %p\n", xop);
1652 while ((hammer2_debug & 0x80000) == 0) {
1653 tsleep(xop, PCATCH, "wait", hz * 10);
1654 }
1655 ip = NULL;
1656 } else {
1657 ip = hammer2_inode_get(pmp, NULL, &xop->head.cluster, -1);
1658 }
1659 }
1660 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1661
1662 if (ip) {
1663 *vpp = hammer2_igetv(ip, &error);
1664 hammer2_inode_unlock(ip);
1665 } else {
1666 *vpp = NULL;
1667 error = ENOENT;
1668 }
1669 return (error);
1670 }
1671
1672 static
1673 int
1674 hammer2_vfs_root(struct mount *mp, struct vnode **vpp)
1675 {
1676 hammer2_pfs_t *pmp;
1677 struct vnode *vp;
1678 int error;
1679
1680 pmp = MPTOPMP(mp);
1681 if (pmp->iroot == NULL) {
1682 *vpp = NULL;
1683 return EINVAL;
1684 }
1685
1686 error = 0;
1687 hammer2_inode_lock(pmp->iroot, HAMMER2_RESOLVE_SHARED);
1688
1689 while (pmp->inode_tid == 0) {
1690 hammer2_xop_ipcluster_t *xop;
1691 hammer2_inode_meta_t *meta;
1692
1693 xop = hammer2_xop_alloc(pmp->iroot, HAMMER2_XOP_MODIFYING);
1694 hammer2_xop_start(&xop->head, hammer2_xop_ipcluster);
1695 error = hammer2_xop_collect(&xop->head, 0);
1696
1697 if (error == 0) {
1698 meta = &xop->head.cluster.focus->data->ipdata.meta;
1699 pmp->iroot->meta = *meta;
1700 pmp->inode_tid = meta->pfs_inum + 1;
1701 if (pmp->inode_tid < HAMMER2_INODE_START)
1702 pmp->inode_tid = HAMMER2_INODE_START;
1703 pmp->modify_tid =
1704 xop->head.cluster.focus->bref.modify_tid + 1;
1705 kprintf("PFS: Starting inode %jd\n",
1706 (intmax_t)pmp->inode_tid);
1707 kprintf("PMP focus good set nextino=%ld mod=%016jx\n",
1708 pmp->inode_tid, pmp->modify_tid);
1709 wakeup(&pmp->iroot);
1710
1711 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1712
1713 /*
1714 * Prime the mount info.
1715 */
1716 hammer2_vfs_statfs(mp, &mp->mnt_stat, NULL);
1717 break;
1718 }
1719
1720 /*
1721 * Loop, try again
1722 */
1723 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
1724 hammer2_inode_unlock(pmp->iroot);
1725 error = tsleep(&pmp->iroot, PCATCH, "h2root", hz);
1726 hammer2_inode_lock(pmp->iroot, HAMMER2_RESOLVE_SHARED);
1727 if (error == EINTR)
1728 break;
1729 }
1730
1731 if (error) {
1732 hammer2_inode_unlock(pmp->iroot);
1733 *vpp = NULL;
1734 } else {
1735 vp = hammer2_igetv(pmp->iroot, &error);
1736 hammer2_inode_unlock(pmp->iroot);
1737 *vpp = vp;
1738 }
1739
1740 return (error);
1741 }
1742
1743 /*
1744 * Filesystem status
1745 *
1746 * XXX incorporate ipdata->meta.inode_quota and data_quota
1747 */
1748 static
1749 int
1750 hammer2_vfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
1751 {
1752 hammer2_pfs_t *pmp;
1753 hammer2_dev_t *hmp;
1754 hammer2_blockref_t bref;
1755 int i;
1756
1757 /*
1758 * NOTE: iroot might not have validated the cluster yet.
