Remove obsolete ieee80211_amrr.9 manual page.
[dragonfly.git] / sys / vfs / hammer2 / hammer2_vfsops.c
CommitLineData
58e43599 1/*
8138a154 2 * Copyright (c) 2011-2014 The DragonFly Project. All rights reserved.
703720e4
MD
3 *
4 * This code is derived from software contributed to The DragonFly Project
5 * by Matthew Dillon <dillon@backplane.com>
355d67fc 6 * by Daniel Flores (GSOC 2013 - mentored by Matthew Dillon, compression)
703720e4
MD
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 */
703720e4
MD
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>
a74bc66c 44#include <sys/vfsops.h>
37aa19df 45#include <sys/sysctl.h>
bfc3a7b1 46#include <sys/socket.h>
355d67fc
MD
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>
703720e4
MD
57
58#include "hammer2.h"
59#include "hammer2_disk.h"
60#include "hammer2_mount.h"
61
355d67fc
MD
62#include "hammer2.h"
63#include "hammer2_lz4.h"
64
65#include "zlib/hammer2_zlib.h"
66
0dea3156
MD
67#define REPORT_REFS_ERRORS 1 /* XXX remove me */
68
355d67fc
MD
69MALLOC_DEFINE(M_OBJCACHE, "objcache", "Object Cache");
70
b7926f31 71struct hammer2_sync_info {
0dea3156 72 hammer2_trans_t trans;
b7926f31
MD
73 int error;
74 int waitfor;
75};
76
e4e20f48 77TAILQ_HEAD(hammer2_mntlist, hammer2_mount);
58e43599 78TAILQ_HEAD(hammer2_pfslist, hammer2_pfsmount);
e4e20f48 79static struct hammer2_mntlist hammer2_mntlist;
58e43599 80static struct hammer2_pfslist hammer2_pfslist;
e4e20f48
MD
81static struct lock hammer2_mntlk;
82
37aa19df 83int hammer2_debug;
a98aa0b0 84int hammer2_cluster_enable = 1;
99535653 85int hammer2_hardlink_enable = 1;
3f5b8b3b 86int hammer2_flush_pipe = 100;
052e0aa0 87int hammer2_synchronous_flush = 1;
837bd39b 88int hammer2_dio_count;
0924b3f8 89long hammer2_limit_dirty_chains;
01eabad4
MD
90long hammer2_iod_file_read;
91long hammer2_iod_meta_read;
92long hammer2_iod_indr_read;
a98aa0b0
MD
93long hammer2_iod_fmap_read;
94long hammer2_iod_volu_read;
01eabad4
MD
95long hammer2_iod_file_write;
96long hammer2_iod_meta_write;
97long hammer2_iod_indr_write;
9061bde5 98long hammer2_iod_fmap_write;
01eabad4
MD
99long hammer2_iod_volu_write;
100long hammer2_ioa_file_read;
101long hammer2_ioa_meta_read;
102long hammer2_ioa_indr_read;
a98aa0b0
MD
103long hammer2_ioa_fmap_read;
104long hammer2_ioa_volu_read;
9061bde5 105long hammer2_ioa_fmap_write;
01eabad4
MD
106long hammer2_ioa_file_write;
107long hammer2_ioa_meta_write;
108long hammer2_ioa_indr_write;
109long hammer2_ioa_volu_write;
37aa19df 110
355d67fc
MD
111MALLOC_DECLARE(C_BUFFER);
112MALLOC_DEFINE(C_BUFFER, "compbuffer", "Buffer used for compression.");
113
114MALLOC_DECLARE(D_BUFFER);
115MALLOC_DEFINE(D_BUFFER, "decompbuffer", "Buffer used for decompression.");
116
37aa19df
MD
117SYSCTL_NODE(_vfs, OID_AUTO, hammer2, CTLFLAG_RW, 0, "HAMMER2 filesystem");
118
119SYSCTL_INT(_vfs_hammer2, OID_AUTO, debug, CTLFLAG_RW,
120 &hammer2_debug, 0, "");
01eabad4
MD
121SYSCTL_INT(_vfs_hammer2, OID_AUTO, cluster_enable, CTLFLAG_RW,
122 &hammer2_cluster_enable, 0, "");
e708f8b9
MD
123SYSCTL_INT(_vfs_hammer2, OID_AUTO, hardlink_enable, CTLFLAG_RW,
124 &hammer2_hardlink_enable, 0, "");
3f5b8b3b
MD
125SYSCTL_INT(_vfs_hammer2, OID_AUTO, flush_pipe, CTLFLAG_RW,
126 &hammer2_flush_pipe, 0, "");
8138a154
MD
127SYSCTL_INT(_vfs_hammer2, OID_AUTO, synchronous_flush, CTLFLAG_RW,
128 &hammer2_synchronous_flush, 0, "");
0924b3f8
MD
129SYSCTL_LONG(_vfs_hammer2, OID_AUTO, limit_dirty_chains, CTLFLAG_RW,
130 &hammer2_limit_dirty_chains, 0, "");
837bd39b
MD
131SYSCTL_INT(_vfs_hammer2, OID_AUTO, dio_count, CTLFLAG_RD,
132 &hammer2_dio_count, 0, "");
a98aa0b0 133
01eabad4
MD
134SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_file_read, CTLFLAG_RW,
135 &hammer2_iod_file_read, 0, "");
136SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_meta_read, CTLFLAG_RW,
137 &hammer2_iod_meta_read, 0, "");
138SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_indr_read, CTLFLAG_RW,
139 &hammer2_iod_indr_read, 0, "");
a98aa0b0
MD
140SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_fmap_read, CTLFLAG_RW,
141 &hammer2_iod_fmap_read, 0, "");
142SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_volu_read, CTLFLAG_RW,
143 &hammer2_iod_volu_read, 0, "");
144
01eabad4
MD
145SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_file_write, CTLFLAG_RW,
146 &hammer2_iod_file_write, 0, "");
147SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_meta_write, CTLFLAG_RW,
148 &hammer2_iod_meta_write, 0, "");
149SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_indr_write, CTLFLAG_RW,
150 &hammer2_iod_indr_write, 0, "");
a98aa0b0
MD
151SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_fmap_write, CTLFLAG_RW,
152 &hammer2_iod_fmap_write, 0, "");
01eabad4
MD
153SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_volu_write, CTLFLAG_RW,
154 &hammer2_iod_volu_write, 0, "");
a98aa0b0 155
01eabad4
MD
156SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_file_read, CTLFLAG_RW,
157 &hammer2_ioa_file_read, 0, "");
158SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_meta_read, CTLFLAG_RW,
159 &hammer2_ioa_meta_read, 0, "");
160SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_indr_read, CTLFLAG_RW,
161 &hammer2_ioa_indr_read, 0, "");
a98aa0b0
MD
162SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_fmap_read, CTLFLAG_RW,
163 &hammer2_ioa_fmap_read, 0, "");
164SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_volu_read, CTLFLAG_RW,
165 &hammer2_ioa_volu_read, 0, "");
166
01eabad4
MD
167SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_file_write, CTLFLAG_RW,
168 &hammer2_ioa_file_write, 0, "");
169SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_meta_write, CTLFLAG_RW,
170 &hammer2_ioa_meta_write, 0, "");
171SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_indr_write, CTLFLAG_RW,
172 &hammer2_ioa_indr_write, 0, "");
a98aa0b0
MD
173SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_fmap_write, CTLFLAG_RW,
174 &hammer2_ioa_fmap_write, 0, "");
01eabad4
MD
175SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_volu_write, CTLFLAG_RW,
176 &hammer2_ioa_volu_write, 0, "");
37aa19df 177
b7926f31 178static int hammer2_vfs_init(struct vfsconf *conf);
355d67fc 179static int hammer2_vfs_uninit(struct vfsconf *vfsp);
b7926f31
MD
180static int hammer2_vfs_mount(struct mount *mp, char *path, caddr_t data,
181 struct ucred *cred);
10136ab6
MD
182static int hammer2_remount(hammer2_mount_t *, struct mount *, char *,
183 struct vnode *, struct ucred *);
184static int hammer2_recovery(hammer2_mount_t *hmp);
b7926f31
MD
185static int hammer2_vfs_unmount(struct mount *mp, int mntflags);
186static int hammer2_vfs_root(struct mount *mp, struct vnode **vpp);
187static int hammer2_vfs_statfs(struct mount *mp, struct statfs *sbp,
703720e4 188 struct ucred *cred);
b7926f31
MD
189static int hammer2_vfs_statvfs(struct mount *mp, struct statvfs *sbp,
190 struct ucred *cred);
b7926f31
MD
191static int hammer2_vfs_vget(struct mount *mp, struct vnode *dvp,
192 ino_t ino, struct vnode **vpp);
193static int hammer2_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
194 struct fid *fhp, struct vnode **vpp);
195static int hammer2_vfs_vptofh(struct vnode *vp, struct fid *fhp);
196static int hammer2_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
197 int *exflagsp, struct ucred **credanonp);
198
199static int hammer2_install_volume_header(hammer2_mount_t *hmp);
b7926f31 200static int hammer2_sync_scan2(struct mount *mp, struct vnode *vp, void *data);
50e4f8f4 201
355d67fc
MD
202static void hammer2_write_thread(void *arg);
203
623d43d4
MD
204static void hammer2_vfs_unmount_hmp1(struct mount *mp, hammer2_mount_t *hmp);
205static void hammer2_vfs_unmount_hmp2(struct mount *mp, hammer2_mount_t *hmp);
206
355d67fc
MD
207/*
208 * Functions for compression in threads,
209 * from hammer2_vnops.c
210 */
1e5c08ba 211static void hammer2_write_file_core(struct buf *bp, hammer2_trans_t *trans,
355d67fc
MD
212 hammer2_inode_t *ip,
213 hammer2_inode_data_t *ipdata,
278ab2b2 214 hammer2_cluster_t *cparent,
355d67fc
MD
215 hammer2_key_t lbase, int ioflag, int pblksize,
216 int *errorp);
1e5c08ba 217static void hammer2_compress_and_write(struct buf *bp, hammer2_trans_t *trans,
355d67fc 218 hammer2_inode_t *ip,
6a5f4fe6 219 const hammer2_inode_data_t *ipdata,
278ab2b2 220 hammer2_cluster_t *cparent,
355d67fc 221 hammer2_key_t lbase, int ioflag,
e07becf8
MD
222 int pblksize, int *errorp,
223 int comp_algo, int check_algo);
1e5c08ba 224static void hammer2_zero_check_and_write(struct buf *bp,
355d67fc 225 hammer2_trans_t *trans, hammer2_inode_t *ip,
6a5f4fe6 226 const hammer2_inode_data_t *ipdata,
278ab2b2 227 hammer2_cluster_t *cparent,
355d67fc 228 hammer2_key_t lbase,
e07becf8
MD
229 int ioflag, int pblksize, int *errorp,
230 int check_algo);
f481450f 231static int test_block_zeros(const char *buf, size_t bytes);
1e5c08ba 232static void zero_write(struct buf *bp, hammer2_trans_t *trans,
355d67fc 233 hammer2_inode_t *ip,
6a5f4fe6 234 const hammer2_inode_data_t *ipdata,
278ab2b2 235 hammer2_cluster_t *cparent,
1e5c08ba
MD
236 hammer2_key_t lbase,
237 int *errorp);
278ab2b2 238static void hammer2_write_bp(hammer2_cluster_t *cluster, struct buf *bp,
e07becf8
MD
239 int ioflag, int pblksize, int *errorp,
240 int check_algo);
355d67fc 241
537d97bc
MD
242static int hammer2_rcvdmsg(kdmsg_msg_t *msg);
243static void hammer2_autodmsg(kdmsg_msg_t *msg);
063639ba 244static int hammer2_lnk_span_reply(kdmsg_state_t *state, kdmsg_msg_t *msg);
bfc3a7b1 245
355d67fc 246
703720e4
MD
247/*
248 * HAMMER2 vfs operations.
249 */
250static struct vfsops hammer2_vfsops = {
066e00cc 251 .vfs_init = hammer2_vfs_init,
58e43599 252 .vfs_uninit = hammer2_vfs_uninit,
066e00cc
MD
253 .vfs_sync = hammer2_vfs_sync,
254 .vfs_mount = hammer2_vfs_mount,
255 .vfs_unmount = hammer2_vfs_unmount,
256 .vfs_root = hammer2_vfs_root,
257 .vfs_statfs = hammer2_vfs_statfs,
258 .vfs_statvfs = hammer2_vfs_statvfs,
259 .vfs_vget = hammer2_vfs_vget,
260 .vfs_vptofh = hammer2_vfs_vptofh,
261 .vfs_fhtovp = hammer2_vfs_fhtovp,
262 .vfs_checkexp = hammer2_vfs_checkexp
703720e4
MD
263};
264
703720e4
MD
265MALLOC_DEFINE(M_HAMMER2, "HAMMER2-mount", "");
266
267VFS_SET(hammer2_vfsops, hammer2, 0);
268MODULE_VERSION(hammer2, 1);
269
e118c14f
MD
270static
271int
066e00cc 272hammer2_vfs_init(struct vfsconf *conf)
703720e4 273{
355d67fc
MD
274 static struct objcache_malloc_args margs_read;
275 static struct objcache_malloc_args margs_write;
276
703720e4
MD
277 int error;
278
279 error = 0;
280
0e92b724 281 if (HAMMER2_BLOCKREF_BYTES != sizeof(struct hammer2_blockref))
703720e4 282 error = EINVAL;
0e92b724 283 if (HAMMER2_INODE_BYTES != sizeof(struct hammer2_inode_data))
703720e4 284 error = EINVAL;
0e92b724 285 if (HAMMER2_VOLUME_BYTES != sizeof(struct hammer2_volume_data))
703720e4
MD
286 error = EINVAL;
287
288 if (error)
289 kprintf("HAMMER2 structure size mismatch; cannot continue.\n");
355d67fc
MD
290
291 margs_read.objsize = 65536;
292 margs_read.mtype = D_BUFFER;
293
294 margs_write.objsize = 32768;
295 margs_write.mtype = C_BUFFER;
296
297 cache_buffer_read = objcache_create(margs_read.mtype->ks_shortdesc,
298 0, 1, NULL, NULL, NULL, objcache_malloc_alloc,
299 objcache_malloc_free, &margs_read);
300 cache_buffer_write = objcache_create(margs_write.mtype->ks_shortdesc,
301 0, 1, NULL, NULL, NULL, objcache_malloc_alloc,
302 objcache_malloc_free, &margs_write);
703720e4 303
e4e20f48
MD
304 lockinit(&hammer2_mntlk, "mntlk", 0, 0);
305 TAILQ_INIT(&hammer2_mntlist);
58e43599 306 TAILQ_INIT(&hammer2_pfslist);
e4e20f48 307
0924b3f8
MD
308 hammer2_limit_dirty_chains = desiredvnodes / 10;
309
50456506
MD
310 hammer2_trans_manage_init();
311
703720e4
MD
312 return (error);
313}
314
355d67fc
MD
315static
316int
317hammer2_vfs_uninit(struct vfsconf *vfsp __unused)
318{
319 objcache_destroy(cache_buffer_read);
320 objcache_destroy(cache_buffer_write);
321 return 0;
322}
323
50456506
MD
324/*
325 * Core PFS allocator. Used to allocate the pmp structure for PFS cluster
326 * mounts and the spmp structure for media (hmp) structures.