1759 */
1760 pmp = MPTOPMP(mp);
1761
1762 mp->mnt_stat.f_files = 0;
1763 mp->mnt_stat.f_ffree = 0;
1764 mp->mnt_stat.f_blocks = 0;
1765 mp->mnt_stat.f_bfree = 0;
1766 mp->mnt_stat.f_bavail = 0;
1767
1768 for (i = 0; i < pmp->iroot->cluster.nchains; ++i) {
1769 hmp = pmp->pfs_hmps[i];
1770 if (hmp == NULL)
1771 continue;
1772 if (pmp->iroot->cluster.array[i].chain)
1773 bref = pmp->iroot->cluster.array[i].chain->bref;
1774 else
1775 bzero(&bref, sizeof(bref));
1776
1777 mp->mnt_stat.f_files = bref.inode_count;
1778 mp->mnt_stat.f_ffree = 0;
1779 #if 0
1780 mp->mnt_stat.f_blocks = (bref.data_count +
1781 hmp->voldata.allocator_free) /
1782 mp->mnt_vstat.f_bsize;
1783 mp->mnt_stat.f_bfree = hmp->voldata.allocator_free /
1784 mp->mnt_vstat.f_bsize;
1785 #endif
1786 mp->mnt_stat.f_blocks = hmp->voldata.allocator_size /
1787 mp->mnt_vstat.f_bsize;
1788 mp->mnt_stat.f_bfree = hmp->voldata.allocator_free /
1789 mp->mnt_vstat.f_bsize;
1790 mp->mnt_stat.f_bavail = mp->mnt_stat.f_bfree;
1791
1792 *sbp = mp->mnt_stat;
1793 }
1794 return (0);
1795 }
1796
1797 static
1798 int
1799 hammer2_vfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred)
1800 {
1801 hammer2_pfs_t *pmp;
1802 hammer2_dev_t *hmp;
1803 hammer2_blockref_t bref;
1804 int i;
1805
1806 /*
1807 * NOTE: iroot might not have validated the cluster yet.
1808 */
1809 pmp = MPTOPMP(mp);
1810
1811 mp->mnt_vstat.f_bsize = 0;
1812 mp->mnt_vstat.f_files = 0;
1813 mp->mnt_vstat.f_ffree = 0;
1814 mp->mnt_vstat.f_blocks = 0;
1815 mp->mnt_vstat.f_bfree = 0;
1816 mp->mnt_vstat.f_bavail = 0;
1817
1818 for (i = 0; i < pmp->iroot->cluster.nchains; ++i) {
1819 hmp = pmp->pfs_hmps[i];
1820 if (hmp == NULL)
1821 continue;
1822 if (pmp->iroot->cluster.array[i].chain)
1823 bref = pmp->iroot->cluster.array[i].chain->bref;
1824 else
1825 bzero(&bref, sizeof(bref));
1826
1827 mp->mnt_vstat.f_bsize = HAMMER2_PBUFSIZE;
1828 mp->mnt_vstat.f_files = bref.inode_count;
1829 mp->mnt_vstat.f_ffree = 0;
1830 #if 0
1831 mp->mnt_vstat.f_blocks = (bref.data_count +
1832 hmp->voldata.allocator_free) /
1833 mp->mnt_vstat.f_bsize;
1834 mp->mnt_vstat.f_bfree = hmp->voldata.allocator_free /
1835 mp->mnt_vstat.f_bsize;
1836 #endif
1837 mp->mnt_vstat.f_blocks = hmp->voldata.allocator_size /
1838 mp->mnt_vstat.f_bsize;
1839 mp->mnt_vstat.f_bfree = hmp->voldata.allocator_free /
1840 mp->mnt_vstat.f_bsize;
1841 mp->mnt_vstat.f_bavail = mp->mnt_vstat.f_bfree;
1842
1843 *sbp = mp->mnt_vstat;
1844 }
1845 return (0);
1846 }
1847
1848 /*
1849 * Mount-time recovery (RW mounts)
1850 *
1851 * Updates to the free block table are allowed to lag flushes by one
1852 * transaction. In case of a crash, then on a fresh mount we must do an
1853 * incremental scan of the last committed transaction id and make sure that
1854 * all related blocks have been marked allocated.
1855 *
1856 * The super-root topology and each PFS has its own transaction id domain,
1857 * so we must track PFS boundary transitions.