327 */
328static hammer2_pfsmount_t *
329hammer2_pfsalloc(const hammer2_inode_data_t *ipdata, hammer2_tid_t alloc_tid)
330{
331 hammer2_pfsmount_t *pmp;
332
333 pmp = kmalloc(sizeof(*pmp), M_HAMMER2, M_WAITOK | M_ZERO);
334 kmalloc_create(&pmp->minode, "HAMMER2-inodes");
335 kmalloc_create(&pmp->mmsg, "HAMMER2-pfsmsg");
336 lockinit(&pmp->lock, "pfslk", 0, 0);
ba87a4ab 337 spin_init(&pmp->inum_spin, "hm2pfsalloc_inum");
50456506
MD
338 RB_INIT(&pmp->inum_tree);
339 TAILQ_INIT(&pmp->unlinkq);
ba87a4ab 340 spin_init(&pmp->list_spin, "hm2pfsalloc_list");
50456506
MD
341
342 pmp->alloc_tid = alloc_tid + 1; /* our first media transaction id */
343 pmp->flush_tid = pmp->alloc_tid;
344 if (ipdata) {
345 pmp->inode_tid = ipdata->pfs_inum + 1;
346 pmp->pfs_clid = ipdata->pfs_clid;
347 }
348 mtx_init(&pmp->wthread_mtx);
349 bioq_init(&pmp->wthread_bioq);
350
351 return pmp;
352}
353
703720e4
MD
354/*
355 * Mount or remount HAMMER2 fileystem from physical media
356 *
357 * mountroot
358 * mp mount point structure
359 * path NULL
360 * data <unused>
361 * cred <unused>
362 *
363 * mount
364 * mp mount point structure
365 * path path to mount point
366 * data pointer to argument structure in user space
367 * volume volume path (device@LABEL form)
368 * hflags user mount flags
369 * cred user credentials
370 *
371 * RETURNS: 0 Success
372 * !0 error number
373 */
e118c14f
MD
374static
375int
066e00cc 376hammer2_vfs_mount(struct mount *mp, char *path, caddr_t data,
bfc3a7b1 377 struct ucred *cred)
703720e4
MD
378{
379 struct hammer2_mount_info info;
e4e20f48 380 hammer2_pfsmount_t *pmp;
50456506 381 hammer2_pfsmount_t *spmp;
54eb943b 382 hammer2_mount_t *hmp;
1897c66e
MD
383 hammer2_key_t key_next;
384 hammer2_key_t key_dummy;
7cfa8da5 385 hammer2_key_t lhc;
703720e4
MD
386 struct vnode *devvp;
387 struct nlookupdata nd;
5c23d7f1 388 hammer2_chain_t *parent;
7cfa8da5 389 hammer2_chain_t *rchain;
278ab2b2
MD
390 hammer2_cluster_t *cluster;
391 hammer2_cluster_t *cparent;
58e43599 392 const hammer2_inode_data_t *ipdata;
50456506 393 hammer2_blockref_t bref;
3a5aa68f 394 struct file *fp;
703720e4
MD
395 char devstr[MNAMELEN];
396 size_t size;
397 size_t done;
50e4f8f4
MD
398 char *dev;
399 char *label;
54eb943b 400 int ronly = 1;
703720e4 401 int error;
1897c66e 402 int cache_index;
278ab2b2 403 int ddflag;
7bed8d7e 404 int i;
703720e4
MD
405
406 hmp = NULL;
e4e20f48 407 pmp = NULL;
50e4f8f4
MD
408 dev = NULL;
409 label = NULL;
703720e4 410 devvp = NULL;
1897c66e 411 cache_index = -1;
703720e4
MD
412
413 kprintf("hammer2_mount\n");
414
415 if (path == NULL) {
416 /*
417 * Root mount
418 */
bfc3a7b1
MD
419 bzero(&info, sizeof(info));
420 info.cluster_fd = -1;
703720e4
MD
421 return (EOPNOTSUPP);
422 } else {
423 /*
424 * Non-root mount or updating a mount
425 */
703720e4
MD
426 error = copyin(data, &info, sizeof(info));
427 if (error)
428 return (error);
429
430 error = copyinstr(info.volume, devstr, MNAMELEN - 1, &done);
431 if (error)
432 return (error);
433
434 /* Extract device and label */
435 dev = devstr;
436 label = strchr(devstr, '@');
437 if (label == NULL ||
50e4f8f4 438 ((label + 1) - dev) > done) {
703720e4 439 return (EINVAL);
50e4f8f4 440 }
703720e4
MD
441 *label = '\0';
442 label++;
443 if (*label == '\0')
444 return (EINVAL);
445
446 if (mp->mnt_flag & MNT_UPDATE) {
447 /* Update mount */
448 /* HAMMER2 implements NFS export via mountctl */
7bed8d7e 449 pmp = MPTOPMP(mp);
58e43599
MD
450 for (i = 0; i < pmp->iroot->cluster.nchains; ++i) {
451 hmp = pmp->iroot->cluster.array[i]->hmp;
7bed8d7e 452 devvp = hmp->devvp;
10136ab6
MD
453 error = hammer2_remount(hmp, mp, path,
454 devvp, cred);
7bed8d7e
MD
455 if (error)
456 break;
457 }
b53e7c4f 458 /*hammer2_inode_install_hidden(pmp);*/
044541cd 459
54eb943b 460 return error;
703720e4
MD
461 }
462 }
463
464 /*
58e43599 465 * HMP device mount
bfc3a7b1
MD
466 *
467 * Lookup name and verify it refers to a block device.
703720e4 468 */
703720e4 469 error = nlookup_init(&nd, dev, UIO_SYSSPACE, NLC_FOLLOW);
54eb943b
MD
470 if (error == 0)
471 error = nlookup(&nd);
472 if (error == 0)
473 error = cache_vref(&nd.nl_nch, nd.nl_cred, &devvp);
703720e4
MD
474 nlookup_done(&nd);
475
54eb943b
MD
476 if (error == 0) {
477 if (vn_isdisk(devvp, &error))
478 error = vfs_mountedon(devvp);
703720e4
MD
479 }
480
481 /*
e4e20f48
MD
482 * Determine if the device has already been mounted. After this
483 * check hmp will be non-NULL if we are doing the second or more
484 * hammer2 mounts from the same device.
703720e4 485 */
e4e20f48
MD
486 lockmgr(&hammer2_mntlk, LK_EXCLUSIVE);
487 TAILQ_FOREACH(hmp, &hammer2_mntlist, mntentry) {
488 if (hmp->devvp == devvp)
489 break;
490 }
491
492 /*
a5913bdf
MD
493 * Open the device if this isn't a secondary mount and construct
494 * the H2 device mount (hmp).
e4e20f48 495 */
a5913bdf 496 if (hmp == NULL) {
50456506
MD
497 hammer2_chain_t *schain;
498 hammer2_xid_t xid;
499
e4e20f48
MD
500 if (error == 0 && vcount(devvp) > 0)
501 error = EBUSY;
502
503 /*
504 * Now open the device
505 */
54eb943b 506 if (error == 0) {
e4e20f48
MD
507 ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
508 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
509 error = vinvalbuf(devvp, V_SAVE, 0, 0);
510 if (error == 0) {
511 error = VOP_OPEN(devvp,
512 ronly ? FREAD : FREAD | FWRITE,
513 FSCRED, NULL);
514 }
515 vn_unlock(devvp);
516 }
517 if (error && devvp) {
518 vrele(devvp);
519 devvp = NULL;
520 }
521 if (error) {
522 lockmgr(&hammer2_mntlk, LK_RELEASE);
523 return error;
54eb943b 524 }
e4e20f48
MD
525 hmp = kmalloc(sizeof(*hmp), M_HAMMER2, M_WAITOK | M_ZERO);
526 hmp->ronly = ronly;
527 hmp->devvp = devvp;
e4e20f48
MD
528 kmalloc_create(&hmp->mchain, "HAMMER2-chains");
529 TAILQ_INSERT_TAIL(&hammer2_mntlist, hmp, mntentry);
fdf62707 530 RB_INIT(&hmp->iotree);
ba87a4ab
SW
531 spin_init(&hmp->io_spin, "hm2mount_io");
532 spin_init(&hmp->list_spin, "hm2mount_list");
da6f36f4 533 TAILQ_INIT(&hmp->flushq);
50e4f8f4 534
50456506 535 lockinit(&hmp->vollk, "h2vol", 0, 0);
0dea3156 536
e4e20f48 537 /*
1a7cfe5a 538 * vchain setup. vchain.data is embedded.
e4e20f48 539 * vchain.refs is initialized and will never drop to 0.
10136ab6
MD
540 *
541 * NOTE! voldata is not yet loaded.
e4e20f48 542 */
0dea3156 543 hmp->vchain.hmp = hmp;
e4e20f48
MD
544 hmp->vchain.refs = 1;
545 hmp->vchain.data = (void *)&hmp->voldata;
004f88b4 546 hmp->vchain.bref.type = HAMMER2_BREF_TYPE_VOLUME;
e4e20f48 547 hmp->vchain.bref.data_off = 0 | HAMMER2_PBUFRADIX;
50456506 548 hmp->vchain.bref.mirror_tid = hmp->voldata.mirror_tid;
10136ab6 549
da6f36f4 550 hammer2_chain_core_alloc(NULL, &hmp->vchain);
e4e20f48 551 /* hmp->vchain.u.xxx is left NULL */
232a50f9 552
1a7cfe5a
MD
553 /*
554 * fchain setup. fchain.data is embedded.
555 * fchain.refs is initialized and will never drop to 0.
556 *
557 * The data is not used but needs to be initialized to
558 * pass assertion muster. We use this chain primarily
559 * as a placeholder for the freemap's top-level RBTREE
560 * so it does not interfere with the volume's topology
561 * RBTREE.
562 */
563 hmp->fchain.hmp = hmp;
564 hmp->fchain.refs = 1;
565 hmp->fchain.data = (void *)&hmp->voldata.freemap_blockset;
566 hmp->fchain.bref.type = HAMMER2_BREF_TYPE_FREEMAP;
567 hmp->fchain.bref.data_off = 0 | HAMMER2_PBUFRADIX;
50456506 568 hmp->fchain.bref.mirror_tid = hmp->voldata.freemap_tid;
512beabd
MD
569 hmp->fchain.bref.methods =
570 HAMMER2_ENC_CHECK(HAMMER2_CHECK_FREEMAP) |
571 HAMMER2_ENC_COMP(HAMMER2_COMP_NONE);
512beabd 572
da6f36f4 573 hammer2_chain_core_alloc(NULL, &hmp->fchain);
1a7cfe5a
MD
574 /* hmp->fchain.u.xxx is left NULL */
575
e4e20f48 576 /*
10136ab6
MD
577 * Install the volume header and initialize fields from
578 * voldata.
e4e20f48
MD
579 */
580 error = hammer2_install_volume_header(hmp);
581 if (error) {
623d43d4
MD
582 ++hmp->pmp_count;
583 hammer2_vfs_unmount_hmp1(mp, hmp);
584 hammer2_vfs_unmount_hmp2(mp, hmp);
2c9c0b65 585 lockmgr(&hammer2_mntlk, LK_RELEASE);
e4e20f48
MD
586 hammer2_vfs_unmount(mp, MNT_FORCE);
587 return error;
588 }
a5913bdf 589
8138a154
MD
590 /*
591 * Really important to get these right or flush will get
592 * confused.
593 */
50456506
MD
594 hmp->spmp = hammer2_pfsalloc(NULL, hmp->voldata.mirror_tid);
595 kprintf("alloc spmp %p tid %016jx\n",
596 hmp->spmp, hmp->voldata.mirror_tid);
597 spmp = hmp->spmp;
598 spmp->inode_tid = 1;
599
600 xid = 0;
a4dc31e0 601 hmp->vchain.bref.mirror_tid = hmp->voldata.mirror_tid;
50456506 602 hmp->vchain.bref.modify_tid = hmp->vchain.bref.mirror_tid;
50456506 603 hmp->vchain.pmp = spmp;
a4dc31e0 604 hmp->fchain.bref.mirror_tid = hmp->voldata.freemap_tid;
50456506 605 hmp->fchain.bref.modify_tid = hmp->fchain.bref.mirror_tid;
50456506 606 hmp->fchain.pmp = spmp;
10136ab6 607
a5913bdf
MD
608 /*
609 * First locate the super-root inode, which is key 0
610 * relative to the volume header's blockset.
611 *
612 * Then locate the root inode by scanning the directory keyspace
613 * represented by the label.
614 */
615 parent = hammer2_chain_lookup_init(&hmp->vchain, 0);
1897c66e
MD
616 schain = hammer2_chain_lookup(&parent, &key_dummy,
617 HAMMER2_SROOT_KEY, HAMMER2_SROOT_KEY,
278ab2b2 618 &cache_index, 0, &ddflag);
a5913bdf
MD
619 hammer2_chain_lookup_done(parent);
620 if (schain == NULL) {
621 kprintf("hammer2_mount: invalid super-root\n");
623d43d4
MD
622 ++hmp->pmp_count;
623 hammer2_vfs_unmount_hmp1(mp, hmp);
624 hammer2_vfs_unmount_hmp2(mp, hmp);
2c9c0b65 625 lockmgr(&hammer2_mntlk, LK_RELEASE);
a5913bdf
MD
626 hammer2_vfs_unmount(mp, MNT_FORCE);
627 return EINVAL;
628 }
7bed8d7e
MD
629
630 /*
50456506
MD
631 * Sanity-check schain's pmp, finish initializing spmp.
632 */
633 KKASSERT(schain->pmp == spmp);
634 spmp->pfs_clid = schain->data->ipdata.pfs_clid;
635
636 /*
7bed8d7e
MD
637 * NOTE: inode_get sucks up schain's lock.
638 */
278ab2b2 639 cluster = hammer2_cluster_from_chain(schain);
50456506
MD
640 spmp->iroot = hammer2_inode_get(spmp, NULL, cluster);
641 spmp->spmp_hmp = hmp;
642 hammer2_inode_ref(spmp->iroot);
643 hammer2_inode_unlock_ex(spmp->iroot, cluster);
a5913bdf 644 schain = NULL;
50456506 645 /* leave spmp->iroot with one ref */
10136ab6
MD
646
647 if ((mp->mnt_flag & MNT_RDONLY) == 0) {
648 error = hammer2_recovery(hmp);
649 /* XXX do something with error */
650 }
063639ba
MD
651 ++hmp->pmp_count;
652
653 /*
654 * XXX RDONLY stuff is totally broken FIXME XXX
655 *
656 * Automatic LNK_CONN
657 * Automatic handling of received LNK_SPAN
658 * Automatic handling of received LNK_CIRC
659 * No automatic LNK_SPAN generation - we do this ourselves
660 * No automatic LNK_CIRC generation - we do this ourselves
661 */
662 kdmsg_iocom_init(&hmp->iocom, hmp,
663 KDMSG_IOCOMF_AUTOCONN |
da6f36f4 664 KDMSG_IOCOMF_AUTORXSPAN,
063639ba
MD
665 hmp->mchain, hammer2_rcvdmsg);
666
667 /*
668 * Ref the cluster management messaging descriptor. The mount
669 * program deals with the other end of the communications pipe.
670 */
671 fp = holdfp(curproc->p_fd, info.cluster_fd, -1);
672 if (fp) {
673 hammer2_cluster_reconnect(hmp, fp);
674 } else {
675 kprintf("hammer2_mount: bad cluster_fd!\n");
676 }
50456506
MD
677 } else {
678 spmp = hmp->spmp;
063639ba 679 ++hmp->pmp_count;
50e4f8f4 680 }
5c23d7f1 681
10252dc7 682 /*
7bed8d7e 683 * Lookup mount point under the media-localized super-root.
50456506
MD
684 *
685 * cluster->pmp will incorrectly point to spmp and must be fixed
686 * up later on.
10252dc7 687 */
50456506 688 cparent = hammer2_inode_lock_ex(spmp->iroot);
e4e20f48 689 lhc = hammer2_dirhash(label, strlen(label));
278ab2b2 690 cluster = hammer2_cluster_lookup(cparent, &key_next,
c667909f 691 lhc, lhc + HAMMER2_DIRHASH_LOMASK,
278ab2b2
MD
692 0, &ddflag);
693 while (cluster) {
694 if (hammer2_cluster_type(cluster) == HAMMER2_BREF_TYPE_INODE &&
695 strcmp(label,
696 hammer2_cluster_data(cluster)->ipdata.filename) == 0) {
7cfa8da5
MD
697 break;
698 }
278ab2b2 699 cluster = hammer2_cluster_next(cparent, cluster, &key_next,
1897c66e 700 key_next,
278ab2b2 701 lhc + HAMMER2_DIRHASH_LOMASK, 0);
7cfa8da5 702 }
50456506 703 hammer2_inode_unlock_ex(spmp->iroot, cparent);
7bed8d7e 704
278ab2b2 705 if (cluster == NULL) {
e4e20f48 706 kprintf("hammer2_mount: PFS label not found\n");
623d43d4
MD
707 hammer2_vfs_unmount_hmp1(mp, hmp);
708 hammer2_vfs_unmount_hmp2(mp, hmp);
58e43599 709 lockmgr(&hammer2_mntlk, LK_RELEASE);
9c2e0de0 710 hammer2_vfs_unmount(mp, MNT_FORCE);
7cfa8da5
MD
711 return EINVAL;
712 }
278ab2b2
MD
713
714 for (i = 0; i < cluster->nchains; ++i) {
715 rchain = cluster->array[i];
716 if (rchain->flags & HAMMER2_CHAIN_MOUNTED) {
717 kprintf("hammer2_mount: PFS label already mounted!\n");
718 hammer2_cluster_unlock(cluster);
719 hammer2_vfs_unmount_hmp1(mp, hmp);
720 hammer2_vfs_unmount_hmp2(mp, hmp);
58e43599 721 lockmgr(&hammer2_mntlk, LK_RELEASE);
278ab2b2
MD
722 hammer2_vfs_unmount(mp, MNT_FORCE);
723 return EBUSY;
724 }
da6f36f4 725 KKASSERT(rchain->pmp == NULL);
9b6b3df4 726#if 0
278ab2b2
MD
727 if (rchain->flags & HAMMER2_CHAIN_RECYCLE) {
728 kprintf("hammer2_mount: PFS label is recycling\n");
729 hammer2_cluster_unlock(cluster);
730 hammer2_vfs_unmount_hmp1(mp, hmp);
731 hammer2_vfs_unmount_hmp2(mp, hmp);
58e43599 732 lockmgr(&hammer2_mntlk, LK_RELEASE);
278ab2b2
MD
733 hammer2_vfs_unmount(mp, MNT_FORCE);
734 return EBUSY;
735 }
9b6b3df4 736#endif
278ab2b2
MD
737 }
738
58e43599
MD
739 /*
740 * Check to see if the cluster id is already mounted at the mount
741 * point. If it is, add us to the cluster.