1858 */
1859 struct hammer2_recovery_elm {
1860 TAILQ_ENTRY(hammer2_recovery_elm) entry;
1861 hammer2_chain_t *chain;
1862 hammer2_tid_t sync_tid;
1863 };
1864
1865 TAILQ_HEAD(hammer2_recovery_list, hammer2_recovery_elm);
1866
1867 struct hammer2_recovery_info {
1868 struct hammer2_recovery_list list;
1869 hammer2_tid_t mtid;
1870 int depth;
1871 };
1872
1873 static int hammer2_recovery_scan(hammer2_dev_t *hmp,
1874 hammer2_chain_t *parent,
1875 struct hammer2_recovery_info *info,
1876 hammer2_tid_t sync_tid);
1877
1878 #define HAMMER2_RECOVERY_MAXDEPTH 10
1879
1880 static
1881 int
1882 hammer2_recovery(hammer2_dev_t *hmp)
1883 {
1884 struct hammer2_recovery_info info;
1885 struct hammer2_recovery_elm *elm;
1886 hammer2_chain_t *parent;
1887 hammer2_tid_t sync_tid;
1888 hammer2_tid_t mirror_tid;
1889 int error;
1890 int cumulative_error = 0;
1891
1892 hammer2_trans_init(hmp->spmp, 0);
1893
1894 sync_tid = hmp->voldata.freemap_tid;
1895 mirror_tid = hmp->voldata.mirror_tid;
1896
1897 kprintf("hammer2 mount \"%s\": ", hmp->devrepname);
1898 if (sync_tid >= mirror_tid) {
1899 kprintf(" no recovery needed\n");
1900 } else {
1901 kprintf(" freemap recovery %016jx-%016jx\n",
1902 sync_tid + 1, mirror_tid);
1903 }
1904
1905 TAILQ_INIT(&info.list);
1906 info.depth = 0;
1907 parent = hammer2_chain_lookup_init(&hmp->vchain, 0);
1908 cumulative_error = hammer2_recovery_scan(hmp, parent, &info, sync_tid);
1909 hammer2_chain_lookup_done(parent);
1910
1911 while ((elm = TAILQ_FIRST(&info.list)) != NULL) {
1912 TAILQ_REMOVE(&info.list, elm, entry);
1913 parent = elm->chain;
1914 sync_tid = elm->sync_tid;
1915 kfree(elm, M_HAMMER2);
1916
1917 hammer2_chain_lock(parent, HAMMER2_RESOLVE_ALWAYS);
1918 error = hammer2_recovery_scan(hmp, parent, &info,
1919 hmp->voldata.freemap_tid);
1920 hammer2_chain_unlock(parent);
1921 hammer2_chain_drop(parent); /* drop elm->chain ref */
1922 if (error)
1923 cumulative_error = error;
1924 }
1925 hammer2_trans_done(hmp->spmp);
1926
1927 return cumulative_error;
1928 }
1929
1930 static
1931 int
1932 hammer2_recovery_scan(hammer2_dev_t *hmp, hammer2_chain_t *parent,
1933 struct hammer2_recovery_info *info,
1934 hammer2_tid_t sync_tid)
1935 {
1936 const hammer2_inode_data_t *ripdata;
1937 hammer2_chain_t *chain;
1938 hammer2_blockref_t bref;
1939 int cache_index;
1940 int cumulative_error = 0;
1941 int error;
1942 int first;
1943
1944 /*
1945 * Adjust freemap to ensure that the block(s) are marked allocated.
1946 */
1947 if (parent->bref.type != HAMMER2_BREF_TYPE_VOLUME) {
1948 hammer2_freemap_adjust(hmp, &parent->bref,
1949 HAMMER2_FREEMAP_DORECOVER);
1950 }
1951
1952 /*
1953 * Check type for recursive scan
1954 */
1955 switch(parent->bref.type) {
1956 case HAMMER2_BREF_TYPE_VOLUME:
1957 /* data already instantiated */
1958 break;
1959 case HAMMER2_BREF_TYPE_INODE:
1960 /*
1961 * Must instantiate data for DIRECTDATA test and also
1962 * for recursion.
1963 */
1964 hammer2_chain_lock(parent, HAMMER2_RESOLVE_ALWAYS);
1965 ripdata = &hammer2_chain_rdata(parent)->ipdata;
1966 if (ripdata->meta.op_flags & HAMMER2_OPFLAG_DIRECTDATA) {
1967 /* not applicable to recovery scan */
1968 hammer2_chain_unlock(parent);
1969 return 0;
1970 }
1971 hammer2_chain_unlock(parent);
1972 break;
1973 case HAMMER2_BREF_TYPE_INDIRECT:
1974 /*
1975 * Must instantiate data for recursion
1976 */
1977 hammer2_chain_lock(parent, HAMMER2_RESOLVE_ALWAYS);
1978 hammer2_chain_unlock(parent);
1979 break;
1980 case HAMMER2_BREF_TYPE_DATA:
1981 case HAMMER2_BREF_TYPE_FREEMAP:
1982 case HAMMER2_BREF_TYPE_FREEMAP_NODE:
1983 case HAMMER2_BREF_TYPE_FREEMAP_LEAF:
1984 /* not applicable to recovery scan */
1985 return 0;
1986 break;
1987 default:
1988 return EDOM;
1989 }
1990
1991 /*
1992 * Defer operation if depth limit reached or if we are crossing a
1993 * PFS boundary.