742 */
743 ipdata = &hammer2_cluster_data(cluster)->ipdata;
50456506 744 hammer2_cluster_bref(cluster, &bref);
58e43599 745 TAILQ_FOREACH(pmp, &hammer2_pfslist, mntentry) {
50456506
MD
746 if (pmp->spmp_hmp == NULL &&
747 bcmp(&pmp->pfs_clid, &ipdata->pfs_clid,
748 sizeof(pmp->pfs_clid)) == 0) {
58e43599 749 break;
50456506 750 }
58e43599 751 }
50456506 752
58e43599
MD
753 if (pmp) {
754 int i;
755 int j;
756
757 hammer2_inode_ref(pmp->iroot);
758 ccms_thread_lock(&pmp->iroot->topo_cst, CCMS_STATE_EXCLUSIVE);
759
760 if (pmp->iroot->cluster.nchains + cluster->nchains >
761 HAMMER2_MAXCLUSTER) {
762 kprintf("hammer2_mount: cluster full!\n");
763
764 ccms_thread_unlock(&pmp->iroot->topo_cst);
765 hammer2_inode_drop(pmp->iroot);
766
767 hammer2_cluster_unlock(cluster);
768 hammer2_vfs_unmount_hmp1(mp, hmp);
769 hammer2_vfs_unmount_hmp2(mp, hmp);
770 lockmgr(&hammer2_mntlk, LK_RELEASE);
771 hammer2_vfs_unmount(mp, MNT_FORCE);
772 return EBUSY;
773 }
774 kprintf("hammer2_vfs_mount: Adding pfs to existing cluster\n");
775 j = pmp->iroot->cluster.nchains;
776 for (i = 0; i < cluster->nchains; ++i) {
50456506
MD
777 rchain = cluster->array[i];
778 KKASSERT(rchain->pmp == NULL);
779 rchain->pmp = pmp;
58e43599
MD
780 hammer2_chain_ref(cluster->array[i]);
781 pmp->iroot->cluster.array[j] = cluster->array[i];
782 ++j;
783 }
784 pmp->iroot->cluster.nchains = j;
785 ccms_thread_unlock(&pmp->iroot->topo_cst);
786 hammer2_inode_drop(pmp->iroot);
787 hammer2_cluster_unlock(cluster);
788 lockmgr(&hammer2_mntlk, LK_RELEASE);
789
790 kprintf("ok\n");
b53e7c4f 791 hammer2_inode_install_hidden(pmp);
58e43599
MD
792
793 return ERANGE;
794 }
795
796 /*
797 * Block device opened successfully, finish initializing the
798 * mount structure.
799 *
800 * From this point on we have to call hammer2_unmount() on failure.
801 */
50456506
MD
802 pmp = hammer2_pfsalloc(ipdata, bref.mirror_tid);
803 kprintf("PMP mirror_tid is %016jx\n", bref.mirror_tid);
804 for (i = 0; i < cluster->nchains; ++i) {
805 rchain = cluster->array[i];
806 KKASSERT(rchain->pmp == NULL);
807 rchain->pmp = pmp;
808 atomic_set_int(&rchain->flags, HAMMER2_CHAIN_MOUNTED);
809 }
810 cluster->pmp = pmp;
58e43599 811
58e43599
MD
812 ccms_domain_init(&pmp->ccms_dom);
813 TAILQ_INSERT_TAIL(&hammer2_pfslist, pmp, mntentry);
814 lockmgr(&hammer2_mntlk, LK_RELEASE);
815
816 kprintf("hammer2_mount hmp=%p pmp=%p pmpcnt=%d\n",
817 hmp, pmp, hmp->pmp_count);
818
819 mp->mnt_flag = MNT_LOCAL;
820 mp->mnt_kern_flag |= MNTK_ALL_MPSAFE; /* all entry pts are SMP */
821 mp->mnt_kern_flag |= MNTK_THR_SYNC; /* new vsyncscan semantics */
822
823 /*
824 * required mount structure initializations
825 */
826 mp->mnt_stat.f_iosize = HAMMER2_PBUFSIZE;
827 mp->mnt_stat.f_bsize = HAMMER2_PBUFSIZE;
828
829 mp->mnt_vstat.f_frsize = HAMMER2_PBUFSIZE;
830 mp->mnt_vstat.f_bsize = HAMMER2_PBUFSIZE;
831
832 /*
833 * Optional fields
834 */
835 mp->mnt_iosize_max = MAXPHYS;
836 mp->mnt_data = (qaddr_t)pmp;
837 pmp->mp = mp;
838
623d43d4
MD
839 /*
840 * After this point hammer2_vfs_unmount() has visibility on hmp
841 * and manual hmp1/hmp2 calls are not needed on fatal errors.
842 */
278ab2b2 843 pmp->iroot = hammer2_inode_get(pmp, NULL, cluster);
476d2aad 844 hammer2_inode_ref(pmp->iroot); /* ref for pmp->iroot */
278ab2b2 845 hammer2_inode_unlock_ex(pmp->iroot, cluster);
609a8021 846
065f4046
MD
847 /*
848 * The logical file buffer bio write thread handles things
849 * like physical block assignment and compression.
50456506
MD
850 *
851 * (only applicable to pfs mounts, not applicable to spmp)
065f4046 852 */
065f4046
MD
853 pmp->wthread_destroy = 0;
854 lwkt_create(hammer2_write_thread, pmp,
855 &pmp->wthread_td, NULL, 0, -1, "hwrite-%s", label);
856
044541cd
MD
857 /*
858 * With the cluster operational install ihidden.
50456506 859 * (only applicable to pfs mounts, not applicable to spmp)
044541cd
MD
860 */
861 hammer2_inode_install_hidden(pmp);
862
bfc3a7b1
MD
863 /*
864 * Finish setup
865 */
f0206a67
VS
866 vfs_getnewfsid(mp);
867 vfs_add_vnodeops(mp, &hammer2_vnode_vops, &mp->mnt_vn_norm_ops);
868 vfs_add_vnodeops(mp, &hammer2_spec_vops, &mp->mnt_vn_spec_ops);
869 vfs_add_vnodeops(mp, &hammer2_fifo_vops, &mp->mnt_vn_fifo_ops);
703720e4 870
54f522df 871 copyinstr(info.volume, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, &size);
703720e4
MD
872 bzero(mp->mnt_stat.f_mntfromname + size, MNAMELEN - size);
873 bzero(mp->mnt_stat.f_mntonname, sizeof(mp->mnt_stat.f_mntonname));
874 copyinstr(path, mp->mnt_stat.f_mntonname,
875 sizeof(mp->mnt_stat.f_mntonname) - 1,
876 &size);
877
bfc3a7b1
MD
878 /*
879 * Initial statfs to prime mnt_stat.
880 */
066e00cc 881 hammer2_vfs_statfs(mp, &mp->mnt_stat, cred);
355d67fc 882
47902fef 883 return 0;
703720e4
MD
884}
885
355d67fc
MD
886/*
887 * Handle bioq for strategy write
888 */
889static
890void
891hammer2_write_thread(void *arg)
892{
065f4046 893 hammer2_pfsmount_t *pmp;
355d67fc
MD
894 struct bio *bio;
895 struct buf *bp;
896 hammer2_trans_t trans;
897 struct vnode *vp;
355d67fc 898 hammer2_inode_t *ip;
278ab2b2 899 hammer2_cluster_t *cparent;
6a5f4fe6 900 hammer2_inode_data_t *wipdata;
355d67fc
MD
901 hammer2_key_t lbase;
902 int lblksize;
903 int pblksize;
904 int error;
905
065f4046 906 pmp = arg;
355d67fc 907
065f4046
MD
908 mtx_lock(&pmp->wthread_mtx);
909 while (pmp->wthread_destroy == 0) {
910 if (bioq_first(&pmp->wthread_bioq) == NULL) {
911 mtxsleep(&pmp->wthread_bioq, &pmp->wthread_mtx,
355d67fc
MD
912 0, "h2bioqw", 0);
913 }
278ab2b2 914 cparent = NULL;
355d67fc 915
50456506 916 hammer2_trans_init(&trans, pmp, HAMMER2_TRANS_BUFCACHE);
a7720be7 917
065f4046 918 while ((bio = bioq_takefirst(&pmp->wthread_bioq)) != NULL) {
a7720be7 919 /*
a4dc31e0
MD
920 * dummy bio for synchronization. The transaction
921 * must be reinitialized.
a7720be7
MD
922 */
923 if (bio->bio_buf == NULL) {
924 bio->bio_flags |= BIO_DONE;
925 wakeup(bio);
a4dc31e0 926 hammer2_trans_done(&trans);
50456506 927 hammer2_trans_init(&trans, pmp,
a4dc31e0 928 HAMMER2_TRANS_BUFCACHE);
a7720be7
MD
929 continue;
930 }
931
932 /*
933 * else normal bio processing
934 */
065f4046 935 mtx_unlock(&pmp->wthread_mtx);
3f5b8b3b
MD
936
937 hammer2_lwinprog_drop(pmp);
355d67fc
MD
938
939 error = 0;
940 bp = bio->bio_buf;
941 vp = bp->b_vp;
942 ip = VTOI(vp);
943
355d67fc
MD
944 /*
945 * Inode is modified, flush size and mtime changes
946 * to ensure that the file size remains consistent
947 * with the buffers being flushed.
72ebfa75
MD
948 *
949 * NOTE: The inode_fsync() call only flushes the
950 * inode's meta-data state, it doesn't try
951 * to flush underlying buffers or chains.
355d67fc 952 */
278ab2b2 953 cparent = hammer2_inode_lock_ex(ip);
355d67fc
MD
954 if (ip->flags & (HAMMER2_INODE_RESIZED |
955 HAMMER2_INODE_MTIME)) {
278ab2b2 956 hammer2_inode_fsync(&trans, ip, cparent);
355d67fc 957 }
6a5f4fe6 958 wipdata = hammer2_cluster_modify_ip(&trans, ip,
278ab2b2 959 cparent, 0);
355d67fc
MD
960 lblksize = hammer2_calc_logical(ip, bio->bio_offset,
961 &lbase, NULL);
6a5f4fe6
MD
962 pblksize = hammer2_calc_physical(ip, wipdata, lbase);
963 hammer2_write_file_core(bp, &trans, ip, wipdata,
278ab2b2 964 cparent,
355d67fc
MD
965 lbase, IO_ASYNC,
966 pblksize, &error);
6a5f4fe6 967 hammer2_cluster_modsync(cparent);
278ab2b2 968 hammer2_inode_unlock_ex(ip, cparent);
355d67fc 969 if (error) {
1e5c08ba
MD
970 kprintf("hammer2: error in buffer write\n");
971 bp->b_flags |= B_ERROR;
972 bp->b_error = EIO;
355d67fc
MD
973 }
974 biodone(bio);
065f4046 975 mtx_lock(&pmp->wthread_mtx);
355d67fc 976 }
a7720be7 977 hammer2_trans_done(&trans);
355d67fc 978 }
065f4046
MD
979 pmp->wthread_destroy = -1;
980 wakeup(&pmp->wthread_destroy);
355d67fc 981
065f4046 982 mtx_unlock(&pmp->wthread_mtx);
355d67fc
MD
983}
984
a7720be7
MD
985void
986hammer2_bioq_sync(hammer2_pfsmount_t *pmp)
987{
988 struct bio sync_bio;
989
990 bzero(&sync_bio, sizeof(sync_bio)); /* dummy with no bio_buf */
991 mtx_lock(&pmp->wthread_mtx);
72ebfa75
MD
992 if (pmp->wthread_destroy == 0 &&
993 TAILQ_FIRST(&pmp->wthread_bioq.queue)) {
994 bioq_insert_tail(&pmp->wthread_bioq, &sync_bio);
a7720be7
MD
995 while ((sync_bio.bio_flags & BIO_DONE) == 0)
996 mtxsleep(&sync_bio, &pmp->wthread_mtx, 0, "h2bioq", 0);
997 }
998 mtx_unlock(&pmp->wthread_mtx);
999}
1000
355d67fc 1001/*
51a0d27c
MD
1002 * Return a chain suitable for I/O, creating the chain if necessary
1003 * and assigning its physical block.
355d67fc
MD
1004 */
1005static
278ab2b2 1006hammer2_cluster_t *
355d67fc 1007hammer2_assign_physical(hammer2_trans_t *trans,
278ab2b2 1008 hammer2_inode_t *ip, hammer2_cluster_t *cparent,
355d67fc
MD
1009 hammer2_key_t lbase, int pblksize, int *errorp)
1010{
278ab2b2 1011 hammer2_cluster_t *cluster;
84e47819 1012 hammer2_cluster_t *dparent;
1897c66e 1013 hammer2_key_t key_dummy;
355d67fc 1014 int pradix = hammer2_getradix(pblksize);
278ab2b2 1015 int ddflag;
355d67fc
MD
1016
1017 /*
1018 * Locate the chain associated with lbase, return a locked chain.
1019 * However, do not instantiate any data reference (which utilizes a
1020 * device buffer) because we will be using direct IO via the
1021 * logical buffer cache buffer.
1022 */
1023 *errorp = 0;
50456506 1024 KKASSERT(pblksize >= HAMMER2_ALLOC_MIN);
355d67fc 1025retry:
84e47819
MD
1026 dparent = hammer2_cluster_lookup_init(cparent, 0);
1027 cluster = hammer2_cluster_lookup(dparent, &key_dummy,
355d67fc 1028 lbase, lbase,
278ab2b2 1029 HAMMER2_LOOKUP_NODATA, &ddflag);
355d67fc 1030
278ab2b2 1031 if (cluster == NULL) {
355d67fc
MD
1032 /*
1033 * We found a hole, create a new chain entry.
1034 *
1035 * NOTE: DATA chains are created without device backing
1036 * store (nor do we want any).
1037 */
84e47819 1038 *errorp = hammer2_cluster_create(trans, dparent, &cluster,
355d67fc
MD
1039 lbase, HAMMER2_PBUFRADIX,
1040 HAMMER2_BREF_TYPE_DATA,
b3659de2 1041 pblksize, 0);
278ab2b2 1042 if (cluster == NULL) {
84e47819 1043 hammer2_cluster_lookup_done(dparent);
278ab2b2 1044 panic("hammer2_cluster_create: par=%p error=%d\n",
84e47819 1045 dparent->focus, *errorp);
355d67fc
MD
1046 goto retry;
1047 }
355d67fc
MD
1048 /*ip->delta_dcount += pblksize;*/
1049 } else {
278ab2b2 1050 switch (hammer2_cluster_type(cluster)) {
355d67fc
MD
1051 case HAMMER2_BREF_TYPE_INODE:
1052 /*
1053 * The data is embedded in the inode. The
1054 * caller is responsible for marking the inode
1055 * modified and copying the data to the embedded
1056 * area.
1057 */
355d67fc
MD
1058 break;
1059 case HAMMER2_BREF_TYPE_DATA:
05dd26e4 1060 if (hammer2_cluster_need_resize(cluster, pblksize)) {
278ab2b2 1061 hammer2_cluster_resize(trans, ip,
84e47819 1062 dparent, cluster,
355d67fc
MD
1063 pradix,
1064 HAMMER2_MODIFY_OPTDATA);
1065 }
05dd26e4
MD
1066
1067 /*
1068 * DATA buffers must be marked modified whether the
1069 * data is in a logical buffer or not. We also have
1070 * to make this call to fixup the chain data pointers
1071 * after resizing in case this is an encrypted or
1072 * compressed buffer.
1073 */
278ab2b2 1074 hammer2_cluster_modify(trans, cluster,
05dd26e4 1075 HAMMER2_MODIFY_OPTDATA);
355d67fc
MD
1076 break;
1077 default:
1078 panic("hammer2_assign_physical: bad type");
1079 /* NOT REACHED */
355d67fc
MD
1080 break;
1081 }
1082 }
1083
1084 /*
84e47819
MD
1085 * Cleanup. If cluster wound up being the inode itself, i.e.
1086 * the DIRECTDATA case for offset 0, then we need to update cparent.
1087 * The caller expects cparent to not become stale.
355d67fc 1088 */
84e47819
MD
1089 hammer2_cluster_lookup_done(dparent);
1090 /* dparent = NULL; safety */
1091 if (cluster && ddflag)
278ab2b2 1092 hammer2_cluster_replace_locked(cparent, cluster);
278ab2b2 1093 return (cluster);
355d67fc
MD
1094}
1095
1096/*
a71db85d
MD
1097 * bio queued from hammer2_vnops.c.
1098 *
355d67fc 1099 * The core write function which determines which path to take
a71db85d
MD
1100 * depending on compression settings. We also have to locate the
1101 * related clusters so we can calculate and set the check data for
1102 * the blockref.
355d67fc
MD
1103 */
1104static
1105void
1e5c08ba 1106hammer2_write_file_core(struct buf *bp, hammer2_trans_t *trans,
355d67fc 1107 hammer2_inode_t *ip, hammer2_inode_data_t *ipdata,
278ab2b2 1108 hammer2_cluster_t *cparent,
355d67fc
MD
1109 hammer2_key_t lbase, int ioflag, int pblksize,
1110 int *errorp)
1111{
278ab2b2 1112 hammer2_cluster_t *cluster;
f481450f 1113
e07becf8 1114 switch(HAMMER2_DEC_ALGO(ipdata->comp_algo)) {
f481450f 1115 case HAMMER2_COMP_NONE:
355d67fc
MD
1116 /*
1117 * We have to assign physical storage to the buffer
1118 * we intend to dirty or write now to avoid deadlocks
1119 * in the strategy code later.
1120 *
1121 * This can return NOOFFSET for inode-embedded data.
1122 * The strategy code will take care of it in that case.
1123 */
278ab2b2 1124 cluster = hammer2_assign_physical(trans, ip, cparent,
355d67fc
MD
1125 lbase, pblksize,
1126 errorp);
e07becf8
MD
1127 hammer2_write_bp(cluster, bp, ioflag, pblksize, errorp,
1128 ipdata->check_algo);
278ab2b2
MD
1129 if (cluster)
1130 hammer2_cluster_unlock(cluster);
f481450f
MD
1131 break;
1132 case HAMMER2_COMP_AUTOZERO:
1133 /*
1134 * Check for zero-fill only
1135 */
1136 hammer2_zero_check_and_write(bp, trans, ip,
278ab2b2 1137 ipdata, cparent, lbase,
e07becf8
MD
1138 ioflag, pblksize, errorp,
1139 ipdata->check_algo);
f481450f
MD
1140 break;
1141 case HAMMER2_COMP_LZ4:
1142 case HAMMER2_COMP_ZLIB:
1143 default:
1144 /*
1145 * Check for zero-fill and attempt compression.
1146 */
1147 hammer2_compress_and_write(bp, trans, ip,
278ab2b2 1148 ipdata, cparent,
f481450f
MD
1149 lbase, ioflag,
1150 pblksize, errorp,
e07becf8
MD
1151 ipdata->comp_algo,
1152 ipdata->check_algo);
f481450f 1153 break;
355d67fc 1154 }
355d67fc
MD
1155}
1156
1157/*
355d67fc
MD
1158 * Generic function that will perform the compression in compression
1159 * write path. The compression algorithm is determined by the settings
1160 * obtained from inode.
1161 */
1162static
1163void
1e5c08ba 1164hammer2_compress_and_write(struct buf *bp, hammer2_trans_t *trans,
6a5f4fe6 1165 hammer2_inode_t *ip, const hammer2_inode_data_t *ipdata,
278ab2b2 1166 hammer2_cluster_t *cparent,
355d67fc 1167 hammer2_key_t lbase, int ioflag, int pblksize,
e07becf8 1168 int *errorp, int comp_algo, int check_algo)
355d67fc 1169{
278ab2b2 1170 hammer2_cluster_t *cluster;
355d67fc 1171 hammer2_chain_t *chain;
f481450f
MD
1172 int comp_size;
1173 int comp_block_size;
278ab2b2 1174 int i;
f481450f 1175 char *comp_buffer;
355d67fc 1176
f481450f 1177 if (test_block_zeros(bp->b_data, pblksize)) {
278ab2b2 1178 zero_write(bp, trans, ip, ipdata, cparent, lbase, errorp);
f481450f
MD
1179 return;
1180 }
1e5c08ba 1181
f481450f
MD
1182 comp_size = 0;
1183 comp_buffer = NULL;
355d67fc 1184
f481450f 1185 KKASSERT(pblksize / 2 <= 32768);
355d67fc 1186
f481450f
MD
1187 if (ip->comp_heuristic < 8 || (ip->comp_heuristic & 7) == 0) {
1188 z_stream strm_compress;
1189 int comp_level;
1190 int ret;
1191
e07becf8 1192 switch(HAMMER2_DEC_ALGO(comp_algo)) {
f481450f
MD
1193 case HAMMER2_COMP_LZ4:
1194 comp_buffer = objcache_get(cache_buffer_write,
1195 M_INTWAIT);
1196 comp_size = LZ4_compress_limitedOutput(
1197 bp->b_data,
1198 &comp_buffer[sizeof(int)],
1199 pblksize,
1200 pblksize / 2 - sizeof(int));
1201 /*
1202 * We need to prefix with the size, LZ4
1203 * doesn't do it for us. Add the related
1204 * overhead.
1205 */
1206 *(int *)comp_buffer = comp_size;
1207 if (comp_size)
1208 comp_size += sizeof(int);
1209 break;
1210 case HAMMER2_COMP_ZLIB:
1211 comp_level = HAMMER2_DEC_LEVEL(comp_algo);
1212 if (comp_level == 0)
1213 comp_level = 6; /* default zlib compression */
1214 else if (comp_level < 6)
1215 comp_level = 6;
1216 else if (comp_level > 9)
1217 comp_level = 9;
1218 ret = deflateInit(&strm_compress, comp_level);
1219 if (ret != Z_OK) {
1220 kprintf("HAMMER2 ZLIB: fatal error "
1221 "on deflateInit.\n");
1222 }
1223
1224 comp_buffer = objcache_get(cache_buffer_write,
1225 M_INTWAIT);
1226 strm_compress.next_in = bp->b_data;
1227 strm_compress.avail_in = pblksize;
1228 strm_compress.next_out = comp_buffer;
1229 strm_compress.avail_out = pblksize / 2;
1230 ret = deflate(&strm_compress, Z_FINISH);
1231 if (ret == Z_STREAM_END) {
1232 comp_size = pblksize / 2 -
1233 strm_compress.avail_out;
1e5c08ba 1234 } else {
f481450f 1235 comp_size = 0;
355d67fc 1236 }
f481450f
MD
1237 ret = deflateEnd(&strm_compress);
1238 break;
1239 default:
1240 kprintf("Error: Unknown compression method.\n");
1241 kprintf("Comp_method = %d.\n", comp_algo);
1242 break;
355d67fc 1243 }
f481450f 1244 }
1e5c08ba 1245
f481450f
MD
1246 if (comp_size == 0) {
1247 /*
1248 * compression failed or turned off
1249 */
1250 comp_block_size = pblksize; /* safety */
1251 if (++ip->comp_heuristic > 128)
1252 ip->comp_heuristic = 8;
1253 } else {
1254 /*
1255 * compression succeeded
1256 */
1257 ip->comp_heuristic = 0;
1258 if (comp_size <= 1024) {
1259 comp_block_size = 1024;
1260 } else if (comp_size <= 2048) {
1261 comp_block_size = 2048;
1262 } else if (comp_size <= 4096) {
1263 comp_block_size = 4096;
1264 } else if (comp_size <= 8192) {
1265 comp_block_size = 8192;
1266 } else if (comp_size <= 16384) {
1267 comp_block_size = 16384;
1268 } else if (comp_size <= 32768) {
1269 comp_block_size = 32768;
355d67fc 1270 } else {
f481450f
MD
1271 panic("hammer2: WRITE PATH: "
1272 "Weird comp_size value.");
1273 /* NOT REACHED */
1274 comp_block_size = pblksize;
1275 }
1276 }
1277
278ab2b2
MD
1278 cluster = hammer2_assign_physical(trans, ip, cparent,
1279 lbase, comp_block_size,
1280 errorp);
84e47819 1281 ipdata = &hammer2_cluster_data(cparent)->ipdata;
f481450f
MD
1282
1283 if (*errorp) {
1284 kprintf("WRITE PATH: An error occurred while "
1285 "assigning physical space.\n");
278ab2b2
MD
1286 KKASSERT(cluster == NULL);
1287 goto done;
1288 }
1289
1290 for (i = 0; i < cluster->nchains; ++i) {
fdf62707
MD
1291 hammer2_io_t *dio;
1292 char *bdata;
f481450f 1293
278ab2b2 1294 chain = cluster->array[i];
f481450f
MD
1295 KKASSERT(chain->flags & HAMMER2_CHAIN_MODIFIED);
1296
1297 switch(chain->bref.type) {
1298 case HAMMER2_BREF_TYPE_INODE:
1299 KKASSERT(chain->data->ipdata.op_flags &
1300 HAMMER2_OPFLAG_DIRECTDATA);
1301 KKASSERT(bp->b_loffset == 0);
1302 bcopy(bp->b_data, chain->data->ipdata.u.data,
1303 HAMMER2_EMBEDDED_BYTES);
1304 break;
1305 case HAMMER2_BREF_TYPE_DATA:
1e5c08ba 1306 /*
f481450f
MD
1307 * Optimize out the read-before-write
1308 * if possible.
1e5c08ba 1309 */
fdf62707
MD
1310 *errorp = hammer2_io_newnz(chain->hmp,
1311 chain->bref.data_off,
1312 chain->bytes,
1313 &dio);
1314 if (*errorp) {
1315 hammer2_io_brelse(&dio);
1316 kprintf("hammer2: WRITE PATH: "
1317 "dbp bread error\n");
1318 break;
355d67fc 1319 }
fdf62707 1320 bdata = hammer2_io_data(dio, chain->bref.data_off);
355d67fc 1321
f481450f
MD
1322 /*
1323 * When loading the block make sure we don't
1324 * leave garbage after the compressed data.
1325 */
1326 if (comp_size) {
1327 chain->bref.methods =
1328 HAMMER2_ENC_COMP(comp_algo) +
e07becf8 1329 HAMMER2_ENC_CHECK(check_algo);
fdf62707 1330 bcopy(comp_buffer, bdata, comp_size);
f481450f 1331 if (comp_size != comp_block_size) {
fdf62707
MD
1332 bzero(bdata + comp_size,
1333 comp_block_size - comp_size);
355d67fc 1334 }
f481450f
MD
1335 } else {
1336 chain->bref.methods =
1337 HAMMER2_ENC_COMP(
1338 HAMMER2_COMP_NONE) +
e07becf8 1339 HAMMER2_ENC_CHECK(check_algo);
fdf62707 1340 bcopy(bp->b_data, bdata, pblksize);
f481450f 1341 }
355d67fc 1342
a71db85d
MD
1343 /*
1344 * The flush code doesn't calculate check codes for
1345 * file data (doing so can result in excessive I/O),
1346 * so we do it here.
1347 */
1348 hammer2_chain_setcheck(chain, bdata);
1349
f481450f 1350 /*
da6f36f4
MD
1351 * Device buffer is now valid, chain is no longer in
1352 * the initial state.
1353 *
1354 * (No blockref table worries with file data)
f481450f 1355 */
fdf62707 1356 atomic_clear_int(&chain->flags, HAMMER2_CHAIN_INITIAL);
355d67fc 1357
f481450f
MD
1358 /* Now write the related bdp. */
1359 if (ioflag & IO_SYNC) {
355d67fc 1360 /*
f481450f 1361 * Synchronous I/O requested.
355d67fc 1362 */
fdf62707 1363 hammer2_io_bwrite(&dio);
f481450f
MD
1364 /*
1365 } else if ((ioflag & IO_DIRECT) &&
1366 loff + n == pblksize) {
fdf62707 1367 hammer2_io_bdwrite(&dio);
f481450f
MD
1368 */
1369 } else if (ioflag & IO_ASYNC) {
fdf62707 1370 hammer2_io_bawrite(&dio);
f481450f 1371 } else {
fdf62707 1372 hammer2_io_bdwrite(&dio);
355d67fc 1373 }
f481450f
MD
1374 break;
1375 default:
1376 panic("hammer2_write_bp: bad chain type %d\n",
1377 chain->bref.type);
1378 /* NOT REACHED */
1379 break;
355d67fc 1380 }
355d67fc 1381 }
278ab2b2 1382done:
837bd39b
MD
1383 if (cluster)
1384 hammer2_cluster_unlock(cluster);
f481450f
MD
1385 if (comp_buffer)
1386 objcache_put(cache_buffer_write, comp_buffer);
355d67fc
MD
1387}
1388
1389/*
1390 * Function that performs zero-checking and writing without compression,
1391 * it corresponds to default zero-checking path.
1392 */
1393static
1394void
1e5c08ba 1395hammer2_zero_check_and_write(struct buf *bp, hammer2_trans_t *trans,
6a5f4fe6 1396 hammer2_inode_t *ip, const hammer2_inode_data_t *ipdata,
278ab2b2 1397 hammer2_cluster_t *cparent,
e07becf8
MD
1398 hammer2_key_t lbase, int ioflag, int pblksize, int *errorp,
1399 int check_algo)
355d67fc 1400{
278ab2b2 1401 hammer2_cluster_t *cluster;
355d67fc 1402
f481450f 1403 if (test_block_zeros(bp->b_data, pblksize)) {
278ab2b2 1404 zero_write(bp, trans, ip, ipdata, cparent, lbase, errorp);
f481450f 1405 } else {
278ab2b2
MD
1406 cluster = hammer2_assign_physical(trans, ip, cparent,
1407 lbase, pblksize, errorp);
e07becf8
MD
1408 hammer2_write_bp(cluster, bp, ioflag, pblksize, errorp,
1409 check_algo);
278ab2b2
MD
1410 if (cluster)
1411 hammer2_cluster_unlock(cluster);
355d67fc
MD
1412 }
1413}
1414
1415/*
1416 * A function to test whether a block of data contains only zeros,
f481450f 1417 * returns TRUE (non-zero) if the block is all zeros.
355d67fc
MD
1418 */
1419static
1420int
f481450f 1421test_block_zeros(const char *buf, size_t bytes)
355d67fc
MD
1422{
1423 size_t i;
1424
1425 for (i = 0; i < bytes; i += sizeof(long)) {
f481450f
MD
1426 if (*(const long *)(buf + i) != 0)
1427 return (0);
355d67fc 1428 }
f481450f 1429 return (1);
355d67fc
MD
1430}
1431
1432/*
1433 * Function to "write" a block that contains only zeros.
1434 */
1435static
1436void
6a5f4fe6 1437zero_write(struct buf *bp, hammer2_trans_t *trans,
da6f36f4
MD
1438 hammer2_inode_t *ip, const hammer2_inode_data_t *ipdata,
1439 hammer2_cluster_t *cparent,
1440 hammer2_key_t lbase, int *errorp __unused)
355d67fc 1441{
278ab2b2
MD
1442 hammer2_cluster_t *cluster;
1443 hammer2_media_data_t *data;
1897c66e 1444 hammer2_key_t key_dummy;
278ab2b2 1445 int ddflag;
355d67fc 1446
278ab2b2
MD
1447 cparent = hammer2_cluster_lookup_init(cparent, 0);
1448 cluster = hammer2_cluster_lookup(cparent, &key_dummy, lbase, lbase,
1449 HAMMER2_LOOKUP_NODATA, &ddflag);
1450 if (cluster) {
6a5f4fe6 1451 data = hammer2_cluster_wdata(cluster);
355d67fc 1452
278ab2b2 1453 if (ddflag) {
6a5f4fe6
MD
1454 KKASSERT(cluster->focus->flags &
1455 HAMMER2_CHAIN_MODIFIED);
278ab2b2 1456 bzero(data->ipdata.u.data, HAMMER2_EMBEDDED_BYTES);
6a5f4fe6 1457 hammer2_cluster_modsync(cluster);
355d67fc 1458 } else {
da6f36f4
MD
1459 hammer2_cluster_delete(trans, cparent, cluster,
1460 HAMMER2_DELETE_PERMANENT);
355d67fc 1461 }
278ab2b2 1462 hammer2_cluster_unlock(cluster);
355d67fc 1463 }
278ab2b2 1464 hammer2_cluster_lookup_done(cparent);
355d67fc
MD
1465}
1466
1467/*
1468 * Function to write the data as it is, without performing any sort of
1469 * compression. This function is used in path without compression and
1470 * default zero-checking path.