1994 */
1995 if (info->depth >= HAMMER2_RECOVERY_MAXDEPTH) {
1996 struct hammer2_recovery_elm *elm;
1997
1998 elm = kmalloc(sizeof(*elm), M_HAMMER2, M_ZERO | M_WAITOK);
1999 elm->chain = parent;
2000 elm->sync_tid = sync_tid;
2001 hammer2_chain_ref(parent);
2002 TAILQ_INSERT_TAIL(&info->list, elm, entry);
2003 /* unlocked by caller */
2004
2005 return(0);
2006 }
2007
2008
2009 /*
2010 * Recursive scan of the last flushed transaction only. We are
2011 * doing this without pmp assignments so don't leave the chains
2012 * hanging around after we are done with them.
2013 */
2014 cache_index = 0;
2015 chain = NULL;
2016 first = 1;
2017
2018 while (hammer2_chain_scan(parent, &chain, &bref,
2019 &first, &cache_index,
2020 HAMMER2_LOOKUP_NODATA) != NULL) {
2021 /*
2022 * If this is a leaf
2023 */
2024 if (chain == NULL) {
2025 if (bref.mirror_tid > sync_tid) {
2026 hammer2_freemap_adjust(hmp, &bref,
2027 HAMMER2_FREEMAP_DORECOVER);
2028 }
2029 continue;
2030 }
2031
2032 /*
2033 * This may or may not be a recursive node.
2034 */
2035 atomic_set_int(&chain->flags, HAMMER2_CHAIN_RELEASE);
2036 if (bref.mirror_tid > sync_tid) {
2037 ++info->depth;
2038 error = hammer2_recovery_scan(hmp, chain,
2039 info, sync_tid);
2040 --info->depth;
2041 if (error)
2042 cumulative_error = error;
2043 }
2044
2045 /*
2046 * Flush the recovery at the PFS boundary to stage it for
2047 * the final flush of the super-root topology.
2048 */
2049 if ((bref.flags & HAMMER2_BREF_FLAG_PFSROOT) &&
2050 (chain->flags & HAMMER2_CHAIN_ONFLUSH)) {
2051 hammer2_flush(chain, HAMMER2_FLUSH_TOP);
2052 }
2053 }
2054
2055 return cumulative_error;
2056 }
2057
2058 /*
2059 * Sync a mount point; this is called on a per-mount basis from the
2060 * filesystem syncer process periodically and whenever a user issues
2061 * a sync.
2062 */
2063 int
2064 hammer2_vfs_sync(struct mount *mp, int waitfor)
2065 {
2066 hammer2_xop_flush_t *xop;
2067 struct hammer2_sync_info info;
2068 hammer2_inode_t *iroot;
2069 hammer2_pfs_t *pmp;
2070 int flags;
2071 int error;
2072
2073 pmp = MPTOPMP(mp);
2074 iroot = pmp->iroot;
2075 KKASSERT(iroot);
2076 KKASSERT(iroot->pmp == pmp);
2077
2078 /*
2079 * We can't acquire locks on existing vnodes while in a transaction
2080 * without risking a deadlock. This assumes that vfsync() can be
2081 * called without the vnode locked (which it can in DragonFly).
2082 * Otherwise we'd have to implement a multi-pass or flag the lock
2083 * failures and retry.
2084 *
2085 * The reclamation code interlocks with the sync list's token
2086 * (by removing the vnode from the scan list) before unlocking
2087 * the inode, giving us time to ref the inode.
2088 */
2089 /*flags = VMSC_GETVP;*/
2090 flags = 0;
2091 if (waitfor & MNT_LAZY)
2092 flags |= VMSC_ONEPASS;
2093
2094 /*
2095 * Preflush the vnodes using a normal transaction before interlocking
2096 * with a flush transaction. We do this to try to run as much of
2097 * the compression as possible outside the flush transaction.
2098 *
2099 * For efficiency do an async pass before making sure with a
2100 * synchronous pass on all related buffer cache buffers.
2101 */
2102 hammer2_trans_init(pmp, 0);
2103 info.error = 0;
2104 info.waitfor = MNT_NOWAIT;
2105 vsyncscan(mp, flags | VMSC_NOWAIT, hammer2_sync_scan2, &info);
2106 info.waitfor = MNT_WAIT;
2107 vsyncscan(mp, flags, hammer2_sync_scan2, &info);
2108 hammer2_trans_done(pmp);
2109
2110 /*
2111 * Start our flush transaction. This does not return until all
2112 * concurrent transactions have completed and will prevent any
2113 * new transactions from running concurrently, except for the
2114 * buffer cache transactions.