1471 */
1472static
1473void
278ab2b2 1474hammer2_write_bp(hammer2_cluster_t *cluster, struct buf *bp, int ioflag,
e07becf8 1475 int pblksize, int *errorp, int check_algo)
355d67fc 1476{
278ab2b2 1477 hammer2_chain_t *chain;
fdf62707
MD
1478 hammer2_io_t *dio;
1479 char *bdata;
355d67fc 1480 int error;
278ab2b2 1481 int i;
355d67fc 1482
278ab2b2 1483 error = 0; /* XXX TODO below */
355d67fc 1484
278ab2b2
MD
1485 for (i = 0; i < cluster->nchains; ++i) {
1486 chain = cluster->array[i];
1487
278ab2b2
MD
1488 KKASSERT(chain->flags & HAMMER2_CHAIN_MODIFIED);
1489
1490 switch(chain->bref.type) {
1491 case HAMMER2_BREF_TYPE_INODE:
1492 KKASSERT(chain->data->ipdata.op_flags &
1493 HAMMER2_OPFLAG_DIRECTDATA);
1494 KKASSERT(bp->b_loffset == 0);
1495 bcopy(bp->b_data, chain->data->ipdata.u.data,
1496 HAMMER2_EMBEDDED_BYTES);
1497 error = 0;
fdf62707 1498 break;
278ab2b2
MD
1499 case HAMMER2_BREF_TYPE_DATA:
1500 error = hammer2_io_newnz(chain->hmp,
1501 chain->bref.data_off,
1502 chain->bytes, &dio);
1503 if (error) {
1504 hammer2_io_bqrelse(&dio);
1505 kprintf("hammer2: WRITE PATH: "
1506 "dbp bread error\n");
1507 break;
1508 }
1509 bdata = hammer2_io_data(dio, chain->bref.data_off);
355d67fc 1510
278ab2b2
MD
1511 chain->bref.methods = HAMMER2_ENC_COMP(
1512 HAMMER2_COMP_NONE) +
e07becf8 1513 HAMMER2_ENC_CHECK(check_algo);
278ab2b2 1514 bcopy(bp->b_data, bdata, chain->bytes);
355d67fc 1515
a71db85d
MD
1516 /*
1517 * The flush code doesn't calculate check codes for
1518 * file data (doing so can result in excessive I/O),
1519 * so we do it here.
1520 */
1521 hammer2_chain_setcheck(chain, bdata);
1522
355d67fc 1523 /*
da6f36f4
MD
1524 * Device buffer is now valid, chain is no longer in
1525 * the initial state.
1526 *
1527 * (No blockref table worries with file data)
355d67fc 1528 */
278ab2b2
MD
1529 atomic_clear_int(&chain->flags, HAMMER2_CHAIN_INITIAL);
1530
1531 if (ioflag & IO_SYNC) {
1532 /*
1533 * Synchronous I/O requested.
1534 */
1535 hammer2_io_bwrite(&dio);
1536 /*
1537 } else if ((ioflag & IO_DIRECT) &&
1538 loff + n == pblksize) {
1539 hammer2_io_bdwrite(&dio);
1540 */
1541 } else if (ioflag & IO_ASYNC) {
1542 hammer2_io_bawrite(&dio);
1543 } else {
1544 hammer2_io_bdwrite(&dio);
1545 }
1546 break;
1547 default:
1548 panic("hammer2_write_bp: bad chain type %d\n",
1549 chain->bref.type);
1550 /* NOT REACHED */
1551 error = 0;
1552 break;
355d67fc 1553 }
278ab2b2 1554 KKASSERT(error == 0); /* XXX TODO */
355d67fc 1555 }
1e5c08ba 1556 *errorp = error;
355d67fc
MD
1557}
1558
e118c14f
MD
1559static
1560int
10136ab6
MD
1561hammer2_remount(hammer2_mount_t *hmp, struct mount *mp, char *path,
1562 struct vnode *devvp, struct ucred *cred)
703720e4 1563{
10136ab6
MD
1564 int error;
1565
1566 if (hmp->ronly && (mp->mnt_kern_flag & MNTK_WANTRDWR)) {
1567 error = hammer2_recovery(hmp);
1568 } else {
1569 error = 0;
1570 }
1571 return error;
703720e4
MD
1572}
1573
e118c14f
MD
1574static
1575int
066e00cc 1576hammer2_vfs_unmount(struct mount *mp, int mntflags)
703720e4 1577{
e4e20f48 1578 hammer2_pfsmount_t *pmp;
54eb943b 1579 hammer2_mount_t *hmp;
7bed8d7e 1580 hammer2_chain_t *rchain;
58e43599 1581 hammer2_cluster_t *cluster;
703720e4 1582 int flags;
50e4f8f4 1583 int error = 0;
7bed8d7e 1584 int i;
703720e4 1585
e4e20f48 1586 pmp = MPTOPMP(mp);
703720e4 1587
623d43d4
MD
1588 if (pmp == NULL)
1589 return(0);
1590
9b6b3df4 1591 lockmgr(&hammer2_mntlk, LK_EXCLUSIVE);
58e43599 1592 TAILQ_REMOVE(&hammer2_pfslist, pmp, mntentry);
9b6b3df4 1593
065f4046
MD
1594 /*
1595 * If mount initialization proceeded far enough we must flush
1596 * its vnodes.
1597 */
1598 if (mntflags & MNT_FORCE)
1599 flags = FORCECLOSE;
1600 else
7bed8d7e 1601 flags = 0;
065f4046
MD
1602 if (pmp->iroot) {
1603 error = vflush(mp, 0, flags);
1604 if (error)
1605 goto failed;
1606 }
50e4f8f4 1607
eae2ed61 1608 ccms_domain_uninit(&pmp->ccms_dom);
eae2ed61 1609
065f4046
MD
1610 if (pmp->wthread_td) {
1611 mtx_lock(&pmp->wthread_mtx);
1612 pmp->wthread_destroy = 1;
1613 wakeup(&pmp->wthread_bioq);
1614 while (pmp->wthread_destroy != -1) {
1615 mtxsleep(&pmp->wthread_destroy,
1616 &pmp->wthread_mtx, 0,
1617 "umount-sleep", 0);
1618 }
1619 mtx_unlock(&pmp->wthread_mtx);
1620 pmp->wthread_td = NULL;
1621 }
e4e20f48 1622
044541cd
MD
1623 /*
1624 * Cleanup our reference on ihidden.
1625 */
1626 if (pmp->ihidden) {
1627 hammer2_inode_drop(pmp->ihidden);
1628 pmp->ihidden = NULL;
1629 }
7bed8d7e 1630
044541cd
MD
1631 /*
1632 * Cleanup our reference on iroot. iroot is (should) not be needed
1633 * by the flush code.
1634 */
1635 if (pmp->iroot) {
58e43599
MD
1636 cluster = &pmp->iroot->cluster;
1637 for (i = 0; i < pmp->iroot->cluster.nchains; ++i) {
1638 rchain = pmp->iroot->cluster.array[i];
1639 if (rchain == NULL)
1640 continue;
1641 hmp = rchain->hmp;
1642 hammer2_vfs_unmount_hmp1(mp, hmp);
7bed8d7e 1643
7bed8d7e 1644 atomic_clear_int(&rchain->flags, HAMMER2_CHAIN_MOUNTED);
0dea3156 1645#if REPORT_REFS_ERRORS
7bed8d7e
MD
1646 if (rchain->refs != 1)
1647 kprintf("PMP->RCHAIN %p REFS WRONG %d\n",
1648 rchain, rchain->refs);
0dea3156 1649#else
7bed8d7e 1650 KKASSERT(rchain->refs == 1);
0dea3156 1651#endif
7bed8d7e 1652 hammer2_chain_drop(rchain);
58e43599
MD
1653 cluster->array[i] = NULL;
1654 hammer2_vfs_unmount_hmp2(mp, hmp);
e4e20f48 1655 }
58e43599 1656 cluster->focus = NULL;
222d9e22 1657
58e43599
MD
1658#if REPORT_REFS_ERRORS
1659 if (pmp->iroot->refs != 1)
1660 kprintf("PMP->IROOT %p REFS WRONG %d\n",
1661 pmp->iroot, pmp->iroot->refs);
1662#else
1663 KKASSERT(pmp->iroot->refs == 1);
1664#endif
1665 /* ref for pmp->iroot */
1666 hammer2_inode_drop(pmp->iroot);
1667 pmp->iroot = NULL;
54eb943b 1668 }
703720e4 1669
e4e20f48 1670 pmp->mp = NULL;
54eb943b 1671 mp->mnt_data = NULL;
703720e4 1672
26bf1a36 1673 kmalloc_destroy(&pmp->mmsg);
99da41ea 1674 kmalloc_destroy(&pmp->minode);
26bf1a36 1675
e4e20f48 1676 kfree(pmp, M_HAMMER2);
9b6b3df4
MD
1677 error = 0;
1678
1679failed:
e4e20f48 1680 lockmgr(&hammer2_mntlk, LK_RELEASE);
a5913bdf 1681
703720e4
MD
1682 return (error);
1683}
1684
623d43d4
MD
1685static
1686void
1687hammer2_vfs_unmount_hmp1(struct mount *mp, hammer2_mount_t *hmp)
1688{
1689 hammer2_mount_exlock(hmp);
1690 --hmp->pmp_count;
1691
1692 kprintf("hammer2_unmount hmp=%p pmpcnt=%d\n", hmp, hmp->pmp_count);
1693
063639ba
MD
1694 kdmsg_iocom_uninit(&hmp->iocom); /* XXX chain depend deadlck? */
1695
623d43d4 1696 /*
850687d2
MD
1697 * Cycle the volume data lock as a safety (probably not needed any
1698 * more). To ensure everything is out we need to flush at least
1699 * three times. (1) The running of the unlinkq can dirty the
1700 * filesystem, (2) A normal flush can dirty the freemap, and
1701 * (3) ensure that the freemap is fully synchronized.
623d43d4 1702 *
850687d2
MD
1703 * The next mount's recovery scan can clean everything up but we want
1704 * to leave the filesystem in a 100% clean state on a normal unmount.
623d43d4
MD
1705 */
1706 hammer2_voldata_lock(hmp);
850687d2 1707 hammer2_voldata_unlock(hmp);
18e8ab5f
MD
1708 if (mp->mnt_data) {
1709 hammer2_vfs_sync(mp, MNT_WAIT);
1710 hammer2_vfs_sync(mp, MNT_WAIT);
1711 hammer2_vfs_sync(mp, MNT_WAIT);
1712 }
850687d2 1713
623d43d4 1714 if (hmp->pmp_count == 0) {
da6f36f4
MD
1715 if ((hmp->vchain.flags | hmp->fchain.flags) &
1716 HAMMER2_CHAIN_FLUSH_MASK) {
623d43d4
MD
1717 kprintf("hammer2_unmount: chains left over "
1718 "after final sync\n");
da6f36f4
MD
1719 kprintf(" vchain %08x\n", hmp->vchain.flags);
1720 kprintf(" fchain %08x\n", hmp->fchain.flags);
0924b3f8 1721
623d43d4
MD
1722 if (hammer2_debug & 0x0010)
1723 Debugger("entered debugger");
1724 }
1725 }
1726}
1727
1728static
1729void
1730hammer2_vfs_unmount_hmp2(struct mount *mp, hammer2_mount_t *hmp)
1731{
50456506 1732 hammer2_pfsmount_t *spmp;
623d43d4
MD
1733 struct vnode *devvp;
1734 int dumpcnt;
1735 int ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
1736
1737 /*
1738 * If no PFS's left drop the master hammer2_mount for the
1739 * device.
1740 */
1741 if (hmp->pmp_count == 0) {
50456506
MD
1742 /*
1743 * Clean up SPMP and the super-root inode
1744 */
1745 spmp = hmp->spmp;
1746 if (spmp) {
1747 if (spmp->iroot) {
1748 hammer2_inode_drop(spmp->iroot);
1749 spmp->iroot = NULL;
1750 }
1751 hmp->spmp = NULL;
837bd39b
MD
1752 kmalloc_destroy(&spmp->mmsg);
1753 kmalloc_destroy(&spmp->minode);
50456506 1754 kfree(spmp, M_HAMMER2);
623d43d4
MD
1755 }
1756
1757 /*
1758 * Finish up with the device vnode
1759 */
1760 if ((devvp = hmp->devvp) != NULL) {
1761 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
1762 vinvalbuf(devvp, (ronly ? 0 : V_SAVE), 0, 0);
1763 hmp->devvp = NULL;
8138a154 1764 VOP_CLOSE(devvp, (ronly ? FREAD : FREAD|FWRITE), NULL);
623d43d4
MD
1765 vn_unlock(devvp);
1766 vrele(devvp);
1767 devvp = NULL;
1768 }
1769
50456506
MD
1770 /*
1771 * Clear vchain/fchain flags that might prevent final cleanup
1772 * of these chains.
1773 */
1774 if (hmp->vchain.flags & HAMMER2_CHAIN_MODIFIED) {
1775 atomic_clear_int(&hmp->vchain.flags,
1776 HAMMER2_CHAIN_MODIFIED);
da6f36f4 1777 hammer2_pfs_memory_wakeup(hmp->vchain.pmp);
50456506
MD
1778 hammer2_chain_drop(&hmp->vchain);
1779 }
da6f36f4 1780 if (hmp->vchain.flags & HAMMER2_CHAIN_UPDATE) {
50456506 1781 atomic_clear_int(&hmp->vchain.flags,
da6f36f4 1782 HAMMER2_CHAIN_UPDATE);
50456506
MD
1783 hammer2_chain_drop(&hmp->vchain);
1784 }
1785
1786 if (hmp->fchain.flags & HAMMER2_CHAIN_MODIFIED) {
1787 atomic_clear_int(&hmp->fchain.flags,
1788 HAMMER2_CHAIN_MODIFIED);
da6f36f4 1789 hammer2_pfs_memory_wakeup(hmp->fchain.pmp);
50456506
MD
1790 hammer2_chain_drop(&hmp->fchain);
1791 }
da6f36f4 1792 if (hmp->fchain.flags & HAMMER2_CHAIN_UPDATE) {
50456506 1793 atomic_clear_int(&hmp->fchain.flags,
da6f36f4 1794 HAMMER2_CHAIN_UPDATE);
50456506
MD
1795 hammer2_chain_drop(&hmp->fchain);
1796 }
1797
623d43d4
MD
1798 /*
1799 * Final drop of embedded freemap root chain to
1800 * clean up fchain.core (fchain structure is not
1801 * flagged ALLOCATED so it is cleaned out and then
1802 * left to rot).
1803 */
1804 hammer2_chain_drop(&hmp->fchain);
1805
1806 /*
1807 * Final drop of embedded volume root chain to clean
1808 * up vchain.core (vchain structure is not flagged
1809 * ALLOCATED so it is cleaned out and then left to
1810 * rot).