2115 *
2116 * NOTE! It is still possible for the paging code to push pages
2117 * out via a UIO_NOCOPY hammer2_vop_write() during the main
2118 * flush.
2119 */
2120 hammer2_trans_init(pmp, HAMMER2_TRANS_ISFLUSH);
2121 hammer2_inode_run_sideq(pmp);
2122
2123 info.error = 0;
2124 info.waitfor = MNT_NOWAIT;
2125 vsyncscan(mp, flags | VMSC_NOWAIT, hammer2_sync_scan2, &info);
2126 info.waitfor = MNT_WAIT;
2127 vsyncscan(mp, flags, hammer2_sync_scan2, &info);
2128 hammer2_bioq_sync(pmp);
2129
2130 /*
2131 * Use the XOP interface to concurrently flush all nodes to
2132 * synchronize the PFSROOT subtopology to the media. A standard
2133 * end-of-scan ENOENT error indicates cluster sufficiency.
2134 *
2135 * Note that this flush will not be visible on crash recovery until
2136 * we flush the super-root topology in the next loop.
2137 *
2138 * XXX For now wait for all flushes to complete.
2139 */
2140 if (iroot) {
2141 xop = hammer2_xop_alloc(iroot, HAMMER2_XOP_MODIFYING);
2142 hammer2_xop_start(&xop->head, hammer2_inode_xop_flush);
2143 error = hammer2_xop_collect(&xop->head,
2144 HAMMER2_XOP_COLLECT_WAITALL);
2145 hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
2146 if (error == ENOENT)
2147 error = 0;
2148 } else {
2149 error = 0;
2150 }
2151 hammer2_trans_done(pmp);
2152
2153 return (error);
2154 }
2155
2156 /*
2157 * Sync passes.
2158 *
2159 * Note that we ignore the tranasction mtid we got above. Instead,
2160 * each vfsync below will ultimately get its own via TRANS_BUFCACHE
2161 * transactions.
2162 */
2163 static int
2164 hammer2_sync_scan2(struct mount *mp, struct vnode *vp, void *data)
2165 {
2166 struct hammer2_sync_info *info = data;
2167 hammer2_inode_t *ip;
2168 int error;
2169
2170 /*
2171 * Degenerate cases. Note that ip == NULL typically means the
2172 * syncer vnode itself and we don't want to vclrisdirty() in that
2173 * situation.
2174 */
2175 ip = VTOI(vp);
2176 if (ip == NULL) {
2177 return(0);
2178 }
2179 if (vp->v_type == VNON || vp->v_type == VBAD) {
2180 vclrisdirty(vp);
2181 return(0);
2182 }
2183
2184 /*
2185 * VOP_FSYNC will start a new transaction so replicate some code
2186 * here to do it inline (see hammer2_vop_fsync()).
2187 *
2188 * WARNING: The vfsync interacts with the buffer cache and might
2189 * block, we can't hold the inode lock at that time.
2190 * However, we MUST ref ip before blocking to ensure that
2191 * it isn't ripped out from under us (since we do not
2192 * hold a lock on the vnode).
2193 */
2194 hammer2_inode_ref(ip);
2195 if ((ip->flags & HAMMER2_INODE_MODIFIED) ||
2196 !RB_EMPTY(&vp->v_rbdirty_tree)) {
2197 vfsync(vp, info->waitfor, 1, NULL, NULL);
2198 if (ip->flags & (HAMMER2_INODE_RESIZED |
2199 HAMMER2_INODE_MODIFIED)) {
2200 hammer2_inode_lock(ip, 0);
2201 hammer2_inode_chain_sync(ip);
2202 hammer2_inode_unlock(ip);
2203 }
2204 }
2205 if ((ip->flags & HAMMER2_INODE_MODIFIED) == 0 &&
2206 RB_EMPTY(&vp->v_rbdirty_tree)) {
2207 vclrisdirty(vp);
2208 }
2209
2210 hammer2_inode_drop(ip);
2211 #if 1
2212 error = 0;
2213 if (error)
2214 info->error = error;
2215 #endif
2216 return(0);
2217 }
2218
2219 static
2220 int
2221 hammer2_vfs_vptofh(struct vnode *vp, struct fid *fhp)
2222 {
2223 hammer2_inode_t *ip;
2224
2225 KKASSERT(MAXFIDSZ >= 16);
2226 ip = VTOI(vp);
2227 fhp->fid_len = offsetof(struct fid, fid_data[16]);
2228 fhp->fid_ext = 0;
2229 ((hammer2_tid_t *)fhp->fid_data)[0] = ip->meta.inum;
2230 ((hammer2_tid_t *)fhp->fid_data)[1] = 0;
2231
2232 return 0;
2233 }
2234
2235 static
2236 int
2237 hammer2_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
2238 struct fid *fhp, struct vnode **vpp)
2239 {
2240 hammer2_pfs_t *pmp;
2241 hammer2_tid_t inum;
2242 int error;
2243
2244 pmp = MPTOPMP(mp);
2245 inum = ((hammer2_tid_t *)fhp->fid_data)[0] & HAMMER2_DIRHASH_USERMSK;
2246 if (vpp) {
2247 if (inum == 1)
2248 error = hammer2_vfs_root(mp, vpp);
2249 else
2250 error = hammer2_vfs_vget(mp, NULL, inum, vpp);
2251 } else {
2252 error = 0;
2253 }
2254 if (error)
2255 kprintf("fhtovp: %016jx -> %p, %d\n", inum, *vpp, error);
2256 return error;
2257 }
2258
2259 static
2260 int
2261 hammer2_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
2262 int *exflagsp, struct ucred **credanonp)
2263 {
2264 hammer2_pfs_t *pmp;
2265 struct netcred *np;
2266 int error;
2267
2268 pmp = MPTOPMP(mp);
2269 np = vfs_export_lookup(mp, &pmp->export, nam);
2270 if (np) {
2271 *exflagsp = np->netc_exflags;
2272 *credanonp = &np->netc_anon;
2273 error = 0;
2274 } else {
2275 error = EACCES;
2276 }
2277 return error;
2278 }
2279
2280 /*
2281 * Support code for hammer2_vfs_mount(). Read, verify, and install the volume
2282 * header into the HMP
2283 *
2284 * XXX read four volhdrs and use the one with the highest TID whos CRC
2285 * matches.
2286 *
2287 * XXX check iCRCs.
2288 *
2289 * XXX For filesystems w/ less than 4 volhdrs, make sure to not write to
2290 * nonexistant locations.
2291 *
2292 * XXX Record selected volhdr and ring updates to each of 4 volhdrs
2293 */
2294 static
2295 int
2296 hammer2_install_volume_header(hammer2_dev_t *hmp)
2297 {
2298 hammer2_volume_data_t *vd;
2299 struct buf *bp;
2300 hammer2_crc32_t crc0, crc, bcrc0, bcrc;
2301 int error_reported;
2302 int error;
2303 int valid;
2304 int i;
2305
2306 error_reported = 0;
2307 error = 0;
2308 valid = 0;
2309 bp = NULL;
2310
2311 /*
2312 * There are up to 4 copies of the volume header (syncs iterate
2313 * between them so there is no single master). We don't trust the
2314 * volu_size field so we don't know precisely how large the filesystem
2315 * is, so depend on the OS to return an error if we go beyond the
2316 * block device's EOF.
2317 */
2318 for (i = 0; i < HAMMER2_NUM_VOLHDRS; i++) {
2319 error = bread(hmp->devvp, i * HAMMER2_ZONE_BYTES64,
2320 HAMMER2_VOLUME_BYTES, &bp);
2321 if (error) {
2322 brelse(bp);
2323 bp = NULL;
2324 continue;
2325 }
2326
2327 vd = (struct hammer2_volume_data *) bp->b_data;
2328 if ((vd->magic != HAMMER2_VOLUME_ID_HBO) &&
2329 (vd->magic != HAMMER2_VOLUME_ID_ABO)) {