1811 */
1812 dumpcnt = 50;
8138a154 1813 hammer2_dump_chain(&hmp->vchain, 0, &dumpcnt, 'v');
623d43d4 1814 dumpcnt = 50;
8138a154 1815 hammer2_dump_chain(&hmp->fchain, 0, &dumpcnt, 'f');
623d43d4
MD
1816 hammer2_mount_unlock(hmp);
1817 hammer2_chain_drop(&hmp->vchain);
1818
1819 hammer2_io_cleanup(hmp, &hmp->iotree);
1820 if (hmp->iofree_count) {
1821 kprintf("io_cleanup: %d I/O's left hanging\n",
1822 hmp->iofree_count);
1823 }
1824
1825 TAILQ_REMOVE(&hammer2_mntlist, hmp, mntentry);
1826 kmalloc_destroy(&hmp->mchain);
1827 kfree(hmp, M_HAMMER2);
1828 } else {
1829 hammer2_mount_unlock(hmp);
1830 }
1831}
1832
e118c14f
MD
1833static
1834int
066e00cc 1835hammer2_vfs_vget(struct mount *mp, struct vnode *dvp,
703720e4
MD
1836 ino_t ino, struct vnode **vpp)
1837{
1838 kprintf("hammer2_vget\n");
1839 return (EOPNOTSUPP);
1840}
1841
e118c14f
MD
1842static
1843int
066e00cc 1844hammer2_vfs_root(struct mount *mp, struct vnode **vpp)
703720e4 1845{
e4e20f48 1846 hammer2_pfsmount_t *pmp;
278ab2b2 1847 hammer2_cluster_t *cparent;
703720e4
MD
1848 int error;
1849 struct vnode *vp;
1850
e4e20f48 1851 pmp = MPTOPMP(mp);
e4e20f48 1852 if (pmp->iroot == NULL) {
703720e4
MD
1853 *vpp = NULL;
1854 error = EINVAL;
1855 } else {
278ab2b2 1856 cparent = hammer2_inode_lock_sh(pmp->iroot);
84e47819 1857 vp = hammer2_igetv(pmp->iroot, cparent, &error);
278ab2b2 1858 hammer2_inode_unlock_sh(pmp->iroot, cparent);
703720e4
MD
1859 *vpp = vp;
1860 if (vp == NULL)
1861 kprintf("vnodefail\n");
1862 }
703720e4
MD
1863
1864 return (error);
1865}
1866
28ee5f14
MD
1867/*
1868 * Filesystem status
1869 *
476d2aad 1870 * XXX incorporate ipdata->inode_quota and data_quota
28ee5f14 1871 */
e118c14f
MD
1872static
1873int
066e00cc 1874hammer2_vfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
703720e4 1875{
28ee5f14 1876 hammer2_pfsmount_t *pmp;
54eb943b 1877 hammer2_mount_t *hmp;
703720e4 1878
28ee5f14 1879 pmp = MPTOPMP(mp);
58e43599
MD
1880 KKASSERT(pmp->iroot->cluster.nchains >= 1);
1881 hmp = pmp->iroot->cluster.focus->hmp; /* XXX */
703720e4 1882
476d2aad 1883 mp->mnt_stat.f_files = pmp->inode_count;
28ee5f14
MD
1884 mp->mnt_stat.f_ffree = 0;
1885 mp->mnt_stat.f_blocks = hmp->voldata.allocator_size / HAMMER2_PBUFSIZE;
9f604b01 1886 mp->mnt_stat.f_bfree = hmp->voldata.allocator_free / HAMMER2_PBUFSIZE;
50e4f8f4 1887 mp->mnt_stat.f_bavail = mp->mnt_stat.f_bfree;
703720e4 1888
50e4f8f4 1889 *sbp = mp->mnt_stat;
703720e4
MD
1890 return (0);
1891}
1892
e118c14f
MD
1893static
1894int
066e00cc 1895hammer2_vfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred)
703720e4 1896{
28ee5f14 1897 hammer2_pfsmount_t *pmp;
50e4f8f4
MD
1898 hammer2_mount_t *hmp;
1899
28ee5f14 1900 pmp = MPTOPMP(mp);
58e43599
MD
1901 KKASSERT(pmp->iroot->cluster.nchains >= 1);
1902 hmp = pmp->iroot->cluster.focus->hmp; /* XXX */
50e4f8f4 1903
88a032af 1904 mp->mnt_vstat.f_bsize = HAMMER2_PBUFSIZE;
476d2aad 1905 mp->mnt_vstat.f_files = pmp->inode_count;
28ee5f14
MD
1906 mp->mnt_vstat.f_ffree = 0;
1907 mp->mnt_vstat.f_blocks = hmp->voldata.allocator_size / HAMMER2_PBUFSIZE;
51f0b4b7 1908 mp->mnt_vstat.f_bfree = hmp->voldata.allocator_free / HAMMER2_PBUFSIZE;
28ee5f14 1909 mp->mnt_vstat.f_bavail = mp->mnt_vstat.f_bfree;
50e4f8f4
MD
1910
1911 *sbp = mp->mnt_vstat;
1912 return (0);
703720e4
MD
1913}
1914
10136ab6
MD
1915/*
1916 * Mount-time recovery (RW mounts)
1917 *
1918 * Updates to the free block table are allowed to lag flushes by one
1919 * transaction. In case of a crash, then on a fresh mount we must do an
50456506
MD
1920 * incremental scan of the last committed transaction id and make sure that
1921 * all related blocks have been marked allocated.
10136ab6 1922 *
50456506
MD
1923 * The super-root topology and each PFS has its own transaction id domain,
1924 * so we must track PFS boundary transitions.
10136ab6
MD
1925 */
1926struct hammer2_recovery_elm {
1927 TAILQ_ENTRY(hammer2_recovery_elm) entry;
1928 hammer2_chain_t *chain;
50456506 1929 hammer2_tid_t sync_tid;
10136ab6
MD
1930};
1931
1932TAILQ_HEAD(hammer2_recovery_list, hammer2_recovery_elm);
1933
50456506
MD
1934struct hammer2_recovery_info {
1935 struct hammer2_recovery_list list;
1936 int depth;
1937};
1938
10136ab6
MD
1939static int hammer2_recovery_scan(hammer2_trans_t *trans, hammer2_mount_t *hmp,
1940 hammer2_chain_t *parent,
50456506
MD
1941 struct hammer2_recovery_info *info,
1942 hammer2_tid_t sync_tid);
10136ab6
MD
1943
1944#define HAMMER2_RECOVERY_MAXDEPTH 10
1945
1946static
1947int
1948hammer2_recovery(hammer2_mount_t *hmp)
1949{
1950 hammer2_trans_t trans;
50456506 1951 struct hammer2_recovery_info info;
10136ab6
MD
1952 struct hammer2_recovery_elm *elm;
1953 hammer2_chain_t *parent;
50456506 1954 hammer2_tid_t sync_tid;
10136ab6
MD
1955 int error;
1956 int cumulative_error = 0;
1957
50456506 1958 hammer2_trans_init(&trans, hmp->spmp, 0);
10136ab6 1959
50456506
MD
1960 sync_tid = 0;
1961 TAILQ_INIT(&info.list);
1962 info.depth = 0;
10136ab6 1963 parent = hammer2_chain_lookup_init(&hmp->vchain, 0);
50456506
MD
1964 cumulative_error = hammer2_recovery_scan(&trans, hmp, parent,
1965 &info, sync_tid);
10136ab6
MD
1966 hammer2_chain_lookup_done(parent);
1967
50456506
MD
1968 while ((elm = TAILQ_FIRST(&info.list)) != NULL) {
1969 TAILQ_REMOVE(&info.list, elm, entry);
10136ab6 1970 parent = elm->chain;
50456506 1971 sync_tid = elm->sync_tid;
10136ab6
MD
1972 kfree(elm, M_HAMMER2);
1973
1974 hammer2_chain_lock(parent, HAMMER2_RESOLVE_ALWAYS |
1975 HAMMER2_RESOLVE_NOREF);
50456506
MD
1976 error = hammer2_recovery_scan(&trans, hmp, parent,
1977 &info, sync_tid);
10136ab6
MD
1978 hammer2_chain_unlock(parent);
1979 if (error)
1980 cumulative_error = error;
1981 }
1982 hammer2_trans_done(&trans);
1983
1984 return cumulative_error;
1985}
1986
1987static
1988int
1989hammer2_recovery_scan(hammer2_trans_t *trans, hammer2_mount_t *hmp,
1990 hammer2_chain_t *parent,
50456506
MD
1991 struct hammer2_recovery_info *info,
1992 hammer2_tid_t sync_tid)
10136ab6
MD
1993{
1994 hammer2_chain_t *chain;
1995 int cache_index;
1996 int cumulative_error = 0;
50456506 1997 int pfs_boundary = 0;
10136ab6
MD
1998 int error;
1999
10136ab6
MD
2000 /*
2001 * Adjust freemap to ensure that the block(s) are marked allocated.
2002 */
2003 if (parent->bref.type != HAMMER2_BREF_TYPE_VOLUME) {
2004 hammer2_freemap_adjust(trans, hmp, &parent->bref,
2005 HAMMER2_FREEMAP_DORECOVER);
2006 }
2007
2008 /*
2009 * Check type for recursive scan
2010 */
2011 switch(parent->bref.type) {
2012 case HAMMER2_BREF_TYPE_VOLUME:
2013 /* data already instantiated */
2014 break;
2015 case HAMMER2_BREF_TYPE_INODE:
2016 /*
2017 * Must instantiate data for DIRECTDATA test and also
2018 * for recursion.
2019 */
2020 hammer2_chain_lock(parent, HAMMER2_RESOLVE_ALWAYS);
10136ab6
MD
2021 if (parent->data->ipdata.op_flags & HAMMER2_OPFLAG_DIRECTDATA) {
2022 /* not applicable to recovery scan */
50456506 2023 hammer2_chain_unlock(parent);
10136ab6
MD
2024 return 0;
2025 }
50456506
MD
2026 if ((parent->data->ipdata.op_flags & HAMMER2_OPFLAG_PFSROOT) &&
2027 info->depth != 0) {
2028 pfs_boundary = 1;
2029 sync_tid = parent->bref.mirror_tid - 1;
2030 }
2031 hammer2_chain_unlock(parent);
10136ab6
MD
2032 break;
2033 case HAMMER2_BREF_TYPE_INDIRECT:
2034 /*
2035 * Must instantiate data for recursion
2036 */
2037 hammer2_chain_lock(parent, HAMMER2_RESOLVE_ALWAYS);
2038 hammer2_chain_unlock(parent);
2039 break;
2040 case HAMMER2_BREF_TYPE_DATA:
2041 case HAMMER2_BREF_TYPE_FREEMAP:
2042 case HAMMER2_BREF_TYPE_FREEMAP_NODE:
2043 case HAMMER2_BREF_TYPE_FREEMAP_LEAF:
2044 /* not applicable to recovery scan */
2045 return 0;
2046 break;
2047 default:
2048 return EDOM;
2049 }
2050
50456506
MD
2051 /*
2052 * Defer operation if depth limit reached or if we are crossing a
2053 * PFS boundary.
2054 */
2055 if (info->depth >= HAMMER2_RECOVERY_MAXDEPTH || pfs_boundary) {
2056 struct hammer2_recovery_elm *elm;
2057
2058 elm = kmalloc(sizeof(*elm), M_HAMMER2, M_ZERO | M_WAITOK);
2059 elm->chain = parent;
2060 elm->sync_tid = sync_tid;
2061 hammer2_chain_ref(parent);
2062 TAILQ_INSERT_TAIL(&info->list, elm, entry);
2063 /* unlocked by caller */
2064
2065 return(0);
2066 }
2067
2068
10136ab6
MD
2069 /*
2070 * Recursive scan of the last flushed transaction only. We are
2071 * doing this without pmp assignments so don't leave the chains
2072 * hanging around after we are done with them.
2073 */
2074 cache_index = 0;
2075 chain = hammer2_chain_scan(parent, NULL, &cache_index,
2076 HAMMER2_LOOKUP_NODATA);
2077 while (chain) {
2078 atomic_set_int(&chain->flags, HAMMER2_CHAIN_RELEASE);
50456506
MD
2079 if (chain->bref.mirror_tid >= sync_tid) {
2080 ++info->depth;
10136ab6 2081 error = hammer2_recovery_scan(trans, hmp, chain,
50456506
MD
2082 info, sync_tid);
2083 --info->depth;
10136ab6
MD
2084 if (error)
2085 cumulative_error = error;
2086 }
2087 chain = hammer2_chain_scan(parent, chain, &cache_index,
2088 HAMMER2_LOOKUP_NODATA);
2089 }
2090
2091 return cumulative_error;
2092}
2093
703720e4
MD
2094/*
2095 * Sync the entire filesystem; this is called from the filesystem syncer
2096 * process periodically and whenever a user calls sync(1) on the hammer
2097 * mountpoint.
2098 *
2099 * Currently is actually called from the syncer! \o/
2100 *
2101 * This task will have to snapshot the state of the dirty inode chain.
2102 * From that, it will have to make sure all of the inodes on the dirty
2103 * chain have IO initiated. We make sure that io is initiated for the root
2104 * block.
2105 *
2106 * If waitfor is set, we wait for media to acknowledge the new rootblock.
2107 *
2108 * THINKS: side A vs side B, to have sync not stall all I/O?
2109 */
e118c14f 2110int
066e00cc 2111hammer2_vfs_sync(struct mount *mp, int waitfor)
703720e4 2112{
b7926f31 2113 struct hammer2_sync_info info;
50456506 2114 hammer2_inode_t *iroot;
a7720be7 2115 hammer2_chain_t *chain;
50456506 2116 hammer2_chain_t *parent;
a5913bdf 2117 hammer2_pfsmount_t *pmp;
54eb943b 2118 hammer2_mount_t *hmp;
b7926f31
MD
2119 int flags;
2120 int error;
7bed8d7e 2121 int total_error;
a7720be7 2122 int force_fchain;
3fc4c63d 2123 int i;
03188ed8 2124 int j;
703720e4 2125
a5913bdf 2126 pmp = MPTOPMP(mp);
50456506
MD
2127 iroot = pmp->iroot;
2128 KKASSERT(iroot);
2129 KKASSERT(iroot->pmp == pmp);
703720e4 2130
355d67fc
MD
2131 /*
2132 * We can't acquire locks on existing vnodes while in a transaction
2133 * without risking a deadlock. This assumes that vfsync() can be
2134 * called without the vnode locked (which it can in DragonFly).
2135 * Otherwise we'd have to implement a multi-pass or flag the lock
2136 * failures and retry.
c057466c
MD
2137 *
2138 * The reclamation code interlocks with the sync list's token
2139 * (by removing the vnode from the scan list) before unlocking
2140 * the inode, giving us time to ref the inode.
355d67fc
MD
2141 */
2142 /*flags = VMSC_GETVP;*/
2143 flags = 0;
b7926f31
MD
2144 if (waitfor & MNT_LAZY)
2145 flags |= VMSC_ONEPASS;
2146
d7bfb2cb 2147 /*
052e0aa0
MD
2148 * Start our flush transaction. This does not return until all
2149 * concurrent transactions have completed and will prevent any
2150 * new transactions from running concurrently, except for the
2151 * buffer cache transactions.
2152 *
2153 * For efficiency do an async pass before making sure with a
2154 * synchronous pass on all related buffer cache buffers. It
2155 * should theoretically not be possible for any new file buffers
2156 * to be instantiated during this sequence.
d7bfb2cb 2157 */
50456506
MD
2158 hammer2_trans_init(&info.trans, pmp, HAMMER2_TRANS_ISFLUSH |
2159 HAMMER2_TRANS_PREFLUSH);
eae2ed61 2160 hammer2_run_unlinkq(&info.trans, pmp);
50456506 2161
b7926f31
MD
2162 info.error = 0;
2163 info.waitfor = MNT_NOWAIT;
eddc656a 2164 vsyncscan(mp, flags | VMSC_NOWAIT, hammer2_sync_scan2, &info);
052e0aa0
MD
2165 info.waitfor = MNT_WAIT;
2166 vsyncscan(mp, flags, hammer2_sync_scan2, &info);
a7720be7 2167
052e0aa0
MD
2168 /*
2169 * Clear PREFLUSH. This prevents (or asserts on) any new logical
2170 * buffer cache flushes which occur during the flush. Device buffers
2171 * are not affected.
2172 */
2173
2174#if 0
b7926f31
MD
2175 if (info.error == 0 && (waitfor & MNT_WAIT)) {
2176 info.waitfor = waitfor;
eddc656a 2177 vsyncscan(mp, flags, hammer2_sync_scan2, &info);
b7926f31
MD
2178
2179 }
052e0aa0 2180#endif
a4dc31e0 2181 hammer2_bioq_sync(info.trans.pmp);
052e0aa0 2182 atomic_clear_int(&info.trans.flags, HAMMER2_TRANS_PREFLUSH);
a7720be7 2183
50456506
MD
2184 total_error = 0;
2185
2186 /*
2187 * Flush all storage elements making up the cluster
2188 *
2189 * We must also flush any deleted siblings because the super-root
2190 * flush won't do it for us. They all must be staged or the
2191 * super-root flush will not be able to update its block table
2192 * properly.
2193 *
2194 * XXX currently done serially instead of concurrently
2195 */
2196 for (i = 0; iroot && i < iroot->cluster.nchains; ++i) {
2197 chain = iroot->cluster.array[i];
2198 if (chain) {
2199 hammer2_chain_lock(chain, HAMMER2_RESOLVE_ALWAYS);
da6f36f4 2200 hammer2_flush(&info.trans, chain);
50456506
MD
2201 hammer2_chain_unlock(chain);
2202 }
50456506 2203 }
052e0aa0 2204#if 0
50456506
MD
2205 hammer2_trans_done(&info.trans);
2206#endif
2207
a7720be7 2208 /*
50456506
MD
2209 * Flush all volume roots to synchronize PFS flushes with the
2210 * storage media. Use a super-root transaction for each one.
2211 *
2212 * The flush code will detect super-root -> pfs-root chain
2213 * transitions using the last pfs-root flush.
a7720be7 2214 */
50456506
MD
2215 for (i = 0; iroot && i < iroot->cluster.nchains; ++i) {
2216 chain = iroot->cluster.array[i];
2217 if (chain == NULL)
2218 continue;
2219
2220 hmp = chain->hmp;
03188ed8
MD
2221
2222 /*
2223 * We only have to flush each hmp once
2224 */
2225 for (j = i - 1; j >= 0; --j) {
2226 if (iroot->cluster.array[j] &&
2227 iroot->cluster.array[j]->hmp == hmp)
2228 break;
2229 }
2230 if (j >= 0)
2231 continue;
2232 hammer2_trans_spmp(&info.trans, hmp->spmp);
a5913bdf 2233
50456506
MD
2234 /*
2235 * Force an update of the XID from the PFS root to the
2236 * topology root. We couldn't do this from the PFS
2237 * transaction because a SPMP transaction is needed.