2330 brelse(bp);
2331 bp = NULL;
2332 continue;
2333 }
2334
2335 if (vd->magic == HAMMER2_VOLUME_ID_ABO) {
2336 /* XXX: Reversed-endianness filesystem */
2337 kprintf("hammer2: reverse-endian filesystem detected");
2338 brelse(bp);
2339 bp = NULL;
2340 continue;
2341 }
2342
2343 crc = vd->icrc_sects[HAMMER2_VOL_ICRC_SECT0];
2344 crc0 = hammer2_icrc32(bp->b_data + HAMMER2_VOLUME_ICRC0_OFF,
2345 HAMMER2_VOLUME_ICRC0_SIZE);
2346 bcrc = vd->icrc_sects[HAMMER2_VOL_ICRC_SECT1];
2347 bcrc0 = hammer2_icrc32(bp->b_data + HAMMER2_VOLUME_ICRC1_OFF,
2348 HAMMER2_VOLUME_ICRC1_SIZE);
2349 if ((crc0 != crc) || (bcrc0 != bcrc)) {
2350 kprintf("hammer2 volume header crc "
2351 "mismatch copy #%d %08x/%08x\n",
2352 i, crc0, crc);
2353 error_reported = 1;
2354 brelse(bp);
2355 bp = NULL;
2356 continue;
2357 }
2358 if (valid == 0 || hmp->voldata.mirror_tid < vd->mirror_tid) {
2359 valid = 1;
2360 hmp->voldata = *vd;
2361 hmp->volhdrno = i;
2362 }
2363 brelse(bp);
2364 bp = NULL;
2365 }
2366 if (valid) {
2367 hmp->volsync = hmp->voldata;
2368 error = 0;
2369 if (error_reported || bootverbose || 1) { /* 1/DEBUG */
2370 kprintf("hammer2: using volume header #%d\n",
2371 hmp->volhdrno);
2372 }
2373 } else {
2374 error = EINVAL;
2375 kprintf("hammer2: no valid volume headers found!\n");
2376 }
2377 return (error);
2378 }
2379
2380 /*
2381 * This handles hysteresis on regular file flushes. Because the BIOs are
2382 * routed to a thread it is possible for an excessive number to build up
2383 * and cause long front-end stalls long before the runningbuffspace limit
2384 * is hit, so we implement hammer2_flush_pipe to control the
2385 * hysteresis.
2386 *
2387 * This is a particular problem when compression is used.
2388 */
2389 void
2390 hammer2_lwinprog_ref(hammer2_pfs_t *pmp)
2391 {
2392 atomic_add_int(&pmp->count_lwinprog, 1);
2393 }
2394
2395 void
2396 hammer2_lwinprog_drop(hammer2_pfs_t *pmp)
2397 {
2398 int lwinprog;
2399
2400 lwinprog = atomic_fetchadd_int(&pmp->count_lwinprog, -1);
2401 if ((lwinprog & HAMMER2_LWINPROG_WAITING) &&
2402 (lwinprog & HAMMER2_LWINPROG_MASK) <= hammer2_flush_pipe * 2 / 3) {
2403 atomic_clear_int(&pmp->count_lwinprog,
2404 HAMMER2_LWINPROG_WAITING);
2405 wakeup(&pmp->count_lwinprog);
2406 }
2407 if ((lwinprog & HAMMER2_LWINPROG_WAITING0) &&
2408 (lwinprog & HAMMER2_LWINPROG_MASK) <= 0) {
2409 atomic_clear_int(&pmp->count_lwinprog,
2410 HAMMER2_LWINPROG_WAITING0);
2411 wakeup(&pmp->count_lwinprog);
2412 }
2413 }
2414
2415 void
2416 hammer2_lwinprog_wait(hammer2_pfs_t *pmp, int flush_pipe)
2417 {
2418 int lwinprog;
2419 int lwflag = (flush_pipe) ? HAMMER2_LWINPROG_WAITING :
2420 HAMMER2_LWINPROG_WAITING0;
2421
2422 for (;;) {
2423 lwinprog = pmp->count_lwinprog;
2424 cpu_ccfence();
2425 if ((lwinprog & HAMMER2_LWINPROG_MASK) <= flush_pipe)
2426 break;
2427 tsleep_interlock(&pmp->count_lwinprog, 0);
2428 atomic_set_int(&pmp->count_lwinprog, lwflag);
2429 lwinprog = pmp->count_lwinprog;
2430 if ((lwinprog & HAMMER2_LWINPROG_MASK) <= flush_pipe)
2431 break;
2432 tsleep(&pmp->count_lwinprog, PINTERLOCKED, "h2wpipe", hz);
2433 }
2434 }
2435
2436 /*
2437 * Manage excessive memory resource use for chain and related
2438 * structures.
2439 */
2440 void
2441 hammer2_pfs_memory_wait(hammer2_pfs_t *pmp)
2442 {
2443 uint32_t waiting;
2444 uint32_t count;
2445 uint32_t limit;
2446 #if 0
2447 static int zzticks;
2448 #endif
2449
2450 /*
2451 * Atomic check condition and wait. Also do an early speedup of
2452 * the syncer to try to avoid hitting the wait.
2453 */
2454 for (;;) {
2455 waiting = pmp->inmem_dirty_chains;
2456 cpu_ccfence();
2457 count = waiting & HAMMER2_DIRTYCHAIN_MASK;
2458
2459 limit = pmp->mp->mnt_nvnodelistsize / 10;
2460 if (limit < hammer2_limit_dirty_chains)
2461 limit = hammer2_limit_dirty_chains;
2462 if (limit < 1000)
2463 limit = 1000;
2464
2465 #if 0
2466 if ((int)(ticks - zzticks) > hz) {
2467 zzticks = ticks;
2468 kprintf("count %ld %ld\n", count, limit);
2469 }
2470 #endif
2471
2472 /*
2473 * Block if there are too many dirty chains present, wait
2474 * for the flush to clean some out.
2475 */
2476 if (count > limit) {
2477 tsleep_interlock(&pmp->inmem_dirty_chains, 0);
2478 if (atomic_cmpset_int(&pmp->inmem_dirty_chains,
2479 waiting,
2480 waiting | HAMMER2_DIRTYCHAIN_WAITING)) {
2481 speedup_syncer(pmp->mp);
2482 tsleep(&pmp->inmem_dirty_chains, PINTERLOCKED,
2483 "chnmem", hz);
2484 }
2485 continue; /* loop on success or fail */
2486 }
2487
2488 /*
2489 * Try to start an early flush before we are forced to block.
2490 */
2491 if (count > limit * 7 / 10)
2492 speedup_syncer(pmp->mp);
2493 break;
2494 }
2495 }
2496
2497 void
2498 hammer2_pfs_memory_inc(hammer2_pfs_t *pmp)
2499 {
2500 if (pmp) {
2501 atomic_add_int(&pmp->inmem_dirty_chains, 1);
2502 }
2503 }
2504
2505 void
2506 hammer2_pfs_memory_wakeup(hammer2_pfs_t *pmp)
2507 {
2508 uint32_t waiting;
2509
2510 if (pmp == NULL)
2511 return;
2512
2513 for (;;) {
2514 waiting = pmp->inmem_dirty_chains;
2515 cpu_ccfence();
2516 if (atomic_cmpset_int(&pmp->inmem_dirty_chains,
2517 waiting,
2518 (waiting - 1) &
2519 ~HAMMER2_DIRTYCHAIN_WAITING)) {
2520 break;
2521 }
2522 }
2523
2524 if (waiting & HAMMER2_DIRTYCHAIN_WAITING)
2525 wakeup(&pmp->inmem_dirty_chains);
2526 }
2527
2528 /*
2529 * Debugging
2530 */
2531 void
2532 hammer2_dump_chain(hammer2_chain_t *chain, int tab, int *countp, char pfx)
2533 {
2534 hammer2_chain_t *scan;
2535 hammer2_chain_t *parent;
2536
2537 --*countp;
2538 if (*countp == 0) {
2539 kprintf("%*.*s...\n", tab, tab, "");
2540 return;
2541 }
2542 if (*countp < 0)
2543 return;
2544 kprintf("%*.*s%c-chain %p.%d %016jx/%d mir=%016jx\n",
2545 tab, tab, "", pfx,
2546 chain, chain->bref.type,
2547 chain->bref.key, chain->bref.keybits,
2548 chain->bref.mirror_tid);
2549
2550 kprintf("%*.*s [%08x] (%s) refs=%d",
2551 tab, tab, "",
2552 chain->flags,
2553 ((chain->bref.type == HAMMER2_BREF_TYPE_INODE &&
2554 chain->data) ? (char *)chain->data->ipdata.filename : "?"),
2555 chain->refs);
2556
2557 parent = chain->parent;
2558 if (parent)
2559 kprintf("\n%*.*s p=%p [pflags %08x prefs %d",
2560 tab, tab, "",
2561 parent, parent->flags, parent->refs);
2562 if (RB_EMPTY(&chain->core.rbtree)) {
2563 kprintf("\n");
2564 } else {
2565 kprintf(" {\n");
2566 RB_FOREACH(scan, hammer2_chain_tree, &chain->core.rbtree)
2567 hammer2_dump_chain(scan, tab + 4, countp, 'a');
2568 if (chain->bref.type == HAMMER2_BREF_TYPE_INODE && chain->data)
2569 kprintf("%*.*s}(%s)\n", tab, tab, "",
2570 chain->data->ipdata.filename);
2571 else
2572 kprintf("%*.*s}\n", tab, tab, "");
2573 }
2574 }
DragonFly BSD