2238 * This does not modify blocks, instead what it does is
2239 * allow the flush code to find the transition point and
2240 * then update on the way back up.
2241 */
da6f36f4 2242 parent = chain->parent;
50456506 2243 KKASSERT(chain->pmp != parent->pmp);
da6f36f4 2244 hammer2_chain_setflush(&info.trans, parent);
a5913bdf 2245
d7bfb2cb
MD
2246 /*
2247 * Media mounts have two 'roots', vchain for the topology
2248 * and fchain for the free block table. Flush both.
2249 *
2250 * Note that the topology and free block table are handled
2251 * independently, so the free block table can wind up being
2252 * ahead of the topology. We depend on the bulk free scan
2253 * code to deal with any loose ends.
2254 */
50456506 2255 hammer2_chain_lock(&hmp->vchain, HAMMER2_RESOLVE_ALWAYS);
10136ab6 2256 hammer2_chain_lock(&hmp->fchain, HAMMER2_RESOLVE_ALWAYS);
da6f36f4 2257 if (hmp->fchain.flags & HAMMER2_CHAIN_FLUSH_MASK) {
50456506
MD
2258 /*
2259 * This will also modify vchain as a side effect,
2260 * mark vchain as modified now.
2261 */
2262 hammer2_voldata_modify(hmp);
10136ab6 2263 chain = &hmp->fchain;
da6f36f4 2264 hammer2_flush(&info.trans, chain);
10136ab6
MD
2265 KKASSERT(chain == &hmp->fchain);
2266 }
2267 hammer2_chain_unlock(&hmp->fchain);
50456506 2268 hammer2_chain_unlock(&hmp->vchain);
10136ab6 2269
7bed8d7e 2270 hammer2_chain_lock(&hmp->vchain, HAMMER2_RESOLVE_ALWAYS);
da6f36f4 2271 if (hmp->vchain.flags & HAMMER2_CHAIN_FLUSH_MASK) {
a7720be7 2272 chain = &hmp->vchain;
da6f36f4 2273 hammer2_flush(&info.trans, chain);
a7720be7
MD
2274 KKASSERT(chain == &hmp->vchain);
2275 force_fchain = 1;
2276 } else {
2277 force_fchain = 0;
7bed8d7e
MD
2278 }
2279 hammer2_chain_unlock(&hmp->vchain);
1a7cfe5a 2280
10136ab6 2281#if 0
7bed8d7e 2282 hammer2_chain_lock(&hmp->fchain, HAMMER2_RESOLVE_ALWAYS);
da6f36f4 2283 if ((hmp->fchain.flags & HAMMER2_CHAIN_FLUSH_MASK) ||
a7720be7 2284 force_fchain) {
d7bfb2cb 2285 /* this will also modify vchain as a side effect */
a7720be7 2286 chain = &hmp->fchain;
da6f36f4 2287 hammer2_flush(&info.trans, chain);
a7720be7 2288 KKASSERT(chain == &hmp->fchain);
7bed8d7e
MD
2289 }
2290 hammer2_chain_unlock(&hmp->fchain);
10136ab6 2291#endif
1a7cfe5a 2292
7bed8d7e 2293 error = 0;
b7926f31 2294
2910a90c 2295 /*
7bed8d7e
MD
2296 * We can't safely flush the volume header until we have
2297 * flushed any device buffers which have built up.
2298 *
2299 * XXX this isn't being incremental
2910a90c 2300 */
7bed8d7e
MD
2301 vn_lock(hmp->devvp, LK_EXCLUSIVE | LK_RETRY);
2302 error = VOP_FSYNC(hmp->devvp, MNT_WAIT, 0);
2303 vn_unlock(hmp->devvp);
b7926f31 2304
2910a90c 2305 /*
7bed8d7e
MD
2306 * The flush code sets CHAIN_VOLUMESYNC to indicate that the
2307 * volume header needs synchronization via hmp->volsync.
2308 *
2309 * XXX synchronize the flag & data with only this flush XXX
2910a90c 2310 */
7bed8d7e
MD
2311 if (error == 0 &&
2312 (hmp->vchain.flags & HAMMER2_CHAIN_VOLUMESYNC)) {
2313 struct buf *bp;
2314
2315 /*
2316 * Synchronize the disk before flushing the volume
2317 * header.
2318 */
2319 bp = getpbuf(NULL);
2320 bp->b_bio1.bio_offset = 0;
2321 bp->b_bufsize = 0;
2322 bp->b_bcount = 0;
2323 bp->b_cmd = BUF_CMD_FLUSH;
2324 bp->b_bio1.bio_done = biodone_sync;
2325 bp->b_bio1.bio_flags |= BIO_SYNC;
2326 vn_strategy(hmp->devvp, &bp->b_bio1);
2327 biowait(&bp->b_bio1, "h2vol");
2328 relpbuf(bp, NULL);
2329
2330 /*
2331 * Then we can safely flush the version of the
2332 * volume header synchronized by the flush code.
2333 */
2334 i = hmp->volhdrno + 1;
2335 if (i >= HAMMER2_NUM_VOLHDRS)
2336 i = 0;
2337 if (i * HAMMER2_ZONE_BYTES64 + HAMMER2_SEGSIZE >
2338 hmp->volsync.volu_size) {
2339 i = 0;
2340 }
2341 kprintf("sync volhdr %d %jd\n",
2342 i, (intmax_t)hmp->volsync.volu_size);
2343 bp = getblk(hmp->devvp, i * HAMMER2_ZONE_BYTES64,
2344 HAMMER2_PBUFSIZE, 0, 0);
2345 atomic_clear_int(&hmp->vchain.flags,
2346 HAMMER2_CHAIN_VOLUMESYNC);
2347 bcopy(&hmp->volsync, bp->b_data, HAMMER2_PBUFSIZE);
2348 bawrite(bp);
2349 hmp->volhdrno = i;
3fc4c63d 2350 }
7bed8d7e
MD
2351 if (error)
2352 total_error = error;
50456506
MD
2353
2354#if 0
2355 hammer2_trans_done(&info.trans);
2356#endif
b7926f31 2357 }
d001f460 2358 hammer2_trans_done(&info.trans);
a4dc31e0 2359
7bed8d7e 2360 return (total_error);
b7926f31 2361}
703720e4 2362
214f4a77
MD
2363/*
2364 * Sync passes.
214f4a77 2365 */
b7926f31
MD
2366static int
2367hammer2_sync_scan2(struct mount *mp, struct vnode *vp, void *data)
2368{
2369 struct hammer2_sync_info *info = data;
2370 hammer2_inode_t *ip;
2371 int error;
2372
c057466c
MD
2373 /*
2374 *
2375 */
b7926f31 2376 ip = VTOI(vp);
eddc656a
MD
2377 if (ip == NULL)
2378 return(0);
2379 if (vp->v_type == VNON || vp->v_type == VBAD) {
2380 vclrisdirty(vp);
2381 return(0);
2382 }
2383 if ((ip->flags & HAMMER2_INODE_MODIFIED) == 0 &&
2384 RB_EMPTY(&vp->v_rbdirty_tree)) {
2385 vclrisdirty(vp);
b7926f31
MD
2386 return(0);
2387 }
d001f460
MD
2388
2389 /*
2390 * VOP_FSYNC will start a new transaction so replicate some code
2391 * here to do it inline (see hammer2_vop_fsync()).
355d67fc
MD
2392 *
2393 * WARNING: The vfsync interacts with the buffer cache and might
2394 * block, we can't hold the inode lock at that time.
c057466c
MD
2395 * However, we MUST ref ip before blocking to ensure that
2396 * it isn't ripped out from under us (since we do not
2397 * hold a lock on the vnode).
d001f460 2398 */
c057466c 2399 hammer2_inode_ref(ip);
9596b8c4 2400 atomic_clear_int(&ip->flags, HAMMER2_INODE_MODIFIED);
c057466c
MD
2401 if (vp)
2402 vfsync(vp, MNT_NOWAIT, 1, NULL, NULL);
a7720be7 2403
c057466c 2404 hammer2_inode_drop(ip);
052e0aa0 2405#if 1
d001f460 2406 error = 0;
b7926f31
MD
2407 if (error)
2408 info->error = error;
052e0aa0 2409#endif
b7926f31 2410 return(0);
703720e4
MD
2411}
2412
e118c14f
MD
2413static
2414int
066e00cc 2415hammer2_vfs_vptofh(struct vnode *vp, struct fid *fhp)
703720e4
MD
2416{
2417 return (0);
2418}
2419
e118c14f
MD
2420static
2421int
066e00cc 2422hammer2_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
703720e4
MD
2423 struct fid *fhp, struct vnode **vpp)
2424{
2425 return (0);
2426}
2427
e118c14f
MD
2428static
2429int
066e00cc 2430hammer2_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
703720e4
MD
2431 int *exflagsp, struct ucred **credanonp)
2432{
2433 return (0);
2434}
50e4f8f4
MD
2435
2436/*
2437 * Support code for hammer2_mount(). Read, verify, and install the volume
2438 * header into the HMP
2439 *
2440 * XXX read four volhdrs and use the one with the highest TID whos CRC
2441 * matches.
2442 *
2443 * XXX check iCRCs.
57381c9e
VS
2444 *
2445 * XXX For filesystems w/ less than 4 volhdrs, make sure to not write to
2446 * nonexistant locations.
2447 *
2448 * XXX Record selected volhdr and ring updates to each of 4 volhdrs
50e4f8f4
MD
2449 */
2450static
2451int
2452hammer2_install_volume_header(hammer2_mount_t *hmp)
2453{
2454 hammer2_volume_data_t *vd;
99924359 2455 struct buf *bp;
60fbd5f4 2456 hammer2_crc32_t crc0, crc, bcrc0, bcrc;
99924359
MD
2457 int error_reported;
2458 int error;
0b3147ba
VS
2459 int valid;
2460 int i;
50e4f8f4 2461
99924359 2462 error_reported = 0;
0b3147ba
VS
2463 error = 0;
2464 valid = 0;
99924359 2465 bp = NULL;
0b3147ba 2466
99924359
MD
2467 /*
2468 * There are up to 4 copies of the volume header (syncs iterate
2469 * between them so there is no single master). We don't trust the
2470 * volu_size field so we don't know precisely how large the filesystem
2471 * is, so depend on the OS to return an error if we go beyond the
2472 * block device's EOF.
2473 */
0b3147ba 2474 for (i = 0; i < HAMMER2_NUM_VOLHDRS; i++) {
62efe6ec 2475 error = bread(hmp->devvp, i * HAMMER2_ZONE_BYTES64,
99924359 2476 HAMMER2_VOLUME_BYTES, &bp);
0b3147ba 2477 if (error) {
99924359
MD
2478 brelse(bp);
2479 bp = NULL;
0b3147ba
VS
2480 continue;
2481 }
2482
88a032af
VS
2483 vd = (struct hammer2_volume_data *) bp->b_data;
2484 if ((vd->magic != HAMMER2_VOLUME_ID_HBO) &&
2485 (vd->magic != HAMMER2_VOLUME_ID_ABO)) {
2486 brelse(bp);
2487 bp = NULL;
0b3147ba 2488 continue;
88a032af
VS
2489 }
2490
2491 if (vd->magic == HAMMER2_VOLUME_ID_ABO) {
2492 /* XXX: Reversed-endianness filesystem */
2493 kprintf("hammer2: reverse-endian filesystem detected");
2494 brelse(bp);
2495 bp = NULL;
2496 continue;
2497 }
0b3147ba
VS
2498
2499 crc = vd->icrc_sects[HAMMER2_VOL_ICRC_SECT0];
60fbd5f4 2500 crc0 = hammer2_icrc32(bp->b_data + HAMMER2_VOLUME_ICRC0_OFF,
99924359 2501 HAMMER2_VOLUME_ICRC0_SIZE);
60fbd5f4
VS
2502 bcrc = vd->icrc_sects[HAMMER2_VOL_ICRC_SECT1];
2503 bcrc0 = hammer2_icrc32(bp->b_data + HAMMER2_VOLUME_ICRC1_OFF,
2504 HAMMER2_VOLUME_ICRC1_SIZE);
2505 if ((crc0 != crc) || (bcrc0 != bcrc)) {
99924359 2506 kprintf("hammer2 volume header crc "
3fc4c63d 2507 "mismatch copy #%d %08x/%08x\n",
60fbd5f4 2508 i, crc0, crc);
99924359
MD
2509 error_reported = 1;
2510 brelse(bp);
2511 bp = NULL;
0b3147ba
VS
2512 continue;
2513 }
4d5318eb 2514 if (valid == 0 || hmp->voldata.mirror_tid < vd->mirror_tid) {
99924359
MD
2515 valid = 1;
2516 hmp->voldata = *vd;
3fc4c63d 2517 hmp->volhdrno = i;
0b3147ba 2518 }
99924359
MD
2519 brelse(bp);
2520 bp = NULL;
50e4f8f4 2521 }
0b3147ba 2522 if (valid) {
3fc4c63d 2523 hmp->volsync = hmp->voldata;
57381c9e 2524 error = 0;
3fc4c63d
MD
2525 if (error_reported || bootverbose || 1) { /* 1/DEBUG */
2526 kprintf("hammer2: using volume header #%d\n",
2527 hmp->volhdrno);
2528 }
0b3147ba
VS
2529 } else {
2530 error = EINVAL;
99924359 2531 kprintf("hammer2: no valid volume headers found!\n");
0b3147ba 2532 }
0b3147ba 2533 return (error);
50e4f8f4 2534}
0b3147ba 2535
1a34728c
MD
2536/*
2537 * Reconnect using the passed file pointer. The caller must ref the
2538 * fp for us.
2539 */
2540void
063639ba 2541hammer2_cluster_reconnect(hammer2_mount_t *hmp, struct file *fp)
1a34728c 2542{
3a5aa68f 2543 size_t name_len;
063639ba 2544 const char *name = "disk-volume";
a5913bdf 2545
70c3c3b7 2546 /*
3a5aa68f
MD
2547 * Closes old comm descriptor, kills threads, cleans up
2548 * states, then installs the new descriptor and creates
2549 * new threads.
1a34728c 2550 */
063639ba 2551 kdmsg_iocom_reconnect(&hmp->iocom, fp, "hammer2");
bfc3a7b1 2552
9b8b748f 2553 /*
063639ba
MD
2554 * Setup LNK_CONN fields for autoinitiated state machine. We
2555 * will use SPANs to advertise multiple PFSs so only pass the
2556 * fsid and HAMMER2_PFSTYPE_SUPROOT for the AUTOCONN.
2557 *
2558 * We are not initiating a LNK_SPAN so we do not have to set-up
2559 * iocom.auto_lnk_span.
537d97bc 2560 */
063639ba
MD
2561 bzero(&hmp->iocom.auto_lnk_conn.pfs_clid,
2562 sizeof(hmp->iocom.auto_lnk_conn.pfs_clid));
2563 hmp->iocom.auto_lnk_conn.pfs_fsid = hmp->voldata.fsid;
2564 hmp->iocom.auto_lnk_conn.pfs_type = HAMMER2_PFSTYPE_SUPROOT;
2565 hmp->iocom.auto_lnk_conn.proto_version = DMSG_SPAN_PROTO_1;
2566#if 0
2567 hmp->iocom.auto_lnk_conn.peer_type = hmp->voldata.peer_type;
2568#endif
2569 hmp->iocom.auto_lnk_conn.peer_type = DMSG_PEER_HAMMER2;
537d97bc
MD
2570
2571 /*
2572 * Filter adjustment. Clients do not need visibility into other
2573 * clients (otherwise millions of clients would present a serious
2574 * problem). The fs_label also serves to restrict the namespace.
9b8b748f 2575 */
063639ba
MD
2576 hmp->iocom.auto_lnk_conn.peer_mask = 1LLU << DMSG_PEER_HAMMER2;
2577 hmp->iocom.auto_lnk_conn.pfs_mask = (uint64_t)-1;
2578
2579#if 0
476d2aad 2580 switch (ipdata->pfs_type) {
537d97bc 2581 case DMSG_PFSTYPE_CLIENT:
063639ba 2582 hmp->iocom.auto_lnk_conn.peer_mask &=
537d97bc
MD
2583 ~(1LLU << DMSG_PFSTYPE_CLIENT);
2584 break;
2585 default:
2586 break;
2587 }
063639ba 2588#endif
537d97bc 2589
063639ba
MD
2590 name_len = strlen(name);
2591 if (name_len >= sizeof(hmp->iocom.auto_lnk_conn.fs_label))
2592 name_len = sizeof(hmp->iocom.auto_lnk_conn.fs_label) - 1;
2593 bcopy(name, hmp->iocom.auto_lnk_conn.fs_label, name_len);
2594 hmp->iocom.auto_lnk_conn.fs_label[name_len] = 0;
537d97bc 2595
063639ba 2596 kdmsg_iocom_autoinitiate(&hmp->iocom, hammer2_autodmsg);
70c3c3b7
MD
2597}
2598
9b8b748f 2599static int
537d97bc 2600hammer2_rcvdmsg(kdmsg_msg_t *msg)
9b8b748f 2601{
063639ba
MD
2602 kprintf("RCVMSG %08x\n", msg->tcmd);
2603
2604 switch(msg->tcmd) {
537d97bc 2605 case DMSG_DBG_SHELL:
10c86c4e 2606 /*
ea7c725f 2607 * (non-transaction)
537d97bc 2608 * Execute shell command (not supported atm)
10c86c4e 2609 */
063639ba 2610 kdmsg_msg_result(msg, DMSG_ERR_NOSUPP);
8c280d5d 2611 break;
537d97bc 2612 case DMSG_DBG_SHELL | DMSGF_REPLY:
ea7c725f
MD
2613 /*
2614 * (non-transaction)
2615 */
537d97bc
MD
2616 if (msg->aux_data) {
2617 msg->aux_data[msg->aux_size - 1] = 0;
2618 kprintf("HAMMER2 DBG: %s\n", msg->aux_data);
2619 }
8c280d5d
MD
2620 break;
2621 default:
3b76886b 2622 /*
ea7c725f 2623 * Unsupported message received. We only need to
3b76886b 2624 * reply if it's a transaction in order to close our end.
063639ba 2625 * Ignore any one-way messages or any further messages
3b76886b
MD
2626 * associated with the transaction.
2627 *
2628 * NOTE: This case also includes DMSG_LNK_ERROR messages
2629 * which might be one-way, replying to those would
2630 * cause an infinite ping-pong.
2631 */
2632 if (msg->any.head.cmd & DMSGF_CREATE)
2633 kdmsg_msg_reply(msg, DMSG_ERR_NOSUPP);
8c280d5d
MD
2634 break;
2635 }
2636 return(0);
2637}
2638
2639/*
537d97bc
MD
2640 * This function is called after KDMSG has automatically handled processing
2641 * of a LNK layer message (typically CONN, SPAN, or CIRC).
8c280d5d 2642 *
537d97bc
MD
2643 * We tag off the LNK_CONN to trigger our LNK_VOLCONF messages which
2644 * advertises all available hammer2 super-root volumes.
8c280d5d 2645 */
da6f36f4
MD
2646static void hammer2_update_spans(hammer2_mount_t *hmp, kdmsg_state_t *state);
2647
537d97bc
MD
2648static void
2649hammer2_autodmsg(kdmsg_msg_t *msg)
8c280d5d 2650{
da6f36f4 2651 hammer2_mount_t *hmp = msg->state->iocom->handle;
1a34728c 2652 int copyid;
8c280d5d 2653
063639ba 2654 kprintf("RCAMSG %08x\n", msg->tcmd);
537d97bc 2655
063639ba
MD
2656 switch(msg->tcmd) {
2657 case DMSG_LNK_CONN | DMSGF_CREATE | DMSGF_REPLY:
2658 case DMSG_LNK_CONN | DMSGF_CREATE | DMSGF_DELETE | DMSGF_REPLY:
2659 if (msg->any.head.cmd & DMSGF_CREATE) {
2660 kprintf("HAMMER2: VOLDATA DUMP\n");
70c3c3b7 2661
063639ba
MD
2662 /*
2663 * Dump the configuration stored in the volume header.
2664 * This will typically be import/export access rights,
2665 * master encryption keys (encrypted), etc.
2666 */
2667 hammer2_voldata_lock(hmp);
2668 copyid = 0;
2669 while (copyid < HAMMER2_COPYID_COUNT) {
2670 if (hmp->voldata.copyinfo[copyid].copyid)
2671 hammer2_volconf_update(hmp, copyid);
2672 ++copyid;
2673 }
2674 hammer2_voldata_unlock(hmp);
1a34728c 2675
063639ba 2676 kprintf("HAMMER2: INITIATE SPANs\n");
da6f36f4 2677 hammer2_update_spans(hmp, msg->state);
1a34728c 2678 }
063639ba
MD
2679 if ((msg->any.head.cmd & DMSGF_DELETE) &&
2680 msg->state && (msg->state->txcmd & DMSGF_DELETE) == 0) {
2681 kprintf("HAMMER2: CONN WAS TERMINATED\n");
2682 }
2683 break;
2684 default:
2685 break;
8c280d5d 2686 }
063639ba
MD
2687}
2688
2689/*
2690 * Update LNK_SPAN state
2691 */
da6f36f4
MD
2692static void
2693hammer2_update_spans(hammer2_mount_t *hmp, kdmsg_state_t *state)
063639ba
MD
2694{
2695 const hammer2_inode_data_t *ipdata;
2696 hammer2_cluster_t *cparent;
2697 hammer2_cluster_t *cluster;
2698 hammer2_pfsmount_t *spmp;
2699 hammer2_key_t key_next;
2700 kdmsg_msg_t *rmsg;
2701 size_t name_len;
2702 int ddflag;
2703
2704 /*
2705 * Lookup mount point under the media-localized super-root.
2706 *
2707 * cluster->pmp will incorrectly point to spmp and must be fixed
2708 * up later on.
2709 */
2710 spmp = hmp->spmp;
2711 cparent = hammer2_inode_lock_ex(spmp->iroot);
2712 cluster = hammer2_cluster_lookup(cparent, &key_next,
2713 HAMMER2_KEY_MIN,
2714 HAMMER2_KEY_MAX,
2715 0, &ddflag);
2716 while (cluster) {
2717 if (hammer2_cluster_type(cluster) != HAMMER2_BREF_TYPE_INODE)
2718 continue;
2719 ipdata = &hammer2_cluster_data(cluster)->ipdata;
2720 kprintf("UPDATE SPANS: %s\n", ipdata->filename);
2721
da6f36f4 2722 rmsg = kdmsg_msg_alloc(state, DMSG_LNK_SPAN | DMSGF_CREATE,
063639ba
MD
2723 hammer2_lnk_span_reply, NULL);
2724 rmsg->any.lnk_span.pfs_clid = ipdata->pfs_clid;
2725 rmsg->any.lnk_span.pfs_fsid = ipdata->pfs_fsid;
2726 rmsg->any.lnk_span.pfs_type = ipdata->pfs_type;
2727 rmsg->any.lnk_span.peer_type = DMSG_PEER_HAMMER2;
2728 rmsg->any.lnk_span.proto_version = DMSG_SPAN_PROTO_1;
2729 name_len = ipdata->name_len;
2730 if (name_len >= sizeof(rmsg->any.lnk_span.fs_label))
2731 name_len = sizeof(rmsg->any.lnk_span.fs_label) - 1;
2732 bcopy(ipdata->filename, rmsg->any.lnk_span.fs_label, name_len);
2733
2734 kdmsg_msg_write(rmsg);
2735
2736 cluster = hammer2_cluster_next(cparent, cluster,
2737 &key_next,
2738 key_next,
2739 HAMMER2_KEY_MAX,
2740 0);
8c280d5d 2741 }
063639ba
MD
2742 hammer2_inode_unlock_ex(spmp->iroot, cparent);
2743}
2744
2745static
2746int
2747hammer2_lnk_span_reply(kdmsg_state_t *state, kdmsg_msg_t *msg)
2748{
2749 if ((state->txcmd & DMSGF_DELETE) == 0 &&
2750 (msg->any.head.cmd & DMSGF_DELETE)) {
2751 kdmsg_msg_reply(msg, 0);
2752 }
2753 return 0;
9b8b748f 2754}
1a34728c
MD
2755
2756/*
2757 * Volume configuration updates are passed onto the userland service
2758 * daemon via the open LNK_CONN transaction.
2759 */
2760void
063639ba 2761hammer2_volconf_update(hammer2_mount_t *hmp, int index)
1a34728c 2762{
3a5aa68f 2763 kdmsg_msg_t *msg;
1a34728c
MD
2764
2765 /* XXX interlock against connection state termination */
063639ba
MD
2766 kprintf("volconf update %p\n", hmp->iocom.conn_state);
2767 if (hmp->iocom.conn_state) {
1a34728c 2768 kprintf("TRANSMIT VOLCONF VIA OPEN CONN TRANSACTION\n");
da6f36f4
MD
2769 msg = kdmsg_msg_alloc(hmp->iocom.conn_state,
2770 DMSG_LNK_HAMMER2_VOLCONF,
2771 NULL, NULL);
1f179146
MD
2772 H2_LNK_VOLCONF(msg)->copy = hmp->voldata.copyinfo[index];
2773 H2_LNK_VOLCONF(msg)->mediaid = hmp->voldata.fsid;
2774 H2_LNK_VOLCONF(msg)->index = index;
3a5aa68f 2775 kdmsg_msg_write(msg);
1a34728c
MD
2776 }
2777}
9797e933 2778
3f5b8b3b
MD
2779/*
2780 * This handles hysteresis on regular file flushes. Because the BIOs are
2781 * routed to a thread it is possible for an excessive number to build up
2782 * and cause long front-end stalls long before the runningbuffspace limit
2783 * is hit, so we implement hammer2_flush_pipe to control the
2784 * hysteresis.
2785 *
2786 * This is a particular problem when compression is used.
2787 */
2788void
2789hammer2_lwinprog_ref(hammer2_pfsmount_t *pmp)
2790{
2791 atomic_add_int(&pmp->count_lwinprog, 1);
2792}
2793
2794void
2795hammer2_lwinprog_drop(hammer2_pfsmount_t *pmp)
2796{
2797 int lwinprog;
2798
2799 lwinprog = atomic_fetchadd_int(&pmp->count_lwinprog, -1);
2800 if ((lwinprog & HAMMER2_LWINPROG_WAITING) &&
2801 (lwinprog & HAMMER2_LWINPROG_MASK) <= hammer2_flush_pipe * 2 / 3) {
2802 atomic_clear_int(&pmp->count_lwinprog,
2803 HAMMER2_LWINPROG_WAITING);
2804 wakeup(&pmp->count_lwinprog);
2805 }
2806}
2807
2808void
2809hammer2_lwinprog_wait(hammer2_pfsmount_t *pmp)
2810{
2811 int lwinprog;
2812
2813 for (;;) {
2814 lwinprog = pmp->count_lwinprog;
2815 cpu_ccfence();
2816 if ((lwinprog & HAMMER2_LWINPROG_MASK) < hammer2_flush_pipe)
2817 break;
2818 tsleep_interlock(&pmp->count_lwinprog, 0);
2819 atomic_set_int(&pmp->count_lwinprog, HAMMER2_LWINPROG_WAITING);
2820 lwinprog = pmp->count_lwinprog;
2821 if ((lwinprog & HAMMER2_LWINPROG_MASK) < hammer2_flush_pipe)
2822 break;
2823 tsleep(&pmp->count_lwinprog, PINTERLOCKED, "h2wpipe", hz);
2824 }
2825}
2826
278ab2b2
MD
2827/*
2828 * Manage excessive memory resource use for chain and related
2829 * structures.
2830 */
2831void
2832hammer2_pfs_memory_wait(hammer2_pfsmount_t *pmp)
2833{
da6f36f4
MD
2834 uint32_t waiting;
2835 uint32_t count;
2836 uint32_t limit;
278ab2b2
MD
2837#if 0
2838 static int zzticks;
2839#endif
2840
2841 /*
2842 * Atomic check condition and wait. Also do an early speedup of
2843 * the syncer to try to avoid hitting the wait.
2844 */
2845 for (;;) {
2846 waiting = pmp->inmem_dirty_chains;
2847 cpu_ccfence();
2848 count = waiting & HAMMER2_DIRTYCHAIN_MASK;
2849
2850 limit = pmp->mp->mnt_nvnodelistsize / 10;
2851 if (limit < hammer2_limit_dirty_chains)
2852 limit = hammer2_limit_dirty_chains;
2853 if (limit < 1000)
2854 limit = 1000;
2855
2856#if 0
2857 if ((int)(ticks - zzticks) > hz) {
2858 zzticks = ticks;
2859 kprintf("count %ld %ld\n", count, limit);
2860 }
2861#endif
2862
2863 /*
2864 * Block if there are too many dirty chains present, wait
2865 * for the flush to clean some out.
2866 */
2867 if (count > limit) {
2868 tsleep_interlock(&pmp->inmem_dirty_chains, 0);
da6f36f4 2869 if (atomic_cmpset_int(&pmp->inmem_dirty_chains,
278ab2b2
MD
2870 waiting,
2871 waiting | HAMMER2_DIRTYCHAIN_WAITING)) {
2872 speedup_syncer(pmp->mp);
2873 tsleep(&pmp->inmem_dirty_chains, PINTERLOCKED,
2874 "chnmem", hz);
2875 }
2876 continue; /* loop on success or fail */
2877 }
2878
2879 /*
2880 * Try to start an early flush before we are forced to block.
2881 */
2882 if (count > limit * 7 / 10)
2883 speedup_syncer(pmp->mp);
2884 break;
2885 }
2886}
2887
2888void
2889hammer2_pfs_memory_inc(hammer2_pfsmount_t *pmp)
2890{
da6f36f4
MD
2891 if (pmp) {
2892 atomic_add_int(&pmp->inmem_dirty_chains, 1);
2893 }
278ab2b2
MD
2894}
2895
2896void
2897hammer2_pfs_memory_wakeup(hammer2_pfsmount_t *pmp)
2898{
da6f36f4 2899 uint32_t waiting;
278ab2b2
MD
2900
2901 if (pmp == NULL)
2902 return;
2903
2904 for (;;) {
2905 waiting = pmp->inmem_dirty_chains;
2906 cpu_ccfence();
da6f36f4 2907 if (atomic_cmpset_int(&pmp->inmem_dirty_chains,
278ab2b2
MD
2908 waiting,
2909 (waiting - 1) &
2910 ~HAMMER2_DIRTYCHAIN_WAITING)) {
2911 break;
2912 }
2913 }
2914
2915 if (waiting & HAMMER2_DIRTYCHAIN_WAITING)
2916 wakeup(&pmp->inmem_dirty_chains);
2917}
2918
2919/*
2920 * Debugging
2921 */
9797e933 2922void
8138a154 2923hammer2_dump_chain(hammer2_chain_t *chain, int tab, int *countp, char pfx)
9797e933
MD
2924{
2925 hammer2_chain_t *scan;
da6f36f4 2926 hammer2_chain_t *parent;
9797e933 2927
a02dfba1
MD
2928 --*countp;
2929 if (*countp == 0) {
2930 kprintf("%*.*s...\n", tab, tab, "");
2931 return;
2932 }
2933 if (*countp < 0)
2934 return;
8138a154
MD
2935 kprintf("%*.*s%c-chain %p.%d %016jx/%d mir=%016jx\n",
2936 tab, tab, "", pfx,
a7720be7
MD
2937 chain, chain->bref.type,
2938 chain->bref.key, chain->bref.keybits,
2939 chain->bref.mirror_tid);
2940
da6f36f4 2941 kprintf("%*.*s [%08x] (%s) refs=%d\n",
a7720be7
MD
2942 tab, tab, "",
2943 chain->flags,
9797e933
MD
2944 ((chain->bref.type == HAMMER2_BREF_TYPE_INODE &&
2945 chain->data) ? (char *)chain->data->ipdata.filename : "?"),
8853dfb5 2946 chain->refs);
a7720be7 2947
da6f36f4 2948 kprintf("%*.*s core [%08x]",
a7720be7 2949 tab, tab, "",
da6f36f4 2950 chain->core.flags);
a7720be7 2951
da6f36f4
MD
2952 parent = chain->parent;
2953 if (parent)
2954 kprintf("\n%*.*s p=%p [pflags %08x prefs %d",
8138a154 2955 tab, tab, "",
da6f36f4
MD
2956 parent, parent->flags, parent->refs);
2957 if (RB_EMPTY(&chain->core.rbtree)) {
9797e933 2958 kprintf("\n");
da6f36f4 2959 } else {
9797e933 2960 kprintf(" {\n");
da6f36f4 2961 RB_FOREACH(scan, hammer2_chain_tree, &chain->core.rbtree)
8138a154 2962 hammer2_dump_chain(scan, tab + 4, countp, 'a');
9797e933
MD
2963 if (chain->bref.type == HAMMER2_BREF_TYPE_INODE && chain->data)
2964 kprintf("%*.*s}(%s)\n", tab, tab, "",
2965 chain->data->ipdata.filename);
2966 else
2967 kprintf("%*.*s}\n", tab, tab, "");
2968 }
2969}